1 /* ---------------------------------------------------------------------------
3 * (c) The GHC Team, 1998-2006
5 * The scheduler and thread-related functionality
7 * --------------------------------------------------------------------------*/
9 #include "PosixSource.h"
10 #define KEEP_LOCKCLOSURE
15 #include "OSThreads.h"
20 #include "StgMiscClosures.h"
21 #include "Interpreter.h"
23 #include "RtsSignals.h"
29 #include "ThreadLabels.h"
30 #include "LdvProfile.h"
32 #include "Proftimer.h"
34 #if defined(GRAN) || defined(PARALLEL_HASKELL)
35 # include "GranSimRts.h"
37 # include "ParallelRts.h"
38 # include "Parallel.h"
39 # include "ParallelDebug.h"
44 #include "Capability.h"
46 #include "AwaitEvent.h"
47 #if defined(mingw32_HOST_OS)
48 #include "win32/IOManager.h"
51 #include "RaiseAsync.h"
53 #include "ThrIOManager.h"
55 #ifdef HAVE_SYS_TYPES_H
56 #include <sys/types.h>
70 // Turn off inlining when debugging - it obfuscates things
73 # define STATIC_INLINE static
76 /* -----------------------------------------------------------------------------
78 * -------------------------------------------------------------------------- */
82 StgTSO* ActiveTSO = NULL; /* for assigning system costs; GranSim-Light only */
83 /* rtsTime TimeOfNextEvent, EndOfTimeSlice; now in GranSim.c */
86 In GranSim we have a runnable and a blocked queue for each processor.
87 In order to minimise code changes new arrays run_queue_hds/tls
88 are created. run_queue_hd is then a short cut (macro) for
89 run_queue_hds[CurrentProc] (see GranSim.h).
92 StgTSO *run_queue_hds[MAX_PROC], *run_queue_tls[MAX_PROC];
93 StgTSO *blocked_queue_hds[MAX_PROC], *blocked_queue_tls[MAX_PROC];
94 StgTSO *ccalling_threadss[MAX_PROC];
95 /* We use the same global list of threads (all_threads) in GranSim as in
96 the std RTS (i.e. we are cheating). However, we don't use this list in
97 the GranSim specific code at the moment (so we are only potentially
102 #if !defined(THREADED_RTS)
103 // Blocked/sleeping thrads
104 StgTSO *blocked_queue_hd = NULL;
105 StgTSO *blocked_queue_tl = NULL;
106 StgTSO *sleeping_queue = NULL; // perhaps replace with a hash table?
109 /* Threads blocked on blackholes.
110 * LOCK: sched_mutex+capability, or all capabilities
112 StgTSO *blackhole_queue = NULL;
115 /* The blackhole_queue should be checked for threads to wake up. See
116 * Schedule.h for more thorough comment.
117 * LOCK: none (doesn't matter if we miss an update)
119 rtsBool blackholes_need_checking = rtsFalse;
121 /* flag set by signal handler to precipitate a context switch
122 * LOCK: none (just an advisory flag)
124 int context_switch = 0;
126 /* flag that tracks whether we have done any execution in this time slice.
127 * LOCK: currently none, perhaps we should lock (but needs to be
128 * updated in the fast path of the scheduler).
130 nat recent_activity = ACTIVITY_YES;
132 /* if this flag is set as well, give up execution
133 * LOCK: none (changes once, from false->true)
135 rtsBool sched_state = SCHED_RUNNING;
141 /* This is used in `TSO.h' and gcc 2.96 insists that this variable actually
142 * exists - earlier gccs apparently didn't.
148 * Set to TRUE when entering a shutdown state (via shutdownHaskellAndExit()) --
149 * in an MT setting, needed to signal that a worker thread shouldn't hang around
150 * in the scheduler when it is out of work.
152 rtsBool shutting_down_scheduler = rtsFalse;
155 * This mutex protects most of the global scheduler data in
156 * the THREADED_RTS runtime.
158 #if defined(THREADED_RTS)
162 #if defined(PARALLEL_HASKELL)
164 rtsTime TimeOfLastYield;
165 rtsBool emitSchedule = rtsTrue;
168 #if !defined(mingw32_HOST_OS)
169 #define FORKPROCESS_PRIMOP_SUPPORTED
172 /* -----------------------------------------------------------------------------
173 * static function prototypes
174 * -------------------------------------------------------------------------- */
176 static Capability *schedule (Capability *initialCapability, Task *task);
179 // These function all encapsulate parts of the scheduler loop, and are
180 // abstracted only to make the structure and control flow of the
181 // scheduler clearer.
183 static void schedulePreLoop (void);
184 #if defined(THREADED_RTS)
185 static void schedulePushWork(Capability *cap, Task *task);
187 static void scheduleStartSignalHandlers (Capability *cap);
188 static void scheduleCheckBlockedThreads (Capability *cap);
189 static void scheduleCheckWakeupThreads(Capability *cap USED_IF_NOT_THREADS);
190 static void scheduleCheckBlackHoles (Capability *cap);
191 static void scheduleDetectDeadlock (Capability *cap, Task *task);
193 static StgTSO *scheduleProcessEvent(rtsEvent *event);
195 #if defined(PARALLEL_HASKELL)
196 static StgTSO *scheduleSendPendingMessages(void);
197 static void scheduleActivateSpark(void);
198 static rtsBool scheduleGetRemoteWork(rtsBool *receivedFinish);
200 #if defined(PAR) || defined(GRAN)
201 static void scheduleGranParReport(void);
203 static void schedulePostRunThread(StgTSO *t);
204 static rtsBool scheduleHandleHeapOverflow( Capability *cap, StgTSO *t );
205 static void scheduleHandleStackOverflow( Capability *cap, Task *task,
207 static rtsBool scheduleHandleYield( Capability *cap, StgTSO *t,
208 nat prev_what_next );
209 static void scheduleHandleThreadBlocked( StgTSO *t );
210 static rtsBool scheduleHandleThreadFinished( Capability *cap, Task *task,
212 static rtsBool scheduleNeedHeapProfile(rtsBool ready_to_gc);
213 static Capability *scheduleDoGC(Capability *cap, Task *task,
214 rtsBool force_major);
216 static rtsBool checkBlackHoles(Capability *cap);
218 static StgTSO *threadStackOverflow(Capability *cap, StgTSO *tso);
219 static StgTSO *threadStackUnderflow(Task *task, StgTSO *tso);
221 static void deleteThread (Capability *cap, StgTSO *tso);
222 static void deleteAllThreads (Capability *cap);
224 #ifdef FORKPROCESS_PRIMOP_SUPPORTED
225 static void deleteThread_(Capability *cap, StgTSO *tso);
228 #if defined(PARALLEL_HASKELL)
229 StgTSO * createSparkThread(rtsSpark spark);
230 StgTSO * activateSpark (rtsSpark spark);
234 static char *whatNext_strs[] = {
244 /* -----------------------------------------------------------------------------
245 * Putting a thread on the run queue: different scheduling policies
246 * -------------------------------------------------------------------------- */
249 addToRunQueue( Capability *cap, StgTSO *t )
251 #if defined(PARALLEL_HASKELL)
252 if (RtsFlags.ParFlags.doFairScheduling) {
253 // this does round-robin scheduling; good for concurrency
254 appendToRunQueue(cap,t);
256 // this does unfair scheduling; good for parallelism
257 pushOnRunQueue(cap,t);
260 // this does round-robin scheduling; good for concurrency
261 appendToRunQueue(cap,t);
265 /* ---------------------------------------------------------------------------
266 Main scheduling loop.
268 We use round-robin scheduling, each thread returning to the
269 scheduler loop when one of these conditions is detected:
272 * timer expires (thread yields)
278 In a GranSim setup this loop iterates over the global event queue.
279 This revolves around the global event queue, which determines what
280 to do next. Therefore, it's more complicated than either the
281 concurrent or the parallel (GUM) setup.
284 GUM iterates over incoming messages.
285 It starts with nothing to do (thus CurrentTSO == END_TSO_QUEUE),
286 and sends out a fish whenever it has nothing to do; in-between
287 doing the actual reductions (shared code below) it processes the
288 incoming messages and deals with delayed operations
289 (see PendingFetches).
290 This is not the ugliest code you could imagine, but it's bloody close.
292 ------------------------------------------------------------------------ */
295 schedule (Capability *initialCapability, Task *task)
299 StgThreadReturnCode ret;
302 #elif defined(PARALLEL_HASKELL)
305 rtsBool receivedFinish = rtsFalse;
307 nat tp_size, sp_size; // stats only
312 #if defined(THREADED_RTS)
313 rtsBool first = rtsTrue;
316 cap = initialCapability;
318 // Pre-condition: this task owns initialCapability.
319 // The sched_mutex is *NOT* held
320 // NB. on return, we still hold a capability.
322 debugTrace (DEBUG_sched,
323 "### NEW SCHEDULER LOOP (task: %p, cap: %p)",
324 task, initialCapability);
328 // -----------------------------------------------------------
329 // Scheduler loop starts here:
331 #if defined(PARALLEL_HASKELL)
332 #define TERMINATION_CONDITION (!receivedFinish)
334 #define TERMINATION_CONDITION ((event = get_next_event()) != (rtsEvent*)NULL)
336 #define TERMINATION_CONDITION rtsTrue
339 while (TERMINATION_CONDITION) {
342 /* Choose the processor with the next event */
343 CurrentProc = event->proc;
344 CurrentTSO = event->tso;
347 #if defined(THREADED_RTS)
349 // don't yield the first time, we want a chance to run this
350 // thread for a bit, even if there are others banging at the
353 ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
355 // Yield the capability to higher-priority tasks if necessary.
356 yieldCapability(&cap, task);
360 #if defined(THREADED_RTS)
361 schedulePushWork(cap,task);
364 // Check whether we have re-entered the RTS from Haskell without
365 // going via suspendThread()/resumeThread (i.e. a 'safe' foreign
367 if (cap->in_haskell) {
368 errorBelch("schedule: re-entered unsafely.\n"
369 " Perhaps a 'foreign import unsafe' should be 'safe'?");
370 stg_exit(EXIT_FAILURE);
373 // The interruption / shutdown sequence.
375 // In order to cleanly shut down the runtime, we want to:
376 // * make sure that all main threads return to their callers
377 // with the state 'Interrupted'.
378 // * clean up all OS threads assocated with the runtime
379 // * free all memory etc.
381 // So the sequence for ^C goes like this:
383 // * ^C handler sets sched_state := SCHED_INTERRUPTING and
384 // arranges for some Capability to wake up
386 // * all threads in the system are halted, and the zombies are
387 // placed on the run queue for cleaning up. We acquire all
388 // the capabilities in order to delete the threads, this is
389 // done by scheduleDoGC() for convenience (because GC already
390 // needs to acquire all the capabilities). We can't kill
391 // threads involved in foreign calls.
393 // * somebody calls shutdownHaskell(), which calls exitScheduler()
395 // * sched_state := SCHED_SHUTTING_DOWN
397 // * all workers exit when the run queue on their capability
398 // drains. All main threads will also exit when their TSO
399 // reaches the head of the run queue and they can return.
401 // * eventually all Capabilities will shut down, and the RTS can
404 // * We might be left with threads blocked in foreign calls,
405 // we should really attempt to kill these somehow (TODO);
407 switch (sched_state) {
410 case SCHED_INTERRUPTING:
411 debugTrace(DEBUG_sched, "SCHED_INTERRUPTING");
412 #if defined(THREADED_RTS)
413 discardSparksCap(cap);
415 /* scheduleDoGC() deletes all the threads */
416 cap = scheduleDoGC(cap,task,rtsFalse);
418 case SCHED_SHUTTING_DOWN:
419 debugTrace(DEBUG_sched, "SCHED_SHUTTING_DOWN");
420 // If we are a worker, just exit. If we're a bound thread
421 // then we will exit below when we've removed our TSO from
423 if (task->tso == NULL && emptyRunQueue(cap)) {
428 barf("sched_state: %d", sched_state);
431 #if defined(THREADED_RTS)
432 // If the run queue is empty, take a spark and turn it into a thread.
434 if (emptyRunQueue(cap)) {
436 spark = findSpark(cap);
438 debugTrace(DEBUG_sched,
439 "turning spark of closure %p into a thread",
440 (StgClosure *)spark);
441 createSparkThread(cap,spark);
445 #endif // THREADED_RTS
447 scheduleStartSignalHandlers(cap);
449 // Only check the black holes here if we've nothing else to do.
450 // During normal execution, the black hole list only gets checked
451 // at GC time, to avoid repeatedly traversing this possibly long
452 // list each time around the scheduler.
453 if (emptyRunQueue(cap)) { scheduleCheckBlackHoles(cap); }
455 scheduleCheckWakeupThreads(cap);
457 scheduleCheckBlockedThreads(cap);
459 scheduleDetectDeadlock(cap,task);
460 #if defined(THREADED_RTS)
461 cap = task->cap; // reload cap, it might have changed
464 // Normally, the only way we can get here with no threads to
465 // run is if a keyboard interrupt received during
466 // scheduleCheckBlockedThreads() or scheduleDetectDeadlock().
467 // Additionally, it is not fatal for the
468 // threaded RTS to reach here with no threads to run.
470 // win32: might be here due to awaitEvent() being abandoned
471 // as a result of a console event having been delivered.
472 if ( emptyRunQueue(cap) ) {
473 #if !defined(THREADED_RTS) && !defined(mingw32_HOST_OS)
474 ASSERT(sched_state >= SCHED_INTERRUPTING);
476 continue; // nothing to do
479 #if defined(PARALLEL_HASKELL)
480 scheduleSendPendingMessages();
481 if (emptyRunQueue(cap) && scheduleActivateSpark())
485 ASSERT(next_fish_to_send_at==0); // i.e. no delayed fishes left!
488 /* If we still have no work we need to send a FISH to get a spark
490 if (emptyRunQueue(cap)) {
491 if (!scheduleGetRemoteWork(&receivedFinish)) continue;
492 ASSERT(rtsFalse); // should not happen at the moment
494 // from here: non-empty run queue.
495 // TODO: merge above case with this, only one call processMessages() !
496 if (PacketsWaiting()) { /* process incoming messages, if
497 any pending... only in else
498 because getRemoteWork waits for
500 receivedFinish = processMessages();
505 scheduleProcessEvent(event);
509 // Get a thread to run
511 t = popRunQueue(cap);
513 #if defined(GRAN) || defined(PAR)
514 scheduleGranParReport(); // some kind of debuging output
516 // Sanity check the thread we're about to run. This can be
517 // expensive if there is lots of thread switching going on...
518 IF_DEBUG(sanity,checkTSO(t));
521 #if defined(THREADED_RTS)
522 // Check whether we can run this thread in the current task.
523 // If not, we have to pass our capability to the right task.
525 Task *bound = t->bound;
529 debugTrace(DEBUG_sched,
530 "### Running thread %lu in bound thread", (unsigned long)t->id);
531 // yes, the Haskell thread is bound to the current native thread
533 debugTrace(DEBUG_sched,
534 "### thread %lu bound to another OS thread", (unsigned long)t->id);
535 // no, bound to a different Haskell thread: pass to that thread
536 pushOnRunQueue(cap,t);
540 // The thread we want to run is unbound.
542 debugTrace(DEBUG_sched,
543 "### this OS thread cannot run thread %lu", (unsigned long)t->id);
544 // no, the current native thread is bound to a different
545 // Haskell thread, so pass it to any worker thread
546 pushOnRunQueue(cap,t);
553 /* context switches are initiated by the timer signal, unless
554 * the user specified "context switch as often as possible", with
557 if (RtsFlags.ConcFlags.ctxtSwitchTicks == 0
558 && !emptyThreadQueues(cap)) {
564 // CurrentTSO is the thread to run. t might be different if we
565 // loop back to run_thread, so make sure to set CurrentTSO after
567 cap->r.rCurrentTSO = t;
569 debugTrace(DEBUG_sched, "-->> running thread %ld %s ...",
570 (long)t->id, whatNext_strs[t->what_next]);
572 startHeapProfTimer();
574 // Check for exceptions blocked on this thread
575 maybePerformBlockedException (cap, t);
577 // ----------------------------------------------------------------------
578 // Run the current thread
580 ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
581 ASSERT(t->cap == cap);
583 prev_what_next = t->what_next;
585 errno = t->saved_errno;
587 SetLastError(t->saved_winerror);
590 cap->in_haskell = rtsTrue;
594 #if defined(THREADED_RTS)
595 if (recent_activity == ACTIVITY_DONE_GC) {
596 // ACTIVITY_DONE_GC means we turned off the timer signal to
597 // conserve power (see #1623). Re-enable it here.
599 prev = xchg((P_)&recent_activity, ACTIVITY_YES);
600 if (prev == ACTIVITY_DONE_GC) {
604 recent_activity = ACTIVITY_YES;
608 switch (prev_what_next) {
612 /* Thread already finished, return to scheduler. */
613 ret = ThreadFinished;
619 r = StgRun((StgFunPtr) stg_returnToStackTop, &cap->r);
620 cap = regTableToCapability(r);
625 case ThreadInterpret:
626 cap = interpretBCO(cap);
631 barf("schedule: invalid what_next field");
634 cap->in_haskell = rtsFalse;
636 // The TSO might have moved, eg. if it re-entered the RTS and a GC
637 // happened. So find the new location:
638 t = cap->r.rCurrentTSO;
640 // We have run some Haskell code: there might be blackhole-blocked
641 // threads to wake up now.
642 // Lock-free test here should be ok, we're just setting a flag.
643 if ( blackhole_queue != END_TSO_QUEUE ) {
644 blackholes_need_checking = rtsTrue;
647 // And save the current errno in this thread.
648 // XXX: possibly bogus for SMP because this thread might already
649 // be running again, see code below.
650 t->saved_errno = errno;
652 // Similarly for Windows error code
653 t->saved_winerror = GetLastError();
656 #if defined(THREADED_RTS)
657 // If ret is ThreadBlocked, and this Task is bound to the TSO that
658 // blocked, we are in limbo - the TSO is now owned by whatever it
659 // is blocked on, and may in fact already have been woken up,
660 // perhaps even on a different Capability. It may be the case
661 // that task->cap != cap. We better yield this Capability
662 // immediately and return to normaility.
663 if (ret == ThreadBlocked) {
664 debugTrace(DEBUG_sched,
665 "--<< thread %lu (%s) stopped: blocked",
666 (unsigned long)t->id, whatNext_strs[t->what_next]);
671 ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
672 ASSERT(t->cap == cap);
674 // ----------------------------------------------------------------------
676 // Costs for the scheduler are assigned to CCS_SYSTEM
678 #if defined(PROFILING)
682 schedulePostRunThread(t);
684 t = threadStackUnderflow(task,t);
686 ready_to_gc = rtsFalse;
690 ready_to_gc = scheduleHandleHeapOverflow(cap,t);
694 scheduleHandleStackOverflow(cap,task,t);
698 if (scheduleHandleYield(cap, t, prev_what_next)) {
699 // shortcut for switching between compiler/interpreter:
705 scheduleHandleThreadBlocked(t);
709 if (scheduleHandleThreadFinished(cap, task, t)) return cap;
710 ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
714 barf("schedule: invalid thread return code %d", (int)ret);
717 if (ready_to_gc || scheduleNeedHeapProfile(ready_to_gc)) {
718 cap = scheduleDoGC(cap,task,rtsFalse);
720 } /* end of while() */
723 /* ----------------------------------------------------------------------------
724 * Setting up the scheduler loop
725 * ------------------------------------------------------------------------- */
728 schedulePreLoop(void)
731 /* set up first event to get things going */
732 /* ToDo: assign costs for system setup and init MainTSO ! */
733 new_event(CurrentProc, CurrentProc, CurrentTime[CurrentProc],
735 CurrentTSO, (StgClosure*)NULL, (rtsSpark*)NULL);
737 debugTrace (DEBUG_gran,
738 "GRAN: Init CurrentTSO (in schedule) = %p",
740 IF_DEBUG(gran, G_TSO(CurrentTSO, 5));
742 if (RtsFlags.GranFlags.Light) {
743 /* Save current time; GranSim Light only */
744 CurrentTSO->gran.clock = CurrentTime[CurrentProc];
749 /* -----------------------------------------------------------------------------
752 * Push work to other Capabilities if we have some.
753 * -------------------------------------------------------------------------- */
755 #if defined(THREADED_RTS)
757 schedulePushWork(Capability *cap USED_IF_THREADS,
758 Task *task USED_IF_THREADS)
760 Capability *free_caps[n_capabilities], *cap0;
763 // migration can be turned off with +RTS -qg
764 if (!RtsFlags.ParFlags.migrate) return;
766 // Check whether we have more threads on our run queue, or sparks
767 // in our pool, that we could hand to another Capability.
768 if ((emptyRunQueue(cap) || cap->run_queue_hd->_link == END_TSO_QUEUE)
769 && sparkPoolSizeCap(cap) < 2) {
773 // First grab as many free Capabilities as we can.
774 for (i=0, n_free_caps=0; i < n_capabilities; i++) {
775 cap0 = &capabilities[i];
776 if (cap != cap0 && tryGrabCapability(cap0,task)) {
777 if (!emptyRunQueue(cap0) || cap->returning_tasks_hd != NULL) {
778 // it already has some work, we just grabbed it at
779 // the wrong moment. Or maybe it's deadlocked!
780 releaseCapability(cap0);
782 free_caps[n_free_caps++] = cap0;
787 // we now have n_free_caps free capabilities stashed in
788 // free_caps[]. Share our run queue equally with them. This is
789 // probably the simplest thing we could do; improvements we might
790 // want to do include:
792 // - giving high priority to moving relatively new threads, on
793 // the gournds that they haven't had time to build up a
794 // working set in the cache on this CPU/Capability.
796 // - giving low priority to moving long-lived threads
798 if (n_free_caps > 0) {
799 StgTSO *prev, *t, *next;
800 rtsBool pushed_to_all;
802 debugTrace(DEBUG_sched, "excess threads on run queue and %d free capabilities, sharing...", n_free_caps);
805 pushed_to_all = rtsFalse;
807 if (cap->run_queue_hd != END_TSO_QUEUE) {
808 prev = cap->run_queue_hd;
810 prev->_link = END_TSO_QUEUE;
811 for (; t != END_TSO_QUEUE; t = next) {
813 t->_link = END_TSO_QUEUE;
814 if (t->what_next == ThreadRelocated
815 || t->bound == task // don't move my bound thread
816 || tsoLocked(t)) { // don't move a locked thread
817 setTSOLink(cap, prev, t);
819 } else if (i == n_free_caps) {
820 pushed_to_all = rtsTrue;
823 setTSOLink(cap, prev, t);
826 debugTrace(DEBUG_sched, "pushing thread %lu to capability %d", (unsigned long)t->id, free_caps[i]->no);
827 appendToRunQueue(free_caps[i],t);
828 if (t->bound) { t->bound->cap = free_caps[i]; }
829 t->cap = free_caps[i];
833 cap->run_queue_tl = prev;
836 // If there are some free capabilities that we didn't push any
837 // threads to, then try to push a spark to each one.
838 if (!pushed_to_all) {
840 // i is the next free capability to push to
841 for (; i < n_free_caps; i++) {
842 if (emptySparkPoolCap(free_caps[i])) {
843 spark = findSpark(cap);
845 debugTrace(DEBUG_sched, "pushing spark %p to capability %d", spark, free_caps[i]->no);
846 newSpark(&(free_caps[i]->r), spark);
852 // release the capabilities
853 for (i = 0; i < n_free_caps; i++) {
854 task->cap = free_caps[i];
855 releaseCapability(free_caps[i]);
858 task->cap = cap; // reset to point to our Capability.
862 /* ----------------------------------------------------------------------------
863 * Start any pending signal handlers
864 * ------------------------------------------------------------------------- */
866 #if defined(RTS_USER_SIGNALS) && !defined(THREADED_RTS)
868 scheduleStartSignalHandlers(Capability *cap)
870 if (RtsFlags.MiscFlags.install_signal_handlers && signals_pending()) {
871 // safe outside the lock
872 startSignalHandlers(cap);
877 scheduleStartSignalHandlers(Capability *cap STG_UNUSED)
882 /* ----------------------------------------------------------------------------
883 * Check for blocked threads that can be woken up.
884 * ------------------------------------------------------------------------- */
887 scheduleCheckBlockedThreads(Capability *cap USED_IF_NOT_THREADS)
889 #if !defined(THREADED_RTS)
891 // Check whether any waiting threads need to be woken up. If the
892 // run queue is empty, and there are no other tasks running, we
893 // can wait indefinitely for something to happen.
895 if ( !emptyQueue(blocked_queue_hd) || !emptyQueue(sleeping_queue) )
897 awaitEvent( emptyRunQueue(cap) && !blackholes_need_checking );
903 /* ----------------------------------------------------------------------------
904 * Check for threads woken up by other Capabilities
905 * ------------------------------------------------------------------------- */
908 scheduleCheckWakeupThreads(Capability *cap USED_IF_THREADS)
910 #if defined(THREADED_RTS)
911 // Any threads that were woken up by other Capabilities get
912 // appended to our run queue.
913 if (!emptyWakeupQueue(cap)) {
914 ACQUIRE_LOCK(&cap->lock);
915 if (emptyRunQueue(cap)) {
916 cap->run_queue_hd = cap->wakeup_queue_hd;
917 cap->run_queue_tl = cap->wakeup_queue_tl;
919 setTSOLink(cap, cap->run_queue_tl, cap->wakeup_queue_hd);
920 cap->run_queue_tl = cap->wakeup_queue_tl;
922 cap->wakeup_queue_hd = cap->wakeup_queue_tl = END_TSO_QUEUE;
923 RELEASE_LOCK(&cap->lock);
928 /* ----------------------------------------------------------------------------
929 * Check for threads blocked on BLACKHOLEs that can be woken up
930 * ------------------------------------------------------------------------- */
932 scheduleCheckBlackHoles (Capability *cap)
934 if ( blackholes_need_checking ) // check without the lock first
936 ACQUIRE_LOCK(&sched_mutex);
937 if ( blackholes_need_checking ) {
938 checkBlackHoles(cap);
939 blackholes_need_checking = rtsFalse;
941 RELEASE_LOCK(&sched_mutex);
945 /* ----------------------------------------------------------------------------
946 * Detect deadlock conditions and attempt to resolve them.
947 * ------------------------------------------------------------------------- */
950 scheduleDetectDeadlock (Capability *cap, Task *task)
953 #if defined(PARALLEL_HASKELL)
954 // ToDo: add deadlock detection in GUM (similar to THREADED_RTS) -- HWL
959 * Detect deadlock: when we have no threads to run, there are no
960 * threads blocked, waiting for I/O, or sleeping, and all the
961 * other tasks are waiting for work, we must have a deadlock of
964 if ( emptyThreadQueues(cap) )
966 #if defined(THREADED_RTS)
968 * In the threaded RTS, we only check for deadlock if there
969 * has been no activity in a complete timeslice. This means
970 * we won't eagerly start a full GC just because we don't have
971 * any threads to run currently.
973 if (recent_activity != ACTIVITY_INACTIVE) return;
976 debugTrace(DEBUG_sched, "deadlocked, forcing major GC...");
978 // Garbage collection can release some new threads due to
979 // either (a) finalizers or (b) threads resurrected because
980 // they are unreachable and will therefore be sent an
981 // exception. Any threads thus released will be immediately
983 cap = scheduleDoGC (cap, task, rtsTrue/*force major GC*/);
985 recent_activity = ACTIVITY_DONE_GC;
986 // disable timer signals (see #1623)
989 if ( !emptyRunQueue(cap) ) return;
991 #if defined(RTS_USER_SIGNALS) && !defined(THREADED_RTS)
992 /* If we have user-installed signal handlers, then wait
993 * for signals to arrive rather then bombing out with a
996 if ( RtsFlags.MiscFlags.install_signal_handlers && anyUserHandlers() ) {
997 debugTrace(DEBUG_sched,
998 "still deadlocked, waiting for signals...");
1002 if (signals_pending()) {
1003 startSignalHandlers(cap);
1006 // either we have threads to run, or we were interrupted:
1007 ASSERT(!emptyRunQueue(cap) || sched_state >= SCHED_INTERRUPTING);
1013 #if !defined(THREADED_RTS)
1014 /* Probably a real deadlock. Send the current main thread the
1015 * Deadlock exception.
1018 switch (task->tso->why_blocked) {
1020 case BlockedOnBlackHole:
1021 case BlockedOnException:
1023 throwToSingleThreaded(cap, task->tso,
1024 (StgClosure *)NonTermination_closure);
1027 barf("deadlock: main thread blocked in a strange way");
1035 /* ----------------------------------------------------------------------------
1036 * Process an event (GRAN only)
1037 * ------------------------------------------------------------------------- */
1041 scheduleProcessEvent(rtsEvent *event)
1045 if (RtsFlags.GranFlags.Light)
1046 GranSimLight_enter_system(event, &ActiveTSO); // adjust ActiveTSO etc
1048 /* adjust time based on time-stamp */
1049 if (event->time > CurrentTime[CurrentProc] &&
1050 event->evttype != ContinueThread)
1051 CurrentTime[CurrentProc] = event->time;
1053 /* Deal with the idle PEs (may issue FindWork or MoveSpark events) */
1054 if (!RtsFlags.GranFlags.Light)
1057 IF_DEBUG(gran, debugBelch("GRAN: switch by event-type\n"));
1059 /* main event dispatcher in GranSim */
1060 switch (event->evttype) {
1061 /* Should just be continuing execution */
1062 case ContinueThread:
1063 IF_DEBUG(gran, debugBelch("GRAN: doing ContinueThread\n"));
1064 /* ToDo: check assertion
1065 ASSERT(run_queue_hd != (StgTSO*)NULL &&
1066 run_queue_hd != END_TSO_QUEUE);
1068 /* Ignore ContinueThreads for fetching threads (if synchr comm) */
1069 if (!RtsFlags.GranFlags.DoAsyncFetch &&
1070 procStatus[CurrentProc]==Fetching) {
1071 debugBelch("ghuH: Spurious ContinueThread while Fetching ignored; TSO %d (%p) [PE %d]\n",
1072 CurrentTSO->id, CurrentTSO, CurrentProc);
1075 /* Ignore ContinueThreads for completed threads */
1076 if (CurrentTSO->what_next == ThreadComplete) {
1077 debugBelch("ghuH: found a ContinueThread event for completed thread %d (%p) [PE %d] (ignoring ContinueThread)\n",
1078 CurrentTSO->id, CurrentTSO, CurrentProc);
1081 /* Ignore ContinueThreads for threads that are being migrated */
1082 if (PROCS(CurrentTSO)==Nowhere) {
1083 debugBelch("ghuH: trying to run the migrating TSO %d (%p) [PE %d] (ignoring ContinueThread)\n",
1084 CurrentTSO->id, CurrentTSO, CurrentProc);
1087 /* The thread should be at the beginning of the run queue */
1088 if (CurrentTSO!=run_queue_hds[CurrentProc]) {
1089 debugBelch("ghuH: TSO %d (%p) [PE %d] is not at the start of the run_queue when doing a ContinueThread\n",
1090 CurrentTSO->id, CurrentTSO, CurrentProc);
1091 break; // run the thread anyway
1094 new_event(proc, proc, CurrentTime[proc],
1096 (StgTSO*)NULL, (StgClosure*)NULL, (rtsSpark*)NULL);
1098 */ /* Catches superfluous CONTINUEs -- should be unnecessary */
1099 break; // now actually run the thread; DaH Qu'vam yImuHbej
1102 do_the_fetchnode(event);
1103 goto next_thread; /* handle next event in event queue */
1106 do_the_globalblock(event);
1107 goto next_thread; /* handle next event in event queue */
1110 do_the_fetchreply(event);
1111 goto next_thread; /* handle next event in event queue */
1113 case UnblockThread: /* Move from the blocked queue to the tail of */
1114 do_the_unblock(event);
1115 goto next_thread; /* handle next event in event queue */
1117 case ResumeThread: /* Move from the blocked queue to the tail of */
1118 /* the runnable queue ( i.e. Qu' SImqa'lu') */
1119 event->tso->gran.blocktime +=
1120 CurrentTime[CurrentProc] - event->tso->gran.blockedat;
1121 do_the_startthread(event);
1122 goto next_thread; /* handle next event in event queue */
1125 do_the_startthread(event);
1126 goto next_thread; /* handle next event in event queue */
1129 do_the_movethread(event);
1130 goto next_thread; /* handle next event in event queue */
1133 do_the_movespark(event);
1134 goto next_thread; /* handle next event in event queue */
1137 do_the_findwork(event);
1138 goto next_thread; /* handle next event in event queue */
1141 barf("Illegal event type %u\n", event->evttype);
1144 /* This point was scheduler_loop in the old RTS */
1146 IF_DEBUG(gran, debugBelch("GRAN: after main switch\n"));
1148 TimeOfLastEvent = CurrentTime[CurrentProc];
1149 TimeOfNextEvent = get_time_of_next_event();
1150 IgnoreEvents=(TimeOfNextEvent==0); // HWL HACK
1151 // CurrentTSO = ThreadQueueHd;
1153 IF_DEBUG(gran, debugBelch("GRAN: time of next event is: %ld\n",
1156 if (RtsFlags.GranFlags.Light)
1157 GranSimLight_leave_system(event, &ActiveTSO);
1159 EndOfTimeSlice = CurrentTime[CurrentProc]+RtsFlags.GranFlags.time_slice;
1162 debugBelch("GRAN: end of time-slice is %#lx\n", EndOfTimeSlice));
1164 /* in a GranSim setup the TSO stays on the run queue */
1166 /* Take a thread from the run queue. */
1167 POP_RUN_QUEUE(t); // take_off_run_queue(t);
1170 debugBelch("GRAN: About to run current thread, which is\n");
1173 context_switch = 0; // turned on via GranYield, checking events and time slice
1176 DumpGranEvent(GR_SCHEDULE, t));
1178 procStatus[CurrentProc] = Busy;
1182 /* ----------------------------------------------------------------------------
1183 * Send pending messages (PARALLEL_HASKELL only)
1184 * ------------------------------------------------------------------------- */
1186 #if defined(PARALLEL_HASKELL)
1188 scheduleSendPendingMessages(void)
1194 # if defined(PAR) // global Mem.Mgmt., omit for now
1195 if (PendingFetches != END_BF_QUEUE) {
1200 if (RtsFlags.ParFlags.BufferTime) {
1201 // if we use message buffering, we must send away all message
1202 // packets which have become too old...
1208 /* ----------------------------------------------------------------------------
1209 * Activate spark threads (PARALLEL_HASKELL only)
1210 * ------------------------------------------------------------------------- */
1212 #if defined(PARALLEL_HASKELL)
1214 scheduleActivateSpark(void)
1217 ASSERT(emptyRunQueue());
1218 /* We get here if the run queue is empty and want some work.
1219 We try to turn a spark into a thread, and add it to the run queue,
1220 from where it will be picked up in the next iteration of the scheduler
1224 /* :-[ no local threads => look out for local sparks */
1225 /* the spark pool for the current PE */
1226 pool = &(cap.r.rSparks); // JB: cap = (old) MainCap
1227 if (advisory_thread_count < RtsFlags.ParFlags.maxThreads &&
1228 pool->hd < pool->tl) {
1230 * ToDo: add GC code check that we really have enough heap afterwards!!
1232 * If we're here (no runnable threads) and we have pending
1233 * sparks, we must have a space problem. Get enough space
1234 * to turn one of those pending sparks into a
1238 spark = findSpark(rtsFalse); /* get a spark */
1239 if (spark != (rtsSpark) NULL) {
1240 tso = createThreadFromSpark(spark); /* turn the spark into a thread */
1241 IF_PAR_DEBUG(fish, // schedule,
1242 debugBelch("==== schedule: Created TSO %d (%p); %d threads active\n",
1243 tso->id, tso, advisory_thread_count));
1245 if (tso==END_TSO_QUEUE) { /* failed to activate spark->back to loop */
1246 IF_PAR_DEBUG(fish, // schedule,
1247 debugBelch("==^^ failed to create thread from spark @ %lx\n",
1249 return rtsFalse; /* failed to generate a thread */
1250 } /* otherwise fall through & pick-up new tso */
1252 IF_PAR_DEBUG(fish, // schedule,
1253 debugBelch("==^^ no local sparks (spark pool contains only NFs: %d)\n",
1254 spark_queue_len(pool)));
1255 return rtsFalse; /* failed to generate a thread */
1257 return rtsTrue; /* success in generating a thread */
1258 } else { /* no more threads permitted or pool empty */
1259 return rtsFalse; /* failed to generateThread */
1262 tso = NULL; // avoid compiler warning only
1263 return rtsFalse; /* dummy in non-PAR setup */
1266 #endif // PARALLEL_HASKELL
1268 /* ----------------------------------------------------------------------------
1269 * Get work from a remote node (PARALLEL_HASKELL only)
1270 * ------------------------------------------------------------------------- */
1272 #if defined(PARALLEL_HASKELL)
1274 scheduleGetRemoteWork(rtsBool *receivedFinish)
1276 ASSERT(emptyRunQueue());
1278 if (RtsFlags.ParFlags.BufferTime) {
1279 IF_PAR_DEBUG(verbose,
1280 debugBelch("...send all pending data,"));
1283 for (i=1; i<=nPEs; i++)
1284 sendImmediately(i); // send all messages away immediately
1288 //++EDEN++ idle() , i.e. send all buffers, wait for work
1289 // suppress fishing in EDEN... just look for incoming messages
1290 // (blocking receive)
1291 IF_PAR_DEBUG(verbose,
1292 debugBelch("...wait for incoming messages...\n"));
1293 *receivedFinish = processMessages(); // blocking receive...
1295 // and reenter scheduling loop after having received something
1296 // (return rtsFalse below)
1298 # else /* activate SPARKS machinery */
1299 /* We get here, if we have no work, tried to activate a local spark, but still
1300 have no work. We try to get a remote spark, by sending a FISH message.
1301 Thread migration should be added here, and triggered when a sequence of
1302 fishes returns without work. */
1303 delay = (RtsFlags.ParFlags.fishDelay!=0ll ? RtsFlags.ParFlags.fishDelay : 0ll);
1305 /* =8-[ no local sparks => look for work on other PEs */
1307 * We really have absolutely no work. Send out a fish
1308 * (there may be some out there already), and wait for
1309 * something to arrive. We clearly can't run any threads
1310 * until a SCHEDULE or RESUME arrives, and so that's what
1311 * we're hoping to see. (Of course, we still have to
1312 * respond to other types of messages.)
1314 rtsTime now = msTime() /*CURRENT_TIME*/;
1315 IF_PAR_DEBUG(verbose,
1316 debugBelch("-- now=%ld\n", now));
1317 IF_PAR_DEBUG(fish, // verbose,
1318 if (outstandingFishes < RtsFlags.ParFlags.maxFishes &&
1319 (last_fish_arrived_at!=0 &&
1320 last_fish_arrived_at+delay > now)) {
1321 debugBelch("--$$ <%llu> delaying FISH until %llu (last fish %llu, delay %llu)\n",
1322 now, last_fish_arrived_at+delay,
1323 last_fish_arrived_at,
1327 if (outstandingFishes < RtsFlags.ParFlags.maxFishes &&
1328 advisory_thread_count < RtsFlags.ParFlags.maxThreads) { // send a FISH, but when?
1329 if (last_fish_arrived_at==0 ||
1330 (last_fish_arrived_at+delay <= now)) { // send FISH now!
1331 /* outstandingFishes is set in sendFish, processFish;
1332 avoid flooding system with fishes via delay */
1333 next_fish_to_send_at = 0;
1335 /* ToDo: this should be done in the main scheduling loop to avoid the
1336 busy wait here; not so bad if fish delay is very small */
1337 int iq = 0; // DEBUGGING -- HWL
1338 next_fish_to_send_at = last_fish_arrived_at+delay; // remember when to send
1339 /* send a fish when ready, but process messages that arrive in the meantime */
1341 if (PacketsWaiting()) {
1343 *receivedFinish = processMessages();
1346 } while (!*receivedFinish || now<next_fish_to_send_at);
1347 // JB: This means the fish could become obsolete, if we receive
1348 // work. Better check for work again?
1349 // last line: while (!receivedFinish || !haveWork || now<...)
1350 // next line: if (receivedFinish || haveWork )
1352 if (*receivedFinish) // no need to send a FISH if we are finishing anyway
1353 return rtsFalse; // NB: this will leave scheduler loop
1354 // immediately after return!
1356 IF_PAR_DEBUG(fish, // verbose,
1357 debugBelch("--$$ <%llu> sent delayed fish (%d processMessages); active/total threads=%d/%d\n",now,iq,run_queue_len(),advisory_thread_count));
1361 // JB: IMHO, this should all be hidden inside sendFish(...)
1363 sendFish(pe, thisPE, NEW_FISH_AGE, NEW_FISH_HISTORY,
1366 // Global statistics: count no. of fishes
1367 if (RtsFlags.ParFlags.ParStats.Global &&
1368 RtsFlags.GcFlags.giveStats > NO_GC_STATS) {
1369 globalParStats.tot_fish_mess++;
1373 /* delayed fishes must have been sent by now! */
1374 next_fish_to_send_at = 0;
1377 *receivedFinish = processMessages();
1378 # endif /* SPARKS */
1381 /* NB: this function always returns rtsFalse, meaning the scheduler
1382 loop continues with the next iteration;
1384 return code means success in finding work; we enter this function
1385 if there is no local work, thus have to send a fish which takes
1386 time until it arrives with work; in the meantime we should process
1387 messages in the main loop;
1390 #endif // PARALLEL_HASKELL
1392 /* ----------------------------------------------------------------------------
1393 * PAR/GRAN: Report stats & debugging info(?)
1394 * ------------------------------------------------------------------------- */
1396 #if defined(PAR) || defined(GRAN)
1398 scheduleGranParReport(void)
1400 ASSERT(run_queue_hd != END_TSO_QUEUE);
1402 /* Take a thread from the run queue, if we have work */
1403 POP_RUN_QUEUE(t); // take_off_run_queue(END_TSO_QUEUE);
1405 /* If this TSO has got its outport closed in the meantime,
1406 * it mustn't be run. Instead, we have to clean it up as if it was finished.
1407 * It has to be marked as TH_DEAD for this purpose.
1408 * If it is TH_TERM instead, it is supposed to have finished in the normal way.
1410 JB: TODO: investigate wether state change field could be nuked
1411 entirely and replaced by the normal tso state (whatnext
1412 field). All we want to do is to kill tsos from outside.
1415 /* ToDo: write something to the log-file
1416 if (RTSflags.ParFlags.granSimStats && !sameThread)
1417 DumpGranEvent(GR_SCHEDULE, RunnableThreadsHd);
1421 /* the spark pool for the current PE */
1422 pool = &(cap.r.rSparks); // cap = (old) MainCap
1425 debugBelch("--=^ %d threads, %d sparks on [%#x]\n",
1426 run_queue_len(), spark_queue_len(pool), CURRENT_PROC));
1429 debugBelch("--=^ %d threads, %d sparks on [%#x]\n",
1430 run_queue_len(), spark_queue_len(pool), CURRENT_PROC));
1432 if (RtsFlags.ParFlags.ParStats.Full &&
1433 (t->par.sparkname != (StgInt)0) && // only log spark generated threads
1434 (emitSchedule || // forced emit
1435 (t && LastTSO && t->id != LastTSO->id))) {
1437 we are running a different TSO, so write a schedule event to log file
1438 NB: If we use fair scheduling we also have to write a deschedule
1439 event for LastTSO; with unfair scheduling we know that the
1440 previous tso has blocked whenever we switch to another tso, so
1441 we don't need it in GUM for now
1443 IF_PAR_DEBUG(fish, // schedule,
1444 debugBelch("____ scheduling spark generated thread %d (%lx) (%lx) via a forced emit\n",t->id,t,t->par.sparkname));
1446 DumpRawGranEvent(CURRENT_PROC, CURRENT_PROC,
1447 GR_SCHEDULE, t, (StgClosure *)NULL, 0, 0);
1448 emitSchedule = rtsFalse;
1453 /* ----------------------------------------------------------------------------
1454 * After running a thread...
1455 * ------------------------------------------------------------------------- */
1458 schedulePostRunThread (StgTSO *t)
1460 // We have to be able to catch transactions that are in an
1461 // infinite loop as a result of seeing an inconsistent view of
1465 // [a,b] <- mapM readTVar [ta,tb]
1466 // when (a == b) loop
1468 // and a is never equal to b given a consistent view of memory.
1470 if (t -> trec != NO_TREC && t -> why_blocked == NotBlocked) {
1471 if (!stmValidateNestOfTransactions (t -> trec)) {
1472 debugTrace(DEBUG_sched | DEBUG_stm,
1473 "trec %p found wasting its time", t);
1475 // strip the stack back to the
1476 // ATOMICALLY_FRAME, aborting the (nested)
1477 // transaction, and saving the stack of any
1478 // partially-evaluated thunks on the heap.
1479 throwToSingleThreaded_(&capabilities[0], t,
1480 NULL, rtsTrue, NULL);
1483 ASSERT(get_itbl((StgClosure *)t->sp)->type == ATOMICALLY_FRAME);
1489 /* HACK 675: if the last thread didn't yield, make sure to print a
1490 SCHEDULE event to the log file when StgRunning the next thread, even
1491 if it is the same one as before */
1493 TimeOfLastYield = CURRENT_TIME;
1496 /* some statistics gathering in the parallel case */
1498 #if defined(GRAN) || defined(PAR) || defined(EDEN)
1502 IF_DEBUG(gran, DumpGranEvent(GR_DESCHEDULE, t));
1503 globalGranStats.tot_heapover++;
1505 globalParStats.tot_heapover++;
1512 DumpGranEvent(GR_DESCHEDULE, t));
1513 globalGranStats.tot_stackover++;
1516 // DumpGranEvent(GR_DESCHEDULE, t);
1517 globalParStats.tot_stackover++;
1521 case ThreadYielding:
1524 DumpGranEvent(GR_DESCHEDULE, t));
1525 globalGranStats.tot_yields++;
1528 // DumpGranEvent(GR_DESCHEDULE, t);
1529 globalParStats.tot_yields++;
1535 debugTrace(DEBUG_sched,
1536 "--<< thread %ld (%p; %s) stopped, blocking on node %p [PE %d] with BQ: ",
1537 t->id, t, whatNext_strs[t->what_next], t->block_info.closure,
1538 (t->block_info.closure==(StgClosure*)NULL ? 99 : where_is(t->block_info.closure)));
1539 if (t->block_info.closure!=(StgClosure*)NULL)
1540 print_bq(t->block_info.closure);
1543 // ??? needed; should emit block before
1545 DumpGranEvent(GR_DESCHEDULE, t));
1546 prune_eventq(t, (StgClosure *)NULL); // prune ContinueThreads for t
1549 ASSERT(procStatus[CurrentProc]==Busy ||
1550 ((procStatus[CurrentProc]==Fetching) &&
1551 (t->block_info.closure!=(StgClosure*)NULL)));
1552 if (run_queue_hds[CurrentProc] == END_TSO_QUEUE &&
1553 !(!RtsFlags.GranFlags.DoAsyncFetch &&
1554 procStatus[CurrentProc]==Fetching))
1555 procStatus[CurrentProc] = Idle;
1558 //++PAR++ blockThread() writes the event (change?)
1562 case ThreadFinished:
1566 barf("parGlobalStats: unknown return code");
1572 /* -----------------------------------------------------------------------------
1573 * Handle a thread that returned to the scheduler with ThreadHeepOverflow
1574 * -------------------------------------------------------------------------- */
1577 scheduleHandleHeapOverflow( Capability *cap, StgTSO *t )
1579 // did the task ask for a large block?
1580 if (cap->r.rHpAlloc > BLOCK_SIZE) {
1581 // if so, get one and push it on the front of the nursery.
1585 blocks = (lnat)BLOCK_ROUND_UP(cap->r.rHpAlloc) / BLOCK_SIZE;
1587 debugTrace(DEBUG_sched,
1588 "--<< thread %ld (%s) stopped: requesting a large block (size %ld)\n",
1589 (long)t->id, whatNext_strs[t->what_next], blocks);
1591 // don't do this if the nursery is (nearly) full, we'll GC first.
1592 if (cap->r.rCurrentNursery->link != NULL ||
1593 cap->r.rNursery->n_blocks == 1) { // paranoia to prevent infinite loop
1594 // if the nursery has only one block.
1597 bd = allocGroup( blocks );
1599 cap->r.rNursery->n_blocks += blocks;
1601 // link the new group into the list
1602 bd->link = cap->r.rCurrentNursery;
1603 bd->u.back = cap->r.rCurrentNursery->u.back;
1604 if (cap->r.rCurrentNursery->u.back != NULL) {
1605 cap->r.rCurrentNursery->u.back->link = bd;
1607 #if !defined(THREADED_RTS)
1608 ASSERT(g0s0->blocks == cap->r.rCurrentNursery &&
1609 g0s0 == cap->r.rNursery);
1611 cap->r.rNursery->blocks = bd;
1613 cap->r.rCurrentNursery->u.back = bd;
1615 // initialise it as a nursery block. We initialise the
1616 // step, gen_no, and flags field of *every* sub-block in
1617 // this large block, because this is easier than making
1618 // sure that we always find the block head of a large
1619 // block whenever we call Bdescr() (eg. evacuate() and
1620 // isAlive() in the GC would both have to do this, at
1624 for (x = bd; x < bd + blocks; x++) {
1625 x->step = cap->r.rNursery;
1631 // This assert can be a killer if the app is doing lots
1632 // of large block allocations.
1633 IF_DEBUG(sanity, checkNurserySanity(cap->r.rNursery));
1635 // now update the nursery to point to the new block
1636 cap->r.rCurrentNursery = bd;
1638 // we might be unlucky and have another thread get on the
1639 // run queue before us and steal the large block, but in that
1640 // case the thread will just end up requesting another large
1642 pushOnRunQueue(cap,t);
1643 return rtsFalse; /* not actually GC'ing */
1647 debugTrace(DEBUG_sched,
1648 "--<< thread %ld (%s) stopped: HeapOverflow",
1649 (long)t->id, whatNext_strs[t->what_next]);
1652 ASSERT(!is_on_queue(t,CurrentProc));
1653 #elif defined(PARALLEL_HASKELL)
1654 /* Currently we emit a DESCHEDULE event before GC in GUM.
1655 ToDo: either add separate event to distinguish SYSTEM time from rest
1656 or just nuke this DESCHEDULE (and the following SCHEDULE) */
1657 if (0 && RtsFlags.ParFlags.ParStats.Full) {
1658 DumpRawGranEvent(CURRENT_PROC, CURRENT_PROC,
1659 GR_DESCHEDULE, t, (StgClosure *)NULL, 0, 0);
1660 emitSchedule = rtsTrue;
1664 if (context_switch) {
1665 // Sometimes we miss a context switch, e.g. when calling
1666 // primitives in a tight loop, MAYBE_GC() doesn't check the
1667 // context switch flag, and we end up waiting for a GC.
1668 // See #1984, and concurrent/should_run/1984
1670 addToRunQueue(cap,t);
1672 pushOnRunQueue(cap,t);
1675 /* actual GC is done at the end of the while loop in schedule() */
1678 /* -----------------------------------------------------------------------------
1679 * Handle a thread that returned to the scheduler with ThreadStackOverflow
1680 * -------------------------------------------------------------------------- */
1683 scheduleHandleStackOverflow (Capability *cap, Task *task, StgTSO *t)
1685 debugTrace (DEBUG_sched,
1686 "--<< thread %ld (%s) stopped, StackOverflow",
1687 (long)t->id, whatNext_strs[t->what_next]);
1689 /* just adjust the stack for this thread, then pop it back
1693 /* enlarge the stack */
1694 StgTSO *new_t = threadStackOverflow(cap, t);
1696 /* The TSO attached to this Task may have moved, so update the
1699 if (task->tso == t) {
1702 pushOnRunQueue(cap,new_t);
1706 /* -----------------------------------------------------------------------------
1707 * Handle a thread that returned to the scheduler with ThreadYielding
1708 * -------------------------------------------------------------------------- */
1711 scheduleHandleYield( Capability *cap, StgTSO *t, nat prev_what_next )
1713 // Reset the context switch flag. We don't do this just before
1714 // running the thread, because that would mean we would lose ticks
1715 // during GC, which can lead to unfair scheduling (a thread hogs
1716 // the CPU because the tick always arrives during GC). This way
1717 // penalises threads that do a lot of allocation, but that seems
1718 // better than the alternative.
1721 /* put the thread back on the run queue. Then, if we're ready to
1722 * GC, check whether this is the last task to stop. If so, wake
1723 * up the GC thread. getThread will block during a GC until the
1727 if (t->what_next != prev_what_next) {
1728 debugTrace(DEBUG_sched,
1729 "--<< thread %ld (%s) stopped to switch evaluators",
1730 (long)t->id, whatNext_strs[t->what_next]);
1732 debugTrace(DEBUG_sched,
1733 "--<< thread %ld (%s) stopped, yielding",
1734 (long)t->id, whatNext_strs[t->what_next]);
1739 //debugBelch("&& Doing sanity check on yielding TSO %ld.", t->id);
1741 ASSERT(t->_link == END_TSO_QUEUE);
1743 // Shortcut if we're just switching evaluators: don't bother
1744 // doing stack squeezing (which can be expensive), just run the
1746 if (t->what_next != prev_what_next) {
1751 ASSERT(!is_on_queue(t,CurrentProc));
1754 //debugBelch("&& Doing sanity check on all ThreadQueues (and their TSOs).");
1755 checkThreadQsSanity(rtsTrue));
1759 addToRunQueue(cap,t);
1762 /* add a ContinueThread event to actually process the thread */
1763 new_event(CurrentProc, CurrentProc, CurrentTime[CurrentProc],
1765 t, (StgClosure*)NULL, (rtsSpark*)NULL);
1767 debugBelch("GRAN: eventq and runnableq after adding yielded thread to queue again:\n");
1774 /* -----------------------------------------------------------------------------
1775 * Handle a thread that returned to the scheduler with ThreadBlocked
1776 * -------------------------------------------------------------------------- */
1779 scheduleHandleThreadBlocked( StgTSO *t
1780 #if !defined(GRAN) && !defined(DEBUG)
1787 debugBelch("--<< thread %ld (%p; %s) stopped, blocking on node %p [PE %d] with BQ: \n",
1788 t->id, t, whatNext_strs[t->what_next], t->block_info.closure, (t->block_info.closure==(StgClosure*)NULL ? 99 : where_is(t->block_info.closure)));
1789 if (t->block_info.closure!=(StgClosure*)NULL) print_bq(t->block_info.closure));
1791 // ??? needed; should emit block before
1793 DumpGranEvent(GR_DESCHEDULE, t));
1794 prune_eventq(t, (StgClosure *)NULL); // prune ContinueThreads for t
1797 ASSERT(procStatus[CurrentProc]==Busy ||
1798 ((procStatus[CurrentProc]==Fetching) &&
1799 (t->block_info.closure!=(StgClosure*)NULL)));
1800 if (run_queue_hds[CurrentProc] == END_TSO_QUEUE &&
1801 !(!RtsFlags.GranFlags.DoAsyncFetch &&
1802 procStatus[CurrentProc]==Fetching))
1803 procStatus[CurrentProc] = Idle;
1807 debugBelch("--<< thread %ld (%p; %s) stopped, blocking on node %p with BQ: \n",
1808 t->id, t, whatNext_strs[t->what_next], t->block_info.closure));
1811 if (t->block_info.closure!=(StgClosure*)NULL)
1812 print_bq(t->block_info.closure));
1814 /* Send a fetch (if BlockedOnGA) and dump event to log file */
1817 /* whatever we schedule next, we must log that schedule */
1818 emitSchedule = rtsTrue;
1822 // We don't need to do anything. The thread is blocked, and it
1823 // has tidied up its stack and placed itself on whatever queue
1824 // it needs to be on.
1826 // ASSERT(t->why_blocked != NotBlocked);
1827 // Not true: for example,
1828 // - in THREADED_RTS, the thread may already have been woken
1829 // up by another Capability. This actually happens: try
1830 // conc023 +RTS -N2.
1831 // - the thread may have woken itself up already, because
1832 // threadPaused() might have raised a blocked throwTo
1833 // exception, see maybePerformBlockedException().
1836 if (traceClass(DEBUG_sched)) {
1837 debugTraceBegin("--<< thread %lu (%s) stopped: ",
1838 (unsigned long)t->id, whatNext_strs[t->what_next]);
1839 printThreadBlockage(t);
1844 /* Only for dumping event to log file
1845 ToDo: do I need this in GranSim, too?
1851 /* -----------------------------------------------------------------------------
1852 * Handle a thread that returned to the scheduler with ThreadFinished
1853 * -------------------------------------------------------------------------- */
1856 scheduleHandleThreadFinished (Capability *cap STG_UNUSED, Task *task, StgTSO *t)
1858 /* Need to check whether this was a main thread, and if so,
1859 * return with the return value.
1861 * We also end up here if the thread kills itself with an
1862 * uncaught exception, see Exception.cmm.
1864 debugTrace(DEBUG_sched, "--++ thread %lu (%s) finished",
1865 (unsigned long)t->id, whatNext_strs[t->what_next]);
1868 endThread(t, CurrentProc); // clean-up the thread
1869 #elif defined(PARALLEL_HASKELL)
1870 /* For now all are advisory -- HWL */
1871 //if(t->priority==AdvisoryPriority) ??
1872 advisory_thread_count--; // JB: Caution with this counter, buggy!
1875 if(t->dist.priority==RevalPriority)
1879 # if defined(EDENOLD)
1880 // the thread could still have an outport... (BUG)
1881 if (t->eden.outport != -1) {
1882 // delete the outport for the tso which has finished...
1883 IF_PAR_DEBUG(eden_ports,
1884 debugBelch("WARNING: Scheduler removes outport %d for TSO %d.\n",
1885 t->eden.outport, t->id));
1888 // thread still in the process (HEAVY BUG! since outport has just been closed...)
1889 if (t->eden.epid != -1) {
1890 IF_PAR_DEBUG(eden_ports,
1891 debugBelch("WARNING: Scheduler removes TSO %d from process %d .\n",
1892 t->id, t->eden.epid));
1893 removeTSOfromProcess(t);
1898 if (RtsFlags.ParFlags.ParStats.Full &&
1899 !RtsFlags.ParFlags.ParStats.Suppressed)
1900 DumpEndEvent(CURRENT_PROC, t, rtsFalse /* not mandatory */);
1902 // t->par only contains statistics: left out for now...
1904 debugBelch("**** end thread: ended sparked thread %d (%lx); sparkname: %lx\n",
1905 t->id,t,t->par.sparkname));
1907 #endif // PARALLEL_HASKELL
1910 // Check whether the thread that just completed was a bound
1911 // thread, and if so return with the result.
1913 // There is an assumption here that all thread completion goes
1914 // through this point; we need to make sure that if a thread
1915 // ends up in the ThreadKilled state, that it stays on the run
1916 // queue so it can be dealt with here.
1921 if (t->bound != task) {
1922 #if !defined(THREADED_RTS)
1923 // Must be a bound thread that is not the topmost one. Leave
1924 // it on the run queue until the stack has unwound to the
1925 // point where we can deal with this. Leaving it on the run
1926 // queue also ensures that the garbage collector knows about
1927 // this thread and its return value (it gets dropped from the
1928 // step->threads list so there's no other way to find it).
1929 appendToRunQueue(cap,t);
1932 // this cannot happen in the threaded RTS, because a
1933 // bound thread can only be run by the appropriate Task.
1934 barf("finished bound thread that isn't mine");
1938 ASSERT(task->tso == t);
1940 if (t->what_next == ThreadComplete) {
1942 // NOTE: return val is tso->sp[1] (see StgStartup.hc)
1943 *(task->ret) = (StgClosure *)task->tso->sp[1];
1945 task->stat = Success;
1948 *(task->ret) = NULL;
1950 if (sched_state >= SCHED_INTERRUPTING) {
1951 task->stat = Interrupted;
1953 task->stat = Killed;
1957 removeThreadLabel((StgWord)task->tso->id);
1959 return rtsTrue; // tells schedule() to return
1965 /* -----------------------------------------------------------------------------
1966 * Perform a heap census
1967 * -------------------------------------------------------------------------- */
1970 scheduleNeedHeapProfile( rtsBool ready_to_gc STG_UNUSED )
1972 // When we have +RTS -i0 and we're heap profiling, do a census at
1973 // every GC. This lets us get repeatable runs for debugging.
1974 if (performHeapProfile ||
1975 (RtsFlags.ProfFlags.profileInterval==0 &&
1976 RtsFlags.ProfFlags.doHeapProfile && ready_to_gc)) {
1983 /* -----------------------------------------------------------------------------
1984 * Perform a garbage collection if necessary
1985 * -------------------------------------------------------------------------- */
1988 scheduleDoGC (Capability *cap, Task *task USED_IF_THREADS, rtsBool force_major)
1991 rtsBool heap_census;
1993 static volatile StgWord waiting_for_gc;
1994 rtsBool was_waiting;
1999 // In order to GC, there must be no threads running Haskell code.
2000 // Therefore, the GC thread needs to hold *all* the capabilities,
2001 // and release them after the GC has completed.
2003 // This seems to be the simplest way: previous attempts involved
2004 // making all the threads with capabilities give up their
2005 // capabilities and sleep except for the *last* one, which
2006 // actually did the GC. But it's quite hard to arrange for all
2007 // the other tasks to sleep and stay asleep.
2010 was_waiting = cas(&waiting_for_gc, 0, 1);
2013 debugTrace(DEBUG_sched, "someone else is trying to GC...");
2014 if (cap) yieldCapability(&cap,task);
2015 } while (waiting_for_gc);
2016 return cap; // NOTE: task->cap might have changed here
2019 for (i=0; i < n_capabilities; i++) {
2020 debugTrace(DEBUG_sched, "ready_to_gc, grabbing all the capabilies (%d/%d)", i, n_capabilities);
2021 if (cap != &capabilities[i]) {
2022 Capability *pcap = &capabilities[i];
2023 // we better hope this task doesn't get migrated to
2024 // another Capability while we're waiting for this one.
2025 // It won't, because load balancing happens while we have
2026 // all the Capabilities, but even so it's a slightly
2027 // unsavoury invariant.
2030 waitForReturnCapability(&pcap, task);
2031 if (pcap != &capabilities[i]) {
2032 barf("scheduleDoGC: got the wrong capability");
2037 waiting_for_gc = rtsFalse;
2040 // so this happens periodically:
2041 if (cap) scheduleCheckBlackHoles(cap);
2043 IF_DEBUG(scheduler, printAllThreads());
2046 * We now have all the capabilities; if we're in an interrupting
2047 * state, then we should take the opportunity to delete all the
2048 * threads in the system.
2050 if (sched_state >= SCHED_INTERRUPTING) {
2051 deleteAllThreads(&capabilities[0]);
2052 sched_state = SCHED_SHUTTING_DOWN;
2055 heap_census = scheduleNeedHeapProfile(rtsTrue);
2057 /* everybody back, start the GC.
2058 * Could do it in this thread, or signal a condition var
2059 * to do it in another thread. Either way, we need to
2060 * broadcast on gc_pending_cond afterward.
2062 #if defined(THREADED_RTS)
2063 debugTrace(DEBUG_sched, "doing GC");
2065 GarbageCollect(force_major || heap_census);
2068 debugTrace(DEBUG_sched, "performing heap census");
2070 performHeapProfile = rtsFalse;
2073 #if defined(THREADED_RTS)
2074 // release our stash of capabilities.
2075 for (i = 0; i < n_capabilities; i++) {
2076 if (cap != &capabilities[i]) {
2077 task->cap = &capabilities[i];
2078 releaseCapability(&capabilities[i]);
2089 /* add a ContinueThread event to continue execution of current thread */
2090 new_event(CurrentProc, CurrentProc, CurrentTime[CurrentProc],
2092 t, (StgClosure*)NULL, (rtsSpark*)NULL);
2094 debugBelch("GRAN: eventq and runnableq after Garbage collection:\n\n");
2102 /* ---------------------------------------------------------------------------
2103 * Singleton fork(). Do not copy any running threads.
2104 * ------------------------------------------------------------------------- */
2107 forkProcess(HsStablePtr *entry
2108 #ifndef FORKPROCESS_PRIMOP_SUPPORTED
2113 #ifdef FORKPROCESS_PRIMOP_SUPPORTED
2120 #if defined(THREADED_RTS)
2121 if (RtsFlags.ParFlags.nNodes > 1) {
2122 errorBelch("forking not supported with +RTS -N<n> greater than 1");
2123 stg_exit(EXIT_FAILURE);
2127 debugTrace(DEBUG_sched, "forking!");
2129 // ToDo: for SMP, we should probably acquire *all* the capabilities
2132 // no funny business: hold locks while we fork, otherwise if some
2133 // other thread is holding a lock when the fork happens, the data
2134 // structure protected by the lock will forever be in an
2135 // inconsistent state in the child. See also #1391.
2136 ACQUIRE_LOCK(&sched_mutex);
2137 ACQUIRE_LOCK(&cap->lock);
2138 ACQUIRE_LOCK(&cap->running_task->lock);
2142 if (pid) { // parent
2144 RELEASE_LOCK(&sched_mutex);
2145 RELEASE_LOCK(&cap->lock);
2146 RELEASE_LOCK(&cap->running_task->lock);
2148 // just return the pid
2154 #if defined(THREADED_RTS)
2155 initMutex(&sched_mutex);
2156 initMutex(&cap->lock);
2157 initMutex(&cap->running_task->lock);
2160 // Now, all OS threads except the thread that forked are
2161 // stopped. We need to stop all Haskell threads, including
2162 // those involved in foreign calls. Also we need to delete
2163 // all Tasks, because they correspond to OS threads that are
2166 for (s = 0; s < total_steps; s++) {
2167 for (t = all_steps[s].threads; t != END_TSO_QUEUE; t = next) {
2168 if (t->what_next == ThreadRelocated) {
2171 next = t->global_link;
2172 // don't allow threads to catch the ThreadKilled
2173 // exception, but we do want to raiseAsync() because these
2174 // threads may be evaluating thunks that we need later.
2175 deleteThread_(cap,t);
2180 // Empty the run queue. It seems tempting to let all the
2181 // killed threads stay on the run queue as zombies to be
2182 // cleaned up later, but some of them correspond to bound
2183 // threads for which the corresponding Task does not exist.
2184 cap->run_queue_hd = END_TSO_QUEUE;
2185 cap->run_queue_tl = END_TSO_QUEUE;
2187 // Any suspended C-calling Tasks are no more, their OS threads
2189 cap->suspended_ccalling_tasks = NULL;
2191 // Empty the threads lists. Otherwise, the garbage
2192 // collector may attempt to resurrect some of these threads.
2193 for (s = 0; s < total_steps; s++) {
2194 all_steps[s].threads = END_TSO_QUEUE;
2197 // Wipe the task list, except the current Task.
2198 ACQUIRE_LOCK(&sched_mutex);
2199 for (task = all_tasks; task != NULL; task=task->all_link) {
2200 if (task != cap->running_task) {
2201 #if defined(THREADED_RTS)
2202 initMutex(&task->lock); // see #1391
2207 RELEASE_LOCK(&sched_mutex);
2209 #if defined(THREADED_RTS)
2210 // Wipe our spare workers list, they no longer exist. New
2211 // workers will be created if necessary.
2212 cap->spare_workers = NULL;
2213 cap->returning_tasks_hd = NULL;
2214 cap->returning_tasks_tl = NULL;
2217 // On Unix, all timers are reset in the child, so we need to start
2222 cap = rts_evalStableIO(cap, entry, NULL); // run the action
2223 rts_checkSchedStatus("forkProcess",cap);
2226 hs_exit(); // clean up and exit
2227 stg_exit(EXIT_SUCCESS);
2229 #else /* !FORKPROCESS_PRIMOP_SUPPORTED */
2230 barf("forkProcess#: primop not supported on this platform, sorry!\n");
2235 /* ---------------------------------------------------------------------------
2236 * Delete all the threads in the system
2237 * ------------------------------------------------------------------------- */
2240 deleteAllThreads ( Capability *cap )
2242 // NOTE: only safe to call if we own all capabilities.
2247 debugTrace(DEBUG_sched,"deleting all threads");
2248 for (s = 0; s < total_steps; s++) {
2249 for (t = all_steps[s].threads; t != END_TSO_QUEUE; t = next) {
2250 if (t->what_next == ThreadRelocated) {
2253 next = t->global_link;
2254 deleteThread(cap,t);
2259 // The run queue now contains a bunch of ThreadKilled threads. We
2260 // must not throw these away: the main thread(s) will be in there
2261 // somewhere, and the main scheduler loop has to deal with it.
2262 // Also, the run queue is the only thing keeping these threads from
2263 // being GC'd, and we don't want the "main thread has been GC'd" panic.
2265 #if !defined(THREADED_RTS)
2266 ASSERT(blocked_queue_hd == END_TSO_QUEUE);
2267 ASSERT(sleeping_queue == END_TSO_QUEUE);
2271 /* -----------------------------------------------------------------------------
2272 Managing the suspended_ccalling_tasks list.
2273 Locks required: sched_mutex
2274 -------------------------------------------------------------------------- */
2277 suspendTask (Capability *cap, Task *task)
2279 ASSERT(task->next == NULL && task->prev == NULL);
2280 task->next = cap->suspended_ccalling_tasks;
2282 if (cap->suspended_ccalling_tasks) {
2283 cap->suspended_ccalling_tasks->prev = task;
2285 cap->suspended_ccalling_tasks = task;
2289 recoverSuspendedTask (Capability *cap, Task *task)
2292 task->prev->next = task->next;
2294 ASSERT(cap->suspended_ccalling_tasks == task);
2295 cap->suspended_ccalling_tasks = task->next;
2298 task->next->prev = task->prev;
2300 task->next = task->prev = NULL;
2303 /* ---------------------------------------------------------------------------
2304 * Suspending & resuming Haskell threads.
2306 * When making a "safe" call to C (aka _ccall_GC), the task gives back
2307 * its capability before calling the C function. This allows another
2308 * task to pick up the capability and carry on running Haskell
2309 * threads. It also means that if the C call blocks, it won't lock
2312 * The Haskell thread making the C call is put to sleep for the
2313 * duration of the call, on the susepended_ccalling_threads queue. We
2314 * give out a token to the task, which it can use to resume the thread
2315 * on return from the C function.
2316 * ------------------------------------------------------------------------- */
2319 suspendThread (StgRegTable *reg)
2326 StgWord32 saved_winerror;
2329 saved_errno = errno;
2331 saved_winerror = GetLastError();
2334 /* assume that *reg is a pointer to the StgRegTable part of a Capability.
2336 cap = regTableToCapability(reg);
2338 task = cap->running_task;
2339 tso = cap->r.rCurrentTSO;
2341 debugTrace(DEBUG_sched,
2342 "thread %lu did a safe foreign call",
2343 (unsigned long)cap->r.rCurrentTSO->id);
2345 // XXX this might not be necessary --SDM
2346 tso->what_next = ThreadRunGHC;
2348 threadPaused(cap,tso);
2350 if ((tso->flags & TSO_BLOCKEX) == 0) {
2351 tso->why_blocked = BlockedOnCCall;
2352 tso->flags |= TSO_BLOCKEX;
2353 tso->flags &= ~TSO_INTERRUPTIBLE;
2355 tso->why_blocked = BlockedOnCCall_NoUnblockExc;
2358 // Hand back capability
2359 task->suspended_tso = tso;
2361 ACQUIRE_LOCK(&cap->lock);
2363 suspendTask(cap,task);
2364 cap->in_haskell = rtsFalse;
2365 releaseCapability_(cap);
2367 RELEASE_LOCK(&cap->lock);
2369 #if defined(THREADED_RTS)
2370 /* Preparing to leave the RTS, so ensure there's a native thread/task
2371 waiting to take over.
2373 debugTrace(DEBUG_sched, "thread %lu: leaving RTS", (unsigned long)tso->id);
2376 errno = saved_errno;
2378 SetLastError(saved_winerror);
2384 resumeThread (void *task_)
2391 StgWord32 saved_winerror;
2394 saved_errno = errno;
2396 saved_winerror = GetLastError();
2400 // Wait for permission to re-enter the RTS with the result.
2401 waitForReturnCapability(&cap,task);
2402 // we might be on a different capability now... but if so, our
2403 // entry on the suspended_ccalling_tasks list will also have been
2406 // Remove the thread from the suspended list
2407 recoverSuspendedTask(cap,task);
2409 tso = task->suspended_tso;
2410 task->suspended_tso = NULL;
2411 tso->_link = END_TSO_QUEUE; // no write barrier reqd
2412 debugTrace(DEBUG_sched, "thread %lu: re-entering RTS", (unsigned long)tso->id);
2414 if (tso->why_blocked == BlockedOnCCall) {
2415 awakenBlockedExceptionQueue(cap,tso);
2416 tso->flags &= ~(TSO_BLOCKEX | TSO_INTERRUPTIBLE);
2419 /* Reset blocking status */
2420 tso->why_blocked = NotBlocked;
2422 cap->r.rCurrentTSO = tso;
2423 cap->in_haskell = rtsTrue;
2424 errno = saved_errno;
2426 SetLastError(saved_winerror);
2429 /* We might have GC'd, mark the TSO dirty again */
2432 IF_DEBUG(sanity, checkTSO(tso));
2437 /* ---------------------------------------------------------------------------
2440 * scheduleThread puts a thread on the end of the runnable queue.
2441 * This will usually be done immediately after a thread is created.
2442 * The caller of scheduleThread must create the thread using e.g.
2443 * createThread and push an appropriate closure
2444 * on this thread's stack before the scheduler is invoked.
2445 * ------------------------------------------------------------------------ */
2448 scheduleThread(Capability *cap, StgTSO *tso)
2450 // The thread goes at the *end* of the run-queue, to avoid possible
2451 // starvation of any threads already on the queue.
2452 appendToRunQueue(cap,tso);
2456 scheduleThreadOn(Capability *cap, StgWord cpu USED_IF_THREADS, StgTSO *tso)
2458 #if defined(THREADED_RTS)
2459 tso->flags |= TSO_LOCKED; // we requested explicit affinity; don't
2460 // move this thread from now on.
2461 cpu %= RtsFlags.ParFlags.nNodes;
2462 if (cpu == cap->no) {
2463 appendToRunQueue(cap,tso);
2465 migrateThreadToCapability_lock(&capabilities[cpu],tso);
2468 appendToRunQueue(cap,tso);
2473 scheduleWaitThread (StgTSO* tso, /*[out]*/HaskellObj* ret, Capability *cap)
2477 // We already created/initialised the Task
2478 task = cap->running_task;
2480 // This TSO is now a bound thread; make the Task and TSO
2481 // point to each other.
2487 task->stat = NoStatus;
2489 appendToRunQueue(cap,tso);
2491 debugTrace(DEBUG_sched, "new bound thread (%lu)", (unsigned long)tso->id);
2494 /* GranSim specific init */
2495 CurrentTSO = m->tso; // the TSO to run
2496 procStatus[MainProc] = Busy; // status of main PE
2497 CurrentProc = MainProc; // PE to run it on
2500 cap = schedule(cap,task);
2502 ASSERT(task->stat != NoStatus);
2503 ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
2505 debugTrace(DEBUG_sched, "bound thread (%lu) finished", (unsigned long)task->tso->id);
2509 /* ----------------------------------------------------------------------------
2511 * ------------------------------------------------------------------------- */
2513 #if defined(THREADED_RTS)
2515 workerStart(Task *task)
2519 // See startWorkerTask().
2520 ACQUIRE_LOCK(&task->lock);
2522 RELEASE_LOCK(&task->lock);
2524 // set the thread-local pointer to the Task:
2527 // schedule() runs without a lock.
2528 cap = schedule(cap,task);
2530 // On exit from schedule(), we have a Capability.
2531 releaseCapability(cap);
2532 workerTaskStop(task);
2536 /* ---------------------------------------------------------------------------
2539 * Initialise the scheduler. This resets all the queues - if the
2540 * queues contained any threads, they'll be garbage collected at the
2543 * ------------------------------------------------------------------------ */
2550 for (i=0; i<=MAX_PROC; i++) {
2551 run_queue_hds[i] = END_TSO_QUEUE;
2552 run_queue_tls[i] = END_TSO_QUEUE;
2553 blocked_queue_hds[i] = END_TSO_QUEUE;
2554 blocked_queue_tls[i] = END_TSO_QUEUE;
2555 ccalling_threadss[i] = END_TSO_QUEUE;
2556 blackhole_queue[i] = END_TSO_QUEUE;
2557 sleeping_queue = END_TSO_QUEUE;
2559 #elif !defined(THREADED_RTS)
2560 blocked_queue_hd = END_TSO_QUEUE;
2561 blocked_queue_tl = END_TSO_QUEUE;
2562 sleeping_queue = END_TSO_QUEUE;
2565 blackhole_queue = END_TSO_QUEUE;
2568 sched_state = SCHED_RUNNING;
2569 recent_activity = ACTIVITY_YES;
2571 #if defined(THREADED_RTS)
2572 /* Initialise the mutex and condition variables used by
2574 initMutex(&sched_mutex);
2577 ACQUIRE_LOCK(&sched_mutex);
2579 /* A capability holds the state a native thread needs in
2580 * order to execute STG code. At least one capability is
2581 * floating around (only THREADED_RTS builds have more than one).
2587 #if defined(THREADED_RTS) || defined(PARALLEL_HASKELL)
2591 #if defined(THREADED_RTS)
2593 * Eagerly start one worker to run each Capability, except for
2594 * Capability 0. The idea is that we're probably going to start a
2595 * bound thread on Capability 0 pretty soon, so we don't want a
2596 * worker task hogging it.
2601 for (i = 1; i < n_capabilities; i++) {
2602 cap = &capabilities[i];
2603 ACQUIRE_LOCK(&cap->lock);
2604 startWorkerTask(cap, workerStart);
2605 RELEASE_LOCK(&cap->lock);
2610 trace(TRACE_sched, "start: %d capabilities", n_capabilities);
2612 RELEASE_LOCK(&sched_mutex);
2617 rtsBool wait_foreign
2618 #if !defined(THREADED_RTS)
2619 __attribute__((unused))
2622 /* see Capability.c, shutdownCapability() */
2626 #if defined(THREADED_RTS)
2627 ACQUIRE_LOCK(&sched_mutex);
2628 task = newBoundTask();
2629 RELEASE_LOCK(&sched_mutex);
2632 // If we haven't killed all the threads yet, do it now.
2633 if (sched_state < SCHED_SHUTTING_DOWN) {
2634 sched_state = SCHED_INTERRUPTING;
2635 scheduleDoGC(NULL,task,rtsFalse);
2637 sched_state = SCHED_SHUTTING_DOWN;
2639 #if defined(THREADED_RTS)
2643 for (i = 0; i < n_capabilities; i++) {
2644 shutdownCapability(&capabilities[i], task, wait_foreign);
2646 boundTaskExiting(task);
2650 freeCapability(&MainCapability);
2655 freeScheduler( void )
2658 if (n_capabilities != 1) {
2659 stgFree(capabilities);
2661 #if defined(THREADED_RTS)
2662 closeMutex(&sched_mutex);
2666 /* -----------------------------------------------------------------------------
2669 This is the interface to the garbage collector from Haskell land.
2670 We provide this so that external C code can allocate and garbage
2671 collect when called from Haskell via _ccall_GC.
2672 -------------------------------------------------------------------------- */
2675 performGC_(rtsBool force_major)
2678 // We must grab a new Task here, because the existing Task may be
2679 // associated with a particular Capability, and chained onto the
2680 // suspended_ccalling_tasks queue.
2681 ACQUIRE_LOCK(&sched_mutex);
2682 task = newBoundTask();
2683 RELEASE_LOCK(&sched_mutex);
2684 scheduleDoGC(NULL,task,force_major);
2685 boundTaskExiting(task);
2691 performGC_(rtsFalse);
2695 performMajorGC(void)
2697 performGC_(rtsTrue);
2700 /* -----------------------------------------------------------------------------
2703 If the thread has reached its maximum stack size, then raise the
2704 StackOverflow exception in the offending thread. Otherwise
2705 relocate the TSO into a larger chunk of memory and adjust its stack
2707 -------------------------------------------------------------------------- */
2710 threadStackOverflow(Capability *cap, StgTSO *tso)
2712 nat new_stack_size, stack_words;
2717 IF_DEBUG(sanity,checkTSO(tso));
2719 // don't allow throwTo() to modify the blocked_exceptions queue
2720 // while we are moving the TSO:
2721 lockClosure((StgClosure *)tso);
2723 if (tso->stack_size >= tso->max_stack_size && !(tso->flags & TSO_BLOCKEX)) {
2724 // NB. never raise a StackOverflow exception if the thread is
2725 // inside Control.Exceptino.block. It is impractical to protect
2726 // against stack overflow exceptions, since virtually anything
2727 // can raise one (even 'catch'), so this is the only sensible
2728 // thing to do here. See bug #767.
2730 debugTrace(DEBUG_gc,
2731 "threadStackOverflow of TSO %ld (%p): stack too large (now %ld; max is %ld)",
2732 (long)tso->id, tso, (long)tso->stack_size, (long)tso->max_stack_size);
2734 /* If we're debugging, just print out the top of the stack */
2735 printStackChunk(tso->sp, stg_min(tso->stack+tso->stack_size,
2738 // Send this thread the StackOverflow exception
2740 throwToSingleThreaded(cap, tso, (StgClosure *)stackOverflow_closure);
2744 /* Try to double the current stack size. If that takes us over the
2745 * maximum stack size for this thread, then use the maximum instead.
2746 * Finally round up so the TSO ends up as a whole number of blocks.
2748 new_stack_size = stg_min(tso->stack_size * 2, tso->max_stack_size);
2749 new_tso_size = (lnat)BLOCK_ROUND_UP(new_stack_size * sizeof(W_) +
2750 TSO_STRUCT_SIZE)/sizeof(W_);
2751 new_tso_size = round_to_mblocks(new_tso_size); /* Be MBLOCK-friendly */
2752 new_stack_size = new_tso_size - TSO_STRUCT_SIZEW;
2754 debugTrace(DEBUG_sched,
2755 "increasing stack size from %ld words to %d.",
2756 (long)tso->stack_size, new_stack_size);
2758 dest = (StgTSO *)allocateLocal(cap,new_tso_size);
2759 TICK_ALLOC_TSO(new_stack_size,0);
2761 /* copy the TSO block and the old stack into the new area */
2762 memcpy(dest,tso,TSO_STRUCT_SIZE);
2763 stack_words = tso->stack + tso->stack_size - tso->sp;
2764 new_sp = (P_)dest + new_tso_size - stack_words;
2765 memcpy(new_sp, tso->sp, stack_words * sizeof(W_));
2767 /* relocate the stack pointers... */
2769 dest->stack_size = new_stack_size;
2771 /* Mark the old TSO as relocated. We have to check for relocated
2772 * TSOs in the garbage collector and any primops that deal with TSOs.
2774 * It's important to set the sp value to just beyond the end
2775 * of the stack, so we don't attempt to scavenge any part of the
2778 tso->what_next = ThreadRelocated;
2779 setTSOLink(cap,tso,dest);
2780 tso->sp = (P_)&(tso->stack[tso->stack_size]);
2781 tso->why_blocked = NotBlocked;
2783 IF_PAR_DEBUG(verbose,
2784 debugBelch("@@ threadStackOverflow of TSO %d (now at %p): stack size increased to %ld\n",
2785 tso->id, tso, tso->stack_size);
2786 /* If we're debugging, just print out the top of the stack */
2787 printStackChunk(tso->sp, stg_min(tso->stack+tso->stack_size,
2793 IF_DEBUG(sanity,checkTSO(dest));
2795 IF_DEBUG(scheduler,printTSO(dest));
2802 threadStackUnderflow (Task *task STG_UNUSED, StgTSO *tso)
2804 bdescr *bd, *new_bd;
2805 lnat new_tso_size_w, tso_size_w;
2808 tso_size_w = tso_sizeW(tso);
2810 if (tso_size_w < MBLOCK_SIZE_W ||
2811 (nat)(tso->stack + tso->stack_size - tso->sp) > tso->stack_size / 4)
2816 // don't allow throwTo() to modify the blocked_exceptions queue
2817 // while we are moving the TSO:
2818 lockClosure((StgClosure *)tso);
2820 new_tso_size_w = round_to_mblocks(tso_size_w/2);
2822 debugTrace(DEBUG_sched, "thread %ld: reducing TSO size from %lu words to %lu",
2823 tso->id, tso_size_w, new_tso_size_w);
2825 bd = Bdescr((StgPtr)tso);
2826 new_bd = splitLargeBlock(bd, new_tso_size_w / BLOCK_SIZE_W);
2827 new_bd->free = bd->free;
2828 bd->free = bd->start + TSO_STRUCT_SIZEW;
2830 new_tso = (StgTSO *)new_bd->start;
2831 memcpy(new_tso,tso,TSO_STRUCT_SIZE);
2832 new_tso->stack_size = new_tso_size_w - TSO_STRUCT_SIZEW;
2834 tso->what_next = ThreadRelocated;
2835 tso->_link = new_tso; // no write barrier reqd: same generation
2837 // The TSO attached to this Task may have moved, so update the
2839 if (task->tso == tso) {
2840 task->tso = new_tso;
2846 IF_DEBUG(sanity,checkTSO(new_tso));
2851 /* ---------------------------------------------------------------------------
2853 - usually called inside a signal handler so it mustn't do anything fancy.
2854 ------------------------------------------------------------------------ */
2857 interruptStgRts(void)
2859 sched_state = SCHED_INTERRUPTING;
2864 /* -----------------------------------------------------------------------------
2867 This function causes at least one OS thread to wake up and run the
2868 scheduler loop. It is invoked when the RTS might be deadlocked, or
2869 an external event has arrived that may need servicing (eg. a
2870 keyboard interrupt).
2872 In the single-threaded RTS we don't do anything here; we only have
2873 one thread anyway, and the event that caused us to want to wake up
2874 will have interrupted any blocking system call in progress anyway.
2875 -------------------------------------------------------------------------- */
2880 #if defined(THREADED_RTS)
2881 // This forces the IO Manager thread to wakeup, which will
2882 // in turn ensure that some OS thread wakes up and runs the
2883 // scheduler loop, which will cause a GC and deadlock check.
2888 /* -----------------------------------------------------------------------------
2891 * Check the blackhole_queue for threads that can be woken up. We do
2892 * this periodically: before every GC, and whenever the run queue is
2895 * An elegant solution might be to just wake up all the blocked
2896 * threads with awakenBlockedQueue occasionally: they'll go back to
2897 * sleep again if the object is still a BLACKHOLE. Unfortunately this
2898 * doesn't give us a way to tell whether we've actually managed to
2899 * wake up any threads, so we would be busy-waiting.
2901 * -------------------------------------------------------------------------- */
2904 checkBlackHoles (Capability *cap)
2907 rtsBool any_woke_up = rtsFalse;
2910 // blackhole_queue is global:
2911 ASSERT_LOCK_HELD(&sched_mutex);
2913 debugTrace(DEBUG_sched, "checking threads blocked on black holes");
2915 // ASSUMES: sched_mutex
2916 prev = &blackhole_queue;
2917 t = blackhole_queue;
2918 while (t != END_TSO_QUEUE) {
2919 ASSERT(t->why_blocked == BlockedOnBlackHole);
2920 type = get_itbl(UNTAG_CLOSURE(t->block_info.closure))->type;
2921 if (type != BLACKHOLE && type != CAF_BLACKHOLE) {
2922 IF_DEBUG(sanity,checkTSO(t));
2923 t = unblockOne(cap, t);
2924 // urk, the threads migrate to the current capability
2925 // here, but we'd like to keep them on the original one.
2927 any_woke_up = rtsTrue;
2937 /* -----------------------------------------------------------------------------
2940 This is used for interruption (^C) and forking, and corresponds to
2941 raising an exception but without letting the thread catch the
2943 -------------------------------------------------------------------------- */
2946 deleteThread (Capability *cap, StgTSO *tso)
2948 // NOTE: must only be called on a TSO that we have exclusive
2949 // access to, because we will call throwToSingleThreaded() below.
2950 // The TSO must be on the run queue of the Capability we own, or
2951 // we must own all Capabilities.
2953 if (tso->why_blocked != BlockedOnCCall &&
2954 tso->why_blocked != BlockedOnCCall_NoUnblockExc) {
2955 throwToSingleThreaded(cap,tso,NULL);
2959 #ifdef FORKPROCESS_PRIMOP_SUPPORTED
2961 deleteThread_(Capability *cap, StgTSO *tso)
2962 { // for forkProcess only:
2963 // like deleteThread(), but we delete threads in foreign calls, too.
2965 if (tso->why_blocked == BlockedOnCCall ||
2966 tso->why_blocked == BlockedOnCCall_NoUnblockExc) {
2967 unblockOne(cap,tso);
2968 tso->what_next = ThreadKilled;
2970 deleteThread(cap,tso);
2975 /* -----------------------------------------------------------------------------
2976 raiseExceptionHelper
2978 This function is called by the raise# primitve, just so that we can
2979 move some of the tricky bits of raising an exception from C-- into
2980 C. Who knows, it might be a useful re-useable thing here too.
2981 -------------------------------------------------------------------------- */
2984 raiseExceptionHelper (StgRegTable *reg, StgTSO *tso, StgClosure *exception)
2986 Capability *cap = regTableToCapability(reg);
2987 StgThunk *raise_closure = NULL;
2989 StgRetInfoTable *info;
2991 // This closure represents the expression 'raise# E' where E
2992 // is the exception raise. It is used to overwrite all the
2993 // thunks which are currently under evaluataion.
2996 // OLD COMMENT (we don't have MIN_UPD_SIZE now):
2997 // LDV profiling: stg_raise_info has THUNK as its closure
2998 // type. Since a THUNK takes at least MIN_UPD_SIZE words in its
2999 // payload, MIN_UPD_SIZE is more approprate than 1. It seems that
3000 // 1 does not cause any problem unless profiling is performed.
3001 // However, when LDV profiling goes on, we need to linearly scan
3002 // small object pool, where raise_closure is stored, so we should
3003 // use MIN_UPD_SIZE.
3005 // raise_closure = (StgClosure *)RET_STGCALL1(P_,allocate,
3006 // sizeofW(StgClosure)+1);
3010 // Walk up the stack, looking for the catch frame. On the way,
3011 // we update any closures pointed to from update frames with the
3012 // raise closure that we just built.
3016 info = get_ret_itbl((StgClosure *)p);
3017 next = p + stack_frame_sizeW((StgClosure *)p);
3018 switch (info->i.type) {
3021 // Only create raise_closure if we need to.
3022 if (raise_closure == NULL) {
3024 (StgThunk *)allocateLocal(cap,sizeofW(StgThunk)+1);
3025 SET_HDR(raise_closure, &stg_raise_info, CCCS);
3026 raise_closure->payload[0] = exception;
3028 UPD_IND(((StgUpdateFrame *)p)->updatee,(StgClosure *)raise_closure);
3032 case ATOMICALLY_FRAME:
3033 debugTrace(DEBUG_stm, "found ATOMICALLY_FRAME at %p", p);
3035 return ATOMICALLY_FRAME;
3041 case CATCH_STM_FRAME:
3042 debugTrace(DEBUG_stm, "found CATCH_STM_FRAME at %p", p);
3044 return CATCH_STM_FRAME;
3050 case CATCH_RETRY_FRAME:
3059 /* -----------------------------------------------------------------------------
3060 findRetryFrameHelper
3062 This function is called by the retry# primitive. It traverses the stack
3063 leaving tso->sp referring to the frame which should handle the retry.
3065 This should either be a CATCH_RETRY_FRAME (if the retry# is within an orElse#)
3066 or should be a ATOMICALLY_FRAME (if the retry# reaches the top level).
3068 We skip CATCH_STM_FRAMEs (aborting and rolling back the nested tx that they
3069 create) because retries are not considered to be exceptions, despite the
3070 similar implementation.
3072 We should not expect to see CATCH_FRAME or STOP_FRAME because those should
3073 not be created within memory transactions.
3074 -------------------------------------------------------------------------- */
3077 findRetryFrameHelper (StgTSO *tso)
3080 StgRetInfoTable *info;
3084 info = get_ret_itbl((StgClosure *)p);
3085 next = p + stack_frame_sizeW((StgClosure *)p);
3086 switch (info->i.type) {
3088 case ATOMICALLY_FRAME:
3089 debugTrace(DEBUG_stm,
3090 "found ATOMICALLY_FRAME at %p during retry", p);
3092 return ATOMICALLY_FRAME;
3094 case CATCH_RETRY_FRAME:
3095 debugTrace(DEBUG_stm,
3096 "found CATCH_RETRY_FRAME at %p during retrry", p);
3098 return CATCH_RETRY_FRAME;
3100 case CATCH_STM_FRAME: {
3101 StgTRecHeader *trec = tso -> trec;
3102 StgTRecHeader *outer = stmGetEnclosingTRec(trec);
3103 debugTrace(DEBUG_stm,
3104 "found CATCH_STM_FRAME at %p during retry", p);
3105 debugTrace(DEBUG_stm, "trec=%p outer=%p", trec, outer);
3106 stmAbortTransaction(tso -> cap, trec);
3107 stmFreeAbortedTRec(tso -> cap, trec);
3108 tso -> trec = outer;
3115 ASSERT(info->i.type != CATCH_FRAME);
3116 ASSERT(info->i.type != STOP_FRAME);
3123 /* -----------------------------------------------------------------------------
3124 resurrectThreads is called after garbage collection on the list of
3125 threads found to be garbage. Each of these threads will be woken
3126 up and sent a signal: BlockedOnDeadMVar if the thread was blocked
3127 on an MVar, or NonTermination if the thread was blocked on a Black
3130 Locks: assumes we hold *all* the capabilities.
3131 -------------------------------------------------------------------------- */
3134 resurrectThreads (StgTSO *threads)
3140 for (tso = threads; tso != END_TSO_QUEUE; tso = next) {
3141 next = tso->global_link;
3143 step = Bdescr((P_)tso)->step;
3144 tso->global_link = step->threads;
3145 step->threads = tso;
3147 debugTrace(DEBUG_sched, "resurrecting thread %lu", (unsigned long)tso->id);
3149 // Wake up the thread on the Capability it was last on
3152 switch (tso->why_blocked) {
3154 case BlockedOnException:
3155 /* Called by GC - sched_mutex lock is currently held. */
3156 throwToSingleThreaded(cap, tso,
3157 (StgClosure *)BlockedOnDeadMVar_closure);
3159 case BlockedOnBlackHole:
3160 throwToSingleThreaded(cap, tso,
3161 (StgClosure *)NonTermination_closure);
3164 throwToSingleThreaded(cap, tso,
3165 (StgClosure *)BlockedIndefinitely_closure);
3168 /* This might happen if the thread was blocked on a black hole
3169 * belonging to a thread that we've just woken up (raiseAsync
3170 * can wake up threads, remember...).
3174 barf("resurrectThreads: thread blocked in a strange way");
3179 /* -----------------------------------------------------------------------------
3180 performPendingThrowTos is called after garbage collection, and
3181 passed a list of threads that were found to have pending throwTos
3182 (tso->blocked_exceptions was not empty), and were blocked.
3183 Normally this doesn't happen, because we would deliver the
3184 exception directly if the target thread is blocked, but there are
3185 small windows where it might occur on a multiprocessor (see
3188 NB. we must be holding all the capabilities at this point, just
3189 like resurrectThreads().
3190 -------------------------------------------------------------------------- */
3193 performPendingThrowTos (StgTSO *threads)
3199 for (tso = threads; tso != END_TSO_QUEUE; tso = next) {
3200 next = tso->global_link;
3202 step = Bdescr((P_)tso)->step;
3203 tso->global_link = step->threads;
3204 step->threads = tso;
3206 debugTrace(DEBUG_sched, "performing blocked throwTo to thread %lu", (unsigned long)tso->id);
3209 maybePerformBlockedException(cap, tso);