aboutsummaryrefslogtreecommitdiffstats
path: root/erts/emulator
diff options
context:
space:
mode:
authorRickard Green <rickard@erlang.org>2012-07-25 03:19:10 +0200
committerRickard Green <rickard@erlang.org>2012-07-31 16:52:40 +0200
commit8457bb3aeb335733d22ab6d517fe173cd90b4f55 (patch)
treebaa5cccaf65c109d3bd13c6211937d79192c58d9 /erts/emulator
parent2e5c1dbb32cfe4be0a782216953b4630338edabc (diff)
downloadotp-8457bb3aeb335733d22ab6d517fe173cd90b4f55.tar.gz
otp-8457bb3aeb335733d22ab6d517fe173cd90b4f55.tar.bz2
otp-8457bb3aeb335733d22ab6d517fe173cd90b4f55.zip
Implement delayed aux work wake up
By using a delayed aux work wake up approach, a memory barrier can be omitted in the delayed dealloc enqueue operation. The amount of operations, on the potentially contended, wake up structure is also reduced.
Diffstat (limited to 'erts/emulator')
-rw-r--r--erts/emulator/beam/erl_process.c92
-rw-r--r--erts/emulator/beam/erl_process.h38
2 files changed, 111 insertions, 19 deletions
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 854360f641..df58d1ba18 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -360,6 +360,7 @@ dbg_chk_aux_work_val(erts_aint32_t value)
valid |= ERTS_SSI_AUX_WORK_ASYNC_READY_CLEAN;
#endif
#ifdef ERTS_SMP
+ valid |= ERTS_SSI_AUX_WORK_DELAYED_AW_WAKEUP;
valid |= ERTS_SSI_AUX_WORK_MISC_THR_PRGR;
valid |= ERTS_SSI_AUX_WORK_DD;
valid |= ERTS_SSI_AUX_WORK_DD_THR_PRGR;
@@ -937,6 +938,45 @@ haw_thr_prgr_current_check_progress(ErtsAuxWorkData *awdp)
}
}
+static ERTS_INLINE erts_aint32_t
+handle_delayed_aux_work_wakeup(ErtsAuxWorkData *awdp, erts_aint32_t aux_work)
+{
+ int jix, max_jix;
+ unset_aux_work_flags(awdp->ssi, ERTS_SSI_AUX_WORK_DELAYED_AW_WAKEUP);
+
+ ERTS_THR_MEMORY_BARRIER;
+
+ max_jix = awdp->delayed_wakeup.jix;
+ awdp->delayed_wakeup.jix = -1;
+ for (jix = 0; jix <= max_jix; jix++) {
+ int sched = awdp->delayed_wakeup.job[jix].sched;
+ erts_aint32_t aux_work = awdp->delayed_wakeup.job[jix].aux_work;
+
+ ASSERT(awdp->delayed_wakeup.sched2jix[sched] == jix);
+ awdp->delayed_wakeup.sched2jix[sched] = -1;
+ set_aux_work_flags_wakeup_nob(ERTS_SCHED_SLEEP_INFO_IX(sched-1),
+ aux_work);
+ }
+ return aux_work & ~ERTS_SSI_AUX_WORK_DELAYED_AW_WAKEUP;
+}
+
+static ERTS_INLINE void
+schedule_aux_work_wakeup(ErtsAuxWorkData *awdp, int sched, erts_aint32_t aux_work)
+{
+ int jix = awdp->delayed_wakeup.sched2jix[sched];
+ if (jix >= 0) {
+ ASSERT(awdp->delayed_wakeup.job[jix].sched == sched);
+ awdp->delayed_wakeup.job[jix].aux_work |= aux_work;
+ }
+ else {
+ jix = ++awdp->delayed_wakeup.jix;
+ awdp->delayed_wakeup.sched2jix[sched] = jix;
+ awdp->delayed_wakeup.job[jix].sched = sched;
+ awdp->delayed_wakeup.job[jix].aux_work = aux_work;
+ }
+ set_aux_work_flags_wakeup_nob(awdp->ssi, ERTS_SSI_AUX_WORK_DELAYED_AW_WAKEUP);
+}
+
#endif
typedef struct erts_misc_aux_work_t_ erts_misc_aux_work_t;
@@ -1193,8 +1233,14 @@ handle_fix_alloc(ErtsAuxWorkData *awdp, erts_aint32_t aux_work)
void
erts_alloc_notify_delayed_dealloc(int ix)
{
- set_aux_work_flags_wakeup_nob(ERTS_SCHED_SLEEP_INFO_IX(ix-1),
- ERTS_SSI_AUX_WORK_DD);
+ ErtsSchedulerData *esdp = erts_get_scheduler_data();
+ if (esdp)
+ schedule_aux_work_wakeup(&esdp->aux_work_data,
+ ix,
+ ERTS_SSI_AUX_WORK_DD);
+ else
+ set_aux_work_flags_wakeup_relb(ERTS_SCHED_SLEEP_INFO_IX(ix-1),
+ ERTS_SSI_AUX_WORK_DD);
}
static ERTS_INLINE erts_aint32_t
@@ -1492,6 +1538,8 @@ handle_aux_work(ErtsAuxWorkData *awdp, erts_aint32_t orig_aux_work, int waiting)
* eachother. Most frequent first.
*/
#ifdef ERTS_SMP
+ HANDLE_AUX_WORK(ERTS_SSI_AUX_WORK_DELAYED_AW_WAKEUP,
+ handle_delayed_aux_work_wakeup);
HANDLE_AUX_WORK(ERTS_SSI_AUX_WORK_DD,
handle_delayed_dealloc);
/* DD must be before DD_THR_PRGR */
@@ -1970,7 +2018,7 @@ thr_prgr_fin_wait(void *vssi)
| ERTS_SSI_FLG_TSE_SLEEPING));
}
-static void init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp);
+static void init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp, char *dawwp);
static void *
aux_thread(void *unused)
@@ -1990,7 +2038,7 @@ aux_thread(void *unused)
callbacks.finalize_wait = thr_prgr_fin_wait;
erts_thr_progress_register_managed_thread(NULL, &callbacks, 1);
- init_aux_work_data(awdp, NULL);
+ init_aux_work_data(awdp, NULL, NULL);
awdp->ssi = ssi;
sched_prep_spin_wait(ssi);
@@ -3606,7 +3654,7 @@ erts_sched_set_wakeup_limit(char *str)
}
static void
-init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp)
+init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp, char *dawwp)
{
awdp->sched_id = esdp ? (int) esdp->no : 0;
awdp->esdp = esdp;
@@ -3624,12 +3672,32 @@ init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp)
#endif
awdp->async_ready.queue = NULL;
#endif
+#ifdef ERTS_SMP
+ if (!dawwp) {
+ awdp->delayed_wakeup.job = NULL;
+ awdp->delayed_wakeup.sched2jix = NULL;
+ awdp->delayed_wakeup.jix = -1;
+ }
+ else {
+ int i;
+ awdp->delayed_wakeup.job = (ErtsDelayedAuxWorkWakeupJob *) dawwp;
+ dawwp += sizeof(ErtsDelayedAuxWorkWakeupJob)*(erts_no_schedulers+1);
+ awdp->delayed_wakeup.sched2jix = (int *) dawwp;
+ awdp->delayed_wakeup.jix = -1;
+ for (i = 0; i <= erts_no_schedulers; i++)
+ awdp->delayed_wakeup.sched2jix[i] = -1;
+ }
+#endif
}
void
erts_init_scheduling(int no_schedulers, int no_schedulers_online)
{
int ix, n, no_ssi;
+ char *daww_ptr;
+#ifdef ERTS_SMP
+ size_t daww_sz;
+#endif
init_misc_op_list_alloc();
@@ -3758,6 +3826,15 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online)
/* Create and initialize scheduler specific data */
+#ifdef ERTS_SMP
+ daww_sz = ERTS_ALC_CACHE_LINE_ALIGN_SIZE((sizeof(ErtsDelayedAuxWorkWakeupJob)
+ + sizeof(int))*(n+1));
+ daww_ptr = erts_alloc_permanent_cache_aligned(ERTS_ALC_T_SCHDLR_DATA,
+ daww_sz*n);
+#else
+ daww_ptr = NULL;
+#endif
+
erts_aligned_scheduler_data =
erts_alloc_permanent_cache_aligned(ERTS_ALC_T_SCHDLR_DATA,
n*sizeof(ErtsAlignedSchedulerData));
@@ -3792,7 +3869,10 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online)
esdp->run_queue = ERTS_RUNQ_IX(ix);
esdp->run_queue->scheduler = esdp;
- init_aux_work_data(&esdp->aux_work_data, esdp);
+ init_aux_work_data(&esdp->aux_work_data, esdp, daww_ptr);
+#ifdef ERTS_SMP
+ daww_ptr += daww_sz;
+#endif
init_sched_wall_time(&esdp->sched_wall_time);
}
diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h
index cff0783bc4..33eda4b1ec 100644
--- a/erts/emulator/beam/erl_process.h
+++ b/erts/emulator/beam/erl_process.h
@@ -253,18 +253,19 @@ typedef enum {
* eachother. Most frequent - lowest bit number.
*/
-#define ERTS_SSI_AUX_WORK_DD (((erts_aint32_t) 1) << 0)
-#define ERTS_SSI_AUX_WORK_DD_THR_PRGR (((erts_aint32_t) 1) << 1)
-#define ERTS_SSI_AUX_WORK_FIX_ALLOC_DEALLOC (((erts_aint32_t) 1) << 2)
-#define ERTS_SSI_AUX_WORK_FIX_ALLOC_LOWER_LIM (((erts_aint32_t) 1) << 3)
-#define ERTS_SSI_AUX_WORK_ASYNC_READY (((erts_aint32_t) 1) << 4)
-#define ERTS_SSI_AUX_WORK_ASYNC_READY_CLEAN (((erts_aint32_t) 1) << 5)
-#define ERTS_SSI_AUX_WORK_MISC_THR_PRGR (((erts_aint32_t) 1) << 6)
-#define ERTS_SSI_AUX_WORK_MISC (((erts_aint32_t) 1) << 7)
-#define ERTS_SSI_AUX_WORK_CHECK_CHILDREN (((erts_aint32_t) 1) << 8)
-#define ERTS_SSI_AUX_WORK_SET_TMO (((erts_aint32_t) 1) << 9)
-#define ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK (((erts_aint32_t) 1) << 10)
-#define ERTS_SSI_AUX_WORK_REAP_PORTS (((erts_aint32_t) 1) << 11)
+#define ERTS_SSI_AUX_WORK_DELAYED_AW_WAKEUP (((erts_aint32_t) 1) << 0)
+#define ERTS_SSI_AUX_WORK_DD (((erts_aint32_t) 1) << 1)
+#define ERTS_SSI_AUX_WORK_DD_THR_PRGR (((erts_aint32_t) 1) << 2)
+#define ERTS_SSI_AUX_WORK_FIX_ALLOC_DEALLOC (((erts_aint32_t) 1) << 3)
+#define ERTS_SSI_AUX_WORK_FIX_ALLOC_LOWER_LIM (((erts_aint32_t) 1) << 4)
+#define ERTS_SSI_AUX_WORK_ASYNC_READY (((erts_aint32_t) 1) << 5)
+#define ERTS_SSI_AUX_WORK_ASYNC_READY_CLEAN (((erts_aint32_t) 1) << 6)
+#define ERTS_SSI_AUX_WORK_MISC_THR_PRGR (((erts_aint32_t) 1) << 7)
+#define ERTS_SSI_AUX_WORK_MISC (((erts_aint32_t) 1) << 8)
+#define ERTS_SSI_AUX_WORK_CHECK_CHILDREN (((erts_aint32_t) 1) << 9)
+#define ERTS_SSI_AUX_WORK_SET_TMO (((erts_aint32_t) 1) << 10)
+#define ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK (((erts_aint32_t) 1) << 11)
+#define ERTS_SSI_AUX_WORK_REAP_PORTS (((erts_aint32_t) 1) << 12)
typedef struct ErtsSchedulerSleepInfo_ ErtsSchedulerSleepInfo;
@@ -403,6 +404,11 @@ typedef struct {
} ErtsSchedWallTime;
typedef struct {
+ int sched;
+ erts_aint32_t aux_work;
+} ErtsDelayedAuxWorkWakeupJob;
+
+typedef struct {
int sched_id;
ErtsSchedulerData *esdp;
ErtsSchedulerSleepInfo *ssi;
@@ -431,6 +437,13 @@ typedef struct {
void *queue;
} async_ready;
#endif
+#ifdef ERTS_SMP
+ struct {
+ int *sched2jix;
+ int jix;
+ ErtsDelayedAuxWorkWakeupJob *job;
+ } delayed_wakeup;
+#endif
} ErtsAuxWorkData;
struct ErtsSchedulerData_ {
@@ -464,7 +477,6 @@ struct ErtsSchedulerData_ {
int virtual_reds;
int cpu_id; /* >= 0 when bound */
ErtsAuxWorkData aux_work_data;
-
ErtsAtomCacheMap atom_cache_map;
ErtsSchedAllocData alloc_data;