diff options
Diffstat (limited to 'erts')
-rw-r--r-- | erts/doc/src/erlang.xml | 115 | ||||
-rw-r--r-- | erts/emulator/beam/atom.names | 4 | ||||
-rw-r--r-- | erts/emulator/beam/beam_bp.c | 99 | ||||
-rw-r--r-- | erts/emulator/beam/beam_bp.h | 7 | ||||
-rw-r--r-- | erts/emulator/beam/erl_bif_info.c | 37 | ||||
-rw-r--r-- | erts/emulator/beam/erl_process.c | 140 | ||||
-rw-r--r-- | erts/emulator/beam/erl_process.h | 50 | ||||
-rw-r--r-- | erts/emulator/test/alloc_SUITE_data/threads.c | 4 | ||||
-rw-r--r-- | erts/emulator/test/statistics_SUITE.erl | 61 | ||||
-rw-r--r-- | erts/epmd/src/epmd_srv.c | 3 | ||||
-rw-r--r-- | erts/preloaded/ebin/erlang.beam | bin | 101796 -> 102000 bytes | |||
-rw-r--r-- | erts/preloaded/src/erlang.erl | 10 |
12 files changed, 354 insertions, 176 deletions
diff --git a/erts/doc/src/erlang.xml b/erts/doc/src/erlang.xml index c37ed3bea5..13b16d1ed3 100644 --- a/erts/doc/src/erlang.xml +++ b/erts/doc/src/erlang.xml @@ -5684,8 +5684,31 @@ true</pre> <anno>Dest</anno>, <anno>Msg</anno>, [])</c></seealso>.</p> </desc> </func> + <func> <name name="statistics" arity="1" clause_i="1"/> + <fsummary>Information about active processes and ports.</fsummary> + <desc><marker id="statistics_active_tasks"></marker> + <p> + Returns a list where each element represents the amount + of active processes and ports on each run queue and its + associated scheduler. That is, the number of processes and + ports that are ready to run, or are currently running. The + element location in the list corresponds to the scheduler + and its run queue. The first element corresponds to scheduler + number 1 and so on. The information is <em>not</em> gathered + atomically. That is, the result is not necessarily a + consistent snapshot of the state, but instead quite + efficiently gathered. See also, + <seealso marker="#statistics_total_active_tasks"><c>statistics(total_active_tasks)</c></seealso>, + <seealso marker="#statistics_run_queue_lengths"><c>statistics(run_queue_lengths)</c></seealso>, and + <seealso marker="#statistics_total_run_queue_lengths"><c>statistics(total_run_queue_lengths)</c></seealso>. + </p> + </desc> + </func> + + <func> + <name name="statistics" arity="1" clause_i="2"/> <fsummary>Information about context switches.</fsummary> <desc> <p>Returns the total number of context switches since the @@ -5694,7 +5717,7 @@ true</pre> </func> <func> - <name name="statistics" arity="1" clause_i="2"/> + <name name="statistics" arity="1" clause_i="3"/> <fsummary>Information about exact reductions.</fsummary> <desc> <marker id="statistics_exact_reductions"></marker> @@ -5708,7 +5731,7 @@ true</pre> </func> <func> - <name name="statistics" arity="1" clause_i="3"/> + <name name="statistics" arity="1" clause_i="4"/> <fsummary>Information about garbage collection.</fsummary> <desc> <p>Returns information about garbage collection, for example:</p> @@ -5720,7 +5743,7 @@ true</pre> </func> <func> - <name name="statistics" arity="1" clause_i="4"/> + <name name="statistics" arity="1" clause_i="5"/> <fsummary>Information about I/O.</fsummary> <desc> <p>Returns <c><anno>Input</anno></c>, @@ -5731,7 +5754,7 @@ true</pre> </func> <func> - <name name="statistics" arity="1" clause_i="5"/> + <name name="statistics" arity="1" clause_i="6"/> <fsummary>Information about reductions.</fsummary> <desc> <marker id="statistics_reductions"></marker> @@ -5749,16 +5772,43 @@ true</pre> </func> <func> - <name name="statistics" arity="1" clause_i="6"/> - <fsummary>Information about the run-queue.</fsummary> - <desc> - <p>Returns the total length of run-queues, that is, the number - of processes that are ready to run on all available run-queues.</p> + <name name="statistics" arity="1" clause_i="7"/> + <fsummary>Information about the run-queues.</fsummary> + <desc><marker id="statistics_run_queue"></marker> + <p> + Returns the total length of the run-queues. That is, the number + of processes and ports that are ready to run on all available + run-queues. The information is gathered atomically. That + is, the result is a consistent snapshot of the state, but + this operation is much more expensive compared to + <seealso marker="#statistics_total_run_queue_lengths"><c>statistics(total_run_queue_lengths)</c></seealso>. + This especially when a large amount of schedulers is used. + </p> </desc> </func> <func> - <name name="statistics" arity="1" clause_i="7"/> + <name name="statistics" arity="1" clause_i="8"/> + <fsummary>Information about the run-queue lengths.</fsummary> + <desc><marker id="statistics_run_queue_lengths"></marker> + <p> + Returns a list where each element represents the amount + of processes and ports ready to run for each run queue. The + element location in the list corresponds to the run queue + of a scheduler. The first element corresponds to the run + queue of scheduler number 1 and so on. The information is + <em>not</em> gathered atomically. That is, the result is + not necessarily a consistent snapshot of the state, but + instead quite efficiently gathered. See also, + <seealso marker="#statistics_total_run_queue_lengths"><c>statistics(total_run_queue_lengths)</c></seealso>, + <seealso marker="#statistics_active_tasks"><c>statistics(active_tasks)</c></seealso>, and + <seealso marker="#statistics_total_active_tasks"><c>statistics(total_active_tasks)</c></seealso>. + </p> + </desc> + </func> + + <func> + <name name="statistics" arity="1" clause_i="9"/> <fsummary>Information about runtime.</fsummary> <desc> <p>Returns information about runtime, in milliseconds.</p> @@ -5773,7 +5823,7 @@ true</pre> </func> <func> - <name name="statistics" arity="1" clause_i="8"/> + <name name="statistics" arity="1" clause_i="10"/> <fsummary>Information about each schedulers work time.</fsummary> <desc> <marker id="statistics_scheduler_wall_time"></marker> @@ -5844,7 +5894,44 @@ ok </func> <func> - <name name="statistics" arity="1" clause_i="9"/> + <name name="statistics" arity="1" clause_i="11"/> + <fsummary>Information about active processes and ports.</fsummary> + <desc><marker id="statistics_total_active_tasks"></marker> + <p> + Returns the total amount of active processes and ports in + the system. That is, the number of processes and ports that + are ready to run, or are currently running. The information + is <em>not</em> gathered atomically. That is, the result + is not necessarily a consistent snapshot of the state, but + instead quite efficiently gathered. See also, + <seealso marker="#statistics_active_tasks"><c>statistics(active_tasks)</c></seealso>, + <seealso marker="#statistics_run_queue_lengths"><c>statistics(run_queue_lengths)</c></seealso>, and + <seealso marker="#statistics_total_run_queue_lengths"><c>statistics(total_run_queue_lengths)</c></seealso>. + </p> + </desc> + </func> + + <func> + <name name="statistics" arity="1" clause_i="12"/> + <fsummary>Information about the run-queue lengths.</fsummary> + <desc><marker id="statistics_total_run_queue_lengths"></marker> + <p> + Returns the total length of the run-queues. That is, the number + of processes and ports that are ready to run on all available + run-queues. The information is <em>not</em> gathered atomically. + That is, the result is not necessarily a consistent snapshot of + the state, but much more efficiently gathered compared to + <seealso marker="#statistics_run_queue"><c>statistics(run_queue)</c></seealso>. + See also, + <seealso marker="#statistics_run_queue_lengths"><c>statistics(run_queue_lengths)</c></seealso>, + <seealso marker="#statistics_total_active_tasks"><c>statistics(total_active_tasks)</c></seealso>, and + <seealso marker="#statistics_active_tasks"><c>statistics(active_tasks)</c></seealso>. + </p> + </desc> + </func> + + <func> + <name name="statistics" arity="1" clause_i="13"/> <fsummary>Information about wall clock.</fsummary> <desc> <p>Returns information about wall clock. <c>wall_clock</c> can @@ -8182,14 +8269,14 @@ timestamp() -> <p>When <c>Pid</c> is scheduled to run. The process runs in function <c>{M, F, Arity}</c>. On some rare occasions, the current function cannot be determined, - then the last element <c>Arity</c> is <c>0</c>.</p> + then the last element is <c>0</c>.</p> </item> <tag><c>{trace, Pid, out, {M, F, Arity} | 0}</c></tag> <item> <p>When <c>Pid</c> is scheduled out. The process was running in function {M, F, Arity}. On some rare occasions, the current function cannot be determined, then the last - element <c>Arity</c> is <c>0</c>.</p> + element is <c>0</c>.</p> </item> <tag><c>{trace, Pid, gc_start, Info}</c></tag> <item> diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names index 190e7817dc..fb3368eae2 100644 --- a/erts/emulator/beam/atom.names +++ b/erts/emulator/beam/atom.names @@ -71,6 +71,7 @@ atom absoluteURI atom ac atom accessor atom active +atom active_tasks atom all atom all_but_first atom all_names @@ -512,6 +513,7 @@ atom return_from atom return_to atom return_trace atom run_queue +atom run_queue_lengths atom runnable atom runnable_ports atom runnable_procs @@ -579,7 +581,9 @@ atom timeout_value atom Times='*' atom timestamp atom total +atom total_active_tasks atom total_heap_size +atom total_run_queue_lengths atom tpkt atom trace trace_ts traced atom trace_control_word diff --git a/erts/emulator/beam/beam_bp.c b/erts/emulator/beam/beam_bp.c index 016d0aaa32..9860968687 100644 --- a/erts/emulator/beam/beam_bp.c +++ b/erts/emulator/beam/beam_bp.c @@ -75,6 +75,16 @@ extern BeamInstr beam_return_time_trace[1]; /* OpCode(i_return_time_trace) */ erts_smp_atomic32_t erts_active_bp_index; erts_smp_atomic32_t erts_staging_bp_index; +/* + * Inlined helpers + */ + +static ERTS_INLINE ErtsMonotonicTime +get_mtime(Process *c_p) +{ + return erts_get_monotonic_time(ERTS_PROC_GET_SCHDATA(c_p)); +} + /* ************************************************************************* ** Local prototypes */ @@ -97,9 +107,6 @@ static int clear_function_break(BeamInstr *pc, Uint break_flags); static BpDataTime* get_time_break(BeamInstr *pc); static GenericBpData* check_break(BeamInstr *pc, Uint break_flags); -static void bp_time_diff(bp_data_time_item_t *item, - process_breakpoint_time_t *pbt, - Uint ms, Uint s, Uint us); static void bp_meta_unref(BpMetaPid* bmp); static void bp_count_unref(BpCount* bcp); @@ -110,13 +117,8 @@ static void uninstall_breakpoint(BeamInstr* pc); /* bp_hash */ #define BP_TIME_ADD(pi0, pi1) \ do { \ - Uint r; \ (pi0)->count += (pi1)->count; \ - (pi0)->s_time += (pi1)->s_time; \ - (pi0)->us_time += (pi1)->us_time; \ - r = (pi0)->us_time / 1000000; \ - (pi0)->s_time += r; \ - (pi0)->us_time = (pi0)->us_time % 1000000; \ + (pi0)->time += (pi1)->time; \ } while(0) static void bp_hash_init(bp_time_hash_t *hash, Uint n); @@ -948,7 +950,7 @@ do_call_trace(Process* c_p, BeamInstr* I, Eterm* reg, void erts_trace_time_call(Process* c_p, BeamInstr* I, BpDataTime* bdt) { - Uint ms,s,us; + ErtsMonotonicTime time; process_breakpoint_time_t *pbt = NULL; bp_data_time_item_t sitem, *item = NULL; bp_time_hash_t *h = NULL; @@ -961,7 +963,7 @@ erts_trace_time_call(Process* c_p, BeamInstr* I, BpDataTime* bdt) * from the process psd */ pbt = ERTS_PROC_GET_CALL_TIME(c_p); - get_sys_now(&ms, &s, &us); + time = get_mtime(c_p); /* get pbt * timestamp = t0 @@ -976,7 +978,7 @@ erts_trace_time_call(Process* c_p, BeamInstr* I, BpDataTime* bdt) } else { ASSERT(pbt->pc); /* add time to previous code */ - bp_time_diff(&sitem, pbt, ms, s, us); + sitem.time = time - pbt->time; sitem.pid = c_p->common.id; sitem.count = 0; @@ -1002,8 +1004,7 @@ erts_trace_time_call(Process* c_p, BeamInstr* I, BpDataTime* bdt) /* Add count to this code */ sitem.pid = c_p->common.id; sitem.count = 1; - sitem.s_time = 0; - sitem.us_time = 0; + sitem.time = 0; /* this breakpoint */ ASSERT(bdt); @@ -1020,15 +1021,13 @@ erts_trace_time_call(Process* c_p, BeamInstr* I, BpDataTime* bdt) } pbt->pc = I; - pbt->ms = ms; - pbt->s = s; - pbt->us = us; + pbt->time = time; } void erts_trace_time_return(Process *p, BeamInstr *pc) { - Uint ms,s,us; + ErtsMonotonicTime time; process_breakpoint_time_t *pbt = NULL; bp_data_time_item_t sitem, *item = NULL; bp_time_hash_t *h = NULL; @@ -1041,7 +1040,7 @@ erts_trace_time_return(Process *p, BeamInstr *pc) * from the process psd */ pbt = ERTS_PROC_GET_CALL_TIME(p); - get_sys_now(&ms,&s,&us); + time = get_mtime(p); /* get pbt * lookup bdt from code @@ -1057,7 +1056,7 @@ erts_trace_time_return(Process *p, BeamInstr *pc) */ ASSERT(pbt->pc); - bp_time_diff(&sitem, pbt, ms, s, us); + sitem.time = time - pbt->time; sitem.pid = p->common.id; sitem.count = 0; @@ -1080,9 +1079,7 @@ erts_trace_time_return(Process *p, BeamInstr *pc) } pbt->pc = pc; - pbt->ms = ms; - pbt->s = s; - pbt->us = us; + pbt->time = time; } } @@ -1183,10 +1180,14 @@ int erts_is_time_break(Process *p, BeamInstr *pc, Eterm *retval) { for(ix = 0; ix < hash.n; ix++) { item = &(hash.item[ix]); if (item->pid != NIL) { + ErtsMonotonicTime sec, usec; + usec = ERTS_MONOTONIC_TO_USEC(item->time); + sec = usec / 1000000; + usec = usec - sec*1000000; t = TUPLE4(hp, item->pid, make_small(item->count), - make_small(item->s_time), - make_small(item->us_time)); + make_small((Uint) sec), + make_small((Uint) usec)); hp += 5; *retval = CONS(hp, t, *retval); hp += 2; } @@ -1266,8 +1267,7 @@ static void bp_hash_rehash(bp_time_hash_t *hash, Uint n) { } item[hval].pid = hash->item[ix].pid; item[hval].count = hash->item[ix].count; - item[hval].s_time = hash->item[ix].s_time; - item[hval].us_time = hash->item[ix].us_time; + item[hval].time = hash->item[ix].time; } } @@ -1315,8 +1315,7 @@ static ERTS_INLINE bp_data_time_item_t * bp_hash_put(bp_time_hash_t *hash, bp_da item = &(hash->item[hval]); item->pid = sitem->pid; - item->s_time = sitem->s_time; - item->us_time = sitem->us_time; + item->time = sitem->time; item->count = sitem->count; hash->used++; @@ -1330,41 +1329,7 @@ static void bp_hash_delete(bp_time_hash_t *hash) { hash->item = NULL; } -static void bp_time_diff(bp_data_time_item_t *item, /* out */ - process_breakpoint_time_t *pbt, /* in */ - Uint ms, Uint s, Uint us) { - int ds,dus; -#ifdef DEBUG - int dms; - - - dms = ms - pbt->ms; -#endif - ds = s - pbt->s; - dus = us - pbt->us; - - /* get_sys_now may return zero difftime, - * this is ok. - */ - -#ifdef DEBUG - ASSERT(dms >= 0 || ds >= 0 || dus >= 0); -#endif - - if (dus < 0) { - dus += 1000000; - ds -= 1; - } - if (ds < 0) { - ds += 1000000; - } - - item->s_time = ds; - item->us_time = dus; -} - void erts_schedule_time_break(Process *p, Uint schedule) { - Uint ms, s, us; process_breakpoint_time_t *pbt = NULL; bp_data_time_item_t sitem, *item = NULL; bp_time_hash_t *h = NULL; @@ -1387,8 +1352,7 @@ void erts_schedule_time_break(Process *p, Uint schedule) { pbdt = get_time_break(pbt->pc); if (pbdt) { - get_sys_now(&ms,&s,&us); - bp_time_diff(&sitem, pbt, ms, s, us); + sitem.time = get_mtime(p) - pbt->time; sitem.pid = p->common.id; sitem.count = 0; @@ -1410,10 +1374,7 @@ void erts_schedule_time_break(Process *p, Uint schedule) { * timestamp it and remove the previous * timestamp in the psd. */ - get_sys_now(&ms,&s,&us); - pbt->ms = ms; - pbt->s = s; - pbt->us = us; + pbt->time = get_mtime(p); break; default : ASSERT(0); diff --git a/erts/emulator/beam/beam_bp.h b/erts/emulator/beam/beam_bp.h index 97d0539ac7..2b89d6fc71 100644 --- a/erts/emulator/beam/beam_bp.h +++ b/erts/emulator/beam/beam_bp.h @@ -29,8 +29,7 @@ typedef struct { Eterm pid; Sint count; - Uint s_time; - Uint us_time; + ErtsMonotonicTime time; } bp_data_time_item_t; typedef struct { @@ -46,9 +45,7 @@ typedef struct bp_data_time { /* Call time */ } BpDataTime; typedef struct { - Uint ms; - Uint s; - Uint us; + ErtsMonotonicTime time; BeamInstr *pc; } process_breakpoint_time_t; /* used within psd */ diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index b44382cde8..414ff6711a 100644 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -3234,6 +3234,39 @@ BIF_RETTYPE statistics_1(BIF_ALIST_1) if (is_non_value(res)) BIF_RET(am_undefined); BIF_TRAP1(gather_sched_wall_time_res_trap, BIF_P, res); + } else if (BIF_ARG_1 == am_total_active_tasks + || BIF_ARG_1 == am_total_run_queue_lengths) { + Uint no = erts_run_queues_len(NULL, 0, BIF_ARG_1 == am_total_active_tasks); + if (IS_USMALL(0, no)) + res = make_small(no); + else { + Eterm *hp = HAlloc(BIF_P, BIG_UINT_HEAP_SIZE); + res = uint_to_big(no, hp); + } + BIF_RET(res); + } else if (BIF_ARG_1 == am_active_tasks + || BIF_ARG_1 == am_run_queue_lengths) { + Eterm res, *hp, **hpp; + Uint sz, *szp; + int no_qs = erts_no_run_queues; + Uint *qszs = erts_alloc(ERTS_ALC_T_TMP,sizeof(Uint)*no_qs*2); + (void) erts_run_queues_len(qszs, 0, BIF_ARG_1 == am_active_tasks); + sz = 0; + szp = &sz; + hpp = NULL; + while (1) { + int i; + for (i = 0; i < no_qs; i++) + qszs[no_qs+i] = erts_bld_uint(hpp, szp, qszs[i]); + res = erts_bld_list(hpp, szp, no_qs, &qszs[no_qs]); + if (hpp) { + erts_free(ERTS_ALC_T_TMP, qszs); + BIF_RET(res); + } + hp = HAlloc(BIF_P, sz); + szp = NULL; + hpp = &hp; + } } else if (BIF_ARG_1 == am_context_switches) { Eterm cs = erts_make_integer(erts_get_total_context_switches(), BIF_P); hp = HAlloc(BIF_P, 3); @@ -3282,7 +3315,7 @@ BIF_RETTYPE statistics_1(BIF_ALIST_1) res = TUPLE2(hp, b1, b2); BIF_RET(res); } else if (BIF_ARG_1 == am_run_queue) { - res = erts_run_queues_len(NULL); + res = erts_run_queues_len(NULL, 1, 0); BIF_RET(make_small(res)); } else if (BIF_ARG_1 == am_wall_clock) { UWord w1, w2; @@ -3302,7 +3335,7 @@ BIF_RETTYPE statistics_1(BIF_ALIST_1) Uint sz, *szp; int no_qs = erts_no_run_queues; Uint *qszs = erts_alloc(ERTS_ALC_T_TMP,sizeof(Uint)*no_qs*2); - (void) erts_run_queues_len(qszs); + (void) erts_run_queues_len(qszs, 0, 0); sz = 0; szp = &sz; hpp = NULL; diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index d583118e7b..dd8bc9a698 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -2239,6 +2239,7 @@ handle_aux_work(ErtsAuxWorkData *awdp, erts_aint32_t orig_aux_work, int waiting) erts_aint32_t aux_work = orig_aux_work; erts_aint32_t ignore = 0; + ASSERT(!awdp->esdp || !ERTS_SCHEDULER_IS_DIRTY(awdp->esdp)); #ifdef ERTS_SMP haw_thr_prgr_current_reset(awdp); #endif @@ -2972,14 +2973,13 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) ErtsMonotonicTime current_time; aux_work = erts_atomic32_read_acqb(&ssi->aux_work); - if (aux_work) { - if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && !thr_prgr_active) { + if (aux_work && !ERTS_SCHEDULER_IS_DIRTY(esdp)) { + if (!thr_prgr_active) { erts_thr_progress_active(esdp, thr_prgr_active = 1); sched_wall_time_change(esdp, 1); } aux_work = handle_aux_work(&esdp->aux_work_data, aux_work, 1); - if (aux_work && !ERTS_SCHEDULER_IS_DIRTY(esdp) - && erts_thr_progress_update(esdp)) + if (aux_work && erts_thr_progress_update(esdp)) erts_thr_progress_leader_update(esdp); } @@ -3131,25 +3131,22 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) #endif aux_work = erts_atomic32_read_acqb(&ssi->aux_work); - if (aux_work) { - if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { - if (!working) - sched_wall_time_change(esdp, working = 1); + if (aux_work && !ERTS_SCHEDULER_IS_DIRTY(esdp)) { + if (!working) + sched_wall_time_change(esdp, working = 1); #ifdef ERTS_SMP - if (!thr_prgr_active) - erts_thr_progress_active(esdp, thr_prgr_active = 1); + if (!thr_prgr_active) + erts_thr_progress_active(esdp, thr_prgr_active = 1); #endif - } aux_work = handle_aux_work(&esdp->aux_work_data, aux_work, 1); #ifdef ERTS_SMP - if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && aux_work && - erts_thr_progress_update(esdp)) + if (aux_work && erts_thr_progress_update(esdp)) erts_thr_progress_leader_update(esdp); #endif } #ifndef ERTS_SMP - if (rq->len != 0 || rq->misc.start) + if (erts_smp_atomic32_read_dirty(&rq->len) != 0 || rq->misc.start) goto sys_woken; #else flgs = erts_smp_atomic32_read_acqb(&ssi->flags); @@ -3248,7 +3245,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) } #ifndef ERTS_SMP - if (rq->len == 0 && !rq->misc.start) + if (erts_smp_atomic32_read_dirty(&rq->len) == 0 && !rq->misc.start) goto sys_aux_work; sys_woken: #else @@ -4965,7 +4962,7 @@ erts_fprintf(stderr, "--------------------------------\n"); rq->out_of_work_count = 0; (void) ERTS_RUNQ_FLGS_READ_BSET(rq, ERTS_RUNQ_FLGS_MIGRATION_INFO, flags); - rq->max_len = rq->len; + rq->max_len = erts_smp_atomic32_read_dirty(&rq->len); for (pix = 0; pix < ERTS_NO_PRIO_LEVELS; pix++) { ErtsRunQueueInfo *rqi; rqi = (pix == ERTS_PORT_PRIO_LEVEL @@ -5123,7 +5120,7 @@ wakeup_other_check(ErtsRunQueue *rq, Uint32 flags) { int wo_reds = rq->wakeup_other_reds; if (wo_reds) { - int left_len = rq->len - 1; + int left_len = erts_smp_atomic32_read_dirty(&rq->len) - 1; if (left_len < 1) { int wo_reduce = wo_reds << wakeup_other.dec_shift; wo_reduce &= wakeup_other.dec_mask; @@ -5196,7 +5193,7 @@ wakeup_other_check_legacy(ErtsRunQueue *rq, Uint32 flags) { int wo_reds = rq->wakeup_other_reds; if (wo_reds) { - erts_aint32_t len = rq->len; + erts_aint32_t len = erts_smp_atomic32_read_dirty(&rq->len); if (len < 2) { rq->wakeup_other -= ERTS_WAKEUP_OTHER_DEC_LEGACY*wo_reds; if (rq->wakeup_other < 0) @@ -5292,7 +5289,7 @@ runq_supervisor(void *unused) ErtsRunQueue *rq = ERTS_RUNQ_IX(ix); if (ERTS_RUNQ_FLGS_GET(rq) & ERTS_RUNQ_FLG_NONEMPTY) { erts_smp_runq_lock(rq); - if (rq->len != 0) + if (erts_smp_atomic32_read_dirty(&rq->len) != 0) wake_scheduler_on_empty_runq(rq); /* forced wakeup... */ erts_smp_runq_unlock(rq); } @@ -5642,7 +5639,7 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online } rq->out_of_work_count = 0; rq->max_len = 0; - rq->len = 0; + erts_smp_atomic32_set_nob(&rq->len, 0); rq->wakeup_other = 0; rq->wakeup_other_reds = 0; rq->halt_in_progress = 0; @@ -6798,18 +6795,19 @@ suspend_scheduler(ErtsSchedulerData *esdp) & ERTS_RUNQ_FLGS_QMASK); aux_work = erts_atomic32_read_acqb(&ssi->aux_work); if (aux_work|qmask) { - if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && !thr_prgr_active) { - erts_thr_progress_active(esdp, thr_prgr_active = 1); - sched_wall_time_change(esdp, 1); + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { + if (!thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 1); + sched_wall_time_change(esdp, 1); + } + if (aux_work) + aux_work = handle_aux_work(&esdp->aux_work_data, + aux_work, + 1); + + if (aux_work && erts_thr_progress_update(esdp)) + erts_thr_progress_leader_update(esdp); } - if (aux_work) - aux_work = handle_aux_work(&esdp->aux_work_data, - aux_work, - 1); - - if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && - (aux_work && erts_thr_progress_update(esdp))) - erts_thr_progress_leader_update(esdp); if (qmask) { #ifdef ERTS_DIRTY_SCHEDULERS if (ERTS_SCHEDULER_IS_DIRTY(esdp)) { @@ -7026,17 +7024,18 @@ suspend_scheduler(ErtsSchedulerData *esdp) & ERTS_RUNQ_FLGS_QMASK); aux_work = erts_atomic32_read_acqb(&ssi->aux_work); if (aux_work|qmask) { - if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && !thr_prgr_active) { - erts_thr_progress_active(esdp, thr_prgr_active = 1); - sched_wall_time_change(esdp, 1); + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { + if (!thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 1); + sched_wall_time_change(esdp, 1); + } + if (aux_work) + aux_work = handle_aux_work(&esdp->aux_work_data, + aux_work, + 1); + if (aux_work && erts_thr_progress_update(esdp)) + erts_thr_progress_leader_update(esdp); } - if (aux_work) - aux_work = handle_aux_work(&esdp->aux_work_data, - aux_work, - 1); - if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && aux_work && - erts_thr_progress_update(esdp)) - erts_thr_progress_leader_update(esdp); if (qmask) { erts_smp_runq_lock(esdp->run_queue); evacuate_run_queue(esdp->run_queue, &sbp); @@ -7939,6 +7938,9 @@ sched_thread_func(void *vesdp) erts_sched_init_time_sup(esdp); + (void) ERTS_RUNQ_FLGS_SET_NOB(esdp->run_queue, + ERTS_RUNQ_FLG_EXEC); + #ifdef ERTS_SMP tse = erts_tse_fetch(); erts_tse_prepare_timed(tse); @@ -8947,24 +8949,39 @@ resume_process_1(BIF_ALIST_1) } Uint -erts_run_queues_len(Uint *qlen) +erts_run_queues_len(Uint *qlen, int atomic_queues_read, int incl_active_sched) { int i = 0; Uint len = 0; - ERTS_ATOMIC_FOREACH_RUNQ(rq, - { - Sint pqlen = 0; - int pix; - for (pix = 0; pix < ERTS_NO_PROC_PRIO_LEVELS; pix++) - pqlen += RUNQ_READ_LEN(&rq->procs.prio_info[pix].len); + if (atomic_queues_read) + ERTS_ATOMIC_FOREACH_RUNQ(rq, + { + Sint rq_len = (Sint) erts_smp_atomic32_read_dirty(&rq->len); + ASSERT(rq_len >= 0); + if (incl_active_sched + && (ERTS_RUNQ_FLGS_GET_NOB(rq) & ERTS_RUNQ_FLG_EXEC)) { + rq_len++; + } + if (qlen) + qlen[i++] = rq_len; + len += (Uint) rq_len; + } + ); + else { + for (i = 0; i < erts_no_run_queues; i++) { + ErtsRunQueue *rq = ERTS_RUNQ_IX(i); + Sint rq_len = (Sint) erts_smp_atomic32_read_nob(&rq->len); + ASSERT(rq_len >= 0); + if (incl_active_sched + && (ERTS_RUNQ_FLGS_GET_NOB(rq) & ERTS_RUNQ_FLG_EXEC)) { + rq_len++; + } + if (qlen) + qlen[i] = rq_len; + len += (Uint) rq_len; + } - if (pqlen < 0) - pqlen = 0; - if (qlen) - qlen[i++] = pqlen; - len += pqlen; } - ); return len; } @@ -9391,8 +9408,10 @@ Process *schedule(Process *p, int calls) if (flags & (ERTS_RUNQ_FLG_CHK_CPU_BIND|ERTS_RUNQ_FLG_SUSPENDED)) { if (flags & ERTS_RUNQ_FLG_SUSPENDED) { + (void) ERTS_RUNQ_FLGS_UNSET_NOB(rq, ERTS_RUNQ_FLG_EXEC); suspend_scheduler(esdp); - flags = ERTS_RUNQ_FLGS_GET_NOB(rq); + flags = ERTS_RUNQ_FLGS_SET_NOB(rq, ERTS_RUNQ_FLG_EXEC); + flags |= ERTS_RUNQ_FLG_EXEC; } if (flags & ERTS_RUNQ_FLG_CHK_CPU_BIND) { flags = ERTS_RUNQ_FLGS_UNSET(rq, ERTS_RUNQ_FLG_CHK_CPU_BIND); @@ -9407,10 +9426,9 @@ Process *schedule(Process *p, int calls) suspend_scheduler(esdp); #endif - { + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { erts_aint32_t aux_work; - int leader_update = ERTS_SCHEDULER_IS_DIRTY(esdp) ? 0 - : erts_thr_progress_update(esdp); + int leader_update = erts_thr_progress_update(esdp); aux_work = erts_atomic32_read_acqb(&esdp->ssi->aux_work); if (aux_work | leader_update | ERTS_SCHED_FAIR) { erts_smp_runq_unlock(rq); @@ -9423,8 +9441,7 @@ Process *schedule(Process *p, int calls) erts_smp_runq_lock(rq); } - ERTS_SMP_LC_ASSERT(ERTS_SCHEDULER_IS_DIRTY(esdp) - || !erts_thr_progress_is_blocking()); + ERTS_SMP_LC_ASSERT(!erts_thr_progress_is_blocking()); } ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); @@ -9483,7 +9500,10 @@ Process *schedule(Process *p, int calls) } #endif + (void) ERTS_RUNQ_FLGS_UNSET(rq, ERTS_RUNQ_FLG_EXEC); scheduler_wait(&fcalls, esdp, rq); + flags = ERTS_RUNQ_FLGS_SET_NOB(rq, ERTS_RUNQ_FLG_EXEC); + flags |= ERTS_RUNQ_FLG_EXEC; #ifdef ERTS_SMP non_empty_runq(rq); diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index 10c6fa4a67..7b19e76352 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.h @@ -170,8 +170,10 @@ extern int erts_sched_thread_suggested_stack_size; (((Uint32) 1) << (ERTS_RUNQ_FLG_BASE2 + 5)) #define ERTS_RUNQ_FLG_PROTECTED \ (((Uint32) 1) << (ERTS_RUNQ_FLG_BASE2 + 6)) +#define ERTS_RUNQ_FLG_EXEC \ + (((Uint32) 1) << (ERTS_RUNQ_FLG_BASE2 + 7)) -#define ERTS_RUNQ_FLG_MAX (ERTS_RUNQ_FLG_BASE2 + 7) +#define ERTS_RUNQ_FLG_MAX (ERTS_RUNQ_FLG_BASE2 + 8) #define ERTS_RUNQ_FLGS_MIGRATION_QMASKS \ (ERTS_RUNQ_FLGS_EMIGRATE_QMASK \ @@ -215,6 +217,9 @@ extern int erts_sched_thread_suggested_stack_size; #define ERTS_RUNQ_FLGS_SET(RQ, FLGS) \ ((Uint32) erts_smp_atomic32_read_bor_relb(&(RQ)->flags, \ (erts_aint32_t) (FLGS))) +#define ERTS_RUNQ_FLGS_SET_NOB(RQ, FLGS) \ + ((Uint32) erts_smp_atomic32_read_bor_nob(&(RQ)->flags, \ + (erts_aint32_t) (FLGS))) #define ERTS_RUNQ_FLGS_BSET(RQ, MSK, FLGS) \ ((Uint32) erts_smp_atomic32_read_bset_relb(&(RQ)->flags, \ (erts_aint32_t) (MSK), \ @@ -222,6 +227,9 @@ extern int erts_sched_thread_suggested_stack_size; #define ERTS_RUNQ_FLGS_UNSET(RQ, FLGS) \ ((Uint32) erts_smp_atomic32_read_band_relb(&(RQ)->flags, \ (erts_aint32_t) ~(FLGS))) +#define ERTS_RUNQ_FLGS_UNSET_NOB(RQ, FLGS) \ + ((Uint32) erts_smp_atomic32_read_band_nob(&(RQ)->flags, \ + (erts_aint32_t) ~(FLGS))) #define ERTS_RUNQ_FLGS_GET(RQ) \ ((Uint32) erts_smp_atomic32_read_acqb(&(RQ)->flags)) #define ERTS_RUNQ_FLGS_GET_NOB(RQ) \ @@ -467,7 +475,7 @@ struct ErtsRunQueue_ { int full_reds_history[ERTS_FULL_REDS_HISTORY_SIZE]; int out_of_work_count; erts_aint32_t max_len; - erts_aint32_t len; + erts_smp_atomic32_t len; int wakeup_other; int wakeup_other_reds; int halt_in_progress; @@ -607,7 +615,7 @@ typedef enum { typedef union { struct { ErtsDirtySchedulerType type: 1; - unsigned num: 31; + Uint num: sizeof(Uint)*8 - 1; } s; Uint no; } ErtsDirtySchedId; @@ -728,7 +736,19 @@ erts_smp_inc_runq_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi, int prio) ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); - len = erts_smp_atomic32_read_nob(&rqi->len); + len = erts_smp_atomic32_read_dirty(&rq->len); + +#ifdef ERTS_SMP + if (len == 0) + erts_non_empty_runq(rq); +#endif + len++; + if (rq->max_len < len) + rq->max_len = len; + ASSERT(len > 0); + erts_smp_atomic32_set_nob(&rq->len, len); + + len = erts_smp_atomic32_read_dirty(&rqi->len); ASSERT(len >= 0); if (len == 0) { ASSERT((erts_smp_atomic32_read_nob(&rq->flags) @@ -741,15 +761,6 @@ erts_smp_inc_runq_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi, int prio) rqi->max_len = len; erts_smp_atomic32_set_relb(&rqi->len, len); - -#ifdef ERTS_SMP - if (rq->len == 0) - erts_non_empty_runq(rq); -#endif - rq->len++; - if (rq->max_len < rq->len) - rq->max_len = len; - ASSERT(rq->len > 0); } ERTS_GLB_INLINE void @@ -759,7 +770,12 @@ erts_smp_dec_runq_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi, int prio) ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); - len = erts_smp_atomic32_read_nob(&rqi->len); + len = erts_smp_atomic32_read_dirty(&rq->len); + len--; + ASSERT(len >= 0); + erts_smp_atomic32_set_nob(&rq->len, len); + + len = erts_smp_atomic32_read_dirty(&rqi->len); len--; ASSERT(len >= 0); if (len == 0) { @@ -770,8 +786,6 @@ erts_smp_dec_runq_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi, int prio) } erts_smp_atomic32_set_relb(&rqi->len, len); - rq->len--; - ASSERT(rq->len >= 0); } ERTS_GLB_INLINE void @@ -781,7 +795,7 @@ erts_smp_reset_max_len(ErtsRunQueue *rq, ErtsRunQueueInfo *rqi) ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); - len = erts_smp_atomic32_read_nob(&rqi->len); + len = erts_smp_atomic32_read_dirty(&rqi->len); ASSERT(rqi->max_len >= len); rqi->max_len = len; } @@ -1678,7 +1692,7 @@ void erts_sched_notify_check_cpu_bind(void); Uint erts_active_schedulers(void); void erts_init_process(int, int, int); Eterm erts_process_status(Process *, ErtsProcLocks, Process *, Eterm); -Uint erts_run_queues_len(Uint *); +Uint erts_run_queues_len(Uint *, int, int); void erts_add_to_runq(Process *); Eterm erts_bound_schedulers_term(Process *c_p); Eterm erts_get_cpu_topology_term(Process *c_p, Eterm which); diff --git a/erts/emulator/test/alloc_SUITE_data/threads.c b/erts/emulator/test/alloc_SUITE_data/threads.c index a8a6a23695..2f5f841e3d 100644 --- a/erts/emulator/test/alloc_SUITE_data/threads.c +++ b/erts/emulator/test/alloc_SUITE_data/threads.c @@ -396,7 +396,7 @@ alloc_op(int t_no, Allctr_t *a, block *bp, int id, int clean_up) bp->p = (unsigned char *) ALLOC(a, bp->s); if(!bp->p) fail(t_no, "ALLOC(%lu) failed [id=%d])\n", bp->s, id); - memset((void *) bp->p, id, (size_t) bp->s); + memset((void *) bp->p, (unsigned char)id, (size_t) bp->s); } else { unsigned char *p = (unsigned char *) REALLOC(a, bp->p, bp->as[bp->i]); @@ -406,7 +406,7 @@ alloc_op(int t_no, Allctr_t *a, block *bp, int id, int clean_up) if(bp->s < bp->as[bp->i]) { CHECK_BLOCK_DATA(t_no, p, bp->s, id); - memset((void *) p, id, (size_t) bp->as[bp->i]); + memset((void *) p, (unsigned char)id, (size_t) bp->as[bp->i]); } else CHECK_BLOCK_DATA(t_no, p, bp->as[bp->i], id); diff --git a/erts/emulator/test/statistics_SUITE.erl b/erts/emulator/test/statistics_SUITE.erl index 56ecf4195a..53c9ba8715 100644 --- a/erts/emulator/test/statistics_SUITE.erl +++ b/erts/emulator/test/statistics_SUITE.erl @@ -32,7 +32,7 @@ run_queue_one/1, scheduler_wall_time/1, reductions/1, reductions_big/1, garbage_collection/1, io/1, - badarg/1]). + badarg/1, run_queues_lengths_active_tasks/1]). %% Internal exports. @@ -54,7 +54,8 @@ suite() -> [{ct_hooks,[ts_install_cth]}]. all() -> [{group, wall_clock}, {group, runtime}, reductions, reductions_big, {group, run_queue}, scheduler_wall_time, - garbage_collection, io, badarg]. + garbage_collection, io, badarg, + run_queues_lengths_active_tasks]. groups() -> [{wall_clock, [], @@ -409,3 +410,59 @@ badarg(Config) when is_list(Config) -> ?line case catch statistics(bad_atom) of {'EXIT', {badarg, _}} -> ok end. + +tok_loop() -> + tok_loop(). + +run_queues_lengths_active_tasks(Config) -> + TokLoops = lists:map(fun (_) -> + spawn_opt(fun () -> + tok_loop() + end, + [link, {priority, low}]) + end, + lists:seq(1,10)), + + TRQLs0 = statistics(total_run_queue_lengths), + TATs0 = statistics(total_active_tasks), + true = is_integer(TRQLs0), + true = is_integer(TATs0), + true = TRQLs0 >= 0, + true = TATs0 >= 11, + + NoScheds = erlang:system_info(schedulers), + RQLs0 = statistics(run_queue_lengths), + ATs0 = statistics(active_tasks), + NoScheds = length(RQLs0), + NoScheds = length(ATs0), + true = lists:sum(RQLs0) >= 0, + true = lists:sum(ATs0) >= 11, + + SO = erlang:system_flag(schedulers_online, 1), + + TRQLs1 = statistics(total_run_queue_lengths), + TATs1 = statistics(total_active_tasks), + true = TRQLs1 >= 10, + true = TATs1 >= 11, + NoScheds = erlang:system_info(schedulers), + + RQLs1 = statistics(run_queue_lengths), + ATs1 = statistics(active_tasks), + NoScheds = length(RQLs1), + NoScheds = length(ATs1), + TRQLs2 = lists:sum(RQLs1), + TATs2 = lists:sum(ATs1), + true = TRQLs2 >= 10, + true = TATs2 >= 11, + [TRQLs2|_] = RQLs1, + [TATs2|_] = ATs1, + + erlang:system_flag(schedulers_online, SO), + + lists:foreach(fun (P) -> + unlink(P), + exit(P, bang) + end, + TokLoops), + + ok. diff --git a/erts/epmd/src/epmd_srv.c b/erts/epmd/src/epmd_srv.c index 8c8d7304f2..5b58554590 100644 --- a/erts/epmd/src/epmd_srv.c +++ b/erts/epmd/src/epmd_srv.c @@ -700,9 +700,6 @@ static void do_request(g, fd, s, buf, bsize) put_int16(node->creation, wbuf+2); } - if (g->delay_write) /* Test of busy server */ - sleep(g->delay_write); - if (reply(g, fd, wbuf, 4) != 4) { dbg_tty_printf(g,1,"** failed to send ALIVE2_RESP for \"%s\"", diff --git a/erts/preloaded/ebin/erlang.beam b/erts/preloaded/ebin/erlang.beam Binary files differindex cd2e7f18a2..58516c0ff3 100644 --- a/erts/preloaded/ebin/erlang.beam +++ b/erts/preloaded/ebin/erlang.beam diff --git a/erts/preloaded/src/erlang.erl b/erts/preloaded/src/erlang.erl index 7280b43502..5fc6d14938 100644 --- a/erts/preloaded/src/erlang.erl +++ b/erts/preloaded/src/erlang.erl @@ -2205,7 +2205,9 @@ setelement(_Index, _Tuple1, _Value) -> spawn_opt(_Tuple) -> erlang:nif_error(undefined). --spec statistics(context_switches) -> {ContextSwitches,0} when +-spec statistics(active_tasks) -> [ActiveTasks] when + ActiveTasks :: non_neg_integer(); + (context_switches) -> {ContextSwitches,0} when ContextSwitches :: non_neg_integer(); (exact_reductions) -> {Total_Exact_Reductions, Exact_Reductions_Since_Last_Call} when @@ -2222,6 +2224,8 @@ spawn_opt(_Tuple) -> Total_Reductions :: non_neg_integer(), Reductions_Since_Last_Call :: non_neg_integer(); (run_queue) -> non_neg_integer(); + (run_queue_lengths) -> [RunQueueLenght] when + RunQueueLenght :: non_neg_integer(); (runtime) -> {Total_Run_Time, Time_Since_Last_Call} when Total_Run_Time :: non_neg_integer(), Time_Since_Last_Call :: non_neg_integer(); @@ -2229,6 +2233,10 @@ spawn_opt(_Tuple) -> SchedulerId :: pos_integer(), ActiveTime :: non_neg_integer(), TotalTime :: non_neg_integer(); + (total_active_tasks) -> ActiveTasks when + ActiveTasks :: non_neg_integer(); + (total_run_queue_lengths) -> TotalRunQueueLenghts when + TotalRunQueueLenghts :: non_neg_integer(); (wall_clock) -> {Total_Wallclock_Time, Wallclock_Time_Since_Last_Call} when Total_Wallclock_Time :: non_neg_integer(), |