From 988f5f5e8061ce2e135a314ca782788eda478a06 Mon Sep 17 00:00:00 2001 From: Lukas Larsson Date: Tue, 30 May 2017 16:35:18 +0200 Subject: erts: Move all I/O polling to a seperate thread --- erts/configure.in | 14 +- erts/doc/src/erl.xml | 35 +- erts/emulator/Makefile.in | 16 +- erts/emulator/beam/erl_alloc.types | 1 - erts/emulator/beam/erl_bif_info.c | 8 +- erts/emulator/beam/erl_init.c | 58 +- erts/emulator/beam/erl_lock_count.c | 20 - erts/emulator/beam/erl_port_task.c | 8 +- erts/emulator/beam/erl_process.c | 700 +++--- erts/emulator/beam/erl_process.h | 8 +- erts/emulator/beam/erl_vm.h | 2 - erts/emulator/beam/global.h | 1 - erts/emulator/beam/sys.h | 4 - erts/emulator/sys/common/erl_check_io.c | 2304 +++++++++---------- erts/emulator/sys/common/erl_check_io.h | 92 +- erts/emulator/sys/common/erl_poll.c | 2682 ++++++++-------------- erts/emulator/sys/common/erl_poll.h | 281 ++- erts/emulator/sys/common/erl_poll_api.h | 122 + erts/emulator/sys/unix/sys.c | 109 +- erts/emulator/sys/win32/erl_poll.c | 259 +-- erts/emulator/sys/win32/sys.c | 15 +- erts/emulator/test/driver_SUITE.erl | 49 +- erts/emulator/test/driver_SUITE_data/chkio_drv.c | 3 + erts/etc/common/erlexec.c | 22 + erts/etc/unix/etp-commands.in | 42 +- 25 files changed, 2893 insertions(+), 3962 deletions(-) create mode 100644 erts/emulator/sys/common/erl_poll_api.h (limited to 'erts') diff --git a/erts/configure.in b/erts/configure.in index 2cb446b470..ca4228270f 100644 --- a/erts/configure.in +++ b/erts/configure.in @@ -3545,6 +3545,7 @@ fi # AC_MSG_CHECKING(whether kernel poll support should be enabled) ERTS_ENABLE_KERNEL_POLL=no +ERTS_BUILD_FALLBACK_POLL=no case $enable_kernel_poll-$have_kernel_poll in no-*) AC_MSG_RESULT(no; disabled by user);; @@ -3555,11 +3556,16 @@ case $enable_kernel_poll-$have_kernel_poll in *) case $have_kernel_poll in epoll) - AC_DEFINE(HAVE_SYS_EPOLL_H, 1, [Define if you have the header file.]);; + AC_DEFINE(HAVE_SYS_EPOLL_H, 1, [Define if you have the header file.]) + ERTS_BUILD_FALLBACK_POLL=yes + ;; /dev/poll) - AC_DEFINE(HAVE_SYS_DEVPOLL_H, 1, [Define if you have header file.]);; + AC_DEFINE(HAVE_SYS_DEVPOLL_H, 1, [Define if you have header file.]) + ;; kqueue) - AC_DEFINE(HAVE_SYS_EVENT_H, 1, [Define if you have header file.]);; + AC_DEFINE(HAVE_SYS_EVENT_H, 1, [Define if you have header file.]) + ERTS_BUILD_FALLBACK_POLL=yes + ;; *) AC_MSG_ERROR(configure.in need to be updated);; esac @@ -3567,7 +3573,7 @@ case $enable_kernel_poll-$have_kernel_poll in AC_DEFINE(ERTS_ENABLE_KERNEL_POLL, 1, [Define if you have kernel poll and want to use it]) AC_MSG_RESULT([yes; $have_kernel_poll]);; esac -AC_SUBST(ERTS_ENABLE_KERNEL_POLL) +AC_SUBST(ERTS_BUILD_FALLBACK_POLL) AC_MSG_CHECKING([whether putenv() stores a copy of the key-value pair]) AC_TRY_RUN([ diff --git a/erts/doc/src/erl.xml b/erts/doc/src/erl.xml index 71fe08d4e6..d24c35d217 100644 --- a/erts/doc/src/erl.xml +++ b/erts/doc/src/erl.xml @@ -762,13 +762,36 @@ process_flag(message_queue_data, MQD).

- + +IOp PollSets -

Enables or disables the kernel poll functionality if supported by - the emulator. Defaults to (disabled). - If the emulator does not support kernel poll, and flag - is passed to the emulator, a warning is - issued at startup.

+

Sets the number of IO pollsets to use when polling for I/O. + This option is only used on platforms that support concurrent + updates of a pollset, otherwise the same number of pollsets + are used as IO poll threads. + The default is 1. +

+ + +IOt PollThreads + +

Sets the number of IO poll threads to use when polling for I/O. + The maximum number of poll threads allowed is 1024. The default is 1. +

+ + +IOPp PollSetsPercentage + +

Similar to +IOp but uses + percentages to set the number of IO pollsets to create, based on the + number of poll threads configured. If both +IOPp and +IOp + are used, +IOPp is ignored. +

+ + +IOPt PollThreadsPercentage + +

Similar to +IOt but uses + percentages to set the number of IO poll threads to create, based on + the number of schedulers configures. If both +IOPt and + +IOt are used, +IOPt is ignored. +

diff --git a/erts/emulator/Makefile.in b/erts/emulator/Makefile.in index 02a9a9a93b..915cad4e18 100644 --- a/erts/emulator/Makefile.in +++ b/erts/emulator/Makefile.in @@ -251,7 +251,7 @@ endif HIPEBEAMLDFLAGS=$($(ARCH)BEAMLDFLAGS) endif -ERTS_ENABLE_KERNEL_POLL=@ERTS_ENABLE_KERNEL_POLL@ +ERTS_BUILD_FALLBACK_POLL=@ERTS_BUILD_FALLBACK_POLL@ # # @@ -928,17 +928,15 @@ $(STATIC_NIF_LIBS) $(STATIC_DRIVER_LIBS): echo "=== Leaving lib after making static libs" endif -ifeq ($(ERTS_ENABLE_KERNEL_POLL),yes) +ifeq ($(ERTS_BUILD_FALLBACK_POLL),yes) OS_OBJS += $(OBJDIR)/erl_poll.kp.o \ - $(OBJDIR)/erl_check_io.kp.o \ - $(OBJDIR)/erl_poll.nkp.o \ - $(OBJDIR)/erl_check_io.nkp.o + $(OBJDIR)/erl_poll.nkp.o else -OS_OBJS += $(OBJDIR)/erl_poll.o \ - $(OBJDIR)/erl_check_io.o +OS_OBJS += $(OBJDIR)/erl_poll.o endif -OS_OBJS += $(OBJDIR)/erl_mseg.o \ +OS_OBJS += $(OBJDIR)/erl_check_io.o \ + $(OBJDIR)/erl_mseg.o \ $(OBJDIR)/erl_mmap.o \ $(OBJDIR)/erl_$(ERLANG_OSTYPE)_sys_ddll.o \ $(OBJDIR)/erl_mtrace_sys_wrap.o \ @@ -1112,7 +1110,7 @@ else SED_PREFIX= endif -ifeq ($(ERTS_ENABLE_KERNEL_POLL),yes) +ifeq ($(ERTS_BUILD_FALLBACK_POLL),yes) SED_SUFFIX=;$(SED_REPL_POLL);$(SED_REPL_CHK_IO) else SED_SUFFIX= diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types index 3ec1389d47..bf5487d31e 100644 --- a/erts/emulator/beam/erl_alloc.types +++ b/erts/emulator/beam/erl_alloc.types @@ -354,7 +354,6 @@ type DRV_EV_STATE LONG_LIVED SYSTEM driver_event_state type DRV_SEL_D_STATE FIXED_SIZE SYSTEM driver_select_data_state type NIF_SEL_D_STATE FIXED_SIZE SYSTEM enif_select_data_state type FD_LIST SHORT_LIVED SYSTEM fd_list -type ACTIVE_FD_ARR SHORT_LIVED SYSTEM active_fd_array type POLLSET LONG_LIVED SYSTEM pollset type POLLSET_UPDREQ SHORT_LIVED SYSTEM pollset_update_req type POLL_FDS LONG_LIVED SYSTEM poll_fds diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c index c782d7a8ce..d0c2accec6 100644 --- a/erts/emulator/beam/erl_bif_info.c +++ b/erts/emulator/beam/erl_bif_info.c @@ -98,9 +98,6 @@ static char erts_system_version[] = ("Erlang/OTP " ERLANG_OTP_RELEASE #ifdef HIPE " [hipe]" #endif -#ifdef ERTS_ENABLE_KERNEL_POLL - " [kernel-poll:%s]" -#endif #ifdef ET_DEBUG #if ET_DEBUG " [type-assertions]" @@ -372,9 +369,6 @@ erts_print_system_version(fmtfn_t to, void *arg, Process *c_p) , total, online , dirty_cpu, dirty_cpu_onln, dirty_io , erts_async_max_threads -#ifdef ERTS_ENABLE_KERNEL_POLL - , erts_use_kernel_poll ? "true" : "false" -#endif ); } @@ -2696,7 +2690,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1) BIF_RET(make_small(CONTEXT_REDS)); } else if (ERTS_IS_ATOM_STR("kernel_poll", BIF_ARG_1)) { #ifdef ERTS_ENABLE_KERNEL_POLL - BIF_RET(erts_use_kernel_poll ? am_true : am_false); + BIF_RET(am_true); #else BIF_RET(am_false); #endif diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c index 8c14c86219..b4d84f6b16 100644 --- a/erts/emulator/beam/erl_init.c +++ b/erts/emulator/beam/erl_init.c @@ -49,6 +49,7 @@ #include "erl_bif_unique.h" #define ERTS_WANT_TIMER_WHEEL_API #include "erl_time.h" +#include "erl_check_io.h" #ifdef HIPE #include "hipe_mode_switch.h" /* for hipe_mode_switch_init() */ @@ -76,8 +77,10 @@ const char etp_otp_release[] = ERLANG_OTP_RELEASE; const char etp_compile_date[] = ERLANG_COMPILE_DATE; const char etp_arch[] = ERLANG_ARCHITECTURE; #ifdef ERTS_ENABLE_KERNEL_POLL +const int erts_use_kernel_poll = 1; const int etp_kernel_poll_support = 1; #else +const int erts_use_kernel_poll = 0; const int etp_kernel_poll_support = 0; #endif #if defined(ARCH_64) @@ -203,16 +206,16 @@ int erts_no_line_info = 0; /* -L: Don't load line information */ */ ErtsModifiedTimings erts_modified_timings[] = { - /* 0 */ {make_small(0), CONTEXT_REDS, INPUT_REDUCTIONS}, - /* 1 */ {make_small(0), (3*CONTEXT_REDS)/4, 2*INPUT_REDUCTIONS}, - /* 2 */ {make_small(0), CONTEXT_REDS/2, INPUT_REDUCTIONS/2}, - /* 3 */ {make_small(0), (7*CONTEXT_REDS)/8, 3*INPUT_REDUCTIONS}, - /* 4 */ {make_small(0), CONTEXT_REDS/3, 3*INPUT_REDUCTIONS}, - /* 5 */ {make_small(0), (10*CONTEXT_REDS)/11, INPUT_REDUCTIONS/2}, - /* 6 */ {make_small(1), CONTEXT_REDS/4, 2*INPUT_REDUCTIONS}, - /* 7 */ {make_small(1), (5*CONTEXT_REDS)/7, INPUT_REDUCTIONS/3}, - /* 8 */ {make_small(10), CONTEXT_REDS/5, 3*INPUT_REDUCTIONS}, - /* 9 */ {make_small(10), (6*CONTEXT_REDS)/7, INPUT_REDUCTIONS/4} + /* 0 */ {make_small(0), CONTEXT_REDS}, + /* 1 */ {make_small(0), (3*CONTEXT_REDS)/4}, + /* 2 */ {make_small(0), CONTEXT_REDS/2}, + /* 3 */ {make_small(0), (7*CONTEXT_REDS)/8}, + /* 4 */ {make_small(0), CONTEXT_REDS/3}, + /* 5 */ {make_small(0), (10*CONTEXT_REDS)/11}, + /* 6 */ {make_small(1), CONTEXT_REDS/4}, + /* 7 */ {make_small(1), (5*CONTEXT_REDS)/7}, + /* 8 */ {make_small(10), CONTEXT_REDS/5}, + /* 9 */ {make_small(10), (6*CONTEXT_REDS)/7} }; #define ERTS_MODIFIED_TIMING_LEVELS \ @@ -310,8 +313,9 @@ erl_init(int ncpu, erts_init_sys_common_misc(); erts_init_process(ncpu, proc_tab_sz, legacy_proc_tab); erts_init_scheduling(no_schedulers, - no_schedulers_online - , no_dirty_cpu_schedulers, + no_schedulers_online, + erts_no_poll_threads, + no_dirty_cpu_schedulers, no_dirty_cpu_schedulers_online, no_dirty_io_schedulers ); @@ -558,9 +562,19 @@ void erts_usage(void) erts_fprintf(stderr, "-hmqd val set default message queue data flag for processes,\n"); erts_fprintf(stderr, " valid values are: off_heap | on_heap\n"); + erts_fprintf(stderr, "-IOp number set number of pollsets to be used to poll for I/O,\n"); + erts_fprintf(stderr, " This value has to be equal or smaller than the\n"); + erts_fprintf(stderr, " number of poll threads. If the current platform\n"); + erts_fprintf(stderr, " does not support concurrent update of pollsets\n"); + erts_fprintf(stderr, " this value is ignored.\n"); + erts_fprintf(stderr, "-IOt number set number of threads to be used to poll for I/O\n"); + erts_fprintf(stderr, "-IOPp number set number of pollsets as a percentage of the\n"); + erts_fprintf(stderr, " number of poll threads."); + erts_fprintf(stderr, "-IOPt number set number of threads to be used to poll for I/O\n"); + erts_fprintf(stderr, " as a percentage of the number of schedulers."); + /* erts_fprintf(stderr, "-i module set the boot module (default init)\n"); */ - erts_fprintf(stderr, "-K boolean enable or disable kernel poll\n"); erts_fprintf(stderr, "-n[s|a|d] Control behavior of signals to ports\n"); erts_fprintf(stderr, " Note that this flag is deprecated!\n"); erts_fprintf(stderr, "-M memory allocator switches,\n"); @@ -727,7 +741,6 @@ early_init(int *argc, char **argv) /* char envbuf[21]; /* enough for any 64-bit integer */ size_t envbufsz; - erts_save_emu_args(*argc, argv); erts_sched_compact_load = 1; @@ -839,6 +852,7 @@ early_init(int *argc, char **argv) /* } break; } + case 'S' : if (argv[i][2] == 'P') { int ptot, ponln; @@ -1100,6 +1114,9 @@ early_init(int *argc, char **argv) /* erts_alloc_init(argc, argv, &alloc_opts); /* Handles (and removes) -M flags. */ /* Require allocators */ + + erts_init_check_io(argc, argv); + /* * Thread progress management: * @@ -1107,13 +1124,14 @@ early_init(int *argc, char **argv) /* * ** Scheduler threads (see erl_process.c) * ** Aux thread (see erl_process.c) * ** Sys message dispatcher thread (see erl_trace.c) + * ** IO Poll threads (see erl_check_io.c) * * * Unmanaged threads that need to register: * ** Async threads (see erl_async.c) * ** Dirty scheduler threads */ erts_thr_progress_init(no_schedulers, - no_schedulers+2, + no_schedulers+2+erts_no_poll_threads, erts_async_max_threads + erts_no_dirty_cpu_schedulers + erts_no_dirty_io_schedulers @@ -1561,16 +1579,6 @@ erl_start(int argc, char **argv) have_break_handler = 0; break; - case 'K': - /* If kernel poll support is present, - erl_sys_args() will remove the K parameter - and value */ - get_arg(argv[i]+2, argv[i+1], &i); - erts_fprintf(stderr, - "kernel-poll not supported; \"K\" parameter ignored\n", - arg); - break; - case 'n': arg = get_arg(argv[i]+2, argv[i+1], &i); switch (arg[0]) { diff --git a/erts/emulator/beam/erl_lock_count.c b/erts/emulator/beam/erl_lock_count.c index 11890e44d1..1ae6076b12 100644 --- a/erts/emulator/beam/erl_lock_count.c +++ b/erts/emulator/beam/erl_lock_count.c @@ -554,16 +554,6 @@ erts_lock_flags_t erts_lcnt_get_category_mask() { return lcnt_category_mask__; } -#ifdef ERTS_ENABLE_KERNEL_POLL -/* erl_poll/erl_check_io only exports one of these variants at a time, and we - * may need to use either one depending on emulator startup flags. */ -void erts_lcnt_update_pollset_locks_nkp(int); -void erts_lcnt_update_pollset_locks_kp(int); - -void erts_lcnt_update_cio_locks_nkp(int); -void erts_lcnt_update_cio_locks_kp(int); -#endif - void erts_lcnt_set_category_mask(erts_lock_flags_t mask) { erts_lock_flags_t changed_categories; @@ -590,16 +580,6 @@ void erts_lcnt_set_category_mask(erts_lock_flags_t mask) { } if(changed_categories & ERTS_LOCK_FLAGS_CATEGORY_IO) { -#ifdef ERTS_ENABLE_KERNEL_POLL - if(erts_use_kernel_poll) { - erts_lcnt_update_pollset_locks_kp(mask & ERTS_LOCK_FLAGS_CATEGORY_IO); - } else { - erts_lcnt_update_pollset_locks_nkp(mask & ERTS_LOCK_FLAGS_CATEGORY_IO); - } -#else - erts_lcnt_update_pollset_locks(mask & ERTS_LOCK_FLAGS_CATEGORY_IO); -#endif - erts_lcnt_update_cio_locks(mask & ERTS_LOCK_FLAGS_CATEGORY_IO); erts_lcnt_update_driver_locks(mask & ERTS_LOCK_FLAGS_CATEGORY_IO); erts_lcnt_update_port_locks(mask & ERTS_LOCK_FLAGS_CATEGORY_IO); diff --git a/erts/emulator/beam/erl_port_task.c b/erts/emulator/beam/erl_port_task.c index 86df0f4cb8..a588477320 100644 --- a/erts/emulator/beam/erl_port_task.c +++ b/erts/emulator/beam/erl_port_task.c @@ -582,12 +582,10 @@ static ERTS_INLINE void reset_executed_io_task_handle(ErtsPortTask *ptp) { if (ptp->u.alive.handle) { - ErtsIoTask *itp = ErtsContainerStruct(ptp->u.alive.handle, ErtsIoTask, task); - ASSERT(ptp == handle2task(ptp->u.alive.handle)); - erts_io_notify_port_task_executed(ptp->u.alive.handle); - reset_port_task_handle(ptp->u.alive.handle); - erts_check_io_interrupt(itp->pollset, 1); + /* The port task handle is reset inside task_executed */ + erts_io_notify_port_task_executed(ptp->type, ptp->u.alive.handle, + reset_port_task_handle); } } diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c index 66dcd85af6..ffb67aefaf 100644 --- a/erts/emulator/beam/erl_process.c +++ b/erts/emulator/beam/erl_process.c @@ -24,6 +24,8 @@ # include "config.h" #endif +#define ERTS_WANT_BREAK_HANDLING + #include /* offsetof() */ #include "sys.h" #include "erl_vm.h" @@ -50,6 +52,7 @@ #include "erl_time.h" #include "erl_nfunc_sched.h" #include "erl_check_io.h" +#include "erl_poll.h" #define ERTS_CHECK_TIME_REDS CONTEXT_REDS #define ERTS_DELAYED_WAKEUP_INFINITY (~(Uint64) 0) @@ -208,6 +211,7 @@ static struct ErtsSchedBusyWait_ { int erts_disable_proc_not_running_opt; static ErtsAuxWorkData *aux_thread_aux_work_data; +static ErtsAuxWorkData *poll_thread_aux_work_data; #define ERTS_SCHDLR_SSPND_CHNG_NMSB (((erts_aint32_t) 1) << 0) #define ERTS_SCHDLR_SSPND_CHNG_MSB (((erts_aint32_t) 1) << 1) @@ -432,6 +436,7 @@ typedef union { static ErtsAlignedSchedulerSleepInfo *aligned_sched_sleep_info; static ErtsAlignedSchedulerSleepInfo *aligned_dirty_cpu_sched_sleep_info; static ErtsAlignedSchedulerSleepInfo *aligned_dirty_io_sched_sleep_info; +static ErtsAlignedSchedulerSleepInfo *aligned_poll_thread_sleep_info; static Uint last_reductions; static Uint last_exact_reductions; @@ -499,9 +504,13 @@ ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(proclist, 200, ERTS_ALC_T_PROC_LIST) +#define ERTS_POLL_THREAD_SLEEP_INFO_IX(IX) \ + (ASSERT(0 <= ((int) (IX)) \ + && ((int) (IX)) < ((int) erts_no_poll_threads)), \ + &aligned_poll_thread_sleep_info[(IX)].ssi) #define ERTS_SCHED_SLEEP_INFO_IX(IX) \ - (ASSERT(-1 <= ((int) (IX)) \ - && ((int) (IX)) < ((int) erts_no_schedulers)), \ + (ASSERT(((int)-1) <= ((int) (IX)) \ + && ((int) (IX)) < ((int) erts_no_schedulers)), \ &aligned_sched_sleep_info[(IX)].ssi) #define ERTS_DIRTY_CPU_SCHED_SLEEP_INFO_IX(IX) \ (ASSERT(0 <= ((int) (IX)) \ @@ -1560,14 +1569,14 @@ erts_sched_finish_poke(ErtsSchedulerSleepInfo *ssi, { switch (flags & ERTS_SSI_FLGS_SLEEP_TYPE) { case ERTS_SSI_FLG_POLL_SLEEPING: - erts_check_io_interrupt(ssi->esdp->pollset, 1); + erts_check_io_interrupt(ssi->psi, 1); break; case ERTS_SSI_FLG_POLL_SLEEPING|ERTS_SSI_FLG_TSE_SLEEPING: /* * Thread progress blocking while poll sleeping; need * to signal on both... */ - erts_check_io_interrupt(ssi->esdp->pollset, 1); + erts_check_io_interrupt(ssi->psi, 1); /* fall through */ case ERTS_SSI_FLG_TSE_SLEEPING: erts_tse_set(ssi->event); @@ -2827,36 +2836,6 @@ erts_set_aux_work_timeout(int ix, erts_aint32_t type, int enable) return old; } - - -static ERTS_INLINE void -sched_waiting_sys(Uint no, ErtsRunQueue *rq) -{ - ERTS_LC_ASSERT(erts_lc_runq_is_locked(rq)); - ASSERT(rq->waiting >= 0); - (void) ERTS_RUNQ_FLGS_SET(rq, (ERTS_RUNQ_FLG_OUT_OF_WORK - | ERTS_RUNQ_FLG_HALFTIME_OUT_OF_WORK)); - rq->waiting++; - rq->waiting *= -1; - rq->woken = 0; - if (erts_system_profile_flags.scheduler) - profile_scheduler(make_small(no), am_inactive); -} - -static ERTS_INLINE void -sched_active_sys(Uint no, ErtsRunQueue *rq) -{ - ERTS_LC_ASSERT(erts_lc_runq_is_locked(rq)); - - ASSERT(!ERTS_RUNQ_IX_IS_DIRTY(rq->ix)); - - ASSERT(rq->waiting < 0); - rq->waiting *= -1; - rq->waiting--; - if (erts_system_profile_flags.scheduler) - profile_scheduler(make_small(no), am_active); -} - Uint erts_active_schedulers(void) { @@ -2867,13 +2846,6 @@ erts_active_schedulers(void) return as; } -static ERTS_INLINE int -prepare_for_sys_schedule(int non_blocking) -{ - return 1; -} - - static ERTS_INLINE void sched_waiting(Uint no, ErtsRunQueue *rq) { @@ -3045,7 +3017,7 @@ sched_set_sleeptype(ErtsSchedulerSleepInfo *ssi, erts_aint32_t sleep_type) erts_tse_reset(ssi->event); else { ASSERT(sleep_type == ERTS_SSI_FLG_POLL_SLEEPING); - erts_check_io_interrupt(ssi->esdp->pollset, 0); + erts_check_io_interrupt(ssi->psi, 0); } while (1) { @@ -3113,6 +3085,12 @@ thr_prgr_fin_wait(void *vssi) static void init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp, char *dawwp); +void +erts_aux_thread_poke() +{ + erts_sched_poke(ERTS_SCHED_SLEEP_INFO_IX(-1)); +} + static void * aux_thread(void *unused) { @@ -3121,6 +3099,7 @@ aux_thread(void *unused) erts_aint32_t aux_work; ErtsThrPrgrCallbacks callbacks; int thr_prgr_active = 1; + ERTS_MSACC_DECLARE_CACHE(); #ifdef ERTS_ENABLE_LOCK_CHECK { @@ -3144,9 +3123,14 @@ aux_thread(void *unused) init_aux_work_data(awdp, NULL, NULL); awdp->ssi = ssi; +#if ERTS_POLL_USE_FALLBACK + ssi->psi = erts_create_pollset_thread(-1); +#endif sched_prep_spin_wait(ssi); + ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_OTHER); + while (1) { erts_aint32_t flgs; @@ -3155,30 +3139,128 @@ aux_thread(void *unused) if (!thr_prgr_active) erts_thr_progress_active(NULL, thr_prgr_active = 1); aux_work = handle_aux_work(awdp, aux_work, 1); + ERTS_MSACC_UPDATE_CACHE(); if (aux_work && erts_thr_progress_update(NULL)) erts_thr_progress_leader_update(NULL); } if (!aux_work) { + +#ifdef ERTS_BREAK_REQUESTED + if (ERTS_BREAK_REQUESTED) + erts_do_break_handling(); +#endif + if (thr_prgr_active) erts_thr_progress_active(NULL, thr_prgr_active = 0); - erts_thr_progress_prepare_wait(NULL); + +#if ERTS_POLL_USE_FALLBACK flgs = sched_spin_wait(ssi, 0); if (flgs & ERTS_SSI_FLG_SLEEPING) { ASSERT(flgs & ERTS_SSI_FLG_WAITING); + flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_POLL_SLEEPING); + if (flgs & ERTS_SSI_FLG_SLEEPING) { + ASSERT(flgs & ERTS_SSI_FLG_POLL_SLEEPING); + ASSERT(flgs & ERTS_SSI_FLG_WAITING); + erts_check_io(ssi->psi); + } + } +#else + erts_thr_progress_prepare_wait(NULL); + + flgs = sched_spin_wait(ssi, 0); + + if (flgs & ERTS_SSI_FLG_SLEEPING) { flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING); if (flgs & ERTS_SSI_FLG_SLEEPING) { - int res; + int res; ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING); ASSERT(flgs & ERTS_SSI_FLG_WAITING); - do { - res = erts_tse_wait(ssi->event); - } while (res == EINTR); + ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_SLEEP); + do { + res = erts_tse_wait(ssi->event); + } while (res == EINTR); + ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_OTHER); + } + } + erts_thr_progress_finalize_wait(NULL); +#endif + } + + flgs = sched_prep_spin_wait(ssi); + } + return NULL; +} + +static void * +poll_thread(void *arg) +{ + int id = (int)(UWord)arg; + ErtsAuxWorkData *awdp = poll_thread_aux_work_data+id; + ErtsSchedulerSleepInfo *ssi = ERTS_POLL_THREAD_SLEEP_INFO_IX(id); + erts_aint32_t aux_work; + ErtsThrPrgrCallbacks callbacks; + int thr_prgr_active = 1; + struct erts_poll_thread *psi = erts_create_pollset_thread(id); + ERTS_MSACC_DECLARE_CACHE(); + +#ifdef ERTS_ENABLE_LOCK_CHECK + { + char buf[] = "poll_thread"; + erts_lc_set_thread_name(buf); + } +#endif + + erts_port_task_pre_alloc_init_thread(); + ssi->event = erts_tse_fetch(); + + erts_msacc_init_thread("poll", id, 0); + + callbacks.arg = (void *) ssi; + callbacks.wakeup = thr_prgr_wakeup; + callbacks.prepare_wait = thr_prgr_prep_wait; + callbacks.wait = thr_prgr_wait; + callbacks.finalize_wait = thr_prgr_fin_wait; + + erts_thr_progress_register_managed_thread(NULL, &callbacks, 0); + init_aux_work_data(awdp, NULL, NULL); + awdp->ssi = ssi; + ssi->psi = psi; + + sched_prep_spin_wait(ssi); + + ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_OTHER); + + while (1) { + erts_aint32_t flgs; + + aux_work = erts_atomic32_read_acqb(&ssi->aux_work); + if (aux_work) { + if (!thr_prgr_active) + erts_thr_progress_active(NULL, thr_prgr_active = 1); + aux_work = handle_aux_work(awdp, aux_work, 1); + ERTS_MSACC_UPDATE_CACHE(); + if (aux_work && erts_thr_progress_update(NULL)) + erts_thr_progress_leader_update(NULL); + } + + if (!aux_work) { + if (thr_prgr_active) + erts_thr_progress_active(NULL, thr_prgr_active = 0); + + flgs = sched_spin_wait(ssi, 0); + + if (flgs & ERTS_SSI_FLG_SLEEPING) { + ASSERT(flgs & ERTS_SSI_FLG_WAITING); + flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_POLL_SLEEPING); + if (flgs & ERTS_SSI_FLG_SLEEPING) { + ASSERT(flgs & ERTS_SSI_FLG_POLL_SLEEPING); + ASSERT(flgs & ERTS_SSI_FLG_WAITING); + erts_check_io(psi); } } - erts_thr_progress_finalize_wait(NULL); } flgs = sched_prep_spin_wait(ssi); @@ -3219,273 +3301,136 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq) dirty_active(esdp, -1); } - /* - * If all schedulers are waiting, one of them *should* - * be waiting in erl_sys_schedule() - */ - - if (ERTS_SCHEDULER_IS_DIRTY(esdp) || !prepare_for_sys_schedule(0)) { + sched_waiting(esdp->no, rq); - sched_waiting(esdp->no, rq); - - erts_runq_unlock(rq); - - spincount = sched_busy_wait.tse; - - if (ERTS_SCHEDULER_IS_DIRTY(esdp)) - dirty_sched_wall_time_change(esdp, working = 0); - else if (thr_prgr_active != working) - sched_wall_time_change(esdp, working = thr_prgr_active); - - while (1) { - ErtsMonotonicTime current_time = 0; + erts_runq_unlock(rq); - aux_work = erts_atomic32_read_acqb(&ssi->aux_work); - if (aux_work && !ERTS_SCHEDULER_IS_DIRTY(esdp)) { - if (!thr_prgr_active) { - erts_thr_progress_active(esdp, thr_prgr_active = 1); - sched_wall_time_change(esdp, 1); - } - aux_work = handle_aux_work(&esdp->aux_work_data, aux_work, 1); - ERTS_MSACC_UPDATE_CACHE(); - if (aux_work && erts_thr_progress_update(esdp)) - erts_thr_progress_leader_update(esdp); - } + spincount = sched_busy_wait.tse; - if (aux_work) { - if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { - flgs = erts_atomic32_read_acqb(&ssi->flags); - current_time = erts_get_monotonic_time(esdp); - if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) { - if (!thr_prgr_active) { - erts_thr_progress_active(esdp, thr_prgr_active = 1); - sched_wall_time_change(esdp, 1); - } - erts_bump_timers(esdp->timer_wheel, current_time); - } - } - } - else { - ErtsMonotonicTime timeout_time; - int do_timeout = 0; - if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { - timeout_time = erts_check_next_timeout_time(esdp); - current_time = erts_get_monotonic_time(esdp); - do_timeout = (current_time >= timeout_time); - } else { - current_time = 0; - timeout_time = ERTS_MONOTONIC_TIME_MAX; - } - if (do_timeout) { - if (!thr_prgr_active) { - erts_thr_progress_active(esdp, thr_prgr_active = 1); - sched_wall_time_change(esdp, 1); - } - } - else { - if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { - if (thr_prgr_active) { - erts_thr_progress_active(esdp, thr_prgr_active = 0); - sched_wall_time_change(esdp, 0); - } - erts_thr_progress_prepare_wait(esdp); - } + if (ERTS_SCHEDULER_IS_DIRTY(esdp)) + dirty_sched_wall_time_change(esdp, working = 0); + else if (thr_prgr_active != working) + sched_wall_time_change(esdp, working = thr_prgr_active); - flgs = sched_spin_wait(ssi, spincount); - if (flgs & ERTS_SSI_FLG_SLEEPING) { - ASSERT(flgs & ERTS_SSI_FLG_WAITING); - flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING); - if (flgs & ERTS_SSI_FLG_SLEEPING) { - int res; - ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING); - ASSERT(flgs & ERTS_SSI_FLG_WAITING); - current_time = ERTS_SCHEDULER_IS_DIRTY(esdp) ? 0 : - erts_get_monotonic_time(esdp); - do { - Sint64 timeout; - if (current_time >= timeout_time) - break; - if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { - timeout = ERTS_MONOTONIC_TO_NSEC(timeout_time - - current_time - - 1) + 1; - } else - timeout = -1; - ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_SLEEP); - res = erts_tse_twait(ssi->event, timeout); - ERTS_MSACC_POP_STATE_M(); - current_time = ERTS_SCHEDULER_IS_DIRTY(esdp) ? 0 : - erts_get_monotonic_time(esdp); - } while (res == EINTR); - } - } - if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) - erts_thr_progress_finalize_wait(esdp); - } - if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && current_time >= timeout_time) - erts_bump_timers(esdp->timer_wheel, current_time); - } + while (1) { + ErtsMonotonicTime current_time = 0; - if (!(flgs & ERTS_SSI_FLG_WAITING)) { - ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING)); - break; - } + aux_work = erts_atomic32_read_acqb(&ssi->aux_work); + if (aux_work && !ERTS_SCHEDULER_IS_DIRTY(esdp)) { + if (!thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 1); + sched_wall_time_change(esdp, 1); + } + aux_work = handle_aux_work(&esdp->aux_work_data, aux_work, 1); + ERTS_MSACC_UPDATE_CACHE(); + if (aux_work && erts_thr_progress_update(esdp)) + erts_thr_progress_leader_update(esdp); + } - flgs = sched_prep_cont_spin_wait(ssi); - spincount = sched_busy_wait.aux_work; + if (aux_work) { + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { + flgs = erts_atomic32_read_acqb(&ssi->flags); + current_time = erts_get_monotonic_time(esdp); + if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) { + if (!thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 1); + sched_wall_time_change(esdp, 1); + } + erts_bump_timers(esdp->timer_wheel, current_time); + } + } + } + else { + ErtsMonotonicTime timeout_time; + int do_timeout = 0; + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { + timeout_time = erts_check_next_timeout_time(esdp); + current_time = erts_get_monotonic_time(esdp); + do_timeout = (current_time >= timeout_time); + } else { + current_time = 0; + timeout_time = ERTS_MONOTONIC_TIME_MAX; + } + if (do_timeout) { + if (!thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 1); + sched_wall_time_change(esdp, 1); + } + } + else { + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { + if (thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 0); + sched_wall_time_change(esdp, 0); + } + erts_thr_progress_prepare_wait(esdp); + } - if (!(flgs & ERTS_SSI_FLG_WAITING)) { - ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING)); - break; - } + flgs = sched_spin_wait(ssi, spincount); + if (flgs & ERTS_SSI_FLG_SLEEPING) { + ASSERT(flgs & ERTS_SSI_FLG_WAITING); + flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING); + if (flgs & ERTS_SSI_FLG_SLEEPING) { + int res; + ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING); + ASSERT(flgs & ERTS_SSI_FLG_WAITING); + current_time = ERTS_SCHEDULER_IS_DIRTY(esdp) ? 0 : + erts_get_monotonic_time(esdp); + do { + Sint64 timeout; + if (current_time >= timeout_time) + break; + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { + timeout = ERTS_MONOTONIC_TO_NSEC(timeout_time + - current_time + - 1) + 1; + } else + timeout = -1; + ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_SLEEP); + res = erts_tse_twait(ssi->event, timeout); + ERTS_MSACC_POP_STATE_M(); + current_time = ERTS_SCHEDULER_IS_DIRTY(esdp) ? 0 : + erts_get_monotonic_time(esdp); + } while (res == EINTR); + } + } + if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) + erts_thr_progress_finalize_wait(esdp); + } + if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && current_time >= timeout_time) + erts_bump_timers(esdp->timer_wheel, current_time); + } - } + if (!(flgs & ERTS_SSI_FLG_WAITING)) { + ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING)); + break; + } - if (flgs & ~(ERTS_SSI_FLG_SUSPENDED|ERTS_SSI_FLG_MSB_EXEC)) - erts_atomic32_read_band_nob(&ssi->flags, - (ERTS_SSI_FLG_SUSPENDED - | ERTS_SSI_FLG_MSB_EXEC)); + flgs = sched_prep_cont_spin_wait(ssi); + spincount = sched_busy_wait.aux_work; - if (ERTS_SCHEDULER_IS_DIRTY(esdp)) - dirty_sched_wall_time_change(esdp, working = 1); - else if (!thr_prgr_active) { - erts_thr_progress_active(esdp, thr_prgr_active = 1); - sched_wall_time_change(esdp, 1); + if (!(flgs & ERTS_SSI_FLG_WAITING)) { + ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING)); + break; } - erts_runq_lock(rq); - sched_active(esdp->no, rq); - } - else - { - - esdp->function_calls = 0; - *fcalls = 0; - - ASSERT(!ERTS_SCHEDULER_IS_DIRTY(esdp)); - - sched_waiting_sys(esdp->no, rq); - - erts_runq_unlock(rq); - - ASSERT(working); - sched_wall_time_change(esdp, working = 0); - spincount = sched_busy_wait.sys_schedule; - if (spincount == 0) - goto sys_aux_work; + if (flgs & ~(ERTS_SSI_FLG_SUSPENDED|ERTS_SSI_FLG_MSB_EXEC)) + erts_atomic32_read_band_nob(&ssi->flags, + (ERTS_SSI_FLG_SUSPENDED + | ERTS_SSI_FLG_MSB_EXEC)); - while (spincount-- > 0) { - ErtsMonotonicTime current_time; - - sys_poll_aux_work: - - if (working) - sched_wall_time_change(esdp, working = 0); - - ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_CHECK_IO); - - LTTNG2(scheduler_poll, esdp->no, 1); - erl_sys_schedule(1); /* Might give us something to do */ - - ERTS_MSACC_POP_STATE_M(); - - if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { - current_time = erts_get_monotonic_time(esdp); - if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) - erts_bump_timers(esdp->timer_wheel, current_time); - } - - sys_aux_work: - - aux_work = erts_atomic32_read_acqb(&ssi->aux_work); - if (aux_work && !ERTS_SCHEDULER_IS_DIRTY(esdp)) { - if (!working) - sched_wall_time_change(esdp, working = 1); - if (!thr_prgr_active) - erts_thr_progress_active(esdp, thr_prgr_active = 1); - aux_work = handle_aux_work(&esdp->aux_work_data, aux_work, 1); - ERTS_MSACC_UPDATE_CACHE(); - if (aux_work && erts_thr_progress_update(esdp)) - erts_thr_progress_leader_update(esdp); - } - - flgs = erts_atomic32_read_acqb(&ssi->flags); - if (!(flgs & ERTS_SSI_FLG_WAITING)) { - ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING)); - goto sys_woken; - } - } - - erts_runq_lock(rq); - - if (aux_work) { - erts_runq_unlock(rq); - goto sys_poll_aux_work; - } - flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_POLL_SLEEPING); - if (!(flgs & ERTS_SSI_FLG_SLEEPING)) { - if (!(flgs & ERTS_SSI_FLG_WAITING)) { - ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING)); - goto sys_locked_woken; - } - erts_runq_unlock(rq); - flgs = sched_prep_cont_spin_wait(ssi); - if (!(flgs & ERTS_SSI_FLG_WAITING)) { - ASSERT(!(flgs & ERTS_SSI_FLG_SLEEPING)); - goto sys_woken; - } - goto sys_poll_aux_work; - } - - ASSERT(flgs & ERTS_SSI_FLG_POLL_SLEEPING); - ASSERT(flgs & ERTS_SSI_FLG_WAITING); - - erts_runq_unlock(rq); - - if (working) - sched_wall_time_change(esdp, working = 0); - - if (thr_prgr_active) - erts_thr_progress_active(esdp, thr_prgr_active = 0); - - ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_CHECK_IO); - LTTNG2(scheduler_poll, esdp->no, 0); - - erl_sys_schedule(0); - - ERTS_MSACC_POP_STATE_M(); - - if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) { - ErtsMonotonicTime current_time = erts_get_monotonic_time(esdp); - if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) - erts_bump_timers(esdp->timer_wheel, current_time); - } - - flgs = sched_prep_cont_spin_wait(ssi); - if (flgs & ERTS_SSI_FLG_WAITING) - goto sys_aux_work; - - sys_woken: - if (!thr_prgr_active) - erts_thr_progress_active(esdp, thr_prgr_active = 1); - erts_runq_lock(rq); - sys_locked_woken: - if (!thr_prgr_active) { - erts_runq_unlock(rq); - erts_thr_progress_active(esdp, thr_prgr_active = 1); - erts_runq_lock(rq); - } - if (flgs & ~(ERTS_SSI_FLG_SUSPENDED|ERTS_SSI_FLG_MSB_EXEC)) - erts_atomic32_read_band_nob(&ssi->flags, - (ERTS_SSI_FLG_SUSPENDED - | ERTS_SSI_FLG_MSB_EXEC)); - if (!working) - sched_wall_time_change(esdp, working = 1); - sched_active_sys(esdp->no, rq); + if (ERTS_SCHEDULER_IS_DIRTY(esdp)) + dirty_sched_wall_time_change(esdp, working = 1); + else if (!thr_prgr_active) { + erts_thr_progress_active(esdp, thr_prgr_active = 1); + sched_wall_time_change(esdp, 1); } + erts_runq_lock(rq); + sched_active(esdp->no, rq); + if (ERTS_SCHEDULER_IS_DIRTY(esdp)) dirty_active(esdp, 1); @@ -5755,7 +5700,6 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num, esdp->run_queue = runq; if (ERTS_RUNQ_IX_IS_DIRTY(runq->ix)) { esdp->no = 0; - esdp->pollset = NULL; if (runq == ERTS_DIRTY_CPU_RUNQ) esdp->type = ERTS_SCHED_DIRTY_CPU; else { @@ -5775,7 +5719,6 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num, esdp->type = ERTS_SCHED_NORMAL; esdp->no = (Uint) num; esdp->dirty_no = 0; - esdp->pollset = erts_get_pollset(num); runq->scheduler = esdp; } esdp->dirty_shadow_process = shadow_proc; @@ -5788,8 +5731,6 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num, shadow_proc->static_flags = ERTS_STC_FLG_SHADOW_PROC; } - esdp->function_calls = 0; - ssi->esdp = esdp; esdp->ssi = ssi; esdp->current_process = NULL; @@ -5821,8 +5762,8 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num, } void -erts_init_scheduling(int no_schedulers, int no_schedulers_online - , int no_dirty_cpu_schedulers, int no_dirty_cpu_schedulers_online, +erts_init_scheduling(int no_schedulers, int no_schedulers_online, int no_poll_threads, + int no_dirty_cpu_schedulers, int no_dirty_cpu_schedulers_online, int no_dirty_io_schedulers ) { @@ -5849,6 +5790,7 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online ASSERT(no_dirty_cpu_schedulers >= 1); ASSERT(no_dirty_cpu_schedulers_online <= no_schedulers_online); ASSERT(no_dirty_cpu_schedulers_online >= 1); + ASSERT(erts_no_poll_threads == no_poll_threads); /* Create and initialize run queues */ @@ -5946,7 +5888,7 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online erts_no_total_schedulers += no_dirty_io_schedulers; /* Create and initialize scheduler sleep info */ - no_ssi = n+1; + no_ssi = n + 1 /* aux thread */; aligned_sched_sleep_info = erts_alloc_permanent_cache_aligned( ERTS_ALC_T_SCHDLR_SLP_INFO, @@ -5957,12 +5899,13 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online ssi->next = NULL; ssi->prev = NULL; #endif + ssi->esdp = NULL; erts_atomic32_init_nob(&ssi->flags, 0); ssi->event = NULL; /* initialized in sched_thread_func */ erts_atomic32_init_nob(&ssi->aux_work, 0); } - aligned_sched_sleep_info++; + aligned_sched_sleep_info += 1 /* aux thread */; aligned_dirty_cpu_sched_sleep_info = erts_alloc_permanent_cache_aligned( @@ -5985,6 +5928,18 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online erts_atomic32_init_nob(&ssi->aux_work, 0); } + aligned_poll_thread_sleep_info = + erts_alloc_permanent_cache_aligned( + ERTS_ALC_T_SCHDLR_SLP_INFO, + no_poll_threads*sizeof(ErtsAlignedSchedulerSleepInfo)); + for (ix = 0; ix < no_poll_threads; ix++) { + ErtsSchedulerSleepInfo *ssi = &aligned_poll_thread_sleep_info[ix].ssi; + ssi->esdp = NULL; + erts_atomic32_init_nob(&ssi->flags, 0); + ssi->event = NULL; /* initialized in poll_thread */ + erts_atomic32_init_nob(&ssi->aux_work, 0); + } + /* Create and initialize scheduler specific data */ daww_sz = ERTS_ALC_CACHE_LINE_ALIGN_SIZE((sizeof(ErtsDelayedAuxWorkWakeupJob) @@ -6041,11 +5996,14 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online erts_atomic32_init_nob(&debug_wait_completed_count, 0); /* debug only */ debug_wait_completed_flags = 0; - aux_thread_aux_work_data = erts_alloc_permanent_cache_aligned(ERTS_ALC_T_SCHDLR_DATA, sizeof(ErtsAuxWorkData)); + poll_thread_aux_work_data = + erts_alloc_permanent_cache_aligned(ERTS_ALC_T_SCHDLR_DATA, + no_poll_threads * sizeof(ErtsAuxWorkData)); + init_no_runqs(no_schedulers_online, no_schedulers_online); balance_info.last_active_runqs = no_schedulers; erts_mtx_init(&balance_info.update_mtx, "migration_info_update", NIL, @@ -7034,12 +6992,8 @@ sched_set_suspended_sleeptype(ErtsSchedulerSleepInfo *ssi, | ERTS_SSI_FLG_WAITING | ERTS_SSI_FLG_SUSPENDED); - if (sleep_type == ERTS_SSI_FLG_TSE_SLEEPING) - erts_tse_reset(ssi->event); - else { - ASSERT(sleep_type == ERTS_SSI_FLG_POLL_SLEEPING); - erts_check_io_interrupt(ssi->esdp->pollset, 0); - } + ASSERT(sleep_type == ERTS_SSI_FLG_TSE_SLEEPING); + erts_tse_reset(ssi->event); while (1) { oflgs = erts_atomic32_cmpxchg_acqb(&ssi->flags, nflgs, xflgs); @@ -7291,17 +7245,6 @@ msb_scheduler_type_switch(ErtsSchedType sched_type, if (exec_type != ERTS_SCHED_NORMAL) schdlr_sspnd.last_msb_dirty_type = exec_type; else { - erts_aint32_t calls; - /* - * Going back to normal scheduler after - * dirty execution; make sure it will check - * for I/O... - */ - if (ERTS_USE_MODIFIED_TIMING()) - calls = ERTS_MODIFIED_TIMING_INPUT_REDS + 1; - else - calls = INPUT_REDUCTIONS + 1; - esdp->function_calls = calls; if ((nrml_prio == ERTS_MSB_NONE_PRIO_BIT) & ((dcpu_prio != ERTS_MSB_NONE_PRIO_BIT) @@ -7376,7 +7319,7 @@ msb_scheduler_type_switch(ErtsSchedType sched_type, } static ERTS_INLINE void -suspend_dirty_scheduler_sleep(ErtsSchedulerData *esdp) +suspend_normal_scheduler_sleep(ErtsSchedulerData *esdp) { ErtsSchedulerSleepInfo *ssi = esdp->ssi; erts_aint32_t flgs = sched_spin_suspended(ssi, @@ -7399,36 +7342,9 @@ suspend_dirty_scheduler_sleep(ErtsSchedulerData *esdp) } static ERTS_INLINE void -suspend_normal_scheduler_sleep(ErtsSchedulerData *esdp) +suspend_dirty_scheduler_sleep(ErtsSchedulerData *esdp) { - ErtsSchedulerSleepInfo *ssi = esdp->ssi; - erts_aint32_t flgs; - int spincount; - - spincount = sched_busy_wait.sys_schedule; - - while (spincount-- > 0) { - - erl_sys_schedule(1); - - flgs = erts_smp_atomic32_read_acqb(&ssi->flags); - if (flgs != (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_WAITING - | ERTS_SSI_FLG_SUSPENDED)) { - return; - } - } - - flgs = sched_set_suspended_sleeptype(ssi, ERTS_SSI_FLG_POLL_SLEEPING); - if (flgs != (ERTS_SSI_FLG_SLEEPING - | ERTS_SSI_FLG_POLL_SLEEPING - | ERTS_SSI_FLG_WAITING - | ERTS_SSI_FLG_SUSPENDED)) { - return; - } - - /* Sleep on pollset... */ - erl_sys_schedule(0); + suspend_normal_scheduler_sleep(esdp); } static void @@ -7662,8 +7578,10 @@ suspend_scheduler(ErtsSchedulerData *esdp) erts_thr_progress_active(esdp, thr_prgr_active = 0); sched_wall_time_change(esdp, 0); } + erts_thr_progress_prepare_wait(NULL); suspend_normal_scheduler_sleep(esdp); - current_time = erts_get_monotonic_time(esdp);; + erts_thr_progress_finalize_wait(NULL); + current_time = erts_get_monotonic_time(esdp); } if (current_time >= timeout_time) { @@ -8493,17 +8411,17 @@ sched_dirty_io_thread_func(void *vesdp) return NULL; } -static ethr_tid aux_tid; - void erts_start_schedulers(void) { + ethr_tid tid; int res = 0; Uint actual; Uint wanted = erts_no_schedulers; Uint wanted_no_schedulers = erts_no_schedulers; char name[16]; ethr_thr_opts opts = ETHR_THR_OPTS_DEFAULT_INITER; + int ix; opts.detached = 1; @@ -8548,7 +8466,6 @@ erts_start_schedulers(void) erts_no_schedulers = actual; { - int ix; for (ix = 0; ix < erts_no_dirty_cpu_schedulers; ix++) { ErtsSchedulerData *esdp = ERTS_DIRTY_CPU_SCHEDULER_IX(ix); erts_snprintf(opts.name, 16, "%d_dirty_cpu_scheduler", ix + 1); @@ -8571,10 +8488,18 @@ erts_start_schedulers(void) erts_snprintf(opts.name, 16, "aux"); - res = ethr_thr_create(&aux_tid, aux_thread, NULL, &opts); + res = ethr_thr_create(&tid, aux_thread, NULL, &opts); if (res != 0) erts_exit(ERTS_ERROR_EXIT, "Failed to create aux thread\n"); + for (ix = 0; ix < erts_no_poll_threads; ix++) { + erts_snprintf(opts.name, 16, "%d_poller", ix); + + res = ethr_thr_create(&tid, poll_thread, (void*)(UWord)ix, &opts); + if (res != 0) + erts_exit(ERTS_ERROR_EXIT, "Failed to create poll thread\n"); + } + if (actual < 1) erts_exit(ERTS_ERROR_EXIT, "Failed to create any scheduler-threads: %s (%d)\n", @@ -9641,7 +9566,6 @@ Process *erts_schedule(ErtsSchedulerData *esdp, Process *p, int calls) ErtsRunQueue *rq; int context_reds; int fcalls; - int input_reductions; int actual_reds; int reds; Uint32 flags; @@ -9661,11 +9585,9 @@ Process *erts_schedule(ErtsSchedulerData *esdp, Process *p, int calls) if (ERTS_USE_MODIFIED_TIMING()) { context_reds = ERTS_MODIFIED_TIMING_CONTEXT_REDS; - input_reductions = ERTS_MODIFIED_TIMING_INPUT_REDS; } else { context_reds = CONTEXT_REDS; - input_reductions = INPUT_REDUCTIONS; } ERTS_LC_ASSERT(ERTS_SCHEDULER_IS_DIRTY(erts_get_scheduler_data()) @@ -9685,7 +9607,6 @@ Process *erts_schedule(ErtsSchedulerData *esdp, Process *p, int calls) } rq = erts_get_runq_current(esdp); ASSERT(esdp); - fcalls = esdp->function_calls; actual_reds = reds = 0; erts_runq_lock(rq); } else { @@ -9711,8 +9632,6 @@ Process *erts_schedule(ErtsSchedulerData *esdp, Process *p, int calls) reds = ERTS_PROC_MIN_CONTEXT_SWITCH_REDS_COST; esdp->virtual_reds = 0; - esdp->function_calls += reds; - fcalls = esdp->function_calls; ASSERT(esdp && esdp == erts_get_scheduler_data()); rq = erts_get_runq_current(esdp); @@ -9802,16 +9721,6 @@ Process *erts_schedule(ErtsSchedulerData *esdp, Process *p, int calls) ERTS_CHK_NO_PROC_LOCKS; - if (is_normal_sched) { - if (esdp->check_time_reds >= ERTS_CHECK_TIME_REDS) - (void) erts_get_monotonic_time(esdp); - - if (esdp->last_monotonic_time >= erts_next_timeout_time(esdp->next_tmo_ref)) { - erts_runq_unlock(rq); - erts_bump_timers(esdp->timer_wheel, esdp->last_monotonic_time); - erts_runq_lock(rq); - } - } } ERTS_LC_ASSERT(!is_normal_sched || !erts_thr_progress_is_blocking()); @@ -9822,6 +9731,16 @@ Process *erts_schedule(ErtsSchedulerData *esdp, Process *p, int calls) ErtsMigrationPath *mp; if (is_normal_sched) { + + if (esdp->check_time_reds >= ERTS_CHECK_TIME_REDS) + (void) erts_get_monotonic_time(esdp); + + if (esdp->last_monotonic_time >= erts_next_timeout_time(esdp->next_tmo_ref)) { + erts_runq_unlock(rq); + erts_bump_timers(esdp->timer_wheel, esdp->last_monotonic_time); + erts_runq_lock(rq); + } + if (rq->check_balance_reds <= 0) check_balance(rq); @@ -9951,37 +9870,6 @@ Process *erts_schedule(ErtsSchedulerData *esdp, Process *p, int calls) goto check_activities_to_run; } - else if (is_normal_sched - && (fcalls > input_reductions - && prepare_for_sys_schedule(!0))) { - ErtsMonotonicTime current_time; - /* - * Schedule system-level activities. - */ - - ERTS_MSACC_PUSH_STATE_CACHED_M(); - - esdp->function_calls = 0; - fcalls = 0; - -#if 0 /* Not needed since we wont wait in sys schedule */ - erts_check_io_interrupt(esdp->pollset, 0); -#endif - erts_runq_unlock(rq); - - ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_CHECK_IO); - LTTNG2(scheduler_poll, esdp->no, 1); - - erl_sys_schedule(1); - ERTS_MSACC_POP_STATE_M(); - - current_time = erts_get_monotonic_time(esdp); - if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) - erts_bump_timers(esdp->timer_wheel, current_time); - - erts_runq_lock(rq); - goto continue_check_activities_to_run; - } if (flags & ERTS_RUNQ_FLG_MISC_OP) exec_misc_ops(rq); diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h index b372146846..b4c06adc95 100644 --- a/erts/emulator/beam/erl_process.h +++ b/erts/emulator/beam/erl_process.h @@ -369,6 +369,7 @@ struct ErtsSchedulerSleepInfo_ { ErtsSchedulerSleepInfo *prev; erts_atomic32_t flags; erts_tse_t *event; + struct erts_poll_thread *psi; erts_atomic32_t aux_work; }; @@ -632,8 +633,6 @@ struct ErtsSchedulerData_ { void *match_pseudo_process; /* erl_db_util.c:db_prog_match() */ Process *free_process; ErtsThrPrgrData thr_progress_data; - struct pollset_info* pollset; - int function_calls; ErtsSchedulerSleepInfo *ssi; Process *current_process; ErtsSchedType type; @@ -1534,9 +1533,7 @@ extern int erts_system_profile_ts_type; void erts_pre_init_process(void); void erts_late_init_process(void); void erts_early_init_scheduling(int); -void erts_init_scheduling(int, int - , int, int, int - ); +void erts_init_scheduling(int, int, int, int, int, int); void erts_execute_dirty_system_task(Process *c_p); int erts_set_gc_state(Process *c_p, int enable); Eterm erts_sched_wall_time_request(Process *c_p, int set, int enable, @@ -2473,6 +2470,7 @@ void erts_notify_inc_runq(ErtsRunQueue *runq); void erts_sched_finish_poke(ErtsSchedulerSleepInfo *, erts_aint32_t); ERTS_GLB_INLINE void erts_sched_poke(ErtsSchedulerSleepInfo *ssi); +void erts_aux_thread_poke(void); #if ERTS_GLB_INLINE_INCL_FUNC_DEF diff --git a/erts/emulator/beam/erl_vm.h b/erts/emulator/beam/erl_vm.h index 82977e62ea..295130f60c 100644 --- a/erts/emulator/beam/erl_vm.h +++ b/erts/emulator/beam/erl_vm.h @@ -46,8 +46,6 @@ */ #define ERTS_X_REGS_ALLOCATED (MAX_REG+3) -#define INPUT_REDUCTIONS (CONTEXT_REDS / 4) - #define H_DEFAULT_SIZE 233 /* default (heap + stack) min size */ #define VH_DEFAULT_SIZE 32768 /* default virtual (bin) heap min size (words) */ #define H_DEFAULT_MAX_SIZE 0 /* default max heap size is off */ diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h index c3f8e92f13..09aeba00fa 100644 --- a/erts/emulator/beam/global.h +++ b/erts/emulator/beam/global.h @@ -1101,7 +1101,6 @@ void erts_save_stacktrace(Process* p, struct StackTrace* s, int depth); typedef struct { Eterm delay_time; int context_reds; - int input_reds; } ErtsModifiedTimings; extern Export *erts_delay_trap; diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h index 0bf2ecfc8b..9106091ca6 100644 --- a/erts/emulator/beam/sys.h +++ b/erts/emulator/beam/sys.h @@ -992,10 +992,6 @@ erts_refc_read(erts_refc_t *refcp, erts_aint_t min_val) #endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */ -#ifdef ERTS_ENABLE_KERNEL_POLL -extern int erts_use_kernel_poll; -#endif - #define sys_memcpy(s1,s2,n) memcpy(s1,s2,n) #define sys_memmove(s1,s2,n) memmove(s1,s2,n) #define sys_memcmp(s1,s2,n) memcmp(s1,s2,n) diff --git a/erts/emulator/sys/common/erl_check_io.c b/erts/emulator/sys/common/erl_check_io.c index ff0f3ea121..3131b96536 100644 --- a/erts/emulator/sys/common/erl_check_io.c +++ b/erts/emulator/sys/common/erl_check_io.c @@ -29,7 +29,6 @@ #endif #define ERL_CHECK_IO_C__ -#define ERTS_WANT_BREAK_HANDLING #ifndef WANT_NONBLOCKING # define WANT_NONBLOCKING #endif @@ -44,151 +43,108 @@ #define ERTS_WANT_TIMER_WHEEL_API #include "erl_time.h" +#if 0 +#define DEBUG_PRINT(FMT, ...) erts_printf(FMT "\r\n", ##__VA_ARGS__) +#define DEBUG_PRINT_FD(FMT, STATE, ...) \ + DEBUG_PRINT("%d: " FMT " (ev=%s, ac=%s, flg=%d)", \ + (STATE) ? (STATE)->fd : (ErtsSysFdType)-1, ##__VA_ARGS__, \ + ev2str((STATE) ? (STATE)->events : ERTS_POLL_EV_NONE), \ + ev2str((STATE) ? (STATE)->active_events : ERTS_POLL_EV_NONE), \ + (STATE) ? (STATE)->flags : ERTS_EV_FLAG_CLEAR) +#define DEBUG_PRINT_MODE +#else +#define DEBUG_PRINT(...) +#endif + +#ifndef DEBUG_PRINT_FD +#define DEBUG_PRINT_FD(...) +#endif + #ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS # include "safe_hash.h" # define DRV_EV_STATE_HTAB_SIZE 1024 #endif -typedef char EventStateType; -#define ERTS_EV_TYPE_NONE ((EventStateType) 0) -#define ERTS_EV_TYPE_DRV_SEL ((EventStateType) 1) /* driver_select */ -#define ERTS_EV_TYPE_STOP_USE ((EventStateType) 2) /* pending stop_select */ -#define ERTS_EV_TYPE_NIF ((EventStateType) 3) /* enif_select */ -#define ERTS_EV_TYPE_STOP_NIF ((EventStateType) 4) /* pending nif stop */ - -typedef char EventStateFlags; -#define ERTS_EV_FLAG_USED ((EventStateFlags) 1) /* ERL_DRV_USE has been turned on */ -#define ERTS_EV_FLAG_DEFER_IN_EV ((EventStateFlags) 2) -#define ERTS_EV_FLAG_DEFER_OUT_EV ((EventStateFlags) 4) +typedef enum { + ERTS_EV_TYPE_NONE = 0, + ERTS_EV_TYPE_DRV_SEL = 1, /* driver_select */ + ERTS_EV_TYPE_STOP_USE = 2, /* pending stop_select */ + ERTS_EV_TYPE_NIF = 3, /* enif_select */ + ERTS_EV_TYPE_STOP_NIF = 4 /* pending nif stop */ +} EventStateType; -#ifdef DEBUG -# define ERTS_ACTIVE_FD_INC 2 +typedef enum { + ERTS_EV_FLAG_CLEAR = 0, + ERTS_EV_FLAG_USED = 1, /* ERL_DRV_USE has been turned on */ +#ifdef ERTS_ENABLE_KERNEL_POLL + ERTS_EV_FLAG_FALLBACK = 2, /* Set when kernel poll rejected fd + and it was put in the nkp version */ #else -# define ERTS_ACTIVE_FD_INC 128 + ERTS_EV_FLAG_FALLBACK = ERTS_EV_FLAG_CLEAR, #endif -#define ERTS_CHECK_IO_POLL_RES_LEN 512 - -#if defined(ERTS_KERNEL_POLL_VERSION) -# define ERTS_CIO_EXPORT(FUNC) FUNC ## _kp -#elif defined(ERTS_NO_KERNEL_POLL_VERSION) -# define ERTS_CIO_EXPORT(FUNC) FUNC ## _nkp -#else -# define ERTS_CIO_EXPORT(FUNC) FUNC -#endif + /* Combinations */ + ERTS_EV_FLAG_USED_FALLBACK = ERTS_EV_FLAG_USED | ERTS_EV_FLAG_FALLBACK +} EventStateFlags; -#define ERTS_CIO_POLL_CTL ERTS_POLL_EXPORT(erts_poll_control) -#define ERTS_CIO_POLL_CTLV ERTS_POLL_EXPORT(erts_poll_controlv) -#define ERTS_CIO_POLL_WAIT ERTS_POLL_EXPORT(erts_poll_wait) -#define ERTS_CIO_POLL_INTR ERTS_POLL_EXPORT(erts_poll_interrupt) -#define ERTS_CIO_POLL_INTR_TMD ERTS_POLL_EXPORT(erts_poll_interrupt_timed) -#define ERTS_CIO_NEW_POLLSET ERTS_POLL_EXPORT(erts_poll_create_pollset) -#define ERTS_CIO_FREE_POLLSET ERTS_POLL_EXPORT(erts_poll_destroy_pollset) -#define ERTS_CIO_POLL_MAX_FDS ERTS_POLL_EXPORT(erts_poll_max_fds) -#define ERTS_CIO_POLL_INIT ERTS_POLL_EXPORT(erts_poll_init) -#define ERTS_CIO_POLL_INFO ERTS_POLL_EXPORT(erts_poll_info) +#define flag2str(flags) \ + ((flags) == ERTS_EV_FLAG_CLEAR ? "CLEAR" : \ + ((flags) == ERTS_EV_FLAG_USED ? "USED" : \ + ((flags) == ERTS_EV_FLAG_FALLBACK ? "FLBK" : \ + ((flags) == ERTS_EV_FLAG_USED_FALLBACK ? "USED|FLBK" : "ERROR")))) -#define GET_FD(fd) fd +/* How many events that can be handled at once by one erts_poll_wait call */ +#define ERTS_CHECK_IO_POLL_RES_LEN 512 -static struct pollset_info +/* Each I/O Poll Thread has one ErtsPollThread each. The ps field + can point to either a private ErtsPollSet or a shared one. + At the moment only kqueue and epoll pollsets can be + shared across threads. +*/ +typedef struct erts_poll_thread { - ErtsPollSet ps; - erts_atomic_t in_poll_wait; /* set while doing poll */ - erts_atomic_t check_io_time; - struct { - int six; /* start index */ - int eix; /* end index */ - erts_atomic32_t no; - int size; - ErtsSysFdType *array; - } active_fd; - erts_atomic_t removed_list; /* struct removed_fd* */ -}*pollsetv; - -#define NUM_OF_POLLSETS 1 + ErtsPollSet *ps; + ErtsPollResFd *pollres; + int pollres_len; +} ErtsPollThread; -#ifdef ERTS_ENABLE_KERNEL_POLL -void erts_init_check_io_kp(void); -void erts_init_check_io_nkp(void); -int ERTS_CIO_EXPORT(driver_select)(ErlDrvPort, ErlDrvEvent, int, int); -int ERTS_CIO_EXPORT(enif_select)(ErlNifEnv*, ErlNifEvent, enum ErlNifSelectFlags, void*, const ErlNifPid*, Eterm); -Uint ERTS_CIO_EXPORT(erts_check_io_size)(void); -Eterm ERTS_CIO_EXPORT(erts_check_io_info)(void *); -int ERTS_CIO_EXPORT(erts_check_io_max_files)(void); -void ERTS_CIO_EXPORT(erts_check_io_interrupt)(struct pollset_info*, int); -void ERTS_CIO_EXPORT(erts_check_io_interrupt_timed)(struct pollset_info*, int, ErtsMonotonicTime); -void ERTS_CIO_EXPORT(erts_check_io)(int); -struct pollset_info* ERTS_CIO_EXPORT(erts_get_pollset)(int); -int ERTS_CIO_EXPORT(erts_check_io_debug)(ErtsCheckIoDebugInfo *); -void ERTS_CIO_EXPORT(erts_io_notify_port_task_executed)(ErtsPortTaskHandle *pthp); -#ifdef ERTS_ENABLE_LOCK_COUNT -void ERTS_CIO_EXPORT(erts_lcnt_update_cio_locks)(int enable); -#endif +/* pollsetv contains pointers to the ErtsPollSets that are in use. + * Which pollset to use is determined by hashing the fd. + */ +static ErtsPollSet **pollsetv; +#if ERTS_POLL_USE_FALLBACK +static ErtsPollSet *flbk_pollset; #endif - -/* ToDo: Was inline in erl_check_io.h but now need struct pollset_info */ -void -ERTS_CIO_EXPORT(erts_io_notify_port_task_executed)(ErtsPortTaskHandle *pthp) -{ - ErtsIoTask *itp = ErtsContainerStruct(pthp, ErtsIoTask, task); - erts_aint_t ci_time = erts_atomic_read_acqb(&itp->pollset->check_io_time); - erts_atomic_set_relb(&itp->executed_time, ci_time); -} - - -struct pollset_info* ERTS_CIO_EXPORT(erts_get_pollset)(int sched_nr) -{ - ASSERT(sched_nr > 0 && sched_nr <= erts_no_schedulers); - return &pollsetv[sched_nr - 1]; -} +static ErtsPollThread *psiv; typedef struct { #ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS SafeHashBucket hb; #endif ErtsSysFdType fd; - struct pollset_info *pollset; struct { ErtsDrvSelectDataState *select; /* ERTS_EV_TYPE_DRV_SEL */ ErtsNifSelectDataState *nif; /* ERTS_EV_TYPE_NIF */ union { erts_driver_t* drv_ptr; /* ERTS_EV_TYPE_STOP_USE */ ErtsResource* resource; /* ERTS_EV_TYPE_STOP_NIF */ - }stop; + } stop; } driver; - ErtsPollEvents events; - unsigned short remove_cnt; /* number of removed_fd's referring to this fd */ + ErtsPollEvents events; /* The events that have been selected upon */ + ErtsPollEvents active_events; /* The events currently active in the pollset */ EventStateType type; EventStateFlags flags; } ErtsDrvEventState; -struct removed_fd { - struct removed_fd *next; -#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - ErtsSysFdType fd; -#else - ErtsDrvEventState* state; - #ifdef DEBUG - ErtsSysFdType fd; - #endif -#endif - -}; - struct drv_ev_state_shared { - /* The layout of this struct must be independent of kp/nkp compilation */ - -#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - int max_fds; -#endif -#define DRV_EV_STATE_LOCK_CNT 128 union { erts_mtx_t lck; byte _cache_line_alignment[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(erts_mtx_t))]; - }locks[DRV_EV_STATE_LOCK_CNT]; + } locks[ERTS_CHECK_IO_DRV_EV_STATE_LOCK_CNT]; #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + int max_fds; erts_atomic_t len; ErtsDrvEventState *v; erts_mtx_t grow_lock; /* prevent lock-hogging of racing growers */ @@ -200,51 +156,121 @@ struct drv_ev_state_shared { #endif }; -#ifndef ERTS_KERNEL_POLL_VERSION +int ERTS_WRITE_UNLIKELY(erts_no_pollsets) = 1; +int ERTS_WRITE_UNLIKELY(erts_no_poll_threads) = 1; struct drv_ev_state_shared drv_ev_state; -#else -extern struct drv_ev_state_shared drv_ev_state; -#endif -static ERTS_INLINE erts_mtx_t* fd_mtx(ErtsSysFdType fd) -{ +static ERTS_INLINE int fd_hash(ErtsSysFdType fd) { int hash = (int)fd; # ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS hash ^= (hash >> 9); # endif - return &drv_ev_state.locks[hash % DRV_EV_STATE_LOCK_CNT].lck; + return hash; } -#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS +static ERTS_INLINE erts_mtx_t* fd_mtx(ErtsSysFdType fd) +{ + return &drv_ev_state.locks[fd_hash(fd) % ERTS_CHECK_IO_DRV_EV_STATE_LOCK_CNT].lck; +} + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS -static ERTS_INLINE ErtsDrvEventState *hash_get_drv_ev_state(ErtsSysFdType fd) +static ERTS_INLINE ErtsDrvEventState *get_drv_ev_state(ErtsSysFdType fd) +{ + return &drv_ev_state.v[(int) fd]; +} + +#define new_drv_ev_state(State, fd) (State) +#define erase_drv_ev_state(State) + +static ERTS_INLINE int grow_drv_ev_state(ErtsSysFdType fd) { + int i; + int old_len; + int new_len; + + if ((unsigned)fd >= (unsigned)erts_atomic_read_nob(&drv_ev_state.len)) { + + if (fd < 0 || fd >= drv_ev_state.max_fds) + return 0; + + erts_mtx_lock(&drv_ev_state.grow_lock); + old_len = erts_atomic_read_nob(&drv_ev_state.len); + if (fd >= old_len) { + new_len = erts_poll_new_table_len(old_len, fd + 1); + if (new_len > drv_ev_state.max_fds) + new_len = drv_ev_state.max_fds; + + for (i=0; iremove_cnt == 0); safe_hash_erase(&drv_ev_state.tab, (void *) state); } +static int drv_ev_state_len(void) +{ + return erts_atomic_read_nob(&drv_ev_state.tab.nitems); +} + #endif /* !ERTS_SYS_CONTINOUS_FD_NUMBERS */ static void stale_drv_select(Eterm id, ErtsDrvEventState *state, int mode); @@ -268,36 +294,39 @@ steal_pending_stop_use(erts_dsprintf_buf_t*, ErlDrvPort, ErtsDrvEventState*, static void steal_pending_stop_nif(erts_dsprintf_buf_t *dsbufp, ErtsResource*, ErtsDrvEventState *state, int mode, int on); - -ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(removed_fd, struct removed_fd, 64, ERTS_ALC_T_FD_LIST) +static ERTS_INLINE void +check_fd_cleanup(ErtsDrvEventState *state, + ErtsDrvSelectDataState **free_select, + ErtsNifSelectDataState **free_nif); +#ifdef DEBUG_PRINT_MODE +static char *drvmode2str(int mode); +static char *nifmode2str(enum ErlNifSelectFlags mode); +#endif static ERTS_INLINE void -init_iotask(ErtsIoTask *io_task, struct pollset_info* psi) +init_iotask(ErtsIoTask *io_task, ErtsSysFdType fd) { erts_port_task_handle_init(&io_task->task); - erts_atomic_init_nob(&io_task->executed_time, ~((erts_aint_t) 0)); - io_task->pollset = psi; + io_task->fd = fd; } static ERTS_INLINE int -is_iotask_active(ErtsIoTask *io_task, erts_aint_t current_cio_time) -{ +is_iotask_active(ErtsIoTask *io_task) +{ if (erts_port_task_is_scheduled(&io_task->task)) return 1; - if (erts_atomic_read_nob(&io_task->executed_time) == current_cio_time) - return 1; return 0; } static ERTS_INLINE ErtsDrvSelectDataState * -alloc_drv_select_data(struct pollset_info* psi) +alloc_drv_select_data(ErtsSysFdType fd) { ErtsDrvSelectDataState *dsp = erts_alloc(ERTS_ALC_T_DRV_SEL_D_STATE, sizeof(ErtsDrvSelectDataState)); dsp->inport = NIL; dsp->outport = NIL; - init_iotask(&dsp->iniotask, psi); - init_iotask(&dsp->outiotask, psi); + init_iotask(&dsp->iniotask, fd); + init_iotask(&dsp->outiotask, fd); return dsp; } @@ -308,8 +337,6 @@ alloc_nif_select_data(void) sizeof(ErtsNifSelectDataState)); dsp->in.pid = NIL; dsp->out.pid = NIL; - dsp->in.ddeselect_cnt = 0; - dsp->out.ddeselect_cnt = 0; return dsp; } @@ -327,176 +354,126 @@ free_nif_select_data(ErtsNifSelectDataState *dsp) erts_free(ERTS_ALC_T_NIF_SEL_D_STATE, dsp); } -static ERTS_INLINE void -remember_removed(ErtsDrvEventState *state) +static ERTS_INLINE int +get_pollset_id(ErtsSysFdType fd) { - struct removed_fd *fdlp; - ERTS_LC_ASSERT(erts_lc_mtx_is_locked(fd_mtx(state->fd))); - if (erts_atomic_read_nob(&state->pollset->in_poll_wait)) { - erts_aint_t was_next, exp_next; - state->remove_cnt++; - ASSERT(state->remove_cnt > 0); - fdlp = removed_fd_alloc(); - #if defined(ERTS_SYS_CONTINOUS_FD_NUMBERS) || defined(DEBUG) - fdlp->fd = state->fd; - #endif - #ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS - fdlp->state = state; - #endif - - /* Lockless atomic insertion in removed_list */ - was_next = erts_atomic_read_acqb(&state->pollset->removed_list); - do { - exp_next = was_next; - fdlp->next = (struct removed_fd*) exp_next; - was_next = erts_atomic_cmpxchg_mb(&state->pollset->removed_list, - (erts_aint_t) fdlp, - exp_next); - }while (was_next != exp_next); - } + return fd_hash(fd) % erts_no_pollsets; } +static ERTS_INLINE ErtsPollSet * +get_pollset(ErtsSysFdType fd) +{ + return pollsetv[get_pollset_id(fd)]; +} -static ERTS_INLINE int -is_removed(ErtsDrvEventState *state) +#if ERTS_POLL_USE_FALLBACK +static ERTS_INLINE ErtsPollSet * +get_fallback(void) { - /* Note that there is a possible race here, where an fd is removed - (increasing remove_cnt) and then added again just before erts_poll_wait - is called by erts_check_io. Any polled event on the re-added fd will then - be falsely ignored. But that does not matter, as the event will trigger - again next time erl_check_io is called. */ - return state->remove_cnt > 0; + return flbk_pollset; } +#endif -static void -forget_removed(struct pollset_info* psi) +/* + * Place a fd within a pollset. This will automatically use + * the fallback ps if needed. + */ +static ERTS_INLINE ErtsPollEvents +erts_io_control_wakeup(ErtsDrvEventState *state, ErtsPollOp op, + ErtsPollEvents pe, int *wake_poller) { - struct removed_fd* fdlp; - struct removed_fd* tofree; + ErtsSysFdType fd = state->fd; + ErtsPollEvents res = 0; + EventStateFlags flags = state->flags; - fdlp = (struct removed_fd*) erts_atomic_xchg_mb(&psi->removed_list, - (erts_aint_t) NULL); + ERTS_LC_ASSERT(erts_lc_mtx_is_locked(fd_mtx(state->fd))); - while (fdlp) { - ErtsResource* resource = NULL; - erts_driver_t* drv_ptr = NULL; - erts_mtx_t* mtx; - ErtsSysFdType fd; - ErtsDrvEventState *state; + if (!(flags & ERTS_EV_FLAG_FALLBACK)) { + res = erts_poll_control(get_pollset(fd), fd, op, pe, wake_poller); -#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - fd = fdlp->fd; - mtx = fd_mtx(fd); - erts_mtx_lock(mtx); - state = &drv_ev_state.v[(int) fd]; -#else - state = fdlp->state; - fd = state->fd; - ASSERT(fd == fdlp->fd); - mtx = fd_mtx(fd); - erts_mtx_lock(mtx); -#endif - ASSERT(state->remove_cnt > 0); - if (--state->remove_cnt == 0) { - switch (state->type) { - case ERTS_EV_TYPE_STOP_NIF: - /* Now we can call stop */ - resource = state->driver.stop.resource; - state->driver.stop.resource = NULL; - ASSERT(resource); - state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; - goto case_ERTS_EV_TYPE_NONE; - - case ERTS_EV_TYPE_STOP_USE: - /* Now we can call stop_select */ - drv_ptr = state->driver.stop.drv_ptr; - ASSERT(drv_ptr); - state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; - state->driver.stop.drv_ptr = NULL; - /* Fall through */ - case ERTS_EV_TYPE_NONE: - case_ERTS_EV_TYPE_NONE: - state->pollset = NULL; -#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS - hash_erase_drv_ev_state(state); -#endif - break; - case ERTS_EV_TYPE_DRV_SEL: - break; - default: - ASSERT(0); - } - } - erts_mtx_unlock(mtx); - if (drv_ptr) { - int was_unmasked = erts_block_fpe(); - DTRACE1(driver_stop_select, drv_ptr->name); - LTTNG1(driver_stop_select, drv_ptr->name); - (*drv_ptr->stop_select) ((ErlDrvEvent) fd, NULL); - erts_unblock_fpe(was_unmasked); - if (drv_ptr->handle) { - erts_ddll_dereference_driver(drv_ptr->handle); - } - } - if (resource) { - erts_resource_stop(resource, (ErlNifEvent)fd, 0); - enif_release_resource(resource->data); +#if ERTS_POLL_USE_FALLBACK + if (op == ERTS_POLL_OP_ADD && res == ERTS_POLL_EV_NVAL) { + /* When an add fails with NVAL, the poll/kevent operation could not + put that fd in the pollset, so we instead put it into a fallback pollset */ + state->flags |= ERTS_EV_FLAG_FALLBACK; + res = erts_poll_control_flbk(get_fallback(), fd, op, pe, wake_poller); } - - tofree = fdlp; - fdlp = fdlp->next; - removed_fd_free(tofree); + } else { + ASSERT(op != ERTS_POLL_OP_ADD); + res = erts_poll_control_flbk(get_fallback(), fd, op, pe, wake_poller); +#endif } + + return res; } -#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS -static void -grow_drv_ev_state(int min_ix) +static ERTS_INLINE ErtsPollEvents +erts_io_control(ErtsDrvEventState *state, ErtsPollOp op, ErtsPollEvents pe) { - int i; - int old_len; - int new_len; + int wake_poller = 0; + return erts_io_control_wakeup(state, op, pe, &wake_poller); +} - erts_mtx_lock(&drv_ev_state.grow_lock); - old_len = erts_atomic_read_nob(&drv_ev_state.len); - if (min_ix >= old_len) { - new_len = erts_poll_new_table_len(old_len, min_ix + 1); - if (new_len > drv_ev_state.max_fds) - new_len = drv_ev_state.max_fds; - - for (i=0; ifd; + erts_mtx_t *mtx = fd_mtx(fd); + int active_events; + ErtsDrvEventState *state; + ErtsDrvSelectDataState *free_select = NULL; + ErtsNifSelectDataState *free_nif = NULL; + + erts_mtx_lock(mtx); + state = get_drv_ev_state(fd); + + active_events = state->active_events; + + switch (type) { + case ERTS_PORT_TASK_INPUT: + + DEBUG_PRINT_FD("executed ready_input", state); + + ASSERT(!(state->active_events & ERTS_POLL_EV_IN)); + if (state->events & ERTS_POLL_EV_IN) + active_events |= ERTS_POLL_EV_IN; + break; + case ERTS_PORT_TASK_OUTPUT: + + DEBUG_PRINT_FD("executed ready_output", state); + + ASSERT(!(state->active_events & ERTS_POLL_EV_OUT)); + if (state->events & ERTS_POLL_EV_OUT) + active_events |= ERTS_POLL_EV_OUT; + break; + default: + erts_exit(ERTS_ABORT_EXIT, "Invalid IO port task type"); + break; } - /*else already grown by racing thread */ - erts_mtx_unlock(&drv_ev_state.grow_lock); -} -#endif /* ERTS_SYS_CONTINOUS_FD_NUMBERS */ + reset_handle(pthp); + + if (active_events) { + /* This is not needed if active_events has not changed */ + if (state->active_events != active_events) { + state->active_events = active_events; + erts_io_control(state, ERTS_POLL_OP_MOD, active_events); + } + } else { + check_fd_cleanup(state, &free_select, &free_nif); + } + erts_mtx_unlock(mtx); + + if (free_select) + free_drv_select_data(free_select); + if (free_nif) + free_nif_select_data(free_nif); +} static ERTS_INLINE void abort_task(Eterm id, ErtsPortTaskHandle *pthp, EventStateType type) @@ -542,16 +519,14 @@ abort_tasks(ErtsDrvEventState *state, int mode) static void deselect(ErtsDrvEventState *state, int mode) { - int do_wake = 0; ErtsPollEvents rm_events; ERTS_LC_ASSERT(erts_lc_mtx_is_locked(fd_mtx(state->fd))); - ASSERT(state->events); abort_tasks(state, mode); - if (!mode) + if (!mode) { rm_events = state->events; - else { + } else { rm_events = 0; ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); if (mode & ERL_DRV_READ) { @@ -564,18 +539,16 @@ deselect(ErtsDrvEventState *state, int mode) } } - ERTS_CIO_POLL_CTL(state->pollset->ps, state->fd, rm_events, - 0, &do_wake); state->events &= ~rm_events; + state->active_events &= ~rm_events; if (!(state->events)) { + erts_io_control(state, ERTS_POLL_OP_DEL, 0); switch (state->type) { case ERTS_EV_TYPE_NIF: state->driver.nif->in.pid = NIL; state->driver.nif->out.pid = NIL; - state->driver.nif->in.ddeselect_cnt = 0; - state->driver.nif->out.ddeselect_cnt = 0; - enif_release_resource(state->driver.stop.resource); + enif_release_resource(state->driver.stop.resource->data); state->driver.stop.resource = NULL; break; case ERTS_EV_TYPE_DRV_SEL: @@ -588,15 +561,15 @@ deselect(ErtsDrvEventState *state, int mode) ASSERT(0); break; } - state->type = ERTS_EV_TYPE_NONE; state->flags = 0; - remember_removed(state); + } else { + erts_io_control(state, ERTS_POLL_OP_MOD, state->active_events); } } #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS -# define IS_FD_UNKNOWN(state) ((state)->type == ERTS_EV_TYPE_NONE && (state)->remove_cnt == 0) +# define IS_FD_UNKNOWN(state) ((state)->type == ERTS_EV_TYPE_NONE) #else # define IS_FD_UNKNOWN(state) ((state) == NULL) #endif @@ -606,17 +579,13 @@ check_fd_cleanup(ErtsDrvEventState *state, ErtsDrvSelectDataState **free_select, ErtsNifSelectDataState **free_nif) { - erts_aint_t current_cio_time; - ERTS_LC_ASSERT(erts_lc_mtx_is_locked(fd_mtx(state->fd))); - - current_cio_time = erts_atomic_read_acqb(&state->pollset->check_io_time); *free_select = NULL; if (state->driver.select && (state->type != ERTS_EV_TYPE_DRV_SEL) - && !is_iotask_active(&state->driver.select->iniotask, current_cio_time) - && !is_iotask_active(&state->driver.select->outiotask, current_cio_time)) { - + && !is_iotask_active(&state->driver.select->iniotask) + && !is_iotask_active(&state->driver.select->outiotask)) { + *free_select = state->driver.select; state->driver.select = NULL; } @@ -628,14 +597,10 @@ check_fd_cleanup(ErtsDrvEventState *state, } if (((state->type != ERTS_EV_TYPE_NONE) - | state->remove_cnt | (state->driver.nif != NULL) | (state->driver.select != NULL)) == 0) { - state->pollset = NULL; -#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS - hash_erase_drv_ev_state(state); -#endif + erase_drv_ev_state(state); } } @@ -645,253 +610,8 @@ check_fd_cleanup(ErtsDrvEventState *state, # define MUST_DEFER(MAY_SLEEP) (MAY_SLEEP) #endif -static ERTS_INLINE int -check_cleanup_active_fd(struct pollset_info* psi, - ErtsSysFdType fd, -#if ERTS_CIO_DEFER_ACTIVE_EVENTS - ErtsPollControlEntry *pce, - int *pce_ix, -#endif - erts_aint_t current_cio_time, - int may_sleep) -{ - ErtsDrvEventState *state; - int active = 0; - erts_mtx_t *mtx = fd_mtx(fd); - void *free_select = NULL; - void *free_nif = NULL; -#if ERTS_CIO_DEFER_ACTIVE_EVENTS - ErtsPollEvents evon = 0, evoff = 0; -#endif - - erts_mtx_lock(mtx); - -#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - state = &drv_ev_state.v[(int) fd]; -#else - state = hash_get_drv_ev_state(fd); /* may be NULL! */ -#endif - if (state && state->pollset == psi) - { - if (state->driver.select) { -#if ERTS_CIO_DEFER_ACTIVE_EVENTS - if (is_iotask_active(&state->driver.select->iniotask, current_cio_time)) { - active = 1; - if (MUST_DEFER(may_sleep) - && (state->events & ERTS_POLL_EV_IN) - && !(state->flags & ERTS_EV_FLAG_DEFER_IN_EV)) { - evoff |= ERTS_POLL_EV_IN; - state->flags |= ERTS_EV_FLAG_DEFER_IN_EV; - } - } - else if (state->flags & ERTS_EV_FLAG_DEFER_IN_EV) { - if (state->events & ERTS_POLL_EV_IN) - evon |= ERTS_POLL_EV_IN; - state->flags &= ~ERTS_EV_FLAG_DEFER_IN_EV; - } - if (is_iotask_active(&state->driver.select->outiotask, current_cio_time)) { - active = 1; - if (MUST_DEFER(may_sleep) - && (state->events & ERTS_POLL_EV_OUT) - && !(state->flags & ERTS_EV_FLAG_DEFER_OUT_EV)) { - evoff |= ERTS_POLL_EV_OUT; - state->flags |= ERTS_EV_FLAG_DEFER_OUT_EV; - } - } - else if (state->flags & ERTS_EV_FLAG_DEFER_OUT_EV) { - if (state->events & ERTS_POLL_EV_OUT) { - evon |= ERTS_POLL_EV_OUT; - } - state->flags &= ~ERTS_EV_FLAG_DEFER_OUT_EV; - } - if (active) - (void) 0; - else -#else - if (is_iotask_active(&state->driver.select->iniotask, current_cio_time) - || is_iotask_active(&state->driver.select->outiotask, current_cio_time)) - active = 1; - else -#endif - if (state->type != ERTS_EV_TYPE_DRV_SEL) { - free_select = state->driver.select; - state->driver.select = NULL; - } -#if ERTS_CIO_DEFER_ACTIVE_EVENTS - if (evon) { - int do_wake = 0; - ERTS_CIO_POLL_CTL(psi->ps, state->fd, evon, 1, &do_wake); - } -#endif - } - - if (state->driver.nif) { - ErtsPollEvents rm_events = 0; - if (state->driver.nif->in.ddeselect_cnt) { - ASSERT(state->type == ERTS_EV_TYPE_NIF); - ASSERT(state->events & ERTS_POLL_EV_IN); - ASSERT(is_nil(state->driver.nif->in.pid)); - if (may_sleep || state->driver.nif->in.ddeselect_cnt == 1) { - rm_events = ERTS_POLL_EV_IN; - state->driver.nif->in.ddeselect_cnt = 0; - } - } - if (state->driver.nif->out.ddeselect_cnt) { - ASSERT(state->type == ERTS_EV_TYPE_NIF); - ASSERT(state->events & ERTS_POLL_EV_OUT); - ASSERT(is_nil(state->driver.nif->out.pid)); - if (may_sleep || state->driver.nif->out.ddeselect_cnt == 1) { - rm_events |= ERTS_POLL_EV_OUT; - state->driver.nif->out.ddeselect_cnt = 0; - } - } - if (rm_events) { - int do_wake = 0; - state->events = ERTS_CIO_POLL_CTL(state->pollset->ps, state->fd, - rm_events, 0, &do_wake); - } - if (state->events) - active = 1; - else if (state->type != ERTS_EV_TYPE_NIF) { - free_nif = state->driver.nif; - state->driver.nif = NULL; - } - } - -#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS - if (((state->type != ERTS_EV_TYPE_NONE) | state->remove_cnt | active) == 0) - hash_erase_drv_ev_state(state); -#endif - } - - erts_mtx_unlock(mtx); - - if (free_select) - free_drv_select_data(free_select); - if (free_nif) - free_nif_select_data(free_nif); - -#if ERTS_CIO_DEFER_ACTIVE_EVENTS - if (evoff) { - ErtsPollControlEntry *pcep = &pce[(*pce_ix)++]; - pcep->fd = fd; - pcep->events = evoff; - pcep->on = 0; - } -#endif - - return active; -} - -static void -check_cleanup_active_fds(struct pollset_info* psi, - erts_aint_t current_cio_time, int may_sleep) -{ - int six = psi->active_fd.six; - int eix = psi->active_fd.eix; - erts_aint32_t no = erts_atomic32_read_dirty(&psi->active_fd.no); - const int size = psi->active_fd.size; - int ix = six; -#if ERTS_CIO_DEFER_ACTIVE_EVENTS - /* every fd might add one entry */ - Uint pce_sz = sizeof(ErtsPollControlEntry)*no; - ErtsPollControlEntry *pctrl_entries = (pce_sz - ? erts_alloc(ERTS_ALC_T_TMP, pce_sz) - : NULL); - int pctrl_ix = 0; -#endif - - while (ix != eix) { - ErtsSysFdType fd = psi->active_fd.array[ix]; - int nix = ix + 1; - if (nix >= size) - nix = 0; - ASSERT(fd != ERTS_SYS_FD_INVALID); - if (!check_cleanup_active_fd(psi, fd, -#if ERTS_CIO_DEFER_ACTIVE_EVENTS - pctrl_entries, - &pctrl_ix, -#endif - current_cio_time, - may_sleep)) { - no--; - if (ix == six) { -#ifdef DEBUG - psi->active_fd.array[ix] = ERTS_SYS_FD_INVALID; -#endif - six = nix; - } - else { - psi->active_fd.array[ix] = psi->active_fd.array[six]; -#ifdef DEBUG - psi->active_fd.array[six] = ERTS_SYS_FD_INVALID; -#endif - six++; - if (six >= size) - six = 0; - } - } - ix = nix; - } - -#if ERTS_CIO_DEFER_ACTIVE_EVENTS - ASSERT(pctrl_ix <= pce_sz/sizeof(ErtsPollControlEntry)); - if (pctrl_ix) - ERTS_CIO_POLL_CTLV(psi->ps, pctrl_entries, pctrl_ix); - if (pctrl_entries) - erts_free(ERTS_ALC_T_TMP, pctrl_entries); -#endif - - psi->active_fd.six = six; - psi->active_fd.eix = eix; - erts_atomic32_set_relb(&psi->active_fd.no, no); -} - -static void grow_active_fds(struct pollset_info *psi) -{ - ASSERT(psi->active_fd.six == psi->active_fd.eix); - psi->active_fd.six = 0; - psi->active_fd.eix = psi->active_fd.size; - psi->active_fd.size += ERTS_ACTIVE_FD_INC; - psi->active_fd.array = erts_realloc(ERTS_ALC_T_ACTIVE_FD_ARR, - psi->active_fd.array, - psi->active_fd.size*sizeof(ErtsSysFdType)); -#ifdef DEBUG - { - int i; - for (i = psi->active_fd.eix + 1; i < psi->active_fd.size; i++) - psi->active_fd.array[i] = ERTS_SYS_FD_INVALID; - } -#endif -} - -static ERTS_INLINE void -add_active_fd(struct pollset_info *psi, ErtsSysFdType fd) -{ - int eix = psi->active_fd.eix; - const int size = psi->active_fd.size; - - psi->active_fd.array[eix] = fd; - - erts_atomic32_set_relb(&psi->active_fd.no, - (erts_atomic32_read_dirty(&psi->active_fd.no) - + 1)); - - eix++; - if (eix >= size) - eix = 0; - psi->active_fd.eix = eix; - - if (psi->active_fd.six == eix) { - grow_active_fds(psi); - } -} - int -ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, - ErlDrvEvent e, - int mode, - int on) +driver_select(ErlDrvPort ix, ErlDrvEvent e, int mode, int on) { void (*stop_select_fn)(ErlDrvEvent, void*) = NULL; Port *prt = erts_drvport2port(ix); @@ -899,6 +619,7 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, ErtsSysFdType fd = (ErtsSysFdType) e; ErtsPollEvents ctl_events = (ErtsPollEvents) 0; ErtsPollEvents old_events; + ErtsPollOp ctl_op = ERTS_POLL_OP_MOD; ErtsDrvEventState *state; int wake_poller = 0; int ret; @@ -907,32 +628,29 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, #ifdef USE_VM_PROBES DTRACE_CHARBUF(name, 64); #endif + ERTS_MSACC_PUSH_AND_SET_STATE(ERTS_MSACC_STATE_CHECK_IO); - if (prt == ERTS_INVALID_ERL_DRV_PORT) + if (prt == ERTS_INVALID_ERL_DRV_PORT) { + ERTS_MSACC_POP_STATE(); return -1; + } ERTS_LC_ASSERT(erts_lc_is_port_locked(prt)); #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - if ((unsigned)fd >= (unsigned)erts_atomic_read_nob(&drv_ev_state.len)) { - if (fd < 0) { - return -1; - } - if (fd >= drv_ev_state.max_fds) { - drv_select_large_fd_error(ix, fd, mode, on); - return -1; - } - grow_drv_ev_state(fd); + if (!grow_drv_ev_state(fd)) { + if (fd > 0) drv_select_large_fd_error(ix, fd, mode, on); + ERTS_MSACC_POP_STATE(); + return -1; } #endif erts_mtx_lock(fd_mtx(fd)); -#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - state = &drv_ev_state.v[(int) fd]; -#else - state = hash_get_drv_ev_state(fd); /* may be NULL! */ -#endif + state = get_drv_ev_state(fd); /* may be NULL! */ + + DEBUG_PRINT_FD("driver_select(%T, %p, %s, %d)", + state, id, fd, drvmode2str(mode), on); if (!on) { if (IS_FD_UNKNOWN(state)) { @@ -950,15 +668,10 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, } else if ((mode&ERL_DRV_USE_NO_CALLBACK) == ERL_DRV_USE) { mode |= (ERL_DRV_READ | ERL_DRV_WRITE); - wake_poller = 1; /* to eject fd from pollset (if needed) */ } } -#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS - if (state == NULL) { - state = hash_new_drv_ev_state(fd); - } -#endif + state = new_drv_ev_state(state, fd); switch (state->type) { case ERTS_EV_TYPE_NIF: @@ -982,7 +695,9 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, ASSERT(state->type == ERTS_EV_TYPE_NONE); break; - }} + } + default: break; + } if (mode & ERL_DRV_READ) { if (state->type == ERTS_EV_TYPE_DRV_SEL) { @@ -999,7 +714,7 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, drv_select_steal(ix, state, mode, on); } ctl_events |= ERTS_POLL_EV_OUT; - } + } ASSERT((state->type == ERTS_EV_TYPE_DRV_SEL) || @@ -1010,32 +725,31 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, if (on) { ctl_events &= ~old_events; state->events |= ctl_events; + if (ctl_events & ERTS_POLL_EV_IN && (!state->driver.select || !is_iotask_active(&state->driver.select->iniotask))) + state->active_events |= ERTS_POLL_EV_IN; + if (ctl_events & ERTS_POLL_EV_OUT && (!state->driver.select || !is_iotask_active(&state->driver.select->outiotask))) + state->active_events |= ERTS_POLL_EV_OUT; + if (old_events == 0 && !(state->flags & ERTS_EV_FLAG_USED)) { + ctl_op = ERTS_POLL_OP_ADD; + } } else { ctl_events &= old_events; state->events &= ~ctl_events; + state->active_events &= ~ctl_events; - if (!(state->flags & ERTS_EV_FLAG_USED) - && old_events && !state->events) { - /* - * Old driver removing all events. At least wake poller. - * It will not make close() 100% safe but it will prevent - * actions delayed by poll timeout. - */ - wake_poller = 1; + if (!state->events) { + if (!(state->flags & ERTS_EV_FLAG_USED) || mode & ERL_DRV_USE) + ctl_op = ERTS_POLL_OP_DEL; } } - if (ctl_events) { + if (ctl_events || ctl_op == ERTS_POLL_OP_DEL) { ErtsPollEvents new_events; - if (!state->pollset) { - ErtsSchedulerData* esdp = erts_get_scheduler_data(); - ASSERT(esdp); - state->pollset = esdp->pollset; - } - - new_events = ERTS_CIO_POLL_CTL(state->pollset->ps, state->fd, ctl_events, on, &wake_poller); + new_events = erts_io_control_wakeup(state, ctl_op, + state->active_events, + &wake_poller); if (new_events & (ERTS_POLL_EV_ERR|ERTS_POLL_EV_NVAL)) { if (state->type == ERTS_EV_TYPE_DRV_SEL && !old_events) { @@ -1048,14 +762,13 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, goto done; } - ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL - || state->type == ERTS_EV_TYPE_NONE); + ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL || state->type == ERTS_EV_TYPE_NONE); } if (on) { if (ctl_events) { if (!state->driver.select) - state->driver.select = alloc_drv_select_data(state->pollset); + state->driver.select = alloc_drv_select_data(state->fd); if (state->type == ERTS_EV_TYPE_NONE) state->type = ERTS_EV_TYPE_DRV_SEL; ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); @@ -1069,47 +782,44 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, } } else { /* off */ - if (state->type == ERTS_EV_TYPE_DRV_SEL) { - if (ctl_events & ERTS_POLL_EV_IN) { - abort_tasks(state, ERL_DRV_READ); - state->driver.select->inport = NIL; - } - if (ctl_events & ERTS_POLL_EV_OUT) { - abort_tasks(state, ERL_DRV_WRITE); - state->driver.select->outport = NIL; - } - if (state->events == 0) { - if (old_events != 0) { - remember_removed(state); - } - if ((mode & ERL_DRV_USE) || !(state->flags & ERTS_EV_FLAG_USED)) { - state->type = ERTS_EV_TYPE_NONE; - state->flags = 0; - } - /*else keep it, as fd will probably be selected upon again */ - } - } - if ((mode & ERL_DRV_USE_NO_CALLBACK) == ERL_DRV_USE) { - erts_driver_t* drv_ptr = prt->drv_ptr; - ASSERT(state->events==0); - if (state->remove_cnt == 0 || !wake_poller) { - /* Safe to close fd now as it is not in pollset - or there was no need to eject fd (kernel poll) */ - stop_select_fn = drv_ptr->stop_select; + if (state->type == ERTS_EV_TYPE_DRV_SEL) { + if (ctl_events & ERTS_POLL_EV_IN) { + abort_tasks(state, ERL_DRV_READ); + state->driver.select->inport = NIL; + } + if (ctl_events & ERTS_POLL_EV_OUT) { + abort_tasks(state, ERL_DRV_WRITE); + state->driver.select->outport = NIL; + } + if (state->events == 0) { + if ((mode & ERL_DRV_USE) || !(state->flags & ERTS_EV_FLAG_USED)) { + state->type = ERTS_EV_TYPE_NONE; + state->flags = 0; + } + /*else keep it, as fd will probably be selected upon again */ + } + } + if ((mode & ERL_DRV_USE_NO_CALLBACK) == ERL_DRV_USE) { + erts_driver_t* drv_ptr = prt->drv_ptr; + ASSERT(state->events==0); + if (!wake_poller) { + /* Safe to close fd now as it is not in pollset + or there was no need to eject fd (kernel poll) */ + stop_select_fn = drv_ptr->stop_select; #ifdef USE_VM_PROBES - strncpy(name, prt->drv_ptr->name, sizeof(name)-1); - name[sizeof(name)-1] = '\0'; + strncpy(name, prt->drv_ptr->name, sizeof(name)-1); + name[sizeof(name)-1] = '\0'; #endif - } - else { - /* Not safe to close fd, postpone stop_select callback. */ - state->type = ERTS_EV_TYPE_STOP_USE; - state->driver.stop.drv_ptr = drv_ptr; - if (drv_ptr->handle) { - erts_ddll_reference_referenced_driver(drv_ptr->handle); - } - } - } + } + else { + /* Not safe to close fd, postpone stop_select callback. */ + state->type = ERTS_EV_TYPE_STOP_USE; + state->driver.stop.drv_ptr = drv_ptr; + if (drv_ptr->handle) { + erts_ddll_reference_referenced_driver(drv_ptr->handle); + } + } + } } ret = 0; @@ -1134,54 +844,47 @@ done_unknown: if (free_nif) free_nif_select_data(free_nif); + ERTS_MSACC_POP_STATE(); + return ret; } int -ERTS_CIO_EXPORT(enif_select)(ErlNifEnv* env, - ErlNifEvent e, - enum ErlNifSelectFlags mode, - void* obj, - const ErlNifPid* pid, - Eterm ref) +enif_select(ErlNifEnv* env, + ErlNifEvent e, + enum ErlNifSelectFlags mode, + void* obj, + const ErlNifPid* pid, + Eterm ref) { int on; ErtsResource* resource = DATA_TO_RESOURCE(obj); ErtsSysFdType fd = (ErtsSysFdType) e; ErtsPollEvents ctl_events = (ErtsPollEvents) 0; ErtsPollEvents old_events; + ErtsPollOp ctl_op = ERTS_POLL_OP_MOD; ErtsDrvEventState *state; - int wake_poller; - int ret; + int ret, wake_poller = 0; enum { NO_STOP=0, CALL_STOP, CALL_STOP_AND_RELEASE } call_stop = NO_STOP; ErtsDrvSelectDataState *free_select = NULL; ErtsNifSelectDataState *free_nif = NULL; -#ifdef USE_VM_PROBES - DTRACE_CHARBUF(name, 64); -#endif ASSERT(!(resource->monitors && resource->monitors->is_dying)); #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - if ((unsigned)fd >= (unsigned)erts_atomic_read_nob(&drv_ev_state.len)) { - if (fd < 0) { - return INT_MIN | ERL_NIF_SELECT_INVALID_EVENT; - } - if (fd >= drv_ev_state.max_fds) { - nif_select_large_fd_error(fd, mode, resource, ref); - return INT_MIN | ERL_NIF_SELECT_INVALID_EVENT; - } - grow_drv_ev_state(fd); + if (!grow_drv_ev_state(fd)) { + if (fd > 0) nif_select_large_fd_error(fd, mode, resource, ref); + return INT_MIN | ERL_NIF_SELECT_INVALID_EVENT; } #endif erts_mtx_lock(fd_mtx(fd)); -#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - state = &drv_ev_state.v[(int) fd]; -#else - state = hash_get_drv_ev_state(fd); /* may be NULL! */ -#endif + state = get_drv_ev_state(fd); /* may be NULL! */ + + DEBUG_PRINT_FD("enif_select(%T, %d, %s, %p, %T, %T)", + state, env->proc->common.id, fd, nifmode2str(mode), resource, + pid ? pid->pid : THE_NON_VALUE, ref); if (mode & ERL_NIF_SELECT_STOP) { ASSERT(resource->type->stop); @@ -1193,13 +896,12 @@ ERTS_CIO_EXPORT(enif_select)(ErlNifEnv* env, } on = 0; mode = ERL_DRV_READ | ERL_DRV_WRITE | ERL_DRV_USE; - wake_poller = 1; /* to eject fd from pollset (if needed) */ ctl_events = ERTS_POLL_EV_IN | ERTS_POLL_EV_OUT; + ctl_op = ERTS_POLL_OP_DEL; } else { on = 1; ASSERT(mode); - wake_poller = 0; if (mode & ERL_DRV_READ) { ctl_events |= ERTS_POLL_EV_IN; } @@ -1208,11 +910,7 @@ ERTS_CIO_EXPORT(enif_select)(ErlNifEnv* env, } } -#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS - if (state == NULL) { - state = hash_new_drv_ev_state(fd); - } -#endif + state = new_drv_ev_state(state,fd); switch (state->type) { case ERTS_EV_TYPE_NIF: @@ -1243,7 +941,9 @@ ERTS_CIO_EXPORT(enif_select)(ErlNifEnv* env, } ASSERT(state->type == ERTS_EV_TYPE_NONE); break; - }} + } + default: break; + } ASSERT((state->type == ERTS_EV_TYPE_NIF) || (state->type == ERTS_EV_TYPE_NONE && !state->events)); @@ -1253,20 +953,22 @@ ERTS_CIO_EXPORT(enif_select)(ErlNifEnv* env, if (on) { ctl_events &= ~old_events; state->events |= ctl_events; + state->active_events |= ctl_events; + if (state->type == ERTS_EV_TYPE_NONE) + ctl_op = ERTS_POLL_OP_ADD; } else { ctl_events &= old_events; state->events &= ~ctl_events; + state->active_events &= ~ctl_events; } - if (ctl_events) { + if (ctl_events || ctl_op == ERTS_POLL_OP_DEL) { ErtsPollEvents new_events; - if (!state->pollset) { - state->pollset = erts_get_scheduler_data()->pollset; - } - - new_events = ERTS_CIO_POLL_CTL(state->pollset->ps, state->fd, ctl_events, on, &wake_poller); + new_events = erts_io_control_wakeup(state, ctl_op, + state->active_events, + &wake_poller); if (new_events & (ERTS_POLL_EV_ERR|ERTS_POLL_EV_NVAL)) { if (state->type == ERTS_EV_TYPE_NIF && !old_events) { @@ -1274,8 +976,6 @@ ERTS_CIO_EXPORT(enif_select)(ErlNifEnv* env, state->flags = 0; state->driver.nif->in.pid = NIL; state->driver.nif->out.pid = NIL; - state->driver.nif->in.ddeselect_cnt = 0; - state->driver.nif->out.ddeselect_cnt = 0; state->driver.stop.resource = NULL; } ret = INT_MIN | ERL_NIF_SELECT_FAILED; @@ -1307,11 +1007,9 @@ ERTS_CIO_EXPORT(enif_select)(ErlNifEnv* env, ASSERT(is_internal_ref(ref)); refn = internal_ref_numbers(ref); state->driver.nif->in.immed = THE_NON_VALUE; - state->driver.nif->in.refn[0] = refn[0]; - state->driver.nif->in.refn[1] = refn[1]; - state->driver.nif->in.refn[2] = refn[2]; + sys_memcpy(state->driver.nif->in.refn, refn, + sizeof(state->driver.nif->in.refn)); } - state->driver.nif->in.ddeselect_cnt = 0; } if (mode & ERL_DRV_WRITE) { state->driver.nif->out.pid = recipient; @@ -1321,11 +1019,9 @@ ERTS_CIO_EXPORT(enif_select)(ErlNifEnv* env, ASSERT(is_internal_ref(ref)); refn = internal_ref_numbers(ref); state->driver.nif->out.immed = THE_NON_VALUE; - state->driver.nif->out.refn[0] = refn[0]; - state->driver.nif->out.refn[1] = refn[1]; - state->driver.nif->out.refn[2] = refn[2]; + sys_memcpy(state->driver.nif->out.refn, refn, + sizeof(state->driver.nif->out.refn)); } - state->driver.nif->out.ddeselect_cnt = 0; } ret = 0; } @@ -1333,14 +1029,9 @@ ERTS_CIO_EXPORT(enif_select)(ErlNifEnv* env, if (state->type == ERTS_EV_TYPE_NIF) { state->driver.nif->in.pid = NIL; state->driver.nif->out.pid = NIL; - state->driver.nif->in.ddeselect_cnt = 0; - state->driver.nif->out.ddeselect_cnt = 0; - if (old_events != 0) { - remember_removed(state); - } } ASSERT(state->events==0); - if (state->remove_cnt == 0 || !wake_poller) { + if (!wake_poller) { /* * Safe to close fd now as it is not in pollset * or there was no need to eject fd (kernel poll) @@ -1455,7 +1146,7 @@ print_driver_name(erts_dsprintf_buf_t *dsbufp, Eterm id) static void steal(erts_dsprintf_buf_t *dsbufp, ErtsDrvEventState *state, int mode) { - erts_dsprintf(dsbufp, "stealing control of fd=%d from ", (int) GET_FD(state->fd)); + erts_dsprintf(dsbufp, "stealing control of fd=%d from ", (int) state->fd); switch (state->type) { case ERTS_EV_TYPE_DRV_SEL: { int deselect_mode = 0; @@ -1479,7 +1170,7 @@ steal(erts_dsprintf_buf_t *dsbufp, ErtsDrvEventState *state, int mode) if (deselect_mode) deselect(state, deselect_mode); else { - erts_dsprintf(dsbufp, "no one", (int) GET_FD(state->fd)); + erts_dsprintf(dsbufp, "no one", (int) state->fd); ASSERT(0); } erts_dsprintf(dsbufp, "\n"); @@ -1510,7 +1201,7 @@ steal(erts_dsprintf_buf_t *dsbufp, ErtsDrvEventState *state, int mode) break; } default: - erts_dsprintf(dsbufp, "no one\n", (int) GET_FD(state->fd)); + erts_dsprintf(dsbufp, "no one\n", (int) state->fd); ASSERT(0); } } @@ -1524,7 +1215,7 @@ print_drv_select_op(erts_dsprintf_buf_t *dsbufp, "driver_select(%p, %d,%s%s%s%s, %d) " "by ", ix, - (int) GET_FD(fd), + (int) fd, mode & ERL_DRV_READ ? " ERL_DRV_READ" : "", mode & ERL_DRV_WRITE ? " ERL_DRV_WRITE" : "", mode & ERL_DRV_USE ? " ERL_DRV_USE" : "", @@ -1541,7 +1232,7 @@ print_nif_select_op(erts_dsprintf_buf_t *dsbufp, { erts_dsprintf(dsbufp, "enif_select(_, %d,%s%s%s, %T:%T, %T) ", - (int) GET_FD(fd), + (int) fd, mode & ERL_NIF_SELECT_READ ? " READ" : "", mode & ERL_NIF_SELECT_WRITE ? " WRITE" : "", mode & ERL_NIF_SELECT_STOP ? " STOP" : "", @@ -1673,7 +1364,7 @@ steal_pending_stop_nif(erts_dsprintf_buf_t *dsbufp, ErtsResource* resource, erts_dsprintf(dsbufp, "called before stop was called for NIF resource %T:%T\n", rt->module, rt->name); - enif_release_resource(state->driver.stop.resource); + enif_release_resource(state->driver.stop.resource->data); state->type = ERTS_EV_TYPE_NONE; state->flags = 0; state->driver.stop.resource = NULL; @@ -1687,8 +1378,7 @@ steal_pending_stop_nif(erts_dsprintf_buf_t *dsbufp, ErtsResource* resource, static ERTS_INLINE int io_task_schedule_allowed(ErtsDrvEventState *state, - ErtsPortTaskType type, - erts_aint_t current_cio_time) + ErtsPortTaskType type) { ErtsIoTask *io_task; @@ -1708,42 +1398,41 @@ io_task_schedule_allowed(ErtsDrvEventState *state, return 0; } - return !is_iotask_active(io_task, current_cio_time); + return !is_iotask_active(io_task); } static ERTS_INLINE void -iready(Eterm id, ErtsDrvEventState *state, erts_aint_t current_cio_time) +iready(Eterm id, ErtsDrvEventState *state) { if (io_task_schedule_allowed(state, - ERTS_PORT_TASK_INPUT, - current_cio_time)) { + ERTS_PORT_TASK_INPUT)) { ErtsIoTask *iotask = &state->driver.select->iniotask; - erts_atomic_set_nob(&iotask->executed_time, current_cio_time); if (erts_port_task_schedule(id, &iotask->task, ERTS_PORT_TASK_INPUT, (ErlDrvEvent) state->fd) != 0) { stale_drv_select(id, state, ERL_DRV_READ); - } - add_active_fd(state->pollset, state->fd); + } else { + DEBUG_PRINT_FD("schedule ready_input(%T, %d)", + state, id, state->fd); + } } } static ERTS_INLINE void -oready(Eterm id, ErtsDrvEventState *state, erts_aint_t current_cio_time) +oready(Eterm id, ErtsDrvEventState *state) { if (io_task_schedule_allowed(state, - ERTS_PORT_TASK_OUTPUT, - current_cio_time)) { + ERTS_PORT_TASK_OUTPUT)) { ErtsIoTask *iotask = &state->driver.select->outiotask; - erts_atomic_set_nob(&iotask->executed_time, current_cio_time); if (erts_port_task_schedule(id, &iotask->task, ERTS_PORT_TASK_OUTPUT, (ErlDrvEvent) state->fd) != 0) { stale_drv_select(id, state, ERL_DRV_WRITE); - } - add_active_fd(state->pollset, state->fd); + } else { + DEBUG_PRINT_FD("schedule ready_output(%T, %d)", state, id, state->fd); + } } } @@ -1798,91 +1487,56 @@ send_event_tuple(struct erts_nif_select_event* e, ErtsResource* resource, static void bad_fd_in_pollset(ErtsDrvEventState *, Eterm inport, Eterm outport); void -ERTS_CIO_EXPORT(erts_check_io_interrupt)(struct pollset_info *psi, int set) +erts_check_io_interrupt(ErtsPollThread *psi, int set) { - ERTS_CIO_POLL_INTR(psi->ps, set); + if (psi) { +#if ERTS_POLL_USE_FALLBACK + if (psi->ps == get_fallback()) { + erts_poll_interrupt_flbk(psi->ps, set); + return; + } +#endif + erts_poll_interrupt(psi->ps, set); + } } -void -ERTS_CIO_EXPORT(erts_check_io_interrupt_timed)(struct pollset_info *psi, - int set, - ErtsMonotonicTime timeout_time) -{ - ERTS_CIO_POLL_INTR_TMD(psi->ps, set, timeout_time); +ErtsPollThread * +erts_create_pollset_thread(int id) { + return psiv+id; } -#ifndef __WIN32__ -/* - * Number of ignored events, for a lingering fd added by enif_select(), - * until we deselect fd-event from pollset. - */ -# define ERTS_NIF_DELAYED_DESELECT 20 -#else -/* Disable delayed deselect as pollset cannot handle active events */ -# define ERTS_NIF_DELAYED_DESELECT 1 -#endif - void -ERTS_CIO_EXPORT(erts_check_io)(int do_wait) +erts_check_io(ErtsPollThread *psi) { - ErtsPollResFd *pollres; int pollres_len; - ErtsMonotonicTime timeout_time; int poll_ret, i; - erts_aint_t current_cio_time; - ErtsSchedulerData *esdp = erts_get_scheduler_data(); - struct pollset_info *psi = esdp->pollset; + ERTS_MSACC_PUSH_AND_SET_STATE(ERTS_MSACC_STATE_CHECK_IO); restart: -#ifdef ERTS_BREAK_REQUESTED - if (ERTS_BREAK_REQUESTED) - erts_do_break_handling(); -#endif - - /* Figure out timeout value */ - timeout_time = (do_wait - ? erts_check_next_timeout_time(esdp) - : ERTS_POLL_NO_TIMEOUT /* poll only */); - - /* - * No need for an atomic inc op when incrementing - * erts_check_io_time, since only one thread can - * check io at a time. - */ - current_cio_time = erts_atomic_read_dirty(&psi->check_io_time); - current_cio_time++; - erts_atomic_set_relb(&psi->check_io_time, current_cio_time); - - check_cleanup_active_fds(psi, - current_cio_time, - timeout_time != ERTS_POLL_NO_TIMEOUT); - #ifdef ERTS_ENABLE_LOCK_CHECK erts_lc_check_exact(NULL, 0); /* No locks should be locked */ #endif - pollres_len = erts_atomic32_read_dirty(&psi->active_fd.no) + ERTS_CHECK_IO_POLL_RES_LEN; + pollres_len = psi->pollres_len; - pollres = erts_alloc(ERTS_ALC_T_TMP, sizeof(ErtsPollResFd)*pollres_len); +#if ERTS_POLL_USE_FALLBACK + if (psi->ps == get_fallback()) { - erts_atomic_set_nob(&psi->in_poll_wait, 1); + poll_ret = erts_poll_wait_flbk(psi->ps, psi->pollres, &pollres_len); - poll_ret = ERTS_CIO_POLL_WAIT(psi->ps, pollres, &pollres_len, timeout_time); + } else +#endif + { + poll_ret = erts_poll_wait(psi->ps, psi->pollres, &pollres_len); + } #ifdef ERTS_ENABLE_LOCK_CHECK erts_lc_check_exact(NULL, 0); /* No locks should be locked */ #endif -#ifdef ERTS_BREAK_REQUESTED - if (ERTS_BREAK_REQUESTED) - erts_do_break_handling(); -#endif - if (poll_ret != 0) { - erts_atomic_set_nob(&psi->in_poll_wait, 0); - forget_removed(psi); - erts_free(ERTS_ALC_T_TMP, pollres); + if (poll_ret == EAGAIN) { goto restart; } @@ -1898,64 +1552,78 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) erl_errno_id(poll_ret), poll_ret); erts_send_error_to_logger_nogl(dsbufp); } + ERTS_MSACC_POP_STATE(); return; } for (i = 0; i < pollres_len; i++) { - ErtsSysFdType fd = (ErtsSysFdType) pollres[i].fd; + erts_driver_t* drv_ptr = NULL; + ErtsResource* resource = NULL; + ErtsDrvSelectDataState *free_select = NULL; + ErtsNifSelectDataState *free_nif = NULL; + ErtsSysFdType fd = (ErtsSysFdType) ERTS_POLL_RES_GET_FD(&psi->pollres[i]); ErtsDrvEventState *state; + ErtsPollEvents revents; erts_mtx_lock(fd_mtx(fd)); + state = get_drv_ev_state(fd); -#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - state = &drv_ev_state.v[ (int) fd]; -#else - state = hash_get_drv_ev_state(fd); if (!state) { - goto next_pollres; + erts_mtx_unlock(fd_mtx(fd)); + continue; } -#endif - /* Skip this fd if it was removed from pollset */ - if (is_removed(state) || state->pollset != psi) { - goto next_pollres; - } + revents = ERTS_POLL_RES_GET_EVTS(&psi->pollres[i]); + + DEBUG_PRINT_FD("triggered %s", state, ev2str(revents)); + + if (revents & ERTS_POLL_EV_ERR) { + /* + * Handle error events by triggering all in/out events + * that has been selected on. + * We *do not* want to call a callback that corresponds + * to an event not selected. + */ + revents = state->active_events; + state->active_events = 0; + } else { + + /* Disregard any events that are not active at the moment, + for instance this could happen if the driver/nif does + select/deselect in rapid succession. */ + revents &= state->active_events | ERTS_POLL_EV_NVAL; + state->active_events &= ~revents; + + /* Reactivate the poll op if there are still active events */ + if (state->active_events) { + DEBUG_PRINT_FD("re-enable %s", state, ev2str(state->active_events)); + erts_io_control(state, ERTS_POLL_OP_MOD, state->active_events); + } + } switch (state->type) { case ERTS_EV_TYPE_DRV_SEL: { /* Requested via driver_select()... */ - ErtsPollEvents revents = pollres[i].events; - - if (revents & ERTS_POLL_EV_ERR) { - /* - * Handle error events by triggering all in/out events - * that the driver has selected. - * We *do not* want to call a callback that corresponds - * to an event not selected. - */ - revents = state->events; - } - else { - revents &= (state->events | ERTS_POLL_EV_NVAL); - } if (revents & (ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT)) { if (revents & ERTS_POLL_EV_OUT) { - oready(state->driver.select->outport, state, current_cio_time); + oready(state->driver.select->outport, state); } /* Someone might have deselected input since revents - was read therefore, update revents... */ - revents &= state->events; + was read (true also on the non-smp emulator since + oready() may have been called); therefore, update + revents... */ + revents &= state->events; if (revents & ERTS_POLL_EV_IN) { - iready(state->driver.select->inport, state, current_cio_time); + iready(state->driver.select->inport, state); } } else if (revents & ERTS_POLL_EV_NVAL) { bad_fd_in_pollset(state, state->driver.select->inport, state->driver.select->outport); - add_active_fd(psi, state->fd); + check_fd_cleanup(state, &free_select, &free_nif); } break; } @@ -1964,89 +1632,116 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait) struct erts_nif_select_event in = {NIL}; struct erts_nif_select_event out = {NIL}; ErtsResource* resource = NULL; - ErtsPollEvents revents = pollres[i].events; - - if (revents & ERTS_POLL_EV_ERR) { - /* - * Handle error events by triggering all in/out events - * that the NIF has selected. - * We *do not* want to send a message that corresponds - * to an event not selected. - */ - revents = state->events; - } - else { - revents &= (state->events | ERTS_POLL_EV_NVAL); - } if (revents & (ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT)) { if (revents & ERTS_POLL_EV_OUT) { if (is_not_nil(state->driver.nif->out.pid)) { out = state->driver.nif->out; resource = state->driver.stop.resource; - state->driver.nif->out.ddeselect_cnt = ERTS_NIF_DELAYED_DESELECT; state->driver.nif->out.pid = NIL; - add_active_fd(psi, state->fd); - } - else { - ASSERT(state->driver.nif->out.ddeselect_cnt >= 2); - state->driver.nif->out.ddeselect_cnt--; } } if (revents & ERTS_POLL_EV_IN) { if (is_not_nil(state->driver.nif->in.pid)) { in = state->driver.nif->in; resource = state->driver.stop.resource; - state->driver.nif->in.ddeselect_cnt = ERTS_NIF_DELAYED_DESELECT; state->driver.nif->in.pid = NIL; - add_active_fd(psi, state->fd); - } - else { - ASSERT(state->driver.nif->in.ddeselect_cnt >= 2); - state->driver.nif->in.ddeselect_cnt--; } } + state->events &= ~revents; } else if (revents & ERTS_POLL_EV_NVAL) { bad_fd_in_pollset(state, NIL, NIL); - add_active_fd(psi, state->fd); + check_fd_cleanup(state, &free_select, &free_nif); } erts_mtx_unlock(fd_mtx(fd)); + if (is_not_nil(in.pid)) { send_event_tuple(&in, resource, am_ready_input); } if (is_not_nil(out.pid)) { send_event_tuple(&out, resource, am_ready_output); } - goto next_pollres_unlocked; + continue; } + case ERTS_EV_TYPE_STOP_NIF: { + resource = state->driver.stop.resource; + state->type = ERTS_EV_TYPE_NONE; + goto case_ERTS_EV_TYPE_NONE; + } + + case ERTS_EV_TYPE_STOP_USE: { +#if ERTS_POLL_USE_FALLBACK + ASSERT(psi->ps == get_fallback()); +#endif + drv_ptr = state->driver.stop.drv_ptr; + state->type = ERTS_EV_TYPE_NONE; + /* fallthrough */ case ERTS_EV_TYPE_NONE: /* Deselected ... */ + case_ERTS_EV_TYPE_NONE: + ASSERT(!state->events && !state->active_events && !state->flags); + check_fd_cleanup(state, &free_select, &free_nif); break; + } default: { /* Error */ erts_dsprintf_buf_t *dsbufp; dsbufp = erts_create_logger_dsbuf(); erts_dsprintf(dsbufp, "Invalid event request type for fd in erts_poll()! " - "fd=%d, event request type=%sd\n", (int) state->fd, + "fd=%d, event request type=%d\n", (int) state->fd, (int) state->type); ASSERT(0); deselect(state, 0); - add_active_fd(psi, state->fd); break; } } - next_pollres:; erts_mtx_unlock(fd_mtx(fd)); - next_pollres_unlocked:; - } - erts_atomic_set_nob(&psi->in_poll_wait, 0); - erts_free(ERTS_ALC_T_TMP, pollres); - forget_removed(psi); + if (drv_ptr) { + int was_unmasked = erts_block_fpe(); + DTRACE1(driver_stop_select, drv_ptr->name); + LTTNG1(driver_stop_select, drv_ptr->name); + (*drv_ptr->stop_select)((ErlDrvEvent) fd, NULL); + erts_unblock_fpe(was_unmasked); + if (drv_ptr->handle) { + erts_ddll_dereference_driver(drv_ptr->handle); + } + } + if (resource) { + erts_resource_stop(resource, (ErlNifEvent)fd, 1); + enif_release_resource(resource->data); + } + if (free_select) + free_drv_select_data(free_select); + if (free_nif) + free_nif_select_data(free_nif); + } + + /* The entire pollres array was filled with events, + * grow it for the next call. We do this for two reasons: + * 1. Pulling out more events in on go will increase throughput + * 2. If the polling implementation is not fair, this will make + * sure that we get all fds that we can. i.e. if 12 fds are + * constantly active, but we only have a pollres_len of 10, + * two of the fds may never be triggered depending on what the + * kernel decides to do. + **/ + if (pollres_len == psi->pollres_len) { + int ev_state_len = drv_ev_state_len(); + erts_free(ERTS_ALC_T_POLLSET, psi->pollres); + psi->pollres_len *= 2; + /* Never grow it larger than the current drv_ev_state.len size */ + if (psi->pollres_len > ev_state_len) + psi->pollres_len = ev_state_len; + psi->pollres = erts_alloc(ERTS_ALC_T_POLLSET, + sizeof(ErtsPollResFd) * psi->pollres_len); + } + + ERTS_MSACC_POP_STATE(); } static void @@ -2167,98 +1862,189 @@ static void drv_ev_state_free(void *des) } #endif -#ifdef ERTS_ENABLE_KERNEL_POLL +#define ERTS_MAX_NO_OF_POLL_THREADS ERTS_MAX_NO_OF_SCHEDULERS -struct io_functions { - int (*select)(ErlDrvPort, ErlDrvEvent, int, int); - int (*enif_select)(ErlNifEnv*, ErlNifEvent, enum ErlNifSelectFlags, void*, const ErlNifPid*, Eterm); - void (*check_io_as_interrupt)(struct pollset_info*); - void (*check_io_interrupt)(struct pollset_info*, int); - void (*check_io_interrupt_tmd)(struct pollset_info*, int, ErtsMonotonicTime); - void (*check_io)(int); - struct pollset_info* (*get_pollset)(int sched_nr); - void (*notify_port_task_executed)(ErtsPortTaskHandle *pthp); - int (*max_files)(void); - Uint (*size)(void); - Eterm (*info)(void *); - int (*check_io_debug)(ErtsCheckIoDebugInfo *); -#ifdef ERTS_ENABLE_LOCK_COUNT - void (*lcnt_update_cio_locks)(int enable); -#endif -}; +static char * +get_arg(char* rest, char** argv, int* ip) +{ + int i = *ip; + if (*rest == '\0') { + if (argv[i+1] == NULL) { + erts_fprintf(stderr, "too few arguments\n"); + erts_usage(); + } + argv[i++] = NULL; + rest = argv[i]; + } + argv[i] = NULL; + *ip = i; + return rest; +} -# ifdef ERTS_KERNEL_POLL_VERSION -struct io_functions erts_io_funcs = {0}; -# else -extern struct io_functions erts_io_funcs; -# endif +static void +parse_args(int *argc, char **argv, int concurrent_waiters) +{ + int i = 0, j; + int no_pollsets = 0, no_poll_threads = 0, + no_pollsets_percentage = 0, + no_poll_threads_percentage = 0; + ASSERT(argc && argv); + while (i < *argc) { + if(argv[i][0] == '-') { + switch (argv[i][1]) { + case 'I': { + if (strncmp(argv[i]+2, "Ot", 2) == 0) { + char *arg = get_arg(argv[i]+4, argv, &i); + if (sscanf(arg, "%d", &no_poll_threads) != 1 || + no_poll_threads < 1 || + ERTS_MAX_NO_OF_POLL_THREADS < no_poll_threads) { + erts_fprintf(stderr,"bad I/O poll threads number: %s\n", arg); + erts_usage(); + } + } else if (strncmp(argv[i]+2, "Op", 3) == 0) { + char *arg = get_arg(argv[i]+4, argv, &i); + if (sscanf(arg, "%d", &no_pollsets) != 1 || + no_pollsets < 1) { + erts_fprintf(stderr,"bad I/O pollset number: %s\n", arg); + erts_usage(); + } + } else if (strncmp(argv[i]+2, "OPt", 4) == 0) { + char *arg = get_arg(argv[i]+5, argv, &i); + if (sscanf(arg, "%d", &no_poll_threads_percentage) != 1 || + no_poll_threads_percentage < 0 || + no_poll_threads_percentage > 100) { + erts_fprintf(stderr,"bad I/O poll thread percentage number: %s\n", arg); + erts_usage(); + } + } else if (strncmp(argv[i]+2, "OPp", 4) == 0) { + char *arg = get_arg(argv[i]+5, argv, &i); + if (sscanf(arg, "%d", &no_pollsets_percentage) != 1 || + no_pollsets_percentage < 0 || + no_pollsets_percentage > 100) { + erts_fprintf(stderr,"bad I/O pollset percentage number: %s\n", arg); + erts_usage(); + } + } else { + break; + } + break; + } + case 'K': + (void)get_arg(argv[i]+2, argv, &i); + break; + case '-': + goto args_parsed; + default: + break; + } + } + i++; + } + +args_parsed: -#endif /* ERTS_ENABLE_KERNEL_POLL */ + if (!concurrent_waiters) { + no_pollsets = no_poll_threads; + no_pollsets_percentage = 100; + } + + if (no_poll_threads == 0) { + if (no_poll_threads_percentage == 0) + no_poll_threads = 1; /* This is the default */ + else { + no_poll_threads = erts_no_schedulers * no_poll_threads_percentage / 100; + if (no_poll_threads < 1) + no_poll_threads = 1; + } + } + + if (no_pollsets == 0) { + if (no_pollsets_percentage == 0) + no_pollsets = 1; /* This is the default */ + else { + no_pollsets = no_poll_threads * no_pollsets_percentage / 100; + if (no_pollsets < 1) + no_pollsets = 1; + } + } + + if (no_poll_threads < no_pollsets) { + erts_fprintf(stderr, + "number of IO poll threads has to be greater or equal to " + "the number of \nIO pollsets. Current values are set to: \n" + " -IOt %d -IOp %d\n", + no_poll_threads, no_pollsets); + erts_usage(); + } + + /* Handled arguments have been marked with NULL. Slide arguments + not handled towards the beginning of argv. */ + for (i = 0, j = 0; i < *argc; i++) { + if (argv[i]) + argv[j++] = argv[i]; + } + *argc = j; + + erts_no_pollsets = no_pollsets; + erts_no_poll_threads = no_poll_threads; +} void -ERTS_CIO_EXPORT(erts_init_check_io)(void) +erts_init_check_io(int *argc, char **argv) { - int j; + int j, concurrent_waiters, no_poll_threads; ERTS_CT_ASSERT((INT_MIN & (ERL_NIF_SELECT_STOP_CALLED | ERL_NIF_SELECT_STOP_SCHEDULED | ERL_NIF_SELECT_INVALID_EVENT | ERL_NIF_SELECT_FAILED)) == 0); -#ifdef ERTS_ENABLE_KERNEL_POLL - ASSERT(erts_io_funcs.select == NULL); - erts_io_funcs.select = ERTS_CIO_EXPORT(driver_select); - erts_io_funcs.enif_select = ERTS_CIO_EXPORT(enif_select); - erts_io_funcs.check_io_interrupt = ERTS_CIO_EXPORT(erts_check_io_interrupt); - erts_io_funcs.check_io_interrupt_tmd= ERTS_CIO_EXPORT(erts_check_io_interrupt_timed); - erts_io_funcs.check_io = ERTS_CIO_EXPORT(erts_check_io); - erts_io_funcs.get_pollset = ERTS_CIO_EXPORT(erts_get_pollset); - erts_io_funcs.notify_port_task_executed = ERTS_CIO_EXPORT(erts_io_notify_port_task_executed); - erts_io_funcs.max_files = ERTS_CIO_EXPORT(erts_check_io_max_files); - erts_io_funcs.size = ERTS_CIO_EXPORT(erts_check_io_size); - erts_io_funcs.info = ERTS_CIO_EXPORT(erts_check_io_info); - erts_io_funcs.check_io_debug = ERTS_CIO_EXPORT(erts_check_io_debug); -#ifdef ERTS_ENABLE_LOCK_COUNT - erts_io_funcs.lcnt_update_cio_locks = ERTS_CIO_EXPORT(erts_lcnt_update_cio_locks); -#endif -#endif - - init_removed_fd_alloc(); - - ERTS_CIO_POLL_INIT(); - pollsetv = erts_alloc(ERTS_ALC_T_POLLSET, - sizeof(struct pollset_info)*erts_no_schedulers); - for (j=0; j < erts_no_schedulers; j++) { - struct pollset_info* psi = &pollsetv[j]; - - erts_atomic_init_nob(&psi->check_io_time, 0); - erts_atomic_init_nob(&psi->in_poll_wait, 0); - psi->ps = ERTS_CIO_NEW_POLLSET(); - psi->active_fd.six = 0; - psi->active_fd.eix = 0; - erts_atomic32_init_nob(&psi->active_fd.no, 0); - psi->active_fd.size = ERTS_ACTIVE_FD_INC; - psi->active_fd.array = erts_alloc(ERTS_ALC_T_ACTIVE_FD_ARR, - sizeof(ErtsSysFdType)*ERTS_ACTIVE_FD_INC); -#ifdef DEBUG - { - int i; - for (i = 0; i < ERTS_ACTIVE_FD_INC; i++) - psi->active_fd.array[i] = ERTS_SYS_FD_INVALID; - } + + erts_poll_init(&concurrent_waiters); +#if ERTS_POLL_USE_FALLBACK + erts_poll_init_flbk(NULL); +#endif + + parse_args(argc, argv, concurrent_waiters); + + /* Create the actual pollsets */ + pollsetv = erts_alloc(ERTS_ALC_T_POLLSET,sizeof(ErtsPollSet *) * erts_no_pollsets); + + for (j=0; j < erts_no_pollsets; j++) + pollsetv[j] = erts_poll_create_pollset(j); + +#if ERTS_POLL_USE_FALLBACK + flbk_pollset = erts_poll_create_pollset_flbk(-1); +#endif + + no_poll_threads = erts_no_poll_threads; +#if ERTS_POLL_USE_FALLBACK + no_poll_threads++; +#endif + + psiv = erts_alloc(ERTS_ALC_T_POLLSET, sizeof(ErtsPollThread) * no_poll_threads); + +#if ERTS_POLL_USE_FALLBACK + psiv[0].pollres_len = ERTS_CHECK_IO_POLL_RES_LEN; + psiv[0].pollres = erts_alloc(ERTS_ALC_T_POLLSET, + sizeof(ErtsPollResFd) * ERTS_CHECK_IO_POLL_RES_LEN); + psiv[0].ps = get_fallback(); + psiv++; #endif - erts_atomic_init_nob(&psi->removed_list, (erts_aint_t)NULL); + for (j = 0; j < erts_no_poll_threads; j++) { + psiv[j].pollres_len = ERTS_CHECK_IO_POLL_RES_LEN; + psiv[j].pollres = erts_alloc(ERTS_ALC_T_POLLSET, + sizeof(ErtsPollResFd) * ERTS_CHECK_IO_POLL_RES_LEN); + psiv[j].ps = pollsetv[j % erts_no_pollsets]; } - { - int i; - for (i=0; icheck_io_time); + piv = erts_alloc(ERTS_ALC_T_TMP, sizeof(ErtsPollInfo) * no_pollsets); - piv[j].active_fds = (int) erts_atomic32_read_acqb(&psi->active_fd.no); - while (1) { - erts_aint_t post_cio_time; - int post_active_fds; +#if ERTS_POLL_USE_FALLBACK + erts_poll_info_flbk(get_fallback(), &piv[0]); + piv[0].poll_threads = 1; + piv[0].active_fds = 0; + piv++; +#endif - ERTS_CIO_POLL_INFO(psi->ps, &piv[j]); + for (j = 0; j < erts_no_pollsets; j++) { + erts_poll_info(pollsetv[j], &piv[j]); + piv[j].active_fds = 0; + piv[j].poll_threads = erts_no_poll_threads / erts_no_pollsets; + if (erts_no_poll_threads % erts_no_pollsets > j) + piv[j].poll_threads++; + } - post_cio_time = erts_atomic_read_mb(&psi->check_io_time); - post_active_fds = (int) erts_atomic32_read_acqb(&psi->active_fd.no); - if (cio_time == post_cio_time && piv[j].active_fds == post_active_fds) - break; - cio_time = post_cio_time; - piv[j].active_fds = post_active_fds; +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + i = 0; + erts_mtx_lock(&drv_ev_state.grow_lock); + len = erts_atomic_read_nob(&drv_ev_state.len); + for (i = 0; i < ERTS_CHECK_IO_DRV_EV_STATE_LOCK_CNT; i++) { + erts_mtx_lock(&drv_ev_state.locks[i].lck); + for (j = i; j < len; j+=ERTS_CHECK_IO_DRV_EV_STATE_LOCK_CNT) { + ErtsDrvEventState *state = get_drv_ev_state(j); + int pollsetid = get_pollset_id(j); + ASSERT(fd_mtx(j) == &drv_ev_state.locks[i].lck); + if (state->flags & ERTS_EV_FLAG_FALLBACK) + pollsetid = -1; + if (state->driver.select + && (state->type == ERTS_EV_TYPE_DRV_SEL) + && (is_iotask_active(&state->driver.select->iniotask) + || is_iotask_active(&state->driver.select->outiotask))) + piv[pollsetid].active_fds++; } + erts_mtx_unlock(&drv_ev_state.locks[i].lck); + } + erts_mtx_unlock(&drv_ev_state.grow_lock); - #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - piv[j].memory_size += sizeof(ErtsDrvEventState) * erts_atomic_read_nob(&drv_ev_state.len); - #else - piv[j].memory_size += safe_hash_table_sz(&drv_ev_state.tab); - { - SafeHashInfo hi; - safe_hash_get_info(&hi, &drv_ev_state.tab); - piv[j].memory_size += hi.objs * sizeof(ErtsDrvEventState); - } - erts_spin_lock(&drv_ev_state.prealloc_lock); - piv[j].memory_size += drv_ev_state.num_prealloc * sizeof(ErtsDrvEventState); - erts_spin_unlock(&drv_ev_state.prealloc_lock); - #endif + piv[0].memory_size += sizeof(ErtsDrvEventState) * erts_atomic_read_nob(&drv_ev_state.len); +#else + piv[0].memory_size += safe_hash_table_sz(&drv_ev_state.tab); + { + SafeHashInfo hi; + safe_hash_get_info(&hi, &drv_ev_state.tab); + piv[0].memory_size += hi.objs * sizeof(ErtsDrvEventState); } + erts_spin_lock(&drv_ev_state.prealloc_lock); + piv[0].memory_size += drv_ev_state.num_prealloc * sizeof(ErtsDrvEventState); + erts_spin_unlock(&drv_ev_state.prealloc_lock); +#endif hpp = NULL; szp = &sz; sz = 0; + piv -= ERTS_POLL_USE_FALLBACK; + bld_it: - for (j = erts_no_schedulers-1; j >= 0; j--) { + for (j = no_pollsets-1; j >= 0; j--) { i = 0; tags[i] = erts_bld_atom(hpp, szp, "name"); @@ -2378,9 +2188,6 @@ ERTS_CIO_EXPORT(erts_check_io_info)(void *proc) tags[i] = erts_bld_atom(hpp, szp, "primary"); values[i++] = erts_bld_atom(hpp, szp, piv[j].primary); - tags[i] = erts_bld_atom(hpp, szp, "fallback"); - values[i++] = erts_bld_atom(hpp, szp, piv[j].fallback ? piv[j].fallback : "false"); - tags[i] = erts_bld_atom(hpp, szp, "kernel_poll"); values[i++] = erts_bld_atom(hpp, szp, piv[j].kernel_poll ? piv[j].kernel_poll : "false"); @@ -2389,20 +2196,13 @@ ERTS_CIO_EXPORT(erts_check_io_info)(void *proc) values[i++] = erts_bld_uint(hpp, szp, piv[j].memory_size); tags[i] = erts_bld_atom(hpp, szp, "total_poll_set_size"); - values[i++] = erts_bld_uint(hpp, szp, (Uint) piv[j].poll_set_size); - - if (piv[j].fallback) { - tags[i] = erts_bld_atom(hpp, szp, "fallback_poll_set_size"); - values[i++] = erts_bld_uint(hpp, szp, (Uint) piv[j].fallback_poll_set_size); - } + values[i++] = erts_bld_uint(hpp, szp, piv[j].poll_set_size); tags[i] = erts_bld_atom(hpp, szp, "lazy_updates"); values[i++] = piv[j].lazy_updates ? am_true : am_false; - if (piv[j].lazy_updates) { - tags[i] = erts_bld_atom(hpp, szp, "pending_updates"); - values[i++] = erts_bld_uint(hpp, szp, (Uint) piv[j].pending_updates); - } + tags[i] = erts_bld_atom(hpp, szp, "pending_updates"); + values[i++] = erts_bld_uint(hpp, szp, piv[j].pending_updates); tags[i] = erts_bld_atom(hpp, szp, "batch_updates"); values[i++] = piv[j].batch_updates ? am_true : am_false; @@ -2410,22 +2210,17 @@ ERTS_CIO_EXPORT(erts_check_io_info)(void *proc) tags[i] = erts_bld_atom(hpp, szp, "concurrent_updates"); values[i++] = piv[j].concurrent_updates ? am_true : am_false; + tags[i] = erts_bld_atom(hpp, szp, "fallback"); + values[i++] = piv[j].is_fallback ? am_true : am_false; + tags[i] = erts_bld_atom(hpp, szp, "max_fds"); - values[i++] = erts_bld_uint(hpp, szp, (Uint) piv[j].max_fds); + values[i++] = erts_bld_uint(hpp, szp, piv[j].max_fds); tags[i] = erts_bld_atom(hpp, szp, "active_fds"); - values[i++] = erts_bld_uint(hpp, szp, (Uint) piv[j].active_fds); - - #ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS - tags[i] = erts_bld_atom(hpp, szp, "no_avoided_wakeups"); - values[i++] = erts_bld_uint(hpp, szp, (Uint) piv[j].no_avoided_wakeups); - - tags[i] = erts_bld_atom(hpp, szp, "no_avoided_interrupts"); - values[i++] = erts_bld_uint(hpp, szp, (Uint) piv[j].no_avoided_interrupts); + values[i++] = erts_bld_uint(hpp, szp, piv[j].active_fds); - tags[i] = erts_bld_atom(hpp, szp, "no_interrupt_timed"); - values[i++] = erts_bld_uint(hpp, szp, (Uint) piv[j].no_interrupt_timed); - #endif + tags[i] = erts_bld_atom(hpp, szp, "poll_threads"); + values[i++] = erts_bld_uint(hpp, szp, piv[j].poll_threads); res = erts_bld_2tup_list(hpp, szp, i, tags, values); @@ -2454,6 +2249,10 @@ static ERTS_INLINE ErtsPollEvents print_events(ErtsPollEvents ev) { int first = 1; + if(ev == ERTS_POLL_EV_NONE) { + erts_printf("N/A"); + return 0; + } if(ev & ERTS_POLL_EV_IN) { ev &= ~ERTS_POLL_EV_IN; erts_printf("%s%s", first ? "" : "|", "IN"); @@ -2486,15 +2285,40 @@ print_flags(EventStateFlags f) erts_printf("%s","USED"); delim = "|"; } - if(f & ERTS_EV_FLAG_DEFER_IN_EV) { - erts_printf("%s%s", delim, "DRIN"); + if(f & ERTS_EV_FLAG_FALLBACK) { + erts_printf("%s%s", delim, "FLBK"); delim = "|"; } - if(f & ERTS_EV_FLAG_DEFER_OUT_EV) { - erts_printf("%s%s", delim, "DROUT"); +} + +#ifdef DEBUG_PRINT_MODE + +static ERTS_INLINE char * +drvmode2str(int mode) { + switch (mode) { + case ERL_DRV_READ|ERL_DRV_USE: return "READ|USE"; + case ERL_DRV_WRITE|ERL_DRV_USE: return "WRITE|USE"; + case ERL_DRV_READ|ERL_DRV_WRITE|ERL_DRV_USE: return "READ|WRITE|USE"; + case ERL_DRV_USE: return "USE"; + case ERL_DRV_READ: return "READ"; + case ERL_DRV_WRITE: return "WRITE"; + case ERL_DRV_READ|ERL_DRV_WRITE: return "READ|WRITE"; + default: return "UNKNOWN"; + } +} + +static ERTS_INLINE char * +nifmode2str(enum ErlNifSelectFlags mode) { + switch (mode) { + case ERL_NIF_SELECT_READ: return "READ"; + case ERL_NIF_SELECT_WRITE: return "WRITE"; + case ERL_NIF_SELECT_STOP: return "STOP"; + default: return "UNKNOWN"; } } +#endif + typedef struct { int used_fds; int num_errors; @@ -2506,53 +2330,32 @@ typedef struct { #endif } IterDebugCounters; -static void doit_erts_check_io_debug(void *vstate, void *vcounters) +static int erts_debug_print_checkio_state(ErtsDrvEventState *state, + ErtsPollEvents ep_events, + int internal) { - ErtsDrvEventState *state = (ErtsDrvEventState *) vstate; - IterDebugCounters *counters = (IterDebugCounters *) vcounters; - ErtsPollEvents cio_events = state->events; - ErtsSysFdType fd = state->fd; -#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - int internal = 0; - ErtsPollEvents ep_events = counters->epep[(int) fd]; -#endif - int err = 0; - #if defined(HAVE_FSTAT) && !defined(NO_FSTAT_ON_SYS_FD_TYPE) struct stat stat_buf; #endif - if (state->driver.select) - counters->no_driver_select_structs++; - if (state->driver.nif) - counters->no_enif_select_structs++; - + ErtsSysFdType fd = state->fd; + ErtsPollEvents cio_events = state->events; + int err = 0; #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - if (state->events || ep_events) { - if (ep_events & ERTS_POLL_EV_NVAL) { - ep_events &= ~ERTS_POLL_EV_NVAL; - internal = 1; - counters->internal_fds++; - } - else - counters->used_fds++; -#else - if (state->events) { - counters->used_fds++; + ErtsPollEvents aio_events = state->active_events; #endif - - erts_printf("pollset=%d fd=%d ", - (int)(state->pollset - pollsetv), (int) fd); - + erts_printf("pollset=%d fd=%d ", + state->flags & ERTS_EV_FLAG_FALLBACK ? -1 : get_pollset_id(fd), (int) fd); + #if defined(HAVE_FSTAT) && !defined(NO_FSTAT_ON_SYS_FD_TYPE) - if (fstat((int) fd, &stat_buf) < 0) - erts_printf("type=unknown "); - else { - erts_printf("type="); + if (fstat((int) fd, &stat_buf) < 0) + erts_printf("type=unknown "); + else { + erts_printf("type="); #ifdef S_ISSOCK - if (S_ISSOCK(stat_buf.st_mode)) - erts_printf("sock "); - else + if (S_ISSOCK(stat_buf.st_mode)) + erts_printf("sock "); + else #endif #ifdef S_ISFIFO if (S_ISFIFO(stat_buf.st_mode)) @@ -2560,196 +2363,239 @@ static void doit_erts_check_io_debug(void *vstate, void *vcounters) else #endif #ifdef S_ISCHR - if (S_ISCHR(stat_buf.st_mode)) - erts_printf("chr "); - else + if (S_ISCHR(stat_buf.st_mode)) + erts_printf("chr "); + else #endif #ifdef S_ISDIR - if (S_ISDIR(stat_buf.st_mode)) - erts_printf("dir "); - else + if (S_ISDIR(stat_buf.st_mode)) + erts_printf("dir "); + else #endif #ifdef S_ISBLK - if (S_ISBLK(stat_buf.st_mode)) - erts_printf("blk "); - else + if (S_ISBLK(stat_buf.st_mode)) + erts_printf("blk "); + else #endif #ifdef S_ISREG - if (S_ISREG(stat_buf.st_mode)) - erts_printf("reg "); - else + if (S_ISREG(stat_buf.st_mode)) + erts_printf("reg "); + else #endif #ifdef S_ISLNK - if (S_ISLNK(stat_buf.st_mode)) - erts_printf("lnk "); - else + if (S_ISLNK(stat_buf.st_mode)) + erts_printf("lnk "); + else #endif #ifdef S_ISDOOR - if (S_ISDOOR(stat_buf.st_mode)) - erts_printf("door "); - else + if (S_ISDOOR(stat_buf.st_mode)) + erts_printf("door "); + else #endif #ifdef S_ISWHT - if (S_ISWHT(stat_buf.st_mode)) - erts_printf("wht "); - else + if (S_ISWHT(stat_buf.st_mode)) + erts_printf("wht "); + else #endif #ifdef S_ISXATTR - if (S_ISXATTR(stat_buf.st_mode)) - erts_printf("xattr "); - else + if (S_ISXATTR(stat_buf.st_mode)) + erts_printf("xattr "); + else #endif - erts_printf("unknown "); - } + erts_printf("unknown "); + } #else - erts_printf("type=unknown "); + erts_printf("type=unknown "); #endif - if (state->type == ERTS_EV_TYPE_DRV_SEL) { - erts_printf("driver_select "); - -#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - if (internal) { - erts_printf("internal "); - err = 1; - } - - if (cio_events == ep_events) { - erts_printf("ev="); - if (print_events(cio_events) != 0) - err = 1; - } - else { - ErtsPollEvents ev = cio_events; -#if ERTS_CIO_DEFER_ACTIVE_EVENTS - if (state->flags & ERTS_EV_FLAG_DEFER_IN_EV) - ev &= ~ERTS_POLL_EV_IN; - if (state->flags & ERTS_EV_FLAG_DEFER_OUT_EV) - ev &= ~ERTS_POLL_EV_OUT; -#endif - if (ev != ep_events) - err = 1; - erts_printf("cio_ev="); - print_events(cio_events); - erts_printf(" ep_ev="); - print_events(ep_events); - } -#else - if (print_events(cio_events) != 0) - err = 1; -#endif - erts_printf(" "); - if (cio_events & ERTS_POLL_EV_IN) { - Eterm id = state->driver.select->inport; - if (is_nil(id)) { - erts_printf("inport=none inname=none indrv=none "); - err = 1; - } - else { - ErtsPortNames *pnp = erts_get_port_names(id, ERTS_INVALID_ERL_DRV_PORT); - erts_printf(" inport=%T inname=%s indrv=%s ", - id, - pnp->name ? pnp->name : "unknown", - (pnp->driver_name - ? pnp->driver_name - : "unknown")); - erts_free_port_names(pnp); - } - } - if (cio_events & ERTS_POLL_EV_OUT) { - Eterm id = state->driver.select->outport; - if (is_nil(id)) { - erts_printf("outport=none outname=none outdrv=none "); - err = 1; - } - else { - ErtsPortNames *pnp = erts_get_port_names(id, ERTS_INVALID_ERL_DRV_PORT); - erts_printf(" outport=%T outname=%s outdrv=%s ", - id, - pnp->name ? pnp->name : "unknown", - (pnp->driver_name - ? pnp->driver_name - : "unknown")); - erts_free_port_names(pnp); - } - } - } - else if (state->type == ERTS_EV_TYPE_NIF) { - ErtsResource* r; - erts_printf("enif_select "); + if (state->type == ERTS_EV_TYPE_DRV_SEL) { + erts_printf("driver_select "); #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - if (internal) { - erts_printf("internal "); - err = 1; - } - + if (internal) { + erts_printf("internal "); + err = 1; + } + if (aio_events == cio_events) { if (cio_events == ep_events) { erts_printf("ev="); if (print_events(cio_events) != 0) err = 1; } else { - err = 1; + ErtsPollEvents ev = cio_events; + if (ev != ep_events && ep_events != ERTS_POLL_EV_NONE) + err = 1; erts_printf("cio_ev="); print_events(cio_events); erts_printf(" ep_ev="); print_events(ep_events); } + } else { + erts_printf("cio_ev="); + print_events(cio_events); + erts_printf(" aio_ev="); + print_events(aio_events); + if ((aio_events != ep_events && ep_events != ERTS_POLL_EV_NONE) || + (aio_events != 0 && ep_events == ERTS_POLL_EV_NONE)) { + erts_printf(" ep_ev="); + print_events(ep_events); + err = 1; + } + } #else + if (print_events(cio_events) != 0) + err = 1; +#endif + erts_printf(" "); + if (cio_events & ERTS_POLL_EV_IN) { + Eterm id = state->driver.select->inport; + if (is_nil(id)) { + erts_printf("inport=none inname=none indrv=none "); + err = 1; + } + else { + ErtsPortNames *pnp = erts_get_port_names(id, ERTS_INVALID_ERL_DRV_PORT); + erts_printf(" inport=%T inname=%s indrv=%s ", + id, + pnp->name ? pnp->name : "unknown", + (pnp->driver_name + ? pnp->driver_name + : "unknown")); + erts_free_port_names(pnp); + } + } + if (cio_events & ERTS_POLL_EV_OUT) { + Eterm id = state->driver.select->outport; + if (is_nil(id)) { + erts_printf("outport=none outname=none outdrv=none "); + err = 1; + } + else { + ErtsPortNames *pnp = erts_get_port_names(id, ERTS_INVALID_ERL_DRV_PORT); + erts_printf(" outport=%T outname=%s outdrv=%s ", + id, + pnp->name ? pnp->name : "unknown", + (pnp->driver_name + ? pnp->driver_name + : "unknown")); + erts_free_port_names(pnp); + } + } + } + else if (state->type == ERTS_EV_TYPE_NIF) { + ErtsResource* r; + erts_printf("enif_select "); + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (internal) { + erts_printf("internal "); + err = 1; + } + + if (cio_events == ep_events) { + erts_printf("ev="); if (print_events(cio_events) != 0) err = 1; + } + else { + err = 1; + erts_printf("cio_ev="); + print_events(cio_events); + erts_printf(" ep_ev="); + print_events(ep_events); + } +#else + if (print_events(cio_events) != 0) + err = 1; #endif - erts_printf(" inpid=%T dd_cnt=%b32d", state->driver.nif->in.pid, - state->driver.nif->in.ddeselect_cnt); - erts_printf(" outpid=%T dd_cnt=%b32d", state->driver.nif->out.pid, - state->driver.nif->out.ddeselect_cnt); - r = state->driver.stop.resource; - erts_printf(" resource=%p(%T:%T)", r, r->type->module, r->type->name); + erts_printf(" inpid=%T", state->driver.nif->in.pid); + erts_printf(" outpid=%T", state->driver.nif->out.pid); + r = state->driver.stop.resource; + erts_printf(" resource=%p(%T:%T)", r, r->type->module, r->type->name); + } +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + else if (internal) { + erts_printf("internal "); + if (cio_events) { + err = 1; + erts_printf("cio_ev="); + print_events(cio_events); } + if (ep_events) { + erts_printf("ep_ev="); + print_events(ep_events); + } + } +#endif + else { + err = 1; + erts_printf("control_type=%d ", (int)state->type); #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - else if (internal) { - erts_printf("internal "); - if (cio_events) { - err = 1; - erts_printf("cio_ev="); - print_events(cio_events); - } - if (ep_events) { - erts_printf("ep_ev="); - print_events(ep_events); - } - } + if (cio_events == ep_events) { + erts_printf("ev="); + print_events(cio_events); + } + else { + erts_printf("cio_ev="); print_events(cio_events); + erts_printf(" ep_ev="); print_events(ep_events); + } +#else + erts_printf("ev=0x%b32x", (Uint32) cio_events); #endif - else { - err = 1; - erts_printf("control_type=%d ", (int)state->type); + } + + erts_printf(" flags="); print_flags(state->flags); + if (err) { + erts_printf(" ERROR"); + } + erts_printf("\r\n"); + return err; +} + +static void doit_erts_check_io_debug(void *vstate, void *vcounters) +{ + ErtsDrvEventState *state = (ErtsDrvEventState *) vstate; + IterDebugCounters *counters = (IterDebugCounters *) vcounters; + int internal = 0; #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS - if (cio_events == ep_events) { - erts_printf("ev="); - print_events(cio_events); - } - else { - erts_printf("cio_ev="); print_events(cio_events); - erts_printf(" ep_ev="); print_events(ep_events); - } + ErtsSysFdType fd = state->fd; + ErtsPollEvents ep_events = counters->epep[(int) fd]; #else - erts_printf("ev=0x%b32x", (Uint32) cio_events); + ErtsPollEvents ep_events = ERTS_POLL_EV_NONE; #endif - } - - erts_printf(" flags="); print_flags(state->flags); - if (err) { + if (state->driver.select) { + counters->no_driver_select_structs++; + ASSERT(state->events || (ep_events != 0 && ep_events != ERTS_POLL_EV_NONE)); + } + if (state->driver.nif) { + counters->no_enif_select_structs++; + ASSERT(state->events || (ep_events != 0 && ep_events != ERTS_POLL_EV_NONE)); + } + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (state->events || (ep_events != 0 && ep_events != ERTS_POLL_EV_NONE)) { + if (ep_events & ERTS_POLL_EV_NVAL) { + ep_events &= ~ERTS_POLL_EV_NVAL; + internal = 1; + counters->internal_fds++; + } + else + counters->used_fds++; +#else + if (state->events) { + counters->used_fds++; +#endif + if (erts_debug_print_checkio_state(state, ep_events, internal)) { counters->num_errors++; - erts_printf(" ERROR"); } - erts_printf("\n"); } } - + +/* ciodpi can be NULL when called from etp-commands */ int -ERTS_CIO_EXPORT(erts_check_io_debug)(ErtsCheckIoDebugInfo *ciodip) +erts_check_io_debug(ErtsCheckIoDebugInfo *ciodip) { #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS int fd, len, i; @@ -2758,12 +2604,10 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(ErtsCheckIoDebugInfo *ciodip) #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS ErtsDrvEventState null_des; - null_des.pollset = NULL; null_des.driver.select = NULL; null_des.driver.nif = NULL; null_des.driver.stop.drv_ptr = NULL; null_des.events = 0; - null_des.remove_cnt = 0; null_des.type = ERTS_EV_TYPE_NONE; null_des.flags = 0; @@ -2771,24 +2615,36 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(ErtsCheckIoDebugInfo *ciodip) sizeof(ErtsPollEvents)*drv_ev_state.max_fds); #endif - erts_printf("--- fds in pollset --------------------------------------\n"); #if defined(ERTS_ENABLE_LOCK_CHECK) erts_lc_check_exact(NULL, 0); /* No locks should be locked */ #endif - erts_thr_progress_block(); /* stop the world to avoid messy locking */ + if (ciodip) + erts_thr_progress_block(); /* stop the world to avoid messy locking */ #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS len = erts_atomic_read_nob(&drv_ev_state.len); - for (i = 0; i < erts_no_schedulers; i++) { - ERTS_POLL_EXPORT(erts_poll_get_selected_events)(pollsetv[i].ps, - counters.epep, - drv_ev_state.max_fds); +#if ERTS_POLL_USE_FALLBACK + erts_printf("--- fds in flbk pollset ---------------------------------\n"); + erts_poll_get_selected_events_flbk(get_fallback(), counters.epep, + drv_ev_state.max_fds); + for (fd = 0; fd < len; fd++) { + if (drv_ev_state.v[fd].flags & ERTS_EV_FLAG_FALLBACK) + doit_erts_check_io_debug(&drv_ev_state.v[fd], &counters); + } +#endif + erts_printf("--- fds in pollset --------------------------------------\n"); + + for (i = 0; i < erts_no_pollsets; i++) { + erts_poll_get_selected_events(pollsetv[i], + counters.epep, + drv_ev_state.max_fds); for (fd = 0; fd < len; fd++) { - if (drv_ev_state.v[fd].pollset == &pollsetv[i]) - doit_erts_check_io_debug(&drv_ev_state.v[fd], &counters); + if (!(drv_ev_state.v[fd].flags & ERTS_EV_FLAG_FALLBACK) + && get_pollset_id(fd) == i) + doit_erts_check_io_debug(&drv_ev_state.v[fd], &counters); } } for (fd = len ; fd < drv_ev_state.max_fds; fd++) { @@ -2799,11 +2655,15 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(ErtsCheckIoDebugInfo *ciodip) safe_hash_for_each(&drv_ev_state.tab, &doit_erts_check_io_debug, &counters); #endif - erts_thr_progress_unblock(); - ciodip->no_used_fds = counters.used_fds; - ciodip->no_driver_select_structs = counters.no_driver_select_structs; - ciodip->no_enif_select_structs = counters.no_enif_select_structs; + if (ciodip) + erts_thr_progress_unblock(); + + if (ciodip) { + ciodip->no_used_fds = counters.used_fds; + ciodip->no_driver_select_structs = counters.no_driver_select_structs; + ciodip->no_enif_select_structs = counters.no_enif_select_structs; + } erts_printf("\n"); erts_printf("used fds=%d\n", counters.used_fds); @@ -2822,97 +2682,19 @@ ERTS_CIO_EXPORT(erts_check_io_debug)(ErtsCheckIoDebugInfo *ciodip) } #ifdef ERTS_ENABLE_LOCK_COUNT -void ERTS_CIO_EXPORT(erts_lcnt_update_cio_locks)(int enable) { +void erts_lcnt_update_cio_locks(int enable) { + int i; #ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS erts_lcnt_enable_hash_lock_count(&drv_ev_state.tab, ERTS_LOCK_FLAGS_CATEGORY_IO, enable); #else (void)enable; #endif -} -#endif /* ERTS_ENABLE_LOCK_COUNT */ - -#ifdef ERTS_ENABLE_KERNEL_POLL -# ifdef ERTS_KERNEL_POLL_VERSION - -/* - * Compile these only once for kp/nkp - */ - -void erts_init_check_io(void) -{ - if (erts_use_kernel_poll) - erts_init_check_io_kp(); - else - erts_init_check_io_nkp(); -} - -int -driver_select(ErlDrvPort port, ErlDrvEvent event, int mode, int on) -{ - return (*erts_io_funcs.select)(port, event, mode, on); -} - -int enif_select(ErlNifEnv* env, ErlNifEvent event, - enum ErlNifSelectFlags flags, void* obj, const ErlNifPid* pid, Eterm ref) -{ - return (*erts_io_funcs.enif_select)(env, event, flags, obj, pid, ref); -} - -struct pollset_info* erts_get_pollset(int sched_nr) -{ - return (*erts_io_funcs.get_pollset)(sched_nr); -} -void erts_io_notify_port_task_executed(ErtsPortTaskHandle *pthp) -{ - erts_io_funcs.notify_port_task_executed(pthp); -} - -void erts_check_io(int do_wait) -{ - erts_io_funcs.check_io(do_wait); -} - -Uint erts_check_io_size(void) -{ - return erts_io_funcs.size(); -} - - -Eterm erts_check_io_info(void *p) -{ - return (*erts_io_funcs.info)(p); -} - -int erts_check_io_max_files(void) -{ - return erts_io_funcs.max_files(); -} - -int -erts_check_io_debug(ErtsCheckIoDebugInfo *ip) -{ - return (*erts_io_funcs.check_io_debug)(ip); -} - -void erts_check_io_interrupt(struct pollset_info* psi, int set) -{ - erts_io_funcs.check_io_interrupt(psi, set); -} - -void erts_check_io_interrupt_timed(struct pollset_info* psi, int set, - ErtsMonotonicTime timeout_time) -{ - erts_io_funcs.check_io_interrupt_tmd(psi, set, timeout_time); -} - -#ifdef ERTS_ENABLE_LOCK_COUNT -void erts_lcnt_update_cio_locks(int enable) -{ - erts_io_funcs.lcnt_update_cio_locks(enable); -} +#if ERTS_POLL_USE_FALLBACK + erts_lcnt_enable_pollset_lock_count_flbk(get_fallback(), enable); #endif -#endif /* ERTS_KERNEL_POLL_VERSION */ - -#endif /* !ERTS_ENABLE_KERNEL_POLL */ + for (i = 0; i < erts_no_pollsets; i++) + erts_lcnt_enable_pollset_lock_count(pollsetv[i], enable); +} +#endif /* ERTS_ENABLE_LOCK_COUNT */ diff --git a/erts/emulator/sys/common/erl_check_io.h b/erts/emulator/sys/common/erl_check_io.h index ab53d91756..443ef1264c 100644 --- a/erts/emulator/sys/common/erl_check_io.h +++ b/erts/emulator/sys/common/erl_check_io.h @@ -18,10 +18,11 @@ * %CopyrightEnd% */ -/* - * Description: Check I/O +/** + * @description Check I/O, a cross platform IO polling framework for ERTS * - * Author: Rickard Green + * @author Rickard Green + * @author Lukas Larsson */ #ifndef ERL_CHECK_IO_H__ @@ -30,35 +31,96 @@ #include "sys.h" #include "erl_sys_driver.h" -struct pollset_info; +/** @brief a structure that is used by each polling thread */ +struct erts_poll_thread; +/** + * Get the memory size of the check io framework + */ Uint erts_check_io_size(void); -Eterm erts_check_io_info(void *); -void erts_io_notify_port_task_executed(ErtsPortTaskHandle *pthp); -void erts_check_io_async_sig_interrupt(struct pollset_info *psi); +/** + * Returns an Eterm with information about all the pollsets active at the + * moment. + * + * @param proc the Process* to allocate the result on. It is passed as + * void * because of header include problems. + */ +Eterm erts_check_io_info(void *proc); +/** + * Should be called when a port IO task has been executed in order to re-enable + * or clear the information about the fd. + * + * @param type The type of event that has been completed. + * @param handle The port task handle of the event. + * @param reset A function pointer to be called when the port task handle + * should be reset. + */ +void erts_io_notify_port_task_executed(ErtsPortTaskType type, + ErtsPortTaskHandle *handle, + void (*reset)(ErtsPortTaskHandle *)); +/** + * Returns the maximum number of fds that the check io framework can handle. + */ int erts_check_io_max_files(void); -void erts_check_io(int); -void erts_init_check_io(void); +/** + * Called by any thread that should check for new IO events. This function will + * not return unless erts_check_io_interrupt(pt, 1) is called by another thread. + * + * @param pt the poll thread structure to use. + */ +void erts_check_io(struct erts_poll_thread *pt); +/** + * Initialize the check io framework. This function will parse the arguments + * and delete any entries that it is interested in. + * + * @param argc the number of arguments + * @param argv an array with the arguments + */ +void erts_init_check_io(int *argc, char **argv); +/** + * Interrupt the poll thread so that it can execute other code. + * + * Should be called with set = 0 by the waiting thread before calling + * erts_check_io. + * + * @param pt the poll thread to wake + * @param set whether to set or clear the interrupt flag + */ +void erts_check_io_interrupt(struct erts_poll_thread *pt, int set); +/** + * Create a new poll thread structure that is associated with the number no. + * It is the callers responsibility that no is unique. + */ +struct erts_poll_thread* erts_create_pollset_thread(int no); #ifdef ERTS_ENABLE_LOCK_COUNT +/** + * Toggle lock counting on all check io locks + */ void erts_lcnt_update_cio_locks(int enable); #endif -void erts_check_io_interrupt(struct pollset_info*, int); -void erts_check_io_interrupt_timed(struct pollset_info*, int, ErtsMonotonicTime); -struct pollset_info* erts_get_pollset(int sched_num); - typedef struct { ErtsPortTaskHandle task; - erts_atomic_t executed_time; - struct pollset_info *pollset; + ErtsSysFdType fd; } ErtsIoTask; + #endif /* ERL_CHECK_IO_H__ */ #if !defined(ERL_CHECK_IO_C__) && !defined(ERTS_ALLOC_C__) #define ERL_CHECK_IO_INTERNAL__ #endif +#define ERTS_CHECK_IO_DRV_EV_STATE_LOCK_CNT 128 + +/* Controls how many pollsets to allocate. Fd's are hashed into + each pollset based on the FD. When doing non-concurrent updates + there will be one pollset per thread. +*/ +extern int erts_no_pollsets; +extern int erts_no_poll_threads; + + #ifndef ERL_CHECK_IO_INTERNAL__ #define ERL_CHECK_IO_INTERNAL__ #include "erl_poll.h" diff --git a/erts/emulator/sys/common/erl_poll.c b/erts/emulator/sys/common/erl_poll.c index 341370ca14..c671ea99f4 100644 --- a/erts/emulator/sys/common/erl_poll.c +++ b/erts/emulator/sys/common/erl_poll.c @@ -18,9 +18,8 @@ * %CopyrightEnd% */ -/* - * Description: Poll interface suitable for ERTS with or without - * SMP support. +/** + * @description Poll interface suitable for ERTS * * The interface is currently implemented using: * - select @@ -29,12 +28,36 @@ * - epoll with poll or select as fallback * - kqueue with poll or select as fallback * - * Some time in the future it will also be - * implemented using Solaris ports. * + * @author Rickard Green + * @author Lukas Larsson + * + * There are two major different implementations off IO polling in this + * file. The concurrent and non-concurrent implementations. + * When available epoll/kqueue are used to implement the concurrent + * versions. poll, select and dev/poll use non-concurrent updates. + * + * Concurrent version: + * In the concurrent version erts_poll_control directly modifies + * the kernel pollset without waking the thread that is waiting + * on events. Also the ErtsPollResFd type is directly mapped to + * the native event type, so no extra copying is needed. Note that + * as no locking at all is done, fds can be triggered that have been + * removed from the pollset. The check_io layer has to deal with this. + * + * Non-concurrent version: + * In the non-concurrent version, the pollset has an internal representation + * of the pollset that is updated by erts_poll_control. When an fd is updated, + * its number is placed in the update request queue and then the waiting thread + * is woken in order to see the change. The internal data in the pollset is + * protected by a mutex that has to be taken by both the modifying and waiting + * thread at different times. * + * The non-concurrent pollset cannot have fd's closed in it while a thread is + * waiting on that fd. In order to fix this, when an ERTS_POLL_OP_DEL command + * is issued, the fd is marked as closing and the waiting thread is woken. The + * fd is then returned in the waiting threads results as ERTS_POLL_EV_NONE. * - * Author: Rickard Green */ #ifdef HAVE_CONFIG_H @@ -62,6 +85,8 @@ # ifdef SYS_SELECT_H # include # endif +#elif defined(_DARWIN_UNLIMITED_SELECT) +# undef _DARWIN_UNLIMITED_SELECT #endif #ifdef NO_SYSCONF # if ERTS_POLL_USE_SELECT @@ -83,20 +108,35 @@ #error "Missing implementation of erts_poll()" #endif -#if defined(ERTS_KERNEL_POLL_VERSION) && !ERTS_POLL_USE_KERNEL_POLL -#error "Missing kernel poll implementation of erts_poll()" -#endif +#if 0 +#define ERTS_POLL_DEBUG_PRINT 1 + +#define DEBUG_PRINT(FMT, PS, ...) \ + do { \ + int myerrno = errno; \ + erts_printf("%d: " FMT "\r\n", (PS)->id, ##__VA_ARGS__); \ + errno = myerrno; \ + } while(0) + +/* Define to print info about modifications done to each fd */ +#define DEBUG_PRINT_FD(FMT, PS, FD, ...) DEBUG_PRINT("%d: " FMT, PS, FD, ##__VA_ARGS__) +/* Define to print entry and exit from erts_poll_wait (can be very spammy) */ +//#define DEBUG_PRINT_WAIT(FMT, PS, ...) DEBUG_PRINT(FMT, PS, ##__VA_ARGS__) -#if defined(ERTS_NO_KERNEL_POLL_VERSION) && ERTS_POLL_USE_KERNEL_POLL -#error "Kernel poll used when it shouldn't be used" +#else +#define ERTS_POLL_DEBUG_PRINT 0 +#define DEBUG_PRINT(...) #endif -#if 0 -#define ERTS_POLL_DEBUG_PRINT +#ifndef DEBUG_PRINT_FD +#define DEBUG_PRINT_FD(...) +#endif +#ifndef DEBUG_PRINT_WAIT +#define DEBUG_PRINT_WAIT(...) #endif -#ifdef _DARWIN_UNLIMITED_SELECT +#if defined(_DARWIN_UNLIMITED_SELECT) && ERTS_POLL_USE_SELECT typedef struct { size_t sz; fd_set* ptr; @@ -142,44 +182,41 @@ int ERTS_SELECT(int nfds, ERTS_fd_set *readfds, ERTS_fd_set *writefds, # define ERTS_SELECT select #endif -#define ERTS_POLL_USE_BATCH_UPDATE_POLLSET (ERTS_POLL_USE_DEVPOLL \ - || ERTS_POLL_USE_KQUEUE) +#define ERTS_POLL_IS_FALLBACK (ERTS_POLL_USE_POLL || ERTS_POLL_USE_SELECT) && ERTS_ENABLE_KERNEL_POLL -#define ERTS_POLL_USE_CONCURRENT_UPDATE ERTS_POLL_USE_EPOLL +#define ERTS_POLL_USE_CONCURRENT_UPDATE (ERTS_POLL_USE_EPOLL || ERTS_POLL_USE_KQUEUE) -#define ERTS_POLL_COALESCE_KP_RES (ERTS_POLL_USE_KQUEUE || ERTS_POLL_USE_EPOLL) +#if !ERTS_POLL_USE_CONCURRENT_UPDATE + +#define ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(PS) \ + erts_atomic32_set_nob(&(PS)->have_update_requests, (erts_aint32_t) 1) +#define ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(PS) \ + erts_atomic32_set_nob(&(PS)->have_update_requests, (erts_aint32_t) 0) +#define ERTS_POLLSET_HAVE_UPDATE_REQUESTS(PS) \ + ((int) erts_atomic32_read_nob(&(PS)->have_update_requests)) #define ERTS_POLLSET_LOCK(PS) \ erts_mtx_lock(&(PS)->mtx) #define ERTS_POLLSET_UNLOCK(PS) \ erts_mtx_unlock(&(PS)->mtx) -#define ERTS_POLLSET_SET_POLLED_CHK(PS) \ - ((int) erts_atomic32_xchg_nob(&(PS)->polled, (erts_aint32_t) 1)) -#define ERTS_POLLSET_UNSET_POLLED(PS) \ - erts_atomic32_set_nob(&(PS)->polled, (erts_aint32_t) 0) -#define ERTS_POLLSET_IS_POLLED(PS) \ - ((int) erts_atomic32_read_nob(&(PS)->polled)) +#else +#define ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(PS) +#define ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(PS) +#define ERTS_POLLSET_HAVE_UPDATE_REQUESTS(PS) 0 -#define ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(PS) \ - erts_atomic32_set_nob(&(PS)->have_update_requests, (erts_aint32_t) 1) -#define ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(PS) \ - erts_atomic32_set_nob(&(PS)->have_update_requests, (erts_aint32_t) 0) -#define ERTS_POLLSET_HAVE_UPDATE_REQUESTS(PS) \ - ((int) erts_atomic32_read_nob(&(PS)->have_update_requests)) +#define ERTS_POLLSET_LOCK(PS) +#define ERTS_POLLSET_UNLOCK(PS) -#if ERTS_POLL_USE_FALLBACK -# if ERTS_POLL_USE_POLL -# define ERTS_POLL_NEED_FALLBACK(PS) ((PS)->no_poll_fds > 1) -# elif ERTS_POLL_USE_SELECT -# define ERTS_POLL_NEED_FALLBACK(PS) ((PS)->no_select_fds > 1) -# endif #endif + /* * --- Data types ------------------------------------------------------------ */ +#if !ERTS_POLL_USE_CONCURRENT_UPDATE + #define ERTS_POLLSET_UPDATE_REQ_BLOCK_SIZE 128 typedef struct ErtsPollSetUpdateRequestsBlock_ ErtsPollSetUpdateRequestsBlock; @@ -189,43 +226,20 @@ struct ErtsPollSetUpdateRequestsBlock_ { int fds[ERTS_POLLSET_UPDATE_REQ_BLOCK_SIZE]; }; - - # define ERTS_POLL_FD_FLG_INURQ (((unsigned short) 1) << 0) -#if ERTS_POLL_USE_FALLBACK -# define ERTS_POLL_FD_FLG_INFLBCK (((unsigned short) 1) << 1) -# define ERTS_POLL_FD_FLG_USEFLBCK (((unsigned short) 1) << 2) -#endif -# define ERTS_POLL_FD_FLG_RST (((unsigned short) 1) << 3) +# define ERTS_POLL_FD_FLG_RST (((unsigned short) 1) << 1) + typedef struct { #if ERTS_POLL_USE_POLL int pix; #endif + ErtsPollEvents used_events; ErtsPollEvents events; -#if ERTS_POLL_COALESCE_KP_RES - unsigned short res_ev_ix; -#endif unsigned short flags; } ErtsFdStatus; - -#if ERTS_POLL_COALESCE_KP_RES -/* res_ev_ix max value */ -#define ERTS_POLL_MAX_RES ((1 << sizeof(unsigned short)*8) - 1) -#endif - -#if ERTS_POLL_USE_KQUEUE - -#define ERTS_POLL_KQ_OP_HANDLED 1 -#define ERTS_POLL_KQ_OP_DEL_R 2 -#define ERTS_POLL_KQ_OP_DEL_W 3 -#define ERTS_POLL_KQ_OP_ADD_R 4 -#define ERTS_POLL_KQ_OP_ADD_W 5 -#define ERTS_POLL_KQ_OP_ADD2_R 6 -#define ERTS_POLL_KQ_OP_ADD2_W 7 - #endif /* @@ -233,196 +247,172 @@ typedef struct { * get unique names in debugger for kp/nkp */ struct ERTS_POLL_EXPORT(erts_pollset) { - ErtsPollSet next; + int id; int internal_fd_limit; - ErtsFdStatus *fds_status; erts_atomic_t no_of_user_fds; - int fds_status_len; + #if ERTS_POLL_USE_KERNEL_POLL int kp_fd; - int res_events_len; -#if ERTS_POLL_USE_EPOLL - struct epoll_event *res_events; -#elif ERTS_POLL_USE_KQUEUE - struct kevent *res_events; -#elif ERTS_POLL_USE_DEVPOLL - struct pollfd *res_events; -#endif #endif /* ERTS_POLL_USE_KERNEL_POLL */ + #if ERTS_POLL_USE_POLL int next_poll_fds_ix; int no_poll_fds; int poll_fds_len; - struct pollfd*poll_fds; + struct pollfd *poll_fds; #elif ERTS_POLL_USE_SELECT int next_sel_fd; int max_fd; -#if ERTS_POLL_USE_FALLBACK - int no_select_fds; -#endif ERTS_fd_set input_fds; ERTS_fd_set res_input_fds; ERTS_fd_set output_fds; ERTS_fd_set res_output_fds; +#elif ERTS_POLL_USE_DEVPOLL + struct pollfd *poll_fds; + int poll_fds_ix; #endif + +#if !ERTS_POLL_USE_CONCURRENT_UPDATE + ErtsFdStatus *fds_status; + int fds_status_len; ErtsPollSetUpdateRequestsBlock update_requests; ErtsPollSetUpdateRequestsBlock *curr_upd_req_block; erts_atomic32_t have_update_requests; - erts_atomic32_t polled; erts_mtx_t mtx; int wake_fds[2]; -#if ERTS_POLL_USE_TIMERFD - int timer_fd; -#endif -#if ERTS_POLL_USE_FALLBACK - int fallback_used; -#endif erts_atomic32_t wakeup_state; - erts_atomic64_t timeout_time; -#ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS - erts_atomic_t no_avoided_wakeups; - erts_atomic_t no_avoided_interrupts; - erts_atomic_t no_interrupt_timed; #endif }; void erts_silence_warn_unused_result(long unused); static void fatal_error(char *format, ...); -static void fatal_error_async_signal_safe(char *error_str); static int max_fds = -1; -static ErtsPollSet pollsets; -static erts_mtx_t pollsets_lock; #if ERTS_POLL_USE_POLL +#if !ERTS_POLL_IS_FALLBACK +static ERTS_INLINE short ev2pollev(ErtsPollEvents ev) +{ + return ERTS_POLL_EV_E2N(ev); +} + +static ERTS_INLINE ErtsPollEvents pollev2ev(short ev) +{ + return ERTS_POLL_EV_N2E(ev); +} + +#else /* ERTS_POLL_IS_FALLBACK */ + static ERTS_INLINE short ev2pollev(ErtsPollEvents ev) { -#if !ERTS_POLL_USE_FALLBACK || ERTS_POLL_USE_KQUEUE - return ERTS_POLL_EV_E2N(ev); -#else /* Note, we only map events we are interested in */ short res_ev = (short) 0; if (ev & ERTS_POLL_EV_IN) - res_ev |= ERTS_POLL_EV_NKP_IN; + res_ev |= ERTS_POLL_EV_NKP_IN; if (ev & ERTS_POLL_EV_OUT) - res_ev |= ERTS_POLL_EV_NKP_OUT; + res_ev |= ERTS_POLL_EV_NKP_OUT; return res_ev; -#endif } static ERTS_INLINE ErtsPollEvents pollev2ev(short ev) { -#if !ERTS_POLL_USE_FALLBACK || ERTS_POLL_USE_KQUEUE - return ERTS_POLL_EV_N2E(ev); -#else /* Note, we only map events we are interested in */ ErtsPollEvents res_ev = (ErtsPollEvents) 0; if (ev & ERTS_POLL_EV_NKP_IN) - res_ev |= ERTS_POLL_EV_IN; + res_ev |= ERTS_POLL_EV_IN; if (ev & ERTS_POLL_EV_NKP_OUT) - res_ev |= ERTS_POLL_EV_OUT; + res_ev |= ERTS_POLL_EV_OUT; if (ev & ERTS_POLL_EV_NKP_ERR) - res_ev |= ERTS_POLL_EV_ERR; + res_ev |= ERTS_POLL_EV_ERR; if (ev & ERTS_POLL_EV_NKP_NVAL) - res_ev |= ERTS_POLL_EV_NVAL; - return res_ev; -#endif + res_ev |= ERTS_POLL_EV_NVAL; + return res_ev; } -#endif +#endif /* !ERTS_POLL_IS_FALLBACK */ + +#endif /* ERTS_POLL_USE_POLL */ + #ifdef HARD_DEBUG static void check_poll_result(ErtsPollResFd pr[], int len); -#if ERTS_POLL_USE_DEVPOLL -static void check_poll_status(ErtsPollSet ps); -#endif /* ERTS_POLL_USE_DEVPOLL */ #endif /* HARD_DEBUG */ -#ifdef ERTS_POLL_DEBUG_PRINT +#if ERTS_POLL_USE_DEVPOLL && defined(DEBUG) +static void check_poll_status(ErtsPollSet *ps); +#endif /* ERTS_POLL_USE_DEVPOLL && DEBUG */ static void print_misc_debug_info(void); +#if ERTS_POLL_USE_EPOLL +uint32_t epoll_events(int kp_fd, int fd); #endif -static ERTS_INLINE void -init_timeout_time(ErtsPollSet ps) -{ - erts_atomic64_init_nob(&ps->timeout_time, - (erts_aint64_t) ERTS_MONOTONIC_TIME_MAX); -} - -static ERTS_INLINE void -set_timeout_time(ErtsPollSet ps, ErtsMonotonicTime time) -{ - erts_atomic64_set_relb(&ps->timeout_time, - (erts_aint64_t) time); -} - -static ERTS_INLINE ErtsMonotonicTime -get_timeout_time(ErtsPollSet ps) -{ - return (ErtsMonotonicTime) erts_atomic64_read_acqb(&ps->timeout_time); -} #define ERTS_POLL_NOT_WOKEN 0 #define ERTS_POLL_WOKEN -1 #define ERTS_POLL_WOKEN_INTR 1 +#if !ERTS_POLL_USE_CONCURRENT_UPDATE static ERTS_INLINE void -reset_wakeup_state(ErtsPollSet ps) +reset_wakeup_state(ErtsPollSet *ps) { erts_atomic32_set_mb(&ps->wakeup_state, ERTS_POLL_NOT_WOKEN); } +#endif static ERTS_INLINE int -is_woken(ErtsPollSet ps) +is_woken(ErtsPollSet *ps) { +#if !ERTS_POLL_USE_CONCURRENT_UPDATE return erts_atomic32_read_acqb(&ps->wakeup_state) != ERTS_POLL_NOT_WOKEN; +#else + return 0; +#endif } static ERTS_INLINE int -is_interrupted_reset(ErtsPollSet ps) +is_interrupted_reset(ErtsPollSet *ps) { +#if !ERTS_POLL_USE_CONCURRENT_UPDATE return (erts_atomic32_xchg_acqb(&ps->wakeup_state, ERTS_POLL_NOT_WOKEN) == ERTS_POLL_WOKEN_INTR); +#else + return 0; +#endif } static ERTS_INLINE void -woke_up(ErtsPollSet ps) +woke_up(ErtsPollSet *ps) { +#if !ERTS_POLL_USE_CONCURRENT_UPDATE erts_aint32_t wakeup_state = erts_atomic32_read_acqb(&ps->wakeup_state); if (wakeup_state == ERTS_POLL_NOT_WOKEN) (void) erts_atomic32_cmpxchg_nob(&ps->wakeup_state, ERTS_POLL_WOKEN, ERTS_POLL_NOT_WOKEN); ASSERT(erts_atomic32_read_nob(&ps->wakeup_state) != ERTS_POLL_NOT_WOKEN); +#endif } /* * --- Wakeup pipe ----------------------------------------------------------- */ +#if !ERTS_POLL_USE_CONCURRENT_UPDATE static ERTS_INLINE void -wake_poller(ErtsPollSet ps, int interrupted, int async_signal_safe) +wake_poller(ErtsPollSet *ps, int interrupted) { int wake; - if (async_signal_safe) - wake = 1; - else { - erts_aint32_t wakeup_state; - if (!interrupted) - wakeup_state = erts_atomic32_cmpxchg_relb(&ps->wakeup_state, - ERTS_POLL_WOKEN, - ERTS_POLL_NOT_WOKEN); - else - wakeup_state = erts_atomic32_xchg_relb(&ps->wakeup_state, - ERTS_POLL_WOKEN_INTR); - wake = wakeup_state == ERTS_POLL_NOT_WOKEN; - } - /* - * NOTE: This function might be called from signal handlers in the - * non-smp case; therefore, it has to be async-signal safe in - * the non-smp case. - */ + erts_aint32_t wakeup_state; + if (!interrupted) + wakeup_state = erts_atomic32_cmpxchg_relb(&ps->wakeup_state, + ERTS_POLL_WOKEN, + ERTS_POLL_NOT_WOKEN); + else + wakeup_state = erts_atomic32_xchg_relb(&ps->wakeup_state, + ERTS_POLL_WOKEN_INTR); + wake = wakeup_state == ERTS_POLL_NOT_WOKEN; + if (wake) { ssize_t res; if (ps->wake_fds[1] < 0) @@ -432,29 +422,27 @@ wake_poller(ErtsPollSet ps, int interrupted, int async_signal_safe) res = write(ps->wake_fds[1], "!", 1); } while (res < 0 && errno == EINTR); if (res <= 0 && errno != ERRNO_BLOCK) { - if (async_signal_safe) - fatal_error_async_signal_safe(__FILE__ - ":XXX:wake_poller(): " - "Failed to write on wakeup pipe\n"); - else - fatal_error("%s:%d:wake_poller(): " - "Failed to write to wakeup pipe fd=%d: " - "%s (%d)\n", - __FILE__, __LINE__, - ps->wake_fds[1], - erl_errno_id(errno), errno); + fatal_error("%s:%d:wake_poller(): " + "Failed to write to wakeup pipe fd=%d: " + "%s (%d)\n", + __FILE__, __LINE__, + ps->wake_fds[1], + erl_errno_id(errno), errno); } } } static ERTS_INLINE void -cleanup_wakeup_pipe(ErtsPollSet ps) +cleanup_wakeup_pipe(ErtsPollSet *ps) { + int intr = 0; int fd = ps->wake_fds[0]; int res; do { char buf[32]; res = read(fd, buf, sizeof(buf)); + if (res > 0) + intr = 1; } while (res > 0 || (res < 0 && errno == EINTR)); if (res < 0 && errno != ERRNO_BLOCK) { fatal_error("%s:%d:cleanup_wakeup_pipe(): " @@ -464,10 +452,12 @@ cleanup_wakeup_pipe(ErtsPollSet ps) fd, erl_errno_id(errno), errno); } + if (intr) + erts_atomic32_set_nob(&ps->wakeup_state, ERTS_POLL_WOKEN_INTR); } static void -create_wakeup_pipe(ErtsPollSet ps) +create_wakeup_pipe(ErtsPollSet *ps) { int do_wake = 0; int wake_fds[2]; @@ -484,20 +474,13 @@ create_wakeup_pipe(ErtsPollSet ps) SET_NONBLOCKING(wake_fds[0]); SET_NONBLOCKING(wake_fds[1]); -#ifdef ERTS_POLL_DEBUG_PRINT - erts_printf("wakeup fds = {%d, %d}\n", wake_fds[0], wake_fds[1]); -#endif + DEBUG_PRINT("wakeup fds = {%d, %d}", ps, wake_fds[0], wake_fds[1]); ERTS_POLL_EXPORT(erts_poll_control)(ps, wake_fds[0], + ERTS_POLL_OP_ADD, ERTS_POLL_EV_IN, - 1, &do_wake); -#if ERTS_POLL_USE_FALLBACK - /* We depend on the wakeup pipe being handled by kernel poll */ - if (ps->fds_status[wake_fds[0]].flags & ERTS_POLL_FD_FLG_INFLBCK) - fatal_error("%s:%d:create_wakeup_pipe(): Internal error\n", - __FILE__, __LINE__); -#endif + &do_wake); if (ps->internal_fd_limit <= wake_fds[1]) ps->internal_fd_limit = wake_fds[1] + 1; if (ps->internal_fd_limit <= wake_fds[0]) @@ -507,81 +490,12 @@ create_wakeup_pipe(ErtsPollSet ps) } -/* - * --- timer fd ----------------------------------------------------------- - */ - -#if ERTS_POLL_USE_TIMERFD - -/* We use the timerfd when using epoll_wait to get high accuracy - timeouts, i.e. we want to sleep with < ms accuracy. */ - -static void -create_timerfd(ErtsPollSet ps) -{ - int do_wake = 0; - int timer_fd; - timer_fd = timerfd_create(CLOCK_MONOTONIC,0); - ERTS_POLL_EXPORT(erts_poll_control)(ps, - timer_fd, - ERTS_POLL_EV_IN, - 1, &do_wake); -#if ERTS_POLL_USE_FALLBACK - /* We depend on the wakeup pipe being handled by kernel poll */ - if (ps->fds_status[timer_fd].flags & ERTS_POLL_FD_FLG_INFLBCK) - fatal_error("%s:%d:create_wakeup_pipe(): Internal error\n", - __FILE__, __LINE__); -#endif - if (ps->internal_fd_limit <= timer_fd) - ps->internal_fd_limit = timer_fd + 1; - ps->timer_fd = timer_fd; -} - -static ERTS_INLINE void -timerfd_set(ErtsPollSet ps, struct itimerspec *its) -{ -#ifdef DEBUG - struct itimerspec old_its; - int res; - res = timerfd_settime(ps->timer_fd, 0, its, &old_its); - ASSERT(res == 0); - ASSERT(old_its.it_interval.tv_sec == 0 && - old_its.it_interval.tv_nsec == 0 && - old_its.it_value.tv_sec == 0 && - old_its.it_value.tv_nsec == 0); - -#else - timerfd_settime(ps->timer_fd, 0, its, NULL); -#endif -} - -static ERTS_INLINE int -timerfd_clear(ErtsPollSet ps, int res, int max_res) { - - struct itimerspec its; - /* we always have to clear the timer */ - its.it_interval.tv_sec = 0; - its.it_interval.tv_nsec = 0; - its.it_value.tv_sec = 0; - its.it_value.tv_nsec = 0; - timerfd_settime(ps->timer_fd, 0, &its, NULL); - - /* only timeout fd triggered */ - if (res == 1 && ps->res_events[0].data.fd == ps->timer_fd) - return 0; - - return res; -} - -#endif /* ERTS_POLL_USE_TIMERFD */ - - /* * --- Poll set update requests ---------------------------------------------- */ static ERTS_INLINE void -enqueue_update_request(ErtsPollSet ps, int fd) +enqueue_update_request(ErtsPollSet *ps, int fd) { ErtsPollSetUpdateRequestsBlock *urqbp; @@ -596,13 +510,11 @@ enqueue_update_request(ErtsPollSet ps, int fd) urqbp = ps->curr_upd_req_block; if (urqbp->len == ERTS_POLLSET_UPDATE_REQ_BLOCK_SIZE) { - ASSERT(!urqbp->next); urqbp = erts_alloc(ERTS_ALC_T_POLLSET_UPDREQ, sizeof(ErtsPollSetUpdateRequestsBlock)); - ps->curr_upd_req_block->next = urqbp; - ps->curr_upd_req_block = urqbp; - urqbp->next = NULL; + urqbp->next = ps->curr_upd_req_block; urqbp->len = 0; + ps->curr_upd_req_block = urqbp; } ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_INURQ; @@ -610,28 +522,29 @@ enqueue_update_request(ErtsPollSet ps, int fd) } static ERTS_INLINE void -free_update_requests_block(ErtsPollSet ps, +free_update_requests_block(ErtsPollSet *ps, ErtsPollSetUpdateRequestsBlock *urqbp) { if (urqbp != &ps->update_requests) erts_free(ERTS_ALC_T_POLLSET_UPDREQ, (void *) urqbp); else { - urqbp->next = NULL; urqbp->len = 0; } } +#endif /* !ERTS_POLL_USE_CONCURRENT_UPDATE */ /* * --- Growing poll set structures ------------------------------------------- */ -#ifndef ERTS_KERNEL_POLL_VERSION /* only one shared implementation */ +#if !ERTS_NO_KERNEL_POLL_VERSION || !ERTS_ENABLE_KERNEL_POLL +/* only one shared implementation */ #define ERTS_FD_TABLE_MIN_LENGTH 1024 #define ERTS_FD_TABLE_EXP_THRESHOLD (2048*1024) -int erts_poll_new_table_len (int old_len, int need_len) +int erts_poll_new_table_len(int old_len, int need_len) { int new_len; @@ -641,7 +554,7 @@ int erts_poll_new_table_len (int old_len, int need_len) } else { new_len = old_len; - do { + do { if (new_len < ERTS_FD_TABLE_EXP_THRESHOLD) new_len *= 2; else @@ -654,30 +567,9 @@ int erts_poll_new_table_len (int old_len, int need_len) } #endif -#if ERTS_POLL_USE_KERNEL_POLL -static void -grow_res_events(ErtsPollSet ps, int new_len) -{ - size_t new_size = sizeof( -#if ERTS_POLL_USE_EPOLL - struct epoll_event -#elif ERTS_POLL_USE_DEVPOLL - struct pollfd -#elif ERTS_POLL_USE_KQUEUE - struct kevent -#endif - ) * erts_poll_new_table_len(ps->res_events_len, new_len); - /* We do not need to save previously stored data */ - if (ps->res_events) - erts_free(ERTS_ALC_T_POLL_RES_EVS, ps->res_events); - ps->res_events = erts_alloc(ERTS_ALC_T_POLL_RES_EVS, new_size); - ps->res_events_len = new_len; -} -#endif /* ERTS_POLL_USE_KERNEL_POLL */ - #if ERTS_POLL_USE_POLL static void -grow_poll_fds(ErtsPollSet ps, int min_ix) +grow_poll_fds(ErtsPollSet *ps, int min_ix) { int i; int new_len = erts_poll_new_table_len(ps->poll_fds_len, min_ix + 1); @@ -725,8 +617,9 @@ ensure_select_fds(int fd, ERTS_fd_set* in, ERTS_fd_set* out) # define ensure_select_fds(fd, in, out) do {} while(0) #endif /* _DARWIN_UNLIMITED_SELECT */ +#if !ERTS_POLL_USE_CONCURRENT_UPDATE static void -grow_fds_status(ErtsPollSet ps, int min_fd) +grow_fds_status(ErtsPollSet *ps, int min_fd) { int i; int new_len = erts_poll_new_table_len(ps->fds_status_len, min_fd + 1); @@ -745,95 +638,290 @@ grow_fds_status(ErtsPollSet ps, int min_fd) #endif ps->fds_status[i].used_events = (ErtsPollEvents) 0; ps->fds_status[i].events = (ErtsPollEvents) 0; -#if ERTS_POLL_COALESCE_KP_RES - ps->fds_status[i].res_ev_ix = (unsigned short) ERTS_POLL_MAX_RES; -#endif ps->fds_status[i].flags = (unsigned short) 0; } ps->fds_status_len = new_len; } +#endif /* * --- Selecting fd to poll on ----------------------------------------------- */ -#if ERTS_POLL_USE_FALLBACK -static int update_fallback_pollset(ErtsPollSet ps, int fd); -#endif - -static ERTS_INLINE int -need_update(ErtsPollSet ps, int fd) +#if ERTS_POLL_USE_EPOLL +static int +update_pollset(ErtsPollSet *ps, int fd, ErtsPollOp op, ErtsPollEvents events) { -#if ERTS_POLL_USE_KERNEL_POLL - int reset; -#endif + int res; + int epoll_op = EPOLL_CTL_MOD; + struct epoll_event epe_templ; + struct epoll_event epe; - ASSERT(fd < ps->fds_status_len); + epe_templ.events = ERTS_POLL_EV_E2N(events) | EPOLLONESHOT; + epe_templ.data.fd = fd; -#if ERTS_POLL_USE_KERNEL_POLL - reset = (int) (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST); - if (reset && !ps->fds_status[fd].used_events) { - ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST; - reset = 0; - } -#else - ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST; +#ifdef VALGRIND + /* Silence invalid valgrind warning ... */ + memset((void *) &epe.data, 0, sizeof(epoll_data_t)); #endif - if (ps->fds_status[fd].used_events != ps->fds_status[fd].events) - return 1; + switch (op) { + case ERTS_POLL_OP_DEL: + /* A note on EPOLL_CTL_DEL: linux kernel versions before 2.6.9 + need a non-NULL event pointer even though it is ignored... */ + epoll_op = EPOLL_CTL_DEL; + epe_templ.events = 0; + erts_atomic_dec_nob(&ps->no_of_user_fds); + break; + case ERTS_POLL_OP_ADD: + epoll_op = EPOLL_CTL_ADD; + erts_atomic_inc_nob(&ps->no_of_user_fds); + break; + case ERTS_POLL_OP_MOD: + epoll_op = EPOLL_CTL_MOD; + break; + default: + ASSERT(0); + break; + } -#if ERTS_POLL_USE_KERNEL_POLL - return reset; -#else - return 0; + do { + /* We init 'epe' every time since epoll_ctl() may modify it + (not declared const and not documented as const). */ + epe.events = epe_templ.events; + epe.data.fd = epe_templ.data.fd; + res = epoll_ctl(ps->kp_fd, epoll_op, fd, &epe); + } while (res != 0 && errno == EINTR); + +#if ERTS_POLL_DEBUG_PRINT + { + int saved_errno = errno; + DEBUG_PRINT_FD("%s = epoll_ctl(%d, %s, %d, {0x%x, %d})", + ps, fd, + res == 0 ? "0" : erl_errno_id(errno), + ps->kp_fd, + (epoll_op == EPOLL_CTL_ADD + ? "EPOLL_CTL_ADD" + : (epoll_op == EPOLL_CTL_MOD + ? "EPOLL_CTL_MOD" + : (epoll_op == EPOLL_CTL_DEL + ? "EPOLL_CTL_DEL" + : "UNKNOWN"))), + fd, + epe_templ.events, + fd); + errno = saved_errno; + } #endif + if (res != 0) { + switch (op) { + case ERTS_POLL_OP_MOD: + epe.events = 0; + do { + /* We init 'epe' every time since epoll_ctl() may modify it + (not declared const and not documented as const). */ + epe.events = 0; + epe.data.fd = fd; + res = epoll_ctl(ps->kp_fd, EPOLL_CTL_DEL, fd, &epe); + } while (res != 0 && errno == EINTR); + /* Fall through ... */ + case ERTS_POLL_OP_ADD: { + erts_atomic_dec_nob(&ps->no_of_user_fds); + res = ERTS_POLL_EV_NVAL; + break; + } + case ERTS_POLL_OP_DEL: { + /* + * Since we use a lazy update approach EPOLL_CTL_DEL will + * frequently fail. This since epoll automatically removes + * a filedescriptor that is closed from the poll set. + */ + res = 0; + break; + } + default: + fatal_error("%s:%d:update_pollset(): Internal error\n", + __FILE__, __LINE__); + break; + } + } else { + res = events; + } + return res; } -#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET +#endif /* ERTS_POLL_USE_EPOLL */ #if ERTS_POLL_USE_KQUEUE -#define ERTS_POLL_MIN_BATCH_BUF_SIZE 128 + +/* Some versions of the EV_SET macro used kevp multiple times, + so we define out own version that make sure that it is safe + to do kevp++ in the argument list. */ +#define ERTS_EV_SET(kevp, a, b, c, f) do { \ + struct kevent *kevp_ = kevp; \ + EV_SET(kevp_, a, b, c, 0, 0, f); \ + } while(0) + +static int +update_pollset(ErtsPollSet *ps, int fd, ErtsPollOp op, ErtsPollEvents events) +{ + int res = 0, len = 0; + struct kevent evts[2]; + struct timespec ts = {0, 0}; + +#ifdef EV_DISPATCH + /* If we have EV_DISPATCH we use it. The kevent descriptions for both + read and write are added on OP_ADD and removed on OP_DEL. And then + after than only EV_ENABLE|EV_DISPATCH are used. + + It could be possible to not modify the pollset when disabling and/or + deleting events, but that may cause the poll threads to be awoken + a lot more than they should so we take the cost here instead of + in the poll thread. + + Note: We need to have EV_DISPATCH both when the event is enabled and + disabled, as otherwise the event may be triggered twice on each re-arm. + Not sure if this is intended or not (can't find anything about it in the + man page), but it seems to be the way it works... + */ + + if (op == ERTS_POLL_OP_DEL) { + erts_atomic_dec_nob(&ps->no_of_user_fds); + /* We could probably skip this delete, do we want to? */ + ERTS_EV_SET(&evts[len++], fd, EVFILT_READ, EV_DELETE, (void *) 0); + ERTS_EV_SET(&evts[len++], fd, EVFILT_WRITE, EV_DELETE, (void *) 0); + } else if (op == ERTS_POLL_OP_ADD) { + uint32_t flags; + erts_atomic_inc_nob(&ps->no_of_user_fds); + + flags = EV_ADD|EV_DISPATCH; + flags |= ((events & ERTS_POLL_EV_IN) ? 0 : EV_DISABLE); + ERTS_EV_SET(&evts[len++], fd, EVFILT_READ, flags, (void *) ERTS_POLL_EV_IN); + + flags = EV_ADD|EV_DISPATCH; + flags |= ((events & ERTS_POLL_EV_OUT) ? 0 : EV_DISABLE); + ERTS_EV_SET(&evts[len++], fd, EVFILT_WRITE, flags, (void *) ERTS_POLL_EV_OUT); + } else { + uint32_t flags; + ASSERT(op == ERTS_POLL_OP_MOD); + + flags = EV_DISPATCH; + flags |= (events & ERTS_POLL_EV_IN) ? EV_ENABLE : EV_DISABLE; + ERTS_EV_SET(&evts[len++], fd, EVFILT_READ, flags, (void *) ERTS_POLL_EV_IN); + + flags = EV_DISPATCH; + flags |= (events & ERTS_POLL_EV_OUT) ? EV_ENABLE : EV_DISABLE; + ERTS_EV_SET(&evts[len++], fd, EVFILT_WRITE, flags, (void *) ERTS_POLL_EV_OUT); + } #else -#define ERTS_POLL_MIN_BATCH_BUF_SIZE 64 -#endif + uint32_t flags = EV_ADD|EV_ONESHOT; -typedef struct { - int len; - int size; -#if ERTS_POLL_USE_DEVPOLL - struct pollfd *buf; -#elif ERTS_POLL_USE_KQUEUE - struct kevent *buf; - struct kevent *ebuf; + if (op == ERTS_POLL_OP_DEL) { + erts_atomic_dec_nob(&ps->no_of_user_fds); + /* We don't do anything when a delete is issued. The fds will be removed + when they are triggered, or when they are closed. */ + events = 0; + } else if (op == ERTS_POLL_OP_ADD) { + erts_atomic_inc_nob(&ps->no_of_user_fds); + } + + if (events & ERTS_POLL_EV_IN) { + ERTS_EV_SET(&evts[len++], fd, EVFILT_READ, flags, (void *) ERTS_POLL_EV_IN); + } + if (events & ERTS_POLL_EV_OUT) { + ERTS_EV_SET(&evts[len++], fd, EVFILT_WRITE, flags, (void *) ERTS_POLL_EV_OUT); + } + +#endif + if (len) + do { + res = kevent(ps->kp_fd, evts, len, NULL, 0, &ts); + } while (res < 0 && errno == EINTR); +#if ERTS_POLL_DEBUG_PRINT + { + int saved_errno = errno, i; + char keventb[255], *keventbp = keventb; + if (res < 0) + keventbp += sprintf(keventbp,"%s = ",erl_errno_id(saved_errno)); + else + keventbp += sprintf(keventbp,"%d = ",res); + keventbp += sprintf(keventbp, "kevent(%d, {",ps->kp_fd); + for (i = 0; i < len; i++) { + const char *flags = "UNKNOWN"; + if (evts[i].flags == EV_DELETE) flags = "EV_DELETE"; + if (evts[i].flags == (EV_ADD|EV_ONESHOT)) flags = "EV_ADD|EV_ONESHOT"; +#ifdef EV_DISPATCH + if (evts[i].flags == (EV_ADD|EV_DISPATCH)) flags = "EV_ADD|EV_DISPATCH"; + if (evts[i].flags == (EV_ADD|EV_DISABLE)) flags = "EV_ADD|EV_DISABLE"; + if (evts[i].flags == (EV_ENABLE|EV_DISPATCH)) flags = "EV_ENABLE|EV_DISPATCH"; + if (evts[i].flags == EV_DISABLE) flags = "EV_DISABLE"; + if (evts[i].flags == (EV_DISABLE|EV_DISPATCH)) flags = "EV_DISABLE|EV_DISABLE"; +#endif + + keventbp += sprintf(keventbp, "%s{%lu, %s, %s}",i > 0 ? ", " : "", + evts[i].ident, + (evts[i].filter == EVFILT_READ + ? "EVFILT_READ" + : (evts[i].filter == EVFILT_WRITE + ? "EVFILT_WRITE" + : "UNKNOWN")), flags); + } + keventbp += sprintf(keventbp, "}, %d)", len); + DEBUG_PRINT_FD("%s", ps, fd, keventb); + errno = saved_errno; + } #endif -} ErtsPollBatchBuf; + if (res < 0) { + if (op != ERTS_POLL_OP_DEL) { +#ifdef EV_RECEIPT + struct kevent receipt_evts[2]; + len = 0; + ERTS_EV_SET(&evts[len++], fd, EVFILT_WRITE, EV_DELETE|EV_RECEIPT, (void *) 0); + ERTS_EV_SET(&evts[len++], fd, EVFILT_READ, EV_DELETE|EV_RECEIPT, (void *) 0); + do { + res = kevent(ps->kp_fd, evts, len, receipt_evts, 2, &ts); + } while (res < 0 && errno == EINTR); +#else + ERTS_EV_SET(&evts[0], fd, EVFILT_WRITE, EV_DELETE, (void *) 0); + do { + res = kevent(ps->kp_fd, evts, 1, NULL, 0, &ts); + } while (res < 0 && errno == EINTR); + ERTS_EV_SET(&evts[0], fd, EVFILT_READ, EV_DELETE, (void *) 0); + do { + res = kevent(ps->kp_fd, evts, 1, NULL, 0, &ts); + } while (res < 0 && errno == EINTR); +#endif + if (op == ERTS_POLL_OP_ADD) + erts_atomic_dec_nob(&ps->no_of_user_fds); + return ERTS_POLL_EV_NVAL; + } + return 0; + } + return events; +} + +#endif /* ERTS_POLL_USE_KQUEUE */ +#if !ERTS_POLL_USE_CONCURRENT_UPDATE static ERTS_INLINE void -setup_batch_buf(ErtsPollSet ps, ErtsPollBatchBuf *bbp) +init_batch_update(ErtsPollSet *ps, int len) { - bbp->len = 0; #if ERTS_POLL_USE_DEVPOLL - bbp->size = ps->res_events_len; - bbp->buf = ps->res_events; -#elif ERTS_POLL_USE_KQUEUE - bbp->size = ps->res_events_len/2; - bbp->buf = ps->res_events; - bbp->ebuf = bbp->buf + bbp->size; + ASSERT(ps->poll_fds == NULL); + ps->poll_fds = erts_alloc(ERTS_ALC_T_TMP, sizeof(struct pollfd) * len); + ps->poll_fds_ix = 0; #endif } - -#if ERTS_POLL_USE_DEVPOLL - -static void -write_batch_buf(ErtsPollSet ps, ErtsPollBatchBuf *bbp) +static ERTS_INLINE void +write_batch_update(ErtsPollSet *ps) { +#if ERTS_POLL_USE_DEVPOLL ssize_t wres; - char *buf = (char *) bbp->buf; - size_t buf_size = sizeof(struct pollfd)*bbp->len; - + char *buf = (char *) ps->poll_fds; + size_t buf_size = sizeof(struct pollfd)*ps->poll_fds_ix; + while (1) { wres = write(ps->kp_fd, (void *) buf, buf_size); if (wres < 0) { @@ -845,9 +933,33 @@ write_batch_buf(ErtsPollSet ps, ErtsPollBatchBuf *bbp) __FILE__, __LINE__, erl_errno_id(errno), errno); } - buf_size -= wres; - if (buf_size <= 0) - break; +#if ERTS_POLL_DEBUG_PRINT + { + int saved_errno = errno, i; + char devpollb[2048], *devpollbp = devpollb; + devpollbp += sprintf(devpollbp, "%d = devpoll(%d, {", wres, ps->kp_fd); + for (i = 0; i < wres / sizeof(struct pollfd); i++) { + if (devpollbp == devpollb) + devpollbp += sprintf(devpollbp, "%d = devpoll(%d, {", wres, ps->kp_fd); + devpollbp += sprintf(devpollbp, "%s{fd = %d, events = %s}", + i > 0 ? ", " : "", + ps->poll_fds[i].fd, + ev2str(ps->poll_fds[i].events)); + if (devpollbp - devpollb > 512) { + devpollbp += sprintf(devpollbp, "}, %d)", ps->poll_fds_ix); + DEBUG_PRINT("%s", ps, devpollb); + devpollbp = devpollb; + } + } + devpollbp += sprintf(devpollbp, "}, %d)", ps->poll_fds_ix); + DEBUG_PRINT("%s", ps, devpollb); + errno = saved_errno; + } +#endif + + buf_size -= wres; + if (buf_size <= 0) + break; buf += wres; } @@ -855,483 +967,54 @@ write_batch_buf(ErtsPollSet ps, ErtsPollBatchBuf *bbp) fatal_error("%s:%d:write_devpoll_buf(): Internal error\n", __FILE__, __LINE__); } - bbp->len = 0; -} - -#elif ERTS_POLL_USE_KQUEUE - -static void -write_batch_buf(ErtsPollSet ps, ErtsPollBatchBuf *bbp) -{ - int res; - int len = bbp->len; - struct kevent *buf = bbp->buf; - struct timespec ts = {0, 0}; - - do { - res = kevent(ps->kp_fd, buf, len, NULL, 0, &ts); - } while (res < 0 && errno == EINTR); - if (res < 0) { - int i; - struct kevent *ebuf = bbp->ebuf; - do { - res = kevent(ps->kp_fd, buf, len, ebuf, len, &ts); - } while (res < 0 && errno == EINTR); - if (res < 0) { - fatal_error("%s:%d: kevent() failed: %s (%d)\n", - __FILE__, __LINE__, erl_errno_id(errno), errno); - } - for (i = 0; i < res; i++) { - if (ebuf[i].flags & EV_ERROR) { - short filter; - int fd = (int) ebuf[i].ident; - - switch ((int) (long) ebuf[i].udata) { - - /* - * Since we use a lazy update approach EV_DELETE will - * frequently fail. This since kqueue automatically - * removes a file descriptor that is closed from the - * poll set. - */ - case ERTS_POLL_KQ_OP_DEL_R: - case ERTS_POLL_KQ_OP_DEL_W: - case ERTS_POLL_KQ_OP_HANDLED: - break; - - /* - * According to the kqueue man page EVFILT_READ support - * does not imply EVFILT_WRITE support; therefore, - * if an EV_ADD fail, we may have to remove other - * events on this fd in the kqueue pollset before - * adding fd to the fallback pollset. - */ - case ERTS_POLL_KQ_OP_ADD_W: - if (ps->fds_status[fd].used_events & ERTS_POLL_EV_IN) { - filter = EVFILT_READ; - goto rm_add_fb; - } - goto add_fb; - case ERTS_POLL_KQ_OP_ADD_R: - if (ps->fds_status[fd].used_events & ERTS_POLL_EV_OUT) { - filter = EVFILT_WRITE; - goto rm_add_fb; - } - goto add_fb; - case ERTS_POLL_KQ_OP_ADD2_W: - case ERTS_POLL_KQ_OP_ADD2_R: { - int j; - for (j = i+1; j < res; j++) { - if (fd == (int) ebuf[j].ident) { - ebuf[j].udata = (void *) ERTS_POLL_KQ_OP_HANDLED; - if (!(ebuf[j].flags & EV_ERROR)) { - switch ((int) (long) ebuf[j].udata) { - case ERTS_POLL_KQ_OP_ADD2_W: - filter = EVFILT_WRITE; - goto rm_add_fb; - case ERTS_POLL_KQ_OP_ADD2_R: - filter = EVFILT_READ; - goto rm_add_fb; - default: - fatal_error("%s:%d:write_batch_buf(): " - "Internal error", - __FILE__, __LINE__); - break; - } - } - goto add_fb; - } - } - /* The other add succeded... */ - filter = ((((int) (long) ebuf[i].udata) - == ERTS_POLL_KQ_OP_ADD2_W) - ? EVFILT_READ - : EVFILT_WRITE); - rm_add_fb: - { - struct kevent kev; - struct timespec ts = {0, 0}; - EV_SET(&kev, fd, filter, EV_DELETE, 0, 0, 0); - (void) kevent(ps->kp_fd, &kev, 1, NULL, 0, &ts); - } - - add_fb: - ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_USEFLBCK; - ASSERT(ps->fds_status[fd].used_events); - ps->fds_status[fd].used_events = 0; - erts_atomic_dec_nob(&ps->no_of_user_fds); - update_fallback_pollset(ps, fd); - ASSERT(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK); - break; - } - default: - fatal_error("%s:%d:write_batch_buf(): Internal error", - __FILE__, __LINE__); - break; - } - } - } - } - bbp->len = 0; -} - -#endif /* ERTS_POLL_USE_KQUEUE */ - -static ERTS_INLINE void -batch_update_pollset(ErtsPollSet ps, int fd, ErtsPollBatchBuf *bbp) -{ - int buf_len; -#if ERTS_POLL_USE_DEVPOLL - short events; - struct pollfd *buf; -#elif ERTS_POLL_USE_KQUEUE - struct kevent *buf; -#endif - -#ifdef ERTS_POLL_DEBUG_PRINT - erts_printf("Doing lazy update on fd=%d\n", fd); -#endif - - if (!need_update(ps, fd)) - return; - - /* Make sure we have room for at least maximum no of entries - per fd */ - if (bbp->size - bbp->len < 2) - write_batch_buf(ps, bbp); - - buf_len = bbp->len; - buf = bbp->buf; - - ASSERT(fd < ps->fds_status_len); - -#if ERTS_POLL_USE_DEVPOLL - events = ERTS_POLL_EV_E2N(ps->fds_status[fd].events); - if (!events) { - buf[buf_len].events = POLLREMOVE; - erts_atomic_dec_nob(&ps->no_of_user_fds); - } - else if (!ps->fds_status[fd].used_events) { - buf[buf_len].events = events; - erts_atomic_inc_nob(&ps->no_of_user_fds); - } - else { - if ((ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST) - || (ps->fds_status[fd].used_events & ~events)) { - /* Reset or removed events... */ - buf[buf_len].fd = fd; - buf[buf_len].events = POLLREMOVE; - buf[buf_len++].revents = 0; - } - buf[buf_len].events = events; - } - buf[buf_len].fd = fd; - buf[buf_len++].revents = 0; - -#elif ERTS_POLL_USE_KQUEUE - - if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK) { - if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_USEFLBCK) - update_fallback_pollset(ps, fd); - else { /* Remove from fallback and try kqueue */ - ErtsPollEvents events = ps->fds_status[fd].events; - ps->fds_status[fd].events = (ErtsPollEvents) 0; - update_fallback_pollset(ps, fd); - ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); - if (events) { - ps->fds_status[fd].events = events; - goto try_kqueue; - } - } - } - else { - ErtsPollEvents events, used_events; - int mod_w, mod_r; - try_kqueue: - events = ERTS_POLL_EV_E2N(ps->fds_status[fd].events); - used_events = ERTS_POLL_EV_E2N(ps->fds_status[fd].used_events); - if (!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST)) { - if (!used_events && - (events & ERTS_POLL_EV_IN) && (events & ERTS_POLL_EV_OUT)) - goto do_add_rw; - mod_r = ((events & ERTS_POLL_EV_IN) - != (used_events & ERTS_POLL_EV_IN)); - mod_w = ((events & ERTS_POLL_EV_OUT) - != (used_events & ERTS_POLL_EV_OUT)); - goto do_mod; - } - else { /* Reset */ - if ((events & ERTS_POLL_EV_IN) && (events & ERTS_POLL_EV_OUT)) { - do_add_rw: - EV_SET(&buf[buf_len], fd, EVFILT_READ, EV_ADD, - 0, 0, (void *) ERTS_POLL_KQ_OP_ADD2_R); - buf_len++; - EV_SET(&buf[buf_len], fd, EVFILT_WRITE, EV_ADD, - 0, 0, (void *) ERTS_POLL_KQ_OP_ADD2_W); - buf_len++; - - } - else { - mod_r = 1; - mod_w = 1; - do_mod: - if (mod_r) { - if (events & ERTS_POLL_EV_IN) { - EV_SET(&buf[buf_len], fd, EVFILT_READ, EV_ADD, - 0, 0, (void *) ERTS_POLL_KQ_OP_ADD_R); - buf_len++; - } - else if (used_events & ERTS_POLL_EV_IN) { - EV_SET(&buf[buf_len], fd, EVFILT_READ, EV_DELETE, - 0, 0, (void *) ERTS_POLL_KQ_OP_DEL_R); - buf_len++; - } - } - if (mod_w) { - if (events & ERTS_POLL_EV_OUT) { - EV_SET(&buf[buf_len], fd, EVFILT_WRITE, EV_ADD, - 0, 0, (void *) ERTS_POLL_KQ_OP_ADD_W); - buf_len++; - } - else if (used_events & ERTS_POLL_EV_OUT) { - EV_SET(&buf[buf_len], fd, EVFILT_WRITE, EV_DELETE, - 0, 0, (void *) ERTS_POLL_KQ_OP_DEL_W); - buf_len++; - } - } - } - } - if (used_events) { - if (!events) { - erts_atomic_dec_nob(&ps->no_of_user_fds); - } - } - else { - if (events) - erts_atomic_inc_nob(&ps->no_of_user_fds); - } - ASSERT((events & ~(ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT)) == 0); - ASSERT((used_events & ~(ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT)) == 0); - } - + erts_free(ERTS_ALC_T_TMP, ps->poll_fds); + ps->poll_fds = NULL; #endif - - ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST; - ps->fds_status[fd].used_events = ps->fds_status[fd].events; - - bbp->len = buf_len; } -#else /* !ERTS_POLL_USE_BATCH_UPDATE_POLLSET */ - -#if ERTS_POLL_USE_EPOLL -static int -#if ERTS_POLL_USE_CONCURRENT_UPDATE -conc_update_pollset(ErtsPollSet ps, int fd, int *update_fallback) -#else -update_pollset(ErtsPollSet ps, int fd) -#endif +static ERTS_INLINE int +need_update(ErtsPollSet *ps, int fd, int *resetp) { - int res; - int op; - struct epoll_event epe_templ; - struct epoll_event epe; - + int reset; ASSERT(fd < ps->fds_status_len); - if (!need_update(ps, fd)) - return 0; - -#ifdef ERTS_POLL_DEBUG_PRINT - erts_printf("Doing update on fd=%d\n", fd); -#endif - if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK) { -#if ERTS_POLL_USE_CONCURRENT_UPDATE - if (!*update_fallback) { - *update_fallback = 1; - return 0; - } -#endif - if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_USEFLBCK) { - return update_fallback_pollset(ps, fd); - } - else { /* Remove from fallback and try epoll */ - ErtsPollEvents events = ps->fds_status[fd].events; - ps->fds_status[fd].events = (ErtsPollEvents) 0; - res = update_fallback_pollset(ps, fd); - ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); - if (!events) - return res; - ps->fds_status[fd].events = events; - } - } - - epe_templ.events = ERTS_POLL_EV_E2N(ps->fds_status[fd].events); - epe_templ.data.fd = fd; - -#ifdef VALGRIND - /* Silence invalid valgrind warning ... */ - memset((void *) &epe.data, 0, sizeof(epoll_data_t)); -#endif - - if (epe_templ.events && ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST) { - do { - /* We init 'epe' every time since epoll_ctl() may modify it - (not declared const and not documented as const). */ - epe.events = epe_templ.events; - epe.data.fd = epe_templ.data.fd; - res = epoll_ctl(ps->kp_fd, EPOLL_CTL_DEL, fd, &epe); - } while (res != 0 && errno == EINTR); - erts_atomic_dec_nob(&ps->no_of_user_fds); - ps->fds_status[fd].used_events = 0; - } + reset = (int) (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST); + ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST; - if (!epe_templ.events) { - /* A note on EPOLL_CTL_DEL: linux kernel versions before 2.6.9 - need a non-NULL event pointer even though it is ignored... */ - op = EPOLL_CTL_DEL; - erts_atomic_dec_nob(&ps->no_of_user_fds); - } - else if (!ps->fds_status[fd].used_events) { - op = EPOLL_CTL_ADD; - erts_atomic_inc_nob(&ps->no_of_user_fds); - } - else { - op = EPOLL_CTL_MOD; - } + *resetp = reset; - do { - /* We init 'epe' every time since epoll_ctl() may modify it - (not declared const and not documented as const). */ - epe.events = epe_templ.events; - epe.data.fd = epe_templ.data.fd; - res = epoll_ctl(ps->kp_fd, op, fd, &epe); - } while (res != 0 && errno == EINTR); + if (reset || ps->fds_status[fd].used_events != ps->fds_status[fd].events) + return 1; -#if defined(ERTS_POLL_DEBUG_PRINT) && 1 - { - int saved_errno = errno; - erts_printf("%s = epoll_ctl(%d, %s, %d, {Ox%x, %d})\n", - res == 0 ? "0" : erl_errno_id(errno), - ps->kp_fd, - (op == EPOLL_CTL_ADD - ? "EPOLL_CTL_ADD" - : (op == EPOLL_CTL_MOD - ? "EPOLL_CTL_MOD" - : (op == EPOLL_CTL_DEL - ? "EPOLL_CTL_DEL" - : "UNKNOWN"))), - fd, - epe_templ.events, - fd); - errno = saved_errno; - } -#endif - if (res == 0) - ps->fds_status[fd].used_events = ps->fds_status[fd].events; - else { - switch (op) { - case EPOLL_CTL_MOD: - epe.events = 0; - do { - /* We init 'epe' every time since epoll_ctl() may modify it - (not declared const and not documented as const). */ - epe.events = 0; - epe.data.fd = fd; - res = epoll_ctl(ps->kp_fd, EPOLL_CTL_DEL, fd, &epe); - } while (res != 0 && errno == EINTR); - ps->fds_status[fd].used_events = 0; - /* Fall through ... */ - case EPOLL_CTL_ADD: { - ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_USEFLBCK; - erts_atomic_dec_nob(&ps->no_of_user_fds); -#if ERTS_POLL_USE_CONCURRENT_UPDATE - if (!*update_fallback) { - *update_fallback = 1; - return 0; - } -#endif - ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); - res = update_fallback_pollset(ps, fd); - ASSERT(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK); - break; - } - case EPOLL_CTL_DEL: { - /* - * Since we use a lazy update approach EPOLL_CTL_DEL will - * frequently fail. This since epoll automatically removes - * a filedescriptor that is closed from the poll set. - */ - ps->fds_status[fd].used_events = 0; - res = 0; - break; - } - default: - fatal_error("%s:%d:update_pollset(): Internal error\n", - __FILE__, __LINE__); - break; - } - } - ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST; - return res; + return 0; } -#if ERTS_POLL_USE_CONCURRENT_UPDATE -static int -update_pollset(ErtsPollSet ps, int fd) +static int update_pollset(ErtsPollSet *ps, ErtsPollResFd pr[], int fd) { - int update_fallback = 1; - return conc_update_pollset(ps, fd, &update_fallback); -} -#endif - -#endif /* ERTS_POLL_USE_EPOLL */ - -#endif /* ERTS_POLL_USE_BATCH_UPDATE_POLLSET */ - -#if ERTS_POLL_USE_POLL || ERTS_POLL_USE_SELECT || ERTS_POLL_USE_FALLBACK - -#if ERTS_POLL_USE_FALLBACK -static int update_fallback_pollset(ErtsPollSet ps, int fd) -#else -static int update_pollset(ErtsPollSet ps, int fd) -#endif -{ -#ifdef ERTS_POLL_DEBUG_PRINT -#if ERTS_POLL_USE_FALLBACK - erts_printf("Doing fallback update on fd=%d\n", fd); -#else - erts_printf("Doing update on fd=%d\n", fd); -#endif -#endif - + int res = 0, reset = 0; + ErtsPollEvents events = ps->fds_status[fd].events; ASSERT(fd < ps->fds_status_len); -#if ERTS_POLL_USE_FALLBACK - ASSERT(ps->fds_status[fd].used_events - ? (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK) - : (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_USEFLBCK)); -#endif - if (!need_update(ps, fd)) - return 0; - -#if ERTS_POLL_USE_FALLBACK - ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST; -#endif + if (!need_update(ps, fd, &reset)) + return res; #if ERTS_POLL_USE_POLL /* --- poll -------------------------------- */ - if (!ps->fds_status[fd].events) { + if (!events) { int pix = ps->fds_status[fd].pix; int last_pix; + + if (reset) { + /* When a fd has been reset, we tell the caller of erts_poll_wait + this by setting the fd as ERTS_POLL_EV_NONE */ + ERTS_POLL_RES_SET_FD(&pr[res], fd); + ERTS_POLL_RES_SET_EVTS(&pr[res], ERTS_POLL_EV_NONE); + DEBUG_PRINT_FD("trig %s (poll)", ps, fd, ev2str(ERTS_POLL_EV_NONE)); + res++; + } + if (pix < 0) { -#if ERTS_POLL_USE_FALLBACK - ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); -#endif - return -1; + return res; } -#if ERTS_POLL_USE_FALLBACK - ASSERT(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK); -#endif erts_atomic_dec_nob(&ps->no_of_user_fds); last_pix = --ps->no_poll_fds; if (pix != last_pix) { @@ -1348,126 +1031,151 @@ static int update_pollset(ErtsPollSet ps, int fd) /* Clear this fd status */ ps->fds_status[fd].pix = -1; ps->fds_status[fd].used_events = (ErtsPollEvents) 0; -#if ERTS_POLL_USE_FALLBACK - ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_INFLBCK; -#endif + } else { int pix = ps->fds_status[fd].pix; if (pix < 0) { -#if ERTS_POLL_USE_FALLBACK - ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK) - || fd == ps->kp_fd); -#endif erts_atomic_inc_nob(&ps->no_of_user_fds); ps->fds_status[fd].pix = pix = ps->no_poll_fds++; if (pix >= ps->poll_fds_len) grow_poll_fds(ps, pix); ps->poll_fds[pix].fd = fd; ps->fds_status[fd].pix = pix; -#if ERTS_POLL_USE_FALLBACK - ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_INFLBCK; -#endif } -#if ERTS_POLL_USE_FALLBACK - ASSERT(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK); -#endif - /* Events to be used in next poll */ - ps->poll_fds[pix].events = ev2pollev(ps->fds_status[fd].events); + ps->poll_fds[pix].events = ev2pollev(events); if (ps->poll_fds[pix].revents) { /* Remove result events that we should not poll for anymore */ ps->poll_fds[pix].revents &= ev2pollev(~(~ps->fds_status[fd].used_events - & ps->fds_status[fd].events)); + & events)); } /* Save events to be used in next poll */ - ps->fds_status[fd].used_events = ps->fds_status[fd].events; + ps->fds_status[fd].used_events = events; } - return 0; + return res; #elif ERTS_POLL_USE_SELECT /* --- select ------------------------------ */ - { - ErtsPollEvents events = ps->fds_status[fd].events; + if (!events) { + + if (reset) { + /* When a fd has been reset, we tell the caller of erts_poll_wait + this by setting the fd as ERTS_POLL_EV_NONE */ + ERTS_POLL_RES_SET_FD(&pr[res], fd); + ERTS_POLL_RES_SET_EVTS(&pr[res], ERTS_POLL_EV_NONE); + DEBUG_PRINT_FD("trig %s (select)", ps, fd, ev2str(ERTS_POLL_EV_NONE)); + res++; + } + + ERTS_FD_CLR(fd, &ps->input_fds); + ERTS_FD_CLR(fd, &ps->output_fds); + + if (ps->fds_status[fd].used_events) { + erts_atomic_dec_nob(&ps->no_of_user_fds); + ps->fds_status[fd].used_events = (ErtsPollEvents) 0; + } + + if (fd == ps->max_fd) { + int max = ps->max_fd; + for (max = ps->max_fd; max >= 0; max--) + if (ps->fds_status[max].used_events) + break; + ps->max_fd = max; + } + + } else { + ensure_select_fds(fd, &ps->input_fds, &ps->output_fds); - if ((ERTS_POLL_EV_IN & events) - != (ERTS_POLL_EV_IN & ps->fds_status[fd].used_events)) { - if (ERTS_POLL_EV_IN & events) { - ERTS_FD_SET(fd, &ps->input_fds); - } - else { - ERTS_FD_CLR(fd, &ps->input_fds); - } - } - if ((ERTS_POLL_EV_OUT & events) - != (ERTS_POLL_EV_OUT & ps->fds_status[fd].used_events)) { - if (ERTS_POLL_EV_OUT & events) { - ERTS_FD_SET(fd, &ps->output_fds); - } - else { - ERTS_FD_CLR(fd, &ps->output_fds); - } - } - if (!ps->fds_status[fd].used_events) { - ASSERT(events); - erts_atomic_inc_nob(&ps->no_of_user_fds); -#if ERTS_POLL_USE_FALLBACK - ps->no_select_fds++; - ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_INFLBCK; -#endif - } - else if (!events) { - ASSERT(ps->fds_status[fd].used_events); - erts_atomic_dec_nob(&ps->no_of_user_fds); - ps->fds_status[fd].events = events; -#if ERTS_POLL_USE_FALLBACK - ps->no_select_fds--; - ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_INFLBCK; -#endif - } + if (!ps->fds_status[fd].used_events) + erts_atomic_inc_nob(&ps->no_of_user_fds); + + if (events & ERTS_POLL_EV_IN) + ERTS_FD_SET(fd, &ps->input_fds); + else + ERTS_FD_CLR(fd, &ps->input_fds); + + if (events & ERTS_POLL_EV_OUT) + ERTS_FD_SET(fd, &ps->output_fds); + else + ERTS_FD_CLR(fd, &ps->output_fds); ps->fds_status[fd].used_events = events; - if (events && fd > ps->max_fd) - ps->max_fd = fd; - else if (!events && fd == ps->max_fd) { - int max = ps->max_fd; - for (max = ps->max_fd; max >= 0; max--) - if (ps->fds_status[max].used_events) - break; - ps->max_fd = max; - } + if (fd > ps->max_fd) + ps->max_fd = fd; } - return 0; -#endif -} -#endif /* ERTS_POLL_USE_POLL || ERTS_POLL_USE_SELECT || ERTS_POLL_USE_FALLBACK */ + return res; +#elif ERTS_POLL_USE_DEVPOLL + + if (!events) { + if (reset) { + /* When a fd has been reset, we tell the caller of erts_poll_wait + this by setting the fd as ERTS_POLL_EV_NONE */ + ERTS_POLL_RES_SET_FD(&pr[res], fd); + ERTS_POLL_RES_SET_EVTS(&pr[res], ERTS_POLL_EV_NONE); + DEBUG_PRINT_FD("trig %s (devpoll)", ps, fd, ev2str(ERTS_POLL_EV_NONE)); + res++; + } -static void -handle_update_requests(ErtsPollSet ps) -{ - ErtsPollSetUpdateRequestsBlock *urqbp = &ps->update_requests; -#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET - ErtsPollBatchBuf bb; - setup_batch_buf(ps, &bb); + ps->poll_fds[ps->poll_fds_ix].fd = fd; + ps->poll_fds[ps->poll_fds_ix].revents = 0; + ps->poll_fds[ps->poll_fds_ix++].events = POLLREMOVE; + + if (ps->fds_status[fd].used_events) { + erts_atomic_dec_nob(&ps->no_of_user_fds); + ps->fds_status[fd].used_events = 0; + } + + } else { + if (!ps->fds_status[fd].used_events) { + erts_atomic_inc_nob(&ps->no_of_user_fds); + } + ps->poll_fds[ps->poll_fds_ix].fd = fd; + ps->poll_fds[ps->poll_fds_ix].revents = 0; + ps->poll_fds[ps->poll_fds_ix++].events = ERTS_POLL_EV_E2N(events); + ps->fds_status[fd].used_events = ps->fds_status[fd].events; + } + + return res; #endif +} + +static int +handle_update_requests(ErtsPollSet *ps, ErtsPollResFd pr[], int no_fds) +{ + int res = 0; + ErtsPollSetUpdateRequestsBlock *urqbp = ps->curr_upd_req_block; while (urqbp) { ErtsPollSetUpdateRequestsBlock *free_urqbp = urqbp; int i; int len = urqbp->len; + + init_batch_update(ps, len); + for (i = 0; i < len; i++) { int fd = urqbp->fds[i]; ASSERT(fd < ps->fds_status_len); - ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_INURQ; -#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET - batch_update_pollset(ps, fd, &bb); -#else - update_pollset(ps, fd); -#endif + ASSERT(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INURQ); + + /* We have run out of PollResFd slots to put results in, + so we yield here and return later for more. */ + if (res == no_fds && pr != NULL) { + memmove(urqbp->fds, urqbp->fds+i, sizeof(int) * (len - i)); + urqbp->len -= i; + ps->curr_upd_req_block = urqbp; + write_batch_update(ps); + return res; + } + + if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INURQ) { + ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_INURQ; + res += update_pollset(ps, pr + res, fd); + } } free_urqbp = urqbp; @@ -1475,12 +1183,9 @@ handle_update_requests(ErtsPollSet ps) free_update_requests_block(ps, free_urqbp); - } + write_batch_update(ps); -#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET - if (bb.len) - write_batch_buf(ps, &bb); -#endif + } ps->curr_upd_req_block = &ps->update_requests; @@ -1489,16 +1194,19 @@ handle_update_requests(ErtsPollSet ps) #endif ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(ps); + return res; } +#endif /* !ERTS_POLL_USE_CONCURRENT_UPDATE */ static ERTS_INLINE ErtsPollEvents -poll_control(ErtsPollSet ps, int fd, ErtsPollEvents events, int on, int *do_wake) +poll_control(ErtsPollSet *ps, int fd, ErtsPollOp op, + ErtsPollEvents events, int *do_wake) { ErtsPollEvents new_events; if (fd < ps->internal_fd_limit || fd >= max_fds) { - if (fd < 0) { + if (fd < 0 || fd >= max_fds) { new_events = ERTS_POLL_EV_ERR; goto done; } @@ -1508,115 +1216,55 @@ poll_control(ErtsPollSet ps, int fd, ErtsPollEvents events, int on, int *do_wake goto done; } #endif +#if !ERTS_POLL_USE_CONCURRENT_UPDATE if (fd == ps->wake_fds[0] || fd == ps->wake_fds[1]) { new_events = ERTS_POLL_EV_NVAL; goto done; } -#if ERTS_POLL_USE_TIMERFD - if (fd == ps->timer_fd) { - new_events = ERTS_POLL_EV_NVAL; - goto done; - } #endif } +#if ERTS_POLL_USE_CONCURRENT_UPDATE + + new_events = update_pollset(ps, fd, op, events); + +#else /* !ERTS_POLL_USE_CONCURRENT_UPDATE */ if (fd >= ps->fds_status_len) grow_fds_status(ps, fd); ASSERT(fd < ps->fds_status_len); - new_events = ps->fds_status[fd].events; - - if (events == 0) { - *do_wake = 0; - goto done; - } - - if (on) - new_events |= events; - else - new_events &= ~events; - - if (new_events == (ErtsPollEvents) 0) { - ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_RST; -#if ERTS_POLL_USE_FALLBACK - ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_USEFLBCK; -#endif - } - - ps->fds_status[fd].events = new_events; - - if (new_events == ps->fds_status[fd].used_events - && !(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST) - ) { - *do_wake = 0; - goto done; - } - - -#if ERTS_POLL_USE_CONCURRENT_UPDATE - if (ERTS_POLLSET_IS_POLLED(ps)) { - int update_fallback = 0; - conc_update_pollset(ps, fd, &update_fallback); - if (!update_fallback) { - *do_wake = 0; /* no need to wake kernel poller */ - goto done; - } + if (op == ERTS_POLL_OP_DEL) { + ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_RST; + ps->fds_status[fd].events = 0; + *do_wake = 1; + } else if (op == ERTS_POLL_OP_ADD) { + ASSERT(ps->fds_status[fd].events == 0); + ps->fds_status[fd].events = events; + *do_wake = 1; + } else { + ASSERT(op == ERTS_POLL_OP_MOD); + ps->fds_status[fd].events = events; + *do_wake = 1; } -#endif + new_events = ps->fds_status[fd].events; enqueue_update_request(ps, fd); - - /* - * If new events have been added, we need to wake up the - * polling thread, but if events have been removed we don't. - */ - if ((new_events && (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST)) - || (~ps->fds_status[fd].used_events & new_events)) - *do_wake = 1; +#endif /* !ERTS_POLL_USE_CONCURRENT_UPDATE */ done: -#ifdef ERTS_POLL_DEBUG_PRINT - erts_printf("0x%x = poll_control(ps, %d, 0x%x, %s) do_wake=%d\n", - (int) new_events, fd, (int) events, (on ? "on" : "off"), *do_wake); -#endif + DEBUG_PRINT_FD("%s = %s(%p, %d, %s, %s) do_wake=%d", + ps, fd, ev2str(new_events), __FUNCTION__, ps, + fd, op2str(op), ev2str(events), *do_wake); return new_events; } -void -ERTS_POLL_EXPORT(erts_poll_controlv)(ErtsPollSet ps, - ErtsPollControlEntry pcev[], - int len) -{ - int i; - int do_wake; - int final_do_wake = 0; - - ERTS_POLLSET_LOCK(ps); - - for (i = 0; i < len; i++) { - do_wake = 0; - pcev[i].events = poll_control(ps, - pcev[i].fd, - pcev[i].events, - pcev[i].on, - &do_wake); - final_do_wake |= do_wake; - } - - ERTS_POLLSET_UNLOCK(ps); - - if (final_do_wake) - wake_poller(ps, 0, 0); - -} - ErtsPollEvents -ERTS_POLL_EXPORT(erts_poll_control)(ErtsPollSet ps, +ERTS_POLL_EXPORT(erts_poll_control)(ErtsPollSet *ps, ErtsSysFdType fd, + ErtsPollOp op, ErtsPollEvents events, - int on, int* do_wake) /* In: Wake up polling thread */ /* Out: Poller is woken */ { @@ -1624,13 +1272,15 @@ ERTS_POLL_EXPORT(erts_poll_control)(ErtsPollSet ps, ERTS_POLLSET_LOCK(ps); - res = poll_control(ps, fd, events, on, do_wake); + res = poll_control(ps, fd, op, events, do_wake); ERTS_POLLSET_UNLOCK(ps); +#if !ERTS_POLL_USE_CONCURRENT_UPDATE if (*do_wake) { - wake_poller(ps, 0, 0); + wake_poller(ps, 0); } +#endif return res; } @@ -1642,180 +1292,60 @@ ERTS_POLL_EXPORT(erts_poll_control)(ErtsPollSet ps, #if ERTS_POLL_USE_KERNEL_POLL static ERTS_INLINE int -save_kp_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, int chk_fds_res) +ERTS_POLL_EXPORT(save_result)(ErtsPollSet *ps, ErtsPollResFd pr[], int max_res, int chk_fds_res, int ebadf) { - int res = 0; - int i; - int n = chk_fds_res < max_res ? chk_fds_res : max_res; +#if !ERTS_POLL_USE_CONCURRENT_UPDATE || ERTS_POLL_DEBUG_PRINT + int n = chk_fds_res < max_res ? chk_fds_res : max_res, i; + int res = n; +#if !ERTS_POLL_USE_CONCURRENT_UPDATE int wake_fd = ps->wake_fds[0]; -#if ERTS_POLL_USE_TIMERFD - int timer_fd = ps->timer_fd; #endif for (i = 0; i < n; i++) { - -#if ERTS_POLL_USE_EPOLL /* --- epoll ------------------------------- */ - - if (ps->res_events[i].events) { - int fd = ps->res_events[i].data.fd; - int ix; - ErtsPollEvents revents; - if (fd == wake_fd) { - cleanup_wakeup_pipe(ps); - continue; - } -#if ERTS_POLL_USE_TIMERFD - if (fd == timer_fd) { - continue; - } + int fd = ERTS_POLL_RES_GET_FD(&pr[i]); +#ifdef DEBUG_PRINT_MODE + ErtsPollEvents evts = ERTS_POLL_RES_GET_EVTS(pr+i); #endif - ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); - /* epoll_wait() can repeat the same fd in result array... */ - ix = (int) ps->fds_status[fd].res_ev_ix; - ASSERT(ix >= 0); - if (ix >= res || pr[ix].fd != fd) { - ix = res; - pr[ix].fd = fd; - pr[ix].events = (ErtsPollEvents) 0; - } - - revents = ERTS_POLL_EV_N2E(ps->res_events[i].events); - pr[ix].events |= revents; - if (revents) { - if (res == ix) { - ps->fds_status[fd].res_ev_ix = (unsigned short) ix; - res++; - } - } - } - -#elif ERTS_POLL_USE_KQUEUE /* --- kqueue ------------------------------ */ - - struct kevent *ev; - int fd; - int ix; - - ev = &ps->res_events[i]; - fd = (int) ev->ident; - ASSERT(fd < ps->fds_status_len); - ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); - ix = (int) ps->fds_status[fd].res_ev_ix; - ASSERT(ix >= 0); - if (ix >= res || pr[ix].fd != fd) { - ix = res; - pr[ix].fd = (int) ev->ident; - pr[ix].events = (ErtsPollEvents) 0; - } - - if (ev->filter == EVFILT_READ) { - if (fd == wake_fd) { - cleanup_wakeup_pipe(ps); - continue; - } - pr[ix].events |= ERTS_POLL_EV_IN; - } - else if (ev->filter == EVFILT_WRITE) - pr[ix].events |= ERTS_POLL_EV_OUT; - if (ev->flags & (EV_ERROR|EV_EOF)) { - if ((ev->flags & EV_ERROR) && (((int) ev->data) == EBADF)) - pr[ix].events |= ERTS_POLL_EV_NVAL; - else - pr[ix].events |= ERTS_POLL_EV_ERR; - } - if (pr[ix].events) { - if (res == ix) { - ps->fds_status[fd].res_ev_ix = (unsigned short) ix; - res++; - } - } - -#elif ERTS_POLL_USE_DEVPOLL /* --- devpoll ----------------------------- */ - - if (ps->res_events[i].revents) { - int fd = ps->res_events[i].fd; - ErtsPollEvents revents; - if (fd == wake_fd) { - cleanup_wakeup_pipe(ps); - continue; - } -#if ERTS_POLL_USE_TIMERFD - if (fd == timer_fd) { - continue; - } -#endif - revents = ERTS_POLL_EV_N2E(ps->res_events[i].events); - pr[res].fd = fd; - pr[res].events = revents; - res++; - } - -#endif - - } - - return res; -} - -#endif /* ERTS_POLL_USE_KERNEL_POLL */ - -#if ERTS_POLL_USE_FALLBACK - -static int -get_kp_results(ErtsPollSet ps, ErtsPollResFd pr[], int max_res) -{ - int res; + DEBUG_PRINT_FD("trig %s (%s)", ps, fd, + ev2str(evts), #if ERTS_POLL_USE_KQUEUE - struct timespec ts = {0, 0}; + "kqueue" +#elif ERTS_POLL_USE_EPOLL + "epoll" +#else + "/dev/poll" #endif + ); - if (max_res > ps->res_events_len) - grow_res_events(ps, max_res); +#if !ERTS_POLL_USE_CONCURRENT_UPDATE - do { -#if ERTS_POLL_USE_EPOLL - res = epoll_wait(ps->kp_fd, ps->res_events, max_res, 0); -#elif ERTS_POLL_USE_KQUEUE - res = kevent(ps->kp_fd, NULL, 0, ps->res_events, max_res, &ts); + if (fd == wake_fd) { + cleanup_wakeup_pipe(ps); + ERTS_POLL_RES_SET_EVTS(&pr[i], ERTS_POLL_EV_NONE); + } else { + /* Reset the events to emulate ONESHOT semantics */ + ps->fds_status[fd].events = 0; + enqueue_update_request(ps, fd); + } #endif - } while (res < 0 && errno == EINTR); - - if (res < 0) { - fatal_error("%s:%d: %s() failed: %s (%d)\n", - __FILE__, __LINE__, -#if ERTS_POLL_USE_EPOLL - "epoll_wait", -#elif ERTS_POLL_USE_KQUEUE - "kevent", -#endif - erl_errno_id(errno), errno); } - return save_kp_result(ps, pr, max_res, res); + return res; +#else + ASSERT(chk_fds_res <= max_res); + return chk_fds_res; +#endif } -#endif /* ERTS_POLL_USE_FALLBACK */ - - +#else /* !ERTS_POLL_USE_KERNEL_POLL */ static ERTS_INLINE int -save_poll_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, - int chk_fds_res, int ebadf) +ERTS_POLL_EXPORT(save_result)(ErtsPollSet *ps, ErtsPollResFd pr[], int max_res, int chk_fds_res, int ebadf) { -#if ERTS_POLL_USE_DEVPOLL - return save_kp_result(ps, pr, max_res, chk_fds_res); -#elif ERTS_POLL_USE_FALLBACK - if (!ps->fallback_used) - return save_kp_result(ps, pr, max_res, chk_fds_res); - else -#endif /* ERTS_POLL_USE_FALLBACK */ - { - #if ERTS_POLL_USE_POLL /* --- poll -------------------------------- */ int res = 0; -#if !ERTS_POLL_USE_FALLBACK int wake_fd = ps->wake_fds[0]; -#endif int i, first_ix, end_ix; /* @@ -1832,23 +1362,30 @@ save_poll_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, if (ps->poll_fds[i].revents != (short) 0) { int fd = ps->poll_fds[i].fd; ErtsPollEvents revents; -#if ERTS_POLL_USE_FALLBACK - if (fd == ps->kp_fd) { - res += get_kp_results(ps, &pr[res], max_res-res); - i++; - continue; - } -#else if (fd == wake_fd) { cleanup_wakeup_pipe(ps); i++; continue; } -#endif revents = pollev2ev(ps->poll_fds[i].revents); - pr[res].fd = fd; - pr[res].events = revents; + ERTS_POLL_RES_SET_FD(&pr[res], fd); + ERTS_POLL_RES_SET_EVTS(&pr[res], revents); + + /* If an fd returns as error, we may want to check the + update_requests queue to see if it has been reset + before delivering the result?!?! This should allow + the user to do driver_dselect + close without waiting + for stop_select... */ + + DEBUG_PRINT_FD("trig %s (poll)", ps, ERTS_POLL_RES_GET_FD(&pr[res]), + ev2str(ERTS_POLL_RES_GET_EVTS(&pr[res]))); + res++; + + /* Clear the events for this fd in order to mimic + how epoll ONESHOT works */ + ps->fds_status[fd].events = 0; + enqueue_update_request(ps, fd); } i++; } @@ -1864,9 +1401,7 @@ save_poll_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, #elif ERTS_POLL_USE_SELECT /* --- select ------------------------------ */ int res = 0; -#if !ERTS_POLL_USE_FALLBACK int wake_fd = ps->wake_fds[0]; -#endif int fd, first_fd, end_fd; /* @@ -1879,29 +1414,23 @@ save_poll_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, if (!ebadf) { while (1) { while (fd < end_fd && res < max_res) { - - pr[res].events = (ErtsPollEvents) 0; + ErtsPollEvents events = 0; if (ERTS_FD_ISSET(fd, &ps->res_input_fds)) { -#if ERTS_POLL_USE_FALLBACK - if (fd == ps->kp_fd) { - res += get_kp_results(ps, &pr[res], max_res-res); - fd++; - continue; - } -#else if (fd == wake_fd) { cleanup_wakeup_pipe(ps); fd++; continue; } -#endif - pr[res].events |= ERTS_POLL_EV_IN; + events |= ERTS_POLL_EV_IN; } if (ERTS_FD_ISSET(fd, &ps->res_output_fds)) - pr[res].events |= ERTS_POLL_EV_OUT; - if (pr[res].events) { - pr[res].fd = fd; + events |= ERTS_POLL_EV_OUT; + if (events) { + ERTS_POLL_RES_SET_FD(&pr[res], fd); + ERTS_POLL_RES_SET_EVTS(&pr[res], events); res++; + ps->fds_status[fd].events = 0; + enqueue_update_request(ps, fd); } fd++; } @@ -1934,7 +1463,7 @@ save_poll_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, if (ps->fds_status[fd].events & ERTS_POLL_EV_OUT) { oset = &ps->res_output_fds; ERTS_FD_ZERO(oset); - ERTS_FD_SET(fd, oset); + ERTS_FD_SET(fd, oset); } do { /* Initiate 'tv' each time; @@ -1943,49 +1472,31 @@ save_poll_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, sres = ERTS_SELECT(ps->max_fd+1, iset, oset, NULL, &tv); } while (sres < 0 && errno == EINTR); if (sres < 0) { -#if ERTS_POLL_USE_FALLBACK - if (fd == ps->kp_fd) { - res += get_kp_results(ps, - &pr[res], - max_res-res); - fd++; - continue; - } -#else if (fd == wake_fd) { cleanup_wakeup_pipe(ps); fd++; continue; } -#endif - pr[res].fd = fd; - pr[res].events = ERTS_POLL_EV_NVAL; + ERTS_POLL_RES_SET_FD(&pr[res], fd); + ERTS_POLL_RES_SET_EVTS(&pr[res], ERTS_POLL_EV_NVAL); res++; } else if (sres > 0) { - pr[res].fd = fd; + ErtsPollEvents events = 0; + ERTS_POLL_RES_SET_FD(&pr[res], fd); if (iset && ERTS_FD_ISSET(fd, iset)) { -#if ERTS_POLL_USE_FALLBACK - if (fd == ps->kp_fd) { - res += get_kp_results(ps, - &pr[res], - max_res-res); - fd++; - continue; - } -#else if (fd == wake_fd) { cleanup_wakeup_pipe(ps); fd++; continue; } -#endif - pr[res].events |= ERTS_POLL_EV_IN; + events |= ERTS_POLL_EV_IN; } if (oset && ERTS_FD_ISSET(fd, oset)) { - pr[res].events |= ERTS_POLL_EV_OUT; + events |= ERTS_POLL_EV_OUT; } - ASSERT(pr[res].events); + ASSERT(events); + ERTS_POLL_RES_SET_EVTS(&pr[res], events); res++; } } @@ -2000,353 +1511,145 @@ save_poll_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, } ps->next_sel_fd = fd; return res; -#endif - } -} - -static ERTS_INLINE ErtsMonotonicTime -get_timeout(ErtsPollSet ps, - int resolution, - ErtsMonotonicTime timeout_time) -{ - ErtsMonotonicTime timeout, save_timeout_time; - - if (timeout_time == ERTS_POLL_NO_TIMEOUT) { - save_timeout_time = ERTS_MONOTONIC_TIME_MIN; - timeout = 0; - } - else { - ErtsMonotonicTime diff_time, current_time; - current_time = erts_get_monotonic_time(NULL); - diff_time = timeout_time - current_time; - if (diff_time <= 0) { - save_timeout_time = ERTS_MONOTONIC_TIME_MIN; - timeout = 0; - } - else { - save_timeout_time = current_time; - switch (resolution) { - case 1000: - /* Round up to nearest even milli second */ - timeout = ERTS_MONOTONIC_TO_MSEC(diff_time - 1) + 1; - if (timeout > (ErtsMonotonicTime) INT_MAX) - timeout = (ErtsMonotonicTime) INT_MAX; - save_timeout_time += ERTS_MSEC_TO_MONOTONIC(timeout); - timeout -= ERTS_PREMATURE_TIMEOUT(timeout, 1000); - break; - case 1000000: - /* Round up to nearest even micro second */ - timeout = ERTS_MONOTONIC_TO_USEC(diff_time - 1) + 1; - save_timeout_time += ERTS_USEC_TO_MONOTONIC(timeout); - timeout -= ERTS_PREMATURE_TIMEOUT(timeout, 1000*1000); - break; - case 1000000000: - /* Round up to nearest even nano second */ - timeout = ERTS_MONOTONIC_TO_NSEC(diff_time - 1) + 1; - save_timeout_time += ERTS_NSEC_TO_MONOTONIC(timeout); - timeout -= ERTS_PREMATURE_TIMEOUT(timeout, 1000*1000*1000); - break; - default: - ERTS_INTERNAL_ERROR("Invalid resolution"); - timeout = 0; - save_timeout_time = 0; - break; - } - } - } - set_timeout_time(ps, save_timeout_time); - return timeout; -} - -#if ERTS_POLL_USE_SELECT - -static ERTS_INLINE int -get_timeout_timeval(ErtsPollSet ps, - SysTimeval *tvp, - ErtsMonotonicTime timeout_time) -{ - ErtsMonotonicTime timeout = get_timeout(ps, - 1000*1000, - timeout_time); - - if (!timeout) { - tvp->tv_sec = 0; - tvp->tv_usec = 0; - - return 0; - } - else { - ErtsMonotonicTime sec = timeout/(1000*1000); - tvp->tv_sec = sec; - tvp->tv_usec = timeout - sec*(1000*1000); - - ASSERT(tvp->tv_sec >= 0); - ASSERT(tvp->tv_usec >= 0); - ASSERT(tvp->tv_usec < 1000*1000); - - return !0; - } - -} - -#endif - -#if ERTS_POLL_USE_KQUEUE || (ERTS_POLL_USE_POLL && defined(HAVE_PPOLL)) || ERTS_POLL_USE_TIMERFD - -static ERTS_INLINE int -get_timeout_timespec(ErtsPollSet ps, - struct timespec *tsp, - ErtsMonotonicTime timeout_time) -{ - ErtsMonotonicTime timeout = get_timeout(ps, - 1000*1000*1000, - timeout_time); - - if (!timeout) { - tsp->tv_sec = 0; - tsp->tv_nsec = 0; - return 0; - } - else { - ErtsMonotonicTime sec = timeout/(1000*1000*1000); - tsp->tv_sec = sec; - tsp->tv_nsec = timeout - sec*(1000*1000*1000); - - ASSERT(tsp->tv_sec >= 0); - ASSERT(tsp->tv_nsec >= 0); - ASSERT(tsp->tv_nsec < 1000*1000*1000); - - return !0; - } -} - -#endif - -#if ERTS_POLL_USE_TIMERFD - -static ERTS_INLINE int -get_timeout_itimerspec(ErtsPollSet ps, - struct itimerspec *itsp, - ErtsMonotonicTime timeout_time) -{ - - itsp->it_interval.tv_sec = 0; - itsp->it_interval.tv_nsec = 0; - - return get_timeout_timespec(ps, &itsp->it_value, timeout_time); +#endif /* ERTS_POLL_USE_SELECT */ } -#endif +#endif /* !ERTS_POLL_USE_KERNEL_POLL */ static ERTS_INLINE int -check_fd_events(ErtsPollSet ps, ErtsMonotonicTime timeout_time, int max_res) +check_fd_events(ErtsPollSet *ps, ErtsPollResFd pr[], int do_wait, int max_res) { int res; - ERTS_MSACC_PUSH_STATE_M(); - if (erts_atomic_read_nob(&ps->no_of_user_fds) == 0 - && timeout_time == ERTS_POLL_NO_TIMEOUT) { - /* Nothing to poll and zero timeout; done... */ - return 0; - } - else { - int timeout; -#if ERTS_POLL_USE_FALLBACK - if (!(ps->fallback_used = ERTS_POLL_NEED_FALLBACK(ps))) { - -#if ERTS_POLL_USE_EPOLL /* --- epoll ------------------------------- */ - if (max_res > ps->res_events_len) - grow_res_events(ps, max_res); -#if ERTS_POLL_USE_TIMERFD - { - struct itimerspec its; - timeout = get_timeout_itimerspec(ps, &its, timeout_time); - if (timeout) { - erts_thr_progress_prepare_wait(NULL); - ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_SLEEP); - timerfd_set(ps, &its); - res = epoll_wait(ps->kp_fd, ps->res_events, max_res, -1); - res = timerfd_clear(ps, res, max_res); - } else { - res = epoll_wait(ps->kp_fd, ps->res_events, max_res, 0); - } - } -#else /* !ERTS_POLL_USE_TIMERFD */ - timeout = (int) get_timeout(ps, 1000, timeout_time); - if (timeout) { - erts_thr_progress_prepare_wait(NULL); - ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_SLEEP); - } - res = epoll_wait(ps->kp_fd, ps->res_events, max_res, timeout); -#endif /* !ERTS_POLL_USE_TIMERFD */ -#elif ERTS_POLL_USE_KQUEUE /* --- kqueue ------------------------------ */ - struct timespec ts; - if (max_res > ps->res_events_len) - grow_res_events(ps, max_res); - timeout = get_timeout_timespec(ps, &ts, timeout_time); - if (timeout) { - erts_thr_progress_prepare_wait(NULL); - ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_SLEEP); - } - res = kevent(ps->kp_fd, NULL, 0, ps->res_events, max_res, &ts); -#endif /* ----------------------------------------- */ - } - else /* use fallback (i.e. poll() or select()) */ -#endif /* ERTS_POLL_USE_FALLBACK */ - { + int timeout = do_wait ? -1 : 0; + DEBUG_PRINT_WAIT("Entering check_fd_events(), do_wait=%d", ps, do_wait); + { +#if ERTS_POLL_USE_EPOLL /* --- epoll ------------------------------- */ + res = epoll_wait(ps->kp_fd, pr, max_res, timeout); + +#elif ERTS_POLL_USE_KQUEUE /* --- kqueue ------------------------------ */ + struct timespec ts = {0, 0}; + struct timespec *tsp = timeout ? NULL : &ts; + res = kevent(ps->kp_fd, NULL, 0, pr, max_res, tsp); +#elif ERTS_POLL_USE_DEVPOLL /* --- devpoll ----------------------------- */ + /* + * The ioctl() will fail with EINVAL on Solaris 10 if dp_nfds + * is set too high. dp_nfds should not be set greater than + * the maximum number of file descriptors in the poll set. + */ + struct dvpoll poll_res; + int nfds = (int) erts_atomic_read_nob(&ps->no_of_user_fds) + 1 /* wakeup pipe */; + poll_res.dp_nfds = nfds < max_res ? nfds : max_res; + poll_res.dp_fds = pr; + poll_res.dp_timeout = timeout; + res = ioctl(ps->kp_fd, DP_POLL, &poll_res); -#if ERTS_POLL_USE_DEVPOLL /* --- devpoll ----------------------------- */ - /* - * The ioctl() will fail with EINVAL on Solaris 10 if dp_nfds - * is set too high. dp_nfds should not be set greater than - * the maximum number of file descriptors in the poll set. - */ - struct dvpoll poll_res; - int nfds = (int) erts_atomic_read_nob(&ps->no_of_user_fds); - nfds++; /* Wakeup pipe */ - timeout = (int) get_timeout(ps, 1000, timeout_time); - poll_res.dp_nfds = nfds < max_res ? nfds : max_res; - if (poll_res.dp_nfds > ps->res_events_len) - grow_res_events(ps, poll_res.dp_nfds); - poll_res.dp_fds = ps->res_events; - if (timeout) { - erts_thr_progress_prepare_wait(NULL); - ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_SLEEP); - } - poll_res.dp_timeout = timeout; - res = ioctl(ps->kp_fd, DP_POLL, &poll_res); -#elif ERTS_POLL_USE_POLL && defined(HAVE_PPOLL) /* --- ppoll ---------------- */ - struct timespec ts; - timeout = get_timeout_timespec(ps, &ts, timeout_time); - if (timeout) { - erts_thr_progress_prepare_wait(NULL); - ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_SLEEP); - } - res = ppoll(ps->poll_fds, ps->no_poll_fds, &ts, NULL); #elif ERTS_POLL_USE_POLL /* --- poll --------------------------------- */ - timeout = (int) get_timeout(ps, 1000, timeout_time); - if (timeout) { - erts_thr_progress_prepare_wait(NULL); - ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_SLEEP); - } - res = poll(ps->poll_fds, ps->no_poll_fds, timeout); + res = poll(ps->poll_fds, ps->no_poll_fds, timeout); + #elif ERTS_POLL_USE_SELECT /* --- select ------------------------------ */ - SysTimeval to; - timeout = get_timeout_timeval(ps, &to, timeout_time); + SysTimeval tv = {0, 0}; + SysTimeval *tvp = timeout ? NULL : &tv; - ERTS_FD_COPY(&ps->input_fds, &ps->res_input_fds); - ERTS_FD_COPY(&ps->output_fds, &ps->res_output_fds); + ERTS_FD_COPY(&ps->input_fds, &ps->res_input_fds); + ERTS_FD_COPY(&ps->output_fds, &ps->res_output_fds); - if (timeout) { - erts_thr_progress_prepare_wait(NULL); - ERTS_MSACC_SET_STATE_CACHED_M(ERTS_MSACC_STATE_SLEEP); - } - res = ERTS_SELECT(ps->max_fd + 1, - &ps->res_input_fds, - &ps->res_output_fds, - NULL, - &to); - if (timeout) { - erts_thr_progress_finalize_wait(NULL); - ERTS_MSACC_POP_STATE_M(); - } - if (res < 0 - && errno == EBADF - && ERTS_POLLSET_HAVE_UPDATE_REQUESTS(ps)) { - /* - * This may have happened because another thread deselected - * a fd in our poll set and then closed it, i.e. the driver - * behaved correctly. We wan't to avoid looking for a bad - * fd, that may even not exist anymore. Therefore, handle - * update requests and try again. - * - * We don't know how much of the timeout is left; therfore, - * we use a zero timeout. If no error occur and no events - * have triggered, we fake an EAGAIN error and let the caller - * restart us. - */ - to.tv_sec = 0; - to.tv_usec = 0; - ERTS_POLLSET_LOCK(ps); - handle_update_requests(ps); - ERTS_POLLSET_UNLOCK(ps); - res = ERTS_SELECT(ps->max_fd + 1, - &ps->res_input_fds, - &ps->res_output_fds, - NULL, - &to); - if (res == 0) { - errno = EAGAIN; - res = -1; - } - } - return res; + res = ERTS_SELECT(ps->max_fd + 1, + &ps->res_input_fds, + &ps->res_output_fds, + NULL, + tvp); #endif /* ----------------------------------------- */ - } - if (timeout) { - erts_thr_progress_finalize_wait(NULL); - ERTS_MSACC_POP_STATE_M(); - } - return res; } + DEBUG_PRINT_WAIT("Leaving check_fd_events(), res=%d", ps, res); + return res; } int -ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps, +ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet *ps, ErtsPollResFd pr[], - int *len, - ErtsMonotonicTime timeout_time) + int *len) { - ErtsMonotonicTime to; - int res, no_fds; + int res, no_fds, used_fds = 0; int ebadf = 0; + int do_wait; int ps_locked = 0; + ERTS_MSACC_DECLARE_CACHE(); no_fds = *len; -#ifdef ERTS_POLL_MAX_RES - if (no_fds >= ERTS_POLL_MAX_RES) - no_fds = ERTS_POLL_MAX_RES; -#endif - *len = 0; + ASSERT(no_fds > 0); -#ifdef ERTS_POLL_DEBUG_PRINT - erts_printf("Entering erts_poll_wait(), timeout_time=%bps\n", - timeout_time); -#endif +#if !ERTS_POLL_USE_CONCURRENT_UPDATE + if (ERTS_POLLSET_HAVE_UPDATE_REQUESTS(ps)) { + ERTS_POLLSET_LOCK(ps); + used_fds = handle_update_requests(ps, pr, no_fds); + ERTS_POLLSET_UNLOCK(ps); - if (ERTS_POLLSET_SET_POLLED_CHK(ps)) { - res = EINVAL; /* Another thread is in erts_poll_wait() - on this pollset... */ - goto done; + if (used_fds == no_fds) { + *len = used_fds; + return 0; + } } +#endif - to = (is_woken(ps) - ? ERTS_POLL_NO_TIMEOUT /* Use zero timeout */ - : timeout_time); + do_wait = !is_woken(ps) && used_fds == 0; - if (ERTS_POLLSET_HAVE_UPDATE_REQUESTS(ps)) { - ERTS_POLLSET_LOCK(ps); - handle_update_requests(ps); - ERTS_POLLSET_UNLOCK(ps); + DEBUG_PRINT_WAIT("Entering %s(), do_wait=%d", ps, __FUNCTION__, do_wait); + + if (do_wait) { + erts_thr_progress_prepare_wait(NULL); + ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_SLEEP); } while (1) { - res = check_fd_events(ps, to, no_fds); + res = check_fd_events(ps, pr + used_fds, do_wait, no_fds - used_fds); + +#if !ERTS_POLL_USE_CONCURRENT_UPDATE + if (res < 0 + && errno == EBADF + && ERTS_POLLSET_HAVE_UPDATE_REQUESTS(ps)) { + /* + * This may have happened because another thread deselected + * a fd in our poll set and then closed it, i.e. the driver + * behaved correctly. We wan't to avoid looking for a bad + * fd, that may even not exist anymore. Therefore, handle + * update requests and try again. This behaviour should only + * happen when using SELECT as the polling mechanism. + */ + ERTS_POLLSET_LOCK(ps); + used_fds += handle_update_requests(ps, pr + used_fds, no_fds - used_fds); + if (used_fds == no_fds) { + *len = used_fds; + ERTS_POLLSET_UNLOCK(ps); + return 0; + } + res = check_fd_events(ps, pr + used_fds, 0, no_fds - used_fds); + /* Keep the lock over the non-blocking poll in order to not + get any nasty races happening. */ + ERTS_POLLSET_UNLOCK(ps); + if (res == 0) { + errno = EAGAIN; + res = -1; + } + } +#endif + if (res != 0) break; - if (to == ERTS_POLL_NO_TIMEOUT) - break; - if (erts_get_monotonic_time(NULL) >= timeout_time) - break; + if (!do_wait) + break; + } + + if (do_wait) { + erts_thr_progress_finalize_wait(NULL); + ERTS_MSACC_UPDATE_CACHE(); + ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_CHECK_IO); } woke_up(ps); - if (res == 0) { - res = ETIMEDOUT; - } - else if (res < 0) { + if (res < 0) { #if ERTS_POLL_USE_SELECT if (errno == EBADF) { ebadf = 1; @@ -2363,26 +1666,21 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps, ps_locked = 1; ERTS_POLLSET_LOCK(ps); - no_fds = save_poll_result(ps, pr, no_fds, res, ebadf); + used_fds += ERTS_POLL_EXPORT(save_result)(ps, pr + used_fds, no_fds - used_fds, res, ebadf); #ifdef HARD_DEBUG - check_poll_result(pr, no_fds); + check_poll_result(pr, used_fds); #endif - res = (no_fds == 0 ? (is_interrupted_reset(ps) ? EINTR : EAGAIN) : 0); - *len = no_fds; + res = (used_fds == 0 ? (is_interrupted_reset(ps) ? EINTR : EAGAIN) : 0); + *len = used_fds; } if (ps_locked) ERTS_POLLSET_UNLOCK(ps); - ERTS_POLLSET_UNSET_POLLED(ps); - done: - set_timeout_time(ps, ERTS_MONOTONIC_TIME_MAX); -#ifdef ERTS_POLL_DEBUG_PRINT - erts_printf("Leaving %s = erts_poll_wait()\n", - res == 0 ? "0" : erl_errno_id(res)); -#endif + DEBUG_PRINT_WAIT("Leaving %s = %s(len = %d)", ps, + res == 0 ? "0" : erl_errno_id(res), __FUNCTION__, *len); return res; } @@ -2392,40 +1690,14 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps, */ void -ERTS_POLL_EXPORT(erts_poll_interrupt)(ErtsPollSet ps, int set) +ERTS_POLL_EXPORT(erts_poll_interrupt)(ErtsPollSet *ps, int set) { +#if !ERTS_POLL_USE_CONCURRENT_UPDATE if (!set) reset_wakeup_state(ps); else - wake_poller(ps, 1, 0); -} - - -/* - * erts_poll_interrupt_timed(): - * If 'set' != 0, interrupt thread blocked in erts_poll_wait() if it - * is not guaranteed that it will timeout before 'msec' milli seconds. - */ -void -ERTS_POLL_EXPORT(erts_poll_interrupt_timed)(ErtsPollSet ps, - int set, - ErtsMonotonicTime timeout_time) -{ - if (!set) - reset_wakeup_state(ps); - else { - ErtsMonotonicTime max_wait_time = get_timeout_time(ps); - if (max_wait_time > timeout_time) - wake_poller(ps, 1, 0); -#ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS - else { - if (ERTS_POLLSET_IS_POLLED(ps)) - erts_atomic_inc_nob(&ps->no_avoided_wakeups); - erts_atomic_inc_nob(&ps->no_avoided_interrupts); - } - erts_atomic_inc_nob(&ps->no_interrupt_timed); + wake_poller(ps, 1); #endif - } } int @@ -2438,14 +1710,19 @@ ERTS_POLL_EXPORT(erts_poll_max_fds)(void) */ void -ERTS_POLL_EXPORT(erts_poll_init)(void) +ERTS_POLL_EXPORT(erts_poll_init)(int *concurrent_updates) { - erts_mtx_init(&pollsets_lock, "pollsets_lock", NIL, - ERTS_LOCK_FLAGS_PROPERTY_STATIC | ERTS_LOCK_FLAGS_CATEGORY_IO); - pollsets = NULL; errno = 0; + if (concurrent_updates) { +#if ERTS_POLL_USE_CONCURRENT_UPDATE + *concurrent_updates = 1; +#else + *concurrent_updates = 0; +#endif + } + #if !defined(NO_SYSCONF) max_fds = sysconf(_SC_OPEN_MAX); #elif ERTS_POLL_USE_SELECT @@ -2464,37 +1741,28 @@ ERTS_POLL_EXPORT(erts_poll_init)(void) fatal_error("erts_poll_init(): Failed to get max number of files: %s\n", erl_errno_id(errno)); -#ifdef ERTS_POLL_DEBUG_PRINT print_misc_debug_info(); -#endif } -ErtsPollSet -ERTS_POLL_EXPORT(erts_poll_create_pollset)(void) +ErtsPollSet * +ERTS_POLL_EXPORT(erts_poll_create_pollset)(int id) { #if ERTS_POLL_USE_KERNEL_POLL int kp_fd; #endif - ErtsPollSet ps = erts_alloc(ERTS_ALC_T_POLLSET, + ErtsPollSet *ps = erts_alloc(ERTS_ALC_T_POLLSET, sizeof(struct ERTS_POLL_EXPORT(erts_pollset))); + ps->id = id; ps->internal_fd_limit = 0; - ps->fds_status = NULL; - ps->fds_status_len = 0; erts_atomic_init_nob(&ps->no_of_user_fds, 0); #if ERTS_POLL_USE_KERNEL_POLL ps->kp_fd = -1; #if ERTS_POLL_USE_EPOLL kp_fd = epoll_create(256); - ps->res_events_len = 0; - ps->res_events = NULL; #elif ERTS_POLL_USE_DEVPOLL kp_fd = open("/dev/poll", O_RDWR); - ps->res_events_len = 0; - ps->res_events = NULL; #elif ERTS_POLL_USE_KQUEUE kp_fd = kqueue(); - ps->res_events_len = 0; - ps->res_events = NULL; #endif if (kp_fd < 0) fatal_error("erts_poll_create_pollset(): Failed to " @@ -2508,10 +1776,6 @@ ERTS_POLL_EXPORT(erts_poll_create_pollset)(void) ": %s (%d)\n", erl_errno_id(errno), errno); #endif /* ERTS_POLL_USE_KERNEL_POLL */ -#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET - /* res_events is also used as write buffer */ - grow_res_events(ps, ERTS_POLL_MIN_BATCH_BUF_SIZE); -#endif #if ERTS_POLL_USE_POLL ps->next_poll_fds_ix = 0; ps->no_poll_fds = 0; @@ -2520,9 +1784,6 @@ ERTS_POLL_EXPORT(erts_poll_create_pollset)(void) #elif ERTS_POLL_USE_SELECT ps->next_sel_fd = 0; ps->max_fd = -1; -#if ERTS_POLL_USE_FALLBACK - ps->no_select_fds = 0; -#endif #ifdef _DARWIN_UNLIMITED_SELECT ps->input_fds.sz = 0; ps->input_fds.ptr = NULL; @@ -2539,140 +1800,48 @@ ERTS_POLL_EXPORT(erts_poll_create_pollset)(void) ERTS_FD_ZERO(&ps->res_output_fds); #endif #endif +#if !ERTS_POLL_USE_CONCURRENT_UPDATE + ps->fds_status = NULL; + ps->fds_status_len = 0; ps->update_requests.next = NULL; ps->update_requests.len = 0; ps->curr_upd_req_block = &ps->update_requests; erts_atomic32_init_nob(&ps->have_update_requests, 0); - erts_atomic32_init_nob(&ps->polled, 0); erts_mtx_init(&ps->mtx, "pollset", NIL, ERTS_LOCK_FLAGS_CATEGORY_IO); - erts_atomic32_init_nob(&ps->wakeup_state, (erts_aint32_t) 0); - create_wakeup_pipe(ps); -#if ERTS_POLL_USE_TIMERFD - create_timerfd(ps); -#endif -#if ERTS_POLL_USE_FALLBACK - if (kp_fd >= ps->fds_status_len) - grow_fds_status(ps, kp_fd); - /* Force kernel poll fd into fallback (poll/select) set */ - ps->fds_status[kp_fd].flags - |= ERTS_POLL_FD_FLG_INFLBCK|ERTS_POLL_FD_FLG_USEFLBCK; - { - int do_wake = 0; - ERTS_POLL_EXPORT(erts_poll_control)(ps, kp_fd, ERTS_POLL_EV_IN, 1, - &do_wake); - } #endif #if ERTS_POLL_USE_KERNEL_POLL if (ps->internal_fd_limit <= kp_fd) ps->internal_fd_limit = kp_fd + 1; ps->kp_fd = kp_fd; #endif - init_timeout_time(ps); -#ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS - erts_atomic_init_nob(&ps->no_avoided_wakeups, 0); - erts_atomic_init_nob(&ps->no_avoided_interrupts, 0); - erts_atomic_init_nob(&ps->no_interrupt_timed, 0); -#endif - handle_update_requests(ps); -#if ERTS_POLL_USE_FALLBACK - ps->fallback_used = 0; +#if !ERTS_POLL_USE_CONCURRENT_UPDATE + erts_atomic32_init_nob(&ps->wakeup_state, (erts_aint32_t) 0); + create_wakeup_pipe(ps); + handle_update_requests(ps, NULL, 0); + cleanup_wakeup_pipe(ps); #endif erts_atomic_set_nob(&ps->no_of_user_fds, 0); /* Don't count wakeup pipe and fallback fd */ - erts_mtx_lock(&pollsets_lock); - ps->next = pollsets; - pollsets = ps; - erts_mtx_unlock(&pollsets_lock); - return ps; } -void -ERTS_POLL_EXPORT(erts_poll_destroy_pollset)(ErtsPollSet ps) -{ - - if (ps->fds_status) - erts_free(ERTS_ALC_T_FD_STATUS, (void *) ps->fds_status); - -#if ERTS_POLL_USE_EPOLL - if (ps->kp_fd >= 0) - close(ps->kp_fd); - if (ps->res_events) - erts_free(ERTS_ALC_T_POLL_RES_EVS, (void *) ps->res_events); -#elif ERTS_POLL_USE_DEVPOLL - if (ps->kp_fd >= 0) - close(ps->kp_fd); - if (ps->res_events) - erts_free(ERTS_ALC_T_POLL_RES_EVS, (void *) ps->res_events); -#elif ERTS_POLL_USE_POLL - if (ps->poll_fds) - erts_free(ERTS_ALC_T_POLL_FDS, (void *) ps->poll_fds); -#elif ERTS_POLL_USE_SELECT -#ifdef _DARWIN_UNLIMITED_SELECT - if (ps->input_fds.ptr) - erts_free(ERTS_ALC_T_SELECT_FDS, (void *) ps->input_fds.ptr); - if (ps->res_input_fds.ptr) - erts_free(ERTS_ALC_T_SELECT_FDS, (void *) ps->res_input_fds.ptr); - if (ps->output_fds.ptr) - erts_free(ERTS_ALC_T_SELECT_FDS, (void *) ps->output_fds.ptr); - if (ps->res_output_fds.ptr) - erts_free(ERTS_ALC_T_SELECT_FDS, (void *) ps->res_output_fds.ptr); -#endif -#endif - { - ErtsPollSetUpdateRequestsBlock *urqbp = ps->update_requests.next; - while (urqbp) { - ErtsPollSetUpdateRequestsBlock *free_urqbp = urqbp; - urqbp = urqbp->next; - free_update_requests_block(ps, free_urqbp); - } - } - erts_mtx_destroy(&ps->mtx); - if (ps->wake_fds[0] >= 0) - close(ps->wake_fds[0]); - if (ps->wake_fds[1] >= 0) - close(ps->wake_fds[1]); -#if ERTS_POLL_USE_TIMERFD - if (ps->timer_fd >= 0) - close(ps->timer_fd); -#endif - - erts_mtx_lock(&pollsets_lock); - if (ps == pollsets) - pollsets = pollsets->next; - else { - ErtsPollSet prev_ps; - for (prev_ps = pollsets; ps != prev_ps->next; prev_ps = prev_ps->next) - ; - ASSERT(ps == prev_ps->next); - prev_ps->next = ps->next; - } - erts_mtx_unlock(&pollsets_lock); - - erts_free(ERTS_ALC_T_POLLSET, (void *) ps); -} - /* * --- Info ------------------------------------------------------------------ */ void -ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet ps, ErtsPollInfo *pip) +ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet *ps, ErtsPollInfo *pip) { +#if !ERTS_POLL_USE_CONCURRENT_UPDATE int pending_updates; +#endif Uint size = 0; ERTS_POLLSET_LOCK(ps); size += sizeof(struct ERTS_POLL_EXPORT(erts_pollset)); +#if !ERTS_POLL_USE_CONCURRENT_UPDATE size += ps->fds_status_len*sizeof(ErtsFdStatus); - -#if ERTS_POLL_USE_EPOLL - size += ps->res_events_len*sizeof(struct epoll_event); -#elif ERTS_POLL_USE_DEVPOLL - size += ps->res_events_len*sizeof(struct pollfd); -#elif ERTS_POLL_USE_KQUEUE - size += ps->res_events_len*sizeof(struct kevent); #endif #if ERTS_POLL_USE_POLL @@ -2684,6 +1853,7 @@ ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet ps, ErtsPollInfo *pip) #endif #endif +#if !ERTS_POLL_USE_CONCURRENT_UPDATE { ErtsPollSetUpdateRequestsBlock *urqbp = ps->update_requests.next; pending_updates = ps->update_requests.len; @@ -2693,8 +1863,9 @@ ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet ps, ErtsPollInfo *pip) urqbp = urqbp->next; } } +#endif - pip->primary = + pip->primary = #if ERTS_POLL_USE_KQUEUE "kqueue" #elif ERTS_POLL_USE_EPOLL @@ -2708,17 +1879,7 @@ ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet ps, ErtsPollInfo *pip) #endif ; - pip->fallback = -#if !ERTS_POLL_USE_FALLBACK - NULL -#elif ERTS_POLL_USE_POLL - "poll" -#elif ERTS_POLL_USE_SELECT - "select" -#endif - ; - - pip->kernel_poll = + pip->kernel_poll = #if !ERTS_POLL_USE_KERNEL_POLL NULL #elif ERTS_POLL_USE_KQUEUE @@ -2733,40 +1894,21 @@ ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet ps, ErtsPollInfo *pip) pip->memory_size = size; pip->poll_set_size = (int) erts_atomic_read_nob(&ps->no_of_user_fds); +#if !ERTS_POLL_USE_CONCURRENT_UPDATE pip->poll_set_size++; /* Wakeup pipe */ -#if ERTS_POLL_USE_TIMERFD - pip->poll_set_size++; /* timerfd */ -#endif - - pip->fallback_poll_set_size = -#if !ERTS_POLL_USE_FALLBACK - 0 -#elif ERTS_POLL_USE_POLL - ps->no_poll_fds -#elif ERTS_POLL_USE_SELECT - ps->no_select_fds -#endif - ; - -#if ERTS_POLL_USE_FALLBACK - /* If only kp_fd is in fallback poll set we don't use fallback... */ - if (pip->fallback_poll_set_size == 1) - pip->fallback_poll_set_size = 0; - else - pip->poll_set_size++; /* kp_fd */ #endif pip->lazy_updates = +#if !ERTS_POLL_USE_CONCURRENT_UPDATE 1 +#else + 0 +#endif ; pip->pending_updates = +#if !ERTS_POLL_USE_CONCURRENT_UPDATE pending_updates - ; - - pip->batch_updates = -#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET - 1 #else 0 #endif @@ -2780,13 +1922,15 @@ ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet ps, ErtsPollInfo *pip) #endif ; - pip->max_fds = max_fds; - -#ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS - pip->no_avoided_wakeups = erts_atomic_read_nob(&ps->no_avoided_wakeups); - pip->no_avoided_interrupts = erts_atomic_read_nob(&ps->no_avoided_interrupts); - pip->no_interrupt_timed = erts_atomic_read_nob(&ps->no_interrupt_timed); + pip->is_fallback = +#if ERTS_POLL_IS_FALLBACK + 1 +#else + 0 #endif + ; + + pip->max_fds = max_fds; ERTS_POLLSET_UNLOCK(ps); @@ -2822,35 +1966,60 @@ fatal_error(char *format, ...) abort(); } -static void -fatal_error_async_signal_safe(char *error_str) +/* + * --- Debug ----------------------------------------------------------------- + */ + +#if ERTS_POLL_USE_EPOLL +uint32_t epoll_events(int kp_fd, int fd) { - if (ERTS_SOMEONE_IS_CRASH_DUMPING || ERTS_GOT_SIGUSR1) { - /* See comment above in fatal_error() */ - return; + /* For epoll we read the information about what is selected upon from the proc fs.*/ + char fname[30]; + FILE *f; + unsigned int pos, flags, mnt_id; + int line = 0; + sprintf(fname,"/proc/%d/fdinfo/%d",getpid(), kp_fd); + f = fopen(fname,"r"); + if (!f) { + fprintf(stderr,"failed to open file %s, errno = %d\n", fname, errno); + ASSERT(0); + return 0; } - if (error_str) { - int len = 0; - while (error_str[len]) - len++; - if (len) { - /* async signal safe */ - erts_silence_warn_unused_result(write(2, error_str, len)); - } + if (fscanf(f,"pos:\t%x\nflags:\t%x", &pos, &flags) != 2) { + fprintf(stderr,"failed to parse file %s, errno = %d\n", fname, errno); + ASSERT(0); + return 0; } - abort(); + if (fscanf(f,"\nmnt_id:\t%x\n", &mnt_id)); + line += 3; + while (!feof(f)) { + /* tfd: 10 events: 40000019 data: 180000000a */ + int ev_fd; + uint32_t events; + uint64_t data; + if (fscanf(f,"tfd:%d events:%x data:%lx\n", &ev_fd, &events, &data) != 3) { + fprintf(stderr,"failed to parse file %s on line %d, errno = %d\n", fname, + line, + errno); + return 0; + } + if (fd == ev_fd) { + fclose(f); + return events; + } + } + fclose(f); + return 0; } - -/* - * --- Debug ----------------------------------------------------------------- - */ +#endif void -ERTS_POLL_EXPORT(erts_poll_get_selected_events)(ErtsPollSet ps, +ERTS_POLL_EXPORT(erts_poll_get_selected_events)(ErtsPollSet *ps, ErtsPollEvents ev[], int len) { int fd; +#if !ERTS_POLL_USE_CONCURRENT_UPDATE ERTS_POLLSET_LOCK(ps); for (fd = 0; fd < len; fd++) { if (fd >= ps->fds_status_len) @@ -2859,9 +2028,6 @@ ERTS_POLL_EXPORT(erts_poll_get_selected_events)(ErtsPollSet ps, ev[fd] = ps->fds_status[fd].events; if ( fd == ps->wake_fds[0] || fd == ps->wake_fds[1] || -#if ERTS_POLL_USE_TIMERFD - fd == ps->timer_fd || -#endif #if ERTS_POLL_USE_KERNEL_POLL fd == ps->kp_fd || #endif @@ -2870,7 +2036,50 @@ ERTS_POLL_EXPORT(erts_poll_get_selected_events)(ErtsPollSet ps, } } ERTS_POLLSET_UNLOCK(ps); +#elif ERTS_POLL_USE_EPOLL + /* For epoll we read the information about what is selected upon from the proc fs.*/ + char fname[30]; + FILE *f; + unsigned int pos, flags, mnt_id; + int line = 0; + sprintf(fname,"/proc/%d/fdinfo/%d",getpid(), ps->kp_fd); + for (fd = 0; fd < len; fd++) + ev[fd] = ERTS_POLL_EV_NONE; + f = fopen(fname,"r"); + if (!f) { + fprintf(stderr,"failed to open file %s, errno = %d\n", fname, errno); + return; + } + if (fscanf(f,"pos:\t%x\nflags:\t%x", &pos, &flags) != 2) { + fprintf(stderr,"failed to parse file %s, errno = %d\n", fname, errno); + ASSERT(0); + return; + } + if (fscanf(f,"\nmnt_id:\t%x\n", &mnt_id)); + line += 3; + while (!feof(f)) { + /* tfd: 10 events: 40000019 data: 180000000a */ + int fd; + uint32_t events; + uint64_t data; + if (fscanf(f,"tfd:%d events:%x data:%lx\n", &fd, &events, &data) != 3) { + fprintf(stderr,"failed to parse file %s on line %d, errno = %d\n", + fname, line, errno); + ASSERT(0); + return; + } + data &= 0xFFFFFFFF; + ASSERT(fd == data); + /* Events are the events that are being monitored, which of course include + error and hup events, but we are only interested in IN/OUT events */ + ev[fd] = (ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT) & ERTS_POLL_EV_N2E(events); + line++; + } +#else + for (fd = 0; fd < len; fd++) + ev[fd] = ERTS_POLL_EV_NONE; +#endif } #ifdef HARD_DEBUG @@ -2890,10 +2099,10 @@ check_poll_result(ErtsPollResFd pr[], int len) } -#if ERTS_POLL_USE_DEVPOLL +#if ERTS_POLL_USE_DEVPOLL && defined(DEBUG) static void -check_poll_status(ErtsPollSet ps) +check_poll_status(ErtsPollSet *ps) { int i; for (i = 0; i < ps->fds_status_len; i++) { @@ -2925,30 +2134,24 @@ check_poll_status(ErtsPollSet ps) #endif /* ERTS_POLL_USE_DEVPOLL */ #endif /* HARD_DEBUG */ -#ifdef ERTS_POLL_DEBUG_PRINT static void print_misc_debug_info(void) { - erts_printf("erts_poll using: %s lazy_updates:%s batch_updates:%s\n", +#if ERTS_POLL_DEBUG_PRINT + erts_printf("erts_poll using: %s lazy_updates:%s\n", #if ERTS_POLL_USE_KQUEUE "kqueue" #elif ERTS_POLL_USE_EPOLL "epoll" #elif ERTS_POLL_USE_DEVPOLL "/dev/poll" -#endif -#if ERTS_POLL_USE_FALLBACK - "-" -#endif -#if ERTS_POLL_USE_POLL +#elif ERTS_POLL_USE_POLL "poll" #elif ERTS_POLL_USE_SELECT "select" #endif , - "true" - , -#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET +#if !ERTS_POLL_USE_CONCURRENT_UPDATE "true" #else "false" @@ -2967,29 +2170,20 @@ print_misc_debug_info(void) #ifdef FD_SETSIZE erts_printf("FD_SETSIZE=%d\n", FD_SETSIZE); #endif -} - #endif +} #ifdef ERTS_ENABLE_LOCK_COUNT -static void erts_lcnt_enable_pollset_lock_count(ErtsPollSet pollset, int enable) { +void ERTS_POLL_EXPORT(erts_lcnt_enable_pollset_lock_count)(ErtsPollSet *pollset, int enable) +{ +#if !ERTS_POLL_USE_CONCURRENT_UPDATE if(enable) { erts_lcnt_install_new_lock_info(&pollset->mtx.lcnt, "pollset_rm", NIL, ERTS_LOCK_TYPE_MUTEX | ERTS_LOCK_FLAGS_CATEGORY_IO); } else { erts_lcnt_uninstall(&pollset->mtx.lcnt); } -} - -void ERTS_POLL_EXPORT(erts_lcnt_update_pollset_locks)(int enable) { - ErtsPollSet iterator; - - erts_mtx_lock(&pollsets_lock); - - for(iterator = pollsets; iterator != NULL; iterator = iterator->next) { - erts_lcnt_enable_pollset_lock_count(iterator, enable); - } - - erts_mtx_unlock(&pollsets_lock); +#endif + return; } #endif diff --git a/erts/emulator/sys/common/erl_poll.h b/erts/emulator/sys/common/erl_poll.h index 6c961205fe..a1e2709a9f 100644 --- a/erts/emulator/sys/common/erl_poll.h +++ b/erts/emulator/sys/common/erl_poll.h @@ -18,11 +18,31 @@ * %CopyrightEnd% */ -/* - * Description: Poll interface suitable for ERTS with or without - * SMP support. +/** + * @description: Poll interface suitable for ERTS with SMP support. + * + * @author: Rickard Green + * @author: Lukas Larsson + * + * This header file exports macros and functions that are used to + * react to I/O polling events from file descriptors or wait-able + * objects. The API exported is the following: * - * Author: Rickard Green + * defines: + * ERTS_POLL_EV_NONE - No events have been set. This is not the same as 0. + * ERTS_POLL_EV_IN - Represent an IN event + * ERTS_POLL_EV_OUT - Represent an OUT event + * ERTS_POLL_EV_ERR - Represent an error event + * ERTS_POLL_EV_NVAL - Represent an invalid event + * + * macro functions: + * ErtsSysFdType ERTS_POLL_RES_GET_FD(ErtsPollResFd *evt); + * void ERTS_POLL_RES_SET_FD(ErtsPollResFd *evt, ErtsSysFdType fd); + * ErtsPollEvents ERTS_POLL_RES_GET_EVTS(ErtsPollResFd *evt) + * void ERTS_POLL_RES_SET_EVTS(ErtsPollResFd *evt, ErtsPollEvents fd); + * + * functions: + * See erl_poll_api.h */ #ifndef ERL_POLL_H__ @@ -32,33 +52,27 @@ #define ERTS_POLL_NO_TIMEOUT ERTS_MONOTONIC_TIME_MIN -#if 0 -#define ERTS_POLL_COUNT_AVOIDED_WAKEUPS -#endif - #ifdef ERTS_ENABLE_KERNEL_POLL -# if defined(ERTS_KERNEL_POLL_VERSION) -# define ERTS_POLL_EXPORT(FUNC) FUNC ## _kp +# undef ERTS_ENABLE_KERNEL_POLL +# define ERTS_ENABLE_KERNEL_POLL 1 +# if defined(ERTS_NO_KERNEL_POLL_VERSION) +# define ERTS_POLL_EXPORT(FUNC) FUNC ## _flbk +# undef ERTS_NO_KERNEL_POLL_VERSION +# define ERTS_NO_KERNEL_POLL_VERSION 1 +# define ERTS_KERNEL_POLL_VERSION 0 # else -# define ERTS_POLL_EXPORT(FUNC) FUNC ## _nkp -# undef ERTS_POLL_DISABLE_KERNEL_POLL -# define ERTS_POLL_DISABLE_KERNEL_POLL +# undef ERTS_KERNEL_POLL_VERSION +# define ERTS_KERNEL_POLL_VERSION 1 +# define ERTS_NO_KERNEL_POLL_VERSION 0 +# define ERTS_POLL_EXPORT(FUNC) FUNC # endif #else # define ERTS_POLL_EXPORT(FUNC) FUNC -# undef ERTS_POLL_DISABLE_KERNEL_POLL -# define ERTS_POLL_DISABLE_KERNEL_POLL -#endif - -#ifdef ERTS_POLL_DISABLE_KERNEL_POLL -# undef HAVE_SYS_EPOLL_H -# undef HAVE_SYS_EVENT_H -# undef HAVE_SYS_DEVPOLL_H +# define ERTS_ENABLE_KERNEL_POLL 0 +# define ERTS_NO_KERNEL_POLL_VERSION 1 +# define ERTS_KERNEL_POLL_VERSION 0 #endif -#undef ERTS_POLL_USE_KERNEL_POLL -#define ERTS_POLL_USE_KERNEL_POLL 0 - #undef ERTS_POLL_USE_KQUEUE #define ERTS_POLL_USE_KQUEUE 0 #undef ERTS_POLL_USE_EPOLL @@ -70,68 +84,96 @@ #undef ERTS_POLL_USE_SELECT #define ERTS_POLL_USE_SELECT 0 -#if defined(HAVE_SYS_EVENT_H) -# undef ERTS_POLL_USE_KQUEUE -# define ERTS_POLL_USE_KQUEUE 1 -# undef ERTS_POLL_USE_KERNEL_POLL -# define ERTS_POLL_USE_KERNEL_POLL 1 -#elif defined(HAVE_SYS_EPOLL_H) -# undef ERTS_POLL_USE_EPOLL -# define ERTS_POLL_USE_EPOLL 1 -# undef ERTS_POLL_USE_KERNEL_POLL -# define ERTS_POLL_USE_KERNEL_POLL 1 -#elif defined(HAVE_SYS_DEVPOLL_H) -# undef ERTS_POLL_USE_DEVPOLL -# define ERTS_POLL_USE_DEVPOLL 1 -# undef ERTS_POLL_USE_KERNEL_POLL -# define ERTS_POLL_USE_KERNEL_POLL 1 +/* Defines which structure that erts_poll_wait should use to wait with + and how events should be represented */ +#define ERTS_POLL_USE_EPOLL_EVS 0 +#define ERTS_POLL_USE_KQUEUE_EVS 0 +#define ERTS_POLL_USE_DEVPOLL_EVS 0 +#define ERTS_POLL_USE_POLL_EVS 0 +#define ERTS_POLL_USE_SELECT_EVS 0 + +#define ERTS_POLL_USE_KERNEL_POLL ERTS_KERNEL_POLL_VERSION + +#if ERTS_ENABLE_KERNEL_POLL +# if defined(HAVE_SYS_EVENT_H) +# undef ERTS_POLL_USE_KQUEUE_EVS +# define ERTS_POLL_USE_KQUEUE_EVS 1 +# undef ERTS_POLL_USE_KQUEUE +# define ERTS_POLL_USE_KQUEUE ERTS_KERNEL_POLL_VERSION +# elif defined(HAVE_SYS_EPOLL_H) +# undef ERTS_POLL_USE_EPOLL_EVS +# define ERTS_POLL_USE_EPOLL_EVS 1 +# undef ERTS_POLL_USE_EPOLL +# define ERTS_POLL_USE_EPOLL ERTS_KERNEL_POLL_VERSION +# elif defined(HAVE_SYS_DEVPOLL_H) +# undef ERTS_POLL_USE_DEVPOLL_EVS +# define ERTS_POLL_USE_DEVPOLL_EVS 1 +# undef ERTS_POLL_USE_DEVPOLL +# define ERTS_POLL_USE_DEVPOLL ERTS_KERNEL_POLL_VERSION +# else +# error "Missing kernel poll implementation of erts_poll()" +# endif #endif -#define ERTS_POLL_USE_FALLBACK (ERTS_POLL_USE_KQUEUE || ERTS_POLL_USE_EPOLL) - -#if !ERTS_POLL_USE_KERNEL_POLL || ERTS_POLL_USE_FALLBACK +#if ERTS_NO_KERNEL_POLL_VERSION # if defined(ERTS_USE_POLL) +# undef ERTS_POLL_USE_POLL_EVS +# define ERTS_POLL_USE_POLL_EVS 1 # undef ERTS_POLL_USE_POLL # define ERTS_POLL_USE_POLL 1 # elif !defined(__WIN32__) +# undef ERTS_POLL_USE_SELECT_EVS +# define ERTS_POLL_USE_SELECT_EVS 1 # undef ERTS_POLL_USE_SELECT # define ERTS_POLL_USE_SELECT 1 # endif #endif -#define ERTS_POLL_USE_TIMERFD 0 +#define ERTS_POLL_USE_FALLBACK (ERTS_POLL_USE_KQUEUE || ERTS_POLL_USE_EPOLL) typedef Uint32 ErtsPollEvents; -#undef ERTS_POLL_EV_E2N + +typedef enum { + ERTS_POLL_OP_ADD = 0, /* Add the FD to the pollset */ + ERTS_POLL_OP_MOD = 1, /* Modify the FD in the pollset */ + ERTS_POLL_OP_DEL = 2 /* Delete the FD from the pollset */ +} ErtsPollOp; + +#define op2str(op) (op == ERTS_POLL_OP_ADD ? "add" : \ + (op == ERTS_POLL_OP_MOD ? "mod" : "del")) #if defined(__WIN32__) /* --- win32 --------------------------------------- */ -#define ERTS_POLL_EV_IN 1 -#define ERTS_POLL_EV_OUT 2 -#define ERTS_POLL_EV_ERR 4 -#define ERTS_POLL_EV_NVAL 8 +#define ERTS_POLL_EV_IN 1 +#define ERTS_POLL_EV_OUT 2 +#define ERTS_POLL_EV_ERR 4 +#define ERTS_POLL_EV_NVAL 8 -#elif ERTS_POLL_USE_EPOLL /* --- epoll ------------------------------- */ +#define ERTS_POLL_EV_E2N(EV) (EV) +#define ERTS_POLL_EV_N2E(EV) (EV) -#include +#elif ERTS_POLL_USE_EPOLL_EVS /* --- epoll ------------------------------- */ -#ifdef HAVE_SYS_TIMERFD_H -#include -#undef ERTS_POLL_USE_TIMERFD -#define ERTS_POLL_USE_TIMERFD 1 -#endif +#include #define ERTS_POLL_EV_E2N(EV) \ ((uint32_t) (EV)) #define ERTS_POLL_EV_N2E(EV) \ - ((ErtsPollEvents) (EV)) + ((ErtsPollEvents) (EV) & ~EPOLLONESHOT) #define ERTS_POLL_EV_IN ERTS_POLL_EV_N2E(EPOLLIN) #define ERTS_POLL_EV_OUT ERTS_POLL_EV_N2E(EPOLLOUT) #define ERTS_POLL_EV_NVAL ERTS_POLL_EV_N2E(EPOLLET) #define ERTS_POLL_EV_ERR ERTS_POLL_EV_N2E(EPOLLERR|EPOLLHUP) -#elif ERTS_POLL_USE_DEVPOLL /* --- devpoll ----------------------------- */ +typedef struct epoll_event ErtsPollResFd; + +#define ERTS_POLL_RES_GET_FD(evt) ((ErtsSysFdType)((evt)->data.fd)) +#define ERTS_POLL_RES_SET_FD(evt, ident) (evt)->data.fd = ident +#define ERTS_POLL_RES_GET_EVTS(evt) ERTS_POLL_EV_N2E((evt)->events) +#define ERTS_POLL_RES_SET_EVTS(evt, evts) (evt)->events = ERTS_POLL_EV_E2N(evts) + +#elif ERTS_POLL_USE_DEVPOLL_EVS /* --- devpoll ----------------------------- */ #include @@ -145,12 +187,37 @@ typedef Uint32 ErtsPollEvents; #define ERTS_POLL_EV_NVAL ERTS_POLL_EV_N2E(POLLNVAL) #define ERTS_POLL_EV_ERR ERTS_POLL_EV_N2E(POLLERR|POLLHUP) -#elif ERTS_POLL_USE_KQUEUE /* --- kqueue ------------------------------ */ +typedef struct pollfd ErtsPollResFd; + +#define ERTS_POLL_RES_GET_FD(evt) ((ErtsSysFdType)((evt)->fd)) +#define ERTS_POLL_RES_SET_FD(evt, ident) (evt)->fd = ident +#define ERTS_POLL_RES_GET_EVTS(evt) ERTS_POLL_EV_N2E((evt)->revents) +#define ERTS_POLL_RES_SET_EVTS(evt, evts) (evt)->revents = ERTS_POLL_EV_E2N(evts) + +#elif ERTS_POLL_USE_KQUEUE_EVS /* --- kqueue ------------------------------ */ /* Kqueue use fallback defines (poll() or select()) */ + +#include + +#ifdef ERTS_USE_POLL +# undef ERTS_POLL_USE_POLL_EVS +# define ERTS_POLL_USE_POLL_EVS 1 +#elif !defined(__WIN32__) +# undef ERTS_POLL_USE_SELECT_EVS +# define ERTS_POLL_USE_SELECT_EVS 1 #endif -#if ERTS_POLL_USE_POLL /* --- poll -------------------------------- */ +typedef struct kevent ErtsPollResFd; + +#define ERTS_POLL_RES_GET_FD(evt) ((ErtsSysFdType)((evt)->ident)) +#define ERTS_POLL_RES_SET_FD(evt, fd) (evt)->ident = fd +#define ERTS_POLL_RES_GET_EVTS(evt) ERTS_POLL_EV_N2E((ErtsPollEvents)(evt)->udata) +#define ERTS_POLL_RES_SET_EVTS(evt, evts) (evt)->udata = (void*)(UWord)(ERTS_POLL_EV_E2N(evts)) + +#endif +#if ERTS_POLL_USE_POLL_EVS + /* --- poll -------------------------------- */ #include #define ERTS_POLL_EV_NKP_E2N(EV) \ @@ -169,7 +236,7 @@ typedef Uint32 ErtsPollEvents; #define ERTS_POLL_EV_NKP_NVAL ERTS_POLL_EV_N2E(POLLNVAL) #define ERTS_POLL_EV_NKP_ERR ERTS_POLL_EV_N2E(POLLERR|POLLHUP) -#elif ERTS_POLL_USE_SELECT /* --- select ------------------------------ */ +#elif ERTS_POLL_USE_SELECT_EVS /* --- select ------------------------------ */ #define ERTS_POLL_EV_NKP_E2N(EV) (EV) #define ERTS_POLL_EV_NKP_N2E(EV) (EV) @@ -195,71 +262,65 @@ typedef Uint32 ErtsPollEvents; #endif -typedef struct ERTS_POLL_EXPORT(erts_pollset) *ErtsPollSet; +#if !ERTS_ENABLE_KERNEL_POLL -typedef struct { - ErtsSysFdType fd; - ErtsPollEvents events; - int on; -} ErtsPollControlEntry; - -typedef struct { +typedef struct _ErtsPollResFd { ErtsSysFdType fd; ErtsPollEvents events; } ErtsPollResFd; +#define ERTS_POLL_RES_GET_FD(evt) (evt)->fd +#define ERTS_POLL_RES_SET_FD(evt, ident) (evt)->fd = (ident) +#define ERTS_POLL_RES_GET_EVTS(evt) ERTS_POLL_EV_N2E((evt)->events) +#define ERTS_POLL_RES_SET_EVTS(evt, evts) (evt)->events = ERTS_POLL_EV_E2N(evts) + +#endif + +#define ERTS_POLL_EV_NONE (UINT_MAX & ~(ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT|ERTS_POLL_EV_NVAL|ERTS_POLL_EV_ERR)) + +#define ev2str(ev) \ + (((ev) == 0 || (ev) == ERTS_POLL_EV_NONE) ? "NONE" : \ + ((ev) == ERTS_POLL_EV_IN ? "IN" : \ + ((ev) == ERTS_POLL_EV_OUT ? "OUT" : \ + ((ev) == (ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT) ? "IN|OUT" : \ + ((ev) & ERTS_POLL_EV_ERR ? "ERR" : \ + ((ev) & ERTS_POLL_EV_NVAL ? "NVAL" : "OTHER")))))) + + +typedef struct ERTS_POLL_EXPORT(erts_pollset) ErtsPollSet; + typedef struct { char *primary; - char *fallback; char *kernel_poll; Uint memory_size; - int poll_set_size; - int fallback_poll_set_size; + Uint poll_set_size; int lazy_updates; - int pending_updates; + Uint pending_updates; int batch_updates; int concurrent_updates; - int max_fds; -#ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS - long no_avoided_wakeups; - long no_avoided_interrupts; - long no_interrupt_timed; -#endif - int active_fds; + int is_fallback; + Uint max_fds; + Uint active_fds; + Uint poll_threads; } ErtsPollInfo; -void ERTS_POLL_EXPORT(erts_poll_interrupt)(ErtsPollSet, - int); -void ERTS_POLL_EXPORT(erts_poll_interrupt_timed)(ErtsPollSet, - int, - ErtsMonotonicTime); -ErtsPollEvents ERTS_POLL_EXPORT(erts_poll_control)(ErtsPollSet, - ErtsSysFdType, - ErtsPollEvents, - int on, - int* wake_poller - ); -void ERTS_POLL_EXPORT(erts_poll_controlv)(ErtsPollSet, - ErtsPollControlEntry [], - int on); -int ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet, - ErtsPollResFd [], - int *, - ErtsMonotonicTime); -int ERTS_POLL_EXPORT(erts_poll_max_fds)(void); -void ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet, - ErtsPollInfo *); -ErtsPollSet ERTS_POLL_EXPORT(erts_poll_create_pollset)(void); -void ERTS_POLL_EXPORT(erts_poll_destroy_pollset)(ErtsPollSet); -void ERTS_POLL_EXPORT(erts_poll_init)(void); -void ERTS_POLL_EXPORT(erts_poll_get_selected_events)(ErtsPollSet, - ErtsPollEvents [], - int); +#if defined(ERTS_POLL_USE_FALLBACK) && ERTS_KERNEL_POLL_VERSION +# undef ERTS_POLL_EXPORT +# define ERTS_POLL_EXPORT(FUNC) FUNC ## _flbk +# include "erl_poll_api.h" +# undef ERTS_POLL_EXPORT +# define ERTS_POLL_EXPORT(FUNC) FUNC +#elif !defined(ERTS_POLL_USE_FALLBACK) +# define ERTS_POLL_USE_FALLBACK 0 +#endif -int erts_poll_new_table_len(int old_len, int need_len); +#include "erl_poll_api.h" -#ifdef ERTS_ENABLE_LOCK_COUNT -void ERTS_POLL_EXPORT(erts_lcnt_update_pollset_locks)(int enable); -#endif +/** + * Get the next size of the array that holds the file descriptors. + * This function is used in order for the check io array and the + * pollset array to be of the same size. + */ +int erts_poll_new_table_len(int old_len, int need_len); #endif /* #ifndef ERL_POLL_H__ */ diff --git a/erts/emulator/sys/common/erl_poll_api.h b/erts/emulator/sys/common/erl_poll_api.h new file mode 100644 index 0000000000..04beb37d1c --- /dev/null +++ b/erts/emulator/sys/common/erl_poll_api.h @@ -0,0 +1,122 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2006-2016. All Rights Reserved. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * + * %CopyrightEnd% + */ +/** + * @description: Poll interface functions + * @author Lukas Larsson + * + * The functions in the header are used to interact with the poll + * implementation. Iff the kernel-poll implementation needs a fallback + * pollset, then all functions are exported twice. Once with a _flbk + * suffix and once without any suffix. If no fallback is needed, then + * only the non-suffix version is exported. + */ + +/** + * Initialize the poll implementation. Has to be called before any other function. + * @param[out] concurrent_waiters if not NULL, set to 1 if more then one thread + * is allowed to wait in the pollsets at the same time. + */ +void ERTS_POLL_EXPORT(erts_poll_init)(int *concurrent_waiters); +/** + * @brief Create a new pollset. + * @param id The unique debug id of this pollset. + */ +ErtsPollSet *ERTS_POLL_EXPORT(erts_poll_create_pollset)(int id); + +/** + * Modify the contents of a pollset. This function can be called while one + * (or possibly more) thread is waiting in the pollset. + * + * @param ps the pollset to modify + * @param fd the file descriptor to modify + * @param op the type of operation to do. Normal usage is ADD,MOD...MOD,DEL. + * @param evts the events that we are changing interest to. Ignored if op is DEL. + * @param[in] wake_poller if set to 1 any thread waiting in the pollset will be woken. + * This parameter is ignored if the pollset supports concurrent waiters. + * @param[out] wake_poller set to 1 if the waiting thread was woken. + * @return The events set, or ERTS_POLL_EV_NVAL if it was not possible to add the + * fd to the pollset. + */ +ErtsPollEvents ERTS_POLL_EXPORT(erts_poll_control)(ErtsPollSet *ps, + ErtsSysFdType fd, + ErtsPollOp op, + ErtsPollEvents evts, + int *wake_poller); + +/** + * Wait for events to be ready in the pollset. If the erts_poll_init call + * set concurrent_waiters to 1, then multiple threads are allowed to call + * this function at the same time. + * + * When an event has been triggered on a fd, that event is disabled. To + * re-enable it the implementation has to call erts_poll_control again. + * + * @param ps the pollset to wait for events in + * @param res an array of fd results that the ready fds are put in. + * @param[in] length the length of the res array + * @param[out] length the number of ready events returned in res + * @return 0 on success, else the ERRNO of the error that happened. + */ +int ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet *ps, + ErtsPollResFd res[], + int *length); +/** + * Interrupt the thread waiting in the pollset. This function should be called + * with set = 0 before any thread calls erts_poll_wait in order to clear any + * interrupts that have happened while the thread was awake. + * + * This function has no effect on pollsets that support concurrent waiters. + * + * @param ps the pollset to wake + * @param set if 1, interrupt the pollset, if 0 clear the interrupt flag. + */ +void ERTS_POLL_EXPORT(erts_poll_interrupt)(ErtsPollSet *ps, int set); + +/* Debug functions */ + +/** + * Get the maximum number of fds supported by the pollset + */ +int ERTS_POLL_EXPORT(erts_poll_max_fds)(void); +/** + * Get information about the given pollset + */ +void ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet *ps, + ErtsPollInfo *info); +/** + * Get information about which events are currently selected. + * + * The unix fd is used to index into the array, so naturally this function does + * not work on windows. If the pollset cannot figure out what the selected + * events for a given fd is, it is set to ERTS_POLL_EV_NONE. + * + * @param ps the pollset to get events from + * @param evts an array of which events are selected on. + */ +void ERTS_POLL_EXPORT(erts_poll_get_selected_events)(ErtsPollSet *ps, + ErtsPollEvents evts[], + int length); + +#ifdef ERTS_ENABLE_LOCK_COUNT +/** + * Enable lock counting of any locks within the pollset. + */ +void ERTS_POLL_EXPORT(erts_lcnt_enable_pollset_lock_count)(ErtsPollSet *, int enable); +#endif diff --git a/erts/emulator/sys/unix/sys.c b/erts/emulator/sys/unix/sys.c index acd7920e86..6315135151 100644 --- a/erts/emulator/sys/unix/sys.c +++ b/erts/emulator/sys/unix/sys.c @@ -499,22 +499,16 @@ static void signal_notify_requested(Eterm type) { static ERTS_INLINE void break_requested(void) { - int i; /* * just set a flag - checked for and handled by * scheduler threads erts_check_io() (not signal handler). */ -#ifdef DEBUG - fprintf(stderr,"break!\n"); -#endif if (ERTS_BREAK_REQUESTED) erts_exit(ERTS_INTR_EXIT, ""); ERTS_SET_BREAK_REQUESTED; - for (i=0; i < erts_no_schedulers; i++) { - /* Make sure we don't sleep in poll */ - erts_check_io_interrupt(ERTS_SCHEDULER_IX(i)->pollset, 1); - } + /* Wake aux thread to get handle break */ + erts_aux_thread_poke(); } static RETSIGTYPE request_break(int signum) @@ -1079,20 +1073,6 @@ erl_debug(char* fmt, ...) #endif /* DEBUG */ -/* - * Called from schedule() when it runs out of runnable processes, - * or when Erlang code has performed INPUT_REDUCTIONS reduction - * steps. runnable == 0 iff there are no runnable Erlang processes. - */ -void -erl_sys_schedule(int runnable) -{ - erts_check_io(!runnable); - ERTS_LC_ASSERT(!erts_thr_progress_is_blocking()); -} - - - static erts_tid_t sig_dispatcher_tid; static void @@ -1281,103 +1261,18 @@ erts_sys_main_thread(void) } -#ifdef ERTS_ENABLE_KERNEL_POLL /* get_value() is currently only used when - kernel-poll is enabled */ - -/* Get arg marks argument as handled by - putting NULL in argv */ -static char * -get_value(char* rest, char** argv, int* ip) -{ - char *param = argv[*ip]+1; - argv[*ip] = NULL; - if (*rest == '\0') { - char *next = argv[*ip + 1]; - if (next[0] == '-' - && next[1] == '-' - && next[2] == '\0') { - erts_fprintf(stderr, "bad \"%s\" value: \n", param); - erts_usage(); - } - (*ip)++; - argv[*ip] = NULL; - return next; - } - return rest; -} - -int erts_use_kernel_poll = 0; - -#endif /* ERTS_ENABLE_KERNEL_POLL */ - void erl_sys_args(int* argc, char** argv) { - int i, j; erts_rwmtx_init(&environ_rwmtx, "environ", NIL, ERTS_LOCK_FLAGS_PROPERTY_STATIC | ERTS_LOCK_FLAGS_CATEGORY_GENERIC); - i = 1; - ASSERT(argc && argv); - while (i < *argc) { - if(argv[i][0] == '-') { - switch (argv[i][1]) { -#ifdef ERTS_ENABLE_KERNEL_POLL - case 'K': { - char *arg = get_value(argv[i] + 2, argv, &i); - if (strcmp("true", arg) == 0) { - erts_use_kernel_poll = 1; - } - else if (strcmp("false", arg) == 0) { - erts_use_kernel_poll = 0; - } - else { - erts_fprintf(stderr, "bad \"K\" value: %s\n", arg); - erts_usage(); - } - break; - } -#endif - case '-': - goto done_parsing; - default: - break; - } - } - i++; - } - - done_parsing: - -#ifdef ERTS_ENABLE_KERNEL_POLL - if (erts_use_kernel_poll) { - char no_kp[10]; - size_t no_kp_sz = sizeof(no_kp); - int res = erts_sys_getenv_raw("ERL_NO_KERNEL_POLL", no_kp, &no_kp_sz); - if (res > 0 - || (res == 0 - && sys_strcmp("false", no_kp) != 0 - && sys_strcmp("FALSE", no_kp) != 0)) { - erts_use_kernel_poll = 0; - } - } -#endif - - erts_init_check_io(); max_files = erts_check_io_max_files(); - init_smp_sig_notify(); init_smp_sig_suspend(); - /* Handled arguments have been marked with NULL. Slide arguments - not handled towards the beginning of argv. */ - for (i = 0, j = 0; i < *argc; i++) { - if (argv[i]) - argv[j++] = argv[i]; - } - *argc = j; } diff --git a/erts/emulator/sys/win32/erl_poll.c b/erts/emulator/sys/win32/erl_poll.c index 7a87e81141..24c8c7bbc7 100644 --- a/erts/emulator/sys/win32/erl_poll.c +++ b/erts/emulator/sys/win32/erl_poll.c @@ -34,6 +34,7 @@ */ /*#define HARDDEBUG */ +/*#define HARDTRACE */ #ifdef HARDDEBUG #ifdef HARDTRACE #define HARDTRACEF(X) my_debug_printf##X @@ -50,7 +51,7 @@ static void my_debug_printf(char *fmt, ...) va_start(args, fmt); erts_vsnprintf(buffer,1024,fmt,args); va_end(args); - erts_fprintf(stderr,"%s\r\n",buffer); + erts_printf("%s\r\n",buffer); } #else #define HARDTRACEF(X) @@ -278,16 +279,13 @@ struct erts_pollset { Waiter** waiter; int allocated_waiters; /* Size ow waiter array */ int num_waiters; /* Number of waiter threads. */ - int restore_events; /* Tells us to restore waiters events - next time around */ HANDLE event_io_ready; /* To be used when waiting for io */ /* These are used to wait for workers to enter standby */ volatile int standby_wait_counter; /* Number of threads to wait for */ CRITICAL_SECTION standby_crit; /* CS to guard the counter */ - HANDLE standby_wait_event; /* Event signalled when counte == 0 */ + HANDLE standby_wait_event; /* Event signalled when counter == 0 */ erts_atomic32_t wakeup_state; erts_mtx_t mtx; - erts_atomic64_t timeout_time; }; @@ -352,39 +350,19 @@ do { \ wait_standby(PS); \ } while(0) -static ERTS_INLINE void -init_timeout_time(ErtsPollSet ps) -{ - erts_atomic64_init_nob(&ps->timeout_time, - (erts_aint64_t) ERTS_MONOTONIC_TIME_MAX); -} - -static ERTS_INLINE void -set_timeout_time(ErtsPollSet ps, ErtsMonotonicTime time) -{ - erts_atomic64_set_relb(&ps->timeout_time, - (erts_aint64_t) time); -} - -static ERTS_INLINE ErtsMonotonicTime -get_timeout_time(ErtsPollSet ps) -{ - return (ErtsMonotonicTime) erts_atomic64_read_acqb(&ps->timeout_time); -} - #define ERTS_POLL_NOT_WOKEN ((erts_aint32_t) 0) #define ERTS_POLL_WOKEN_IO_READY ((erts_aint32_t) 1) #define ERTS_POLL_WOKEN_INTR ((erts_aint32_t) 2) #define ERTS_POLL_WOKEN_TIMEDOUT ((erts_aint32_t) 3) static ERTS_INLINE int -is_io_ready(ErtsPollSet ps) +is_io_ready(ErtsPollSet *ps) { return erts_atomic32_read_nob(&ps->wakeup_state) == ERTS_POLL_WOKEN_IO_READY; } static ERTS_INLINE void -woke_up(ErtsPollSet ps) +woke_up(ErtsPollSet *ps) { if (erts_atomic32_read_nob(&ps->wakeup_state) == ERTS_POLL_NOT_WOKEN) erts_atomic32_cmpxchg_nob(&ps->wakeup_state, @@ -407,7 +385,7 @@ woke_up(ErtsPollSet ps) } static ERTS_INLINE int -wakeup_cause(ErtsPollSet ps) +wakeup_cause(ErtsPollSet *ps) { int res; erts_aint32_t wakeup_state = erts_atomic32_read_acqb(&ps->wakeup_state); @@ -430,46 +408,8 @@ wakeup_cause(ErtsPollSet ps) return res; } -static ERTS_INLINE DWORD -poll_wait_timeout(ErtsPollSet ps, ErtsMonotonicTime timeout_time) -{ - ErtsMonotonicTime current_time, diff_time, timeout; - - if (timeout_time == ERTS_POLL_NO_TIMEOUT) { - no_timeout: - set_timeout_time(ps, ERTS_MONOTONIC_TIME_MIN); - woke_up(ps); - return (DWORD) 0; - } - - current_time = erts_get_monotonic_time(NULL); - diff_time = timeout_time - current_time; - if (diff_time <= 0) - goto no_timeout; - - /* Round up to nearest milli second */ - timeout = (ERTS_MONOTONIC_TO_MSEC(diff_time - 1) + 1); - if (timeout > INT_MAX) - timeout = INT_MAX; /* Also prevents DWORD overflow */ - - set_timeout_time(ps, current_time + ERTS_MSEC_TO_MONOTONIC(timeout)); - - ResetEvent(ps->event_io_ready); - /* - * Since we don't know the internals of ResetEvent() we issue - * a memory barrier as a safety precaution ensuring that - * the load of wakeup_state wont be reordered with stores made - * by ResetEvent(). - */ - ERTS_THR_MEMORY_BARRIER; - if (erts_atomic32_read_nob(&ps->wakeup_state) != ERTS_POLL_NOT_WOKEN) - return (DWORD) 0; - - return (DWORD) timeout; -} - static ERTS_INLINE void -wake_poller(ErtsPollSet ps, int io_ready) +wake_poller(ErtsPollSet *ps, int io_ready) { erts_aint32_t wakeup_state; if (io_ready) { @@ -504,13 +444,13 @@ wake_poller(ErtsPollSet ps, int io_ready) } static ERTS_INLINE void -reset_io_ready(ErtsPollSet ps) +reset_io_ready(ErtsPollSet *ps) { erts_atomic32_set_nob(&ps->wakeup_state, ERTS_POLL_NOT_WOKEN); } static ERTS_INLINE void -restore_io_ready(ErtsPollSet ps) +restore_io_ready(ErtsPollSet *ps) { erts_atomic32_set_nob(&ps->wakeup_state, ERTS_POLL_WOKEN_IO_READY); } @@ -520,13 +460,13 @@ restore_io_ready(ErtsPollSet ps) * notifying a poller thread about I/O ready. */ static ERTS_INLINE void -notify_io_ready(ErtsPollSet ps) +notify_io_ready(ErtsPollSet *ps) { wake_poller(ps, 1); } static ERTS_INLINE void -reset_interrupt(ErtsPollSet ps) +reset_interrupt(ErtsPollSet *ps) { /* We need to keep io-ready if set */ erts_aint32_t wakeup_state = erts_atomic32_read_nob(&ps->wakeup_state); @@ -543,12 +483,12 @@ reset_interrupt(ErtsPollSet ps) } static ERTS_INLINE void -set_interrupt(ErtsPollSet ps) +set_interrupt(ErtsPollSet *ps) { wake_poller(ps, 0); } -static void setup_standby_wait(ErtsPollSet ps, int num_threads) +static void setup_standby_wait(ErtsPollSet *ps, int num_threads) { EnterCriticalSection(&(ps->standby_crit)); ps->standby_wait_counter = num_threads; @@ -556,7 +496,7 @@ static void setup_standby_wait(ErtsPollSet ps, int num_threads) LeaveCriticalSection(&(ps->standby_crit)); } -static void signal_standby(ErtsPollSet ps) +static void signal_standby(ErtsPollSet *ps) { EnterCriticalSection(&(ps->standby_crit)); --(ps->standby_wait_counter); @@ -570,7 +510,7 @@ static void signal_standby(ErtsPollSet ps) LeaveCriticalSection(&(ps->standby_crit)); } -static void wait_standby(ErtsPollSet ps) +static void wait_standby(ErtsPollSet *ps) { WaitForSingleObject(ps->standby_wait_event,INFINITE); } @@ -638,7 +578,7 @@ static void consistency_check(Waiter* w) #endif -static void new_waiter(ErtsPollSet ps) +static void new_waiter(ErtsPollSet *ps) { register Waiter* w; DWORD tid; /* Id for thread. */ @@ -732,7 +672,7 @@ static void *break_waiter(void *param) static void *threaded_waiter(void *param) { register Waiter* w = (Waiter *) param; - ErtsPollSet ps = (ErtsPollSet) w->xdata; + ErtsPollSet *ps = (ErtsPollSet*) w->xdata; #ifdef HARD_POLL_DEBUG2 HANDLE oold_fired[64]; int num_oold_fired; @@ -835,9 +775,9 @@ event_happened: ASSERT(i >= WAIT_OBJECT_0+1); i -= WAIT_OBJECT_0; ASSERT(i >= 1); - w->active_events--; HARDDEBUGF(("i = %d, a,h,t = %d,%d,%d",i, w->active_events, w->highwater, w->total_events)); + w->active_events--; #ifdef HARD_POLL_DEBUG2 fired[num_fired++] = w->events[i]; #endif @@ -867,7 +807,7 @@ event_happened: * The actual adding and removing from pollset utilities */ -static int set_driver_select(ErtsPollSet ps, HANDLE event, ErtsPollEvents mode) +static int set_driver_select(ErtsPollSet *ps, HANDLE event, ErtsPollEvents mode) { int i; int best_waiter = -1; /* The waiter with lowest number of events. */ @@ -957,13 +897,13 @@ static int set_driver_select(ErtsPollSet ps, HANDLE event, ErtsPollEvents mode) #endif erts_mtx_unlock(&w->mtx); START_WAITER(ps,w); - HARDDEBUGF(("add select %d %d %d %d",best_waiter, + HARDDEBUGF(("%d: add select %d %d %d %d", event, best_waiter, w->active_events,w->highwater,w->total_events)); return mode; } -static int cancel_driver_select(ErtsPollSet ps, HANDLE event) +static int cancel_driver_select(ErtsPollSet *ps, HANDLE event) { int i; @@ -1018,7 +958,7 @@ static int cancel_driver_select(ErtsPollSet ps, HANDLE event) * Interface functions */ -void erts_poll_interrupt(ErtsPollSet ps, int set /* bool */) +void erts_poll_interrupt(ErtsPollSet *ps, int set /* bool */) { HARDTRACEF(("In erts_poll_interrupt(%d)",set)); if (!set) @@ -1028,35 +968,23 @@ void erts_poll_interrupt(ErtsPollSet ps, int set /* bool */) HARDTRACEF(("Out erts_poll_interrupt(%d)",set)); } -void erts_poll_interrupt_timed(ErtsPollSet ps, - int set /* bool */, - ErtsMonotonicTime timeout_time) -{ - HARDTRACEF(("In erts_poll_interrupt_timed(%d,%ld)",set,timeout_time)); - if (!set) - reset_interrupt(ps); - else if (get_timeout_time(ps) > timeout_time) - set_interrupt(ps); - HARDTRACEF(("Out erts_poll_interrupt_timed")); -} - /* * Windows is special, there is actually only one event type, and * the only difference between ERTS_POLL_EV_IN and ERTS_POLL_EV_OUT * is which driver callback will eventually be called. */ -static ErtsPollEvents do_poll_control(ErtsPollSet ps, - ErtsSysFdType fd, - ErtsPollEvents pe, - int on /* bool */) +static ErtsPollEvents do_poll_control(ErtsPollSet *ps, + ErtsSysFdType fd, + ErtsPollOp op, + ErtsPollEvents pe) { HANDLE event = (HANDLE) fd; ErtsPollEvents mode; ErtsPollEvents result; ASSERT(event != INVALID_HANDLE_VALUE); - if (on) { + if (op != ERTS_POLL_OP_DEL) { if (pe & ERTS_POLL_EV_IN || !(pe & ERTS_POLL_EV_OUT )) { mode = ERTS_POLL_EV_IN; } else { @@ -1069,51 +997,30 @@ static ErtsPollEvents do_poll_control(ErtsPollSet ps, return result; } -ErtsPollEvents erts_poll_control(ErtsPollSet ps, +ErtsPollEvents erts_poll_control(ErtsPollSet *ps, ErtsSysFdType fd, + ErtsPollOp op, ErtsPollEvents pe, - int on, int* do_wake) /* In: Wake up polling thread */ /* Out: Poller is woken */ { ErtsPollEvents result; - HARDTRACEF(("In erts_poll_control(0x%08X, %u, %d)",(unsigned long) fd, (unsigned) pe, on)); + HARDTRACEF(("In erts_poll_control(0x%08X, %s, %s)", + (unsigned long) fd, op2str(op), ev2str(pe))); ERTS_POLLSET_LOCK(ps); - result=do_poll_control(ps,fd,pe,on); + result=do_poll_control(ps, fd, op, pe); ERTS_POLLSET_UNLOCK(ps); *do_wake = 0; /* Never any need to wake polling threads on windows */ HARDTRACEF(("Out erts_poll_control -> %u",(unsigned) result)); return result; } -void erts_poll_controlv(ErtsPollSet ps, - ErtsPollControlEntry pcev[], - int len) -{ - int i; - int hshur = 0; - int do_wake = 0; - - HARDTRACEF(("In erts_poll_controlv(%d)",len)); - ERTS_POLLSET_LOCK(ps); - - for (i = 0; i < len; i++) { - pcev[i].events = do_poll_control(ps, - pcev[i].fd, - pcev[i].events, - pcev[i].on); - } - ERTS_POLLSET_UNLOCK(ps); - HARDTRACEF(("Out erts_poll_controlv")); -} - -int erts_poll_wait(ErtsPollSet ps, +int erts_poll_wait(ErtsPollSet *ps, ErtsPollResFd pr[], - int *len, - ErtsMonotonicTime timeout_time) + int *len) { int no_fds; - DWORD timeout; + DWORD timeout = INFINITE; EventData* ev; int res = 0; int num = 0; @@ -1124,42 +1031,6 @@ int erts_poll_wait(ErtsPollSet ps, HARDTRACEF(("In erts_poll_wait")); ERTS_POLLSET_LOCK(ps); - if (!is_io_ready(ps) && ps->restore_events) { - HARDDEBUGF(("Restore events: %d",ps->num_waiters)); - ps->restore_events = 0; - for (i = 0; i < ps->num_waiters; ++i) { - Waiter* w = ps->waiter[i]; - erts_mtx_lock(&w->mtx); - HARDDEBUGF(("Maybe reset %d %d %d %d",i, - w->active_events,w->highwater,w->total_events)); - if (w->active_events < w->total_events) { - erts_mtx_unlock(&w->mtx); - STOP_WAITER(ps,w); - HARDDEBUGF(("Need reset %d %d %d %d",i, - w->active_events,w->highwater,w->total_events)); - erts_mtx_lock(&w->mtx); - /* Need reset, just check that it doesn't have got more to tell */ - if (w->highwater != w->active_events) { - HARDDEBUGF(("Oups!")); - /* Oups, got signalled before we took the lock, can't reset */ - if(!is_io_ready(ps)) { - erts_exit(ERTS_ERROR_EXIT,"Internal error: " - "Inconsistent io structures in erl_poll.\n"); - } - START_WAITER(ps,w); - erts_mtx_unlock(&w->mtx); - ps->restore_events = 1; - continue; - } - w->active_events = w->highwater = w->total_events; - START_WAITER(ps,w); - erts_mtx_unlock(&w->mtx); - } else { - erts_mtx_unlock(&w->mtx); - } - } - } - no_fds = *len; #ifdef ERTS_POLL_MAX_RES @@ -1167,11 +1038,18 @@ int erts_poll_wait(ErtsPollSet ps, no_fds = ERTS_POLL_MAX_RES; #endif - timeout = poll_wait_timeout(ps, timeout_time); - - /*HARDDEBUGF(("timeout = %ld",(long) timeout));*/ + ResetEvent(ps->event_io_ready); + /* + * Since we don't know the internals of ResetEvent() we issue + * a memory barrier as a safety precaution ensuring that + * the load of wakeup_state wont be reordered with stores made + * by ResetEvent(). + */ + ERTS_THR_MEMORY_BARRIER; + if (erts_atomic32_read_nob(&ps->wakeup_state) != ERTS_POLL_NOT_WOKEN) + timeout = (DWORD) 0; - if (timeout > 0 && !erts_atomic32_read_nob(&break_waiter_state)) { + if (!erts_atomic32_read_nob(&break_waiter_state)) { HANDLE harr[2] = {ps->event_io_ready, break_happened_event}; int num_h = 2; ERTS_MSACC_PUSH_STATE_M(); @@ -1215,7 +1093,7 @@ int erts_poll_wait(ErtsPollSet ps, reset_io_ready(ps); - n = ps->num_waiters; + n = ps->num_waiters; for (i = 0; i < n; i++) { Waiter* w = ps->waiter[i]; @@ -1241,11 +1119,10 @@ int erts_poll_wait(ErtsPollSet ps, HARDDEBUGF(("To many FD's to report!")); goto done; } - HARDDEBUGF(("SET! Restore events")); - ps->restore_events = 1; HARDDEBUGF(("Report %d,%d",i,j)); - pr[num].fd = (ErtsSysFdType) w->events[j]; - pr[num].events = w->evdata[j]->mode; + ERTS_POLL_RES_SET_FD(&pr[num], w->events[j]); + ERTS_POLL_RES_SET_EVTS(&pr[num], w->evdata[j]->mode); + remove_event_from_set(w, j); #ifdef HARD_POLL_DEBUG poll_debug_reported(w->events[j],w->highwater | (j << 16)); poll_debug_reported(w->events[j],first | (last << 16)); @@ -1253,13 +1130,14 @@ int erts_poll_wait(ErtsPollSet ps, ++num; } + w->total_events = w->highwater = w->active_events; + #ifdef DEBUG consistency_check(w); #endif erts_mtx_unlock(&w->mtx); } done: - set_timeout_time(ps, ERTS_MONOTONIC_TIME_MAX); *len = num; ERTS_POLLSET_UNLOCK(ps); HARDTRACEF(("Out erts_poll_wait")); @@ -1274,7 +1152,7 @@ int erts_poll_max_fds(void) return res; } -void erts_poll_info(ErtsPollSet ps, +void erts_poll_info(ErtsPollSet *ps, ErtsPollInfo *pip) { Uint size = 0; @@ -1299,16 +1177,12 @@ void erts_poll_info(ErtsPollSet ps, pip->primary = "WaitForMultipleObjects"; - pip->fallback = NULL; - pip->kernel_poll = NULL; pip->memory_size = size; pip->poll_set_size = num_events; - pip->fallback_poll_set_size = 0; - pip->lazy_updates = 0; pip->pending_updates = 0; @@ -1316,6 +1190,8 @@ void erts_poll_info(ErtsPollSet ps, pip->batch_updates = 0; pip->concurrent_updates = 0; + + pip->is_fallback = 0; ERTS_POLLSET_UNLOCK(ps); pip->max_fds = erts_poll_max_fds(); @@ -1323,9 +1199,9 @@ void erts_poll_info(ErtsPollSet ps, } -ErtsPollSet erts_poll_create_pollset(void) +ErtsPollSet *erts_poll_create_pollset(int no) { - ErtsPollSet ps = SEL_ALLOC(ERTS_ALC_T_POLLSET, + ErtsPollSet *ps = SEL_ALLOC(ERTS_ALC_T_POLLSET, sizeof(struct erts_pollset)); HARDTRACEF(("In erts_poll_create_pollset")); @@ -1337,17 +1213,15 @@ ErtsPollSet erts_poll_create_pollset(void) ps->standby_wait_counter = 0; ps->event_io_ready = CreateManualEvent(FALSE); ps->standby_wait_event = CreateManualEvent(FALSE); - ps->restore_events = 0; erts_atomic32_init_nob(&ps->wakeup_state, ERTS_POLL_NOT_WOKEN); erts_mtx_init(&ps->mtx, "pollset", NIL, ERTS_LOCK_FLAGS_CATEGORY_IO); - init_timeout_time(ps); HARDTRACEF(("Out erts_poll_create_pollset")); return ps; } -void erts_poll_destroy_pollset(ErtsPollSet ps) +void erts_poll_destroy_pollset(ErtsPollSet *ps) { int i; HARDTRACEF(("In erts_poll_destroy_pollset")); @@ -1378,14 +1252,16 @@ void erts_poll_destroy_pollset(ErtsPollSet ps) /* * Actually mostly initializes the friend module sys_interrupt... */ -void erts_poll_init(void) +void erts_poll_init(int *concurrent_updates) { - erts_tid_t thread; #ifdef HARD_POLL_DEBUG poll_debug_init(); #endif + if (concurrent_updates) + *concurrent_updates = 0; + HARDTRACEF(("In erts_poll_init")); erts_sys_break_event = CreateManualEvent(FALSE); @@ -1394,21 +1270,26 @@ void erts_poll_init(void) break_happened_event = CreateManualEvent(FALSE); erts_atomic32_init_nob(&break_waiter_state, 0); + HARDTRACEF(("Out erts_poll_init")); +} + +void erts_poll_late_init(void) +{ + erts_tid_t thread; erts_thr_create(&thread, &break_waiter, NULL, NULL); ERTS_UNSET_BREAK_REQUESTED; - HARDTRACEF(("Out erts_poll_init")); } /* * Non windows friendly interface, not used when fd's are not continous */ -void erts_poll_get_selected_events(ErtsPollSet ps, +void erts_poll_get_selected_events(ErtsPollSet *ps, ErtsPollEvents ev[], int len) { int i; HARDTRACEF(("In erts_poll_get_selected_events")); for (i = 0; i < len; ++i) - ev[i] = 0; + ev[i] = ERTS_POLL_EV_NONE; HARDTRACEF(("Out erts_poll_get_selected_events")); } diff --git a/erts/emulator/sys/win32/sys.c b/erts/emulator/sys/win32/sys.c index de391f078b..0598a12351 100644 --- a/erts/emulator/sys/win32/sys.c +++ b/erts/emulator/sys/win32/sys.c @@ -3235,27 +3235,16 @@ void erl_sys_init(void) SetStdHandle(STD_ERROR_HANDLE, GetStdHandle(STD_OUTPUT_HANDLE)); } erts_sys_init_float(); - erts_init_check_io(); /* Suppress windows error message popups */ SetErrorMode(SetErrorMode(0) | SEM_FAILCRITICALERRORS | SEM_NOOPENFILEERRORBOX); } +void erts_poll_late_init(void); void erl_sys_late_init(void) { /* do nothing */ -} - -/* - * Called from schedule() when it runs out of runnable processes, - * or when Erlang code has performed INPUT_REDUCTIONS reduction - * steps. runnable == 0 iff there are no runnable Erlang processes. - */ -void -erl_sys_schedule(int runnable) -{ - erts_check_io(!runnable); - ERTS_LC_ASSERT(!erts_thr_progress_is_blocking()); + erts_poll_late_init(); } diff --git a/erts/emulator/test/driver_SUITE.erl b/erts/emulator/test/driver_SUITE.erl index 332d11345b..0e691b0c80 100644 --- a/erts/emulator/test/driver_SUITE.erl +++ b/erts/emulator/test/driver_SUITE.erl @@ -136,7 +136,7 @@ suite() -> {timetrap, {minutes, 1}}]. all() -> %% Keep a_test first and z_test last... - [a_test, outputv_errors, outputv_echo, queue_echo, {group, timer}, + [a_test, outputv_errors, outputv_echo, queue_echo, %{group, timer}, driver_unloaded, io_ready_exit, use_fallback_pollset, bad_fd_in_pollset, fd_change, steal_control, otp_6602, driver_system_info_base_ver, @@ -785,12 +785,9 @@ io_ready_exit(Config) when is_list(Config) -> use_fallback_pollset(Config) when is_list(Config) -> FlbkFun = fun () -> - ChkIoDuring = get_check_io_total(erlang:system_info(check_io)), - case lists:keysearch(fallback_poll_set_size, - 1, - ChkIoDuring) of - {value, - {fallback_poll_set_size, N}} when N > 0 -> + {Flbk, _} = get_fallback(erlang:system_info(check_io)), + case lists:keysearch(total_poll_set_size, 1, Flbk) of + {value, {total_poll_set_size, N}} when N > 0 -> ok; Error -> ct:fail({failed_to_use_fallback, Error}) @@ -1013,14 +1010,15 @@ get_stable_check_io_info() -> %% Merge return from erlang:system_info(check_io) %% as if it was one big pollset. get_check_io_total(ChkIo) -> - lists:foldl(fun(Pollset, Acc) -> - lists:zipwith(fun(A, B) -> - add_pollset_infos(A,B) - end, - Pollset, Acc) - end, - hd(ChkIo), - tl(ChkIo)). + {Fallback, Rest} = get_fallback(ChkIo), + add_fallback_infos(Fallback, + lists:foldl(fun(Pollset, Acc) -> + lists:zipwith(fun(A, B) -> + add_pollset_infos(A,B) + end, + Pollset, Acc) + end, + hd(Rest), tl(Rest))). add_pollset_infos({Tag, A}=TA , {Tag, B}=TB) -> case tag_type(Tag) of @@ -1035,13 +1033,32 @@ add_pollset_infos({Tag, A}=TA , {Tag, B}=TB) -> end end. +get_fallback([MaybeFallback | ChkIo] = AllChkIo) -> + case proplists:get_value(fallback, MaybeFallback) of + true -> + {MaybeFallback, ChkIo}; + false -> + {undefined, AllChkIo} + end. + +add_fallback_infos(undefined, Acc) -> + Acc; +add_fallback_infos(Flbk, Acc) -> + lists:zipwith(fun({Tag, A}=TA, {Tag, B}=TB) -> + case tag_type(Tag) of + sum -> {Tag, A + B}; + const when Tag =:= fallback -> TA; + const -> TB + end + end, + Flbk, Acc). + tag_type(name) -> const; tag_type(primary) -> const; tag_type(fallback) -> const; tag_type(kernel_poll) -> const; tag_type(memory_size) -> sum; tag_type(total_poll_set_size) -> sum; -tag_type(fallback_poll_set_size) -> sum; tag_type(lazy_updates) -> const; tag_type(pending_updates) -> sum; tag_type(batch_updates) -> const; diff --git a/erts/emulator/test/driver_SUITE_data/chkio_drv.c b/erts/emulator/test/driver_SUITE_data/chkio_drv.c index 5167036602..d548c4b1dc 100644 --- a/erts/emulator/test/driver_SUITE_data/chkio_drv.c +++ b/erts/emulator/test/driver_SUITE_data/chkio_drv.c @@ -511,6 +511,9 @@ chkio_drv_ready_input(ErlDrvData drv_data, ErlDrvEvent event) driver_failure_atom(cddp->port, "input_fd_not_found"); break; } + case CHKIO_FD_CHANGE: + /* This may be triggered when an fd is closed while being selected on. */ + break; case CHKIO_STEAL: break; case CHKIO_STEAL_AUX: diff --git a/erts/etc/common/erlexec.c b/erts/etc/common/erlexec.c index 6b194e25da..d61a3cbf95 100644 --- a/erts/etc/common/erlexec.c +++ b/erts/etc/common/erlexec.c @@ -810,6 +810,28 @@ int main(int argc, char **argv) add_Eargs(argv[i+1]); i++; break; + case 'I': + if (argv[i][2] == 'O' && (argv[i][3] == 't' || argv[i][3] == 'p')) { + if (argv[i][4] != '\0') + goto the_default; + argv[i][0] = '-'; + add_Eargs(argv[i]); + add_Eargs(argv[i+1]); + i++; + break; + } + if (argv[i][2] == 'O' && argv[i][3] == 'P' && + (argv[i][4] == 't' || argv[i][4] == 'p')) { + if (argv[i][5] != '\0') + goto the_default; + argv[i][0] = '-'; + add_Eargs(argv[i]); + add_Eargs(argv[i+1]); + i++; + break; + } + usage(argv[i]); + break; case 'S': if (argv[i][2] == 'P') { if (argv[i][3] != '\0') diff --git a/erts/etc/unix/etp-commands.in b/erts/etc/unix/etp-commands.in index 8f70f879d5..95e065b156 100644 --- a/erts/etc/unix/etp-commands.in +++ b/erts/etc/unix/etp-commands.in @@ -2316,40 +2316,46 @@ end define etp-port-sched-flags-int # Args: int # - if ($arg0 & 0x1) + if ($arg0 & (1 << 0)) printf " in-run-queue" end - if ($arg0 & 0x2) + if ($arg0 & (1 << 1)) printf " executing" end - if ($arg0 & 0x4) + if ($arg0 & (1 << 2)) printf " have-tasks" end - if ($arg0 & 0x8) + if ($arg0 & (1 << 3)) printf " exited" end - if ($arg0 & 0x10) + if ($arg0 & (1 << 4)) printf " busy-port" end - if ($arg0 & 0x20) + if ($arg0 & (1 << 5)) printf " busy-port-q" end - if ($arg0 & 0x40) + if ($arg0 & (1 << 6)) printf " chk-unset-busy-port-q" end - if ($arg0 & 0x80) + if ($arg0 & (1 << 7)) printf " have-busy-tasks" end - if ($arg0 & 0x100) + if ($arg0 & (1 << 8)) printf " have-nosuspend-tasks" end - if ($arg0 & 0x200) + if ($arg0 & (1 << 9)) printf " parallelism" end - if ($arg0 & 0x400) + if ($arg0 & (1 << 10)) printf " force-sched" end - if ($arg0 & 0xfffff800) + if ($arg0 & (1 << 11)) + printf " exiting" + end + if ($arg0 & (1 << 12)) + printf " exec-imm" + end + if ($arg0 & 0xffffc000) printf " GARBAGE" end printf "\n" @@ -2850,6 +2856,14 @@ define etp-run-queue-info-internal printf " Pointer: (ErtsRunQueue *) %p\n", $runq end +define etp-fds + if $_exitsignal == -1 + call erts_check_io_debug(0) + else + printf "Not yet implemented for core files" + end +end + define etp-disasm-1 set $code_ptr = ((BeamInstr*)$arg0) set $addr = *$code_ptr @@ -4314,6 +4328,10 @@ document etp-init %--------------------------------------------------------------------------- end +define hook-run + set $_exitsignal = -1 +end + etp-init help etp-init etp-show -- cgit v1.2.3