diff options
Diffstat (limited to 'erts/emulator/beam')
61 files changed, 4229 insertions, 1526 deletions
diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names index d28e519ae1..5d06a32941 100644 --- a/erts/emulator/beam/atom.names +++ b/erts/emulator/beam/atom.names @@ -116,6 +116,7 @@ atom binary_longest_prefix_trap atom binary_longest_suffix_trap atom binary_match_trap atom binary_matches_trap +atom binary_to_list_continue atom binary_to_term_trap atom block atom blocked @@ -315,6 +316,7 @@ atom line_length atom linked_in_driver atom links atom list +atom list_to_binary_continue atom little atom loaded atom load_cancelled diff --git a/erts/emulator/beam/beam_bp.c b/erts/emulator/beam/beam_bp.c index 49a34ab4ad..4e711c89e0 100644 --- a/erts/emulator/beam/beam_bp.c +++ b/erts/emulator/beam/beam_bp.c @@ -642,7 +642,7 @@ erts_generic_breakpoint(Process* c_p, BeamInstr* I, Eterm* reg) erts_smp_atomic_inc_nob(&bp->count->acount); } - if (bp_flags & ERTS_BPF_TIME_TRACE_ACTIVE) { + if (bp_flags & ERTS_BPF_TIME_TRACE_ACTIVE && erts_is_tracer_proc_valid(c_p)) { Eterm w; erts_trace_time_call(c_p, I, bp->time); w = (BeamInstr) *c_p->cp; @@ -730,7 +730,8 @@ erts_bif_trace(int bif_index, Process* p, Eterm* args, BeamInstr* I) } } if (bp_flags & ERTS_BPF_TIME_TRACE_ACTIVE && - IS_TRACED_FL(p, F_TRACE_CALLS)) { + IS_TRACED_FL(p, F_TRACE_CALLS) && + erts_is_tracer_proc_valid(p)) { BeamInstr *pc = (BeamInstr *)ep->code+3; erts_trace_time_call(p, pc, bp->time); } diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index 0cec9ea3ec..8bfb7d2ad2 100644 --- a/erts/emulator/beam/beam_emu.c +++ b/erts/emulator/beam/beam_emu.c @@ -1301,7 +1301,7 @@ void process_main(void) (Eterm)fptr[1], (Uint)fptr[2], NULL, fun_buf); } else { - erts_snprintf(fun_buf, sizeof(fun_buf), + erts_snprintf(fun_buf, sizeof(DTRACE_CHARBUF_NAME(fun_buf)), "<unknown/%p>", next); } } @@ -3503,6 +3503,7 @@ get_map_elements_fail: * I[0]: &&call_nif * I[1]: Function pointer to NIF function * I[2]: Pointer to erl_module_nif + * I[3]: Function pointer to dirty NIF */ BifFunction vbf; @@ -3523,13 +3524,6 @@ get_map_elements_fail: reg[0] = r(0); nif_bif_result = (*fp)(&env, bif_nif_arity, reg); erts_post_nif(&env); -#ifdef ERTS_DIRTY_SCHEDULERS - if (is_non_value(nif_bif_result) && c_p->freason == TRAP) { - Export* ep = (Export*) c_p->psd->data[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT]; - ep->code[0] = I[-3]; - ep->code[1] = I[-2]; - } -#endif } ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(nif_bif_result)); PROCESS_MAIN_CHK_LOCKS(c_p); diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c index e96177cfd9..cfc6146b0a 100644 --- a/erts/emulator/beam/beam_load.c +++ b/erts/emulator/beam/beam_load.c @@ -2363,7 +2363,11 @@ load_code(LoaderState* stp) if (stp->may_load_nif) { const int finfo_ix = ci - FUNC_INFO_SZ; - enum { MIN_FUNC_SZ = 3 }; +#ifdef ERTS_DIRTY_SCHEDULERS + enum { MIN_FUNC_SZ = 4 }; +#else + enum { MIN_FUNC_SZ = 3 }; +#endif if (finfo_ix - last_func_start < MIN_FUNC_SZ && last_func_start) { /* Must make room for call_nif op */ int pad = MIN_FUNC_SZ - (finfo_ix - last_func_start); diff --git a/erts/emulator/beam/beam_ranges.c b/erts/emulator/beam/beam_ranges.c index 0f2d5d0c2a..cb6470638f 100644 --- a/erts/emulator/beam/beam_ranges.c +++ b/erts/emulator/beam/beam_ranges.c @@ -282,7 +282,7 @@ find_range(BeamInstr* pc) while (low < high) { if (pc < mid->start) { high = mid; - } else if (pc > RANGE_END(mid)) { + } else if (pc >= RANGE_END(mid)) { low = mid + 1; } else { erts_smp_atomic_set_nob(&r[active].mid, (erts_aint_t) mid); diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c index 06a1230ca0..42dd160e38 100644 --- a/erts/emulator/beam/bif.c +++ b/erts/emulator/beam/bif.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2013. All Rights Reserved. + * Copyright Ericsson AB 1996-2014. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -1869,6 +1869,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { } else if (is_external_pid(to)) { dep = external_pid_dist_entry(to); if(dep == erts_this_dist_entry) { +#if DEBUG erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); erts_dsprintf(dsbufp, "Discarding message %T from %T to %T in an old " @@ -1879,6 +1880,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { external_pid_creation(to), erts_this_node->creation); erts_send_error_to_logger(p->group_leader, dsbufp); +#endif return 0; } return remote_send(p, dep, to, to, msg, suspend); @@ -1886,8 +1888,13 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { Eterm id = erts_whereis_name_to_id(p, to); rp = erts_proc_lookup(id); - if (rp) + if (rp) { + if (IS_TRACED(p)) + trace_send(p, to, msg); + if (ERTS_PROC_GET_SAVED_CALLS_BUF(p)) + save_calls(p, &exp_send); goto send_message; + } pt = erts_port_lookup(id, (erts_port_synchronous_ops @@ -1907,6 +1914,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { } else if (is_external_port(to) && (external_port_dist_entry(to) == erts_this_dist_entry)) { +#if DEBUG erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); erts_dsprintf(dsbufp, "Discarding message %T from %T to %T in an old " @@ -1917,6 +1925,7 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend, Eterm *refp) { external_port_creation(to), erts_this_node->creation); erts_send_error_to_logger(p->group_leader, dsbufp); +#endif return 0; } else if (is_internal_port(to)) { int ret_val; @@ -2763,6 +2772,7 @@ static int do_list_to_integer(Process *p, Eterm orig_list, Eterm *integer, Eterm *rest) { Sint i = 0; + Uint ui = 0; int skip = 0; int neg = 0; int n = 0; @@ -2816,8 +2826,8 @@ static int do_list_to_integer(Process *p, Eterm orig_list, unsigned_val(CAR(list_val(lst))) > '9') { break; } - i = i * 10; - i = i + unsigned_val(CAR(list_val(lst))) - '0'; + ui = ui * 10; + ui = ui + unsigned_val(CAR(list_val(lst))) - '0'; n++; lst = CDR(list_val(lst)); if (is_nil(lst)) { @@ -2841,7 +2851,8 @@ static int do_list_to_integer(Process *p, Eterm orig_list, */ if (n <= SMALL_DIGITS) { /* It must be small */ - if (neg) i = -i; + if (neg) i = -(Sint)ui; + else i = (Sint)ui; res = make_small(i); } else { lg2 = (n+1)*230/69+1; @@ -2882,9 +2893,6 @@ static int do_list_to_integer(Process *p, Eterm orig_list, res = big_plus_small(res, m, hp); } - if (is_big(res)) /* check if small */ - res = big_plus_small(res, 0, hp); /* includes conversion to small */ - if (neg) { if (is_small(res)) res = make_small(-signed_val(res)); @@ -2894,8 +2902,12 @@ static int do_list_to_integer(Process *p, Eterm orig_list, } } - if (is_big(res)) { - hp += (big_arity(res)+1); + if (is_not_small(res)) { + res = big_plus_small(res, 0, hp); /* includes conversion to small */ + + if (is_not_small(res)) { + hp += (big_arity(res)+1); + } } HRelease(p,hp_end,hp); } diff --git a/erts/emulator/beam/bif.h b/erts/emulator/beam/bif.h index 51b77a95ed..72c55ccb55 100644 --- a/erts/emulator/beam/bif.h +++ b/erts/emulator/beam/bif.h @@ -124,12 +124,85 @@ do { \ return THE_NON_VALUE; \ } while(0) +#define ERTS_BIF_ERROR_TRAPPED0(Proc, Reason, Bif) \ +do { \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + return THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_ERROR_TRAPPED1(Proc, Reason, Bif, A0) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + return THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_ERROR_TRAPPED2(Proc, Reason, Bif, A0, A1) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + reg[1] = (Eterm) (A1); \ + return THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_ERROR_TRAPPED3(Proc, Reason, Bif, A0, A1, A2) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + reg[1] = (Eterm) (A1); \ + reg[2] = (Eterm) (A2); \ + return THE_NON_VALUE; \ +} while (0) + #define ERTS_BIF_PREP_ERROR(Ret, Proc, Reason) \ do { \ (Proc)->freason = (Reason); \ (Ret) = THE_NON_VALUE; \ } while (0) +#define ERTS_BIF_PREP_ERROR_TRAPPED0(Ret, Proc, Reason, Bif) \ +do { \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + (Ret) = THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_PREP_ERROR_TRAPPED1(Ret, Proc, Reason, Bif, A0) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + (Ret) = THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_PREP_ERROR_TRAPPED2(Ret, Proc, Reason, Bif, A0, A1) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + reg[1] = (Eterm) (A1); \ + (Ret) = THE_NON_VALUE; \ +} while (0) + +#define ERTS_BIF_PREP_ERROR_TRAPPED3(Ret, Proc, Reason, Bif, A0, A1, A2) \ +do { \ + Eterm* reg = ERTS_PROC_GET_SCHDATA((Proc))->x_reg_array; \ + (Proc)->freason = (Reason); \ + (Proc)->current = (Bif)->code; \ + reg[0] = (Eterm) (A0); \ + reg[1] = (Eterm) (A1); \ + reg[2] = (Eterm) (A2); \ + (Ret) = THE_NON_VALUE; \ +} while (0) #define ERTS_BIF_PREP_TRAP0(Ret, Trap, Proc) \ do { \ @@ -392,6 +465,51 @@ erts_bif_prep_await_proc_exit_apply_trap(Process *c_p, Eterm args[], int nargs); +#ifndef HIPE + +#define HIPE_WRAPPER_BIF_DISABLE_GC(BIF_NAME, ARITY) + +#else + +#include "erl_fun.h" +#include "hipe_mode_switch.h" + +/* + * Hipe wrappers used by native code for BIFs that disable GC while trapping. + * Also add usage of the wrapper in ../hipe/hipe_bif_list.m4 + * + * Problem: + * When native code calls a BIF that traps, hipe_mode_switch will push a + * "trap frame" on the Erlang stack in order to find its way back from beam_emu + * back to native caller when finally done. If GC is disabled and stack/heap + * is full there is no place to push the "trap frame". + * + * Solution: + * We reserve space on stack for the "trap frame" here before the BIF is called. + * If the BIF does not trap, the space is reclaimed here before returning. + * If the BIF traps, hipe_push_beam_trap_frame() will detect that a "trap frame" + * already is reserved and use it. + */ + + +#define HIPE_WRAPPER_BIF_DISABLE_GC(BIF_NAME, ARITY) \ +BIF_RETTYPE hipe_wrapper_ ## BIF_NAME ## _ ## ARITY (Process* c_p, \ + Eterm* args); \ +BIF_RETTYPE hipe_wrapper_ ## BIF_NAME ## _ ## ARITY (Process* c_p, \ + Eterm* args) \ +{ \ + BIF_RETTYPE res; \ + hipe_reserve_beam_trap_frame(c_p, args, ARITY); \ + res = BIF_NAME ## _ ## ARITY (c_p, args); \ + if (is_value(res) || c_p->freason != TRAP) { \ + hipe_unreserve_beam_trap_frame(c_p); \ + } \ + return res; \ +} + +#endif + + #include "erl_bif_table.h" #endif diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab index 2d888862bf..e68b8e6274 100644 --- a/erts/emulator/beam/bif.tab +++ b/erts/emulator/beam/bif.tab @@ -45,6 +45,7 @@ bif erlang:apply/3 bif erlang:atom_to_list/1 bif erlang:binary_to_list/1 bif erlang:binary_to_list/3 +bif erlang:binary_to_term/1 bif erlang:crc32/1 bif erlang:crc32/2 bif erlang:crc32_combine/3 @@ -151,12 +152,11 @@ bif erts_internal:port_command/3 bif erts_internal:port_control/3 bif erts_internal:port_close/1 bif erts_internal:port_connect/2 -bif erts_internal:binary_to_term/1 -bif erts_internal:binary_to_term/2 bif erts_internal:request_system_task/3 bif erts_internal:check_process_code/2 +bif erts_internal:map_to_tuple_keys/1 # inet_db support bif erlang:port_set_data/2 @@ -480,6 +480,11 @@ bif erlang:call_on_load_function/1 bif erlang:finish_after_on_load/2 # +# New Bifs in R13B04 +# +bif erlang:binary_to_term/2 + +# # The binary match bifs (New in R14A - EEP9) # @@ -596,6 +601,10 @@ bif maps:values/1 bif erts_internal:cmp_term/2 # +# New in 17.1. +# +bif erlang:fun_info_mfa/1 +# # Obsolete # diff --git a/erts/emulator/beam/big.c b/erts/emulator/beam/big.c index 41a041eba6..de7d370938 100644 --- a/erts/emulator/beam/big.c +++ b/erts/emulator/beam/big.c @@ -274,7 +274,9 @@ _b = _b << _s; \ _vn1 = _b >> H_EXP; \ _vn0 = _b & LO_MASK; \ - _un32 = (_a1 << _s) | ((_a0>>(D_EXP-_s)) & (-_s >> (D_EXP-1))); \ + /* If needed to avoid undefined behaviour */ \ + if (_s) _un32 = (_a1 << _s) | ((_a0>>(D_EXP-_s)) & (-_s >> (D_EXP-1))); \ + else _un32 = _a1; \ _un10 = _a0 << _s; \ _un1 = _un10 >> H_EXP; \ _un0 = _un10 & LO_MASK; \ @@ -1506,13 +1508,15 @@ Eterm uword_to_big(UWord x, Eterm *y) */ Eterm small_to_big(Sint x, Eterm *y) { + Uint xu; if (x >= 0) { + xu = x; *y = make_pos_bignum_header(1); } else { - x = -x; + xu = -(Uint)x; *y = make_neg_bignum_header(1); } - BIG_DIGIT(y, 0) = x; + BIG_DIGIT(y, 0) = xu; return make_big(y); } @@ -1540,21 +1544,24 @@ Eterm erts_uint64_to_big(Uint64 x, Eterm **hpp) Eterm erts_sint64_to_big(Sint64 x, Eterm **hpp) { Eterm *hp = *hpp; + Uint64 ux; int neg; - if (x >= 0) + if (x >= 0) { neg = 0; + ux = x; + } else { neg = 1; - x = -x; + ux = -(Uint64)x; } #if defined(ARCH_32) || HALFWORD_HEAP - if (x >= (((Uint64) 1) << 32)) { + if (ux >= (((Uint64) 1) << 32)) { if (neg) *hp = make_neg_bignum_header(2); else *hp = make_pos_bignum_header(2); - BIG_DIGIT(hp, 0) = (Uint) (x & ((Uint) 0xffffffff)); - BIG_DIGIT(hp, 1) = (Uint) ((x >> 32) & ((Uint) 0xffffffff)); + BIG_DIGIT(hp, 0) = (Uint) (ux & ((Uint) 0xffffffff)); + BIG_DIGIT(hp, 1) = (Uint) ((ux >> 32) & ((Uint) 0xffffffff)); *hpp += 3; } else @@ -1564,7 +1571,7 @@ Eterm erts_sint64_to_big(Sint64 x, Eterm **hpp) *hp = make_neg_bignum_header(1); else *hp = make_pos_bignum_header(1); - BIG_DIGIT(hp, 0) = (Uint) x; + BIG_DIGIT(hp, 0) = (Uint) ux; *hpp += 2; } return make_big(hp); @@ -2667,9 +2674,6 @@ Eterm erts_chars_to_integer(Process *BIF_P, char *bytes, res = big_plus_small(res, m, hp); } - if (is_big(res)) /* check if small */ - res = big_plus_small(res, 0, hp); /* includes conversion to small */ - if (neg) { if (is_small(res)) res = make_small(-signed_val(res)); @@ -2679,8 +2683,12 @@ Eterm erts_chars_to_integer(Process *BIF_P, char *bytes, } } - if (is_big(res)) { - hp += (big_arity(res) + 1); + if (is_not_small(res)) { + res = big_plus_small(res, 0, hp); /* includes conversion to small */ + + if (is_not_small(res)) { + hp += (big_arity(res) + 1); + } } HRelease(BIF_P, hp_end, hp); goto bytebuf_to_integer_1_done; diff --git a/erts/emulator/beam/big.h b/erts/emulator/beam/big.h index d80111822e..da31876d75 100644 --- a/erts/emulator/beam/big.h +++ b/erts/emulator/beam/big.h @@ -101,7 +101,7 @@ typedef Uint dsize_t; /* Vector size type */ #define ERTS_SINT64_HEAP_SIZE(X) \ (IS_SSMALL((X)) \ ? 0 \ - : ERTS_UINT64_BIG_HEAP_SIZE__((X) >= 0 ? (X) : -(X))) + : ERTS_UINT64_BIG_HEAP_SIZE__((X) >= 0 ? (X) : -(Uint64)(X))) #define ERTS_UINT64_HEAP_SIZE(X) \ (IS_USMALL(0, (X)) ? 0 : ERTS_UINT64_BIG_HEAP_SIZE__((X))) diff --git a/erts/emulator/beam/binary.c b/erts/emulator/beam/binary.c index c7926f18af..f50d484576 100644 --- a/erts/emulator/beam/binary.c +++ b/erts/emulator/beam/binary.c @@ -31,12 +31,11 @@ #include "erl_binary.h" #include "erl_bits.h" -#ifdef DEBUG -static int list_to_bitstr_buf(Eterm obj, char* buf, Uint len); -#else -static int list_to_bitstr_buf(Eterm obj, char* buf); -#endif -static int bitstr_list_len(Eterm obj, Uint* num_bytes); +static Export binary_to_list_continue_export; +static Export list_to_binary_continue_export; + +static BIF_RETTYPE binary_to_list_continue(BIF_ALIST_1); +static BIF_RETTYPE list_to_binary_continue(BIF_ALIST_1); void erts_init_binary(void) @@ -49,6 +48,15 @@ erts_init_binary(void) "Internal error: Address of orig_bytes[0] of a Binary" " is *not* 8-byte aligned\n"); } + + erts_init_trap_export(&binary_to_list_continue_export, + am_erts_internal, am_binary_to_list_continue, 1, + &binary_to_list_continue); + + erts_init_trap_export(&list_to_binary_continue_export, + am_erts_internal, am_list_to_binary_continue, 1, + &list_to_binary_continue); + } /* @@ -333,6 +341,132 @@ BIF_RETTYPE integer_to_binary_1(BIF_ALIST_1) BIF_RET(res); } +#define ERTS_B2L_BYTES_PER_REDUCTION 256 + +typedef struct { + Eterm res; + Eterm *hp; +#ifdef DEBUG + Eterm *hp_end; +#endif + byte *bytes; + Uint size; + Uint bitoffs; +} ErtsB2LState; + +static void b2l_state_destructor(Binary *mbp) +{ + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(mbp) == b2l_state_destructor); +} + +static BIF_RETTYPE +binary_to_list_chunk(Process *c_p, + Eterm mb_eterm, + ErtsB2LState* sp, + int reds_left, + int gc_disabled) +{ + BIF_RETTYPE ret; + int bump_reds; + Uint size; + byte *bytes; + + size = (reds_left + 1)*ERTS_B2L_BYTES_PER_REDUCTION; + if (size > sp->size) + size = sp->size; + bytes = sp->bytes + (sp->size - size); + + bump_reds = (size - 1)/ERTS_B2L_BYTES_PER_REDUCTION + 1; + BUMP_REDS(c_p, bump_reds); + + ASSERT(is_list(sp->res) || is_nil(sp->res)); + + sp->res = erts_bin_bytes_to_list(sp->res, + sp->hp, + bytes, + size, + sp->bitoffs); + sp->size -= size; + sp->hp += 2*size; + + if (sp->size > 0) { + + if (!gc_disabled) + erts_set_gc_state(c_p, 0); + + ASSERT(c_p->flags & F_DISABLE_GC); + ASSERT(is_value(mb_eterm)); + ERTS_BIF_PREP_TRAP1(ret, + &binary_to_list_continue_export, + c_p, + mb_eterm); + } + else { + + ASSERT(sp->hp == sp->hp_end); + ASSERT(sp->size == 0); + + if (!gc_disabled || !erts_set_gc_state(c_p, 1)) + ERTS_BIF_PREP_RET(ret, sp->res); + else + ERTS_BIF_PREP_YIELD_RETURN(ret, c_p, sp->res); + ASSERT(!(c_p->flags & F_DISABLE_GC)); + } + + return ret; +} + +static ERTS_INLINE BIF_RETTYPE +binary_to_list(Process *c_p, Eterm *hp, Eterm tail, byte *bytes, Uint size, Uint bitoffs) +{ + int reds_left = ERTS_BIF_REDS_LEFT(c_p); + if (size < reds_left*ERTS_B2L_BYTES_PER_REDUCTION) { + Eterm res; + BIF_RETTYPE ret; + int bump_reds = (size - 1)/ERTS_B2L_BYTES_PER_REDUCTION + 1; + BUMP_REDS(c_p, bump_reds); + res = erts_bin_bytes_to_list(tail, hp, bytes, size, bitoffs); + ERTS_BIF_PREP_RET(ret, res); + return ret; + } + else { + Binary *mbp = erts_create_magic_binary(sizeof(ErtsB2LState), + b2l_state_destructor); + ErtsB2LState *sp = ERTS_MAGIC_BIN_DATA(mbp); + Eterm mb; + + sp->res = tail; + sp->hp = hp; +#ifdef DEBUG + sp->hp_end = sp->hp + 2*size; +#endif + sp->bytes = bytes; + sp->size = size; + sp->bitoffs = bitoffs; + + hp = HAlloc(c_p, PROC_BIN_SIZE); + mb = erts_mk_magic_binary_term(&hp, &MSO(c_p), mbp); + return binary_to_list_chunk(c_p, mb, sp, reds_left, 0); + } +} + +static BIF_RETTYPE binary_to_list_continue(BIF_ALIST_1) +{ + Binary *mbp = ((ProcBin *) binary_val(BIF_ARG_1))->val; + + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(mbp) == b2l_state_destructor); + + ASSERT(BIF_P->flags & F_DISABLE_GC); + + return binary_to_list_chunk(BIF_P, + BIF_ARG_1, + (ErtsB2LState*) ERTS_MAGIC_BIN_DATA(mbp), + ERTS_BIF_REDS_LEFT(BIF_P), + 1); +} + +HIPE_WRAPPER_BIF_DISABLE_GC(binary_to_list, 1) + BIF_RETTYPE binary_to_list_1(BIF_ALIST_1) { Eterm real_bin; @@ -354,14 +488,15 @@ BIF_RETTYPE binary_to_list_1(BIF_ALIST_1) } else { Eterm* hp = HAlloc(BIF_P, 2 * size); byte* bytes = binary_bytes(real_bin)+offset; - - BIF_RET(erts_bin_bytes_to_list(NIL, hp, bytes, size, bitoffs)); + return binary_to_list(BIF_P, hp, NIL, bytes, size, bitoffs); } error: BIF_ERROR(BIF_P, BADARG); } +HIPE_WRAPPER_BIF_DISABLE_GC(binary_to_list, 3) + BIF_RETTYPE binary_to_list_3(BIF_ALIST_3) { byte* bytes; @@ -387,12 +522,13 @@ BIF_RETTYPE binary_to_list_3(BIF_ALIST_3) } i = stop-start+1; hp = HAlloc(BIF_P, 2*i); - BIF_RET(erts_bin_bytes_to_list(NIL, hp, bytes+start-1, i, bitoffs)); - + return binary_to_list(BIF_P, hp, NIL, bytes+start-1, i, bitoffs); error: BIF_ERROR(BIF_P, BADARG); } +HIPE_WRAPPER_BIF_DISABLE_GC(bitstring_to_list, 1) + BIF_RETTYPE bitstring_to_list_1(BIF_ALIST_1) { Eterm real_bin; @@ -431,124 +567,441 @@ BIF_RETTYPE bitstring_to_list_1(BIF_ALIST_1) previous = CONS(hp, make_binary(last), previous); hp += 2; } - BIF_RET(erts_bin_bytes_to_list(previous, hp, bytes, size, bitoffs)); + + return binary_to_list(BIF_P, hp, previous, bytes, size, bitoffs); } /* Turn a possibly deep list of ints (and binaries) into */ /* One large binary object */ -/* - * This bif also exists in the binary module, under the name - * binary:list_to_bin/1, why it's divided into interface and - * implementation. Also the backend for iolist_to_binary_1. - */ +typedef enum { + ERTS_L2B_OK, + ERTS_L2B_YIELD, + ERTS_L2B_TYPE_ERROR, + ERTS_L2B_OVERFLOW_ERROR +} ErtsL2BResult; -BIF_RETTYPE erts_list_to_binary_bif(Process *p, Eterm arg) -{ +#define ERTS_L2B_STATE_INITER(C_P, ARG, BIF, SZFunc, TBufFunc) \ + {ERTS_IOLIST2BUF_STATE_INITER((C_P), (ARG)), \ + (ARG), THE_NON_VALUE, (BIF), (SZFunc), (TBufFunc)} + +#define ERTS_L2B_STATE_MOVE(TO, FROM) \ + sys_memcpy((void *) (TO), (void *) (FROM), sizeof(ErtsL2BState)) + +typedef struct ErtsL2BState_ ErtsL2BState; + +struct ErtsL2BState_ { + ErtsIOList2BufState buf; + Eterm arg; Eterm bin; - Eterm h,t; - ErlDrvSizeT size; - byte* bytes; -#ifdef DEBUG - ErlDrvSizeT offset; -#endif + Export *bif; + int (*iolist_to_buf_size)(ErtsIOListState *); + ErlDrvSizeT (*iolist_to_buf)(ErtsIOList2BufState *); +}; + +static ERTS_INLINE ErtsL2BResult +list_to_binary_engine(ErtsL2BState *sp) +{ + ErlDrvSizeT res; + Process *c_p = sp->buf.iolist.c_p; + + /* + * have_size == 0 while sp->iolist_to_buf_size() + * has not finished the calculation. + */ + + if (!sp->buf.iolist.have_size) { + switch (sp->iolist_to_buf_size(&sp->buf.iolist)) { + case ERTS_IOLIST_YIELD: + return ERTS_L2B_YIELD; + case ERTS_IOLIST_OVERFLOW: + return ERTS_L2B_OVERFLOW_ERROR; + case ERTS_IOLIST_TYPE: + return ERTS_L2B_TYPE_ERROR; + case ERTS_IOLIST_OK: + break; + default: + ASSERT(0); + break; + } + + ASSERT(sp->buf.iolist.have_size); + + /* + * Size calculated... Setup state for + * sp->iolist_to_buf_*() + */ + + sp->bin = new_binary(c_p, + (byte *) NULL, + sp->buf.iolist.size); + + if (sp->buf.iolist.size == 0) + return ERTS_L2B_OK; + + sp->buf.buf = (char *) binary_bytes(sp->bin); + sp->buf.len = sp->buf.iolist.size; + sp->buf.iolist.obj = sp->arg; - if (is_nil(arg)) { - BIF_RET(new_binary(p,(byte*)"",0)); + if (sp->buf.iolist.reds_left <= 0) { + BUMP_ALL_REDS(c_p); + return ERTS_L2B_YIELD; + } } - if (is_not_list(arg)) { - goto error; + + ASSERT(sp->buf.iolist.size != 0); + ASSERT(is_value(sp->bin)); + ASSERT(sp->buf.buf); + + res = sp->iolist_to_buf(&sp->buf); + + if (!ERTS_IOLIST_TO_BUF_FAILED(res)) { + ASSERT(res == 0); + return ERTS_L2B_OK; } - /* check for [binary()] case */ - h = CAR(list_val(arg)); - t = CDR(list_val(arg)); - if (is_binary(h) && is_nil(t) && !( - HEADER_SUB_BIN == *(binary_val(h)) && ( - ((ErlSubBin *)binary_val(h))->bitoffs != 0 || - ((ErlSubBin *)binary_val(h))->bitsize != 0 - ))) { - return h; - } - switch (erts_iolist_size(arg, &size)) { - case ERTS_IOLIST_OVERFLOW: BIF_ERROR(p, SYSTEM_LIMIT); - case ERTS_IOLIST_TYPE: goto error; - default: ; - } - bin = new_binary(p, (byte *)NULL, size); - bytes = binary_bytes(bin); -#ifdef DEBUG - offset = -#endif - erts_iolist_to_buf(arg, (char*) bytes, size); - ASSERT(offset == 0); - BIF_RET(bin); + switch (res) { + case ERTS_IOLIST_TO_BUF_YIELD: + return ERTS_L2B_YIELD; + case ERTS_IOLIST_TO_BUF_OVERFLOW: + return ERTS_L2B_OVERFLOW_ERROR; + case ERTS_IOLIST_TO_BUF_TYPE_ERROR: + return ERTS_L2B_TYPE_ERROR; + default: + ERTS_INTERNAL_ERROR("Invalid return value from iolist_to_buf_yielding()"); + return ERTS_L2B_TYPE_ERROR; + } +} + +static void +l2b_state_destructor(Binary *mbp) +{ + ErtsL2BState *sp = ERTS_MAGIC_BIN_DATA(mbp); + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(mbp) == l2b_state_destructor); + DESTROY_SAVED_ESTACK(&sp->buf.iolist.estack); +} + +static ERTS_INLINE Eterm +l2b_final_touch(Process *c_p, ErtsL2BState *sp) +{ + Eterm *hp; + ErlSubBin* sbin; + if (sp->buf.offset == 0) + return sp->bin; + + hp = HAlloc(c_p, ERL_SUB_BIN_SIZE); + ASSERT(sp->buf.offset > 0); + sbin = (ErlSubBin *) hp; + sbin->thing_word = HEADER_SUB_BIN; + sbin->size = sp->buf.iolist.size-1; + sbin->offs = 0; + sbin->orig = sp->bin; + sbin->bitoffs = 0; + sbin->bitsize = sp->buf.offset; + sbin->is_writable = 0; + return make_binary(sbin); +} + +static BIF_RETTYPE +list_to_binary_chunk(Eterm mb_eterm, + ErtsL2BState* sp, + int reds_left, + int gc_disabled) +{ + Eterm err = BADARG; + BIF_RETTYPE ret; + Process *c_p = sp->buf.iolist.c_p; + + sp->buf.iolist.reds_left = reds_left; - error: - BIF_ERROR(p, BADARG); + switch (list_to_binary_engine(sp)) { + + case ERTS_L2B_OK: { + Eterm result = l2b_final_touch(c_p, sp); + if (!gc_disabled || !erts_set_gc_state(c_p, 1)) + ERTS_BIF_PREP_RET(ret, result); + else + ERTS_BIF_PREP_YIELD_RETURN(ret, c_p, result); + ASSERT(!(c_p->flags & F_DISABLE_GC)); + break; + } + case ERTS_L2B_YIELD: + if (!gc_disabled) { + /* first yield... */ + Eterm *hp; + Binary *mbp = erts_create_magic_binary(sizeof(ErtsL2BState), + l2b_state_destructor); + ErtsL2BState *new_sp = ERTS_MAGIC_BIN_DATA(mbp); + + ERTS_L2B_STATE_MOVE(new_sp, sp); + sp = new_sp; + + hp = HAlloc(c_p, PROC_BIN_SIZE); + mb_eterm = erts_mk_magic_binary_term(&hp, &MSO(c_p), mbp); + + ASSERT(is_value(mb_eterm)); + + erts_set_gc_state(c_p, 0); + } + + ASSERT(c_p->flags & F_DISABLE_GC); + + ERTS_BIF_PREP_TRAP1(ret, + &list_to_binary_continue_export, + c_p, + mb_eterm); + break; + + case ERTS_L2B_OVERFLOW_ERROR: + err = SYSTEM_LIMIT; + /* fall through */ + + case ERTS_L2B_TYPE_ERROR: + if (!gc_disabled) + ERTS_BIF_PREP_ERROR(ret, c_p, err); + else { + if (erts_set_gc_state(c_p, 1)) + ERTS_VBUMP_ALL_REDS(c_p); + + ERTS_BIF_PREP_ERROR_TRAPPED1(ret, + c_p, + err, + sp->bif, + sp->arg); + } + + ASSERT(!(c_p->flags & F_DISABLE_GC)); + break; + + default: + ERTS_INTERNAL_ERROR("Invalid return value from list_to_binary_engine()"); + ERTS_BIF_PREP_ERROR(ret,c_p, EXC_INTERNAL_ERROR); + break; + } + return ret; } +static BIF_RETTYPE list_to_binary_continue(BIF_ALIST_1) +{ + Binary *mbp = ((ProcBin *) binary_val(BIF_ARG_1))->val; + ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(mbp) == l2b_state_destructor); + + ASSERT(BIF_P->flags & F_DISABLE_GC); + + return list_to_binary_chunk(BIF_ARG_1, + ERTS_MAGIC_BIN_DATA(mbp), + ERTS_BIF_REDS_LEFT(BIF_P), + 1); +} + +BIF_RETTYPE erts_list_to_binary_bif(Process *c_p, Eterm arg, Export *bif) +{ + BIF_RETTYPE ret; + + if (is_nil(arg)) + ERTS_BIF_PREP_RET(ret, new_binary(c_p, (byte *) "", 0)); + else if (is_not_list(arg)) + ERTS_BIF_PREP_ERROR(ret, c_p, BADARG); + else { + /* check for [binary()] case */ + Eterm h = CAR(list_val(arg)); + Eterm t = CDR(list_val(arg)); + if (is_binary(h) + && is_nil(t) + && !(HEADER_SUB_BIN == *(binary_val(h)) + && (((ErlSubBin *)binary_val(h))->bitoffs != 0 + || ((ErlSubBin *)binary_val(h))->bitsize != 0))) { + ERTS_BIF_PREP_RET(ret, h); + } + else { + ErtsL2BState state = ERTS_L2B_STATE_INITER(c_p, + arg, + bif, + erts_iolist_size_yielding, + erts_iolist_to_buf_yielding); + int orig_reds_left = ERTS_BIF_REDS_LEFT(c_p); + + /* + * First try to do it all at once without having to use + * yielding iolist_to_buf(). + */ + state.buf.iolist.reds_left = orig_reds_left; + switch (erts_iolist_size_yielding(&state.buf.iolist)) { + case ERTS_IOLIST_OK: { + ErlDrvSizeT size = state.buf.iolist.size; + Eterm bin; + char *buf; + + if (size == 0) { + ERTS_BIF_PREP_RET(ret, new_binary(c_p, (byte *) NULL, 0)); + break; /* done */ + } + + bin = new_binary(c_p, (byte *) NULL, size); + buf = (char *) binary_bytes(bin); + + if (size < ERTS_IOLIST_TO_BUF_BYTES_PER_RED*CONTEXT_REDS) { + /* An (over) estimation of reductions needed */ + int reds_left = state.buf.iolist.reds_left; + int to_buf_reds = orig_reds_left - reds_left; + to_buf_reds += size/ERTS_IOLIST_TO_BUF_BYTES_PER_RED; + if (to_buf_reds <= reds_left) { + ErlDrvSizeT res; + + res = erts_iolist_to_buf(arg, buf, size); + if (res == 0) { + BUMP_REDS(c_p, to_buf_reds); + ERTS_BIF_PREP_RET(ret, bin); + break; /* done */ + } + if (!ERTS_IOLIST_TO_BUF_FAILED(res)) + ERTS_INTERNAL_ERROR("iolist_size/iolist_to_buf missmatch"); + if (res == ERTS_IOLIST_TO_BUF_OVERFLOW) + goto overflow; + goto type_error; + } + } + /* + * Since size has been computed list_to_binary_chunk() expects + * state prepared for iolist_to_buf. + */ + state.bin = bin; + state.buf.buf = buf; + state.buf.len = size; + state.buf.iolist.obj = arg; + /* Fall through... */ + } + case ERTS_IOLIST_YIELD: + ret = list_to_binary_chunk(THE_NON_VALUE, + &state, + state.buf.iolist.reds_left, + 0); + break; + case ERTS_IOLIST_OVERFLOW: + overflow: + ERTS_BIF_PREP_ERROR(ret, c_p, SYSTEM_LIMIT); + break; + case ERTS_IOLIST_TYPE: + type_error: + default: + ERTS_BIF_PREP_ERROR(ret, c_p, BADARG); + break; + } + } + } + return ret; +} + +HIPE_WRAPPER_BIF_DISABLE_GC(list_to_binary, 1) + BIF_RETTYPE list_to_binary_1(BIF_ALIST_1) { - return erts_list_to_binary_bif(BIF_P, BIF_ARG_1); + return erts_list_to_binary_bif(BIF_P, BIF_ARG_1, bif_export[BIF_list_to_binary_1]); } -/* Turn a possibly deep list of ints (and binaries) into */ -/* One large binary object */ +HIPE_WRAPPER_BIF_DISABLE_GC(iolist_to_binary, 1) BIF_RETTYPE iolist_to_binary_1(BIF_ALIST_1) { if (is_binary(BIF_ARG_1)) { BIF_RET(BIF_ARG_1); } - return erts_list_to_binary_bif(BIF_P, BIF_ARG_1); + return erts_list_to_binary_bif(BIF_P, BIF_ARG_1, bif_export[BIF_iolist_to_binary_1]); } +static int bitstr_list_len(ErtsIOListState *); +static ErlDrvSizeT list_to_bitstr_buf_yielding(ErtsIOList2BufState *); +static ErlDrvSizeT list_to_bitstr_buf_not_yielding(ErtsIOList2BufState *); + +HIPE_WRAPPER_BIF_DISABLE_GC(list_to_bitstring, 1) + BIF_RETTYPE list_to_bitstring_1(BIF_ALIST_1) { - Eterm bin; - Uint sz; - int offset; - byte* bytes; - ErlSubBin* sb1; - Eterm* hp; - - if (is_nil(BIF_ARG_1)) { - BIF_RET(new_binary(BIF_P,(byte*)"",0)); - } - if (is_not_list(BIF_ARG_1)) { - error: - BIF_ERROR(BIF_P, BADARG); - } - switch (bitstr_list_len(BIF_ARG_1, &sz)) { - case ERTS_IOLIST_TYPE: - goto error; - case ERTS_IOLIST_OVERFLOW: - BIF_ERROR(BIF_P, SYSTEM_LIMIT); - } - bin = new_binary(BIF_P, (byte *)NULL, sz); - bytes = binary_bytes(bin); -#ifdef DEBUG - offset = list_to_bitstr_buf(BIF_ARG_1, (char*) bytes, sz); -#else - offset = list_to_bitstr_buf(BIF_ARG_1, (char*) bytes); -#endif - ASSERT(offset >= 0); - if (offset > 0) { - hp = HAlloc(BIF_P, ERL_SUB_BIN_SIZE); - sb1 = (ErlSubBin *) hp; - sb1->thing_word = HEADER_SUB_BIN; - sb1->size = sz-1; - sb1->offs = 0; - sb1->orig = bin; - sb1->bitoffs = 0; - sb1->bitsize = offset; - sb1->is_writable = 0; - bin = make_binary(sb1); + BIF_RETTYPE ret; + + if (is_nil(BIF_ARG_1)) + ERTS_BIF_PREP_RET(ret, new_binary(BIF_P, (byte *) "", 0)); + else if (is_not_list(BIF_ARG_1)) + ERTS_BIF_PREP_ERROR(ret, BIF_P, BADARG); + else { + /* check for [bitstring()] case */ + Eterm h = CAR(list_val(BIF_ARG_1)); + Eterm t = CDR(list_val(BIF_ARG_1)); + if (is_binary(h) && is_nil(t)) { + ERTS_BIF_PREP_RET(ret, h); + } + else { + ErtsL2BState state = ERTS_L2B_STATE_INITER(BIF_P, + BIF_ARG_1, + bif_export[BIF_list_to_bitstring_1], + bitstr_list_len, + list_to_bitstr_buf_yielding); + int orig_reds_left = ERTS_BIF_REDS_LEFT(BIF_P); + + /* + * First try to do it all at once without having to use + * yielding list_to_bitstr_buf(). + */ + state.buf.iolist.reds_left = orig_reds_left; + switch (bitstr_list_len(&state.buf.iolist)) { + case ERTS_IOLIST_OK: { + ErlDrvSizeT size = state.buf.iolist.size; + + state.bin = new_binary(BIF_P, (byte *) NULL, size); + state.buf.buf = (char *) binary_bytes(state.bin); + state.buf.len = size; + state.buf.iolist.obj = BIF_ARG_1; + + if (size < ERTS_IOLIST_TO_BUF_BYTES_PER_RED*CONTEXT_REDS) { + /* An (over) estimation of reductions needed */ + int reds_left = state.buf.iolist.reds_left; + int to_buf_reds = orig_reds_left - reds_left; + to_buf_reds += size/ERTS_IOLIST_TO_BUF_BYTES_PER_RED; + if (to_buf_reds <= reds_left) { + ErlDrvSizeT res; + + res = list_to_bitstr_buf_not_yielding(&state.buf); + if (res == 0) { + Eterm res_bin = l2b_final_touch(BIF_P, &state); + BUMP_REDS(BIF_P, to_buf_reds); + ERTS_BIF_PREP_RET(ret, res_bin); + break; /* done */ + } + if (!ERTS_IOLIST_TO_BUF_FAILED(res)) + ERTS_INTERNAL_ERROR("iolist_size/iolist_to_buf missmatch"); + if (res == ERTS_IOLIST_TO_BUF_OVERFLOW) + goto overflow; + goto type_error; + } + } + /* + * Since size has been computed list_to_binary_chunk() expects + * the state prepared for list_to_bitstr_buf. + */ + + /* Fall through... */ + } + case ERTS_IOLIST_YIELD: + ret = list_to_binary_chunk(THE_NON_VALUE, + &state, + state.buf.iolist.reds_left, + 0); + break; + case ERTS_IOLIST_OVERFLOW: + overflow: + ERTS_BIF_PREP_ERROR(ret, BIF_P, SYSTEM_LIMIT); + break; + case ERTS_IOLIST_TYPE: + type_error: + default: + ERTS_BIF_PREP_ERROR(ret, BIF_P, BADARG); + break; + } + } } - - BIF_RET(bin); + + return ret; } BIF_RETTYPE split_binary_2(BIF_ALIST_2) @@ -605,123 +1058,353 @@ BIF_RETTYPE split_binary_2(BIF_ALIST_2) * Local functions. */ +static int +list_to_bitstr_buf_bcopy(ErtsIOList2BufState *state, Eterm obj, int *yield_countp); + /* * The input list is assumed to be type-correct and the buffer is * assumed to be of sufficient size. Those assumptions are verified in * the DEBUG-built emulator. */ -static int +static ErlDrvSizeT +list_to_bitstr_buf(int yield_support, ErtsIOList2BufState *state) +{ + +#undef LIST_TO_BITSTR_BUF_BCOPY_DBG +#undef LIST_TO_BITSTR_BUF_BCOPY #ifdef DEBUG -list_to_bitstr_buf(Eterm obj, char* buf, Uint len) +#define LIST_TO_BITSTR_BUF_BCOPY_DBG \ + len -= size + (offset>7); #else -list_to_bitstr_buf(Eterm obj, char* buf) +#define LIST_TO_BITSTR_BUF_BCOPY_DBG #endif -{ - Eterm* objp; - int offset = 0; +#define LIST_TO_BITSTR_BUF_BCOPY(CONSP) \ + do { \ + byte* bptr; \ + Uint bitsize; \ + Uint bitoffs; \ + Uint num_bits; \ + size_t size = binary_size(obj); \ + if (yield_support) { \ + size_t max_size = ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; \ + if (yield_count > 0) \ + max_size *= yield_count+1; \ + if (size > max_size) { \ + state->objp = CONSP; \ + goto L_bcopy_yield; \ + } \ + if (size >= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT) { \ + int cost = (int) size; \ + cost /= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; \ + yield_count -= cost; \ + } \ + } \ + ASSERT(size <= len); \ + ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); \ + num_bits = 8*size+bitsize; \ + copy_binary_to_buffer(buf, offset, bptr, bitoffs, num_bits); \ + offset += bitsize; \ + buf += size + (offset>7); \ + LIST_TO_BITSTR_BUF_BCOPY_DBG; \ + offset = offset & 7; \ + } while(0) + +#ifdef DEBUG + ErlDrvSizeT len; +#endif + Eterm obj; + char *buf; + Eterm *objp = NULL; + int offset; + int init_yield_count = 0, yield_count; DECLARE_ESTACK(s); - goto L_again; - - while (!ESTACK_ISEMPTY(s)) { - obj = ESTACK_POP(s); - L_again: - if (is_list(obj)) { - L_iter_list: - objp = list_val(obj); - obj = CAR(objp); - if (is_byte(obj)) { - ASSERT(len > 0); - if (offset == 0) { - *buf++ = unsigned_val(obj); - } else { - *buf = (char)((unsigned_val(obj) >> offset) | - ((*buf >> (8-offset)) << (8-offset))); - buf++; - *buf = (unsigned_val(obj) << (8-offset)); - } + + obj = state->iolist.obj; + buf = state->buf; + offset = state->offset; #ifdef DEBUG - len--; + len = state->len; #endif - } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - - ASSERT(size <= len); - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - num_bits = 8*size+bitsize; - copy_binary_to_buffer(buf, offset, bptr, bitoffs, num_bits); - offset += bitsize; - buf += size + (offset>7); + + if (!yield_support) { + yield_count = init_yield_count = 0; /* Shut up faulty warning... >:-( */ + goto L_again; + } + else { + + if (state->iolist.reds_left <= 0) + return ERTS_IOLIST_TO_BUF_YIELD; + + ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK); + init_yield_count = (ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED + * state->iolist.reds_left); + yield_count = init_yield_count; + + if (!state->iolist.estack.start) + goto L_again; + else { + int chk_stack; + /* Restart; restore state... */ + ESTACK_RESTORE(s, &state->iolist.estack); + + if (!state->bcopy.bptr) + chk_stack = 0; + else { + chk_stack = 1; + if (list_to_bitstr_buf_bcopy(state, THE_NON_VALUE, &yield_count)) { + /* Yield again... */ + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + ESTACK_SAVE(s, &state->iolist.estack); + return ERTS_IOLIST_TO_BUF_YIELD; + } + buf = state->buf; + offset = state->offset; #ifdef DEBUG - len -= size + (offset>7); + len = state->len; #endif - offset = offset & 7; - } else if (is_list(obj)) { - ESTACK_PUSH(s, CDR(objp)); - goto L_iter_list; /* on head */ - } else { - ASSERT(is_nil(obj)); } - obj = CDR(objp); - if (is_list(obj)) { - goto L_iter_list; /* on tail */ - } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - - ASSERT(size <= len); - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - num_bits = 8*size+bitsize; - copy_binary_to_buffer(buf, offset, bptr, bitoffs, num_bits); - offset += bitsize; - buf += size+(offset>7); + objp = state->objp; + state->objp = NULL; + + if (objp) + goto L_tail; + if (!chk_stack) + goto L_again; + /* check stack */ + } + } + + while (!ESTACK_ISEMPTY(s)) { + obj = ESTACK_POP(s); + L_again: + if (is_list(obj)) { + while (1) { /* Tail loop */ + while (1) { /* Head loop */ + if (yield_support && --yield_count <= 0) + goto L_yield; + objp = list_val(obj); + obj = CAR(objp); + if (is_byte(obj)) { + ASSERT(len > 0); + if (offset == 0) { + *buf++ = unsigned_val(obj); + } else { + *buf = (char)((unsigned_val(obj) >> offset) | + ((*buf >> (8-offset)) << (8-offset))); + buf++; + *buf = (unsigned_val(obj) << (8-offset)); + } #ifdef DEBUG - len -= size+(offset>7); + len--; #endif - offset = offset & 7; - } else { - ASSERT(is_nil(obj)); + } else if (is_binary(obj)) { + LIST_TO_BITSTR_BUF_BCOPY(objp); + } else if (is_list(obj)) { + ESTACK_PUSH(s, CDR(objp)); + continue; /* Head loop */ + } else { + ASSERT(is_nil(obj)); + } + break; + } + + L_tail: + + obj = CDR(objp); + if (is_list(obj)) { + continue; /* Tail loop */ + } else if (is_binary(obj)) { + LIST_TO_BITSTR_BUF_BCOPY(NULL); + } else { + ASSERT(is_nil(obj)); + } + break; } } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - - ASSERT(size <= len); - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - num_bits = 8*size+bitsize; - copy_binary_to_buffer(buf, offset, bptr, bitoffs, num_bits); - offset += bitsize; - buf += size + (offset>7); -#ifdef DEBUG - len -= size + (offset>7); -#endif - offset = offset & 7; + LIST_TO_BITSTR_BUF_BCOPY(NULL); } else { + if (yield_support && --yield_count <= 0) + goto L_yield; ASSERT(is_nil(obj)); } } DESTROY_ESTACK(s); - return offset; + + if (yield_support) { + int reds; + CLEAR_SAVED_ESTACK(&state->iolist.estack); + reds = ((init_yield_count - yield_count - 1) + / ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED) + 1; + BUMP_REDS(state->iolist.c_p, reds); + state->iolist.reds_left -= reds; + if (state->iolist.reds_left < 0) + state->iolist.reds_left = 0; + } + state->buf = buf; + state->offset = offset; + return 0; + +L_bcopy_yield: + + state->buf = buf; + state->offset = offset; +#ifdef DEBUG + state->len = len; +#endif + + if (list_to_bitstr_buf_bcopy(state, obj, &yield_count) == 0) + ERTS_INTERNAL_ERROR("Missing yield"); + + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + ESTACK_SAVE(s, &state->iolist.estack); + return ERTS_IOLIST_TO_BUF_YIELD; + +L_yield: + + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + state->iolist.obj = obj; + state->buf = buf; + state->offset = offset; + ESTACK_SAVE(s, &state->iolist.estack); +#ifdef DEBUG + state->len = len; +#endif + return ERTS_IOLIST_TO_BUF_YIELD; + + +#undef LIST_TO_BITSTR_BUF_BCOPY_DBG +#undef LIST_TO_BITSTR_BUF_BCOPY + +} + +static ErlDrvSizeT +list_to_bitstr_buf_yielding(ErtsIOList2BufState *state) +{ + return list_to_bitstr_buf(1, state); +} + +static ErlDrvSizeT +list_to_bitstr_buf_not_yielding(ErtsIOList2BufState *state) +{ + return list_to_bitstr_buf(0, state); } static int -bitstr_list_len(Eterm obj, Uint* num_bytes) +list_to_bitstr_buf_bcopy(ErtsIOList2BufState *state, Eterm obj, int *yield_countp) +{ + int res; + char *buf = state->buf; + char *next_buf; + int offset = state->offset; + int next_offset; +#ifdef DEBUG + ErlDrvSizeT len = state->len; + ErlDrvSizeT next_len; +#endif + byte* bptr; + size_t size; + size_t max_size; + Uint bitoffs; + Uint num_bits; + Uint bitsize; + int yield_count = *yield_countp; + + if (state->bcopy.bptr) { + bptr = state->bcopy.bptr; + size = state->bcopy.size; + bitoffs = state->bcopy.bitoffs; + bitsize = state->bcopy.bitsize; + state->bcopy.bptr = NULL; + } + else { + + ASSERT(is_binary(obj)); + + size = binary_size(obj); + + ASSERT(size <= len); + + ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); + } + + max_size = (size_t) ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; + if (yield_count > 0) + max_size *= (size_t) (yield_count+1); + + if (size <= max_size) { + if (size >= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT) { + int cost = (int) size; + cost /= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; + yield_count -= cost; + } + next_offset = offset + bitsize; + next_buf = buf + size+(next_offset>7); +#ifdef DEBUG + next_len = len - size+(next_offset>7); +#endif + next_offset &= 7; + num_bits = 8*size+bitsize; + res = 0; + } + else { + ASSERT(0 < max_size && max_size < size); + yield_count = 0; + state->bcopy.bptr = bptr + max_size; + state->bcopy.bitoffs = bitoffs; + state->bcopy.bitsize = bitsize; + state->bcopy.size = size - max_size; + next_buf = buf + max_size; +#ifdef DEBUG + next_len = len - max_size; +#endif + next_offset = offset; + num_bits = 8*max_size; + size = max_size; + res = 1; + } + + copy_binary_to_buffer(buf, offset, bptr, bitoffs, num_bits); + + state->offset = next_offset; + state->buf = next_buf; +#ifdef DEBUG + state->len = next_len; +#endif + *yield_countp = yield_count; + + return res; +} + +static int +bitstr_list_len(ErtsIOListState *state) { Eterm* objp; - Uint len = 0; - Uint offs = 0; + Eterm obj; + Uint len, offs; + int res, init_yield_count, yield_count; DECLARE_ESTACK(s); + + if (state->reds_left <= 0) + return ERTS_IOLIST_YIELD; + + len = (Uint) state->size; + offs = state->offs; + obj = state->obj; + + ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK); + init_yield_count = ERTS_IOLIST_SIZE_YIELDS_COUNT_PER_RED; + init_yield_count *= state->reds_left; + yield_count = init_yield_count; + if (state->estack.start) { + /* Restart; restore estack... */ + ESTACK_RESTORE(s, &state->estack); + } + goto L_again; #define SAFE_ADD(Var, Val) \ @@ -748,46 +1431,55 @@ bitstr_list_len(Eterm obj, Uint* num_bytes) obj = ESTACK_POP(s); L_again: if (is_list(obj)) { - L_iter_list: - objp = list_val(obj); - /* Head */ - obj = CAR(objp); - if (is_byte(obj)) { - len++; - if (len == 0) { - goto L_overflow_error; + while (1) { /* Tail loop */ + while (1) { /* Head loop */ + if (--yield_count <= 0) + goto L_yield; + objp = list_val(obj); + /* Head */ + obj = CAR(objp); + if (is_byte(obj)) { + len++; + if (len == 0) { + goto L_overflow_error; + } + } else if (is_binary(obj)) { + SAFE_ADD(len, binary_size(obj)); + SAFE_ADD_BITSIZE(offs, obj); + } else if (is_list(obj)) { + ESTACK_PUSH(s, CDR(objp)); + continue; /* Head loop */ + } else if (is_not_nil(obj)) { + goto L_type_error; + } + break; } - } else if (is_binary(obj)) { - SAFE_ADD(len, binary_size(obj)); - SAFE_ADD_BITSIZE(offs, obj); - } else if (is_list(obj)) { - ESTACK_PUSH(s, CDR(objp)); - goto L_iter_list; /* on head */ - } else if (is_not_nil(obj)) { - goto L_type_error; + /* Tail */ + obj = CDR(objp); + if (is_list(obj)) + continue; /* Tail loop */ + else if (is_binary(obj)) { + SAFE_ADD(len, binary_size(obj)); + SAFE_ADD_BITSIZE(offs, obj); + } else if (is_not_nil(obj)) { + goto L_type_error; + } + break; } - /* Tail */ - obj = CDR(objp); - if (is_list(obj)) - goto L_iter_list; /* on tail */ - else if (is_binary(obj)) { + } else { + if (--yield_count <= 0) + goto L_yield; + if (is_binary(obj)) { SAFE_ADD(len, binary_size(obj)); SAFE_ADD_BITSIZE(offs, obj); } else if (is_not_nil(obj)) { goto L_type_error; } - } else if (is_binary(obj)) { - SAFE_ADD(len, binary_size(obj)); - SAFE_ADD_BITSIZE(offs, obj); - } else if (is_not_nil(obj)) { - goto L_type_error; } } #undef SAFE_ADD #undef SAFE_ADD_BITSIZE - DESTROY_ESTACK(s); - /* * Make sure that the number of bits in the bitstring will fit * in an Uint to ensure that the binary can be matched using @@ -800,15 +1492,42 @@ bitstr_list_len(Eterm obj, Uint* num_bytes) if (len << 3 < len) { goto L_overflow_error; } - *num_bytes = len; - return ERTS_IOLIST_OK; + state->size = len; - L_type_error: - DESTROY_ESTACK(s); - return ERTS_IOLIST_TYPE; + res = ERTS_IOLIST_OK; + + L_return: { + int yc = init_yield_count - yield_count; + int reds; + + DESTROY_ESTACK(s); + CLEAR_SAVED_ESTACK(&state->estack); + + reds = (yc - 1)/ERTS_IOLIST_SIZE_YIELDS_COUNT_PER_RED + 1; + BUMP_REDS(state->c_p, reds); + state->reds_left -= reds; + state->size = (ErlDrvSizeT) len; + state->have_size = 1; + return res; + } L_overflow_error: - DESTROY_ESTACK(s); - return ERTS_IOLIST_OVERFLOW; + res = ERTS_IOLIST_OVERFLOW; + len = 0; + goto L_return; + + L_type_error: + res = ERTS_IOLIST_TYPE; + len = 0; + goto L_return; + + L_yield: + BUMP_ALL_REDS(state->c_p); + state->reds_left = 0; + state->size = len; + state->offs = offs; + state->obj = obj; + ESTACK_SAVE(s, &state->estack); + return ERTS_IOLIST_YIELD; } diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c index 7d4f52ee23..08265b590d 100644 --- a/erts/emulator/beam/break.c +++ b/erts/emulator/beam/break.c @@ -256,6 +256,7 @@ print_process_info(int to, void *to_arg, Process *p) p->current[1], p->current[2]); } + erts_print(to, to_arg, "Run queue: %d\n", erts_get_runq_proc(p)->ix); erts_print(to, to_arg, "Spawned by: %T\n", p->parent); approx_started = (time_t) p->approx_started; diff --git a/erts/emulator/beam/copy.c b/erts/emulator/beam/copy.c index 3a987e213b..50548850eb 100644 --- a/erts/emulator/beam/copy.c +++ b/erts/emulator/beam/copy.c @@ -48,7 +48,8 @@ copy_object(Eterm obj, Process* to) if (DTRACE_ENABLED(copy_object)) { DTRACE_CHARBUF(proc_name, 64); - erts_snprintf(proc_name, sizeof(proc_name), "%T", to->common.id); + erts_snprintf(proc_name, sizeof(DTRACE_CHARBUF_NAME(proc_name)), + "%T", to->common.id); DTRACE2(copy_object, proc_name, size); } #endif diff --git a/erts/emulator/beam/dist.c b/erts/emulator/beam/dist.c index 6ecf3f0722..ec07ddcd9c 100644 --- a/erts/emulator/beam/dist.c +++ b/erts/emulator/beam/dist.c @@ -851,9 +851,12 @@ erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message) #ifdef USE_VM_PROBES *node_name = *sender_name = *receiver_name = '\0'; if (DTRACE_ENABLED(message_send) || DTRACE_ENABLED(message_send_remote)) { - erts_snprintf(node_name, sizeof(node_name), "%T", dsdp->dep->sysname); - erts_snprintf(sender_name, sizeof(sender_name), "%T", sender->common.id); - erts_snprintf(receiver_name, sizeof(receiver_name), "%T", remote); + erts_snprintf(node_name, sizeof(DTRACE_CHARBUF_NAME(node_name)), + "%T", dsdp->dep->sysname); + erts_snprintf(sender_name, sizeof(DTRACE_CHARBUF_NAME(sender_name)), + "%T", sender->common.id); + erts_snprintf(receiver_name, sizeof(DTRACE_CHARBUF_NAME(receiver_name)), + "%T", remote); msize = size_object(message); if (token != NIL && token != am_have_dt_utag) { tok_label = signed_val(SEQ_TRACE_T_LABEL(token)); @@ -908,9 +911,11 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message) #ifdef USE_VM_PROBES *node_name = *sender_name = *receiver_name = '\0'; if (DTRACE_ENABLED(message_send) || DTRACE_ENABLED(message_send_remote)) { - erts_snprintf(node_name, sizeof(node_name), "%T", dsdp->dep->sysname); - erts_snprintf(sender_name, sizeof(sender_name), "%T", sender->common.id); - erts_snprintf(receiver_name, sizeof(receiver_name), + erts_snprintf(node_name, sizeof(DTRACE_CHARBUF_NAME(node_name)), + "%T", dsdp->dep->sysname); + erts_snprintf(sender_name, sizeof(DTRACE_CHARBUF_NAME(sender_name)), + "%T", sender->common.id); + erts_snprintf(receiver_name, sizeof(DTRACE_CHARBUF_NAME(receiver_name)), "{%T,%s}", remote_name, node_name); msize = size_object(message); if (token != NIL && token != am_have_dt_utag) { @@ -971,11 +976,14 @@ erts_dsig_send_exit_tt(ErtsDSigData *dsdp, Eterm local, Eterm remote, #ifdef USE_VM_PROBES *node_name = *sender_name = *remote_name = '\0'; if (DTRACE_ENABLED(process_exit_signal_remote)) { - erts_snprintf(node_name, sizeof(node_name), "%T", dsdp->dep->sysname); - erts_snprintf(sender_name, sizeof(sender_name), "%T", sender->common.id); - erts_snprintf(remote_name, sizeof(remote_name), + erts_snprintf(node_name, sizeof(DTRACE_CHARBUF_NAME(node_name)), + "%T", dsdp->dep->sysname); + erts_snprintf(sender_name, sizeof(DTRACE_CHARBUF_NAME(sender_name)), + "%T", sender->common.id); + erts_snprintf(remote_name, sizeof(DTRACE_CHARBUF_NAME(remote_name)), "{%T,%s}", remote, node_name); - erts_snprintf(reason_str, sizeof(reason), "%T", reason); + erts_snprintf(reason_str, sizeof(DTRACE_CHARBUF_NAME(reason_str)), + "%T", reason); if (token != NIL && token != am_have_dt_utag) { tok_label = signed_val(SEQ_TRACE_T_LABEL(token)); tok_lastcnt = signed_val(SEQ_TRACE_T_LASTCNT(token)); @@ -1797,8 +1805,9 @@ dsig_send(ErtsDSigData *dsdp, Eterm ctl, Eterm msg, int force_busy) DTRACE_CHARBUF(port_str, 64); DTRACE_CHARBUF(remote_str, 64); - erts_snprintf(port_str, sizeof(port_str), "%T", cid); - erts_snprintf(remote_str, sizeof(remote_str), + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), + "%T", cid); + erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), "%T", dep->sysname); DTRACE3(dist_port_not_busy, erts_this_node_sysname, port_str, remote_str); @@ -1855,9 +1864,11 @@ dsig_send(ErtsDSigData *dsdp, Eterm ctl, Eterm msg, int force_busy) DTRACE_CHARBUF(remote_str, 64); DTRACE_CHARBUF(pid_str, 16); - erts_snprintf(port_str, sizeof(port_str), "%T", cid); - erts_snprintf(remote_str, sizeof(remote_str), "%T", dep->sysname); - erts_snprintf(pid_str, sizeof(pid_str), "%T", c_p->common.id); + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", cid); + erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), + "%T", dep->sysname); + erts_snprintf(pid_str, sizeof(DTRACE_CHARBUF_NAME(pid_str)), + "%T", c_p->common.id); DTRACE4(dist_port_busy, erts_this_node_sysname, port_str, remote_str, pid_str); } @@ -1890,8 +1901,9 @@ dist_port_command(Port *prt, ErtsDistOutputBuf *obuf) DTRACE_CHARBUF(port_str, 64); DTRACE_CHARBUF(remote_str, 64); - erts_snprintf(port_str, sizeof(port_str), "%T", prt->common.id); - erts_snprintf(remote_str, sizeof(remote_str), + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), + "%T", prt->common.id); + erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), "%T", prt->dist_entry->sysname); DTRACE4(dist_output, erts_this_node_sysname, port_str, remote_str, size); @@ -1944,8 +1956,9 @@ dist_port_commandv(Port *prt, ErtsDistOutputBuf *obuf) DTRACE_CHARBUF(port_str, 64); DTRACE_CHARBUF(remote_str, 64); - erts_snprintf(port_str, sizeof(port_str), "%T", prt->common.id); - erts_snprintf(remote_str, sizeof(remote_str), + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), + "%T", prt->common.id); + erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), "%T", prt->dist_entry->sysname); DTRACE4(dist_outputv, erts_this_node_sysname, port_str, remote_str, size); @@ -2280,8 +2293,9 @@ erts_dist_port_not_busy(Port *prt) DTRACE_CHARBUF(port_str, 64); DTRACE_CHARBUF(remote_str, 64); - erts_snprintf(port_str, sizeof(port_str), "%T", prt->common.id); - erts_snprintf(remote_str, sizeof(remote_str), + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), + "%T", prt->common.id); + erts_snprintf(remote_str, sizeof(DTRACE_CHARBUF_NAME(remote_str)), "%T", prt->dist_entry->sysname); DTRACE3(dist_port_not_busy, erts_this_node_sysname, port_str, remote_str); @@ -3246,10 +3260,10 @@ send_nodes_mon_msgs(Process *c_p, Eterm what, Eterm node, Eterm type, Eterm reas DTRACE_CHARBUF(type_str, 12); DTRACE_CHARBUF(reason_str, 64); - erts_snprintf(what_str, sizeof(what_str), "%T", what); - erts_snprintf(node_str, sizeof(node_str), "%T", node); - erts_snprintf(type_str, sizeof(type_str), "%T", type); - erts_snprintf(reason_str, sizeof(reason_str), "%T", reason); + erts_snprintf(what_str, sizeof(DTRACE_CHARBUF_NAME(what_str)), "%T", what); + erts_snprintf(node_str, sizeof(DTRACE_CHARBUF_NAME(node_str)), "%T", node); + erts_snprintf(type_str, sizeof(DTRACE_CHARBUF_NAME(type_str)), "%T", type); + erts_snprintf(reason_str, sizeof(DTRACE_CHARBUF_NAME(reason_str)), "%T", reason); DTRACE5(dist_monitor, erts_this_node_sysname, what_str, node_str, type_str, reason_str); } diff --git a/erts/emulator/beam/dist.h b/erts/emulator/beam/dist.h index 0519a9225e..f32b999198 100644 --- a/erts/emulator/beam/dist.h +++ b/erts/emulator/beam/dist.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 @@ -40,6 +40,7 @@ #define DFLAG_SMALL_ATOM_TAGS 0x4000 #define DFLAG_INTERNAL_TAGS 0x8000 #define DFLAG_UTF8_ATOMS 0x10000 +#define DFLAG_MAP_TAG 0x20000 /* All flags that should be enabled when term_to_binary/1 is used. */ #define TERM_TO_BINARY_DFLAGS (DFLAG_EXTENDED_REFERENCES \ @@ -47,7 +48,8 @@ | DFLAG_NEW_FLOATS \ | DFLAG_EXTENDED_PIDS_PORTS \ | DFLAG_EXPORT_PTR_TAG \ - | DFLAG_BIT_BINARIES) + | DFLAG_BIT_BINARIES \ + | DFLAG_MAP_TAG) /* opcodes used in distribution messages */ #define DOP_LINK 1 diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c index 05ac24e04d..90cd227fae 100644 --- a/erts/emulator/beam/erl_alloc.c +++ b/erts/emulator/beam/erl_alloc.c @@ -1873,8 +1873,8 @@ erts_alc_fatal_error(int error, int func, ErtsAlcType_t n, ...) size = va_arg(argp, Uint); va_end(argp); erl_exit(1, - "%s: Cannot %s %lu bytes of memory (of type \"%s\").\n", - allctr_str, op, size, t_str); + "%s: Cannot %s %lu bytes of memory (of type \"%s\", thread %d).\n", + allctr_str, op, size, t_str, ERTS_ALC_GET_THR_IX()); break; } case ERTS_ALC_E_NOALLCTR: diff --git a/erts/emulator/beam/erl_alloc.h b/erts/emulator/beam/erl_alloc.h index 942eaa47d0..d3109b9432 100644 --- a/erts/emulator/beam/erl_alloc.h +++ b/erts/emulator/beam/erl_alloc.h @@ -492,7 +492,7 @@ static TYPE * \ NAME##_alloc(void) \ { \ ErtsSchedulerData *esdp = erts_get_scheduler_data(); \ - if (!esdp) \ + if (!esdp || ERTS_SCHEDULER_IS_DIRTY(esdp)) \ return NULL; \ return (TYPE *) erts_sspa_alloc(sspa_data_##NAME##__, \ (int) esdp->no - 1); \ diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types index 17ac6316b7..21434eb117 100644 --- a/erts/emulator/beam/erl_alloc.types +++ b/erts/emulator/beam/erl_alloc.types @@ -267,7 +267,6 @@ type CODE_IX_LOCK_Q SHORT_LIVED SYSTEM code_ix_lock_q type PROC_INTERVAL LONG_LIVED SYSTEM process_interval type BUSY_CALLER_TAB SHORT_LIVED SYSTEM busy_caller_table type BUSY_CALLER SHORT_LIVED SYSTEM busy_caller -type PORT_DATA_HEAP STANDARD SYSTEM port_data_heap type PROC_SYS_TSK SHORT_LIVED PROCESSES proc_sys_task type PROC_SYS_TSK_QS SHORT_LIVED PROCESSES proc_sys_task_queues @@ -357,12 +356,14 @@ type DB_MS_PSDO_PROC LONG_LIVED_LOW ETS db_match_pseudo_proc type SCHDLR_DATA LONG_LIVED_LOW SYSTEM scheduler_data type LL_TEMP_TERM LONG_LIVED_LOW SYSTEM ll_temp_term +type NIF_TRAP_EXPORT STANDARD_LOW CODE nif_trap_export_entry type EXPORT LONG_LIVED_LOW CODE export_entry type MONITOR_SH STANDARD_LOW PROCESSES monitor_sh type NLINK_SH STANDARD_LOW PROCESSES nlink_sh type AINFO_REQ STANDARD_LOW SYSTEM alloc_info_request type SCHED_WTIME_REQ STANDARD_LOW SYSTEM sched_wall_time_request type GC_INFO_REQ STANDARD_LOW SYSTEM gc_info_request +type PORT_DATA_HEAP STANDARD_LOW SYSTEM port_data_heap +else # "fullword" @@ -375,12 +376,14 @@ type DB_MS_PSDO_PROC LONG_LIVED ETS db_match_pseudo_proc type SCHDLR_DATA LONG_LIVED SYSTEM scheduler_data type LL_TEMP_TERM LONG_LIVED SYSTEM ll_temp_term +type NIF_TRAP_EXPORT STANDARD CODE nif_trap_export_entry type EXPORT LONG_LIVED CODE export_entry type MONITOR_SH FIXED_SIZE PROCESSES monitor_sh type NLINK_SH FIXED_SIZE PROCESSES nlink_sh type AINFO_REQ SHORT_LIVED SYSTEM alloc_info_request type SCHED_WTIME_REQ SHORT_LIVED SYSTEM sched_wall_time_request type GC_INFO_REQ SHORT_LIVED SYSTEM gc_info_request +type PORT_DATA_HEAP STANDARD SYSTEM port_data_heap +endif @@ -395,6 +398,7 @@ type DRV_EV_STATE LONG_LIVED SYSTEM driver_event_state type DRV_EV_D_STATE FIXED_SIZE SYSTEM driver_event_data_state type DRV_SEL_D_STATE FIXED_SIZE SYSTEM driver_select_data_state type FD_LIST SHORT_LIVED SYSTEM fd_list +type ACTIVE_FD_ARR SHORT_LIVED SYSTEM active_fd_array type POLLSET LONG_LIVED SYSTEM pollset type POLLSET_UPDREQ SHORT_LIVED SYSTEM pollset_update_req type POLL_FDS LONG_LIVED SYSTEM poll_fds diff --git a/erts/emulator/beam/erl_alloc_util.c b/erts/emulator/beam/erl_alloc_util.c index 45f0cc4312..e3172dc4fb 100644 --- a/erts/emulator/beam/erl_alloc_util.c +++ b/erts/emulator/beam/erl_alloc_util.c @@ -205,7 +205,7 @@ MBC after deallocating first block: ASSERT(((UWord)(F) & (~FLG_MASK|THIS_FREE_BLK_HDR_FLG|PREV_FREE_BLK_HDR_FLG)) == THIS_FREE_BLK_HDR_FLG), \ (B)->bhdr = ((Sz) | (F)), \ (B)->u.carrier = (C)) - + # define IS_MBC_FIRST_ABLK(AP,B) \ ((((UWord)(B) & ~ERTS_SACRR_UNIT_MASK) == MBC_HEADER_SIZE(AP)) \ && ((B)->bhdr & MBC_ABLK_OFFSET_MASK) == 0) @@ -378,9 +378,8 @@ do { \ #ifdef ERTS_SMP #define SBC_HEADER_SIZE \ - (UNIT_CEILING(sizeof(Carrier_t) \ - - sizeof(ErtsAlcCPoolData_t) \ - + ABLK_HDR_SZ) \ + (UNIT_CEILING(offsetof(Carrier_t, cpool) \ + + ABLK_HDR_SZ) \ - ABLK_HDR_SZ) #else #define SBC_HEADER_SIZE \ @@ -929,6 +928,88 @@ unlink_carrier(CarrierList_t *cl, Carrier_t *crr) #ifdef ERTS_SMP +#ifdef DEBUG +static int is_in_list(ErtsDoubleLink_t* sentinel, ErtsDoubleLink_t* node) +{ + ErtsDoubleLink_t* p; + + ASSERT(node != sentinel); + for (p = sentinel->next; p != sentinel; p = p->next) { + if (p == node) + return 1; + } + return 0; +} +#endif /* DEBUG */ + +static ERTS_INLINE void +link_edl_after(ErtsDoubleLink_t* after_me, ErtsDoubleLink_t* node) +{ + ErtsDoubleLink_t* before_me = after_me->next; + ASSERT(node != after_me && node != before_me); + node->next = before_me; + node->prev = after_me; + before_me->prev = node; + after_me->next = node; +} + +static ERTS_INLINE void +link_edl_before(ErtsDoubleLink_t* before_me, ErtsDoubleLink_t* node) +{ + ErtsDoubleLink_t* after_me = before_me->prev; + ASSERT(node != before_me && node != after_me); + node->next = before_me; + node->prev = after_me; + before_me->prev = node; + after_me->next = node; +} + +static ERTS_INLINE void +unlink_edl(ErtsDoubleLink_t* node) +{ + node->next->prev = node->prev; + node->prev->next = node->next; +} + +static ERTS_INLINE void +relink_edl_before(ErtsDoubleLink_t* before_me, ErtsDoubleLink_t* node) +{ + if (node != before_me && node != before_me->prev) { + unlink_edl(node); + link_edl_before(before_me, node); + } +} + +static ERTS_INLINE int is_abandoned(Carrier_t *crr) +{ + return crr->cpool.abandoned.next != NULL; +} + +static ERTS_INLINE void +link_abandoned_carrier(ErtsDoubleLink_t* list, Carrier_t *crr) +{ + ASSERT(!is_abandoned(crr)); + + link_edl_after(list, &crr->cpool.abandoned); + + ASSERT(crr->cpool.abandoned.next != &crr->cpool.abandoned); + ASSERT(crr->cpool.abandoned.prev != &crr->cpool.abandoned); +} + +static ERTS_INLINE void +unlink_abandoned_carrier(Carrier_t *crr) +{ + ASSERT(is_in_list(&crr->cpool.orig_allctr->cpool.pooled_list, + &crr->cpool.abandoned) || + is_in_list(&crr->cpool.orig_allctr->cpool.traitor_list, + &crr->cpool.abandoned)); + + unlink_edl(&crr->cpool.abandoned); + + crr->cpool.abandoned.next = NULL; + crr->cpool.abandoned.prev = NULL; +} + static ERTS_INLINE void clear_busy_pool_carrier(Allctr_t *allctr, Carrier_t *crr) { @@ -955,7 +1036,7 @@ clear_busy_pool_carrier(Allctr_t *allctr, Carrier_t *crr) } } -#endif +#endif /* ERTS_SMP */ #if 0 #define ERTS_DBG_CHK_FIX_LIST(A, FIX, IX, B) \ @@ -1775,6 +1856,18 @@ handle_delayed_dealloc(Allctr_t *allctr, * data has been overwritten by the queue. */ Carrier_t *crr = FIRST_BLK_TO_MBC(allctr, blk); + + /* Restore word overwritten by the dd-queue as it will be read + * if this carrier is pulled from dc_list by cpool_fetch() + */ + ERTS_ALC_CPOOL_ASSERT(FBLK_TO_MBC(blk) != crr); + ERTS_ALC_CPOOL_ASSERT(sizeof(ErtsAllctrDDBlock_t) == sizeof(void*)); +#ifdef MBC_ABLK_OFFSET_BITS + blk->u.carrier = crr; +#else + blk->carrier = crr; +#endif + ERTS_ALC_CPOOL_ASSERT(ERTS_ALC_IS_CPOOL_ENABLED(allctr)); ERTS_ALC_CPOOL_ASSERT(allctr == crr->cpool.orig_allctr); ERTS_ALC_CPOOL_ASSERT(((erts_aint_t) allctr) @@ -2563,10 +2656,9 @@ mbc_realloc(Allctr_t *allctr, void *p, Uint size, Uint32 alcu_flgs, #ifdef ERTS_SMP #define ERTS_ALC_MAX_DEALLOC_CARRIER 10 -#define ERTS_ALC_CPOOL_MAX_FETCH_INSPECT 10 +#define ERTS_ALC_CPOOL_MAX_FETCH_INSPECT 20 +#define ERTS_ALC_CPOOL_MAX_TRAITOR_INSPECT 10 #define ERTS_ALC_CPOOL_CHECK_LIMIT_COUNT 100 -#define ERTS_ALC_CPOOL_MAX_NO_CARRIERS 5 -#define ERTS_ALC_CPOOL_INSERT_ALLOWED_OFFSET 100 #define ERTS_ALC_CPOOL_MAX_FAILED_STAT_READS 3 #define ERTS_ALC_CPOOL_PTR_MOD_MRK (((erts_aint_t) 1) << 0) @@ -2743,9 +2835,6 @@ cpool_insert(Allctr_t *allctr, Carrier_t *crr) (erts_aint_t) CARRIER_SZ(crr)); erts_atomic_inc_nob(&allctr->cpool.stat.no_carriers); - erts_smp_atomic_set_nob(&crr->allctr, - ((erts_aint_t) allctr)|ERTS_CRR_ALCTR_FLG_IN_POOL); - /* * We search in 'next' direction and begin by passing * one element before trying to insert. This in order to @@ -2804,6 +2893,9 @@ cpool_insert(Allctr_t *allctr, Carrier_t *crr) cpool_set_mod_marked(&cpd2p->prev, (erts_aint_t) &crr->cpool, (erts_aint_t) cpd1p); + + erts_smp_atomic_set_wb(&crr->allctr, + ((erts_aint_t) allctr)|ERTS_CRR_ALCTR_FLG_IN_POOL); } static void @@ -2904,59 +2996,163 @@ cpool_delete(Allctr_t *allctr, Allctr_t *prev_allctr, Carrier_t *crr) static Carrier_t * cpool_fetch(Allctr_t *allctr, UWord size) { - int i; + int i, i_stop, has_passed_sentinel; Carrier_t *crr; ErtsAlcCPoolData_t *cpdp; - ErtsAlcCPoolData_t *sentinel = &carrier_pool[allctr->alloc_no].sentinel; + ErtsAlcCPoolData_t *cpool_entrance; + ErtsAlcCPoolData_t *sentinel; + ErtsDoubleLink_t* dl; + ErtsDoubleLink_t* first_old_traitor; ERTS_ALC_CPOOL_ASSERT(allctr->alloc_no == ERTS_ALC_A_INVALID /* testcase */ || erts_thr_progress_is_managed_thread()); - i = 0; + i = ERTS_ALC_CPOOL_MAX_FETCH_INSPECT; + first_old_traitor = allctr->cpool.traitor_list.next; + cpool_entrance = NULL; - /* First; check our own pending dealloc carrier list... */ - crr = allctr->cpool.dc_list.last; - while (crr && i < ERTS_ALC_CPOOL_MAX_FETCH_INSPECT) { - if (erts_atomic_read_nob(&crr->cpool.max_size) >= size) { - unlink_carrier(&allctr->cpool.dc_list, crr); -#ifdef ERTS_ALC_CPOOL_DEBUG - ERTS_ALC_CPOOL_ASSERT(erts_smp_atomic_xchg_nob(&crr->allctr, - ((erts_aint_t) allctr)) - == (((erts_aint_t) allctr) & ~ERTS_CRR_ALCTR_FLG_MASK)); -#else - erts_smp_atomic_set_nob(&crr->allctr, ((erts_aint_t) allctr)); -#endif - return crr; + /* + * Search my own pooled_list, + * i.e my abandoned carriers that were in the pool last time I checked. + */ + + dl = allctr->cpool.pooled_list.next; + while(dl != &allctr->cpool.pooled_list) { + erts_aint_t exp, act; + crr = (Carrier_t *) (((char *) dl) - offsetof(Carrier_t, cpool.abandoned)); + + ASSERT(!is_in_list(&allctr->cpool.traitor_list, dl)); + ASSERT(crr->cpool.orig_allctr == allctr); + dl = dl->next; + exp = erts_smp_atomic_read_rb(&crr->allctr); + if ((exp & ERTS_CRR_ALCTR_FLG_MASK) == ERTS_CRR_ALCTR_FLG_IN_POOL + && erts_atomic_read_nob(&crr->cpool.max_size) >= size) { + /* Try to fetch it... */ + act = erts_smp_atomic_cmpxchg_mb(&crr->allctr, + (erts_aint_t) allctr, + exp); + if (act == exp) { + cpool_delete(allctr, ((Allctr_t *) (act & ~ERTS_CRR_ALCTR_FLG_MASK)), crr); + unlink_abandoned_carrier(crr); + + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.pooled_list); + return crr; + } + exp = act; + } + if (exp & ERTS_CRR_ALCTR_FLG_IN_POOL) { + if (!cpool_entrance) + cpool_entrance = &crr->cpool; + } + else { /* Not in pool, move to traitor_list */ + unlink_abandoned_carrier(crr); + link_abandoned_carrier(&allctr->cpool.traitor_list, crr); + } + if (--i <= 0) { + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.pooled_list); + return NULL; } - crr = crr->prev; - i++; } - /* ... then the pool ... */ + /* Now search traitor_list. + * i.e carriers employed by other allocators last time I checked. + * They might have been abandoned since then. + */ + + i_stop = (i < ERTS_ALC_CPOOL_MAX_TRAITOR_INSPECT ? + 0 : i - ERTS_ALC_CPOOL_MAX_TRAITOR_INSPECT); + dl = first_old_traitor; + while(dl != &allctr->cpool.traitor_list) { + erts_aint_t exp, act; + crr = (Carrier_t *) (((char *) dl) - offsetof(Carrier_t, cpool.abandoned)); + ASSERT(dl != &allctr->cpool.pooled_list); + ASSERT(crr->cpool.orig_allctr == allctr); + dl = dl->next; + exp = erts_smp_atomic_read_rb(&crr->allctr); + if (exp & ERTS_CRR_ALCTR_FLG_IN_POOL) { + if (!(exp & ERTS_CRR_ALCTR_FLG_BUSY) + && erts_atomic_read_nob(&crr->cpool.max_size) >= size) { + /* Try to fetch it... */ + act = erts_smp_atomic_cmpxchg_mb(&crr->allctr, + (erts_aint_t) allctr, + exp); + if (act == exp) { + cpool_delete(allctr, ((Allctr_t *) (act & ~ERTS_CRR_ALCTR_FLG_MASK)), crr); + unlink_abandoned_carrier(crr); + + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.traitor_list); + return crr; + } + exp = act; + } + if (exp & ERTS_CRR_ALCTR_FLG_IN_POOL) { + if (!cpool_entrance) + cpool_entrance = &crr->cpool; + + /* Move to pooled_list */ + unlink_abandoned_carrier(crr); + link_abandoned_carrier(&allctr->cpool.pooled_list, crr); + } + } + if (--i <= i_stop) { + /* Move sentinel to continue next search from here */ + relink_edl_before(dl, &allctr->cpool.traitor_list); + if (i > 0) + break; + else + return NULL; + } + } /* - * We search in 'prev' direction and begin by passing - * one element before trying to fetch. This in order to - * avoid contention with threads inserting elements. + * Finally search the shared pool and try employ foreign carriers */ - cpdp = cpool_aint2cpd(cpool_read(&sentinel->prev)); - if (cpdp == sentinel) - return NULL; + sentinel = &carrier_pool[allctr->alloc_no].sentinel; + if (cpool_entrance) { + /* We saw a pooled carried above, use it as entrance into the pool + */ + cpdp = cpool_entrance; + } + else { + /* No pooled carried seen above. Start search at cpool sentinel, + * but begin by passing one element before trying to fetch. + * This in order to avoid contention with threads inserting elements. + */ + cpool_entrance = sentinel; + cpdp = cpool_aint2cpd(cpool_read(&cpool_entrance->prev)); + if (cpdp == sentinel) + return NULL; + } - while (i < ERTS_ALC_CPOOL_MAX_FETCH_INSPECT) { + has_passed_sentinel = 0; + while (1) { erts_aint_t exp; cpdp = cpool_aint2cpd(cpool_read(&cpdp->prev)); - if (cpdp == sentinel) { + if (cpdp == cpool_entrance) { + if (cpool_entrance == sentinel) { + cpdp = cpool_aint2cpd(cpool_read(&cpdp->prev)); + if (cpdp == sentinel) + return NULL; + } + i = 0; /* Last one to inspect */ + } + else if (cpdp == sentinel) { + if (has_passed_sentinel) { + /* We been here before. cpool_entrance must have been removed */ + return NULL; + } cpdp = cpool_aint2cpd(cpool_read(&cpdp->prev)); if (cpdp == sentinel) return NULL; - i = ERTS_ALC_CPOOL_MAX_FETCH_INSPECT; /* Last one to inspect */ + has_passed_sentinel = 1; } - crr = (Carrier_t *) (((char *) cpdp) - offsetof(Carrier_t, cpool)); + crr = (Carrier_t *)(((char *)cpdp) - offsetof(Carrier_t, cpool)); exp = erts_smp_atomic_read_rb(&crr->allctr); - if (((exp & (ERTS_CRR_ALCTR_FLG_IN_POOL|ERTS_CRR_ALCTR_FLG_BUSY)) - == ERTS_CRR_ALCTR_FLG_IN_POOL) + if (((exp & (ERTS_CRR_ALCTR_FLG_MASK)) == ERTS_CRR_ALCTR_FLG_IN_POOL) && (erts_atomic_read_nob(&cpdp->max_size) >= size)) { erts_aint_t act; /* Try to fetch it... */ @@ -2965,11 +3161,35 @@ cpool_fetch(Allctr_t *allctr, UWord size) exp); if (act == exp) { cpool_delete(allctr, ((Allctr_t *) (act & ~ERTS_CRR_ALCTR_FLG_MASK)), crr); + if (crr->cpool.orig_allctr == allctr) { + unlink_abandoned_carrier(crr); + } return crr; } } - i++; + if (--i <= 0) + return NULL; } + + /* Last; check our own pending dealloc carrier list... */ + crr = allctr->cpool.dc_list.last; + while (crr) { + if (erts_atomic_read_nob(&crr->cpool.max_size) >= size) { + unlink_carrier(&allctr->cpool.dc_list, crr); +#ifdef ERTS_ALC_CPOOL_DEBUG + ERTS_ALC_CPOOL_ASSERT(erts_smp_atomic_xchg_nob(&crr->allctr, + ((erts_aint_t) allctr)) + == (((erts_aint_t) allctr) & ~ERTS_CRR_ALCTR_FLG_MASK)); +#else + erts_smp_atomic_set_nob(&crr->allctr, ((erts_aint_t) allctr)); +#endif + return crr; + } + crr = crr->prev; + if (--i <= 0) + return NULL; + } + return NULL; } @@ -3066,6 +3286,9 @@ schedule_dealloc_carrier(Allctr_t *allctr, Carrier_t *crr) return; } + if (is_abandoned(crr)) + unlink_abandoned_carrier(crr); + if (crr->cpool.thr_prgr == ERTS_THR_PRGR_INVALID || erts_thr_progress_has_reached(crr->cpool.thr_prgr)) { dealloc_carrier(allctr, crr, 1); @@ -3112,6 +3335,8 @@ cpool_init_carrier_data(Allctr_t *allctr, Carrier_t *crr) limit = (csz/100)*allctr->cpool.util_limit; crr->cpool.abandon_limit = limit; } + crr->cpool.abandoned.next = NULL; + crr->cpool.abandoned.prev = NULL; } static void @@ -3142,6 +3367,9 @@ abandon_carrier(Allctr_t *allctr, Carrier_t *crr) STAT_MBC_CPOOL_INSERT(allctr, crr); unlink_carrier(&allctr->mbc_list, crr); + if (crr->cpool.orig_allctr == allctr) { + link_abandoned_carrier(&allctr->cpool.pooled_list, crr); + } allctr->remove_mbc(allctr, crr); @@ -3274,6 +3502,15 @@ create_carrier(Allctr_t *allctr, Uint umem_sz, UWord flags) ASSERT(!(flags & CFLG_FORCE_MSEG && flags & CFLG_FORCE_SYS_ALLOC)); + if (umem_sz > (ERTS_UINT_MAX - ERTS_UINT_MAX/100)) { + /* Do an overly conservative _overflow_ check here so we don't + * have to deal with it from here on. I guess we could be more accurate + * but I don't think the need to allocate over 99% of the address space + * will ever arise on any machine, neither 32 nor 64 bit. + */ + return NULL; + } + blk_sz = UMEMSZ2BLKSZ(allctr, umem_sz); #ifdef ERTS_SMP @@ -3640,6 +3877,11 @@ destroy_carrier(Allctr_t *allctr, Block_t *blk, Carrier_t **busy_pcrr_pp) if (busy_pcrr_pp && *busy_pcrr_pp) { ERTS_ALC_CPOOL_ASSERT(*busy_pcrr_pp == crr); *busy_pcrr_pp = NULL; + ERTS_ALC_CPOOL_ASSERT(erts_smp_atomic_read_nob(&crr->allctr) + == (((erts_aint_t) allctr) + | ERTS_CRR_ALCTR_FLG_IN_POOL + | ERTS_CRR_ALCTR_FLG_BUSY)); + erts_smp_atomic_set_nob(&crr->allctr, ((erts_aint_t) allctr)); cpool_delete(allctr, allctr, crr); } else @@ -5519,6 +5761,10 @@ erts_alcu_start(Allctr_t *allctr, AllctrInit_t *init) allctr->min_block_size = sz; } + allctr->cpool.pooled_list.next = &allctr->cpool.pooled_list; + allctr->cpool.pooled_list.prev = &allctr->cpool.pooled_list; + allctr->cpool.traitor_list.next = &allctr->cpool.traitor_list; + allctr->cpool.traitor_list.prev = &allctr->cpool.traitor_list; allctr->cpool.dc_list.first = NULL; allctr->cpool.dc_list.last = NULL; allctr->cpool.abandon_limit = 0; diff --git a/erts/emulator/beam/erl_alloc_util.h b/erts/emulator/beam/erl_alloc_util.h index 7be6b1ed9d..eee920e66c 100644 --- a/erts/emulator/beam/erl_alloc_util.h +++ b/erts/emulator/beam/erl_alloc_util.h @@ -268,6 +268,11 @@ typedef union {char c[ERTS_ALLOC_ALIGN_BYTES]; long l; double d;} Unit_t; #ifdef ERTS_SMP +typedef struct ErtsDoubleLink_t_ { + struct ErtsDoubleLink_t_ *next; + struct ErtsDoubleLink_t_ *prev; +}ErtsDoubleLink_t; + typedef struct { erts_atomic_t next; erts_atomic_t prev; @@ -277,6 +282,7 @@ typedef struct { UWord abandon_limit; UWord blocks; UWord blocks_size; + ErtsDoubleLink_t abandoned; /* node in pooled_list or traitor_list */ } ErtsAlcCPoolData_t; #endif @@ -500,7 +506,12 @@ struct Allctr_t_ { CarrierList_t sbc_list; #ifdef ERTS_SMP struct { - CarrierList_t dc_list; + /* pooled_list, traitor list and dc_list contain only + carriers _created_ by this allocator */ + ErtsDoubleLink_t pooled_list; + ErtsDoubleLink_t traitor_list; + CarrierList_t dc_list; + UWord abandon_limit; int disable_abandon; int check_limit_count; diff --git a/erts/emulator/beam/erl_async.c b/erts/emulator/beam/erl_async.c index b3dc327704..decae6b2ca 100644 --- a/erts/emulator/beam/erl_async.c +++ b/erts/emulator/beam/erl_async.c @@ -292,7 +292,8 @@ static ERTS_INLINE void async_add(ErtsAsync *a, ErtsAsyncQ* q) if (DTRACE_ENABLED(aio_pool_add)) { DTRACE_CHARBUF(port_str, 16); - erts_snprintf(port_str, sizeof(port_str), "%T", a->port); + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), + "%T", a->port); /* DTRACE TODO: Get the queue length from erts_thr_q_enqueue() ? */ len = -1; DTRACE2(aio_pool_add, port_str, len); @@ -327,7 +328,8 @@ static ERTS_INLINE ErtsAsync *async_get(ErtsThrQ_t *q, if (DTRACE_ENABLED(aio_pool_get)) { DTRACE_CHARBUF(port_str, 16); - erts_snprintf(port_str, sizeof(port_str), "%T", a->port); + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), + "%T", a->port); /* DTRACE TODO: Get the length from erts_thr_q_dequeue() ? */ len = -1; DTRACE2(aio_pool_get, port_str, len); diff --git a/erts/emulator/beam/erl_bif_binary.c b/erts/emulator/beam/erl_bif_binary.c index ff775691b3..3bf78adce7 100644 --- a/erts/emulator/beam/erl_bif_binary.c +++ b/erts/emulator/beam/erl_bif_binary.c @@ -1324,9 +1324,9 @@ static int parse_match_opts_list(Eterm l, Eterm bin, Uint *posp, Uint *endp) goto badarg; } if (len < 0) { - Sint lentmp = -len; + Uint lentmp = -(Uint)len; /* overflow */ - if (lentmp == len || lentmp < 0 || -lentmp != len) { + if ((Sint)lentmp < 0) { goto badarg; } len = lentmp; @@ -1555,9 +1555,9 @@ BIF_RETTYPE erts_binary_part(Process *p, Eterm binary, Eterm epos, Eterm elen) goto badarg; } if (len < 0) { - Sint lentmp = -len; + Uint lentmp = -(Uint)len; /* overflow */ - if (lentmp == len || lentmp < 0 || -lentmp != len) { + if ((Sint)lentmp < 0) { goto badarg; } len = lentmp; @@ -1644,9 +1644,9 @@ BIF_RETTYPE erts_gc_binary_part(Process *p, Eterm *reg, Eterm live, int range_is goto badarg; } if (len < 0) { - Sint lentmp = -len; + Uint lentmp = -(Uint)len; /* overflow */ - if (lentmp == len || lentmp < 0 || -lentmp != len) { + if ((Sint)lentmp < 0) { goto badarg; } len = lentmp; @@ -2213,9 +2213,9 @@ static BIF_RETTYPE binary_bin_to_list_common(Process *p, goto badarg; } if (len < 0) { - Sint lentmp = -len; + Uint lentmp = -(Uint)len; /* overflow */ - if (lentmp == len || lentmp < 0 || -lentmp != len) { + if ((Sint)lentmp < 0) { goto badarg; } len = lentmp; @@ -2294,18 +2294,11 @@ BIF_RETTYPE binary_bin_to_list_1(BIF_ALIST_1) BIF_ERROR(BIF_P,BADARG); } -/* - * Ok, erlang:list_to_binary does not interrupt, and we really don't want - * an alternative implementation for the exact same thing, why we - * have descided to use the old non-restarting implementation for now. - * In reality, there are seldom many iterations involved in doing this, so the - * problem of long-running bifs is not really that big in this case. - * So, for now we use the old implementation also in the module binary. - */ +HIPE_WRAPPER_BIF_DISABLE_GC(binary_list_to_bin, 1) BIF_RETTYPE binary_list_to_bin_1(BIF_ALIST_1) { - return erts_list_to_binary_bif(BIF_P, BIF_ARG_1); + return erts_list_to_binary_bif(BIF_P, BIF_ARG_1, bif_export[BIF_binary_list_to_bin_1]); } typedef struct { diff --git a/erts/emulator/beam/erl_bif_ddll.c b/erts/emulator/beam/erl_bif_ddll.c index 1728b200f7..56cd2ba04f 100644 --- a/erts/emulator/beam/erl_bif_ddll.c +++ b/erts/emulator/beam/erl_bif_ddll.c @@ -1548,8 +1548,10 @@ static int do_load_driver_entry(DE_Handle *dh, char *path, char *name) switch (dp->extended_marker) { case ERL_DRV_EXTENDED_MARKER: - if (ERL_DRV_EXTENDED_MAJOR_VERSION != dp->major_version - || ERL_DRV_EXTENDED_MINOR_VERSION < dp->minor_version) { + if (dp->major_version < ERL_DRV_MIN_REQUIRED_MAJOR_VERSION_ON_LOAD + || (ERL_DRV_EXTENDED_MAJOR_VERSION < dp->major_version + || (ERL_DRV_EXTENDED_MAJOR_VERSION == dp->major_version + && ERL_DRV_EXTENDED_MINOR_VERSION < dp->minor_version))) { /* Incompatible driver version */ res = ERL_DE_LOAD_ERROR_INCORRECT_VERSION; goto error; diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index 2adba9b240..b90362d82c 100755..100644 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -27,6 +27,7 @@ #include "erl_process.h" #include "error.h" #include "erl_driver.h" +#include "erl_nif.h" #include "bif.h" #include "big.h" #include "erl_version.h" @@ -90,7 +91,7 @@ static char erts_system_version[] = ("Erlang/OTP " ERLANG_OTP_RELEASE " [smp:%beu:%beu]" #endif #ifdef USE_THREADS -#ifdef ERTS_DIRTY_SCHEDULERS +#if defined(ERTS_DIRTY_SCHEDULERS) && defined(ERTS_SMP) " [ds:%beu:%beu:%beu]" #endif " [async-threads:%d]" @@ -2459,6 +2460,13 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) ERL_DRV_EXTENDED_MINOR_VERSION); hp = HAlloc(BIF_P, 2*n); BIF_RET(buf_to_intlist(&hp, buf, n, NIL)); + } else if (ERTS_IS_ATOM_STR("nif_version", BIF_ARG_1)) { + char buf[42]; + int n = erts_snprintf(buf, 42, "%d.%d", + ERL_NIF_MAJOR_VERSION, + ERL_NIF_MINOR_VERSION); + hp = HAlloc(BIF_P, 2*n); + BIF_RET(buf_to_intlist(&hp, buf, n, NIL)); } else if (ERTS_IS_ATOM_STR("smp_support", BIF_ARG_1)) { #ifdef ERTS_SMP BIF_RET(am_true); @@ -2691,6 +2699,14 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) else if (ERTS_IS_ATOM_STR("ets_limit",BIF_ARG_1)) { BIF_RET(make_small(erts_db_get_max_tabs())); } + else if (ERTS_IS_ATOM_STR("tolerant_timeofday",BIF_ARG_1)) { + BIF_RET(erts_disable_tolerant_timeofday + ? am_disabled + : am_enabled); + } + else if (ERTS_IS_ATOM_STR("eager_check_io",BIF_ARG_1)) { + BIF_RET(erts_eager_check_io ? am_true : am_false); + } BIF_ERROR(BIF_P, BADARG); } @@ -3050,6 +3066,25 @@ fun_info_2(BIF_ALIST_2) return TUPLE2(hp, what, val); } +BIF_RETTYPE +fun_info_mfa_1(BIF_ALIST_1) +{ + Process* p = BIF_P; + Eterm fun = BIF_ARG_1; + Eterm* hp; + + if (is_fun(fun)) { + ErlFunThing* funp = (ErlFunThing *) fun_val(fun); + hp = HAlloc(p, 4); + BIF_RET(TUPLE3(hp,funp->fe->module,funp->fe->address[-2],make_small(funp->arity))); + } else if (is_export(fun)) { + Export* exp = (Export *) ((UWord) (export_val(fun))[1]); + hp = HAlloc(p, 4); + BIF_RET(TUPLE3(hp,exp->code[0],exp->code[1],make_small(exp->code[2]))); + } + BIF_ERROR(p, BADARG); +} + BIF_RETTYPE is_process_alive_1(BIF_ALIST_1) { if(is_internal_pid(BIF_ARG_1)) { @@ -3280,17 +3315,38 @@ BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1) BIF_RET(make_small((Uint) words)); } else if (ERTS_IS_ATOM_STR("check_io_debug", BIF_ARG_1)) { - /* Used by (emulator) */ - int res; + /* Used by driver_SUITE (emulator) */ + Uint sz, *szp; + Eterm res, *hp, **hpp; + int no_errors; + ErtsCheckIoDebugInfo ciodi = {0}; #ifdef HAVE_ERTS_CHECK_IO_DEBUG erts_smp_proc_unlock(BIF_P,ERTS_PROC_LOCK_MAIN); - res = erts_check_io_debug(); + no_errors = erts_check_io_debug(&ciodi); erts_smp_proc_lock(BIF_P,ERTS_PROC_LOCK_MAIN); #else - res = 0; + no_errors = 0; #endif - ASSERT(res >= 0); - BIF_RET(erts_make_integer((Uint) res, BIF_P)); + sz = 0; + szp = &sz; + hpp = NULL; + while (1) { + res = erts_bld_tuple(hpp, szp, 4, + erts_bld_uint(hpp, szp, + (Uint) no_errors), + erts_bld_uint(hpp, szp, + (Uint) ciodi.no_used_fds), + erts_bld_uint(hpp, szp, + (Uint) ciodi.no_driver_select_structs), + erts_bld_uint(hpp, szp, + (Uint) ciodi.no_driver_event_structs)); + if (hpp) + break; + hp = HAlloc(BIF_P, sz); + szp = NULL; + hpp = &hp; + } + BIF_RET(res); } else if (ERTS_IS_ATOM_STR("process_info_args", BIF_ARG_1)) { /* Used by process_SUITE (emulator) */ @@ -3851,16 +3907,19 @@ static Eterm lcnt_build_lock_stats_term(Eterm **hpp, Uint *szp, erts_lcnt_lock_s Uint tries = 0, colls = 0; unsigned long timer_s = 0, timer_ns = 0, timer_n = 0; unsigned int line = 0; + unsigned int i; Eterm af, uil; Eterm uit, uic; Eterm uits, uitns, uitn; Eterm tt, tstat, tloc, t; + Eterm thist, vhist[ERTS_LCNT_HISTOGRAM_SLOT_SIZE]; /* term: - * [{{file, line}, {tries, colls, {seconds, nanoseconds, n_blocks}}}] + * [{{file, line}, {tries, colls, {seconds, nanoseconds, n_blocks}}, + * { .. histogram .. }] */ - + tries = (Uint) ethr_atomic_read(&stats->tries); colls = (Uint) ethr_atomic_read(&stats->colls); @@ -3869,23 +3928,27 @@ static Eterm lcnt_build_lock_stats_term(Eterm **hpp, Uint *szp, erts_lcnt_lock_s timer_ns = stats->timer.ns; timer_n = stats->timer_n; - af = erts_atom_put(stats->file, strlen(stats->file), ERTS_ATOM_ENC_LATIN1, 1); + af = erts_atom_put((byte *)stats->file, strlen(stats->file), ERTS_ATOM_ENC_LATIN1, 1); uil = erts_bld_uint( hpp, szp, line); tloc = erts_bld_tuple(hpp, szp, 2, af, uil); - uit = erts_bld_uint( hpp, szp, tries); - uic = erts_bld_uint( hpp, szp, colls); - + uit = erts_bld_uint( hpp, szp, tries); + uic = erts_bld_uint( hpp, szp, colls); + uits = erts_bld_uint( hpp, szp, timer_s); uitns = erts_bld_uint( hpp, szp, timer_ns); uitn = erts_bld_uint( hpp, szp, timer_n); tt = erts_bld_tuple(hpp, szp, 3, uits, uitns, uitn); tstat = erts_bld_tuple(hpp, szp, 3, uit, uic, tt); - - t = erts_bld_tuple(hpp, szp, 2, tloc, tstat); - - res = erts_bld_cons( hpp, szp, t, res); + + for(i = 0; i < ERTS_LCNT_HISTOGRAM_SLOT_SIZE; i++) { + vhist[i] = erts_bld_uint(hpp, szp, stats->hist.ns[i]); + } + thist = erts_bld_tuplev(hpp, szp, ERTS_LCNT_HISTOGRAM_SLOT_SIZE, vhist); + + t = erts_bld_tuple(hpp, szp, 3, tloc, tstat, thist); + res = erts_bld_cons( hpp, szp, t, res); return res; } @@ -3906,13 +3969,13 @@ static Eterm lcnt_build_lock_term(Eterm **hpp, Uint *szp, erts_lcnt_lock_t *lock ASSERT(ltype); - type = erts_atom_put(ltype, strlen(ltype), ERTS_ATOM_ENC_LATIN1, 1); - name = erts_atom_put(lock->name, strlen(lock->name), ERTS_ATOM_ENC_LATIN1, 1); + type = erts_atom_put((byte *)ltype, strlen(ltype), ERTS_ATOM_ENC_LATIN1, 1); + name = erts_atom_put((byte *)lock->name, strlen(lock->name), ERTS_ATOM_ENC_LATIN1, 1); if (lock->flag & ERTS_LCNT_LT_ALLOC) { /* use allocator types names as id's for allocator locks */ ltype = (char *) ERTS_ALC_A2AD(signed_val(lock->id)); - id = erts_atom_put(ltype, strlen(ltype), ERTS_ATOM_ENC_LATIN1, 1); + id = erts_atom_put((byte *)ltype, strlen(ltype), ERTS_ATOM_ENC_LATIN1, 1); } else if (lock->flag & ERTS_LCNT_LT_PROCLOCK) { /* use registered names as id's for process locks if available */ proc = erts_proc_lookup(lock->id); @@ -3923,16 +3986,15 @@ static Eterm lcnt_build_lock_term(Eterm **hpp, Uint *szp, erts_lcnt_lock_t *lock id = lock->id; } } else { - id = lock->id; + id = lock->id; } - + for (i = 0; i < lock->n_stats; i++) { stats = lcnt_build_lock_stats_term(hpp, szp, &(lock->stats[i]), stats); } - - t = erts_bld_tuple(hpp, szp, 4, name, id, type, stats); - - res = erts_bld_cons( hpp, szp, t, res); + + t = erts_bld_tuple(hpp, szp, 4, name, id, type, stats); + res = erts_bld_cons( hpp, szp, t, res); return res; } @@ -3952,12 +4014,12 @@ static Eterm lcnt_build_result_term(Eterm **hpp, Uint *szp, erts_lcnt_data_t *da dtns = erts_bld_uint( hpp, szp, data->duration.ns); tdt = erts_bld_tuple(hpp, szp, 2, dts, dtns); - adur = erts_atom_put(str_duration, strlen(str_duration), ERTS_ATOM_ENC_LATIN1, 1); + adur = erts_atom_put((byte *)str_duration, strlen(str_duration), ERTS_ATOM_ENC_LATIN1, 1); tdur = erts_bld_tuple(hpp, szp, 2, adur, tdt); /* lock tuple */ - aloc = erts_atom_put(str_locks, strlen(str_locks), ERTS_ATOM_ENC_LATIN1, 1); + aloc = erts_atom_put((byte *)str_locks, strlen(str_locks), ERTS_ATOM_ENC_LATIN1, 1); for (lock = data->current_locks->head; lock != NULL ; lock = lock->next ) { lloc = lcnt_build_lock_term(hpp, szp, lock, lloc); diff --git a/erts/emulator/beam/erl_bif_port.c b/erts/emulator/beam/erl_bif_port.c index 77627a6897..7ce950e090 100644 --- a/erts/emulator/beam/erl_bif_port.c +++ b/erts/emulator/beam/erl_bif_port.c @@ -472,7 +472,7 @@ cleanup_old_port_data(erts_aint_t data) ErtsPortDataHeap *pdhp = (ErtsPortDataHeap *) data; size_t size; ERTS_SMP_DATA_DEPENDENCY_READ_MEMORY_BARRIER; - size = sizeof(ErtsPortDataHeap) + pdhp->hsize*(sizeof(Eterm) - 1); + size = sizeof(ErtsPortDataHeap) + (pdhp->hsize-1)*sizeof(Eterm); erts_schedule_thr_prgr_later_cleanup_op(free_port_data_heap, (void *) pdhp, &pdhp->later_op, @@ -493,8 +493,8 @@ void erts_cleanup_port_data(Port *prt) { ASSERT(erts_atomic32_read_nob(&prt->state) & ERTS_PORT_SFLGS_INVALID_LOOKUP); - cleanup_old_port_data(erts_smp_atomic_read_nob(&prt->data)); - erts_smp_atomic_set_nob(&prt->data, (erts_aint_t) THE_NON_VALUE); + cleanup_old_port_data(erts_smp_atomic_xchg_nob(&prt->data, + (erts_aint_t) NULL)); } Uint @@ -508,7 +508,7 @@ erts_port_data_size(Port *prt) } else { ErtsPortDataHeap *pdhp = (ErtsPortDataHeap *) data; - return (Uint) sizeof(ErtsPortDataHeap) + pdhp->hsize*(sizeof(Eterm)-1); + return (Uint) sizeof(ErtsPortDataHeap) + (pdhp->hsize-1)*sizeof(Eterm); } } @@ -550,10 +550,11 @@ BIF_RETTYPE port_set_data_2(BIF_ALIST_2) hsize = size_object(BIF_ARG_2); pdhp = erts_alloc(ERTS_ALC_T_PORT_DATA_HEAP, - sizeof(ErtsPortDataHeap) + hsize*(sizeof(Eterm)-1)); + sizeof(ErtsPortDataHeap) + (hsize-1)*sizeof(Eterm)); hp = &pdhp->heap[0]; pdhp->off_heap.first = NULL; pdhp->off_heap.overhead = 0; + pdhp->hsize = hsize; pdhp->data = copy_struct(BIF_ARG_2, hsize, &hp, &pdhp->off_heap); data = (erts_aint_t) pdhp; ASSERT((data & 0x3) == 0); @@ -561,8 +562,14 @@ BIF_RETTYPE port_set_data_2(BIF_ALIST_2) data = erts_smp_atomic_xchg_wb(&prt->data, data); + if (data == (erts_aint_t)NULL) { + /* Port terminated by racing thread */ + data = erts_smp_atomic_xchg_wb(&prt->data, data); + ASSERT(data != (erts_aint_t)NULL); + cleanup_old_port_data(data); + BIF_ERROR(BIF_P, BADARG); + } cleanup_old_port_data(data); - BIF_RET(am_true); } @@ -581,6 +588,8 @@ BIF_RETTYPE port_get_data_1(BIF_ALIST_1) BIF_ERROR(BIF_P, BADARG); data = erts_smp_atomic_read_ddrb(&prt->data); + if (data == (erts_aint_t)NULL) + BIF_ERROR(BIF_P, BADARG); /* Port terminated by racing thread */ if ((data & 0x3) != 0) { res = (Eterm) (UWord) data; @@ -882,7 +891,7 @@ open_port(Process* p, Eterm name, Eterm settings, int *err_typep, int *err_nump) DTRACE_CHARBUF(port_str, DTRACE_TERM_BUF_SIZE); dtrace_proc_str(p, process_str); - erts_snprintf(port_str, sizeof(port_str), "%T", port->common.id); + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", port->common.id); DTRACE3(port_open, process_str, name_buf, port_str); } #endif diff --git a/erts/emulator/beam/erl_binary.h b/erts/emulator/beam/erl_binary.h index 819b19e566..06dfeb1260 100644 --- a/erts/emulator/beam/erl_binary.h +++ b/erts/emulator/beam/erl_binary.h @@ -166,7 +166,7 @@ Eterm erts_bin_bytes_to_list(Eterm previous, Eterm* hp, byte* bytes, Uint size, * Common implementation for erlang:list_to_binary/1 and binary:list_to_bin/1 */ -BIF_RETTYPE erts_list_to_binary_bif(Process *p, Eterm arg); +BIF_RETTYPE erts_list_to_binary_bif(Process *p, Eterm arg, Export *bif); BIF_RETTYPE erts_gc_binary_part(Process *p, Eterm *reg, Eterm live, int range_is_tuple); BIF_RETTYPE erts_binary_part(Process *p, Eterm binary, Eterm epos, Eterm elen); @@ -236,6 +236,8 @@ erts_bin_drv_alloc_fnf(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; void *res; + if (bsize < size) /* overflow */ + return NULL; res = erts_alloc_fnf(ERTS_ALC_T_DRV_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); return (Binary *) res; @@ -246,6 +248,8 @@ erts_bin_drv_alloc(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; void *res; + if (bsize < size) /* overflow */ + erts_alloc_enomem(ERTS_ALC_T_DRV_BINARY, size); res = erts_alloc(ERTS_ALC_T_DRV_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); return (Binary *) res; @@ -257,6 +261,8 @@ erts_bin_nrml_alloc(Uint size) { Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; void *res; + if (bsize < size) /* overflow */ + erts_alloc_enomem(ERTS_ALC_T_BINARY, size); res = erts_alloc(ERTS_ALC_T_BINARY, bsize); ERTS_CHK_BIN_ALIGNMENT(res); return (Binary *) res; @@ -267,11 +273,12 @@ erts_bin_realloc_fnf(Binary *bp, Uint size) { Binary *nbp; Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; + ErtsAlcType_t type = (bp->flags & BIN_FLAG_DRV) ? ERTS_ALC_T_DRV_BINARY + : ERTS_ALC_T_BINARY; ASSERT((bp->flags & BIN_FLAG_MAGIC) == 0); - if (bp->flags & BIN_FLAG_DRV) - nbp = erts_realloc_fnf(ERTS_ALC_T_DRV_BINARY, (void *) bp, bsize); - else - nbp = erts_realloc_fnf(ERTS_ALC_T_BINARY, (void *) bp, bsize); + if (bsize < size) /* overflow */ + return NULL; + nbp = erts_realloc_fnf(type, (void *) bp, bsize); ERTS_CHK_BIN_ALIGNMENT(nbp); return nbp; } @@ -281,17 +288,14 @@ erts_bin_realloc(Binary *bp, Uint size) { Binary *nbp; Uint bsize = ERTS_SIZEOF_Binary(size) + CHICKEN_PAD; + ErtsAlcType_t type = (bp->flags & BIN_FLAG_DRV) ? ERTS_ALC_T_DRV_BINARY + : ERTS_ALC_T_BINARY; ASSERT((bp->flags & BIN_FLAG_MAGIC) == 0); - if (bp->flags & BIN_FLAG_DRV) - nbp = erts_realloc_fnf(ERTS_ALC_T_DRV_BINARY, (void *) bp, bsize); - else - nbp = erts_realloc_fnf(ERTS_ALC_T_BINARY, (void *) bp, bsize); + if (bsize < size) /* overflow */ + erts_realloc_enomem(type, bp, size); + nbp = erts_realloc_fnf(type, (void *) bp, bsize); if (!nbp) - erts_realloc_n_enomem(ERTS_ALC_T2N(bp->flags & BIN_FLAG_DRV - ? ERTS_ALC_T_DRV_BINARY - : ERTS_ALC_T_BINARY), - bp, - bsize); + erts_realloc_enomem(type, bp, bsize); ERTS_CHK_BIN_ALIGNMENT(nbp); return nbp; } @@ -312,6 +316,7 @@ erts_create_magic_binary(Uint size, void (*destructor)(Binary *)) { Uint bsize = ERTS_MAGIC_BIN_SIZE(size); Binary* bptr = erts_alloc_fnf(ERTS_ALC_T_BINARY, bsize); + ASSERT(bsize > size); if (!bptr) erts_alloc_n_enomem(ERTS_ALC_T2N(ERTS_ALC_T_BINARY), bsize); ERTS_CHK_BIN_ALIGNMENT(bptr); diff --git a/erts/emulator/beam/erl_db.c b/erts/emulator/beam/erl_db.c index a5d67571e2..8f246ffa07 100644 --- a/erts/emulator/beam/erl_db.c +++ b/erts/emulator/beam/erl_db.c @@ -259,10 +259,11 @@ static void schedule_free_dbtable(DbTable* tb) /* * NON-SMP case: Caller is *not* allowed to access the *tb * structure after this function has returned! - * SMP case: Caller is allowed to access the *tb structure - * until the bif has returned (we typically - * need to unlock the table lock after this - * function has returned). + * SMP case: Caller is allowed to access the *common* part of the *tb + * structure until the bif has returned (we typically need to + * unlock the table lock after this function has returned). + * Caller is *not* allowed to access the specialized part + * (hash or tree) of *tb after this function has returned. */ ASSERT(erts_refc_read(&tb->common.ref, 0) == 0); erts_schedule_thr_prgr_later_cleanup_op(free_dbtable, @@ -3279,34 +3280,37 @@ erts_db_process_exiting(Process *c_p, ErtsProcLocks c_p_locks) } erts_smp_rwmtx_runlock(mmtl); if (tb) { - int reds; - DbFixation** pp; + int reds = 0; db_lock(tb, LCK_WRITE_REC); - #ifdef ERTS_SMP - erts_smp_mtx_lock(&tb->common.fixlock); - #endif - reds = 10; - - for (pp = &tb->common.fixations; *pp != NULL; - pp = &(*pp)->next) { - if ((*pp)->pid == pid) { - DbFixation* fix = *pp; - erts_aint_t diff = -((erts_aint_t) fix->counter); - erts_refc_add(&tb->common.ref,diff,0); - *pp = fix->next; - erts_db_free(ERTS_ALC_T_DB_FIXATION, - tb, fix, sizeof(DbFixation)); - ERTS_ETS_MISC_MEM_ADD(-sizeof(DbFixation)); - break; + if (!(tb->common.status & DB_DELETE)) { + DbFixation** pp; + + #ifdef ERTS_SMP + erts_smp_mtx_lock(&tb->common.fixlock); + #endif + reds = 10; + + for (pp = &tb->common.fixations; *pp != NULL; + pp = &(*pp)->next) { + if ((*pp)->pid == pid) { + DbFixation* fix = *pp; + erts_aint_t diff = -((erts_aint_t) fix->counter); + erts_refc_add(&tb->common.ref,diff,0); + *pp = fix->next; + erts_db_free(ERTS_ALC_T_DB_FIXATION, + tb, fix, sizeof(DbFixation)); + ERTS_ETS_MISC_MEM_ADD(-sizeof(DbFixation)); + break; + } + } + #ifdef ERTS_SMP + erts_smp_mtx_unlock(&tb->common.fixlock); + #endif + if (!IS_FIXED(tb) && IS_HASH_TABLE(tb->common.status)) { + db_unfix_table_hash(&(tb->hash)); + reds += 40; } - } - #ifdef ERTS_SMP - erts_smp_mtx_unlock(&tb->common.fixlock); - #endif - if (!IS_FIXED(tb) && IS_HASH_TABLE(tb->common.status)) { - db_unfix_table_hash(&(tb->hash)); - reds += 40; } db_unlock(tb, LCK_WRITE_REC); BUMP_REDS(c_p, reds); diff --git a/erts/emulator/beam/erl_db_tree.c b/erts/emulator/beam/erl_db_tree.c index 25029ba90f..a62a83a928 100644 --- a/erts/emulator/beam/erl_db_tree.c +++ b/erts/emulator/beam/erl_db_tree.c @@ -485,7 +485,7 @@ static int db_first_tree(Process *p, DbTable *tbl, Eterm *ret) *ret = am_EOT; return DB_ERROR_NONE; } - /* Walk down to the tree to the left */ + /* Walk down the tree to the left */ if ((stack = get_static_stack(tb)) != NULL) { stack->pos = stack->slot = 0; } @@ -531,7 +531,7 @@ static int db_last_tree(Process *p, DbTable *tbl, Eterm *ret) *ret = am_EOT; return DB_ERROR_NONE; } - /* Walk down to the tree to the left */ + /* Walk down the tree to the right */ if ((stack = get_static_stack(tb)) != NULL) { stack->pos = stack->slot = 0; } diff --git a/erts/emulator/beam/erl_driver.h b/erts/emulator/beam/erl_driver.h index 5517c26ba4..e498ac70ec 100644 --- a/erts/emulator/beam/erl_driver.h +++ b/erts/emulator/beam/erl_driver.h @@ -133,7 +133,23 @@ typedef struct { #define ERL_DRV_EXTENDED_MARKER (0xfeeeeeed) #define ERL_DRV_EXTENDED_MAJOR_VERSION 3 -#define ERL_DRV_EXTENDED_MINOR_VERSION 0 +#define ERL_DRV_EXTENDED_MINOR_VERSION 2 + +/* + * The emulator will refuse to load a driver with a major version + * lower than ERL_DRV_MIN_REQUIRED_MAJOR_VERSION_ON_LOAD. The load + * may however fail if user have not removed use of deprecated + * symbols. + * + * The ERL_DRV_MIN_REQUIRED_MAJOR_VERSION_ON_LOAD have to allow + * loading of drivers built at least two major OTP releases + * ago. + * + * Bump of major version to 3 happened in OTP 17. That is, in + * OTP 19 we can increase ERL_DRV_MIN_REQUIRED_MAJOR_VERSION_ON_LOAD + * to 3. + */ +#define ERL_DRV_MIN_REQUIRED_MAJOR_VERSION_ON_LOAD 2 /* * The emulator will refuse to load a driver with different major @@ -182,7 +198,7 @@ typedef long long ErlDrvSInt64; #error No 64-bit integer type #endif -#if defined(__WIN32__) +#if defined(__WIN32__) || defined(_WIN32) typedef ErlDrvUInt ErlDrvSizeT; typedef ErlDrvSInt ErlDrvSSizeT; #else @@ -345,6 +361,9 @@ typedef struct erl_drv_entry { /* Called on behalf of driver_select when it is safe to release 'event'. A typical unix driver would call close(event) */ + void (*emergency_close)(ErlDrvData drv_data); + /* called when the port is closed abruptly. + specifically when erl_crash_dump is called. */ /* When adding entries here, dont forget to pad in obsolete/driver.h */ } ErlDrvEntry; diff --git a/erts/emulator/beam/erl_drv_nif.h b/erts/emulator/beam/erl_drv_nif.h index 3f829ea7ea..4e8c6dc68b 100644 --- a/erts/emulator/beam/erl_drv_nif.h +++ b/erts/emulator/beam/erl_drv_nif.h @@ -35,6 +35,7 @@ typedef struct { int scheduler_threads; int nif_major_version; int nif_minor_version; + int dirty_scheduler_support; } ErlDrvSysInfo; typedef struct { diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c index aa15d2cc57..0db42d4325 100644 --- a/erts/emulator/beam/erl_gc.c +++ b/erts/emulator/beam/erl_gc.c @@ -2018,6 +2018,20 @@ setup_rootset(Process *p, Eterm *objv, int nobj, Rootset *rootset) roots[n].sz = 1; n++; } + + /* + * If a NIF has saved arguments, they need to be added + */ + if (ERTS_PROC_GET_NIF_TRAP_EXPORT(p)) { + Eterm* argv; + int argc; + if (erts_setup_nif_gc(p, &argv, &argc)) { + roots[n].v = argv; + roots[n].sz = argc; + n++; + } + } + ASSERT(n <= rootset->size); mp = p->msg.first; diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c index d54658f1ea..61f8385efc 100644 --- a/erts/emulator/beam/erl_init.c +++ b/erts/emulator/beam/erl_init.c @@ -324,7 +324,6 @@ erl_init(int ncpu, BIN_VH_MIN_SIZE = erts_next_heap_size(BIN_VH_MIN_SIZE, 0); erts_init_trace(); - erts_init_binary(); erts_init_bits(); erts_code_ix_init(); erts_init_fun_table(); @@ -337,6 +336,7 @@ erl_init(int ncpu, erts_ddll_init(); init_emulator(); erts_ptab_init(); /* Must be after init_emulator() */ + erts_init_binary(); /* Must be after init_emulator() */ erts_bp_init(); init_db(); /* Must be after init_emulator */ erts_bif_timer_init(); @@ -548,6 +548,8 @@ void erts_usage(void) erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); erts_fprintf(stderr, "-sct cput set cpu topology,\n"); erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); + erts_fprintf(stderr, "-secio bool enable/disable eager check I/O scheduling,\n"); + erts_fprintf(stderr, " see the erl(1) documentation for more info.\n"); #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT erts_fprintf(stderr, "-sub bool enable/disable scheduler utilization balancing,\n"); #else @@ -1674,6 +1676,22 @@ erl_start(int argc, char **argv) erts_usage(); } } + else if (has_prefix("ecio", sub_param)) { + arg = get_arg(sub_param+4, argv[i+1], &i); +#ifndef __OSE__ + if (sys_strcmp("true", arg) == 0) + erts_eager_check_io = 1; + else +#endif + if (sys_strcmp("false", arg) == 0) + erts_eager_check_io = 0; + else { + erts_fprintf(stderr, + "bad schedule eager check I/O value '%s'\n", + arg); + erts_usage(); + } + } else if (has_prefix("pp", sub_param)) { arg = get_arg(sub_param+2, argv[i+1], &i); if (sys_strcmp(arg, "true") == 0) @@ -2066,8 +2084,10 @@ erl_exit_vv(int n, int flush_async, char *fmt, va_list args1, va_list args2) system_cleanup(flush_async); save_statistics(); - - an = abs(n); + if (n < 0) + an = -(unsigned int)n; + else + an = n; if (erts_mtrace_enabled) erts_mtrace_exit((Uint32) an); diff --git a/erts/emulator/beam/erl_lock_check.c b/erts/emulator/beam/erl_lock_check.c index 7e3a90779d..b105ece6f1 100644 --- a/erts/emulator/beam/erl_lock_check.c +++ b/erts/emulator/beam/erl_lock_check.c @@ -139,7 +139,6 @@ static erts_lc_lock_order_t erts_lock_order[] = { { "drv_tsd", NULL }, { "async_enq_mtx", NULL }, #ifdef ERTS_SMP - { "sys_msg_q", NULL }, { "atom_tab", NULL }, { "make_ref", NULL }, { "misc_op_list_pre_alloc_lock", "address" }, @@ -148,6 +147,7 @@ static erts_lc_lock_order_t erts_lock_order[] = { { "btm_pre_alloc_lock", NULL, }, { "dist_entry_out_queue", "address" }, { "port_sched_lock", "port_id" }, + { "sys_msg_q", NULL }, { "port_table", NULL }, #endif { "mtrace_op", NULL }, @@ -227,8 +227,7 @@ rw_op_str(Uint16 flags) case ERTS_LC_FLG_LO_READ: return " (r)"; case ERTS_LC_FLG_LO_WRITE: - erts_fprintf(stderr, "\nInternal error\n"); - lc_abort(); + ERTS_INTERNAL_ERROR("Only write flag present"); default: break; } @@ -270,9 +269,9 @@ union erts_lc_free_block_t_ { static ethr_tsd_key locks_key; -static erts_lc_locked_locks_t *erts_locked_locks; +static erts_lc_locked_locks_t *erts_locked_locks = NULL; -static erts_lc_free_block_t *free_blocks; +static erts_lc_free_block_t *free_blocks = NULL; #ifdef ERTS_LC_STATIC_ALLOC #define ERTS_LC_FB_CHUNK_SIZE 10000 @@ -311,8 +310,7 @@ static ERTS_INLINE void lc_free(void *p) static void *lc_core_alloc(void) { lc_unlock(); - erts_fprintf(stderr, "Lock checker out of memory!\n"); - lc_abort(); + ERTS_INTERNAL_ERROR("Lock checker out of memory!\n"); } #else @@ -325,8 +323,7 @@ static void *lc_core_alloc(void) fbs = (erts_lc_free_block_t *) malloc(sizeof(erts_lc_free_block_t) * ERTS_LC_FB_CHUNK_SIZE); if (!fbs) { - erts_fprintf(stderr, "Lock checker failed to allocate memory!\n"); - lc_abort(); + ERTS_INTERNAL_ERROR("Lock checker failed to allocate memory!"); } for (i = 1; i < ERTS_LC_FB_CHUNK_SIZE - 1; i++) { #ifdef DEBUG @@ -366,11 +363,11 @@ create_locked_locks(char *thread_name) { erts_lc_locked_locks_t *l_lcks = malloc(sizeof(erts_lc_locked_locks_t)); if (!l_lcks) - lc_abort(); + ERTS_INTERNAL_ERROR("Lock checker failed to allocate memory!"); l_lcks->thread_name = strdup(thread_name ? thread_name : "unknown"); if (!l_lcks->thread_name) - lc_abort(); + ERTS_INTERNAL_ERROR("Lock checker failed to allocate memory!"); l_lcks->emu_thread = 0; l_lcks->tid = erts_thr_self(); @@ -691,7 +688,7 @@ erts_lc_set_thread_name(char *thread_name) free((void *) l_lcks->thread_name); l_lcks->thread_name = strdup(thread_name ? thread_name : "unknown"); if (!l_lcks->thread_name) - lc_abort(); + ERTS_INTERNAL_ERROR("strdup failed"); } l_lcks->emu_thread = 1; } @@ -1330,7 +1327,7 @@ erts_lc_init(void) #endif /* #ifdef ERTS_LC_STATIC_ALLOC */ if (ethr_spinlock_init(&free_blocks_lock) != 0) - lc_abort(); + ERTS_INTERNAL_ERROR("spinlock_init failed"); erts_tsd_key_create(&locks_key,"erts_lock_check_key"); } diff --git a/erts/emulator/beam/erl_lock_count.c b/erts/emulator/beam/erl_lock_count.c index 6f44bf097b..cf6996ea06 100644 --- a/erts/emulator/beam/erl_lock_count.c +++ b/erts/emulator/beam/erl_lock_count.c @@ -61,6 +61,25 @@ static ERTS_INLINE void lcnt_unlock(void) { ethr_mutex_unlock(&lcnt_data_lock); } +const int log2_tab64[64] = { + 63, 0, 58, 1, 59, 47, 53, 2, + 60, 39, 48, 27, 54, 33, 42, 3, + 61, 51, 37, 40, 49, 18, 28, 20, + 55, 30, 34, 11, 43, 14, 22, 4, + 62, 57, 46, 52, 38, 26, 32, 41, + 50, 36, 17, 19, 29, 10, 13, 21, + 56, 45, 25, 31, 35, 16, 9, 12, + 44, 24, 15, 8, 23, 7, 6, 5}; + +static ERTS_INLINE int lcnt_log2(Uint64 v) { + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + v |= v >> 32; + return log2_tab64[((Uint64)((v - (v >> 1))*0x07EDD5E59A4E28C2)) >> 58]; +} static char* lcnt_lock_type(Uint16 flag) { switch(flag & ERTS_LCNT_LT_ALL) { @@ -81,19 +100,20 @@ static void lcnt_clear_stats(erts_lcnt_lock_stats_t *stats) { stats->timer_n = 0; stats->file = (char *)str_undefined; stats->line = 0; + sys_memzero(stats->hist.ns, sizeof(stats->hist.ns)); } static void lcnt_time(erts_lcnt_time_t *time) { -#ifdef HAVE_GETHRTIME +#if 0 || defined(HAVE_GETHRTIME) SysHrTime hr_time; hr_time = sys_gethrtime(); time->s = (unsigned long)(hr_time / 1000000000LL); time->ns = (unsigned long)(hr_time - 1000000000LL*time->s); -#else - SysTimeval tv; - sys_gettimeofday(&tv); - time->s = tv.tv_sec; - time->ns = tv.tv_usec*1000LL; +#else + SysTimeval tv; + sys_gettimeofday(&tv); + time->s = tv.tv_sec; + time->ns = tv.tv_usec*1000LL; #endif } @@ -111,28 +131,29 @@ static void lcnt_time_diff(erts_lcnt_time_t *d, erts_lcnt_time_t *t1, erts_lcnt_ dns += 1000000000LL; } + ASSERT(ds >= 0); + d->s = ds; d->ns = dns; } -/* difference d must be positive */ +/* difference d must be non-negative */ static void lcnt_time_add(erts_lcnt_time_t *t, erts_lcnt_time_t *d) { - unsigned long ngns = 0; - t->s += d->s; t->ns += d->ns; - ngns = t->ns / 1000000000LL; + t->s += t->ns / 1000000000LL; t->ns = t->ns % 1000000000LL; - - t->s += ngns; } static erts_lcnt_thread_data_t *lcnt_thread_data_alloc(void) { erts_lcnt_thread_data_t *eltd; eltd = (erts_lcnt_thread_data_t*)malloc(sizeof(erts_lcnt_thread_data_t)); + if (!eltd) { + ERTS_INTERNAL_ERROR("Lock counter failed to allocate memory!"); + } eltd->timer_set = 0; eltd->lock_in_conflict = 0; @@ -158,59 +179,64 @@ static char* lock_opt(Uint16 flag) { return "--"; } -static void print_lock_x(erts_lcnt_lock_t *lock, Uint16 flag, char *action, char *extra) { - erts_aint_t colls, tries, w_state, r_state; - erts_lcnt_lock_stats_t *stats = NULL; - +static void print_lock_x(erts_lcnt_lock_t *lock, Uint16 flag, char *action) { + erts_aint_t w_state, r_state; char *type; - int i; - + + if (strcmp(lock->name, "run_queue") != 0) return; type = lcnt_lock_type(lock->flag); r_state = ethr_atomic_read(&lock->r_state); w_state = ethr_atomic_read(&lock->w_state); - if (lock->flag & flag) { - erts_printf("%20s [%30s] [r/w state %4ld/%4ld] id %T %s\r\n", - action, - lock->name, - r_state, - w_state, - lock->id, - extra); + erts_fprintf(stderr,"%10s [%24s] [r/w state %4ld/%4ld] %2s id %T\r\n", + action, + lock->name, + r_state, + w_state, + type, + lock->id); } } - -static void print_lock(erts_lcnt_lock_t *lock, char *action) { - if (strcmp(lock->name, "proc_main") == 0) { - print_lock_x(lock, ERTS_LCNT_LT_ALL, action, ""); - } -} - #endif static erts_lcnt_lock_stats_t *lcnt_get_lock_stats(erts_lcnt_lock_t *lock, char *file, unsigned int line) { unsigned int i; erts_lcnt_lock_stats_t *stats = NULL; - - for (i = 0; i < lock->n_stats; i++) { - if ((lock->stats[i].file == file) && (lock->stats[i].line == line)) { - return &(lock->stats[i]); - } - } - if (lock->n_stats < ERTS_LCNT_MAX_LOCK_LOCATIONS) { - stats = &lock->stats[lock->n_stats]; - lock->n_stats++; - stats->file = file; - stats->line = line; - return stats; + if (erts_lcnt_rt_options & ERTS_LCNT_OPT_LOCATION) { + for (i = 0; i < lock->n_stats; i++) { + if ((lock->stats[i].file == file) && (lock->stats[i].line == line)) { + return &(lock->stats[i]); + } + } + if (lock->n_stats < ERTS_LCNT_MAX_LOCK_LOCATIONS) { + stats = &lock->stats[lock->n_stats]; + lock->n_stats++; + stats->file = file; + stats->line = line; + return stats; + } } return &lock->stats[0]; +} +static void lcnt_update_stats_hist(erts_lcnt_hist_t *hist, erts_lcnt_time_t *time_wait) { + int idx; + unsigned long r; + + if (time_wait->s > 0 || time_wait->ns > ERTS_LCNT_HISTOGRAM_MAX_NS) { + idx = ERTS_LCNT_HISTOGRAM_SLOT_SIZE - 1; + } else { + r = time_wait->ns >> ERTS_LCNT_HISTOGRAM_RSHIFT; + if (r) idx = lcnt_log2(r); + else idx = 0; + } + hist->ns[idx]++; } -static void lcnt_update_stats(erts_lcnt_lock_stats_t *stats, int lock_in_conflict, erts_lcnt_time_t *time_wait) { +static void lcnt_update_stats(erts_lcnt_lock_stats_t *stats, int lock_in_conflict, + erts_lcnt_time_t *time_wait) { ethr_atomic_inc(&stats->tries); @@ -220,6 +246,7 @@ static void lcnt_update_stats(erts_lcnt_lock_stats_t *stats, int lock_in_conflic if (time_wait) { lcnt_time_add(&(stats->timer), time_wait); stats->timer_n++; + lcnt_update_stats_hist(&stats->hist,time_wait); } } @@ -248,6 +275,9 @@ void erts_lcnt_init() { /* init lcnt structure */ erts_lcnt_data = (erts_lcnt_data_t*)malloc(sizeof(erts_lcnt_data_t)); + if (!erts_lcnt_data) { + ERTS_INTERNAL_ERROR("Lock counter failed to allocate memory!"); + } erts_lcnt_data->current_locks = erts_lcnt_list_init(); erts_lcnt_data->deleted_locks = erts_lcnt_list_init(); @@ -269,6 +299,9 @@ erts_lcnt_lock_list_t *erts_lcnt_list_init(void) { erts_lcnt_lock_list_t *list; list = (erts_lcnt_lock_list_t*)malloc(sizeof(erts_lcnt_lock_list_t)); + if (!list) { + ERTS_INTERNAL_ERROR("Lock counter failed to allocate memory!"); + } list->head = NULL; list->tail = NULL; list->n = 0; @@ -330,8 +363,9 @@ void erts_lcnt_list_delete(erts_lcnt_lock_list_t *list, erts_lcnt_lock_t *lock) /* interface to erl_threads.h */ /* only lock on init and destroy, all others should use atomics */ void erts_lcnt_init_lock(erts_lcnt_lock_t *lock, char *name, Uint16 flag ) { - erts_lcnt_init_lock_x(lock, name, flag, am_undefined); + erts_lcnt_init_lock_x(lock, name, flag, NIL); } + void erts_lcnt_init_lock_x(erts_lcnt_lock_t *lock, char *name, Uint16 flag, Eterm id) { int i; if (!name) { @@ -360,7 +394,6 @@ void erts_lcnt_init_lock_x(erts_lcnt_lock_t *lock, char *name, Uint16 flag, Eter } erts_lcnt_list_insert(erts_lcnt_data->current_locks, lock); - lcnt_unlock(); } @@ -375,6 +408,9 @@ void erts_lcnt_destroy_lock(erts_lcnt_lock_t *lock) { /* copy structure and insert the copy */ deleted_lock = (erts_lcnt_lock_t*)malloc(sizeof(erts_lcnt_lock_t)); + if (!deleted_lock) { + ERTS_INTERNAL_ERROR("Lock counter failed to allocate memory!"); + } memcpy(deleted_lock, lock, sizeof(erts_lcnt_lock_t)); deleted_lock->next = NULL; @@ -417,8 +453,9 @@ void erts_lcnt_lock_opt(erts_lcnt_lock_t *lock, Uint16 option) { if ((w_state > 0) || (r_state > 0)) { eltd->lock_in_conflict = 1; - if (eltd->timer_set == 0) + if (eltd->timer_set == 0) { lcnt_time(&eltd->timer); + } eltd->timer_set++; } else { eltd->lock_in_conflict = 0; @@ -433,7 +470,7 @@ void erts_lcnt_lock(erts_lcnt_lock_t *lock) { if (!ERTS_LCNT_LOCK_TYPE(lock)) return; w_state = ethr_atomic_read(&lock->w_state); - ethr_atomic_inc( &lock->w_state); + ethr_atomic_inc(&lock->w_state); eltd = lcnt_get_thread_data(); @@ -446,10 +483,10 @@ void erts_lcnt_lock(erts_lcnt_lock_t *lock) { * 'atomicly'. All other locks will block the thread if w_state > 0 * i.e. locked. */ - if (eltd->timer_set == 0) + if (eltd->timer_set == 0) { lcnt_time(&eltd->timer); + } eltd->timer_set++; - } else { eltd->lock_in_conflict = 0; } @@ -459,11 +496,10 @@ void erts_lcnt_lock(erts_lcnt_lock_t *lock) { void erts_lcnt_lock_unaquire(erts_lcnt_lock_t *lock) { /* should check if this thread was "waiting" */ - if (erts_lcnt_rt_options & ERTS_LCNT_OPT_SUSPEND) return; if (!ERTS_LCNT_LOCK_TYPE(lock)) return; - ethr_atomic_dec( &lock->w_state); + ethr_atomic_dec(&lock->w_state); } /* erts_lcnt_lock_post @@ -491,7 +527,7 @@ void erts_lcnt_lock_post_x(erts_lcnt_lock_t *lock, char *file, unsigned int line if (!(lock->flag & (ERTS_LCNT_LT_RWMUTEX | ERTS_LCNT_LT_RWSPINLOCK))) { flowstate = ethr_atomic_read(&lock->flowstate); ASSERT(flowstate == 0); - ethr_atomic_inc( &lock->flowstate); + ethr_atomic_inc(&lock->flowstate); } #endif @@ -500,19 +536,12 @@ void erts_lcnt_lock_post_x(erts_lcnt_lock_t *lock, char *file, unsigned int line ASSERT(eltd); /* if lock was in conflict, time it */ - - if (erts_lcnt_rt_options & ERTS_LCNT_OPT_LOCATION) { - stats = lcnt_get_lock_stats(lock, file, line); - } else { - stats = &lock->stats[0]; - } - + stats = lcnt_get_lock_stats(lock, file, line); if (eltd->timer_set) { lcnt_time(&timer); lcnt_time_diff(&time_wait, &timer, &(eltd->timer)); lcnt_update_stats(stats, eltd->lock_in_conflict, &time_wait); - eltd->timer_set--; ASSERT(eltd->timer_set >= 0); } else { @@ -541,11 +570,11 @@ void erts_lcnt_unlock(erts_lcnt_lock_t *lock) { /* flowstate */ flowstate = ethr_atomic_read(&lock->flowstate); ASSERT(flowstate == 1); - ethr_atomic_dec( &lock->flowstate); + ethr_atomic_dec(&lock->flowstate); /* write state */ w_state = ethr_atomic_read(&lock->w_state); - ASSERT(w_state > 0) + ASSERT(w_state > 0); #endif ethr_atomic_dec(&lock->w_state); } @@ -582,9 +611,7 @@ void erts_lcnt_trylock(erts_lcnt_lock_t *lock, int res) { ethr_atomic_inc( &lock->flowstate); #endif ethr_atomic_inc(&lock->w_state); - lcnt_update_stats(&(lock->stats[0]), 0, NULL); - } else { ethr_atomic_inc(&lock->stats[0].tries); ethr_atomic_inc(&lock->stats[0].colls); diff --git a/erts/emulator/beam/erl_lock_count.h b/erts/emulator/beam/erl_lock_count.h index 75f7cd028b..ffbb93da1b 100644 --- a/erts/emulator/beam/erl_lock_count.h +++ b/erts/emulator/beam/erl_lock_count.h @@ -35,6 +35,10 @@ * | | | - collisions (including trylock busy) * | | | - timer (time spent in waiting for lock) * | | | - n_timer (collisions excluding trylock busy) + * | | | - histogram + * | | | | - # 0 = log2(lock wait_time ns) + * | | | | - ... + * | | | | - # n = log2(lock wait_time ns) * * Each instance of a lock is the unique lock, i.e. set and id in that set. * For each lock there is a set of statistics with where and what impact @@ -68,8 +72,17 @@ #include "ethread.h" +#define ERTS_LCNT_MAX_LOCK_LOCATIONS (10) -#define ERTS_LCNT_MAX_LOCK_LOCATIONS (10) +/* histogram */ +#define ERTS_LCNT_HISTOGRAM_MAX_NS (((unsigned long)1LL << 28) - 1) +#if 0 || defined(HAVE_GETHRTIME) +#define ERTS_LCNT_HISTOGRAM_SLOT_SIZE (30) +#define ERTS_LCNT_HISTOGRAM_RSHIFT (0) +#else +#define ERTS_LCNT_HISTOGRAM_SLOT_SIZE (20) +#define ERTS_LCNT_HISTOGRAM_RSHIFT (10) +#endif #define ERTS_LCNT_LT_SPINLOCK (((Uint16) 1) << 0) #define ERTS_LCNT_LT_RWSPINLOCK (((Uint16) 1) << 1) @@ -104,6 +117,10 @@ typedef struct { extern erts_lcnt_time_t timer_start; +typedef struct { + Uint32 ns[ERTS_LCNT_HISTOGRAM_SLOT_SIZE]; /* log2 array of nano seconds occurences */ +} erts_lcnt_hist_t; + typedef struct erts_lcnt_lock_stats_s { /* "tries" and "colls" needs to be atomic since * trylock busy does not aquire a lock and there @@ -118,6 +135,7 @@ typedef struct erts_lcnt_lock_stats_s { unsigned long timer_n; /* #times waited for lock */ erts_lcnt_time_t timer; /* total wait time for lock */ + erts_lcnt_hist_t hist; } erts_lcnt_lock_stats_t; /* rw locks uses both states, other locks only uses w_state */ diff --git a/erts/emulator/beam/erl_map.c b/erts/emulator/beam/erl_map.c index 2fff7f9390..5e740aacdd 100644 --- a/erts/emulator/beam/erl_map.c +++ b/erts/emulator/beam/erl_map.c @@ -58,6 +58,8 @@ * - maps:size/1 * - maps:without/2 * + * DEBUG: for sharing calculation + * - erts_internal:map_to_tuple_keys/1 */ /* erlang:map_size/1 @@ -647,22 +649,24 @@ int erts_maps_remove(Process *p, Eterm key, Eterm map, Eterm *res) { *mhp++ = tup; if (is_immed(key)) { - while(n--) { + while (1) { if (*ks == key) { goto found_key; - } else { + } else if (--n) { *mhp++ = *vs++; *thp++ = *ks++; - } + } else + break; } } else { - while(n--) { + while(1) { if (EQ(*ks, key)) { goto found_key; - } else { + } else if (--n) { *mhp++ = *vs++; *thp++ = *ks++; - } + } else + break; } } @@ -676,7 +680,7 @@ int erts_maps_remove(Process *p, Eterm key, Eterm map, Eterm *res) { found_key: /* Copy rest of keys and values */ - if (n) { + if (--n) { sys_memcpy(mhp, vs+1, n*sizeof(Eterm)); sys_memcpy(thp, ks+1, n*sizeof(Eterm)); } @@ -817,3 +821,17 @@ int erts_validate_and_sort_map(map_t* mp) } return 1; } + +/* + * erts_internal:map_to_tuple_keys/1 + * + * Used in erts_debug:size/1 + */ + +BIF_RETTYPE erts_internal_map_to_tuple_keys_1(BIF_ALIST_1) { + if (is_map(BIF_ARG_1)) { + map_t *mp = (map_t*)map_val(BIF_ARG_1); + BIF_RET(mp->keys); + } + BIF_ERROR(BIF_P, BADARG); +} diff --git a/erts/emulator/beam/erl_message.c b/erts/emulator/beam/erl_message.c index 6a9030fd99..8870fac7d9 100644 --- a/erts/emulator/beam/erl_message.c +++ b/erts/emulator/beam/erl_message.c @@ -415,7 +415,13 @@ erts_queue_dist_message(Process *rcvr, if (!(*rcvr_locks & ERTS_PROC_LOCK_MSGQ)) erts_smp_proc_unlock(rcvr, ERTS_PROC_LOCK_MSGQ); - erts_proc_notify_new_message(rcvr); + erts_proc_notify_new_message(rcvr, +#ifdef ERTS_SMP + *rcvr_locks +#else + 0 +#endif + ); } } @@ -542,7 +548,13 @@ queue_message(Process *c_p, if (locked_msgq) erts_smp_proc_unlock(receiver, ERTS_PROC_LOCK_MSGQ); - erts_proc_notify_new_message(receiver); + erts_proc_notify_new_message(receiver, +#ifdef ERTS_SMP + *receiver_locks +#else + 0 +#endif + ); #ifndef ERTS_SMP ERTS_HOLE_CHECK(receiver); @@ -896,8 +908,10 @@ erts_send_message(Process* sender, #ifdef USE_VM_PROBES *sender_name = *receiver_name = '\0'; if (DTRACE_ENABLED(message_send)) { - erts_snprintf(sender_name, sizeof(DTRACE_CHARBUF_NAME(sender_name)), "%T", sender->common.id); - erts_snprintf(receiver_name, sizeof(DTRACE_CHARBUF_NAME(receiver_name)), "%T", receiver->common.id); + erts_snprintf(sender_name, sizeof(DTRACE_CHARBUF_NAME(sender_name)), + "%T", sender->common.id); + erts_snprintf(receiver_name, sizeof(DTRACE_CHARBUF_NAME(receiver_name)), + "%T", receiver->common.id); } #endif if (SEQ_TRACE_TOKEN(sender) != NIL && !(flags & ERTS_SND_FLG_NO_SEQ_TRACE)) { @@ -1030,7 +1044,6 @@ erts_send_message(Process* sender, } BM_SWAP_TIMER(send,system); } else { -#ifdef ERTS_SMP ErlOffHeap *ohp; Eterm *hp; erts_aint32_t state; @@ -1062,42 +1075,6 @@ erts_send_message(Process* sender, #endif ); BM_SWAP_TIMER(send,system); -#else - ErlMessage* mp = message_alloc(); - Eterm *hp; - BM_SWAP_TIMER(send,size); - msize = size_object(message); - BM_SWAP_TIMER(size,send); - - if (receiver->stop - receiver->htop <= msize) { - BM_SWAP_TIMER(send,system); - erts_garbage_collect(receiver, msize, receiver->arg_reg, receiver->arity); - BM_SWAP_TIMER(system,send); - } - hp = receiver->htop; - receiver->htop = hp + msize; - BM_SWAP_TIMER(send,copy); - message = copy_struct(message, msize, &hp, &receiver->off_heap); - BM_MESSAGE_COPIED(msize); - BM_SWAP_TIMER(copy,send); - DTRACE6(message_send, sender_name, receiver_name, - (uint32_t)msize, tok_label, tok_lastcnt, tok_serial); - ERL_MESSAGE_TERM(mp) = message; - ERL_MESSAGE_TOKEN(mp) = NIL; -#ifdef USE_VM_PROBES - ERL_MESSAGE_DT_UTAG(mp) = NIL; -#endif - mp->next = NULL; - mp->data.attached = NULL; - LINK_MESSAGE(receiver, mp); - res = receiver->msg.len; - erts_proc_notify_new_message(receiver); - - if (IS_TRACED_FL(receiver, F_TRACE_RECEIVE)) { - trace_receive(receiver, message); - } - BM_SWAP_TIMER(send,system); -#endif /* #ifndef ERTS_SMP */ } return res; } diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c index 40860e141c..adc3520ebb 100644 --- a/erts/emulator/beam/erl_nif.c +++ b/erts/emulator/beam/erl_nif.c @@ -472,6 +472,18 @@ int enif_inspect_binary(ErlNifEnv* env, Eterm bin_term, ErlNifBinary* bin) struct enif_tmp_obj_t* tmp; byte* raw_ptr; }u; + + if (is_boxed(bin_term) && *binary_val(bin_term) == HEADER_SUB_BIN) { + ErlSubBin* sb = (ErlSubBin*) binary_val(bin_term); + if (sb->is_writable) { + ProcBin* pb = (ProcBin*) binary_val(sb->orig); + ASSERT(pb->thing_word == HEADER_PROC_BIN); + if (pb->flags) { + erts_emasculate_writable_binary(pb); + sb->is_writable = 0; + } + } + } u.tmp = NULL; bin->data = erts_get_aligned_binary_bytes_extra(bin_term, &u.raw_ptr, allocator, sizeof(struct enif_tmp_obj_t)); @@ -1513,59 +1525,264 @@ int enif_consume_timeslice(ErlNifEnv* env, int percent) return ERTS_BIF_REDS_LEFT(env->proc) == 0; } -#ifdef ERTS_DIRTY_SCHEDULERS +/* + * NIF exports need a few more items than the Export struct provides, + * including the erl_module_nif* and a NIF function pointer, so the + * NifExport below adds those. The Export member must be first in the + * struct. The saved_mfa, saved_argc, nif_level, alloced_argv_sz and argv + * members are used to track the MFA and arguments of the top NIF in case a + * chain of one or more enif_schedule_nif() calls results in an exception, + * since in that case the original MFA and registers have to be restored + * before returning to Erlang to ensure stacktrace information associated + * with the exception is correct. + */ +typedef ERL_NIF_TERM (*NativeFunPtr)(ErlNifEnv*, int, const ERL_NIF_TERM[]); + +typedef struct { + Export exp; + struct erl_module_nif* m; + NativeFunPtr fp; + Eterm saved_mfa[3]; + int saved_argc; + int alloced_argv_sz; + Eterm argv[1]; +} NifExport; -static void -alloc_proc_psd(Process* proc, Export **ep) +/* + * If a process has saved arguments, they need to be part of the GC + * rootset. The function below is called from setup_rootset() in + * erl_gc.c. This function is declared in erl_process.h. + */ +int +erts_setup_nif_gc(Process* proc, Eterm** objv, int* nobj) { + NifExport* ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + int gc = (ep && ep->saved_argc > 0); + + if (gc) { + *objv = ep->argv; + *nobj = ep->saved_argc; + } + return gc; +} + +/* + * Allocate a NifExport and set it in proc specific data + */ +static NifExport* +allocate_nif_sched_data(Process* proc, int argc) +{ + NifExport* ep; + size_t argv_extra, total; int i; - if (!*ep) { - *ep = erts_alloc(ERTS_ALC_T_PSD, sizeof(Export)); - sys_memset((void*) *ep, 0, sizeof(Export)); - for (i=0; i<ERTS_NUM_CODE_IX; i++) { - (*ep)->addressv[i] = &(*ep)->code[3]; - } - (*ep)->code[3] = (BeamInstr) em_call_nif; + + argv_extra = argc > 1 ? sizeof(Eterm)*(argc-1) : 0; + total = sizeof(NifExport) + argv_extra; + ep = erts_alloc(ERTS_ALC_T_NIF_TRAP_EXPORT, total); + sys_memset((void*) ep, 0, total); + ep->alloced_argv_sz = argc; + for (i=0; i<ERTS_NUM_CODE_IX; i++) { + ep->exp.addressv[i] = &ep->exp.code[3]; } - (void) ERTS_PROC_SET_DIRTY_SCHED_TRAP_EXPORT(proc, ERTS_PROC_LOCK_MAIN, *ep); + ep->exp.code[3] = (BeamInstr) em_call_nif; + (void) ERTS_PROC_SET_NIF_TRAP_EXPORT(proc, ERTS_PROC_LOCK_MAIN, ep); + return ep; +} + +static ERTS_INLINE void +destroy_nif_export(NifExport *nif_export) +{ + erts_free(ERTS_ALC_T_NIF_TRAP_EXPORT, (void *) nif_export); } +void +erts_destroy_nif_export(void *nif_export) +{ + destroy_nif_export((NifExport *) nif_export); +} + +/* + * Initialize a NifExport struct. Create it if needed and store it in the + * proc. The direct_fp function is what will be invoked by op_call_nif, and + * the indirect_fp function, if not NULL, is what the direct_fp function + * will call. If the allocated NifExport isn't enough to hold all of argv, + * allocate a larger one. Save MFA and registers only if the need_save + * parameter is true. + */ static ERL_NIF_TERM -execute_dirty_nif_finalizer(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +init_nif_sched_data(ErlNifEnv* env, NativeFunPtr direct_fp, NativeFunPtr indirect_fp, + int need_save, int argc, const ERL_NIF_TERM argv[]) { - Eterm* reg = ERTS_PROC_GET_SCHDATA(env->proc)->x_reg_array; - ERL_NIF_TERM result = (ERL_NIF_TERM) reg[0]; - typedef ERL_NIF_TERM (*FinalizerFP)(ErlNifEnv*, ERL_NIF_TERM); - FinalizerFP fp; -#if HAVE_INT64 && SIZEOF_LONG != 8 - ASSERT(sizeof(fp) <= sizeof(ErlNifUInt64)); - enif_get_uint64(env, reg[1], (ErlNifUInt64 *) &fp); -#else - ASSERT(sizeof(fp) <= sizeof(unsigned long)); - enif_get_ulong(env, reg[1], (unsigned long *) &fp); -#endif - return (*fp)(env, result); + Process* proc = env->proc; + Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; + NifExport* ep; + int i; + + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + if (!ep) + ep = allocate_nif_sched_data(proc, argc); + else if (need_save && ep->alloced_argv_sz < argc) { + NifExport* new_ep = allocate_nif_sched_data(proc, argc); + destroy_nif_export(ep); + ep = new_ep; + } + ERTS_VBUMP_ALL_REDS(proc); + for (i = 0; i < argc; i++) { + if (need_save) + ep->argv[i] = reg[i]; + reg[i] = (Eterm) argv[i]; + } + if (need_save) { + ep->saved_mfa[0] = proc->current[0]; + ep->saved_mfa[1] = proc->current[1]; + ep->saved_mfa[2] = proc->current[2]; + ep->saved_argc = argc; + } + proc->i = (BeamInstr*) ep->exp.addressv[0]; + ep->exp.code[0] = (BeamInstr) proc->current[0]; + ep->exp.code[1] = (BeamInstr) proc->current[1]; + ep->exp.code[2] = argc; + ep->exp.code[4] = (BeamInstr) direct_fp; + ep->m = env->mod_nif; + ep->fp = indirect_fp; + proc->freason = TRAP; + proc->arity = argc; + return THE_NON_VALUE; } -#endif /* ERTS_DIRTY_SCHEDULERS */ +/* + * Restore saved MFA and registers. Registers are restored only when the + * exception flag is true. + */ +static void +restore_nif_mfa(Process* proc, NifExport* ep, int exception) +{ + int i; + Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; -#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + proc->current[0] = ep->saved_mfa[0]; + proc->current[1] = ep->saved_mfa[1]; + proc->current[2] = ep->saved_mfa[2]; + if (exception) + for (i = 0; i < ep->saved_argc; i++) + reg[i] = ep->argv[i]; + ep->saved_argc = 0; + ep->saved_mfa[0] = THE_NON_VALUE; +} -ERL_NIF_TERM -enif_schedule_dirty_nif(ErlNifEnv* env, int flags, - ERL_NIF_TERM (*fp)(ErlNifEnv*, int, const ERL_NIF_TERM[]), - int argc, const ERL_NIF_TERM argv[]) +#ifdef ERTS_DIRTY_SCHEDULERS + +/* + * Finalize a dirty NIF call. This function is scheduled to cause the VM to + * switch the process off a dirty scheduler thread and back onto a regular + * scheduler thread, and then return the result from the dirty NIF. It also + * restores the original NIF MFA when necessary based on the value of + * ep->fp set by execute_dirty_nif via init_nif_sched_data -- non-NULL + * means restore, NULL means do not restore. + */ +static ERL_NIF_TERM +dirty_nif_finalizer(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NifExport* ep; + + ASSERT(argc == 1); + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data)); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + if (ep->fp) + restore_nif_mfa(proc, ep, 0); + return argv[0]; +} + +/* Finalize a dirty NIF call that raised an exception. Otherwise same as + * the dirty_nif_finalizer() function. + */ +static ERL_NIF_TERM +dirty_nif_exception(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NifExport* ep; + + ASSERT(!ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data)); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + if (ep->fp) + restore_nif_mfa(proc, ep, 1); + return enif_make_badarg(env); +} + +/* + * Dirty NIF execution wrapper function. Invoke an application's dirty NIF, + * then check the result and schedule the appropriate finalizer function + * where needed. Also restore the original NIF MFA when appropriate. + */ +static ERL_NIF_TERM +execute_dirty_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NativeFunPtr fp = (NativeFunPtr) proc->current[6]; + NifExport* ep; + ERL_NIF_TERM result; + + ASSERT(ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data)); + + /* + * Set ep->fp to NULL before the native call so we know later whether it scheduled another NIF for execution + */ + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + ep->fp = NULL; + result = (*fp)(env, argc, argv); + erts_smp_atomic32_read_band_mb(&proc->state, + ~(ERTS_PSFLG_DIRTY_CPU_PROC + |ERTS_PSFLG_DIRTY_IO_PROC + |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q + |ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); + if (erts_refc_dectest(&env->mod_nif->rt_dtor_cnt, 0) == 0 && env->mod_nif->mod == NULL) + close_lib(env->mod_nif); + /* + * If no more NIFs were scheduled by the native call via + * enif_schedule_nif(), then ep->fp will still be NULL as set above, in + * which case we need to restore the original NIF calling + * context. Reuse fp essentially as a boolean for this, passing it to + * init_nif_sched_data below. Both dirty_nif_exception and + * dirty_nif_finalizer then check ep->fp to decide whether or not to + * restore the original calling context. + */ + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + if (ep->fp) + fp = NULL; + if (is_non_value(result)) { + if (proc->freason != TRAP) { + ASSERT(proc->freason == BADARG); + return init_nif_sched_data(env, dirty_nif_exception, fp, 0, argc, argv); + } else { + if (ep->fp == NULL) + restore_nif_mfa(proc, ep, 1); + return result; + } + } + else + return init_nif_sched_data(env, dirty_nif_finalizer, fp, 0, 1, &result); +} + +/* + * Dirty NIF scheduling wrapper function. Schedule a dirty NIF to execute + * via the execute_dirty_nif() wrapper function. The dirty scheduler thread + * type (CPU or I/O) is indicated in flags parameter. + */ +static ERTS_INLINE ERL_NIF_TERM +schedule_dirty_nif(ErlNifEnv* env, int flags, int argc, const ERL_NIF_TERM argv[]) { -#ifdef USE_THREADS erts_aint32_t state, n, a; Process* proc = env->proc; - Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; - Export* ep = NULL; - int i; + NativeFunPtr fp = (NativeFunPtr) proc->current[6]; + NifExport* ep; + int need_save; - int chkflgs = (flags & (ERL_NIF_DIRTY_JOB_IO_BOUND|ERL_NIF_DIRTY_JOB_CPU_BOUND)); - if (chkflgs != ERL_NIF_DIRTY_JOB_IO_BOUND && chkflgs != ERL_NIF_DIRTY_JOB_CPU_BOUND) - return enif_make_badarg(env); + ASSERT(flags==ERL_NIF_DIRTY_JOB_IO_BOUND || flags==ERL_NIF_DIRTY_JOB_CPU_BOUND); a = erts_smp_atomic32_read_acqb(&proc->state); while (1) { @@ -1577,7 +1794,7 @@ enif_schedule_dirty_nif(ErlNifEnv* env, int flags, */ n &= ~(ERTS_PSFLG_DIRTY_CPU_PROC|ERTS_PSFLG_DIRTY_IO_PROC |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q); - if (chkflgs == ERL_NIF_DIRTY_JOB_CPU_BOUND) + if (flags == ERL_NIF_DIRTY_JOB_CPU_BOUND) n |= ERTS_PSFLG_DIRTY_CPU_PROC; else n |= ERTS_PSFLG_DIRTY_IO_PROC; @@ -1585,82 +1802,106 @@ enif_schedule_dirty_nif(ErlNifEnv* env, int flags, if (a == state) break; } - if (!(ep = ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(proc))) - alloc_proc_psd(proc, &ep); - ERTS_VBUMP_ALL_REDS(proc); - ep->code[2] = argc; - for (i = 0; i < argc; i++) { - reg[i] = (Eterm) argv[i]; - } - proc->i = (BeamInstr*) ep->addressv[0]; - ep->code[4] = (BeamInstr) fp; - proc->freason = TRAP; + erts_refc_inc(&env->mod_nif->rt_dtor_cnt, 1); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + need_save = (ep == NULL || is_non_value(ep->saved_mfa[0])); + return init_nif_sched_data(env, execute_dirty_nif, fp, need_save, argc, argv); +} - return THE_NON_VALUE; -#else - return (*fp)(env, argc, argv); -#endif +static ERL_NIF_TERM +schedule_dirty_io_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + return schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_IO_BOUND, argc, argv); +} + +static ERL_NIF_TERM +schedule_dirty_cpu_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + return schedule_dirty_nif(env, ERL_NIF_DIRTY_JOB_CPU_BOUND, argc, argv); +} + +#endif /* ERTS_DIRTY_SCHEDULERS */ + +/* + * NIF execution wrapper used by enif_schedule_nif() for regular NIFs. It + * calls the actual NIF, restores original NIF MFA if necessary, and + * then returns the NIF result. + */ +static ERL_NIF_TERM +execute_nif(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]) +{ + Process* proc = env->proc; + NativeFunPtr fp = (NativeFunPtr) proc->current[6]; + NifExport* ep; + ERL_NIF_TERM result; + + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + ep->fp = NULL; + result = (*fp)(env, argc, argv); + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + /* + * If no NIFs were scheduled by the native call via + * enif_schedule_nif(), then ep->fp will still be NULL as set above, in + * which case we need to restore the original NIF MFA. + */ + if (ep->fp == NULL) + restore_nif_mfa(proc, ep, is_non_value(result) && proc->freason != TRAP); + return result; } ERL_NIF_TERM -enif_schedule_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result, - ERL_NIF_TERM (*fp)(ErlNifEnv*, ERL_NIF_TERM)) +enif_schedule_nif(ErlNifEnv* env, const char* fun_name, int flags, + ERL_NIF_TERM (*fp)(ErlNifEnv*, int, const ERL_NIF_TERM[]), + int argc, const ERL_NIF_TERM argv[]) { -#ifdef USE_THREADS Process* proc = env->proc; - Eterm* reg = ERTS_PROC_GET_SCHDATA(proc)->x_reg_array; - Export* ep; + NifExport* ep; + ERL_NIF_TERM fun_name_atom, result; + int need_save; - erts_smp_atomic32_read_band_mb(&proc->state, - ~(ERTS_PSFLG_DIRTY_CPU_PROC - |ERTS_PSFLG_DIRTY_IO_PROC - |ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q - |ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); - if (!(ep = ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(proc))) - alloc_proc_psd(proc, &ep); - ERTS_VBUMP_ALL_REDS(proc); - ep->code[2] = 2; - reg[0] = (Eterm) result; -#if HAVE_INT64 && SIZEOF_LONG != 8 - ASSERT(sizeof(fp) <= sizeof(ErlNifUInt64)); - reg[1] = (Eterm) enif_make_uint64(env, (ErlNifUInt64) fp); -#else - ASSERT(sizeof(fp) <= sizeof(unsigned long)); - reg[1] = (Eterm) enif_make_ulong(env, (unsigned long) fp); -#endif - proc->i = (BeamInstr*) ep->addressv[0]; - ep->code[4] = (BeamInstr) execute_dirty_nif_finalizer; - proc->freason = TRAP; + if (argc > MAX_ARG) + return enif_make_badarg(env); + fun_name_atom = enif_make_atom(env, fun_name); + if (enif_is_exception(env, fun_name_atom)) + return fun_name_atom; - return THE_NON_VALUE; + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + need_save = (ep == NULL || is_non_value(ep->saved_mfa[0])); + + if (flags) { +#ifdef ERTS_DIRTY_SCHEDULERS + NativeFunPtr sched_fun; + int chkflgs = (flags & (ERL_NIF_DIRTY_JOB_IO_BOUND|ERL_NIF_DIRTY_JOB_CPU_BOUND)); + if (chkflgs == ERL_NIF_DIRTY_JOB_IO_BOUND) + sched_fun = schedule_dirty_io_nif; + else if (chkflgs == ERL_NIF_DIRTY_JOB_CPU_BOUND) + sched_fun = schedule_dirty_cpu_nif; + else + return enif_make_badarg(env); + result = init_nif_sched_data(env, sched_fun, fp, need_save, argc, argv); #else - return (*fp)(env, result); + return enif_make_badarg(env); #endif -} + } + else + result = init_nif_sched_data(env, execute_nif, fp, need_save, argc, argv); -/* A simple finalizer that just returns its result argument */ -ERL_NIF_TERM -enif_dirty_nif_finalizer(ErlNifEnv* env, ERL_NIF_TERM result) -{ + ep = (NifExport*) ERTS_PROC_GET_NIF_TRAP_EXPORT(proc); + ASSERT(ep); + ep->exp.code[1] = (BeamInstr) fun_name_atom; return result; } +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + int enif_is_on_dirty_scheduler(ErlNifEnv* env) { return ERTS_SCHEDULER_IS_DIRTY(env->proc->scheduler_data); } -int -enif_have_dirty_schedulers() -{ -#ifdef USE_THREADS - return 1; -#else - return 0; -#endif -} - #endif /* ERL_NIF_DIRTY_SCHEDULER_SUPPORT */ /* Maps */ @@ -1961,6 +2202,35 @@ static Eterm load_nif_error(Process* p, const char* atom, const char* format, .. return ret; } +/* + * The function below is for looping through ErlNifFunc arrays, helping + * provide backwards compatibility across the version 2.7 change that added + * the "flags" field to ErlNifFunc. + */ +static ErlNifFunc* next_func(ErlNifEntry* entry, int* incrp, ErlNifFunc* func) +{ + ASSERT(incrp); + if (!*incrp) { + if (entry->major > 2 || (entry->major == 2 && entry->minor >= 7)) + *incrp = sizeof(ErlNifFunc); + else { + /* + * ErlNifFuncV1 below is what ErlNifFunc was before the + * addition of the flags field for 2.7, and is needed to handle + * backward compatibility. + */ + typedef struct { + const char* name; + unsigned arity; + ERL_NIF_TERM (*fptr)(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]); + }ErlNifFuncV1; + *incrp = sizeof(ErlNifFuncV1); + } + } + return (ErlNifFunc*) ((char*)func + *incrp); +} + + BIF_RETTYPE load_nif_2(BIF_ALIST_2) { static const char bad_lib[] = "bad_lib"; @@ -2049,8 +2319,10 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) (entry = erts_sys_ddll_call_nif_init(init_func)) == NULL)) { ret = load_nif_error(BIF_P, bad_lib, "Library init-call unsuccessful"); } - else if (entry->major != ERL_NIF_MAJOR_VERSION - || entry->minor > ERL_NIF_MINOR_VERSION + else if (entry->major < ERL_NIF_MIN_REQUIRED_MAJOR_VERSION_ON_LOAD + || (ERL_NIF_MAJOR_VERSION < entry->major + || (ERL_NIF_MAJOR_VERSION == entry->major + && ERL_NIF_MINOR_VERSION < entry->minor)) || (entry->major==2 && entry->minor == 5)) { /* experimental maps */ ret = load_nif_error(BIF_P, bad_lib, "Library version (%d.%d) not compatible (with %d.%d).", @@ -2068,22 +2340,48 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) } else { /*erts_fprintf(stderr, "Found module %T\r\n", mod_atom);*/ - + + int maybe_dirty_nifs = ((entry->major > 2 || (entry->major == 2 && entry->minor >= 7)) + && (entry->options & ERL_NIF_DIRTY_NIF_OPTION)); + int incr = 0; + ErlNifFunc* f = entry->funcs; for (i=0; i < entry->num_of_funcs && ret==am_ok; i++) { BeamInstr** code_pp; - ErlNifFunc* f = &entry->funcs[i]; if (!erts_atom_get(f->name, sys_strlen(f->name), &f_atom, ERTS_ATOM_ENC_LATIN1) || (code_pp = get_func_pp(mod->curr.code, f_atom, f->arity))==NULL) { ret = load_nif_error(BIF_P,bad_lib,"Function not found %T:%s/%u", mod_atom, f->name, f->arity); - } - else if (code_pp[1] - code_pp[0] < (5+3)) { + } + else if (maybe_dirty_nifs && f->flags) { + /* + * If the flags field is non-zero and this emulator was + * built with dirty scheduler support, check that the flags + * value is legal. But if this emulator was built without + * dirty scheduler support, treat a non-zero flags field as + * a load error. + */ +#ifdef ERTS_DIRTY_SCHEDULERS + if (f->flags != ERL_NIF_DIRTY_JOB_IO_BOUND && f->flags != ERL_NIF_DIRTY_JOB_CPU_BOUND) + ret = load_nif_error(BIF_P, bad_lib, "Illegal flags field value %d for NIF %T:%s/%u", + f->flags, mod_atom, f->name, f->arity); +#else + ret = load_nif_error(BIF_P, bad_lib, "NIF %T:%s/%u requires a runtime with dirty scheduler support.", + mod_atom, f->name, f->arity); +#endif + } +#ifdef ERTS_DIRTY_SCHEDULERS + else if (code_pp[1] - code_pp[0] < (5+4)) +#else + else if (code_pp[1] - code_pp[0] < (5+3)) +#endif + { ret = load_nif_error(BIF_P,bad_lib,"No explicit call to load_nif" - " in module (%T:%s/%u to small)", - mod_atom, entry->funcs[i].name, entry->funcs[i].arity); + " in module (%T:%s/%u too small)", + mod_atom, f->name, f->arity); } /*erts_fprintf(stderr, "Found NIF %T:%s/%u\r\n", - mod_atom, entry->funcs[i].name, entry->funcs[i].arity);*/ + mod_atom, f->name, f->arity);*/ + f = next_func(entry, &incr, f); } } @@ -2109,7 +2407,8 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) * is deprecated and was only ment as a development feature not to * be used in production systems. (See warning below) */ - int k; + int k, old_incr = 0; + ErlNifFunc* old_func; lib->priv_data = mod->curr.nif->priv_data; ASSERT(mod->curr.nif->entry != NULL); @@ -2118,13 +2417,16 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) goto error; } /* Check that no NIF is removed */ + old_func = mod->curr.nif->entry->funcs; for (k=0; k < mod->curr.nif->entry->num_of_funcs; k++) { - ErlNifFunc* old_func = &mod->curr.nif->entry->funcs[k]; + int incr = 0; + ErlNifFunc* f = entry->funcs; for (i=0; i < entry->num_of_funcs; i++) { - if (old_func->arity == entry->funcs[i].arity - && sys_strcmp(old_func->name, entry->funcs[i].name) == 0) { + if (old_func->arity == f->arity + && sys_strcmp(old_func->name, f->name) == 0) { break; } + f = next_func(entry, &incr, f); } if (i == entry->num_of_funcs) { ret = load_nif_error(BIF_P,reload,"Reloaded library missing " @@ -2132,7 +2434,8 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) old_func->name, old_func->arity); goto error; } - } + old_func = next_func(mod->curr.nif->entry, &old_incr, old_func); + } erts_pre_nif(&env, BIF_P, lib); veto = entry->reload(&env, &lib->priv_data, BIF_ARG_2); erts_post_nif(&env); @@ -2179,13 +2482,17 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) /* ** Everything ok, patch the beam code with op_call_nif */ - mod->curr.nif = lib; + + int incr = 0; + ErlNifFunc* f = entry->funcs; + + mod->curr.nif = lib; for (i=0; i < entry->num_of_funcs; i++) { BeamInstr* code_ptr; - erts_atom_get(entry->funcs[i].name, sys_strlen(entry->funcs[i].name), &f_atom, ERTS_ATOM_ENC_LATIN1); - code_ptr = *get_func_pp(mod->curr.code, f_atom, entry->funcs[i].arity); - + erts_atom_get(f->name, sys_strlen(f->name), &f_atom, ERTS_ATOM_ENC_LATIN1); + code_ptr = *get_func_pp(mod->curr.code, f_atom, f->arity); + if (code_ptr[1] == 0) { code_ptr[5+0] = (BeamInstr) BeamOp(op_call_nif); } @@ -2193,10 +2500,21 @@ BIF_RETTYPE load_nif_2(BIF_ALIST_2) GenericBp* g = (GenericBp *) code_ptr[1]; ASSERT(code_ptr[5+0] == (BeamInstr) BeamOp(op_i_generic_breakpoint)); - g->orig_instr = (BeamInstr) BeamOp(op_call_nif); - } - code_ptr[5+1] = (BeamInstr) entry->funcs[i].fptr; + g->orig_instr = (BeamInstr) BeamOp(op_call_nif); + } + if ((entry->major > 2 || (entry->major == 2 && entry->minor >= 7)) + && (entry->options & ERL_NIF_DIRTY_NIF_OPTION) && f->flags) { +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT + code_ptr[5+3] = (BeamInstr) f->fptr; + code_ptr[5+1] = (f->flags == ERL_NIF_DIRTY_JOB_IO_BOUND) ? + (BeamInstr) schedule_dirty_io_nif : + (BeamInstr) schedule_dirty_cpu_nif; +#endif + } + else + code_ptr[5+1] = (BeamInstr) f->fptr; code_ptr[5+2] = (BeamInstr) lib; + f = next_func(entry, &incr, f); } } else { diff --git a/erts/emulator/beam/erl_nif.h b/erts/emulator/beam/erl_nif.h index c12ba4d554..849024453c 100644 --- a/erts/emulator/beam/erl_nif.h +++ b/erts/emulator/beam/erl_nif.h @@ -42,9 +42,25 @@ ** 2.5: R17 Maps API additions ** 2.6: R17 with maps ** R17 dirty schedulers +** 2.7: 17.3 add enif_schedule_nif +** remove enif_schedule_dirty_nif, enif_schedule_dirty_nif_finalizer, enif_dirty_nif_finalizer +** add ErlNifEntry options +** add ErlNifFunc flags */ #define ERL_NIF_MAJOR_VERSION 2 -#define ERL_NIF_MINOR_VERSION 6 +#define ERL_NIF_MINOR_VERSION 7 + +/* + * The emulator will refuse to load a nif-lib with a major version + * lower than ERL_NIF_MIN_REQUIRED_MAJOR_VERSION_ON_LOAD. The load + * may however fail if user have not removed use of deprecated + * symbols. + * + * The ERL_NIF_MIN_REQUIRED_MAJOR_VERSION_ON_LOAD have to allow + * loading of nif-libs built at least two major OTP releases + * ago. + */ +#define ERL_NIF_MIN_REQUIRED_MAJOR_VERSION_ON_LOAD 2 #include <stdlib.h> @@ -113,8 +129,10 @@ typedef struct const char* name; unsigned arity; ERL_NIF_TERM (*fptr)(ErlNifEnv* env, int argc, const ERL_NIF_TERM argv[]); + unsigned flags; }ErlNifFunc; + typedef struct enif_entry_t { int major; @@ -127,8 +145,11 @@ typedef struct enif_entry_t int (*upgrade)(ErlNifEnv*, void** priv_data, void** old_priv_data, ERL_NIF_TERM load_info); void (*unload) (ErlNifEnv*, void* priv_data); const char* vm_variant; + unsigned options; }ErlNifEntry; +/* Field bits for ErlNifEntry options */ +#define ERL_NIF_DIRTY_NIF_OPTION 1 typedef struct @@ -231,6 +252,11 @@ extern TWinDynNifCallbacks WinDynNifCallbacks; # endif #endif +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +# define ERL_NIF_ENTRY_OPTIONS ERL_NIF_DIRTY_NIF_OPTION +#else +# define ERL_NIF_ENTRY_OPTIONS 0 +#endif #ifdef __cplusplus } @@ -256,7 +282,8 @@ ERL_NIF_INIT_DECL(NAME) \ sizeof(FUNCS) / sizeof(*FUNCS), \ FUNCS, \ LOAD, RELOAD, UPGRADE, UNLOAD, \ - ERL_NIF_VM_VARIANT \ + ERL_NIF_VM_VARIANT, \ + ERL_NIF_ENTRY_OPTIONS \ }; \ ERL_NIF_INIT_BODY; \ return &entry; \ diff --git a/erts/emulator/beam/erl_nif_api_funcs.h b/erts/emulator/beam/erl_nif_api_funcs.h index d7c554e60b..630cefae93 100644 --- a/erts/emulator/beam/erl_nif_api_funcs.h +++ b/erts/emulator/beam/erl_nif_api_funcs.h @@ -22,7 +22,7 @@ #endif /* -** WARNING: add new ERL_NIF_API_FUNC_DECL entries at the bottom of the list +** WARNING: Add new ERL_NIF_API_FUNC_DECL entries at the bottom of the list ** to keep compatibility on Windows!!! ** ** And don't forget to increase ERL_NIF_MINOR_VERSION in erl_nif.h @@ -141,14 +141,6 @@ ERL_NIF_API_FUNC_DECL(int,enif_is_number,(ErlNifEnv*, ERL_NIF_TERM term)); ERL_NIF_API_FUNC_DECL(void*,enif_dlopen,(const char* lib, void (*err_handler)(void*,const char*), void* err_arg)); ERL_NIF_API_FUNC_DECL(void*,enif_dlsym,(void* handle, const char* symbol, void (*err_handler)(void*,const char*), void* err_arg)); ERL_NIF_API_FUNC_DECL(int,enif_consume_timeslice,(ErlNifEnv*, int percent)); -#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT -ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_dirty_nif,(ErlNifEnv*,int,ERL_NIF_TERM (*)(ErlNifEnv*,int,const ERL_NIF_TERM[]),int,const ERL_NIF_TERM[])); -ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_dirty_nif_finalizer,(ErlNifEnv*,ERL_NIF_TERM,ERL_NIF_TERM (*)(ErlNifEnv*,ERL_NIF_TERM))); -ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_dirty_nif_finalizer,(ErlNifEnv*,ERL_NIF_TERM)); -ERL_NIF_API_FUNC_DECL(int,enif_is_on_dirty_scheduler,(ErlNifEnv*)); -ERL_NIF_API_FUNC_DECL(int,enif_have_dirty_schedulers,(void)); -#endif - ERL_NIF_API_FUNC_DECL(int, enif_is_map, (ErlNifEnv* env, ERL_NIF_TERM term)); ERL_NIF_API_FUNC_DECL(int, enif_get_map_size, (ErlNifEnv* env, ERL_NIF_TERM term, size_t *size)); ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM, enif_make_new_map, (ErlNifEnv* env)); @@ -163,12 +155,22 @@ ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_is_tail, (ErlNifEnv *env, ErlNifMap ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_next, (ErlNifEnv *env, ErlNifMapIterator *iter)); ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_prev, (ErlNifEnv *env, ErlNifMapIterator *iter)); ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_get_pair, (ErlNifEnv *env, ErlNifMapIterator *iter, ERL_NIF_TERM *key, ERL_NIF_TERM *value)); - +ERL_NIF_API_FUNC_DECL(ERL_NIF_TERM,enif_schedule_nif,(ErlNifEnv*,const char*,int,ERL_NIF_TERM (*)(ErlNifEnv*,int,const ERL_NIF_TERM[]),int,const ERL_NIF_TERM[])); /* -** Add new entries here to keep compatibility on Windows!!! +** ADD NEW ENTRIES HERE (before this comment) !!! */ + + +/* + * Conditional EXPERIMENTAL stuff always last. + * Must be moved up and made unconditional to support binary backward + * compatibility on Windows. + */ +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +ERL_NIF_API_FUNC_DECL(int,enif_is_on_dirty_scheduler,(ErlNifEnv*)); #endif +#endif /* ERL_NIF_API_FUNC_DECL */ /* ** Please keep the ERL_NIF_API_FUNC_MACRO list below in the same order @@ -282,21 +284,12 @@ ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_get_pair, (ErlNifEnv *env, ErlNifMa # define enif_make_int64 ERL_NIF_API_FUNC_MACRO(enif_make_int64) # define enif_make_uint64 ERL_NIF_API_FUNC_MACRO(enif_make_uint64) #endif - # define enif_is_exception ERL_NIF_API_FUNC_MACRO(enif_is_exception) # define enif_make_reverse_list ERL_NIF_API_FUNC_MACRO(enif_make_reverse_list) # define enif_is_number ERL_NIF_API_FUNC_MACRO(enif_is_number) # define enif_dlopen ERL_NIF_API_FUNC_MACRO(enif_dlopen) # define enif_dlsym ERL_NIF_API_FUNC_MACRO(enif_dlsym) # define enif_consume_timeslice ERL_NIF_API_FUNC_MACRO(enif_consume_timeslice) -#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT -# define enif_schedule_dirty_nif ERL_NIF_API_FUNC_MACRO(enif_schedule_dirty_nif) -# define enif_schedule_dirty_nif_finalizer ERL_NIF_API_FUNC_MACRO(enif_schedule_dirty_nif_finalizer) -# define enif_dirty_nif_finalizer ERL_NIF_API_FUNC_MACRO(enif_dirty_nif_finalizer) -# define enif_is_on_dirty_scheduler ERL_NIF_API_FUNC_MACRO(enif_is_on_dirty_scheduler) -# define enif_have_dirty_schedulers ERL_NIF_API_FUNC_MACRO(enif_have_dirty_schedulers) -#endif - # define enif_is_map ERL_NIF_API_FUNC_MACRO(enif_is_map) # define enif_get_map_size ERL_NIF_API_FUNC_MACRO(enif_get_map_size) # define enif_make_new_map ERL_NIF_API_FUNC_MACRO(enif_make_new_map) @@ -311,11 +304,21 @@ ERL_NIF_API_FUNC_DECL(int, enif_map_iterator_get_pair, (ErlNifEnv *env, ErlNifMa # define enif_map_iterator_next ERL_NIF_API_FUNC_MACRO(enif_map_iterator_next) # define enif_map_iterator_prev ERL_NIF_API_FUNC_MACRO(enif_map_iterator_prev) # define enif_map_iterator_get_pair ERL_NIF_API_FUNC_MACRO(enif_map_iterator_get_pair) +# define enif_schedule_nif ERL_NIF_API_FUNC_MACRO(enif_schedule_nif) /* -** Add new entries here +** ADD NEW ENTRIES HERE (before this comment) */ + +/* + * Conditional EXPERIMENTAL stuff always last + * Must be moved up and made unconditional to support binary backward + * compatibility on Windows. + */ +#ifdef ERL_NIF_DIRTY_SCHEDULER_SUPPORT +# define enif_is_on_dirty_scheduler ERL_NIF_API_FUNC_MACRO(enif_is_on_dirty_scheduler) #endif +#endif /* ERL_NIF_API_FUNC_MACRO */ #if defined(__GNUC__) && !(defined(__WIN32__) || defined(_WIN32) || defined(_WIN32_)) diff --git a/erts/emulator/beam/erl_port_task.c b/erts/emulator/beam/erl_port_task.c index fb6048b41f..2aa0a27197 100644 --- a/erts/emulator/beam/erl_port_task.c +++ b/erts/emulator/beam/erl_port_task.c @@ -32,6 +32,7 @@ #include "global.h" #include "erl_port_task.h" #include "dist.h" +#include "erl_check_io.h" #include "dtrace-wrapper.h" #include <stdarg.h> @@ -68,6 +69,13 @@ static void chk_task_queues(Port *pp, ErtsPortTask *execq, int processing_busy_q #define DTRACE_DRIVER(PROBE_NAME, PP) do {} while(0) #endif +#define ERTS_SMP_LC_VERIFY_RQ(RQ, PP) \ + do { \ + ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(runq)); \ + ERTS_SMP_LC_ASSERT((RQ) == ((ErtsRunQueue *) \ + erts_smp_atomic_read_nob(&(PP)->run_queue))); \ + } while (0) + erts_smp_atomic_t erts_port_task_outstanding_io_tasks; #define ERTS_PT_STATE_SCHEDULED 0 @@ -543,6 +551,16 @@ reset_handle(ErtsPortTask *ptp) } static ERTS_INLINE void +reset_executed_io_task_handle(ErtsPortTask *ptp) +{ + if (ptp->u.alive.handle) { + ASSERT(ptp == handle2task(ptp->u.alive.handle)); + erts_io_notify_port_task_executed(ptp->u.alive.handle); + reset_port_task_handle(ptp->u.alive.handle); + } +} + +static ERTS_INLINE void set_handle(ErtsPortTask *ptp, ErtsPortTaskHandle *pthp) { ptp->u.alive.handle = pthp; @@ -798,12 +816,13 @@ schedule_port_task_handle_list_free(ErtsPortTaskHandleList *pthlp) static ERTS_INLINE void abort_nosuspend_task(Port *pp, ErtsPortTaskType type, - ErtsPortTaskTypeData *tdp) + ErtsPortTaskTypeData *tdp, + int bpq_data) { ASSERT(type == ERTS_PORT_TASK_PROC_SIG); - if (!pp->sched.taskq.bpq) + if (!bpq_data) tdp->psig.callback(NULL, ERTS_PORT_SFLG_INVALID, ERTS_PROC2PORT_SIG_ABORT_NOSUSPEND, @@ -991,6 +1010,7 @@ static ERTS_INLINE int finalize_exec(Port *pp, ErtsPortTask **execq, int processing_busy_q) { erts_aint32_t act; + unsigned int prof_runnable_ports; if (!processing_busy_q) pp->sched.taskq.local.first = *execq; @@ -1007,6 +1027,10 @@ finalize_exec(Port *pp, ErtsPortTask **execq, int processing_busy_q) if (act & ERTS_PTS_FLG_CHK_UNSET_BUSY_PORT_Q) act = check_unset_busy_port_q(pp, act, pp->sched.taskq.bpq); + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&pp->sched); + while (1) { erts_aint32_t new, exp; @@ -1018,12 +1042,24 @@ finalize_exec(Port *pp, ErtsPortTask **execq, int processing_busy_q) act = erts_smp_atomic32_cmpxchg_relb(&pp->sched.flags, new, exp); - ASSERT(!(act & ERTS_PTS_FLG_IN_RUNQ)); + ERTS_LC_ASSERT(!(act & ERTS_PTS_FLG_IN_RUNQ)); + ERTS_LC_ASSERT(!(act & ERTS_PTS_FLG_EXEC_IMM)); if (exp == act) break; } + if (prof_runnable_ports | IS_TRACED_FL(pp, F_TRACE_SCHED_PORTS)) { + /* trace port scheduling, out */ + if (IS_TRACED_FL(pp, F_TRACE_SCHED_PORTS)) + trace_sched_ports(pp, am_out); + if (prof_runnable_ports) { + if (!(act & (ERTS_PTS_FLG_EXEC_IMM|ERTS_PTS_FLG_HAVE_TASKS))) + profile_runnable_port(pp, am_inactive); + erts_port_task_sched_unlock(&pp->sched); + } + } + return (act & ERTS_PTS_FLG_HAVE_TASKS) != 0; } @@ -1345,7 +1381,7 @@ erts_port_task_abort_nosuspend_tasks(Port *pp) #endif schedule_port_task_handle_list_free(pthlp); - abort_nosuspend_task(pp, type, &td); + abort_nosuspend_task(pp, type, &td, pp->sched.taskq.bpq != NULL); } } @@ -1369,11 +1405,9 @@ erts_port_task_schedule(Eterm id, Port *pp; ErtsPortTask *ptp = NULL; erts_aint32_t act, add_flags; + unsigned int prof_runnable_ports; - if (pthp && erts_port_task_is_scheduled(pthp)) { - ASSERT(0); - erts_port_task_abort(pthp); - } + ERTS_LC_ASSERT(!pthp || !erts_port_task_is_scheduled(pthp)); ASSERT(is_internal_port(id)); @@ -1457,6 +1491,10 @@ erts_port_task_schedule(Eterm id, if (ns_pthlp) add_flags |= ERTS_PTS_FLG_HAVE_NS_TASKS; + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&pp->sched); + while (1) { erts_aint32_t new, exp; @@ -1481,6 +1519,13 @@ erts_port_task_schedule(Eterm id, goto done; /* Died after our task insert... */ } + if (prof_runnable_ports) { + if (!(act & ERTS_PTS_FLG_EXEC_IMM)) + profile_runnable_port(pp, am_active); + erts_port_task_sched_unlock(&pp->sched); + prof_runnable_ports = 0; + } + /* Enqueue port on run-queue */ runq = erts_port_runq(pp); @@ -1489,8 +1534,10 @@ erts_port_task_schedule(Eterm id, #ifdef ERTS_SMP xrunq = erts_check_emigration_need(runq, ERTS_PORT_PRIO_LEVEL); + ERTS_SMP_LC_ASSERT(runq != xrunq); + ERTS_SMP_LC_VERIFY_RQ(runq, pp); if (xrunq) { - /* Port emigrated ... */ + /* Emigrate port ... */ erts_smp_atomic_set_nob(&pp->run_queue, (erts_aint_t) xrunq); erts_smp_runq_unlock(runq); runq = erts_port_runq(pp); @@ -1500,10 +1547,6 @@ erts_port_task_schedule(Eterm id, #endif enqueue_port(runq, pp); - - if (erts_system_profile_flags.runnable_ports) { - profile_runnable_port(pp, am_active); - } erts_smp_runq_unlock(runq); @@ -1511,6 +1554,9 @@ erts_port_task_schedule(Eterm id, done: + if (prof_runnable_ports) + erts_port_task_sched_unlock(&pp->sched); + #ifdef ERTS_SMP if (dhndl != ERTS_THR_PRGR_DHANDLE_MANAGED) erts_port_dec_refc(pp); @@ -1525,7 +1571,7 @@ abort_nosuspend: erts_port_dec_refc(pp); #endif - abort_nosuspend_task(pp, ptp->type, &ptp->u.alive.td); + abort_nosuspend_task(pp, ptp->type, &ptp->u.alive.td, 0); ASSERT(ns_pthlp); erts_free(ERTS_ALC_T_PT_HNDL_LIST, ns_pthlp); @@ -1609,6 +1655,8 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) goto done; } + ERTS_SMP_LC_VERIFY_RQ(runq, pp); + erts_smp_runq_unlock(runq); *curr_port_pp = pp; @@ -1659,8 +1707,6 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) goto aborted_port_task; } - reset_handle(ptp); - if (erts_system_monitor_long_schedule != 0) { start_time = erts_timestamp_millis(); } @@ -1671,6 +1717,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) switch (ptp->type) { case ERTS_PORT_TASK_TIMEOUT: + reset_handle(ptp); reds = ERTS_PORT_REDS_TIMEOUT; if (!(state & ERTS_PORT_SFLGS_DEAD)) { DTRACE_DRIVER(driver_timeout, pp); @@ -1685,6 +1732,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) for input and output */ (*pp->drv_ptr->ready_input)((ErlDrvData) pp->drv_data, ptp->u.alive.td.io.event); + reset_executed_io_task_handle(ptp); io_tasks_executed++; break; case ERTS_PORT_TASK_OUTPUT: @@ -1693,6 +1741,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) DTRACE_DRIVER(driver_ready_output, pp); (*pp->drv_ptr->ready_output)((ErlDrvData) pp->drv_data, ptp->u.alive.td.io.event); + reset_executed_io_task_handle(ptp); io_tasks_executed++; break; case ERTS_PORT_TASK_EVENT: @@ -1702,10 +1751,12 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) (*pp->drv_ptr->event)((ErlDrvData) pp->drv_data, ptp->u.alive.td.io.event, ptp->u.alive.td.io.event_data); + reset_executed_io_task_handle(ptp); io_tasks_executed++; break; case ERTS_PORT_TASK_PROC_SIG: { ErtsProc2PortSigData *sigdp = &ptp->u.alive.td.psig.data; + reset_handle(ptp); ASSERT((state & ERTS_PORT_SFLGS_DEAD) == 0); if (!pp->sched.taskq.bpq) reds = ptp->u.alive.td.psig.callback(pp, @@ -1723,6 +1774,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) break; } case ERTS_PORT_TASK_DIST_CMD: + reset_handle(ptp); reds = erts_dist_command(pp, CONTEXT_REDS - pp->reds); break; default: @@ -1765,10 +1817,6 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) erts_unblock_fpe(fpe_was_unmasked); - /* trace port scheduling, out */ - if (IS_TRACED_FL(pp, F_TRACE_SCHED_PORTS)) { - trace_sched_ports(pp, am_out); - } if (io_tasks_executed) { ASSERT(erts_smp_atomic_read_nob(&erts_port_task_outstanding_io_tasks) @@ -1791,11 +1839,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) erts_smp_runq_lock(runq); - if (!active) { - if (erts_system_profile_flags.runnable_ports) - profile_runnable_port(pp, am_inactive); - } - else { + if (active) { #ifdef ERTS_SMP ErtsRunQueue *xrunq; #endif @@ -1804,6 +1848,8 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) #ifdef ERTS_SMP xrunq = erts_check_emigration_need(runq, ERTS_PORT_PRIO_LEVEL); + ERTS_SMP_LC_ASSERT(runq != xrunq); + ERTS_SMP_LC_VERIFY_RQ(runq, pp); if (!xrunq) { #endif enqueue_port(runq, pp); @@ -1811,7 +1857,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) #ifdef ERTS_SMP } else { - /* Port emigrated ... */ + /* Emigrate port... */ erts_smp_atomic_set_nob(&pp->run_queue, (erts_aint_t) xrunq); erts_smp_runq_unlock(runq); @@ -2035,9 +2081,9 @@ begin_port_cleanup(Port *pp, ErtsPortTask **execqp, int *processing_busy_q_p) DTRACE_CHARBUF(pid_str, 16); ErtsProcList* plp2 = plp; - erts_snprintf(port_str, sizeof(port_str), "%T", pp->common.id); + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", pp->common.id); while (plp2 != NULL) { - erts_snprintf(pid_str, sizeof(pid_str), "%T", plp2->pid); + erts_snprintf(pid_str, sizeof(DTRACE_CHARBUF_NAME(pid_str)), "%T", plp2->pid); DTRACE2(process_port_unblocked, pid_str, port_str); } } diff --git a/erts/emulator/beam/erl_port_task.h b/erts/emulator/beam/erl_port_task.h index 1d30465ec9..406cd3c492 100644 --- a/erts/emulator/beam/erl_port_task.h +++ b/erts/emulator/beam/erl_port_task.h @@ -78,6 +78,7 @@ extern erts_smp_atomic_t erts_port_task_outstanding_io_tasks; #define ERTS_PTS_FLG_PARALLELISM (((erts_aint32_t) 1) << 9) #define ERTS_PTS_FLG_FORCE_SCHED (((erts_aint32_t) 1) << 10) #define ERTS_PTS_FLG_EXITING (((erts_aint32_t) 1) << 11) +#define ERTS_PTS_FLG_EXEC_IMM (((erts_aint32_t) 1) << 12) #define ERTS_PTS_FLGS_BUSY \ (ERTS_PTS_FLG_BUSY_PORT | ERTS_PTS_FLG_BUSY_PORT_Q) @@ -87,6 +88,7 @@ extern erts_smp_atomic_t erts_port_task_outstanding_io_tasks; | ERTS_PTS_FLG_HAVE_BUSY_TASKS \ | ERTS_PTS_FLG_HAVE_TASKS \ | ERTS_PTS_FLG_EXEC \ + | ERTS_PTS_FLG_EXEC_IMM \ | ERTS_PTS_FLG_FORCE_SCHED \ | ERTS_PTS_FLG_EXITING) @@ -154,7 +156,7 @@ erts_port_task_handle_init(ErtsPortTaskHandle *pthp) ERTS_GLB_INLINE int erts_port_task_is_scheduled(ErtsPortTaskHandle *pthp) { - return ((void *) erts_smp_atomic_read_nob(pthp)) != NULL; + return ((void *) erts_smp_atomic_read_acqb(pthp)) != NULL; } ERTS_GLB_INLINE void erts_port_task_pre_init_sched(ErtsPortTaskSched *ptsp, diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index 37e1d07107..ea63d20dfa 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -148,6 +148,12 @@ extern BeamInstr beam_apply[]; extern BeamInstr beam_exit[]; extern BeamInstr beam_continue_exit[]; +#ifdef __OSE__ +/* Eager check I/O not supported on OSE yet. */ +int erts_eager_check_io = 0; +#else +int erts_eager_check_io = 0; +#endif int erts_sched_compact_load; int erts_sched_balance_util = 0; Uint erts_no_schedulers; @@ -590,12 +596,10 @@ erts_pre_init_process(void) erts_psd_required_locks[ERTS_PSD_DELAYED_GC_TASK_QS].set_locks = ERTS_PSD_DELAYED_GC_TASK_QS_SET_LOCKS; -#ifdef ERTS_DIRTY_SCHEDULERS - erts_psd_required_locks[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT].get_locks - = ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_GET_LOCKS; - erts_psd_required_locks[ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT].set_locks - = ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_SET_LOCKS; -#endif + erts_psd_required_locks[ERTS_PSD_NIF_TRAP_EXPORT].get_locks + = ERTS_PSD_NIF_TRAP_EXPORT_GET_LOCKS; + erts_psd_required_locks[ERTS_PSD_NIF_TRAP_EXPORT].set_locks + = ERTS_PSD_NIF_TRAP_EXPORT_SET_LOCKS; /* Check that we have locks for all entries */ for (ix = 0; ix < ERTS_PSD_SIZE; ix++) { @@ -712,72 +716,24 @@ sched_wall_time_ts(void) #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT -#ifdef ARCH_64 - -static ERTS_INLINE Uint64 -aschedtime_read(ErtsAtomicSchedTime *var) -{ - return (Uint64) erts_atomic_read_nob((erts_atomic_t *) var); -} - -static ERTS_INLINE void -aschedtime_set(ErtsAtomicSchedTime *var, Uint64 val) -{ - erts_atomic_set_nob((erts_atomic_t *) var, (erts_aint_t) val); -} - -static ERTS_INLINE void -aschedtime_init(ErtsAtomicSchedTime *var) -{ - erts_atomic_init_nob((erts_atomic_t *) var, (erts_aint_t) 0); -} - -#elif defined(ARCH_32) - static ERTS_INLINE Uint64 aschedtime_read(ErtsAtomicSchedTime *var) { - erts_dw_aint_t dw; - erts_dw_atomic_read_nob((erts_dw_atomic_t *) var, &dw); -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw.dw_sint; -#else - { - Uint64 res; - res = (Uint64) ((Uint32) dw.sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw.sint[ERTS_DW_AINT_LOW_WORD]); - return res; - } -#endif + return (Uint64) erts_atomic64_read_nob((erts_atomic64_t *) var); } static ERTS_INLINE void aschedtime_set(ErtsAtomicSchedTime *var, Uint64 val) { - erts_dw_aint_t dw; -#ifdef ETHR_SU_DW_NAINT_T__ - dw.dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw.sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) (val & 0xffffffff); - dw.sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif - erts_dw_atomic_set_nob((erts_dw_atomic_t *) var, &dw); + erts_atomic64_set_nob((erts_atomic64_t *) var, (erts_aint64_t) val); } static ERTS_INLINE void aschedtime_init(ErtsAtomicSchedTime *var) { - erts_dw_aint_t dw; - dw.sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) 0; - dw.sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) 0; - erts_dw_atomic_init_nob((erts_dw_atomic_t *) var, &dw); + erts_atomic64_init_nob((erts_atomic64_t *) var, (erts_aint64_t) 0); } -#else -# error :-/ -#endif - #define ERTS_GET_AVG_MAX_UNLOCKED_TRY 50 #define ERTS_SCHED_AVG_UTIL_WRITE_MARKER (~((Uint64) 0)) @@ -2211,6 +2167,9 @@ aux_work_timeout_early_init(int no_schedulers) p = (UWord) malloc((sizeof(ErtsAuxWorkTmo) + sizeof(erts_atomic32_t)*(no_schedulers+1)) + ERTS_CACHE_LINE_SIZE-1); + if (!p) { + ERTS_INTERNAL_ERROR("malloc failed to allocate memory!"); + } if (p & ERTS_CACHE_LINE_MASK) p = (p & ~ERTS_CACHE_LINE_MASK) + ERTS_CACHE_LINE_SIZE; ASSERT((p & ERTS_CACHE_LINE_MASK) == 0); @@ -2380,29 +2339,47 @@ try_set_sys_scheduling(void) #endif static ERTS_INLINE int -prepare_for_sys_schedule(ErtsSchedulerData *esdp) +prepare_for_sys_schedule(ErtsSchedulerData *esdp, int non_blocking) { + if (non_blocking && erts_eager_check_io) { #ifdef ERTS_SMP - while (!erts_port_task_have_outstanding_io_tasks() - && try_set_sys_scheduling()) { #ifdef ERTS_SCHED_ONLY_POLL_SCHED_1 - if (esdp->no != 1) { - /* If we are not scheduler 1 and ERTS_SCHED_ONLY_POLL_SCHED_1 is used - then we make sure to wake scheduler 1 */ - ErtsRunQueue *rq = ERTS_RUNQ_IX(0); - clear_sys_scheduling(); - wake_scheduler(rq); - return 0; - } + if (esdp->no != 1) { + /* If we are not scheduler 1 and ERTS_SCHED_ONLY_POLL_SCHED_1 is used + then we make sure to wake scheduler 1 */ + ErtsRunQueue *rq = ERTS_RUNQ_IX(0); + wake_scheduler(rq); + return 0; + } #endif - if (!erts_port_task_have_outstanding_io_tasks()) + return try_set_sys_scheduling(); +#else return 1; - clear_sys_scheduling(); +#endif } - return 0; + else { +#ifdef ERTS_SMP + while (!erts_port_task_have_outstanding_io_tasks() + && try_set_sys_scheduling()) { +#ifdef ERTS_SCHED_ONLY_POLL_SCHED_1 + if (esdp->no != 1) { + /* If we are not scheduler 1 and ERTS_SCHED_ONLY_POLL_SCHED_1 is used + then we make sure to wake scheduler 1 */ + ErtsRunQueue *rq = ERTS_RUNQ_IX(0); + clear_sys_scheduling(); + wake_scheduler(rq); + return 0; + } +#endif + if (!erts_port_task_have_outstanding_io_tasks()) + return 1; + clear_sys_scheduling(); + } + return 0; #else - return !erts_port_task_have_outstanding_io_tasks(); + return !erts_port_task_have_outstanding_io_tasks(); #endif + } } #ifdef ERTS_SMP @@ -2672,6 +2649,13 @@ aux_thread(void *unused) ErtsThrPrgrCallbacks callbacks; int thr_prgr_active = 1; +#ifdef ERTS_ENABLE_LOCK_CHECK + { + char buf[] = "aux_thread"; + erts_lc_set_thread_name(buf); + } +#endif + ssi->event = erts_tse_fetch(); callbacks.arg = (void *) ssi; @@ -2772,7 +2756,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * be waiting in erl_sys_schedule() */ - if (ERTS_SCHEDULER_IS_DIRTY(esdp) || !prepare_for_sys_schedule(esdp)) { + if (ERTS_SCHEDULER_IS_DIRTY(esdp) || !prepare_for_sys_schedule(esdp, 0)) { sched_waiting(esdp->no, rq); @@ -2936,7 +2920,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * Got to check that we still got I/O tasks; otherwise * we have to continue checking for I/O... */ - if (!prepare_for_sys_schedule(esdp)) { + if (!prepare_for_sys_schedule(esdp, 0)) { spincount *= ERTS_SCHED_TSE_SLEEP_SPINCOUNT_FACT; goto tse_wait; } @@ -2958,7 +2942,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * Got to check that we still got I/O tasks; otherwise * we have to wait in erl_sys_schedule() after all... */ - if (!prepare_for_sys_schedule(esdp)) { + if (!prepare_for_sys_schedule(esdp, 0)) { /* * Not allowed to wait in erl_sys_schedule; * do tse wait instead... @@ -3212,11 +3196,11 @@ chk_wake_sched(ErtsRunQueue *crq, int ix, int activate) return 0; wrq = ERTS_RUNQ_IX(ix); flags = ERTS_RUNQ_FLGS_GET(wrq); + if (activate && !(flags & ERTS_RUNQ_FLG_SUSPENDED)) { + if (try_inc_no_active_runqs(ix+1)) + (void) ERTS_RUNQ_FLGS_UNSET(wrq, ERTS_RUNQ_FLG_INACTIVE); + } if (!(flags & (ERTS_RUNQ_FLG_SUSPENDED|ERTS_RUNQ_FLG_NONEMPTY))) { - if (activate) { - if (try_inc_no_active_runqs(ix+1)) - (void) ERTS_RUNQ_FLGS_UNSET(wrq, ERTS_RUNQ_FLG_INACTIVE); - } wake_scheduler(wrq); return 1; } @@ -3748,17 +3732,25 @@ evacuate_run_queue(ErtsRunQueue *rq, } #ifdef ERTS_DIRTY_SCHEDULERS else if (state & ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q) { - erts_aint32_t old; - old = erts_smp_atomic32_read_band_nob(&proc->state, - ~(ERTS_PSFLG_DIRTY_CPU_PROC - | ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q)); +#ifdef DEBUG + erts_aint32_t old = +#else + (void) +#endif + erts_smp_atomic32_read_band_nob(&proc->state, + ~(ERTS_PSFLG_DIRTY_CPU_PROC + | ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q)); /* assert that no other dirty flags are set */ ASSERT(!(old & (ERTS_PSFLG_DIRTY_IO_PROC|ERTS_PSFLG_DIRTY_IO_PROC_IN_Q))); } else if (state & ERTS_PSFLG_DIRTY_IO_PROC_IN_Q) { - erts_aint32_t old; - old = erts_smp_atomic32_read_band_nob(&proc->state, - ~(ERTS_PSFLG_DIRTY_IO_PROC - | ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); +#ifdef DEBUG + erts_aint32_t old = +#else + (void) +#endif + erts_smp_atomic32_read_band_nob(&proc->state, + ~(ERTS_PSFLG_DIRTY_IO_PROC + | ERTS_PSFLG_DIRTY_IO_PROC_IN_Q)); /* assert that no other dirty flags are set */ ASSERT(!(old & (ERTS_PSFLG_DIRTY_CPU_PROC|ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q))); } @@ -5867,6 +5859,9 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p, Proces case ERTS_ENQUEUE_NOT: if (erts_system_profile_flags.runnable_procs) { + /* Status lock prevents out of order "runnable proc" trace msgs */ + ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p)); + if (!(a & ERTS_PSFLG_ACTIVE_SYS) && (!(a & ERTS_PSFLG_ACTIVE) || (a & ERTS_PSFLG_SUSPENDED))) { @@ -5980,7 +5975,8 @@ change_proc_schedule_state(Process *p, erts_aint32_t clear_state_flags, erts_aint32_t set_state_flags, erts_aint32_t *statep, - erts_aint32_t *enq_prio_p) + erts_aint32_t *enq_prio_p, + ErtsProcLocks locks) { /* * NOTE: ERTS_PSFLG_RUNNING, ERTS_PSFLG_RUNNING_SYS and @@ -5989,6 +5985,11 @@ change_proc_schedule_state(Process *p, */ erts_aint32_t a = *statep, n; int enqueue; /* < 0 -> use proxy */ + unsigned int prof_runnable_procs = erts_system_profile_flags.runnable_procs; + unsigned int lock_status = (prof_runnable_procs + && !(locks & ERTS_PROC_LOCK_STATUS)); + + ERTS_SMP_LC_ASSERT(locks == erts_proc_lc_my_proc_locks(p)); ASSERT(!(a & ERTS_PSFLG_PROXY)); ASSERT((clear_state_flags & (ERTS_PSFLG_RUNNING @@ -5998,6 +5999,9 @@ change_proc_schedule_state(Process *p, | ERTS_PSFLG_RUNNING_SYS | ERTS_PSFLG_ACTIVE_SYS)) == 0); + if (lock_status) + erts_smp_proc_lock(p, ERTS_PROC_LOCK_STATUS); + while (1) { erts_aint32_t e; n = e = a; @@ -6033,7 +6037,9 @@ change_proc_schedule_state(Process *p, break; } - if (erts_system_profile_flags.runnable_procs) { + if (prof_runnable_procs) { + + /* Status lock prevents out of order "runnable proc" trace msgs */ if (((n & (ERTS_PSFLG_SUSPENDED | ERTS_PSFLG_ACTIVE)) == ERTS_PSFLG_ACTIVE) @@ -6046,15 +6052,18 @@ change_proc_schedule_state(Process *p, profile_runnable_proc(p, am_active); } + if (lock_status) + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); } + *statep = a; return enqueue; } static ERTS_INLINE void -schedule_process(Process *p, erts_aint32_t in_state) +schedule_process(Process *p, erts_aint32_t in_state, ErtsProcLocks locks) { erts_aint32_t enq_prio = -1; erts_aint32_t state = in_state; @@ -6062,7 +6071,8 @@ schedule_process(Process *p, erts_aint32_t in_state) 0, ERTS_PSFLG_ACTIVE, &state, - &enq_prio); + &enq_prio, + locks); if (enqueue != ERTS_ENQUEUE_NOT) add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio), state, @@ -6070,16 +6080,27 @@ schedule_process(Process *p, erts_aint32_t in_state) } void -erts_schedule_process(Process *p, erts_aint32_t state) +erts_schedule_process(Process *p, erts_aint32_t state, ErtsProcLocks locks) { - schedule_process(p, state); + schedule_process(p, state, locks); } static void schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) { + /* + * Expects status lock to be locked when called, and + * returns with status lock unlocked... + */ erts_aint32_t a = state, n, enq_prio = -1; int enqueue; /* < 0 -> use proxy */ + unsigned int prof_runnable_procs = erts_system_profile_flags.runnable_procs; + + /* Status lock prevents out of order "runnable proc" trace msgs */ + ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p)); + + if (!prof_runnable_procs) + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); ASSERT(!(state & ERTS_PSFLG_PROXY)); @@ -6088,7 +6109,7 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) n = e = a; if (a & ERTS_PSFLG_FREE) - return; /* We don't want to schedule free processes... */ + goto cleanup; /* We don't want to schedule free processes... */ enqueue = ERTS_ENQUEUE_NOT; n |= ERTS_PSFLG_ACTIVE_SYS; @@ -6101,7 +6122,7 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) goto cleanup; } - if (erts_system_profile_flags.runnable_procs) { + if (prof_runnable_procs) { if (!(a & (ERTS_PSFLG_ACTIVE_SYS | ERTS_PSFLG_RUNNING @@ -6111,6 +6132,8 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) profile_runnable_proc(p, am_active); } + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); + prof_runnable_procs = 0; } if (enqueue != ERTS_ENQUEUE_NOT) { @@ -6125,8 +6148,14 @@ schedule_process_sys_task(Process *p, erts_aint32_t state, Process *proxy) } cleanup: + + if (prof_runnable_procs) + erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS); + if (proxy) free_proxy_proc(proxy); + + ERTS_SMP_LC_ASSERT(!(ERTS_PROC_LOCK_STATUS & erts_proc_lc_my_proc_locks(p))); } static ERTS_INLINE int @@ -6144,7 +6173,7 @@ suspend_process(Process *c_p, Process *p) if (c_p == p) { state = erts_smp_atomic32_read_bor_relb(&p->state, ERTS_PSFLG_SUSPENDED); - ASSERT(state & ERTS_PSFLG_RUNNING); + ASSERT(state & (ERTS_PSFLG_RUNNING|ERTS_PSFLG_RUNNING_SYS)); suspended = (state & ERTS_PSFLG_SUSPENDED) ? -1: 1; } else { @@ -6193,7 +6222,7 @@ suspend_process(Process *c_p, Process *p) } static ERTS_INLINE void -resume_process(Process *p) +resume_process(Process *p, ErtsProcLocks locks) { erts_aint32_t state, enq_prio = -1; int enqueue; @@ -6210,7 +6239,8 @@ resume_process(Process *p) ERTS_PSFLG_SUSPENDED, 0, &state, - &enq_prio); + &enq_prio, + locks); if (enqueue) add2runq(enqueue > 0 ? p : make_proxy_proc(NULL, p, enq_prio), state, @@ -7811,6 +7841,9 @@ erts_start_schedulers(void) #ifdef ETHR_HAVE_THREAD_NAMES opts.name = malloc(80); + if (!opts.name) { + ERTS_INTERNAL_ERROR("malloc failed to allocate memory!"); + } #endif #ifdef ERTS_SMP @@ -8023,7 +8056,8 @@ handle_pend_sync_suspend(Process *suspendee, } /* suspender is suspended waiting for suspendee to suspend; resume suspender */ - resume_process(suspender); + ASSERT(suspendee != suspender); + resume_process(suspender, ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspender, ERTS_PROC_LOCK_STATUS); } } @@ -8058,7 +8092,7 @@ pid2proc_not_running(Process *c_p, ErtsProcLocks c_p_locks, ASSERT(c_p->flags & F_P2PNR_RESCHED); c_p->flags &= ~F_P2PNR_RESCHED; if (!suspend && rp) - resume_process(rp); + resume_process(rp, rp_locks); } else { @@ -8216,7 +8250,8 @@ handle_pend_bif_sync_suspend(Process *suspendee, } /* suspender is suspended waiting for suspendee to suspend; resume suspender */ - resume_process(suspender); + ASSERT(suspender != suspendee); + resume_process(suspender, ERTS_PROC_LOCK_LINK|ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspender, ERTS_PROC_LOCK_LINK|ERTS_PROC_LOCK_STATUS); } @@ -8576,7 +8611,8 @@ resume_process_1(BIF_ALIST_1) ASSERT(ERTS_PSFLG_SUSPENDED & erts_smp_atomic32_read_nob(&suspendee->state)); - resume_process(suspendee); + ASSERT(BIF_P != suspendee); + resume_process(suspendee, ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspendee, ERTS_PROC_LOCK_STATUS); } @@ -8706,7 +8742,7 @@ erts_resume(Process* process, ErtsProcLocks process_locks) ERTS_SMP_LC_ASSERT(process_locks == erts_proc_lc_my_proc_locks(process)); if (!(process_locks & ERTS_PROC_LOCK_STATUS)) erts_smp_proc_lock(process, ERTS_PROC_LOCK_STATUS); - resume_process(process); + resume_process(process, process_locks|ERTS_PROC_LOCK_STATUS); if (!(process_locks & ERTS_PROC_LOCK_STATUS)) erts_smp_proc_unlock(process, ERTS_PROC_LOCK_STATUS); } @@ -8725,7 +8761,7 @@ erts_resume_processes(ErtsProcList *list) proc = erts_pid2proc(NULL, 0, plp->pid, ERTS_PROC_LOCK_STATUS); if (proc) { if (erts_proclist_same(plp, proc)) { - resume_process(proc); + resume_process(proc, ERTS_PROC_LOCK_STATUS); nresumed++; } erts_smp_proc_unlock(proc, ERTS_PROC_LOCK_STATUS); @@ -9140,7 +9176,7 @@ Process *schedule(Process *p, int calls) } else if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && (fcalls > input_reductions && - prepare_for_sys_schedule(esdp))) { + prepare_for_sys_schedule(esdp, !0))) { /* * Schedule system-level activities. */ @@ -9148,8 +9184,6 @@ Process *schedule(Process *p, int calls) erts_smp_atomic32_set_relb(&function_calls, 0); fcalls = 0; - ASSERT(!erts_port_task_have_outstanding_io_tasks()); - #if 0 /* Not needed since we wont wait in sys schedule */ erts_sys_schedule_interrupt(0); #endif @@ -9181,7 +9215,9 @@ Process *schedule(Process *p, int calls) if (RUNQ_READ_LEN(&rq->ports.info.len)) { int have_outstanding_io; have_outstanding_io = erts_port_task_execute(rq, &esdp->current_port); - if ((have_outstanding_io && fcalls > 2*input_reductions) + if ((!erts_eager_check_io + && have_outstanding_io + && fcalls > 2*input_reductions) || rq->halt_in_progress) { /* * If we have performed more than 2*INPUT_REDUCTIONS since @@ -9207,7 +9243,6 @@ Process *schedule(Process *p, int calls) */ pick_next_process: { erts_aint32_t psflg_band_mask; - erts_aint32_t running_flag; int prio_q; int qmask; @@ -9269,12 +9304,6 @@ Process *schedule(Process *p, int calls) 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; @@ -9284,8 +9313,12 @@ Process *schedule(Process *p, int calls) tmp = state & (ERTS_PSFLG_SUSPENDED | ERTS_PSFLG_PENDING_EXIT | ERTS_PSFLG_ACTIVE_SYS); - if (tmp != ERTS_PSFLG_SUSPENDED) - new |= running_flag; + if (tmp != ERTS_PSFLG_SUSPENDED) { + if (state & ERTS_PSFLG_ACTIVE_SYS) + new |= ERTS_PSFLG_RUNNING_SYS; + else + new |= ERTS_PSFLG_RUNNING; + } } state = erts_smp_atomic32_cmpxchg_relb(&p->state, new, exp); if (state == exp) { @@ -9964,8 +9997,10 @@ erts_internal_request_system_task_3(BIF_ALIST_3) rp_state = n; } - erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_STATUS); - + /* + * schedule_process_sys_task() unlocks status + * lock on process. + */ schedule_process_sys_task(rp, rp_state, NULL); if (free_stqs) @@ -10710,7 +10745,7 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). * Schedule process for execution. */ - schedule_process(p, state); + schedule_process(p, state, 0); VERBOSE(DEBUG_PROCESSES, ("Created a new process: %T\n",p->common.id)); @@ -11031,7 +11066,8 @@ set_proc_exiting(Process *p, ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT, ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE, &state, - &enq_prio); + &enq_prio, + ERTS_PROC_LOCKS_ALL); p->fvalue = reason; if (bp) @@ -11072,7 +11108,8 @@ set_proc_self_exiting(Process *c_p) ERTS_PSFLG_SUSPENDED|ERTS_PSFLG_PENDING_EXIT, ERTS_PSFLG_EXITING|ERTS_PSFLG_ACTIVE, &state, - &enq_prio); + &enq_prio, + ERTS_PROC_LOCKS_ALL); ASSERT(!enqueue); return state; @@ -11313,7 +11350,7 @@ send_exit_signal(Process *c_p, /* current process if and only dtrace_pid_str(from, sender_str); dtrace_proc_str(rp, receiver_str); - erts_snprintf(reason_buf, sizeof(reason_buf) - 1, "%T", reason); + erts_snprintf(reason_buf, sizeof(DTRACE_CHARBUF_NAME(reason_buf)) - 1, "%T", reason); DTRACE3(process_exit_signal, sender_str, receiver_str, reason_buf); } #endif @@ -11717,8 +11754,9 @@ resume_suspend_monitor(ErtsSuspendMonitor *smon, void *vc_p) Process *suspendee = erts_pid2proc((Process *) vc_p, ERTS_PROC_LOCK_MAIN, smon->pid, ERTS_PROC_LOCK_STATUS); if (suspendee) { + ASSERT(suspendee != vc_p); if (smon->active) - resume_process(suspendee); + resume_process(suspendee, ERTS_PROC_LOCK_STATUS); erts_smp_proc_unlock(suspendee, ERTS_PROC_LOCK_STATUS); } erts_destroy_suspend_monitor(smon); @@ -11810,6 +11848,7 @@ erts_continue_exit_process(Process *p) struct saved_calls *scb; process_breakpoint_time_t *pbt; erts_aint32_t state; + void *nif_export; #ifdef DEBUG int yield_allowed = 1; @@ -11960,6 +11999,7 @@ erts_continue_exit_process(Process *p) : NULL); scb = ERTS_PROC_SET_SAVED_CALLS_BUF(p, ERTS_PROC_LOCKS_ALL, NULL); pbt = ERTS_PROC_SET_CALL_TIME(p, ERTS_PROC_LOCKS_ALL, NULL); + nif_export = ERTS_PROC_SET_NIF_TRAP_EXPORT(p, ERTS_PROC_LOCKS_ALL, NULL); erts_smp_proc_unlock(p, ERTS_PROC_LOCKS_ALL); #ifdef BM_COUNTERS @@ -12007,6 +12047,9 @@ erts_continue_exit_process(Process *p) if (pbt) erts_free(ERTS_ALC_T_BPD, (void *) pbt); + if (nif_export) + erts_destroy_nif_export(nif_export); + delete_process(p); #ifdef ERTS_SMP @@ -12051,7 +12094,7 @@ timeout_proc(Process* p) state = erts_smp_atomic32_read_acqb(&p->state); if (!(state & ERTS_PSFLG_ACTIVE)) - schedule_process(p, state); + schedule_process(p, state, ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_STATUS); } diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index ed6dadbffa..f50b217d4a 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.h @@ -104,6 +104,7 @@ struct saved_calls { }; extern Export exp_send, exp_receive, exp_timeout; +extern int erts_eager_check_io; extern int erts_sched_compact_load; extern int erts_sched_balance_util; extern Uint erts_no_schedulers; @@ -350,13 +351,7 @@ typedef struct { #undef ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT #define ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT_OPT -#ifdef ARCH_64 -typedef erts_atomic_t ErtsAtomicSchedTime; -#elif defined(ARCH_32) -typedef erts_dw_atomic_t ErtsAtomicSchedTime; -#else -# error :-/ -#endif +typedef erts_atomic64_t ErtsAtomicSchedTime; #if ERTS_HAVE_SCHED_UTIL_BALANCING_SUPPORT typedef struct { @@ -734,13 +729,9 @@ erts_smp_reset_max_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi) #define ERTS_PSD_DIST_ENTRY 3 #define ERTS_PSD_CALL_TIME_BP 4 #define ERTS_PSD_DELAYED_GC_TASK_QS 5 -#ifdef ERTS_DIRTY_SCHEDULERS -#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT 6 +#define ERTS_PSD_NIF_TRAP_EXPORT 6 #define ERTS_PSD_SIZE 7 -#else -#define ERTS_PSD_SIZE 6 -#endif typedef struct { void *data[ERTS_PSD_SIZE]; @@ -767,10 +758,8 @@ typedef struct { #define ERTS_PSD_DELAYED_GC_TASK_QS_GET_LOCKS ERTS_PROC_LOCK_MAIN #define ERTS_PSD_DELAYED_GC_TASK_QS_SET_LOCKS ERTS_PROC_LOCK_MAIN -#ifdef ERTS_DIRTY_SCHEDULERS -#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_GET_LOCKS ERTS_PROC_LOCK_MAIN -#define ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT_SET_LOCKS ERTS_PROC_LOCK_MAIN -#endif +#define ERTS_PSD_NIF_TRAP_EXPORT_GET_LOCKS ERTS_PROC_LOCK_MAIN +#define ERTS_PSD_NIF_TRAP_EXPORT_SET_LOCKS ERTS_PROC_LOCK_MAIN typedef struct { ErtsProcLocks get_locks; @@ -1367,6 +1356,9 @@ Uint64 erts_get_proc_interval(void); Uint64 erts_ensure_later_proc_interval(Uint64); Uint64 erts_step_proc_interval(void); +int erts_setup_nif_gc(Process* proc, Eterm** objv, int* nobj); /* see erl_nif.c */ +void erts_destroy_nif_export(void *); /* see erl_nif.c */ + ErtsProcList *erts_proclist_create(Process *); void erts_proclist_destroy(ErtsProcList *); @@ -1704,17 +1696,17 @@ ErtsSchedulerData *erts_get_scheduler_data(void) #endif #endif -void erts_schedule_process(Process *, erts_aint32_t); +void erts_schedule_process(Process *, erts_aint32_t, ErtsProcLocks); -ERTS_GLB_INLINE void erts_proc_notify_new_message(Process *p); +ERTS_GLB_INLINE void erts_proc_notify_new_message(Process *p, ErtsProcLocks locks); #if ERTS_GLB_INLINE_INCL_FUNC_DEF ERTS_GLB_INLINE void -erts_proc_notify_new_message(Process *p) +erts_proc_notify_new_message(Process *p, ErtsProcLocks locks) { /* No barrier needed, due to msg lock */ erts_aint32_t state = erts_smp_atomic32_read_nob(&p->state); if (!(state & ERTS_PSFLG_ACTIVE)) - erts_schedule_process(p, state); + erts_schedule_process(p, state, locks); } #endif @@ -1817,12 +1809,10 @@ erts_psd_set(Process *p, ErtsProcLocks plocks, int ix, void *data) #define ERTS_PROC_SET_DELAYED_GC_TASK_QS(P, L, PBT) \ ((ErtsProcSysTaskQs *) erts_psd_set((P), (L), ERTS_PSD_DELAYED_GC_TASK_QS, (void *) (PBT))) -#ifdef ERTS_DIRTY_SCHEDULERS -#define ERTS_PROC_GET_DIRTY_SCHED_TRAP_EXPORT(P) \ - ((Export *) erts_psd_get((P), ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT)) -#define ERTS_PROC_SET_DIRTY_SCHED_TRAP_EXPORT(P, L, DSTE) \ - ((Export *) erts_psd_set((P), (L), ERTS_PSD_DIRTY_SCHED_TRAP_EXPORT, (void *) (DSTE))) -#endif +#define ERTS_PROC_GET_NIF_TRAP_EXPORT(P) \ + erts_psd_get((P), ERTS_PSD_NIF_TRAP_EXPORT) +#define ERTS_PROC_SET_NIF_TRAP_EXPORT(P, L, NTE) \ + erts_psd_set((P), (L), ERTS_PSD_NIF_TRAP_EXPORT, (void *) (NTE)) ERTS_GLB_INLINE Eterm erts_proc_get_error_handler(Process *p); diff --git a/erts/emulator/beam/erl_ptab.c b/erts/emulator/beam/erl_ptab.c index eabf016081..02943ee683 100644 --- a/erts/emulator/beam/erl_ptab.c +++ b/erts/emulator/beam/erl_ptab.c @@ -280,124 +280,38 @@ struct ErtsPTabListBifData_ { }; -#ifdef ARCH_32 - -static ERTS_INLINE Uint64 -dw_aint_to_uint64(erts_dw_aint_t *dw) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw->dw_sint; -#else - Uint64 res; - res = (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - -static void -unint64_to_dw_aint(erts_dw_aint_t *dw, Uint64 val) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - dw->dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw->sint[ERTS_DW_AINT_LOW_WORD] = (erts_aint_t) (val & 0xffffffff); - dw->sint[ERTS_DW_AINT_HIGH_WORD] = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif -} - static ERTS_INLINE void last_data_init_nob(ErtsPTab *ptab, Uint64 val) { - erts_dw_aint_t dw; - unint64_to_dw_aint(&dw, val); - erts_smp_dw_atomic_init_nob(&ptab->vola.tile.last_data, &dw); + erts_smp_atomic64_init_nob(&ptab->vola.tile.last_data, (erts_aint64_t) val); } static ERTS_INLINE void last_data_set_relb(ErtsPTab *ptab, Uint64 val) { - erts_dw_aint_t dw; - unint64_to_dw_aint(&dw, val); - erts_smp_dw_atomic_set_relb(&ptab->vola.tile.last_data, &dw); + erts_smp_atomic64_set_relb(&ptab->vola.tile.last_data, (erts_aint64_t) val); } static ERTS_INLINE Uint64 last_data_read_nob(ErtsPTab *ptab) { - erts_dw_aint_t dw; - erts_smp_dw_atomic_read_nob(&ptab->vola.tile.last_data, &dw); - return dw_aint_to_uint64(&dw); + return (Uint64) erts_smp_atomic64_read_nob(&ptab->vola.tile.last_data); } static ERTS_INLINE Uint64 last_data_read_acqb(ErtsPTab *ptab) { - erts_dw_aint_t dw; - erts_smp_dw_atomic_read_acqb(&ptab->vola.tile.last_data, &dw); - return dw_aint_to_uint64(&dw); + return (Uint64) erts_smp_atomic64_read_acqb(&ptab->vola.tile.last_data); } static ERTS_INLINE Uint64 last_data_cmpxchg_relb(ErtsPTab *ptab, Uint64 new, Uint64 exp) { - erts_dw_aint_t dw_new, dw_xchg; - - unint64_to_dw_aint(&dw_new, new); - unint64_to_dw_aint(&dw_xchg, exp); - - if (erts_smp_dw_atomic_cmpxchg_relb(&ptab->vola.tile.last_data, - &dw_new, - &dw_xchg)) - return exp; - else - return dw_aint_to_uint64(&dw_xchg); -} - -#elif defined(ARCH_64) - -union { - erts_smp_atomic_t pid_data; - char align[ERTS_CACHE_LINE_SIZE]; -} last erts_align_attribute(ERTS_CACHE_LINE_SIZE); - -static ERTS_INLINE void -last_data_init_nob(ErtsPTab *ptab, Uint64 val) -{ - erts_smp_atomic_init_nob(&ptab->vola.tile.last_data, (erts_aint_t) val); + return (Uint64) erts_smp_atomic64_cmpxchg_relb(&ptab->vola.tile.last_data, + (erts_aint64_t) new, + (erts_aint64_t) exp); } -static ERTS_INLINE void -last_data_set_relb(ErtsPTab *ptab, Uint64 val) -{ - erts_smp_atomic_set_relb(&ptab->vola.tile.last_data, (erts_aint_t) val); -} - -static ERTS_INLINE Uint64 -last_data_read_nob(ErtsPTab *ptab) -{ - return (Uint64) erts_smp_atomic_read_nob(&ptab->vola.tile.last_data); -} - -static ERTS_INLINE Uint64 -last_data_read_acqb(ErtsPTab *ptab) -{ - return (Uint64) erts_smp_atomic_read_acqb(&ptab->vola.tile.last_data); -} - -static ERTS_INLINE Uint64 -last_data_cmpxchg_relb(ErtsPTab *ptab, Uint64 new, Uint64 exp) -{ - return (Uint64) erts_smp_atomic_cmpxchg_relb(&ptab->vola.tile.last_data, - (erts_aint_t) new, - (erts_aint_t) exp); -} - -#else -# error "Not 64-bit, nor 32-bit architecture..." -#endif - static ERTS_INLINE int last_data_cmp(Uint64 ld1, Uint64 ld2) { diff --git a/erts/emulator/beam/erl_ptab.h b/erts/emulator/beam/erl_ptab.h index e3e05f14af..876241159b 100644 --- a/erts/emulator/beam/erl_ptab.h +++ b/erts/emulator/beam/erl_ptab.h @@ -88,11 +88,7 @@ typedef struct { } ErtsPTabListData; typedef struct { -#ifdef ARCH_32 - erts_smp_dw_atomic_t last_data; -#else - erts_smp_atomic_t last_data; -#endif + erts_smp_atomic64_t last_data; erts_smp_atomic32_t count; erts_smp_atomic32_t aid_ix; erts_smp_atomic32_t fid_ix; diff --git a/erts/emulator/beam/erl_smp.h b/erts/emulator/beam/erl_smp.h index c38ef47d87..6c40edeb3e 100644 --- a/erts/emulator/beam/erl_smp.h +++ b/erts/emulator/beam/erl_smp.h @@ -60,6 +60,7 @@ typedef erts_tsd_key_t erts_smp_tsd_key_t; #define erts_smp_dw_atomic_t erts_dw_atomic_t #define erts_smp_atomic_t erts_atomic_t #define erts_smp_atomic32_t erts_atomic32_t +#define erts_smp_atomic64_t erts_atomic64_t typedef erts_spinlock_t erts_smp_spinlock_t; typedef erts_rwlock_t erts_smp_rwlock_t; void erts_thr_fatal_error(int, char *); /* implemented in erl_init.c */ @@ -95,6 +96,7 @@ typedef int erts_smp_tsd_key_t; #define erts_smp_dw_atomic_t erts_no_dw_atomic_t #define erts_smp_atomic_t erts_no_atomic_t #define erts_smp_atomic32_t erts_no_atomic32_t +#define erts_smp_atomic64_t erts_no_atomic64_t #if __GNUC__ > 2 typedef struct { } erts_smp_spinlock_t; typedef struct { } erts_smp_rwlock_t; @@ -489,6 +491,116 @@ ERTS_GLB_INLINE void erts_smp_thr_sigwait(const sigset_t *set, int *sig); #define erts_smp_atomic32_set_dirty erts_atomic32_set_dirty #define erts_smp_atomic32_read_dirty erts_atomic32_read_dirty +/* 64-bit atomics */ + +#define erts_smp_atomic64_init_nob erts_atomic64_init_nob +#define erts_smp_atomic64_set_nob erts_atomic64_set_nob +#define erts_smp_atomic64_read_nob erts_atomic64_read_nob +#define erts_smp_atomic64_inc_read_nob erts_atomic64_inc_read_nob +#define erts_smp_atomic64_dec_read_nob erts_atomic64_dec_read_nob +#define erts_smp_atomic64_inc_nob erts_atomic64_inc_nob +#define erts_smp_atomic64_dec_nob erts_atomic64_dec_nob +#define erts_smp_atomic64_add_read_nob erts_atomic64_add_read_nob +#define erts_smp_atomic64_add_nob erts_atomic64_add_nob +#define erts_smp_atomic64_read_bor_nob erts_atomic64_read_bor_nob +#define erts_smp_atomic64_read_band_nob erts_atomic64_read_band_nob +#define erts_smp_atomic64_xchg_nob erts_atomic64_xchg_nob +#define erts_smp_atomic64_cmpxchg_nob erts_atomic64_cmpxchg_nob +#define erts_smp_atomic64_read_bset_nob erts_atomic64_read_bset_nob + +#define erts_smp_atomic64_init_mb erts_atomic64_init_mb +#define erts_smp_atomic64_set_mb erts_atomic64_set_mb +#define erts_smp_atomic64_read_mb erts_atomic64_read_mb +#define erts_smp_atomic64_inc_read_mb erts_atomic64_inc_read_mb +#define erts_smp_atomic64_dec_read_mb erts_atomic64_dec_read_mb +#define erts_smp_atomic64_inc_mb erts_atomic64_inc_mb +#define erts_smp_atomic64_dec_mb erts_atomic64_dec_mb +#define erts_smp_atomic64_add_read_mb erts_atomic64_add_read_mb +#define erts_smp_atomic64_add_mb erts_atomic64_add_mb +#define erts_smp_atomic64_read_bor_mb erts_atomic64_read_bor_mb +#define erts_smp_atomic64_read_band_mb erts_atomic64_read_band_mb +#define erts_smp_atomic64_xchg_mb erts_atomic64_xchg_mb +#define erts_smp_atomic64_cmpxchg_mb erts_atomic64_cmpxchg_mb +#define erts_smp_atomic64_read_bset_mb erts_atomic64_read_bset_mb + +#define erts_smp_atomic64_init_acqb erts_atomic64_init_acqb +#define erts_smp_atomic64_set_acqb erts_atomic64_set_acqb +#define erts_smp_atomic64_read_acqb erts_atomic64_read_acqb +#define erts_smp_atomic64_inc_read_acqb erts_atomic64_inc_read_acqb +#define erts_smp_atomic64_dec_read_acqb erts_atomic64_dec_read_acqb +#define erts_smp_atomic64_inc_acqb erts_atomic64_inc_acqb +#define erts_smp_atomic64_dec_acqb erts_atomic64_dec_acqb +#define erts_smp_atomic64_add_read_acqb erts_atomic64_add_read_acqb +#define erts_smp_atomic64_add_acqb erts_atomic64_add_acqb +#define erts_smp_atomic64_read_bor_acqb erts_atomic64_read_bor_acqb +#define erts_smp_atomic64_read_band_acqb erts_atomic64_read_band_acqb +#define erts_smp_atomic64_xchg_acqb erts_atomic64_xchg_acqb +#define erts_smp_atomic64_cmpxchg_acqb erts_atomic64_cmpxchg_acqb +#define erts_smp_atomic64_read_bset_acqb erts_atomic64_read_bset_acqb + +#define erts_smp_atomic64_init_relb erts_atomic64_init_relb +#define erts_smp_atomic64_set_relb erts_atomic64_set_relb +#define erts_smp_atomic64_read_relb erts_atomic64_read_relb +#define erts_smp_atomic64_inc_read_relb erts_atomic64_inc_read_relb +#define erts_smp_atomic64_dec_read_relb erts_atomic64_dec_read_relb +#define erts_smp_atomic64_inc_relb erts_atomic64_inc_relb +#define erts_smp_atomic64_dec_relb erts_atomic64_dec_relb +#define erts_smp_atomic64_add_read_relb erts_atomic64_add_read_relb +#define erts_smp_atomic64_add_relb erts_atomic64_add_relb +#define erts_smp_atomic64_read_bor_relb erts_atomic64_read_bor_relb +#define erts_smp_atomic64_read_band_relb erts_atomic64_read_band_relb +#define erts_smp_atomic64_xchg_relb erts_atomic64_xchg_relb +#define erts_smp_atomic64_cmpxchg_relb erts_atomic64_cmpxchg_relb +#define erts_smp_atomic64_read_bset_relb erts_atomic64_read_bset_relb + +#define erts_smp_atomic64_init_ddrb erts_atomic64_init_ddrb +#define erts_smp_atomic64_set_ddrb erts_atomic64_set_ddrb +#define erts_smp_atomic64_read_ddrb erts_atomic64_read_ddrb +#define erts_smp_atomic64_inc_read_ddrb erts_atomic64_inc_read_ddrb +#define erts_smp_atomic64_dec_read_ddrb erts_atomic64_dec_read_ddrb +#define erts_smp_atomic64_inc_ddrb erts_atomic64_inc_ddrb +#define erts_smp_atomic64_dec_ddrb erts_atomic64_dec_ddrb +#define erts_smp_atomic64_add_read_ddrb erts_atomic64_add_read_ddrb +#define erts_smp_atomic64_add_ddrb erts_atomic64_add_ddrb +#define erts_smp_atomic64_read_bor_ddrb erts_atomic64_read_bor_ddrb +#define erts_smp_atomic64_read_band_ddrb erts_atomic64_read_band_ddrb +#define erts_smp_atomic64_xchg_ddrb erts_atomic64_xchg_ddrb +#define erts_smp_atomic64_cmpxchg_ddrb erts_atomic64_cmpxchg_ddrb +#define erts_smp_atomic64_read_bset_ddrb erts_atomic64_read_bset_ddrb + +#define erts_smp_atomic64_init_rb erts_atomic64_init_rb +#define erts_smp_atomic64_set_rb erts_atomic64_set_rb +#define erts_smp_atomic64_read_rb erts_atomic64_read_rb +#define erts_smp_atomic64_inc_read_rb erts_atomic64_inc_read_rb +#define erts_smp_atomic64_dec_read_rb erts_atomic64_dec_read_rb +#define erts_smp_atomic64_inc_rb erts_atomic64_inc_rb +#define erts_smp_atomic64_dec_rb erts_atomic64_dec_rb +#define erts_smp_atomic64_add_read_rb erts_atomic64_add_read_rb +#define erts_smp_atomic64_add_rb erts_atomic64_add_rb +#define erts_smp_atomic64_read_bor_rb erts_atomic64_read_bor_rb +#define erts_smp_atomic64_read_band_rb erts_atomic64_read_band_rb +#define erts_smp_atomic64_xchg_rb erts_atomic64_xchg_rb +#define erts_smp_atomic64_cmpxchg_rb erts_atomic64_cmpxchg_rb +#define erts_smp_atomic64_read_bset_rb erts_atomic64_read_bset_rb + +#define erts_smp_atomic64_init_wb erts_atomic64_init_wb +#define erts_smp_atomic64_set_wb erts_atomic64_set_wb +#define erts_smp_atomic64_read_wb erts_atomic64_read_wb +#define erts_smp_atomic64_inc_read_wb erts_atomic64_inc_read_wb +#define erts_smp_atomic64_dec_read_wb erts_atomic64_dec_read_wb +#define erts_smp_atomic64_inc_wb erts_atomic64_inc_wb +#define erts_smp_atomic64_dec_wb erts_atomic64_dec_wb +#define erts_smp_atomic64_add_read_wb erts_atomic64_add_read_wb +#define erts_smp_atomic64_add_wb erts_atomic64_add_wb +#define erts_smp_atomic64_read_bor_wb erts_atomic64_read_bor_wb +#define erts_smp_atomic64_read_band_wb erts_atomic64_read_band_wb +#define erts_smp_atomic64_xchg_wb erts_atomic64_xchg_wb +#define erts_smp_atomic64_cmpxchg_wb erts_atomic64_cmpxchg_wb +#define erts_smp_atomic64_read_bset_wb erts_atomic64_read_bset_wb + +#define erts_smp_atomic64_set_dirty erts_atomic64_set_dirty +#define erts_smp_atomic64_read_dirty erts_atomic64_read_dirty + #else /* !ERTS_SMP */ /* Double word size atomics */ @@ -751,6 +863,116 @@ ERTS_GLB_INLINE void erts_smp_thr_sigwait(const sigset_t *set, int *sig); #define erts_smp_atomic32_set_dirty erts_no_atomic32_set #define erts_smp_atomic32_read_dirty erts_no_atomic32_read +/* 64-bit atomics */ + +#define erts_smp_atomic64_init_nob erts_no_atomic64_set +#define erts_smp_atomic64_set_nob erts_no_atomic64_set +#define erts_smp_atomic64_read_nob erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_nob erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_nob erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_nob erts_no_atomic64_inc +#define erts_smp_atomic64_dec_nob erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_nob erts_no_atomic64_add_read +#define erts_smp_atomic64_add_nob erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_nob erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_nob erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_nob erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_nob erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_nob erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_mb erts_no_atomic64_set +#define erts_smp_atomic64_set_mb erts_no_atomic64_set +#define erts_smp_atomic64_read_mb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_mb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_mb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_mb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_mb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_mb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_mb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_mb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_mb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_mb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_mb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_mb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_acqb erts_no_atomic64_set +#define erts_smp_atomic64_set_acqb erts_no_atomic64_set +#define erts_smp_atomic64_read_acqb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_acqb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_acqb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_acqb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_acqb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_acqb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_acqb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_acqb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_acqb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_acqb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_acqb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_acqb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_relb erts_no_atomic64_set +#define erts_smp_atomic64_set_relb erts_no_atomic64_set +#define erts_smp_atomic64_read_relb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_relb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_relb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_relb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_relb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_relb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_relb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_relb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_relb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_relb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_relb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_relb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_ddrb erts_no_atomic64_set +#define erts_smp_atomic64_set_ddrb erts_no_atomic64_set +#define erts_smp_atomic64_read_ddrb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_ddrb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_ddrb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_ddrb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_ddrb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_ddrb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_ddrb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_ddrb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_ddrb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_ddrb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_ddrb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_ddrb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_rb erts_no_atomic64_set +#define erts_smp_atomic64_set_rb erts_no_atomic64_set +#define erts_smp_atomic64_read_rb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_rb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_rb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_rb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_rb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_rb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_rb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_rb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_rb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_rb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_rb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_rb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_init_wb erts_no_atomic64_set +#define erts_smp_atomic64_set_wb erts_no_atomic64_set +#define erts_smp_atomic64_read_wb erts_no_atomic64_read +#define erts_smp_atomic64_inc_read_wb erts_no_atomic64_inc_read +#define erts_smp_atomic64_dec_read_wb erts_no_atomic64_dec_read +#define erts_smp_atomic64_inc_wb erts_no_atomic64_inc +#define erts_smp_atomic64_dec_wb erts_no_atomic64_dec +#define erts_smp_atomic64_add_read_wb erts_no_atomic64_add_read +#define erts_smp_atomic64_add_wb erts_no_atomic64_add +#define erts_smp_atomic64_read_bor_wb erts_no_atomic64_read_bor +#define erts_smp_atomic64_read_band_wb erts_no_atomic64_read_band +#define erts_smp_atomic64_xchg_wb erts_no_atomic64_xchg +#define erts_smp_atomic64_cmpxchg_wb erts_no_atomic64_cmpxchg +#define erts_smp_atomic64_read_bset_wb erts_no_atomic64_read_bset + +#define erts_smp_atomic64_set_dirty erts_no_atomic64_set +#define erts_smp_atomic64_read_dirty erts_no_atomic64_read + #endif /* !ERTS_SMP */ #if ERTS_GLB_INLINE_INCL_FUNC_DEF diff --git a/erts/emulator/beam/erl_term.h b/erts/emulator/beam/erl_term.h index f10a3a9d38..37014ccf94 100644 --- a/erts/emulator/beam/erl_term.h +++ b/erts/emulator/beam/erl_term.h @@ -112,11 +112,11 @@ struct erl_node_; /* Declared in erl_node_tables.h */ * 1000 REFC_BINARY | | * 1001 HEAP_BINARY | BINARIES | * 1010 SUB_BINARY | | - * 1011 Not used + * 1011 Not used; see comment below * 1100 EXTERNAL_PID | | * 1101 EXTERNAL_PORT | EXTERNAL THINGS | * 1110 EXTERNAL_REF | | - * 1111 Not used + * 1111 MAP * * COMMENTS: * @@ -140,10 +140,11 @@ struct erl_node_; /* Declared in erl_node_tables.h */ #define REFC_BINARY_SUBTAG (0x8 << _TAG_PRIMARY_SIZE) /* BINARY */ #define HEAP_BINARY_SUBTAG (0x9 << _TAG_PRIMARY_SIZE) /* BINARY */ #define SUB_BINARY_SUBTAG (0xA << _TAG_PRIMARY_SIZE) /* BINARY */ -#define MAP_SUBTAG (0xB << _TAG_PRIMARY_SIZE) /* MAP */ +/* _BINARY_XXX_MASK depends on 0xB being unused */ #define EXTERNAL_PID_SUBTAG (0xC << _TAG_PRIMARY_SIZE) /* EXTERNAL_PID */ #define EXTERNAL_PORT_SUBTAG (0xD << _TAG_PRIMARY_SIZE) /* EXTERNAL_PORT */ #define EXTERNAL_REF_SUBTAG (0xE << _TAG_PRIMARY_SIZE) /* EXTERNAL_REF */ +#define MAP_SUBTAG (0xF << _TAG_PRIMARY_SIZE) /* MAP */ #define _TAG_HEADER_ARITYVAL (TAG_PRIMARY_HEADER|ARITYVAL_SUBTAG) @@ -156,11 +157,11 @@ struct erl_node_; /* Declared in erl_node_tables.h */ #define _TAG_HEADER_REFC_BIN (TAG_PRIMARY_HEADER|REFC_BINARY_SUBTAG) #define _TAG_HEADER_HEAP_BIN (TAG_PRIMARY_HEADER|HEAP_BINARY_SUBTAG) #define _TAG_HEADER_SUB_BIN (TAG_PRIMARY_HEADER|SUB_BINARY_SUBTAG) -#define _TAG_HEADER_MAP (TAG_PRIMARY_HEADER|MAP_SUBTAG) #define _TAG_HEADER_EXTERNAL_PID (TAG_PRIMARY_HEADER|EXTERNAL_PID_SUBTAG) #define _TAG_HEADER_EXTERNAL_PORT (TAG_PRIMARY_HEADER|EXTERNAL_PORT_SUBTAG) #define _TAG_HEADER_EXTERNAL_REF (TAG_PRIMARY_HEADER|EXTERNAL_REF_SUBTAG) #define _TAG_HEADER_BIN_MATCHSTATE (TAG_PRIMARY_HEADER|BIN_MATCHSTATE_SUBTAG) +#define _TAG_HEADER_MAP (TAG_PRIMARY_HEADER|MAP_SUBTAG) #define _TAG_HEADER_MASK 0x3F @@ -892,7 +893,8 @@ typedef struct external_thing_ { (((x) & _TAG_HEADER_MASK) == _TAG_HEADER_EXTERNAL_REF) #define is_external_header(x) \ - (((x) & (_TAG_HEADER_MASK-_BINARY_XXX_MASK)) == _TAG_HEADER_EXTERNAL_PID) + (((x) & (_TAG_HEADER_MASK-_BINARY_XXX_MASK)) == _TAG_HEADER_EXTERNAL_PID \ + && ((x) & _TAG_HEADER_MASK) != _TAG_HEADER_MAP) #define is_external(x) (is_boxed((x)) && is_external_header(*boxed_val((x)))) diff --git a/erts/emulator/beam/erl_thr_progress.c b/erts/emulator/beam/erl_thr_progress.c index 545a0343d0..664c479eb6 100644 --- a/erts/emulator/beam/erl_thr_progress.c +++ b/erts/emulator/beam/erl_thr_progress.c @@ -115,70 +115,24 @@ #undef read_nob #define read_nob erts_thr_prgr_read_nob__ -#ifdef ARCH_64 - static ERTS_INLINE void set_mb(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_set_mb(atmc, val); + erts_atomic64_set_mb(atmc, (erts_aint64_t) val); } static ERTS_INLINE void set_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_set_nob(atmc, val); + erts_atomic64_set_nob(atmc, (erts_aint64_t) val); } static ERTS_INLINE void init_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) { - erts_atomic_init_nob(atmc, val); -} - -#else - -#undef dw_aint_to_val -#define dw_aint_to_val erts_thr_prgr_dw_aint_to_val__ - -static void -val_to_dw_aint(erts_dw_aint_t *dw_aint, ErtsThrPrgrVal val) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - dw_aint->dw_sint = (ETHR_SU_DW_NAINT_T__) val; -#else - dw_aint->sint[ERTS_DW_AINT_LOW_WORD] - = (erts_aint_t) (val & 0xffffffff); - dw_aint->sint[ERTS_DW_AINT_HIGH_WORD] - = (erts_aint_t) ((val >> 32) & 0xffffffff); -#endif + erts_atomic64_init_nob(atmc, (erts_aint64_t) val); } -static ERTS_INLINE void -set_mb(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_set_mb(atmc, &dw_aint); -} - -static ERTS_INLINE void -set_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_set_nob(atmc, &dw_aint); -} - -static ERTS_INLINE void -init_nob(ERTS_THR_PRGR_ATOMIC *atmc, ErtsThrPrgrVal val) -{ - erts_dw_aint_t dw_aint; - val_to_dw_aint(&dw_aint, val); - erts_dw_atomic_init_nob(atmc, &dw_aint); -} - -#endif - /* #define ERTS_THR_PROGRESS_STATE_DEBUG */ #ifdef ERTS_THR_PROGRESS_STATE_DEBUG diff --git a/erts/emulator/beam/erl_thr_progress.h b/erts/emulator/beam/erl_thr_progress.h index 5f392944c2..03ddbd467c 100644 --- a/erts/emulator/beam/erl_thr_progress.h +++ b/erts/emulator/beam/erl_thr_progress.h @@ -115,11 +115,7 @@ struct ErtsThrPrgrLaterOp_ { extern erts_tsd_key_t erts_thr_prgr_data_key__; -#ifdef ARCH_64 -# define ERTS_THR_PRGR_ATOMIC erts_atomic_t -#else /* ARCH_32 */ -# define ERTS_THR_PRGR_ATOMIC erts_dw_atomic_t -#endif +#define ERTS_THR_PRGR_ATOMIC erts_atomic64_t typedef struct { void *arg; @@ -158,10 +154,6 @@ void erts_thr_progress_unmanaged_continue__(int umrefc_ix); void erts_thr_progress_dbg_print_state(void); -#ifdef ARCH_32 -#define ERTS_THR_PRGR_ATOMIC erts_dw_atomic_t -ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_dw_aint_to_val__(erts_dw_aint_t *dw_aint); -#endif ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc); ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc); ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc); @@ -184,68 +176,24 @@ ERTS_GLB_INLINE int erts_thr_progress_has_reached(ErtsThrPrgrVal val); #if ERTS_GLB_INLINE_INCL_FUNC_DEF -#ifdef ARCH_64 - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_nob(atmc); -} - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_acqb(atmc); -} - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc) -{ - return (ErtsThrPrgrVal) erts_atomic_read_mb(atmc); -} - -#else /* ARCH_32 */ - -ERTS_GLB_INLINE ErtsThrPrgrVal -erts_thr_prgr_dw_aint_to_val__(erts_dw_aint_t *dw_aint) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (ErtsThrPrgrVal) dw_aint->dw_sint; -#else - ErtsThrPrgrVal res; - res = (ErtsThrPrgrVal) ((Uint32) dw_aint->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (ErtsThrPrgrVal) ((Uint32) dw_aint->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_nob(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_nob(atmc); } ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_acqb(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_acqb(atmc); } ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc) { - erts_dw_aint_t dw_aint; - erts_dw_atomic_read_mb(atmc, &dw_aint); - return erts_thr_prgr_dw_aint_to_val__(&dw_aint); + return (ErtsThrPrgrVal) erts_atomic64_read_mb(atmc); } -#endif - ERTS_GLB_INLINE int erts_thr_progress_is_managed_thread(void) { diff --git a/erts/emulator/beam/erl_threads.h b/erts/emulator/beam/erl_threads.h index 80026104db..7214f3ea33 100644 --- a/erts/emulator/beam/erl_threads.h +++ b/erts/emulator/beam/erl_threads.h @@ -344,6 +344,16 @@ typedef ethr_ts_event erts_tse_t; #define erts_aint32_t ethr_sint32_t #define erts_atomic32_t ethr_atomic32_t +#if defined(ARCH_32) +# define erts_atomic64_t ethr_dw_atomic_t +# define erts_aint64_t ethr_sint64_t +#elif defined(ARCH_64) +# define erts_atomic64_t ethr_atomic_t +# define erts_aint64_t ethr_sint_t +#else +# error "Not supported architecture" +#endif + #define ERTS_DW_AINT_HIGH_WORD ETHR_DW_SINT_HIGH_WORD #define ERTS_DW_AINT_LOW_WORD ETHR_DW_SINT_LOW_WORD @@ -414,10 +424,12 @@ typedef int erts_tse_t; typedef struct { SWord sint[2]; } erts_dw_aint_t; typedef SWord erts_aint_t; typedef Sint32 erts_aint32_t; +typedef Sint64 erts_aint64_t; #define erts_dw_atomic_t erts_dw_aint_t #define erts_atomic_t erts_aint_t #define erts_atomic32_t erts_aint32_t +#define erts_atomic64_t erts_aint64_t #if __GNUC__ > 2 typedef struct { } erts_spinlock_t; @@ -446,6 +458,7 @@ typedef struct { int gcc_is_buggy; } erts_rwlock_t; #define erts_no_dw_atomic_t erts_dw_aint_t #define erts_no_atomic_t erts_aint_t #define erts_no_atomic32_t erts_aint32_t +#define erts_no_atomic64_t erts_aint64_t #define ERTS_AINT_NULL ((erts_aint_t) NULL) @@ -570,6 +583,29 @@ ERTS_GLB_INLINE erts_aint32_t erts_no_atomic32_cmpxchg(erts_no_atomic32_t *xchgp ERTS_GLB_INLINE erts_aint32_t erts_no_atomic32_read_bset(erts_no_atomic32_t *var, erts_aint32_t mask, erts_aint32_t set); +ERTS_GLB_INLINE void erts_no_atomic64_set(erts_no_atomic64_t *var, + erts_aint64_t i); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read(erts_no_atomic64_t *var); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_inc_read(erts_no_atomic64_t *incp); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_dec_read(erts_no_atomic64_t *decp); +ERTS_GLB_INLINE void erts_no_atomic64_inc(erts_no_atomic64_t *incp); +ERTS_GLB_INLINE void erts_no_atomic64_dec(erts_no_atomic64_t *decp); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_add_read(erts_no_atomic64_t *addp, + erts_aint64_t i); +ERTS_GLB_INLINE void erts_no_atomic64_add(erts_no_atomic64_t *addp, + erts_aint64_t i); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_bor(erts_no_atomic64_t *var, + erts_aint64_t mask); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_band(erts_no_atomic64_t *var, + erts_aint64_t mask); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_xchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_cmpxchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new, + erts_aint64_t expected); +ERTS_GLB_INLINE erts_aint64_t erts_no_atomic64_read_bset(erts_no_atomic64_t *var, + erts_aint64_t mask, + erts_aint64_t set); ERTS_GLB_INLINE void erts_spinlock_init_x_opt(erts_spinlock_t *lock, char *name, @@ -1200,6 +1236,441 @@ erts_atomic32_read_dirty(erts_atomic32_t *var) #endif +/* 64-bit atomics */ + +#if defined(ARCH_64) + +#define erts_atomic64_init_nob ethr_atomic_init +#define erts_atomic64_set_nob ethr_atomic_set +#define erts_atomic64_read_nob ethr_atomic_read +#define erts_atomic64_inc_read_nob ethr_atomic_inc_read +#define erts_atomic64_dec_read_nob ethr_atomic_dec_read +#define erts_atomic64_inc_nob ethr_atomic_inc +#define erts_atomic64_dec_nob ethr_atomic_dec +#define erts_atomic64_add_read_nob ethr_atomic_add_read +#define erts_atomic64_add_nob ethr_atomic_add +#define erts_atomic64_read_bor_nob ethr_atomic_read_bor +#define erts_atomic64_read_band_nob ethr_atomic_read_band +#define erts_atomic64_xchg_nob ethr_atomic_xchg +#define erts_atomic64_cmpxchg_nob ethr_atomic_cmpxchg +#define erts_atomic64_read_bset_nob erts_atomic_read_bset_nob + +#define erts_atomic64_init_mb ethr_atomic_init_mb +#define erts_atomic64_set_mb ethr_atomic_set_mb +#define erts_atomic64_read_mb ethr_atomic_read_mb +#define erts_atomic64_inc_read_mb ethr_atomic_inc_read_mb +#define erts_atomic64_dec_read_mb ethr_atomic_dec_read_mb +#define erts_atomic64_inc_mb ethr_atomic_inc_mb +#define erts_atomic64_dec_mb ethr_atomic_dec_mb +#define erts_atomic64_add_read_mb ethr_atomic_add_read_mb +#define erts_atomic64_add_mb ethr_atomic_add_mb +#define erts_atomic64_read_bor_mb ethr_atomic_read_bor_mb +#define erts_atomic64_read_band_mb ethr_atomic_read_band_mb +#define erts_atomic64_xchg_mb ethr_atomic_xchg_mb +#define erts_atomic64_cmpxchg_mb ethr_atomic_cmpxchg_mb +#define erts_atomic64_read_bset_mb erts_atomic_read_bset_mb + +#define erts_atomic64_init_acqb ethr_atomic_init_acqb +#define erts_atomic64_set_acqb ethr_atomic_set_acqb +#define erts_atomic64_read_acqb ethr_atomic_read_acqb +#define erts_atomic64_inc_read_acqb ethr_atomic_inc_read_acqb +#define erts_atomic64_dec_read_acqb ethr_atomic_dec_read_acqb +#define erts_atomic64_inc_acqb ethr_atomic_inc_acqb +#define erts_atomic64_dec_acqb ethr_atomic_dec_acqb +#define erts_atomic64_add_read_acqb ethr_atomic_add_read_acqb +#define erts_atomic64_add_acqb ethr_atomic_add_acqb +#define erts_atomic64_read_bor_acqb ethr_atomic_read_bor_acqb +#define erts_atomic64_read_band_acqb ethr_atomic_read_band_acqb +#define erts_atomic64_xchg_acqb ethr_atomic_xchg_acqb +#define erts_atomic64_cmpxchg_acqb ethr_atomic_cmpxchg_acqb +#define erts_atomic64_read_bset_acqb erts_atomic_read_bset_acqb + +#define erts_atomic64_init_relb ethr_atomic_init_relb +#define erts_atomic64_set_relb ethr_atomic_set_relb +#define erts_atomic64_read_relb ethr_atomic_read_relb +#define erts_atomic64_inc_read_relb ethr_atomic_inc_read_relb +#define erts_atomic64_dec_read_relb ethr_atomic_dec_read_relb +#define erts_atomic64_inc_relb ethr_atomic_inc_relb +#define erts_atomic64_dec_relb ethr_atomic_dec_relb +#define erts_atomic64_add_read_relb ethr_atomic_add_read_relb +#define erts_atomic64_add_relb ethr_atomic_add_relb +#define erts_atomic64_read_bor_relb ethr_atomic_read_bor_relb +#define erts_atomic64_read_band_relb ethr_atomic_read_band_relb +#define erts_atomic64_xchg_relb ethr_atomic_xchg_relb +#define erts_atomic64_cmpxchg_relb ethr_atomic_cmpxchg_relb +#define erts_atomic64_read_bset_relb erts_atomic_read_bset_relb + +#define erts_atomic64_init_ddrb ethr_atomic_init_ddrb +#define erts_atomic64_set_ddrb ethr_atomic_set_ddrb +#define erts_atomic64_read_ddrb ethr_atomic_read_ddrb +#define erts_atomic64_inc_read_ddrb ethr_atomic_inc_read_ddrb +#define erts_atomic64_dec_read_ddrb ethr_atomic_dec_read_ddrb +#define erts_atomic64_inc_ddrb ethr_atomic_inc_ddrb +#define erts_atomic64_dec_ddrb ethr_atomic_dec_ddrb +#define erts_atomic64_add_read_ddrb ethr_atomic_add_read_ddrb +#define erts_atomic64_add_ddrb ethr_atomic_add_ddrb +#define erts_atomic64_read_bor_ddrb ethr_atomic_read_bor_ddrb +#define erts_atomic64_read_band_ddrb ethr_atomic_read_band_ddrb +#define erts_atomic64_xchg_ddrb ethr_atomic_xchg_ddrb +#define erts_atomic64_cmpxchg_ddrb ethr_atomic_cmpxchg_ddrb +#define erts_atomic64_read_bset_ddrb erts_atomic_read_bset_ddrb + +#define erts_atomic64_init_rb ethr_atomic_init_rb +#define erts_atomic64_set_rb ethr_atomic_set_rb +#define erts_atomic64_read_rb ethr_atomic_read_rb +#define erts_atomic64_inc_read_rb ethr_atomic_inc_read_rb +#define erts_atomic64_dec_read_rb ethr_atomic_dec_read_rb +#define erts_atomic64_inc_rb ethr_atomic_inc_rb +#define erts_atomic64_dec_rb ethr_atomic_dec_rb +#define erts_atomic64_add_read_rb ethr_atomic_add_read_rb +#define erts_atomic64_add_rb ethr_atomic_add_rb +#define erts_atomic64_read_bor_rb ethr_atomic_read_bor_rb +#define erts_atomic64_read_band_rb ethr_atomic_read_band_rb +#define erts_atomic64_xchg_rb ethr_atomic_xchg_rb +#define erts_atomic64_cmpxchg_rb ethr_atomic_cmpxchg_rb +#define erts_atomic64_read_bset_rb erts_atomic_read_bset_rb + +#define erts_atomic64_init_wb ethr_atomic_init_wb +#define erts_atomic64_set_wb ethr_atomic_set_wb +#define erts_atomic64_read_wb ethr_atomic_read_wb +#define erts_atomic64_inc_read_wb ethr_atomic_inc_read_wb +#define erts_atomic64_dec_read_wb ethr_atomic_dec_read_wb +#define erts_atomic64_inc_wb ethr_atomic_inc_wb +#define erts_atomic64_dec_wb ethr_atomic_dec_wb +#define erts_atomic64_add_read_wb ethr_atomic_add_read_wb +#define erts_atomic64_add_wb ethr_atomic_add_wb +#define erts_atomic64_read_bor_wb ethr_atomic_read_bor_wb +#define erts_atomic64_read_band_wb ethr_atomic_read_band_wb +#define erts_atomic64_xchg_wb ethr_atomic_xchg_wb +#define erts_atomic64_cmpxchg_wb ethr_atomic_cmpxchg_wb +#define erts_atomic64_read_bset_wb erts_atomic_read_bset_wb + +#define erts_atomic64_set_dirty erts_atomic_set_dirty +#define erts_atomic64_read_dirty erts_atomic_read_dirty + +#elif defined(ARCH_32) + +#undef ERTS_ATOMIC64_OPS_DECL__ + +#define ERTS_ATOMIC64_OPS_DECL__(BARRIER) \ +ERTS_GLB_INLINE void \ +erts_atomic64_init_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE void \ +erts_atomic64_set_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_inc_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_dec_read_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE void \ +erts_atomic64_inc_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE void \ +erts_atomic64_dec_ ## BARRIER(erts_atomic64_t *var); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_add_read_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE void \ +erts_atomic64_add_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bor_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_band_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_xchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_cmpxchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t new, \ + erts_aint64_t exp); \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bset_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t mask, \ + erts_aint64_t set) + +ERTS_ATOMIC64_OPS_DECL__(nob); +ERTS_ATOMIC64_OPS_DECL__(mb); +ERTS_ATOMIC64_OPS_DECL__(acqb); +ERTS_ATOMIC64_OPS_DECL__(relb); +ERTS_ATOMIC64_OPS_DECL__(ddrb); +ERTS_ATOMIC64_OPS_DECL__(rb); +ERTS_ATOMIC64_OPS_DECL__(wb); + +#undef ERTS_ATOMIC64_OPS_DECL__ + +ERTS_GLB_INLINE void +erts_atomic64_set_dirty(erts_atomic64_t *var, erts_aint64_t val); +ERTS_GLB_INLINE erts_aint64_t +erts_atomic64_read_dirty(erts_atomic64_t *var); + +#if ERTS_GLB_INLINE_INCL_FUNC_DEF + +/* + * The ethr_dw_atomic_*_nob() functions below + * are here to make it possible for the + * ERTS_ATOMIC64_OPS_IMPL__() to map erts + * barriers to ethread barriers... + */ +static ERTS_INLINE void +ethr_dw_atomic_init_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_init(var, val); +} + +static ERTS_INLINE void +ethr_dw_atomic_set_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_set(var, val); +} + +static ERTS_INLINE void +ethr_dw_atomic_read_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *val) +{ + ethr_dw_atomic_read(var, val); +} + +static ERTS_INLINE int +ethr_dw_atomic_cmpxchg_nob(ethr_dw_atomic_t *var, + ethr_dw_sint_t *new, + ethr_dw_sint_t *xchg) +{ + return ethr_dw_atomic_cmpxchg(var, new, xchg); +} + +#undef ERTS_ATOMIC64_OPS_IMPL__ +#undef ERTS_ATOMIC64_DW_CMPXCHG_IMPL__ +#undef ERTS_DW_SINT_TO_AINT64__ +#undef ERTS_AINT64_TO_DW_SINT__ + +#ifdef ETHR_SU_DW_NAINT_T__ +#define ERTS_DW_SINT_TO_AINT64__(DW) \ + ((erts_aint64_t) DW.dw_sint) +#define ERTS_AINT64_TO_DW_SINT__(DW, AINT64) \ + (DW.dw_sint = (ETHR_SU_DW_NAINT_T__) AINT64) +#else /* !ETHR_SU_DW_NAINT_T__ */ +#define ERTS_DW_SINT_TO_AINT64__(DW) \ + ((((erts_aint64_t) DW.sint[ETHR_DW_SINT_HIGH_WORD]) << 32) \ + | (((erts_aint64_t) DW.sint[ETHR_DW_SINT_LOW_WORD]) \ + & ((erts_aint64_t) 0xffffffff))) +#define ERTS_AINT64_TO_DW_SINT__(DW, AINT64) \ + do { \ + DW.sint[ETHR_DW_SINT_LOW_WORD] = \ + (ethr_sint_t) (AINT64 & 0xffffffff); \ + DW.sint[ETHR_DW_SINT_HIGH_WORD] = \ + (ethr_sint_t) ((AINT64 >> 32) & 0xffffffff); \ + } while (0) +#endif /* !ETHR_SU_DW_NAINT_T__ */ + +#define ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(CmpXchgOp, \ + AVarP, XchgVar, NewVar, \ + ModificationCode) \ +do { \ + ethr_dw_sint_t dw_xchg__, dw_new__; \ + ethr_dw_atomic_read(AVarP, &dw_xchg__); \ + do { \ + XchgVar = ERTS_DW_SINT_TO_AINT64__(dw_xchg__); \ + { \ + ModificationCode; \ + } \ + ERTS_AINT64_TO_DW_SINT__(dw_new__, NewVar); \ + } while (!CmpXchgOp((AVarP), &dw_new__, &dw_xchg__)); \ +} while (0) + +#define ERTS_ATOMIC64_OPS_IMPL__(BARRIER) \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_init_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + ethr_dw_sint_t dw; \ + ERTS_AINT64_TO_DW_SINT__(dw, val); \ + ethr_dw_atomic_init_ ## BARRIER(var, &dw); \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_set_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + ethr_dw_sint_t dw; \ + ERTS_AINT64_TO_DW_SINT__(dw, val); \ + ethr_dw_atomic_set_ ## BARRIER(var, &dw); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + ethr_dw_sint_t dw; \ + ethr_dw_atomic_read_ ## BARRIER(var, &dw); \ + return ERTS_DW_SINT_TO_AINT64__(dw); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_inc_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + 1); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_dec_read_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg - 1); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_inc_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + 1); \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_dec_ ## BARRIER(erts_atomic64_t *var) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg - 1); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_add_read_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + val); \ + return new; \ +} \ + \ +ERTS_GLB_INLINE void \ +erts_atomic64_add_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg + val); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bor_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg | val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_band_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = xchg & val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_xchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t val) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + new = val); \ + return xchg; \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_cmpxchg_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t new, \ + erts_aint64_t exp) \ +{ \ + ethr_dw_sint_t dw_xchg, dw_new; \ + ERTS_AINT64_TO_DW_SINT__(dw_xchg, exp); \ + ERTS_AINT64_TO_DW_SINT__(dw_new, new); \ + if (ethr_dw_atomic_cmpxchg_ ## BARRIER(var, &dw_new, &dw_xchg)) \ + return exp; \ + return ERTS_DW_SINT_TO_AINT64__(dw_xchg); \ +} \ + \ +ERTS_GLB_INLINE erts_aint64_t \ +erts_atomic64_read_bset_ ## BARRIER(erts_atomic64_t *var, \ + erts_aint64_t mask, \ + erts_aint64_t set) \ +{ \ + erts_aint64_t xchg, new; \ + ERTS_ATOMIC64_DW_CMPXCHG_IMPL__(ethr_dw_atomic_cmpxchg_ ## BARRIER, \ + var, xchg, new, \ + { \ + new = xchg & ~mask; \ + new |= mask & set; \ + }); \ + return xchg; \ +} + +ERTS_ATOMIC64_OPS_IMPL__(nob) +ERTS_ATOMIC64_OPS_IMPL__(mb) +ERTS_ATOMIC64_OPS_IMPL__(acqb) +ERTS_ATOMIC64_OPS_IMPL__(relb) +ERTS_ATOMIC64_OPS_IMPL__(ddrb) +ERTS_ATOMIC64_OPS_IMPL__(rb) +ERTS_ATOMIC64_OPS_IMPL__(wb) + +#undef ERTS_ATOMIC64_OPS_IMPL__ +#undef ERTS_ATOMIC64_DW_CMPXCHG_IMPL__ + +ERTS_GLB_INLINE void +erts_atomic64_set_dirty(erts_atomic64_t *var, erts_aint64_t val) +{ + ethr_sint_t *sint = ethr_dw_atomic_addr(var); + ethr_dw_sint_t dw; + ERTS_AINT64_TO_DW_SINT__(dw, val); + sint[0] = dw.sint[0]; + sint[1] = dw.sint[1]; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_atomic64_read_dirty(erts_atomic64_t *var) +{ + ethr_sint_t *sint; + ethr_dw_sint_t dw; + sint = ethr_dw_atomic_addr(var); + dw.sint[0] = sint[0]; + dw.sint[1] = sint[1]; + return ERTS_DW_SINT_TO_AINT64__(dw); +} + +#undef ERTS_DW_SINT_TO_AINT64__ +#undef ERTS_AINT64_TO_DW_SINT__ + +#endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */ + +#endif /* ARCH_32 */ + #else /* !USE_THREADS */ /* Double word size atomics */ @@ -1462,6 +1933,116 @@ erts_atomic32_read_dirty(erts_atomic32_t *var) #define erts_atomic32_set_dirty erts_no_atomic32_set #define erts_atomic32_read_dirty erts_no_atomic32_read +/* 64-bit atomics */ + +#define erts_atomic64_init_nob erts_no_atomic64_set +#define erts_atomic64_set_nob erts_no_atomic64_set +#define erts_atomic64_read_nob erts_no_atomic64_read +#define erts_atomic64_inc_read_nob erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_nob erts_no_atomic64_dec_read +#define erts_atomic64_inc_nob erts_no_atomic64_inc +#define erts_atomic64_dec_nob erts_no_atomic64_dec +#define erts_atomic64_add_read_nob erts_no_atomic64_add_read +#define erts_atomic64_add_nob erts_no_atomic64_add +#define erts_atomic64_read_bor_nob erts_no_atomic64_read_bor +#define erts_atomic64_read_band_nob erts_no_atomic64_read_band +#define erts_atomic64_xchg_nob erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_nob erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_nob erts_no_atomic64_read_bset + +#define erts_atomic64_init_mb erts_no_atomic64_set +#define erts_atomic64_set_mb erts_no_atomic64_set +#define erts_atomic64_read_mb erts_no_atomic64_read +#define erts_atomic64_inc_read_mb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_mb erts_no_atomic64_dec_read +#define erts_atomic64_inc_mb erts_no_atomic64_inc +#define erts_atomic64_dec_mb erts_no_atomic64_dec +#define erts_atomic64_add_read_mb erts_no_atomic64_add_read +#define erts_atomic64_add_mb erts_no_atomic64_add +#define erts_atomic64_read_bor_mb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_mb erts_no_atomic64_read_band +#define erts_atomic64_xchg_mb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_mb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_mb erts_no_atomic64_read_bset + +#define erts_atomic64_init_acqb erts_no_atomic64_set +#define erts_atomic64_set_acqb erts_no_atomic64_set +#define erts_atomic64_read_acqb erts_no_atomic64_read +#define erts_atomic64_inc_read_acqb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_acqb erts_no_atomic64_dec_read +#define erts_atomic64_inc_acqb erts_no_atomic64_inc +#define erts_atomic64_dec_acqb erts_no_atomic64_dec +#define erts_atomic64_add_read_acqb erts_no_atomic64_add_read +#define erts_atomic64_add_acqb erts_no_atomic64_add +#define erts_atomic64_read_bor_acqb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_acqb erts_no_atomic64_read_band +#define erts_atomic64_xchg_acqb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_acqb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_acqb erts_no_atomic64_read_bset + +#define erts_atomic64_init_relb erts_no_atomic64_set +#define erts_atomic64_set_relb erts_no_atomic64_set +#define erts_atomic64_read_relb erts_no_atomic64_read +#define erts_atomic64_inc_read_relb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_relb erts_no_atomic64_dec_read +#define erts_atomic64_inc_relb erts_no_atomic64_inc +#define erts_atomic64_dec_relb erts_no_atomic64_dec +#define erts_atomic64_add_read_relb erts_no_atomic64_add_read +#define erts_atomic64_add_relb erts_no_atomic64_add +#define erts_atomic64_read_bor_relb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_relb erts_no_atomic64_read_band +#define erts_atomic64_xchg_relb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_relb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_relb erts_no_atomic64_read_bset + +#define erts_atomic64_init_ddrb erts_no_atomic64_set +#define erts_atomic64_set_ddrb erts_no_atomic64_set +#define erts_atomic64_read_ddrb erts_no_atomic64_read +#define erts_atomic64_inc_read_ddrb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_ddrb erts_no_atomic64_dec_read +#define erts_atomic64_inc_ddrb erts_no_atomic64_inc +#define erts_atomic64_dec_ddrb erts_no_atomic64_dec +#define erts_atomic64_add_read_ddrb erts_no_atomic64_add_read +#define erts_atomic64_add_ddrb erts_no_atomic64_add +#define erts_atomic64_read_bor_ddrb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_ddrb erts_no_atomic64_read_band +#define erts_atomic64_xchg_ddrb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_ddrb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_ddrb erts_no_atomic64_read_bset + +#define erts_atomic64_init_rb erts_no_atomic64_set +#define erts_atomic64_set_rb erts_no_atomic64_set +#define erts_atomic64_read_rb erts_no_atomic64_read +#define erts_atomic64_inc_read_rb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_rb erts_no_atomic64_dec_read +#define erts_atomic64_inc_rb erts_no_atomic64_inc +#define erts_atomic64_dec_rb erts_no_atomic64_dec +#define erts_atomic64_add_read_rb erts_no_atomic64_add_read +#define erts_atomic64_add_rb erts_no_atomic64_add +#define erts_atomic64_read_bor_rb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_rb erts_no_atomic64_read_band +#define erts_atomic64_xchg_rb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_rb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_rb erts_no_atomic64_read_bset + +#define erts_atomic64_init_wb erts_no_atomic64_set +#define erts_atomic64_set_wb erts_no_atomic64_set +#define erts_atomic64_read_wb erts_no_atomic64_read +#define erts_atomic64_inc_read_wb erts_no_atomic64_inc_read +#define erts_atomic64_dec_read_wb erts_no_atomic64_dec_read +#define erts_atomic64_inc_wb erts_no_atomic64_inc +#define erts_atomic64_dec_wb erts_no_atomic64_dec +#define erts_atomic64_add_read_wb erts_no_atomic64_add_read +#define erts_atomic64_add_wb erts_no_atomic64_add +#define erts_atomic64_read_bor_wb erts_no_atomic64_read_bor +#define erts_atomic64_read_band_wb erts_no_atomic64_read_band +#define erts_atomic64_xchg_wb erts_no_atomic64_xchg +#define erts_atomic64_cmpxchg_wb erts_no_atomic64_cmpxchg +#define erts_atomic64_read_bset_wb erts_no_atomic64_read_bset + +#define erts_atomic64_set_dirty erts_no_atomic64_set +#define erts_atomic64_read_dirty erts_no_atomic64_read + #endif /* !USE_THREADS */ #if ERTS_GLB_INLINE_INCL_FUNC_DEF @@ -2383,6 +2964,104 @@ erts_no_atomic32_read_bset(erts_no_atomic32_t *var, return old; } +/* atomic64 */ + +ERTS_GLB_INLINE void +erts_no_atomic64_set(erts_no_atomic64_t *var, erts_aint64_t i) +{ + *var = i; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read(erts_no_atomic64_t *var) +{ + return *var; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_inc_read(erts_no_atomic64_t *incp) +{ + return ++(*incp); +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_dec_read(erts_no_atomic64_t *decp) +{ + return --(*decp); +} + +ERTS_GLB_INLINE void +erts_no_atomic64_inc(erts_no_atomic64_t *incp) +{ + ++(*incp); +} + +ERTS_GLB_INLINE void +erts_no_atomic64_dec(erts_no_atomic64_t *decp) +{ + --(*decp); +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_add_read(erts_no_atomic64_t *addp, erts_aint64_t i) +{ + return *addp += i; +} + +ERTS_GLB_INLINE void +erts_no_atomic64_add(erts_no_atomic64_t *addp, erts_aint64_t i) +{ + *addp += i; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_bor(erts_no_atomic64_t *var, erts_aint64_t mask) +{ + erts_aint64_t old; + old = *var; + *var |= mask; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_band(erts_no_atomic64_t *var, erts_aint64_t mask) +{ + erts_aint64_t old; + old = *var; + *var &= mask; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_xchg(erts_no_atomic64_t *xchgp, erts_aint64_t new) +{ + erts_aint64_t old = *xchgp; + *xchgp = new; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_cmpxchg(erts_no_atomic64_t *xchgp, + erts_aint64_t new, + erts_aint64_t expected) +{ + erts_aint64_t old = *xchgp; + if (old == expected) + *xchgp = new; + return old; +} + +ERTS_GLB_INLINE erts_aint64_t +erts_no_atomic64_read_bset(erts_no_atomic64_t *var, + erts_aint64_t mask, + erts_aint64_t set) +{ + erts_aint64_t old = *var; + *var &= ~mask; + *var |= (mask & set); + return old; +} + /* spinlock */ ERTS_GLB_INLINE void diff --git a/erts/emulator/beam/erl_trace.c b/erts/emulator/beam/erl_trace.c index 6978a5f11a..ea5c850a30 100644 --- a/erts/emulator/beam/erl_trace.c +++ b/erts/emulator/beam/erl_trace.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1999-2013. All Rights Reserved. + * Copyright Ericsson AB 1999-2014. 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 @@ -151,6 +151,11 @@ do { \ message dispatcher thread takes care of that). */ #define ERTS_GET_TRACER_REF(RES, TPID, TRACEE_FLGS) \ do { (RES) = (TPID); } while(0) +int +erts_is_tracer_proc_valid(Process* p) +{ + return 1; +} #else #define ERTS_NULL_TRACER_REF NULL #define ERTS_TRACER_REF_TYPE Process * @@ -163,6 +168,20 @@ do { \ return; \ } \ } while (0) +int +erts_is_tracer_proc_valid(Process* p) +{ + Process* tracer; + + tracer = erts_proc_lookup(ERTS_TRACER_PROC(p)); + if (tracer && ERTS_TRACE_FLAGS(tracer) & F_TRACER) { + return 1; + } else { + ERTS_TRACER_PROC(p) = NIL; + ERTS_TRACE_FLAGS(p) = ~TRACEE_FLAGS; + return 0; + } +} #endif static Uint active_sched; @@ -2505,7 +2524,7 @@ monitor_large_heap(Process *p) { #ifndef ERTS_SMP ASSERT(is_internal_pid(system_monitor)); monitor_p = erts_proc_lookup(system_monitor); - if (monitor_p || p == monitor_p) { + if (!monitor_p || p == monitor_p) { return; } #endif diff --git a/erts/emulator/beam/erl_trace.h b/erts/emulator/beam/erl_trace.h index 853c6cb0d8..4f2c70d6e7 100644 --- a/erts/emulator/beam/erl_trace.h +++ b/erts/emulator/beam/erl_trace.h @@ -39,6 +39,7 @@ void erts_change_default_tracing(int setflags, Uint *flagsp, Eterm *tracerp); void erts_get_default_tracing(Uint *flagsp, Eterm *tracerp); void erts_set_system_monitor(Eterm monitor); Eterm erts_get_system_monitor(void); +int erts_is_tracer_proc_valid(Process* p); #ifdef ERTS_SMP void erts_check_my_tracer_proc(Process *); diff --git a/erts/emulator/beam/erl_unicode.c b/erts/emulator/beam/erl_unicode.c index 3a968594f3..f8e1431a53 100644 --- a/erts/emulator/beam/erl_unicode.c +++ b/erts/emulator/beam/erl_unicode.c @@ -2126,6 +2126,8 @@ Eterm erts_convert_native_to_filename(Process *p, byte *bytes) mac = 1; case ERL_FILENAME_UTF8: size = strlen((char *) bytes); + if (size == 0) + return NIL; if (erts_analyze_utf8(bytes,size,&err_pos,&num_chars,NULL) != ERTS_UTF8_OK) { goto noconvert; } diff --git a/erts/emulator/beam/erl_utils.h b/erts/emulator/beam/erl_utils.h index 0807649ea1..c32f8fd61c 100644 --- a/erts/emulator/beam/erl_utils.h +++ b/erts/emulator/beam/erl_utils.h @@ -32,11 +32,7 @@ typedef struct { #endif union { Uint64 not_atomic; -#ifdef ARCH_64 - erts_atomic_t atomic; -#else - erts_dw_atomic_t atomic; -#endif + erts_atomic64_t atomic; } counter; } erts_interval_t; @@ -50,9 +46,6 @@ Uint64 erts_ensure_later_interval_nob(erts_interval_t *, Uint64); Uint64 erts_ensure_later_interval_acqb(erts_interval_t *, Uint64); Uint64 erts_smp_ensure_later_interval_nob(erts_interval_t *, Uint64); Uint64 erts_smp_ensure_later_interval_acqb(erts_interval_t *, Uint64); -#ifdef ARCH_32 -ERTS_GLB_INLINE Uint64 erts_interval_dw_aint_to_val__(erts_dw_aint_t *); -#endif ERTS_GLB_INLINE Uint64 erts_current_interval_nob__(erts_interval_t *); ERTS_GLB_INLINE Uint64 erts_current_interval_acqb__(erts_interval_t *); ERTS_GLB_INLINE Uint64 erts_current_interval_nob(erts_interval_t *); @@ -62,46 +55,16 @@ ERTS_GLB_INLINE Uint64 erts_smp_current_interval_acqb(erts_interval_t *); #if ERTS_GLB_INLINE_INCL_FUNC_DEF -#ifdef ARCH_32 - -ERTS_GLB_INLINE Uint64 -erts_interval_dw_aint_to_val__(erts_dw_aint_t *dw) -{ -#ifdef ETHR_SU_DW_NAINT_T__ - return (Uint64) dw->dw_sint; -#else - Uint64 res; - res = (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_HIGH_WORD]); - res <<= 32; - res |= (Uint64) ((Uint32) dw->sint[ERTS_DW_AINT_LOW_WORD]); - return res; -#endif -} - -#endif - ERTS_GLB_INLINE Uint64 erts_current_interval_nob__(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t dw; - erts_dw_atomic_read_nob(&icp->counter.atomic, &dw); - return erts_interval_dw_aint_to_val__(&dw); -#endif + return (Uint64) erts_atomic64_read_nob(&icp->counter.atomic); } ERTS_GLB_INLINE Uint64 erts_current_interval_acqb__(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_read_acqb(&icp->counter.atomic); -#else - erts_dw_aint_t dw; - erts_dw_atomic_read_acqb(&icp->counter.atomic, &dw); - return erts_interval_dw_aint_to_val__(&dw); -#endif + return (Uint64) erts_atomic64_read_acqb(&icp->counter.atomic); } ERTS_GLB_INLINE Uint64 diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c index 9671cde228..9b9b4b2a62 100644 --- a/erts/emulator/beam/external.c +++ b/erts/emulator/beam/external.c @@ -44,9 +44,6 @@ #include "erl_zlib.h" #include "erl_map.h" -#ifdef HIPE -#include "hipe_mode_switch.h" -#endif #define in_area(ptr,start,nbytes) ((UWord)((char*)(ptr) - (char*)(start)) < (nbytes)) #define MAX_STRING_LEN 0xffff @@ -111,26 +108,17 @@ static int encode_size_struct_int(struct TTBSizeContext_*, ErtsAtomCacheMap *acm static Export binary_to_term_trap_export; static BIF_RETTYPE binary_to_term_trap_1(BIF_ALIST_1); -static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* context_b); +static BIF_RETTYPE binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* context_b, + Export *bif, Eterm arg0, Eterm arg1); void erts_init_external(void) { -#if 1 /* In R16 */ erts_init_trap_export(&term_to_binary_trap_export, - am_erlang, am_term_to_binary_trap, 1, + am_erts_internal, am_term_to_binary_trap, 1, &term_to_binary_trap_1); erts_init_trap_export(&binary_to_term_trap_export, - am_erlang, am_binary_to_term_trap, 1, + am_erts_internal, am_binary_to_term_trap, 1, &binary_to_term_trap_1); -#else - sys_memset((void *) &term_to_binary_trap_export, 0, sizeof(Export)); - term_to_binary_trap_export.address = &term_to_binary_trap_export.code[3]; - term_to_binary_trap_export.code[0] = am_erlang; - term_to_binary_trap_export.code[1] = am_term_to_binary_trap; - term_to_binary_trap_export.code[2] = 1; - term_to_binary_trap_export.code[3] = (BeamInstr) em_apply_bif; - term_to_binary_trap_export.code[4] = (BeamInstr) &term_to_binary_trap_1; -#endif return; } @@ -1069,6 +1057,8 @@ static BIF_RETTYPE term_to_binary_trap_1(BIF_ALIST_1) } } +HIPE_WRAPPER_BIF_DISABLE_GC(term_to_binary, 1) + BIF_RETTYPE term_to_binary_1(BIF_ALIST_1) { Eterm res = erts_term_to_binary_int(BIF_P, BIF_ARG_1, 0, TERM_TO_BINARY_DFLAGS, NULL); @@ -1081,6 +1071,8 @@ BIF_RETTYPE term_to_binary_1(BIF_ALIST_1) } } +HIPE_WRAPPER_BIF_DISABLE_GC(term_to_binary, 2) + BIF_RETTYPE term_to_binary_2(BIF_ALIST_2) { Process* p = BIF_P; @@ -1185,6 +1177,8 @@ typedef struct B2TContext_t { Uint32 flags; SWord reds; Eterm trap_bin; + Export *bif; + Eterm arg[2]; enum B2TState state; union { B2TSizeContext sc; @@ -1356,7 +1350,8 @@ static BIF_RETTYPE binary_to_term_trap_1(BIF_ALIST_1) Binary *context_bin = ((ProcBin *) binary_val(BIF_ARG_1))->val; ASSERT(ERTS_MAGIC_BIN_DESTRUCTOR(context_bin) == b2t_context_destructor); - return binary_to_term_int(BIF_P, 0, THE_NON_VALUE, context_bin); + return binary_to_term_int(BIF_P, 0, THE_NON_VALUE, context_bin, NULL, + THE_NON_VALUE, THE_NON_VALUE); } @@ -1391,8 +1386,10 @@ static B2TContext* b2t_export_context(Process* p, B2TContext* src) return ctx; } -static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* context_b) +static BIF_RETTYPE binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* context_b, + Export *bif_init, Eterm arg0, Eterm arg1) { + BIF_RETTYPE ret_val; #ifdef EXTREME_B2T_TRAPPING SWord initial_reds = 1 + b2t_rand() % 4; #else @@ -1409,6 +1406,9 @@ static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* con ctx->state = B2TPrepare; ctx->aligned_alloc = NULL; ctx->flags = flags; + ctx->bif = bif_init; + ctx->arg[0] = arg0; + ctx->arg[1] = arg1; IF_DEBUG(ctx->trap_bin = THE_NON_VALUE;) } else { is_first_call = 0; @@ -1504,12 +1504,24 @@ static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* con HRelease(p, ctx->u.dc.hp_end, ctx->u.dc.hp_start); /*fall through*/ case B2TBadArg: - b2t_destroy_context(ctx); - if (!is_first_call) { - erts_set_gc_state(p, 1); - } BUMP_REDS(p, (initial_reds - ctx->reds) / B2T_BYTES_PER_REDUCTION); - BIF_ERROR(p, BADARG & ~EXF_SAVETRACE); + + ASSERT(ctx->bif == bif_export[BIF_binary_to_term_1] + || ctx->bif == bif_export[BIF_binary_to_term_2]); + + if (is_first_call) + ERTS_BIF_PREP_ERROR(ret_val, p, BADARG); + else { + erts_set_gc_state(p, 1); + if (is_non_value(ctx->arg[1])) + ERTS_BIF_PREP_ERROR_TRAPPED1(ret_val, p, BADARG, ctx->bif, + ctx->arg[0]); + else + ERTS_BIF_PREP_ERROR_TRAPPED2(ret_val, p, BADARG, ctx->bif, + ctx->arg[0], ctx->arg[1]); + } + b2t_destroy_context(ctx); + return ret_val; case B2TDone: b2t_destroy_context(ctx); @@ -1524,7 +1536,8 @@ static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* con erts_set_gc_state(p, 1); } BUMP_REDS(p, (initial_reds - ctx->reds) / B2T_BYTES_PER_REDUCTION); - return ctx->u.dc.res; + ERTS_BIF_PREP_RET(ret_val, ctx->u.dc.res); + return ret_val; default: ASSERT(!"Unknown state in binary_to_term"); @@ -1541,15 +1554,24 @@ static Eterm binary_to_term_int(Process* p, Uint32 flags, Eterm bin, Binary* con erts_set_gc_state(p, 0); } BUMP_ALL_REDS(p); - BIF_TRAP1(&binary_to_term_trap_export, p, ctx->trap_bin); + + ERTS_BIF_PREP_TRAP1(ret_val, &binary_to_term_trap_export, + p, ctx->trap_bin); + + return ret_val; } -BIF_RETTYPE erts_internal_binary_to_term_1(BIF_ALIST_1) +HIPE_WRAPPER_BIF_DISABLE_GC(binary_to_term, 1) + +BIF_RETTYPE binary_to_term_1(BIF_ALIST_1) { - return binary_to_term_int(BIF_P, 0, BIF_ARG_1, NULL); + return binary_to_term_int(BIF_P, 0, BIF_ARG_1, NULL, bif_export[BIF_binary_to_term_1], + BIF_ARG_1, THE_NON_VALUE); } -BIF_RETTYPE erts_internal_binary_to_term_2(BIF_ALIST_2) +HIPE_WRAPPER_BIF_DISABLE_GC(binary_to_term, 2) + +BIF_RETTYPE binary_to_term_2(BIF_ALIST_2) { Eterm opts; Eterm opt; @@ -1570,7 +1592,8 @@ BIF_RETTYPE erts_internal_binary_to_term_2(BIF_ALIST_2) if (is_not_nil(opts)) goto error; - return binary_to_term_int(BIF_P, flags, BIF_ARG_1, NULL); + return binary_to_term_int(BIF_P, flags, BIF_ARG_1, NULL, bif_export[BIF_binary_to_term_2], + BIF_ARG_1, BIF_ARG_2); error: BIF_ERROR(BIF_P, BADARG); @@ -1902,6 +1925,7 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla } real_size = endp - bytes; result_bin = erts_bin_realloc(context->s.ec.result_bin,real_size); + result_bin->orig_size = real_size; level = context->s.ec.level; BUMP_REDS(p, (initial_reds - reds) / TERM_TO_BINARY_LOOP_FACTOR); if (level == 0 || real_size < 6) { /* We are done */ @@ -1981,6 +2005,7 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla erl_zlib_deflate_finish(&(context->s.cc.stream)); result_bin = erts_bin_realloc(context->s.cc.destination_bin, context->s.cc.dest_len+6); + result_bin->orig_size = context->s.cc.dest_len+6; context->s.cc.destination_bin = NULL; pb = (ProcBin *) HAlloc(p, PROC_BIN_SIZE); pb->thing_word = HEADER_PROC_BIN; @@ -2562,29 +2587,25 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep, { map_t *mp = (map_t*)map_val(obj); Uint size = map_get_size(mp); - Eterm *mptr; *ep++ = MAP_EXT; put_int32(size, ep); ep += 4; - /* Push values first */ if (size > 0) { - mptr = map_get_values(mp); + Eterm *kptr = map_get_keys(mp); + Eterm *vptr = map_get_values(mp); + for (i = size-1; i >= 1; i--) { WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) mptr[i]); + WSTACK_PUSH(s, (UWord) vptr[i]); + WSTACK_PUSH(s, ENC_TERM); + WSTACK_PUSH(s, (UWord) kptr[i]); } WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) mptr[0]); + WSTACK_PUSH(s, (UWord) vptr[0]); - mptr = map_get_keys(mp); - for (i = size-1; i >= 1; i--) { - WSTACK_PUSH(s, ENC_TERM); - WSTACK_PUSH(s, (UWord) mptr[i]); - } - - obj = mptr[0]; + obj = kptr[0]; goto L_jump_start; } } @@ -3518,16 +3539,16 @@ dec_term_atom_common: keys = make_tuple(hp); *hp++ = make_arityval(size); - kptr = hp; hp += size; + kptr = hp - 1; mp = (map_t*)hp; hp += MAP_HEADER_SIZE; - vptr = hp; hp += size; + vptr = hp - 1; - /* kptr, first word for keys - * vptr, first word for values + /* kptr, last word for keys + * vptr, last word for values */ /* @@ -3542,27 +3563,12 @@ dec_term_atom_common: mp->keys = keys; *objp = make_map(mp); - /* We assume the map is wellformed, meaning: - * - ascending key order - * - unique keys - */ - - objp = vptr + size - 1; - n = size; - - while (n-- > 0) { - *objp = (Eterm) COMPRESS_POINTER(next); - next = objp; - objp--; - } - - objp = kptr + size - 1; - n = size; - - while (n-- > 0) { - *objp = (Eterm) COMPRESS_POINTER(next); - next = objp; - objp--; + for (n = size; n; n--) { + *vptr = (Eterm) COMPRESS_POINTER(next); + *kptr = (Eterm) COMPRESS_POINTER(vptr); + next = kptr; + vptr--; + kptr--; } } break; @@ -4459,66 +4465,3 @@ error: #undef SKIP2 #undef CHKSIZE } - - -#ifdef HIPE -BIF_RETTYPE hipe_wrapper_term_to_binary_1(BIF_ALIST_1); -BIF_RETTYPE hipe_wrapper_term_to_binary_2(BIF_ALIST_2); -BIF_RETTYPE hipe_wrapper_erts_internal_binary_to_term_1(BIF_ALIST_1); -BIF_RETTYPE hipe_wrapper_erts_internal_binary_to_term_2(BIF_ALIST_2); - -/* Hipe wrappers used by native code for BIFs that disable GC while trapping. - * - * Problem: - * When native code calls a BIF that traps, hipe_mode_switch will push a - * "trap frame" on the Erlang stack in order to find its way back from beam_emu - * back to native caller when finally done. If GC is disabled and stack/heap - * is full there is no place to push the "trap frame". - * - * Solution: - * We reserve space on stack for the "trap frame" here before the BIF is called. - * If the BIF does not trap, the space is reclaimed here before returning. - * If the BIF traps, hipe_push_beam_trap_frame() will detect that a "trap frame" - * already is reserved and use it. - */ -BIF_RETTYPE hipe_wrapper_term_to_binary_1(BIF_ALIST_1) -{ - Eterm res; - hipe_reserve_beam_trap_frame(BIF_P, BIF__ARGS, 1); - res = term_to_binary_1(BIF_P, BIF__ARGS); - if (is_value(res) || BIF_P->freason != TRAP) { - hipe_unreserve_beam_trap_frame(BIF_P); - } - return res; -} -BIF_RETTYPE hipe_wrapper_term_to_binary_2(BIF_ALIST_2) -{ - Eterm res; - hipe_reserve_beam_trap_frame(BIF_P, BIF__ARGS, 2); - res = term_to_binary_2(BIF_P, BIF__ARGS); - if (is_value(res) || BIF_P->freason != TRAP) { - hipe_unreserve_beam_trap_frame(BIF_P); - } - return res; -} -BIF_RETTYPE hipe_wrapper_erts_internal_binary_to_term_1(BIF_ALIST_1) -{ - Eterm res; - hipe_reserve_beam_trap_frame(BIF_P, BIF__ARGS, 1); - res = erts_internal_binary_to_term_1(BIF_P, BIF__ARGS); - if (is_value(res) || BIF_P->freason != TRAP) { - hipe_unreserve_beam_trap_frame(BIF_P); - } - return res; -} -BIF_RETTYPE hipe_wrapper_erts_internal_binary_to_term_2(BIF_ALIST_2) -{ - Eterm res; - hipe_reserve_beam_trap_frame(BIF_P, BIF__ARGS, 2); - res = erts_internal_binary_to_term_2(BIF_P, BIF__ARGS); - if (is_value(res) || BIF_P->freason != TRAP) { - hipe_unreserve_beam_trap_frame(BIF_P); - } - return res; -} -#endif /*HIPE*/ diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h index 8fcb95d0e2..32a2dc43e8 100755..100644 --- a/erts/emulator/beam/global.h +++ b/erts/emulator/beam/global.h @@ -160,6 +160,7 @@ struct erts_driver_t_ { void (*ready_async)(ErlDrvData drv_data, ErlDrvThreadData thread_data); /* Might be NULL */ void (*process_exit)(ErlDrvData drv_data, ErlDrvMonitor *monitor); void (*stop_select)(ErlDrvEvent event, void*); /* Might be NULL */ + void (*emergency_close)(ErlDrvData drv_data); /* Might be NULL */ }; extern erts_driver_t *driver_list; @@ -435,6 +436,8 @@ do {\ }\ } while(0) +#define CLEAR_SAVED_ESTACK(estack) ((void) ((estack)->start = NULL)) + /* * Use on empty stack, only the allocator can be changed before this. * The src stack is reset to NULL. @@ -551,6 +554,8 @@ do {\ }\ } while(0) +#define CLEAR_SAVED_WSTACK(wstack) ((void) ((wstack)->wstart = NULL)) + /* * Use on empty stack, only the allocator can be changed before this. * The src stack is reset to NULL. @@ -848,6 +853,7 @@ Uint erts_port_ioq_size(Port *pp); void erts_stale_drv_select(Eterm, ErlDrvPort, ErlDrvEvent, int, int); Port *erts_get_heart_port(void); +void erts_emergency_close_ports(void); #if defined(ERTS_SMP) && defined(ERTS_ENABLE_LOCK_COUNT) void erts_lcnt_enable_io_lock_count(int enable); @@ -951,20 +957,67 @@ struct Sint_buf { }; char* Sint_to_buf(Sint, struct Sint_buf*); +#define ERTS_IOLIST_STATE_INITER(C_P, OBJ) \ + {(C_P), 0, 0, (OBJ), {NULL, NULL, NULL, ERTS_ALC_T_INVALID}, 0, 0} + +#define ERTS_IOLIST_STATE_MOVE(TO, FROM) \ + sys_memcpy((void *) (TO), (void *) (FROM), sizeof(ErtsIOListState)) + +#define ERTS_IOLIST_SIZE_YIELDS_COUNT_PER_RED 8 + +typedef struct { + Process *c_p; + ErlDrvSizeT size; + Uint offs; + Eterm obj; + ErtsEStack estack; + int reds_left; + int have_size; +} ErtsIOListState; + +#define ERTS_IOLIST2BUF_STATE_INITER(C_P, OBJ) \ + {ERTS_IOLIST_STATE_INITER((C_P), (OBJ)), {NULL, 0, 0, 0}, NULL, 0, NULL, 0} + +#define ERTS_IOLIST2BUF_STATE_MOVE(TO, FROM) \ + sys_memcpy((void *) (TO), (void *) (FROM), sizeof(ErtsIOList2BufState)) + +#define ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT 32 +#define ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED 8 +#define ERTS_IOLIST_TO_BUF_BYTES_PER_RED \ + (ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED*ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT) + +typedef struct { + ErtsIOListState iolist; + struct { + byte *bptr; + size_t size; + Uint bitoffs; + Uint bitsize; + } bcopy; + char *buf; + ErlDrvSizeT len; + Eterm *objp; + int offset; +} ErtsIOList2BufState; + #define ERTS_IOLIST_OK 0 #define ERTS_IOLIST_OVERFLOW 1 #define ERTS_IOLIST_TYPE 2 +#define ERTS_IOLIST_YIELD 3 Eterm buf_to_intlist(Eterm**, const char*, size_t, Eterm); /* most callers pass plain char*'s */ #define ERTS_IOLIST_TO_BUF_OVERFLOW (~((ErlDrvSizeT) 0)) #define ERTS_IOLIST_TO_BUF_TYPE_ERROR (~((ErlDrvSizeT) 1)) +#define ERTS_IOLIST_TO_BUF_YIELD (~((ErlDrvSizeT) 2)) #define ERTS_IOLIST_TO_BUF_FAILED(R) \ - (((R) & (~((ErlDrvSizeT) 1))) == (~((ErlDrvSizeT) 1))) + (((R) & (~((ErlDrvSizeT) 3))) == (~((ErlDrvSizeT) 3))) #define ERTS_IOLIST_TO_BUF_SUCCEEDED(R) \ (!ERTS_IOLIST_TO_BUF_FAILED((R))) ErlDrvSizeT erts_iolist_to_buf(Eterm, char*, ErlDrvSizeT); +ErlDrvSizeT erts_iolist_to_buf_yielding(ErtsIOList2BufState *); +int erts_iolist_size_yielding(ErtsIOListState *state); int erts_iolist_size(Eterm, ErlDrvSizeT *); int is_string(Eterm); void erl_at_exit(void (*) (void*), void*); diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c index cd5060ebb3..012a7d1a4b 100644 --- a/erts/emulator/beam/io.c +++ b/erts/emulator/beam/io.c @@ -914,8 +914,8 @@ int erts_port_handle_xports(Port *prt) (iov)->iov_base = (ptr); \ (iov)->iov_len = (len); \ if (sizeof((iov)->iov_len) < sizeof(len) \ - /* Check if (len) overflowed (iov)->iov_len */ \ - && ((len) >> (sizeof((iov)->iov_len)*CHAR_BIT)) != 0) { \ + /* Check if (len) overflowed (iov)->iov_len */ \ + && (iov)->iov_len != (len)) { \ goto L_overflow; \ } \ *(bv)++ = (bin); \ @@ -1218,9 +1218,10 @@ typedef struct { static ERTS_INLINE ErtsTryImmDrvCallResult try_imm_drv_call(ErtsTryImmDrvCallState *sp) { + unsigned int prof_runnable_ports; ErtsTryImmDrvCallResult res; int reds_left_in; - erts_aint32_t invalid_state, invalid_sched_flags; + erts_aint32_t act, exp, invalid_state, invalid_sched_flags; Port *prt = sp->port; Process *c_p = sp->c_p; @@ -1247,18 +1248,39 @@ try_imm_drv_call(ErtsTryImmDrvCallState *sp) goto locked_fail; } - sp->sched_flags = erts_smp_atomic32_read_nob(&prt->sched.flags); - if (sp->sched_flags & invalid_sched_flags) { - res = ERTS_TRY_IMM_DRV_CALL_INVALID_SCHED_FLAGS; - goto locked_fail; - } + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&prt->sched); + act = erts_smp_atomic32_read_nob(&prt->sched.flags); + + do { + erts_aint32_t new; + + if (act & invalid_sched_flags) { + res = ERTS_TRY_IMM_DRV_CALL_INVALID_SCHED_FLAGS; + sp->sched_flags = act; + goto locked_fail; + } + exp = act; + new = act | ERTS_PTS_FLG_EXEC_IMM; + act = erts_smp_atomic32_cmpxchg_mb(&prt->sched.flags, new, exp); + } while (act != exp); + + sp->sched_flags = act; if (!c_p) reds_left_in = CONTEXT_REDS/10; else { if (IS_TRACED_FL(c_p, F_TRACE_SCHED_PROCS)) trace_virtual_sched(c_p, am_out); + /* + * No status lock held while sending runnable + * proc trace messages. It is however not needed + * in this case, since only this thread can send + * such messages for this process until the process + * has been scheduled out. + */ if (erts_system_profile_flags.runnable_procs && erts_system_profile_flags.exclusive) profile_runnable_proc(c_p, am_inactive); @@ -1273,11 +1295,14 @@ try_imm_drv_call(ErtsTryImmDrvCallState *sp) ERTS_SMP_CHK_NO_PROC_LOCKS; - if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) - trace_sched_ports_where(prt, am_in, sp->port_op); - if (erts_system_profile_flags.runnable_ports - && !erts_port_is_scheduled(prt)) - profile_runnable_port(prt, am_active); + if (prof_runnable_ports | IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) { + if (prof_runnable_ports && !(act & (ERTS_PTS_FLG_IN_RUNQ|ERTS_PTS_FLG_EXEC))) + profile_runnable_port(prt, am_active); + if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) + trace_sched_ports_where(prt, am_in, sp->port_op); + if (prof_runnable_ports) + erts_port_task_sched_unlock(&prt->sched); + } sp->fpe_was_unmasked = erts_block_fpe(); @@ -1294,17 +1319,31 @@ finalize_imm_drv_call(ErtsTryImmDrvCallState *sp) int reds; Port *prt = sp->port; Process *c_p = sp->c_p; + erts_aint32_t act; + unsigned int prof_runnable_ports; reds = prt->reds; reds += erts_port_driver_callback_epilogue(prt, NULL); erts_unblock_fpe(sp->fpe_was_unmasked); - if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) - trace_sched_ports_where(prt, am_out, sp->port_op); - if (erts_system_profile_flags.runnable_ports - && !erts_port_is_scheduled(prt)) - profile_runnable_port(prt, am_inactive); + prof_runnable_ports = erts_system_profile_flags.runnable_ports; + if (prof_runnable_ports) + erts_port_task_sched_lock(&prt->sched); + + act = erts_smp_atomic32_read_band_mb(&prt->sched.flags, + ~ERTS_PTS_FLG_EXEC_IMM); + ERTS_SMP_LC_ASSERT(act & ERTS_PTS_FLG_EXEC_IMM); + + if (prof_runnable_ports | IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) { + if (IS_TRACED_FL(prt, F_TRACE_SCHED_PORTS)) + trace_sched_ports_where(prt, am_out, sp->port_op); + if (prof_runnable_ports) { + if (!(act & (ERTS_PTS_FLG_IN_RUNQ|ERTS_PTS_FLG_EXEC))) + profile_runnable_port(prt, am_inactive); + erts_port_task_sched_unlock(&prt->sched); + } + } erts_port_release(prt); @@ -1319,6 +1358,13 @@ finalize_imm_drv_call(ErtsTryImmDrvCallState *sp) if (IS_TRACED_FL(c_p, F_TRACE_SCHED_PROCS)) trace_virtual_sched(c_p, am_in); + /* + * No status lock held while sending runnable + * proc trace messages. It is however not needed + * in this case, since only this thread can send + * such messages for this process until the process + * has been scheduled out. + */ if (erts_system_profile_flags.runnable_procs && erts_system_profile_flags.exclusive) profile_runnable_proc(c_p, am_active); @@ -2473,7 +2519,7 @@ set_port_connected(int bang_op, DTRACE_CHARBUF(newprocess_str, DTRACE_TERM_BUF_SIZE); dtrace_pid_str(connect, process_str); - erts_snprintf(port_str, sizeof(port_str), "%T", prt->common.id); + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", prt->common.id); dtrace_proc_str(rp, newprocess_str); DTRACE4(port_connect, process_str, port_str, prt->name, newprocess_str); } @@ -3591,9 +3637,9 @@ erts_deliver_port_exit(Port *p, Eterm from, Eterm reason, int send_closed) DTRACE_CHARBUF(port_str, DTRACE_TERM_BUF_SIZE); DTRACE_CHARBUF(rreason_str, 64); - erts_snprintf(from_str, sizeof(from_str), "%T", from); + erts_snprintf(from_str, sizeof(DTRACE_CHARBUF_NAME(from_str)), "%T", from); dtrace_port_str(p, port_str); - erts_snprintf(rreason_str, sizeof(rreason_str), "%T", rreason); + erts_snprintf(rreason_str, sizeof(DTRACE_CHARBUF_NAME(rreason_str)), "%T", rreason); DTRACE4(port_exit, from_str, port_str, p->name, rreason_str); } #endif @@ -4660,7 +4706,7 @@ set_busy_port(ErlDrvPort dprt, int on) #ifdef USE_VM_PROBES if (DTRACE_ENABLED(port_busy)) { - erts_snprintf(port_str, sizeof(port_str), + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", prt->common.id); DTRACE1(port_busy, port_str); } @@ -4673,7 +4719,7 @@ set_busy_port(ErlDrvPort dprt, int on) #ifdef USE_VM_PROBES if (DTRACE_ENABLED(port_not_busy)) { - erts_snprintf(port_str, sizeof(port_str), + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", prt->common.id); DTRACE1(port_not_busy, port_str); } @@ -4725,9 +4771,9 @@ erts_port_resume_procs(Port *prt) DTRACE_CHARBUF(pid_str, 16); ErtsProcList* plp2 = plp; - erts_snprintf(port_str, sizeof(port_str), "%T", prt->common.id); + erts_snprintf(port_str, sizeof(DTRACE_CHARBUF_NAME(port_str)), "%T", prt->common.id); while (plp2 != NULL) { - erts_snprintf(pid_str, sizeof(pid_str), "%T", plp2->pid); + erts_snprintf(pid_str, sizeof(DTRACE_CHARBUF_NAME(pid_str)), "%T", plp2->pid); DTRACE2(process_port_unblocked, pid_str, port_str); } } @@ -6129,7 +6175,7 @@ driver_pdl_create(ErlDrvPort dp) return NULL; pdl = erts_alloc(ERTS_ALC_T_PORT_DATA_LOCK, sizeof(struct erl_drv_port_data_lock)); - erts_mtx_init(&pdl->mtx, "port_data_lock"); + erts_mtx_init_x(&pdl->mtx, "port_data_lock", pp->common.id, 1); pdl_init_refc(pdl); erts_port_inc_refc(pp); pdl->prt = pp; @@ -7166,7 +7212,7 @@ char *driver_dl_error(void) #define ERL_DRV_SYS_INFO_SIZE(LAST_FIELD) \ - (((size_t) &((ErlDrvSysInfo *) 0)->LAST_FIELD) \ + (offsetof(ErlDrvSysInfo, LAST_FIELD) \ + sizeof(((ErlDrvSysInfo *) 0)->LAST_FIELD)) void @@ -7228,6 +7274,18 @@ driver_system_info(ErlDrvSysInfo *sip, size_t si_size) sip->nif_major_version = ERL_NIF_MAJOR_VERSION; sip->nif_minor_version = ERL_NIF_MINOR_VERSION; } + /* + * 'dirty_scheduler_support' is the last field in the 4th version + * (driver version 3.1, NIF version 2.7) + */ + if (si_size >= ERL_DRV_SYS_INFO_SIZE(dirty_scheduler_support)) { +#if defined(ERL_NIF_DIRTY_SCHEDULER_SUPPORT) && defined(USE_THREADS) + sip->dirty_scheduler_support = 1; +#else + sip->dirty_scheduler_support = 0; +#endif + } + } @@ -7291,6 +7349,8 @@ no_stop_select_callback(ErlDrvEvent event, void* private) erts_send_error_to_logger_nogl(dsbufp); } +#define IS_DRIVER_VERSION_GE(DE,MAJOR,MINOR) \ + ((DE)->major_version >= (MAJOR) && (DE)->minor_version >= (MINOR)) static int init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) @@ -7338,6 +7398,7 @@ init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) drv->timeout = de->timeout ? de->timeout : no_timeout_callback; drv->ready_async = de->ready_async; drv->process_exit = de->process_exit; + drv->emergency_close = IS_DRIVER_VERSION_GE(de,3,2) ? de->emergency_close : NULL; if (de->stop_select) drv->stop_select = de->stop_select; else @@ -7356,6 +7417,8 @@ init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle) } } +#undef IS_DRIVER_VERSION_GE + void erts_destroy_driver(erts_driver_t *drv) { @@ -7499,7 +7562,7 @@ Port *erts_get_heart_port(void) if (!port) continue; /* only examine undead or alive ports */ - if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_DEAD) + if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP) continue; /* immediate atom compare */ reg = port->common.u.alive.reg; @@ -7510,3 +7573,23 @@ Port *erts_get_heart_port(void) return NULL; } + +void erts_emergency_close_ports(void) +{ + int ix, max = erts_ptab_max(&erts_port); + + for (ix = 0; ix < max; ix++) { + Port *port = erts_pix2port(ix); + + if (!port) + continue; + /* only examine undead or alive ports */ + if (erts_atomic32_read_nob(&port->state) & ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP) + continue; + + /* emergency close socket */ + if (port->drv_ptr->emergency_close) { + port->drv_ptr->emergency_close((ErlDrvData) port->drv_data); + } + } +} diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab index 73630fda8e..68fcc177ae 100644 --- a/erts/emulator/beam/ops.tab +++ b/erts/emulator/beam/ops.tab @@ -1484,7 +1484,8 @@ new_map j d I I update_map_assoc j s d I I update_map_exact j s d I I -is_map Fail cq => jump Fail +is_map Fail Literal=q => move Literal x | is_map Fail x +is_map Fail c => jump Fail %macro: is_map IsMap -fail_action is_map f r diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h index e273056a2b..c29d4b3777 100644 --- a/erts/emulator/beam/sys.h +++ b/erts/emulator/beam/sys.h @@ -66,8 +66,12 @@ */ #ifndef ERTS_SYS_FD_TYPE +#define ERTS_SYS_FD_INVALID ((ErtsSysFdType) -1) typedef int ErtsSysFdType; #else +#ifndef ERTS_SYS_FD_INVALID +# error missing ERTS_SYS_FD_INVALID +#endif typedef ERTS_SYS_FD_TYPE ErtsSysFdType; #endif @@ -154,10 +158,14 @@ typedef ERTS_SYS_FD_TYPE ErtsSysFdType; /* In VC++, noreturn is a declspec that has to be before the types, * but in GNUC it is an att ribute to be placed between return type * and function name, hence __decl_noreturn <types> __noreturn <function name> + * + * at some platforms (e.g. Android) __noreturn is defined at sys/cdef.h */ #if __GNUC__ # define __decl_noreturn -# define __noreturn __attribute__((noreturn)) +# ifndef __noreturn +# define __noreturn __attribute__((noreturn)) +# endif #else # if defined(__WIN32__) && defined(_MSC_VER) # define __noreturn @@ -270,6 +278,7 @@ __decl_noreturn void __noreturn erl_assert_error(const char* expr, const char *f typedef unsigned int Eterm; typedef unsigned int Uint; typedef int Sint; +#define ERTS_UINT_MAX UINT_MAX #define ERTS_SIZEOF_ETERM SIZEOF_INT #define ErtsStrToSint strtol #else @@ -343,6 +352,7 @@ typedef long long Sint; typedef Uint UWord; typedef Sint SWord; +#define ERTS_UINT_MAX ERTS_UWORD_MAX #endif /* HALFWORD_HEAP */ @@ -495,7 +505,7 @@ extern volatile int erts_writing_erl_crash_dump; # define NO_ERF # define NO_ERFC /* This definition doesn't take NaN into account, but matherr() gets those */ -# define finite(x) (fabs(x) != HUGE_VAL) +# define isfinite(x) (fabs(x) != HUGE_VAL) # define USE_MATHERR # define HAVE_FINITE #endif @@ -738,6 +748,14 @@ void init_getenv_state(GETENV_STATE *); char * getenv_string(GETENV_STATE *); void fini_getenv_state(GETENV_STATE *); +#define HAVE_ERTS_CHECK_IO_DEBUG +typedef struct { + int no_used_fds; + int no_driver_select_structs; + int no_driver_event_structs; +} ErtsCheckIoDebugInfo; +int erts_check_io_debug(ErtsCheckIoDebugInfo *ip); + /* xxxP */ #define SYS_DEFAULT_FLOAT_DECIMALS 20 void init_sys_float(void); diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c index 738f793020..c505c44905 100644 --- a/erts/emulator/beam/utils.c +++ b/erts/emulator/beam/utils.c @@ -48,6 +48,7 @@ #include "erl_sched_spec_pre_alloc.h" #include "beam_bp.h" #include "erl_ptab.h" +#include "erl_check_io.h" #undef M_TRIM_THRESHOLD #undef M_TOP_PAD @@ -3197,106 +3198,303 @@ buf_to_intlist(Eterm** hpp, const char *buf, size_t len, Eterm tail) ** */ -ErlDrvSizeT erts_iolist_to_buf(Eterm obj, char* buf, ErlDrvSizeT alloced_len) +typedef enum { + ERTS_IL2B_BCOPY_OK, + ERTS_IL2B_BCOPY_YIELD, + ERTS_IL2B_BCOPY_OVERFLOW, + ERTS_IL2B_BCOPY_TYPE_ERROR +} ErtsIL2BBCopyRes; + +static ErtsIL2BBCopyRes +iolist_to_buf_bcopy(ErtsIOList2BufState *state, Eterm obj, int *yield_countp); + +static ERTS_INLINE ErlDrvSizeT +iolist_to_buf(const int yield_support, + ErtsIOList2BufState *state, + Eterm obj, + char* buf, + ErlDrvSizeT alloced_len) { - ErlDrvSizeT len = (ErlDrvSizeT) alloced_len; - Eterm* objp; +#undef IOLIST_TO_BUF_BCOPY +#define IOLIST_TO_BUF_BCOPY(CONSP) \ +do { \ + size_t size = binary_size(obj); \ + if (size > 0) { \ + Uint bitsize; \ + byte* bptr; \ + Uint bitoffs; \ + Uint num_bits; \ + if (yield_support) { \ + size_t max_size = ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; \ + if (yield_count > 0) \ + max_size *= yield_count+1; \ + if (size > max_size) { \ + state->objp = CONSP; \ + goto L_bcopy_yield; \ + } \ + if (size >= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT) { \ + int cost = (int) size; \ + cost /= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; \ + yield_count -= cost; \ + } \ + } \ + if (len < size) \ + goto L_overflow; \ + ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); \ + if (bitsize != 0) \ + goto L_type_error; \ + num_bits = 8*size; \ + copy_binary_to_buffer(buf, 0, bptr, bitoffs, num_bits); \ + buf += size; \ + len -= size; \ + } \ +} while (0) + + ErlDrvSizeT res, len; + Eterm* objp = NULL; + int init_yield_count; + int yield_count; DECLARE_ESTACK(s); - goto L_again; - - while (!ESTACK_ISEMPTY(s)) { - obj = ESTACK_POP(s); - L_again: - if (is_list(obj)) { - L_iter_list: - objp = list_val(obj); - obj = CAR(objp); - if (is_byte(obj)) { - if (len == 0) { - goto L_overflow; - } - *buf++ = unsigned_val(obj); - len--; - } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - - if (len < size) { + + len = (ErlDrvSizeT) alloced_len; + + if (!yield_support) { + yield_count = init_yield_count = 0; /* Shut up faulty warning... >:-( */ + goto L_again; + } + else { + + if (state->iolist.reds_left <= 0) + return ERTS_IOLIST_TO_BUF_YIELD; + + ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK); + init_yield_count = (ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED + * state->iolist.reds_left); + yield_count = init_yield_count; + + if (!state->iolist.estack.start) + goto L_again; + else { + int chk_stack; + /* Restart; restore state... */ + ESTACK_RESTORE(s, &state->iolist.estack); + + if (!state->bcopy.bptr) + chk_stack = 0; + else { + chk_stack = 1; + switch (iolist_to_buf_bcopy(state, THE_NON_VALUE, &yield_count)) { + case ERTS_IL2B_BCOPY_OK: + break; + case ERTS_IL2B_BCOPY_YIELD: + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + ESTACK_SAVE(s, &state->iolist.estack); + return ERTS_IOLIST_TO_BUF_YIELD; + case ERTS_IL2B_BCOPY_OVERFLOW: goto L_overflow; - } - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - if (bitsize != 0) { + case ERTS_IL2B_BCOPY_TYPE_ERROR: goto L_type_error; } - num_bits = 8*size; - copy_binary_to_buffer(buf, 0, bptr, bitoffs, num_bits); - buf += size; - len -= size; - } else if (is_list(obj)) { - ESTACK_PUSH(s, CDR(objp)); - goto L_iter_list; /* on head */ - } else if (is_not_nil(obj)) { - goto L_type_error; } - obj = CDR(objp); - if (is_list(obj)) { - goto L_iter_list; /* on tail */ - } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - if (len < size) { - goto L_overflow; + obj = state->iolist.obj; + buf = state->buf; + len = state->len; + objp = state->objp; + state->objp = NULL; + if (objp) + goto L_tail; + if (!chk_stack) + goto L_again; + /* check stack */ + } + } + + while (!ESTACK_ISEMPTY(s)) { + obj = ESTACK_POP(s); + L_again: + if (is_list(obj)) { + while (1) { /* Tail loop */ + while (1) { /* Head loop */ + if (yield_support && --yield_count <= 0) + goto L_yield; + objp = list_val(obj); + obj = CAR(objp); + if (is_byte(obj)) { + if (len == 0) { + goto L_overflow; + } + *buf++ = unsigned_val(obj); + len--; + } else if (is_binary(obj)) { + IOLIST_TO_BUF_BCOPY(objp); + } else if (is_list(obj)) { + ESTACK_PUSH(s, CDR(objp)); + continue; /* Head loop */ + } else if (is_not_nil(obj)) { + goto L_type_error; + } + break; } - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - if (bitsize != 0) { + + L_tail: + + obj = CDR(objp); + + if (is_list(obj)) { + continue; /* Tail loop */ + } else if (is_binary(obj)) { + IOLIST_TO_BUF_BCOPY(NULL); + } else if (is_not_nil(obj)) { goto L_type_error; } - num_bits = 8*size; - copy_binary_to_buffer(buf, 0, bptr, bitoffs, num_bits); - buf += size; - len -= size; - } else if (is_not_nil(obj)) { - goto L_type_error; + break; } } else if (is_binary(obj)) { - byte* bptr; - size_t size = binary_size(obj); - Uint bitsize; - Uint bitoffs; - Uint num_bits; - if (len < size) { - goto L_overflow; - } - ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); - if (bitsize != 0) { - goto L_type_error; - } - num_bits = 8*size; - copy_binary_to_buffer(buf, 0, bptr, bitoffs, num_bits); - buf += size; - len -= size; + IOLIST_TO_BUF_BCOPY(NULL); } else if (is_not_nil(obj)) { goto L_type_error; - } + } else if (yield_support && --yield_count <= 0) + goto L_yield; } + res = len; + + L_return: + DESTROY_ESTACK(s); - return len; + + if (yield_support) { + int reds; + CLEAR_SAVED_ESTACK(&state->iolist.estack); + reds = ((init_yield_count - yield_count - 1) + / ERTS_IOLIST_TO_BUF_YIELD_COUNT_PER_RED) + 1; + BUMP_REDS(state->iolist.c_p, reds); + state->iolist.reds_left -= reds; + if (state->iolist.reds_left < 0) + state->iolist.reds_left = 0; + } + + + return res; L_type_error: - DESTROY_ESTACK(s); - return ERTS_IOLIST_TO_BUF_TYPE_ERROR; + res = ERTS_IOLIST_TO_BUF_TYPE_ERROR; + goto L_return; L_overflow: - DESTROY_ESTACK(s); - return ERTS_IOLIST_TO_BUF_OVERFLOW; + res = ERTS_IOLIST_TO_BUF_OVERFLOW; + goto L_return; + + L_bcopy_yield: + + state->buf = buf; + state->len = len; + + switch (iolist_to_buf_bcopy(state, obj, &yield_count)) { + case ERTS_IL2B_BCOPY_OK: + ERTS_INTERNAL_ERROR("Missing yield"); + case ERTS_IL2B_BCOPY_YIELD: + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + ESTACK_SAVE(s, &state->iolist.estack); + return ERTS_IOLIST_TO_BUF_YIELD; + case ERTS_IL2B_BCOPY_OVERFLOW: + goto L_overflow; + case ERTS_IL2B_BCOPY_TYPE_ERROR: + goto L_type_error; + } + + L_yield: + + BUMP_ALL_REDS(state->iolist.c_p); + state->iolist.reds_left = 0; + state->iolist.obj = obj; + state->buf = buf; + state->len = len; + ESTACK_SAVE(s, &state->iolist.estack); + return ERTS_IOLIST_TO_BUF_YIELD; + +#undef IOLIST_TO_BUF_BCOPY +} + +static ErtsIL2BBCopyRes +iolist_to_buf_bcopy(ErtsIOList2BufState *state, Eterm obj, int *yield_countp) +{ + ErtsIL2BBCopyRes res; + char *buf = state->buf; + ErlDrvSizeT len = state->len; + byte* bptr; + size_t size; + size_t max_size; + Uint bitoffs; + Uint num_bits; + int yield_count = *yield_countp; + + if (state->bcopy.bptr) { + bptr = state->bcopy.bptr; + size = state->bcopy.size; + bitoffs = state->bcopy.bitoffs; + state->bcopy.bptr = NULL; + } + else { + Uint bitsize; + + ASSERT(is_binary(obj)); + + size = binary_size(obj); + if (size <= 0) + return ERTS_IL2B_BCOPY_OK; + + if (len < size) + return ERTS_IL2B_BCOPY_OVERFLOW; + + ERTS_GET_BINARY_BYTES(obj, bptr, bitoffs, bitsize); + if (bitsize != 0) + return ERTS_IL2B_BCOPY_TYPE_ERROR; + } + + ASSERT(size > 0); + max_size = (size_t) ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; + if (yield_count > 0) + max_size *= (size_t) (yield_count+1); + + if (size <= max_size) { + if (size >= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT) { + int cost = (int) size; + cost /= ERTS_IOLIST_TO_BUF_BYTES_PER_YIELD_COUNT; + yield_count -= cost; + } + res = ERTS_IL2B_BCOPY_OK; + } + else { + ASSERT(0 < max_size && max_size < size); + yield_count = 0; + state->bcopy.bptr = bptr + max_size; + state->bcopy.bitoffs = bitoffs; + state->bcopy.size = size - max_size; + size = max_size; + res = ERTS_IL2B_BCOPY_YIELD; + } + + num_bits = 8*size; + copy_binary_to_buffer(buf, 0, bptr, bitoffs, num_bits); + state->buf += size; + state->len -= size; + *yield_countp = yield_count; + + return res; +} + +ErlDrvSizeT erts_iolist_to_buf_yielding(ErtsIOList2BufState *state) +{ + return iolist_to_buf(1, state, state->iolist.obj, state->buf, state->len); +} + +ErlDrvSizeT erts_iolist_to_buf(Eterm obj, char* buf, ErlDrvSizeT alloced_len) +{ + return iolist_to_buf(0, NULL, obj, buf, alloced_len); } /* @@ -3307,11 +3505,32 @@ ErlDrvSizeT erts_iolist_to_buf(Eterm obj, char* buf, ErlDrvSizeT alloced_len) * Any input term error detected in erts_iolist_to_buf should also * be detected in this function! */ -int erts_iolist_size(Eterm obj, ErlDrvSizeT* sizep) + +static ERTS_INLINE int +iolist_size(const int yield_support, ErtsIOListState *state, Eterm obj, ErlDrvSizeT* sizep) { + int res, init_yield_count, yield_count; Eterm* objp; - Uint size = 0; /* Intentionally Uint due to halfword heap */ + Uint size = (Uint) *sizep; /* Intentionally Uint due to halfword heap */ DECLARE_ESTACK(s); + + if (!yield_support) + yield_count = init_yield_count = 0; /* Shut up faulty warning... >:-( */ + else { + if (state->reds_left <= 0) + return ERTS_IOLIST_YIELD; + ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK); + init_yield_count = ERTS_IOLIST_SIZE_YIELDS_COUNT_PER_RED; + init_yield_count *= state->reds_left; + yield_count = init_yield_count; + if (state->estack.start) { + /* Restart; restore state... */ + ESTACK_RESTORE(s, &state->estack); + size = (Uint) state->size; + obj = state->obj; + } + } + goto L_again; #define SAFE_ADD(Var, Val) \ @@ -3327,51 +3546,101 @@ int erts_iolist_size(Eterm obj, ErlDrvSizeT* sizep) obj = ESTACK_POP(s); L_again: if (is_list(obj)) { - L_iter_list: - objp = list_val(obj); - /* Head */ - obj = CAR(objp); - if (is_byte(obj)) { - size++; - if (size == 0) { - goto L_overflow_error; + while (1) { /* Tail loop */ + while (1) { /* Head loop */ + if (yield_support && --yield_count <= 0) + goto L_yield; + objp = list_val(obj); + /* Head */ + obj = CAR(objp); + if (is_byte(obj)) { + size++; + if (size == 0) { + goto L_overflow_error; + } + } else if (is_binary(obj) && binary_bitsize(obj) == 0) { + SAFE_ADD(size, binary_size(obj)); + } else if (is_list(obj)) { + ESTACK_PUSH(s, CDR(objp)); + continue; /* Head loop */ + } else if (is_not_nil(obj)) { + goto L_type_error; + } + break; } - } else if (is_binary(obj) && binary_bitsize(obj) == 0) { - SAFE_ADD(size, binary_size(obj)); - } else if (is_list(obj)) { - ESTACK_PUSH(s, CDR(objp)); - goto L_iter_list; /* on head */ - } else if (is_not_nil(obj)) { - goto L_type_error; + /* Tail */ + obj = CDR(objp); + if (is_list(obj)) + continue; /* Tail loop */ + else if (is_binary(obj) && binary_bitsize(obj) == 0) { + SAFE_ADD(size, binary_size(obj)); + } else if (is_not_nil(obj)) { + goto L_type_error; + } + break; } - /* Tail */ - obj = CDR(objp); - if (is_list(obj)) - goto L_iter_list; /* on tail */ - else if (is_binary(obj) && binary_bitsize(obj) == 0) { + } else { + if (yield_support && --yield_count <= 0) + goto L_yield; + if (is_binary(obj) && binary_bitsize(obj) == 0) { /* Tail was binary */ SAFE_ADD(size, binary_size(obj)); } else if (is_not_nil(obj)) { goto L_type_error; } - } else if (is_binary(obj) && binary_bitsize(obj) == 0) { /* Tail was binary */ - SAFE_ADD(size, binary_size(obj)); - } else if (is_not_nil(obj)) { - goto L_type_error; } } #undef SAFE_ADD - DESTROY_ESTACK(s); *sizep = (ErlDrvSizeT) size; - return ERTS_IOLIST_OK; - L_overflow_error: + res = ERTS_IOLIST_OK; + + L_return: + DESTROY_ESTACK(s); - return ERTS_IOLIST_OVERFLOW; + + if (yield_support) { + int yc, reds; + CLEAR_SAVED_ESTACK(&state->estack); + yc = init_yield_count - yield_count; + reds = ((yc - 1) / ERTS_IOLIST_SIZE_YIELDS_COUNT_PER_RED) + 1; + BUMP_REDS(state->c_p, reds); + state->reds_left -= reds; + state->size = (ErlDrvSizeT) size; + state->have_size = 1; + } + + return res; + + L_overflow_error: + res = ERTS_IOLIST_OVERFLOW; + size = 0; + goto L_return; L_type_error: - DESTROY_ESTACK(s); - return ERTS_IOLIST_TYPE; + res = ERTS_IOLIST_TYPE; + size = 0; + goto L_return; + + L_yield: + BUMP_ALL_REDS(state->c_p); + state->reds_left = 0; + state->size = size; + state->obj = obj; + ESTACK_SAVE(s, &state->estack); + return ERTS_IOLIST_YIELD; +} + +int erts_iolist_size_yielding(ErtsIOListState *state) +{ + ErlDrvSizeT size = state->size; + return iolist_size(1, state, state->obj, &size); +} + +int erts_iolist_size(Eterm obj, ErlDrvSizeT* sizep) +{ + *sizep = 0; + return iolist_size(0, NULL, obj, sizep); } /* return 0 if item is not a non-empty flat list of bytes */ @@ -3680,6 +3949,9 @@ erts_save_emu_args(int argc, char **argv) size += sz+1; } ptr = (char *) malloc(size); + if (!ptr) { + ERTS_INTERNAL_ERROR("malloc failed to allocate memory!"); + } #ifdef DEBUG end_ptr = ptr + size; #endif @@ -3944,19 +4216,7 @@ void erts_silence_warn_unused_result(long unused) void erts_interval_init(erts_interval_t *icp) { -#ifdef ARCH_64 - erts_atomic_init_nob(&icp->counter.atomic, 0); -#else - erts_dw_aint_t dw; -#ifdef ETHR_SU_DW_NAINT_T__ - dw.dw_sint = 0; -#else - dw.sint[ERTS_DW_AINT_HIGH_WORD] = 0; - dw.sint[ERTS_DW_AINT_LOW_WORD] = 0; -#endif - erts_dw_atomic_init_nob(&icp->counter.atomic, &dw); - -#endif + erts_atomic64_init_nob(&icp->counter.atomic, 0); #ifdef DEBUG icp->smp_api = 0; #endif @@ -3978,55 +4238,13 @@ erts_smp_interval_init(erts_interval_t *icp) static ERTS_INLINE Uint64 step_interval_nob(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_inc_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_nob(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - } -#endif + return (Uint64) erts_atomic64_inc_read_nob(&icp->counter.atomic); } static ERTS_INLINE Uint64 step_interval_relb(erts_interval_t *icp) { -#ifdef ARCH_64 - return (Uint64) erts_atomic_inc_read_relb(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_relb(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - } -#endif + return (Uint64) erts_atomic64_inc_read_relb(&icp->counter.atomic); } @@ -4034,38 +4252,10 @@ static ERTS_INLINE Uint64 ensure_later_interval_nob(erts_interval_t *icp, Uint64 ic) { Uint64 curr_ic; -#ifdef ARCH_64 - curr_ic = (Uint64) erts_atomic_read_nob(&icp->counter.atomic); - if (curr_ic > ic) - return curr_ic; - return (Uint64) erts_atomic_inc_read_nob(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_nob(&icp->counter.atomic, &exp); - curr_ic = erts_interval_dw_aint_to_val__(&exp); + curr_ic = (Uint64) erts_atomic64_read_nob(&icp->counter.atomic); if (curr_ic > ic) return curr_ic; - - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_nob(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - } -#endif + return (Uint64) erts_atomic64_inc_read_nob(&icp->counter.atomic); } @@ -4073,38 +4263,10 @@ static ERTS_INLINE Uint64 ensure_later_interval_acqb(erts_interval_t *icp, Uint64 ic) { Uint64 curr_ic; -#ifdef ARCH_64 - curr_ic = (Uint64) erts_atomic_read_acqb(&icp->counter.atomic); + curr_ic = (Uint64) erts_atomic64_read_acqb(&icp->counter.atomic); if (curr_ic > ic) return curr_ic; - return (Uint64) erts_atomic_inc_read_acqb(&icp->counter.atomic); -#else - erts_dw_aint_t exp; - - erts_dw_atomic_read_acqb(&icp->counter.atomic, &exp); - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - - while (1) { - erts_dw_aint_t new = exp; - -#ifdef ETHR_SU_DW_NAINT_T__ - new.dw_sint++; -#else - new.sint[ERTS_DW_AINT_LOW_WORD]++; - if (new.sint[ERTS_DW_AINT_LOW_WORD] == 0) - new.sint[ERTS_DW_AINT_HIGH_WORD]++; -#endif - - if (erts_dw_atomic_cmpxchg_acqb(&icp->counter.atomic, &new, &exp)) - return erts_interval_dw_aint_to_val__(&new); - - curr_ic = erts_interval_dw_aint_to_val__(&exp); - if (curr_ic > ic) - return curr_ic; - } -#endif + return (Uint64) erts_atomic64_inc_read_acqb(&icp->counter.atomic); } Uint64 |