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"
35 /* PARALLEL_HASKELL includes go here */
38 #include "Capability.h"
40 #include "AwaitEvent.h"
41 #if defined(mingw32_HOST_OS)
42 #include "win32/IOManager.h"
45 #include "RaiseAsync.h"
47 #include "ThrIOManager.h"
49 #ifdef HAVE_SYS_TYPES_H
50 #include <sys/types.h>
64 // Turn off inlining when debugging - it obfuscates things
67 # define STATIC_INLINE static
70 /* -----------------------------------------------------------------------------
72 * -------------------------------------------------------------------------- */
74 #if !defined(THREADED_RTS)
75 // Blocked/sleeping thrads
76 StgTSO *blocked_queue_hd = NULL;
77 StgTSO *blocked_queue_tl = NULL;
78 StgTSO *sleeping_queue = NULL; // perhaps replace with a hash table?
81 /* Threads blocked on blackholes.
82 * LOCK: sched_mutex+capability, or all capabilities
84 StgTSO *blackhole_queue = NULL;
86 /* The blackhole_queue should be checked for threads to wake up. See
87 * Schedule.h for more thorough comment.
88 * LOCK: none (doesn't matter if we miss an update)
90 rtsBool blackholes_need_checking = rtsFalse;
92 /* flag that tracks whether we have done any execution in this time slice.
93 * LOCK: currently none, perhaps we should lock (but needs to be
94 * updated in the fast path of the scheduler).
96 nat recent_activity = ACTIVITY_YES;
98 /* if this flag is set as well, give up execution
99 * LOCK: none (changes once, from false->true)
101 rtsBool sched_state = SCHED_RUNNING;
103 /* This is used in `TSO.h' and gcc 2.96 insists that this variable actually
104 * exists - earlier gccs apparently didn't.
110 * Set to TRUE when entering a shutdown state (via shutdownHaskellAndExit()) --
111 * in an MT setting, needed to signal that a worker thread shouldn't hang around
112 * in the scheduler when it is out of work.
114 rtsBool shutting_down_scheduler = rtsFalse;
117 * This mutex protects most of the global scheduler data in
118 * the THREADED_RTS runtime.
120 #if defined(THREADED_RTS)
124 #if !defined(mingw32_HOST_OS)
125 #define FORKPROCESS_PRIMOP_SUPPORTED
128 /* -----------------------------------------------------------------------------
129 * static function prototypes
130 * -------------------------------------------------------------------------- */
132 static Capability *schedule (Capability *initialCapability, Task *task);
135 // These function all encapsulate parts of the scheduler loop, and are
136 // abstracted only to make the structure and control flow of the
137 // scheduler clearer.
139 static void schedulePreLoop (void);
140 static void scheduleFindWork (Capability *cap);
141 #if defined(THREADED_RTS)
142 static void scheduleYield (Capability **pcap, Task *task);
144 static void scheduleStartSignalHandlers (Capability *cap);
145 static void scheduleCheckBlockedThreads (Capability *cap);
146 static void scheduleCheckWakeupThreads(Capability *cap USED_IF_NOT_THREADS);
147 static void scheduleCheckBlackHoles (Capability *cap);
148 static void scheduleDetectDeadlock (Capability *cap, Task *task);
149 static void schedulePushWork(Capability *cap, Task *task);
150 #if defined(PARALLEL_HASKELL)
151 static rtsBool scheduleGetRemoteWork(Capability *cap);
152 static void scheduleSendPendingMessages(void);
154 #if defined(PARALLEL_HASKELL) || defined(THREADED_RTS)
155 static void scheduleActivateSpark(Capability *cap);
157 static void schedulePostRunThread(Capability *cap, StgTSO *t);
158 static rtsBool scheduleHandleHeapOverflow( Capability *cap, StgTSO *t );
159 static void scheduleHandleStackOverflow( Capability *cap, Task *task,
161 static rtsBool scheduleHandleYield( Capability *cap, StgTSO *t,
162 nat prev_what_next );
163 static void scheduleHandleThreadBlocked( StgTSO *t );
164 static rtsBool scheduleHandleThreadFinished( Capability *cap, Task *task,
166 static rtsBool scheduleNeedHeapProfile(rtsBool ready_to_gc);
167 static Capability *scheduleDoGC(Capability *cap, Task *task,
168 rtsBool force_major);
170 static rtsBool checkBlackHoles(Capability *cap);
172 static StgTSO *threadStackOverflow(Capability *cap, StgTSO *tso);
173 static StgTSO *threadStackUnderflow(Task *task, StgTSO *tso);
175 static void deleteThread (Capability *cap, StgTSO *tso);
176 static void deleteAllThreads (Capability *cap);
178 #ifdef FORKPROCESS_PRIMOP_SUPPORTED
179 static void deleteThread_(Capability *cap, StgTSO *tso);
183 static char *whatNext_strs[] = {
193 /* -----------------------------------------------------------------------------
194 * Putting a thread on the run queue: different scheduling policies
195 * -------------------------------------------------------------------------- */
198 addToRunQueue( Capability *cap, StgTSO *t )
200 #if defined(PARALLEL_HASKELL)
201 if (RtsFlags.ParFlags.doFairScheduling) {
202 // this does round-robin scheduling; good for concurrency
203 appendToRunQueue(cap,t);
205 // this does unfair scheduling; good for parallelism
206 pushOnRunQueue(cap,t);
209 // this does round-robin scheduling; good for concurrency
210 appendToRunQueue(cap,t);
214 /* ---------------------------------------------------------------------------
215 Main scheduling loop.
217 We use round-robin scheduling, each thread returning to the
218 scheduler loop when one of these conditions is detected:
221 * timer expires (thread yields)
227 In a GranSim setup this loop iterates over the global event queue.
228 This revolves around the global event queue, which determines what
229 to do next. Therefore, it's more complicated than either the
230 concurrent or the parallel (GUM) setup.
231 This version has been entirely removed (JB 2008/08).
234 GUM iterates over incoming messages.
235 It starts with nothing to do (thus CurrentTSO == END_TSO_QUEUE),
236 and sends out a fish whenever it has nothing to do; in-between
237 doing the actual reductions (shared code below) it processes the
238 incoming messages and deals with delayed operations
239 (see PendingFetches).
240 This is not the ugliest code you could imagine, but it's bloody close.
242 (JB 2008/08) This version was formerly indicated by a PP-Flag PAR,
243 now by PP-flag PARALLEL_HASKELL. The Eden RTS (in GHC-6.x) uses it,
244 as well as future GUM versions. This file has been refurbished to
245 only contain valid code, which is however incomplete, refers to
246 invalid includes etc.
248 ------------------------------------------------------------------------ */
251 schedule (Capability *initialCapability, Task *task)
255 StgThreadReturnCode ret;
256 #if defined(PARALLEL_HASKELL)
257 rtsBool receivedFinish = rtsFalse;
261 #if defined(THREADED_RTS)
262 rtsBool first = rtsTrue;
265 cap = initialCapability;
267 // Pre-condition: this task owns initialCapability.
268 // The sched_mutex is *NOT* held
269 // NB. on return, we still hold a capability.
271 debugTrace (DEBUG_sched,
272 "### NEW SCHEDULER LOOP (task: %p, cap: %p)",
273 task, initialCapability);
277 // -----------------------------------------------------------
278 // Scheduler loop starts here:
280 #if defined(PARALLEL_HASKELL)
281 #define TERMINATION_CONDITION (!receivedFinish)
283 #define TERMINATION_CONDITION rtsTrue
286 while (TERMINATION_CONDITION) {
288 // Check whether we have re-entered the RTS from Haskell without
289 // going via suspendThread()/resumeThread (i.e. a 'safe' foreign
291 if (cap->in_haskell) {
292 errorBelch("schedule: re-entered unsafely.\n"
293 " Perhaps a 'foreign import unsafe' should be 'safe'?");
294 stg_exit(EXIT_FAILURE);
297 // The interruption / shutdown sequence.
299 // In order to cleanly shut down the runtime, we want to:
300 // * make sure that all main threads return to their callers
301 // with the state 'Interrupted'.
302 // * clean up all OS threads assocated with the runtime
303 // * free all memory etc.
305 // So the sequence for ^C goes like this:
307 // * ^C handler sets sched_state := SCHED_INTERRUPTING and
308 // arranges for some Capability to wake up
310 // * all threads in the system are halted, and the zombies are
311 // placed on the run queue for cleaning up. We acquire all
312 // the capabilities in order to delete the threads, this is
313 // done by scheduleDoGC() for convenience (because GC already
314 // needs to acquire all the capabilities). We can't kill
315 // threads involved in foreign calls.
317 // * somebody calls shutdownHaskell(), which calls exitScheduler()
319 // * sched_state := SCHED_SHUTTING_DOWN
321 // * all workers exit when the run queue on their capability
322 // drains. All main threads will also exit when their TSO
323 // reaches the head of the run queue and they can return.
325 // * eventually all Capabilities will shut down, and the RTS can
328 // * We might be left with threads blocked in foreign calls,
329 // we should really attempt to kill these somehow (TODO);
331 switch (sched_state) {
334 case SCHED_INTERRUPTING:
335 debugTrace(DEBUG_sched, "SCHED_INTERRUPTING");
336 #if defined(THREADED_RTS)
337 discardSparksCap(cap);
339 /* scheduleDoGC() deletes all the threads */
340 cap = scheduleDoGC(cap,task,rtsFalse);
342 case SCHED_SHUTTING_DOWN:
343 debugTrace(DEBUG_sched, "SCHED_SHUTTING_DOWN");
344 // If we are a worker, just exit. If we're a bound thread
345 // then we will exit below when we've removed our TSO from
347 if (task->tso == NULL && emptyRunQueue(cap)) {
352 barf("sched_state: %d", sched_state);
355 scheduleFindWork(cap);
357 /* work pushing, currently relevant only for THREADED_RTS:
358 (pushes threads, wakes up idle capabilities for stealing) */
359 schedulePushWork(cap,task);
361 #if defined(PARALLEL_HASKELL)
362 /* since we perform a blocking receive and continue otherwise,
363 either we never reach here or we definitely have work! */
364 // from here: non-empty run queue
365 ASSERT(!emptyRunQueue(cap));
367 if (PacketsWaiting()) { /* now process incoming messages, if any
370 CAUTION: scheduleGetRemoteWork called
371 above, waits for messages as well! */
372 processMessages(cap, &receivedFinish);
374 #endif // PARALLEL_HASKELL: non-empty run queue!
376 scheduleDetectDeadlock(cap,task);
378 #if defined(THREADED_RTS)
379 cap = task->cap; // reload cap, it might have changed
382 // Normally, the only way we can get here with no threads to
383 // run is if a keyboard interrupt received during
384 // scheduleCheckBlockedThreads() or scheduleDetectDeadlock().
385 // Additionally, it is not fatal for the
386 // threaded RTS to reach here with no threads to run.
388 // win32: might be here due to awaitEvent() being abandoned
389 // as a result of a console event having been delivered.
391 #if defined(THREADED_RTS)
395 // // don't yield the first time, we want a chance to run this
396 // // thread for a bit, even if there are others banging at the
399 // ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
403 scheduleYield(&cap,task);
404 if (emptyRunQueue(cap)) continue; // look for work again
407 #if !defined(THREADED_RTS) && !defined(mingw32_HOST_OS)
408 if ( emptyRunQueue(cap) ) {
409 ASSERT(sched_state >= SCHED_INTERRUPTING);
414 // Get a thread to run
416 t = popRunQueue(cap);
418 // Sanity check the thread we're about to run. This can be
419 // expensive if there is lots of thread switching going on...
420 IF_DEBUG(sanity,checkTSO(t));
422 #if defined(THREADED_RTS)
423 // Check whether we can run this thread in the current task.
424 // If not, we have to pass our capability to the right task.
426 Task *bound = t->bound;
430 debugTrace(DEBUG_sched,
431 "### Running thread %lu in bound thread", (unsigned long)t->id);
432 // yes, the Haskell thread is bound to the current native thread
434 debugTrace(DEBUG_sched,
435 "### thread %lu bound to another OS thread", (unsigned long)t->id);
436 // no, bound to a different Haskell thread: pass to that thread
437 pushOnRunQueue(cap,t);
441 // The thread we want to run is unbound.
443 debugTrace(DEBUG_sched,
444 "### this OS thread cannot run thread %lu", (unsigned long)t->id);
445 // no, the current native thread is bound to a different
446 // Haskell thread, so pass it to any worker thread
447 pushOnRunQueue(cap,t);
454 /* context switches are initiated by the timer signal, unless
455 * the user specified "context switch as often as possible", with
458 if (RtsFlags.ConcFlags.ctxtSwitchTicks == 0
459 && !emptyThreadQueues(cap)) {
460 cap->context_switch = 1;
465 // CurrentTSO is the thread to run. t might be different if we
466 // loop back to run_thread, so make sure to set CurrentTSO after
468 cap->r.rCurrentTSO = t;
470 debugTrace(DEBUG_sched, "-->> running thread %ld %s ...",
471 (long)t->id, whatNext_strs[t->what_next]);
473 startHeapProfTimer();
475 // Check for exceptions blocked on this thread
476 maybePerformBlockedException (cap, t);
478 // ----------------------------------------------------------------------
479 // Run the current thread
481 ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
482 ASSERT(t->cap == cap);
483 ASSERT(t->bound ? t->bound->cap == cap : 1);
485 prev_what_next = t->what_next;
487 errno = t->saved_errno;
489 SetLastError(t->saved_winerror);
492 cap->in_haskell = rtsTrue;
496 #if defined(THREADED_RTS)
497 if (recent_activity == ACTIVITY_DONE_GC) {
498 // ACTIVITY_DONE_GC means we turned off the timer signal to
499 // conserve power (see #1623). Re-enable it here.
501 prev = xchg((P_)&recent_activity, ACTIVITY_YES);
502 if (prev == ACTIVITY_DONE_GC) {
506 recent_activity = ACTIVITY_YES;
510 switch (prev_what_next) {
514 /* Thread already finished, return to scheduler. */
515 ret = ThreadFinished;
521 r = StgRun((StgFunPtr) stg_returnToStackTop, &cap->r);
522 cap = regTableToCapability(r);
527 case ThreadInterpret:
528 cap = interpretBCO(cap);
533 barf("schedule: invalid what_next field");
536 cap->in_haskell = rtsFalse;
538 // The TSO might have moved, eg. if it re-entered the RTS and a GC
539 // happened. So find the new location:
540 t = cap->r.rCurrentTSO;
542 // We have run some Haskell code: there might be blackhole-blocked
543 // threads to wake up now.
544 // Lock-free test here should be ok, we're just setting a flag.
545 if ( blackhole_queue != END_TSO_QUEUE ) {
546 blackholes_need_checking = rtsTrue;
549 // And save the current errno in this thread.
550 // XXX: possibly bogus for SMP because this thread might already
551 // be running again, see code below.
552 t->saved_errno = errno;
554 // Similarly for Windows error code
555 t->saved_winerror = GetLastError();
558 #if defined(THREADED_RTS)
559 // If ret is ThreadBlocked, and this Task is bound to the TSO that
560 // blocked, we are in limbo - the TSO is now owned by whatever it
561 // is blocked on, and may in fact already have been woken up,
562 // perhaps even on a different Capability. It may be the case
563 // that task->cap != cap. We better yield this Capability
564 // immediately and return to normaility.
565 if (ret == ThreadBlocked) {
566 debugTrace(DEBUG_sched,
567 "--<< thread %lu (%s) stopped: blocked",
568 (unsigned long)t->id, whatNext_strs[t->what_next]);
573 ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
574 ASSERT(t->cap == cap);
576 // ----------------------------------------------------------------------
578 // Costs for the scheduler are assigned to CCS_SYSTEM
580 #if defined(PROFILING)
584 schedulePostRunThread(cap,t);
586 t = threadStackUnderflow(task,t);
588 ready_to_gc = rtsFalse;
592 ready_to_gc = scheduleHandleHeapOverflow(cap,t);
596 scheduleHandleStackOverflow(cap,task,t);
600 if (scheduleHandleYield(cap, t, prev_what_next)) {
601 // shortcut for switching between compiler/interpreter:
607 scheduleHandleThreadBlocked(t);
611 if (scheduleHandleThreadFinished(cap, task, t)) return cap;
612 ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
616 barf("schedule: invalid thread return code %d", (int)ret);
619 if (ready_to_gc || scheduleNeedHeapProfile(ready_to_gc)) {
620 cap = scheduleDoGC(cap,task,rtsFalse);
622 } /* end of while() */
625 /* ----------------------------------------------------------------------------
626 * Setting up the scheduler loop
627 * ------------------------------------------------------------------------- */
630 schedulePreLoop(void)
632 // initialisation for scheduler - what cannot go into initScheduler()
635 /* -----------------------------------------------------------------------------
638 * Search for work to do, and handle messages from elsewhere.
639 * -------------------------------------------------------------------------- */
642 scheduleFindWork (Capability *cap)
644 scheduleStartSignalHandlers(cap);
646 // Only check the black holes here if we've nothing else to do.
647 // During normal execution, the black hole list only gets checked
648 // at GC time, to avoid repeatedly traversing this possibly long
649 // list each time around the scheduler.
650 if (emptyRunQueue(cap)) { scheduleCheckBlackHoles(cap); }
652 scheduleCheckWakeupThreads(cap);
654 scheduleCheckBlockedThreads(cap);
656 #if defined(THREADED_RTS) || defined(PARALLEL_HASKELL)
657 // Try to activate one of our own sparks
658 if (emptyRunQueue(cap)) { scheduleActivateSpark(cap); }
661 #if defined(THREADED_RTS)
662 // Try to steak work if we don't have any
663 if (emptyRunQueue(cap)) { stealWork(cap); }
666 #if defined(PARALLEL_HASKELL)
667 // if messages have been buffered...
668 scheduleSendPendingMessages();
671 #if defined(PARALLEL_HASKELL)
672 if (emptyRunQueue(cap)) {
673 receivedFinish = scheduleGetRemoteWork(cap);
674 continue; // a new round, (hopefully) with new work
676 in GUM, this a) sends out a FISH and returns IF no fish is
678 b) (blocking) awaits and receives messages
680 in Eden, this is only the blocking receive, as b) in GUM.
686 #if defined(THREADED_RTS)
687 STATIC_INLINE rtsBool
688 shouldYieldCapability (Capability *cap, Task *task)
690 // we need to yield this capability to someone else if..
691 // - another thread is initiating a GC
692 // - another Task is returning from a foreign call
693 // - the thread at the head of the run queue cannot be run
694 // by this Task (it is bound to another Task, or it is unbound
695 // and this task it bound).
696 return (waiting_for_gc ||
697 cap->returning_tasks_hd != NULL ||
698 (!emptyRunQueue(cap) && (task->tso == NULL
699 ? cap->run_queue_hd->bound != NULL
700 : cap->run_queue_hd->bound != task)));
703 // This is the single place where a Task goes to sleep. There are
704 // two reasons it might need to sleep:
705 // - there are no threads to run
706 // - we need to yield this Capability to someone else
707 // (see shouldYieldCapability())
709 // The return value indicates whether
712 scheduleYield (Capability **pcap, Task *task)
714 Capability *cap = *pcap;
716 // if we have work, and we don't need to give up the Capability, continue.
717 if (!emptyRunQueue(cap) && !shouldYieldCapability(cap,task))
720 // otherwise yield (sleep), and keep yielding if necessary.
722 yieldCapability(&cap,task);
724 while (shouldYieldCapability(cap,task));
726 // note there may still be no threads on the run queue at this
727 // point, the caller has to check.
734 /* -----------------------------------------------------------------------------
737 * Push work to other Capabilities if we have some.
738 * -------------------------------------------------------------------------- */
741 schedulePushWork(Capability *cap USED_IF_THREADS,
742 Task *task USED_IF_THREADS)
744 /* following code not for PARALLEL_HASKELL. I kept the call general,
745 future GUM versions might use pushing in a distributed setup */
746 #if defined(THREADED_RTS)
748 Capability *free_caps[n_capabilities], *cap0;
751 // migration can be turned off with +RTS -qg
752 if (!RtsFlags.ParFlags.migrate) return;
754 // Check whether we have more threads on our run queue, or sparks
755 // in our pool, that we could hand to another Capability.
756 if ((emptyRunQueue(cap) || cap->run_queue_hd->_link == END_TSO_QUEUE)
757 && sparkPoolSizeCap(cap) < 2) {
761 // First grab as many free Capabilities as we can.
762 for (i=0, n_free_caps=0; i < n_capabilities; i++) {
763 cap0 = &capabilities[i];
764 if (cap != cap0 && tryGrabCapability(cap0,task)) {
765 if (!emptyRunQueue(cap0) || cap->returning_tasks_hd != NULL) {
766 // it already has some work, we just grabbed it at
767 // the wrong moment. Or maybe it's deadlocked!
768 releaseCapability(cap0);
770 free_caps[n_free_caps++] = cap0;
775 // we now have n_free_caps free capabilities stashed in
776 // free_caps[]. Share our run queue equally with them. This is
777 // probably the simplest thing we could do; improvements we might
778 // want to do include:
780 // - giving high priority to moving relatively new threads, on
781 // the gournds that they haven't had time to build up a
782 // working set in the cache on this CPU/Capability.
784 // - giving low priority to moving long-lived threads
786 if (n_free_caps > 0) {
787 StgTSO *prev, *t, *next;
788 rtsBool pushed_to_all;
790 debugTrace(DEBUG_sched,
791 "cap %d: %s and %d free capabilities, sharing...",
793 (!emptyRunQueue(cap) && cap->run_queue_hd->_link != END_TSO_QUEUE)?
794 "excess threads on run queue":"sparks to share (>=2)",
798 pushed_to_all = rtsFalse;
800 if (cap->run_queue_hd != END_TSO_QUEUE) {
801 prev = cap->run_queue_hd;
803 prev->_link = END_TSO_QUEUE;
804 for (; t != END_TSO_QUEUE; t = next) {
806 t->_link = END_TSO_QUEUE;
807 if (t->what_next == ThreadRelocated
808 || t->bound == task // don't move my bound thread
809 || tsoLocked(t)) { // don't move a locked thread
810 setTSOLink(cap, prev, t);
812 } else if (i == n_free_caps) {
813 pushed_to_all = rtsTrue;
816 setTSOLink(cap, prev, t);
819 debugTrace(DEBUG_sched, "pushing thread %lu to capability %d", (unsigned long)t->id, free_caps[i]->no);
820 appendToRunQueue(free_caps[i],t);
821 if (t->bound) { t->bound->cap = free_caps[i]; }
822 t->cap = free_caps[i];
826 cap->run_queue_tl = prev;
830 /* JB I left this code in place, it would work but is not necessary */
832 // If there are some free capabilities that we didn't push any
833 // threads to, then try to push a spark to each one.
834 if (!pushed_to_all) {
836 // i is the next free capability to push to
837 for (; i < n_free_caps; i++) {
838 if (emptySparkPoolCap(free_caps[i])) {
839 spark = tryStealSpark(cap->sparks);
841 debugTrace(DEBUG_sched, "pushing spark %p to capability %d", spark, free_caps[i]->no);
842 newSpark(&(free_caps[i]->r), spark);
847 #endif /* SPARK_PUSHING */
849 // release the capabilities
850 for (i = 0; i < n_free_caps; i++) {
851 task->cap = free_caps[i];
852 releaseAndWakeupCapability(free_caps[i]);
855 task->cap = cap; // reset to point to our Capability.
857 #endif /* THREADED_RTS */
861 /* ----------------------------------------------------------------------------
862 * Start any pending signal handlers
863 * ------------------------------------------------------------------------- */
865 #if defined(RTS_USER_SIGNALS) && !defined(THREADED_RTS)
867 scheduleStartSignalHandlers(Capability *cap)
869 if (RtsFlags.MiscFlags.install_signal_handlers && signals_pending()) {
870 // safe outside the lock
871 startSignalHandlers(cap);
876 scheduleStartSignalHandlers(Capability *cap STG_UNUSED)
881 /* ----------------------------------------------------------------------------
882 * Check for blocked threads that can be woken up.
883 * ------------------------------------------------------------------------- */
886 scheduleCheckBlockedThreads(Capability *cap USED_IF_NOT_THREADS)
888 #if !defined(THREADED_RTS)
890 // Check whether any waiting threads need to be woken up. If the
891 // run queue is empty, and there are no other tasks running, we
892 // can wait indefinitely for something to happen.
894 if ( !emptyQueue(blocked_queue_hd) || !emptyQueue(sleeping_queue) )
896 awaitEvent( emptyRunQueue(cap) && !blackholes_need_checking );
902 /* ----------------------------------------------------------------------------
903 * Check for threads woken up by other Capabilities
904 * ------------------------------------------------------------------------- */
907 scheduleCheckWakeupThreads(Capability *cap USED_IF_THREADS)
909 #if defined(THREADED_RTS)
910 // Any threads that were woken up by other Capabilities get
911 // appended to our run queue.
912 if (!emptyWakeupQueue(cap)) {
913 ACQUIRE_LOCK(&cap->lock);
914 if (emptyRunQueue(cap)) {
915 cap->run_queue_hd = cap->wakeup_queue_hd;
916 cap->run_queue_tl = cap->wakeup_queue_tl;
918 setTSOLink(cap, cap->run_queue_tl, cap->wakeup_queue_hd);
919 cap->run_queue_tl = cap->wakeup_queue_tl;
921 cap->wakeup_queue_hd = cap->wakeup_queue_tl = END_TSO_QUEUE;
922 RELEASE_LOCK(&cap->lock);
927 /* ----------------------------------------------------------------------------
928 * Check for threads blocked on BLACKHOLEs that can be woken up
929 * ------------------------------------------------------------------------- */
931 scheduleCheckBlackHoles (Capability *cap)
933 if ( blackholes_need_checking ) // check without the lock first
935 ACQUIRE_LOCK(&sched_mutex);
936 if ( blackholes_need_checking ) {
937 blackholes_need_checking = rtsFalse;
938 // important that we reset the flag *before* checking the
939 // blackhole queue, otherwise we could get deadlock. This
940 // happens as follows: we wake up a thread that
941 // immediately runs on another Capability, blocks on a
942 // blackhole, and then we reset the blackholes_need_checking flag.
943 checkBlackHoles(cap);
945 RELEASE_LOCK(&sched_mutex);
949 /* ----------------------------------------------------------------------------
950 * Detect deadlock conditions and attempt to resolve them.
951 * ------------------------------------------------------------------------- */
954 scheduleDetectDeadlock (Capability *cap, Task *task)
957 #if defined(PARALLEL_HASKELL)
958 // ToDo: add deadlock detection in GUM (similar to THREADED_RTS) -- HWL
963 * Detect deadlock: when we have no threads to run, there are no
964 * threads blocked, waiting for I/O, or sleeping, and all the
965 * other tasks are waiting for work, we must have a deadlock of
968 if ( emptyThreadQueues(cap) )
970 #if defined(THREADED_RTS)
972 * In the threaded RTS, we only check for deadlock if there
973 * has been no activity in a complete timeslice. This means
974 * we won't eagerly start a full GC just because we don't have
975 * any threads to run currently.
977 if (recent_activity != ACTIVITY_INACTIVE) return;
980 debugTrace(DEBUG_sched, "deadlocked, forcing major GC...");
982 // Garbage collection can release some new threads due to
983 // either (a) finalizers or (b) threads resurrected because
984 // they are unreachable and will therefore be sent an
985 // exception. Any threads thus released will be immediately
987 cap = scheduleDoGC (cap, task, rtsTrue/*force major GC*/);
989 recent_activity = ACTIVITY_DONE_GC;
990 // disable timer signals (see #1623)
993 if ( !emptyRunQueue(cap) ) return;
995 #if defined(RTS_USER_SIGNALS) && !defined(THREADED_RTS)
996 /* If we have user-installed signal handlers, then wait
997 * for signals to arrive rather then bombing out with a
1000 if ( RtsFlags.MiscFlags.install_signal_handlers && anyUserHandlers() ) {
1001 debugTrace(DEBUG_sched,
1002 "still deadlocked, waiting for signals...");
1006 if (signals_pending()) {
1007 startSignalHandlers(cap);
1010 // either we have threads to run, or we were interrupted:
1011 ASSERT(!emptyRunQueue(cap) || sched_state >= SCHED_INTERRUPTING);
1017 #if !defined(THREADED_RTS)
1018 /* Probably a real deadlock. Send the current main thread the
1019 * Deadlock exception.
1022 switch (task->tso->why_blocked) {
1024 case BlockedOnBlackHole:
1025 case BlockedOnException:
1027 throwToSingleThreaded(cap, task->tso,
1028 (StgClosure *)nonTermination_closure);
1031 barf("deadlock: main thread blocked in a strange way");
1040 /* ----------------------------------------------------------------------------
1041 * Send pending messages (PARALLEL_HASKELL only)
1042 * ------------------------------------------------------------------------- */
1044 #if defined(PARALLEL_HASKELL)
1046 scheduleSendPendingMessages(void)
1049 # if defined(PAR) // global Mem.Mgmt., omit for now
1050 if (PendingFetches != END_BF_QUEUE) {
1055 if (RtsFlags.ParFlags.BufferTime) {
1056 // if we use message buffering, we must send away all message
1057 // packets which have become too old...
1063 /* ----------------------------------------------------------------------------
1064 * Activate spark threads (PARALLEL_HASKELL and THREADED_RTS)
1065 * ------------------------------------------------------------------------- */
1067 #if defined(PARALLEL_HASKELL) || defined(THREADED_RTS)
1069 scheduleActivateSpark(Capability *cap)
1073 /* We only want to stay here if the run queue is empty and we want some
1074 work. We try to turn a spark into a thread, and add it to the run
1075 queue, from where it will be picked up in the next iteration of the
1078 if (!emptyRunQueue(cap))
1079 /* In the threaded RTS, another task might have pushed a thread
1080 on our run queue in the meantime ? But would need a lock.. */
1084 // Really we should be using reclaimSpark() here, but
1085 // experimentally it doesn't seem to perform as well as just
1086 // stealing from our own spark pool:
1087 // spark = reclaimSpark(cap->sparks);
1088 spark = tryStealSpark(cap->sparks); // defined in Sparks.c
1090 if (spark != NULL) {
1091 debugTrace(DEBUG_sched,
1092 "turning spark of closure %p into a thread",
1093 (StgClosure *)spark);
1094 createSparkThread(cap,spark); // defined in Sparks.c
1097 #endif // PARALLEL_HASKELL || THREADED_RTS
1099 /* ----------------------------------------------------------------------------
1100 * Get work from a remote node (PARALLEL_HASKELL only)
1101 * ------------------------------------------------------------------------- */
1103 #if defined(PARALLEL_HASKELL)
1104 static rtsBool /* return value used in PARALLEL_HASKELL only */
1105 scheduleGetRemoteWork (Capability *cap STG_UNUSED)
1107 #if defined(PARALLEL_HASKELL)
1108 rtsBool receivedFinish = rtsFalse;
1110 // idle() , i.e. send all buffers, wait for work
1111 if (RtsFlags.ParFlags.BufferTime) {
1112 IF_PAR_DEBUG(verbose,
1113 debugBelch("...send all pending data,"));
1116 for (i=1; i<=nPEs; i++)
1117 sendImmediately(i); // send all messages away immediately
1121 /* this would be the place for fishing in GUM...
1123 if (no-earlier-fish-around)
1124 sendFish(choosePe());
1127 // Eden:just look for incoming messages (blocking receive)
1128 IF_PAR_DEBUG(verbose,
1129 debugBelch("...wait for incoming messages...\n"));
1130 processMessages(cap, &receivedFinish); // blocking receive...
1133 return receivedFinish;
1134 // reenter scheduling look after having received something
1136 #else /* !PARALLEL_HASKELL, i.e. THREADED_RTS */
1138 return rtsFalse; /* return value unused in THREADED_RTS */
1140 #endif /* PARALLEL_HASKELL */
1142 #endif // PARALLEL_HASKELL || THREADED_RTS
1144 /* ----------------------------------------------------------------------------
1145 * After running a thread...
1146 * ------------------------------------------------------------------------- */
1149 schedulePostRunThread (Capability *cap, StgTSO *t)
1151 // We have to be able to catch transactions that are in an
1152 // infinite loop as a result of seeing an inconsistent view of
1156 // [a,b] <- mapM readTVar [ta,tb]
1157 // when (a == b) loop
1159 // and a is never equal to b given a consistent view of memory.
1161 if (t -> trec != NO_TREC && t -> why_blocked == NotBlocked) {
1162 if (!stmValidateNestOfTransactions (t -> trec)) {
1163 debugTrace(DEBUG_sched | DEBUG_stm,
1164 "trec %p found wasting its time", t);
1166 // strip the stack back to the
1167 // ATOMICALLY_FRAME, aborting the (nested)
1168 // transaction, and saving the stack of any
1169 // partially-evaluated thunks on the heap.
1170 throwToSingleThreaded_(cap, t, NULL, rtsTrue, NULL);
1172 ASSERT(get_itbl((StgClosure *)t->sp)->type == ATOMICALLY_FRAME);
1176 /* some statistics gathering in the parallel case */
1179 /* -----------------------------------------------------------------------------
1180 * Handle a thread that returned to the scheduler with ThreadHeepOverflow
1181 * -------------------------------------------------------------------------- */
1184 scheduleHandleHeapOverflow( Capability *cap, StgTSO *t )
1186 // did the task ask for a large block?
1187 if (cap->r.rHpAlloc > BLOCK_SIZE) {
1188 // if so, get one and push it on the front of the nursery.
1192 blocks = (lnat)BLOCK_ROUND_UP(cap->r.rHpAlloc) / BLOCK_SIZE;
1194 debugTrace(DEBUG_sched,
1195 "--<< thread %ld (%s) stopped: requesting a large block (size %ld)\n",
1196 (long)t->id, whatNext_strs[t->what_next], blocks);
1198 // don't do this if the nursery is (nearly) full, we'll GC first.
1199 if (cap->r.rCurrentNursery->link != NULL ||
1200 cap->r.rNursery->n_blocks == 1) { // paranoia to prevent infinite loop
1201 // if the nursery has only one block.
1204 bd = allocGroup( blocks );
1206 cap->r.rNursery->n_blocks += blocks;
1208 // link the new group into the list
1209 bd->link = cap->r.rCurrentNursery;
1210 bd->u.back = cap->r.rCurrentNursery->u.back;
1211 if (cap->r.rCurrentNursery->u.back != NULL) {
1212 cap->r.rCurrentNursery->u.back->link = bd;
1214 #if !defined(THREADED_RTS)
1215 ASSERT(g0s0->blocks == cap->r.rCurrentNursery &&
1216 g0s0 == cap->r.rNursery);
1218 cap->r.rNursery->blocks = bd;
1220 cap->r.rCurrentNursery->u.back = bd;
1222 // initialise it as a nursery block. We initialise the
1223 // step, gen_no, and flags field of *every* sub-block in
1224 // this large block, because this is easier than making
1225 // sure that we always find the block head of a large
1226 // block whenever we call Bdescr() (eg. evacuate() and
1227 // isAlive() in the GC would both have to do this, at
1231 for (x = bd; x < bd + blocks; x++) {
1232 x->step = cap->r.rNursery;
1238 // This assert can be a killer if the app is doing lots
1239 // of large block allocations.
1240 IF_DEBUG(sanity, checkNurserySanity(cap->r.rNursery));
1242 // now update the nursery to point to the new block
1243 cap->r.rCurrentNursery = bd;
1245 // we might be unlucky and have another thread get on the
1246 // run queue before us and steal the large block, but in that
1247 // case the thread will just end up requesting another large
1249 pushOnRunQueue(cap,t);
1250 return rtsFalse; /* not actually GC'ing */
1254 debugTrace(DEBUG_sched,
1255 "--<< thread %ld (%s) stopped: HeapOverflow",
1256 (long)t->id, whatNext_strs[t->what_next]);
1258 if (cap->context_switch) {
1259 // Sometimes we miss a context switch, e.g. when calling
1260 // primitives in a tight loop, MAYBE_GC() doesn't check the
1261 // context switch flag, and we end up waiting for a GC.
1262 // See #1984, and concurrent/should_run/1984
1263 cap->context_switch = 0;
1264 addToRunQueue(cap,t);
1266 pushOnRunQueue(cap,t);
1269 /* actual GC is done at the end of the while loop in schedule() */
1272 /* -----------------------------------------------------------------------------
1273 * Handle a thread that returned to the scheduler with ThreadStackOverflow
1274 * -------------------------------------------------------------------------- */
1277 scheduleHandleStackOverflow (Capability *cap, Task *task, StgTSO *t)
1279 debugTrace (DEBUG_sched,
1280 "--<< thread %ld (%s) stopped, StackOverflow",
1281 (long)t->id, whatNext_strs[t->what_next]);
1283 /* just adjust the stack for this thread, then pop it back
1287 /* enlarge the stack */
1288 StgTSO *new_t = threadStackOverflow(cap, t);
1290 /* The TSO attached to this Task may have moved, so update the
1293 if (task->tso == t) {
1296 pushOnRunQueue(cap,new_t);
1300 /* -----------------------------------------------------------------------------
1301 * Handle a thread that returned to the scheduler with ThreadYielding
1302 * -------------------------------------------------------------------------- */
1305 scheduleHandleYield( Capability *cap, StgTSO *t, nat prev_what_next )
1307 // Reset the context switch flag. We don't do this just before
1308 // running the thread, because that would mean we would lose ticks
1309 // during GC, which can lead to unfair scheduling (a thread hogs
1310 // the CPU because the tick always arrives during GC). This way
1311 // penalises threads that do a lot of allocation, but that seems
1312 // better than the alternative.
1313 cap->context_switch = 0;
1315 /* put the thread back on the run queue. Then, if we're ready to
1316 * GC, check whether this is the last task to stop. If so, wake
1317 * up the GC thread. getThread will block during a GC until the
1321 if (t->what_next != prev_what_next) {
1322 debugTrace(DEBUG_sched,
1323 "--<< thread %ld (%s) stopped to switch evaluators",
1324 (long)t->id, whatNext_strs[t->what_next]);
1326 debugTrace(DEBUG_sched,
1327 "--<< thread %ld (%s) stopped, yielding",
1328 (long)t->id, whatNext_strs[t->what_next]);
1333 //debugBelch("&& Doing sanity check on yielding TSO %ld.", t->id);
1335 ASSERT(t->_link == END_TSO_QUEUE);
1337 // Shortcut if we're just switching evaluators: don't bother
1338 // doing stack squeezing (which can be expensive), just run the
1340 if (t->what_next != prev_what_next) {
1344 addToRunQueue(cap,t);
1349 /* -----------------------------------------------------------------------------
1350 * Handle a thread that returned to the scheduler with ThreadBlocked
1351 * -------------------------------------------------------------------------- */
1354 scheduleHandleThreadBlocked( StgTSO *t
1355 #if !defined(GRAN) && !defined(DEBUG)
1361 // We don't need to do anything. The thread is blocked, and it
1362 // has tidied up its stack and placed itself on whatever queue
1363 // it needs to be on.
1365 // ASSERT(t->why_blocked != NotBlocked);
1366 // Not true: for example,
1367 // - in THREADED_RTS, the thread may already have been woken
1368 // up by another Capability. This actually happens: try
1369 // conc023 +RTS -N2.
1370 // - the thread may have woken itself up already, because
1371 // threadPaused() might have raised a blocked throwTo
1372 // exception, see maybePerformBlockedException().
1375 if (traceClass(DEBUG_sched)) {
1376 debugTraceBegin("--<< thread %lu (%s) stopped: ",
1377 (unsigned long)t->id, whatNext_strs[t->what_next]);
1378 printThreadBlockage(t);
1384 /* -----------------------------------------------------------------------------
1385 * Handle a thread that returned to the scheduler with ThreadFinished
1386 * -------------------------------------------------------------------------- */
1389 scheduleHandleThreadFinished (Capability *cap STG_UNUSED, Task *task, StgTSO *t)
1391 /* Need to check whether this was a main thread, and if so,
1392 * return with the return value.
1394 * We also end up here if the thread kills itself with an
1395 * uncaught exception, see Exception.cmm.
1397 debugTrace(DEBUG_sched, "--++ thread %lu (%s) finished",
1398 (unsigned long)t->id, whatNext_strs[t->what_next]);
1401 // Check whether the thread that just completed was a bound
1402 // thread, and if so return with the result.
1404 // There is an assumption here that all thread completion goes
1405 // through this point; we need to make sure that if a thread
1406 // ends up in the ThreadKilled state, that it stays on the run
1407 // queue so it can be dealt with here.
1412 if (t->bound != task) {
1413 #if !defined(THREADED_RTS)
1414 // Must be a bound thread that is not the topmost one. Leave
1415 // it on the run queue until the stack has unwound to the
1416 // point where we can deal with this. Leaving it on the run
1417 // queue also ensures that the garbage collector knows about
1418 // this thread and its return value (it gets dropped from the
1419 // step->threads list so there's no other way to find it).
1420 appendToRunQueue(cap,t);
1423 // this cannot happen in the threaded RTS, because a
1424 // bound thread can only be run by the appropriate Task.
1425 barf("finished bound thread that isn't mine");
1429 ASSERT(task->tso == t);
1431 if (t->what_next == ThreadComplete) {
1433 // NOTE: return val is tso->sp[1] (see StgStartup.hc)
1434 *(task->ret) = (StgClosure *)task->tso->sp[1];
1436 task->stat = Success;
1439 *(task->ret) = NULL;
1441 if (sched_state >= SCHED_INTERRUPTING) {
1442 task->stat = Interrupted;
1444 task->stat = Killed;
1448 removeThreadLabel((StgWord)task->tso->id);
1450 return rtsTrue; // tells schedule() to return
1456 /* -----------------------------------------------------------------------------
1457 * Perform a heap census
1458 * -------------------------------------------------------------------------- */
1461 scheduleNeedHeapProfile( rtsBool ready_to_gc STG_UNUSED )
1463 // When we have +RTS -i0 and we're heap profiling, do a census at
1464 // every GC. This lets us get repeatable runs for debugging.
1465 if (performHeapProfile ||
1466 (RtsFlags.ProfFlags.profileInterval==0 &&
1467 RtsFlags.ProfFlags.doHeapProfile && ready_to_gc)) {
1474 /* -----------------------------------------------------------------------------
1475 * Perform a garbage collection if necessary
1476 * -------------------------------------------------------------------------- */
1479 scheduleDoGC (Capability *cap, Task *task USED_IF_THREADS, rtsBool force_major)
1481 rtsBool heap_census;
1483 /* extern static volatile StgWord waiting_for_gc;
1484 lives inside capability.c */
1485 rtsBool was_waiting;
1490 // In order to GC, there must be no threads running Haskell code.
1491 // Therefore, the GC thread needs to hold *all* the capabilities,
1492 // and release them after the GC has completed.
1494 // This seems to be the simplest way: previous attempts involved
1495 // making all the threads with capabilities give up their
1496 // capabilities and sleep except for the *last* one, which
1497 // actually did the GC. But it's quite hard to arrange for all
1498 // the other tasks to sleep and stay asleep.
1501 /* Other capabilities are prevented from running yet more Haskell
1502 threads if waiting_for_gc is set. Tested inside
1503 yieldCapability() and releaseCapability() in Capability.c */
1505 was_waiting = cas(&waiting_for_gc, 0, 1);
1508 debugTrace(DEBUG_sched, "someone else is trying to GC...");
1509 if (cap) yieldCapability(&cap,task);
1510 } while (waiting_for_gc);
1511 return cap; // NOTE: task->cap might have changed here
1514 setContextSwitches();
1515 for (i=0; i < n_capabilities; i++) {
1516 debugTrace(DEBUG_sched, "ready_to_gc, grabbing all the capabilies (%d/%d)", i, n_capabilities);
1517 if (cap != &capabilities[i]) {
1518 Capability *pcap = &capabilities[i];
1519 // we better hope this task doesn't get migrated to
1520 // another Capability while we're waiting for this one.
1521 // It won't, because load balancing happens while we have
1522 // all the Capabilities, but even so it's a slightly
1523 // unsavoury invariant.
1525 waitForReturnCapability(&pcap, task);
1526 if (pcap != &capabilities[i]) {
1527 barf("scheduleDoGC: got the wrong capability");
1532 waiting_for_gc = rtsFalse;
1535 // so this happens periodically:
1536 if (cap) scheduleCheckBlackHoles(cap);
1538 IF_DEBUG(scheduler, printAllThreads());
1541 * We now have all the capabilities; if we're in an interrupting
1542 * state, then we should take the opportunity to delete all the
1543 * threads in the system.
1545 if (sched_state >= SCHED_INTERRUPTING) {
1546 deleteAllThreads(&capabilities[0]);
1547 sched_state = SCHED_SHUTTING_DOWN;
1550 heap_census = scheduleNeedHeapProfile(rtsTrue);
1552 /* everybody back, start the GC.
1553 * Could do it in this thread, or signal a condition var
1554 * to do it in another thread. Either way, we need to
1555 * broadcast on gc_pending_cond afterward.
1557 #if defined(THREADED_RTS)
1558 debugTrace(DEBUG_sched, "doing GC");
1560 GarbageCollect(force_major || heap_census);
1563 debugTrace(DEBUG_sched, "performing heap census");
1565 performHeapProfile = rtsFalse;
1570 Once we are all together... this would be the place to balance all
1571 spark pools. No concurrent stealing or adding of new sparks can
1572 occur. Should be defined in Sparks.c. */
1573 balanceSparkPoolsCaps(n_capabilities, capabilities);
1576 #if defined(THREADED_RTS)
1577 // release our stash of capabilities.
1578 for (i = 0; i < n_capabilities; i++) {
1579 if (cap != &capabilities[i]) {
1580 task->cap = &capabilities[i];
1581 releaseCapability(&capabilities[i]);
1594 /* ---------------------------------------------------------------------------
1595 * Singleton fork(). Do not copy any running threads.
1596 * ------------------------------------------------------------------------- */
1599 forkProcess(HsStablePtr *entry
1600 #ifndef FORKPROCESS_PRIMOP_SUPPORTED
1605 #ifdef FORKPROCESS_PRIMOP_SUPPORTED
1612 #if defined(THREADED_RTS)
1613 if (RtsFlags.ParFlags.nNodes > 1) {
1614 errorBelch("forking not supported with +RTS -N<n> greater than 1");
1615 stg_exit(EXIT_FAILURE);
1619 debugTrace(DEBUG_sched, "forking!");
1621 // ToDo: for SMP, we should probably acquire *all* the capabilities
1624 // no funny business: hold locks while we fork, otherwise if some
1625 // other thread is holding a lock when the fork happens, the data
1626 // structure protected by the lock will forever be in an
1627 // inconsistent state in the child. See also #1391.
1628 ACQUIRE_LOCK(&sched_mutex);
1629 ACQUIRE_LOCK(&cap->lock);
1630 ACQUIRE_LOCK(&cap->running_task->lock);
1634 if (pid) { // parent
1636 RELEASE_LOCK(&sched_mutex);
1637 RELEASE_LOCK(&cap->lock);
1638 RELEASE_LOCK(&cap->running_task->lock);
1640 // just return the pid
1646 #if defined(THREADED_RTS)
1647 initMutex(&sched_mutex);
1648 initMutex(&cap->lock);
1649 initMutex(&cap->running_task->lock);
1652 // Now, all OS threads except the thread that forked are
1653 // stopped. We need to stop all Haskell threads, including
1654 // those involved in foreign calls. Also we need to delete
1655 // all Tasks, because they correspond to OS threads that are
1658 for (s = 0; s < total_steps; s++) {
1659 for (t = all_steps[s].threads; t != END_TSO_QUEUE; t = next) {
1660 if (t->what_next == ThreadRelocated) {
1663 next = t->global_link;
1664 // don't allow threads to catch the ThreadKilled
1665 // exception, but we do want to raiseAsync() because these
1666 // threads may be evaluating thunks that we need later.
1667 deleteThread_(cap,t);
1672 // Empty the run queue. It seems tempting to let all the
1673 // killed threads stay on the run queue as zombies to be
1674 // cleaned up later, but some of them correspond to bound
1675 // threads for which the corresponding Task does not exist.
1676 cap->run_queue_hd = END_TSO_QUEUE;
1677 cap->run_queue_tl = END_TSO_QUEUE;
1679 // Any suspended C-calling Tasks are no more, their OS threads
1681 cap->suspended_ccalling_tasks = NULL;
1683 // Empty the threads lists. Otherwise, the garbage
1684 // collector may attempt to resurrect some of these threads.
1685 for (s = 0; s < total_steps; s++) {
1686 all_steps[s].threads = END_TSO_QUEUE;
1689 // Wipe the task list, except the current Task.
1690 ACQUIRE_LOCK(&sched_mutex);
1691 for (task = all_tasks; task != NULL; task=task->all_link) {
1692 if (task != cap->running_task) {
1693 #if defined(THREADED_RTS)
1694 initMutex(&task->lock); // see #1391
1699 RELEASE_LOCK(&sched_mutex);
1701 #if defined(THREADED_RTS)
1702 // Wipe our spare workers list, they no longer exist. New
1703 // workers will be created if necessary.
1704 cap->spare_workers = NULL;
1705 cap->returning_tasks_hd = NULL;
1706 cap->returning_tasks_tl = NULL;
1709 // On Unix, all timers are reset in the child, so we need to start
1714 cap = rts_evalStableIO(cap, entry, NULL); // run the action
1715 rts_checkSchedStatus("forkProcess",cap);
1718 hs_exit(); // clean up and exit
1719 stg_exit(EXIT_SUCCESS);
1721 #else /* !FORKPROCESS_PRIMOP_SUPPORTED */
1722 barf("forkProcess#: primop not supported on this platform, sorry!\n");
1727 /* ---------------------------------------------------------------------------
1728 * Delete all the threads in the system
1729 * ------------------------------------------------------------------------- */
1732 deleteAllThreads ( Capability *cap )
1734 // NOTE: only safe to call if we own all capabilities.
1739 debugTrace(DEBUG_sched,"deleting all threads");
1740 for (s = 0; s < total_steps; s++) {
1741 for (t = all_steps[s].threads; t != END_TSO_QUEUE; t = next) {
1742 if (t->what_next == ThreadRelocated) {
1745 next = t->global_link;
1746 deleteThread(cap,t);
1751 // The run queue now contains a bunch of ThreadKilled threads. We
1752 // must not throw these away: the main thread(s) will be in there
1753 // somewhere, and the main scheduler loop has to deal with it.
1754 // Also, the run queue is the only thing keeping these threads from
1755 // being GC'd, and we don't want the "main thread has been GC'd" panic.
1757 #if !defined(THREADED_RTS)
1758 ASSERT(blocked_queue_hd == END_TSO_QUEUE);
1759 ASSERT(sleeping_queue == END_TSO_QUEUE);
1763 /* -----------------------------------------------------------------------------
1764 Managing the suspended_ccalling_tasks list.
1765 Locks required: sched_mutex
1766 -------------------------------------------------------------------------- */
1769 suspendTask (Capability *cap, Task *task)
1771 ASSERT(task->next == NULL && task->prev == NULL);
1772 task->next = cap->suspended_ccalling_tasks;
1774 if (cap->suspended_ccalling_tasks) {
1775 cap->suspended_ccalling_tasks->prev = task;
1777 cap->suspended_ccalling_tasks = task;
1781 recoverSuspendedTask (Capability *cap, Task *task)
1784 task->prev->next = task->next;
1786 ASSERT(cap->suspended_ccalling_tasks == task);
1787 cap->suspended_ccalling_tasks = task->next;
1790 task->next->prev = task->prev;
1792 task->next = task->prev = NULL;
1795 /* ---------------------------------------------------------------------------
1796 * Suspending & resuming Haskell threads.
1798 * When making a "safe" call to C (aka _ccall_GC), the task gives back
1799 * its capability before calling the C function. This allows another
1800 * task to pick up the capability and carry on running Haskell
1801 * threads. It also means that if the C call blocks, it won't lock
1804 * The Haskell thread making the C call is put to sleep for the
1805 * duration of the call, on the susepended_ccalling_threads queue. We
1806 * give out a token to the task, which it can use to resume the thread
1807 * on return from the C function.
1808 * ------------------------------------------------------------------------- */
1811 suspendThread (StgRegTable *reg)
1818 StgWord32 saved_winerror;
1821 saved_errno = errno;
1823 saved_winerror = GetLastError();
1826 /* assume that *reg is a pointer to the StgRegTable part of a Capability.
1828 cap = regTableToCapability(reg);
1830 task = cap->running_task;
1831 tso = cap->r.rCurrentTSO;
1833 debugTrace(DEBUG_sched,
1834 "thread %lu did a safe foreign call",
1835 (unsigned long)cap->r.rCurrentTSO->id);
1837 // XXX this might not be necessary --SDM
1838 tso->what_next = ThreadRunGHC;
1840 threadPaused(cap,tso);
1842 if ((tso->flags & TSO_BLOCKEX) == 0) {
1843 tso->why_blocked = BlockedOnCCall;
1844 tso->flags |= TSO_BLOCKEX;
1845 tso->flags &= ~TSO_INTERRUPTIBLE;
1847 tso->why_blocked = BlockedOnCCall_NoUnblockExc;
1850 // Hand back capability
1851 task->suspended_tso = tso;
1853 ACQUIRE_LOCK(&cap->lock);
1855 suspendTask(cap,task);
1856 cap->in_haskell = rtsFalse;
1857 releaseCapability_(cap,rtsFalse);
1859 RELEASE_LOCK(&cap->lock);
1861 #if defined(THREADED_RTS)
1862 /* Preparing to leave the RTS, so ensure there's a native thread/task
1863 waiting to take over.
1865 debugTrace(DEBUG_sched, "thread %lu: leaving RTS", (unsigned long)tso->id);
1868 errno = saved_errno;
1870 SetLastError(saved_winerror);
1876 resumeThread (void *task_)
1883 StgWord32 saved_winerror;
1886 saved_errno = errno;
1888 saved_winerror = GetLastError();
1892 // Wait for permission to re-enter the RTS with the result.
1893 waitForReturnCapability(&cap,task);
1894 // we might be on a different capability now... but if so, our
1895 // entry on the suspended_ccalling_tasks list will also have been
1898 // Remove the thread from the suspended list
1899 recoverSuspendedTask(cap,task);
1901 tso = task->suspended_tso;
1902 task->suspended_tso = NULL;
1903 tso->_link = END_TSO_QUEUE; // no write barrier reqd
1904 debugTrace(DEBUG_sched, "thread %lu: re-entering RTS", (unsigned long)tso->id);
1906 if (tso->why_blocked == BlockedOnCCall) {
1907 awakenBlockedExceptionQueue(cap,tso);
1908 tso->flags &= ~(TSO_BLOCKEX | TSO_INTERRUPTIBLE);
1911 /* Reset blocking status */
1912 tso->why_blocked = NotBlocked;
1914 cap->r.rCurrentTSO = tso;
1915 cap->in_haskell = rtsTrue;
1916 errno = saved_errno;
1918 SetLastError(saved_winerror);
1921 /* We might have GC'd, mark the TSO dirty again */
1924 IF_DEBUG(sanity, checkTSO(tso));
1929 /* ---------------------------------------------------------------------------
1932 * scheduleThread puts a thread on the end of the runnable queue.
1933 * This will usually be done immediately after a thread is created.
1934 * The caller of scheduleThread must create the thread using e.g.
1935 * createThread and push an appropriate closure
1936 * on this thread's stack before the scheduler is invoked.
1937 * ------------------------------------------------------------------------ */
1940 scheduleThread(Capability *cap, StgTSO *tso)
1942 // The thread goes at the *end* of the run-queue, to avoid possible
1943 // starvation of any threads already on the queue.
1944 appendToRunQueue(cap,tso);
1948 scheduleThreadOn(Capability *cap, StgWord cpu USED_IF_THREADS, StgTSO *tso)
1950 #if defined(THREADED_RTS)
1951 tso->flags |= TSO_LOCKED; // we requested explicit affinity; don't
1952 // move this thread from now on.
1953 cpu %= RtsFlags.ParFlags.nNodes;
1954 if (cpu == cap->no) {
1955 appendToRunQueue(cap,tso);
1957 wakeupThreadOnCapability(cap, &capabilities[cpu], tso);
1960 appendToRunQueue(cap,tso);
1965 scheduleWaitThread (StgTSO* tso, /*[out]*/HaskellObj* ret, Capability *cap)
1969 // We already created/initialised the Task
1970 task = cap->running_task;
1972 // This TSO is now a bound thread; make the Task and TSO
1973 // point to each other.
1979 task->stat = NoStatus;
1981 appendToRunQueue(cap,tso);
1983 debugTrace(DEBUG_sched, "new bound thread (%lu)", (unsigned long)tso->id);
1985 cap = schedule(cap,task);
1987 ASSERT(task->stat != NoStatus);
1988 ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
1990 debugTrace(DEBUG_sched, "bound thread (%lu) finished", (unsigned long)task->tso->id);
1994 /* ----------------------------------------------------------------------------
1996 * ------------------------------------------------------------------------- */
1998 #if defined(THREADED_RTS)
1999 void OSThreadProcAttr
2000 workerStart(Task *task)
2004 // See startWorkerTask().
2005 ACQUIRE_LOCK(&task->lock);
2007 RELEASE_LOCK(&task->lock);
2009 // set the thread-local pointer to the Task:
2012 // schedule() runs without a lock.
2013 cap = schedule(cap,task);
2015 // On exit from schedule(), we have a Capability.
2016 releaseCapability(cap);
2017 workerTaskStop(task);
2021 /* ---------------------------------------------------------------------------
2024 * Initialise the scheduler. This resets all the queues - if the
2025 * queues contained any threads, they'll be garbage collected at the
2028 * ------------------------------------------------------------------------ */
2033 #if !defined(THREADED_RTS)
2034 blocked_queue_hd = END_TSO_QUEUE;
2035 blocked_queue_tl = END_TSO_QUEUE;
2036 sleeping_queue = END_TSO_QUEUE;
2039 blackhole_queue = END_TSO_QUEUE;
2041 sched_state = SCHED_RUNNING;
2042 recent_activity = ACTIVITY_YES;
2044 #if defined(THREADED_RTS)
2045 /* Initialise the mutex and condition variables used by
2047 initMutex(&sched_mutex);
2050 ACQUIRE_LOCK(&sched_mutex);
2052 /* A capability holds the state a native thread needs in
2053 * order to execute STG code. At least one capability is
2054 * floating around (only THREADED_RTS builds have more than one).
2060 #if defined(THREADED_RTS) || defined(PARALLEL_HASKELL)
2064 #if defined(THREADED_RTS)
2066 * Eagerly start one worker to run each Capability, except for
2067 * Capability 0. The idea is that we're probably going to start a
2068 * bound thread on Capability 0 pretty soon, so we don't want a
2069 * worker task hogging it.
2074 for (i = 1; i < n_capabilities; i++) {
2075 cap = &capabilities[i];
2076 ACQUIRE_LOCK(&cap->lock);
2077 startWorkerTask(cap, workerStart);
2078 RELEASE_LOCK(&cap->lock);
2083 trace(TRACE_sched, "start: %d capabilities", n_capabilities);
2085 RELEASE_LOCK(&sched_mutex);
2090 rtsBool wait_foreign
2091 #if !defined(THREADED_RTS)
2092 __attribute__((unused))
2095 /* see Capability.c, shutdownCapability() */
2099 #if defined(THREADED_RTS)
2100 ACQUIRE_LOCK(&sched_mutex);
2101 task = newBoundTask();
2102 RELEASE_LOCK(&sched_mutex);
2105 // If we haven't killed all the threads yet, do it now.
2106 if (sched_state < SCHED_SHUTTING_DOWN) {
2107 sched_state = SCHED_INTERRUPTING;
2108 scheduleDoGC(NULL,task,rtsFalse);
2110 sched_state = SCHED_SHUTTING_DOWN;
2112 #if defined(THREADED_RTS)
2116 for (i = 0; i < n_capabilities; i++) {
2117 shutdownCapability(&capabilities[i], task, wait_foreign);
2119 boundTaskExiting(task);
2126 freeScheduler( void )
2130 if (n_capabilities != 1) {
2131 stgFree(capabilities);
2133 #if defined(THREADED_RTS)
2134 closeMutex(&sched_mutex);
2138 /* -----------------------------------------------------------------------------
2141 This is the interface to the garbage collector from Haskell land.
2142 We provide this so that external C code can allocate and garbage
2143 collect when called from Haskell via _ccall_GC.
2144 -------------------------------------------------------------------------- */
2147 performGC_(rtsBool force_major)
2150 // We must grab a new Task here, because the existing Task may be
2151 // associated with a particular Capability, and chained onto the
2152 // suspended_ccalling_tasks queue.
2153 ACQUIRE_LOCK(&sched_mutex);
2154 task = newBoundTask();
2155 RELEASE_LOCK(&sched_mutex);
2156 scheduleDoGC(NULL,task,force_major);
2157 boundTaskExiting(task);
2163 performGC_(rtsFalse);
2167 performMajorGC(void)
2169 performGC_(rtsTrue);
2172 /* -----------------------------------------------------------------------------
2175 If the thread has reached its maximum stack size, then raise the
2176 StackOverflow exception in the offending thread. Otherwise
2177 relocate the TSO into a larger chunk of memory and adjust its stack
2179 -------------------------------------------------------------------------- */
2182 threadStackOverflow(Capability *cap, StgTSO *tso)
2184 nat new_stack_size, stack_words;
2189 IF_DEBUG(sanity,checkTSO(tso));
2191 // don't allow throwTo() to modify the blocked_exceptions queue
2192 // while we are moving the TSO:
2193 lockClosure((StgClosure *)tso);
2195 if (tso->stack_size >= tso->max_stack_size && !(tso->flags & TSO_BLOCKEX)) {
2196 // NB. never raise a StackOverflow exception if the thread is
2197 // inside Control.Exceptino.block. It is impractical to protect
2198 // against stack overflow exceptions, since virtually anything
2199 // can raise one (even 'catch'), so this is the only sensible
2200 // thing to do here. See bug #767.
2202 debugTrace(DEBUG_gc,
2203 "threadStackOverflow of TSO %ld (%p): stack too large (now %ld; max is %ld)",
2204 (long)tso->id, tso, (long)tso->stack_size, (long)tso->max_stack_size);
2206 /* If we're debugging, just print out the top of the stack */
2207 printStackChunk(tso->sp, stg_min(tso->stack+tso->stack_size,
2210 // Send this thread the StackOverflow exception
2212 throwToSingleThreaded(cap, tso, (StgClosure *)stackOverflow_closure);
2216 /* Try to double the current stack size. If that takes us over the
2217 * maximum stack size for this thread, then use the maximum instead
2218 * (that is, unless we're already at or over the max size and we
2219 * can't raise the StackOverflow exception (see above), in which
2220 * case just double the size). Finally round up so the TSO ends up as
2221 * a whole number of blocks.
2223 if (tso->stack_size >= tso->max_stack_size) {
2224 new_stack_size = tso->stack_size * 2;
2226 new_stack_size = stg_min(tso->stack_size * 2, tso->max_stack_size);
2228 new_tso_size = (lnat)BLOCK_ROUND_UP(new_stack_size * sizeof(W_) +
2229 TSO_STRUCT_SIZE)/sizeof(W_);
2230 new_tso_size = round_to_mblocks(new_tso_size); /* Be MBLOCK-friendly */
2231 new_stack_size = new_tso_size - TSO_STRUCT_SIZEW;
2233 debugTrace(DEBUG_sched,
2234 "increasing stack size from %ld words to %d.",
2235 (long)tso->stack_size, new_stack_size);
2237 dest = (StgTSO *)allocateLocal(cap,new_tso_size);
2238 TICK_ALLOC_TSO(new_stack_size,0);
2240 /* copy the TSO block and the old stack into the new area */
2241 memcpy(dest,tso,TSO_STRUCT_SIZE);
2242 stack_words = tso->stack + tso->stack_size - tso->sp;
2243 new_sp = (P_)dest + new_tso_size - stack_words;
2244 memcpy(new_sp, tso->sp, stack_words * sizeof(W_));
2246 /* relocate the stack pointers... */
2248 dest->stack_size = new_stack_size;
2250 /* Mark the old TSO as relocated. We have to check for relocated
2251 * TSOs in the garbage collector and any primops that deal with TSOs.
2253 * It's important to set the sp value to just beyond the end
2254 * of the stack, so we don't attempt to scavenge any part of the
2257 tso->what_next = ThreadRelocated;
2258 setTSOLink(cap,tso,dest);
2259 tso->sp = (P_)&(tso->stack[tso->stack_size]);
2260 tso->why_blocked = NotBlocked;
2262 IF_PAR_DEBUG(verbose,
2263 debugBelch("@@ threadStackOverflow of TSO %d (now at %p): stack size increased to %ld\n",
2264 tso->id, tso, tso->stack_size);
2265 /* If we're debugging, just print out the top of the stack */
2266 printStackChunk(tso->sp, stg_min(tso->stack+tso->stack_size,
2272 IF_DEBUG(sanity,checkTSO(dest));
2274 IF_DEBUG(scheduler,printTSO(dest));
2281 threadStackUnderflow (Task *task STG_UNUSED, StgTSO *tso)
2283 bdescr *bd, *new_bd;
2284 lnat free_w, tso_size_w;
2287 tso_size_w = tso_sizeW(tso);
2289 if (tso_size_w < MBLOCK_SIZE_W ||
2290 (nat)(tso->stack + tso->stack_size - tso->sp) > tso->stack_size / 4)
2295 // don't allow throwTo() to modify the blocked_exceptions queue
2296 // while we are moving the TSO:
2297 lockClosure((StgClosure *)tso);
2299 // this is the number of words we'll free
2300 free_w = round_to_mblocks(tso_size_w/2);
2302 bd = Bdescr((StgPtr)tso);
2303 new_bd = splitLargeBlock(bd, free_w / BLOCK_SIZE_W);
2304 bd->free = bd->start + TSO_STRUCT_SIZEW;
2306 new_tso = (StgTSO *)new_bd->start;
2307 memcpy(new_tso,tso,TSO_STRUCT_SIZE);
2308 new_tso->stack_size = new_bd->free - new_tso->stack;
2310 debugTrace(DEBUG_sched, "thread %ld: reducing TSO size from %lu words to %lu",
2311 (long)tso->id, tso_size_w, tso_sizeW(new_tso));
2313 tso->what_next = ThreadRelocated;
2314 tso->_link = new_tso; // no write barrier reqd: same generation
2316 // The TSO attached to this Task may have moved, so update the
2318 if (task->tso == tso) {
2319 task->tso = new_tso;
2325 IF_DEBUG(sanity,checkTSO(new_tso));
2330 /* ---------------------------------------------------------------------------
2332 - usually called inside a signal handler so it mustn't do anything fancy.
2333 ------------------------------------------------------------------------ */
2336 interruptStgRts(void)
2338 sched_state = SCHED_INTERRUPTING;
2339 setContextSwitches();
2343 /* -----------------------------------------------------------------------------
2346 This function causes at least one OS thread to wake up and run the
2347 scheduler loop. It is invoked when the RTS might be deadlocked, or
2348 an external event has arrived that may need servicing (eg. a
2349 keyboard interrupt).
2351 In the single-threaded RTS we don't do anything here; we only have
2352 one thread anyway, and the event that caused us to want to wake up
2353 will have interrupted any blocking system call in progress anyway.
2354 -------------------------------------------------------------------------- */
2359 #if defined(THREADED_RTS)
2360 // This forces the IO Manager thread to wakeup, which will
2361 // in turn ensure that some OS thread wakes up and runs the
2362 // scheduler loop, which will cause a GC and deadlock check.
2367 /* -----------------------------------------------------------------------------
2370 * Check the blackhole_queue for threads that can be woken up. We do
2371 * this periodically: before every GC, and whenever the run queue is
2374 * An elegant solution might be to just wake up all the blocked
2375 * threads with awakenBlockedQueue occasionally: they'll go back to
2376 * sleep again if the object is still a BLACKHOLE. Unfortunately this
2377 * doesn't give us a way to tell whether we've actually managed to
2378 * wake up any threads, so we would be busy-waiting.
2380 * -------------------------------------------------------------------------- */
2383 checkBlackHoles (Capability *cap)
2386 rtsBool any_woke_up = rtsFalse;
2389 // blackhole_queue is global:
2390 ASSERT_LOCK_HELD(&sched_mutex);
2392 debugTrace(DEBUG_sched, "checking threads blocked on black holes");
2394 // ASSUMES: sched_mutex
2395 prev = &blackhole_queue;
2396 t = blackhole_queue;
2397 while (t != END_TSO_QUEUE) {
2398 ASSERT(t->why_blocked == BlockedOnBlackHole);
2399 type = get_itbl(UNTAG_CLOSURE(t->block_info.closure))->type;
2400 if (type != BLACKHOLE && type != CAF_BLACKHOLE) {
2401 IF_DEBUG(sanity,checkTSO(t));
2402 t = unblockOne(cap, t);
2404 any_woke_up = rtsTrue;
2414 /* -----------------------------------------------------------------------------
2417 This is used for interruption (^C) and forking, and corresponds to
2418 raising an exception but without letting the thread catch the
2420 -------------------------------------------------------------------------- */
2423 deleteThread (Capability *cap, StgTSO *tso)
2425 // NOTE: must only be called on a TSO that we have exclusive
2426 // access to, because we will call throwToSingleThreaded() below.
2427 // The TSO must be on the run queue of the Capability we own, or
2428 // we must own all Capabilities.
2430 if (tso->why_blocked != BlockedOnCCall &&
2431 tso->why_blocked != BlockedOnCCall_NoUnblockExc) {
2432 throwToSingleThreaded(cap,tso,NULL);
2436 #ifdef FORKPROCESS_PRIMOP_SUPPORTED
2438 deleteThread_(Capability *cap, StgTSO *tso)
2439 { // for forkProcess only:
2440 // like deleteThread(), but we delete threads in foreign calls, too.
2442 if (tso->why_blocked == BlockedOnCCall ||
2443 tso->why_blocked == BlockedOnCCall_NoUnblockExc) {
2444 unblockOne(cap,tso);
2445 tso->what_next = ThreadKilled;
2447 deleteThread(cap,tso);
2452 /* -----------------------------------------------------------------------------
2453 raiseExceptionHelper
2455 This function is called by the raise# primitve, just so that we can
2456 move some of the tricky bits of raising an exception from C-- into
2457 C. Who knows, it might be a useful re-useable thing here too.
2458 -------------------------------------------------------------------------- */
2461 raiseExceptionHelper (StgRegTable *reg, StgTSO *tso, StgClosure *exception)
2463 Capability *cap = regTableToCapability(reg);
2464 StgThunk *raise_closure = NULL;
2466 StgRetInfoTable *info;
2468 // This closure represents the expression 'raise# E' where E
2469 // is the exception raise. It is used to overwrite all the
2470 // thunks which are currently under evaluataion.
2473 // OLD COMMENT (we don't have MIN_UPD_SIZE now):
2474 // LDV profiling: stg_raise_info has THUNK as its closure
2475 // type. Since a THUNK takes at least MIN_UPD_SIZE words in its
2476 // payload, MIN_UPD_SIZE is more approprate than 1. It seems that
2477 // 1 does not cause any problem unless profiling is performed.
2478 // However, when LDV profiling goes on, we need to linearly scan
2479 // small object pool, where raise_closure is stored, so we should
2480 // use MIN_UPD_SIZE.
2482 // raise_closure = (StgClosure *)RET_STGCALL1(P_,allocate,
2483 // sizeofW(StgClosure)+1);
2487 // Walk up the stack, looking for the catch frame. On the way,
2488 // we update any closures pointed to from update frames with the
2489 // raise closure that we just built.
2493 info = get_ret_itbl((StgClosure *)p);
2494 next = p + stack_frame_sizeW((StgClosure *)p);
2495 switch (info->i.type) {
2498 // Only create raise_closure if we need to.
2499 if (raise_closure == NULL) {
2501 (StgThunk *)allocateLocal(cap,sizeofW(StgThunk)+1);
2502 SET_HDR(raise_closure, &stg_raise_info, CCCS);
2503 raise_closure->payload[0] = exception;
2505 UPD_IND(((StgUpdateFrame *)p)->updatee,(StgClosure *)raise_closure);
2509 case ATOMICALLY_FRAME:
2510 debugTrace(DEBUG_stm, "found ATOMICALLY_FRAME at %p", p);
2512 return ATOMICALLY_FRAME;
2518 case CATCH_STM_FRAME:
2519 debugTrace(DEBUG_stm, "found CATCH_STM_FRAME at %p", p);
2521 return CATCH_STM_FRAME;
2527 case CATCH_RETRY_FRAME:
2536 /* -----------------------------------------------------------------------------
2537 findRetryFrameHelper
2539 This function is called by the retry# primitive. It traverses the stack
2540 leaving tso->sp referring to the frame which should handle the retry.
2542 This should either be a CATCH_RETRY_FRAME (if the retry# is within an orElse#)
2543 or should be a ATOMICALLY_FRAME (if the retry# reaches the top level).
2545 We skip CATCH_STM_FRAMEs (aborting and rolling back the nested tx that they
2546 create) because retries are not considered to be exceptions, despite the
2547 similar implementation.
2549 We should not expect to see CATCH_FRAME or STOP_FRAME because those should
2550 not be created within memory transactions.
2551 -------------------------------------------------------------------------- */
2554 findRetryFrameHelper (StgTSO *tso)
2557 StgRetInfoTable *info;
2561 info = get_ret_itbl((StgClosure *)p);
2562 next = p + stack_frame_sizeW((StgClosure *)p);
2563 switch (info->i.type) {
2565 case ATOMICALLY_FRAME:
2566 debugTrace(DEBUG_stm,
2567 "found ATOMICALLY_FRAME at %p during retry", p);
2569 return ATOMICALLY_FRAME;
2571 case CATCH_RETRY_FRAME:
2572 debugTrace(DEBUG_stm,
2573 "found CATCH_RETRY_FRAME at %p during retrry", p);
2575 return CATCH_RETRY_FRAME;
2577 case CATCH_STM_FRAME: {
2578 StgTRecHeader *trec = tso -> trec;
2579 StgTRecHeader *outer = stmGetEnclosingTRec(trec);
2580 debugTrace(DEBUG_stm,
2581 "found CATCH_STM_FRAME at %p during retry", p);
2582 debugTrace(DEBUG_stm, "trec=%p outer=%p", trec, outer);
2583 stmAbortTransaction(tso -> cap, trec);
2584 stmFreeAbortedTRec(tso -> cap, trec);
2585 tso -> trec = outer;
2592 ASSERT(info->i.type != CATCH_FRAME);
2593 ASSERT(info->i.type != STOP_FRAME);
2600 /* -----------------------------------------------------------------------------
2601 resurrectThreads is called after garbage collection on the list of
2602 threads found to be garbage. Each of these threads will be woken
2603 up and sent a signal: BlockedOnDeadMVar if the thread was blocked
2604 on an MVar, or NonTermination if the thread was blocked on a Black
2607 Locks: assumes we hold *all* the capabilities.
2608 -------------------------------------------------------------------------- */
2611 resurrectThreads (StgTSO *threads)
2617 for (tso = threads; tso != END_TSO_QUEUE; tso = next) {
2618 next = tso->global_link;
2620 step = Bdescr((P_)tso)->step;
2621 tso->global_link = step->threads;
2622 step->threads = tso;
2624 debugTrace(DEBUG_sched, "resurrecting thread %lu", (unsigned long)tso->id);
2626 // Wake up the thread on the Capability it was last on
2629 switch (tso->why_blocked) {
2631 case BlockedOnException:
2632 /* Called by GC - sched_mutex lock is currently held. */
2633 throwToSingleThreaded(cap, tso,
2634 (StgClosure *)blockedOnDeadMVar_closure);
2636 case BlockedOnBlackHole:
2637 throwToSingleThreaded(cap, tso,
2638 (StgClosure *)nonTermination_closure);
2641 throwToSingleThreaded(cap, tso,
2642 (StgClosure *)blockedIndefinitely_closure);
2645 /* This might happen if the thread was blocked on a black hole
2646 * belonging to a thread that we've just woken up (raiseAsync
2647 * can wake up threads, remember...).
2651 barf("resurrectThreads: thread blocked in a strange way");
2656 /* -----------------------------------------------------------------------------
2657 performPendingThrowTos is called after garbage collection, and
2658 passed a list of threads that were found to have pending throwTos
2659 (tso->blocked_exceptions was not empty), and were blocked.
2660 Normally this doesn't happen, because we would deliver the
2661 exception directly if the target thread is blocked, but there are
2662 small windows where it might occur on a multiprocessor (see
2665 NB. we must be holding all the capabilities at this point, just
2666 like resurrectThreads().
2667 -------------------------------------------------------------------------- */
2670 performPendingThrowTos (StgTSO *threads)
2676 for (tso = threads; tso != END_TSO_QUEUE; tso = next) {
2677 next = tso->global_link;
2679 step = Bdescr((P_)tso)->step;
2680 tso->global_link = step->threads;
2681 step->threads = tso;
2683 debugTrace(DEBUG_sched, "performing blocked throwTo to thread %lu", (unsigned long)tso->id);
2686 maybePerformBlockedException(cap, tso);