aboutsummaryrefslogtreecommitdiffstats
path: root/erts/emulator/beam/erl_process.c
diff options
context:
space:
mode:
authorRickard Green <[email protected]>2010-09-15 22:14:51 +0200
committerRickard Green <[email protected]>2011-11-13 20:39:30 +0100
commita67e91e658bdbba24fcc3c79b06fdf10ff830bc9 (patch)
tree07f9e6b1fd715d516d2571521307fe1b9d7c3948 /erts/emulator/beam/erl_process.c
parent55358c54778ead444e51f565d00175ba887ef182 (diff)
downloadotp-a67e91e658bdbba24fcc3c79b06fdf10ff830bc9.tar.gz
otp-a67e91e658bdbba24fcc3c79b06fdf10ff830bc9.tar.bz2
otp-a67e91e658bdbba24fcc3c79b06fdf10ff830bc9.zip
Optimize memory allocation
A number of memory allocation optimizations have been implemented. Most optimizations reduce contention caused by synchronization between threads during allocation and deallocation of memory. Most notably: * Synchronization of memory management in scheduler specific allocator instances has been rewritten to use lock-free synchronization. * Synchronization of memory management in scheduler specific pre-allocators has been rewritten to use lock-free synchronization. * The 'mseg_alloc' memory segment allocator now use scheduler specific instances instead of one instance. Apart from reducing contention this also ensures that memory allocators always create memory segments on the local NUMA node on a NUMA system.
Diffstat (limited to 'erts/emulator/beam/erl_process.c')
-rw-r--r--erts/emulator/beam/erl_process.c1157
1 files changed, 881 insertions, 276 deletions
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 9574435415..4dda17f559 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -39,6 +39,7 @@
#include "erl_binary.h"
#include "beam_bp.h"
#include "erl_cpu_topology.h"
+#include "erl_thr_progress.h"
#define ERTS_RUNQ_CHECK_BALANCE_REDS_PER_SCHED (2000*CONTEXT_REDS)
#define ERTS_RUNQ_CALL_CHECK_BALANCE_REDS \
@@ -127,6 +128,8 @@ ErtsLcPSDLocks erts_psd_required_locks[ERTS_PSD_SIZE];
int erts_disable_proc_not_running_opt;
+static ErtsAuxWorkData *aux_thread_aux_work_data;
+
#define ERTS_SCHDLR_SSPND_CHNG_WAITER (((erts_aint32_t) 1) << 0)
#define ERTS_SCHDLR_SSPND_CHNG_MSB (((erts_aint32_t) 1) << 1)
#define ERTS_SCHDLR_SSPND_CHNG_ONLN (((erts_aint32_t) 1) << 2)
@@ -152,6 +155,7 @@ do { \
static struct {
erts_smp_mtx_t mtx;
erts_smp_cnd_t cnd;
+ int aux_thread;
int online;
int curr_online;
int wait_curr_online;
@@ -213,8 +217,6 @@ Uint erts_no_run_queues;
ErtsAlignedSchedulerData *erts_aligned_scheduler_data;
-#ifdef ERTS_SMP
-
typedef union {
ErtsSchedulerSleepInfo ssi;
char align[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(ErtsSchedulerSleepInfo))];
@@ -222,8 +224,6 @@ typedef union {
static ErtsAlignedSchedulerSleepInfo *aligned_sched_sleep_info;
-#endif
-
#ifndef BM_COUNTERS
static int processes_busy;
#endif
@@ -285,8 +285,9 @@ ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(proclist,
ERTS_ALC_T_PROC_LIST)
#define ERTS_SCHED_SLEEP_INFO_IX(IX) \
- (ASSERT_EXPR(0 <= (IX) && (IX) < erts_no_schedulers), \
- &aligned_sched_sleep_info[(IX)].ssi)
+ (ASSERT_EXPR(-1 <= ((int) (IX)) \
+ && ((int) (IX)) < ((int) erts_no_schedulers)), \
+ &aligned_sched_sleep_info[(IX)].ssi)
#define ERTS_FOREACH_RUNQ(RQVAR, DO) \
do { \
@@ -339,6 +340,57 @@ static void exec_misc_ops(ErtsRunQueue *);
static void print_function_from_pc(int to, void *to_arg, BeamInstr* x);
static int stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp,
int yreg);
+
+static void aux_work_timeout(void *unused);
+static void aux_work_timeout_early_init(int no_schedulers);
+static void aux_work_timeout_late_init(void);
+static void setup_aux_work_timer(void);
+
+#if defined(DEBUG) || 0
+#define ERTS_DBG_CHK_AUX_WORK_VAL(V) dbg_chk_aux_work_val((V))
+static void
+dbg_chk_aux_work_val(erts_aint32_t value)
+{
+ erts_aint32_t valid = 0;
+
+#ifdef ERTS_SSI_AUX_WORK_SET_TMO
+ valid |= ERTS_SSI_AUX_WORK_SET_TMO;
+#endif
+#ifdef ERTS_SSI_AUX_WORK_CHECK_CHILDREN
+ valid |= ERTS_SSI_AUX_WORK_CHECK_CHILDREN;
+#endif
+#ifdef ERTS_SSI_AUX_WORK_MISC
+ valid |= ERTS_SSI_AUX_WORK_MISC;
+#endif
+
+#ifdef ERTS_SSI_AUX_WORK_FIX_ALLOC_LOWER_LIM
+ valid |= ERTS_SSI_AUX_WORK_FIX_ALLOC_LOWER_LIM;
+#endif
+#ifdef ERTS_SSI_AUX_WORK_FIX_ALLOC_DEALLOC
+ valid |= ERTS_SSI_AUX_WORK_FIX_ALLOC_DEALLOC;
+#endif
+#ifdef ERTS_SSI_AUX_WORK_DD
+ valid |= ERTS_SSI_AUX_WORK_DD;
+#endif
+#ifdef ERTS_SSI_AUX_WORK_DD
+ valid |= ERTS_SSI_AUX_WORK_DD_THR_PRGR;
+#endif
+#ifdef ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK
+ valid |= ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK;
+#endif
+
+ if (~valid & value)
+ erl_exit(ERTS_ABORT_EXIT,
+ "Invalid aux_work value found: 0x%x\n",
+ ~valid & value);
+}
+#define ERTS_DBG_CHK_SSI_AUX_WORK(SSI) \
+ ERTS_DBG_CHK_AUX_WORK_VAL(erts_atomic32_read_nob(&(SSI)->aux_work))
+#else
+#define ERTS_DBG_CHK_AUX_WORK_VAL(V)
+#define ERTS_DBG_CHK_SSI_AUX_WORK(SSI)
+#endif
+
#ifdef ERTS_SMP
static void handle_pending_exiters(ErtsProcList *);
@@ -570,6 +622,18 @@ erts_psd_set_init(Process *p, ErtsProcLocks plocks, int ix, void *data)
#ifdef ERTS_SMP
+static void
+prepare_for_block(void *vrq)
+{
+ erts_smp_runq_unlock((ErtsRunQueue *) vrq);
+}
+
+static void
+resume_after_block(void *vrq)
+{
+ erts_smp_runq_lock((ErtsRunQueue *) vrq);
+}
+
void
erts_sched_finish_poke(ErtsSchedulerSleepInfo *ssi, erts_aint32_t flags)
{
@@ -589,6 +653,68 @@ erts_sched_finish_poke(ErtsSchedulerSleepInfo *ssi, erts_aint32_t flags)
}
}
+#endif
+
+static ERTS_INLINE void
+set_aux_work_flags_wakeup_nob(ErtsSchedulerSleepInfo *ssi,
+ erts_aint32_t flgs)
+{
+ erts_aint32_t old_flgs;
+
+ ERTS_DBG_CHK_SSI_AUX_WORK(ssi);
+
+ old_flgs = erts_atomic32_read_nob(&ssi->aux_work);
+ if ((old_flgs & flgs) == 0) {
+
+ old_flgs = erts_atomic32_read_bor_nob(&ssi->aux_work, flgs);
+
+ if ((old_flgs & flgs) == 0) {
+#ifdef ERTS_SMP
+ erts_sched_poke(ssi);
+#else
+ erts_sys_schedule_interrupt(1);
+#endif
+ }
+ }
+}
+
+#if 0 /* Currently not used */
+
+static ERTS_INLINE void
+set_aux_work_flags_wakeup_relb(ErtsSchedulerSleepInfo *ssi,
+ erts_aint32_t flgs)
+{
+ erts_aint32_t old_flgs;
+
+ ERTS_DBG_CHK_SSI_AUX_WORK(ssi);
+
+ old_flgs = erts_atomic32_read_bor_relb(&ssi->aux_work, flgs);
+
+ if ((old_flgs & flgs) == 0) {
+#ifdef ERTS_SMP
+ erts_sched_poke(ssi);
+#else
+ erts_sys_schedule_interrupt(1);
+#endif
+ }
+}
+
+#endif
+
+static ERTS_INLINE erts_aint32_t
+set_aux_work_flags(ErtsSchedulerSleepInfo *ssi, erts_aint32_t flgs)
+{
+ return erts_atomic32_read_bor_nob(&ssi->aux_work, flgs);
+}
+
+static ERTS_INLINE erts_aint32_t
+unset_aux_work_flags(ErtsSchedulerSleepInfo *ssi, erts_aint32_t flgs)
+{
+ return erts_atomic32_read_band_nob(&ssi->aux_work, ~flgs);
+}
+
+#ifdef ERTS_SMP
+
typedef struct erts_misc_aux_work_t_ erts_misc_aux_work_t;
struct erts_misc_aux_work_t_ {
erts_misc_aux_work_t *next;
@@ -623,24 +749,27 @@ init_misc_aux_work(void)
misc_aux_work_queues =
erts_alloc_permanent_cache_aligned(ERTS_ALC_T_MISC_AUX_WORK_Q,
- erts_no_schedulers *
- sizeof(erts_algnd_misc_aux_work_q_t));
+ sizeof(erts_algnd_misc_aux_work_q_t)
+ * (erts_no_schedulers+1));
- for (ix = 0; ix < erts_no_schedulers; ix++) {
+ for (ix = 0; ix <= erts_no_schedulers; ix++) {
erts_smp_mtx_init_x(&misc_aux_work_queues[ix].data.mtx,
"misc_aux_work_queue",
- make_small(ix + 1));
+ make_small(ix));
misc_aux_work_queues[ix].data.first = NULL;
misc_aux_work_queues[ix].data.last = NULL;
}
}
-static void
-handle_misc_aux_work(ErtsSchedulerData *esdp)
+static erts_aint32_t
+handle_misc_aux_work(ErtsAuxWorkData *awdp,
+ erts_aint32_t aux_work)
{
- int ix = (int) esdp->no - 1;
+ int ix = (int) awdp->sched_id;
erts_misc_aux_work_t *mawp;
+ unset_aux_work_flags(awdp->ssi, ERTS_SSI_AUX_WORK_MISC);
+
erts_smp_mtx_lock(&misc_aux_work_queues[ix].data.mtx);
mawp = misc_aux_work_queues[ix].data.first;
misc_aux_work_queues[ix].data.first = NULL;
@@ -654,6 +783,35 @@ handle_misc_aux_work(ErtsSchedulerData *esdp)
mawp = mawp->next;
misc_aux_work_free(free_mawp);
}
+
+ return aux_work & ~ERTS_SSI_AUX_WORK_MISC;
+}
+
+static void
+smp_schedule_misc_aux_work(int ix,
+ void (*func)(void *),
+ void *arg)
+{
+ erts_aint32_t aux_work;
+ erts_misc_aux_work_t *mawp;
+ ErtsSchedulerSleepInfo *ssi;
+
+ mawp = misc_aux_work_alloc();
+
+ mawp->func = func;
+ mawp->arg = arg;
+ mawp->next = NULL;
+
+ erts_smp_mtx_lock(&misc_aux_work_queues[ix].data.mtx);
+ if (!misc_aux_work_queues[ix].data.last)
+ misc_aux_work_queues[ix].data.first = mawp;
+ else
+ misc_aux_work_queues[ix].data.last->next = mawp;
+ misc_aux_work_queues[ix].data.last = mawp;
+ erts_smp_mtx_unlock(&misc_aux_work_queues[ix].data.mtx);
+
+ set_aux_work_flags_wakeup_nob(ERTS_SCHED_SLEEP_INFO_IX(ix-1),
+ ERTS_SSI_AUX_WORK_MISC);
}
void
@@ -667,111 +825,425 @@ erts_smp_schedule_misc_aux_work(int ignore_self,
if (ignore_self) {
ErtsSchedulerData *esdp = erts_get_scheduler_data();
if (esdp)
- ignore_ix = (int) esdp->no - 1;
+ ignore_ix = (int) esdp->no;
}
- ASSERT(0 <= max_sched && max_sched <= erts_no_schedulers);
+ ASSERT(0 < max_sched && max_sched <= erts_no_schedulers);
- for (ix = 0; ix < max_sched; ix++) {
- erts_aint32_t aux_work;
- erts_misc_aux_work_t *mawp;
- ErtsSchedulerSleepInfo *ssi;
+ for (ix = 1; ix <= max_sched; ix++) {
if (ix == ignore_ix)
continue;
+ smp_schedule_misc_aux_work(ix, func, arg);
+ }
+}
- mawp = misc_aux_work_alloc();
+#endif
- mawp->func = func;
- mawp->arg = arg;
- mawp->next = NULL;
+static erts_aint32_t
+handle_fix_alloc(ErtsAuxWorkData *awdp, erts_aint32_t aux_work)
+{
+ ErtsSchedulerSleepInfo *ssi = awdp->ssi;
+ erts_aint32_t res;
- erts_smp_mtx_lock(&misc_aux_work_queues[ix].data.mtx);
- if (!misc_aux_work_queues[ix].data.last)
- misc_aux_work_queues[ix].data.first = mawp;
- else
- misc_aux_work_queues[ix].data.last->next = mawp;
- misc_aux_work_queues[ix].data.last = mawp;
- erts_smp_mtx_unlock(&misc_aux_work_queues[ix].data.mtx);
-
- ssi = ERTS_SCHED_SLEEP_INFO_IX(ix);
- aux_work = erts_smp_atomic32_read_bor_nob(&ssi->aux_work,
- ERTS_SSI_AUX_WORK_MISC);
- if ((aux_work & ERTS_SSI_AUX_WORK_MISC) == 0)
- erts_sched_poke(ssi);
- }
+ unset_aux_work_flags(ssi, (ERTS_SSI_AUX_WORK_FIX_ALLOC_LOWER_LIM
+ | ERTS_SSI_AUX_WORK_FIX_ALLOC_DEALLOC));
+ aux_work &= ~(ERTS_SSI_AUX_WORK_FIX_ALLOC_LOWER_LIM
+ | ERTS_SSI_AUX_WORK_FIX_ALLOC_DEALLOC);
+ res = erts_alloc_fix_alloc_shrink(awdp->sched_id, aux_work);
+ if (res) {
+ set_aux_work_flags(ssi, res);
+ aux_work |= res;
+ }
+
+ return aux_work;
+}
+
+#ifdef ERTS_SMP
+
+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);
+}
+
+static erts_aint32_t
+handle_delayed_dealloc(ErtsAuxWorkData *awdp, erts_aint32_t aux_work)
+{
+ ErtsSchedulerSleepInfo *ssi = awdp->ssi;
+ int need_thr_progress = 0;
+ int more_work = 0;
+
+ unset_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_DD);
+ erts_alloc_scheduler_handle_delayed_dealloc((void *) awdp->esdp,
+ &need_thr_progress,
+ &more_work);
+ if (more_work) {
+ if (set_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_DD)
+ & ERTS_SSI_AUX_WORK_DD_THR_PRGR) {
+ unset_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_DD_THR_PRGR);
+ aux_work &= ~ERTS_SSI_AUX_WORK_DD_THR_PRGR;
+ }
+ return aux_work;
+ }
+
+ if (need_thr_progress) {
+ set_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_DD_THR_PRGR);
+ awdp->dd.thr_prgr = erts_thr_progress_later();
+ erts_thr_progress_wakeup(awdp->esdp, awdp->dd.thr_prgr);
+ }
+ else if (awdp->dd.completed_callback) {
+ awdp->dd.completed_callback(awdp->dd.completed_arg);
+ awdp->dd.completed_callback = NULL;
+ awdp->dd.completed_arg = NULL;
+ }
+ return aux_work & ~ERTS_SSI_AUX_WORK_DD;
}
+static erts_aint32_t
+handle_delayed_dealloc_thr_prgr(ErtsAuxWorkData *awdp, erts_aint32_t aux_work)
+{
+ ErtsSchedulerSleepInfo *ssi;
+ int need_thr_progress;
+ int more_work;
+
+ if (!erts_thr_progress_has_reached(awdp->dd.thr_prgr))
+ return aux_work & ~ERTS_SSI_AUX_WORK_DD_THR_PRGR;
+
+ ssi = awdp->ssi;
+ need_thr_progress = 0;
+ more_work = 0;
+
+ erts_alloc_scheduler_handle_delayed_dealloc((void *) awdp->esdp,
+ &need_thr_progress,
+ &more_work);
+ if (more_work) {
+ set_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_DD);
+ unset_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_DD_THR_PRGR);
+ return ((aux_work & ~ERTS_SSI_AUX_WORK_DD_THR_PRGR)
+ | ERTS_SSI_AUX_WORK_DD);
+ }
+
+ if (need_thr_progress) {
+ awdp->dd.thr_prgr = erts_thr_progress_later();
+ erts_thr_progress_wakeup(awdp->esdp, awdp->dd.thr_prgr);
+ }
+ else {
+ unset_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_DD_THR_PRGR);
+ if (awdp->dd.completed_callback) {
+ awdp->dd.completed_callback(awdp->dd.completed_arg);
+ awdp->dd.completed_callback = NULL;
+ awdp->dd.completed_arg = NULL;
+ }
+ }
+
+ return aux_work & ~ERTS_SSI_AUX_WORK_DD_THR_PRGR;
+}
+
+static erts_atomic32_t completed_dealloc_count;
+
+static void
+completed_dealloc(void *vproc)
+{
+ if (erts_atomic32_dec_read_mb(&completed_dealloc_count) == 0) {
+ erts_resume((Process *) vproc, (ErtsProcLocks) 0);
+ erts_smp_proc_dec_refc((Process *) vproc);
+ }
+}
+
+static void
+setup_completed_dealloc(void *vproc)
+{
+ ErtsSchedulerData *esdp = erts_get_scheduler_data();
+ ErtsAuxWorkData *awdp = (esdp
+ ? &esdp->aux_work_data
+ : aux_thread_aux_work_data);
+ erts_alloc_fix_alloc_shrink(awdp->sched_id, 0);
+ set_aux_work_flags_wakeup_nob(awdp->ssi, ERTS_SSI_AUX_WORK_DD);
+ awdp->dd.completed_callback = completed_dealloc;
+ awdp->dd.completed_arg = vproc;
+}
+
+static void
+prep_setup_completed_dealloc(void *vproc)
+{
+ erts_aint32_t count = (erts_aint32_t) (erts_no_schedulers+1);
+ if (erts_atomic32_dec_read_mb(&completed_dealloc_count) == count) {
+ /* scheduler threads */
+ erts_smp_schedule_misc_aux_work(0,
+ erts_no_schedulers,
+ setup_completed_dealloc,
+ vproc);
+ /* aux_thread */
+ smp_schedule_misc_aux_work(0,
+ setup_completed_dealloc,
+ vproc);
+ }
+}
+
+#endif /* ERTS_SMP */
+
+int
+erts_debug_wait_deallocations(Process *c_p)
+{
+#ifndef ERTS_SMP
+ erts_alloc_fix_alloc_shrink(1, 0);
+ return 1;
+#else
+ /* Only one process at a time can do this */
+ erts_aint32_t count = (erts_aint32_t) (2*(erts_no_schedulers+1));
+ if (0 == erts_atomic32_cmpxchg_mb(&completed_dealloc_count,
+ count,
+ 0)) {
+ erts_suspend(c_p, ERTS_PROC_LOCK_MAIN, NULL);
+ erts_smp_proc_inc_refc(c_p);
+ /* scheduler threads */
+ erts_smp_schedule_misc_aux_work(0,
+ erts_no_schedulers,
+ prep_setup_completed_dealloc,
+ (void *) c_p);
+ /* aux_thread */
+ smp_schedule_misc_aux_work(0,
+ prep_setup_completed_dealloc,
+ (void *) c_p);
+ return 1;
+ }
+ return 0;
+#endif
+}
+
+
#ifdef ERTS_SMP_SCHEDULERS_NEED_TO_CHECK_CHILDREN
void
erts_smp_notify_check_children_needed(void)
{
int i;
+ for (i = 0; i < erts_no_schedulers; i++)
+ set_aux_work_flags_wakeup_nob(ERTS_SCHED_SLEEP_INFO_IX(i),
+ ERTS_SSI_AUX_WORK_CHECK_CHILDREN);
+}
- for (i = 0; i < erts_no_schedulers; i++) {
- erts_aint32_t aux_work;
- ErtsSchedulerSleepInfo *ssi;
- ssi = ERTS_SCHED_SLEEP_INFO_IX(i);
- aux_work = erts_smp_atomic32_read_bor_nob(&ssi->aux_work,
- ERTS_SSI_AUX_WORK_CHECK_CHILDREN);
- if (!(aux_work & ERTS_SSI_AUX_WORK_CHECK_CHILDREN))
- erts_sched_poke(ssi);
- }
+static erts_aint32_t
+handle_check_children(ErtsAuxWorkData *awdp, erts_aint32_t aux_work)
+{
+ unset_aux_work_flags(awdp->ssi, ERTS_SSI_AUX_WORK_CHECK_CHILDREN);
+ erts_check_children();
+ return aux_work & ~ERTS_SSI_AUX_WORK_CHECK_CHILDREN;
}
+
#endif
-#ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK
+#ifdef ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK
+
+static erts_aint32_t
+handle_mseg_cache_check(ErtsAuxWorkData *awdp, erts_aint32_t aux_work)
+{
+ unset_aux_work_flags(awdp->ssi, ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK);
+ erts_mseg_cache_check();
+ return aux_work & ~ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK;
+}
+
+#endif
+
+static erts_aint32_t
+handle_setup_aux_work_timer(ErtsAuxWorkData *awdp, erts_aint32_t aux_work)
+{
+ unset_aux_work_flags(awdp->ssi, ERTS_SSI_AUX_WORK_SET_TMO);
+ setup_aux_work_timer();
+ return aux_work & ~ERTS_SSI_AUX_WORK_SET_TMO;
+}
+
static ERTS_INLINE erts_aint32_t
-blockable_aux_work(ErtsSchedulerData *esdp,
- ErtsSchedulerSleepInfo *ssi,
- erts_aint32_t aux_work)
+handle_aux_work(ErtsAuxWorkData *awdp, erts_aint32_t aux_work)
{
- if (aux_work & ERTS_SSI_BLOCKABLE_AUX_WORK_MASK) {
- if (aux_work & ERTS_SSI_AUX_WORK_MISC) {
- aux_work = erts_smp_atomic32_read_band_nob(&ssi->aux_work,
- ~ERTS_SSI_AUX_WORK_MISC);
- aux_work &= ~ERTS_SSI_AUX_WORK_MISC;
- handle_misc_aux_work(esdp);
- }
+ /*
+ * Handlers are *only* allowed to modify flags in return value
+ * and ssi flags that are explicity handled by the handler.
+ * Handlers are, e.g., not allowed to read the ssi flag field and
+ * then unconditionally return that value.
+ */
+ ERTS_DBG_CHK_AUX_WORK_VAL(aux_work);
+ if (aux_work & ERTS_SSI_AUX_WORK_SET_TMO) {
+ aux_work = handle_setup_aux_work_timer(awdp, aux_work);
+ ERTS_DBG_CHK_AUX_WORK_VAL(aux_work);
+ }
+#ifdef ERTS_SMP
+ if (aux_work & ERTS_SSI_AUX_WORK_MISC) {
+ aux_work = handle_misc_aux_work(awdp, aux_work);
+ ERTS_DBG_CHK_AUX_WORK_VAL(aux_work);
+ }
+#endif
#ifdef ERTS_SMP_SCHEDULERS_NEED_TO_CHECK_CHILDREN
- if (aux_work & ERTS_SSI_AUX_WORK_CHECK_CHILDREN) {
- aux_work = erts_smp_atomic32_band_nob(&ssi->aux_work,
- ~ERTS_SSI_AUX_WORK_CHECK_CHILDREN);
- aux_work &= ~ERTS_SSI_AUX_WORK_CHECK_CHILDREN;
- erts_check_children();
- }
+ if (aux_work & ERTS_SSI_AUX_WORK_CHECK_CHILDREN) {
+ aux_work = handle_check_children(awdp, aux_work);
+ ERTS_DBG_CHK_AUX_WORK_VAL(aux_work);
+ }
+#endif
+ if (aux_work & (ERTS_SSI_AUX_WORK_FIX_ALLOC_LOWER_LIM
+ | ERTS_SSI_AUX_WORK_FIX_ALLOC_DEALLOC)) {
+ aux_work = handle_fix_alloc(awdp, aux_work);
+ ERTS_DBG_CHK_AUX_WORK_VAL(aux_work);
+ }
+#ifdef ERTS_SMP
+ if (aux_work & ERTS_SSI_AUX_WORK_DD) {
+ aux_work = handle_delayed_dealloc(awdp, aux_work);
+ ERTS_DBG_CHK_AUX_WORK_VAL(aux_work);
+ }
+ if (aux_work & ERTS_SSI_AUX_WORK_DD_THR_PRGR) {
+ aux_work = handle_delayed_dealloc_thr_prgr(awdp, aux_work);
+ ERTS_DBG_CHK_AUX_WORK_VAL(aux_work);
+ }
#endif
+#ifdef ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK
+ if (aux_work & ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK) {
+ aux_work = handle_mseg_cache_check(awdp, aux_work);
+ ERTS_DBG_CHK_AUX_WORK_VAL(aux_work);
}
+#endif
+ ERTS_DBG_CHK_AUX_WORK_VAL(aux_work);
return aux_work;
}
-#endif
+typedef struct {
+ union {
+ ErlTimer data;
+ char align__[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(ErlTimer))];
+ } timer;
-#ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK
-static ERTS_INLINE erts_aint32_t
-nonblockable_aux_work(ErtsSchedulerData *esdp,
- ErtsSchedulerSleepInfo *ssi,
- erts_aint32_t aux_work)
+ int initialized;
+ erts_atomic32_t refc;
+ erts_atomic32_t type[1];
+} ErtsAuxWorkTmo;
+
+static ErtsAuxWorkTmo *aux_work_tmo;
+
+static void
+aux_work_timeout_early_init(int no_schedulers)
{
- if (aux_work & ERTS_SSI_NONBLOCKABLE_AUX_WORK_MASK) {
+ int i;
+ UWord p;
+
+ /*
+ * This is done really early. Our own allocators have
+ * not been started yet.
+ */
+
+ p = (UWord) malloc((sizeof(ErtsAuxWorkTmo)
+ + sizeof(erts_atomic32_t)*(no_schedulers+1))
+ + ERTS_CACHE_LINE_SIZE-1);
+ if (p & ERTS_CACHE_LINE_MASK)
+ p = (p & ~ERTS_CACHE_LINE_MASK) + ERTS_CACHE_LINE_SIZE;
+ ASSERT((p & ERTS_CACHE_LINE_MASK) == 0);
+
+ aux_work_tmo = (ErtsAuxWorkTmo *) p;
+ aux_work_tmo->initialized = 0;
+ erts_atomic32_init_nob(&aux_work_tmo->refc, 0);
+ for (i = 0; i <= no_schedulers; i++)
+ erts_atomic32_init_nob(&aux_work_tmo->type[i], 0);
+}
+void
+aux_work_timeout_late_init(void)
+{
+ aux_work_tmo->initialized = 1;
+ if (erts_atomic32_read_nob(&aux_work_tmo->refc)) {
+ aux_work_tmo->timer.data.active = 0;
+ erts_set_timer(&aux_work_tmo->timer.data,
+ aux_work_timeout,
+ NULL,
+ NULL,
+ 1000);
}
}
-#endif
static void
-prepare_for_block(void *vrq)
+aux_work_timeout(void *unused)
{
- erts_smp_runq_unlock((ErtsRunQueue *) vrq);
+ erts_aint32_t refc;
+ int i;
+#ifdef ERTS_SMP
+ i = 0;
+#else
+ i = 1;
+#endif
+
+ for (; i <= erts_no_schedulers; i++) {
+ erts_aint32_t type;
+ type = erts_atomic32_read_acqb(&aux_work_tmo->type[i]);
+ if (type)
+ set_aux_work_flags_wakeup_nob(ERTS_SCHED_SLEEP_INFO_IX(i-1),
+ type);
+ }
+
+ refc = erts_atomic32_read_nob(&aux_work_tmo->refc);
+ ASSERT(refc >= 1);
+ if (refc != 1
+ || 1 != erts_atomic32_cmpxchg_relb(&aux_work_tmo->refc, 0, 1)) {
+ /* Setup next timeout... */
+ aux_work_tmo->timer.data.active = 0;
+ erts_set_timer(&aux_work_tmo->timer.data,
+ aux_work_timeout,
+ NULL,
+ NULL,
+ 1000);
+ }
}
static void
-resume_after_block(void *vrq)
+setup_aux_work_timer(void)
{
- erts_smp_runq_lock((ErtsRunQueue *) vrq);
+#ifndef ERTS_SMP
+ if (!erts_get_scheduler_data())
+ set_aux_work_flags_wakeup_nob(ERTS_SCHED_SLEEP_INFO_IX(0),
+ ERTS_SSI_AUX_WORK_SET_TMO);
+ else
+#endif
+ {
+ aux_work_tmo->timer.data.active = 0;
+ erts_set_timer(&aux_work_tmo->timer.data,
+ aux_work_timeout,
+ NULL,
+ NULL,
+ 1000);
+ }
}
+erts_aint32_t
+erts_set_aux_work_timeout(int ix, erts_aint32_t type, int enable)
+{
+ erts_aint32_t old, refc;
+
+#ifndef ERTS_SMP
+ ix = 1;
#endif
+ ERTS_DBG_CHK_AUX_WORK_VAL(type);
+ ERTS_DBG_CHK_AUX_WORK_VAL(erts_atomic32_read_nob(&aux_work_tmo->type[ix]));
+// erts_fprintf(stderr, "t(%d, 0x%x, %d)\n", ix, type, enable);
+
+ if (!enable) {
+ old = erts_atomic32_read_band_mb(&aux_work_tmo->type[ix], ~type);
+ ERTS_DBG_CHK_AUX_WORK_VAL(erts_atomic32_read_nob(&aux_work_tmo->type[ix]));
+ if (old != 0 && (old & ~type) == 0)
+ erts_atomic32_dec_relb(&aux_work_tmo->refc);
+ return old;
+ }
+
+ old = erts_atomic32_read_bor_mb(&aux_work_tmo->type[ix], type);
+ ERTS_DBG_CHK_AUX_WORK_VAL(erts_atomic32_read_nob(&aux_work_tmo->type[ix]));
+ if (old == 0 && type != 0) {
+ refc = erts_atomic32_inc_read_acqb(&aux_work_tmo->refc);
+ if (refc == 1) {
+ erts_atomic32_inc_acqb(&aux_work_tmo->refc);
+ if (aux_work_tmo->initialized)
+ setup_aux_work_timer();
+ }
+ }
+ return old;
+}
+
+
+
static ERTS_INLINE void
sched_waiting_sys(Uint no, ErtsRunQueue *rq)
{
@@ -800,8 +1272,6 @@ sched_active_sys(Uint no, ErtsRunQueue *rq)
Uint
erts_active_schedulers(void)
{
- /* RRRRRRRRR */
-
Uint as = erts_no_schedulers;
ERTS_ATOMIC_FOREACH_RUNQ(rq, as -= abs(rq->waiting));
@@ -988,6 +1458,10 @@ sched_set_sleeptype(ErtsSchedulerSleepInfo *ssi, erts_aint32_t sleep_type)
if (sleep_type == ERTS_SSI_FLG_TSE_SLEEPING)
erts_tse_reset(ssi->event);
+ else {
+ ASSERT(sleep_type == ERTS_SSI_FLG_POLL_SLEEPING);
+ erts_sys_schedule_interrupt(0);
+ }
while (1) {
oflgs = erts_smp_atomic32_cmpxchg_acqb(&ssi->flags, nflgs, xflgs);
@@ -1006,16 +1480,96 @@ sched_set_sleeptype(ErtsSchedulerSleepInfo *ssi, erts_aint32_t sleep_type)
(((FLGS) & (ERTS_SSI_FLG_WAITING|ERTS_SSI_FLG_SUSPENDED)) \
!= ERTS_SSI_FLG_WAITING)
+
+static void
+poke_ssi(void *vssi)
+{
+ erts_sched_poke((ErtsSchedulerSleepInfo *) vssi);
+}
+
+static void init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp);
+
+static void *
+aux_thread(void *unused)
+{
+ ErtsAuxWorkData *awdp = aux_thread_aux_work_data;
+ ErtsSchedulerSleepInfo *ssi = ERTS_SCHED_SLEEP_INFO_IX(-1);
+ erts_aint32_t aux_work;
+ ErtsThrPrgrWakeupCallback wake_me;
+ int thr_prgr_active = 1;
+
+ wake_me.wakeup = poke_ssi;
+ wake_me.arg = (void *) ssi;
+
+ erts_thr_progress_register_managed_thread(NULL, &wake_me, 1);
+ init_aux_work_data(awdp, NULL);
+ awdp->ssi = ssi;
+
+ erts_register_blockable_thread();
+
+ ssi->event = erts_tse_fetch();
+
+ erts_smp_mtx_lock(&schdlr_sspnd.mtx);
+ schdlr_sspnd.aux_thread = 1;
+ erts_smp_cnd_signal(&schdlr_sspnd.cnd);
+ erts_smp_mtx_unlock(&schdlr_sspnd.mtx);
+
+ sched_prep_spin_wait(ssi);
+
+ while (1) {
+ erts_aint32_t flgs;
+
+ aux_work = erts_atomic32_read_acqb(&ssi->aux_work);
+ if (aux_work) {
+ if (!thr_prgr_active)
+ erts_thr_progress_active(NULL, thr_prgr_active = 1);
+ aux_work = handle_aux_work(awdp, aux_work);
+ if (aux_work && erts_thr_progress_update(NULL))
+ erts_thr_progress_leader_update(NULL);
+ }
+
+ erts_smp_activity_begin(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
+
+ if (!aux_work) {
+ if (thr_prgr_active)
+ erts_thr_progress_active(NULL, thr_prgr_active = 0);
+ erts_thr_progress_prepare_wait(NULL);
+
+ flgs = sched_spin_wait(ssi, 0);
+
+ if (flgs & ERTS_SSI_FLG_SLEEPING) {
+ ASSERT(flgs & ERTS_SSI_FLG_WAITING);
+ flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING);
+ if (flgs & ERTS_SSI_FLG_SLEEPING) {
+ int res;
+ ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING);
+ ASSERT(flgs & ERTS_SSI_FLG_WAITING);
+ do {
+ res = erts_tse_wait(ssi->event);
+ } while (res == EINTR);
+ }
+ }
+ erts_thr_progress_finalize_wait(NULL);
+ }
+
+ erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
+
+ flgs = sched_prep_spin_wait(ssi);
+ }
+ return NULL;
+}
+
+#endif /* ERTS_SMP */
+
static void
scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
{
ErtsSchedulerSleepInfo *ssi = esdp->ssi;
int spincount;
+ erts_aint32_t aux_work = 0;
+#ifdef ERTS_SMP
+ int thr_prgr_active = 1;
erts_aint32_t flgs;
-#if defined(ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK) \
- || defined(ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK)
- erts_aint32_t aux_work;
-#endif
ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq));
@@ -1049,36 +1603,44 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
tse_wait:
-#ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK
- aux_work = erts_smp_atomic32_read_nob(&ssi->aux_work);
- tse_blockable_aux_work:
- aux_work = blockable_aux_work(esdp, ssi, aux_work);
-#endif
- erts_smp_activity_begin(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
-
while (1) {
-#ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK
-#ifndef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK
- aux_work = erts_smp_atomic32_read_nob(&ssi->aux_work);
-#endif
- nonblockable_aux_work(esdp, ssi, aux_work);
-#endif
+ aux_work = erts_atomic32_read_acqb(&ssi->aux_work);
+ if (aux_work) {
+ if (!thr_prgr_active)
+ erts_thr_progress_active(esdp, thr_prgr_active = 1);
+ aux_work = handle_aux_work(&esdp->aux_work_data, aux_work);
+ if (aux_work && erts_thr_progress_update(esdp))
+ erts_thr_progress_leader_update(esdp);
+ }
- flgs = sched_spin_wait(ssi, spincount);
- if (flgs & ERTS_SSI_FLG_SLEEPING) {
- ASSERT(flgs & ERTS_SSI_FLG_WAITING);
- flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING);
+ erts_smp_activity_begin(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
+
+ if (aux_work)
+ flgs = erts_smp_atomic32_read_acqb(&ssi->flags);
+ else {
+ if (thr_prgr_active)
+ erts_thr_progress_active(esdp, thr_prgr_active = 0);
+ erts_thr_progress_prepare_wait(esdp);
+
+ flgs = sched_spin_wait(ssi, spincount);
if (flgs & ERTS_SSI_FLG_SLEEPING) {
- int res;
- ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING);
ASSERT(flgs & ERTS_SSI_FLG_WAITING);
- do {
- res = erts_tse_wait(ssi->event);
- } while (res == EINTR);
+ flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING);
+ if (flgs & ERTS_SSI_FLG_SLEEPING) {
+ int res;
+ ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING);
+ ASSERT(flgs & ERTS_SSI_FLG_WAITING);
+ do {
+ res = erts_tse_wait(ssi->event);
+ } while (res == EINTR);
+ }
}
+ erts_thr_progress_finalize_wait(esdp);
}
+ erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
+
if (!(flgs & ERTS_SSI_FLG_WAITING)) {
ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING));
break;
@@ -1092,18 +1654,8 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
break;
}
-#ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK
- aux_work = erts_smp_atomic32_read_nob(&ssi->aux_work);
- if (aux_work & ERTS_SSI_BLOCKABLE_AUX_WORK_MASK) {
- erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
- goto tse_blockable_aux_work;
- }
-#endif
-
}
- erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
-
if (flgs & ~ERTS_SSI_FLG_SUSPENDED)
erts_smp_atomic32_read_band_nob(&ssi->flags, ERTS_SSI_FLG_SUSPENDED);
@@ -1111,7 +1663,9 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
sched_active(esdp->no, rq);
}
- else {
+ else
+#endif
+ {
erts_aint_t dt;
erts_smp_atomic32_set_relb(&function_calls, 0);
@@ -1135,18 +1689,27 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
if (dt) erts_bump_timer(dt);
sys_aux_work:
-
-#ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK
- aux_work = erts_smp_atomic32_read_nob(&ssi->aux_work);
- aux_work = blockable_aux_work(esdp, ssi, aux_work);
+#ifndef ERTS_SMP
+ erts_sys_schedule_interrupt(0);
#endif
-#ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK
-#ifndef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK
- aux_work = erts_smp_atomic32_read_nob(&ssi->aux_work);
+
+ aux_work = erts_atomic32_read_acqb(&ssi->aux_work);
+ if (aux_work) {
+#ifdef ERTS_SMP
+ if (!thr_prgr_active)
+ erts_thr_progress_active(esdp, thr_prgr_active = 1);
#endif
- nonblockable_aux_work(esdp, ssi, aux_work);
+ aux_work = handle_aux_work(&esdp->aux_work_data, aux_work);
+#ifdef ERTS_SMP
+ 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)
+ goto sys_woken;
+#else
flgs = erts_smp_atomic32_read_acqb(&ssi->flags);
if (!(flgs & ERTS_SSI_FLG_WAITING)) {
ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING));
@@ -1168,10 +1731,12 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
goto tse_wait;
}
}
+#endif
}
erts_smp_runq_lock(rq);
+#ifdef ERTS_SMP
/*
* If we got new I/O tasks we aren't allowed to
* sleep in erl_sys_schedule().
@@ -1183,64 +1748,81 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
* Got to check that we still got I/O tasks; otherwise
* we have to wait in erl_sys_schedule() after all...
*/
- if (prepare_for_sys_schedule())
- goto do_sys_schedule;
-
- /*
- * Not allowed to wait in erl_sys_schedule;
- * do tse wait instead...
- */
- sched_change_waiting_sys_to_waiting(esdp->no, rq);
+ if (!prepare_for_sys_schedule()) {
+ /*
+ * Not allowed to wait in erl_sys_schedule;
+ * do tse wait instead...
+ */
+ sched_change_waiting_sys_to_waiting(esdp->no, rq);
+ erts_smp_runq_unlock(rq);
+ spincount = 0;
+ goto tse_wait;
+ }
+ }
+#endif
+ if (aux_work) {
erts_smp_runq_unlock(rq);
- spincount = 0;
- goto tse_wait;
+ goto sys_poll_aux_work;
}
- else {
- do_sys_schedule:
- erts_sys_schedule_interrupt(0);
- flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_POLL_SLEEPING);
- if (!(flgs & ERTS_SSI_FLG_SLEEPING)) {
- if (!(flgs & ERTS_SSI_FLG_WAITING))
- goto sys_locked_woken;
- erts_smp_runq_unlock(rq);
- flgs = sched_prep_cont_spin_wait(ssi);
- if (!(flgs & ERTS_SSI_FLG_WAITING)) {
- ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING));
- goto sys_woken;
- }
- ASSERT(!erts_port_task_have_outstanding_io_tasks());
- goto sys_poll_aux_work;
+#ifdef ERTS_SMP
+ if (thr_prgr_active)
+ erts_thr_progress_active(esdp, thr_prgr_active = 0);
+ flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_POLL_SLEEPING);
+ if (!(flgs & ERTS_SSI_FLG_SLEEPING)) {
+ if (!(flgs & ERTS_SSI_FLG_WAITING))
+ goto sys_locked_woken;
+ erts_smp_runq_unlock(rq);
+ flgs = sched_prep_cont_spin_wait(ssi);
+ if (!(flgs & ERTS_SSI_FLG_WAITING)) {
+ ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING));
+ goto sys_woken;
}
+ ASSERT(!erts_port_task_have_outstanding_io_tasks());
+ goto sys_poll_aux_work;
+ }
- ASSERT(flgs & ERTS_SSI_FLG_POLL_SLEEPING);
- ASSERT(flgs & ERTS_SSI_FLG_WAITING);
+ ASSERT(flgs & ERTS_SSI_FLG_POLL_SLEEPING);
+ ASSERT(flgs & ERTS_SSI_FLG_WAITING);
+#endif
- erts_smp_runq_unlock(rq);
+ erts_smp_runq_unlock(rq);
- ASSERT(!erts_port_task_have_outstanding_io_tasks());
+ ASSERT(!erts_port_task_have_outstanding_io_tasks());
- erl_sys_schedule(0);
+ erl_sys_schedule(0);
- dt = erts_do_time_read_and_reset();
- if (dt) erts_bump_timer(dt);
+ dt = erts_do_time_read_and_reset();
+ if (dt) erts_bump_timer(dt);
- flgs = sched_prep_cont_spin_wait(ssi);
- if (flgs & ERTS_SSI_FLG_WAITING)
- goto sys_aux_work;
+#ifndef ERTS_SMP
+ if (rq->len == 0 && !rq->misc.start)
+ goto sys_aux_work;
+ sys_woken:
+#else
+ flgs = sched_prep_cont_spin_wait(ssi);
+ if (flgs & ERTS_SSI_FLG_WAITING)
+ goto sys_aux_work;
- sys_woken:
- erts_smp_runq_lock(rq);
- sys_locked_woken:
- clear_sys_scheduling();
- if (flgs & ~ERTS_SSI_FLG_SUSPENDED)
- erts_smp_atomic32_read_band_nob(&ssi->flags, ERTS_SSI_FLG_SUSPENDED);
- sched_active_sys(esdp->no, rq);
- }
+ sys_woken:
+ erts_smp_runq_lock(rq);
+ sys_locked_woken:
+ clear_sys_scheduling();
+ if (flgs & ~ERTS_SSI_FLG_SUSPENDED)
+ erts_smp_atomic32_read_band_nob(&ssi->flags, ERTS_SSI_FLG_SUSPENDED);
+#endif
+ sched_active_sys(esdp->no, rq);
}
+#ifdef ERTS_SMP
+ if (!thr_prgr_active)
+ erts_thr_progress_active(esdp, thr_prgr_active = 1);
+
ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq));
+#endif
}
+#ifdef ERTS_SMP
+
static ERTS_INLINE erts_aint32_t
ssi_flags_set_wake(ErtsSchedulerSleepInfo *ssi)
{
@@ -2547,8 +3129,9 @@ erts_debug_nbalance(void)
}
void
-erts_early_init_scheduling(void)
+erts_early_init_scheduling(int no_schedulers)
{
+ aux_work_timeout_early_init(no_schedulers);
wakeup_other_limit = ERTS_WAKEUP_OTHER_LIMIT_MEDIUM;
}
@@ -2569,12 +3152,24 @@ erts_sched_set_wakeup_limit(char *str)
return EINVAL;
return 0;
}
-
+
+static void
+init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp)
+{
+ awdp->sched_id = esdp ? (int) esdp->no : 0;
+ awdp->esdp = esdp;
+ awdp->ssi = esdp ? esdp->ssi : NULL;
+#ifdef ERTS_SMP
+ awdp->dd.thr_prgr = ERTS_THR_PRGR_VAL_WAITING;
+ awdp->dd.completed_callback = NULL;
+ awdp->dd.completed_arg = NULL;
+#endif
+}
void
erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online)
{
- int ix, n;
+ int ix, n, no_ssi;
#ifndef ERTS_SMP
mrq = 0;
@@ -2684,23 +3279,31 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online)
n = (int) no_schedulers;
erts_no_schedulers = n;
-#ifdef ERTS_SMP
/* Create and initialize scheduler sleep info */
-
+#ifdef ERTS_SMP
+ no_ssi = n+1;
+#else
+ no_ssi = 1;
+#endif
aligned_sched_sleep_info =
- erts_alloc_permanent_cache_aligned(ERTS_ALC_T_SCHDLR_SLP_INFO,
- n * sizeof(ErtsAlignedSchedulerSleepInfo));
-
- for (ix = 0; ix < n; ix++) {
- ErtsSchedulerSleepInfo *ssi = ERTS_SCHED_SLEEP_INFO_IX(ix);
+ erts_alloc_permanent_cache_aligned(
+ ERTS_ALC_T_SCHDLR_SLP_INFO,
+ no_ssi*sizeof(ErtsAlignedSchedulerSleepInfo));
+ for (ix = 0; ix < no_ssi; ix++) {
+ ErtsSchedulerSleepInfo *ssi = &aligned_sched_sleep_info[ix].ssi;
+#ifdef ERTS_SMP
#if 0 /* no need to initialize these... */
ssi->next = NULL;
ssi->prev = NULL;
#endif
erts_smp_atomic32_init_nob(&ssi->flags, 0);
ssi->event = NULL; /* initialized in sched_thread_func */
- erts_smp_atomic32_init_nob(&ssi->aux_work, 0);
+#endif
+ erts_atomic32_init_nob(&ssi->aux_work, 0);
}
+
+#ifdef ERTS_SMP
+ aligned_sched_sleep_info++;
#endif
/* Create and initialize scheduler specific data */
@@ -2714,7 +3317,6 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online)
#ifdef ERTS_SMP
erts_bits_init_state(&esdp->erl_bits_state);
esdp->match_pseudo_process = NULL;
- esdp->ssi = ERTS_SCHED_SLEEP_INFO_IX(ix);
esdp->free_process = NULL;
#if HALFWORD_HEAP
/* Registers need to be heap allocated (correct memory range) for tracing to work */
@@ -2725,6 +3327,7 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online)
esdp->num_tmp_heap_used = 0;
#endif
esdp->no = (Uint) ix+1;
+ esdp->ssi = ERTS_SCHED_SLEEP_INFO_IX(ix);
esdp->current_process = NULL;
esdp->current_port = NULL;
@@ -2745,13 +3348,23 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online)
#ifdef ERTS_SMP
erts_smp_atomic32_init_nob(&esdp->chk_cpu_bind, 0);
#endif
+ init_aux_work_data(&esdp->aux_work_data, esdp);
}
#ifdef ERTS_SMP
+ init_misc_aux_work();
+
+ erts_atomic32_init_nob(&completed_dealloc_count, 0); /* debug only */
+
+ aux_thread_aux_work_data =
+ erts_alloc_permanent_cache_aligned(ERTS_ALC_T_SCHDLR_DATA,
+ sizeof(ErtsAuxWorkData));
+
erts_smp_mtx_init(&schdlr_sspnd.mtx, "schdlr_sspnd");
erts_smp_cnd_init(&schdlr_sspnd.cnd);
erts_smp_atomic32_init_nob(&schdlr_sspnd.changing, 0);
+ schdlr_sspnd.aux_thread = 0;
schdlr_sspnd.online = no_schedulers_online;
schdlr_sspnd.curr_online = no_schedulers;
schdlr_sspnd.msb.ongoing = 0;
@@ -2809,6 +3422,8 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online)
/* init port tasks */
erts_port_task_init();
+ aux_work_timeout_late_init();
+
#ifndef ERTS_SMP
#ifdef ERTS_DO_VERIFY_UNUSED_TEMP_ALLOC
erts_scheduler_data->verify_unused_temp_alloc
@@ -3058,10 +3673,8 @@ suspend_scheduler(ErtsSchedulerData *esdp)
long active_schedulers;
int curr_online = 1;
int wake = 0;
-#if defined(ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK) \
- || defined(ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK)
erts_aint32_t aux_work;
-#endif
+ int thr_prgr_active = 1;
/*
* Schedulers may be suspended in two different ways:
@@ -3134,40 +3747,46 @@ suspend_scheduler(ErtsSchedulerData *esdp)
break;
erts_smp_mtx_unlock(&schdlr_sspnd.mtx);
-
-#ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK
- aux_work = erts_smp_atomic32_read_nob(&ssi->aux_work);
- blockable_aux_work:
- blockable_aux_work(esdp, ssi, aux_work);
-#endif
-
- erts_smp_activity_begin(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
while (1) {
erts_aint32_t flgs;
-#ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK
-#ifndef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK
- aux_work = erts_smp_atomic32_read(&ssi->aux_work);
-#endif
- nonblockable_aux_work(esdp, ssi, aux_work);
-#endif
- flgs = sched_spin_suspended(ssi,
- ERTS_SCHED_SUSPEND_SLEEP_SPINCOUNT);
- if (flgs == (ERTS_SSI_FLG_SLEEPING
- | ERTS_SSI_FLG_WAITING
- | ERTS_SSI_FLG_SUSPENDED)) {
- flgs = sched_set_suspended_sleeptype(ssi);
+ aux_work = erts_atomic32_read_acqb(&ssi->aux_work);
+ if (aux_work) {
+ if (!thr_prgr_active)
+ erts_thr_progress_active(esdp, thr_prgr_active = 1);
+ aux_work = handle_aux_work(&esdp->aux_work_data, aux_work);
+ if (aux_work && erts_thr_progress_update(esdp))
+ erts_thr_progress_leader_update(esdp);
+ }
+
+ erts_smp_activity_begin(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
+
+ if (!aux_work) {
+ if (thr_prgr_active)
+ erts_thr_progress_active(esdp, thr_prgr_active = 0);
+ erts_thr_progress_prepare_wait(esdp);
+ flgs = sched_spin_suspended(ssi,
+ ERTS_SCHED_SUSPEND_SLEEP_SPINCOUNT);
if (flgs == (ERTS_SSI_FLG_SLEEPING
- | ERTS_SSI_FLG_TSE_SLEEPING
| ERTS_SSI_FLG_WAITING
| ERTS_SSI_FLG_SUSPENDED)) {
- int res;
- do {
- res = erts_tse_wait(ssi->event);
- } while (res == EINTR);
+ flgs = sched_set_suspended_sleeptype(ssi);
+ if (flgs == (ERTS_SSI_FLG_SLEEPING
+ | ERTS_SSI_FLG_TSE_SLEEPING
+ | ERTS_SSI_FLG_WAITING
+ | ERTS_SSI_FLG_SUSPENDED)) {
+ int res;
+
+ do {
+ res = erts_tse_wait(ssi->event);
+ } while (res == EINTR);
+ }
}
+ erts_thr_progress_finalize_wait(esdp);
}
+ erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
+
flgs = sched_prep_spin_suspended(ssi, (ERTS_SSI_FLG_WAITING
| ERTS_SSI_FLG_SUSPENDED));
if (!(flgs & ERTS_SSI_FLG_SUSPENDED))
@@ -3175,19 +3794,8 @@ suspend_scheduler(ErtsSchedulerData *esdp)
changing = erts_smp_atomic32_read_nob(&schdlr_sspnd.changing);
if (changing & ~ERTS_SCHDLR_SSPND_CHNG_WAITER)
break;
-
-
-#ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK
- aux_work = erts_smp_atomic32_read_nob(&ssi->aux_work);
- if (aux_work & ERTS_SSI_BLOCKABLE_AUX_WORK_MASK) {
- erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
- goto blockable_aux_work;
- }
-#endif
-
}
- erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL);
erts_smp_mtx_lock(&schdlr_sspnd.mtx);
changing = erts_smp_atomic32_read_nob(&schdlr_sspnd.changing);
@@ -3213,6 +3821,9 @@ suspend_scheduler(ErtsSchedulerData *esdp)
if (erts_system_profile_flags.scheduler)
profile_scheduler(make_small(esdp->no), am_active);
+ if (!thr_prgr_active)
+ erts_thr_progress_active(esdp, thr_prgr_active = 1);
+
erts_smp_runq_lock(esdp->run_queue);
non_empty_runq(esdp->run_queue);
@@ -3720,8 +4331,14 @@ erts_multi_scheduling_blockers(Process *p)
static void *
sched_thread_func(void *vesdp)
{
+ ErtsThrPrgrWakeupCallback wake_me;
+ ErtsSchedulerData *esdp = vesdp;
+ Uint no = esdp->no;
#ifdef ERTS_SMP
- Uint no = ((ErtsSchedulerData *) vesdp)->no;
+ wake_me.wakeup = poke_ssi;
+ wake_me.arg = (void *) esdp->ssi;
+ erts_thr_progress_register_managed_thread(esdp, &wake_me, 0);
+ erts_alloc_register_scheduler(vesdp);
#endif
#ifdef ERTS_ENABLE_LOCK_CHECK
{
@@ -3730,11 +4347,13 @@ sched_thread_func(void *vesdp)
erts_lc_set_thread_name(&buf[0]);
}
#endif
- erts_alloc_reg_scheduler_id(no);
erts_tsd_set(sched_data_key, vesdp);
#ifdef ERTS_SMP
+#if HAVE_ERTS_MSEG
+ erts_mseg_late_init();
+#endif
- erts_sched_init_check_cpu_bind((ErtsSchedulerData *) vesdp);
+ erts_sched_init_check_cpu_bind(esdp);
erts_proc_lock_prepare_proc_lock_waiter();
ERTS_SCHED_SLEEP_INFO_IX(no - 1)->event = erts_tse_fetch();
@@ -3754,17 +4373,18 @@ sched_thread_func(void *vesdp)
if (--schdlr_sspnd.curr_online == schdlr_sspnd.wait_curr_online) {
erts_smp_atomic32_read_band_nob(&schdlr_sspnd.changing,
~ERTS_SCHDLR_SSPND_CHNG_ONLN);
- if (((ErtsSchedulerData *) vesdp)->no != 1)
+ if (no != 1)
erts_smp_cnd_signal(&schdlr_sspnd.cnd);
}
- if (((ErtsSchedulerData *) vesdp)->no == 1) {
+ if (no == 1) {
if (schdlr_sspnd.curr_online != schdlr_sspnd.wait_curr_online) {
erts_smp_activity_begin(ERTS_ACTIVITY_WAIT,
susp_sched_prep_block,
susp_sched_resume_block,
NULL);
- while (schdlr_sspnd.curr_online != schdlr_sspnd.wait_curr_online)
+ while (schdlr_sspnd.curr_online != schdlr_sspnd.wait_curr_online
+ || !schdlr_sspnd.aux_thread)
erts_smp_cnd_wait(&schdlr_sspnd.cnd, &schdlr_sspnd.mtx);
erts_smp_activity_end(ERTS_ACTIVITY_WAIT,
susp_sched_prep_block,
@@ -3776,19 +4396,22 @@ sched_thread_func(void *vesdp)
erts_smp_mtx_unlock(&schdlr_sspnd.mtx);
#ifdef ERTS_DO_VERIFY_UNUSED_TEMP_ALLOC
- ((ErtsSchedulerData *) vesdp)->verify_unused_temp_alloc
+ esdp->verify_unused_temp_alloc
= erts_alloc_get_verify_unused_temp_alloc(
- &((ErtsSchedulerData *) vesdp)->verify_unused_temp_alloc_data);
+ &esdp->verify_unused_temp_alloc_data);
ERTS_VERIFY_UNUSED_TEMP_ALLOC(NULL);
#endif
process_main();
/* No schedulers should *ever* terminate */
- erl_exit(ERTS_ABORT_EXIT, "Scheduler thread number %beu terminated\n",
- ((ErtsSchedulerData *) vesdp)->no);
+ erl_exit(ERTS_ABORT_EXIT,
+ "Scheduler thread number %beu terminated\n",
+ no);
return NULL;
}
+static ethr_tid aux_tid;
+
void
erts_start_schedulers(void)
{
@@ -3810,6 +4433,10 @@ erts_start_schedulers(void)
erts_block_system(0);
+ res = ethr_thr_create(&aux_tid, aux_thread, NULL, &opts);
+ if (res != 0)
+ erl_exit(1, "Failed to create aux thread\n");
+
while (actual < wanted) {
ErtsSchedulerData *esdp = ERTS_SCHEDULER_IX(actual);
actual++;
@@ -5355,23 +5982,19 @@ Process *schedule(Process *p, int calls)
}
}
-#if defined(ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK) \
- || defined(ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK)
{
- ErtsSchedulerSleepInfo *ssi = esdp->ssi;
- erts_aint32_t aux_work = erts_smp_atomic32_read_nob(&ssi->aux_work);
- if (aux_work) {
+ erts_aint32_t aux_work;
+ int leader_update = erts_thr_progress_update(esdp);
+ aux_work = erts_atomic32_read_acqb(&esdp->ssi->aux_work);
+ if (aux_work | leader_update) {
erts_smp_runq_unlock(rq);
-#ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK
- aux_work = blockable_aux_work(esdp, ssi, aux_work);
-#endif
-#ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK
- nonblockable_aux_work(esdp, ssi, aux_work);
-#endif
+ if (leader_update)
+ erts_thr_progress_leader_update(esdp);
+ if (aux_work)
+ handle_aux_work(&esdp->aux_work_data, aux_work);
erts_smp_runq_lock(rq);
}
}
-#endif
erts_smp_chk_system_block(prepare_for_block,
resume_after_block,
@@ -5380,17 +6003,21 @@ Process *schedule(Process *p, int calls)
ERTS_SMP_LC_ASSERT(!ERTS_LC_IS_BLOCKING);
ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq));
-#endif
+#else /* ERTS_SMP */
+ {
+ erts_aint32_t aux_work;
+ aux_work = erts_atomic32_read_acqb(&esdp->ssi->aux_work);
+ if (aux_work)
+ handle_aux_work(&esdp->aux_work_data, aux_work);
+ }
+#endif /* ERTS_SMP */
ASSERT(rq->len == rq->procs.len + rq->ports.info.len);
-#ifndef ERTS_SMP
+ if (rq->len == 0 && !rq->misc.start) {
- if (rq->len == 0 && !rq->misc.start)
- goto do_sys_schedule;
+#ifdef ERTS_SMP
-#else /* ERTS_SMP */
- if (rq->len == 0 && !rq->misc.start) {
ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq));
rq->wakeup_other = 0;
@@ -5421,26 +6048,17 @@ Process *schedule(Process *p, int calls)
}
}
+#endif
+
scheduler_wait(&fcalls, esdp, rq);
+#ifdef ERTS_SMP
non_empty_runq(rq);
+#endif
goto check_activities_to_run;
}
- else
-#endif /* ERTS_SMP */
- if (fcalls > input_reductions && prepare_for_sys_schedule()) {
- int runnable;
-
-#ifdef ERTS_SMP
- runnable = 1;
-#else
- do_sys_schedule:
- runnable = rq->len != 0;
- if (!runnable)
- sched_waiting_sys(esdp->no, rq);
-#endif
-
+ else if (fcalls > input_reductions && prepare_for_sys_schedule()) {
/*
* Schedule system-level activities.
*/
@@ -5450,11 +6068,11 @@ Process *schedule(Process *p, int calls)
ASSERT(!erts_port_task_have_outstanding_io_tasks());
-#ifdef ERTS_SMP
- /* erts_sys_schedule_interrupt(0); */
+#if 0 /* Not needed since we wont wait in sys schedule */
+ erts_sys_schedule_interrupt(0);
#endif
erts_smp_runq_unlock(rq);
- erl_sys_schedule(runnable);
+ erl_sys_schedule(1);
dt = erts_do_time_read_and_reset();
if (dt) erts_bump_timer(dt);
#ifdef ERTS_SMP
@@ -5462,8 +6080,6 @@ Process *schedule(Process *p, int calls)
clear_sys_scheduling();
goto continue_check_activities_to_run;
#else
- if (!runnable)
- sched_active_sys(esdp->no, rq);
goto check_activities_to_run;
#endif
}
@@ -5778,20 +6394,9 @@ erts_sched_stat_term(Process *p, int total)
void
erts_schedule_misc_op(void (*func)(void *), void *arg)
{
- ErtsRunQueue *rq;
- ErtsMiscOpList *molp = misc_op_list_alloc();
ErtsSchedulerData *esdp = erts_get_scheduler_data();
-
- if (esdp) {
- rq = esdp->run_queue;
- } else {
- /*
- * This can only happen when the sys msg dispatcher
- * thread schedules misc ops (this happens *very*
- * seldom; only when trace drivers are unloaded).
- */
- rq = ERTS_RUNQ_IX(0);
- }
+ ErtsRunQueue *rq = esdp ? esdp->run_queue : ERTS_RUNQ_IX(0);
+ ErtsMiscOpList *molp = misc_op_list_alloc();
erts_smp_runq_lock(rq);