From 0c1ceeb0a30609a105dd69e000d1c391e2d2927e Mon Sep 17 00:00:00 2001 From: Patrik Nyblom Date: Thu, 2 May 2013 18:49:45 +0200 Subject: Add system_monitor of long_schedule --- erts/emulator/beam/atom.names | 4 ++ erts/emulator/beam/beam_emu.c | 22 +++++++ erts/emulator/beam/erl_bif_trace.c | 23 ++++++- erts/emulator/beam/erl_port_task.c | 13 ++++ erts/emulator/beam/erl_process.c | 1 + erts/emulator/beam/erl_process.h | 1 + erts/emulator/beam/erl_trace.c | 128 +++++++++++++++++++++++++++++++++++++ erts/emulator/beam/erl_trace.h | 2 + erts/emulator/beam/global.h | 2 + erts/emulator/beam/utils.c | 18 ++++++ erts/emulator/test/trace_SUITE.erl | 69 ++++++++++++++++++++ 11 files changed, 281 insertions(+), 2 deletions(-) (limited to 'erts/emulator') diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names index ce60bb9bbc..80208d35e1 100644 --- a/erts/emulator/beam/atom.names +++ b/erts/emulator/beam/atom.names @@ -178,6 +178,7 @@ atom disable_trace atom disabled atom display_items atom dist +atom dist_cmd atom Div='/' atom div atom dlink @@ -312,6 +313,7 @@ atom load_cancelled atom load_failure atom local atom long_gc +atom long_schedule atom low atom Lt='<' atom machine @@ -430,6 +432,7 @@ atom port atom ports atom port_count atom port_limit +atom port_op atom print atom priority atom private @@ -441,6 +444,7 @@ atom process_display atom process_limit atom process_dump atom procs +atom proc_sig atom profile atom protected atom protection diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index 944ed6da81..5781009f58 100644 --- a/erts/emulator/beam/beam_emu.c +++ b/erts/emulator/beam/beam_emu.c @@ -924,6 +924,7 @@ extern int count_instructions; # define NOINLINE #endif + /* * The following functions are called directly by process_main(). * Don't inline them. @@ -1153,6 +1154,9 @@ void process_main(void) Eterm pt_arity; /* Used by do_put_tuple */ + Uint64 start_time = 0; /* Monitor long schedule */ + BeamInstr* start_time_i = NULL; + ERL_BITS_DECLARE_STATEP; /* Has to be last declaration */ @@ -1175,6 +1179,16 @@ void process_main(void) do_schedule: reds_used = REDS_IN(c_p) - FCALLS; do_schedule1: + + if (start_time != 0) { + Sint64 diff = erts_timestamp_millis() - start_time; + if (diff > 0 && (Uint) diff > erts_system_monitor_long_schedule) { + BeamInstr *inptr = find_function_from_pc(start_time_i); + BeamInstr *outptr = find_function_from_pc(c_p->i); + monitor_long_schedule_proc(c_p,inptr,outptr,(Uint) diff); + } + } + PROCESS_MAIN_CHK_LOCKS(c_p); ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p); #if HALFWORD_HEAP @@ -1183,11 +1197,18 @@ void process_main(void) ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p); c_p = schedule(c_p, reds_used); ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p); + start_time = 0; #ifdef DEBUG pid = c_p->common.id; /* Save for debugging purpouses */ #endif ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p); PROCESS_MAIN_CHK_LOCKS(c_p); + + if (erts_system_monitor_long_schedule != 0) { + start_time = erts_timestamp_millis(); + start_time_i = c_p->i; + } + reg = ERTS_PROC_GET_SCHDATA(c_p)->x_reg_array; freg = ERTS_PROC_GET_SCHDATA(c_p)->f_reg_array; #if !HEAP_ON_C_STACK @@ -6151,6 +6172,7 @@ apply_fun(Process* p, Eterm fun, Eterm args, Eterm* reg) } + static Eterm new_fun(Process* p, Eterm* reg, ErlFunEntry* fe, int num_free) { diff --git a/erts/emulator/beam/erl_bif_trace.c b/erts/emulator/beam/erl_bif_trace.c index 41c98bbffb..14a40dcd7e 100644 --- a/erts/emulator/beam/erl_bif_trace.c +++ b/erts/emulator/beam/erl_bif_trace.c @@ -2014,6 +2014,7 @@ void erts_system_monitor_clear(Process *c_p) { #endif erts_set_system_monitor(NIL); erts_system_monitor_long_gc = 0; + erts_system_monitor_long_schedule = 0; erts_system_monitor_large_heap = 0; erts_system_monitor_flags.busy_port = 0; erts_system_monitor_flags.busy_dist_port = 0; @@ -2038,12 +2039,17 @@ static Eterm system_monitor_get(Process *p) Uint hsz = 3 + (erts_system_monitor_flags.busy_dist_port ? 2 : 0) + (erts_system_monitor_flags.busy_port ? 2 : 0); Eterm long_gc = NIL; + Eterm long_schedule = NIL; Eterm large_heap = NIL; if (erts_system_monitor_long_gc != 0) { hsz += 2+3; (void) erts_bld_uint(NULL, &hsz, erts_system_monitor_long_gc); } + if (erts_system_monitor_long_schedule != 0) { + hsz += 2+3; + (void) erts_bld_uint(NULL, &hsz, erts_system_monitor_long_schedule); + } if (erts_system_monitor_large_heap != 0) { hsz += 2+3; (void) erts_bld_uint(NULL, &hsz, erts_system_monitor_large_heap); @@ -2053,6 +2059,10 @@ static Eterm system_monitor_get(Process *p) if (erts_system_monitor_long_gc != 0) { long_gc = erts_bld_uint(&hp, NULL, erts_system_monitor_long_gc); } + if (erts_system_monitor_long_schedule != 0) { + long_schedule = erts_bld_uint(&hp, NULL, + erts_system_monitor_long_schedule); + } if (erts_system_monitor_large_heap != 0) { large_heap = erts_bld_uint(&hp, NULL, erts_system_monitor_large_heap); } @@ -2061,6 +2071,10 @@ static Eterm system_monitor_get(Process *p) Eterm t = TUPLE2(hp, am_long_gc, long_gc); hp += 3; res = CONS(hp, t, res); hp += 2; } + if (long_schedule != NIL) { + Eterm t = TUPLE2(hp, am_long_schedule, long_schedule); hp += 3; + res = CONS(hp, t, res); hp += 2; + } if (large_heap != NIL) { Eterm t = TUPLE2(hp, am_large_heap, large_heap); hp += 3; res = CONS(hp, t, res); hp += 2; @@ -2115,7 +2129,7 @@ system_monitor(Process *p, Eterm monitor_pid, Eterm list) } if (is_not_list(list)) goto error; else { - Uint long_gc, large_heap; + Uint long_gc, long_schedule, large_heap; int busy_port, busy_dist_port; system_blocked = 1; @@ -2125,7 +2139,8 @@ system_monitor(Process *p, Eterm monitor_pid, Eterm list) if (!erts_pid2proc(p, ERTS_PROC_LOCK_MAIN, monitor_pid, 0)) goto error; - for (long_gc = 0, large_heap = 0, busy_port = 0, busy_dist_port = 0; + for (long_gc = 0, long_schedule = 0, large_heap = 0, + busy_port = 0, busy_dist_port = 0; is_list(list); list = CDR(list_val(list))) { Eterm t = CAR(list_val(list)); @@ -2135,6 +2150,9 @@ system_monitor(Process *p, Eterm monitor_pid, Eterm list) if (tp[1] == am_long_gc) { if (! term_to_Uint(tp[2], &long_gc)) goto error; if (long_gc < 1) long_gc = 1; + } else if (tp[1] == am_long_schedule) { + if (! term_to_Uint(tp[2], &long_schedule)) goto error; + if (long_schedule < 1) long_schedule = 1; } else if (tp[1] == am_large_heap) { if (! term_to_Uint(tp[2], &large_heap)) goto error; if (large_heap < 16384) large_heap = 16384; @@ -2150,6 +2168,7 @@ system_monitor(Process *p, Eterm monitor_pid, Eterm list) prev = system_monitor_get(p); erts_set_system_monitor(monitor_pid); erts_system_monitor_long_gc = long_gc; + erts_system_monitor_long_schedule = long_schedule; erts_system_monitor_large_heap = large_heap; erts_system_monitor_flags.busy_port = !!busy_port; erts_system_monitor_flags.busy_dist_port = !!busy_dist_port; diff --git a/erts/emulator/beam/erl_port_task.c b/erts/emulator/beam/erl_port_task.c index 0ed08bee01..9ba7704106 100644 --- a/erts/emulator/beam/erl_port_task.c +++ b/erts/emulator/beam/erl_port_task.c @@ -1578,6 +1578,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) int fpe_was_unmasked; erts_aint32_t state; int active; + Uint64 start_time = 0; ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(runq)); @@ -1639,6 +1640,10 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) reset_handle(ptp); + if (erts_system_monitor_long_schedule != 0) { + start_time = erts_timestamp_millis(); + } + ERTS_SMP_LC_ASSERT(erts_lc_is_port_locked(pp)); ERTS_SMP_CHK_NO_PROC_LOCKS; ASSERT(pp->drv_ptr); @@ -1707,6 +1712,14 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp) reds += erts_port_driver_callback_epilogue(pp, &state); + if (start_time != 0) { + Sint64 diff = erts_timestamp_millis() - start_time; + if (diff > 0 && (Uint) diff > erts_system_monitor_long_schedule) { + monitor_long_schedule_port(pp,ptp->type,(Uint) diff); + } + } + start_time = 0; + aborted_port_task: schedule_port_task_free(ptp); diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index 5d9c5af55f..9618a5555b 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -267,6 +267,7 @@ static Uint last_exact_reductions; Uint erts_default_process_flags; Eterm erts_system_monitor; Eterm erts_system_monitor_long_gc; +Uint erts_system_monitor_long_schedule; Eterm erts_system_monitor_large_heap; struct erts_system_monitor_flags_t erts_system_monitor_flags; diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index 3d3579fa7e..797f9e687b 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.h @@ -1002,6 +1002,7 @@ extern erts_smp_rwmtx_t erts_cpu_bind_rwmtx; */ extern Eterm erts_system_monitor; extern Uint erts_system_monitor_long_gc; +extern Uint erts_system_monitor_long_schedule; extern Uint erts_system_monitor_large_heap; struct erts_system_monitor_flags_t { unsigned int busy_port : 1; diff --git a/erts/emulator/beam/erl_trace.c b/erts/emulator/beam/erl_trace.c index 848877d43e..bb6ed44523 100644 --- a/erts/emulator/beam/erl_trace.c +++ b/erts/emulator/beam/erl_trace.c @@ -2268,7 +2268,134 @@ trace_gc(Process *p, Eterm what) #undef LOCAL_HEAP_SIZE } +void +monitor_long_schedule_proc(Process *p, BeamInstr *in_fp, BeamInstr *out_fp, Uint time) +{ + ErlHeapFragment *bp; + ErlOffHeap *off_heap; +#ifndef ERTS_SMP + Process *monitor_p; +#endif + Uint hsz; + Eterm *hp, list, in_mfa = am_undefined, out_mfa = am_undefined; + Eterm in_tpl, out_tpl, tmo_tpl, tmo, msg; + + +#ifndef ERTS_SMP + ASSERT(is_internal_pid(system_monitor)); + monitor_p = erts_proc_lookup(system_monitor); + if (!monitor_p || p == monitor_p) { + return; + } +#endif + /* + * Size: {monitor, pid, long_schedule, [{timeout, T}, {in, {M,F,A}},{out,{M,F,A}}]} -> + * 5 (top tuple of 4), (3 (elements) * 2 (cons)) + 3 (timeout tuple of 2) + size of Timeout + + * (2 * 3 (in/out tuple of 2)) + + * 0 (unknown) or 4 (MFA tuple of 3) + 0 (unknown) or 4 (MFA tuple of 3) + * = 20 + (in_fp != NULL) ? 4 : 0 + (out_fp != NULL) ? 4 : 0 + size of Timeout + */ + hsz = 20 + ((in_fp != NULL) ? 4 : 0) + ((out_fp != NULL) ? 4 : 0); + (void) erts_bld_uint(NULL, &hsz, time); + hp = ERTS_ALLOC_SYSMSG_HEAP(hsz, &bp, &off_heap, monitor_p); + tmo = erts_bld_uint(&hp, NULL, time); + if (in_fp != NULL) { + in_mfa = TUPLE3(hp,(Eterm) in_fp[0], (Eterm) in_fp[1], make_small(in_fp[2])); + hp +=4; + } + if (out_fp != NULL) { + out_mfa = TUPLE3(hp,(Eterm) out_fp[0], (Eterm) out_fp[1], make_small(out_fp[2])); + hp +=4; + } + tmo_tpl = TUPLE2(hp,am_timeout, tmo); + hp += 3; + in_tpl = TUPLE2(hp,am_in,in_mfa); + hp += 3; + out_tpl = TUPLE2(hp,am_out,out_mfa); + hp += 3; + list = CONS(hp,out_tpl,NIL); + hp += 2; + list = CONS(hp,in_tpl,list); + hp += 2; + list = CONS(hp,tmo_tpl,list); + hp += 2; + msg = TUPLE4(hp, am_monitor, p->common.id, am_long_schedule, list); + hp += 5; +#ifdef ERTS_SMP + enqueue_sys_msg(SYS_MSG_TYPE_SYSMON, p->common.id, NIL, msg, bp); +#else + erts_queue_message(monitor_p, NULL, bp, msg, NIL +#ifdef USE_VM_PROBES + , NIL +#endif + ); +#endif +} +void +monitor_long_schedule_port(Port *pp, ErtsPortTaskType type, Uint time) +{ + ErlHeapFragment *bp; + ErlOffHeap *off_heap; +#ifndef ERTS_SMP + Process *monitor_p; +#endif + Uint hsz; + Eterm *hp, list, op; + Eterm op_tpl, tmo_tpl, tmo, msg; + + +#ifndef ERTS_SMP + ASSERT(is_internal_pid(system_monitor)); + monitor_p = erts_proc_lookup(system_monitor); + if (!monitor_p) { + return; + } +#endif + /* + * Size: {monitor, port, long_schedule, [{timeout, T}, {op, Operation}]} -> + * 5 (top tuple of 4), (2 (elements) * 2 (cons)) + 3 (timeout tuple of 2) + * + size of Timeout + 3 (op tuple of 2 atoms) + * = 15 + size of Timeout + */ + hsz = 15; + (void) erts_bld_uint(NULL, &hsz, time); + + hp = ERTS_ALLOC_SYSMSG_HEAP(hsz, &bp, &off_heap, monitor_p); + switch (type) { + case ERTS_PORT_TASK_PROC_SIG: op = am_proc_sig; break; + case ERTS_PORT_TASK_TIMEOUT: op = am_timeout; break; + case ERTS_PORT_TASK_INPUT: op = am_input; break; + case ERTS_PORT_TASK_OUTPUT: op = am_output; break; + case ERTS_PORT_TASK_EVENT: op = am_event; break; + case ERTS_PORT_TASK_DIST_CMD: op = am_dist_cmd; break; + default: op = am_undefined; break; + } + + tmo = erts_bld_uint(&hp, NULL, time); + + op_tpl = TUPLE2(hp,am_port_op,op); + hp += 3; + + tmo_tpl = TUPLE2(hp,am_timeout, tmo); + hp += 3; + + list = CONS(hp,op_tpl,NIL); + hp += 2; + list = CONS(hp,tmo_tpl,list); + hp += 2; + msg = TUPLE4(hp, am_monitor, pp->common.id, am_long_schedule, list); + hp += 5; +#ifdef ERTS_SMP + enqueue_sys_msg(SYS_MSG_TYPE_SYSMON, pp->common.id, NIL, msg, bp); +#else + erts_queue_message(monitor_p, NULL, bp, msg, NIL +#ifdef USE_VM_PROBES + , NIL +#endif + ); +#endif +} void monitor_long_gc(Process *p, Uint time) { @@ -3011,6 +3138,7 @@ sys_msg_disp_failure(ErtsSysMsgQ *smqp, Eterm receiver) case SYS_MSG_TYPE_SYSMON: if (receiver == NIL && !erts_system_monitor_long_gc + && !erts_system_monitor_long_schedule && !erts_system_monitor_large_heap && !erts_system_monitor_flags.busy_port && !erts_system_monitor_flags.busy_dist_port) diff --git a/erts/emulator/beam/erl_trace.h b/erts/emulator/beam/erl_trace.h index 50fb27aab0..54d3aafdda 100644 --- a/erts/emulator/beam/erl_trace.h +++ b/erts/emulator/beam/erl_trace.h @@ -83,6 +83,8 @@ void erts_system_profile_setup_active_schedulers(void); /* system_monitor */ void monitor_long_gc(Process *p, Uint time); +void monitor_long_schedule_proc(Process *p, BeamInstr *in_i, BeamInstr *out_i, Uint time); +void monitor_long_schedule_port(Port *pp, ErtsPortTaskType type, Uint time); void monitor_large_heap(Process *p); void monitor_generic(Process *p, Eterm type, Eterm spec); Uint erts_trace_flag2bit(Eterm flag); diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h index 012c1c7e6a..12826cf428 100755 --- a/erts/emulator/beam/global.h +++ b/erts/emulator/beam/global.h @@ -746,6 +746,8 @@ void erl_drv_thr_init(void); /* utils.c */ void erts_cleanup_offheap(ErlOffHeap *offheap); +Uint64 erts_timestamp_millis(void); + Export* erts_find_function(Eterm, Eterm, unsigned int, ErtsCodeIndex); Eterm store_external_or_ref_in_proc_(Process *, Eterm); diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c index d5d97d748a..0a833f7e66 100644 --- a/erts/emulator/beam/utils.c +++ b/erts/emulator/beam/utils.c @@ -3722,6 +3722,24 @@ erts_smp_ensure_later_interval_acqb(erts_interval_t *icp, Uint64 ic) } +/* + * A millisecond timestamp without time correction where there's no hrtime + * - for tracing on "long" things... + */ +Uint64 erts_timestamp_millis(void) +{ +#ifdef HAVE_GETHRTIME + return (Uint64) (sys_gethrtime() / 1000000); +#else + Uint64 res; + SysTimeval tv; + sys_gettimeofday(&tv); + res = (Uint64) tv.tv_sec*1000000; + res += (Uint64) tv.tv_usec; + return (res / 1000); +#endif +} + #ifdef DEBUG /* * Handy functions when using a debugger - don't use in the code! diff --git a/erts/emulator/test/trace_SUITE.erl b/erts/emulator/test/trace_SUITE.erl index 221b65309a..caa58ae281 100644 --- a/erts/emulator/test/trace_SUITE.erl +++ b/erts/emulator/test/trace_SUITE.erl @@ -34,6 +34,7 @@ system_monitor_args/1, more_system_monitor_args/1, system_monitor_long_gc_1/1, system_monitor_long_gc_2/1, system_monitor_large_heap_1/1, system_monitor_large_heap_2/1, + system_monitor_long_schedule/1, bad_flag/1, trace_delivered/1]). -include_lib("test_server/include/test_server.hrl"). @@ -52,6 +53,7 @@ all() -> set_on_first_spawn, system_monitor_args, more_system_monitor_args, system_monitor_long_gc_1, system_monitor_long_gc_2, system_monitor_large_heap_1, + system_monitor_long_schedule, system_monitor_large_heap_2, bad_flag, trace_delivered]. groups() -> @@ -508,6 +510,65 @@ try_l(Val) -> ?line {Self,Comb1} = erlang:system_monitor(undefined), ?line [{large_heap,Val},{long_gc,Arbitrary2}] = lists:sort(Comb1). +monitor_sys(Parent) -> + receive + {monitor,Pid,long_schedule,Data} when is_pid(Pid) -> + io:format("Long schedule of ~w: ~w~n",[Pid,Data]), + Parent ! {Pid,Data}, + monitor_sys(Parent); + {monitor,Port,long_schedule,Data} when is_port(Port) -> + {name,Name} = erlang:port_info(Port,name), + io:format("Long schedule of ~w (~p): ~w~n",[Port,Name,Data]), + Parent ! {Port,Data}, + monitor_sys(Parent); + Other -> + erlang:display(Other) + end. + +start_monitor() -> + Parent = self(), + Mpid = spawn_link(fun() -> monitor_sys(Parent) end), + erlang:system_monitor(Mpid,[{long_schedule,100}]), + erlang:yield(), % Need to be rescheduled for the trace to take + ok. + +system_monitor_long_schedule(suite) -> + []; +system_monitor_long_schedule(doc) -> + ["Tests erlang:system_monitor(Pid, [{long_schedule,Time}])"]; +system_monitor_long_schedule(Config) when is_list(Config) -> + Path = ?config(data_dir, Config), + erl_ddll:start(), + case (catch load_driver(Path, slow_drv)) of + ok -> + do_system_monitor_long_schedule(); + _Error -> + {skip, "Unable to load slow_drv (windows or no usleep()?)"} + end. +do_system_monitor_long_schedule() -> + start_monitor(), + Port = open_port({spawn_driver,slow_drv}, []), + "ok" = erlang:port_control(Port,0,[]), + Self = self(), + receive + {Self,L} when is_list(L) -> + ok + after 1000 -> + ?t:fail(no_trace_of_pid) + end, + "ok" = erlang:port_control(Port,1,[]), + "ok" = erlang:port_control(Port,2,[]), + receive + {Port,LL} when is_list(LL) -> + ok + after 1000 -> + ?t:fail(no_trace_of_port) + end, + port_close(Port), + erlang:system_monitor(undefined), + ok. + + -define(LONG_GC_SLEEP, 670). system_monitor_long_gc_1(suite) -> @@ -1521,3 +1582,11 @@ issue_non_empty_runq_warning(DeadLine, RQLen) -> " Processes info: ~p~n", [DeadLine div 1000, RQLen, self(), PIs]), receive after 1000 -> ok end. + +load_driver(Dir, Driver) -> + case erl_ddll:load_driver(Dir, Driver) of + ok -> ok; + {error, Error} = Res -> + io:format("~s\n", [erl_ddll:format_error(Error)]), + Res + end. -- cgit v1.2.3