diff options
Diffstat (limited to 'erts/emulator')
132 files changed, 5862 insertions, 2350 deletions
diff --git a/erts/emulator/Makefile.in b/erts/emulator/Makefile.in index 7145824f91..53fc7bd713 100644 --- a/erts/emulator/Makefile.in +++ b/erts/emulator/Makefile.in @@ -112,18 +112,24 @@ NO_INLINE_FUNCTIONS=true else ifeq ($(TYPE),lcnt) -PURIFY = +PURIFY = TYPEMARKER = .lcnt TYPE_FLAGS = @CFLAGS@ -DERTS_ENABLE_LOCK_COUNT else ifeq ($(TYPE),frmptr) -PURIFY = +PURIFY = OMIT_OMIT_FP=yes TYPEMARKER = .frmptr TYPE_FLAGS = @CFLAGS@ -DERTS_FRMPTR else +ifeq ($(TYPE),icount) +PURIFY = +TYPEMARKER = .icount +TYPE_FLAGS = @CFLAGS@ -DERTS_OPCODE_COUNTER_SUPPORT +else + # If type isn't one of the above, it *is* opt type... override TYPE=opt PURIFY = @@ -138,6 +144,7 @@ endif endif endif endif +endif comma:=, space:= diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names index 5d06a32941..c097866c7e 100644 --- a/erts/emulator/beam/atom.names +++ b/erts/emulator/beam/atom.names @@ -198,6 +198,7 @@ atom dotall atom driver atom driver_options atom dsend +atom dsend_continue_trap atom dunlink atom duplicate_bag atom dupnames @@ -279,7 +280,6 @@ 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 @@ -335,6 +335,7 @@ atom max atom maximum atom max_tables max_processes atom mbuf_size +atom md5 atom memory atom memory_internal atom memory_types diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index 1026e5f649..e9f5fd798b 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 @@ -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 */ @@ -3503,6 +3500,7 @@ 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 +3521,6 @@ 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); @@ -3788,8 +3779,6 @@ 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 +3878,6 @@ 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 +4980,14 @@ 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 +5021,10 @@ 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 +5141,16 @@ 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) { #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); diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c index e96177cfd9..07654f6d5c 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); @@ -4038,7 +4044,7 @@ 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 +4253,20 @@ 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); /* @@ -5103,10 +5118,11 @@ 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 +5132,8 @@ 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 +5322,7 @@ 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 +5361,7 @@ 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 +5384,33 @@ 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 +5586,7 @@ 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 +5982,7 @@ 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/bif.c b/erts/emulator/beam/bif.c index fcbeb6cf5c..49996e7f0b 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" @@ -46,7 +48,7 @@ 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; static Export *await_sched_wall_time_mod_trap; static erts_smp_atomic32_t sched_wall_time; @@ -1777,6 +1779,8 @@ 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 +1799,36 @@ 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 +1836,8 @@ 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 +1858,8 @@ 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 +1878,7 @@ 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 " @@ -1879,9 +1889,10 @@ 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); @@ -1912,6 +1923,7 @@ 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 " @@ -1922,6 +1934,7 @@ 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; @@ -1936,7 +1949,7 @@ 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)) { @@ -1945,12 +1958,12 @@ 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; @@ -2030,9 +2043,14 @@ 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 ??? */ @@ -2063,9 +2081,11 @@ 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; @@ -2073,34 +2093,44 @@ 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) { @@ -2108,68 +2138,127 @@ 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) { @@ -2177,35 +2266,46 @@ 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; } /**********************************************************************/ @@ -2768,6 +2868,7 @@ 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; @@ -2821,8 +2922,8 @@ 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)) { @@ -2846,7 +2947,8 @@ 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; @@ -2887,9 +2989,6 @@ 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)); @@ -2899,8 +2998,12 @@ 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); } @@ -3989,16 +4092,19 @@ 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); } /**********************************************************************/ @@ -4809,6 +4915,10 @@ void erts_init_bif(void) #endif , &bif_return_trap); + erts_init_trap_export(&dsend_continue_trap_export, + am_erts_internal, am_dsend_continue_trap, 1, + dsend_continue_trap_1); + flush_monitor_message_trap = erts_export_put(am_erlang, am_flush_monitor_message, 2); diff --git a/erts/emulator/beam/bif.h b/erts/emulator/beam/bif.h index 72c55ccb55..7b69b39511 100644 --- a/erts/emulator/beam/bif.h +++ b/erts/emulator/beam/bif.h @@ -465,6 +465,8 @@ 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) @@ -509,6 +511,7 @@ BIF_RETTYPE hipe_wrapper_ ## BIF_NAME ## _ ## ARITY (Process* c_p, \ #endif +#endif /* ERL_WANT_HIPE_BIF_WRAPPER__ */ #include "erl_bif_table.h" diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab index 011e49f1fe..55ac778475 100644 --- a/erts/emulator/beam/bif.tab +++ b/erts/emulator/beam/bif.tab @@ -578,7 +578,7 @@ bif io:printable_range/0 bif os:unsetenv/1 # -# New in R17A +# New in 17.0 # bif re:inspect/2 @@ -601,6 +601,17 @@ bif maps:values/1 bif erts_internal:cmp_term/2 # +# New in 17.1 +# + +bif erlang:fun_info_mfa/1 + +# New in 18.0 +# + +bif erlang:get_keys/0 + +# # Obsolete # diff --git a/erts/emulator/beam/big.c b/erts/emulator/beam/big.c index 41a041eba6..de7d370938 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 @@ -1564,7 +1571,7 @@ 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); @@ -2667,9 +2674,6 @@ 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 +2683,12 @@ 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..da31876d75 100644 --- a/erts/emulator/beam/big.h +++ b/erts/emulator/beam/big.h @@ -101,7 +101,7 @@ 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))) diff --git a/erts/emulator/beam/binary.c b/erts/emulator/beam/binary.c index f50d484576..cc0b3b9b6c 100644 --- a/erts/emulator/beam/binary.c +++ b/erts/emulator/beam/binary.c @@ -26,7 +26,9 @@ #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 "erl_binary.h" #include "erl_bits.h" @@ -83,8 +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); @@ -122,8 +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); @@ -177,7 +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; diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c index 7d4f52ee23..5aee85174f 100644 --- a/erts/emulator/beam/break.c +++ b/erts/emulator/beam/break.c @@ -256,6 +256,7 @@ 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; @@ -753,6 +754,8 @@ 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) @@ -803,7 +806,7 @@ 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..0010f6a440 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" 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..90cd227fae 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: diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types index d23d26a0fc..61def65235 100644 --- a/erts/emulator/beam/erl_alloc.types +++ b/erts/emulator/beam/erl_alloc.types @@ -267,7 +267,6 @@ 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 @@ -357,12 +356,14 @@ 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 +else # "fullword" @@ -375,12 +376,14 @@ 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 +endif @@ -395,6 +398,7 @@ 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 diff --git a/erts/emulator/beam/erl_alloc_util.c b/erts/emulator/beam/erl_alloc_util.c index 45f0cc4312..e3172dc4fb 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) \ @@ -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_ALC_CPOOL_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; 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_bif_binary.c b/erts/emulator/beam/erl_bif_binary.c index 7e0e825a0d..bd0d7c71cc 100644 --- a/erts/emulator/beam/erl_bif_binary.c +++ b/erts/emulator/beam/erl_bif_binary.c @@ -32,7 +32,9 @@ #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 "erl_binary.h" #include "erl_bits.h" @@ -1324,9 +1326,9 @@ 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 +1557,9 @@ 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 +1646,9 @@ 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 +2215,9 @@ 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; @@ -2424,8 +2426,6 @@ 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_info.c b/erts/emulator/beam/erl_bif_info.c index 4d5e55aaf5..d2ee5e4224 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" @@ -90,7 +91,7 @@ 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 +116,9 @@ 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 @@ -2300,7 +2304,7 @@ 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 +2463,13 @@ 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); @@ -2696,6 +2707,9 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) ? am_disabled : am_enabled); } + 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); } @@ -3055,6 +3069,25 @@ 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)) { @@ -3285,17 +3318,38 @@ 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) */ @@ -3856,16 +3910,19 @@ 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); @@ -3874,23 +3931,27 @@ 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; } @@ -3911,13 +3972,13 @@ 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); @@ -3928,16 +3989,15 @@ 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; } @@ -3957,12 +4017,12 @@ 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_binary.h b/erts/emulator/beam/erl_binary.h index 6c9f53ce87..8d264d166e 100644 --- a/erts/emulator/beam/erl_binary.h +++ b/erts/emulator/beam/erl_binary.h @@ -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..71d31c01aa 100644 --- a/erts/emulator/beam/erl_bits.c +++ b/erts/emulator/beam/erl_bits.c @@ -1299,7 +1299,6 @@ 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 +1370,6 @@ 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 +1472,6 @@ 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 +1484,6 @@ 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 +1531,6 @@ 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 +1577,7 @@ 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_hash.h b/erts/emulator/beam/erl_db_hash.h index 908cec11d4..e68081a5b1 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; diff --git a/erts/emulator/beam/erl_db_util.c b/erts/emulator/beam/erl_db_util.c index 3927615e04..7eb80e3bb1 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) @@ -218,6 +213,8 @@ typedef enum { matchTuple, matchPushT, matchPushL, + matchPushM, + matchPushK, matchPop, matchBind, matchCmp, @@ -227,11 +224,13 @@ typedef enum { matchEqRef, matchEq, matchList, + matchMap, matchSkip, matchPushC, matchConsA, /* Car is below Cdr */ matchConsB, /* Cdr is below Car (unusual) */ matchMkTuple, + matchMkMap, matchCall0, matchCall1, matchCall2, @@ -856,6 +855,13 @@ 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 +875,9 @@ 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 +897,14 @@ 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 +1375,51 @@ restart: for (;;) { switch (t & _TAG_PRIMARY_MASK) { case TAG_PRIMARY_BOXED: - if (!BOXED_IS_TUPLE(t)) { + if (is_map(t)) { + num_iters = map_get_size(map_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 = map_get_keys(map_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, matchPushK); + ++(context.stack_used); + DMC_PUSH(text, dmc_private_copy(&context, key)); + } + if (context.stack_used > context.stack_need) { + context.stack_need = context.stack_used; + } + for (i = num_iters; i--; ) { + Eterm value = map_get_values(map_val(t))[i]; + DMC_PUSH(text, matchPop); + --(context.stack_used); + res = dmc_one_term(&context, &heap, &stack, &text, + value); + ASSERT(res != retFail); + if (res == retRestart) { + goto restart; + } + } + break; + } + if (!is_tuple(t)) { goto simple_term; } num_iters = arityval(*tuple_val(t)); @@ -1715,10 +1770,8 @@ 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 +1861,7 @@ 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; @@ -1887,6 +1940,34 @@ restart: *sp++ = list_val_rel(*ep,base); ++ep; break; + case matchMap: + if (!is_map_rel(*ep, base)) { + FAIL(); + } + n = *pc++; + if (map_get_size(map_val_rel(*ep, base)) < n) { + FAIL(); + } + ep = map_val_rel(*ep, base); + break; + case matchPushM: + if (!is_map_rel(*ep, base)) { + FAIL(); + } + n = *pc++; + if (map_get_size(map_val_rel(*ep, base)) < n) { + FAIL(); + } + *sp++ = map_val_rel(*ep++, base); + break; + case matchPushK: + t = (Eterm) *pc++; + tp = erts_maps_get_rel(t, make_map_rel(ep, base), base); + if (!tp) { + FAIL(); + } + *sp++ = tp; + break; case matchPop: ep = *(--sp); break; @@ -1987,6 +2068,23 @@ restart: } *esp++ = t; break; + case matchMkMap: + n = *pc++; + ehp = HAllocX(build_proc, 1 + MAP_HEADER_SIZE + n, HEAP_XTRA); + t = *ehp++ = *--esp; + { + map_t *m = (map_t *)ehp; + m->thing_word = MAP_HEADER; + m->size = n; + m->keys = t; + } + t = make_map(ehp); + ehp += MAP_HEADER_SIZE; + while (n--) { + *ehp++ = *--esp; + } + *esp++ = t; + break; case matchCall0: bif = (Eterm (*)(Process*, ...)) *pc++; t = (*bif)(build_proc, bif_args); @@ -3156,34 +3254,38 @@ 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)); + } + } + 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; } @@ -3243,7 +3345,6 @@ static DMCRet dmc_one_term(DMCContext *context, { Sint n; Eterm *hp; - ErlHeapFragment *tmp_mb; Uint sz, sz2, sz3; Uint i, j; @@ -3334,6 +3435,13 @@ static DMCRet dmc_one_term(DMCContext *context, DMC_PUSH(*text, n); DMC_PUSH(*stack, c); break; + case (_TAG_HEADER_MAP >> _TAG_PRIMARY_SIZE): + n = map_get_size(map_val(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 +3523,8 @@ 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 +3537,22 @@ 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 +3643,78 @@ 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 +3723,36 @@ 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) +{ + map_t *m = (map_t *)map_val(t); + Eterm *values = map_get_values(m); + int nelems = map_get_size(m); + int constant_values; + DMCRet ret; + + 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, matchMkMap); + DMC_PUSH(*text, nelems); + context->stack_used -= nelems; + *constant = 0; + return retOk; +} + static DMCRet dmc_whole_expression(DMCContext *context, DMCHeap *heap, DMC_STACK_TYPE(UWord) *text, @@ -4580,7 +4747,10 @@ 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 +5025,7 @@ 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 +5296,12 @@ 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 matchPushT: ++t; n = *t; @@ -5136,6 +5312,18 @@ 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 matchPushK: + ++t; + p = (Eterm) *t; + ++t; + erts_printf("PushK\t%p (%T)\n", t, p); + break; case matchPop: ++t; erts_printf("Pop\n"); @@ -5252,6 +5440,12 @@ void db_match_dis(Binary *bp) ++t; erts_printf("MkTuple\t%beu\n", n); break; + case matchMkMap: + ++t; + n = *t; + ++t; + erts_printf("MkMapA\t%beu\n", n); + break; case matchOr: ++t; n = *t; diff --git a/erts/emulator/beam/erl_driver.h b/erts/emulator/beam/erl_driver.h index 5ced8c5ca0..f9938fc66c 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 1 /* * The emulator will refuse to load a driver with a major version 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_gc.c b/erts/emulator/beam/erl_gc.c index aa15d2cc57..5f78a7b532 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" @@ -2018,6 +2020,20 @@ 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 +2402,6 @@ 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..bf0496c112 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 @@ -67,8 +69,6 @@ do { \ #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 +97,33 @@ 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 5e6d812242..77445ef1ff 100644 --- a/erts/emulator/beam/erl_init.c +++ b/erts/emulator/beam/erl_init.c @@ -161,9 +161,6 @@ 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. */ @@ -548,6 +545,8 @@ 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 @@ -1674,6 +1673,22 @@ 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) @@ -1882,11 +1897,6 @@ erl_start(int argc, char **argv) if (argv[i][2] == 0) { /* -c: documented option */ erts_disable_tolerant_timeofday = 1; } -#ifdef ERTS_OPCODE_COUNTER_SUPPORT - else if (argv[i][2] == 'i') { /* -ci: undcoumented option*/ - count_instructions = 1; - } -#endif break; case 'W': arg = get_arg(argv[i]+2, argv[i+1], &i); @@ -2066,8 +2076,10 @@ 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_lock_check.c b/erts/emulator/beam/erl_lock_check.c index c13eb87012..b105ece6f1 100644 --- a/erts/emulator/beam/erl_lock_check.c +++ b/erts/emulator/beam/erl_lock_check.c @@ -139,7 +139,6 @@ 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" }, @@ -148,6 +147,7 @@ static erts_lc_lock_order_t erts_lock_order[] = { { "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 }, @@ -227,8 +227,7 @@ 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; } @@ -311,8 +310,7 @@ 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 +323,7 @@ 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 +363,11 @@ 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 +688,7 @@ 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 +1327,7 @@ 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..cf6996ea06 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) { @@ -81,19 +100,20 @@ 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 +#if 0 || defined(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; +#else + SysTimeval tv; + sys_gettimeofday(&tv); + time->s = tv.tv_sec; + time->ns = tv.tv_usec*1000LL; #endif } @@ -111,28 +131,29 @@ 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 +179,64 @@ 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); + 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); } } - -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, ""); - } -} - #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 +246,7 @@ 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 +275,9 @@ 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 +299,9 @@ 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 +363,9 @@ 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 +394,6 @@ 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 +408,9 @@ 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 +453,9 @@ 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 +470,7 @@ 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 +483,10 @@ 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 +496,10 @@ 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 +527,7 @@ 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 +536,12 @@ 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 +570,11 @@ 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 +611,7 @@ 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..ffbb93da1b 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(HAVE_GETHRTIME) +#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 5e740aacdd..b2a16eb5ed 100644 --- a/erts/emulator/beam/erl_map.c +++ b/erts/emulator/beam/erl_map.c @@ -113,36 +113,55 @@ 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) { - - Eterm *ks,*vs; +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 +{ + Eterm *ks, *vs; map_t *mp; - Uint n,i; + Uint n, i; - mp = (map_t*)map_val(map); + mp = (map_t *)map_val_rel(map, map_base); n = map_get_size(mp); - ks = map_get_keys(mp); + + if (n == 0) { + return NULL; + } + + ks = (Eterm *)tuple_val_rel(mp->keys, map_base) + 1; vs = map_get_values(mp); - for( i = 0; i < n; i++) { - if (EQ(ks[i], key)) { - *value = vs[i]; - return 1; - } + 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], NULL, key, map_base)) { + return &vs[i]; + } } - return 0; + return NULL; } 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); } @@ -150,52 +169,22 @@ BIF_RETTYPE maps_find_2(BIF_ALIST_2) { } 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"; diff --git a/erts/emulator/beam/erl_map.h b/erts/emulator/beam/erl_map.h index cfacb2ec28..2e02ca4677 100644 --- a/erts/emulator/beam/erl_map.h +++ b/erts/emulator/beam/erl_map.h @@ -64,9 +64,17 @@ typedef struct map_s { 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); + +#if HALFWORD_HEAP +const Eterm * +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 +const Eterm * +erts_maps_get(Eterm key, Eterm map); +# define erts_maps_get_rel(A, B, B_BASE) erts_maps_get(A, B) #endif +#endif diff --git a/erts/emulator/beam/erl_message.c b/erts/emulator/beam/erl_message.c index 59a677a12c..8870fac7d9 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); diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c index ff551ea3af..198acfd128 100644 --- a/erts/emulator/beam/erl_nif.c +++ b/erts/emulator/beam/erl_nif.c @@ -472,6 +472,18 @@ 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 +551,7 @@ 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 +571,6 @@ 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 +1522,264 @@ 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 +1791,7 @@ 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 +1799,106 @@ 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 */ @@ -1733,10 +1955,16 @@ int enif_get_map_value(ErlNifEnv* env, Eterm key, Eterm *value) { + const Eterm *ret; if (is_not_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, @@ -1977,6 +2205,35 @@ 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 +2343,48 @@ 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 +2410,8 @@ 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 +2420,16 @@ 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 +2437,8 @@ 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 +2485,17 @@ 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 +2503,21 @@ 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..849024453c 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 @@ -243,6 +252,11 @@ 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 +282,8 @@ 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..74e38c13df 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) { @@ -283,13 +303,9 @@ 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 +315,7 @@ 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 +445,7 @@ 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 +459,65 @@ 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; @@ -502,19 +567,10 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount, 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); - + WSTACK_PUSH5(s, vs[n], PRT_TERM, PRT_ASSOC, ks[n], 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); + WSTACK_PUSH6(s, PRT_COMMA, vs[n], PRT_TERM, PRT_ASSOC, + ks[n], PRT_TERM); } } } diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index b73f9b7f92..b0276fbae0 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -148,6 +148,12 @@ 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; @@ -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++) { @@ -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); + return (Uint64) erts_atomic64_read_nob((erts_atomic64_t *) var); } static ERTS_INLINE void aschedtime_set(ErtsAtomicSchedTime *var, Uint64 val) { - erts_atomic_set_nob((erts_atomic_t *) var, (erts_aint_t) val); + erts_atomic64_set_nob((erts_atomic64_t *) var, (erts_aint64_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 + erts_atomic64_init_nob((erts_atomic64_t *) var, (erts_aint64_t) 0); } -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); -} - -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); -} - -#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); @@ -2380,29 +2339,47 @@ 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 @@ -2779,7 +2756,7 @@ 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); @@ -2943,7 +2920,7 @@ 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; } @@ -2965,7 +2942,7 @@ 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... @@ -3219,11 +3196,11 @@ 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; } @@ -3755,17 +3732,25 @@ 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))); } @@ -5874,6 +5859,9 @@ 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))) { @@ -5987,7 +5975,8 @@ 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 @@ -5996,6 +5985,11 @@ 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 @@ -6005,6 +5999,9 @@ 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; @@ -6040,7 +6037,9 @@ 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) @@ -6053,15 +6052,18 @@ 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; @@ -6069,7 +6071,8 @@ 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, @@ -6077,16 +6080,27 @@ 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)); @@ -6095,7 +6109,7 @@ 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; @@ -6108,7 +6122,7 @@ 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 @@ -6118,6 +6132,8 @@ 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) { @@ -6132,8 +6148,14 @@ 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 @@ -6200,7 +6222,7 @@ 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; @@ -6217,7 +6239,8 @@ 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, @@ -7818,6 +7841,9 @@ erts_start_schedulers(void) #ifdef ETHR_HAVE_THREAD_NAMES opts.name = malloc(80); + if (!opts.name) { + ERTS_INTERNAL_ERROR("malloc failed to allocate memory!"); + } #endif #ifdef ERTS_SMP @@ -8030,7 +8056,8 @@ 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); } } @@ -8065,7 +8092,7 @@ 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 { @@ -8223,7 +8250,8 @@ 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); } @@ -8583,7 +8611,8 @@ 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); } @@ -8713,7 +8742,7 @@ 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); } @@ -8732,7 +8761,7 @@ 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); @@ -9147,7 +9176,7 @@ 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))) { /* * Schedule system-level activities. */ @@ -9155,8 +9184,6 @@ 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 @@ -9188,7 +9215,9 @@ 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 @@ -9968,8 +9997,10 @@ 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) @@ -10714,7 +10745,7 @@ 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)); @@ -11035,7 +11066,8 @@ 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) @@ -11076,7 +11108,8 @@ 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; @@ -11721,8 +11754,9 @@ 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); @@ -11814,6 +11848,7 @@ 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; @@ -11964,6 +11999,7 @@ 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 @@ -12011,6 +12047,9 @@ 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 @@ -12055,7 +12094,7 @@ 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); } diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index ed6dadbffa..55009ad904 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; @@ -350,13 +352,7 @@ 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 +484,6 @@ typedef struct { } ErtsSchedWallTime; typedef struct { - Uint64 reclaimed; - Uint64 garbage_cols; -} ErtsGCInfo; - -typedef struct { int sched; erts_aint32_t aux_work; } ErtsDelayedAuxWorkWakeupJob; @@ -734,13 +725,9 @@ 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 +754,8 @@ 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; @@ -1367,6 +1352,9 @@ 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 *); @@ -1704,17 +1692,17 @@ 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 +1805,10 @@ 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); diff --git a/erts/emulator/beam/erl_process_dict.c b/erts/emulator/beam/erl_process_dict.c index 23e5bf737f..3ce707efda 100644 --- a/erts/emulator/beam/erl_process_dict.c +++ b/erts/emulator/beam/erl_process_dict.c @@ -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_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_thr_progress.c b/erts/emulator/beam/erl_thr_progress.c index 545a0343d0..664c479eb6 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 + erts_atomic64_init_nob(atmc, (erts_aint64_t) val); } -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); -} - -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 diff --git a/erts/emulator/beam/erl_thr_progress.h b/erts/emulator/beam/erl_thr_progress.h index 5f392944c2..03ddbd467c 100644 --- a/erts/emulator/beam/erl_thr_progress.h +++ b/erts/emulator/beam/erl_thr_progress.h @@ -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..7214f3ea33 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) @@ -570,6 +583,29 @@ 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, @@ -1200,6 +1236,441 @@ 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 +1933,116 @@ 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 @@ -2383,6 +2964,104 @@ 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 diff --git a/erts/emulator/beam/erl_utils.h b/erts/emulator/beam/erl_utils.h index 0807649ea1..c32f8fd61c 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 diff --git a/erts/emulator/beam/erl_vm.h b/erts/emulator/beam/erl_vm.h index b7de8208ad..78d98229d8 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. diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c index 8d240355b0..601cbe9d7d 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" @@ -498,15 +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) @@ -527,19 +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) @@ -1740,54 +1761,14 @@ 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) { @@ -1899,8 +1880,6 @@ 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 */ @@ -1961,8 +1940,6 @@ 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; @@ -2325,8 +2302,9 @@ 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_LAST_ARRAY_ELEMENT ((Eterm) 5) static byte* enc_term(ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dflags, @@ -2362,6 +2340,9 @@ 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; + } } } @@ -2385,8 +2366,8 @@ 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: @@ -2399,6 +2380,39 @@ 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_LAST_ARRAY_ELEMENT: /* obj is the tuple */ { @@ -2417,17 +2431,16 @@ 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); @@ -2576,8 +2589,7 @@ 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; @@ -2593,18 +2605,8 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, 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]); - } - - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) vptr[0]); - - obj = kptr[0]; - goto L_jump_start; + WSTACK_PUSH4(s, (UWord)kptr, (UWord)vptr, + ENC_MAP_PAIR, size); } } break; @@ -2642,6 +2644,7 @@ 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) { @@ -2687,7 +2690,7 @@ 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. */ @@ -2697,7 +2700,7 @@ 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 { /* @@ -2711,11 +2714,18 @@ 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: @@ -2744,13 +2754,12 @@ 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; @@ -2764,16 +2773,6 @@ 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 @@ -2805,7 +2804,13 @@ 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; @@ -3385,8 +3390,6 @@ 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; @@ -3439,8 +3442,6 @@ 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; diff --git a/erts/emulator/beam/external.h b/erts/emulator/beam/external.h index bf00958eb1..f120e96e3b 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 @@ -150,6 +150,7 @@ typedef struct { Uint extsize; } ErtsBinary2TermState; + /* -------------------------------------------------------------------------- */ void erts_init_atom_cache_map(ErtsAtomCacheMap *); @@ -161,8 +162,12 @@ 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 891046a8b5..ec8c1e3ccb 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 */ { @@ -479,6 +480,17 @@ do { \ *s.sp++ = (z); \ } while(0) +#define ESTACK_PUSH4(s, E1, E2, E3, E4) \ +do { \ + if (s.sp > s.end - 4) { \ + erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ + } \ + *s.sp++ = (E1); \ + *s.sp++ = (E2); \ + *s.sp++ = (E3); \ + *s.sp++ = (E4); \ +} while(0) + #define ESTACK_COUNT(s) (s.sp - s.start) #define ESTACK_ISEMPTY(s) (s.sp == s.start) #define ESTACK_POP(s) (*(--s.sp)) @@ -597,6 +609,42 @@ do { \ *s.wsp++ = (z); \ } while(0) +#define WSTACK_PUSH4(s, A1, A2, A3, A4) \ +do { \ + if (s.wsp > s.wend - 4) { \ + erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + } \ + *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, WSTK_DEF_STACK(s)); \ + } \ + *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, WSTK_DEF_STACK(s)); \ + } \ + *s.wsp++ = (A1); \ + *s.wsp++ = (A2); \ + *s.wsp++ = (A3); \ + *s.wsp++ = (A4); \ + *s.wsp++ = (A5); \ + *s.wsp++ = (A6); \ +} while(0) + #define WSTACK_COUNT(s) (s.wsp - s.wstart) #define WSTACK_ISEMPTY(s) (s.wsp == s.wstart) #define WSTACK_POP(s) (*(--s.wsp)) @@ -809,23 +857,6 @@ 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 */ diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c index edf4a28784..4d262ff022 100644 --- a/erts/emulator/beam/io.c +++ b/erts/emulator/beam/io.c @@ -1218,9 +1218,10 @@ 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 +1248,39 @@ 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 +1295,14 @@ 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 +1319,31 @@ 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 +1358,13 @@ 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 +3153,6 @@ 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); @@ -5460,8 +5504,6 @@ 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; @@ -5953,9 +5995,7 @@ 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 +6025,6 @@ ErlDrvBinary* driver_realloc_binary(ErlDrvBinary* bin, ErlDrvSizeT size) if (!newbin) return NULL; - newbin->orig_size = size; return Binary2ErlDrvBinary(newbin); } @@ -6129,7 +6168,7 @@ 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; @@ -7166,7 +7205,7 @@ 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 +7267,18 @@ 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 + } + } diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h index 05f07e57b2..c29d4b3777 100644 --- a/erts/emulator/beam/sys.h +++ b/erts/emulator/beam/sys.h @@ -66,8 +66,12 @@ */ #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 @@ -274,6 +278,7 @@ __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 @@ -347,6 +352,7 @@ typedef long long Sint; typedef Uint UWord; typedef Sint SWord; +#define ERTS_UINT_MAX ERTS_UWORD_MAX #endif /* HALFWORD_HEAP */ @@ -499,7 +505,7 @@ 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 @@ -742,6 +748,14 @@ 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); + /* xxxP */ #define SYS_DEFAULT_FLOAT_DECIMALS 20 void init_sys_float(void); diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c index 72092ec7b0..b341c4d949 100644 --- a/erts/emulator/beam/utils.c +++ b/erts/emulator/beam/utils.c @@ -48,6 +48,10 @@ #include "erl_sched_spec_pre_alloc.h" #include "beam_bp.h" #include "erl_ptab.h" +#include "erl_check_io.h" +#ifdef HIPE +# include "hipe_mode_switch.h" +#endif #undef M_TRIM_THRESHOLD #undef M_TOP_PAD @@ -1228,25 +1232,20 @@ make_hash2(Eterm term) 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); + ESTACK_PUSH4(s, hash_xor_values, hash_xor_keys, hash, 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); + ESTACK_PUSH2(s, HASH_MAP_VAL, 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); + ESTACK_PUSH2(s, HASH_MAP_KEY, tmp); } goto hash2_common; } @@ -3948,6 +3947,9 @@ 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 @@ -4212,19 +4214,7 @@ 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 @@ -4246,55 +4236,13 @@ 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); } @@ -4302,38 +4250,10 @@ 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); } @@ -4341,38 +4261,10 @@ 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 diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c index 09bada457d..db8a251fdd 100644 --- a/erts/emulator/drivers/common/inet_drv.c +++ b/erts/emulator/drivers/common/inet_drv.c @@ -4372,7 +4372,7 @@ static int erl_inet_close(inet_descriptor* desc) desc_close(desc); desc->state = INET_STATE_CLOSED; } else if (desc->prebound && (desc->s != INVALID_SOCKET)) { - sock_select(desc, FD_READ | FD_WRITE | FD_CLOSE, 0); + sock_select(desc, FD_READ | FD_WRITE | FD_CLOSE | ERL_DRV_USE_NO_CALLBACK, 0); desc->event_mask = 0; #ifdef __WIN32__ desc->forced_events = 0; @@ -4536,16 +4536,19 @@ static ErlDrvSSizeT inet_ctl_open(inet_descriptor* desc, int domain, int type, /* as inet_open but pass in an open socket (MUST BE OF RIGHT TYPE) */ static ErlDrvSSizeT inet_ctl_fdopen(inet_descriptor* desc, int domain, int type, - SOCKET s, char** rbuf, ErlDrvSizeT rsize) + SOCKET s, Uint32 bound, + char** rbuf, ErlDrvSizeT rsize) { inet_address name; unsigned int sz = sizeof(name); - /* check that it is a socket and that the socket is bound */ - if (IS_SOCKET_ERROR(sock_name(s, (struct sockaddr*) &name, &sz))) - return ctl_error(sock_errno(), rbuf, rsize); - if (name.sa.sa_family != domain) - return ctl_error(EINVAL, rbuf, rsize); + if (bound) { + /* check that it is a socket and that the socket is bound */ + if (IS_SOCKET_ERROR(sock_name(s, (struct sockaddr*) &name, &sz))) + return ctl_error(sock_errno(), rbuf, rsize); + if (name.sa.sa_family != domain) + return ctl_error(EINVAL, rbuf, rsize); + } #ifdef __OSE__ /* for fdopen duplicating the sd will allow to uniquely identify the signal from OSE with erlang port */ @@ -4560,7 +4563,12 @@ static ErlDrvSSizeT inet_ctl_fdopen(inet_descriptor* desc, int domain, int type, #ifdef __WIN32__ driver_select(desc->port, desc->event, ERL_DRV_READ, 1); #endif - desc->state = INET_STATE_BOUND; /* assume bound */ + + if (bound) + desc->state = INET_STATE_BOUND; + else + desc->state = INET_STATE_OPEN; + if (type == SOCK_STREAM) { /* check if connected */ sz = sizeof(name); if (!IS_SOCKET_ERROR(sock_peer(s, (struct sockaddr*) &name, &sz))) { @@ -5772,7 +5780,7 @@ done: ia_p->Ipv6IfIndex && ia_p->Ipv6IfIndex != index) { - /* Oops, there was an other interface for IPv6. Possible? XXX */ + /* Oops, there was another interface for IPv6. Possible? XXX */ index = ia_p->Ipv6IfIndex; goto index; } @@ -9121,10 +9129,11 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData e, unsigned int cmd, break; } - case INET_REQ_FDOPEN: { /* pass in an open socket */ + case INET_REQ_FDOPEN: { /* pass in an open (and optionally bound) socket */ int domain; + int bound; DEBUGF(("tcp_inet_ctl(%ld): FDOPEN\r\n", (long)desc->inet.port)); - if (len != 6) return ctl_error(EINVAL, rbuf, rsize); + if (len != 6 && len != 10) return ctl_error(EINVAL, rbuf, rsize); switch(buf[0]) { case INET_AF_INET: domain = AF_INET; @@ -9142,8 +9151,13 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData e, unsigned int cmd, return ctl_error(EINVAL, rbuf, rsize); } if (buf[1] != INET_TYPE_STREAM) return ctl_error(EINVAL, rbuf, rsize); + + if (len == 6) bound = 1; + else bound = get_int32(buf+2+4); + return inet_ctl_fdopen(INETP(desc), domain, SOCK_STREAM, - (SOCKET) get_int32(buf+2), rbuf, rsize); + (SOCKET) get_int32(buf+2), + bound, rbuf, rsize); break; } @@ -11116,10 +11130,11 @@ static ErlDrvSSizeT packet_inet_ctl(ErlDrvData e, unsigned int cmd, char* buf, return replen; - case INET_REQ_FDOPEN: { /* pass in an open (and bound) socket */ + case INET_REQ_FDOPEN: { /* pass in an open (and optionally bound) socket */ SOCKET s; + int bound; DEBUGF(("packet inet_ctl(%ld): FDOPEN\r\n", (long)desc->port)); - if (len != 6) { + if (len != 6 && len != 10) { return ctl_error(EINVAL, rbuf, rsize); } switch (buf[0]) { @@ -11144,7 +11159,11 @@ static ErlDrvSSizeT packet_inet_ctl(ErlDrvData e, unsigned int cmd, char* buf, return ctl_error(EINVAL, rbuf, rsize); } s = (SOCKET)get_int32(buf+2); - replen = inet_ctl_fdopen(desc, af, type, s, rbuf, rsize); + + if (len == 6) bound = 1; + else bound = get_int32(buf+2+4); + + replen = inet_ctl_fdopen(desc, af, type, s, bound, rbuf, rsize); if ((*rbuf)[0] != INET_REP_ERROR) { if (desc->active) diff --git a/erts/emulator/drivers/unix/multi_drv.c b/erts/emulator/drivers/unix/multi_drv.c index 822c96730c..724d325ed5 100644 --- a/erts/emulator/drivers/unix/multi_drv.c +++ b/erts/emulator/drivers/unix/multi_drv.c @@ -20,7 +20,7 @@ /* Purpose: Multidriver interface This is an example of a driver which allows multiple instances of itself. I.e have one erlang process execute open_port(multi......) and - at the same time have an other erlang process open an other port + at the same time have another erlang process open another port running multi there as well. */ diff --git a/erts/emulator/drivers/win32/win_efile.c b/erts/emulator/drivers/win32/win_efile.c index a321bb9641..7e4043fc1b 100644 --- a/erts/emulator/drivers/win32/win_efile.c +++ b/erts/emulator/drivers/win32/win_efile.c @@ -1288,6 +1288,10 @@ do_fileinfo(Efile_call_state* state, Efile_info* pInfo, { HANDLE handle; /* Handle returned by CreateFile() */ BY_HANDLE_FILE_INFORMATION fileInfo; /* from CreateFile() */ + + /* We initialise nNumberOfLinks as GetFileInformationByHandle + does not always initialise this field */ + fileInfo.nNumberOfLinks = 1; if (handle = CreateFileW(name, GENERIC_READ, FILE_SHARE_FLAGS, NULL, OPEN_EXISTING, 0, NULL)) { GetFileInformationByHandle(handle, &fileInfo); diff --git a/erts/emulator/hipe/hipe_amd64.c b/erts/emulator/hipe/hipe_amd64.c index b5dff06987..16c597e7b4 100644 --- a/erts/emulator/hipe/hipe_amd64.c +++ b/erts/emulator/hipe/hipe_amd64.c @@ -224,18 +224,19 @@ void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process * return alloc_code(nrbytes); } -/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2() - and hipe_bif0.c:hipe_make_stub() */ -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) + +/* Make stub for native code calling exported beam function. +*/ +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { /* * This creates a native code stub with the following contents: * - * movq $Address, P_BEAM_IP(%ebp) %% Actually two movl + * movq $Address, P_CALLEE_EXP(%ebp) %% Actually two movl * movb $Arity, P_ARITY(%ebp) * jmp callemu * - * The stub has variable size, depending on whether the P_BEAM_IP + * The stub has variable size, depending on whether the P_CALLEE_EXP * and P_ARITY offsets fit in 8-bit signed displacements or not. * The rel32 offset in the final jmp depends on its actual location, * which also depends on the size of the previous instructions. @@ -248,49 +249,49 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) codeSize = /* 23, 26, 29, or 32 bytes */ 23 + /* 23 when all offsets are 8-bit */ - (P_BEAM_IP >= 128 ? 3 : 0) + - ((P_BEAM_IP + 4) >= 128 ? 3 : 0) + + (P_CALLEE_EXP >= 128 ? 3 : 0) + + ((P_CALLEE_EXP + 4) >= 128 ? 3 : 0) + (P_ARITY >= 128 ? 3 : 0); codep = code = alloc_code(codeSize); - /* movl $beamAddress, P_BEAM_IP(%ebp); 3 or 6 bytes, plus 4 */ + /* movl $callee_exp, P_CALLEE_EXP(%ebp); 3 or 6 bytes, plus 4 */ codep[0] = 0xc7; -#if P_BEAM_IP >= 128 +#if P_CALLEE_EXP >= 128 codep[1] = 0x85; /* disp32[EBP] */ - codep[2] = P_BEAM_IP & 0xFF; - codep[3] = (P_BEAM_IP >> 8) & 0xFF; - codep[4] = (P_BEAM_IP >> 16) & 0xFF; - codep[5] = (P_BEAM_IP >> 24) & 0xFF; + codep[2] = P_CALLEE_EXP & 0xFF; + codep[3] = (P_CALLEE_EXP >> 8) & 0xFF; + codep[4] = (P_CALLEE_EXP >> 16) & 0xFF; + codep[5] = (P_CALLEE_EXP >> 24) & 0xFF; codep += 6; #else codep[1] = 0x45; /* disp8[EBP] */ - codep[2] = P_BEAM_IP; + codep[2] = P_CALLEE_EXP; codep += 3; #endif - codep[0] = ((unsigned long)beamAddress ) & 0xFF; - codep[1] = ((unsigned long)beamAddress >> 8) & 0xFF; - codep[2] = ((unsigned long)beamAddress >> 16) & 0xFF; - codep[3] = ((unsigned long)beamAddress >> 24) & 0xFF; + codep[0] = ((unsigned long)callee_exp ) & 0xFF; + codep[1] = ((unsigned long)callee_exp >> 8) & 0xFF; + codep[2] = ((unsigned long)callee_exp >> 16) & 0xFF; + codep[3] = ((unsigned long)callee_exp >> 24) & 0xFF; codep += 4; - /* movl (shl 32 $beamAddress), P_BEAM_IP+4(%ebp); 3 or 6 bytes, plus 4 */ + /* movl (shl 32 $callee_exp), P_CALLEE_EXP+4(%ebp); 3 or 6 bytes, plus 4 */ codep[0] = 0xc7; -#if P_BEAM_IP+4 >= 128 +#if P_CALLEE_EXP+4 >= 128 codep[1] = 0x85; /* disp32[EBP] */ - codep[2] = (P_BEAM_IP+4) & 0xFF; - codep[3] = ((P_BEAM_IP+4) >> 8) & 0xFF; - codep[4] = ((P_BEAM_IP+4) >> 16) & 0xFF; - codep[5] = ((P_BEAM_IP+4) >> 24) & 0xFF; + codep[2] = (P_CALLEE_EXP+4) & 0xFF; + codep[3] = ((P_CALLEE_EXP+4) >> 8) & 0xFF; + codep[4] = ((P_CALLEE_EXP+4) >> 16) & 0xFF; + codep[5] = ((P_CALLEE_EXP+4) >> 24) & 0xFF; codep += 6; #else codep[1] = 0x45; /* disp8[EBP] */ - codep[2] = (P_BEAM_IP+4); + codep[2] = (P_CALLEE_EXP+4); codep += 3; #endif - codep[0] = ((unsigned long)beamAddress >> 32) & 0xFF; - codep[1] = ((unsigned long)beamAddress >> 40) & 0xFF; - codep[2] = ((unsigned long)beamAddress >> 48) & 0xFF; - codep[3] = ((unsigned long)beamAddress >> 56) & 0xFF; + codep[0] = ((unsigned long)callee_exp >> 32) & 0xFF; + codep[1] = ((unsigned long)callee_exp >> 40) & 0xFF; + codep[2] = ((unsigned long)callee_exp >> 48) & 0xFF; + codep[3] = ((unsigned long)callee_exp >> 56) & 0xFF; codep += 4; /* movb $beamArity, P_ARITY(%ebp); 3 or 6 bytes */ diff --git a/erts/emulator/hipe/hipe_amd64_asm.m4 b/erts/emulator/hipe/hipe_amd64_asm.m4 index 7c81040b8b..b4b3c073ab 100644 --- a/erts/emulator/hipe/hipe_amd64_asm.m4 +++ b/erts/emulator/hipe/hipe_amd64_asm.m4 @@ -33,7 +33,35 @@ define(HEAP_LIMIT_IN_REGISTER,0)dnl global for HL define(SIMULATE_NSP,0)dnl change to 1 to simulate call/ret insns `#define AMD64_LEAF_WORDS 'LEAF_WORDS -`#define LEAF_WORDS 'LEAF_WORDS +`#define LEAF_WORDS 'LEAF_WORDS +`#define AMD64_NR_ARG_REGS 'NR_ARG_REGS +`#define NR_ARG_REGS 'NR_ARG_REGS + +`#define AMD64_HP_IN_REGISTER 'HP_IN_REGISTER +`#if AMD64_HP_IN_REGISTER' +`#define AMD64_HEAP_POINTER 15' +define(HP,%r15)dnl Only change this together with above +`#endif' + +`#define AMD64_FCALLS_IN_REGISTER 'FCALLS_IN_REGISTER +`#if AMD64_FCALLS_IN_REGISTER' +`#define AMD64_FCALLS_REGISTER 11' +define(FCALLS,%r11)dnl This goes together with line above +`#endif' + +`#define AMD64_HEAP_LIMIT_IN_REGISTER 'HEAP_LIMIT_IN_REGISTER +`#if AMD64_HEAP_LIMIT_IN_REGISTER' +`#define AMD64_HEAP_LIMIT_REGISTER 12' +define(HEAP_LIMIT,%r12)dnl Change this together with line above +`#endif' + +`#define AMD64_SIMULATE_NSP 'SIMULATE_NSP + + +`#ifdef ASM' +/* + * Only assembler stuff from here on (when included from *.S) + */ /* * Workarounds for Darwin. @@ -63,33 +91,24 @@ ifelse(OPSYS,darwin,`` */ `#define P %rbp' -`#define AMD64_HP_IN_REGISTER 'HP_IN_REGISTER `#if AMD64_HP_IN_REGISTER -#define AMD64_HEAP_POINTER 15' -define(HP,%r15)dnl Only change this together with above -`#define SAVE_HP movq 'HP`, P_HP(P) +#define SAVE_HP movq 'HP`, P_HP(P) #define RESTORE_HP movq P_HP(P), 'HP` #else #define SAVE_HP /*empty*/ #define RESTORE_HP /*empty*/ #endif' -`#define AMD64_FCALLS_IN_REGISTER 'FCALLS_IN_REGISTER `#if AMD64_FCALLS_IN_REGISTER -#define AMD64_FCALLS_REGISTER 11' -define(FCALLS,%r11)dnl This goes together with line above -`#define SAVE_FCALLS movq 'FCALLS`, P_FCALLS(P) +#define SAVE_FCALLS movq 'FCALLS`, P_FCALLS(P) #define RESTORE_FCALLS movq P_FCALLS(P), 'FCALLS` #else #define SAVE_FCALLS /*empty*/ #define RESTORE_FCALLS /*empty*/ #endif' -`#define AMD64_HEAP_LIMIT_IN_REGISTER 'HEAP_LIMIT_IN_REGISTER `#if AMD64_HEAP_LIMIT_IN_REGISTER -#define AMD64_HEAP_LIMIT_REGISTER 12' -define(HEAP_LIMIT,%r12)dnl Change this together with line above -`#define RESTORE_HEAP_LIMIT movq P_HP_LIMIT(P), 'HEAP_LIMIT` +#define RESTORE_HEAP_LIMIT movq P_HP_LIMIT(P), 'HEAP_LIMIT` #else #define RESTORE_HEAP_LIMIT /*empty*/ #endif' @@ -99,7 +118,6 @@ define(NSP,%rsp)dnl `#define SAVE_CSP movq %rsp, P_CSP(P) #define RESTORE_CSP movq P_CSP(P), %rsp' -`#define AMD64_SIMULATE_NSP 'SIMULATE_NSP /* * Context switching macros. @@ -132,8 +150,6 @@ define(NSP,%rsp)dnl /* * Argument (parameter) registers. */ -`#define AMD64_NR_ARG_REGS 'NR_ARG_REGS -`#define NR_ARG_REGS 'NR_ARG_REGS define(defarg,`define(ARG$1,`$2')dnl #`define ARG'$1 $2' @@ -263,4 +279,6 @@ define(NBIF_RET,`NBIF_RET_N(eval(RET_POP($1)))')dnl `/* #define NBIF_RET_3 'NBIF_RET(3)` */' `/* #define NBIF_RET_5 'NBIF_RET(5)` */' +`#endif /* ASM */' + `#endif /* HIPE_AMD64_ASM_H */' diff --git a/erts/emulator/hipe/hipe_amd64_bifs.m4 b/erts/emulator/hipe/hipe_amd64_bifs.m4 index 0de69a617f..7a4bb30447 100644 --- a/erts/emulator/hipe/hipe_amd64_bifs.m4 +++ b/erts/emulator/hipe/hipe_amd64_bifs.m4 @@ -18,7 +18,7 @@ changecom(`/*', `*/')dnl * %CopyrightEnd% */ - +#`define ASM' include(`hipe/hipe_amd64_asm.m4') #`include' "config.h" #`include' "hipe_literals.h" @@ -39,7 +39,10 @@ define(HANDLE_GOT_MBUF,` jmp 2b') `#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP) -# define CALL_BIF(F) movq $CSYM(F), P_BIF_CALLEE(P); call CSYM(hipe_debug_bif_wrapper) +# define CALL_BIF(F) \ + movq CSYM(F)@GOTPCREL(%rip), %r11; \ + movq %r11, P_BIF_CALLEE(P); \ + call CSYM(hipe_debug_bif_wrapper) #else # define CALL_BIF(F) call CSYM(F) #endif' diff --git a/erts/emulator/hipe/hipe_amd64_glue.S b/erts/emulator/hipe/hipe_amd64_glue.S index 8816906870..955f7362b4 100644 --- a/erts/emulator/hipe/hipe_amd64_glue.S +++ b/erts/emulator/hipe/hipe_amd64_glue.S @@ -17,10 +17,9 @@ * %CopyrightEnd% */ - +#define ASM #include "hipe_amd64_asm.h" #include "hipe_literals.h" -#define ASM #include "hipe_mode_switch.h" /* @@ -109,7 +108,7 @@ ASYM(nbif_return): * stub (hipe_x86_loader.erl) which should look as follows: * * stub for f/N: - * movq $<f's BEAM code address>, P_BEAM_IP(P) + * movq $<f's export entry address>, P_CALLEE_EXP(P) * movb $<N>, P_ARITY(P) * jmp nbif_callemu * @@ -119,7 +118,7 @@ ASYM(nbif_return): GLOBAL(ASYM(nbif_callemu)) ASYM(nbif_callemu): STORE_ARG_REGS - movl $HIPE_MODE_SWITCH_RES_CALL, %eax + movl $HIPE_MODE_SWITCH_RES_CALL_EXPORTED, %eax jmp .suspend_exit /* diff --git a/erts/emulator/hipe/hipe_arm.c b/erts/emulator/hipe/hipe_arm.c index 3db3ffe9b1..165eb543c8 100644 --- a/erts/emulator/hipe/hipe_arm.c +++ b/erts/emulator/hipe/hipe_arm.c @@ -260,9 +260,9 @@ int hipe_patch_insn(void *address, Uint32 value, Eterm type) return 0; } -/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2() - and hipe_bif0.c:hipe_make_stub() */ -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) +/* Make stub for native code calling exported beam function +*/ +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { unsigned int *code; unsigned int *tramp_callemu; @@ -272,9 +272,9 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) * Native code calls BEAM via a stub looking as follows: * * mov r0, #beamArity - * ldr r8, [pc,#0] // beamAddress + * ldr r8, [pc,#0] // callee_exp * b nbif_callemu - * .long beamAddress + * .long callee_exp * * I'm using r0 and r8 since they aren't used for * parameter passing in native code. The branch to @@ -292,12 +292,12 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) /* mov r0, #beamArity */ code[0] = 0xE3A00000 | (beamArity & 0xFF); - /* ldr r8, [pc,#0] // beamAddress */ + /* ldr r8, [pc,#0] // callee_exp */ code[1] = 0xE59F8000; /* b nbif_callemu */ code[2] = 0xEA000000 | (callemu_offset & 0x00FFFFFF); - /* .long beamAddress */ - code[3] = (unsigned int)beamAddress; + /* .long callee_exp */ + code[3] = (unsigned int)callee_exp; hipe_flush_icache_range(code, 4*sizeof(int)); diff --git a/erts/emulator/hipe/hipe_arm_asm.m4 b/erts/emulator/hipe/hipe_arm_asm.m4 index 85dc84973d..b2e3f83d1e 100644 --- a/erts/emulator/hipe/hipe_arm_asm.m4 +++ b/erts/emulator/hipe/hipe_arm_asm.m4 @@ -29,6 +29,14 @@ define(LEAF_WORDS,16)dnl number of stack words for leaf functions define(NR_ARG_REGS,3)dnl admissible values are 0 to 6, inclusive `#define ARM_LEAF_WORDS 'LEAF_WORDS +`#define ARM_NR_ARG_REGS 'NR_ARG_REGS +`#define NR_ARG_REGS 'NR_ARG_REGS + + +`#ifdef ASM' +/* + * Only assembler stuff from here on (when included from *.S) + */ /* * Reserved registers. @@ -77,8 +85,6 @@ define(NR_ARG_REGS,3)dnl admissible values are 0 to 6, inclusive /* * Argument (parameter) registers. */ -`#define ARM_NR_ARG_REGS 'NR_ARG_REGS -`#define NR_ARG_REGS 'NR_ARG_REGS define(defarg,`define(ARG$1,`$2')dnl #`define ARG'$1 $2' @@ -195,4 +201,6 @@ define(QUICK_CALL_RET,`NBIF_POP_N(eval(RET_POP($2)))b $1')dnl `/* #define QUICK_CALL_RET_F_3 'QUICK_CALL_RET(F,3)` */' `/* #define QUICK_CALL_RET_F_5 'QUICK_CALL_RET(F,5)` */' +`#endif /* ASM */' + `#endif /* HIPE_ARM_ASM_H */' diff --git a/erts/emulator/hipe/hipe_arm_bifs.m4 b/erts/emulator/hipe/hipe_arm_bifs.m4 index bd8bc5ab6b..57e51bb8b1 100644 --- a/erts/emulator/hipe/hipe_arm_bifs.m4 +++ b/erts/emulator/hipe/hipe_arm_bifs.m4 @@ -19,6 +19,7 @@ changecom(`/*', `*/')dnl */ +#`define ASM' include(`hipe/hipe_arm_asm.m4') #`include' "config.h" #`include' "hipe_literals.h" diff --git a/erts/emulator/hipe/hipe_arm_glue.S b/erts/emulator/hipe/hipe_arm_glue.S index 2e2b8604a6..069cb4512e 100644 --- a/erts/emulator/hipe/hipe_arm_glue.S +++ b/erts/emulator/hipe/hipe_arm_glue.S @@ -17,10 +17,9 @@ * %CopyrightEnd% */ - +#define ASM #include "hipe_arm_asm.h" #include "hipe_literals.h" -#define ASM #include "hipe_mode_switch.h" .text @@ -135,7 +134,7 @@ hipe_arm_throw_to_native: * which should look as follows: * * stub for f/N: - * <set r8 to f's BEAM code address> + * <set r8 to f's export entry address> * <set r0 to N> * b nbif_callemu * @@ -143,10 +142,10 @@ hipe_arm_throw_to_native: */ .global nbif_callemu nbif_callemu: - str r8, [P, #P_BEAM_IP] + str r8, [P, #P_CALLEE_EXP] str r0, [P, #P_ARITY] STORE_ARG_REGS - mov r0, #HIPE_MODE_SWITCH_RES_CALL + mov r0, #HIPE_MODE_SWITCH_RES_CALL_EXPORTED b .suspend_exit /* diff --git a/erts/emulator/hipe/hipe_bif0.c b/erts/emulator/hipe/hipe_bif0.c index 2497d51df1..9eb0b88ced 100644 --- a/erts/emulator/hipe/hipe_bif0.c +++ b/erts/emulator/hipe/hipe_bif0.c @@ -89,25 +89,6 @@ static Eterm address_to_term(const void *address, Process *p) /* * BIFs for reading and writing memory. Used internally by HiPE. */ -#if 0 /* XXX: unused */ -BIF_RETTYPE hipe_bifs_read_u8_1(BIF_ALIST_1) -{ - unsigned char *address = term_to_address(BIF_ARG_1); - if (!address) - BIF_ERROR(BIF_P, BADARG); - BIF_RET(make_small(*address)); -} -#endif - -#if 0 /* XXX: unused */ -BIF_RETTYPE hipe_bifs_read_u32_1(BIF_ALIST_1) -{ - Uint32 *address = term_to_address(BIF_ARG_1); - if (!address || !hipe_word32_address_ok(address)) - BIF_ERROR(BIF_P, BADARG); - BIF_RET(Uint_to_term(*address, BIF_P)); -} -#endif BIF_RETTYPE hipe_bifs_write_u8_2(BIF_ALIST_2) { @@ -120,22 +101,6 @@ BIF_RETTYPE hipe_bifs_write_u8_2(BIF_ALIST_2) BIF_RET(NIL); } -#if 0 /* XXX: unused */ -BIF_RETTYPE hipe_bifs_write_s32_2(BIF_ALIST_2) -{ - Sint32 *address; - Sint value; - - address = term_to_address(BIF_ARG_1); - if (!address || !hipe_word32_address_ok(address)) - BIF_ERROR(BIF_P, BADARG); - if (!term_to_Sint32(BIF_ARG_2, &value)) - BIF_ERROR(BIF_P, BADARG); - *address = value; - BIF_RET(NIL); -} -#endif - BIF_RETTYPE hipe_bifs_write_u32_2(BIF_ALIST_2) { Uint32 *address; @@ -639,33 +604,6 @@ BIF_RETTYPE hipe_bifs_fun_to_address_1(BIF_ALIST_1) BIF_RET(address_to_term(pc, BIF_P)); } -static void *hipe_get_emu_address(Eterm m, Eterm f, unsigned int arity, int is_remote) -{ - void *address = NULL; - if (!is_remote) - address = hipe_find_emu_address(m, f, arity); - if (!address) { - /* if not found, stub it via the export entry */ - /* no lock needed around erts_export_get_or_make_stub() */ - Export *export_entry = erts_export_get_or_make_stub(m, f, arity); - address = export_entry->addressv[erts_active_code_ix()]; - } - return address; -} - -#if 0 /* XXX: unused */ -BIF_RETTYPE hipe_bifs_get_emu_address_1(BIF_ALIST_1) -{ - struct mfa mfa; - void *address; - - if (!term_to_mfa(BIF_ARG_1, &mfa)) - BIF_ERROR(BIF_P, BADARG); - address = hipe_get_emu_address(mfa.mod, mfa.fun, mfa.ari); - BIF_RET(address_to_term(address, BIF_P)); -} -#endif - BIF_RETTYPE hipe_bifs_set_native_address_3(BIF_ALIST_3) { Eterm *pc; @@ -713,33 +651,6 @@ BIF_RETTYPE hipe_bifs_set_native_address_3(BIF_ALIST_3) BIF_RET(am_false); } -#if 0 /* XXX: unused */ -/* - * hipe_bifs_address_to_fun(Address) - * - Address is the address of the start of a emu function's code - * - returns {Module, Function, Arity} - */ -BIF_RETTYPE hipe_bifs_address_to_fun_1(BIF_ALIST_1) -{ - Eterm *pc; - Eterm *funcinfo; - Eterm *hp; - - pc = term_to_address(BIF_ARG_1); - if (!pc) - BIF_ERROR(BIF_P, BADARG); - funcinfo = find_function_from_pc(pc); - if (!funcinfo) - BIF_RET(am_false); - hp = HAlloc(BIF_P, 4); - hp[0] = make_arityval(3); - hp[1] = funcinfo[0]; - hp[2] = funcinfo[1]; - hp[3] = make_small(funcinfo[2]); - BIF_RET(make_tuple(hp)); -} -#endif - BIF_RETTYPE hipe_bifs_enter_sdesc_1(BIF_ALIST_1) { struct sdesc *sdesc; @@ -948,37 +859,6 @@ BIF_RETTYPE hipe_bifs_primop_address_1(BIF_ALIST_1) BIF_RET(address_to_term(primop->address, BIF_P)); } -#if 0 /* XXX: unused */ -/* - * hipe_bifs_gbif_address(F,A) -> address or false - */ -#define GBIF_LIST(ATOM,ARY,CFUN) extern Eterm gbif_##CFUN(void); -#include "hipe_gbif_list.h" -#undef GBIF_LIST - -BIF_RETTYPE hipe_bifs_gbif_address_2(BIF_ALIST_2) -{ - Uint arity; - void *address; - - if (is_not_atom(BIF_ARG_1) || is_not_small(BIF_ARG_2)) - BIF_RET(am_false); /* error or false, does it matter? */ - arity = signed_val(BIF_ARG_2); - /* XXX: replace with a hash table later */ - do { /* trick to let us use 'break' instead of 'goto' */ -#define GBIF_LIST(ATOM,ARY,CFUN) if (BIF_ARG_1 == ATOM && arity == ARY) { address = CFUN; break; } -#include "hipe_gbif_list.h" -#undef GBIF_LIST - printf("\r\n%s: guard BIF ", __FUNCTION__); - fflush(stdout); - erts_printf("%T", BIF_ARG_1); - printf("/%lu isn't listed in hipe_gbif_list.h\r\n", arity); - BIF_RET(am_false); - } while (0); - BIF_RET(address_to_term(address, BIF_P)); -} -#endif - BIF_RETTYPE hipe_bifs_atom_to_word_1(BIF_ALIST_1) { if (is_not_atom(BIF_ARG_1)) @@ -1022,83 +902,12 @@ BIF_RETTYPE hipe_conv_big_to_float(BIF_ALIST_1) */ void hipe_emulate_fpe(Process* p) { - if (!finite(p->hipe.float_result)) { + if (!isfinite(p->hipe.float_result)) { p->fp_exception = 1; } } #endif -#if 0 /* XXX: unused */ -/* - * At least parts of this should be inlined in native code. - * The rest could be made a primop used by both the emulator and - * native code... - */ -BIF_RETTYPE hipe_bifs_make_fun_3(BIF_ALIST_3) -{ - Eterm free_vars; - Eterm mod; - Eterm *tp; - Uint index; - Uint uniq; - Uint num_free; - Eterm tmp_var; - Uint *tmp_ptr; - unsigned needed; - ErlFunThing *funp; - Eterm *hp; - int i; - - if (is_not_list(BIF_ARG_1) && is_not_nil(BIF_ARG_1)) - BIF_ERROR(BIF_P, BADARG); - free_vars = BIF_ARG_1; - - if (is_not_atom(BIF_ARG_2)) - BIF_ERROR(BIF_P, BADARG); - mod = BIF_ARG_2; - - if (is_not_tuple(BIF_ARG_3) || - (arityval(*tuple_val(BIF_ARG_3)) != 3)) - BIF_ERROR(BIF_P, BADARG); - tp = tuple_val(BIF_ARG_3); - - if (term_to_Uint(tp[1], &index) == 0) - BIF_ERROR(BIF_P, BADARG); - if (term_to_Uint(tp[2], &uniq) == 0) - BIF_ERROR(BIF_P, BADARG); - if (term_to_Uint(tp[3], &num_free) == 0) - BIF_ERROR(BIF_P, BADARG); - - needed = ERL_FUN_SIZE + num_free; - funp = (ErlFunThing *) HAlloc(BIF_P, needed); - hp = funp->env; - - funp->thing_word = HEADER_FUN; - - /* Need a ErlFunEntry *fe - * fe->refc++; - * funp->fe = fe; - */ - - funp->num_free = num_free; - funp->creator = BIF_P->id; - for (i = 0; i < num_free; i++) { - if (is_nil(free_vars)) - BIF_ERROR(BIF_P, BADARG); - tmp_ptr = list_val(free_vars); - tmp_var = CAR(tmp_ptr); - free_vars = CDR(tmp_ptr); - *hp++ = tmp_var; - } - if (is_not_nil(free_vars)) - BIF_ERROR(BIF_P, BADARG); - - funp->next = MSO(BIF_P).funs; - MSO(BIF_P).funs = funp; - - BIF_RET(make_fun(funp)); -} -#endif /* * args: Module, {Uniq, Index, BeamAddress} @@ -1163,22 +972,6 @@ BIF_RETTYPE hipe_bifs_set_native_address_in_fe_2(BIF_ALIST_2) BIF_RET(am_true); } -#if 0 /* XXX: unused */ -BIF_RETTYPE hipe_bifs_make_native_stub_2(BIF_ALIST_2) -{ - void *beamAddress; - Uint beamArity; - void *stubAddress; - - if ((beamAddress = term_to_address(BIF_ARG_1)) == 0 || - is_not_small(BIF_ARG_2) || - (beamArity = unsigned_val(BIF_ARG_2)) >= 256) - BIF_ERROR(BIF_P, BADARG); - stubAddress = hipe_make_native_stub(beamAddress, beamArity); - BIF_RET(address_to_term(stubAddress, BIF_P)); -} -#endif - /* * MFA info hash table: * - maps MFA to native code entry point @@ -1323,16 +1116,6 @@ static inline struct hipe_mfa_info *hipe_mfa_info_table_get_locked(Eterm m, Eter return NULL; } -#if 0 /* XXX: unused */ -void *hipe_mfa_find_na(Eterm m, Eterm f, unsigned int arity) -{ - const struct hipe_mfa_info *p; - - p = hipe_mfa_info_table_get(m, f, arity); - return p ? p->address : NULL; -} -#endif - static struct hipe_mfa_info *hipe_mfa_info_table_put_locked(Eterm m, Eterm f, unsigned int arity) { unsigned long h; @@ -1490,18 +1273,13 @@ void hipe_mfa_save_orig_beam_op(Eterm mod, Eterm fun, unsigned int ari, Eterm *p static void *hipe_make_stub(Eterm m, Eterm f, unsigned int arity, int is_remote) { - void *BEAMAddress; + Export *export_entry; void *StubAddress; -#if 0 - if (is_not_atom(m) || is_not_atom(f) || arity > 255) - return NULL; -#endif - BEAMAddress = hipe_get_emu_address(m, f, arity, is_remote); - StubAddress = hipe_make_native_stub(BEAMAddress, arity); -#if 0 - hipe_mfa_set_na(m, f, arity, StubAddress); -#endif + ASSERT(is_remote); + + export_entry = erts_export_get_or_make_stub(m, f, arity); + StubAddress = hipe_make_native_stub(export_entry, arity); return StubAddress; } diff --git a/erts/emulator/hipe/hipe_bif_list.m4 b/erts/emulator/hipe/hipe_bif_list.m4 index 5f92b6bac4..96a849621f 100644 --- a/erts/emulator/hipe/hipe_bif_list.m4 +++ b/erts/emulator/hipe/hipe_bif_list.m4 @@ -277,7 +277,10 @@ ifelse($1,list_to_binary_1,hipe_wrapper_list_to_binary_1, ifelse($1,iolist_to_binary_1,hipe_wrapper_iolist_to_binary_1, ifelse($1,binary_list_to_bin_1,hipe_wrapper_binary_list_to_bin_1, ifelse($1,list_to_bitstring_1,hipe_wrapper_list_to_bitstring_1, -$1)))))))))))') +ifelse($1,send_2,hipe_wrapper_send_2, +ifelse($1,send_3,hipe_wrapper_send_3, +ifelse($1,ebif_bang_2,hipe_wrapper_ebif_bang_2, +$1))))))))))))))') define(BIF_LIST,`standard_bif_interface_$3(nbif_$4, CFUN($4))') include(TARGET/`erl_bif_list.h') diff --git a/erts/emulator/hipe/hipe_debug.c b/erts/emulator/hipe/hipe_debug.c index 32694a8f97..61406b92af 100644 --- a/erts/emulator/hipe/hipe_debug.c +++ b/erts/emulator/hipe/hipe_debug.c @@ -172,8 +172,10 @@ void hipe_print_pcb(Process *p) printf("P: 0x%0*lx\r\n", 2*(int)sizeof(long), (unsigned long)p); printf("-----------------------------------------------\r\n"); printf("Offset| Name | Value | *Value |\r\n"); +#undef U #define U(n,x) \ printf(" % 4d | %s | 0x%0*lx | |\r\n", (int)offsetof(Process,x), n, 2*(int)sizeof(long), (unsigned long)p->x) +#undef P #define P(n,x) \ printf(" % 4d | %s | 0x%0*lx | 0x%0*lx |\r\n", (int)offsetof(Process,x), n, 2*(int)sizeof(long), (unsigned long)p->x, 2*(int)sizeof(long), p->x ? (unsigned long)*(p->x) : -1UL) @@ -231,7 +233,7 @@ void hipe_print_pcb(Process *p) U("nsp ", hipe.nsp); U("nstack ", hipe.nstack); U("nstend ", hipe.nstend); - U("ncallee ", hipe.ncallee); + U("ncallee ", hipe.u.ncallee); hipe_arch_print_pcb(&p->hipe); #endif /* HIPE */ #undef U diff --git a/erts/emulator/hipe/hipe_gc.c b/erts/emulator/hipe/hipe_gc.c index 86c4068072..b10263f6e2 100644 --- a/erts/emulator/hipe/hipe_gc.c +++ b/erts/emulator/hipe/hipe_gc.c @@ -22,6 +22,9 @@ #ifdef HAVE_CONFIG_H #include "config.h" #endif + +#define ERL_WANT_GC_INTERNALS__ + #include "global.h" #include "erl_gc.h" diff --git a/erts/emulator/hipe/hipe_mkliterals.c b/erts/emulator/hipe/hipe_mkliterals.c index 0e287908b1..ed355ce264 100644 --- a/erts/emulator/hipe/hipe_mkliterals.c +++ b/erts/emulator/hipe/hipe_mkliterals.c @@ -498,8 +498,8 @@ static const struct rts_param rts_params[] = { { 38, "P_ARG4", 1, offsetof(struct process, def_arg_reg[4]) }, { 39, "P_ARG5", 1, offsetof(struct process, def_arg_reg[5]) }, { 40, "P_NSP", 1, offsetof(struct process, hipe.nsp) }, - { 41, "P_NCALLEE", 1, offsetof(struct process, hipe.ncallee) }, - { 42, "P_CLOSURE", 1, offsetof(struct process, hipe.closure) }, + { 41, "P_NCALLEE", 1, offsetof(struct process, hipe.u.ncallee) }, + { 42, "P_CLOSURE", 1, offsetof(struct process, hipe.u.closure) }, { 43, "P_NSP_LIMIT", 1, offsetof(struct process, hipe.nstack) }, { 44, "P_CSP", #if defined(__i386__) || defined(__x86_64__) @@ -524,6 +524,7 @@ static const struct rts_param rts_params[] = { }, { 49, "P_MSG_FIRST", 1, offsetof(struct process, msg.first) }, { 50, "P_MSG_SAVE", 1, offsetof(struct process, msg.save) }, + { 51, "P_CALLEE_EXP", 1, offsetof(struct process, hipe.u.callee_exp) }, }; #define NR_PARAMS ARRAY_SIZE(rts_params) diff --git a/erts/emulator/hipe/hipe_mode_switch.c b/erts/emulator/hipe/hipe_mode_switch.c index 4ddc2790b1..8c73312d45 100644 --- a/erts/emulator/hipe/hipe_mode_switch.c +++ b/erts/emulator/hipe/hipe_mode_switch.c @@ -2,7 +2,7 @@ * %CopyrightBegin% * - * Copyright Ericsson AB 2001-2013. All Rights Reserved. + * Copyright Ericsson AB 2001-2014. 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 @@ -140,7 +140,6 @@ void hipe_check_pcb(Process *p, const char *file, unsigned line) #endif /* HIPE_DEBUG > 0 */ /* ensure that at least nwords words are available on the native stack */ -static void hipe_check_nstack(Process *p, unsigned nwords); #if defined(__sparc__) #include "hipe_sparc_glue.h" @@ -159,7 +158,7 @@ static void hipe_check_nstack(Process *p, unsigned nwords); Uint hipe_beam_pc_return[1]; /* needed in hipe_debug.c */ Uint hipe_beam_pc_throw[1]; /* needed in hipe_debug.c */ Uint hipe_beam_pc_resume[1]; /* needed by hipe_set_timeout() */ -static Eterm hipe_beam_catch_throw; +Eterm hipe_beam_catch_throw; void hipe_mode_switch_init(void) { @@ -185,48 +184,31 @@ void hipe_set_call_trap(Uint *bfun, void *nfun, int is_closure) bfun[-4] = (Uint)nfun; } -void hipe_reserve_beam_trap_frame(Process *p, Eterm reg[], unsigned arity) -{ - /* ensure that at least 2 words are available on the BEAM stack */ - if ((p->stop - 2) < p->htop) { - DPRINTF("calling gc to reserve BEAM stack size"); - p->fcalls -= erts_garbage_collect(p, 2, reg, arity); - ASSERT(!((p->stop - 2) < p->htop)); - } - p->stop -= 2; - p->stop[0] = NIL; - p->stop[1] = NIL; -} - static __inline__ void hipe_push_beam_trap_frame(Process *p, Eterm reg[], unsigned arity) { - if (p->flags & F_DISABLE_GC) { + if (&p->stop[1] < p->hend && p->stop[1] == hipe_beam_catch_throw) { /* Trap frame already reserved */ - ASSERT(p->stop[0] == NIL && p->stop[1] == NIL); + ASSERT(p->stop[0] == NIL); } else { + ASSERT(!(p->flags & F_DISABLE_GC)); if ((p->stop - 2) < p->htop) { DPRINTF("calling gc to increase BEAM stack size"); p->fcalls -= erts_garbage_collect(p, 2, reg, arity); ASSERT(!((p->stop - 2) < p->htop)); } p->stop -= 2; + p->stop[1] = hipe_beam_catch_throw; } - p->stop[1] = hipe_beam_catch_throw; p->stop[0] = make_cp(p->cp); ++p->catches; p->cp = hipe_beam_pc_return; } -void hipe_unreserve_beam_trap_frame(Process *p) -{ - ASSERT(p->stop[0] == NIL && p->stop[1] == NIL); - p->stop += 2; -} - static __inline__ void hipe_pop_beam_trap_frame(Process *p) { + ASSERT(p->stop[1] == hipe_beam_catch_throw); p->cp = cp_val(p->stop[0]); --p->catches; p->stop += 2; @@ -257,14 +239,14 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) /* BEAM calls a native code function */ unsigned arity = cmd >> 8; - /* p->hipe.ncallee set in beam_emu */ + /* p->hipe.u.ncallee set in beam_emu */ if (p->cp == hipe_beam_pc_return) { /* Native called BEAM, which now tailcalls native. */ hipe_pop_beam_trap_frame(p); result = hipe_tailcall_to_native(p, arity, reg); break; } - DPRINTF("calling %#lx/%u", (long)p->hipe.ncallee, arity); + DPRINTF("calling %#lx/%u", (long)p->hipe.u.ncallee, arity); result = hipe_call_to_native(p, arity, reg); break; } @@ -282,18 +264,18 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) arity -= funp->num_free; /* arity == #formals */ reg[arity] = fun; ++arity; /* correct for having added the closure */ - /* HIPE_ASSERT(p->hipe.ncallee == (void(*)(void))funp->native_address); */ + /* HIPE_ASSERT(p->hipe.u.ncallee == (void(*)(void))funp->native_address); */ /* just like a normal call from now on */ - /* p->hipe.ncallee set in beam_emu */ + /* p->hipe.u.ncallee set in beam_emu */ if (p->cp == hipe_beam_pc_return) { /* Native called BEAM, which now tailcalls native. */ hipe_pop_beam_trap_frame(p); result = hipe_tailcall_to_native(p, arity, reg); break; } - DPRINTF("calling %#lx/%u", (long)p->hipe.ncallee, arity); + DPRINTF("calling %#lx/%u", (long)p->hipe.u.ncallee, arity); result = hipe_call_to_native(p, arity, reg); break; } @@ -396,13 +378,13 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) if (is_recursive) hipe_push_beam_trap_frame(p, reg, p->arity); - result = HIPE_MODE_SWITCH_RES_CALL; + result = HIPE_MODE_SWITCH_RES_CALL_BEAM; break; } - case HIPE_MODE_SWITCH_RES_CALL: { + case HIPE_MODE_SWITCH_RES_CALL_EXPORTED: { /* Native code calls or tailcalls BEAM. * - * p->i is the callee's BEAM code + * p->hipe.u.callee_exp is the callee's export entry * p->arity is the callee's arity * p->def_arg_reg[] contains the register parameters * p->hipe.nsp[] contains the stacked parameters @@ -422,15 +404,15 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) * F(A1, ..., AN, FV1, ..., FVM, Closure) * (Where Ai is argument i and FVj is free variable j) * - * p->hipe.closure contains the closure + * p->hipe.u.closure contains the closure * p->def_arg_reg[] contains the register parameters * p->hipe.nsp[] contains the stacked parameters */ ErlFunThing *closure; unsigned num_free, arity, i, is_recursive; - HIPE_ASSERT(is_fun(p->hipe.closure)); - closure = (ErlFunThing*)fun_val(p->hipe.closure); + HIPE_ASSERT(is_fun(p->hipe.u.closure)); + closure = (ErlFunThing*)fun_val(p->hipe.u.closure); num_free = closure->num_free; arity = closure->fe->arity; @@ -460,10 +442,10 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) p->i = closure->fe->address; /* Change result code to the faster plain CALL type. */ - result = HIPE_MODE_SWITCH_RES_CALL; + result = HIPE_MODE_SWITCH_RES_CALL_BEAM; } /* Append the closure as the last parameter. Don't increment arity. */ - reg[arity] = p->hipe.closure; + reg[arity] = p->hipe.u.closure; if (is_recursive) { /* BEAM called native, which now calls BEAM. @@ -541,7 +523,7 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) } } HIPE_CHECK_PCB(p); - result = HIPE_MODE_SWITCH_RES_CALL; + result = HIPE_MODE_SWITCH_RES_CALL_BEAM; p->def_arg_reg[3] = result; return p; } @@ -569,7 +551,7 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[]) address = hipe_get_remote_na(mfa[0], mfa[1], arity); if (!address) goto do_apply_fail; - p->hipe.ncallee = (void(*)(void)) address; + p->hipe.u.ncallee = (void(*)(void)) address; result = hipe_tailcall_to_native(p, arity, reg); goto do_return_from_native; do_apply_fail: @@ -599,7 +581,6 @@ static unsigned hipe_next_nstack_size(unsigned size) } #if 0 && defined(HIPE_NSTACK_GROWS_UP) -#define hipe_nstack_avail(p) ((p)->hipe.nstend - (p)->hipe.nsp) void hipe_inc_nstack(Process *p) { Eterm *old_nstack = p->hipe.nstack; @@ -623,7 +604,6 @@ void hipe_inc_nstack(Process *p) #endif #if defined(HIPE_NSTACK_GROWS_DOWN) -#define hipe_nstack_avail(p) ((unsigned)((p)->hipe.nsp - (p)->hipe.nstack)) void hipe_inc_nstack(Process *p) { unsigned old_size = p->hipe.nstend - p->hipe.nstack; @@ -655,12 +635,6 @@ void hipe_empty_nstack(Process *p) p->hipe.nstend = NULL; } -static void hipe_check_nstack(Process *p, unsigned nwords) -{ - while (hipe_nstack_avail(p) < nwords) - hipe_inc_nstack(p); -} - void hipe_set_closure_stub(ErlFunEntry *fe, unsigned num_free) { unsigned arity; diff --git a/erts/emulator/hipe/hipe_mode_switch.h b/erts/emulator/hipe/hipe_mode_switch.h index 06721e3c04..b8de12fcbb 100644 --- a/erts/emulator/hipe/hipe_mode_switch.h +++ b/erts/emulator/hipe/hipe_mode_switch.h @@ -31,7 +31,7 @@ /* result codes for beam_emu <- hipe_mode_switch() return */ #define HIPE_MODE_SWITCH_RES_RETURN 4 -#define HIPE_MODE_SWITCH_RES_CALL 5 +#define HIPE_MODE_SWITCH_RES_CALL_EXPORTED 5 #define HIPE_MODE_SWITCH_RES_THROW 6 /* additional result codes for hipe_mode_switch() <- native return */ @@ -45,6 +45,8 @@ #define HIPE_MODE_SWITCH_RES_APPLY 13 /* mode_switch <- native */ +#define HIPE_MODE_SWITCH_RES_CALL_BEAM 14 + #ifndef ASM #include "error.h" @@ -59,13 +61,58 @@ void hipe_empty_nstack(Process *p); void hipe_set_closure_stub(ErlFunEntry *fe, unsigned num_free); Eterm hipe_build_stacktrace(Process *p, struct StackTrace *s); -void hipe_reserve_beam_trap_frame(Process*, Eterm reg[], unsigned arity); -void hipe_unreserve_beam_trap_frame(Process*); +ERTS_GLB_INLINE void hipe_reserve_beam_trap_frame(Process*, Eterm reg[], unsigned arity); +ERTS_GLB_INLINE void hipe_unreserve_beam_trap_frame(Process*); extern Uint hipe_beam_pc_return[]; extern Uint hipe_beam_pc_throw[]; extern Uint hipe_beam_pc_resume[]; +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +#include "erl_gc.h" +#include "hipe_stack.h" + +#if defined(__sparc__) +#include "hipe_sparc_glue.h" +#elif defined(__i386__) +#include "hipe_x86_glue.h" +#elif defined(__x86_64__) +#include "hipe_amd64_glue.h" +#elif defined(__powerpc__) || defined(__ppc__) || defined(__powerpc64__) +#include "hipe_ppc_glue.h" +#elif defined(__arm__) +#include "hipe_arm_glue.h" +#endif + +extern Eterm hipe_beam_catch_throw; + +ERTS_GLB_INLINE void hipe_reserve_beam_trap_frame(Process *p, Eterm reg[], unsigned arity) +{ + if (!hipe_bifcall_from_native_is_recursive(p)) + return; + + /* ensure that at least 2 words are available on the BEAM stack */ + if ((p->stop - 2) < p->htop) { + p->fcalls -= erts_garbage_collect(p, 2, reg, arity); + ASSERT(!((p->stop - 2) < p->htop)); + } + p->stop -= 2; + p->stop[0] = NIL; + p->stop[1] = hipe_beam_catch_throw; +} + +ERTS_GLB_INLINE void hipe_unreserve_beam_trap_frame(Process *p) +{ + if (!hipe_bifcall_from_native_is_recursive(p)) + return; + + ASSERT(p->stop[0] == NIL && p->stop[1] == hipe_beam_catch_throw); + p->stop += 2; +} + +#endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */ + #endif /* ASM */ #endif /* HIPE_MODE_SWITCH_H */ diff --git a/erts/emulator/hipe/hipe_native_bif.c b/erts/emulator/hipe/hipe_native_bif.c index 7d343dd91e..7e8632b50d 100644 --- a/erts/emulator/hipe/hipe_native_bif.c +++ b/erts/emulator/hipe/hipe_native_bif.c @@ -330,8 +330,6 @@ char *hipe_bs_allocate(int len) Binary *bptr; bptr = erts_bin_nrml_alloc(len); - bptr->flags = 0; - bptr->orig_size = len; erts_smp_atomic_init_nob(&bptr->refc, 1); return bptr->orig_bytes; } @@ -341,7 +339,6 @@ Binary *hipe_bs_reallocate(Binary* oldbptr, int newsize) Binary *bptr; bptr = erts_bin_realloc(oldbptr, newsize); - bptr->orig_size = newsize; return bptr; } diff --git a/erts/emulator/hipe/hipe_ppc.c b/erts/emulator/hipe/hipe_ppc.c index 2d8fd61e1e..4dc26cdbc8 100644 --- a/erts/emulator/hipe/hipe_ppc.c +++ b/erts/emulator/hipe/hipe_ppc.c @@ -285,7 +285,7 @@ int hipe_patch_insn(void *address, Uint64 value, Eterm type) } } -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { unsigned int *code; @@ -294,16 +294,16 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) code = alloc_stub(7); - /* addis r12,0,beamAddress@highest */ - code[0] = 0x3d800000 | (((unsigned long)beamAddress >> 48) & 0xffff); - /* ori r12,r12,beamAddress@higher */ - code[1] = 0x618c0000 | (((unsigned long)beamAddress >> 32) & 0xffff); + /* addis r12,0,callee_exp@highest */ + code[0] = 0x3d800000 | (((unsigned long)callee_exp >> 48) & 0xffff); + /* ori r12,r12,callee_exp@higher */ + code[1] = 0x618c0000 | (((unsigned long)callee_exp >> 32) & 0xffff); /* sldi r12,r12,32 (rldicr r12,r12,32,31) */ code[2] = 0x798c07c6; - /* oris r12,r12,beamAddress@h */ - code[3] = 0x658c0000 | (((unsigned long)beamAddress >> 16) & 0xffff); - /* ori r12,r12,beamAddress@l */ - code[4] = 0x618c0000 | ((unsigned long)beamAddress & 0xffff); + /* oris r12,r12,callee_exp@h */ + code[3] = 0x658c0000 | (((unsigned long)callee_exp >> 16) & 0xffff); + /* ori r12,r12,callee_exp@l */ + code[4] = 0x618c0000 | ((unsigned long)callee_exp & 0xffff); /* addi r0,0,beamArity */ code[5] = 0x38000000 | (beamArity & 0x7FFF); /* ba nbif_callemu */ @@ -355,18 +355,16 @@ int hipe_patch_insn(void *address, Uint32 value, Eterm type) return 0; } -/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2() - and hipe_bif0.c:hipe_make_stub() */ -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { unsigned int *code; /* * Native code calls BEAM via a stub looking as follows: * - * addi r12,0,beamAddress@l + * addi r12,0,callee_exp@l * addi r0,0,beamArity - * addis r12,r12,beamAddress@ha + * addis r12,r12,callee_exp@ha * ba nbif_callemu * * I'm using r0 and r12 since the standard SVR4 ABI allows @@ -384,12 +382,12 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) code = alloc_stub(4); - /* addi r12,0,beamAddress@l */ - code[0] = 0x39800000 | ((unsigned long)beamAddress & 0xFFFF); + /* addi r12,0,callee_exp@l */ + code[0] = 0x39800000 | ((unsigned long)callee_exp & 0xFFFF); /* addi r0,0,beamArity */ code[1] = 0x38000000 | (beamArity & 0x7FFF); - /* addis r12,r12,beamAddress@ha */ - code[2] = 0x3D8C0000 | at_ha((unsigned long)beamAddress); + /* addis r12,r12,callee_exp@ha */ + code[2] = 0x3D8C0000 | at_ha((unsigned long)callee_exp); /* ba nbif_callemu */ code[3] = 0x48000002 | (unsigned long)&nbif_callemu; diff --git a/erts/emulator/hipe/hipe_ppc_asm.m4 b/erts/emulator/hipe/hipe_ppc_asm.m4 index 343402f9f0..4a1caa1543 100644 --- a/erts/emulator/hipe/hipe_ppc_asm.m4 +++ b/erts/emulator/hipe/hipe_ppc_asm.m4 @@ -23,6 +23,22 @@ changecom(`/*', `*/')dnl #define HIPE_PPC_ASM_H' /* + * Tunables. + */ +define(LEAF_WORDS,16)dnl number of stack words for leaf functions +define(NR_ARG_REGS,4)dnl admissible values are 0 to 6, inclusive + +`#define PPC_LEAF_WORDS 'LEAF_WORDS +`#define PPC_NR_ARG_REGS 'NR_ARG_REGS +`#define NR_ARG_REGS 'NR_ARG_REGS + + +`#ifdef ASM' +/* + * Only assembler stuff from here on (when included from *.S) + */ + +/* * Handle 32 vs 64-bit. */ ifelse(ARCH,ppc64,` @@ -53,13 +69,6 @@ define(WSIZE,4)dnl `#define STORE 'STORE `#define CMPI 'CMPI -/* - * Tunables. - */ -define(LEAF_WORDS,16)dnl number of stack words for leaf functions -define(NR_ARG_REGS,4)dnl admissible values are 0 to 6, inclusive - -`#define PPC_LEAF_WORDS 'LEAF_WORDS /* * Workarounds for Darwin. @@ -193,8 +202,6 @@ NAME: \ /* * Argument (parameter) registers. */ -`#define PPC_NR_ARG_REGS 'NR_ARG_REGS -`#define NR_ARG_REGS 'NR_ARG_REGS define(defarg,`define(ARG$1,`$2')dnl #`define ARG'$1 $2' @@ -309,4 +316,6 @@ define(QUICK_CALL_RET,`NBIF_POP_N(eval(RET_POP($2)))b $1')dnl `/* #define QUICK_CALL_RET_F_3 'QUICK_CALL_RET(F,3)` */' `/* #define QUICK_CALL_RET_F_5 'QUICK_CALL_RET(F,5)` */' +`#endif /* ASM */' + `#endif /* HIPE_PPC_ASM_H */' diff --git a/erts/emulator/hipe/hipe_ppc_bifs.m4 b/erts/emulator/hipe/hipe_ppc_bifs.m4 index 7cc2b5c7b6..f53b79b52e 100644 --- a/erts/emulator/hipe/hipe_ppc_bifs.m4 +++ b/erts/emulator/hipe/hipe_ppc_bifs.m4 @@ -19,6 +19,7 @@ changecom(`/*', `*/')dnl */ +#`define ASM' include(`hipe/hipe_ppc_asm.m4') #`include' "config.h" #`include' "hipe_literals.h" diff --git a/erts/emulator/hipe/hipe_ppc_glue.S b/erts/emulator/hipe/hipe_ppc_glue.S index 6f0217c738..c48fb150af 100644 --- a/erts/emulator/hipe/hipe_ppc_glue.S +++ b/erts/emulator/hipe/hipe_ppc_glue.S @@ -17,10 +17,9 @@ * %CopyrightEnd% */ - +#define ASM #include "hipe_ppc_asm.h" #include "hipe_literals.h" -#define ASM #include "hipe_mode_switch.h" .text @@ -296,7 +295,7 @@ CSYM(hipe_ppc_throw_to_native): * which should look as follows: * * stub for f/N: - * <set r12 to f's BEAM code address> + * <set r12 to f's export entry address> * <set r0 to N> * b nbif_callemu * @@ -312,10 +311,10 @@ CSYM(hipe_ppc_throw_to_native): */ GLOBAL(ASYM(nbif_callemu)) ASYM(nbif_callemu): - STORE r12, P_BEAM_IP(P) + STORE r12, P_CALLEE_EXP(P) STORE r0, P_ARITY(P) STORE_ARG_REGS - li r3, HIPE_MODE_SWITCH_RES_CALL + li r3, HIPE_MODE_SWITCH_RES_CALL_EXPORTED b .suspend_exit /* diff --git a/erts/emulator/hipe/hipe_process.h b/erts/emulator/hipe/hipe_process.h index 4ee99d78a2..86655ad42c 100644 --- a/erts/emulator/hipe/hipe_process.h +++ b/erts/emulator/hipe/hipe_process.h @@ -23,14 +23,17 @@ #define HIPE_PROCESS_H #include "erl_alloc.h" +#include "export.h" struct hipe_process_state { Eterm *nsp; /* Native stack pointer. */ Eterm *nstack; /* Native stack block start. */ Eterm *nstend; /* Native stack block end (start+size). */ - /* XXX: ncallee and closure could share space in a union */ - void (*ncallee)(void); /* Native code callee (label) to invoke. */ - Eterm closure; /* Used to pass a closure from native code. */ + union { + void (*ncallee)(void); /* Native code callee (label) to invoke. */ + Eterm closure; /* Used to pass a closure from native code. */ + Export* callee_exp; /* Used to pass export entry from native code */ + }u; Eterm *nstgraylim; /* Gray/white stack boundary. */ Eterm *nstblacklim; /* Black/gray stack boundary. Must exist if graylim exists. Ignored if no graylim. */ diff --git a/erts/emulator/hipe/hipe_risc_glue.h b/erts/emulator/hipe/hipe_risc_glue.h index cc2671c016..dbb7086dae 100644 --- a/erts/emulator/hipe/hipe_risc_glue.h +++ b/erts/emulator/hipe/hipe_risc_glue.h @@ -214,6 +214,14 @@ hipe_trap_from_native_is_recursive(Process *p) return 0; } +/* Native called BIF. Is it a recursive call? + i.e should we return back to native when BIF is done? */ +static __inline__ int +hipe_bifcall_from_native_is_recursive(Process *p) +{ + return (p->hipe.nra != (void(*)(void))&nbif_return); +} + /* Native makes a call which needs to unload the parameters. This differs from hipe_call_from_native_is_recursive() in diff --git a/erts/emulator/hipe/hipe_risc_stack.c b/erts/emulator/hipe/hipe_risc_stack.c index 1183856c7e..bea3a0fecd 100644 --- a/erts/emulator/hipe/hipe_risc_stack.c +++ b/erts/emulator/hipe/hipe_risc_stack.c @@ -226,7 +226,7 @@ void (*hipe_handle_stack_trap(Process *p))(void) * The native stack MUST contain a stack frame as it appears on * entry to a function (actuals, caller's frame, caller's return address). * p->hipe.narity MUST contain the arity (number of actuals). - * On exit, p->hipe.ncallee is set to the handler's PC and p->hipe.nsp + * On exit, p->hipe.u.ncallee is set to the handler's PC and p->hipe.nsp * is set to its SP (low address of its stack frame). */ void hipe_find_handler(Process *p) @@ -254,7 +254,7 @@ void hipe_find_handler(Process *p) if ((exnra = sdesc_exnra(sdesc)) != 0 && (p->catches >= 0 || exnra == (unsigned long)&nbif_fail)) { - p->hipe.ncallee = (void(*)(void)) exnra; + p->hipe.u.ncallee = (void(*)(void)) exnra; p->hipe.nsp = nsp; p->hipe.narity = 0; /* update the gray/white boundary if we threw past it */ diff --git a/erts/emulator/hipe/hipe_sparc.c b/erts/emulator/hipe/hipe_sparc.c index 49d4da7bab..2052aa8498 100644 --- a/erts/emulator/hipe/hipe_sparc.c +++ b/erts/emulator/hipe/hipe_sparc.c @@ -204,9 +204,7 @@ void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process * return alloc_code(nrbytes); } -/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2() - and hipe_bif0.c:hipe_make_stub() */ -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { unsigned int *code; unsigned int callEmuOffset; @@ -215,11 +213,11 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) code = alloc_code(5*sizeof(int)); /* sethi %hi(Address), %i4 */ - code[0] = 0x39000000 | (((unsigned int)beamAddress >> 10) & 0x3FFFFF); + code[0] = 0x39000000 | (((unsigned int)callee_exp >> 10) & 0x3FFFFF); /* or %g0, %o7, %i3 ! mov %o7, %i3 */ code[1] = 0xB610000F; /* or %i4, %lo(Address), %i4 */ - code[2] = 0xB8172000 | ((unsigned int)beamAddress & 0x3FF); + code[2] = 0xB8172000 | ((unsigned int)callee_exp & 0x3FF); /* call callemu */ callEmuOffset = (char*)nbif_callemu - (char*)&code[3]; code[3] = (1 << 30) | ((callEmuOffset >> 2) & 0x3FFFFFFF); diff --git a/erts/emulator/hipe/hipe_sparc_asm.m4 b/erts/emulator/hipe/hipe_sparc_asm.m4 index 227d10ed80..c3c3bcb74a 100644 --- a/erts/emulator/hipe/hipe_sparc_asm.m4 +++ b/erts/emulator/hipe/hipe_sparc_asm.m4 @@ -29,6 +29,14 @@ define(LEAF_WORDS,16)dnl number of stack words for leaf functions define(NR_ARG_REGS,4)dnl admissible values are 0 to 6, inclusive `#define SPARC_LEAF_WORDS 'LEAF_WORDS +`#define SPARC_NR_ARG_REGS 'NR_ARG_REGS +`#define NR_ARG_REGS 'NR_ARG_REGS + + +`#ifdef ASM' +/* + * Only assembler stuff from here on (when included from *.S) + */ /* * Reserved registers. @@ -80,9 +88,6 @@ define(NR_ARG_REGS,4)dnl admissible values are 0 to 6, inclusive /* * Argument (parameter) registers. */ -`#define SPARC_NR_ARG_REGS 'NR_ARG_REGS -`#define NR_ARG_REGS 'NR_ARG_REGS - define(defarg,`define(ARG$1,`$2')dnl #`define ARG'$1 $2' )dnl @@ -210,4 +215,6 @@ define(QUICK_CALL_RET,`ba $1; NBIF_POP_N(eval(RET_POP($2)))')dnl `/* #define QUICK_CALL_RET_F_3 'QUICK_CALL_RET(F,3)` */' `/* #define QUICK_CALL_RET_F_5 'QUICK_CALL_RET(F,5)` */' +`#endif /* ASM */' + `#endif /* HIPE_SPARC_ASM_H */' diff --git a/erts/emulator/hipe/hipe_sparc_bifs.m4 b/erts/emulator/hipe/hipe_sparc_bifs.m4 index ca5af45d58..2bfe3a4646 100644 --- a/erts/emulator/hipe/hipe_sparc_bifs.m4 +++ b/erts/emulator/hipe/hipe_sparc_bifs.m4 @@ -19,6 +19,7 @@ changecom(`/*', `*/')dnl */ +#`define ASM' include(`hipe/hipe_sparc_asm.m4') #`include' "config.h" #`include' "hipe_literals.h" diff --git a/erts/emulator/hipe/hipe_sparc_glue.S b/erts/emulator/hipe/hipe_sparc_glue.S index 44bdf1bc7e..6c8c841194 100644 --- a/erts/emulator/hipe/hipe_sparc_glue.S +++ b/erts/emulator/hipe/hipe_sparc_glue.S @@ -18,10 +18,9 @@ * %CopyrightEnd% */ - +#define ASM #include "hipe_sparc_asm.h" #include "hipe_literals.h" -#define ASM #include "hipe_mode_switch.h" .section ".text" @@ -155,9 +154,9 @@ hipe_sparc_throw_to_native: * which should look as follows: * * stub for f/N: - * sethi %hi(f's BEAM code address), TEMP_ARG0 + * sethi %hi(f's export entry address), TEMP_ARG0 * mov RA, TEMP_RA ! because the call below clobbers RA (%o7) - * or TEMP_ARG0, %lo(f's BEAM code address), TEMP_ARG0 + * or TEMP_ARG0, %lo(f's export entry address), TEMP_ARG0 * call nbif_callemu ! clobbers RA! * mov N, TEMP_ARG1 ! delay slot: TEMP_ARG1 := ARITY * @@ -165,12 +164,12 @@ hipe_sparc_throw_to_native: */ .global nbif_callemu nbif_callemu: - st TEMP_ARG0, [P+P_BEAM_IP] + st TEMP_ARG0, [P+P_CALLEE_EXP] st TEMP_ARG1, [P+P_ARITY] st TEMP_RA, [P+P_NRA] STORE_ARG_REGS ba .flush_exit - mov HIPE_MODE_SWITCH_RES_CALL, %o0 + mov HIPE_MODE_SWITCH_RES_CALL_EXPORTED, %o0 /* * nbif_apply diff --git a/erts/emulator/hipe/hipe_stack.h b/erts/emulator/hipe/hipe_stack.h index 66f9f04c73..4cfdb54dd8 100644 --- a/erts/emulator/hipe/hipe_stack.h +++ b/erts/emulator/hipe/hipe_stack.h @@ -108,12 +108,23 @@ extern int hipe_fill_stacktrace(Process*, int, Eterm**); #if 0 && defined(HIPE_NSTACK_GROWS_UP) #define hipe_nstack_start(p) ((p)->hipe.nstack) #define hipe_nstack_used(p) ((p)->hipe.nsp - (p)->hipe.nstack) +#define hipe_nstack_avail(p) ((p)->hipe.nstend - (p)->hipe.nsp) #endif #if defined(HIPE_NSTACK_GROWS_DOWN) #define hipe_nstack_start(p) ((p)->hipe.nsp) #define hipe_nstack_used(p) ((p)->hipe.nstend - (p)->hipe.nsp) +#define hipe_nstack_avail(p) ((unsigned)((p)->hipe.nsp - (p)->hipe.nstack)) #endif +/* ensure that at least nwords words are available on the native stack */ +static __inline__ void hipe_check_nstack(Process *p, unsigned nwords) +{ + extern void hipe_inc_nstack(Process *p); + + while (hipe_nstack_avail(p) < nwords) + hipe_inc_nstack(p); +} + /* * GC support procedures */ diff --git a/erts/emulator/hipe/hipe_x86.c b/erts/emulator/hipe/hipe_x86.c index 327c74e9aa..314f6b597c 100644 --- a/erts/emulator/hipe/hipe_x86.c +++ b/erts/emulator/hipe/hipe_x86.c @@ -182,18 +182,16 @@ void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process * return alloc_code(nrbytes); } -/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2() - and hipe_bif0.c:hipe_make_stub() */ -void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) +void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity) { /* * This creates a native code stub with the following contents: * - * movl $Address, P_BEAM_IP(%ebp) + * movl $Address, P_CALLEE_EXP(%ebp) * movb $Arity, P_ARITY(%ebp) * jmp callemu * - * The stub has variable size, depending on whether the P_BEAM_IP + * The stub has variable size, depending on whether the P_CALLEE_EXP * and P_ARITY offsets fit in 8-bit signed displacements or not. * The rel32 offset in the final jmp depends on its actual location, * which also depends on the size of the previous instructions. @@ -206,28 +204,28 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity) codeSize = /* 16, 19, or 22 bytes */ 16 + /* 16 when both offsets are 8-bit */ - (P_BEAM_IP >= 128 ? 3 : 0) + + (P_CALLEE_EXP >= 128 ? 3 : 0) + (P_ARITY >= 128 ? 3 : 0); codep = code = alloc_code(codeSize); - /* movl $beamAddress, P_BEAM_IP(%ebp); 3 or 6 bytes, plus 4 */ + /* movl $beamAddress, P_CALLEE_EXP(%ebp); 3 or 6 bytes, plus 4 */ codep[0] = 0xc7; -#if P_BEAM_IP >= 128 +#if P_CALLEE_EXP >= 128 codep[1] = 0x85; /* disp32[EBP] */ - codep[2] = P_BEAM_IP & 0xFF; - codep[3] = (P_BEAM_IP >> 8) & 0xFF; - codep[4] = (P_BEAM_IP >> 16) & 0xFF; - codep[5] = (P_BEAM_IP >> 24) & 0xFF; + codep[2] = P_CALLEE_EXP & 0xFF; + codep[3] = (P_CALLEE_EXP >> 8) & 0xFF; + codep[4] = (P_CALLEE_EXP >> 16) & 0xFF; + codep[5] = (P_CALLEE_EXP >> 24) & 0xFF; codep += 6; #else codep[1] = 0x45; /* disp8[EBP] */ - codep[2] = P_BEAM_IP; + codep[2] = P_CALLEE_EXP; codep += 3; #endif - codep[0] = ((unsigned int)beamAddress) & 0xFF; - codep[1] = ((unsigned int)beamAddress >> 8) & 0xFF; - codep[2] = ((unsigned int)beamAddress >> 16) & 0xFF; - codep[3] = ((unsigned int)beamAddress >> 24) & 0xFF; + codep[0] = ((unsigned int)callee_exp) & 0xFF; + codep[1] = ((unsigned int)callee_exp >> 8) & 0xFF; + codep[2] = ((unsigned int)callee_exp >> 16) & 0xFF; + codep[3] = ((unsigned int)callee_exp >> 24) & 0xFF; codep += 4; /* movb $beamArity, P_ARITY(%ebp); 3 or 6 bytes */ diff --git a/erts/emulator/hipe/hipe_x86_asm.m4 b/erts/emulator/hipe/hipe_x86_asm.m4 index 020ccf8d4b..39c5cb1044 100644 --- a/erts/emulator/hipe/hipe_x86_asm.m4 +++ b/erts/emulator/hipe/hipe_x86_asm.m4 @@ -33,6 +33,18 @@ define(SIMULATE_NSP,0)dnl change to 1 to simulate call/ret insns `#define X86_LEAF_WORDS 'LEAF_WORDS `#define LEAF_WORDS 'LEAF_WORDS +`#define X86_NR_ARG_REGS 'NR_ARG_REGS +`#define NR_ARG_REGS 'NR_ARG_REGS + +`#define X86_HP_IN_ESI 'HP_IN_ESI +`#define X86_SIMULATE_NSP 'SIMULATE_NSP + + +`#ifdef ASM' +/* + * Only assembler stuff from here on (when included from *.S) + */ + /* * Workarounds for Darwin. */ @@ -60,7 +72,6 @@ ifelse(OPSYS,darwin,`` */ `#define P %ebp' -`#define X86_HP_IN_ESI 'HP_IN_ESI `#if X86_HP_IN_ESI #define SAVE_HP movl %esi, P_HP(P) #define RESTORE_HP movl P_HP(P), %esi @@ -73,7 +84,6 @@ ifelse(OPSYS,darwin,`` #define SAVE_CSP movl %esp, P_CSP(P) #define RESTORE_CSP movl P_CSP(P), %esp' -`#define X86_SIMULATE_NSP 'SIMULATE_NSP /* * Context switching macros. @@ -100,12 +110,10 @@ ifelse(OPSYS,darwin,`` SAVE_CACHED_STATE; \ SWITCH_ERLANG_TO_C_QUICK' + /* * Argument (parameter) registers. */ -`#define X86_NR_ARG_REGS 'NR_ARG_REGS -`#define NR_ARG_REGS 'NR_ARG_REGS - ifelse(eval(NR_ARG_REGS >= 1),0,, ``#define ARG0 %eax '')dnl @@ -282,4 +290,6 @@ define(LOAD_CALLER_SAVE,`LAR_N(eval(NR_CALLER_SAVE-1))')dnl `#define STORE_CALLER_SAVE 'STORE_CALLER_SAVE `#define LOAD_CALLER_SAVE 'LOAD_CALLER_SAVE +`#endif /* ASM */' + `#endif /* HIPE_X86_ASM_H */' diff --git a/erts/emulator/hipe/hipe_x86_bifs.m4 b/erts/emulator/hipe/hipe_x86_bifs.m4 index dd6980f555..a0f16efa33 100644 --- a/erts/emulator/hipe/hipe_x86_bifs.m4 +++ b/erts/emulator/hipe/hipe_x86_bifs.m4 @@ -19,6 +19,7 @@ changecom(`/*', `*/')dnl */ +#`define ASM' include(`hipe/hipe_x86_asm.m4') #`include' "config.h" #`include' "hipe_literals.h" diff --git a/erts/emulator/hipe/hipe_x86_glue.S b/erts/emulator/hipe/hipe_x86_glue.S index 88b86f4de7..9d38eaaafd 100644 --- a/erts/emulator/hipe/hipe_x86_glue.S +++ b/erts/emulator/hipe/hipe_x86_glue.S @@ -18,10 +18,9 @@ * %CopyrightEnd% */ - +#define ASM #include "hipe_x86_asm.h" #include "hipe_literals.h" -#define ASM #include "hipe_mode_switch.h" /* @@ -104,7 +103,7 @@ ASYM(nbif_return): * stub (hipe_x86_loader.erl) which should look as follows: * * stub for f/N: - * movl $<f's BEAM code address>, P_BEAM_IP(P) + * movl $<f's export entry address>, P_CALLEE_EXP(P) * movb $<N>, P_ARITY(P) * jmp nbif_callemu * @@ -114,7 +113,7 @@ ASYM(nbif_return): GLOBAL(ASYM(nbif_callemu)) ASYM(nbif_callemu): STORE_ARG_REGS - movl $HIPE_MODE_SWITCH_RES_CALL, %eax + movl $HIPE_MODE_SWITCH_RES_CALL_EXPORTED, %eax jmp .suspend_exit /* diff --git a/erts/emulator/hipe/hipe_x86_glue.h b/erts/emulator/hipe/hipe_x86_glue.h index 63ad250d60..4b6e495b9a 100644 --- a/erts/emulator/hipe/hipe_x86_glue.h +++ b/erts/emulator/hipe/hipe_x86_glue.h @@ -207,6 +207,14 @@ hipe_trap_from_native_is_recursive(Process *p) return 0; } +/* Native called BIF. Is it a recursive call? + i.e should we return back to native when BIF is done? */ +static __inline__ int +hipe_bifcall_from_native_is_recursive(Process *p) +{ + return (*p->hipe.nsp != (Eterm)nbif_return); +} + /* Native makes a call which needs to unload the parameters. This differs from hipe_call_from_native_is_recursive() in diff --git a/erts/emulator/hipe/hipe_x86_stack.c b/erts/emulator/hipe/hipe_x86_stack.c index 9ad3fa9d31..7f1c2f7d41 100644 --- a/erts/emulator/hipe/hipe_x86_stack.c +++ b/erts/emulator/hipe/hipe_x86_stack.c @@ -209,7 +209,7 @@ void (*hipe_handle_stack_trap(Process *p))(void) * The native stack MUST contain a stack frame as it appears on * entry to a function (return address, actuals, caller's frame). * p->hipe.narity MUST contain the arity (number of actuals). - * On exit, p->hipe.ncallee is set to the handler's PC and p->hipe.nsp + * On exit, p->hipe.u.ncallee is set to the handler's PC and p->hipe.nsp * is set to its SP (low address of its stack frame). */ void hipe_find_handler(Process *p) @@ -240,7 +240,7 @@ void hipe_find_handler(Process *p) if ((exnra = sdesc_exnra(sdesc)) != 0 && (p->catches >= 0 || exnra == (unsigned long)nbif_fail)) { - p->hipe.ncallee = (void(*)(void)) exnra; + p->hipe.u.ncallee = (void(*)(void)) exnra; p->hipe.nsp = nsp; p->hipe.narity = 0; /* update the gray/white boundary if we threw past it */ diff --git a/erts/emulator/internal_doc/CarrierMigration.md b/erts/emulator/internal_doc/CarrierMigration.md index b93c11c6ec..2a9594db25 100644 --- a/erts/emulator/internal_doc/CarrierMigration.md +++ b/erts/emulator/internal_doc/CarrierMigration.md @@ -16,12 +16,12 @@ When a carrier is empty, i.e. contains only one large free block, it is deallocated. Since multiblock carriers can contain both allocated blocks and free blocks at the same time, an allocator instance might be stuck with a large amount of poorly utilized carriers if the memory -load decrease. After a peak in memory usage it is expected that not -all memory can be returned since the blocks still allocated is likely +load decreases. After a peak in memory usage it is expected that not +all memory can be returned since the blocks still allocated are likely to be dispersed over multiple carriers. Such poorly utilized carriers -can usually be reused if the memory load increase again. However, +can usually be reused if the memory load increases again. However, since each scheduler thread manages its own set of allocator -instances, and memory load is not necessarily connected to CPU load we +instances, and memory load is not necessarily correlated to CPU load, we might get into a situation where there are lots of poorly utilized multiblock carriers on some allocator instances while we need to allocate new multiblock carriers on other allocator instances. In @@ -50,13 +50,13 @@ the allocator instance manages. Free blocks in one specific carrier can be referred to from potentially every other carrier that is managed, and the amount of such references can be huge. That is, the work of removing the free blocks of such a carrier from the search -tree will be huge. One way of solving this could be to not migrate +tree will be huge. One way of solving this could be not to migrate carriers that contain lots of free blocks, but this would prevent us -from migrating carriers that potentially needs to be migrated in order +from migrating carriers that potentially need to be migrated in order to solve the problem we set out to solve. By using one data structure of free blocks in each carrier and an -allocator instance wide data structure of carriers managed by the +allocator instance-wide data structure of carriers managed by the allocator instance, the work needed in order to remove and add carriers can be kept to a minimum. When migration of carriers is enabled on a specific allocator type, we require that an allocation @@ -76,9 +76,9 @@ through a pool of carriers. In order for a carrier migration to complete, one scheduler needs to move the carrier into the pool, and another scheduler needs to take the carrier out of the pool. -The pool is implemented as a lock free, circular, double linked, +The pool is implemented as a lock-free, circular, double linked, list. The list contains a sentinel which is used as the starting point -when inserting to, or fetching from the pool. Carriers in the pool are +when inserting to, or fetching from, the pool. Carriers in the pool are elements in this list. The list can be modified by all scheduler threads @@ -108,19 +108,19 @@ all search operations need to read the content of the sentinel. If we were to modify the sentinel, the cache line containing the sentinel would unnecessarily be bounced between processors. -The `prev`, and `next` fields in the elements of the list contains the +The `prev` and `next` fields in the elements of the list contain the value of the pointer, a modification marker, and a deleted marker. Memory operations on these fields are done using atomic memory operations. When a thread has set the modification marker in a field, no-one except the thread that set the marker is allowed to modify the -field. If multiple modification markers needs to be set, we always +field. If multiple modification markers need to be set, we always begin with `next` fields followed by `prev` fields in the order following the actual pointers. This guarantees that no deadlocks will occur. When a carrier is being removed from a pool, we mark it with a thread progress value that needs to be reached before we are allowed to -modify the `next`, and `prev` fields. That is, until we reach this +modify the `next` and `prev` fields. That is, until we reach this thread progress we are not allowed to insert the carrier into the pool again, and we are not allowed to deallocate the carrier. This ensures that threads inspecting the pool always will be able to traverse the @@ -130,12 +130,12 @@ threads may have references to it via the pool. ### Migration ### -There exist one pool for each allocator type enabling migration of +There exists one pool for each allocator type enabling migration of carriers between scheduler specific allocator instances of the same allocator type. Each allocator instance keeps track of the current utilization of its -multiblock carriers. When the utilization falls below the "abandon +multiblock carriers. When the total utilization falls below the "abandon carrier utilization limit" it starts to inspect the utilization of the current carrier when deallocations are made. If also the utilization of the carrier falls below the "abandon carrier utilization limit" it @@ -146,28 +146,53 @@ Since the carrier has been unlinked from the data structure of available free blocks, no more allocations will be made in the carrier. The allocator instance putting the carrier into the pool, however, still has the responsibility of performing deallocations in -it while it remains in the pool. +it while it remains in the pool. The allocator instance with this +deallocation responsibility is here called the **employer**. -Each carrier has a flag field containing information about allocator -instance owning the carrier, a flag indicating if the carrier is in +Each carrier has a flag field containing information about the +employing allocator instance, a flag indicating if the carrier is in the pool or not, and a flag indicating if it is busy or not. When the -carrier is in the pool, the owning allocator instance needs to mark it +carrier is in the pool, the employing allocator instance needs to mark it as busy while operating on it. If another thread inspects it in order -to try to fetch it from the pool, it will abort the fetch if it is -busy. When fetching the carrier from the pool, ownership will changed -and further deallocations in the carrier will be redirected to the new -owner using the delayed dealloc functionality. +to try to fetch it from the pool, it will skip it if it is busy. When +fetching the carrier from the pool, employment will change and further +deallocations in the carrier will be redirected to the new +employer using the delayed dealloc functionality. If a carrier in the pool becomes empty, it will be withdrawn from the pool. All carriers that become empty are also always passed to its -originating allocator instance for deallocation using the delayed +**owning** allocator instance for deallocation using the delayed dealloc functionality. Since carriers this way always will be -deallocated by the allocator instance that allocated the carrier the +deallocated by the owner that allocated the carrier, the underlying functionality of allocating and deallocating carriers can remain simple and doesn't have to bother about multiple threads. In a NUMA system we will also not mix carriers originating from multiple NUMA nodes. +In short: + +* The allocator instance that created a carrier **owns** it. +* An empty carrier is always deallocated by its **owner**. +* **Ownership** never changes. +* The allocator instance that uses a carrier **employs** it. +* An **employer** can abandon a carrier into the pool. +* Pooled carriers are not allocated from. +* Deallocation in a pooled carrier is still performed by its **employer**. +* **Employment** can only change when a carrier is fetched from the pool. + +### Searching the pool ### + +To harbor real time characteristics, searching the pool is +limited. We only inspect a limited number of carriers. If none of +those carriers had a free block large enough to satisfy the allocation +request, the search will fail. A carrier in the pool can also be busy +if another thread is currently doing block deallocation work on the +carrier. A busy carrier will also be skipped by the search as it can +not satisfy the request. The pool is lock-free and we do not want to +block, waiting for the other thread to finish. + +#### Before OTP 17.4 #### + When an allocator instance needs more carrier space, it always begins by inspecting its own carriers that are waiting for thread progress before they can be deallocated. If no such carrier could be found, it @@ -176,10 +201,69 @@ it will allocate a new carrier. Regardless of where the allocator instance gets the carrier from it the just links in the carrier into its data structure of free blocks. +#### After OTP 17.4 #### + +The old search algorithm had a problem as the search always started at +the same position in the pool, the sentinel. This could lead to +contention from concurrent searching processes. But even worse, it +could lead to a "bad" state when searches fail with a high rate +leading to new carriers instead being allocated. These new carriers +may later be inserted into the pool due to bad utilization. If the +frequency of insertions into the pool is higher than successful +fetching from the pool, memory will eventually get exhausted. + +This "bad" state consists of a cluster of small and/or highly +fragmented carriers located at the sentinel in the pool. The largest free +block in such a "bad" carrier is rather small, making it unable to satisfy +most allocation requests. As the search always started at the +sentinel, any such "bad" carriers that had been left in the pool would +eventually cluster together at the sentinel. All searches first +have to skip past this cluster of "bad" carriers to reach a "good" +carrier. When the cluster gets to the same size as the search limit, +all searches will essentially fail. + +To counter the "bad cluster" problem and also ease the contention, the +search will now always start by first looking at the allocators **own** +carriers. That is, carriers that were initially created by the +allocator itself and later had been abandoned to the pool. If none of +our own abandoned carrier would do, then the search continues into the +pool, as before, to look for carriers created by other +allocators. However, if we have at least one abandoned carrier of our +own that could not satisfy the request, we can use that as entry point +into the pool. + +The result is that we prefer carriers created by the thread itself, +which is good for NUMA performance. And we get more entry points when +searching the pool, which will ease contention and clustering. + +To do the first search among own carriers, every allocator instance +has two new lists: `pooled_list` and `traitor_list`. These lists are only +accessed by the allocator itself and they only contain the allocator's +own carriers. When an owned carrier is abandoned and put in the +pool, it is also linked into `pooled_list`. When we search our +`pooled_list` and find a carrier that is no longer in the pool, we +move that carrier from `pooled_list` to `traitor_list` as it is now +employed by another allocator. If searching `pooled_list` fails, we +also do a limited search of `traitor_list`. When finding an abandoned +carrier in `traitor_list` it is either employed or moved back to +`pooled_list` if it could not satisfy the allocation request. + +When searching `pooled_list` and `traitor_list` we always start at the +point where the last search ended. This to avoid clustering +problems and increase the probability to find a "good" carrier. As +`pooled_list` and `traitor_list` are only accessed by the owning +allocator instance, they need no thread synchronization at all. + +Furthermore, the search for own carriers that are scheduled +for deallocation is now done as the last search option. The idea is +that it is better to reuse a poorly utilized carrier than to +resurrect an empty carrier that was just about to be released back to +the OS. + ### Result ### The use of this strategy of abandoning carriers with poor utilization -and reusing these in allocator instances with an increased carrier +and reusing them in allocator instances with an increased carrier demand is extremely effective and completely eliminates the problems that otherwise sometimes occurred when CPU load dropped while memory load did not. diff --git a/erts/emulator/internal_doc/SuperCarrier.md b/erts/emulator/internal_doc/SuperCarrier.md new file mode 100644 index 0000000000..0ad6af41de --- /dev/null +++ b/erts/emulator/internal_doc/SuperCarrier.md @@ -0,0 +1,191 @@ +Super Carrier +============= + +A super carrier is large memory area, allocated at VM start, which can +be used during runtime to allocate normal carriers from. + +The super carrier feature was introduced in OTP R16B03. It is +enabled with command line option +MMscs <size in Mb> +and can be configured with other options. + +Problem +------- + +The initial motivation for this feature was customers asking for a way +to pre-allocate physcial memory at VM start for it to use. + +Other problems were different experienced limitations of the OS +implementation of mmap: + +* Increasingly bad performance of mmap/munmap as the number of mmap'ed areas grow. +* Fragmentation problem between mmap'ed areas. + +A third problem was management of low memory in the halfword +emulator. The implementation used a naive linear search structure to +hold free segments which would lead to poor performance when +fragmentation increased. + + +Solution +-------- + +Allocate one large continious area of address space at VM start and +then use that area to satisfy our dynamic memory need during +runtime. In other words: implement our own mmap. + +### Use cases ### + +If command line option +MMscrpm (Reserve Physical Memory) is set to +false, only virtual space is allocated for the super carrier from +start. The super carrier then acts as an "alternative mmap" implementation +without changing the consumption of physical memory pages. Physical +pages will be reserved on demand when an allocation is done from the super +carrier and be unreserved when the memory is released back to the +super carrier. + +If +MMscrpm is set to true, which is default, the initial allocation +will reserve physical memory for the entire super carrier. This can be +used by users that want to ensure a certain *minimum* amount of +physical memory for the VM. + +However, what reservation of physical memory actually means highly +depends on the operating system, and how it is configured. For +example, different memory overcommit settings on Linux drastically +change the behaviour. + +A third feature is to have the super carrier limit the *maximum* +amount of memory used by the VM. If +MMsco (Super Carrier Only) is set +to true, which is default, allocations will only be done from the +super carrier. When the super carrier gets full, the VM will fail due +to out of memory. +If +MMsco is false, allocations will use mmap directly if the super +carrier is full. + + + +### Implementation ### + +The entire super carrier implementation is kept in erl_mmap.c. The +name suggest that it can be viewed as our own mmap implementation. + +A super carrier needs to satisfy two slightly different kinds of +allocation requests; multi block carriers (MBC) and single block +carriers (SBC). They are both rather large blocks of continious +memory, but MBCs and SBCs have different demands on alignment and +size. + +SBCs can have arbitrary size and do only need minimum 8-byte +alignment. + +MBCs are more restricted. They can only have a number of fixed +sizes that are powers of 2. The start address need to have a very +large aligment (currently 256 kb, called "super alignment"). This is a +design choice that allows very low overhead per allocated block in the +MBC. + +To reduce fragmentation within the super carrier, it is good to keep SBCs +and MBCs apart. MBCs with their uniform alignment and sizes can be +packed very efficiently together. SBCs without demand for aligment can +also be allocated quite efficiently together. But mixing them can lead +to a lot of memory wasted when we need to create large holes of +padding to the next alignment limit. + +The super carrier thus contains two areas. One area for MBCs growing from +the bottom and up. And one area for SBCs growing from the top and +down. Like a process with a heap and a stack growing towards each +other. + + +### Data structures ### + +The MBC area is called **sa** as in super aligned and the SBC area is +called **sua** as in super un-aligned. + +Note that the "super" in super alignment and the "super" in super +carrier has nothing to do with each other. We could have choosen +another naming to avoid confusion, such as "meta" carrier or "giant" +aligment. + + +-------+ <---- sua.top + | sua | + | | + |-------| <---- sua.bot + | | + | | + | | + |-------| <---- sa.top + | | + | sa | + | | + +-------+ <---- sa.bot + + +When a carrier is deallocated a free memory segment will be created +inside the corresponding area, unless the carrier was at the very top +(in `sa`) or bottom (in `sua`) in which case the area will just shrink +down or up. + +We need to keep track of all the free segments in order to reuse them +for new carrier allocations. One initial idea was to use the same +mechanism that is used to keep track of free blocks within MBCs +(alloc_util and the different strategies). However, that would not be +as straight forward as one can think and can also waste quite a lot of +memory as it uses prepended block headers. The granularity of the +super carrier is one memory page (usually 4kb). We want to allocate +and free entire pages and we don't want to waste an entire page just +to hold the block header of the following pages. + +Instead we store the meta information about all the free segments in a +dedicated area apart from the `sa` and `sua` areas. Every free segment is +represented by a descriptor struct (`ErtsFreeSegDesc`). + + typedef struct { + RBTNode snode; /* node in 'stree' */ + RBTNode anode; /* node in 'atree' */ + char* start; + char* end; + }ErtsFreeSegDesc; + +To find the smallest free segment that will satisfy a carrier allocation +(best fit), the free segments are organized in a tree sorted by +size (`stree`). We search in this tree at allocation. If no free segment of +sufficient size was found, the area (`sa` or `sua`) is instead expanded. +If two or more free segments with equal size exist, the one at lowest +address is choosen for `sa` and highest address for `sua`. + +At carrier deallocation, we want to coalesce with any adjacent free +segments, to form one large free segment. To do that, all free +segments are also organized in a tree sorted in address order (`atree`). + +So, in total we keep four trees of free descriptors for the super +carrier; two for `sa` and two for `sua`. They all use the same +red-black-tree implementation that support the different sorting +orders used. + +When allocating a new MBC we first search after a free segment in `sa`, +then try to raise `sa.top`, and then as a fallback try to search after a +free segment in `sua`. When an MBC is allocated in `sua`, a larger segment +is allocated which is then trimmed to obtain the right +alignment. Allocation search for an SBC is done in reverse order. When +an SBC is allocated in `sa`, the size is aligned up to super aligned +size. + +### The free descriptor area ### + +As mentioned above, the descriptors for the free segments are +allocated in a separate area. This area has a constant configurable +size (+MMscrfsd) that defaults to 65536 descriptors. This should be +more than enough in most cases. If the descriptors area should fill up, +new descriptor areas will be allocated first directly from the OS, and +then from `sua` and `sa` in the super carrier, and lastly from the memory +segment itself which is being deallocated. Allocating free descriptor +areas from the super carrier is only a last resort, and should be +avoided, as it creates fragmentation. + +### Halfword emulator ### + +The halfword emulator uses the super carrier implementation to manage +its low memory mappings thar are needed for all term storage. The +super carrier can here not be configured by command line options. One +could imagine a second configurable instance of the super carrier used +by high memory allocation, but that has not been implemented. diff --git a/erts/emulator/sys/common/erl_check_io.c b/erts/emulator/sys/common/erl_check_io.c index 1db673e7f3..0051b45b31 100644 --- a/erts/emulator/sys/common/erl_check_io.c +++ b/erts/emulator/sys/common/erl_check_io.c @@ -52,8 +52,17 @@ typedef char EventStateType; #define ERTS_EV_TYPE_STOP_USE ((EventStateType) 3) /* pending stop_select */ typedef char EventStateFlags; -#define ERTS_EV_FLAG_USED ((EventStateFlags) 1) /* ERL_DRV_USE has been turned on */ +#define ERTS_EV_FLAG_USED ((EventStateFlags) 1) /* ERL_DRV_USE has been turned on */ +#define ERTS_EV_FLAG_DEFER_IN_EV ((EventStateFlags) 2) +#define ERTS_EV_FLAG_DEFER_OUT_EV ((EventStateFlags) 4) +#ifdef DEBUG +# define ERTS_ACTIVE_FD_INC 2 +#else +# define ERTS_ACTIVE_FD_INC 128 +#endif + +#define ERTS_CHECK_IO_POLL_RES_LEN 512 #if defined(ERTS_KERNEL_POLL_VERSION) # define ERTS_CIO_EXPORT(FUNC) FUNC ## _kp @@ -67,6 +76,7 @@ typedef char EventStateFlags; (ERTS_POLL_USE_POLL && !ERTS_POLL_USE_KERNEL_POLL) #define ERTS_CIO_POLL_CTL ERTS_POLL_EXPORT(erts_poll_control) +#define ERTS_CIO_POLL_CTLV ERTS_POLL_EXPORT(erts_poll_controlv) #define ERTS_CIO_POLL_WAIT ERTS_POLL_EXPORT(erts_poll_wait) #ifdef ERTS_POLL_NEED_ASYNC_INTERRUPT_SUPPORT #define ERTS_CIO_POLL_AS_INTR ERTS_POLL_EXPORT(erts_poll_async_sig_interrupt) @@ -85,6 +95,13 @@ static struct pollset_info { ErtsPollSet ps; erts_smp_atomic_t in_poll_wait; /* set while doing poll */ + struct { + int six; /* start index */ + int eix; /* end index */ + erts_smp_atomic32_t no; + int size; + ErtsSysFdType *array; + } active_fd; #ifdef ERTS_SMP struct removed_fd* removed_list; /* list of deselected fd's*/ erts_smp_spinlock_t removed_list_lock; @@ -97,9 +114,11 @@ typedef struct { SafeHashBucket hb; #endif ErtsSysFdType fd; - union { - ErtsDrvEventDataState *event; /* ERTS_EV_TYPE_DRV_EV */ + struct { ErtsDrvSelectDataState *select; /* ERTS_EV_TYPE_DRV_SEL */ +#if ERTS_CIO_HAVE_DRV_EVENT + ErtsDrvEventDataState *event; /* ERTS_EV_TYPE_DRV_EV */ +#endif erts_driver_t* drv_ptr; /* ERTS_EV_TYPE_STOP_USE */ } driver; ErtsPollEvents events; @@ -169,6 +188,10 @@ static ERTS_INLINE ErtsDrvEventState* hash_new_drv_ev_state(ErtsSysFdType fd) ErtsDrvEventState tmpl; tmpl.fd = fd; tmpl.driver.select = NULL; +#if ERTS_CIO_HAVE_DRV_EVENT + tmpl.driver.event = NULL; +#endif + tmpl.driver.drv_ptr = NULL; tmpl.events = 0; tmpl.remove_cnt = 0; tmpl.type = ERTS_EV_TYPE_NONE; @@ -209,6 +232,69 @@ ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(removed_fd, struct removed_fd, 64, ERTS_ALC_T_F #endif static ERTS_INLINE void +init_iotask(ErtsIoTask *io_task) +{ + erts_port_task_handle_init(&io_task->task); + erts_smp_atomic_init_nob(&io_task->executed_time, ~((erts_aint_t) 0)); +} + +static ERTS_INLINE int +is_iotask_active(ErtsIoTask *io_task, erts_aint_t current_cio_time) +{ + if (erts_port_task_is_scheduled(&io_task->task)) + return 1; + if (erts_smp_atomic_read_nob(&io_task->executed_time) == current_cio_time) + return 1; + return 0; +} + +static ERTS_INLINE ErtsDrvSelectDataState * +alloc_drv_select_data(void) +{ + ErtsDrvSelectDataState *dsp = erts_alloc(ERTS_ALC_T_DRV_SEL_D_STATE, + sizeof(ErtsDrvSelectDataState)); + dsp->inport = NIL; + dsp->outport = NIL; + init_iotask(&dsp->iniotask); + init_iotask(&dsp->outiotask); + return dsp; +} + +static ERTS_INLINE void +free_drv_select_data(ErtsDrvSelectDataState *dsp) +{ + ASSERT(!erts_port_task_is_scheduled(&dsp->iniotask.task)); + ASSERT(!erts_port_task_is_scheduled(&dsp->outiotask.task)); + erts_free(ERTS_ALC_T_DRV_SEL_D_STATE, dsp); +} + +#if ERTS_CIO_HAVE_DRV_EVENT + +static ERTS_INLINE ErtsDrvEventDataState * +alloc_drv_event_data(void) +{ + ErtsDrvEventDataState *dep = erts_alloc(ERTS_ALC_T_DRV_EV_D_STATE, + sizeof(ErtsDrvEventDataState)); + dep->port = NIL; + dep->data = NULL; + dep->removed_events = 0; +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + dep->deferred_events = 0; +#endif + init_iotask(&dep->iotask); + return dep; +} + +static ERTS_INLINE void +free_drv_event_data(ErtsDrvEventDataState *dep) +{ + ASSERT(!erts_port_task_is_scheduled(&dep->iotask.task)); + erts_free(ERTS_ALC_T_DRV_EV_D_STATE, dep); +} + +#endif /* ERTS_CIO_HAVE_DRV_EVENT */ + +static ERTS_INLINE void remember_removed(ErtsDrvEventState *state, struct pollset_info* psi) { #ifdef ERTS_SMP @@ -288,7 +374,7 @@ forget_removed(struct pollset_info* psi) drv_ptr = state->driver.drv_ptr; ASSERT(drv_ptr); state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; + state->flags &= ~ERTS_EV_FLAG_USED; state->driver.drv_ptr = NULL; /* Fall through */ case ERTS_EV_TYPE_NONE: @@ -345,6 +431,10 @@ grow_drv_ev_state(int min_ix) for (i = erts_smp_atomic_read_nob(&drv_ev_state_len); i < new_len; i++) { drv_ev_state[i].fd = (ErtsSysFdType) i; drv_ev_state[i].driver.select = NULL; +#if ERTS_CIO_HAVE_DRV_EVENT + drv_ev_state[i].driver.event = NULL; +#endif + drv_ev_state[i].driver.drv_ptr = NULL; drv_ev_state[i].events = 0; drv_ev_state[i].remove_cnt = 0; drv_ev_state[i].type = ERTS_EV_TYPE_NONE; @@ -365,11 +455,7 @@ grow_drv_ev_state(int min_ix) static ERTS_INLINE void abort_task(Eterm id, ErtsPortTaskHandle *pthp, EventStateType type) { - if (is_nil(id)) { - ASSERT(type == ERTS_EV_TYPE_NONE - || !erts_port_task_is_scheduled(pthp)); - } - else if (erts_port_task_is_scheduled(pthp)) { + if (is_not_nil(id) && erts_port_task_is_scheduled(pthp)) { erts_port_task_abort(pthp); ASSERT(erts_is_port_alive(id)); } @@ -384,7 +470,7 @@ abort_tasks(ErtsDrvEventState *state, int mode) #if ERTS_CIO_HAVE_DRV_EVENT case ERTS_EV_TYPE_DRV_EV: abort_task(state->driver.event->port, - &state->driver.event->task, + &state->driver.event->iotask.task, ERTS_EV_TYPE_DRV_EV); return; #endif @@ -398,14 +484,14 @@ abort_tasks(ErtsDrvEventState *state, int mode) case ERL_DRV_WRITE: ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); abort_task(state->driver.select->outport, - &state->driver.select->outtask, + &state->driver.select->outiotask.task, state->type); if (mode == ERL_DRV_WRITE) break; case ERL_DRV_READ: ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); abort_task(state->driver.select->inport, - &state->driver.select->intask, + &state->driver.select->iniotask.task, state->type); break; default: @@ -443,16 +529,14 @@ deselect(ErtsDrvEventState *state, int mode) if (!(state->events)) { switch (state->type) { case ERTS_EV_TYPE_DRV_SEL: - ASSERT(!erts_port_task_is_scheduled(&state->driver.select->intask)); - ASSERT(!erts_port_task_is_scheduled(&state->driver.select->outtask)); - erts_free(ERTS_ALC_T_DRV_SEL_D_STATE, - state->driver.select); + state->driver.select->inport = NIL; + state->driver.select->outport = NIL; break; #if ERTS_CIO_HAVE_DRV_EVENT case ERTS_EV_TYPE_DRV_EV: - ASSERT(!erts_port_task_is_scheduled(&state->driver.event->task)); - erts_free(ERTS_ALC_T_DRV_EV_D_STATE, - state->driver.event); + state->driver.event->port = NIL; + state->driver.event->data = NULL; + state->driver.event->removed_events = (ErtsPollEvents) 0; break; #endif case ERTS_EV_TYPE_NONE: @@ -462,20 +546,297 @@ deselect(ErtsDrvEventState *state, int mode) break; } - state->driver.select = NULL; state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; + state->flags &= ~ERTS_EV_FLAG_USED; remember_removed(state, &pollset); } } - #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS # define IS_FD_UNKNOWN(state) ((state)->type == ERTS_EV_TYPE_NONE && (state)->remove_cnt == 0) #else # define IS_FD_UNKNOWN(state) ((state) == NULL) #endif +static ERTS_INLINE void +check_fd_cleanup(ErtsDrvEventState *state, +#if ERTS_CIO_HAVE_DRV_EVENT + ErtsDrvEventDataState **free_event, +#endif + ErtsDrvSelectDataState **free_select) +{ + erts_aint_t current_cio_time; + + ERTS_SMP_LC_ASSERT(erts_smp_lc_mtx_is_locked(fd_mtx(state->fd))); + + current_cio_time = erts_smp_atomic_read_acqb(&erts_check_io_time); + *free_select = NULL; + if (state->driver.select + && (state->type != ERTS_EV_TYPE_DRV_SEL) + && !is_iotask_active(&state->driver.select->iniotask, current_cio_time) + && !is_iotask_active(&state->driver.select->outiotask, current_cio_time)) { + + *free_select = state->driver.select; + state->driver.select = NULL; + } + +#if ERTS_CIO_HAVE_DRV_EVENT + *free_event = NULL; + if (state->driver.event + && (state->type != ERTS_EV_TYPE_DRV_EV) + && !is_iotask_active(&state->driver.event->iotask, current_cio_time)) { + + *free_event = state->driver.event; + state->driver.event = NULL; + } +#endif + +#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (((state->type != ERTS_EV_TYPE_NONE) + | state->remove_cnt +#if ERTS_CIO_HAVE_DRV_EVENT + | (state->driver.event != NULL) +#endif + | (state->driver.select != NULL)) == 0) { + + hash_erase_drv_ev_state(state); + + } +#endif +} + +static ERTS_INLINE int +check_cleanup_active_fd(ErtsSysFdType fd, +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + ErtsPollControlEntry *pce, + int *pce_ix, +#endif + erts_aint_t current_cio_time) +{ + ErtsDrvEventState *state; + int active = 0; + erts_smp_mtx_t *mtx = fd_mtx(fd); + void *free_select = NULL; +#if ERTS_CIO_HAVE_DRV_EVENT + void *free_event = NULL; +#endif +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + ErtsPollEvents evon = 0, evoff = 0; +#endif + + erts_smp_mtx_lock(mtx); + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + state = &drv_ev_state[(int) fd]; +#else + state = hash_get_drv_ev_state(fd); /* may be NULL! */ + if (state) +#endif + { + if (state->driver.select) { +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + if (is_iotask_active(&state->driver.select->iniotask, current_cio_time)) { + active = 1; + if ((state->events & ERTS_POLL_EV_IN) + && !(state->flags & ERTS_EV_FLAG_DEFER_IN_EV)) { + evoff |= ERTS_POLL_EV_IN; + state->flags |= ERTS_EV_FLAG_DEFER_IN_EV; + } + } + else if (state->flags & ERTS_EV_FLAG_DEFER_IN_EV) { + if (state->events & ERTS_POLL_EV_IN) + evon |= ERTS_POLL_EV_IN; + state->flags &= ~ERTS_EV_FLAG_DEFER_IN_EV; + } + if (is_iotask_active(&state->driver.select->outiotask, current_cio_time)) { + active = 1; + if ((state->events & ERTS_POLL_EV_OUT) + && !(state->flags & ERTS_EV_FLAG_DEFER_OUT_EV)) { + evoff |= ERTS_POLL_EV_OUT; + state->flags |= ERTS_EV_FLAG_DEFER_OUT_EV; + } + } + else if (state->flags & ERTS_EV_FLAG_DEFER_OUT_EV) { + if (state->events & ERTS_POLL_EV_OUT) + evon |= ERTS_POLL_EV_OUT; + state->flags &= ~ERTS_EV_FLAG_DEFER_OUT_EV; + } + if (active) + (void) 0; + else +#else + if (is_iotask_active(&state->driver.select->iniotask, current_cio_time) + || is_iotask_active(&state->driver.select->outiotask, current_cio_time)) + active = 1; + else +#endif + if (state->type != ERTS_EV_TYPE_DRV_SEL) { + free_select = state->driver.select; + state->driver.select = NULL; + } + } + +#if ERTS_CIO_HAVE_DRV_EVENT + if (state->driver.event) { + if (is_iotask_active(&state->driver.event->iotask, current_cio_time)) { +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + ErtsPollEvents evs = state->events & ~state->driver.event->deferred_events; + if (evs) { + evoff |= evs; + state->driver.event->deferred_events |= evs; + } +#endif + active = 1; + } + else if (state->type != ERTS_EV_TYPE_DRV_EV) { + free_event = state->driver.event; + state->driver.event = NULL; + } +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + else { + ErtsPollEvents evs = state->events & state->driver.event->deferred_events; + if (evs) { + evon |= evs; + state->driver.event->deferred_events = 0; + } + } +#endif + + } +#endif + +#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (((state->type != ERTS_EV_TYPE_NONE) | state->remove_cnt | active) == 0) + hash_erase_drv_ev_state(state); +#endif + + } + + erts_smp_mtx_unlock(mtx); + + if (free_select) + free_drv_select_data(free_select); +#if ERTS_CIO_HAVE_DRV_EVENT + if (free_event) + free_drv_event_data(free_event); +#endif + +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + if (evoff) { + ErtsPollControlEntry *pcep = &pce[(*pce_ix)++]; + pcep->fd = fd; + pcep->events = evoff; + pcep->on = 0; + } + if (evon) { + ErtsPollControlEntry *pcep = &pce[(*pce_ix)++]; + pcep->fd = fd; + pcep->events = evon; + pcep->on = 1; + } +#endif + + return active; +} + +static void +check_cleanup_active_fds(erts_aint_t current_cio_time) +{ + int six = pollset.active_fd.six; + int eix = pollset.active_fd.eix; + erts_aint32_t no = erts_smp_atomic32_read_dirty(&pollset.active_fd.no); + int size = pollset.active_fd.size; + int ix = six; +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + /* every fd might add two entries */ + Uint pce_sz = 2*sizeof(ErtsPollControlEntry)*no; + ErtsPollControlEntry *pctrl_entries = (pce_sz + ? erts_alloc(ERTS_ALC_T_TMP, pce_sz) + : NULL); + int pctrl_ix = 0; +#endif + + while (ix != eix) { + ErtsSysFdType fd = pollset.active_fd.array[ix]; + int nix = ix + 1; + if (nix >= size) + nix = 0; + ASSERT(fd != ERTS_SYS_FD_INVALID); + if (!check_cleanup_active_fd(fd, +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + pctrl_entries, + &pctrl_ix, +#endif + current_cio_time)) { + no--; + if (ix == six) { +#ifdef DEBUG + pollset.active_fd.array[ix] = ERTS_SYS_FD_INVALID; +#endif + six = nix; + } + else { + pollset.active_fd.array[ix] = pollset.active_fd.array[six]; +#ifdef DEBUG + pollset.active_fd.array[six] = ERTS_SYS_FD_INVALID; +#endif + six++; + if (six >= size) + six = 0; + } + } + ix = nix; + } + +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + ASSERT(pctrl_ix <= pce_sz/sizeof(ErtsPollControlEntry)); + if (pctrl_ix) + ERTS_CIO_POLL_CTLV(pollset.ps, pctrl_entries, pctrl_ix); + if (pctrl_entries) + erts_free(ERTS_ALC_T_TMP, pctrl_entries); +#endif + + pollset.active_fd.six = six; + pollset.active_fd.eix = eix; + erts_smp_atomic32_set_relb(&pollset.active_fd.no, no); +} + +static ERTS_INLINE void +add_active_fd(ErtsSysFdType fd) +{ + int eix = pollset.active_fd.eix; + int size = pollset.active_fd.size; + + + pollset.active_fd.array[eix] = fd; + + erts_smp_atomic32_set_relb(&pollset.active_fd.no, + (erts_smp_atomic32_read_dirty(&pollset.active_fd.no) + + 1)); + + eix++; + if (eix >= size) + eix = 0; + if (pollset.active_fd.six == eix) { + pollset.active_fd.six = 0; + eix = size; + size += ERTS_ACTIVE_FD_INC; + pollset.active_fd.array = erts_realloc(ERTS_ALC_T_ACTIVE_FD_ARR, + pollset.active_fd.array, + sizeof(ErtsSysFdType)*size); + pollset.active_fd.size = size; +#ifdef DEBUG + { + int i; + for (i = eix + 1; i < size; i++) + pollset.active_fd.array[i] = ERTS_SYS_FD_INVALID; + } +#endif + + } + + pollset.active_fd.eix = eix; +} int ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, @@ -492,6 +853,10 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, ErtsDrvEventState *state; int wake_poller; int ret; +#if ERTS_CIO_HAVE_DRV_EVENT + ErtsDrvEventDataState *free_event = NULL; +#endif + ErtsDrvSelectDataState *free_select = NULL; #ifdef USE_VM_PROBES DTRACE_CHARBUF(name, 64); #endif @@ -593,9 +958,9 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, if (new_events & (ERTS_POLL_EV_ERR|ERTS_POLL_EV_NVAL)) { if (state->type == ERTS_EV_TYPE_DRV_SEL && !state->events) { state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; - erts_free(ERTS_ALC_T_DRV_SEL_D_STATE, state->driver.select); - state->driver.select = NULL; + state->flags &= ~ERTS_EV_FLAG_USED; + state->driver.select->inport = NIL; + state->driver.select->outport = NIL; } ret = -1; goto done; @@ -613,18 +978,10 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, state->events = new_events; if (ctl_events) { if (on) { - if (state->type == ERTS_EV_TYPE_NONE) { - ErtsDrvSelectDataState *dsdsp - = erts_alloc(ERTS_ALC_T_DRV_SEL_D_STATE, - sizeof(ErtsDrvSelectDataState)); - dsdsp->inport = NIL; - dsdsp->outport = NIL; - erts_port_task_handle_init(&dsdsp->intask); - erts_port_task_handle_init(&dsdsp->outtask); - ASSERT(state->driver.select == NULL); - state->driver.select = dsdsp; + if (!state->driver.select) + state->driver.select = alloc_drv_select_data(); + if (state->type == ERTS_EV_TYPE_NONE) state->type = ERTS_EV_TYPE_DRV_SEL; - } ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); if (ctl_events & ERTS_POLL_EV_IN) state->driver.select->inport = id; @@ -645,17 +1002,12 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, state->driver.select->outport = NIL; } if (new_events == 0) { - ASSERT(!erts_port_task_is_scheduled(&state->driver.select->intask)); - ASSERT(!erts_port_task_is_scheduled(&state->driver.select->outtask)); if (old_events != 0) { remember_removed(state, &pollset); } if ((mode & ERL_DRV_USE) || !(state->flags & ERTS_EV_FLAG_USED)) { state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; - erts_free(ERTS_ALC_T_DRV_SEL_D_STATE, - state->driver.select); - state->driver.select = NULL; + state->flags &= ~ERTS_EV_FLAG_USED; } /*else keep it, as fd will probably be selected upon again */ } @@ -686,13 +1038,15 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, ret = 0; -done:; -#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS - if (state->type == ERTS_EV_TYPE_NONE && state->remove_cnt == 0) { - hash_erase_drv_ev_state(state); - } +done: + + check_fd_cleanup(state, +#if ERTS_CIO_HAVE_DRV_EVENT + &free_event, #endif -done_unknown: + &free_select); + +done_unknown: erts_smp_mtx_unlock(fd_mtx(fd)); if (stop_select_fn) { int was_unmasked = erts_block_fpe(); @@ -700,6 +1054,12 @@ done_unknown: (*stop_select_fn)(e, NULL); erts_unblock_fpe(was_unmasked); } + if (free_select) + free_drv_select_data(free_select); +#if ERTS_CIO_HAVE_DRV_EVENT + if (free_event) + free_drv_event_data(free_event); +#endif return ret; } @@ -719,6 +1079,10 @@ ERTS_CIO_EXPORT(driver_event)(ErlDrvPort ix, ErtsDrvEventState *state; int do_wake = 0; int ret; +#if ERTS_CIO_HAVE_DRV_EVENT + ErtsDrvEventDataState *free_event; +#endif + ErtsDrvSelectDataState *free_select; Port *prt = erts_drvport2port(ix); if (prt == ERTS_INVALID_ERL_DRV_PORT) @@ -799,10 +1163,8 @@ ERTS_CIO_EXPORT(driver_event)(ErlDrvPort ix, state->driver.event->removed_events |= remove_events; } else { - state->driver.event - = erts_alloc(ERTS_ALC_T_DRV_EV_D_STATE, - sizeof(ErtsDrvEventDataState)); - erts_port_task_handle_init(&state->driver.event->task); + if (!state->driver.event) + state->driver.event = alloc_drv_event_data(); state->driver.event->port = id; state->driver.event->removed_events = (ErtsPollEvents) 0; state->type = ERTS_EV_TYPE_DRV_EV; @@ -812,10 +1174,10 @@ ERTS_CIO_EXPORT(driver_event)(ErlDrvPort ix, else { if (state->type == ERTS_EV_TYPE_DRV_EV) { abort_tasks(state, 0); - erts_free(ERTS_ALC_T_DRV_EV_D_STATE, - state->driver.event); + state->driver.event->port = NIL; + state->driver.event->data = NULL; + state->driver.event->removed_events = (ErtsPollEvents) 0; } - state->driver.select = NULL; state->type = ERTS_EV_TYPE_NONE; remember_removed(state, &pollset); } @@ -825,12 +1187,22 @@ ERTS_CIO_EXPORT(driver_event)(ErlDrvPort ix, ret = 0; done: -#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS - if (state->type == ERTS_EV_TYPE_NONE && state->remove_cnt == 0) { - hash_erase_drv_ev_state(state); - } + + check_fd_cleanup(state, +#if ERTS_CIO_HAVE_DRV_EVENT + &free_event, #endif + &free_select); + erts_smp_mtx_unlock(fd_mtx(fd)); + + if (free_select) + free_drv_select_data(free_select); +#if ERTS_CIO_HAVE_DRV_EVENT + if (free_event) + free_drv_event_data(free_event); +#endif + return ret; #endif } @@ -1027,7 +1399,7 @@ steal_pending_stop_select(erts_dsprintf_buf_t *dsbufp, ErlDrvPort ix, * In either case stop_select should not be called. */ state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; + state->flags &= ~ERTS_EV_FLAG_USED; if (state->driver.drv_ptr->handle) { erts_ddll_dereference_driver(state->driver.drv_ptr->handle); } @@ -1099,38 +1471,103 @@ event_large_fd_error(ErlDrvPort ix, ErtsSysFdType fd, ErlDrvEventData event_data #endif #endif +static ERTS_INLINE int +io_task_schedule_allowed(ErtsDrvEventState *state, + ErtsPortTaskType type, + erts_aint_t current_cio_time) +{ + ErtsIoTask *io_task; + + switch (type) { + case ERTS_PORT_TASK_INPUT: + if (!state->driver.select) + return 0; +#if ERTS_CIO_HAVE_DRV_EVENT + if (state->driver.event) + return 0; +#endif + io_task = &state->driver.select->iniotask; + break; + case ERTS_PORT_TASK_OUTPUT: + if (!state->driver.select) + return 0; +#if ERTS_CIO_HAVE_DRV_EVENT + if (state->driver.event) + return 0; +#endif + io_task = &state->driver.select->outiotask; + break; +#if ERTS_CIO_HAVE_DRV_EVENT + case ERTS_PORT_TASK_EVENT: + if (!state->driver.event) + return 0; + if (state->driver.select) + return 0; + io_task = &state->driver.event->iotask; + break; +#endif + default: + ERTS_INTERNAL_ERROR("Invalid I/O-task type"); + return 0; + } + + return !is_iotask_active(io_task, current_cio_time); +} + static ERTS_INLINE void -iready(Eterm id, ErtsDrvEventState *state) +iready(Eterm id, ErtsDrvEventState *state, erts_aint_t current_cio_time) { - if (erts_port_task_schedule(id, - &state->driver.select->intask, - ERTS_PORT_TASK_INPUT, - (ErlDrvEvent) state->fd) != 0) { - stale_drv_select(id, state, ERL_DRV_READ); + if (io_task_schedule_allowed(state, + ERTS_PORT_TASK_INPUT, + current_cio_time)) { + ErtsIoTask *iotask = &state->driver.select->iniotask; + erts_smp_atomic_set_nob(&iotask->executed_time, current_cio_time); + if (erts_port_task_schedule(id, + &iotask->task, + ERTS_PORT_TASK_INPUT, + (ErlDrvEvent) state->fd) != 0) { + stale_drv_select(id, state, ERL_DRV_READ); + } + add_active_fd(state->fd); } } static ERTS_INLINE void -oready(Eterm id, ErtsDrvEventState *state) +oready(Eterm id, ErtsDrvEventState *state, erts_aint_t current_cio_time) { - if (erts_port_task_schedule(id, - &state->driver.select->outtask, - ERTS_PORT_TASK_OUTPUT, - (ErlDrvEvent) state->fd) != 0) { - stale_drv_select(id, state, ERL_DRV_WRITE); + if (io_task_schedule_allowed(state, + ERTS_PORT_TASK_OUTPUT, + current_cio_time)) { + ErtsIoTask *iotask = &state->driver.select->outiotask; + erts_smp_atomic_set_nob(&iotask->executed_time, current_cio_time); + if (erts_port_task_schedule(id, + &iotask->task, + ERTS_PORT_TASK_OUTPUT, + (ErlDrvEvent) state->fd) != 0) { + stale_drv_select(id, state, ERL_DRV_WRITE); + } + add_active_fd(state->fd); } } #if ERTS_CIO_HAVE_DRV_EVENT static ERTS_INLINE void -eready(Eterm id, ErtsDrvEventState *state, ErlDrvEventData event_data) +eready(Eterm id, ErtsDrvEventState *state, ErlDrvEventData event_data, + erts_aint_t current_cio_time) { - if (erts_port_task_schedule(id, - &state->driver.event->task, - ERTS_PORT_TASK_EVENT, - (ErlDrvEvent) state->fd, - event_data) != 0) { - stale_drv_select(id, state, 0); + if (io_task_schedule_allowed(state, + ERTS_PORT_TASK_EVENT, + current_cio_time)) { + ErtsIoTask *iotask = &state->driver.event->iotask; + erts_smp_atomic_set_nob(&iotask->executed_time, current_cio_time); + if (erts_port_task_schedule(id, + &iotask->task, + ERTS_PORT_TASK_EVENT, + (ErlDrvEvent) state->fd, + event_data) != 0) { + stale_drv_select(id, state, 0); + } + add_active_fd(state->fd); } } #endif @@ -1161,10 +1598,11 @@ ERTS_CIO_EXPORT(erts_check_io_interrupt_timed)(int set, void ERTS_CIO_EXPORT(erts_check_io)(int do_wait) { - ErtsPollResFd pollres[256]; + ErtsPollResFd *pollres; int pollres_len; SysTimeval wait_time; int poll_ret, i; + erts_aint_t current_cio_time; restart: @@ -1181,10 +1619,24 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) wait_time.tv_usec = 0; } + /* + * No need for an atomic inc op when incrementing + * erts_check_io_time, since only one thread can + * check io at a time. + */ + current_cio_time = erts_smp_atomic_read_dirty(&erts_check_io_time); + current_cio_time++; + erts_smp_atomic_set_relb(&erts_check_io_time, current_cio_time); + + check_cleanup_active_fds(current_cio_time); + #ifdef ERTS_ENABLE_LOCK_CHECK erts_lc_check_exact(NULL, 0); /* No locks should be locked */ #endif - pollres_len = sizeof(pollres)/sizeof(ErtsPollResFd); + + pollres_len = erts_smp_atomic32_read_dirty(&pollset.active_fd.no) + ERTS_CHECK_IO_POLL_RES_LEN; + + pollres = erts_alloc(ERTS_ALC_T_TMP, sizeof(ErtsPollResFd)*pollres_len); erts_smp_atomic_set_nob(&pollset.in_poll_wait, 1); @@ -1204,6 +1656,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) if (poll_ret != 0) { erts_smp_atomic_set_nob(&pollset.in_poll_wait, 0); forget_removed(&pollset); + erts_free(ERTS_ALC_T_TMP, pollres); if (poll_ret == EAGAIN) { goto restart; } @@ -1263,15 +1716,15 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) if ((revents & ERTS_POLL_EV_IN) || (!(revents & ERTS_POLL_EV_OUT) && state->events & ERTS_POLL_EV_IN)) { - iready(state->driver.select->inport, state); + iready(state->driver.select->inport, state, current_cio_time); } else if (state->events & ERTS_POLL_EV_OUT) { - oready(state->driver.select->outport, state); + oready(state->driver.select->outport, state, current_cio_time); } } else if (revents & (ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT)) { if (revents & ERTS_POLL_EV_OUT) { - oready(state->driver.select->outport, state); + oready(state->driver.select->outport, state, current_cio_time); } /* Someone might have deselected input since revents was read (true also on the non-smp emulator since @@ -1279,7 +1732,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) revents... */ revents &= ~(~state->events & ERTS_POLL_EV_IN); if (revents & ERTS_POLL_EV_IN) { - iready(state->driver.select->inport, state); + iready(state->driver.select->inport, state, current_cio_time); } } else if (revents & ERTS_POLL_EV_NVAL) { @@ -1287,6 +1740,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) state->driver.select->inport, state->driver.select->outport, state->events); + add_active_fd(state->fd); } break; } @@ -1304,8 +1758,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) if (revents) { event_data->events = state->events; event_data->revents = revents; - - eready(state->driver.event->port, state, event_data); + eready(state->driver.event->port, state, event_data, current_cio_time); } break; } @@ -1323,6 +1776,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) (int) state->type); ASSERT(0); deselect(state, 0); + add_active_fd(state->fd); break; } } @@ -1334,6 +1788,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) } erts_smp_atomic_set_nob(&pollset.in_poll_wait, 0); + erts_free(ERTS_ALC_T_TMP, pollres); forget_removed(&pollset); } @@ -1469,10 +1924,27 @@ static void drv_ev_state_free(void *des) void ERTS_CIO_EXPORT(erts_init_check_io)(void) { + erts_smp_atomic_init_nob(&erts_check_io_time, 0); erts_smp_atomic_init_nob(&pollset.in_poll_wait, 0); + ERTS_CIO_POLL_INIT(); pollset.ps = ERTS_CIO_NEW_POLLSET(); + pollset.active_fd.six = 0; + pollset.active_fd.eix = 0; + erts_smp_atomic32_init_nob(&pollset.active_fd.no, 0); + pollset.active_fd.size = ERTS_ACTIVE_FD_INC; + pollset.active_fd.array = erts_alloc(ERTS_ALC_T_ACTIVE_FD_ARR, + sizeof(ErtsSysFdType)*ERTS_ACTIVE_FD_INC); +#ifdef DEBUG + { + int i; + for (i = 0; i < ERTS_ACTIVE_FD_INC; i++) + pollset.active_fd.array[i] = ERTS_SYS_FD_INVALID; + } +#endif + + #ifdef ERTS_SMP init_removed_fd_alloc(); pollset.removed_list = NULL; @@ -1548,12 +2020,27 @@ Eterm ERTS_CIO_EXPORT(erts_check_io_info)(void *proc) { Process *p = (Process *) proc; - Eterm tags[15], values[15], res; + Eterm tags[16], values[16], res; Uint sz, *szp, *hp, **hpp, memory_size; Sint i; ErtsPollInfo pi; - - ERTS_CIO_POLL_INFO(pollset.ps, &pi); + erts_aint_t cio_time = erts_smp_atomic_read_acqb(&erts_check_io_time); + int active_fds = (int) erts_smp_atomic32_read_acqb(&pollset.active_fd.no); + + while (1) { + erts_aint_t post_cio_time; + int post_active_fds; + + ERTS_CIO_POLL_INFO(pollset.ps, &pi); + + post_cio_time = erts_smp_atomic_read_mb(&erts_check_io_time); + post_active_fds = (int) erts_smp_atomic32_read_acqb(&pollset.active_fd.no); + if (cio_time == post_cio_time && active_fds == post_active_fds) + break; + cio_time = post_cio_time; + active_fds = post_active_fds; + } + memory_size = pi.memory_size; #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS memory_size += sizeof(ErtsDrvEventState) * erts_smp_atomic_read_nob(&drv_ev_state_len); @@ -1617,6 +2104,9 @@ ERTS_CIO_EXPORT(erts_check_io_info)(void *proc) tags[i] = erts_bld_atom(hpp, szp, "max_fds"); values[i++] = erts_bld_uint(hpp, szp, (Uint) pi.max_fds); + tags[i] = erts_bld_atom(hpp, szp, "active_fds"); + values[i++] = erts_bld_uint(hpp, szp, (Uint) active_fds); + #ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS tags[i] = erts_bld_atom(hpp, szp, "no_avoided_wakeups"); values[i++] = erts_bld_uint(hpp, szp, (Uint) pi.no_avoided_wakeups); @@ -1671,6 +2161,8 @@ print_events(ErtsPollEvents ev) typedef struct { int used_fds; int num_errors; + int no_driver_select_structs; + int no_driver_event_structs; #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS int internal_fds; ErtsPollEvents *epep; @@ -1693,6 +2185,13 @@ static void doit_erts_check_io_debug(void *vstate, void *vcounters) struct stat stat_buf; #endif + if (state->driver.select) + counters->no_driver_select_structs++; +#if ERTS_CIO_HAVE_DRV_EVENT + if (state->driver.event) + counters->no_driver_event_structs++; +#endif + #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS if (state->events || ep_events) { if (ep_events & ERTS_POLL_EV_NVAL) { @@ -1831,6 +2330,7 @@ static void doit_erts_check_io_debug(void *vstate, void *vcounters) } } } +#if ERTS_CIO_HAVE_DRV_EVENT else if (state->type == ERTS_EV_TYPE_DRV_EV) { Eterm id; erts_printf("driver_event "); @@ -1866,6 +2366,7 @@ static void doit_erts_check_io_debug(void *vstate, void *vcounters) erts_free_port_names(pnp); } } +#endif #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS else if (internal) { erts_printf("internal "); @@ -1905,7 +2406,7 @@ static void doit_erts_check_io_debug(void *vstate, void *vcounters) } int -ERTS_CIO_EXPORT(erts_check_io_debug)(void) +ERTS_CIO_EXPORT(erts_check_io_debug)(ErtsCheckIoDebugInfo *ciodip) { #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS int fd, len; @@ -1915,6 +2416,10 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(void) ErtsDrvEventState null_des; null_des.driver.select = NULL; +#if ERTS_CIO_HAVE_DRV_EVENT + null_des.driver.event = NULL; +#endif + null_des.driver.drv_ptr = NULL; null_des.events = 0; null_des.remove_cnt = 0; null_des.type = ERTS_EV_TYPE_NONE; @@ -1935,6 +2440,8 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(void) #endif counters.used_fds = 0; counters.num_errors = 0; + counters.no_driver_select_structs = 0; + counters.no_driver_event_structs = 0; #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS len = erts_smp_atomic_read_nob(&drv_ev_state_len); @@ -1951,8 +2458,16 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(void) erts_smp_thr_progress_unblock(); + ciodip->no_used_fds = counters.used_fds; + ciodip->no_driver_select_structs = counters.no_driver_select_structs; + ciodip->no_driver_event_structs = counters.no_driver_event_structs; + erts_printf("\n"); erts_printf("used fds=%d\n", counters.used_fds); + erts_printf("Number of driver_select() structures=%d\n", counters.no_driver_select_structs); +#if ERTS_CIO_HAVE_DRV_EVENT + erts_printf("Number of driver_event() structures=%d\n", counters.no_driver_event_structs); +#endif #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS erts_printf("internal fds=%d\n", counters.internal_fds); #endif @@ -1961,6 +2476,7 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(void) #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS erts_free(ERTS_ALC_T_TMP, (void *) counters.epep); #endif + return counters.num_errors; } diff --git a/erts/emulator/sys/common/erl_check_io.h b/erts/emulator/sys/common/erl_check_io.h index edab7947ba..d01297d55c 100644 --- a/erts/emulator/sys/common/erl_check_io.h +++ b/erts/emulator/sys/common/erl_check_io.h @@ -26,6 +26,7 @@ #ifndef ERL_CHECK_IO_H__ #define ERL_CHECK_IO_H__ +#include "sys.h" #include "erl_sys_driver.h" #ifdef ERTS_ENABLE_KERNEL_POLL @@ -52,8 +53,8 @@ void erts_check_io_kp(int); void erts_check_io_nkp(int); void erts_init_check_io_kp(void); void erts_init_check_io_nkp(void); -int erts_check_io_debug_kp(void); -int erts_check_io_debug_nkp(void); +int erts_check_io_debug_kp(ErtsCheckIoDebugInfo *); +int erts_check_io_debug_nkp(ErtsCheckIoDebugInfo *); #else /* !ERTS_ENABLE_KERNEL_POLL */ @@ -70,6 +71,27 @@ void erts_init_check_io(void); #endif +extern erts_smp_atomic_t erts_check_io_time; + +typedef struct { + ErtsPortTaskHandle task; + erts_smp_atomic_t executed_time; +} ErtsIoTask; + +ERTS_GLB_INLINE void erts_io_notify_port_task_executed(ErtsPortTaskHandle *pthp); + +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +ERTS_GLB_INLINE void +erts_io_notify_port_task_executed(ErtsPortTaskHandle *pthp) +{ + ErtsIoTask *itp = (ErtsIoTask *) (((char *) pthp) - offsetof(ErtsIoTask, task)); + erts_aint_t ci_time = erts_smp_atomic_read_acqb(&erts_check_io_time); + erts_smp_atomic_set_relb(&itp->executed_time, ci_time); +} + +#endif + #endif /* ERL_CHECK_IO_H__ */ #if !defined(ERL_CHECK_IO_C__) && !defined(ERTS_ALLOC_C__) @@ -81,6 +103,16 @@ void erts_init_check_io(void); #include "erl_poll.h" #include "erl_port_task.h" +#ifdef __WIN32__ +/* + * Current erts_poll implementation for Windows cannot handle + * active events in the set of events polled. + */ +# define ERTS_CIO_DEFER_ACTIVE_EVENTS 1 +#else +# define ERTS_CIO_DEFER_ACTIVE_EVENTS 0 +#endif + /* * ErtsDrvEventDataState is used by driver_event() which is almost never * used. We allocate ErtsDrvEventDataState separate since we dont wan't @@ -91,13 +123,16 @@ typedef struct { Eterm port; ErlDrvEventData data; ErtsPollEvents removed_events; - ErtsPortTaskHandle task; +#if ERTS_CIO_DEFER_ACTIVE_EVENTS + ErtsPollEvents deferred_events; +#endif + ErtsIoTask iotask; } ErtsDrvEventDataState; typedef struct { Eterm inport; Eterm outport; - ErtsPortTaskHandle intask; - ErtsPortTaskHandle outtask; + ErtsIoTask iniotask; + ErtsIoTask outiotask; } ErtsDrvSelectDataState; #endif /* #ifndef ERL_CHECK_IO_INTERNAL__ */ diff --git a/erts/emulator/sys/common/erl_poll.c b/erts/emulator/sys/common/erl_poll.c index 0a58a625b2..aa412a20c8 100644 --- a/erts/emulator/sys/common/erl_poll.c +++ b/erts/emulator/sys/common/erl_poll.c @@ -2157,7 +2157,7 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps, #ifdef ERTS_POLL_DEBUG_PRINT erts_printf("Entering erts_poll_wait(), timeout=%d\n", - (int) tv->tv_sec*1000 + tv->tv_usec/1000); + (int) tvp->tv_sec*1000 + tvp->tv_usec/1000); #endif if (ERTS_POLLSET_SET_POLLED_CHK(ps)) { diff --git a/erts/emulator/sys/common/erl_sys_common_misc.c b/erts/emulator/sys/common/erl_sys_common_misc.c index e3ba741058..e63f0bda54 100644 --- a/erts/emulator/sys/common/erl_sys_common_misc.c +++ b/erts/emulator/sys/common/erl_sys_common_misc.c @@ -44,6 +44,14 @@ #endif #endif +/* + * erts_check_io_time is used by the erl_check_io implementation. The + * global erts_check_io_time variable is declared here since there + * (often) exist two versions of erl_check_io (kernel-poll and + * non-kernel-poll), and we dont want two versions of this variable. + */ +erts_smp_atomic_t erts_check_io_time; + /* Written once and only once */ static int filename_encoding = ERL_FILENAME_UNKNOWN; diff --git a/erts/emulator/sys/unix/erl_unix_sys.h b/erts/emulator/sys/unix/erl_unix_sys.h index 176fc049a7..26ed2fb558 100644 --- a/erts/emulator/sys/unix/erl_unix_sys.h +++ b/erts/emulator/sys/unix/erl_unix_sys.h @@ -135,9 +135,6 @@ /* File descriptors are numbers anc consecutively allocated on Unix */ #define ERTS_SYS_CONTINOUS_FD_NUMBERS -#define HAVE_ERTS_CHECK_IO_DEBUG -int erts_check_io_debug(void); - #ifndef ERTS_SMP # undef ERTS_POLL_NEED_ASYNC_INTERRUPT_SUPPORT # define ERTS_POLL_NEED_ASYNC_INTERRUPT_SUPPORT @@ -230,8 +227,24 @@ extern void sys_stop_cat(void); */ #ifdef USE_ISINF_ISNAN /* simulate finite() */ -# define finite(f) (!isinf(f) && !isnan(f)) -# define HAVE_FINITE +# define isfinite(f) (!isinf(f) && !isnan(f)) +# define HAVE_ISFINITE +#elif (defined(__GNUC__) && !defined(__llvm__)) && defined(HAVE_FINITE) +/* We use finite in gcc as it emits assembler instead of + the function call that isfinite emits. The assembler is + significantly faster. */ +# ifdef isfinite +# undef isfinite +# endif +# define isfinite finite +# ifndef HAVE_ISFINITE +# define HAVE_ISFINITE +# endif +#elif defined(isfinite) && !defined(HAVE_ISFINITE) +# define HAVE_ISFINITE +#elif !defined(HAVE_ISFINITE) && defined(HAVE_FINITE) +# define isfinite finite +# define HAVE_ISFINITE #endif #ifdef NO_FPE_SIGNALS @@ -241,7 +254,7 @@ extern void sys_stop_cat(void); #define erts_thread_init_fp_exception() do{}while(0) #endif # define __ERTS_FP_CHECK_INIT(fpexnp) do {} while (0) -# define __ERTS_FP_ERROR(fpexnp, f, Action) if (!finite(f)) { Action; } else {} +# define __ERTS_FP_ERROR(fpexnp, f, Action) if (!isfinite(f)) { Action; } else {} # define __ERTS_FP_ERROR_THOROUGH(fpexnp, f, Action) __ERTS_FP_ERROR(fpexnp, f, Action) # define __ERTS_SAVE_FP_EXCEPTION(fpexnp) # define __ERTS_RESTORE_FP_EXCEPTION(fpexnp) @@ -305,7 +318,7 @@ static __inline__ void __ERTS_FP_CHECK_INIT(volatile unsigned long *fp_exception code to always throw floating-point exceptions on errors. */ static __inline__ int erts_check_fpe_thorough(volatile unsigned long *fp_exception, double f) { - return erts_check_fpe(fp_exception, f) || !finite(f); + return erts_check_fpe(fp_exception, f) || !isfinite(f); } # define __ERTS_FP_ERROR_THOROUGH(fpexnp, f, Action) \ do { if (erts_check_fpe_thorough((fpexnp),(f))) { Action; } } while (0) diff --git a/erts/emulator/sys/unix/sys.c b/erts/emulator/sys/unix/sys.c index c51954be38..47acd132ca 100644 --- a/erts/emulator/sys/unix/sys.c +++ b/erts/emulator/sys/unix/sys.c @@ -306,7 +306,7 @@ struct { void (*check_io)(int); Uint (*size)(void); Eterm (*info)(void *); - int (*check_io_debug)(void); + int (*check_io_debug)(ErtsCheckIoDebugInfo *); } io_func = {0}; @@ -328,9 +328,9 @@ Eterm erts_check_io_info(void *p) } int -erts_check_io_debug(void) +erts_check_io_debug(ErtsCheckIoDebugInfo *ip) { - return (*io_func.check_io_debug)(); + return (*io_func.check_io_debug)(ip); } diff --git a/erts/emulator/sys/win32/erl_poll.c b/erts/emulator/sys/win32/erl_poll.c index 7a1d129cd5..972170d465 100644 --- a/erts/emulator/sys/win32/erl_poll.c +++ b/erts/emulator/sys/win32/erl_poll.c @@ -1085,7 +1085,7 @@ void erts_poll_controlv(ErtsPollSet ps, pcev[i].events, pcev[i].on); } - ERTS_POLLSET_LOCK(ps); + ERTS_POLLSET_UNLOCK(ps); HARDTRACEF(("Out erts_poll_controlv")); } diff --git a/erts/emulator/sys/win32/erl_win_sys.h b/erts/emulator/sys/win32/erl_win_sys.h index a78dbf64af..838f0c61eb 100644 --- a/erts/emulator/sys/win32/erl_win_sys.h +++ b/erts/emulator/sys/win32/erl_win_sys.h @@ -113,12 +113,10 @@ /* * Our own type of "FD's" */ +#define ERTS_SYS_FD_INVALID INVALID_HANDLE_VALUE #define ERTS_SYS_FD_TYPE HANDLE #define NO_FSTAT_ON_SYS_FD_TYPE 1 /* They are events, not files */ -#define HAVE_ERTS_CHECK_IO_DEBUG -int erts_check_io_debug(void); - /* * For erl_time_sup */ diff --git a/erts/emulator/sys/win32/sys.c b/erts/emulator/sys/win32/sys.c index 0ded6b274e..164ef95629 100644 --- a/erts/emulator/sys/win32/sys.c +++ b/erts/emulator/sys/win32/sys.c @@ -1392,39 +1392,46 @@ int parse_command(wchar_t* cmd){ return i; } -static BOOL need_quotes(wchar_t *str) -{ - int in_quote = 0; - int backslashed = 0; - int naked_space = 0; - while (*str != L'\0') { - switch (*str) { - case L'\\' : - backslashed = !backslashed; - break; - case L'"': - if (backslashed) { - backslashed=0; - } else { - in_quote = !in_quote; - } - break; - case L' ': - backslashed = 0; - if (!(backslashed || in_quote)) { - naked_space++; - } - break; - default: - backslashed = 0; +/* + * Translating of command line arguments to correct format. In the examples + * below the '' are not part of the actual string. + * 'io:format("hello").' -> 'io:format(\"hello\").' + * 'io:format("is anybody in there?").' -> '"io:format(\"is anybody in there?\")."' + * 'Just nod if you can hear me.' -> '"Just nod if you can hear me."' + * 'Is there ""anyone at home?' -> '"Is there \"\"anyone at home?"' + * 'Relax."' -> 'Relax.\"' + * + * If new == NULL we just calculate the length. + * + * The reason for having to quote all of the is becasue CreateProcessW removes + * one level of escaping since it takes a single long command line rather + * than the argument chunks that unix uses. + */ +static int escape_and_quote(wchar_t *str, wchar_t *new, BOOL *quoted) { + int i, j = 0; + if (new == NULL) + *quoted = FALSE; + else if (*quoted) + new[j++] = L'"'; + for ( i = 0; str[i] != L'\0'; i++,j++) { + if (str[i] == L' ' && new == NULL && *quoted == FALSE) { + *quoted = TRUE; + j++; + } + /* check if we have to escape quotes */ + if (str[i] == L'"') { + if (new) new[j] = L'\\'; + j++; } - ++str; + if (new) new[j] = str[i]; } - return (naked_space > 0); + if (*quoted) { + if (new) new[j] = L'"'; + j++; + } + return j; } - - /* *---------------------------------------------------------------------- @@ -1585,31 +1592,24 @@ create_child_process wcscpy(appname, execPath); } if (argv == NULL) { - BOOL orig_need_q = need_quotes(execPath); + BOOL orig_need_q; wchar_t *ptr; - int ocl = wcslen(execPath); + int ocl = escape_and_quote(execPath, NULL, &orig_need_q); if (run_cmd) { newcmdline = (wchar_t *) erts_alloc(ERTS_ALC_T_TMP, - (ocl + ((orig_need_q) ? 3 : 1) - + 11)*sizeof(wchar_t)); + (ocl + 1 + 11)*sizeof(wchar_t)); memcpy(newcmdline,L"cmd.exe /c ",11*sizeof(wchar_t)); ptr = newcmdline + 11; } else { newcmdline = (wchar_t *) erts_alloc(ERTS_ALC_T_TMP, - (ocl + ((orig_need_q) ? 3 : 1))*sizeof(wchar_t)); + (ocl + 1)*sizeof(wchar_t)); ptr = (wchar_t *) newcmdline; } - if (orig_need_q) { - *ptr++ = L'"'; - } - memcpy(ptr,execPath,ocl*sizeof(wchar_t)); - ptr += ocl; - if (orig_need_q) { - *ptr++ = L'"'; - } - *ptr = L'\0'; + ptr += escape_and_quote(execPath, ptr, &orig_need_q); + ptr[0] = L'\0'; } else { - int sum = 1; /* '\0' */ + int sum = 0; + BOOL *qte = NULL; wchar_t **ar = argv; wchar_t *n; wchar_t *save_arg0 = NULL; @@ -1620,11 +1620,13 @@ create_child_process if (run_cmd) { sum += 11; /* cmd.exe /c */ } + + while (*ar != NULL) ar++; + qte = erts_alloc(ERTS_ALC_T_TMP, (ar - argv)*sizeof(BOOL)); + + ar = argv; while (*ar != NULL) { - sum += wcslen(*ar); - if (need_quotes(*ar)) { - sum += 2; /* quotes */ - } + sum += escape_and_quote(*ar,NULL,qte+(ar - argv)); sum++; /* space */ ++ar; } @@ -1636,26 +1638,18 @@ create_child_process n += 11; } while (*ar != NULL) { - int q = need_quotes(*ar); - sum = wcslen(*ar); - if (q) { - *n++ = L'"'; - } - memcpy(n,*ar,sum*sizeof(wchar_t)); - n += sum; - if (q) { - *n++ = L'"'; - } + n += escape_and_quote(*ar,n,qte+(ar - argv)); *n++ = L' '; ++ar; } - *(n-1) = L'\0'; + *(n-1) = L'\0'; /* overwrite last space with '\0' */ if (save_arg0 != NULL) { argv[0] = save_arg0; } + erts_free(ERTS_ALC_T_TMP, qte); } - DEBUGF(("Creating child process: %s, createFlags = %d\n", newcmdline, createFlags)); + DEBUGF((stderr,"Creating child process: %S, createFlags = %d\n", newcmdline, createFlags)); ok = CreateProcessW((wchar_t *) appname, (wchar_t *) newcmdline, NULL, @@ -2190,7 +2184,7 @@ static void fd_stop(ErlDrvData data) ASSERT(dp->out.flushEvent); SetEvent(dp->out.flushEvent); } while (WaitForSingleObject(dp->out.flushReplyEvent, 10) == WAIT_TIMEOUT - || !(dp->out.flags & DF_THREAD_FLUSHED)); + && !(dp->out.flags & DF_THREAD_FLUSHED)); } } diff --git a/erts/emulator/test/Makefile b/erts/emulator/test/Makefile index 0b0568c31a..dfbe47786a 100644 --- a/erts/emulator/test/Makefile +++ b/erts/emulator/test/Makefile @@ -31,6 +31,7 @@ MODULES= \ a_SUITE \ after_SUITE \ alloc_SUITE \ + async_ports_SUITE \ beam_SUITE \ beam_literals_SUITE \ bif_SUITE \ diff --git a/erts/emulator/test/a_SUITE.erl b/erts/emulator/test/a_SUITE.erl index 195c9c0a5f..17579be416 100644 --- a/erts/emulator/test/a_SUITE.erl +++ b/erts/emulator/test/a_SUITE.erl @@ -97,23 +97,13 @@ display_check_io(ChkIo) -> catch erlang:display('--- CHECK IO INFO ---'), catch erlang:display(ChkIo), catch erts_debug:set_internal_state(available_internal_state, true), - NoOfErrorFds = (catch erts_debug:get_internal_state(check_io_debug)), + NoOfErrorFds = (catch element(1, erts_debug:get_internal_state(check_io_debug))), catch erlang:display({'NoOfErrorFds', NoOfErrorFds}), catch erts_debug:set_internal_state(available_internal_state, false), catch erlang:display('--- CHECK IO INFO ---'), ok. get_check_io_info() -> - ChkIo = erlang:system_info(check_io), - case lists:keysearch(pending_updates, 1, ChkIo) of - {value, {pending_updates, 0}} -> - display_check_io(ChkIo), - ChkIo; - false -> - ChkIo; - _ -> - receive after 10 -> ok end, - get_check_io_info() - end. + z_SUITE:get_check_io_info(). diff --git a/erts/emulator/test/async_ports_SUITE.erl b/erts/emulator/test/async_ports_SUITE.erl new file mode 100644 index 0000000000..c89b3655ff --- /dev/null +++ b/erts/emulator/test/async_ports_SUITE.erl @@ -0,0 +1,118 @@ +-module(async_ports_SUITE). + +-include_lib("common_test/include/ct.hrl"). + +-compile(export_all). + +-define(PACKET_SIZE, (10 * 1024 * 8)). +-define(CPORT_DELAY, 100). +-define(TEST_LOOPS_COUNT, 100000). +-define(SLEEP_BEFORE_CHECK, 1000). +-define(TEST_PROCS_COUNT, 2). +-define(TC_TIMETRAP_SECONDS, 10). + +suite() -> [{ct_hooks,[ts_install_cth]}]. + +all() -> + [ + permanent_busy_test + ]. + +permanent_busy_test(Config) -> + ct:timetrap({seconds, ?TC_TIMETRAP_SECONDS}), + ExePath = filename:join(?config(data_dir, Config), "cport"), + + Self = self(), + spawn_link( + fun() -> + Block = <<0:?PACKET_SIZE>>, + + Port = open_port(ExePath), + + Testers = + lists:map( + fun(_) -> + erlang:spawn_link(?MODULE, run_loop, + [Self, + Port, + Block, + ?TEST_LOOPS_COUNT, + 0]) + end, + lists:seq(1, ?TEST_PROCS_COUNT)), + Self ! {test_info, Port, Testers}, + endless_flush(Port) + end), + + receive + {test_info, Port, Testers} -> + MaxWaitTime = round(0.7 * ?TC_TIMETRAP_SECONDS * 1000), + ct:log("wait testers, maximum ~w mcsec~n", [MaxWaitTime]), + ok = wait_testers(MaxWaitTime, Testers), + timer:sleep(?SLEEP_BEFORE_CHECK), + case erlang:port_command(Port, <<"test">>, [nosuspend]) of + false -> + exit(port_dead); + true -> + ok + end + end. + +wait_testers(Timeout, Testers) -> + lists:foldl( + fun(Pid, AccIn) -> + StartWait = os:timestamp(), + receive + {Pid, port_dead} -> + recalc_timeout(AccIn, StartWait) + after AccIn -> + Pid ! stop, + recalc_timeout(AccIn, StartWait) + end + end, Timeout, Testers), + ok. + +recalc_timeout(TimeoutIn, WaitStart) -> + erlang:max(0, TimeoutIn - round(timer:now_diff(os:timestamp(), WaitStart)) div 1000). + +open_port(ExePath) -> + erlang:open_port({spawn, ExePath ++ " 100"}, [{packet, 4}, eof, exit_status, use_stdio, binary]). + +run_loop(RootProc, Port, Block, CheckLimit, BusyCnt) -> + receive + stop -> + ok + after 0 -> + case erlang:port_command(Port, Block, [nosuspend]) of + true -> + run_loop(RootProc, Port, Block, CheckLimit, 0); + false -> + if + BusyCnt + 1 > CheckLimit -> + check_dead(RootProc, Port, Block, CheckLimit); + true -> + run_loop(RootProc, Port, Block, CheckLimit, BusyCnt + 1) + end + end + end. + +check_dead(RootProc, Port, Block, CheckLimit) -> + ct:log("~p: check port dead~n", [self()]), + timer:sleep(?SLEEP_BEFORE_CHECK), + case erlang:port_command(Port, Block, [nosuspend]) of + true -> + ct:log("not dead~n"), + run_loop(RootProc, Port, Block, CheckLimit, 0); + false -> + ct:log("port dead: ~p~n", [Port]), + RootProc ! {self(), port_dead}, + ok + end. + +endless_flush(Port) -> + receive + {Port, {data, _}} -> + endless_flush(Port); + {Port, SomethingWrong} -> + erlang:error({someting_wrong, SomethingWrong}) + end. diff --git a/erts/emulator/test/async_ports_SUITE_data/Makefile.src b/erts/emulator/test/async_ports_SUITE_data/Makefile.src new file mode 100644 index 0000000000..56da3fbe12 --- /dev/null +++ b/erts/emulator/test/async_ports_SUITE_data/Makefile.src @@ -0,0 +1,15 @@ +CC = @CC@ +LD = @LD@ +CFLAGS = @CFLAGS@ @DEFS@ +CROSSLDFLAGS = @CROSSLDFLAGS@ + +PROGS = cport@exe@ + + +all: $(PROGS) + +cport@exe@: cport@obj@ + $(LD) $(CROSSLDFLAGS) -o cport cport@obj@ @LIBS@ + +cport@obj@: cport.c + $(CC) -c -o cport@obj@ $(CFLAGS) cport.c diff --git a/erts/emulator/test/async_ports_SUITE_data/cport.c b/erts/emulator/test/async_ports_SUITE_data/cport.c new file mode 100644 index 0000000000..033aff382a --- /dev/null +++ b/erts/emulator/test/async_ports_SUITE_data/cport.c @@ -0,0 +1,81 @@ +#include <stdlib.h> +#include <stdio.h> +#include <errno.h> +#include <string.h> +#ifdef __WIN32__ +# include "windows.h" +# include "winbase.h" +#else +# include <unistd.h> +#endif + +typedef unsigned char byte; + +int read_cmd(byte *buf) +{ + int len; + if (read_exact(buf, 4) != 4) + return(-1); + + len = (buf[0] << 24) | (buf[1] << 16) | (buf[2] << 8) | buf[3]; + return read_exact(buf, len); +} + +int write_cmd(byte *buf, int len) +{ + byte li[4]; + li[0] = (len >> 24) & 0xff; + li[1] = (len >> 16) & 0xff; + li[2] = (len >> 8) & 0xff; + li[3] = len & 0xff; + write_exact(&li, 4); + + return write_exact(buf, len); +} + +int read_exact(byte *buf, int len) +{ + int i, got=0; + do { + if ((i = read(0, buf+got, len-got)) <= 0) + { + return(i); + } + got += i; + } while (got<len); + return len; +} + +int write_exact(byte *buf, int len) +{ + int i, wrote = 0; + do { + if ((i = write(1, buf+wrote, len-wrote)) < 0) + return (i); + wrote += i; + } while (wrote<len); + return len; +} + +byte static_buf[31457280]; // 30 mb + +int main(int argc, char **argv) { + int sleep_time = atoi(argv[1]); + int fn, arg, res; + byte *buf = &static_buf[0]; + int len = 0; + if (sleep_time <= 0) + sleep_time = 0; +#ifdef __WIN32__ + else + sleep_time = ((sleep_time - 1) / 1000) + 1; /* Milli seconds */ +#endif + while ((len = read_cmd(buf)) > 0) { +#ifdef __WIN32__ + Sleep((DWORD) sleep_time); +#else + usleep(sleep_time); +#endif + write_cmd(buf, len); + } +} diff --git a/erts/emulator/test/busy_port_SUITE.erl b/erts/emulator/test/busy_port_SUITE.erl index 4b4af0babe..2ed5aaa0d0 100644 --- a/erts/emulator/test/busy_port_SUITE.erl +++ b/erts/emulator/test/busy_port_SUITE.erl @@ -98,8 +98,10 @@ generator(0, Writer, _Data) -> %% Calling process_info(Pid, current_function) on a suspended process %% used to crash Beam. - {current_function, {erlang, send, 2}} = - process_info(Writer, current_function), + case process_info(Writer, [status,current_function]) of + [{status,suspended},{current_function,{erlang,send,2}}] -> ok; + [{status,suspended},{current_function,{erlang,bif_return_trap,_}}] -> ok + end, unlock_slave(); generator(N, Writer, Data) -> Writer ! {exec, Data}, diff --git a/erts/emulator/test/driver_SUITE.erl b/erts/emulator/test/driver_SUITE.erl index c62bc0c454..623d62f876 100644 --- a/erts/emulator/test/driver_SUITE.erl +++ b/erts/emulator/test/driver_SUITE.erl @@ -31,8 +31,9 @@ end_per_suite/1, init_per_group/2,end_per_group/2, init_per_testcase/2, end_per_testcase/2, + + a_test/1, outputv_echo/1, - timer_measure/1, timer_cancel/1, timer_change/1, @@ -79,7 +80,8 @@ thr_free_drv/1, async_blast/1, thr_msg_blast/1, - consume_timeslice/1]). + consume_timeslice/1, + z_test/1]). -export([bin_prefix/2]). @@ -122,19 +124,19 @@ init_per_testcase(Case, Config) when is_atom(Case), is_list(Config) -> _ -> erts_debug:set_internal_state(available_internal_state, true) end, erlang:display({init_per_testcase, Case}), - ?line 0 = erts_debug:get_internal_state(check_io_debug), + ?line 0 = element(1, erts_debug:get_internal_state(check_io_debug)), [{watchdog, Dog},{testcase, Case}|Config]. end_per_testcase(Case, Config) -> Dog = ?config(watchdog, Config), erlang:display({end_per_testcase, Case}), - ?line 0 = erts_debug:get_internal_state(check_io_debug), + ?line 0 = element(1, erts_debug:get_internal_state(check_io_debug)), ?t:timetrap_cancel(Dog). suite() -> [{ct_hooks,[ts_install_cth]}]. -all() -> - [outputv_errors, outputv_echo, queue_echo, {group, timer}, +all() -> %% Keep a_test first and z_test last... + [a_test, outputv_errors, outputv_echo, queue_echo, {group, timer}, driver_unloaded, io_ready_exit, use_fallback_pollset, bad_fd_in_pollset, driver_event, fd_change, steal_control, otp_6602, driver_system_info_base_ver, @@ -151,7 +153,8 @@ all() -> thr_free_drv, async_blast, thr_msg_blast, - consume_timeslice]. + consume_timeslice, + z_test]. groups() -> [{timer, [], @@ -917,8 +920,7 @@ steal_control_test(Hndl = {erts_poll_info, Before}) -> end. chkio_test_init(Config) when is_list(Config) -> - ?line wait_until_no_pending_updates(), - ?line ChkIo = erlang:system_info(check_io), + ?line ChkIo = get_stable_check_io_info(), ?line case catch lists:keysearch(name, 1, ChkIo) of {value, {name, erts_poll}} -> ?line ?t:format("Before test: ~p~n", [ChkIo]), @@ -937,8 +939,7 @@ chkio_test_fini({skipped, _} = Res) -> chkio_test_fini({chkio_test_result, Res, Before}) -> ?line ok = erl_ddll:unload_driver('chkio_drv'), ?line ok = erl_ddll:stop(), - ?line wait_until_no_pending_updates(), - ?line After = erlang:system_info(check_io), + ?line After = get_stable_check_io_info(), ?line ?t:format("After test: ~p~n", [After]), ?line verify_chkio_state(Before, After), ?line Res. @@ -985,7 +986,7 @@ chkio_test({erts_poll_info, Before}, ?line Fun(), ?line During = erlang:system_info(check_io), ?line erlang:display(During), - ?line 0 = erts_debug:get_internal_state(check_io_debug), + ?line 0 = element(1, erts_debug:get_internal_state(check_io_debug)), ?line ?t:format("During test: ~p~n", [During]), ?line chk_chkio_port(Port), ?line case erlang:port_control(Port, ?CHKIO_STOP, "") of @@ -1034,18 +1035,22 @@ verify_chkio_state(Before, After) -> After) end, ?line ok. - - -wait_until_no_pending_updates() -> - case lists:keysearch(pending_updates, 1, erlang:system_info(check_io)) of - {value, {pending_updates, 0}} -> - ok; - false -> - ok; +get_stable_check_io_info() -> + ChkIo = erlang:system_info(check_io), + PendUpdNo = case lists:keysearch(pending_updates, 1, ChkIo) of + {value, {pending_updates, PendNo}} -> + PendNo; + false -> + 0 + end, + {value, {active_fds, ActFds}} = lists:keysearch(active_fds, 1, ChkIo), + case {PendUpdNo, ActFds} of + {0, 0} -> + ChkIo; _ -> receive after 10 -> ok end, - wait_until_no_pending_updates() + get_stable_check_io_info() end. otp_6602(doc) -> ["Missed port lock when stealing control of fd from a " @@ -1062,10 +1067,9 @@ otp_6602(Config) when is_list(Config) -> %% Inet driver use port locking... {ok, S} = gen_udp:open(0), {ok, Fd} = inet:getfd(S), - {ok, Port} = inet:port(S), %% Steal fd (lock checker used to %% trigger here). - {ok, _S2} = gen_udp:open(Port,[{fd,Fd}]), + {ok, _S2} = gen_udp:open(0,[{fd,Fd}]), Parent ! Done end), ?line receive Done -> ok end, @@ -1085,7 +1089,15 @@ otp_6602(Config) when is_list(Config) -> ["async_thrs", "sched_thrs"])). --define(EXPECTED_SYSTEM_INFO_NAMES, ?EXPECTED_SYSTEM_INFO_NAMES2). +-define(EXPECTED_SYSTEM_INFO_NAMES3, + (?EXPECTED_SYSTEM_INFO_NAMES2 ++ + ["emu_nif_vsn"])). + +-define(EXPECTED_SYSTEM_INFO_NAMES4, + (?EXPECTED_SYSTEM_INFO_NAMES3 ++ + ["dirty_sched"])). + +-define(EXPECTED_SYSTEM_INFO_NAMES, ?EXPECTED_SYSTEM_INFO_NAMES4). 'driver_system_info_base_ver'(doc) -> []; @@ -1133,16 +1145,18 @@ check_driver_system_info_result(Result) -> drv_vsn_str2tup(erlang:system_info(driver_version))} of {DDVSN, DDVSN} -> ?line [] = Ns; - {{1, 0}, _} -> - ?line ExpNs = lists:sort(?EXPECTED_SYSTEM_INFO_NAMES - -- ?EXPECTED_SYSTEM_INFO_NAMES1), - ?line ExpNs = lists:sort(Ns); - {{1, 1}, _} -> + %% {{1, 0}, _} -> + %% ?line ExpNs = lists:sort(?EXPECTED_SYSTEM_INFO_NAMES + %% -- ?EXPECTED_SYSTEM_INFO_NAMES1), + %% ?line ExpNs = lists:sort(Ns); + %% {{1, 1}, _} -> + %% ?line ExpNs = lists:sort(?EXPECTED_SYSTEM_INFO_NAMES + %% -- ?EXPECTED_SYSTEM_INFO_NAMES2), + %% ?line ExpNs = lists:sort(Ns); + {{3, 0}, _} -> ?line ExpNs = lists:sort(?EXPECTED_SYSTEM_INFO_NAMES - -- ?EXPECTED_SYSTEM_INFO_NAMES2), - ?line ExpNs = lists:sort(Ns); - {{2, 0}, _} -> - ?line [] = Ns + -- ?EXPECTED_SYSTEM_INFO_NAMES3), + ?line ExpNs = lists:sort(Ns) end. chk_sis(SIs, Ns) -> @@ -1189,6 +1203,14 @@ check_si_res(["async_thrs", Value]) -> check_si_res(["sched_thrs", Value]) -> ?line Value = integer_to_list(erlang:system_info(schedulers)); +%% Data added in 3rd version of driver_system_info() (driver version 1.5) +check_si_res(["emu_nif_vsn", Value]) -> + ?line Value = erlang:system_info(nif_version); + +%% Data added in 4th version of driver_system_info() (driver version 3.1) +check_si_res(["dirty_sched", _Value]) -> + true; + check_si_res(Unexpected) -> ?line ?t:fail({unexpected_result, Unexpected}). @@ -2370,10 +2392,25 @@ count_proc_sched(Ps, PNs) -> PNs end. +a_test(Config) when is_list(Config) -> + check_io_debug(). + +z_test(Config) when is_list(Config) -> + check_io_debug(). + %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% Utilities %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% +check_io_debug() -> + get_stable_check_io_info(), + {NoErrorFds, NoUsedFds, NoDrvSelStructs, NoDrvEvStructs} + = erts_debug:get_internal_state(check_io_debug), + 0 = NoErrorFds, + NoUsedFds = NoDrvSelStructs, + 0 = NoDrvEvStructs, + ok. + %flush_msgs() -> % receive % M -> diff --git a/erts/emulator/test/driver_SUITE_data/sys_info_base_drv.c b/erts/emulator/test/driver_SUITE_data/sys_info_base_drv.c index e44c7dbd5e..964034f5a6 100644 --- a/erts/emulator/test/driver_SUITE_data/sys_info_base_drv.c +++ b/erts/emulator/test/driver_SUITE_data/sys_info_base_drv.c @@ -41,7 +41,9 @@ "thread=%s " \ "smp=%s " \ "async_thrs=%d " \ - "sched_thrs=%d" + "sched_thrs=%d " \ + "emu_nif_vsn=%d.%d" + static size_t sys_info_drv_max_res_len(ErlDrvSysInfo *sip) @@ -55,6 +57,7 @@ sys_info_drv_max_res_len(ErlDrvSysInfo *sip) slen += 5; /* smp */ slen += 20; /* async_thrs */ slen += 20; /* sched_thrs */ + slen += 2*20; /* emu_nif_vsn */ return slen; } @@ -72,7 +75,9 @@ sys_info_drv_sprintf_sys_info(ErlDrvSysInfo *sip, char *str) sip->thread_support ? "true" : "false", sip->smp_support ? "true" : "false", sip->async_threads, - sip->scheduler_threads); + sip->scheduler_threads, + sip->nif_major_version, + sip->nif_minor_version); } #include "sys_info_drv_impl.c" diff --git a/erts/emulator/test/driver_SUITE_data/sys_info_curr_drv.c b/erts/emulator/test/driver_SUITE_data/sys_info_curr_drv.c index 5bbc966932..6d2c47fdaf 100644 --- a/erts/emulator/test/driver_SUITE_data/sys_info_curr_drv.c +++ b/erts/emulator/test/driver_SUITE_data/sys_info_curr_drv.c @@ -40,7 +40,9 @@ "thread=%s " \ "smp=%s " \ "async_thrs=%d " \ - "sched_thrs=%d" + "sched_thrs=%d " \ + "emu_nif_vsn=%d.%d " \ + "dirty_sched=%s" static size_t sys_info_drv_max_res_len(ErlDrvSysInfo *sip) @@ -54,6 +56,8 @@ sys_info_drv_max_res_len(ErlDrvSysInfo *sip) slen += 5; /* smp */ slen += 20; /* async_thrs */ slen += 20; /* sched_thrs */ + slen += 2*20; /* emu_nif_vsn */ + slen += 5; /* dirty_sched */ return slen; } @@ -71,7 +75,10 @@ sys_info_drv_sprintf_sys_info(ErlDrvSysInfo *sip, char *str) sip->thread_support ? "true" : "false", sip->smp_support ? "true" : "false", sip->async_threads, - sip->scheduler_threads); + sip->scheduler_threads, + sip->nif_major_version, + sip->nif_minor_version, + sip->dirty_scheduler_support ? "true" : "false"); } #include "sys_info_drv_impl.c" diff --git a/erts/emulator/test/driver_SUITE_data/sys_info_prev_drv.c b/erts/emulator/test/driver_SUITE_data/sys_info_prev_drv.c index 63c69f751c..2271d7027b 100644 --- a/erts/emulator/test/driver_SUITE_data/sys_info_prev_drv.c +++ b/erts/emulator/test/driver_SUITE_data/sys_info_prev_drv.c @@ -41,7 +41,9 @@ "thread=%s " \ "smp=%s " \ "async_thrs=%d " \ - "sched_thrs=%d" + "sched_thrs=%d " \ + "emu_nif_vsn=%d.%d" + static size_t sys_info_drv_max_res_len(ErlDrvSysInfo *sip) @@ -55,6 +57,7 @@ sys_info_drv_max_res_len(ErlDrvSysInfo *sip) slen += 5; /* smp */ slen += 20; /* async_thrs */ slen += 20; /* sched_thrs */ + slen += 2*20; /* emu_nif_vsn */ return slen; } @@ -72,7 +75,9 @@ sys_info_drv_sprintf_sys_info(ErlDrvSysInfo *sip, char *str) sip->thread_support ? "true" : "false", sip->smp_support ? "true" : "false", sip->async_threads, - sip->scheduler_threads); + sip->scheduler_threads, + sip->nif_major_version, + sip->nif_minor_version); } #include "sys_info_drv_impl.c" diff --git a/erts/emulator/test/float_SUITE_data/fp_drv.c b/erts/emulator/test/float_SUITE_data/fp_drv.c index b80385c3f9..82d18d6440 100644 --- a/erts/emulator/test/float_SUITE_data/fp_drv.c +++ b/erts/emulator/test/float_SUITE_data/fp_drv.c @@ -29,9 +29,14 @@ #if defined (__GNUC__) int _finite(double x); #endif -#ifndef finite -#define finite _finite +#ifndef isfinite +#define isfinite _finite #endif +#elif !defined(HAVE_ISFINITE) && defined(HAVE_FINITE) +/* If not windows and we do not have isfinite */ +#define isfinite finite +#elif !defined(HAVE_ISFINITE) +# error "No finite function found!" #endif #include "erl_driver.h" @@ -79,21 +84,21 @@ do_test(void *unused) x = 3.23e133; y = 3.57e257; z = x*y; - if (finite(z)) + if (isfinite(z)) return "is finite (1)"; x = 5.0; y = 0.0; z = x/y; - if (finite(z)) + if (isfinite(z)) return "is finite (2)"; z = log(-1.0); - if (finite(z)) + if (isfinite(z)) return "is finite (3)"; z = log(0.0); - if (finite(z)) + if (isfinite(z)) return "is finite (4)"; return "ok"; diff --git a/erts/emulator/test/fun_SUITE.erl b/erts/emulator/test/fun_SUITE.erl index 8ad5f290ed..2968f5bebb 100644 --- a/erts/emulator/test/fun_SUITE.erl +++ b/erts/emulator/test/fun_SUITE.erl @@ -30,7 +30,7 @@ fun_to_port/1,t_hash/1,t_phash/1,t_phash2/1,md5/1, refc/1,refc_ets/1,refc_dist/1, const_propagation/1,t_arity/1,t_is_function2/1, - t_fun_info/1]). + t_fun_info/1,t_fun_info_mfa/1]). -export([nothing/0]). @@ -42,7 +42,8 @@ all() -> [bad_apply, bad_fun_call, badarity, ext_badarity, equality, ordering, fun_to_port, t_hash, t_phash, t_phash2, md5, refc, refc_ets, refc_dist, - const_propagation, t_arity, t_is_function2, t_fun_info]. + const_propagation, t_arity, t_is_function2, t_fun_info, + t_fun_info_mfa]. groups() -> []. @@ -824,6 +825,24 @@ t_fun_info(Config) when is_list(Config) -> ?line bad_info(<<1,2>>), ok. +t_fun_info_mfa(Config) when is_list(Config) -> + Fun1 = fun spawn_call/2, + {module,M1} = erlang:fun_info(Fun1, module), + {name,F1} = erlang:fun_info(Fun1, name), + {arity,A1} = erlang:fun_info(Fun1, arity), + {M1,F1,A1=2} = erlang:fun_info_mfa(Fun1), + %% Module fun. + Fun2 = fun ?MODULE:t_fun_info/1, + {module,M2} = erlang:fun_info(Fun2, module), + {name,F2} = erlang:fun_info(Fun2, name), + {arity,A2} = erlang:fun_info(Fun2, arity), + {M2,F2,A2=1} = erlang:fun_info_mfa(Fun2), + + %% Not fun. + {'EXIT',_} = (catch erlang:fun_info_mfa(id(d))), + ok. + + bad_info(Term) -> try erlang:fun_info(Term, module) of Any -> diff --git a/erts/emulator/test/match_spec_SUITE.erl b/erts/emulator/test/match_spec_SUITE.erl index fdce157abc..fc4a5028e1 100644 --- a/erts/emulator/test/match_spec_SUITE.erl +++ b/erts/emulator/test/match_spec_SUITE.erl @@ -30,6 +30,7 @@ -export([fpe/1]). -export([otp_9422/1]). -export([faulty_seq_trace/1, do_faulty_seq_trace/0]). +-export([maps/1]). -export([runner/2, loop_runner/3]). -export([f1/1, f2/2, f3/2, fn/1, fn/2, fn/3]). -export([do_boxed_and_small/0]). @@ -62,7 +63,8 @@ all() -> moving_labels, faulty_seq_trace, empty_list, - otp_9422]; + otp_9422, + maps]; true -> [not_run] end. @@ -899,6 +901,31 @@ fpe(Config) when is_list(Config) -> _ -> ok end. +maps(Config) when is_list(Config) -> + {ok,#{},[],[]} = erlang:match_spec_test(#{}, [{'_',[],['$_']}], table), + {ok,#{},[],[]} = erlang:match_spec_test(#{}, [{#{},[],['$_']}], table), + {ok,false,[],[]} = + erlang:match_spec_test(#{}, [{not_a_map,[],['$_']}], table), + {ok,bar,[],[]} = + erlang:match_spec_test(#{foo => bar}, + [{#{foo => '$1'},[],['$1']}], + table), + {ok,false,[],[]} = + erlang:match_spec_test(#{foo => bar}, + [{#{foo => qux},[],[qux]}], + table), + {ok,false,[],[]} = + erlang:match_spec_test(#{}, [{#{foo => '_'},[],[foo]}], table), + {error,_} = + erlang:match_spec_test(#{}, [{#{'$1' => '_'},[],[foo]}], table), + {ok,bar,[],[]} = + erlang:match_spec_test({#{foo => bar}}, + [{{#{foo => '$1'}},[],['$1']}], + table), + {ok,#{foo := 3},[],[]} = + erlang:match_spec_test({}, [{{},[],[#{foo => {'+',1,2}}]}], table), + ok. + empty_list(Config) when is_list(Config) -> Val=[{'$1',[], [{message,'$1'},{message,{caller}},{return_trace}]}], %% Did crash debug VM in faulty assert: diff --git a/erts/emulator/test/module_info_SUITE.erl b/erts/emulator/test/module_info_SUITE.erl index 8a63d9fe3e..f3986f0c4f 100644 --- a/erts/emulator/test/module_info_SUITE.erl +++ b/erts/emulator/test/module_info_SUITE.erl @@ -24,7 +24,7 @@ -export([all/0, suite/0,groups/0,init_per_suite/1, end_per_suite/1, init_per_group/2,end_per_group/2, init_per_testcase/2,end_per_testcase/2, - exports/1,functions/1,native/1]). + exports/1,functions/1,native/1,info/1]). %%-compile(native). @@ -52,8 +52,8 @@ end_per_group(_GroupName, Config) -> Config. -modules() -> - [exports, functions, native]. +modules() -> + [exports, functions, native, info]. init_per_testcase(Func, Config) when is_atom(Func), is_list(Config) -> Dog = ?t:timetrap(?t:minutes(3)), @@ -122,6 +122,22 @@ native_proj({Name,Arity,Addr}) -> native_filter(Set) -> sofs:no_elements(Set) =/= 1. +%% Test that the module info of this module is correct. Use +%% erlang:get_module_info(?MODULE) to avoid compiler optimization tricks. +info(Config) when is_list(Config) -> + Info = erlang:get_module_info(?MODULE), + All = all_exported(), + {ok,{?MODULE,MD5}} = beam_lib:md5(code:which(?MODULE)), + {module, ?MODULE} = lists:keyfind(module, 1, Info), + {md5, MD5} = lists:keyfind(md5, 1, Info), + {exports, Exports} = lists:keyfind(exports, 1, Info), + All = lists:sort(Exports), + {attributes, Attrs} = lists:keyfind(attributes, 1, Info), + {vsn,_} = lists:keyfind(vsn, 1, Attrs), + {compile, Compile} = lists:keyfind(compile, 1, Info), + {options,_} = lists:keyfind(options, 1, Compile), + ok. + %% Helper functions (local). add_arity(L) -> diff --git a/erts/emulator/test/nif_SUITE.erl b/erts/emulator/test/nif_SUITE.erl index b2da6f58af..4560077a51 100644 --- a/erts/emulator/test/nif_SUITE.erl +++ b/erts/emulator/test/nif_SUITE.erl @@ -37,7 +37,9 @@ threading/1, send/1, send2/1, send3/1, send_threaded/1, neg/1, is_checks/1, get_length/1, make_atom/1, make_string/1, reverse_list_test/1, - otp_9668/1, consume_timeslice/1, dirty_nif/1, dirty_nif_send/1 + otp_9828/1, + otp_9668/1, consume_timeslice/1, dirty_nif/1, dirty_nif_send/1, + dirty_nif_exception/1, nif_schedule/1 ]). -export([many_args_100/100]). @@ -64,7 +66,9 @@ all() -> resource_takeover, threading, send, send2, send3, send_threaded, neg, is_checks, get_length, make_atom, make_string,reverse_list_test, - otp_9668, consume_timeslice, dirty_nif, dirty_nif_send + otp_9828, + otp_9668, consume_timeslice, + nif_schedule, dirty_nif, dirty_nif_send, dirty_nif_exception ]. groups() -> @@ -1440,6 +1444,20 @@ otp_9668(Config) -> ?line verify_tmpmem(TmpMem), ok. +otp_9828(doc) -> ["Copy of writable binary"]; +otp_9828(Config) -> + ensure_lib_loaded(Config, 1), + + otp_9828_loop(<<"I'm alive!">>, 1000). + +otp_9828_loop(Bin, 0) -> + ok; +otp_9828_loop(Bin, Val) -> + WrtBin = <<Bin/binary, Val:32>>, + ok = otp_9828_nif(WrtBin), + otp_9828_loop(WrtBin, Val-1). + + consume_timeslice(Config) when is_list(Config) -> CONTEXT_REDS = 2000, Me = self(), @@ -1524,6 +1542,20 @@ consume_timeslice(Config) when is_list(Config) -> ok. +nif_schedule(Config) when is_list(Config) -> + ensure_lib_loaded(Config), + A = "this is a string", + B = {this,is,a,tuple}, + {B,A} = call_nif_schedule(A, B), + ok = try call_nif_schedule(1, 2) + catch + error:badarg -> + [{?MODULE,call_nif_schedule,[1,2],_}|_] = + erlang:get_stacktrace(), + ok + end, + ok. + dirty_nif(Config) when is_list(Config) -> try erlang:system_info(dirty_cpu_schedulers) of N when is_integer(N) -> @@ -1532,6 +1564,8 @@ dirty_nif(Config) when is_list(Config) -> Val2 = "Erlang", Val3 = list_to_binary([Val2, 0]), {Val1, Val2, Val3} = call_dirty_nif(Val1, Val2, Val3), + LargeArray = lists:duplicate(1000, ok), + LargeArray = call_dirty_nif_zero_args(), ok catch error:badarg -> @@ -1556,6 +1590,24 @@ dirty_nif_send(Config) when is_list(Config) -> {skipped,"No dirty scheduler support"} end. +dirty_nif_exception(Config) when is_list(Config) -> + try erlang:system_info(dirty_cpu_schedulers) of + N when is_integer(N) -> + ensure_lib_loaded(Config), + try + call_dirty_nif_exception(), + ?t:fail(expected_badarg) + catch + error:badarg -> + [{?MODULE,call_dirty_nif_exception,[],_}|_] = + erlang:get_stacktrace(), + ok + end + catch + error:badarg -> + {skipped,"No dirty scheduler support"} + end. + next_msg(_Pid) -> receive M -> M @@ -1684,9 +1736,13 @@ reverse_list(_) -> ?nif_stub. echo_int(_) -> ?nif_stub. type_sizes() -> ?nif_stub. otp_9668_nif(_) -> ?nif_stub. +otp_9828_nif(_) -> ?nif_stub. consume_timeslice_nif(_,_) -> ?nif_stub. +call_nif_schedule(_,_) -> ?nif_stub. call_dirty_nif(_,_,_) -> ?nif_stub. send_from_dirty_nif(_) -> ?nif_stub. +call_dirty_nif_exception() -> ?nif_stub. +call_dirty_nif_zero_args() -> ?nif_stub. %% maps is_map_nif(_) -> ?nif_stub. diff --git a/erts/emulator/test/nif_SUITE_data/nif_SUITE.c b/erts/emulator/test/nif_SUITE_data/nif_SUITE.c index 955dc64189..85544db2ab 100644 --- a/erts/emulator/test/nif_SUITE_data/nif_SUITE.c +++ b/erts/emulator/test/nif_SUITE_data/nif_SUITE.c @@ -1473,6 +1473,26 @@ static ERL_NIF_TERM otp_9668_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM ar return atom_ok; } +static ERL_NIF_TERM otp_9828_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + /* copy a writable binary could reallocate it due to "emasculation" + and thereby render a previous inspection invalid. + */ + ErlNifBinary bin1; + ErlNifEnv* myenv; + + if (!enif_inspect_binary(env, argv[0], &bin1)) { + return enif_make_badarg(env); + } + + myenv = enif_alloc_env(); + enif_make_copy(myenv, argv[0]); + enif_free_env(myenv); + + return memcmp(bin1.data, "I'm alive!", 10)==0 ? atom_ok : atom_false; +} + + static ERL_NIF_TERM consume_timeslice_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) { int percent; @@ -1493,25 +1513,57 @@ static ERL_NIF_TERM consume_timeslice_nif(ErlNifEnv* env, int argc, const ERL_NI } } +static ERL_NIF_TERM nif_sched2(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + char s[64]; + if (!enif_get_string(env, argv[2], s, sizeof s, ERL_NIF_LATIN1)) + return enif_make_badarg(env); + return enif_make_tuple2(env, argv[3], argv[2]); +} + +static ERL_NIF_TERM nif_sched1(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + ERL_NIF_TERM new_argv[4]; + new_argv[0] = enif_make_atom(env, "garbage0"); + new_argv[1] = enif_make_atom(env, "garbage1"); + new_argv[2] = argv[0]; + new_argv[3] = argv[1]; + return enif_schedule_nif(env, "nif_sched2", 0, nif_sched2, 4, new_argv); +} + +static ERL_NIF_TERM call_nif_schedule(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + if (argc != 2) + return enif_make_atom(env, "false"); + return enif_schedule_nif(env, "nif_sched1", 0, nif_sched1, argc, argv); +} + #ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + +static int have_dirty_schedulers(void) +{ + ErlNifSysInfo si; + enif_system_info(&si, sizeof(si)); + return si.dirty_scheduler_support; +} + static ERL_NIF_TERM dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) { int n; char s[10]; ErlNifBinary b; ERL_NIF_TERM result; - if (enif_have_dirty_schedulers()) { + if (have_dirty_schedulers()) { assert(enif_is_on_dirty_scheduler(env)); } assert(argc == 3); enif_get_int(env, argv[0], &n); enif_get_string(env, argv[1], s, sizeof s, ERL_NIF_LATIN1); enif_inspect_binary(env, argv[2], &b); - result = enif_make_tuple3(env, - enif_make_int(env, n), - enif_make_string(env, s, ERL_NIF_LATIN1), - enif_make_binary(env, &b)); - return enif_schedule_dirty_nif_finalizer(env, result, enif_dirty_nif_finalizer); + return enif_make_tuple3(env, + enif_make_int(env, n), + enif_make_string(env, s, ERL_NIF_LATIN1), + enif_make_binary(env, &b)); } static ERL_NIF_TERM call_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) @@ -1522,11 +1574,11 @@ static ERL_NIF_TERM call_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM assert(!enif_is_on_dirty_scheduler(env)); if (argc != 3) return enif_make_badarg(env); - if (enif_have_dirty_schedulers()) { + if (have_dirty_schedulers()) { if (enif_get_int(env, argv[0], &n) && enif_get_string(env, argv[1], s, sizeof s, ERL_NIF_LATIN1) && enif_inspect_binary(env, argv[2], &b)) - return enif_schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_CPU_BOUND, dirty_nif, argc, argv); + return enif_schedule_nif(env, "call_dirty_nif", ERL_NIF_DIRTY_JOB_CPU_BOUND, dirty_nif, argc, argv); else return enif_make_badarg(env); } else { @@ -1534,35 +1586,54 @@ static ERL_NIF_TERM call_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM } } -static ERL_NIF_TERM dirty_sender(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +static ERL_NIF_TERM send_from_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) { ERL_NIF_TERM result; ErlNifPid pid; ErlNifEnv* menv; int res; - enif_get_local_pid(env, argv[0], &pid); + if (!enif_get_local_pid(env, argv[0], &pid)) + return enif_make_badarg(env); result = enif_make_tuple2(env, enif_make_atom(env, "ok"), enif_make_pid(env, &pid)); menv = enif_alloc_env(); res = enif_send(env, &pid, menv, result); enif_free_env(menv); if (!res) - /* Note the next line will crash, since dirty nifs can't return exceptions. - * This is intentional, since enif_send should not fail if the test succeeds. - */ - return enif_schedule_dirty_nif_finalizer(env, enif_make_badarg(env), enif_dirty_nif_finalizer); + return enif_make_badarg(env); else - return enif_schedule_dirty_nif_finalizer(env, result, enif_dirty_nif_finalizer); + return result; } -static ERL_NIF_TERM send_from_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +static ERL_NIF_TERM call_dirty_nif_exception(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) { - ERL_NIF_TERM result; - ErlNifPid pid; - - if (!enif_get_local_pid(env, argv[0], &pid)) + switch (argc) { + case 0: { + ERL_NIF_TERM args[255]; + int i; + for (i = 0; i < 255; i++) + args[i] = enif_make_int(env, i); + return enif_schedule_nif(env, "call_dirty_nif_exception", ERL_NIF_DIRTY_JOB_CPU_BOUND, + call_dirty_nif_exception, 255, argv); + } + case 1: return enif_make_badarg(env); - return enif_schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_CPU_BOUND, dirty_sender, argc, argv); + default: + return enif_schedule_nif(env, "call_dirty_nif_exception", ERL_NIF_DIRTY_JOB_CPU_BOUND, + call_dirty_nif_exception, argc-1, argv); + } +} + +static ERL_NIF_TERM call_dirty_nif_zero_args(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + int i; + ERL_NIF_TERM result[1000]; + ERL_NIF_TERM ok = enif_make_atom(env, "ok"); + assert(argc == 0); + for (i = 0; i < sizeof(result)/sizeof(*result); i++) { + result[i] = ok; + } + return enif_make_list_from_array(env, result, i); } #endif @@ -1741,10 +1812,14 @@ static ErlNifFunc nif_funcs[] = {"echo_int", 1, echo_int}, {"type_sizes", 0, type_sizes}, {"otp_9668_nif", 1, otp_9668_nif}, + {"otp_9828_nif", 1, otp_9828_nif}, {"consume_timeslice_nif", 2, consume_timeslice_nif}, + {"call_nif_schedule", 2, call_nif_schedule}, #ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT {"call_dirty_nif", 3, call_dirty_nif}, - {"send_from_dirty_nif", 1, send_from_dirty_nif}, + {"send_from_dirty_nif", 1, send_from_dirty_nif, ERL_NIF_DIRTY_JOB_CPU_BOUND}, + {"call_dirty_nif_exception", 0, call_dirty_nif_exception, ERL_NIF_DIRTY_JOB_IO_BOUND}, + {"call_dirty_nif_zero_args", 0, call_dirty_nif_zero_args, ERL_NIF_DIRTY_JOB_CPU_BOUND}, #endif {"is_map_nif", 1, is_map_nif}, {"get_map_size_nif", 1, get_map_size_nif}, diff --git a/erts/emulator/test/num_bif_SUITE.erl b/erts/emulator/test/num_bif_SUITE.erl index ff8d18eef8..8cf8377c30 100644 --- a/erts/emulator/test/num_bif_SUITE.erl +++ b/erts/emulator/test/num_bif_SUITE.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1997-2013. All Rights Reserved. +%% Copyright Ericsson AB 1997-2014. 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 @@ -394,18 +394,15 @@ t_string_to_integer(Config) when is_list(Config) -> test_sti(268435455), test_sti(-268435455), - %% 1 bsl 28 - 1, just before 32 bit bignum - test_sti(1 bsl 28 - 1), - %% 1 bsl 28, just beyond 32 bit small - test_sti(1 bsl 28), - %% 1 bsl 33, just beyond 32 bit - test_sti(1 bsl 33), - %% 1 bsl 60 - 1, just before 64 bit bignum - test_sti(1 bsl 60 - 1), - %% 1 bsl 60, just beyond 64 bit small - test_sti(1 bsl 60), - %% 1 bsl 65, just beyond 64 bit - test_sti(1 bsl 65), + % Interesting values around 2-pows, such as MIN_SMALL and MAX_SMALL. + lists:foreach(fun(Bits) -> + N = 1 bsl Bits, + test_sti(N - 1), + test_sti(N), + test_sti(N + 1) + end, + lists:seq(16, 130)), + %% Bignums. test_sti(123456932798748738738,16), test_sti(list_to_integer(lists:duplicate(2000, $1))), @@ -454,10 +451,11 @@ test_sti(Num) -> end|| Base <- lists:seq(2,36)]. test_sti(Num,Base) -> - Num = list_to_integer(int2list(Num,Base),Base), - Num = -1*list_to_integer(int2list(Num*-1,Base),Base), - Num = binary_to_integer(int2bin(Num,Base),Base), - Num = -1*binary_to_integer(int2bin(Num*-1,Base),Base). + Neg = -Num, + Num = list_to_integer(int2list(Num,Base),Base), + Neg = list_to_integer(int2list(Num*-1,Base),Base), + Num = binary_to_integer(int2bin(Num,Base),Base), + Neg = binary_to_integer(int2bin(Num*-1,Base),Base). % Calling this function (which is not supposed to be inlined) prevents % the compiler from calculating the answer, so we don't test the compiler diff --git a/erts/emulator/test/port_SUITE.erl b/erts/emulator/test/port_SUITE.erl index e01b2f253b..6bbf93b7d7 100644 --- a/erts/emulator/test/port_SUITE.erl +++ b/erts/emulator/test/port_SUITE.erl @@ -90,6 +90,7 @@ mix_up_ports/1, otp_5112/1, otp_5119/1, otp_6224/1, exit_status_multi_scheduling_block/1, ports/1, spawn_driver/1, spawn_executable/1, close_deaf_port/1, + port_setget_data/1, unregister_name/1, parallelism_option/1]). -export([do_iter_max_ports/2]). @@ -115,6 +116,7 @@ all() -> mix_up_ports, otp_5112, otp_5119, exit_status_multi_scheduling_block, ports, spawn_driver, spawn_executable, close_deaf_port, unregister_name, + port_setget_data, parallelism_option]. groups() -> @@ -1405,6 +1407,12 @@ spawn_executable(Config) when is_list(Config) -> run_echo_args(SpaceDir,[ExactFile2,"hello world","dlrow olleh"]), [ExactFile2,"hello world","dlrow olleh"] = run_echo_args(SpaceDir,[binary, ExactFile2,"hello world","dlrow olleh"]), + + [ExactFile2,"hello \"world\"","\"dlrow\" olleh"] = + run_echo_args(SpaceDir,[binary, ExactFile2,"hello \"world\"","\"dlrow\" olleh"]), + [ExactFile2,"hello \"world\"","\"dlrow\" olleh"] = + run_echo_args(SpaceDir,[binary, ExactFile2,"hello \"world\"","\"dlrow\" olleh"]), + [ExactFile2] = run_echo_args(SpaceDir,[default]), [ExactFile2,"hello world","dlrow olleh"] = run_echo_args(SpaceDir,[switch_order,ExactFile2,"hello world", @@ -2333,6 +2341,61 @@ close_deaf_port_1(N, Cmd) -> {comment, "Could not spawn more than " ++ integer_to_list(N) ++ " OS processes."} end. +%% Test undocumented port_set_data/2 and port_get_data/1 +%% Hammer from multiple processes a while +%% and then abrubtly close the port (OTP-12208). +port_setget_data(Config) when is_list(Config) -> + ok = load_driver(?config(data_dir, Config), "echo_drv"), + Port = erlang:open_port({spawn_driver, "echo_drv"}, []), + + NSched = erlang:system_info(schedulers_online), + HeapData = {1,2,3,<<"A heap binary">>,fun()->"This is fun"end, + list_to_binary(lists:seq(1,100))}, + PRs = lists:map(fun(I) -> + spawn_opt(fun() -> port_setget_data_hammer(Port,HeapData,false,1) end, + [monitor, {scheduler, I rem NSched}]) + end, + lists:seq(1,10)), + receive after 100 -> ok end, + Papa = self(), + lists:foreach(fun({Pid,_}) -> Pid ! {Papa,prepare_for_close} end, PRs), + lists:foreach(fun({Pid,_}) -> + receive {Pid,prepare_for_close} -> ok end + end, + PRs), + port_close(Port), + lists:foreach(fun({Pid,Ref}) -> + receive {'DOWN', Ref, process, Pid, normal} -> ok end + end, + PRs), + ok. + +port_setget_data_hammer(Port, HeapData, IsSet0, N) -> + Rand = random:uniform(3), + IsSet1 = try case Rand of + 1 -> true = erlang:port_set_data(Port, atom), true; + 2 -> true = erlang:port_set_data(Port, HeapData), true; + 3 -> case erlang:port_get_data(Port) of + atom -> true; + HeapData -> true; + undefined -> false=IsSet0 + end + end + catch + error:badarg -> + true = get(prepare_for_close), + io:format("~p did ~p rounds before port closed\n", [self(), N]), + exit(normal) + end, + receive {Papa, prepare_for_close} -> + put(prepare_for_close, true), + Papa ! {self(),prepare_for_close} + after 0 -> + ok + end, + port_setget_data_hammer(Port, HeapData, IsSet1, N+1). + + wait_until(Fun) -> case catch Fun() of true -> diff --git a/erts/emulator/test/tuple_SUITE.erl b/erts/emulator/test/tuple_SUITE.erl index 46ece41096..f627eea07f 100644 --- a/erts/emulator/test/tuple_SUITE.erl +++ b/erts/emulator/test/tuple_SUITE.erl @@ -21,8 +21,9 @@ init_per_group/2,end_per_group/2, t_size/1, t_tuple_size/1, t_element/1, t_setelement/1, t_insert_element/1, t_delete_element/1, - t_list_to_tuple/1, t_tuple_to_list/1, - t_make_tuple_2/1, t_make_tuple_3/1, t_append_element/1, + t_list_to_tuple/1, t_list_to_upper_boundry_tuple/1, t_tuple_to_list/1, + t_make_tuple_2/1, t_make_upper_boundry_tuple_2/1, t_make_tuple_3/1, + t_append_element/1, t_append_element_upper_boundry/1, build_and_match/1, tuple_with_case/1, tuple_in_guard/1]). -include_lib("test_server/include/test_server.hrl"). @@ -40,8 +41,10 @@ suite() -> [{ct_hooks,[ts_install_cth]}]. all() -> [build_and_match, t_size, t_tuple_size, t_list_to_tuple, + t_list_to_upper_boundry_tuple, t_tuple_to_list, t_element, t_setelement, - t_make_tuple_2, t_make_tuple_3, t_append_element, + t_make_tuple_2, t_make_upper_boundry_tuple_2, t_make_tuple_3, + t_append_element, t_append_element_upper_boundry, t_insert_element, t_delete_element, tuple_with_case, tuple_in_guard]. @@ -49,11 +52,21 @@ groups() -> []. init_per_suite(Config) -> + A0 = case application:start(sasl) of + ok -> [sasl]; + _ -> [] + end, + A = case application:start(os_mon) of + ok -> [os_mon|A0]; + _ -> A0 + end, + [{started_apps, A}|Config]. + +end_per_suite(Config) -> + As = ?config(started_apps, Config), + lists:foreach(fun (A) -> application:stop(A) end, As), Config. -end_per_suite(_Config) -> - ok. - init_per_group(_GroupName, Config) -> Config. @@ -176,14 +189,19 @@ t_list_to_tuple(Config) when is_list(Config) -> {'EXIT', {badarg, _}} = (catch list_to_tuple(id([a|b]))), {'EXIT', {badarg, _}} = (catch list_to_tuple(id([a|b]))), - % test upper boundry, 16777215 elements - MaxSize = 1 bsl 24 - 1, - MaxTuple = list_to_tuple(lists:seq(1, MaxSize)), - MaxSize = size(MaxTuple), - {'EXIT', {badarg,_}} = (catch list_to_tuple(lists:seq(1, 1 bsl 24))), ok. +t_list_to_upper_boundry_tuple(Config) when is_list(Config) -> + sys_mem_cond_run(2048, + fun () -> + %% test upper boundry, 16777215 elements + MaxSize = 1 bsl 24 - 1, + MaxTuple = list_to_tuple(lists:seq(1, MaxSize)), + MaxSize = size(MaxTuple), + ok + end). + %% Tests tuple_to_list/1. t_tuple_to_list(Config) when is_list(Config) -> @@ -214,8 +232,6 @@ t_make_tuple_2(Config) when is_list(Config) -> t_make_tuple1({a}), t_make_tuple1(erlang:make_tuple(400, [])), - % test upper boundry, 16777215 elements - t_make_tuple(1 bsl 24 - 1, a), {'EXIT', {badarg,_}} = (catch erlang:make_tuple(1 bsl 24, a)), {'EXIT', {badarg,_}} = (catch erlang:make_tuple(-1, a)), @@ -225,6 +241,13 @@ t_make_tuple_2(Config) when is_list(Config) -> {'EXIT', {badarg,_}} = (catch erlang:make_tuple(1 bsl 65 + 3, a)), ok. +t_make_upper_boundry_tuple_2(Config) when is_list(Config) -> + sys_mem_cond_run(2048, + fun () -> + %% test upper boundry, 16777215 elements + t_make_tuple(1 bsl 24 - 1, a) + end). + t_make_tuple1(Element) -> lists:foreach(fun(Size) -> t_make_tuple(Size, Element) end, [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 255, 256, 511, 512, 999, @@ -309,13 +332,17 @@ t_delete_element(Config) when is_list(Config) -> %% Tests the append_element/2 BIF. t_append_element(Config) when is_list(Config) -> - ok = t_append_element({}, 2048, 2048), - - % test upper boundry, 16777215 elements - MaxSize = 1 bsl 24 - 1, - MaxTuple = list_to_tuple(lists:seq(1, MaxSize)), - {'EXIT',{badarg,_}} = (catch erlang:append_element(MaxTuple, a)), - ok. + ok = t_append_element({}, 2048, 2048). + +t_append_element_upper_boundry(Config) when is_list(Config) -> + sys_mem_cond_run(2048, + fun () -> + %% test upper boundry, 16777215 elements + MaxSize = 1 bsl 24 - 1, + MaxTuple = list_to_tuple(lists:seq(1, MaxSize)), + {'EXIT',{badarg,_}} = (catch erlang:append_element(MaxTuple, a)), + ok + end). t_append_element(_Tuple, 0, _High) -> ok; t_append_element(Tuple, N, High) -> @@ -371,3 +398,31 @@ tuple_in_guard(Config) when is_list(Config) -> %% Use this function to avoid compile-time evaluation of an expression. id(I) -> I. + +sys_mem_cond_run(ReqSizeMB, TestFun) when is_integer(ReqSizeMB) -> + case total_memory() of + TotMem when is_integer(TotMem), TotMem >= ReqSizeMB -> + TestFun(); + TotMem when is_integer(TotMem) -> + {skipped, "Not enough memory ("++integer_to_list(TotMem)++" MB)"}; + undefined -> + {skipped, "Could not retrieve memory information"} + end. + + +total_memory() -> + %% Totat memory in MB. + try + MemoryData = memsup:get_system_memory_data(), + case lists:keysearch(total_memory, 1, MemoryData) of + {value, {total_memory, TM}} -> + TM div (1024*1024); + false -> + {value, {system_total_memory, STM}} = + lists:keysearch(system_total_memory, 1, MemoryData), + STM div (1024*1024) + end + catch + _ : _ -> + undefined + end. diff --git a/erts/emulator/test/z_SUITE.erl b/erts/emulator/test/z_SUITE.erl index 4b3075a164..b0c6224dfe 100644 --- a/erts/emulator/test/z_SUITE.erl +++ b/erts/emulator/test/z_SUITE.erl @@ -38,7 +38,7 @@ -export([schedulers_alive/1, node_container_refc_check/1, long_timers/1, pollset_size/1, - check_io_debug/1]). + check_io_debug/1, get_check_io_info/0]). -define(DEFAULT_TIMEOUT, ?t:minutes(5)). @@ -288,11 +288,14 @@ check_io_debug(Config) when is_list(Config) -> end. check_io_debug_test() -> + ?line erlang:display(get_check_io_info()), ?line erts_debug:set_internal_state(available_internal_state, true), - ?line erlang:display(erlang:system_info(check_io)), - ?line NoOfErrorFds = erts_debug:get_internal_state(check_io_debug), + ?line {NoErrorFds, NoUsedFds, NoDrvSelStructs, NoDrvEvStructs} + = erts_debug:get_internal_state(check_io_debug), ?line erts_debug:set_internal_state(available_internal_state, false), - ?line 0 = NoOfErrorFds, + ?line 0 = NoErrorFds, + ?line NoUsedFds = NoDrvSelStructs, + ?line 0 = NoDrvEvStructs, ?line ok. @@ -305,7 +308,7 @@ display_check_io(ChkIo) -> catch erlang:display('--- CHECK IO INFO ---'), catch erlang:display(ChkIo), catch erts_debug:set_internal_state(available_internal_state, true), - NoOfErrorFds = (catch erts_debug:get_internal_state(check_io_debug)), + NoOfErrorFds = (catch element(1, erts_debug:get_internal_state(check_io_debug))), catch erlang:display({'NoOfErrorFds', NoOfErrorFds}), catch erts_debug:set_internal_state(available_internal_state, false), catch erlang:display('--- CHECK IO INFO ---'), @@ -313,14 +316,19 @@ display_check_io(ChkIo) -> get_check_io_info() -> ChkIo = erlang:system_info(check_io), - case lists:keysearch(pending_updates, 1, ChkIo) of - {value, {pending_updates, 0}} -> + PendUpdNo = case lists:keysearch(pending_updates, 1, ChkIo) of + {value, {pending_updates, PendNo}} -> + PendNo; + false -> + 0 + end, + {value, {active_fds, ActFds}} = lists:keysearch(active_fds, 1, ChkIo), + case {PendUpdNo, ActFds} of + {0, 0} -> display_check_io(ChkIo), ChkIo; - false -> - ChkIo; _ -> - receive after 10 -> ok end, + receive after 100 -> ok end, get_check_io_info() end. diff --git a/erts/emulator/utils/make_compiler_flags b/erts/emulator/utils/make_compiler_flags index cebe8cd0c5..ca1bc47113 100755 --- a/erts/emulator/utils/make_compiler_flags +++ b/erts/emulator/utils/make_compiler_flags @@ -70,7 +70,7 @@ my($prog) = $prog[$#prog]; print "/* Warning: Do not edit this file.\n"; print " Auto-generated by '$prog'.*/\n"; -foreach(keys %constants) { +foreach (sort(keys %constants)) { print "const char* erts_build_flags_$_ = \"$constants{$_}\";\n" } diff --git a/erts/emulator/valgrind/suppress.patched.3.6.0 b/erts/emulator/valgrind/suppress.patched.3.6.0 index b3507bdba7..f79e3ff634 100644 --- a/erts/emulator/valgrind/suppress.patched.3.6.0 +++ b/erts/emulator/valgrind/suppress.patched.3.6.0 @@ -273,6 +273,11 @@ obj:*/ssleay.* fun:AES_cbc_encrypt ... } +{ + crypto RC4 can do harmless word aligned read past end of input + Memcheck:Addr8 + fun:RC4 +} { erts_bits_init_state; Why is this needed? diff --git a/erts/emulator/valgrind/suppress.standard b/erts/emulator/valgrind/suppress.standard index a4da31a61d..b3c77119fb 100644 --- a/erts/emulator/valgrind/suppress.standard +++ b/erts/emulator/valgrind/suppress.standard @@ -260,6 +260,11 @@ obj:*/ssleay.* fun:AES_cbc_encrypt ... } +{ + crypto RC4 can do harmless word aligned read past end of input + Memcheck:Addr8 + fun:RC4 +} { Prebuilt constant terms in os_info_init (PossiblyLost) |