diff options
Diffstat (limited to 'erts/emulator/beam')
34 files changed, 3712 insertions, 1033 deletions
diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names index c2f32ba089..96547ba743 100644 --- a/erts/emulator/beam/atom.names +++ b/erts/emulator/beam/atom.names @@ -71,6 +71,7 @@ atom ac atom active atom all atom all_but_first +atom all_names atom alloc_info atom alloc_sizes atom allocated @@ -78,6 +79,7 @@ atom allocated_areas atom allocator atom allocator_sizes atom alloc_util_allocators +atom allow_gc atom allow_passive_connect atom already_loaded atom amd64 @@ -114,6 +116,7 @@ atom binary_longest_prefix_trap atom binary_longest_suffix_trap atom binary_match_trap atom binary_matches_trap +atom binary_to_term_trap atom block atom blocked atom bm @@ -322,6 +325,8 @@ atom low atom Lt='<' atom machine atom match +atom match_limit +atom match_limit_recursion atom match_spec atom max atom maximum @@ -351,11 +356,13 @@ atom multi_scheduling atom multiline atom name atom named_table +atom namelist atom native_addresses atom Neq='=/=' atom Neqeq='/=' atom net_kernel atom net_kernel_terminated +atom never_utf atom new atom new_index atom new_uniq @@ -381,6 +388,7 @@ atom nosuspend atom no_float atom no_integer atom no_network +atom no_start_optimize atom not atom not_a_list atom not_loaded @@ -391,6 +399,7 @@ atom notalive atom notbol atom noteol atom notempty +atom notempty_atstart atom notify atom notsup atom nouse_stdio @@ -475,6 +484,7 @@ atom register atom registered_name atom reload atom rem +atom report_errors atom reset atom restart atom return_from @@ -557,6 +567,7 @@ atom true atom tuple atom type atom ucompile +atom ucp atom undef atom ungreedy atom unicode diff --git a/erts/emulator/beam/beam_bif_load.c b/erts/emulator/beam/beam_bif_load.c index 649594a334..3f92c5b025 100644 --- a/erts/emulator/beam/beam_bif_load.c +++ b/erts/emulator/beam/beam_bif_load.c @@ -36,7 +36,7 @@ #include "erl_thr_progress.h" static void set_default_trace_pattern(Eterm module); -static Eterm check_process_code(Process* rp, Module* modp); +static Eterm check_process_code(Process* rp, Module* modp, int allow_gc, int *redsp); static void delete_code(Module* modp); static void decrement_refc(BeamInstr* code); static int is_native(BeamInstr* code); @@ -427,69 +427,82 @@ check_old_code_1(BIF_ALIST_1) } Eterm -check_process_code_2(BIF_ALIST_2) +erts_check_process_code(Process *c_p, Eterm module, int allow_gc, int *redsp) { - Process* rp; Module* modp; + Eterm res; + ErtsCodeIndex code_ix; - if (is_not_atom(BIF_ARG_2)) { - goto error; - } - if (is_internal_pid(BIF_ARG_1)) { - Eterm res; - ErtsCodeIndex code_ix; - - code_ix = erts_active_code_ix(); - modp = erts_get_module(BIF_ARG_2, code_ix); - if (modp == NULL) { /* Doesn't exist. */ - return am_false; - } - erts_rlock_old_code(code_ix); - if (modp->old.code == NULL) { /* No old code. */ - erts_runlock_old_code(code_ix); - return am_false; - } - erts_runlock_old_code(code_ix); - -#ifdef ERTS_SMP - rp = erts_pid2proc_suspend(BIF_P, ERTS_PROC_LOCK_MAIN, - BIF_ARG_1, ERTS_PROC_LOCK_MAIN); -#else - rp = erts_pid2proc(BIF_P, 0, BIF_ARG_1, 0); -#endif - if (!rp) { - BIF_RET(am_false); - } - if (rp == ERTS_PROC_LOCK_BUSY) { - ERTS_BIF_YIELD2(bif_export[BIF_check_process_code_2], BIF_P, - BIF_ARG_1, BIF_ARG_2); - } - erts_rlock_old_code(code_ix); - if (modp->old.code != NULL) { /* must check again */ - res = check_process_code(rp, modp); - } - else { - res = am_false; - } - erts_runlock_old_code(code_ix); -#ifdef ERTS_SMP - if (BIF_P != rp) { - erts_resume(rp, ERTS_PROC_LOCK_MAIN); - erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_MAIN); + (*redsp)++; + + ASSERT(is_atom(module)); + + code_ix = erts_active_code_ix(); + modp = erts_get_module(module, code_ix); + if (!modp) + return am_false; + erts_rlock_old_code(code_ix); + res = modp->old.code ? check_process_code(c_p, modp, allow_gc, redsp) : am_false; + erts_runlock_old_code(code_ix); + + return res; +} + +BIF_RETTYPE erts_internal_check_process_code_2(BIF_ALIST_2) +{ + int reds = 0; + Eterm res; + Eterm olist = BIF_ARG_2; + int allow_gc = 1; + + if (is_not_atom(BIF_ARG_1)) + goto badarg; + + while (is_list(olist)) { + Eterm *lp = list_val(olist); + Eterm opt = CAR(lp); + if (is_tuple(opt)) { + Eterm* tp = tuple_val(opt); + switch (arityval(tp[0])) { + case 2: + switch (tp[1]) { + case am_allow_gc: + switch (tp[2]) { + case am_false: + allow_gc = 0; + break; + case am_true: + allow_gc = 1; + break; + default: + goto badarg; + } + break; + default: + goto badarg; + } + break; + default: + goto badarg; + } } -#endif - BIF_RET(res); - } - else if (is_external_pid(BIF_ARG_1) - && external_pid_dist_entry(BIF_ARG_1) == erts_this_dist_entry) { - BIF_RET(am_false); + else + goto badarg; + olist = CDR(lp); } + if (is_not_nil(olist)) + goto badarg; + + res = erts_check_process_code(BIF_P, BIF_ARG_1, allow_gc, &reds); + + ASSERT(is_value(res)); + + BIF_RET2(res, reds); - error: +badarg: BIF_ERROR(BIF_P, BADARG); } - BIF_RETTYPE delete_module_1(BIF_ALIST_1) { ErtsCodeIndex code_ix; @@ -710,7 +723,7 @@ set_default_trace_pattern(Eterm module) } static Eterm -check_process_code(Process* rp, Module* modp) +check_process_code(Process* rp, Module* modp, int allow_gc, int *redsp) { BeamInstr* start; char* mod_start; @@ -773,6 +786,16 @@ check_process_code(Process* rp, Module* modp) } } + if (rp->flags & F_DISABLE_GC) { + /* + * Cannot proceed. Process has disabled gc in order to + * safely leave inconsistent data on the heap and/or + * off heap lists. Need to wait for gc to be enabled + * again. + */ + return THE_NON_VALUE; + } + /* * See if there are funs that refer to the old version of the module. */ @@ -786,6 +809,8 @@ check_process_code(Process* rp, Module* modp) if (done_gc) { return am_true; } else { + if (!allow_gc) + return am_aborted; /* * Try to get rid of this fun by garbage collecting. * Clear both fvalue and ftrace to make sure they @@ -796,7 +821,7 @@ check_process_code(Process* rp, Module* modp) rp->ftrace = NIL; done_gc = 1; FLAGS(rp) |= F_NEED_FULLSWEEP; - (void) erts_garbage_collect(rp, 0, rp->arg_reg, rp->arity); + *redsp += erts_garbage_collect(rp, 0, rp->arg_reg, rp->arity); goto rescan; } } @@ -850,6 +875,9 @@ check_process_code(Process* rp, Module* modp) Uint lit_size; struct erl_off_heap_header* oh; + if (!allow_gc) + return am_aborted; + /* * Try to get rid of constants by by garbage collecting. * Clear both fvalue and ftrace. @@ -859,11 +887,12 @@ check_process_code(Process* rp, Module* modp) rp->ftrace = NIL; done_gc = 1; FLAGS(rp) |= F_NEED_FULLSWEEP; - (void) erts_garbage_collect(rp, 0, rp->arg_reg, rp->arity); + *redsp += erts_garbage_collect(rp, 0, rp->arg_reg, rp->arity); literals = (Eterm *) modp->old.code[MI_LITERALS_START]; lit_size = (Eterm *) modp->old.code[MI_LITERALS_END] - literals; oh = (struct erl_off_heap_header *) modp->old.code[MI_LITERALS_OFF_HEAP]; + *redsp += lit_size / 10; /* Need, better value... */ erts_garbage_collect_literals(rp, literals, lit_size, oh); } } diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index 78ab6fa30f..b413f0e859 100644 --- a/erts/emulator/beam/beam_emu.c +++ b/erts/emulator/beam/beam_emu.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 @@ -48,7 +48,7 @@ # define OpCase(OpCode) case op_##OpCode # define CountCase(OpCode) case op_count_##OpCode # define OpCode(OpCode) ((Uint*)op_##OpCode) -# define Goto(Rel) {Go = (int)(Rel); goto emulator_loop;} +# define Goto(Rel) {Go = (int)(UWord)(Rel); goto emulator_loop;} # define LabelAddr(Addr) &&##Addr #else # define OpCase(OpCode) lb_##OpCode @@ -133,7 +133,7 @@ do { \ /* We don't check the range if an ordinary switch is used */ #ifdef NO_JUMP_TABLE -#define VALID_INSTR(IP) (0 <= (int)(IP) && ((int)(IP) < (NUMBER_OF_OPCODES*2+10))) +#define VALID_INSTR(IP) ((UWord)(IP) < (NUMBER_OF_OPCODES*2+10)) #else #define VALID_INSTR(IP) \ ((SWord)LabelAddr(emulator_loop) <= (SWord)(IP) && \ @@ -4326,7 +4326,19 @@ void process_main(void) flags = Arg(2); BsGetFieldSize(tmp_arg2, (flags >> 3), ClauseFail(), size); if (size >= SMALL_BITS) { - Uint wordsneeded = 1+WSIZE(NBYTES((Uint) size)); + Uint wordsneeded; + /* check bits size before potential gc. + * We do not want a gc and then realize we don't need + * the allocated space (i.e. if the op fails) + * + * remember to reacquire the matchbuffer after gc. + */ + + mb = ms_matchbuffer(tmp_arg1); + if (mb->size - mb->offset < size) { + ClauseFail(); + } + wordsneeded = 1+WSIZE(NBYTES((Uint) size)); TestHeapPreserve(wordsneeded, Arg(1), tmp_arg1); } mb = ms_matchbuffer(tmp_arg1); diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c index 13d31285b2..61c1abedb5 100644 --- a/erts/emulator/beam/bif.c +++ b/erts/emulator/beam/bif.c @@ -3764,45 +3764,6 @@ BIF_RETTYPE now_0(BIF_ALIST_0) /**********************************************************************/ -BIF_RETTYPE garbage_collect_1(BIF_ALIST_1) -{ - int reds; - Process *rp; - - if (is_not_pid(BIF_ARG_1)) { - BIF_ERROR(BIF_P, BADARG); - } - - if (BIF_P->common.id == BIF_ARG_1) - rp = BIF_P; - else { -#ifdef ERTS_SMP - rp = erts_pid2proc_suspend(BIF_P, ERTS_PROC_LOCK_MAIN, - BIF_ARG_1, ERTS_PROC_LOCK_MAIN); - if (rp == ERTS_PROC_LOCK_BUSY) - ERTS_BIF_YIELD1(bif_export[BIF_garbage_collect_1], BIF_P, BIF_ARG_1); -#else - rp = erts_proc_lookup(BIF_ARG_1); -#endif - if (!rp) - BIF_RET(am_false); - } - - /* The GC cost is taken for the process executing this BIF. */ - - FLAGS(rp) |= F_NEED_FULLSWEEP; - reds = erts_garbage_collect(rp, 0, rp->arg_reg, rp->arity); - -#ifdef ERTS_SMP - if (BIF_P != rp) { - erts_resume(rp, ERTS_PROC_LOCK_MAIN); - erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_MAIN); - } -#endif - - BIF_RET2(am_true, reds); -} - BIF_RETTYPE garbage_collect_0(BIF_ALIST_0) { int reds; @@ -4527,7 +4488,7 @@ BIF_RETTYPE system_flag_2(BIF_ALIST_2) BIF_P->group_leader, "A call to erlang:system_flag(cpu_topology, _) was made.\n" "The cpu_topology argument is deprecated and scheduled\n" - "for removal in erts-5.10/OTP-R16. For more information\n" + "for removal in Erlang/OTP 18. For more information\n" "see the erlang:system_flag/2 documentation.\n"); BIF_TRAP1(set_cpu_topology_trap, BIF_P, BIF_ARG_2); } else if (ERTS_IS_ATOM_STR("scheduler_bind_type", BIF_ARG_1)) { @@ -4535,7 +4496,7 @@ BIF_RETTYPE system_flag_2(BIF_ALIST_2) BIF_P->group_leader, "A call to erlang:system_flag(scheduler_bind_type, _) was\n" "made. The scheduler_bind_type argument is deprecated and\n" - "scheduled for removal in erts-5.10/OTP-R16. For more\n" + "scheduled for removal in Erlang/OTP 18. For more\n" "information see the erlang:system_flag/2 documentation.\n"); return erts_bind_schedulers(BIF_P, BIF_ARG_2); } diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab index 6037c08dd8..3ec534f0bc 100644 --- a/erts/emulator/beam/bif.tab +++ b/erts/emulator/beam/bif.tab @@ -45,8 +45,6 @@ bif erlang:apply/3 bif erlang:atom_to_list/1 bif erlang:binary_to_list/1 bif erlang:binary_to_list/3 -bif erlang:binary_to_term/1 -bif erlang:check_process_code/2 bif erlang:crc32/1 bif erlang:crc32/2 bif erlang:crc32_combine/3 @@ -67,7 +65,6 @@ bif erlang:float_to_list/1 bif erlang:float_to_list/2 bif erlang:fun_info/2 bif erlang:garbage_collect/0 -bif erlang:garbage_collect/1 bif erlang:get/0 bif erlang:get/1 bif erlang:get_keys/1 @@ -154,6 +151,12 @@ bif erts_internal:port_command/3 bif erts_internal:port_control/3 bif erts_internal:port_close/1 bif erts_internal:port_connect/2 +bif erts_internal:binary_to_term/1 +bif erts_internal:binary_to_term/2 + +bif erts_internal:request_system_task/3 +bif erts_internal:check_process_code/2 + # inet_db support bif erlang:port_set_data/2 @@ -477,11 +480,6 @@ bif erlang:call_on_load_function/1 bif erlang:finish_after_on_load/2 # -# New Bifs in R13B4 -# -bif erlang:binary_to_term/2 - -# # The binary match bifs (New in R14A - EEP9) # @@ -575,6 +573,12 @@ bif io:printable_range/0 bif os:unsetenv/1 # +# New in R17A +# + +bif re:inspect/2 + +# # Obsolete # diff --git a/erts/emulator/beam/binary.c b/erts/emulator/beam/binary.c index 33abac2f3d..c7926f18af 100644 --- a/erts/emulator/beam/binary.c +++ b/erts/emulator/beam/binary.c @@ -447,6 +447,7 @@ BIF_RETTYPE bitstring_to_list_1(BIF_ALIST_1) BIF_RETTYPE erts_list_to_binary_bif(Process *p, Eterm arg) { Eterm bin; + Eterm h,t; ErlDrvSizeT size; byte* bytes; #ifdef DEBUG @@ -459,6 +460,16 @@ BIF_RETTYPE erts_list_to_binary_bif(Process *p, Eterm arg) if (is_not_list(arg)) { goto error; } + /* check for [binary()] case */ + h = CAR(list_val(arg)); + t = CDR(list_val(arg)); + if (is_binary(h) && is_nil(t) && !( + HEADER_SUB_BIN == *(binary_val(h)) && ( + ((ErlSubBin *)binary_val(h))->bitoffs != 0 || + ((ErlSubBin *)binary_val(h))->bitsize != 0 + ))) { + return h; + } switch (erts_iolist_size(arg, &size)) { case ERTS_IOLIST_OVERFLOW: BIF_ERROR(p, SYSTEM_LIMIT); case ERTS_IOLIST_TYPE: goto error; diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c index b7e1092907..7d4f52ee23 100644 --- a/erts/emulator/beam/break.c +++ b/erts/emulator/beam/break.c @@ -112,10 +112,12 @@ process_killer(void) erts_smp_proc_lock(rp, rp_locks); state = erts_smp_atomic32_read_acqb(&rp->state); if (state & (ERTS_PSFLG_FREE - | ERTS_PSFLG_EXITING - | ERTS_PSFLG_ACTIVE - | ERTS_PSFLG_IN_RUNQ - | ERTS_PSFLG_RUNNING)) { + | ERTS_PSFLG_EXITING + | ERTS_PSFLG_ACTIVE + | ERTS_PSFLG_ACTIVE_SYS + | ERTS_PSFLG_IN_RUNQ + | ERTS_PSFLG_RUNNING + | ERTS_PSFLG_RUNNING_SYS)) { erts_printf("Can only kill WAITING processes this way\n"); } else { diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c index b5ba9bb94a..8094c6ee2e 100644 --- a/erts/emulator/beam/erl_alloc.c +++ b/erts/emulator/beam/erl_alloc.c @@ -75,9 +75,9 @@ #define ERTS_ALC_DEFAULT_ENABLED_ACUL_EHEAP_ALLOC 45 #define ERTS_ALC_DEFAULT_ENABLED_ACUL_LL_ALLOC 85 -#define ERTS_ALC_DEFAULT_ACUL 0 -#define ERTS_ALC_DEFAULT_ACUL_EHEAP_ALLOC 0 -#define ERTS_ALC_DEFAULT_ACUL_LL_ALLOC 0 +#define ERTS_ALC_DEFAULT_ACUL ERTS_ALC_DEFAULT_ENABLED_ACUL +#define ERTS_ALC_DEFAULT_ACUL_EHEAP_ALLOC ERTS_ALC_DEFAULT_ENABLED_ACUL_EHEAP_ALLOC +#define ERTS_ALC_DEFAULT_ACUL_LL_ALLOC ERTS_ALC_DEFAULT_ENABLED_ACUL_LL_ALLOC #ifndef ERTS_SMP # undef ERTS_ALC_DEFAULT_ACUL diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types index bb5eba80be..b4e52770e3 100644 --- a/erts/emulator/beam/erl_alloc.types +++ b/erts/emulator/beam/erl_alloc.types @@ -1,7 +1,7 @@ # # %CopyrightBegin% # -# Copyright Ericsson AB 2003-2013. All Rights Reserved. +# Copyright Ericsson AB 2003-2014. 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 @@ -150,7 +150,7 @@ type LINK_LH STANDARD PROCESSES link_lh type SUSPEND_MON STANDARD PROCESSES suspend_monitor type PEND_SUSPEND SHORT_LIVED PROCESSES pending_suspend type PROC_LIST SHORT_LIVED PROCESSES proc_list -type EXTRA_ROOT SHORT_LIVED PROCESSES extra_root +type SAVED_ESTACK SHORT_LIVED PROCESSES saved_estack type FUN_ENTRY LONG_LIVED CODE fun_entry type ATOM_TXT LONG_LIVED ATOM atom_text type BEAM_REGISTER EHEAP PROCESSES beam_register @@ -268,6 +268,8 @@ 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 +if threads_no_smp # Need thread safe allocs, but std_alloc and fix_alloc are not; diff --git a/erts/emulator/beam/erl_bif_ddll.c b/erts/emulator/beam/erl_bif_ddll.c index 1c3e955f47..1728b200f7 100644 --- a/erts/emulator/beam/erl_bif_ddll.c +++ b/erts/emulator/beam/erl_bif_ddll.c @@ -182,7 +182,7 @@ BIF_RETTYPE erl_ddll_try_load_3(BIF_ALIST_3) Eterm name_term = BIF_ARG_2; Eterm options = BIF_ARG_3; char *path = NULL; - ErlDrvSizeT path_len; + Sint path_len; char *name = NULL; DE_Handle *dh; erts_driver_t *drv; @@ -198,6 +198,7 @@ BIF_RETTYPE erl_ddll_try_load_3(BIF_ALIST_3) int kill_ports = 0; int do_build_load_error = 0; int build_this_load_error = 0; + int encoding; for(l = options; is_list(l); l = CDR(list_val(l))) { Eterm opt = CAR(list_val(l)); @@ -257,18 +258,23 @@ BIF_RETTYPE erl_ddll_try_load_3(BIF_ALIST_3) goto error; } - if (erts_iolist_size(path_term, &path_len)) { - goto error; + encoding = erts_get_native_filename_encoding(); + if (encoding == ERL_FILENAME_WIN_WCHAR) { + /* Do not convert the lib name to utf-16le yet, do that in win32 specific code */ + /* since lib_name is used in error messages */ + encoding = ERL_FILENAME_UTF8; } - path = erts_alloc(ERTS_ALC_T_DDLL_TMP_BUF, path_len + 1 /* might need path separator */ + sys_strlen(name) + 1); - if (erts_iolist_to_buf(path_term, path, path_len) != 0) { + path = erts_convert_filename_to_encoding(path_term, NULL, 0, + ERTS_ALC_T_DDLL_TMP_BUF, 1, 0, + encoding, &path_len, + sys_strlen(name) + 2); /* might need path separator */ + if (!path) { goto error; } - while (path_len > 0 && (path[path_len-1] == '\\' || path[path_len-1] == '/')) { - --path_len; - } + ASSERT(path_len > 0 && path[path_len-1] == 0); + while (--path_len > 0 && (path[path_len-1] == '\\' || path[path_len-1] == '/')) + ; path[path_len++] = '/'; - /*path[path_len] = '\0';*/ sys_strcpy(path+path_len,name); #if DDLL_SMP @@ -1524,7 +1530,7 @@ static int do_load_driver_entry(DE_Handle *dh, char *path, char *name) assert_drv_list_rwlocked(); - if ((res = erts_sys_ddll_open(path, &(dh->handle))) != ERL_DE_NO_ERROR) { + if ((res = erts_sys_ddll_open(path, &(dh->handle), NULL)) != ERL_DE_NO_ERROR) { return res; } diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index d7f1e2d971..e0b654cb22 100755 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -64,9 +64,11 @@ static Export *gather_gc_info_res_trap; #define DECL_AM(S) Eterm AM_ ## S = am_atom_put(#S, sizeof(#S) - 1) +static char otp_correction_package[] = ERLANG_OTP_CORRECTION_PACKAGE; /* Keep erts_system_version as a global variable for easy access from a core */ -static char erts_system_version[] = ("Erlang " ERLANG_OTP_RELEASE - " (erts-" ERLANG_VERSION ")" +static char erts_system_version[] = ("Erlang/OTP " ERLANG_OTP_RELEASE + "%s" + " [erts-" ERLANG_VERSION "]" #if !HEAP_ON_C_STACK && !HALFWORD_HEAP " [no-c-stack-objects]" #endif @@ -304,11 +306,28 @@ make_link_list(Process *p, ErtsLink *root, Eterm tail) int erts_print_system_version(int to, void *arg, Process *c_p) { + int i, rc = -1; + char *rc_str = ""; + char rc_buf[100]; + char *ocp = otp_correction_package; #ifdef ERTS_SMP Uint total, online, active; (void) erts_schedulers_state(&total, &online, &active, 0); #endif - return erts_print(to, arg, erts_system_version + for (i = 0; i < sizeof(otp_correction_package)-4; i++) { + if (ocp[i] == '-' && ocp[i+1] == 'r' && ocp[i+2] == 'c') + rc = atoi(&ocp[i+3]); + } + if (rc >= 0) { + if (rc == 0) + rc_str = " [DEVELOPMENT]"; + else { + erts_snprintf(rc_buf, sizeof(rc_buf), " [RELEASE CANDIDATE %d]", rc); + rc_str = rc_buf; + } + } + return erts_print(to, arg, erts_system_version, + rc_str #ifdef ERTS_SMP , total, online #endif @@ -2417,6 +2436,10 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) DECL_AM(unknown); BIF_RET(AM_unknown); } + } else if (ERTS_IS_ATOM_STR("otp_correction_package", BIF_ARG_1)) { + int n = sizeof(ERLANG_OTP_CORRECTION_PACKAGE)-1; + hp = HAlloc(BIF_P, 2*n); + BIF_RET(buf_to_intlist(&hp, ERLANG_OTP_CORRECTION_PACKAGE, n, NIL)); } else if (ERTS_IS_ATOM_STR("otp_release", BIF_ARG_1)) { int n = sizeof(ERLANG_OTP_RELEASE)-1; hp = HAlloc(BIF_P, 2*n); @@ -3603,6 +3626,20 @@ BIF_RETTYPE erts_debug_set_internal_state_2(BIF_ALIST_2) BIF_RET(am_true); } } + else if (ERTS_IS_ATOM_STR("gc_state", BIF_ARG_1)) { + /* Used by process_SUITE (emulator) */ + int res, enable; + + switch (BIF_ARG_2) { + case am_true: enable = 1; break; + case am_false: enable = 0; break; + default: BIF_ERROR(BIF_P, BADARG); break; + } + + res = (BIF_P->flags & F_DISABLE_GC) ? am_false : am_true; + erts_set_gc_state(BIF_P, enable); + BIF_RET(res); + } else if (ERTS_IS_ATOM_STR("send_fake_exit_signal", BIF_ARG_1)) { /* Used by signal_SUITE (emulator) */ diff --git a/erts/emulator/beam/erl_bif_port.c b/erts/emulator/beam/erl_bif_port.c index 3cd53ef65d..f298422267 100644 --- a/erts/emulator/beam/erl_bif_port.c +++ b/erts/emulator/beam/erl_bif_port.c @@ -796,43 +796,29 @@ open_port(Process* p, Eterm name, Eterm settings, int *err_typep, int *err_nump) goto badarg; } - if (*tp == am_spawn || *tp == am_spawn_driver) { /* A process port */ + if (*tp == am_spawn || *tp == am_spawn_driver || *tp == am_spawn_executable) { /* A process port */ + int encoding; if (arity != make_arityval(2)) { goto badarg; } name = tp[1]; - if (is_atom(name)) { - name_buf = (char *) erts_alloc(ERTS_ALC_T_TMP, - atom_tab(atom_val(name))->len+1); - sys_memcpy((void *) name_buf, - (void *) atom_tab(atom_val(name))->name, - atom_tab(atom_val(name))->len); - name_buf[atom_tab(atom_val(name))->len] = '\0'; - } else if ((i = is_string(name))) { - name_buf = (char *) erts_alloc(ERTS_ALC_T_TMP, i + 1); - if (intlist_to_buf(name, name_buf, i) != i) - erl_exit(1, "%s:%d: Internal error\n", __FILE__, __LINE__); - name_buf[i] = '\0'; - } else { + encoding = erts_get_native_filename_encoding(); + /* Do not convert the command to utf-16le yet, do that in win32 specific code */ + /* since the cmd is used for comparsion with drivers names and copied to port info */ + if (encoding == ERL_FILENAME_WIN_WCHAR) { + encoding = ERL_FILENAME_UTF8; + } + if ((name_buf = erts_convert_filename_to_encoding(name, NULL, 0, ERTS_ALC_T_TMP,0,1, encoding, NULL, 0)) + == NULL) { goto badarg; } + if (*tp == am_spawn_driver) { opts.spawn_type = ERTS_SPAWN_DRIVER; + } else if (*tp == am_spawn_executable) { + opts.spawn_type = ERTS_SPAWN_EXECUTABLE; } - driver = &spawn_driver; - } else if (*tp == am_spawn_executable) { /* A program */ - /* - * {spawn_executable,Progname} - */ - - if (arity != make_arityval(2)) { - goto badarg; - } - name = tp[1]; - if ((name_buf = erts_convert_filename_to_native(name, NULL, 0, ERTS_ALC_T_TMP,0,1, NULL)) == NULL) { - goto badarg; - } - opts.spawn_type = ERTS_SPAWN_EXECUTABLE; + driver = &spawn_driver; } else if (*tp == am_fd) { /* An fd port */ int n; @@ -873,29 +859,8 @@ open_port(Process* p, Eterm name, Eterm settings, int *err_typep, int *err_nump) } if (edir != NIL) { - /* A working directory is expressed differently if spawn_executable, i.e. Unicode is handles - for spawn_executable... */ - if (opts.spawn_type != ERTS_SPAWN_EXECUTABLE) { - Eterm iolist; - DeclareTmpHeap(heap,4,p); - int r; - - UseTmpHeap(4,p); - heap[0] = edir; - heap[1] = make_list(heap+2); - heap[2] = make_small(0); - heap[3] = NIL; - iolist = make_list(heap); - r = erts_iolist_to_buf(iolist, (char*) dir, MAXPATHLEN); - UnUseTmpHeap(4,p); - if (ERTS_IOLIST_TO_BUF_FAILED(r)) { - goto badarg; - } - opts.wd = (char *) dir; - } else { - if ((opts.wd = erts_convert_filename_to_native(edir, NULL, 0, ERTS_ALC_T_TMP,0,1,NULL)) == NULL) { - goto badarg; - } + if ((opts.wd = erts_convert_filename_to_native(edir, NULL, 0, ERTS_ALC_T_TMP,0,1,NULL)) == NULL) { + goto badarg; } } @@ -973,11 +938,11 @@ static char **convert_args(Eterm l) int n; int i = 0; Eterm str; - /* We require at least one element in list (argv[0]) */ if (is_not_list(l) && is_not_nil(l)) { return NULL; } n = list_length(l); + /* We require at least one element in argv[0] + NULL at end */ pp = erts_alloc(ERTS_ALC_T_TMP, (n + 2) * sizeof(char **)); pp[i++] = erts_default_arg0; while (is_list(l)) { diff --git a/erts/emulator/beam/erl_bif_re.c b/erts/emulator/beam/erl_bif_re.c index 3d34c2a77f..448c6f6f6d 100644 --- a/erts/emulator/beam/erl_bif_re.c +++ b/erts/emulator/beam/erl_bif_re.c @@ -180,6 +180,9 @@ static Eterm make_signed_integer(int x, Process *p) #define PARSE_FLAG_STARTOFFSET 8 #define PARSE_FLAG_CAPTURE_OPT 16 #define PARSE_FLAG_GLOBAL 32 +#define PARSE_FLAG_REPORT_ERRORS 64 +#define PARSE_FLAG_MATCH_LIMIT 128 +#define PARSE_FLAG_MATCH_LIMIT_RECURSION 256 #define CAPSPEC_VALUES 0 #define CAPSPEC_TYPE 1 @@ -192,7 +195,9 @@ parse_options(Eterm listp, /* in */ int *exec_options, /* out */ int *flags,/* out */ int *startoffset, /* out */ - Eterm *capture_spec) /* capture_spec[CAPSPEC_SIZE] */ /* out */ + Eterm *capture_spec, /* capture_spec[CAPSPEC_SIZE] */ /* out */ + int *match_limit, /* out */ + int *match_limit_recursion) /* out */ { int copt,eopt,fl; Eterm item; @@ -234,7 +239,7 @@ parse_options(Eterm listp, /* in */ case am_offset: { int tmp; - if (!term_to_int(tp[2],&tmp)) { + if (!term_to_int(tp[2],&tmp) || tmp < 0) { return -1; } if (startoffset != NULL) { @@ -243,6 +248,31 @@ parse_options(Eterm listp, /* in */ } fl |= (PARSE_FLAG_UNIQUE_EXEC_OPT|PARSE_FLAG_STARTOFFSET); break; + case am_match_limit: + { + int tmp; + if (!term_to_int(tp[2],&tmp) || tmp < 0) { + return -1; + } + if (match_limit != NULL) { + *match_limit = tmp; + } + } + fl |= (PARSE_FLAG_UNIQUE_EXEC_OPT|PARSE_FLAG_MATCH_LIMIT); + break; + case am_match_limit_recursion: + { + int tmp; + if (!term_to_int(tp[2],&tmp) || tmp < 0) { + return -1; + } + if (match_limit_recursion != NULL) { + *match_limit_recursion = tmp; + } + } + fl |= (PARSE_FLAG_UNIQUE_EXEC_OPT| + PARSE_FLAG_MATCH_LIMIT_RECURSION); + break; case am_newline: if (!is_atom(tp[2])) { return -1; @@ -276,7 +306,7 @@ parse_options(Eterm listp, /* in */ default: return -1; } - }else if (is_not_atom(item)) { + } else if (is_not_atom(item)) { return -1; } else { switch(item) { @@ -288,6 +318,10 @@ parse_options(Eterm listp, /* in */ eopt |= PCRE_NOTEMPTY; fl |= PARSE_FLAG_UNIQUE_EXEC_OPT; break; + case am_notempty_atstart: + eopt |= PCRE_NOTEMPTY_ATSTART; + fl |= PARSE_FLAG_UNIQUE_EXEC_OPT; + break; case am_notbol: eopt |= PCRE_NOTBOL; fl |= PARSE_FLAG_UNIQUE_EXEC_OPT; @@ -296,6 +330,10 @@ parse_options(Eterm listp, /* in */ eopt |= PCRE_NOTEOL; fl |= PARSE_FLAG_UNIQUE_EXEC_OPT; break; + case am_no_start_optimize: + copt |= PCRE_NO_START_OPTIMIZE; + fl |= PARSE_FLAG_UNIQUE_COMPILE_OPT; + break; case am_caseless: copt |= PCRE_CASELESS; fl |= PARSE_FLAG_UNIQUE_COMPILE_OPT; @@ -332,6 +370,18 @@ parse_options(Eterm listp, /* in */ copt |= PCRE_UNGREEDY; fl |= PARSE_FLAG_UNIQUE_COMPILE_OPT; break; + case am_ucp: + copt |= PCRE_UCP; + fl |= PARSE_FLAG_UNIQUE_COMPILE_OPT; + break; + case am_never_utf: + copt |= PCRE_NEVER_UTF; + fl |= PARSE_FLAG_UNIQUE_COMPILE_OPT; + break; + case am_report_errors: + fl |= (PARSE_FLAG_UNIQUE_EXEC_OPT | + PARSE_FLAG_REPORT_ERRORS); + break; case am_unicode: copt |= PCRE_UTF8; fl |= (PARSE_FLAG_UNIQUE_COMPILE_OPT | PARSE_FLAG_UNICODE); @@ -359,7 +409,7 @@ parse_options(Eterm listp, /* in */ if (compile_options != NULL) { *compile_options = copt; } - if (exec_options != NULL) { + if (exec_options != NULL) { *exec_options = eopt; } if (flags != NULL) { @@ -373,34 +423,49 @@ parse_options(Eterm listp, /* in */ */ static Eterm -build_compile_result(Process *p, Eterm error_tag, pcre *result, int errcode, const char *errstr, int errofset, int unicode, int with_ok) +build_compile_result(Process *p, Eterm error_tag, pcre *result, int errcode, const char *errstr, int errofset, int unicode, int with_ok, Eterm extra_err_tag) { Eterm *hp; Eterm ret; size_t pattern_size; int capture_count; + int use_crlf; + unsigned long options; if (!result) { /* Return {error_tag, {Code, String, Offset}} */ int elen = sys_strlen(errstr); int need = 3 /* tuple of 2 */ + 3 /* tuple of 2 */ + - (2 * elen) /* The error string list */; + (2 * elen) /* The error string list */ + + ((extra_err_tag != NIL) ? 3 : 0); hp = HAlloc(p, need); ret = buf_to_intlist(&hp, (char *) errstr, elen, NIL); ret = TUPLE2(hp, ret, make_small(errofset)); hp += 3; + if (extra_err_tag != NIL) { + /* Return {error_tag, {extra_tag, + {Code, String, Offset}}} instead */ + ret = TUPLE2(hp, extra_err_tag, ret); + hp += 3; + } ret = TUPLE2(hp, error_tag, ret); } else { erts_pcre_fullinfo(result, NULL, PCRE_INFO_SIZE, &pattern_size); erts_pcre_fullinfo(result, NULL, PCRE_INFO_CAPTURECOUNT, &capture_count); + erts_pcre_fullinfo(result, NULL, PCRE_INFO_OPTIONS, &options); + options &= PCRE_NEWLINE_CR|PCRE_NEWLINE_LF | PCRE_NEWLINE_CRLF | + PCRE_NEWLINE_ANY | PCRE_NEWLINE_ANYCRLF; + use_crlf = (options == PCRE_NEWLINE_ANY || + options == PCRE_NEWLINE_CRLF || + options == PCRE_NEWLINE_ANYCRLF); /* XXX: Optimize - keep in offheap binary to allow this to be kept across traps w/o need of copying */ ret = new_binary(p, (byte *) result, pattern_size); erts_pcre_free(result); - hp = HAlloc(p, (with_ok) ? (3+5) : 5); - ret = TUPLE4(hp,am_re_pattern, make_small(capture_count), make_small(unicode),ret); + hp = HAlloc(p, (with_ok) ? (3+6) : 6); + ret = TUPLE5(hp,am_re_pattern, make_small(capture_count), make_small(unicode),make_small(use_crlf),ret); if (with_ok) { - hp += 5; + hp += 6; ret = TUPLE2(hp,am_ok,ret); } } @@ -424,9 +489,12 @@ re_compile(Process* p, Eterm arg1, Eterm arg2) int options = 0; int pflags = 0; int unicode = 0; +#ifdef DEBUG + int buffres; +#endif - if (parse_options(arg2,&options,NULL,&pflags,NULL,NULL) + if (parse_options(arg2,&options,NULL,&pflags,NULL,NULL,NULL,NULL) < 0) { BIF_ERROR(p,BADARG); } @@ -445,16 +513,19 @@ re_compile(Process* p, Eterm arg1, Eterm arg2) BIF_ERROR(p,BADARG); } expr = erts_alloc(ERTS_ALC_T_RE_TMP_BUF, slen + 1); - if (erts_iolist_to_buf(arg1, expr, slen) != 0) { - erts_free(ERTS_ALC_T_RE_TMP_BUF, expr); - BIF_ERROR(p,BADARG); - } +#ifdef DEBUG + buffres = +#endif + erts_iolist_to_buf(arg1, expr, slen); + + ASSERT(buffres >= 0); + expr[slen]='\0'; result = erts_pcre_compile2(expr, options, &errcode, &errstr, &errofset, default_table); ret = build_compile_result(p, am_error, result, errcode, - errstr, errofset, unicode, 1); + errstr, errofset, unicode, 1, NIL); erts_free(ERTS_ALC_T_RE_TMP_BUF, expr); BIF_RET(ret); } @@ -492,7 +563,7 @@ typedef struct _return_info { } ReturnInfo; typedef struct _restart_context { - pcre_extra extra; + erts_pcre_extra extra; void *restart_data; Uint32 flags; char *subject; /* to be able to free it when done */ @@ -502,6 +573,7 @@ typedef struct _restart_context { } RestartContext; #define RESTART_FLAG_SUBJECT_IN_BINARY 0x1 +#define RESTART_FLAG_REPORT_MATCH_LIMIT 0x2 static void cleanup_restart_context(RestartContext *rc) { @@ -542,13 +614,29 @@ static Eterm build_exec_return(Process *p, int rc, RestartContext *restartp, Ete Eterm res; Eterm *hp; if (rc <= 0) { - res = am_nomatch; + if (restartp->flags & RESTART_FLAG_REPORT_MATCH_LIMIT) { + if (rc == PCRE_ERROR_MATCHLIMIT) { + hp = HAlloc(p,3); + res = TUPLE2(hp,am_error,am_match_limit); + } else if (rc == PCRE_ERROR_RECURSIONLIMIT) { + hp = HAlloc(p,3); + res = TUPLE2(hp,am_error,am_match_limit_recursion); + } else { + res = am_nomatch; + } + } else { + res = am_nomatch; + } } else { - ReturnInfo *ri = restartp->ret_info; + ReturnInfo *ri; ReturnInfo defri = {RetIndex,0,{0}}; - if (ri == NULL) { + + if (restartp->ret_info == NULL) { ri = &defri; + } else { + ri = restartp->ret_info; } + if (ri->type == RetNone) { res = am_match; } else if (ri->type == RetIndex){ @@ -577,6 +665,17 @@ static Eterm build_exec_return(Process *p, int rc, RestartContext *restartp, Ete ri->num_spec * 2 * sizeof(Eterm)); for (i = 0; i < ri->num_spec; ++i) { x = ri->v[i]; + if (x < -1) { + int n = i-x+1; + int j; + for (j = i+1; j < ri->num_spec && j < n; ++j) { + if (restartp->ovector[(ri->v[j])*2] >= 0) { + x = ri->v[j]; + break; + } + } + i = n-1; + } if (x < rc && x >= 0) { tmp_vect[n*2] = make_signed_integer(restartp->ovector[x*2],p); tmp_vect[n*2+1] = make_signed_integer(restartp->ovector[x*2+1]-restartp->ovector[x*2],p); @@ -658,6 +757,17 @@ static Eterm build_exec_return(Process *p, int rc, RestartContext *restartp, Ete ri->num_spec * sizeof(Eterm)); for (i = 0; i < ri->num_spec; ++i) { x = ri->v[i]; + if (x < -1) { + int n = i-x+1; + int j; + for (j = i+1; j < ri->num_spec && j < n; ++j) { + if (restartp->ovector[(ri->v[j])*2] >= 0) { + x = ri->v[j]; + break; + } + } + i = n-1; + } if (x < rc && x >= 0) { char *cp; int len; @@ -722,6 +832,49 @@ static Eterm build_exec_return(Process *p, int rc, RestartContext *restartp, Ete */ #define RINFO_SIZ(Num) (sizeof(ReturnInfo) + (sizeof(int) * (Num - 1))) +#define PICK_INDEX(NameEntry) \ + ((int) ((((unsigned) ((unsigned char *) (NameEntry))[0]) << 8) + \ + ((unsigned) ((unsigned char *) (NameEntry))[1]))) + + +static void build_one_capture(const pcre *code, ReturnInfo **ri, int *sallocated, int has_dupnames, char *name) +{ + ReturnInfo *r = (*ri); + if (has_dupnames) { + /* Build a sequence of positions, starting with -size if + more than one, otherwise just put the index there... */ + char *first,*last; + int esize = erts_pcre_get_stringtable_entries(code,name,&first,&last); + if (esize == PCRE_ERROR_NOSUBSTRING) { + r->v[r->num_spec - 1] = -1; + } else if(last == first) { + r->v[r->num_spec - 1] = PICK_INDEX(first); + } else { + int num = ((last - first) / esize) + 1; + int i; + ASSERT(num > 1); + r->v[r->num_spec - 1] = -num; /* A value less than -1 means + multiple indexes for same name */ + for (i = 0; i < num; ++i) { + ++(r->num_spec); + if(r->num_spec > (*sallocated)) { + (*sallocated) += 10; + r = erts_realloc(ERTS_ALC_T_RE_SUBJECT, r, + RINFO_SIZ((*sallocated))); + } + r->v[r->num_spec - 1] = PICK_INDEX(first); + first += esize; + } + } + } else { + /* Use the faster binary search if no duplicate names are present */ + if ((r->v[r->num_spec - 1] = erts_pcre_get_stringnumber(code,name)) == + PCRE_ERROR_NOSUBSTRING) { + r->v[r->num_spec - 1] = -1; + } + } + *ri = r; +} static ReturnInfo * build_capture(Eterm capture_spec[CAPSPEC_SIZE], const pcre *code) @@ -770,13 +923,58 @@ build_capture(Eterm capture_spec[CAPSPEC_SIZE], const pcre *code) } ri->v[ri->num_spec - 1] = 0; break; + case am_all_names: + { + int rc,i,top; + int entrysize; + unsigned char *nametable, *last = NULL; + int has_dupnames; + unsigned long options; + + if (erts_pcre_fullinfo(code, NULL, PCRE_INFO_OPTIONS, &options) != 0) + goto error; + if ((rc = erts_pcre_fullinfo(code, NULL, PCRE_INFO_NAMECOUNT, &top)) != 0) + goto error; + if (top <= 0) { + ri->num_spec = 0; + ri->type = RetNone; + break; + } + if (erts_pcre_fullinfo(code, NULL, PCRE_INFO_NAMEENTRYSIZE, &entrysize) != 0) + goto error; + if (erts_pcre_fullinfo(code, NULL, PCRE_INFO_NAMETABLE, &nametable) != 0) + goto error; + + has_dupnames = ((options & PCRE_DUPNAMES) != 0); + + for(i=0;i<top;++i) { + if (last == NULL || !has_dupnames || strcmp((char *) last+2,(char *) nametable+2)) { + ASSERT(ri->num_spec >= 0); + ++(ri->num_spec); + if(ri->num_spec > sallocated) { + sallocated += 10; + ri = erts_realloc(ERTS_ALC_T_RE_SUBJECT, ri, RINFO_SIZ(sallocated)); + } + if (has_dupnames) { + /* This could be more effective, we actually have + the names and could fill in the vector + immediately. Now we lookup the name again. */ + build_one_capture(code,&ri,&sallocated,has_dupnames,(char *) nametable+2); + } else { + ri->v[ri->num_spec - 1] = PICK_INDEX(nametable); + } + } + last = nametable; + nametable += entrysize; + } + break; + } default: if (is_list(capture_spec[CAPSPEC_VALUES])) { for(l=capture_spec[CAPSPEC_VALUES];is_list(l);l = CDR(list_val(l))) { int x; Eterm val = CAR(list_val(l)); - if (ri->num_spec < 0) - ri->num_spec = 0; + ASSERT(ri->num_spec >= 0); ++(ri->num_spec); if(ri->num_spec > sallocated) { sallocated += 10; @@ -785,6 +983,11 @@ build_capture(Eterm capture_spec[CAPSPEC_SIZE], const pcre *code) if (term_to_int(val,&x)) { ri->v[ri->num_spec - 1] = x; } else if (is_atom(val) || is_binary(val) || is_list(val)) { + int has_dupnames; + unsigned long options; + if (erts_pcre_fullinfo(code, NULL, PCRE_INFO_OPTIONS, &options) != 0) + goto error; + has_dupnames = ((options & PCRE_DUPNAMES) != 0); if (is_atom(val)) { Atom *ap = atom_tab(atom_val(val)); if ((ap->len + 1) > tmpbsiz) { @@ -799,6 +1002,10 @@ build_capture(Eterm capture_spec[CAPSPEC_SIZE], const pcre *code) tmpb[ap->len] = '\0'; } else { ErlDrvSizeT slen; +#ifdef DEBUG + int buffres; +#endif + if (erts_iolist_size(val, &slen)) { goto error; } @@ -810,15 +1017,15 @@ build_capture(Eterm capture_spec[CAPSPEC_SIZE], const pcre *code) (tmpbsiz = slen + 1)); } } - if (erts_iolist_to_buf(val, tmpb, slen) != 0) { - goto error; - } + +#ifdef DEBUG + buffres = +#endif + erts_iolist_to_buf(val, tmpb, slen); + ASSERT(buffres >= 0); tmpb[slen] = '\0'; } - if ((ri->v[ri->num_spec - 1] = erts_pcre_get_stringnumber(code,tmpb)) == - PCRE_ERROR_NOSUBSTRING) { - ri->v[ri->num_spec - 1] = -1; - } + build_one_capture(code,&ri,&sallocated,has_dupnames,tmpb); } else { goto error; } @@ -867,15 +1074,18 @@ re_run(Process *p, Eterm arg1, Eterm arg2, Eterm arg3) unsigned long loop_count; Eterm capture[CAPSPEC_SIZE] = CAPSPEC_INIT; int is_list_cap; + int match_limit = 0; + int match_limit_recursion = 0; - if (parse_options(arg3,&comp_options,&options,&pflags,&startoffset,capture) + if (parse_options(arg3,&comp_options,&options,&pflags,&startoffset,capture, + &match_limit,&match_limit_recursion) < 0) { BIF_ERROR(p,BADARG); } is_list_cap = ((pflags & PARSE_FLAG_CAPTURE_OPT) && (capture[CAPSPEC_TYPE] == am_list)); - if (is_not_tuple(arg2) || (arityval(*tuple_val(arg2)) != 4)) { + if (is_not_tuple(arg2) || (arityval(*tuple_val(arg2)) != 5)) { if (is_binary(arg2) || is_list(arg2) || is_nil(arg2)) { /* Compile from textual RE */ ErlDrvSizeT slen; @@ -885,6 +1095,9 @@ re_run(Process *p, Eterm arg1, Eterm arg2, Eterm arg3) const char *errstr = ""; int errofset = 0; int capture_count; +#ifdef DEBUG + int buffres; +#endif if (pflags & PARSE_FLAG_UNICODE && (!is_binary(arg2) || !is_binary(arg1) || @@ -897,18 +1110,32 @@ re_run(Process *p, Eterm arg1, Eterm arg2, Eterm arg3) } expr = erts_alloc(ERTS_ALC_T_RE_TMP_BUF, slen + 1); - if (erts_iolist_to_buf(arg2, expr, slen) != 0) { - erts_free(ERTS_ALC_T_RE_TMP_BUF, expr); - BIF_ERROR(p,BADARG); - } + +#ifdef DEBUG + buffres = +#endif + erts_iolist_to_buf(arg2, expr, slen); + + ASSERT(buffres >= 0); + expr[slen]='\0'; result = erts_pcre_compile2(expr, comp_options, &errcode, &errstr, &errofset, default_table); if (!result) { - erts_free(ERTS_ALC_T_RE_TMP_BUF, expr); /* Compilation error gives badarg except in the compile - function */ - BIF_ERROR(p,BADARG); + function or if we have PARSE_FLAG_REPORT_ERRORS */ + if (pflags & PARSE_FLAG_REPORT_ERRORS) { + res = build_compile_result(p, am_error, result, errcode, + errstr, errofset, + (pflags & + PARSE_FLAG_UNICODE) ? 1 : 0, + 1, am_compile); + erts_free(ERTS_ALC_T_RE_TMP_BUF, expr); + BIF_RET(res); + } else { + erts_free(ERTS_ALC_T_RE_TMP_BUF, expr); + BIF_ERROR(p,BADARG); + } } if (pflags & PARSE_FLAG_GLOBAL) { Eterm precompiled = @@ -917,7 +1144,7 @@ re_run(Process *p, Eterm arg1, Eterm arg2, Eterm arg3) errstr, errofset, (pflags & PARSE_FLAG_UNICODE) ? 1 : 0, - 0); + 0, NIL); Eterm *hp,r; erts_free(ERTS_ALC_T_RE_TMP_BUF, expr); hp = HAlloc(p,4); @@ -947,7 +1174,8 @@ re_run(Process *p, Eterm arg1, Eterm arg2, Eterm arg3) tp = tuple_val(arg2); if (tp[1] != am_re_pattern || is_not_small(tp[2]) || - is_not_small(tp[3]) || is_not_binary(tp[4])) { + is_not_small(tp[3]) || is_not_small(tp[4]) || + is_not_binary(tp[5])) { BIF_ERROR(p,BADARG); } @@ -967,9 +1195,9 @@ re_run(Process *p, Eterm arg1, Eterm arg2, Eterm arg3) } ovsize = 3*(unsigned_val(tp[2])+1); - code_size = binary_size(tp[4]); - if ((code_tmp = (const pcre *) - erts_get_aligned_binary_bytes(tp[4], &temp_alloc)) == NULL) { + code_size = binary_size(tp[5]); + code_tmp = (const pcre *) erts_get_aligned_binary_bytes(tp[5], &temp_alloc); + if (code_tmp == NULL || code_size < 4) { erts_free_aligned_binary_bytes(temp_alloc); BIF_ERROR(p, BADARG); } @@ -994,6 +1222,16 @@ re_run(Process *p, Eterm arg1, Eterm arg2, Eterm arg3) restart.extra.restart_flags = 0; restart.extra.loop_counter_return = &loop_count; restart.ret_info = NULL; + + if (pflags & PARSE_FLAG_MATCH_LIMIT) { + restart.extra.flags |= PCRE_EXTRA_MATCH_LIMIT; + restart.extra.match_limit = match_limit; + } + + if (pflags & PARSE_FLAG_MATCH_LIMIT_RECURSION) { + restart.extra.flags |= PCRE_EXTRA_MATCH_LIMIT_RECURSION; + restart.extra.match_limit_recursion = match_limit_recursion; + } if (pflags & PARSE_FLAG_CAPTURE_OPT) { if ((restart.ret_info = build_capture(capture,restart.code)) == NULL) { @@ -1002,7 +1240,7 @@ re_run(Process *p, Eterm arg1, Eterm arg2, Eterm arg3) BIF_ERROR(p,BADARG); } } - + /* Optimized - if already in binary off heap, keep that and avoid copying, also binary returns can be sub binaries in that case */ @@ -1029,6 +1267,9 @@ re_run(Process *p, Eterm arg1, Eterm arg2, Eterm arg3) restart.subject = (char *) (pb->bytes+offset); restart.flags |= RESTART_FLAG_SUBJECT_IN_BINARY; } else { +#ifdef DEBUG + int buffres; +#endif handle_iolist: if (erts_iolist_size(arg1, &slength)) { erts_free(ERTS_ALC_T_RE_SUBJECT, restart.ovector); @@ -1040,24 +1281,30 @@ handle_iolist: } restart.subject = erts_alloc(ERTS_ALC_T_RE_SUBJECT, slength); - if (erts_iolist_to_buf(arg1, restart.subject, slength) != 0) { - erts_free(ERTS_ALC_T_RE_SUBJECT, restart.ovector); - erts_free(ERTS_ALC_T_RE_SUBJECT, restart.code); - erts_free(ERTS_ALC_T_RE_SUBJECT, restart.subject); - if (restart.ret_info != NULL) { - erts_free(ERTS_ALC_T_RE_SUBJECT, restart.ret_info); - } - BIF_ERROR(p,BADARG); - } +#ifdef DEBUG + buffres = +#endif + erts_iolist_to_buf(arg1, restart.subject, slength); + ASSERT(buffres >= 0); } + if (pflags & PARSE_FLAG_REPORT_ERRORS) { + restart.flags |= RESTART_FLAG_REPORT_MATCH_LIMIT; + } #ifdef DEBUG loop_count = 0xFFFFFFFF; #endif + + rc = erts_pcre_exec(restart.code, &(restart.extra), restart.subject, + slength, startoffset, + options, restart.ovector, ovsize); + + if (rc == PCRE_ERROR_BADENDIANNESS || rc == PCRE_ERROR_BADMAGIC) { + cleanup_restart_context(&restart); + BIF_ERROR(p,BADARG); + } - rc = erts_pcre_exec(restart.code, &(restart.extra), restart.subject, slength, startoffset, - options, restart.ovector, ovsize); ASSERT(loop_count != 0xFFFFFFFF); BUMP_REDS(p, loop_count / LOOP_FACTOR); if (rc == PCRE_ERROR_LOOP_LIMIT) { @@ -1077,7 +1324,7 @@ handle_iolist: arg2 /* To avoid GC of precompiled code, XXX: not utilized yet */, magic_bin); } - + res = build_exec_return(p, rc, &restart, arg1); cleanup_restart_context(&restart); @@ -1149,6 +1396,120 @@ static BIF_RETTYPE re_exec_trap(BIF_ALIST_3) BIF_RET(res); } +BIF_RETTYPE +re_inspect_2(BIF_ALIST_2) +{ + Eterm *tp,*tmp_vec,*hp; + int i,top,j; + int entrysize; + unsigned char *nametable, *last,*name; + int has_dupnames; + unsigned long options; + int num_names; + Eterm res; + const pcre *code; + byte *temp_alloc = NULL; +#ifdef DEBUG + int infores; +#endif + + + if (is_not_tuple(BIF_ARG_1) || (arityval(*tuple_val(BIF_ARG_1)) != 5)) { + goto error; + } + tp = tuple_val(BIF_ARG_1); + if (tp[1] != am_re_pattern || is_not_small(tp[2]) || + is_not_small(tp[3]) || is_not_small(tp[4]) || + is_not_binary(tp[5])) { + goto error; + } + if (BIF_ARG_2 != am_namelist) { + goto error; + } + if ((code = (const pcre *) + erts_get_aligned_binary_bytes(tp[5], &temp_alloc)) == NULL) { + goto error; + } + + /* OK, so let's try to get some info */ + + if (erts_pcre_fullinfo(code, NULL, PCRE_INFO_OPTIONS, &options) != 0) + goto error; + +#ifdef DEBUG + infores = +#endif + erts_pcre_fullinfo(code, NULL, PCRE_INFO_NAMECOUNT, &top); + + ASSERT(infores == 0); + + if (top <= 0) { + hp = HAlloc(BIF_P, 3); + res = TUPLE2(hp,am_namelist,NIL); + erts_free_aligned_binary_bytes(temp_alloc); + BIF_RET(res); + } +#ifdef DEBUG + infores = +#endif + erts_pcre_fullinfo(code, NULL, PCRE_INFO_NAMEENTRYSIZE, &entrysize); + + ASSERT(infores == 0); + +#ifdef DEBUG + infores = +#endif + erts_pcre_fullinfo(code, NULL, PCRE_INFO_NAMETABLE, &nametable); + + ASSERT(infores == 0); + + has_dupnames = ((options & PCRE_DUPNAMES) != 0); + /* First, count the names */ + num_names = 0; + last = NULL; + name = nametable; + for(i=0;i<top;++i) { + if (last == NULL || !has_dupnames || strcmp((char *) last+2, + (char *) name+2)) { + ++num_names; + } + last = name; + name += entrysize; + } + tmp_vec = erts_alloc(ERTS_ALC_T_RE_TMP_BUF, + num_names * sizeof(Eterm)); + /* Re-iterate and fill tmp_vec */ + last = NULL; + name = nametable; + j = 0; + for(i=0;i<top;++i) { + if (last == NULL || !has_dupnames || strcmp((char *) last+2, + (char *) name+2)) { + tmp_vec[j++] = new_binary(BIF_P, (byte *) name+2, strlen((char *) name+2)); + } + last = name; + name += entrysize; + } + ASSERT(j == num_names); + hp = HAlloc(BIF_P, 3+2*j); + res = NIL; + for(i = j-1 ;i >= 0; --i) { + res = CONS(hp,tmp_vec[i],res); + hp += 2; + } + res = TUPLE2(hp,am_namelist,res); + erts_free_aligned_binary_bytes(temp_alloc); + erts_free(ERTS_ALC_T_RE_TMP_BUF, tmp_vec); + BIF_RET(res); + + error: + /* tmp_vec never allocated when we reach here */ + erts_free_aligned_binary_bytes(temp_alloc); + BIF_ERROR(BIF_P,BADARG); +} + + + diff --git a/erts/emulator/beam/erl_bif_timer.c b/erts/emulator/beam/erl_bif_timer.c index d67695e533..03ac97283c 100644 --- a/erts/emulator/beam/erl_bif_timer.c +++ b/erts/emulator/beam/erl_bif_timer.c @@ -616,7 +616,7 @@ erts_print_bif_timer_info(int to, void *to_arg) : btm->receiver.proc.ess->common.id); erts_print(to, to_arg, "=timer:%T\n", receiver); erts_print(to, to_arg, "Message: %T\n", btm->message); - erts_print(to, to_arg, "Time left: %u ms\n", + erts_print(to, to_arg, "Time left: %u\n", erts_time_left(&btm->tm)); } } diff --git a/erts/emulator/beam/erl_db_hash.c b/erts/emulator/beam/erl_db_hash.c index 2fea4671e1..06dac8f161 100644 --- a/erts/emulator/beam/erl_db_hash.c +++ b/erts/emulator/beam/erl_db_hash.c @@ -2106,7 +2106,7 @@ static void db_print_hash(int to, void *to_arg, int show, DbTable *tbl) DbTableHash *tb = &tbl->hash; int i; - erts_print(to, to_arg, "Buckets: %d \n", NACTIVE(tb)); + erts_print(to, to_arg, "Buckets: %d\n", NACTIVE(tb)); if (show) { for (i = 0; i < NACTIVE(tb); i++) { diff --git a/erts/emulator/beam/erl_debug.c b/erts/emulator/beam/erl_debug.c index b90d00f236..873a9860da 100644 --- a/erts/emulator/beam/erl_debug.c +++ b/erts/emulator/beam/erl_debug.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1998-2012. All Rights Reserved. + * Copyright Ericsson AB 1998-2013. 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 @@ -299,6 +299,9 @@ void erts_check_for_holes(Process* p) ErlHeapFragment* hf; Eterm* start; + if (p->flags & F_DISABLE_GC) + return; + start = p->last_htop ? p->last_htop : HEAP_START(p); check_memory(start, HEAP_TOP(p)); p->last_htop = HEAP_TOP(p); diff --git a/erts/emulator/beam/erl_driver.h b/erts/emulator/beam/erl_driver.h index b68fd46fcc..5cffae92be 100644 --- a/erts/emulator/beam/erl_driver.h +++ b/erts/emulator/beam/erl_driver.h @@ -365,17 +365,23 @@ typedef struct erl_drv_entry { * It must initialize a ErlDrvEntry structure and return a pointer to it. */ +#ifdef STATIC_ERLANG_DRIVER +# define ERLANG_DRIVER_NAME(NAME) NAME ## _driver_init +#else +# define ERLANG_DRIVER_NAME(NAME) driver_init +#endif + /* For windows dynamic drivers */ #ifndef ERL_DRIVER_TYPES_ONLY #if defined(__WIN32__) # define DRIVER_INIT(DRIVER_NAME) \ - __declspec(dllexport) ErlDrvEntry* driver_init(void); \ - __declspec(dllexport) ErlDrvEntry* driver_init(void) + __declspec(dllexport) ErlDrvEntry* ERLANG_DRIVER_NAME(DRIVER_NAME)(void); \ + __declspec(dllexport) ErlDrvEntry* ERLANG_DRIVER_NAME(DRIVER_NAME)(void) #else # define DRIVER_INIT(DRIVER_NAME) \ - ErlDrvEntry* driver_init(void); \ - ErlDrvEntry* driver_init(void) + ErlDrvEntry* ERLANG_DRIVER_NAME(DRIVER_NAME)(void); \ + ErlDrvEntry* ERLANG_DRIVER_NAME(DRIVER_NAME)(void) #endif #define ERL_DRV_BUSY_MSGQ_DISABLED (~((ErlDrvSizeT) 0)) diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c index e89725c190..8ff6f9a3b9 100644 --- a/erts/emulator/beam/erl_gc.c +++ b/erts/emulator/beam/erl_gc.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 2002-2013. All Rights Reserved. + * Copyright Ericsson AB 2002-2014. 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 @@ -400,10 +400,16 @@ erts_garbage_collect(Process* p, int need, Eterm* objv, int nobj) Uint reclaimed_now = 0; int done = 0; Uint ms1, s1, us1; - ErtsSchedulerData *esdp = erts_get_scheduler_data(); + ErtsSchedulerData *esdp; #ifdef USE_VM_PROBES DTRACE_CHARBUF(pidbuf, DTRACE_TERM_BUF_SIZE); #endif + + if (p->flags & F_DISABLE_GC) + return 1; + + esdp = erts_get_scheduler_data(); + if (IS_TRACED_FL(p, F_TRACE_GC)) { trace_gc(p, am_gc_start); } @@ -532,6 +538,9 @@ erts_garbage_collect_hibernate(Process* p) Uint area_size; Sint offs; + if (p->flags & F_DISABLE_GC) + ERTS_INTERNAL_ERROR("GC disabled"); + /* * Preliminaries. */ @@ -667,6 +676,8 @@ erts_garbage_collect_literals(Process* p, Eterm* literals, Uint n; struct erl_off_heap_header** prev; + if (p->flags & F_DISABLE_GC) + return; /* * Set GC state. */ @@ -1964,17 +1975,6 @@ setup_rootset(Process *p, Eterm *objv, int nobj, Rootset *rootset) ++n; } - /* - * A trapping BIF can add to rootset by setting the extra_root - * in the process_structure. - */ - if (p->extra_root != NULL) { - roots[n].v = p->extra_root->objv; - roots[n].sz = p->extra_root->sz; - ++n; - } - - ASSERT((is_nil(p->seq_trace_token) || is_tuple(follow_moved(p->seq_trace_token)) || is_atom(p->seq_trace_token))); @@ -2552,11 +2552,6 @@ offset_one_rootset(Process *p, Sint offs, char* area, Uint area_size, p->dictionary->used, offs, area, area_size); } - if (p->extra_root != NULL) { - offset_heap_ptr(p->extra_root->objv, - p->extra_root->sz, - offs, area, area_size); - } offset_heap_ptr(&p->fvalue, 1, offs, area, area_size); offset_heap_ptr(&p->ftrace, 1, offs, area, area_size); diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c index 8c4fffa75b..19088fd913 100644 --- a/erts/emulator/beam/erl_init.c +++ b/erts/emulator/beam/erl_init.c @@ -537,6 +537,12 @@ 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"); +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT + erts_fprintf(stderr, "-sub bool enable/disable scheduler utilization balancing,\n"); +#else + erts_fprintf(stderr, "-sub false disable scheduler utilization balancing,\n"); +#endif + erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); erts_fprintf(stderr, "-sws val set scheduler wakeup strategy, valid values are:\n"); erts_fprintf(stderr, " default|legacy.\n"); erts_fprintf(stderr, "-swct val set scheduler wake cleanup threshold, valid values are:\n"); @@ -553,8 +559,8 @@ void erts_usage(void) erts_fprintf(stderr, " numbers is %d\n", ERTS_MAX_NO_OF_SCHEDULERS); erts_fprintf(stderr, "-SP p1:p2 specify schedulers (p1) and schedulers online (p2)\n"); - erts_fprintf(stderr, " as percentages of logical processors configured and logical\n"); - erts_fprintf(stderr, " processors available, respectively\n"); + erts_fprintf(stderr, " as percentages of logical processors configured and logical\n"); + erts_fprintf(stderr, " processors available, respectively\n"); erts_fprintf(stderr, "-t size set the maximum number of atoms the " "emulator can handle\n"); erts_fprintf(stderr, " valid range is [%d-%d]\n", @@ -1433,8 +1439,10 @@ erl_start(int argc, char **argv) } else if (has_prefix("cl", sub_param)) { arg = get_arg(sub_param+2, argv[i+1], &i); - if (sys_strcmp("true", arg) == 0) + if (sys_strcmp("true", arg) == 0) { erts_sched_compact_load = 1; + erts_sched_balance_util = 0; + } else if (sys_strcmp("false", arg) == 0) erts_sched_compact_load = 0; else { @@ -1512,6 +1520,26 @@ erl_start(int argc, char **argv) erts_usage(); } } + else if (has_prefix("ub", sub_param)) { + arg = get_arg(sub_param+2, argv[i+1], &i); + if (sys_strcmp("true", arg) == 0) { +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT + erts_sched_balance_util = 1; +#else + erts_fprintf(stderr, + "scheduler utilization balancing not " + "supported on this system\n"); + erts_usage(); +#endif + } + else if (sys_strcmp("false", arg) == 0) + erts_sched_balance_util = 0; + else { + erts_fprintf(stderr, "bad scheduler utilization balancing " + " value '%s'\n", arg); + erts_usage(); + } + } else if (has_prefix("wct", sub_param)) { arg = get_arg(sub_param+3, argv[i+1], &i); if (erts_sched_set_wake_cleanup_threshold(arg) != 0) { diff --git a/erts/emulator/beam/erl_message.h b/erts/emulator/beam/erl_message.h index 771eba431f..0f3bb8d281 100644 --- a/erts/emulator/beam/erl_message.h +++ b/erts/emulator/beam/erl_message.h @@ -46,6 +46,11 @@ typedef struct erl_off_heap { Uint64 overhead; /* Administrative overhead (used to force GC). */ } ErlOffHeap; +#define ERTS_INIT_OFF_HEAP(OHP) \ + do { \ + (OHP)->first = NULL; \ + (OHP)->overhead = 0; \ + } while (0) #include "external.h" #include "erl_process.h" diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c index 48f8be8dd3..dc285b3cf7 100644 --- a/erts/emulator/beam/erl_nif.c +++ b/erts/emulator/beam/erl_nif.c @@ -1213,7 +1213,8 @@ static void close_lib(struct erl_module_nif* lib) lib->entry->unload(&env, lib->priv_data); post_nif_noproc(&env); } - erts_sys_ddll_close(lib->handle); + if (!erts_is_static_nif(lib->handle)) + erts_sys_ddll_close(lib->handle); lib->handle = NULL; } @@ -1406,7 +1407,7 @@ void* enif_dlopen(const char* lib, ErtsSysDdllError errdesc = ERTS_SYS_DDLL_ERROR_INIT; void* handle; void* init_func; - if (erts_sys_ddll_open2(lib, &handle, &errdesc) == ERL_DE_NO_ERROR) { + if (erts_sys_ddll_open(lib, &handle, &errdesc) == ERL_DE_NO_ERROR) { if (erts_sys_ddll_load_nif_init(handle, &init_func, &errdesc) == ERL_DE_NO_ERROR) { erts_sys_ddll_call_nif_init(init_func); } @@ -1564,12 +1565,13 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) static const char upgrade[] = "upgrade"; char* lib_name = NULL; void* handle = NULL; - void* init_func; + void* init_func = NULL; ErlNifEntry* entry = NULL; ErlNifEnv env; - int len, i, err; + int i, err, encoding; Module* mod; Eterm mod_atom; + const Atom* mod_atomp; Eterm f_atom; BeamInstr* caller; ErtsSysDdllError errdesc = ERTS_SYS_DDLL_ERROR_INIT; @@ -1578,18 +1580,18 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) struct erl_module_nif* lib = NULL; int reload_warning = 0; - len = list_length(BIF_ARG_1); - if (len < 0) { - BIF_ERROR(BIF_P, BADARG); + encoding = erts_get_native_filename_encoding(); + if (encoding == ERL_FILENAME_WIN_WCHAR) { + /* Do not convert the lib name to utf-16le yet, do that in win32 specific code */ + /* since lib_name is used in error messages */ + encoding = ERL_FILENAME_UTF8; } - - lib_name = (char *) erts_alloc(ERTS_ALC_T_TMP, len + 1); - - if (intlist_to_buf(BIF_ARG_1, lib_name, len) != len) { - erts_free(ERTS_ALC_T_TMP, lib_name); + lib_name = erts_convert_filename_to_encoding(BIF_ARG_1, NULL, 0, + ERTS_ALC_T_TMP, 1, 0, encoding, + NULL, 0); + if (!lib_name) { BIF_ERROR(BIF_P, BADARG); } - lib_name[len] = '\0'; if (!erts_try_seize_code_write_permission(BIF_P)) { erts_free(ERTS_ALC_T_TMP, lib_name); @@ -1613,13 +1615,19 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) mod=erts_get_module(mod_atom, erts_active_code_ix()); ASSERT(mod != NULL); + mod_atomp = atom_tab(atom_val(mod_atom)); + init_func = erts_static_nif_get_nif_init((char*)mod_atomp->name, mod_atomp->len); + if (init_func != NULL) + handle = init_func; + if (!in_area(caller, mod->curr.code, mod->curr.code_length)) { ASSERT(in_area(caller, mod->old.code, mod->old.code_length)); ret = load_nif_error(BIF_P, "old_code", "Calling load_nif from old " "module '%T' not allowed", mod_atom); } - else if ((err=erts_sys_ddll_open2(lib_name, &handle, &errdesc)) != ERL_DE_NO_ERROR) { + else if (init_func == NULL && + (err=erts_sys_ddll_open(lib_name, &handle, &errdesc)) != ERL_DE_NO_ERROR) { const char slogan[] = "Failed to load NIF library"; if (strstr(errdesc.str, lib_name) != NULL) { ret = load_nif_error(BIF_P, "load_failed", "%s: '%s'", slogan, errdesc.str); @@ -1628,7 +1636,8 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) ret = load_nif_error(BIF_P, "load_failed", "%s %s: '%s'", slogan, lib_name, errdesc.str); } } - else if (erts_sys_ddll_load_nif_init(handle, &init_func, &errdesc) != ERL_DE_NO_ERROR) { + else if (init_func == NULL && + erts_sys_ddll_load_nif_init(handle, &init_func, &errdesc) != ERL_DE_NO_ERROR) { ret = load_nif_error(BIF_P, bad_lib, "Failed to find library init" " function: '%s'", errdesc.str); @@ -1784,7 +1793,7 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) if (lib != NULL) { erts_free(ERTS_ALC_T_NIF, lib); } - if (handle != NULL) { + if (handle != NULL && !erts_is_static_nif(handle)) { erts_sys_ddll_close(handle); } erts_sys_ddll_free_error(&errdesc); diff --git a/erts/emulator/beam/erl_nif.h b/erts/emulator/beam/erl_nif.h index 8006741a63..5f4dc21d5c 100644 --- a/erts/emulator/beam/erl_nif.h +++ b/erts/emulator/beam/erl_nif.h @@ -168,7 +168,7 @@ extern TWinDynNifCallbacks WinDynNifCallbacks; # undef ERL_NIF_API_FUNC_DECL #endif -#if (defined(__WIN32__) || defined(_WIN32) || defined(_WIN32_)) && !defined(STATIC_ERLANG_DRIVER) +#if (defined(__WIN32__) || defined(_WIN32) || defined(_WIN32_)) && !defined(STATIC_ERLANG_DRIVER) && !defined(STATIC_ERLANG_NIF) # define ERL_NIF_API_FUNC_MACRO(NAME) (WinDynNifCallbacks.NAME) # include "erl_nif_api_funcs.h" /* note that we have to keep ERL_NIF_API_FUNC_MACRO defined */ @@ -180,15 +180,22 @@ extern TWinDynNifCallbacks WinDynNifCallbacks; # undef ERL_NIF_API_FUNC_DECL #endif - #if (defined(__WIN32__) || defined(_WIN32) || defined(_WIN32_)) # define ERL_NIF_INIT_GLOB TWinDynNifCallbacks WinDynNifCallbacks; -# define ERL_NIF_INIT_DECL(MODNAME) __declspec(dllexport) ErlNifEntry* nif_init(TWinDynNifCallbacks* callbacks) +# ifdef STATIC_ERLANG_NIF +# define ERL_NIF_INIT_DECL(MODNAME) __declspec(dllexport) ErlNifEntry* MODNAME ## _nif_init(TWinDynNifCallbacks* callbacks) +# else +# define ERL_NIF_INIT_DECL(MODNAME) __declspec(dllexport) ErlNifEntry* nif_init(TWinDynNifCallbacks* callbacks) +# endif # define ERL_NIF_INIT_BODY memcpy(&WinDynNifCallbacks,callbacks,sizeof(TWinDynNifCallbacks)) #else # define ERL_NIF_INIT_GLOB # define ERL_NIF_INIT_BODY -# define ERL_NIF_INIT_DECL(MODNAME) ErlNifEntry* nif_init(void) +# ifdef STATIC_ERLANG_NIF +# define ERL_NIF_INIT_DECL(MODNAME) ErlNifEntry* MODNAME ## _nif_init(void) +# else +# define ERL_NIF_INIT_DECL(MODNAME) ErlNifEntry* nif_init(void) +# endif #endif diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index 1efd070afd..2f383f4c01 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.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 @@ -144,6 +144,7 @@ extern BeamInstr beam_exit[]; extern BeamInstr beam_continue_exit[]; int erts_sched_compact_load; +int erts_sched_balance_util = 0; Uint erts_no_schedulers; #define ERTS_THR_PRGR_LATER_CLEANUP_OP_THRESHOLD_VERY_LAZY (4*1024*1024) @@ -283,6 +284,40 @@ struct erts_system_profile_flags_t erts_system_profile_flags; #error "Need to store process_count in another type" #endif +typedef enum { + ERTS_PSTT_GC, /* Garbage Collect */ + ERTS_PSTT_CPC /* Check Process Code */ +} ErtsProcSysTaskType; + +#define ERTS_MAX_PROC_SYS_TASK_ARGS 2 + +struct ErtsProcSysTask_ { + ErtsProcSysTask *next; + ErtsProcSysTask *prev; + ErtsProcSysTaskType type; + Eterm requester; + Eterm reply_tag; + Eterm req_id; + Uint req_id_sz; + Eterm arg[ERTS_MAX_PROC_SYS_TASK_ARGS]; + ErlOffHeap off_heap; + Eterm heap[1]; +}; + +#define ERTS_PROC_SYS_TASK_SIZE(HSz) \ + (sizeof(ErtsProcSysTask) - sizeof(Eterm) + sizeof(Eterm)*(HSz)) + +struct ErtsProcSysTaskQs_ { + int qmask; + int ncount; + ErtsProcSysTask *q[ERTS_NO_PROC_PRIO_LEVELS]; +}; + +ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(proc_sys_task_queues, + ErtsProcSysTaskQs, + 50, + ERTS_ALC_T_PROC_SYS_TSK_QS) + ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(misc_op_list, ErtsMiscOpList, 10, @@ -353,6 +388,14 @@ static void aux_work_timeout_early_init(int no_schedulers); static void aux_work_timeout_late_init(void); static void setup_aux_work_timer(void); +static int execute_sys_tasks(Process *c_p, + erts_aint32_t *statep, + int in_reds); +static int cleanup_sys_tasks(Process *c_p, + erts_aint32_t in_state, + int in_reds); + + #if defined(DEBUG) || 0 #define ERTS_DBG_CHK_AUX_WORK_VAL(V) dbg_chk_aux_work_val((V)) static void @@ -471,6 +514,11 @@ erts_pre_init_process(void) erts_psd_required_locks[ERTS_PSD_CALL_TIME_BP].set_locks = ERTS_PSD_CALL_TIME_BP_SET_LOCKS; + erts_psd_required_locks[ERTS_PSD_DELAYED_GC_TASK_QS].get_locks + = ERTS_PSD_DELAYED_GC_TASK_QS_GET_LOCKS; + erts_psd_required_locks[ERTS_PSD_DELAYED_GC_TASK_QS].set_locks + = ERTS_PSD_DELAYED_GC_TASK_QS_SET_LOCKS; + /* Check that we have locks for all entries */ for (ix = 0; ix < ERTS_PSD_SIZE; ix++) { ERTS_SMP_LC_ASSERT(erts_psd_required_locks[ix].get_locks); @@ -561,6 +609,7 @@ erts_late_init_process(void) static void init_sched_wall_time(ErtsSchedWallTime *swtp) { + swtp->need = erts_sched_balance_util; swtp->enabled = 0; swtp->start = 0; swtp->working.total = 0; @@ -583,27 +632,253 @@ sched_wall_time_ts(void) #endif } +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + +#ifdef ARCH_64 + +static ERTS_INLINE Uint64 +aschedtime_read(ErtsAtomicSchedTime *var) +{ + return (Uint64) erts_atomic_read_nob((erts_atomic_t *) var); +} + +static ERTS_INLINE void +aschedtime_set(ErtsAtomicSchedTime *var, Uint64 val) +{ + erts_atomic_set_nob((erts_atomic_t *) var, (erts_aint_t) val); +} + +static ERTS_INLINE void +aschedtime_init(ErtsAtomicSchedTime *var) +{ + erts_atomic_init_nob((erts_atomic_t *) var, (erts_aint_t) 0); +} + +#elif defined(ARCH_32) + +static ERTS_INLINE Uint64 +aschedtime_read(ErtsAtomicSchedTime *var) +{ + erts_dw_aint_t dw; + erts_dw_atomic_read_nob((erts_dw_atomic_t *) var, &dw); +#ifdef ETHR_SU_DW_NAINT_T__ + return (Uint64) dw.dw_sint; +#else + { + Uint64 res; + res = (Uint64) ((Uint32) dw.sint[ERTS_DW_AINT_HIGH_WORD]); + res <<= 32; + res |= (Uint64) ((Uint32) dw.sint[ERTS_DW_AINT_LOW_WORD]); + return res; + } +#endif +} + +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)) + +/* Intervals in nanoseconds */ +#define ERTS_SCHED_UTIL_SHORT_INTERVAL ((Uint64) 1*1000*1000*1000) +#define ERTS_SCHED_UTIL_LONG_INTERVAL ((Uint64) 10*1000*1000*1000) + + +#define ERTS_SCHED_UTIL_IGNORE_IMBALANCE_DIFF 5000 /* ppm */ + +static ERTS_INLINE Uint64 +calc_sched_worktime(int is_working, Uint64 now, Uint64 last, + Uint64 interval, Uint64 old_worktime) +{ + Uint64 worktime; + Uint64 new; + + if (now <= last) + return old_worktime; + + new = now - last; + + if (new >= interval) + return is_working ? interval : (Uint64) 0; + + + /* + * Division by 1000 in order to avoid + * overflow. If changed update assertions + * in init_runq_sched_util(). + */ + worktime = old_worktime; + worktime *= (interval - new)/1000; + worktime /= (interval/1000); + if (is_working) + worktime += new; + + ASSERT(0 <= worktime && worktime <= interval); + + return worktime; +} + +static ERTS_INLINE void +update_avg_sched_util(ErtsSchedulerData *esdp, Uint64 now, int is_working) +{ + ErtsRunQueue *rq; + int worked; + Uint64 swt, lwt, last; + + rq = esdp->run_queue; + last = aschedtime_read(&rq->sched_util.last); + + if (now <= last) { + ASSERT(last == ERTS_SCHED_AVG_UTIL_WRITE_MARKER); + return; + } + + ASSERT(now >= last); + + worked = rq->sched_util.is_working; + + swt = calc_sched_worktime(worked, now, last, ERTS_SCHED_UTIL_SHORT_INTERVAL, + rq->sched_util.worktime.short_interval); + lwt = calc_sched_worktime(worked, now, last, ERTS_SCHED_UTIL_LONG_INTERVAL, + rq->sched_util.worktime.long_interval); + + aschedtime_set(&rq->sched_util.last, ERTS_SCHED_AVG_UTIL_WRITE_MARKER); + ERTS_THR_WRITE_MEMORY_BARRIER; + rq->sched_util.is_working = is_working; + rq->sched_util.worktime.short_interval = swt; + rq->sched_util.worktime.long_interval = lwt; + ERTS_THR_WRITE_MEMORY_BARRIER; + aschedtime_set(&rq->sched_util.last, now); +} + +int +erts_get_sched_util(ErtsRunQueue *rq, int initially_locked, int short_interval) +{ + /* Average scheduler utilization in ppm */ + int util, is_working, try = 0, locked = initially_locked; + Uint64 worktime, old_worktime, now, last, interval, *old_worktimep; + + if (short_interval) { + old_worktimep = &rq->sched_util.worktime.short_interval; + interval = ERTS_SCHED_UTIL_SHORT_INTERVAL; + } + else { + old_worktimep = &rq->sched_util.worktime.long_interval; + interval = ERTS_SCHED_UTIL_LONG_INTERVAL; + } + + while (1) { + Uint64 chk_last; + last = aschedtime_read(&rq->sched_util.last); + ERTS_THR_READ_MEMORY_BARRIER; + is_working = rq->sched_util.is_working; + old_worktime = *old_worktimep; + ERTS_THR_READ_MEMORY_BARRIER; + chk_last = aschedtime_read(&rq->sched_util.last); + if (chk_last == last) + break; + if (!locked) { + if (++try >= ERTS_GET_AVG_MAX_UNLOCKED_TRY) { + /* Writer will eventually block on runq-lock */ + erts_smp_runq_lock(rq); + locked = 1; + } + } + } + + if (!initially_locked && locked) + erts_smp_runq_unlock(rq); + + now = sched_wall_time_ts(); + worktime = calc_sched_worktime(is_working, now, last, interval, old_worktime); + + util = (int) ((worktime * 1000000)/interval); + + ASSERT(0 <= util && util <= 1000000); + + return util; +} + +static void +init_runq_sched_util(ErtsRunQueueSchedUtil *rqsu, int enabled) +{ + aschedtime_init(&rqsu->last); + if (!enabled) + aschedtime_set(&rqsu->last, ERTS_SCHED_AVG_UTIL_WRITE_MARKER); + rqsu->is_working = 0; + rqsu->worktime.short_interval = (Uint64) 0; + rqsu->worktime.long_interval = (Uint64) 0; + +#ifdef DEBUG + { + Uint64 intrvl; + /* + * If one of these asserts fail we may have + * overflow in calc_sched_worktime(). Which + * have to be fixed either by shrinking + * interval size, or fix calculation of + * worktime in calc_sched_worktime(). + */ + intrvl = ERTS_SCHED_UTIL_SHORT_INTERVAL; + ASSERT(intrvl*(intrvl/1000) > intrvl); + intrvl = ERTS_SCHED_UTIL_LONG_INTERVAL; + ASSERT(intrvl*(intrvl/1000) > intrvl); + } +#endif +} + +#endif /* ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT */ + static ERTS_INLINE void sched_wall_time_change(ErtsSchedulerData *esdp, int working) { - if (esdp->sched_wall_time.enabled) { + if (esdp->sched_wall_time.need) { Uint64 ts = sched_wall_time_ts(); - if (working) { +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + update_avg_sched_util(esdp, ts, working); +#endif + if (esdp->sched_wall_time.enabled) { + if (working) { #ifdef DEBUG - ASSERT(!esdp->sched_wall_time.working.currently); - esdp->sched_wall_time.working.currently = 1; + ASSERT(!esdp->sched_wall_time.working.currently); + esdp->sched_wall_time.working.currently = 1; #endif - ts -= esdp->sched_wall_time.start; - esdp->sched_wall_time.working.start = ts; - } - else { + ts -= esdp->sched_wall_time.start; + esdp->sched_wall_time.working.start = ts; + } + else { #ifdef DEBUG - ASSERT(esdp->sched_wall_time.working.currently); - esdp->sched_wall_time.working.currently = 0; + ASSERT(esdp->sched_wall_time.working.currently); + esdp->sched_wall_time.working.currently = 0; #endif - ts -= esdp->sched_wall_time.start; - ts -= esdp->sched_wall_time.working.start; - esdp->sched_wall_time.working.total += ts; + ts -= esdp->sched_wall_time.start; + ts -= esdp->sched_wall_time.working.start; + esdp->sched_wall_time.working.total += ts; + } } } } @@ -658,10 +933,13 @@ reply_sched_wall_time(void *vswtrp) ASSERT(esdp); if (swtrp->set) { - if (!swtrp->enable && esdp->sched_wall_time.enabled) + if (!swtrp->enable && esdp->sched_wall_time.enabled) { + esdp->sched_wall_time.need = erts_sched_balance_util; esdp->sched_wall_time.enabled = 0; + } else if (swtrp->enable && !esdp->sched_wall_time.enabled) { Uint64 ts = sched_wall_time_ts(); + esdp->sched_wall_time.need = 1; esdp->sched_wall_time.enabled = 1; esdp->sched_wall_time.start = ts; esdp->sched_wall_time.working.total = 0; @@ -2037,9 +2315,8 @@ ongoing_multi_scheduling_block(void) } static ERTS_INLINE void -empty_runq(ErtsRunQueue *rq) +empty_runq_aux(ErtsRunQueue *rq, Uint32 old_flags) { - Uint32 old_flags = ERTS_RUNQ_FLGS_UNSET(rq, ERTS_RUNQ_FLG_NONEMPTY|ERTS_RUNQ_FLG_PROTECTED); if (old_flags & ERTS_RUNQ_FLG_NONEMPTY) { #ifdef DEBUG erts_aint32_t empty = erts_smp_atomic32_read_nob(&no_empty_run_queues); @@ -2060,6 +2337,23 @@ empty_runq(ErtsRunQueue *rq) } static ERTS_INLINE void +empty_runq(ErtsRunQueue *rq) +{ + Uint32 old_flags = ERTS_RUNQ_FLGS_UNSET(rq, ERTS_RUNQ_FLG_NONEMPTY|ERTS_RUNQ_FLG_PROTECTED); + empty_runq_aux(rq, old_flags); +} + +static ERTS_INLINE Uint32 +empty_protected_runq(ErtsRunQueue *rq) +{ + Uint32 old_flags = ERTS_RUNQ_FLGS_BSET(rq, + ERTS_RUNQ_FLG_NONEMPTY|ERTS_RUNQ_FLG_PROTECTED, + ERTS_RUNQ_FLG_PROTECTED); + empty_runq_aux(rq, old_flags); + return old_flags; +} + +static ERTS_INLINE void non_empty_runq(ErtsRunQueue *rq) { Uint32 old_flags = ERTS_RUNQ_FLGS_SET(rq, ERTS_RUNQ_FLG_NONEMPTY); @@ -2083,6 +2377,18 @@ non_empty_runq(ErtsRunQueue *rq) } } +void +erts_empty_runq(ErtsRunQueue *rq) +{ + empty_runq(rq); +} + +void +erts_non_empty_runq(ErtsRunQueue *rq) +{ + non_empty_runq(rq); +} + static erts_aint32_t sched_prep_spin_wait(ErtsSchedulerSleepInfo *ssi) { @@ -2585,7 +2891,7 @@ ssi_flags_set_wake(ErtsSchedulerSleepInfo *ssi) } static void -wake_scheduler(ErtsRunQueue *rq, int incq) +wake_scheduler(ErtsRunQueue *rq) { ErtsSchedulerSleepInfo *ssi; erts_aint32_t flgs; @@ -2604,9 +2910,6 @@ wake_scheduler(ErtsRunQueue *rq, int incq) flgs = ssi_flags_set_wake(ssi); erts_sched_finish_poke(ssi, flgs); - - if (incq && (flgs & ERTS_SSI_FLG_WAITING)) - non_empty_runq(rq); } #define ERTS_NO_USED_RUNQS_SHIFT 16 @@ -2697,7 +3000,7 @@ chk_wake_sched(ErtsRunQueue *crq, int ix, int activate) if (try_inc_no_active_runqs(ix+1)) (void) ERTS_RUNQ_FLGS_UNSET(wrq, ERTS_RUNQ_FLG_INACTIVE); } - wake_scheduler(wrq, 0); + wake_scheduler(wrq); return 1; } return 0; @@ -2745,7 +3048,7 @@ smp_notify_inc_runq(ErtsRunQueue *runq) { #ifdef ERTS_SMP if (runq) - wake_scheduler(runq, 1); + wake_scheduler(runq); #endif } @@ -2763,7 +3066,7 @@ erts_sched_notify_check_cpu_bind(void) for (ix = 0; ix < erts_no_run_queues; ix++) { ErtsRunQueue *rq = ERTS_RUNQ_IX(ix); (void) ERTS_RUNQ_FLGS_SET(rq, ERTS_RUNQ_FLG_CHK_CPU_BIND); - wake_scheduler(rq, 0); + wake_scheduler(rq); } #else erts_sched_check_cpu_bind(erts_get_scheduler_data()); @@ -2848,7 +3151,7 @@ dequeue_process(ErtsRunQueue *runq, int prio_q, erts_aint32_t *statep) if (statep) *statep = state; - prio = (int) (ERTS_PSFLG_PRIO_MASK & state); + prio = (int) ERTS_PSFLGS_GET_PRQ_PRIO(state); rqi = &runq->procs.prio_info[prio]; @@ -2891,6 +3194,11 @@ check_immigration_need(ErtsRunQueue *c_rq, ErtsMigrationPath *mp, int prio) if (!f_rq) return NULL; +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + if (mp->sched_util) + return NULL; +#endif + f_rq_flags = ERTS_RUNQ_FLGS_GET(f_rq); if (f_rq_flags & ERTS_RUNQ_FLG_PROTECTED) return NULL; @@ -2997,7 +3305,7 @@ immigrate(ErtsRunQueue *c_rq, ErtsMigrationPath *mp) erts_aint32_t state; state = erts_smp_atomic32_read_acqb(&proc->state); if (!(ERTS_PSFLG_BOUND & state) - && (prio == (int) (ERTS_PSFLG_PRIO_MASK & state))) { + && (prio == (int) ERTS_PSFLGS_GET_PRQ_PRIO(state))) { ErtsRunQueueInfo *rqi = &rq->procs.prio_info[prio]; unqueue_process(rq, rpq, rqi, prio, prev_proc, proc); erts_smp_runq_unlock(rq); @@ -3030,7 +3338,7 @@ suspend_run_queue(ErtsRunQueue *rq) ERTS_SSI_FLG_SUSPENDED); (void) ERTS_RUNQ_FLGS_SET(rq, ERTS_RUNQ_FLG_SUSPENDED); - wake_scheduler(rq, 0); + wake_scheduler(rq); } static void scheduler_ix_resume_wake(Uint ix); @@ -3079,7 +3387,7 @@ schedule_bound_processes(ErtsRunQueue *rq, while (proc) { erts_aint32_t state = erts_smp_atomic32_read_acqb(&proc->state); next = proc->next; - enqueue_process(rq, (int) (ERTS_PSFLG_PRIO_MASK & state), proc); + enqueue_process(rq, (int) ERTS_PSFLGS_GET_PRQ_PRIO(state), proc); proc = next; } } @@ -3122,6 +3430,9 @@ evacuate_run_queue(ErtsRunQueue *rq, to_rq->misc.start = start; to_rq->misc.end = end; + + non_empty_runq(to_rq); + erts_smp_runq_unlock(to_rq); smp_notify_inc_runq(to_rq); erts_smp_runq_lock(to_rq); @@ -3184,7 +3495,7 @@ evacuate_run_queue(ErtsRunQueue *rq, sbpp->last = proc; } else { - int prio = (int) (ERTS_PSFLG_PRIO_MASK & state); + int prio = (int) ERTS_PSFLGS_GET_PRQ_PRIO(state); erts_smp_runq_unlock(rq); to_rq = mp->prio[prio].runq; @@ -3257,7 +3568,7 @@ try_steal_task_from_victim(ErtsRunQueue *rq, int *rq_lockedp, ErtsRunQueue *vrq, erts_aint32_t state = erts_smp_atomic32_read_acqb(&proc->state); if (!(ERTS_PSFLG_BOUND & state)) { /* Steal process */ - int prio = (int) (ERTS_PSFLG_PRIO_MASK & state); + int prio = (int) ERTS_PSFLGS_GET_PRQ_PRIO(state); ErtsRunQueueInfo *rqi = &vrq->procs.prio_info[prio]; unqueue_process(vrq, rpq, rqi, prio, prev_proc, proc); erts_smp_runq_unlock(vrq); @@ -3334,7 +3645,7 @@ try_steal_task(ErtsRunQueue *rq) Uint32 flags; /* Protect jobs we steal from getting stolen from us... */ - flags = ERTS_RUNQ_FLGS_SET(rq, ERTS_RUNQ_FLG_PROTECTED); + flags = empty_protected_runq(rq); if (flags & ERTS_RUNQ_FLG_SUSPENDED) return 0; /* go suspend instead... */ @@ -3413,6 +3724,9 @@ typedef struct { int full_reds_history_change; int oowc; int max_len; +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + int sched_util; +#endif } ErtsRunQueueBalance; static ErtsRunQueueBalance *run_queue_info; @@ -3576,6 +3890,9 @@ check_balance(ErtsRunQueue *c_rq) Sint64 scheds_reds, full_scheds_reds; int forced, active, current_active, oowc, half_full_scheds, full_scheds, mmax_len, blnc_no_rqs, qix, pix, freds_hist_ix; +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + int sched_util_balancing; +#endif if (erts_smp_atomic32_xchg_nob(&balance_info.checking_balance, 1)) { c_rq->check_balance_reds = INT_MAX; @@ -3631,6 +3948,10 @@ check_balance(ErtsRunQueue *c_rq) return; } +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + sched_util_balancing = 0; +#endif + freds_hist_ix = balance_info.full_reds_history_index; balance_info.full_reds_history_index++; if (balance_info.full_reds_history_index >= ERTS_FULL_REDS_HISTORY_SIZE) @@ -3661,7 +3982,12 @@ check_balance(ErtsRunQueue *c_rq) run_queue_info[qix].oowc = rq->out_of_work_count; run_queue_info[qix].max_len = rq->max_len; rq->check_balance_reds = INT_MAX; - + +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + if (erts_sched_balance_util) + run_queue_info[qix].sched_util = erts_get_sched_util(rq, 1, 0); +#endif + erts_smp_runq_unlock(rq); } @@ -3731,8 +4057,38 @@ check_balance(ErtsRunQueue *c_rq) mmax_len = run_queue_info[qix].max_len; } - if (!erts_sched_compact_load) + if (!erts_sched_compact_load) { +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + if (erts_sched_balance_util && full_scheds < blnc_no_rqs) { + int avg_util = 0; + + for (qix = 0; qix < blnc_no_rqs; qix++) + avg_util += run_queue_info[qix].sched_util; + + avg_util /= blnc_no_rqs; /* in ppm */ + + sched_util_balancing = 1; + /* + * In order to avoid renaming a large amount of fields + * we write utilization values instead of lenght values + * in the 'max_len' and 'migration_limit' fields... + */ + for (qix = 0; qix < blnc_no_rqs; qix++) { + run_queue_info[qix].flags = 0; /* Reset for later use... */ + for (pix = 0; pix < ERTS_NO_PRIO_LEVELS; pix++) { + run_queue_info[qix].prio[pix].emigrate_to = -1; + run_queue_info[qix].prio[pix].immigrate_from = -1; + run_queue_info[qix].prio[pix].avail = 100; + run_queue_info[qix].prio[pix].max_len = run_queue_info[qix].sched_util; + run_queue_info[qix].prio[pix].migration_limit = avg_util; + } + } + active = blnc_no_rqs; + goto setup_migration_paths; + } +#endif goto all_active; + } if (!forced && half_full_scheds != blnc_no_rqs) { int min = 1; @@ -3849,15 +4205,30 @@ check_balance(ErtsRunQueue *c_rq) } } +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + setup_migration_paths: +#endif + /* Setup migration paths for all priorities */ for (pix = 0; pix < ERTS_NO_PRIO_LEVELS; pix++) { int low = 0, high = 0; for (qix = 0; qix < blnc_no_rqs; qix++) { int len_diff = run_queue_info[qix].prio[pix].max_len; len_diff -= run_queue_info[qix].prio[pix].migration_limit; + #ifdef DBG_PRINT if (pix == 2) erts_fprintf(stderr, "%d ", len_diff); #endif + +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + if (sched_util_balancing + && -ERTS_SCHED_UTIL_IGNORE_IMBALANCE_DIFF <= len_diff + && len_diff <= ERTS_SCHED_UTIL_IGNORE_IMBALANCE_DIFF) { + /* ignore minor imbalance */ + len_diff = 0; + } +#endif + run_queue_compare[qix].qix = qix; run_queue_compare[qix].len = len_diff; if (len_diff != 0) { @@ -3984,6 +4355,9 @@ erts_fprintf(stderr, "--------------------------------\n"); Uint32 flags = run_queue_info[qix].flags; ErtsMigrationPath *mp = &new_mpaths->mpath[qix]; +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + mp->sched_util = sched_util_balancing; +#endif mp->flags = flags; mp->misc_evac_runq = NULL; @@ -4575,11 +4949,17 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online) #endif init_misc_op_list_alloc(); + init_proc_sys_task_queues_alloc(); #ifdef ERTS_SMP set_wakeup_other_data(); #endif +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + if (erts_sched_balance_util) + erts_sched_compact_load = 0; +#endif + ASSERT(no_schedulers_online <= no_schedulers); ASSERT(no_schedulers_online >= 1); ASSERT(no_schedulers >= 1); @@ -4648,6 +5028,11 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online) rq->ports.info.reds = 0; rq->ports.start = NULL; rq->ports.end = NULL; + +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + init_runq_sched_util(&rq->sched_util, erts_sched_balance_util); +#endif + } #ifdef ERTS_SMP @@ -4746,6 +5131,7 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online) esdp->reductions = 0; init_sched_wall_time(&esdp->sched_wall_time); + erts_port_task_handle_init(&esdp->nosuspend_port_task_handle); } @@ -4858,46 +5244,166 @@ erts_get_scheduler_data(void) #endif +static Process * +make_proxy_proc(Process *prev_proxy, Process *proc, erts_aint32_t prio) +{ + erts_aint32_t state; + Process *proxy; +#ifdef ERTS_SMP + ErtsRunQueue *rq = RUNQ_READ_RQ(&proc->run_queue); +#endif + + state = (ERTS_PSFLG_PROXY + | ERTS_PSFLG_IN_RUNQ + | (((erts_aint32_t) 1) << (prio + ERTS_PSFLGS_IN_PRQ_MASK_OFFSET)) + | (prio << ERTS_PSFLGS_PRQ_PRIO_OFFSET) + | (prio << ERTS_PSFLGS_USR_PRIO_OFFSET) + | (prio << ERTS_PSFLGS_ACT_PRIO_OFFSET)); + + if (prev_proxy) { + proxy = prev_proxy; + ASSERT(erts_smp_atomic32_read_nob(&proxy->state) & ERTS_PSFLG_PROXY); + erts_smp_atomic32_set_nob(&proxy->state, state); +#ifdef ERTS_SMP + RUNQ_SET_RQ(&proc->run_queue, rq); +#endif + } + else { + proxy = erts_alloc(ERTS_ALC_T_PROC, sizeof(Process)); +#ifdef DEBUG + { + int i; + Uint32 *ui32 = (Uint32 *) (char *) proxy; + for (i = 0; i < sizeof(Process)/sizeof(Uint32); i++) + ui32[i] = (Uint32) 0xdeadbeef; + } +#endif + erts_smp_atomic32_init_nob(&proxy->state, state); +#ifdef ERTS_SMP + erts_smp_atomic_init_nob(&proxy->run_queue, + erts_smp_atomic_read_nob(&proc->run_queue)); +#endif + } + + proxy->common.id = proc->common.id; + + return proxy; +} + +static ERTS_INLINE void +free_proxy_proc(Process *proxy) +{ + ASSERT(erts_smp_atomic32_read_nob(&proxy->state) & ERTS_PSFLG_PROXY); + erts_free(ERTS_ALC_T_PROC, proxy); +} + + +static ERTS_INLINE int +check_enqueue_in_prio_queue(erts_aint32_t *prq_prio_p, + erts_aint32_t *newp, + erts_aint32_t actual) +{ + erts_aint32_t aprio, qbit, max_qbit; + + aprio = (actual >> ERTS_PSFLGS_ACT_PRIO_OFFSET) & ERTS_PSFLGS_PRIO_MASK; + qbit = 1 << aprio; + + *prq_prio_p = aprio; + + max_qbit = (actual >> ERTS_PSFLGS_IN_PRQ_MASK_OFFSET) & ERTS_PSFLGS_QMASK; + max_qbit |= 1 << ERTS_PSFLGS_QMASK_BITS; + max_qbit &= -max_qbit; + /* + * max_qbit now either contain bit set for highest prio queue or a bit + * out of range (which will have a value larger than valid range). + */ + + if (qbit >= max_qbit) + return 0; /* Already queued in higher or equal prio */ + + /* Need to enqueue (if already enqueued, it is in lower prio) */ + *newp |= qbit << ERTS_PSFLGS_IN_PRQ_MASK_OFFSET; + + if ((actual & (ERTS_PSFLG_IN_RUNQ|ERTS_PSFLGS_USR_PRIO_MASK)) + != (aprio << ERTS_PSFLGS_USR_PRIO_OFFSET)) { + /* + * Process struct already enqueued, or actual prio not + * equal to user prio, i.e., enqueue using proxy. + */ + return -1; + } + + /* + * Enqueue using process struct. + */ + *newp &= ~ERTS_PSFLGS_PRQ_PRIO_MASK; + *newp |= ERTS_PSFLG_IN_RUNQ | (aprio << ERTS_PSFLGS_PRQ_PRIO_OFFSET); + return 1; +} + /* * scheduler_out_process() return with c_rq locked. */ static ERTS_INLINE int -schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p) +schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p, Process *proxy) { - erts_aint32_t a, e, n; + erts_aint32_t a, e, n, enq_prio = -1; int res = 0; + int enqueue; /* < 0 -> use proxy */ a = state; while (1) { n = e = a; - ASSERT(a & ERTS_PSFLG_RUNNING); - ASSERT(!(a & ERTS_PSFLG_IN_RUNQ)); + ASSERT(a & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS)); + + enqueue = 0; - n &= ~ERTS_PSFLG_RUNNING; - if ((a & (ERTS_PSFLG_ACTIVE|ERTS_PSFLG_SUSPENDED)) == ERTS_PSFLG_ACTIVE) - n |= ERTS_PSFLG_IN_RUNQ; + n &= ~(ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS); + if (a & ERTS_PSFLG_ACTIVE_SYS + || (a & (ERTS_PSFLG_ACTIVE|ERTS_PSFLG_SUSPENDED)) == ERTS_PSFLG_ACTIVE) { + enqueue = check_enqueue_in_prio_queue(&enq_prio, &n, a); + } a = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e); if (a == e) break; } - if (!(n & ERTS_PSFLG_IN_RUNQ)) { - if (erts_system_profile_flags.runnable_procs) - profile_runnable_proc(p, am_inactive); + if (!enqueue) { + + if (erts_system_profile_flags.runnable_procs) { + + if (!(a & ERTS_PSFLG_ACTIVE_SYS) + && (!(a & ERTS_PSFLG_ACTIVE) + || (a & ERTS_PSFLG_SUSPENDED))) { + /* Process inactive */ + profile_runnable_proc(p, am_inactive); + } + } + + if (proxy) + free_proxy_proc(proxy); } else { - int prio = (int) (ERTS_PSFLG_PRIO_MASK & n); + Process *sched_p; ErtsRunQueue *runq = erts_get_runq_proc(p); - ASSERT(!(n & ERTS_PSFLG_SUSPENDED)); + ASSERT(!(n & ERTS_PSFLG_SUSPENDED) || (n & ERTS_PSFLG_ACTIVE_SYS)); + + if (enqueue < 0) + sched_p = make_proxy_proc(proxy, p, enq_prio); + else { + sched_p = p; + if (proxy) + free_proxy_proc(proxy); + } #ifdef ERTS_SMP if (!(ERTS_PSFLG_BOUND & n)) { - ErtsRunQueue *new_runq = erts_check_emigration_need(runq, prio); + ErtsRunQueue *new_runq = erts_check_emigration_need(runq, enq_prio); if (new_runq) { - RUNQ_SET_RQ(&p->run_queue, new_runq); + RUNQ_SET_RQ(&sched_p->run_queue, new_runq); runq = new_runq; } } @@ -4908,7 +5414,7 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p) erts_smp_runq_lock(runq); /* Enqueue the process */ - enqueue_process(runq, prio, p); + enqueue_process(runq, (int) enq_prio, sched_p); if (runq == c_rq) return res; @@ -4920,14 +5426,13 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p) } static ERTS_INLINE void -add2runq(Process *p, erts_aint32_t state) +add2runq(Process *p, erts_aint32_t state, erts_aint32_t prio) { - int prio = (int) (ERTS_PSFLG_PRIO_MASK & state); ErtsRunQueue *runq = erts_get_runq_proc(p); #ifdef ERTS_SMP if (!(ERTS_PSFLG_BOUND & state)) { - ErtsRunQueue *new_runq = erts_check_emigration_need(runq, prio); + ErtsRunQueue *new_runq = erts_check_emigration_need(runq, (int) prio); if (new_runq) { RUNQ_SET_RQ(&p->run_queue, new_runq); runq = new_runq; @@ -4939,101 +5444,236 @@ add2runq(Process *p, erts_aint32_t state) erts_smp_runq_lock(runq); /* Enqueue the process */ - enqueue_process(runq, prio, p); + enqueue_process(runq, (int) prio, p); erts_smp_runq_unlock(runq); smp_notify_inc_runq(runq); } -static ERTS_INLINE void -schedule_process(Process *p, erts_aint32_t state, int active_enq) +static ERTS_INLINE int +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 a = state, n; + /* + * NOTE: ERTS_PSFLG_RUNNING, ERTS_PSFLG_RUNNING_SYS and + * ERTS_PSFLG_ACTIVE_SYS are not allowed to be + * altered by this function! + */ + erts_aint32_t a = *statep, n; + int enqueue; /* < 0 -> use proxy */ + + ASSERT(!(a & ERTS_PSFLG_PROXY)); + ASSERT((clear_state_flags & (ERTS_PSFLG_RUNNING + | ERTS_PSFLG_RUNNING_SYS + | ERTS_PSFLG_ACTIVE_SYS)) == 0); + ASSERT((set_state_flags & (ERTS_PSFLG_RUNNING + | ERTS_PSFLG_RUNNING_SYS + | ERTS_PSFLG_ACTIVE_SYS)) == 0); while (1) { erts_aint32_t e; n = e = a; + enqueue = 0; + if (a & ERTS_PSFLG_FREE) - return; /* We don't want to schedule free processes... */ - n |= ERTS_PSFLG_ACTIVE; - if (!(a & (ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_RUNNING))) - n |= ERTS_PSFLG_IN_RUNQ; + break; /* We don't want to schedule free processes... */ + + if (clear_state_flags) + n &= ~clear_state_flags; + + if (set_state_flags) + n |= set_state_flags; + + if ((n & (ERTS_PSFLG_SUSPENDED + | ERTS_PSFLG_RUNNING + | ERTS_PSFLG_RUNNING_SYS + | ERTS_PSFLG_IN_RUNQ + | ERTS_PSFLG_ACTIVE)) == ERTS_PSFLG_ACTIVE) { + /* + * Active and seemingly need to be enqueued, but + * process may be in a run queue via proxy, need + * further inspection... + */ + enqueue = check_enqueue_in_prio_queue(enq_prio_p, &n, a); + } + a = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e); if (a == e) break; - if (!active_enq && (a & ERTS_PSFLG_ACTIVE)) - return; /* Someone else activated process ... */ + if (enqueue == 0 && n == a) + break; } - if (erts_system_profile_flags.runnable_procs - && !(a & (ERTS_PSFLG_ACTIVE|ERTS_PSFLG_SUSPENDED))) { - profile_runnable_proc(p, am_active); + if (erts_system_profile_flags.runnable_procs) { + + if (((n & (ERTS_PSFLG_SUSPENDED + | ERTS_PSFLG_ACTIVE)) == ERTS_PSFLG_ACTIVE) + && (!(a & (ERTS_PSFLG_ACTIVE_SYS + | ERTS_PSFLG_RUNNING + | ERTS_PSFLG_RUNNING_SYS) + && (!(a & ERTS_PSFLG_ACTIVE) + || (a & ERTS_PSFLG_SUSPENDED))))) { + /* We activated a prevously inactive process */ + profile_runnable_proc(p, am_active); + } + } - if ((n & ERTS_PSFLG_IN_RUNQ) && !(a & ERTS_PSFLG_IN_RUNQ)) - add2runq(p, n); + *statep = a; + + return enqueue; +} + +static ERTS_INLINE void +schedule_process(Process *p, erts_aint32_t in_state) +{ + erts_aint32_t enq_prio = -1; + erts_aint32_t state = in_state; + int enqueue = change_proc_schedule_state(p, + 0, + ERTS_PSFLG_ACTIVE, + &state, + &enq_prio); + if (enqueue) + add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio), + state, + enq_prio); } void erts_schedule_process(Process *p, erts_aint32_t state) { - schedule_process(p, state, 0); + schedule_process(p, state); +} + +static void +schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) +{ + erts_aint32_t a = state, n, enq_prio = -1; + int enqueue; /* < 0 -> use proxy */ + + ASSERT(!(state & ERTS_PSFLG_PROXY)); + + while (1) { + erts_aint32_t e; + n = e = a; + + if (a & ERTS_PSFLG_FREE) + return; /* We don't want to schedule free processes... */ + + enqueue = 0; + n |= ERTS_PSFLG_ACTIVE_SYS; + if (!(a & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS))) + enqueue = check_enqueue_in_prio_queue(&enq_prio, &n, a); + a = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e); + if (a == e) + break; + if (a == n && !enqueue) + goto cleanup; + } + + if (erts_system_profile_flags.runnable_procs) { + + if (!(a & (ERTS_PSFLG_ACTIVE_SYS + | ERTS_PSFLG_RUNNING + | ERTS_PSFLG_RUNNING_SYS)) + && (!(a & ERTS_PSFLG_ACTIVE) || (a & ERTS_PSFLG_SUSPENDED))) { + /* We activated a prevously inactive process */ + profile_runnable_proc(p, am_active); + } + + } + + if (enqueue) { + Process *sched_p; + if (enqueue > 0) + sched_p = p; + else { + sched_p = make_proxy_proc(proxy, p, enq_prio); + proxy = NULL; + } + add2runq(sched_p, n, enq_prio); + } + +cleanup: + if (proxy) + free_proxy_proc(proxy); } static ERTS_INLINE int suspend_process(Process *c_p, Process *p) { - erts_aint32_t state = erts_smp_atomic32_read_acqb(&p->state); + erts_aint32_t state; int suspended = 0; ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p)); + state = erts_smp_atomic32_read_acqb(&p->state); + if ((state & ERTS_PSFLG_SUSPENDED)) suspended = -1; else { if (c_p == p) { state = erts_smp_atomic32_read_bor_relb(&p->state, ERTS_PSFLG_SUSPENDED); - state |= ERTS_PSFLG_SUSPENDED; ASSERT(state & ERTS_PSFLG_RUNNING); - suspended = 1; + suspended = (state & ERTS_PSFLG_SUSPENDED) ? -1: 1; } else { while (!(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_EXITING))) { - erts_aint32_t e, n; + erts_aint32_t n, e; + n = e = state; n |= ERTS_PSFLG_SUSPENDED; state = erts_smp_atomic32_cmpxchg_relb(&p->state, n, e); if (state == e) { - state = n; suspended = 1; break; } + if (state & ERTS_PSFLG_SUSPENDED) { + suspended = -1; + break; + } } } } - if (state & ERTS_PSFLG_SUSPENDED) { + if (suspended) { ASSERT(!(ERTS_PSFLG_RUNNING & state) || p == erts_get_current_process()); - if (erts_system_profile_flags.runnable_procs - && (p->rcount == 0) - && (state & ERTS_PSFLG_ACTIVE)) { - profile_runnable_proc(p, am_inactive); + if (suspended > 0 && erts_system_profile_flags.runnable_procs) { + + /* 'state' is before our change... */ + + if ((state & (ERTS_PSFLG_ACTIVE + | ERTS_PSFLG_ACTIVE_SYS + | ERTS_PSFLG_RUNNING + | ERTS_PSFLG_RUNNING_SYS + | ERTS_PSFLG_SUSPENDED)) == ERTS_PSFLG_ACTIVE) { + /* We made process inactive */ + profile_runnable_proc(p, am_inactive); + } + } p->rcount++; /* count number of suspend */ } + return suspended; } static ERTS_INLINE void resume_process(Process *p) { - erts_aint32_t state; + erts_aint32_t state, enq_prio = -1; + int enqueue; + ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p)); ASSERT(p->rcount > 0); @@ -5041,14 +5681,16 @@ resume_process(Process *p) if (--p->rcount > 0) /* multiple suspend */ return; - state = erts_smp_atomic32_read_band_mb(&p->state, ~ERTS_PSFLG_SUSPENDED); - state &= ~ERTS_PSFLG_SUSPENDED; - if ((state & (ERTS_PSFLG_EXITING - | ERTS_PSFLG_ACTIVE - | ERTS_PSFLG_IN_RUNQ - | ERTS_PSFLG_RUNNING)) == ERTS_PSFLG_ACTIVE) { - schedule_process(p, state, 1); - } + state = erts_smp_atomic32_read_nob(&p->state); + enqueue = change_proc_schedule_state(p, + ERTS_PSFLG_SUSPENDED, + 0, + &state, + &enq_prio); + if (enqueue) + add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio), + state, + enq_prio); } int @@ -5457,7 +6099,7 @@ erts_set_schedulers_online(Process *p, for (ix = no; ix < online; ix++) { ErtsRunQueue *rq = ERTS_RUNQ_IX(ix); - wake_scheduler(rq, 0); + wake_scheduler(rq); } } } @@ -5556,7 +6198,7 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) for (ix = 1; ix < online; ix++) { ErtsRunQueue *rq = ERTS_RUNQ_IX(ix); - wake_scheduler(rq, 0); + wake_scheduler(rq); } if (erts_smp_atomic32_read_nob(&schdlr_sspnd.active) @@ -6032,7 +6674,8 @@ pid2proc_not_running(Process *c_p, ErtsProcLocks c_p_locks, goto done; } else { - if (!(ERTS_PSFLG_RUNNING & erts_smp_atomic32_read_acqb(&rp->state))) + if (!((ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS) + & erts_smp_atomic32_read_acqb(&rp->state))) goto done; } @@ -6695,7 +7338,7 @@ Eterm erts_get_process_priority(Process *p) { erts_aint32_t state = erts_smp_atomic32_read_nob(&p->state); - switch (state & ERTS_PSFLG_PRIO_MASK) { + switch (ERTS_PSFLGS_GET_USR_PRIO(state)) { case PRIORITY_MAX: return am_max; case PRIORITY_HIGH: return am_high; case PRIORITY_NORMAL: return am_normal; @@ -6718,18 +7361,68 @@ erts_set_process_priority(Process *p, Eterm value) } a = erts_smp_atomic32_read_nob(&p->state); - if (nprio == (a & ERTS_PSFLG_PRIO_MASK)) + if (nprio == ERTS_PSFLGS_GET_USR_PRIO(a)) oprio = nprio; else { - erts_aint32_t e, n; + int slocked = 0; + erts_aint32_t e, n, aprio; + + if (a & ERTS_PSFLG_ACTIVE_SYS) { + erts_smp_proc_lock(p, ERTS_PROC_LOCK_STATUS); + slocked = 1; + } + do { - oprio = a & ERTS_PSFLG_PRIO_MASK; + oprio = ERTS_PSFLGS_GET_USR_PRIO(a); n = e = a; - ASSERT(!(a & ERTS_PSFLG_IN_RUNQ)); + if (!(a & (ERTS_PSFLG_ACTIVE_SYS|ERTS_PSFLG_DELAYED_SYS))) + aprio = nprio; + else { + int max_qbit; + + if (!slocked) { + erts_smp_proc_lock(p, ERTS_PROC_LOCK_STATUS); + slocked = 1; + } + + max_qbit = 0; + if (a & ERTS_PSFLG_ACTIVE_SYS) + max_qbit |= p->sys_task_qs->qmask; + if (a & ERTS_PSFLG_DELAYED_SYS) { + ErtsProcSysTaskQs *qs; + qs = ERTS_PROC_GET_DELAYED_GC_TASK_QS(p); + ASSERT(qs); + max_qbit |= qs->qmask; + } + max_qbit &= -max_qbit; + switch (max_qbit) { + case MAX_BIT: + aprio = PRIORITY_MAX; + break; + case HIGH_BIT: + aprio = PRIORITY_HIGH; + break; + case NORMAL_BIT: + aprio = PRIORITY_NORMAL; + break; + case LOW_BIT: + aprio = PRIORITY_LOW; + break; + default: + ERTS_INTERNAL_ERROR("Invalid qmask"); + aprio = -1; + } + + if (aprio > nprio) /* low value -> high prio */ + aprio = nprio; + } + + n &= ~(ERTS_PSFLGS_USR_PRIO_MASK + | ERTS_PSFLGS_ACT_PRIO_MASK); + n |= ((nprio << ERTS_PSFLGS_USR_PRIO_OFFSET) + | (aprio << ERTS_PSFLGS_ACT_PRIO_OFFSET)); - n &= ~ERTS_PSFLG_PRIO_MASK; - n |= nprio; a = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e); } while (a != e); } @@ -6763,6 +7456,7 @@ erts_set_process_priority(Process *p, Eterm value) Process *schedule(Process *p, int calls) { + Process *proxy_p = NULL; ErtsRunQueue *rq; erts_aint_t dt; ErtsSchedulerData *esdp; @@ -6805,6 +7499,8 @@ Process *schedule(Process *p, int calls) actual_reds = reds = 0; erts_smp_runq_lock(rq); } else { + sched_out_proc: + #ifdef ERTS_SMP ERTS_SMP_CHK_HAVE_ONLY_MAIN_PROC_LOCK(p); esdp = p->scheduler_data; @@ -6858,10 +7554,11 @@ Process *schedule(Process *p, int calls) esdp->reductions += reds; - schedule_out_process(rq, state, p); /* Returns with rq locked! */ + schedule_out_process(rq, state, p, proxy_p); /* Returns with rq locked! */ + proxy_p = NULL; ERTS_PROC_REDUCTIONS_EXECUTED(rq, - (int) (state & ERTS_PSFLG_PRIO_MASK), + (int) ERTS_PSFLGS_GET_USR_PRIO(state), reds, actual_reds); @@ -6870,18 +7567,19 @@ Process *schedule(Process *p, int calls) p->scheduler_data = NULL; #endif + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_STATUS); if (state & ERTS_PSFLG_FREE) { #ifdef ERTS_SMP ASSERT(esdp->free_process == p); esdp->free_process = NULL; -#else - erts_free_proc(p); +#else + state = erts_smp_atomic32_read_nob(&p->state); + if (!(state & ERTS_PSFLG_IN_RUNQ)) + erts_free_proc(p); #endif } - erts_smp_proc_unlock(p, ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_STATUS); - #ifdef ERTS_SMP ASSERT(!esdp->free_process); #endif @@ -6934,7 +7632,7 @@ Process *schedule(Process *p, int calls) continue_check_activities_to_run: flags = ERTS_RUNQ_FLGS_GET_NOB(rq); continue_check_activities_to_run_known_flags: - + ASSERT(flags & ERTS_RUNQ_FLG_NONEMPTY); if (flags & (ERTS_RUNQ_FLG_CHK_CPU_BIND|ERTS_RUNQ_FLG_SUSPENDED)) { @@ -6986,20 +7684,16 @@ Process *schedule(Process *p, int calls) rq->wakeup_other = 0; rq->wakeup_other_reds = 0; - empty_runq(rq); - flags = ERTS_RUNQ_FLGS_GET_NOB(rq); - if (flags & ERTS_RUNQ_FLG_SUSPENDED) { - non_empty_runq(rq); + if (flags & ERTS_RUNQ_FLG_SUSPENDED) goto continue_check_activities_to_run_known_flags; - } - else if (!(flags & ERTS_RUNQ_FLG_INACTIVE)) { - if (try_steal_task(rq)) { - non_empty_runq(rq); + if (flags & ERTS_RUNQ_FLG_INACTIVE) + empty_runq(rq); + else { + if (try_steal_task(rq)) goto continue_check_activities_to_run; - } - (void) ERTS_RUNQ_FLGS_UNSET(rq, ERTS_RUNQ_FLG_PROTECTED); + empty_runq(rq); /* * Check for ERTS_RUNQ_FLG_SUSPENDED has to be done @@ -7011,7 +7705,6 @@ Process *schedule(Process *p, int calls) goto continue_check_activities_to_run_known_flags; } } - #endif scheduler_wait(&fcalls, esdp, rq); @@ -7088,6 +7781,8 @@ Process *schedule(Process *p, int calls) * Find a new process to run. */ pick_next_process: { + erts_aint32_t psflg_band_mask; + erts_aint32_t running_flag; int prio_q; int qmask; @@ -7121,18 +7816,62 @@ Process *schedule(Process *p, int calls) ASSERT(p); /* Wrong qmask in rq->flags? */ + psflg_band_mask = ~(((erts_aint32_t) 1) << (ERTS_PSFLGS_GET_PRQ_PRIO(state) + + ERTS_PSFLGS_IN_PRQ_MASK_OFFSET)); + + if (!(state & ERTS_PSFLG_PROXY)) + psflg_band_mask &= ~ERTS_PSFLG_IN_RUNQ; + else { + proxy_p = p; + p = erts_proc_lookup_raw(proxy_p->common.id); + if (!p) { + free_proxy_proc(proxy_p); + proxy_p = NULL; + goto pick_next_process; + } + state = erts_smp_atomic32_read_nob(&p->state); + } + + + if (state & ERTS_PSFLG_ACTIVE_SYS) + running_flag = ERTS_PSFLG_RUNNING_SYS; + else + running_flag = ERTS_PSFLG_RUNNING; + while (1) { erts_aint32_t exp, new, tmp; tmp = new = exp = state; - new &= ~ERTS_PSFLG_IN_RUNQ; - tmp = state & (ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT); - if (tmp != ERTS_PSFLG_SUSPENDED) - new |= ERTS_PSFLG_RUNNING; + new &= psflg_band_mask; + if (!(state & (ERTS_PSFLG_RUNNING + | ERTS_PSFLG_RUNNING_SYS))) { + tmp = state & (ERTS_PSFLG_SUSPENDED + | ERTS_PSFLG_PENDING_EXIT + | ERTS_PSFLG_ACTIVE_SYS); + if (tmp != ERTS_PSFLG_SUSPENDED) + new |= running_flag; + } state = erts_smp_atomic32_cmpxchg_relb(&p->state, new, exp); if (state == exp) { - tmp = state & (ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT); - if (tmp == ERTS_PSFLG_SUSPENDED) + if ((state & (ERTS_PSFLG_RUNNING + | ERTS_PSFLG_RUNNING_SYS + | ERTS_PSFLG_FREE)) + || ((state & (ERTS_PSFLG_SUSPENDED + | ERTS_PSFLG_PENDING_EXIT + | ERTS_PSFLG_ACTIVE_SYS)) + == ERTS_PSFLG_SUSPENDED)) { + if (state & ERTS_PSFLG_FREE) { +#ifdef ERTS_SMP + erts_smp_proc_dec_refc(p); +#else + erts_free_proc(p); +#endif + } + if (proxy_p) { + free_proxy_proc(proxy_p); + proxy_p = NULL; + } goto pick_next_process; + } state = new; break; } @@ -7162,7 +7901,7 @@ Process *schedule(Process *p, int calls) (UWord) esdp->no); int migrated = old && old != esdp->no; - prio = (int) (state & ERTS_PSFLG_PRIO_MASK); + prio = (int) ERTS_PSFLGS_GET_USR_PRIO(state); erts_smp_spin_lock(&erts_sched_stat.lock); erts_sched_stat.prio[prio].total_executed++; @@ -7182,9 +7921,6 @@ Process *schedule(Process *p, int calls) ASSERT(!p->scheduler_data); p->scheduler_data = esdp; #endif - /* Never run a suspended process */ - ASSERT(!(ERTS_PSFLG_SUSPENDED & erts_smp_atomic32_read_nob(&p->state))); - reds = context_reds; if (IS_TRACED(p)) { @@ -7210,21 +7946,780 @@ Process *schedule(Process *p, int calls) erts_check_my_tracer_proc(p); #endif + if (state & ERTS_PSFLG_RUNNING_SYS) { + reds -= execute_sys_tasks(p, &state, reds); + if (reds <= 0) { + p->fcalls = reds; + goto sched_out_proc; + } + + ASSERT(state & ERTS_PSFLG_RUNNING_SYS); + ASSERT(!(state & ERTS_PSFLG_RUNNING)); + + while (1) { + erts_aint32_t n, e; + + if (((state & (ERTS_PSFLG_SUSPENDED + | ERTS_PSFLG_ACTIVE)) != ERTS_PSFLG_ACTIVE) + && !(state & ERTS_PSFLG_EXITING)) + goto sched_out_proc; + + n = e = state; + n &= ~ERTS_PSFLG_RUNNING_SYS; + n |= ERTS_PSFLG_RUNNING; + + state = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e); + if (state == e) { + state = n; + break; + } + + ASSERT(state & ERTS_PSFLG_RUNNING_SYS); + ASSERT(!(state & ERTS_PSFLG_RUNNING)); + } + } + if (!(state & ERTS_PSFLG_EXITING) && ((FLAGS(p) & F_FORCE_GC) || (MSO(p).overhead > BIN_VHEAP_SZ(p)))) { reds -= erts_garbage_collect(p, 0, p->arg_reg, p->arity); - if (reds < 0) { - reds = 1; + if (reds <= 0) { + p->fcalls = reds; + goto sched_out_proc; } } + + if (proxy_p) { + free_proxy_proc(proxy_p); + proxy_p = NULL; + } p->fcalls = reds; ERTS_SMP_CHK_HAVE_ONLY_MAIN_PROC_LOCK(p); + + /* Never run a suspended process */ + ASSERT(!(ERTS_PSFLG_SUSPENDED & erts_smp_atomic32_read_nob(&p->state))); + return p; } } +static int +notify_sys_task_executed(Process *c_p, ErtsProcSysTask *st, Eterm st_result) +{ + Process *rp = erts_proc_lookup(st->requester); + if (rp) { + ErtsProcLocks rp_locks; + ErlOffHeap *ohp; + ErlHeapFragment* bp; + Eterm *hp, msg, req_id, result; + Uint st_result_sz, hsz; +#ifdef DEBUG + Eterm *hp_start; +#endif + + rp_locks = (c_p == rp) ? ERTS_PROC_LOCK_MAIN : 0; + + st_result_sz = is_immed(st_result) ? 0 : size_object(st_result); + hsz = st->req_id_sz + st_result_sz + 4 /* 3-tuple */; + + hp = erts_alloc_message_heap(hsz, + &bp, + &ohp, + rp, + &rp_locks); + +#ifdef DEBUG + hp_start = hp; +#endif + + req_id = st->req_id_sz == 0 ? st->req_id : copy_struct(st->req_id, + st->req_id_sz, + &hp, + ohp); + + result = st_result_sz == 0 ? st_result : copy_struct(st_result, + st_result_sz, + &hp, + ohp); + + ASSERT(is_immed(st->reply_tag)); + + msg = TUPLE3(hp, st->reply_tag, req_id, result); + +#ifdef DEBUG + hp += 4; + ASSERT(hp_start + hsz == hp); +#endif + + erts_queue_message(rp, + &rp_locks, + bp, + msg, + NIL +#ifdef USE_VM_PROBES + , NIL +#endif + ); + + if (c_p == rp) + rp_locks &= ~ERTS_PROC_LOCK_MAIN; + + if (rp_locks) + erts_smp_proc_unlock(rp, rp_locks); + } + + erts_cleanup_offheap(&st->off_heap); + + erts_free(ERTS_ALC_T_PROC_SYS_TSK, st); + + return rp ? 1 : 0; +} + +static ERTS_INLINE ErtsProcSysTask * +fetch_sys_task(Process *c_p, erts_aint32_t state, int *qmaskp, int *priop) +{ + ErtsProcSysTaskQs *unused_qs = NULL; + int qbit, qmask; + ErtsProcSysTask *st, **qp; + + *priop = -1; /* Shut up annoying erroneous warning */ + + erts_smp_proc_lock(c_p, ERTS_PROC_LOCK_STATUS); + + if (!c_p->sys_task_qs) { + qmask = 0; + st = NULL; + goto update_state; + } + + qmask = c_p->sys_task_qs->qmask; + + if ((state & (ERTS_PSFLG_ACTIVE + | ERTS_PSFLG_EXITING + | ERTS_PSFLG_SUSPENDED)) == ERTS_PSFLG_ACTIVE) { + /* No sys tasks if we got exclusively higher prio user work to do */ + st = NULL; + switch (ERTS_PSFLGS_GET_USR_PRIO(state)) { + case PRIORITY_MAX: + if (!(qmask & MAX_BIT)) + goto done; + break; + case PRIORITY_HIGH: + if (!(qmask & (MAX_BIT|HIGH_BIT))) + goto done; + break; + default: + break; + } + } + + qbit = qmask & -qmask; + switch (qbit) { + case MAX_BIT: + qp = &c_p->sys_task_qs->q[PRIORITY_MAX]; + *priop = PRIORITY_MAX; + break; + case HIGH_BIT: + qp = &c_p->sys_task_qs->q[PRIORITY_HIGH]; + *priop = PRIORITY_HIGH; + break; + case NORMAL_BIT: + if (!(qmask & PRIORITY_LOW) + || ++c_p->sys_task_qs->ncount <= RESCHEDULE_LOW) { + qp = &c_p->sys_task_qs->q[PRIORITY_NORMAL]; + *priop = PRIORITY_NORMAL; + break; + } + c_p->sys_task_qs->ncount = 0; + /* Fall through */ + case LOW_BIT: + qp = &c_p->sys_task_qs->q[PRIORITY_LOW]; + *priop = PRIORITY_LOW; + break; + default: + ERTS_INTERNAL_ERROR("Invalid qmask"); + } + + st = *qp; + ASSERT(st); + if (st->next != st) { + *qp = st->next; + st->next->prev = st->prev; + st->prev->next = st->next; + } + else { + erts_aint32_t a, e, n, st_prio, qmask2; + + *qp = NULL; + qmask &= ~qbit; + c_p->sys_task_qs->qmask = qmask; + + update_state: + + qmask2 = qmask; + + if (state & ERTS_PSFLG_DELAYED_SYS) { + ErtsProcSysTaskQs *qs = ERTS_PROC_GET_DELAYED_GC_TASK_QS(c_p); + ASSERT(qs); + qmask2 |= qs->qmask; + } + + switch (qmask2 & -qmask2) { + case MAX_BIT: + st_prio = PRIORITY_MAX; + break; + case HIGH_BIT: + st_prio = PRIORITY_HIGH; + break; + case NORMAL_BIT: + st_prio = PRIORITY_NORMAL; + break; + case LOW_BIT: + case 0: + st_prio = PRIORITY_LOW; + break; + default: + ERTS_INTERNAL_ERROR("Invalid qmask"); + } + + if (!qmask) { + unused_qs = c_p->sys_task_qs; + c_p->sys_task_qs = NULL; + } + + a = state; + do { + erts_aint32_t prio = ERTS_PSFLGS_GET_USR_PRIO(a); + + if (prio > st_prio) + prio = st_prio; + + n = e = a; + + n &= ~ERTS_PSFLGS_ACT_PRIO_MASK; + n |= (prio << ERTS_PSFLGS_ACT_PRIO_OFFSET); + + if (!qmask) + n &= ~ERTS_PSFLG_ACTIVE_SYS; + + if (a == n) + break; + a = erts_smp_atomic32_cmpxchg_nob(&c_p->state, n, e); + } while (a != e); + } + +done: + + erts_smp_proc_unlock(c_p, ERTS_PROC_LOCK_STATUS); + + if (unused_qs) + proc_sys_task_queues_free(unused_qs); + + *qmaskp = qmask; + + return st; +} + +static void save_gc_task(Process *c_p, ErtsProcSysTask *st, int prio); + +static int +execute_sys_tasks(Process *c_p, erts_aint32_t *statep, int in_reds) +{ + int garbage_collected = 0; + erts_aint32_t state = *statep; + int max_reds = in_reds; + int reds = 0; + int qmask = 0; + + ERTS_SMP_LC_ASSERT(erts_proc_lc_my_proc_locks(c_p) == ERTS_PROC_LOCK_MAIN); + + do { + ErtsProcSysTask *st; + int st_prio; + Eterm st_res; + + if (state & (ERTS_PSFLG_EXITING|ERTS_PSFLG_PENDING_EXIT)) { +#ifdef ERTS_SMP + if (state & ERTS_PSFLG_PENDING_EXIT) + erts_handle_pending_exit(c_p, ERTS_PROC_LOCK_MAIN); +#endif + ASSERT(ERTS_PROC_IS_EXITING(c_p)); + break; + } + + st = fetch_sys_task(c_p, state, &qmask, &st_prio); + if (!st) + break; + + switch (st->type) { + case ERTS_PSTT_GC: + if (c_p->flags & F_DISABLE_GC) { + save_gc_task(c_p, st, st_prio); + st = NULL; + reds++; + } + else { + if (!garbage_collected) { + FLAGS(c_p) |= F_NEED_FULLSWEEP; + reds += erts_garbage_collect(c_p, + 0, + c_p->arg_reg, + c_p->arity); + garbage_collected = 1; + } + st_res = am_true; + } + break; + case ERTS_PSTT_CPC: + st_res = erts_check_process_code(c_p, + st->arg[0], + st->arg[1] == am_true, + &reds); + if (is_non_value(st_res)) { + /* Needed gc, but gc was disabled */ + save_gc_task(c_p, st, st_prio); + st = NULL; + } + break; + default: + ERTS_INTERNAL_ERROR("Invalid process sys task type"); + st_res = am_false; + } + + if (st) + reds += notify_sys_task_executed(c_p, st, st_res); + + state = erts_smp_atomic32_read_acqb(&c_p->state); + } while (qmask && reds < max_reds); + + *statep = state; + + return reds; +} + +static int +cleanup_sys_tasks(Process *c_p, erts_aint32_t in_state, int in_reds) +{ + erts_aint32_t state = in_state; + int max_reds = in_reds; + int reds = 0; + int qmask = 0; + + ERTS_SMP_LC_ASSERT(erts_proc_lc_my_proc_locks(c_p) == ERTS_PROC_LOCK_MAIN); + + do { + ErtsProcSysTask *st; + Eterm st_res; + int st_prio; + + st = fetch_sys_task(c_p, state, &qmask, &st_prio); + if (!st) + break; + + switch (st->type) { + case ERTS_PSTT_GC: + st_res = am_false; + break; + case ERTS_PSTT_CPC: + st_res = am_false; + break; + default: + ERTS_INTERNAL_ERROR("Invalid process sys task type"); + st_res = am_false; + break; + } + + reds += notify_sys_task_executed(c_p, st, st_res); + + state = erts_smp_atomic32_read_acqb(&c_p->state); + } while (qmask && reds < max_reds); + + return reds; +} + +BIF_RETTYPE +erts_internal_request_system_task_3(BIF_ALIST_3) +{ + Process *rp = erts_proc_lookup(BIF_ARG_1); + ErtsProcSysTaskQs *stqs, *free_stqs = NULL; + ErtsProcSysTask *st = NULL; + erts_aint32_t prio, rp_state; + int rp_locked; + Eterm noproc_res, req_type; + + if (!rp && !is_internal_pid(BIF_ARG_1)) { + if (!is_external_pid(BIF_ARG_1)) + goto badarg; + if (external_pid_dist_entry(BIF_ARG_1) != erts_this_dist_entry) + goto badarg; + } + + switch (BIF_ARG_2) { + case am_max: prio = PRIORITY_MAX; break; + case am_high: prio = PRIORITY_HIGH; break; + case am_normal: prio = PRIORITY_NORMAL; break; + case am_low: prio = PRIORITY_LOW; break; + default: goto badarg; + } + + if (is_not_tuple(BIF_ARG_3)) + goto badarg; + else { + int i; + Eterm *tp = tuple_val(BIF_ARG_3); + Uint arity = arityval(*tp); + Eterm req_id; + Uint req_id_sz; + Eterm arg[ERTS_MAX_PROC_SYS_TASK_ARGS]; + Uint arg_sz[ERTS_MAX_PROC_SYS_TASK_ARGS]; + Uint tot_sz; + Eterm *hp; + + if (arity < 2) + goto badarg; + if (arity > 2 + ERTS_MAX_PROC_SYS_TASK_ARGS) + goto badarg; + req_type = tp[1]; + req_id = tp[2]; + req_id_sz = is_immed(req_id) ? req_id : size_object(req_id); + tot_sz = req_id_sz; + for (i = 0; i < ERTS_MAX_PROC_SYS_TASK_ARGS; i++) { + int tix = 3 + i; + if (tix > arity) { + arg[i] = THE_NON_VALUE; + arg_sz[i] = 0; + } + else { + arg[i] = tp[tix]; + if (is_immed(arg[i])) + arg_sz[i] = 0; + else { + arg_sz[i] = size_object(arg[i]); + tot_sz += arg_sz[i]; + } + } + } + st = erts_alloc(ERTS_ALC_T_PROC_SYS_TSK, + ERTS_PROC_SYS_TASK_SIZE(tot_sz)); + st->next = st->prev = st; /* Prep for empty prio queue */ + ERTS_INIT_OFF_HEAP(&st->off_heap); + hp = &st->heap[0]; + + st->requester = BIF_P->common.id; + st->reply_tag = req_type; + st->req_id_sz = req_id_sz; + st->req_id = req_id_sz == 0 ? req_id : copy_struct(req_id, + req_id_sz, + &hp, + &st->off_heap); + + for (i = 0; i < ERTS_MAX_PROC_SYS_TASK_ARGS; i++) + st->arg[i] = arg_sz[i] == 0 ? arg[i] : copy_struct(arg[i], + arg_sz[i], + &hp, + &st->off_heap); + ASSERT(&st->heap[0] + tot_sz == hp); + } + + switch (req_type) { + + case am_garbage_collect: + st->type = ERTS_PSTT_GC; + noproc_res = am_false; + if (!rp) + goto noproc; + break; + + case am_check_process_code: + if (is_not_atom(st->arg[0])) + goto badarg; + if (st->arg[1] != am_true && st->arg[1] != am_false) + goto badarg; + noproc_res = am_false; + st->type = ERTS_PSTT_CPC; + if (!rp) + goto noproc; + break; + + default: + goto badarg; + } + + rp_state = erts_smp_atomic32_read_nob(&rp->state); + + rp_locked = 0; + + free_stqs = NULL; + if (rp_state & ERTS_PSFLG_ACTIVE_SYS) + stqs = NULL; + else { + alloc_qs: + stqs = proc_sys_task_queues_alloc(); + stqs->qmask = 1 << prio; + stqs->ncount = 0; + stqs->q[PRIORITY_MAX] = NULL; + stqs->q[PRIORITY_HIGH] = NULL; + stqs->q[PRIORITY_NORMAL] = NULL; + stqs->q[PRIORITY_LOW] = NULL; + stqs->q[prio] = st; + } + + if (!rp_locked) { + rp_locked = 1; + erts_smp_proc_lock(rp, ERTS_PROC_LOCK_STATUS); + + rp_state = erts_smp_atomic32_read_nob(&rp->state); + if (rp_state & ERTS_PSFLG_EXITING) { + erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_STATUS); + rp = NULL; + free_stqs = stqs; + goto noproc; + } + } + + if (!rp->sys_task_qs) { + if (stqs) + rp->sys_task_qs = stqs; + else + goto alloc_qs; + } + else { + if (stqs) + free_stqs = stqs; + stqs = rp->sys_task_qs; + if (!stqs->q[prio]) { + stqs->q[prio] = st; + stqs->qmask |= 1 << prio; + } + else { + st->next = stqs->q[prio]; + st->prev = stqs->q[prio]->prev; + st->next->prev = st; + st->prev->next = st; + ASSERT(stqs->qmask & (1 << prio)); + } + } + + if (ERTS_PSFLGS_GET_ACT_PRIO(rp_state) > prio) { + erts_aint32_t n, a, e; + /* Need to elevate actual prio */ + + a = rp_state; + do { + if (ERTS_PSFLGS_GET_ACT_PRIO(a) <= prio) { + n = a; + break; + } + n = e = a; + n &= ~ERTS_PSFLGS_ACT_PRIO_MASK; + n |= (prio << ERTS_PSFLGS_ACT_PRIO_OFFSET); + a = erts_smp_atomic32_cmpxchg_nob(&rp->state, n, e); + } while (a != e); + rp_state = n; + } + + erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_STATUS); + + schedule_process_sys_task(rp, rp_state, NULL); + + if (free_stqs) + proc_sys_task_queues_free(free_stqs); + + BIF_RET(am_ok); + +noproc: + + notify_sys_task_executed(BIF_P, st, noproc_res); + if (free_stqs) + proc_sys_task_queues_free(free_stqs); + BIF_RET(am_ok); + +badarg: + + if (st) { + erts_cleanup_offheap(&st->off_heap); + erts_free(ERTS_ALC_T_PROC_SYS_TSK, st); + } + if (free_stqs) + proc_sys_task_queues_free(free_stqs); + BIF_ERROR(BIF_P, BADARG); +} + +static void +save_gc_task(Process *c_p, ErtsProcSysTask *st, int prio) +{ + erts_aint32_t state; + ErtsProcSysTaskQs *qs; + + ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_MAIN == erts_proc_lc_my_proc_locks(c_p)); + + qs = ERTS_PROC_GET_DELAYED_GC_TASK_QS(c_p); + if (!qs) { + st->next = st->prev = st; + qs = proc_sys_task_queues_alloc(); + qs->qmask = 1 << prio; + qs->ncount = 0; + qs->q[PRIORITY_MAX] = NULL; + qs->q[PRIORITY_HIGH] = NULL; + qs->q[PRIORITY_NORMAL] = NULL; + qs->q[PRIORITY_LOW] = NULL; + qs->q[prio] = st; + (void) ERTS_PROC_SET_DELAYED_GC_TASK_QS(c_p, ERTS_PROC_LOCK_MAIN, qs); + } + else { + if (!qs->q[prio]) { + st->next = st->prev = st; + qs->q[prio] = st; + qs->qmask |= 1 << prio; + } + else { + st->next = qs->q[prio]; + st->prev = qs->q[prio]->prev; + st->next->prev = st; + st->prev->next = st; + ASSERT(qs->qmask & (1 << prio)); + } + } + + state = erts_smp_atomic32_read_nob(&c_p->state); + ASSERT((ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS) & state); + + while (!(state & ERTS_PSFLG_DELAYED_SYS) + || prio < ERTS_PSFLGS_GET_ACT_PRIO(state)) { + erts_aint32_t n, e; + + n = e = state; + n |= ERTS_PSFLG_DELAYED_SYS; + if (prio < ERTS_PSFLGS_GET_ACT_PRIO(state)) { + n &= ~ERTS_PSFLGS_ACT_PRIO_MASK; + n |= prio << ERTS_PSFLGS_ACT_PRIO_OFFSET; + } + state = erts_smp_atomic32_cmpxchg_relb(&c_p->state, n, e); + if (state == e) + break; + } +} + +int +erts_set_gc_state(Process *c_p, int enable) +{ + ErtsProcSysTaskQs *dgc_tsk_qs; + ASSERT(c_p == erts_get_current_process()); + ASSERT((ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS) + & erts_smp_atomic32_read_nob(&c_p->state)); + ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_MAIN == erts_proc_lc_my_proc_locks(c_p)); + + if (!enable) { + c_p->flags |= F_DISABLE_GC; + return 0; + } + + c_p->flags &= ~F_DISABLE_GC; + + dgc_tsk_qs = ERTS_PROC_GET_DELAYED_GC_TASK_QS(c_p); + if (!dgc_tsk_qs) + return 0; + + /* Move delayed gc tasks into sys tasks queues. */ + + erts_smp_proc_lock(c_p, ERTS_PROC_LOCK_STATUS); + + if (!c_p->sys_task_qs) { + c_p->sys_task_qs = dgc_tsk_qs; + dgc_tsk_qs = NULL; + } + else { + ErtsProcSysTaskQs *stsk_qs; + int prio; + + /* + * We push delayed tasks to the front of the queue + * since they have already made it to the front + * once and then been delayed after that. + */ + + stsk_qs = c_p->sys_task_qs; + + while (dgc_tsk_qs->qmask) { + int qbit = dgc_tsk_qs->qmask & -dgc_tsk_qs->qmask; + dgc_tsk_qs->qmask &= ~qbit; + switch (qbit) { + case MAX_BIT: + prio = PRIORITY_MAX; + break; + case HIGH_BIT: + prio = PRIORITY_HIGH; + break; + case NORMAL_BIT: + prio = PRIORITY_NORMAL; + break; + case LOW_BIT: + prio = PRIORITY_LOW; + break; + default: + ERTS_INTERNAL_ERROR("Invalid qmask"); + prio = -1; + break; + } + + ASSERT(dgc_tsk_qs->q[prio]); + + if (!stsk_qs->q[prio]) { + stsk_qs->q[prio] = dgc_tsk_qs->q[prio]; + stsk_qs->qmask |= 1 << prio; + } + else { + ErtsProcSysTask *first1, *last1, *first2, *last2; + + ASSERT(stsk_qs->qmask & (1 << prio)); + first1 = dgc_tsk_qs->q[prio]; + last1 = first1->prev; + first2 = stsk_qs->q[prio]; + last2 = first1->prev; + + last1->next = first2; + first2->prev = last1; + + first1->prev = last2; + last2->next = first1; + + stsk_qs->q[prio] = first1; + } + + } + } + +#ifdef DEBUG + { + int qmask; + erts_aint32_t aprio, state = +#endif + + erts_smp_atomic32_read_bset_nob(&c_p->state, + (ERTS_PSFLG_DELAYED_SYS + | ERTS_PSFLG_ACTIVE_SYS), + ERTS_PSFLG_ACTIVE_SYS); + +#ifdef DEBUG + ASSERT(state & ERTS_PSFLG_DELAYED_SYS); + qmask = c_p->sys_task_qs->qmask; + aprio = ERTS_PSFLGS_GET_ACT_PRIO(state); + ASSERT(ERTS_PSFLGS_GET_USR_PRIO(state) >= aprio); + ASSERT((qmask & -qmask) >= (1 << aprio)); + } +#endif + + erts_smp_proc_unlock(c_p, ERTS_PROC_LOCK_STATUS); + + (void) ERTS_PROC_SET_DELAYED_GC_TASK_QS(c_p, ERTS_PROC_LOCK_MAIN, NULL); + + if (dgc_tsk_qs) + proc_sys_task_queues_free(dgc_tsk_qs); + + return 1; +} + void erts_sched_stat_modify(int what) { @@ -7324,6 +8819,10 @@ erts_schedule_misc_op(void (*func)(void *), void *arg) rq->misc.start = molp; rq->misc.end = molp; +#ifdef ERTS_SMP + non_empty_runq(rq); +#endif + erts_smp_runq_unlock(rq); smp_notify_inc_runq(rq); @@ -7521,7 +9020,8 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). prio = (erts_aint32_t) so->priority; } - state |= (prio & ERTS_PSFLG_PRIO_MASK); + state |= (((prio & ERTS_PSFLGS_PRIO_MASK) << ERTS_PSFLGS_ACT_PRIO_OFFSET) + | ((prio & ERTS_PSFLGS_PRIO_MASK) << ERTS_PSFLGS_USR_PRIO_OFFSET)); if (!rq) rq = erts_get_runq_proc(parent); @@ -7592,13 +9092,14 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). p->htop = p->heap; p->heap_sz = sz; p->catches = 0; - p->extra_root = NULL; p->bin_vheap_sz = p->min_vheap_size; p->bin_old_vheap_sz = p->min_vheap_size; p->bin_old_vheap = 0; p->bin_vheap_mature = 0; + p->sys_task_qs = NULL; + /* No need to initialize p->fcalls. */ p->current = p->initial+INITIAL_MOD; @@ -7764,7 +9265,7 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). * Schedule process for execution. */ - schedule_process(p, state, 0); + schedule_process(p, state); VERBOSE(DEBUG_PROCESSES, ("Created a new process: %T\n",p->common.id)); @@ -7815,6 +9316,7 @@ void erts_init_empty_process(Process *p) p->bin_vheap_sz = BIN_VH_MIN_SIZE; p->bin_old_vheap_sz = BIN_VH_MIN_SIZE; p->bin_old_vheap = 0; + p->sys_task_qs = NULL; p->bin_vheap_mature = 0; #ifdef ERTS_SMP p->common.u.alive.ptimer = NULL; @@ -8070,33 +9572,21 @@ delete_process(Process* p) p->fvalue = NIL; } -static ERTS_INLINE erts_aint32_t -set_proc_exiting_state(Process *p, erts_aint32_t state) -{ - erts_aint32_t a, n, e; - a = state; - while (1) { - n = e = a; - n &= ~(ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT); - n |= ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE; - if (!(a & (ERTS_PSFLG_IN_RUNQ|ERTS_PSFLG_RUNNING))) - n |= ERTS_PSFLG_IN_RUNQ; - a = erts_smp_atomic32_cmpxchg_relb(&p->state, n, e); - if (a == e) - break; - } - return a; -} - static ERTS_INLINE void set_proc_exiting(Process *p, - erts_aint32_t state, + erts_aint32_t in_state, Eterm reason, ErlHeapFragment *bp) { + erts_aint32_t state = in_state, enq_prio = -1; + int enqueue; ERTS_SMP_LC_ASSERT(erts_proc_lc_my_proc_locks(p) == ERTS_PROC_LOCKS_ALL); - state = set_proc_exiting_state(p, state); + enqueue = change_proc_schedule_state(p, + ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT, + ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE, + &state, + &enq_prio); p->fvalue = reason; if (bp) @@ -8111,15 +9601,37 @@ set_proc_exiting(Process *p, cancel_timer(p); p->i = (BeamInstr *) beam_exit; - if (erts_system_profile_flags.runnable_procs - && !(state & (ERTS_PSFLG_ACTIVE|ERTS_PSFLG_SUSPENDED))) { - profile_runnable_proc(p, am_active); - } - - if (!(state & (ERTS_PSFLG_IN_RUNQ|ERTS_PSFLG_RUNNING))) - add2runq(p, state); + if (enqueue) + add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio), + state, + enq_prio); } +static ERTS_INLINE erts_aint32_t +set_proc_self_exiting(Process *c_p) +{ +#ifdef DEBUG + int enqueue; +#endif + erts_aint32_t state, enq_prio = -1; + + ERTS_SMP_LC_ASSERT(erts_proc_lc_my_proc_locks(c_p) == ERTS_PROC_LOCKS_ALL); + + state = erts_smp_atomic32_read_nob(&c_p->state); + ASSERT(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS)); + +#ifdef DEBUG + enqueue = +#endif + change_proc_schedule_state(c_p, + ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT, + ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE, + &state, + &enq_prio); + + ASSERT(!enqueue); + return state; +} #ifdef ERTS_SMP @@ -8167,7 +9679,7 @@ handle_pending_exiters(ErtsProcList *pnd_xtrs) if (p) { if (erts_proclist_same(plp, p)) { erts_aint32_t state = erts_smp_atomic32_read_acqb(&p->state); - if (!(state & ERTS_PSFLG_RUNNING)) { + if (!(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS))) { ASSERT(state & ERTS_PSFLG_PENDING_EXIT); erts_handle_pending_exit(p, ERTS_PROC_LOCKS_ALL); } @@ -8196,8 +9708,11 @@ save_pending_exiter(Process *p) erts_proclist_store_last(&rq->procs.pending_exiters, plp); + non_empty_runq(rq); + erts_smp_runq_unlock(rq); - wake_scheduler(rq, 1); + + wake_scheduler(rq); } #endif @@ -8388,7 +9903,7 @@ send_exit_signal(Process *c_p, /* current process if and only } set_proc_exiting(c_p, state, rsn, NULL); } - else if (!(state & ERTS_PSFLG_RUNNING)) { + else if (!(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS))) { /* Process not running ... */ ErtsProcLocks need_locks = ~(*rp_locks) & ERTS_PROC_LOCKS_ALL; if (need_locks @@ -8765,9 +10280,6 @@ resume_suspend_monitor(ErtsSuspendMonitor *smon, void *vc_p) void erts_do_exit_process(Process* p, Eterm reason) { -#ifdef ERTS_SMP - erts_aint32_t state; -#endif p->arity = 0; /* No live registers */ p->fvalue = reason; @@ -8792,10 +10304,9 @@ erts_do_exit_process(Process* p, Eterm reason) #endif #ifndef ERTS_SMP - set_proc_exiting_state(p, erts_smp_atomic32_read_nob(&p->state)); + set_proc_self_exiting(p); #else - state = set_proc_exiting_state(p, erts_smp_atomic32_read_nob(&p->state)); - if (state & ERTS_PSFLG_PENDING_EXIT) { + if (ERTS_PSFLG_PENDING_EXIT & set_proc_self_exiting(p)) { /* Process exited before pending exit was received... */ p->pending_exit.reason = THE_NON_VALUE; if (p->pending_exit.bp) { @@ -8849,6 +10360,7 @@ erts_continue_exit_process(Process *p) DistEntry *dep; struct saved_calls *scb; process_breakpoint_time_t *pbt; + erts_aint32_t state; #ifdef DEBUG int yield_allowed = 1; @@ -8885,6 +10397,13 @@ erts_continue_exit_process(Process *p) p->flags &= ~F_USING_DB; } + erts_set_gc_state(p, 1); + state = erts_smp_atomic32_read_acqb(&p->state); + if (state & ERTS_PSFLG_ACTIVE_SYS) { + if (cleanup_sys_tasks(p, state, CONTEXT_REDS) >= CONTEXT_REDS/2) + goto yield; + } + if (p->flags & F_USING_DDLL) { erts_ddll_proc_dead(p, ERTS_PROC_LOCK_MAIN); p->flags &= ~F_USING_DDLL; @@ -8962,17 +10481,31 @@ erts_continue_exit_process(Process *p) { /* Inactivate and notify free */ erts_aint32_t n, e, a = erts_smp_atomic32_read_nob(&p->state); +#ifdef ERTS_SMP + int refc_inced = 0; +#endif while (1) { n = e = a; ASSERT(a & ERTS_PSFLG_EXITING); n |= ERTS_PSFLG_FREE; n &= ~ERTS_PSFLG_ACTIVE; +#ifdef ERTS_SMP + if ((n & ERTS_PSFLG_IN_RUNQ) && !refc_inced) { + erts_smp_proc_inc_refc(p); + refc_inced = 1; + } +#endif a = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e); if (a == e) break; } - } +#ifdef ERTS_SMP + if (refc_inced && !(n & ERTS_PSFLG_IN_RUNQ)) + erts_smp_proc_dec_refc(p); +#endif + } + dep = ((p->flags & F_DISTRIBUTION) ? ERTS_PROC_SET_DIST_ENTRY(p, ERTS_PROC_LOCKS_ALL, NULL) : NULL); @@ -9025,12 +10558,6 @@ erts_continue_exit_process(Process *p) if (pbt) erts_free(ERTS_ALC_T_BPD, (void *) pbt); - if (p->extra_root != NULL) { - (p->extra_root->cleanup)(p->extra_root); /* Should deallocate - whole structure */ - p->extra_root = NULL; - } - delete_process(p); #ifdef ERTS_SMP @@ -9075,7 +10602,7 @@ timeout_proc(Process* p) state = erts_smp_atomic32_read_acqb(&p->state); if (!(state & ERTS_PSFLG_ACTIVE)) - schedule_process(p, state, 0); + schedule_process(p, state); } @@ -9153,7 +10680,9 @@ erts_program_counter_info(int to, void *to_arg, Process *p) print_function_from_pc(to, to_arg, p->cp); erts_print(to, to_arg, ")\n"); state = erts_smp_atomic32_read_acqb(&p->state); - if (!(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_GC))) { + if (!(state & (ERTS_PSFLG_RUNNING + | ERTS_PSFLG_RUNNING_SYS + | ERTS_PSFLG_GC))) { erts_print(to, to_arg, "arity = %d\n",p->arity); if (!ERTS_IS_CRASH_DUMPING) { /* diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index 8d136f6e8b..6155f99b85 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.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 @@ -70,6 +70,9 @@ typedef struct process Process; struct ErtsNodesMonitor_; +#define ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT 0 +#define ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT 0 + #define ERTS_MAX_NO_OF_SCHEDULERS 1024 #define ERTS_DEFAULT_MAX_PROCESSES (1 << 18) @@ -98,6 +101,7 @@ struct saved_calls { extern Export exp_send, exp_receive, exp_timeout; extern int erts_sched_compact_load; +extern int erts_sched_balance_util; extern Uint erts_no_schedulers; extern Uint erts_no_run_queues; extern int erts_sched_thread_suggested_stack_size; @@ -198,6 +202,10 @@ extern int erts_sched_thread_suggested_stack_size; #define ERTS_RUNQ_FLGS_SET(RQ, FLGS) \ ((Uint32) erts_smp_atomic32_read_bor_relb(&(RQ)->flags, \ (erts_aint32_t) (FLGS))) +#define ERTS_RUNQ_FLGS_BSET(RQ, MSK, FLGS) \ + ((Uint32) erts_smp_atomic32_read_bset_relb(&(RQ)->flags, \ + (erts_aint32_t) (MSK), \ + (erts_aint32_t) (FLGS))) #define ERTS_RUNQ_FLGS_UNSET(RQ, FLGS) \ ((Uint32) erts_smp_atomic32_read_band_relb(&(RQ)->flags, \ (erts_aint32_t) ~(FLGS))) @@ -316,9 +324,40 @@ typedef struct { int reds; } ErtsRunQueueInfo; + +#ifdef HAVE_GETHRTIME +# undef ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT +# define ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT 1 +#endif + #ifdef ERTS_SMP +#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 + +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT +typedef struct { + ErtsAtomicSchedTime last; + struct { + Uint64 short_interval; + Uint64 long_interval; + } worktime; + int is_working; +} ErtsRunQueueSchedUtil; +#endif + typedef struct { +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + int sched_util; +#endif Uint32 flags; ErtsRunQueue *misc_evac_runq; struct { @@ -385,6 +424,9 @@ struct ErtsRunQueue_ { Port *start; Port *end; } ports; +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + ErtsRunQueueSchedUtil sched_util; +#endif }; #ifdef ERTS_SMP @@ -414,6 +456,7 @@ do { \ } while (0) typedef struct { + int need; /* "+sbu true" or scheduler_wall_time enabled */ int enabled; Uint64 start; struct { @@ -542,6 +585,12 @@ int erts_smp_lc_runq_is_locked(ErtsRunQueue *); #ifdef ERTS_INCLUDE_SCHEDULER_INTERNALS +#ifdef ERTS_SMP +void erts_empty_runq(ErtsRunQueue *rq); +void erts_non_empty_runq(ErtsRunQueue *rq); +#endif + + /* * Run queue locked during modifications. We use atomic ops since * other threads peek at values without run queue lock. @@ -574,6 +623,10 @@ erts_smp_inc_runq_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi, int prio) erts_smp_atomic32_set_relb(&rqi->len, len); +#ifdef ERTS_SMP + if (rq->len == 0) + erts_non_empty_runq(rq); +#endif rq->len++; if (rq->max_len < rq->len) rq->max_len = len; @@ -631,8 +684,9 @@ erts_smp_reset_max_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi) #define ERTS_PSD_SCHED_ID 2 #define ERTS_PSD_DIST_ENTRY 3 #define ERTS_PSD_CALL_TIME_BP 4 +#define ERTS_PSD_DELAYED_GC_TASK_QS 5 -#define ERTS_PSD_SIZE 5 +#define ERTS_PSD_SIZE 6 typedef struct { void *data[ERTS_PSD_SIZE]; @@ -656,6 +710,9 @@ typedef struct { #define ERTS_PSD_CALL_TIME_BP_GET_LOCKS ERTS_PROC_LOCK_MAIN #define ERTS_PSD_CALL_TIME_BP_SET_LOCKS ERTS_PROC_LOCK_MAIN +#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 + typedef struct { ErtsProcLocks get_locks; ErtsProcLocks set_locks; @@ -688,6 +745,9 @@ typedef struct { ErlHeapFragment *bp; } ErtsPendExit; +typedef struct ErtsProcSysTask_ ErtsProcSysTask; +typedef struct ErtsProcSysTaskQs_ ErtsProcSysTaskQs; + #ifdef ERTS_SMP typedef struct ErtsPendingSuspend_ ErtsPendingSuspend; @@ -704,13 +764,6 @@ struct ErtsPendingSuspend_ { #endif -typedef struct ErlExtraRootSet_ ErlExtraRootSet; -struct ErlExtraRootSet_ { - Eterm *objv; - Uint sz; - void (*cleanup)(ErlExtraRootSet *); -}; - /* Defines to ease the change of memory architecture */ # define HEAP_START(p) (p)->heap # define HEAP_TOP(p) (p)->htop @@ -804,8 +857,6 @@ struct process { ErlMessageQueue msg; /* Message queue */ - ErlExtraRootSet *extra_root; /* Used by trapping BIF's */ - union { ErtsBifTimer *bif_timers; /* Bif timers aiming at this process */ void *terminate; @@ -855,6 +906,8 @@ struct process { Uint64 bin_old_vheap_sz; /* Virtual old heap block size for binaries */ Uint64 bin_old_vheap; /* Virtual old heap size for binaries */ + ErtsProcSysTaskQs *sys_task_qs; + erts_smp_atomic32_t state; /* Process state flags (see ERTS_PSFLG_*) */ #ifdef ERTS_SMP @@ -924,24 +977,67 @@ void erts_check_for_holes(Process* p); # error "Need to increase ERTS_PSFLG_PRIO_SHIFT" #endif -#define ERTS_PSFLG_PRIO_SHIFT 2 +#define ERTS_PSFLGS_PRIO_BITS 2 +#define ERTS_PSFLGS_PRIO_MASK \ + ((((erts_aint32_t) 1) << ERTS_PSFLGS_PRIO_BITS) - 1) -#define ERTS_PSFLG_BIT(N) \ - (((erts_aint32_t) 1) << (ERTS_PSFLG_PRIO_SHIFT + (N))) +#define ERTS_PSFLGS_ACT_PRIO_OFFSET (0*ERTS_PSFLGS_PRIO_BITS) +#define ERTS_PSFLGS_USR_PRIO_OFFSET (1*ERTS_PSFLGS_PRIO_BITS) +#define ERTS_PSFLGS_PRQ_PRIO_OFFSET (2*ERTS_PSFLGS_PRIO_BITS) +#define ERTS_PSFLGS_ZERO_BIT_OFFSET (3*ERTS_PSFLGS_PRIO_BITS) -#define ERTS_PSFLG_PRIO_MASK (ERTS_PSFLG_BIT(0) - 1) +#define ERTS_PSFLGS_QMASK_BITS 4 +#define ERTS_PSFLGS_QMASK \ + ((((erts_aint32_t) 1) << ERTS_PSFLGS_QMASK_BITS) - 1) +#define ERTS_PSFLGS_IN_PRQ_MASK_OFFSET \ + ERTS_PSFLGS_ZERO_BIT_OFFSET -#define ERTS_PSFLG_FREE ERTS_PSFLG_BIT(0) -#define ERTS_PSFLG_EXITING ERTS_PSFLG_BIT(1) -#define ERTS_PSFLG_PENDING_EXIT ERTS_PSFLG_BIT(2) -#define ERTS_PSFLG_ACTIVE ERTS_PSFLG_BIT(3) -#define ERTS_PSFLG_IN_RUNQ ERTS_PSFLG_BIT(4) -#define ERTS_PSFLG_RUNNING ERTS_PSFLG_BIT(5) -#define ERTS_PSFLG_SUSPENDED ERTS_PSFLG_BIT(6) -#define ERTS_PSFLG_GC ERTS_PSFLG_BIT(7) -#define ERTS_PSFLG_BOUND ERTS_PSFLG_BIT(8) -#define ERTS_PSFLG_TRAP_EXIT ERTS_PSFLG_BIT(9) +#define ERTS_PSFLG_BIT(N) \ + (((erts_aint32_t) 1) << (ERTS_PSFLGS_ZERO_BIT_OFFSET + (N))) +/* + * ACT_PRIO -> Active prio, i.e., currently active prio. This + * prio may be higher than user prio. + * USR_PRIO -> User prio. i.e., prio the user has set. + * PRQ_PRIO -> Prio queue prio, i.e., prio queue currently + * enqueued in. + */ +#define ERTS_PSFLGS_ACT_PRIO_MASK \ + (ERTS_PSFLGS_PRIO_MASK << ERTS_PSFLGS_ACT_PRIO_OFFSET) +#define ERTS_PSFLGS_USR_PRIO_MASK \ + (ERTS_PSFLGS_PRIO_MASK << ERTS_PSFLGS_USR_PRIO_OFFSET) +#define ERTS_PSFLGS_PRQ_PRIO_MASK \ + (ERTS_PSFLGS_PRIO_MASK << ERTS_PSFLGS_PRQ_PRIO_OFFSET) +#define ERTS_PSFLG_IN_PRQ_MAX ERTS_PSFLG_BIT(0) +#define ERTS_PSFLG_IN_PRQ_HIGH ERTS_PSFLG_BIT(1) +#define ERTS_PSFLG_IN_PRQ_NORMAL ERTS_PSFLG_BIT(2) +#define ERTS_PSFLG_IN_PRQ_LOW ERTS_PSFLG_BIT(3) +#define ERTS_PSFLG_FREE ERTS_PSFLG_BIT(4) +#define ERTS_PSFLG_EXITING ERTS_PSFLG_BIT(5) +#define ERTS_PSFLG_PENDING_EXIT ERTS_PSFLG_BIT(6) +#define ERTS_PSFLG_ACTIVE ERTS_PSFLG_BIT(7) +#define ERTS_PSFLG_IN_RUNQ ERTS_PSFLG_BIT(8) +#define ERTS_PSFLG_RUNNING ERTS_PSFLG_BIT(9) +#define ERTS_PSFLG_SUSPENDED ERTS_PSFLG_BIT(10) +#define ERTS_PSFLG_GC ERTS_PSFLG_BIT(11) +#define ERTS_PSFLG_BOUND ERTS_PSFLG_BIT(12) +#define ERTS_PSFLG_TRAP_EXIT ERTS_PSFLG_BIT(13) +#define ERTS_PSFLG_ACTIVE_SYS ERTS_PSFLG_BIT(14) +#define ERTS_PSFLG_RUNNING_SYS ERTS_PSFLG_BIT(15) +#define ERTS_PSFLG_PROXY ERTS_PSFLG_BIT(16) +#define ERTS_PSFLG_DELAYED_SYS ERTS_PSFLG_BIT(17) + +#define ERTS_PSFLGS_IN_PRQ_MASK (ERTS_PSFLG_IN_PRQ_MAX \ + | ERTS_PSFLG_IN_PRQ_HIGH \ + | ERTS_PSFLG_IN_PRQ_NORMAL \ + | ERTS_PSFLG_IN_PRQ_LOW) + +#define ERTS_PSFLGS_GET_ACT_PRIO(PSFLGS) \ + (((PSFLGS) >> ERTS_PSFLGS_ACT_PRIO_OFFSET) & ERTS_PSFLGS_PRIO_MASK) +#define ERTS_PSFLGS_GET_USR_PRIO(PSFLGS) \ + (((PSFLGS) >> ERTS_PSFLGS_USR_PRIO_OFFSET) & ERTS_PSFLGS_PRIO_MASK) +#define ERTS_PSFLGS_GET_PRQ_PRIO(PSFLGS) \ + (((PSFLGS) >> ERTS_PSFLGS_USR_PRIO_OFFSET) & ERTS_PSFLGS_PRIO_MASK) /* The sequential tracing token is a tuple of size 5: * @@ -1056,6 +1152,7 @@ extern struct erts_system_profile_flags_t erts_system_profile_flags; #define F_HAVE_BLCKD_MSCHED (1 << 8) /* Process has blocked multi-scheduling */ #define F_P2PNR_RESCHED (1 << 9) /* Process has been rescheduled via erts_pid2proc_not_running() */ #define F_FORCE_GC (1 << 10) /* Force gc at process in-scheduling */ +#define F_DISABLE_GC (1 << 11) /* Disable GC */ /* process trace_flags */ #define F_SENSITIVE (1 << 0) @@ -1146,6 +1243,7 @@ void erts_late_init_process(void); void erts_early_init_scheduling(int); void erts_init_scheduling(int, int); +int erts_set_gc_state(Process *c_p, int enable); Eterm erts_sched_wall_time_request(Process *c_p, int set, int enable); Eterm erts_gc_info_request(Process *c_p); Uint64 erts_get_proc_interval(void); @@ -1591,6 +1689,11 @@ erts_psd_set(Process *p, ErtsProcLocks plocks, int ix, void *data) #define ERTS_PROC_SET_CALL_TIME(P, L, PBT) \ ((process_breakpoint_time_t *) erts_psd_set((P), (L), ERTS_PSD_CALL_TIME_BP, (void *) (PBT))) +#define ERTS_PROC_GET_DELAYED_GC_TASK_QS(P) \ + ((ErtsProcSysTaskQs *) erts_psd_get((P), ERTS_PSD_DELAYED_GC_TASK_QS)) +#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))) + ERTS_GLB_INLINE Eterm erts_proc_get_error_handler(Process *p); ERTS_GLB_INLINE Eterm erts_proc_set_error_handler(Process *p, @@ -1636,6 +1739,13 @@ erts_proc_set_error_handler(Process *p, ErtsProcLocks plocks, Eterm handler) extern erts_atomic_t erts_migration_paths; +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT +int erts_get_sched_util(ErtsRunQueue *rq, + int initially_locked, + int short_interval); +#endif + + ERTS_GLB_INLINE ErtsMigrationPaths *erts_get_migration_paths_managed(void); ERTS_GLB_INLINE ErtsMigrationPaths *erts_get_migration_paths(void); ERTS_GLB_INLINE ErtsRunQueue *erts_check_emigration_need(ErtsRunQueue *c_rq, @@ -1687,22 +1797,36 @@ erts_check_emigration_need(ErtsRunQueue *c_rq, int prio) return mp->prio[prio].runq; } - - if (prio == ERTS_PORT_PRIO_LEVEL) - len = RUNQ_READ_LEN(&c_rq->ports.info.len); +#if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT + if (mp->sched_util) { + ErtsRunQueue *rq = mp->prio[prio].runq; + /* No migration if other is non-empty */ + if (!(ERTS_RUNQ_FLGS_GET(rq) & ERTS_RUNQ_FLG_NONEMPTY) + && erts_get_sched_util(rq, 0, 1) < mp->prio[prio].limit.other + && erts_get_sched_util(c_rq, 0, 1) > mp->prio[prio].limit.this) { + return rq; + } + } else - len = RUNQ_READ_LEN(&c_rq->procs.prio_info[prio].len); - - if (len > mp->prio[prio].limit.this) { - ErtsRunQueue *n_rq = mp->prio[prio].runq; - if (n_rq) { - if (prio == ERTS_PORT_PRIO_LEVEL) - len = RUNQ_READ_LEN(&n_rq->ports.info.len); - else - len = RUNQ_READ_LEN(&n_rq->procs.prio_info[prio].len); - - if (len < mp->prio[prio].limit.other) - return n_rq; +#endif + { + + if (prio == ERTS_PORT_PRIO_LEVEL) + len = RUNQ_READ_LEN(&c_rq->ports.info.len); + else + len = RUNQ_READ_LEN(&c_rq->procs.prio_info[prio].len); + + if (len > mp->prio[prio].limit.this) { + ErtsRunQueue *n_rq = mp->prio[prio].runq; + if (n_rq) { + if (prio == ERTS_PORT_PRIO_LEVEL) + len = RUNQ_READ_LEN(&n_rq->ports.info.len); + else + len = RUNQ_READ_LEN(&n_rq->procs.prio_info[prio].len); + + if (len < mp->prio[prio].limit.other) + return n_rq; + } } } } diff --git a/erts/emulator/beam/erl_unicode.c b/erts/emulator/beam/erl_unicode.c index 569c0a7d31..3a968594f3 100644 --- a/erts/emulator/beam/erl_unicode.c +++ b/erts/emulator/beam/erl_unicode.c @@ -1984,9 +1984,21 @@ BIF_RETTYPE binary_to_existing_atom_2(BIF_ALIST_2) * string routines, that will certainly fail on some OS. */ -char *erts_convert_filename_to_native(Eterm name, char *statbuf, size_t statbuf_size, ErtsAlcType_t alloc_type, int allow_empty, int allow_atom, Sint *used) +char *erts_convert_filename_to_native(Eterm name, char *statbuf, size_t statbuf_size, + ErtsAlcType_t alloc_type, int allow_empty, + int allow_atom, Sint *used) { int encoding = erts_get_native_filename_encoding(); + return erts_convert_filename_to_encoding(name, statbuf, statbuf_size, alloc_type, + allow_empty, allow_atom, encoding, + used, 0); +} + +char *erts_convert_filename_to_encoding(Eterm name, char *statbuf, size_t statbuf_size, + ErtsAlcType_t alloc_type, int allow_empty, + int allow_atom, int encoding, Sint *used, + Uint extra) +{ char* name_buf = NULL; if ((allow_atom && is_atom(name)) || @@ -1998,13 +2010,14 @@ char *erts_convert_filename_to_native(Eterm name, char *statbuf, size_t statbuf_ } if (encoding == ERL_FILENAME_WIN_WCHAR) { need += 2; + extra *= 2; } else { ++need; } if (used) *used = (Sint) need; - if (need > statbuf_size) { - name_buf = (char *) erts_alloc(alloc_type, need); + if (need+extra > statbuf_size) { + name_buf = (char *) erts_alloc(alloc_type, need+extra); } else { name_buf = statbuf; } @@ -2016,52 +2029,27 @@ char *erts_convert_filename_to_native(Eterm name, char *statbuf, size_t statbuf_ } else if (is_binary(name)) { byte *temp_alloc = NULL; byte *bytes; - byte *err_pos; - Uint size,num_chars; + Uint size; size = binary_size(name); bytes = erts_get_aligned_binary_bytes(name, &temp_alloc); + if (encoding != ERL_FILENAME_WIN_WCHAR) { /*Add 0 termination only*/ if (used) *used = (Sint) size+1; - if (size+1 > statbuf_size) { - name_buf = (char *) erts_alloc(alloc_type, size+1); + if (size+1+extra > statbuf_size) { + name_buf = (char *) erts_alloc(alloc_type, size+1+extra); } else { name_buf = statbuf; } memcpy(name_buf,bytes,size); name_buf[size]=0; - } else if (erts_analyze_utf8(bytes,size,&err_pos,&num_chars,NULL) != ERTS_UTF8_OK || - erts_get_user_requested_filename_encoding() == ERL_FILENAME_LATIN1) { - byte *p; - /* What to do now? Maybe latin1, so just take byte for byte instead */ - if (used) - *used = (Sint) (size+1)*2; - if ((size+1)*2 > statbuf_size) { - name_buf = (char *) erts_alloc(alloc_type, (size+1)*2); - } else { - name_buf = statbuf; - } - p = (byte *) name_buf; - while (size--) { - *p++ = *bytes++; - *p++ = 0; - } - *p++ = 0; - *p++ = 0; - } else { /* WIN_WCHAR and valid UTF8 */ - if (used) - *used = (Sint) (num_chars+1)*2; - if ((num_chars+1)*2 > statbuf_size) { - name_buf = (char *) erts_alloc(alloc_type, (num_chars+1)*2); - } else { - name_buf = statbuf; - } - erts_copy_utf8_to_utf16_little((byte *) name_buf, bytes, num_chars); - name_buf[num_chars*2] = 0; - name_buf[num_chars*2+1] = 0; - } + } else { + name_buf = erts_convert_filename_to_wchar(bytes, size, + statbuf, statbuf_size, + alloc_type, used, extra); + } erts_free_aligned_binary_bytes(temp_alloc); } else { return NULL; @@ -2069,6 +2057,50 @@ char *erts_convert_filename_to_native(Eterm name, char *statbuf, size_t statbuf_ return name_buf; } +char* erts_convert_filename_to_wchar(byte* bytes, Uint size, + char *statbuf, size_t statbuf_size, + ErtsAlcType_t alloc_type, Sint* used, + Uint extra_wchars) +{ + byte *err_pos; + Uint num_chars; + char* name_buf = NULL; + Sint need; + char *p; + + if (erts_analyze_utf8(bytes,size,&err_pos,&num_chars,NULL) != ERTS_UTF8_OK || + erts_get_user_requested_filename_encoding() == ERL_FILENAME_LATIN1) { + + /* What to do now? Maybe latin1, so just take byte for byte instead */ + need = (Sint) (size + extra_wchars + 1) * 2; + if (need > statbuf_size) { + name_buf = (char *) erts_alloc(alloc_type, need); + } else { + name_buf = statbuf; + } + p = name_buf; + while (size--) { + *p++ = *bytes++; + *p++ = 0; + } + } else { /* WIN_WCHAR and valid UTF8 */ + need = (Sint) (num_chars + extra_wchars + 1) * 2; + if (need > statbuf_size) { + name_buf = (char *) erts_alloc(alloc_type, need); + } else { + name_buf = statbuf; + } + erts_copy_utf8_to_utf16_little((byte *) name_buf, bytes, num_chars); + p = name_buf + num_chars*2; + } + *p++ = 0; + *p++ = 0; + if (used) + *used = p - name_buf; + return name_buf; +} + + static int filename_len_16bit(byte *str) { byte *p = str; @@ -2148,16 +2180,31 @@ Sint erts_native_filename_need(Eterm ioterm, int encoding) ap = atom_tab(atom_val(ioterm)); switch (encoding) { case ERL_FILENAME_LATIN1: - need = ap->len; + need = ap->latin1_chars; /* May be -1 */ break; case ERL_FILENAME_UTF8_MAC: case ERL_FILENAME_UTF8: - for (i = 0; i < ap->len; i++) { - need += (ap->name[i] >= 0x80) ? 2 : 1; - } + need = ap->len; break; case ERL_FILENAME_WIN_WCHAR: - need = 2*(ap->len); + if (ap->latin1_chars >= 0) { + need = 2* ap->latin1_chars; + } + else { + for (i = 0; i < ap->len; ) { + if (ap->name[i] < 0x80) { + i++; + } else if (ap->name[i] < 0xE0) { + i += 2; + } else if (ap->name[i] < 0xF0) { + i += 3; + } else { + need = -1; + break; + } + need += 2; + } + } break; default: need = -1; @@ -2287,26 +2334,36 @@ void erts_native_filename_put(Eterm ioterm, int encoding, byte *p) switch (encoding) { case ERL_FILENAME_LATIN1: for (i = 0; i < ap->len; i++) { - *p++ = ap->name[i]; - } - break; - case ERL_FILENAME_UTF8_MAC: - case ERL_FILENAME_UTF8: - for (i = 0; i < ap->len; i++) { - if(ap->name[i] < 0x80) { + if (ap->name[i] < 0x80) { *p++ = ap->name[i]; } else { - *p++ = (((ap->name[i]) >> 6) | ((byte) 0xC0)); - *p++ = (((ap->name[i]) & 0x3F) | ((byte) 0x80)); + ASSERT(ap->name[i] < 0xC4); + *p++ = ((ap->name[i] & 3) << 6) | (ap->name[i+1] & 0x3F); + i++; } } break; + case ERL_FILENAME_UTF8_MAC: + case ERL_FILENAME_UTF8: + sys_memcpy(p, ap->name, ap->len); + break; case ERL_FILENAME_WIN_WCHAR: for (i = 0; i < ap->len; i++) { /* Little endian */ - *p++ = ap->name[i]; - *p++ = 0; - } + if (ap->name[i] < 0x80) { + *p++ = ap->name[i]; + *p++ = 0; + } else if (ap->name[i] < 0xE0) { + *p++ = ((ap->name[i] & 3) << 6) | (ap->name[i+1] & 0x3F); + *p++ = ((ap->name[i] & 0x1C) >> 2); + i++; + } else { + ASSERT(ap->name[i] < 0xF0); + *p++ = ((ap->name[i+1] & 3) << 6) | (ap->name[i+2] & 0x3C); + *p++ = ((ap->name[i] & 0xF) << 4) | ((ap->name[i+1] & 0x3C) >> 2); + i += 2; + } + } break; default: ASSERT(0); diff --git a/erts/emulator/beam/erl_zlib.c b/erts/emulator/beam/erl_zlib.c index 47fd92988e..8e33144f96 100644 --- a/erts/emulator/beam/erl_zlib.c +++ b/erts/emulator/beam/erl_zlib.c @@ -87,6 +87,46 @@ int ZEXPORT erl_zlib_deflate_finish(z_stream *streamp) return deflateEnd(streamp); } +int ZEXPORT erl_zlib_inflate_start(z_stream *streamp, const Bytef* source, + uLong sourceLen) +{ + streamp->next_in = (Bytef*)source; + streamp->avail_in = (uInt)sourceLen; + streamp->total_out = streamp->avail_out = 0; + streamp->next_out = NULL; + erl_zlib_alloc_init(streamp); + return inflateInit(streamp); +} +/* + * Inflate a chunk, The destination length is the limit. + * Returns Z_OK if more to process, Z_STREAM_END if we are done. + */ +int ZEXPORT erl_zlib_inflate_chunk(z_stream *streamp, Bytef* dest, uLongf* destLen) +{ + int err; + uLongf last_tot = streamp->total_out; + + streamp->next_out = dest; + streamp->avail_out = (uInt)*destLen; + + if ((uLong)streamp->avail_out != *destLen) return Z_BUF_ERROR; + + err = inflate(streamp, Z_NO_FLUSH); + ASSERT(err != Z_STREAM_ERROR); + *destLen = streamp->total_out - last_tot; + return err; +} + +/* + * When we are done, free up the inflate structure + * Retyurns Z_OK or Error + */ +int ZEXPORT erl_zlib_inflate_finish(z_stream *streamp) +{ + return inflateEnd(streamp); +} + + int ZEXPORT erl_zlib_compress2 (Bytef* dest, uLongf* destLen, const Bytef* source, uLong sourceLen, int level) diff --git a/erts/emulator/beam/erl_zlib.h b/erts/emulator/beam/erl_zlib.h index 5ac849d21c..160166c66b 100644 --- a/erts/emulator/beam/erl_zlib.h +++ b/erts/emulator/beam/erl_zlib.h @@ -39,6 +39,12 @@ int ZEXPORT erl_zlib_deflate_start(z_stream *streamp, const Bytef* source, int ZEXPORT erl_zlib_deflate_chunk(z_stream *streamp, Bytef* dest, uLongf* destLen); int ZEXPORT erl_zlib_deflate_finish(z_stream *streamp); +int ZEXPORT erl_zlib_inflate_start(z_stream *streamp, const Bytef* source, + uLong sourceLen); +int ZEXPORT erl_zlib_inflate_chunk(z_stream *streamp, Bytef* dest, uLongf* destLen); +int ZEXPORT erl_zlib_inflate_finish(z_stream *streamp); + + /* Use instead of compress */ #define erl_zlib_compress(dest,destLen,source,sourceLen) \ diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c index 22b0a02937..5e7a5cab6e 100644 --- a/erts/emulator/beam/external.c +++ b/erts/emulator/beam/external.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 @@ -61,6 +61,9 @@ */ # define ERTS_DEBUG_USE_DIST_SEP # endif +# define IF_DEBUG(X) X +#else +# define IF_DEBUG(X) #endif /* Does Sint fit in Sint32? @@ -84,29 +87,40 @@ static Export term_to_binary_trap_export; static byte* enc_term(ErtsAtomCacheMap *, Eterm, byte*, Uint32, struct erl_off_heap_header** off_heap); -static int enc_term_int(Process *p,ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dflags, +struct TTBEncodeContext_; +static int enc_term_int(struct TTBEncodeContext_*,ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dflags, struct erl_off_heap_header** off_heap, Sint *reds, byte **res); static Uint is_external_string(Eterm obj, int* p_is_string); static byte* enc_atom(ErtsAtomCacheMap *, Eterm, byte*, Uint32); static byte* enc_pid(ErtsAtomCacheMap *, Eterm, byte*, Uint32); -static byte* dec_term(ErtsDistExternal *, Eterm**, byte*, ErlOffHeap*, Eterm*); +struct B2TContext_t; +static byte* dec_term(ErtsDistExternal *, Eterm**, byte*, ErlOffHeap*, Eterm*, struct B2TContext_t*); static byte* dec_atom(ErtsDistExternal *, byte*, Eterm*); static byte* dec_pid(ErtsDistExternal *, Eterm**, byte*, ErlOffHeap*, Eterm*); -static Sint decoded_size(byte *ep, byte* endp, int internal_tags); +static Sint decoded_size(byte *ep, byte* endp, int internal_tags, struct B2TContext_t*); static BIF_RETTYPE term_to_binary_trap_1(BIF_ALIST_1); static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint flags, Binary *context_b); static Uint encode_size_struct2(ErtsAtomCacheMap *, Eterm, unsigned); -static int encode_size_struct_int(Process *p, ErtsAtomCacheMap *acmp, Eterm obj, +struct TTBSizeContext_; +static int encode_size_struct_int(struct TTBSizeContext_*, ErtsAtomCacheMap *acmp, Eterm obj, unsigned dflags, Sint *reds, Uint *res); +static Export binary_to_term_trap_export; +static BIF_RETTYPE binary_to_term_trap_1(BIF_ALIST_1); +static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* context_b); + void erts_init_external(void) { #if 1 /* In R16 */ erts_init_trap_export(&term_to_binary_trap_export, am_erlang, am_term_to_binary_trap, 1, &term_to_binary_trap_1); + + erts_init_trap_export(&binary_to_term_trap_export, + am_erlang, am_binary_to_term_trap, 1, + &binary_to_term_trap_1); #else sys_memset((void *) &term_to_binary_trap_export, 0, sizeof(Export)); term_to_binary_trap_export.address = &term_to_binary_trap_export.code[3]; @@ -877,7 +891,7 @@ erts_decode_dist_ext_size(ErtsDistExternal *edep) goto fail; ep = edep->extp+1; } - res = decoded_size(ep, edep->ext_endp, 0); + res = decoded_size(ep, edep->ext_endp, 0, NULL); if (res >= 0) return res; fail: @@ -889,12 +903,12 @@ Sint erts_decode_ext_size(byte *ext, Uint size) { if (size == 0 || *ext != VERSION_MAGIC) return -1; - return decoded_size(ext+1, ext+size, 0); + return decoded_size(ext+1, ext+size, 0, NULL); } Sint erts_decode_ext_size_ets(byte *ext, Uint size) { - Sint sz = decoded_size(ext, ext+size, 1); + Sint sz = decoded_size(ext, ext+size, 1, NULL); ASSERT(sz >= 0); return sz; } @@ -927,7 +941,7 @@ erts_decode_dist_ext(Eterm** hpp, goto error; ep++; } - ep = dec_term(edep, hpp, ep, off_heap, &obj); + ep = dec_term(edep, hpp, ep, off_heap, &obj, NULL); if (!ep) goto error; @@ -948,7 +962,7 @@ Eterm erts_decode_ext(Eterm **hpp, ErlOffHeap *off_heap, byte **ext) byte *ep = *ext; if (*ep++ != VERSION_MAGIC) return THE_NON_VALUE; - ep = dec_term(NULL, hpp, ep, off_heap, &obj); + ep = dec_term(NULL, hpp, ep, off_heap, &obj, NULL); if (!ep) { #ifdef DEBUG bin_write(ERTS_PRINT_STDERR,NULL,*ext,500); @@ -962,7 +976,7 @@ Eterm erts_decode_ext(Eterm **hpp, ErlOffHeap *off_heap, byte **ext) Eterm erts_decode_ext_ets(Eterm **hpp, ErlOffHeap *off_heap, byte *ext) { Eterm obj; - ext = dec_term(NULL, hpp, ext, off_heap, &obj); + ext = dec_term(NULL, hpp, ext, off_heap, &obj, NULL); ASSERT(ext); return obj; } @@ -1043,9 +1057,14 @@ static BIF_RETTYPE term_to_binary_trap_1(BIF_ALIST_1) Binary *bin = ((ProcBin *) binary_val(bt))->val; Eterm res = erts_term_to_binary_int(BIF_P, Term, 0, 0,bin); if (is_tuple(res)) { + ASSERT(BIF_P->flags & F_DISABLE_GC); BIF_TRAP1(&term_to_binary_trap_export,BIF_P,res); } else { - BIF_RET(res); + if (erts_set_gc_state(BIF_P, 1) + || MSO(BIF_P).overhead > BIN_VHEAP_SZ(BIF_P)) + ERTS_BIF_YIELD_RETURN(BIF_P, res); + else + BIF_RET(res); } } @@ -1053,8 +1072,10 @@ BIF_RETTYPE term_to_binary_1(BIF_ALIST_1) { Eterm res = erts_term_to_binary_int(BIF_P, BIF_ARG_1, 0, TERM_TO_BINARY_DFLAGS, NULL); if (is_tuple(res)) { + erts_set_gc_state(BIF_P, 0); BIF_TRAP1(&term_to_binary_trap_export,BIF_P,res); } else { + ASSERT(!(BIF_P->flags & F_DISABLE_GC)); BIF_RET(res); } } @@ -1067,7 +1088,6 @@ BIF_RETTYPE term_to_binary_2(BIF_ALIST_2) int level = 0; Uint flags = TERM_TO_BINARY_DFLAGS; Eterm res; - Binary *bin = NULL; while (is_list(Flags)) { Eterm arg = CAR(list_val(Flags)); @@ -1104,14 +1124,74 @@ BIF_RETTYPE term_to_binary_2(BIF_ALIST_2) goto error; } - res = erts_term_to_binary_int(p, Term, level, flags, bin); + res = erts_term_to_binary_int(p, Term, level, flags, NULL); if (is_tuple(res)) { + erts_set_gc_state(p, 0); BIF_TRAP1(&term_to_binary_trap_export,BIF_P,res); } else { + ASSERT(!(BIF_P->flags & F_DISABLE_GC)); BIF_RET(res); } } + +enum B2TState { /* order is somewhat significant */ + B2TPrepare, + B2TUncompressChunk, + B2TSizeInit, + B2TSize, + B2TDecodeInit, + B2TDecode, + B2TDecodeList, + B2TDecodeTuple, + B2TDecodeString, + B2TDecodeBinary, + + B2TDone, + B2TDecodeFail, + B2TBadArg +}; + +typedef struct { + int heap_size; + int terms; + byte* ep; + int atom_extra_skip; +} B2TSizeContext; + +typedef struct { + byte* ep; + Eterm res; + Eterm* next; + Eterm* hp_start; + Eterm* hp; + Eterm* hp_end; + int remaining_n; + char* remaining_bytes; +} B2TDecodeContext; + +typedef struct { + z_stream stream; + byte* dbytes; + Uint dleft; +} B2TUncompressContext; + +typedef struct B2TContext_t { + Sint heap_size; + byte* aligned_alloc; + ErtsBinary2TermState b2ts; + Uint32 flags; + SWord reds; + Eterm trap_bin; + enum B2TState state; + union { + B2TSizeContext sc; + B2TDecodeContext dc; + B2TUncompressContext uc; + } u; +} B2TContext; + + static uLongf binary2term_uncomp_size(byte* data, Sint size) { z_stream stream; @@ -1141,48 +1221,62 @@ static uLongf binary2term_uncomp_size(byte* data, Sint size) return err == Z_STREAM_END ? uncomp_size : 0; } -static ERTS_INLINE Sint -binary2term_prepare(ErtsBinary2TermState *state, byte *data, Sint data_size) +static ERTS_INLINE int +binary2term_prepare(ErtsBinary2TermState *state, byte *data, Sint data_size, + B2TContext* ctx) { - Sint res; byte *bytes = data; Sint size = data_size; state->exttmp = 0; if (size < 1 || *bytes != VERSION_MAGIC) { - error: - if (state->exttmp) - erts_free(ERTS_ALC_T_TMP, state->extp); - state->extp = NULL; - state->exttmp = 0; return -1; } bytes++; size--; if (size < 5 || *bytes != COMPRESSED) { state->extp = bytes; + if (ctx) + ctx->state = B2TSizeInit; } else { uLongf dest_len = (Uint32) get_int32(bytes+1); bytes += 5; size -= 5; if (dest_len > 32*1024*1024 - || (state->extp = erts_alloc_fnf(ERTS_ALC_T_TMP, dest_len)) == NULL) { + || (state->extp = erts_alloc_fnf(ERTS_ALC_T_EXT_TERM_DATA, dest_len)) == NULL) { + /* + * Try avoid out-of-memory crash due to corrupted 'dest_len' + * by checking the actual length of the uncompressed data. + * The only way to do that is to uncompress it. Sad but true. + */ if (dest_len != binary2term_uncomp_size(bytes, size)) { - goto error; + return -1; } - state->extp = erts_alloc(ERTS_ALC_T_TMP, dest_len); + state->extp = erts_alloc(ERTS_ALC_T_EXT_TERM_DATA, dest_len); + ctx->reds -= dest_len; } state->exttmp = 1; - if (erl_zlib_uncompress(state->extp, &dest_len, bytes, size) != Z_OK) - goto error; + if (ctx) { + if (erl_zlib_inflate_start(&ctx->u.uc.stream, bytes, size) != Z_OK) + return -1; + + ctx->u.uc.dbytes = state->extp; + ctx->u.uc.dleft = dest_len; + ctx->state = B2TUncompressChunk; + } + else { + uLongf dlen = dest_len; + if (erl_zlib_uncompress(state->extp, &dlen, bytes, size) != Z_OK + || dlen != dest_len) { + return -1; + } + } size = (Sint) dest_len; } - res = decoded_size(state->extp, state->extp + size, 0); - if (res < 0) - goto error; - return res; + state->extsize = size; + return 0; } static ERTS_INLINE void @@ -1190,7 +1284,7 @@ binary2term_abort(ErtsBinary2TermState *state) { if (state->exttmp) { state->exttmp = 0; - erts_free(ERTS_ALC_T_TMP, state->extp); + erts_free(ERTS_ALC_T_EXT_TERM_DATA, state->extp); } } @@ -1198,11 +1292,11 @@ static ERTS_INLINE Eterm binary2term_create(ErtsDistExternal *edep, ErtsBinary2TermState *state, Eterm **hpp, ErlOffHeap *ohp) { Eterm res; - if (!dec_term(edep, hpp, state->extp, ohp, &res)) + if (!dec_term(edep, hpp, state->extp, ohp, &res, NULL)) res = THE_NON_VALUE; if (state->exttmp) { state->exttmp = 0; - erts_free(ERTS_ALC_T_TMP, state->extp); + erts_free(ERTS_ALC_T_EXT_TERM_DATA, state->extp); } return res; } @@ -1210,7 +1304,18 @@ binary2term_create(ErtsDistExternal *edep, ErtsBinary2TermState *state, Eterm ** Sint erts_binary2term_prepare(ErtsBinary2TermState *state, byte *data, Sint data_size) { - return binary2term_prepare(state, data, data_size); + Sint res; + + if (binary2term_prepare(state, data, data_size, NULL) < 0 || + (res=decoded_size(state->extp, state->extp + state->extsize, 0, NULL)) < 0) { + + if (state->exttmp) + erts_free(ERTS_ALC_T_EXT_TERM_DATA, state->extp); + state->extp = NULL; + state->exttmp = 0; + return -1; + } + return res; } void @@ -1225,68 +1330,233 @@ erts_binary2term_create(ErtsBinary2TermState *state, Eterm **hpp, ErlOffHeap *oh return binary2term_create(NULL,state, hpp, ohp); } -BIF_RETTYPE binary_to_term_1(BIF_ALIST_1) +static void b2t_destroy_context(B2TContext* context) { - Sint heap_size; - Eterm res; + erts_free_aligned_binary_bytes_extra(context->aligned_alloc, + ERTS_ALC_T_EXT_TERM_DATA); + context->aligned_alloc = NULL; + binary2term_abort(&context->b2ts); + if (context->state == B2TUncompressChunk) { + erl_zlib_inflate_finish(&context->u.uc.stream); + } +} + +static void b2t_context_destructor(Binary *context_bin) +{ + B2TContext* ctx = (B2TContext*) ERTS_MAGIC_BIN_DATA(context_bin); + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(context_bin) == b2t_context_destructor); + + b2t_destroy_context(ctx); +} + +static BIF_RETTYPE binary_to_term_trap_1(BIF_ALIST_1) +{ + Binary *context_bin = ((ProcBin *) binary_val(BIF_ARG_1))->val; + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(context_bin) == b2t_context_destructor); + + return binary_to_term_int(BIF_P, 0, THE_NON_VALUE, context_bin); +} + + +#define B2T_BYTES_PER_REDUCTION 128 +#define B2T_MEMCPY_FACTOR 8 + +/* Define for testing */ +/*#define EXTREME_B2T_TRAPPING 1*/ + +#ifdef EXTREME_B2T_TRAPPING +static unsigned b2t_rand(void) +{ + static unsigned prev = 17; + prev = (prev * 214013 + 2531011); + return prev; +} +#endif + + +static B2TContext* b2t_export_context(Process* p, B2TContext* src) +{ + Binary* context_b = erts_create_magic_binary(sizeof(B2TContext), + b2t_context_destructor); + B2TContext* ctx = ERTS_MAGIC_BIN_DATA(context_b); Eterm* hp; - Eterm* endp; - Sint size; - byte* bytes; - byte* temp_alloc = NULL; - ErtsBinary2TermState b2ts; + sys_memcpy(ctx, src, sizeof(B2TContext)); + if (ctx->state >= B2TDecode && ctx->u.dc.next == &src->u.dc.res) { + ctx->u.dc.next = &ctx->u.dc.res; + } + hp = HAlloc(p, PROC_BIN_SIZE); + ctx->trap_bin = erts_mk_magic_binary_term(&hp, &MSO(p), context_b); + return ctx; +} - if ((bytes = erts_get_aligned_binary_bytes(BIF_ARG_1, &temp_alloc)) == NULL) { - error: - erts_free_aligned_binary_bytes(temp_alloc); - BIF_ERROR(BIF_P, BADARG); +static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* context_b) +{ +#ifdef EXTREME_B2T_TRAPPING + SWord initial_reds = 1 + b2t_rand() % 4; +#else + SWord initial_reds = (Uint)(ERTS_BIF_REDS_LEFT(p) * B2T_BYTES_PER_REDUCTION); +#endif + B2TContext c_buff; + B2TContext *ctx; + int is_first_call; + + if (context_b == NULL) { + /* Setup enough to get started */ + is_first_call = 1; + ctx = &c_buff; + ctx->state = B2TPrepare; + ctx->aligned_alloc = NULL; + ctx->flags = flags; + IF_DEBUG(ctx->trap_bin = THE_NON_VALUE;) + } else { + is_first_call = 0; + ctx = ERTS_MAGIC_BIN_DATA(context_b); + ASSERT(ctx->state != B2TPrepare); } - size = binary_size(BIF_ARG_1); + ctx->reds = initial_reds; + + do { + switch (ctx->state) { + case B2TPrepare: { + byte* bytes; + Uint bin_size; + bytes = erts_get_aligned_binary_bytes_extra(bin, + &ctx->aligned_alloc, + ERTS_ALC_T_EXT_TERM_DATA, + 0); + if (bytes == NULL) { + ctx->b2ts.exttmp = 0; + ctx->state = B2TBadArg; + break; + } + bin_size = binary_size(bin); + if (ctx->aligned_alloc) { + ctx->reds -= bin_size / 8; + } + if (binary2term_prepare(&ctx->b2ts, bytes, bin_size, ctx) < 0) { + ctx->state = B2TBadArg; + } + break; + } + case B2TUncompressChunk: { + uLongf chunk = ctx->reds; + int zret; + + if (chunk > ctx->u.uc.dleft) + chunk = ctx->u.uc.dleft; + zret = erl_zlib_inflate_chunk(&ctx->u.uc.stream, + ctx->u.uc.dbytes, &chunk); + ctx->u.uc.dbytes += chunk; + ctx->u.uc.dleft -= chunk; + if (zret == Z_OK && ctx->u.uc.dleft > 0) { + ctx->reds = 0; + } + else if (erl_zlib_inflate_finish(&ctx->u.uc.stream) == Z_OK + && zret == Z_STREAM_END + && ctx->u.uc.dleft == 0) { + ctx->reds -= chunk; + ctx->state = B2TSizeInit; + } + else { + ctx->state = B2TBadArg; + } + break; + } + case B2TSizeInit: + ctx->u.sc.ep = NULL; + ctx->state = B2TSize; + /*fall through*/ + case B2TSize: + ctx->heap_size = decoded_size(ctx->b2ts.extp, + ctx->b2ts.extp + ctx->b2ts.extsize, + 0, ctx); + break; + + case B2TDecodeInit: + if (ctx == &c_buff && ctx->b2ts.extsize > ctx->reds) { + /* dec_term will maybe trap, allocate space for magic bin + before result term to make it easy to trim with HRelease. + */ + ctx = b2t_export_context(p, &c_buff); + } + ctx->u.dc.ep = ctx->b2ts.extp; + ctx->u.dc.res = (Eterm) (UWord) NULL; + ctx->u.dc.next = &ctx->u.dc.res; + ctx->u.dc.hp_start = HAlloc(p, ctx->heap_size); + ctx->u.dc.hp = ctx->u.dc.hp_start; + ctx->u.dc.hp_end = ctx->u.dc.hp_start + ctx->heap_size; + ctx->state = B2TDecode; + /*fall through*/ + case B2TDecode: + case B2TDecodeList: + case B2TDecodeTuple: + case B2TDecodeString: + case B2TDecodeBinary: { + ErtsDistExternal fakedep; + fakedep.flags = ctx->flags; + dec_term(&fakedep, NULL, NULL, &MSO(p), NULL, ctx); + break; + } + case B2TDecodeFail: + HRelease(p, ctx->u.dc.hp_end, ctx->u.dc.hp_start); + /*fall through*/ + case B2TBadArg: + b2t_destroy_context(ctx); + if (!is_first_call) { + erts_set_gc_state(p, 1); + } + BUMP_REDS(p, (initial_reds - ctx->reds) / B2T_BYTES_PER_REDUCTION); + BIF_ERROR(p, BADARG & ~EXF_SAVETRACE); - heap_size = binary2term_prepare(&b2ts, bytes, size); - if (heap_size < 0) - goto error; + case B2TDone: + b2t_destroy_context(ctx); - hp = HAlloc(BIF_P, heap_size); - endp = hp + heap_size; + if (ctx->u.dc.hp > ctx->u.dc.hp_end) { + erl_exit(1, ":%s, line %d: heap overrun by %d words(s)\n", + __FILE__, __LINE__, ctx->u.dc.hp - ctx->u.dc.hp_end); + } + HRelease(p, ctx->u.dc.hp_end, ctx->u.dc.hp); - res = binary2term_create(NULL, &b2ts, &hp, &MSO(BIF_P)); + if (!is_first_call) { + erts_set_gc_state(p, 1); + } + BUMP_REDS(p, (initial_reds - ctx->reds) / B2T_BYTES_PER_REDUCTION); + return ctx->u.dc.res; - erts_free_aligned_binary_bytes(temp_alloc); + default: + ASSERT(!"Unknown state in binary_to_term"); + } + }while (ctx->reds > 0 || ctx->state >= B2TDone); - if (hp > endp) { - erl_exit(1, ":%s, line %d: heap overrun by %d words(s)\n", - __FILE__, __LINE__, hp-endp); + if (ctx == &c_buff) { + ASSERT(ctx->trap_bin == THE_NON_VALUE); + ctx = b2t_export_context(p, &c_buff); } + ASSERT(ctx->trap_bin != THE_NON_VALUE); - HRelease(BIF_P, endp, hp); - - if (res == THE_NON_VALUE) - goto error; + if (is_first_call) { + erts_set_gc_state(p, 0); + } + BUMP_ALL_REDS(p); + BIF_TRAP1(&binary_to_term_trap_export, p, ctx->trap_bin); +} - return res; +BIF_RETTYPE erts_internal_binary_to_term_1(BIF_ALIST_1) +{ + return binary_to_term_int(BIF_P, 0, BIF_ARG_1, NULL); } -BIF_RETTYPE binary_to_term_2(BIF_ALIST_2) +BIF_RETTYPE erts_internal_binary_to_term_2(BIF_ALIST_2) { - Sint heap_size; - Eterm res; Eterm opts; Eterm opt; - Eterm* hp; - Eterm* endp; - Sint size; - byte* bytes; - byte* temp_alloc = NULL; - ErtsBinary2TermState b2ts; - ErtsDistExternal fakedep; + Uint32 flags = 0; - fakedep.flags = 0; opts = BIF_ARG_2; while (is_list(opts)) { opt = CAR(list_val(opts)); if (opt == am_safe) { - fakedep.flags |= ERTS_DIST_EXT_BTT_SAFE; + flags |= ERTS_DIST_EXT_BTT_SAFE; } else { goto error; @@ -1297,35 +1567,10 @@ BIF_RETTYPE binary_to_term_2(BIF_ALIST_2) if (is_not_nil(opts)) goto error; - if ((bytes = erts_get_aligned_binary_bytes(BIF_ARG_1, &temp_alloc)) == NULL) { - error: - erts_free_aligned_binary_bytes(temp_alloc); - BIF_ERROR(BIF_P, BADARG); - } - size = binary_size(BIF_ARG_1); - - heap_size = binary2term_prepare(&b2ts, bytes, size); - if (heap_size < 0) - goto error; - - hp = HAlloc(BIF_P, heap_size); - endp = hp + heap_size; - - res = binary2term_create(&fakedep, &b2ts, &hp, &MSO(BIF_P)); - - erts_free_aligned_binary_bytes(temp_alloc); - - if (hp > endp) { - erl_exit(1, ":%s, line %d: heap overrun by %d words(s)\n", - __FILE__, __LINE__, hp-endp); - } - - HRelease(BIF_P, endp, hp); - - if (res == THE_NON_VALUE) - goto error; + return binary_to_term_int(BIF_P, flags, BIF_ARG_1, NULL); - return res; +error: + BIF_ERROR(BIF_P, BADARG); } Eterm @@ -1473,25 +1718,29 @@ erts_term_to_binary(Process* p, Eterm Term, int level, Uint flags) { /* #define EXTREME_TTB_TRAPPING 1 */ #ifndef EXTREME_TTB_TRAPPING -#define TERM_TO_BINARY_LOOP_FACTOR 500 -#define TERM_TO_BINARY_SIZE_FACTOR 500000 -#define TERM_TO_BINARY_COMPRESS_CHUNK 500000 +#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_SIZE_FACTOR 10 #define TERM_TO_BINARY_COMPRESS_CHUNK 10 #endif typedef enum { TTBSize, TTBEncode, TTBCompress } TTBState; -typedef struct { +typedef struct TTBSizeContext_ { Uint flags; int level; + Uint result; + Eterm obj; + ErtsEStack estack; } TTBSizeContext; -typedef struct { +typedef struct TTBEncodeContext_ { Uint flags; int level; + byte* ep; + Eterm obj; + ErtsWStack wstack; Binary *result_bin; } TTBEncodeContext; @@ -1514,15 +1763,17 @@ typedef struct { } s; } TTBContext; -static void context_destructor(Binary *context_bin) +static void ttb_context_destructor(Binary *context_bin) { TTBContext *context = ERTS_MAGIC_BIN_DATA(context_bin); if (context->alive) { context->alive = 0; switch (context->state) { case TTBSize: + DESTROY_SAVED_ESTACK(&context->s.sc.estack); break; case TTBEncode: + DESTROY_SAVED_WSTACK(&context->s.ec.wstack); if (context->s.ec.result_bin != NULL) { /* Set to NULL if ever made alive! */ ASSERT(erts_refc_read(&(context->s.ec.result_bin->refc),0) == 0); erts_bin_free(context->s.ec.result_bin); @@ -1567,7 +1818,7 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla do { \ if (context_b == NULL) { \ context_b = erts_create_magic_binary(sizeof(TTBContext), \ - context_destructor); \ + ttb_context_destructor); \ context = ERTS_MAGIC_BIN_DATA(context_b); \ memcpy(context,&c_buff,sizeof(TTBContext)); \ } \ @@ -1587,6 +1838,7 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla /* Setup enough to get started */ context->state = TTBSize; context->alive = 1; + context->s.sc.estack.start = NULL; context->s.sc.flags = flags; context->s.sc.level = level; } else { @@ -1602,7 +1854,8 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla int level; Uint flags; /* Try for fast path */ - if (encode_size_struct_int(p, NULL, Term, context->s.sc.flags, &reds, &size) < 0) { + if (encode_size_struct_int(&context->s.sc, NULL, Term, + context->s.sc.flags, &reds, &size) < 0) { EXPORT_CONTEXT(); /* Same state */ RETURN_STATE(); @@ -1615,7 +1868,7 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla /* Finish in one go */ res = erts_term_to_binary_simple(p, Term, size, level, flags); - BUMP_REDS(p, size / TERM_TO_BINARY_SIZE_FACTOR); + BUMP_REDS(p, 1); return res; } @@ -1628,6 +1881,7 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla context->state = TTBEncode; context->s.ec.flags = flags; context->s.ec.level = level; + context->s.ec.wstack.wstart = NULL; context->s.ec.result_bin = result_bin; break; } @@ -1639,7 +1893,7 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla Binary *result_bin; flags = context->s.ec.flags; - if (enc_term_int(p,NULL,Term, bytes+1, flags, NULL, &reds, &endp) < 0) { + if (enc_term_int(&context->s.ec, NULL,Term, bytes+1, flags, NULL, &reds, &endp) < 0) { EXPORT_CONTEXT(); RETURN_STATE(); } @@ -2047,27 +2301,6 @@ dec_pid(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, Ete #define ENC_PATCH_FUN_SIZE ((Eterm) 2) #define ENC_LAST_ARRAY_ELEMENT ((Eterm) 3) -/* Free extra rootset (used when trapping) */ -static void cleanup_ttb_extra_root(ErlExtraRootSet *rs) -{ - if (rs->objv != NULL) { - erts_free(ERTS_ALC_T_EXTRA_ROOT, rs->objv); - } - erts_free(ERTS_ALC_T_EXTRA_ROOT, rs); -} - -/* Same as above, but we have an extra "stack" beyond GC reach, i.e. an array of two extra roots */ -static void cleanup_ttb_extra_root_2(ErlExtraRootSet *rs) -{ - if (rs->objv != NULL) { - erts_free(ERTS_ALC_T_EXTRA_ROOT, rs->objv); - } - if (rs[1].objv != NULL) { - erts_free(ERTS_ALC_T_EXTRA_ROOT, rs[1].objv); - } - - erts_free(ERTS_ALC_T_EXTRA_ROOT, rs); -} static byte* enc_term(ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dflags, @@ -2079,39 +2312,43 @@ enc_term(ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dflags, } static int -enc_term_int(Process *p,ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dflags, +enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dflags, struct erl_off_heap_header** off_heap, Sint *reds, byte **res) { - DECLARE_ESTACK(s); - DECLARE_WSTACK(com); + DECLARE_WSTACK(s); Uint n; Uint i; Uint j; Uint* ptr; Eterm val; FloatDef f; - int count_reds = (p != NULL && reds != NULL); Sint r = 0; +#if HALFWORD_HEAP + UWord wobj; +#endif - if (count_reds) { - ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_EXTRA_ROOT); - WSTACK_CHANGE_ALLOCATOR(com, ERTS_ALC_T_EXTRA_ROOT); + + if (ctx) { + WSTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK); r = *reds; - } - if (p && p->extra_root) { /* restore saved stacks and byte pointer */ - ESTACK_RESTORE(s,p->extra_root[0].objv, p->extra_root[0].sz); - obj = ESTACK_POP(s); - WSTACK_RESTORE(com, p->extra_root[1].objv, p->extra_root[1].sz); - ep = (byte *) WSTACK_POP(com); + if (ctx->wstack.wstart) { /* restore saved stacks and byte pointer */ + WSTACK_RESTORE(s, &ctx->wstack); + ep = ctx->ep; + obj = ctx->obj; + } } goto L_jump_start; outer_loop: - while (!ESTACK_ISEMPTY(s)) { - obj = ESTACK_POP(s); - switch (val = WSTACK_POP(com)) { + while (!WSTACK_ISEMPTY(s)) { +#if HALFWORD_HEAP + obj = (Eterm) (wobj = WSTACK_POP(s)); +#else + obj = WSTACK_POP(s); +#endif + switch (val = WSTACK_POP(s)) { case ENC_TERM: break; case ENC_ONE_CONS: @@ -2122,55 +2359,52 @@ enc_term_int(Process *p,ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dfla obj = CAR(cons); tl = CDR(cons); - WSTACK_PUSH(com, is_list(tl) ? ENC_ONE_CONS : ENC_TERM); - ESTACK_PUSH(s, tl); + WSTACK_PUSH(s, is_list(tl) ? ENC_ONE_CONS : ENC_TERM); + WSTACK_PUSH(s, tl); } break; case ENC_PATCH_FUN_SIZE: - /* obj will be discarded, it was NIL */ { - byte* size_p = (byte *) WSTACK_POP(com); +#if HALFWORD_HEAP + byte* size_p = (byte *) wobj; +#else + byte* size_p = (byte *) obj; +#endif put_int32(ep - size_p, size_p); } goto outer_loop; case ENC_LAST_ARRAY_ELEMENT: /* obj is the tuple */ { - Eterm* ptr = tuple_val(obj); - i = arityval(*ptr); - obj = ptr[i]; +#if HALFWORD_HEAP + Eterm* ptr = (Eterm *) wobj; +#else + Eterm* ptr = (Eterm *) obj; +#endif + obj = *ptr; } break; default: /* ENC_LAST_ARRAY_ELEMENT+1 and upwards */ { - Eterm* ptr = tuple_val(obj); - i = arityval(*ptr); - ESTACK_PUSH(s, obj); /* put back tuple and next element index */ - WSTACK_PUSH(com, val-1); - obj = ptr[i - (val - ENC_LAST_ARRAY_ELEMENT)]; /* the index is counting down */ +#if HALFWORD_HEAP + Eterm* ptr = (Eterm *) wobj; +#else + Eterm* ptr = (Eterm *) obj; +#endif + WSTACK_PUSH(s, val-1); + obj = *ptr++; + WSTACK_PUSH(s, (UWord)ptr); } break; } L_jump_start: - if (count_reds && --r == 0) { + if (ctx && --r == 0) { *reds = r; - ESTACK_PUSH(s,obj); /* push back current object, to be popped on restore */ - WSTACK_PUSH(com,((UWord) ep)); - if (p->extra_root == NULL) { - /* NB. Allocate an array of two "extra-roots", of which only the first element - is seen and handled by the GC. Index 1 holds the Wstack. */ - p->extra_root = erts_alloc(ERTS_ALC_T_EXTRA_ROOT, sizeof(ErlExtraRootSet)*2); - p->extra_root->objv = NULL; - p->extra_root->sz = 0; - p->extra_root->cleanup = cleanup_ttb_extra_root_2; - p->extra_root[1].objv = NULL; - p->extra_root[1].sz = 0; - p->extra_root[1].cleanup = NULL; /* Never used */ - } - ESTACK_SAVE(s, p->extra_root[0].objv, p->extra_root[0].sz); - WSTACK_SAVE(com, p->extra_root[1].objv, (p->extra_root[1].sz)); + ctx->obj = obj; + ctx->ep = ep; + WSTACK_SAVE(s, &ctx->wstack); return -1; } switch(tag_val_def(obj)) { @@ -2316,8 +2550,8 @@ enc_term_int(Process *p,ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dfla ep += 4; } if (i > 0) { - WSTACK_PUSH(com, ENC_LAST_ARRAY_ELEMENT+i-1); - ESTACK_PUSH(s, obj); + WSTACK_PUSH(s, ENC_LAST_ARRAY_ELEMENT+i-1); + WSTACK_PUSH(s, (UWord)ptr); } break; @@ -2461,9 +2695,8 @@ enc_term_int(Process *p,ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dfla int ei; *ep++ = NEW_FUN_EXT; - WSTACK_PUSH(com, (UWord) ep); /* Position for patching in size */ - WSTACK_PUSH(com, ENC_PATCH_FUN_SIZE); - ESTACK_PUSH(s,NIL); /* Will be thrown away */ + WSTACK_PUSH(s, ENC_PATCH_FUN_SIZE); + WSTACK_PUSH(s, (UWord) ep); /* Position for patching in size */ ep += 4; *ep = funp->arity; ep += 1; @@ -2480,8 +2713,8 @@ enc_term_int(Process *p,ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dfla fun_env: for (ei = funp->num_free-1; ei > 0; ei--) { - WSTACK_PUSH(com, ENC_TERM); - ESTACK_PUSH(s, (UWord) funp->env[ei]); + WSTACK_PUSH(s, ENC_TERM); + WSTACK_PUSH(s, (UWord) funp->env[ei]); } if (funp->num_free != 0) { obj = funp->env[0]; @@ -2524,13 +2757,9 @@ enc_term_int(Process *p,ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, Uint32 dfla break; } } - DESTROY_ESTACK(s); - DESTROY_WSTACK(com); - if (p && p->extra_root) { - cleanup_ttb_extra_root_2(p->extra_root); - p->extra_root = NULL; - } - if (count_reds) { + DESTROY_WSTACK(s); + if (ctx) { + ASSERT(ctx->wstack.wstart == NULL); *reds = r; } *res = ep; @@ -2604,21 +2833,112 @@ undo_offheap_in_area(ErlOffHeap* off_heap, Eterm* start, Eterm* end) #endif /* DEBUG */ } + /* Decode term from external format into *objp. ** On failure return NULL and (R13B04) *hpp will be unchanged. */ static byte* -dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, Eterm* objp) +dec_term(ErtsDistExternal *edep, Eterm** hpp, byte* ep, ErlOffHeap* off_heap, + Eterm* objp, B2TContext* ctx) { - Eterm* hp_saved = *hpp; + Eterm* hp_saved; int n; ErtsAtomEncoding char_enc; - register Eterm* hp = *hpp; /* Please don't take the address of hp */ - Eterm* next = objp; + register Eterm* hp; /* Please don't take the address of hp */ + Eterm* next; + SWord reds; + + if (ctx) { + hp_saved = ctx->u.dc.hp_start; + reds = ctx->reds; + next = ctx->u.dc.next; + ep = ctx->u.dc.ep; + hpp = &ctx->u.dc.hp; + + if (ctx->state != B2TDecode) { + int n_limit = reds; + + n = ctx->u.dc.remaining_n; + if (ctx->state == B2TDecodeBinary) { + n_limit *= B2T_MEMCPY_FACTOR; + ASSERT(n_limit >= reds); + reds -= n / B2T_MEMCPY_FACTOR; + } + else + reds -= n; - *next = (Eterm) (UWord) NULL; + if (n > n_limit) { + ctx->u.dc.remaining_n -= n_limit; + n = n_limit; + reds = 0; + } + else { + ctx->u.dc.remaining_n = 0; + } + + switch (ctx->state) { + case B2TDecodeList: + objp = next - 2; + while (n > 0) { + objp[0] = (Eterm) COMPRESS_POINTER(next); + objp[1] = make_list(next); + next = objp; + objp -= 2; + n--; + } + break; + + case B2TDecodeTuple: + objp = next - 1; + while (n-- > 0) { + objp[0] = (Eterm) COMPRESS_POINTER(next); + next = objp; + objp--; + } + break; + + case B2TDecodeString: + hp = *hpp; + hp[-1] = make_list(hp); /* overwrite the premature NIL */ + while (n-- > 0) { + hp[0] = make_small(*ep++); + hp[1] = make_list(hp+2); + hp += 2; + } + hp[-1] = NIL; + *hpp = hp; + break; + + case B2TDecodeBinary: + sys_memcpy(ctx->u.dc.remaining_bytes, ep, n); + ctx->u.dc.remaining_bytes += n; + ep += n; + break; + + default: + ASSERT(!"Unknown state"); + } + if (!ctx->u.dc.remaining_n) { + ctx->state = B2TDecode; + } + if (reds <= 0) { + ctx->u.dc.next = next; + ctx->u.dc.ep = ep; + ctx->reds = 0; + return NULL; + } + } + } + else { + hp_saved = *hpp; + reds = ERTS_SWORD_MAX; + next = objp; + *next = (Eterm) (UWord) NULL; + } + hp = *hpp; while (next != NULL) { + objp = next; next = (Eterm *) EXPAND_POINTER(*objp); @@ -2738,7 +3058,16 @@ dec_term_atom_common: *objp = make_tuple(hp); *hp++ = make_arityval(n); hp += n; - objp = hp - 1; + objp = hp - 1; + if (ctx) { + if (reds < n) { + ASSERT(reds > 0); + ctx->state = B2TDecodeTuple; + ctx->u.dc.remaining_n = n - reds; + n = reds; + } + reds -= n; + } while (n-- > 0) { objp[0] = (Eterm) COMPRESS_POINTER(next); next = objp; @@ -2756,17 +3085,27 @@ dec_term_atom_common: break; } *objp = make_list(hp); - hp += 2*n; + hp += 2 * n; objp = hp - 2; objp[0] = (Eterm) COMPRESS_POINTER((objp+1)); objp[1] = (Eterm) COMPRESS_POINTER(next); next = objp; objp -= 2; - while (--n > 0) { + n--; + if (ctx) { + if (reds < n) { + ctx->state = B2TDecodeList; + ctx->u.dc.remaining_n = n - reds; + n = reds; + } + reds -= n; + } + while (n > 0) { objp[0] = (Eterm) COMPRESS_POINTER(next); - objp[1] = make_list(objp + 2); + objp[1] = make_list(next); next = objp; objp -= 2; + n--; } break; case STRING_EXT: @@ -2777,6 +3116,14 @@ dec_term_atom_common: break; } *objp = make_list(hp); + if (ctx) { + if (reds < n) { + ctx->state = B2TDecodeString; + ctx->u.dc.remaining_n = n - reds; + n = reds; + } + reds -= n; + } while (n-- > 0) { hp[0] = make_small(*ep++); hp[1] = make_list(hp+2); @@ -2984,7 +3331,6 @@ dec_term_atom_common: dbin->flags = 0; dbin->orig_size = n; erts_refc_init(&dbin->refc, 1); - sys_memcpy(dbin->orig_bytes, ep, n); pb = (ProcBin *) hp; hp += PROC_BIN_SIZE; pb->thing_word = HEADER_PROC_BIN; @@ -2995,7 +3341,20 @@ dec_term_atom_common: pb->bytes = (byte*) dbin->orig_bytes; pb->flags = 0; *objp = make_binary(pb); - } + if (ctx) { + int n_limit = reds * B2T_MEMCPY_FACTOR; + if (n > n_limit) { + ctx->state = B2TDecodeBinary; + ctx->u.dc.remaining_n = n - n_limit; + ctx->u.dc.remaining_bytes = dbin->orig_bytes + n_limit; + n = n_limit; + reds = 0; + } + else + reds -= n / B2T_MEMCPY_FACTOR; + } + sys_memcpy(dbin->orig_bytes, ep, n); + } ep += n; break; } @@ -3018,13 +3377,14 @@ dec_term_atom_common: sys_memcpy(hb->data, ep, n); bin = make_binary(hb); hp += heap_bin_size(n); + ep += n; } else { Binary* dbin = erts_bin_nrml_alloc(n); ProcBin* pb; + dbin->flags = 0; dbin->orig_size = n; erts_refc_init(&dbin->refc, 1); - sys_memcpy(dbin->orig_bytes, ep, n); pb = (ProcBin *) hp; pb->thing_word = HEADER_PROC_BIN; pb->size = n; @@ -3035,8 +3395,23 @@ dec_term_atom_common: pb->flags = 0; bin = make_binary(pb); hp += PROC_BIN_SIZE; - } - ep += n; + if (ctx) { + int n_limit = reds * B2T_MEMCPY_FACTOR; + if (n > n_limit) { + ctx->state = B2TDecodeBinary; + ctx->u.dc.remaining_n = n - n_limit; + ctx->u.dc.remaining_bytes = dbin->orig_bytes + n_limit; + n = n_limit; + reds = 0; + } + else + reds -= n / B2T_MEMCPY_FACTOR; + } + sys_memcpy(dbin->orig_bytes, ep, n); + ep += n; + n = pb->size; + } + if (bitsize == 8 || n == 0) { *objp = bin; } else { @@ -3067,7 +3442,7 @@ dec_term_atom_common: goto error; } *hpp = hp; - ep = dec_term(edep, hpp, ep, off_heap, &temp); + ep = dec_term(edep, hpp, ep, off_heap, &temp, NULL); hp = *hpp; if (ep == NULL) { goto error; @@ -3127,7 +3502,7 @@ dec_term_atom_common: } *hpp = hp; /* Index */ - if ((ep = dec_term(edep, hpp, ep, off_heap, &temp)) == NULL) { + if ((ep = dec_term(edep, hpp, ep, off_heap, &temp, NULL)) == NULL) { goto error; } if (!is_small(temp)) { @@ -3136,7 +3511,7 @@ dec_term_atom_common: old_index = unsigned_val(temp); /* Uniq */ - if ((ep = dec_term(edep, hpp, ep, off_heap, &temp)) == NULL) { + if ((ep = dec_term(edep, hpp, ep, off_heap, &temp, NULL)) == NULL) { goto error; } if (!is_small(temp)) { @@ -3204,7 +3579,7 @@ dec_term_atom_common: } /* Index */ - if ((ep = dec_term(edep, hpp, ep, off_heap, &temp)) == NULL) { + if ((ep = dec_term(edep, hpp, ep, off_heap, &temp, NULL)) == NULL) { goto error; } if (!is_small(temp)) { @@ -3213,7 +3588,7 @@ dec_term_atom_common: old_index = unsigned_val(temp); /* Uniq */ - if ((ep = dec_term(edep, hpp, ep, off_heap, &temp)) == NULL) { + if ((ep = dec_term(edep, hpp, ep, off_heap, &temp, NULL)) == NULL) { goto error; } if (!is_small(temp)) { @@ -3313,8 +3688,31 @@ dec_term_atom_common: } undo_offheap_in_area(off_heap, hp_saved, hp); *hpp = hp_saved; - return NULL; + if (ctx) { + ctx->state = B2TDecodeFail; + ctx->reds = reds; + } + return NULL; } + + if (--reds <= 0) { + if (ctx) { + if (next || ctx->state != B2TDecode) { + ctx->u.dc.ep = ep; + ctx->u.dc.next = next; + ctx->u.dc.hp = hp; + ctx->reds = 0; + return NULL; + } + } + else { + reds = ERTS_SWORD_MAX; + } + } + } + if (ctx) { + ctx->state = B2TDone; + ctx->reds = reds; } *hpp = hp; return ep; @@ -3331,26 +3729,24 @@ static Uint encode_size_struct2(ErtsAtomCacheMap *acmp, Eterm obj, unsigned dfla } static int -encode_size_struct_int(Process *p, ErtsAtomCacheMap *acmp, Eterm obj, +encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, unsigned dflags, Sint *reds, Uint *res) { DECLARE_ESTACK(s); Uint m, i, arity; Uint result = 0; - int count_reds = (p != NULL && reds != 0); Sint r = 0; - if (count_reds) { - ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_EXTRA_ROOT); + if (ctx) { + ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK); r = *reds; - } - if (p && p->extra_root) { /* restore saved stack */ - ESTACK_RESTORE(s,p->extra_root->objv, p->extra_root->sz + 1); - result = ESTACK_POP(s); /*Untagged, beyond p->extra_root->sz */ - obj = ESTACK_POP(s); - - } + if (ctx->estack.start) { /* restore saved stack */ + ESTACK_RESTORE(s, &ctx->estack); + result = ctx->result; + obj = ctx->obj; + } + } goto L_jump_start; @@ -3376,18 +3772,11 @@ encode_size_struct_int(Process *p, ErtsAtomCacheMap *acmp, Eterm obj, } L_jump_start: - if (count_reds && --r == 0) { + if (ctx && --r == 0) { *reds = r; - ESTACK_PUSH(s,obj); /* push back current object */ - ESTACK_PUSH(s,result); /* Untagged, will be out of GC reach */ - if (p->extra_root == NULL) { - p->extra_root = erts_alloc(ERTS_ALC_T_EXTRA_ROOT, sizeof(ErlExtraRootSet)); - p->extra_root->objv = NULL; - p->extra_root->sz = 0; - p->extra_root->cleanup = cleanup_ttb_extra_root; - } - ESTACK_SAVE(s, p->extra_root->objv, p->extra_root->sz); - --p->extra_root->sz; /* Hide result from GC */ + ctx->obj = obj; + ctx->result = result; + ESTACK_SAVE(s, &ctx->estack); return -1; } switch (tag_val_def(obj)) { @@ -3590,11 +3979,8 @@ encode_size_struct_int(Process *p, ErtsAtomCacheMap *acmp, Eterm obj, } DESTROY_ESTACK(s); - if (p && p->extra_root) { - cleanup_ttb_extra_root(p->extra_root); - p->extra_root = NULL; - } - if (count_reds) { + if (ctx) { + ASSERT(ctx->estack.start == NULL); *reds = r; } *res = result; @@ -3602,18 +3988,37 @@ encode_size_struct_int(Process *p, ErtsAtomCacheMap *acmp, Eterm obj, } static Sint -decoded_size(byte *ep, byte* endp, int internal_tags) +decoded_size(byte *ep, byte* endp, int internal_tags, B2TContext* ctx) { - int heap_size = 0; + int heap_size; int terms; - int atom_extra_skip = 0; + int atom_extra_skip; Uint n; + SWord reds; + + if (ctx) { + reds = ctx->reds; + if (ctx->u.sc.ep) { + heap_size = ctx->u.sc.heap_size; + terms = ctx->u.sc.terms; + ep = ctx->u.sc.ep; + atom_extra_skip = ctx->u.sc.atom_extra_skip; + goto init_done; + } + } + else + reds = 0; /* not used but compiler warns anyway */ + + heap_size = 0; + terms = 1; + atom_extra_skip = 0; +init_done: #define SKIP(sz) \ do { \ if ((sz) <= endp-ep) { \ ep += (sz); \ - } else { return -1; }; \ + } else { goto error; }; \ } while (0) #define SKIP2(sz1, sz2) \ @@ -3621,31 +4026,32 @@ decoded_size(byte *ep, byte* endp, int internal_tags) Uint sz = (sz1) + (sz2); \ if (sz1 < sz && (sz) <= endp-ep) { \ ep += (sz); \ - } else { return -1; } \ + } else { goto error; } \ } while (0) #define CHKSIZE(sz) \ do { \ - if ((sz) > endp-ep) { return -1; } \ + if ((sz) > endp-ep) { goto error; } \ } while (0) #define ADDTERMS(n) \ do { \ int before = terms; \ terms += (n); \ - if (terms < before) return -1; \ + if (terms < before) goto error; \ } while (0) - - for (terms=1; terms > 0; terms--) { - int tag; - + ASSERT(terms > 0); + do { + int tag; CHKSIZE(1); tag = ep++[0]; switch (tag) { case INTEGER_EXT: SKIP(4); +#if !defined(ARCH_64) || HALFWORD_HEAP heap_size += BIG_UINT_HEAP_SIZE; +#endif break; case SMALL_INTEGER_EXT: SKIP(1); @@ -3660,7 +4066,7 @@ decoded_size(byte *ep, byte* endp, int internal_tags) CHKSIZE(4); n = get_int32(ep); if (n > BIG_ARITY_MAX*sizeof(ErtsDigit)) { - return -1; + goto error; } SKIP2(n,4+1); /* skip, size,sign,digits */ heap_size += 1+1+(n+sizeof(Eterm)-1)/sizeof(Eterm); /* XXX: 1 too much? */ @@ -3669,7 +4075,7 @@ decoded_size(byte *ep, byte* endp, int internal_tags) CHKSIZE(2); n = get_int16(ep); if (n > MAX_ATOM_CHARACTERS) { - return -1; + goto error; } SKIP(n+2+atom_extra_skip); atom_extra_skip = 0; @@ -3679,7 +4085,7 @@ decoded_size(byte *ep, byte* endp, int internal_tags) n = get_int16(ep); ep += 2; if (n > MAX_ATOM_SZ_LIMIT) { - return -1; + goto error; } SKIP(n+atom_extra_skip); atom_extra_skip = 0; @@ -3688,7 +4094,7 @@ decoded_size(byte *ep, byte* endp, int internal_tags) CHKSIZE(1); n = get_int8(ep); if (n > MAX_ATOM_CHARACTERS) { - return -1; + goto error; } SKIP(n+1+atom_extra_skip); atom_extra_skip = 0; @@ -3698,7 +4104,7 @@ decoded_size(byte *ep, byte* endp, int internal_tags) n = get_int8(ep); ep++; if (n > MAX_ATOM_SZ_LIMIT) { - return -1; + goto error; } SKIP(n+atom_extra_skip); atom_extra_skip = 0; @@ -3727,7 +4133,7 @@ decoded_size(byte *ep, byte* endp, int internal_tags) id_words = get_int16(ep); if (id_words > ERTS_MAX_REF_NUMBERS) - return -1; + goto error; ep += 2; atom_extra_skip = 1 + 4*id_words; @@ -3829,7 +4235,7 @@ decoded_size(byte *ep, byte* endp, int internal_tags) num_free = get_int32(ep); ep += 4; if (num_free > MAX_ARG) { - return -1; + goto error; } terms += 4 + num_free; heap_size += ERL_FUN_SIZE + num_free; @@ -3846,24 +4252,47 @@ decoded_size(byte *ep, byte* endp, int internal_tags) case BINARY_INTERNAL_REF: if (!internal_tags) { - return -1; + goto error; } SKIP(sizeof(ProcBin)); heap_size += PROC_BIN_SIZE; break; case BIT_BINARY_INTERNAL_REF: if (!internal_tags) { - return -1; + goto error; } SKIP(2+sizeof(ProcBin)); heap_size += PROC_BIN_SIZE + ERL_SUB_BIN_SIZE; break; default: - return -1; + goto error; } - } + terms--; + + if (ctx && --reds <= 0 && terms > 0) { + ctx->u.sc.heap_size = heap_size; + ctx->u.sc.terms = terms; + ctx->u.sc.ep = ep; + ctx->u.sc.atom_extra_skip = atom_extra_skip; + ctx->reds = 0; + return 0; + } + }while (terms > 0); + /* 'terms' may be non-zero if it has wrapped around */ - return terms==0 ? heap_size : -1; + if (terms == 0) { + if (ctx) { + ctx->state = B2TDecodeInit; + ctx->reds = reds; + } + return heap_size; + } + +error: + if (ctx) { + ctx->state = B2TBadArg; + } + return -1; #undef SKIP #undef SKIP2 #undef CHKSIZE diff --git a/erts/emulator/beam/external.h b/erts/emulator/beam/external.h index ff29e84972..83001b2c7e 100644 --- a/erts/emulator/beam/external.h +++ b/erts/emulator/beam/external.h @@ -146,6 +146,7 @@ typedef struct { typedef struct { byte *extp; int exttmp; + Uint extsize; } ErtsBinary2TermState; /* -------------------------------------------------------------------------- */ diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h index 063d16c0c7..c183c519ff 100755 --- a/erts/emulator/beam/global.h +++ b/erts/emulator/beam/global.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 @@ -186,11 +186,6 @@ extern void erts_ddll_remove_monitor(Process *p, extern Eterm erts_ddll_monitor_driver(Process *p, Eterm description, ErtsProcLocks plocks); -/* - * Max no. of drivers (linked in and dynamically loaded). Each table - * entry uses 4 bytes. - */ -#define DRIVER_TAB_SIZE 32 /* ** Just like the driver binary but with initial flags @@ -375,231 +370,233 @@ extern int stackdump_on_exit; * DESTROY_ESTACK(Stack) */ +typedef struct { + UWord* start; + UWord* sp; + UWord* end; + ErtsAlcType_t alloc_type; +}ErtsEStack; -void erl_grow_stack(ErtsAlcType_t a_type, Eterm** start, Eterm** sp, Eterm** end); -#define ESTK_CONCAT(a,b) a##b -#define ESTK_SUBSCRIPT(s,i) *((Eterm *)((byte *)ESTK_CONCAT(s,_start) + (i))) #define DEF_ESTACK_SIZE (16) -#define DECLARE_ESTACK(s) \ - Eterm ESTK_CONCAT(s,_default_stack)[DEF_ESTACK_SIZE]; \ - Eterm* ESTK_CONCAT(s,_start) = ESTK_CONCAT(s,_default_stack); \ - Eterm* ESTK_CONCAT(s,_sp) = ESTK_CONCAT(s,_start); \ - Eterm* ESTK_CONCAT(s,_end) = ESTK_CONCAT(s,_start) + DEF_ESTACK_SIZE;\ - ErtsAlcType_t ESTK_CONCAT(s,_alloc_type) = ERTS_ALC_T_ESTACK +void erl_grow_estack(ErtsEStack*, Eterm* def_stack); +#define ESTK_CONCAT(a,b) a##b +#define ESTK_DEF_STACK(s) ESTK_CONCAT(s,_default_estack) + +#define DECLARE_ESTACK(s) \ + UWord ESTK_DEF_STACK(s)[DEF_ESTACK_SIZE]; \ + ErtsEStack s = { \ + ESTK_DEF_STACK(s), /* start */ \ + ESTK_DEF_STACK(s), /* sp */ \ + ESTK_DEF_STACK(s) + DEF_ESTACK_SIZE, /* end */ \ + ERTS_ALC_T_ESTACK /* alloc_type */ \ + } #define ESTACK_CHANGE_ALLOCATOR(s,t) \ do { \ - if (ESTK_CONCAT(s,_start) != ESTK_CONCAT(s,_default_stack)) { \ + if (s.start != ESTK_DEF_STACK(s)) { \ erl_exit(1, "Internal error - trying to change allocator " \ "type of active estack\n"); \ } \ - ESTK_CONCAT(s,_alloc_type) = (t); \ + s.alloc_type = (t); \ } while (0) +#define DESTROY_ESTACK(s) \ +do { \ + if (s.start != ESTK_DEF_STACK(s)) { \ + erts_free(s.alloc_type, s.start); \ + } \ +} while(0) + + /* - * Do not free the stack after this, it may have pointers into what - * was saved in 'v'. 'v' and 'vsize' are changed by this macro. If - * 'v' points to anything, it should have been allocated by a previous - * call to this macro. Be careful to set a correct allocator prior to - * saving. - * 'v' can be any lvalue pointer, it will point to an array of UWord - * after calling this macro. + * Do not free the stack after this, it may have pointers into what + * was saved in 'dst'. */ -#define ESTACK_SAVE(s,v,vsize) /* v and vsize are "name parameters" */ \ -do { \ - Uint _esz = ESTACK_COUNT(s); \ - if (ESTK_CONCAT(s,_start) == ESTK_CONCAT(s,_default_stack)) { \ - if ((v) == NULL) { \ - (v) = erts_alloc(ESTK_CONCAT(s,_alloc_type), \ - DEF_ESTACK_SIZE * sizeof(Eterm)); \ - } \ - memcpy((v),ESTK_CONCAT(s,_start),_esz*sizeof(Eterm)); \ - } else { \ - (v) = (void *) ESTK_CONCAT(s,_start); \ - } \ - (vsize) = _esz; \ +#define ESTACK_SAVE(s,dst)\ +do {\ + if (s.start == ESTK_DEF_STACK(s)) {\ + UWord _wsz = ESTACK_COUNT(s);\ + (dst)->start = erts_alloc(s.alloc_type,\ + DEF_ESTACK_SIZE * sizeof(UWord));\ + memcpy((dst)->start, s.start,_wsz*sizeof(UWord));\ + (dst)->sp = (dst)->start + _wsz;\ + (dst)->end = (dst)->start + DEF_ESTACK_SIZE;\ + (dst)->alloc_type = s.alloc_type;\ + } else\ + *(dst) = s;\ } while (0) -/* - * Use on empty stack, only the allocator can be changed before this - * The vector parameter is reset to NULL if the vector is moved to stack, - * otherwise it's kept for reuse, so a saved and restored vector might - * need freeing using the correct allocator parameter. - * 'v' can be any lvalue pointer, it's cast to an (Eterm *). +#define DESTROY_SAVED_ESTACK(estack)\ +do {\ + if ((estack)->start) {\ + erts_free((estack)->alloc_type, (estack)->start);\ + (estack)->start = NULL;\ + }\ +} while(0) + +/* + * Use on empty stack, only the allocator can be changed before this. + * The src stack is reset to NULL. */ -#define ESTACK_RESTORE(s, v, vsize) /*v is a "name parameter"*/ \ -do { \ - if ((vsize) > DEF_ESTACK_SIZE) { \ - Uint _ca = DEF_ESTACK_SIZE; \ - while (_ca < (vsize)) \ - _ca = _ca * 2; \ - ESTK_CONCAT(s,_start) = (Eterm *) (v); \ - ESTK_CONCAT(s,_end) = ((Eterm *)(v)) + _ca; \ - ESTK_CONCAT(s,_sp) = ESTK_CONCAT(s,_start) + (vsize); \ - (v) = NULL; \ - } else { \ - memcpy(ESTK_CONCAT(s,_start),(v),(vsize)*sizeof(Eterm));\ - ESTK_CONCAT(s,_sp) = ESTK_CONCAT(s,_start) + (vsize); \ - } \ - } while (0) +#define ESTACK_RESTORE(s, src) \ +do { \ + ASSERT(s.start == ESTK_DEF_STACK(s)); \ + s = *(src); /* struct copy */ \ + (src)->start = NULL; \ + ASSERT(s.sp >= s.start); \ + ASSERT(s.sp <= s.end); \ +} while (0) -#define ESTACK_IS_STATIC(s) (ESTK_CONCAT(s,_start) == ESTK_CONCAT(s,_default_stack)) +#define ESTACK_IS_STATIC(s) (s.start == ESTK_DEF_STACK(s))) -#define DESTROY_ESTACK(s) \ -do { \ - if (ESTK_CONCAT(s,_start) != ESTK_CONCAT(s,_default_stack)) { \ - erts_free(ESTK_CONCAT(s,_alloc_type), ESTK_CONCAT(s,_start)); \ - } \ +#define ESTACK_PUSH(s, x) \ +do { \ + if (s.sp == s.end) { \ + erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ + } \ + *s.sp++ = (x); \ } while(0) -#define ESTACK_PUSH(s, x) \ -do { \ - if (ESTK_CONCAT(s,_sp) == ESTK_CONCAT(s,_end)) { \ - erl_grow_stack(ESTK_CONCAT(s,_alloc_type),&ESTK_CONCAT(s,_start), \ - &ESTK_CONCAT(s,_sp), &ESTK_CONCAT(s,_end)); \ - } \ - *ESTK_CONCAT(s,_sp)++ = (x); \ +#define ESTACK_PUSH2(s, x, y) \ +do { \ + if (s.sp > s.end - 2) { \ + erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ + } \ + *s.sp++ = (x); \ + *s.sp++ = (y); \ } while(0) -#define ESTACK_PUSH2(s, x, y) \ -do { \ - if (ESTK_CONCAT(s,_sp) > ESTK_CONCAT(s,_end) - 2) { \ - erl_grow_stack(ESTK_CONCAT(s,_alloc_type),&ESTK_CONCAT(s,_start), \ - &ESTK_CONCAT(s,_sp), &ESTK_CONCAT(s,_end)); \ - } \ - *ESTK_CONCAT(s,_sp)++ = (x); \ - *ESTK_CONCAT(s,_sp)++ = (y); \ +#define ESTACK_PUSH3(s, x, y, z) \ +do { \ + if (s.sp > s.end - 3) { \ + erl_grow_estack(&s, ESTK_DEF_STACK(s)); \ + } \ + *s.sp++ = (x); \ + *s.sp++ = (y); \ + *s.sp++ = (z); \ } while(0) -#define ESTACK_PUSH3(s, x, y, z) \ -do { \ - if (ESTK_CONCAT(s,_sp) > ESTK_CONCAT(s,_end) - 3) { \ - erl_grow_stack(&ESTK_CONCAT(s,_start), &ESTK_CONCAT(s,_sp), \ - &ESTK_CONCAT(s,_end)); \ - } \ - *ESTK_CONCAT(s,_sp)++ = (x); \ - *ESTK_CONCAT(s,_sp)++ = (y); \ - *ESTK_CONCAT(s,_sp)++ = (z); \ -} while(0) +#define ESTACK_COUNT(s) (s.sp - s.start) +#define ESTACK_ISEMPTY(s) (s.sp == s.start) +#define ESTACK_POP(s) (*(--s.sp)) -#define ESTACK_COUNT(s) (ESTK_CONCAT(s,_sp) - ESTK_CONCAT(s,_start)) -#define ESTACK_ISEMPTY(s) (ESTK_CONCAT(s,_sp) == ESTK_CONCAT(s,_start)) -#define ESTACK_POP(s) (*(--ESTK_CONCAT(s,_sp))) +/* + * WSTACK: same as ESTACK but with UWord instead of Eterm + */ +typedef struct { + UWord* wstart; + UWord* wsp; + UWord* wend; + ErtsAlcType_t alloc_type; +}ErtsWStack; -void erl_grow_wstack(ErtsAlcType_t a_type, UWord** start, UWord** sp, UWord** end); -#define WSTK_CONCAT(a,b) a##b -#define WSTK_SUBSCRIPT(s,i) *((UWord *)((byte *)WSTK_CONCAT(s,_start) + (i))) #define DEF_WSTACK_SIZE (16) -#define DECLARE_WSTACK(s) \ - UWord WSTK_CONCAT(s,_default_stack)[DEF_WSTACK_SIZE]; \ - UWord* WSTK_CONCAT(s,_start) = WSTK_CONCAT(s,_default_stack); \ - UWord* WSTK_CONCAT(s,_sp) = WSTK_CONCAT(s,_start); \ - UWord* WSTK_CONCAT(s,_end) = WSTK_CONCAT(s,_start) + DEF_WSTACK_SIZE; \ - ErtsAlcType_t WSTK_CONCAT(s,_alloc_type) = ERTS_ALC_T_ESTACK +void erl_grow_wstack(ErtsWStack*, Eterm* def_stack); +#define WSTK_CONCAT(a,b) a##b +#define WSTK_DEF_STACK(s) WSTK_CONCAT(s,_default_wstack) + +#define DECLARE_WSTACK(s) \ + UWord WSTK_DEF_STACK(s)[DEF_WSTACK_SIZE]; \ + ErtsWStack s = { \ + WSTK_DEF_STACK(s), /* wstart */ \ + WSTK_DEF_STACK(s), /* wsp */ \ + WSTK_DEF_STACK(s) + DEF_WSTACK_SIZE, /* wend */ \ + ERTS_ALC_T_ESTACK /* alloc_type */ \ + } #define WSTACK_CHANGE_ALLOCATOR(s,t) \ do { \ - if (WSTK_CONCAT(s,_start) != WSTK_CONCAT(s,_default_stack)) { \ + if (s.wstart != WSTK_DEF_STACK(s)) { \ erl_exit(1, "Internal error - trying to change allocator " \ "type of active wstack\n"); \ } \ - WSTK_CONCAT(s,_alloc_type) = (t); \ + s.alloc_type = (t); \ } while (0) -#define DESTROY_WSTACK(s) \ -do { \ - if (WSTK_CONCAT(s,_start) != WSTK_CONCAT(s,_default_stack)) { \ - erts_free(WSTK_CONCAT(s,_alloc_type), WSTK_CONCAT(s,_start)); \ - } \ +#define DESTROY_WSTACK(s) \ +do { \ + if (s.wstart != WSTK_DEF_STACK(s)) { \ + erts_free(s.alloc_type, s.wstart); \ + } \ } while(0) + /* - * Do not free the stack after this, it may have pointers into what - * was saved in 'v'. 'v' and 'vsize' are changed by this macro. If - * 'v' points to anything, it should have been allocated by a previous - * call to this macro. Be careful to set a correct allocator prior to - * saving. - * 'v' can be any lvalue pointer, it will point to an array of UWord - * after calling this macro. + * Do not free the stack after this, it may have pointers into what + * was saved in 'dst'. */ -#define WSTACK_SAVE(s,v,vsize) /* v and vsize are "name parameters" */ \ -do { \ - Uint _wsz = WSTACK_COUNT(s); \ - if (WSTK_CONCAT(s,_start) == WSTK_CONCAT(s,_default_stack)) { \ - if ((v) == NULL) { \ - (v) = erts_alloc(WSTK_CONCAT(s,_alloc_type), \ - DEF_WSTACK_SIZE * sizeof(UWord)); \ - } \ - memcpy((v),WSTK_CONCAT(s,_start),_wsz*sizeof(UWord)); \ - } else { \ - (v) = (void *) WSTK_CONCAT(s,_start); \ - } \ - (vsize) = _wsz; \ +#define WSTACK_SAVE(s,dst)\ +do {\ + if (s.wstart == WSTK_DEF_STACK(s)) {\ + UWord _wsz = WSTACK_COUNT(s);\ + (dst)->wstart = erts_alloc(s.alloc_type,\ + DEF_WSTACK_SIZE * sizeof(UWord));\ + memcpy((dst)->wstart, s.wstart,_wsz*sizeof(UWord));\ + (dst)->wsp = (dst)->wstart + _wsz;\ + (dst)->wend = (dst)->wstart + DEF_WSTACK_SIZE;\ + (dst)->alloc_type = s.alloc_type;\ + } else\ + *(dst) = s;\ } while (0) -/* - * Use on empty stack, only the allocator can be changed before this - * The vector parameter is reset to NULL if the vector is moved to stack, - * otherwise it's kept for reuse, so a saved and restored vector might - * need freeing using the correct allocator parameter. - * 'v' can be any lvalue pointer, it's cast to an (UWord *). +#define DESTROY_SAVED_WSTACK(wstack)\ +do {\ + if ((wstack)->wstart) {\ + erts_free((wstack)->alloc_type, (wstack)->wstart);\ + (wstack)->wstart = NULL;\ + }\ +} while(0) + +/* + * Use on empty stack, only the allocator can be changed before this. + * The src stack is reset to NULL. */ -#define WSTACK_RESTORE(s, v, vsize) /*v is a "name parameter"*/ \ -do { \ - if ((vsize) > DEF_WSTACK_SIZE) { \ - Uint _ca = DEF_WSTACK_SIZE; \ - while (_ca < (vsize)) \ - _ca = _ca * 2; \ - WSTK_CONCAT(s,_start) = (UWord *) (v); \ - WSTK_CONCAT(s,_end) = ((UWord *)(v)) + _ca; \ - WSTK_CONCAT(s,_sp) = WSTK_CONCAT(s,_start) + (vsize); \ - (v) = NULL; \ - } else { \ - memcpy(WSTK_CONCAT(s,_start),(v),(vsize)*sizeof(UWord));\ - WSTK_CONCAT(s,_sp) = WSTK_CONCAT(s,_start) + (vsize); \ - } \ - } while (0) +#define WSTACK_RESTORE(s, src) \ +do { \ + ASSERT(s.wstart == WSTK_DEF_STACK(s)); \ + s = *(src); /* struct copy */ \ + (src)->wstart = NULL; \ + ASSERT(s.wsp >= s.wstart); \ + ASSERT(s.wsp <= s.wend); \ +} while (0) -#define WSTACK_IS_STATIC(s) (WSTK_CONCAT(s,_start) == WSTK_CONCAT(s,_default_stack)) +#define WSTACK_IS_STATIC(s) (s.wstart == WSTK_DEF_STACK(s))) -#define WSTACK_PUSH(s, x) \ -do { \ - if (WSTK_CONCAT(s,_sp) == WSTK_CONCAT(s,_end)) { \ - erl_grow_wstack(WSTK_CONCAT(s,_alloc_type), &WSTK_CONCAT(s,_start), \ - &WSTK_CONCAT(s,_sp), &WSTK_CONCAT(s,_end)); \ - } \ - *WSTK_CONCAT(s,_sp)++ = (x); \ +#define WSTACK_PUSH(s, x) \ +do { \ + if (s.wsp == s.wend) { \ + erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + } \ + *s.wsp++ = (x); \ } while(0) -#define WSTACK_PUSH2(s, x, y) \ -do { \ - if (WSTK_CONCAT(s,_sp) > WSTK_CONCAT(s,_end) - 2) { \ - erl_grow_wstack(WSTK_CONCAT(s,_alloc_type), &WSTK_CONCAT(s,_start), \ - &WSTK_CONCAT(s,_sp), &WSTK_CONCAT(s,_end)); \ - } \ - *WSTK_CONCAT(s,_sp)++ = (x); \ - *WSTK_CONCAT(s,_sp)++ = (y); \ +#define WSTACK_PUSH2(s, x, y) \ +do { \ + if (s.wsp > s.wend - 2) { \ + erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + } \ + *s.wsp++ = (x); \ + *s.wsp++ = (y); \ } while(0) -#define WSTACK_PUSH3(s, x, y, z) \ -do { \ - if (WSTK_CONCAT(s,_sp) > WSTK_CONCAT(s,_end) - 3) { \ - erl_grow_wstack(WSTK_CONCAT(s,_alloc_type), &WSTK_CONCAT(s,_start), \ - &WSTK_CONCAT(s,_sp), &WSTK_CONCAT(s,_end)); \ - } \ - *WSTK_CONCAT(s,_sp)++ = (x); \ - *WSTK_CONCAT(s,_sp)++ = (y); \ - *WSTK_CONCAT(s,_sp)++ = (z); \ +#define WSTACK_PUSH3(s, x, y, z) \ +do { \ + if (s.wsp > s.wend - 3) { \ + erl_grow_wstack(&s, WSTK_DEF_STACK(s)); \ + } \ + *s.wsp++ = (x); \ + *s.wsp++ = (y); \ + *s.wsp++ = (z); \ } while(0) -#define WSTACK_COUNT(s) (WSTK_CONCAT(s,_sp) - WSTK_CONCAT(s,_start)) +#define WSTACK_COUNT(s) (s.wsp - s.wstart) +#define WSTACK_ISEMPTY(s) (s.wsp == s.wstart) +#define WSTACK_POP(s) (*(--s.wsp)) -#define WSTACK_ISEMPTY(s) (WSTK_CONCAT(s,_sp) == WSTK_CONCAT(s,_start)) -#define WSTACK_POP(s) (*(--WSTK_CONCAT(s,_sp))) /* binary.c */ @@ -655,6 +652,10 @@ Eterm erl_send(Process *p, Eterm to, Eterm msg); Eterm erl_is_function(Process* p, Eterm arg1, Eterm arg2); +/* beam_bif_load.c */ +Eterm erts_check_process_code(Process *c_p, Eterm module, int allow_gc, int *redsp); + + /* beam_load.c */ typedef struct { BeamInstr* current; /* Pointer to: Mod, Name, Arity */ @@ -852,6 +853,12 @@ Port *erts_get_heart_port(void); void erts_lcnt_enable_io_lock_count(int enable); #endif +/* driver_tab.c */ +typedef void *(*ErtsStaticNifInitFPtr)(void); +ErtsStaticNifInitFPtr erts_static_nif_get_nif_init(const char *name, int len); +int erts_is_static_nif(void *handle); +void erts_init_static_drivers(void); + /* erl_drv_thread.c */ void erl_drv_thr_init(void); @@ -911,6 +918,17 @@ char *erts_convert_filename_to_native(Eterm name, char *statbuf, ErtsAlcType_t alloc_type, int allow_empty, int allow_atom, Sint *used /* out */); +char *erts_convert_filename_to_encoding(Eterm name, char *statbuf, + size_t statbuf_size, + ErtsAlcType_t alloc_type, + int allow_empty, int allow_atom, + int encoding, + Sint *used /* out */, + Uint extra); +char* erts_convert_filename_to_wchar(byte* bytes, Uint size, + char *statbuf, size_t statbuf_size, + ErtsAlcType_t alloc_type, Sint* used, + Uint extra_wchars); Eterm erts_convert_native_to_filename(Process *p, byte *bytes); Eterm erts_utf8_to_list(Process *p, Uint num, byte *bytes, Uint sz, Uint left, Uint *num_built, Uint *num_eaten, Eterm tail); @@ -1110,7 +1128,12 @@ erts_alloc_message_heap_state(Uint size, if (statep) *statep = state; if ((state & (ERTS_PSFLG_EXITING|ERTS_PSFLG_PENDING_EXIT)) + || (receiver->flags & F_DISABLE_GC) || HEAP_LIMIT(receiver) - HEAP_TOP(receiver) <= size) { + /* + * The heap is either potentially in an inconsistent + * state, or not large enough. + */ #ifdef ERTS_SMP if (locked_main) { *receiver_locks &= ~ERTS_PROC_LOCK_MAIN; diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c index 9076bbe73c..49af86b36a 100644 --- a/erts/emulator/beam/io.c +++ b/erts/emulator/beam/io.c @@ -2765,6 +2765,7 @@ void erts_init_io(int port_tab_size, init_driver(&fd_driver, &fd_driver_entry, NULL); init_driver(&vanilla_driver, &vanilla_driver_entry, NULL); init_driver(&spawn_driver, &spawn_driver_entry, NULL); + erts_init_static_drivers(); for (dp = driver_tab; *dp != NULL; dp++) erts_add_driver_entry(*dp, NULL, 1); @@ -7060,7 +7061,7 @@ void *driver_dl_open(char * path) int res; int *last_error_p = erts_smp_tsd_get(driver_list_last_error_key); int locked = maybe_lock_driver_list(); - if ((res = erts_sys_ddll_open(path, &ptr)) == 0) { + if ((res = erts_sys_ddll_open(path, &ptr, NULL)) == 0) { maybe_unlock_driver_list(locked); return ptr; } else { diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab index 1e5ae46bfa..c29f3f9b1b 100644 --- a/erts/emulator/beam/ops.tab +++ b/erts/emulator/beam/ops.tab @@ -763,17 +763,17 @@ allocate_init t I y ################################################################# # -# The BIFs erlang:check_process_code/2 must be called like a function, +# The BIFs erts_internal:check_process_code/2 must be called like a function, # to ensure that c_p->i (program counter) is set correctly (an ordinary # BIF call doesn't set it). # -call_ext u==2 Bif=u$bif:erlang:check_process_code/2 => i_call_ext Bif -call_ext_last u==2 Bif=u$bif:erlang:check_process_code/2 D => i_call_ext_last Bif D -call_ext_only u==2 Bif=u$bif:erlang:check_process_code/2 => i_call_ext_only Bif +call_ext u==2 Bif=u$bif:erts_internal:check_process_code/2 => i_call_ext Bif +call_ext_last u==2 Bif=u$bif:erts_internal:check_process_code/2 D => i_call_ext_last Bif D +call_ext_only u==2 Bif=u$bif:erts_internal:check_process_code/2 => i_call_ext_only Bif # -# The BIFs erlang:garbage_collect/0,1 must be called like functions, +# The BIFs erlang:garbage_collect/0 must be called like a function, # to allow them to invoke the garbage collector. (The stack pointer must # be saved and p->arity must be zeroed, which is not done on ordinary BIF calls.) # @@ -782,10 +782,6 @@ call_ext u==0 Bif=u$bif:erlang:garbage_collect/0 => i_call_ext Bif call_ext_last u==0 Bif=u$bif:erlang:garbage_collect/0 D => i_call_ext_last Bif D call_ext_only u==0 Bif=u$bif:erlang:garbage_collect/0 => i_call_ext_only Bif -call_ext u==1 Bif=u$bif:erlang:garbage_collect/1 => i_call_ext Bif -call_ext_last u==1 Bif=u$bif:erlang:garbage_collect/1 D => i_call_ext_last Bif D -call_ext_only u==1 Bif=u$bif:erlang:garbage_collect/1 => i_call_ext_only Bif - # # put/2 and erase/1 must be able to do garbage collection, so we must call # them like functions. diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h index 31252ed78f..189d9ebac8 100644 --- a/erts/emulator/beam/sys.h +++ b/erts/emulator/beam/sys.h @@ -279,18 +279,21 @@ typedef unsigned long UWord; typedef long SWord; #define SWORD_CONSTANT(Const) Const##L #define UWORD_CONSTANT(Const) Const##UL +#define ERTS_UWORD_MAX ULONG_MAX #define ERTS_SWORD_MAX LONG_MAX #elif SIZEOF_VOID_P == SIZEOF_INT typedef unsigned int UWord; typedef int SWord; #define SWORD_CONSTANT(Const) Const #define UWORD_CONSTANT(Const) Const##U +#define ERTS_UWORD_MAX UINT_MAX #define ERTS_SWORD_MAX INT_MAX #elif SIZEOF_VOID_P == SIZEOF_LONG_LONG typedef unsigned long long UWord; typedef long long SWord; #define SWORD_CONSTANT(Const) Const##LL #define UWORD_CONSTANT(Const) Const##ULL +#define ERTS_UWORD_MAX ULLONG_MAX #define ERTS_SWORD_MAX LLONG_MAX #else #error Found no appropriate type to use for 'Eterm', 'Uint' and 'Sint' @@ -304,6 +307,7 @@ typedef unsigned long Uint; typedef long Sint; #define SWORD_CONSTANT(Const) Const##L #define UWORD_CONSTANT(Const) Const##UL +#define ERTS_UWORD_MAX ULONG_MAX #define ERTS_SWORD_MAX LONG_MAX #define ERTS_SIZEOF_ETERM SIZEOF_LONG #define ErtsStrToSint strtol @@ -313,6 +317,7 @@ typedef unsigned int Uint; typedef int Sint; #define SWORD_CONSTANT(Const) Const #define UWORD_CONSTANT(Const) Const##U +#define ERTS_UWORD_MAX UINT_MAX #define ERTS_SWORD_MAX INT_MAX #define ERTS_SIZEOF_ETERM SIZEOF_INT #define ErtsStrToSint strtol @@ -322,6 +327,7 @@ typedef unsigned long long Uint; typedef long long Sint; #define SWORD_CONSTANT(Const) Const##LL #define UWORD_CONSTANT(Const) Const##ULL +#define ERTS_UWORD_MAX ULLONG_MAX #define ERTS_SWORD_MAX LLONG_MAX #define ERTS_SIZEOF_ETERM SIZEOF_LONG_LONG #if defined(__WIN32__) @@ -661,8 +667,7 @@ typedef struct { #define ERTS_SYS_DDLL_ERROR_INIT {NULL} extern void erts_sys_ddll_free_error(ErtsSysDdllError*); extern void erl_sys_ddll_init(void); /* to initialize mutexes etc */ -extern int erts_sys_ddll_open2(const char *path, void **handle, ErtsSysDdllError*); -#define erts_sys_ddll_open(P,H) erts_sys_ddll_open2(P,H,NULL) +extern int erts_sys_ddll_open(const char *path, void **handle, ErtsSysDdllError*); extern int erts_sys_ddll_open_noext(char *path, void **handle, ErtsSysDdllError*); extern int erts_sys_ddll_load_driver_init(void *handle, void **function); extern int erts_sys_ddll_load_nif_init(void *handle, void **function,ErtsSysDdllError*); diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c index 0d75bbcc77..7f8bdcb2ca 100644 --- a/erts/emulator/beam/utils.c +++ b/erts/emulator/beam/utils.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 @@ -185,39 +185,41 @@ erts_set_hole_marker(Eterm* ptr, Uint sz) * Helper function for the ESTACK macros defined in global.h. */ void -erl_grow_stack(ErtsAlcType_t a_type, Eterm** start, Eterm** sp, Eterm** end) +erl_grow_estack(ErtsEStack* s, Eterm* default_estack) { - Uint old_size = (*end - *start); + Uint old_size = (s->end - s->start); Uint new_size = old_size * 2; - Uint sp_offs = *sp - *start; - if (new_size > 2 * DEF_ESTACK_SIZE) { - *start = erts_realloc(a_type, (void *) *start, new_size*sizeof(Eterm)); + Uint sp_offs = s->sp - s->start; + if (s->start != default_estack) { + s->start = erts_realloc(s->alloc_type, s->start, + new_size*sizeof(Eterm)); } else { - Eterm* new_ptr = erts_alloc(a_type, new_size*sizeof(Eterm)); - sys_memcpy(new_ptr, *start, old_size*sizeof(Eterm)); - *start = new_ptr; + Eterm* new_ptr = erts_alloc(s->alloc_type, new_size*sizeof(Eterm)); + sys_memcpy(new_ptr, s->start, old_size*sizeof(Eterm)); + s->start = new_ptr; } - *end = *start + new_size; - *sp = *start + sp_offs; + s->end = s->start + new_size; + s->sp = s->start + sp_offs; } /* - * Helper function for the ESTACK macros defined in global.h. + * Helper function for the WSTACK macros defined in global.h. */ void -erl_grow_wstack(ErtsAlcType_t a_type, UWord** start, UWord** sp, UWord** end) +erl_grow_wstack(ErtsWStack* s, UWord* default_wstack) { - Uint old_size = (*end - *start); + Uint old_size = (s->wend - s->wstart); Uint new_size = old_size * 2; - Uint sp_offs = *sp - *start; - if (new_size > 2 * DEF_ESTACK_SIZE) { - *start = erts_realloc(a_type, (void *) *start, new_size*sizeof(UWord)); + Uint sp_offs = s->wsp - s->wstart; + if (s->wstart != default_wstack) { + s->wstart = erts_realloc(s->alloc_type, s->wstart, + new_size*sizeof(UWord)); } else { - UWord* new_ptr = erts_alloc(a_type, new_size*sizeof(UWord)); - sys_memcpy(new_ptr, *start, old_size*sizeof(UWord)); - *start = new_ptr; + UWord* new_ptr = erts_alloc(s->alloc_type, new_size*sizeof(UWord)); + sys_memcpy(new_ptr, s->wstart, old_size*sizeof(UWord)); + s->wstart = new_ptr; } - *end = *start + new_size; - *sp = *start + sp_offs; + s->wend = s->wstart + new_size; + s->wsp = s->wstart + sp_offs; } /* CTYPE macros */ @@ -1675,7 +1677,7 @@ static int do_send_to_logger(Eterm tag, Eterm gleader, char *buf, int len) p = erts_whereis_process(NULL, 0, am_error_logger, 0, 0); if (p) { erts_aint32_t state = erts_smp_atomic32_read_acqb(&p->state); - if (state & ERTS_PSFLG_RUNNING) + if (state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS)) p = NULL; } } @@ -2846,7 +2848,7 @@ pop_next: return 0; not_equal: - DESTROY_ESTACK(stack); + DESTROY_WSTACK(stack); return j; #undef CMP_NODES @@ -3021,6 +3023,14 @@ buf_to_intlist(Eterm** hpp, const char *buf, size_t len, Eterm tail) ** Return remaining bytes in buffer on success ** ERTS_IOLIST_TO_BUF_OVERFLOW on overflow ** ERTS_IOLIST_TO_BUF_TYPE_ERROR on type error (including that result would not be a whole number of bytes) +** +** Note! +** Do not detect indata errors in this fiunction that are not detected by erts_iolist_size! +** +** A caller should be able to rely on a successful return from erts_iolist_to_buf +** if erts_iolist_size is previously successfully called and erts_iolist_to_buf +** is called with a buffer at least as large as the value given by erts_iolist_size. +** */ ErlDrvSizeT erts_iolist_to_buf(Eterm obj, char* buf, ErlDrvSizeT alloced_len) @@ -3127,6 +3137,11 @@ ErlDrvSizeT erts_iolist_to_buf(Eterm obj, char* buf, ErlDrvSizeT alloced_len) /* * Return 0 if successful, and non-zero if unsuccessful. + * + * It is vital that if erts_iolist_to_buf would return an error for + * any type of term data, this function should do so as well. + * Any input term error detected in erts_iolist_to_buf should also + * be detected in this function! */ int erts_iolist_size(Eterm obj, ErlDrvSizeT* sizep) { @@ -4006,7 +4021,6 @@ erts_smp_ensure_later_interval_acqb(erts_interval_t *icp, Uint64 ic) #endif } - /* * A millisecond timestamp without time correction where there's no hrtime * - for tracing on "long" things... |