diff options
Diffstat (limited to 'erts/emulator/beam/erl_process.c')
| -rw-r--r-- | erts/emulator/beam/erl_process.c | 2772 |
1 files changed, 486 insertions, 2286 deletions
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index b47ce97c46..21ee72edf2 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2010. All Rights Reserved. + * Copyright Ericsson AB 1996-2011. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in @@ -24,7 +24,6 @@ #endif #include <stddef.h> /* offsetof() */ -#include <ctype.h> #include "sys.h" #include "erl_vm.h" #include "global.h" @@ -39,6 +38,7 @@ #include "erl_threads.h" #include "erl_binary.h" #include "beam_bp.h" +#include "erl_cpu_topology.h" #define ERTS_RUNQ_CHECK_BALANCE_REDS_PER_SCHED (2000*CONTEXT_REDS) #define ERTS_RUNQ_CALL_CHECK_BALANCE_REDS \ @@ -63,8 +63,6 @@ #define ERTS_WAKEUP_OTHER_DEC 10 #define ERTS_WAKEUP_OTHER_FIXED_INC (CONTEXT_REDS/10) -#define ERTS_MAX_CPU_TOPOLOGY_ID ((int) 0xffff) - #if 0 || defined(DEBUG) #define ERTS_FAKE_SCHED_BIND_PRINT_SORTED_CPU_DATA #endif @@ -119,10 +117,6 @@ Uint erts_process_tab_index_mask; static int wakeup_other_limit; -#ifdef ERTS_SMP -Uint erts_max_main_threads; -#endif - int erts_sched_thread_suggested_stack_size = -1; #ifdef ERTS_ENABLE_LOCK_CHECK @@ -133,21 +127,22 @@ ErtsLcPSDLocks erts_psd_required_locks[ERTS_PSD_SIZE]; int erts_disable_proc_not_running_opt; -#define ERTS_SCHDLR_SSPND_CHNG_WAITER (((long) 1) << 0) -#define ERTS_SCHDLR_SSPND_CHNG_MSB (((long) 1) << 1) -#define ERTS_SCHDLR_SSPND_CHNG_ONLN (((long) 1) << 2) +#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) #ifndef DEBUG #define ERTS_SCHDLR_SSPND_CHNG_SET(VAL, OLD_VAL) \ - erts_smp_atomic_set(&schdlr_sspnd.changing, (VAL)) + erts_smp_atomic32_set(&schdlr_sspnd.changing, (VAL)) #else #define ERTS_SCHDLR_SSPND_CHNG_SET(VAL, OLD_VAL) \ do { \ - long old_val__ = erts_smp_atomic_xchg(&schdlr_sspnd.changing, \ - (VAL)); \ + erts_aint32_t old_val__; \ + old_val__ = erts_smp_atomic32_xchg(&schdlr_sspnd.changing, \ + (VAL)); \ ASSERT(old_val__ == (OLD_VAL)); \ } while (0) @@ -160,10 +155,10 @@ static struct { int online; int curr_online; int wait_curr_online; - erts_smp_atomic_t changing; - erts_smp_atomic_t active; + erts_smp_atomic32_t changing; + erts_smp_atomic32_t active; struct { - erts_smp_atomic_t ongoing; + erts_smp_atomic32_t ongoing; long wait_active; ErtsProcList *procs; } msb; /* Multi Scheduling Block */ @@ -171,11 +166,11 @@ static struct { static struct { erts_smp_mtx_t update_mtx; - erts_smp_atomic_t active_runqs; + erts_smp_atomic32_t active_runqs; int last_active_runqs; - erts_smp_atomic_t used_runqs; + erts_smp_atomic32_t used_runqs; int forced_check_balance; - erts_smp_atomic_t checking_balance; + erts_smp_atomic32_t checking_balance; int halftime; int full_reds_history_index; struct { @@ -195,48 +190,6 @@ do { \ #endif -/* - * Cpu topology hierarchy. - */ -#define ERTS_TOPOLOGY_NODE 0 -#define ERTS_TOPOLOGY_PROCESSOR 1 -#define ERTS_TOPOLOGY_PROCESSOR_NODE 2 -#define ERTS_TOPOLOGY_CORE 3 -#define ERTS_TOPOLOGY_THREAD 4 -#define ERTS_TOPOLOGY_LOGICAL 5 - -#define ERTS_TOPOLOGY_MAX_DEPTH 6 - -typedef struct { - int bind_id; - int bound_id; -} ErtsCpuBindData; - -static ErtsCpuBindData *scheduler2cpu_map; -erts_smp_rwmtx_t erts_cpu_bind_rwmtx; - -typedef enum { - ERTS_CPU_BIND_UNDEFINED, - ERTS_CPU_BIND_SPREAD, - ERTS_CPU_BIND_PROCESSOR_SPREAD, - ERTS_CPU_BIND_THREAD_SPREAD, - ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD, - ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD, - ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD, - ERTS_CPU_BIND_NO_SPREAD, - ERTS_CPU_BIND_NONE -} ErtsCpuBindOrder; - -#define ERTS_CPU_BIND_DEFAULT_BIND \ - ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD - -ErtsCpuBindOrder cpu_bind_order; - -static erts_cpu_topology_t *user_cpudata; -static int user_cpudata_size; -static erts_cpu_topology_t *system_cpudata; -static int system_cpudata_size; - erts_sched_stat_t erts_sched_stat; ErtsRunQueue *erts_common_run_queue; @@ -247,11 +200,11 @@ static erts_tsd_key_t sched_data_key; static erts_smp_mtx_t proc_tab_mtx; -static erts_smp_atomic_t function_calls; +static erts_smp_atomic32_t function_calls; #ifdef ERTS_SMP -static erts_smp_atomic_t doing_sys_schedule; -static erts_smp_atomic_t no_empty_run_queues; +static erts_smp_atomic32_t doing_sys_schedule; +static erts_smp_atomic32_t no_empty_run_queues; #else /* !ERTS_SMP */ ErtsSchedulerData *erts_scheduler_data; #endif @@ -259,11 +212,6 @@ ErtsSchedulerData *erts_scheduler_data; ErtsAlignedRunQueue *erts_aligned_run_queues; Uint erts_no_run_queues; -typedef union { - ErtsSchedulerData esd; - char align[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(ErtsSchedulerData))]; -} ErtsAlignedSchedulerData; - ErtsAlignedSchedulerData *erts_aligned_scheduler_data; #ifdef ERTS_SMP @@ -300,7 +248,10 @@ Uint erts_num_active_procs; Process** erts_active_procs; #endif -static erts_smp_atomic_t process_count; +#if ERTS_MAX_PROCESSES > 0x7fffffff +#error "Need to store process_count in another type" +#endif +static erts_smp_atomic32_t process_count; typedef struct ErtsTermProcElement_ ErtsTermProcElement; struct ErtsTermProcElement_ { @@ -334,12 +285,6 @@ ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(proclist, 200, ERTS_ALC_T_PROC_LIST) -#define ERTS_RUNQ_IX(IX) \ - (ASSERT_EXPR(0 <= (IX) && (IX) < erts_no_run_queues), \ - &erts_aligned_run_queues[(IX)].runq) -#define ERTS_SCHEDULER_IX(IX) \ - (ASSERT_EXPR(0 <= (IX) && (IX) < erts_no_schedulers), \ - &erts_aligned_scheduler_data[(IX)].esd) #define ERTS_SCHED_SLEEP_INFO_IX(IX) \ (ASSERT_EXPR(0 <= (IX) && (IX) < erts_no_schedulers), \ &aligned_sched_sleep_info[(IX)].ssi) @@ -398,22 +343,8 @@ static int stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp, #ifdef ERTS_SMP static void handle_pending_exiters(ErtsProcList *); -static void cpu_bind_order_sort(erts_cpu_topology_t *cpudata, - int size, - ErtsCpuBindOrder bind_order, - int mk_seq); -static void signal_schedulers_bind_change(erts_cpu_topology_t *cpudata, int size); - #endif -static int reader_group_lookup(int logical); -static void create_tmp_cpu_topology_copy(erts_cpu_topology_t **cpudata, - int *cpudata_size); -static void destroy_tmp_cpu_topology_copy(erts_cpu_topology_t *cpudata); - -static void early_cpu_bind_init(void); -static void late_cpu_bind_init(void); - #if defined(ERTS_SMP) && defined(ERTS_ENABLE_LOCK_CHECK) int erts_smp_lc_runq_is_locked(ErtsRunQueue *runq) @@ -469,18 +400,18 @@ erts_pre_init_process(void) /* initialize the scheduler */ void -erts_init_process(void) +erts_init_process(int ncpu) { Uint proc_bits = ERTS_PROC_BITS; #ifdef ERTS_SMP erts_disable_proc_not_running_opt = 0; - erts_init_proc_lock(); + erts_init_proc_lock(ncpu); #endif init_proclist_alloc(); - erts_smp_atomic_init(&process_count, 0); + erts_smp_atomic32_init(&process_count, 0); if (erts_use_r9_pids_ports) { proc_bits = ERTS_R9_PROC_BITS; @@ -641,7 +572,7 @@ erts_psd_set_init(Process *p, ErtsProcLocks plocks, int ix, void *data) #ifdef ERTS_SMP void -erts_sched_finish_poke(ErtsSchedulerSleepInfo *ssi, long flags) +erts_sched_finish_poke(ErtsSchedulerSleepInfo *ssi, erts_aint32_t flags) { switch (flags & ERTS_SSI_FLGS_SLEEP_TYPE) { case ERTS_SSI_FLG_POLL_SLEEPING: @@ -659,6 +590,118 @@ erts_sched_finish_poke(ErtsSchedulerSleepInfo *ssi, long flags) } } +typedef struct erts_misc_aux_work_t_ erts_misc_aux_work_t; +struct erts_misc_aux_work_t_ { + erts_misc_aux_work_t *next; + void (*func)(void *); + void *arg; +}; + +typedef struct { + erts_smp_mtx_t mtx; + erts_misc_aux_work_t *first; + erts_misc_aux_work_t *last; +} erts_misc_aux_work_q_t; + +typedef union { + erts_misc_aux_work_q_t data; + char align[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(erts_misc_aux_work_q_t))]; +} erts_algnd_misc_aux_work_q_t; + +static erts_algnd_misc_aux_work_q_t *misc_aux_work_queues; + +ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(misc_aux_work, + erts_misc_aux_work_t, + 200, + ERTS_ALC_T_MISC_AUX_WORK) + +static void +init_misc_aux_work(void) +{ + int ix; + + init_misc_aux_work_alloc(); + + 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)); + + 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)); + misc_aux_work_queues[ix].data.first = NULL; + misc_aux_work_queues[ix].data.last = NULL; + } +} + +static void +handle_misc_aux_work(ErtsSchedulerData *esdp) +{ + int ix = (int) esdp->no - 1; + erts_misc_aux_work_t *mawp; + + 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; + misc_aux_work_queues[ix].data.last = NULL; + erts_smp_mtx_unlock(&misc_aux_work_queues[ix].data.mtx); + + while (mawp) { + erts_misc_aux_work_t *free_mawp; + mawp->func(mawp->arg); + free_mawp = mawp; + mawp = mawp->next; + misc_aux_work_free(free_mawp); + } +} + +void +erts_smp_schedule_misc_aux_work(int ignore_self, + int max_sched, + void (*func)(void *), + void *arg) +{ + int ix, ignore_ix = -1; + + if (ignore_self) { + ErtsSchedulerData *esdp = erts_get_scheduler_data(); + if (esdp) + ignore_ix = (int) esdp->no - 1; + } + + 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; + if (ix == ignore_ix) + continue; + + 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); + + ssi = ERTS_SCHED_SLEEP_INFO_IX(ix); + aux_work = erts_smp_atomic32_bor(&ssi->aux_work, + ERTS_SSI_AUX_WORK_MISC); + if ((aux_work & ERTS_SSI_AUX_WORK_MISC) == 0) + erts_sched_poke(ssi); + } +} + #ifdef ERTS_SMP_SCHEDULERS_NEED_TO_CHECK_CHILDREN void erts_smp_notify_check_children_needed(void) @@ -666,11 +709,11 @@ erts_smp_notify_check_children_needed(void) int i; for (i = 0; i < erts_no_schedulers; i++) { - long aux_work; + erts_aint32_t aux_work; ErtsSchedulerSleepInfo *ssi; ssi = ERTS_SCHED_SLEEP_INFO_IX(i); - aux_work = erts_smp_atomic_bor(&ssi->aux_work, - ERTS_SSI_AUX_WORK_CHECK_CHILDREN); + aux_work = erts_smp_atomic32_bor(&ssi->aux_work, + ERTS_SSI_AUX_WORK_CHECK_CHILDREN); if (!(aux_work & ERTS_SSI_AUX_WORK_CHECK_CHILDREN)) erts_sched_poke(ssi); } @@ -678,16 +721,22 @@ erts_smp_notify_check_children_needed(void) #endif #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK -static ERTS_INLINE long +static ERTS_INLINE erts_aint32_t blockable_aux_work(ErtsSchedulerData *esdp, ErtsSchedulerSleepInfo *ssi, - long aux_work) + 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_band(&ssi->aux_work, + ~ERTS_SSI_AUX_WORK_MISC); + aux_work &= ~ERTS_SSI_AUX_WORK_MISC; + handle_misc_aux_work(esdp); + } #ifdef ERTS_SMP_SCHEDULERS_NEED_TO_CHECK_CHILDREN if (aux_work & ERTS_SSI_AUX_WORK_CHECK_CHILDREN) { - aux_work = erts_smp_atomic_band(&ssi->aux_work, - ~ERTS_SSI_AUX_WORK_CHECK_CHILDREN); + aux_work = erts_smp_atomic32_band(&ssi->aux_work, + ~ERTS_SSI_AUX_WORK_CHECK_CHILDREN); aux_work &= ~ERTS_SSI_AUX_WORK_CHECK_CHILDREN; erts_check_children(); } @@ -699,10 +748,10 @@ blockable_aux_work(ErtsSchedulerData *esdp, #endif #ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK -static ERTS_INLINE long +static ERTS_INLINE erts_aint32_t nonblockable_aux_work(ErtsSchedulerData *esdp, ErtsSchedulerSleepInfo *ssi, - long aux_work) + erts_aint32_t aux_work) { if (aux_work & ERTS_SSI_NONBLOCKABLE_AUX_WORK_MASK) { @@ -762,15 +811,31 @@ erts_active_schedulers(void) return as; } +#ifdef ERTS_SMP + +static ERTS_INLINE void +clear_sys_scheduling(void) +{ + erts_smp_atomic32_set_relb(&doing_sys_schedule, 0); +} + +static ERTS_INLINE int +try_set_sys_scheduling(void) +{ + return 0 == erts_smp_atomic32_cmpxchg_acqb(&doing_sys_schedule, 1, 0); +} + +#endif + static ERTS_INLINE int prepare_for_sys_schedule(void) { #ifdef ERTS_SMP while (!erts_port_task_have_outstanding_io_tasks() - && !erts_smp_atomic_xchg(&doing_sys_schedule, 1)) { + && try_set_sys_scheduling()) { if (!erts_port_task_have_outstanding_io_tasks()) return 1; - erts_smp_atomic_set(&doing_sys_schedule, 0); + clear_sys_scheduling(); } return 0; #else @@ -818,53 +883,55 @@ sched_active(Uint no, ErtsRunQueue *rq) static int ERTS_INLINE ongoing_multi_scheduling_block(void) { - return erts_smp_atomic_read(&schdlr_sspnd.msb.ongoing) != 0; + return erts_smp_atomic32_read(&schdlr_sspnd.msb.ongoing) != 0; } static ERTS_INLINE void empty_runq(ErtsRunQueue *rq) { - long oifls = erts_smp_atomic_band(&rq->info_flags, ~ERTS_RUNQ_IFLG_NONEMPTY); + erts_aint32_t oifls = erts_smp_atomic32_band(&rq->info_flags, + ~ERTS_RUNQ_IFLG_NONEMPTY); if (oifls & ERTS_RUNQ_IFLG_NONEMPTY) { #ifdef DEBUG - long empty = erts_smp_atomic_read(&no_empty_run_queues); + erts_aint32_t empty = erts_smp_atomic32_read(&no_empty_run_queues); /* * For a short period of time no_empty_run_queues may have * been increased twice for a specific run queue. */ ASSERT(0 <= empty && empty < 2*erts_no_run_queues); #endif - erts_smp_atomic_inc(&no_empty_run_queues); + erts_smp_atomic32_inc(&no_empty_run_queues); } } static ERTS_INLINE void non_empty_runq(ErtsRunQueue *rq) { - long oifls = erts_smp_atomic_bor(&rq->info_flags, ERTS_RUNQ_IFLG_NONEMPTY); + erts_aint32_t oifls = erts_smp_atomic32_bor(&rq->info_flags, + ERTS_RUNQ_IFLG_NONEMPTY); if (!(oifls & ERTS_RUNQ_IFLG_NONEMPTY)) { #ifdef DEBUG - long empty = erts_smp_atomic_read(&no_empty_run_queues); + erts_aint32_t empty = erts_smp_atomic32_read(&no_empty_run_queues); /* * For a short period of time no_empty_run_queues may have * been increased twice for a specific run queue. */ ASSERT(0 < empty && empty <= 2*erts_no_run_queues); #endif - erts_smp_atomic_dec(&no_empty_run_queues); + erts_smp_atomic32_dec(&no_empty_run_queues); } } -static long +static erts_aint32_t sched_prep_spin_wait(ErtsSchedulerSleepInfo *ssi) { - long oflgs; - long nflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_WAITING); - long xflgs = 0; + erts_aint32_t oflgs; + erts_aint32_t nflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_WAITING); + erts_aint32_t xflgs = 0; do { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return nflgs; xflgs = oflgs; @@ -872,16 +939,16 @@ sched_prep_spin_wait(ErtsSchedulerSleepInfo *ssi) return oflgs; } -static long +static erts_aint32_t sched_prep_cont_spin_wait(ErtsSchedulerSleepInfo *ssi) { - long oflgs; - long nflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_WAITING); - long xflgs = ERTS_SSI_FLG_WAITING; + erts_aint32_t oflgs; + erts_aint32_t nflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_WAITING); + erts_aint32_t xflgs = ERTS_SSI_FLG_WAITING; do { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return nflgs; xflgs = oflgs; @@ -890,15 +957,15 @@ sched_prep_cont_spin_wait(ErtsSchedulerSleepInfo *ssi) return oflgs; } -static long +static erts_aint32_t sched_spin_wait(ErtsSchedulerSleepInfo *ssi, int spincount) { - long until_yield = ERTS_SCHED_SPIN_UNTIL_YIELD; + int until_yield = ERTS_SCHED_SPIN_UNTIL_YIELD; int sc = spincount; - long flgs; + erts_aint32_t flgs; do { - flgs = erts_smp_atomic_read(&ssi->flags); + flgs = erts_smp_atomic32_read(&ssi->flags); if ((flgs & (ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING)) != (ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING)) { break; @@ -912,18 +979,18 @@ sched_spin_wait(ErtsSchedulerSleepInfo *ssi, int spincount) return flgs; } -static long -sched_set_sleeptype(ErtsSchedulerSleepInfo *ssi, long sleep_type) +static erts_aint32_t +sched_set_sleeptype(ErtsSchedulerSleepInfo *ssi, erts_aint32_t sleep_type) { - long oflgs; - long nflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING|sleep_type; - long xflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING; + erts_aint32_t oflgs; + erts_aint32_t nflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING|sleep_type; + erts_aint32_t xflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING; if (sleep_type == ERTS_SSI_FLG_TSE_SLEEPING) erts_tse_reset(ssi->event); while (1) { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return nflgs; if ((oflgs & (ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING)) @@ -940,14 +1007,14 @@ sched_set_sleeptype(ErtsSchedulerSleepInfo *ssi, long sleep_type) != ERTS_SSI_FLG_WAITING) static void -scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) +scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) { ErtsSchedulerSleepInfo *ssi = esdp->ssi; int spincount; - long flgs; + erts_aint32_t flgs; #if defined(ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK) \ || defined(ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK) - long aux_work; + erts_aint32_t aux_work; #endif ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); @@ -983,7 +1050,7 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) tse_wait: #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); tse_blockable_aux_work: aux_work = blockable_aux_work(esdp, ssi, aux_work); #endif @@ -993,7 +1060,7 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) #ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK #ifndef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); #endif nonblockable_aux_work(esdp, ssi, aux_work); #endif @@ -1026,7 +1093,7 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) } #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&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; @@ -1038,16 +1105,16 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL); if (flgs & ~ERTS_SSI_FLG_SUSPENDED) - erts_smp_atomic_band(&ssi->flags, ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_band(&ssi->flags, ERTS_SSI_FLG_SUSPENDED); erts_smp_runq_lock(rq); sched_active(esdp->no, rq); } else { - long dt; + erts_aint_t dt; - erts_smp_atomic_set(&function_calls, 0); + erts_smp_atomic32_set(&function_calls, 0); *fcalls = 0; sched_waiting_sys(esdp->no, rq); @@ -1060,25 +1127,27 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) sys_poll_aux_work: + ASSERT(!erts_port_task_have_outstanding_io_tasks()); + erl_sys_schedule(1); /* Might give us something to do */ - dt = do_time_read_and_reset(); - if (dt) bump_timer(dt); + dt = erts_do_time_read_and_reset(); + if (dt) erts_bump_timer(dt); sys_aux_work: #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); aux_work = blockable_aux_work(esdp, ssi, aux_work); #endif #ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK #ifndef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&ssi->aux_work); #endif nonblockable_aux_work(esdp, ssi, aux_work); #endif - flgs = erts_smp_atomic_read(&ssi->flags); + flgs = erts_smp_atomic32_read(&ssi->flags); if (!(flgs & ERTS_SSI_FLG_WAITING)) { ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING)); goto sys_woken; @@ -1096,7 +1165,7 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * call erl_sys_schedule() until it is handled. */ if (erts_port_task_have_outstanding_io_tasks()) { - erts_smp_atomic_set(&doing_sys_schedule, 0); + clear_sys_scheduling(); /* * Got to check that we still got I/O tasks; otherwise * we have to continue checking for I/O... @@ -1115,7 +1184,7 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) * sleep in erl_sys_schedule(). */ if (erts_port_task_have_outstanding_io_tasks()) { - erts_smp_atomic_set(&doing_sys_schedule, 0); + clear_sys_scheduling(); /* * Got to check that we still got I/O tasks; otherwise @@ -1155,10 +1224,12 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) erts_smp_runq_unlock(rq); + ASSERT(!erts_port_task_have_outstanding_io_tasks()); + erl_sys_schedule(0); - dt = do_time_read_and_reset(); - if (dt) 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) @@ -1167,9 +1238,9 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) sys_woken: erts_smp_runq_lock(rq); sys_locked_woken: - erts_smp_atomic_set(&doing_sys_schedule, 0); + clear_sys_scheduling(); if (flgs & ~ERTS_SSI_FLG_SUSPENDED) - erts_smp_atomic_band(&ssi->flags, ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_band(&ssi->flags, ERTS_SSI_FLG_SUSPENDED); sched_active_sys(esdp->no, rq); } } @@ -1177,15 +1248,15 @@ scheduler_wait(long *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(rq)); } -static ERTS_INLINE long +static ERTS_INLINE erts_aint32_t ssi_flags_set_wake(ErtsSchedulerSleepInfo *ssi) { /* reset all flags but suspended */ - long oflgs; - long nflgs = 0; - long xflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING; + erts_aint32_t oflgs; + erts_aint32_t nflgs = 0; + erts_aint32_t xflgs = ERTS_SSI_FLG_SLEEPING|ERTS_SSI_FLG_WAITING; while (1) { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return oflgs; nflgs = oflgs & ERTS_SSI_FLG_SUSPENDED; @@ -1196,7 +1267,6 @@ ssi_flags_set_wake(ErtsSchedulerSleepInfo *ssi) static void wake_scheduler(ErtsRunQueue *rq, int incq, int one) { - int res; ErtsSchedulerSleepInfo *ssi; ErtsSchedulerSleepList *sl; @@ -1217,7 +1287,7 @@ wake_scheduler(ErtsRunQueue *rq, int incq, int one) if (!ssi) erts_smp_spin_unlock(&sl->lock); else if (one) { - long flgs; + erts_aint32_t flgs; if (ssi->prev) ssi->prev->next = ssi->next; else { @@ -1227,9 +1297,9 @@ wake_scheduler(ErtsRunQueue *rq, int incq, int one) if (ssi->next) ssi->next->prev = ssi->prev; - res = sl->list != NULL; erts_smp_spin_unlock(&sl->lock); + ERTS_THR_MEMORY_BARRIER; flgs = ssi_flags_set_wake(ssi); erts_sched_finish_poke(ssi, flgs); @@ -1239,10 +1309,12 @@ wake_scheduler(ErtsRunQueue *rq, int incq, int one) else { sl->list = NULL; erts_smp_spin_unlock(&sl->lock); + + ERTS_THR_MEMORY_BARRIER; do { ErtsSchedulerSleepInfo *wake_ssi = ssi; ssi = ssi->next; - erts_sched_finish_poke(ssi, ssi_flags_set_wake(wake_ssi)); + erts_sched_finish_poke(wake_ssi, ssi_flags_set_wake(wake_ssi)); } while (ssi); } } @@ -1264,15 +1336,17 @@ wake_all_schedulers(void) static ERTS_INLINE int chk_wake_sched(ErtsRunQueue *crq, int ix, int activate) { - long iflgs; + erts_aint32_t iflgs; ErtsRunQueue *wrq; if (crq->ix == ix) return 0; wrq = ERTS_RUNQ_IX(ix); - iflgs = erts_smp_atomic_read(&wrq->info_flags); + iflgs = erts_smp_atomic32_read(&wrq->info_flags); if (!(iflgs & (ERTS_RUNQ_IFLG_SUSPENDED|ERTS_RUNQ_IFLG_NONEMPTY))) { if (activate) { - if (ix == erts_smp_atomic_cmpxchg(&balance_info.active_runqs, ix+1, ix)) { + if (ix == erts_smp_atomic32_cmpxchg(&balance_info.active_runqs, + ix+1, + ix)) { erts_smp_xrunq_lock(crq, wrq); wrq->flags &= ~ERTS_RUNQ_FLG_INACTIVE; erts_smp_xrunq_unlock(crq, wrq); @@ -1289,8 +1363,8 @@ wake_scheduler_on_empty_runq(ErtsRunQueue *crq) { int ix = crq->ix; int stop_ix = ix; - int active_ix = erts_smp_atomic_read(&balance_info.active_runqs); - int balance_ix = erts_smp_atomic_read(&balance_info.used_runqs); + int active_ix = erts_smp_atomic32_read(&balance_info.active_runqs); + int balance_ix = erts_smp_atomic32_read(&balance_info.used_runqs); if (active_ix > balance_ix) active_ix = balance_ix; @@ -1335,6 +1409,31 @@ erts_smp_notify_inc_runq(ErtsRunQueue *runq) smp_notify_inc_runq(runq); } +void +erts_sched_notify_check_cpu_bind(void) +{ +#ifdef ERTS_SMP + int ix; + if (erts_common_run_queue) { + for (ix = 0; ix < erts_no_schedulers; ix++) + erts_smp_atomic32_set(&ERTS_SCHEDULER_IX(ix)->chk_cpu_bind, 1); + wake_all_schedulers(); + } + else { + for (ix = 0; ix < erts_no_run_queues; ix++) { + ErtsRunQueue *rq = ERTS_RUNQ_IX(ix); + erts_smp_runq_lock(rq); + rq->flags |= ERTS_RUNQ_FLG_CHK_CPU_BIND; + erts_smp_runq_unlock(rq); + wake_scheduler(rq, 0, 1); + }; + } +#else + erts_sched_check_cpu_bind(erts_get_scheduler_data()); +#endif +} + + #ifdef ERTS_SMP ErtsRunQueue * @@ -1485,14 +1584,15 @@ evacuate_run_queue(ErtsRunQueue *evac_rq, ErtsRunQueue *rq) erts_smp_runq_lock(evac_rq); - erts_smp_atomic_bor(&evac_rq->scheduler->ssi->flags, ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_bor(&evac_rq->scheduler->ssi->flags, + ERTS_SSI_FLG_SUSPENDED); evac_rq->flags &= ~ERTS_RUNQ_FLGS_IMMIGRATE_QMASK; evac_rq->flags |= (ERTS_RUNQ_FLGS_EMIGRATE_QMASK | ERTS_RUNQ_FLGS_EVACUATE_QMASK | ERTS_RUNQ_FLG_SUSPENDED); - erts_smp_atomic_bor(&evac_rq->info_flags, ERTS_RUNQ_IFLG_SUSPENDED); + erts_smp_atomic32_bor(&evac_rq->info_flags, ERTS_RUNQ_IFLG_SUSPENDED); /* * Need to set up evacuation paths first since we * may release the run queue lock on evac_rq @@ -1741,7 +1841,7 @@ static ERTS_INLINE int check_possible_steal_victim(ErtsRunQueue *rq, int *rq_lockedp, int vix) { ErtsRunQueue *vrq = ERTS_RUNQ_IX(vix); - long iflgs = erts_smp_atomic_read(&vrq->info_flags); + erts_aint32_t iflgs = erts_smp_atomic32_read(&vrq->info_flags); if (iflgs & ERTS_RUNQ_IFLG_NONEMPTY) return try_steal_task_from_victim(rq, rq_lockedp, vrq); else @@ -1771,8 +1871,8 @@ try_steal_task(ErtsRunQueue *rq) ERTS_SMP_LC_CHK_RUNQ_LOCK(rq, rq_locked); - active_rqs = erts_smp_atomic_read(&balance_info.active_runqs); - blnc_rqs = erts_smp_atomic_read(&balance_info.used_runqs); + active_rqs = erts_smp_atomic32_read(&balance_info.active_runqs); + blnc_rqs = erts_smp_atomic32_read(&balance_info.used_runqs); if (active_rqs > blnc_rqs) active_rqs = blnc_rqs; @@ -1783,7 +1883,7 @@ try_steal_task(ErtsRunQueue *rq) if (active_rqs < blnc_rqs) { int no = blnc_rqs - active_rqs; int stop_ix = vix = active_rqs + rq->ix % no; - while (erts_smp_atomic_read(&no_empty_run_queues) < blnc_rqs) { + while (erts_smp_atomic32_read(&no_empty_run_queues) < blnc_rqs) { res = check_possible_steal_victim(rq, &rq_locked, vix); if (res) goto done; @@ -1798,7 +1898,7 @@ try_steal_task(ErtsRunQueue *rq) vix = rq->ix; /* ... then try to steal a job from another active queue... */ - while (erts_smp_atomic_read(&no_empty_run_queues) < blnc_rqs) { + while (erts_smp_atomic32_read(&no_empty_run_queues) < blnc_rqs) { vix++; if (vix >= active_rqs) vix = 0; @@ -1886,20 +1986,23 @@ do { \ static void check_balance(ErtsRunQueue *c_rq) { +#if ERTS_MAX_PROCESSES >= (1 << 27) +# error check_balance() assumes ERTS_MAX_PROCESS < (1 << 27) +#endif ErtsRunQueueBalance avg = {0}; Sint64 scheds_reds, full_scheds_reds; int forced, active, current_active, oowc, half_full_scheds, full_scheds, mmax_len, blnc_no_rqs, qix, pix, freds_hist_ix; - if (erts_smp_atomic_xchg(&balance_info.checking_balance, 1)) { + if (erts_smp_atomic32_xchg(&balance_info.checking_balance, 1)) { c_rq->check_balance_reds = INT_MAX; return; } - blnc_no_rqs = (int) erts_smp_atomic_read(&balance_info.used_runqs); + blnc_no_rqs = (int) erts_smp_atomic32_read(&balance_info.used_runqs); if (blnc_no_rqs == 1) { c_rq->check_balance_reds = INT_MAX; - erts_smp_atomic_set(&balance_info.checking_balance, 0); + erts_smp_atomic32_set(&balance_info.checking_balance, 0); return; } @@ -1907,7 +2010,7 @@ check_balance(ErtsRunQueue *c_rq) if (balance_info.halftime) { balance_info.halftime = 0; - erts_smp_atomic_set(&balance_info.checking_balance, 0); + erts_smp_atomic32_set(&balance_info.checking_balance, 0); ERTS_FOREACH_RUNQ(rq, { if (rq->waiting) @@ -1935,12 +2038,12 @@ check_balance(ErtsRunQueue *c_rq) forced = balance_info.forced_check_balance; balance_info.forced_check_balance = 0; - blnc_no_rqs = (int) erts_smp_atomic_read(&balance_info.used_runqs); + blnc_no_rqs = (int) erts_smp_atomic32_read(&balance_info.used_runqs); if (blnc_no_rqs == 1) { erts_smp_mtx_unlock(&balance_info.update_mtx); erts_smp_runq_lock(c_rq); c_rq->check_balance_reds = INT_MAX; - erts_smp_atomic_set(&balance_info.checking_balance, 0); + erts_smp_atomic32_set(&balance_info.checking_balance, 0); return; } @@ -1949,7 +2052,7 @@ check_balance(ErtsRunQueue *c_rq) if (balance_info.full_reds_history_index >= ERTS_FULL_REDS_HISTORY_SIZE) balance_info.full_reds_history_index = 0; - current_active = erts_smp_atomic_read(&balance_info.active_runqs); + current_active = erts_smp_atomic32_read(&balance_info.active_runqs); /* Read balance information for all run queues */ for (qix = 0; qix < blnc_no_rqs; qix++) { @@ -2009,12 +2112,14 @@ check_balance(ErtsRunQueue *c_rq) run_queue_info[qix].prio[pix].avail = 0; } else { - int xreds = 0; - int procreds = treds; - procreds -= run_queue_info[qix].prio[ERTS_PORT_PRIO_LEVEL].reds; + Sint64 xreds = 0; + Sint64 procreds = treds; + procreds -= + ((Sint64) + run_queue_info[qix].prio[ERTS_PORT_PRIO_LEVEL].reds); for (pix = 0; pix < ERTS_NO_PROC_PRIO_LEVELS; pix++) { - int av; + Sint64 av; if (xreds == 0) av = 100; @@ -2025,9 +2130,10 @@ check_balance(ErtsRunQueue *c_rq) if (av == 0) av = 1; } - run_queue_info[qix].prio[pix].avail = av; + run_queue_info[qix].prio[pix].avail = (int) av; + ASSERT(run_queue_info[qix].prio[pix].avail >= 0); if (pix < PRIORITY_NORMAL) /* ie., max or high */ - xreds += run_queue_info[qix].prio[pix].reds; + xreds += (Sint64) run_queue_info[qix].prio[pix].reds; } run_queue_info[qix].prio[ERTS_PORT_PRIO_LEVEL].avail = 100; } @@ -2132,7 +2238,8 @@ check_balance(ErtsRunQueue *c_rq) if (max_len != 0) { int avail = avg.prio[pix].avail; if (avail != 0) { - max_len = ((100*max_len - 1) / avail) + 1; + max_len = (int) ((100*((Sint64) max_len) - 1) + / ((Sint64) avail)) + 1; avg.prio[pix].max_len = max_len; ASSERT(max_len >= 0); } @@ -2149,9 +2256,10 @@ check_balance(ErtsRunQueue *c_rq) || run_queue_info[qix].prio[pix].avail == 0) limit = 0; else - limit = (((avg.prio[pix].max_len - * run_queue_info[qix].prio[pix].avail) - 1) - / 100 + 1); + limit = (int) (((((Sint64) avg.prio[pix].max_len) + * ((Sint64) run_queue_info[qix].prio[pix].avail)) + - 1) + / 100 + 1); run_queue_info[qix].prio[pix].migration_limit = limit; } } @@ -2279,10 +2387,10 @@ erts_fprintf(stderr, "--------------------------------\n"); } balance_info.last_active_runqs = active; - erts_smp_atomic_set(&balance_info.active_runqs, active); + erts_smp_atomic32_set(&balance_info.active_runqs, active); balance_info.halftime = 1; - erts_smp_atomic_set(&balance_info.checking_balance, 0); + erts_smp_atomic32_set(&balance_info.checking_balance, 0); /* Write migration paths and reset balance statistics in all queues */ for (qix = 0; qix < blnc_no_rqs; qix++) { @@ -2379,7 +2487,6 @@ erts_debug_nbalance(void) void erts_early_init_scheduling(void) { - early_cpu_bind_init(); wakeup_other_limit = ERTS_WAKEUP_OTHER_LIMIT_MEDIUM; } @@ -2421,18 +2528,11 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) n = (int) (mrq ? no_schedulers : 1); - erts_aligned_run_queues = erts_alloc(ERTS_ALC_T_RUNQS, - (sizeof(ErtsAlignedRunQueue)*(n+1))); - if ((((UWord) erts_aligned_run_queues) & ERTS_CACHE_LINE_MASK) != 0) - erts_aligned_run_queues = ((ErtsAlignedRunQueue *) - ((((UWord) erts_aligned_run_queues) - & ~ERTS_CACHE_LINE_MASK) - + ERTS_CACHE_LINE_SIZE)); - - ASSERT((((UWord) erts_aligned_run_queues) & ERTS_CACHE_LINE_MASK) == 0); - + erts_aligned_run_queues = + erts_alloc_permanent_cache_aligned(ERTS_ALC_T_RUNQS, + sizeof(ErtsAlignedRunQueue) * n); #ifdef ERTS_SMP - erts_smp_atomic_init(&no_empty_run_queues, 0); + erts_smp_atomic32_init(&no_empty_run_queues, 0); #endif erts_no_run_queues = n; @@ -2442,7 +2542,7 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) ErtsRunQueue *rq = ERTS_RUNQ_IX(ix); rq->ix = ix; - erts_smp_atomic_init(&rq->info_flags, ERTS_RUNQ_IFLG_NONEMPTY); + erts_smp_atomic32_init(&rq->info_flags, ERTS_RUNQ_IFLG_NONEMPTY); /* make sure that the "extra" id correponds to the schedulers * id if the esdp->no <-> ix+1 mapping change. @@ -2525,38 +2625,27 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) #ifdef ERTS_SMP /* Create and initialize scheduler sleep info */ - aligned_sched_sleep_info = erts_alloc(ERTS_ALC_T_SCHDLR_SLP_INFO, - (sizeof(ErtsAlignedSchedulerSleepInfo) - *(n+1))); - if ((((UWord) aligned_sched_sleep_info) & ERTS_CACHE_LINE_MASK) == 0) - aligned_sched_sleep_info = ((ErtsAlignedSchedulerSleepInfo *) - ((((UWord) aligned_sched_sleep_info) - & ~ERTS_CACHE_LINE_MASK) - + ERTS_CACHE_LINE_SIZE)); + 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); #if 0 /* no need to initialize these... */ ssi->next = NULL; ssi->prev = NULL; #endif - erts_smp_atomic_init(&ssi->flags, 0); + erts_smp_atomic32_init(&ssi->flags, 0); ssi->event = NULL; /* initialized in sched_thread_func */ - erts_smp_atomic_init(&ssi->aux_work, 0); + erts_smp_atomic32_init(&ssi->aux_work, 0); } #endif /* Create and initialize scheduler specific data */ - erts_aligned_scheduler_data = erts_alloc(ERTS_ALC_T_SCHDLR_DATA, - (sizeof(ErtsAlignedSchedulerData) - *(n+1))); - if ((((UWord) erts_aligned_scheduler_data) & ERTS_CACHE_LINE_MASK) != 0) - erts_aligned_scheduler_data = ((ErtsAlignedSchedulerData *) - ((((UWord) erts_aligned_scheduler_data) - & ~ERTS_CACHE_LINE_MASK) - + ERTS_CACHE_LINE_SIZE)); - - ASSERT((((UWord) erts_aligned_scheduler_data) & ERTS_CACHE_LINE_MASK) == 0); + erts_aligned_scheduler_data = + erts_alloc_permanent_cache_aligned(ERTS_ALC_T_SCHDLR_DATA, + n*sizeof(ErtsAlignedSchedulerData)); for (ix = 0; ix < n; ix++) { ErtsSchedulerData *esdp = ERTS_SCHEDULER_IX(ix); @@ -2592,7 +2681,7 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) } #ifdef ERTS_SMP - erts_smp_atomic_init(&esdp->chk_cpu_bind, 0); + erts_smp_atomic32_init(&esdp->chk_cpu_bind, 0); #endif } @@ -2600,21 +2689,21 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) erts_smp_mtx_init(&schdlr_sspnd.mtx, "schdlr_sspnd"); erts_smp_cnd_init(&schdlr_sspnd.cnd); - erts_smp_atomic_init(&schdlr_sspnd.changing, 0); + erts_smp_atomic32_init(&schdlr_sspnd.changing, 0); schdlr_sspnd.online = no_schedulers_online; schdlr_sspnd.curr_online = no_schedulers; - erts_smp_atomic_init(&schdlr_sspnd.msb.ongoing, 0); - erts_smp_atomic_init(&schdlr_sspnd.active, no_schedulers); + erts_smp_atomic32_init(&schdlr_sspnd.msb.ongoing, 0); + erts_smp_atomic32_init(&schdlr_sspnd.active, no_schedulers); schdlr_sspnd.msb.procs = NULL; - erts_smp_atomic_set(&balance_info.used_runqs, - erts_common_run_queue ? 1 : no_schedulers_online); - erts_smp_atomic_init(&balance_info.active_runqs, no_schedulers); + erts_smp_atomic32_set(&balance_info.used_runqs, + erts_common_run_queue ? 1 : no_schedulers_online); + erts_smp_atomic32_init(&balance_info.active_runqs, no_schedulers); balance_info.last_active_runqs = no_schedulers; erts_smp_mtx_init(&balance_info.update_mtx, "migration_info_update"); balance_info.forced_check_balance = 0; balance_info.halftime = 1; balance_info.full_reds_history_index = 0; - erts_smp_atomic_init(&balance_info.checking_balance, 0); + erts_smp_atomic32_init(&balance_info.checking_balance, 0); balance_info.prev_rise.active_runqs = 0; balance_info.prev_rise.max_len = 0; balance_info.prev_rise.reds = 0; @@ -2623,8 +2712,8 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) if (no_schedulers_online < no_schedulers) { if (erts_common_run_queue) { for (ix = no_schedulers_online; ix < no_schedulers; ix++) - erts_smp_atomic_bor(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, - ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_bor(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, + ERTS_SSI_FLG_SUSPENDED); } else { for (ix = no_schedulers_online; ix < erts_no_run_queues; ix++) @@ -2638,7 +2727,9 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) ERTS_SCHDLR_SSPND_CHNG_SET((ERTS_SCHDLR_SSPND_CHNG_ONLN | ERTS_SCHDLR_SSPND_CHNG_WAITER), 0); - erts_smp_atomic_init(&doing_sys_schedule, 0); + erts_smp_atomic32_init(&doing_sys_schedule, 0); + + init_misc_aux_work(); #else /* !ERTS_SMP */ { @@ -2652,12 +2743,19 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online) erts_no_schedulers = 1; #endif - erts_smp_atomic_init(&function_calls, 0); + erts_smp_atomic32_init(&function_calls, 0); /* init port tasks */ erts_port_task_init(); - late_cpu_bind_init(); +#ifndef ERTS_SMP +#ifdef ERTS_DO_VERIFY_UNUSED_TEMP_ALLOC + erts_scheduler_data->verify_unused_temp_alloc + = erts_alloc_get_verify_unused_temp_alloc( + &erts_scheduler_data->verify_unused_temp_alloc_data); + ERTS_VERIFY_UNUSED_TEMP_ALLOC(NULL); +#endif +#endif } ErtsRunQueue * @@ -2769,6 +2867,19 @@ resume_process(Process *p) p->rstatus = P_FREE; } +int +erts_get_max_no_executing_schedulers(void) +{ +#ifdef ERTS_SMP + if (erts_smp_atomic32_read(&schdlr_sspnd.changing)) + return (int) erts_no_schedulers; + ERTS_THR_MEMORY_BARRIER; + return (int) erts_smp_atomic32_read(&schdlr_sspnd.active); +#else + return 1; +#endif +} + #ifdef ERTS_SMP static void @@ -2787,13 +2898,13 @@ static void scheduler_ix_resume_wake(Uint ix) { ErtsSchedulerSleepInfo *ssi = ERTS_SCHED_SLEEP_INFO_IX(ix); - long xflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_TSE_SLEEPING - | ERTS_SSI_FLG_WAITING - | ERTS_SSI_FLG_SUSPENDED); - long oflgs; + erts_aint32_t xflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_TSE_SLEEPING + | ERTS_SSI_FLG_WAITING + | ERTS_SSI_FLG_SUSPENDED); + erts_aint32_t oflgs; do { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, 0, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, 0, xflgs); if (oflgs == xflgs) { erts_sched_finish_poke(ssi, oflgs); break; @@ -2802,17 +2913,17 @@ scheduler_ix_resume_wake(Uint ix) } while (oflgs & ERTS_SSI_FLG_SUSPENDED); } -static long -sched_prep_spin_suspended(ErtsSchedulerSleepInfo *ssi, long xpct) +static erts_aint32_t +sched_prep_spin_suspended(ErtsSchedulerSleepInfo *ssi, erts_aint32_t xpct) { - long oflgs; - long nflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_WAITING - | ERTS_SSI_FLG_SUSPENDED); - long xflgs = xpct; + erts_aint32_t oflgs; + erts_aint32_t nflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_WAITING + | ERTS_SSI_FLG_SUSPENDED); + erts_aint32_t xflgs = xpct; do { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return nflgs; xflgs = oflgs; @@ -2821,15 +2932,15 @@ sched_prep_spin_suspended(ErtsSchedulerSleepInfo *ssi, long xpct) return oflgs; } -static long +static erts_aint32_t sched_spin_suspended(ErtsSchedulerSleepInfo *ssi, int spincount) { int until_yield = ERTS_SCHED_SPIN_UNTIL_YIELD; int sc = spincount; - long flgs; + erts_aint32_t flgs; do { - flgs = erts_smp_atomic_read(&ssi->flags); + flgs = erts_smp_atomic32_read(&ssi->flags); if ((flgs & (ERTS_SSI_FLG_SLEEPING | ERTS_SSI_FLG_WAITING | ERTS_SSI_FLG_SUSPENDED)) @@ -2847,22 +2958,22 @@ sched_spin_suspended(ErtsSchedulerSleepInfo *ssi, int spincount) return flgs; } -static long +static erts_aint32_t sched_set_suspended_sleeptype(ErtsSchedulerSleepInfo *ssi) { - long oflgs; - long nflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_TSE_SLEEPING - | ERTS_SSI_FLG_WAITING - | ERTS_SSI_FLG_SUSPENDED); - long xflgs = (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_WAITING - | ERTS_SSI_FLG_SUSPENDED); + erts_aint32_t oflgs; + erts_aint32_t nflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_TSE_SLEEPING + | ERTS_SSI_FLG_WAITING + | ERTS_SSI_FLG_SUSPENDED); + erts_aint32_t xflgs = (ERTS_SSI_FLG_SLEEPING + | ERTS_SSI_FLG_WAITING + | ERTS_SSI_FLG_SUSPENDED); erts_tse_reset(ssi->event); while (1) { - oflgs = erts_smp_atomic_cmpxchg(&ssi->flags, nflgs, xflgs); + oflgs = erts_smp_atomic32_cmpxchg(&ssi->flags, nflgs, xflgs); if (oflgs == xflgs) return nflgs; if ((oflgs & (ERTS_SSI_FLG_SLEEPING @@ -2880,18 +2991,16 @@ sched_set_suspended_sleeptype(ErtsSchedulerSleepInfo *ssi) static void suspend_scheduler(ErtsSchedulerData *esdp) { - long flgs; - int changing; + erts_aint32_t flgs; + erts_aint32_t changing; long no = (long) esdp->no; - ErtsRunQueue *rq = esdp->run_queue; ErtsSchedulerSleepInfo *ssi = esdp->ssi; long active_schedulers; int curr_online = 1; int wake = 0; - int reset_read_group = 0; #if defined(ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK) \ || defined(ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK) - long aux_work; + erts_aint32_t aux_work; #endif /* @@ -2909,20 +3018,7 @@ suspend_scheduler(ErtsSchedulerData *esdp) erts_smp_runq_unlock(esdp->run_queue); - /* Unbind from cpu */ - erts_smp_rwmtx_rwlock(&erts_cpu_bind_rwmtx); - if (scheduler2cpu_map[esdp->no].bound_id >= 0 - && erts_unbind_from_cpu(erts_cpuinfo) == 0) { - esdp->cpu_id = scheduler2cpu_map[esdp->no].bound_id = -1; - reset_read_group = 1; - } - erts_smp_rwmtx_rwunlock(&erts_cpu_bind_rwmtx); - - if (reset_read_group) - erts_smp_rwmtx_set_reader_group(0); - - if (esdp->no <= erts_max_main_threads) - erts_thr_set_main_status(0, 0); + erts_sched_check_cpu_bind_prep_suspend(esdp); if (erts_system_profile_flags.scheduler) profile_scheduler(make_small(esdp->no), am_inactive); @@ -2932,15 +3028,15 @@ suspend_scheduler(ErtsSchedulerData *esdp) flgs = sched_prep_spin_suspended(ssi, ERTS_SSI_FLG_SUSPENDED); if (flgs & ERTS_SSI_FLG_SUSPENDED) { - active_schedulers = erts_smp_atomic_dectest(&schdlr_sspnd.active); + active_schedulers = erts_smp_atomic32_dectest(&schdlr_sspnd.active); ASSERT(active_schedulers >= 1); - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if (changing & ERTS_SCHDLR_SSPND_CHNG_MSB) { if (active_schedulers == schdlr_sspnd.msb.wait_active) wake = 1; if (active_schedulers == 1) { - changing = erts_smp_atomic_band(&schdlr_sspnd.changing, - ~ERTS_SCHDLR_SSPND_CHNG_MSB); + changing = erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_MSB); changing &= ~ERTS_SCHDLR_SSPND_CHNG_MSB; } } @@ -2962,8 +3058,8 @@ suspend_scheduler(ErtsSchedulerData *esdp) && schdlr_sspnd.curr_online == schdlr_sspnd.wait_curr_online) wake = 1; if (schdlr_sspnd.online == schdlr_sspnd.curr_online) { - changing = erts_smp_atomic_band(&schdlr_sspnd.changing, - ~ERTS_SCHDLR_SSPND_CHNG_ONLN); + changing = erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_ONLN); changing &= ~ERTS_SCHDLR_SSPND_CHNG_ONLN; } } @@ -2973,29 +3069,30 @@ suspend_scheduler(ErtsSchedulerData *esdp) wake = 0; } - flgs = erts_smp_atomic_read(&ssi->flags); + flgs = erts_smp_atomic32_read(&ssi->flags); if (!(flgs & ERTS_SSI_FLG_SUSPENDED)) break; erts_smp_mtx_unlock(&schdlr_sspnd.mtx); #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&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) { - long flgs; + erts_aint32_t flgs; #ifdef ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK #ifndef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->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); + flgs = sched_spin_suspended(ssi, + ERTS_SCHED_SUSPEND_SLEEP_SPINCOUNT); if (flgs == (ERTS_SSI_FLG_SLEEPING | ERTS_SSI_FLG_WAITING | ERTS_SSI_FLG_SUSPENDED)) { @@ -3015,13 +3112,13 @@ suspend_scheduler(ErtsSchedulerData *esdp) | ERTS_SSI_FLG_SUSPENDED)); if (!(flgs & ERTS_SSI_FLG_SUSPENDED)) break; - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if (changing & ~ERTS_SCHDLR_SSPND_CHNG_WAITER) break; #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK - aux_work = erts_smp_atomic_read(&ssi->aux_work); + aux_work = erts_smp_atomic32_read(&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; @@ -3033,19 +3130,19 @@ suspend_scheduler(ErtsSchedulerData *esdp) erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL); erts_smp_mtx_lock(&schdlr_sspnd.mtx); - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); } - active_schedulers = erts_smp_atomic_inctest(&schdlr_sspnd.active); - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + active_schedulers = erts_smp_atomic32_inctest(&schdlr_sspnd.active); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if ((changing & ERTS_SCHDLR_SSPND_CHNG_MSB) && schdlr_sspnd.online == active_schedulers) { - erts_smp_atomic_band(&schdlr_sspnd.changing, - ~ERTS_SCHDLR_SSPND_CHNG_MSB); + erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_MSB); } ASSERT(no <= schdlr_sspnd.online); - ASSERT(!erts_smp_atomic_read(&schdlr_sspnd.msb.ongoing)); + ASSERT(!erts_smp_atomic32_read(&schdlr_sspnd.msb.ongoing)); } @@ -3056,17 +3153,10 @@ suspend_scheduler(ErtsSchedulerData *esdp) if (erts_system_profile_flags.scheduler) profile_scheduler(make_small(esdp->no), am_active); - if (esdp->no <= erts_max_main_threads) - erts_thr_set_main_status(1, (int) esdp->no); - erts_smp_runq_lock(esdp->run_queue); non_empty_runq(esdp->run_queue); - /* Make sure we check if we should bind to a cpu or not... */ - if (rq->flags & ERTS_RUNQ_FLG_SHARED_RUNQ) - erts_smp_atomic_set(&esdp->chk_cpu_bind, 1); - else - rq->flags |= ERTS_RUNQ_FLG_CHK_CPU_BIND; + erts_sched_check_cpu_bind_post_suspend(esdp); } #define ERTS_RUNQ_RESET_SUSPEND_INFO(RQ, DBG_ID) \ @@ -3081,7 +3171,7 @@ do { \ (RQ)->flags |= (ERTS_RUNQ_FLG_OUT_OF_WORK \ | ERTS_RUNQ_FLG_HALFTIME_OUT_OF_WORK); \ (RQ)->check_balance_reds = ERTS_RUNQ_CALL_CHECK_BALANCE_REDS; \ - erts_smp_atomic_band(&(RQ)->info_flags, ~ERTS_RUNQ_IFLG_SUSPENDED); \ + erts_smp_atomic32_band(&(RQ)->info_flags, ~ERTS_RUNQ_IFLG_SUSPENDED);\ for (pix__ = 0; pix__ < ERTS_NO_PROC_PRIO_LEVELS; pix__++) { \ (RQ)->procs.prio_info[pix__].max_len = 0; \ (RQ)->procs.prio_info[pix__].reds = 0; \ @@ -3123,9 +3213,9 @@ erts_schedulers_state(Uint *total, int yield_allowed) { int res; - long changing; + erts_aint32_t changing; erts_smp_mtx_lock(&schdlr_sspnd.mtx); - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if (yield_allowed && (changing & ~ERTS_SCHDLR_SSPND_CHNG_WAITER)) res = ERTS_SCHDLR_SSPND_YIELD_RESTART; else { @@ -3146,7 +3236,7 @@ erts_set_schedulers_online(Process *p, Sint *old_no) { int ix, res, no, have_unlocked_plocks; - long changing; + erts_aint32_t changing; if (new_no < 1 || erts_no_schedulers < new_no) return ERTS_SCHDLR_SSPND_EINVAL; @@ -3156,7 +3246,7 @@ erts_set_schedulers_online(Process *p, have_unlocked_plocks = 0; no = (int) new_no; - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if (changing) { res = ERTS_SCHDLR_SSPND_YIELD_RESTART; } @@ -3203,7 +3293,7 @@ erts_set_schedulers_online(Process *p, ErtsRunQueue *to_rq = ERTS_RUNQ_IX(ix % no); evacuate_run_queue(from_rq, to_rq); } - erts_smp_atomic_set(&balance_info.used_runqs, no); + erts_smp_atomic32_set(&balance_info.used_runqs, no); erts_smp_mtx_unlock(&balance_info.update_mtx); erts_smp_mtx_lock(&schdlr_sspnd.mtx); } @@ -3231,8 +3321,8 @@ erts_set_schedulers_online(Process *p, for (ix = no; ix < online; ix++) { ErtsSchedulerSleepInfo *ssi; ssi = ERTS_SCHED_SLEEP_INFO_IX(ix); - erts_smp_atomic_bor(&ssi->flags, - ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_bor(&ssi->flags, + ERTS_SSI_FLG_SUSPENDED); } wake_all_schedulers(); } @@ -3257,7 +3347,7 @@ erts_set_schedulers_online(Process *p, for (ix = erts_no_run_queues-1; ix >= no; ix--) evacuate_run_queue(ERTS_RUNQ_IX(ix), ERTS_RUNQ_IX(ix % no)); - erts_smp_atomic_set(&balance_info.used_runqs, no); + erts_smp_atomic32_set(&balance_info.used_runqs, no); erts_smp_mtx_unlock(&balance_info.update_mtx); erts_smp_mtx_lock(&schdlr_sspnd.mtx); for (ix = no; ix < online; ix++) { @@ -3279,10 +3369,11 @@ erts_set_schedulers_online(Process *p, NULL); ASSERT(res != ERTS_SCHDLR_SSPND_DONE ? (ERTS_SCHDLR_SSPND_CHNG_WAITER - & erts_smp_atomic_read(&schdlr_sspnd.changing)) + & erts_smp_atomic32_read(&schdlr_sspnd.changing)) : (ERTS_SCHDLR_SSPND_CHNG_WAITER - == erts_smp_atomic_read(&schdlr_sspnd.changing))); - erts_smp_atomic_band(&schdlr_sspnd.changing, ~ERTS_SCHDLR_SSPND_CHNG_WAITER); + == erts_smp_atomic32_read(&schdlr_sspnd.changing))); + erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_WAITER); } } @@ -3297,11 +3388,11 @@ ErtsSchedSuspendResult erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) { int ix, res, have_unlocked_plocks = 0; - long changing; + erts_aint32_t changing; ErtsProcList *plp; erts_smp_mtx_lock(&schdlr_sspnd.mtx); - changing = erts_smp_atomic_read(&schdlr_sspnd.changing); + changing = erts_smp_atomic32_read(&schdlr_sspnd.changing); if (changing) { res = ERTS_SCHDLR_SSPND_YIELD_RESTART; /* Yield */ } @@ -3311,7 +3402,7 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) plp->next = schdlr_sspnd.msb.procs; schdlr_sspnd.msb.procs = plp; p->flags |= F_HAVE_BLCKD_MSCHED; - ASSERT(erts_smp_atomic_read(&schdlr_sspnd.active) == 1); + ASSERT(erts_smp_atomic32_read(&schdlr_sspnd.active) == 1); ASSERT(p->scheduler_data->no == 1); res = ERTS_SCHDLR_SSPND_DONE_MSCHED_BLOCKED; } @@ -3322,11 +3413,11 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) have_unlocked_plocks = 1; erts_smp_proc_unlock(p, plocks); } - ASSERT(0 == erts_smp_atomic_read(&schdlr_sspnd.msb.ongoing)); - erts_smp_atomic_set(&schdlr_sspnd.msb.ongoing, 1); + ASSERT(0 == erts_smp_atomic32_read(&schdlr_sspnd.msb.ongoing)); + erts_smp_atomic32_set(&schdlr_sspnd.msb.ongoing, 1); if (online == 1) { res = ERTS_SCHDLR_SSPND_DONE_MSCHED_BLOCKED; - ASSERT(erts_smp_atomic_read(&schdlr_sspnd.active) == 1); + ASSERT(erts_smp_atomic32_read(&schdlr_sspnd.active) == 1); ASSERT(p->scheduler_data->no == 1); } else { @@ -3346,14 +3437,14 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) } if (erts_common_run_queue) { for (ix = 1; ix < online; ix++) - erts_smp_atomic_bor(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, - ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_bor(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, + ERTS_SSI_FLG_SUSPENDED); wake_all_schedulers(); } else { erts_smp_mtx_unlock(&schdlr_sspnd.mtx); erts_smp_mtx_lock(&balance_info.update_mtx); - erts_smp_atomic_set(&balance_info.used_runqs, 1); + erts_smp_atomic32_set(&balance_info.used_runqs, 1); for (ix = 0; ix < online; ix++) { ErtsRunQueue *rq = ERTS_RUNQ_IX(ix); erts_smp_runq_lock(rq); @@ -3375,7 +3466,7 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) susp_sched_prep_block, susp_sched_resume_block, NULL); - while (erts_smp_atomic_read(&schdlr_sspnd.active) + while (erts_smp_atomic32_read(&schdlr_sspnd.active) != schdlr_sspnd.msb.wait_active) erts_smp_cnd_wait(&schdlr_sspnd.cnd, &schdlr_sspnd.mtx); erts_smp_activity_end(ERTS_ACTIVITY_WAIT, @@ -3384,11 +3475,11 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) NULL); ASSERT(res != ERTS_SCHDLR_SSPND_DONE_MSCHED_BLOCKED ? (ERTS_SCHDLR_SSPND_CHNG_WAITER - & erts_smp_atomic_read(&schdlr_sspnd.changing)) + & erts_smp_atomic32_read(&schdlr_sspnd.changing)) : (ERTS_SCHDLR_SSPND_CHNG_WAITER - == erts_smp_atomic_read(&schdlr_sspnd.changing))); - erts_smp_atomic_band(&schdlr_sspnd.changing, - ~ERTS_SCHDLR_SSPND_CHNG_WAITER); + == erts_smp_atomic32_read(&schdlr_sspnd.changing))); + erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_WAITER); } plp = proclist_create(p); plp->next = schdlr_sspnd.msb.procs; @@ -3455,16 +3546,16 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) }); #endif p->flags &= ~F_HAVE_BLCKD_MSCHED; - erts_smp_atomic_set(&schdlr_sspnd.msb.ongoing, 0); + erts_smp_atomic32_set(&schdlr_sspnd.msb.ongoing, 0); if (schdlr_sspnd.online == 1) { /* No schedulers to resume */ - ASSERT(erts_smp_atomic_read(&schdlr_sspnd.active) == 1); + ASSERT(erts_smp_atomic32_read(&schdlr_sspnd.active) == 1); ERTS_SCHDLR_SSPND_CHNG_SET(0, ERTS_SCHDLR_SSPND_CHNG_MSB); } else if (erts_common_run_queue) { for (ix = 1; ix < schdlr_sspnd.online; ix++) - erts_smp_atomic_band(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, - ~ERTS_SSI_FLG_SUSPENDED); + erts_smp_atomic32_band(&ERTS_SCHED_SLEEP_INFO_IX(ix)->flags, + ~ERTS_SSI_FLG_SUSPENDED); wake_all_schedulers(); } else { @@ -3490,7 +3581,7 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int all) evacuate_run_queue(ERTS_RUNQ_IX(ix), ERTS_RUNQ_IX(ix % online)); - erts_smp_atomic_set(&balance_info.used_runqs, online); + erts_smp_atomic32_set(&balance_info.used_runqs, online); /* Make sure that we balance soon... */ balance_info.forced_check_balance = 1; erts_smp_runq_lock(ERTS_RUNQ_IX(0)); @@ -3514,7 +3605,7 @@ void erts_dbg_multi_scheduling_return_trap(Process *p, Eterm return_value) { if (return_value == am_blocked) { - long active = erts_smp_atomic_read(&schdlr_sspnd.active); + erts_aint32_t active = erts_smp_atomic32_read(&schdlr_sspnd.active); ASSERT(1 <= active && active <= 2); ASSERT(ERTS_PROC_GET_SCHDATA(p)->no == 1); } @@ -3583,15 +3674,7 @@ sched_thread_func(void *vesdp) erts_tsd_set(sched_data_key, vesdp); #ifdef ERTS_SMP - if (no <= erts_max_main_threads) { - erts_thr_set_main_status(1, (int) no); - if (erts_reader_groups) { - int rg = (int) no; - if (rg > erts_reader_groups) - rg = (((int) no) - 1) % erts_reader_groups + 1; - erts_smp_rwmtx_set_reader_group(rg); - } - } + erts_sched_init_check_cpu_bind((ErtsSchedulerData *) vesdp); erts_proc_lock_prepare_proc_lock_waiter(); ERTS_SCHED_SLEEP_INFO_IX(no - 1)->event = erts_tse_fetch(); @@ -3605,12 +3688,12 @@ sched_thread_func(void *vesdp) erts_thread_init_float(); erts_smp_mtx_lock(&schdlr_sspnd.mtx); - ASSERT(erts_smp_atomic_read(&schdlr_sspnd.changing) + ASSERT(erts_smp_atomic32_read(&schdlr_sspnd.changing) & ERTS_SCHDLR_SSPND_CHNG_ONLN); if (--schdlr_sspnd.curr_online == schdlr_sspnd.wait_curr_online) { - erts_smp_atomic_band(&schdlr_sspnd.changing, - ~ERTS_SCHDLR_SSPND_CHNG_ONLN); + erts_smp_atomic32_band(&schdlr_sspnd.changing, + ~ERTS_SCHDLR_SSPND_CHNG_ONLN); if (((ErtsSchedulerData *) vesdp)->no != 1) erts_smp_cnd_signal(&schdlr_sspnd.cnd); } @@ -3632,6 +3715,13 @@ 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 + = erts_alloc_get_verify_unused_temp_alloc( + &((ErtsSchedulerData *) vesdp)->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 %bpu terminated\n", @@ -3693,1907 +3783,6 @@ erts_start_schedulers(void) #endif /* ERTS_SMP */ -static int -int_cmp(const void *vx, const void *vy) -{ - return *((int *) vx) - *((int *) vy); -} - -static int -cpu_spread_order_cmp(const void *vx, const void *vy) -{ - erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx; - erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy; - - if (x->thread != y->thread) - return x->thread - y->thread; - if (x->core != y->core) - return x->core - y->core; - if (x->processor_node != y->processor_node) - return x->processor_node - y->processor_node; - if (x->processor != y->processor) - return x->processor - y->processor; - if (x->node != y->node) - return x->node - y->node; - return 0; -} - -static int -cpu_processor_spread_order_cmp(const void *vx, const void *vy) -{ - erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx; - erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy; - - if (x->thread != y->thread) - return x->thread - y->thread; - if (x->processor_node != y->processor_node) - return x->processor_node - y->processor_node; - if (x->core != y->core) - return x->core - y->core; - if (x->node != y->node) - return x->node - y->node; - if (x->processor != y->processor) - return x->processor - y->processor; - return 0; -} - -static int -cpu_thread_spread_order_cmp(const void *vx, const void *vy) -{ - erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx; - erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy; - - if (x->thread != y->thread) - return x->thread - y->thread; - if (x->node != y->node) - return x->node - y->node; - if (x->processor != y->processor) - return x->processor - y->processor; - if (x->processor_node != y->processor_node) - return x->processor_node - y->processor_node; - if (x->core != y->core) - return x->core - y->core; - return 0; -} - -static int -cpu_thread_no_node_processor_spread_order_cmp(const void *vx, const void *vy) -{ - erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx; - erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy; - - if (x->thread != y->thread) - return x->thread - y->thread; - if (x->node != y->node) - return x->node - y->node; - if (x->core != y->core) - return x->core - y->core; - if (x->processor != y->processor) - return x->processor - y->processor; - return 0; -} - -static int -cpu_no_node_processor_spread_order_cmp(const void *vx, const void *vy) -{ - erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx; - erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy; - - if (x->node != y->node) - return x->node - y->node; - if (x->thread != y->thread) - return x->thread - y->thread; - if (x->core != y->core) - return x->core - y->core; - if (x->processor != y->processor) - return x->processor - y->processor; - return 0; -} - -static int -cpu_no_node_thread_spread_order_cmp(const void *vx, const void *vy) -{ - erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx; - erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy; - - if (x->node != y->node) - return x->node - y->node; - if (x->thread != y->thread) - return x->thread - y->thread; - if (x->processor != y->processor) - return x->processor - y->processor; - if (x->core != y->core) - return x->core - y->core; - return 0; -} - -static int -cpu_no_spread_order_cmp(const void *vx, const void *vy) -{ - erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx; - erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy; - - if (x->node != y->node) - return x->node - y->node; - if (x->processor != y->processor) - return x->processor - y->processor; - if (x->processor_node != y->processor_node) - return x->processor_node - y->processor_node; - if (x->core != y->core) - return x->core - y->core; - if (x->thread != y->thread) - return x->thread - y->thread; - return 0; -} - -static ERTS_INLINE void -make_cpudata_id_seq(erts_cpu_topology_t *cpudata, int size, int no_node) -{ - int ix; - int node = -1; - int processor = -1; - int processor_node = -1; - int processor_node_node = -1; - int core = -1; - int thread = -1; - int old_node = -1; - int old_processor = -1; - int old_processor_node = -1; - int old_core = -1; - int old_thread = -1; - - for (ix = 0; ix < size; ix++) { - if (!no_node || cpudata[ix].node >= 0) { - if (old_node == cpudata[ix].node) - cpudata[ix].node = node; - else { - old_node = cpudata[ix].node; - old_processor = processor = -1; - if (!no_node) - old_processor_node = processor_node = -1; - old_core = core = -1; - old_thread = thread = -1; - if (no_node || cpudata[ix].node >= 0) - cpudata[ix].node = ++node; - } - } - if (old_processor == cpudata[ix].processor) - cpudata[ix].processor = processor; - else { - old_processor = cpudata[ix].processor; - if (!no_node) - processor_node_node = old_processor_node = processor_node = -1; - old_core = core = -1; - old_thread = thread = -1; - cpudata[ix].processor = ++processor; - } - if (no_node && cpudata[ix].processor_node < 0) - old_processor_node = -1; - else { - if (old_processor_node == cpudata[ix].processor_node) { - if (no_node) - cpudata[ix].node = cpudata[ix].processor_node = node; - else { - if (processor_node_node >= 0) - cpudata[ix].node = processor_node_node; - cpudata[ix].processor_node = processor_node; - } - } - else { - old_processor_node = cpudata[ix].processor_node; - old_core = core = -1; - old_thread = thread = -1; - if (no_node) - cpudata[ix].node = cpudata[ix].processor_node = ++node; - else { - cpudata[ix].node = processor_node_node = ++node; - cpudata[ix].processor_node = ++processor_node; - } - } - } - if (!no_node && cpudata[ix].processor_node < 0) - cpudata[ix].processor_node = 0; - if (old_core == cpudata[ix].core) - cpudata[ix].core = core; - else { - old_core = cpudata[ix].core; - old_thread = thread = -1; - cpudata[ix].core = ++core; - } - if (old_thread == cpudata[ix].thread) - cpudata[ix].thread = thread; - else - old_thread = cpudata[ix].thread = ++thread; - } -} - -static void -cpu_bind_order_sort(erts_cpu_topology_t *cpudata, - int size, - ErtsCpuBindOrder bind_order, - int mk_seq) -{ - if (size > 1) { - int no_node = 0; - int (*cmp_func)(const void *, const void *); - switch (bind_order) { - case ERTS_CPU_BIND_SPREAD: - cmp_func = cpu_spread_order_cmp; - break; - case ERTS_CPU_BIND_PROCESSOR_SPREAD: - cmp_func = cpu_processor_spread_order_cmp; - break; - case ERTS_CPU_BIND_THREAD_SPREAD: - cmp_func = cpu_thread_spread_order_cmp; - break; - case ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD: - no_node = 1; - cmp_func = cpu_thread_no_node_processor_spread_order_cmp; - break; - case ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD: - no_node = 1; - cmp_func = cpu_no_node_processor_spread_order_cmp; - break; - case ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD: - no_node = 1; - cmp_func = cpu_no_node_thread_spread_order_cmp; - break; - case ERTS_CPU_BIND_NO_SPREAD: - cmp_func = cpu_no_spread_order_cmp; - break; - default: - cmp_func = NULL; - erl_exit(ERTS_ABORT_EXIT, - "Bad cpu bind type: %d\n", - (int) cpu_bind_order); - break; - } - - if (mk_seq) - make_cpudata_id_seq(cpudata, size, no_node); - - qsort(cpudata, size, sizeof(erts_cpu_topology_t), cmp_func); - } -} - -static int -processor_order_cmp(const void *vx, const void *vy) -{ - erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx; - erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy; - - if (x->processor != y->processor) - return x->processor - y->processor; - if (x->node != y->node) - return x->node - y->node; - if (x->processor_node != y->processor_node) - return x->processor_node - y->processor_node; - if (x->core != y->core) - return x->core - y->core; - if (x->thread != y->thread) - return x->thread - y->thread; - return 0; -} - -static void -check_cpu_bind(ErtsSchedulerData *esdp) -{ - int rg = 0; - int res; - int cpu_id; - erts_smp_runq_unlock(esdp->run_queue); - erts_smp_rwmtx_rwlock(&erts_cpu_bind_rwmtx); - cpu_id = scheduler2cpu_map[esdp->no].bind_id; - if (cpu_id >= 0 && cpu_id != scheduler2cpu_map[esdp->no].bound_id) { - res = erts_bind_to_cpu(erts_cpuinfo, cpu_id); - if (res == 0) - esdp->cpu_id = scheduler2cpu_map[esdp->no].bound_id = cpu_id; - else { - erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); - erts_dsprintf(dsbufp, "Scheduler %d failed to bind to cpu %d: %s\n", - (int) esdp->no, cpu_id, erl_errno_id(-res)); - erts_send_error_to_logger_nogl(dsbufp); - if (scheduler2cpu_map[esdp->no].bound_id >= 0) - goto unbind; - } - } - else if (cpu_id < 0) { - unbind: - /* Get rid of old binding */ - res = erts_unbind_from_cpu(erts_cpuinfo); - if (res == 0) - esdp->cpu_id = scheduler2cpu_map[esdp->no].bound_id = -1; - else if (res != -ENOTSUP) { - erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); - erts_dsprintf(dsbufp, "Scheduler %d failed to unbind from cpu %d: %s\n", - (int) esdp->no, cpu_id, erl_errno_id(-res)); - erts_send_error_to_logger_nogl(dsbufp); - } - } - if (erts_reader_groups) { - if (esdp->cpu_id >= 0) - rg = reader_group_lookup(esdp->cpu_id); - else - rg = (((int) esdp->no) - 1) % erts_reader_groups + 1; - } - erts_smp_runq_lock(esdp->run_queue); -#ifdef ERTS_SMP - if (erts_common_run_queue) - erts_smp_atomic_set(&esdp->chk_cpu_bind, 0); - else { - esdp->run_queue->flags &= ~ERTS_RUNQ_FLG_CHK_CPU_BIND; - } -#endif - erts_smp_rwmtx_rwunlock(&erts_cpu_bind_rwmtx); - - if (erts_reader_groups) - erts_smp_rwmtx_set_reader_group(rg); -} - -static void -signal_schedulers_bind_change(erts_cpu_topology_t *cpudata, int size) -{ - int s_ix = 1; - int cpu_ix; - - if (cpu_bind_order != ERTS_CPU_BIND_NONE && size) { - - cpu_bind_order_sort(cpudata, size, cpu_bind_order, 1); - - for (cpu_ix = 0; cpu_ix < size && cpu_ix < erts_no_schedulers; cpu_ix++) - if (erts_is_cpu_available(erts_cpuinfo, cpudata[cpu_ix].logical)) - scheduler2cpu_map[s_ix++].bind_id = cpudata[cpu_ix].logical; - } - - if (s_ix <= erts_no_schedulers) - for (; s_ix <= erts_no_schedulers; s_ix++) - scheduler2cpu_map[s_ix].bind_id = -1; - -#ifdef ERTS_SMP - if (erts_common_run_queue) { - for (s_ix = 0; s_ix < erts_no_schedulers; s_ix++) - erts_smp_atomic_set(&ERTS_SCHEDULER_IX(s_ix)->chk_cpu_bind, 1); - wake_all_schedulers(); - } - else { - for (s_ix = 0; s_ix < erts_no_run_queues; s_ix++) { - ErtsRunQueue *rq = ERTS_RUNQ_IX(s_ix); - erts_smp_runq_lock(rq); - rq->flags |= ERTS_RUNQ_FLG_CHK_CPU_BIND; - erts_smp_runq_unlock(rq); - wake_scheduler(rq, 0, 1); - }; - } -#else - check_cpu_bind(erts_get_scheduler_data()); -#endif -} - -int -erts_init_scheduler_bind_type(char *how) -{ - if (erts_bind_to_cpu(erts_cpuinfo, -1) == -ENOTSUP) - return ERTS_INIT_SCHED_BIND_TYPE_NOT_SUPPORTED; - - if (!system_cpudata && !user_cpudata) - return ERTS_INIT_SCHED_BIND_TYPE_ERROR_NO_CPU_TOPOLOGY; - - if (sys_strcmp(how, "db") == 0) - cpu_bind_order = ERTS_CPU_BIND_DEFAULT_BIND; - else if (sys_strcmp(how, "s") == 0) - cpu_bind_order = ERTS_CPU_BIND_SPREAD; - else if (sys_strcmp(how, "ps") == 0) - cpu_bind_order = ERTS_CPU_BIND_PROCESSOR_SPREAD; - else if (sys_strcmp(how, "ts") == 0) - cpu_bind_order = ERTS_CPU_BIND_THREAD_SPREAD; - else if (sys_strcmp(how, "tnnps") == 0) - cpu_bind_order = ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD; - else if (sys_strcmp(how, "nnps") == 0) - cpu_bind_order = ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD; - else if (sys_strcmp(how, "nnts") == 0) - cpu_bind_order = ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD; - else if (sys_strcmp(how, "ns") == 0) - cpu_bind_order = ERTS_CPU_BIND_NO_SPREAD; - else if (sys_strcmp(how, "u") == 0) - cpu_bind_order = ERTS_CPU_BIND_NONE; - else - return ERTS_INIT_SCHED_BIND_TYPE_ERROR_NO_BAD_TYPE; - - return ERTS_INIT_SCHED_BIND_TYPE_SUCCESS; -} - -/* - * reader groups map - */ - -typedef struct { - int level[ERTS_TOPOLOGY_MAX_DEPTH+1]; -} erts_avail_cput; - -typedef struct { - int *map; - int size; - int groups; -} erts_reader_groups_map_test; - -typedef struct { - int id; - int sub_levels; - int reader_groups; -} erts_rg_count_t; - -typedef struct { - int logical; - int reader_group; -} erts_reader_groups_map_t; - -typedef struct { - erts_reader_groups_map_t *map; - int map_size; - int logical_processors; - int groups; -} erts_make_reader_groups_map_test; - -static int reader_groups_available_cpu_check; -static int reader_groups_logical_processors; -static int reader_groups_map_size; -static erts_reader_groups_map_t *reader_groups_map; - -#define ERTS_TOPOLOGY_RG ERTS_TOPOLOGY_MAX_DEPTH - -static void -make_reader_groups_map(erts_make_reader_groups_map_test *test); - -static Eterm -get_reader_groups_map(Process *c_p, - erts_reader_groups_map_t *map, - int map_size, - int logical_processors) -{ -#ifdef DEBUG - Eterm *endp; -#endif - Eterm res = NIL, tuple; - Eterm *hp; - int i; - - hp = HAlloc(c_p, logical_processors*(2+3)); -#ifdef DEBUG - endp = hp + logical_processors*(2+3); -#endif - for (i = map_size - 1; i >= 0; i--) { - if (map[i].logical >= 0) { - tuple = TUPLE2(hp, - make_small(map[i].logical), - make_small(map[i].reader_group)); - hp += 3; - res = CONS(hp, tuple, res); - hp += 2; - } - } - ASSERT(hp == endp); - return res; -} - -Eterm -erts_debug_reader_groups_map(Process *c_p, int groups) -{ - Eterm res; - erts_make_reader_groups_map_test test; - - test.groups = groups; - make_reader_groups_map(&test); - if (!test.map) - res = NIL; - else { - res = get_reader_groups_map(c_p, - test.map, - test.map_size, - test.logical_processors); - erts_free(ERTS_ALC_T_TMP, test.map); - } - return res; -} - - -Eterm -erts_get_reader_groups_map(Process *c_p) -{ - Eterm res; - erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx); - res = get_reader_groups_map(c_p, - reader_groups_map, - reader_groups_map_size, - reader_groups_logical_processors); - erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx); - return res; -} - -static void -make_available_cpu_topology(erts_avail_cput *no, - erts_avail_cput *avail, - erts_cpu_topology_t *cpudata, - int *size, - int test) -{ - int len = *size; - erts_cpu_topology_t last; - int a, i, j; - - no->level[ERTS_TOPOLOGY_NODE] = -1; - no->level[ERTS_TOPOLOGY_PROCESSOR] = -1; - no->level[ERTS_TOPOLOGY_PROCESSOR_NODE] = -1; - no->level[ERTS_TOPOLOGY_CORE] = -1; - no->level[ERTS_TOPOLOGY_THREAD] = -1; - no->level[ERTS_TOPOLOGY_LOGICAL] = -1; - - last.node = INT_MIN; - last.processor = INT_MIN; - last.processor_node = INT_MIN; - last.core = INT_MIN; - last.thread = INT_MIN; - last.logical = INT_MIN; - - a = 0; - - for (i = 0; i < len; i++) { - - if (!test && !erts_is_cpu_available(erts_cpuinfo, cpudata[i].logical)) - continue; - - if (last.node != cpudata[i].node) - goto node; - if (last.processor != cpudata[i].processor) - goto processor; - if (last.processor_node != cpudata[i].processor_node) - goto processor_node; - if (last.core != cpudata[i].core) - goto core; - ASSERT(last.thread != cpudata[i].thread); - goto thread; - - node: - no->level[ERTS_TOPOLOGY_NODE]++; - processor: - no->level[ERTS_TOPOLOGY_PROCESSOR]++; - processor_node: - no->level[ERTS_TOPOLOGY_PROCESSOR_NODE]++; - core: - no->level[ERTS_TOPOLOGY_CORE]++; - thread: - no->level[ERTS_TOPOLOGY_THREAD]++; - - no->level[ERTS_TOPOLOGY_LOGICAL]++; - - for (j = 0; j < ERTS_TOPOLOGY_LOGICAL; j++) - avail[a].level[j] = no->level[j]; - - avail[a].level[ERTS_TOPOLOGY_LOGICAL] = cpudata[i].logical; - avail[a].level[ERTS_TOPOLOGY_RG] = 0; - - ASSERT(last.logical != cpudata[a].logical); - - last = cpudata[i]; - a++; - } - - no->level[ERTS_TOPOLOGY_NODE]++; - no->level[ERTS_TOPOLOGY_PROCESSOR]++; - no->level[ERTS_TOPOLOGY_PROCESSOR_NODE]++; - no->level[ERTS_TOPOLOGY_CORE]++; - no->level[ERTS_TOPOLOGY_THREAD]++; - no->level[ERTS_TOPOLOGY_LOGICAL]++; - - *size = a; -} - -static int -reader_group_lookup(int logical) -{ - int start = logical % reader_groups_map_size; - int ix = start; - - do { - if (reader_groups_map[ix].logical == logical) { - ASSERT(reader_groups_map[ix].reader_group > 0); - return reader_groups_map[ix].reader_group; - } - ix++; - if (ix == reader_groups_map_size) - ix = 0; - } while (ix != start); - - erl_exit(ERTS_ABORT_EXIT, "Logical cpu id %d not found\n", logical); -} - -static void -reader_group_insert(erts_reader_groups_map_t *map, int map_size, - int logical, int reader_group) -{ - int start = logical % map_size; - int ix = start; - - do { - if (map[ix].logical < 0) { - map[ix].logical = logical; - map[ix].reader_group = reader_group; - return; - } - ix++; - if (ix == map_size) - ix = 0; - } while (ix != start); - - erl_exit(ERTS_ABORT_EXIT, "Reader groups map full\n"); -} - - -static int -sub_levels(erts_rg_count_t *rgc, int level, int aix, int avail_sz, erts_avail_cput *avail) -{ - int sub_level = level+1; - int last = -1; - rgc->sub_levels = 0; - - do { - if (last != avail[aix].level[sub_level]) { - rgc->sub_levels++; - last = avail[aix].level[sub_level]; - } - aix++; - } - while (aix < avail_sz && rgc->id == avail[aix].level[level]); - rgc->reader_groups = 0; - return aix; -} - -static int -write_reader_groups(int *rgp, erts_rg_count_t *rgcp, - int level, int a, - int avail_sz, erts_avail_cput *avail) -{ - int rg = *rgp; - int sub_level = level+1; - int sl_per_gr = rgcp->sub_levels / rgcp->reader_groups; - int xsl = rgcp->sub_levels % rgcp->reader_groups; - int sls = 0; - int last = -1; - int xsl_rg_lim = (rgcp->reader_groups - xsl) + rg + 1; - - ASSERT(level < 0 || avail[a].level[level] == rgcp->id) - - do { - if (last != avail[a].level[sub_level]) { - if (!sls) { - sls = sl_per_gr; - rg++; - if (rg >= xsl_rg_lim) - sls++; - } - last = avail[a].level[sub_level]; - sls--; - } - avail[a].level[ERTS_TOPOLOGY_RG] = rg; - a++; - } while (a < avail_sz && (level < 0 - || avail[a].level[level] == rgcp->id)); - - ASSERT(rgcp->reader_groups == rg - *rgp); - - *rgp = rg; - - return a; -} - -static int -rg_count_sub_levels_compare(const void *vx, const void *vy) -{ - erts_rg_count_t *x = (erts_rg_count_t *) vx; - erts_rg_count_t *y = (erts_rg_count_t *) vy; - if (x->sub_levels != y->sub_levels) - return y->sub_levels - x->sub_levels; - return x->id - y->id; -} - -static int -rg_count_id_compare(const void *vx, const void *vy) -{ - erts_rg_count_t *x = (erts_rg_count_t *) vx; - erts_rg_count_t *y = (erts_rg_count_t *) vy; - return x->id - y->id; -} - -static void -make_reader_groups_map(erts_make_reader_groups_map_test *test) -{ - int i, spread_level, avail_sz; - erts_avail_cput no, *avail; - erts_cpu_topology_t *cpudata; - erts_reader_groups_map_t *map; - int map_sz; - int groups = erts_reader_groups; - - if (test) { - test->map = NULL; - test->map_size = 0; - groups = test->groups; - } - - if (!groups) - return; - - if (!test) { - if (reader_groups_map) - erts_free(ERTS_ALC_T_RDR_GRPS_MAP, reader_groups_map); - - reader_groups_logical_processors = 0; - reader_groups_map_size = 0; - reader_groups_map = NULL; - } - - create_tmp_cpu_topology_copy(&cpudata, &avail_sz); - - if (!cpudata) - return; - - cpu_bind_order_sort(cpudata, - avail_sz, - ERTS_CPU_BIND_NO_SPREAD, - 1); - - avail = erts_alloc(ERTS_ALC_T_TMP, - sizeof(erts_avail_cput)*avail_sz); - - make_available_cpu_topology(&no, avail, cpudata, - &avail_sz, test != NULL); - - destroy_tmp_cpu_topology_copy(cpudata); - - map_sz = avail_sz*2+1; - - if (test) { - map = erts_alloc(ERTS_ALC_T_TMP, - (sizeof(erts_reader_groups_map_t) - * map_sz)); - test->map = map; - test->map_size = map_sz; - test->logical_processors = avail_sz; - } - else { - map = erts_alloc(ERTS_ALC_T_RDR_GRPS_MAP, - (sizeof(erts_reader_groups_map_t) - * map_sz)); - reader_groups_map = map; - reader_groups_logical_processors = avail_sz; - reader_groups_map_size = map_sz; - - } - - for (i = 0; i < map_sz; i++) { - map[i].logical = -1; - map[i].reader_group = 0; - } - - spread_level = ERTS_TOPOLOGY_CORE; - for (i = ERTS_TOPOLOGY_NODE; i < ERTS_TOPOLOGY_THREAD; i++) { - if (no.level[i] > groups) { - spread_level = i; - break; - } - } - - if (no.level[spread_level] <= groups) { - int a, rg, last = -1; - rg = 0; - ASSERT(spread_level == ERTS_TOPOLOGY_CORE); - for (a = 0; a < avail_sz; a++) { - if (last != avail[a].level[spread_level]) { - rg++; - last = avail[a].level[spread_level]; - } - reader_group_insert(map, - map_sz, - avail[a].level[ERTS_TOPOLOGY_LOGICAL], - rg); - } - } - else { /* groups < no.level[spread_level] */ - erts_rg_count_t *rg_count; - int a, rg, tl, toplevels; - - tl = spread_level-1; - - if (spread_level == ERTS_TOPOLOGY_NODE) - toplevels = 1; - else - toplevels = no.level[tl]; - - rg_count = erts_alloc(ERTS_ALC_T_TMP, - toplevels*sizeof(erts_rg_count_t)); - - if (toplevels == 1) { - rg_count[0].id = 0; - rg_count[0].sub_levels = no.level[spread_level]; - rg_count[0].reader_groups = groups; - } - else { - int rgs_per_tl, rgs; - rgs = groups; - rgs_per_tl = rgs / toplevels; - - a = 0; - for (i = 0; i < toplevels; i++) { - rg_count[i].id = avail[a].level[tl]; - a = sub_levels(&rg_count[i], tl, a, avail_sz, avail); - } - - qsort(rg_count, - toplevels, - sizeof(erts_rg_count_t), - rg_count_sub_levels_compare); - - for (i = 0; i < toplevels; i++) { - if (rg_count[i].sub_levels < rgs_per_tl) { - rg_count[i].reader_groups = rg_count[i].sub_levels; - rgs -= rg_count[i].sub_levels; - } - else { - rg_count[i].reader_groups = rgs_per_tl; - rgs -= rgs_per_tl; - } - } - - while (rgs > 0) { - for (i = 0; i < toplevels; i++) { - if (rg_count[i].sub_levels == rg_count[i].reader_groups) - break; - else { - rg_count[i].reader_groups++; - if (--rgs == 0) - break; - } - } - } - - qsort(rg_count, - toplevels, - sizeof(erts_rg_count_t), - rg_count_id_compare); - } - - a = i = rg = 0; - while (a < avail_sz) { - a = write_reader_groups(&rg, &rg_count[i], tl, - a, avail_sz, avail); - i++; - } - - ASSERT(groups == rg); - - for (a = 0; a < avail_sz; a++) - reader_group_insert(map, - map_sz, - avail[a].level[ERTS_TOPOLOGY_LOGICAL], - avail[a].level[ERTS_TOPOLOGY_RG]); - - erts_free(ERTS_ALC_T_TMP, rg_count); - } - - erts_free(ERTS_ALC_T_TMP, avail); -} - -/* - * CPU topology - */ - -typedef struct { - int *id; - int used; - int size; -} ErtsCpuTopIdSeq; - -typedef struct { - ErtsCpuTopIdSeq logical; - ErtsCpuTopIdSeq thread; - ErtsCpuTopIdSeq core; - ErtsCpuTopIdSeq processor_node; - ErtsCpuTopIdSeq processor; - ErtsCpuTopIdSeq node; -} ErtsCpuTopEntry; - -static void -init_cpu_top_entry(ErtsCpuTopEntry *cte) -{ - int size = 10; - cte->logical.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS, - sizeof(int)*size); - cte->logical.size = size; - cte->thread.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS, - sizeof(int)*size); - cte->thread.size = size; - cte->core.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS, - sizeof(int)*size); - cte->core.size = size; - cte->processor_node.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS, - sizeof(int)*size); - cte->processor_node.size = size; - cte->processor.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS, - sizeof(int)*size); - cte->processor.size = size; - cte->node.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS, - sizeof(int)*size); - cte->node.size = size; -} - -static void -destroy_cpu_top_entry(ErtsCpuTopEntry *cte) -{ - erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->logical.id); - erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->thread.id); - erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->core.id); - erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->processor_node.id); - erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->processor.id); - erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->node.id); -} - -static int -get_cput_value_or_range(int *v, int *vr, char **str) -{ - long l; - char *c = *str; - errno = 0; - if (!isdigit((unsigned char)*c)) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID; - l = strtol(c, &c, 10); - if (errno != 0 || l < 0 || ERTS_MAX_CPU_TOPOLOGY_ID < l) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID; - *v = (int) l; - if (*c == '-') { - c++; - if (!isdigit((unsigned char)*c)) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE; - l = strtol(c, &c, 10); - if (errno != 0 || l < 0 || ERTS_MAX_CPU_TOPOLOGY_ID < l) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE; - *vr = (int) l; - } - *str = c; - return ERTS_INIT_CPU_TOPOLOGY_OK; -} - -static int -get_cput_id_seq(ErtsCpuTopIdSeq *idseq, char **str) -{ - int ix = 0; - int need_size = 0; - char *c = *str; - - while (1) { - int res; - int val; - int nids; - int val_range = -1; - res = get_cput_value_or_range(&val, &val_range, &c); - if (res != ERTS_INIT_CPU_TOPOLOGY_OK) - return res; - if (val_range < 0 || val_range == val) - nids = 1; - else { - if (val_range > val) - nids = val_range - val + 1; - else - nids = val - val_range + 1; - } - need_size += nids; - if (need_size > idseq->size) { - idseq->size = need_size + 10; - idseq->id = erts_realloc(ERTS_ALC_T_TMP_CPU_IDS, - idseq->id, - sizeof(int)*idseq->size); - } - if (nids == 1) - idseq->id[ix++] = val; - else if (val_range > val) { - for (; val <= val_range; val++) - idseq->id[ix++] = val; - } - else { - for (; val >= val_range; val--) - idseq->id[ix++] = val; - } - if (*c != ',') - break; - c++; - } - *str = c; - idseq->used = ix; - return ERTS_INIT_CPU_TOPOLOGY_OK; -} - -static int -get_cput_entry(ErtsCpuTopEntry *cput, char **str) -{ - int h; - char *c = *str; - - cput->logical.used = 0; - cput->thread.id[0] = 0; - cput->thread.used = 1; - cput->core.id[0] = 0; - cput->core.used = 1; - cput->processor_node.id[0] = -1; - cput->processor_node.used = 1; - cput->processor.id[0] = 0; - cput->processor.used = 1; - cput->node.id[0] = -1; - cput->node.used = 1; - - h = ERTS_TOPOLOGY_MAX_DEPTH; - while (*c != ':' && *c != '\0') { - int res; - ErtsCpuTopIdSeq *idseqp; - switch (*c++) { - case 'L': - if (h <= ERTS_TOPOLOGY_LOGICAL) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY; - idseqp = &cput->logical; - h = ERTS_TOPOLOGY_LOGICAL; - break; - case 't': - case 'T': - if (h <= ERTS_TOPOLOGY_THREAD) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY; - idseqp = &cput->thread; - h = ERTS_TOPOLOGY_THREAD; - break; - case 'c': - case 'C': - if (h <= ERTS_TOPOLOGY_CORE) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY; - idseqp = &cput->core; - h = ERTS_TOPOLOGY_CORE; - break; - case 'p': - case 'P': - if (h <= ERTS_TOPOLOGY_PROCESSOR) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY; - idseqp = &cput->processor; - h = ERTS_TOPOLOGY_PROCESSOR; - break; - case 'n': - case 'N': - if (h <= ERTS_TOPOLOGY_PROCESSOR) { - do_node: - if (h <= ERTS_TOPOLOGY_NODE) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY; - idseqp = &cput->node; - h = ERTS_TOPOLOGY_NODE; - } - else { - int p_node = 0; - char *p_chk = c; - while (*p_chk != '\0' && *p_chk != ':') { - if (*p_chk == 'p' || *p_chk == 'P') { - p_node = 1; - break; - } - p_chk++; - } - if (!p_node) - goto do_node; - if (h <= ERTS_TOPOLOGY_PROCESSOR_NODE) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY; - idseqp = &cput->processor_node; - h = ERTS_TOPOLOGY_PROCESSOR_NODE; - } - break; - default: - return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_TYPE; - } - res = get_cput_id_seq(idseqp, &c); - if (res != ERTS_INIT_CPU_TOPOLOGY_OK) - return res; - } - - if (cput->logical.used < 1) - return ERTS_INIT_CPU_TOPOLOGY_MISSING_LID; - - if (*c == ':') { - c++; - } - - if (cput->thread.used != 1 - && cput->thread.used != cput->logical.used) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE; - if (cput->core.used != 1 - && cput->core.used != cput->logical.used) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE; - if (cput->processor_node.used != 1 - && cput->processor_node.used != cput->logical.used) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE; - if (cput->processor.used != 1 - && cput->processor.used != cput->logical.used) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE; - if (cput->node.used != 1 - && cput->node.used != cput->logical.used) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE; - - *str = c; - return ERTS_INIT_CPU_TOPOLOGY_OK; -} - -static int -verify_topology(erts_cpu_topology_t *cpudata, int size) -{ - if (size > 0) { - int *logical; - int node, processor, no_nodes, i; - - /* Verify logical ids */ - logical = erts_alloc(ERTS_ALC_T_TMP, sizeof(int)*size); - - for (i = 0; i < size; i++) - logical[i] = cpudata[i].logical; - - qsort(logical, size, sizeof(int), int_cmp); - for (i = 0; i < size-1; i++) { - if (logical[i] == logical[i+1]) { - erts_free(ERTS_ALC_T_TMP, logical); - return ERTS_INIT_CPU_TOPOLOGY_NOT_UNIQUE_LIDS; - } - } - - erts_free(ERTS_ALC_T_TMP, logical); - - qsort(cpudata, size, sizeof(erts_cpu_topology_t), processor_order_cmp); - - /* Verify unique entities */ - - for (i = 1; i < size; i++) { - if (cpudata[i-1].processor == cpudata[i].processor - && cpudata[i-1].node == cpudata[i].node - && (cpudata[i-1].processor_node - == cpudata[i].processor_node) - && cpudata[i-1].core == cpudata[i].core - && cpudata[i-1].thread == cpudata[i].thread) { - return ERTS_INIT_CPU_TOPOLOGY_NOT_UNIQUE_ENTITIES; - } - } - - /* Verify numa nodes */ - node = cpudata[0].node; - processor = cpudata[0].processor; - no_nodes = cpudata[0].node < 0 && cpudata[0].processor_node < 0; - for (i = 1; i < size; i++) { - if (no_nodes) { - if (cpudata[i].node >= 0 || cpudata[i].processor_node >= 0) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_NODES; - } - else { - if (cpudata[i].processor == processor && cpudata[i].node != node) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_NODES; - node = cpudata[i].node; - processor = cpudata[i].processor; - if (node >= 0 && cpudata[i].processor_node >= 0) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_NODES; - if (node < 0 && cpudata[i].processor_node < 0) - return ERTS_INIT_CPU_TOPOLOGY_INVALID_NODES; - } - } - } - - return ERTS_INIT_CPU_TOPOLOGY_OK; -} - -int -erts_init_cpu_topology(char *topology_str) -{ - ErtsCpuTopEntry cput; - int need_size; - char *c; - int ix; - int error = ERTS_INIT_CPU_TOPOLOGY_OK; - - if (user_cpudata) - erts_free(ERTS_ALC_T_CPUDATA, user_cpudata); - user_cpudata_size = 10; - - user_cpudata = erts_alloc(ERTS_ALC_T_CPUDATA, - (sizeof(erts_cpu_topology_t) - * user_cpudata_size)); - - init_cpu_top_entry(&cput); - - ix = 0; - need_size = 0; - - c = topology_str; - if (*c == '\0') { - error = ERTS_INIT_CPU_TOPOLOGY_MISSING; - goto fail; - } - do { - int r; - error = get_cput_entry(&cput, &c); - if (error != ERTS_INIT_CPU_TOPOLOGY_OK) - goto fail; - need_size += cput.logical.used; - if (user_cpudata_size < need_size) { - user_cpudata_size = need_size + 10; - user_cpudata = erts_realloc(ERTS_ALC_T_CPUDATA, - user_cpudata, - (sizeof(erts_cpu_topology_t) - * user_cpudata_size)); - } - - ASSERT(cput.thread.used == 1 - || cput.thread.used == cput.logical.used); - ASSERT(cput.core.used == 1 - || cput.core.used == cput.logical.used); - ASSERT(cput.processor_node.used == 1 - || cput.processor_node.used == cput.logical.used); - ASSERT(cput.processor.used == 1 - || cput.processor.used == cput.logical.used); - ASSERT(cput.node.used == 1 - || cput.node.used == cput.logical.used); - - for (r = 0; r < cput.logical.used; r++) { - user_cpudata[ix].logical = cput.logical.id[r]; - user_cpudata[ix].thread = - cput.thread.id[cput.thread.used == 1 ? 0 : r]; - user_cpudata[ix].core = - cput.core.id[cput.core.used == 1 ? 0 : r]; - user_cpudata[ix].processor_node = - cput.processor_node.id[cput.processor_node.used == 1 ? 0 : r]; - user_cpudata[ix].processor = - cput.processor.id[cput.processor.used == 1 ? 0 : r]; - user_cpudata[ix].node = - cput.node.id[cput.node.used == 1 ? 0 : r]; - ix++; - } - } while (*c != '\0'); - - if (user_cpudata_size != ix) { - user_cpudata_size = ix; - user_cpudata = erts_realloc(ERTS_ALC_T_CPUDATA, - user_cpudata, - (sizeof(erts_cpu_topology_t) - * user_cpudata_size)); - } - - error = verify_topology(user_cpudata, user_cpudata_size); - if (error == ERTS_INIT_CPU_TOPOLOGY_OK) { - destroy_cpu_top_entry(&cput); - return ERTS_INIT_CPU_TOPOLOGY_OK; - } - - fail: - if (user_cpudata) - erts_free(ERTS_ALC_T_CPUDATA, user_cpudata); - user_cpudata_size = 0; - destroy_cpu_top_entry(&cput); - return error; -} - -#define ERTS_GET_CPU_TOPOLOGY_ERROR -1 -#define ERTS_GET_USED_CPU_TOPOLOGY 0 -#define ERTS_GET_DETECTED_CPU_TOPOLOGY 1 -#define ERTS_GET_DEFINED_CPU_TOPOLOGY 2 - -static Eterm get_cpu_topology_term(Process *c_p, int type); - -Eterm -erts_set_cpu_topology(Process *c_p, Eterm term) -{ - erts_cpu_topology_t *cpudata = NULL; - int cpudata_size = 0; - Eterm res; - - erts_smp_rwmtx_rwlock(&erts_cpu_bind_rwmtx); - res = get_cpu_topology_term(c_p, ERTS_GET_USED_CPU_TOPOLOGY); - if (term == am_undefined) { - if (user_cpudata) - erts_free(ERTS_ALC_T_CPUDATA, user_cpudata); - user_cpudata = NULL; - user_cpudata_size = 0; - - if (cpu_bind_order != ERTS_CPU_BIND_NONE && system_cpudata) { - cpudata_size = system_cpudata_size; - cpudata = erts_alloc(ERTS_ALC_T_TMP, - (sizeof(erts_cpu_topology_t) - * cpudata_size)); - - sys_memcpy((void *) cpudata, - (void *) system_cpudata, - sizeof(erts_cpu_topology_t)*cpudata_size); - } - } - else if (is_not_list(term)) { - error: - res = THE_NON_VALUE; - goto done; - } - else { - Eterm list = term; - int ix = 0; - - cpudata_size = 100; - cpudata = erts_alloc(ERTS_ALC_T_TMP, - (sizeof(erts_cpu_topology_t) - * cpudata_size)); - - while (is_list(list)) { - Eterm *lp = list_val(list); - Eterm cpu = CAR(lp); - Eterm* tp; - Sint id; - - if (is_not_tuple(cpu)) - goto error; - - tp = tuple_val(cpu); - - if (arityval(tp[0]) != 7 || tp[1] != am_cpu) - goto error; - - if (ix >= cpudata_size) { - cpudata_size += 100; - cpudata = erts_realloc(ERTS_ALC_T_TMP, - cpudata, - (sizeof(erts_cpu_topology_t) - * cpudata_size)); - } - - id = signed_val(tp[2]); - if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id) - goto error; - cpudata[ix].node = (int) id; - - id = signed_val(tp[3]); - if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id) - goto error; - cpudata[ix].processor = (int) id; - - id = signed_val(tp[4]); - if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id) - goto error; - cpudata[ix].processor_node = (int) id; - - id = signed_val(tp[5]); - if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id) - goto error; - cpudata[ix].core = (int) id; - - id = signed_val(tp[6]); - if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id) - goto error; - cpudata[ix].thread = (int) id; - - id = signed_val(tp[7]); - if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id) - goto error; - cpudata[ix].logical = (int) id; - - list = CDR(lp); - ix++; - } - - if (is_not_nil(list)) - goto error; - - cpudata_size = ix; - - if (ERTS_INIT_CPU_TOPOLOGY_OK != verify_topology(cpudata, cpudata_size)) - goto error; - - if (user_cpudata_size != cpudata_size) { - if (user_cpudata) - erts_free(ERTS_ALC_T_CPUDATA, user_cpudata); - user_cpudata = erts_alloc(ERTS_ALC_T_CPUDATA, - sizeof(erts_cpu_topology_t)*cpudata_size); - user_cpudata_size = cpudata_size; - } - - sys_memcpy((void *) user_cpudata, - (void *) cpudata, - sizeof(erts_cpu_topology_t)*cpudata_size); - } - - make_reader_groups_map(NULL); - - signal_schedulers_bind_change(cpudata, cpudata_size); - - done: - erts_smp_rwmtx_rwunlock(&erts_cpu_bind_rwmtx); - - if (cpudata) - erts_free(ERTS_ALC_T_TMP, cpudata); - - return res; -} - -static Eterm -bound_schedulers_term(ErtsCpuBindOrder order) -{ - switch (order) { - case ERTS_CPU_BIND_SPREAD: { - ERTS_DECL_AM(spread); - return AM_spread; - } - case ERTS_CPU_BIND_PROCESSOR_SPREAD: { - ERTS_DECL_AM(processor_spread); - return AM_processor_spread; - } - case ERTS_CPU_BIND_THREAD_SPREAD: { - ERTS_DECL_AM(thread_spread); - return AM_thread_spread; - } - case ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD: { - ERTS_DECL_AM(thread_no_node_processor_spread); - return AM_thread_no_node_processor_spread; - } - case ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD: { - ERTS_DECL_AM(no_node_processor_spread); - return AM_no_node_processor_spread; - } - case ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD: { - ERTS_DECL_AM(no_node_thread_spread); - return AM_no_node_thread_spread; - } - case ERTS_CPU_BIND_NO_SPREAD: { - ERTS_DECL_AM(no_spread); - return AM_no_spread; - } - case ERTS_CPU_BIND_NONE: { - ERTS_DECL_AM(unbound); - return AM_unbound; - } - default: - ASSERT(0); - return THE_NON_VALUE; - } -} - -Eterm -erts_bound_schedulers_term(Process *c_p) -{ - ErtsCpuBindOrder order; - erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx); - order = cpu_bind_order; - erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx); - return bound_schedulers_term(order); -} - -static void -create_tmp_cpu_topology_copy(erts_cpu_topology_t **cpudata, int *cpudata_size) -{ - if (user_cpudata) { - *cpudata_size = user_cpudata_size; - *cpudata = erts_alloc(ERTS_ALC_T_TMP, - (sizeof(erts_cpu_topology_t) - * (*cpudata_size))); - sys_memcpy((void *) *cpudata, - (void *) user_cpudata, - sizeof(erts_cpu_topology_t)*(*cpudata_size)); - } - else if (system_cpudata) { - *cpudata_size = system_cpudata_size; - *cpudata = erts_alloc(ERTS_ALC_T_TMP, - (sizeof(erts_cpu_topology_t) - * (*cpudata_size))); - sys_memcpy((void *) *cpudata, - (void *) system_cpudata, - sizeof(erts_cpu_topology_t)*(*cpudata_size)); - } - else { - *cpudata = NULL; - *cpudata_size = 0; - } -} - -static void -destroy_tmp_cpu_topology_copy(erts_cpu_topology_t *cpudata) -{ - if (cpudata) - erts_free(ERTS_ALC_T_TMP, cpudata); -} - -Eterm -erts_bind_schedulers(Process *c_p, Eterm how) -{ - Eterm res; - erts_cpu_topology_t *cpudata; - int cpudata_size; - ErtsCpuBindOrder old_cpu_bind_order; - - erts_smp_rwmtx_rwlock(&erts_cpu_bind_rwmtx); - - if (erts_bind_to_cpu(erts_cpuinfo, -1) == -ENOTSUP) { - ERTS_BIF_PREP_ERROR(res, c_p, EXC_NOTSUP); - } - else { - - old_cpu_bind_order = cpu_bind_order; - - if (ERTS_IS_ATOM_STR("default_bind", how)) - cpu_bind_order = ERTS_CPU_BIND_DEFAULT_BIND; - else if (ERTS_IS_ATOM_STR("spread", how)) - cpu_bind_order = ERTS_CPU_BIND_SPREAD; - else if (ERTS_IS_ATOM_STR("processor_spread", how)) - cpu_bind_order = ERTS_CPU_BIND_PROCESSOR_SPREAD; - else if (ERTS_IS_ATOM_STR("thread_spread", how)) - cpu_bind_order = ERTS_CPU_BIND_THREAD_SPREAD; - else if (ERTS_IS_ATOM_STR("thread_no_node_processor_spread", how)) - cpu_bind_order = ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD; - else if (ERTS_IS_ATOM_STR("no_node_processor_spread", how)) - cpu_bind_order = ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD; - else if (ERTS_IS_ATOM_STR("no_node_thread_spread", how)) - cpu_bind_order = ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD; - else if (ERTS_IS_ATOM_STR("no_spread", how)) - cpu_bind_order = ERTS_CPU_BIND_NO_SPREAD; - else if (ERTS_IS_ATOM_STR("unbound", how)) - cpu_bind_order = ERTS_CPU_BIND_NONE; - else { - cpu_bind_order = old_cpu_bind_order; - ERTS_BIF_PREP_ERROR(res, c_p, BADARG); - goto done; - } - - create_tmp_cpu_topology_copy(&cpudata, &cpudata_size); - - if (!cpudata) { - cpu_bind_order = old_cpu_bind_order; - ERTS_BIF_PREP_ERROR(res, c_p, BADARG); - goto done; - } - - signal_schedulers_bind_change(cpudata, cpudata_size); - - destroy_tmp_cpu_topology_copy(cpudata); - - res = bound_schedulers_term(old_cpu_bind_order); - } - - done: - - erts_smp_rwmtx_rwunlock(&erts_cpu_bind_rwmtx); - - return res; -} - -Eterm -erts_fake_scheduler_bindings(Process *p, Eterm how) -{ - ErtsCpuBindOrder fake_cpu_bind_order; - erts_cpu_topology_t *cpudata; - int cpudata_size; - Eterm res; - - if (ERTS_IS_ATOM_STR("default_bind", how)) - fake_cpu_bind_order = ERTS_CPU_BIND_DEFAULT_BIND; - else if (ERTS_IS_ATOM_STR("spread", how)) - fake_cpu_bind_order = ERTS_CPU_BIND_SPREAD; - else if (ERTS_IS_ATOM_STR("processor_spread", how)) - fake_cpu_bind_order = ERTS_CPU_BIND_PROCESSOR_SPREAD; - else if (ERTS_IS_ATOM_STR("thread_spread", how)) - fake_cpu_bind_order = ERTS_CPU_BIND_THREAD_SPREAD; - else if (ERTS_IS_ATOM_STR("thread_no_node_processor_spread", how)) - fake_cpu_bind_order = ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD; - else if (ERTS_IS_ATOM_STR("no_node_processor_spread", how)) - fake_cpu_bind_order = ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD; - else if (ERTS_IS_ATOM_STR("no_node_thread_spread", how)) - fake_cpu_bind_order = ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD; - else if (ERTS_IS_ATOM_STR("no_spread", how)) - fake_cpu_bind_order = ERTS_CPU_BIND_NO_SPREAD; - else if (ERTS_IS_ATOM_STR("unbound", how)) - fake_cpu_bind_order = ERTS_CPU_BIND_NONE; - else { - ERTS_BIF_PREP_ERROR(res, p, BADARG); - return res; - } - - erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx); - create_tmp_cpu_topology_copy(&cpudata, &cpudata_size); - erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx); - - if (!cpudata || fake_cpu_bind_order == ERTS_CPU_BIND_NONE) - ERTS_BIF_PREP_RET(res, am_false); - else { - int i; - Eterm *hp; - - cpu_bind_order_sort(cpudata, cpudata_size, fake_cpu_bind_order, 1); - -#ifdef ERTS_FAKE_SCHED_BIND_PRINT_SORTED_CPU_DATA - - erts_fprintf(stderr, "node: "); - for (i = 0; i < cpudata_size; i++) - erts_fprintf(stderr, " %2d", cpudata[i].node); - erts_fprintf(stderr, "\n"); - erts_fprintf(stderr, "processor: "); - for (i = 0; i < cpudata_size; i++) - erts_fprintf(stderr, " %2d", cpudata[i].processor); - erts_fprintf(stderr, "\n"); - if (fake_cpu_bind_order != ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD - && fake_cpu_bind_order != ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD - && fake_cpu_bind_order != ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD) { - erts_fprintf(stderr, "processor_node:"); - for (i = 0; i < cpudata_size; i++) - erts_fprintf(stderr, " %2d", cpudata[i].processor_node); - erts_fprintf(stderr, "\n"); - } - erts_fprintf(stderr, "core: "); - for (i = 0; i < cpudata_size; i++) - erts_fprintf(stderr, " %2d", cpudata[i].core); - erts_fprintf(stderr, "\n"); - erts_fprintf(stderr, "thread: "); - for (i = 0; i < cpudata_size; i++) - erts_fprintf(stderr, " %2d", cpudata[i].thread); - erts_fprintf(stderr, "\n"); - erts_fprintf(stderr, "logical: "); - for (i = 0; i < cpudata_size; i++) - erts_fprintf(stderr, " %2d", cpudata[i].logical); - erts_fprintf(stderr, "\n"); -#endif - - hp = HAlloc(p, cpudata_size+1); - ERTS_BIF_PREP_RET(res, make_tuple(hp)); - *hp++ = make_arityval((Uint) cpudata_size); - for (i = 0; i < cpudata_size; i++) - *hp++ = make_small((Uint) cpudata[i].logical); - } - - destroy_tmp_cpu_topology_copy(cpudata); - - return res; -} - -Eterm -erts_get_schedulers_binds(Process *c_p) -{ - int ix; - ERTS_DECL_AM(unbound); - Eterm *hp = HAlloc(c_p, erts_no_schedulers+1); - Eterm res = make_tuple(hp); - - *(hp++) = make_arityval(erts_no_schedulers); - erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx); - for (ix = 1; ix <= erts_no_schedulers; ix++) - *(hp++) = (scheduler2cpu_map[ix].bound_id >= 0 - ? make_small(scheduler2cpu_map[ix].bound_id) - : AM_unbound); - erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx); - return res; -} - -static Eterm -bld_topology_term(Eterm **hpp, - Uint *hszp, - erts_cpu_topology_t *cpudata, - int size) -{ - Eterm res = NIL; - int i; - - if (size == 0) - return am_undefined; - - for (i = size-1; i >= 0; i--) { - res = erts_bld_cons(hpp, - hszp, - erts_bld_tuple(hpp, - hszp, - 7, - am_cpu, - make_small(cpudata[i].node), - make_small(cpudata[i].processor), - make_small(cpudata[i].processor_node), - make_small(cpudata[i].core), - make_small(cpudata[i].thread), - make_small(cpudata[i].logical)), - res); - } - return res; -} - -static Eterm -get_cpu_topology_term(Process *c_p, int type) -{ -#ifdef DEBUG - Eterm *hp_end; -#endif - Eterm *hp; - Uint hsz; - Eterm res = THE_NON_VALUE; - erts_cpu_topology_t *cpudata = NULL; - int size = 0; - - switch (type) { - case ERTS_GET_USED_CPU_TOPOLOGY: - if (user_cpudata) - goto defined; - else - goto detected; - case ERTS_GET_DETECTED_CPU_TOPOLOGY: - detected: - if (!system_cpudata) - res = am_undefined; - else { - size = system_cpudata_size; - cpudata = erts_alloc(ERTS_ALC_T_TMP, - (sizeof(erts_cpu_topology_t) - * size)); - sys_memcpy((void *) cpudata, - (void *) system_cpudata, - sizeof(erts_cpu_topology_t)*size); - } - break; - case ERTS_GET_DEFINED_CPU_TOPOLOGY: - defined: - if (!user_cpudata) - res = am_undefined; - else { - size = user_cpudata_size; - cpudata = user_cpudata; - } - break; - default: - erl_exit(ERTS_ABORT_EXIT, "Bad cpu topology type: %d\n", type); - break; - } - - if (res == am_undefined) { - ASSERT(!cpudata); - return res; - } - - hsz = 0; - - bld_topology_term(NULL, &hsz, - cpudata, size); - - hp = HAlloc(c_p, hsz); - -#ifdef DEBUG - hp_end = hp + hsz; -#endif - - res = bld_topology_term(&hp, NULL, - cpudata, size); - - ASSERT(hp_end == hp); - - if (cpudata && cpudata != system_cpudata && cpudata != user_cpudata) - erts_free(ERTS_ALC_T_TMP, cpudata); - - return res; -} - -Eterm -erts_get_cpu_topology_term(Process *c_p, Eterm which) -{ - Eterm res; - int type; - erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx); - if (ERTS_IS_ATOM_STR("used", which)) - type = ERTS_GET_USED_CPU_TOPOLOGY; - else if (ERTS_IS_ATOM_STR("detected", which)) - type = ERTS_GET_DETECTED_CPU_TOPOLOGY; - else if (ERTS_IS_ATOM_STR("defined", which)) - type = ERTS_GET_DEFINED_CPU_TOPOLOGY; - else - type = ERTS_GET_CPU_TOPOLOGY_ERROR; - if (type == ERTS_GET_CPU_TOPOLOGY_ERROR) - res = THE_NON_VALUE; - else - res = get_cpu_topology_term(c_p, type); - erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx); - return res; -} - -static void -early_cpu_bind_init(void) -{ - user_cpudata = NULL; - user_cpudata_size = 0; - - system_cpudata_size = erts_get_cpu_topology_size(erts_cpuinfo); - system_cpudata = erts_alloc(ERTS_ALC_T_CPUDATA, - (sizeof(erts_cpu_topology_t) - * system_cpudata_size)); - - cpu_bind_order = ERTS_CPU_BIND_UNDEFINED; - - reader_groups_available_cpu_check = 1; - reader_groups_logical_processors = 0; - reader_groups_map_size = 0; - reader_groups_map = NULL; - - if (!erts_get_cpu_topology(erts_cpuinfo, system_cpudata) - || ERTS_INIT_CPU_TOPOLOGY_OK != verify_topology(system_cpudata, - system_cpudata_size)) { - erts_free(ERTS_ALC_T_CPUDATA, system_cpudata); - system_cpudata = NULL; - system_cpudata_size = 0; - } -} - -static void -late_cpu_bind_init(void) -{ - int ix; - - erts_smp_rwmtx_init(&erts_cpu_bind_rwmtx, "cpu_bind"); - - scheduler2cpu_map = erts_alloc(ERTS_ALC_T_CPUDATA, - (sizeof(ErtsCpuBindData) - * (erts_no_schedulers+1))); - for (ix = 1; ix <= erts_no_schedulers; ix++) { - scheduler2cpu_map[ix].bind_id = -1; - scheduler2cpu_map[ix].bound_id = -1; - } - - if (cpu_bind_order == ERTS_CPU_BIND_UNDEFINED) { - int ncpus = erts_get_cpu_configured(erts_cpuinfo); - if (ncpus < 1 || erts_no_schedulers < ncpus) - cpu_bind_order = ERTS_CPU_BIND_NONE; - else - cpu_bind_order = ((system_cpudata || user_cpudata) - && (erts_bind_to_cpu(erts_cpuinfo, -1) != -ENOTSUP) - ? ERTS_CPU_BIND_DEFAULT_BIND - : ERTS_CPU_BIND_NONE); - } - - make_reader_groups_map(NULL); - - if (cpu_bind_order != ERTS_CPU_BIND_NONE) { - erts_cpu_topology_t *cpudata; - int cpudata_size; - create_tmp_cpu_topology_copy(&cpudata, &cpudata_size); - signal_schedulers_bind_change(cpudata, cpudata_size); - destroy_tmp_cpu_topology_copy(cpudata); - } -} - -int -erts_update_cpu_info(void) -{ - int changed; - erts_smp_rwmtx_rwlock(&erts_cpu_bind_rwmtx); - changed = erts_cpu_info_update(erts_cpuinfo); - if (changed) { - erts_cpu_topology_t *cpudata; - int cpudata_size; - - if (system_cpudata) - erts_free(ERTS_ALC_T_CPUDATA, system_cpudata); - - system_cpudata_size = erts_get_cpu_topology_size(erts_cpuinfo); - if (!system_cpudata_size) - system_cpudata = NULL; - else { - system_cpudata = erts_alloc(ERTS_ALC_T_CPUDATA, - (sizeof(erts_cpu_topology_t) - * system_cpudata_size)); - - if (!erts_get_cpu_topology(erts_cpuinfo, system_cpudata) - || (ERTS_INIT_CPU_TOPOLOGY_OK - != verify_topology(system_cpudata, - system_cpudata_size))) { - erts_free(ERTS_ALC_T_CPUDATA, system_cpudata); - system_cpudata = NULL; - system_cpudata_size = 0; - } - } - - create_tmp_cpu_topology_copy(&cpudata, &cpudata_size); - signal_schedulers_bind_change(cpudata, cpudata_size); - destroy_tmp_cpu_topology_copy(cpudata); - } - erts_smp_rwmtx_rwunlock(&erts_cpu_bind_rwmtx); - return changed; -} - #ifdef ERTS_SMP static void @@ -6884,10 +5073,10 @@ Process *schedule(Process *p, int calls) { ErtsRunQueue *rq; ErtsRunPrioQueue *rpq; - long dt; + erts_aint_t dt; ErtsSchedulerData *esdp; int context_reds; - long fcalls; + int fcalls; int input_reductions; int actual_reds; int reds; @@ -6910,7 +5099,7 @@ Process *schedule(Process *p, int calls) esdp = erts_get_scheduler_data(); rq = erts_get_runq_current(esdp); ASSERT(esdp); - fcalls = erts_smp_atomic_read(&function_calls); + fcalls = (int) erts_smp_atomic32_read(&function_calls); actual_reds = reds = 0; erts_smp_runq_lock(rq); } else { @@ -6928,7 +5117,7 @@ Process *schedule(Process *p, int calls) reds = ERTS_PROC_MIN_CONTEXT_SWITCH_REDS_COST; esdp->virtual_reds = 0; - fcalls = erts_smp_atomic_addtest(&function_calls, reds); + fcalls = (int) erts_smp_atomic32_addtest(&function_calls, reds); ASSERT(esdp && esdp == erts_get_scheduler_data()); rq = erts_get_runq_current(esdp); @@ -7029,10 +5218,10 @@ Process *schedule(Process *p, int calls) ERTS_SMP_CHK_NO_PROC_LOCKS; - dt = do_time_read_and_reset(); + dt = erts_do_time_read_and_reset(); if (dt) { erts_smp_runq_unlock(rq); - bump_timer(dt); + erts_bump_timer(dt); erts_smp_runq_lock(rq); } BM_STOP_TIMER(system); @@ -7061,15 +5250,15 @@ Process *schedule(Process *p, int calls) | ERTS_RUNQ_FLG_CHK_CPU_BIND | ERTS_RUNQ_FLG_SUSPENDED)) { if ((rq->flags & ERTS_RUNQ_FLG_SUSPENDED) - || (erts_smp_atomic_read(&esdp->ssi->flags) + || (erts_smp_atomic32_read(&esdp->ssi->flags) & ERTS_SSI_FLG_SUSPENDED)) { - ASSERT(erts_smp_atomic_read(&esdp->ssi->flags) + ASSERT(erts_smp_atomic32_read(&esdp->ssi->flags) & ERTS_SSI_FLG_SUSPENDED); suspend_scheduler(esdp); } if ((rq->flags & ERTS_RUNQ_FLG_CHK_CPU_BIND) - || erts_smp_atomic_read(&esdp->chk_cpu_bind)) { - check_cpu_bind(esdp); + || erts_smp_atomic32_read(&esdp->chk_cpu_bind)) { + erts_sched_check_cpu_bind(esdp); } } @@ -7077,7 +5266,7 @@ Process *schedule(Process *p, int calls) || defined(ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK) { ErtsSchedulerSleepInfo *ssi = esdp->ssi; - long aux_work = erts_smp_atomic_read(&ssi->aux_work); + erts_aint32_t aux_work = erts_smp_atomic32_read(&ssi->aux_work); if (aux_work) { erts_smp_runq_unlock(rq); #ifdef ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK @@ -7119,9 +5308,9 @@ Process *schedule(Process *p, int calls) if (rq->flags & (ERTS_RUNQ_FLG_SHARED_RUNQ | ERTS_RUNQ_FLG_SUSPENDED)) { if ((rq->flags & ERTS_RUNQ_FLG_SUSPENDED) - || (erts_smp_atomic_read(&esdp->ssi->flags) + || (erts_smp_atomic32_read(&esdp->ssi->flags) & ERTS_SSI_FLG_SUSPENDED)) { - ASSERT(erts_smp_atomic_read(&esdp->ssi->flags) + ASSERT(erts_smp_atomic32_read(&esdp->ssi->flags) & ERTS_SSI_FLG_SUSPENDED); non_empty_runq(rq); goto continue_check_activities_to_run; @@ -7163,19 +5352,21 @@ Process *schedule(Process *p, int calls) * Schedule system-level activities. */ - erts_smp_atomic_set(&function_calls, 0); + erts_smp_atomic32_set(&function_calls, 0); fcalls = 0; + ASSERT(!erts_port_task_have_outstanding_io_tasks()); + #ifdef ERTS_SMP /* erts_sys_schedule_interrupt(0); */ #endif erts_smp_runq_unlock(rq); erl_sys_schedule(runnable); - dt = do_time_read_and_reset(); - if (dt) bump_timer(dt); + dt = erts_do_time_read_and_reset(); + if (dt) erts_bump_timer(dt); #ifdef ERTS_SMP erts_smp_runq_lock(rq); - erts_smp_atomic_set(&doing_sys_schedule, 0); + clear_sys_scheduling(); goto continue_check_activities_to_run; #else if (!runnable) @@ -7203,7 +5394,7 @@ Process *schedule(Process *p, int calls) if (erts_common_run_queue->waiting) wake_scheduler(erts_common_run_queue, 0, 1); } - else if (erts_smp_atomic_read(&no_empty_run_queues) != 0) { + else if (erts_smp_atomic32_read(&no_empty_run_queues) != 0) { wake_scheduler_on_empty_runq(rq); rq->wakeup_other = 0; } @@ -7497,6 +5688,15 @@ erts_schedule_misc_op(void (*func)(void *), void *arg) ErtsRunQueue *rq = erts_get_runq_current(NULL); ErtsMiscOpList *molp = misc_op_list_alloc(); + if (!rq) { + /* + * 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); + } + erts_smp_runq_lock(rq); while (rq->misc.evac_runq) { @@ -7651,7 +5851,7 @@ erts_test_next_pid(int set, Uint next) Uint erts_process_count(void) { - long res = erts_smp_atomic_read(&process_count); + erts_aint32_t res = erts_smp_atomic32_read(&process_count); ASSERT(res >= 0); return (Uint) res; } @@ -7700,7 +5900,7 @@ alloc_process(void) ASSERT(!process_tab[p_next]); process_tab[p_next] = p; - erts_smp_atomic_inc(&process_count); + erts_smp_atomic32_inc(&process_count); p->id = make_internal_pid(p_serial << p_serial_shift | p_next); if (p->id == ERTS_INVALID_PID) { /* Do not use the invalid pid; change serial */ @@ -7826,7 +6026,7 @@ erl_create_process(Process* parent, /* Parent of process (default group leader). p->min_heap_size = H_MIN_SIZE; p->min_vheap_size = BIN_VH_MIN_SIZE; p->prio = PRIORITY_NORMAL; - p->max_gen_gcs = (Uint16) erts_smp_atomic_read(&erts_max_gen_gcs); + p->max_gen_gcs = (Uint16) erts_smp_atomic32_read(&erts_max_gen_gcs); } p->skipped = 0; ASSERT(p->min_heap_size == erts_next_heap_size(p->min_heap_size, 0)); @@ -8879,11 +7079,11 @@ static void doit_exit_monitor(ErtsMonitor *mon, void *vpcontext) DeclareTmpHeapNoproc(lhp,3); ErtsProcLocks rp_locks = (ERTS_PROC_LOCK_LINK | ERTS_PROC_LOCKS_MSG_SEND); - UseTmpHeapNoproc(3); rp = erts_pid2proc(NULL, 0, mon->pid, rp_locks); if (rp == NULL) { goto done; } + UseTmpHeapNoproc(3); rmon = erts_remove_monitor(&(rp->monitors),mon->ref); if (rmon) { erts_destroy_monitor(rmon); @@ -9283,8 +7483,8 @@ continue_exit_process(Process *p p->status_flags = 0; #endif process_tab[pix] = NULL; /* Time of death! */ - ASSERT(erts_smp_atomic_read(&process_count) > 0); - erts_smp_atomic_dec(&process_count); + ASSERT(erts_smp_atomic32_read(&process_count) > 0); + erts_smp_atomic32_dec(&process_count); #ifdef ERTS_SMP erts_pix_unlock(pix_lock); @@ -9424,7 +7624,7 @@ cancel_timer(Process* p) #ifdef ERTS_SMP erts_cancel_smp_ptimer(p->u.ptimer); #else - erl_cancel_timer(&p->u.tm); + erts_cancel_timer(&p->u.tm); #endif } @@ -9450,7 +7650,7 @@ set_timer(Process* p, Uint timeout) (ErlTimeoutProc) timeout_proc, timeout); #else - erl_set_timer(&p->u.tm, + erts_set_timer(&p->u.tm, (ErlTimeoutProc) timeout_proc, NULL, (void*) p, @@ -9498,7 +7698,7 @@ erts_program_counter_info(int to, void *to_arg, Process *p) * only cause problems. */ for (i = 0; i < p->arity; i++) - erts_print(to, to_arg, " %T\n", p->arg_reg[i]); + erts_print(to, to_arg, " %.*T\n", INT_MAX, p->arg_reg[i]); } } } |