1 /* ---------------------------------------------------------------------------
2 * $Id: Schedule.c,v 1.169 2003/05/14 09:11:49 simonmar Exp $
4 * (c) The GHC Team, 1998-2000
8 * Different GHC ways use this scheduler quite differently (see comments below)
9 * Here is the global picture:
11 * WAY Name CPP flag What's it for
12 * --------------------------------------
13 * mp GUM PAR Parallel execution on a distributed memory machine
14 * s SMP SMP Parallel execution on a shared memory machine
15 * mg GranSim GRAN Simulation of parallel execution
16 * md GUM/GdH DIST Distributed execution (based on GUM)
18 * --------------------------------------------------------------------------*/
20 //@node Main scheduling code, , ,
21 //@section Main scheduling code
24 * Version with scheduler monitor support for SMPs (WAY=s):
26 This design provides a high-level API to create and schedule threads etc.
27 as documented in the SMP design document.
29 It uses a monitor design controlled by a single mutex to exercise control
30 over accesses to shared data structures, and builds on the Posix threads
33 The majority of state is shared. In order to keep essential per-task state,
34 there is a Capability structure, which contains all the information
35 needed to run a thread: its STG registers, a pointer to its TSO, a
36 nursery etc. During STG execution, a pointer to the capability is
37 kept in a register (BaseReg).
39 In a non-SMP build, there is one global capability, namely MainRegTable.
43 * Version with support for distributed memory parallelism aka GUM (WAY=mp):
45 The main scheduling loop in GUM iterates until a finish message is received.
46 In that case a global flag @receivedFinish@ is set and this instance of
47 the RTS shuts down. See ghc/rts/parallel/HLComms.c:processMessages()
48 for the handling of incoming messages, such as PP_FINISH.
49 Note that in the parallel case we have a system manager that coordinates
50 different PEs, each of which are running one instance of the RTS.
51 See ghc/rts/parallel/SysMan.c for the main routine of the parallel program.
52 From this routine processes executing ghc/rts/Main.c are spawned. -- HWL
54 * Version with support for simulating parallel execution aka GranSim (WAY=mg):
56 The main scheduling code in GranSim is quite different from that in std
57 (concurrent) Haskell: while concurrent Haskell just iterates over the
58 threads in the runnable queue, GranSim is event driven, i.e. it iterates
59 over the events in the global event queue. -- HWL
64 //* Variables and Data structures::
65 //* Main scheduling loop::
66 //* Suspend and Resume::
68 //* Garbage Collextion Routines::
69 //* Blocking Queue Routines::
70 //* Exception Handling Routines::
71 //* Debugging Routines::
75 //@node Includes, Variables and Data structures, Main scheduling code, Main scheduling code
76 //@subsection Includes
78 #include "PosixSource.h"
85 #include "StgStartup.h"
87 #define COMPILING_SCHEDULER
89 #include "StgMiscClosures.h"
91 #include "Interpreter.h"
92 #include "Exception.h"
99 #include "ThreadLabels.h"
101 #include "Proftimer.h"
102 #include "ProfHeap.h"
104 #if defined(GRAN) || defined(PAR)
105 # include "GranSimRts.h"
106 # include "GranSim.h"
107 # include "ParallelRts.h"
108 # include "Parallel.h"
109 # include "ParallelDebug.h"
110 # include "FetchMe.h"
114 #include "Capability.h"
115 #include "OSThreads.h"
118 #ifdef HAVE_SYS_TYPES_H
119 #include <sys/types.h>
129 //@node Variables and Data structures, Prototypes, Includes, Main scheduling code
130 //@subsection Variables and Data structures
132 /* Main thread queue.
133 * Locks required: sched_mutex.
135 StgMainThread *main_threads = NULL;
138 // Pointer to the thread that executes main
139 // When this thread is finished, the program terminates
140 // by calling shutdownHaskellAndExit.
141 // It would be better to add a call to shutdownHaskellAndExit
142 // to the Main.main wrapper and to remove this hack.
143 StgMainThread *main_main_thread = NULL;
147 * Locks required: sched_mutex.
151 StgTSO* ActiveTSO = NULL; /* for assigning system costs; GranSim-Light only */
152 /* rtsTime TimeOfNextEvent, EndOfTimeSlice; now in GranSim.c */
155 In GranSim we have a runnable and a blocked queue for each processor.
156 In order to minimise code changes new arrays run_queue_hds/tls
157 are created. run_queue_hd is then a short cut (macro) for
158 run_queue_hds[CurrentProc] (see GranSim.h).
161 StgTSO *run_queue_hds[MAX_PROC], *run_queue_tls[MAX_PROC];
162 StgTSO *blocked_queue_hds[MAX_PROC], *blocked_queue_tls[MAX_PROC];
163 StgTSO *ccalling_threadss[MAX_PROC];
164 /* We use the same global list of threads (all_threads) in GranSim as in
165 the std RTS (i.e. we are cheating). However, we don't use this list in
166 the GranSim specific code at the moment (so we are only potentially
171 StgTSO *run_queue_hd = NULL;
172 StgTSO *run_queue_tl = NULL;
173 StgTSO *blocked_queue_hd = NULL;
174 StgTSO *blocked_queue_tl = NULL;
175 StgTSO *sleeping_queue = NULL; /* perhaps replace with a hash table? */
179 /* Linked list of all threads.
180 * Used for detecting garbage collected threads.
182 StgTSO *all_threads = NULL;
184 /* When a thread performs a safe C call (_ccall_GC, using old
185 * terminology), it gets put on the suspended_ccalling_threads
186 * list. Used by the garbage collector.
188 static StgTSO *suspended_ccalling_threads;
190 static StgTSO *threadStackOverflow(StgTSO *tso);
192 /* KH: The following two flags are shared memory locations. There is no need
193 to lock them, since they are only unset at the end of a scheduler
197 /* flag set by signal handler to precipitate a context switch */
198 //@cindex context_switch
199 nat context_switch = 0;
201 /* if this flag is set as well, give up execution */
202 //@cindex interrupted
203 rtsBool interrupted = rtsFalse;
205 /* Next thread ID to allocate.
206 * Locks required: thread_id_mutex
208 //@cindex next_thread_id
209 static StgThreadID next_thread_id = 1;
212 * Pointers to the state of the current thread.
213 * Rule of thumb: if CurrentTSO != NULL, then we're running a Haskell
214 * thread. If CurrentTSO == NULL, then we're at the scheduler level.
217 /* The smallest stack size that makes any sense is:
218 * RESERVED_STACK_WORDS (so we can get back from the stack overflow)
219 * + sizeofW(StgStopFrame) (the stg_stop_thread_info frame)
220 * + 1 (the closure to enter)
222 * + 1 (spare slot req'd by stg_ap_v_ret)
224 * A thread with this stack will bomb immediately with a stack
225 * overflow, which will increase its stack size.
228 #define MIN_STACK_WORDS (RESERVED_STACK_WORDS + sizeofW(StgStopFrame) + 3)
235 /* This is used in `TSO.h' and gcc 2.96 insists that this variable actually
236 * exists - earlier gccs apparently didn't.
241 static rtsBool ready_to_gc;
244 * Set to TRUE when entering a shutdown state (via shutdownHaskellAndExit()) --
245 * in an MT setting, needed to signal that a worker thread shouldn't hang around
246 * in the scheduler when it is out of work.
248 static rtsBool shutting_down_scheduler = rtsFalse;
250 void addToBlockedQueue ( StgTSO *tso );
252 static void schedule ( void );
253 void interruptStgRts ( void );
255 static void detectBlackHoles ( void );
258 static void sched_belch(char *s, ...);
261 #if defined(RTS_SUPPORTS_THREADS)
262 /* ToDo: carefully document the invariants that go together
263 * with these synchronisation objects.
265 Mutex sched_mutex = INIT_MUTEX_VAR;
266 Mutex term_mutex = INIT_MUTEX_VAR;
269 * A heavyweight solution to the problem of protecting
270 * the thread_id from concurrent update.
272 Mutex thread_id_mutex = INIT_MUTEX_VAR;
276 static Condition gc_pending_cond = INIT_COND_VAR;
280 #endif /* RTS_SUPPORTS_THREADS */
284 rtsTime TimeOfLastYield;
285 rtsBool emitSchedule = rtsTrue;
289 static char *whatNext_strs[] = {
299 StgTSO * createSparkThread(rtsSpark spark);
300 StgTSO * activateSpark (rtsSpark spark);
304 * The thread state for the main thread.
305 // ToDo: check whether not needed any more
309 #if defined(PAR) || defined(RTS_SUPPORTS_THREADS)
310 static void taskStart(void);
318 #if defined(RTS_SUPPORTS_THREADS)
320 startSchedulerTask(void)
322 startTask(taskStart);
326 //@node Main scheduling loop, Suspend and Resume, Prototypes, Main scheduling code
327 //@subsection Main scheduling loop
329 /* ---------------------------------------------------------------------------
330 Main scheduling loop.
332 We use round-robin scheduling, each thread returning to the
333 scheduler loop when one of these conditions is detected:
336 * timer expires (thread yields)
341 Locking notes: we acquire the scheduler lock once at the beginning
342 of the scheduler loop, and release it when
344 * running a thread, or
345 * waiting for work, or
346 * waiting for a GC to complete.
349 In a GranSim setup this loop iterates over the global event queue.
350 This revolves around the global event queue, which determines what
351 to do next. Therefore, it's more complicated than either the
352 concurrent or the parallel (GUM) setup.
355 GUM iterates over incoming messages.
356 It starts with nothing to do (thus CurrentTSO == END_TSO_QUEUE),
357 and sends out a fish whenever it has nothing to do; in-between
358 doing the actual reductions (shared code below) it processes the
359 incoming messages and deals with delayed operations
360 (see PendingFetches).
361 This is not the ugliest code you could imagine, but it's bloody close.
363 ------------------------------------------------------------------------ */
370 StgThreadReturnCode ret;
378 rtsBool receivedFinish = rtsFalse;
380 nat tp_size, sp_size; // stats only
383 rtsBool was_interrupted = rtsFalse;
384 StgTSOWhatNext prev_what_next;
386 ACQUIRE_LOCK(&sched_mutex);
388 #if defined(RTS_SUPPORTS_THREADS)
389 waitForWorkCapability(&sched_mutex, &cap, rtsFalse);
390 IF_DEBUG(scheduler, sched_belch("worker thread (osthread %p): entering RTS", osThreadId()));
392 /* simply initialise it in the non-threaded case */
393 grabCapability(&cap);
397 /* set up first event to get things going */
398 /* ToDo: assign costs for system setup and init MainTSO ! */
399 new_event(CurrentProc, CurrentProc, CurrentTime[CurrentProc],
401 CurrentTSO, (StgClosure*)NULL, (rtsSpark*)NULL);
404 fprintf(stderr, "GRAN: Init CurrentTSO (in schedule) = %p\n", CurrentTSO);
405 G_TSO(CurrentTSO, 5));
407 if (RtsFlags.GranFlags.Light) {
408 /* Save current time; GranSim Light only */
409 CurrentTSO->gran.clock = CurrentTime[CurrentProc];
412 event = get_next_event();
414 while (event!=(rtsEvent*)NULL) {
415 /* Choose the processor with the next event */
416 CurrentProc = event->proc;
417 CurrentTSO = event->tso;
421 while (!receivedFinish) { /* set by processMessages */
422 /* when receiving PP_FINISH message */
429 IF_DEBUG(scheduler, printAllThreads());
431 #if defined(RTS_SUPPORTS_THREADS)
432 /* Check to see whether there are any worker threads
433 waiting to deposit external call results. If so,
434 yield our capability */
435 yieldToReturningWorker(&sched_mutex, &cap);
438 /* If we're interrupted (the user pressed ^C, or some other
439 * termination condition occurred), kill all the currently running
443 IF_DEBUG(scheduler, sched_belch("interrupted"));
444 interrupted = rtsFalse;
445 was_interrupted = rtsTrue;
446 #if defined(RTS_SUPPORTS_THREADS)
447 // In the threaded RTS, deadlock detection doesn't work,
448 // so just exit right away.
449 prog_belch("interrupted");
450 releaseCapability(cap);
451 startTask(taskStart); // thread-safe-call to shutdownHaskellAndExit
452 RELEASE_LOCK(&sched_mutex);
453 shutdownHaskellAndExit(EXIT_SUCCESS);
459 /* Go through the list of main threads and wake up any
460 * clients whose computations have finished. ToDo: this
461 * should be done more efficiently without a linear scan
462 * of the main threads list, somehow...
464 #if defined(RTS_SUPPORTS_THREADS)
466 StgMainThread *m, **prev;
467 prev = &main_threads;
468 for (m = main_threads; m != NULL; prev = &m->link, m = m->link) {
469 switch (m->tso->what_next) {
472 // NOTE: return val is tso->sp[1] (see StgStartup.hc)
473 *(m->ret) = (StgClosure *)m->tso->sp[1];
477 broadcastCondition(&m->wakeup);
479 removeThreadLabel((StgWord)m->tso);
481 if(m == main_main_thread)
483 releaseCapability(cap);
484 startTask(taskStart); // thread-safe-call to shutdownHaskellAndExit
485 RELEASE_LOCK(&sched_mutex);
486 shutdownHaskellAndExit(EXIT_SUCCESS);
490 if (m->ret) *(m->ret) = NULL;
492 if (was_interrupted) {
493 m->stat = Interrupted;
497 broadcastCondition(&m->wakeup);
499 removeThreadLabel((StgWord)m->tso);
501 if(m == main_main_thread)
503 releaseCapability(cap);
504 startTask(taskStart); // thread-safe-call to shutdownHaskellAndExit
505 RELEASE_LOCK(&sched_mutex);
506 shutdownHaskellAndExit(EXIT_SUCCESS);
515 #else /* not threaded */
518 /* in GUM do this only on the Main PE */
521 /* If our main thread has finished or been killed, return.
524 StgMainThread *m = main_threads;
525 if (m->tso->what_next == ThreadComplete
526 || m->tso->what_next == ThreadKilled) {
528 removeThreadLabel((StgWord)m->tso);
530 main_threads = main_threads->link;
531 if (m->tso->what_next == ThreadComplete) {
532 // We finished successfully, fill in the return value
533 // NOTE: return val is tso->sp[1] (see StgStartup.hc)
534 if (m->ret) { *(m->ret) = (StgClosure *)m->tso->sp[1]; };
538 if (m->ret) { *(m->ret) = NULL; };
539 if (was_interrupted) {
540 m->stat = Interrupted;
550 /* Top up the run queue from our spark pool. We try to make the
551 * number of threads in the run queue equal to the number of
554 * Disable spark support in SMP for now, non-essential & requires
555 * a little bit of work to make it compile cleanly. -- sof 1/02.
557 #if 0 /* defined(SMP) */
559 nat n = getFreeCapabilities();
560 StgTSO *tso = run_queue_hd;
562 /* Count the run queue */
563 while (n > 0 && tso != END_TSO_QUEUE) {
570 spark = findSpark(rtsFalse);
572 break; /* no more sparks in the pool */
574 /* I'd prefer this to be done in activateSpark -- HWL */
575 /* tricky - it needs to hold the scheduler lock and
576 * not try to re-acquire it -- SDM */
577 createSparkThread(spark);
579 sched_belch("==^^ turning spark of closure %p into a thread",
580 (StgClosure *)spark));
583 /* We need to wake up the other tasks if we just created some
586 if (getFreeCapabilities() - n > 1) {
587 signalCondition( &thread_ready_cond );
592 /* check for signals each time around the scheduler */
593 #if defined(RTS_USER_SIGNALS)
594 if (signals_pending()) {
595 RELEASE_LOCK(&sched_mutex); /* ToDo: kill */
596 startSignalHandlers();
597 ACQUIRE_LOCK(&sched_mutex);
601 /* Check whether any waiting threads need to be woken up. If the
602 * run queue is empty, and there are no other tasks running, we
603 * can wait indefinitely for something to happen.
605 if ( !EMPTY_QUEUE(blocked_queue_hd) || !EMPTY_QUEUE(sleeping_queue)
606 #if defined(RTS_SUPPORTS_THREADS) && !defined(SMP)
611 awaitEvent( EMPTY_RUN_QUEUE()
613 && allFreeCapabilities()
617 /* we can be interrupted while waiting for I/O... */
618 if (interrupted) continue;
621 * Detect deadlock: when we have no threads to run, there are no
622 * threads waiting on I/O or sleeping, and all the other tasks are
623 * waiting for work, we must have a deadlock of some description.
625 * We first try to find threads blocked on themselves (ie. black
626 * holes), and generate NonTermination exceptions where necessary.
628 * If no threads are black holed, we have a deadlock situation, so
629 * inform all the main threads.
631 #if !defined(PAR) && !defined(RTS_SUPPORTS_THREADS)
632 if ( EMPTY_THREAD_QUEUES()
633 #if defined(RTS_SUPPORTS_THREADS)
634 && EMPTY_QUEUE(suspended_ccalling_threads)
637 && allFreeCapabilities()
641 IF_DEBUG(scheduler, sched_belch("deadlocked, forcing major GC..."));
642 #if defined(THREADED_RTS)
643 /* and SMP mode ..? */
644 releaseCapability(cap);
646 // Garbage collection can release some new threads due to
647 // either (a) finalizers or (b) threads resurrected because
648 // they are about to be send BlockedOnDeadMVar. Any threads
649 // thus released will be immediately runnable.
650 GarbageCollect(GetRoots,rtsTrue);
652 if ( !EMPTY_RUN_QUEUE() ) { goto not_deadlocked; }
655 sched_belch("still deadlocked, checking for black holes..."));
658 if ( !EMPTY_RUN_QUEUE() ) { goto not_deadlocked; }
660 #if defined(RTS_USER_SIGNALS)
661 /* If we have user-installed signal handlers, then wait
662 * for signals to arrive rather then bombing out with a
665 #if defined(RTS_SUPPORTS_THREADS)
666 if ( 0 ) { /* hmm..what to do? Simply stop waiting for
667 a signal with no runnable threads (or I/O
668 suspended ones) leads nowhere quick.
669 For now, simply shut down when we reach this
672 ToDo: define precisely under what conditions
673 the Scheduler should shut down in an MT setting.
676 if ( anyUserHandlers() ) {
679 sched_belch("still deadlocked, waiting for signals..."));
683 // we might be interrupted...
684 if (interrupted) { continue; }
686 if (signals_pending()) {
687 RELEASE_LOCK(&sched_mutex);
688 startSignalHandlers();
689 ACQUIRE_LOCK(&sched_mutex);
691 ASSERT(!EMPTY_RUN_QUEUE());
696 /* Probably a real deadlock. Send the current main thread the
697 * Deadlock exception (or in the SMP build, send *all* main
698 * threads the deadlock exception, since none of them can make
703 #if defined(RTS_SUPPORTS_THREADS)
704 for (m = main_threads; m != NULL; m = m->link) {
705 switch (m->tso->why_blocked) {
706 case BlockedOnBlackHole:
707 raiseAsync(m->tso, (StgClosure *)NonTermination_closure);
709 case BlockedOnException:
711 raiseAsync(m->tso, (StgClosure *)Deadlock_closure);
714 barf("deadlock: main thread blocked in a strange way");
719 switch (m->tso->why_blocked) {
720 case BlockedOnBlackHole:
721 raiseAsync(m->tso, (StgClosure *)NonTermination_closure);
723 case BlockedOnException:
725 raiseAsync(m->tso, (StgClosure *)Deadlock_closure);
728 barf("deadlock: main thread blocked in a strange way");
733 #if defined(RTS_SUPPORTS_THREADS)
734 /* ToDo: revisit conditions (and mechanism) for shutting
735 down a multi-threaded world */
736 IF_DEBUG(scheduler, sched_belch("all done, i think...shutting down."));
737 RELEASE_LOCK(&sched_mutex);
744 #elif defined(RTS_SUPPORTS_THREADS)
745 /* ToDo: add deadlock detection in threaded RTS */
747 /* ToDo: add deadlock detection in GUM (similar to SMP) -- HWL */
751 /* If there's a GC pending, don't do anything until it has
755 IF_DEBUG(scheduler,sched_belch("waiting for GC"));
756 waitCondition( &gc_pending_cond, &sched_mutex );
760 #if defined(RTS_SUPPORTS_THREADS)
762 /* block until we've got a thread on the run queue and a free
766 if ( EMPTY_RUN_QUEUE() ) {
767 /* Give up our capability */
768 releaseCapability(cap);
770 /* If we're in the process of shutting down (& running the
771 * a batch of finalisers), don't wait around.
773 if ( shutting_down_scheduler ) {
774 RELEASE_LOCK(&sched_mutex);
777 IF_DEBUG(scheduler, sched_belch("thread %d: waiting for work", osThreadId()));
778 waitForWorkCapability(&sched_mutex, &cap, rtsTrue);
779 IF_DEBUG(scheduler, sched_belch("thread %d: work now available", osThreadId()));
782 if ( EMPTY_RUN_QUEUE() ) {
783 continue; // nothing to do
789 if (RtsFlags.GranFlags.Light)
790 GranSimLight_enter_system(event, &ActiveTSO); // adjust ActiveTSO etc
792 /* adjust time based on time-stamp */
793 if (event->time > CurrentTime[CurrentProc] &&
794 event->evttype != ContinueThread)
795 CurrentTime[CurrentProc] = event->time;
797 /* Deal with the idle PEs (may issue FindWork or MoveSpark events) */
798 if (!RtsFlags.GranFlags.Light)
801 IF_DEBUG(gran, fprintf(stderr, "GRAN: switch by event-type\n"));
803 /* main event dispatcher in GranSim */
804 switch (event->evttype) {
805 /* Should just be continuing execution */
807 IF_DEBUG(gran, fprintf(stderr, "GRAN: doing ContinueThread\n"));
808 /* ToDo: check assertion
809 ASSERT(run_queue_hd != (StgTSO*)NULL &&
810 run_queue_hd != END_TSO_QUEUE);
812 /* Ignore ContinueThreads for fetching threads (if synchr comm) */
813 if (!RtsFlags.GranFlags.DoAsyncFetch &&
814 procStatus[CurrentProc]==Fetching) {
815 belch("ghuH: Spurious ContinueThread while Fetching ignored; TSO %d (%p) [PE %d]",
816 CurrentTSO->id, CurrentTSO, CurrentProc);
819 /* Ignore ContinueThreads for completed threads */
820 if (CurrentTSO->what_next == ThreadComplete) {
821 belch("ghuH: found a ContinueThread event for completed thread %d (%p) [PE %d] (ignoring ContinueThread)",
822 CurrentTSO->id, CurrentTSO, CurrentProc);
825 /* Ignore ContinueThreads for threads that are being migrated */
826 if (PROCS(CurrentTSO)==Nowhere) {
827 belch("ghuH: trying to run the migrating TSO %d (%p) [PE %d] (ignoring ContinueThread)",
828 CurrentTSO->id, CurrentTSO, CurrentProc);
831 /* The thread should be at the beginning of the run queue */
832 if (CurrentTSO!=run_queue_hds[CurrentProc]) {
833 belch("ghuH: TSO %d (%p) [PE %d] is not at the start of the run_queue when doing a ContinueThread",
834 CurrentTSO->id, CurrentTSO, CurrentProc);
835 break; // run the thread anyway
838 new_event(proc, proc, CurrentTime[proc],
840 (StgTSO*)NULL, (StgClosure*)NULL, (rtsSpark*)NULL);
842 */ /* Catches superfluous CONTINUEs -- should be unnecessary */
843 break; // now actually run the thread; DaH Qu'vam yImuHbej
846 do_the_fetchnode(event);
847 goto next_thread; /* handle next event in event queue */
850 do_the_globalblock(event);
851 goto next_thread; /* handle next event in event queue */
854 do_the_fetchreply(event);
855 goto next_thread; /* handle next event in event queue */
857 case UnblockThread: /* Move from the blocked queue to the tail of */
858 do_the_unblock(event);
859 goto next_thread; /* handle next event in event queue */
861 case ResumeThread: /* Move from the blocked queue to the tail of */
862 /* the runnable queue ( i.e. Qu' SImqa'lu') */
863 event->tso->gran.blocktime +=
864 CurrentTime[CurrentProc] - event->tso->gran.blockedat;
865 do_the_startthread(event);
866 goto next_thread; /* handle next event in event queue */
869 do_the_startthread(event);
870 goto next_thread; /* handle next event in event queue */
873 do_the_movethread(event);
874 goto next_thread; /* handle next event in event queue */
877 do_the_movespark(event);
878 goto next_thread; /* handle next event in event queue */
881 do_the_findwork(event);
882 goto next_thread; /* handle next event in event queue */
885 barf("Illegal event type %u\n", event->evttype);
888 /* This point was scheduler_loop in the old RTS */
890 IF_DEBUG(gran, belch("GRAN: after main switch"));
892 TimeOfLastEvent = CurrentTime[CurrentProc];
893 TimeOfNextEvent = get_time_of_next_event();
894 IgnoreEvents=(TimeOfNextEvent==0); // HWL HACK
895 // CurrentTSO = ThreadQueueHd;
897 IF_DEBUG(gran, belch("GRAN: time of next event is: %ld",
900 if (RtsFlags.GranFlags.Light)
901 GranSimLight_leave_system(event, &ActiveTSO);
903 EndOfTimeSlice = CurrentTime[CurrentProc]+RtsFlags.GranFlags.time_slice;
906 belch("GRAN: end of time-slice is %#lx", EndOfTimeSlice));
908 /* in a GranSim setup the TSO stays on the run queue */
910 /* Take a thread from the run queue. */
911 t = POP_RUN_QUEUE(); // take_off_run_queue(t);
914 fprintf(stderr, "GRAN: About to run current thread, which is\n");
917 context_switch = 0; // turned on via GranYield, checking events and time slice
920 DumpGranEvent(GR_SCHEDULE, t));
922 procStatus[CurrentProc] = Busy;
925 if (PendingFetches != END_BF_QUEUE) {
929 /* ToDo: phps merge with spark activation above */
930 /* check whether we have local work and send requests if we have none */
931 if (EMPTY_RUN_QUEUE()) { /* no runnable threads */
932 /* :-[ no local threads => look out for local sparks */
933 /* the spark pool for the current PE */
934 pool = &(MainRegTable.rSparks); // generalise to cap = &MainRegTable
935 if (advisory_thread_count < RtsFlags.ParFlags.maxThreads &&
936 pool->hd < pool->tl) {
938 * ToDo: add GC code check that we really have enough heap afterwards!!
940 * If we're here (no runnable threads) and we have pending
941 * sparks, we must have a space problem. Get enough space
942 * to turn one of those pending sparks into a
946 spark = findSpark(rtsFalse); /* get a spark */
947 if (spark != (rtsSpark) NULL) {
948 tso = activateSpark(spark); /* turn the spark into a thread */
949 IF_PAR_DEBUG(schedule,
950 belch("==== schedule: Created TSO %d (%p); %d threads active",
951 tso->id, tso, advisory_thread_count));
953 if (tso==END_TSO_QUEUE) { /* failed to activate spark->back to loop */
954 belch("==^^ failed to activate spark");
956 } /* otherwise fall through & pick-up new tso */
958 IF_PAR_DEBUG(verbose,
959 belch("==^^ no local sparks (spark pool contains only NFs: %d)",
960 spark_queue_len(pool)));
965 /* If we still have no work we need to send a FISH to get a spark
968 if (EMPTY_RUN_QUEUE()) {
969 /* =8-[ no local sparks => look for work on other PEs */
971 * We really have absolutely no work. Send out a fish
972 * (there may be some out there already), and wait for
973 * something to arrive. We clearly can't run any threads
974 * until a SCHEDULE or RESUME arrives, and so that's what
975 * we're hoping to see. (Of course, we still have to
976 * respond to other types of messages.)
978 TIME now = msTime() /*CURRENT_TIME*/;
979 IF_PAR_DEBUG(verbose,
980 belch("-- now=%ld", now));
981 IF_PAR_DEBUG(verbose,
982 if (outstandingFishes < RtsFlags.ParFlags.maxFishes &&
983 (last_fish_arrived_at!=0 &&
984 last_fish_arrived_at+RtsFlags.ParFlags.fishDelay > now)) {
985 belch("--$$ delaying FISH until %ld (last fish %ld, delay %ld, now %ld)",
986 last_fish_arrived_at+RtsFlags.ParFlags.fishDelay,
987 last_fish_arrived_at,
988 RtsFlags.ParFlags.fishDelay, now);
991 if (outstandingFishes < RtsFlags.ParFlags.maxFishes &&
992 (last_fish_arrived_at==0 ||
993 (last_fish_arrived_at+RtsFlags.ParFlags.fishDelay <= now))) {
994 /* outstandingFishes is set in sendFish, processFish;
995 avoid flooding system with fishes via delay */
997 sendFish(pe, mytid, NEW_FISH_AGE, NEW_FISH_HISTORY,
1000 // Global statistics: count no. of fishes
1001 if (RtsFlags.ParFlags.ParStats.Global &&
1002 RtsFlags.GcFlags.giveStats > NO_GC_STATS) {
1003 globalParStats.tot_fish_mess++;
1007 receivedFinish = processMessages();
1010 } else if (PacketsWaiting()) { /* Look for incoming messages */
1011 receivedFinish = processMessages();
1014 /* Now we are sure that we have some work available */
1015 ASSERT(run_queue_hd != END_TSO_QUEUE);
1017 /* Take a thread from the run queue, if we have work */
1018 t = POP_RUN_QUEUE(); // take_off_run_queue(END_TSO_QUEUE);
1019 IF_DEBUG(sanity,checkTSO(t));
1021 /* ToDo: write something to the log-file
1022 if (RTSflags.ParFlags.granSimStats && !sameThread)
1023 DumpGranEvent(GR_SCHEDULE, RunnableThreadsHd);
1027 /* the spark pool for the current PE */
1028 pool = &(MainRegTable.rSparks); // generalise to cap = &MainRegTable
1031 belch("--=^ %d threads, %d sparks on [%#x]",
1032 run_queue_len(), spark_queue_len(pool), CURRENT_PROC));
1035 if (0 && RtsFlags.ParFlags.ParStats.Full &&
1036 t && LastTSO && t->id != LastTSO->id &&
1037 LastTSO->why_blocked == NotBlocked &&
1038 LastTSO->what_next != ThreadComplete) {
1039 // if previously scheduled TSO not blocked we have to record the context switch
1040 DumpVeryRawGranEvent(TimeOfLastYield, CURRENT_PROC, CURRENT_PROC,
1041 GR_DESCHEDULE, LastTSO, (StgClosure *)NULL, 0, 0);
1044 if (RtsFlags.ParFlags.ParStats.Full &&
1045 (emitSchedule /* forced emit */ ||
1046 (t && LastTSO && t->id != LastTSO->id))) {
1048 we are running a different TSO, so write a schedule event to log file
1049 NB: If we use fair scheduling we also have to write a deschedule
1050 event for LastTSO; with unfair scheduling we know that the
1051 previous tso has blocked whenever we switch to another tso, so
1052 we don't need it in GUM for now
1054 DumpRawGranEvent(CURRENT_PROC, CURRENT_PROC,
1055 GR_SCHEDULE, t, (StgClosure *)NULL, 0, 0);
1056 emitSchedule = rtsFalse;
1060 #else /* !GRAN && !PAR */
1062 /* grab a thread from the run queue */
1063 ASSERT(run_queue_hd != END_TSO_QUEUE);
1064 t = POP_RUN_QUEUE();
1065 // Sanity check the thread we're about to run. This can be
1066 // expensive if there is lots of thread switching going on...
1067 IF_DEBUG(sanity,checkTSO(t));
1070 cap->r.rCurrentTSO = t;
1072 /* context switches are now initiated by the timer signal, unless
1073 * the user specified "context switch as often as possible", with
1076 if ((RtsFlags.ConcFlags.ctxtSwitchTicks == 0
1077 && (run_queue_hd != END_TSO_QUEUE
1078 || blocked_queue_hd != END_TSO_QUEUE
1079 || sleeping_queue != END_TSO_QUEUE)))
1086 RELEASE_LOCK(&sched_mutex);
1088 IF_DEBUG(scheduler, sched_belch("-->> running thread %ld %s ...",
1089 t->id, whatNext_strs[t->what_next]));
1092 startHeapProfTimer();
1095 /* +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ */
1096 /* Run the current thread
1098 prev_what_next = t->what_next;
1099 switch (prev_what_next) {
1101 case ThreadComplete:
1102 /* Thread already finished, return to scheduler. */
1103 ret = ThreadFinished;
1106 ret = StgRun((StgFunPtr) stg_returnToStackTop, &cap->r);
1108 case ThreadInterpret:
1109 ret = interpretBCO(cap);
1112 barf("schedule: invalid what_next field");
1114 /* +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ */
1116 /* Costs for the scheduler are assigned to CCS_SYSTEM */
1118 stopHeapProfTimer();
1122 ACQUIRE_LOCK(&sched_mutex);
1124 #ifdef RTS_SUPPORTS_THREADS
1125 IF_DEBUG(scheduler,fprintf(stderr,"scheduler (task %ld): ", osThreadId()););
1126 #elif !defined(GRAN) && !defined(PAR)
1127 IF_DEBUG(scheduler,fprintf(stderr,"scheduler: "););
1129 t = cap->r.rCurrentTSO;
1132 /* HACK 675: if the last thread didn't yield, make sure to print a
1133 SCHEDULE event to the log file when StgRunning the next thread, even
1134 if it is the same one as before */
1136 TimeOfLastYield = CURRENT_TIME;
1142 IF_DEBUG(gran, DumpGranEvent(GR_DESCHEDULE, t));
1143 globalGranStats.tot_heapover++;
1145 globalParStats.tot_heapover++;
1148 // did the task ask for a large block?
1149 if (cap->r.rHpAlloc > BLOCK_SIZE_W) {
1150 // if so, get one and push it on the front of the nursery.
1154 blocks = (nat)BLOCK_ROUND_UP(cap->r.rHpAlloc * sizeof(W_)) / BLOCK_SIZE;
1156 IF_DEBUG(scheduler,belch("--<< thread %ld (%s) stopped: requesting a large block (size %d)",
1157 t->id, whatNext_strs[t->what_next], blocks));
1159 // don't do this if it would push us over the
1160 // alloc_blocks_lim limit; we'll GC first.
1161 if (alloc_blocks + blocks < alloc_blocks_lim) {
1163 alloc_blocks += blocks;
1164 bd = allocGroup( blocks );
1166 // link the new group into the list
1167 bd->link = cap->r.rCurrentNursery;
1168 bd->u.back = cap->r.rCurrentNursery->u.back;
1169 if (cap->r.rCurrentNursery->u.back != NULL) {
1170 cap->r.rCurrentNursery->u.back->link = bd;
1172 ASSERT(g0s0->blocks == cap->r.rCurrentNursery &&
1173 g0s0->blocks == cap->r.rNursery);
1174 cap->r.rNursery = g0s0->blocks = bd;
1176 cap->r.rCurrentNursery->u.back = bd;
1178 // initialise it as a nursery block. We initialise the
1179 // step, gen_no, and flags field of *every* sub-block in
1180 // this large block, because this is easier than making
1181 // sure that we always find the block head of a large
1182 // block whenever we call Bdescr() (eg. evacuate() and
1183 // isAlive() in the GC would both have to do this, at
1187 for (x = bd; x < bd + blocks; x++) {
1194 // don't forget to update the block count in g0s0.
1195 g0s0->n_blocks += blocks;
1196 // This assert can be a killer if the app is doing lots
1197 // of large block allocations.
1198 ASSERT(countBlocks(g0s0->blocks) == g0s0->n_blocks);
1200 // now update the nursery to point to the new block
1201 cap->r.rCurrentNursery = bd;
1203 // we might be unlucky and have another thread get on the
1204 // run queue before us and steal the large block, but in that
1205 // case the thread will just end up requesting another large
1207 PUSH_ON_RUN_QUEUE(t);
1212 /* make all the running tasks block on a condition variable,
1213 * maybe set context_switch and wait till they all pile in,
1214 * then have them wait on a GC condition variable.
1216 IF_DEBUG(scheduler,belch("--<< thread %ld (%s) stopped: HeapOverflow",
1217 t->id, whatNext_strs[t->what_next]));
1220 ASSERT(!is_on_queue(t,CurrentProc));
1222 /* Currently we emit a DESCHEDULE event before GC in GUM.
1223 ToDo: either add separate event to distinguish SYSTEM time from rest
1224 or just nuke this DESCHEDULE (and the following SCHEDULE) */
1225 if (0 && RtsFlags.ParFlags.ParStats.Full) {
1226 DumpRawGranEvent(CURRENT_PROC, CURRENT_PROC,
1227 GR_DESCHEDULE, t, (StgClosure *)NULL, 0, 0);
1228 emitSchedule = rtsTrue;
1232 ready_to_gc = rtsTrue;
1233 context_switch = 1; /* stop other threads ASAP */
1234 PUSH_ON_RUN_QUEUE(t);
1235 /* actual GC is done at the end of the while loop */
1241 DumpGranEvent(GR_DESCHEDULE, t));
1242 globalGranStats.tot_stackover++;
1245 // DumpGranEvent(GR_DESCHEDULE, t);
1246 globalParStats.tot_stackover++;
1248 IF_DEBUG(scheduler,belch("--<< thread %ld (%s) stopped, StackOverflow",
1249 t->id, whatNext_strs[t->what_next]));
1250 /* just adjust the stack for this thread, then pop it back
1256 /* enlarge the stack */
1257 StgTSO *new_t = threadStackOverflow(t);
1259 /* This TSO has moved, so update any pointers to it from the
1260 * main thread stack. It better not be on any other queues...
1261 * (it shouldn't be).
1263 for (m = main_threads; m != NULL; m = m->link) {
1268 threadPaused(new_t);
1269 PUSH_ON_RUN_QUEUE(new_t);
1273 case ThreadYielding:
1276 DumpGranEvent(GR_DESCHEDULE, t));
1277 globalGranStats.tot_yields++;
1280 // DumpGranEvent(GR_DESCHEDULE, t);
1281 globalParStats.tot_yields++;
1283 /* put the thread back on the run queue. Then, if we're ready to
1284 * GC, check whether this is the last task to stop. If so, wake
1285 * up the GC thread. getThread will block during a GC until the
1289 if (t->what_next != prev_what_next) {
1290 belch("--<< thread %ld (%s) stopped to switch evaluators",
1291 t->id, whatNext_strs[t->what_next]);
1293 belch("--<< thread %ld (%s) stopped, yielding",
1294 t->id, whatNext_strs[t->what_next]);
1299 //belch("&& Doing sanity check on yielding TSO %ld.", t->id);
1301 ASSERT(t->link == END_TSO_QUEUE);
1303 // Shortcut if we're just switching evaluators: don't bother
1304 // doing stack squeezing (which can be expensive), just run the
1306 if (t->what_next != prev_what_next) {
1313 ASSERT(!is_on_queue(t,CurrentProc));
1316 //belch("&& Doing sanity check on all ThreadQueues (and their TSOs).");
1317 checkThreadQsSanity(rtsTrue));
1321 if (RtsFlags.ParFlags.doFairScheduling) {
1322 /* this does round-robin scheduling; good for concurrency */
1323 APPEND_TO_RUN_QUEUE(t);
1325 /* this does unfair scheduling; good for parallelism */
1326 PUSH_ON_RUN_QUEUE(t);
1329 // this does round-robin scheduling; good for concurrency
1330 APPEND_TO_RUN_QUEUE(t);
1334 /* add a ContinueThread event to actually process the thread */
1335 new_event(CurrentProc, CurrentProc, CurrentTime[CurrentProc],
1337 t, (StgClosure*)NULL, (rtsSpark*)NULL);
1339 belch("GRAN: eventq and runnableq after adding yielded thread to queue again:");
1348 belch("--<< thread %ld (%p; %s) stopped, blocking on node %p [PE %d] with BQ: ",
1349 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)));
1350 if (t->block_info.closure!=(StgClosure*)NULL) print_bq(t->block_info.closure));
1352 // ??? needed; should emit block before
1354 DumpGranEvent(GR_DESCHEDULE, t));
1355 prune_eventq(t, (StgClosure *)NULL); // prune ContinueThreads for t
1358 ASSERT(procStatus[CurrentProc]==Busy ||
1359 ((procStatus[CurrentProc]==Fetching) &&
1360 (t->block_info.closure!=(StgClosure*)NULL)));
1361 if (run_queue_hds[CurrentProc] == END_TSO_QUEUE &&
1362 !(!RtsFlags.GranFlags.DoAsyncFetch &&
1363 procStatus[CurrentProc]==Fetching))
1364 procStatus[CurrentProc] = Idle;
1368 belch("--<< thread %ld (%p; %s) stopped, blocking on node %p with BQ: ",
1369 t->id, t, whatNext_strs[t->what_next], t->block_info.closure));
1372 if (t->block_info.closure!=(StgClosure*)NULL)
1373 print_bq(t->block_info.closure));
1375 /* Send a fetch (if BlockedOnGA) and dump event to log file */
1378 /* whatever we schedule next, we must log that schedule */
1379 emitSchedule = rtsTrue;
1382 /* don't need to do anything. Either the thread is blocked on
1383 * I/O, in which case we'll have called addToBlockedQueue
1384 * previously, or it's blocked on an MVar or Blackhole, in which
1385 * case it'll be on the relevant queue already.
1388 fprintf(stderr, "--<< thread %d (%s) stopped: ",
1389 t->id, whatNext_strs[t->what_next]);
1390 printThreadBlockage(t);
1391 fprintf(stderr, "\n"));
1393 /* Only for dumping event to log file
1394 ToDo: do I need this in GranSim, too?
1401 case ThreadFinished:
1402 /* Need to check whether this was a main thread, and if so, signal
1403 * the task that started it with the return value. If we have no
1404 * more main threads, we probably need to stop all the tasks until
1407 /* We also end up here if the thread kills itself with an
1408 * uncaught exception, see Exception.hc.
1410 IF_DEBUG(scheduler,belch("--++ thread %d (%s) finished",
1411 t->id, whatNext_strs[t->what_next]));
1413 endThread(t, CurrentProc); // clean-up the thread
1415 /* For now all are advisory -- HWL */
1416 //if(t->priority==AdvisoryPriority) ??
1417 advisory_thread_count--;
1420 if(t->dist.priority==RevalPriority)
1424 if (RtsFlags.ParFlags.ParStats.Full &&
1425 !RtsFlags.ParFlags.ParStats.Suppressed)
1426 DumpEndEvent(CURRENT_PROC, t, rtsFalse /* not mandatory */);
1431 barf("schedule: invalid thread return code %d", (int)ret);
1435 // When we have +RTS -i0 and we're heap profiling, do a census at
1436 // every GC. This lets us get repeatable runs for debugging.
1437 if (performHeapProfile ||
1438 (RtsFlags.ProfFlags.profileInterval==0 &&
1439 RtsFlags.ProfFlags.doHeapProfile && ready_to_gc)) {
1440 GarbageCollect(GetRoots, rtsTrue);
1442 performHeapProfile = rtsFalse;
1443 ready_to_gc = rtsFalse; // we already GC'd
1449 && allFreeCapabilities()
1452 /* everybody back, start the GC.
1453 * Could do it in this thread, or signal a condition var
1454 * to do it in another thread. Either way, we need to
1455 * broadcast on gc_pending_cond afterward.
1457 #if defined(RTS_SUPPORTS_THREADS)
1458 IF_DEBUG(scheduler,sched_belch("doing GC"));
1460 GarbageCollect(GetRoots,rtsFalse);
1461 ready_to_gc = rtsFalse;
1463 broadcastCondition(&gc_pending_cond);
1466 /* add a ContinueThread event to continue execution of current thread */
1467 new_event(CurrentProc, CurrentProc, CurrentTime[CurrentProc],
1469 t, (StgClosure*)NULL, (rtsSpark*)NULL);
1471 fprintf(stderr, "GRAN: eventq and runnableq after Garbage collection:\n");
1479 IF_GRAN_DEBUG(unused,
1480 print_eventq(EventHd));
1482 event = get_next_event();
1485 /* ToDo: wait for next message to arrive rather than busy wait */
1488 } /* end of while(1) */
1490 IF_PAR_DEBUG(verbose,
1491 belch("== Leaving schedule() after having received Finish"));
1494 /* ---------------------------------------------------------------------------
1495 * Singleton fork(). Do not copy any running threads.
1496 * ------------------------------------------------------------------------- */
1498 StgInt forkProcess(StgTSO* tso) {
1500 #ifndef mingw32_TARGET_OS
1506 IF_DEBUG(scheduler,sched_belch("forking!"));
1509 if (pid) { /* parent */
1511 /* just return the pid */
1513 } else { /* child */
1514 /* wipe all other threads */
1515 run_queue_hd = run_queue_tl = tso;
1516 tso->link = END_TSO_QUEUE;
1518 /* When clearing out the threads, we need to ensure
1519 that a 'main thread' is left behind; if there isn't,
1520 the Scheduler will shutdown next time it is entered.
1522 ==> we don't kill a thread that's on the main_threads
1523 list (nor the current thread.)
1525 [ Attempts at implementing the more ambitious scheme of
1526 killing the main_threads also, and then adding the
1527 current thread onto the main_threads list if it wasn't
1528 there already, failed -- waitThread() (for one) wasn't
1529 up to it. If it proves to be desirable to also kill
1530 the main threads, then this scheme will have to be
1531 revisited (and fully debugged!)
1536 /* DO NOT TOUCH THE QUEUES directly because most of the code around
1537 us is picky about finding the thread still in its queue when
1538 handling the deleteThread() */
1540 for (t = all_threads; t != END_TSO_QUEUE; t = next) {
1543 /* Don't kill the current thread.. */
1544 if (t->id == tso->id) continue;
1546 /* ..or a main thread */
1547 for (m = main_threads; m != NULL; m = m->link) {
1548 if (m->tso->id == t->id) {
1560 barf("forkProcess#: primop not implemented for mingw32, sorry! (%u)\n", tso->id);
1561 /* pointlessly printing out the TSOs 'id' to avoid CC unused warning. */
1563 #endif /* mingw32 */
1566 /* ---------------------------------------------------------------------------
1567 * deleteAllThreads(): kill all the live threads.
1569 * This is used when we catch a user interrupt (^C), before performing
1570 * any necessary cleanups and running finalizers.
1572 * Locks: sched_mutex held.
1573 * ------------------------------------------------------------------------- */
1575 void deleteAllThreads ( void )
1578 IF_DEBUG(scheduler,sched_belch("deleting all threads"));
1579 for (t = all_threads; t != END_TSO_QUEUE; t = next) {
1580 next = t->global_link;
1583 run_queue_hd = run_queue_tl = END_TSO_QUEUE;
1584 blocked_queue_hd = blocked_queue_tl = END_TSO_QUEUE;
1585 sleeping_queue = END_TSO_QUEUE;
1588 /* startThread and insertThread are now in GranSim.c -- HWL */
1591 //@node Suspend and Resume, Run queue code, Main scheduling loop, Main scheduling code
1592 //@subsection Suspend and Resume
1594 /* ---------------------------------------------------------------------------
1595 * Suspending & resuming Haskell threads.
1597 * When making a "safe" call to C (aka _ccall_GC), the task gives back
1598 * its capability before calling the C function. This allows another
1599 * task to pick up the capability and carry on running Haskell
1600 * threads. It also means that if the C call blocks, it won't lock
1603 * The Haskell thread making the C call is put to sleep for the
1604 * duration of the call, on the susepended_ccalling_threads queue. We
1605 * give out a token to the task, which it can use to resume the thread
1606 * on return from the C function.
1607 * ------------------------------------------------------------------------- */
1610 suspendThread( StgRegTable *reg,
1612 #if !defined(RTS_SUPPORTS_THREADS) && !defined(DEBUG)
1620 /* assume that *reg is a pointer to the StgRegTable part
1623 cap = (Capability *)((void *)reg - sizeof(StgFunTable));
1625 ACQUIRE_LOCK(&sched_mutex);
1628 sched_belch("thread %d did a _ccall_gc (is_concurrent: %d)", cap->r.rCurrentTSO->id,concCall));
1630 // XXX this might not be necessary --SDM
1631 cap->r.rCurrentTSO->what_next = ThreadRunGHC;
1633 threadPaused(cap->r.rCurrentTSO);
1634 cap->r.rCurrentTSO->link = suspended_ccalling_threads;
1635 suspended_ccalling_threads = cap->r.rCurrentTSO;
1637 #if defined(RTS_SUPPORTS_THREADS)
1638 if(cap->r.rCurrentTSO->blocked_exceptions == NULL)
1640 cap->r.rCurrentTSO->why_blocked = BlockedOnCCall;
1641 cap->r.rCurrentTSO->blocked_exceptions = END_TSO_QUEUE;
1645 cap->r.rCurrentTSO->why_blocked = BlockedOnCCall_NoUnblockExc;
1649 /* Use the thread ID as the token; it should be unique */
1650 tok = cap->r.rCurrentTSO->id;
1652 /* Hand back capability */
1653 releaseCapability(cap);
1655 #if defined(RTS_SUPPORTS_THREADS)
1656 /* Preparing to leave the RTS, so ensure there's a native thread/task
1657 waiting to take over.
1659 IF_DEBUG(scheduler, sched_belch("worker thread (%d, osthread %p): leaving RTS", tok, osThreadId()));
1660 //if (concCall) { // implementing "safe" as opposed to "threadsafe" is more difficult
1661 startTask(taskStart);
1665 /* Other threads _might_ be available for execution; signal this */
1667 RELEASE_LOCK(&sched_mutex);
1672 resumeThread( StgInt tok,
1673 rtsBool concCall STG_UNUSED )
1675 StgTSO *tso, **prev;
1678 #if defined(RTS_SUPPORTS_THREADS)
1679 /* Wait for permission to re-enter the RTS with the result. */
1680 ACQUIRE_LOCK(&sched_mutex);
1681 grabReturnCapability(&sched_mutex, &cap);
1683 IF_DEBUG(scheduler, sched_belch("worker thread (%d, osthread %p): re-entering RTS", tok, osThreadId()));
1685 grabCapability(&cap);
1688 /* Remove the thread off of the suspended list */
1689 prev = &suspended_ccalling_threads;
1690 for (tso = suspended_ccalling_threads;
1691 tso != END_TSO_QUEUE;
1692 prev = &tso->link, tso = tso->link) {
1693 if (tso->id == (StgThreadID)tok) {
1698 if (tso == END_TSO_QUEUE) {
1699 barf("resumeThread: thread not found");
1701 tso->link = END_TSO_QUEUE;
1703 #if defined(RTS_SUPPORTS_THREADS)
1704 if(tso->why_blocked == BlockedOnCCall)
1706 awakenBlockedQueueNoLock(tso->blocked_exceptions);
1707 tso->blocked_exceptions = NULL;
1711 /* Reset blocking status */
1712 tso->why_blocked = NotBlocked;
1714 cap->r.rCurrentTSO = tso;
1715 #if defined(RTS_SUPPORTS_THREADS)
1716 RELEASE_LOCK(&sched_mutex);
1722 /* ---------------------------------------------------------------------------
1724 * ------------------------------------------------------------------------ */
1725 static void unblockThread(StgTSO *tso);
1727 /* ---------------------------------------------------------------------------
1728 * Comparing Thread ids.
1730 * This is used from STG land in the implementation of the
1731 * instances of Eq/Ord for ThreadIds.
1732 * ------------------------------------------------------------------------ */
1735 cmp_thread(StgPtr tso1, StgPtr tso2)
1737 StgThreadID id1 = ((StgTSO *)tso1)->id;
1738 StgThreadID id2 = ((StgTSO *)tso2)->id;
1740 if (id1 < id2) return (-1);
1741 if (id1 > id2) return 1;
1745 /* ---------------------------------------------------------------------------
1746 * Fetching the ThreadID from an StgTSO.
1748 * This is used in the implementation of Show for ThreadIds.
1749 * ------------------------------------------------------------------------ */
1751 rts_getThreadId(StgPtr tso)
1753 return ((StgTSO *)tso)->id;
1758 labelThread(StgPtr tso, char *label)
1763 /* Caveat: Once set, you can only set the thread name to "" */
1764 len = strlen(label)+1;
1765 buf = stgMallocBytes(len * sizeof(char), "Schedule.c:labelThread()");
1766 strncpy(buf,label,len);
1767 /* Update will free the old memory for us */
1768 updateThreadLabel((StgWord)tso,buf);
1772 /* ---------------------------------------------------------------------------
1773 Create a new thread.
1775 The new thread starts with the given stack size. Before the
1776 scheduler can run, however, this thread needs to have a closure
1777 (and possibly some arguments) pushed on its stack. See
1778 pushClosure() in Schedule.h.
1780 createGenThread() and createIOThread() (in SchedAPI.h) are
1781 convenient packaged versions of this function.
1783 currently pri (priority) is only used in a GRAN setup -- HWL
1784 ------------------------------------------------------------------------ */
1785 //@cindex createThread
1787 /* currently pri (priority) is only used in a GRAN setup -- HWL */
1789 createThread(nat size, StgInt pri)
1792 createThread(nat size)
1799 /* First check whether we should create a thread at all */
1801 /* check that no more than RtsFlags.ParFlags.maxThreads threads are created */
1802 if (advisory_thread_count >= RtsFlags.ParFlags.maxThreads) {
1804 belch("{createThread}Daq ghuH: refusing to create another thread; no more than %d threads allowed (currently %d)",
1805 RtsFlags.ParFlags.maxThreads, advisory_thread_count);
1806 return END_TSO_QUEUE;
1812 ASSERT(!RtsFlags.GranFlags.Light || CurrentProc==0);
1815 // ToDo: check whether size = stack_size - TSO_STRUCT_SIZEW
1817 /* catch ridiculously small stack sizes */
1818 if (size < MIN_STACK_WORDS + TSO_STRUCT_SIZEW) {
1819 size = MIN_STACK_WORDS + TSO_STRUCT_SIZEW;
1822 stack_size = size - TSO_STRUCT_SIZEW;
1824 tso = (StgTSO *)allocate(size);
1825 TICK_ALLOC_TSO(stack_size, 0);
1827 SET_HDR(tso, &stg_TSO_info, CCS_SYSTEM);
1829 SET_GRAN_HDR(tso, ThisPE);
1832 // Always start with the compiled code evaluator
1833 tso->what_next = ThreadRunGHC;
1835 /* tso->id needs to be unique. For now we use a heavyweight mutex to
1836 * protect the increment operation on next_thread_id.
1837 * In future, we could use an atomic increment instead.
1839 ACQUIRE_LOCK(&thread_id_mutex);
1840 tso->id = next_thread_id++;
1841 RELEASE_LOCK(&thread_id_mutex);
1843 tso->why_blocked = NotBlocked;
1844 tso->blocked_exceptions = NULL;
1846 tso->stack_size = stack_size;
1847 tso->max_stack_size = round_to_mblocks(RtsFlags.GcFlags.maxStkSize)
1849 tso->sp = (P_)&(tso->stack) + stack_size;
1852 tso->prof.CCCS = CCS_MAIN;
1855 /* put a stop frame on the stack */
1856 tso->sp -= sizeofW(StgStopFrame);
1857 SET_HDR((StgClosure*)tso->sp,(StgInfoTable *)&stg_stop_thread_info,CCS_SYSTEM);
1860 tso->link = END_TSO_QUEUE;
1861 /* uses more flexible routine in GranSim */
1862 insertThread(tso, CurrentProc);
1864 /* In a non-GranSim setup the pushing of a TSO onto the runq is separated
1870 if (RtsFlags.GranFlags.GranSimStats.Full)
1871 DumpGranEvent(GR_START,tso);
1873 if (RtsFlags.ParFlags.ParStats.Full)
1874 DumpGranEvent(GR_STARTQ,tso);
1875 /* HACk to avoid SCHEDULE
1879 /* Link the new thread on the global thread list.
1881 tso->global_link = all_threads;
1885 tso->dist.priority = MandatoryPriority; //by default that is...
1889 tso->gran.pri = pri;
1891 tso->gran.magic = TSO_MAGIC; // debugging only
1893 tso->gran.sparkname = 0;
1894 tso->gran.startedat = CURRENT_TIME;
1895 tso->gran.exported = 0;
1896 tso->gran.basicblocks = 0;
1897 tso->gran.allocs = 0;
1898 tso->gran.exectime = 0;
1899 tso->gran.fetchtime = 0;
1900 tso->gran.fetchcount = 0;
1901 tso->gran.blocktime = 0;
1902 tso->gran.blockcount = 0;
1903 tso->gran.blockedat = 0;
1904 tso->gran.globalsparks = 0;
1905 tso->gran.localsparks = 0;
1906 if (RtsFlags.GranFlags.Light)
1907 tso->gran.clock = Now; /* local clock */
1909 tso->gran.clock = 0;
1911 IF_DEBUG(gran,printTSO(tso));
1914 tso->par.magic = TSO_MAGIC; // debugging only
1916 tso->par.sparkname = 0;
1917 tso->par.startedat = CURRENT_TIME;
1918 tso->par.exported = 0;
1919 tso->par.basicblocks = 0;
1920 tso->par.allocs = 0;
1921 tso->par.exectime = 0;
1922 tso->par.fetchtime = 0;
1923 tso->par.fetchcount = 0;
1924 tso->par.blocktime = 0;
1925 tso->par.blockcount = 0;
1926 tso->par.blockedat = 0;
1927 tso->par.globalsparks = 0;
1928 tso->par.localsparks = 0;
1932 globalGranStats.tot_threads_created++;
1933 globalGranStats.threads_created_on_PE[CurrentProc]++;
1934 globalGranStats.tot_sq_len += spark_queue_len(CurrentProc);
1935 globalGranStats.tot_sq_probes++;
1937 // collect parallel global statistics (currently done together with GC stats)
1938 if (RtsFlags.ParFlags.ParStats.Global &&
1939 RtsFlags.GcFlags.giveStats > NO_GC_STATS) {
1940 //fprintf(stderr, "Creating thread %d @ %11.2f\n", tso->id, usertime());
1941 globalParStats.tot_threads_created++;
1947 belch("==__ schedule: Created TSO %d (%p);",
1948 CurrentProc, tso, tso->id));
1950 IF_PAR_DEBUG(verbose,
1951 belch("==__ schedule: Created TSO %d (%p); %d threads active",
1952 tso->id, tso, advisory_thread_count));
1954 IF_DEBUG(scheduler,sched_belch("created thread %ld, stack size = %lx words",
1955 tso->id, tso->stack_size));
1962 all parallel thread creation calls should fall through the following routine.
1965 createSparkThread(rtsSpark spark)
1967 ASSERT(spark != (rtsSpark)NULL);
1968 if (advisory_thread_count >= RtsFlags.ParFlags.maxThreads)
1970 barf("{createSparkThread}Daq ghuH: refusing to create another thread; no more than %d threads allowed (currently %d)",
1971 RtsFlags.ParFlags.maxThreads, advisory_thread_count);
1972 return END_TSO_QUEUE;
1976 tso = createThread(RtsFlags.GcFlags.initialStkSize);
1977 if (tso==END_TSO_QUEUE)
1978 barf("createSparkThread: Cannot create TSO");
1980 tso->priority = AdvisoryPriority;
1982 pushClosure(tso,spark);
1983 PUSH_ON_RUN_QUEUE(tso);
1984 advisory_thread_count++;
1991 Turn a spark into a thread.
1992 ToDo: fix for SMP (needs to acquire SCHED_MUTEX!)
1995 //@cindex activateSpark
1997 activateSpark (rtsSpark spark)
2001 tso = createSparkThread(spark);
2002 if (RtsFlags.ParFlags.ParStats.Full) {
2003 //ASSERT(run_queue_hd == END_TSO_QUEUE); // I think ...
2004 IF_PAR_DEBUG(verbose,
2005 belch("==^^ activateSpark: turning spark of closure %p (%s) into a thread",
2006 (StgClosure *)spark, info_type((StgClosure *)spark)));
2008 // ToDo: fwd info on local/global spark to thread -- HWL
2009 // tso->gran.exported = spark->exported;
2010 // tso->gran.locked = !spark->global;
2011 // tso->gran.sparkname = spark->name;
2017 static SchedulerStatus waitThread_(/*out*/StgMainThread* m
2018 #if defined(THREADED_RTS)
2019 , rtsBool blockWaiting
2024 /* ---------------------------------------------------------------------------
2027 * scheduleThread puts a thread on the head of the runnable queue.
2028 * This will usually be done immediately after a thread is created.
2029 * The caller of scheduleThread must create the thread using e.g.
2030 * createThread and push an appropriate closure
2031 * on this thread's stack before the scheduler is invoked.
2032 * ------------------------------------------------------------------------ */
2034 static void scheduleThread_ (StgTSO* tso);
2037 scheduleThread_(StgTSO *tso)
2039 // Precondition: sched_mutex must be held.
2041 /* Put the new thread on the head of the runnable queue. The caller
2042 * better push an appropriate closure on this thread's stack
2043 * beforehand. In the SMP case, the thread may start running as
2044 * soon as we release the scheduler lock below.
2046 PUSH_ON_RUN_QUEUE(tso);
2050 IF_DEBUG(scheduler,printTSO(tso));
2054 void scheduleThread(StgTSO* tso)
2056 ACQUIRE_LOCK(&sched_mutex);
2057 scheduleThread_(tso);
2058 RELEASE_LOCK(&sched_mutex);
2062 scheduleWaitThread(StgTSO* tso, /*[out]*/HaskellObj* ret)
2063 { // Precondition: sched_mutex must be held
2066 m = stgMallocBytes(sizeof(StgMainThread), "waitThread");
2070 #if defined(RTS_SUPPORTS_THREADS)
2071 initCondition(&m->wakeup);
2074 /* Put the thread on the main-threads list prior to scheduling the TSO.
2075 Failure to do so introduces a race condition in the MT case (as
2076 identified by Wolfgang Thaller), whereby the new task/OS thread
2077 created by scheduleThread_() would complete prior to the thread
2078 that spawned it managed to put 'itself' on the main-threads list.
2079 The upshot of it all being that the worker thread wouldn't get to
2080 signal the completion of the its work item for the main thread to
2081 see (==> it got stuck waiting.) -- sof 6/02.
2083 IF_DEBUG(scheduler, sched_belch("waiting for thread (%d)\n", tso->id));
2085 m->link = main_threads;
2088 scheduleThread_(tso);
2089 #if defined(THREADED_RTS)
2090 return waitThread_(m, rtsTrue);
2092 return waitThread_(m);
2096 /* ---------------------------------------------------------------------------
2099 * Initialise the scheduler. This resets all the queues - if the
2100 * queues contained any threads, they'll be garbage collected at the
2103 * ------------------------------------------------------------------------ */
2107 term_handler(int sig STG_UNUSED)
2110 ACQUIRE_LOCK(&term_mutex);
2112 RELEASE_LOCK(&term_mutex);
2123 for (i=0; i<=MAX_PROC; i++) {
2124 run_queue_hds[i] = END_TSO_QUEUE;
2125 run_queue_tls[i] = END_TSO_QUEUE;
2126 blocked_queue_hds[i] = END_TSO_QUEUE;
2127 blocked_queue_tls[i] = END_TSO_QUEUE;
2128 ccalling_threadss[i] = END_TSO_QUEUE;
2129 sleeping_queue = END_TSO_QUEUE;
2132 run_queue_hd = END_TSO_QUEUE;
2133 run_queue_tl = END_TSO_QUEUE;
2134 blocked_queue_hd = END_TSO_QUEUE;
2135 blocked_queue_tl = END_TSO_QUEUE;
2136 sleeping_queue = END_TSO_QUEUE;
2139 suspended_ccalling_threads = END_TSO_QUEUE;
2141 main_threads = NULL;
2142 all_threads = END_TSO_QUEUE;
2147 RtsFlags.ConcFlags.ctxtSwitchTicks =
2148 RtsFlags.ConcFlags.ctxtSwitchTime / TICK_MILLISECS;
2150 #if defined(RTS_SUPPORTS_THREADS)
2151 /* Initialise the mutex and condition variables used by
2153 initMutex(&sched_mutex);
2154 initMutex(&term_mutex);
2155 initMutex(&thread_id_mutex);
2157 initCondition(&thread_ready_cond);
2161 initCondition(&gc_pending_cond);
2164 #if defined(RTS_SUPPORTS_THREADS)
2165 ACQUIRE_LOCK(&sched_mutex);
2168 /* Install the SIGHUP handler */
2171 struct sigaction action,oact;
2173 action.sa_handler = term_handler;
2174 sigemptyset(&action.sa_mask);
2175 action.sa_flags = 0;
2176 if (sigaction(SIGTERM, &action, &oact) != 0) {
2177 barf("can't install TERM handler");
2182 /* A capability holds the state a native thread needs in
2183 * order to execute STG code. At least one capability is
2184 * floating around (only SMP builds have more than one).
2188 #if defined(RTS_SUPPORTS_THREADS)
2189 /* start our haskell execution tasks */
2191 startTaskManager(RtsFlags.ParFlags.nNodes, taskStart);
2193 startTaskManager(0,taskStart);
2197 #if /* defined(SMP) ||*/ defined(PAR)
2201 #if defined(RTS_SUPPORTS_THREADS)
2202 RELEASE_LOCK(&sched_mutex);
2208 exitScheduler( void )
2210 #if defined(RTS_SUPPORTS_THREADS)
2213 shutting_down_scheduler = rtsTrue;
2216 /* -----------------------------------------------------------------------------
2217 Managing the per-task allocation areas.
2219 Each capability comes with an allocation area. These are
2220 fixed-length block lists into which allocation can be done.
2222 ToDo: no support for two-space collection at the moment???
2223 -------------------------------------------------------------------------- */
2225 /* -----------------------------------------------------------------------------
2226 * waitThread is the external interface for running a new computation
2227 * and waiting for the result.
2229 * In the non-SMP case, we create a new main thread, push it on the
2230 * main-thread stack, and invoke the scheduler to run it. The
2231 * scheduler will return when the top main thread on the stack has
2232 * completed or died, and fill in the necessary fields of the
2233 * main_thread structure.
2235 * In the SMP case, we create a main thread as before, but we then
2236 * create a new condition variable and sleep on it. When our new
2237 * main thread has completed, we'll be woken up and the status/result
2238 * will be in the main_thread struct.
2239 * -------------------------------------------------------------------------- */
2242 howManyThreadsAvail ( void )
2246 for (q = run_queue_hd; q != END_TSO_QUEUE; q = q->link)
2248 for (q = blocked_queue_hd; q != END_TSO_QUEUE; q = q->link)
2250 for (q = sleeping_queue; q != END_TSO_QUEUE; q = q->link)
2256 finishAllThreads ( void )
2259 while (run_queue_hd != END_TSO_QUEUE) {
2260 waitThread ( run_queue_hd, NULL);
2262 while (blocked_queue_hd != END_TSO_QUEUE) {
2263 waitThread ( blocked_queue_hd, NULL);
2265 while (sleeping_queue != END_TSO_QUEUE) {
2266 waitThread ( blocked_queue_hd, NULL);
2269 (blocked_queue_hd != END_TSO_QUEUE ||
2270 run_queue_hd != END_TSO_QUEUE ||
2271 sleeping_queue != END_TSO_QUEUE);
2275 waitThread(StgTSO *tso, /*out*/StgClosure **ret)
2278 SchedulerStatus stat;
2280 m = stgMallocBytes(sizeof(StgMainThread), "waitThread");
2284 #if defined(RTS_SUPPORTS_THREADS)
2285 initCondition(&m->wakeup);
2288 /* see scheduleWaitThread() comment */
2289 ACQUIRE_LOCK(&sched_mutex);
2290 m->link = main_threads;
2293 IF_DEBUG(scheduler, sched_belch("waiting for thread %d", tso->id));
2294 #if defined(THREADED_RTS)
2295 stat = waitThread_(m, rtsFalse);
2297 stat = waitThread_(m);
2299 RELEASE_LOCK(&sched_mutex);
2305 waitThread_(StgMainThread* m
2306 #if defined(THREADED_RTS)
2307 , rtsBool blockWaiting
2311 SchedulerStatus stat;
2313 // Precondition: sched_mutex must be held.
2314 IF_DEBUG(scheduler, sched_belch("== scheduler: new main thread (%d)\n", m->tso->id));
2316 #if defined(RTS_SUPPORTS_THREADS)
2318 # if defined(THREADED_RTS)
2319 if (!blockWaiting) {
2320 /* In the threaded case, the OS thread that called main()
2321 * gets to enter the RTS directly without going via another
2324 main_main_thread = m;
2325 RELEASE_LOCK(&sched_mutex);
2327 ACQUIRE_LOCK(&sched_mutex);
2328 main_main_thread = NULL;
2329 ASSERT(m->stat != NoStatus);
2334 waitCondition(&m->wakeup, &sched_mutex);
2335 } while (m->stat == NoStatus);
2338 /* GranSim specific init */
2339 CurrentTSO = m->tso; // the TSO to run
2340 procStatus[MainProc] = Busy; // status of main PE
2341 CurrentProc = MainProc; // PE to run it on
2343 RELEASE_LOCK(&sched_mutex);
2346 RELEASE_LOCK(&sched_mutex);
2348 ASSERT(m->stat != NoStatus);
2353 #if defined(RTS_SUPPORTS_THREADS)
2354 closeCondition(&m->wakeup);
2357 IF_DEBUG(scheduler, fprintf(stderr, "== scheduler: main thread (%d) finished\n",
2361 // Postcondition: sched_mutex still held
2365 //@node Run queue code, Garbage Collextion Routines, Suspend and Resume, Main scheduling code
2366 //@subsection Run queue code
2370 NB: In GranSim we have many run queues; run_queue_hd is actually a macro
2371 unfolding to run_queue_hds[CurrentProc], thus CurrentProc is an
2372 implicit global variable that has to be correct when calling these
2376 /* Put the new thread on the head of the runnable queue.
2377 * The caller of createThread better push an appropriate closure
2378 * on this thread's stack before the scheduler is invoked.
2380 static /* inline */ void
2381 add_to_run_queue(tso)
2384 ASSERT(tso!=run_queue_hd && tso!=run_queue_tl);
2385 tso->link = run_queue_hd;
2387 if (run_queue_tl == END_TSO_QUEUE) {
2392 /* Put the new thread at the end of the runnable queue. */
2393 static /* inline */ void
2394 push_on_run_queue(tso)
2397 ASSERT(get_itbl((StgClosure *)tso)->type == TSO);
2398 ASSERT(run_queue_hd!=NULL && run_queue_tl!=NULL);
2399 ASSERT(tso!=run_queue_hd && tso!=run_queue_tl);
2400 if (run_queue_hd == END_TSO_QUEUE) {
2403 run_queue_tl->link = tso;
2409 Should be inlined because it's used very often in schedule. The tso
2410 argument is actually only needed in GranSim, where we want to have the
2411 possibility to schedule *any* TSO on the run queue, irrespective of the
2412 actual ordering. Therefore, if tso is not the nil TSO then we traverse
2413 the run queue and dequeue the tso, adjusting the links in the queue.
2415 //@cindex take_off_run_queue
2416 static /* inline */ StgTSO*
2417 take_off_run_queue(StgTSO *tso) {
2421 qetlaHbogh Qu' ngaSbogh ghomDaQ {tso} yIteq!
2423 if tso is specified, unlink that tso from the run_queue (doesn't have
2424 to be at the beginning of the queue); GranSim only
2426 if (tso!=END_TSO_QUEUE) {
2427 /* find tso in queue */
2428 for (t=run_queue_hd, prev=END_TSO_QUEUE;
2429 t!=END_TSO_QUEUE && t!=tso;
2433 /* now actually dequeue the tso */
2434 if (prev!=END_TSO_QUEUE) {
2435 ASSERT(run_queue_hd!=t);
2436 prev->link = t->link;
2438 /* t is at beginning of thread queue */
2439 ASSERT(run_queue_hd==t);
2440 run_queue_hd = t->link;
2442 /* t is at end of thread queue */
2443 if (t->link==END_TSO_QUEUE) {
2444 ASSERT(t==run_queue_tl);
2445 run_queue_tl = prev;
2447 ASSERT(run_queue_tl!=t);
2449 t->link = END_TSO_QUEUE;
2451 /* take tso from the beginning of the queue; std concurrent code */
2453 if (t != END_TSO_QUEUE) {
2454 run_queue_hd = t->link;
2455 t->link = END_TSO_QUEUE;
2456 if (run_queue_hd == END_TSO_QUEUE) {
2457 run_queue_tl = END_TSO_QUEUE;
2466 //@node Garbage Collextion Routines, Blocking Queue Routines, Run queue code, Main scheduling code
2467 //@subsection Garbage Collextion Routines
2469 /* ---------------------------------------------------------------------------
2470 Where are the roots that we know about?
2472 - all the threads on the runnable queue
2473 - all the threads on the blocked queue
2474 - all the threads on the sleeping queue
2475 - all the thread currently executing a _ccall_GC
2476 - all the "main threads"
2478 ------------------------------------------------------------------------ */
2480 /* This has to be protected either by the scheduler monitor, or by the
2481 garbage collection monitor (probably the latter).
2486 GetRoots(evac_fn evac)
2491 for (i=0; i<=RtsFlags.GranFlags.proc; i++) {
2492 if ((run_queue_hds[i] != END_TSO_QUEUE) && ((run_queue_hds[i] != NULL)))
2493 evac((StgClosure **)&run_queue_hds[i]);
2494 if ((run_queue_tls[i] != END_TSO_QUEUE) && ((run_queue_tls[i] != NULL)))
2495 evac((StgClosure **)&run_queue_tls[i]);
2497 if ((blocked_queue_hds[i] != END_TSO_QUEUE) && ((blocked_queue_hds[i] != NULL)))
2498 evac((StgClosure **)&blocked_queue_hds[i]);
2499 if ((blocked_queue_tls[i] != END_TSO_QUEUE) && ((blocked_queue_tls[i] != NULL)))
2500 evac((StgClosure **)&blocked_queue_tls[i]);
2501 if ((ccalling_threadss[i] != END_TSO_QUEUE) && ((ccalling_threadss[i] != NULL)))
2502 evac((StgClosure **)&ccalling_threads[i]);
2509 if (run_queue_hd != END_TSO_QUEUE) {
2510 ASSERT(run_queue_tl != END_TSO_QUEUE);
2511 evac((StgClosure **)&run_queue_hd);
2512 evac((StgClosure **)&run_queue_tl);
2515 if (blocked_queue_hd != END_TSO_QUEUE) {
2516 ASSERT(blocked_queue_tl != END_TSO_QUEUE);
2517 evac((StgClosure **)&blocked_queue_hd);
2518 evac((StgClosure **)&blocked_queue_tl);
2521 if (sleeping_queue != END_TSO_QUEUE) {
2522 evac((StgClosure **)&sleeping_queue);
2526 if (suspended_ccalling_threads != END_TSO_QUEUE) {
2527 evac((StgClosure **)&suspended_ccalling_threads);
2530 #if defined(PAR) || defined(GRAN)
2531 markSparkQueue(evac);
2534 #if defined(RTS_USER_SIGNALS)
2535 // mark the signal handlers (signals should be already blocked)
2536 markSignalHandlers(evac);
2539 // main threads which have completed need to be retained until they
2540 // are dealt with in the main scheduler loop. They won't be
2541 // retained any other way: the GC will drop them from the
2542 // all_threads list, so we have to be careful to treat them as roots
2546 for (m = main_threads; m != NULL; m = m->link) {
2547 switch (m->tso->what_next) {
2548 case ThreadComplete:
2550 evac((StgClosure **)&m->tso);
2559 /* -----------------------------------------------------------------------------
2562 This is the interface to the garbage collector from Haskell land.
2563 We provide this so that external C code can allocate and garbage
2564 collect when called from Haskell via _ccall_GC.
2566 It might be useful to provide an interface whereby the programmer
2567 can specify more roots (ToDo).
2569 This needs to be protected by the GC condition variable above. KH.
2570 -------------------------------------------------------------------------- */
2572 static void (*extra_roots)(evac_fn);
2577 /* Obligated to hold this lock upon entry */
2578 ACQUIRE_LOCK(&sched_mutex);
2579 GarbageCollect(GetRoots,rtsFalse);
2580 RELEASE_LOCK(&sched_mutex);
2584 performMajorGC(void)
2586 ACQUIRE_LOCK(&sched_mutex);
2587 GarbageCollect(GetRoots,rtsTrue);
2588 RELEASE_LOCK(&sched_mutex);
2592 AllRoots(evac_fn evac)
2594 GetRoots(evac); // the scheduler's roots
2595 extra_roots(evac); // the user's roots
2599 performGCWithRoots(void (*get_roots)(evac_fn))
2601 ACQUIRE_LOCK(&sched_mutex);
2602 extra_roots = get_roots;
2603 GarbageCollect(AllRoots,rtsFalse);
2604 RELEASE_LOCK(&sched_mutex);
2607 /* -----------------------------------------------------------------------------
2610 If the thread has reached its maximum stack size, then raise the
2611 StackOverflow exception in the offending thread. Otherwise
2612 relocate the TSO into a larger chunk of memory and adjust its stack
2614 -------------------------------------------------------------------------- */
2617 threadStackOverflow(StgTSO *tso)
2619 nat new_stack_size, new_tso_size, stack_words;
2623 IF_DEBUG(sanity,checkTSO(tso));
2624 if (tso->stack_size >= tso->max_stack_size) {
2627 belch("@@ threadStackOverflow of TSO %d (%p): stack too large (now %ld; max is %ld",
2628 tso->id, tso, tso->stack_size, tso->max_stack_size);
2629 /* If we're debugging, just print out the top of the stack */
2630 printStackChunk(tso->sp, stg_min(tso->stack+tso->stack_size,
2633 /* Send this thread the StackOverflow exception */
2634 raiseAsync(tso, (StgClosure *)stackOverflow_closure);
2638 /* Try to double the current stack size. If that takes us over the
2639 * maximum stack size for this thread, then use the maximum instead.
2640 * Finally round up so the TSO ends up as a whole number of blocks.
2642 new_stack_size = stg_min(tso->stack_size * 2, tso->max_stack_size);
2643 new_tso_size = (nat)BLOCK_ROUND_UP(new_stack_size * sizeof(W_) +
2644 TSO_STRUCT_SIZE)/sizeof(W_);
2645 new_tso_size = round_to_mblocks(new_tso_size); /* Be MBLOCK-friendly */
2646 new_stack_size = new_tso_size - TSO_STRUCT_SIZEW;
2648 IF_DEBUG(scheduler, fprintf(stderr,"== scheduler: increasing stack size from %d words to %d.\n", tso->stack_size, new_stack_size));
2650 dest = (StgTSO *)allocate(new_tso_size);
2651 TICK_ALLOC_TSO(new_stack_size,0);
2653 /* copy the TSO block and the old stack into the new area */
2654 memcpy(dest,tso,TSO_STRUCT_SIZE);
2655 stack_words = tso->stack + tso->stack_size - tso->sp;
2656 new_sp = (P_)dest + new_tso_size - stack_words;
2657 memcpy(new_sp, tso->sp, stack_words * sizeof(W_));
2659 /* relocate the stack pointers... */
2661 dest->stack_size = new_stack_size;
2663 /* Mark the old TSO as relocated. We have to check for relocated
2664 * TSOs in the garbage collector and any primops that deal with TSOs.
2666 * It's important to set the sp value to just beyond the end
2667 * of the stack, so we don't attempt to scavenge any part of the
2670 tso->what_next = ThreadRelocated;
2672 tso->sp = (P_)&(tso->stack[tso->stack_size]);
2673 tso->why_blocked = NotBlocked;
2674 dest->mut_link = NULL;
2676 IF_PAR_DEBUG(verbose,
2677 belch("@@ threadStackOverflow of TSO %d (now at %p): stack size increased to %ld",
2678 tso->id, tso, tso->stack_size);
2679 /* If we're debugging, just print out the top of the stack */
2680 printStackChunk(tso->sp, stg_min(tso->stack+tso->stack_size,
2683 IF_DEBUG(sanity,checkTSO(tso));
2685 IF_DEBUG(scheduler,printTSO(dest));
2691 //@node Blocking Queue Routines, Exception Handling Routines, Garbage Collextion Routines, Main scheduling code
2692 //@subsection Blocking Queue Routines
2694 /* ---------------------------------------------------------------------------
2695 Wake up a queue that was blocked on some resource.
2696 ------------------------------------------------------------------------ */
2700 unblockCount ( StgBlockingQueueElement *bqe, StgClosure *node )
2705 unblockCount ( StgBlockingQueueElement *bqe, StgClosure *node )
2707 /* write RESUME events to log file and
2708 update blocked and fetch time (depending on type of the orig closure) */
2709 if (RtsFlags.ParFlags.ParStats.Full) {
2710 DumpRawGranEvent(CURRENT_PROC, CURRENT_PROC,
2711 GR_RESUMEQ, ((StgTSO *)bqe), ((StgTSO *)bqe)->block_info.closure,
2712 0, 0 /* spark_queue_len(ADVISORY_POOL) */);
2713 if (EMPTY_RUN_QUEUE())
2714 emitSchedule = rtsTrue;
2716 switch (get_itbl(node)->type) {
2718 ((StgTSO *)bqe)->par.fetchtime += CURRENT_TIME-((StgTSO *)bqe)->par.blockedat;
2723 ((StgTSO *)bqe)->par.blocktime += CURRENT_TIME-((StgTSO *)bqe)->par.blockedat;
2730 barf("{unblockOneLocked}Daq Qagh: unexpected closure in blocking queue");
2737 static StgBlockingQueueElement *
2738 unblockOneLocked(StgBlockingQueueElement *bqe, StgClosure *node)
2741 PEs node_loc, tso_loc;
2743 node_loc = where_is(node); // should be lifted out of loop
2744 tso = (StgTSO *)bqe; // wastes an assignment to get the type right
2745 tso_loc = where_is((StgClosure *)tso);
2746 if (IS_LOCAL_TO(PROCS(node),tso_loc)) { // TSO is local
2747 /* !fake_fetch => TSO is on CurrentProc is same as IS_LOCAL_TO */
2748 ASSERT(CurrentProc!=node_loc || tso_loc==CurrentProc);
2749 CurrentTime[CurrentProc] += RtsFlags.GranFlags.Costs.lunblocktime;
2750 // insertThread(tso, node_loc);
2751 new_event(tso_loc, tso_loc, CurrentTime[CurrentProc],
2753 tso, node, (rtsSpark*)NULL);
2754 tso->link = END_TSO_QUEUE; // overwrite link just to be sure
2757 } else { // TSO is remote (actually should be FMBQ)
2758 CurrentTime[CurrentProc] += RtsFlags.GranFlags.Costs.mpacktime +
2759 RtsFlags.GranFlags.Costs.gunblocktime +
2760 RtsFlags.GranFlags.Costs.latency;
2761 new_event(tso_loc, CurrentProc, CurrentTime[CurrentProc],
2763 tso, node, (rtsSpark*)NULL);
2764 tso->link = END_TSO_QUEUE; // overwrite link just to be sure
2767 /* the thread-queue-overhead is accounted for in either Resume or UnblockThread */
2769 fprintf(stderr," %s TSO %d (%p) [PE %d] (block_info.closure=%p) (next=%p) ,",
2770 (node_loc==tso_loc ? "Local" : "Global"),
2771 tso->id, tso, CurrentProc, tso->block_info.closure, tso->link));
2772 tso->block_info.closure = NULL;
2773 IF_DEBUG(scheduler,belch("-- Waking up thread %ld (%p)",
2777 static StgBlockingQueueElement *
2778 unblockOneLocked(StgBlockingQueueElement *bqe, StgClosure *node)
2780 StgBlockingQueueElement *next;
2782 switch (get_itbl(bqe)->type) {
2784 ASSERT(((StgTSO *)bqe)->why_blocked != NotBlocked);
2785 /* if it's a TSO just push it onto the run_queue */
2787 // ((StgTSO *)bqe)->link = END_TSO_QUEUE; // debugging?
2788 PUSH_ON_RUN_QUEUE((StgTSO *)bqe);
2790 unblockCount(bqe, node);
2791 /* reset blocking status after dumping event */
2792 ((StgTSO *)bqe)->why_blocked = NotBlocked;
2796 /* if it's a BLOCKED_FETCH put it on the PendingFetches list */
2798 bqe->link = (StgBlockingQueueElement *)PendingFetches;
2799 PendingFetches = (StgBlockedFetch *)bqe;
2803 /* can ignore this case in a non-debugging setup;
2804 see comments on RBHSave closures above */
2806 /* check that the closure is an RBHSave closure */
2807 ASSERT(get_itbl((StgClosure *)bqe) == &stg_RBH_Save_0_info ||
2808 get_itbl((StgClosure *)bqe) == &stg_RBH_Save_1_info ||
2809 get_itbl((StgClosure *)bqe) == &stg_RBH_Save_2_info);
2813 barf("{unblockOneLocked}Daq Qagh: Unexpected IP (%#lx; %s) in blocking queue at %#lx\n",
2814 get_itbl((StgClosure *)bqe), info_type((StgClosure *)bqe),
2818 IF_PAR_DEBUG(bq, fprintf(stderr, ", %p (%s)", bqe, info_type((StgClosure*)bqe)));
2822 #else /* !GRAN && !PAR */
2824 unblockOneLocked(StgTSO *tso)
2828 ASSERT(get_itbl(tso)->type == TSO);
2829 ASSERT(tso->why_blocked != NotBlocked);
2830 tso->why_blocked = NotBlocked;
2832 PUSH_ON_RUN_QUEUE(tso);
2834 IF_DEBUG(scheduler,sched_belch("waking up thread %ld", tso->id));
2839 #if defined(GRAN) || defined(PAR)
2840 inline StgBlockingQueueElement *
2841 unblockOne(StgBlockingQueueElement *bqe, StgClosure *node)
2843 ACQUIRE_LOCK(&sched_mutex);
2844 bqe = unblockOneLocked(bqe, node);
2845 RELEASE_LOCK(&sched_mutex);
2850 unblockOne(StgTSO *tso)
2852 ACQUIRE_LOCK(&sched_mutex);
2853 tso = unblockOneLocked(tso);
2854 RELEASE_LOCK(&sched_mutex);
2861 awakenBlockedQueue(StgBlockingQueueElement *q, StgClosure *node)
2863 StgBlockingQueueElement *bqe;
2868 belch("##-_ AwBQ for node %p on PE %d @ %ld by TSO %d (%p): ", \
2869 node, CurrentProc, CurrentTime[CurrentProc],
2870 CurrentTSO->id, CurrentTSO));
2872 node_loc = where_is(node);
2874 ASSERT(q == END_BQ_QUEUE ||
2875 get_itbl(q)->type == TSO || // q is either a TSO or an RBHSave
2876 get_itbl(q)->type == CONSTR); // closure (type constructor)
2877 ASSERT(is_unique(node));
2879 /* FAKE FETCH: magically copy the node to the tso's proc;
2880 no Fetch necessary because in reality the node should not have been
2881 moved to the other PE in the first place
2883 if (CurrentProc!=node_loc) {
2885 belch("## node %p is on PE %d but CurrentProc is %d (TSO %d); assuming fake fetch and adjusting bitmask (old: %#x)",
2886 node, node_loc, CurrentProc, CurrentTSO->id,
2887 // CurrentTSO, where_is(CurrentTSO),
2888 node->header.gran.procs));
2889 node->header.gran.procs = (node->header.gran.procs) | PE_NUMBER(CurrentProc);
2891 belch("## new bitmask of node %p is %#x",
2892 node, node->header.gran.procs));
2893 if (RtsFlags.GranFlags.GranSimStats.Global) {
2894 globalGranStats.tot_fake_fetches++;
2899 // ToDo: check: ASSERT(CurrentProc==node_loc);
2900 while (get_itbl(bqe)->type==TSO) { // q != END_TSO_QUEUE) {
2903 bqe points to the current element in the queue
2904 next points to the next element in the queue
2906 //tso = (StgTSO *)bqe; // wastes an assignment to get the type right
2907 //tso_loc = where_is(tso);
2909 bqe = unblockOneLocked(bqe, node);
2912 /* if this is the BQ of an RBH, we have to put back the info ripped out of
2913 the closure to make room for the anchor of the BQ */
2914 if (bqe!=END_BQ_QUEUE) {
2915 ASSERT(get_itbl(node)->type == RBH && get_itbl(bqe)->type == CONSTR);
2917 ASSERT((info_ptr==&RBH_Save_0_info) ||
2918 (info_ptr==&RBH_Save_1_info) ||
2919 (info_ptr==&RBH_Save_2_info));
2921 /* cf. convertToRBH in RBH.c for writing the RBHSave closure */
2922 ((StgRBH *)node)->blocking_queue = (StgBlockingQueueElement *)((StgRBHSave *)bqe)->payload[0];
2923 ((StgRBH *)node)->mut_link = (StgMutClosure *)((StgRBHSave *)bqe)->payload[1];
2926 belch("## Filled in RBH_Save for %p (%s) at end of AwBQ",
2927 node, info_type(node)));
2930 /* statistics gathering */
2931 if (RtsFlags.GranFlags.GranSimStats.Global) {
2932 // globalGranStats.tot_bq_processing_time += bq_processing_time;
2933 globalGranStats.tot_bq_len += len; // total length of all bqs awakened
2934 // globalGranStats.tot_bq_len_local += len_local; // same for local TSOs only
2935 globalGranStats.tot_awbq++; // total no. of bqs awakened
2938 fprintf(stderr,"## BQ Stats of %p: [%d entries] %s\n",
2939 node, len, (bqe!=END_BQ_QUEUE) ? "RBH" : ""));
2943 awakenBlockedQueue(StgBlockingQueueElement *q, StgClosure *node)
2945 StgBlockingQueueElement *bqe;
2947 ACQUIRE_LOCK(&sched_mutex);
2949 IF_PAR_DEBUG(verbose,
2950 belch("##-_ AwBQ for node %p on [%x]: ",
2954 if(get_itbl(q)->type == CONSTR || q==END_BQ_QUEUE) {
2955 IF_PAR_DEBUG(verbose, belch("## ... nothing to unblock so lets just return. RFP (BUG?)"));
2960 ASSERT(q == END_BQ_QUEUE ||
2961 get_itbl(q)->type == TSO ||
2962 get_itbl(q)->type == BLOCKED_FETCH ||
2963 get_itbl(q)->type == CONSTR);
2966 while (get_itbl(bqe)->type==TSO ||
2967 get_itbl(bqe)->type==BLOCKED_FETCH) {
2968 bqe = unblockOneLocked(bqe, node);
2970 RELEASE_LOCK(&sched_mutex);
2973 #else /* !GRAN && !PAR */
2975 #ifdef RTS_SUPPORTS_THREADS
2977 awakenBlockedQueueNoLock(StgTSO *tso)
2979 while (tso != END_TSO_QUEUE) {
2980 tso = unblockOneLocked(tso);
2986 awakenBlockedQueue(StgTSO *tso)
2988 ACQUIRE_LOCK(&sched_mutex);
2989 while (tso != END_TSO_QUEUE) {
2990 tso = unblockOneLocked(tso);
2992 RELEASE_LOCK(&sched_mutex);
2996 //@node Exception Handling Routines, Debugging Routines, Blocking Queue Routines, Main scheduling code
2997 //@subsection Exception Handling Routines
2999 /* ---------------------------------------------------------------------------
3001 - usually called inside a signal handler so it mustn't do anything fancy.
3002 ------------------------------------------------------------------------ */
3005 interruptStgRts(void)
3011 /* -----------------------------------------------------------------------------
3014 This is for use when we raise an exception in another thread, which
3016 This has nothing to do with the UnblockThread event in GranSim. -- HWL
3017 -------------------------------------------------------------------------- */
3019 #if defined(GRAN) || defined(PAR)
3021 NB: only the type of the blocking queue is different in GranSim and GUM
3022 the operations on the queue-elements are the same
3023 long live polymorphism!
3025 Locks: sched_mutex is held upon entry and exit.
3029 unblockThread(StgTSO *tso)
3031 StgBlockingQueueElement *t, **last;
3033 switch (tso->why_blocked) {
3036 return; /* not blocked */
3039 ASSERT(get_itbl(tso->block_info.closure)->type == MVAR);
3041 StgBlockingQueueElement *last_tso = END_BQ_QUEUE;
3042 StgMVar *mvar = (StgMVar *)(tso->block_info.closure);
3044 last = (StgBlockingQueueElement **)&mvar->head;
3045 for (t = (StgBlockingQueueElement *)mvar->head;
3047 last = &t->link, last_tso = t, t = t->link) {
3048 if (t == (StgBlockingQueueElement *)tso) {
3049 *last = (StgBlockingQueueElement *)tso->link;
3050 if (mvar->tail == tso) {
3051 mvar->tail = (StgTSO *)last_tso;
3056 barf("unblockThread (MVAR): TSO not found");
3059 case BlockedOnBlackHole:
3060 ASSERT(get_itbl(tso->block_info.closure)->type == BLACKHOLE_BQ);
3062 StgBlockingQueue *bq = (StgBlockingQueue *)(tso->block_info.closure);
3064 last = &bq->blocking_queue;
3065 for (t = bq->blocking_queue;
3067 last = &t->link, t = t->link) {
3068 if (t == (StgBlockingQueueElement *)tso) {
3069 *last = (StgBlockingQueueElement *)tso->link;
3073 barf("unblockThread (BLACKHOLE): TSO not found");
3076 case BlockedOnException:
3078 StgTSO *target = tso->block_info.tso;
3080 ASSERT(get_itbl(target)->type == TSO);
3082 if (target->what_next == ThreadRelocated) {
3083 target = target->link;
3084 ASSERT(get_itbl(target)->type == TSO);
3087 ASSERT(target->blocked_exceptions != NULL);
3089 last = (StgBlockingQueueElement **)&target->blocked_exceptions;
3090 for (t = (StgBlockingQueueElement *)target->blocked_exceptions;
3092 last = &t->link, t = t->link) {
3093 ASSERT(get_itbl(t)->type == TSO);
3094 if (t == (StgBlockingQueueElement *)tso) {
3095 *last = (StgBlockingQueueElement *)tso->link;
3099 barf("unblockThread (Exception): TSO not found");
3103 case BlockedOnWrite:
3105 /* take TSO off blocked_queue */
3106 StgBlockingQueueElement *prev = NULL;
3107 for (t = (StgBlockingQueueElement *)blocked_queue_hd; t != END_BQ_QUEUE;
3108 prev = t, t = t->link) {
3109 if (t == (StgBlockingQueueElement *)tso) {
3111 blocked_queue_hd = (StgTSO *)t->link;
3112 if ((StgBlockingQueueElement *)blocked_queue_tl == t) {
3113 blocked_queue_tl = END_TSO_QUEUE;
3116 prev->link = t->link;
3117 if ((StgBlockingQueueElement *)blocked_queue_tl == t) {
3118 blocked_queue_tl = (StgTSO *)prev;
3124 barf("unblockThread (I/O): TSO not found");
3127 case BlockedOnDelay:
3129 /* take TSO off sleeping_queue */
3130 StgBlockingQueueElement *prev = NULL;
3131 for (t = (StgBlockingQueueElement *)sleeping_queue; t != END_BQ_QUEUE;
3132 prev = t, t = t->link) {
3133 if (t == (StgBlockingQueueElement *)tso) {
3135 sleeping_queue = (StgTSO *)t->link;
3137 prev->link = t->link;
3142 barf("unblockThread (I/O): TSO not found");
3146 barf("unblockThread");
3150 tso->link = END_TSO_QUEUE;
3151 tso->why_blocked = NotBlocked;
3152 tso->block_info.closure = NULL;
3153 PUSH_ON_RUN_QUEUE(tso);
3157 unblockThread(StgTSO *tso)
3161 /* To avoid locking unnecessarily. */
3162 if (tso->why_blocked == NotBlocked) {
3166 switch (tso->why_blocked) {
3169 ASSERT(get_itbl(tso->block_info.closure)->type == MVAR);
3171 StgTSO *last_tso = END_TSO_QUEUE;
3172 StgMVar *mvar = (StgMVar *)(tso->block_info.closure);
3175 for (t = mvar->head; t != END_TSO_QUEUE;
3176 last = &t->link, last_tso = t, t = t->link) {
3179 if (mvar->tail == tso) {
3180 mvar->tail = last_tso;
3185 barf("unblockThread (MVAR): TSO not found");
3188 case BlockedOnBlackHole:
3189 ASSERT(get_itbl(tso->block_info.closure)->type == BLACKHOLE_BQ);
3191 StgBlockingQueue *bq = (StgBlockingQueue *)(tso->block_info.closure);
3193 last = &bq->blocking_queue;
3194 for (t = bq->blocking_queue; t != END_TSO_QUEUE;
3195 last = &t->link, t = t->link) {
3201 barf("unblockThread (BLACKHOLE): TSO not found");
3204 case BlockedOnException:
3206 StgTSO *target = tso->block_info.tso;
3208 ASSERT(get_itbl(target)->type == TSO);
3210 while (target->what_next == ThreadRelocated) {
3211 target = target->link;
3212 ASSERT(get_itbl(target)->type == TSO);
3215 ASSERT(target->blocked_exceptions != NULL);
3217 last = &target->blocked_exceptions;
3218 for (t = target->blocked_exceptions; t != END_TSO_QUEUE;
3219 last = &t->link, t = t->link) {
3220 ASSERT(get_itbl(t)->type == TSO);
3226 barf("unblockThread (Exception): TSO not found");
3230 case BlockedOnWrite:
3232 StgTSO *prev = NULL;
3233 for (t = blocked_queue_hd; t != END_TSO_QUEUE;
3234 prev = t, t = t->link) {
3237 blocked_queue_hd = t->link;
3238 if (blocked_queue_tl == t) {
3239 blocked_queue_tl = END_TSO_QUEUE;
3242 prev->link = t->link;
3243 if (blocked_queue_tl == t) {
3244 blocked_queue_tl = prev;
3250 barf("unblockThread (I/O): TSO not found");
3253 case BlockedOnDelay:
3255 StgTSO *prev = NULL;
3256 for (t = sleeping_queue; t != END_TSO_QUEUE;
3257 prev = t, t = t->link) {
3260 sleeping_queue = t->link;
3262 prev->link = t->link;
3267 barf("unblockThread (I/O): TSO not found");
3271 barf("unblockThread");
3275 tso->link = END_TSO_QUEUE;
3276 tso->why_blocked = NotBlocked;
3277 tso->block_info.closure = NULL;
3278 PUSH_ON_RUN_QUEUE(tso);
3282 /* -----------------------------------------------------------------------------
3285 * The following function implements the magic for raising an
3286 * asynchronous exception in an existing thread.
3288 * We first remove the thread from any queue on which it might be
3289 * blocked. The possible blockages are MVARs and BLACKHOLE_BQs.
3291 * We strip the stack down to the innermost CATCH_FRAME, building
3292 * thunks in the heap for all the active computations, so they can
3293 * be restarted if necessary. When we reach a CATCH_FRAME, we build
3294 * an application of the handler to the exception, and push it on
3295 * the top of the stack.
3297 * How exactly do we save all the active computations? We create an
3298 * AP_STACK for every UpdateFrame on the stack. Entering one of these
3299 * AP_STACKs pushes everything from the corresponding update frame
3300 * upwards onto the stack. (Actually, it pushes everything up to the
3301 * next update frame plus a pointer to the next AP_STACK object.
3302 * Entering the next AP_STACK object pushes more onto the stack until we
3303 * reach the last AP_STACK object - at which point the stack should look
3304 * exactly as it did when we killed the TSO and we can continue
3305 * execution by entering the closure on top of the stack.
3307 * We can also kill a thread entirely - this happens if either (a) the
3308 * exception passed to raiseAsync is NULL, or (b) there's no
3309 * CATCH_FRAME on the stack. In either case, we strip the entire
3310 * stack and replace the thread with a zombie.
3312 * Locks: sched_mutex held upon entry nor exit.
3314 * -------------------------------------------------------------------------- */
3317 deleteThread(StgTSO *tso)
3319 raiseAsync(tso,NULL);
3323 raiseAsyncWithLock(StgTSO *tso, StgClosure *exception)
3325 /* When raising async exs from contexts where sched_mutex isn't held;
3326 use raiseAsyncWithLock(). */
3327 ACQUIRE_LOCK(&sched_mutex);
3328 raiseAsync(tso,exception);
3329 RELEASE_LOCK(&sched_mutex);
3333 raiseAsync(StgTSO *tso, StgClosure *exception)
3335 StgRetInfoTable *info;
3338 // Thread already dead?
3339 if (tso->what_next == ThreadComplete || tso->what_next == ThreadKilled) {
3344 sched_belch("raising exception in thread %ld.", tso->id));
3346 // Remove it from any blocking queues
3351 // The stack freezing code assumes there's a closure pointer on
3352 // the top of the stack, so we have to arrange that this is the case...
3354 if (sp[0] == (W_)&stg_enter_info) {
3358 sp[0] = (W_)&stg_dummy_ret_closure;
3364 // 1. Let the top of the stack be the "current closure"
3366 // 2. Walk up the stack until we find either an UPDATE_FRAME or a
3369 // 3. If it's an UPDATE_FRAME, then make an AP_STACK containing the
3370 // current closure applied to the chunk of stack up to (but not
3371 // including) the update frame. This closure becomes the "current
3372 // closure". Go back to step 2.
3374 // 4. If it's a CATCH_FRAME, then leave the exception handler on
3375 // top of the stack applied to the exception.
3377 // 5. If it's a STOP_FRAME, then kill the thread.
3382 info = get_ret_itbl((StgClosure *)frame);
3384 while (info->i.type != UPDATE_FRAME
3385 && (info->i.type != CATCH_FRAME || exception == NULL)
3386 && info->i.type != STOP_FRAME) {
3387 frame += stack_frame_sizeW((StgClosure *)frame);
3388 info = get_ret_itbl((StgClosure *)frame);
3391 switch (info->i.type) {
3394 // If we find a CATCH_FRAME, and we've got an exception to raise,
3395 // then build the THUNK raise(exception), and leave it on
3396 // top of the CATCH_FRAME ready to enter.
3400 StgCatchFrame *cf = (StgCatchFrame *)frame;
3404 // we've got an exception to raise, so let's pass it to the
3405 // handler in this frame.
3407 raise = (StgClosure *)allocate(sizeofW(StgClosure)+1);
3408 TICK_ALLOC_SE_THK(1,0);
3409 SET_HDR(raise,&stg_raise_info,cf->header.prof.ccs);
3410 raise->payload[0] = exception;
3412 // throw away the stack from Sp up to the CATCH_FRAME.
3416 /* Ensure that async excpetions are blocked now, so we don't get
3417 * a surprise exception before we get around to executing the
3420 if (tso->blocked_exceptions == NULL) {
3421 tso->blocked_exceptions = END_TSO_QUEUE;
3424 /* Put the newly-built THUNK on top of the stack, ready to execute
3425 * when the thread restarts.
3428 sp[-1] = (W_)&stg_enter_info;
3430 tso->what_next = ThreadRunGHC;
3431 IF_DEBUG(sanity, checkTSO(tso));
3440 // First build an AP_STACK consisting of the stack chunk above the
3441 // current update frame, with the top word on the stack as the
3444 words = frame - sp - 1;
3445 ap = (StgAP_STACK *)allocate(PAP_sizeW(words));
3448 ap->fun = (StgClosure *)sp[0];
3450 for(i=0; i < (nat)words; ++i) {
3451 ap->payload[i] = (StgClosure *)*sp++;
3454 SET_HDR(ap,&stg_AP_STACK_info,
3455 ((StgClosure *)frame)->header.prof.ccs /* ToDo */);
3456 TICK_ALLOC_UP_THK(words+1,0);
3459 fprintf(stderr, "scheduler: Updating ");
3460 printPtr((P_)((StgUpdateFrame *)frame)->updatee);
3461 fprintf(stderr, " with ");
3462 printObj((StgClosure *)ap);
3465 // Replace the updatee with an indirection - happily
3466 // this will also wake up any threads currently
3467 // waiting on the result.
3469 // Warning: if we're in a loop, more than one update frame on
3470 // the stack may point to the same object. Be careful not to
3471 // overwrite an IND_OLDGEN in this case, because we'll screw
3472 // up the mutable lists. To be on the safe side, don't
3473 // overwrite any kind of indirection at all. See also
3474 // threadSqueezeStack in GC.c, where we have to make a similar
3477 if (!closure_IND(((StgUpdateFrame *)frame)->updatee)) {
3478 // revert the black hole
3479 UPD_IND_NOLOCK(((StgUpdateFrame *)frame)->updatee,ap);
3481 sp += sizeofW(StgUpdateFrame) - 1;
3482 sp[0] = (W_)ap; // push onto stack
3487 // We've stripped the entire stack, the thread is now dead.
3488 sp += sizeofW(StgStopFrame);
3489 tso->what_next = ThreadKilled;
3500 /* -----------------------------------------------------------------------------
3501 resurrectThreads is called after garbage collection on the list of
3502 threads found to be garbage. Each of these threads will be woken
3503 up and sent a signal: BlockedOnDeadMVar if the thread was blocked
3504 on an MVar, or NonTermination if the thread was blocked on a Black
3507 Locks: sched_mutex isn't held upon entry nor exit.
3508 -------------------------------------------------------------------------- */
3511 resurrectThreads( StgTSO *threads )
3515 for (tso = threads; tso != END_TSO_QUEUE; tso = next) {
3516 next = tso->global_link;
3517 tso->global_link = all_threads;
3519 IF_DEBUG(scheduler, sched_belch("resurrecting thread %d", tso->id));
3521 switch (tso->why_blocked) {
3523 case BlockedOnException:
3524 /* Called by GC - sched_mutex lock is currently held. */
3525 raiseAsync(tso,(StgClosure *)BlockedOnDeadMVar_closure);
3527 case BlockedOnBlackHole:
3528 raiseAsync(tso,(StgClosure *)NonTermination_closure);
3531 /* This might happen if the thread was blocked on a black hole
3532 * belonging to a thread that we've just woken up (raiseAsync
3533 * can wake up threads, remember...).
3537 barf("resurrectThreads: thread blocked in a strange way");
3542 /* -----------------------------------------------------------------------------
3543 * Blackhole detection: if we reach a deadlock, test whether any
3544 * threads are blocked on themselves. Any threads which are found to
3545 * be self-blocked get sent a NonTermination exception.
3547 * This is only done in a deadlock situation in order to avoid
3548 * performance overhead in the normal case.
3550 * Locks: sched_mutex is held upon entry and exit.
3551 * -------------------------------------------------------------------------- */
3554 detectBlackHoles( void )
3556 StgTSO *tso = all_threads;
3558 StgClosure *blocked_on;
3559 StgRetInfoTable *info;
3561 for (tso = all_threads; tso != END_TSO_QUEUE; tso = tso->global_link) {
3563 while (tso->what_next == ThreadRelocated) {
3565 ASSERT(get_itbl(tso)->type == TSO);
3568 if (tso->why_blocked != BlockedOnBlackHole) {
3571 blocked_on = tso->block_info.closure;
3573 frame = (StgClosure *)tso->sp;
3576 info = get_ret_itbl(frame);
3577 switch (info->i.type) {
3579 if (((StgUpdateFrame *)frame)->updatee == blocked_on) {
3580 /* We are blocking on one of our own computations, so
3581 * send this thread the NonTermination exception.
3584 sched_belch("thread %d is blocked on itself", tso->id));
3585 raiseAsync(tso, (StgClosure *)NonTermination_closure);
3589 frame = (StgClosure *) ((StgUpdateFrame *)frame + 1);
3595 // normal stack frames; do nothing except advance the pointer
3597 (StgPtr)frame += stack_frame_sizeW(frame);
3604 //@node Debugging Routines, Index, Exception Handling Routines, Main scheduling code
3605 //@subsection Debugging Routines
3607 /* -----------------------------------------------------------------------------
3608 * Debugging: why is a thread blocked
3609 * [Also provides useful information when debugging threaded programs
3610 * at the Haskell source code level, so enable outside of DEBUG. --sof 7/02]
3611 -------------------------------------------------------------------------- */
3615 printThreadBlockage(StgTSO *tso)
3617 switch (tso->why_blocked) {
3619 fprintf(stderr,"is blocked on read from fd %d", tso->block_info.fd);
3621 case BlockedOnWrite:
3622 fprintf(stderr,"is blocked on write to fd %d", tso->block_info.fd);
3624 case BlockedOnDelay:
3625 fprintf(stderr,"is blocked until %d", tso->block_info.target);
3628 fprintf(stderr,"is blocked on an MVar");
3630 case BlockedOnException:
3631 fprintf(stderr,"is blocked on delivering an exception to thread %d",
3632 tso->block_info.tso->id);
3634 case BlockedOnBlackHole:
3635 fprintf(stderr,"is blocked on a black hole");
3638 fprintf(stderr,"is not blocked");
3642 fprintf(stderr,"is blocked on global address; local FM_BQ is %p (%s)",
3643 tso->block_info.closure, info_type(tso->block_info.closure));
3645 case BlockedOnGA_NoSend:
3646 fprintf(stderr,"is blocked on global address (no send); local FM_BQ is %p (%s)",
3647 tso->block_info.closure, info_type(tso->block_info.closure));
3650 #if defined(RTS_SUPPORTS_THREADS)
3651 case BlockedOnCCall:
3652 fprintf(stderr,"is blocked on an external call");
3654 case BlockedOnCCall_NoUnblockExc:
3655 fprintf(stderr,"is blocked on an external call (exceptions were already blocked)");
3659 barf("printThreadBlockage: strange tso->why_blocked: %d for TSO %d (%d)",
3660 tso->why_blocked, tso->id, tso);
3666 printThreadStatus(StgTSO *tso)
3668 switch (tso->what_next) {
3670 fprintf(stderr,"has been killed");
3672 case ThreadComplete:
3673 fprintf(stderr,"has completed");
3676 printThreadBlockage(tso);
3681 printAllThreads(void)
3687 char time_string[TIME_STR_LEN], node_str[NODE_STR_LEN];
3688 ullong_format_string(TIME_ON_PROC(CurrentProc),
3689 time_string, rtsFalse/*no commas!*/);
3691 fprintf(stderr, "all threads at [%s]:\n", time_string);
3693 char time_string[TIME_STR_LEN], node_str[NODE_STR_LEN];
3694 ullong_format_string(CURRENT_TIME,
3695 time_string, rtsFalse/*no commas!*/);
3697 fprintf(stderr,"all threads at [%s]:\n", time_string);
3699 fprintf(stderr,"all threads:\n");
3702 for (t = all_threads; t != END_TSO_QUEUE; t = t->global_link) {
3703 fprintf(stderr, "\tthread %d @ %p ", t->id, (void *)t);
3704 label = lookupThreadLabel((StgWord)t);
3705 if (label) fprintf(stderr,"[\"%s\"] ",(char *)label);
3706 printThreadStatus(t);
3707 fprintf(stderr,"\n");
3714 Print a whole blocking queue attached to node (debugging only).
3719 print_bq (StgClosure *node)
3721 StgBlockingQueueElement *bqe;
3725 fprintf(stderr,"## BQ of closure %p (%s): ",
3726 node, info_type(node));
3728 /* should cover all closures that may have a blocking queue */
3729 ASSERT(get_itbl(node)->type == BLACKHOLE_BQ ||
3730 get_itbl(node)->type == FETCH_ME_BQ ||
3731 get_itbl(node)->type == RBH ||
3732 get_itbl(node)->type == MVAR);
3734 ASSERT(node!=(StgClosure*)NULL); // sanity check
3736 print_bqe(((StgBlockingQueue*)node)->blocking_queue);
3740 Print a whole blocking queue starting with the element bqe.
3743 print_bqe (StgBlockingQueueElement *bqe)
3748 NB: In a parallel setup a BQ of an RBH must end with an RBH_Save closure;
3750 for (end = (bqe==END_BQ_QUEUE);
3751 !end; // iterate until bqe points to a CONSTR
3752 end = (get_itbl(bqe)->type == CONSTR) || (bqe->link==END_BQ_QUEUE),
3753 bqe = end ? END_BQ_QUEUE : bqe->link) {
3754 ASSERT(bqe != END_BQ_QUEUE); // sanity check
3755 ASSERT(bqe != (StgBlockingQueueElement *)NULL); // sanity check
3756 /* types of closures that may appear in a blocking queue */
3757 ASSERT(get_itbl(bqe)->type == TSO ||
3758 get_itbl(bqe)->type == BLOCKED_FETCH ||
3759 get_itbl(bqe)->type == CONSTR);
3760 /* only BQs of an RBH end with an RBH_Save closure */
3761 //ASSERT(get_itbl(bqe)->type != CONSTR || get_itbl(node)->type == RBH);
3763 switch (get_itbl(bqe)->type) {
3765 fprintf(stderr," TSO %u (%x),",
3766 ((StgTSO *)bqe)->id, ((StgTSO *)bqe));
3769 fprintf(stderr," BF (node=%p, ga=((%x, %d, %x)),",
3770 ((StgBlockedFetch *)bqe)->node,
3771 ((StgBlockedFetch *)bqe)->ga.payload.gc.gtid,
3772 ((StgBlockedFetch *)bqe)->ga.payload.gc.slot,
3773 ((StgBlockedFetch *)bqe)->ga.weight);
3776 fprintf(stderr," %s (IP %p),",
3777 (get_itbl(bqe) == &stg_RBH_Save_0_info ? "RBH_Save_0" :
3778 get_itbl(bqe) == &stg_RBH_Save_1_info ? "RBH_Save_1" :
3779 get_itbl(bqe) == &stg_RBH_Save_2_info ? "RBH_Save_2" :
3780 "RBH_Save_?"), get_itbl(bqe));
3783 barf("Unexpected closure type %s in blocking queue", // of %p (%s)",
3784 info_type((StgClosure *)bqe)); // , node, info_type(node));
3788 fputc('\n', stderr);
3790 # elif defined(GRAN)
3792 print_bq (StgClosure *node)
3794 StgBlockingQueueElement *bqe;
3795 PEs node_loc, tso_loc;
3798 /* should cover all closures that may have a blocking queue */
3799 ASSERT(get_itbl(node)->type == BLACKHOLE_BQ ||
3800 get_itbl(node)->type == FETCH_ME_BQ ||
3801 get_itbl(node)->type == RBH);
3803 ASSERT(node!=(StgClosure*)NULL); // sanity check
3804 node_loc = where_is(node);
3806 fprintf(stderr,"## BQ of closure %p (%s) on [PE %d]: ",
3807 node, info_type(node), node_loc);
3810 NB: In a parallel setup a BQ of an RBH must end with an RBH_Save closure;
3812 for (bqe = ((StgBlockingQueue*)node)->blocking_queue, end = (bqe==END_BQ_QUEUE);
3813 !end; // iterate until bqe points to a CONSTR
3814 end = (get_itbl(bqe)->type == CONSTR) || (bqe->link==END_BQ_QUEUE), bqe = end ? END_BQ_QUEUE : bqe->link) {
3815 ASSERT(bqe != END_BQ_QUEUE); // sanity check
3816 ASSERT(bqe != (StgBlockingQueueElement *)NULL); // sanity check
3817 /* types of closures that may appear in a blocking queue */
3818 ASSERT(get_itbl(bqe)->type == TSO ||
3819 get_itbl(bqe)->type == CONSTR);
3820 /* only BQs of an RBH end with an RBH_Save closure */
3821 ASSERT(get_itbl(bqe)->type != CONSTR || get_itbl(node)->type == RBH);
3823 tso_loc = where_is((StgClosure *)bqe);
3824 switch (get_itbl(bqe)->type) {
3826 fprintf(stderr," TSO %d (%p) on [PE %d],",
3827 ((StgTSO *)bqe)->id, (StgTSO *)bqe, tso_loc);
3830 fprintf(stderr," %s (IP %p),",
3831 (get_itbl(bqe) == &stg_RBH_Save_0_info ? "RBH_Save_0" :
3832 get_itbl(bqe) == &stg_RBH_Save_1_info ? "RBH_Save_1" :
3833 get_itbl(bqe) == &stg_RBH_Save_2_info ? "RBH_Save_2" :
3834 "RBH_Save_?"), get_itbl(bqe));
3837 barf("Unexpected closure type %s in blocking queue of %p (%s)",
3838 info_type((StgClosure *)bqe), node, info_type(node));
3842 fputc('\n', stderr);
3846 Nice and easy: only TSOs on the blocking queue
3849 print_bq (StgClosure *node)
3853 ASSERT(node!=(StgClosure*)NULL); // sanity check
3854 for (tso = ((StgBlockingQueue*)node)->blocking_queue;
3855 tso != END_TSO_QUEUE;
3857 ASSERT(tso!=NULL && tso!=END_TSO_QUEUE); // sanity check
3858 ASSERT(get_itbl(tso)->type == TSO); // guess what, sanity check
3859 fprintf(stderr," TSO %d (%p),", tso->id, tso);
3861 fputc('\n', stderr);
3872 for (i=0, tso=run_queue_hd;
3873 tso != END_TSO_QUEUE;
3882 sched_belch(char *s, ...)
3887 fprintf(stderr, "scheduler (task %ld): ", osThreadId());
3889 fprintf(stderr, "== ");
3891 fprintf(stderr, "scheduler: ");
3893 vfprintf(stderr, s, ap);
3894 fprintf(stderr, "\n");
3901 //@node Index, , Debugging Routines, Main scheduling code
3905 //* StgMainThread:: @cindex\s-+StgMainThread
3906 //* awaken_blocked_queue:: @cindex\s-+awaken_blocked_queue
3907 //* blocked_queue_hd:: @cindex\s-+blocked_queue_hd
3908 //* blocked_queue_tl:: @cindex\s-+blocked_queue_tl
3909 //* context_switch:: @cindex\s-+context_switch
3910 //* createThread:: @cindex\s-+createThread
3911 //* gc_pending_cond:: @cindex\s-+gc_pending_cond
3912 //* initScheduler:: @cindex\s-+initScheduler
3913 //* interrupted:: @cindex\s-+interrupted
3914 //* next_thread_id:: @cindex\s-+next_thread_id
3915 //* print_bq:: @cindex\s-+print_bq
3916 //* run_queue_hd:: @cindex\s-+run_queue_hd
3917 //* run_queue_tl:: @cindex\s-+run_queue_tl
3918 //* sched_mutex:: @cindex\s-+sched_mutex
3919 //* schedule:: @cindex\s-+schedule
3920 //* take_off_run_queue:: @cindex\s-+take_off_run_queue
3921 //* term_mutex:: @cindex\s-+term_mutex