diff options
Diffstat (limited to 'erts/emulator/beam')
-rw-r--r-- | erts/emulator/beam/arith_instrs.tab | 83 | ||||
-rw-r--r-- | erts/emulator/beam/beam_emu.c | 39 | ||||
-rw-r--r-- | erts/emulator/beam/dist.c | 30 | ||||
-rw-r--r-- | erts/emulator/beam/erl_bif_info.c | 2 | ||||
-rw-r--r-- | erts/emulator/beam/erl_node_tables.h | 1 | ||||
-rw-r--r-- | erts/emulator/beam/erl_proc_sig_queue.c | 2 | ||||
-rw-r--r-- | erts/emulator/beam/erl_process.c | 6 | ||||
-rw-r--r-- | erts/emulator/beam/external.c | 6 | ||||
-rw-r--r-- | erts/emulator/beam/external.h | 8 | ||||
-rw-r--r-- | erts/emulator/beam/global.h | 9 | ||||
-rw-r--r-- | erts/emulator/beam/sys.h | 8 |
11 files changed, 136 insertions, 58 deletions
diff --git a/erts/emulator/beam/arith_instrs.tab b/erts/emulator/beam/arith_instrs.tab index 5f23b2c168..f14b376419 100644 --- a/erts/emulator/beam/arith_instrs.tab +++ b/erts/emulator/beam/arith_instrs.tab @@ -51,11 +51,50 @@ plus.fetch(Op1, Op2) { plus.execute(Fail, Dst) { if (ERTS_LIKELY(is_both_small(PlusOp1, PlusOp2))) { +#ifdef HAVE_OVERFLOW_CHECK_BUILTINS + Sint lhs_tagged, rhs_untagged, res; + + /* The value part of immediate integers start right after the tag and + * occupy the rest of the word, so if you squint a bit they look like + * fixed-point integers; as long as you mask the tag away you will get + * correct results from addition/subtraction since they share the same + * notion of zero. It's fairly easy to see that the following holds + * when (a + b) is in range: + * + * (a >> s) + (b >> s) == ((a & ~m) + (b & ~m)) >> s + * + * Where 's' is the tag size and 'm' is the tag mask. + * + * The left-hand side is our fallback in the #else clause and is the + * fastest way to do this safely in plain C. The actual addition will + * never overflow since `Sint` has a much greater range than our + * smalls, so we can use the IS_SSMALL macro to see if the result is + * within range. + * + * What we're doing below is an extension of the right-hand side. By + * treating `a` and `b` as fixed-point integers, all additions whose + * result is out of range will also overflow `Sint` and we can use the + * compiler's overflow intrinsics to check for this condition. + * + * In addition, since the tag lives in the lowest bits we can further + * optimize this by only stripping the tag from either side. The higher + * bits can't influence the tag bits since we bail on overflow, so the + * tag bits from the tagged side will simply appear in the result. */ + lhs_tagged = PlusOp1; + rhs_untagged = PlusOp2 & ~_TAG_IMMED1_MASK; + + if (ERTS_LIKELY(!__builtin_add_overflow(lhs_tagged, rhs_untagged, &res))) { + ASSERT(is_small(res)); + $Dst = res; + $NEXT0(); + } +#else Sint i = signed_val(PlusOp1) + signed_val(PlusOp2); if (ERTS_LIKELY(IS_SSMALL(i))) { $Dst = make_small(i); $NEXT0(); } +#endif } $OUTLINED_ARITH_2($Fail, mixed_plus, BIF_splus_2, PlusOp1, PlusOp2, $Dst); } @@ -73,11 +112,26 @@ minus.fetch(Op1, Op2) { minus.execute(Fail, Dst) { if (ERTS_LIKELY(is_both_small(MinusOp1, MinusOp2))) { +#ifdef HAVE_OVERFLOW_CHECK_BUILTINS + Sint lhs_tagged, rhs_untagged, res; + + /* See plus.execute */ + lhs_tagged = MinusOp1; + rhs_untagged = MinusOp2 & ~_TAG_IMMED1_MASK; + + if (ERTS_LIKELY(!__builtin_sub_overflow(lhs_tagged, rhs_untagged, &res))) { + ASSERT(is_small(res)); + $Dst = res; + $NEXT0(); + } +#else Sint i = signed_val(MinusOp1) - signed_val(MinusOp2); + if (ERTS_LIKELY(IS_SSMALL(i))) { $Dst = make_small(i); $NEXT0(); } +#endif } $OUTLINED_ARITH_2($Fail, mixed_minus, BIF_sminus_2, MinusOp1, MinusOp2, $Dst); } @@ -97,12 +151,27 @@ increment.execute(IncrementVal, Dst) { Eterm result; if (ERTS_LIKELY(is_small(increment_reg_val))) { +#ifdef HAVE_OVERFLOW_CHECK_BUILTINS + Sint lhs_tagged, rhs_untagged, res; + + /* See plus.execute */ + lhs_tagged = increment_reg_val; + rhs_untagged = (Sint)increment_val << _TAG_IMMED1_SIZE; + + if (ERTS_LIKELY(!__builtin_add_overflow(lhs_tagged, rhs_untagged, &res))) { + ASSERT(is_small(res)); + $Dst = res; + $NEXT0(); + } +#else Sint i = signed_val(increment_reg_val) + increment_val; if (ERTS_LIKELY(IS_SSMALL(i))) { $Dst = make_small(i); $NEXT0(); } +#endif } + result = erts_mixed_plus(c_p, increment_reg_val, make_small(increment_val)); ERTS_HOLE_CHECK(c_p); if (ERTS_LIKELY(is_value(result))) { @@ -118,11 +187,15 @@ i_times(Fail, Op1, Op2, Dst) { Eterm op2 = $Op2; #ifdef HAVE_OVERFLOW_CHECK_BUILTINS if (ERTS_LIKELY(is_both_small(op1, op2))) { - Sint a = signed_val(op1); - Sint b = signed_val(op2); - Sint res; - if (ERTS_LIKELY(!__builtin_mul_overflow(a, b, &res) && IS_SSMALL(res))) { - $Dst = make_small(res); + /* See plus.execute */ + Sint lhs_untagged, rhs_actual, res; + + lhs_untagged = op1 & ~_TAG_IMMED1_MASK; + rhs_actual = signed_val(op2); + + if (ERTS_LIKELY(!__builtin_mul_overflow(lhs_untagged, rhs_actual, &res))) { + ASSERT(!(res & _TAG_IMMED1_MASK)); + $Dst = res | _TAG_IMMED1_SMALL; $NEXT0(); } } diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index ea01ce597d..8e93e53003 100644 --- a/erts/emulator/beam/beam_emu.c +++ b/erts/emulator/beam/beam_emu.c @@ -375,44 +375,33 @@ do { \ /* * process_main() is already huge, so we want to avoid inlining - * into it. Especially functions that are seldom used. + * seldom used functions into it. */ -#ifdef __GNUC__ -# define NOINLINE __attribute__((__noinline__)) -#else -# define NOINLINE -#endif - - -/* - * The following functions are called directly by process_main(). - * Don't inline them. - */ -static void init_emulator_finish(void) NOINLINE; -static ErtsCodeMFA *ubif2mfa(void* uf) NOINLINE; +static void init_emulator_finish(void) ERTS_NOINLINE; +static ErtsCodeMFA *ubif2mfa(void* uf) ERTS_NOINLINE; static BeamInstr* handle_error(Process* c_p, BeamInstr* pc, - Eterm* reg, ErtsCodeMFA* bif_mfa) NOINLINE; + Eterm* reg, ErtsCodeMFA* bif_mfa) ERTS_NOINLINE; static BeamInstr* call_error_handler(Process* p, ErtsCodeMFA* mfa, - Eterm* reg, Eterm func) NOINLINE; + Eterm* reg, Eterm func) ERTS_NOINLINE; static BeamInstr* fixed_apply(Process* p, Eterm* reg, Uint arity, - BeamInstr *I, Uint offs) NOINLINE; + BeamInstr *I, Uint offs) ERTS_NOINLINE; static BeamInstr* apply(Process* p, Eterm* reg, - BeamInstr *I, Uint offs) NOINLINE; + BeamInstr *I, Uint offs) ERTS_NOINLINE; static BeamInstr* call_fun(Process* p, int arity, - Eterm* reg, Eterm args) NOINLINE; + Eterm* reg, Eterm args) ERTS_NOINLINE; static BeamInstr* apply_fun(Process* p, Eterm fun, - Eterm args, Eterm* reg) NOINLINE; + Eterm args, Eterm* reg) ERTS_NOINLINE; static Eterm new_fun(Process* p, Eterm* reg, - ErlFunEntry* fe, int num_free) NOINLINE; + ErlFunEntry* fe, int num_free) ERTS_NOINLINE; static int is_function2(Eterm Term, Uint arity); static Eterm erts_gc_new_map(Process* p, Eterm* reg, Uint live, - Uint n, BeamInstr* ptr) NOINLINE; + Uint n, BeamInstr* ptr) ERTS_NOINLINE; static Eterm erts_gc_new_small_map_lit(Process* p, Eterm* reg, Eterm keys_literal, - Uint live, BeamInstr* ptr) NOINLINE; + Uint live, BeamInstr* ptr) ERTS_NOINLINE; static Eterm erts_gc_update_map_assoc(Process* p, Eterm* reg, Uint live, - Uint n, BeamInstr* new_p) NOINLINE; + Uint n, BeamInstr* new_p) ERTS_NOINLINE; static Eterm erts_gc_update_map_exact(Process* p, Eterm* reg, Uint live, - Uint n, Eterm* new_p) NOINLINE; + Uint n, Eterm* new_p) ERTS_NOINLINE; static Eterm get_map_element(Eterm map, Eterm key); static Eterm get_map_element_hash(Eterm map, Eterm key, Uint32 hx); diff --git a/erts/emulator/beam/dist.c b/erts/emulator/beam/dist.c index 8bbe6450eb..456b6fdac0 100644 --- a/erts/emulator/beam/dist.c +++ b/erts/emulator/beam/dist.c @@ -775,19 +775,25 @@ void init_dist(void) static ERTS_INLINE ErtsDistOutputBuf * alloc_dist_obuf(Uint size, Uint headers) { - int i; + Uint obuf_size = sizeof(ErtsDistOutputBuf)*(headers); ErtsDistOutputBuf *obuf; - Uint obuf_size = sizeof(ErtsDistOutputBuf)*(headers) + - sizeof(byte)*size; - Binary *bin = erts_bin_drv_alloc(obuf_size); - obuf = (ErtsDistOutputBuf *) &bin->orig_bytes[size]; + Binary *bin; + byte *extp; + int i; + + bin = erts_bin_drv_alloc(obuf_size + size); erts_refc_add(&bin->intern.refc, headers - 1, 1); + + obuf = (ErtsDistOutputBuf *)&bin->orig_bytes[0]; + extp = (byte *)&bin->orig_bytes[obuf_size]; + for (i = 0; i < headers; i++) { obuf[i].bin = bin; - obuf[i].extp = (byte *)&bin->orig_bytes[0]; + obuf[i].extp = extp; #ifdef DEBUG obuf[i].dbg_pattern = ERTS_DIST_OUTPUT_BUF_DBG_PATTERN; - obuf[i].alloc_endp = obuf->extp + size; + obuf[i].ext_startp = extp; + obuf[i].alloc_endp = &extp[size]; ASSERT(bin == ErtsDistOutputBuf2Binary(obuf)); #endif } @@ -1360,7 +1366,7 @@ erts_dist_seq_tree_foreach_delete_yielding(DistSeqNode **root, limit); if (res > 0) { if (ysp != &ys) - erts_free(ERTS_ALC_T_ML_YIELD_STATE, ysp); + erts_free(ERTS_ALC_T_SEQ_YIELD_STATE, ysp); *vyspp = NULL; } else { @@ -2341,7 +2347,8 @@ erts_dsig_send(ErtsDSigSendContext *ctx) (ctx->fragments-1) * ERTS_DIST_FRAGMENT_HEADER_SIZE, ctx->fragments); ctx->obuf->ext_start = &ctx->obuf->extp[0]; - ctx->obuf->ext_endp = &ctx->obuf->extp[0] + ctx->max_finalize_prepend + ctx->dhdr_ext_size; + ctx->obuf->ext_endp = &ctx->obuf->extp[0] + ctx->max_finalize_prepend + + ctx->dhdr_ext_size; /* Encode internal version of dist header */ ctx->obuf->extp = erts_encode_ext_dist_header_setup( @@ -2380,8 +2387,8 @@ erts_dsig_send(ErtsDSigSendContext *ctx) case ERTS_DSIG_SEND_PHASE_FIN: { ASSERT(ctx->obuf->extp < ctx->obuf->ext_endp); - ASSERT(((byte*)&ctx->obuf->bin->orig_bytes[0]) <= ctx->obuf->extp - ctx->max_finalize_prepend); - ASSERT(ctx->obuf->ext_endp <= ((byte*)ctx->obuf->bin->orig_bytes) + ctx->data_size + ctx->dhdr_ext_size); + ASSERT(ctx->obuf->ext_startp <= ctx->obuf->extp - ctx->max_finalize_prepend); + ASSERT(ctx->obuf->ext_endp <= (byte*)ctx->obuf->ext_startp + ctx->data_size + ctx->dhdr_ext_size); ctx->data_size = ctx->obuf->ext_endp - ctx->obuf->extp; @@ -3457,6 +3464,7 @@ dist_ctrl_get_data_1(BIF_ALIST_1) pb->bytes = (byte*) obuf->extp; pb->flags = 0; res = make_binary(pb); + hp += PROC_BIN_SIZE; } else { hp = HAlloc(BIF_P, PROC_BIN_SIZE * 2 + 4 + hsz); pb = (ProcBin *) (char *) hp; diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index a7424bbcb8..39d42d9757 100644 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -2579,6 +2579,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) /* Need to be the only thread running... */ erts_proc_unlock(BIF_P, ERTS_PROC_LOCK_MAIN); + BIF_P->scheduler_data->current_process = NULL; erts_thr_progress_block(); if (BIF_ARG_1 == am_info) @@ -2592,6 +2593,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) erts_thr_progress_unblock(); erts_proc_lock(BIF_P, ERTS_PROC_LOCK_MAIN); + BIF_P->scheduler_data->current_process = BIF_P; ASSERT(dsbufp && dsbufp->str); res = new_binary(BIF_P, (byte *) dsbufp->str, dsbufp->str_len); diff --git a/erts/emulator/beam/erl_node_tables.h b/erts/emulator/beam/erl_node_tables.h index c434926142..fc3e117463 100644 --- a/erts/emulator/beam/erl_node_tables.h +++ b/erts/emulator/beam/erl_node_tables.h @@ -95,6 +95,7 @@ enum dist_entry_state { struct ErtsDistOutputBuf_ { #ifdef DEBUG Uint dbg_pattern; + byte *ext_startp; byte *alloc_endp; #endif ErtsDistOutputBuf *next; diff --git a/erts/emulator/beam/erl_proc_sig_queue.c b/erts/emulator/beam/erl_proc_sig_queue.c index 4e9f177e51..f58a606d57 100644 --- a/erts/emulator/beam/erl_proc_sig_queue.c +++ b/erts/emulator/beam/erl_proc_sig_queue.c @@ -1019,6 +1019,8 @@ send_gen_exit_signal(Process *c_p, Eterm from_tag, ref_sz = size_object(ref); hsz += ref_sz; + reason_sz = 0; /* Set to silence gcc warning */ + /* The reason was part of the control message, just use copy it into the xsigd */ if (is_value(reason)) { diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index 9e662632b4..2b45d2d353 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -12141,10 +12141,9 @@ erts_proc_exit_handle_dist_monitor(ErtsMonitor *mon, void *vctxt, Sint reds) reason); switch (code) { case ERTS_DSIG_SEND_CONTINUE: + case ERTS_DSIG_SEND_YIELD: erts_set_gc_state(c_p, 0); ctxt->dist_state = erts_dsend_export_trap_context(c_p, &ctx); - /* fall-through */ - case ERTS_DSIG_SEND_YIELD: break; case ERTS_DSIG_SEND_OK: break; @@ -12388,11 +12387,10 @@ erts_proc_exit_handle_dist_link(ErtsLink *lnk, void *vctxt, Sint reds) reason, SEQ_TRACE_TOKEN(c_p)); switch (code) { + case ERTS_DSIG_SEND_YIELD: case ERTS_DSIG_SEND_CONTINUE: erts_set_gc_state(c_p, 0); ctxt->dist_state = erts_dsend_export_trap_context(c_p, &ctx); - /* fall-through */ - case ERTS_DSIG_SEND_YIELD: break; case ERTS_DSIG_SEND_OK: break; diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c index 471c1c3938..fa8314c115 100644 --- a/erts/emulator/beam/external.c +++ b/erts/emulator/beam/external.c @@ -699,6 +699,7 @@ dist_ext_size(ErtsDistExternal *edep) } else { sz -= sizeof(ErtsAtomTranslationTable); } + ASSERT(sz % 4 == 0); return sz; } @@ -706,8 +707,9 @@ Uint erts_dist_ext_size(ErtsDistExternal *edep) { Uint sz = dist_ext_size(edep); + sz += 4; /* may need to pad to 8-byte-align ErtsDistExternalData */ sz += edep->data[0].frag_id * sizeof(ErtsDistExternalData); - return sz + ERTS_EXTRA_DATA_ALIGN_SZ(sz); + return sz; } Uint @@ -749,6 +751,8 @@ erts_make_dist_ext_copy(ErtsDistExternal *edep, ErtsDistExternal *new_edep) erts_ref_dist_entry(new_edep->dep); ep += dist_ext_sz; + ep += (UWord)ep & 4; /* 8-byte alignment for ErtsDistExternalData */ + ASSERT((UWord)ep % 8 == 0); new_edep->data = (ErtsDistExternalData*)ep; sys_memzero(new_edep->data, sizeof(ErtsDistExternalData) * edep->data->frag_id); diff --git a/erts/emulator/beam/external.h b/erts/emulator/beam/external.h index 396cd9f802..f2cc9bf98f 100644 --- a/erts/emulator/beam/external.h +++ b/erts/emulator/beam/external.h @@ -144,14 +144,6 @@ typedef struct erl_dist_external { ErtsAtomTranslationTable attab; } ErtsDistExternal; -#define ERTS_DIST_EXT_SIZE(EDEP) \ - (sizeof(ErtsDistExternal) \ - - (((EDEP)->flags & ERTS_DIST_EXT_ATOM_TRANS_TAB) \ - ? (ASSERT(0 <= (EDEP)->attab.size \ - && (EDEP)->attab.size <= ERTS_ATOM_CACHE_SIZE), \ - sizeof(Eterm)*(ERTS_ATOM_CACHE_SIZE - (EDEP)->attab.size)) \ - : sizeof(ErtsAtomTranslationTable))) - typedef struct { byte *extp; int exttmp; diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h index f9bbe4167f..4c8d3d3dbe 100644 --- a/erts/emulator/beam/global.h +++ b/erts/emulator/beam/global.h @@ -1216,10 +1216,11 @@ Uint64 erts_timestamp_millis(void); Export* erts_find_function(Eterm, Eterm, unsigned int, ErtsCodeIndex); -void *erts_calc_stacklimit(char *prev_c, UWord stacksize); -int erts_check_below_limit(char *ptr, char *limit); -int erts_check_above_limit(char *ptr, char *limit); -void *erts_ptr_id(void *ptr); +/* ERTS_NOINLINE prevents link-time optimization across modules */ +void *erts_calc_stacklimit(char *prev_c, UWord stacksize) ERTS_NOINLINE; +int erts_check_below_limit(char *ptr, char *limit) ERTS_NOINLINE; +int erts_check_above_limit(char *ptr, char *limit) ERTS_NOINLINE; +void *erts_ptr_id(void *ptr) ERTS_NOINLINE; Eterm store_external_or_ref_in_proc_(Process *, Eterm); Eterm store_external_or_ref_(Uint **, ErlOffHeap*, Eterm); diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h index a6312293cc..c261c8e117 100644 --- a/erts/emulator/beam/sys.h +++ b/erts/emulator/beam/sys.h @@ -63,6 +63,14 @@ # endif #endif +#ifndef ERTS_NOINLINE +# if ERTS_AT_LEAST_GCC_VSN__(3,1,1) +# define ERTS_NOINLINE __attribute__((__noinline__)) +# else +# define ERTS_NOINLINE +# endif +#endif + #if defined(DEBUG) || defined(ERTS_ENABLE_LOCK_CHECK) # undef ERTS_CAN_INLINE # define ERTS_CAN_INLINE 0 |