erts: Add microstate accounting

Microstate accounting is a way to track which state the
different threads within ERTS are in. The main usage area
is to pin point performance bottlenecks by checking which
states the threads are in and then from there figuring out
why and where to optimize.

Since checking whether microstate accounting is on or off is
relatively expensive if done in a short loop only a few of the
states are enabled by default and more states can be enabled
through configure.

I've done some benchmarking and the overhead with it turned off
is not noticible and with it on it is a fraction of a percent.
If you enable the extra states, depending on the benchmark,
the ovehead when turned off is about 1% and when turned on
somewhere inbetween 5-15%.

OTP-12345

1 files changed, 46 insertions, 2 deletions

diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 3902632ef8..8e1ebe795c 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -941,6 +941,12 @@ sched_wall_time_change(ErtsSchedulerData *esdp, int working)
 	    }
 	}
     }
+    if (!working) {
+        ERTS_MSACC_SET_STATE_M_X(ERTS_MSACC_STATE_BUSY_WAIT);
+    } else {
+        ERTS_MSACC_SET_STATE_M_X(ERTS_MSACC_STATE_OTHER);
+    }
+
 }
 
 typedef struct {
@@ -1696,15 +1702,17 @@ handle_delayed_dealloc(ErtsAuxWorkData *awdp, erts_aint32_t aux_work, int waitin
     int need_thr_progress = 0;
     ErtsThrPrgrVal wakeup = ERTS_THR_PRGR_INVALID;
     int more_work = 0;
-
+    ERTS_MSACC_PUSH_STATE_M_X();
 #ifdef ERTS_DIRTY_SCHEDULERS
     ASSERT(!awdp->esdp || !ERTS_SCHEDULER_IS_DIRTY(awdp->esdp));
 #endif
     unset_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_DD);
+    ERTS_MSACC_SET_STATE_CACHED_M_X(ERTS_MSACC_STATE_ALLOC);
     erts_alloc_scheduler_handle_delayed_dealloc((void *) awdp->esdp,
 						&need_thr_progress,
 						&wakeup,
 						&more_work);
+    ERTS_MSACC_POP_STATE_M_X();
     if (more_work) {
 	if (set_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_DD)
 	    & ERTS_SSI_AUX_WORK_DD_THR_PRGR) {
@@ -2176,12 +2184,15 @@ handle_aux_work(ErtsAuxWorkData *awdp, erts_aint32_t orig_aux_work, int waiting)
 	ERTS_DBG_CHK_AUX_WORK_VAL(aux_work); \
 	if (!(aux_work & ~ignore)) { \
 	    ERTS_DBG_CHK_AUX_WORK_VAL(aux_work); \
+            ERTS_MSACC_UPDATE_CACHE();       \
+            ERTS_MSACC_POP_STATE_M();              \
 	    return aux_work; \
 	} \
     }
 
     erts_aint32_t aux_work = orig_aux_work;
     erts_aint32_t ignore = 0;
+    ERTS_MSACC_PUSH_AND_SET_STATE_M(ERTS_MSACC_STATE_AUX);
 
     ASSERT(!awdp->esdp || !ERTS_SCHEDULER_IS_DIRTY(awdp->esdp));
 #ifdef ERTS_SMP
@@ -2272,6 +2283,8 @@ handle_aux_work(ErtsAuxWorkData *awdp, erts_aint32_t orig_aux_work, int waiting)
 	haw_thr_prgr_current_check_progress(awdp);
 #endif
 
+    ERTS_MSACC_UPDATE_CACHE();
+    ERTS_MSACC_POP_STATE_M();
     return aux_work;
 
 #undef HANDLE_AUX_WORK
@@ -2779,6 +2792,8 @@ aux_thread(void *unused)
 
     ssi->event = erts_tse_fetch();
 
+    erts_msacc_init_thread("aux", 1, 1);
+
     callbacks.arg = (void *) ssi;
     callbacks.wakeup = thr_prgr_wakeup;
     callbacks.prepare_wait = thr_prgr_prep_wait;
@@ -2789,6 +2804,7 @@ aux_thread(void *unused)
     init_aux_work_data(awdp, NULL, NULL);
     awdp->ssi = ssi;
 
+
     sched_prep_spin_wait(ssi);
 
     while (1) {
@@ -2842,6 +2858,9 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 #ifdef ERTS_SMP
     int thr_prgr_active = 1;
     erts_aint32_t flgs;
+#endif
+    ERTS_MSACC_PUSH_STATE_M();
+#ifdef ERTS_SMP
 
     ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq));
 
@@ -2898,6 +2917,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 		    sched_wall_time_change(esdp, 1);
 		}
 		aux_work = handle_aux_work(&esdp->aux_work_data, aux_work, 1);
+                ERTS_MSACC_UPDATE_CACHE();
 		if (aux_work && erts_thr_progress_update(esdp))
 		    erts_thr_progress_leader_update(esdp);
 	    }
@@ -2959,7 +2979,9 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 								     - 1) + 1;
 				} else
 				    timeout = -1;
+                                ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_SLEEP);
 				res = erts_tse_twait(ssi->event, timeout);
+                                ERTS_MSACC_POP_STATE_M();
 				current_time = ERTS_SCHEDULER_IS_DIRTY(esdp) ? 0 :
 				    erts_get_monotonic_time(esdp);
 			    } while (res == EINTR);
@@ -3028,9 +3050,13 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 	    if (working)
 		sched_wall_time_change(esdp, working = 0);
 
+	    ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_CHECK_IO);
+
 	    ASSERT(!erts_port_task_have_outstanding_io_tasks());
 	    erl_sys_schedule(1); /* Might give us something to do */
 
+	    ERTS_MSACC_POP_STATE_M();
+
 	    if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) {
 		current_time = erts_get_monotonic_time(esdp);
 		if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref))
@@ -3051,6 +3077,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 		    erts_thr_progress_active(esdp, thr_prgr_active = 1);
 #endif
 		aux_work = handle_aux_work(&esdp->aux_work_data, aux_work, 1);
+                ERTS_MSACC_UPDATE_CACHE();
 #ifdef ERTS_SMP
 		if (aux_work && erts_thr_progress_update(esdp))
 		    erts_thr_progress_leader_update(esdp);
@@ -3148,8 +3175,12 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 
 	ASSERT(!erts_port_task_have_outstanding_io_tasks());
 
+	ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_CHECK_IO);
+
 	erl_sys_schedule(0);
 
+	ERTS_MSACC_POP_STATE_M();
+
 	if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) {
 	    ErtsMonotonicTime current_time = erts_get_monotonic_time(esdp);
 	    if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref))
@@ -7926,6 +7957,8 @@ sched_thread_func(void *vesdp)
     callbacks.wait = thr_prgr_wait;
     callbacks.finalize_wait = thr_prgr_fin_wait;
 
+    erts_msacc_init_thread("scheduler", no, 1);
+
     erts_thr_progress_register_managed_thread(esdp, &callbacks, 0);
     erts_alloc_register_scheduler(vesdp);
 #endif
@@ -9204,6 +9237,8 @@ Process *schedule(Process *p, int calls)
     Uint32 flags;
     erts_aint32_t state = 0; /* Supress warning... */
 
+    ERTS_MSACC_DECLARE_CACHE();
+
 #ifdef USE_VM_PROBES
     if (p != NULL && DTRACE_ENABLED(process_unscheduled)) {
         DTRACE_CHARBUF(process_buf, DTRACE_TERM_BUF_SIZE);
@@ -9308,6 +9343,8 @@ Process *schedule(Process *p, int calls)
 
 	erts_smp_proc_unlock(p, ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_STATUS);
 
+        ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_OTHER);
+
 	if (state & ERTS_PSFLG_FREE) {
 #ifdef ERTS_SMP
 	    ASSERT(esdp->free_process == p);
@@ -9468,7 +9505,7 @@ Process *schedule(Process *p, int calls)
 	    scheduler_wait(&fcalls, esdp, rq);
 	    flags = ERTS_RUNQ_FLGS_SET_NOB(rq, ERTS_RUNQ_FLG_EXEC);
 	    flags |= ERTS_RUNQ_FLG_EXEC;
-
+            ERTS_MSACC_UPDATE_CACHE();
 #ifdef ERTS_SMP
 	    non_empty_runq(rq);
 #endif
@@ -9483,6 +9520,8 @@ Process *schedule(Process *p, int calls)
 	     * Schedule system-level activities.
 	     */
 
+	    ERTS_MSACC_PUSH_STATE_CACHED_M();
+
 	    erts_smp_atomic32_set_relb(&function_calls, 0);
 	    fcalls = 0;
 
@@ -9490,7 +9529,9 @@ Process *schedule(Process *p, int calls)
 	    erts_sys_schedule_interrupt(0);
 #endif
 	    erts_smp_runq_unlock(rq);
+	    ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_CHECK_IO);
 	    erl_sys_schedule(1);
+	    ERTS_MSACC_POP_STATE_M();
 
 	    current_time = erts_get_monotonic_time(esdp);
 	    if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref))
@@ -9568,9 +9609,12 @@ Process *schedule(Process *p, int calls)
 	    case 0:			/* No process at all */
 	    default:
 		ASSERT(qmask == 0);
+                ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_OTHER);
 		goto check_activities_to_run;
 	    }
 
+            ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_EMULATOR);
+
 	    BM_START_TIMER(system);
 
 	    /*