diff options
Diffstat (limited to 'erts/emulator/beam')
90 files changed, 15233 insertions, 4532 deletions
diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names index d28e519ae1..ae3f30d82f 100644 --- a/erts/emulator/beam/atom.names +++ b/erts/emulator/beam/atom.names @@ -109,6 +109,7 @@ atom bag atom band atom big atom bif_return_trap +atom bif_timer_server atom binary atom binary_bin_to_list_trap atom binary_copy_trap @@ -116,6 +117,7 @@ atom binary_longest_prefix_trap atom binary_longest_suffix_trap atom binary_match_trap atom binary_matches_trap +atom binary_to_list_continue atom binary_to_term_trap atom block atom blocked @@ -143,9 +145,11 @@ atom catchlevel atom cd atom cdr atom cflags +atom CHANGE='CHANGE' atom characters_to_binary_int atom characters_to_list_int atom clear +atom clock_service atom close atom closed atom code @@ -155,6 +159,7 @@ atom compat_rel atom compile atom compressed atom config_h +atom convert_time_unit atom connect atom connected atom connection_closed @@ -197,6 +202,7 @@ atom dotall atom driver atom driver_options atom dsend +atom dsend_continue_trap atom dunlink atom duplicate_bag atom dupnames @@ -235,7 +241,7 @@ atom first atom firstline atom flags atom flush -atom flush_monitor_message +atom flush_monitor_messages atom force atom format_cpu_topology atom free @@ -278,7 +284,6 @@ atom http httph https http_response http_request http_header http_eoh http_error atom id atom if_clause atom ignore -atom imports atom in atom in_exiting atom inactive @@ -315,6 +320,7 @@ atom line_length atom linked_in_driver atom links atom list +atom list_to_binary_continue atom little atom loaded atom load_cancelled @@ -333,6 +339,7 @@ atom max atom maximum atom max_tables max_processes atom mbuf_size +atom md5 atom memory atom memory_internal atom memory_types @@ -342,6 +349,8 @@ atom message_queue_len atom messages atom meta atom meta_match_spec +atom micro_seconds +atom milli_seconds atom min_heap_size atom min_bin_vheap_size atom minor_version @@ -352,12 +361,15 @@ atom monitored_by atom monitor atom monitor_nodes atom monitors +atom monotonic atom more atom multi_scheduling atom multiline +atom nano_seconds atom name atom named_table atom namelist +atom native atom native_addresses atom Neq='=/=' atom Neqeq='/=' @@ -448,6 +460,7 @@ atom ports atom port_count atom port_limit atom port_op +atom positive atom print atom priority atom private @@ -507,6 +520,7 @@ atom schedulers_online atom scheme atom scientific atom scope +atom seconds atom sensitive atom sequential_tracer atom sequential_trace_token @@ -552,6 +566,7 @@ atom term_to_binary_trap atom this atom thread_pool_size atom threads +atom time_offset atom timeout atom timeout_value atom Times='*' diff --git a/erts/emulator/beam/beam_debug.c b/erts/emulator/beam/beam_debug.c index a3cd08834f..6bb987985d 100644 --- a/erts/emulator/beam/beam_debug.c +++ b/erts/emulator/beam/beam_debug.c @@ -579,9 +579,29 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) unpacked = ap; ap = addr + size; switch (op) { - case op_i_select_val_rfI: - case op_i_select_val_xfI: - case op_i_select_val_yfI: + case op_i_select_val_lins_rfI: + case op_i_select_val_lins_xfI: + case op_i_select_val_lins_yfI: + { + int n = ap[-1]; + int ix = n; + + while (ix--) { + erts_print(to, to_arg, "%T ", (Eterm) ap[0]); + ap++; + size++; + } + ix = n; + while (ix--) { + erts_print(to, to_arg, "f(" HEXF ") ", (Eterm) ap[0]); + ap++; + size++; + } + } + break; + case op_i_select_val_bins_rfI: + case op_i_select_val_bins_xfI: + case op_i_select_val_bins_yfI: { int n = ap[-1]; @@ -598,13 +618,19 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr) case op_i_select_tuple_arity_yfI: { int n = ap[-1]; + int ix = n; - while (n > 0) { + while (ix--) { Uint arity = arityval(ap[0]); - erts_print(to, to_arg, " {%d} f(" HEXF ")", arity, ap[1]); - ap += 2; - size += 2; - n--; + erts_print(to, to_arg, "{%d} ", arity, ap[1]); + ap++; + size++; + } + ix = n; + while (ix--) { + erts_print(to, to_arg, "f(" HEXF ") ", ap[0]); + ap++; + size++; } } break; diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index 1026e5f649..8fcdc72895 100644 --- a/erts/emulator/beam/beam_emu.c +++ b/erts/emulator/beam/beam_emu.c @@ -241,10 +241,6 @@ BeamInstr beam_return_time_trace[1]; /* OpCode(i_return_time_trace) */ void** beam_ops; #endif -#ifndef ERTS_SMP /* Not supported with smp emulator */ -extern int count_instructions; -#endif - #define SWAPIN \ HTOP = HEAP_TOP(c_p); \ E = c_p->stop @@ -703,7 +699,7 @@ extern int count_instructions; Fail; \ } -#define IsMap(Src, Fail) if (is_not_map(Src)) { Fail; } +#define IsMap(Src, Fail) if (!is_map(Src)) { Fail; } #define HasMapField(Src, Key, Fail) if (has_not_map_field(Src, Key)) { Fail; } @@ -1163,14 +1159,15 @@ void process_main(void) Eterm (*arith_func)(Process* p, Eterm* reg, Uint live); -#ifndef NO_JUMP_TABLE - static void* opcodes[] = { DEFINE_OPCODES }; #ifdef ERTS_OPCODE_COUNTER_SUPPORT static void* counting_opcodes[] = { DEFINE_COUNTING_OPCODES }; -#endif +#else +#ifndef NO_JUMP_TABLE + static void* opcodes[] = { DEFINE_OPCODES }; #else int Go; #endif +#endif Uint temp_bits; /* Temporary used by BsSkipBits2 & BsGetInteger2 */ @@ -2141,19 +2138,18 @@ void process_main(void) NextPF(0, next); } - { Eterm select_val2; - OpCase(i_select_tuple_arity2_yfAfAf): + OpCase(i_select_tuple_arity2_yfAAff): select_val2 = yb(Arg(0)); goto do_select_tuple_arity2; - OpCase(i_select_tuple_arity2_xfAfAf): + OpCase(i_select_tuple_arity2_xfAAff): select_val2 = xb(Arg(0)); goto do_select_tuple_arity2; - OpCase(i_select_tuple_arity2_rfAfAf): + OpCase(i_select_tuple_arity2_rfAAff): select_val2 = r(0); I--; @@ -2164,22 +2160,22 @@ void process_main(void) select_val2 = *tuple_val(select_val2); goto do_select_val2; - OpCase(i_select_val2_yfcfcf): + OpCase(i_select_val2_yfccff): select_val2 = yb(Arg(0)); goto do_select_val2; - OpCase(i_select_val2_xfcfcf): + OpCase(i_select_val2_xfccff): select_val2 = xb(Arg(0)); goto do_select_val2; - OpCase(i_select_val2_rfcfcf): + OpCase(i_select_val2_rfccff): select_val2 = r(0); I--; do_select_val2: if (select_val2 == Arg(2)) { - I += 2; - } else if (select_val2 == Arg(4)) { + I += 3; + } else if (select_val2 == Arg(3)) { I += 4; } @@ -2206,20 +2202,50 @@ void process_main(void) do_select_tuple_arity: if (is_tuple(select_val)) { select_val = *tuple_val(select_val); - goto do_binary_search; + goto do_linear_search; } SET_I((BeamInstr *) Arg(1)); Goto(*I); - OpCase(i_select_val_xfI): + OpCase(i_select_val_lins_xfI): + select_val = xb(Arg(0)); + goto do_linear_search; + + OpCase(i_select_val_lins_yfI): + select_val = yb(Arg(0)); + goto do_linear_search; + + OpCase(i_select_val_lins_rfI): + select_val = r(0); + I--; + + do_linear_search: { + BeamInstr *vs = &Arg(3); + int ix = 0; + + for(;;) { + if (vs[ix+0] >= select_val) { ix += 0; break; } + if (vs[ix+1] >= select_val) { ix += 1; break; } + ix += 2; + } + + if (vs[ix] == select_val) { + I += ix + Arg(2) + 2; + } + + SET_I((BeamInstr *) Arg(1)); + Goto(*I); + } + + OpCase(i_select_val_bins_xfI): select_val = xb(Arg(0)); goto do_binary_search; - OpCase(i_select_val_yfI): + OpCase(i_select_val_bins_yfI): select_val = yb(Arg(0)); goto do_binary_search; - OpCase(i_select_val_rfI): + OpCase(i_select_val_bins_rfI): select_val = r(0); I--; @@ -2366,7 +2392,7 @@ void process_main(void) } OpCase(i_has_map_fields_fsI): { - map_t* mp; + flatmap_t* mp; Eterm map; Eterm field; Eterm *ks; @@ -2374,22 +2400,34 @@ void process_main(void) Uint sz,n; GetArg1(1, map); + n = (Uint)Arg(2); + fs = &Arg(3); /* pattern fields */ - /* this instruction assumes Arg1 is a map, - * i.e. that it follows a test is_map if needed. - */ + /* get term from field? */ + if (is_hashmap(map)) { + Uint32 hx; + while(n--) { + field = *fs++; + hx = hashmap_make_hash(field); + if (!erts_hashmap_get(hx,field,map)) { + SET_I((BeamInstr *) Arg(0)); + goto has_map_fields_fail; + } + } + goto has_map_fields_ok; + } - mp = (map_t *)map_val(map); - sz = map_get_size(mp); + ASSERT(is_flatmap(map)); + + mp = (flatmap_t *)flatmap_val(map); + sz = flatmap_get_size(mp); if (sz == 0) { SET_I((BeamInstr *) Arg(0)); goto has_map_fields_fail; } - ks = map_get_keys(mp); - n = (Uint)Arg(2); - fs = &Arg(3); /* pattern fields */ + ks = flatmap_get_keys(mp); ASSERT(n>0); @@ -2407,7 +2445,7 @@ void process_main(void) SET_I((BeamInstr *) Arg(0)); goto has_map_fields_fail; } - +has_map_fields_ok: I += 4 + Arg(2); has_map_fields_fail: ASSERT(VALID_INSTR(*I)); @@ -2434,12 +2472,8 @@ do { \ OpCase(i_get_map_elements_fsI): { Eterm map; - map_t *mp; - Eterm field; - Eterm *ks; - Eterm *vs; BeamInstr *fs; - Uint sz,n; + Uint sz, n; GetArg1(1, map); @@ -2447,36 +2481,56 @@ do { \ * i.e. that it follows a test is_map if needed. */ - mp = (map_t *)map_val(map); - sz = map_get_size(mp); - - if (sz == 0) { - SET_I((BeamInstr *) Arg(0)); - goto get_map_elements_fail; - } - n = (Uint)Arg(2) / 2; fs = &Arg(3); /* pattern fields and target registers */ - ks = map_get_keys(mp); - vs = map_get_values(mp); - while(sz) { - field = (Eterm)*fs; - if (EQ(field,*ks)) { - PUT_TERM_REG(*vs, fs[1]); - n--; + if (is_flatmap(map)) { + flatmap_t *mp; + Eterm *ks; + Eterm *vs; + + mp = (flatmap_t *)flatmap_val(map); + sz = flatmap_get_size(mp); + + if (sz == 0) { + SET_I((BeamInstr *) Arg(0)); + goto get_map_elements_fail; + } + + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + while(sz) { + if (EQ((Eterm)*fs,*ks)) { + PUT_TERM_REG(*vs, fs[1]); + n--; + fs += 2; + /* no more values to fetch, we are done */ + if (n == 0) break; + } + ks++; sz--; + vs++; + } + + if (n) { + SET_I((BeamInstr *) Arg(0)); + goto get_map_elements_fail; + } + } else { + const Eterm *v; + Uint32 hx; + ASSERT(is_hashmap(map)); + while(n--) { + hx = hashmap_make_hash((Eterm)*fs); + if ((v = erts_hashmap_get(hx,(Eterm)*fs, map)) == NULL) { + SET_I((BeamInstr *) Arg(0)); + goto get_map_elements_fail; + } + PUT_TERM_REG(*v, fs[1]); fs += 2; - /* no more values to fetch, we are done */ - if (n == 0) break; } - ks++; sz--; - vs++; } - if (n) { - SET_I((BeamInstr *) Arg(0)); - goto get_map_elements_fail; - } I += 4 + Arg(2); get_map_elements_fail: @@ -2775,6 +2829,7 @@ get_map_elements_fail: } PreFetch(1, next); ASSERT(!ERTS_PROC_IS_EXITING(c_p)); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p); reg[0] = r(0); result = (*bf)(c_p, reg, I); ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(result)); @@ -3503,6 +3558,7 @@ get_map_elements_fail: * I[0]: &&call_nif * I[1]: Function pointer to NIF function * I[2]: Pointer to erl_module_nif + * I[3]: Function pointer to dirty NIF */ BifFunction vbf; @@ -3523,13 +3579,6 @@ get_map_elements_fail: reg[0] = r(0); nif_bif_result = (*fp)(&env, bif_nif_arity, reg); erts_post_nif(&env); -#ifdef ERTS_DIRTY_SCHEDULERS - if (is_non_value(nif_bif_result) && c_p->freason == TRAP) { - Export* ep = ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(c_p); - ep->code[0] = I[-3]; - ep->code[1] = I[-2]; - } -#endif } ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(nif_bif_result)); PROCESS_MAIN_CHK_LOCKS(c_p); @@ -3562,7 +3611,7 @@ get_map_elements_fail: vbf = (BifFunction) Arg(0); PROCESS_MAIN_CHK_LOCKS(c_p); bif_nif_arity = I[-1]; - ASSERT(bif_nif_arity <= 3); + ASSERT(bif_nif_arity <= 4); ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p); reg[0] = r(0); { @@ -3788,8 +3837,6 @@ get_map_elements_fail: * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(num_bytes); - bptr->flags = 0; - bptr->orig_size = num_bytes; erts_refc_init(&bptr->refc, 1); erts_current_bin = (byte *) bptr->orig_bytes; @@ -3889,8 +3936,6 @@ get_map_elements_fail: * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(tmp_arg1); - bptr->flags = 0; - bptr->orig_size = tmp_arg1; erts_refc_init(&bptr->refc, 1); erts_current_bin = (byte *) bptr->orig_bytes; @@ -4993,14 +5038,14 @@ get_map_elements_fail: * ... remainder of original BEAM code */ ASSERT(I[-5] == (Uint) OpCode(i_func_info_IaaI)); - c_p->hipe.ncallee = (void(*)(void)) I[-4]; + c_p->hipe.u.ncallee = (void(*)(void)) I[-4]; cmd = HIPE_MODE_SWITCH_CMD_CALL | (I[-1] << 8); ++hipe_trap_count; goto L_hipe_mode_switch; } OpCase(hipe_trap_call_closure): { ASSERT(I[-5] == (Uint) OpCode(i_func_info_IaaI)); - c_p->hipe.ncallee = (void(*)(void)) I[-4]; + c_p->hipe.u.ncallee = (void(*)(void)) I[-4]; cmd = HIPE_MODE_SWITCH_CMD_CALL_CLOSURE | (I[-1] << 8); ++hipe_trap_count; goto L_hipe_mode_switch; @@ -5034,7 +5079,10 @@ get_map_elements_fail: case HIPE_MODE_SWITCH_RES_RETURN: ASSERT(is_value(reg[0])); MoveReturn(reg[0], r(0)); - case HIPE_MODE_SWITCH_RES_CALL: + case HIPE_MODE_SWITCH_RES_CALL_EXPORTED: + c_p->i = c_p->hipe.u.callee_exp->addressv[erts_active_code_ix()]; + /*fall through*/ + case HIPE_MODE_SWITCH_RES_CALL_BEAM: SET_I(c_p->i); r(0) = reg[0]; Dispatch(); @@ -5151,22 +5199,16 @@ get_map_elements_fail: #ifndef NO_JUMP_TABLE #ifdef ERTS_OPCODE_COUNTER_SUPPORT - /* Are tables correctly generated by beam_makeops? */ - ASSERT(sizeof(counting_opcodes) == sizeof(opcodes)); - - if (count_instructions) { + ERTS_CT_ASSERT(sizeof(counting_opcodes) == sizeof(opcodes)); #ifdef DEBUG - counting_opcodes[op_catch_end_y] = LabelAddr(lb_catch_end_y); + counting_opcodes[op_catch_end_y] = LabelAddr(lb_catch_end_y); #endif - counting_opcodes[op_i_func_info_IaaI] = LabelAddr(lb_i_func_info_IaaI); - beam_ops = counting_opcodes; - } - else -#endif /* #ifndef ERTS_OPCODE_COUNTER_SUPPORT */ - { - beam_ops = opcodes; - } + counting_opcodes[op_i_func_info_IaaI] = LabelAddr(lb_i_func_info_IaaI); + beam_ops = counting_opcodes; +#else /* #ifndef ERTS_OPCODE_COUNTER_SUPPORT */ + beam_ops = opcodes; +#endif /* ERTS_OPCODE_COUNTER_SUPPORT */ #endif /* NO_JUMP_TABLE */ em_call_error_handler = OpCode(call_error_handler); @@ -6430,55 +6472,69 @@ new_fun(Process* p, Eterm* reg, ErlFunEntry* fe, int num_free) static int has_not_map_field(Eterm map, Eterm key) { - map_t* mp; - Eterm* keys; - Uint i; - Uint n; - - mp = (map_t *)map_val(map); - keys = map_get_keys(mp); - n = map_get_size(mp); - if (is_immed(key)) { - for (i = 0; i < n; i++) { - if (keys[i] == key) { - return 0; + Uint32 hx; + if (is_flatmap(map)) { + flatmap_t* mp; + Eterm* keys; + Uint i; + Uint n; + + mp = (flatmap_t *)flatmap_val(map); + keys = flatmap_get_keys(mp); + n = flatmap_get_size(mp); + if (is_immed(key)) { + for (i = 0; i < n; i++) { + if (keys[i] == key) { + return 0; + } } - } - } else { - for (i = 0; i < n; i++) { - if (EQ(keys[i], key)) { - return 0; + } else { + for (i = 0; i < n; i++) { + if (EQ(keys[i], key)) { + return 0; + } } } + return 1; } - return 1; + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + return erts_hashmap_get(hx,key,map) ? 0 : 1; } static Eterm get_map_element(Eterm map, Eterm key) { - map_t *mp; - Eterm* ks, *vs; - Uint i; - Uint n; - - mp = (map_t *)map_val(map); - ks = map_get_keys(mp); - vs = map_get_values(mp); - n = map_get_size(mp); - if (is_immed(key)) { - for (i = 0; i < n; i++) { - if (ks[i] == key) { - return vs[i]; + Uint32 hx; + const Eterm *vs; + if (is_flatmap(map)) { + flatmap_t *mp; + Eterm *ks; + Uint i; + Uint n; + + mp = (flatmap_t *)flatmap_val(map); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + n = flatmap_get_size(mp); + if (is_immed(key)) { + for (i = 0; i < n; i++) { + if (ks[i] == key) { + return vs[i]; + } } - } - } else { - for (i = 0; i < n; i++) { - if (EQ(ks[i], key)) { - return vs[i]; + } else { + for (i = 0; i < n; i++) { + if (EQ(ks[i], key)) { + return vs[i]; + } } } + return THE_NON_VALUE; } - return THE_NON_VALUE; + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + vs = erts_hashmap_get(hx,key,map); + return vs ? *vs : THE_NON_VALUE; } #define GET_TERM(term, dest) \ @@ -6511,7 +6567,30 @@ new_map(Process* p, Eterm* reg, BeamInstr* I) Eterm *mhp,*thp; Eterm *E; BeamInstr *ptr; - map_t *mp; + flatmap_t *mp; + ErtsHeapFactory factory; + + ptr = &Arg(4); + + if (n > 2*MAP_SMALL_MAP_LIMIT) { + if (HeapWordsLeft(p) < n) { + erts_garbage_collect(p, n, reg, Arg(2)); + } + + mhp = p->htop; + thp = p->htop; + E = p->stop; + + for (i = 0; i < n/2; i++) { + GET_TERM(*ptr++, *mhp++); + GET_TERM(*ptr++, *mhp++); + } + + p->htop = mhp; + + factory.p = p; + return erts_hashmap_from_array(&factory, thp, n/2, 0); + } if (HeapWordsLeft(p) < need) { erts_garbage_collect(p, need, reg, Arg(2)); @@ -6520,12 +6599,11 @@ new_map(Process* p, Eterm* reg, BeamInstr* I) thp = p->htop; mhp = thp + 1 + n/2; E = p->stop; - ptr = &Arg(4); keys = make_tuple(thp); *thp++ = make_arityval(n/2); - mp = (map_t *)mhp; mhp += MAP_HEADER_SIZE; - mp->thing_word = MAP_HEADER; + mp = (flatmap_t *)mhp; mhp += MAP_HEADER_FLATMAP_SZ; + mp->thing_word = MAP_HEADER_FLATMAP; mp->size = n/2; mp->keys = keys; @@ -6534,7 +6612,7 @@ new_map(Process* p, Eterm* reg, BeamInstr* I) GET_TERM(*ptr++, *mhp++); } p->htop = mhp; - return make_map(mp); + return make_flatmap(mp); } static Eterm @@ -6544,7 +6622,7 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) Uint num_old; Uint num_updates; Uint need; - map_t *old_mp, *mp; + flatmap_t *old_mp, *mp; Eterm res; Eterm* hp; Eterm* E; @@ -6554,12 +6632,44 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) Eterm new_key; Eterm* kp; - if (is_not_map(map)) { - return THE_NON_VALUE; + new_p = &Arg(5); + num_updates = Arg(4) / 2; + + if (is_not_flatmap(map)) { + Uint32 hx; + Eterm val; + + /* apparently the compiler does not emit is_map instructions, + * bad compiler */ + + if (is_not_hashmap(map)) + return THE_NON_VALUE; + + res = map; + E = p->stop; + while(num_updates--) { + /* assoc can't fail */ + GET_TERM(new_p[0], new_key); + GET_TERM(new_p[1], val); + hx = hashmap_make_hash(new_key); + + res = erts_hashmap_insert(p, hx, new_key, val, res, 0); + if (p->mbuf) { + Uint live = Arg(3); + reg[live] = res; + erts_garbage_collect(p, 0, reg, live+1); + res = reg[live]; + } + + E = p->stop; + + new_p += 2; + } + return res; } - old_mp = (map_t *) map_val(map); - num_old = map_get_size(old_mp); + old_mp = (flatmap_t *) flatmap_val(map); + num_old = flatmap_get_size(old_mp); /* * If the old map is empty, create a new map. @@ -6574,14 +6684,13 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) * update list are new). */ - num_updates = Arg(4) / 2; - need = 2*(num_old+num_updates) + 1 + MAP_HEADER_SIZE; + need = 2*(num_old+num_updates) + 1 + MAP_HEADER_FLATMAP_SZ; if (HeapWordsLeft(p) < need) { Uint live = Arg(3); reg[live] = map; erts_garbage_collect(p, need, reg, live+1); map = reg[live]; - old_mp = (map_t *)map_val(map); + old_mp = (flatmap_t *)flatmap_val(map); } /* @@ -6612,16 +6721,15 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) kp = p->htop + 1; /* Point to first key */ hp = kp + num_old + num_updates; - res = make_map(hp); - mp = (map_t *)hp; - hp += MAP_HEADER_SIZE; - mp->thing_word = MAP_HEADER; + res = make_flatmap(hp); + mp = (flatmap_t *)hp; + hp += MAP_HEADER_FLATMAP_SZ; + mp->thing_word = MAP_HEADER_FLATMAP; mp->keys = make_tuple(kp-1); - old_vals = map_get_values(old_mp); - old_keys = map_get_keys(old_mp); + old_vals = flatmap_get_values(old_mp); + old_keys = flatmap_get_keys(old_mp); - new_p = &Arg(5); GET_TERM(*new_p, new_key); n = num_updates; @@ -6707,8 +6815,19 @@ update_map_assoc(Process* p, Eterm* reg, Eterm map, BeamInstr* I) n = kp - p->htop - 1; /* Actual number of keys/values */ *p->htop = make_arityval(n); + p->htop = hp; mp->size = n; - p->htop = hp; + + /* The expensive case, need to build a hashmap */ + if (n > MAP_SMALL_MAP_LIMIT) { + res = erts_hashmap_from_ks_and_vs(p,flatmap_get_keys(mp),flatmap_get_values(mp),n); + if (p->mbuf) { + Uint live = Arg(3); + reg[live] = res; + erts_garbage_collect(p, 0, reg, live+1); + res = reg[live]; + } + } return res; } @@ -6723,7 +6842,7 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) Uint i; Uint num_old; Uint need; - map_t *old_mp, *mp; + flatmap_t *old_mp, *mp; Eterm res; Eterm* hp; Eterm* E; @@ -6732,12 +6851,48 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) BeamInstr* new_p; Eterm new_key; - if (is_not_map(map)) { - return THE_NON_VALUE; + new_p = &Arg(5); + n = Arg(4) / 2; /* Number of values to be updated */ + ASSERT(n > 0); + + if (is_not_flatmap(map)) { + Uint32 hx; + Eterm val; + + /* apparently the compiler does not emit is_map instructions, + * bad compiler */ + + if (is_not_hashmap(map)) + return THE_NON_VALUE; + + res = map; + E = p->stop; + while(n--) { + /* assoc can't fail */ + GET_TERM(new_p[0], new_key); + GET_TERM(new_p[1], val); + hx = hashmap_make_hash(new_key); + + res = erts_hashmap_insert(p, hx, new_key, val, res, 1); + if (is_non_value(res)) + return res; + + if (p->mbuf) { + Uint live = Arg(3); + reg[live] = res; + erts_garbage_collect(p, 0, reg, live+1); + res = reg[live]; + } + + E = p->stop; + + new_p += 2; + } + return res; } - old_mp = (map_t *) map_val(map); - num_old = map_get_size(old_mp); + old_mp = (flatmap_t *) flatmap_val(map); + num_old = flatmap_get_size(old_mp); /* * If the old map is empty, create a new map. @@ -6751,13 +6906,13 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) * Allocate the exact heap space needed. */ - need = num_old + MAP_HEADER_SIZE; + need = num_old + MAP_HEADER_FLATMAP_SZ; if (HeapWordsLeft(p) < need) { Uint live = Arg(3); reg[live] = map; erts_garbage_collect(p, need, reg, live+1); map = reg[live]; - old_mp = (map_t *)map_val(map); + old_mp = (flatmap_t *)flatmap_val(map); } /* @@ -6767,23 +6922,20 @@ update_map_exact(Process* p, Eterm* reg, Eterm map, BeamInstr* I) hp = p->htop; E = p->stop; - old_vals = map_get_values(old_mp); - old_keys = map_get_keys(old_mp); + old_vals = flatmap_get_values(old_mp); + old_keys = flatmap_get_keys(old_mp); - res = make_map(hp); - mp = (map_t *)hp; - hp += MAP_HEADER_SIZE; - mp->thing_word = MAP_HEADER; + res = make_flatmap(hp); + mp = (flatmap_t *)hp; + hp += MAP_HEADER_FLATMAP_SZ; + mp->thing_word = MAP_HEADER_FLATMAP; mp->size = num_old; mp->keys = old_mp->keys; /* Get array of key/value pairs to be updated */ - new_p = &Arg(5); GET_TERM(*new_p, new_key); /* Update all values */ - n = Arg(4) / 2; /* Number of values to be updated */ - ASSERT(n > 0); for (i = 0; i < num_old; i++) { if (!EQ(*old_keys, new_key)) { /* Not same keys */ diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c index e96177cfd9..02689e5b19 100644 --- a/erts/emulator/beam/beam_load.c +++ b/erts/emulator/beam/beam_load.c @@ -245,7 +245,7 @@ typedef struct { /* * This structure contains all information about the module being loaded. */ - +#define MD5_SIZE 16 typedef struct LoaderState { /* * The current logical file within the binary. @@ -292,7 +292,7 @@ typedef struct LoaderState { StringPatch* string_patches; /* Linked list of position into string table to patch. */ BeamInstr catches; /* Linked list of catch_yf instructions. */ unsigned loaded_size; /* Final size of code when loaded. */ - byte mod_md5[16]; /* MD5 for module code. */ + byte mod_md5[MD5_SIZE]; /* MD5 for module code. */ int may_load_nif; /* true if NIFs may later be loaded for this module */ int on_load; /* Index in the code for the on_load function * (or 0 if there is no on_load function) @@ -528,6 +528,7 @@ static Eterm exported_from_module(Process* p, Eterm mod); static Eterm functions_in_module(Process* p, Eterm mod); static Eterm attributes_for_module(Process* p, Eterm mod); static Eterm compilation_info_for_module(Process* p, Eterm mod); +static Eterm md5_of_module(Process* p, Eterm mod); static Eterm native_addresses(Process* p, Eterm mod); int patch_funentries(Eterm Patchlist); int patch(Eterm Addresses, Uint fe); @@ -648,6 +649,7 @@ erts_prepare_loading(Binary* magic, Process *c_p, Eterm group_leader, stp->code[MI_COMPILE_PTR] = 0; stp->code[MI_COMPILE_SIZE] = 0; stp->code[MI_COMPILE_SIZE_ON_HEAP] = 0; + stp->code[MI_MD5_PTR] = 0; /* * Read the atom table. @@ -2363,7 +2365,11 @@ load_code(LoaderState* stp) if (stp->may_load_nif) { const int finfo_ix = ci - FUNC_INFO_SZ; - enum { MIN_FUNC_SZ = 3 }; +#ifdef ERTS_DIRTY_SCHEDULERS + enum { MIN_FUNC_SZ = 4 }; +#else + enum { MIN_FUNC_SZ = 3 }; +#endif if (finfo_ix - last_func_start < MIN_FUNC_SZ && last_func_start) { /* Must make room for call_nif op */ int pad = MIN_FUNC_SZ - (finfo_ix - last_func_start); @@ -3315,9 +3321,10 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, { GenOp* op; + GenOpArg *tmp; int arity = Size.val + 3; int size = Size.val / 2; - int i; + int i, j, align = 0; /* * Verify the validity of the list. @@ -3332,9 +3339,37 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, } /* + * Use a special-cased instruction if there are only two values. + */ + if (size == 2) { + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_tuple_arity2_6; + GENOP_ARITY(op, arity - 1); + op->a[0] = S; + op->a[1] = Fail; + op->a[2].type = TAG_u; + op->a[2].val = Rest[0].val; + op->a[3].type = TAG_u; + op->a[3].val = Rest[2].val; + op->a[4] = Rest[1]; + op->a[5] = Rest[3]; + + return op; + } + + /* * Generate the generic instruction. + * Assumption: + * Few different tuple arities to select on (fewer than 20). + * Use linear scan approach. */ + align = 1; + + arity += 2*align; + size += align; + NEW_GENOP(stp, op); op->next = NULL; op->op = genop_i_select_tuple_arity_3; @@ -3342,39 +3377,36 @@ gen_select_tuple_arity(LoaderState* stp, GenOpArg S, GenOpArg Fail, op->a[0] = S; op->a[1] = Fail; op->a[2].type = TAG_u; - op->a[2].val = Size.val / 2; - for (i = 0; i < Size.val; i += 2) { - op->a[i+3].type = TAG_v; - op->a[i+3].val = make_arityval(Rest[i].val); - op->a[i+4] = Rest[i+1]; - } + op->a[2].val = size; - /* - * Sort the values to make them useful for a binary search. - */ + tmp = (GenOpArg *) erts_alloc(ERTS_ALC_T_LOADER_TMP, sizeof(GenOpArg)*(arity-2*align)); - qsort(op->a+3, size, 2*sizeof(GenOpArg), - (int (*)(const void *, const void *)) genopargcompare); -#ifdef DEBUG - for (i = 3; i < arity-2; i += 2) { - ASSERT(op->a[i].val < op->a[i+2].val); + for (i = 3; i < arity - 2*align; i+=2) { + tmp[i-3].type = TAG_v; + tmp[i-3].val = make_arityval(Rest[i-3].val); + tmp[i-2] = Rest[i-2]; } -#endif /* - * Use a special-cased instruction if there are only two values. + * Sort the values to make them useful for a sentinel search */ - if (size == 2) { - op->op = genop_i_select_tuple_arity2_6; - op->arity--; - op->a[2].type = TAG_u; - op->a[2].val = arityval(op->a[3].val); - op->a[3] = op->a[4]; - op->a[4].type = TAG_u; - op->a[4].val = arityval(op->a[5].val); - op->a[5] = op->a[6]; + + qsort(tmp, size - align, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); + + j = 3; + for (i = 3; i < arity - 2*align; i += 2) { + op->a[j] = tmp[i-3]; + op->a[j + size] = tmp[i-2]; + j++; } + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) tmp); + + op->a[j].type = TAG_u; + op->a[j].val = ~((BeamInstr)0); + op->a[j+size] = Fail; + return op; } @@ -3596,45 +3628,109 @@ gen_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail, GenOpArg Size, GenOpArg* Rest) { GenOp* op; + GenOpArg *tmp; int arity = Size.val + 3; int size = Size.val / 2; - int i; + int i, j, align = 0; + + if (size == 2) { + + /* + * Use a special-cased instruction if there are only two values. + */ + + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_val2_6; + GENOP_ARITY(op, arity - 1); + op->a[0] = S; + op->a[1] = Fail; + op->a[2] = Rest[0]; + op->a[3] = Rest[2]; + op->a[4] = Rest[1]; + op->a[5] = Rest[3]; + + return op; + + } else if (size > 10) { + + /* binary search instruction */ + + NEW_GENOP(stp, op); + op->next = NULL; + op->op = genop_i_select_val_bins_3; + GENOP_ARITY(op, arity); + op->a[0] = S; + op->a[1] = Fail; + op->a[2].type = TAG_u; + op->a[2].val = size; + for (i = 3; i < arity; i++) { + op->a[i] = Rest[i-3]; + } + + /* + * Sort the values to make them useful for a binary search. + */ + + qsort(op->a+3, size, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); +#ifdef DEBUG + for (i = 3; i < arity-2; i += 2) { + ASSERT(op->a[i].val < op->a[i+2].val); + } +#endif + return op; + } + + /* linear search instruction */ + + align = 1; + + arity += 2*align; + size += align; NEW_GENOP(stp, op); op->next = NULL; - op->op = genop_i_select_val_3; + op->op = genop_i_select_val_lins_3; GENOP_ARITY(op, arity); op->a[0] = S; op->a[1] = Fail; op->a[2].type = TAG_u; op->a[2].val = size; - for (i = 3; i < arity; i++) { - op->a[i] = Rest[i-3]; + + tmp = (GenOpArg *) erts_alloc(ERTS_ALC_T_LOADER_TMP, sizeof(GenOpArg)*(arity-2*align)); + + for (i = 3; i < arity - 2*align; i++) { + tmp[i-3] = Rest[i-3]; } /* - * Sort the values to make them useful for a binary search. + * Sort the values to make them useful for a sentinel search */ - qsort(op->a+3, size, 2*sizeof(GenOpArg), - (int (*)(const void *, const void *)) genopargcompare); -#ifdef DEBUG - for (i = 3; i < arity-2; i += 2) { - ASSERT(op->a[i].val < op->a[i+2].val); + qsort(tmp, size - align, 2*sizeof(GenOpArg), + (int (*)(const void *, const void *)) genopargcompare); + + j = 3; + for (i = 3; i < arity - 2*align; i += 2) { + op->a[j] = tmp[i-3]; + op->a[j+size] = tmp[i-2]; + j++; } -#endif - /* - * Use a special-cased instruction if there are only two values. - */ - if (size == 2) { - op->op = genop_i_select_val2_6; - op->arity--; - op->a[2] = op->a[3]; - op->a[3] = op->a[4]; - op->a[4] = op->a[5]; - op->a[5] = op->a[6]; + erts_free(ERTS_ALC_T_LOADER_TMP, (void *) tmp); + + /* add sentinel */ + + op->a[j].type = TAG_u; + op->a[j].val = ~((BeamInstr)0); + op->a[j+size] = Fail; + +#ifdef DEBUG + for (i = 0; i < size - 1; i++) { + ASSERT(op->a[i+3].val <= op->a[i+4].val); } +#endif return op; } @@ -4038,7 +4134,7 @@ freeze_code(LoaderState* stp) } size = (stp->ci * sizeof(BeamInstr)) + (stp->total_literal_size * sizeof(Eterm)) + - strtab_size + attr_size + compile_size + line_size; + strtab_size + attr_size + compile_size + MD5_SIZE + line_size; /* * Move the code to its final location. @@ -4247,11 +4343,20 @@ freeze_code(LoaderState* stp) code[MI_COMPILE_SIZE_ON_HEAP] = decoded_size; } CHKBLK(ERTS_ALC_T_CODE,code); + { + byte* md5_sum = str_table + strtab_size + attr_size + compile_size; + CHKBLK(ERTS_ALC_T_CODE,code); + sys_memcpy(md5_sum, stp->mod_md5, MD5_SIZE); + CHKBLK(ERTS_ALC_T_CODE,code); + code[MI_MD5_PTR] = (BeamInstr) md5_sum; + CHKBLK(ERTS_ALC_T_CODE,code); + } + CHKBLK(ERTS_ALC_T_CODE,code); /* * Make sure that we have not overflowed the allocated code space. */ - ASSERT(str_table + strtab_size + attr_size + compile_size == + ASSERT(str_table + strtab_size + attr_size + compile_size + MD5_SIZE == ((byte *) code) + size); /* @@ -4967,7 +5072,8 @@ get_tag_and_value(LoaderState* stp, Uint len_code, arity = count/sizeof(Eterm); *result = new_literal(stp, &hp, arity+1); - (void) bytes_to_big(bigbuf, count, neg, hp); + if (is_nil(bytes_to_big(bigbuf, count, neg, hp))) + goto load_error; if (bigbuf != default_buf) { erts_free(ERTS_ALC_T_LOADER_TMP, (void *) bigbuf); @@ -5103,10 +5209,11 @@ erts_module_info_0(Process* p, Eterm module) hp += 3; \ list = CONS(hp, tup, list) + BUILD_INFO(am_md5); BUILD_INFO(am_compile); BUILD_INFO(am_attributes); - BUILD_INFO(am_imports); BUILD_INFO(am_exports); + BUILD_INFO(am_module); #undef BUILD_INFO return list; } @@ -5116,8 +5223,8 @@ erts_module_info_1(Process* p, Eterm module, Eterm what) { if (what == am_module) { return module; - } else if (what == am_imports) { - return NIL; + } else if (what == am_md5) { + return md5_of_module(p, module); } else if (what == am_exports) { return exported_from_module(p, module); } else if (what == am_functions) { @@ -5306,7 +5413,7 @@ attributes_for_module(Process* p, /* Process whose heap to use. */ Eterm result = NIL; Eterm* end; - if (is_not_atom(mod) || (is_not_list(result) && is_not_nil(result))) { + if (is_not_atom(mod)) { return THE_NON_VALUE; } @@ -5345,7 +5452,7 @@ compilation_info_for_module(Process* p, /* Process whose heap to use. */ Eterm result = NIL; Eterm* end; - if (is_not_atom(mod) || (is_not_list(result) && is_not_nil(result))) { + if (is_not_atom(mod)) { return THE_NON_VALUE; } @@ -5368,6 +5475,33 @@ compilation_info_for_module(Process* p, /* Process whose heap to use. */ } /* + * Returns the MD5 checksum for a module + * + * Returns a tagged term, or 0 on error. + */ + +Eterm +md5_of_module(Process* p, /* Process whose heap to use. */ + Eterm mod) /* Tagged atom for module. */ +{ + Module* modp; + BeamInstr* code; + Eterm res = NIL; + + if (is_not_atom(mod)) { + return THE_NON_VALUE; + } + + modp = erts_get_module(mod, erts_active_code_ix()); + if (modp == NULL) { + return THE_NON_VALUE; + } + code = modp->curr.code; + res = new_binary(p, (byte *) code[MI_MD5_PTR], MD5_SIZE); + return res; +} + +/* * Build a single {M,F,A,Loction} item to be part of * a stack trace. */ @@ -5543,7 +5677,7 @@ code_module_md5_1(BIF_ALIST_1) res = am_undefined; goto done; } - res = new_binary(p, stp->mod_md5, sizeof(stp->mod_md5)); + res = new_binary(p, stp->mod_md5, MD5_SIZE); done: erts_free_aligned_binary_bytes(temp_alloc); @@ -5939,6 +6073,7 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info) code[MI_LITERALS_END] = 0; code[MI_LITERALS_OFF_HEAP] = 0; code[MI_ON_LOAD_FUNCTION_PTR] = 0; + code[MI_MD5_PTR] = 0; ci = MI_FUNCTIONS + n + 1; /* diff --git a/erts/emulator/beam/beam_load.h b/erts/emulator/beam/beam_load.h index bd22b0c4de..0e3ca0bdb0 100644 --- a/erts/emulator/beam/beam_load.h +++ b/erts/emulator/beam/beam_load.h @@ -91,7 +91,6 @@ extern Uint erts_total_code_size; #define MI_LITERALS_END 8 #define MI_LITERALS_OFF_HEAP 9 - /* * Pointer to the on_load function (or NULL if none). */ @@ -103,6 +102,11 @@ extern Uint erts_total_code_size; #define MI_LINE_TABLE 11 /* + * Pointer to the module MD5 sum (16 bytes) + */ +#define MI_MD5_PTR 12 + +/* * Start of function pointer table. This table contains pointers to * all functions in the module plus an additional pointer just beyond * the end of the last function. @@ -111,7 +115,7 @@ extern Uint erts_total_code_size; * this table. */ -#define MI_FUNCTIONS 12 +#define MI_FUNCTIONS 13 /* * Layout of the line table. diff --git a/erts/emulator/beam/beam_ranges.c b/erts/emulator/beam/beam_ranges.c index 0f2d5d0c2a..cb6470638f 100644 --- a/erts/emulator/beam/beam_ranges.c +++ b/erts/emulator/beam/beam_ranges.c @@ -282,7 +282,7 @@ find_range(BeamInstr* pc) while (low < high) { if (pc < mid->start) { high = mid; - } else if (pc > RANGE_END(mid)) { + } else if (pc >= RANGE_END(mid)) { low = mid + 1; } else { erts_smp_atomic_set_nob(&r[active].mid, (erts_aint_t) mid); diff --git a/erts/emulator/beam/benchmark.c b/erts/emulator/beam/benchmark.c index 8613131176..b16fe6b271 100644 --- a/erts/emulator/beam/benchmark.c +++ b/erts/emulator/beam/benchmark.c @@ -37,37 +37,9 @@ unsigned long long major_gc; #ifdef BM_TIMERS -#if (defined(__i386__) || defined(__x86_64__)) && USE_PERFCTR - -#include "libperfctr.h" -struct vperfctr *system_clock; -double cpu_khz; -BM_NEW_TIMER(start); - -static double get_hrvtime(void) -{ - unsigned long long ticks; - double milli_seconds; - - ticks = vperfctr_read_tsc(system_clock); - milli_seconds = (double)ticks / cpu_khz; - return milli_seconds; -} - -static void stop_hrvtime(void) -{ - if(system_clock) - { - vperfctr_stop(system_clock); - vperfctr_close(system_clock); - system_clock = NULL; - } -} - -#else /* not perfctr, asuming Solaris */ +/* assuming Solaris */ #include <time.h> BM_TIMER_T system_clock; -#endif unsigned long local_pause_times[MAX_PAUSE_TIME]; unsigned long pause_times[MAX_PAUSE_TIME]; @@ -117,40 +89,6 @@ unsigned long long message_sizes[1000]; void init_benchmarking() { #ifdef BM_TIMERS -#if (defined(__i386__) || defined(__x86_64__)) && USE_PERFCTR - /* pass `--with-perfctr=/path/to/perfctr' when configuring */ - struct perfctr_info info; - struct vperfctr_control control; - int i; - - system_clock = vperfctr_open(); - if (system_clock != NULL) - { - if (vperfctr_info(system_clock,&info) >= 0) - { - cpu_khz = (double)info.cpu_khz; - if (info.cpu_features & PERFCTR_FEATURE_RDTSC) - { - memset(&control,0,sizeof control); - control.cpu_control.tsc_on = 1; - } - } - if (vperfctr_control(system_clock,&control) < 0) - { - vperfctr_close(system_clock); - system_clock = NULL; - } - } - - for (i = 0; i < 1000; i++) - { - BM_START_TIMER(system); - BM_STOP_TIMER(system); - } - - timer_time = system_time / 1000; - start_time = 0; -#else int i; for (i = 0; i < 1000; i++) { @@ -158,7 +96,6 @@ void init_benchmarking() BM_STOP_TIMER(system); } timer_time = system_time / 1000; -#endif for (i = 0; i < MAX_PAUSE_TIME; i++) { local_pause_times[i] = 0; diff --git a/erts/emulator/beam/benchmark.h b/erts/emulator/beam/benchmark.h index 766edaac42..7fc3933f3d 100644 --- a/erts/emulator/beam/benchmark.h +++ b/erts/emulator/beam/benchmark.h @@ -37,10 +37,7 @@ /* BM_TIMERS keeps track of the time spent in diferent parts of the * system. It only measures accual active time, not time spent in idle - * mode. These timers requires hardware support. For Linux, use the - * package perfctr from user.it.uu.se/~mikpe/linux/perfctr. If this - * package is not specified when configuring the system - * (--with-perfctr=PATH), the Solaris hrtime_t will be used. + * mode. Currently, the Solaris hrtime_t will be used. * To add new timers look below. */ #define BM_TIMERS @@ -142,43 +139,12 @@ extern unsigned long long major_gc; * meassure (send time in shared heap for instance). */ -#if (defined(__i386__) || defined(__x86_64__)) && USE_PERFCTR -#include "libperfctr.h" +/* (Assuming Solaris) */ -#define BM_TIMER_T double - -extern struct vperfctr *system_clock; -extern double cpu_khz; -extern BM_TIMER_T start_time; - -#define BM_START_TIMER(t) start_time = \ - (BM_TIMER_T)vperfctr_read_tsc(system_clock) / \ - cpu_khz; - -#define BM_STOP_TIMER(t) do { \ - BM_TIMER_T tmp = ((BM_TIMER_T)vperfctr_read_tsc(system_clock) / cpu_khz); \ - tmp -= (start_time + timer_time); \ - t##_time += (tmp > 0 ? tmp : 0); \ -} while(0) - -#define BM_TIME_PRINTER(str,time) do { \ - int min,sec,milli,micro; \ - BM_TIMER_T tmp = (time) * 1000; \ - micro = (uint)(tmp - ((int)(tmp / 1000)) * 1000); \ - tmp /= 1000; \ - milli = (uint)(tmp - ((int)(tmp / 1000)) * 1000); \ - tmp /= 1000; \ - sec = (uint)(tmp - ((int)(tmp / 60)) * 60); \ - min = (uint)tmp / 60; \ - erts_fprintf(file,str": %d:%02d.%03d %03d\n",min,sec,milli,micro); \ -} while(0) - -#else /* !USE_PERFCTR (Assuming Solaris) */ - -#define BM_TIMER_T hrtime_t -#define BM_START_TIMER(t) system_clock = sys_gethrtime() +#define BM_TIMER_T ErtsMonotonicTime +#define BM_START_TIMER(t) system_clock = ERTS_MONOTONIC_TO_NSEC(erts_os_monotonic_time()) #define BM_STOP_TIMER(t) do { \ - BM_TIMER_T tmp = (sys_gethrtime() - system_clock) - timer_time; \ + BM_TIMER_T tmp = (ERTS_MONOTONIC_TO_NSEC(erts_os_monotonic_time()) - system_clock) - timer_time; \ t##_time += (tmp > 0 ? tmp : 0); \ } while(0) @@ -196,7 +162,6 @@ extern BM_TIMER_T start_time; } while(0) extern BM_TIMER_T system_clock; -#endif /* USE_PERFCTR */ extern BM_TIMER_T timer_time; extern BM_TIMER_T system_time; diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c index 06a1230ca0..cc20ec7440 100644 --- a/erts/emulator/beam/bif.c +++ b/erts/emulator/beam/bif.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2013. All Rights Reserved. + * Copyright Ericsson AB 1996-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -28,7 +28,9 @@ #include "global.h" #include "erl_process.h" #include "error.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "dist.h" #include "erl_version.h" @@ -40,16 +42,22 @@ #define ERTS_PTAB_WANT_BIF_IMPL__ #include "erl_ptab.h" #include "erl_bits.h" +#include "erl_bif_unique.h" -static Export* flush_monitor_message_trap = NULL; +static Export* flush_monitor_messages_trap = NULL; static Export* set_cpu_topology_trap = NULL; static Export* await_proc_exit_trap = NULL; static Export* await_port_send_result_trap = NULL; Export* erts_format_cpu_topology_trap = NULL; +static Export dsend_continue_trap_export; +Export *erts_convert_time_unit_trap = NULL; static Export *await_sched_wall_time_mod_trap; static erts_smp_atomic32_t sched_wall_time; +static erts_smp_mtx_t ports_snapshot_mtx; +erts_smp_atomic_t erts_dead_ports_ptr; /* To store dying ports during snapshot */ + #define DECL_AM(S) Eterm AM_ ## S = am_atom_put(#S, sizeof(#S) - 1) /* @@ -391,7 +399,7 @@ remote_demonitor(Process *c_p, DistEntry *dep, Eterm ref, Eterm to) return res; } -static int demonitor(Process *c_p, Eterm ref) +static int demonitor(Process *c_p, Eterm ref, Eterm *multip) { ErtsMonitor *mon = NULL; /* The monitor entry to delete */ Process *rp; /* Local target process */ @@ -415,65 +423,73 @@ static int demonitor(Process *c_p, Eterm ref) goto done; } - if (mon->type != MON_ORIGIN) { - res = ERTS_DEMONITOR_BADARG; - goto done; - } - to = mon->pid; - - if (is_atom(to)) { - /* Monitoring a name at node to */ - ASSERT(is_node_name_atom(to)); - dep = erts_sysname_to_connected_dist_entry(to); - ASSERT(dep != erts_this_dist_entry); - if (dep) - deref_de = 1; - } else { - ASSERT(is_pid(to)); - dep = pid_dist_entry(to); - } - if (dep != erts_this_dist_entry) { - res = remote_demonitor(c_p, dep, ref, to); - /* remote_demonitor() unlocks link lock on c_p */ - unlock_link = 0; - } - else { /* Local monitor */ - if (deref_de) { - deref_de = 0; - erts_deref_dist_entry(dep); + switch (mon->type) { + case MON_TIME_OFFSET: + *multip = am_true; + erts_demonitor_time_offset(ref); + res = ERTS_DEMONITOR_TRUE; + break; + case MON_ORIGIN: + to = mon->pid; + *multip = am_false; + if (is_atom(to)) { + /* Monitoring a name at node to */ + ASSERT(is_node_name_atom(to)); + dep = erts_sysname_to_connected_dist_entry(to); + ASSERT(dep != erts_this_dist_entry); + if (dep) + deref_de = 1; + } else { + ASSERT(is_pid(to)); + dep = pid_dist_entry(to); } - dep = NULL; - rp = erts_pid2proc_opt(c_p, - ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_LINK, - to, - ERTS_PROC_LOCK_LINK, - ERTS_P2P_FLG_ALLOW_OTHER_X); - mon = erts_remove_monitor(&ERTS_P_MONITORS(c_p), ref); + if (dep != erts_this_dist_entry) { + res = remote_demonitor(c_p, dep, ref, to); + /* remote_demonitor() unlocks link lock on c_p */ + unlock_link = 0; + } + else { /* Local monitor */ + if (deref_de) { + deref_de = 0; + erts_deref_dist_entry(dep); + } + dep = NULL; + rp = erts_pid2proc_opt(c_p, + ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_LINK, + to, + ERTS_PROC_LOCK_LINK, + ERTS_P2P_FLG_ALLOW_OTHER_X); + mon = erts_remove_monitor(&ERTS_P_MONITORS(c_p), ref); #ifndef ERTS_SMP - ASSERT(mon); + ASSERT(mon); #else - if (!mon) - res = ERTS_DEMONITOR_FALSE; - else + if (!mon) + res = ERTS_DEMONITOR_FALSE; + else #endif - { - res = ERTS_DEMONITOR_TRUE; - erts_destroy_monitor(mon); - } - if (rp) { - ErtsMonitor *rmon; - rmon = erts_remove_monitor(&ERTS_P_MONITORS(rp), ref); - if (rp != c_p) - erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_LINK); - if (rmon != NULL) - erts_destroy_monitor(rmon); - } - else { - ERTS_SMP_ASSERT_IS_NOT_EXITING(c_p); - } + { + res = ERTS_DEMONITOR_TRUE; + erts_destroy_monitor(mon); + } + if (rp) { + ErtsMonitor *rmon; + rmon = erts_remove_monitor(&ERTS_P_MONITORS(rp), ref); + if (rp != c_p) + erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_LINK); + if (rmon != NULL) + erts_destroy_monitor(rmon); + } + else { + ERTS_SMP_ASSERT_IS_NOT_EXITING(c_p); + } + } + break; + default: + res = ERTS_DEMONITOR_BADARG; + *multip = am_false; + break; } - done: if (unlock_link) @@ -490,7 +506,8 @@ static int demonitor(Process *c_p, Eterm ref) BIF_RETTYPE demonitor_1(BIF_ALIST_1) { - switch (demonitor(BIF_P, BIF_ARG_1)) { + Eterm multi; + switch (demonitor(BIF_P, BIF_ARG_1, &multi)) { case ERTS_DEMONITOR_FALSE: case ERTS_DEMONITOR_TRUE: BIF_RET(am_true); @@ -508,6 +525,7 @@ BIF_RETTYPE demonitor_1(BIF_ALIST_1) BIF_RETTYPE demonitor_2(BIF_ALIST_2) { Eterm res = am_true; + Eterm multi = am_false; int info = 0; int flush = 0; Eterm list = BIF_ARG_2; @@ -530,13 +548,18 @@ BIF_RETTYPE demonitor_2(BIF_ALIST_2) if (is_not_nil(list)) goto badarg; - switch (demonitor(BIF_P, BIF_ARG_1)) { + switch (demonitor(BIF_P, BIF_ARG_1, &multi)) { case ERTS_DEMONITOR_FALSE: if (info) res = am_false; - if (flush) - BIF_TRAP2(flush_monitor_message_trap, BIF_P, BIF_ARG_1, res); + if (flush) { + flush_messages: + BIF_TRAP3(flush_monitor_messages_trap, BIF_P, + BIF_ARG_1, multi, res); + } case ERTS_DEMONITOR_TRUE: + if (multi == am_true && flush) + goto flush_messages; BIF_RET(res); case ERTS_DEMONITOR_YIELD_TRUE: ERTS_BIF_YIELD_RETURN(BIF_P, am_true); @@ -595,14 +618,12 @@ erts_queue_monitor_message(Process *p, } static BIF_RETTYPE -local_pid_monitor(Process *p, Eterm target) +local_pid_monitor(Process *p, Eterm target, Eterm mon_ref, int bool) { BIF_RETTYPE ret; - Eterm mon_ref; Process *rp; ErtsProcLocks p_locks = ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_LINK; - mon_ref = erts_make_ref(p); ERTS_BIF_PREP_RET(ret, mon_ref); if (target == p->common.id) { return ret; @@ -615,12 +636,18 @@ local_pid_monitor(Process *p, Eterm target) if (!rp) { erts_smp_proc_unlock(p, ERTS_PROC_LOCK_LINK); p_locks &= ~ERTS_PROC_LOCK_LINK; - erts_queue_monitor_message(p, &p_locks, - mon_ref, am_process, target, am_noproc); + if (bool) + ret = am_false; + else + erts_queue_monitor_message(p, &p_locks, + mon_ref, am_process, target, am_noproc); } else { ASSERT(rp != p); + if (bool) + ret = am_true; + erts_add_monitor(&ERTS_P_MONITORS(p), MON_ORIGIN, mon_ref, target, NIL); erts_add_monitor(&ERTS_P_MONITORS(rp), MON_TARGET, mon_ref, p->common.id, NIL); @@ -744,13 +771,28 @@ BIF_RETTYPE monitor_2(BIF_ALIST_2) int deref_de = 0; /* Only process monitors are implemented */ - if (BIF_ARG_1 != am_process) { + switch (BIF_ARG_1) { + case am_time_offset: { + Eterm ref; + if (BIF_ARG_2 != am_clock_service) + goto error; + ref = erts_make_ref(BIF_P); + erts_smp_proc_lock(BIF_P, ERTS_PROC_LOCK_LINK); + erts_add_monitor(&ERTS_P_MONITORS(BIF_P), MON_TIME_OFFSET, + ref, am_clock_service, NIL); + erts_smp_proc_unlock(BIF_P, ERTS_PROC_LOCK_LINK); + erts_monitor_time_offset(BIF_P->common.id, ref); + BIF_RET(ref); + } + case am_process: + break; + default: goto error; } if (is_internal_pid(target)) { local_pid: - ret = local_pid_monitor(BIF_P, target); + ret = local_pid_monitor(BIF_P, target, erts_make_ref(BIF_P), 0); } else if (is_external_pid(target)) { dep = external_pid_dist_entry(target); if (dep == erts_this_dist_entry) @@ -793,6 +835,25 @@ BIF_RETTYPE monitor_2(BIF_ALIST_2) return ret; } +BIF_RETTYPE erts_internal_monitor_process_2(BIF_ALIST_2) +{ + if (is_not_internal_pid(BIF_ARG_1)) { + if (is_external_pid(BIF_ARG_1) + && (external_pid_dist_entry(BIF_ARG_1) + == erts_this_dist_entry)) { + BIF_RET(am_false); + } + goto badarg; + } + + if (is_not_internal_ref(BIF_ARG_2)) + goto badarg; + + BIF_RET(local_pid_monitor(BIF_P, BIF_ARG_1, BIF_ARG_2, 1)); + +badarg: + BIF_ERROR(BIF_P, BADARG); +} /**********************************************************************/ /* this is a combination of the spawn and link BIFs */ @@ -1777,6 +1838,8 @@ BIF_RETTYPE whereis_1(BIF_ALIST_1) * erlang:'!'/2 */ +HIPE_WRAPPER_BIF_DISABLE_GC(ebif_bang, 2) + BIF_RETTYPE ebif_bang_2(BIF_ALIST_2) { @@ -1795,34 +1858,36 @@ ebif_bang_2(BIF_ALIST_2) #define SEND_USER_ERROR (-5) #define SEND_INTERNAL_ERROR (-6) #define SEND_AWAIT_RESULT (-7) +#define SEND_YIELD_CONTINUE (-8) + -Sint do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp); +Sint do_send(Process *p, Eterm to, Eterm msg, Eterm *refp, ErtsSendContext*); static Sint remote_send(Process *p, DistEntry *dep, - Eterm to, Eterm full_to, Eterm msg, int suspend) + Eterm to, Eterm full_to, Eterm msg, + ErtsSendContext* ctx) { Sint res; int code; - ErtsDSigData dsd; ASSERT(is_atom(to) || is_external_pid(to)); - code = erts_dsig_prepare(&dsd, dep, p, ERTS_DSP_NO_LOCK, !suspend); + code = erts_dsig_prepare(&ctx->dsd, dep, p, ERTS_DSP_NO_LOCK, !ctx->suspend); switch (code) { case ERTS_DSIG_PREP_NOT_ALIVE: case ERTS_DSIG_PREP_NOT_CONNECTED: res = SEND_TRAP; break; case ERTS_DSIG_PREP_WOULD_SUSPEND: - ASSERT(!suspend); + ASSERT(!ctx->suspend); res = SEND_YIELD; break; case ERTS_DSIG_PREP_CONNECTED: { if (is_atom(to)) - code = erts_dsig_send_reg_msg(&dsd, to, msg); + code = erts_dsig_send_reg_msg(to, msg, ctx); else - code = erts_dsig_send_msg(&dsd, to, msg); + code = erts_dsig_send_msg(to, msg, ctx); /* * Note that reductions have been bumped on calling * process by erts_dsig_send_reg_msg() or @@ -1830,6 +1895,8 @@ static Sint remote_send(Process *p, DistEntry *dep, */ if (code == ERTS_DSIG_SEND_YIELD) res = SEND_YIELD_RETURN; + else if (code == ERTS_DSIG_SEND_CONTINUE) + res = SEND_YIELD_CONTINUE; else res = 0; break; @@ -1850,7 +1917,8 @@ static Sint remote_send(Process *p, DistEntry *dep, } Sint -do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { +do_send(Process *p, Eterm to, Eterm msg, Eterm *refp, ErtsSendContext* ctx) +{ Eterm portid; Port *pt; Process* rp; @@ -1869,6 +1937,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { } else if (is_external_pid(to)) { dep = external_pid_dist_entry(to); if(dep == erts_this_dist_entry) { +#if DEBUG erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); erts_dsprintf(dsbufp, "Discarding message %T from %T to %T in an old " @@ -1879,15 +1948,21 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { external_pid_creation(to), erts_this_node->creation); erts_send_error_to_logger(p->group_leader, dsbufp); +#endif return 0; } - return remote_send(p, dep, to, to, msg, suspend); + return remote_send(p, dep, to, to, msg, ctx); } else if (is_atom(to)) { Eterm id = erts_whereis_name_to_id(p, to); rp = erts_proc_lookup(id); - if (rp) + if (rp) { + if (IS_TRACED(p)) + trace_send(p, to, msg); + if (ERTS_PROC_GET_SAVED_CALLS_BUF(p)) + save_calls(p, &exp_send); goto send_message; + } pt = erts_port_lookup(id, (erts_port_synchronous_ops @@ -1907,6 +1982,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { } else if (is_external_port(to) && (external_port_dist_entry(to) == erts_this_dist_entry)) { +#if DEBUG erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); erts_dsprintf(dsbufp, "Discarding message %T from %T to %T in an old " @@ -1917,6 +1993,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { external_port_creation(to), erts_this_node->creation); erts_send_error_to_logger(p->group_leader, dsbufp); +#endif return 0; } else if (is_internal_port(to)) { int ret_val; @@ -1931,7 +2008,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { ret_val = 0; if (pt) { - int ps_flags = suspend ? 0 : ERTS_PORT_SIG_FLG_NOSUSPEND; + int ps_flags = ctx->suspend ? 0 : ERTS_PORT_SIG_FLG_NOSUSPEND; *refp = NIL; switch (erts_port_command(p, ps_flags, pt, msg, refp)) { @@ -1940,12 +2017,12 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { return SEND_USER_ERROR; case ERTS_PORT_OP_BUSY: /* Nothing has been sent */ - if (suspend) + if (ctx->suspend) erts_suspend(p, ERTS_PROC_LOCK_MAIN, pt); return SEND_YIELD; case ERTS_PORT_OP_BUSY_SCHEDULED: /* Message was sent */ - if (suspend) { + if (ctx->suspend) { erts_suspend(p, ERTS_PROC_LOCK_MAIN, pt); ret_val = SEND_YIELD_RETURN; break; @@ -2025,9 +2102,14 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { return 0; } - ret = remote_send(p, dep, tp[1], to, msg, suspend); - if (dep) - erts_deref_dist_entry(dep); + ret = remote_send(p, dep, tp[1], to, msg, ctx); + if (ret != SEND_YIELD_CONTINUE) { + if (dep) { + erts_deref_dist_entry(dep); + } + } else { + ctx->dep_to_deref = dep; + } return ret; } else { if (IS_TRACED(p)) /* XXX Is this really neccessary ??? */ @@ -2058,9 +2140,11 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { } } +HIPE_WRAPPER_BIF_DISABLE_GC(send, 3) BIF_RETTYPE send_3(BIF_ALIST_3) { + BIF_RETTYPE retval; Eterm ref; Process *p = BIF_P; Eterm to = BIF_ARG_1; @@ -2068,34 +2152,44 @@ BIF_RETTYPE send_3(BIF_ALIST_3) Eterm opts = BIF_ARG_3; int connect = !0; - int suspend = !0; Eterm l = opts; Sint result; - + DeclareTypedTmpHeap(ErtsSendContext, ctx, BIF_P); + UseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), BIF_P); + + ctx->suspend = !0; + ctx->dep_to_deref = NULL; + ctx->return_term = am_ok; + ctx->dss.reds = (Sint) (ERTS_BIF_REDS_LEFT(p) * TERM_TO_BINARY_LOOP_FACTOR); + ctx->dss.phase = ERTS_DSIG_SEND_PHASE_INIT; + while (is_list(l)) { if (CAR(list_val(l)) == am_noconnect) { connect = 0; } else if (CAR(list_val(l)) == am_nosuspend) { - suspend = 0; + ctx->suspend = 0; } else { - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); + goto done; } l = CDR(list_val(l)); } if(!is_nil(l)) { - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); + goto done; } #ifdef DEBUG ref = NIL; #endif - result = do_send(p, to, msg, suspend, &ref); + result = do_send(p, to, msg, &ref, ctx); if (result > 0) { ERTS_VBUMP_REDS(p, result); if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(am_ok); + ERTS_BIF_PREP_RET(retval, am_ok); + goto done; } switch (result) { @@ -2103,68 +2197,127 @@ BIF_RETTYPE send_3(BIF_ALIST_3) /* May need to yield even though we do not bump reds here... */ if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(am_ok); + ERTS_BIF_PREP_RET(retval, am_ok); break; case SEND_TRAP: if (connect) { - BIF_TRAP3(dsend3_trap, p, to, msg, opts); + ERTS_BIF_PREP_TRAP3(retval, dsend3_trap, p, to, msg, opts); } else { - BIF_RET(am_noconnect); + ERTS_BIF_PREP_RET(retval, am_noconnect); } break; case SEND_YIELD: - if (suspend) { - ERTS_BIF_YIELD3(bif_export[BIF_send_3], p, to, msg, opts); + if (ctx->suspend) { + ERTS_BIF_PREP_YIELD3(retval, + bif_export[BIF_send_3], p, to, msg, opts); } else { - BIF_RET(am_nosuspend); + ERTS_BIF_PREP_RET(retval, am_nosuspend); } break; case SEND_YIELD_RETURN: - if (!suspend) - BIF_RET(am_nosuspend); + if (!ctx->suspend) { + ERTS_BIF_PREP_RET(retval, am_nosuspend); + break; + } yield_return: - ERTS_BIF_YIELD_RETURN(p, am_ok); + ERTS_BIF_PREP_YIELD_RETURN(retval, p, am_ok); + break; case SEND_AWAIT_RESULT: ASSERT(is_internal_ref(ref)); - BIF_TRAP3(await_port_send_result_trap, p, ref, am_nosuspend, am_ok); + ERTS_BIF_PREP_TRAP3(retval, await_port_send_result_trap, p, ref, am_nosuspend, am_ok); + break; case SEND_BADARG: - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); break; case SEND_USER_ERROR: - BIF_ERROR(p, EXC_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_ERROR); break; case SEND_INTERNAL_ERROR: - BIF_ERROR(p, EXC_INTERNAL_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_INTERNAL_ERROR); + break; + case SEND_YIELD_CONTINUE: + BUMP_ALL_REDS(p); + erts_set_gc_state(p, 0); + ERTS_BIF_PREP_TRAP1(retval, &dsend_continue_trap_export, p, + erts_dsend_export_trap_context(p, ctx)); break; default: - ASSERT(! "Illegal send result"); + erl_exit(ERTS_ABORT_EXIT, "send_3 invalid result %d\n", (int)result); break; } - ASSERT(! "Can not arrive here"); - BIF_ERROR(p, BADARG); + +done: + UnUseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), BIF_P); + return retval; } +HIPE_WRAPPER_BIF_DISABLE_GC(send, 2) + BIF_RETTYPE send_2(BIF_ALIST_2) { return erl_send(BIF_P, BIF_ARG_1, BIF_ARG_2); } +static BIF_RETTYPE dsend_continue_trap_1(BIF_ALIST_1) +{ + Binary* bin = ((ProcBin*) binary_val(BIF_ARG_1))->val; + ErtsSendContext* ctx = (ErtsSendContext*) ERTS_MAGIC_BIN_DATA(bin); + Sint initial_reds = (Sint) (ERTS_BIF_REDS_LEFT(BIF_P) * TERM_TO_BINARY_LOOP_FACTOR); + int result; + + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(bin) == erts_dsend_context_dtor); + + ctx->dss.reds = initial_reds; + result = erts_dsig_send(&ctx->dsd, &ctx->dss); + + switch (result) { + case ERTS_DSIG_SEND_OK: + erts_set_gc_state(BIF_P, 1); + BIF_RET(ctx->return_term); + break; + case ERTS_DSIG_SEND_YIELD: /*SEND_YIELD_RETURN*/ + erts_set_gc_state(BIF_P, 1); + if (!ctx->suspend) + BIF_RET(am_nosuspend); + ERTS_BIF_YIELD_RETURN(BIF_P, ctx->return_term); + + case ERTS_DSIG_SEND_CONTINUE: { /*SEND_YIELD_CONTINUE*/ + BUMP_ALL_REDS(BIF_P); + BIF_TRAP1(&dsend_continue_trap_export, BIF_P, BIF_ARG_1); + } + default: + erl_exit(ERTS_ABORT_EXIT, "dsend_continue_trap invalid result %d\n", (int)result); + break; + } + ASSERT(! "Can not arrive here"); + BIF_ERROR(BIF_P, BADARG); +} + Eterm erl_send(Process *p, Eterm to, Eterm msg) { + Eterm retval; Eterm ref; Sint result; + DeclareTypedTmpHeap(ErtsSendContext, ctx, p); + UseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), p); #ifdef DEBUG ref = NIL; #endif + ctx->suspend = !0; + ctx->dep_to_deref = NULL; + ctx->return_term = msg; + ctx->dss.reds = (Sint) (ERTS_BIF_REDS_LEFT(p) * TERM_TO_BINARY_LOOP_FACTOR); + ctx->dss.phase = ERTS_DSIG_SEND_PHASE_INIT; - result = do_send(p, to, msg, !0, &ref); + result = do_send(p, to, msg, &ref, ctx); if (result > 0) { ERTS_VBUMP_REDS(p, result); if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(msg); + ERTS_BIF_PREP_RET(retval, msg); + goto done; } switch (result) { @@ -2172,35 +2325,46 @@ Eterm erl_send(Process *p, Eterm to, Eterm msg) /* May need to yield even though we do not bump reds here... */ if (ERTS_IS_PROC_OUT_OF_REDS(p)) goto yield_return; - BIF_RET(msg); + ERTS_BIF_PREP_RET(retval, msg); break; case SEND_TRAP: - BIF_TRAP2(dsend2_trap, p, to, msg); + ERTS_BIF_PREP_TRAP2(retval, dsend2_trap, p, to, msg); break; case SEND_YIELD: - ERTS_BIF_YIELD2(bif_export[BIF_send_2], p, to, msg); + ERTS_BIF_PREP_YIELD2(retval, bif_export[BIF_send_2], p, to, msg); break; case SEND_YIELD_RETURN: yield_return: - ERTS_BIF_YIELD_RETURN(p, msg); + ERTS_BIF_PREP_YIELD_RETURN(retval, p, msg); + break; case SEND_AWAIT_RESULT: ASSERT(is_internal_ref(ref)); - BIF_TRAP3(await_port_send_result_trap, p, ref, msg, msg); + ERTS_BIF_PREP_TRAP3(retval, + await_port_send_result_trap, p, ref, msg, msg); + break; case SEND_BADARG: - BIF_ERROR(p, BADARG); + ERTS_BIF_PREP_ERROR(retval, p, BADARG); break; case SEND_USER_ERROR: - BIF_ERROR(p, EXC_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_ERROR); break; case SEND_INTERNAL_ERROR: - BIF_ERROR(p, EXC_INTERNAL_ERROR); + ERTS_BIF_PREP_ERROR(retval, p, EXC_INTERNAL_ERROR); + break; + case SEND_YIELD_CONTINUE: + BUMP_ALL_REDS(p); + erts_set_gc_state(p, 0); + ERTS_BIF_PREP_TRAP1(retval, &dsend_continue_trap_export, p, + erts_dsend_export_trap_context(p, ctx)); break; default: - ASSERT(! "Illegal send result"); + erl_exit(ERTS_ABORT_EXIT, "invalid send result %d\n", (int)result); break; } - ASSERT(! "Can not arrive here"); - BIF_ERROR(p, BADARG); + +done: + UnUseTmpHeap(sizeof(ErtsSendContext)/sizeof(Eterm), p); + return retval; } /**********************************************************************/ @@ -2763,6 +2927,7 @@ static int do_list_to_integer(Process *p, Eterm orig_list, Eterm *integer, Eterm *rest) { Sint i = 0; + Uint ui = 0; int skip = 0; int neg = 0; int n = 0; @@ -2816,8 +2981,8 @@ static int do_list_to_integer(Process *p, Eterm orig_list, unsigned_val(CAR(list_val(lst))) > '9') { break; } - i = i * 10; - i = i + unsigned_val(CAR(list_val(lst))) - '0'; + ui = ui * 10; + ui = ui + unsigned_val(CAR(list_val(lst))) - '0'; n++; lst = CDR(list_val(lst)); if (is_nil(lst)) { @@ -2841,7 +3006,8 @@ static int do_list_to_integer(Process *p, Eterm orig_list, */ if (n <= SMALL_DIGITS) { /* It must be small */ - if (neg) i = -i; + if (neg) i = -(Sint)ui; + else i = (Sint)ui; res = make_small(i); } else { lg2 = (n+1)*230/69+1; @@ -2882,9 +3048,6 @@ static int do_list_to_integer(Process *p, Eterm orig_list, res = big_plus_small(res, m, hp); } - if (is_big(res)) /* check if small */ - res = big_plus_small(res, 0, hp); /* includes conversion to small */ - if (neg) { if (is_small(res)) res = make_small(-signed_val(res)); @@ -2894,8 +3057,12 @@ static int do_list_to_integer(Process *p, Eterm orig_list, } } - if (is_big(res)) { - hp += (big_arity(res)+1); + if (is_not_small(res)) { + res = big_plus_small(res, 0, hp); /* includes conversion to small */ + + if (is_not_small(res)) { + hp += (big_arity(res)+1); + } } HRelease(p,hp_end,hp); } @@ -3434,91 +3601,6 @@ BIF_RETTYPE self_0(BIF_ALIST_0) /**********************************************************************/ -/* - New representation of refs in R9, see erl_term.h - - In the first data word, only the usual 18 bits are used. Ordinarily, - in "long refs" all words are used (in other words, practically never - wrap around), but for compatibility with older nodes, "short refs" - exist. Short refs come into being by being converted from the old - external format for refs (tag REFERENCE_EXT). Short refs are - converted back to the old external format. - - When converting a long ref to the external format in the case of - preparing for sending to an older node, the ref is truncated by only - using the first word (with 18 significant bits), and using the old tag - REFERENCE_EXT. - - When comparing refs or different size, only the parts up to the length - of the shorter operand are used. This has the desirable effect that a - long ref sent to an old node and back will be treated as equal to - the original, although some of the bits have been lost. - - The hash value for a ref always considers only the first word, since - in the above scenario, the original and the copy should have the same - hash value. -*/ - -static Uint32 reference0; /* Initialized in erts_init_bif */ -static Uint32 reference1; -static Uint32 reference2; -static erts_smp_spinlock_t make_ref_lock; -static erts_smp_mtx_t ports_snapshot_mtx; -erts_smp_atomic_t erts_dead_ports_ptr; /* To store dying ports during snapshot */ - -void -erts_make_ref_in_array(Uint32 ref[ERTS_MAX_REF_NUMBERS]) -{ - erts_smp_spin_lock(&make_ref_lock); - - reference0++; - if (reference0 >= MAX_REFERENCE) { - reference0 = 0; - reference1++; - if (reference1 == 0) { - reference2++; - } - } - - ref[0] = reference0; - ref[1] = reference1; - ref[2] = reference2; - - erts_smp_spin_unlock(&make_ref_lock); -} - -Eterm erts_make_ref_in_buffer(Eterm buffer[REF_THING_SIZE]) -{ - Eterm* hp = buffer; - Uint32 ref[ERTS_MAX_REF_NUMBERS]; - - erts_make_ref_in_array(ref); - write_ref_thing(hp, ref[0], ref[1], ref[2]); - return make_internal_ref(hp); -} - -Eterm erts_make_ref(Process *p) -{ - Eterm* hp; - Uint32 ref[ERTS_MAX_REF_NUMBERS]; - - ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_MAIN & erts_proc_lc_my_proc_locks(p)); - - hp = HAlloc(p, REF_THING_SIZE); - - erts_make_ref_in_array(ref); - write_ref_thing(hp, ref[0], ref[1], ref[2]); - - return make_internal_ref(hp); -} - -BIF_RETTYPE make_ref_0(BIF_ALIST_0) -{ - return erts_make_ref(BIF_P); -} - -/**********************************************************************/ - /* return the time of day */ BIF_RETTYPE time_0(BIF_ALIST_0) @@ -3984,16 +4066,19 @@ BIF_RETTYPE halt_2(BIF_ALIST_2) BIF_RETTYPE function_exported_3(BIF_ALIST_3) { + int arity; if (is_not_atom(BIF_ARG_1) || is_not_atom(BIF_ARG_2) || is_not_small(BIF_ARG_3)) { BIF_ERROR(BIF_P, BADARG); } - if (erts_find_function(BIF_ARG_1, BIF_ARG_2, signed_val(BIF_ARG_3), - erts_active_code_ix()) == NULL) { - BIF_RET(am_false); + arity = signed_val(BIF_ARG_3); + if (erts_find_function(BIF_ARG_1, BIF_ARG_2, arity, + erts_active_code_ix()) != NULL || + erts_is_builtin(BIF_ARG_1, BIF_ARG_2, arity)) { + BIF_RET(am_true); } - BIF_RET(am_true); + BIF_RET(am_false); } /**********************************************************************/ @@ -4496,6 +4581,28 @@ BIF_RETTYPE system_flag_2(BIF_ALIST_2) break; } #endif + } else if (BIF_ARG_1 == am_time_offset + && ERTS_IS_ATOM_STR("finalize", BIF_ARG_2)) { + ErtsTimeOffsetState res; + erts_smp_proc_unlock(BIF_P, ERTS_PROC_LOCK_MAIN); + res = erts_finalize_time_offset(); + erts_smp_proc_lock(BIF_P, ERTS_PROC_LOCK_MAIN); + switch (res) { + case ERTS_TIME_OFFSET_PRELIMINARY: { + DECL_AM(preliminary); + BIF_RET(AM_preliminary); + } + case ERTS_TIME_OFFSET_FINAL: { + DECL_AM(final); + BIF_RET(AM_final); + } + case ERTS_TIME_OFFSET_VOLATILE: { + DECL_AM(volatile); + BIF_RET(AM_volatile); + } + default: + ERTS_INTERNAL_ERROR("Unknown state"); + } } else if (ERTS_IS_ATOM_STR("scheduling_statistics", BIF_ARG_1)) { int what; if (ERTS_IS_ATOM_STR("disable", BIF_ARG_2)) @@ -4647,7 +4754,7 @@ BIF_RETTYPE bump_reductions_1(BIF_ALIST_1) } BIF_RETTYPE erts_internal_cmp_term_2(BIF_ALIST_2) { - int res = CMP_TERM(BIF_ARG_1,BIF_ARG_2); + Sint res = CMP_TERM(BIF_ARG_1,BIF_ARG_2); /* ensure -1, 0, 1 result */ if (res < 0) { @@ -4783,11 +4890,6 @@ void erts_init_trap_export(Export* ep, Eterm m, Eterm f, Uint a, void erts_init_bif(void) { - reference0 = 0; - reference1 = 0; - reference2 = 0; - - erts_smp_spinlock_init(&make_ref_lock, "make_ref"); erts_smp_mtx_init(&ports_snapshot_mtx, "ports_snapshot"); erts_smp_atomic_init_nob(&erts_dead_ports_ptr, (erts_aint_t) NULL); @@ -4804,9 +4906,17 @@ void erts_init_bif(void) #endif , &bif_return_trap); - flush_monitor_message_trap = erts_export_put(am_erlang, - am_flush_monitor_message, - 2); + erts_init_trap_export(&dsend_continue_trap_export, + am_erts_internal, am_dsend_continue_trap, 1, + dsend_continue_trap_1); + + flush_monitor_messages_trap = erts_export_put(am_erts_internal, + am_flush_monitor_messages, + 3); + + erts_convert_time_unit_trap = erts_export_put(am_erlang, + am_convert_time_unit, + 3); set_cpu_topology_trap = erts_export_put(am_erlang, am_set_cpu_topology, diff --git a/erts/emulator/beam/bif.h b/erts/emulator/beam/bif.h index 51b77a95ed..d461c3f479 100644 --- a/erts/emulator/beam/bif.h +++ b/erts/emulator/beam/bif.h @@ -21,6 +21,7 @@ #define __BIF_H__ extern Export* erts_format_cpu_topology_trap; +extern Export *erts_convert_time_unit_trap; #define BIF_RETTYPE Eterm @@ -30,10 +31,12 @@ extern Export* erts_format_cpu_topology_trap; #define BIF_ALIST_1 Process* A__p, Eterm* BIF__ARGS #define BIF_ALIST_2 Process* A__p, Eterm* BIF__ARGS #define BIF_ALIST_3 Process* A__p, Eterm* BIF__ARGS +#define BIF_ALIST_4 Process* A__p, Eterm* BIF__ARGS #define BIF_ARG_1 (BIF__ARGS[0]) #define BIF_ARG_2 (BIF__ARGS[1]) #define BIF_ARG_3 (BIF__ARGS[2]) +#define BIF_ARG_4 (BIF__ARGS[3]) #define ERTS_IS_PROC_OUT_OF_REDS(p) \ ((p)->fcalls > 0 \ @@ -124,12 +127,85 @@ do { \ return THE_NON_VALUE; \ } while(0) +#define ERTS_BIF_ERROR_TRAPPED0(Proc, Reason, Bif) \ +do { \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + return THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_ERROR_TRAPPED1(Proc, Reason, Bif, A0) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + return THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_ERROR_TRAPPED2(Proc, Reason, Bif, A0, A1) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + reg[1] = (Eterm) (A1); \ + return THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_ERROR_TRAPPED3(Proc, Reason, Bif, A0, A1, A2) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + reg[1] = (Eterm) (A1); \ + reg[2] = (Eterm) (A2); \ + return THE_NON_VALUE; \ +} while (0) + #define ERTS_BIF_PREP_ERROR(Ret, Proc, Reason) \ do { \ (Proc)->freason = (Reason); \ (Ret) = THE_NON_VALUE; \ } while (0) +#define ERTS_BIF_PREP_ERROR_TRAPPED0(Ret, Proc, Reason, Bif) \ +do { \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + (Ret) = THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_PREP_ERROR_TRAPPED1(Ret, Proc, Reason, Bif, A0) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + (Ret) = THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_PREP_ERROR_TRAPPED2(Ret, Proc, Reason, Bif, A0, A1) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + reg[1] = (Eterm) (A1); \ + (Ret) = THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_PREP_ERROR_TRAPPED3(Ret, Proc, Reason, Bif, A0, A1, A2) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + reg[1] = (Eterm) (A1); \ + reg[2] = (Eterm) (A2); \ + (Ret) = THE_NON_VALUE; \ +} while (0) #define ERTS_BIF_PREP_TRAP0(Ret, Trap, Proc) \ do { \ @@ -392,6 +468,54 @@ erts_bif_prep_await_proc_exit_apply_trap(Process *c_p, Eterm args[], int nargs); +#ifdef ERL_WANT_HIPE_BIF_WRAPPER__ + +#ifndef HIPE + +#define HIPE_WRAPPER_BIF_DISABLE_GC(BIF_NAME, ARITY) + +#else + +#include "erl_fun.h" +#include "hipe_mode_switch.h" + +/* + * Hipe wrappers used by native code for BIFs that disable GC while trapping. + * Also add usage of the wrapper in ../hipe/hipe_bif_list.m4 + * + * Problem: + * When native code calls a BIF that traps, hipe_mode_switch will push a + * "trap frame" on the Erlang stack in order to find its way back from beam_emu + * back to native caller when finally done. If GC is disabled and stack/heap + * is full there is no place to push the "trap frame". + * + * Solution: + * We reserve space on stack for the "trap frame" here before the BIF is called. + * If the BIF does not trap, the space is reclaimed here before returning. + * If the BIF traps, hipe_push_beam_trap_frame() will detect that a "trap frame" + * already is reserved and use it. + */ + + +#define HIPE_WRAPPER_BIF_DISABLE_GC(BIF_NAME, ARITY) \ +BIF_RETTYPE hipe_wrapper_ ## BIF_NAME ## _ ## ARITY (Process* c_p, \ + Eterm* args); \ +BIF_RETTYPE hipe_wrapper_ ## BIF_NAME ## _ ## ARITY (Process* c_p, \ + Eterm* args) \ +{ \ + BIF_RETTYPE res; \ + hipe_reserve_beam_trap_frame(c_p, args, ARITY); \ + res = BIF_NAME ## _ ## ARITY (c_p, args); \ + if (is_value(res) || c_p->freason != TRAP) { \ + hipe_unreserve_beam_trap_frame(c_p); \ + } \ + return res; \ +} + +#endif + +#endif /* ERL_WANT_HIPE_BIF_WRAPPER__ */ + #include "erl_bif_table.h" #endif diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab index 2d888862bf..471f687101 100644 --- a/erts/emulator/beam/bif.tab +++ b/erts/emulator/beam/bif.tab @@ -45,6 +45,7 @@ bif erlang:apply/3 bif erlang:atom_to_list/1 bif erlang:binary_to_list/1 bif erlang:binary_to_list/3 +bif erlang:binary_to_term/1 bif erlang:crc32/1 bif erlang:crc32/2 bif erlang:crc32_combine/3 @@ -91,6 +92,8 @@ bif erlang:loaded/0 bif erlang:localtime/0 bif erlang:localtime_to_universaltime/2 bif erlang:make_ref/0 +bif erlang:unique_integer/0 +bif erlang:unique_integer/1 bif erlang:md5/1 bif erlang:md5_init/0 bif erlang:md5_update/2 @@ -103,6 +106,13 @@ ubif erlang:node/1 ubif erlang:node/0 bif erlang:nodes/1 bif erlang:now/0 +bif erlang:monotonic_time/0 +bif erlang:monotonic_time/1 +bif erlang:system_time/0 +bif erlang:system_time/1 +bif erlang:time_offset/0 +bif erlang:time_offset/1 +bif erlang:timestamp/0 bif erlang:open_port/2 @@ -151,12 +161,22 @@ bif erts_internal:port_command/3 bif erts_internal:port_control/3 bif erts_internal:port_close/1 bif erts_internal:port_connect/2 -bif erts_internal:binary_to_term/1 -bif erts_internal:binary_to_term/2 bif erts_internal:request_system_task/3 bif erts_internal:check_process_code/2 +bif erts_internal:map_to_tuple_keys/1 +bif erts_internal:map_type/1 +bif erts_internal:map_hashmap_children/1 + +bif erts_internal:time_unit/0 + +bif erts_internal:get_bif_timer_servers/0 +bif erts_internal:create_bif_timer/0 +bif erts_internal:access_bif_timer/1 + +bif erts_internal:monitor_process/2 +bif erts_internal:is_system_process/1 # inet_db support bif erlang:port_set_data/2 @@ -194,16 +214,12 @@ bif math:erf/1 bif math:erfc/1 bif math:exp/1 bif math:log/1 +bif math:log2/1 bif math:log10/1 bif math:sqrt/1 bif math:atan2/2 bif math:pow/2 -bif erlang:start_timer/3 -bif erlang:send_after/3 -bif erlang:cancel_timer/1 -bif erlang:read_timer/1 - bif erlang:make_tuple/2 bif erlang:append_element/2 bif erlang:make_tuple/3 @@ -347,6 +363,8 @@ bif os:getenv/0 bif os:getenv/1 bif os:getpid/0 bif os:timestamp/0 +bif os:system_time/0 +bif os:system_time/1 # # Bifs in the erl_ddll module (the module actually does not exist) @@ -480,6 +498,11 @@ bif erlang:call_on_load_function/1 bif erlang:finish_after_on_load/2 # +# New Bifs in R13B04 +# +bif erlang:binary_to_term/2 + +# # The binary match bifs (New in R14A - EEP9) # @@ -573,7 +596,7 @@ bif io:printable_range/0 bif os:unsetenv/1 # -# New in R17A +# New in 17.0 # bif re:inspect/2 @@ -595,6 +618,21 @@ bif maps:values/1 bif erts_internal:cmp_term/2 +bif ets:take/2 + +# +# New in 17.1 +# + +bif erlang:fun_info_mfa/1 + +# New in 18.0 +# + +bif erlang:get_keys/0 +bif ets:update_counter/4 +bif erts_debug:map_info/1 + # # Obsolete # diff --git a/erts/emulator/beam/big.c b/erts/emulator/beam/big.c index 41a041eba6..a4ea9c59ca 100644 --- a/erts/emulator/beam/big.c +++ b/erts/emulator/beam/big.c @@ -274,7 +274,9 @@ _b = _b << _s; \ _vn1 = _b >> H_EXP; \ _vn0 = _b & LO_MASK; \ - _un32 = (_a1 << _s) | ((_a0>>(D_EXP-_s)) & (-_s >> (D_EXP-1))); \ + /* If needed to avoid undefined behaviour */ \ + if (_s) _un32 = (_a1 << _s) | ((_a0>>(D_EXP-_s)) & (-_s >> (D_EXP-1))); \ + else _un32 = _a1; \ _un10 = _a0 << _s; \ _un1 = _un10 >> H_EXP; \ _un0 = _un10 & LO_MASK; \ @@ -1506,13 +1508,15 @@ Eterm uword_to_big(UWord x, Eterm *y) */ Eterm small_to_big(Sint x, Eterm *y) { + Uint xu; if (x >= 0) { + xu = x; *y = make_pos_bignum_header(1); } else { - x = -x; + xu = -(Uint)x; *y = make_neg_bignum_header(1); } - BIG_DIGIT(y, 0) = x; + BIG_DIGIT(y, 0) = xu; return make_big(y); } @@ -1540,21 +1544,24 @@ Eterm erts_uint64_to_big(Uint64 x, Eterm **hpp) Eterm erts_sint64_to_big(Sint64 x, Eterm **hpp) { Eterm *hp = *hpp; + Uint64 ux; int neg; - if (x >= 0) + if (x >= 0) { neg = 0; + ux = x; + } else { neg = 1; - x = -x; + ux = -(Uint64)x; } #if defined(ARCH_32) || HALFWORD_HEAP - if (x >= (((Uint64) 1) << 32)) { + if (ux >= (((Uint64) 1) << 32)) { if (neg) *hp = make_neg_bignum_header(2); else *hp = make_pos_bignum_header(2); - BIG_DIGIT(hp, 0) = (Uint) (x & ((Uint) 0xffffffff)); - BIG_DIGIT(hp, 1) = (Uint) ((x >> 32) & ((Uint) 0xffffffff)); + BIG_DIGIT(hp, 0) = (Uint) (ux & ((Uint) 0xffffffff)); + BIG_DIGIT(hp, 1) = (Uint) ((ux >> 32) & ((Uint) 0xffffffff)); *hpp += 3; } else @@ -1564,12 +1571,52 @@ Eterm erts_sint64_to_big(Sint64 x, Eterm **hpp) *hp = make_neg_bignum_header(1); else *hp = make_pos_bignum_header(1); - BIG_DIGIT(hp, 0) = (Uint) x; + BIG_DIGIT(hp, 0) = (Uint) ux; *hpp += 2; } return make_big(hp); } +Eterm +erts_uint64_array_to_big(Uint **hpp, int neg, int len, Uint64 *array) +{ + Uint *headerp; + int i, pot_digits, digits; + + headerp = *hpp; + + pot_digits = digits = 0; + for (i = 0; i < len; i++) { +#if defined(ARCH_32) || HALFWORD_HEAP + Uint low_val = array[i] & ((Uint) 0xffffffff); + Uint high_val = (array[i] >> 32) & ((Uint) 0xffffffff); + BIG_DIGIT(headerp, pot_digits) = low_val; + pot_digits++; + if (low_val) + digits = pot_digits; + BIG_DIGIT(headerp, pot_digits) = high_val; + pot_digits++; + if (high_val) + digits = pot_digits; +#else + Uint val = array[i]; + BIG_DIGIT(headerp, pot_digits) = val; + pot_digits++; + if (val) + digits = pot_digits; +#endif + } + + if (neg) + *headerp = make_neg_bignum_header(digits); + else + *headerp = make_pos_bignum_header(digits); + + *hpp = headerp + 1 + digits; + + return make_big(headerp); +} + /* ** Convert a bignum to a double float */ @@ -1893,6 +1940,8 @@ Eterm bytes_to_big(byte *xp, dsize_t xsz, int xsgn, Eterm *r) *rwp = d; rwp++; } + if (rsz > BIG_ARITY_MAX) + return NIL; if (xsgn) { *r = make_neg_bignum_header(rsz); } @@ -2667,9 +2716,6 @@ Eterm erts_chars_to_integer(Process *BIF_P, char *bytes, res = big_plus_small(res, m, hp); } - if (is_big(res)) /* check if small */ - res = big_plus_small(res, 0, hp); /* includes conversion to small */ - if (neg) { if (is_small(res)) res = make_small(-signed_val(res)); @@ -2679,8 +2725,12 @@ Eterm erts_chars_to_integer(Process *BIF_P, char *bytes, } } - if (is_big(res)) { - hp += (big_arity(res) + 1); + if (is_not_small(res)) { + res = big_plus_small(res, 0, hp); /* includes conversion to small */ + + if (is_not_small(res)) { + hp += (big_arity(res) + 1); + } } HRelease(BIF_P, hp_end, hp); goto bytebuf_to_integer_1_done; diff --git a/erts/emulator/beam/big.h b/erts/emulator/beam/big.h index d80111822e..4e4611de16 100644 --- a/erts/emulator/beam/big.h +++ b/erts/emulator/beam/big.h @@ -101,9 +101,12 @@ typedef Uint dsize_t; /* Vector size type */ #define ERTS_SINT64_HEAP_SIZE(X) \ (IS_SSMALL((X)) \ ? 0 \ - : ERTS_UINT64_BIG_HEAP_SIZE__((X) >= 0 ? (X) : -(X))) + : ERTS_UINT64_BIG_HEAP_SIZE__((X) >= 0 ? (X) : -(Uint64)(X))) #define ERTS_UINT64_HEAP_SIZE(X) \ (IS_USMALL(0, (X)) ? 0 : ERTS_UINT64_BIG_HEAP_SIZE__((X))) +#define ERTS_MAX_SINT64_HEAP_SIZE (1 + 2) +#define ERTS_MAX_UINT64_HEAP_SIZE (1 + 2) +#define ERTS_UINT64_ARRAY_TO_BIG_MAX_HEAP_SZ(LEN) (2*(LEN)+1) #else @@ -111,6 +114,9 @@ typedef Uint dsize_t; /* Vector size type */ (IS_SSMALL((X)) ? 0 : (1 + 1)) #define ERTS_UINT64_HEAP_SIZE(X) \ (IS_USMALL(0, (X)) ? 0 : (1 + 1)) +#define ERTS_MAX_SINT64_HEAP_SIZE (1 + 1) +#define ERTS_MAX_UINT64_HEAP_SIZE (1 + 1) +#define ERTS_UINT64_ARRAY_TO_BIG_MAX_HEAP_SZ(LEN) ((LEN)+1) #endif @@ -156,6 +162,7 @@ int term_to_Uint(Eterm, Uint*); int term_to_UWord(Eterm, UWord*); int term_to_Sint(Eterm, Sint*); #if HAVE_INT64 +Eterm erts_uint64_array_to_big(Uint **, int, int, Uint64 *); int term_to_Uint64(Eterm, Uint64*); int term_to_Sint64(Eterm, Sint64*); #endif diff --git a/erts/emulator/beam/binary.c b/erts/emulator/beam/binary.c index c7926f18af..cc0b3b9b6c 100644 --- a/erts/emulator/beam/binary.c +++ b/erts/emulator/beam/binary.c @@ -26,17 +26,18 @@ #include "global.h" #include "erl_process.h" #include "error.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "erl_binary.h" #include "erl_bits.h" -#ifdef DEBUG -static int list_to_bitstr_buf(Eterm obj, char* buf, Uint len); -#else -static int list_to_bitstr_buf(Eterm obj, char* buf); -#endif -static int bitstr_list_len(Eterm obj, Uint* num_bytes); +static Export binary_to_list_continue_export; +static Export list_to_binary_continue_export; + +static BIF_RETTYPE binary_to_list_continue(BIF_ALIST_1); +static BIF_RETTYPE list_to_binary_continue(BIF_ALIST_1); void erts_init_binary(void) @@ -49,6 +50,15 @@ erts_init_binary(void) "Internal error: Address of orig_bytes[0] of a Binary" " is *not* 8-byte aligned\n"); } + + erts_init_trap_export(&binary_to_list_continue_export, + am_erts_internal, am_binary_to_list_continue, 1, + &binary_to_list_continue); + + erts_init_trap_export(&list_to_binary_continue_export, + am_erts_internal, am_list_to_binary_continue, 1, + &list_to_binary_continue); + } /* @@ -75,8 +85,6 @@ new_binary(Process *p, byte *buf, Uint len) * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(len); - bptr->flags = 0; - bptr->orig_size = len; erts_refc_init(&bptr->refc, 1); if (buf != NULL) { sys_memcpy(bptr->orig_bytes, buf, len); @@ -114,8 +122,6 @@ Eterm erts_new_mso_binary(Process *p, byte *buf, int len) * Allocate the binary struct itself. */ bptr = erts_bin_nrml_alloc(len); - bptr->flags = 0; - bptr->orig_size = len; erts_refc_init(&bptr->refc, 1); if (buf != NULL) { sys_memcpy(bptr->orig_bytes, buf, len); @@ -169,7 +175,6 @@ erts_realloc_binary(Eterm bin, size_t size) } else { /* REFC */ ProcBin* pb = (ProcBin *) bval; Binary* newbin = erts_bin_realloc(pb->val, size); - newbin->orig_size = size; pb->val = newbin; pb->size = size; pb->bytes = (byte*) newbin->orig_bytes; @@ -333,6 +338,132 @@ BIF_RETTYPE integer_to_binary_1(BIF_ALIST_1) BIF_RET(res); } +#define ERTS_B2L_BYTES_PER_REDUCTION 256 + +typedef struct { + Eterm res; + Eterm *hp; +#ifdef DEBUG + Eterm *hp_end; +#endif + byte *bytes; + Uint size; + Uint bitoffs; +} ErtsB2LState; + +static void b2l_state_destructor(Binary *mbp) +{ + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(mbp) == b2l_state_destructor); +} + +static BIF_RETTYPE +binary_to_list_chunk(Process *c_p, + Eterm mb_eterm, + ErtsB2LState* sp, + int reds_left, + int gc_disabled) +{ + BIF_RETTYPE ret; + int bump_reds; + Uint size; + byte *bytes; + + size = (reds_left + 1)*ERTS_B2L_BYTES_PER_REDUCTION; + if (size > sp->size) + size = sp->size; + bytes = sp->bytes + (sp->size - size); + + bump_reds = (size - 1)/ERTS_B2L_BYTES_PER_REDUCTION + 1; + BUMP_REDS(c_p, bump_reds); + + ASSERT(is_list(sp->res) || is_nil(sp->res)); + + sp->res = erts_bin_bytes_to_list(sp->res, + sp->hp, + bytes, + size, + sp->bitoffs); + sp->size -= size; + sp->hp += 2*size; + + if (sp->size > 0) { + + if (!gc_disabled) + erts_set_gc_state(c_p, 0); + + ASSERT(c_p->flags & F_DISABLE_GC); + ASSERT(is_value(mb_eterm)); + ERTS_BIF_PREP_TRAP1(ret, + &binary_to_list_continue_export, + c_p, + mb_eterm); + } + else { + + ASSERT(sp->hp == sp->hp_end); + ASSERT(sp->size == 0); + + if (!gc_disabled || !erts_set_gc_state(c_p, 1)) + ERTS_BIF_PREP_RET(ret, sp->res); + else + ERTS_BIF_PREP_YIELD_RETURN(ret, c_p, sp->res); + ASSERT(!(c_p->flags & F_DISABLE_GC)); + } + + return ret; +} + +static ERTS_INLINE BIF_RETTYPE +binary_to_list(Process *c_p, Eterm *hp, Eterm tail, byte *bytes, Uint size, Uint bitoffs) +{ + int reds_left = ERTS_BIF_REDS_LEFT(c_p); + if (size < reds_left*ERTS_B2L_BYTES_PER_REDUCTION) { + Eterm res; + BIF_RETTYPE ret; + int bump_reds = (size - 1)/ERTS_B2L_BYTES_PER_REDUCTION + 1; + BUMP_REDS(c_p, bump_reds); + res = erts_bin_bytes_to_list(tail, hp, bytes, size, bitoffs); + ERTS_BIF_PREP_RET(ret, res); + return ret; + } + else { + Binary *mbp = erts_create_magic_binary(sizeof(ErtsB2LState), + b2l_state_destructor); + ErtsB2LState *sp = ERTS_MAGIC_BIN_DATA(mbp); + Eterm mb; + + sp->res = tail; + sp->hp = hp; +#ifdef DEBUG + sp->hp_end = sp->hp + 2*size; +#endif + sp->bytes = bytes; + sp->size = size; + sp->bitoffs = bitoffs; + + hp = HAlloc(c_p, PROC_BIN_SIZE); + mb = erts_mk_magic_binary_term(&hp, &MSO(c_p), mbp); + return binary_to_list_chunk(c_p, mb, sp, reds_left, 0); + } +} + +static BIF_RETTYPE binary_to_list_continue(BIF_ALIST_1) +{ + Binary *mbp = ((ProcBin *) binary_val(BIF_ARG_1))->val; + + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(mbp) == b2l_state_destructor); + + ASSERT(BIF_P->flags & F_DISABLE_GC); + + return binary_to_list_chunk(BIF_P, + BIF_ARG_1, + (ErtsB2LState*) ERTS_MAGIC_BIN_DATA(mbp), + ERTS_BIF_REDS_LEFT(BIF_P), + 1); +} + +HIPE_WRAPPER_BIF_DISABLE_GC(binary_to_list, 1) + BIF_RETTYPE binary_to_list_1(BIF_ALIST_1) { Eterm real_bin; @@ -354,14 +485,15 @@ BIF_RETTYPE binary_to_list_1(BIF_ALIST_1) } else { Eterm* hp = HAlloc(BIF_P, 2 * size); byte* bytes = binary_bytes(real_bin)+offset; - - BIF_RET(erts_bin_bytes_to_list(NIL, hp, bytes, size, bitoffs)); + return binary_to_list(BIF_P, hp, NIL, bytes, size, bitoffs); } error: BIF_ERROR(BIF_P, BADARG); } +HIPE_WRAPPER_BIF_DISABLE_GC(binary_to_list, 3) + BIF_RETTYPE binary_to_list_3(BIF_ALIST_3) { byte* bytes; @@ -387,12 +519,13 @@ BIF_RETTYPE binary_to_list_3(BIF_ALIST_3) } i = stop-start+1; hp = HAlloc(BIF_P, 2*i); - BIF_RET(erts_bin_bytes_to_list(NIL, hp, bytes+start-1, i, bitoffs)); - + return binary_to_list(BIF_P, hp, NIL, bytes+start-1, i, bitoffs); error: BIF_ERROR(BIF_P, BADARG); } +HIPE_WRAPPER_BIF_DISABLE_GC(bitstring_to_list, 1) + BIF_RETTYPE bitstring_to_list_1(BIF_ALIST_1) { Eterm real_bin; @@ -431,124 +564,441 @@ BIF_RETTYPE bitstring_to_list_1(BIF_ALIST_1) previous = CONS(hp, make_binary(last), previous); hp += 2; } - BIF_RET(erts_bin_bytes_to_list(previous, hp, bytes, size, bitoffs)); + + return binary_to_list(BIF_P, hp, previous, bytes, size, bitoffs); } /* Turn a possibly deep list of ints (and binaries) into */ /* One large binary object */ -/* - * This bif also exists in the binary module, under the name - * binary:list_to_bin/1, why it's divided into interface and - * implementation. Also the backend for iolist_to_binary_1. - */ +typedef enum { + ERTS_L2B_OK, + ERTS_L2B_YIELD, + ERTS_L2B_TYPE_ERROR, + ERTS_L2B_OVERFLOW_ERROR +} ErtsL2BResult; -BIF_RETTYPE erts_list_to_binary_bif(Process *p, Eterm arg) -{ +#define ERTS_L2B_STATE_INITER(C_P, ARG, BIF, SZFunc, TBufFunc) \ + {ERTS_IOLIST2BUF_STATE_INITER((C_P), (ARG)), \ + (ARG), THE_NON_VALUE, (BIF), (SZFunc), (TBufFunc)} + +#define ERTS_L2B_STATE_MOVE(TO, FROM) \ + sys_memcpy((void *) (TO), (void *) (FROM), sizeof(ErtsL2BState)) + +typedef struct ErtsL2BState_ ErtsL2BState; + +struct ErtsL2BState_ { + ErtsIOList2BufState buf; + Eterm arg; Eterm bin; - Eterm h,t; - ErlDrvSizeT size; - byte* bytes; -#ifdef DEBUG - ErlDrvSizeT offset; -#endif + Export *bif; + int (*iolist_to_buf_size)(ErtsIOListState *); + ErlDrvSizeT (*iolist_to_buf)(ErtsIOList2BufState *); +}; + +static ERTS_INLINE ErtsL2BResult +list_to_binary_engine(ErtsL2BState *sp) +{ + ErlDrvSizeT res; + Process *c_p = sp->buf.iolist.c_p; + + /* + * have_size == 0 while sp->iolist_to_buf_size() + * has not finished the calculation. + */ + + if (!sp->buf.iolist.have_size) { + switch (sp->iolist_to_buf_size(&sp->buf.iolist)) { + case ERTS_IOLIST_YIELD: + return ERTS_L2B_YIELD; + case ERTS_IOLIST_OVERFLOW: + return ERTS_L2B_OVERFLOW_ERROR; + case ERTS_IOLIST_TYPE: + return ERTS_L2B_TYPE_ERROR; + case ERTS_IOLIST_OK: + break; + default: + ASSERT(0); + break; + } + + ASSERT(sp->buf.iolist.have_size); + + /* + * Size calculated... Setup state for + * sp->iolist_to_buf_*() + */ + + sp->bin = new_binary(c_p, + (byte *) NULL, + sp->buf.iolist.size); + + if (sp->buf.iolist.size == 0) + return ERTS_L2B_OK; + + sp->buf.buf = (char *) binary_bytes(sp->bin); + sp->buf.len = sp->buf.iolist.size; + sp->buf.iolist.obj = sp->arg; - if (is_nil(arg)) { - BIF_RET(new_binary(p,(byte*)"",0)); + if (sp->buf.iolist.reds_left <= 0) { + BUMP_ALL_REDS(c_p); + return ERTS_L2B_YIELD; + } } - if (is_not_list(arg)) { - goto error; + + ASSERT(sp->buf.iolist.size != 0); + ASSERT(is_value(sp->bin)); + ASSERT(sp->buf.buf); + + res = sp->iolist_to_buf(&sp->buf); + + if (!ERTS_IOLIST_TO_BUF_FAILED(res)) { + ASSERT(res == 0); + return ERTS_L2B_OK; } - /* check for [binary()] case */ - h = CAR(list_val(arg)); - t = CDR(list_val(arg)); - if (is_binary(h) && is_nil(t) && !( - HEADER_SUB_BIN == *(binary_val(h)) && ( - ((ErlSubBin *)binary_val(h))->bitoffs != 0 || - ((ErlSubBin *)binary_val(h))->bitsize != 0 - ))) { - return h; - } - switch (erts_iolist_size(arg, &size)) { - case ERTS_IOLIST_OVERFLOW: BIF_ERROR(p, SYSTEM_LIMIT); - case ERTS_IOLIST_TYPE: goto error; - default: ; - } - bin = new_binary(p, (byte *)NULL, size); - bytes = binary_bytes(bin); -#ifdef DEBUG - offset = -#endif - erts_iolist_to_buf(arg, (char*) bytes, size); - ASSERT(offset == 0); - BIF_RET(bin); + switch (res) { + case ERTS_IOLIST_TO_BUF_YIELD: + return ERTS_L2B_YIELD; + case ERTS_IOLIST_TO_BUF_OVERFLOW: + return ERTS_L2B_OVERFLOW_ERROR; + case ERTS_IOLIST_TO_BUF_TYPE_ERROR: + return ERTS_L2B_TYPE_ERROR; + default: + ERTS_INTERNAL_ERROR("Invalid return value from iolist_to_buf_yielding()"); + return ERTS_L2B_TYPE_ERROR; + } +} + +static void +l2b_state_destructor(Binary *mbp) +{ + ErtsL2BState *sp = ERTS_MAGIC_BIN_DATA(mbp); + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(mbp) == l2b_state_destructor); + DESTROY_SAVED_ESTACK(&sp->buf.iolist.estack); +} + +static ERTS_INLINE Eterm +l2b_final_touch(Process *c_p, ErtsL2BState *sp) +{ + Eterm *hp; + ErlSubBin* sbin; + if (sp->buf.offset == 0) + return sp->bin; + + hp = HAlloc(c_p, ERL_SUB_BIN_SIZE); + ASSERT(sp->buf.offset > 0); + sbin = (ErlSubBin *) hp; + sbin->thing_word = HEADER_SUB_BIN; + sbin->size = sp->buf.iolist.size-1; + sbin->offs = 0; + sbin->orig = sp->bin; + sbin->bitoffs = 0; + sbin->bitsize = sp->buf.offset; + sbin->is_writable = 0; + return make_binary(sbin); +} + +static BIF_RETTYPE +list_to_binary_chunk(Eterm mb_eterm, + ErtsL2BState* sp, + int reds_left, + int gc_disabled) +{ + Eterm err = BADARG; + BIF_RETTYPE ret; + Process *c_p = sp->buf.iolist.c_p; + + sp->buf.iolist.reds_left = reds_left; - error: - BIF_ERROR(p, BADARG); + switch (list_to_binary_engine(sp)) { + + case ERTS_L2B_OK: { + Eterm result = l2b_final_touch(c_p, sp); + if (!gc_disabled || !erts_set_gc_state(c_p, 1)) + ERTS_BIF_PREP_RET(ret, result); + else + ERTS_BIF_PREP_YIELD_RETURN(ret, c_p, result); + ASSERT(!(c_p->flags & F_DISABLE_GC)); + break; + } + case ERTS_L2B_YIELD: + if (!gc_disabled) { + /* first yield... */ + Eterm *hp; + Binary *mbp = erts_create_magic_binary(sizeof(ErtsL2BState), + l2b_state_destructor); + ErtsL2BState *new_sp = ERTS_MAGIC_BIN_DATA(mbp); + + ERTS_L2B_STATE_MOVE(new_sp, sp); + sp = new_sp; + + hp = HAlloc(c_p, PROC_BIN_SIZE); + mb_eterm = erts_mk_magic_binary_term(&hp, &MSO(c_p), mbp); + + ASSERT(is_value(mb_eterm)); + + erts_set_gc_state(c_p, 0); + } + + ASSERT(c_p->flags & F_DISABLE_GC); + + ERTS_BIF_PREP_TRAP1(ret, + &list_to_binary_continue_export, + c_p, + mb_eterm); + break; + + case ERTS_L2B_OVERFLOW_ERROR: + err = SYSTEM_LIMIT; + /* fall through */ + + case ERTS_L2B_TYPE_ERROR: + if (!gc_disabled) + ERTS_BIF_PREP_ERROR(ret, c_p, err); + else { + if (erts_set_gc_state(c_p, 1)) + ERTS_VBUMP_ALL_REDS(c_p); + + ERTS_BIF_PREP_ERROR_TRAPPED1(ret, + c_p, + err, + sp->bif, + sp->arg); + } + + ASSERT(!(c_p->flags & F_DISABLE_GC)); + break; + + default: + ERTS_INTERNAL_ERROR("Invalid return value from list_to_binary_engine()"); + ERTS_BIF_PREP_ERROR(ret,c_p, EXC_INTERNAL_ERROR); + break; + } + return ret; } +static BIF_RETTYPE list_to_binary_continue(BIF_ALIST_1) +{ + Binary *mbp = ((ProcBin *) binary_val(BIF_ARG_1))->val; + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(mbp) == l2b_state_destructor); + + ASSERT(BIF_P->flags & F_DISABLE_GC); + + return list_to_binary_chunk(BIF_ARG_1, + ERTS_MAGIC_BIN_DATA(mbp), + ERTS_BIF_REDS_LEFT(BIF_P), + 1); +} + +BIF_RETTYPE erts_list_to_binary_bif(Process *c_p, Eterm arg, Export *bif) +{ + BIF_RETTYPE ret; + + if (is_nil(arg)) + ERTS_BIF_PREP_RET(ret, new_binary(c_p, (byte *) "", 0)); + else if (is_not_list(arg)) + ERTS_BIF_PREP_ERROR(ret, c_p, BADARG); + else { + /* check for [binary()] case */ + Eterm h = CAR(list_val(arg)); + Eterm t = CDR(list_val(arg)); + if (is_binary(h) + && is_nil(t) + && !(HEADER_SUB_BIN == *(binary_val(h)) + && (((ErlSubBin *)binary_val(h))->bitoffs != 0 + || ((ErlSubBin *)binary_val(h))->bitsize != 0))) { + ERTS_BIF_PREP_RET(ret, h); + } + else { + ErtsL2BState state = ERTS_L2B_STATE_INITER(c_p, + arg, + bif, + erts_iolist_size_yielding, + erts_iolist_to_buf_yielding); + int orig_reds_left = ERTS_BIF_REDS_LEFT(c_p); + + /* + * First try to do it all at once without having to use + * yielding iolist_to_buf(). + */ + state.buf.iolist.reds_left = orig_reds_left; + switch (erts_iolist_size_yielding(&state.buf.iolist)) { + case ERTS_IOLIST_OK: { + ErlDrvSizeT size = state.buf.iolist.size; + Eterm bin; + char *buf; + + if (size == 0) { + ERTS_BIF_PREP_RET(ret, new_binary(c_p, (byte *) NULL, 0)); + break; /* done */ + } + + bin = new_binary(c_p, (byte *) NULL, size); + buf = (char *) binary_bytes(bin); + + if (size < ERTS_IOLIST_TO_BUF_BYTES_PER_RED*CONTEXT_REDS) { + /* An (over) estimation of reductions needed */ + int reds_left = state.buf.iolist.reds_left; + int to_buf_reds = orig_reds_left - reds_left; + to_buf_reds += size/ERTS_IOLIST_TO_BUF_BYTES_PER_RED; + if (to_buf_reds <= reds_left) { + ErlDrvSizeT res; + + res = erts_iolist_to_buf(arg, buf, size); + if (res == 0) { + BUMP_REDS(c_p, to_buf_reds); + ERTS_BIF_PREP_RET(ret, bin); + break; /* done */ + } + if (!ERTS_IOLIST_TO_BUF_FAILED(res)) + ERTS_INTERNAL_ERROR("iolist_size/iolist_to_buf missmatch"); + if (res == ERTS_IOLIST_TO_BUF_OVERFLOW) + goto overflow; + goto type_error; + } + } + /* + * Since size has been computed list_to_binary_chunk() expects + * state prepared for iolist_to_buf. + */ + state.bin = bin; + state.buf.buf = buf; + state.buf.len = size; + state.buf.iolist.obj = arg; + /* Fall through... */ + } + case ERTS_IOLIST_YIELD: + ret = list_to_binary_chunk(THE_NON_VALUE, + &state, + state.buf.iolist.reds_left, + 0); + break; + case ERTS_IOLIST_OVERFLOW: + overflow: + ERTS_BIF_PREP_ERROR(ret, c_p, SYSTEM_LIMIT); + break; + case ERTS_IOLIST_TYPE: + type_error: + default: + ERTS_BIF_PREP_ERROR(ret, c_p, BADARG); + break; + } + } + } + return ret; +} + +HIPE_WRAPPER_BIF_DISABLE_GC(list_to_binary, 1) + BIF_RETTYPE list_to_binary_1(BIF_ALIST_1) { - return erts_list_to_binary_bif(BIF_P, BIF_ARG_1); + return erts_list_to_binary_bif(BIF_P, BIF_ARG_1, bif_export[BIF_list_to_binary_1]); } -/* Turn a possibly deep list of ints (and binaries) into */ -/* One large binary object */ +HIPE_WRAPPER_BIF_DISABLE_GC(iolist_to_binary, 1) BIF_RETTYPE iolist_to_binary_1(BIF_ALIST_1) { if (is_binary(BIF_ARG_1)) { BIF_RET(BIF_ARG_1); } - return erts_list_to_binary_bif(BIF_P, BIF_ARG_1); + return erts_list_to_binary_bif(BIF_P, BIF_ARG_1, bif_export[BIF_iolist_to_binary_1]); } +static int bitstr_list_len(ErtsIOListState *); +static ErlDrvSizeT list_to_bitstr_buf_yielding(ErtsIOList2BufState *); +static ErlDrvSizeT list_to_bitstr_buf_not_yielding(ErtsIOList2BufState *); + +HIPE_WRAPPER_BIF_DISABLE_GC(list_to_bitstring, 1) + BIF_RETTYPE list_to_bitstring_1(BIF_ALIST_1) { - Eterm bin; - Uint sz; - int offset; - byte* bytes; - ErlSubBin* sb1; - Eterm* hp; - - if (is_nil(BIF_ARG_1)) { - BIF_RET(new_binary(BIF_P,(byte*)"",0)); - } - if (is_not_list(BIF_ARG_1)) { - error: - BIF_ERROR(BIF_P, BADARG); - } - switch (bitstr_list_len(BIF_ARG_1, &sz)) { - case ERTS_IOLIST_TYPE: - goto error; - case ERTS_IOLIST_OVERFLOW: - BIF_ERROR(BIF_P, SYSTEM_LIMIT); - } - bin = new_binary(BIF_P, (byte *)NULL, sz); - bytes = binary_bytes(bin); -#ifdef DEBUG - offset = list_to_bitstr_buf(BIF_ARG_1, (char*) bytes, sz); -#else - offset = list_to_bitstr_buf(BIF_ARG_1, (char*) bytes); -#endif - ASSERT(offset >= 0); - if (offset > 0) { - hp = HAlloc(BIF_P, ERL_SUB_BIN_SIZE); - sb1 = (ErlSubBin *) hp; - sb1->thing_word = HEADER_SUB_BIN; - sb1->size = sz-1; - sb1->offs = 0; - sb1->orig = bin; - sb1->bitoffs = 0; - sb1->bitsize = offset; - sb1->is_writable = 0; - bin = make_binary(sb1); + BIF_RETTYPE ret; + + if (is_nil(BIF_ARG_1)) + ERTS_BIF_PREP_RET(ret, new_binary(BIF_P, (byte *) "", 0)); + else if (is_not_list(BIF_ARG_1)) + ERTS_BIF_PREP_ERROR(ret, BIF_P, BADARG); + else { + /* check for [bitstring()] case */ + Eterm h = CAR(list_val(BIF_ARG_1)); + Eterm t = CDR(list_val(BIF_ARG_1)); + if (is_binary(h) && is_nil(t)) { + ERTS_BIF_PREP_RET(ret, h); + } + else { + ErtsL2BState state = ERTS_L2B_STATE_INITER(BIF_P, + BIF_ARG_1, + bif_export[BIF_list_to_bitstring_1], + bitstr_list_len, + list_to_bitstr_buf_yielding); + int orig_reds_left = ERTS_BIF_REDS_LEFT(BIF_P); + + /* + * First try to do it all at once without having to use + * yielding list_to_bitstr_buf(). + */ + state.buf.iolist.reds_left = orig_reds_left; + switch (bitstr_list_len(&state.buf.iolist)) { + case ERTS_IOLIST_OK: { + ErlDrvSizeT size = state.buf.iolist.size; + + state.bin = new_binary(BIF_P, (byte *) NULL, size); + state.buf.buf = (char *) binary_bytes(state.bin); + state.buf.len = size; + state.buf.iolist.obj = BIF_ARG_1; + + if (size < ERTS_IOLIST_TO_BUF_BYTES_PER_RED*CONTEXT_REDS) { + /* An (over) estimation of reductions needed */ + int reds_left = state.buf.iolist.reds_left; + int to_buf_reds = orig_reds_left - reds_left; + to_buf_reds += size/ERTS_IOLIST_TO_BUF_BYTES_PER_RED; + if (to_buf_reds <= reds_left) { + ErlDrvSizeT res; + + res = list_to_bitstr_buf_not_yielding(&state.buf); + if (res == 0) { + Eterm res_bin = l2b_final_touch(BIF_P, &state); + BUMP_REDS(BIF_P, to_buf_reds); + ERTS_BIF_PREP_RET(ret, res_bin); + break; /* done */ + } + if (!ERTS_IOLIST_TO_BUF_FAILED(res)) + ERTS_INTERNAL_ERROR("iolist_size/iolist_to_buf missmatch"); + if (res == ERTS_IOLIST_TO_BUF_OVERFLOW) + goto overflow; + goto type_error; + } + } + /* + * Since size has been computed list_to_binary_chunk() expects + * the state prepared for list_to_bitstr_buf. + */ + + /* Fall through... */ + } + case ERTS_IOLIST_YIELD: + ret = list_to_binary_chunk(THE_NON_VALUE, + &state, + state.buf.iolist.reds_left, + 0); + break; + case ERTS_IOLIST_OVERFLOW: + overflow: + ERTS_BIF_PREP_ERROR(ret, BIF_P, SYSTEM_LIMIT); + break; + case ERTS_IOLIST_TYPE: + type_error: + default: + ERTS_BIF_PREP_ERROR(ret, BIF_P, BADARG); + break; + } + } } - - BIF_RET(bin); + + return ret; } BIF_RETTYPE split_binary_2(BIF_ALIST_2) @@ -605,123 +1055,353 @@ BIF_RETTYPE split_binary_2(BIF_ALIST_2) * Local functions. */ +static int +list_to_bitstr_buf_bcopy(ErtsIOList2BufState *state, Eterm obj, int *yield_countp); + /* * The input list is assumed to be type-correct and the buffer is * assumed to be of sufficient size. Those assumptions are verified in * the DEBUG-built emulator. */ -static int +static ErlDrvSizeT +list_to_bitstr_buf(int yield_support, ErtsIOList2BufState *state) +{ + +#undef LIST_TO_BITSTR_BUF_BCOPY_DBG +#undef LIST_TO_BITSTR_BUF_BCOPY #ifdef DEBUG -list_to_bitstr_buf(Eterm obj, char* buf, Uint len) +#define LIST_TO_BITSTR_BUF_BCOPY_DBG \ + len -= size + (offset>7); #else -list_to_bitstr_buf(Eterm obj, char* buf) +#define LIST_TO_BITSTR_BUF_BCOPY_DBG #endif -{ - Eterm* objp; - int offset = 0; +#define LIST_TO_BITSTR_BUF_BCOPY(CONSP) \ + do { \ + byte* bptr; \ + Uint bitsize; \ + Uint bitoffs; \ + Uint num_bits; \ + size_t size = binary_size(obj); \ + if (yield_support) { \ + size_t max_size = ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; \ + if (yield_count > 0) \ + max_size *= yield_count+1; \ + if (size > max_size) { \ + state->objp = CONSP; \ + goto L_bcopy_yield; \ + } \ + if (size >= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT) { \ + int cost = (int) size; \ + cost /= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; \ + yield_count -= cost; \ + } \ + } \ + ASSERT(size <= len); \ + ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); \ + num_bits = 8*size+bitsize; \ + copy_binary_to_buffer(buf, offset, bptr, bitoffs, num_bits); \ + offset += bitsize; \ + buf += size + (offset>7); \ + LIST_TO_BITSTR_BUF_BCOPY_DBG; \ + offset = offset & 7; \ + } while(0) + +#ifdef DEBUG + ErlDrvSizeT len; +#endif + Eterm obj; + char *buf; + Eterm *objp = NULL; + int offset; + int init_yield_count = 0, yield_count; DECLARE_ESTACK(s); - goto L_again; - - while (!ESTACK_ISEMPTY(s)) { - obj = ESTACK_POP(s); - L_again: - if (is_list(obj)) { - L_iter_list: - objp = list_val(obj); - obj = CAR(objp); - if (is_byte(obj)) { - ASSERT(len > 0); - if (offset == 0) { - *buf++ = unsigned_val(obj); - } else { - *buf = (char)((unsigned_val(obj) >> offset) | - ((*buf >> (8-offset)) << (8-offset))); - buf++; - *buf = (unsigned_val(obj) << (8-offset)); - } + + obj = state->iolist.obj; + buf = state->buf; + offset = state->offset; #ifdef DEBUG - len--; + len = state->len; #endif - } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - - ASSERT(size <= len); - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - num_bits = 8*size+bitsize; - copy_binary_to_buffer(buf, offset, bptr, bitoffs, num_bits); - offset += bitsize; - buf += size + (offset>7); + + if (!yield_support) { + yield_count = init_yield_count = 0; /* Shut up faulty warning... >:-( */ + goto L_again; + } + else { + + if (state->iolist.reds_left <= 0) + return ERTS_IOLIST_TO_BUF_YIELD; + + ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK); + init_yield_count = (ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED + * state->iolist.reds_left); + yield_count = init_yield_count; + + if (!state->iolist.estack.start) + goto L_again; + else { + int chk_stack; + /* Restart; restore state... */ + ESTACK_RESTORE(s, &state->iolist.estack); + + if (!state->bcopy.bptr) + chk_stack = 0; + else { + chk_stack = 1; + if (list_to_bitstr_buf_bcopy(state, THE_NON_VALUE, &yield_count)) { + /* Yield again... */ + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + ESTACK_SAVE(s, &state->iolist.estack); + return ERTS_IOLIST_TO_BUF_YIELD; + } + buf = state->buf; + offset = state->offset; #ifdef DEBUG - len -= size + (offset>7); + len = state->len; #endif - offset = offset & 7; - } else if (is_list(obj)) { - ESTACK_PUSH(s, CDR(objp)); - goto L_iter_list; /* on head */ - } else { - ASSERT(is_nil(obj)); } - obj = CDR(objp); - if (is_list(obj)) { - goto L_iter_list; /* on tail */ - } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - - ASSERT(size <= len); - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - num_bits = 8*size+bitsize; - copy_binary_to_buffer(buf, offset, bptr, bitoffs, num_bits); - offset += bitsize; - buf += size+(offset>7); + objp = state->objp; + state->objp = NULL; + + if (objp) + goto L_tail; + if (!chk_stack) + goto L_again; + /* check stack */ + } + } + + while (!ESTACK_ISEMPTY(s)) { + obj = ESTACK_POP(s); + L_again: + if (is_list(obj)) { + while (1) { /* Tail loop */ + while (1) { /* Head loop */ + if (yield_support && --yield_count <= 0) + goto L_yield; + objp = list_val(obj); + obj = CAR(objp); + if (is_byte(obj)) { + ASSERT(len > 0); + if (offset == 0) { + *buf++ = unsigned_val(obj); + } else { + *buf = (char)((unsigned_val(obj) >> offset) | + ((*buf >> (8-offset)) << (8-offset))); + buf++; + *buf = (unsigned_val(obj) << (8-offset)); + } #ifdef DEBUG - len -= size+(offset>7); + len--; #endif - offset = offset & 7; - } else { - ASSERT(is_nil(obj)); + } else if (is_binary(obj)) { + LIST_TO_BITSTR_BUF_BCOPY(objp); + } else if (is_list(obj)) { + ESTACK_PUSH(s, CDR(objp)); + continue; /* Head loop */ + } else { + ASSERT(is_nil(obj)); + } + break; + } + + L_tail: + + obj = CDR(objp); + if (is_list(obj)) { + continue; /* Tail loop */ + } else if (is_binary(obj)) { + LIST_TO_BITSTR_BUF_BCOPY(NULL); + } else { + ASSERT(is_nil(obj)); + } + break; } } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - - ASSERT(size <= len); - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - num_bits = 8*size+bitsize; - copy_binary_to_buffer(buf, offset, bptr, bitoffs, num_bits); - offset += bitsize; - buf += size + (offset>7); -#ifdef DEBUG - len -= size + (offset>7); -#endif - offset = offset & 7; + LIST_TO_BITSTR_BUF_BCOPY(NULL); } else { + if (yield_support && --yield_count <= 0) + goto L_yield; ASSERT(is_nil(obj)); } } DESTROY_ESTACK(s); - return offset; + + if (yield_support) { + int reds; + CLEAR_SAVED_ESTACK(&state->iolist.estack); + reds = ((init_yield_count - yield_count - 1) + / ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED) + 1; + BUMP_REDS(state->iolist.c_p, reds); + state->iolist.reds_left -= reds; + if (state->iolist.reds_left < 0) + state->iolist.reds_left = 0; + } + state->buf = buf; + state->offset = offset; + return 0; + +L_bcopy_yield: + + state->buf = buf; + state->offset = offset; +#ifdef DEBUG + state->len = len; +#endif + + if (list_to_bitstr_buf_bcopy(state, obj, &yield_count) == 0) + ERTS_INTERNAL_ERROR("Missing yield"); + + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + ESTACK_SAVE(s, &state->iolist.estack); + return ERTS_IOLIST_TO_BUF_YIELD; + +L_yield: + + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + state->iolist.obj = obj; + state->buf = buf; + state->offset = offset; + ESTACK_SAVE(s, &state->iolist.estack); +#ifdef DEBUG + state->len = len; +#endif + return ERTS_IOLIST_TO_BUF_YIELD; + + +#undef LIST_TO_BITSTR_BUF_BCOPY_DBG +#undef LIST_TO_BITSTR_BUF_BCOPY + +} + +static ErlDrvSizeT +list_to_bitstr_buf_yielding(ErtsIOList2BufState *state) +{ + return list_to_bitstr_buf(1, state); +} + +static ErlDrvSizeT +list_to_bitstr_buf_not_yielding(ErtsIOList2BufState *state) +{ + return list_to_bitstr_buf(0, state); } static int -bitstr_list_len(Eterm obj, Uint* num_bytes) +list_to_bitstr_buf_bcopy(ErtsIOList2BufState *state, Eterm obj, int *yield_countp) +{ + int res; + char *buf = state->buf; + char *next_buf; + int offset = state->offset; + int next_offset; +#ifdef DEBUG + ErlDrvSizeT len = state->len; + ErlDrvSizeT next_len; +#endif + byte* bptr; + size_t size; + size_t max_size; + Uint bitoffs; + Uint num_bits; + Uint bitsize; + int yield_count = *yield_countp; + + if (state->bcopy.bptr) { + bptr = state->bcopy.bptr; + size = state->bcopy.size; + bitoffs = state->bcopy.bitoffs; + bitsize = state->bcopy.bitsize; + state->bcopy.bptr = NULL; + } + else { + + ASSERT(is_binary(obj)); + + size = binary_size(obj); + + ASSERT(size <= len); + + ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); + } + + max_size = (size_t) ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; + if (yield_count > 0) + max_size *= (size_t) (yield_count+1); + + if (size <= max_size) { + if (size >= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT) { + int cost = (int) size; + cost /= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; + yield_count -= cost; + } + next_offset = offset + bitsize; + next_buf = buf + size+(next_offset>7); +#ifdef DEBUG + next_len = len - size+(next_offset>7); +#endif + next_offset &= 7; + num_bits = 8*size+bitsize; + res = 0; + } + else { + ASSERT(0 < max_size && max_size < size); + yield_count = 0; + state->bcopy.bptr = bptr + max_size; + state->bcopy.bitoffs = bitoffs; + state->bcopy.bitsize = bitsize; + state->bcopy.size = size - max_size; + next_buf = buf + max_size; +#ifdef DEBUG + next_len = len - max_size; +#endif + next_offset = offset; + num_bits = 8*max_size; + size = max_size; + res = 1; + } + + copy_binary_to_buffer(buf, offset, bptr, bitoffs, num_bits); + + state->offset = next_offset; + state->buf = next_buf; +#ifdef DEBUG + state->len = next_len; +#endif + *yield_countp = yield_count; + + return res; +} + +static int +bitstr_list_len(ErtsIOListState *state) { Eterm* objp; - Uint len = 0; - Uint offs = 0; + Eterm obj; + Uint len, offs; + int res, init_yield_count, yield_count; DECLARE_ESTACK(s); + + if (state->reds_left <= 0) + return ERTS_IOLIST_YIELD; + + len = (Uint) state->size; + offs = state->offs; + obj = state->obj; + + ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK); + init_yield_count = ERTS_IOLIST_SIZE_YIELDS_COUNT_PER_RED; + init_yield_count *= state->reds_left; + yield_count = init_yield_count; + if (state->estack.start) { + /* Restart; restore estack... */ + ESTACK_RESTORE(s, &state->estack); + } + goto L_again; #define SAFE_ADD(Var, Val) \ @@ -748,46 +1428,55 @@ bitstr_list_len(Eterm obj, Uint* num_bytes) obj = ESTACK_POP(s); L_again: if (is_list(obj)) { - L_iter_list: - objp = list_val(obj); - /* Head */ - obj = CAR(objp); - if (is_byte(obj)) { - len++; - if (len == 0) { - goto L_overflow_error; + while (1) { /* Tail loop */ + while (1) { /* Head loop */ + if (--yield_count <= 0) + goto L_yield; + objp = list_val(obj); + /* Head */ + obj = CAR(objp); + if (is_byte(obj)) { + len++; + if (len == 0) { + goto L_overflow_error; + } + } else if (is_binary(obj)) { + SAFE_ADD(len, binary_size(obj)); + SAFE_ADD_BITSIZE(offs, obj); + } else if (is_list(obj)) { + ESTACK_PUSH(s, CDR(objp)); + continue; /* Head loop */ + } else if (is_not_nil(obj)) { + goto L_type_error; + } + break; } - } else if (is_binary(obj)) { - SAFE_ADD(len, binary_size(obj)); - SAFE_ADD_BITSIZE(offs, obj); - } else if (is_list(obj)) { - ESTACK_PUSH(s, CDR(objp)); - goto L_iter_list; /* on head */ - } else if (is_not_nil(obj)) { - goto L_type_error; + /* Tail */ + obj = CDR(objp); + if (is_list(obj)) + continue; /* Tail loop */ + else if (is_binary(obj)) { + SAFE_ADD(len, binary_size(obj)); + SAFE_ADD_BITSIZE(offs, obj); + } else if (is_not_nil(obj)) { + goto L_type_error; + } + break; } - /* Tail */ - obj = CDR(objp); - if (is_list(obj)) - goto L_iter_list; /* on tail */ - else if (is_binary(obj)) { + } else { + if (--yield_count <= 0) + goto L_yield; + if (is_binary(obj)) { SAFE_ADD(len, binary_size(obj)); SAFE_ADD_BITSIZE(offs, obj); } else if (is_not_nil(obj)) { goto L_type_error; } - } else if (is_binary(obj)) { - SAFE_ADD(len, binary_size(obj)); - SAFE_ADD_BITSIZE(offs, obj); - } else if (is_not_nil(obj)) { - goto L_type_error; } } #undef SAFE_ADD #undef SAFE_ADD_BITSIZE - DESTROY_ESTACK(s); - /* * Make sure that the number of bits in the bitstring will fit * in an Uint to ensure that the binary can be matched using @@ -800,15 +1489,42 @@ bitstr_list_len(Eterm obj, Uint* num_bytes) if (len << 3 < len) { goto L_overflow_error; } - *num_bytes = len; - return ERTS_IOLIST_OK; + state->size = len; - L_type_error: - DESTROY_ESTACK(s); - return ERTS_IOLIST_TYPE; + res = ERTS_IOLIST_OK; + + L_return: { + int yc = init_yield_count - yield_count; + int reds; + + DESTROY_ESTACK(s); + CLEAR_SAVED_ESTACK(&state->estack); + + reds = (yc - 1)/ERTS_IOLIST_SIZE_YIELDS_COUNT_PER_RED + 1; + BUMP_REDS(state->c_p, reds); + state->reds_left -= reds; + state->size = (ErlDrvSizeT) len; + state->have_size = 1; + return res; + } L_overflow_error: - DESTROY_ESTACK(s); - return ERTS_IOLIST_OVERFLOW; + res = ERTS_IOLIST_OVERFLOW; + len = 0; + goto L_return; + + L_type_error: + res = ERTS_IOLIST_TYPE; + len = 0; + goto L_return; + + L_yield: + BUMP_ALL_REDS(state->c_p); + state->reds_left = 0; + state->size = len; + state->offs = offs; + state->obj = obj; + ESTACK_SAVE(s, &state->estack); + return ERTS_IOLIST_YIELD; } diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c index 7d4f52ee23..e2fa572546 100644 --- a/erts/emulator/beam/break.c +++ b/erts/emulator/beam/break.c @@ -181,7 +181,6 @@ static void doit_print_monitor(ErtsMonitor *mon, void *vpcontext) ASSERT(is_node_name_atom(mon->pid)); erts_print(to, to_arg, "%s{to,{%T,%T},%T}", prefix, mon->name, mon->pid, mon->ref); - erts_print(to, to_arg,"}"); } else if (is_atom(mon->name)){ /* local by name */ erts_print(to, to_arg, "%s{to,{%T,%T},%T}", prefix, mon->name, erts_this_dist_entry->sysname, mon->ref); @@ -210,25 +209,12 @@ print_process_info(int to, void *to_arg, Process *p) erts_print(to, to_arg, "State: "); state = erts_smp_atomic32_read_acqb(&p->state); - if (state & ERTS_PSFLG_FREE) - erts_print(to, to_arg, "Non Existing\n"); /* Should never happen */ - else if (state & ERTS_PSFLG_EXITING) - erts_print(to, to_arg, "Exiting\n"); - else if (state & ERTS_PSFLG_GC) { - garbing = 1; - running = 1; - erts_print(to, to_arg, "Garbing\n"); - } - else if (state & ERTS_PSFLG_SUSPENDED) - erts_print(to, to_arg, "Suspended\n"); - else if (state & ERTS_PSFLG_RUNNING) { - running = 1; - erts_print(to, to_arg, "Running\n"); - } - else if (state & ERTS_PSFLG_ACTIVE) - erts_print(to, to_arg, "Scheduled\n"); - else - erts_print(to, to_arg, "Waiting\n"); + erts_dump_process_state(to, to_arg, state); + if (state & ERTS_PSFLG_GC) { + garbing = 1; + running = 1; + } else if (state & ERTS_PSFLG_RUNNING) + running = 1; /* * If the process is registered as a global process, display the @@ -256,6 +242,7 @@ print_process_info(int to, void *to_arg, Process *p) p->current[1], p->current[2]); } + erts_print(to, to_arg, "Run queue: %d\n", erts_get_runq_proc(p)->ix); erts_print(to, to_arg, "Spawned by: %T\n", p->parent); approx_started = (time_t) p->approx_started; @@ -351,6 +338,10 @@ print_process_info(int to, void *to_arg, Process *p) #endif erts_stack_dump(to, to_arg, p); } + + /* Display all states */ + erts_print(to, to_arg, "Internal State: "); + erts_dump_extended_process_state(to, to_arg, state); } static void @@ -680,27 +671,39 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) char* dumpname; int secs; int env_erl_crash_dump_seconds_set = 1; + int i; if (ERTS_SOMEONE_IS_CRASH_DUMPING) return; #ifdef ERTS_SMP + /* Order all managed threads to block, this has to be done + first to guarantee that this is the only thread to generate + crash dump. */ + erts_thr_progress_fatal_error_block(&tpd_buf); + +#ifdef ERTS_THR_HAVE_SIG_FUNCS /* - * Wait for all managed threads to block. If all threads haven't blocked - * after a minute, we go anyway and hope for the best... - * - * We do not release system again. We expect an exit() or abort() after - * dump has been written. + * We suspend all scheduler threads so that we can dump some + * data about the currently running processes and scheduler data. + * We have to be very very careful when doing this as the schedulers + * could be anywhere. */ - erts_thr_progress_fatal_error_block(60000, &tpd_buf); - /* Either worked or not... */ + for (i = 0; i < erts_no_schedulers; i++) { + erts_tid_t tid = ERTS_SCHEDULER_IX(i)->tid; + if (!erts_equal_tids(tid,erts_thr_self())) + sys_thr_suspend(tid); + } + +#endif /* Allow us to pass certain places without locking... */ erts_smp_atomic32_set_mb(&erts_writing_erl_crash_dump, 1); erts_smp_tsd_set(erts_is_crash_dumping_key, (void *) 1); -#else + +#else /* !ERTS_SMP */ erts_writing_erl_crash_dump = 1; -#endif +#endif /* ERTS_SMP */ envsz = sizeof(env); /* ERL_CRASH_DUMP_SECONDS not set @@ -753,11 +756,12 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) dumpname = "erl_crash.dump"; else dumpname = &dumpnamebuf[0]; + + erts_fprintf(stderr,"\nCrash dump is being written to: %s...", dumpname); fd = open(dumpname,O_WRONLY | O_CREAT | O_TRUNC,0640); - if (fd < 0) + if (fd < 0) return; /* Can't create the crash dump, skip it */ - time(&now); erts_fdprintf(fd, "=erl_crash_dump:0.3\n%s", ctime(&now)); @@ -771,9 +775,74 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) erts_fdprintf(fd, "System version: "); erts_print_system_version(fd, NULL, NULL); erts_fdprintf(fd, "%s\n", "Compiled: " ERLANG_COMPILE_DATE); + erts_fdprintf(fd, "Taints: "); erts_print_nif_taints(fd, NULL); erts_fdprintf(fd, "Atoms: %d\n", atom_table_size()); + +#ifdef USE_THREADS + /* We want to note which thread it was that called erl_exit */ + if (erts_get_scheduler_data()) { + erts_fdprintf(fd, "Calling Thread: scheduler:%d\n", + erts_get_scheduler_data()->no); + } else { + if (!erts_thr_getname(erts_thr_self(), dumpnamebuf, MAXPATHLEN)) + erts_fdprintf(fd, "Calling Thread: %s\n", dumpnamebuf); + else + erts_fdprintf(fd, "Calling Thread: %p\n", erts_thr_self()); + } +#else + erts_fdprintf(fd, "Calling Thread: scheduler:1\n"); +#endif + +#if defined(ERTS_HAVE_TRY_CATCH) + + /* + * erts_print_scheduler_info is not guaranteed to be safe to call + * here for all schedulers as we may have suspended a scheduler + * in the middle of updating the STACK_TOP and STACK_START + * variables and thus when scanning the stack we could get + * segmentation faults. We protect against this very unlikely + * scenario by using the ERTS_SYS_TRY_CATCH. + */ + for (i = 0; i < erts_no_schedulers; i++) { + ERTS_SYS_TRY_CATCH( + erts_print_scheduler_info(fd, NULL, ERTS_SCHEDULER_IX(i)), + erts_fdprintf(fd, "** crashed **\n")); + } +#endif + +#ifdef ERTS_SMP + +#if defined(ERTS_THR_HAVE_SIG_FUNCS) + + /* We resume all schedulers so that we are in a known safe state + when we write the rest of the crash dump */ + for (i = 0; i < erts_no_schedulers; i++) { + erts_tid_t tid = ERTS_SCHEDULER_IX(i)->tid; + if (!erts_equal_tids(tid,erts_thr_self())) + sys_thr_resume(tid); + } +#endif + + /* + * Wait for all managed threads to block. If all threads haven't blocked + * after a minute, we go anyway and hope for the best... + * + * We do not release system again. We expect an exit() or abort() after + * dump has been written. + */ + erts_thr_progress_fatal_error_wait(60000); + /* Either worked or not... */ +#endif + +#ifndef ERTS_HAVE_TRY_CATCH + /* This is safe to call here, as all schedulers are blocked */ + for (i = 0; i < erts_no_schedulers; i++) { + erts_print_scheduler_info(fd, NULL, ERTS_SCHEDULER_IX(i)); + } +#endif + info(fd, NULL); /* General system info */ if (erts_ptab_initialized(&erts_proc)) process_info(fd, NULL); /* Info about each process and port */ @@ -803,7 +872,7 @@ erl_crash_dump_v(char *file, int line, char* fmt, va_list args) erts_fdprintf(fd, "=end\n"); close(fd); - erts_fprintf(stderr,"\nCrash dump was written to: %s\n", dumpname); + erts_fprintf(stderr,"done\n"); } void diff --git a/erts/emulator/beam/copy.c b/erts/emulator/beam/copy.c index 50548850eb..4d12dae787 100644 --- a/erts/emulator/beam/copy.c +++ b/erts/emulator/beam/copy.c @@ -21,6 +21,8 @@ # include "config.h" #endif +#define ERL_WANT_GC_INTERNALS__ + #include "sys.h" #include "erl_vm.h" #include "global.h" @@ -125,6 +127,52 @@ Uint size_object(Eterm obj) obj = *bptr; break; } + case MAP_SUBTAG: + switch (MAP_HEADER_TYPE(hdr)) { + case MAP_HEADER_TAG_FLATMAP_HEAD : + { + Uint n; + flatmap_t *mp; + mp = (flatmap_t*)flatmap_val_rel(obj,base); + ptr = (Eterm *)mp; + n = flatmap_get_size(mp) + 1; + sum += n + 2; + ptr += 2; /* hdr + size words */ + while (n--) { + obj = *ptr++; + if (!IS_CONST(obj)) { + ESTACK_PUSH(s, obj); + } + } + goto pop_next; + } + case MAP_HEADER_TAG_HAMT_HEAD_BITMAP : + case MAP_HEADER_TAG_HAMT_HEAD_ARRAY : + case MAP_HEADER_TAG_HAMT_NODE_BITMAP : + { + Eterm *head; + Uint sz; + head = hashmap_val_rel(obj, base); + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + sum += 1 + sz + header_arity(hdr); + head += 1 + header_arity(hdr); + + if (sz == 0) { + goto pop_next; + } + while(sz-- > 1) { + obj = head[sz]; + if (!IS_CONST(obj)) { + ESTACK_PUSH(s, obj); + } + } + obj = head[0]; + } + break; + default: + erl_exit(ERTS_ABORT_EXIT, "size_object: bad hashmap type %d\n", MAP_HEADER_TYPE(hdr)); + } + break; case SUB_BINARY_SUBTAG: { Eterm real_bin; @@ -152,25 +200,7 @@ Uint size_object(Eterm obj) goto pop_next; } break; - case MAP_SUBTAG: - { - Uint n; - map_t *mp; - mp = (map_t*)map_val_rel(obj,base); - ptr = (Eterm *)mp; - n = map_get_size(mp) + 1; - sum += n + 2; - ptr += 2; /* hdr + size words */ - while (n--) { - obj = *ptr++; - if (!IS_CONST(obj)) { - ESTACK_PUSH(s, obj); - } - } - goto pop_next; - } - break; - case BIN_MATCHSTATE_SUBTAG: + case BIN_MATCHSTATE_SUBTAG: erl_exit(ERTS_ABORT_EXIT, "size_object: matchstate term not allowed"); default: @@ -338,15 +368,6 @@ Eterm copy_struct(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap) } } break; - case MAP_SUBTAG: - { - i = map_get_size(objp) + 3; - *argp = make_map_rel(htop, dst_base); - while (i--) { - *htop++ = *objp++; - } - } - break; case REFC_BINARY_SUBTAG: { ProcBin* pb; @@ -457,7 +478,7 @@ Eterm copy_struct(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap) { ExternalThing *etp = (ExternalThing *) htop; - i = thing_arityval(hdr) + 1; + i = thing_arityval(hdr) + 1; tp = htop; while (i--) { @@ -471,6 +492,28 @@ Eterm copy_struct(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap) *argp = make_external_rel(tp, dst_base); } break; + case MAP_SUBTAG: + tp = htop; + switch (MAP_HEADER_TYPE(hdr)) { + case MAP_HEADER_TAG_FLATMAP_HEAD : + i = flatmap_get_size(objp) + 3; + *argp = make_flatmap_rel(htop, dst_base); + while (i--) { + *htop++ = *objp++; + } + break; + case MAP_HEADER_TAG_HAMT_HEAD_BITMAP : + case MAP_HEADER_TAG_HAMT_HEAD_ARRAY : + *htop++ = *objp++; + case MAP_HEADER_TAG_HAMT_NODE_BITMAP : + i = 1 + hashmap_bitcount(MAP_HEADER_VAL(hdr)); + while (i--) { *htop++ = *objp++; } + *argp = make_hashmap_rel(tp, dst_base); + break; + default: + erl_exit(ERTS_ABORT_EXIT, "copy_struct: bad hashmap type %d\n", MAP_HEADER_TYPE(hdr)); + } + break; case BIN_MATCHSTATE_SUBTAG: erl_exit(ERTS_ABORT_EXIT, "copy_struct: matchstate term not allowed"); diff --git a/erts/emulator/beam/dist.c b/erts/emulator/beam/dist.c index ec07ddcd9c..bfecac1612 100644 --- a/erts/emulator/beam/dist.c +++ b/erts/emulator/beam/dist.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2013. All Rights Reserved. + * Copyright Ericsson AB 1996-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -119,7 +119,7 @@ Export* dmonitor_p_trap = NULL; /* forward declarations */ static void clear_dist_entry(DistEntry*); -static int dsig_send(ErtsDSigData *, Eterm, Eterm, int); +static int dsig_send_ctl(ErtsDSigData* dsdp, Eterm ctl, int force_busy); static void send_nodes_mon_msgs(Process *, Eterm, Eterm, Eterm, Eterm); static void init_nodes_monitors(void); @@ -622,9 +622,7 @@ alloc_dist_obuf(Uint size) ErtsDistOutputBuf *obuf; Uint obuf_size = sizeof(ErtsDistOutputBuf)+sizeof(byte)*(size-1); Binary *bin = erts_bin_drv_alloc(obuf_size); - bin->flags = BIN_FLAG_DRV; erts_refc_init(&bin->refc, 1); - bin->orig_size = (SWord) obuf_size; obuf = (ErtsDistOutputBuf *) &bin->orig_bytes[0]; #ifdef DEBUG obuf->dbg_pattern = ERTS_DIST_OUTPUT_BUF_DBG_PATTERN; @@ -709,6 +707,55 @@ static void clear_dist_entry(DistEntry *dep) } } +void erts_dsend_context_dtor(Binary* ctx_bin) +{ + ErtsSendContext* ctx = ERTS_MAGIC_BIN_DATA(ctx_bin); + switch (ctx->dss.phase) { + case ERTS_DSIG_SEND_PHASE_MSG_SIZE: + DESTROY_SAVED_ESTACK(&ctx->dss.u.sc.estack); + break; + case ERTS_DSIG_SEND_PHASE_MSG_ENCODE: + DESTROY_SAVED_WSTACK(&ctx->dss.u.ec.wstack); + break; + default:; + } + if (ctx->dss.phase >= ERTS_DSIG_SEND_PHASE_ALLOC && ctx->dss.obuf) { + free_dist_obuf(ctx->dss.obuf); + } + if (ctx->dep_to_deref) + erts_deref_dist_entry(ctx->dep_to_deref); +} + +Eterm erts_dsend_export_trap_context(Process* p, ErtsSendContext* ctx) +{ + struct exported_ctx { + ErtsSendContext ctx; + ErtsAtomCacheMap acm; + }; + Binary* ctx_bin = erts_create_magic_binary(sizeof(struct exported_ctx), + erts_dsend_context_dtor); + struct exported_ctx* dst = ERTS_MAGIC_BIN_DATA(ctx_bin); + Uint ctl_size = !HALFWORD_HEAP ? 0 : (arityval(ctx->ctl_heap[0]) + 1); + Eterm* hp = HAlloc(p, ctl_size + PROC_BIN_SIZE); + + sys_memcpy(&dst->ctx, ctx, sizeof(ErtsSendContext)); + ASSERT(ctx->dss.ctl == make_tuple(ctx->ctl_heap)); +#if !HALFWORD_HEAP + dst->ctx.dss.ctl = make_tuple(dst->ctx.ctl_heap); +#else + /* Must put control tuple in low mem */ + sys_memcpy(hp, ctx->ctl_heap, ctl_size*sizeof(Eterm)); + dst->ctx.dss.ctl = make_tuple(hp); + hp += ctl_size; +#endif + if (ctx->dss.acmp) { + sys_memcpy(&dst->acm, ctx->dss.acmp, sizeof(ErtsAtomCacheMap)); + dst->ctx.dss.acmp = &dst->acm; + } + return erts_mk_magic_binary_term(&hp, &MSO(p), ctx_bin); +} + + /* * The erts_dsig_send_*() functions implemented below, sends asynchronous * distributed signals to other Erlang nodes. Before sending a distributed @@ -731,7 +778,7 @@ erts_dsig_send_link(ErtsDSigData *dsdp, Eterm local, Eterm remote) int res; UseTmpHeapNoproc(4); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(4); return res; } @@ -744,7 +791,7 @@ erts_dsig_send_unlink(ErtsDSigData *dsdp, Eterm local, Eterm remote) int res; UseTmpHeapNoproc(4); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(4); return res; } @@ -772,7 +819,7 @@ erts_dsig_send_m_exit(ErtsDSigData *dsdp, Eterm watcher, Eterm watched, erts_smp_de_links_unlock(dsdp->dep); #endif - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 1); + res = dsig_send_ctl(dsdp, ctl, 1); UnUseTmpHeapNoproc(6); return res; } @@ -793,7 +840,7 @@ erts_dsig_send_monitor(ErtsDSigData *dsdp, Eterm watcher, Eterm watched, make_small(DOP_MONITOR_P), watcher, watched, ref); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(5); return res; } @@ -815,18 +862,17 @@ erts_dsig_send_demonitor(ErtsDSigData *dsdp, Eterm watcher, make_small(DOP_DEMONITOR_P), watcher, watched, ref); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, force); + res = dsig_send_ctl(dsdp, ctl, force); UnUseTmpHeapNoproc(5); return res; } int -erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) +erts_dsig_send_msg(Eterm remote, Eterm message, ErtsSendContext* ctx) { Eterm ctl; - DeclareTmpHeapNoproc(ctl_heap,5); Eterm token = NIL; - Process *sender = dsdp->proc; + Process *sender = ctx->dsd.proc; int res; #ifdef USE_VM_PROBES Sint tok_label = 0; @@ -838,8 +884,7 @@ erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) DTRACE_CHARBUF(receiver_name, 64); #endif - UseTmpHeapNoproc(5); - if (SEQ_TRACE_TOKEN(sender) != NIL + if (SEQ_TRACE_TOKEN(sender) != NIL #ifdef USE_VM_PROBES && SEQ_TRACE_TOKEN(sender) != am_have_dt_utag #endif @@ -852,7 +897,7 @@ erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) *node_name = *sender_name = *receiver_name = '\0'; if (DTRACE_ENABLED(message_send) || DTRACE_ENABLED(message_send_remote)) { erts_snprintf(node_name, sizeof(DTRACE_CHARBUF_NAME(node_name)), - "%T", dsdp->dep->sysname); + "%T", ctx->dsd.dep->sysname); erts_snprintf(sender_name, sizeof(DTRACE_CHARBUF_NAME(sender_name)), "%T", sender->common.id); erts_snprintf(receiver_name, sizeof(DTRACE_CHARBUF_NAME(receiver_name)), @@ -867,26 +912,28 @@ erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) #endif if (token != NIL) - ctl = TUPLE4(&ctl_heap[0], + ctl = TUPLE4(&ctx->ctl_heap[0], make_small(DOP_SEND_TT), am_Cookie, remote, token); else - ctl = TUPLE3(&ctl_heap[0], make_small(DOP_SEND), am_Cookie, remote); + ctl = TUPLE3(&ctx->ctl_heap[0], make_small(DOP_SEND), am_Cookie, remote); DTRACE6(message_send, sender_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); DTRACE7(message_send_remote, sender_name, node_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); - res = dsig_send(dsdp, ctl, message, 0); - UnUseTmpHeapNoproc(5); + ctx->dss.ctl = ctl; + ctx->dss.msg = message; + ctx->dss.force_busy = 0; + res = erts_dsig_send(&ctx->dsd, &ctx->dss); return res; } int -erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) +erts_dsig_send_reg_msg(Eterm remote_name, Eterm message, + ErtsSendContext* ctx) { Eterm ctl; - DeclareTmpHeapNoproc(ctl_heap,6); Eterm token = NIL; - Process *sender = dsdp->proc; + Process *sender = ctx->dsd.proc; int res; #ifdef USE_VM_PROBES Sint tok_label = 0; @@ -898,7 +945,6 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) DTRACE_CHARBUF(receiver_name, 128); #endif - UseTmpHeapNoproc(6); if (SEQ_TRACE_TOKEN(sender) != NIL #ifdef USE_VM_PROBES && SEQ_TRACE_TOKEN(sender) != am_have_dt_utag @@ -912,7 +958,7 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) *node_name = *sender_name = *receiver_name = '\0'; if (DTRACE_ENABLED(message_send) || DTRACE_ENABLED(message_send_remote)) { erts_snprintf(node_name, sizeof(DTRACE_CHARBUF_NAME(node_name)), - "%T", dsdp->dep->sysname); + "%T", ctx->dsd.dep->sysname); erts_snprintf(sender_name, sizeof(DTRACE_CHARBUF_NAME(sender_name)), "%T", sender->common.id); erts_snprintf(receiver_name, sizeof(DTRACE_CHARBUF_NAME(receiver_name)), @@ -927,17 +973,19 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) #endif if (token != NIL) - ctl = TUPLE5(&ctl_heap[0], make_small(DOP_REG_SEND_TT), + ctl = TUPLE5(&ctx->ctl_heap[0], make_small(DOP_REG_SEND_TT), sender->common.id, am_Cookie, remote_name, token); else - ctl = TUPLE4(&ctl_heap[0], make_small(DOP_REG_SEND), + ctl = TUPLE4(&ctx->ctl_heap[0], make_small(DOP_REG_SEND), sender->common.id, am_Cookie, remote_name); DTRACE6(message_send, sender_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); DTRACE7(message_send_remote, sender_name, node_name, receiver_name, msize, tok_label, tok_lastcnt, tok_serial); - res = dsig_send(dsdp, ctl, message, 0); - UnUseTmpHeapNoproc(6); + ctx->dss.ctl = ctl; + ctx->dss.msg = message; + ctx->dss.force_busy = 0; + res = erts_dsig_send(&ctx->dsd, &ctx->dss); return res; } @@ -994,7 +1042,7 @@ erts_dsig_send_exit_tt(ErtsDSigData *dsdp, Eterm local, Eterm remote, DTRACE7(process_exit_signal_remote, sender_name, node_name, remote_name, reason_str, tok_label, tok_lastcnt, tok_serial); /* forced, i.e ignore busy */ - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 1); + res = dsig_send_ctl(dsdp, ctl, 1); UnUseTmpHeapNoproc(6); return res; } @@ -1010,7 +1058,7 @@ erts_dsig_send_exit(ErtsDSigData *dsdp, Eterm local, Eterm remote, Eterm reason) ctl = TUPLE4(&ctl_heap[0], make_small(DOP_EXIT), local, remote, reason); /* forced, i.e ignore busy */ - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 1); + res = dsig_send_ctl(dsdp, ctl, 1); UnUseTmpHeapNoproc(5); return res; } @@ -1026,7 +1074,7 @@ erts_dsig_send_exit2(ErtsDSigData *dsdp, Eterm local, Eterm remote, Eterm reason ctl = TUPLE4(&ctl_heap[0], make_small(DOP_EXIT2), local, remote, reason); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(5); return res; } @@ -1043,7 +1091,7 @@ erts_dsig_send_group_leader(ErtsDSigData *dsdp, Eterm leader, Eterm remote) ctl = TUPLE3(&ctl_heap[0], make_small(DOP_GROUP_LEADER), leader, remote); - res = dsig_send(dsdp, ctl, THE_NON_VALUE, 0); + res = dsig_send_ctl(dsdp, ctl, 0); UnUseTmpHeapNoproc(4); return res; } @@ -1693,194 +1741,235 @@ int erts_net_message(Port *prt, return -1; } -static int -dsig_send(ErtsDSigData *dsdp, Eterm ctl, Eterm msg, int force_busy) +static int dsig_send_ctl(ErtsDSigData* dsdp, Eterm ctl, int force_busy) { + struct erts_dsig_send_context ctx; + int ret; + ctx.ctl = ctl; + ctx.msg = THE_NON_VALUE; + ctx.force_busy = force_busy; + ctx.phase = ERTS_DSIG_SEND_PHASE_INIT; +#ifdef DEBUG + ctx.reds = 1; /* provoke assert below (no reduction count without msg) */ +#endif + ret = erts_dsig_send(dsdp, &ctx); + ASSERT(ret != ERTS_DSIG_SEND_CONTINUE); + return ret; +} + +int +erts_dsig_send(ErtsDSigData *dsdp, struct erts_dsig_send_context* ctx) +{ + int retval; + Sint initial_reds = ctx->reds; Eterm cid; - int suspended = 0; - int resume = 0; - Uint32 pass_through_size; - Uint data_size, dhdr_ext_size; - ErtsAtomCacheMap *acmp; - ErtsDistOutputBuf *obuf; - DistEntry *dep = dsdp->dep; - Uint32 flags = dep->flags; - Process *c_p = dsdp->proc; - if (!c_p || dsdp->no_suspend) - force_busy = 1; + while (1) { + switch (ctx->phase) { + case ERTS_DSIG_SEND_PHASE_INIT: + ctx->flags = dsdp->dep->flags; + ctx->c_p = dsdp->proc; - ERTS_SMP_LC_ASSERT(!c_p - || (ERTS_PROC_LOCK_MAIN - == erts_proc_lc_my_proc_locks(c_p))); + if (!ctx->c_p || dsdp->no_suspend) + ctx->force_busy = 1; - if (!erts_is_alive) - return ERTS_DSIG_SEND_OK; + ERTS_SMP_LC_ASSERT(!ctx->c_p + || (ERTS_PROC_LOCK_MAIN + == erts_proc_lc_my_proc_locks(ctx->c_p))); - if (flags & DFLAG_DIST_HDR_ATOM_CACHE) { - acmp = erts_get_atom_cache_map(c_p); - pass_through_size = 0; - } - else { - acmp = NULL; - pass_through_size = 1; - } + if (!erts_is_alive) + return ERTS_DSIG_SEND_OK; -#ifdef ERTS_DIST_MSG_DBG - erts_fprintf(stderr, ">>%s CTL: %T\n", pass_through_size ? "P" : " ", ctl); - if (is_value(msg)) - erts_fprintf(stderr, " MSG: %T\n", msg); -#endif + if (ctx->flags & DFLAG_DIST_HDR_ATOM_CACHE) { + ctx->acmp = erts_get_atom_cache_map(ctx->c_p); + ctx->pass_through_size = 0; + } + else { + ctx->acmp = NULL; + ctx->pass_through_size = 1; + } - data_size = pass_through_size; - erts_reset_atom_cache_map(acmp); - data_size += erts_encode_dist_ext_size(ctl, flags, acmp); - if (is_value(msg)) - data_size += erts_encode_dist_ext_size(msg, flags, acmp); - erts_finalize_atom_cache_map(acmp, flags); + #ifdef ERTS_DIST_MSG_DBG + erts_fprintf(stderr, ">>%s CTL: %T\n", ctx->pass_through_size ? "P" : " ", ctx->ctl); + if (is_value(msg)) + erts_fprintf(stderr, " MSG: %T\n", msg); + #endif + + ctx->data_size = ctx->pass_through_size; + erts_reset_atom_cache_map(ctx->acmp); + erts_encode_dist_ext_size(ctx->ctl, ctx->flags, ctx->acmp, &ctx->data_size); + + if (is_value(ctx->msg)) { + ctx->u.sc.estack.start = NULL; + ctx->u.sc.flags = ctx->flags; + ctx->u.sc.level = 0; + ctx->phase = ERTS_DSIG_SEND_PHASE_MSG_SIZE; + } else { + ctx->phase = ERTS_DSIG_SEND_PHASE_ALLOC; + } + break; - dhdr_ext_size = erts_encode_ext_dist_header_size(acmp); - data_size += dhdr_ext_size; + case ERTS_DSIG_SEND_PHASE_MSG_SIZE: + if (erts_encode_dist_ext_size_int(ctx->msg, ctx, &ctx->data_size)) { + retval = ERTS_DSIG_SEND_CONTINUE; + goto done; + } - obuf = alloc_dist_obuf(data_size); - obuf->ext_endp = &obuf->data[0] + pass_through_size + dhdr_ext_size; + ctx->phase = ERTS_DSIG_SEND_PHASE_ALLOC; + case ERTS_DSIG_SEND_PHASE_ALLOC: + erts_finalize_atom_cache_map(ctx->acmp, ctx->flags); + + ctx->dhdr_ext_size = erts_encode_ext_dist_header_size(ctx->acmp); + ctx->data_size += ctx->dhdr_ext_size; + + ctx->obuf = alloc_dist_obuf(ctx->data_size); + ctx->obuf->ext_endp = &ctx->obuf->data[0] + ctx->pass_through_size + ctx->dhdr_ext_size; + + /* Encode internal version of dist header */ + ctx->obuf->extp = erts_encode_ext_dist_header_setup(ctx->obuf->ext_endp, ctx->acmp); + /* Encode control message */ + erts_encode_dist_ext(ctx->ctl, &ctx->obuf->ext_endp, ctx->flags, ctx->acmp, NULL, NULL); + if (is_value(ctx->msg)) { + ctx->u.ec.flags = ctx->flags; + ctx->u.ec.level = 0; + ctx->u.ec.wstack.wstart = NULL; + ctx->phase = ERTS_DSIG_SEND_PHASE_MSG_ENCODE; + } else { + ctx->phase = ERTS_DSIG_SEND_PHASE_FIN; + } + break; - /* Encode internal version of dist header */ - obuf->extp = erts_encode_ext_dist_header_setup(obuf->ext_endp, acmp); - /* Encode control message */ - erts_encode_dist_ext(ctl, &obuf->ext_endp, flags, acmp); - if (is_value(msg)) { - /* Encode message */ - erts_encode_dist_ext(msg, &obuf->ext_endp, flags, acmp); - } + case ERTS_DSIG_SEND_PHASE_MSG_ENCODE: + if (erts_encode_dist_ext(ctx->msg, &ctx->obuf->ext_endp, ctx->flags, ctx->acmp, &ctx->u.ec, &ctx->reds)) { + retval = ERTS_DSIG_SEND_CONTINUE; + goto done; + } - ASSERT(obuf->extp < obuf->ext_endp); - ASSERT(&obuf->data[0] <= obuf->extp - pass_through_size); - ASSERT(obuf->ext_endp <= &obuf->data[0] + data_size); + ctx->phase = ERTS_DSIG_SEND_PHASE_FIN; + case ERTS_DSIG_SEND_PHASE_FIN: { + DistEntry *dep = dsdp->dep; + int suspended = 0; + int resume = 0; - data_size = obuf->ext_endp - obuf->extp; + ASSERT(ctx->obuf->extp < ctx->obuf->ext_endp); + ASSERT(&ctx->obuf->data[0] <= ctx->obuf->extp - ctx->pass_through_size); + ASSERT(ctx->obuf->ext_endp <= &ctx->obuf->data[0] + ctx->data_size); - /* - * Signal encoded; now verify that the connection still exists, - * and if so enqueue the signal and schedule it for send. - */ - obuf->next = NULL; - erts_smp_de_rlock(dep); - cid = dep->cid; - if (cid != dsdp->cid - || dep->connection_id != dsdp->connection_id - || dep->status & ERTS_DE_SFLG_EXITING) { - /* Not the same connection as when we started; drop message... */ - erts_smp_de_runlock(dep); - free_dist_obuf(obuf); - } - else { - ErtsProcList *plp = NULL; - erts_smp_mtx_lock(&dep->qlock); - dep->qsize += size_obuf(obuf); - if (dep->qsize >= erts_dist_buf_busy_limit) - dep->qflgs |= ERTS_DE_QFLG_BUSY; - if (!force_busy && (dep->qflgs & ERTS_DE_QFLG_BUSY)) { - erts_smp_mtx_unlock(&dep->qlock); + ctx->data_size = ctx->obuf->ext_endp - ctx->obuf->extp; - plp = erts_proclist_create(c_p); - erts_suspend(c_p, ERTS_PROC_LOCK_MAIN, NULL); - suspended = 1; - erts_smp_mtx_lock(&dep->qlock); - } + /* + * Signal encoded; now verify that the connection still exists, + * and if so enqueue the signal and schedule it for send. + */ + ctx->obuf->next = NULL; + erts_smp_de_rlock(dep); + cid = dep->cid; + if (cid != dsdp->cid + || dep->connection_id != dsdp->connection_id + || dep->status & ERTS_DE_SFLG_EXITING) { + /* Not the same connection as when we started; drop message... */ + erts_smp_de_runlock(dep); + free_dist_obuf(ctx->obuf); + } + else { + ErtsProcList *plp = NULL; + erts_smp_mtx_lock(&dep->qlock); + dep->qsize += size_obuf(ctx->obuf); + if (dep->qsize >= erts_dist_buf_busy_limit) + dep->qflgs |= ERTS_DE_QFLG_BUSY; + if (!ctx->force_busy && (dep->qflgs & ERTS_DE_QFLG_BUSY)) { + erts_smp_mtx_unlock(&dep->qlock); + + plp = erts_proclist_create(ctx->c_p); + erts_suspend(ctx->c_p, ERTS_PROC_LOCK_MAIN, NULL); + suspended = 1; + erts_smp_mtx_lock(&dep->qlock); + } - /* Enqueue obuf on dist entry */ - if (dep->out_queue.last) - dep->out_queue.last->next = obuf; - else - dep->out_queue.first = obuf; - dep->out_queue.last = obuf; + /* Enqueue obuf on dist entry */ + if (dep->out_queue.last) + dep->out_queue.last->next = ctx->obuf; + else + dep->out_queue.first = ctx->obuf; + dep->out_queue.last = ctx->obuf; + + if (!ctx->force_busy) { + if (!(dep->qflgs & ERTS_DE_QFLG_BUSY)) { + if (suspended) + resume = 1; /* was busy when we started, but isn't now */ + #ifdef USE_VM_PROBES + if (resume && DTRACE_ENABLED(dist_port_not_busy)) { + DTRACE_CHARBUF(port_str, 64); + DTRACE_CHARBUF(remote_str, 64); + + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), + "%T", cid); + erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), + "%T", dep->sysname); + DTRACE3(dist_port_not_busy, erts_this_node_sysname, + port_str, remote_str); + } + #endif + } + else { + /* Enqueue suspended process on dist entry */ + ASSERT(plp); + erts_proclist_store_last(&dep->suspended, plp); + } + } - if (!force_busy) { - if (!(dep->qflgs & ERTS_DE_QFLG_BUSY)) { - if (suspended) - resume = 1; /* was busy when we started, but isn't now */ -#ifdef USE_VM_PROBES - if (resume && DTRACE_ENABLED(dist_port_not_busy)) { - DTRACE_CHARBUF(port_str, 64); - DTRACE_CHARBUF(remote_str, 64); - - erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), - "%T", cid); - erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), - "%T", dep->sysname); - DTRACE3(dist_port_not_busy, erts_this_node_sysname, - port_str, remote_str); - } -#endif + erts_smp_mtx_unlock(&dep->qlock); + erts_schedule_dist_command(NULL, dep); + erts_smp_de_runlock(dep); + + if (resume) { + erts_resume(ctx->c_p, ERTS_PROC_LOCK_MAIN); + erts_proclist_destroy(plp); + /* + * Note that the calling process still have to yield as if it + * suspended. If not, the calling process could later be + * erroneously scheduled when it shouldn't be. + */ + } } - else { - /* Enqueue suspended process on dist entry */ - ASSERT(plp); - erts_proclist_store_last(&dep->suspended, plp); + ctx->obuf = NULL; + + if (suspended) { + #ifdef USE_VM_PROBES + if (!resume && DTRACE_ENABLED(dist_port_busy)) { + DTRACE_CHARBUF(port_str, 64); + DTRACE_CHARBUF(remote_str, 64); + DTRACE_CHARBUF(pid_str, 16); + + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", cid); + erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), + "%T", dep->sysname); + erts_snprintf(pid_str, sizeof(DTRACE_CHARBUF_NAME(pid_str)), + "%T", ctx->c_p->common.id); + DTRACE4(dist_port_busy, erts_this_node_sysname, + port_str, remote_str, pid_str); + } + #endif + if (!resume && erts_system_monitor_flags.busy_dist_port) + monitor_generic(ctx->c_p, am_busy_dist_port, cid); + retval = ERTS_DSIG_SEND_YIELD; + } else { + retval = ERTS_DSIG_SEND_OK; } + goto done; } - - erts_smp_mtx_unlock(&dep->qlock); - erts_schedule_dist_command(NULL, dep); - erts_smp_de_runlock(dep); - - if (resume) { - erts_resume(c_p, ERTS_PROC_LOCK_MAIN); - erts_proclist_destroy(plp); - /* - * Note that the calling process still have to yield as if it - * suspended. If not, the calling process could later be - * erroneously scheduled when it shouldn't be. - */ + default: + erl_exit(ERTS_ABORT_EXIT, "dsig_send invalid phase (%d)\n", (int)ctx->phase); } } - if (c_p) { - int reds; - /* - * Bump reductions on calling process. - * - * This is the reduction cost: Always a base cost of 8 reductions - * plus 16 reductions per kilobyte generated external data. - */ - - data_size >>= (10-4); -#if defined(ARCH_64) && !HALFWORD_HEAP - data_size &= 0x003fffffffffffff; -#elif defined(ARCH_32) || HALFWORD_HEAP - data_size &= 0x003fffff; -#else -# error "Ohh come on ... !?!" -#endif - reds = 8 + ((int) data_size > 1000000 ? 1000000 : (int) data_size); - BUMP_REDS(c_p, reds); - } - - if (suspended) { -#ifdef USE_VM_PROBES - if (!resume && DTRACE_ENABLED(dist_port_busy)) { - DTRACE_CHARBUF(port_str, 64); - DTRACE_CHARBUF(remote_str, 64); - DTRACE_CHARBUF(pid_str, 16); - - erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", cid); - erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), - "%T", dep->sysname); - erts_snprintf(pid_str, sizeof(DTRACE_CHARBUF_NAME(pid_str)), - "%T", c_p->common.id); - DTRACE4(dist_port_busy, erts_this_node_sysname, - port_str, remote_str, pid_str); - } -#endif - if (!resume && erts_system_monitor_flags.busy_dist_port) - monitor_generic(c_p, am_busy_dist_port, cid); - return ERTS_DSIG_SEND_YIELD; +done: + if (ctx->msg && ctx->c_p) { + BUMP_REDS(ctx->c_p, (initial_reds - ctx->reds) / TERM_TO_BINARY_LOOP_FACTOR); } - return ERTS_DSIG_SEND_OK; + return retval; } - static Uint dist_port_command(Port *prt, ErtsDistOutputBuf *obuf) { diff --git a/erts/emulator/beam/dist.h b/erts/emulator/beam/dist.h index f32b999198..2a2ba0c83f 100644 --- a/erts/emulator/beam/dist.h +++ b/erts/emulator/beam/dist.h @@ -22,6 +22,7 @@ #include "erl_process.h" #include "erl_node_tables.h" +#include "zlib.h" #define DFLAG_PUBLISHED 0x01 #define DFLAG_ATOM_CACHE 0x02 @@ -264,17 +265,105 @@ erts_destroy_dist_link(ErtsDistLinkData *dldp) #endif + + +/* Define for testing */ +/* #define EXTREME_TTB_TRAPPING 1 */ + +#ifndef EXTREME_TTB_TRAPPING +#define TERM_TO_BINARY_LOOP_FACTOR 32 +#else +#define TERM_TO_BINARY_LOOP_FACTOR 1 +#endif + +typedef enum { TTBSize, TTBEncode, TTBCompress } TTBState; +typedef struct TTBSizeContext_ { + Uint flags; + int level; + Uint result; + Eterm obj; + ErtsEStack estack; +} TTBSizeContext; + +typedef struct TTBEncodeContext_ { + Uint flags; + int level; + byte* ep; + Eterm obj; + ErtsWStack wstack; + Binary *result_bin; +} TTBEncodeContext; + +typedef struct { + Uint real_size; + Uint dest_len; + byte *dbytes; + Binary *result_bin; + Binary *destination_bin; + z_stream stream; +} TTBCompressContext; + +typedef struct { + int alive; + TTBState state; + union { + TTBSizeContext sc; + TTBEncodeContext ec; + TTBCompressContext cc; + } s; +} TTBContext; + +enum erts_dsig_send_phase { + ERTS_DSIG_SEND_PHASE_INIT, + ERTS_DSIG_SEND_PHASE_MSG_SIZE, + ERTS_DSIG_SEND_PHASE_ALLOC, + ERTS_DSIG_SEND_PHASE_MSG_ENCODE, + ERTS_DSIG_SEND_PHASE_FIN +}; + +struct erts_dsig_send_context { + enum erts_dsig_send_phase phase; + Sint reds; + + Eterm ctl; + Eterm msg; + int force_busy; + Uint32 pass_through_size; + Uint data_size, dhdr_ext_size; + ErtsAtomCacheMap *acmp; + ErtsDistOutputBuf *obuf; + Uint32 flags; + Process *c_p; + union { + TTBSizeContext sc; + TTBEncodeContext ec; + }u; +}; + +typedef struct { + int suspend; + + Eterm ctl_heap[6]; + ErtsDSigData dsd; + DistEntry* dep_to_deref; + struct erts_dsig_send_context dss; + + Eterm return_term; +}ErtsSendContext; + + /* * erts_dsig_send_* return values. */ #define ERTS_DSIG_SEND_OK 0 #define ERTS_DSIG_SEND_YIELD 1 +#define ERTS_DSIG_SEND_CONTINUE 2 extern int erts_dsig_send_link(ErtsDSigData *, Eterm, Eterm); -extern int erts_dsig_send_msg(ErtsDSigData *, Eterm, Eterm); +extern int erts_dsig_send_msg(Eterm, Eterm, ErtsSendContext*); extern int erts_dsig_send_exit_tt(ErtsDSigData *, Eterm, Eterm, Eterm, Eterm); extern int erts_dsig_send_unlink(ErtsDSigData *, Eterm, Eterm); -extern int erts_dsig_send_reg_msg(ErtsDSigData *, Eterm, Eterm); +extern int erts_dsig_send_reg_msg(Eterm, Eterm, ErtsSendContext*); extern int erts_dsig_send_group_leader(ErtsDSigData *, Eterm, Eterm); extern int erts_dsig_send_exit(ErtsDSigData *, Eterm, Eterm, Eterm); extern int erts_dsig_send_exit2(ErtsDSigData *, Eterm, Eterm, Eterm); @@ -282,6 +371,10 @@ extern int erts_dsig_send_demonitor(ErtsDSigData *, Eterm, Eterm, Eterm, int); extern int erts_dsig_send_monitor(ErtsDSigData *, Eterm, Eterm, Eterm); extern int erts_dsig_send_m_exit(ErtsDSigData *, Eterm, Eterm, Eterm, Eterm); +extern int erts_dsig_send(ErtsDSigData *dsdp, struct erts_dsig_send_context* ctx); +extern void erts_dsend_context_dtor(Binary*); +extern Eterm erts_dsend_export_trap_context(Process* p, ErtsSendContext* ctx); + extern int erts_dist_command(Port *prt, int reds); extern void erts_dist_port_not_busy(Port *prt); extern void erts_kill_dist_connection(DistEntry *dep, Uint32); diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c index 05ac24e04d..f2bceff4eb 100644 --- a/erts/emulator/beam/erl_alloc.c +++ b/erts/emulator/beam/erl_alloc.c @@ -1873,8 +1873,8 @@ erts_alc_fatal_error(int error, int func, ErtsAlcType_t n, ...) size = va_arg(argp, Uint); va_end(argp); erl_exit(1, - "%s: Cannot %s %lu bytes of memory (of type \"%s\").\n", - allctr_str, op, size, t_str); + "%s: Cannot %s %lu bytes of memory (of type \"%s\", thread %d).\n", + allctr_str, op, size, t_str, ERTS_ALC_GET_THR_IX()); break; } case ERTS_ALC_E_NOALLCTR: @@ -3939,7 +3939,7 @@ static Uint install_debug_functions(void) { int i; - ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); + ERTS_CT_ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); sys_memcpy((void *)real_allctrs,(void *)erts_allctrs,sizeof(erts_allctrs)); diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types index 17ac6316b7..e2f8da38b9 100644 --- a/erts/emulator/beam/erl_alloc.types +++ b/erts/emulator/beam/erl_alloc.types @@ -267,9 +267,9 @@ type CODE_IX_LOCK_Q SHORT_LIVED SYSTEM code_ix_lock_q type PROC_INTERVAL LONG_LIVED SYSTEM process_interval type BUSY_CALLER_TAB SHORT_LIVED SYSTEM busy_caller_table type BUSY_CALLER SHORT_LIVED SYSTEM busy_caller -type PORT_DATA_HEAP STANDARD SYSTEM port_data_heap type PROC_SYS_TSK SHORT_LIVED PROCESSES proc_sys_task type PROC_SYS_TSK_QS SHORT_LIVED PROCESSES proc_sys_task_queues +type NEW_TIME_OFFSET SHORT_LIVED SYSTEM new_time_offset +if threads_no_smp # Need thread safe allocs, but std_alloc and fix_alloc are not; @@ -357,12 +357,15 @@ type DB_MS_PSDO_PROC LONG_LIVED_LOW ETS db_match_pseudo_proc type SCHDLR_DATA LONG_LIVED_LOW SYSTEM scheduler_data type LL_TEMP_TERM LONG_LIVED_LOW SYSTEM ll_temp_term +type NIF_TRAP_EXPORT STANDARD_LOW CODE nif_trap_export_entry type EXPORT LONG_LIVED_LOW CODE export_entry type MONITOR_SH STANDARD_LOW PROCESSES monitor_sh type NLINK_SH STANDARD_LOW PROCESSES nlink_sh type AINFO_REQ STANDARD_LOW SYSTEM alloc_info_request type SCHED_WTIME_REQ STANDARD_LOW SYSTEM sched_wall_time_request type GC_INFO_REQ STANDARD_LOW SYSTEM gc_info_request +type PORT_DATA_HEAP STANDARD_LOW SYSTEM port_data_heap +type BIF_TIMER_DATA LONG_LIVED_LOW SYSTEM bif_timer_data +else # "fullword" @@ -375,12 +378,15 @@ type DB_MS_PSDO_PROC LONG_LIVED ETS db_match_pseudo_proc type SCHDLR_DATA LONG_LIVED SYSTEM scheduler_data type LL_TEMP_TERM LONG_LIVED SYSTEM ll_temp_term +type NIF_TRAP_EXPORT STANDARD CODE nif_trap_export_entry type EXPORT LONG_LIVED CODE export_entry type MONITOR_SH FIXED_SIZE PROCESSES monitor_sh type NLINK_SH FIXED_SIZE PROCESSES nlink_sh type AINFO_REQ SHORT_LIVED SYSTEM alloc_info_request type SCHED_WTIME_REQ SHORT_LIVED SYSTEM sched_wall_time_request type GC_INFO_REQ SHORT_LIVED SYSTEM gc_info_request +type PORT_DATA_HEAP STANDARD SYSTEM port_data_heap +type BIF_TIMER_DATA LONG_LIVED SYSTEM bif_timer_data +endif @@ -395,6 +401,7 @@ type DRV_EV_STATE LONG_LIVED SYSTEM driver_event_state type DRV_EV_D_STATE FIXED_SIZE SYSTEM driver_event_data_state type DRV_SEL_D_STATE FIXED_SIZE SYSTEM driver_select_data_state type FD_LIST SHORT_LIVED SYSTEM fd_list +type ACTIVE_FD_ARR SHORT_LIVED SYSTEM active_fd_array type POLLSET LONG_LIVED SYSTEM pollset type POLLSET_UPDREQ SHORT_LIVED SYSTEM pollset_update_req type POLL_FDS LONG_LIVED SYSTEM poll_fds @@ -411,6 +418,8 @@ type CS_PROG_PATH LONG_LIVED SYSTEM cs_prog_path type ENVIRONMENT TEMPORARY SYSTEM environment type PUTENV_STR SYSTEM SYSTEM putenv_string type PRT_REP_EXIT STANDARD SYSTEM port_report_exit +type SYS_BLOCKING STANDARD SYSTEM sys_blocking +type SYS_WRITE_BUF TEMPORARY SYSTEM sys_write_buf +endif diff --git a/erts/emulator/beam/erl_alloc_util.c b/erts/emulator/beam/erl_alloc_util.c index 45f0cc4312..2f277690e4 100644 --- a/erts/emulator/beam/erl_alloc_util.c +++ b/erts/emulator/beam/erl_alloc_util.c @@ -205,7 +205,7 @@ MBC after deallocating first block: ASSERT(((UWord)(F) & (~FLG_MASK|THIS_FREE_BLK_HDR_FLG|PREV_FREE_BLK_HDR_FLG)) == THIS_FREE_BLK_HDR_FLG), \ (B)->bhdr = ((Sz) | (F)), \ (B)->u.carrier = (C)) - + # define IS_MBC_FIRST_ABLK(AP,B) \ ((((UWord)(B) & ~ERTS_SACRR_UNIT_MASK) == MBC_HEADER_SIZE(AP)) \ && ((B)->bhdr & MBC_ABLK_OFFSET_MASK) == 0) @@ -378,9 +378,8 @@ do { \ #ifdef ERTS_SMP #define SBC_HEADER_SIZE \ - (UNIT_CEILING(sizeof(Carrier_t) \ - - sizeof(ErtsAlcCPoolData_t) \ - + ABLK_HDR_SZ) \ + (UNIT_CEILING(offsetof(Carrier_t, cpool) \ + + ABLK_HDR_SZ) \ - ABLK_HDR_SZ) #else #define SBC_HEADER_SIZE \ @@ -929,6 +928,88 @@ unlink_carrier(CarrierList_t *cl, Carrier_t *crr) #ifdef ERTS_SMP +#ifdef DEBUG +static int is_in_list(ErtsDoubleLink_t* sentinel, ErtsDoubleLink_t* node) +{ + ErtsDoubleLink_t* p; + + ASSERT(node != sentinel); + for (p = sentinel->next; p != sentinel; p = p->next) { + if (p == node) + return 1; + } + return 0; +} +#endif /* DEBUG */ + +static ERTS_INLINE void +link_edl_after(ErtsDoubleLink_t* after_me, ErtsDoubleLink_t* node) +{ + ErtsDoubleLink_t* before_me = after_me->next; + ASSERT(node != after_me && node != before_me); + node->next = before_me; + node->prev = after_me; + before_me->prev = node; + after_me->next = node; +} + +static ERTS_INLINE void +link_edl_before(ErtsDoubleLink_t* before_me, ErtsDoubleLink_t* node) +{ + ErtsDoubleLink_t* after_me = before_me->prev; + ASSERT(node != before_me && node != after_me); + node->next = before_me; + node->prev = after_me; + before_me->prev = node; + after_me->next = node; +} + +static ERTS_INLINE void +unlink_edl(ErtsDoubleLink_t* node) +{ + node->next->prev = node->prev; + node->prev->next = node->next; +} + +static ERTS_INLINE void +relink_edl_before(ErtsDoubleLink_t* before_me, ErtsDoubleLink_t* node) +{ + if (node != before_me && node != before_me->prev) { + unlink_edl(node); + link_edl_before(before_me, node); + } +} + +static ERTS_INLINE int is_abandoned(Carrier_t *crr) +{ + return crr->cpool.abandoned.next != NULL; +} + +static ERTS_INLINE void +link_abandoned_carrier(ErtsDoubleLink_t* list, Carrier_t *crr) +{ + ASSERT(!is_abandoned(crr)); + + link_edl_after(list, &crr->cpool.abandoned); + + ASSERT(crr->cpool.abandoned.next != &crr->cpool.abandoned); + ASSERT(crr->cpool.abandoned.prev != &crr->cpool.abandoned); +} + +static ERTS_INLINE void +unlink_abandoned_carrier(Carrier_t *crr) +{ + ASSERT(is_in_list(&crr->cpool.orig_allctr->cpool.pooled_list, + &crr->cpool.abandoned) || + is_in_list(&crr->cpool.orig_allctr->cpool.traitor_list, + &crr->cpool.abandoned)); + + unlink_edl(&crr->cpool.abandoned); + + crr->cpool.abandoned.next = NULL; + crr->cpool.abandoned.prev = NULL; +} + static ERTS_INLINE void clear_busy_pool_carrier(Allctr_t *allctr, Carrier_t *crr) { @@ -955,7 +1036,7 @@ clear_busy_pool_carrier(Allctr_t *allctr, Carrier_t *crr) } } -#endif +#endif /* ERTS_SMP */ #if 0 #define ERTS_DBG_CHK_FIX_LIST(A, FIX, IX, B) \ @@ -1361,7 +1442,7 @@ get_pref_allctr(void *extra) pref_ix = ERTS_ALC_GET_THR_IX(); - ASSERT(sizeof(UWord) == sizeof(Allctr_t *)); + ERTS_CT_ASSERT(sizeof(UWord) == sizeof(Allctr_t *)); ASSERT(0 <= pref_ix && pref_ix < tspec->size); return tspec->allctr[pref_ix]; @@ -1775,6 +1856,18 @@ handle_delayed_dealloc(Allctr_t *allctr, * data has been overwritten by the queue. */ Carrier_t *crr = FIRST_BLK_TO_MBC(allctr, blk); + + /* Restore word overwritten by the dd-queue as it will be read + * if this carrier is pulled from dc_list by cpool_fetch() + */ + ERTS_ALC_CPOOL_ASSERT(FBLK_TO_MBC(blk) != crr); + ERTS_CT_ASSERT(sizeof(ErtsAllctrDDBlock_t) == sizeof(void*)); +#ifdef MBC_ABLK_OFFSET_BITS + blk->u.carrier = crr; +#else + blk->carrier = crr; +#endif + ERTS_ALC_CPOOL_ASSERT(ERTS_ALC_IS_CPOOL_ENABLED(allctr)); ERTS_ALC_CPOOL_ASSERT(allctr == crr->cpool.orig_allctr); ERTS_ALC_CPOOL_ASSERT(((erts_aint_t) allctr) @@ -2563,10 +2656,9 @@ mbc_realloc(Allctr_t *allctr, void *p, Uint size, Uint32 alcu_flgs, #ifdef ERTS_SMP #define ERTS_ALC_MAX_DEALLOC_CARRIER 10 -#define ERTS_ALC_CPOOL_MAX_FETCH_INSPECT 10 +#define ERTS_ALC_CPOOL_MAX_FETCH_INSPECT 20 +#define ERTS_ALC_CPOOL_MAX_TRAITOR_INSPECT 10 #define ERTS_ALC_CPOOL_CHECK_LIMIT_COUNT 100 -#define ERTS_ALC_CPOOL_MAX_NO_CARRIERS 5 -#define ERTS_ALC_CPOOL_INSERT_ALLOWED_OFFSET 100 #define ERTS_ALC_CPOOL_MAX_FAILED_STAT_READS 3 #define ERTS_ALC_CPOOL_PTR_MOD_MRK (((erts_aint_t) 1) << 0) @@ -2743,9 +2835,6 @@ cpool_insert(Allctr_t *allctr, Carrier_t *crr) (erts_aint_t) CARRIER_SZ(crr)); erts_atomic_inc_nob(&allctr->cpool.stat.no_carriers); - erts_smp_atomic_set_nob(&crr->allctr, - ((erts_aint_t) allctr)|ERTS_CRR_ALCTR_FLG_IN_POOL); - /* * We search in 'next' direction and begin by passing * one element before trying to insert. This in order to @@ -2804,6 +2893,9 @@ cpool_insert(Allctr_t *allctr, Carrier_t *crr) cpool_set_mod_marked(&cpd2p->prev, (erts_aint_t) &crr->cpool, (erts_aint_t) cpd1p); + + erts_smp_atomic_set_wb(&crr->allctr, + ((erts_aint_t) allctr)|ERTS_CRR_ALCTR_FLG_IN_POOL); } static void @@ -2904,59 +2996,163 @@ cpool_delete(Allctr_t *allctr, Allctr_t *prev_allctr, Carrier_t *crr) static Carrier_t * cpool_fetch(Allctr_t *allctr, UWord size) { - int i; + int i, i_stop, has_passed_sentinel; Carrier_t *crr; ErtsAlcCPoolData_t *cpdp; - ErtsAlcCPoolData_t *sentinel = &carrier_pool[allctr->alloc_no].sentinel; + ErtsAlcCPoolData_t *cpool_entrance; + ErtsAlcCPoolData_t *sentinel; + ErtsDoubleLink_t* dl; + ErtsDoubleLink_t* first_old_traitor; ERTS_ALC_CPOOL_ASSERT(allctr->alloc_no == ERTS_ALC_A_INVALID /* testcase */ || erts_thr_progress_is_managed_thread()); - i = 0; + i = ERTS_ALC_CPOOL_MAX_FETCH_INSPECT; + first_old_traitor = allctr->cpool.traitor_list.next; + cpool_entrance = NULL; - /* First; check our own pending dealloc carrier list... */ - crr = allctr->cpool.dc_list.last; - while (crr && i < ERTS_ALC_CPOOL_MAX_FETCH_INSPECT) { - if (erts_atomic_read_nob(&crr->cpool.max_size) >= size) { - unlink_carrier(&allctr->cpool.dc_list, crr); -#ifdef ERTS_ALC_CPOOL_DEBUG - ERTS_ALC_CPOOL_ASSERT(erts_smp_atomic_xchg_nob(&crr->allctr, - ((erts_aint_t) allctr)) - == (((erts_aint_t) allctr) & ~ERTS_CRR_ALCTR_FLG_MASK)); -#else - erts_smp_atomic_set_nob(&crr->allctr, ((erts_aint_t) allctr)); -#endif - return crr; + /* + * Search my own pooled_list, + * i.e my abandoned carriers that were in the pool last time I checked. + */ + + dl = allctr->cpool.pooled_list.next; + while(dl != &allctr->cpool.pooled_list) { + erts_aint_t exp, act; + crr = (Carrier_t *) (((char *) dl) - offsetof(Carrier_t, cpool.abandoned)); + + ASSERT(!is_in_list(&allctr->cpool.traitor_list, dl)); + ASSERT(crr->cpool.orig_allctr == allctr); + dl = dl->next; + exp = erts_smp_atomic_read_rb(&crr->allctr); + if ((exp & ERTS_CRR_ALCTR_FLG_MASK) == ERTS_CRR_ALCTR_FLG_IN_POOL + && erts_atomic_read_nob(&crr->cpool.max_size) >= size) { + /* Try to fetch it... */ + act = erts_smp_atomic_cmpxchg_mb(&crr->allctr, + (erts_aint_t) allctr, + exp); + if (act == exp) { + cpool_delete(allctr, ((Allctr_t *) (act & ~ERTS_CRR_ALCTR_FLG_MASK)), crr); + unlink_abandoned_carrier(crr); + + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.pooled_list); + return crr; + } + exp = act; + } + if (exp & ERTS_CRR_ALCTR_FLG_IN_POOL) { + if (!cpool_entrance) + cpool_entrance = &crr->cpool; + } + else { /* Not in pool, move to traitor_list */ + unlink_abandoned_carrier(crr); + link_abandoned_carrier(&allctr->cpool.traitor_list, crr); + } + if (--i <= 0) { + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.pooled_list); + return NULL; } - crr = crr->prev; - i++; } - /* ... then the pool ... */ + /* Now search traitor_list. + * i.e carriers employed by other allocators last time I checked. + * They might have been abandoned since then. + */ + + i_stop = (i < ERTS_ALC_CPOOL_MAX_TRAITOR_INSPECT ? + 0 : i - ERTS_ALC_CPOOL_MAX_TRAITOR_INSPECT); + dl = first_old_traitor; + while(dl != &allctr->cpool.traitor_list) { + erts_aint_t exp, act; + crr = (Carrier_t *) (((char *) dl) - offsetof(Carrier_t, cpool.abandoned)); + ASSERT(dl != &allctr->cpool.pooled_list); + ASSERT(crr->cpool.orig_allctr == allctr); + dl = dl->next; + exp = erts_smp_atomic_read_rb(&crr->allctr); + if (exp & ERTS_CRR_ALCTR_FLG_IN_POOL) { + if (!(exp & ERTS_CRR_ALCTR_FLG_BUSY) + && erts_atomic_read_nob(&crr->cpool.max_size) >= size) { + /* Try to fetch it... */ + act = erts_smp_atomic_cmpxchg_mb(&crr->allctr, + (erts_aint_t) allctr, + exp); + if (act == exp) { + cpool_delete(allctr, ((Allctr_t *) (act & ~ERTS_CRR_ALCTR_FLG_MASK)), crr); + unlink_abandoned_carrier(crr); + + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.traitor_list); + return crr; + } + exp = act; + } + if (exp & ERTS_CRR_ALCTR_FLG_IN_POOL) { + if (!cpool_entrance) + cpool_entrance = &crr->cpool; + + /* Move to pooled_list */ + unlink_abandoned_carrier(crr); + link_abandoned_carrier(&allctr->cpool.pooled_list, crr); + } + } + if (--i <= i_stop) { + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.traitor_list); + if (i > 0) + break; + else + return NULL; + } + } /* - * We search in 'prev' direction and begin by passing - * one element before trying to fetch. This in order to - * avoid contention with threads inserting elements. + * Finally search the shared pool and try employ foreign carriers */ - cpdp = cpool_aint2cpd(cpool_read(&sentinel->prev)); - if (cpdp == sentinel) - return NULL; + sentinel = &carrier_pool[allctr->alloc_no].sentinel; + if (cpool_entrance) { + /* We saw a pooled carried above, use it as entrance into the pool + */ + cpdp = cpool_entrance; + } + else { + /* No pooled carried seen above. Start search at cpool sentinel, + * but begin by passing one element before trying to fetch. + * This in order to avoid contention with threads inserting elements. + */ + cpool_entrance = sentinel; + cpdp = cpool_aint2cpd(cpool_read(&cpool_entrance->prev)); + if (cpdp == sentinel) + return NULL; + } - while (i < ERTS_ALC_CPOOL_MAX_FETCH_INSPECT) { + has_passed_sentinel = 0; + while (1) { erts_aint_t exp; cpdp = cpool_aint2cpd(cpool_read(&cpdp->prev)); - if (cpdp == sentinel) { + if (cpdp == cpool_entrance) { + if (cpool_entrance == sentinel) { + cpdp = cpool_aint2cpd(cpool_read(&cpdp->prev)); + if (cpdp == sentinel) + return NULL; + } + i = 0; /* Last one to inspect */ + } + else if (cpdp == sentinel) { + if (has_passed_sentinel) { + /* We been here before. cpool_entrance must have been removed */ + return NULL; + } cpdp = cpool_aint2cpd(cpool_read(&cpdp->prev)); if (cpdp == sentinel) return NULL; - i = ERTS_ALC_CPOOL_MAX_FETCH_INSPECT; /* Last one to inspect */ + has_passed_sentinel = 1; } - crr = (Carrier_t *) (((char *) cpdp) - offsetof(Carrier_t, cpool)); + crr = (Carrier_t *)(((char *)cpdp) - offsetof(Carrier_t, cpool)); exp = erts_smp_atomic_read_rb(&crr->allctr); - if (((exp & (ERTS_CRR_ALCTR_FLG_IN_POOL|ERTS_CRR_ALCTR_FLG_BUSY)) - == ERTS_CRR_ALCTR_FLG_IN_POOL) + if (((exp & (ERTS_CRR_ALCTR_FLG_MASK)) == ERTS_CRR_ALCTR_FLG_IN_POOL) && (erts_atomic_read_nob(&cpdp->max_size) >= size)) { erts_aint_t act; /* Try to fetch it... */ @@ -2965,11 +3161,35 @@ cpool_fetch(Allctr_t *allctr, UWord size) exp); if (act == exp) { cpool_delete(allctr, ((Allctr_t *) (act & ~ERTS_CRR_ALCTR_FLG_MASK)), crr); + if (crr->cpool.orig_allctr == allctr) { + unlink_abandoned_carrier(crr); + } return crr; } } - i++; + if (--i <= 0) + return NULL; } + + /* Last; check our own pending dealloc carrier list... */ + crr = allctr->cpool.dc_list.last; + while (crr) { + if (erts_atomic_read_nob(&crr->cpool.max_size) >= size) { + unlink_carrier(&allctr->cpool.dc_list, crr); +#ifdef ERTS_ALC_CPOOL_DEBUG + ERTS_ALC_CPOOL_ASSERT(erts_smp_atomic_xchg_nob(&crr->allctr, + ((erts_aint_t) allctr)) + == (((erts_aint_t) allctr) & ~ERTS_CRR_ALCTR_FLG_MASK)); +#else + erts_smp_atomic_set_nob(&crr->allctr, ((erts_aint_t) allctr)); +#endif + return crr; + } + crr = crr->prev; + if (--i <= 0) + return NULL; + } + return NULL; } @@ -3066,6 +3286,9 @@ schedule_dealloc_carrier(Allctr_t *allctr, Carrier_t *crr) return; } + if (is_abandoned(crr)) + unlink_abandoned_carrier(crr); + if (crr->cpool.thr_prgr == ERTS_THR_PRGR_INVALID || erts_thr_progress_has_reached(crr->cpool.thr_prgr)) { dealloc_carrier(allctr, crr, 1); @@ -3112,6 +3335,8 @@ cpool_init_carrier_data(Allctr_t *allctr, Carrier_t *crr) limit = (csz/100)*allctr->cpool.util_limit; crr->cpool.abandon_limit = limit; } + crr->cpool.abandoned.next = NULL; + crr->cpool.abandoned.prev = NULL; } static void @@ -3142,6 +3367,9 @@ abandon_carrier(Allctr_t *allctr, Carrier_t *crr) STAT_MBC_CPOOL_INSERT(allctr, crr); unlink_carrier(&allctr->mbc_list, crr); + if (crr->cpool.orig_allctr == allctr) { + link_abandoned_carrier(&allctr->cpool.pooled_list, crr); + } allctr->remove_mbc(allctr, crr); @@ -3274,6 +3502,15 @@ create_carrier(Allctr_t *allctr, Uint umem_sz, UWord flags) ASSERT(!(flags & CFLG_FORCE_MSEG && flags & CFLG_FORCE_SYS_ALLOC)); + if (umem_sz > (ERTS_UINT_MAX - ERTS_UINT_MAX/100)) { + /* Do an overly conservative _overflow_ check here so we don't + * have to deal with it from here on. I guess we could be more accurate + * but I don't think the need to allocate over 99% of the address space + * will ever arise on any machine, neither 32 nor 64 bit. + */ + return NULL; + } + blk_sz = UMEMSZ2BLKSZ(allctr, umem_sz); #ifdef ERTS_SMP @@ -3640,6 +3877,11 @@ destroy_carrier(Allctr_t *allctr, Block_t *blk, Carrier_t **busy_pcrr_pp) if (busy_pcrr_pp && *busy_pcrr_pp) { ERTS_ALC_CPOOL_ASSERT(*busy_pcrr_pp == crr); *busy_pcrr_pp = NULL; + ERTS_ALC_CPOOL_ASSERT(erts_smp_atomic_read_nob(&crr->allctr) + == (((erts_aint_t) allctr) + | ERTS_CRR_ALCTR_FLG_IN_POOL + | ERTS_CRR_ALCTR_FLG_BUSY)); + erts_smp_atomic_set_nob(&crr->allctr, ((erts_aint_t) allctr)); cpool_delete(allctr, allctr, crr); } else @@ -5519,6 +5761,10 @@ erts_alcu_start(Allctr_t *allctr, AllctrInit_t *init) allctr->min_block_size = sz; } + allctr->cpool.pooled_list.next = &allctr->cpool.pooled_list; + allctr->cpool.pooled_list.prev = &allctr->cpool.pooled_list; + allctr->cpool.traitor_list.next = &allctr->cpool.traitor_list; + allctr->cpool.traitor_list.prev = &allctr->cpool.traitor_list; allctr->cpool.dc_list.first = NULL; allctr->cpool.dc_list.last = NULL; allctr->cpool.abandon_limit = 0; @@ -5696,7 +5942,7 @@ erts_alcu_init(AlcUInit_t *init) erts_atomic_init_nob(&sentinel->prev, (erts_aint_t) sentinel); } #endif - ASSERT(SBC_BLK_SZ_MASK == MBC_FBLK_SZ_MASK); /* see BLK_SZ */ + ERTS_CT_ASSERT(SBC_BLK_SZ_MASK == MBC_FBLK_SZ_MASK); /* see BLK_SZ */ #if HAVE_ERTS_MSEG ASSERT(erts_mseg_unit_size() == ERTS_SACRR_UNIT_SZ); max_mseg_carriers = init->mmc; diff --git a/erts/emulator/beam/erl_alloc_util.h b/erts/emulator/beam/erl_alloc_util.h index 7be6b1ed9d..eee920e66c 100644 --- a/erts/emulator/beam/erl_alloc_util.h +++ b/erts/emulator/beam/erl_alloc_util.h @@ -268,6 +268,11 @@ typedef union {char c[ERTS_ALLOC_ALIGN_BYTES]; long l; double d;} Unit_t; #ifdef ERTS_SMP +typedef struct ErtsDoubleLink_t_ { + struct ErtsDoubleLink_t_ *next; + struct ErtsDoubleLink_t_ *prev; +}ErtsDoubleLink_t; + typedef struct { erts_atomic_t next; erts_atomic_t prev; @@ -277,6 +282,7 @@ typedef struct { UWord abandon_limit; UWord blocks; UWord blocks_size; + ErtsDoubleLink_t abandoned; /* node in pooled_list or traitor_list */ } ErtsAlcCPoolData_t; #endif @@ -500,7 +506,12 @@ struct Allctr_t_ { CarrierList_t sbc_list; #ifdef ERTS_SMP struct { - CarrierList_t dc_list; + /* pooled_list, traitor list and dc_list contain only + carriers _created_ by this allocator */ + ErtsDoubleLink_t pooled_list; + ErtsDoubleLink_t traitor_list; + CarrierList_t dc_list; + UWord abandon_limit; int disable_abandon; int check_limit_count; diff --git a/erts/emulator/beam/erl_async.c b/erts/emulator/beam/erl_async.c index decae6b2ca..bc06d41720 100644 --- a/erts/emulator/beam/erl_async.c +++ b/erts/emulator/beam/erl_async.c @@ -176,7 +176,7 @@ erts_init_async(void) ErtsThrQInit_t qinit = ERTS_THR_Q_INIT_DEFAULT; #endif erts_thr_opts_t thr_opts = ERTS_THR_OPTS_DEFAULT_INITER; - char *ptr; + char *ptr, thr_name[16]; size_t tot_size = 0; int i; @@ -227,23 +227,16 @@ erts_init_async(void) thr_opts.suggested_stack_size = erts_async_thread_suggested_stack_size; -#ifdef ETHR_HAVE_THREAD_NAMES - thr_opts.name = malloc(sizeof(char)*(strlen("async_XXXX")+1)); -#endif + thr_opts.name = thr_name; for (i = 0; i < erts_async_max_threads; i++) { ErtsAsyncQ *aq = async_q(i); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(thr_opts.name, "async_%d", i+1); -#endif + erts_snprintf(thr_opts.name, 16, "async_%d", i+1); erts_thr_create(&aq->thr_id, async_main, (void*) aq, &thr_opts); } -#ifdef ETHR_HAVE_THREAD_NAMES - free(thr_opts.name); -#endif /* Wait for async threads to initialize... */ erts_mtx_lock(&async->init.data.mtx); diff --git a/erts/emulator/beam/erl_bif_binary.c b/erts/emulator/beam/erl_bif_binary.c index ff775691b3..934904d58e 100644 --- a/erts/emulator/beam/erl_bif_binary.c +++ b/erts/emulator/beam/erl_bif_binary.c @@ -32,10 +32,13 @@ #include "global.h" #include "erl_process.h" #include "error.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "erl_binary.h" #include "erl_bits.h" +#include "erl_bif_unique.h" /* @@ -1324,9 +1327,9 @@ static int parse_match_opts_list(Eterm l, Eterm bin, Uint *posp, Uint *endp) goto badarg; } if (len < 0) { - Sint lentmp = -len; + Uint lentmp = -(Uint)len; /* overflow */ - if (lentmp == len || lentmp < 0 || -lentmp != len) { + if ((Sint)lentmp < 0) { goto badarg; } len = lentmp; @@ -1555,9 +1558,9 @@ BIF_RETTYPE erts_binary_part(Process *p, Eterm binary, Eterm epos, Eterm elen) goto badarg; } if (len < 0) { - Sint lentmp = -len; + Uint lentmp = -(Uint)len; /* overflow */ - if (lentmp == len || lentmp < 0 || -lentmp != len) { + if ((Sint)lentmp < 0) { goto badarg; } len = lentmp; @@ -1644,9 +1647,9 @@ BIF_RETTYPE erts_gc_binary_part(Process *p, Eterm *reg, Eterm live, int range_is goto badarg; } if (len < 0) { - Sint lentmp = -len; + Uint lentmp = -(Uint)len; /* overflow */ - if (lentmp == len || lentmp < 0 || -lentmp != len) { + if ((Sint)lentmp < 0) { goto badarg; } len = lentmp; @@ -2213,9 +2216,9 @@ static BIF_RETTYPE binary_bin_to_list_common(Process *p, goto badarg; } if (len < 0) { - Sint lentmp = -len; + Uint lentmp = -(Uint)len; /* overflow */ - if (lentmp == len || lentmp < 0 || -lentmp != len) { + if ((Sint)lentmp < 0) { goto badarg; } len = lentmp; @@ -2294,18 +2297,11 @@ BIF_RETTYPE binary_bin_to_list_1(BIF_ALIST_1) BIF_ERROR(BIF_P,BADARG); } -/* - * Ok, erlang:list_to_binary does not interrupt, and we really don't want - * an alternative implementation for the exact same thing, why we - * have descided to use the old non-restarting implementation for now. - * In reality, there are seldom many iterations involved in doing this, so the - * problem of long-running bifs is not really that big in this case. - * So, for now we use the old implementation also in the module binary. - */ +HIPE_WRAPPER_BIF_DISABLE_GC(binary_list_to_bin, 1) BIF_RETTYPE binary_list_to_bin_1(BIF_ALIST_1) { - return erts_list_to_binary_bif(BIF_P, BIF_ARG_1); + return erts_list_to_binary_bif(BIF_P, BIF_ARG_1, bif_export[BIF_binary_list_to_bin_1]); } typedef struct { @@ -2431,8 +2427,6 @@ static BIF_RETTYPE do_binary_copy(Process *p, Eterm bin, Eterm en) } cbs->result = erts_bin_nrml_alloc(target_size); /* Always offheap if trapping */ - cbs->result->flags = 0; - cbs->result->orig_size = target_size; erts_refc_init(&(cbs->result->refc), 1); t = (byte *) cbs->result->orig_bytes; /* No offset or anything */ pos = 0; diff --git a/erts/emulator/beam/erl_bif_ddll.c b/erts/emulator/beam/erl_bif_ddll.c index 56cd2ba04f..fc4f819f56 100644 --- a/erts/emulator/beam/erl_bif_ddll.c +++ b/erts/emulator/beam/erl_bif_ddll.c @@ -45,6 +45,7 @@ #include "big.h" #include "dist.h" #include "erl_version.h" +#include "erl_bif_unique.h" #include "dtrace-wrapper.h" #ifdef ERTS_SMP diff --git a/erts/emulator/beam/erl_bif_guard.c b/erts/emulator/beam/erl_bif_guard.c index bbd8aa31d9..e7d84ebda1 100644 --- a/erts/emulator/beam/erl_bif_guard.c +++ b/erts/emulator/beam/erl_bif_guard.c @@ -459,23 +459,25 @@ Eterm erts_gc_byte_size_1(Process* p, Eterm* reg, Uint live) Eterm erts_gc_map_size_1(Process* p, Eterm* reg, Uint live) { Eterm arg = reg[live]; - if (is_map(arg)) { - map_t *mp = (map_t*)map_val(arg); - Uint size = map_get_size(mp); - if (IS_USMALL(0, size)) { - return make_small(size); - } else { - Eterm* hp; - if (ERTS_NEED_GC(p, BIG_UINT_HEAP_SIZE)) { - erts_garbage_collect(p, BIG_UINT_HEAP_SIZE, reg, live); - } - hp = p->htop; - p->htop += BIG_UINT_HEAP_SIZE; - return uint_to_big(size, hp); - } + Eterm* hp; + Uint size; + if (is_flatmap(arg)) { + flatmap_t *mp = (flatmap_t*)flatmap_val(arg); + size = flatmap_get_size(mp); + } else if (is_hashmap(arg)) { + size = hashmap_size(arg); } else { BIF_ERROR(p, BADARG); } + if (IS_USMALL(0, size)) { + return make_small(size); + } + if (ERTS_NEED_GC(p, BIG_UINT_HEAP_SIZE)) { + erts_garbage_collect(p, BIG_UINT_HEAP_SIZE, reg, live); + } + hp = p->htop; + p->htop += BIG_UINT_HEAP_SIZE; + return uint_to_big(size, hp); } Eterm erts_gc_abs_1(Process* p, Eterm* reg, Uint live) diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index 2adba9b240..b2658a1fd6 100755..100644 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -27,6 +27,7 @@ #include "erl_process.h" #include "error.h" #include "erl_driver.h" +#include "erl_nif.h" #include "bif.h" #include "big.h" #include "erl_version.h" @@ -41,6 +42,7 @@ #include "erl_cpu_topology.h" #include "erl_async.h" #include "erl_thr_progress.h" +#include "erl_bif_unique.h" #define ERTS_PTAB_WANT_DEBUG_FUNCS__ #include "erl_ptab.h" #ifdef HIPE @@ -90,7 +92,7 @@ static char erts_system_version[] = ("Erlang/OTP " ERLANG_OTP_RELEASE " [smp:%beu:%beu]" #endif #ifdef USE_THREADS -#ifdef ERTS_DIRTY_SCHEDULERS +#if defined(ERTS_DIRTY_SCHEDULERS) && defined(ERTS_SMP) " [ds:%beu:%beu:%beu]" #endif " [async-threads:%d]" @@ -115,6 +117,9 @@ static char erts_system_version[] = ("Erlang/OTP " ERLANG_OTP_RELEASE #ifdef ERTS_ENABLE_LOCK_COUNT " [lock-counting]" #endif +#ifdef ERTS_OPCODE_COUNTER_SUPPORT + " [instruction-counting]" +#endif #ifdef PURIFY " [purify-compiled]" #endif @@ -536,6 +541,7 @@ pi_locks(Eterm info) switch (info) { case am_status: case am_priority: + case am_trap_exit: return ERTS_PROC_LOCK_STATUS; case am_links: case am_monitors: @@ -588,7 +594,7 @@ static Eterm pi_args[] = { am_min_bin_vheap_size, am_current_location, am_current_stacktrace, -}; +}; #define ERTS_PI_ARGS ((int) (sizeof(pi_args)/sizeof(Eterm))) @@ -2098,6 +2104,48 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) BIF_RET(am_opt); #endif BIF_RET(res); + } else if (BIF_ARG_1 == am_time_offset) { + switch (erts_time_offset_state()) { + case ERTS_TIME_OFFSET_PRELIMINARY: { + ERTS_DECL_AM(preliminary); + BIF_RET(AM_preliminary); + } + case ERTS_TIME_OFFSET_FINAL: { + ERTS_DECL_AM(final); + BIF_RET(AM_final); + } + case ERTS_TIME_OFFSET_VOLATILE: { + ERTS_DECL_AM(volatile); + BIF_RET(AM_volatile); + } + default: + ERTS_INTERNAL_ERROR("Invalid time offset state"); + } + } else if (ERTS_IS_ATOM_STR("os_monotonic_time_source", BIF_ARG_1)) { + BIF_RET(erts_monotonic_time_source(BIF_P)); + } else if (ERTS_IS_ATOM_STR("os_system_time_source", BIF_ARG_1)) { + BIF_RET(erts_system_time_source(BIF_P)); + } else if (ERTS_IS_ATOM_STR("time_correction", BIF_ARG_1)) { + BIF_RET(erts_has_time_correction() ? am_true : am_false); + } else if (ERTS_IS_ATOM_STR("start_time", BIF_ARG_1)) { + BIF_RET(erts_get_monotonic_start_time(BIF_P)); + } else if (ERTS_IS_ATOM_STR("time_warp_mode", BIF_ARG_1)) { + switch (erts_time_warp_mode()) { + case ERTS_NO_TIME_WARP_MODE: { + ERTS_DECL_AM(no_time_warp); + BIF_RET(AM_no_time_warp); + } + case ERTS_SINGLE_TIME_WARP_MODE: { + ERTS_DECL_AM(single_time_warp); + BIF_RET(AM_single_time_warp); + } + case ERTS_MULTI_TIME_WARP_MODE: { + ERTS_DECL_AM(multi_time_warp); + BIF_RET(AM_multi_time_warp); + } + default: + ERTS_INTERNAL_ERROR("Invalid time warp mode"); + } } else if (BIF_ARG_1 == am_allocated_areas) { res = erts_allocated_areas(NULL, NULL, BIF_P); BIF_RET(res); @@ -2300,7 +2348,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) for (i = num_instructions-1; i >= 0; i--) { res = erts_bld_cons(hpp, hszp, erts_bld_tuple(hpp, hszp, 2, - erts_atom_put(opc[i].name, + erts_atom_put((byte *)opc[i].name, strlen(opc[i].name), ERTS_ATOM_ENC_LATIN1, 1), @@ -2459,6 +2507,13 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) ERL_DRV_EXTENDED_MINOR_VERSION); hp = HAlloc(BIF_P, 2*n); BIF_RET(buf_to_intlist(&hp, buf, n, NIL)); + } else if (ERTS_IS_ATOM_STR("nif_version", BIF_ARG_1)) { + char buf[42]; + int n = erts_snprintf(buf, 42, "%d.%d", + ERL_NIF_MAJOR_VERSION, + ERL_NIF_MINOR_VERSION); + hp = HAlloc(BIF_P, 2*n); + BIF_RET(buf_to_intlist(&hp, buf, n, NIL)); } else if (ERTS_IS_ATOM_STR("smp_support", BIF_ARG_1)) { #ifdef ERTS_SMP BIF_RET(am_true); @@ -2691,6 +2746,16 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) else if (ERTS_IS_ATOM_STR("ets_limit",BIF_ARG_1)) { BIF_RET(make_small(erts_db_get_max_tabs())); } + else if (ERTS_IS_ATOM_STR("tolerant_timeofday",BIF_ARG_1)) { + if (erts_has_time_correction() + && erts_time_offset_state() == ERTS_TIME_OFFSET_FINAL) { + BIF_RET(am_enabled); + } + BIF_RET(am_disabled); + } + else if (ERTS_IS_ATOM_STR("eager_check_io",BIF_ARG_1)) { + BIF_RET(erts_eager_check_io ? am_true : am_false); + } BIF_ERROR(BIF_P, BADARG); } @@ -3050,6 +3115,25 @@ fun_info_2(BIF_ALIST_2) return TUPLE2(hp, what, val); } +BIF_RETTYPE +fun_info_mfa_1(BIF_ALIST_1) +{ + Process* p = BIF_P; + Eterm fun = BIF_ARG_1; + Eterm* hp; + + if (is_fun(fun)) { + ErlFunThing* funp = (ErlFunThing *) fun_val(fun); + hp = HAlloc(p, 4); + BIF_RET(TUPLE3(hp,funp->fe->module,funp->fe->address[-2],make_small(funp->arity))); + } else if (is_export(fun)) { + Export* exp = (Export *) ((UWord) (export_val(fun))[1]); + hp = HAlloc(p, 4); + BIF_RET(TUPLE3(hp,exp->code[0],exp->code[1],make_small(exp->code[2]))); + } + BIF_ERROR(p, BADARG); +} + BIF_RETTYPE is_process_alive_1(BIF_ALIST_1) { if(is_internal_pid(BIF_ARG_1)) { @@ -3280,17 +3364,38 @@ BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1) BIF_RET(make_small((Uint) words)); } else if (ERTS_IS_ATOM_STR("check_io_debug", BIF_ARG_1)) { - /* Used by (emulator) */ - int res; + /* Used by driver_SUITE (emulator) */ + Uint sz, *szp; + Eterm res, *hp, **hpp; + int no_errors; + ErtsCheckIoDebugInfo ciodi = {0}; #ifdef HAVE_ERTS_CHECK_IO_DEBUG erts_smp_proc_unlock(BIF_P,ERTS_PROC_LOCK_MAIN); - res = erts_check_io_debug(); + no_errors = erts_check_io_debug(&ciodi); erts_smp_proc_lock(BIF_P,ERTS_PROC_LOCK_MAIN); #else - res = 0; + no_errors = 0; #endif - ASSERT(res >= 0); - BIF_RET(erts_make_integer((Uint) res, BIF_P)); + sz = 0; + szp = &sz; + hpp = NULL; + while (1) { + res = erts_bld_tuple(hpp, szp, 4, + erts_bld_uint(hpp, szp, + (Uint) no_errors), + erts_bld_uint(hpp, szp, + (Uint) ciodi.no_used_fds), + erts_bld_uint(hpp, szp, + (Uint) ciodi.no_driver_select_structs), + erts_bld_uint(hpp, szp, + (Uint) ciodi.no_driver_event_structs)); + if (hpp) + break; + hp = HAlloc(BIF_P, sz); + szp = NULL; + hpp = &hp; + } + BIF_RET(res); } else if (ERTS_IS_ATOM_STR("process_info_args", BIF_ARG_1)) { /* Used by process_SUITE (emulator) */ @@ -3344,6 +3449,29 @@ BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1) else if (ERTS_IS_ATOM_STR("mmap", BIF_ARG_1)) { BIF_RET(erts_mmap_debug_info(BIF_P)); } + else if (ERTS_IS_ATOM_STR("unique_monotonic_integer_state", BIF_ARG_1)) { + BIF_RET(erts_debug_get_unique_monotonic_integer_state(BIF_P)); + } + else if (ERTS_IS_ATOM_STR("min_unique_monotonic_integer", BIF_ARG_1)) { + Sint64 value = erts_get_min_unique_monotonic_integer(); + if (IS_SSMALL(value)) + BIF_RET(make_small(value)); + else { + Uint hsz = ERTS_SINT64_HEAP_SIZE(value); + Eterm *hp = HAlloc(BIF_P, hsz); + BIF_RET(erts_sint64_to_big(value, &hp)); + } + } + else if (ERTS_IS_ATOM_STR("min_unique_integer", BIF_ARG_1)) { + Sint64 value = erts_get_min_unique_integer(); + if (IS_SSMALL(value)) + BIF_RET(make_small(value)); + else { + Uint hsz = ERTS_SINT64_HEAP_SIZE(value); + Eterm *hp = HAlloc(BIF_P, hsz); + BIF_RET(erts_sint64_to_big(value, &hp)); + } + } } else if (is_tuple(BIF_ARG_1)) { Eterm* tp = tuple_val(BIF_ARG_1); @@ -3538,6 +3666,58 @@ BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1) BIF_RET(erts_debug_reader_groups_map(BIF_P, (int) groups)); } + else if (ERTS_IS_ATOM_STR("internal_hash", tp[1])) { + Uint hash = (Uint) make_internal_hash(tp[2]); + Uint hsz = 0; + Eterm* hp; + erts_bld_uint(NULL, &hsz, hash); + hp = HAlloc(BIF_P,hsz); + return erts_bld_uint(&hp, NULL, hash); + } + else if (ERTS_IS_ATOM_STR("atom", tp[1])) { + Uint ix; + if (!term_to_Uint(tp[2], &ix)) + BIF_ERROR(BIF_P, BADARG); + while (ix >= atom_table_size()) { + char tmp[20]; + erts_snprintf(tmp, sizeof(tmp), "am%x", atom_table_size()); + erts_atom_put((byte *) tmp, strlen(tmp), ERTS_ATOM_ENC_LATIN1, 1); + } + return make_atom(ix); + } + + break; + } + case 3: { + if (ERTS_IS_ATOM_STR("check_time_config", tp[1])) { + int res, time_correction; + ErtsTimeWarpMode time_warp_mode; + if (tp[2] == am_true) + time_correction = !0; + else if (tp[2] == am_false) + time_correction = 0; + else + break; + if (ERTS_IS_ATOM_STR("no_time_warp", tp[3])) + time_warp_mode = ERTS_NO_TIME_WARP_MODE; + else if (ERTS_IS_ATOM_STR("single_time_warp", tp[3])) + time_warp_mode = ERTS_SINGLE_TIME_WARP_MODE; + else if (ERTS_IS_ATOM_STR("multi_time_warp", tp[3])) + time_warp_mode = ERTS_MULTI_TIME_WARP_MODE; + else + break; + res = erts_check_time_adj_support(time_correction, + time_warp_mode); + BIF_RET(res ? am_true : am_false); + } + else if (ERTS_IS_ATOM_STR("make_unique_integer", tp[1])) { + Eterm res = erts_debug_make_unique_integer(BIF_P, + tp[2], + tp[3]); + if (is_non_value(res)) + break; + BIF_RET(res); + } break; } default: @@ -3547,6 +3727,24 @@ BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1) BIF_ERROR(BIF_P, BADARG); } +BIF_RETTYPE erts_internal_is_system_process_1(BIF_ALIST_1) +{ + if (is_internal_pid(BIF_ARG_1)) { + Process *rp = erts_proc_lookup(BIF_ARG_1); + if (rp && (rp->static_flags & ERTS_STC_FLG_SYSTEM_PROC)) + BIF_RET(am_true); + BIF_RET(am_false); + } + + if (is_external_pid(BIF_ARG_1) + && external_pid_dist_entry(BIF_ARG_1) == erts_this_dist_entry) { + BIF_RET(am_false); + } + + BIF_ERROR(BIF_P, BADARG); +} + + static erts_smp_atomic_t hipe_test_reschedule_flag; @@ -3841,6 +4039,17 @@ BIF_RETTYPE erts_debug_set_internal_state_2(BIF_ALIST_2) } } } + else if (ERTS_IS_ATOM_STR("broken_halt", BIF_ARG_1)) { + /* Ugly ugly code used by bif_SUITE:erlang_halt/1 */ +#if defined(ERTS_HAVE_TRY_CATCH) + erts_get_scheduler_data()->run_queue = NULL; +#endif + erl_exit(ERTS_DUMP_EXIT, "%T", BIF_ARG_2); + } + else if (ERTS_IS_ATOM_STR("unique_monotonic_integer_state", BIF_ARG_1)) { + int res = erts_debug_set_unique_monotonic_integer_state(BIF_ARG_2); + BIF_RET(res ? am_true : am_false); + } } BIF_ERROR(BIF_P, BADARG); @@ -3851,16 +4060,19 @@ static Eterm lcnt_build_lock_stats_term(Eterm **hpp, Uint *szp, erts_lcnt_lock_s Uint tries = 0, colls = 0; unsigned long timer_s = 0, timer_ns = 0, timer_n = 0; unsigned int line = 0; + unsigned int i; Eterm af, uil; Eterm uit, uic; Eterm uits, uitns, uitn; Eterm tt, tstat, tloc, t; + Eterm thist, vhist[ERTS_LCNT_HISTOGRAM_SLOT_SIZE]; /* term: - * [{{file, line}, {tries, colls, {seconds, nanoseconds, n_blocks}}}] + * [{{file, line}, {tries, colls, {seconds, nanoseconds, n_blocks}}, + * { .. histogram .. }] */ - + tries = (Uint) ethr_atomic_read(&stats->tries); colls = (Uint) ethr_atomic_read(&stats->colls); @@ -3869,23 +4081,27 @@ static Eterm lcnt_build_lock_stats_term(Eterm **hpp, Uint *szp, erts_lcnt_lock_s timer_ns = stats->timer.ns; timer_n = stats->timer_n; - af = erts_atom_put(stats->file, strlen(stats->file), ERTS_ATOM_ENC_LATIN1, 1); + af = erts_atom_put((byte *)stats->file, strlen(stats->file), ERTS_ATOM_ENC_LATIN1, 1); uil = erts_bld_uint( hpp, szp, line); tloc = erts_bld_tuple(hpp, szp, 2, af, uil); - uit = erts_bld_uint( hpp, szp, tries); - uic = erts_bld_uint( hpp, szp, colls); - + uit = erts_bld_uint( hpp, szp, tries); + uic = erts_bld_uint( hpp, szp, colls); + uits = erts_bld_uint( hpp, szp, timer_s); uitns = erts_bld_uint( hpp, szp, timer_ns); uitn = erts_bld_uint( hpp, szp, timer_n); tt = erts_bld_tuple(hpp, szp, 3, uits, uitns, uitn); tstat = erts_bld_tuple(hpp, szp, 3, uit, uic, tt); - - t = erts_bld_tuple(hpp, szp, 2, tloc, tstat); - - res = erts_bld_cons( hpp, szp, t, res); + + for(i = 0; i < ERTS_LCNT_HISTOGRAM_SLOT_SIZE; i++) { + vhist[i] = erts_bld_uint(hpp, szp, stats->hist.ns[i]); + } + thist = erts_bld_tuplev(hpp, szp, ERTS_LCNT_HISTOGRAM_SLOT_SIZE, vhist); + + t = erts_bld_tuple(hpp, szp, 3, tloc, tstat, thist); + res = erts_bld_cons( hpp, szp, t, res); return res; } @@ -3906,13 +4122,13 @@ static Eterm lcnt_build_lock_term(Eterm **hpp, Uint *szp, erts_lcnt_lock_t *lock ASSERT(ltype); - type = erts_atom_put(ltype, strlen(ltype), ERTS_ATOM_ENC_LATIN1, 1); - name = erts_atom_put(lock->name, strlen(lock->name), ERTS_ATOM_ENC_LATIN1, 1); + type = erts_atom_put((byte *)ltype, strlen(ltype), ERTS_ATOM_ENC_LATIN1, 1); + name = erts_atom_put((byte *)lock->name, strlen(lock->name), ERTS_ATOM_ENC_LATIN1, 1); if (lock->flag & ERTS_LCNT_LT_ALLOC) { /* use allocator types names as id's for allocator locks */ ltype = (char *) ERTS_ALC_A2AD(signed_val(lock->id)); - id = erts_atom_put(ltype, strlen(ltype), ERTS_ATOM_ENC_LATIN1, 1); + id = erts_atom_put((byte *)ltype, strlen(ltype), ERTS_ATOM_ENC_LATIN1, 1); } else if (lock->flag & ERTS_LCNT_LT_PROCLOCK) { /* use registered names as id's for process locks if available */ proc = erts_proc_lookup(lock->id); @@ -3923,16 +4139,15 @@ static Eterm lcnt_build_lock_term(Eterm **hpp, Uint *szp, erts_lcnt_lock_t *lock id = lock->id; } } else { - id = lock->id; + id = lock->id; } - + for (i = 0; i < lock->n_stats; i++) { stats = lcnt_build_lock_stats_term(hpp, szp, &(lock->stats[i]), stats); } - - t = erts_bld_tuple(hpp, szp, 4, name, id, type, stats); - - res = erts_bld_cons( hpp, szp, t, res); + + t = erts_bld_tuple(hpp, szp, 4, name, id, type, stats); + res = erts_bld_cons( hpp, szp, t, res); return res; } @@ -3952,12 +4167,12 @@ static Eterm lcnt_build_result_term(Eterm **hpp, Uint *szp, erts_lcnt_data_t *da dtns = erts_bld_uint( hpp, szp, data->duration.ns); tdt = erts_bld_tuple(hpp, szp, 2, dts, dtns); - adur = erts_atom_put(str_duration, strlen(str_duration), ERTS_ATOM_ENC_LATIN1, 1); + adur = erts_atom_put((byte *)str_duration, strlen(str_duration), ERTS_ATOM_ENC_LATIN1, 1); tdur = erts_bld_tuple(hpp, szp, 2, adur, tdt); /* lock tuple */ - aloc = erts_atom_put(str_locks, strlen(str_locks), ERTS_ATOM_ENC_LATIN1, 1); + aloc = erts_atom_put((byte *)str_locks, strlen(str_locks), ERTS_ATOM_ENC_LATIN1, 1); for (lock = data->current_locks->head; lock != NULL ; lock = lock->next ) { lloc = lcnt_build_lock_term(hpp, szp, lock, lloc); diff --git a/erts/emulator/beam/erl_bif_port.c b/erts/emulator/beam/erl_bif_port.c index afb33c1cdb..7ce950e090 100644 --- a/erts/emulator/beam/erl_bif_port.c +++ b/erts/emulator/beam/erl_bif_port.c @@ -472,7 +472,7 @@ cleanup_old_port_data(erts_aint_t data) ErtsPortDataHeap *pdhp = (ErtsPortDataHeap *) data; size_t size; ERTS_SMP_DATA_DEPENDENCY_READ_MEMORY_BARRIER; - size = sizeof(ErtsPortDataHeap) + pdhp->hsize*(sizeof(Eterm) - 1); + size = sizeof(ErtsPortDataHeap) + (pdhp->hsize-1)*sizeof(Eterm); erts_schedule_thr_prgr_later_cleanup_op(free_port_data_heap, (void *) pdhp, &pdhp->later_op, @@ -493,8 +493,8 @@ void erts_cleanup_port_data(Port *prt) { ASSERT(erts_atomic32_read_nob(&prt->state) & ERTS_PORT_SFLGS_INVALID_LOOKUP); - cleanup_old_port_data(erts_smp_atomic_read_nob(&prt->data)); - erts_smp_atomic_set_nob(&prt->data, (erts_aint_t) THE_NON_VALUE); + cleanup_old_port_data(erts_smp_atomic_xchg_nob(&prt->data, + (erts_aint_t) NULL)); } Uint @@ -508,7 +508,7 @@ erts_port_data_size(Port *prt) } else { ErtsPortDataHeap *pdhp = (ErtsPortDataHeap *) data; - return (Uint) sizeof(ErtsPortDataHeap) + pdhp->hsize*(sizeof(Eterm)-1); + return (Uint) sizeof(ErtsPortDataHeap) + (pdhp->hsize-1)*sizeof(Eterm); } } @@ -550,10 +550,11 @@ BIF_RETTYPE port_set_data_2(BIF_ALIST_2) hsize = size_object(BIF_ARG_2); pdhp = erts_alloc(ERTS_ALC_T_PORT_DATA_HEAP, - sizeof(ErtsPortDataHeap) + hsize*(sizeof(Eterm)-1)); + sizeof(ErtsPortDataHeap) + (hsize-1)*sizeof(Eterm)); hp = &pdhp->heap[0]; pdhp->off_heap.first = NULL; pdhp->off_heap.overhead = 0; + pdhp->hsize = hsize; pdhp->data = copy_struct(BIF_ARG_2, hsize, &hp, &pdhp->off_heap); data = (erts_aint_t) pdhp; ASSERT((data & 0x3) == 0); @@ -561,8 +562,14 @@ BIF_RETTYPE port_set_data_2(BIF_ALIST_2) data = erts_smp_atomic_xchg_wb(&prt->data, data); + if (data == (erts_aint_t)NULL) { + /* Port terminated by racing thread */ + data = erts_smp_atomic_xchg_wb(&prt->data, data); + ASSERT(data != (erts_aint_t)NULL); + cleanup_old_port_data(data); + BIF_ERROR(BIF_P, BADARG); + } cleanup_old_port_data(data); - BIF_RET(am_true); } @@ -581,6 +588,8 @@ BIF_RETTYPE port_get_data_1(BIF_ALIST_1) BIF_ERROR(BIF_P, BADARG); data = erts_smp_atomic_read_ddrb(&prt->data); + if (data == (erts_aint_t)NULL) + BIF_ERROR(BIF_P, BADARG); /* Port terminated by racing thread */ if ((data & 0x3) != 0) { res = (Eterm) (UWord) data; diff --git a/erts/emulator/beam/erl_bif_timer.c b/erts/emulator/beam/erl_bif_timer.c index 03ac97283c..0bd8d20c34 100644 --- a/erts/emulator/beam/erl_bif_timer.c +++ b/erts/emulator/beam/erl_bif_timer.c @@ -27,6 +27,7 @@ #include "error.h" #include "big.h" #include "erl_thr_progress.h" +#include "erl_bif_unique.h" /**************************************************************************** ** BIF Timer support @@ -480,7 +481,7 @@ setup_bif_timer(Uint32 xflags, tab_insert(btm); ASSERT(btm == tab_find(ref)); - btm->tm.active = 0; /* MUST be initalized */ + erts_init_timer(&btm->tm); erts_set_timer(&btm->tm, (ErlTimeoutProc) bif_timer_timeout, (ErlCancelProc) bif_timer_cleanup, @@ -489,8 +490,9 @@ setup_bif_timer(Uint32 xflags, return ref; } +BIF_RETTYPE old_send_after_3(BIF_ALIST_3); /* send_after(Time, Pid, Message) -> Ref */ -BIF_RETTYPE send_after_3(BIF_ALIST_3) +BIF_RETTYPE old_send_after_3(BIF_ALIST_3) { Eterm res; @@ -510,8 +512,9 @@ BIF_RETTYPE send_after_3(BIF_ALIST_3) } } +BIF_RETTYPE old_start_timer_3(BIF_ALIST_3); /* start_timer(Time, Pid, Message) -> Ref */ -BIF_RETTYPE start_timer_3(BIF_ALIST_3) +BIF_RETTYPE old_start_timer_3(BIF_ALIST_3) { Eterm res; @@ -531,8 +534,9 @@ BIF_RETTYPE start_timer_3(BIF_ALIST_3) } } +BIF_RETTYPE old_cancel_timer_1(BIF_ALIST_1); /* cancel_timer(Ref) -> false | RemainingTime */ -BIF_RETTYPE cancel_timer_1(BIF_ALIST_1) +BIF_RETTYPE old_cancel_timer_1(BIF_ALIST_1) { Eterm res; ErtsBifTimer *btm; @@ -569,8 +573,9 @@ BIF_RETTYPE cancel_timer_1(BIF_ALIST_1) BIF_RET(res); } +BIF_RETTYPE old_read_timer_1(BIF_ALIST_1); /* read_timer(Ref) -> false | RemainingTime */ -BIF_RETTYPE read_timer_1(BIF_ALIST_1) +BIF_RETTYPE old_read_timer_1(BIF_ALIST_1) { Eterm res; ErtsBifTimer *btm; @@ -652,7 +657,7 @@ erts_cancel_bif_timers(Process *p, ErtsProcLocks plocks) erts_smp_btm_rwunlock(); } -void erts_bif_timer_init(void) +static void erts_old_bif_timer_init(void) { int i; no_bif_timers = 0; @@ -703,3 +708,146 @@ erts_bif_timer_foreach(void (*func)(Eterm, Eterm, ErlHeapFragment *, void *), } } } + +typedef struct { + Uint ref_heap[REF_THING_SIZE]; + Eterm pid[1]; +} ErtsBifTimerServers; + +static ErtsBifTimerServers *bif_timer_servers; + +void erts_bif_timer_init(void) +{ + erts_old_bif_timer_init(); +} + +void +erts_bif_timer_start_servers(Eterm parent) +{ + Process *parent_proc; + Eterm *hp, btr_ref, arg_list_end; + ErlSpawnOpts so; + int i; + + bif_timer_servers = erts_alloc(ERTS_ALC_T_BIF_TIMER_DATA, + (sizeof(ErtsBifTimerServers) + + (sizeof(Eterm)*(erts_no_schedulers-1)))); + + so.flags = SPO_USE_ARGS|SPO_SYSTEM_PROC|SPO_PREFER_SCHED|SPO_OFF_HEAP_MSGS; + so.min_heap_size = H_MIN_SIZE; + so.min_vheap_size = BIN_VH_MIN_SIZE; + so.priority = PRIORITY_MAX; + so.max_gen_gcs = (Uint16) erts_smp_atomic32_read_nob(&erts_max_gen_gcs); + + /* + * Parent is "init" and schedulers have not yet been started, so it + * *should* be alive and well... + */ + ASSERT(is_internal_pid(parent)); + parent_proc = (Process *) erts_ptab_pix2intptr_ddrb(&erts_proc, + internal_pid_index(parent)); + ASSERT(parent_proc); + ASSERT(parent_proc->common.id == parent); + ASSERT(!ERTS_PROC_IS_EXITING(parent_proc)); + + erts_smp_proc_lock(parent_proc, ERTS_PROC_LOCK_MAIN); + + hp = HAlloc(parent_proc, 2*erts_no_schedulers + 2 + REF_THING_SIZE); + + btr_ref = erts_make_ref_in_buffer(hp); + hp += REF_THING_SIZE; + + arg_list_end = CONS(hp, btr_ref, NIL); + hp += 2; + + for (i = 0; i < erts_no_schedulers; i++) { + int sched = i+1; + Eterm arg_list = CONS(hp, make_small(i+1), arg_list_end); + hp += 2; + + so.scheduler = sched; /* Preferred scheduler */ + + bif_timer_servers->pid[i] = erl_create_process(parent_proc, + am_erts_internal, + am_bif_timer_server, + arg_list, + &so); + } + + erts_smp_proc_unlock(parent_proc, ERTS_PROC_LOCK_MAIN); + + hp = internal_ref_val(btr_ref); + for (i = 0; i < REF_THING_SIZE; i++) + bif_timer_servers->ref_heap[i] = hp[i]; +} + +BIF_RETTYPE +erts_internal_get_bif_timer_servers_0(BIF_ALIST_0) +{ + int i; + Eterm *hp, res = NIL; + + hp = HAlloc(BIF_P, erts_no_schedulers*2); + for (i = erts_no_schedulers-1; i >= 0; i--) { + res = CONS(hp, bif_timer_servers->pid[i], res); + hp += 2; + } + BIF_RET(res); +} + +BIF_RETTYPE +erts_internal_access_bif_timer_1(BIF_ALIST_1) +{ + int ix; + Uint32 *rdp; + Eterm ref, pid, *hp, res; + + if (is_not_internal_ref(BIF_ARG_1)) { + if (is_not_ref(BIF_ARG_1)) + BIF_ERROR(BIF_P, BADARG); + BIF_RET(am_undefined); + } + + rdp = internal_ref_numbers(BIF_ARG_1); + ix = (int) erts_get_ref_numbers_thr_id(rdp); + if (ix < 1 || erts_no_schedulers < ix) + BIF_RET(am_undefined); + + pid = bif_timer_servers->pid[ix-1]; + ASSERT(is_internal_pid(pid)); + + hp = HAlloc(BIF_P, 3 /* 2-tuple */ + REF_THING_SIZE); + for (ix = 0; ix < REF_THING_SIZE; ix++) + hp[ix] = bif_timer_servers->ref_heap[ix]; + ref = make_internal_ref(&hp[0]); + hp += REF_THING_SIZE; + + res = TUPLE2(hp, ref, pid); + BIF_RET(res); +} + +BIF_RETTYPE +erts_internal_create_bif_timer_0(BIF_ALIST_0) +{ + ErtsSchedulerData *esdp = ERTS_PROC_GET_SCHDATA(BIF_P); + Eterm *hp, btr_ref, t_ref, pid, res; + int ix; + + hp = HAlloc(BIF_P, 4 /* 3-tuple */ + 2*REF_THING_SIZE); + for (ix = 0; ix < REF_THING_SIZE; ix++) + hp[ix] = bif_timer_servers->ref_heap[ix]; + btr_ref = make_internal_ref(&hp[0]); + hp += REF_THING_SIZE; + + t_ref = erts_sched_make_ref_in_buffer(esdp, hp); + hp += REF_THING_SIZE; + + ASSERT(erts_get_ref_numbers_thr_id(internal_ref_numbers(t_ref)) + == (Uint32) esdp->no); + + pid = bif_timer_servers->pid[((int) esdp->no) - 1]; + + res = TUPLE3(hp, btr_ref, pid, t_ref); + + BIF_RET(res); +} diff --git a/erts/emulator/beam/erl_bif_timer.h b/erts/emulator/beam/erl_bif_timer.h index 1197c176f5..c2f5dfd3c3 100644 --- a/erts/emulator/beam/erl_bif_timer.h +++ b/erts/emulator/beam/erl_bif_timer.h @@ -33,4 +33,5 @@ void erts_cancel_bif_timers(Process *p, ErtsProcLocks plocks); void erts_bif_timer_init(void); void erts_bif_timer_foreach(void (*func)(Eterm,Eterm,ErlHeapFragment *,void *), void *arg); +void erts_bif_timer_start_servers(Eterm); #endif diff --git a/erts/emulator/beam/erl_bif_trace.c b/erts/emulator/beam/erl_bif_trace.c index 06fbbea123..ac57205c47 100644 --- a/erts/emulator/beam/erl_bif_trace.c +++ b/erts/emulator/beam/erl_bif_trace.c @@ -38,6 +38,7 @@ #include "beam_bp.h" #include "erl_binary.h" #include "erl_thr_progress.h" +#include "erl_bif_unique.h" #define DECL_AM(S) Eterm AM_ ## S = am_atom_put(#S, sizeof(#S) - 1) @@ -651,7 +652,7 @@ Eterm trace_3(BIF_ALIST_3) if (pid_spec == am_all) { if (on) { if (!erts_cpu_timestamp) { -#ifdef HAVE_CLOCK_GETTIME +#ifdef HAVE_CLOCK_GETTIME_CPU_TIME /* Perhaps clock_gettime was found during config on a different machine than this. We check @@ -678,7 +679,7 @@ Eterm trace_3(BIF_ALIST_3) if (erts_start_now_cpu() < 0) { goto error; } -#endif /* HAVE_CLOCK_GETTIME */ +#endif /* HAVE_CLOCK_GETTIME_CPU_TIME */ erts_cpu_timestamp = !0; } } diff --git a/erts/emulator/beam/erl_bif_unique.c b/erts/emulator/beam/erl_bif_unique.c new file mode 100644 index 0000000000..57b0bab72f --- /dev/null +++ b/erts/emulator/beam/erl_bif_unique.c @@ -0,0 +1,556 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2014. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "sys.h" +#include "erl_vm.h" +#include "erl_alloc.h" +#include "export.h" +#include "bif.h" +#include "erl_bif_unique.h" + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * Reference * +\* */ + +static union { + erts_atomic64_t count; + char align__[ERTS_CACHE_LINE_SIZE]; +} global_reference erts_align_attribute(ERTS_CACHE_LINE_SIZE); + + +/* + * ref[0] indicate thread creating reference as follows: + * + * - ref[0] == 0 => Non-scheduler thread; + * - else; ref[0] <= erts_no_schedulers => + * ordinary scheduler with id == ref[0]; + * - else; ref[0] <= erts_no_schedulers + * + erts_no_dirty_cpu_schedulers => + * dirty cpu scheduler with id == 'ref[0] - erts_no_schedulers'; + * - else => + * dirty io scheduler with id == 'ref[0] + * - erts_no_schedulers + * - erts_no_dirty_cpu_schedulers' + */ + +#ifdef DEBUG +static Uint32 max_thr_id; +#endif + +static void +init_reference(void) +{ +#ifdef DEBUG + max_thr_id = (Uint32) erts_no_schedulers; +#ifdef ERTS_DIRTY_SCHEDULERS + max_thr_id += (Uint32) erts_no_dirty_cpu_schedulers; + max_thr_id += (Uint32) erts_no_dirty_io_schedulers; +#endif +#endif + erts_atomic64_init_nob(&global_reference.count, 0); +} + +static ERTS_INLINE void +global_make_ref_in_array(Uint32 thr_id, Uint32 ref[ERTS_MAX_REF_NUMBERS]) +{ + Uint64 value; + + value = (Uint64) erts_atomic64_inc_read_mb(&global_reference.count); + + erts_set_ref_numbers(ref, thr_id, value); +} + +static ERTS_INLINE void +make_ref_in_array(Uint32 ref[ERTS_MAX_REF_NUMBERS]) +{ + ErtsSchedulerData *esdp = erts_get_scheduler_data(); + if (esdp) + erts_sched_make_ref_in_array(esdp, ref); + else + global_make_ref_in_array(0, ref); +} + +void +erts_make_ref_in_array(Uint32 ref[ERTS_MAX_REF_NUMBERS]) +{ + make_ref_in_array(ref); +} + +Eterm erts_make_ref_in_buffer(Eterm buffer[REF_THING_SIZE]) +{ + Eterm* hp = buffer; + Uint32 ref[ERTS_MAX_REF_NUMBERS]; + + make_ref_in_array(ref); + write_ref_thing(hp, ref[0], ref[1], ref[2]); + return make_internal_ref(hp); +} + +Eterm erts_make_ref(Process *c_p) +{ + Eterm* hp; + Uint32 ref[ERTS_MAX_REF_NUMBERS]; + + ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_MAIN & erts_proc_lc_my_proc_locks(c_p)); + + hp = HAlloc(c_p, REF_THING_SIZE); + + make_ref_in_array(ref); + write_ref_thing(hp, ref[0], ref[1], ref[2]); + + return make_internal_ref(hp); +} + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * Unique Integer * +\* */ + +static struct { + union { + struct { + int left_shift; + int right_shift; + Uint64 mask; + Uint64 val0_max; + } o; + char align__[ERTS_CACHE_LINE_SIZE]; + } r; + union { + erts_atomic64_t val1; + char align__[ERTS_CACHE_LINE_SIZE]; + } w; +} unique_data erts_align_attribute(ERTS_CACHE_LINE_SIZE); + +static void +init_unique_integer(void) +{ + int bits; + unique_data.r.o.val0_max = (Uint64) erts_no_schedulers; +#ifdef ERTS_DIRTY_SCHEDULERS + unique_data.r.o.val0_max += (Uint64) erts_no_dirty_cpu_schedulers; + unique_data.r.o.val0_max += (Uint64) erts_no_dirty_io_schedulers; +#endif + bits = erts_fit_in_bits_int64(unique_data.r.o.val0_max); + unique_data.r.o.left_shift = bits; + unique_data.r.o.right_shift = 64 - bits; + unique_data.r.o.mask = (((Uint64) 1) << bits) - 1; + erts_atomic64_init_nob(&unique_data.w.val1, -1); +} + +#define ERTS_MAX_UNIQUE_INT_HEAP_SIZE ERTS_UINT64_ARRAY_TO_BIG_MAX_HEAP_SZ(2) + +static ERTS_INLINE Eterm +bld_unique_integer_term(Eterm **hpp, Uint *szp, + Uint64 val0, Uint64 val1, + int positive) +{ + Uint hsz; + Uint64 unique_val[2]; + + unique_val[0] = ((Uint64) val0); + unique_val[0] |= ((Uint64) val1) << unique_data.r.o.left_shift; + unique_val[1] = ((Uint64) val1) >> unique_data.r.o.right_shift; + unique_val[1] &= unique_data.r.o.mask; + + if (positive) { + unique_val[0]++; + if (unique_val[0] == 0) + unique_val[1]++; + } + else { + ASSERT(MIN_SMALL < 0); + if (unique_val[1] == 0 + && unique_val[0] < ((Uint64) -1*((Sint64) MIN_SMALL))) { + Sint64 s_unique_val = (Sint64) unique_val[0]; + s_unique_val += MIN_SMALL; + ASSERT(MIN_SMALL <= s_unique_val && s_unique_val < 0); + if (szp) + *szp = 0; + if (!hpp) + return THE_NON_VALUE; + return make_small((Sint) s_unique_val); + } + if (unique_val[0] < ((Uint64) -1*((Sint64) MIN_SMALL))) { + ASSERT(unique_val[1] != 0); + unique_val[1] -= 1; + } + unique_val[0] += MIN_SMALL; + } + + if (!unique_val[1]) { + if (unique_val[0] <= MAX_SMALL) { + if (szp) + *szp = 0; + if (!hpp) + return THE_NON_VALUE; + return make_small((Uint) unique_val[0]); + } + + if (szp) + *szp = ERTS_UINT64_HEAP_SIZE(unique_val[0]); + if (!hpp) + return THE_NON_VALUE; + return erts_uint64_to_big(unique_val[0], hpp); + } + else { + Eterm tmp, *tmp_hp, res; + DeclareTmpHeapNoproc(local_heap, 2*ERTS_MAX_UNIQUE_INT_HEAP_SIZE); + + UseTmpHeapNoproc(2*ERTS_MAX_UNIQUE_INT_HEAP_SIZE); + + tmp_hp = local_heap; + + tmp = erts_uint64_array_to_big(&tmp_hp, 0, 2, unique_val); + ASSERT(is_big(tmp)); + + hsz = big_arity(tmp) + 1; + + ASSERT(hsz <= ERTS_MAX_UNIQUE_INT_HEAP_SIZE); + + if (szp) + *szp = hsz; + + if (!hpp) + res = THE_NON_VALUE; + else { + int hix; + Eterm *hp = *hpp; + tmp_hp = big_val(tmp); + for (hix = 0; hix < hsz; hix++) + hp[hix] = tmp_hp[hix]; + + *hpp = hp + hsz; + res = make_big(hp); + } + + UnUseTmpHeapNoproc(2*ERTS_MAX_UNIQUE_INT_HEAP_SIZE); + + return res; + } +} + +static ERTS_INLINE Eterm unique_integer_bif(Process *c_p, int positive) +{ + ErtsSchedulerData *esdp; + Uint64 thr_id, unique; + Uint hsz; + Eterm *hp; + + esdp = ERTS_PROC_GET_SCHDATA(c_p); + thr_id = (Uint64) esdp->thr_id; + unique = esdp->unique++; + bld_unique_integer_term(NULL, &hsz, thr_id, unique, positive); + hp = hsz ? HAlloc(c_p, hsz) : NULL; + return bld_unique_integer_term(&hp, NULL, thr_id, unique, positive); +} + +Uint +erts_raw_unique_integer_heap_size(Uint64 val[ERTS_UNIQUE_INT_RAW_VALUES]) +{ + Uint sz; + bld_unique_integer_term(NULL, &sz, val[0], val[1], 0); + return sz; +} + +Eterm +erts_raw_make_unique_integer(Eterm **hpp, Uint64 val[ERTS_UNIQUE_INT_RAW_VALUES]) +{ + return bld_unique_integer_term(hpp, NULL, val[0], val[1], 0); +} + +void +erts_raw_get_unique_integer(Uint64 val[ERTS_UNIQUE_INT_RAW_VALUES]) +{ + ErtsSchedulerData *esdp = erts_get_scheduler_data(); + if (esdp) { + val[0] = (Uint64) esdp->thr_id; + val[1] = esdp->unique++; + } + else { + val[0] = (Uint64) 0; + val[1] = (Uint64) erts_atomic64_inc_read_nob(&unique_data.w.val1); + } +} + + +Sint64 +erts_get_min_unique_integer(void) +{ + return (Sint64) MIN_SMALL; +} + +/* --- Debug --- */ + +Eterm +erts_debug_make_unique_integer(Process *c_p, Eterm etval0, Eterm etval1) +{ + Uint64 val0, val1; + Uint hsz; + Eterm res, *hp, *end_hp; + + if (!term_to_Uint64(etval0, &val0)) + return THE_NON_VALUE; + + if (!term_to_Uint64(etval1, &val1)) + return THE_NON_VALUE; + + bld_unique_integer_term(NULL, &hsz, val0, val1, 0); + + hp = HAlloc(c_p, hsz); + end_hp = hp + hsz; + + res = bld_unique_integer_term(&hp, NULL, val0, val1, 0); + if (hp != end_hp) + ERTS_INTERNAL_ERROR("Heap allocation error"); + + return res; +} + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * Strict Monotonic Counter * +\* */ + +static struct { + union { + erts_atomic64_t value; + char align__[ERTS_CACHE_LINE_SIZE]; + } w; +} raw_unique_monotonic_integer erts_align_attribute(ERTS_CACHE_LINE_SIZE); + +#if defined(ARCH_32) || HALFWORD_HEAP +# define ERTS_UNIQUE_MONOTONIC_OFFSET ERTS_SINT64_MIN +#else +# define ERTS_UNIQUE_MONOTONIC_OFFSET MIN_SMALL +#endif + +static void +init_unique_monotonic_integer(void) +{ + erts_atomic64_init_nob(&raw_unique_monotonic_integer.w.value, + (erts_aint64_t) -1); +} + +static ERTS_INLINE Uint64 +get_raw_unique_monotonic_integer(void) +{ + return (Uint64) erts_atomic64_inc_read_mb(&raw_unique_monotonic_integer.w.value); +} + +static ERTS_INLINE Uint +get_unique_monotonic_integer_heap_size(Uint64 raw, int positive) +{ + if (positive) { + Uint64 value = raw+1; + return ERTS_UINT64_HEAP_SIZE(value); + } + else { + Sint64 value = ((Sint64) raw) + ERTS_UNIQUE_MONOTONIC_OFFSET; + if (IS_SSMALL(value)) + return 0; +#if defined(ARCH_32) || HALFWORD_HEAP + return ERTS_SINT64_HEAP_SIZE(value); +#else + return ERTS_UINT64_HEAP_SIZE((Uint64) value); +#endif + } +} + +static ERTS_INLINE Eterm +make_unique_monotonic_integer_value(Eterm *hp, Uint hsz, Uint64 raw, int positive) +{ + Eterm res; +#ifdef DEBUG + Eterm *end_hp = hp + hsz; +#endif + + if (positive) { + Uint64 value = raw+1; + res = hsz ? erts_uint64_to_big(value, &hp) : make_small(value); + } + else { + Sint64 value = ((Sint64) raw) + ERTS_UNIQUE_MONOTONIC_OFFSET; + if (hsz == 0) + res = make_small(value); + else { +#if defined(ARCH_32) || HALFWORD_HEAP + res = erts_sint64_to_big(value, &hp); +#else + res = erts_uint64_to_big((Uint64) value, &hp); +#endif + } + } + + ASSERT(end_hp == hp); + + return res; +} + +static ERTS_INLINE Eterm +unique_monotonic_integer_bif(Process *c_p, int positive) +{ + Uint64 raw; + Uint hsz; + Eterm *hp; + + raw = get_raw_unique_monotonic_integer(); + hsz = get_unique_monotonic_integer_heap_size(raw, positive); + hp = hsz ? HAlloc(c_p, hsz) : NULL; + return make_unique_monotonic_integer_value(hp, hsz, raw, positive); +} + +Sint64 +erts_raw_get_unique_monotonic_integer(void) +{ + return get_raw_unique_monotonic_integer(); +} + +Uint +erts_raw_unique_monotonic_integer_heap_size(Sint64 raw) +{ + return get_unique_monotonic_integer_heap_size(raw, 0); +} + +Eterm +erts_raw_make_unique_monotonic_integer_value(Eterm **hpp, Sint64 raw) +{ + Uint hsz = get_unique_monotonic_integer_heap_size(raw, 0); + Eterm res = make_unique_monotonic_integer_value(*hpp, hsz, raw, 0); + *hpp += hsz; + return res; +} + +Sint64 +erts_get_min_unique_monotonic_integer(void) +{ + return ERTS_UNIQUE_MONOTONIC_OFFSET; +} + +/* --- Debug --- */ + +int +erts_debug_set_unique_monotonic_integer_state(Eterm et_value) +{ + Sint64 value; + + if (!term_to_Sint64(et_value, &value)) { + Uint64 uvalue; + if (!term_to_Uint64(et_value, &uvalue)) + return 0; + value = (Sint64) uvalue; + } + + erts_atomic64_set_mb(&raw_unique_monotonic_integer.w.value, + (erts_aint64_t) value); + return 1; +} + +Eterm +erts_debug_get_unique_monotonic_integer_state(Process *c_p) +{ + Uint64 value; + Eterm hsz, *hp; + + value = (Uint64) erts_atomic64_read_mb(&raw_unique_monotonic_integer.w.value); + + if (IS_USMALL(0, value)) + return make_small(value); + hsz = ERTS_UINT64_HEAP_SIZE(value); + hp = HAlloc(c_p, hsz); + return erts_uint64_to_big(value, &hp); +} + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * Initilazation * +\* */ + +void +erts_bif_unique_init(void) +{ + init_reference(); + init_unique_monotonic_integer(); + init_unique_integer(); +} + +void +erts_sched_bif_unique_init(ErtsSchedulerData *esdp) +{ + esdp->unique = (Uint64) 0; + esdp->ref = (Uint64) 0; +} + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * The BIFs * +\* */ + + +BIF_RETTYPE make_ref_0(BIF_ALIST_0) +{ + BIF_RETTYPE res; + Eterm* hp; + + ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_MAIN & erts_proc_lc_my_proc_locks(BIF_P)); + + hp = HAlloc(BIF_P, REF_THING_SIZE); + + res = erts_sched_make_ref_in_buffer(ERTS_PROC_GET_SCHDATA(BIF_P), hp); + + BIF_RET(res); +} + +BIF_RETTYPE unique_integer_0(BIF_ALIST_0) +{ + BIF_RET(unique_integer_bif(BIF_P, 0)); +} + +BIF_RETTYPE unique_integer_1(BIF_ALIST_1) +{ + Eterm modlist = BIF_ARG_1; + int monotonic = 0; + int positive = 0; + BIF_RETTYPE res; + + while (is_list(modlist)) { + Eterm *consp = list_val(modlist); + switch (CAR(consp)) { + case am_monotonic: + monotonic = 1; + break; + case am_positive: + positive = 1; + break; + default: + BIF_ERROR(BIF_P, BADARG); + } + modlist = CDR(consp); + } + + if (is_not_nil(modlist)) + BIF_ERROR(BIF_P, BADARG); + + if (monotonic) + res = unique_monotonic_integer_bif(BIF_P, positive); + else + res = unique_integer_bif(BIF_P, positive); + + BIF_RET(res); +} diff --git a/erts/emulator/beam/erl_bif_unique.h b/erts/emulator/beam/erl_bif_unique.h new file mode 100644 index 0000000000..cd001172a1 --- /dev/null +++ b/erts/emulator/beam/erl_bif_unique.h @@ -0,0 +1,131 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2014. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +#ifndef ERTS_BIF_UNIQUE_H__ +#define ERTS_BIF_UNIQUE_H__ + +#include "erl_process.h" +#include "big.h" + +void erts_bif_unique_init(void); +void erts_sched_bif_unique_init(ErtsSchedulerData *esdp); + +/* reference */ +Eterm erts_make_ref(Process *); +Eterm erts_make_ref_in_buffer(Eterm buffer[REF_THING_SIZE]); +void erts_make_ref_in_array(Uint32 ref[ERTS_MAX_REF_NUMBERS]); + +/* strict monotonic counter */ + +#define ERTS_MAX_UNIQUE_MONOTONIC_INTEGER_HEAP_SIZE ERTS_MAX_UINT64_HEAP_SIZE + +/* + * Note that a raw value is an intermediate value that + * not necessarily correspond to the end result. + */ +Sint64 erts_raw_get_unique_monotonic_integer(void); +Uint erts_raw_unique_monotonic_integer_heap_size(Sint64 raw); +Eterm erts_raw_make_unique_monotonic_integer_value(Eterm **hpp, Sint64 raw); + +Sint64 erts_get_min_unique_monotonic_integer(void); + +int erts_debug_set_unique_monotonic_integer_state(Eterm et_value); +Eterm erts_debug_get_unique_monotonic_integer_state(Process *c_p); + +/* unique integer */ +#define ERTS_UNIQUE_INT_RAW_VALUES 2 +#define ERTS_MAX_UNIQUE_INT_HEAP_SIZE ERTS_UINT64_ARRAY_TO_BIG_MAX_HEAP_SZ(2) + +Uint erts_raw_unique_integer_heap_size(Uint64 val[ERTS_UNIQUE_INT_RAW_VALUES]); +Eterm erts_raw_make_unique_integer(Eterm **hpp, Uint64 val[ERTS_UNIQUE_INT_RAW_VALUES]); +void erts_raw_get_unique_integer(Uint64 val[ERTS_UNIQUE_INT_RAW_VALUES]); +Sint64 erts_get_min_unique_integer(void); + +Eterm erts_debug_make_unique_integer(Process *c_p, + Eterm etval0, + Eterm etval1); + + +ERTS_GLB_INLINE void erts_set_ref_numbers(Uint32 ref[ERTS_MAX_REF_NUMBERS], + Uint32 thr_id, Uint64 value); +ERTS_GLB_INLINE Uint32 erts_get_ref_numbers_thr_id(Uint32 ref[ERTS_MAX_REF_NUMBERS]); +ERTS_GLB_INLINE Uint64 erts_get_ref_numbers_value(Uint32 ref[ERTS_MAX_REF_NUMBERS]); +ERTS_GLB_INLINE void erts_sched_make_ref_in_array(ErtsSchedulerData *esdp, + Uint32 ref[ERTS_MAX_REF_NUMBERS]); +ERTS_GLB_INLINE Eterm erts_sched_make_ref_in_buffer(ErtsSchedulerData *esdp, + Eterm buffer[REF_THING_SIZE]); + +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +ERTS_GLB_INLINE void +erts_set_ref_numbers(Uint32 ref[ERTS_MAX_REF_NUMBERS], Uint32 thr_id, Uint64 value) +{ + /* + * We cannot use thread id in the first 18-bit word since + * the hash/phash/phash2 BIFs only hash on this word. If + * we did, we would get really poor hash values. Instead + * we have to shuffle the bits a bit. + */ + ASSERT(thr_id == (thr_id & ((Uint32) 0x3ffff))); + ref[0] = (Uint32) (value & ((Uint64) 0x3ffff)); + ref[1] = (((Uint32) (value & ((Uint64) 0xfffc0000))) + | (thr_id & ((Uint32) 0x3ffff))); + ref[2] = (Uint32) ((value >> 32) & ((Uint64) 0xffffffff)); +} + +ERTS_GLB_INLINE Uint32 +erts_get_ref_numbers_thr_id(Uint32 ref[ERTS_MAX_REF_NUMBERS]) +{ + return ref[1] & ((Uint32) 0x3ffff); +} + +ERTS_GLB_INLINE Uint64 +erts_get_ref_numbers_value(Uint32 ref[ERTS_MAX_REF_NUMBERS]) +{ + return (((((Uint64) ref[2]) & ((Uint64) 0xffffffff)) << 32) + | (((Uint64) ref[1]) & ((Uint64) 0xfffc0000)) + | (((Uint64) ref[0]) & ((Uint64) 0x3ffff))); +} + +ERTS_GLB_INLINE void +erts_sched_make_ref_in_array(ErtsSchedulerData *esdp, + Uint32 ref[ERTS_MAX_REF_NUMBERS]) +{ + Uint64 value; + + ASSERT(esdp); + value = esdp->ref++; + erts_set_ref_numbers(ref, (Uint32) esdp->thr_id, value); +} + +ERTS_GLB_INLINE Eterm +erts_sched_make_ref_in_buffer(ErtsSchedulerData *esdp, + Eterm buffer[REF_THING_SIZE]) +{ + Eterm* hp = buffer; + Uint32 ref[ERTS_MAX_REF_NUMBERS]; + + erts_sched_make_ref_in_array(esdp, ref); + write_ref_thing(hp, ref[0], ref[1], ref[2]); + return make_internal_ref(hp); +} + +#endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */ + +#endif /* ERTS_BIF_UNIQUE_H__ */ diff --git a/erts/emulator/beam/erl_binary.h b/erts/emulator/beam/erl_binary.h index 819b19e566..8d264d166e 100644 --- a/erts/emulator/beam/erl_binary.h +++ b/erts/emulator/beam/erl_binary.h @@ -166,7 +166,7 @@ Eterm erts_bin_bytes_to_list(Eterm previous, Eterm* hp, byte* bytes, Uint size, * Common implementation for erlang:list_to_binary/1 and binary:list_to_bin/1 */ -BIF_RETTYPE erts_list_to_binary_bif(Process *p, Eterm arg); +BIF_RETTYPE erts_list_to_binary_bif(Process *p, Eterm arg, Export *bif); BIF_RETTYPE erts_gc_binary_part(Process *p, Eterm *reg, Eterm live, int range_is_tuple); BIF_RETTYPE erts_binary_part(Process *p, Eterm binary, Eterm epos, Eterm elen); @@ -231,35 +231,58 @@ erts_free_aligned_binary_bytes(byte* buf) # define CHICKEN_PAD (sizeof(void*) - 1) #endif +/* Caller must initialize 'refc' +*/ ERTS_GLB_INLINE Binary * erts_bin_drv_alloc_fnf(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; - void *res; + Binary *res; + + if (bsize < size) /* overflow */ + return NULL; res = erts_alloc_fnf(ERTS_ALC_T_DRV_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); - return (Binary *) res; + if (res) { + res->orig_size = size; + res->flags = BIN_FLAG_DRV; + } + return res; } +/* Caller must initialize 'refc' +*/ ERTS_GLB_INLINE Binary * erts_bin_drv_alloc(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; - void *res; + Binary *res; + + if (bsize < size) /* overflow */ + erts_alloc_enomem(ERTS_ALC_T_DRV_BINARY, size); res = erts_alloc(ERTS_ALC_T_DRV_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); - return (Binary *) res; + res->orig_size = size; + res->flags = BIN_FLAG_DRV; + return res; } +/* Caller must initialize 'refc' +*/ ERTS_GLB_INLINE Binary * erts_bin_nrml_alloc(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; - void *res; + Binary *res; + + if (bsize < size) /* overflow */ + erts_alloc_enomem(ERTS_ALC_T_BINARY, size); res = erts_alloc(ERTS_ALC_T_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); - return (Binary *) res; + res->orig_size = size; + res->flags = 0; + return res; } ERTS_GLB_INLINE Binary * @@ -267,12 +290,15 @@ erts_bin_realloc_fnf(Binary *bp, Uint size) { Binary *nbp; Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; + ErtsAlcType_t type = (bp->flags & BIN_FLAG_DRV) ? ERTS_ALC_T_DRV_BINARY + : ERTS_ALC_T_BINARY; ASSERT((bp->flags & BIN_FLAG_MAGIC) == 0); - if (bp->flags & BIN_FLAG_DRV) - nbp = erts_realloc_fnf(ERTS_ALC_T_DRV_BINARY, (void *) bp, bsize); - else - nbp = erts_realloc_fnf(ERTS_ALC_T_BINARY, (void *) bp, bsize); + if (bsize < size) /* overflow */ + return NULL; + nbp = erts_realloc_fnf(type, (void *) bp, bsize); ERTS_CHK_BIN_ALIGNMENT(nbp); + if (nbp) + nbp->orig_size = size; return nbp; } @@ -281,18 +307,16 @@ erts_bin_realloc(Binary *bp, Uint size) { Binary *nbp; Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; + ErtsAlcType_t type = (bp->flags & BIN_FLAG_DRV) ? ERTS_ALC_T_DRV_BINARY + : ERTS_ALC_T_BINARY; ASSERT((bp->flags & BIN_FLAG_MAGIC) == 0); - if (bp->flags & BIN_FLAG_DRV) - nbp = erts_realloc_fnf(ERTS_ALC_T_DRV_BINARY, (void *) bp, bsize); - else - nbp = erts_realloc_fnf(ERTS_ALC_T_BINARY, (void *) bp, bsize); + if (bsize < size) /* overflow */ + erts_realloc_enomem(type, bp, size); + nbp = erts_realloc_fnf(type, (void *) bp, bsize); if (!nbp) - erts_realloc_n_enomem(ERTS_ALC_T2N(bp->flags & BIN_FLAG_DRV - ? ERTS_ALC_T_DRV_BINARY - : ERTS_ALC_T_BINARY), - bp, - bsize); + erts_realloc_enomem(type, bp, bsize); ERTS_CHK_BIN_ALIGNMENT(nbp); + nbp->orig_size = size; return nbp; } @@ -312,6 +336,7 @@ erts_create_magic_binary(Uint size, void (*destructor)(Binary *)) { Uint bsize = ERTS_MAGIC_BIN_SIZE(size); Binary* bptr = erts_alloc_fnf(ERTS_ALC_T_BINARY, bsize); + ASSERT(bsize > size); if (!bptr) erts_alloc_n_enomem(ERTS_ALC_T2N(ERTS_ALC_T_BINARY), bsize); ERTS_CHK_BIN_ALIGNMENT(bptr); @@ -324,4 +349,4 @@ erts_create_magic_binary(Uint size, void (*destructor)(Binary *)) #endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */ -#endif +#endif /* !__ERL_BINARY_H */ diff --git a/erts/emulator/beam/erl_bits.c b/erts/emulator/beam/erl_bits.c index 73765772c8..5cc0a23dc9 100644 --- a/erts/emulator/beam/erl_bits.c +++ b/erts/emulator/beam/erl_bits.c @@ -403,7 +403,10 @@ erts_bs_get_integer_2(Process *p, Uint num_bits, unsigned flags, ErlBinMatchBuff words_needed = 1+WSIZE(bytes); hp = HeapOnlyAlloc(p, words_needed); res = bytes_to_big(LSB, bytes, sgn, hp); - if (is_small(res)) { + if (is_nil(res)) { + p->htop = hp; + res = THE_NON_VALUE; + } else if (is_small(res)) { p->htop = hp; } else if ((actual = bignum_header_arity(*hp)+1) < words_needed) { p->htop = hp + actual; @@ -1299,7 +1302,6 @@ erts_bs_append(Process* c_p, Eterm* reg, Uint live, Eterm build_size_term, if (binp->orig_size < pb->size) { Uint new_size = 2*pb->size; binp = erts_bin_realloc(binp, new_size); - binp->orig_size = new_size; pb->val = binp; pb->bytes = (byte *) binp->orig_bytes; } @@ -1371,8 +1373,6 @@ erts_bs_append(Process* c_p, Eterm* reg, Uint live, Eterm build_size_term, * Allocate the binary data struct itself. */ bptr = erts_bin_nrml_alloc(bin_size); - bptr->flags = 0; - bptr->orig_size = bin_size; erts_refc_init(&bptr->refc, 1); erts_current_bin = (byte *) bptr->orig_bytes; @@ -1475,7 +1475,6 @@ erts_bs_private_append(Process* p, Eterm bin, Eterm build_size_term, Uint unit) * is safe to reallocate it. */ binp = erts_bin_realloc(binp, new_size); - binp->orig_size = new_size; pb->val = binp; pb->bytes = (byte *) binp->orig_bytes; } else { @@ -1488,8 +1487,6 @@ erts_bs_private_append(Process* p, Eterm bin, Eterm build_size_term, Uint unit) * binary and copy the contents of the old binary into it. */ Binary* bptr = erts_bin_nrml_alloc(new_size); - bptr->flags = 0; - bptr->orig_size = new_size; erts_refc_init(&bptr->refc, 1); sys_memcpy(bptr->orig_bytes, binp->orig_bytes, binp->orig_size); pb->flags |= PB_IS_WRITABLE | PB_ACTIVE_WRITER; @@ -1537,8 +1534,6 @@ erts_bs_init_writable(Process* p, Eterm sz) * Allocate the binary data struct itself. */ bptr = erts_bin_nrml_alloc(bin_size); - bptr->flags = 0; - bptr->orig_size = bin_size; erts_refc_init(&bptr->refc, 1); /* @@ -1585,9 +1580,7 @@ erts_emasculate_writable_binary(ProcBin* pb) /* Our allocators are 8 byte aligned, i.e., shrinking with less than 8 bytes will have no real effect */ if (unused >= 8) { - Uint new_size = pb->size; binp = erts_bin_realloc(binp, pb->size); - binp->orig_size = new_size; pb->val = binp; pb->bytes = (byte *) binp->orig_bytes; } diff --git a/erts/emulator/beam/erl_db.c b/erts/emulator/beam/erl_db.c index a5d67571e2..fff892ae54 100644 --- a/erts/emulator/beam/erl_db.c +++ b/erts/emulator/beam/erl_db.c @@ -259,10 +259,11 @@ static void schedule_free_dbtable(DbTable* tb) /* * NON-SMP case: Caller is *not* allowed to access the *tb * structure after this function has returned! - * SMP case: Caller is allowed to access the *tb structure - * until the bif has returned (we typically - * need to unlock the table lock after this - * function has returned). + * SMP case: Caller is allowed to access the *common* part of the *tb + * structure until the bif has returned (we typically need to + * unlock the table lock after this function has returned). + * Caller is *not* allowed to access the specialized part + * (hash or tree) of *tb after this function has returned. */ ASSERT(erts_refc_read(&tb->common.ref, 0) == 0); erts_schedule_thr_prgr_later_cleanup_op(free_dbtable, @@ -752,6 +753,31 @@ BIF_RETTYPE ets_prev_2(BIF_ALIST_2) BIF_RET(ret); } +/* +** take(Tab, Key) +*/ +BIF_RETTYPE ets_take_2(BIF_ALIST_2) +{ + DbTable* tb; +#ifdef DEBUG + int cret; +#endif + Eterm ret; + CHECK_TABLES(); + + tb = db_get_table(BIF_P, BIF_ARG_1, DB_WRITE, LCK_WRITE_REC); + if (!tb) { + BIF_ERROR(BIF_P, BADARG); + } +#ifdef DEBUG + cret = +#endif + tb->common.meth->db_take(BIF_P, tb, BIF_ARG_2, &ret); + ASSERT(cret == DB_ERROR_NONE); + db_unlock(tb, LCK_WRITE_REC); + BIF_RET(ret); +} + /* ** update_element(Tab, Key, {Pos, Value}) ** update_element(Tab, Key, [{Pos, Value}]) @@ -779,7 +805,7 @@ BIF_RETTYPE ets_update_element_3(BIF_ALIST_3) list = BIF_ARG_3; } - if (!tb->common.meth->db_lookup_dbterm(tb, BIF_ARG_2, &handle)) { + if (!tb->common.meth->db_lookup_dbterm(BIF_P, tb, BIF_ARG_2, THE_NON_VALUE, &handle)) { cret = DB_ERROR_BADKEY; goto bail_out; } @@ -818,7 +844,7 @@ BIF_RETTYPE ets_update_element_3(BIF_ALIST_3) } finalize: - tb->common.meth->db_finalize_dbterm(&handle); + tb->common.meth->db_finalize_dbterm(cret, &handle); bail_out: UnUseTmpHeap(2,BIF_P); @@ -837,14 +863,8 @@ bail_out: } } -/* -** update_counter(Tab, Key, Incr) -** update_counter(Tab, Key, {Upop}) -** update_counter(Tab, Key, [{Upop}]) -** Upop = {Pos,Incr} | {Pos,Incr,Threshold,WarpTo} -** Returns new value(s) (integer or [integer]) -*/ -BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) +static BIF_RETTYPE +do_update_counter(Process *p, Eterm arg1, Eterm arg2, Eterm arg3, Eterm arg4) { DbTable* tb; int cret = DB_ERROR_BADITEM; @@ -854,7 +874,7 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) Eterm* ret_list_currp = NULL; Eterm* ret_list_prevp = NULL; Eterm iter; - DeclareTmpHeap(cell,5,BIF_P); + DeclareTmpHeap(cell, 5, p); Eterm *tuple = cell+2; DbUpdateHandle handle; Uint halloc_size = 0; /* overestimated heap usage */ @@ -862,28 +882,29 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) Eterm* hstart; Eterm* hend; - if ((tb = db_get_table(BIF_P, BIF_ARG_1, DB_WRITE, LCK_WRITE_REC)) == NULL) { - BIF_ERROR(BIF_P, BADARG); + if ((tb = db_get_table(p, arg1, DB_WRITE, LCK_WRITE_REC)) == NULL) { + BIF_ERROR(p, BADARG); } - UseTmpHeap(5,BIF_P); + UseTmpHeap(5, p); if (!(tb->common.status & (DB_SET | DB_ORDERED_SET))) { goto bail_out; } - if (is_integer(BIF_ARG_3)) { /* Incr */ - upop_list = CONS(cell, TUPLE2(tuple, make_small(tb->common.keypos+1), - BIF_ARG_3), NIL); + if (is_integer(arg3)) { /* Incr */ + upop_list = CONS(cell, + TUPLE2(tuple, make_small(tb->common.keypos+1), arg3), + NIL); } - else if (is_tuple(BIF_ARG_3)) { /* {Upop} */ - upop_list = CONS(cell, BIF_ARG_3, NIL); + else if (is_tuple(arg3)) { /* {Upop} */ + upop_list = CONS(cell, arg3, NIL); } else { /* [{Upop}] (probably) */ - upop_list = BIF_ARG_3; + upop_list = arg3; ret_list_prevp = &ret; } - if (!tb->common.meth->db_lookup_dbterm(tb, BIF_ARG_2, &handle)) { + if (!tb->common.meth->db_lookup_dbterm(p, tb, arg2, arg4, &handle)) { goto bail_out; /* key not found */ } @@ -956,13 +977,13 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) if (ret_list_prevp) { /* Prepare to return a list */ ret = NIL; halloc_size += list_size; - hstart = HAlloc(BIF_P, halloc_size); + hstart = HAlloc(p, halloc_size); ret_list_currp = hstart; htop = hstart + list_size; hend = hstart + halloc_size; } else { - hstart = htop = HAlloc(BIF_P, halloc_size); + hstart = htop = HAlloc(p, halloc_size); } hend = hstart + halloc_size; @@ -1009,26 +1030,54 @@ BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) (is_list(ret) && (list_val(ret)+list_size)==ret_list_currp)); ASSERT(htop <= hend); - HRelease(BIF_P,hend,htop); + HRelease(p, hend, htop); finalize: - tb->common.meth->db_finalize_dbterm(&handle); + tb->common.meth->db_finalize_dbterm(cret, &handle); bail_out: - UnUseTmpHeap(5,BIF_P); + UnUseTmpHeap(5, p); db_unlock(tb, LCK_WRITE_REC); switch (cret) { case DB_ERROR_NONE: BIF_RET(ret); case DB_ERROR_SYSRES: - BIF_ERROR(BIF_P, SYSTEM_LIMIT); + BIF_ERROR(p, SYSTEM_LIMIT); default: - BIF_ERROR(BIF_P, BADARG); + BIF_ERROR(p, BADARG); break; } } +/* +** update_counter(Tab, Key, Incr) +** update_counter(Tab, Key, Upop) +** update_counter(Tab, Key, [{Upop}]) +** Upop = {Pos,Incr} | {Pos,Incr,Threshold,WarpTo} +** Returns new value(s) (integer or [integer]) +*/ +BIF_RETTYPE ets_update_counter_3(BIF_ALIST_3) +{ + return do_update_counter(BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3, THE_NON_VALUE); +} + +/* +** update_counter(Tab, Key, Incr, Default) +** update_counter(Tab, Key, Upop, Default) +** update_counter(Tab, Key, [{Upop}], Default) +** Upop = {Pos,Incr} | {Pos,Incr,Threshold,WarpTo} +** Returns new value(s) (integer or [integer]) +*/ +BIF_RETTYPE ets_update_counter_4(BIF_ALIST_4) +{ + if (is_not_tuple(BIF_ARG_4)) { + BIF_ERROR(BIF_P, BADARG); + } + return do_update_counter(BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3, BIF_ARG_4); +} + + /* ** The put BIF */ @@ -2642,7 +2691,9 @@ BIF_RETTYPE ets_match_object_3(BIF_ALIST_3) BIF_RETTYPE ets_info_1(BIF_ALIST_1) { static Eterm fields[] = {am_protection, am_keypos, am_type, am_named_table, - am_node, am_size, am_name, am_heir, am_owner, am_memory, am_compressed}; + am_node, am_size, am_name, am_heir, am_owner, am_memory, am_compressed, + am_write_concurrency, + am_read_concurrency}; Eterm results[sizeof(fields)/sizeof(Eterm)]; DbTable* tb; Eterm res; @@ -3279,34 +3330,37 @@ erts_db_process_exiting(Process *c_p, ErtsProcLocks c_p_locks) } erts_smp_rwmtx_runlock(mmtl); if (tb) { - int reds; - DbFixation** pp; + int reds = 0; db_lock(tb, LCK_WRITE_REC); - #ifdef ERTS_SMP - erts_smp_mtx_lock(&tb->common.fixlock); - #endif - reds = 10; - - for (pp = &tb->common.fixations; *pp != NULL; - pp = &(*pp)->next) { - if ((*pp)->pid == pid) { - DbFixation* fix = *pp; - erts_aint_t diff = -((erts_aint_t) fix->counter); - erts_refc_add(&tb->common.ref,diff,0); - *pp = fix->next; - erts_db_free(ERTS_ALC_T_DB_FIXATION, - tb, fix, sizeof(DbFixation)); - ERTS_ETS_MISC_MEM_ADD(-sizeof(DbFixation)); - break; + if (!(tb->common.status & DB_DELETE)) { + DbFixation** pp; + + #ifdef ERTS_SMP + erts_smp_mtx_lock(&tb->common.fixlock); + #endif + reds = 10; + + for (pp = &tb->common.fixations; *pp != NULL; + pp = &(*pp)->next) { + if ((*pp)->pid == pid) { + DbFixation* fix = *pp; + erts_aint_t diff = -((erts_aint_t) fix->counter); + erts_refc_add(&tb->common.ref,diff,0); + *pp = fix->next; + erts_db_free(ERTS_ALC_T_DB_FIXATION, + tb, fix, sizeof(DbFixation)); + ERTS_ETS_MISC_MEM_ADD(-sizeof(DbFixation)); + break; + } + } + #ifdef ERTS_SMP + erts_smp_mtx_unlock(&tb->common.fixlock); + #endif + if (!IS_FIXED(tb) && IS_HASH_TABLE(tb->common.status)) { + db_unfix_table_hash(&(tb->hash)); + reds += 40; } - } - #ifdef ERTS_SMP - erts_smp_mtx_unlock(&tb->common.fixlock); - #endif - if (!IS_FIXED(tb) && IS_HASH_TABLE(tb->common.status)) { - db_unfix_table_hash(&(tb->hash)); - reds += 40; } db_unlock(tb, LCK_WRITE_REC); BUMP_REDS(c_p, reds); @@ -3666,6 +3720,10 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What) ret = am_protected; else if (tb->common.status & DB_PUBLIC) ret = am_public; + } else if (What == am_write_concurrency) { + ret = tb->common.status & DB_FINE_LOCKED ? am_true : am_false; + } else if (What == am_read_concurrency) { + ret = tb->common.status & DB_FREQ_READ ? am_true : am_false; } else if (What == am_name) { ret = tb->common.the_name; } else if (What == am_keypos) { @@ -3748,7 +3806,7 @@ static Eterm table_info(Process* p, DbTable* tb, Eterm What) avg, std_dev_real, std_dev_exp, make_small(stats.min_chain_len), make_small(stats.max_chain_len), - make_small(db_kept_items_hash(&tb->hash))); + make_small(stats.kept_items)); } else { ret = am_false; @@ -3770,6 +3828,11 @@ static void print_table(int to, void *to_arg, int show, DbTable* tb) + sizeof(Uint) - 1) / sizeof(Uint))); + erts_print(to, to_arg, "Type: %T\n", table_info(NULL, tb, am_type)); + erts_print(to, to_arg, "Protection: %T\n", table_info(NULL, tb, am_protection)); + erts_print(to, to_arg, "Compressed: %T\n", table_info(NULL, tb, am_compressed)); + erts_print(to, to_arg, "Write Concurrency: %T\n", table_info(NULL, tb, am_write_concurrency)); + erts_print(to, to_arg, "Read Concurrency: %T\n", table_info(NULL, tb, am_read_concurrency)); } void db_info(int to, void *to_arg, int show) /* Called by break handler */ diff --git a/erts/emulator/beam/erl_db_hash.c b/erts/emulator/beam/erl_db_hash.c index 06dac8f161..045c8ae135 100644 --- a/erts/emulator/beam/erl_db_hash.c +++ b/erts/emulator/beam/erl_db_hash.c @@ -174,7 +174,7 @@ static ERTS_INLINE void add_fixed_deletion(DbTableHash* tb, int ix) /* optimised version of make_hash (normal case? atomic key) */ #define MAKE_HASH(term) \ ((is_atom(term) ? (atom_tab(atom_val(term))->slot.bucket.hvalue) : \ - make_hash2(term)) % MAX_HASH) + make_internal_hash(term)) % MAX_HASH) #ifdef ERTS_SMP # define DB_HASH_LOCK_MASK (DB_HASH_LOCK_CNT-1) @@ -382,7 +382,7 @@ static HashDbTerm* search_list(DbTableHash* tb, Eterm key, static void shrink(DbTableHash* tb, int nactive); static void grow(DbTableHash* tb, int nactive); static Eterm build_term_list(Process* p, HashDbTerm* ptr1, HashDbTerm* ptr2, - DbTableHash*); + Uint sz, DbTableHash*); static int analyze_pattern(DbTableHash *tb, Eterm pattern, struct mp_info *mpi); @@ -426,6 +426,7 @@ static int db_select_count_continue_hash(Process *p, DbTable *tbl, static int db_select_delete_continue_hash(Process *p, DbTable *tbl, Eterm continuation, Eterm *ret); +static int db_take_hash(Process *, DbTable *, Eterm, Eterm *); static void db_print_hash(int to, void *to_arg, int show, @@ -443,8 +444,11 @@ static int db_delete_all_objects_hash(Process* p, DbTable* tbl); #ifdef HARDDEBUG static void db_check_table_hash(DbTableHash *tb); #endif -static int db_lookup_dbterm_hash(DbTable *tbl, Eterm key, DbUpdateHandle* handle); -static void db_finalize_dbterm_hash(DbUpdateHandle* handle); +static int +db_lookup_dbterm_hash(Process *p, DbTable *tbl, Eterm key, Eterm obj, + DbUpdateHandle* handle); +static void +db_finalize_dbterm_hash(int cret, DbUpdateHandle* handle); static ERTS_INLINE void try_shrink(DbTableHash* tb) { @@ -536,6 +540,7 @@ DbTableMethod db_hash = db_select_delete_continue_hash, db_select_count_hash, db_select_count_continue_hash, + db_take_hash, db_delete_all_objects_hash, db_free_table_hash, db_free_table_continue_hash, @@ -646,25 +651,6 @@ restart: /* ToDo: Maybe try grow/shrink the table as well */ } -/* Only used by tests -*/ -Uint db_kept_items_hash(DbTableHash *tb) -{ - Uint kept_items = 0; - Uint ix = 0; - erts_smp_rwmtx_t* lck = RLOCK_HASH(tb,ix); - HashDbTerm* b; - do { - for (b = BUCKET(tb, ix); b != NULL; b = b->next) { - if (b->hvalue == INVALID_HASH) { - ++kept_items; - } - } - ix = next_slot(tb, ix, &lck); - }while (ix); - return kept_items; -} - int db_create_hash(Process *p, DbTable *tbl) { DbTableHash *tb = &tbl->hash; @@ -879,34 +865,49 @@ Ldone: return ret; } +static Eterm +get_term_list(Process *p, DbTableHash *tb, Eterm key, HashValue hval, + HashDbTerm *b1, HashDbTerm **bend) +{ + HashDbTerm* b2 = b1->next; + Eterm copy; + Uint sz = b1->dbterm.size + 2; + + if (tb->common.status & (DB_BAG | DB_DUPLICATE_BAG)) { + while (b2 && has_key(tb, b2, key, hval)) { + if (b2->hvalue != INVALID_HASH) + sz += b2->dbterm.size + 2; + + b2 = b2->next; + } + } + copy = build_term_list(p, b1, b2, sz, tb); + CHECK_TABLES(); + if (bend) { + *bend = b2; + } + return copy; +} + int db_get_hash(Process *p, DbTable *tbl, Eterm key, Eterm *ret) { DbTableHash *tb = &tbl->hash; HashValue hval; int ix; - HashDbTerm* b1; + HashDbTerm* b; erts_smp_rwmtx_t* lck; hval = MAKE_HASH(key); lck = RLOCK_HASH(tb,hval); ix = hash_to_ix(tb, hval); - b1 = BUCKET(tb, ix); - - while(b1 != 0) { - if (has_live_key(tb,b1,key,hval)) { - HashDbTerm* b2 = b1->next; - Eterm copy; + b = BUCKET(tb, ix); - if (tb->common.status & (DB_BAG | DB_DUPLICATE_BAG)) { - while(b2 != NULL && has_key(tb,b2,key,hval)) - b2 = b2->next; - } - copy = build_term_list(p, b1, b2, tb); - CHECK_TABLES(); - *ret = copy; + while(b != 0) { + if (has_live_key(tb, b, key, hval)) { + *ret = get_term_list(p, tb, key, hval, b, NULL); goto done; } - b1 = b1->next; + b = b->next; } *ret = NIL; done: @@ -1240,7 +1241,7 @@ static int db_slot_hash(Process *p, DbTable *tbl, Eterm slot_term, Eterm *ret) lck = RLOCK_HASH(tb, slot); nactive = NACTIVE(tb); if (slot < nactive) { - *ret = build_term_list(p, BUCKET(tb, slot), 0, tb); + *ret = build_term_list(p, BUCKET(tb, slot), NULL, 0, tb); retval = DB_ERROR_NONE; } else if (slot == nactive) { @@ -2069,6 +2070,46 @@ trap: } +static int db_take_hash(Process *p, DbTable *tbl, Eterm key, Eterm *ret) +{ + DbTableHash *tb = &tbl->hash; + HashDbTerm **bp, *b; + HashValue hval = MAKE_HASH(key); + erts_smp_rwmtx_t *lck = WLOCK_HASH(tb, hval); + int ix = hash_to_ix(tb, hval); + int nitems_diff = 0; + + *ret = NIL; + for (bp = &BUCKET(tb, ix), b = *bp; b; bp = &b->next, b = b->next) { + if (has_live_key(tb, b, key, hval)) { + HashDbTerm *bend; + + *ret = get_term_list(p, tb, key, hval, b, &bend); + while (b != bend) { + --nitems_diff; + if (nitems_diff == -1 && IS_FIXED(tb)) { + /* Pseudo remove (no need to keep several of same key) */ + add_fixed_deletion(tb, ix); + bp = &b->next; + b->hvalue = INVALID_HASH; + b = b->next; + } else { + *bp = b->next; + free_term(tb, b); + b = *bp; + } + } + break; + } + } + WUNLOCK_HASH(lck); + if (nitems_diff) { + erts_smp_atomic_add_nob(&tb->common.nitems, nitems_diff); + try_shrink(tb); + } + return DB_ERROR_NONE; +} + /* ** Other interface routines (not directly coupled to one bif) */ @@ -2104,10 +2145,38 @@ int db_mark_all_deleted_hash(DbTable *tbl) static void db_print_hash(int to, void *to_arg, int show, DbTable *tbl) { DbTableHash *tb = &tbl->hash; + DbHashStats stats; int i; erts_print(to, to_arg, "Buckets: %d\n", NACTIVE(tb)); - + +#ifdef ERTS_SMP + i = tbl->common.is_thread_safe; + /* If crash dumping we set table to thread safe in order to + avoid taking any locks */ + if (ERTS_IS_CRASH_DUMPING) + tbl->common.is_thread_safe = 1; + + db_calc_stats_hash(&tbl->hash, &stats); + + tbl->common.is_thread_safe = i; +#else + db_calc_stats_hash(&tbl->hash, &stats); +#endif + + erts_print(to, to_arg, "Chain Length Avg: %f\n", stats.avg_chain_len); + erts_print(to, to_arg, "Chain Length Max: %d\n", stats.max_chain_len); + erts_print(to, to_arg, "Chain Length Min: %d\n", stats.min_chain_len); + erts_print(to, to_arg, "Chain Length Std Dev: %f\n", + stats.std_dev_chain_len); + erts_print(to, to_arg, "Chain Length Expected Std Dev: %f\n", + stats.std_dev_expected); + + if (IS_FIXED(tb)) + erts_print(to, to_arg, "Fixed: %d\n", stats.kept_items); + else + erts_print(to, to_arg, "Fixed: false\n"); + if (show) { for (i = 0; i < NACTIVE(tb); i++) { HashDbTerm* list = BUCKET(tb,i); @@ -2483,23 +2552,23 @@ static int free_seg(DbTableHash *tb, int free_records) ** Copy terms from ptr1 until ptr2 ** works for ptr1 == ptr2 == 0 => [] ** or ptr2 == 0 +** sz is either precalculated heap size or 0 if not known */ static Eterm build_term_list(Process* p, HashDbTerm* ptr1, HashDbTerm* ptr2, - DbTableHash* tb) + Uint sz, DbTableHash* tb) { - int sz = 0; HashDbTerm* ptr; Eterm list = NIL; Eterm copy; Eterm *hp, *hend; - ptr = ptr1; - while(ptr != ptr2) { - - if (ptr->hvalue != INVALID_HASH) - sz += ptr->dbterm.size + 2; - - ptr = ptr->next; + if (!sz) { + ptr = ptr1; + while(ptr != ptr2) { + if (ptr->hvalue != INVALID_HASH) + sz += ptr->dbterm.size + 2; + ptr = ptr->next; + } } hp = HAlloc(p, sz); @@ -2730,59 +2799,129 @@ static HashDbTerm* next(DbTableHash *tb, Uint *iptr, erts_smp_rwmtx_t** lck_ptr, return NULL; } -static int db_lookup_dbterm_hash(DbTable *tbl, Eterm key, DbUpdateHandle* handle) +static int +db_lookup_dbterm_hash(Process *p, DbTable *tbl, Eterm key, Eterm obj, + DbUpdateHandle* handle) { DbTableHash *tb = &tbl->hash; - HashDbTerm* b; - HashDbTerm** prevp; - int ix; HashValue hval; + HashDbTerm **bp, *b; erts_smp_rwmtx_t* lck; + int flags = 0; + + ASSERT(tb->common.status & DB_SET); hval = MAKE_HASH(key); - lck = WLOCK_HASH(tb,hval); - ix = hash_to_ix(tb, hval); - prevp = &BUCKET(tb, ix); - b = *prevp; + lck = WLOCK_HASH(tb, hval); + bp = &BUCKET(tb, hash_to_ix(tb, hval)); + b = *bp; - while (b != 0) { - if (has_live_key(tb,b,key,hval)) { - handle->tb = tbl; - handle->bp = (void**) prevp; - handle->dbterm = &b->dbterm; - handle->mustResize = 0; - handle->new_size = b->dbterm.size; - #if HALFWORD_HEAP - handle->abs_vec = NULL; - #endif - handle->lck = lck; - /* KEEP hval WLOCKED, db_finalize_dbterm_hash will WUNLOCK */ - return 1; - } - prevp = &b->next; - b = *prevp; + for (;;) { + if (b == NULL) { + break; + } + if (has_key(tb, b, key, hval)) { + if (b->hvalue != INVALID_HASH) { + goto Ldone; + } + break; + } + bp = &b->next; + b = *bp; } - WUNLOCK_HASH(lck); - return 0; + + if (obj == THE_NON_VALUE) { + WUNLOCK_HASH(lck); + return 0; + } + + { + Eterm *objp = tuple_val(obj); + int arity = arityval(*objp); + Eterm *htop, *hend; + + ASSERT(arity >= tb->common.keypos); + htop = HAlloc(p, arity + 1); + hend = htop + arity + 1; + sys_memcpy(htop, objp, sizeof(Eterm) * (arity + 1)); + htop[tb->common.keypos] = key; + obj = make_tuple(htop); + + if (b == NULL) { + HashDbTerm *q = new_dbterm(tb, obj); + + q->hvalue = hval; + q->next = NULL; + *bp = b = q; + + { + int nitems = erts_smp_atomic_inc_read_nob(&tb->common.nitems); + int nactive = NACTIVE(tb); + + if (nitems > nactive * (CHAIN_LEN + 1) && !IS_FIXED(tb)) { + grow(tb, nactive); + } + } + } else { + HashDbTerm *q, *next = b->next; + + ASSERT(b->hvalue == INVALID_HASH); + q = replace_dbterm(tb, b, obj); + q->next = next; + q->hvalue = hval; + *bp = b = q; + erts_smp_atomic_inc_nob(&tb->common.nitems); + } + + HRelease(p, hend, htop); + flags |= DB_NEW_OBJECT; + } + +Ldone: + handle->tb = tbl; + handle->bp = (void **)bp; + handle->dbterm = &b->dbterm; + handle->flags = flags; + handle->new_size = b->dbterm.size; +#if HALFWORD_HEAP + handle->abs_vec = NULL; +#endif + handle->lck = lck; + return 1; } /* Must be called after call to db_lookup_dbterm */ -static void db_finalize_dbterm_hash(DbUpdateHandle* handle) +static void +db_finalize_dbterm_hash(int cret, DbUpdateHandle* handle) { DbTable* tbl = handle->tb; - HashDbTerm* oldp = (HashDbTerm*) *(handle->bp); + DbTableHash *tb = &tbl->hash; + HashDbTerm **bp = (HashDbTerm **) handle->bp; + HashDbTerm *b = *bp; erts_smp_rwmtx_t* lck = (erts_smp_rwmtx_t*) handle->lck; - ERTS_SMP_LC_ASSERT(IS_HASH_WLOCKED(&tbl->hash,lck)); /* locked by db_lookup_dbterm_hash */ + ERTS_SMP_LC_ASSERT(IS_HASH_WLOCKED(tb, lck)); /* locked by db_lookup_dbterm_hash */ - ASSERT((&oldp->dbterm == handle->dbterm) == !(tbl->common.compress && handle->mustResize)); + ASSERT((&b->dbterm == handle->dbterm) == !(tb->common.compress && handle->flags & DB_MUST_RESIZE)); + + if (handle->flags & DB_NEW_OBJECT && cret != DB_ERROR_NONE) { + if (IS_FIXED(tb)) { + add_fixed_deletion(tb, hash_to_ix(tb, b->hvalue)); + b->hvalue = INVALID_HASH; + } else { + *bp = b->next; + free_term(tb, b); + } - if (handle->mustResize) { + WUNLOCK_HASH(lck); + erts_smp_atomic_dec_nob(&tb->common.nitems); + try_shrink(tb); + } else if (handle->flags & DB_MUST_RESIZE) { db_finalize_resize(handle, offsetof(HashDbTerm,dbterm)); WUNLOCK_HASH(lck); - free_term(&tbl->hash, oldp); + free_term(tb, b); } else { WUNLOCK_HASH(lck); @@ -2833,6 +2972,7 @@ void db_calc_stats_hash(DbTableHash* tb, DbHashStats* stats) erts_smp_rwmtx_t* lck; int sum = 0; int sq_sum = 0; + int kept_items = 0; int ix; int len; @@ -2844,6 +2984,8 @@ void db_calc_stats_hash(DbTableHash* tb, DbHashStats* stats) len = 0; for (b = BUCKET(tb,ix); b!=NULL; b=b->next) { len++; + if (b->hvalue == INVALID_HASH) + ++kept_items; } sum += len; sq_sum += len*len; @@ -2855,7 +2997,8 @@ void db_calc_stats_hash(DbTableHash* tb, DbHashStats* stats) stats->std_dev_chain_len = sqrt((sq_sum - stats->avg_chain_len*sum) / NACTIVE(tb)); /* Expected standard deviation from a good uniform hash function, ie binomial distribution (not taking the linear hashing into acount) */ - stats->std_dev_expected = sqrt(stats->avg_chain_len * (1 - 1.0/NACTIVE(tb))); + stats->std_dev_expected = sqrt(stats->avg_chain_len * (1 - 1.0/NACTIVE(tb))); + stats->kept_items = kept_items; } #ifdef HARDDEBUG diff --git a/erts/emulator/beam/erl_db_hash.h b/erts/emulator/beam/erl_db_hash.h index 908cec11d4..f12cd363b0 100644 --- a/erts/emulator/beam/erl_db_hash.h +++ b/erts/emulator/beam/erl_db_hash.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1998-2013. All Rights Reserved. + * Copyright Ericsson AB 1998-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -42,7 +42,7 @@ typedef struct hash_db_term { typedef struct db_table_hash_fine_locks { union { erts_smp_rwmtx_t lck; - byte _cache_line_alignment[64]; + byte _cache_line_alignment[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(erts_smp_rwmtx_t))]; }lck_vec[DB_HASH_LOCK_CNT]; } DbTableHashFineLocks; @@ -104,6 +104,7 @@ typedef struct { float std_dev_expected; int max_chain_len; int min_chain_len; + int kept_items; }DbHashStats; void db_calc_stats_hash(DbTableHash* tb, DbHashStats*); diff --git a/erts/emulator/beam/erl_db_tree.c b/erts/emulator/beam/erl_db_tree.c index a62a83a928..577da35b75 100644 --- a/erts/emulator/beam/erl_db_tree.c +++ b/erts/emulator/beam/erl_db_tree.c @@ -383,6 +383,7 @@ static int db_select_delete_tree(Process *p, DbTable *tbl, Eterm pattern, Eterm *ret); static int db_select_delete_continue_tree(Process *p, DbTable *tbl, Eterm continuation, Eterm *ret); +static int db_take_tree(Process *, DbTable *, Eterm, Eterm *); static void db_print_tree(int to, void *to_arg, int show, DbTable *tbl); static int db_free_table_tree(DbTable *tbl); @@ -398,8 +399,11 @@ static int db_delete_all_objects_tree(Process* p, DbTable* tbl); #ifdef HARDDEBUG static void db_check_table_tree(DbTable *tbl); #endif -static int db_lookup_dbterm_tree(DbTable *, Eterm key, DbUpdateHandle*); -static void db_finalize_dbterm_tree(DbUpdateHandle*); +static int +db_lookup_dbterm_tree(Process *, DbTable *, Eterm key, Eterm obj, + DbUpdateHandle*); +static void +db_finalize_dbterm_tree(int cret, DbUpdateHandle *); /* ** Static variables @@ -431,6 +435,7 @@ DbTableMethod db_tree = db_select_delete_continue_tree, db_select_count_tree, db_select_count_continue_tree, + db_take_tree, db_delete_all_objects_tree, db_free_table_tree, db_free_table_continue_tree, @@ -1722,6 +1727,28 @@ static int db_select_delete_tree(Process *p, DbTable *tbl, } +static int db_take_tree(Process *p, DbTable *tbl, Eterm key, Eterm *ret) +{ + DbTableTree *tb = &tbl->tree; + TreeDbTerm *this; + + *ret = NIL; + this = linkout_tree(tb, key, NULL); + if (this) { + Eterm copy, *hp, *hend; + + hp = HAlloc(p, this->dbterm.size + 2); + hend = hp + this->dbterm.size + 2; + copy = db_copy_object_from_ets(&tb->common, + &this->dbterm, &hp, &MSO(p)); + *ret = CONS(hp, copy, NIL); + hp += 2; + HRelease(p, hend, hp); + free_term(tb, this); + } + return DB_ERROR_NONE; +} + /* ** Other interface routines (not directly coupled to one bif) */ @@ -2522,16 +2549,43 @@ static TreeDbTerm **find_node2(DbTableTree *tb, Eterm key) return this; } -static int db_lookup_dbterm_tree(DbTable *tbl, Eterm key, DbUpdateHandle* handle) +static int +db_lookup_dbterm_tree(Process *p, DbTable *tbl, Eterm key, Eterm obj, + DbUpdateHandle* handle) { DbTableTree *tb = &tbl->tree; TreeDbTerm **pp = find_node2(tb, key); - - if (pp == NULL) return 0; + int flags = 0; + + if (pp == NULL) { + if (obj == THE_NON_VALUE) { + return 0; + } else { + Eterm *objp = tuple_val(obj); + int arity = arityval(*objp); + Eterm *htop, *hend; + + ASSERT(arity >= tb->common.keypos); + htop = HAlloc(p, arity + 1); + hend = htop + arity + 1; + sys_memcpy(htop, objp, sizeof(Eterm) * (arity + 1)); + htop[tb->common.keypos] = key; + obj = make_tuple(htop); + + if (db_put_tree(tbl, obj, 1) != DB_ERROR_NONE) { + return 0; + } + + pp = find_node2(tb, key); + ASSERT(pp != NULL); + HRelease(p, hend, htop); + flags |= DB_NEW_OBJECT; + } + } handle->tb = tbl; handle->dbterm = &(*pp)->dbterm; - handle->mustResize = 0; + handle->flags = flags; handle->bp = (void**) pp; handle->new_size = (*pp)->dbterm.size; #if HALFWORD_HEAP @@ -2540,15 +2594,21 @@ static int db_lookup_dbterm_tree(DbTable *tbl, Eterm key, DbUpdateHandle* handle return 1; } -static void db_finalize_dbterm_tree(DbUpdateHandle* handle) +static void +db_finalize_dbterm_tree(int cret, DbUpdateHandle *handle) { - if (handle->mustResize) { - TreeDbTerm* oldp = (TreeDbTerm*) *handle->bp; + DbTable *tbl = handle->tb; + DbTableTree *tb = &tbl->tree; + TreeDbTerm *bp = (TreeDbTerm *) *handle->bp; + if (handle->flags & DB_NEW_OBJECT && cret != DB_ERROR_NONE) { + Eterm ret; + db_erase_tree(tbl, GETKEY(tb, bp->dbterm.tpl), &ret); + } else if (handle->flags & DB_MUST_RESIZE) { db_finalize_resize(handle, offsetof(TreeDbTerm,dbterm)); - reset_static_stack(&handle->tb->tree); + reset_static_stack(tb); - free_term(&handle->tb->tree, oldp); + free_term(tb, bp); } #ifdef DEBUG handle->dbterm = 0; diff --git a/erts/emulator/beam/erl_db_util.c b/erts/emulator/beam/erl_db_util.c index 3927615e04..0bf562d937 100644 --- a/erts/emulator/beam/erl_db_util.c +++ b/erts/emulator/beam/erl_db_util.c @@ -198,11 +198,6 @@ set_match_trace(Process *tracee_p, Eterm fail_term, Eterm tracer, return ret; } - -/* Type checking... */ - -#define BOXED_IS_TUPLE(Boxed) is_arity_value(*boxed_val((Boxed))) - /* ** ** Types and enum's (compiled matches) @@ -218,7 +213,9 @@ typedef enum { matchTuple, matchPushT, matchPushL, + matchPushM, matchPop, + matchSwap, matchBind, matchCmp, matchEqBin, @@ -227,11 +224,15 @@ typedef enum { matchEqRef, matchEq, matchList, + matchMap, + matchKey, matchSkip, matchPushC, matchConsA, /* Car is below Cdr */ matchConsB, /* Cdr is below Car (unusual) */ matchMkTuple, + matchMkFlatMap, + matchMkHashMap, matchCall0, matchCall1, matchCall2, @@ -856,6 +857,13 @@ static int match_compact(ErlHeapFragment *expr, DMCErrInfo *err_info); static Uint my_size_object(Eterm t); static Eterm my_copy_struct(Eterm t, Eterm **hp, ErlOffHeap* off_heap); +/* Guard subroutines */ +static void +dmc_rearrange_constants(DMCContext *context, DMC_STACK_TYPE(UWord) *text, + int textpos, Eterm *p, Uint nelems); +static DMCRet +dmc_array(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm *p, Uint nelems, int *constant); /* Guard compilation */ static void do_emit_constant(DMCContext *context, DMC_STACK_TYPE(UWord) *text, Eterm t); @@ -869,6 +877,9 @@ static DMCRet dmc_tuple(DMCContext *context, DMC_STACK_TYPE(UWord) *text, Eterm t, int *constant); +static DMCRet +dmc_map(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm t, int *constant); static DMCRet dmc_variable(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, @@ -888,12 +899,14 @@ static DMCRet compile_guard_expr(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, Eterm t); -/* match expression subroutine */ +/* match expression subroutines */ static DMCRet dmc_one_term(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(Eterm) *stack, DMC_STACK_TYPE(UWord) *text, Eterm c); +static Eterm +dmc_private_copy(DMCContext *context, Eterm c); #ifdef DMC_DEBUG @@ -1364,7 +1377,112 @@ restart: for (;;) { switch (t & _TAG_PRIMARY_MASK) { case TAG_PRIMARY_BOXED: - if (!BOXED_IS_TUPLE(t)) { + if (is_flatmap(t)) { + num_iters = flatmap_get_size(flatmap_val(t)); + if (!structure_checked) { + DMC_PUSH(text, matchMap); + DMC_PUSH(text, num_iters); + } + structure_checked = 0; + for (i = 0; i < num_iters; ++i) { + Eterm key = flatmap_get_keys(flatmap_val(t))[i]; + if (db_is_variable(key) >= 0) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Variable found in map key.", + -1, 0UL, dmcError); + } + goto error; + } else if (key == am_Underscore) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Underscore found in map key.", + -1, 0UL, dmcError); + } + goto error; + } + DMC_PUSH(text, matchKey); + DMC_PUSH(text, dmc_private_copy(&context, key)); + { + int old_stack = ++(context.stack_used); + Eterm value = flatmap_get_values(flatmap_val(t))[i]; + res = dmc_one_term(&context, &heap, &stack, &text, + value); + ASSERT(res != retFail); + if (res == retRestart) { + goto restart; + } + if (old_stack != context.stack_used) { + ASSERT(old_stack + 1 == context.stack_used); + DMC_PUSH(text, matchSwap); + } + if (context.stack_used > context.stack_need) { + context.stack_need = context.stack_used; + } + DMC_PUSH(text, matchPop); + --(context.stack_used); + } + } + break; + } + if (is_hashmap(t)) { + DECLARE_WSTACK(wstack); + Eterm *kv; + num_iters = hashmap_size(t); + if (!structure_checked) { + DMC_PUSH(text, matchMap); + DMC_PUSH(text, num_iters); + } + structure_checked = 0; + + hashmap_iterator_init(&wstack, t, 0); + + while ((kv=hashmap_iterator_next(&wstack)) != NULL) { + Eterm key = CAR(kv); + Eterm value = CDR(kv); + if (db_is_variable(key) >= 0) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Variable found in map key.", + -1, 0UL, dmcError); + } + DESTROY_WSTACK(wstack); + goto error; + } else if (key == am_Underscore) { + if (context.err_info) { + add_dmc_err(context.err_info, + "Underscore found in map key.", + -1, 0UL, dmcError); + } + DESTROY_WSTACK(wstack); + goto error; + } + DMC_PUSH(text, matchKey); + DMC_PUSH(text, dmc_private_copy(&context, key)); + { + int old_stack = ++(context.stack_used); + res = dmc_one_term(&context, &heap, &stack, &text, + value); + ASSERT(res != retFail); + if (res == retRestart) { + DESTROY_WSTACK(wstack); + goto restart; + } + if (old_stack != context.stack_used) { + ASSERT(old_stack + 1 == context.stack_used); + DMC_PUSH(text, matchSwap); + } + if (context.stack_used > context.stack_need) { + context.stack_need = context.stack_used; + } + DMC_PUSH(text, matchPop); + --(context.stack_used); + } + } + DESTROY_WSTACK(wstack); + break; + } + if (!is_tuple(t)) { goto simple_term; } num_iters = arityval(*tuple_val(t)); @@ -1715,10 +1833,8 @@ Eterm db_prog_match(Process *c_p, Binary *bprog, Uint32 *return_flags) { MatchProg *prog = Binary2MatchProg(bprog); - Eterm *ep; - Eterm *tp; + const Eterm *ep, *tp, **sp; Eterm t; - Eterm **sp; Eterm *esp; MatchVariable* variables; BeamInstr *cp; @@ -1808,7 +1924,7 @@ Eterm db_prog_match(Process *c_p, Binary *bprog, restart: ep = &term; esp = (Eterm*)((char*)mpsp->u.heap + prog->stack_offset); - sp = (Eterm **) esp; + sp = (const Eterm **)esp; ret = am_true; do_catch = 0; fail_label = -1; @@ -1887,9 +2003,57 @@ restart: *sp++ = list_val_rel(*ep,base); ++ep; break; + case matchMap: + if (!is_map_rel(*ep, base)) { + FAIL(); + } + n = *pc++; + if (is_flatmap_rel(*ep,base)) { + if (flatmap_get_size(flatmap_val_rel(*ep, base)) < n) { + FAIL(); + } + } else { + ASSERT(is_hashmap_rel(*ep,base)); + if (hashmap_size_rel(*ep, base) < n) { + FAIL(); + } + } + ep = flatmap_val_rel(*ep, base); + break; + case matchPushM: + if (!is_map_rel(*ep, base)) { + FAIL(); + } + n = *pc++; + if (is_flatmap_rel(*ep,base)) { + if (flatmap_get_size(flatmap_val_rel(*ep, base)) < n) { + FAIL(); + } + } else { + ASSERT(is_hashmap_rel(*ep,base)); + if (hashmap_size_rel(*ep, base) < n) { + FAIL(); + } + } + *sp++ = flatmap_val_rel(*ep++, base); + break; + case matchKey: + t = (Eterm) *pc++; + tp = erts_maps_get_rel(t, make_flatmap_rel(ep, base), base); + if (!tp) { + FAIL(); + } + *sp++ = ep; + ep = tp; + break; case matchPop: ep = *(--sp); break; + case matchSwap: + tp = sp[-1]; + sp[-1] = sp[-2]; + sp[-2] = tp; + break; case matchBind: n = *pc++; variables[n].term = *ep++; @@ -1987,6 +2151,38 @@ restart: } *esp++ = t; break; + case matchMkFlatMap: + n = *pc++; + ehp = HAllocX(build_proc, 1 + MAP_HEADER_FLATMAP_SZ + n, HEAP_XTRA); + t = *ehp++ = *--esp; + { + flatmap_t *m = (flatmap_t *)ehp; + m->thing_word = MAP_HEADER_FLATMAP; + m->size = n; + m->keys = t; + } + t = make_flatmap(ehp); + ehp += MAP_HEADER_FLATMAP_SZ; + while (n--) { + *ehp++ = *--esp; + } + *esp++ = t; + break; + case matchMkHashMap: + n = *pc++; + esp -= 2*n; + ehp = HAllocX(build_proc, 2*n, HEAP_XTRA); + { + ErtsHeapFactory factory; + Uint ix; + factory.p = build_proc; + for (ix = 0; ix < 2*n; ix++){ + ehp[ix] = esp[ix]; + } + t = erts_hashmap_from_array(&factory, ehp, n, 0); + } + *esp++ = t; + break; case matchCall0: bif = (Eterm (*)(Process*, ...)) *pc++; t = (*bif)(build_proc, bif_args); @@ -2601,10 +2797,10 @@ Wterm db_do_read_element(DbUpdateHandle* handle, Sint position) } ASSERT(((DbTableCommon*)handle->tb)->compress); - ASSERT(!handle->mustResize); + ASSERT(!(handle->flags & DB_MUST_RESIZE)); handle->dbterm = db_alloc_tmp_uncompressed(&handle->tb->common, handle->dbterm); - handle->mustResize = 1; + handle->flags |= DB_MUST_RESIZE; return handle->dbterm->tpl[position]; } @@ -2637,11 +2833,11 @@ void db_do_update_element(DbUpdateHandle* handle, #endif return; } - if (!handle->mustResize) { + if (!(handle->flags & DB_MUST_RESIZE)) { if (handle->tb->common.compress) { handle->dbterm = db_alloc_tmp_uncompressed(&handle->tb->common, handle->dbterm); - handle->mustResize = 1; + handle->flags |= DB_MUST_RESIZE; oldval = handle->dbterm->tpl[position]; #if HALFWORD_HEAP old_base = NULL; @@ -2701,7 +2897,7 @@ both_size_set: /* write new value in old dbterm, finalize will make a flat copy */ handle->dbterm->tpl[position] = newval; - handle->mustResize = 1; + handle->flags |= DB_MUST_RESIZE; #if HALFWORD_HEAP if (old_base && newval_sz > 0) { @@ -3156,34 +3352,45 @@ int db_is_variable(Eterm obj) /* return 1 if obj contains a variable or underscore */ /* return 0 if obj is fully ground */ -int db_has_variable(Eterm obj) -{ - switch(obj & _TAG_PRIMARY_MASK) { - case TAG_PRIMARY_LIST: { - while (is_list(obj)) { - if (db_has_variable(CAR(list_val(obj)))) +int db_has_variable(Eterm node) { + DECLARE_ESTACK(s); + + ESTACK_PUSH(s,node); + while (!ESTACK_ISEMPTY(s)) { + node = ESTACK_POP(s); + switch(node & _TAG_PRIMARY_MASK) { + case TAG_PRIMARY_LIST: + while (is_list(node)) { + ESTACK_PUSH(s,CAR(list_val(node))); + node = CDR(list_val(node)); + } + ESTACK_PUSH(s,node); /* Non wellformed list or [] */ + break; + case TAG_PRIMARY_BOXED: + if (is_tuple(node)) { + Eterm *tuple = tuple_val(node); + int arity = arityval(*tuple); + while(arity--) { + ESTACK_PUSH(s,*(++tuple)); + } + } else if (is_flatmap(node)) { + Eterm *values = flatmap_get_values(flatmap_val(node)); + Uint size = flatmap_get_size(flatmap_val(node)); + ESTACK_PUSH(s, ((flatmap_t *) flatmap_val(node))->keys); + while (size--) { + ESTACK_PUSH(s, *(values++)); + } + } + break; + case TAG_PRIMARY_IMMED1: + if (node == am_Underscore || db_is_variable(node) >= 0) { + DESTROY_ESTACK(s); return 1; - obj = CDR(list_val(obj)); - } - return(db_has_variable(obj)); /* Non wellformed list or [] */ - } - case TAG_PRIMARY_BOXED: - if (!BOXED_IS_TUPLE(obj)) { - return 0; - } else { - Eterm *tuple = tuple_val(obj); - int arity = arityval(*tuple++); - while(arity--) { - if (db_has_variable(*tuple)) - return 1; - tuple++; } - return(0); + break; } - case TAG_PRIMARY_IMMED1: - if (obj == am_Underscore || db_is_variable(obj) >= 0) - return 1; } + DESTROY_ESTACK(s); return 0; } @@ -3243,11 +3450,9 @@ static DMCRet dmc_one_term(DMCContext *context, { Sint n; Eterm *hp; - ErlHeapFragment *tmp_mb; Uint sz, sz2, sz3; Uint i, j; - switch (c & _TAG_PRIMARY_MASK) { case TAG_PRIMARY_IMMED1: if ((n = db_is_variable(c)) >= 0) { /* variable */ @@ -3334,6 +3539,16 @@ static DMCRet dmc_one_term(DMCContext *context, DMC_PUSH(*text, n); DMC_PUSH(*stack, c); break; + case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE): + if (is_flatmap(c)) + n = flatmap_get_size(flatmap_val(c)); + else + n = hashmap_size(c); + DMC_PUSH(*text, matchPushM); + ++(context->stack_used); + DMC_PUSH(*text, n); + DMC_PUSH(*stack, c); + break; case (_TAG_HEADER_REF >> _TAG_PRIMARY_SIZE): { Eterm* ref_val = internal_ref_val(c); @@ -3415,16 +3630,8 @@ static DMCRet dmc_one_term(DMCContext *context, #endif break; default: /* BINARY, FUN, VECTOR, or EXTERNAL */ - /* - ** Make a private copy... - */ - n = size_object(c); - tmp_mb = new_message_buffer(n); - hp = tmp_mb->mem; DMC_PUSH(*text, matchEqBin); - DMC_PUSH(*text, copy_struct(c, n, &hp, &(tmp_mb->off_heap))); - tmp_mb->next = context->save; - context->save = tmp_mb; + DMC_PUSH(*text, dmc_private_copy(context, c)); break; } break; @@ -3437,6 +3644,22 @@ static DMCRet dmc_one_term(DMCContext *context, } /* +** Make a private copy of a term in a context. +*/ + +static Eterm +dmc_private_copy(DMCContext *context, Eterm c) +{ + Uint n = size_object(c); + ErlHeapFragment *tmp_mb = new_message_buffer(n); + Eterm *hp = tmp_mb->mem; + Eterm copy = copy_struct(c, n, &hp, &(tmp_mb->off_heap)); + tmp_mb->next = context->save; + context->save = tmp_mb; + return copy; +} + +/* ** Match guard compilation */ @@ -3527,57 +3750,78 @@ static DMCRet dmc_list(DMCContext *context, return retOk; } -static DMCRet dmc_tuple(DMCContext *context, - DMCHeap *heap, - DMC_STACK_TYPE(UWord) *text, - Eterm t, - int *constant) +static void +dmc_rearrange_constants(DMCContext *context, DMC_STACK_TYPE(UWord) *text, + int textpos, Eterm *p, Uint nelems) { DMC_STACK_TYPE(UWord) instr_save; + Uint i; + + DMC_INIT_STACK(instr_save); + while (DMC_STACK_NUM(*text) > textpos) { + DMC_PUSH(instr_save, DMC_POP(*text)); + } + for (i = nelems; i--;) { + do_emit_constant(context, text, p[i]); + } + while(!DMC_EMPTY(instr_save)) { + DMC_PUSH(*text, DMC_POP(instr_save)); + } + DMC_FREE(instr_save); +} + +static DMCRet +dmc_array(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm *p, Uint nelems, int *constant) +{ int all_constant = 1; int textpos = DMC_STACK_NUM(*text); - Eterm *p = tuple_val(t); - Uint nelems = arityval(*p); Uint i; - int c; - DMCRet ret; /* - ** We remember where we started to layout code, + ** We remember where we started to layout code, ** assume all is constant and back up and restart if not so. - ** The tuple should be laid out with the last element first, - ** so we can memcpy the tuple to the eheap. + ** The array should be laid out with the last element first, + ** so we can memcpy it to the eheap. */ - for (i = nelems; i > 0; --i) { - if ((ret = dmc_expr(context, heap, text, p[i], &c)) != retOk) - return ret; - if (!c && all_constant) { - all_constant = 0; - if (i < nelems) { - Uint j; + for (i = nelems; i--;) { + DMCRet ret; + int c; - /* - * Oops, we need to relayout the constants. - * Save the already laid out instructions. - */ - DMC_INIT_STACK(instr_save); - while (DMC_STACK_NUM(*text) > textpos) - DMC_PUSH(instr_save, DMC_POP(*text)); - for (j = nelems; j > i; --j) - do_emit_constant(context, text, p[j]); - while(!DMC_EMPTY(instr_save)) - DMC_PUSH(*text, DMC_POP(instr_save)); - DMC_FREE(instr_save); - } - } else if (c && !all_constant) { - /* push a constant */ - do_emit_constant(context, text, p[i]); - } + ret = dmc_expr(context, heap, text, p[i], &c); + if (ret != retOk) { + return ret; + } + if (!c && all_constant) { + all_constant = 0; + if (i < nelems - 1) { + dmc_rearrange_constants(context, text, textpos, + p + i + 1, nelems - i - 1); + } + } else if (c && !all_constant) { + do_emit_constant(context, text, p[i]); + } + } + *constant = all_constant; + return retOk; +} + +static DMCRet +dmc_tuple(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm t, int *constant) +{ + int all_constant; + Eterm *p = tuple_val(t); + Uint nelems = arityval(*p); + DMCRet ret; + + ret = dmc_array(context, heap, text, p + 1, nelems, &all_constant); + if (ret != retOk) { + return ret; } - if (all_constant) { - *constant = 1; - return retOk; + *constant = 1; + return retOk; } DMC_PUSH(*text, matchMkTuple); DMC_PUSH(*text, nelems); @@ -3586,6 +3830,93 @@ static DMCRet dmc_tuple(DMCContext *context, return retOk; } +static DMCRet +dmc_map(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, + Eterm t, int *constant) +{ + int nelems; + int constant_values; + DMCRet ret; + if (is_flatmap(t)) { + flatmap_t *m = (flatmap_t *)flatmap_val(t); + Eterm *values = flatmap_get_values(m); + + nelems = flatmap_get_size(m); + ret = dmc_array(context, heap, text, values, nelems, &constant_values); + + if (ret != retOk) { + return ret; + } + if (constant_values) { + *constant = 1; + return retOk; + } + DMC_PUSH(*text, matchPushC); + DMC_PUSH(*text, dmc_private_copy(context, m->keys)); + if (++context->stack_used > context->stack_need) { + context->stack_need = context->stack_used; + } + DMC_PUSH(*text, matchMkFlatMap); + DMC_PUSH(*text, nelems); + context->stack_used -= nelems; + *constant = 0; + return retOk; + } else { + DECLARE_WSTACK(wstack); + Eterm *kv; + int c; + + ASSERT(is_hashmap(t)); + + hashmap_iterator_init(&wstack, t, 1); + constant_values = 1; + nelems = hashmap_size(t); + + while ((kv=hashmap_iterator_prev(&wstack)) != NULL) { + if ((ret = dmc_expr(context, heap, text, CDR(kv), &c)) != retOk) { + DESTROY_WSTACK(wstack); + return ret; + } + if (!c) + constant_values = 0; + } + + if (constant_values) { + *constant = 1; + DESTROY_WSTACK(wstack); + return retOk; + } + + *constant = 0; + + hashmap_iterator_init(&wstack, t, 1); + + while ((kv=hashmap_iterator_prev(&wstack)) != NULL) { + /* push key */ + if ((ret = dmc_expr(context, heap, text, CAR(kv), &c)) != retOk) { + DESTROY_WSTACK(wstack); + return ret; + } + if (c) { + do_emit_constant(context, text, CAR(kv)); + } + /* push value */ + if ((ret = dmc_expr(context, heap, text, CDR(kv), &c)) != retOk) { + DESTROY_WSTACK(wstack); + return ret; + } + if (c) { + do_emit_constant(context, text, CDR(kv)); + } + } + DMC_PUSH(*text, matchMkHashMap); + DMC_PUSH(*text, nelems); + context->stack_used -= nelems; + DESTROY_WSTACK(wstack); + return retOk; + } +} + static DMCRet dmc_whole_expression(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, @@ -4580,7 +4911,10 @@ static DMCRet dmc_expr(DMCContext *context, return ret; break; case TAG_PRIMARY_BOXED: - if (!BOXED_IS_TUPLE(t)) { + if (is_map(t)) { + return dmc_map(context, heap, text, t, constant); + } + if (!is_tuple(t)) { goto simple_term; } p = tuple_val(t); @@ -4855,7 +5189,7 @@ static Eterm my_copy_struct(Eterm t, Eterm **hp, ErlOffHeap* off_heap) *hp += 2; break; case TAG_PRIMARY_BOXED: - if (BOXED_IS_TUPLE(t)) { + if (is_tuple(t)) { if (arityval(*tuple_val(t)) == 1 && is_tuple(a = tuple_val(t)[1])) { Uint i,n; @@ -5126,6 +5460,18 @@ void db_match_dis(Binary *bp) ++t; erts_printf("Tuple\t%beu\n", n); break; + case matchMap: + ++t; + n = *t; + ++t; + erts_printf("Map\t%beu\n", n); + break; + case matchKey: + ++t; + p = (Eterm) *t; + ++t; + erts_printf("Key\t%p (%T)\n", t, p); + break; case matchPushT: ++t; n = *t; @@ -5136,10 +5482,20 @@ void db_match_dis(Binary *bp) ++t; erts_printf("PushL\n"); break; + case matchPushM: + ++t; + n = *t; + ++t; + erts_printf("PushM\t%beu\n", n); + break; case matchPop: ++t; erts_printf("Pop\n"); break; + case matchSwap: + ++t; + erts_printf("Swap\n"); + break; case matchBind: ++t; n = *t; @@ -5252,6 +5608,18 @@ void db_match_dis(Binary *bp) ++t; erts_printf("MkTuple\t%beu\n", n); break; + case matchMkFlatMap: + ++t; + n = *t; + ++t; + erts_printf("MkFlatMap\t%beu\n", n); + break; + case matchMkHashMap: + ++t; + n = *t; + ++t; + erts_printf("MkHashMap\t%beu\n", n); + break; case matchOr: ++t; n = *t; diff --git a/erts/emulator/beam/erl_db_util.h b/erts/emulator/beam/erl_db_util.h index 328b19dfc9..ca206c7f58 100644 --- a/erts/emulator/beam/erl_db_util.h +++ b/erts/emulator/beam/erl_db_util.h @@ -76,6 +76,9 @@ typedef struct db_term { union db_table; typedef union db_table DbTable; +#define DB_MUST_RESIZE 1 +#define DB_NEW_OBJECT 2 + /* Info about a database entry while it's being updated * (by update_counter or update_element) */ @@ -84,7 +87,7 @@ typedef struct { DbTerm* dbterm; void** bp; /* {Hash|Tree}DbTerm** */ Uint new_size; - int mustResize; + int flags; void* lck; #if HALFWORD_HEAP unsigned char* abs_vec; /* [i] true if dbterm->tpl[i] is absolute Eterm */ @@ -165,6 +168,7 @@ typedef struct db_table_method DbTable* tb, /* [in out] */ Eterm continuation, Eterm* ret); + int (*db_take)(Process *, DbTable *, Eterm, Eterm *); int (*db_delete_all_objects)(Process* p, DbTable* db /* [in out] */ ); @@ -182,15 +186,14 @@ typedef struct db_table_method void *arg); void (*db_check_table)(DbTable* tb); - /* Lookup a dbterm for updating. Return false if not found. - */ - int (*db_lookup_dbterm)(DbTable*, Eterm key, - DbUpdateHandle* handle); /* [out] */ + /* Lookup a dbterm for updating. Return false if not found. */ + int (*db_lookup_dbterm)(Process *, DbTable *, Eterm key, Eterm obj, + DbUpdateHandle* handle); - /* Must be called for each db_lookup_dbterm that returned true, - ** even if dbterm was not updated. - */ - void (*db_finalize_dbterm)(DbUpdateHandle* handle); + /* Must be called for each db_lookup_dbterm that returned true, even if + ** dbterm was not updated. If the handle was of a new object and cret is + ** not DB_ERROR_NONE, the object is removed from the table. */ + void (*db_finalize_dbterm)(int cret, DbUpdateHandle* handle); } DbTableMethod; diff --git a/erts/emulator/beam/erl_driver.h b/erts/emulator/beam/erl_driver.h index 3ecb379326..e498ac70ec 100644 --- a/erts/emulator/beam/erl_driver.h +++ b/erts/emulator/beam/erl_driver.h @@ -133,7 +133,7 @@ typedef struct { #define ERL_DRV_EXTENDED_MARKER (0xfeeeeeed) #define ERL_DRV_EXTENDED_MAJOR_VERSION 3 -#define ERL_DRV_EXTENDED_MINOR_VERSION 0 +#define ERL_DRV_EXTENDED_MINOR_VERSION 2 /* * The emulator will refuse to load a driver with a major version @@ -198,7 +198,7 @@ typedef long long ErlDrvSInt64; #error No 64-bit integer type #endif -#if defined(__WIN32__) +#if defined(__WIN32__) || defined(_WIN32) typedef ErlDrvUInt ErlDrvSizeT; typedef ErlDrvSInt ErlDrvSSizeT; #else @@ -361,6 +361,9 @@ typedef struct erl_drv_entry { /* Called on behalf of driver_select when it is safe to release 'event'. A typical unix driver would call close(event) */ + void (*emergency_close)(ErlDrvData drv_data); + /* called when the port is closed abruptly. + specifically when erl_crash_dump is called. */ /* When adding entries here, dont forget to pad in obsolete/driver.h */ } ErlDrvEntry; diff --git a/erts/emulator/beam/erl_drv_nif.h b/erts/emulator/beam/erl_drv_nif.h index 3f829ea7ea..4e8c6dc68b 100644 --- a/erts/emulator/beam/erl_drv_nif.h +++ b/erts/emulator/beam/erl_drv_nif.h @@ -35,6 +35,7 @@ typedef struct { int scheduler_threads; int nif_major_version; int nif_minor_version; + int dirty_scheduler_support; } ErlDrvSysInfo; typedef struct { diff --git a/erts/emulator/beam/erl_drv_thread.c b/erts/emulator/beam/erl_drv_thread.c index 147249f751..240faa823d 100644 --- a/erts/emulator/beam/erl_drv_thread.c +++ b/erts/emulator/beam/erl_drv_thread.c @@ -601,17 +601,16 @@ erl_drv_thread_create(char *name, #ifdef USE_THREADS int res; struct ErlDrvTid_ *dtid; - ethr_thr_opts ethr_opts; + ethr_thr_opts ethr_opts = ETHR_THR_OPTS_DEFAULT_INITER; ethr_thr_opts *use_opts; - ethr_thr_opts def_ethr_opts = ETHR_THR_OPTS_DEFAULT_INITER; - if (!opts) + if (!opts && !name) use_opts = NULL; else { - sys_memcpy((void *) ðr_opts, - (void *) &def_ethr_opts, - sizeof(ethr_thr_opts)); - ethr_opts.suggested_stack_size = opts->suggested_stack_size; + if(opts) + ethr_opts.suggested_stack_size = opts->suggested_stack_size; + + ethr_opts.name = name; use_opts = ðr_opts; } diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c index aa15d2cc57..4a116c0740 100644 --- a/erts/emulator/beam/erl_gc.c +++ b/erts/emulator/beam/erl_gc.c @@ -20,6 +20,8 @@ # include "config.h" #endif +#define ERL_WANT_GC_INTERNALS__ + #include "sys.h" #include "erl_vm.h" #include "global.h" @@ -37,6 +39,7 @@ #include "hipe_mode_switch.h" #endif #include "dtrace-wrapper.h" +#include "erl_bif_unique.h" #define ERTS_INACT_WR_PB_LEAVE_MUCH_LIMIT 1 #define ERTS_INACT_WR_PB_LEAVE_MUCH_PERCENTAGE 20 @@ -94,10 +97,10 @@ typedef struct { static Uint setup_rootset(Process*, Eterm*, int, Rootset*); static void cleanup_rootset(Rootset *rootset); -static Uint combined_message_size(Process* p); +static Uint combined_message_size(Process* p, int off_heap_msgs); static void remove_message_buffers(Process* p); -static int major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl); -static int minor_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl); +static int major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl, int off_heap_msgs); +static int minor_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl, int off_heap_msgs); static void do_minor(Process *p, Uint new_sz, Eterm* objv, int nobj); static Eterm* sweep_rootset(Rootset *rootset, Eterm* htop, char* src, Uint src_size); static Eterm* sweep_one_area(Eterm* n_hp, Eterm* n_htop, char* src, Uint src_size); @@ -173,15 +176,15 @@ erts_init_gc(void) int i = 0, ix; Sint max_heap_size = 0; - ASSERT(offsetof(ProcBin,thing_word) == offsetof(struct erl_off_heap_header,thing_word)); - ASSERT(offsetof(ProcBin,thing_word) == offsetof(ErlFunThing,thing_word)); - ASSERT(offsetof(ProcBin,thing_word) == offsetof(ExternalThing,header)); - ASSERT(offsetof(ProcBin,size) == offsetof(struct erl_off_heap_header,size)); - ASSERT(offsetof(ProcBin,size) == offsetof(ErlSubBin,size)); - ASSERT(offsetof(ProcBin,size) == offsetof(ErlHeapBin,size)); - ASSERT(offsetof(ProcBin,next) == offsetof(struct erl_off_heap_header,next)); - ASSERT(offsetof(ProcBin,next) == offsetof(ErlFunThing,next)); - ASSERT(offsetof(ProcBin,next) == offsetof(ExternalThing,next)); + ERTS_CT_ASSERT(offsetof(ProcBin,thing_word) == offsetof(struct erl_off_heap_header,thing_word)); + ERTS_CT_ASSERT(offsetof(ProcBin,thing_word) == offsetof(ErlFunThing,thing_word)); + ERTS_CT_ASSERT(offsetof(ProcBin,thing_word) == offsetof(ExternalThing,header)); + ERTS_CT_ASSERT(offsetof(ProcBin,size) == offsetof(struct erl_off_heap_header,size)); + ERTS_CT_ASSERT(offsetof(ProcBin,size) == offsetof(ErlSubBin,size)); + ERTS_CT_ASSERT(offsetof(ProcBin,size) == offsetof(ErlHeapBin,size)); + ERTS_CT_ASSERT(offsetof(ProcBin,next) == offsetof(struct erl_off_heap_header,next)); + ERTS_CT_ASSERT(offsetof(ProcBin,next) == offsetof(ErlFunThing,next)); + ERTS_CT_ASSERT(offsetof(ProcBin,next) == offsetof(ExternalThing,next)); erts_test_long_gc_sleep = 0; @@ -400,7 +403,9 @@ erts_garbage_collect(Process* p, int need, Eterm* objv, int nobj) { Uint reclaimed_now = 0; int done = 0; + int off_heap_msgs; Uint ms1, s1, us1; + erts_aint32_t state; ErtsSchedulerData *esdp; #ifdef USE_VM_PROBES DTRACE_CHARBUF(pidbuf, DTRACE_TERM_BUF_SIZE); @@ -417,7 +422,8 @@ erts_garbage_collect(Process* p, int need, Eterm* objv, int nobj) trace_gc(p, am_gc_start); } - erts_smp_atomic32_read_bor_nob(&p->state, ERTS_PSFLG_GC); + state = erts_smp_atomic32_read_bor_nob(&p->state, ERTS_PSFLG_GC); + off_heap_msgs = state & ERTS_PSFLG_OFF_HEAP_MSGS; if (erts_system_monitor_long_gc != 0) { get_now(&ms1, &s1, &us1); } @@ -443,11 +449,11 @@ erts_garbage_collect(Process* p, int need, Eterm* objv, int nobj) while (!done) { if ((FLAGS(p) & F_NEED_FULLSWEEP) != 0) { DTRACE2(gc_major_start, pidbuf, need); - done = major_collection(p, need, objv, nobj, &reclaimed_now); + done = major_collection(p, need, objv, nobj, &reclaimed_now, off_heap_msgs); DTRACE2(gc_major_end, pidbuf, reclaimed_now); } else { DTRACE2(gc_minor_start, pidbuf, need); - done = minor_collection(p, need, objv, nobj, &reclaimed_now); + done = minor_collection(p, need, objv, nobj, &reclaimed_now, off_heap_msgs); DTRACE2(gc_minor_end, pidbuf, reclaimed_now); } } @@ -830,7 +836,7 @@ erts_garbage_collect_literals(Process* p, Eterm* literals, } static int -minor_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl) +minor_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl, int off_heap_msgs) { Uint mature = HIGH_WATER(p) - HEAP_START(p); @@ -869,20 +875,22 @@ minor_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl) Uint size_after; Uint need_after; Uint stack_size = STACK_SZ_ON_HEAP(p); - Uint fragments = MBUF_SIZE(p) + combined_message_size(p); + Uint fragments = MBUF_SIZE(p) + combined_message_size(p, off_heap_msgs); Uint size_before = fragments + (HEAP_TOP(p) - HEAP_START(p)); Uint new_sz = next_heap_size(p, HEAP_SIZE(p) + fragments, 0); do_minor(p, new_sz, objv, nobj); - /* - * Copy newly received message onto the end of the new heap. - */ - ErtsGcQuickSanityCheck(p); - for (msgp = p->msg.first; msgp; msgp = msgp->next) { - if (msgp->data.attached) { - erts_move_msg_attached_data_to_heap(&p->htop, &p->off_heap, msgp); - ErtsGcQuickSanityCheck(p); + if (!off_heap_msgs) { + /* + * Copy newly received message onto the end of the new heap. + */ + ErtsGcQuickSanityCheck(p); + for (msgp = p->msg.first; msgp; msgp = msgp->next) { + if (msgp->data.attached) { + erts_move_msg_attached_data_to_heap(&p->htop, &p->off_heap, msgp); + ErtsGcQuickSanityCheck(p); + } } } ErtsGcQuickSanityCheck(p); @@ -1208,7 +1216,7 @@ do_minor(Process *p, Uint new_sz, Eterm* objv, int nobj) */ static int -major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl) +major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl, int off_heap_msgs) { Rootset rootset; Roots* roots; @@ -1221,8 +1229,7 @@ major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl) Uint oh_size = (char *) OLD_HTOP(p) - oh; Uint n; Uint new_sz; - Uint fragments = MBUF_SIZE(p) + combined_message_size(p); - ErlMessage *msgp; + Uint fragments = MBUF_SIZE(p) + combined_message_size(p, off_heap_msgs); size_before = fragments + (HEAP_TOP(p) - HEAP_START(p)); @@ -1432,13 +1439,16 @@ major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl) ErtsGcQuickSanityCheck(p); - /* - * Copy newly received message onto the end of the new heap. - */ - for (msgp = p->msg.first; msgp; msgp = msgp->next) { - if (msgp->data.attached) { - erts_move_msg_attached_data_to_heap(&p->htop, &p->off_heap, msgp); - ErtsGcQuickSanityCheck(p); + if (!off_heap_msgs) { + ErlMessage *msgp; + /* + * Copy newly received message onto the end of the new heap. + */ + for (msgp = p->msg.first; msgp; msgp = msgp->next) { + if (msgp->data.attached) { + erts_move_msg_attached_data_to_heap(&p->htop, &p->off_heap, msgp); + ErtsGcQuickSanityCheck(p); + } } } @@ -1499,15 +1509,17 @@ adjust_after_fullsweep(Process *p, Uint size_before, int need, Eterm *objv, int * mbuf list. */ static Uint -combined_message_size(Process* p) +combined_message_size(Process* p, int off_heap_msgs) { - Uint sz = 0; + Uint sz; ErlMessage *msgp; - for (msgp = p->msg.first; msgp; msgp = msgp->next) { - if (msgp->data.attached) { + if (off_heap_msgs) + return 0; + + for (sz = 0, msgp = p->msg.first; msgp; msgp = msgp->next) { + if (msgp->data.attached) sz += erts_msg_attached_data_size(msgp); - } } return sz; } @@ -2018,6 +2030,20 @@ setup_rootset(Process *p, Eterm *objv, int nobj, Rootset *rootset) roots[n].sz = 1; n++; } + + /* + * If a NIF has saved arguments, they need to be added + */ + if (ERTS_PROC_GET_NIF_TRAP_EXPORT(p)) { + Eterm* argv; + int argc; + if (erts_setup_nif_gc(p, &argv, &argc)) { + roots[n].v = argv; + roots[n].sz = argc; + n++; + } + } + ASSERT(n <= rootset->size); mp = p->msg.first; @@ -2386,7 +2412,6 @@ sweep_off_heap(Process *p, int fullsweep) } pb->val = erts_bin_realloc(pb->val, new_size); - pb->val->orig_size = new_size; pb->bytes = (byte *) pb->val->orig_bytes; } } diff --git a/erts/emulator/beam/erl_gc.h b/erts/emulator/beam/erl_gc.h index 5203dda263..bd6dcc9078 100644 --- a/erts/emulator/beam/erl_gc.h +++ b/erts/emulator/beam/erl_gc.h @@ -20,10 +20,12 @@ #ifndef __ERL_GC_H__ #define __ERL_GC_H__ -#include "erl_map.h" +#if defined(ERL_WANT_GC_INTERNALS__) || defined(ERTS_DO_INCL_GLB_INLINE_FUNC_DEF) /* GC declarations shared by beam/erl_gc.c and hipe/hipe_gc.c */ +#include "erl_map.h" + #if defined(DEBUG) && !ERTS_GLB_INLINE_INCL_FUNC_DEF # define HARDDEBUG 1 #endif @@ -53,7 +55,10 @@ do { \ nelts = header_arity(HDR); \ switch ((HDR) & _HEADER_SUBTAG_MASK) { \ case SUB_BINARY_SUBTAG: nelts++; break; \ - case MAP_SUBTAG: nelts+=map_get_size(PTR) + 1; break; \ + case MAP_SUBTAG: \ + if (is_flatmap_header(HDR)) nelts+=flatmap_get_size(PTR) + 1; \ + else nelts += hashmap_bitcount(MAP_HEADER_VAL(HDR)); \ + break; \ case FUN_SUBTAG: nelts+=((ErlFunThing*)(PTR))->num_free+1; break; \ } \ gval = make_boxed(HTOP); \ @@ -61,14 +66,11 @@ do { \ *HTOP++ = HDR; \ *PTR++ = gval; \ while (nelts--) *HTOP++ = *PTR++; \ - \ } while(0) #define in_area(ptr,start,nbytes) \ ((UWord)((char*)(ptr) - (char*)(start)) < (nbytes)) -extern Uint erts_test_long_gc_sleep; - #if defined(DEBUG) || defined(ERTS_OFFHEAP_DEBUG) int within(Eterm *ptr, Process *p); #endif @@ -97,4 +99,33 @@ ERTS_GLB_INLINE Eterm follow_moved(Eterm term) } #endif +#endif /* ERL_GC_C__ || HIPE_GC_C__ */ + +/* + * Global exported + */ + +extern Uint erts_test_long_gc_sleep; + +typedef struct { + Uint64 reclaimed; + Uint64 garbage_cols; +} ErtsGCInfo; + +void erts_gc_info(ErtsGCInfo *gcip); +void erts_init_gc(void); +int erts_garbage_collect(struct process*, int, Eterm*, int); +void erts_garbage_collect_hibernate(struct process* p); +Eterm erts_gc_after_bif_call(struct process* p, Eterm result, Eterm* regs, Uint arity); +void erts_garbage_collect_literals(struct process* p, Eterm* literals, + Uint lit_size, + struct erl_off_heap_header* oh); +Uint erts_next_heap_size(Uint, Uint); +Eterm erts_heap_sizes(struct process* p); + +void erts_offset_off_heap(struct erl_off_heap*, Sint, Eterm*, Eterm*); +void erts_offset_heap_ptr(Eterm*, Uint, Sint, Eterm*, Eterm*); +void erts_offset_heap(Eterm*, Uint, Sint, Eterm*, Eterm*); +void erts_free_heap_frags(struct process* p); + #endif /* __ERL_GC_H__ */ diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c index d54658f1ea..86d3416423 100644 --- a/erts/emulator/beam/erl_init.c +++ b/erts/emulator/beam/erl_init.c @@ -45,6 +45,7 @@ #include "erl_thr_queue.h" #include "erl_async.h" #include "erl_ptab.h" +#include "erl_bif_unique.h" #ifdef HIPE #include "hipe_mode_switch.h" /* for hipe_mode_switch_init() */ @@ -134,7 +135,9 @@ static void erl_init(int ncpu, int legacy_proc_tab, int port_tab_sz, int port_tab_sz_ignore_files, - int legacy_port_tab); + int legacy_port_tab, + int time_correction, + ErtsTimeWarpMode time_warp_mode); static erts_atomic_t exiting; @@ -161,9 +164,6 @@ int H_MIN_SIZE; /* The minimum heap grain */ int BIN_VH_MIN_SIZE; /* The minimum binary virtual*/ Uint32 erts_debug_flags; /* Debug flags. */ -#ifdef ERTS_OPCODE_COUNTER_SUPPORT -int count_instructions; -#endif int erts_backtrace_depth; /* How many functions to show in a backtrace * in error codes. */ @@ -188,12 +188,10 @@ static int no_dirty_io_schedulers; Uint32 verbose; /* See erl_debug.h for information about verbose */ #endif -int erts_disable_tolerant_timeofday; /* Time correction can be disabled it is - * not and/or it is too slow. - */ - int erts_atom_table_size = ATOM_LIMIT; /* Maximum number of atoms */ +int erts_pd_initial_size = 10; + int erts_modified_timing_level; int erts_no_crash_dump = 0; /* Use -d to suppress crash dump. */ @@ -269,6 +267,19 @@ this_rel_num(void) return this_rel; } +static ERTS_INLINE void +set_default_time_adj(int *time_correction_p, ErtsTimeWarpMode *time_warp_mode_p) +{ + *time_correction_p = 1; + *time_warp_mode_p = ERTS_NO_TIME_WARP_MODE; + if (!erts_check_time_adj_support(*time_correction_p, + *time_warp_mode_p)) { + *time_correction_p = 0; + ASSERT(erts_check_time_adj_support(*time_correction_p, + *time_warp_mode_p)); + } +} + /* * Common error printout function, all error messages * that don't go to the error logger go through here. @@ -284,13 +295,22 @@ static int early_init(int *argc, char **argv); void erts_short_init(void) { - int ncpu = early_init(NULL, NULL); + + int ncpu; + int time_correction; + ErtsTimeWarpMode time_warp_mode; + + set_default_time_adj(&time_correction, + &time_warp_mode); + ncpu = early_init(NULL, NULL); erl_init(ncpu, ERTS_DEFAULT_MAX_PROCESSES, 0, ERTS_DEFAULT_MAX_PORTS, 0, - 0); + 0, + time_correction, + time_warp_mode); erts_initialized = 1; } @@ -300,12 +320,15 @@ erl_init(int ncpu, int legacy_proc_tab, int port_tab_sz, int port_tab_sz_ignore_files, - int legacy_port_tab) + int legacy_port_tab, + int time_correction, + ErtsTimeWarpMode time_warp_mode) { init_benchmarking(); + erts_bif_unique_init(); erts_init_monitors(); - erts_init_time(); + erts_init_time(time_correction, time_warp_mode); erts_init_sys_common_misc(); erts_init_process(ncpu, proc_tab_sz, legacy_proc_tab); erts_init_scheduling(no_schedulers, @@ -316,6 +339,7 @@ erl_init(int ncpu, no_dirty_io_schedulers #endif ); + erts_late_init_time_sup(); erts_init_cpu_topology(); /* Must be after init_scheduling */ erts_init_gc(); /* Must be after init_scheduling */ erts_alloc_late_init(); @@ -324,7 +348,6 @@ erl_init(int ncpu, BIN_VH_MIN_SIZE = erts_next_heap_size(BIN_VH_MIN_SIZE, 0); erts_init_trace(); - erts_init_binary(); erts_init_bits(); erts_code_ix_init(); erts_init_fun_table(); @@ -337,6 +360,7 @@ erl_init(int ncpu, erts_ddll_init(); init_emulator(); erts_ptab_init(); /* Must be after init_emulator() */ + erts_init_binary(); /* Must be after init_emulator() */ erts_bp_init(); init_db(); /* Must be after init_emulator */ erts_bif_timer_init(); @@ -365,12 +389,13 @@ erl_init(int ncpu, erl_nif_init(); } -static void +static Eterm erl_first_process_otp(char* modname, void* code, unsigned size, int argc, char** argv) { int i; Eterm start_mod; Eterm args; + Eterm res; Eterm* hp; Process parent; ErlSpawnOpts so; @@ -400,10 +425,11 @@ erl_first_process_otp(char* modname, void* code, unsigned size, int argc, char** hp += 2; args = CONS(hp, env, args); - so.flags = 0; - (void) erl_create_process(&parent, start_mod, am_start, args, &so); + so.flags = SPO_SYSTEM_PROC; + res = erl_create_process(&parent, start_mod, am_start, args, &so); erts_smp_proc_unlock(&parent, ERTS_PROC_LOCK_MAIN); erts_cleanup_empty_process(&parent); + return res; } Eterm @@ -509,9 +535,9 @@ void erts_usage(void) /* erts_fprintf(stderr, "-b func set the boot function (default boot)\n"); */ - erts_fprintf(stderr, "-c disable continuous date/time correction with\n"); - erts_fprintf(stderr, " respect to uptime\n"); - + erts_fprintf(stderr, "-c bool enable or disable time correction\n"); + erts_fprintf(stderr, "-C mode set time warp mode; valid modes are:\n"); + erts_fprintf(stderr, " no_time_warp|single_time_warp|multi_time_warp\n"); erts_fprintf(stderr, "-d don't write a crash dump for internally detected errors\n"); erts_fprintf(stderr, " (halt(String) will still produce a crash dump)\n"); erts_fprintf(stderr, "-fn[u|a|l] Control how filenames are interpreted\n"); @@ -519,6 +545,8 @@ void erts_usage(void) H_DEFAULT_SIZE); erts_fprintf(stderr, "-hmbs size set minimum binary virtual heap size in words (default %d)\n", VH_DEFAULT_SIZE); + erts_fprintf(stderr, "-hpds size initial process dictionary size (default %d)\n", + erts_pd_initial_size); /* erts_fprintf(stderr, "-i module set the boot module (default init)\n"); */ @@ -548,6 +576,8 @@ void erts_usage(void) erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); erts_fprintf(stderr, "-sct cput set cpu topology,\n"); erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); + erts_fprintf(stderr, "-secio bool enable/disable eager check I/O scheduling,\n"); + erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT erts_fprintf(stderr, "-sub bool enable/disable scheduler utilization balancing,\n"); #else @@ -679,7 +709,6 @@ early_init(int *argc, char **argv) /* erts_sched_compact_load = 1; erts_printf_eterm_func = erts_printf_term; - erts_disable_tolerant_timeofday = 0; display_items = 200; erts_backtrace_depth = DEFAULT_BACKTRACE_SIZE; erts_async_max_threads = ERTS_DEFAULT_NO_ASYNC_THREADS; @@ -1185,7 +1214,11 @@ erl_start(int argc, char **argv) int port_tab_sz_ignore_files = 0; int legacy_proc_tab = 0; int legacy_port_tab = 0; + int time_correction; + ErtsTimeWarpMode time_warp_mode; + set_default_time_adj(&time_correction, + &time_warp_mode); envbufsz = sizeof(envbuf); if (erts_sys_getenv_raw(ERL_MAX_ETS_TABLES_ENV, envbuf, &envbufsz) == 0) @@ -1406,6 +1439,7 @@ erl_start(int argc, char **argv) * * h|ms - min_heap_size * h|mbs - min_bin_vheap_size + * h|pds - erts_pd_initial_size * */ if (has_prefix("mbs", sub_param)) { @@ -1423,6 +1457,14 @@ erl_start(int argc, char **argv) erts_usage(); } VERBOSE(DEBUG_SYSTEM, ("using minimum heap size %d\n", H_MIN_SIZE)); + } else if (has_prefix("pds", sub_param)) { + arg = get_arg(sub_param+3, argv[i+1], &i); + if ((erts_pd_initial_size = atoi(arg)) <= 0) { + erts_fprintf(stderr, "bad initial process dictionary size %s\n", arg); + erts_usage(); + } + VERBOSE(DEBUG_SYSTEM, ("using initial process dictionary size %d\n", + erts_pd_initial_size)); } else { /* backward compatibility */ arg = get_arg(argv[i]+2, argv[i+1], &i); @@ -1674,6 +1716,22 @@ erl_start(int argc, char **argv) erts_usage(); } } + else if (has_prefix("ecio", sub_param)) { + arg = get_arg(sub_param+4, argv[i+1], &i); +#ifndef __OSE__ + if (sys_strcmp("true", arg) == 0) + erts_eager_check_io = 1; + else +#endif + if (sys_strcmp("false", arg) == 0) + erts_eager_check_io = 0; + else { + erts_fprintf(stderr, + "bad schedule eager check I/O value '%s'\n", + arg); + erts_usage(); + } + } else if (has_prefix("pp", sub_param)) { arg = get_arg(sub_param+2, argv[i+1], &i); if (sys_strcmp(arg, "true") == 0) @@ -1878,15 +1936,56 @@ erl_start(int argc, char **argv) } break; } - case 'c': - if (argv[i][2] == 0) { /* -c: documented option */ - erts_disable_tolerant_timeofday = 1; + case 'C': + arg = get_arg(argv[i]+2, argv[i+1], &i); + if (sys_strcmp(arg, "no_time_warp") == 0) + time_warp_mode = ERTS_NO_TIME_WARP_MODE; + else if (sys_strcmp(arg, "single_time_warp") == 0) + time_warp_mode = ERTS_SINGLE_TIME_WARP_MODE; + else if (sys_strcmp(arg, "multi_time_warp") == 0) + time_warp_mode = ERTS_MULTI_TIME_WARP_MODE; + else { + erts_fprintf(stderr, + "Invalid time warp mode: %s\n", arg); + erts_usage(); } + break; + case 'c': + if (sys_strcmp(argv[i]+2, "false") == 0) + goto time_correction_false; + else if (sys_strcmp(argv[i]+2, "true") == 0) + goto time_correction_true; #ifdef ERTS_OPCODE_COUNTER_SUPPORT else if (argv[i][2] == 'i') { /* -ci: undcoumented option*/ count_instructions = 1; } #endif + else if (argv[i][2] == '\0') { + if (i + 1 >= argc) + goto time_correction_false; + else { + if (sys_strcmp(argv[i+1], "false") == 0) { + (void) get_arg(argv[i]+2, argv[i+1], &i); + goto time_correction_false; + } + else if (sys_strcmp(argv[i+1], "true") == 0) { + (void) get_arg(argv[i]+2, argv[i+1], &i); + time_correction_true: + time_correction = 1; + break; + } + else { + time_correction_false: + time_correction = 0; + break; + } + } + } + else { + arg = get_arg(argv[i]+2, argv[i+1], &i); + erts_fprintf(stderr, "Invalid time correnction value: %s\n", arg); + erts_usage(); + } break; case 'W': arg = get_arg(argv[i]+2, argv[i+1], &i); @@ -1932,6 +2031,30 @@ erl_start(int argc, char **argv) i++; } + if (!erts_check_time_adj_support(time_correction, time_warp_mode)) { + char *time_correction_str = time_correction ? "Enabled" : "Disabled"; + char *time_warp_str = "undefined"; + switch (time_warp_mode) { + case ERTS_NO_TIME_WARP_MODE: + time_warp_str = "no"; + break; + case ERTS_SINGLE_TIME_WARP_MODE: + time_warp_str = "single"; + break; + case ERTS_MULTI_TIME_WARP_MODE: + time_warp_str = "multi"; + break; + default: + time_warp_str = "undefined"; + break; + } + erts_fprintf(stderr, "%s time correction with %s time warp mode " + "is not supported on this platform\n", + time_correction_str, + time_warp_str); + erts_usage(); + } + /* Output format on windows for sprintf defaults to three exponents. * We use two-exponent to mimic normal sprintf behaviour. */ @@ -1965,7 +2088,9 @@ erl_start(int argc, char **argv) legacy_proc_tab, port_tab_sz, port_tab_sz_ignore_files, - legacy_port_tab); + legacy_port_tab, + time_correction, + time_warp_mode); load_preloaded(); erts_end_staging_code_ix(); @@ -1973,7 +2098,11 @@ erl_start(int argc, char **argv) erts_initialized = 1; - erl_first_process_otp("otp_ring0", NULL, 0, boot_argc, boot_argv); + { + Eterm init = erl_first_process_otp("otp_ring0", NULL, 0, + boot_argc, boot_argv); + erts_bif_timer_start_servers(init); + } #ifdef ERTS_SMP erts_start_schedulers(); @@ -2066,8 +2195,10 @@ erl_exit_vv(int n, int flush_async, char *fmt, va_list args1, va_list args2) system_cleanup(flush_async); save_statistics(); - - an = abs(n); + if (n < 0) + an = -(unsigned int)n; + else + an = n; if (erts_mtrace_enabled) erts_mtrace_exit((Uint32) an); diff --git a/erts/emulator/beam/erl_instrument.c b/erts/emulator/beam/erl_instrument.c index df7c443387..da85b86c87 100644 --- a/erts/emulator/beam/erl_instrument.c +++ b/erts/emulator/beam/erl_instrument.c @@ -1226,7 +1226,7 @@ erts_instr_init(int stat, int map_stat) mem_anchor = NULL; /* Install instrumentation functions */ - ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); + ERTS_CT_ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); sys_memcpy((void *)real_allctrs,(void *)erts_allctrs,sizeof(erts_allctrs)); diff --git a/erts/emulator/beam/erl_lock_check.c b/erts/emulator/beam/erl_lock_check.c index 7e3a90779d..261460d054 100644 --- a/erts/emulator/beam/erl_lock_check.c +++ b/erts/emulator/beam/erl_lock_check.c @@ -139,15 +139,14 @@ static erts_lc_lock_order_t erts_lock_order[] = { { "drv_tsd", NULL }, { "async_enq_mtx", NULL }, #ifdef ERTS_SMP - { "sys_msg_q", NULL }, { "atom_tab", NULL }, - { "make_ref", NULL }, { "misc_op_list_pre_alloc_lock", "address" }, { "message_pre_alloc_lock", "address" }, { "ptimer_pre_alloc_lock", "address", }, { "btm_pre_alloc_lock", NULL, }, { "dist_entry_out_queue", "address" }, { "port_sched_lock", "port_id" }, + { "sys_msg_q", NULL }, { "port_table", NULL }, #endif { "mtrace_op", NULL }, @@ -168,6 +167,8 @@ static erts_lc_lock_order_t erts_lock_order[] = { { "timer_wheel", NULL }, { "system_block", NULL }, { "timeofday", NULL }, + { "get_time", NULL }, + { "get_corrected_time", NULL }, { "breakpoints", NULL }, { "pollsets_lock", NULL }, { "pix_lock", "address" }, @@ -184,10 +185,8 @@ static erts_lc_lock_order_t erts_lock_order[] = { { "efile_drv dtrace mutex", NULL }, #endif { "mtrace_buf", NULL }, -#ifdef __WIN32__ #ifdef ERTS_SMP - { "sys_gethrtime", NULL }, -#endif + { "os_monotonic_time", NULL }, #endif { "erts_alloc_hard_debug", NULL }, { "hard_dbg_mseg", NULL }, @@ -227,8 +226,7 @@ rw_op_str(Uint16 flags) case ERTS_LC_FLG_LO_READ: return " (r)"; case ERTS_LC_FLG_LO_WRITE: - erts_fprintf(stderr, "\nInternal error\n"); - lc_abort(); + ERTS_INTERNAL_ERROR("Only write flag present"); default: break; } @@ -270,9 +268,9 @@ union erts_lc_free_block_t_ { static ethr_tsd_key locks_key; -static erts_lc_locked_locks_t *erts_locked_locks; +static erts_lc_locked_locks_t *erts_locked_locks = NULL; -static erts_lc_free_block_t *free_blocks; +static erts_lc_free_block_t *free_blocks = NULL; #ifdef ERTS_LC_STATIC_ALLOC #define ERTS_LC_FB_CHUNK_SIZE 10000 @@ -311,8 +309,7 @@ static ERTS_INLINE void lc_free(void *p) static void *lc_core_alloc(void) { lc_unlock(); - erts_fprintf(stderr, "Lock checker out of memory!\n"); - lc_abort(); + ERTS_INTERNAL_ERROR("Lock checker out of memory!\n"); } #else @@ -325,8 +322,7 @@ static void *lc_core_alloc(void) fbs = (erts_lc_free_block_t *) malloc(sizeof(erts_lc_free_block_t) * ERTS_LC_FB_CHUNK_SIZE); if (!fbs) { - erts_fprintf(stderr, "Lock checker failed to allocate memory!\n"); - lc_abort(); + ERTS_INTERNAL_ERROR("Lock checker failed to allocate memory!"); } for (i = 1; i < ERTS_LC_FB_CHUNK_SIZE - 1; i++) { #ifdef DEBUG @@ -366,11 +362,11 @@ create_locked_locks(char *thread_name) { erts_lc_locked_locks_t *l_lcks = malloc(sizeof(erts_lc_locked_locks_t)); if (!l_lcks) - lc_abort(); + ERTS_INTERNAL_ERROR("Lock checker failed to allocate memory!"); l_lcks->thread_name = strdup(thread_name ? thread_name : "unknown"); if (!l_lcks->thread_name) - lc_abort(); + ERTS_INTERNAL_ERROR("Lock checker failed to allocate memory!"); l_lcks->emu_thread = 0; l_lcks->tid = erts_thr_self(); @@ -691,7 +687,7 @@ erts_lc_set_thread_name(char *thread_name) free((void *) l_lcks->thread_name); l_lcks->thread_name = strdup(thread_name ? thread_name : "unknown"); if (!l_lcks->thread_name) - lc_abort(); + ERTS_INTERNAL_ERROR("strdup failed"); } l_lcks->emu_thread = 1; } @@ -1330,7 +1326,7 @@ erts_lc_init(void) #endif /* #ifdef ERTS_LC_STATIC_ALLOC */ if (ethr_spinlock_init(&free_blocks_lock) != 0) - lc_abort(); + ERTS_INTERNAL_ERROR("spinlock_init failed"); erts_tsd_key_create(&locks_key,"erts_lock_check_key"); } diff --git a/erts/emulator/beam/erl_lock_count.c b/erts/emulator/beam/erl_lock_count.c index 6f44bf097b..c6d8f4df95 100644 --- a/erts/emulator/beam/erl_lock_count.c +++ b/erts/emulator/beam/erl_lock_count.c @@ -61,6 +61,25 @@ static ERTS_INLINE void lcnt_unlock(void) { ethr_mutex_unlock(&lcnt_data_lock); } +const int log2_tab64[64] = { + 63, 0, 58, 1, 59, 47, 53, 2, + 60, 39, 48, 27, 54, 33, 42, 3, + 61, 51, 37, 40, 49, 18, 28, 20, + 55, 30, 34, 11, 43, 14, 22, 4, + 62, 57, 46, 52, 38, 26, 32, 41, + 50, 36, 17, 19, 29, 10, 13, 21, + 56, 45, 25, 31, 35, 16, 9, 12, + 44, 24, 15, 8, 23, 7, 6, 5}; + +static ERTS_INLINE int lcnt_log2(Uint64 v) { + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + v |= v >> 32; + return log2_tab64[((Uint64)((v - (v >> 1))*0x07EDD5E59A4E28C2)) >> 58]; +} static char* lcnt_lock_type(Uint16 flag) { switch(flag & ERTS_LCNT_LT_ALL) { @@ -81,20 +100,17 @@ static void lcnt_clear_stats(erts_lcnt_lock_stats_t *stats) { stats->timer_n = 0; stats->file = (char *)str_undefined; stats->line = 0; + sys_memzero(stats->hist.ns, sizeof(stats->hist.ns)); } static void lcnt_time(erts_lcnt_time_t *time) { -#ifdef HAVE_GETHRTIME - SysHrTime hr_time; - hr_time = sys_gethrtime(); - time->s = (unsigned long)(hr_time / 1000000000LL); - time->ns = (unsigned long)(hr_time - 1000000000LL*time->s); -#else - SysTimeval tv; - sys_gettimeofday(&tv); - time->s = tv.tv_sec; - time->ns = tv.tv_usec*1000LL; -#endif + /* + * erts_sys_hrtime() is the highest resolution + * we could find, it may or may not be monotonic... + */ + ErtsMonotonicTime mtime = erts_sys_hrtime(); + time->s = (unsigned long) (mtime / 1000000000LL); + time->ns = (unsigned long) (mtime - 1000000000LL*time->s); } static void lcnt_time_diff(erts_lcnt_time_t *d, erts_lcnt_time_t *t1, erts_lcnt_time_t *t0) { @@ -111,28 +127,29 @@ static void lcnt_time_diff(erts_lcnt_time_t *d, erts_lcnt_time_t *t1, erts_lcnt_ dns += 1000000000LL; } + ASSERT(ds >= 0); + d->s = ds; d->ns = dns; } -/* difference d must be positive */ +/* difference d must be non-negative */ static void lcnt_time_add(erts_lcnt_time_t *t, erts_lcnt_time_t *d) { - unsigned long ngns = 0; - t->s += d->s; t->ns += d->ns; - ngns = t->ns / 1000000000LL; + t->s += t->ns / 1000000000LL; t->ns = t->ns % 1000000000LL; - - t->s += ngns; } static erts_lcnt_thread_data_t *lcnt_thread_data_alloc(void) { erts_lcnt_thread_data_t *eltd; eltd = (erts_lcnt_thread_data_t*)malloc(sizeof(erts_lcnt_thread_data_t)); + if (!eltd) { + ERTS_INTERNAL_ERROR("Lock counter failed to allocate memory!"); + } eltd->timer_set = 0; eltd->lock_in_conflict = 0; @@ -158,59 +175,64 @@ static char* lock_opt(Uint16 flag) { return "--"; } -static void print_lock_x(erts_lcnt_lock_t *lock, Uint16 flag, char *action, char *extra) { - erts_aint_t colls, tries, w_state, r_state; - erts_lcnt_lock_stats_t *stats = NULL; - +static void print_lock_x(erts_lcnt_lock_t *lock, Uint16 flag, char *action) { + erts_aint_t w_state, r_state; char *type; - int i; - + + if (strcmp(lock->name, "run_queue") != 0) return; type = lcnt_lock_type(lock->flag); r_state = ethr_atomic_read(&lock->r_state); w_state = ethr_atomic_read(&lock->w_state); - if (lock->flag & flag) { - erts_printf("%20s [%30s] [r/w state %4ld/%4ld] id %T %s\r\n", - action, - lock->name, - r_state, - w_state, - lock->id, - extra); - } -} - -static void print_lock(erts_lcnt_lock_t *lock, char *action) { - if (strcmp(lock->name, "proc_main") == 0) { - print_lock_x(lock, ERTS_LCNT_LT_ALL, action, ""); + erts_fprintf(stderr,"%10s [%24s] [r/w state %4ld/%4ld] %2s id %T\r\n", + action, + lock->name, + r_state, + w_state, + type, + lock->id); } } - #endif static erts_lcnt_lock_stats_t *lcnt_get_lock_stats(erts_lcnt_lock_t *lock, char *file, unsigned int line) { unsigned int i; erts_lcnt_lock_stats_t *stats = NULL; - - for (i = 0; i < lock->n_stats; i++) { - if ((lock->stats[i].file == file) && (lock->stats[i].line == line)) { - return &(lock->stats[i]); - } - } - if (lock->n_stats < ERTS_LCNT_MAX_LOCK_LOCATIONS) { - stats = &lock->stats[lock->n_stats]; - lock->n_stats++; - stats->file = file; - stats->line = line; - return stats; + if (erts_lcnt_rt_options & ERTS_LCNT_OPT_LOCATION) { + for (i = 0; i < lock->n_stats; i++) { + if ((lock->stats[i].file == file) && (lock->stats[i].line == line)) { + return &(lock->stats[i]); + } + } + if (lock->n_stats < ERTS_LCNT_MAX_LOCK_LOCATIONS) { + stats = &lock->stats[lock->n_stats]; + lock->n_stats++; + stats->file = file; + stats->line = line; + return stats; + } } return &lock->stats[0]; +} + +static void lcnt_update_stats_hist(erts_lcnt_hist_t *hist, erts_lcnt_time_t *time_wait) { + int idx; + unsigned long r; + if (time_wait->s > 0 || time_wait->ns > ERTS_LCNT_HISTOGRAM_MAX_NS) { + idx = ERTS_LCNT_HISTOGRAM_SLOT_SIZE - 1; + } else { + r = time_wait->ns >> ERTS_LCNT_HISTOGRAM_RSHIFT; + if (r) idx = lcnt_log2(r); + else idx = 0; + } + hist->ns[idx]++; } -static void lcnt_update_stats(erts_lcnt_lock_stats_t *stats, int lock_in_conflict, erts_lcnt_time_t *time_wait) { +static void lcnt_update_stats(erts_lcnt_lock_stats_t *stats, int lock_in_conflict, + erts_lcnt_time_t *time_wait) { ethr_atomic_inc(&stats->tries); @@ -220,6 +242,7 @@ static void lcnt_update_stats(erts_lcnt_lock_stats_t *stats, int lock_in_conflic if (time_wait) { lcnt_time_add(&(stats->timer), time_wait); stats->timer_n++; + lcnt_update_stats_hist(&stats->hist,time_wait); } } @@ -248,6 +271,9 @@ void erts_lcnt_init() { /* init lcnt structure */ erts_lcnt_data = (erts_lcnt_data_t*)malloc(sizeof(erts_lcnt_data_t)); + if (!erts_lcnt_data) { + ERTS_INTERNAL_ERROR("Lock counter failed to allocate memory!"); + } erts_lcnt_data->current_locks = erts_lcnt_list_init(); erts_lcnt_data->deleted_locks = erts_lcnt_list_init(); @@ -269,6 +295,9 @@ erts_lcnt_lock_list_t *erts_lcnt_list_init(void) { erts_lcnt_lock_list_t *list; list = (erts_lcnt_lock_list_t*)malloc(sizeof(erts_lcnt_lock_list_t)); + if (!list) { + ERTS_INTERNAL_ERROR("Lock counter failed to allocate memory!"); + } list->head = NULL; list->tail = NULL; list->n = 0; @@ -330,8 +359,9 @@ void erts_lcnt_list_delete(erts_lcnt_lock_list_t *list, erts_lcnt_lock_t *lock) /* interface to erl_threads.h */ /* only lock on init and destroy, all others should use atomics */ void erts_lcnt_init_lock(erts_lcnt_lock_t *lock, char *name, Uint16 flag ) { - erts_lcnt_init_lock_x(lock, name, flag, am_undefined); + erts_lcnt_init_lock_x(lock, name, flag, NIL); } + void erts_lcnt_init_lock_x(erts_lcnt_lock_t *lock, char *name, Uint16 flag, Eterm id) { int i; if (!name) { @@ -360,7 +390,6 @@ void erts_lcnt_init_lock_x(erts_lcnt_lock_t *lock, char *name, Uint16 flag, Eter } erts_lcnt_list_insert(erts_lcnt_data->current_locks, lock); - lcnt_unlock(); } @@ -375,6 +404,9 @@ void erts_lcnt_destroy_lock(erts_lcnt_lock_t *lock) { /* copy structure and insert the copy */ deleted_lock = (erts_lcnt_lock_t*)malloc(sizeof(erts_lcnt_lock_t)); + if (!deleted_lock) { + ERTS_INTERNAL_ERROR("Lock counter failed to allocate memory!"); + } memcpy(deleted_lock, lock, sizeof(erts_lcnt_lock_t)); deleted_lock->next = NULL; @@ -417,8 +449,9 @@ void erts_lcnt_lock_opt(erts_lcnt_lock_t *lock, Uint16 option) { if ((w_state > 0) || (r_state > 0)) { eltd->lock_in_conflict = 1; - if (eltd->timer_set == 0) + if (eltd->timer_set == 0) { lcnt_time(&eltd->timer); + } eltd->timer_set++; } else { eltd->lock_in_conflict = 0; @@ -433,7 +466,7 @@ void erts_lcnt_lock(erts_lcnt_lock_t *lock) { if (!ERTS_LCNT_LOCK_TYPE(lock)) return; w_state = ethr_atomic_read(&lock->w_state); - ethr_atomic_inc( &lock->w_state); + ethr_atomic_inc(&lock->w_state); eltd = lcnt_get_thread_data(); @@ -446,10 +479,10 @@ void erts_lcnt_lock(erts_lcnt_lock_t *lock) { * 'atomicly'. All other locks will block the thread if w_state > 0 * i.e. locked. */ - if (eltd->timer_set == 0) + if (eltd->timer_set == 0) { lcnt_time(&eltd->timer); + } eltd->timer_set++; - } else { eltd->lock_in_conflict = 0; } @@ -459,11 +492,10 @@ void erts_lcnt_lock(erts_lcnt_lock_t *lock) { void erts_lcnt_lock_unaquire(erts_lcnt_lock_t *lock) { /* should check if this thread was "waiting" */ - if (erts_lcnt_rt_options & ERTS_LCNT_OPT_SUSPEND) return; if (!ERTS_LCNT_LOCK_TYPE(lock)) return; - ethr_atomic_dec( &lock->w_state); + ethr_atomic_dec(&lock->w_state); } /* erts_lcnt_lock_post @@ -491,7 +523,7 @@ void erts_lcnt_lock_post_x(erts_lcnt_lock_t *lock, char *file, unsigned int line if (!(lock->flag & (ERTS_LCNT_LT_RWMUTEX | ERTS_LCNT_LT_RWSPINLOCK))) { flowstate = ethr_atomic_read(&lock->flowstate); ASSERT(flowstate == 0); - ethr_atomic_inc( &lock->flowstate); + ethr_atomic_inc(&lock->flowstate); } #endif @@ -500,19 +532,12 @@ void erts_lcnt_lock_post_x(erts_lcnt_lock_t *lock, char *file, unsigned int line ASSERT(eltd); /* if lock was in conflict, time it */ - - if (erts_lcnt_rt_options & ERTS_LCNT_OPT_LOCATION) { - stats = lcnt_get_lock_stats(lock, file, line); - } else { - stats = &lock->stats[0]; - } - + stats = lcnt_get_lock_stats(lock, file, line); if (eltd->timer_set) { lcnt_time(&timer); lcnt_time_diff(&time_wait, &timer, &(eltd->timer)); lcnt_update_stats(stats, eltd->lock_in_conflict, &time_wait); - eltd->timer_set--; ASSERT(eltd->timer_set >= 0); } else { @@ -541,11 +566,11 @@ void erts_lcnt_unlock(erts_lcnt_lock_t *lock) { /* flowstate */ flowstate = ethr_atomic_read(&lock->flowstate); ASSERT(flowstate == 1); - ethr_atomic_dec( &lock->flowstate); + ethr_atomic_dec(&lock->flowstate); /* write state */ w_state = ethr_atomic_read(&lock->w_state); - ASSERT(w_state > 0) + ASSERT(w_state > 0); #endif ethr_atomic_dec(&lock->w_state); } @@ -582,9 +607,7 @@ void erts_lcnt_trylock(erts_lcnt_lock_t *lock, int res) { ethr_atomic_inc( &lock->flowstate); #endif ethr_atomic_inc(&lock->w_state); - lcnt_update_stats(&(lock->stats[0]), 0, NULL); - } else { ethr_atomic_inc(&lock->stats[0].tries); ethr_atomic_inc(&lock->stats[0].colls); diff --git a/erts/emulator/beam/erl_lock_count.h b/erts/emulator/beam/erl_lock_count.h index 75f7cd028b..09fadd7e9e 100644 --- a/erts/emulator/beam/erl_lock_count.h +++ b/erts/emulator/beam/erl_lock_count.h @@ -35,6 +35,10 @@ * | | | - collisions (including trylock busy) * | | | - timer (time spent in waiting for lock) * | | | - n_timer (collisions excluding trylock busy) + * | | | - histogram + * | | | | - # 0 = log2(lock wait_time ns) + * | | | | - ... + * | | | | - # n = log2(lock wait_time ns) * * Each instance of a lock is the unique lock, i.e. set and id in that set. * For each lock there is a set of statistics with where and what impact @@ -68,8 +72,17 @@ #include "ethread.h" +#define ERTS_LCNT_MAX_LOCK_LOCATIONS (10) -#define ERTS_LCNT_MAX_LOCK_LOCATIONS (10) +/* histogram */ +#define ERTS_LCNT_HISTOGRAM_MAX_NS (((unsigned long)1LL << 28) - 1) +#if 0 || defined(ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT) +#define ERTS_LCNT_HISTOGRAM_SLOT_SIZE (30) +#define ERTS_LCNT_HISTOGRAM_RSHIFT (0) +#else +#define ERTS_LCNT_HISTOGRAM_SLOT_SIZE (20) +#define ERTS_LCNT_HISTOGRAM_RSHIFT (10) +#endif #define ERTS_LCNT_LT_SPINLOCK (((Uint16) 1) << 0) #define ERTS_LCNT_LT_RWSPINLOCK (((Uint16) 1) << 1) @@ -104,6 +117,10 @@ typedef struct { extern erts_lcnt_time_t timer_start; +typedef struct { + Uint32 ns[ERTS_LCNT_HISTOGRAM_SLOT_SIZE]; /* log2 array of nano seconds occurences */ +} erts_lcnt_hist_t; + typedef struct erts_lcnt_lock_stats_s { /* "tries" and "colls" needs to be atomic since * trylock busy does not aquire a lock and there @@ -118,6 +135,7 @@ typedef struct erts_lcnt_lock_stats_s { unsigned long timer_n; /* #times waited for lock */ erts_lcnt_time_t timer; /* total wait time for lock */ + erts_lcnt_hist_t hist; } erts_lcnt_lock_stats_t; /* rw locks uses both states, other locks only uses w_state */ diff --git a/erts/emulator/beam/erl_map.c b/erts/emulator/beam/erl_map.c index fdd2d0c0f6..ab40f47919 100644 --- a/erts/emulator/beam/erl_map.c +++ b/erts/emulator/beam/erl_map.c @@ -16,6 +16,9 @@ * * %CopyrightEnd% * + * hashmaps are an adaption of Rich Hickeys Persistent HashMaps + * which were an adaption of Phil Bagwells - Hash Array Mapped Tries + * * Author: Björn-Egil Dahlberg */ @@ -58,41 +61,82 @@ * - maps:size/1 * - maps:without/2 * + * DEBUG: for sharing calculation + * - erts_internal:map_to_tuple_keys/1 */ +#ifndef DECL_AM +#define DECL_AM(S) Eterm AM_ ## S = am_atom_put(#S, sizeof(#S) - 1) +#endif + +/* for hashmap_from_list/1 */ +typedef struct { + Uint32 hx; + Uint32 skip; + Uint i; + Eterm val; +} hxnode_t; + + +static Eterm flatmap_merge(Process *p, Eterm nodeA, Eterm nodeB); +static Eterm map_merge_mixed(Process *p, Eterm flat, Eterm tree, int swap_args); +static Eterm hashmap_merge(Process *p, Eterm nodeA, Eterm nodeB); +static Eterm hashmap_to_list(Process *p, Eterm map); +static Eterm hashmap_keys(Process *p, Eterm map); +static Eterm hashmap_values(Process *p, Eterm map); +static Eterm hashmap_delete(Process *p, Uint32 hx, Eterm key, Eterm node); +static Eterm flatmap_from_validated_list(Process *p, Eterm list, Uint size); +static Eterm hashmap_from_validated_list(Process *p, Eterm list, Uint size); +static Eterm hashmap_from_unsorted_array(ErtsHeapFactory*, hxnode_t *hxns, Uint n, int reject_dupkeys); +static Eterm hashmap_from_sorted_unique_array(ErtsHeapFactory*, hxnode_t *hxns, Uint n, int is_root); +static Eterm hashmap_from_chunked_array(ErtsHeapFactory*, hxnode_t *hxns, Uint n, int is_root); +static Eterm hashmap_info(Process *p, Eterm node); +static Eterm hashmap_bld_tuple_uint(Uint **hpp, Uint *szp, Uint n, Uint nums[]); +static int hxnodecmp(hxnode_t* a, hxnode_t* b); +static int hxnodecmpkey(hxnode_t* a, hxnode_t* b); + /* erlang:map_size/1 * the corresponding instruction is implemented in: * beam/erl_bif_guard.c */ BIF_RETTYPE map_size_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_1)) { Eterm *hp; Uint hsz = 0; - map_t *mp = (map_t*)map_val(BIF_ARG_1); - Uint n = map_get_size(mp); + flatmap_t *mp = (flatmap_t*)flatmap_val(BIF_ARG_1); + Uint n = flatmap_get_size(mp); erts_bld_uint(NULL, &hsz, n); hp = HAlloc(BIF_P, hsz); BIF_RET(erts_bld_uint(&hp, NULL, n)); + } else if (is_hashmap(BIF_ARG_1)) { + Eterm *head, *hp, res; + Uint size, hsz=0; + + head = hashmap_val(BIF_ARG_1); + size = head[1]; + (void) erts_bld_uint(NULL, &hsz, size); + hp = HAlloc(BIF_P, hsz); + res = erts_bld_uint(&hp, NULL, size); + BIF_RET(res); } BIF_ERROR(BIF_P, BADARG); } -/* maps:to_list/1 - */ +/* maps:to_list/1 */ BIF_RETTYPE maps_to_list_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_1)) { Uint n; Eterm* hp; Eterm *ks,*vs, res, tup; - map_t *mp = (map_t*)map_val(BIF_ARG_1); + flatmap_t *mp = (flatmap_t*)flatmap_val(BIF_ARG_1); - ks = map_get_keys(mp); - vs = map_get_values(mp); - n = map_get_size(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + n = flatmap_get_size(mp); hp = HAlloc(BIF_P, (2 + 3) * n); res = NIL; @@ -102,6 +146,8 @@ BIF_RETTYPE maps_to_list_1(BIF_ALIST_1) { } BIF_RET(res); + } else if (is_hashmap(BIF_ARG_1)) { + return hashmap_to_list(BIF_P, BIF_ARG_1); } BIF_ERROR(BIF_P, BADARG); @@ -111,89 +157,84 @@ BIF_RETTYPE maps_to_list_1(BIF_ALIST_1) { * return value if key *matches* a key in the map */ -int erts_maps_find(Eterm key, Eterm map, Eterm *value) { +const Eterm * +#if HALFWORD_HEAP +erts_maps_get_rel(Eterm key, Eterm map, Eterm *map_base) +#else +erts_maps_get(Eterm key, Eterm map) +#endif +{ + Uint32 hx; + if (is_flatmap_rel(map, map_base)) { + Eterm *ks, *vs; + flatmap_t *mp; + Uint n, i; - Eterm *ks,*vs; - map_t *mp; - Uint n,i; + mp = (flatmap_t *)flatmap_val_rel(map, map_base); + n = flatmap_get_size(mp); - mp = (map_t*)map_val(map); - n = map_get_size(mp); - ks = map_get_keys(mp); - vs = map_get_values(mp); + if (n == 0) { + return NULL; + } - for( i = 0; i < n; i++) { - if (EQ(ks[i], key)) { - *value = vs[i]; - return 1; + ks = (Eterm *)tuple_val_rel(mp->keys, map_base) + 1; + vs = flatmap_get_values(mp); + + if (is_immed(key)) { + for (i = 0; i < n; i++) { + if (ks[i] == key) { + return &vs[i]; + } + } } + + for (i = 0; i < n; i++) { + if (eq_rel(ks[i], map_base, key, NULL)) { + return &vs[i]; + } + } + return NULL; } - return 0; + ASSERT(is_hashmap_rel(map, map_base)); + hx = hashmap_make_hash(key); + + return erts_hashmap_get_rel(hx, key, map, map_base); } BIF_RETTYPE maps_find_2(BIF_ALIST_2) { if (is_map(BIF_ARG_2)) { - Eterm *hp, value,res; + Eterm *hp, res; + const Eterm *value; - if (erts_maps_find(BIF_ARG_1, BIF_ARG_2, &value)) { + value = erts_maps_get(BIF_ARG_1, BIF_ARG_2); + if (value) { hp = HAlloc(BIF_P, 3); res = make_tuple(hp); *hp++ = make_arityval(2); *hp++ = am_ok; - *hp++ = value; + *hp++ = *value; BIF_RET(res); } - BIF_RET(am_error); } BIF_ERROR(BIF_P, BADARG); } + /* maps:get/2 * return value if key *matches* a key in the map * exception bad_key if none matches */ - -int erts_maps_get(Eterm key, Eterm map, Eterm *value) { - Eterm *ks,*vs; - map_t *mp; - Uint n,i; - - mp = (map_t*)map_val(map); - n = map_get_size(mp); - - if (n == 0) - return 0; - - ks = map_get_keys(mp); - vs = map_get_values(mp); - - if (is_immed(key)) { - for( i = 0; i < n; i++) { - if (ks[i] == key) { - *value = vs[i]; - return 1; - } - } - } - - for( i = 0; i < n; i++) { - if (EQ(ks[i], key)) { - *value = vs[i]; - return 1; - } - } - return 0; -} - BIF_RETTYPE maps_get_2(BIF_ALIST_2) { if (is_map(BIF_ARG_2)) { Eterm *hp; - Eterm value, error; + Eterm error; + const Eterm *value; char *s_error; - if (erts_maps_get(BIF_ARG_1, BIF_ARG_2, &value)) { - BIF_RET(value); + value = erts_maps_get(BIF_ARG_1, BIF_ARG_2); + if (value) { + BIF_RET(*value); } s_error = "bad_key"; @@ -211,13 +252,8 @@ BIF_RETTYPE maps_get_2(BIF_ALIST_2) { */ BIF_RETTYPE maps_from_list_1(BIF_ALIST_1) { - Eterm *kv, item = BIF_ARG_1; - Eterm *hp, *thp,*vs, *ks, keys, res; - map_t *mp; - Uint size = 0, unused_size = 0; - Sint c = 0; - Sint idx = 0; - + Eterm item = BIF_ARG_1, res, *kv; + Uint size = 0; if (is_list(item) || is_nil(item)) { /* Calculate size and check validity */ @@ -238,368 +274,1085 @@ BIF_RETTYPE maps_from_list_1(BIF_ALIST_1) { if (is_not_nil(item)) goto error; - hp = HAlloc(BIF_P, 3 + 1 + (2 * size)); - thp = hp; + if (size > MAP_SMALL_MAP_LIMIT) { + BIF_RET(hashmap_from_validated_list(BIF_P, BIF_ARG_1, size)); + } else { + BIF_RET(flatmap_from_validated_list(BIF_P, BIF_ARG_1, size)); + } + } + +error: + + BIF_ERROR(BIF_P, BADARG); +} + +static Eterm flatmap_from_validated_list(Process *p, Eterm list, Uint size) { + Eterm *kv, item = list; + Eterm *hp, *thp,*vs, *ks, keys, res; + flatmap_t *mp; + Uint unused_size = 0; + Sint c = 0; + Sint idx = 0; + + + hp = HAlloc(p, 3 + 1 + (2 * size)); + thp = hp; + keys = make_tuple(hp); + *hp++ = make_arityval(size); + ks = hp; + hp += size; + mp = (flatmap_t*)hp; + res = make_flatmap(mp); + hp += MAP_HEADER_FLATMAP_SZ; + vs = hp; + + mp->thing_word = MAP_HEADER_FLATMAP; + mp->size = size; /* set later, might shrink*/ + mp->keys = keys; + + if (size == 0) + return res; + + /* first entry */ + kv = tuple_val(CAR(list_val(item))); + ks[0] = kv[1]; + vs[0] = kv[2]; + size = 1; + item = CDR(list_val(item)); + + /* insert sort key/value pairs */ + while(is_list(item)) { + + kv = tuple_val(CAR(list_val(item))); + + /* compare ks backwards + * idx represent word index to be written (hole position). + * We cannot copy the elements when searching since we might + * have an equal key. So we search for just the index first =( + * + * It is perhaps faster to move the values in the first pass. + * Check for uniqueness during insert phase and then have a + * second phace compacting the map if duplicates are found + * during insert. .. or do someother sort .. shell-sort perhaps. + */ + + idx = size; + + while(idx > 0 && (c = CMP_TERM(kv[1],ks[idx-1])) < 0) { idx--; } + + if (c == 0) { + /* last compare was equal, + * i.e. we have to release memory + * and overwrite that key/value + */ + ks[idx-1] = kv[1]; + vs[idx-1] = kv[2]; + unused_size++; + } else { + Uint i = size; + while(i > idx) { + ks[i] = ks[i-1]; + vs[i] = vs[i-1]; + i--; + } + ks[idx] = kv[1]; + vs[idx] = kv[2]; + size++; + } + item = CDR(list_val(item)); + } + + if (unused_size) { + /* the key tuple is embedded in the heap + * write a bignum to clear it. + */ + /* release values as normal since they are on the top of the heap */ + + ks[size] = make_pos_bignum_header(unused_size - 1); + HRelease(p, vs + size + unused_size, vs + size); + } + + *thp = make_arityval(size); + mp->size = size; + return res; +} + +#define swizzle32(D,S) \ + do { \ + (D) = ((S) & 0x0000000f) << 28 | ((S) & 0x000000f0) << 20 \ + | ((S) & 0x00000f00) << 12 | ((S) & 0x0000f000) << 4 \ + | ((S) & 0x000f0000) >> 4 | ((S) & 0x00f00000) >> 12 \ + | ((S) & 0x0f000000) >> 20 | ((S) & 0xf0000000) >> 28; \ + } while(0) + +#define maskval(V,L) (((V) >> ((7 - (L))*4)) & 0xf) +#define cdepth(V1,V2) (hashmap_clz((V1) ^ (V2)) >> 2) + +static Eterm hashmap_from_validated_list(Process *p, Eterm list, Uint size) { + Eterm item = list; + Eterm *hp; + Eterm *kv, res; + Eterm tmp[2]; + Uint32 sw, hx; + Uint ix = 0; + hxnode_t *hxns; + ErtsHeapFactory factory; + + ASSERT(size > 0); + + hp = HAlloc(p, (2 * size)); + + /* create tmp hx values and leaf ptrs */ + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, size * sizeof(hxnode_t)); + + while(is_list(item)) { + res = CAR(list_val(item)); + kv = tuple_val(res); + hx = hashmap_restore_hash(tmp,0,kv[1]); + swizzle32(sw,hx); + hxns[ix].hx = sw; + hxns[ix].val = CONS(hp, kv[1], kv[2]); hp += 2; + hxns[ix].skip = 1; /* will be reassigned in from_array */ + hxns[ix].i = ix; + ix++; + item = CDR(list_val(item)); + } + + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, size, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + + if (hashmap_size(res) <= MAP_SMALL_MAP_LIMIT) { + DECLARE_WSTACK(wstack); + Eterm *kv, *ks, *vs; + flatmap_t *mp; + Eterm keys; + Uint n = hashmap_size(res); + + /* build flat structure */ + hp = HAlloc(p, 3 + 1 + (2 * n)); keys = make_tuple(hp); - *hp++ = make_arityval(size); + *hp++ = make_arityval(n); ks = hp; - hp += size; - mp = (map_t*)hp; - res = make_map(mp); - hp += MAP_HEADER_SIZE; + hp += n; + mp = (flatmap_t*)hp; + hp += MAP_HEADER_FLATMAP_SZ; vs = hp; - mp->thing_word = MAP_HEADER; - mp->size = size; /* set later, might shrink*/ + mp->thing_word = MAP_HEADER_FLATMAP; + mp->size = n; mp->keys = keys; - if (size == 0) - BIF_RET(res); + hashmap_iterator_init(&wstack, res, 0); - item = BIF_ARG_1; + while ((kv=hashmap_iterator_next(&wstack)) != NULL) { + *ks++ = CAR(kv); + *vs++ = CDR(kv); + } - /* first entry */ - kv = tuple_val(CAR(list_val(item))); - ks[0] = kv[1]; - vs[0] = kv[2]; - size = 1; - item = CDR(list_val(item)); + /* it cannot have multiple keys */ + erts_validate_and_sort_flatmap(mp); - /* insert sort key/value pairs */ - while(is_list(item)) { + DESTROY_WSTACK(wstack); + return make_flatmap(mp); + } - kv = tuple_val(CAR(list_val(item))); - - /* compare ks backwards - * idx represent word index to be written (hole position). - * We cannot copy the elements when searching since we might - * have an equal key. So we search for just the index first =( - * - * It is perhaps faster to move the values in the first pass. - * Check for uniqueness during insert phase and then have a - * second phace compacting the map if duplicates are found - * during insert. .. or do someother sort .. shell-sort perhaps. - */ + return res; +} - idx = size; +Eterm erts_hashmap_from_array(ErtsHeapFactory* factory, Eterm *leafs, Uint n, + int reject_dupkeys) { + Uint32 sw, hx; + Uint ix; + hxnode_t *hxns; + Eterm res; + + /* create tmp hx values and leaf ptrs */ + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, n * sizeof(hxnode_t)); + + for (ix = 0; ix < n; ix++) { + hx = hashmap_make_hash(*leafs); + swizzle32(sw,hx); + hxns[ix].hx = sw; + hxns[ix].val = make_list(leafs); + hxns[ix].skip = 1; + hxns[ix].i = ix; + leafs += 2; + } - while(idx > 0 && (c = CMP_TERM(kv[1],ks[idx-1])) < 0) { idx--; } + res = hashmap_from_unsorted_array(factory, hxns, n, reject_dupkeys); - if (c == 0) { - /* last compare was equal, - * i.e. we have to release memory - * and overwrite that key/value - */ - ks[idx-1] = kv[1]; - vs[idx-1] = kv[2]; - unused_size++; - } else { - Uint i = size; - while(i > idx) { - ks[i] = ks[i-1]; - vs[i] = vs[i-1]; - i--; + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + + return res; +} + + +Eterm erts_hashmap_from_ks_and_vs_extra(Process *p, Eterm *ks, Eterm *vs, Uint n, + Eterm key, Eterm value) { + Uint32 sw, hx; + Uint i,sz; + hxnode_t *hxns; + ErtsHeapFactory factory; + Eterm *hp, res; + + sz = (key == THE_NON_VALUE) ? n : (n + 1); + ASSERT(sz > MAP_SMALL_MAP_LIMIT); + hp = HAlloc(p, 2 * sz); + + /* create tmp hx values and leaf ptrs */ + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, sz * sizeof(hxnode_t)); + + for(i = 0; i < n; i++) { + hx = hashmap_make_hash(ks[i]); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, ks[i], vs[i]); hp += 2; + hxns[i].skip = 1; /* will be reassigned in from_array */ + hxns[i].i = i; + } + + if (key != THE_NON_VALUE) { + hx = hashmap_make_hash(key); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, key, value); hp += 2; + hxns[i].skip = 1; + hxns[i].i = i; + } + + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, sz, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + + return res; +} + +static Eterm hashmap_from_unsorted_array(ErtsHeapFactory* factory, + hxnode_t *hxns, Uint n, + int reject_dupkeys) { + Uint jx = 0, ix = 0, lx, cx; + Eterm res; + + if (n == 0) { + Eterm *hp; + hp = erts_produce_heap(factory, 2, 0); + hp[0] = MAP_HEADER_HAMT_HEAD_BITMAP(0); + hp[1] = 0; + + return make_hashmap(hp); + } + + /* sort and compact array (remove non-unique entries) */ + qsort(hxns, n, sizeof(hxnode_t), (int (*)(const void *, const void *)) hxnodecmp); + + ix = 0, cx = 0; + while(ix < n - 1) { + if (hxns[ix].hx == hxns[ix+1].hx) { + + /* find region of equal hash values */ + jx = ix + 1; + while(jx < n && hxns[ix].hx == hxns[jx].hx) { jx++; } + /* find all correct keys from region + * (last in list but now hash sorted so we check highest id instead) */ + + /* resort with keys instead of hash value within region */ + + qsort(&hxns[ix], jx - ix, sizeof(hxnode_t), + (int (*)(const void *, const void *)) hxnodecmpkey); + + while(ix < jx) { + lx = ix; + while(ix < jx && EQ(CAR(list_val(hxns[ix].val)), CAR(list_val(hxns[lx].val)))) { + if (reject_dupkeys) + return THE_NON_VALUE; + + if (hxns[ix].i > hxns[lx].i) { + lx = ix; + } + ix++; } - ks[idx] = kv[1]; - vs[idx] = kv[2]; - size++; + hxns[cx].hx = hxns[lx].hx; + hxns[cx].val = hxns[lx].val; + cx++; } - item = CDR(list_val(item)); + ix = jx; + continue; + } + if (ix > cx) { + hxns[cx].hx = hxns[ix].hx; + hxns[cx].val = hxns[ix].val; } + cx++; + ix++; + } - if (unused_size) { - /* the key tuple is embedded in the heap - * write a bignum to clear it. - */ - /* release values as normal since they are on the top of the heap */ + if (ix < n) { + hxns[cx].hx = hxns[ix].hx; + hxns[cx].val = hxns[ix].val; + cx++; + } - ks[size] = make_pos_bignum_header(unused_size - 1); - HRelease(BIF_P, vs + size + unused_size, vs + size); - } + if (cx > 1) { + /* recursive decompose array */ + res = hashmap_from_sorted_unique_array(factory, hxns, cx, 0); + } else { + Eterm *hp; - *thp = make_arityval(size); - mp->size = size; - BIF_RET(res); + /* we only have one item, either because n was 1 or + * because we hade multiples of the same key. + * + * hash value has been swizzled, need to drag it down to get the + * correct slot. */ + + hp = erts_produce_heap(factory, HAMT_HEAD_BITMAP_SZ(1), 0); + hp[0] = MAP_HEADER_HAMT_HEAD_BITMAP(1 << ((hxns[0].hx >> 0x1c) & 0xf)); + hp[1] = 1; + hp[2] = hxns[0].val; + res = make_hashmap(hp); } -error: + return res; +} - BIF_ERROR(BIF_P, BADARG); +static Eterm hashmap_from_sorted_unique_array(ErtsHeapFactory* factory, + hxnode_t *hxns, Uint n, int lvl) { + Eterm res = NIL; + Uint i,ix,jx,elems; + Uint32 sw, hx; + Eterm val; + Eterm th[2]; + hxnode_t *tmp; + + ASSERT(lvl < 32); + ix = 0; + elems = 1; + while (ix < n - 1) { + if (hxns[ix].hx == hxns[ix+1].hx) { + jx = ix + 1; + while (jx < n && hxns[ix].hx == hxns[jx].hx) { jx++; } + tmp = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, ((jx - ix)) * sizeof(hxnode_t)); + + for(i = 0; i < jx - ix; i++) { + val = hxns[i + ix].val; + hx = hashmap_restore_hash(th, lvl + 8, CAR(list_val(val))); + swizzle32(sw,hx); + tmp[i].hx = sw; + tmp[i].val = val; + tmp[i].i = i; + tmp[i].skip = 1; + } + + qsort(tmp, jx - ix, sizeof(hxnode_t), (int (*)(const void *, const void *)) hxnodecmp); + + hxns[ix].skip = jx - ix; + hxns[ix].val = hashmap_from_sorted_unique_array(factory, tmp, jx - ix, lvl + 8); + erts_free(ERTS_ALC_T_TMP, (void *) tmp); + ix = jx; + if (ix < n) { elems++; } + continue; + } + hxns[ix].skip = 1; + elems++; + ix++; + } + + res = hashmap_from_chunked_array(factory, hxns, elems, !lvl); + + ERTS_FACTORY_HOLE_CHECK(factory); + + return res; } -/* maps:is_key/2 - */ +#define HALLOC_EXTRA 200 +static Eterm hashmap_from_chunked_array(ErtsHeapFactory *factory, + hxnode_t *hxns, Uint n, int is_root) { + Uint ix, d, dn, dc, slot, elems; + Uint32 v, vp, vn, hdr; + Uint bp, sz; + DECLARE_ESTACK(stack); + Eterm res = NIL, *hp = NULL, *nhp; -BIF_RETTYPE maps_is_key_2(BIF_ALIST_2) { - if (is_map(BIF_ARG_2)) { - Eterm *ks, key; - map_t *mp; - Uint n,i; + ASSERT(n > 1); - mp = (map_t*)map_val(BIF_ARG_2); - key = BIF_ARG_1; - n = map_get_size(mp); - ks = map_get_keys(mp); + /* push initial nodes on the stack, + * this is the starting depth */ - if (n == 0) - BIF_RET(am_false); + ix = 0; + d = 0; + vp = hxns[ix].hx; + v = hxns[ix + hxns[ix].skip].hx; - if (is_immed(key)) { - for( i = 0; i < n; i++) { - if (ks[i] == key) { - BIF_RET(am_true); - } + ASSERT(vp > v); + slot = maskval(vp,d); + + while(slot == maskval(v,d)) { + ESTACK_PUSH(stack, 1 << slot); + d++; + slot = maskval(vp,d); + } + + res = hxns[ix].val; + + if (hxns[ix].skip > 1) { + dc = 7; + /* build collision nodes */ + while (dc > d) { + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(1), HALLOC_EXTRA); + hp[0] = MAP_HEADER_HAMT_NODE_BITMAP(1 << maskval(vp,dc)); + hp[1] = res; + res = make_hashmap(hp); + dc--; + } + } + + ESTACK_PUSH2(stack,res,1 << slot); + + /* all of the other nodes .. */ + elems = n - 2; /* remove first and last elements */ + while(elems--) { + hdr = ESTACK_POP(stack); + ix = ix + hxns[ix].skip; + + /* determine if node or subtree should be built by looking + * at the next value. */ + + vn = hxns[ix + hxns[ix].skip].hx; + dn = cdepth(v,vn); + ASSERT(v > vn); + + res = hxns[ix].val; + + if (hxns[ix].skip > 1) { + int wat = (d > dn) ? d : dn; + dc = 7; + /* build collision nodes */ + while (dc > wat) { + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(1), HALLOC_EXTRA); + hp[0] = MAP_HEADER_HAMT_NODE_BITMAP(1 << maskval(v,dc)); + hp[1] = res; + res = make_hashmap(hp); + dc--; } } - for( i = 0; i < n; i++) { - if (EQ(ks[i], key)) { - BIF_RET(am_true); + /* next depth is higher (implies collision) */ + if (d < dn) { + /* hdr is the popped one initially */ + while(d < dn) { + slot = maskval(v, d); + bp = 1 << slot; + ESTACK_PUSH(stack, hdr | bp); + d++; + hdr = 0; /* clear hdr for all other collisions */ } + + slot = maskval(v, d); + bp = 1 << slot; + /* no more collisions */ + ESTACK_PUSH2(stack,res,bp); + } else if (d == dn) { + /* no collisions at all */ + slot = maskval(v, d); + bp = 1 << slot; + ESTACK_PUSH2(stack,res,hdr | bp); + } else { + /* dn < n, we have a drop and we are done + * build nodes and subtree */ + while (dn != d) { + slot = maskval(v, d); + bp = 1 << slot; + /* OR bitposition before sz calculation to handle + * redundant collisions */ + hdr |= bp; + sz = hashmap_bitcount(hdr); + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(sz), HALLOC_EXTRA); + nhp = hp; + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hdr); + *hp++ = res; sz--; + while (sz--) { *hp++ = ESTACK_POP(stack); } + ASSERT((hp - nhp) < 18); + res = make_hashmap(nhp); + + /* we need to pop the next hdr and push if we don't need it */ + + hdr = ESTACK_POP(stack); + d--; + } + ESTACK_PUSH2(stack,res,hdr); } - BIF_RET(am_false); + + vp = v; + v = vn; + d = dn; + ERTS_FACTORY_HOLE_CHECK(factory); + } + + /* v and vp are reused from above */ + dn = cdepth(vp,v); + ix = ix + hxns[ix].skip; + res = hxns[ix].val; + + if (hxns[ix].skip > 1) { + dc = 7; + /* build collision nodes */ + while (dc > dn) { + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(1), HALLOC_EXTRA); + hp[0] = MAP_HEADER_HAMT_NODE_BITMAP(1 << maskval(v,dc)); + hp[1] = res; + res = make_hashmap(hp); + dc--; + } + } + + hdr = ESTACK_POP(stack); + /* pop remaining subtree if any */ + while (dn) { + slot = maskval(v, dn); + bp = 1 << slot; + /* OR bitposition before sz calculation to handle + * redundant collisions */ + hdr |= bp; + sz = hashmap_bitcount(hdr); + hp = erts_produce_heap(factory, HAMT_NODE_BITMAP_SZ(sz), HALLOC_EXTRA); + nhp = hp; + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hdr); + *hp++ = res; sz--; + + while (sz--) { *hp++ = ESTACK_POP(stack); } + res = make_hashmap(nhp); + hdr = ESTACK_POP(stack); + dn--; + } + + /* and finally the root .. */ + + slot = maskval(v, dn); + bp = 1 << slot; + hdr |= bp; + sz = hashmap_bitcount(hdr); + hp = erts_produce_heap(factory, sz + /* hdr + item */ (is_root ? 2 : 1), 0); + nhp = hp; + + if (is_root) { + *hp++ = (hdr == 0xffff) ? MAP_HEADER_HAMT_HEAD_ARRAY : MAP_HEADER_HAMT_HEAD_BITMAP(hdr); + *hp++ = n; + } else { + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hdr); + } + + *hp++ = res; sz--; + while (sz--) { *hp++ = ESTACK_POP(stack); } + + res = make_hashmap(nhp); + + ASSERT(ESTACK_COUNT(stack) == 0); + DESTROY_ESTACK(stack); + ERTS_FACTORY_HOLE_CHECK(factory); + return res; +} +#undef HALLOC_EXTRA + +static int hxnodecmpkey(hxnode_t *a, hxnode_t *b) { + return CMP_TERM(CAR(list_val(a->val)), CAR(list_val(b->val))); +} + +static int hxnodecmp(hxnode_t *a, hxnode_t *b) { + if (a->hx < b->hx) + return 1; + else if (a->hx == b->hx) + return 0; + else + return -1; +} + +/* maps:is_key/2 */ + +BIF_RETTYPE maps_is_key_2(BIF_ALIST_2) { + if (is_map(BIF_ARG_2)) { + BIF_RET(erts_maps_get(BIF_ARG_1, BIF_ARG_2) ? am_true : am_false); } BIF_ERROR(BIF_P, BADARG); } -/* maps:keys/1 - */ +/* maps:keys/1 */ BIF_RETTYPE maps_keys_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_1)) { Eterm *hp, *ks, res = NIL; - map_t *mp; + flatmap_t *mp; Uint n; - mp = (map_t*)map_val(BIF_ARG_1); - n = map_get_size(mp); + mp = (flatmap_t*)flatmap_val(BIF_ARG_1); + n = flatmap_get_size(mp); if (n == 0) BIF_RET(res); hp = HAlloc(BIF_P, (2 * n)); - ks = map_get_keys(mp); + ks = flatmap_get_keys(mp); while(n--) { res = CONS(hp, ks[n], res); hp += 2; } BIF_RET(res); + } else if (is_hashmap(BIF_ARG_1)) { + BIF_RET(hashmap_keys(BIF_P, BIF_ARG_1)); } BIF_ERROR(BIF_P, BADARG); } -/* maps:merge/2 - */ +/* maps:merge/2 */ BIF_RETTYPE maps_merge_2(BIF_ALIST_2) { - if (is_map(BIF_ARG_1) && is_map(BIF_ARG_2)) { - Eterm *hp,*thp; - Eterm tup; - Eterm *ks,*vs,*ks1,*vs1,*ks2,*vs2; - map_t *mp1,*mp2,*mp_new; - Uint n1,n2,i1,i2,need,unused_size=0; - int c = 0; - - mp1 = (map_t*)map_val(BIF_ARG_1); - mp2 = (map_t*)map_val(BIF_ARG_2); - n1 = map_get_size(mp1); - n2 = map_get_size(mp2); - - need = MAP_HEADER_SIZE + 1 + 2*(n1 + n2); - - hp = HAlloc(BIF_P, need); - thp = hp; - tup = make_tuple(thp); - ks = hp + 1; hp += 1 + n1 + n2; - mp_new = (map_t*)hp; hp += MAP_HEADER_SIZE; - vs = hp; hp += n1 + n2; - - mp_new->thing_word = MAP_HEADER; - mp_new->size = 0; - mp_new->keys = tup; - - i1 = 0; i2 = 0; - ks1 = map_get_keys(mp1); - vs1 = map_get_values(mp1); - ks2 = map_get_keys(mp2); - vs2 = map_get_values(mp2); - - while(i1 < n1 && i2 < n2) { - c = CMP_TERM(ks1[i1],ks2[i2]); - if ( c == 0) { - /* use righthand side arguments map value, - * but advance both maps */ - *ks++ = ks2[i2]; - *vs++ = vs2[i2]; - i1++, i2++, unused_size++; - } else if ( c < 0) { - *ks++ = ks1[i1]; - *vs++ = vs1[i1]; - i1++; - } else { - *ks++ = ks2[i2]; - *vs++ = vs2[i2]; - i2++; - } + if (is_flatmap(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_2)) { + BIF_RET(flatmap_merge(BIF_P, BIF_ARG_1, BIF_ARG_2)); + } else if (is_hashmap(BIF_ARG_2)) { + /* Will always become a tree */ + BIF_RET(map_merge_mixed(BIF_P, BIF_ARG_1, BIF_ARG_2, 0)); } + } else if (is_hashmap(BIF_ARG_1)) { + if (is_hashmap(BIF_ARG_2)) { + BIF_RET(hashmap_merge(BIF_P, BIF_ARG_1, BIF_ARG_2)); + } else if (is_flatmap(BIF_ARG_2)) { + /* Will always become a tree */ + BIF_RET(map_merge_mixed(BIF_P, BIF_ARG_2, BIF_ARG_1, 1)); + } + } + BIF_ERROR(BIF_P, BADARG); +} + +static Eterm flatmap_merge(Process *p, Eterm nodeA, Eterm nodeB) { + Eterm *hp,*thp; + Eterm tup; + Eterm *ks,*vs,*ks1,*vs1,*ks2,*vs2; + flatmap_t *mp1,*mp2,*mp_new; + Uint n,n1,n2,i1,i2,need,unused_size=0; + Sint c = 0; - /* copy remaining */ - while (i1 < n1) { + mp1 = (flatmap_t*)flatmap_val(nodeA); + mp2 = (flatmap_t*)flatmap_val(nodeB); + n1 = flatmap_get_size(mp1); + n2 = flatmap_get_size(mp2); + + need = MAP_HEADER_FLATMAP_SZ + 1 + 2 * (n1 + n2); + + hp = HAlloc(p, need); + thp = hp; + tup = make_tuple(thp); + ks = hp + 1; hp += 1 + n1 + n2; + mp_new = (flatmap_t*)hp; hp += MAP_HEADER_FLATMAP_SZ; + vs = hp; hp += n1 + n2; + + mp_new->thing_word = MAP_HEADER_FLATMAP; + mp_new->size = 0; + mp_new->keys = tup; + + i1 = 0; i2 = 0; + ks1 = flatmap_get_keys(mp1); + vs1 = flatmap_get_values(mp1); + ks2 = flatmap_get_keys(mp2); + vs2 = flatmap_get_values(mp2); + + while(i1 < n1 && i2 < n2) { + c = CMP_TERM(ks1[i1],ks2[i2]); + if (c == 0) { + /* use righthand side arguments map value, + * but advance both maps */ + *ks++ = ks2[i2]; + *vs++ = vs2[i2]; + i1++, i2++, unused_size++; + } else if (c < 0) { *ks++ = ks1[i1]; *vs++ = vs1[i1]; i1++; - } - - while (i2 < n2) { + } else { *ks++ = ks2[i2]; *vs++ = vs2[i2]; i2++; } + } - if (unused_size) { - /* the key tuple is embedded in the heap, write a bignum to clear it. - * - * release values as normal since they are on the top of the heap - * size = n1 + n1 - unused_size - */ + /* copy remaining */ + while (i1 < n1) { + *ks++ = ks1[i1]; + *vs++ = vs1[i1]; + i1++; + } - *ks = make_pos_bignum_header(unused_size - 1); - HRelease(BIF_P, vs + unused_size, vs); - } + while (i2 < n2) { + *ks++ = ks2[i2]; + *vs++ = vs2[i2]; + i2++; + } - mp_new->size = n1 + n2 - unused_size; - *thp = make_arityval(n1 + n2 - unused_size); + if (unused_size) { + /* the key tuple is embedded in the heap, write a bignum to clear it. + * + * release values as normal since they are on the top of the heap + * size = n1 + n1 - unused_size + */ - BIF_RET(make_map(mp_new)); + *ks = make_pos_bignum_header(unused_size - 1); + HRelease(p, vs + unused_size, vs); } - BIF_ERROR(BIF_P, BADARG); -} -/* maps:new/2 - */ -BIF_RETTYPE maps_new_0(BIF_ALIST_0) { - Eterm* hp; - Eterm tup; - map_t *mp; + n = n1 + n2 - unused_size; + *thp = make_arityval(n); + mp_new->size = n; - hp = HAlloc(BIF_P, (MAP_HEADER_SIZE + 1)); - tup = make_tuple(hp); - *hp++ = make_arityval(0); + /* Reshape map to a hashmap if the map exceeds the limit */ - mp = (map_t*)hp; - mp->thing_word = MAP_HEADER; - mp->size = 0; - mp->keys = tup; + if (n > MAP_SMALL_MAP_LIMIT) { + Uint32 hx,sw; + Uint i; + Eterm res; + hxnode_t *hxns; + ErtsHeapFactory factory; - BIF_RET(make_map(mp)); -} + ks = flatmap_get_keys(mp_new); + vs = flatmap_get_values(mp_new); -/* maps:put/3 - */ + hp = HAlloc(p, 2 * n); -Eterm erts_maps_put(Process *p, Eterm key, Eterm value, Eterm map) { - Sint n,i; - Sint c = 0; - Eterm* hp, *shp; - Eterm *ks,*vs, res, tup; - map_t *mp = (map_t*)map_val(map); + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP,n * sizeof(hxnode_t)); - n = map_get_size(mp); + for (i = 0; i < n; i++) { + hx = hashmap_make_hash(ks[i]); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, ks[i], vs[i]); hp += 2; + hxns[i].skip = 1; + hxns[i].i = i; + } - if (n == 0) { - hp = HAlloc(p, MAP_HEADER_SIZE + 1 + 2); - tup = make_tuple(hp); - *hp++ = make_arityval(1); - *hp++ = key; - res = make_map(hp); - *hp++ = MAP_HEADER; - *hp++ = 1; - *hp++ = tup; - *hp++ = value; + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, n, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); return res; } - ks = map_get_keys(mp); - vs = map_get_values(mp); + return make_flatmap(mp_new); +} + +static Eterm map_merge_mixed(Process *p, Eterm flat, Eterm tree, int swap_args) { + Eterm *ks, *vs, *hp, res; + flatmap_t *mp; + Uint n, i; + hxnode_t *hxns; + Uint32 sw, hx; + ErtsHeapFactory factory; + + /* convert flat to tree */ + + ASSERT(is_flatmap(flat)); + ASSERT(is_hashmap(tree)); + + mp = (flatmap_t*)flatmap_val(flat); + n = flatmap_get_size(mp); + + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + hp = HAlloc(p, 2 * n); - /* only allocate for values, - * assume key-tuple will be intact + hxns = (hxnode_t *)erts_alloc(ERTS_ALC_T_TMP, n * sizeof(hxnode_t)); + + for (i = 0; i < n; i++) { + hx = hashmap_make_hash(ks[i]); + swizzle32(sw,hx); + hxns[i].hx = sw; + hxns[i].val = CONS(hp, ks[i], vs[i]); hp += 2; + hxns[i].skip = 1; + hxns[i].i = i; + } + + factory.p = p; + res = hashmap_from_unsorted_array(&factory, hxns, n, 0); + + erts_free(ERTS_ALC_T_TMP, (void *) hxns); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + + return swap_args ? hashmap_merge(p, tree, res) : hashmap_merge(p, res, tree); +} + +#define HALLOC_EXTRA 200 + +static Eterm hashmap_merge(Process *p, Eterm nodeA, Eterm nodeB) { +#define PSTACK_TYPE struct HashmapMergePStackType + struct HashmapMergePStackType { + Eterm *srcA, *srcB; + Uint32 abm, bbm, rbm; /* node bitmaps */ + int keepA; + int ix; + Eterm array[16]; + }; + PSTACK_DECLARE(s, 4); + struct HashmapMergePStackType* sp = PSTACK_PUSH(s); + Eterm *hp, *nhp; + Eterm hdrA, hdrB; + Eterm th[2]; + Uint32 ahx, bhx; + Uint size; /* total key-value counter */ + int keepA = 0; + unsigned lvl = 0; + Eterm res = THE_NON_VALUE; + + /* + * Strategy: Do depth-first traversal of both trees (at the same time) + * and merge each pair of nodes. */ - hp = HAlloc(p, MAP_HEADER_SIZE + n); - shp = hp; /* save hp, used if optimistic update fails */ - res = make_map(hp); - *hp++ = MAP_HEADER; - *hp++ = n; - *hp++ = mp->keys; - - if (is_immed(key)) { - for( i = 0; i < n; i ++) { - if (ks[i] == key) { - *hp++ = value; - vs++; - c = 1; + { + hashmap_head_t* a = (hashmap_head_t*) hashmap_val(nodeA); + hashmap_head_t* b = (hashmap_head_t*) hashmap_val(nodeB); + size = a->size + b->size; + } + +recurse: + + if (primary_tag(nodeA) == TAG_PRIMARY_BOXED && + primary_tag(nodeB) == TAG_PRIMARY_LIST) { + /* Avoid implementing this combination by switching places */ + Eterm tmp = nodeA; + nodeA = nodeB; + nodeB = tmp; + keepA = !keepA; + } + + switch (primary_tag(nodeA)) { + case TAG_PRIMARY_LIST: { + sp->srcA = list_val(nodeA); + switch (primary_tag(nodeB)) { + case TAG_PRIMARY_LIST: { /* LEAF + LEAF */ + sp->srcB = list_val(nodeB); + + if (EQ(CAR(sp->srcA), CAR(sp->srcB))) { + --size; + res = keepA ? nodeA : nodeB; } else { - *hp++ = *vs++; + ahx = hashmap_restore_hash(th, lvl, CAR(sp->srcA)); + bhx = hashmap_restore_hash(th, lvl, CAR(sp->srcB)); + sp->abm = 1 << hashmap_index(ahx); + sp->bbm = 1 << hashmap_index(bhx); + + sp->srcA = &nodeA; + sp->srcB = &nodeB; } + break; } - } else { - for( i = 0; i < n; i ++) { - if (EQ(ks[i], key)) { - *hp++ = value; - vs++; - c = 1; - } else { - *hp++ = *vs++; + case TAG_PRIMARY_BOXED: { /* LEAF + NODE */ + sp->srcB = boxed_val(nodeB); + ASSERT(is_header(*sp->srcB)); + hdrB = *sp->srcB++; + + ahx = hashmap_restore_hash(th, lvl, CAR(sp->srcA)); + sp->abm = 1 << hashmap_index(ahx); + sp->srcA = &nodeA; + switch(hdrB & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + sp->srcB++; + sp->bbm = 0xffff; + break; + + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcB++; + case HAMT_SUBTAG_NODE_BITMAP: + sp->bbm = MAP_HEADER_VAL(hdrB); + break; + + default: + erl_exit(1, "bad header tag %ld\r\n", *sp->srcB & _HEADER_MAP_SUBTAG_MASK); + break; } + break; } + default: + erl_exit(1, "bad primary tag %ld\r\n", nodeB); + } + break; } + case TAG_PRIMARY_BOXED: { /* NODE + NODE */ + sp->srcA = boxed_val(nodeA); + hdrA = *sp->srcA++; + ASSERT(is_header(hdrA)); + switch (hdrA & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: { + sp->srcA++; + ASSERT(primary_tag(nodeB) == TAG_PRIMARY_BOXED); + sp->abm = 0xffff; + sp->srcB = boxed_val(nodeB); + hdrB = *sp->srcB++; + ASSERT(is_header(hdrB)); + switch (hdrB & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + sp->srcB++; + sp->bbm = 0xffff; + break; + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcB++; + case HAMT_SUBTAG_NODE_BITMAP: + sp->bbm = MAP_HEADER_VAL(hdrB); + break; + default: + erl_exit(1, "bad header tag %ld\r\n", *sp->srcB & _HEADER_MAP_SUBTAG_MASK); + } + break; + } + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcA++; + case HAMT_SUBTAG_NODE_BITMAP: { + ASSERT(primary_tag(nodeB) == TAG_PRIMARY_BOXED); + sp->abm = MAP_HEADER_VAL(hdrA); + sp->srcB = boxed_val(nodeB); + hdrB = *sp->srcB++; + ASSERT(is_header(hdrB)); + switch (hdrB & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + sp->srcB++; + sp->bbm = 0xffff; + break; + case HAMT_SUBTAG_HEAD_BITMAP: sp->srcB++; + case HAMT_SUBTAG_NODE_BITMAP: + sp->bbm = MAP_HEADER_VAL(hdrB); + break; - if (c) - return res; + default: + erl_exit(1, "bad header tag %ld\r\n", *sp->srcB & _HEADER_MAP_SUBTAG_MASK); + } + break; + } + default: + erl_exit(1, "bad primary tag %ld\r\n", nodeA); + } + break; + } + default: + erl_exit(1, "bad primary tag %ld\r\n", nodeA); + } - /* need to make a new tuple, - * use old hp since it needs to be recreated anyway. - */ - tup = make_tuple(shp); - *shp++ = make_arityval(n+1); + for (;;) { + if (is_value(res)) { /* We have a complete (sub-)tree or leaf */ + if (lvl == 0) + break; - hp = HAlloc(p, 3 + n + 1); - res = make_map(hp); - *hp++ = MAP_HEADER; - *hp++ = n + 1; - *hp++ = tup; + /* Pop from stack and continue build parent node */ + lvl--; + sp = PSTACK_POP(s); + sp->array[sp->ix++] = res; + res = THE_NON_VALUE; + if (sp->rbm) { + sp->srcA++; + sp->srcB++; + keepA = sp->keepA; + } + } else { /* Start build a node */ + sp->ix = 0; + sp->rbm = sp->abm | sp->bbm; + ASSERT(!(sp->rbm == 0 && lvl > 0)); + } - ks = map_get_keys(mp); - vs = map_get_values(mp); + while (sp->rbm) { + Uint32 next = sp->rbm & (sp->rbm-1); + Uint32 bit = sp->rbm ^ next; + sp->rbm = next; + if (sp->abm & bit) { + if (sp->bbm & bit) { + /* Bit clash. Push and resolve by recursive merge */ + if (sp->rbm) { + sp->keepA = keepA; + } + nodeA = *sp->srcA; + nodeB = *sp->srcB; + lvl++; + sp = PSTACK_PUSH(s); + goto recurse; + } else { + sp->array[sp->ix++] = *sp->srcA++; + } + } else { + ASSERT(sp->bbm & bit); + sp->array[sp->ix++] = *sp->srcB++; + } + } - ASSERT(n >= 0); + ASSERT(sp->ix == hashmap_bitcount(sp->abm | sp->bbm)); + if (lvl == 0) { + nhp = HAllocX(p, HAMT_HEAD_BITMAP_SZ(sp->ix), HALLOC_EXTRA); + hp = nhp; + *hp++ = (sp->ix == 16 ? MAP_HEADER_HAMT_HEAD_ARRAY + : MAP_HEADER_HAMT_HEAD_BITMAP(sp->abm | sp->bbm)); + *hp++ = size; + } else { + nhp = HAllocX(p, HAMT_NODE_BITMAP_SZ(sp->ix), HALLOC_EXTRA); + hp = nhp; + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(sp->abm | sp->bbm); + } + memcpy(hp, sp->array, sp->ix * sizeof(Eterm)); + res = make_boxed(nhp); + } + PSTACK_DESTROY(s); + return res; +} - /* copy map in order */ - while (n && ((c = CMP_TERM(*ks, key)) < 0)) { - *shp++ = *ks++; - *hp++ = *vs++; - n--; +static int hash_cmp(Uint32 ha, Uint32 hb) +{ + int i; + for (i=0; i<8; i++) { + int cmp = (int)(ha & 0xF) - (int)(hb & 0xF); + if (cmp) + return cmp; + ha >>= 4; + hb >>= 4; } + return 0; +} - *shp++ = key; - *hp++ = value; +int hashmap_key_hash_cmp(Eterm* ap, Eterm* bp) +{ + Eterm th[2]; + unsigned lvl = 0; + + if (ap && bp) { + ASSERT(CMP_TERM(CAR(ap), CAR(bp)) != 0); + for (;;) { + Uint32 ha = hashmap_restore_hash(th, lvl, CAR(ap)); + Uint32 hb = hashmap_restore_hash(th, lvl, CAR(bp)); + int cmp = hash_cmp(ha, hb); + if (cmp) + return cmp; + lvl += 8; + } + } + return ap ? -1 : 1; +} - ASSERT(n >= 0); +/* maps:new/0 */ - while(n--) { - *shp++ = *ks++; - *hp++ = *vs++; - } - /* we have one word remaining - * this will work out fine once we get the size word - * in the header. - */ - *shp = make_pos_bignum_header(0); - return res; +BIF_RETTYPE maps_new_0(BIF_ALIST_0) { + Eterm* hp; + Eterm tup; + flatmap_t *mp; + + hp = HAlloc(BIF_P, (MAP_HEADER_FLATMAP_SZ + 1)); + tup = make_tuple(hp); + *hp++ = make_arityval(0); + + mp = (flatmap_t*)hp; + mp->thing_word = MAP_HEADER_FLATMAP; + mp->size = 0; + mp->keys = tup; + + BIF_RET(make_flatmap(mp)); } +/* maps:put/3 */ + BIF_RETTYPE maps_put_3(BIF_ALIST_3) { if (is_map(BIF_ARG_3)) { BIF_RET(erts_maps_put(BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3)); @@ -607,81 +1360,87 @@ BIF_RETTYPE maps_put_3(BIF_ALIST_3) { BIF_ERROR(BIF_P, BADARG); } -/* maps:remove/3 - */ +/* maps:remove/3 */ int erts_maps_remove(Process *p, Eterm key, Eterm map, Eterm *res) { - Sint n; - Uint need; - Eterm *hp_start; - Eterm *thp, *mhp; - Eterm *ks, *vs, tup; - map_t *mp = (map_t*)map_val(map); + Uint32 hx; + if (is_flatmap(map)) { + Sint n; + Uint need; + Eterm *hp_start; + Eterm *thp, *mhp; + Eterm *ks, *vs, tup; + flatmap_t *mp = (flatmap_t*)flatmap_val(map); + + n = flatmap_get_size(mp); + + if (n == 0) { + *res = map; + return 1; + } - n = map_get_size(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); - if (n == 0) { - *res = map; - return 1; - } + /* Assume key exists. + * Release allocated if it didn't. + * Allocate key tuple first. + */ - ks = map_get_keys(mp); - vs = map_get_values(mp); + need = n + 1 - 1 + 3 + n - 1; /* tuple - 1 + map - 1 */ + hp_start = HAlloc(p, need); + thp = hp_start; + mhp = thp + n; /* offset with tuple heap size */ - /* Assume key exists. - * Release allocated if it didn't. - * Allocate key tuple first. - */ + tup = make_tuple(thp); + *thp++ = make_arityval(n - 1); - need = n + 1 - 1 + 3 + n - 1; /* tuple - 1 + map - 1 */ - hp_start = HAlloc(p, need); - thp = hp_start; - mhp = thp + n; /* offset with tuple heap size */ + *res = make_flatmap(mhp); + *mhp++ = MAP_HEADER_FLATMAP; + *mhp++ = n - 1; + *mhp++ = tup; - tup = make_tuple(thp); - *thp++ = make_arityval(n - 1); - - *res = make_map(mhp); - *mhp++ = MAP_HEADER; - *mhp++ = n - 1; - *mhp++ = tup; - - if (is_immed(key)) { - while (1) { - if (*ks == key) { - goto found_key; - } else if (--n) { - *mhp++ = *vs++; - *thp++ = *ks++; - } else - break; - } - } else { - while(1) { - if (EQ(*ks, key)) { - goto found_key; - } else if (--n) { - *mhp++ = *vs++; - *thp++ = *ks++; - } else - break; + if (is_immed(key)) { + while (1) { + if (*ks == key) { + goto found_key; + } else if (--n) { + *mhp++ = *vs++; + *thp++ = *ks++; + } else + break; + } + } else { + while(1) { + if (EQ(*ks, key)) { + goto found_key; + } else if (--n) { + *mhp++ = *vs++; + *thp++ = *ks++; + } else + break; + } } - } - /* Not found, remove allocated memory - * and return previous map. - */ - HRelease(p, hp_start + need, hp_start); + /* Not found, remove allocated memory + * and return previous map. + */ + HRelease(p, hp_start + need, hp_start); - *res = map; - return 1; + *res = map; + return 1; found_key: - /* Copy rest of keys and values */ - if (--n) { - sys_memcpy(mhp, vs+1, n*sizeof(Eterm)); - sys_memcpy(thp, ks+1, n*sizeof(Eterm)); + /* Copy rest of keys and values */ + if (--n) { + sys_memcpy(mhp, vs+1, n*sizeof(Eterm)); + sys_memcpy(thp, ks+1, n*sizeof(Eterm)); + } + return 1; } + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + *res = hashmap_delete(p, hx, key, map); return 1; } @@ -695,29 +1454,28 @@ BIF_RETTYPE maps_remove_2(BIF_ALIST_2) { BIF_ERROR(BIF_P, BADARG); } -/* maps:update/3 - */ - int erts_maps_update(Process *p, Eterm key, Eterm value, Eterm map, Eterm *res) { + Uint32 hx; + if (is_flatmap(map)) { Sint n,i; Eterm* hp,*shp; Eterm *ks,*vs; - map_t *mp = (map_t*)map_val(map); + flatmap_t *mp = (flatmap_t*)flatmap_val(map); - if ((n = map_get_size(mp)) == 0) { + if ((n = flatmap_get_size(mp)) == 0) { return 0; } - ks = map_get_keys(mp); - vs = map_get_values(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); /* only allocate for values, * assume key-tuple will be intact */ - hp = HAlloc(p, MAP_HEADER_SIZE + n); + hp = HAlloc(p, MAP_HEADER_FLATMAP_SZ + n); shp = hp; - *hp++ = MAP_HEADER; + *hp++ = MAP_HEADER_FLATMAP; *hp++ = n; *hp++ = mp->keys; @@ -739,7 +1497,7 @@ int erts_maps_update(Process *p, Eterm key, Eterm value, Eterm map, Eterm *res) } } - HRelease(p, shp + MAP_HEADER_SIZE + n, shp); + HRelease(p, shp + MAP_HEADER_FLATMAP_SZ + n, shp); return 0; found_key: @@ -747,10 +1505,147 @@ found_key: vs++; if (++i < n) sys_memcpy(hp, vs, (n - i)*sizeof(Eterm)); - *res = make_map(shp); + *res = make_flatmap(shp); return 1; + } + + ASSERT(is_hashmap(map)); + hx = hashmap_make_hash(key); + *res = erts_hashmap_insert(p, hx, key, value, map, 1); + if (is_value(*res)) + return 1; + + return 0; +} + +Eterm erts_maps_put(Process *p, Eterm key, Eterm value, Eterm map) { + Uint32 hx; + Eterm res; + if (is_flatmap(map)) { + Sint n,i; + Sint c = 0; + Eterm* hp, *shp; + Eterm *ks, *vs, tup; + flatmap_t *mp = (flatmap_t*)flatmap_val(map); + + n = flatmap_get_size(mp); + + if (n == 0) { + hp = HAlloc(p, MAP_HEADER_FLATMAP_SZ + 1 + 2); + tup = make_tuple(hp); + *hp++ = make_arityval(1); + *hp++ = key; + res = make_flatmap(hp); + *hp++ = MAP_HEADER_FLATMAP; + *hp++ = 1; + *hp++ = tup; + *hp++ = value; + + return res; + } + + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + /* only allocate for values, + * assume key-tuple will be intact + */ + + hp = HAlloc(p, MAP_HEADER_FLATMAP_SZ + n); + shp = hp; /* save hp, used if optimistic update fails */ + res = make_flatmap(hp); + *hp++ = MAP_HEADER_FLATMAP; + *hp++ = n; + *hp++ = mp->keys; + + if (is_immed(key)) { + for( i = 0; i < n; i ++) { + if (ks[i] == key) { + *hp++ = value; + vs++; + c = 1; + } else { + *hp++ = *vs++; + } + } + } else { + for( i = 0; i < n; i ++) { + if (EQ(ks[i], key)) { + *hp++ = value; + vs++; + c = 1; + } else { + *hp++ = *vs++; + } + } + } + + if (c) + return res; + + /* the map will grow */ + + if (n >= MAP_SMALL_MAP_LIMIT) { + HRelease(p, shp + MAP_HEADER_FLATMAP_SZ + n, shp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + res = erts_hashmap_from_ks_and_vs_extra(p,ks,vs,n,key,value); + + return res; + } + + /* still a small map. need to make a new tuple, + * use old hp since it needs to be recreated anyway. */ + + tup = make_tuple(shp); + *shp++ = make_arityval(n+1); + + hp = HAlloc(p, 3 + n + 1); + res = make_flatmap(hp); + *hp++ = MAP_HEADER_FLATMAP; + *hp++ = n + 1; + *hp++ = tup; + + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + ASSERT(n >= 0); + + /* copy map in order */ + while (n && ((c = CMP_TERM(*ks, key)) < 0)) { + *shp++ = *ks++; + *hp++ = *vs++; + n--; + } + + *shp++ = key; + *hp++ = value; + + ASSERT(n >= 0); + + while(n--) { + *shp++ = *ks++; + *hp++ = *vs++; + } + /* we have one word remaining + * this will work out fine once we get the size word + * in the header. + */ + *shp = make_pos_bignum_header(0); + return res; + } + ASSERT(is_hashmap(map)); + + hx = hashmap_make_hash(key); + res = erts_hashmap_insert(p, hx, key, value, map, 0); + ASSERT(is_hashmap(res)); + + return res; } +/* maps:update/3 */ + BIF_RETTYPE maps_update_3(BIF_ALIST_3) { if (is_map(BIF_ARG_3)) { Eterm res; @@ -762,41 +1657,769 @@ BIF_RETTYPE maps_update_3(BIF_ALIST_3) { } -/* maps:values/1 - */ +/* maps:values/1 */ BIF_RETTYPE maps_values_1(BIF_ALIST_1) { - if (is_map(BIF_ARG_1)) { + if (is_flatmap(BIF_ARG_1)) { Eterm *hp, *vs, res = NIL; - map_t *mp; + flatmap_t *mp; Uint n; - mp = (map_t*)map_val(BIF_ARG_1); - n = map_get_size(mp); + mp = (flatmap_t*)flatmap_val(BIF_ARG_1); + n = flatmap_get_size(mp); if (n == 0) BIF_RET(res); hp = HAlloc(BIF_P, (2 * n)); - vs = map_get_values(mp); + vs = flatmap_get_values(mp); while(n--) { res = CONS(hp, vs[n], res); hp += 2; } BIF_RET(res); + } else if (is_hashmap(BIF_ARG_1)) { + BIF_RET(hashmap_values(BIF_P, BIF_ARG_1)); } BIF_ERROR(BIF_P, BADARG); } -int erts_validate_and_sort_map(map_t* mp) +static Eterm hashmap_to_list(Process *p, Eterm node) { + DECLARE_WSTACK(stack); + Eterm *hp, *kv; + Eterm res = NIL; + + hp = HAlloc(p, hashmap_size(node) * (2 + 3)); + hashmap_iterator_init(&stack, node, 0); + while ((kv=hashmap_iterator_next(&stack)) != NULL) { + Eterm tup = TUPLE2(hp, CAR(kv), CDR(kv)); + hp += 3; + res = CONS(hp, tup, res); + hp += 2; + } + DESTROY_WSTACK(stack); + return res; +} + +void hashmap_iterator_init(ErtsWStack* s, Eterm node, int reverse) { + Eterm hdr = *hashmap_val(node); + Uint sz; + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + break; + default: + erl_exit(1, "bad header"); + } + + WSTACK_PUSH3((*s), (UWord)THE_NON_VALUE, /* end marker */ + (UWord)(!reverse ? 0 : sz+1), + (UWord)node); +} + +Eterm* hashmap_iterator_next(ErtsWStack* s) { + Eterm node, *ptr, hdr; + Uint32 sz; + Uint idx; + + for (;;) { + ASSERT(!WSTACK_ISEMPTY((*s))); + node = (Eterm) WSTACK_POP((*s)); + if (is_non_value(node)) { + return NULL; + } + idx = (Uint) WSTACK_POP((*s)); + for (;;) { + ASSERT(is_boxed(node)); + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + ptr++; + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + ptr++; + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz < 17); + break; + default: + erl_exit(1, "bad header"); + } + + idx++; + + if (idx <= sz) { + WSTACK_PUSH2((*s), (UWord)idx, (UWord)node); + + if (is_list(ptr[idx])) { + return list_val(ptr[idx]); + } + ASSERT(is_boxed(ptr[idx])); + node = ptr[idx]; + idx = 0; + } + else + break; /* and pop parent node */ + } + } +} + +Eterm* hashmap_iterator_prev(ErtsWStack* s) { + Eterm node, *ptr, hdr; + Uint32 sz; + Uint idx; + + for (;;) { + ASSERT(!WSTACK_ISEMPTY((*s))); + node = (Eterm) WSTACK_POP((*s)); + if (is_non_value(node)) { + return NULL; + } + idx = (Uint) WSTACK_POP((*s)); + for (;;) { + ASSERT(is_boxed(node)); + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + ptr++; + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + ptr++; + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz < 17); + break; + default: + erl_exit(1, "bad header"); + } + + if (idx > sz) + idx = sz; + else + idx--; + + if (idx >= 1) { + WSTACK_PUSH2((*s), (UWord)idx, (UWord)node); + + if (is_list(ptr[idx])) { + return list_val(ptr[idx]); + } + ASSERT(is_boxed(ptr[idx])); + node = ptr[idx]; + idx = 17; + } + else + break; /* and pop parent node */ + } + } +} + +const Eterm * +#if HALFWORD_HEAP +erts_hashmap_get_rel(Uint32 hx, Eterm key, Eterm node, Eterm *map_base) +#else +erts_hashmap_get(Uint32 hx, Eterm key, Eterm node) +#endif { - Eterm *ks = map_get_keys(mp); - Eterm *vs = map_get_values(mp); - Uint sz = map_get_size(mp); + Eterm *ptr, hdr, *res; + Uint ix, lvl = 0; + Uint32 hval,bp; + DeclareTmpHeapNoproc(th,2); + UseTmpHeapNoproc(2); + + ASSERT(is_boxed(node)); + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + ASSERT(is_hashmap_header_head(hdr)); + ptr++; + + for (;;) { + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + if (hval != 0xffff) { + bp = 1 << ix; + if (!(bp & hval)) { + /* not occupied */ + res = NULL; + break; + } + ix = hashmap_bitcount(hval & (bp - 1)); + } + node = ptr[ix+1]; + + if (is_list(node)) { /* LEAF NODE [K|V] */ + ptr = list_val(node); + + res = eq_rel(CAR(ptr), map_base, key, NULL) ? &(CDR(ptr)) : NULL; + break; + } + + hx = hashmap_shift_hash(th,hx,lvl,key); + + ASSERT(is_boxed(node)); + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + ASSERT(!is_hashmap_header_head(hdr)); + } + + UnUseTmpHeapNoproc(2); + return res; +} + +Eterm erts_hashmap_insert(Process *p, Uint32 hx, Eterm key, Eterm value, + Eterm map, int is_update) { + Uint size, upsz; + Eterm *hp, res = THE_NON_VALUE; + DECLARE_ESTACK(stack); + if (erts_hashmap_insert_down(hx, key, map, &size, &upsz, &stack, is_update)) { + hp = HAlloc(p, size); + res = erts_hashmap_insert_up(hp, key, value, &upsz, &stack); + } + + DESTROY_ESTACK(stack); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + ERTS_HOLE_CHECK(p); + + return res; +} + + +int erts_hashmap_insert_down(Uint32 hx, Eterm key, Eterm node, Uint *sz, + Uint *update_size, ErtsEStack *sp, int is_update) { + Eterm *ptr; + Eterm hdr, ckey; + Eterm th[2]; + Uint32 ix, cix, bp, hval, chx; + Uint slot, lvl = 0, clvl; + Uint size = 0, n = 0; + + *update_size = 1; + + for (;;) { + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: /* LEAF NODE [K|V] */ + ptr = list_val(node); + ckey = CAR(ptr); + if (EQ(ckey, key)) { + *update_size = 0; + goto unroll; + } + if (is_update) { + return 0; + } + goto insert_subnodes; + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + size += HAMT_HEAD_ARRAY_SZ; + ESTACK_PUSH2(*sp, ix, node); + node = ptr[ix+2]; + break; + case HAMT_SUBTAG_NODE_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH4(*sp, n, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+1]; + ASSERT(HAMT_NODE_BITMAP_SZ(n) <= 17); + size += HAMT_NODE_BITMAP_SZ(n); + break; + } + /* not occupied */ + if (is_update) { + return 0; + } + size += HAMT_NODE_BITMAP_SZ(n+1); + goto unroll; + case HAMT_SUBTAG_HEAD_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH4(*sp, n, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+2]; + ASSERT(HAMT_HEAD_BITMAP_SZ(n) <= 18); + size += HAMT_HEAD_BITMAP_SZ(n); + break; + } + /* not occupied */ + if (is_update) { + return 0; + } + size += HAMT_HEAD_BITMAP_SZ(n+1); + goto unroll; + default: + erl_exit(1, "bad header tag %ld\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + break; + default: + erl_exit(1, "bad primary tag %p\r\n", node); + break; + } + } +insert_subnodes: + clvl = lvl; + chx = hashmap_restore_hash(th,clvl,ckey); + size += HAMT_NODE_BITMAP_SZ(2); + ix = hashmap_index(hx); + cix = hashmap_index(chx); + + while (cix == ix) { + ESTACK_PUSH4(*sp, 0, 1 << ix, 0, MAP_HEADER_HAMT_NODE_BITMAP(0)); + size += HAMT_NODE_BITMAP_SZ(1); + hx = hashmap_shift_hash(th,hx,lvl,key); + chx = hashmap_shift_hash(th,chx,clvl,ckey); + ix = hashmap_index(hx); + cix = hashmap_index(chx); + } + ESTACK_PUSH3(*sp, cix, ix, node); + +unroll: + *sz = size + /* res cons */ 2; + return 1; +} + +Eterm erts_hashmap_insert_up(Eterm *hp, Eterm key, Eterm value, + Uint *update_size, ErtsEStack *sp) { + Eterm node, *ptr, hdr; + Eterm res; + Eterm *nhp = NULL; + Uint32 ix, cix, bp, hval; + Uint slot, n; + /* Needed for halfword */ + DeclareTmpHeapNoproc(fake,1); + UseTmpHeapNoproc(1); + + res = CONS(hp, key, value); hp += 2; + + do { + node = ESTACK_POP(*sp); + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: + ix = (Uint32) ESTACK_POP(*sp); + cix = (Uint32) ESTACK_POP(*sp); + + nhp = hp; + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP((1 << ix) | (1 << cix)); + if (ix < cix) { + *hp++ = res; + *hp++ = node; + } else { + *hp++ = node; + *hp++ = res; + } + res = make_hashmap(nhp); + break; + case TAG_PRIMARY_HEADER: + /* subnodes, fake it */ + *fake = node; + node = make_boxed(fake); + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + slot = (Uint) ESTACK_POP(*sp); + nhp = hp; + n = HAMT_HEAD_ARRAY_SZ - 2; + *hp++ = MAP_HEADER_HAMT_HEAD_ARRAY; ptr++; + *hp++ = (*ptr++) + *update_size; + while(n--) { *hp++ = *ptr++; } + nhp[slot+2] = res; + res = make_hashmap(nhp); + break; + case HAMT_SUBTAG_NODE_BITMAP: + slot = (Uint) ESTACK_POP(*sp); + bp = (Uint32) ESTACK_POP(*sp); + n = (Uint32) ESTACK_POP(*sp); + hval = MAP_HEADER_VAL(hdr); + nhp = hp; + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hval | bp); ptr++; + + n -= slot; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + if (hval & bp) { ptr++; n--; } + while(n--) { *hp++ = *ptr++; } + + res = make_hashmap(nhp); + break; + case HAMT_SUBTAG_HEAD_BITMAP: + slot = (Uint) ESTACK_POP(*sp); + bp = (Uint32) ESTACK_POP(*sp); + n = (Uint32) ESTACK_POP(*sp); + hval = MAP_HEADER_VAL(hdr); + nhp = hp; + *hp++ = MAP_HEADER_HAMT_HEAD_BITMAP(hval | bp); ptr++; + *hp++ = (*ptr++) + *update_size; + + n -= slot; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + if (hval & bp) { ptr++; n--; } + while(n--) { *hp++ = *ptr++; } + + if ((hval | bp) == 0xffff) { + *nhp = MAP_HEADER_HAMT_HEAD_ARRAY; + } + res = make_hashmap(nhp); + break; + default: + erl_exit(1, "bad header tag %x\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + break; + default: + erl_exit(1, "bad primary tag %x\r\n", primary_tag(node)); + break; + } + + } while(!ESTACK_ISEMPTY(*sp)); + + UnUseTmpHeapNoproc(1); + return res; +} + +static Eterm hashmap_keys(Process* p, Eterm node) { + DECLARE_WSTACK(stack); + hashmap_head_t* root; + Eterm *hp, *kv; + Eterm res = NIL; + + root = (hashmap_head_t*) boxed_val(node); + hp = HAlloc(p, root->size * 2); + hashmap_iterator_init(&stack, node, 0); + while ((kv=hashmap_iterator_next(&stack)) != NULL) { + res = CONS(hp, CAR(kv), res); + hp += 2; + } + DESTROY_WSTACK(stack); + return res; +} + +static Eterm hashmap_values(Process* p, Eterm node) { + DECLARE_WSTACK(stack); + hashmap_head_t* root; + Eterm *hp, *kv; + Eterm res = NIL; + + root = (hashmap_head_t*) boxed_val(node); + hp = HAlloc(p, root->size * 2); + hashmap_iterator_init(&stack, node, 0); + while ((kv=hashmap_iterator_next(&stack)) != NULL) { + res = CONS(hp, CDR(kv), res); + hp += 2; + } + DESTROY_WSTACK(stack); + return res; +} + +static Eterm hashmap_delete(Process *p, Uint32 hx, Eterm key, Eterm map) { + Eterm *hp = NULL, *nhp = NULL, *hp_end = NULL; + Eterm th[2]; + Eterm *ptr; + Eterm hdr, res = map, node = map; + Uint32 ix, bp, hval; + Uint slot, lvl = 0; + Uint size = 0, n = 0; + DECLARE_ESTACK(stack); + + for (;;) { + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: + if (EQ(CAR(list_val(node)), key)) { + goto unroll; + } + goto not_found; + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + ix = hashmap_index(hx); + hx = hashmap_shift_hash(th,hx,lvl,key); + size += HAMT_HEAD_ARRAY_SZ; + ESTACK_PUSH2(stack, ix, node); + node = ptr[ix+2]; + break; + case HAMT_SUBTAG_NODE_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH4(stack, n, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+1]; + ASSERT(HAMT_NODE_BITMAP_SZ(n) <= 17); + size += HAMT_NODE_BITMAP_SZ(n); + break; + } + /* not occupied */ + goto not_found; + case HAMT_SUBTAG_HEAD_BITMAP: + hval = MAP_HEADER_VAL(hdr); + ix = hashmap_index(hx); + bp = 1 << ix; + slot = hashmap_bitcount(hval & (bp - 1)); + n = hashmap_bitcount(hval); + + ESTACK_PUSH4(stack, n, bp, slot, node); + + /* occupied */ + if (bp & hval) { + hx = hashmap_shift_hash(th,hx,lvl,key); + node = ptr[slot+2]; + ASSERT(HAMT_HEAD_BITMAP_SZ(n) <= 18); + size += HAMT_HEAD_BITMAP_SZ(n); + break; + } + /* not occupied */ + goto not_found; + default: + erl_exit(1, "bad header tag %ld\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + break; + default: + erl_exit(1, "bad primary tag %p\r\n", node); + break; + } + } + +unroll: + /* the size is bounded and atleast one less than the previous size */ + size -= 1; + n = hashmap_size(map) - 1; + + if (n <= MAP_SMALL_MAP_LIMIT) { + DECLARE_WSTACK(wstack); + Eterm *kv, *ks, *vs; + flatmap_t *mp; + Eterm keys; + + DESTROY_ESTACK(stack); + + /* build flat structure */ + hp = HAlloc(p, 3 + 1 + (2 * n)); + keys = make_tuple(hp); + *hp++ = make_arityval(n); + ks = hp; + hp += n; + mp = (flatmap_t*)hp; + hp += MAP_HEADER_FLATMAP_SZ; + vs = hp; + + mp->thing_word = MAP_HEADER_FLATMAP; + mp->size = n; + mp->keys = keys; + + hashmap_iterator_init(&wstack, map, 0); + + while ((kv=hashmap_iterator_next(&wstack)) != NULL) { + if (EQ(CAR(kv),key)) + continue; + *ks++ = CAR(kv); + *vs++ = CDR(kv); + } + + /* it cannot have multiple keys */ + erts_validate_and_sort_flatmap(mp); + + DESTROY_WSTACK(wstack); + return make_flatmap(mp); + } + + hp = HAlloc(p, size); + hp_end = hp + size; + res = THE_NON_VALUE; + + do { + node = ESTACK_POP(stack); + + /* all nodes are things */ + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + ix = (Uint) ESTACK_POP(stack); + nhp = hp; + if (res == THE_NON_VALUE) { + n = 16; + n -= ix; + *hp++ = MAP_HEADER_HAMT_HEAD_BITMAP(0xffff ^ (1 << ix)); ptr++; + *hp++ = (*ptr++) - 1; + while(ix--) { *hp++ = *ptr++; } + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + res = make_hashmap(nhp); + } else { + n = 16; + *hp++ = MAP_HEADER_HAMT_HEAD_ARRAY; ptr++; + *hp++ = (*ptr++) - 1; + while(n--) { *hp++ = *ptr++; } + nhp[ix+2] = res; + res = make_hashmap(nhp); + } + break; + case HAMT_SUBTAG_NODE_BITMAP: + slot = (Uint) ESTACK_POP(stack); + bp = (Uint32) ESTACK_POP(stack); + n = (Uint32) ESTACK_POP(stack); + nhp = hp; + + /* bitmap change matrix + * res | none leaf bitmap + * ---------------------------- + * n=1 | remove remove keep + * n=2 | other keep keep + * n>2 | shrink keep keep + * + * other: (remember, n is 2) + * shrink if the other bitmap value is a bitmap node + * remove if the other bitmap value is a leaf + * + * remove: + * this bitmap node is removed, res is moved up in tree (could be none) + * this is a special case of shrink + * + * keep: + * the current path index is still used down in the tree, need to keep it + * copy as usual with the updated res + * + * shrink: + * the current path index is no longer used down in the tree, remove it (shrink) + */ + if (res == THE_NON_VALUE) { + if (n == 1) { + break; + } else if (n == 2) { + if (slot == 0) { + ix = 2; /* off by one 'cause hdr */ + } else { + ix = 1; /* off by one 'cause hdr */ + } + if (primary_tag(ptr[ix]) == TAG_PRIMARY_LIST) { + res = ptr[ix]; + } else { + hval = MAP_HEADER_VAL(hdr); + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hval ^ bp); + *hp++ = ptr[ix]; + res = make_hashmap(nhp); + } + } else { + /* n > 2 */ + hval = MAP_HEADER_VAL(hdr); + *hp++ = MAP_HEADER_HAMT_NODE_BITMAP(hval ^ bp); ptr++; + n -= slot; + while(slot--) { *hp++ = *ptr++; } + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + res = make_hashmap(nhp); + } + } else if (primary_tag(res) == TAG_PRIMARY_LIST && n == 1) { + break; + } else { + /* res is bitmap or leaf && n > 1, keep */ + n -= slot; + *hp++ = *ptr++; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + res = make_hashmap(nhp); + } + break; + case HAMT_SUBTAG_HEAD_BITMAP: + slot = (Uint) ESTACK_POP(stack); + bp = (Uint32) ESTACK_POP(stack); + n = (Uint32) ESTACK_POP(stack); + nhp = hp; + + if (res != THE_NON_VALUE) { + *hp++ = *ptr++; + *hp++ = (*ptr++) - 1; + n -= slot; + while(slot--) { *hp++ = *ptr++; } + *hp++ = res; + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + } else { + hval = MAP_HEADER_VAL(hdr); + *hp++ = MAP_HEADER_HAMT_HEAD_BITMAP(hval ^ bp); ptr++; + *hp++ = (*ptr++) - 1; + n -= slot; + while(slot--) { *hp++ = *ptr++; } + ptr++; n--; + while(n--) { *hp++ = *ptr++; } + } + res = make_hashmap(nhp); + break; + default: + erl_exit(1, "bad header tag %x\r\n", hdr & _HEADER_MAP_SUBTAG_MASK); + break; + } + } while(!ESTACK_ISEMPTY(stack)); + HRelease(p, hp_end, hp); +not_found: + DESTROY_ESTACK(stack); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(p); + ERTS_HOLE_CHECK(p); + return res; +} + + +int erts_validate_and_sort_flatmap(flatmap_t* mp) +{ + Eterm *ks = flatmap_get_keys(mp); + Eterm *vs = flatmap_get_values(mp); + Uint sz = flatmap_get_size(mp); Uint ix,jx; Eterm tmp; - int c; + Sint c; /* sort */ @@ -819,3 +2442,277 @@ int erts_validate_and_sort_map(map_t* mp) } return 1; } + +/* Really rough estimate of sqrt(x) + * Guaranteed not to be less than sqrt(x) + */ +static int int_sqrt_ceiling(Uint x) +{ + int n; + + if (x <= 2) + return x; + + n = erts_fit_in_bits_uint(x-1); + if (n & 1) { + /* Calc: sqrt(2^n) = 2^(n/2) * sqrt(2) ~= 2^(n/2) * 3 / 2 */ + return (1 << (n/2 - 1)) * 3; + } + else { + /* Calc: sqrt(2^n) = 2^(n/2) */ + return 1 << (n / 2); + } +} + +Uint hashmap_over_estimated_heap_size(Uint k) +{ + /* k is nr of key-value pairs. + N(k) is expected nr of nodes in hamt. + + Observation: + For uniformly distributed hash values, average of N varies between + 0.3*k and 0.4*k (with a beautiful sine curve) + and standard deviation of N is about sqrt(k)/3. + + Assuming normal probability distribution, we overestimate nr of nodes + by 15 std.devs above the average, which gives a probability for overrun + less than 1.0e-49 (same magnitude as a git SHA1 collision). + */ + Uint max_nodes = 2*k/5 + (15/3)*int_sqrt_ceiling(k); + return (k*2 + /* leaf cons cells */ + k + /* leaf list terms */ + max_nodes*2); /* headers + parent boxed terms */ +} + + +BIF_RETTYPE erts_debug_map_info_1(BIF_ALIST_1) { + if (is_hashmap(BIF_ARG_1)) { + BIF_RET(hashmap_info(BIF_P,BIF_ARG_1)); + } + BIF_ERROR(BIF_P, BADARG); +} + +/* + * erts_internal:map_to_tuple_keys/1 + * + * Used in erts_debug:size/1 + */ + +BIF_RETTYPE erts_internal_map_to_tuple_keys_1(BIF_ALIST_1) { + if (is_flatmap(BIF_ARG_1)) { + flatmap_t *mp = (flatmap_t*)flatmap_val(BIF_ARG_1); + BIF_RET(mp->keys); + } + BIF_ERROR(BIF_P, BADARG); +} + +/* + * erts_internal:map_type/1 + * + * Used in erts_debug:size/1 + */ + +BIF_RETTYPE erts_internal_map_type_1(BIF_ALIST_1) { + DECL_AM(hashmap); + DECL_AM(hashmap_node); + DECL_AM(flatmap); + if (is_flatmap(BIF_ARG_1)) { + BIF_RET(AM_flatmap); + } else if (is_hashmap(BIF_ARG_1)) { + Eterm hdr = *(boxed_val(BIF_ARG_1)); + ASSERT(is_header(hdr)); + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_HEAD_BITMAP: + BIF_RET(AM_hashmap); + case HAMT_SUBTAG_NODE_BITMAP: + BIF_RET(AM_hashmap_node); + default: + erl_exit(1, "bad header"); + } + } + BIF_ERROR(BIF_P, BADARG); +} + +/* + * erts_internal:map_hashmap_children/1 + * + * Used in erts_debug:size/1 + */ + +BIF_RETTYPE erts_internal_map_hashmap_children_1(BIF_ALIST_1) { + if (is_hashmap(BIF_ARG_1)) { + Eterm node = BIF_ARG_1; + Eterm *ptr, hdr, *hp, res = NIL; + Uint sz = 0; + ptr = boxed_val(node); + hdr = *ptr; + + ASSERT(is_header(hdr)); + + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ptr += 1; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ptr += 2; + break; + case HAMT_SUBTAG_HEAD_ARRAY: + sz = 16; + ptr += 2; + break; + default: + erl_exit(1, "bad header\r\n"); + break; + } + ASSERT(sz < 17); + hp = HAlloc(BIF_P, 2*sz); + while(sz--) { res = CONS(hp, *ptr++, res); hp += 2; } + BIF_RET(res); + } + BIF_ERROR(BIF_P, BADARG); +} + + +static Eterm hashmap_info(Process *p, Eterm node) { + Eterm *hp; + Eterm res = NIL, info = NIL; + Eterm *ptr, tup, hdr; + Uint sz; + DECL_AM(depth); + DECL_AM(leafs); + DECL_AM(bitmaps); + DECL_AM(arrays); + Uint nleaf=0, nbitmap=0, narray=0; + Uint bitmap_usage[16], leaf_usage[16]; + Uint lvl = 0, clvl; + DECLARE_ESTACK(stack); + + for (sz = 0; sz < 16; sz++) { + bitmap_usage[sz] = 0; + leaf_usage[sz] = 0; + } + + ptr = boxed_val(node); + ESTACK_PUSH(stack, 0); + ESTACK_PUSH(stack, node); + do { + node = ESTACK_POP(stack); + clvl = ESTACK_POP(stack); + if (lvl < clvl) + lvl = clvl; + switch(primary_tag(node)) { + case TAG_PRIMARY_LIST: + nleaf++; + leaf_usage[clvl] += 1; + break; + case TAG_PRIMARY_BOXED: + ptr = boxed_val(node); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_NODE_BITMAP: + nbitmap++; + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz < 17); + bitmap_usage[sz-1] += 1; + while(sz--) { + ESTACK_PUSH(stack, clvl + 1); + ESTACK_PUSH(stack, ptr[sz+1]); + } + break; + case HAMT_SUBTAG_HEAD_BITMAP: + nbitmap++; + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + bitmap_usage[sz-1] += 1; + while(sz--) { + ESTACK_PUSH(stack, clvl + 1); + ESTACK_PUSH(stack, ptr[sz+2]); + } + break; + case HAMT_SUBTAG_HEAD_ARRAY: + narray++; + sz = 16; + while(sz--) { + ESTACK_PUSH(stack, clvl + 1); + ESTACK_PUSH(stack, ptr[sz+2]); + } + break; + default: + erl_exit(1, "bad header\r\n"); + break; + } + } + } while(!ESTACK_ISEMPTY(stack)); + + + /* size */ + sz = 0; + hashmap_bld_tuple_uint(NULL,&sz,16,leaf_usage); + hashmap_bld_tuple_uint(NULL,&sz,16,bitmap_usage); + + /* alloc */ + hp = HAlloc(p, 2+3 + 3*(2+4) + sz); + + info = hashmap_bld_tuple_uint(&hp,NULL,16,leaf_usage); + tup = TUPLE3(hp, AM_leafs, make_small(nleaf),info); hp += 4; + res = CONS(hp, tup, res); hp += 2; + + info = hashmap_bld_tuple_uint(&hp,NULL,16,bitmap_usage); + tup = TUPLE3(hp, AM_bitmaps, make_small(nbitmap), info); hp += 4; + res = CONS(hp, tup, res); hp += 2; + + tup = TUPLE3(hp, AM_arrays, make_small(narray),NIL); hp += 4; + res = CONS(hp, tup, res); hp += 2; + + tup = TUPLE2(hp, AM_depth, make_small(lvl)); hp += 3; + res = CONS(hp, tup, res); hp += 2; + + DESTROY_ESTACK(stack); + ERTS_HOLE_CHECK(p); + return res; +} + +static Eterm hashmap_bld_tuple_uint(Uint **hpp, Uint *szp, Uint n, Uint nums[]) { + Eterm res = THE_NON_VALUE; + Eterm *ts = (Eterm *)erts_alloc(ERTS_ALC_T_TMP, n * sizeof(Eterm)); + Uint i; + + for (i = 0; i < n; i++) { + ts[i] = erts_bld_uint(hpp, szp, nums[i]); + } + res = erts_bld_tuplev(hpp, szp, n, ts); + erts_free(ERTS_ALC_T_TMP, (void *) ts); + return res; +} + + +/* implementation of builtin emulations */ + +#if !ERTS_AT_LEAST_GCC_VSN__(3, 4, 0) +/* Count leading zeros emulation */ +Uint32 hashmap_clz(Uint32 x) { + Uint32 y; + int n = 32; + y = x >>16; if (y != 0) {n = n -16; x = y;} + y = x >> 8; if (y != 0) {n = n - 8; x = y;} + y = x >> 4; if (y != 0) {n = n - 4; x = y;} + y = x >> 2; if (y != 0) {n = n - 2; x = y;} + y = x >> 1; if (y != 0) return n - 2; + return n - x; +} + +const Uint32 SK5 = 0x55555555, SK3 = 0x33333333; +const Uint32 SKF0 = 0xF0F0F0F, SKFF = 0xFF00FF; + +/* CTPOP emulation */ +Uint32 hashmap_bitcount(Uint32 x) { + x -= ((x >> 1 ) & SK5); + x = (x & SK3 ) + ((x >> 2 ) & SK3 ); + x = (x & SKF0) + ((x >> 4 ) & SKF0); + x += x >> 8; + return (x + (x >> 16)) & 0x3F; +} +#endif diff --git a/erts/emulator/beam/erl_map.h b/erts/emulator/beam/erl_map.h index cfacb2ec28..2cc6768bfc 100644 --- a/erts/emulator/beam/erl_map.h +++ b/erts/emulator/beam/erl_map.h @@ -22,13 +22,23 @@ #define __ERL_MAP_H__ #include "sys.h" + +/* instrinsic wrappers */ +#if ERTS_AT_LEAST_GCC_VSN__(3, 4, 0) +#define hashmap_clz(x) ((Uint32) __builtin_clz((unsigned int)(x))) +#define hashmap_bitcount(x) ((Uint32) __builtin_popcount((unsigned int) (x))) +#else +Uint32 hashmap_clz(Uint32 x); +Uint32 hashmap_bitcount(Uint32 x); +#endif + /* MAP */ -typedef struct map_s { +typedef struct flatmap_s { Eterm thing_word; Uint size; Eterm keys; /* tuple */ -} map_t; +} flatmap_t; /* map node * * ----------- @@ -42,31 +52,144 @@ typedef struct map_s { * ----------- */ +/* the head-node is a bitmap or array with an untagged size */ + + +#define hashmap_size(x) (((hashmap_head_t*) hashmap_val(x))->size) +#define hashmap_size_rel(RTERM, BASE) hashmap_size(rterm2wterm(RTERM, BASE)) +#define hashmap_make_hash(Key) make_internal_hash(Key) + +#define hashmap_restore_hash(Heap,Lvl,Key) \ + (((Lvl) < 8) ? hashmap_make_hash(Key) >> (4*(Lvl)) : hashmap_make_hash(CONS(Heap, make_small((Lvl)>>3), (Key))) >> (4*((Lvl) & 7))) +#define hashmap_shift_hash(Heap,Hx,Lvl,Key) \ + (((++(Lvl)) & 7) ? (Hx) >> 4 : hashmap_make_hash(CONS(Heap, make_small((Lvl)>>3), Key))) /* erl_term.h stuff */ -#define make_map(x) make_boxed((Eterm*)(x)) -#define make_map_rel(x, BASE) make_boxed_rel((Eterm*)(x),(BASE)) -#define is_map(x) (is_boxed((x)) && is_map_header(*boxed_val((x)))) -#define is_map_rel(RTERM,BASE) is_map(rterm2wterm(RTERM,BASE)) -#define is_not_map(x) (!is_map((x))) -#define is_map_header(x) (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_MAP) -#define header_is_map(x) ((((x) & (_HEADER_SUBTAG_MASK)) == MAP_SUBTAG)) -#define map_val(x) (_unchecked_boxed_val((x))) -#define map_val_rel(RTERM, BASE) map_val(rterm2wterm(RTERM, BASE)) - -#define map_get_values(x) (((Eterm *)(x)) + 3) -#define map_get_keys(x) (((Eterm *)tuple_val(((map_t *)(x))->keys)) + 1) -#define map_get_size(x) (((map_t*)(x))->size) - -#define MAP_HEADER _make_header(1,_TAG_HEADER_MAP) -#define MAP_HEADER_SIZE (sizeof(map_t) / sizeof(Eterm)) - -Eterm erts_maps_put(Process *p, Eterm key, Eterm value, Eterm map); -int erts_maps_update(Process *p, Eterm key, Eterm value, Eterm map, Eterm *res); -int erts_maps_find(Eterm key, Eterm map, Eterm *value); -int erts_maps_get(Eterm key, Eterm map, Eterm *value); -int erts_maps_remove(Process *p, Eterm key, Eterm map, Eterm *res); -int erts_validate_and_sort_map(map_t* map); +#define flatmap_get_values(x) (((Eterm *)(x)) + 3) +#define flatmap_get_keys(x) (((Eterm *)tuple_val(((flatmap_t *)(x))->keys)) + 1) +#define flatmap_get_size(x) (((flatmap_t*)(x))->size) + +#ifdef DEBUG +#define MAP_SMALL_MAP_LIMIT (3) +#else +#define MAP_SMALL_MAP_LIMIT (32) +#endif + +struct ErtsWStack_; +struct ErtsEStack_; + +Eterm erts_maps_put(Process *p, Eterm key, Eterm value, Eterm map); +int erts_maps_update(Process *p, Eterm key, Eterm value, Eterm map, Eterm *res); +int erts_maps_remove(Process *p, Eterm key, Eterm map, Eterm *res); + +Eterm erts_hashmap_insert(Process *p, Uint32 hx, Eterm key, Eterm value, + Eterm node, int is_update); +int erts_hashmap_insert_down(Uint32 hx, Eterm key, Eterm node, Uint *sz, + Uint *upsz, struct ErtsEStack_ *sp, int is_update); +Eterm erts_hashmap_insert_up(Eterm *hp, Eterm key, Eterm value, + Uint *upsz, struct ErtsEStack_ *sp); + +int erts_validate_and_sort_flatmap(flatmap_t* map); +Uint hashmap_over_estimated_heap_size(Uint n); +void hashmap_iterator_init(struct ErtsWStack_* s, Eterm node, int reverse); +Eterm* hashmap_iterator_next(struct ErtsWStack_* s); +Eterm* hashmap_iterator_prev(struct ErtsWStack_* s); +int hashmap_key_hash_cmp(Eterm* ap, Eterm* bp); +Eterm erts_hashmap_from_array(ErtsHeapFactory*, Eterm *leafs, Uint n, int reject_dupkeys); + +#define erts_hashmap_from_ks_and_vs(P, KS, VS, N) \ + erts_hashmap_from_ks_and_vs_extra((P), (KS), (VS), (N), THE_NON_VALUE, THE_NON_VALUE); + +Eterm erts_hashmap_from_ks_and_vs_extra(Process *p, Eterm *ks, Eterm *vs, Uint n, + Eterm k, Eterm v); + +const Eterm * +#if HALFWORD_HEAP +erts_maps_get_rel(Eterm key, Eterm map, Eterm *map_base); +# define erts_maps_get(A, B) erts_maps_get_rel(A, B, NULL) +#else +erts_maps_get(Eterm key, Eterm map); +# define erts_maps_get_rel(A, B, B_BASE) erts_maps_get(A, B) +#endif + +const Eterm * +#if HALFWORD_HEAP +erts_hashmap_get_rel(Uint32 hx, Eterm key, Eterm node, Eterm *map_base); +# define erts_hashmap_get(Hx, K, M) erts_hashmap_get_rel(Hx, K, M, NULL) +#else +erts_hashmap_get(Uint32 hx, Eterm key, Eterm map); +# define erts_hashmap_get_rel(Hx, K, M, M_BASE) erts_hashmap_get(Hx, K, M) #endif +/* hamt nodes v2.0 + * + * node :: leaf | array | bitmap + * head + */ +typedef struct hashmap_head_s { + Eterm thing_word; + Uint size; + Eterm items[1]; +} hashmap_head_t; + +/* thing_word tagscheme + * Need two bits for map subtags + * + * Original HEADER representation: + * + * aaaaaaaaaaaaaaaa aaaaaaaaaatttt00 arity:26, tag:4 + * + * For maps we have: + * + * vvvvvvvvvvvvvvvv aaaaaaaamm111100 val:16, arity:8, mtype:2 + * + * unsure about trailing zeros + * + * map-tag: + * 00 - flat map tag (non-hamt) -> val:16 = #items + * 01 - map-node bitmap tag -> val:16 = bitmap + * 10 - map-head (array-node) -> val:16 = 0xffff + * 11 - map-head (bitmap-node) -> val:16 = bitmap + */ + +/* erl_map.h stuff */ + +#define is_hashmap_header_head(x) ((MAP_HEADER_TYPE(x) & (0x2))) + +#define MAKE_MAP_HEADER(Type,Arity,Val) \ + (_make_header(((((Uint16)(Val)) << MAP_HEADER_ARITY_SZ) | (Arity)) << MAP_HEADER_TAG_SZ | (Type) , _TAG_HEADER_MAP)) + +#define MAP_HEADER_FLATMAP \ + MAKE_MAP_HEADER(MAP_HEADER_TAG_FLATMAP_HEAD,0x1,0x0) + +#define MAP_HEADER_HAMT_HEAD_ARRAY \ + MAKE_MAP_HEADER(MAP_HEADER_TAG_HAMT_HEAD_ARRAY,0x1,0xffff) + +#define MAP_HEADER_HAMT_HEAD_BITMAP(Bmp) \ + MAKE_MAP_HEADER(MAP_HEADER_TAG_HAMT_HEAD_BITMAP,0x1,Bmp) + +#define MAP_HEADER_HAMT_NODE_BITMAP(Bmp) \ + MAKE_MAP_HEADER(MAP_HEADER_TAG_HAMT_NODE_BITMAP,0x0,Bmp) + +#define MAP_HEADER_FLATMAP_SZ (sizeof(flatmap_t) / sizeof(Eterm)) + +#define HAMT_NODE_ARRAY_SZ (17) +#define HAMT_HEAD_ARRAY_SZ (18) +#define HAMT_NODE_BITMAP_SZ(n) (1 + n) +#define HAMT_HEAD_BITMAP_SZ(n) (2 + n) + +/* 2 bits maps tag + 4 bits subtag + 2 ignore bits */ +#define _HEADER_MAP_SUBTAG_MASK (0xfc) +/* 1 bit map tag + 1 ignore bit + 4 bits subtag + 2 ignore bits */ +#define _HEADER_MAP_HASHMAP_HEAD_MASK (0xbc) + +#define HAMT_SUBTAG_NODE_BITMAP ((MAP_HEADER_TAG_HAMT_NODE_BITMAP << _HEADER_ARITY_OFFS) | MAP_SUBTAG) +#define HAMT_SUBTAG_HEAD_ARRAY ((MAP_HEADER_TAG_HAMT_HEAD_ARRAY << _HEADER_ARITY_OFFS) | MAP_SUBTAG) +#define HAMT_SUBTAG_HEAD_BITMAP ((MAP_HEADER_TAG_HAMT_HEAD_BITMAP << _HEADER_ARITY_OFFS) | MAP_SUBTAG) +#define HAMT_SUBTAG_HEAD_FLATMAP ((MAP_HEADER_TAG_FLATMAP_HEAD << _HEADER_ARITY_OFFS) | MAP_SUBTAG) + +#define hashmap_index(hash) (((Uint32)hash) & 0xf) + + +#endif diff --git a/erts/emulator/beam/erl_math.c b/erts/emulator/beam/erl_math.c index 16d4fdc09c..9b864628db 100644 --- a/erts/emulator/beam/erl_math.c +++ b/erts/emulator/beam/erl_math.c @@ -207,6 +207,24 @@ BIF_RETTYPE math_log_1(BIF_ALIST_1) return math_call_1(BIF_P, log, BIF_ARG_1); } +#ifdef HAVE_LOG2 +static double +log2_wrapper(double x) +{ + return log2(x); +} +#else +static double +log2_wrapper(double x) +{ + return log(x) / 0.6931471805599453; /* log(2.0); */ +} +#endif + +BIF_RETTYPE math_log2_1(BIF_ALIST_1) +{ + return math_call_1(BIF_P, log2_wrapper, BIF_ARG_1); +} BIF_RETTYPE math_log10_1(BIF_ALIST_1) { diff --git a/erts/emulator/beam/erl_message.c b/erts/emulator/beam/erl_message.c index 0eb8117980..22cbae10d1 100644 --- a/erts/emulator/beam/erl_message.c +++ b/erts/emulator/beam/erl_message.c @@ -415,7 +415,13 @@ erts_queue_dist_message(Process *rcvr, if (!(*rcvr_locks & ERTS_PROC_LOCK_MSGQ)) erts_smp_proc_unlock(rcvr, ERTS_PROC_LOCK_MSGQ); - erts_proc_notify_new_message(rcvr); + erts_proc_notify_new_message(rcvr, +#ifdef ERTS_SMP + *rcvr_locks +#else + 0 +#endif + ); } } @@ -542,7 +548,13 @@ queue_message(Process *c_p, if (locked_msgq) erts_smp_proc_unlock(receiver, ERTS_PROC_LOCK_MSGQ); - erts_proc_notify_new_message(receiver); + erts_proc_notify_new_message(receiver, +#ifdef ERTS_SMP + *receiver_locks +#else + 0 +#endif + ); #ifndef ERTS_SMP ERTS_HOLE_CHECK(receiver); @@ -982,7 +994,7 @@ erts_send_message(Process* sender, #endif ); BM_SWAP_TIMER(send,system); - } else if (sender == receiver) { + } else if (sender == receiver && !(sender->flags & F_OFF_HEAP_MSGS)) { /* Drop message if receiver has a pending exit ... */ #ifdef ERTS_SMP ErtsProcLocks need_locks = (~(*receiver_locks) @@ -1032,7 +1044,6 @@ erts_send_message(Process* sender, } BM_SWAP_TIMER(send,system); } else { -#ifdef ERTS_SMP ErlOffHeap *ohp; Eterm *hp; erts_aint32_t state; @@ -1064,42 +1075,6 @@ erts_send_message(Process* sender, #endif ); BM_SWAP_TIMER(send,system); -#else - ErlMessage* mp = message_alloc(); - Eterm *hp; - BM_SWAP_TIMER(send,size); - msize = size_object(message); - BM_SWAP_TIMER(size,send); - - if (receiver->stop - receiver->htop <= msize) { - BM_SWAP_TIMER(send,system); - erts_garbage_collect(receiver, msize, receiver->arg_reg, receiver->arity); - BM_SWAP_TIMER(system,send); - } - hp = receiver->htop; - receiver->htop = hp + msize; - BM_SWAP_TIMER(send,copy); - message = copy_struct(message, msize, &hp, &receiver->off_heap); - BM_MESSAGE_COPIED(msize); - BM_SWAP_TIMER(copy,send); - DTRACE6(message_send, sender_name, receiver_name, - (uint32_t)msize, tok_label, tok_lastcnt, tok_serial); - ERL_MESSAGE_TERM(mp) = message; - ERL_MESSAGE_TOKEN(mp) = NIL; -#ifdef USE_VM_PROBES - ERL_MESSAGE_DT_UTAG(mp) = NIL; -#endif - mp->next = NULL; - mp->data.attached = NULL; - LINK_MESSAGE(receiver, mp); - res = receiver->msg.len; - erts_proc_notify_new_message(receiver); - - if (IS_TRACED_FL(receiver, F_TRACE_RECEIVE)) { - trace_receive(receiver, message); - } - BM_SWAP_TIMER(send,system); -#endif /* #ifndef ERTS_SMP */ } return res; } @@ -1171,3 +1146,15 @@ erts_deliver_exit_message(Eterm from, Process *to, ErtsProcLocks *to_locksp, } } +Eterm* erts_produce_heap(ErtsHeapFactory* factory, Uint need, Uint xtra) +{ + Eterm* res; + if (factory->p) { + res = HAllocX(factory->p, need, xtra); + } else { + res = factory->hp; + factory->hp += need; + } + return res; +} + diff --git a/erts/emulator/beam/erl_message.h b/erts/emulator/beam/erl_message.h index 0f3bb8d281..8713941769 100644 --- a/erts/emulator/beam/erl_message.h +++ b/erts/emulator/beam/erl_message.h @@ -68,6 +68,21 @@ struct erl_heap_fragment { Eterm mem[1]; /* Data */ }; +typedef struct { + Process* p; + Eterm* hp; +} ErtsHeapFactory; + +Eterm* erts_produce_heap(ErtsHeapFactory*, Uint need, Uint xtra); +#ifdef CHECK_FOR_HOLES +# define ERTS_FACTORY_HOLE_CHECK(f) do { \ + if ((f)->p) erts_check_for_holes((f)->p); \ + } while (0) +#else +# define ERTS_FACTORY_HOLE_CHECK(p) +#endif + + typedef struct erl_mesg { struct erl_mesg* next; /* Next message */ union { @@ -198,15 +213,25 @@ do { \ if ((M)->data.attached) { \ Uint need__ = erts_msg_attached_data_size((M)); \ if ((ST) - (HT) >= need__) { \ - Uint *htop__ = (HT); \ + Uint *htop__; \ + move__attached__msg__data____: \ + htop__ = (HT); \ erts_move_msg_attached_data_to_heap(&htop__, &MSO((P)), (M));\ ASSERT(htop__ - (HT) <= need__); \ (HT) = htop__; \ } \ else { \ + int off_heap_msgs__ = (int) (P)->flags & F_OFF_HEAP_MSGS; \ + if (!off_heap_msgs__) \ + need__ = 0; \ { SWPO ; } \ - (FC) -= erts_garbage_collect((P), 0, NULL, 0); \ + (FC) -= erts_garbage_collect((P), need__, NULL, 0); \ { SWPI ; } \ + if (off_heap_msgs__) { \ + ASSERT((M)->data.attached); \ + ASSERT((ST) - (HT) >= need__); \ + goto move__attached__msg__data____; \ + } \ } \ ASSERT(!(M)->data.attached); \ } \ diff --git a/erts/emulator/beam/erl_monitors.h b/erts/emulator/beam/erl_monitors.h index fb11dbbd22..9972890db7 100644 --- a/erts/emulator/beam/erl_monitors.h +++ b/erts/emulator/beam/erl_monitors.h @@ -82,6 +82,7 @@ /* Type tags for monitors */ #define MON_ORIGIN 1 #define MON_TARGET 3 +#define MON_TIME_OFFSET 7 /* Type tags for links */ #define LINK_PID 1 /* ...Or port */ @@ -103,7 +104,7 @@ typedef struct erts_monitor_or_link { typedef struct erts_monitor { struct erts_monitor *left, *right; Sint16 balance; - Uint16 type; /* MON_ORIGIN | MON_TARGET */ + Uint16 type; /* MON_ORIGIN | MON_TARGET | MON_TIME_OFFSET */ Eterm ref; Eterm pid; /* In case of distributed named monitor, this is the nodename atom in MON_ORIGIN process, otherwise a pid or diff --git a/erts/emulator/beam/erl_mtrace.c b/erts/emulator/beam/erl_mtrace.c index c8bb126687..fa1bde1c87 100644 --- a/erts/emulator/beam/erl_mtrace.c +++ b/erts/emulator/beam/erl_mtrace.c @@ -627,7 +627,7 @@ erts_mtrace_install_wrapper_functions(void) if (erts_mtrace_enabled) { int i; /* Install trace functions */ - ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); + ERTS_CT_ASSERT(sizeof(erts_allctrs) == sizeof(real_allctrs)); sys_memcpy((void *) real_allctrs, (void *) erts_allctrs, diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c index ff551ea3af..660f446a52 100644 --- a/erts/emulator/beam/erl_nif.c +++ b/erts/emulator/beam/erl_nif.c @@ -36,6 +36,7 @@ #include "erl_thr_progress.h" #include "dtrace-wrapper.h" #include "erl_process.h" +#include "erl_bif_unique.h" #if defined(USE_DYNAMIC_TRACE) && (defined(USE_DTRACE) || defined(USE_SYSTEMTAP)) #define HAVE_USE_DTRACE 1 #endif @@ -472,6 +473,18 @@ int enif_inspect_binary(ErlNifEnv* env, Eterm bin_term, ErlNifBinary* bin) struct enif_tmp_obj_t* tmp; byte* raw_ptr; }u; + + if (is_boxed(bin_term) && *binary_val(bin_term) == HEADER_SUB_BIN) { + ErlSubBin* sb = (ErlSubBin*) binary_val(bin_term); + if (sb->is_writable) { + ProcBin* pb = (ProcBin*) binary_val(sb->orig); + ASSERT(pb->thing_word == HEADER_PROC_BIN); + if (pb->flags) { + erts_emasculate_writable_binary(pb); + sb->is_writable = 0; + } + } + } u.tmp = NULL; bin->data = erts_get_aligned_binary_bytes_extra(bin_term, &u.raw_ptr, allocator, sizeof(struct enif_tmp_obj_t)); @@ -539,9 +552,7 @@ int enif_alloc_binary(size_t size, ErlNifBinary* bin) if (refbin == NULL) { return 0; /* The NIF must take action */ } - refbin->flags = BIN_FLAG_DRV; /* BUGBUG: Flag? */ erts_refc_init(&refbin->refc, 1); - refbin->orig_size = (SWord) size; bin->size = size; bin->data = (unsigned char*) refbin->orig_bytes; @@ -561,7 +572,6 @@ int enif_realloc_binary(ErlNifBinary* bin, size_t size) if (!newbin) { return 0; } - newbin->orig_size = size; bin->ref_bin = newbin; bin->data = (unsigned char*) newbin->orig_bytes; bin->size = size; @@ -1513,72 +1523,264 @@ int enif_consume_timeslice(ErlNifEnv* env, int percent) return ERTS_BIF_REDS_LEFT(env->proc) == 0; } -#ifdef ERTS_DIRTY_SCHEDULERS - -/* NIFs exports need one more item than the Export struct provides, the - * erl_module_nif*, so the DirtyNifExport below adds that. The Export - * member must be first in the struct. +/* + * NIF exports need a few more items than the Export struct provides, + * including the erl_module_nif* and a NIF function pointer, so the + * NifExport below adds those. The Export member must be first in the + * struct. The saved_mfa, saved_argc, nif_level, alloced_argv_sz and argv + * members are used to track the MFA and arguments of the top NIF in case a + * chain of one or more enif_schedule_nif() calls results in an exception, + * since in that case the original MFA and registers have to be restored + * before returning to Erlang to ensure stacktrace information associated + * with the exception is correct. */ +typedef ERL_NIF_TERM (*NativeFunPtr)(ErlNifEnv*, int, const ERL_NIF_TERM[]); + typedef struct { Export exp; struct erl_module_nif* m; -} DirtyNifExport; + NativeFunPtr fp; + Eterm saved_mfa[3]; + int saved_argc; + int alloced_argv_sz; + Eterm argv[1]; +} NifExport; -static void -alloc_proc_psd(Process* proc, DirtyNifExport **ep) +/* + * If a process has saved arguments, they need to be part of the GC + * rootset. The function below is called from setup_rootset() in + * erl_gc.c. This function is declared in erl_process.h. + */ +int +erts_setup_nif_gc(Process* proc, Eterm** objv, int* nobj) +{ + NifExport* ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + int gc = (ep && ep->saved_argc > 0); + + if (gc) { + *objv = ep->argv; + *nobj = ep->saved_argc; + } + return gc; +} + +/* + * Allocate a NifExport and set it in proc specific data + */ +static NifExport* +allocate_nif_sched_data(Process* proc, int argc) { + NifExport* ep; + size_t argv_extra, total; int i; - if (!*ep) { - *ep = erts_alloc(ERTS_ALC_T_PSD, sizeof(DirtyNifExport)); - sys_memset((void*) *ep, 0, sizeof(DirtyNifExport)); - for (i=0; i<ERTS_NUM_CODE_IX; i++) { - (*ep)->exp.addressv[i] = &(*ep)->exp.code[3]; - } - (*ep)->exp.code[3] = (BeamInstr) em_call_nif; + + argv_extra = argc > 1 ? sizeof(Eterm)*(argc-1) : 0; + total = sizeof(NifExport) + argv_extra; + ep = erts_alloc(ERTS_ALC_T_NIF_TRAP_EXPORT, total); + sys_memset((void*) ep, 0, total); + ep->alloced_argv_sz = argc; + for (i=0; i<ERTS_NUM_CODE_IX; i++) { + ep->exp.addressv[i] = &ep->exp.code[3]; } - (void) ERTS_PROC_SET_DIRTY_SCHED_TRAP_EXPORT(proc, ERTS_PROC_LOCK_MAIN, &(*ep)->exp); + ep->exp.code[3] = (BeamInstr) em_call_nif; + (void) ERTS_PROC_SET_NIF_TRAP_EXPORT(proc, ERTS_PROC_LOCK_MAIN, ep); + return ep; +} + +static ERTS_INLINE void +destroy_nif_export(NifExport *nif_export) +{ + erts_free(ERTS_ALC_T_NIF_TRAP_EXPORT, (void *) nif_export); +} + +void +erts_destroy_nif_export(void *nif_export) +{ + destroy_nif_export((NifExport *) nif_export); } +/* + * Initialize a NifExport struct. Create it if needed and store it in the + * proc. The direct_fp function is what will be invoked by op_call_nif, and + * the indirect_fp function, if not NULL, is what the direct_fp function + * will call. If the allocated NifExport isn't enough to hold all of argv, + * allocate a larger one. Save MFA and registers only if the need_save + * parameter is true. + */ static ERL_NIF_TERM -execute_dirty_nif_finalizer(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +init_nif_sched_data(ErlNifEnv* env, NativeFunPtr direct_fp, NativeFunPtr indirect_fp, + int need_save, int argc, const ERL_NIF_TERM argv[]) { - Eterm* reg = ERTS_PROC_GET_SCHDATA(env->proc)->x_reg_array; - ERL_NIF_TERM result, dirty_result = (ERL_NIF_TERM) reg[0]; - typedef ERL_NIF_TERM (*FinalizerFP)(ErlNifEnv*, ERL_NIF_TERM); - FinalizerFP fp; -#if HAVE_INT64 && SIZEOF_LONG != 8 - ASSERT(sizeof(fp) <= sizeof(ErlNifUInt64)); - enif_get_uint64(env, reg[1], (ErlNifUInt64 *) &fp); -#else - ASSERT(sizeof(fp) <= sizeof(unsigned long)); - enif_get_ulong(env, reg[1], (unsigned long *) &fp); -#endif - result = (*fp)(env, dirty_result); - if (erts_refc_dectest(&env->mod_nif->rt_dtor_cnt, 0) == 0 - && env->mod_nif->mod == NULL) - close_lib(env->mod_nif); - return result; + Process* proc = env->proc; + Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; + NifExport* ep; + int i; + + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + if (!ep) + ep = allocate_nif_sched_data(proc, argc); + else if (need_save && ep->alloced_argv_sz < argc) { + NifExport* new_ep = allocate_nif_sched_data(proc, argc); + destroy_nif_export(ep); + ep = new_ep; + } + ERTS_VBUMP_ALL_REDS(proc); + for (i = 0; i < argc; i++) { + if (need_save) + ep->argv[i] = reg[i]; + reg[i] = (Eterm) argv[i]; + } + if (need_save) { + ep->saved_mfa[0] = proc->current[0]; + ep->saved_mfa[1] = proc->current[1]; + ep->saved_mfa[2] = proc->current[2]; + ep->saved_argc = argc; + } + proc->i = (BeamInstr*) ep->exp.addressv[0]; + ep->exp.code[0] = (BeamInstr) proc->current[0]; + ep->exp.code[1] = (BeamInstr) proc->current[1]; + ep->exp.code[2] = argc; + ep->exp.code[4] = (BeamInstr) direct_fp; + ep->m = env->mod_nif; + ep->fp = indirect_fp; + proc->freason = TRAP; + proc->arity = argc; + return THE_NON_VALUE; } -#endif /* ERTS_DIRTY_SCHEDULERS */ +/* + * Restore saved MFA and registers. Registers are restored only when the + * exception flag is true. + */ +static void +restore_nif_mfa(Process* proc, NifExport* ep, int exception) +{ + int i; + Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; -#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + proc->current[0] = ep->saved_mfa[0]; + proc->current[1] = ep->saved_mfa[1]; + proc->current[2] = ep->saved_mfa[2]; + if (exception) + for (i = 0; i < ep->saved_argc; i++) + reg[i] = ep->argv[i]; + ep->saved_argc = 0; + ep->saved_mfa[0] = THE_NON_VALUE; +} -ERL_NIF_TERM -enif_schedule_dirty_nif(ErlNifEnv* env, int flags, - ERL_NIF_TERM (*fp)(ErlNifEnv*, int, const ERL_NIF_TERM[]), - int argc, const ERL_NIF_TERM argv[]) +#ifdef ERTS_DIRTY_SCHEDULERS + +/* + * Finalize a dirty NIF call. This function is scheduled to cause the VM to + * switch the process off a dirty scheduler thread and back onto a regular + * scheduler thread, and then return the result from the dirty NIF. It also + * restores the original NIF MFA when necessary based on the value of + * ep->fp set by execute_dirty_nif via init_nif_sched_data -- non-NULL + * means restore, NULL means do not restore. + */ +static ERL_NIF_TERM +dirty_nif_finalizer(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NifExport* ep; + + ASSERT(argc == 1); + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data)); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + if (ep->fp) + restore_nif_mfa(proc, ep, 0); + return argv[0]; +} + +/* Finalize a dirty NIF call that raised an exception. Otherwise same as + * the dirty_nif_finalizer() function. + */ +static ERL_NIF_TERM +dirty_nif_exception(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NifExport* ep; + + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data)); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + if (ep->fp) + restore_nif_mfa(proc, ep, 1); + return enif_make_badarg(env); +} + +/* + * Dirty NIF execution wrapper function. Invoke an application's dirty NIF, + * then check the result and schedule the appropriate finalizer function + * where needed. Also restore the original NIF MFA when appropriate. + */ +static ERL_NIF_TERM +execute_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NativeFunPtr fp = (NativeFunPtr) proc->current[6]; + NifExport* ep; + ERL_NIF_TERM result; + + ASSERT(ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data)); + + /* + * Set ep->fp to NULL before the native call so we know later whether it scheduled another NIF for execution + */ + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + ep->fp = NULL; + result = (*fp)(env, argc, argv); + erts_smp_atomic32_read_band_mb(&proc->state, + ~(ERTS_PSFLG_DIRTY_CPU_PROC + |ERTS_PSFLG_DIRTY_IO_PROC + |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q + |ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); + if (erts_refc_dectest(&env->mod_nif->rt_dtor_cnt, 0) == 0 && env->mod_nif->mod == NULL) + close_lib(env->mod_nif); + /* + * If no more NIFs were scheduled by the native call via + * enif_schedule_nif(), then ep->fp will still be NULL as set above, in + * which case we need to restore the original NIF calling + * context. Reuse fp essentially as a boolean for this, passing it to + * init_nif_sched_data below. Both dirty_nif_exception and + * dirty_nif_finalizer then check ep->fp to decide whether or not to + * restore the original calling context. + */ + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + if (ep->fp) + fp = NULL; + if (is_non_value(result)) { + if (proc->freason != TRAP) { + ASSERT(proc->freason == BADARG); + return init_nif_sched_data(env, dirty_nif_exception, fp, 0, argc, argv); + } else { + if (ep->fp == NULL) + restore_nif_mfa(proc, ep, 1); + return result; + } + } + else + return init_nif_sched_data(env, dirty_nif_finalizer, fp, 0, 1, &result); +} + +/* + * Dirty NIF scheduling wrapper function. Schedule a dirty NIF to execute + * via the execute_dirty_nif() wrapper function. The dirty scheduler thread + * type (CPU or I/O) is indicated in flags parameter. + */ +static ERTS_INLINE ERL_NIF_TERM +schedule_dirty_nif(ErlNifEnv* env, int flags, int argc, const ERL_NIF_TERM argv[]) { -#ifdef USE_THREADS erts_aint32_t state, n, a; Process* proc = env->proc; - Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; - DirtyNifExport* ep = NULL; - int i; + NativeFunPtr fp = (NativeFunPtr) proc->current[6]; + NifExport* ep; + int need_save; - int chkflgs = (flags & (ERL_NIF_DIRTY_JOB_IO_BOUND|ERL_NIF_DIRTY_JOB_CPU_BOUND)); - if (chkflgs != ERL_NIF_DIRTY_JOB_IO_BOUND && chkflgs != ERL_NIF_DIRTY_JOB_CPU_BOUND) - return enif_make_badarg(env); + ASSERT(flags==ERL_NIF_DIRTY_JOB_IO_BOUND || flags==ERL_NIF_DIRTY_JOB_CPU_BOUND); a = erts_smp_atomic32_read_acqb(&proc->state); while (1) { @@ -1590,7 +1792,7 @@ enif_schedule_dirty_nif(ErlNifEnv* env, int flags, */ n &= ~(ERTS_PSFLG_DIRTY_CPU_PROC|ERTS_PSFLG_DIRTY_IO_PROC |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q); - if (chkflgs == ERL_NIF_DIRTY_JOB_CPU_BOUND) + if (flags == ERL_NIF_DIRTY_JOB_CPU_BOUND) n |= ERTS_PSFLG_DIRTY_CPU_PROC; else n |= ERTS_PSFLG_DIRTY_IO_PROC; @@ -1598,85 +1800,106 @@ enif_schedule_dirty_nif(ErlNifEnv* env, int flags, if (a == state) break; } - if (!(ep = (DirtyNifExport*) ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(proc))) - alloc_proc_psd(proc, &ep); - ERTS_VBUMP_ALL_REDS(proc); - ep->exp.code[2] = argc; - for (i = 0; i < argc; i++) { - reg[i] = (Eterm) argv[i]; - } - proc->i = (BeamInstr*) ep->exp.addressv[0]; - ep->exp.code[4] = (BeamInstr) fp; - ep->m = env->mod_nif; - proc->freason = TRAP; - erts_refc_inc(&env->mod_nif->rt_dtor_cnt, 1); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + need_save = (ep == NULL || is_non_value(ep->saved_mfa[0])); + return init_nif_sched_data(env, execute_dirty_nif, fp, need_save, argc, argv); +} - return THE_NON_VALUE; -#else - return (*fp)(env, argc, argv); -#endif +static ERL_NIF_TERM +schedule_dirty_io_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + return schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_IO_BOUND, argc, argv); +} + +static ERL_NIF_TERM +schedule_dirty_cpu_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + return schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_CPU_BOUND, argc, argv); +} + +#endif /* ERTS_DIRTY_SCHEDULERS */ + +/* + * NIF execution wrapper used by enif_schedule_nif() for regular NIFs. It + * calls the actual NIF, restores original NIF MFA if necessary, and + * then returns the NIF result. + */ +static ERL_NIF_TERM +execute_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NativeFunPtr fp = (NativeFunPtr) proc->current[6]; + NifExport* ep; + ERL_NIF_TERM result; + + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + ep->fp = NULL; + result = (*fp)(env, argc, argv); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + /* + * If no NIFs were scheduled by the native call via + * enif_schedule_nif(), then ep->fp will still be NULL as set above, in + * which case we need to restore the original NIF MFA. + */ + if (ep->fp == NULL) + restore_nif_mfa(proc, ep, is_non_value(result) && proc->freason != TRAP); + return result; } ERL_NIF_TERM -enif_schedule_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result, - ERL_NIF_TERM (*fp)(ErlNifEnv*, ERL_NIF_TERM)) +enif_schedule_nif(ErlNifEnv* env, const char* fun_name, int flags, + ERL_NIF_TERM (*fp)(ErlNifEnv*, int, const ERL_NIF_TERM[]), + int argc, const ERL_NIF_TERM argv[]) { -#ifdef USE_THREADS Process* proc = env->proc; - Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; - DirtyNifExport* ep; + NifExport* ep; + ERL_NIF_TERM fun_name_atom, result; + int need_save; - erts_smp_atomic32_read_band_mb(&proc->state, - ~(ERTS_PSFLG_DIRTY_CPU_PROC - |ERTS_PSFLG_DIRTY_IO_PROC - |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q - |ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); - if (!(ep = (DirtyNifExport*) ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(proc))) - alloc_proc_psd(proc, &ep); - ERTS_VBUMP_ALL_REDS(proc); - ep->exp.code[2] = 2; - reg[0] = (Eterm) result; -#if HAVE_INT64 && SIZEOF_LONG != 8 - ASSERT(sizeof(fp) <= sizeof(ErlNifUInt64)); - reg[1] = (Eterm) enif_make_uint64(env, (ErlNifUInt64) fp); -#else - ASSERT(sizeof(fp) <= sizeof(unsigned long)); - reg[1] = (Eterm) enif_make_ulong(env, (unsigned long) fp); -#endif - proc->i = (BeamInstr*) ep->exp.addressv[0]; - ep->exp.code[4] = (BeamInstr) execute_dirty_nif_finalizer; - proc->freason = TRAP; + if (argc > MAX_ARG) + return enif_make_badarg(env); + fun_name_atom = enif_make_atom(env, fun_name); + if (enif_is_exception(env, fun_name_atom)) + return fun_name_atom; - return THE_NON_VALUE; + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + need_save = (ep == NULL || is_non_value(ep->saved_mfa[0])); + + if (flags) { +#ifdef ERTS_DIRTY_SCHEDULERS + NativeFunPtr sched_fun; + int chkflgs = (flags & (ERL_NIF_DIRTY_JOB_IO_BOUND|ERL_NIF_DIRTY_JOB_CPU_BOUND)); + if (chkflgs == ERL_NIF_DIRTY_JOB_IO_BOUND) + sched_fun = schedule_dirty_io_nif; + else if (chkflgs == ERL_NIF_DIRTY_JOB_CPU_BOUND) + sched_fun = schedule_dirty_cpu_nif; + else + return enif_make_badarg(env); + result = init_nif_sched_data(env, sched_fun, fp, need_save, argc, argv); #else - return (*fp)(env, result); + return enif_make_badarg(env); #endif -} + } + else + result = init_nif_sched_data(env, execute_nif, fp, need_save, argc, argv); -/* A simple finalizer that just returns its result argument */ -ERL_NIF_TERM -enif_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result) -{ + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + ep->exp.code[1] = (BeamInstr) fun_name_atom; return result; } +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + int enif_is_on_dirty_scheduler(ErlNifEnv* env) { return ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data); } -int -enif_have_dirty_schedulers() -{ -#ifdef USE_THREADS - return 1; -#else - return 0; -#endif -} - #endif /* ERL_NIF_DIRTY_SCHEDULER_SUPPORT */ /* Maps */ @@ -1688,29 +1911,33 @@ int enif_is_map(ErlNifEnv* env, ERL_NIF_TERM term) int enif_get_map_size(ErlNifEnv* env, ERL_NIF_TERM term, size_t *size) { - if (is_map(term)) { - map_t *mp; - mp = (map_t*)map_val(term); - *size = map_get_size(mp); + if (is_flatmap(term)) { + flatmap_t *mp; + mp = (flatmap_t*)flatmap_val(term); + *size = flatmap_get_size(mp); return 1; } + else if (is_hashmap(term)) { + *size = hashmap_size(term); + return 1; + } return 0; } ERL_NIF_TERM enif_make_new_map(ErlNifEnv* env) { - Eterm* hp = alloc_heap(env,MAP_HEADER_SIZE+1); + Eterm* hp = alloc_heap(env,MAP_HEADER_FLATMAP_SZ+1); Eterm tup; - map_t *mp; + flatmap_t *mp; tup = make_tuple(hp); *hp++ = make_arityval(0); - mp = (map_t*)hp; - mp->thing_word = MAP_HEADER; + mp = (flatmap_t*)hp; + mp->thing_word = MAP_HEADER_FLATMAP; mp->size = 0; mp->keys = tup; - return make_map(mp); + return make_flatmap(mp); } int enif_make_map_put(ErlNifEnv* env, @@ -1719,7 +1946,7 @@ int enif_make_map_put(ErlNifEnv* env, Eterm value, Eterm *map_out) { - if (is_not_map(map_in)) { + if (!is_map(map_in)) { return 0; } flush_env(env); @@ -1733,10 +1960,16 @@ int enif_get_map_value(ErlNifEnv* env, Eterm key, Eterm *value) { - if (is_not_map(map)) { + const Eterm *ret; + if (!is_map(map)) { return 0; } - return erts_maps_get(key, map, value); + ret = erts_maps_get(key, map); + if (ret) { + *value = *ret; + return 1; + } + return 0; } int enif_make_map_update(ErlNifEnv* env, @@ -1746,7 +1979,7 @@ int enif_make_map_update(ErlNifEnv* env, Eterm *map_out) { int res; - if (is_not_map(map_in)) { + if (!is_map(map_in)) { return 0; } @@ -1762,7 +1995,7 @@ int enif_make_map_remove(ErlNifEnv* env, Eterm *map_out) { int res; - if (is_not_map(map_in)) { + if (!is_map(map_in)) { return 0; } flush_env(env); @@ -1776,13 +2009,13 @@ int enif_map_iterator_create(ErlNifEnv *env, ErlNifMapIterator *iter, ErlNifMapIteratorEntry entry) { - if (is_map(map)) { - map_t *mp = (map_t*)map_val(map); + if (is_flatmap(map)) { + flatmap_t *mp = (flatmap_t*)flatmap_val(map); size_t offset; switch (entry) { case ERL_NIF_MAP_ITERATOR_HEAD: offset = 0; break; - case ERL_NIF_MAP_ITERATOR_TAIL: offset = map_get_size(mp) - 1; break; + case ERL_NIF_MAP_ITERATOR_TAIL: offset = flatmap_get_size(mp) - 1; break; default: goto error; } @@ -1791,14 +2024,37 @@ int enif_map_iterator_create(ErlNifEnv *env, */ iter->map = map; - iter->ks = ((Eterm *)map_get_keys(mp)) + offset; - iter->vs = ((Eterm *)map_get_values(mp)) + offset; - iter->t_limit = map_get_size(mp) + 1; + iter->u.flat.ks = ((Eterm *)flatmap_get_keys(mp)) + offset; + iter->u.flat.vs = ((Eterm *)flatmap_get_values(mp)) + offset; + iter->size = flatmap_get_size(mp); iter->idx = offset + 1; return 1; } - + else if (is_hashmap(map)) { + iter->map = map; + iter->size = hashmap_size(map); + iter->u.hash.wstack = erts_alloc(ERTS_ALC_T_NIF, sizeof(ErtsDynamicWStack)); + WSTACK_INIT(iter->u.hash.wstack, ERTS_ALC_T_NIF); + + switch (entry) { + case ERL_NIF_MAP_ITERATOR_HEAD: + iter->idx = 1; + hashmap_iterator_init(&iter->u.hash.wstack->ws, map, 0); + iter->u.hash.kv = hashmap_iterator_next(&iter->u.hash.wstack->ws); + break; + case ERL_NIF_MAP_ITERATOR_TAIL: + iter->idx = hashmap_size(map); + hashmap_iterator_init(&iter->u.hash.wstack->ws, map, 1); + iter->u.hash.kv = hashmap_iterator_prev(&iter->u.hash.wstack->ws); + break; + default: + goto error; + } + ASSERT(!!iter->u.hash.kv == (iter->idx >= 1 && + iter->idx <= iter->size)); + return 1; + } error: #ifdef DEBUG iter->map = THE_NON_VALUE; @@ -1808,48 +2064,97 @@ error: void enif_map_iterator_destroy(ErlNifEnv *env, ErlNifMapIterator *iter) { - /* not used */ + if (is_hashmap(iter->map)) { + WSTACK_DESTROY(iter->u.hash.wstack->ws); + erts_free(ERTS_ALC_T_NIF, iter->u.hash.wstack); + } + else + ASSERT(is_flatmap(iter->map)); + #ifdef DEBUG iter->map = THE_NON_VALUE; #endif - } int enif_map_iterator_is_tail(ErlNifEnv *env, ErlNifMapIterator *iter) { - ASSERT(iter && is_map(iter->map)); - ASSERT(iter->idx >= 0 && (iter->idx <= map_get_size(map_val(iter->map)) + 1)); - return (iter->t_limit == 1 || iter->idx == iter->t_limit); + ASSERT(iter); + if (is_flatmap(iter->map)) { + ASSERT(iter->idx >= 0); + ASSERT(iter->idx <= flatmap_get_size(flatmap_val(iter->map)) + 1); + return (iter->size == 0 || iter->idx > iter->size); + } + else { + ASSERT(is_hashmap(iter->map)); + return iter->idx > iter->size; + } } int enif_map_iterator_is_head(ErlNifEnv *env, ErlNifMapIterator *iter) { - ASSERT(iter && is_map(iter->map)); - ASSERT(iter->idx >= 0 && (iter->idx <= map_get_size(map_val(iter->map)) + 1)); - return (iter->t_limit == 1 || iter->idx == 0); + ASSERT(iter); + if (is_flatmap(iter->map)) { + ASSERT(iter->idx >= 0); + ASSERT(iter->idx <= flatmap_get_size(flatmap_val(iter->map)) + 1); + return (iter->size == 0 || iter->idx == 0); + } + else { + ASSERT(is_hashmap(iter->map)); + return iter->idx == 0; + } } int enif_map_iterator_next(ErlNifEnv *env, ErlNifMapIterator *iter) { - ASSERT(iter && is_map(iter->map)); - if (iter->idx < iter->t_limit) { - iter->idx++; - iter->ks++; - iter->vs++; + ASSERT(iter); + if (is_flatmap(iter->map)) { + if (iter->idx <= iter->size) { + iter->idx++; + iter->u.flat.ks++; + iter->u.flat.vs++; + } + return (iter->idx <= iter->size); + } + else { + ASSERT(is_hashmap(iter->map)); + + if (iter->idx <= hashmap_size(iter->map)) { + if (iter->idx < 1) { + hashmap_iterator_init(&iter->u.hash.wstack->ws, iter->map, 0); + } + iter->u.hash.kv = hashmap_iterator_next(&iter->u.hash.wstack->ws); + iter->idx++; + ASSERT(!!iter->u.hash.kv == (iter->idx <= iter->size)); + } + return iter->idx <= iter->size; } - return (iter->idx != iter->t_limit); } int enif_map_iterator_prev(ErlNifEnv *env, ErlNifMapIterator *iter) { - ASSERT(iter && is_map(iter->map)); - if (iter->idx > 0) { - iter->idx--; - iter->ks--; - iter->vs--; + ASSERT(iter); + if (is_flatmap(iter->map)) { + if (iter->idx > 0) { + iter->idx--; + iter->u.flat.ks--; + iter->u.flat.vs--; + } + return iter->idx > 0; + } + else { + ASSERT(is_hashmap(iter->map)); + + if (iter->idx > 0) { + if (iter->idx > iter->size) { + hashmap_iterator_init(&iter->u.hash.wstack->ws, iter->map, 1); + } + iter->u.hash.kv = hashmap_iterator_prev(&iter->u.hash.wstack->ws); + iter->idx--; + ASSERT(!!iter->u.hash.kv == (iter->idx > 0)); + } + return iter->idx > 0; } - return (iter->idx > 0); } int enif_map_iterator_get_pair(ErlNifEnv *env, @@ -1857,15 +2162,25 @@ int enif_map_iterator_get_pair(ErlNifEnv *env, Eterm *key, Eterm *value) { - ASSERT(iter && is_map(iter->map)); - if (iter->idx > 0 && iter->idx < iter->t_limit) { - ASSERT(iter->ks >= map_get_keys(map_val(iter->map)) && - iter->ks < (map_get_keys(map_val(iter->map)) + map_get_size(map_val(iter->map)))); - ASSERT(iter->vs >= map_get_values(map_val(iter->map)) && - iter->vs < (map_get_values(map_val(iter->map)) + map_get_size(map_val(iter->map)))); - *key = *(iter->ks); - *value = *(iter->vs); - return 1; + ASSERT(iter); + if (is_flatmap(iter->map)) { + if (iter->idx > 0 && iter->idx <= iter->size) { + ASSERT(iter->u.flat.ks >= flatmap_get_keys(flatmap_val(iter->map)) && + iter->u.flat.ks < (flatmap_get_keys(flatmap_val(iter->map)) + flatmap_get_size(flatmap_val(iter->map)))); + ASSERT(iter->u.flat.vs >= flatmap_get_values(flatmap_val(iter->map)) && + iter->u.flat.vs < (flatmap_get_values(flatmap_val(iter->map)) + flatmap_get_size(flatmap_val(iter->map)))); + *key = *(iter->u.flat.ks); + *value = *(iter->u.flat.vs); + return 1; + } + } + else { + ASSERT(is_hashmap(iter->map)); + if (iter->idx > 0 && iter->idx <= iter->size) { + *key = CAR(iter->u.hash.kv); + *value = CDR(iter->u.hash.kv); + return 1; + } } return 0; } @@ -1977,6 +2292,35 @@ static Eterm load_nif_error(Process* p, const char* atom, const char* format, .. return ret; } +/* + * The function below is for looping through ErlNifFunc arrays, helping + * provide backwards compatibility across the version 2.7 change that added + * the "flags" field to ErlNifFunc. + */ +static ErlNifFunc* next_func(ErlNifEntry* entry, int* incrp, ErlNifFunc* func) +{ + ASSERT(incrp); + if (!*incrp) { + if (entry->major > 2 || (entry->major == 2 && entry->minor >= 7)) + *incrp = sizeof(ErlNifFunc); + else { + /* + * ErlNifFuncV1 below is what ErlNifFunc was before the + * addition of the flags field for 2.7, and is needed to handle + * backward compatibility. + */ + typedef struct { + const char* name; + unsigned arity; + ERL_NIF_TERM (*fptr)(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]); + }ErlNifFuncV1; + *incrp = sizeof(ErlNifFuncV1); + } + } + return (ErlNifFunc*) ((char*)func + *incrp); +} + + BIF_RETTYPE load_nif_2(BIF_ALIST_2) { static const char bad_lib[] = "bad_lib"; @@ -2086,22 +2430,48 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) } else { /*erts_fprintf(stderr, "Found module %T\r\n", mod_atom);*/ - + + int maybe_dirty_nifs = ((entry->major > 2 || (entry->major == 2 && entry->minor >= 7)) + && (entry->options & ERL_NIF_DIRTY_NIF_OPTION)); + int incr = 0; + ErlNifFunc* f = entry->funcs; for (i=0; i < entry->num_of_funcs && ret==am_ok; i++) { BeamInstr** code_pp; - ErlNifFunc* f = &entry->funcs[i]; if (!erts_atom_get(f->name, sys_strlen(f->name), &f_atom, ERTS_ATOM_ENC_LATIN1) || (code_pp = get_func_pp(mod->curr.code, f_atom, f->arity))==NULL) { ret = load_nif_error(BIF_P,bad_lib,"Function not found %T:%s/%u", mod_atom, f->name, f->arity); - } - else if (code_pp[1] - code_pp[0] < (5+3)) { + } + else if (maybe_dirty_nifs && f->flags) { + /* + * If the flags field is non-zero and this emulator was + * built with dirty scheduler support, check that the flags + * value is legal. But if this emulator was built without + * dirty scheduler support, treat a non-zero flags field as + * a load error. + */ +#ifdef ERTS_DIRTY_SCHEDULERS + if (f->flags != ERL_NIF_DIRTY_JOB_IO_BOUND && f->flags != ERL_NIF_DIRTY_JOB_CPU_BOUND) + ret = load_nif_error(BIF_P, bad_lib, "Illegal flags field value %d for NIF %T:%s/%u", + f->flags, mod_atom, f->name, f->arity); +#else + ret = load_nif_error(BIF_P, bad_lib, "NIF %T:%s/%u requires a runtime with dirty scheduler support.", + mod_atom, f->name, f->arity); +#endif + } +#ifdef ERTS_DIRTY_SCHEDULERS + else if (code_pp[1] - code_pp[0] < (5+4)) +#else + else if (code_pp[1] - code_pp[0] < (5+3)) +#endif + { ret = load_nif_error(BIF_P,bad_lib,"No explicit call to load_nif" - " in module (%T:%s/%u to small)", - mod_atom, entry->funcs[i].name, entry->funcs[i].arity); + " in module (%T:%s/%u too small)", + mod_atom, f->name, f->arity); } /*erts_fprintf(stderr, "Found NIF %T:%s/%u\r\n", - mod_atom, entry->funcs[i].name, entry->funcs[i].arity);*/ + mod_atom, f->name, f->arity);*/ + f = next_func(entry, &incr, f); } } @@ -2127,7 +2497,8 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) * is deprecated and was only ment as a development feature not to * be used in production systems. (See warning below) */ - int k; + int k, old_incr = 0; + ErlNifFunc* old_func; lib->priv_data = mod->curr.nif->priv_data; ASSERT(mod->curr.nif->entry != NULL); @@ -2136,13 +2507,16 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) goto error; } /* Check that no NIF is removed */ + old_func = mod->curr.nif->entry->funcs; for (k=0; k < mod->curr.nif->entry->num_of_funcs; k++) { - ErlNifFunc* old_func = &mod->curr.nif->entry->funcs[k]; + int incr = 0; + ErlNifFunc* f = entry->funcs; for (i=0; i < entry->num_of_funcs; i++) { - if (old_func->arity == entry->funcs[i].arity - && sys_strcmp(old_func->name, entry->funcs[i].name) == 0) { + if (old_func->arity == f->arity + && sys_strcmp(old_func->name, f->name) == 0) { break; } + f = next_func(entry, &incr, f); } if (i == entry->num_of_funcs) { ret = load_nif_error(BIF_P,reload,"Reloaded library missing " @@ -2150,7 +2524,8 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) old_func->name, old_func->arity); goto error; } - } + old_func = next_func(mod->curr.nif->entry, &old_incr, old_func); + } erts_pre_nif(&env, BIF_P, lib); veto = entry->reload(&env, &lib->priv_data, BIF_ARG_2); erts_post_nif(&env); @@ -2197,13 +2572,17 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) /* ** Everything ok, patch the beam code with op_call_nif */ - mod->curr.nif = lib; + + int incr = 0; + ErlNifFunc* f = entry->funcs; + + mod->curr.nif = lib; for (i=0; i < entry->num_of_funcs; i++) { BeamInstr* code_ptr; - erts_atom_get(entry->funcs[i].name, sys_strlen(entry->funcs[i].name), &f_atom, ERTS_ATOM_ENC_LATIN1); - code_ptr = *get_func_pp(mod->curr.code, f_atom, entry->funcs[i].arity); - + erts_atom_get(f->name, sys_strlen(f->name), &f_atom, ERTS_ATOM_ENC_LATIN1); + code_ptr = *get_func_pp(mod->curr.code, f_atom, f->arity); + if (code_ptr[1] == 0) { code_ptr[5+0] = (BeamInstr) BeamOp(op_call_nif); } @@ -2211,10 +2590,21 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) GenericBp* g = (GenericBp *) code_ptr[1]; ASSERT(code_ptr[5+0] == (BeamInstr) BeamOp(op_i_generic_breakpoint)); - g->orig_instr = (BeamInstr) BeamOp(op_call_nif); - } - code_ptr[5+1] = (BeamInstr) entry->funcs[i].fptr; + g->orig_instr = (BeamInstr) BeamOp(op_call_nif); + } + if ((entry->major > 2 || (entry->major == 2 && entry->minor >= 7)) + && (entry->options & ERL_NIF_DIRTY_NIF_OPTION) && f->flags) { +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + code_ptr[5+3] = (BeamInstr) f->fptr; + code_ptr[5+1] = (f->flags == ERL_NIF_DIRTY_JOB_IO_BOUND) ? + (BeamInstr) schedule_dirty_io_nif : + (BeamInstr) schedule_dirty_cpu_nif; +#endif + } + else + code_ptr[5+1] = (BeamInstr) f->fptr; code_ptr[5+2] = (BeamInstr) lib; + f = next_func(entry, &incr, f); } } else { diff --git a/erts/emulator/beam/erl_nif.h b/erts/emulator/beam/erl_nif.h index 5b93c2398e..9b2b90c82d 100644 --- a/erts/emulator/beam/erl_nif.h +++ b/erts/emulator/beam/erl_nif.h @@ -42,9 +42,13 @@ ** 2.5: R17 Maps API additions ** 2.6: R17 with maps ** R17 dirty schedulers +** 2.7: 17.3 add enif_schedule_nif +** remove enif_schedule_dirty_nif, enif_schedule_dirty_nif_finalizer, enif_dirty_nif_finalizer +** add ErlNifEntry options +** add ErlNifFunc flags */ #define ERL_NIF_MAJOR_VERSION 2 -#define ERL_NIF_MINOR_VERSION 6 +#define ERL_NIF_MINOR_VERSION 7 /* * The emulator will refuse to load a nif-lib with a major version @@ -125,8 +129,10 @@ typedef struct const char* name; unsigned arity; ERL_NIF_TERM (*fptr)(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]); + unsigned flags; }ErlNifFunc; + typedef struct enif_entry_t { int major; @@ -139,8 +145,11 @@ typedef struct enif_entry_t int (*upgrade)(ErlNifEnv*, void** priv_data, void** old_priv_data, ERL_NIF_TERM load_info); void (*unload) (ErlNifEnv*, void* priv_data); const char* vm_variant; + unsigned options; }ErlNifEntry; +/* Field bits for ErlNifEntry options */ +#define ERL_NIF_DIRTY_NIF_OPTION 1 typedef struct @@ -192,10 +201,18 @@ typedef enum typedef struct /* All fields all internal and may change */ { ERL_NIF_TERM map; - ERL_NIF_UINT t_limit; + ERL_NIF_UINT size; ERL_NIF_UINT idx; - ERL_NIF_TERM *ks; - ERL_NIF_TERM *vs; + union { + struct { + ERL_NIF_TERM *ks; + ERL_NIF_TERM *vs; + }flat; + struct { + struct ErtsDynamicWStack_* wstack; + ERL_NIF_TERM* kv; + }hash; + }u; void* __spare__[2]; /* for future additions to be ABI compatible (same struct size) */ } ErlNifMapIterator; @@ -243,6 +260,11 @@ extern TWinDynNifCallbacks WinDynNifCallbacks; # endif #endif +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +# define ERL_NIF_ENTRY_OPTIONS ERL_NIF_DIRTY_NIF_OPTION +#else +# define ERL_NIF_ENTRY_OPTIONS 0 +#endif #ifdef __cplusplus } @@ -268,7 +290,8 @@ ERL_NIF_INIT_DECL(NAME) \ sizeof(FUNCS) / sizeof(*FUNCS), \ FUNCS, \ LOAD, RELOAD, UPGRADE, UNLOAD, \ - ERL_NIF_VM_VARIANT \ + ERL_NIF_VM_VARIANT, \ + ERL_NIF_ENTRY_OPTIONS \ }; \ ERL_NIF_INIT_BODY; \ return &entry; \ diff --git a/erts/emulator/beam/erl_nif_api_funcs.h b/erts/emulator/beam/erl_nif_api_funcs.h index d7c554e60b..630cefae93 100644 --- a/erts/emulator/beam/erl_nif_api_funcs.h +++ b/erts/emulator/beam/erl_nif_api_funcs.h @@ -22,7 +22,7 @@ #endif /* -** WARNING: add new ERL_NIF_API_FUNC_DECL entries at the bottom of the list +** WARNING: Add new ERL_NIF_API_FUNC_DECL entries at the bottom of the list ** to keep compatibility on Windows!!! ** ** And don't forget to increase ERL_NIF_MINOR_VERSION in erl_nif.h @@ -141,14 +141,6 @@ ERL_NIF_API_FUNC_DECL(int,enif_is_number,(ErlNifEnv*, ERL_NIF_TERM term)); ERL_NIF_API_FUNC_DECL(void*,enif_dlopen,(const char* lib, void (*err_handler)(void*,const char*), void* err_arg)); ERL_NIF_API_FUNC_DECL(void*,enif_dlsym,(void* handle, const char* symbol, void (*err_handler)(void*,const char*), void* err_arg)); ERL_NIF_API_FUNC_DECL(int,enif_consume_timeslice,(ErlNifEnv*, int percent)); -#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT -ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_dirty_nif,(ErlNifEnv*,int,ERL_NIF_TERM (*)(ErlNifEnv*,int,const ERL_NIF_TERM[]),int,const ERL_NIF_TERM[])); -ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_dirty_nif_finalizer,(ErlNifEnv*,ERL_NIF_TERM,ERL_NIF_TERM (*)(ErlNifEnv*,ERL_NIF_TERM))); -ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_dirty_nif_finalizer,(ErlNifEnv*,ERL_NIF_TERM)); -ERL_NIF_API_FUNC_DECL(int,enif_is_on_dirty_scheduler,(ErlNifEnv*)); -ERL_NIF_API_FUNC_DECL(int,enif_have_dirty_schedulers,(void)); -#endif - ERL_NIF_API_FUNC_DECL(int, enif_is_map, (ErlNifEnv* env, ERL_NIF_TERM term)); ERL_NIF_API_FUNC_DECL(int, enif_get_map_size, (ErlNifEnv* env, ERL_NIF_TERM term, size_t *size)); ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM, enif_make_new_map, (ErlNifEnv* env)); @@ -163,12 +155,22 @@ ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_is_tail, (ErlNifEnv *env, ErlNifMap ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_next, (ErlNifEnv *env, ErlNifMapIterator *iter)); ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_prev, (ErlNifEnv *env, ErlNifMapIterator *iter)); ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_get_pair, (ErlNifEnv *env, ErlNifMapIterator *iter, ERL_NIF_TERM *key, ERL_NIF_TERM *value)); - +ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_nif,(ErlNifEnv*,const char*,int,ERL_NIF_TERM (*)(ErlNifEnv*,int,const ERL_NIF_TERM[]),int,const ERL_NIF_TERM[])); /* -** Add new entries here to keep compatibility on Windows!!! +** ADD NEW ENTRIES HERE (before this comment) !!! */ + + +/* + * Conditional EXPERIMENTAL stuff always last. + * Must be moved up and made unconditional to support binary backward + * compatibility on Windows. + */ +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +ERL_NIF_API_FUNC_DECL(int,enif_is_on_dirty_scheduler,(ErlNifEnv*)); #endif +#endif /* ERL_NIF_API_FUNC_DECL */ /* ** Please keep the ERL_NIF_API_FUNC_MACRO list below in the same order @@ -282,21 +284,12 @@ ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_get_pair, (ErlNifEnv *env, ErlNifMa # define enif_make_int64 ERL_NIF_API_FUNC_MACRO(enif_make_int64) # define enif_make_uint64 ERL_NIF_API_FUNC_MACRO(enif_make_uint64) #endif - # define enif_is_exception ERL_NIF_API_FUNC_MACRO(enif_is_exception) # define enif_make_reverse_list ERL_NIF_API_FUNC_MACRO(enif_make_reverse_list) # define enif_is_number ERL_NIF_API_FUNC_MACRO(enif_is_number) # define enif_dlopen ERL_NIF_API_FUNC_MACRO(enif_dlopen) # define enif_dlsym ERL_NIF_API_FUNC_MACRO(enif_dlsym) # define enif_consume_timeslice ERL_NIF_API_FUNC_MACRO(enif_consume_timeslice) -#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT -# define enif_schedule_dirty_nif ERL_NIF_API_FUNC_MACRO(enif_schedule_dirty_nif) -# define enif_schedule_dirty_nif_finalizer ERL_NIF_API_FUNC_MACRO(enif_schedule_dirty_nif_finalizer) -# define enif_dirty_nif_finalizer ERL_NIF_API_FUNC_MACRO(enif_dirty_nif_finalizer) -# define enif_is_on_dirty_scheduler ERL_NIF_API_FUNC_MACRO(enif_is_on_dirty_scheduler) -# define enif_have_dirty_schedulers ERL_NIF_API_FUNC_MACRO(enif_have_dirty_schedulers) -#endif - # define enif_is_map ERL_NIF_API_FUNC_MACRO(enif_is_map) # define enif_get_map_size ERL_NIF_API_FUNC_MACRO(enif_get_map_size) # define enif_make_new_map ERL_NIF_API_FUNC_MACRO(enif_make_new_map) @@ -311,11 +304,21 @@ ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_get_pair, (ErlNifEnv *env, ErlNifMa # define enif_map_iterator_next ERL_NIF_API_FUNC_MACRO(enif_map_iterator_next) # define enif_map_iterator_prev ERL_NIF_API_FUNC_MACRO(enif_map_iterator_prev) # define enif_map_iterator_get_pair ERL_NIF_API_FUNC_MACRO(enif_map_iterator_get_pair) +# define enif_schedule_nif ERL_NIF_API_FUNC_MACRO(enif_schedule_nif) /* -** Add new entries here +** ADD NEW ENTRIES HERE (before this comment) */ + +/* + * Conditional EXPERIMENTAL stuff always last + * Must be moved up and made unconditional to support binary backward + * compatibility on Windows. + */ +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +# define enif_is_on_dirty_scheduler ERL_NIF_API_FUNC_MACRO(enif_is_on_dirty_scheduler) #endif +#endif /* ERL_NIF_API_FUNC_MACRO */ #if defined(__GNUC__) && !(defined(__WIN32__) || defined(_WIN32) || defined(_WIN32_)) diff --git a/erts/emulator/beam/erl_port_task.c b/erts/emulator/beam/erl_port_task.c index 31d9a1e26e..2aa0a27197 100644 --- a/erts/emulator/beam/erl_port_task.c +++ b/erts/emulator/beam/erl_port_task.c @@ -32,6 +32,7 @@ #include "global.h" #include "erl_port_task.h" #include "dist.h" +#include "erl_check_io.h" #include "dtrace-wrapper.h" #include <stdarg.h> @@ -68,6 +69,13 @@ static void chk_task_queues(Port *pp, ErtsPortTask *execq, int processing_busy_q #define DTRACE_DRIVER(PROBE_NAME, PP) do {} while(0) #endif +#define ERTS_SMP_LC_VERIFY_RQ(RQ, PP) \ + do { \ + ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(runq)); \ + ERTS_SMP_LC_ASSERT((RQ) == ((ErtsRunQueue *) \ + erts_smp_atomic_read_nob(&(PP)->run_queue))); \ + } while (0) + erts_smp_atomic_t erts_port_task_outstanding_io_tasks; #define ERTS_PT_STATE_SCHEDULED 0 @@ -543,6 +551,16 @@ reset_handle(ErtsPortTask *ptp) } static ERTS_INLINE void +reset_executed_io_task_handle(ErtsPortTask *ptp) +{ + if (ptp->u.alive.handle) { + ASSERT(ptp == handle2task(ptp->u.alive.handle)); + erts_io_notify_port_task_executed(ptp->u.alive.handle); + reset_port_task_handle(ptp->u.alive.handle); + } +} + +static ERTS_INLINE void set_handle(ErtsPortTask *ptp, ErtsPortTaskHandle *pthp) { ptp->u.alive.handle = pthp; @@ -798,12 +816,13 @@ schedule_port_task_handle_list_free(ErtsPortTaskHandleList *pthlp) static ERTS_INLINE void abort_nosuspend_task(Port *pp, ErtsPortTaskType type, - ErtsPortTaskTypeData *tdp) + ErtsPortTaskTypeData *tdp, + int bpq_data) { ASSERT(type == ERTS_PORT_TASK_PROC_SIG); - if (!pp->sched.taskq.bpq) + if (!bpq_data) tdp->psig.callback(NULL, ERTS_PORT_SFLG_INVALID, ERTS_PROC2PORT_SIG_ABORT_NOSUSPEND, @@ -991,6 +1010,7 @@ static ERTS_INLINE int finalize_exec(Port *pp, ErtsPortTask **execq, int processing_busy_q) { erts_aint32_t act; + unsigned int prof_runnable_ports; if (!processing_busy_q) pp->sched.taskq.local.first = *execq; @@ -1007,6 +1027,10 @@ finalize_exec(Port *pp, ErtsPortTask **execq, int processing_busy_q) if (act & ERTS_PTS_FLG_CHK_UNSET_BUSY_PORT_Q) act = check_unset_busy_port_q(pp, act, pp->sched.taskq.bpq); + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&pp->sched); + while (1) { erts_aint32_t new, exp; @@ -1018,12 +1042,24 @@ finalize_exec(Port *pp, ErtsPortTask **execq, int processing_busy_q) act = erts_smp_atomic32_cmpxchg_relb(&pp->sched.flags, new, exp); - ASSERT(!(act & ERTS_PTS_FLG_IN_RUNQ)); + ERTS_LC_ASSERT(!(act & ERTS_PTS_FLG_IN_RUNQ)); + ERTS_LC_ASSERT(!(act & ERTS_PTS_FLG_EXEC_IMM)); if (exp == act) break; } + if (prof_runnable_ports | IS_TRACED_FL(pp, F_TRACE_SCHED_PORTS)) { + /* trace port scheduling, out */ + if (IS_TRACED_FL(pp, F_TRACE_SCHED_PORTS)) + trace_sched_ports(pp, am_out); + if (prof_runnable_ports) { + if (!(act & (ERTS_PTS_FLG_EXEC_IMM|ERTS_PTS_FLG_HAVE_TASKS))) + profile_runnable_port(pp, am_inactive); + erts_port_task_sched_unlock(&pp->sched); + } + } + return (act & ERTS_PTS_FLG_HAVE_TASKS) != 0; } @@ -1345,7 +1381,7 @@ erts_port_task_abort_nosuspend_tasks(Port *pp) #endif schedule_port_task_handle_list_free(pthlp); - abort_nosuspend_task(pp, type, &td); + abort_nosuspend_task(pp, type, &td, pp->sched.taskq.bpq != NULL); } } @@ -1369,11 +1405,9 @@ erts_port_task_schedule(Eterm id, Port *pp; ErtsPortTask *ptp = NULL; erts_aint32_t act, add_flags; + unsigned int prof_runnable_ports; - if (pthp && erts_port_task_is_scheduled(pthp)) { - ASSERT(0); - erts_port_task_abort(pthp); - } + ERTS_LC_ASSERT(!pthp || !erts_port_task_is_scheduled(pthp)); ASSERT(is_internal_port(id)); @@ -1457,6 +1491,10 @@ erts_port_task_schedule(Eterm id, if (ns_pthlp) add_flags |= ERTS_PTS_FLG_HAVE_NS_TASKS; + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&pp->sched); + while (1) { erts_aint32_t new, exp; @@ -1481,6 +1519,13 @@ erts_port_task_schedule(Eterm id, goto done; /* Died after our task insert... */ } + if (prof_runnable_ports) { + if (!(act & ERTS_PTS_FLG_EXEC_IMM)) + profile_runnable_port(pp, am_active); + erts_port_task_sched_unlock(&pp->sched); + prof_runnable_ports = 0; + } + /* Enqueue port on run-queue */ runq = erts_port_runq(pp); @@ -1489,8 +1534,10 @@ erts_port_task_schedule(Eterm id, #ifdef ERTS_SMP xrunq = erts_check_emigration_need(runq, ERTS_PORT_PRIO_LEVEL); + ERTS_SMP_LC_ASSERT(runq != xrunq); + ERTS_SMP_LC_VERIFY_RQ(runq, pp); if (xrunq) { - /* Port emigrated ... */ + /* Emigrate port ... */ erts_smp_atomic_set_nob(&pp->run_queue, (erts_aint_t) xrunq); erts_smp_runq_unlock(runq); runq = erts_port_runq(pp); @@ -1500,10 +1547,6 @@ erts_port_task_schedule(Eterm id, #endif enqueue_port(runq, pp); - - if (erts_system_profile_flags.runnable_ports) { - profile_runnable_port(pp, am_active); - } erts_smp_runq_unlock(runq); @@ -1511,6 +1554,9 @@ erts_port_task_schedule(Eterm id, done: + if (prof_runnable_ports) + erts_port_task_sched_unlock(&pp->sched); + #ifdef ERTS_SMP if (dhndl != ERTS_THR_PRGR_DHANDLE_MANAGED) erts_port_dec_refc(pp); @@ -1525,7 +1571,7 @@ abort_nosuspend: erts_port_dec_refc(pp); #endif - abort_nosuspend_task(pp, ptp->type, &ptp->u.alive.td); + abort_nosuspend_task(pp, ptp->type, &ptp->u.alive.td, 0); ASSERT(ns_pthlp); erts_free(ERTS_ALC_T_PT_HNDL_LIST, ns_pthlp); @@ -1609,6 +1655,8 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) goto done; } + ERTS_SMP_LC_VERIFY_RQ(runq, pp); + erts_smp_runq_unlock(runq); *curr_port_pp = pp; @@ -1659,8 +1707,6 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) goto aborted_port_task; } - reset_handle(ptp); - if (erts_system_monitor_long_schedule != 0) { start_time = erts_timestamp_millis(); } @@ -1671,6 +1717,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) switch (ptp->type) { case ERTS_PORT_TASK_TIMEOUT: + reset_handle(ptp); reds = ERTS_PORT_REDS_TIMEOUT; if (!(state & ERTS_PORT_SFLGS_DEAD)) { DTRACE_DRIVER(driver_timeout, pp); @@ -1685,6 +1732,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) for input and output */ (*pp->drv_ptr->ready_input)((ErlDrvData) pp->drv_data, ptp->u.alive.td.io.event); + reset_executed_io_task_handle(ptp); io_tasks_executed++; break; case ERTS_PORT_TASK_OUTPUT: @@ -1693,6 +1741,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) DTRACE_DRIVER(driver_ready_output, pp); (*pp->drv_ptr->ready_output)((ErlDrvData) pp->drv_data, ptp->u.alive.td.io.event); + reset_executed_io_task_handle(ptp); io_tasks_executed++; break; case ERTS_PORT_TASK_EVENT: @@ -1702,10 +1751,12 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) (*pp->drv_ptr->event)((ErlDrvData) pp->drv_data, ptp->u.alive.td.io.event, ptp->u.alive.td.io.event_data); + reset_executed_io_task_handle(ptp); io_tasks_executed++; break; case ERTS_PORT_TASK_PROC_SIG: { ErtsProc2PortSigData *sigdp = &ptp->u.alive.td.psig.data; + reset_handle(ptp); ASSERT((state & ERTS_PORT_SFLGS_DEAD) == 0); if (!pp->sched.taskq.bpq) reds = ptp->u.alive.td.psig.callback(pp, @@ -1723,6 +1774,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) break; } case ERTS_PORT_TASK_DIST_CMD: + reset_handle(ptp); reds = erts_dist_command(pp, CONTEXT_REDS - pp->reds); break; default: @@ -1765,10 +1817,6 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) erts_unblock_fpe(fpe_was_unmasked); - /* trace port scheduling, out */ - if (IS_TRACED_FL(pp, F_TRACE_SCHED_PORTS)) { - trace_sched_ports(pp, am_out); - } if (io_tasks_executed) { ASSERT(erts_smp_atomic_read_nob(&erts_port_task_outstanding_io_tasks) @@ -1791,11 +1839,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) erts_smp_runq_lock(runq); - if (!active) { - if (erts_system_profile_flags.runnable_ports) - profile_runnable_port(pp, am_inactive); - } - else { + if (active) { #ifdef ERTS_SMP ErtsRunQueue *xrunq; #endif @@ -1804,6 +1848,8 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) #ifdef ERTS_SMP xrunq = erts_check_emigration_need(runq, ERTS_PORT_PRIO_LEVEL); + ERTS_SMP_LC_ASSERT(runq != xrunq); + ERTS_SMP_LC_VERIFY_RQ(runq, pp); if (!xrunq) { #endif enqueue_port(runq, pp); @@ -1811,7 +1857,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) #ifdef ERTS_SMP } else { - /* Port emigrated ... */ + /* Emigrate port... */ erts_smp_atomic_set_nob(&pp->run_queue, (erts_aint_t) xrunq); erts_smp_runq_unlock(runq); diff --git a/erts/emulator/beam/erl_port_task.h b/erts/emulator/beam/erl_port_task.h index 1d30465ec9..406cd3c492 100644 --- a/erts/emulator/beam/erl_port_task.h +++ b/erts/emulator/beam/erl_port_task.h @@ -78,6 +78,7 @@ extern erts_smp_atomic_t erts_port_task_outstanding_io_tasks; #define ERTS_PTS_FLG_PARALLELISM (((erts_aint32_t) 1) << 9) #define ERTS_PTS_FLG_FORCE_SCHED (((erts_aint32_t) 1) << 10) #define ERTS_PTS_FLG_EXITING (((erts_aint32_t) 1) << 11) +#define ERTS_PTS_FLG_EXEC_IMM (((erts_aint32_t) 1) << 12) #define ERTS_PTS_FLGS_BUSY \ (ERTS_PTS_FLG_BUSY_PORT | ERTS_PTS_FLG_BUSY_PORT_Q) @@ -87,6 +88,7 @@ extern erts_smp_atomic_t erts_port_task_outstanding_io_tasks; | ERTS_PTS_FLG_HAVE_BUSY_TASKS \ | ERTS_PTS_FLG_HAVE_TASKS \ | ERTS_PTS_FLG_EXEC \ + | ERTS_PTS_FLG_EXEC_IMM \ | ERTS_PTS_FLG_FORCE_SCHED \ | ERTS_PTS_FLG_EXITING) @@ -154,7 +156,7 @@ erts_port_task_handle_init(ErtsPortTaskHandle *pthp) ERTS_GLB_INLINE int erts_port_task_is_scheduled(ErtsPortTaskHandle *pthp) { - return ((void *) erts_smp_atomic_read_nob(pthp)) != NULL; + return ((void *) erts_smp_atomic_read_acqb(pthp)) != NULL; } ERTS_GLB_INLINE void erts_port_task_pre_init_sched(ErtsPortTaskSched *ptsp, diff --git a/erts/emulator/beam/erl_printf_term.c b/erts/emulator/beam/erl_printf_term.c index d18760dc43..e0dfbd31b8 100644 --- a/erts/emulator/beam/erl_printf_term.c +++ b/erts/emulator/beam/erl_printf_term.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2005-2013. All Rights Reserved. + * Copyright Ericsson AB 2005-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -25,6 +25,7 @@ #include "sys.h" #include "big.h" #include "erl_map.h" +#include "erl_binary.h" #define PRINT_CHAR(CNT, FN, ARG, C) \ do { \ @@ -138,6 +139,25 @@ is_printable_string(Eterm list, Eterm* base) return 0; } +static int is_printable_ascii(byte* bytep, Uint bytesize, Uint bitoffs) +{ + if (!bitoffs) { + while (bytesize--) { + if (*bytep < ' ' || *bytep >= 127) + return 0; + bytep++; + } + } else { + while (bytesize--) { + byte octet = (bytep[0] << bitoffs) | (bytep[1] >> (8-bitoffs)); + if (octet < ' ' || octet >= 127) + return 0; + bytep++; + } + } + return 1; +} + /* print a atom doing what quoting is necessary */ static int print_atom_name(fmtfn_t fn, void* arg, Eterm atom, long *dcount) { @@ -227,6 +247,17 @@ static int print_atom_name(fmtfn_t fn, void* arg, Eterm atom, long *dcount) #define PRT_PATCH_FUN_SIZE ((Eterm) 7) #define PRT_LAST_ARRAY_ELEMENT ((Eterm) 8) /* Note! Must be last... */ +#if 0 +static char *format_binary(Uint16 x, char *b) { + int z; + b[16] = '\0'; + for (z = 0; z < 16; z++) { + b[15-z] = ((x>>z) & 0x1) ? '1' : '0'; + } + return b; +} +#endif + static int print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, Eterm* obj_base) /* ignored if !HALFWORD_HEAP */ @@ -283,13 +314,9 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, tl = CDR(cons); if (is_not_nil(tl)) { if (is_list(tl)) { - WSTACK_PUSH(s, tl); - WSTACK_PUSH(s, PRT_ONE_CONS); - WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH3(s, tl, PRT_ONE_CONS, PRT_COMMA); } else { - WSTACK_PUSH(s, tl); - WSTACK_PUSH(s, PRT_TERM); - WSTACK_PUSH(s, PRT_BAR); + WSTACK_PUSH3(s, tl, PRT_TERM, PRT_BAR); } } } @@ -299,9 +326,7 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, break; default: /* PRT_LAST_ARRAY_ELEMENT+1 and upwards */ obj = *popped.ptr; - WSTACK_PUSH(s, (UWord) (popped.ptr + 1)); - WSTACK_PUSH(s, val-1); - WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH3(s, (UWord) (popped.ptr + 1), val-1, PRT_COMMA); break; } break; @@ -431,8 +456,7 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, WSTACK_PUSH(s,PRT_CLOSE_TUPLE); ++nobj; if (i > 0) { - WSTACK_PUSH(s, (UWord) nobj); - WSTACK_PUSH(s, PRT_LAST_ARRAY_ELEMENT+i-1); + WSTACK_PUSH2(s, (UWord) nobj, PRT_LAST_ARRAY_ELEMENT+i-1); } break; case FLOAT_DEF: { @@ -446,13 +470,65 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, PRINT_STRING(res, fn, arg, "#MatchState"); } else { - ProcBin* pb = (ProcBin *) binary_val(wobj); - if (pb->size == 1) - PRINT_STRING(res, fn, arg, "<<1 byte>>"); - else { + byte* bytep; + Uint bytesize = binary_size_rel(obj,obj_base); + Uint bitoffs; + Uint bitsize; + byte octet; + ERTS_GET_BINARY_BYTES_REL(obj, bytep, bitoffs, bitsize, obj_base); + + if (bitsize || !bytesize + || !is_printable_ascii(bytep, bytesize, bitoffs)) { + int is_first = 1; PRINT_STRING(res, fn, arg, "<<"); - PRINT_UWORD(res, fn, arg, 'u', 0, 1, (ErlPfUWord) pb->size); - PRINT_STRING(res, fn, arg, " bytes>>"); + while (bytesize) { + if (is_first) + is_first = 0; + else + PRINT_CHAR(res, fn, arg, ','); + if (bitoffs) + octet = (bytep[0] << bitoffs) | (bytep[1] >> (8-bitoffs)); + else + octet = bytep[0]; + PRINT_UWORD(res, fn, arg, 'u', 0, 1, octet); + ++bytep; + --bytesize; + } + if (bitsize) { + Uint bits = bitoffs + bitsize; + octet = bytep[0]; + if (bits < 8) + octet >>= 8 - bits; + else if (bits > 8) { + bits -= 8; /* bits in last byte */ + octet <<= bits; + octet |= bytep[1] >> (8 - bits); + } + octet &= (1 << bitsize) - 1; + if (is_first) + is_first = 0; + else + PRINT_CHAR(res, fn, arg, ','); + PRINT_UWORD(res, fn, arg, 'u', 0, 1, octet); + PRINT_CHAR(res, fn, arg, ':'); + PRINT_UWORD(res, fn, arg, 'u', 0, 1, bitsize); + } + PRINT_STRING(res, fn, arg, ">>"); + } + else { + PRINT_STRING(res, fn, arg, "<<\""); + while (bytesize) { + if (bitoffs) + octet = (bytep[0] << bitoffs) | (bytep[1] >> (8-bitoffs)); + else + octet = bytep[0]; + if (octet == '"') + PRINT_CHAR(res, fn, arg, '\\'); + PRINT_CHAR(res, fn, arg, octet); + ++bytep; + --bytesize; + } + PRINT_STRING(res, fn, arg, "\">>"); } } break; @@ -489,37 +565,74 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, } break; case MAP_DEF: - { - Uint n; - Eterm *ks, *vs; - map_t *mp = (map_t *)map_val(wobj); - n = map_get_size(mp); - ks = map_get_keys(mp); - vs = map_get_values(mp); - - PRINT_CHAR(res, fn, arg, '#'); - PRINT_CHAR(res, fn, arg, '{'); - WSTACK_PUSH(s, PRT_CLOSE_TUPLE); - if (n > 0) { - n--; - WSTACK_PUSH(s, vs[n]); - WSTACK_PUSH(s, PRT_TERM); - WSTACK_PUSH(s, PRT_ASSOC); - WSTACK_PUSH(s, ks[n]); - WSTACK_PUSH(s, PRT_TERM); - - while (n--) { - WSTACK_PUSH(s, PRT_COMMA); - WSTACK_PUSH(s, vs[n]); - WSTACK_PUSH(s, PRT_TERM); - WSTACK_PUSH(s, PRT_ASSOC); - WSTACK_PUSH(s, ks[n]); - WSTACK_PUSH(s, PRT_TERM); - } - } - } - break; - default: + if (is_flatmap(wobj)) { + Uint n; + Eterm *ks, *vs; + flatmap_t *mp = (flatmap_t *)flatmap_val(wobj); + n = flatmap_get_size(mp); + ks = flatmap_get_keys(mp); + vs = flatmap_get_values(mp); + + PRINT_CHAR(res, fn, arg, '#'); + PRINT_CHAR(res, fn, arg, '{'); + WSTACK_PUSH(s, PRT_CLOSE_TUPLE); + if (n > 0) { + n--; + WSTACK_PUSH5(s, vs[n], PRT_TERM, PRT_ASSOC, ks[n], PRT_TERM); + while (n--) { + WSTACK_PUSH6(s, PRT_COMMA, vs[n], PRT_TERM, PRT_ASSOC, + ks[n], PRT_TERM); + } + } + } else { + Uint n, mapval; + Eterm *head; + head = hashmap_val(wobj); + mapval = MAP_HEADER_VAL(*head); + switch (MAP_HEADER_TYPE(*head)) { + case MAP_HEADER_TAG_HAMT_HEAD_ARRAY: + case MAP_HEADER_TAG_HAMT_HEAD_BITMAP: + PRINT_STRING(res, fn, arg, "#<"); + PRINT_UWORD(res, fn, arg, 'x', 0, 1, mapval); + PRINT_STRING(res, fn, arg, ">{"); + WSTACK_PUSH(s,PRT_CLOSE_TUPLE); + n = hashmap_bitcount(mapval); + ASSERT(n < 17); + head += 2; + if (n > 0) { + n--; + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + while (n--) { + WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + } + } + break; + case MAP_HEADER_TAG_HAMT_NODE_BITMAP: + n = hashmap_bitcount(mapval); + head++; + PRINT_CHAR(res, fn, arg, '<'); + PRINT_UWORD(res, fn, arg, 'x', 0, 1, mapval); + PRINT_STRING(res, fn, arg, ">{"); + WSTACK_PUSH(s,PRT_CLOSE_TUPLE); + ASSERT(n < 17); + if (n > 0) { + n--; + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + while (n--) { + WSTACK_PUSH(s, PRT_COMMA); + WSTACK_PUSH(s, head[n]); + WSTACK_PUSH(s, PRT_TERM); + } + } + break; + } + } + break; + default: PRINT_STRING(res, fn, arg, "<unknown:"); PRINT_POINTER(res, fn, arg, wobj); PRINT_CHAR(res, fn, arg, '>'); @@ -528,17 +641,17 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, } L_done: - DESTROY_WSTACK(s); return res; } + int erts_printf_term(fmtfn_t fn, void* arg, ErlPfEterm term, long precision, ErlPfEterm* term_base) { int res; - ASSERT(sizeof(ErlPfEterm) == sizeof(Eterm)); + ERTS_CT_ASSERT(sizeof(ErlPfEterm) == sizeof(Eterm)); res = print_term(fn, arg, (Eterm)term, &precision, (Eterm*)term_base); if (res < 0) diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index b4b97d7df1..f74a2ee54c 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -43,6 +43,7 @@ #include "erl_async.h" #include "dtrace-wrapper.h" #include "erl_ptab.h" +#include "erl_bif_unique.h" #define ERTS_DELAYED_WAKEUP_INFINITY (~(Uint64) 0) @@ -148,6 +149,12 @@ extern BeamInstr beam_apply[]; extern BeamInstr beam_exit[]; extern BeamInstr beam_continue_exit[]; +#ifdef __OSE__ +/* Eager check I/O not supported on OSE yet. */ +int erts_eager_check_io = 0; +#else +int erts_eager_check_io = 1; +#endif int erts_sched_compact_load; int erts_sched_balance_util = 0; Uint erts_no_schedulers; @@ -451,8 +458,7 @@ do { \ static void exec_misc_ops(ErtsRunQueue *); static void print_function_from_pc(int to, void *to_arg, BeamInstr* x); -static int stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, - int yreg); +static int stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg); static void aux_work_timeout(void *unused); static void aux_work_timeout_early_init(int no_schedulers); @@ -590,12 +596,10 @@ erts_pre_init_process(void) erts_psd_required_locks[ERTS_PSD_DELAYED_GC_TASK_QS].set_locks = ERTS_PSD_DELAYED_GC_TASK_QS_SET_LOCKS; -#ifdef ERTS_DIRTY_SCHEDULERS - erts_psd_required_locks[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT].get_locks - = ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_GET_LOCKS; - erts_psd_required_locks[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT].set_locks - = ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_SET_LOCKS; -#endif + erts_psd_required_locks[ERTS_PSD_NIF_TRAP_EXPORT].get_locks + = ERTS_PSD_NIF_TRAP_EXPORT_GET_LOCKS; + erts_psd_required_locks[ERTS_PSD_NIF_TRAP_EXPORT].set_locks + = ERTS_PSD_NIF_TRAP_EXPORT_SET_LOCKS; /* Check that we have locks for all entries */ for (ix = 0; ix < ERTS_PSD_SIZE; ix++) { @@ -698,8 +702,8 @@ init_sched_wall_time(ErtsSchedWallTime *swtp) static ERTS_INLINE Uint64 sched_wall_time_ts(void) { -#ifdef HAVE_GETHRTIME - return (Uint64) sys_gethrtime(); +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + return (Uint64) erts_os_monotonic_time(); #else Uint64 res; SysTimeval tv; @@ -712,72 +716,24 @@ sched_wall_time_ts(void) #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT -#ifdef ARCH_64 - -static ERTS_INLINE Uint64 -aschedtime_read(ErtsAtomicSchedTime *var) -{ - return (Uint64) erts_atomic_read_nob((erts_atomic_t *) var); -} - -static ERTS_INLINE void -aschedtime_set(ErtsAtomicSchedTime *var, Uint64 val) -{ - erts_atomic_set_nob((erts_atomic_t *) var, (erts_aint_t) val); -} - -static ERTS_INLINE void -aschedtime_init(ErtsAtomicSchedTime *var) -{ - erts_atomic_init_nob((erts_atomic_t *) var, (erts_aint_t) 0); -} - -#elif defined(ARCH_32) - static ERTS_INLINE Uint64 aschedtime_read(ErtsAtomicSchedTime *var) { - erts_dw_aint_t dw; - erts_dw_atomic_read_nob((erts_dw_atomic_t *) var, &dw); -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw.dw_sint; -#else - { - Uint64 res; - res = (Uint64) ((Uint32) dw.sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw.sint[ERTS_DW_AINT_LOW_WORD]); - return res; - } -#endif + return (Uint64) erts_atomic64_read_nob((erts_atomic64_t *) var); } static ERTS_INLINE void aschedtime_set(ErtsAtomicSchedTime *var, Uint64 val) { - erts_dw_aint_t dw; -#ifdef ETHR_SU_DW_NAINT_T__ - dw.dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw.sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) (val & 0xffffffff); - dw.sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif - erts_dw_atomic_set_nob((erts_dw_atomic_t *) var, &dw); + erts_atomic64_set_nob((erts_atomic64_t *) var, (erts_aint64_t) val); } static ERTS_INLINE void aschedtime_init(ErtsAtomicSchedTime *var) { - erts_dw_aint_t dw; - dw.sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) 0; - dw.sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) 0; - erts_dw_atomic_init_nob((erts_dw_atomic_t *) var, &dw); + erts_atomic64_init_nob((erts_atomic64_t *) var, (erts_aint64_t) 0); } -#else -# error :-/ -#endif - #define ERTS_GET_AVG_MAX_UNLOCKED_TRY 50 #define ERTS_SCHED_AVG_UTIL_WRITE_MARKER (~((Uint64) 0)) @@ -2211,6 +2167,9 @@ aux_work_timeout_early_init(int no_schedulers) p = (UWord) malloc((sizeof(ErtsAuxWorkTmo) + sizeof(erts_atomic32_t)*(no_schedulers+1)) + ERTS_CACHE_LINE_SIZE-1); + if (!p) { + ERTS_INTERNAL_ERROR("malloc failed to allocate memory!"); + } if (p & ERTS_CACHE_LINE_MASK) p = (p & ~ERTS_CACHE_LINE_MASK) + ERTS_CACHE_LINE_SIZE; ASSERT((p & ERTS_CACHE_LINE_MASK) == 0); @@ -2227,7 +2186,7 @@ aux_work_timeout_late_init(void) { aux_work_tmo->initialized = 1; if (erts_atomic32_read_nob(&aux_work_tmo->refc)) { - aux_work_tmo->timer.data.active = 0; + erts_init_timer(&aux_work_tmo->timer.data); erts_set_timer(&aux_work_tmo->timer.data, aux_work_timeout, NULL, @@ -2260,7 +2219,6 @@ aux_work_timeout(void *unused) if (refc != 1 || 1 != erts_atomic32_cmpxchg_relb(&aux_work_tmo->refc, 0, 1)) { /* Setup next timeout... */ - aux_work_tmo->timer.data.active = 0; erts_set_timer(&aux_work_tmo->timer.data, aux_work_timeout, NULL, @@ -2279,7 +2237,7 @@ setup_aux_work_timer(void) else #endif { - aux_work_tmo->timer.data.active = 0; + erts_init_timer(&aux_work_tmo->timer.data); erts_set_timer(&aux_work_tmo->timer.data, aux_work_timeout, NULL, @@ -2359,7 +2317,6 @@ erts_active_schedulers(void) ERTS_ATOMIC_FOREACH_RUNQ(rq, as -= abs(rq->waiting)); - ASSERT(as >= 0); return as; } @@ -2380,29 +2337,47 @@ try_set_sys_scheduling(void) #endif static ERTS_INLINE int -prepare_for_sys_schedule(ErtsSchedulerData *esdp) +prepare_for_sys_schedule(ErtsSchedulerData *esdp, int non_blocking) { + if (non_blocking && erts_eager_check_io) { #ifdef ERTS_SMP - while (!erts_port_task_have_outstanding_io_tasks() - && try_set_sys_scheduling()) { #ifdef ERTS_SCHED_ONLY_POLL_SCHED_1 - if (esdp->no != 1) { - /* If we are not scheduler 1 and ERTS_SCHED_ONLY_POLL_SCHED_1 is used - then we make sure to wake scheduler 1 */ - ErtsRunQueue *rq = ERTS_RUNQ_IX(0); - clear_sys_scheduling(); - wake_scheduler(rq); - return 0; - } + if (esdp->no != 1) { + /* If we are not scheduler 1 and ERTS_SCHED_ONLY_POLL_SCHED_1 is used + then we make sure to wake scheduler 1 */ + ErtsRunQueue *rq = ERTS_RUNQ_IX(0); + wake_scheduler(rq); + return 0; + } #endif - if (!erts_port_task_have_outstanding_io_tasks()) + return try_set_sys_scheduling(); +#else return 1; - clear_sys_scheduling(); +#endif } - return 0; + else { +#ifdef ERTS_SMP + while (!erts_port_task_have_outstanding_io_tasks() + && try_set_sys_scheduling()) { +#ifdef ERTS_SCHED_ONLY_POLL_SCHED_1 + if (esdp->no != 1) { + /* If we are not scheduler 1 and ERTS_SCHED_ONLY_POLL_SCHED_1 is used + then we make sure to wake scheduler 1 */ + ErtsRunQueue *rq = ERTS_RUNQ_IX(0); + clear_sys_scheduling(); + wake_scheduler(rq); + return 0; + } +#endif + if (!erts_port_task_have_outstanding_io_tasks()) + return 1; + clear_sys_scheduling(); + } + return 0; #else - return !erts_port_task_have_outstanding_io_tasks(); + return !erts_port_task_have_outstanding_io_tasks(); #endif + } } #ifdef ERTS_SMP @@ -2663,6 +2638,13 @@ thr_prgr_fin_wait(void *vssi) static void init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp, char *dawwp); +void +erts_interupt_aux_thread_timed(ErtsMonotonicTime timeout_time) +{ + /* TODO only poke when needed (based on timeout_time) */ + erts_sched_poke(ERTS_SCHED_SLEEP_INFO_IX(-1)); +} + static void * aux_thread(void *unused) { @@ -2671,6 +2653,18 @@ aux_thread(void *unused) erts_aint32_t aux_work; ErtsThrPrgrCallbacks callbacks; int thr_prgr_active = 1; + ErtsTimerWheel *timer_wheel = erts_default_timer_wheel; + ErtsNextTimeoutRef nxt_tmo_ref = erts_get_next_timeout_reference(timer_wheel); + + if (!timer_wheel) + ERTS_INTERNAL_ERROR("Missing aux timer wheel"); + +#ifdef ERTS_ENABLE_LOCK_CHECK + { + char buf[] = "aux_thread"; + erts_lc_set_thread_name(buf); + } +#endif ssi->event = erts_tse_fetch(); @@ -2687,6 +2681,7 @@ aux_thread(void *unused) sched_prep_spin_wait(ssi); while (1) { + ErtsMonotonicTime current_time; erts_aint32_t flgs; aux_work = erts_atomic32_read_acqb(&ssi->aux_work); @@ -2698,28 +2693,56 @@ aux_thread(void *unused) erts_thr_progress_leader_update(NULL); } - if (!aux_work) { - if (thr_prgr_active) - erts_thr_progress_active(NULL, thr_prgr_active = 0); - erts_thr_progress_prepare_wait(NULL); + if (aux_work) { + current_time = erts_get_monotonic_time(); + if (current_time >= erts_next_timeout_time(nxt_tmo_ref)) { + if (!thr_prgr_active) + erts_thr_progress_active(NULL, thr_prgr_active = 1); + erts_bump_timers(timer_wheel, current_time); + } + } + else { + ErtsMonotonicTime timeout_time; + timeout_time = erts_check_next_timeout_time(timer_wheel, + ERTS_SEC_TO_MONOTONIC(10*60)); + current_time = erts_get_monotonic_time(); + if (current_time >= timeout_time) { + if (!thr_prgr_active) + erts_thr_progress_active(NULL, thr_prgr_active = 1); + } + else { + if (thr_prgr_active) + erts_thr_progress_active(NULL, thr_prgr_active = 0); + erts_thr_progress_prepare_wait(NULL); - ERTS_SCHED_FAIR_YIELD(); + ERTS_SCHED_FAIR_YIELD(); - flgs = sched_spin_wait(ssi, 0); + flgs = sched_spin_wait(ssi, 0); - if (flgs & ERTS_SSI_FLG_SLEEPING) { - ASSERT(flgs & ERTS_SSI_FLG_WAITING); - flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING); if (flgs & ERTS_SSI_FLG_SLEEPING) { - int res; - ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING); ASSERT(flgs & ERTS_SSI_FLG_WAITING); - do { - res = erts_tse_wait(ssi->event); - } while (res == EINTR); + flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING); + if (flgs & ERTS_SSI_FLG_SLEEPING) { + int res; + ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING); + ASSERT(flgs & ERTS_SSI_FLG_WAITING); + current_time = erts_get_monotonic_time(); + do { + Sint64 timeout; + if (current_time >= timeout_time) + break; + timeout = ERTS_MONOTONIC_TO_NSEC(timeout_time + - current_time + - 1) + 1; + res = erts_tse_twait(ssi->event, timeout); + current_time = erts_get_monotonic_time(); + } while (res == EINTR); + } } + erts_thr_progress_finalize_wait(NULL); } - erts_thr_progress_finalize_wait(NULL); + if (current_time >= timeout_time) + erts_bump_timers(timer_wheel, current_time); } flgs = sched_prep_spin_wait(ssi); @@ -2772,7 +2795,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * be waiting in erl_sys_schedule() */ - if (ERTS_SCHEDULER_IS_DIRTY(esdp) || !prepare_for_sys_schedule(esdp)) { + if (ERTS_SCHEDULER_IS_DIRTY(esdp) || !prepare_for_sys_schedule(esdp, 0)) { sched_waiting(esdp->no, rq); @@ -2786,6 +2809,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) sched_wall_time_change(esdp, thr_prgr_active); while (1) { + ErtsMonotonicTime current_time; aux_work = erts_atomic32_read_acqb(&ssi->aux_work); if (aux_work) { @@ -2799,34 +2823,65 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) erts_thr_progress_leader_update(esdp); } - if (aux_work) + if (aux_work) { flgs = erts_smp_atomic32_read_acqb(&ssi->flags); + current_time = erts_get_monotonic_time(); + if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) { + if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && !thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 1); + sched_wall_time_change(esdp, 1); + } + erts_bump_timers(esdp->timer_wheel, current_time); + } + } else { - if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { - if (thr_prgr_active) { - erts_thr_progress_active(esdp, thr_prgr_active = 0); - sched_wall_time_change(esdp, 0); + ErtsMonotonicTime timeout_time; + timeout_time = erts_check_next_timeout_time(esdp->timer_wheel, + ERTS_SEC_TO_MONOTONIC(10*60)); + current_time = erts_get_monotonic_time(); + if (current_time >= timeout_time) { + if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && !thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 1); + sched_wall_time_change(esdp, 1); } - erts_thr_progress_prepare_wait(esdp); } + else { + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { + if (thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 0); + sched_wall_time_change(esdp, 0); + } + erts_thr_progress_prepare_wait(esdp); + } - ERTS_SCHED_FAIR_YIELD(); + ERTS_SCHED_FAIR_YIELD(); - flgs = sched_spin_wait(ssi, spincount); - if (flgs & ERTS_SSI_FLG_SLEEPING) { - ASSERT(flgs & ERTS_SSI_FLG_WAITING); - flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING); + flgs = sched_spin_wait(ssi, spincount); if (flgs & ERTS_SSI_FLG_SLEEPING) { - int res; - ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING); ASSERT(flgs & ERTS_SSI_FLG_WAITING); - do { - res = erts_tse_wait(ssi->event); - } while (res == EINTR); + flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING); + if (flgs & ERTS_SSI_FLG_SLEEPING) { + int res; + ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING); + ASSERT(flgs & ERTS_SSI_FLG_WAITING); + current_time = erts_get_monotonic_time(); + do { + Sint64 timeout; + if (current_time >= timeout_time) + break; + timeout = ERTS_MONOTONIC_TO_NSEC(timeout_time + - current_time + - 1) + 1; + res = erts_tse_twait(ssi->event, timeout); + current_time = erts_get_monotonic_time(); + } while (res == EINTR); + } } + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) + erts_thr_progress_finalize_wait(esdp); } - if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) - erts_thr_progress_finalize_wait(esdp); + if (current_time >= timeout_time) + erts_bump_timers(esdp->timer_wheel, current_time); } if (!(flgs & ERTS_SSI_FLG_WAITING)) { @@ -2859,7 +2914,6 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) else #endif { - erts_aint_t dt; erts_smp_atomic32_set_relb(&function_calls, 0); *fcalls = 0; @@ -2884,6 +2938,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) goto sys_aux_work; while (spincount-- > 0) { + ErtsMonotonicTime current_time; sys_poll_aux_work: @@ -2893,8 +2948,9 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) ASSERT(!erts_port_task_have_outstanding_io_tasks()); erl_sys_schedule(1); /* Might give us something to do */ - dt = erts_do_time_read_and_reset(); - if (dt) erts_bump_timer(dt); + current_time = erts_get_monotonic_time(); + if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) + erts_bump_timers(esdp->timer_wheel, current_time); sys_aux_work: #ifndef ERTS_SMP @@ -2936,7 +2992,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * Got to check that we still got I/O tasks; otherwise * we have to continue checking for I/O... */ - if (!prepare_for_sys_schedule(esdp)) { + if (!prepare_for_sys_schedule(esdp, 0)) { spincount *= ERTS_SCHED_TSE_SLEEP_SPINCOUNT_FACT; goto tse_wait; } @@ -2958,7 +3014,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * Got to check that we still got I/O tasks; otherwise * we have to wait in erl_sys_schedule() after all... */ - if (!prepare_for_sys_schedule(esdp)) { + if (!prepare_for_sys_schedule(esdp, 0)) { /* * Not allowed to wait in erl_sys_schedule; * do tse wait instead... @@ -3009,8 +3065,11 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) erl_sys_schedule(0); - dt = erts_do_time_read_and_reset(); - if (dt) erts_bump_timer(dt); + { + ErtsMonotonicTime current_time = erts_get_monotonic_time(); + if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) + erts_bump_timers(esdp->timer_wheel, current_time); + } #ifndef ERTS_SMP if (rq->len == 0 && !rq->misc.start) @@ -3212,11 +3271,11 @@ chk_wake_sched(ErtsRunQueue *crq, int ix, int activate) return 0; wrq = ERTS_RUNQ_IX(ix); flags = ERTS_RUNQ_FLGS_GET(wrq); + if (activate && !(flags & ERTS_RUNQ_FLG_SUSPENDED)) { + if (try_inc_no_active_runqs(ix+1)) + (void) ERTS_RUNQ_FLGS_UNSET(wrq, ERTS_RUNQ_FLG_INACTIVE); + } if (!(flags & (ERTS_RUNQ_FLG_SUSPENDED|ERTS_RUNQ_FLG_NONEMPTY))) { - if (activate) { - if (try_inc_no_active_runqs(ix+1)) - (void) ERTS_RUNQ_FLGS_UNSET(wrq, ERTS_RUNQ_FLG_INACTIVE); - } wake_scheduler(wrq); return 1; } @@ -3748,17 +3807,25 @@ evacuate_run_queue(ErtsRunQueue *rq, } #ifdef ERTS_DIRTY_SCHEDULERS else if (state & ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q) { - erts_aint32_t old; - old = erts_smp_atomic32_read_band_nob(&proc->state, - ~(ERTS_PSFLG_DIRTY_CPU_PROC - | ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q)); +#ifdef DEBUG + erts_aint32_t old = +#else + (void) +#endif + erts_smp_atomic32_read_band_nob(&proc->state, + ~(ERTS_PSFLG_DIRTY_CPU_PROC + | ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q)); /* assert that no other dirty flags are set */ ASSERT(!(old & (ERTS_PSFLG_DIRTY_IO_PROC|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q))); } else if (state & ERTS_PSFLG_DIRTY_IO_PROC_IN_Q) { - erts_aint32_t old; - old = erts_smp_atomic32_read_band_nob(&proc->state, - ~(ERTS_PSFLG_DIRTY_IO_PROC - | ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); +#ifdef DEBUG + erts_aint32_t old = +#else + (void) +#endif + erts_smp_atomic32_read_band_nob(&proc->state, + ~(ERTS_PSFLG_DIRTY_IO_PROC + | ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); /* assert that no other dirty flags are set */ ASSERT(!(old & (ERTS_PSFLG_DIRTY_CPU_PROC|ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q))); } @@ -5272,6 +5339,10 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num, #else esdp->no = (Uint) num; #endif + + esdp->timer_wheel = erts_default_timer_wheel; + esdp->next_tmo_ref = erts_get_next_timeout_reference(esdp->timer_wheel); + esdp->ssi = ssi; esdp->current_process = NULL; esdp->current_port = NULL; @@ -5284,6 +5355,9 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num, esdp->run_queue = runq; esdp->run_queue->scheduler = esdp; + esdp->thr_id = (Uint32) num; + erts_sched_bif_unique_init(esdp); + if (daww_ptr) { init_aux_work_data(&esdp->aux_work_data, esdp, *daww_ptr); #ifdef ERTS_SMP @@ -5844,6 +5918,13 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p, Proces int check_emigration_need; #endif +#ifdef ERTS_SMP + if ((p->static_flags & ERTS_STC_FLG_PREFER_SCHED) + && p->preferred_run_queue != RUNQ_READ_RQ(&p->run_queue)) { + RUNQ_SET_RQ(&p->run_queue, p->preferred_run_queue); + } +#endif + a = state; while (1) { @@ -5867,6 +5948,9 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p, Proces case ERTS_ENQUEUE_NOT: if (erts_system_profile_flags.runnable_procs) { + /* Status lock prevents out of order "runnable proc" trace msgs */ + ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p)); + if (!(a & ERTS_PSFLG_ACTIVE_SYS) && (!(a & ERTS_PSFLG_ACTIVE) || (a & ERTS_PSFLG_SUSPENDED))) { @@ -5879,6 +5963,7 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p, Proces free_proxy_proc(proxy); erts_smp_runq_lock(c_rq); + return 0; #ifdef ERTS_DIRTY_SCHEDULERS @@ -5980,7 +6065,8 @@ change_proc_schedule_state(Process *p, erts_aint32_t clear_state_flags, erts_aint32_t set_state_flags, erts_aint32_t *statep, - erts_aint32_t *enq_prio_p) + erts_aint32_t *enq_prio_p, + ErtsProcLocks locks) { /* * NOTE: ERTS_PSFLG_RUNNING, ERTS_PSFLG_RUNNING_SYS and @@ -5989,6 +6075,11 @@ change_proc_schedule_state(Process *p, */ erts_aint32_t a = *statep, n; int enqueue; /* < 0 -> use proxy */ + unsigned int prof_runnable_procs = erts_system_profile_flags.runnable_procs; + unsigned int lock_status = (prof_runnable_procs + && !(locks & ERTS_PROC_LOCK_STATUS)); + + ERTS_SMP_LC_ASSERT(locks == erts_proc_lc_my_proc_locks(p)); ASSERT(!(a & ERTS_PSFLG_PROXY)); ASSERT((clear_state_flags & (ERTS_PSFLG_RUNNING @@ -5998,6 +6089,9 @@ change_proc_schedule_state(Process *p, | ERTS_PSFLG_RUNNING_SYS | ERTS_PSFLG_ACTIVE_SYS)) == 0); + if (lock_status) + erts_smp_proc_lock(p, ERTS_PROC_LOCK_STATUS); + while (1) { erts_aint32_t e; n = e = a; @@ -6033,7 +6127,9 @@ change_proc_schedule_state(Process *p, break; } - if (erts_system_profile_flags.runnable_procs) { + if (prof_runnable_procs) { + + /* Status lock prevents out of order "runnable proc" trace msgs */ if (((n & (ERTS_PSFLG_SUSPENDED | ERTS_PSFLG_ACTIVE)) == ERTS_PSFLG_ACTIVE) @@ -6046,15 +6142,18 @@ change_proc_schedule_state(Process *p, profile_runnable_proc(p, am_active); } + if (lock_status) + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); } + *statep = a; return enqueue; } static ERTS_INLINE void -schedule_process(Process *p, erts_aint32_t in_state) +schedule_process(Process *p, erts_aint32_t in_state, ErtsProcLocks locks) { erts_aint32_t enq_prio = -1; erts_aint32_t state = in_state; @@ -6062,7 +6161,8 @@ schedule_process(Process *p, erts_aint32_t in_state) 0, ERTS_PSFLG_ACTIVE, &state, - &enq_prio); + &enq_prio, + locks); if (enqueue != ERTS_ENQUEUE_NOT) add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio), state, @@ -6070,16 +6170,27 @@ schedule_process(Process *p, erts_aint32_t in_state) } void -erts_schedule_process(Process *p, erts_aint32_t state) +erts_schedule_process(Process *p, erts_aint32_t state, ErtsProcLocks locks) { - schedule_process(p, state); + schedule_process(p, state, locks); } static void schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) { + /* + * Expects status lock to be locked when called, and + * returns with status lock unlocked... + */ erts_aint32_t a = state, n, enq_prio = -1; int enqueue; /* < 0 -> use proxy */ + unsigned int prof_runnable_procs = erts_system_profile_flags.runnable_procs; + + /* Status lock prevents out of order "runnable proc" trace msgs */ + ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p)); + + if (!prof_runnable_procs) + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); ASSERT(!(state & ERTS_PSFLG_PROXY)); @@ -6088,7 +6199,7 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) n = e = a; if (a & ERTS_PSFLG_FREE) - return; /* We don't want to schedule free processes... */ + goto cleanup; /* We don't want to schedule free processes... */ enqueue = ERTS_ENQUEUE_NOT; n |= ERTS_PSFLG_ACTIVE_SYS; @@ -6101,7 +6212,7 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) goto cleanup; } - if (erts_system_profile_flags.runnable_procs) { + if (prof_runnable_procs) { if (!(a & (ERTS_PSFLG_ACTIVE_SYS | ERTS_PSFLG_RUNNING @@ -6111,6 +6222,8 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) profile_runnable_proc(p, am_active); } + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); + prof_runnable_procs = 0; } if (enqueue != ERTS_ENQUEUE_NOT) { @@ -6125,8 +6238,14 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) } cleanup: + + if (prof_runnable_procs) + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); + if (proxy) free_proxy_proc(proxy); + + ERTS_SMP_LC_ASSERT(!(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p))); } static ERTS_INLINE int @@ -6144,7 +6263,7 @@ suspend_process(Process *c_p, Process *p) if (c_p == p) { state = erts_smp_atomic32_read_bor_relb(&p->state, ERTS_PSFLG_SUSPENDED); - ASSERT(state & ERTS_PSFLG_RUNNING); + ASSERT(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS)); suspended = (state & ERTS_PSFLG_SUSPENDED) ? -1: 1; } else { @@ -6193,7 +6312,7 @@ suspend_process(Process *c_p, Process *p) } static ERTS_INLINE void -resume_process(Process *p) +resume_process(Process *p, ErtsProcLocks locks) { erts_aint32_t state, enq_prio = -1; int enqueue; @@ -6210,7 +6329,8 @@ resume_process(Process *p) ERTS_PSFLG_SUSPENDED, 0, &state, - &enq_prio); + &enq_prio, + locks); if (enqueue) add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio), state, @@ -6556,7 +6676,7 @@ suspend_scheduler(ErtsSchedulerData *esdp) int res; do { - res = erts_tse_wait(ssi->event); + res = erts_tse_twait(ssi->event, -1); } while (res == EINTR); } } @@ -6719,6 +6839,7 @@ suspend_scheduler(ErtsSchedulerData *esdp) erts_smp_mtx_unlock(&schdlr_sspnd.mtx); while (1) { + ErtsMonotonicTime current_time; erts_aint32_t qmask; erts_aint32_t flgs; @@ -6743,30 +6864,64 @@ suspend_scheduler(ErtsSchedulerData *esdp) } } - if (!aux_work) { - if (thr_prgr_active) { - erts_thr_progress_active(esdp, thr_prgr_active = 0); - sched_wall_time_change(esdp, 0); + if (aux_work) { + current_time = erts_get_monotonic_time(); + if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) { + if (!thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 1); + sched_wall_time_change(esdp, 1); + } + erts_bump_timers(esdp->timer_wheel, current_time); } - erts_thr_progress_prepare_wait(esdp); - flgs = sched_spin_suspended(ssi, - ERTS_SCHED_SUSPEND_SLEEP_SPINCOUNT); - if (flgs == (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_WAITING - | ERTS_SSI_FLG_SUSPENDED)) { - flgs = sched_set_suspended_sleeptype(ssi); + } + else { + ErtsMonotonicTime timeout_time; + timeout_time = erts_check_next_timeout_time(esdp->timer_wheel, + ERTS_SEC_TO_MONOTONIC(60*60)); + current_time = erts_get_monotonic_time(); + + if (current_time >= timeout_time) { + if (!thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 1); + sched_wall_time_change(esdp, 1); + } + } + else { + if (thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 0); + sched_wall_time_change(esdp, 0); + } + erts_thr_progress_prepare_wait(esdp); + flgs = sched_spin_suspended(ssi, + ERTS_SCHED_SUSPEND_SLEEP_SPINCOUNT); if (flgs == (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_TSE_SLEEPING | ERTS_SSI_FLG_WAITING | ERTS_SSI_FLG_SUSPENDED)) { - int res; - - do { - res = erts_tse_wait(ssi->event); - } while (res == EINTR); + flgs = sched_set_suspended_sleeptype(ssi); + if (flgs == (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_TSE_SLEEPING + | ERTS_SSI_FLG_WAITING + | ERTS_SSI_FLG_SUSPENDED)) { + int res; + + current_time = erts_get_monotonic_time(); + do { + Sint64 timeout; + if (current_time >= timeout_time) + break; + timeout = ERTS_MONOTONIC_TO_NSEC(timeout_time + - current_time + - 1) + 1; + res = erts_tse_twait(ssi->event, timeout); + current_time = erts_get_monotonic_time(); + } while (res == EINTR); + } } + erts_thr_progress_finalize_wait(esdp); } - erts_thr_progress_finalize_wait(esdp); + + if (current_time >= timeout_time) + erts_bump_timers(esdp->timer_wheel, current_time); } flgs = sched_prep_spin_suspended(ssi, (ERTS_SSI_FLG_WAITING @@ -7585,6 +7740,9 @@ sched_thread_func(void *vesdp) ErtsThrPrgrCallbacks callbacks; ErtsSchedulerData *esdp = vesdp; Uint no = esdp->no; + + esdp->timer_wheel = erts_create_timer_wheel((int) no); + esdp->next_tmo_ref = erts_get_next_timeout_reference(esdp->timer_wheel); #ifdef ERTS_SMP ERTS_SCHED_SLEEP_INFO_IX(no - 1)->event = erts_tse_fetch(); callbacks.arg = (void *) esdp->ssi; @@ -7687,6 +7845,8 @@ sched_dirty_cpu_thread_func(void *vesdp) callbacks.wait = NULL; callbacks.finalize_wait = NULL; + esdp->thr_id += erts_no_schedulers; + erts_thr_progress_register_unmanaged_thread(&callbacks); #ifdef ERTS_ENABLE_LOCK_CHECK { @@ -7748,6 +7908,8 @@ sched_dirty_io_thread_func(void *vesdp) callbacks.wait = NULL; callbacks.finalize_wait = NULL; + esdp->thr_id += erts_no_schedulers + erts_no_dirty_cpu_schedulers; + erts_thr_progress_register_unmanaged_thread(&callbacks); #ifdef ERTS_ENABLE_LOCK_CHECK { @@ -7805,20 +7967,17 @@ erts_start_schedulers(void) Uint actual; Uint wanted = erts_no_schedulers; Uint wanted_no_schedulers = erts_no_schedulers; + char name[16]; ethr_thr_opts opts = ETHR_THR_OPTS_DEFAULT_INITER; opts.detached = 1; -#ifdef ETHR_HAVE_THREAD_NAMES - opts.name = malloc(80); -#endif + opts.name = name; #ifdef ERTS_SMP if (erts_runq_supervision_interval) { opts.suggested_stack_size = 16; -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name, "runq_supervisor"); -#endif + erts_snprintf(opts.name, 16, "runq_supervisor"); erts_atomic_init_nob(&runq_supervisor_sleeping, 0); if (0 != ethr_event_init(&runq_supervision_event)) erl_exit(1, "Failed to create run-queue supervision event\n"); @@ -7845,9 +8004,7 @@ erts_start_schedulers(void) ASSERT(actual == esdp->no - 1); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name, "scheduler_%d", actual + 1); -#endif + erts_snprintf(opts.name, 16, "%lu_scheduler", actual + 1); #ifdef __OSE__ /* This should be done in the bind strategy */ @@ -7869,18 +8026,14 @@ erts_start_schedulers(void) int ix; for (ix = 0; ix < erts_no_dirty_cpu_schedulers; ix++) { ErtsSchedulerData *esdp = ERTS_DIRTY_CPU_SCHEDULER_IX(ix); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name,"dirty_cpu_scheduler_%d", ix + 1); -#endif + erts_snprintf(opts.name, 16, "%d_dirty_cpu_scheduler", ix + 1); res = ethr_thr_create(&esdp->tid,sched_dirty_cpu_thread_func,(void*)esdp,&opts); if (res != 0) erl_exit(1, "Failed to create dirty cpu scheduler thread %d\n", ix); } for (ix = 0; ix < erts_no_dirty_io_schedulers; ix++) { ErtsSchedulerData *esdp = ERTS_DIRTY_IO_SCHEDULER_IX(ix); -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name,"dirty_io_scheduler_%d", ix + 1); -#endif + erts_snprintf(opts.name, 16, "%d_dirty_io_scheduler", ix + 1); res = ethr_thr_create(&esdp->tid,sched_dirty_io_thread_func,(void*)esdp,&opts); if (res != 0) erl_exit(1, "Failed to create dirty io scheduler thread %d\n", ix); @@ -7891,9 +8044,7 @@ erts_start_schedulers(void) ERTS_THR_MEMORY_BARRIER; -#ifdef ETHR_HAVE_THREAD_NAMES - sprintf(opts.name, "aux"); -#endif + erts_snprintf(opts.name, 16, "aux"); #ifdef __OSE__ opts.coreNo = 0; @@ -7919,9 +8070,6 @@ erts_start_schedulers(void) erts_send_error_to_logger_nogl(dsbufp); } -#ifdef ETHR_HAVE_THREAD_NAMES - free(opts.name); -#endif } #endif /* ERTS_SMP */ @@ -8023,7 +8171,8 @@ handle_pend_sync_suspend(Process *suspendee, } /* suspender is suspended waiting for suspendee to suspend; resume suspender */ - resume_process(suspender); + ASSERT(suspendee != suspender); + resume_process(suspender, ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspender, ERTS_PROC_LOCK_STATUS); } } @@ -8058,7 +8207,7 @@ pid2proc_not_running(Process *c_p, ErtsProcLocks c_p_locks, ASSERT(c_p->flags & F_P2PNR_RESCHED); c_p->flags &= ~F_P2PNR_RESCHED; if (!suspend && rp) - resume_process(rp); + resume_process(rp, rp_locks); } else { @@ -8216,7 +8365,8 @@ handle_pend_bif_sync_suspend(Process *suspendee, } /* suspender is suspended waiting for suspendee to suspend; resume suspender */ - resume_process(suspender); + ASSERT(suspender != suspendee); + resume_process(suspender, ERTS_PROC_LOCK_LINK|ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspender, ERTS_PROC_LOCK_LINK|ERTS_PROC_LOCK_STATUS); } @@ -8576,7 +8726,8 @@ resume_process_1(BIF_ALIST_1) ASSERT(ERTS_PSFLG_SUSPENDED & erts_smp_atomic32_read_nob(&suspendee->state)); - resume_process(suspendee); + ASSERT(BIF_P != suspendee); + resume_process(suspendee, ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspendee, ERTS_PROC_LOCK_STATUS); } @@ -8706,7 +8857,7 @@ erts_resume(Process* process, ErtsProcLocks process_locks) ERTS_SMP_LC_ASSERT(process_locks == erts_proc_lc_my_proc_locks(process)); if (!(process_locks & ERTS_PROC_LOCK_STATUS)) erts_smp_proc_lock(process, ERTS_PROC_LOCK_STATUS); - resume_process(process); + resume_process(process, process_locks|ERTS_PROC_LOCK_STATUS); if (!(process_locks & ERTS_PROC_LOCK_STATUS)) erts_smp_proc_unlock(process, ERTS_PROC_LOCK_STATUS); } @@ -8725,7 +8876,7 @@ erts_resume_processes(ErtsProcList *list) proc = erts_pid2proc(NULL, 0, plp->pid, ERTS_PROC_LOCK_STATUS); if (proc) { if (erts_proclist_same(plp, proc)) { - resume_process(proc); + resume_process(proc, ERTS_PROC_LOCK_STATUS); nresumed++; } erts_smp_proc_unlock(proc, ERTS_PROC_LOCK_STATUS); @@ -8861,7 +9012,6 @@ Process *schedule(Process *p, int calls) { Process *proxy_p = NULL; ErtsRunQueue *rq; - erts_aint_t dt; ErtsSchedulerData *esdp; int context_reds; int fcalls; @@ -8991,11 +9141,13 @@ Process *schedule(Process *p, int calls) ERTS_SMP_CHK_NO_PROC_LOCKS; - dt = erts_do_time_read_and_reset(); - if (dt) { - erts_smp_runq_unlock(rq); - erts_bump_timer(dt); - erts_smp_runq_lock(rq); + { + ErtsMonotonicTime current_time = erts_get_monotonic_time(); + if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) { + erts_smp_runq_unlock(rq); + erts_bump_timers(esdp->timer_wheel, current_time); + erts_smp_runq_lock(rq); + } } BM_STOP_TIMER(system); @@ -9140,7 +9292,8 @@ Process *schedule(Process *p, int calls) } else if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && (fcalls > input_reductions && - prepare_for_sys_schedule(esdp))) { + prepare_for_sys_schedule(esdp, !0))) { + ErtsMonotonicTime current_time; /* * Schedule system-level activities. */ @@ -9148,15 +9301,15 @@ Process *schedule(Process *p, int calls) erts_smp_atomic32_set_relb(&function_calls, 0); fcalls = 0; - ASSERT(!erts_port_task_have_outstanding_io_tasks()); - #if 0 /* Not needed since we wont wait in sys schedule */ erts_sys_schedule_interrupt(0); #endif erts_smp_runq_unlock(rq); erl_sys_schedule(1); - dt = erts_do_time_read_and_reset(); - if (dt) erts_bump_timer(dt); + + current_time = erts_get_monotonic_time(); + if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) + erts_bump_timers(esdp->timer_wheel, current_time); #ifdef ERTS_SMP erts_smp_runq_lock(rq); @@ -9181,7 +9334,9 @@ Process *schedule(Process *p, int calls) if (RUNQ_READ_LEN(&rq->ports.info.len)) { int have_outstanding_io; have_outstanding_io = erts_port_task_execute(rq, &esdp->current_port); - if ((have_outstanding_io && fcalls > 2*input_reductions) + if ((!erts_eager_check_io + && have_outstanding_io + && fcalls > 2*input_reductions) || rq->halt_in_progress) { /* * If we have performed more than 2*INPUT_REDUCTIONS since @@ -9961,8 +10116,10 @@ erts_internal_request_system_task_3(BIF_ALIST_3) rp_state = n; } - erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_STATUS); - + /* + * schedule_process_sys_task() unlocks status + * lock on process. + */ schedule_process_sys_task(rp, rp_state, NULL); if (free_stqs) @@ -10398,7 +10555,7 @@ alloc_process(ErtsRunQueue *rq, erts_aint32_t state) init_arg.run_queue = rq; init_arg.state = state; - ASSERT(((char *) p) == ((char *) &p->common)); + ERTS_CT_ASSERT(offsetof(Process,common) == 0); if (!erts_ptab_new_element(&erts_proc, &p->common, @@ -10457,7 +10614,10 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). int ix = so->scheduler-1; ASSERT(0 <= ix && ix < erts_no_run_queues); rq = ERTS_RUNQ_IX(ix); - state |= ERTS_PSFLG_BOUND; + if (!(so->flags & SPO_PREFER_SCHED)) { + /* Unsupported feature... */ + state |= ERTS_PSFLG_BOUND; + } } prio = (erts_aint32_t) so->priority; } @@ -10465,6 +10625,9 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). state |= (((prio & ERTS_PSFLGS_PRIO_MASK) << ERTS_PSFLGS_ACT_PRIO_OFFSET) | ((prio & ERTS_PSFLGS_PRIO_MASK) << ERTS_PSFLGS_USR_PRIO_OFFSET)); + if (so->flags & SPO_OFF_HEAP_MSGS) + state |= ERTS_PSFLG_OFF_HEAP_MSGS; + if (!rq) rq = erts_get_runq_proc(parent); @@ -10488,11 +10651,25 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). heap_need = arg_size; p->flags = erts_default_process_flags; + if (so->flags & SPO_OFF_HEAP_MSGS) + p->flags |= F_OFF_HEAP_MSGS; +#ifdef ERTS_SMP + p->preferred_run_queue = NULL; +#endif + p->static_flags = 0; + if (so->flags & SPO_SYSTEM_PROC) + p->static_flags |= ERTS_STC_FLG_SYSTEM_PROC; if (so->flags & SPO_USE_ARGS) { p->min_heap_size = so->min_heap_size; p->min_vheap_size = so->min_vheap_size; p->max_gen_gcs = so->max_gen_gcs; + if (so->flags & SPO_PREFER_SCHED) { +#ifdef ERTS_SMP + p->preferred_run_queue = rq; +#endif + p->static_flags |= ERTS_STC_FLG_PREFER_SCHED; + } } else { p->min_heap_size = H_MIN_SIZE; p->min_vheap_size = BIN_VH_MIN_SIZE; @@ -10570,7 +10747,7 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). #ifdef ERTS_SMP p->common.u.alive.ptimer = NULL; #else - sys_memset(&p->common.u.alive.tm, 0, sizeof(ErlTimer)); + erts_init_timer(&p->common.u.alive.tm); #endif p->common.u.alive.reg = NULL; @@ -10707,7 +10884,7 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). * Schedule process for execution. */ - schedule_process(p, state); + schedule_process(p, state, 0); VERBOSE(DEBUG_PROCESSES, ("Created a new process: %T\n",p->common.id)); @@ -10763,7 +10940,7 @@ void erts_init_empty_process(Process *p) #ifdef ERTS_SMP p->common.u.alive.ptimer = NULL; #else - memset(&(p->common.u.alive.tm), 0, sizeof(ErlTimer)); + erts_init_timer(&p->common.u.alive.tm); #endif p->next = NULL; p->off_heap.first = NULL; @@ -10813,6 +10990,8 @@ void erts_init_empty_process(Process *p) p->parent = NIL; p->approx_started = 0; + p->static_flags = 0; + p->common.u.alive.started_interval = 0; #ifdef HIPE @@ -10838,6 +11017,7 @@ void erts_init_empty_process(Process *p) p->pending_suspenders = NULL; p->pending_exit.reason = THE_NON_VALUE; p->pending_exit.bp = NULL; + p->preferred_run_queue = NULL; erts_proc_lock_init(p); erts_smp_proc_unlock(p, ERTS_PROC_LOCKS_ALL); RUNQ_SET_RQ(&p->run_queue, ERTS_RUNQ_IX(0)); @@ -11028,7 +11208,8 @@ set_proc_exiting(Process *p, ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT, ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE, &state, - &enq_prio); + &enq_prio, + ERTS_PROC_LOCKS_ALL); p->fvalue = reason; if (bp) @@ -11069,7 +11250,8 @@ set_proc_self_exiting(Process *c_p) ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT, ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE, &state, - &enq_prio); + &enq_prio, + ERTS_PROC_LOCKS_ALL); ASSERT(!enqueue); return state; @@ -11461,7 +11643,8 @@ static void doit_exit_monitor(ErtsMonitor *mon, void *vpcontext) ErtsMonitor *rmon; Process *rp; - if (mon->type == MON_ORIGIN) { + switch (mon->type) { + case MON_ORIGIN: /* We are monitoring someone else, we need to demonitor that one.. */ if (is_atom(mon->pid)) { /* remote by name */ ASSERT(is_node_name_atom(mon->pid)); @@ -11524,7 +11707,8 @@ static void doit_exit_monitor(ErtsMonitor *mon, void *vpcontext) } } } - } else { /* type == MON_TARGET */ + break; + case MON_TARGET: ASSERT(mon->type == MON_TARGET); ASSERT(is_pid(mon->pid) || is_internal_port(mon->pid)); if (is_internal_port(mon->pid)) { @@ -11583,6 +11767,12 @@ static void doit_exit_monitor(ErtsMonitor *mon, void *vpcontext) } } } + break; + case MON_TIME_OFFSET: + erts_demonitor_time_offset(mon->ref); + break; + default: + ERTS_INTERNAL_ERROR("Invalid monitor type"); } done: /* As the monitors are previously removed from the process, @@ -11714,8 +11904,9 @@ resume_suspend_monitor(ErtsSuspendMonitor *smon, void *vc_p) Process *suspendee = erts_pid2proc((Process *) vc_p, ERTS_PROC_LOCK_MAIN, smon->pid, ERTS_PROC_LOCK_STATUS); if (suspendee) { + ASSERT(suspendee != vc_p); if (smon->active) - resume_process(suspendee); + resume_process(suspendee, ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspendee, ERTS_PROC_LOCK_STATUS); } erts_destroy_suspend_monitor(smon); @@ -11740,6 +11931,9 @@ erts_do_exit_process(Process* p, Eterm reason) } #endif + if (p->static_flags & ERTS_STC_FLG_SYSTEM_PROC) + erl_exit(1, "System process %T terminated: %T\n", p->common.id, reason); + #ifdef ERTS_SMP ERTS_SMP_CHK_HAVE_ONLY_MAIN_PROC_LOCK(p); /* By locking all locks (main lock is already locked) when going @@ -11807,6 +12001,7 @@ erts_continue_exit_process(Process *p) struct saved_calls *scb; process_breakpoint_time_t *pbt; erts_aint32_t state; + void *nif_export; #ifdef DEBUG int yield_allowed = 1; @@ -11957,6 +12152,7 @@ erts_continue_exit_process(Process *p) : NULL); scb = ERTS_PROC_SET_SAVED_CALLS_BUF(p, ERTS_PROC_LOCKS_ALL, NULL); pbt = ERTS_PROC_SET_CALL_TIME(p, ERTS_PROC_LOCKS_ALL, NULL); + nif_export = ERTS_PROC_SET_NIF_TRAP_EXPORT(p, ERTS_PROC_LOCKS_ALL, NULL); erts_smp_proc_unlock(p, ERTS_PROC_LOCKS_ALL); #ifdef BM_COUNTERS @@ -12004,6 +12200,9 @@ erts_continue_exit_process(Process *p) if (pbt) erts_free(ERTS_ALC_T_BPD, (void *) pbt); + if (nif_export) + erts_destroy_nif_export(nif_export); + delete_process(p); #ifdef ERTS_SMP @@ -12048,7 +12247,7 @@ timeout_proc(Process* p) state = erts_smp_atomic32_read_acqb(&p->state); if (!(state & ERTS_PSFLG_ACTIVE)) - schedule_process(p, state); + schedule_process(p, state, ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_STATUS); } @@ -12109,7 +12308,7 @@ erts_stack_dump(int to, void *to_arg, Process *p) } erts_program_counter_info(to, to_arg, p); for (sp = p->stop; sp < STACK_START(p); sp++) { - yreg = stack_element_dump(to, to_arg, p, sp, yreg); + yreg = stack_element_dump(to, to_arg, sp, yreg); } } @@ -12166,7 +12365,7 @@ print_function_from_pc(int to, void *to_arg, BeamInstr* x) } static int -stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, int yreg) +stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg) { Eterm x = *sp; @@ -12195,6 +12394,214 @@ stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, int yreg) } /* + * Print scheduler information + */ +void +erts_print_scheduler_info(int to, void *to_arg, ErtsSchedulerData *esdp) { + int i; + erts_aint32_t flg; + Process *p; + + erts_print(to, to_arg, "=scheduler:%u\n", esdp->no); + +#ifdef ERTS_SMP + flg = erts_smp_atomic32_read_dirty(&esdp->ssi->flags); + erts_print(to, to_arg, "Scheduler Sleep Info Flags: "); + for (i = 0; i < ERTS_SSI_FLGS_MAX && flg; i++) { + erts_aint32_t chk = (1 << i); + if (flg & chk) { + switch (chk) { + case ERTS_SSI_FLG_SLEEPING: + erts_print(to, to_arg, "SLEEPING"); break; + case ERTS_SSI_FLG_POLL_SLEEPING: + erts_print(to, to_arg, "POLL_SLEEPING"); break; + case ERTS_SSI_FLG_TSE_SLEEPING: + erts_print(to, to_arg, "TSE_SLEEPING"); break; + case ERTS_SSI_FLG_WAITING: + erts_print(to, to_arg, "WAITING"); break; + case ERTS_SSI_FLG_SUSPENDED: + erts_print(to, to_arg, "SUSPENDED"); break; + default: + erts_print(to, to_arg, "UNKNOWN(%d)", flg); break; + } + if (flg > chk) + erts_print(to, to_arg, " | "); + flg -= chk; + } + } + erts_print(to, to_arg, "\n"); +#endif + + flg = erts_atomic32_read_dirty(&esdp->ssi->aux_work); + erts_print(to, to_arg, "Scheduler Sleep Info Aux Work: "); + for (i = 0; i < ERTS_SSI_AUX_WORK_MAX && flg; i++) { + erts_aint32_t chk = (1 << i); + if (flg & chk) { + switch (chk) { + case ERTS_SSI_AUX_WORK_DELAYED_AW_WAKEUP: + erts_print(to, to_arg, "DELAYED_AW_WAKEUP"); break; + case ERTS_SSI_AUX_WORK_DD: + erts_print(to, to_arg, "DELAYED_DEALLOC"); break; + case ERTS_SSI_AUX_WORK_DD_THR_PRGR: + erts_print(to, to_arg, "DELAYED_DEALLOC_THR_PRGR"); break; + case ERTS_SSI_AUX_WORK_FIX_ALLOC_DEALLOC: + erts_print(to, to_arg, "FIX_ALLOC_DEALLOC"); break; + case ERTS_SSI_AUX_WORK_FIX_ALLOC_LOWER_LIM: + erts_print(to, to_arg, "FIX_ALLOC_LOWER_LIM"); break; + case ERTS_SSI_AUX_WORK_THR_PRGR_LATER_OP: + erts_print(to, to_arg, "THR_PRGR_LATER_OP"); break; + case ERTS_SSI_AUX_WORK_ASYNC_READY: + erts_print(to, to_arg, "ASYNC_READY"); break; + case ERTS_SSI_AUX_WORK_ASYNC_READY_CLEAN: + erts_print(to, to_arg, "ASYNC_READY_CLEAN"); break; + case ERTS_SSI_AUX_WORK_MISC_THR_PRGR: + erts_print(to, to_arg, "MISC_THR_PRGR"); break; + case ERTS_SSI_AUX_WORK_MISC: + erts_print(to, to_arg, "MISC"); break; + case ERTS_SSI_AUX_WORK_CHECK_CHILDREN: + erts_print(to, to_arg, "CHECK_CHILDREN"); break; + case ERTS_SSI_AUX_WORK_SET_TMO: + erts_print(to, to_arg, "SET_TMO"); break; + case ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK: + erts_print(to, to_arg, "MSEG_CACHE_CHECK"); break; + case ERTS_SSI_AUX_WORK_REAP_PORTS: + erts_print(to, to_arg, "REAP_PORTS"); break; + default: + erts_print(to, to_arg, "UNKNOWN(%d)", flg); break; + } + if (flg > chk) + erts_print(to, to_arg, " | "); + flg -= chk; + } + } + erts_print(to, to_arg, "\n"); + + erts_print(to, to_arg, "Current Port: "); + if (esdp->current_port) + erts_print(to, to_arg, "%T", esdp->current_port->common.id); + erts_print(to, to_arg, "\n"); + + p = esdp->current_process; + erts_print(to, to_arg, "Current Process: "); + if (esdp->current_process && !(ERTS_TRACE_FLAGS(p) & F_SENSITIVE)) { + flg = erts_smp_atomic32_read_dirty(&p->state); + erts_print(to, to_arg, "%T\n", p->common.id); + + erts_print(to, to_arg, "Current Process State: "); + erts_dump_process_state(to, to_arg, flg); + + erts_print(to, to_arg, "Current Process Internal State: "); + erts_dump_extended_process_state(to, to_arg, flg); + + erts_print(to, to_arg, "Current Process Program counter: %p (", p->i); + print_function_from_pc(to, to_arg, p->i); + erts_print(to, to_arg, ")\n"); + erts_print(to, to_arg, "Current Process CP: %p (", p->cp); + print_function_from_pc(to, to_arg, p->cp); + erts_print(to, to_arg, ")\n"); + + /* Getting this stacktrace can segfault if we are very very + unlucky if called while a process is being garbage collected. + Therefore we only call this on other schedulers if we either + have protection against segfaults, or we know that the process + is not garbage collecting. It *should* always be safe to call + on a process owned by us, even if it is currently being garbage + collected. + */ + erts_print(to, to_arg, "Current Process Limited Stack Trace:\n"); + erts_limited_stack_trace(to, to_arg, p); + } else + erts_print(to, to_arg, "\n"); + + for (i = 0; i < ERTS_NO_PROC_PRIO_LEVELS; i++) { + erts_print(to, to_arg, "Run Queue "); + switch (i) { + case PRIORITY_MAX: + erts_print(to, to_arg, "Max "); + break; + case PRIORITY_HIGH: + erts_print(to, to_arg, "High "); + break; + case PRIORITY_NORMAL: + erts_print(to, to_arg, "Normal "); + break; + case PRIORITY_LOW: + erts_print(to, to_arg, "Low "); + break; + default: + erts_print(to, to_arg, "Unknown "); + break; + } + erts_print(to, to_arg, "Length: %d\n", + erts_smp_atomic32_read_dirty(&esdp->run_queue->procs.prio_info[i].len)); + } + erts_print(to, to_arg, "Run Queue Port Length: %d\n", + erts_smp_atomic32_read_dirty(&esdp->run_queue->ports.info.len)); + + flg = erts_smp_atomic32_read_dirty(&esdp->run_queue->flags); + erts_print(to, to_arg, "Run Queue Flags: "); + for (i = 0; i < ERTS_RUNQ_FLG_MAX && flg; i++) { + erts_aint32_t chk = (1 << i); + if (flg & chk) { + switch (chk) { + case (1 << PRIORITY_MAX): + erts_print(to, to_arg, "NONEMPTY_MAX"); break; + case (1 << PRIORITY_HIGH): + erts_print(to, to_arg, "NONEMPTY_HIGH"); break; + case (1 << PRIORITY_NORMAL): + erts_print(to, to_arg, "NONEMPTY_NORMAL"); break; + case (1 << PRIORITY_LOW): + erts_print(to, to_arg, "NONEMPTY_LOW"); break; + case (1 << (PRIORITY_MAX + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_MAX"); break; + case (1 << (PRIORITY_HIGH + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_HIGH"); break; + case (1 << (PRIORITY_NORMAL + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_NORMAL"); break; + case (1 << (PRIORITY_LOW + ERTS_RUNQ_FLGS_EMIGRATE_SHFT)): + erts_print(to, to_arg, "EMIGRATE_LOW"); break; + case (1 << (PRIORITY_MAX + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_MAX"); break; + case (1 << (PRIORITY_HIGH + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_HIGH"); break; + case (1 << (PRIORITY_NORMAL + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_NORMAL"); break; + case (1 << (PRIORITY_LOW + ERTS_RUNQ_FLGS_IMMIGRATE_SHFT)): + erts_print(to, to_arg, "IMMIGRATE_LOW"); break; + case (1 << (PRIORITY_MAX + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_MAX"); break; + case (1 << (PRIORITY_HIGH + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_HIGH"); break; + case (1 << (PRIORITY_NORMAL + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_NORMAL"); break; + case (1 << (PRIORITY_LOW + ERTS_RUNQ_FLGS_EVACUATE_SHFT)): + erts_print(to, to_arg, "EVACUATE_LOW"); break; + case ERTS_RUNQ_FLG_OUT_OF_WORK: + erts_print(to, to_arg, "OUT_OF_WORK"); break; + case ERTS_RUNQ_FLG_HALFTIME_OUT_OF_WORK: + erts_print(to, to_arg, "HALFTIME_OUT_OF_WORK"); break; + case ERTS_RUNQ_FLG_SUSPENDED: + erts_print(to, to_arg, "SUSPENDED"); break; + case ERTS_RUNQ_FLG_CHK_CPU_BIND: + erts_print(to, to_arg, "CHK_CPU_BIND"); break; + case ERTS_RUNQ_FLG_INACTIVE: + erts_print(to, to_arg, "INACTIVE"); break; + case ERTS_RUNQ_FLG_NONEMPTY: + erts_print(to, to_arg, "NONEMPTY"); break; + case ERTS_RUNQ_FLG_PROTECTED: + erts_print(to, to_arg, "PROTECTED"); break; + default: + erts_print(to, to_arg, "UNKNOWN(%d)", flg); break; + } + if (flg > chk) + erts_print(to, to_arg, " | "); + flg -= chk; + } + } + erts_print(to, to_arg, "\n"); +} + +/* * A nice system halt closing all open port goes as follows: * 1) This function schedules the aux work ERTS_SSI_AUX_WORK_REAP_PORTS * on all schedulers, then schedules itself out. diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index ed6dadbffa..743711cc3b 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.h @@ -58,6 +58,7 @@ typedef struct process Process; #include "external.h" #include "erl_mseg.h" #include "erl_async.h" +#include "erl_gc.h" #ifdef HIPE #include "hipe_process.h" @@ -104,6 +105,7 @@ struct saved_calls { }; extern Export exp_send, exp_receive, exp_timeout; +extern int erts_eager_check_io; extern int erts_sched_compact_load; extern int erts_sched_balance_util; extern Uint erts_no_schedulers; @@ -168,6 +170,8 @@ extern int erts_sched_thread_suggested_stack_size; #define ERTS_RUNQ_FLG_PROTECTED \ (((Uint32) 1) << (ERTS_RUNQ_FLG_BASE2 + 6)) +#define ERTS_RUNQ_FLG_MAX (ERTS_RUNQ_FLG_BASE2 + 7) + #define ERTS_RUNQ_FLGS_MIGRATION_QMASKS \ (ERTS_RUNQ_FLGS_EMIGRATE_QMASK \ | ERTS_RUNQ_FLGS_IMMIGRATE_QMASK \ @@ -250,6 +254,8 @@ typedef enum { #define ERTS_SSI_FLG_WAITING (((erts_aint32_t) 1) << 3) #define ERTS_SSI_FLG_SUSPENDED (((erts_aint32_t) 1) << 4) +#define ERTS_SSI_FLGS_MAX 5 + #define ERTS_SSI_FLGS_SLEEP_TYPE \ (ERTS_SSI_FLG_TSE_SLEEPING|ERTS_SSI_FLG_POLL_SLEEPING) @@ -281,6 +287,8 @@ typedef enum { #define ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK (((erts_aint32_t) 1) << 12) #define ERTS_SSI_AUX_WORK_REAP_PORTS (((erts_aint32_t) 1) << 13) +#define ERTS_SSI_AUX_WORK_MAX 14 + typedef struct ErtsSchedulerSleepInfo_ ErtsSchedulerSleepInfo; #ifdef ERTS_DIRTY_SCHEDULERS @@ -340,7 +348,7 @@ typedef struct { } ErtsRunQueueInfo; -#ifdef HAVE_GETHRTIME +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT # undef ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT # define ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT 1 #endif @@ -350,13 +358,7 @@ typedef struct { #undef ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT #define ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT -#ifdef ARCH_64 -typedef erts_atomic_t ErtsAtomicSchedTime; -#elif defined(ARCH_32) -typedef erts_dw_atomic_t ErtsAtomicSchedTime; -#else -# error :-/ -#endif +typedef erts_atomic64_t ErtsAtomicSchedTime; #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT typedef struct { @@ -488,11 +490,6 @@ typedef struct { } ErtsSchedWallTime; typedef struct { - Uint64 reclaimed; - Uint64 garbage_cols; -} ErtsGCInfo; - -typedef struct { int sched; erts_aint32_t aux_work; } ErtsDelayedAuxWorkWakeupJob; @@ -567,6 +564,8 @@ struct ErtsSchedulerData_ { Eterm* x_reg_array; /* X registers */ FloatDef* f_reg_array; /* Floating point registers. */ + ErtsTimerWheel *timer_wheel; + ErtsNextTimeoutRef next_tmo_ref; #ifdef ERTS_SMP ethr_tid tid; /* Thread id */ struct erl_bits_state erl_bits_state; /* erl_bits.c state */ @@ -593,6 +592,10 @@ struct ErtsSchedulerData_ { ErtsAuxWorkData aux_work_data; ErtsAtomCacheMap atom_cache_map; + Uint32 thr_id; + Uint64 unique; + Uint64 ref; + ErtsSchedAllocData alloc_data; Uint64 reductions; @@ -734,13 +737,9 @@ erts_smp_reset_max_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi) #define ERTS_PSD_DIST_ENTRY 3 #define ERTS_PSD_CALL_TIME_BP 4 #define ERTS_PSD_DELAYED_GC_TASK_QS 5 -#ifdef ERTS_DIRTY_SCHEDULERS -#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT 6 +#define ERTS_PSD_NIF_TRAP_EXPORT 6 #define ERTS_PSD_SIZE 7 -#else -#define ERTS_PSD_SIZE 6 -#endif typedef struct { void *data[ERTS_PSD_SIZE]; @@ -767,10 +766,8 @@ typedef struct { #define ERTS_PSD_DELAYED_GC_TASK_QS_GET_LOCKS ERTS_PROC_LOCK_MAIN #define ERTS_PSD_DELAYED_GC_TASK_QS_SET_LOCKS ERTS_PROC_LOCK_MAIN -#ifdef ERTS_DIRTY_SCHEDULERS -#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_GET_LOCKS ERTS_PROC_LOCK_MAIN -#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_SET_LOCKS ERTS_PROC_LOCK_MAIN -#endif +#define ERTS_PSD_NIF_TRAP_EXPORT_GET_LOCKS ERTS_PROC_LOCK_MAIN +#define ERTS_PSD_NIF_TRAP_EXPORT_SET_LOCKS ERTS_PROC_LOCK_MAIN typedef struct { ErtsProcLocks get_locks; @@ -945,6 +942,8 @@ struct process { Eterm parent; /* Pid of process that created this process. */ erts_approx_time_t approx_started; /* Time when started. */ + Uint32 static_flags; /* Flags that do *not* change */ + /* This is the place, where all fields that differs between memory * architectures, have gone to. */ @@ -976,6 +975,7 @@ struct process { ErtsSchedulerData *scheduler_data; Eterm suspendee; ErtsPendingSuspend *pending_suspenders; + ErtsRunQueue *preferred_run_queue; erts_smp_atomic_t run_queue; #ifdef HIPE struct hipe_process_state_smp hipe_smp; @@ -1085,11 +1085,15 @@ void erts_check_for_holes(Process* p); #define ERTS_PSFLG_RUNNING_SYS ERTS_PSFLG_BIT(15) #define ERTS_PSFLG_PROXY ERTS_PSFLG_BIT(16) #define ERTS_PSFLG_DELAYED_SYS ERTS_PSFLG_BIT(17) +#define ERTS_PSFLG_OFF_HEAP_MSGS ERTS_PSFLG_BIT(18) #ifdef ERTS_DIRTY_SCHEDULERS -#define ERTS_PSFLG_DIRTY_CPU_PROC ERTS_PSFLG_BIT(18) -#define ERTS_PSFLG_DIRTY_IO_PROC ERTS_PSFLG_BIT(19) -#define ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q ERTS_PSFLG_BIT(20) -#define ERTS_PSFLG_DIRTY_IO_PROC_IN_Q ERTS_PSFLG_BIT(21) +#define ERTS_PSFLG_DIRTY_CPU_PROC ERTS_PSFLG_BIT(19) +#define ERTS_PSFLG_DIRTY_IO_PROC ERTS_PSFLG_BIT(20) +#define ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q ERTS_PSFLG_BIT(21) +#define ERTS_PSFLG_DIRTY_IO_PROC_IN_Q ERTS_PSFLG_BIT(22) +#define ERTS_PSFLG_MAX (ERTS_PSFLGS_ZERO_BIT_OFFSET + 23) +#else +#define ERTS_PSFLG_MAX (ERTS_PSFLGS_ZERO_BIT_OFFSET + 19) #endif #define ERTS_PSFLGS_IN_PRQ_MASK (ERTS_PSFLG_IN_PRQ_MAX \ @@ -1104,6 +1108,12 @@ void erts_check_for_holes(Process* p); #define ERTS_PSFLGS_GET_PRQ_PRIO(PSFLGS) \ (((PSFLGS) >> ERTS_PSFLGS_USR_PRIO_OFFSET) & ERTS_PSFLGS_PRIO_MASK) +/* + * Static flags that do not change after process creation. + */ +#define ERTS_STC_FLG_SYSTEM_PROC (((Uint32) 1) << 0) +#define ERTS_STC_FLG_PREFER_SCHED (((Uint32) 1) << 1) + /* The sequential tracing token is a tuple of size 5: * * {Flags, Label, Serial, Sender} @@ -1131,6 +1141,9 @@ void erts_check_for_holes(Process* p); #define SPO_LINK 1 #define SPO_USE_ARGS 2 #define SPO_MONITOR 4 +#define SPO_OFF_HEAP_MSGS 8 +#define SPO_SYSTEM_PROC 16 +#define SPO_PREFER_SCHED 32 /* * The following struct contains options for a process to be spawned. @@ -1218,6 +1231,7 @@ extern struct erts_system_profile_flags_t erts_system_profile_flags; #define F_P2PNR_RESCHED (1 << 9) /* Process has been rescheduled via erts_pid2proc_not_running() */ #define F_FORCE_GC (1 << 10) /* Force gc at process in-scheduling */ #define F_DISABLE_GC (1 << 11) /* Disable GC */ +#define F_OFF_HEAP_MSGS (1 << 12) /* process trace_flags */ #define F_SENSITIVE (1 << 0) @@ -1367,6 +1381,9 @@ Uint64 erts_get_proc_interval(void); Uint64 erts_ensure_later_proc_interval(Uint64); Uint64 erts_step_proc_interval(void); +int erts_setup_nif_gc(Process* proc, Eterm** objv, int* nobj); /* see erl_nif.c */ +void erts_destroy_nif_export(void *); /* see erl_nif.c */ + ErtsProcList *erts_proclist_create(Process *); void erts_proclist_destroy(ErtsProcList *); @@ -1622,7 +1639,11 @@ void erts_cleanup_empty_process(Process* p); void erts_debug_verify_clean_empty_process(Process* p); #endif void erts_stack_dump(int to, void *to_arg, Process *); +void erts_limited_stack_trace(int to, void *to_arg, Process *); void erts_program_counter_info(int to, void *to_arg, Process *); +void erts_print_scheduler_info(int to, void *to_arg, ErtsSchedulerData *esdp); +void erts_dump_extended_process_state(int to, void *to_arg, erts_aint32_t psflg); +void erts_dump_process_state(int to, void *to_arg, erts_aint32_t psflg); Eterm erts_get_process_priority(Process *p); Eterm erts_set_process_priority(Process *p, Eterm prio); @@ -1704,17 +1725,17 @@ ErtsSchedulerData *erts_get_scheduler_data(void) #endif #endif -void erts_schedule_process(Process *, erts_aint32_t); +void erts_schedule_process(Process *, erts_aint32_t, ErtsProcLocks); -ERTS_GLB_INLINE void erts_proc_notify_new_message(Process *p); +ERTS_GLB_INLINE void erts_proc_notify_new_message(Process *p, ErtsProcLocks locks); #if ERTS_GLB_INLINE_INCL_FUNC_DEF ERTS_GLB_INLINE void -erts_proc_notify_new_message(Process *p) +erts_proc_notify_new_message(Process *p, ErtsProcLocks locks) { /* No barrier needed, due to msg lock */ erts_aint32_t state = erts_smp_atomic32_read_nob(&p->state); if (!(state & ERTS_PSFLG_ACTIVE)) - erts_schedule_process(p, state); + erts_schedule_process(p, state, locks); } #endif @@ -1817,12 +1838,10 @@ erts_psd_set(Process *p, ErtsProcLocks plocks, int ix, void *data) #define ERTS_PROC_SET_DELAYED_GC_TASK_QS(P, L, PBT) \ ((ErtsProcSysTaskQs *) erts_psd_set((P), (L), ERTS_PSD_DELAYED_GC_TASK_QS, (void *) (PBT))) -#ifdef ERTS_DIRTY_SCHEDULERS -#define ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(P) \ - ((Export *) erts_psd_get((P), ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT)) -#define ERTS_PROC_SET_DIRTY_SCHED_TRAP_EXPORT(P, L, DSTE) \ - ((Export *) erts_psd_set((P), (L), ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT, (void *) (DSTE))) -#endif +#define ERTS_PROC_GET_NIF_TRAP_EXPORT(P) \ + erts_psd_get((P), ERTS_PSD_NIF_TRAP_EXPORT) +#define ERTS_PROC_SET_NIF_TRAP_EXPORT(P, L, NTE) \ + erts_psd_set((P), (L), ERTS_PSD_NIF_TRAP_EXPORT, (void *) (NTE)) ERTS_GLB_INLINE Eterm erts_proc_get_error_handler(Process *p); @@ -2229,6 +2248,8 @@ extern int erts_disable_proc_not_running_opt; void erts_smp_notify_inc_runq(ErtsRunQueue *runq); +void erts_interupt_aux_thread_timed(ErtsMonotonicTime timeout_time); + #ifdef ERTS_SMP void erts_sched_finish_poke(ErtsSchedulerSleepInfo *, erts_aint32_t); ERTS_GLB_INLINE void erts_sched_poke(ErtsSchedulerSleepInfo *ssi); diff --git a/erts/emulator/beam/erl_process_dict.c b/erts/emulator/beam/erl_process_dict.c index 23e5bf737f..00761f2d0e 100644 --- a/erts/emulator/beam/erl_process_dict.c +++ b/erts/emulator/beam/erl_process_dict.c @@ -47,7 +47,7 @@ /* Hash constant macros */ #define MAX_HASH 1342177280UL -#define INITIAL_SIZE 10 +#define INITIAL_SIZE (erts_pd_initial_size) /* Hash utility macros */ #define HASH_RANGE(PDict) ((PDict)->homeSize + (PDict)->splitPosition) @@ -82,6 +82,7 @@ static void pd_hash_erase(Process *p, Eterm id, Eterm *ret); static void pd_hash_erase_all(Process *p); static Eterm pd_hash_get_keys(Process *p, Eterm value); +static Eterm pd_hash_get_all_keys(Process *p, ProcDict *pd); static Eterm pd_hash_get_all(Process *p, ProcDict *pd); static Eterm pd_hash_put(Process *p, Eterm id, Eterm value); @@ -275,6 +276,16 @@ BIF_RETTYPE get_1(BIF_ALIST_1) BIF_RET(ret); } +BIF_RETTYPE get_keys_0(BIF_ALIST_0) +{ + Eterm ret; + + PD_CHECK(BIF_P->dictionary); + ret = pd_hash_get_all_keys(BIF_P,BIF_P->dictionary); + PD_CHECK(BIF_P->dictionary); + BIF_RET(ret); +} + BIF_RETTYPE get_keys_1(BIF_ALIST_1) { Eterm ret; @@ -412,6 +423,47 @@ Eterm erts_pd_hash_get(Process *p, Eterm id) return am_undefined; } +#define PD_GET_TKEY(Dst,Src) \ +do { \ + ASSERT(is_tuple((Src))); \ + ASSERT(arityval(*((Eterm*)tuple_val((Src)))) == 2); \ + (Dst) = ((Eterm*)tuple_val((Src)))[1]; \ +} while(0) + +static Eterm pd_hash_get_all_keys(Process *p, ProcDict *pd) { + Eterm* hp; + Eterm res = NIL; + Eterm tmp, tmp2; + unsigned int i; + unsigned int num; + + if (pd == NULL) { + return res; + } + + num = HASH_RANGE(pd); + hp = HAlloc(p, pd->numElements * 2); + + for (i = 0; i < num; ++i) { + tmp = ARRAY_GET(pd, i); + if (is_boxed(tmp)) { + PD_GET_TKEY(tmp,tmp); + res = CONS(hp, tmp, res); + hp += 2; + } else if (is_list(tmp)) { + while (tmp != NIL) { + tmp2 = TCAR(tmp); + PD_GET_TKEY(tmp2,tmp2); + res = CONS(hp, tmp2, res); + hp += 2; + tmp = TCDR(tmp); + } + } + } + return res; +} +#undef PD_GET_TKEY + static Eterm pd_hash_get_keys(Process *p, Eterm value) { Eterm *hp; diff --git a/erts/emulator/beam/erl_process_dump.c b/erts/emulator/beam/erl_process_dump.c index 2f3cf23b00..36bb6b2f0e 100644 --- a/erts/emulator/beam/erl_process_dump.c +++ b/erts/emulator/beam/erl_process_dump.c @@ -43,8 +43,9 @@ static void dump_process_info(int to, void *to_arg, Process *p); static void dump_element(int to, void *to_arg, Eterm x); static void dump_dist_ext(int to, void *to_arg, ErtsDistExternal *edep); static void dump_element_nl(int to, void *to_arg, Eterm x); -static int stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, +static int stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg); +static void stack_trace_dump(int to, void *to_arg, Eterm* sp); static void print_function_from_pc(int to, void *to_arg, BeamInstr* x); static void heap_dump(int to, void *to_arg, Eterm x); static void dump_binaries(int to, void *to_arg, Binary* root); @@ -148,7 +149,7 @@ dump_process_info(int to, void *to_arg, Process *p) if ((ERTS_TRACE_FLAGS(p) & F_SENSITIVE) == 0) { erts_print(to, to_arg, "=proc_stack:%T\n", p->common.id); for (sp = p->stop; sp < STACK_START(p); sp++) { - yreg = stack_element_dump(to, to_arg, p, sp, yreg); + yreg = stack_element_dump(to, to_arg, sp, yreg); } erts_print(to, to_arg, "=proc_heap:%T\n", p->common.id); @@ -243,9 +244,65 @@ dump_element_nl(int to, void *to_arg, Eterm x) erts_putc(to, to_arg, '\n'); } +static void +stack_trace_dump(int to, void *to_arg, Eterm *sp) { + Eterm x = *sp; + if (is_CP(x)) { + erts_print(to, to_arg, "%p:", sp); + erts_print(to, to_arg, "SReturn addr 0x%X (", cp_val(x)); + print_function_from_pc(to, to_arg, cp_val(x)); + erts_print(to, to_arg, ")\n"); + } +} + +void +erts_limited_stack_trace(int to, void *to_arg, Process *p) +{ + Eterm* sp; + + + if (ERTS_TRACE_FLAGS(p) & F_SENSITIVE) { + return; + } + + if (STACK_START(p) < STACK_TOP(p)) { + return; + } + + if ((STACK_START(p) - STACK_TOP(p)) < 512) { + if (erts_sys_is_area_readable((char*)STACK_TOP(p), + (char*)STACK_START(p))) + for (sp = STACK_TOP(p); sp < STACK_START(p); sp++) + stack_trace_dump(to, to_arg, sp); + else + erts_print(to, to_arg, "Could not read from stack memory: %p - %p\n", + STACK_TOP(p), STACK_START(p)); + } else { + sp = STACK_TOP(p); + if (erts_sys_is_area_readable((char*)STACK_TOP(p), + (char*)(STACK_TOP(p) + 25))) + for (; sp < (STACK_TOP(p) + 256); sp++) + stack_trace_dump(to, to_arg, sp); + else + erts_print(to, to_arg, "Could not read from stack memory: %p - %p\n", + STACK_TOP(p), STACK_TOP(p) + 256); + + erts_print(to, to_arg, "%p: skipping %d frames\n", + sp, STACK_START(p) - STACK_TOP(p) - 512); + + if (erts_sys_is_area_readable((char*)(STACK_START(p) - 256), + (char*)STACK_START(p))) + for (sp = STACK_START(p) - 256; sp < STACK_START(p); sp++) + stack_trace_dump(to, to_arg, sp); + else + erts_print(to, to_arg, "Could not read from stack memory: %p - %p\n", + STACK_START(p) - 256, STACK_START(p)); + } + +} static int -stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, int yreg) +stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg) { Eterm x = *sp; @@ -508,3 +565,114 @@ dump_externally(int to, void *to_arg, Eterm term) erts_print(to, to_arg, "%02X", *s++); } } + +void erts_dump_process_state(int to, void *to_arg, erts_aint32_t psflg) { + if (psflg & ERTS_PSFLG_FREE) + erts_print(to, to_arg, "Non Existing\n"); /* Should never happen */ + else if (psflg & ERTS_PSFLG_EXITING) + erts_print(to, to_arg, "Exiting\n"); + else if (psflg & ERTS_PSFLG_GC) { + erts_print(to, to_arg, "Garbing\n"); + } + else if (psflg & ERTS_PSFLG_SUSPENDED) + erts_print(to, to_arg, "Suspended\n"); + else if (psflg & ERTS_PSFLG_RUNNING) { + erts_print(to, to_arg, "Running\n"); + } + else if (psflg & ERTS_PSFLG_ACTIVE) + erts_print(to, to_arg, "Scheduled\n"); + else + erts_print(to, to_arg, "Waiting\n"); +} + +void +erts_dump_extended_process_state(int to, void *to_arg, erts_aint32_t psflg) { + + int i; + + switch (ERTS_PSFLGS_GET_ACT_PRIO(psflg)) { + case PRIORITY_MAX: erts_print(to, to_arg, "ACT_PRIO_MAX | "); break; + case PRIORITY_HIGH: erts_print(to, to_arg, "ACT_PRIO_HIGH | "); break; + case PRIORITY_NORMAL: erts_print(to, to_arg, "ACT_PRIO_NORMAL | "); break; + case PRIORITY_LOW: erts_print(to, to_arg, "ACT_PRIO_LOW | "); break; + } + switch (ERTS_PSFLGS_GET_USR_PRIO(psflg)) { + case PRIORITY_MAX: erts_print(to, to_arg, "USR_PRIO_MAX | "); break; + case PRIORITY_HIGH: erts_print(to, to_arg, "USR_PRIO_HIGH | "); break; + case PRIORITY_NORMAL: erts_print(to, to_arg, "USR_PRIO_NORMAL | "); break; + case PRIORITY_LOW: erts_print(to, to_arg, "USR_PRIO_LOW | "); break; + } + switch (ERTS_PSFLGS_GET_PRQ_PRIO(psflg)) { + case PRIORITY_MAX: erts_print(to, to_arg, "PRQ_PRIO_MAX"); break; + case PRIORITY_HIGH: erts_print(to, to_arg, "PRQ_PRIO_HIGH"); break; + case PRIORITY_NORMAL: erts_print(to, to_arg, "PRQ_PRIO_NORMAL"); break; + case PRIORITY_LOW: erts_print(to, to_arg, "PRQ_PRIO_LOW"); break; + } + + psflg &= ~(ERTS_PSFLGS_ACT_PRIO_MASK | + ERTS_PSFLGS_USR_PRIO_MASK | + ERTS_PSFLGS_PRQ_PRIO_MASK); + + if (psflg) + erts_print(to, to_arg, " | "); + + for (i = 0; i < ERTS_PSFLG_MAX && psflg; i++) { + erts_aint32_t chk = (1 << i); + if (psflg & chk) { + switch (chk) { + case ERTS_PSFLG_IN_PRQ_MAX: + erts_print(to, to_arg, "IN_PRQ_MAX"); break; + case ERTS_PSFLG_IN_PRQ_HIGH: + erts_print(to, to_arg, "IN_PRQ_HIGH"); break; + case ERTS_PSFLG_IN_PRQ_NORMAL: + erts_print(to, to_arg, "IN_PRQ_NORMAL"); break; + case ERTS_PSFLG_IN_PRQ_LOW: + erts_print(to, to_arg, "IN_PRQ_LOW"); break; + case ERTS_PSFLG_FREE: + erts_print(to, to_arg, "FREE"); break; + case ERTS_PSFLG_EXITING: + erts_print(to, to_arg, "EXITING"); break; + case ERTS_PSFLG_PENDING_EXIT: + erts_print(to, to_arg, "PENDING_EXIT"); break; + case ERTS_PSFLG_ACTIVE: + erts_print(to, to_arg, "ACTIVE"); break; + case ERTS_PSFLG_IN_RUNQ: + erts_print(to, to_arg, "IN_RUNQ"); break; + case ERTS_PSFLG_RUNNING: + erts_print(to, to_arg, "RUNNING"); break; + case ERTS_PSFLG_SUSPENDED: + erts_print(to, to_arg, "SUSPENDED"); break; + case ERTS_PSFLG_GC: + erts_print(to, to_arg, "GC"); break; + case ERTS_PSFLG_BOUND: + erts_print(to, to_arg, "BOUND"); break; + case ERTS_PSFLG_TRAP_EXIT: + erts_print(to, to_arg, "TRAP_EXIT"); break; + case ERTS_PSFLG_ACTIVE_SYS: + erts_print(to, to_arg, "ACTIVE_SYS"); break; + case ERTS_PSFLG_RUNNING_SYS: + erts_print(to, to_arg, "RUNNING_SYS"); break; + case ERTS_PSFLG_PROXY: + erts_print(to, to_arg, "PROXY"); break; + case ERTS_PSFLG_DELAYED_SYS: + erts_print(to, to_arg, "DELAYED_SYS"); break; +#ifdef ERTS_DIRTY_SCHEDULERS + case ERTS_PSFLG_DIRTY_CPU_PROC: + erts_print(to, to_arg, "DIRTY_CPU_PROC"); break; + case ERTS_PSFLG_DIRTY_IO_PROC: + erts_print(to, to_arg, "DIRTY_IO_PROC"); break; + case ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q: + erts_print(to, to_arg, "DIRTY_CPU_PROC_IN_Q"); break; + case ERTS_PSFLG_DIRTY_IO_PROC_IN_Q: + erts_print(to, to_arg, "DIRTY_IO_PROC_IN_Q"); break; +#endif + default: + erts_print(to, to_arg, "UNKNOWN(%d)", chk); break; + } + if (psflg > chk) + erts_print(to, to_arg, " | "); + psflg -= chk; + } + } + erts_print(to, to_arg, "\n"); +} diff --git a/erts/emulator/beam/erl_ptab.c b/erts/emulator/beam/erl_ptab.c index eabf016081..02943ee683 100644 --- a/erts/emulator/beam/erl_ptab.c +++ b/erts/emulator/beam/erl_ptab.c @@ -280,124 +280,38 @@ struct ErtsPTabListBifData_ { }; -#ifdef ARCH_32 - -static ERTS_INLINE Uint64 -dw_aint_to_uint64(erts_dw_aint_t *dw) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw->dw_sint; -#else - Uint64 res; - res = (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - -static void -unint64_to_dw_aint(erts_dw_aint_t *dw, Uint64 val) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - dw->dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw->sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) (val & 0xffffffff); - dw->sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif -} - static ERTS_INLINE void last_data_init_nob(ErtsPTab *ptab, Uint64 val) { - erts_dw_aint_t dw; - unint64_to_dw_aint(&dw, val); - erts_smp_dw_atomic_init_nob(&ptab->vola.tile.last_data, &dw); + erts_smp_atomic64_init_nob(&ptab->vola.tile.last_data, (erts_aint64_t) val); } static ERTS_INLINE void last_data_set_relb(ErtsPTab *ptab, Uint64 val) { - erts_dw_aint_t dw; - unint64_to_dw_aint(&dw, val); - erts_smp_dw_atomic_set_relb(&ptab->vola.tile.last_data, &dw); + erts_smp_atomic64_set_relb(&ptab->vola.tile.last_data, (erts_aint64_t) val); } static ERTS_INLINE Uint64 last_data_read_nob(ErtsPTab *ptab) { - erts_dw_aint_t dw; - erts_smp_dw_atomic_read_nob(&ptab->vola.tile.last_data, &dw); - return dw_aint_to_uint64(&dw); + return (Uint64) erts_smp_atomic64_read_nob(&ptab->vola.tile.last_data); } static ERTS_INLINE Uint64 last_data_read_acqb(ErtsPTab *ptab) { - erts_dw_aint_t dw; - erts_smp_dw_atomic_read_acqb(&ptab->vola.tile.last_data, &dw); - return dw_aint_to_uint64(&dw); + return (Uint64) erts_smp_atomic64_read_acqb(&ptab->vola.tile.last_data); } static ERTS_INLINE Uint64 last_data_cmpxchg_relb(ErtsPTab *ptab, Uint64 new, Uint64 exp) { - erts_dw_aint_t dw_new, dw_xchg; - - unint64_to_dw_aint(&dw_new, new); - unint64_to_dw_aint(&dw_xchg, exp); - - if (erts_smp_dw_atomic_cmpxchg_relb(&ptab->vola.tile.last_data, - &dw_new, - &dw_xchg)) - return exp; - else - return dw_aint_to_uint64(&dw_xchg); -} - -#elif defined(ARCH_64) - -union { - erts_smp_atomic_t pid_data; - char align[ERTS_CACHE_LINE_SIZE]; -} last erts_align_attribute(ERTS_CACHE_LINE_SIZE); - -static ERTS_INLINE void -last_data_init_nob(ErtsPTab *ptab, Uint64 val) -{ - erts_smp_atomic_init_nob(&ptab->vola.tile.last_data, (erts_aint_t) val); + return (Uint64) erts_smp_atomic64_cmpxchg_relb(&ptab->vola.tile.last_data, + (erts_aint64_t) new, + (erts_aint64_t) exp); } -static ERTS_INLINE void -last_data_set_relb(ErtsPTab *ptab, Uint64 val) -{ - erts_smp_atomic_set_relb(&ptab->vola.tile.last_data, (erts_aint_t) val); -} - -static ERTS_INLINE Uint64 -last_data_read_nob(ErtsPTab *ptab) -{ - return (Uint64) erts_smp_atomic_read_nob(&ptab->vola.tile.last_data); -} - -static ERTS_INLINE Uint64 -last_data_read_acqb(ErtsPTab *ptab) -{ - return (Uint64) erts_smp_atomic_read_acqb(&ptab->vola.tile.last_data); -} - -static ERTS_INLINE Uint64 -last_data_cmpxchg_relb(ErtsPTab *ptab, Uint64 new, Uint64 exp) -{ - return (Uint64) erts_smp_atomic_cmpxchg_relb(&ptab->vola.tile.last_data, - (erts_aint_t) new, - (erts_aint_t) exp); -} - -#else -# error "Not 64-bit, nor 32-bit architecture..." -#endif - static ERTS_INLINE int last_data_cmp(Uint64 ld1, Uint64 ld2) { diff --git a/erts/emulator/beam/erl_ptab.h b/erts/emulator/beam/erl_ptab.h index e3e05f14af..876241159b 100644 --- a/erts/emulator/beam/erl_ptab.h +++ b/erts/emulator/beam/erl_ptab.h @@ -88,11 +88,7 @@ typedef struct { } ErtsPTabListData; typedef struct { -#ifdef ARCH_32 - erts_smp_dw_atomic_t last_data; -#else - erts_smp_atomic_t last_data; -#endif + erts_smp_atomic64_t last_data; erts_smp_atomic32_t count; erts_smp_atomic32_t aid_ix; erts_smp_atomic32_t fid_ix; diff --git a/erts/emulator/beam/erl_smp.h b/erts/emulator/beam/erl_smp.h index c38ef47d87..6c40edeb3e 100644 --- a/erts/emulator/beam/erl_smp.h +++ b/erts/emulator/beam/erl_smp.h @@ -60,6 +60,7 @@ typedef erts_tsd_key_t erts_smp_tsd_key_t; #define erts_smp_dw_atomic_t erts_dw_atomic_t #define erts_smp_atomic_t erts_atomic_t #define erts_smp_atomic32_t erts_atomic32_t +#define erts_smp_atomic64_t erts_atomic64_t typedef erts_spinlock_t erts_smp_spinlock_t; typedef erts_rwlock_t erts_smp_rwlock_t; void erts_thr_fatal_error(int, char *); /* implemented in erl_init.c */ @@ -95,6 +96,7 @@ typedef int erts_smp_tsd_key_t; #define erts_smp_dw_atomic_t erts_no_dw_atomic_t #define erts_smp_atomic_t erts_no_atomic_t #define erts_smp_atomic32_t erts_no_atomic32_t +#define erts_smp_atomic64_t erts_no_atomic64_t #if __GNUC__ > 2 typedef struct { } erts_smp_spinlock_t; typedef struct { } erts_smp_rwlock_t; @@ -489,6 +491,116 @@ ERTS_GLB_INLINE void erts_smp_thr_sigwait(const sigset_t *set, int *sig); #define erts_smp_atomic32_set_dirty erts_atomic32_set_dirty #define erts_smp_atomic32_read_dirty erts_atomic32_read_dirty +/* 64-bit atomics */ + +#define erts_smp_atomic64_init_nob erts_atomic64_init_nob +#define erts_smp_atomic64_set_nob erts_atomic64_set_nob +#define erts_smp_atomic64_read_nob erts_atomic64_read_nob +#define erts_smp_atomic64_inc_read_nob erts_atomic64_inc_read_nob +#define erts_smp_atomic64_dec_read_nob erts_atomic64_dec_read_nob +#define erts_smp_atomic64_inc_nob erts_atomic64_inc_nob +#define erts_smp_atomic64_dec_nob erts_atomic64_dec_nob +#define erts_smp_atomic64_add_read_nob erts_atomic64_add_read_nob +#define erts_smp_atomic64_add_nob erts_atomic64_add_nob +#define erts_smp_atomic64_read_bor_nob erts_atomic64_read_bor_nob +#define erts_smp_atomic64_read_band_nob erts_atomic64_read_band_nob +#define erts_smp_atomic64_xchg_nob erts_atomic64_xchg_nob +#define erts_smp_atomic64_cmpxchg_nob erts_atomic64_cmpxchg_nob +#define erts_smp_atomic64_read_bset_nob erts_atomic64_read_bset_nob + +#define erts_smp_atomic64_init_mb erts_atomic64_init_mb +#define erts_smp_atomic64_set_mb erts_atomic64_set_mb +#define erts_smp_atomic64_read_mb erts_atomic64_read_mb +#define erts_smp_atomic64_inc_read_mb erts_atomic64_inc_read_mb +#define erts_smp_atomic64_dec_read_mb erts_atomic64_dec_read_mb +#define erts_smp_atomic64_inc_mb erts_atomic64_inc_mb +#define erts_smp_atomic64_dec_mb erts_atomic64_dec_mb +#define erts_smp_atomic64_add_read_mb erts_atomic64_add_read_mb +#define erts_smp_atomic64_add_mb erts_atomic64_add_mb +#define erts_smp_atomic64_read_bor_mb erts_atomic64_read_bor_mb +#define erts_smp_atomic64_read_band_mb erts_atomic64_read_band_mb +#define erts_smp_atomic64_xchg_mb erts_atomic64_xchg_mb +#define erts_smp_atomic64_cmpxchg_mb erts_atomic64_cmpxchg_mb +#define erts_smp_atomic64_read_bset_mb erts_atomic64_read_bset_mb + +#define erts_smp_atomic64_init_acqb erts_atomic64_init_acqb +#define erts_smp_atomic64_set_acqb erts_atomic64_set_acqb +#define erts_smp_atomic64_read_acqb erts_atomic64_read_acqb +#define erts_smp_atomic64_inc_read_acqb erts_atomic64_inc_read_acqb +#define erts_smp_atomic64_dec_read_acqb erts_atomic64_dec_read_acqb +#define erts_smp_atomic64_inc_acqb erts_atomic64_inc_acqb +#define erts_smp_atomic64_dec_acqb erts_atomic64_dec_acqb +#define erts_smp_atomic64_add_read_acqb erts_atomic64_add_read_acqb +#define erts_smp_atomic64_add_acqb erts_atomic64_add_acqb +#define erts_smp_atomic64_read_bor_acqb erts_atomic64_read_bor_acqb +#define erts_smp_atomic64_read_band_acqb erts_atomic64_read_band_acqb +#define erts_smp_atomic64_xchg_acqb erts_atomic64_xchg_acqb +#define erts_smp_atomic64_cmpxchg_acqb erts_atomic64_cmpxchg_acqb +#define erts_smp_atomic64_read_bset_acqb erts_atomic64_read_bset_acqb + +#define erts_smp_atomic64_init_relb erts_atomic64_init_relb +#define erts_smp_atomic64_set_relb erts_atomic64_set_relb +#define erts_smp_atomic64_read_relb erts_atomic64_read_relb +#define erts_smp_atomic64_inc_read_relb erts_atomic64_inc_read_relb +#define erts_smp_atomic64_dec_read_relb erts_atomic64_dec_read_relb +#define erts_smp_atomic64_inc_relb erts_atomic64_inc_relb +#define erts_smp_atomic64_dec_relb erts_atomic64_dec_relb +#define erts_smp_atomic64_add_read_relb erts_atomic64_add_read_relb +#define erts_smp_atomic64_add_relb erts_atomic64_add_relb +#define erts_smp_atomic64_read_bor_relb erts_atomic64_read_bor_relb +#define erts_smp_atomic64_read_band_relb erts_atomic64_read_band_relb +#define erts_smp_atomic64_xchg_relb erts_atomic64_xchg_relb +#define erts_smp_atomic64_cmpxchg_relb erts_atomic64_cmpxchg_relb +#define erts_smp_atomic64_read_bset_relb erts_atomic64_read_bset_relb + +#define erts_smp_atomic64_init_ddrb erts_atomic64_init_ddrb +#define erts_smp_atomic64_set_ddrb erts_atomic64_set_ddrb +#define erts_smp_atomic64_read_ddrb erts_atomic64_read_ddrb +#define erts_smp_atomic64_inc_read_ddrb erts_atomic64_inc_read_ddrb +#define erts_smp_atomic64_dec_read_ddrb erts_atomic64_dec_read_ddrb +#define erts_smp_atomic64_inc_ddrb erts_atomic64_inc_ddrb +#define erts_smp_atomic64_dec_ddrb erts_atomic64_dec_ddrb +#define erts_smp_atomic64_add_read_ddrb erts_atomic64_add_read_ddrb +#define erts_smp_atomic64_add_ddrb erts_atomic64_add_ddrb +#define erts_smp_atomic64_read_bor_ddrb erts_atomic64_read_bor_ddrb +#define erts_smp_atomic64_read_band_ddrb erts_atomic64_read_band_ddrb +#define erts_smp_atomic64_xchg_ddrb erts_atomic64_xchg_ddrb +#define erts_smp_atomic64_cmpxchg_ddrb erts_atomic64_cmpxchg_ddrb +#define erts_smp_atomic64_read_bset_ddrb erts_atomic64_read_bset_ddrb + +#define erts_smp_atomic64_init_rb erts_atomic64_init_rb +#define erts_smp_atomic64_set_rb erts_atomic64_set_rb +#define erts_smp_atomic64_read_rb erts_atomic64_read_rb +#define erts_smp_atomic64_inc_read_rb erts_atomic64_inc_read_rb +#define erts_smp_atomic64_dec_read_rb erts_atomic64_dec_read_rb +#define erts_smp_atomic64_inc_rb erts_atomic64_inc_rb +#define erts_smp_atomic64_dec_rb erts_atomic64_dec_rb +#define erts_smp_atomic64_add_read_rb erts_atomic64_add_read_rb +#define erts_smp_atomic64_add_rb erts_atomic64_add_rb +#define erts_smp_atomic64_read_bor_rb erts_atomic64_read_bor_rb +#define erts_smp_atomic64_read_band_rb erts_atomic64_read_band_rb +#define erts_smp_atomic64_xchg_rb erts_atomic64_xchg_rb +#define erts_smp_atomic64_cmpxchg_rb erts_atomic64_cmpxchg_rb +#define erts_smp_atomic64_read_bset_rb erts_atomic64_read_bset_rb + +#define erts_smp_atomic64_init_wb erts_atomic64_init_wb +#define erts_smp_atomic64_set_wb erts_atomic64_set_wb +#define erts_smp_atomic64_read_wb erts_atomic64_read_wb +#define erts_smp_atomic64_inc_read_wb erts_atomic64_inc_read_wb +#define erts_smp_atomic64_dec_read_wb erts_atomic64_dec_read_wb +#define erts_smp_atomic64_inc_wb erts_atomic64_inc_wb +#define erts_smp_atomic64_dec_wb erts_atomic64_dec_wb +#define erts_smp_atomic64_add_read_wb erts_atomic64_add_read_wb +#define erts_smp_atomic64_add_wb erts_atomic64_add_wb +#define erts_smp_atomic64_read_bor_wb erts_atomic64_read_bor_wb +#define erts_smp_atomic64_read_band_wb erts_atomic64_read_band_wb +#define erts_smp_atomic64_xchg_wb erts_atomic64_xchg_wb +#define erts_smp_atomic64_cmpxchg_wb erts_atomic64_cmpxchg_wb +#define erts_smp_atomic64_read_bset_wb erts_atomic64_read_bset_wb + +#define erts_smp_atomic64_set_dirty erts_atomic64_set_dirty +#define erts_smp_atomic64_read_dirty erts_atomic64_read_dirty + #else /* !ERTS_SMP */ /* Double word size atomics */ @@ -751,6 +863,116 @@ ERTS_GLB_INLINE void erts_smp_thr_sigwait(const sigset_t *set, int *sig); #define erts_smp_atomic32_set_dirty erts_no_atomic32_set #define erts_smp_atomic32_read_dirty erts_no_atomic32_read +/* 64-bit atomics */ + +#define erts_smp_atomic64_init_nob erts_no_atomic64_set +#define erts_smp_atomic64_set_nob erts_no_atomic64_set +#define erts_smp_atomic64_read_nob erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_nob erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_nob erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_nob erts_no_atomic64_inc +#define erts_smp_atomic64_dec_nob erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_nob erts_no_atomic64_add_read +#define erts_smp_atomic64_add_nob erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_nob erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_nob erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_nob erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_nob erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_nob erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_mb erts_no_atomic64_set +#define erts_smp_atomic64_set_mb erts_no_atomic64_set +#define erts_smp_atomic64_read_mb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_mb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_mb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_mb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_mb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_mb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_mb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_mb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_mb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_mb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_mb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_mb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_acqb erts_no_atomic64_set +#define erts_smp_atomic64_set_acqb erts_no_atomic64_set +#define erts_smp_atomic64_read_acqb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_acqb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_acqb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_acqb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_acqb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_acqb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_acqb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_acqb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_acqb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_acqb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_acqb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_acqb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_relb erts_no_atomic64_set +#define erts_smp_atomic64_set_relb erts_no_atomic64_set +#define erts_smp_atomic64_read_relb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_relb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_relb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_relb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_relb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_relb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_relb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_relb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_relb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_relb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_relb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_relb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_ddrb erts_no_atomic64_set +#define erts_smp_atomic64_set_ddrb erts_no_atomic64_set +#define erts_smp_atomic64_read_ddrb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_ddrb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_ddrb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_ddrb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_ddrb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_ddrb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_ddrb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_ddrb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_ddrb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_ddrb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_ddrb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_ddrb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_rb erts_no_atomic64_set +#define erts_smp_atomic64_set_rb erts_no_atomic64_set +#define erts_smp_atomic64_read_rb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_rb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_rb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_rb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_rb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_rb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_rb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_rb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_rb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_rb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_rb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_rb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_wb erts_no_atomic64_set +#define erts_smp_atomic64_set_wb erts_no_atomic64_set +#define erts_smp_atomic64_read_wb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_wb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_wb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_wb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_wb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_wb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_wb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_wb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_wb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_wb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_wb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_wb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_set_dirty erts_no_atomic64_set +#define erts_smp_atomic64_read_dirty erts_no_atomic64_read + #endif /* !ERTS_SMP */ #if ERTS_GLB_INLINE_INCL_FUNC_DEF diff --git a/erts/emulator/beam/erl_term.c b/erts/emulator/beam/erl_term.c index 28cbe7004f..bc04d7b78e 100644 --- a/erts/emulator/beam/erl_term.c +++ b/erts/emulator/beam/erl_term.c @@ -86,11 +86,12 @@ unsigned tag_val_def(Wterm x) case (_TAG_HEADER_EXTERNAL_PID >> _TAG_PRIMARY_SIZE): return EXTERNAL_PID_DEF; case (_TAG_HEADER_EXTERNAL_PORT >> _TAG_PRIMARY_SIZE): return EXTERNAL_PORT_DEF; case (_TAG_HEADER_EXTERNAL_REF >> _TAG_PRIMARY_SIZE): return EXTERNAL_REF_DEF; + case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE): return MAP_DEF; case (_TAG_HEADER_REFC_BIN >> _TAG_PRIMARY_SIZE): return BINARY_DEF; case (_TAG_HEADER_HEAP_BIN >> _TAG_PRIMARY_SIZE): return BINARY_DEF; case (_TAG_HEADER_SUB_BIN >> _TAG_PRIMARY_SIZE): return BINARY_DEF; - case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE): return MAP_DEF; } + break; } case TAG_PRIMARY_IMMED1: { diff --git a/erts/emulator/beam/erl_term.h b/erts/emulator/beam/erl_term.h index 37014ccf94..095aa54ddd 100644 --- a/erts/emulator/beam/erl_term.h +++ b/erts/emulator/beam/erl_term.h @@ -147,21 +147,21 @@ struct erl_node_; /* Declared in erl_node_tables.h */ #define MAP_SUBTAG (0xF << _TAG_PRIMARY_SIZE) /* MAP */ -#define _TAG_HEADER_ARITYVAL (TAG_PRIMARY_HEADER|ARITYVAL_SUBTAG) -#define _TAG_HEADER_FUN (TAG_PRIMARY_HEADER|FUN_SUBTAG) -#define _TAG_HEADER_POS_BIG (TAG_PRIMARY_HEADER|POS_BIG_SUBTAG) -#define _TAG_HEADER_NEG_BIG (TAG_PRIMARY_HEADER|NEG_BIG_SUBTAG) -#define _TAG_HEADER_FLOAT (TAG_PRIMARY_HEADER|FLOAT_SUBTAG) -#define _TAG_HEADER_EXPORT (TAG_PRIMARY_HEADER|EXPORT_SUBTAG) -#define _TAG_HEADER_REF (TAG_PRIMARY_HEADER|REF_SUBTAG) -#define _TAG_HEADER_REFC_BIN (TAG_PRIMARY_HEADER|REFC_BINARY_SUBTAG) -#define _TAG_HEADER_HEAP_BIN (TAG_PRIMARY_HEADER|HEAP_BINARY_SUBTAG) -#define _TAG_HEADER_SUB_BIN (TAG_PRIMARY_HEADER|SUB_BINARY_SUBTAG) -#define _TAG_HEADER_EXTERNAL_PID (TAG_PRIMARY_HEADER|EXTERNAL_PID_SUBTAG) -#define _TAG_HEADER_EXTERNAL_PORT (TAG_PRIMARY_HEADER|EXTERNAL_PORT_SUBTAG) -#define _TAG_HEADER_EXTERNAL_REF (TAG_PRIMARY_HEADER|EXTERNAL_REF_SUBTAG) +#define _TAG_HEADER_ARITYVAL (TAG_PRIMARY_HEADER|ARITYVAL_SUBTAG) +#define _TAG_HEADER_FUN (TAG_PRIMARY_HEADER|FUN_SUBTAG) +#define _TAG_HEADER_POS_BIG (TAG_PRIMARY_HEADER|POS_BIG_SUBTAG) +#define _TAG_HEADER_NEG_BIG (TAG_PRIMARY_HEADER|NEG_BIG_SUBTAG) +#define _TAG_HEADER_FLOAT (TAG_PRIMARY_HEADER|FLOAT_SUBTAG) +#define _TAG_HEADER_EXPORT (TAG_PRIMARY_HEADER|EXPORT_SUBTAG) +#define _TAG_HEADER_REF (TAG_PRIMARY_HEADER|REF_SUBTAG) +#define _TAG_HEADER_REFC_BIN (TAG_PRIMARY_HEADER|REFC_BINARY_SUBTAG) +#define _TAG_HEADER_HEAP_BIN (TAG_PRIMARY_HEADER|HEAP_BINARY_SUBTAG) +#define _TAG_HEADER_SUB_BIN (TAG_PRIMARY_HEADER|SUB_BINARY_SUBTAG) +#define _TAG_HEADER_EXTERNAL_PID (TAG_PRIMARY_HEADER|EXTERNAL_PID_SUBTAG) +#define _TAG_HEADER_EXTERNAL_PORT (TAG_PRIMARY_HEADER|EXTERNAL_PORT_SUBTAG) +#define _TAG_HEADER_EXTERNAL_REF (TAG_PRIMARY_HEADER|EXTERNAL_REF_SUBTAG) #define _TAG_HEADER_BIN_MATCHSTATE (TAG_PRIMARY_HEADER|BIN_MATCHSTATE_SUBTAG) -#define _TAG_HEADER_MAP (TAG_PRIMARY_HEADER|MAP_SUBTAG) +#define _TAG_HEADER_MAP (TAG_PRIMARY_HEADER|MAP_SUBTAG) #define _TAG_HEADER_MASK 0x3F @@ -296,9 +296,11 @@ _ET_DECLARE_CHECKED(Uint,atom_val,Eterm) #define atom_val(x) _ET_APPLY(atom_val,(x)) /* header (arityval or thing) access methods */ -#define _make_header(sz,tag) ((Uint)(((sz) << _HEADER_ARITY_OFFS) + (tag))) +#define _make_header(sz,tag) ((Uint)(((Uint)(sz) << _HEADER_ARITY_OFFS) + (tag))) #define is_header(x) (((x) & _TAG_PRIMARY_MASK) == TAG_PRIMARY_HEADER) -#define _unchecked_header_arity(x) ((x) >> _HEADER_ARITY_OFFS) +//#define _unchecked_header_arity(x) ((x) >> _HEADER_ARITY_OFFS) +#define _unchecked_header_arity(x) \ + (is_map_header(x) ? MAP_HEADER_ARITY(x) : ((x) >> _HEADER_ARITY_OFFS)) _ET_DECLARE_CHECKED(Uint,header_arity,Eterm) #define header_arity(x) _ET_APPLY(header_arity,(x)) @@ -361,6 +363,7 @@ _ET_DECLARE_CHECKED(Uint,thing_subtag,Eterm) ((((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_REFC_BIN) || \ (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_HEAP_BIN) || \ (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_SUB_BIN)) + #define make_binary(x) make_boxed((Eterm*)(x)) #define is_binary(x) (is_boxed((x)) && is_binary_header(*boxed_val((x)))) #define is_not_binary(x) (!is_binary((x))) @@ -990,6 +993,43 @@ _ET_DECLARE_CHECKED(Uint32*,external_ref_data,Wterm) _ET_DECLARE_CHECKED(struct erl_node_*,external_ref_node,Eterm) #define external_ref_node(x) _ET_APPLY(external_ref_node,(x)) +/* maps */ + +#define MAP_HEADER_TAG_SZ (2) +#define MAP_HEADER_ARITY_SZ (8) +#define MAP_HEADER_VAL_SZ (16) + +#define MAP_HEADER_TAG_FLATMAP_HEAD (0x0) +#define MAP_HEADER_TAG_HAMT_NODE_BITMAP (0x1) +#define MAP_HEADER_TAG_HAMT_HEAD_ARRAY (0x2) +#define MAP_HEADER_TAG_HAMT_HEAD_BITMAP (0x3) + +#define MAP_HEADER_TYPE(Hdr) (((Hdr) >> (_HEADER_ARITY_OFFS)) & (0x3)) +#define MAP_HEADER_ARITY(Hdr) (((Hdr) >> (_HEADER_ARITY_OFFS + MAP_HEADER_TAG_SZ)) & (0xff)) +#define MAP_HEADER_VAL(Hdr) (((Hdr) >> (_HEADER_ARITY_OFFS + MAP_HEADER_TAG_SZ + MAP_HEADER_ARITY_SZ)) & (0xffff)) + +#define make_hashmap(x) make_boxed((Eterm*)(x)) +#define make_hashmap_rel make_boxed_rel +#define is_hashmap(x) (is_boxed((x)) && is_hashmap_header(*boxed_val((x)))) +#define is_not_hashmap(x) (!is_hashmap(x)) +#define is_hashmap_rel(RTERM,BASE) is_hashmap(rterm2wterm(RTERM,BASE)) +#define is_hashmap_header(x) (((x) & (_HEADER_MAP_HASHMAP_HEAD_MASK)) == HAMT_SUBTAG_HEAD_ARRAY) +#define hashmap_val(x) _unchecked_boxed_val((x)) +#define hashmap_val_rel(RTERM, BASE) hashmap_val(rterm2wterm(RTERM, BASE)) + +#define make_flatmap(x) make_boxed((Eterm*)(x)) +#define make_flatmap_rel(x, BASE) make_boxed_rel((Eterm*)(x),(BASE)) +#define is_flatmap(x) (is_boxed((x)) && is_flatmap_header(*boxed_val((x)))) +#define is_flatmap_rel(RTERM,BASE) is_flatmap(rterm2wterm(RTERM,BASE)) +#define is_not_flatmap(x) (!is_flatmap((x))) +#define is_flatmap_header(x) (((x) & (_HEADER_MAP_SUBTAG_MASK)) == HAMT_SUBTAG_HEAD_FLATMAP) +#define flatmap_val(x) (_unchecked_boxed_val((x))) +#define flatmap_val_rel(RTERM, BASE) flatmap_val(rterm2wterm(RTERM, BASE)) + +#define is_map_header(x) (((x) & (_TAG_HEADER_MASK)) == _TAG_HEADER_MAP) +#define is_map(x) (is_boxed((x)) && is_map_header(*boxed_val(x))) +#define is_map_rel(RTERM,BASE) is_map(rterm2wterm(RTERM,BASE)) + /* number tests */ #define is_integer(x) (is_small(x) || is_big(x)) @@ -1096,6 +1136,8 @@ _ET_DECLARE_CHECKED(Uint,y_reg_index,Uint) #define BIG_DEF 0xf #define SMALL_DEF 0x10 +#define FIRST_VACANT_TAG_DEF 0x11 + #if ET_DEBUG extern unsigned tag_val_def_debug(Wterm, const char*, unsigned); #define tag_val_def(x) tag_val_def_debug((x),__FILE__,__LINE__) diff --git a/erts/emulator/beam/erl_thr_progress.c b/erts/emulator/beam/erl_thr_progress.c index 545a0343d0..4c9b00d2ee 100644 --- a/erts/emulator/beam/erl_thr_progress.c +++ b/erts/emulator/beam/erl_thr_progress.c @@ -115,70 +115,24 @@ #undef read_nob #define read_nob erts_thr_prgr_read_nob__ -#ifdef ARCH_64 - static ERTS_INLINE void set_mb(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_set_mb(atmc, val); + erts_atomic64_set_mb(atmc, (erts_aint64_t) val); } static ERTS_INLINE void set_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_set_nob(atmc, val); + erts_atomic64_set_nob(atmc, (erts_aint64_t) val); } static ERTS_INLINE void init_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_init_nob(atmc, val); -} - -#else - -#undef dw_aint_to_val -#define dw_aint_to_val erts_thr_prgr_dw_aint_to_val__ - -static void -val_to_dw_aint(erts_dw_aint_t *dw_aint, ErtsThrPrgrVal val) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - dw_aint->dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw_aint->sint[ERTS_DW_AINT_LOW_WORD] - = (erts_aint_t) (val & 0xffffffff); - dw_aint->sint[ERTS_DW_AINT_HIGH_WORD] - = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif -} - -static ERTS_INLINE void -set_mb(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_set_mb(atmc, &dw_aint); -} - -static ERTS_INLINE void -set_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_set_nob(atmc, &dw_aint); + erts_atomic64_init_nob(atmc, (erts_aint64_t) val); } -static ERTS_INLINE void -init_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_init_nob(atmc, &dw_aint); -} - -#endif - /* #define ERTS_THR_PROGRESS_STATE_DEBUG */ #ifdef ERTS_THR_PROGRESS_STATE_DEBUG @@ -1381,25 +1335,10 @@ erts_thr_progress_block(void) thr_progress_block(tmp_thr_prgr_data(NULL), 1); } -void -erts_thr_progress_fatal_error_block(SWord timeout, - ErtsThrPrgrData *tmp_tpd_bufp) +int +erts_thr_progress_fatal_error_block(ErtsThrPrgrData *tmp_tpd_bufp) { ErtsThrPrgrData *tpd = perhaps_thr_prgr_data(NULL); - erts_aint32_t bc; - SWord time_left = timeout; - SysTimeval to; - - /* - * Counting poll intervals may give us a too long timeout - * if cpu is busy. If we got tolerant time of day we use it - * to prevent this. - */ - if (!erts_disable_tolerant_timeofday) { - erts_get_timeval(&to); - to.tv_sec += timeout / 1000; - to.tv_sec += timeout % 1000; - } if (!tpd) { /* @@ -1412,9 +1351,24 @@ erts_thr_progress_fatal_error_block(SWord timeout, init_tmp_thr_prgr_data(tpd); } - bc = thr_progress_block(tpd, 0); - if (bc == 0) - return; /* Succefully blocked all managed threads */ + /* Returns number of threads that have not yes been blocked */ + return thr_progress_block(tpd, 0); +} + +void +erts_thr_progress_fatal_error_wait(SWord timeout) { + erts_aint32_t bc; + SWord time_left = timeout; + ErtsMonotonicTime timeout_time; + + /* + * Counting poll intervals may give us a too long timeout + * if cpu is busy. We use timeout time to try to prevent + * this. In case we havn't got time correction this may + * however fail too... + */ + timeout_time = erts_get_monotonic_time(); + timeout_time += ERTS_MSEC_TO_MONOTONIC((ErtsMonotonicTime) timeout); while (1) { if (erts_milli_sleep(ERTS_THR_PRGR_FTL_ERR_BLCK_POLL_INTERVAL) == 0) @@ -1424,14 +1378,8 @@ erts_thr_progress_fatal_error_block(SWord timeout, break; /* Succefully blocked all managed threads */ if (time_left <= 0) break; /* Timeout */ - if (!erts_disable_tolerant_timeofday) { - SysTimeval now; - erts_get_timeval(&now); - if (now.tv_sec > to.tv_sec) - break; /* Timeout */ - if (now.tv_sec == to.tv_sec && now.tv_usec >= to.tv_usec) - break; /* Timeout */ - } + if (timeout_time <= erts_get_monotonic_time()) + break; /* Timeout */ } } diff --git a/erts/emulator/beam/erl_thr_progress.h b/erts/emulator/beam/erl_thr_progress.h index 5f392944c2..cf11c4e114 100644 --- a/erts/emulator/beam/erl_thr_progress.h +++ b/erts/emulator/beam/erl_thr_progress.h @@ -83,8 +83,8 @@ typedef struct { ErtsThrPrgrLeaderState leader_state; } ErtsThrPrgrData; -void erts_thr_progress_fatal_error_block(SWord timeout, - ErtsThrPrgrData *tmp_tpd_bufp); +int erts_thr_progress_fatal_error_block(ErtsThrPrgrData *tmp_tpd_bufp); +void erts_thr_progress_fatal_error_wait(SWord timeout); #endif /* ERTS_SMP */ @@ -115,11 +115,7 @@ struct ErtsThrPrgrLaterOp_ { extern erts_tsd_key_t erts_thr_prgr_data_key__; -#ifdef ARCH_64 -# define ERTS_THR_PRGR_ATOMIC erts_atomic_t -#else /* ARCH_32 */ -# define ERTS_THR_PRGR_ATOMIC erts_dw_atomic_t -#endif +#define ERTS_THR_PRGR_ATOMIC erts_atomic64_t typedef struct { void *arg; @@ -158,10 +154,6 @@ void erts_thr_progress_unmanaged_continue__(int umrefc_ix); void erts_thr_progress_dbg_print_state(void); -#ifdef ARCH_32 -#define ERTS_THR_PRGR_ATOMIC erts_dw_atomic_t -ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_dw_aint_to_val__(erts_dw_aint_t *dw_aint); -#endif ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc); ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc); ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc); @@ -184,68 +176,24 @@ ERTS_GLB_INLINE int erts_thr_progress_has_reached(ErtsThrPrgrVal val); #if ERTS_GLB_INLINE_INCL_FUNC_DEF -#ifdef ARCH_64 - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_nob(atmc); -} - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_acqb(atmc); -} - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_mb(atmc); -} - -#else /* ARCH_32 */ - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_dw_aint_to_val__(erts_dw_aint_t *dw_aint) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (ErtsThrPrgrVal) dw_aint->dw_sint; -#else - ErtsThrPrgrVal res; - res = (ErtsThrPrgrVal) ((Uint32) dw_aint->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (ErtsThrPrgrVal) ((Uint32) dw_aint->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_nob(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_nob(atmc); } ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_acqb(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_acqb(atmc); } ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_mb(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_mb(atmc); } -#endif - ERTS_GLB_INLINE int erts_thr_progress_is_managed_thread(void) { diff --git a/erts/emulator/beam/erl_threads.h b/erts/emulator/beam/erl_threads.h index 80026104db..dc20ac207f 100644 --- a/erts/emulator/beam/erl_threads.h +++ b/erts/emulator/beam/erl_threads.h @@ -344,6 +344,16 @@ typedef ethr_ts_event erts_tse_t; #define erts_aint32_t ethr_sint32_t #define erts_atomic32_t ethr_atomic32_t +#if defined(ARCH_32) +# define erts_atomic64_t ethr_dw_atomic_t +# define erts_aint64_t ethr_sint64_t +#elif defined(ARCH_64) +# define erts_atomic64_t ethr_atomic_t +# define erts_aint64_t ethr_sint_t +#else +# error "Not supported architecture" +#endif + #define ERTS_DW_AINT_HIGH_WORD ETHR_DW_SINT_HIGH_WORD #define ERTS_DW_AINT_LOW_WORD ETHR_DW_SINT_LOW_WORD @@ -414,10 +424,12 @@ typedef int erts_tse_t; typedef struct { SWord sint[2]; } erts_dw_aint_t; typedef SWord erts_aint_t; typedef Sint32 erts_aint32_t; +typedef Sint64 erts_aint64_t; #define erts_dw_atomic_t erts_dw_aint_t #define erts_atomic_t erts_aint_t #define erts_atomic32_t erts_aint32_t +#define erts_atomic64_t erts_aint64_t #if __GNUC__ > 2 typedef struct { } erts_spinlock_t; @@ -446,6 +458,7 @@ typedef struct { int gcc_is_buggy; } erts_rwlock_t; #define erts_no_dw_atomic_t erts_dw_aint_t #define erts_no_atomic_t erts_aint_t #define erts_no_atomic32_t erts_aint32_t +#define erts_no_atomic64_t erts_aint64_t #define ERTS_AINT_NULL ((erts_aint_t) NULL) @@ -463,6 +476,7 @@ ERTS_GLB_INLINE void erts_thr_detach(erts_tid_t tid); ERTS_GLB_INLINE void erts_thr_exit(void *res); ERTS_GLB_INLINE void erts_thr_install_exit_handler(void (*exit_handler)(void)); ERTS_GLB_INLINE erts_tid_t erts_thr_self(void); +ERTS_GLB_INLINE int erts_thr_getname(erts_tid_t tid, char *buf, size_t len); ERTS_GLB_INLINE int erts_equal_tids(erts_tid_t x, erts_tid_t y); ERTS_GLB_INLINE void erts_mtx_init_x(erts_mtx_t *mtx, char *name, Eterm extra, int enable_lcnt); @@ -570,6 +584,29 @@ ERTS_GLB_INLINE erts_aint32_t erts_no_atomic32_cmpxchg(erts_no_atomic32_t *xchgp ERTS_GLB_INLINE erts_aint32_t erts_no_atomic32_read_bset(erts_no_atomic32_t *var, erts_aint32_t mask, erts_aint32_t set); +ERTS_GLB_INLINE void erts_no_atomic64_set(erts_no_atomic64_t *var, + erts_aint64_t i); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read(erts_no_atomic64_t *var); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_inc_read(erts_no_atomic64_t *incp); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_dec_read(erts_no_atomic64_t *decp); +ERTS_GLB_INLINE void erts_no_atomic64_inc(erts_no_atomic64_t *incp); +ERTS_GLB_INLINE void erts_no_atomic64_dec(erts_no_atomic64_t *decp); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_add_read(erts_no_atomic64_t *addp, + erts_aint64_t i); +ERTS_GLB_INLINE void erts_no_atomic64_add(erts_no_atomic64_t *addp, + erts_aint64_t i); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_bor(erts_no_atomic64_t *var, + erts_aint64_t mask); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_band(erts_no_atomic64_t *var, + erts_aint64_t mask); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_xchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_cmpxchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new, + erts_aint64_t expected); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_bset(erts_no_atomic64_t *var, + erts_aint64_t mask, + erts_aint64_t set); ERTS_GLB_INLINE void erts_spinlock_init_x_opt(erts_spinlock_t *lock, char *name, @@ -615,16 +652,24 @@ ERTS_GLB_INLINE void erts_tse_set(erts_tse_t *ep); ERTS_GLB_INLINE void erts_tse_reset(erts_tse_t *ep); ERTS_GLB_INLINE int erts_tse_wait(erts_tse_t *ep); ERTS_GLB_INLINE int erts_tse_swait(erts_tse_t *ep, int spincount); +ERTS_GLB_INLINE int erts_tse_twait(erts_tse_t *ep, Sint64 tmo); +ERTS_GLB_INLINE int erts_tse_stwait(erts_tse_t *ep, int spincount, Sint64 tmo); ERTS_GLB_INLINE int erts_tse_is_tmp(erts_tse_t *ep); ERTS_GLB_INLINE void erts_thr_set_main_status(int, int); ERTS_GLB_INLINE int erts_thr_get_main_status(void); ERTS_GLB_INLINE void erts_thr_yield(void); + #ifdef ETHR_HAVE_ETHR_SIG_FUNCS #define ERTS_THR_HAVE_SIG_FUNCS 1 ERTS_GLB_INLINE void erts_thr_sigmask(int how, const sigset_t *set, sigset_t *oset); ERTS_GLB_INLINE void erts_thr_sigwait(const sigset_t *set, int *sig); + +#ifdef USE_THREADS +ERTS_GLB_INLINE void erts_thr_kill(erts_tid_t tid, int sig); +#endif + #endif /* #ifdef HAVE_ETHR_SIG_FUNCS */ #ifdef USE_THREADS @@ -1200,6 +1245,441 @@ erts_atomic32_read_dirty(erts_atomic32_t *var) #endif +/* 64-bit atomics */ + +#if defined(ARCH_64) + +#define erts_atomic64_init_nob ethr_atomic_init +#define erts_atomic64_set_nob ethr_atomic_set +#define erts_atomic64_read_nob ethr_atomic_read +#define erts_atomic64_inc_read_nob ethr_atomic_inc_read +#define erts_atomic64_dec_read_nob ethr_atomic_dec_read +#define erts_atomic64_inc_nob ethr_atomic_inc +#define erts_atomic64_dec_nob ethr_atomic_dec +#define erts_atomic64_add_read_nob ethr_atomic_add_read +#define erts_atomic64_add_nob ethr_atomic_add +#define erts_atomic64_read_bor_nob ethr_atomic_read_bor +#define erts_atomic64_read_band_nob ethr_atomic_read_band +#define erts_atomic64_xchg_nob ethr_atomic_xchg +#define erts_atomic64_cmpxchg_nob ethr_atomic_cmpxchg +#define erts_atomic64_read_bset_nob erts_atomic_read_bset_nob + +#define erts_atomic64_init_mb ethr_atomic_init_mb +#define erts_atomic64_set_mb ethr_atomic_set_mb +#define erts_atomic64_read_mb ethr_atomic_read_mb +#define erts_atomic64_inc_read_mb ethr_atomic_inc_read_mb +#define erts_atomic64_dec_read_mb ethr_atomic_dec_read_mb +#define erts_atomic64_inc_mb ethr_atomic_inc_mb +#define erts_atomic64_dec_mb ethr_atomic_dec_mb +#define erts_atomic64_add_read_mb ethr_atomic_add_read_mb +#define erts_atomic64_add_mb ethr_atomic_add_mb +#define erts_atomic64_read_bor_mb ethr_atomic_read_bor_mb +#define erts_atomic64_read_band_mb ethr_atomic_read_band_mb +#define erts_atomic64_xchg_mb ethr_atomic_xchg_mb +#define erts_atomic64_cmpxchg_mb ethr_atomic_cmpxchg_mb +#define erts_atomic64_read_bset_mb erts_atomic_read_bset_mb + +#define erts_atomic64_init_acqb ethr_atomic_init_acqb +#define erts_atomic64_set_acqb ethr_atomic_set_acqb +#define erts_atomic64_read_acqb ethr_atomic_read_acqb +#define erts_atomic64_inc_read_acqb ethr_atomic_inc_read_acqb +#define erts_atomic64_dec_read_acqb ethr_atomic_dec_read_acqb +#define erts_atomic64_inc_acqb ethr_atomic_inc_acqb +#define erts_atomic64_dec_acqb ethr_atomic_dec_acqb +#define erts_atomic64_add_read_acqb ethr_atomic_add_read_acqb +#define erts_atomic64_add_acqb ethr_atomic_add_acqb +#define erts_atomic64_read_bor_acqb ethr_atomic_read_bor_acqb +#define erts_atomic64_read_band_acqb ethr_atomic_read_band_acqb +#define erts_atomic64_xchg_acqb ethr_atomic_xchg_acqb +#define erts_atomic64_cmpxchg_acqb ethr_atomic_cmpxchg_acqb +#define erts_atomic64_read_bset_acqb erts_atomic_read_bset_acqb + +#define erts_atomic64_init_relb ethr_atomic_init_relb +#define erts_atomic64_set_relb ethr_atomic_set_relb +#define erts_atomic64_read_relb ethr_atomic_read_relb +#define erts_atomic64_inc_read_relb ethr_atomic_inc_read_relb +#define erts_atomic64_dec_read_relb ethr_atomic_dec_read_relb +#define erts_atomic64_inc_relb ethr_atomic_inc_relb +#define erts_atomic64_dec_relb ethr_atomic_dec_relb +#define erts_atomic64_add_read_relb ethr_atomic_add_read_relb +#define erts_atomic64_add_relb ethr_atomic_add_relb +#define erts_atomic64_read_bor_relb ethr_atomic_read_bor_relb +#define erts_atomic64_read_band_relb ethr_atomic_read_band_relb +#define erts_atomic64_xchg_relb ethr_atomic_xchg_relb +#define erts_atomic64_cmpxchg_relb ethr_atomic_cmpxchg_relb +#define erts_atomic64_read_bset_relb erts_atomic_read_bset_relb + +#define erts_atomic64_init_ddrb ethr_atomic_init_ddrb +#define erts_atomic64_set_ddrb ethr_atomic_set_ddrb +#define erts_atomic64_read_ddrb ethr_atomic_read_ddrb +#define erts_atomic64_inc_read_ddrb ethr_atomic_inc_read_ddrb +#define erts_atomic64_dec_read_ddrb ethr_atomic_dec_read_ddrb +#define erts_atomic64_inc_ddrb ethr_atomic_inc_ddrb +#define erts_atomic64_dec_ddrb ethr_atomic_dec_ddrb +#define erts_atomic64_add_read_ddrb ethr_atomic_add_read_ddrb +#define erts_atomic64_add_ddrb ethr_atomic_add_ddrb +#define erts_atomic64_read_bor_ddrb ethr_atomic_read_bor_ddrb +#define erts_atomic64_read_band_ddrb ethr_atomic_read_band_ddrb +#define erts_atomic64_xchg_ddrb ethr_atomic_xchg_ddrb +#define erts_atomic64_cmpxchg_ddrb ethr_atomic_cmpxchg_ddrb +#define erts_atomic64_read_bset_ddrb erts_atomic_read_bset_ddrb + +#define erts_atomic64_init_rb ethr_atomic_init_rb +#define erts_atomic64_set_rb ethr_atomic_set_rb +#define erts_atomic64_read_rb ethr_atomic_read_rb +#define erts_atomic64_inc_read_rb ethr_atomic_inc_read_rb +#define erts_atomic64_dec_read_rb ethr_atomic_dec_read_rb +#define erts_atomic64_inc_rb ethr_atomic_inc_rb +#define erts_atomic64_dec_rb ethr_atomic_dec_rb +#define erts_atomic64_add_read_rb ethr_atomic_add_read_rb +#define erts_atomic64_add_rb ethr_atomic_add_rb +#define erts_atomic64_read_bor_rb ethr_atomic_read_bor_rb +#define erts_atomic64_read_band_rb ethr_atomic_read_band_rb +#define erts_atomic64_xchg_rb ethr_atomic_xchg_rb +#define erts_atomic64_cmpxchg_rb ethr_atomic_cmpxchg_rb +#define erts_atomic64_read_bset_rb erts_atomic_read_bset_rb + +#define erts_atomic64_init_wb ethr_atomic_init_wb +#define erts_atomic64_set_wb ethr_atomic_set_wb +#define erts_atomic64_read_wb ethr_atomic_read_wb +#define erts_atomic64_inc_read_wb ethr_atomic_inc_read_wb +#define erts_atomic64_dec_read_wb ethr_atomic_dec_read_wb +#define erts_atomic64_inc_wb ethr_atomic_inc_wb +#define erts_atomic64_dec_wb ethr_atomic_dec_wb +#define erts_atomic64_add_read_wb ethr_atomic_add_read_wb +#define erts_atomic64_add_wb ethr_atomic_add_wb +#define erts_atomic64_read_bor_wb ethr_atomic_read_bor_wb +#define erts_atomic64_read_band_wb ethr_atomic_read_band_wb +#define erts_atomic64_xchg_wb ethr_atomic_xchg_wb +#define erts_atomic64_cmpxchg_wb ethr_atomic_cmpxchg_wb +#define erts_atomic64_read_bset_wb erts_atomic_read_bset_wb + +#define erts_atomic64_set_dirty erts_atomic_set_dirty +#define erts_atomic64_read_dirty erts_atomic_read_dirty + +#elif defined(ARCH_32) + +#undef ERTS_ATOMIC64_OPS_DECL__ + +#define ERTS_ATOMIC64_OPS_DECL__(BARRIER) \ +ERTS_GLB_INLINE void \ +erts_atomic64_init_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE void \ +erts_atomic64_set_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_inc_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_dec_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE void \ +erts_atomic64_inc_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE void \ +erts_atomic64_dec_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_add_read_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE void \ +erts_atomic64_add_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bor_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_band_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_xchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_cmpxchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t new, \ + erts_aint64_t exp); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bset_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t mask, \ + erts_aint64_t set) + +ERTS_ATOMIC64_OPS_DECL__(nob); +ERTS_ATOMIC64_OPS_DECL__(mb); +ERTS_ATOMIC64_OPS_DECL__(acqb); +ERTS_ATOMIC64_OPS_DECL__(relb); +ERTS_ATOMIC64_OPS_DECL__(ddrb); +ERTS_ATOMIC64_OPS_DECL__(rb); +ERTS_ATOMIC64_OPS_DECL__(wb); + +#undef ERTS_ATOMIC64_OPS_DECL__ + +ERTS_GLB_INLINE void +erts_atomic64_set_dirty(erts_atomic64_t *var, erts_aint64_t val); +ERTS_GLB_INLINE erts_aint64_t +erts_atomic64_read_dirty(erts_atomic64_t *var); + +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +/* + * The ethr_dw_atomic_*_nob() functions below + * are here to make it possible for the + * ERTS_ATOMIC64_OPS_IMPL__() to map erts + * barriers to ethread barriers... + */ +static ERTS_INLINE void +ethr_dw_atomic_init_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_init(var, val); +} + +static ERTS_INLINE void +ethr_dw_atomic_set_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_set(var, val); +} + +static ERTS_INLINE void +ethr_dw_atomic_read_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_read(var, val); +} + +static ERTS_INLINE int +ethr_dw_atomic_cmpxchg_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *new, + ethr_dw_sint_t *xchg) +{ + return ethr_dw_atomic_cmpxchg(var, new, xchg); +} + +#undef ERTS_ATOMIC64_OPS_IMPL__ +#undef ERTS_ATOMIC64_DW_CMPXCHG_IMPL__ +#undef ERTS_DW_SINT_TO_AINT64__ +#undef ERTS_AINT64_TO_DW_SINT__ + +#ifdef ETHR_SU_DW_NAINT_T__ +#define ERTS_DW_SINT_TO_AINT64__(DW) \ + ((erts_aint64_t) DW.dw_sint) +#define ERTS_AINT64_TO_DW_SINT__(DW, AINT64) \ + (DW.dw_sint = (ETHR_SU_DW_NAINT_T__) AINT64) +#else /* !ETHR_SU_DW_NAINT_T__ */ +#define ERTS_DW_SINT_TO_AINT64__(DW) \ + ((((erts_aint64_t) DW.sint[ETHR_DW_SINT_HIGH_WORD]) << 32) \ + | (((erts_aint64_t) DW.sint[ETHR_DW_SINT_LOW_WORD]) \ + & ((erts_aint64_t) 0xffffffff))) +#define ERTS_AINT64_TO_DW_SINT__(DW, AINT64) \ + do { \ + DW.sint[ETHR_DW_SINT_LOW_WORD] = \ + (ethr_sint_t) (AINT64 & 0xffffffff); \ + DW.sint[ETHR_DW_SINT_HIGH_WORD] = \ + (ethr_sint_t) ((AINT64 >> 32) & 0xffffffff); \ + } while (0) +#endif /* !ETHR_SU_DW_NAINT_T__ */ + +#define ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(CmpXchgOp, \ + AVarP, XchgVar, NewVar, \ + ModificationCode) \ +do { \ + ethr_dw_sint_t dw_xchg__, dw_new__; \ + ethr_dw_atomic_read(AVarP, &dw_xchg__); \ + do { \ + XchgVar = ERTS_DW_SINT_TO_AINT64__(dw_xchg__); \ + { \ + ModificationCode; \ + } \ + ERTS_AINT64_TO_DW_SINT__(dw_new__, NewVar); \ + } while (!CmpXchgOp((AVarP), &dw_new__, &dw_xchg__)); \ +} while (0) + +#define ERTS_ATOMIC64_OPS_IMPL__(BARRIER) \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_init_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + ethr_dw_sint_t dw; \ + ERTS_AINT64_TO_DW_SINT__(dw, val); \ + ethr_dw_atomic_init_ ## BARRIER(var, &dw); \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_set_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + ethr_dw_sint_t dw; \ + ERTS_AINT64_TO_DW_SINT__(dw, val); \ + ethr_dw_atomic_set_ ## BARRIER(var, &dw); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + ethr_dw_sint_t dw; \ + ethr_dw_atomic_read_ ## BARRIER(var, &dw); \ + return ERTS_DW_SINT_TO_AINT64__(dw); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_inc_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + 1); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_dec_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg - 1); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_inc_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + 1); \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_dec_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg - 1); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_add_read_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + val); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_add_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + val); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bor_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg | val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_band_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg & val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_xchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_cmpxchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t new, \ + erts_aint64_t exp) \ +{ \ + ethr_dw_sint_t dw_xchg, dw_new; \ + ERTS_AINT64_TO_DW_SINT__(dw_xchg, exp); \ + ERTS_AINT64_TO_DW_SINT__(dw_new, new); \ + if (ethr_dw_atomic_cmpxchg_ ## BARRIER(var, &dw_new, &dw_xchg)) \ + return exp; \ + return ERTS_DW_SINT_TO_AINT64__(dw_xchg); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bset_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t mask, \ + erts_aint64_t set) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + { \ + new = xchg & ~mask; \ + new |= mask & set; \ + }); \ + return xchg; \ +} + +ERTS_ATOMIC64_OPS_IMPL__(nob) +ERTS_ATOMIC64_OPS_IMPL__(mb) +ERTS_ATOMIC64_OPS_IMPL__(acqb) +ERTS_ATOMIC64_OPS_IMPL__(relb) +ERTS_ATOMIC64_OPS_IMPL__(ddrb) +ERTS_ATOMIC64_OPS_IMPL__(rb) +ERTS_ATOMIC64_OPS_IMPL__(wb) + +#undef ERTS_ATOMIC64_OPS_IMPL__ +#undef ERTS_ATOMIC64_DW_CMPXCHG_IMPL__ + +ERTS_GLB_INLINE void +erts_atomic64_set_dirty(erts_atomic64_t *var, erts_aint64_t val) +{ + ethr_sint_t *sint = ethr_dw_atomic_addr(var); + ethr_dw_sint_t dw; + ERTS_AINT64_TO_DW_SINT__(dw, val); + sint[0] = dw.sint[0]; + sint[1] = dw.sint[1]; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_atomic64_read_dirty(erts_atomic64_t *var) +{ + ethr_sint_t *sint; + ethr_dw_sint_t dw; + sint = ethr_dw_atomic_addr(var); + dw.sint[0] = sint[0]; + dw.sint[1] = sint[1]; + return ERTS_DW_SINT_TO_AINT64__(dw); +} + +#undef ERTS_DW_SINT_TO_AINT64__ +#undef ERTS_AINT64_TO_DW_SINT__ + +#endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */ + +#endif /* ARCH_32 */ + #else /* !USE_THREADS */ /* Double word size atomics */ @@ -1462,6 +1942,116 @@ erts_atomic32_read_dirty(erts_atomic32_t *var) #define erts_atomic32_set_dirty erts_no_atomic32_set #define erts_atomic32_read_dirty erts_no_atomic32_read +/* 64-bit atomics */ + +#define erts_atomic64_init_nob erts_no_atomic64_set +#define erts_atomic64_set_nob erts_no_atomic64_set +#define erts_atomic64_read_nob erts_no_atomic64_read +#define erts_atomic64_inc_read_nob erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_nob erts_no_atomic64_dec_read +#define erts_atomic64_inc_nob erts_no_atomic64_inc +#define erts_atomic64_dec_nob erts_no_atomic64_dec +#define erts_atomic64_add_read_nob erts_no_atomic64_add_read +#define erts_atomic64_add_nob erts_no_atomic64_add +#define erts_atomic64_read_bor_nob erts_no_atomic64_read_bor +#define erts_atomic64_read_band_nob erts_no_atomic64_read_band +#define erts_atomic64_xchg_nob erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_nob erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_nob erts_no_atomic64_read_bset + +#define erts_atomic64_init_mb erts_no_atomic64_set +#define erts_atomic64_set_mb erts_no_atomic64_set +#define erts_atomic64_read_mb erts_no_atomic64_read +#define erts_atomic64_inc_read_mb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_mb erts_no_atomic64_dec_read +#define erts_atomic64_inc_mb erts_no_atomic64_inc +#define erts_atomic64_dec_mb erts_no_atomic64_dec +#define erts_atomic64_add_read_mb erts_no_atomic64_add_read +#define erts_atomic64_add_mb erts_no_atomic64_add +#define erts_atomic64_read_bor_mb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_mb erts_no_atomic64_read_band +#define erts_atomic64_xchg_mb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_mb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_mb erts_no_atomic64_read_bset + +#define erts_atomic64_init_acqb erts_no_atomic64_set +#define erts_atomic64_set_acqb erts_no_atomic64_set +#define erts_atomic64_read_acqb erts_no_atomic64_read +#define erts_atomic64_inc_read_acqb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_acqb erts_no_atomic64_dec_read +#define erts_atomic64_inc_acqb erts_no_atomic64_inc +#define erts_atomic64_dec_acqb erts_no_atomic64_dec +#define erts_atomic64_add_read_acqb erts_no_atomic64_add_read +#define erts_atomic64_add_acqb erts_no_atomic64_add +#define erts_atomic64_read_bor_acqb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_acqb erts_no_atomic64_read_band +#define erts_atomic64_xchg_acqb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_acqb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_acqb erts_no_atomic64_read_bset + +#define erts_atomic64_init_relb erts_no_atomic64_set +#define erts_atomic64_set_relb erts_no_atomic64_set +#define erts_atomic64_read_relb erts_no_atomic64_read +#define erts_atomic64_inc_read_relb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_relb erts_no_atomic64_dec_read +#define erts_atomic64_inc_relb erts_no_atomic64_inc +#define erts_atomic64_dec_relb erts_no_atomic64_dec +#define erts_atomic64_add_read_relb erts_no_atomic64_add_read +#define erts_atomic64_add_relb erts_no_atomic64_add +#define erts_atomic64_read_bor_relb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_relb erts_no_atomic64_read_band +#define erts_atomic64_xchg_relb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_relb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_relb erts_no_atomic64_read_bset + +#define erts_atomic64_init_ddrb erts_no_atomic64_set +#define erts_atomic64_set_ddrb erts_no_atomic64_set +#define erts_atomic64_read_ddrb erts_no_atomic64_read +#define erts_atomic64_inc_read_ddrb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_ddrb erts_no_atomic64_dec_read +#define erts_atomic64_inc_ddrb erts_no_atomic64_inc +#define erts_atomic64_dec_ddrb erts_no_atomic64_dec +#define erts_atomic64_add_read_ddrb erts_no_atomic64_add_read +#define erts_atomic64_add_ddrb erts_no_atomic64_add +#define erts_atomic64_read_bor_ddrb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_ddrb erts_no_atomic64_read_band +#define erts_atomic64_xchg_ddrb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_ddrb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_ddrb erts_no_atomic64_read_bset + +#define erts_atomic64_init_rb erts_no_atomic64_set +#define erts_atomic64_set_rb erts_no_atomic64_set +#define erts_atomic64_read_rb erts_no_atomic64_read +#define erts_atomic64_inc_read_rb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_rb erts_no_atomic64_dec_read +#define erts_atomic64_inc_rb erts_no_atomic64_inc +#define erts_atomic64_dec_rb erts_no_atomic64_dec +#define erts_atomic64_add_read_rb erts_no_atomic64_add_read +#define erts_atomic64_add_rb erts_no_atomic64_add +#define erts_atomic64_read_bor_rb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_rb erts_no_atomic64_read_band +#define erts_atomic64_xchg_rb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_rb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_rb erts_no_atomic64_read_bset + +#define erts_atomic64_init_wb erts_no_atomic64_set +#define erts_atomic64_set_wb erts_no_atomic64_set +#define erts_atomic64_read_wb erts_no_atomic64_read +#define erts_atomic64_inc_read_wb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_wb erts_no_atomic64_dec_read +#define erts_atomic64_inc_wb erts_no_atomic64_inc +#define erts_atomic64_dec_wb erts_no_atomic64_dec +#define erts_atomic64_add_read_wb erts_no_atomic64_add_read +#define erts_atomic64_add_wb erts_no_atomic64_add +#define erts_atomic64_read_bor_wb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_wb erts_no_atomic64_read_band +#define erts_atomic64_xchg_wb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_wb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_wb erts_no_atomic64_read_bset + +#define erts_atomic64_set_dirty erts_no_atomic64_set +#define erts_atomic64_read_dirty erts_no_atomic64_read + #endif /* !USE_THREADS */ #if ERTS_GLB_INLINE_INCL_FUNC_DEF @@ -1548,6 +2138,16 @@ erts_thr_self(void) #endif } +ERTS_GLB_INLINE int +erts_thr_getname(erts_tid_t tid, char *buf, size_t len) +{ +#ifdef USE_THREADS + return ethr_getname(tid, buf, len); +#else + return -1; +#endif +} + ERTS_GLB_INLINE int erts_equal_tids(erts_tid_t x, erts_tid_t y) @@ -2383,6 +2983,104 @@ erts_no_atomic32_read_bset(erts_no_atomic32_t *var, return old; } +/* atomic64 */ + +ERTS_GLB_INLINE void +erts_no_atomic64_set(erts_no_atomic64_t *var, erts_aint64_t i) +{ + *var = i; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read(erts_no_atomic64_t *var) +{ + return *var; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_inc_read(erts_no_atomic64_t *incp) +{ + return ++(*incp); +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_dec_read(erts_no_atomic64_t *decp) +{ + return --(*decp); +} + +ERTS_GLB_INLINE void +erts_no_atomic64_inc(erts_no_atomic64_t *incp) +{ + ++(*incp); +} + +ERTS_GLB_INLINE void +erts_no_atomic64_dec(erts_no_atomic64_t *decp) +{ + --(*decp); +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_add_read(erts_no_atomic64_t *addp, erts_aint64_t i) +{ + return *addp += i; +} + +ERTS_GLB_INLINE void +erts_no_atomic64_add(erts_no_atomic64_t *addp, erts_aint64_t i) +{ + *addp += i; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_bor(erts_no_atomic64_t *var, erts_aint64_t mask) +{ + erts_aint64_t old; + old = *var; + *var |= mask; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_band(erts_no_atomic64_t *var, erts_aint64_t mask) +{ + erts_aint64_t old; + old = *var; + *var &= mask; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_xchg(erts_no_atomic64_t *xchgp, erts_aint64_t new) +{ + erts_aint64_t old = *xchgp; + *xchgp = new; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_cmpxchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new, + erts_aint64_t expected) +{ + erts_aint64_t old = *xchgp; + if (old == expected) + *xchgp = new; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_bset(erts_no_atomic64_t *var, + erts_aint64_t mask, + erts_aint64_t set) +{ + erts_aint64_t old = *var; + *var &= ~mask; + *var |= (mask & set); + return old; +} + /* spinlock */ ERTS_GLB_INLINE void @@ -2794,6 +3492,27 @@ ERTS_GLB_INLINE int erts_tse_swait(erts_tse_t *ep, int spincount) #endif } +ERTS_GLB_INLINE int erts_tse_twait(erts_tse_t *ep, Sint64 tmo) +{ +#ifdef USE_THREADS + return ethr_event_twait(&((ethr_ts_event *) ep)->event, + (ethr_sint64_t) tmo); +#else + return ENOTSUP; +#endif +} + +ERTS_GLB_INLINE int erts_tse_stwait(erts_tse_t *ep, int spincount, Sint64 tmo) +{ +#ifdef USE_THREADS + return ethr_event_stwait(&((ethr_ts_event *) ep)->event, + spincount, + (ethr_sint64_t) tmo); +#else + return ENOTSUP; +#endif +} + ERTS_GLB_INLINE int erts_tse_is_tmp(erts_tse_t *ep) { #ifdef USE_THREADS @@ -2838,6 +3557,15 @@ ERTS_GLB_INLINE void erts_thr_yield(void) #ifdef ETHR_HAVE_ETHR_SIG_FUNCS ERTS_GLB_INLINE void +erts_thr_kill(erts_tid_t tid, int sig) { +#ifdef USE_THREADS + int res = ethr_kill((ethr_tid)tid, sig); + if (res) + erts_thr_fatal_error(res, "killing thread"); +#endif +} + +ERTS_GLB_INLINE void erts_thr_sigmask(int how, const sigset_t *set, sigset_t *oset) { #ifdef USE_THREADS diff --git a/erts/emulator/beam/erl_time.h b/erts/emulator/beam/erl_time.h index 4bbdcaa3e3..cb7764addc 100644 --- a/erts/emulator/beam/erl_time.h +++ b/erts/emulator/beam/erl_time.h @@ -20,11 +20,16 @@ #ifndef ERL_TIME_H__ #define ERL_TIME_H__ -#define ERTS_SHORT_TIME_T_MAX ERTS_AINT32_T_MAX -#define ERTS_SHORT_TIME_T_MIN ERTS_AINT32_T_MIN -typedef erts_aint32_t erts_short_time_t; +#if defined(DEBUG) || 0 +#define ERTS_TIME_ASSERT(B) ERTS_ASSERT(B) +#else +#define ERTS_TIME_ASSERT(B) ((void) 1) +#endif + +typedef struct ErtsTimerWheel_ ErtsTimerWheel; +typedef erts_atomic64_t * ErtsNextTimeoutRef; +extern ErtsTimerWheel *erts_default_timer_wheel; -extern erts_smp_atomic32_t do_time; /* set at clock interrupt */ extern SysTimeval erts_first_emu_time; /* @@ -34,8 +39,8 @@ typedef struct erl_timer { struct erl_timer* next; /* next entry tiw slot or chain */ struct erl_timer* prev; /* prev entry tiw slot or chain */ Uint slot; /* slot in timer wheel */ - Uint count; /* number of loops remaining */ - int active; /* 1=activated, 0=deactivated */ + erts_smp_atomic_t wheel; + ErtsMonotonicTime timeout_pos; /* Timeout in absolute clock ticks */ /* called when timeout */ void (*timeout)(void*); /* called when cancel (may be NULL) */ @@ -62,7 +67,6 @@ union ErtsSmpPTimer_ { ErtsSmpPTimer *next; }; - void erts_create_smp_ptimer(ErtsSmpPTimer **timer_ref, Eterm id, ErlTimeoutProc timeout_func, @@ -70,36 +74,42 @@ void erts_create_smp_ptimer(ErtsSmpPTimer **timer_ref, void erts_cancel_smp_ptimer(ErtsSmpPTimer *ptimer); #endif +void erts_monitor_time_offset(Eterm id, Eterm ref); +int erts_demonitor_time_offset(Eterm ref); + +void erts_late_init_time_sup(void); + /* timer-wheel api */ -void erts_init_time(void); +ErtsTimerWheel *erts_create_timer_wheel(int); +ErtsNextTimeoutRef erts_get_next_timeout_reference(ErtsTimerWheel *); +void erts_init_time(int time_correction, ErtsTimeWarpMode time_warp_mode); void erts_set_timer(ErlTimer*, ErlTimeoutProc, ErlCancelProc, void*, Uint); void erts_cancel_timer(ErlTimer*); -void erts_bump_timer(erts_short_time_t); -Uint erts_timer_wheel_memory_size(void); Uint erts_time_left(ErlTimer *); -erts_short_time_t erts_next_time(void); +void erts_bump_timers(ErtsTimerWheel *, ErtsMonotonicTime); +Uint erts_timer_wheel_memory_size(void); #ifdef DEBUG void erts_p_slpq(void); #endif -ERTS_GLB_INLINE erts_short_time_t erts_do_time_read_and_reset(void); -ERTS_GLB_INLINE void erts_do_time_add(erts_short_time_t); +ErtsMonotonicTime erts_check_next_timeout_time(ErtsTimerWheel *, + ErtsMonotonicTime); + +ERTS_GLB_INLINE void erts_init_timer(ErlTimer *p); +ERTS_GLB_INLINE ErtsMonotonicTime erts_next_timeout_time(ErtsNextTimeoutRef); #if ERTS_GLB_INLINE_INCL_FUNC_DEF -ERTS_GLB_INLINE erts_short_time_t erts_do_time_read_and_reset(void) +ERTS_GLB_INLINE void erts_init_timer(ErlTimer *p) { - erts_short_time_t time = erts_smp_atomic32_xchg_acqb(&do_time, 0); - if (time < 0) - erl_exit(ERTS_ABORT_EXIT, "Internal time management error\n"); - return time; + erts_smp_atomic_init_nob(&p->wheel, (erts_aint_t) NULL); } -ERTS_GLB_INLINE void erts_do_time_add(erts_short_time_t elapsed) +ERTS_GLB_INLINE ErtsMonotonicTime erts_next_timeout_time(ErtsNextTimeoutRef nxt_tmo_ref) { - erts_smp_atomic32_add_relb(&do_time, elapsed); + return (ErtsMonotonicTime) erts_atomic64_read_acqb((erts_atomic64_t *) nxt_tmo_ref); } #endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */ @@ -107,7 +117,7 @@ ERTS_GLB_INLINE void erts_do_time_add(erts_short_time_t elapsed) /* time_sup */ -#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME)) +#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME_CPU_TIME)) # ifndef HAVE_ERTS_NOW_CPU # define HAVE_ERTS_NOW_CPU # ifdef HAVE_GETHRVTIME @@ -121,25 +131,220 @@ void erts_get_now_cpu(Uint* megasec, Uint* sec, Uint* microsec); typedef UWord erts_approx_time_t; erts_approx_time_t erts_get_approx_time(void); -void erts_get_timeval(SysTimeval *tv); -erts_time_t erts_get_time(void); +int erts_has_time_correction(void); +int erts_check_time_adj_support(int time_correction, + ErtsTimeWarpMode time_warp_mode); + +ErtsTimeWarpMode erts_time_warp_mode(void); -ERTS_GLB_INLINE int erts_cmp_timeval(SysTimeval *t1p, SysTimeval *t2p); +typedef enum { + ERTS_TIME_OFFSET_PRELIMINARY, + ERTS_TIME_OFFSET_FINAL, + ERTS_TIME_OFFSET_VOLATILE +} ErtsTimeOffsetState; + +ErtsTimeOffsetState erts_time_offset_state(void); +ErtsTimeOffsetState erts_finalize_time_offset(void); +struct process; +Eterm erts_get_monotonic_start_time(struct process *c_p); +Eterm erts_monotonic_time_source(struct process*c_p); +Eterm erts_system_time_source(struct process*c_p); + +#ifdef SYS_CLOCK_RESOLUTION +#define ERTS_CLKTCK_RESOLUTION ((ErtsMonotonicTime) (SYS_CLOCK_RESOLUTION*1000)) +#else +#define ERTS_CLKTCK_RESOLUTION (erts_time_sup__.r.o.clktck_resolution) +#endif + +struct erts_time_sup_read_only__ { + ErtsMonotonicTime monotonic_time_unit; +#ifndef SYS_CLOCK_RESOLUTION + ErtsMonotonicTime clktck_resolution; +#endif +}; + +typedef struct { + union { + struct erts_time_sup_read_only__ o; + char align__[(((sizeof(struct erts_time_sup_read_only__) - 1) + / ASSUMED_CACHE_LINE_SIZE) + 1) + * ASSUMED_CACHE_LINE_SIZE]; + } r; +} ErtsTimeSupData; + +extern ErtsTimeSupData erts_time_sup__; + +ERTS_GLB_INLINE Uint64 +erts_time_unit_conversion(Uint64 value, + Uint32 from_time_unit, + Uint32 to_time_unit); #if ERTS_GLB_INLINE_INCL_FUNC_DEF -ERTS_GLB_INLINE int -erts_cmp_timeval(SysTimeval *t1p, SysTimeval *t2p) +ERTS_GLB_INLINE Uint64 +erts_time_unit_conversion(Uint64 value, + Uint32 from_time_unit, + Uint32 to_time_unit) { - if (t1p->tv_sec == t2p->tv_sec) { - if (t1p->tv_usec < t2p->tv_usec) - return -1; - else if (t1p->tv_usec > t2p->tv_usec) - return 1; - return 0; - } - return t1p->tv_sec < t2p->tv_sec ? -1 : 1; + Uint64 high, low, result; + if (value <= ~((Uint64) 0)/to_time_unit) + return (value*to_time_unit)/from_time_unit; + + low = value & ((Uint64) 0xffffffff); + high = (value >> 32) & ((Uint64) 0xffffffff); + + low *= to_time_unit; + high *= to_time_unit; + + high += (low >> 32) & ((Uint64) 0xffffffff); + low &= ((Uint64) 0xffffffff); + + result = high % from_time_unit; + high /= from_time_unit; + high <<= 32; + + result <<= 32; + result += low; + result /= from_time_unit; + result += high; + + return result; } -#endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */ +#endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */ + +#if ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT + +/* + * If the monotonic time unit is a compile time constant, + * it is assumed (and need) to be a power of 10. + */ + +#if ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT < 1000*1000 +# error Compile time time unit needs to be at least 1000000 +#endif + +#define ERTS_MONOTONIC_TIME_UNIT \ + ((ErtsMonotonicTime) ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT) + +#if ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT == 1000*1000*1000 +/* Nano-second time unit */ + +#define ERTS_MONOTONIC_TO_SEC__(NSEC) ((NSEC) / (1000*1000*1000)) +#define ERTS_MONOTONIC_TO_MSEC__(NSEC) ((NSEC) / (1000*1000)) +#define ERTS_MONOTONIC_TO_USEC__(NSEC) ((NSEC) / 1000) +#define ERTS_MONOTONIC_TO_NSEC__(NSEC) (NSEC) + +#define ERTS_SEC_TO_MONOTONIC__(SEC) (((ErtsMonotonicTime) (SEC))*(1000*1000*1000)) +#define ERTS_MSEC_TO_MONOTONIC__(MSEC) (((ErtsMonotonicTime) (MSEC))*(1000*1000)) +#define ERTS_USEC_TO_MONOTONIC__(USEC) (((ErtsMonotonicTime) (USEC))*1000) +#define ERTS_NSEC_TO_MONOTONIC__(NSEC) ((ErtsMonotonicTime) (NSEC)) + +#elif ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT == 1000*1000 +/* Micro-second time unit */ + +#define ERTS_MONOTONIC_TO_SEC__(USEC) ((USEC) / (1000*1000)) +#define ERTS_MONOTONIC_TO_MSEC__(USEC) ((USEC) / 1000) +#define ERTS_MONOTONIC_TO_USEC__(USEC) (USEC) +#define ERTS_MONOTONIC_TO_NSEC__(USEC) ((USEC)*1000) + +#define ERTS_SEC_TO_MONOTONIC__(SEC) (((ErtsMonotonicTime) (SEC))*(1000*1000)) +#define ERTS_MSEC_TO_MONOTONIC__(MSEC) (((ErtsMonotonicTime) (MSEC))*1000) +#define ERTS_USEC_TO_MONOTONIC__(USEC) ((ErtsMonotonicTime) (USEC)) +#define ERTS_NSEC_TO_MONOTONIC__(NSEC) (((ErtsMonotonicTime) (NSEC))/1000) + +#else +#error Missing implementation for monotonic time unit +#endif + +#define ERTS_MONOTONIC_TO_CLKTCKS__(MON) \ + ((MON) / (ERTS_MONOTONIC_TIME_UNIT/ERTS_CLKTCK_RESOLUTION)) +#define ERTS_CLKTCKS_TO_MONOTONIC__(TCKS) \ + ((TCKS) * (ERTS_MONOTONIC_TIME_UNIT/ERTS_CLKTCK_RESOLUTION)) + +#else /* !ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT */ + +#define ERTS_MONOTONIC_TIME_UNIT (erts_time_sup__.r.o.monotonic_time_unit) + +#define ERTS_CONV_FROM_MON_UNIT___(M, TO) \ + ((ErtsMonotonicTime) \ + erts_time_unit_conversion((Uint64) (M), \ + (Uint32) ERTS_MONOTONIC_TIME_UNIT, \ + (Uint32) (TO))) + +#define ERTS_CONV_TO_MON_UNIT___(M, FROM) \ + ((ErtsMonotonicTime) \ + erts_time_unit_conversion((Uint64) (M), \ + (Uint32) (FROM), \ + (Uint32) ERTS_MONOTONIC_TIME_UNIT)) \ + +#define ERTS_MONOTONIC_TO_SEC__(M) \ + ERTS_CONV_FROM_MON_UNIT___((M), 1) +#define ERTS_MONOTONIC_TO_MSEC__(M) \ + ERTS_CONV_FROM_MON_UNIT___((M), 1000) +#define ERTS_MONOTONIC_TO_USEC__(M) \ + ERTS_CONV_FROM_MON_UNIT___((M), 1000*1000) +#define ERTS_MONOTONIC_TO_NSEC__(M) \ + ERTS_CONV_FROM_MON_UNIT___((M), 1000*1000*1000) + +#define ERTS_SEC_TO_MONOTONIC__(SEC) \ + ERTS_CONV_TO_MON_UNIT___((SEC), 1) +#define ERTS_MSEC_TO_MONOTONIC__(MSEC) \ + ERTS_CONV_TO_MON_UNIT___((MSEC), 1000) +#define ERTS_USEC_TO_MONOTONIC__(USEC) \ + ERTS_CONV_TO_MON_UNIT___((USEC), 1000*1000) +#define ERTS_NSEC_TO_MONOTONIC__(NSEC) \ + ERTS_CONV_TO_MON_UNIT___((NSEC), 1000*1000*1000) + +#define ERTS_MONOTONIC_TO_CLKTCKS__(MON) \ + ERTS_CONV_FROM_MON_UNIT___((MON), ERTS_CLKTCK_RESOLUTION) +#define ERTS_CLKTCKS_TO_MONOTONIC__(TCKS) \ + ERTS_CONV_TO_MON_UNIT___((TCKS), ERTS_CLKTCK_RESOLUTION) + +#endif /* !ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT */ + +#define ERTS_MSEC_TO_CLKTCKS__(MON) \ + ((MON) * (ERTS_CLKTCK_RESOLUTION/1000)) +#define ERTS_CLKTCKS_TO_MSEC__(TCKS) \ + ((TCKS) / (ERTS_CLKTCK_RESOLUTION/1000)) + +#define ERTS_MONOTONIC_TO_SEC(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_MONOTONIC_TO_SEC__((X))) +#define ERTS_MONOTONIC_TO_MSEC(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_MONOTONIC_TO_MSEC__((X))) +#define ERTS_MONOTONIC_TO_USEC(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_MONOTONIC_TO_USEC__((X))) +#define ERTS_MONOTONIC_TO_NSEC(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_MONOTONIC_TO_NSEC__((X))) +#define ERTS_SEC_TO_MONOTONIC(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_SEC_TO_MONOTONIC__((X))) +#define ERTS_MSEC_TO_MONOTONIC(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_MSEC_TO_MONOTONIC__((X))) +#define ERTS_USEC_TO_MONOTONIC(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_USEC_TO_MONOTONIC__((X))) +#define ERTS_NSEC_TO_MONOTONIC(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_NSEC_TO_MONOTONIC__((X))) + +#define ERTS_MONOTONIC_TO_CLKTCKS(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_MONOTONIC_TO_CLKTCKS__((X))) +#define ERTS_CLKTCKS_TO_MONOTONIC(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_CLKTCKS_TO_MONOTONIC__((X))) + +#define ERTS_MSEC_TO_CLKTCKS(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_MSEC_TO_CLKTCKS__((X))) +#define ERTS_CLKTCKS_TO_MSEC(X) \ + (ERTS_TIME_ASSERT((X) >= 0), \ + ERTS_CLKTCKS_TO_MSEC__((X))) + #endif /* ERL_TIME_H__ */ diff --git a/erts/emulator/beam/erl_time_sup.c b/erts/emulator/beam/erl_time_sup.c index 3272a5326d..bbdedcc128 100644 --- a/erts/emulator/beam/erl_time_sup.c +++ b/erts/emulator/beam/erl_time_sup.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1999-2012. All Rights Reserved. + * Copyright Ericsson AB 1999-2015. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -18,60 +18,10 @@ */ /* -** Support routines for the timer wheel -** -** This code contains two strategies for dealing with -** date/time changes in the system. -** If the system has some kind of high resolution timer (HAVE_GETHRTIME), -** the high resolution timer is used to correct the time-of-day and the -** timeouts, the base source is the hrtimer, but at certain intervals the -** OS time-of-day is checked and if it is not within certain bounds, the -** delivered time gets slowly adjusted for each call until -** it corresponds to the system time (built-in adjtime...). -** The call gethrtime() is detected by autoconf on Unix, but other -** platforms may define it in erl_*_sys.h and implement -** their own high resolution timer. The high resolution timer -** strategy is (probably) best on all systems where the timer have -** a resolution higher or equal to gettimeofday (or what's implemented -** is sys_gettimeofday()). The actual resolution is the interesting thing, -** not the unit's thats used (i.e. on VxWorks, nanoseconds can be -** retrieved in terms of units, but the actual resolution is the same as -** for the clock ticks). -** If the systems best timer routine is kernel ticks returned from -** sys_times(), and the actual resolution of sys_gettimeofday() is -** better (like most unixes that does not have any realtime extensions), -** another strategy is used. The tolerant gettimeofday() corrects -** the value with respect to uptime (sys_times() return value) and checks -** for correction both when delivering timeticks and delivering nowtime. -** this strategy is slower, but accurate on systems without better timer -** routines. The kernel tick resolution is not enough to implement -** a gethrtime routine. On Linux and other non solaris unix-boxes the second -** strategy is used, on all other platforms we use the first. -** -** The following is expected (from sys.[ch] and erl_*_sys.h): -** -** 64 bit integers. So it is, and so it will be. -** -** sys_init_time(), will return the clock resolution in MS and -** that's about it. More could be added of course -** If the clock-rate is constant (i.e. 1 ms) one can define -** SYS_CLOCK_RESOLUTION (to 1), -** which makes erts_deliver_time/erts_time_remaining a bit faster. -** -** if HAVE_GETHRTIME is defined: -** sys_gethrtime() will return a SysHrTime (long long) representing -** nanoseconds, sys_init_hrtime() will do any initialization. -** else -** a long (64bit) integer type called Sint64 should be defined. -** -** sys_times() will return clock_ticks since start and -** fill in a SysTimes structure (struct tms). Instead of CLK_TCK, -** SYS_CLK_TCK is used to determine the resolution of kernel ticks. -** -** sys_gettimeofday() will take a SysTimeval (a struct timeval) as parameter -** and fill it in as gettimeofday(X,NULL). -** -*/ + * Support routines for the time + */ + +/* #define ERTS_TIME_CORRECTION_PRINT */ #ifdef HAVE_CONFIG_H # include "config.h" @@ -80,384 +30,1107 @@ #include "sys.h" #include "erl_vm.h" #include "global.h" - + static erts_smp_mtx_t erts_timeofday_mtx; - -static SysTimeval inittv; /* Used everywhere, the initial time-of-day */ +static erts_smp_mtx_t erts_get_time_mtx; static SysTimes t_start; /* Used in elapsed_time_both */ -static SysTimeval gtv; /* Used in wall_clock_elapsed_time_both */ -static SysTimeval then; /* Used in get_now */ -static SysTimeval last_emu_time; /* Used in erts_get_emu_time() */ -SysTimeval erts_first_emu_time; /* Used in erts_get_emu_time() */ +static ErtsMonotonicTime prev_wall_clock_elapsed; /* Used in wall_clock_elapsed_time_both */ +static ErtsMonotonicTime previous_now; /* Used in get_now */ -union { - erts_smp_atomic_t time; - char align[ERTS_CACHE_LINE_SIZE]; -} approx erts_align_attribute(ERTS_CACHE_LINE_SIZE); +static ErtsMonitor *time_offset_monitors = NULL; +static Uint no_time_offset_monitors = 0; + +#ifdef DEBUG +static int time_sup_initialized = 0; +#endif + +#define ERTS_MONOTONIC_TIME_KILO \ + ((ErtsMonotonicTime) 1000) +#define ERTS_MONOTONIC_TIME_MEGA \ + (ERTS_MONOTONIC_TIME_KILO*ERTS_MONOTONIC_TIME_KILO) +#define ERTS_MONOTONIC_TIME_GIGA \ + (ERTS_MONOTONIC_TIME_MEGA*ERTS_MONOTONIC_TIME_KILO) +#define ERTS_MONOTONIC_TIME_TERA \ + (ERTS_MONOTONIC_TIME_GIGA*ERTS_MONOTONIC_TIME_KILO) static void -init_approx_time(void) +schedule_send_time_offset_changed_notifications(ErtsMonotonicTime new_offset); + +/* + * NOTE! ERTS_MONOTONIC_TIME_START *need* to be a multiple + * of ERTS_MONOTONIC_TIME_UNIT. + */ + +#if ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT + +#ifdef ARCH_32 +/* + * Want to use a big-num of arity 2 as long as possible (584 years + * in the nano-second time unit case). + */ +#define ERTS_MONOTONIC_TIME_START \ + (((((((ErtsMonotonicTime) 1) << 32)-1) \ + / ERTS_MONOTONIC_TIME_UNIT) \ + * ERTS_MONOTONIC_TIME_UNIT) \ + + ERTS_MONOTONIC_TIME_UNIT) + +#else /* ARCH_64 */ + +#if ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT <= 10*1000*1000 + +/* + * Using micro second time unit or lower. Start at zero since + * time will remain an immediate for a very long time anyway + * (1827 years in the 10 micro second case)... + */ +#define ERTS_MONOTONIC_TIME_START ((ErtsMonotonicTime) 0) + +#else /* ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT > 1000*1000 */ + +/* + * Want to use an immediate as long as possible (36 years in the + * nano-second time unit case). +*/ +#define ERTS_MONOTONIC_TIME_START \ + ((((ErtsMonotonicTime) MIN_SMALL) \ + / ERTS_MONOTONIC_TIME_UNIT) \ + * ERTS_MONOTONIC_TIME_UNIT) + +#endif /* ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT > 1000*1000 */ + +#endif /* ARCH_64 */ + +#define ERTS_MONOTONIC_OFFSET_NATIVE \ + (ERTS_MONOTONIC_TIME_START - ERTS_MONOTONIC_TIME_UNIT) +#define ERTS_MONOTONIC_OFFSET_NSEC \ + ERTS_MONOTONIC_TO_NSEC__(ERTS_MONOTONIC_OFFSET_NATIVE) +#define ERTS_MONOTONIC_OFFSET_USEC \ + ERTS_MONOTONIC_TO_USEC__(ERTS_MONOTONIC_OFFSET_NATIVE) +#define ERTS_MONOTONIC_OFFSET_MSEC \ + ERTS_MONOTONIC_TO_MSEC__(ERTS_MONOTONIC_OFFSET_NATIVE) +#define ERTS_MONOTONIC_OFFSET_SEC \ + ERTS_MONOTONIC_TO_SEC__(ERTS_MONOTONIC_OFFSET_NATIVE) + +#else /* ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT */ + +/* + * Initialized in erts_init_time_sup()... + */ + +#define ERTS_MONOTONIC_TIME_START (time_sup.r.o.start) +#define ERTS_MONOTONIC_OFFSET_NATIVE (time_sup.r.o.start_offset.native) +#define ERTS_MONOTONIC_OFFSET_NSEC (time_sup.r.o.start_offset.nsec) +#define ERTS_MONOTONIC_OFFSET_USEC (time_sup.r.o.start_offset.usec) +#define ERTS_MONOTONIC_OFFSET_MSEC (time_sup.r.o.start_offset.msec) +#define ERTS_MONOTONIC_OFFSET_SEC (time_sup.r.o.start_offset.sec) + +#endif /* ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT */ + +struct time_sup_read_only__ { + ErtsMonotonicTime (*get_time)(void); + int correction; + ErtsTimeWarpMode warp_mode; +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + ErtsMonotonicTime moffset; + int os_monotonic_time_disable; + char *os_monotonic_time_func; + char *os_monotonic_time_clock_id; + int os_monotonic_time_locked; + Uint64 os_monotonic_time_resolution; + Uint64 os_monotonic_time_extended; +#endif + char *os_system_time_func; + char *os_system_time_clock_id; + int os_system_time_locked; + Uint64 os_system_time_resolution; + Uint64 os_system_time_extended; +#if !ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT + ErtsMonotonicTime start; + struct { + ErtsMonotonicTime native; + ErtsMonotonicTime nsec; + ErtsMonotonicTime usec; + ErtsMonotonicTime msec; + ErtsMonotonicTime sec; + } start_offset; +#endif + struct { + ErtsMonotonicTime large_diff; + ErtsMonotonicTime small_diff; + } adj; +}; + +typedef struct { +#ifndef ERTS_HAVE_CORRECTED_OS_MONOTONIC + ErtsMonotonicTime drift; /* Correction for os monotonic drift */ +#endif + ErtsMonotonicTime error; /* Correction for error between system times */ +} ErtsMonotonicCorrection; + +typedef struct { + ErtsMonotonicTime erl_mtime; + ErtsMonotonicTime os_mtime; + ErtsMonotonicCorrection correction; +} ErtsMonotonicCorrectionInstance; + +#define ERTS_DRIFT_INTERVALS 5 +typedef struct { + struct { + struct { + ErtsMonotonicTime sys; + ErtsMonotonicTime mon; + } diff; + struct { + ErtsMonotonicTime sys; + ErtsMonotonicTime mon; + } time; + } intervals[ERTS_DRIFT_INTERVALS]; + struct { + ErtsMonotonicTime sys; + ErtsMonotonicTime mon; + } acc; + int ix; + int dirty_counter; +} ErtsMonotonicDriftData; + +typedef struct { + ErtsMonotonicCorrectionInstance prev; + ErtsMonotonicCorrectionInstance curr; +#ifndef ERTS_HAVE_CORRECTED_OS_MONOTONIC + ErtsMonotonicDriftData drift; +#endif + ErtsMonotonicTime last_check; + int short_check_interval; +} ErtsMonotonicCorrectionData; + +struct time_sup_infrequently_changed__ { +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + struct { + erts_smp_rwmtx_t rwmtx; + ErlTimer timer; + ErtsMonotonicCorrectionData cdata; + } parmon; + ErtsMonotonicTime minit; +#endif + int finalized_offset; + ErtsSystemTime sinit; + ErtsMonotonicTime not_corrected_moffset; + erts_atomic64_t offset; +}; + +struct time_sup_frequently_changed__ { + ErtsMonotonicTime last_not_corrected_time; +}; + +static struct { + union { + struct time_sup_read_only__ o; + char align__[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(struct time_sup_read_only__))]; + } r; + union { + struct time_sup_infrequently_changed__ c; + char align__[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(struct time_sup_infrequently_changed__))]; + } inf; + union { + struct time_sup_frequently_changed__ c; + char align__[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(struct time_sup_frequently_changed__))]; + } f; +} time_sup erts_align_attribute(ERTS_CACHE_LINE_SIZE); + +ErtsTimeSupData erts_time_sup__ erts_align_attribute(ERTS_CACHE_LINE_SIZE); + +/* + * erts_get_approx_time() returns an *approximate* time + * in seconds. NOTE that this time may jump backwards!!! + */ +erts_approx_time_t +erts_get_approx_time(void) { - erts_smp_atomic_init_nob(&approx.time, 0); + ErtsSystemTime stime = erts_os_system_time(); + return (erts_approx_time_t) ERTS_MONOTONIC_TO_SEC(stime); } -static ERTS_INLINE erts_approx_time_t -get_approx_time(void) +static ERTS_INLINE void +init_time_offset(ErtsMonotonicTime offset) { - return (erts_approx_time_t) erts_smp_atomic_read_nob(&approx.time); + erts_atomic64_init_nob(&time_sup.inf.c.offset, (erts_aint64_t) offset); } static ERTS_INLINE void -update_approx_time(SysTimeval *tv) +set_time_offset(ErtsMonotonicTime offset) { - erts_approx_time_t new_secs = (erts_approx_time_t) tv->tv_sec; - erts_approx_time_t old_secs = get_approx_time(); - if (old_secs != new_secs) - erts_smp_atomic_set_nob(&approx.time, new_secs); + erts_atomic64_set_relb(&time_sup.inf.c.offset, (erts_aint64_t) offset); } -/* - * erts_get_approx_time() returns an *approximate* time - * in seconds. NOTE that this time may jump backwards!!! - */ -erts_approx_time_t -erts_get_approx_time(void) +static ERTS_INLINE ErtsMonotonicTime +get_time_offset(void) { - return get_approx_time(); + return (ErtsMonotonicTime) erts_atomic64_read_acqb(&time_sup.inf.c.offset); } -#ifdef HAVE_GETHRTIME -int erts_disable_tolerant_timeofday; +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT -static SysHrTime hr_init_time, hr_last_correction_check, - hr_correction, hr_last_time; +/* + * Time correction adjustments made due to + * error between Erlang system time and OS + * system time: + * - Large adjustment ~1% + * - Small adjustment ~0.05% + */ +#define ERTS_TCORR_ERR_UNIT 2048 +#define ERTS_TCORR_ERR_LARGE_ADJ 20 +#define ERTS_TCORR_ERR_SMALL_ADJ 1 + +#define ERTS_INIT_SHORT_INTERVAL_COUNTER 10 +#define ERTS_LONG_TIME_CORRECTION_CHECK ERTS_SEC_TO_MONOTONIC(60) +#define ERTS_SHORT_TIME_CORRECTION_CHECK ERTS_SEC_TO_MONOTONIC(15) + +#define ERTS_TIME_DRIFT_MAX_ADJ_DIFF ERTS_USEC_TO_MONOTONIC(50) +#define ERTS_TIME_DRIFT_MIN_ADJ_DIFF ERTS_USEC_TO_MONOTONIC(5) -static void init_tolerant_timeofday(void) +static ERTS_INLINE ErtsMonotonicTime +calc_corrected_erl_mtime(ErtsMonotonicTime os_mtime, + ErtsMonotonicCorrectionInstance *cip, + ErtsMonotonicTime *os_mdiff_p) { - /* Should be in sys.c */ -#if defined(HAVE_SYSCONF) && defined(_SC_NPROCESSORS_CONF) - if (sysconf(_SC_NPROCESSORS_CONF) > 1) { - char b[1024]; - int maj,min,build; - os_flavor(b,1024); - os_version(&maj,&min,&build); - if (!strcmp(b,"sunos") && maj <= 5 && min <= 7) { - erts_disable_tolerant_timeofday = 1; - } - } + ErtsMonotonicTime erl_mtime, diff = os_mtime - cip->os_mtime; + ERTS_TIME_ASSERT(diff >= 0); +#ifndef ERTS_HAVE_CORRECTED_OS_MONOTONIC + diff += (cip->correction.drift*diff)/ERTS_MONOTONIC_TIME_UNIT; #endif - hr_init_time = sys_gethrtime(); - hr_last_correction_check = hr_last_time = hr_init_time; - hr_correction = 0; + erl_mtime = cip->erl_mtime; + erl_mtime += diff; + erl_mtime += cip->correction.error*(diff/ERTS_TCORR_ERR_UNIT); + if (os_mdiff_p) + *os_mdiff_p = diff; + return erl_mtime; } -static void get_tolerant_timeofday(SysTimeval *tv) +static ErtsMonotonicTime get_corrected_time(void) { - SysHrTime diff_time, curr; + ErtsMonotonicTime os_mtime; + ErtsMonotonicCorrectionData cdata; + ErtsMonotonicCorrectionInstance *cip; + + erts_smp_rwmtx_rlock(&time_sup.inf.c.parmon.rwmtx); - if (erts_disable_tolerant_timeofday) { - sys_gettimeofday(tv); - return; + os_mtime = erts_os_monotonic_time(); + + cdata = time_sup.inf.c.parmon.cdata; + + erts_smp_rwmtx_runlock(&time_sup.inf.c.parmon.rwmtx); + + if (os_mtime >= cdata.curr.os_mtime) + cip = &cdata.curr; + else { + if (os_mtime < cdata.prev.os_mtime) + erl_exit(ERTS_ABORT_EXIT, + "OS monotonic time stepped backwards\n"); + cip = &cdata.prev; } - *tv = inittv; - diff_time = ((curr = sys_gethrtime()) + hr_correction - hr_init_time) / 1000; - if (curr < hr_init_time) { - erl_exit(1,"Unexpected behaviour from operating system high " - "resolution timer"); + return calc_corrected_erl_mtime(os_mtime, cip, NULL); +} + +#ifdef ERTS_TIME_CORRECTION_PRINT + +static ERTS_INLINE void +print_correction(int change, + ErtsMonotonicTime sdiff, + ErtsMonotonicTime old_ecorr, + ErtsMonotonicTime old_dcorr, + ErtsMonotonicTime new_ecorr, + ErtsMonotonicTime new_dcorr, + Uint tmo) +{ + ErtsMonotonicTime usec_sdiff; + if (sdiff < 0) + usec_sdiff = -1*ERTS_MONOTONIC_TO_USEC(-1*sdiff); + else + usec_sdiff = ERTS_MONOTONIC_TO_USEC(sdiff); + + if (!change) + fprintf(stderr, + "sdiff = %lld usec : [ec=%lld ppm, dc=%lld ppb] : " + "tmo = %lld msec\r\n", + (long long) usec_sdiff, + (long long) (1000000*old_ecorr) / ERTS_TCORR_ERR_UNIT, + (long long) (1000000000*old_dcorr) / ERTS_MONOTONIC_TIME_UNIT, + (long long) tmo); + else + fprintf(stderr, + "sdiff = %lld usec : [ec=%lld ppm, dc=%lld ppb] " + "-> [ec=%lld ppm, dc=%lld ppb] : tmo = %lld msec\r\n", + (long long) usec_sdiff, + (long long) (1000000*old_ecorr) / ERTS_TCORR_ERR_UNIT, + (long long) (1000000000*old_dcorr) / ERTS_MONOTONIC_TIME_UNIT, + (long long) (1000000*new_ecorr) / ERTS_TCORR_ERR_UNIT, + (long long) (1000000000*new_dcorr) / ERTS_MONOTONIC_TIME_UNIT, + (long long) tmo); + +} + +#endif + +static void +check_time_correction(void *unused) +{ +#ifndef ERTS_TIME_CORRECTION_PRINT +# define ERTS_PRINT_CORRECTION +#else +# ifdef ERTS_HAVE_CORRECTED_OS_MONOTONIC +# define ERTS_PRINT_CORRECTION \ + print_correction(set_new_correction, \ + sdiff, \ + cip->correction.error, \ + 0, \ + new_correction.error, \ + 0, \ + timeout) +# else +# define ERTS_PRINT_CORRECTION \ + print_correction(set_new_correction, \ + sdiff, \ + cip->correction.error, \ + cip->correction.drift, \ + new_correction.error, \ + new_correction.drift, \ + timeout) +# endif +#endif + ErtsMonotonicCorrectionData cdata; + ErtsMonotonicCorrection new_correction; + ErtsMonotonicCorrectionInstance *cip; + ErtsMonotonicTime mdiff, sdiff, os_mtime, erl_mtime, os_stime, + erl_stime, time_offset; + Uint timeout; + int set_new_correction, begin_short_intervals = 0; + + erts_smp_rwmtx_rlock(&time_sup.inf.c.parmon.rwmtx); + + ASSERT(time_sup.inf.c.finalized_offset); + + erts_os_times(&os_mtime, &os_stime); + + cdata = time_sup.inf.c.parmon.cdata; + + erts_smp_rwmtx_runlock(&time_sup.inf.c.parmon.rwmtx); + + if (os_mtime < cdata.curr.os_mtime) + erl_exit(ERTS_ABORT_EXIT, + "OS monotonic time stepped backwards\n"); + cip = &cdata.curr; + + erl_mtime = calc_corrected_erl_mtime(os_mtime, cip, &mdiff); + time_offset = get_time_offset(); + erl_stime = erl_mtime + time_offset; + + sdiff = erl_stime - os_stime; + + new_correction = cip->correction; + + if (time_sup.r.o.warp_mode == ERTS_MULTI_TIME_WARP_MODE + && (sdiff < -2*time_sup.r.o.adj.small_diff + || 2*time_sup.r.o.adj.small_diff < sdiff)) { + /* System time diff exeeded limits; change time offset... */ + time_offset -= sdiff; + sdiff = 0; + set_time_offset(time_offset); + schedule_send_time_offset_changed_notifications(time_offset); + begin_short_intervals = 1; + if (cdata.curr.correction.error == 0) + set_new_correction = 0; + else { + set_new_correction = 1; + new_correction.error = 0; + } + } + else if (cdata.curr.correction.error == 0) { + if (sdiff < -time_sup.r.o.adj.small_diff) { + set_new_correction = 1; + if (sdiff < -time_sup.r.o.adj.large_diff) + new_correction.error = ERTS_TCORR_ERR_LARGE_ADJ; + else + new_correction.error = ERTS_TCORR_ERR_SMALL_ADJ; + } + else if (sdiff > time_sup.r.o.adj.small_diff) { + set_new_correction = 1; + if (sdiff > time_sup.r.o.adj.large_diff) + new_correction.error = -ERTS_TCORR_ERR_LARGE_ADJ; + else + new_correction.error = -ERTS_TCORR_ERR_SMALL_ADJ; + } + else { + set_new_correction = 0; + } + } + else if (cdata.curr.correction.error > 0) { + if (sdiff < 0) { + if (cdata.curr.correction.error == ERTS_TCORR_ERR_LARGE_ADJ + || -time_sup.r.o.adj.large_diff <= sdiff) + set_new_correction = 0; + else { + new_correction.error = ERTS_TCORR_ERR_LARGE_ADJ; + set_new_correction = 1; + } + } + else if (sdiff > time_sup.r.o.adj.small_diff) { + set_new_correction = 1; + if (sdiff > time_sup.r.o.adj.large_diff) + new_correction.error = -ERTS_TCORR_ERR_LARGE_ADJ; + else + new_correction.error = -ERTS_TCORR_ERR_SMALL_ADJ; + } + else { + set_new_correction = 1; + new_correction.error = 0; + } + } + else /* if (cdata.curr.correction.error < 0) */ { + if (0 < sdiff) { + if (cdata.curr.correction.error == -ERTS_TCORR_ERR_LARGE_ADJ + || sdiff <= time_sup.r.o.adj.large_diff) + set_new_correction = 0; + else { + new_correction.error = -ERTS_TCORR_ERR_LARGE_ADJ; + set_new_correction = 1; + } + set_new_correction = 0; + } + else if (sdiff < -time_sup.r.o.adj.small_diff) { + set_new_correction = 1; + if (sdiff < -time_sup.r.o.adj.large_diff) + new_correction.error = ERTS_TCORR_ERR_LARGE_ADJ; + else + new_correction.error = ERTS_TCORR_ERR_SMALL_ADJ; + } + else { + set_new_correction = 1; + new_correction.error = 0; + } } - if ((curr - hr_last_correction_check) / 1000 > 1000000) { - /* Check the correction need */ - SysHrTime tv_diff, diffdiff; - SysTimeval tmp; - int done = 0; - - sys_gettimeofday(&tmp); - tv_diff = ((SysHrTime) tmp.tv_sec) * 1000000 + tmp.tv_usec; - tv_diff -= ((SysHrTime) inittv.tv_sec) * 1000000 + inittv.tv_usec; - diffdiff = diff_time - tv_diff; - if (diffdiff > 10000) { - SysHrTime corr = (curr - hr_last_time) / 100; - if (corr / 1000 >= diffdiff) { - ++done; - hr_correction -= ((SysHrTime)diffdiff) * 1000; - } else { - hr_correction -= corr; +#ifndef ERTS_HAVE_CORRECTED_OS_MONOTONIC + { + ErtsMonotonicDriftData *ddp = &time_sup.inf.c.parmon.cdata.drift; + int ix = ddp->ix; + ErtsMonotonicTime mtime_diff, old_os_mtime; + + old_os_mtime = ddp->intervals[ix].time.mon; + mtime_diff = os_mtime - old_os_mtime; + + if (mtime_diff >= ERTS_SEC_TO_MONOTONIC(10)) { + ErtsMonotonicTime drift_adj, drift_adj_diff, old_os_stime, + stime_diff, mtime_acc, stime_acc, avg_drift_adj; + + old_os_stime = ddp->intervals[ix].time.sys; + + mtime_acc = ddp->acc.mon; + stime_acc = ddp->acc.sys; + + avg_drift_adj = (((stime_acc - mtime_acc)*ERTS_MONOTONIC_TIME_UNIT) + / mtime_acc); + + mtime_diff = os_mtime - old_os_mtime; + stime_diff = os_stime - old_os_stime; + drift_adj = (((stime_diff - mtime_diff)*ERTS_MONOTONIC_TIME_UNIT) + / mtime_diff); + + ix++; + if (ix >= ERTS_DRIFT_INTERVALS) + ix = 0; + mtime_acc -= ddp->intervals[ix].diff.mon; + mtime_acc += mtime_diff; + stime_acc -= ddp->intervals[ix].diff.sys; + stime_acc += stime_diff; + + ddp->intervals[ix].diff.mon = mtime_diff; + ddp->intervals[ix].diff.sys = stime_diff; + ddp->intervals[ix].time.mon = os_mtime; + ddp->intervals[ix].time.sys = os_stime; + + ddp->ix = ix; + ddp->acc.mon = mtime_acc; + ddp->acc.sys = stime_acc; + + drift_adj_diff = avg_drift_adj - drift_adj; + if (drift_adj_diff < -ERTS_TIME_DRIFT_MAX_ADJ_DIFF + || ERTS_TIME_DRIFT_MAX_ADJ_DIFF < drift_adj_diff) { + ddp->dirty_counter = ERTS_DRIFT_INTERVALS; + begin_short_intervals = 1; } - diff_time = (curr + hr_correction - hr_init_time) / 1000; - } else if (diffdiff < -10000) { - SysHrTime corr = (curr - hr_last_time) / 100; - if (corr / 1000 >= -diffdiff) { - ++done; - hr_correction -= ((SysHrTime)diffdiff) * 1000; - } else { - hr_correction += corr; + else { + if (ddp->dirty_counter <= 0) { + drift_adj = ((stime_acc - mtime_acc) + *ERTS_MONOTONIC_TIME_UNIT) / mtime_acc; + } + if (ddp->dirty_counter >= 0) { + if (ddp->dirty_counter == 0) { + /* Force set new drift correction... */ + set_new_correction = 1; + } + ddp->dirty_counter--; + } + drift_adj_diff = drift_adj - new_correction.drift; + if (drift_adj_diff) { + if (drift_adj_diff > ERTS_TIME_DRIFT_MAX_ADJ_DIFF) + drift_adj_diff = ERTS_TIME_DRIFT_MAX_ADJ_DIFF; + else if (drift_adj_diff < -ERTS_TIME_DRIFT_MAX_ADJ_DIFF) + drift_adj_diff = -ERTS_TIME_DRIFT_MAX_ADJ_DIFF; + new_correction.drift += drift_adj_diff; + + if (drift_adj_diff < -ERTS_TIME_DRIFT_MIN_ADJ_DIFF + || ERTS_TIME_DRIFT_MIN_ADJ_DIFF < drift_adj_diff) { + set_new_correction = 1; + } + } } - diff_time = (curr + hr_correction - hr_init_time) / 1000; - } else { - ++done; } - if (done) { - hr_last_correction_check = curr; + } +#endif + + begin_short_intervals |= set_new_correction; + + if (begin_short_intervals) { + time_sup.inf.c.parmon.cdata.short_check_interval + = ERTS_INIT_SHORT_INTERVAL_COUNTER; + } + else if ((os_mtime - time_sup.inf.c.parmon.cdata.last_check + >= ERTS_SHORT_TIME_CORRECTION_CHECK - ERTS_MONOTONIC_TIME_UNIT) + && time_sup.inf.c.parmon.cdata.short_check_interval > 0) { + time_sup.inf.c.parmon.cdata.short_check_interval--; + } + time_sup.inf.c.parmon.cdata.last_check = os_mtime; + + if (new_correction.error == 0) + timeout = ERTS_MONOTONIC_TO_MSEC(ERTS_LONG_TIME_CORRECTION_CHECK); + else { + ErtsMonotonicTime ecorr = new_correction.error; + if (sdiff < 0) + sdiff = -1*sdiff; + if (ecorr < 0) + ecorr = -1*ecorr; + if (sdiff > ecorr*(ERTS_LONG_TIME_CORRECTION_CHECK/ERTS_TCORR_ERR_UNIT)) + timeout = ERTS_MONOTONIC_TO_MSEC(ERTS_LONG_TIME_CORRECTION_CHECK); + else { + timeout = ERTS_MONOTONIC_TO_MSEC((ERTS_TCORR_ERR_UNIT*sdiff)/ecorr); + if (timeout < 10) + timeout = 10; } } - tv->tv_sec += (int) (diff_time / ((SysHrTime) 1000000)); - tv->tv_usec += (int) (diff_time % ((SysHrTime) 1000000)); - if (tv->tv_usec >= 1000000) { - tv->tv_usec -= 1000000; - tv->tv_sec += 1; + + if (timeout > ERTS_MONOTONIC_TO_MSEC(ERTS_SHORT_TIME_CORRECTION_CHECK) + && time_sup.inf.c.parmon.cdata.short_check_interval) { + timeout = ERTS_MONOTONIC_TO_MSEC(ERTS_SHORT_TIME_CORRECTION_CHECK); + } + + ERTS_PRINT_CORRECTION; + + if (set_new_correction) { + erts_smp_rwmtx_rwlock(&time_sup.inf.c.parmon.rwmtx); + + os_mtime = erts_os_monotonic_time(); + + /* Save previous correction instance */ + time_sup.inf.c.parmon.cdata.prev = *cip; + + /* + * Current correction instance begin when + * OS monotonic time has increased one unit. + */ + os_mtime++; + + /* + * Erlang monotonic time corresponding to + * next OS monotonic time using previous + * correction. + */ + erl_mtime = calc_corrected_erl_mtime(os_mtime, cip, NULL); + + /* + * Save new current correction instance. + */ + time_sup.inf.c.parmon.cdata.curr.erl_mtime = erl_mtime; + time_sup.inf.c.parmon.cdata.curr.os_mtime = os_mtime; + time_sup.inf.c.parmon.cdata.curr.correction = new_correction; + + erts_smp_rwmtx_rwunlock(&time_sup.inf.c.parmon.rwmtx); } - hr_last_time = curr; + + erts_set_timer(&time_sup.inf.c.parmon.timer, + check_time_correction, + NULL, + NULL, + timeout); + +#undef ERTS_PRINT_CORRECTION } -#define correction (hr_correction/1000000) +#ifndef ERTS_HAVE_CORRECTED_OS_MONOTONIC -#else /* !HAVE_GETHRTIME */ -#if !defined(CORRECT_USING_TIMES) -#define init_tolerant_timeofday() -#define get_tolerant_timeofday(tvp) sys_gettimeofday(tvp) -#else +static void +init_check_time_correction(void *unused) +{ + ErtsMonotonicDriftData *ddp; + ErtsMonotonicTime old_mtime, old_stime, mtime, stime, mtime_diff, + stime_diff; + int ix; + + ddp = &time_sup.inf.c.parmon.cdata.drift; + ix = ddp->ix; + old_mtime = ddp->intervals[0].time.mon; + old_stime = ddp->intervals[0].time.sys; + + erts_os_times(&mtime, &stime); + + mtime_diff = mtime - old_mtime; + stime_diff = stime - old_stime; + if (100*stime_diff < 80*mtime_diff || 120*mtime_diff < 100*stime_diff ) { + /* Had a system time leap... pretend no drift... */ + stime_diff = mtime_diff; + } + + /* + * We use old time values in order to trigger + * a drift adjustment, and repeat this interval + * in all slots... + */ + for (ix = 0; ix < ERTS_DRIFT_INTERVALS; ix++) { + ddp->intervals[ix].diff.mon = mtime_diff; + ddp->intervals[ix].diff.sys = stime_diff; + ddp->intervals[ix].time.mon = old_mtime; + ddp->intervals[ix].time.sys = old_stime; + } -typedef Sint64 Milli; - -static clock_t init_ct; -static Sint64 ct_wrap; -static Milli init_tv_m; -static Milli correction_supress; -static Milli last_ct_diff; -static Milli last_cc; -static clock_t last_ct; - -/* sys_times() might need to be wrapped and the values shifted (right) - a bit to cope with newer linux (2.5.*) kernels, this has to be taken care - of dynamically to start with, a special version that uses - the times() return value as a high resolution timer can be made - to fully utilize the faster ticks, like on windows, but for now, we'll - settle with this silly workaround */ -#ifdef ERTS_WRAP_SYS_TIMES -#define KERNEL_TICKS() (sys_times_wrap() & \ - ((1UL << ((sizeof(clock_t) * 8) - 1)) - 1)) -#else -SysTimes dummy_tms; + ddp->acc.sys = stime_diff*ERTS_DRIFT_INTERVALS; + ddp->acc.mon = mtime_diff*ERTS_DRIFT_INTERVALS; + ddp->ix = 0; + ddp->dirty_counter = ERTS_DRIFT_INTERVALS; -#define KERNEL_TICKS() (sys_times(&dummy_tms) & \ - ((1UL << ((sizeof(clock_t) * 8) - 1)) - 1)) + check_time_correction(NULL); +} #endif -static void init_tolerant_timeofday(void) +static ErtsMonotonicTime +finalize_corrected_time_offset(ErtsSystemTime *stimep) { - last_ct = init_ct = KERNEL_TICKS(); - last_cc = 0; - init_tv_m = (((Milli) inittv.tv_sec) * 1000) + - (inittv.tv_usec / 1000); - ct_wrap = 0; - correction_supress = 0; -} + ErtsMonotonicTime os_mtime; + ErtsMonotonicCorrectionData cdata; + ErtsMonotonicCorrectionInstance *cip; + erts_smp_rwmtx_rlock(&time_sup.inf.c.parmon.rwmtx); + + erts_os_times(&os_mtime, stimep); + + cdata = time_sup.inf.c.parmon.cdata; + + erts_smp_rwmtx_runlock(&time_sup.inf.c.parmon.rwmtx); + + if (os_mtime < cdata.curr.os_mtime) + erl_exit(ERTS_ABORT_EXIT, + "OS monotonic time stepped backwards\n"); + cip = &cdata.curr; + + return calc_corrected_erl_mtime(os_mtime, cip, NULL); +} -static void get_tolerant_timeofday(SysTimeval *tvp) +static void +late_init_time_correction(void) { - clock_t current_ct; - SysTimeval current_tv; - Milli ct_diff; - Milli tv_diff; - Milli current_correction; - Milli act_correction; /* long shown to be too small */ - Milli max_adjust; - - if (erts_disable_tolerant_timeofday) { - sys_gettimeofday(tvp); - return; - } + if (time_sup.inf.c.finalized_offset) { -#ifdef ERTS_WRAP_SYS_TIMES -#define TICK_MS (1000 / SYS_CLK_TCK_WRAP) + erts_init_timer(&time_sup.inf.c.parmon.timer); + erts_set_timer(&time_sup.inf.c.parmon.timer, +#ifndef ERTS_HAVE_CORRECTED_OS_MONOTONIC + init_check_time_correction, #else -#define TICK_MS (1000 / SYS_CLK_TCK) + check_time_correction, #endif - current_ct = KERNEL_TICKS(); - sys_gettimeofday(¤t_tv); - - /* I dont know if uptime can move some units backwards - on some systems, but I allow for small backward - jumps to avoid such problems if they exist...*/ - if (last_ct > 100 && current_ct < (last_ct - 100)) { - ct_wrap += ((Sint64) 1) << ((sizeof(clock_t) * 8) - 1); + NULL, + NULL, + ERTS_MONOTONIC_TO_MSEC(ERTS_SHORT_TIME_CORRECTION_CHECK)); } - last_ct = current_ct; - ct_diff = ((ct_wrap + current_ct) - init_ct) * TICK_MS; +} - /* - * We will adjust the time in milliseconds and we allow for 1% - * adjustments, but if this function is called more often then every 100 - * millisecond (which is obviously possible), we will never adjust, so - * we accumulate small times by setting last_ct_diff iff max_adjust > 0 - */ - if ((max_adjust = (ct_diff - last_ct_diff)/100) > 0) - last_ct_diff = ct_diff; +#endif /* ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT */ - tv_diff = ((((Milli) current_tv.tv_sec) * 1000) + - (current_tv.tv_usec / 1000)) - init_tv_m; +static ErtsMonotonicTime get_not_corrected_time(void) +{ + ErtsMonotonicTime stime, mtime; - current_correction = ((ct_diff - tv_diff) / TICK_MS) * TICK_MS; /* trunc */ + erts_smp_mtx_lock(&erts_get_time_mtx); - /* - * We allow the current_correction value to wobble a little, as it - * suffers from the low resolution of the kernel ticks. - * if it hasn't changed more than one tick in either direction, - * we will keep the old value. - */ - if ((last_cc > current_correction + TICK_MS) || - (last_cc < current_correction - TICK_MS)) { - last_cc = current_correction; - } else { - current_correction = last_cc; - } - - /* - * As time goes, we try to get the actual correction to 0, - * that is, make erlangs time correspond to the systems dito. - * The act correction is what we seem to need (current_correction) - * minus the correction suppression. The correction supression - * will change slowly (max 1% of elapsed time) but in millisecond steps. - */ - act_correction = current_correction - correction_supress; - if (max_adjust > 0) { - /* - * Here we slowly adjust erlangs time to correspond with the - * system time by changing the correction_supress variable. - * It can change max_adjust milliseconds which is 1% of elapsed time - */ - if (act_correction > 0) { - if (current_correction - correction_supress > max_adjust) { - correction_supress += max_adjust; - } else { - correction_supress = current_correction; - } - act_correction = current_correction - correction_supress; - } else if (act_correction < 0) { - if (correction_supress - current_correction > max_adjust) { - correction_supress -= max_adjust; - } else { - correction_supress = current_correction; + stime = erts_os_system_time(); + + mtime = stime - time_sup.inf.c.not_corrected_moffset; + + if (mtime >= time_sup.f.c.last_not_corrected_time) + time_sup.f.c.last_not_corrected_time = mtime; + else { + mtime = time_sup.f.c.last_not_corrected_time; + + if (time_sup.r.o.warp_mode == ERTS_MULTI_TIME_WARP_MODE) { + ErtsMonotonicTime new_offset = stime - mtime; + new_offset = ERTS_MONOTONIC_TO_USEC(new_offset); + new_offset = ERTS_USEC_TO_MONOTONIC(new_offset); + if (time_sup.inf.c.not_corrected_moffset != new_offset) { + time_sup.inf.c.not_corrected_moffset = new_offset; + set_time_offset(new_offset); + schedule_send_time_offset_changed_notifications(new_offset); } - act_correction = current_correction - correction_supress; } + } - /* - * The actual correction will correct the timeval so that system - * time warps gets smothed down. - */ - current_tv.tv_sec += act_correction / 1000; - current_tv.tv_usec += (act_correction % 1000) * 1000; - - if (current_tv.tv_usec >= 1000000) { - ++current_tv.tv_sec ; - current_tv.tv_usec -= 1000000; - } else if (current_tv.tv_usec < 0) { - --current_tv.tv_sec; - current_tv.tv_usec += 1000000; - } - *tvp = current_tv; -#undef TICK_MS + + ASSERT(stime == mtime + time_sup.inf.c.not_corrected_moffset); + + erts_smp_mtx_unlock(&erts_get_time_mtx); + + return mtime; } -#endif /* CORRECT_USING_TIMES */ -#endif /* !HAVE_GETHRTIME */ +int erts_check_time_adj_support(int time_correction, + ErtsTimeWarpMode time_warp_mode) +{ + if (!time_correction) + return 1; + + /* User wants time correction */ -/* -** Why this? Well, most platforms have a constant clock resolution of 1, -** we dont want the deliver_time/time_remaining routines to waste -** time dividing and multiplying by/with a variable that's always one. -** so the return value of sys_init_time is ignored on those platforms. -*/ - -#ifndef SYS_CLOCK_RESOLUTION -static int clock_resolution; -#define CLOCK_RESOLUTION clock_resolution +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + return !time_sup.r.o.os_monotonic_time_disable; #else -#define CLOCK_RESOLUTION SYS_CLOCK_RESOLUTION + return 0; #endif +} -/* -** The clock resolution should really be the resolution of the -** time function in use, which on most platforms -** is 1. On VxWorks the resolution should be -** the number of ticks per second (or 1, which would work nicely to). -** -** Setting lower resolutions is mostly interesting when timers are used -** instead of something like select. -*/ - -static SysTimeval last_delivered; - -static void init_erts_deliver_time(const SysTimeval *inittv) +int +erts_has_time_correction(void) { - /* We set the initial values for deliver_time here */ - last_delivered = *inittv; - last_delivered.tv_usec = 1000 * (last_delivered.tv_usec / 1000); - /* ms resolution */ + return time_sup.r.o.correction; } -static void do_erts_deliver_time(const SysTimeval *current) +void erts_init_sys_time_sup(void) { - SysTimeval cur_time; - erts_time_t elapsed; - - /* calculate and deliver appropriate number of ticks */ - cur_time = *current; - cur_time.tv_usec = 1000 * (cur_time.tv_usec / 1000); /* ms resolution */ - elapsed = (1000 * (cur_time.tv_sec - last_delivered.tv_sec) + - (cur_time.tv_usec - last_delivered.tv_usec) / 1000) / - CLOCK_RESOLUTION; + ErtsSysInitTimeResult sys_init_time_res + = ERTS_SYS_INIT_TIME_RESULT_INITER; - /* Sometimes the time jump backwards, - resulting in a negative elapsed time. We compensate for - this by simply pretend as if the time stood still. :) */ + sys_init_time(&sys_init_time_res); - if (elapsed > 0) { + erts_time_sup__.r.o.monotonic_time_unit + = sys_init_time_res.os_monotonic_time_unit; - ASSERT(elapsed < ((erts_time_t) ERTS_SHORT_TIME_T_MAX)); +#ifndef SYS_CLOCK_RESOLUTION + erts_time_sup__.r.o.clktck_resolution + = sys_init_time_res.sys_clock_resolution; + erts_time_sup__.r.o.clktck_resolution *= 1000; +#endif - erts_do_time_add((erts_short_time_t) elapsed); - last_delivered = cur_time; - } +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + time_sup.r.o.os_monotonic_time_disable + = !sys_init_time_res.have_os_monotonic_time; + time_sup.r.o.os_monotonic_time_func + = sys_init_time_res.os_monotonic_time_info.func; + time_sup.r.o.os_monotonic_time_clock_id + = sys_init_time_res.os_monotonic_time_info.clock_id; + time_sup.r.o.os_monotonic_time_locked + = sys_init_time_res.os_monotonic_time_info.locked_use; + time_sup.r.o.os_monotonic_time_resolution + = sys_init_time_res.os_monotonic_time_info.resolution; + time_sup.r.o.os_monotonic_time_extended + = sys_init_time_res.os_monotonic_time_info.extended; +#endif + time_sup.r.o.os_system_time_func + = sys_init_time_res.os_system_time_info.func; + time_sup.r.o.os_system_time_clock_id + = sys_init_time_res.os_system_time_info.clock_id; + time_sup.r.o.os_system_time_locked + = sys_init_time_res.os_system_time_info.locked_use; + time_sup.r.o.os_system_time_resolution + = sys_init_time_res.os_system_time_info.resolution; } int -erts_init_time_sup(void) +erts_init_time_sup(int time_correction, ErtsTimeWarpMode time_warp_mode) { + ErtsMonotonicTime resolution; +#if !ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT + ErtsMonotonicTime abs_native_offset, native_offset; +#endif + + ASSERT(ERTS_MONOTONIC_TIME_MIN < ERTS_MONOTONIC_TIME_MAX); + erts_smp_mtx_init(&erts_timeofday_mtx, "timeofday"); + erts_smp_mtx_init(&erts_get_time_mtx, "get_time"); - init_approx_time(); + time_sup.r.o.correction = time_correction; + time_sup.r.o.warp_mode = time_warp_mode; + + if (time_warp_mode == ERTS_SINGLE_TIME_WARP_MODE) + time_sup.inf.c.finalized_offset = 0; + else + time_sup.inf.c.finalized_offset = ~0; + +#if !ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT + +#ifdef ARCH_32 + time_sup.r.o.start = ((((ErtsMonotonicTime) 1) << 32)-1); + time_sup.r.o.start /= ERTS_MONOTONIC_TIME_UNIT; + time_sup.r.o.start *= ERTS_MONOTONIC_TIME_UNIT; + time_sup.r.o.start += ERTS_MONOTONIC_TIME_UNIT; + native_offset = time_sup.r.o.start - ERTS_MONOTONIC_TIME_UNIT; + native_offset = native_offset; +#else /* ARCH_64 */ + if (ERTS_MONOTONIC_TIME_UNIT <= 10*1000*1000) { + time_sup.r.o.start = 0; + native_offset = -ERTS_MONOTONIC_TIME_UNIT; + abs_native_offset = ERTS_MONOTONIC_TIME_UNIT; + } + else { + time_sup.r.o.start = ((ErtsMonotonicTime) MIN_SMALL); + time_sup.r.o.start /= ERTS_MONOTONIC_TIME_UNIT; + time_sup.r.o.start *= ERTS_MONOTONIC_TIME_UNIT; + native_offset = time_sup.r.o.start - ERTS_MONOTONIC_TIME_UNIT; + abs_native_offset = -1*native_offset; + } +#endif - last_emu_time.tv_sec = 0; - last_emu_time.tv_usec = 0; + time_sup.r.o.start_offset.native = (time_sup.r.o.start + - ERTS_MONOTONIC_TIME_UNIT); + time_sup.r.o.start_offset.nsec = (ErtsMonotonicTime) + erts_time_unit_conversion((Uint64) abs_native_offset, + (Uint32) ERTS_MONOTONIC_TIME_UNIT, + (Uint32) 1000*1000*1000); + time_sup.r.o.start_offset.usec = (ErtsMonotonicTime) + erts_time_unit_conversion((Uint64) abs_native_offset, + (Uint32) ERTS_MONOTONIC_TIME_UNIT, + (Uint32) 1000*1000); + time_sup.r.o.start_offset.msec = (ErtsMonotonicTime) + erts_time_unit_conversion((Uint64) abs_native_offset, + (Uint32) ERTS_MONOTONIC_TIME_UNIT, + (Uint32) 1000); + time_sup.r.o.start_offset.sec = (ErtsMonotonicTime) + erts_time_unit_conversion((Uint64) abs_native_offset, + (Uint32) ERTS_MONOTONIC_TIME_UNIT, + (Uint32) 1); + if (native_offset < 0) { + time_sup.r.o.start_offset.nsec *= -1; + time_sup.r.o.start_offset.usec *= -1; + time_sup.r.o.start_offset.msec *= -1; + time_sup.r.o.start_offset.sec *= -1; + } -#ifndef SYS_CLOCK_RESOLUTION - clock_resolution = sys_init_time(); -#else - (void) sys_init_time(); #endif - sys_gettimeofday(&inittv); + + resolution = time_sup.r.o.os_system_time_resolution; +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + if (resolution > time_sup.r.o.os_monotonic_time_resolution) + resolution = time_sup.r.o.os_monotonic_time_resolution; +#endif + + time_sup.r.o.adj.large_diff = erts_time_sup__.r.o.monotonic_time_unit; + time_sup.r.o.adj.large_diff *= 50; + time_sup.r.o.adj.large_diff /= resolution; + if (time_sup.r.o.adj.large_diff < ERTS_USEC_TO_MONOTONIC(500)) + time_sup.r.o.adj.large_diff = ERTS_USEC_TO_MONOTONIC(500); + time_sup.r.o.adj.small_diff = time_sup.r.o.adj.large_diff/10; + +#ifdef ERTS_TIME_CORRECTION_PRINT + fprintf(stderr, "start = %lld\n\r", (long long) ERTS_MONOTONIC_TIME_START); + fprintf(stderr, "native offset = %lld\n\r", (long long) ERTS_MONOTONIC_OFFSET_NATIVE); + fprintf(stderr, "nsec offset = %lld\n\r", (long long) ERTS_MONOTONIC_OFFSET_NSEC); + fprintf(stderr, "usec offset = %lld\n\r", (long long) ERTS_MONOTONIC_OFFSET_USEC); + fprintf(stderr, "msec offset = %lld\n\r", (long long) ERTS_MONOTONIC_OFFSET_MSEC); + fprintf(stderr, "sec offset = %lld\n\r", (long long) ERTS_MONOTONIC_OFFSET_SEC); + fprintf(stderr, "large diff = %lld usec\r\n", + (long long) ERTS_MONOTONIC_TO_USEC(time_sup.r.o.adj.large_diff)); + fprintf(stderr, "small diff = %lld usec\r\n", + (long long) ERTS_MONOTONIC_TO_USEC(time_sup.r.o.adj.small_diff)); +#endif + + if (ERTS_MONOTONIC_TIME_UNIT < ERTS_CLKTCK_RESOLUTION) + ERTS_INTERNAL_ERROR("Too small monotonic time time unit"); + +#ifndef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + time_sup.r.o.correction = 0; +#else + if (time_sup.r.o.os_monotonic_time_disable) + time_sup.r.o.correction = 0; + + if (time_sup.r.o.correction) { + ErtsMonotonicCorrectionData *cdatap; + erts_smp_rwmtx_opt_t rwmtx_opts = ERTS_SMP_RWMTX_OPT_DEFAULT_INITER; + ErtsMonotonicTime offset; + erts_os_times(&time_sup.inf.c.minit, + &time_sup.inf.c.sinit); + time_sup.r.o.moffset = -1*time_sup.inf.c.minit; + offset = time_sup.inf.c.sinit; + offset -= ERTS_MONOTONIC_TIME_UNIT; + init_time_offset(offset); + + rwmtx_opts.type = ERTS_SMP_RWMTX_TYPE_EXTREMELY_FREQUENT_READ; + rwmtx_opts.lived = ERTS_SMP_RWMTX_LONG_LIVED; + + erts_smp_rwmtx_init_opt(&time_sup.inf.c.parmon.rwmtx, + &rwmtx_opts, "get_corrected_time"); + + cdatap = &time_sup.inf.c.parmon.cdata; -#ifdef HAVE_GETHRTIME - sys_init_hrtime(); +#ifndef ERTS_HAVE_CORRECTED_OS_MONOTONIC + cdatap->drift.intervals[0].time.sys = time_sup.inf.c.sinit; + cdatap->drift.intervals[0].time.mon = time_sup.inf.c.minit; + cdatap->curr.correction.drift = 0; #endif - init_tolerant_timeofday(); + cdatap->curr.correction.error = 0; + cdatap->curr.erl_mtime = ERTS_MONOTONIC_TIME_UNIT; + cdatap->curr.os_mtime = time_sup.inf.c.minit; + cdatap->last_check = time_sup.inf.c.minit; + cdatap->short_check_interval = ERTS_INIT_SHORT_INTERVAL_COUNTER; + cdatap->prev = cdatap->curr; + + time_sup.r.o.get_time = get_corrected_time; + } + else +#endif + { + ErtsMonotonicTime stime, offset; + time_sup.r.o.get_time = get_not_corrected_time; + stime = time_sup.inf.c.sinit = erts_os_system_time(); + offset = stime - ERTS_MONOTONIC_TIME_UNIT; + time_sup.inf.c.not_corrected_moffset = offset; + init_time_offset(offset); + time_sup.f.c.last_not_corrected_time = 0; + } - init_erts_deliver_time(&inittv); - gtv = inittv; - then.tv_sec = then.tv_usec = 0; + prev_wall_clock_elapsed = 0; - erts_deliver_time(); + previous_now = ERTS_MONOTONIC_TO_USEC(get_time_offset()); - return CLOCK_RESOLUTION; +#ifdef DEBUG + time_sup_initialized = 1; +#endif + + return ERTS_CLKTCK_RESOLUTION/1000; } + +void +erts_late_init_time_sup(void) +{ +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + /* Timer wheel must be initialized */ + if (time_sup.r.o.get_time == get_corrected_time) + late_init_time_correction(); +#endif + erts_late_sys_init_time(); +} + +ErtsTimeWarpMode erts_time_warp_mode(void) +{ + return time_sup.r.o.warp_mode; +} + +ErtsTimeOffsetState erts_time_offset_state(void) +{ + switch (time_sup.r.o.warp_mode) { + case ERTS_NO_TIME_WARP_MODE: + return ERTS_TIME_OFFSET_FINAL; + case ERTS_SINGLE_TIME_WARP_MODE: + if (time_sup.inf.c.finalized_offset) + return ERTS_TIME_OFFSET_FINAL; + return ERTS_TIME_OFFSET_PRELIMINARY; + case ERTS_MULTI_TIME_WARP_MODE: + return ERTS_TIME_OFFSET_VOLATILE; + default: + ERTS_INTERNAL_ERROR("Invalid time warp mode"); + return ERTS_TIME_OFFSET_VOLATILE; + } +} + +/* + * erts_finalize_time_offset() will only change time offset + * the first time it is called when the emulator has been + * started in "single time warp" mode. Returns previous + * state: + * * ERTS_TIME_OFFSET_PRELIMINARY - Finalization performed + * * ERTS_TIME_OFFSET_FINAL - Already finialized; nothing changed + * * ERTS_TIME_OFFSET_VOLATILE - Not supported, either in + * * no correction mode (or multi time warp mode; not yet implemented). + */ + +ErtsTimeOffsetState +erts_finalize_time_offset(void) +{ + switch (time_sup.r.o.warp_mode) { + case ERTS_NO_TIME_WARP_MODE: + return ERTS_TIME_OFFSET_FINAL; + case ERTS_MULTI_TIME_WARP_MODE: + return ERTS_TIME_OFFSET_VOLATILE; + case ERTS_SINGLE_TIME_WARP_MODE: { + ErtsTimeOffsetState res = ERTS_TIME_OFFSET_FINAL; + + erts_smp_mtx_lock(&erts_get_time_mtx); + + if (!time_sup.inf.c.finalized_offset) { + ErtsMonotonicTime mtime, new_offset; + +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + if (!time_sup.r.o.correction) +#endif + { + ErtsMonotonicTime stime = erts_os_system_time(); + + mtime = stime - time_sup.inf.c.not_corrected_moffset; + + if (mtime >= time_sup.f.c.last_not_corrected_time) { + time_sup.f.c.last_not_corrected_time = mtime; + new_offset = time_sup.inf.c.not_corrected_moffset; + } + else { + mtime = time_sup.f.c.last_not_corrected_time; + + ASSERT(time_sup.inf.c.not_corrected_moffset != stime - mtime); + new_offset = stime - mtime; + time_sup.inf.c.not_corrected_moffset = new_offset; + } + + } +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + else { + ErtsSystemTime stime; + mtime = finalize_corrected_time_offset(&stime); + new_offset = stime - mtime; + } +#endif + new_offset = ERTS_MONOTONIC_TO_USEC(new_offset); + new_offset = ERTS_USEC_TO_MONOTONIC(new_offset); + + set_time_offset(new_offset); + schedule_send_time_offset_changed_notifications(new_offset); + + time_sup.inf.c.finalized_offset = ~0; + res = ERTS_TIME_OFFSET_PRELIMINARY; + } + + erts_smp_mtx_unlock(&erts_get_time_mtx); + +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + if (res == ERTS_TIME_OFFSET_PRELIMINARY + && time_sup.r.o.get_time == get_corrected_time) { + late_init_time_correction(); + } +#endif + + return res; + } + default: + ERTS_INTERNAL_ERROR("Invalid time warp mode"); + return ERTS_TIME_OFFSET_VOLATILE; + } +} + /* info functions */ void @@ -498,23 +1171,16 @@ elapsed_time_both(UWord *ms_user, UWord *ms_sys, void wall_clock_elapsed_time_both(UWord *ms_total, UWord *ms_diff) { - UWord prev_total; - SysTimeval tv; + ErtsMonotonicTime now, elapsed; erts_smp_mtx_lock(&erts_timeofday_mtx); - get_tolerant_timeofday(&tv); + now = time_sup.r.o.get_time(); - *ms_total = 1000 * (tv.tv_sec - inittv.tv_sec) + - (tv.tv_usec - inittv.tv_usec) / 1000; - - prev_total = 1000 * (gtv.tv_sec - inittv.tv_sec) + - (gtv.tv_usec - inittv.tv_usec) / 1000; - *ms_diff = *ms_total - prev_total; - gtv = tv; - - /* must sync the machine's idea of time here */ - do_erts_deliver_time(&tv); + elapsed = ERTS_MONOTONIC_TO_MSEC(now); + *ms_total = (UWord) elapsed; + *ms_diff = (UWord) (elapsed - prev_wall_clock_elapsed); + prev_wall_clock_elapsed = elapsed; erts_smp_mtx_unlock(&erts_timeofday_mtx); } @@ -890,146 +1556,555 @@ univ_to_local(Sint *year, Sint *month, Sint *day, return 0; } - /* get a timestamp */ void get_now(Uint* megasec, Uint* sec, Uint* microsec) { - SysTimeval now; + ErtsMonotonicTime now_megasec, now_sec, now, mtime, time_offset; + mtime = time_sup.r.o.get_time(); + time_offset = get_time_offset(); + now = ERTS_MONOTONIC_TO_USEC(mtime + time_offset); + erts_smp_mtx_lock(&erts_timeofday_mtx); - - get_tolerant_timeofday(&now); - do_erts_deliver_time(&now); - - /* Make sure time is later than last */ - if (then.tv_sec > now.tv_sec || - (then.tv_sec == now.tv_sec && then.tv_usec >= now.tv_usec)) { - now = then; - now.tv_usec++; - } - /* Check for carry from above + general reasonability */ - if (now.tv_usec >= 1000000) { - now.tv_usec = 0; - now.tv_sec++; - } - then = now; + + /* Make sure now time is later than last time */ + if (now <= previous_now) + now = previous_now + 1; + + previous_now = now; erts_smp_mtx_unlock(&erts_timeofday_mtx); - - *megasec = (Uint) (now.tv_sec / 1000000); - *sec = (Uint) (now.tv_sec % 1000000); - *microsec = (Uint) (now.tv_usec); - update_approx_time(&now); + now_megasec = now / ERTS_MONOTONIC_TIME_TERA; + now_sec = now / ERTS_MONOTONIC_TIME_MEGA; + *megasec = (Uint) now_megasec; + *sec = (Uint) (now_sec - now_megasec*ERTS_MONOTONIC_TIME_MEGA); + *microsec = (Uint) (now - now_sec*ERTS_MONOTONIC_TIME_MEGA); + + ASSERT(((ErtsMonotonicTime) *megasec)*ERTS_MONOTONIC_TIME_TERA + + ((ErtsMonotonicTime) *sec)*ERTS_MONOTONIC_TIME_MEGA + + ((ErtsMonotonicTime) *microsec) == now); +} + +ErtsMonotonicTime +erts_get_monotonic_time(void) +{ + return time_sup.r.o.get_time(); } void get_sys_now(Uint* megasec, Uint* sec, Uint* microsec) { - SysTimeval now; - - sys_gettimeofday(&now); - - *megasec = (Uint) (now.tv_sec / 1000000); - *sec = (Uint) (now.tv_sec % 1000000); - *microsec = (Uint) (now.tv_usec); + ErtsSystemTime stime = erts_os_system_time(); + ErtsSystemTime ms, s, us; + + us = ERTS_MONOTONIC_TO_USEC(stime); + s = us / (1000*1000); + ms = s / (1000*1000); - update_approx_time(&now); + *megasec = (Uint) ms; + *sec = (Uint) (s - ms*(1000*1000)); + *microsec = (Uint) (us - s*(1000*1000)); } +#ifdef HAVE_ERTS_NOW_CPU +void erts_get_now_cpu(Uint* megasec, Uint* sec, Uint* microsec) { + SysCpuTime t; + SysTimespec tp; + + sys_get_proc_cputime(t, tp); + *microsec = (Uint)(tp.tv_nsec / 1000); + t = (tp.tv_sec / 1000000); + *megasec = (Uint)(t % 1000000); + *sec = (Uint)(tp.tv_sec % 1000000); +} +#endif -/* deliver elapsed *ticks* to the machine - takes a pointer - to a struct timeval representing current time (to save - a gettimeofday() where possible) or NULL */ +#include "big.h" -void erts_deliver_time(void) { - SysTimeval now; - - erts_smp_mtx_lock(&erts_timeofday_mtx); - - get_tolerant_timeofday(&now); - do_erts_deliver_time(&now); - - erts_smp_mtx_unlock(&erts_timeofday_mtx); +void +erts_monitor_time_offset(Eterm id, Eterm ref) +{ + erts_smp_mtx_lock(&erts_get_time_mtx); + erts_add_monitor(&time_offset_monitors, MON_TIME_OFFSET, ref, id, NIL); + no_time_offset_monitors++; + erts_smp_mtx_unlock(&erts_get_time_mtx); +} - update_approx_time(&now); +int +erts_demonitor_time_offset(Eterm ref) +{ + int res; + ErtsMonitor *mon; + ASSERT(is_internal_ref(ref)); + erts_smp_mtx_lock(&erts_get_time_mtx); + mon = erts_remove_monitor(&time_offset_monitors, ref); + if (!mon) + res = 0; + else { + ASSERT(no_time_offset_monitors > 0); + no_time_offset_monitors--; + res = 1; + } + erts_smp_mtx_unlock(&erts_get_time_mtx); + if (res) + erts_destroy_monitor(mon); + return res; } -/* get *real* time (not ticks) remaining until next timeout - if there - isn't one, give a "long" time, that is guaranteed - to not cause overflow when we report elapsed time later on */ +typedef struct { + Eterm pid; + Eterm ref; + Eterm heap[REF_THING_SIZE]; +} ErtsTimeOffsetMonitorInfo; -void erts_time_remaining(SysTimeval *rem_time) +typedef struct { + Uint ix; + ErtsTimeOffsetMonitorInfo *to_mon_info; +} ErtsTimeOffsetMonitorContext; + +static void +save_time_offset_monitor(ErtsMonitor *mon, void *vcntxt) { - erts_time_t ticks; - SysTimeval cur_time; - erts_time_t elapsed; - - /* erts_next_time() returns no of ticks to next timeout or -1 if none */ - - ticks = (erts_time_t) erts_next_time(); - if (ticks == (erts_time_t) -1) { - /* timer queue empty */ - /* this will cause at most 100000000 ticks */ - rem_time->tv_sec = 100000; - rem_time->tv_usec = 0; - } else { - /* next timeout after ticks ticks */ - ticks *= CLOCK_RESOLUTION; - - erts_smp_mtx_lock(&erts_timeofday_mtx); - - get_tolerant_timeofday(&cur_time); - cur_time.tv_usec = 1000 * - (cur_time.tv_usec / 1000);/* ms resolution*/ - elapsed = 1000 * (cur_time.tv_sec - last_delivered.tv_sec) + - (cur_time.tv_usec - last_delivered.tv_usec) / 1000; - - erts_smp_mtx_unlock(&erts_timeofday_mtx); + ErtsTimeOffsetMonitorContext *cntxt; + Eterm *from_hp, *to_hp; + Uint mix; + int hix; + + cntxt = (ErtsTimeOffsetMonitorContext *) vcntxt; + mix = (cntxt->ix)++; + cntxt->to_mon_info[mix].pid = mon->pid; + to_hp = &cntxt->to_mon_info[mix].heap[0]; + + ASSERT(is_internal_ref(mon->ref)); + from_hp = internal_ref_val(mon->ref); + ASSERT(thing_arityval(*from_hp) + 1 == REF_THING_SIZE); + + for (hix = 0; hix < REF_THING_SIZE; hix++) + to_hp[hix] = from_hp[hix]; + + cntxt->to_mon_info[mix].ref + = make_internal_ref(&cntxt->to_mon_info[mix].heap[0]); + +} + +static void +send_time_offset_changed_notifications(void *new_offsetp) +{ + ErtsMonotonicTime new_offset; + ErtsTimeOffsetMonitorInfo *to_mon_info = NULL; /* Shut up faulty warning */ + Uint no_monitors; + char *tmp = NULL; + +#ifdef ARCH_64 + new_offset = (ErtsMonotonicTime) new_offsetp; +#else + new_offset = *((ErtsMonotonicTime *) new_offsetp); + erts_free(ERTS_ALC_T_NEW_TIME_OFFSET, new_offsetp); +#endif + new_offset -= ERTS_MONOTONIC_OFFSET_NATIVE; + + erts_smp_mtx_lock(&erts_get_time_mtx); + + no_monitors = no_time_offset_monitors; + if (no_monitors) { + ErtsTimeOffsetMonitorContext cntxt; + Uint alloc_sz; - if (ticks <= elapsed) { /* Ooops, better hurry */ - rem_time->tv_sec = rem_time->tv_usec = 0; - return; + /* Monitor info array size */ + alloc_sz = no_monitors*sizeof(ErtsTimeOffsetMonitorInfo); + /* + template max size */ + alloc_sz += 6*sizeof(Eterm); /* 5-tuple */ + alloc_sz += ERTS_MAX_SINT64_HEAP_SIZE*sizeof(Eterm); /* max offset size */ + tmp = erts_alloc(ERTS_ALC_T_TMP, alloc_sz); + + to_mon_info = (ErtsTimeOffsetMonitorInfo *) tmp; + cntxt.ix = 0; + cntxt.to_mon_info = to_mon_info; + + erts_doforall_monitors(time_offset_monitors, + save_time_offset_monitor, + &cntxt); + + ASSERT(cntxt.ix == no_monitors); + } + + erts_smp_mtx_unlock(&erts_get_time_mtx); + + if (no_monitors) { + Eterm *hp, *patch_refp, new_offset_term, message_template; + Uint mix, hsz; + + /* Make message template */ + + hp = (Eterm *) (tmp + no_monitors*sizeof(ErtsTimeOffsetMonitorInfo)); + + hsz = 6; /* 5-tuple */ + hsz += REF_THING_SIZE; + hsz += ERTS_SINT64_HEAP_SIZE(new_offset); + + if (IS_SSMALL(new_offset)) + new_offset_term = make_small(new_offset); + else + new_offset_term = erts_sint64_to_big(new_offset, &hp); + message_template = TUPLE5(hp, + am_CHANGE, + THE_NON_VALUE, /* Patch point for ref */ + am_time_offset, + am_clock_service, + new_offset_term); + patch_refp = &hp[2]; + + ASSERT(*patch_refp == THE_NON_VALUE); + + for (mix = 0; mix < no_monitors; mix++) { + Process *rp = erts_proc_lookup(to_mon_info[mix].pid); + if (rp) { + Eterm ref = to_mon_info[mix].ref; + ErtsProcLocks rp_locks = ERTS_PROC_LOCK_LINK; + erts_smp_proc_lock(rp, ERTS_PROC_LOCK_LINK); + if (erts_lookup_monitor(ERTS_P_MONITORS(rp), ref)) { + ErlHeapFragment *bp; + ErlOffHeap *ohp; + Eterm message; + + hp = erts_alloc_message_heap(hsz, &bp, &ohp, rp, &rp_locks); + *patch_refp = ref; + ASSERT(hsz == size_object(message_template)); + message = copy_struct(message_template, hsz, &hp, ohp); + erts_queue_message(rp, &rp_locks, bp, message, NIL +#ifdef USE_VM_PROBES + , NIL +#endif + ); + } + erts_smp_proc_unlock(rp, rp_locks); + } } - rem_time->tv_sec = (ticks - elapsed) / 1000; - rem_time->tv_usec = 1000 * ((ticks - elapsed) % 1000); + + erts_free(ERTS_ALC_T_TMP, tmp); } } -void erts_get_timeval(SysTimeval *tv) +static void +schedule_send_time_offset_changed_notifications(ErtsMonotonicTime new_offset) { - erts_smp_mtx_lock(&erts_timeofday_mtx); - get_tolerant_timeofday(tv); - erts_smp_mtx_unlock(&erts_timeofday_mtx); - update_approx_time(tv); +#ifdef ARCH_64 + void *new_offsetp = (void *) new_offset; + ASSERT(sizeof(void *) == sizeof(ErtsMonotonicTime)); +#else + void *new_offsetp = erts_alloc(ERTS_ALC_T_NEW_TIME_OFFSET, + sizeof(ErtsMonotonicTime)); + *((ErtsMonotonicTime *) new_offsetp) = new_offset; +#endif + erts_schedule_misc_aux_work(1, + send_time_offset_changed_notifications, + new_offsetp); } -erts_time_t -erts_get_time(void) +static ERTS_INLINE Eterm +make_time_val(Process *c_p, ErtsMonotonicTime time_val) { - SysTimeval sys_tv; - - erts_smp_mtx_lock(&erts_timeofday_mtx); - - get_tolerant_timeofday(&sys_tv); - - erts_smp_mtx_unlock(&erts_timeofday_mtx); + Sint64 val = (Sint64) time_val; + Eterm *hp; + Uint sz; - update_approx_time(&sys_tv); + if (IS_SSMALL(val)) + return make_small(val); - return sys_tv.tv_sec; + sz = ERTS_SINT64_HEAP_SIZE(val); + hp = HAlloc(c_p, sz); + return erts_sint64_to_big(val, &hp); } -#ifdef HAVE_ERTS_NOW_CPU -void erts_get_now_cpu(Uint* megasec, Uint* sec, Uint* microsec) { - SysCpuTime t; - SysTimespec tp; +Eterm +erts_get_monotonic_start_time(struct process *c_p) +{ + return make_time_val(c_p, ERTS_MONOTONIC_TIME_START); +} - sys_get_proc_cputime(t, tp); - *microsec = (Uint)(tp.tv_nsec / 1000); - t = (tp.tv_sec / 1000000); - *megasec = (Uint)(t % 1000000); - *sec = (Uint)(tp.tv_sec % 1000000); +static Eterm +bld_monotonic_time_source(Uint **hpp, Uint *szp, Sint64 os_mtime) +{ +#ifndef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + return NIL; +#else + int i = 0; + Eterm k[6]; + Eterm v[6]; + + if (time_sup.r.o.os_monotonic_time_disable) + return NIL; + + k[i] = erts_bld_atom(hpp, szp, "function"); + v[i++] = erts_bld_atom(hpp, szp, + time_sup.r.o.os_monotonic_time_func); + + if (time_sup.r.o.os_monotonic_time_clock_id) { + k[i] = erts_bld_atom(hpp, szp, "clock_id"); + v[i++] = erts_bld_atom(hpp, szp, + time_sup.r.o.os_monotonic_time_clock_id); + } + + k[i] = erts_bld_atom(hpp, szp, "resolution"); + v[i++] = erts_bld_uint64(hpp, szp, + time_sup.r.o.os_monotonic_time_resolution); + + k[i] = erts_bld_atom(hpp, szp, "extended"); + v[i++] = time_sup.r.o.os_monotonic_time_extended ? am_yes : am_no; + + k[i] = erts_bld_atom(hpp, szp, "parallel"); + v[i++] = time_sup.r.o.os_monotonic_time_locked ? am_no : am_yes; + + k[i] = erts_bld_atom(hpp, szp, "time"); + v[i++] = erts_bld_sint64(hpp, szp, os_mtime); + + return erts_bld_2tup_list(hpp, szp, (Sint) i, k, v); +#endif +} + +Eterm +erts_monotonic_time_source(struct process *c_p) +{ + Uint hsz = 0; + Eterm *hp = NULL; + Sint64 os_mtime = 0; +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + if (!time_sup.r.o.os_monotonic_time_disable) + os_mtime = (Sint64) erts_os_monotonic_time(); +#endif + + bld_monotonic_time_source(NULL, &hsz, os_mtime); + if (hsz) + hp = HAlloc(c_p, hsz); + return bld_monotonic_time_source(&hp, NULL, os_mtime); +} + +static Eterm +bld_system_time_source(Uint **hpp, Uint *szp, Sint64 os_stime) +{ + int i = 0; + Eterm k[5]; + Eterm v[5]; + + k[i] = erts_bld_atom(hpp, szp, "function"); + v[i++] = erts_bld_atom(hpp, szp, + time_sup.r.o.os_system_time_func); + + if (time_sup.r.o.os_system_time_clock_id) { + k[i] = erts_bld_atom(hpp, szp, "clock_id"); + v[i++] = erts_bld_atom(hpp, szp, + time_sup.r.o.os_system_time_clock_id); + } + + k[i] = erts_bld_atom(hpp, szp, "resolution"); + v[i++] = erts_bld_uint64(hpp, szp, + time_sup.r.o.os_system_time_resolution); + + k[i] = erts_bld_atom(hpp, szp, "parallel"); + v[i++] = am_yes; + + k[i] = erts_bld_atom(hpp, szp, "time"); + v[i++] = erts_bld_sint64(hpp, szp, os_stime); + + return erts_bld_2tup_list(hpp, szp, (Sint) i, k, v); } + +Eterm +erts_system_time_source(struct process *c_p) +{ + Uint hsz = 0; + Eterm *hp = NULL; + Sint64 os_stime = (Sint64) erts_os_system_time(); + + bld_system_time_source(NULL, &hsz, os_stime); + if (hsz) + hp = HAlloc(c_p, hsz); + return bld_system_time_source(&hp, NULL, os_stime); +} + + +#include "bif.h" + +static ERTS_INLINE Eterm +time_unit_conversion(Process *c_p, Eterm term, ErtsMonotonicTime val, ErtsMonotonicTime muloff) +{ + ErtsMonotonicTime result; + BIF_RETTYPE ret; + + if (val < 0) + goto trap_to_erlang_code; + + /* Convert to common user specified time units */ + switch (term) { + case am_seconds: + case make_small(1): + result = ERTS_MONOTONIC_TO_SEC(val) + muloff*ERTS_MONOTONIC_OFFSET_SEC; + ERTS_BIF_PREP_RET(ret, make_time_val(c_p, result)); + break; + case am_milli_seconds: + case make_small(1000): + result = ERTS_MONOTONIC_TO_MSEC(val) + muloff*ERTS_MONOTONIC_OFFSET_MSEC; + ERTS_BIF_PREP_RET(ret, make_time_val(c_p, result)); + break; + case am_micro_seconds: + case make_small(1000*1000): + result = ERTS_MONOTONIC_TO_USEC(val) + muloff*ERTS_MONOTONIC_OFFSET_USEC; + ERTS_BIF_PREP_RET(ret, make_time_val(c_p, result)); + break; +#ifdef ARCH_64 + case am_nano_seconds: + case make_small(1000*1000*1000): + result = ERTS_MONOTONIC_TO_NSEC(val) + muloff*ERTS_MONOTONIC_OFFSET_NSEC; + ERTS_BIF_PREP_RET(ret, make_time_val(c_p, result)); + break; +#endif + default: { + Eterm value, native_res; +#ifndef ARCH_64 + Sint user_res; + if (term == am_nano_seconds) + goto to_nano_seconds; + if (term_to_Sint(term, &user_res)) { + if (user_res == 1000*1000*1000) { + to_nano_seconds: + result = (ERTS_MONOTONIC_TO_NSEC(val) + + muloff*ERTS_MONOTONIC_OFFSET_NSEC); + ERTS_BIF_PREP_RET(ret, make_time_val(c_p, result)); + break; + } + if (user_res <= 0) + goto badarg; + } +#else + if (is_small(term)) { + if (signed_val(term) <= 0) + goto badarg; + } #endif + else if (is_big(term)) { + if (big_sign(term)) + goto badarg; + } + else { + badarg: + ERTS_BIF_PREP_ERROR(ret, c_p, BADARG); + break; + } + + trap_to_erlang_code: + /* Do it in erlang code instead; pass along values to use... */ + value = make_time_val(c_p, val + muloff*ERTS_MONOTONIC_OFFSET_NATIVE); + native_res = make_time_val(c_p, ERTS_MONOTONIC_TIME_UNIT); + + ERTS_BIF_PREP_TRAP3(ret, erts_convert_time_unit_trap, c_p, + value, native_res, term); + + break; + } + } + + return ret; +} + +/* Built in functions */ + +BIF_RETTYPE monotonic_time_0(BIF_ALIST_0) +{ + ErtsMonotonicTime mtime = time_sup.r.o.get_time(); + mtime += ERTS_MONOTONIC_OFFSET_NATIVE; + BIF_RET(make_time_val(BIF_P, mtime)); +} + +BIF_RETTYPE monotonic_time_1(BIF_ALIST_1) +{ + BIF_RET(time_unit_conversion(BIF_P, BIF_ARG_1, time_sup.r.o.get_time(), 1)); +} + +BIF_RETTYPE system_time_0(BIF_ALIST_0) +{ + ErtsMonotonicTime mtime, offset; + mtime = time_sup.r.o.get_time(); + offset = get_time_offset(); + BIF_RET(make_time_val(BIF_P, mtime + offset)); +} + +BIF_RETTYPE system_time_1(BIF_ALIST_0) +{ + ErtsMonotonicTime mtime, offset; + mtime = time_sup.r.o.get_time(); + offset = get_time_offset(); + BIF_RET(time_unit_conversion(BIF_P, BIF_ARG_1, mtime + offset, 0)); +} + +BIF_RETTYPE erts_internal_time_unit_0(BIF_ALIST_0) +{ + BIF_RET(make_time_val(BIF_P, ERTS_MONOTONIC_TIME_UNIT)); +} + +BIF_RETTYPE time_offset_0(BIF_ALIST_0) +{ + ErtsMonotonicTime time_offset = get_time_offset(); + time_offset -= ERTS_MONOTONIC_OFFSET_NATIVE; + BIF_RET(make_time_val(BIF_P, time_offset)); +} + +BIF_RETTYPE time_offset_1(BIF_ALIST_1) +{ + BIF_RET(time_unit_conversion(BIF_P, BIF_ARG_1, get_time_offset(), -1)); +} + + +BIF_RETTYPE timestamp_0(BIF_ALIST_0) +{ + Eterm *hp, res; + ErtsMonotonicTime stime, mtime, all_sec, offset; + Uint mega_sec, sec, micro_sec; + + mtime = time_sup.r.o.get_time(); + offset = get_time_offset(); + stime = ERTS_MONOTONIC_TO_USEC(mtime + offset); + all_sec = stime / ERTS_MONOTONIC_TIME_MEGA; + mega_sec = (Uint) (stime / ERTS_MONOTONIC_TIME_TERA); + sec = (Uint) (all_sec - (((ErtsMonotonicTime) mega_sec) + * ERTS_MONOTONIC_TIME_MEGA)); + micro_sec = (Uint) (stime - all_sec*ERTS_MONOTONIC_TIME_MEGA); + + ASSERT(((ErtsMonotonicTime) mega_sec)*ERTS_MONOTONIC_TIME_TERA + + ((ErtsMonotonicTime) sec)*ERTS_MONOTONIC_TIME_MEGA + + micro_sec == stime); + + /* + * Mega seconds is the only value that potentially + * ever could be a bignum. However, that wont happen + * during at least the next 4 million years... + * + * (System time will also have wrapped in the + * 64-bit integer before we get there...) + */ + + ASSERT(IS_USMALL(0, mega_sec)); + ASSERT(IS_USMALL(0, sec)); + ASSERT(IS_USMALL(0, micro_sec)); + + hp = HAlloc(BIF_P, 4); + res = TUPLE3(hp, + make_small(mega_sec), + make_small(sec), + make_small(micro_sec)); + BIF_RET(res); +} + +BIF_RETTYPE os_system_time_0(BIF_ALIST_0) +{ + ErtsSystemTime stime = erts_os_system_time(); + BIF_RET(make_time_val(BIF_P, stime)); +} + +BIF_RETTYPE os_system_time_1(BIF_ALIST_0) +{ + ErtsSystemTime stime = erts_os_system_time(); + BIF_RET(time_unit_conversion(BIF_P, BIF_ARG_1, stime, 0)); +} + diff --git a/erts/emulator/beam/erl_trace.c b/erts/emulator/beam/erl_trace.c index ea5c850a30..2f9969b0e7 100644 --- a/erts/emulator/beam/erl_trace.c +++ b/erts/emulator/beam/erl_trace.c @@ -2225,7 +2225,7 @@ trace_gc(Process *p, Eterm what) Eterm* limit; #endif - ASSERT(sizeof(values)/sizeof(*values) == sizeof(tags)/sizeof(Eterm)); + ERTS_CT_ASSERT(sizeof(values)/sizeof(*values) == sizeof(tags)/sizeof(Eterm)); UseTmpHeap(LOCAL_HEAP_SIZE,p); @@ -3492,16 +3492,13 @@ init_sys_msg_dispatcher(void) thr_opts.coreNo = 0; #endif thr_opts.detached = 1; + thr_opts.name = "sys_msg_dispatcher"; init_smq_element_alloc(); sys_message_queue = NULL; sys_message_queue_end = NULL; erts_smp_cnd_init(&smq_cnd); erts_smp_mtx_init(&smq_mtx, "sys_msg_q"); -#ifdef ETHR_HAVE_THREAD_NAMES - thr_opts.name = "sys_msg_dispatcher"; -#endif - erts_smp_thr_create(&sys_msg_dispatcher_tid, sys_msg_dispatcher_func, NULL, diff --git a/erts/emulator/beam/erl_unicode.c b/erts/emulator/beam/erl_unicode.c index 3a968594f3..f8e1431a53 100644 --- a/erts/emulator/beam/erl_unicode.c +++ b/erts/emulator/beam/erl_unicode.c @@ -2126,6 +2126,8 @@ Eterm erts_convert_native_to_filename(Process *p, byte *bytes) mac = 1; case ERL_FILENAME_UTF8: size = strlen((char *) bytes); + if (size == 0) + return NIL; if (erts_analyze_utf8(bytes,size,&err_pos,&num_chars,NULL) != ERTS_UTF8_OK) { goto noconvert; } diff --git a/erts/emulator/beam/erl_utils.h b/erts/emulator/beam/erl_utils.h index 0807649ea1..7cb8972e29 100644 --- a/erts/emulator/beam/erl_utils.h +++ b/erts/emulator/beam/erl_utils.h @@ -32,11 +32,7 @@ typedef struct { #endif union { Uint64 not_atomic; -#ifdef ARCH_64 - erts_atomic_t atomic; -#else - erts_dw_atomic_t atomic; -#endif + erts_atomic64_t atomic; } counter; } erts_interval_t; @@ -50,9 +46,6 @@ Uint64 erts_ensure_later_interval_nob(erts_interval_t *, Uint64); Uint64 erts_ensure_later_interval_acqb(erts_interval_t *, Uint64); Uint64 erts_smp_ensure_later_interval_nob(erts_interval_t *, Uint64); Uint64 erts_smp_ensure_later_interval_acqb(erts_interval_t *, Uint64); -#ifdef ARCH_32 -ERTS_GLB_INLINE Uint64 erts_interval_dw_aint_to_val__(erts_dw_aint_t *); -#endif ERTS_GLB_INLINE Uint64 erts_current_interval_nob__(erts_interval_t *); ERTS_GLB_INLINE Uint64 erts_current_interval_acqb__(erts_interval_t *); ERTS_GLB_INLINE Uint64 erts_current_interval_nob(erts_interval_t *); @@ -62,46 +55,16 @@ ERTS_GLB_INLINE Uint64 erts_smp_current_interval_acqb(erts_interval_t *); #if ERTS_GLB_INLINE_INCL_FUNC_DEF -#ifdef ARCH_32 - -ERTS_GLB_INLINE Uint64 -erts_interval_dw_aint_to_val__(erts_dw_aint_t *dw) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw->dw_sint; -#else - Uint64 res; - res = (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - -#endif - ERTS_GLB_INLINE Uint64 erts_current_interval_nob__(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t dw; - erts_dw_atomic_read_nob(&icp->counter.atomic, &dw); - return erts_interval_dw_aint_to_val__(&dw); -#endif + return (Uint64) erts_atomic64_read_nob(&icp->counter.atomic); } ERTS_GLB_INLINE Uint64 erts_current_interval_acqb__(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_read_acqb(&icp->counter.atomic); -#else - erts_dw_aint_t dw; - erts_dw_atomic_read_acqb(&icp->counter.atomic, &dw); - return erts_interval_dw_aint_to_val__(&dw); -#endif + return (Uint64) erts_atomic64_read_acqb(&icp->counter.atomic); } ERTS_GLB_INLINE Uint64 @@ -150,12 +113,14 @@ void erts_silence_warn_unused_result(long unused); int erts_fit_in_bits_int64(Sint64); int erts_fit_in_bits_int32(Sint32); +int erts_fit_in_bits_uint(Uint); int erts_list_length(Eterm); int erts_is_builtin(Eterm, Eterm, int); Uint32 make_broken_hash(Eterm); Uint32 block_hash(byte *, unsigned, Uint32); Uint32 make_hash2(Eterm); Uint32 make_hash(Eterm); +Uint32 make_internal_hash(Eterm); void erts_save_emu_args(int argc, char **argv); Eterm erts_get_emu_args(struct process *c_p); diff --git a/erts/emulator/beam/erl_vm.h b/erts/emulator/beam/erl_vm.h index b7de8208ad..3a9fb1e07b 100644 --- a/erts/emulator/beam/erl_vm.h +++ b/erts/emulator/beam/erl_vm.h @@ -20,8 +20,6 @@ #ifndef __ERL_VM_H__ #define __ERL_VM_H__ -/* #define ERTS_OPCODE_COUNTER_SUPPORT */ - /* FORCE_HEAP_FRAGS: * Debug provocation to make HAlloc always create heap fragments (if allowed) * even if there is room on heap. @@ -119,9 +117,9 @@ #if defined(DEBUG) || defined(CHECK_FOR_HOLES) #if HALFWORD_HEAP -# define ERTS_HOLE_MARKER (0xaf5e78ccU) +# define ERTS_HOLE_MARKER (0xdeadbeef) #else -# define ERTS_HOLE_MARKER (((0xaf5e78ccUL << 24) << 8) | 0xaf5e78ccUL) +# define ERTS_HOLE_MARKER (((0xdeadbeef << 24) << 8) | 0xdeadbeef) #endif #endif @@ -174,6 +172,7 @@ extern int H_MIN_SIZE; /* minimum (heap + stack) */ extern int BIN_VH_MIN_SIZE; /* minimum virtual (bin) heap */ extern int erts_atom_table_size;/* Atom table size */ +extern int erts_pd_initial_size;/* Initial Process dictionary table size */ #define ORIG_CREATION 0 #define INTERNAL_CREATION 255 diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c index 656de7c49a..2117dbec62 100644 --- a/erts/emulator/beam/external.c +++ b/erts/emulator/beam/external.c @@ -36,7 +36,9 @@ #include "erl_process.h" #include "error.h" #include "external.h" +#define ERL_WANT_HIPE_BIF_WRAPPER__ #include "bif.h" +#undef ERL_WANT_HIPE_BIF_WRAPPER__ #include "big.h" #include "dist.h" #include "erl_binary.h" @@ -44,9 +46,6 @@ #include "erl_zlib.h" #include "erl_map.h" -#ifdef HIPE -#include "hipe_mode_switch.h" -#endif #define in_area(ptr,start,nbytes) ((UWord)((char*)(ptr) - (char*)(start)) < (nbytes)) #define MAX_STRING_LEN 0xffff @@ -111,26 +110,17 @@ static int encode_size_struct_int(struct TTBSizeContext_*, ErtsAtomCacheMap *acm static Export binary_to_term_trap_export; static BIF_RETTYPE binary_to_term_trap_1(BIF_ALIST_1); -static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* context_b); +static BIF_RETTYPE binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* context_b, + Export *bif, Eterm arg0, Eterm arg1); void erts_init_external(void) { -#if 1 /* In R16 */ erts_init_trap_export(&term_to_binary_trap_export, - am_erlang, am_term_to_binary_trap, 1, + am_erts_internal, am_term_to_binary_trap, 1, &term_to_binary_trap_1); erts_init_trap_export(&binary_to_term_trap_export, - am_erlang, am_binary_to_term_trap, 1, + am_erts_internal, am_binary_to_term_trap, 1, &binary_to_term_trap_1); -#else - sys_memset((void *) &term_to_binary_trap_export, 0, sizeof(Export)); - term_to_binary_trap_export.address = &term_to_binary_trap_export.code[3]; - term_to_binary_trap_export.code[0] = am_erlang; - term_to_binary_trap_export.code[1] = am_term_to_binary_trap; - term_to_binary_trap_export.code[2] = 1; - term_to_binary_trap_export.code[3] = (BeamInstr) em_apply_bif; - term_to_binary_trap_export.code[4] = (BeamInstr) &term_to_binary_trap_1; -#endif return; } @@ -510,15 +500,37 @@ byte *erts_encode_ext_dist_header_finalize(byte *ext, ErtsAtomCache *cache, Uint return ep; } -Uint erts_encode_dist_ext_size(Eterm term, Uint32 flags, ErtsAtomCacheMap *acmp) +int erts_encode_dist_ext_size(Eterm term, Uint32 flags, ErtsAtomCacheMap *acmp, + Uint* szp) { - Uint sz = 0; + Uint sz; + if (encode_size_struct_int(NULL, acmp, term, flags, NULL, &sz)) { + return -1; + } else { #ifndef ERTS_DEBUG_USE_DIST_SEP - if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) + if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) #endif - sz++ /* VERSION_MAGIC */; - sz += encode_size_struct2(acmp, term, flags); - return sz; + sz++ /* VERSION_MAGIC */; + + *szp += sz; + return 0; + } +} + +int erts_encode_dist_ext_size_int(Eterm term, struct erts_dsig_send_context* ctx, Uint* szp) +{ + Uint sz; + if (encode_size_struct_int(&ctx->u.sc, ctx->acmp, term, ctx->flags, &ctx->reds, &sz)) { + return -1; + } else { +#ifndef ERTS_DEBUG_USE_DIST_SEP + if (!(ctx->flags & DFLAG_DIST_HDR_ATOM_CACHE)) +#endif + sz++ /* VERSION_MAGIC */; + + *szp += sz; + return 0; + } } Uint erts_encode_ext_size(Eterm term) @@ -539,19 +551,16 @@ Uint erts_encode_ext_size_ets(Eterm term) } -void erts_encode_dist_ext(Eterm term, byte **ext, Uint32 flags, ErtsAtomCacheMap *acmp) +int erts_encode_dist_ext(Eterm term, byte **ext, Uint32 flags, ErtsAtomCacheMap *acmp, + TTBEncodeContext* ctx, Sint* reds) { - byte *ep = *ext; -#ifndef ERTS_DEBUG_USE_DIST_SEP - if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) -#endif - *ep++ = VERSION_MAGIC; - ep = enc_term(acmp, term, ep, flags, NULL); - if (!ep) - erl_exit(ERTS_ABORT_EXIT, - "%s:%d:erts_encode_dist_ext(): Internal data structure error\n", - __FILE__, __LINE__); - *ext = ep; + if (!ctx || !ctx->wstack.wstart) { + #ifndef ERTS_DEBUG_USE_DIST_SEP + if (!(flags & DFLAG_DIST_HDR_ATOM_CACHE)) + #endif + *(*ext)++ = VERSION_MAGIC; + } + return enc_term_int(ctx, acmp, term, *ext, flags, NULL, reds, ext); } void erts_encode_ext(Eterm term, byte **ext) @@ -1069,6 +1078,8 @@ static BIF_RETTYPE term_to_binary_trap_1(BIF_ALIST_1) } } +HIPE_WRAPPER_BIF_DISABLE_GC(term_to_binary, 1) + BIF_RETTYPE term_to_binary_1(BIF_ALIST_1) { Eterm res = erts_term_to_binary_int(BIF_P, BIF_ARG_1, 0, TERM_TO_BINARY_DFLAGS, NULL); @@ -1081,6 +1092,8 @@ BIF_RETTYPE term_to_binary_1(BIF_ALIST_1) } } +HIPE_WRAPPER_BIF_DISABLE_GC(term_to_binary, 2) + BIF_RETTYPE term_to_binary_2(BIF_ALIST_2) { Process* p = BIF_P; @@ -1169,7 +1182,8 @@ typedef struct { Eterm* hp_end; int remaining_n; char* remaining_bytes; - Eterm* maps_head; + Eterm* maps_list; + struct dec_term_hamt_placeholder* hamt_list; } B2TDecodeContext; typedef struct { @@ -1185,6 +1199,8 @@ typedef struct B2TContext_t { Uint32 flags; SWord reds; Eterm trap_bin; + Export *bif; + Eterm arg[2]; enum B2TState state; union { B2TSizeContext sc; @@ -1356,7 +1372,8 @@ static BIF_RETTYPE binary_to_term_trap_1(BIF_ALIST_1) Binary *context_bin = ((ProcBin *) binary_val(BIF_ARG_1))->val; ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(context_bin) == b2t_context_destructor); - return binary_to_term_int(BIF_P, 0, THE_NON_VALUE, context_bin); + return binary_to_term_int(BIF_P, 0, THE_NON_VALUE, context_bin, NULL, + THE_NON_VALUE, THE_NON_VALUE); } @@ -1391,8 +1408,10 @@ static B2TContext* b2t_export_context(Process* p, B2TContext* src) return ctx; } -static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* context_b) +static BIF_RETTYPE binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* context_b, + Export *bif_init, Eterm arg0, Eterm arg1) { + BIF_RETTYPE ret_val; #ifdef EXTREME_B2T_TRAPPING SWord initial_reds = 1 + b2t_rand() % 4; #else @@ -1409,6 +1428,9 @@ static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* con ctx->state = B2TPrepare; ctx->aligned_alloc = NULL; ctx->flags = flags; + ctx->bif = bif_init; + ctx->arg[0] = arg0; + ctx->arg[1] = arg1; IF_DEBUG(ctx->trap_bin = THE_NON_VALUE;) } else { is_first_call = 0; @@ -1487,7 +1509,8 @@ static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* con ctx->u.dc.hp_start = HAlloc(p, ctx->heap_size); ctx->u.dc.hp = ctx->u.dc.hp_start; ctx->u.dc.hp_end = ctx->u.dc.hp_start + ctx->heap_size; - ctx->u.dc.maps_head = NULL; + ctx->u.dc.maps_list = NULL; + ctx->u.dc.hamt_list = NULL; ctx->state = B2TDecode; /*fall through*/ case B2TDecode: @@ -1504,12 +1527,24 @@ static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* con HRelease(p, ctx->u.dc.hp_end, ctx->u.dc.hp_start); /*fall through*/ case B2TBadArg: - b2t_destroy_context(ctx); - if (!is_first_call) { - erts_set_gc_state(p, 1); - } BUMP_REDS(p, (initial_reds - ctx->reds) / B2T_BYTES_PER_REDUCTION); - BIF_ERROR(p, BADARG & ~EXF_SAVETRACE); + + ASSERT(ctx->bif == bif_export[BIF_binary_to_term_1] + || ctx->bif == bif_export[BIF_binary_to_term_2]); + + if (is_first_call) + ERTS_BIF_PREP_ERROR(ret_val, p, BADARG); + else { + erts_set_gc_state(p, 1); + if (is_non_value(ctx->arg[1])) + ERTS_BIF_PREP_ERROR_TRAPPED1(ret_val, p, BADARG, ctx->bif, + ctx->arg[0]); + else + ERTS_BIF_PREP_ERROR_TRAPPED2(ret_val, p, BADARG, ctx->bif, + ctx->arg[0], ctx->arg[1]); + } + b2t_destroy_context(ctx); + return ret_val; case B2TDone: b2t_destroy_context(ctx); @@ -1524,7 +1559,8 @@ static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* con erts_set_gc_state(p, 1); } BUMP_REDS(p, (initial_reds - ctx->reds) / B2T_BYTES_PER_REDUCTION); - return ctx->u.dc.res; + ERTS_BIF_PREP_RET(ret_val, ctx->u.dc.res); + return ret_val; default: ASSERT(!"Unknown state in binary_to_term"); @@ -1541,15 +1577,24 @@ static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* con erts_set_gc_state(p, 0); } BUMP_ALL_REDS(p); - BIF_TRAP1(&binary_to_term_trap_export, p, ctx->trap_bin); + + ERTS_BIF_PREP_TRAP1(ret_val, &binary_to_term_trap_export, + p, ctx->trap_bin); + + return ret_val; } -BIF_RETTYPE erts_internal_binary_to_term_1(BIF_ALIST_1) +HIPE_WRAPPER_BIF_DISABLE_GC(binary_to_term, 1) + +BIF_RETTYPE binary_to_term_1(BIF_ALIST_1) { - return binary_to_term_int(BIF_P, 0, BIF_ARG_1, NULL); + return binary_to_term_int(BIF_P, 0, BIF_ARG_1, NULL, bif_export[BIF_binary_to_term_1], + BIF_ARG_1, THE_NON_VALUE); } -BIF_RETTYPE erts_internal_binary_to_term_2(BIF_ALIST_2) +HIPE_WRAPPER_BIF_DISABLE_GC(binary_to_term, 2) + +BIF_RETTYPE binary_to_term_2(BIF_ALIST_2) { Eterm opts; Eterm opt; @@ -1570,7 +1615,8 @@ BIF_RETTYPE erts_internal_binary_to_term_2(BIF_ALIST_2) if (is_not_nil(opts)) goto error; - return binary_to_term_int(BIF_P, flags, BIF_ARG_1, NULL); + return binary_to_term_int(BIF_P, flags, BIF_ARG_1, NULL, bif_export[BIF_binary_to_term_2], + BIF_ARG_1, BIF_ARG_2); error: BIF_ERROR(BIF_P, BADARG); @@ -1717,54 +1763,14 @@ erts_term_to_binary(Process* p, Eterm Term, int level, Uint flags) { return erts_term_to_binary_simple(p, Term, size, level, flags); } -/* Define for testing */ -/* #define EXTREME_TTB_TRAPPING 1 */ +/* Define EXTREME_TTB_TRAPPING for testing in dist.h */ #ifndef EXTREME_TTB_TRAPPING -#define TERM_TO_BINARY_LOOP_FACTOR 32 #define TERM_TO_BINARY_COMPRESS_CHUNK (1 << 18) #else -#define TERM_TO_BINARY_LOOP_FACTOR 1 #define TERM_TO_BINARY_COMPRESS_CHUNK 10 #endif - - -typedef enum { TTBSize, TTBEncode, TTBCompress } TTBState; -typedef struct TTBSizeContext_ { - Uint flags; - int level; - Uint result; - Eterm obj; - ErtsEStack estack; -} TTBSizeContext; - -typedef struct TTBEncodeContext_ { - Uint flags; - int level; - byte* ep; - Eterm obj; - ErtsWStack wstack; - Binary *result_bin; -} TTBEncodeContext; - -typedef struct { - Uint real_size; - Uint dest_len; - byte *dbytes; - Binary *result_bin; - Binary *destination_bin; - z_stream stream; -} TTBCompressContext; - -typedef struct { - int alive; - TTBState state; - union { - TTBSizeContext sc; - TTBEncodeContext ec; - TTBCompressContext cc; - } s; -} TTBContext; +#define TERM_TO_BINARY_MEMCPY_FACTOR 8 static void ttb_context_destructor(Binary *context_bin) { @@ -1876,8 +1882,6 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla } result_bin = erts_bin_nrml_alloc(size); - result_bin->flags = 0; - result_bin->orig_size = size; erts_refc_init(&result_bin->refc, 0); result_bin->orig_bytes[0] = VERSION_MAGIC; /* Next state immediately, no need to export context */ @@ -1938,8 +1942,6 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla context->s.cc.result_bin = result_bin; result_bin = erts_bin_nrml_alloc(real_size); - result_bin->flags = 0; - result_bin->orig_size = real_size; erts_refc_init(&result_bin->refc, 0); result_bin->orig_bytes[0] = VERSION_MAGIC; @@ -2302,8 +2304,10 @@ dec_pid(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, Ete #define ENC_TERM ((Eterm) 0) #define ENC_ONE_CONS ((Eterm) 1) #define ENC_PATCH_FUN_SIZE ((Eterm) 2) -#define ENC_LAST_ARRAY_ELEMENT ((Eterm) 3) - +#define ENC_BIN_COPY ((Eterm) 3) +#define ENC_MAP_PAIR ((Eterm) 4) +#define ENC_HASHMAP_NODE ((Eterm) 5) +#define ENC_LAST_ARRAY_ELEMENT ((Eterm) 6) static byte* enc_term(ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dflags, @@ -2339,6 +2343,9 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, WSTACK_RESTORE(s, &ctx->wstack); ep = ctx->ep; obj = ctx->obj; + if (is_non_value(obj)) { + goto outer_loop; + } } } @@ -2362,8 +2369,8 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, obj = CAR(cons); tl = CDR(cons); - WSTACK_PUSH(s, is_list(tl) ? ENC_ONE_CONS : ENC_TERM); - WSTACK_PUSH(s, tl); + WSTACK_PUSH2(s, (is_list(tl) ? ENC_ONE_CONS : ENC_TERM), + tl); } break; case ENC_PATCH_FUN_SIZE: @@ -2376,6 +2383,46 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, put_int32(ep - size_p, size_p); } goto outer_loop; + case ENC_BIN_COPY: { + Uint bits = (Uint)obj; + Uint bitoffs = WSTACK_POP(s); + byte* bytes = (byte*) WSTACK_POP(s); + byte* dst = (byte*) WSTACK_POP(s); + if (bits > r * (TERM_TO_BINARY_MEMCPY_FACTOR * 8)) { + Uint n = r * TERM_TO_BINARY_MEMCPY_FACTOR; + WSTACK_PUSH5(s, (UWord)(dst + n), (UWord)(bytes + n), bitoffs, + ENC_BIN_COPY, bits - 8*n); + bits = 8*n; + copy_binary_to_buffer(dst, 0, bytes, bitoffs, bits); + obj = THE_NON_VALUE; + r = 0; /* yield */ + break; + } else { + copy_binary_to_buffer(dst, 0, bytes, bitoffs, bits); + r -= bits / (TERM_TO_BINARY_MEMCPY_FACTOR * 8); + goto outer_loop; + } + } + case ENC_MAP_PAIR: { + Uint pairs_left = obj; + Eterm *vptr = (Eterm*) WSTACK_POP(s); + Eterm *kptr = (Eterm*) WSTACK_POP(s); + + obj = *kptr; + if (--pairs_left > 0) { + WSTACK_PUSH4(s, (UWord)(kptr+1), (UWord)(vptr+1), + ENC_MAP_PAIR, pairs_left); + } + WSTACK_PUSH2(s, ENC_TERM, *vptr); + break; + } + case ENC_HASHMAP_NODE: + if (is_list(obj)) { /* leaf node [K|V] */ + ptr = list_val(obj); + WSTACK_PUSH2(s, ENC_TERM, CDR(ptr)); + obj = CAR(ptr); + } + break; case ENC_LAST_ARRAY_ELEMENT: /* obj is the tuple */ { @@ -2394,17 +2441,16 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, #else Eterm* ptr = (Eterm *) obj; #endif - WSTACK_PUSH(s, val-1); obj = *ptr++; - WSTACK_PUSH(s, (UWord)ptr); + WSTACK_PUSH2(s, val-1, (UWord)ptr); } break; } L_jump_start: - if (ctx && --r == 0) { - *reds = r; + if (ctx && --r <= 0) { + *reds = 0; ctx->obj = obj; ctx->ep = ep; WSTACK_SAVE(s, &ctx->wstack); @@ -2553,39 +2599,59 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, ep += 4; } if (i > 0) { - WSTACK_PUSH(s, ENC_LAST_ARRAY_ELEMENT+i-1); - WSTACK_PUSH(s, (UWord)ptr); + WSTACK_PUSH2(s, ENC_LAST_ARRAY_ELEMENT+i-1, (UWord)ptr); } break; case MAP_DEF: - { - map_t *mp = (map_t*)map_val(obj); - Uint size = map_get_size(mp); + if (is_flatmap(obj)) { + flatmap_t *mp = (flatmap_t*)flatmap_val(obj); + Uint size = flatmap_get_size(mp); *ep++ = MAP_EXT; put_int32(size, ep); ep += 4; if (size > 0) { - Eterm *kptr = map_get_keys(mp); - Eterm *vptr = map_get_values(mp); - - for (i = size-1; i >= 1; i--) { - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) vptr[i]); - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) kptr[i]); - } + Eterm *kptr = flatmap_get_keys(mp); + Eterm *vptr = flatmap_get_values(mp); - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) vptr[0]); + WSTACK_PUSH4(s, (UWord)kptr, (UWord)vptr, + ENC_MAP_PAIR, size); + } + } else { + Eterm hdr; + Uint node_sz; + ptr = boxed_val(obj); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + *ep++ = MAP_EXT; + ptr++; + put_int32(*ptr, ep); ep += 4; + node_sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + *ep++ = MAP_EXT; + ptr++; + put_int32(*ptr, ep); ep += 4; + /*fall through*/ + case HAMT_SUBTAG_NODE_BITMAP: + node_sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(node_sz < 17); + break; + default: + erl_exit(1, "bad header\r\n"); + } - obj = kptr[0]; - goto L_jump_start; + ptr++; + WSTACK_RESERVE(s, node_sz*2); + while(node_sz--) { + WSTACK_FAST_PUSH(s, ENC_HASHMAP_NODE); + WSTACK_FAST_PUSH(s, *ptr++); } } break; - case FLOAT_DEF: GET_DOUBLE(obj, f); if (dflags & DFLAG_NEW_FLOATS) { @@ -2619,6 +2685,7 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint bitoffs; Uint bitsize; byte* bytes; + byte* data_dst; ERTS_GET_BINARY_BYTES(obj, bytes, bitoffs, bitsize); if (dflags & DFLAG_INTERNAL_TAGS) { @@ -2664,7 +2731,7 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, j = binary_size(obj); put_int32(j, ep); ep += 4; - copy_binary_to_buffer(ep, 0, bytes, bitoffs, 8*j); + data_dst = ep; ep += j; } else if (dflags & DFLAG_BIT_BINARIES) { /* Bit-level binary. */ @@ -2674,7 +2741,7 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, ep += 4; *ep++ = bitsize; ep[j] = 0; /* Zero unused bits at end of binary */ - copy_binary_to_buffer(ep, 0, bytes, bitoffs, 8*j+bitsize); + data_dst = ep; ep += j + 1; } else { /* @@ -2688,11 +2755,18 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, put_int32((j+1), ep); ep += 4; ep[j] = 0; /* Zero unused bits at end of binary */ - copy_binary_to_buffer(ep, 0, bytes, bitoffs, 8*j+bitsize); + data_dst = ep; ep += j+1; *ep++ = SMALL_INTEGER_EXT; *ep++ = bitsize; } + if (ctx && j > r * TERM_TO_BINARY_MEMCPY_FACTOR) { + WSTACK_PUSH5(s, (UWord)data_dst, (UWord)bytes, bitoffs, + ENC_BIN_COPY, 8*j + bitsize); + } else { + copy_binary_to_buffer(data_dst, 0, bytes, bitoffs, + 8 * j + bitsize); + } } break; case EXPORT_DEF: @@ -2721,13 +2795,12 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, case FUN_DEF: { ErlFunThing* funp = (ErlFunThing *) fun_val(obj); + int ei; if ((dflags & DFLAG_NEW_FUN_TAGS) != 0) { - int ei; - *ep++ = NEW_FUN_EXT; - WSTACK_PUSH(s, ENC_PATCH_FUN_SIZE); - WSTACK_PUSH(s, (UWord) ep); /* Position for patching in size */ + WSTACK_PUSH2(s, ENC_PATCH_FUN_SIZE, + (UWord) ep); /* Position for patching in size */ ep += 4; *ep = funp->arity; ep += 1; @@ -2741,16 +2814,6 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, ep = enc_term(acmp, make_small(funp->fe->old_index), ep, dflags, off_heap); ep = enc_term(acmp, make_small(funp->fe->old_uniq), ep, dflags, off_heap); ep = enc_pid(acmp, funp->creator, ep, dflags); - - fun_env: - for (ei = funp->num_free-1; ei > 0; ei--) { - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) funp->env[ei]); - } - if (funp->num_free != 0) { - obj = funp->env[0]; - goto L_jump_start; - } } else { /* * Communicating with an obsolete erl_interface or @@ -2782,7 +2845,13 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, *ep++ = SMALL_TUPLE_EXT; put_int8(funp->num_free, ep); ep += 1; - goto fun_env; + } + for (ei = funp->num_free-1; ei > 0; ei--) { + WSTACK_PUSH2(s, ENC_TERM, (UWord) funp->env[ei]); + } + if (funp->num_free != 0) { + obj = funp->env[0]; + goto L_jump_start; } } break; @@ -2864,9 +2933,19 @@ undo_offheap_in_area(ErlOffHeap* off_heap, Eterm* start, Eterm* end) #endif /* DEBUG */ } +struct dec_term_hamt_placeholder +{ + struct dec_term_hamt_placeholder* next; + Eterm* objp; /* write result here */ + Uint size; /* nr of leafs */ + Eterm leafs[1]; +}; + +#define DEC_TERM_HAMT_PLACEHOLDER_SIZE \ + (offsetof(struct dec_term_hamt_placeholder, leafs) / sizeof(Eterm)) /* Decode term from external format into *objp. -** On failure return NULL and (R13B04) *hpp will be unchanged. +** On failure return NULL and *hpp will be unchanged. */ static byte* dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, @@ -2876,7 +2955,8 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, int n; ErtsAtomEncoding char_enc; register Eterm* hp; /* Please don't take the address of hp */ - Eterm *maps_head; /* for validation of maps */ + Eterm *maps_list; /* for preprocessing of small maps */ + struct dec_term_hamt_placeholder* hamt_list; /* for preprocessing of big maps */ Eterm* next; SWord reds; @@ -2886,7 +2966,8 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, next = ctx->u.dc.next; ep = ctx->u.dc.ep; hpp = &ctx->u.dc.hp; - maps_head = ctx->u.dc.maps_head; + maps_list = ctx->u.dc.maps_list; + hamt_list = ctx->u.dc.hamt_list; if (ctx->state != B2TDecode) { int n_limit = reds; @@ -2967,7 +3048,8 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, reds = ERTS_SWORD_MAX; next = objp; *next = (Eterm) (UWord) NULL; - maps_head = NULL; + maps_list = NULL; + hamt_list = NULL; } hp = *hpp; @@ -3031,6 +3113,8 @@ dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, big = make_small(0); } else { big = bytes_to_big(first, n, neg, hp); + if (is_nil(big)) + goto error; if (is_big(big)) { hp += big_arity(big) + 1; } @@ -3362,8 +3446,6 @@ dec_term_atom_common: } else { Binary* dbin = erts_bin_nrml_alloc(n); ProcBin* pb; - dbin->flags = 0; - dbin->orig_size = n; erts_refc_init(&dbin->refc, 1); pb = (ProcBin *) hp; hp += PROC_BIN_SIZE; @@ -3371,6 +3453,7 @@ dec_term_atom_common: pb->size = n; pb->next = off_heap->first; off_heap->first = (struct erl_off_heap_header*)pb; + OH_OVERHEAD(off_heap, pb->size / sizeof(Eterm)); pb->val = dbin; pb->bytes = (byte*) dbin->orig_bytes; pb->flags = 0; @@ -3416,14 +3499,13 @@ dec_term_atom_common: Binary* dbin = erts_bin_nrml_alloc(n); ProcBin* pb; - dbin->flags = 0; - dbin->orig_size = n; erts_refc_init(&dbin->refc, 1); pb = (ProcBin *) hp; pb->thing_word = HEADER_PROC_BIN; pb->size = n; pb->next = off_heap->first; off_heap->first = (struct erl_off_heap_header*)pb; + OH_OVERHEAD(off_heap, pb->size / sizeof(Eterm)); pb->val = dbin; pb->bytes = (byte*) dbin->orig_bytes; pb->flags = 0; @@ -3505,46 +3587,67 @@ dec_term_atom_common: break; case MAP_EXT: { - map_t *mp; Uint32 size,n; Eterm *kptr,*vptr; Eterm keys; size = get_int32(ep); ep += 4; - keys = make_tuple(hp); - *hp++ = make_arityval(size); - hp += size; - kptr = hp - 1; - - mp = (map_t*)hp; - hp += MAP_HEADER_SIZE; - hp += size; - vptr = hp - 1; - - /* kptr, last word for keys - * vptr, last word for values - */ - - /* - * Use thing_word to link through decoded maps. - * The list of maps is for later validation. - */ - - mp->thing_word = (Eterm) COMPRESS_POINTER(maps_head); - maps_head = (Eterm *) mp; - - mp->size = size; - mp->keys = keys; - *objp = make_map(mp); - - for (n = size; n; n--) { - *vptr = (Eterm) COMPRESS_POINTER(next); - *kptr = (Eterm) COMPRESS_POINTER(vptr); - next = kptr; - vptr--; - kptr--; - } + if (size <= MAP_SMALL_MAP_LIMIT) { + flatmap_t *mp; + + keys = make_tuple(hp); + *hp++ = make_arityval(size); + hp += size; + kptr = hp - 1; + + mp = (flatmap_t*)hp; + hp += MAP_HEADER_FLATMAP_SZ; + hp += size; + vptr = hp - 1; + + /* kptr, last word for keys + * vptr, last word for values + */ + + /* + * Use thing_word to link through decoded maps. + * The list of maps is for later validation. + */ + + mp->thing_word = (Eterm) COMPRESS_POINTER(maps_list); + maps_list = (Eterm *) mp; + + mp->size = size; + mp->keys = keys; + *objp = make_flatmap(mp); + + for (n = size; n; n--) { + *vptr = (Eterm) COMPRESS_POINTER(next); + *kptr = (Eterm) COMPRESS_POINTER(vptr); + next = kptr; + vptr--; + kptr--; + } + } + else { /* Make hamt */ + struct dec_term_hamt_placeholder* holder = + (struct dec_term_hamt_placeholder*) hp; + + holder->next = hamt_list; + hamt_list = holder; + holder->objp = objp; + holder->size = size; + + hp += DEC_TERM_HAMT_PLACEHOLDER_SIZE; + + for (n = size; n; n--) { + CDR(hp) = (Eterm) COMPRESS_POINTER(next); + CAR(hp) = (Eterm) COMPRESS_POINTER(&CDR(hp)); + next = &CAR(hp); + hp += 2; + } + } } break; case NEW_FUN_EXT: @@ -3722,6 +3825,7 @@ dec_term_atom_common: hp += PROC_BIN_SIZE; pb->next = off_heap->first; off_heap->first = (struct erl_off_heap_header*)pb; + OH_OVERHEAD(off_heap, pb->size / sizeof(Eterm)); pb->flags = 0; *objp = make_binary(pb); break; @@ -3739,6 +3843,7 @@ dec_term_atom_common: hp += PROC_BIN_SIZE; pb->next = off_heap->first; off_heap->first = (struct erl_off_heap_header*)pb; + OH_OVERHEAD(off_heap, pb->size / sizeof(Eterm)); pb->flags = 0; sub = (ErlSubBin*)hp; @@ -3765,7 +3870,8 @@ dec_term_atom_common: ctx->u.dc.ep = ep; ctx->u.dc.next = next; ctx->u.dc.hp = hp; - ctx->u.dc.maps_head = maps_head; + ctx->u.dc.maps_list = maps_list; + ctx->u.dc.hamt_list = hamt_list; ctx->reds = 0; return NULL; } @@ -3780,12 +3886,40 @@ dec_term_atom_common: * - done here for when we know it is complete. */ - while (maps_head) { - next = (Eterm *)(EXPAND_POINTER(*maps_head)); - *maps_head = MAP_HEADER; - if (!erts_validate_and_sort_map((map_t*)maps_head)) + while (maps_list) { + next = (Eterm *)(EXPAND_POINTER(*maps_list)); + *maps_list = MAP_HEADER_FLATMAP; + if (!erts_validate_and_sort_flatmap((flatmap_t*)maps_list)) goto error; - maps_head = next; + maps_list = next; + } + + /* Iterate through all the hamts and build tree nodes. + */ + if (hamt_list) { + ErtsHeapFactory factory; + + factory.p = NULL; + factory.hp = hp; + /* We assume heap will suffice (see hashmap_over_estimated_heap_size) */ + + do { + struct dec_term_hamt_placeholder* hamt = hamt_list; + *hamt->objp = erts_hashmap_from_array(&factory, + hamt->leafs, + hamt->size, + 1); + if (is_non_value(*hamt->objp)) + goto error; + + hamt_list = hamt->next; + + /* Yes, we waste a couple of heap words per hamt + for the temporary placeholder */ + *(Eterm*)hamt = make_pos_bignum_header(DEC_TERM_HAMT_PLACEHOLDER_SIZE-1); + } while (hamt_list); + + hp = factory.hp; } if (ctx) { @@ -3970,8 +4104,9 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, } for (i = 1; i <= arity; ++i) { if (is_list(ptr[i])) { - if ((m = is_string(obj)) && (m < MAX_STRING_LEN)) { + if ((m = is_string(ptr[i])) && (m < MAX_STRING_LEN)) { result += m + 2 + 1; + continue; } else { result += 5; } @@ -3982,43 +4117,99 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, } break; case MAP_DEF: - { - map_t *mp = (map_t*)map_val(obj); - Uint size = map_get_size(mp); + if (is_flatmap(obj)) { + flatmap_t *mp = (flatmap_t*)flatmap_val(obj); + Uint size = flatmap_get_size(mp); Uint i; Eterm *ptr; result += 1 + 4; /* tag + 4 bytes size */ /* push values first */ - ptr = map_get_values(mp); - i = size; - while(i--) { + ptr = flatmap_get_values(mp); + for (i = size; i; i--, ptr++) { if (is_list(*ptr)) { if ((m = is_string(*ptr)) && (m < MAX_STRING_LEN)) { result += m + 2 + 1; + continue; } else { result += 5; } } ESTACK_PUSH(s,*ptr); - ++ptr; } - ptr = map_get_keys(mp); - i = size; - while(i--) { + ptr = flatmap_get_keys(mp); + for (i = size; i; i--, ptr++) { if (is_list(*ptr)) { if ((m = is_string(*ptr)) && (m < MAX_STRING_LEN)) { result += m + 2 + 1; + continue; } else { result += 5; } } ESTACK_PUSH(s,*ptr); - ++ptr; } goto outer_loop; + } else { + Eterm *ptr; + Eterm hdr; + Uint node_sz; + ptr = boxed_val(obj); + hdr = *ptr; + ASSERT(is_header(hdr)); + switch(hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + ptr++; + node_sz = 16; + result += 1 + 4; /* tag + 4 bytes size */ + break; + case HAMT_SUBTAG_HEAD_BITMAP: + ptr++; + result += 1 + 4; /* tag + 4 bytes size */ + /*fall through*/ + case HAMT_SUBTAG_NODE_BITMAP: + node_sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(node_sz < 17); + break; + default: + erl_exit(1, "bad header\r\n"); + } + + ptr++; + ESTACK_RESERVE(s, node_sz*2); + while(node_sz--) { + if (is_list(*ptr)) { + Eterm* leaf = list_val(*ptr); + if (is_not_list(CAR(leaf))) { + ESTACK_FAST_PUSH(s, CAR(leaf)); + } + else { + if ((m = is_string(CAR(leaf))) && (m < MAX_STRING_LEN)) { + result += m + 2 + 1; + } else { + result += 5; + ESTACK_FAST_PUSH(s, CAR(leaf)); + } + } + if (is_not_list(CDR(leaf))) { + ESTACK_FAST_PUSH(s, CDR(leaf)); + } + else { + if ((m = is_string(CDR(leaf))) && (m < MAX_STRING_LEN)) { + result += m + 2 + 1; + } else { + result += 5; + ESTACK_FAST_PUSH(s, CDR(leaf)); + } + } + } + else { + ESTACK_FAST_PUSH(s, *ptr); + } + ptr++; + } } break; case FLOAT_DEF: @@ -4123,6 +4314,8 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, return 0; } + + static Sint decoded_size(byte *ep, byte* endp, int internal_tags, B2TContext* ctx) { @@ -4316,7 +4509,11 @@ init_done: n = get_int32(ep); ep += 4; ADDTERMS(2*n); - heap_size += 3 + n + 1 + n; + if (n <= MAP_SMALL_MAP_LIMIT) { + heap_size += 3 + n + 1 + n; + } else { + heap_size += hashmap_over_estimated_heap_size(n); + } break; case STRING_EXT: CHKSIZE(2); @@ -4440,66 +4637,3 @@ error: #undef SKIP2 #undef CHKSIZE } - - -#ifdef HIPE -BIF_RETTYPE hipe_wrapper_term_to_binary_1(BIF_ALIST_1); -BIF_RETTYPE hipe_wrapper_term_to_binary_2(BIF_ALIST_2); -BIF_RETTYPE hipe_wrapper_erts_internal_binary_to_term_1(BIF_ALIST_1); -BIF_RETTYPE hipe_wrapper_erts_internal_binary_to_term_2(BIF_ALIST_2); - -/* Hipe wrappers used by native code for BIFs that disable GC while trapping. - * - * Problem: - * When native code calls a BIF that traps, hipe_mode_switch will push a - * "trap frame" on the Erlang stack in order to find its way back from beam_emu - * back to native caller when finally done. If GC is disabled and stack/heap - * is full there is no place to push the "trap frame". - * - * Solution: - * We reserve space on stack for the "trap frame" here before the BIF is called. - * If the BIF does not trap, the space is reclaimed here before returning. - * If the BIF traps, hipe_push_beam_trap_frame() will detect that a "trap frame" - * already is reserved and use it. - */ -BIF_RETTYPE hipe_wrapper_term_to_binary_1(BIF_ALIST_1) -{ - Eterm res; - hipe_reserve_beam_trap_frame(BIF_P, BIF__ARGS, 1); - res = term_to_binary_1(BIF_P, BIF__ARGS); - if (is_value(res) || BIF_P->freason != TRAP) { - hipe_unreserve_beam_trap_frame(BIF_P); - } - return res; -} -BIF_RETTYPE hipe_wrapper_term_to_binary_2(BIF_ALIST_2) -{ - Eterm res; - hipe_reserve_beam_trap_frame(BIF_P, BIF__ARGS, 2); - res = term_to_binary_2(BIF_P, BIF__ARGS); - if (is_value(res) || BIF_P->freason != TRAP) { - hipe_unreserve_beam_trap_frame(BIF_P); - } - return res; -} -BIF_RETTYPE hipe_wrapper_erts_internal_binary_to_term_1(BIF_ALIST_1) -{ - Eterm res; - hipe_reserve_beam_trap_frame(BIF_P, BIF__ARGS, 1); - res = erts_internal_binary_to_term_1(BIF_P, BIF__ARGS); - if (is_value(res) || BIF_P->freason != TRAP) { - hipe_unreserve_beam_trap_frame(BIF_P); - } - return res; -} -BIF_RETTYPE hipe_wrapper_erts_internal_binary_to_term_2(BIF_ALIST_2) -{ - Eterm res; - hipe_reserve_beam_trap_frame(BIF_P, BIF__ARGS, 2); - res = erts_internal_binary_to_term_2(BIF_P, BIF__ARGS); - if (is_value(res) || BIF_P->freason != TRAP) { - hipe_unreserve_beam_trap_frame(BIF_P); - } - return res; -} -#endif /*HIPE*/ diff --git a/erts/emulator/beam/external.h b/erts/emulator/beam/external.h index bf00958eb1..50fcfa04d6 100644 --- a/erts/emulator/beam/external.h +++ b/erts/emulator/beam/external.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2013. All Rights Reserved. + * Copyright Ericsson AB 1996-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -150,19 +150,23 @@ typedef struct { Uint extsize; } ErtsBinary2TermState; + /* -------------------------------------------------------------------------- */ void erts_init_atom_cache_map(ErtsAtomCacheMap *); void erts_reset_atom_cache_map(ErtsAtomCacheMap *); void erts_destroy_atom_cache_map(ErtsAtomCacheMap *); void erts_finalize_atom_cache_map(ErtsAtomCacheMap *, Uint32); -Uint erts_encode_ext_dist_header_size(ErtsAtomCacheMap *); Uint erts_encode_ext_dist_header_size(ErtsAtomCacheMap *); byte *erts_encode_ext_dist_header_setup(byte *, ErtsAtomCacheMap *); byte *erts_encode_ext_dist_header_finalize(byte *, ErtsAtomCache *, Uint32); -Uint erts_encode_dist_ext_size(Eterm, Uint32, ErtsAtomCacheMap *); -void erts_encode_dist_ext(Eterm, byte **, Uint32, ErtsAtomCacheMap *); +struct erts_dsig_send_context; +int erts_encode_dist_ext_size(Eterm, Uint32, ErtsAtomCacheMap*, Uint* szp); +int erts_encode_dist_ext_size_int(Eterm term, struct erts_dsig_send_context* ctx, Uint* szp); +struct TTBEncodeContext_; +int erts_encode_dist_ext(Eterm, byte **, Uint32, ErtsAtomCacheMap *, + struct TTBEncodeContext_ *, Sint* reds); Uint erts_encode_ext_size(Eterm); Uint erts_encode_ext_size_2(Eterm, unsigned); diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h index 8fcb95d0e2..634fe533d0 100755..100644 --- a/erts/emulator/beam/global.h +++ b/erts/emulator/beam/global.h @@ -41,6 +41,7 @@ #include "error.h" #include "erl_utils.h" #include "erl_port.h" +#include "erl_gc.h" struct enif_environment_t /* ErlNifEnv */ { @@ -160,6 +161,7 @@ struct erts_driver_t_ { void (*ready_async)(ErlDrvData drv_data, ErlDrvThreadData thread_data); /* Might be NULL */ void (*process_exit)(ErlDrvData drv_data, ErlDrvMonitor *monitor); void (*stop_select)(ErlDrvEvent event, void*); /* Might be NULL */ + void (*emergency_close)(ErlDrvData drv_data); /* Might be NULL */ }; extern erts_driver_t *driver_list; @@ -346,8 +348,6 @@ extern Uint display_items; /* no of items to display in traces etc */ extern int erts_backtrace_depth; extern erts_smp_atomic32_t erts_max_gen_gcs; -extern int erts_disable_tolerant_timeofday; - extern int bif_reductions; /* reductions + fcalls (when doing call_bif) */ extern int stackdump_on_exit; @@ -370,16 +370,17 @@ extern int stackdump_on_exit; * DESTROY_ESTACK(Stack) */ -typedef struct { +typedef struct ErtsEStack_ { Eterm* start; Eterm* sp; Eterm* end; + Eterm* edefault; ErtsAlcType_t alloc_type; }ErtsEStack; #define DEF_ESTACK_SIZE (16) -void erl_grow_estack(ErtsEStack*, Eterm* def_stack); +void erl_grow_estack(ErtsEStack*, Uint need); #define ESTK_CONCAT(a,b) a##b #define ESTK_DEF_STACK(s) ESTK_CONCAT(s,_default_estack) @@ -389,22 +390,23 @@ void erl_grow_estack(ErtsEStack*, Eterm* def_stack); ESTK_DEF_STACK(s), /* start */ \ ESTK_DEF_STACK(s), /* sp */ \ ESTK_DEF_STACK(s) + DEF_ESTACK_SIZE, /* end */ \ + ESTK_DEF_STACK(s), /* default */ \ ERTS_ALC_T_ESTACK /* alloc_type */ \ } #define ESTACK_CHANGE_ALLOCATOR(s,t) \ do { \ - if (s.start != ESTK_DEF_STACK(s)) { \ + if ((s).start != ESTK_DEF_STACK(s)) { \ erl_exit(1, "Internal error - trying to change allocator " \ "type of active estack\n"); \ } \ - s.alloc_type = (t); \ + (s).alloc_type = (t); \ } while (0) #define DESTROY_ESTACK(s) \ do { \ - if (s.start != ESTK_DEF_STACK(s)) { \ - erts_free(s.alloc_type, s.start); \ + if ((s).start != ESTK_DEF_STACK(s)) { \ + erts_free((s).alloc_type, (s).start); \ } \ } while(0) @@ -415,16 +417,17 @@ do { \ */ #define ESTACK_SAVE(s,dst)\ do {\ - if (s.start == ESTK_DEF_STACK(s)) {\ + if ((s).start == ESTK_DEF_STACK(s)) {\ UWord _wsz = ESTACK_COUNT(s);\ - (dst)->start = erts_alloc(s.alloc_type,\ + (dst)->start = erts_alloc((s).alloc_type,\ DEF_ESTACK_SIZE * sizeof(Eterm));\ - memcpy((dst)->start, s.start,_wsz*sizeof(Eterm));\ + memcpy((dst)->start, (s).start,_wsz*sizeof(Eterm));\ (dst)->sp = (dst)->start + _wsz;\ (dst)->end = (dst)->start + DEF_ESTACK_SIZE;\ - (dst)->alloc_type = s.alloc_type;\ + (dst)->edefault = NULL;\ + (dst)->alloc_type = (s).alloc_type;\ } else\ - *(dst) = s;\ + *(dst) = (s);\ } while (0) #define DESTROY_SAVED_ESTACK(estack)\ @@ -435,78 +438,122 @@ do {\ }\ } while(0) +#define CLEAR_SAVED_ESTACK(estack) ((void) ((estack)->start = NULL)) + /* * Use on empty stack, only the allocator can be changed before this. * The src stack is reset to NULL. */ #define ESTACK_RESTORE(s, src) \ do { \ - ASSERT(s.start == ESTK_DEF_STACK(s)); \ - s = *(src); /* struct copy */ \ + ASSERT((s).start == ESTK_DEF_STACK(s)); \ + (s) = *(src); /* struct copy */ \ (src)->start = NULL; \ - ASSERT(s.sp >= s.start); \ - ASSERT(s.sp <= s.end); \ + ASSERT((s).sp >= (s).start); \ + ASSERT((s).sp <= (s).end); \ } while (0) -#define ESTACK_IS_STATIC(s) (s.start == ESTK_DEF_STACK(s))) +#define ESTACK_IS_STATIC(s) ((s).start == ESTK_DEF_STACK(s)) -#define ESTACK_PUSH(s, x) \ -do { \ - if (s.sp == s.end) { \ - erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ - } \ - *s.sp++ = (x); \ +#define ESTACK_PUSH(s, x) \ +do { \ + if ((s).sp == (s).end) { \ + erl_grow_estack(&(s), 1); \ + } \ + *(s).sp++ = (x); \ } while(0) #define ESTACK_PUSH2(s, x, y) \ do { \ - if (s.sp > s.end - 2) { \ - erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ + if ((s).sp > (s).end - 2) { \ + erl_grow_estack(&(s), 2); \ } \ - *s.sp++ = (x); \ - *s.sp++ = (y); \ + *(s).sp++ = (x); \ + *(s).sp++ = (y); \ } while(0) #define ESTACK_PUSH3(s, x, y, z) \ do { \ - if (s.sp > s.end - 3) { \ - erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ + if ((s).sp > (s).end - 3) { \ + erl_grow_estack(&s, 3); \ } \ - *s.sp++ = (x); \ - *s.sp++ = (y); \ - *s.sp++ = (z); \ + *(s).sp++ = (x); \ + *(s).sp++ = (y); \ + *(s).sp++ = (z); \ +} while(0) + +#define ESTACK_PUSH4(s, E1, E2, E3, E4) \ +do { \ + if ((s).sp > (s).end - 4) { \ + erl_grow_estack(&s, 4); \ + } \ + *(s).sp++ = (E1); \ + *(s).sp++ = (E2); \ + *(s).sp++ = (E3); \ + *(s).sp++ = (E4); \ +} while(0) + +#define ESTACK_RESERVE(s, push_cnt) \ +do { \ + if ((s).sp > (s).end - (push_cnt)) { \ + erl_grow_estack(&(s), (push_cnt)); \ + } \ } while(0) -#define ESTACK_COUNT(s) (s.sp - s.start) -#define ESTACK_ISEMPTY(s) (s.sp == s.start) -#define ESTACK_POP(s) (*(--s.sp)) +/* Must be preceded by ESTACK_RESERVE */ +#define ESTACK_FAST_PUSH(s, x) \ +do { \ + ASSERT((s).sp < (s).end); \ + *s.sp++ = (x); \ +} while(0) + +#define ESTACK_COUNT(s) ((s).sp - (s).start) +#define ESTACK_ISEMPTY(s) ((s).sp == (s).start) +#define ESTACK_POP(s) (*(--(s).sp)) /* * WSTACK: same as ESTACK but with UWord instead of Eterm */ -typedef struct { +typedef struct ErtsWStack_ { UWord* wstart; UWord* wsp; UWord* wend; + UWord* wdefault; ErtsAlcType_t alloc_type; }ErtsWStack; #define DEF_WSTACK_SIZE (16) -void erl_grow_wstack(ErtsWStack*, UWord* def_stack); +void erl_grow_wstack(ErtsWStack*, Uint need); #define WSTK_CONCAT(a,b) a##b #define WSTK_DEF_STACK(s) WSTK_CONCAT(s,_default_wstack) -#define DECLARE_WSTACK(s) \ +#define WSTACK_DECLARE(s) \ UWord WSTK_DEF_STACK(s)[DEF_WSTACK_SIZE]; \ ErtsWStack s = { \ WSTK_DEF_STACK(s), /* wstart */ \ WSTK_DEF_STACK(s), /* wsp */ \ WSTK_DEF_STACK(s) + DEF_WSTACK_SIZE, /* wend */ \ + WSTK_DEF_STACK(s), /* wdflt */ \ ERTS_ALC_T_ESTACK /* alloc_type */ \ } +#define DECLARE_WSTACK WSTACK_DECLARE + +typedef struct ErtsDynamicWStack_ { + UWord default_stack[DEF_WSTACK_SIZE]; + ErtsWStack ws; +}ErtsDynamicWStack; + +#define WSTACK_INIT(dwsp, ALC_TYPE) \ +do { \ + (dwsp)->ws.wstart = (dwsp)->default_stack; \ + (dwsp)->ws.wsp = (dwsp)->default_stack; \ + (dwsp)->ws.wend = (dwsp)->default_stack + DEF_WSTACK_SIZE;\ + (dwsp)->ws.wdefault = (dwsp)->default_stack; \ + (dwsp)->ws.alloc_type = ALC_TYPE; \ +} while (0) #define WSTACK_CHANGE_ALLOCATOR(s,t) \ do { \ @@ -517,13 +564,20 @@ do { \ s.alloc_type = (t); \ } while (0) -#define DESTROY_WSTACK(s) \ +#define WSTACK_DESTROY(s) \ do { \ - if (s.wstart != WSTK_DEF_STACK(s)) { \ + if (s.wstart != s.wdefault) { \ erts_free(s.alloc_type, s.wstart); \ } \ } while(0) +#define DESTROY_WSTACK WSTACK_DESTROY +#define WSTACK_DEBUG(s) \ + do { \ + fprintf(stderr, "wstack size = %ld\r\n", s.wsp - s.wstart); \ + fprintf(stderr, "wstack wstart = %p\r\n", s.wstart); \ + fprintf(stderr, "wstack wsp = %p\r\n", s.wsp); \ + } while(0) /* * Do not free the stack after this, it may have pointers into what @@ -538,6 +592,7 @@ do {\ memcpy((dst)->wstart, s.wstart,_wsz*sizeof(UWord));\ (dst)->wsp = (dst)->wstart + _wsz;\ (dst)->wend = (dst)->wstart + DEF_WSTACK_SIZE;\ + (dst)->wdefault = NULL;\ (dst)->alloc_type = s.alloc_type;\ } else\ *(dst) = s;\ @@ -551,6 +606,8 @@ do {\ }\ } while(0) +#define CLEAR_SAVED_WSTACK(wstack) ((void) ((wstack)->wstart = NULL)) + /* * Use on empty stack, only the allocator can be changed before this. * The src stack is reset to NULL. @@ -564,12 +621,12 @@ do { \ ASSERT(s.wsp <= s.wend); \ } while (0) -#define WSTACK_IS_STATIC(s) (s.wstart == WSTK_DEF_STACK(s))) +#define WSTACK_IS_STATIC(s) (s.wstart == WSTK_DEF_STACK(s)) #define WSTACK_PUSH(s, x) \ do { \ if (s.wsp == s.wend) { \ - erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + erl_grow_wstack(&s, 1); \ } \ *s.wsp++ = (x); \ } while(0) @@ -577,7 +634,7 @@ do { \ #define WSTACK_PUSH2(s, x, y) \ do { \ if (s.wsp > s.wend - 2) { \ - erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + erl_grow_wstack(&s, 2); \ } \ *s.wsp++ = (x); \ *s.wsp++ = (y); \ @@ -585,17 +642,129 @@ do { \ #define WSTACK_PUSH3(s, x, y, z) \ do { \ - if (s.wsp > s.wend - 3) { \ - erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + if (s.wsp > s.wend - 3) { \ + erl_grow_wstack(&s, 3); \ } \ *s.wsp++ = (x); \ *s.wsp++ = (y); \ *s.wsp++ = (z); \ } while(0) +#define WSTACK_PUSH4(s, A1, A2, A3, A4) \ +do { \ + if (s.wsp > s.wend - 4) { \ + erl_grow_wstack(&s, 4); \ + } \ + *s.wsp++ = (A1); \ + *s.wsp++ = (A2); \ + *s.wsp++ = (A3); \ + *s.wsp++ = (A4); \ +} while(0) + +#define WSTACK_PUSH5(s, A1, A2, A3, A4, A5) \ +do { \ + if (s.wsp > s.wend - 5) { \ + erl_grow_wstack(&s, 5); \ + } \ + *s.wsp++ = (A1); \ + *s.wsp++ = (A2); \ + *s.wsp++ = (A3); \ + *s.wsp++ = (A4); \ + *s.wsp++ = (A5); \ +} while(0) + +#define WSTACK_PUSH6(s, A1, A2, A3, A4, A5, A6) \ +do { \ + if (s.wsp > s.wend - 6) { \ + erl_grow_wstack(&s, 6); \ + } \ + *s.wsp++ = (A1); \ + *s.wsp++ = (A2); \ + *s.wsp++ = (A3); \ + *s.wsp++ = (A4); \ + *s.wsp++ = (A5); \ + *s.wsp++ = (A6); \ +} while(0) + +#define WSTACK_RESERVE(s, push_cnt) \ +do { \ + if (s.wsp > s.wend - (push_cnt)) { \ + erl_grow_wstack(&s, (push_cnt)); \ + } \ +} while(0) + +/* Must be preceded by WSTACK_RESERVE */ +#define WSTACK_FAST_PUSH(s, x) \ +do { \ + ASSERT(s.wsp < s.wend); \ + *s.wsp++ = (x); \ +} while(0) + #define WSTACK_COUNT(s) (s.wsp - s.wstart) #define WSTACK_ISEMPTY(s) (s.wsp == s.wstart) -#define WSTACK_POP(s) (*(--s.wsp)) +#define WSTACK_POP(s) ((ASSERT(s.wsp > s.wstart)),*(--s.wsp)) + +#define WSTACK_ROLLBACK(s, count) (ASSERT(WSTACK_COUNT(s) >= (count)), \ + s.wsp = s.wstart + (count)) + +/* PSTACK - Stack of any type. + * Usage: + * { + * #define PSTACK_TYPE MyType + * PSTACK_DECLARE(s,16); + * MyType *sp = PSTACK_PUSH(s); + * + * sp->x = .... + * sp->y = .... + * sp = PSTACK_PUSH(s); + * ... + * sp = PSTACK_POP(s); + * if (PSTACK_IS_EMPTY(s)) { + * // sp is invalid when stack is empty after pop + * } + * + * PSTACK_DESTROY(s); + * } + */ + + +typedef struct ErtsPStack_ { + byte* pstart; + byte* psp; + byte* pend; + ErtsAlcType_t alloc_type; +}ErtsPStack; + +void erl_grow_pstack(ErtsPStack* s, void* default_pstack, unsigned need_bytes); +#define PSTK_CONCAT(a,b) a##b +#define PSTK_DEF_STACK(s) PSTK_CONCAT(s,_default_pstack) + +#define PSTACK_DECLARE(s, DEF_PSTACK_SIZE) \ +PSTACK_TYPE PSTK_DEF_STACK(s)[DEF_PSTACK_SIZE]; \ +ErtsPStack s = { (byte*)PSTK_DEF_STACK(s), /* pstart */ \ + (byte*)(PSTK_DEF_STACK(s) - 1), /* psp */ \ + (byte*)(PSTK_DEF_STACK(s) + (DEF_PSTACK_SIZE)), /* pend */\ + ERTS_ALC_T_ESTACK /* alloc_type */ \ +} + +#define PSTACK_DESTROY(s) \ +do { \ + if (s.pstart != (byte*)PSTK_DEF_STACK(s)) { \ + erts_free(s.alloc_type, s.pstart); \ + } \ +} while(0) + +#define PSTACK_IS_EMPTY(s) (s.psp < s.pstart) + +#define PSTACK_TOP(s) (ASSERT(!PSTACK_IS_EMPTY(s)), (PSTACK_TYPE*)(s.psp)) + +#define PSTACK_PUSH(s) \ + (s.psp += sizeof(PSTACK_TYPE), \ + ((s.psp == s.pend) ? erl_grow_pstack(&s, PSTK_DEF_STACK(s), \ + sizeof(PSTACK_TYPE)) : (void)0), \ + ((PSTACK_TYPE*) s.psp)) + +#define PSTACK_POP(s) ((PSTACK_TYPE*) (s.psp -= sizeof(PSTACK_TYPE))) /* binary.c */ @@ -618,9 +787,6 @@ erts_bld_port_info(Eterm **hpp, void erts_bif_info_init(void); /* bif.c */ -Eterm erts_make_ref(Process *); -Eterm erts_make_ref_in_buffer(Eterm buffer[REF_THING_SIZE]); -void erts_make_ref_in_array(Uint32 ref[ERTS_MAX_REF_NUMBERS]); ERTS_GLB_INLINE Eterm erts_proc_store_ref(Process *c_p, Uint32 ref[ERTS_MAX_REF_NUMBERS]); @@ -805,23 +971,6 @@ void MD5Init(MD5_CTX *); void MD5Update(MD5_CTX *, unsigned char *, unsigned int); void MD5Final(unsigned char [16], MD5_CTX *); -/* ggc.c */ - -void erts_gc_info(ErtsGCInfo *gcip); -void erts_init_gc(void); -int erts_garbage_collect(Process*, int, Eterm*, int); -void erts_garbage_collect_hibernate(Process* p); -Eterm erts_gc_after_bif_call(Process* p, Eterm result, Eterm* regs, Uint arity); -void erts_garbage_collect_literals(Process* p, Eterm* literals, - Uint lit_size, - struct erl_off_heap_header* oh); -Uint erts_next_heap_size(Uint, Uint); -Eterm erts_heap_sizes(Process* p); - -void erts_offset_off_heap(ErlOffHeap *, Sint, Eterm*, Eterm*); -void erts_offset_heap_ptr(Eterm*, Uint, Sint, Eterm*, Eterm*); -void erts_offset_heap(Eterm*, Uint, Sint, Eterm*, Eterm*); -void erts_free_heap_frags(Process* p); /* io.c */ @@ -848,6 +997,7 @@ Uint erts_port_ioq_size(Port *pp); void erts_stale_drv_select(Eterm, ErlDrvPort, ErlDrvEvent, int, int); Port *erts_get_heart_port(void); +void erts_emergency_close_ports(void); #if defined(ERTS_SMP) && defined(ERTS_ENABLE_LOCK_COUNT) void erts_lcnt_enable_io_lock_count(int enable); @@ -951,20 +1101,67 @@ struct Sint_buf { }; char* Sint_to_buf(Sint, struct Sint_buf*); +#define ERTS_IOLIST_STATE_INITER(C_P, OBJ) \ + {(C_P), 0, 0, (OBJ), {NULL, NULL, NULL, ERTS_ALC_T_INVALID}, 0, 0} + +#define ERTS_IOLIST_STATE_MOVE(TO, FROM) \ + sys_memcpy((void *) (TO), (void *) (FROM), sizeof(ErtsIOListState)) + +#define ERTS_IOLIST_SIZE_YIELDS_COUNT_PER_RED 8 + +typedef struct { + Process *c_p; + ErlDrvSizeT size; + Uint offs; + Eterm obj; + ErtsEStack estack; + int reds_left; + int have_size; +} ErtsIOListState; + +#define ERTS_IOLIST2BUF_STATE_INITER(C_P, OBJ) \ + {ERTS_IOLIST_STATE_INITER((C_P), (OBJ)), {NULL, 0, 0, 0}, NULL, 0, NULL, 0} + +#define ERTS_IOLIST2BUF_STATE_MOVE(TO, FROM) \ + sys_memcpy((void *) (TO), (void *) (FROM), sizeof(ErtsIOList2BufState)) + +#define ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT 32 +#define ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED 8 +#define ERTS_IOLIST_TO_BUF_BYTES_PER_RED \ + (ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED*ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT) + +typedef struct { + ErtsIOListState iolist; + struct { + byte *bptr; + size_t size; + Uint bitoffs; + Uint bitsize; + } bcopy; + char *buf; + ErlDrvSizeT len; + Eterm *objp; + int offset; +} ErtsIOList2BufState; + #define ERTS_IOLIST_OK 0 #define ERTS_IOLIST_OVERFLOW 1 #define ERTS_IOLIST_TYPE 2 +#define ERTS_IOLIST_YIELD 3 Eterm buf_to_intlist(Eterm**, const char*, size_t, Eterm); /* most callers pass plain char*'s */ #define ERTS_IOLIST_TO_BUF_OVERFLOW (~((ErlDrvSizeT) 0)) #define ERTS_IOLIST_TO_BUF_TYPE_ERROR (~((ErlDrvSizeT) 1)) +#define ERTS_IOLIST_TO_BUF_YIELD (~((ErlDrvSizeT) 2)) #define ERTS_IOLIST_TO_BUF_FAILED(R) \ - (((R) & (~((ErlDrvSizeT) 1))) == (~((ErlDrvSizeT) 1))) + (((R) & (~((ErlDrvSizeT) 3))) == (~((ErlDrvSizeT) 3))) #define ERTS_IOLIST_TO_BUF_SUCCEEDED(R) \ (!ERTS_IOLIST_TO_BUF_FAILED((R))) ErlDrvSizeT erts_iolist_to_buf(Eterm, char*, ErlDrvSizeT); +ErlDrvSizeT erts_iolist_to_buf_yielding(ErtsIOList2BufState *); +int erts_iolist_size_yielding(ErtsIOListState *state); int erts_iolist_size(Eterm, ErlDrvSizeT *); int is_string(Eterm); void erl_at_exit(void (*) (void*), void*); @@ -1108,7 +1305,9 @@ erts_alloc_message_heap_state(Uint size, state = erts_smp_atomic32_read_acqb(&receiver->state); if (statep) *statep = state; - if (state & (ERTS_PSFLG_EXITING|ERTS_PSFLG_PENDING_EXIT)) + if (state & (ERTS_PSFLG_OFF_HEAP_MSGS + | ERTS_PSFLG_EXITING + | ERTS_PSFLG_PENDING_EXIT)) goto allocate_in_mbuf; #endif @@ -1128,7 +1327,9 @@ erts_alloc_message_heap_state(Uint size, state = erts_smp_atomic32_read_nob(&receiver->state); if (statep) *statep = state; - if ((state & (ERTS_PSFLG_EXITING|ERTS_PSFLG_PENDING_EXIT)) + if ((state & (ERTS_PSFLG_OFF_HEAP_MSGS + | ERTS_PSFLG_EXITING + | ERTS_PSFLG_PENDING_EXIT)) || (receiver->flags & F_DISABLE_GC) || HEAP_LIMIT(receiver) - HEAP_TOP(receiver) <= size) { /* diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c index edf4a28784..1db3a9fba7 100644 --- a/erts/emulator/beam/io.c +++ b/erts/emulator/beam/io.c @@ -47,6 +47,7 @@ #include "external.h" #include "dtrace-wrapper.h" #include "erl_map.h" +#include "erl_bif_unique.h" extern ErlDrvEntry fd_driver_entry; #ifndef __OSE__ @@ -391,7 +392,7 @@ static Port *create_port(char *name, /* Set default tracing */ erts_get_default_tracing(&ERTS_TRACE_FLAGS(prt), &ERTS_TRACER_PROC(prt)); - ASSERT(((char *) prt) == ((char *) &prt->common)); + ERTS_CT_ASSERT(offsetof(Port,common) == 0); #if !ERTS_PORT_INIT_INSTR_NEED_ID /* @@ -1218,9 +1219,10 @@ typedef struct { static ERTS_INLINE ErtsTryImmDrvCallResult try_imm_drv_call(ErtsTryImmDrvCallState *sp) { + unsigned int prof_runnable_ports; ErtsTryImmDrvCallResult res; int reds_left_in; - erts_aint32_t invalid_state, invalid_sched_flags; + erts_aint32_t act, exp, invalid_state, invalid_sched_flags; Port *prt = sp->port; Process *c_p = sp->c_p; @@ -1247,18 +1249,39 @@ try_imm_drv_call(ErtsTryImmDrvCallState *sp) goto locked_fail; } - sp->sched_flags = erts_smp_atomic32_read_nob(&prt->sched.flags); - if (sp->sched_flags & invalid_sched_flags) { - res = ERTS_TRY_IMM_DRV_CALL_INVALID_SCHED_FLAGS; - goto locked_fail; - } + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&prt->sched); + act = erts_smp_atomic32_read_nob(&prt->sched.flags); + + do { + erts_aint32_t new; + + if (act & invalid_sched_flags) { + res = ERTS_TRY_IMM_DRV_CALL_INVALID_SCHED_FLAGS; + sp->sched_flags = act; + goto locked_fail; + } + exp = act; + new = act | ERTS_PTS_FLG_EXEC_IMM; + act = erts_smp_atomic32_cmpxchg_mb(&prt->sched.flags, new, exp); + } while (act != exp); + + sp->sched_flags = act; if (!c_p) reds_left_in = CONTEXT_REDS/10; else { if (IS_TRACED_FL(c_p, F_TRACE_SCHED_PROCS)) trace_virtual_sched(c_p, am_out); + /* + * No status lock held while sending runnable + * proc trace messages. It is however not needed + * in this case, since only this thread can send + * such messages for this process until the process + * has been scheduled out. + */ if (erts_system_profile_flags.runnable_procs && erts_system_profile_flags.exclusive) profile_runnable_proc(c_p, am_inactive); @@ -1273,11 +1296,14 @@ try_imm_drv_call(ErtsTryImmDrvCallState *sp) ERTS_SMP_CHK_NO_PROC_LOCKS; - if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) - trace_sched_ports_where(prt, am_in, sp->port_op); - if (erts_system_profile_flags.runnable_ports - && !erts_port_is_scheduled(prt)) - profile_runnable_port(prt, am_active); + if (prof_runnable_ports | IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) { + if (prof_runnable_ports && !(act & (ERTS_PTS_FLG_IN_RUNQ|ERTS_PTS_FLG_EXEC))) + profile_runnable_port(prt, am_active); + if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) + trace_sched_ports_where(prt, am_in, sp->port_op); + if (prof_runnable_ports) + erts_port_task_sched_unlock(&prt->sched); + } sp->fpe_was_unmasked = erts_block_fpe(); @@ -1294,17 +1320,31 @@ finalize_imm_drv_call(ErtsTryImmDrvCallState *sp) int reds; Port *prt = sp->port; Process *c_p = sp->c_p; + erts_aint32_t act; + unsigned int prof_runnable_ports; reds = prt->reds; reds += erts_port_driver_callback_epilogue(prt, NULL); erts_unblock_fpe(sp->fpe_was_unmasked); - if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) - trace_sched_ports_where(prt, am_out, sp->port_op); - if (erts_system_profile_flags.runnable_ports - && !erts_port_is_scheduled(prt)) - profile_runnable_port(prt, am_inactive); + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&prt->sched); + + act = erts_smp_atomic32_read_band_mb(&prt->sched.flags, + ~ERTS_PTS_FLG_EXEC_IMM); + ERTS_SMP_LC_ASSERT(act & ERTS_PTS_FLG_EXEC_IMM); + + if (prof_runnable_ports | IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) { + if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) + trace_sched_ports_where(prt, am_out, sp->port_op); + if (prof_runnable_ports) { + if (!(act & (ERTS_PTS_FLG_IN_RUNQ|ERTS_PTS_FLG_EXEC))) + profile_runnable_port(prt, am_inactive); + erts_port_task_sched_unlock(&prt->sched); + } + } erts_port_release(prt); @@ -1319,6 +1359,13 @@ finalize_imm_drv_call(ErtsTryImmDrvCallState *sp) if (IS_TRACED_FL(c_p, F_TRACE_SCHED_PROCS)) trace_virtual_sched(c_p, am_in); + /* + * No status lock held while sending runnable + * proc trace messages. It is however not needed + * in this case, since only this thread can send + * such messages for this process until the process + * has been scheduled out. + */ if (erts_system_profile_flags.runnable_procs && erts_system_profile_flags.exclusive) profile_runnable_proc(c_p, am_active); @@ -3107,8 +3154,6 @@ static void deliver_read_message(Port* prt, erts_aint32_t state, Eterm to, Binary* bptr; bptr = erts_bin_nrml_alloc(len); - bptr->flags = 0; - bptr->orig_size = len; erts_refc_init(&bptr->refc, 1); sys_memcpy(bptr->orig_bytes, buf, len); @@ -4041,6 +4086,9 @@ erts_port_control(Process* c_p, size, &resp_bufp, &resp_size); + + control_flags = prt->control_flags; + finalize_imm_drv_call(&try_call_state); if (tmp_alloced) erts_free(ERTS_ALC_T_TMP, bufp); @@ -4048,8 +4096,6 @@ erts_port_control(Process* c_p, return ERTS_PORT_OP_BADARG; } - control_flags = prt->control_flags; - hsz = port_control_result_size(control_flags, resp_bufp, &resp_size, @@ -4438,7 +4484,7 @@ make_port_info_term(Eterm **hpp_start, int len; int start; static Eterm item[] = ERTS_PORT_INFO_1_ITEMS; - static Eterm value[sizeof(item)/sizeof(item[0])]; + Eterm value[sizeof(item)/sizeof(item[0])]; start = 0; len = sizeof(item)/sizeof(item[0]); @@ -5305,7 +5351,11 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len) case ERL_DRV_MAP: { /* int */ ERTS_DDT_CHK_ENOUGH_ARGS(1); if ((int) ptr[0] < 0) ERTS_DDT_FAIL; - need += MAP_HEADER_SIZE + 1 + 2*ptr[0]; + if (ptr[0] > MAP_SMALL_MAP_LIMIT) { + need += hashmap_over_estimated_heap_size(ptr[0]); + } else { + need += MAP_HEADER_FLATMAP_SZ + 1 + 2*ptr[0]; + } depth -= 2*ptr[0]; if (depth < 0) ERTS_DDT_FAIL; ptr++; @@ -5460,8 +5510,6 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len) ProcBin* pbp; Binary* bp = erts_bin_nrml_alloc(size); ASSERT(bufp); - bp->flags = 0; - bp->orig_size = (SWord) size; erts_refc_init(&bp->refc, 1); sys_memcpy((void *) bp->orig_bytes, (void *) bufp, size); pbp = (ProcBin *) hp; @@ -5551,31 +5599,52 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len) case ERL_DRV_MAP: { /* int */ int size = (int)ptr[0]; - Eterm* tp = hp; - Eterm* vp; - map_t *mp; - - *tp = make_arityval(size); - - hp += 1 + size; - mp = (map_t*)hp; - mp->thing_word = MAP_HEADER; - mp->size = size; - mp->keys = make_tuple(tp); - mess = make_map(mp); - - hp += MAP_HEADER_SIZE + size; /* advance "heap" pointer */ - - tp += size; /* point at last key */ - vp = hp - 1; /* point at last value */ - - while(size--) { - *vp-- = ESTACK_POP(stack); - *tp-- = ESTACK_POP(stack); - } - if (!erts_validate_and_sort_map(mp)) - ERTS_DDT_FAIL; - ptr++; + if (size > MAP_SMALL_MAP_LIMIT) { + int ix = 2*size; + ErtsHeapFactory factory; + Eterm* leafs = hp; + + hp += 2*size; + while(ix--) { *--hp = ESTACK_POP(stack); } + + hp += 2*size; + factory.p = NULL; + factory.hp = hp; + /* We assume heap will suffice (see hashmap_over_estimated_heap_size) */ + + mess = erts_hashmap_from_array(&factory, leafs, size, 1); + + if (is_non_value(mess)) + ERTS_DDT_FAIL; + + hp = factory.hp; + } else { + Eterm* tp = hp; + Eterm* vp; + flatmap_t *mp; + + *tp = make_arityval(size); + + hp += 1 + size; + mp = (flatmap_t*)hp; + mp->thing_word = MAP_HEADER_FLATMAP; + mp->size = size; + mp->keys = make_tuple(tp); + mess = make_flatmap(mp); + + hp += MAP_HEADER_FLATMAP_SZ + size; + + tp += size; /* point at last key */ + vp = hp - 1; /* point at last value */ + + while(size--) { + *vp-- = ESTACK_POP(stack); + *tp-- = ESTACK_POP(stack); + } + if (!erts_validate_and_sort_flatmap(mp)) + ERTS_DDT_FAIL; + } + ptr++; break; } @@ -5953,9 +6022,7 @@ driver_alloc_binary(ErlDrvSizeT size) bin = erts_bin_drv_alloc_fnf((Uint) size); if (!bin) return NULL; /* The driver write must take action */ - bin->flags = BIN_FLAG_DRV; erts_refc_init(&bin->refc, 1); - bin->orig_size = (SWord) size; return Binary2ErlDrvBinary(bin); } @@ -5985,7 +6052,6 @@ ErlDrvBinary* driver_realloc_binary(ErlDrvBinary* bin, ErlDrvSizeT size) if (!newbin) return NULL; - newbin->orig_size = size; return Binary2ErlDrvBinary(newbin); } @@ -6129,7 +6195,7 @@ driver_pdl_create(ErlDrvPort dp) return NULL; pdl = erts_alloc(ERTS_ALC_T_PORT_DATA_LOCK, sizeof(struct erl_drv_port_data_lock)); - erts_mtx_init(&pdl->mtx, "port_data_lock"); + erts_mtx_init_x(&pdl->mtx, "port_data_lock", pp->common.id, 1); pdl_init_refc(pdl); erts_port_inc_refc(pp); pdl->prt = pp; @@ -6659,7 +6725,7 @@ static void ref_to_driver_monitor(Eterm ref, ErlDrvMonitor *mon) { RefThing *refp; ASSERT(is_internal_ref(ref)); - ASSERT(sizeof(RefThing) <= sizeof(ErlDrvMonitor)); + ERTS_CT_ASSERT(sizeof(RefThing) <= sizeof(ErlDrvMonitor)); refp = ref_thing_ptr(ref); memset(mon,0,sizeof(ErlDrvMonitor)); memcpy(mon,refp,sizeof(RefThing)); @@ -7166,7 +7232,7 @@ char *driver_dl_error(void) #define ERL_DRV_SYS_INFO_SIZE(LAST_FIELD) \ - (((size_t) &((ErlDrvSysInfo *) 0)->LAST_FIELD) \ + (offsetof(ErlDrvSysInfo, LAST_FIELD) \ + sizeof(((ErlDrvSysInfo *) 0)->LAST_FIELD)) void @@ -7228,6 +7294,18 @@ driver_system_info(ErlDrvSysInfo *sip, size_t si_size) sip->nif_major_version = ERL_NIF_MAJOR_VERSION; sip->nif_minor_version = ERL_NIF_MINOR_VERSION; } + /* + * 'dirty_scheduler_support' is the last field in the 4th version + * (driver version 3.1, NIF version 2.7) + */ + if (si_size >= ERL_DRV_SYS_INFO_SIZE(dirty_scheduler_support)) { +#if defined(ERL_NIF_DIRTY_SCHEDULER_SUPPORT) && defined(USE_THREADS) + sip->dirty_scheduler_support = 1; +#else + sip->dirty_scheduler_support = 0; +#endif + } + } @@ -7291,6 +7369,8 @@ no_stop_select_callback(ErlDrvEvent event, void* private) erts_send_error_to_logger_nogl(dsbufp); } +#define IS_DRIVER_VERSION_GE(DE,MAJOR,MINOR) \ + ((DE)->major_version >= (MAJOR) && (DE)->minor_version >= (MINOR)) static int init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) @@ -7338,6 +7418,7 @@ init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) drv->timeout = de->timeout ? de->timeout : no_timeout_callback; drv->ready_async = de->ready_async; drv->process_exit = de->process_exit; + drv->emergency_close = IS_DRIVER_VERSION_GE(de,3,2) ? de->emergency_close : NULL; if (de->stop_select) drv->stop_select = de->stop_select; else @@ -7356,6 +7437,8 @@ init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) } } +#undef IS_DRIVER_VERSION_GE + void erts_destroy_driver(erts_driver_t *drv) { @@ -7499,7 +7582,7 @@ Port *erts_get_heart_port(void) if (!port) continue; /* only examine undead or alive ports */ - if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_DEAD) + if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP) continue; /* immediate atom compare */ reg = port->common.u.alive.reg; @@ -7510,3 +7593,23 @@ Port *erts_get_heart_port(void) return NULL; } + +void erts_emergency_close_ports(void) +{ + int ix, max = erts_ptab_max(&erts_port); + + for (ix = 0; ix < max; ix++) { + Port *port = erts_pix2port(ix); + + if (!port) + continue; + /* only examine undead or alive ports */ + if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP) + continue; + + /* emergency close socket */ + if (port->drv_ptr->emergency_close) { + port->drv_ptr->emergency_close((ErlDrvData) port->drv_data); + } + } +} diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab index 68fcc177ae..d3649080dc 100644 --- a/erts/emulator/beam/ops.tab +++ b/erts/emulator/beam/ops.tab @@ -166,22 +166,26 @@ is_tuple Fail=f S | select_tuple_arity S=d Fail=f Size=u Rest=* => \ select_tuple_arity S=d Fail=f Size=u Rest=* => \ gen_select_tuple_arity(S, Fail, Size, Rest) -i_select_val r f I -i_select_val x f I -i_select_val y f I +i_select_val_bins r f I +i_select_val_bins x f I +i_select_val_bins y f I -i_select_val2 r f c f c f -i_select_val2 x f c f c f -i_select_val2 y f c f c f +i_select_val_lins r f I +i_select_val_lins x f I +i_select_val_lins y f I -i_select_tuple_arity2 r f A f A f -i_select_tuple_arity2 x f A f A f -i_select_tuple_arity2 y f A f A f +i_select_val2 r f c c f f +i_select_val2 x f c c f f +i_select_val2 y f c c f f i_select_tuple_arity r f I i_select_tuple_arity x f I i_select_tuple_arity y f I +i_select_tuple_arity2 r f A A f f +i_select_tuple_arity2 x f A A f f +i_select_tuple_arity2 y f A A f f + i_jump_on_val_zero r f I i_jump_on_val_zero x f I i_jump_on_val_zero y f I diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h index 05f07e57b2..251b39508f 100644 --- a/erts/emulator/beam/sys.h +++ b/erts/emulator/beam/sys.h @@ -21,6 +21,25 @@ #define __SYS_H__ +#if defined(DEBUG) || defined(ERTS_ENABLE_LOCK_CHECK) +# undef ERTS_CAN_INLINE +# define ERTS_CAN_INLINE 0 +# undef ERTS_INLINE +# define ERTS_INLINE +#endif + +#if ERTS_CAN_INLINE +#define ERTS_GLB_INLINE static ERTS_INLINE +#else +#define ERTS_GLB_INLINE +#endif + +#if ERTS_CAN_INLINE || defined(ERTS_DO_INCL_GLB_INLINE_FUNC_DEF) +# define ERTS_GLB_INLINE_INCL_FUNC_DEF 1 +#else +# define ERTS_GLB_INLINE_INCL_FUNC_DEF 0 +#endif + #if defined(VALGRIND) && !defined(NO_FPE_SIGNALS) # define NO_FPE_SIGNALS #endif @@ -66,8 +85,12 @@ */ #ifndef ERTS_SYS_FD_TYPE +#define ERTS_SYS_FD_INVALID ((ErtsSysFdType) -1) typedef int ErtsSysFdType; #else +#ifndef ERTS_SYS_FD_INVALID +# error missing ERTS_SYS_FD_INVALID +#endif typedef ERTS_SYS_FD_TYPE ErtsSysFdType; #endif @@ -128,24 +151,8 @@ typedef ERTS_SYS_FD_TYPE ErtsSysFdType; # endif #endif -#if defined(DEBUG) || defined(ERTS_ENABLE_LOCK_CHECK) -# undef ERTS_CAN_INLINE -# define ERTS_CAN_INLINE 0 -# undef ERTS_INLINE -# define ERTS_INLINE -#endif - -#if ERTS_CAN_INLINE -#define ERTS_GLB_INLINE static ERTS_INLINE -#else -#define ERTS_GLB_INLINE -#endif - -#if ERTS_CAN_INLINE || defined(ERTS_DO_INCL_GLB_INLINE_FUNC_DEF) -# define ERTS_GLB_INLINE_INCL_FUNC_DEF 1 -#else -# define ERTS_GLB_INLINE_INCL_FUNC_DEF 0 -#endif +#define ERTS_MK_VSN_INT(Major, Minor, Build) \ + ((((Major) & 0x3ff) << 20) | (((Minor) & 0x3ff) << 10) | ((Build) & 0x3ff)) #ifndef ERTS_EXIT_AFTER_DUMP # define ERTS_EXIT_AFTER_DUMP exit @@ -184,6 +191,32 @@ __decl_noreturn void __noreturn erl_assert_error(const char* expr, const char *f # define ASSERT(e) ((void) 1) #endif +/* ERTS_UNDEF can be used to silence false warnings about + * "variable may be used uninitialized" while keeping the variable + * marked as undefined by valgrind. + */ +#ifdef VALGRIND +# define ERTS_UNDEF(V,I) +#else +# define ERTS_UNDEF(V,I) V = I +#endif + +/* + * Compile time assert + * (the actual compiler error msg can be a bit confusing) + */ +#if ERTS_AT_LEAST_GCC_VSN__(3,1,1) +# define ERTS_CT_ASSERT(e) \ + do { \ + enum { compile_time_assert__ = __builtin_choose_expr((e),0,(void)0) }; \ + } while(0) +#else +# define ERTS_CT_ASSERT(e) \ + do { \ + enum { compile_time_assert__ = 1/(e) }; \ + } while (0) +#endif + /* * Microsoft C/C++: We certainly want to use stdarg.h and prototypes. * But MSC doesn't define __STDC__, unless we compile with the -Za @@ -274,6 +307,7 @@ __decl_noreturn void __noreturn erl_assert_error(const char* expr, const char *f typedef unsigned int Eterm; typedef unsigned int Uint; typedef int Sint; +#define ERTS_UINT_MAX UINT_MAX #define ERTS_SIZEOF_ETERM SIZEOF_INT #define ErtsStrToSint strtol #else @@ -347,23 +381,52 @@ typedef long long Sint; typedef Uint UWord; typedef Sint SWord; +#define ERTS_UINT_MAX ERTS_UWORD_MAX #endif /* HALFWORD_HEAP */ typedef UWord BeamInstr; #ifndef HAVE_INT64 -#if SIZEOF_LONG == 8 -#define HAVE_INT64 1 +# if SIZEOF_LONG == 8 +# define HAVE_INT64 1 typedef unsigned long Uint64; typedef long Sint64; -#elif SIZEOF_LONG_LONG == 8 -#define HAVE_INT64 1 +# ifdef ULONG_MAX +# define ERTS_UINT64_MAX ULONG_MAX +# endif +# ifdef LONG_MAX +# define ERTS_SINT64_MAX LONG_MAX +# endif +# ifdef LONG_MIN +# define ERTS_SINT64_MIN LONG_MIN +# endif +# elif SIZEOF_LONG_LONG == 8 +# define HAVE_INT64 1 typedef unsigned long long Uint64; typedef long long Sint64; -#else -#define HAVE_INT64 0 +# ifdef ULLONG_MAX +# define ERTS_UINT64_MAX ULLONG_MAX +# endif +# ifdef LLONG_MAX +# define ERTS_SINT64_MAX LLONG_MAX +# endif +# ifdef LLONG_MIN +# define ERTS_SINT64_MIN LLONG_MIN +# endif +# else +# error "No 64-bit integer type found" +# endif +#endif + +#ifndef ERTS_UINT64_MAX +# define ERTS_UINT64_MAX (~((Uint64) 0)) +#endif +#ifndef ERTS_SINT64_MAX +# define ERTS_SINT64_MAX ((Sint64) ((((Uint64) 1) << 63)-1)) #endif +#ifndef ERTS_SINT64_MIN +# define ERTS_SINT64_MIN (-1*(((Sint64) 1) << 63)) #endif #if SIZEOF_LONG == 4 @@ -499,7 +562,7 @@ extern volatile int erts_writing_erl_crash_dump; # define NO_ERF # define NO_ERFC /* This definition doesn't take NaN into account, but matherr() gets those */ -# define finite(x) (fabs(x) != HUGE_VAL) +# define isfinite(x) (fabs(x) != HUGE_VAL) # define USE_MATHERR # define HAVE_FINITE #endif @@ -640,14 +703,43 @@ extern char *erts_default_arg0; extern char os_type[]; -extern int sys_init_time(void); +typedef enum { + ERTS_NO_TIME_WARP_MODE, + ERTS_SINGLE_TIME_WARP_MODE, + ERTS_MULTI_TIME_WARP_MODE +} ErtsTimeWarpMode; + +typedef struct { + int have_os_monotonic_time; + ErtsMonotonicTime os_monotonic_time_unit; + ErtsMonotonicTime sys_clock_resolution; + struct { + Uint64 resolution; + char *func; + char *clock_id; + int locked_use; + int extended; + } os_monotonic_time_info; + struct { + Uint64 resolution; + char *func; + char *clock_id; + int locked_use; + } os_system_time_info; +} ErtsSysInitTimeResult; + +#define ERTS_SYS_INIT_TIME_RESULT_INITER \ + {0, (ErtsMonotonicTime) -1, (ErtsMonotonicTime) 1} + +extern void erts_init_sys_time_sup(void); +extern void sys_init_time(ErtsSysInitTimeResult *); +extern void erts_late_sys_init_time(void); extern void erts_deliver_time(void); extern void erts_time_remaining(SysTimeval *); -extern int erts_init_time_sup(void); +extern int erts_init_time_sup(int, ErtsTimeWarpMode); extern void erts_sys_init_float(void); extern void erts_thread_init_float(void); extern void erts_thread_disable_fpe(void); - ERTS_GLB_INLINE int erts_block_fpe(void); ERTS_GLB_INLINE void erts_unblock_fpe(int); @@ -694,7 +786,7 @@ extern char *erts_sys_ddll_error(int code); void erts_sys_schedule_interrupt(int set); #ifdef ERTS_SMP -void erts_sys_schedule_interrupt_timed(int set, erts_short_time_t msec); +void erts_sys_schedule_interrupt_timed(int, ErtsMonotonicTime); void erts_sys_main_thread(void); #endif @@ -733,6 +825,7 @@ int univ_to_local( int local_to_univ(Sint *year, Sint *month, Sint *day, Sint *hour, Sint *minute, Sint *second, int isdst); void get_now(Uint*, Uint*, Uint*); +ErtsMonotonicTime erts_get_monotonic_time(void); void get_sys_now(Uint*, Uint*, Uint*); void set_break_quit(void (*)(void), void (*)(void)); @@ -742,6 +835,16 @@ void init_getenv_state(GETENV_STATE *); char * getenv_string(GETENV_STATE *); void fini_getenv_state(GETENV_STATE *); +#define HAVE_ERTS_CHECK_IO_DEBUG +typedef struct { + int no_used_fds; + int no_driver_select_structs; + int no_driver_event_structs; +} ErtsCheckIoDebugInfo; +int erts_check_io_debug(ErtsCheckIoDebugInfo *ip); + +int erts_sys_is_area_readable(char *start, char *stop); + /* xxxP */ #define SYS_DEFAULT_FLOAT_DECIMALS 20 void init_sys_float(void); @@ -770,6 +873,11 @@ int erts_sys_unsetenv(char *key); char *erts_read_env(char *key); void erts_free_read_env(void *value); +#if defined(ERTS_THR_HAVE_SIG_FUNCS) && !defined(ETHR_UNUSABLE_SIGUSRX) +extern void sys_thr_resume(erts_tid_t tid); +extern void sys_thr_suspend(erts_tid_t tid); +#endif + /* utils.c */ /* Options to sys_alloc_opt */ diff --git a/erts/emulator/beam/time.c b/erts/emulator/beam/time.c index 2fd8e0cf00..2bdda6c8af 100644 --- a/erts/emulator/beam/time.c +++ b/erts/emulator/beam/time.c @@ -83,7 +83,8 @@ #define ASSERT_NO_LOCKED_LOCKS #endif -static erts_smp_mtx_t tiw_lock; +#define ERTS_MONOTONIC_DAY ERTS_SEC_TO_MONOTONIC(60*60*24) +#define ERTS_CLKTCKS_DAY ERTS_MONOTONIC_TO_CLKTCKS(ERTS_MONOTONIC_DAY) /* BEGIN tiw_lock protected variables @@ -91,18 +92,12 @@ static erts_smp_mtx_t tiw_lock; ** The individual timer cells in tiw are also protected by the same mutex. */ +/* timing wheel size NEED to be a power of 2 */ #ifdef SMALL_MEMORY -#define TIW_SIZE 8192 +#define TIW_SIZE (1 << 13) #else -#define TIW_SIZE 65536 /* timing wheel size (should be a power of 2) */ +#define TIW_SIZE (1 << 20) #endif -static ErlTimer** tiw; /* the timing wheel, allocated in init_time() */ -static Uint tiw_pos; /* current position in wheel */ -static Uint tiw_nto; /* number of timeouts in wheel */ -static Uint tiw_min; -static ErlTimer *tiw_min_ptr; - -/* END tiw_lock protected variables */ /* Actual interval time chosen by sys_init_time() */ @@ -114,83 +109,135 @@ static int tiw_itime; /* Constant after init */ # define TIW_ITIME tiw_itime #endif -erts_smp_atomic32_t do_time; /* set at clock interrupt */ -static ERTS_INLINE erts_short_time_t do_time_read(void) +struct ErtsTimerWheel_ { + ErlTimer *w[TIW_SIZE]; + ErtsMonotonicTime pos; + Uint nto; + struct { + ErlTimer *head; + ErlTimer **tail; + Uint nto; + } at_once; + int true_next_timeout_time; + ErtsMonotonicTime next_timeout_time; + erts_atomic64_t next_timeout; + erts_smp_atomic32_t is_bumping; + erts_smp_mtx_t lock; +}; + +ErtsTimerWheel *erts_default_timer_wheel; /* managed by aux thread */ + +static ERTS_INLINE ErtsTimerWheel * +get_timer_wheel(ErlTimer *p) +{ + return (ErtsTimerWheel *) erts_smp_atomic_read_acqb(&p->wheel); +} + +static ERTS_INLINE void +set_timer_wheel(ErlTimer *p, ErtsTimerWheel *tiw) { - return erts_smp_atomic32_read_acqb(&do_time); + erts_smp_atomic_set_relb(&p->wheel, (erts_aint_t) tiw); } -static ERTS_INLINE erts_short_time_t do_time_update(void) +static ERTS_INLINE void +init_next_timeout(ErtsTimerWheel *tiw, + ErtsMonotonicTime time) { - return do_time_read(); + erts_atomic64_init_nob(&tiw->next_timeout, + (erts_aint64_t) time); } -static ERTS_INLINE void do_time_init(void) +static ERTS_INLINE void +set_next_timeout(ErtsTimerWheel *tiw, + ErtsMonotonicTime time, + int true_timeout) { - erts_smp_atomic32_init_nob(&do_time, 0); + tiw->true_next_timeout_time = true_timeout; + tiw->next_timeout_time = time; + erts_atomic64_set_relb(&tiw->next_timeout, + (erts_aint64_t) time); } /* get the time (in units of TIW_ITIME) to the next timeout, or -1 if there are no timeouts */ -static erts_short_time_t next_time_internal(void) /* PRE: tiw_lock taken by caller */ +static ERTS_INLINE ErtsMonotonicTime +find_next_timeout(ErtsTimerWheel *tiw, + ErtsMonotonicTime curr_time, + ErtsMonotonicTime max_search_time) { - int i, tm, nto; - Uint32 min; - ErlTimer* p; - erts_short_time_t dt; - - if (tiw_nto == 0) - return -1; /* no timeouts in wheel */ + int start_ix, tiw_pos_ix; + ErlTimer *p; + int true_min_timeout; + ErtsMonotonicTime min_timeout, min_timeout_pos, slot_timeout_pos, timeout_limit; + + ERTS_SMP_LC_ASSERT(erts_smp_lc_mtx_is_locked(&tiw->lock)); + + if (tiw->true_next_timeout_time) + return tiw->next_timeout_time; - if (tiw_min_ptr) { - min = tiw_min; - dt = do_time_read(); - return ((min >= dt) ? (min - dt) : 0); + /* We never set next timeout beyond timeout_limit */ + timeout_limit = curr_time + ERTS_MONOTONIC_DAY; + + if (tiw->nto == 0) { /* no timeouts in wheel */ + true_min_timeout = tiw->true_next_timeout_time = 0; + min_timeout_pos = ERTS_MONOTONIC_TO_CLKTCKS(timeout_limit); + goto found_next; } - - /* start going through wheel to find next timeout */ - tm = nto = 0; - min = (Uint32) -1; /* max Uint32 */ - i = tiw_pos; + + /* + * Don't want others entering trying to bump + * timers while we are checking... + */ + set_next_timeout(tiw, timeout_limit, 0); + + true_min_timeout = 1; + slot_timeout_pos = tiw->pos; + min_timeout_pos = ERTS_MONOTONIC_TO_CLKTCKS(curr_time + max_search_time); + + start_ix = tiw_pos_ix = (int) (tiw->pos & (TIW_SIZE-1)); + do { - p = tiw[i]; - while (p != NULL) { - nto++; - if (p->count == 0) { - /* found next timeout */ - dt = do_time_read(); - /* p->count is zero */ - tiw_min_ptr = p; - tiw_min = tm; - return ((tm >= dt) ? (tm - dt) : 0); - } else { - /* keep shortest time in 'min' */ - if (tm + p->count*TIW_SIZE < min) { - min = tm + p->count*TIW_SIZE; - tiw_min_ptr = p; - tiw_min = min; - } + slot_timeout_pos++; + if (slot_timeout_pos >= min_timeout_pos) { + true_min_timeout = 0; + break; + } + + p = tiw->w[tiw_pos_ix]; + + while (p) { + ErtsMonotonicTime timeout_pos; + ASSERT(p != p->next); + timeout_pos = p->timeout_pos; + if (min_timeout_pos > timeout_pos) { + min_timeout_pos = timeout_pos; + if (min_timeout_pos <= slot_timeout_pos) + goto found_next; } p = p->next; } - /* when we have found all timeouts the shortest time will be in min */ - if (nto == tiw_nto) break; - tm++; - i = (i + 1) % TIW_SIZE; - } while (i != tiw_pos); - dt = do_time_read(); - if (min <= (Uint32) dt) - return 0; - if ((min - (Uint32) dt) > (Uint32) ERTS_SHORT_TIME_T_MAX) - return ERTS_SHORT_TIME_T_MAX; - return (erts_short_time_t) (min - (Uint32) dt); + + tiw_pos_ix++; + if (tiw_pos_ix == TIW_SIZE) + tiw_pos_ix = 0; + } while (start_ix != tiw_pos_ix); + +found_next: + + min_timeout = ERTS_CLKTCKS_TO_MONOTONIC(min_timeout_pos); + if (min_timeout != tiw->next_timeout_time) + set_next_timeout(tiw, min_timeout, true_min_timeout); + + return min_timeout; } -static void remove_timer(ErlTimer *p) { +static void +remove_timer(ErtsTimerWheel *tiw, ErlTimer *p) +{ /* first */ if (!p->prev) { - tiw[p->slot] = p->next; + tiw->w[p->slot] = p->next; if(p->next) p->next->prev = NULL; } else { @@ -207,79 +254,164 @@ static void remove_timer(ErlTimer *p) { p->next = NULL; p->prev = NULL; - /* Make sure cancel callback isn't called */ - p->active = 0; - tiw_nto--; + + set_timer_wheel(p, NULL); + tiw->nto--; +} + +ErtsMonotonicTime +erts_check_next_timeout_time(ErtsTimerWheel *tiw, + ErtsMonotonicTime max_search_time) +{ + ErtsMonotonicTime next, curr; + + curr = erts_get_monotonic_time(); + + erts_smp_mtx_lock(&tiw->lock); + + next = find_next_timeout(tiw, curr, max_search_time); + + erts_smp_mtx_unlock(&tiw->lock); + + return next; } -/* Private export to erl_time_sup.c */ -erts_short_time_t erts_next_time(void) +#ifndef DEBUG +#define ERTS_DBG_CHK_SAFE_TO_SKIP_TO(TIW, TO) ((void) 0) +#else +#define ERTS_DBG_CHK_SAFE_TO_SKIP_TO(TIW, TO) debug_check_safe_to_skip_to((TIW), (TO)) +static void +debug_check_safe_to_skip_to(ErtsTimerWheel *tiw, ErtsMonotonicTime skip_to_pos) { - erts_short_time_t ret; + int slots, ix; + ErlTimer *tmr; + ErtsMonotonicTime tmp; + + ix = (int) (tiw->pos & (TIW_SIZE-1)); + tmp = skip_to_pos - tiw->pos; + ASSERT(tmp >= 0); + if (tmp < (ErtsMonotonicTime) TIW_SIZE) + slots = (int) tmp; + else + slots = TIW_SIZE; - erts_smp_mtx_lock(&tiw_lock); - (void)do_time_update(); - ret = next_time_internal(); - erts_smp_mtx_unlock(&tiw_lock); - return ret; + while (slots > 0) { + tmr = tiw->w[ix]; + while (tmr) { + ASSERT(tmr->timeout_pos > skip_to_pos); + tmr = tmr->next; + } + ix++; + if (ix == TIW_SIZE) + ix = 0; + slots--; + } } +#endif -static ERTS_INLINE void bump_timer_internal(erts_short_time_t dt) /* PRE: tiw_lock is write-locked */ +void +erts_bump_timers(ErtsTimerWheel *tiw, ErtsMonotonicTime curr_time) { - Uint keep_pos; - Uint count; - ErlTimer *p, **prev, *timeout_head, **timeout_tail; - Uint dtime = (Uint) dt; - - /* no need to bump the position if there aren't any timeouts */ - if (tiw_nto == 0) { - erts_smp_mtx_unlock(&tiw_lock); - return; + int tiw_pos_ix, slots; + ErlTimer *p, *timeout_head, **timeout_tail; + ErtsMonotonicTime bump_to, tmp_slots; + + if (erts_smp_atomic32_cmpxchg_nob(&tiw->is_bumping, 1, 0) != 0) + return; /* Another thread is currently bumping... */ + + bump_to = ERTS_MONOTONIC_TO_CLKTCKS(curr_time); + + erts_smp_mtx_lock(&tiw->lock); + + if (tiw->pos >= bump_to) { + timeout_head = NULL; + goto done; } - /* if do_time > TIW_SIZE we want to go around just once */ - count = (Uint)(dtime / TIW_SIZE) + 1; - keep_pos = (tiw_pos + dtime) % TIW_SIZE; - if (dtime > TIW_SIZE) dtime = TIW_SIZE; - - timeout_head = NULL; - timeout_tail = &timeout_head; - while (dtime > 0) { - /* this is to decrease the counters with the right amount */ - /* when dtime >= TIW_SIZE */ - if (tiw_pos == keep_pos) count--; - prev = &tiw[tiw_pos]; - while ((p = *prev) != NULL) { - ASSERT( p != p->next); - if (p->count < count) { /* we have a timeout */ - /* remove min time */ - if (tiw_min_ptr == p) { - tiw_min_ptr = NULL; - tiw_min = 0; - } + /* Don't want others here while we are bumping... */ + set_next_timeout(tiw, curr_time + ERTS_MONOTONIC_DAY, 0); + if (!tiw->at_once.head) { + timeout_head = NULL; + timeout_tail = &timeout_head; + } + else { + ASSERT(tiw->nto >= tiw->at_once.nto); + timeout_head = tiw->at_once.head; + timeout_tail = tiw->at_once.tail; + tiw->nto -= tiw->at_once.nto; + tiw->at_once.head = NULL; + tiw->at_once.tail = &tiw->at_once.head; + tiw->at_once.nto = 0; + } + + if (tiw->nto == 0) { + ERTS_DBG_CHK_SAFE_TO_SKIP_TO(tiw, bump_to); + tiw->pos = bump_to; + goto done; + } + + if (tiw->true_next_timeout_time) { + ErtsMonotonicTime skip_until_pos; + /* + * No need inspecting slots where we know no timeouts + * to trigger should reside. + */ + + skip_until_pos = ERTS_MONOTONIC_TO_CLKTCKS(tiw->next_timeout_time); + if (skip_until_pos > bump_to) + skip_until_pos = bump_to; + + ERTS_DBG_CHK_SAFE_TO_SKIP_TO(tiw, skip_until_pos); + ASSERT(skip_until_pos > tiw->pos); + + tiw->pos = skip_until_pos - 1; + } + + tiw_pos_ix = (int) ((tiw->pos+1) & (TIW_SIZE-1)); + tmp_slots = (bump_to - tiw->pos); + if (tmp_slots < (ErtsMonotonicTime) TIW_SIZE) + slots = (int) tmp_slots; + else + slots = TIW_SIZE; + + while (slots > 0) { + p = tiw->w[tiw_pos_ix]; + while (p) { + ErlTimer *next = p->next; + ASSERT(p != next); + if (p->timeout_pos <= bump_to) { /* we have a timeout */ /* Remove from list */ - remove_timer(p); + remove_timer(tiw, p); *timeout_tail = p; /* Insert in timeout queue */ timeout_tail = &p->next; } - else { - /* no timeout, just decrease counter */ - p->count -= count; - prev = &p->next; - } + p = next; } - tiw_pos = (tiw_pos + 1) % TIW_SIZE; - dtime--; + tiw_pos_ix++; + if (tiw_pos_ix == TIW_SIZE) + tiw_pos_ix = 0; + slots--; } - tiw_pos = keep_pos; - if (tiw_min_ptr) - tiw_min -= dt; - - erts_smp_mtx_unlock(&tiw_lock); + + ASSERT(tmp_slots >= (ErtsMonotonicTime) TIW_SIZE + || tiw_pos_ix == (int) ((bump_to+1) & (TIW_SIZE-1))); + + tiw->pos = bump_to; + + /* Search at most two seconds ahead... */ + (void) find_next_timeout(tiw, curr_time, ERTS_SEC_TO_MONOTONIC(2)); + +done: + + erts_smp_mtx_unlock(&tiw->lock); + erts_smp_atomic32_set_nob(&tiw->is_bumping, 0); + /* Call timedout timers callbacks */ while (timeout_head) { + ErlTimeoutProc timeout; + void *arg; p = timeout_head; timeout_head = p->next; /* Here comes hairy use of the timer fields! @@ -288,35 +420,69 @@ static ERTS_INLINE void bump_timer_internal(erts_short_time_t dt) /* PRE: tiw_lo * accesses any field until the ->timeout * callback is called. */ + ASSERT(p->timeout_pos <= bump_to); p->next = NULL; p->prev = NULL; p->slot = 0; - (*p->timeout)(p->arg); + timeout = p->timeout; + arg = p->arg; + (*timeout)(arg); } } -void erts_bump_timer(erts_short_time_t dt) /* dt is value from do_time */ +Uint +erts_timer_wheel_memory_size(void) +{ +#ifdef ERTS_SMP + return sizeof(ErtsTimerWheel)*(1 + erts_no_schedulers); +#else + return sizeof(ErtsTimerWheel); +#endif +} + +ErtsTimerWheel * +erts_create_timer_wheel(int no) { - erts_smp_mtx_lock(&tiw_lock); - bump_timer_internal(dt); + ErtsMonotonicTime mtime; + int i; + ErtsTimerWheel *tiw; + tiw = (ErtsTimerWheel *) erts_alloc(ERTS_ALC_T_TIMER_WHEEL, + sizeof(ErtsTimerWheel)); + for(i = 0; i < TIW_SIZE; i++) + tiw->w[i] = NULL; + + erts_smp_atomic32_init_nob(&tiw->is_bumping, 0); + erts_smp_mtx_init_x(&tiw->lock, "timer_wheel", make_small(no)); + + mtime = erts_get_monotonic_time(); + tiw->pos = ERTS_MONOTONIC_TO_CLKTCKS(mtime); + tiw->nto = 0; + tiw->at_once.head = NULL; + tiw->at_once.tail = &tiw->at_once.head; + tiw->at_once.nto = 0; + tiw->true_next_timeout_time = 0; + tiw->next_timeout_time = mtime + ERTS_MONOTONIC_DAY; + init_next_timeout(tiw, mtime + ERTS_MONOTONIC_DAY); + return tiw; } -Uint -erts_timer_wheel_memory_size(void) +ErtsNextTimeoutRef +erts_get_next_timeout_reference(ErtsTimerWheel *tiw) { - return (Uint) TIW_SIZE * sizeof(ErlTimer*); + return (ErtsNextTimeoutRef) &tiw->next_timeout; } + /* this routine links the time cells into a free list at the start and sets the time queue as empty */ void -erts_init_time(void) +erts_init_time(int time_correction, ErtsTimeWarpMode time_warp_mode) { - int i, itime; + int itime; /* system dependent init; must be done before do_time_init() if timer thread is enabled */ - itime = erts_init_time_sup(); + itime = erts_init_time_sup(time_correction, time_warp_mode); #ifdef TIW_ITIME_IS_CONSTANT if (itime != TIW_ITIME) { erl_exit(ERTS_ABORT_EXIT, "timer resolution mismatch %d != %d", itime, TIW_ITIME); @@ -325,117 +491,111 @@ erts_init_time(void) tiw_itime = itime; #endif - erts_smp_mtx_init(&tiw_lock, "timer_wheel"); - - tiw = (ErlTimer**) erts_alloc(ERTS_ALC_T_TIMER_WHEEL, - TIW_SIZE * sizeof(ErlTimer*)); - for(i = 0; i < TIW_SIZE; i++) - tiw[i] = NULL; - do_time_init(); - tiw_pos = tiw_nto = 0; - tiw_min_ptr = NULL; - tiw_min = 0; + erts_default_timer_wheel = erts_create_timer_wheel(0); } +void +erts_set_timer(ErlTimer *p, ErlTimeoutProc timeout, + ErlCancelProc cancel, void *arg, Uint to) +{ + ErtsMonotonicTime timeout_time, timeout_pos; + ErtsMonotonicTime curr_time; + ErtsTimerWheel *tiw; + ErtsSchedulerData *esdp; + + curr_time = erts_get_monotonic_time(); + esdp = erts_get_scheduler_data(); + if (esdp) + tiw = esdp->timer_wheel; + else + tiw = erts_default_timer_wheel; + erts_smp_mtx_lock(&tiw->lock); + if (get_timer_wheel(p)) + ERTS_INTERNAL_ERROR("Double set timer"); -/* -** Insert a process into the time queue, with a timeout 't' -*/ -static void -insert_timer(ErlTimer* p, Uint t) -{ - Uint tm; - Uint64 ticks; + p->timeout = timeout; + p->cancel = cancel; + p->arg = arg; - /* The current slot (tiw_pos) in timing wheel is the next slot to be - * be processed. Hence no extra time tick is needed. - * - * (x + y - 1)/y is precisely the "number of bins" formula. - */ - ticks = (t + (TIW_ITIME - 1)) / TIW_ITIME; + if (to == 0) { + timeout_pos = ERTS_MONOTONIC_TO_CLKTCKS(curr_time); + tiw->nto++; + tiw->at_once.nto++; + *tiw->at_once.tail = p; + tiw->at_once.tail = &p->next; + p->next = NULL; + p->timeout_pos = timeout_pos; + timeout_time = ERTS_CLKTCKS_TO_MONOTONIC(timeout_pos); + } + else { + int tm; + ErtsMonotonicTime ticks; - /* - * Ticks must be a Uint64, or the addition may overflow here, - * resulting in an incorrect value for p->count below. - */ - ticks += do_time_update(); /* Add backlog of unprocessed time */ - - /* calculate slot */ - tm = (ticks + tiw_pos) % TIW_SIZE; - p->slot = (Uint) tm; - p->count = (Uint) (ticks / TIW_SIZE); + ticks = ERTS_MSEC_TO_CLKTCKS(to); + timeout_pos = ERTS_MONOTONIC_TO_CLKTCKS(curr_time - 1) + 1 + ticks; + + /* calculate slot */ + tm = (int) (timeout_pos & (TIW_SIZE-1)); + p->slot = (Uint) tm; - /* insert at head of list at slot */ - p->next = tiw[tm]; - p->prev = NULL; - if (p->next != NULL) - p->next->prev = p; - tiw[tm] = p; + /* insert at head of list at slot */ + p->next = tiw->w[tm]; + p->prev = NULL; + if (p->next != NULL) + p->next->prev = p; + tiw->w[tm] = p; + tiw->nto++; - /* insert min time */ - if ((tiw_nto == 0) || ((tiw_min_ptr != NULL) && (ticks < tiw_min))) { - tiw_min = ticks; - tiw_min_ptr = p; - } - if ((tiw_min_ptr == p) && (ticks > tiw_min)) { - /* some other timer might be 'min' now */ - tiw_min = 0; - tiw_min_ptr = NULL; + timeout_time = ERTS_CLKTCKS_TO_MONOTONIC(timeout_pos); + p->timeout_pos = timeout_pos; + + ASSERT(ERTS_MSEC_TO_MONOTONIC(to) <= timeout_time - curr_time); + ASSERT(timeout_time - curr_time + < ERTS_MSEC_TO_MONOTONIC(to) + ERTS_CLKTCKS_TO_MONOTONIC(1)); } - tiw_nto++; -} + if (timeout_time < tiw->next_timeout_time) + set_next_timeout(tiw, timeout_time, 1); -void -erts_set_timer(ErlTimer* p, ErlTimeoutProc timeout, ErlCancelProc cancel, - void* arg, Uint t) -{ + set_timer_wheel(p, tiw); + + erts_smp_mtx_unlock(&tiw->lock); - erts_deliver_time(); - erts_smp_mtx_lock(&tiw_lock); - if (p->active) { /* XXX assert ? */ - erts_smp_mtx_unlock(&tiw_lock); - return; - } - p->timeout = timeout; - p->cancel = cancel; - p->arg = arg; - p->active = 1; - insert_timer(p, t); - erts_smp_mtx_unlock(&tiw_lock); #if defined(ERTS_SMP) - if (t <= (Uint) ERTS_SHORT_TIME_T_MAX) - erts_sys_schedule_interrupt_timed(1, (erts_short_time_t) t); + if (tiw == erts_default_timer_wheel) + erts_interupt_aux_thread_timed(timeout_time); #endif + } void -erts_cancel_timer(ErlTimer* p) +erts_cancel_timer(ErlTimer *p) { - erts_smp_mtx_lock(&tiw_lock); - if (!p->active) { /* allow repeated cancel (drivers) */ - erts_smp_mtx_unlock(&tiw_lock); + ErtsTimerWheel *tiw; + ErlCancelProc cancel; + void *arg = NULL; /* Shut up faulty warning... */ + + tiw = get_timer_wheel(p); + if (!tiw) return; - } + + erts_smp_mtx_lock(&tiw->lock); + if (tiw != get_timer_wheel(p)) + cancel = NULL; + else { + remove_timer(tiw, p); + p->slot = 0; - /* is it the 'min' timer, remove min */ - if (p == tiw_min_ptr) { - tiw_min_ptr = NULL; - tiw_min = 0; + cancel = p->cancel; + arg = p->arg; } + erts_smp_mtx_unlock(&tiw->lock); - remove_timer(p); - p->slot = p->count = 0; - - if (p->cancel != NULL) { - erts_smp_mtx_unlock(&tiw_lock); - (*p->cancel)(p->arg); - return; - } - erts_smp_mtx_unlock(&tiw_lock); + if (cancel) + (*cancel)(arg); } /* @@ -447,59 +607,58 @@ erts_cancel_timer(ErlTimer* p) Uint erts_time_left(ErlTimer *p) { - Uint left; - erts_short_time_t dt; + ErtsTimerWheel *tiw; + ErtsMonotonicTime current_time, timeout_time; - erts_smp_mtx_lock(&tiw_lock); - - if (!p->active) { - erts_smp_mtx_unlock(&tiw_lock); + tiw = get_timer_wheel(p); + if (!tiw) return 0; - } - if (p->slot < tiw_pos) - left = (p->count + 1) * TIW_SIZE + p->slot - tiw_pos; + erts_smp_mtx_lock(&tiw->lock); + if (tiw != get_timer_wheel(p)) + timeout_time = ERTS_MONOTONIC_TIME_MIN; else - left = p->count * TIW_SIZE + p->slot - tiw_pos; - dt = do_time_read(); - if (left < dt) - left = 0; - else - left -= dt; - - erts_smp_mtx_unlock(&tiw_lock); + timeout_time = ERTS_CLKTCKS_TO_MONOTONIC(p->timeout_pos); + erts_smp_mtx_unlock(&tiw->lock); - return (Uint) left * TIW_ITIME; + current_time = erts_get_monotonic_time(); + if (timeout_time <= current_time) + return 0; + return (Uint) ERTS_MONOTONIC_TO_MSEC(timeout_time - current_time); } #ifdef DEBUG void erts_p_slpq(void) { + ErtsTimerWheel *tiw = erts_default_timer_wheel; + ErtsMonotonicTime current_time = erts_get_monotonic_time(); int i; ErlTimer* p; - erts_smp_mtx_lock(&tiw_lock); + erts_smp_mtx_lock(&tiw->lock); /* print the whole wheel, starting at the current position */ - erts_printf("\ntiw_pos = %d tiw_nto %d\n", tiw_pos, tiw_nto); - i = tiw_pos; - if (tiw[i] != NULL) { + erts_printf("\ncurrent time = %bps tiw_pos = %d tiw_nto %d\n", + current_time, tiw->pos, tiw->nto); + i = tiw->pos; + if (tiw->w[i] != NULL) { erts_printf("%d:\n", i); - for(p = tiw[i]; p != NULL; p = p->next) { - erts_printf(" (count %d, slot %d)\n", - p->count, p->slot); + for(p = tiw->w[i]; p != NULL; p = p->next) { + erts_printf(" (timeout time %bps, slot %d)\n", + ERTS_CLKTCKS_TO_MONOTONIC(p->timeout_pos), + p->slot); } } - for(i = (i+1)%TIW_SIZE; i != tiw_pos; i = (i+1)%TIW_SIZE) { - if (tiw[i] != NULL) { + for(i = ((i+1) & (TIW_SIZE-1)); i != (tiw->pos & (TIW_SIZE-1)); i = ((i+1) & (TIW_SIZE-1))) { + if (tiw->w[i] != NULL) { erts_printf("%d:\n", i); - for(p = tiw[i]; p != NULL; p = p->next) { - erts_printf(" (count %d, slot %d)\n", - p->count, p->slot); + for(p = tiw->w[i]; p != NULL; p = p->next) { + erts_printf(" (timeout time %bps, slot %d)\n", + ERTS_CLKTCKS_TO_MONOTONIC(p->timeout_pos), p->slot); } } } - erts_smp_mtx_unlock(&tiw_lock); + erts_smp_mtx_unlock(&tiw->lock); } #endif /* DEBUG */ diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c index 738f793020..8fc8962e4f 100644 --- a/erts/emulator/beam/utils.c +++ b/erts/emulator/beam/utils.c @@ -48,6 +48,11 @@ #include "erl_sched_spec_pre_alloc.h" #include "beam_bp.h" #include "erl_ptab.h" +#include "erl_check_io.h" +#include "erl_bif_unique.h" +#ifdef HIPE +# include "hipe_mode_switch.h" +#endif #undef M_TRIM_THRESHOLD #undef M_TOP_PAD @@ -186,12 +191,18 @@ erts_set_hole_marker(Eterm* ptr, Uint sz) * Helper function for the ESTACK macros defined in global.h. */ void -erl_grow_estack(ErtsEStack* s, Eterm* default_estack) +erl_grow_estack(ErtsEStack* s, Uint need) { Uint old_size = (s->end - s->start); - Uint new_size = old_size * 2; + Uint new_size; Uint sp_offs = s->sp - s->start; - if (s->start != default_estack) { + + if (need < old_size) + new_size = 2*old_size; + else + new_size = ((need / old_size) + 2) * old_size; + + if (s->start != s->edefault) { s->start = erts_realloc(s->alloc_type, s->start, new_size*sizeof(Eterm)); } else { @@ -206,12 +217,18 @@ erl_grow_estack(ErtsEStack* s, Eterm* default_estack) * Helper function for the WSTACK macros defined in global.h. */ void -erl_grow_wstack(ErtsWStack* s, UWord* default_wstack) +erl_grow_wstack(ErtsWStack* s, Uint need) { Uint old_size = (s->wend - s->wstart); - Uint new_size = old_size * 2; + Uint new_size; Uint sp_offs = s->wsp - s->wstart; - if (s->wstart != default_wstack) { + + if (need < old_size) + new_size = 2 * old_size; + else + new_size = ((need / old_size) + 2) * old_size; + + if (s->wstart != s->wdefault) { s->wstart = erts_realloc(s->alloc_type, s->wstart, new_size*sizeof(UWord)); } else { @@ -223,6 +240,32 @@ erl_grow_wstack(ErtsWStack* s, UWord* default_wstack) s->wsp = s->wstart + sp_offs; } +/* + * Helper function for the PSTACK macros defined in global.h. + */ +void +erl_grow_pstack(ErtsPStack* s, void* default_pstack, unsigned need_bytes) +{ + Uint old_size = s->pend - s->pstart; + Uint new_size; + Uint sp_offs = s->psp - s->pstart; + + if (need_bytes < old_size) + new_size = 2 * old_size; + else + new_size = ((need_bytes / old_size) + 2) * old_size; + + if (s->pstart != default_pstack) { + s->pstart = erts_realloc(s->alloc_type, s->pstart, new_size); + } else { + byte* new_ptr = erts_alloc(s->alloc_type, new_size); + sys_memcpy(new_ptr, s->pstart, old_size); + s->pstart = new_ptr; + } + s->pend = s->pstart + new_size; + s->psp = s->pstart + sp_offs; +} + /* CTYPE macros */ #define LATIN1 @@ -310,6 +353,17 @@ int erts_fit_in_bits_int32(Sint32 value) return fit_in_bits((Sint64) (Uint32) value, 4); } +int erts_fit_in_bits_uint(Uint value) +{ +#if ERTS_SIZEOF_ETERM == 4 + return fit_in_bits((Sint64) (Uint32) value, 4); +#elif ERTS_SIZEOF_ETERM == 8 + return fit_in_bits(value, 5); +#else +# error "No way, Jose" +#endif +} + int erts_print(int to, void *arg, char *format, ...) { @@ -706,7 +760,7 @@ erts_bld_atom_2uint_3tup_list(Uint **hpp, Uint *szp, Sint length, ** If N < 0, Y = FUNNY_NUMBER4 else Y = FUNNY_NUMBER3. ** The hash value is Y*h(J) mod 2^32 where h(J) is calculated like ** h(0) = <initial hash> -** h(i) = h(i-i)*X + B(i-1) +** h(i) = h(i-1)*X + B(i-1) ** The above should hold regardless of internal representation. ** Pids are hashed like small numbers but with differrent constants, as are ** ports. @@ -788,10 +842,10 @@ Uint32 make_hash(Eterm term_arg) unsigned op; /* Must not collide with the real tag_val_def's: */ -#define MAKE_HASH_TUPLE_OP 0x11 -#define MAKE_HASH_TERM_ARRAY_OP 0x12 -#define MAKE_HASH_CDR_PRE_OP 0x13 -#define MAKE_HASH_CDR_POST_OP 0x14 +#define MAKE_HASH_TUPLE_OP (FIRST_VACANT_TAG_DEF) +#define MAKE_HASH_TERM_ARRAY_OP (FIRST_VACANT_TAG_DEF+1) +#define MAKE_HASH_CDR_PRE_OP (FIRST_VACANT_TAG_DEF+2) +#define MAKE_HASH_CDR_POST_OP (FIRST_VACANT_TAG_DEF+3) /* ** Convenience macro for calculating a bytewise hash on an unsigned 32 bit @@ -971,23 +1025,8 @@ tail_recur: break; } case MAP_DEF: - { - map_t *mp = (map_t *)map_val(term); - int size = map_get_size(mp); - Eterm *ks = map_get_keys(mp); - Eterm *vs = map_get_values(mp); - - /* Use a prime with size to remedy some of - * the {} and <<>> hash problems */ - hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + size; - if (size == 0) - break; - - /* push values first */ - WSTACK_PUSH3(stack, (UWord)vs, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); - WSTACK_PUSH3(stack, (UWord)ks, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); - break; - } + hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + make_hash2(term); + break; case TUPLE_DEF: { Eterm* ptr = tuple_val(term); @@ -1091,10 +1130,11 @@ Uint32 make_hash2(Eterm term) { Uint32 hash; - Uint32 hash_xor_keys = 0; - Uint32 hash_xor_values = 0; + Uint32 hash_xor_pairs; DeclareTmpHeapNoproc(tmp_big,2); + ERTS_UNDEF(hash_xor_pairs, 0); + /* (HCONST * {2, ..., 16}) mod 2^32 */ #define HCONST_2 0x3c6ef372UL #define HCONST_3 0xdaa66d2bUL @@ -1111,10 +1151,15 @@ make_hash2(Eterm term) #define HCONST_14 0xa708a81eUL #define HCONST_15 0x454021d7UL #define HCONST_16 0xe3779b90UL +#define HCONST_17 0x81af1549UL +#define HCONST_18 0x1fe68f02UL +#define HCONST_19 0xbe1e08bbUL +#define HCONST_20 0x5c558274UL +#define HCONST_21 0xfa8cfc2dUL #define HASH_MAP_TAIL (_make_header(1,_TAG_HEADER_REF)) -#define HASH_MAP_KEY (_make_header(2,_TAG_HEADER_REF)) -#define HASH_MAP_VAL (_make_header(3,_TAG_HEADER_REF)) +#define HASH_MAP_PAIR (_make_header(2,_TAG_HEADER_REF)) +#define HASH_CDR (_make_header(3,_TAG_HEADER_REF)) #define UINT32_HASH_2(Expr1, Expr2, AConst) \ do { \ @@ -1137,6 +1182,13 @@ make_hash2(Eterm term) } while(0) #define IS_SSMALL28(x) (((Uint) (((x) >> (28-1)) + 1)) < 2) + +#ifdef ARCH_64 +# define POINTER_HASH(Ptr, AConst) UINT32_HASH_2((Uint32)(UWord)(Ptr), (((UWord)(Ptr)) >> 32), AConst) +#else +# define POINTER_HASH(Ptr, AConst) UINT32_HASH(Ptr, AConst) +#endif + /* Optimization. Simple cases before declaration of estack. */ if (primary_tag(term) == TAG_PRIMARY_IMMED1) { switch (term & _TAG_IMMED1_MASK) { @@ -1191,9 +1243,9 @@ make_hash2(Eterm term) if (c > 0) UINT32_HASH(sh, HCONST_4); if (is_list(term)) { - term = *ptr; - tmp = *++ptr; - ESTACK_PUSH(s, tmp); + tmp = CDR(ptr); + ESTACK_PUSH(s, tmp); + term = CAR(ptr); } } break; @@ -1210,51 +1262,90 @@ make_hash2(Eterm term) UINT32_HASH(arity, HCONST_9); if (arity == 0) /* Empty tuple */ goto hash2_common; - for (i = arity; i >= 1; i--) { - tmp = elem[i]; - ESTACK_PUSH(s, tmp); - } - goto hash2_common; - } - break; - case MAP_SUBTAG: - { - map_t *mp = (map_t *)map_val(term); - int i; - int size = map_get_size(mp); - Eterm *ks = map_get_keys(mp); - Eterm *vs = map_get_values(mp); - UINT32_HASH(size, HCONST_16); - if (size == 0) { - goto hash2_common; - } - ESTACK_PUSH(s, hash_xor_values); - ESTACK_PUSH(s, hash_xor_keys); - ESTACK_PUSH(s, hash); - ESTACK_PUSH(s, HASH_MAP_TAIL); - hash = 0; - hash_xor_keys = 0; - hash_xor_values = 0; - for (i = size - 1; i >= 0; i--) { - tmp = vs[i]; - ESTACK_PUSH(s, HASH_MAP_VAL); - ESTACK_PUSH(s, tmp); - } - /* We do not want to expose the tuple representation. - * Do not push the keys as a tuple. - */ - for (i = size - 1; i >= 0; i--) { - tmp = ks[i]; - ESTACK_PUSH(s, HASH_MAP_KEY); - ESTACK_PUSH(s, tmp); + for (i = arity; ; i--) { + term = elem[i]; + if (i == 1) + break; + ESTACK_PUSH(s, term); } - goto hash2_common; } break; + case MAP_SUBTAG: + { + Eterm* ptr = boxed_val(term) + 1; + Uint size; + int i; + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_FLATMAP: + { + flatmap_t *mp = (flatmap_t *)flatmap_val(term); + Eterm *ks = flatmap_get_keys(mp); + Eterm *vs = flatmap_get_values(mp); + size = flatmap_get_size(mp); + UINT32_HASH(size, HCONST_16); + if (size == 0) + goto hash2_common; + + /* We want a portable hash function that is *independent* of + * the order in which keys and values are encountered. + * We therefore calculate context independent hashes for all . + * key-value pairs and then xor them together. + */ + ESTACK_PUSH(s, hash_xor_pairs); + ESTACK_PUSH(s, hash); + ESTACK_PUSH(s, HASH_MAP_TAIL); + hash = 0; + hash_xor_pairs = 0; + for (i = size - 1; i >= 0; i--) { + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, vs[i]); + ESTACK_PUSH(s, ks[i]); + } + goto hash2_common; + } + + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_HEAD_BITMAP: + size = *ptr++; + UINT32_HASH(size, HCONST_16); + if (size == 0) + goto hash2_common; + ESTACK_PUSH(s, hash_xor_pairs); + ESTACK_PUSH(s, hash); + ESTACK_PUSH(s, HASH_MAP_TAIL); + hash = 0; + hash_xor_pairs = 0; + } + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + i = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + case HAMT_SUBTAG_NODE_BITMAP: + i = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + break; + default: + erl_exit(1, "bad header"); + } + while (i) { + if (is_list(*ptr)) { + Eterm* cons = list_val(*ptr); + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, CDR(cons)); + ESTACK_PUSH(s, CAR(cons)); + } + else { + ASSERT(is_boxed(*ptr)); + ESTACK_PUSH(s, *ptr); + } + i--; ptr++; + } + goto hash2_common; + } + break; case EXPORT_SUBTAG: { Export* ep = *((Export **) (export_val(term) + 1)); - UINT32_HASH_2 (ep->code[2], atom_tab(atom_val(ep->code[0]))->slot.bucket.hvalue, @@ -1269,7 +1360,6 @@ make_hash2(Eterm term) { ErlFunThing* funp = (ErlFunThing *) fun_val(term); Uint num_free = funp->num_free; - UINT32_HASH_2 (num_free, atom_tab(atom_val(funp->fe->module))->slot.bucket.hvalue, @@ -1350,7 +1440,8 @@ make_hash2(Eterm term) do { Uint t; Uint32 x, y; - t = i < n ? BIG_DIGIT(ptr, i++) : 0; + ASSERT(i < n); + t = BIG_DIGIT(ptr, i++); x = t & 0xffffffff; y = t >> 32; UINT32_HASH_2(x, y, con); @@ -1458,20 +1549,395 @@ make_hash2(Eterm term) switch (term) { case HASH_MAP_TAIL: { hash = (Uint32) ESTACK_POP(s); - UINT32_HASH(hash_xor_keys, HCONST_16); - UINT32_HASH(hash_xor_values, HCONST_16); - hash_xor_keys = (Uint32) ESTACK_POP(s); - hash_xor_values = (Uint32) ESTACK_POP(s); + UINT32_HASH(hash_xor_pairs, HCONST_19); + hash_xor_pairs = (Uint32) ESTACK_POP(s); goto hash2_common; } - case HASH_MAP_KEY: - hash_xor_keys ^= hash; + case HASH_MAP_PAIR: + hash_xor_pairs ^= hash; hash = 0; goto hash2_common; - case HASH_MAP_VAL: - hash_xor_values ^= hash; + default: + break; + } + } + } + } +} + +/* Term hash function for internal use. + * + * Limitation #1: Is not "portable" in any way between different VM instances. + * + * Limitation #2: The hash value is only valid as long as the term exists + * somewhere in the VM. Why? Because external pids, ports and refs are hashed + * by mixing the node *pointer* value. If a node disappears and later reappears + * with a new ErlNode struct, externals from that node will hash different than + * before. + * + * One IMPORTANT property must hold (for hamt). + * EVERY BIT of the term that is significant for equality (see EQ) + * MUST BE USED AS INPUT FOR THE HASH. Two different terms must always have a + * chance of hashing different when salted: hash([Salt|A]) vs hash([Salt|B]). + * + * This is why we can not use cached hash values for atoms for example. + * + */ + +#define CONST_HASH(AConst) \ +do { /* Lightweight mixing of constant (type info) */ \ + hash ^= AConst; \ + hash = (hash << 17) ^ (hash >> (32-17)); \ +} while (0) + +Uint32 +make_internal_hash(Eterm term) +{ + Uint32 hash; + Uint32 hash_xor_pairs; + + ERTS_UNDEF(hash_xor_pairs, 0); + + /* Optimization. Simple cases before declaration of estack. */ + if (primary_tag(term) == TAG_PRIMARY_IMMED1) { + hash = 0; + #if ERTS_SIZEOF_ETERM == 8 + UINT32_HASH_2((Uint32)term, (Uint32)(term >> 32), HCONST); + #elif ERTS_SIZEOF_ETERM == 4 + UINT32_HASH(term, HCONST); + #else + # error "No you don't" + #endif + return hash; + } + { + Eterm tmp; + DECLARE_ESTACK(s); + + UseTmpHeapNoproc(2); + hash = 0; + for (;;) { + switch (primary_tag(term)) { + case TAG_PRIMARY_LIST: + { + int c = 0; + Uint32 sh = 0; + Eterm* ptr = list_val(term); + while (is_byte(*ptr)) { + /* Optimization for strings. */ + sh = (sh << 8) + unsigned_val(*ptr); + if (c == 3) { + UINT32_HASH(sh, HCONST_4); + c = sh = 0; + } else { + c++; + } + term = CDR(ptr); + if (is_not_list(term)) + break; + ptr = list_val(term); + } + if (c > 0) + UINT32_HASH(sh, HCONST_4); + if (is_list(term)) { + tmp = CDR(ptr); + CONST_HASH(HCONST_17); /* Hash CAR in cons cell */ + ESTACK_PUSH(s, tmp); + if (is_not_list(tmp)) { + ESTACK_PUSH(s, HASH_CDR); + } + term = CAR(ptr); + } + } + break; + case TAG_PRIMARY_BOXED: + { + Eterm hdr = *boxed_val(term); + ASSERT(is_header(hdr)); + switch (hdr & _TAG_HEADER_MASK) { + case ARITYVAL_SUBTAG: + { + int i; + int arity = header_arity(hdr); + Eterm* elem = tuple_val(term); + UINT32_HASH(arity, HCONST_9); + if (arity == 0) /* Empty tuple */ + goto pop_next; + for (i = arity; ; i--) { + term = elem[i]; + if (i == 1) + break; + ESTACK_PUSH(s, term); + } + } + break; + + case MAP_SUBTAG: + { + Eterm* ptr = boxed_val(term) + 1; + Uint size; + int i; + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_FLATMAP: + { + flatmap_t *mp = (flatmap_t *)flatmap_val(term); + Eterm *ks = flatmap_get_keys(mp); + Eterm *vs = flatmap_get_values(mp); + size = flatmap_get_size(mp); + UINT32_HASH(size, HCONST_16); + if (size == 0) + goto pop_next; + + /* We want a hash function that is *independent* of + * the order in which keys and values are encountered. + * We therefore calculate context independent hashes for all . + * key-value pairs and then xor them together. + */ + ESTACK_PUSH(s, hash_xor_pairs); + ESTACK_PUSH(s, hash); + ESTACK_PUSH(s, HASH_MAP_TAIL); + hash = 0; + hash_xor_pairs = 0; + for (i = size - 1; i >= 0; i--) { + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, vs[i]); + ESTACK_PUSH(s, ks[i]); + } + goto pop_next; + } + case HAMT_SUBTAG_HEAD_ARRAY: + case HAMT_SUBTAG_HEAD_BITMAP: + size = *ptr++; + UINT32_HASH(size, HCONST_16); + if (size == 0) + goto pop_next; + ESTACK_PUSH(s, hash_xor_pairs); + ESTACK_PUSH(s, hash); + ESTACK_PUSH(s, HASH_MAP_TAIL); + hash = 0; + hash_xor_pairs = 0; + } + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + i = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + case HAMT_SUBTAG_NODE_BITMAP: + i = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + break; + default: + erl_exit(1, "bad header"); + } + while (i) { + if (is_list(*ptr)) { + Eterm* cons = list_val(*ptr); + ESTACK_PUSH(s, HASH_MAP_PAIR); + ESTACK_PUSH(s, CDR(cons)); + ESTACK_PUSH(s, CAR(cons)); + } + else { + ASSERT(is_boxed(*ptr)); + ESTACK_PUSH(s, *ptr); + } + i--; ptr++; + } + goto pop_next; + } + break; + case EXPORT_SUBTAG: + { + Export* ep = *((Export **) (export_val(term) + 1)); + /* Assumes Export entries never moves */ + POINTER_HASH(ep, HCONST_14); + goto pop_next; + } + + case FUN_SUBTAG: + { + ErlFunThing* funp = (ErlFunThing *) fun_val(term); + Uint num_free = funp->num_free; + UINT32_HASH_2(num_free, funp->fe->module, HCONST_20); + UINT32_HASH_2(funp->fe->old_index, funp->fe->old_uniq, HCONST_21); + if (num_free == 0) { + goto pop_next; + } else { + Eterm* bptr = funp->env + num_free - 1; + while (num_free-- > 1) { + term = *bptr--; + ESTACK_PUSH(s, term); + } + term = *bptr; + } + } + break; + case REFC_BINARY_SUBTAG: + case HEAP_BINARY_SUBTAG: + case SUB_BINARY_SUBTAG: + { + byte* bptr; + unsigned sz = binary_size(term); + Uint32 con = HCONST_13 + hash; + Uint bitoffs; + Uint bitsize; + + ERTS_GET_BINARY_BYTES(term, bptr, bitoffs, bitsize); + if (sz == 0 && bitsize == 0) { + hash = con; + } else { + if (bitoffs == 0) { + hash = block_hash(bptr, sz, con); + if (bitsize > 0) { + UINT32_HASH_2(bitsize, (bptr[sz] >> (8 - bitsize)), + HCONST_15); + } + } else { + byte* buf = (byte *) erts_alloc(ERTS_ALC_T_TMP, + sz + (bitsize != 0)); + erts_copy_bits(bptr, bitoffs, 1, buf, 0, 1, sz*8+bitsize); + hash = block_hash(buf, sz, con); + if (bitsize > 0) { + UINT32_HASH_2(bitsize, (buf[sz] >> (8 - bitsize)), + HCONST_15); + } + erts_free(ERTS_ALC_T_TMP, (void *) buf); + } + } + goto pop_next; + } + break; + case POS_BIG_SUBTAG: + case NEG_BIG_SUBTAG: + { + Eterm* ptr = big_val(term); + Uint i = 0; + Uint n = BIG_SIZE(ptr); + Uint32 con = BIG_SIGN(ptr) ? HCONST_10 : HCONST_11; +#if D_EXP == 16 + do { + Uint32 x, y; + x = i < n ? BIG_DIGIT(ptr, i++) : 0; + x += (Uint32)(i < n ? BIG_DIGIT(ptr, i++) : 0) << 16; + y = i < n ? BIG_DIGIT(ptr, i++) : 0; + y += (Uint32)(i < n ? BIG_DIGIT(ptr, i++) : 0) << 16; + UINT32_HASH_2(x, y, con); + } while (i < n); +#elif D_EXP == 32 + do { + Uint32 x, y; + x = i < n ? BIG_DIGIT(ptr, i++) : 0; + y = i < n ? BIG_DIGIT(ptr, i++) : 0; + UINT32_HASH_2(x, y, con); + } while (i < n); +#elif D_EXP == 64 + do { + Uint t; + Uint32 x, y; + ASSERT(i < n); + t = BIG_DIGIT(ptr, i++); + x = t & 0xffffffff; + y = t >> 32; + UINT32_HASH_2(x, y, con); + } while (i < n); +#else +#error "unsupported D_EXP size" +#endif + goto pop_next; + } + break; + case REF_SUBTAG: + UINT32_HASH(internal_ref_numbers(term)[0], HCONST_7); + ASSERT(internal_ref_no_of_numbers(term) == 3); + UINT32_HASH_2(internal_ref_numbers(term)[1], + internal_ref_numbers(term)[2], HCONST_8); + goto pop_next; + + case EXTERNAL_REF_SUBTAG: + { + ExternalThing* thing = external_thing_ptr(term); + + ASSERT(external_thing_ref_no_of_numbers(thing) == 3); + /* See limitation #2 */ + #ifdef ARCH_64 + POINTER_HASH(thing->node, HCONST_7); + UINT32_HASH(external_thing_ref_numbers(thing)[0], HCONST_7); + #else + UINT32_HASH_2(thing->node, + external_thing_ref_numbers(thing)[0], HCONST_7); + #endif + UINT32_HASH_2(external_thing_ref_numbers(thing)[1], + external_thing_ref_numbers(thing)[2], HCONST_8); + goto pop_next; + } + case EXTERNAL_PID_SUBTAG: { + ExternalThing* thing = external_thing_ptr(term); + /* See limitation #2 */ + #ifdef ARCH_64 + POINTER_HASH(thing->node, HCONST_5); + UINT32_HASH(thing->data.ui[0], HCONST_5); + #else + UINT32_HASH_2(thing->node, thing->data.ui[0], HCONST_5); + #endif + goto pop_next; + } + case EXTERNAL_PORT_SUBTAG: { + ExternalThing* thing = external_thing_ptr(term); + /* See limitation #2 */ + #ifdef ARCH_64 + POINTER_HASH(thing->node, HCONST_6); + UINT32_HASH(thing->data.ui[0], HCONST_6); + #else + UINT32_HASH_2(thing->node, thing->data.ui[0], HCONST_6); + #endif + goto pop_next; + } + case FLOAT_SUBTAG: + { + FloatDef ff; + GET_DOUBLE(term, ff); + UINT32_HASH_2(ff.fw[0], ff.fw[1], HCONST_12); + goto pop_next; + } + + default: + erl_exit(1, "Invalid tag in make_hash2(0x%X)\n", term); + } + } + break; + case TAG_PRIMARY_IMMED1: + #if ERTS_SIZEOF_ETERM == 8 + UINT32_HASH_2((Uint32)term, (Uint32)(term >> 32), HCONST); + #else + UINT32_HASH(term, HCONST); + #endif + goto pop_next; + + default: + erl_exit(1, "Invalid tag in make_hash2(0x%X)\n", term); + + pop_next: + if (ESTACK_ISEMPTY(s)) { + DESTROY_ESTACK(s); + UnUseTmpHeapNoproc(2); + return hash; + } + + term = ESTACK_POP(s); + + switch (term) { + case HASH_MAP_TAIL: { + hash = (Uint32) ESTACK_POP(s); + UINT32_HASH(hash_xor_pairs, HCONST_19); + hash_xor_pairs = (Uint32) ESTACK_POP(s); + goto pop_next; + } + case HASH_MAP_PAIR: + hash_xor_pairs ^= hash; hash = 0; - goto hash2_common; + goto pop_next; + + case HASH_CDR: + CONST_HASH(HCONST_18); /* Hash CDR i cons cell */ + goto pop_next; default: break; } @@ -1479,9 +1945,10 @@ make_hash2(Eterm term) } } +#undef CONST_HASH #undef HASH_MAP_TAIL -#undef HASH_MAP_KEY -#undef HASH_MAP_VAL +#undef HASH_MAP_PAIR +#undef HASH_CDR #undef UINT32_HASH_2 #undef UINT32_HASH @@ -1698,23 +2165,8 @@ tail_recur: break; case MAP_DEF: - { - map_t *mp = (map_t *)map_val(term); - int size = map_get_size(mp); - Eterm *ks = map_get_keys(mp); - Eterm *vs = map_get_values(mp); - - /* Use a prime with size to remedy some of - * the {} and <<>> hash problems */ - hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + size; - if (size == 0) - break; - - /* push values first */ - WSTACK_PUSH3(stack, (UWord)vs, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); - WSTACK_PUSH3(stack, (UWord)ks, (UWord) size, MAKE_HASH_TERM_ARRAY_OP); - break; - } + hash = hash*FUNNY_NUMBER13 + FUNNY_NUMBER14 + make_hash2(term); + break; case TUPLE_DEF: { Eterm* ptr = tuple_val(term); @@ -2119,22 +2571,6 @@ tailrecur_ne: ++bb; goto term_array; } - case MAP_SUBTAG: - { - aa = map_val_rel(a, a_base); - if (!is_boxed(b) || *boxed_val_rel(b,b_base) != *aa) - goto not_equal; - bb = map_val_rel(b,b_base); - sz = map_get_size((map_t*)aa); - - if (sz != map_get_size((map_t*)bb)) goto not_equal; - if (sz == 0) goto pop_next; - - aa += 2; - bb += 2; - sz += 1; /* increment for tuple-keys */ - goto term_array; - } case REFC_BINARY_SUBTAG: case HEAP_BINARY_SUBTAG: case SUB_BINARY_SUBTAG: @@ -2326,6 +2762,46 @@ tailrecur_ne: } break; /* not equal */ } + case MAP_SUBTAG: + if (is_flatmap_rel(a, a_base)) { + aa = flatmap_val_rel(a, a_base); + if (!is_boxed(b) || *boxed_val_rel(b,b_base) != *aa) + goto not_equal; + bb = flatmap_val_rel(b,b_base); + sz = flatmap_get_size((flatmap_t*)aa); + + if (sz != flatmap_get_size((flatmap_t*)bb)) goto not_equal; + if (sz == 0) goto pop_next; + + aa += 2; + bb += 2; + sz += 1; /* increment for tuple-keys */ + goto term_array; + + } else { + if (!is_boxed(b) || *boxed_val_rel(b,b_base) != hdr) + goto not_equal; + + aa = hashmap_val_rel(a, a_base) + 1; + bb = hashmap_val_rel(b, b_base) + 1; + switch (hdr & _HEADER_MAP_SUBTAG_MASK) { + case HAMT_SUBTAG_HEAD_ARRAY: + aa++; bb++; + sz = 16; + break; + case HAMT_SUBTAG_HEAD_BITMAP: + aa++; bb++; + case HAMT_SUBTAG_NODE_BITMAP: + sz = hashmap_bitcount(MAP_HEADER_VAL(hdr)); + ASSERT(sz > 0 && sz < 17); + break; + default: + erl_exit(1, "Unknown hashmap subsubtag\n"); + } + goto term_array; + } + default: + ASSERT(!"Unknown boxed subtab in EQ"); } break; } @@ -2449,7 +2925,18 @@ Sint erts_cmp_rel_opt(Eterm a, Eterm* a_base, Eterm b, Eterm* b_base, int exact) Sint erts_cmp(Eterm a, Eterm b, int exact) #endif { - DECLARE_WSTACK(stack); +#define PSTACK_TYPE struct erts_cmp_hashmap_state + struct erts_cmp_hashmap_state { + Sint wstack_rollback; + int was_exact; + Eterm *ap; + Eterm *bp; + Eterm min_key; + Sint cmp_res; /* result so far -1,0,+1 */ + }; + PSTACK_DECLARE(hmap_stack, 1); + WSTACK_DECLARE(stack); + WSTACK_DECLARE(b_stack); /* only used by hashmaps */ Eterm* aa; Eterm* bb; int i; @@ -2465,6 +2952,26 @@ Sint erts_cmp(Eterm a, Eterm b, int exact) Uint32 *anum; Uint32 *bnum; +/* The WSTACK contains naked Eterms and Operations marked with header-tags */ +#define OP_BITS 4 +#define OP_MASK 0xF +#define TERM_ARRAY_OP 0 +#define SWITCH_EXACT_OFF_OP 1 +#define HASHMAP_PHASE1_ARE_KEYS_EQUAL 2 +#define HASHMAP_PHASE1_IS_MIN_KEY 3 +#define HASHMAP_PHASE1_CMP_VALUES 4 +#define HASHMAP_PHASE2_ARE_KEYS_EQUAL 5 +#define HASHMAP_PHASE2_IS_MIN_KEY_A 6 +#define HASHMAP_PHASE2_IS_MIN_KEY_B 7 + + +#define OP_WORD(OP) (((OP) << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER) +#define TERM_ARRAY_OP_WORD(SZ) OP_WORD(((SZ) << OP_BITS) | TERM_ARRAY_OP) + +#define GET_OP(WORD) (ASSERT(is_header(WORD)), ((WORD) >> _TAG_PRIMARY_SIZE) & OP_MASK) +#define GET_OP_ARG(WORD) (ASSERT(is_header(WORD)), ((WORD) >> (OP_BITS + _TAG_PRIMARY_SIZE))) + + #define RETURN_NEQ(cmp) { j=(cmp); ASSERT(j != 0); goto not_equal; } #define ON_CMP_GOTO(cmp) if ((j=(cmp)) == 0) goto pop_next; else goto not_equal @@ -2480,6 +2987,8 @@ Sint erts_cmp(Eterm a, Eterm b, int exact) } while (0) +bodyrecur: + j = 0; tailrecur: if (is_same(a,a_base,b,b_base)) { /* Equal values or pointers. */ goto pop_next; @@ -2606,25 +3115,96 @@ tailrecur_ne: ++aa; ++bb; goto term_array; - case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE) : - if (!is_map_rel(b,b_base)) { - a_tag = MAP_DEF; - goto mixed_types; - } - aa = (Eterm *)map_val_rel(a,a_base); - bb = (Eterm *)map_val_rel(b,b_base); - - i = map_get_size((map_t*)aa); - if (i != map_get_size((map_t*)bb)) { - RETURN_NEQ((int)(i - map_get_size((map_t*)bb))); - } - if (i == 0) { - goto pop_next; + case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE) : + { + struct erts_cmp_hashmap_state* sp; + if (is_flatmap_header(ahdr)) { + if (!is_flatmap_rel(b,b_base)) { + if (is_hashmap_rel(b,b_base)) { + aa = (Eterm *)flatmap_val_rel(a,a_base); + i = flatmap_get_size((flatmap_t*)aa) - hashmap_size_rel(b,b_base); + ASSERT(i != 0); + RETURN_NEQ(i); + } + a_tag = MAP_DEF; + goto mixed_types; + } + aa = (Eterm *)flatmap_val_rel(a,a_base); + bb = (Eterm *)flatmap_val_rel(b,b_base); + + i = flatmap_get_size((flatmap_t*)aa); + if (i != flatmap_get_size((flatmap_t*)bb)) { + RETURN_NEQ((int)(i - flatmap_get_size((flatmap_t*)bb))); + } + if (i == 0) { + goto pop_next; + } + aa += 2; + bb += 2; + if (exact) { + i += 1; /* increment for tuple-keys */ + goto term_array; + } + else { + /* Value array */ + WSTACK_PUSH3(stack,(UWord)(bb+1),(UWord)(aa+1),TERM_ARRAY_OP_WORD(i)); + /* Switch back from 'exact' key compare */ + WSTACK_PUSH(stack,OP_WORD(SWITCH_EXACT_OFF_OP)); + /* Now do 'exact' compare of key tuples */ + a = *aa; + b = *bb; + exact = 1; + goto bodyrecur; + } + } + if (!is_hashmap_rel(b,b_base)) { + if (is_flatmap_rel(b,b_base)) { + bb = (Eterm *)flatmap_val_rel(b,b_base); + i = hashmap_size_rel(a,a_base) - flatmap_get_size((flatmap_t*)bb); + ASSERT(i != 0); + RETURN_NEQ(i); + } + a_tag = MAP_DEF; + goto mixed_types; + } + i = hashmap_size_rel(a,a_base) - hashmap_size_rel(b,b_base); + if (i) { + RETURN_NEQ(i); + } + if (hashmap_size_rel(a,a_base) == 0) { + goto pop_next; + } + + /* Hashmap compare strategy: + Phase 1. While keys are identical + Do synchronous stepping through leafs of both trees in hash + order. Maintain value compare result of minimal key. + + Phase 2. If key diff was found in phase 1 + Ignore values from now on. + Continue iterate trees by always advancing the one + lagging behind hash-wise. Identical keys are skipped. + A minimal key can only be candidate as tie-breaker if we + have passed that hash value in the other tree (which means + the key did not exist in the other tree). + */ + + sp = PSTACK_PUSH(hmap_stack); + hashmap_iterator_init(&stack, a, 0); + hashmap_iterator_init(&b_stack, b, 0); + sp->ap = hashmap_iterator_next(&stack); + sp->bp = hashmap_iterator_next(&b_stack); + sp->cmp_res = 0; + ASSERT(sp->ap && sp->bp); + + a = CAR(sp->ap); + b = CAR(sp->bp); + sp->was_exact = exact; + exact = 1; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_ARE_KEYS_EQUAL)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; } - aa += 2; - bb += 2; - i += 1; /* increment for tuple-keys */ - goto term_array; case (_TAG_HEADER_FLOAT >> _TAG_PRIMARY_SIZE): if (!is_float_rel(b,b_base)) { a_tag = FLOAT_DEF; @@ -2984,8 +3564,7 @@ term_array: /* arrays in 'aa' and 'bb', length in 'i' */ goto not_equal; } } else { - /* (ab)Use TAG_PRIMARY_HEADER to recognize a term_array */ - WSTACK_PUSH3(stack, i, (UWord)bb, (UWord)aa | TAG_PRIMARY_HEADER); + WSTACK_PUSH3(stack, (UWord)bb, (UWord)aa, TERM_ARRAY_OP_WORD(i)); goto tailrecur_ne; } } @@ -2997,22 +3576,179 @@ term_array: /* arrays in 'aa' and 'bb', length in 'i' */ pop_next: if (!WSTACK_ISEMPTY(stack)) { UWord something = WSTACK_POP(stack); - if (primary_tag((Eterm) something) == TAG_PRIMARY_HEADER) { /* a term_array */ - aa = (Eterm*) something; - bb = (Eterm*) WSTACK_POP(stack); - i = WSTACK_POP(stack); - goto term_array; + struct erts_cmp_hashmap_state* sp; + if (primary_tag((Eterm) something) == TAG_PRIMARY_HEADER) { /* an operation */ + switch (GET_OP(something)) { + case TERM_ARRAY_OP: + i = GET_OP_ARG(something); + aa = (Eterm*)WSTACK_POP(stack); + bb = (Eterm*) WSTACK_POP(stack); + goto term_array; + + case SWITCH_EXACT_OFF_OP: + /* Done with exact compare of map keys, switch back */ + ASSERT(exact); + exact = 0; + goto pop_next; + + case HASHMAP_PHASE1_ARE_KEYS_EQUAL: { + sp = PSTACK_TOP(hmap_stack); + if (j) { + /* Key diff found, enter phase 2 */ + if (hashmap_key_hash_cmp(sp->ap, sp->bp) < 0) { + sp->min_key = CAR(sp->ap); + sp->cmp_res = -1; + sp->ap = hashmap_iterator_next(&stack); + } + else { + sp->min_key = CAR(sp->bp); + sp->cmp_res = 1; + sp->bp = hashmap_iterator_next(&b_stack); + } + exact = 1; /* only exact key compares in phase 2 */ + goto case_HASHMAP_PHASE2_LOOP; + } + + /* No key diff found so far, compare values if min key */ + + if (sp->cmp_res) { + a = CAR(sp->ap); + b = sp->min_key; + exact = 1; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_IS_MIN_KEY)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + /* no min key-value found yet */ + a = CDR(sp->ap); + b = CDR(sp->bp); + exact = sp->was_exact; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_CMP_VALUES)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + case HASHMAP_PHASE1_IS_MIN_KEY: + sp = PSTACK_TOP(hmap_stack); + if (j < 0) { + a = CDR(sp->ap); + b = CDR(sp->bp); + exact = sp->was_exact; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_CMP_VALUES)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + goto case_HASHMAP_PHASE1_LOOP; + + case HASHMAP_PHASE1_CMP_VALUES: + sp = PSTACK_TOP(hmap_stack); + if (j) { + sp->cmp_res = j; + sp->min_key = CAR(sp->ap); + } + case_HASHMAP_PHASE1_LOOP: + sp->ap = hashmap_iterator_next(&stack); + sp->bp = hashmap_iterator_next(&b_stack); + if (!sp->ap) { + /* end of maps with identical keys */ + ASSERT(!sp->bp); + j = sp->cmp_res; + exact = sp->was_exact; + (void) PSTACK_POP(hmap_stack); + ON_CMP_GOTO(j); + } + a = CAR(sp->ap); + b = CAR(sp->bp); + exact = 1; + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE1_ARE_KEYS_EQUAL)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + + case_HASHMAP_PHASE2_LOOP: + if (sp->ap && sp->bp) { + a = CAR(sp->ap); + b = CAR(sp->bp); + ASSERT(exact); + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE2_ARE_KEYS_EQUAL)); + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + goto case_HASHMAP_PHASE2_NEXT_STEP; + + case HASHMAP_PHASE2_ARE_KEYS_EQUAL: + sp = PSTACK_TOP(hmap_stack); + if (j == 0) { + /* keys are equal, skip them */ + sp->ap = hashmap_iterator_next(&stack); + sp->bp = hashmap_iterator_next(&b_stack); + goto case_HASHMAP_PHASE2_LOOP; + } + /* fall through */ + case_HASHMAP_PHASE2_NEXT_STEP: + if (sp->ap || sp->bp) { + if (hashmap_key_hash_cmp(sp->ap, sp->bp) < 0) { + ASSERT(sp->ap); + a = CAR(sp->ap); + b = sp->min_key; + ASSERT(exact); + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE2_IS_MIN_KEY_A)); + } + else { /* hash_cmp > 0 */ + ASSERT(sp->bp); + a = CAR(sp->bp); + b = sp->min_key; + ASSERT(exact); + WSTACK_PUSH(stack, OP_WORD(HASHMAP_PHASE2_IS_MIN_KEY_B)); + } + sp->wstack_rollback = WSTACK_COUNT(stack); + goto bodyrecur; + } + /* End of both maps */ + j = sp->cmp_res; + exact = sp->was_exact; + (void) PSTACK_POP(hmap_stack); + ON_CMP_GOTO(j); + + case HASHMAP_PHASE2_IS_MIN_KEY_A: + sp = PSTACK_TOP(hmap_stack); + if (j < 0) { + sp->min_key = CAR(sp->ap); + sp->cmp_res = -1; + } + sp->ap = hashmap_iterator_next(&stack); + goto case_HASHMAP_PHASE2_LOOP; + + case HASHMAP_PHASE2_IS_MIN_KEY_B: + sp = PSTACK_TOP(hmap_stack); + if (j < 0) { + sp->min_key = CAR(sp->bp); + sp->cmp_res = 1; + } + sp->bp = hashmap_iterator_next(&b_stack); + goto case_HASHMAP_PHASE2_LOOP; + + default: + ASSERT(!"Invalid cmp op"); + } /* switch */ } a = (Eterm) something; b = (Eterm) WSTACK_POP(stack); goto tailrecur; } - DESTROY_WSTACK(stack); + ASSERT(PSTACK_IS_EMPTY(hmap_stack)); + PSTACK_DESTROY(hmap_stack); + WSTACK_DESTROY(stack); + WSTACK_DESTROY(b_stack); return 0; not_equal: - DESTROY_WSTACK(stack); + if (!PSTACK_IS_EMPTY(hmap_stack)) { + WSTACK_ROLLBACK(stack, PSTACK_TOP(hmap_stack)->wstack_rollback); + goto pop_next; + } + PSTACK_DESTROY(hmap_stack); + WSTACK_DESTROY(stack); + WSTACK_DESTROY(b_stack); return j; #undef CMP_NODES @@ -3197,106 +3933,303 @@ buf_to_intlist(Eterm** hpp, const char *buf, size_t len, Eterm tail) ** */ -ErlDrvSizeT erts_iolist_to_buf(Eterm obj, char* buf, ErlDrvSizeT alloced_len) +typedef enum { + ERTS_IL2B_BCOPY_OK, + ERTS_IL2B_BCOPY_YIELD, + ERTS_IL2B_BCOPY_OVERFLOW, + ERTS_IL2B_BCOPY_TYPE_ERROR +} ErtsIL2BBCopyRes; + +static ErtsIL2BBCopyRes +iolist_to_buf_bcopy(ErtsIOList2BufState *state, Eterm obj, int *yield_countp); + +static ERTS_INLINE ErlDrvSizeT +iolist_to_buf(const int yield_support, + ErtsIOList2BufState *state, + Eterm obj, + char* buf, + ErlDrvSizeT alloced_len) { - ErlDrvSizeT len = (ErlDrvSizeT) alloced_len; - Eterm* objp; +#undef IOLIST_TO_BUF_BCOPY +#define IOLIST_TO_BUF_BCOPY(CONSP) \ +do { \ + size_t size = binary_size(obj); \ + if (size > 0) { \ + Uint bitsize; \ + byte* bptr; \ + Uint bitoffs; \ + Uint num_bits; \ + if (yield_support) { \ + size_t max_size = ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; \ + if (yield_count > 0) \ + max_size *= yield_count+1; \ + if (size > max_size) { \ + state->objp = CONSP; \ + goto L_bcopy_yield; \ + } \ + if (size >= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT) { \ + int cost = (int) size; \ + cost /= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; \ + yield_count -= cost; \ + } \ + } \ + if (len < size) \ + goto L_overflow; \ + ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); \ + if (bitsize != 0) \ + goto L_type_error; \ + num_bits = 8*size; \ + copy_binary_to_buffer(buf, 0, bptr, bitoffs, num_bits); \ + buf += size; \ + len -= size; \ + } \ +} while (0) + + ErlDrvSizeT res, len; + Eterm* objp = NULL; + int init_yield_count; + int yield_count; DECLARE_ESTACK(s); - goto L_again; - - while (!ESTACK_ISEMPTY(s)) { - obj = ESTACK_POP(s); - L_again: - if (is_list(obj)) { - L_iter_list: - objp = list_val(obj); - obj = CAR(objp); - if (is_byte(obj)) { - if (len == 0) { - goto L_overflow; - } - *buf++ = unsigned_val(obj); - len--; - } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - - if (len < size) { + + len = (ErlDrvSizeT) alloced_len; + + if (!yield_support) { + yield_count = init_yield_count = 0; /* Shut up faulty warning... >:-( */ + goto L_again; + } + else { + + if (state->iolist.reds_left <= 0) + return ERTS_IOLIST_TO_BUF_YIELD; + + ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK); + init_yield_count = (ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED + * state->iolist.reds_left); + yield_count = init_yield_count; + + if (!state->iolist.estack.start) + goto L_again; + else { + int chk_stack; + /* Restart; restore state... */ + ESTACK_RESTORE(s, &state->iolist.estack); + + if (!state->bcopy.bptr) + chk_stack = 0; + else { + chk_stack = 1; + switch (iolist_to_buf_bcopy(state, THE_NON_VALUE, &yield_count)) { + case ERTS_IL2B_BCOPY_OK: + break; + case ERTS_IL2B_BCOPY_YIELD: + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + ESTACK_SAVE(s, &state->iolist.estack); + return ERTS_IOLIST_TO_BUF_YIELD; + case ERTS_IL2B_BCOPY_OVERFLOW: goto L_overflow; - } - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - if (bitsize != 0) { + case ERTS_IL2B_BCOPY_TYPE_ERROR: goto L_type_error; } - num_bits = 8*size; - copy_binary_to_buffer(buf, 0, bptr, bitoffs, num_bits); - buf += size; - len -= size; - } else if (is_list(obj)) { - ESTACK_PUSH(s, CDR(objp)); - goto L_iter_list; /* on head */ - } else if (is_not_nil(obj)) { - goto L_type_error; } - obj = CDR(objp); - if (is_list(obj)) { - goto L_iter_list; /* on tail */ - } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - if (len < size) { - goto L_overflow; + obj = state->iolist.obj; + buf = state->buf; + len = state->len; + objp = state->objp; + state->objp = NULL; + if (objp) + goto L_tail; + if (!chk_stack) + goto L_again; + /* check stack */ + } + } + + while (!ESTACK_ISEMPTY(s)) { + obj = ESTACK_POP(s); + L_again: + if (is_list(obj)) { + while (1) { /* Tail loop */ + while (1) { /* Head loop */ + if (yield_support && --yield_count <= 0) + goto L_yield; + objp = list_val(obj); + obj = CAR(objp); + if (is_byte(obj)) { + if (len == 0) { + goto L_overflow; + } + *buf++ = unsigned_val(obj); + len--; + } else if (is_binary(obj)) { + IOLIST_TO_BUF_BCOPY(objp); + } else if (is_list(obj)) { + ESTACK_PUSH(s, CDR(objp)); + continue; /* Head loop */ + } else if (is_not_nil(obj)) { + goto L_type_error; + } + break; } - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - if (bitsize != 0) { + + L_tail: + + obj = CDR(objp); + + if (is_list(obj)) { + continue; /* Tail loop */ + } else if (is_binary(obj)) { + IOLIST_TO_BUF_BCOPY(NULL); + } else if (is_not_nil(obj)) { goto L_type_error; } - num_bits = 8*size; - copy_binary_to_buffer(buf, 0, bptr, bitoffs, num_bits); - buf += size; - len -= size; - } else if (is_not_nil(obj)) { - goto L_type_error; + break; } } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - if (len < size) { - goto L_overflow; - } - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - if (bitsize != 0) { - goto L_type_error; - } - num_bits = 8*size; - copy_binary_to_buffer(buf, 0, bptr, bitoffs, num_bits); - buf += size; - len -= size; + IOLIST_TO_BUF_BCOPY(NULL); } else if (is_not_nil(obj)) { goto L_type_error; - } + } else if (yield_support && --yield_count <= 0) + goto L_yield; } + res = len; + + L_return: + DESTROY_ESTACK(s); - return len; + + if (yield_support) { + int reds; + CLEAR_SAVED_ESTACK(&state->iolist.estack); + reds = ((init_yield_count - yield_count - 1) + / ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED) + 1; + BUMP_REDS(state->iolist.c_p, reds); + state->iolist.reds_left -= reds; + if (state->iolist.reds_left < 0) + state->iolist.reds_left = 0; + } + + + return res; L_type_error: - DESTROY_ESTACK(s); - return ERTS_IOLIST_TO_BUF_TYPE_ERROR; + res = ERTS_IOLIST_TO_BUF_TYPE_ERROR; + goto L_return; L_overflow: - DESTROY_ESTACK(s); - return ERTS_IOLIST_TO_BUF_OVERFLOW; + res = ERTS_IOLIST_TO_BUF_OVERFLOW; + goto L_return; + + L_bcopy_yield: + + state->buf = buf; + state->len = len; + + switch (iolist_to_buf_bcopy(state, obj, &yield_count)) { + case ERTS_IL2B_BCOPY_OK: + ERTS_INTERNAL_ERROR("Missing yield"); + case ERTS_IL2B_BCOPY_YIELD: + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + ESTACK_SAVE(s, &state->iolist.estack); + return ERTS_IOLIST_TO_BUF_YIELD; + case ERTS_IL2B_BCOPY_OVERFLOW: + goto L_overflow; + case ERTS_IL2B_BCOPY_TYPE_ERROR: + goto L_type_error; + } + + L_yield: + + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + state->iolist.obj = obj; + state->buf = buf; + state->len = len; + ESTACK_SAVE(s, &state->iolist.estack); + return ERTS_IOLIST_TO_BUF_YIELD; + +#undef IOLIST_TO_BUF_BCOPY +} + +static ErtsIL2BBCopyRes +iolist_to_buf_bcopy(ErtsIOList2BufState *state, Eterm obj, int *yield_countp) +{ + ErtsIL2BBCopyRes res; + char *buf = state->buf; + ErlDrvSizeT len = state->len; + byte* bptr; + size_t size; + size_t max_size; + Uint bitoffs; + Uint num_bits; + int yield_count = *yield_countp; + + if (state->bcopy.bptr) { + bptr = state->bcopy.bptr; + size = state->bcopy.size; + bitoffs = state->bcopy.bitoffs; + state->bcopy.bptr = NULL; + } + else { + Uint bitsize; + + ASSERT(is_binary(obj)); + + size = binary_size(obj); + if (size <= 0) + return ERTS_IL2B_BCOPY_OK; + + if (len < size) + return ERTS_IL2B_BCOPY_OVERFLOW; + + ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); + if (bitsize != 0) + return ERTS_IL2B_BCOPY_TYPE_ERROR; + } + + ASSERT(size > 0); + max_size = (size_t) ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; + if (yield_count > 0) + max_size *= (size_t) (yield_count+1); + + if (size <= max_size) { + if (size >= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT) { + int cost = (int) size; + cost /= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; + yield_count -= cost; + } + res = ERTS_IL2B_BCOPY_OK; + } + else { + ASSERT(0 < max_size && max_size < size); + yield_count = 0; + state->bcopy.bptr = bptr + max_size; + state->bcopy.bitoffs = bitoffs; + state->bcopy.size = size - max_size; + size = max_size; + res = ERTS_IL2B_BCOPY_YIELD; + } + + num_bits = 8*size; + copy_binary_to_buffer(buf, 0, bptr, bitoffs, num_bits); + state->buf += size; + state->len -= size; + *yield_countp = yield_count; + + return res; +} + +ErlDrvSizeT erts_iolist_to_buf_yielding(ErtsIOList2BufState *state) +{ + return iolist_to_buf(1, state, state->iolist.obj, state->buf, state->len); +} + +ErlDrvSizeT erts_iolist_to_buf(Eterm obj, char* buf, ErlDrvSizeT alloced_len) +{ + return iolist_to_buf(0, NULL, obj, buf, alloced_len); } /* @@ -3307,11 +4240,32 @@ ErlDrvSizeT erts_iolist_to_buf(Eterm obj, char* buf, ErlDrvSizeT alloced_len) * Any input term error detected in erts_iolist_to_buf should also * be detected in this function! */ -int erts_iolist_size(Eterm obj, ErlDrvSizeT* sizep) + +static ERTS_INLINE int +iolist_size(const int yield_support, ErtsIOListState *state, Eterm obj, ErlDrvSizeT* sizep) { + int res, init_yield_count, yield_count; Eterm* objp; - Uint size = 0; /* Intentionally Uint due to halfword heap */ + Uint size = (Uint) *sizep; /* Intentionally Uint due to halfword heap */ DECLARE_ESTACK(s); + + if (!yield_support) + yield_count = init_yield_count = 0; /* Shut up faulty warning... >:-( */ + else { + if (state->reds_left <= 0) + return ERTS_IOLIST_YIELD; + ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK); + init_yield_count = ERTS_IOLIST_SIZE_YIELDS_COUNT_PER_RED; + init_yield_count *= state->reds_left; + yield_count = init_yield_count; + if (state->estack.start) { + /* Restart; restore state... */ + ESTACK_RESTORE(s, &state->estack); + size = (Uint) state->size; + obj = state->obj; + } + } + goto L_again; #define SAFE_ADD(Var, Val) \ @@ -3327,51 +4281,101 @@ int erts_iolist_size(Eterm obj, ErlDrvSizeT* sizep) obj = ESTACK_POP(s); L_again: if (is_list(obj)) { - L_iter_list: - objp = list_val(obj); - /* Head */ - obj = CAR(objp); - if (is_byte(obj)) { - size++; - if (size == 0) { - goto L_overflow_error; + while (1) { /* Tail loop */ + while (1) { /* Head loop */ + if (yield_support && --yield_count <= 0) + goto L_yield; + objp = list_val(obj); + /* Head */ + obj = CAR(objp); + if (is_byte(obj)) { + size++; + if (size == 0) { + goto L_overflow_error; + } + } else if (is_binary(obj) && binary_bitsize(obj) == 0) { + SAFE_ADD(size, binary_size(obj)); + } else if (is_list(obj)) { + ESTACK_PUSH(s, CDR(objp)); + continue; /* Head loop */ + } else if (is_not_nil(obj)) { + goto L_type_error; + } + break; } - } else if (is_binary(obj) && binary_bitsize(obj) == 0) { - SAFE_ADD(size, binary_size(obj)); - } else if (is_list(obj)) { - ESTACK_PUSH(s, CDR(objp)); - goto L_iter_list; /* on head */ - } else if (is_not_nil(obj)) { - goto L_type_error; + /* Tail */ + obj = CDR(objp); + if (is_list(obj)) + continue; /* Tail loop */ + else if (is_binary(obj) && binary_bitsize(obj) == 0) { + SAFE_ADD(size, binary_size(obj)); + } else if (is_not_nil(obj)) { + goto L_type_error; + } + break; } - /* Tail */ - obj = CDR(objp); - if (is_list(obj)) - goto L_iter_list; /* on tail */ - else if (is_binary(obj) && binary_bitsize(obj) == 0) { + } else { + if (yield_support && --yield_count <= 0) + goto L_yield; + if (is_binary(obj) && binary_bitsize(obj) == 0) { /* Tail was binary */ SAFE_ADD(size, binary_size(obj)); } else if (is_not_nil(obj)) { goto L_type_error; } - } else if (is_binary(obj) && binary_bitsize(obj) == 0) { /* Tail was binary */ - SAFE_ADD(size, binary_size(obj)); - } else if (is_not_nil(obj)) { - goto L_type_error; } } #undef SAFE_ADD - DESTROY_ESTACK(s); *sizep = (ErlDrvSizeT) size; - return ERTS_IOLIST_OK; - L_overflow_error: + res = ERTS_IOLIST_OK; + + L_return: + DESTROY_ESTACK(s); - return ERTS_IOLIST_OVERFLOW; + + if (yield_support) { + int yc, reds; + CLEAR_SAVED_ESTACK(&state->estack); + yc = init_yield_count - yield_count; + reds = ((yc - 1) / ERTS_IOLIST_SIZE_YIELDS_COUNT_PER_RED) + 1; + BUMP_REDS(state->c_p, reds); + state->reds_left -= reds; + state->size = (ErlDrvSizeT) size; + state->have_size = 1; + } + + return res; + + L_overflow_error: + res = ERTS_IOLIST_OVERFLOW; + size = 0; + goto L_return; L_type_error: - DESTROY_ESTACK(s); - return ERTS_IOLIST_TYPE; + res = ERTS_IOLIST_TYPE; + size = 0; + goto L_return; + + L_yield: + BUMP_ALL_REDS(state->c_p); + state->reds_left = 0; + state->size = size; + state->obj = obj; + ESTACK_SAVE(s, &state->estack); + return ERTS_IOLIST_YIELD; +} + +int erts_iolist_size_yielding(ErtsIOListState *state) +{ + ErlDrvSizeT size = state->size; + return iolist_size(1, state, state->obj, &size); +} + +int erts_iolist_size(Eterm obj, ErlDrvSizeT* sizep) +{ + *sizep = 0; + return iolist_size(0, NULL, obj, sizep); } /* return 0 if item is not a non-empty flat list of bytes */ @@ -3507,7 +4511,7 @@ erts_create_smp_ptimer(ErtsSmpPTimer **timer_ref, res->timer.timeout_func = timeout_func; res->timer.timer_ref = timer_ref; res->timer.id = id; - res->timer.tm.active = 0; /* MUST be initalized */ + erts_init_timer(&res->timer.tm); ASSERT(!*timer_ref); @@ -3680,6 +4684,9 @@ erts_save_emu_args(int argc, char **argv) size += sz+1; } ptr = (char *) malloc(size); + if (!ptr) { + ERTS_INTERNAL_ERROR("malloc failed to allocate memory!"); + } #ifdef DEBUG end_ptr = ptr + size; #endif @@ -3944,19 +4951,7 @@ void erts_silence_warn_unused_result(long unused) void erts_interval_init(erts_interval_t *icp) { -#ifdef ARCH_64 - erts_atomic_init_nob(&icp->counter.atomic, 0); -#else - erts_dw_aint_t dw; -#ifdef ETHR_SU_DW_NAINT_T__ - dw.dw_sint = 0; -#else - dw.sint[ERTS_DW_AINT_HIGH_WORD] = 0; - dw.sint[ERTS_DW_AINT_LOW_WORD] = 0; -#endif - erts_dw_atomic_init_nob(&icp->counter.atomic, &dw); - -#endif + erts_atomic64_init_nob(&icp->counter.atomic, 0); #ifdef DEBUG icp->smp_api = 0; #endif @@ -3978,55 +4973,13 @@ erts_smp_interval_init(erts_interval_t *icp) static ERTS_INLINE Uint64 step_interval_nob(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_inc_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_nob(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - } -#endif + return (Uint64) erts_atomic64_inc_read_nob(&icp->counter.atomic); } static ERTS_INLINE Uint64 step_interval_relb(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_inc_read_relb(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_relb(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - } -#endif + return (Uint64) erts_atomic64_inc_read_relb(&icp->counter.atomic); } @@ -4034,38 +4987,10 @@ static ERTS_INLINE Uint64 ensure_later_interval_nob(erts_interval_t *icp, Uint64 ic) { Uint64 curr_ic; -#ifdef ARCH_64 - curr_ic = (Uint64) erts_atomic_read_nob(&icp->counter.atomic); - if (curr_ic > ic) - return curr_ic; - return (Uint64) erts_atomic_inc_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - curr_ic = erts_interval_dw_aint_to_val__(&exp); + curr_ic = (Uint64) erts_atomic64_read_nob(&icp->counter.atomic); if (curr_ic > ic) return curr_ic; - - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_nob(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - } -#endif + return (Uint64) erts_atomic64_inc_read_nob(&icp->counter.atomic); } @@ -4073,38 +4998,10 @@ static ERTS_INLINE Uint64 ensure_later_interval_acqb(erts_interval_t *icp, Uint64 ic) { Uint64 curr_ic; -#ifdef ARCH_64 - curr_ic = (Uint64) erts_atomic_read_acqb(&icp->counter.atomic); - if (curr_ic > ic) - return curr_ic; - return (Uint64) erts_atomic_inc_read_acqb(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_acqb(&icp->counter.atomic, &exp); - curr_ic = erts_interval_dw_aint_to_val__(&exp); + curr_ic = (Uint64) erts_atomic64_read_acqb(&icp->counter.atomic); if (curr_ic > ic) return curr_ic; - - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_acqb(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - } -#endif + return (Uint64) erts_atomic64_inc_read_acqb(&icp->counter.atomic); } Uint64 @@ -4191,8 +5088,8 @@ erts_smp_ensure_later_interval_acqb(erts_interval_t *icp, Uint64 ic) */ Uint64 erts_timestamp_millis(void) { -#ifdef HAVE_GETHRTIME - return (Uint64) (sys_gethrtime() / 1000000); +#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT + return ERTS_MONOTONIC_TO_MSEC(erts_os_monotonic_time()); #else Uint64 res; SysTimeval tv; |