1 /* ---------------------------------------------------------------------------
2 * $Id: Schedule.c,v 1.115 2002/02/04 20:40:36 sof 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"
88 #include "StgMiscClosures.h"
90 #include "Interpreter.h"
91 #include "Exception.h"
100 #include "Proftimer.h"
101 #include "ProfHeap.h"
103 #if defined(GRAN) || defined(PAR)
104 # include "GranSimRts.h"
105 # include "GranSim.h"
106 # include "ParallelRts.h"
107 # include "Parallel.h"
108 # include "ParallelDebug.h"
109 # include "FetchMe.h"
113 #include "Capability.h"
114 #include "OSThreads.h"
119 //@node Variables and Data structures, Prototypes, Includes, Main scheduling code
120 //@subsection Variables and Data structures
124 * These are the threads which clients have requested that we run.
126 * In a 'threaded' build, we might have several concurrent clients all
127 * waiting for results, and each one will wait on a condition variable
128 * until the result is available.
130 * In non-SMP, clients are strictly nested: the first client calls
131 * into the RTS, which might call out again to C with a _ccall_GC, and
132 * eventually re-enter the RTS.
134 * Main threads information is kept in a linked list:
136 //@cindex StgMainThread
137 typedef struct StgMainThread_ {
139 SchedulerStatus stat;
141 #if defined(RTS_SUPPORTS_THREADS)
144 struct StgMainThread_ *link;
147 /* Main thread queue.
148 * Locks required: sched_mutex.
150 static StgMainThread *main_threads;
153 * Locks required: sched_mutex.
157 StgTSO* ActiveTSO = NULL; /* for assigning system costs; GranSim-Light only */
158 /* rtsTime TimeOfNextEvent, EndOfTimeSlice; now in GranSim.c */
161 In GranSim we have a runable and a blocked queue for each processor.
162 In order to minimise code changes new arrays run_queue_hds/tls
163 are created. run_queue_hd is then a short cut (macro) for
164 run_queue_hds[CurrentProc] (see GranSim.h).
167 StgTSO *run_queue_hds[MAX_PROC], *run_queue_tls[MAX_PROC];
168 StgTSO *blocked_queue_hds[MAX_PROC], *blocked_queue_tls[MAX_PROC];
169 StgTSO *ccalling_threadss[MAX_PROC];
170 /* We use the same global list of threads (all_threads) in GranSim as in
171 the std RTS (i.e. we are cheating). However, we don't use this list in
172 the GranSim specific code at the moment (so we are only potentially
177 StgTSO *run_queue_hd, *run_queue_tl;
178 StgTSO *blocked_queue_hd, *blocked_queue_tl;
179 StgTSO *sleeping_queue; /* perhaps replace with a hash table? */
183 /* Linked list of all threads.
184 * Used for detecting garbage collected threads.
188 /* Threads suspended in _ccall_GC.
190 static StgTSO *suspended_ccalling_threads;
192 static StgTSO *threadStackOverflow(StgTSO *tso);
194 /* KH: The following two flags are shared memory locations. There is no need
195 to lock them, since they are only unset at the end of a scheduler
199 /* flag set by signal handler to precipitate a context switch */
200 //@cindex context_switch
203 /* if this flag is set as well, give up execution */
204 //@cindex interrupted
207 /* Next thread ID to allocate.
208 * Locks required: sched_mutex
210 //@cindex next_thread_id
211 StgThreadID next_thread_id = 1;
214 * Pointers to the state of the current thread.
215 * Rule of thumb: if CurrentTSO != NULL, then we're running a Haskell
216 * thread. If CurrentTSO == NULL, then we're at the scheduler level.
219 /* The smallest stack size that makes any sense is:
220 * RESERVED_STACK_WORDS (so we can get back from the stack overflow)
221 * + sizeofW(StgStopFrame) (the stg_stop_thread_info frame)
222 * + 1 (the realworld token for an IO thread)
223 * + 1 (the closure to enter)
225 * A thread with this stack will bomb immediately with a stack
226 * overflow, which will increase its stack size.
229 #define MIN_STACK_WORDS (RESERVED_STACK_WORDS + sizeofW(StgStopFrame) + 2)
236 /* This is used in `TSO.h' and gcc 2.96 insists that this variable actually
237 * exists - earlier gccs apparently didn't.
244 void addToBlockedQueue ( StgTSO *tso );
246 static void schedule ( void );
247 void interruptStgRts ( void );
249 static StgTSO * createThread_ ( nat size, rtsBool have_lock, StgInt pri );
251 static StgTSO * createThread_ ( nat size, rtsBool have_lock );
254 static void detectBlackHoles ( void );
257 static void sched_belch(char *s, ...);
260 #if defined(RTS_SUPPORTS_THREADS)
261 /* ToDo: carefully document the invariants that go together
262 * with these synchronisation objects.
264 Mutex sched_mutex = INIT_MUTEX_VAR;
265 Mutex term_mutex = INIT_MUTEX_VAR;
266 #if defined(THREADED_RTS)
268 * The rts_mutex is the 'big lock' that the active native
269 * thread within the RTS holds while executing code
270 * within the RTS. It is given up when the thread makes a
271 * transition out of the RTS (e.g., to perform an external
272 * C call), hopefully for another thread to enter the RTS.
275 Mutex rts_mutex = INIT_MUTEX_VAR;
277 * When a native thread has completed executing an external
278 * call, it needs to communicate the result back to the
279 * (Haskell) thread that made the call. Do this as follows:
281 * - in resumeThread(), the thread increments the counter
282 * ext_threads_waiting, and then blocks on the
284 * - upon entry to the scheduler, the thread that's currently
285 * holding the RTS lock checks ext_threads_waiting. If there
286 * are native threads waiting, it gives up its RTS lock
287 * and tries to re-grab the RTS lock [perhaps after having
288 * waited for a bit..?]
289 * - care must be taken to deal with the case where more than
290 * one external thread are waiting on the lock. [ToDo: more]
294 static nat ext_threads_waiting = 0;
296 * thread_ready_aux_mutex is used to handle the scenario where the
297 * the RTS executing thread runs out of work, but there are
298 * active external threads. The RTS executing thread gives up
299 * its RTS mutex, and blocks waiting for the thread_ready_cond.
300 * Unfortunately, a condition variable needs to be associated
301 * with a mutex in pthreads, so rts_thread_waiting_mutex is
302 * used for just this purpose.
305 Mutex thread_ready_aux_mutex = INIT_MUTEX_VAR;
309 /* thread_ready_cond: when signalled, a thread has
310 * become runnable. When used?
312 Condition thread_ready_cond = INIT_COND_VAR;
313 Condition gc_pending_cond = INIT_COND_VAR;
320 rtsTime TimeOfLastYield;
321 rtsBool emitSchedule = rtsTrue;
325 char *whatNext_strs[] = {
333 char *threadReturnCode_strs[] = {
334 "HeapOverflow", /* might also be StackOverflow */
343 StgTSO * createSparkThread(rtsSpark spark);
344 StgTSO * activateSpark (rtsSpark spark);
348 * The thread state for the main thread.
349 // ToDo: check whether not needed any more
353 #if defined(PAR) || defined(RTS_SUPPORTS_THREADS)
354 static void taskStart(void);
358 /* threads start up using 'taskStart', so make them
359 them grab the RTS lock. */
360 #if defined(THREADED_RTS)
361 ACQUIRE_LOCK(&rts_mutex);
370 //@node Main scheduling loop, Suspend and Resume, Prototypes, Main scheduling code
371 //@subsection Main scheduling loop
373 /* ---------------------------------------------------------------------------
374 Main scheduling loop.
376 We use round-robin scheduling, each thread returning to the
377 scheduler loop when one of these conditions is detected:
380 * timer expires (thread yields)
385 Locking notes: we acquire the scheduler lock once at the beginning
386 of the scheduler loop, and release it when
388 * running a thread, or
389 * waiting for work, or
390 * waiting for a GC to complete.
393 In a GranSim setup this loop iterates over the global event queue.
394 This revolves around the global event queue, which determines what
395 to do next. Therefore, it's more complicated than either the
396 concurrent or the parallel (GUM) setup.
399 GUM iterates over incoming messages.
400 It starts with nothing to do (thus CurrentTSO == END_TSO_QUEUE),
401 and sends out a fish whenever it has nothing to do; in-between
402 doing the actual reductions (shared code below) it processes the
403 incoming messages and deals with delayed operations
404 (see PendingFetches).
405 This is not the ugliest code you could imagine, but it's bloody close.
407 ------------------------------------------------------------------------ */
414 StgThreadReturnCode ret;
422 rtsBool receivedFinish = rtsFalse;
424 nat tp_size, sp_size; // stats only
427 rtsBool was_interrupted = rtsFalse;
429 ACQUIRE_LOCK(&sched_mutex);
431 #if defined(THREADED_RTS)
432 /* ToDo: consider SMP support */
433 if (ext_threads_waiting > 0) {
434 /* (At least) one external thread is waiting to
435 * to deposit the result of an external call.
436 * Give way to one of them by giving up the RTS
439 RELEASE_LOCK(&sched_mutex);
440 RELEASE_LOCK(&rts_mutex);
441 /* ToDo: come up with mechanism that guarantees that
442 * the main thread doesn't loop here.
445 /* ToDo: longjmp() */
452 /* set up first event to get things going */
453 /* ToDo: assign costs for system setup and init MainTSO ! */
454 new_event(CurrentProc, CurrentProc, CurrentTime[CurrentProc],
456 CurrentTSO, (StgClosure*)NULL, (rtsSpark*)NULL);
459 fprintf(stderr, "GRAN: Init CurrentTSO (in schedule) = %p\n", CurrentTSO);
460 G_TSO(CurrentTSO, 5));
462 if (RtsFlags.GranFlags.Light) {
463 /* Save current time; GranSim Light only */
464 CurrentTSO->gran.clock = CurrentTime[CurrentProc];
467 event = get_next_event();
469 while (event!=(rtsEvent*)NULL) {
470 /* Choose the processor with the next event */
471 CurrentProc = event->proc;
472 CurrentTSO = event->tso;
476 while (!receivedFinish) { /* set by processMessages */
477 /* when receiving PP_FINISH message */
484 IF_DEBUG(scheduler, printAllThreads());
486 /* If we're interrupted (the user pressed ^C, or some other
487 * termination condition occurred), kill all the currently running
491 IF_DEBUG(scheduler, sched_belch("interrupted"));
493 interrupted = rtsFalse;
494 was_interrupted = rtsTrue;
497 /* Go through the list of main threads and wake up any
498 * clients whose computations have finished. ToDo: this
499 * should be done more efficiently without a linear scan
500 * of the main threads list, somehow...
502 #if defined(RTS_SUPPORTS_THREADS)
504 StgMainThread *m, **prev;
505 prev = &main_threads;
506 for (m = main_threads; m != NULL; m = m->link) {
507 switch (m->tso->what_next) {
510 *(m->ret) = (StgClosure *)m->tso->sp[0];
514 broadcastCondition(&m->wakeup);
517 if (m->ret) *(m->ret) = NULL;
519 if (was_interrupted) {
520 m->stat = Interrupted;
524 broadcastCondition(&m->wakeup);
532 #else /* not threaded */
535 /* in GUM do this only on the Main PE */
538 /* If our main thread has finished or been killed, return.
541 StgMainThread *m = main_threads;
542 if (m->tso->what_next == ThreadComplete
543 || m->tso->what_next == ThreadKilled) {
544 main_threads = main_threads->link;
545 if (m->tso->what_next == ThreadComplete) {
546 /* we finished successfully, fill in the return value */
547 if (m->ret) { *(m->ret) = (StgClosure *)m->tso->sp[0]; };
551 if (m->ret) { *(m->ret) = NULL; };
552 if (was_interrupted) {
553 m->stat = Interrupted;
563 /* Top up the run queue from our spark pool. We try to make the
564 * number of threads in the run queue equal to the number of
567 * Disable spark support in SMP for now, non-essential & requires
568 * a little bit of work to make it compile cleanly. -- sof 1/02.
570 #if 0 /* defined(SMP) */
572 nat n = getFreeCapabilities();
573 StgTSO *tso = run_queue_hd;
575 /* Count the run queue */
576 while (n > 0 && tso != END_TSO_QUEUE) {
583 spark = findSpark(rtsFalse);
585 break; /* no more sparks in the pool */
587 /* I'd prefer this to be done in activateSpark -- HWL */
588 /* tricky - it needs to hold the scheduler lock and
589 * not try to re-acquire it -- SDM */
590 createSparkThread(spark);
592 sched_belch("==^^ turning spark of closure %p into a thread",
593 (StgClosure *)spark));
596 /* We need to wake up the other tasks if we just created some
599 if (getFreeCapabilities() - n > 1) {
600 signalCondition( &thread_ready_cond );
605 /* check for signals each time around the scheduler */
606 #ifndef mingw32_TARGET_OS
607 if (signals_pending()) {
608 startSignalHandlers();
612 /* Check whether any waiting threads need to be woken up. If the
613 * run queue is empty, and there are no other tasks running, we
614 * can wait indefinitely for something to happen.
615 * ToDo: what if another client comes along & requests another
618 if (blocked_queue_hd != END_TSO_QUEUE || sleeping_queue != END_TSO_QUEUE) {
620 (run_queue_hd == END_TSO_QUEUE)
622 && allFreeCapabilities()
626 /* we can be interrupted while waiting for I/O... */
627 if (interrupted) continue;
630 * Detect deadlock: when we have no threads to run, there are no
631 * threads waiting on I/O or sleeping, and all the other tasks are
632 * waiting for work, we must have a deadlock of some description.
634 * We first try to find threads blocked on themselves (ie. black
635 * holes), and generate NonTermination exceptions where necessary.
637 * If no threads are black holed, we have a deadlock situation, so
638 * inform all the main threads.
641 if (blocked_queue_hd == END_TSO_QUEUE
642 && run_queue_hd == END_TSO_QUEUE
643 && sleeping_queue == END_TSO_QUEUE
645 && allFreeCapabilities()
646 #elif defined(THREADED_RTS)
647 && suspended_ccalling_threads == END_TSO_QUEUE
651 IF_DEBUG(scheduler, sched_belch("deadlocked, forcing major GC..."));
652 RELEASE_LOCK(&sched_mutex);
653 GarbageCollect(GetRoots,rtsTrue);
654 ACQUIRE_LOCK(&sched_mutex);
655 IF_DEBUG(scheduler, sched_belch("GC done."));
656 if (blocked_queue_hd == END_TSO_QUEUE
657 && run_queue_hd == END_TSO_QUEUE
658 && sleeping_queue == END_TSO_QUEUE) {
660 IF_DEBUG(scheduler, sched_belch("still deadlocked, checking for black holes..."));
663 /* No black holes, so probably a real deadlock. Send the
664 * current main thread the Deadlock exception (or in the SMP
665 * build, send *all* main threads the deadlock exception,
666 * since none of them can make progress).
668 if (run_queue_hd == END_TSO_QUEUE) {
670 #if defined(RTS_SUPPORTS_THREADS)
671 for (m = main_threads; m != NULL; m = m->link) {
672 switch (m->tso->why_blocked) {
673 case BlockedOnBlackHole:
674 raiseAsync(m->tso, (StgClosure *)NonTermination_closure);
676 case BlockedOnException:
678 raiseAsync(m->tso, (StgClosure *)Deadlock_closure);
681 barf("deadlock: main thread blocked in a strange way");
686 switch (m->tso->why_blocked) {
687 case BlockedOnBlackHole:
688 raiseAsync(m->tso, (StgClosure *)NonTermination_closure);
690 case BlockedOnException:
692 raiseAsync(m->tso, (StgClosure *)Deadlock_closure);
695 barf("deadlock: main thread blocked in a strange way");
699 #if defined(RTS_SUPPORTS_THREADS)
700 if ( run_queue_hd == END_TSO_QUEUE ) {
701 IF_DEBUG(scheduler, sched_belch("all done, it seems...shut down."));
702 shutdownHaskellAndExit(0);
706 ASSERT( run_queue_hd != END_TSO_QUEUE );
710 /* ToDo: add deadlock detection in GUM (similar to SMP) -- HWL */
714 /* If there's a GC pending, don't do anything until it has
718 IF_DEBUG(scheduler,sched_belch("waiting for GC"));
719 waitCondition( &gc_pending_cond, &sched_mutex );
724 /* block until we've got a thread on the run queue and a free
727 while ( run_queue_hd == END_TSO_QUEUE
728 || noFreeCapabilities()
730 IF_DEBUG(scheduler, sched_belch("waiting for work"));
731 waitCondition( &thread_ready_cond, &sched_mutex );
732 IF_DEBUG(scheduler, sched_belch("work now available"));
734 #elif defined(THREADED_RTS)
735 if ( run_queue_hd == END_TSO_QUEUE ) {
736 /* no work available, wait for external calls to complete. */
737 IF_DEBUG(scheduler, sched_belch("worker thread (%d): waiting for external thread to complete..", osThreadId()));
738 RELEASE_LOCK(&sched_mutex);
739 RELEASE_LOCK(&rts_mutex);
740 /* Sigh - need to have a mutex locked in order to wait on the
741 condition variable. */
742 ACQUIRE_LOCK(&thread_ready_aux_mutex);
743 waitCondition(&thread_ready_cond, &thread_ready_aux_mutex);
744 RELEASE_LOCK(&thread_ready_aux_mutex);
745 IF_DEBUG(scheduler, sched_belch("worker thread (%d): re-awakened from no-work slumber..\n", osThreadId()));
746 /* ToDo: longjmp() */
754 if (RtsFlags.GranFlags.Light)
755 GranSimLight_enter_system(event, &ActiveTSO); // adjust ActiveTSO etc
757 /* adjust time based on time-stamp */
758 if (event->time > CurrentTime[CurrentProc] &&
759 event->evttype != ContinueThread)
760 CurrentTime[CurrentProc] = event->time;
762 /* Deal with the idle PEs (may issue FindWork or MoveSpark events) */
763 if (!RtsFlags.GranFlags.Light)
766 IF_DEBUG(gran, fprintf(stderr, "GRAN: switch by event-type\n"));
768 /* main event dispatcher in GranSim */
769 switch (event->evttype) {
770 /* Should just be continuing execution */
772 IF_DEBUG(gran, fprintf(stderr, "GRAN: doing ContinueThread\n"));
773 /* ToDo: check assertion
774 ASSERT(run_queue_hd != (StgTSO*)NULL &&
775 run_queue_hd != END_TSO_QUEUE);
777 /* Ignore ContinueThreads for fetching threads (if synchr comm) */
778 if (!RtsFlags.GranFlags.DoAsyncFetch &&
779 procStatus[CurrentProc]==Fetching) {
780 belch("ghuH: Spurious ContinueThread while Fetching ignored; TSO %d (%p) [PE %d]",
781 CurrentTSO->id, CurrentTSO, CurrentProc);
784 /* Ignore ContinueThreads for completed threads */
785 if (CurrentTSO->what_next == ThreadComplete) {
786 belch("ghuH: found a ContinueThread event for completed thread %d (%p) [PE %d] (ignoring ContinueThread)",
787 CurrentTSO->id, CurrentTSO, CurrentProc);
790 /* Ignore ContinueThreads for threads that are being migrated */
791 if (PROCS(CurrentTSO)==Nowhere) {
792 belch("ghuH: trying to run the migrating TSO %d (%p) [PE %d] (ignoring ContinueThread)",
793 CurrentTSO->id, CurrentTSO, CurrentProc);
796 /* The thread should be at the beginning of the run queue */
797 if (CurrentTSO!=run_queue_hds[CurrentProc]) {
798 belch("ghuH: TSO %d (%p) [PE %d] is not at the start of the run_queue when doing a ContinueThread",
799 CurrentTSO->id, CurrentTSO, CurrentProc);
800 break; // run the thread anyway
803 new_event(proc, proc, CurrentTime[proc],
805 (StgTSO*)NULL, (StgClosure*)NULL, (rtsSpark*)NULL);
807 */ /* Catches superfluous CONTINUEs -- should be unnecessary */
808 break; // now actually run the thread; DaH Qu'vam yImuHbej
811 do_the_fetchnode(event);
812 goto next_thread; /* handle next event in event queue */
815 do_the_globalblock(event);
816 goto next_thread; /* handle next event in event queue */
819 do_the_fetchreply(event);
820 goto next_thread; /* handle next event in event queue */
822 case UnblockThread: /* Move from the blocked queue to the tail of */
823 do_the_unblock(event);
824 goto next_thread; /* handle next event in event queue */
826 case ResumeThread: /* Move from the blocked queue to the tail of */
827 /* the runnable queue ( i.e. Qu' SImqa'lu') */
828 event->tso->gran.blocktime +=
829 CurrentTime[CurrentProc] - event->tso->gran.blockedat;
830 do_the_startthread(event);
831 goto next_thread; /* handle next event in event queue */
834 do_the_startthread(event);
835 goto next_thread; /* handle next event in event queue */
838 do_the_movethread(event);
839 goto next_thread; /* handle next event in event queue */
842 do_the_movespark(event);
843 goto next_thread; /* handle next event in event queue */
846 do_the_findwork(event);
847 goto next_thread; /* handle next event in event queue */
850 barf("Illegal event type %u\n", event->evttype);
853 /* This point was scheduler_loop in the old RTS */
855 IF_DEBUG(gran, belch("GRAN: after main switch"));
857 TimeOfLastEvent = CurrentTime[CurrentProc];
858 TimeOfNextEvent = get_time_of_next_event();
859 IgnoreEvents=(TimeOfNextEvent==0); // HWL HACK
860 // CurrentTSO = ThreadQueueHd;
862 IF_DEBUG(gran, belch("GRAN: time of next event is: %ld",
865 if (RtsFlags.GranFlags.Light)
866 GranSimLight_leave_system(event, &ActiveTSO);
868 EndOfTimeSlice = CurrentTime[CurrentProc]+RtsFlags.GranFlags.time_slice;
871 belch("GRAN: end of time-slice is %#lx", EndOfTimeSlice));
873 /* in a GranSim setup the TSO stays on the run queue */
875 /* Take a thread from the run queue. */
876 t = POP_RUN_QUEUE(); // take_off_run_queue(t);
879 fprintf(stderr, "GRAN: About to run current thread, which is\n");
882 context_switch = 0; // turned on via GranYield, checking events and time slice
885 DumpGranEvent(GR_SCHEDULE, t));
887 procStatus[CurrentProc] = Busy;
890 if (PendingFetches != END_BF_QUEUE) {
894 /* ToDo: phps merge with spark activation above */
895 /* check whether we have local work and send requests if we have none */
896 if (EMPTY_RUN_QUEUE()) { /* no runnable threads */
897 /* :-[ no local threads => look out for local sparks */
898 /* the spark pool for the current PE */
899 pool = &(MainRegTable.rSparks); // generalise to cap = &MainRegTable
900 if (advisory_thread_count < RtsFlags.ParFlags.maxThreads &&
901 pool->hd < pool->tl) {
903 * ToDo: add GC code check that we really have enough heap afterwards!!
905 * If we're here (no runnable threads) and we have pending
906 * sparks, we must have a space problem. Get enough space
907 * to turn one of those pending sparks into a
911 spark = findSpark(rtsFalse); /* get a spark */
912 if (spark != (rtsSpark) NULL) {
913 tso = activateSpark(spark); /* turn the spark into a thread */
914 IF_PAR_DEBUG(schedule,
915 belch("==== schedule: Created TSO %d (%p); %d threads active",
916 tso->id, tso, advisory_thread_count));
918 if (tso==END_TSO_QUEUE) { /* failed to activate spark->back to loop */
919 belch("==^^ failed to activate spark");
921 } /* otherwise fall through & pick-up new tso */
923 IF_PAR_DEBUG(verbose,
924 belch("==^^ no local sparks (spark pool contains only NFs: %d)",
925 spark_queue_len(pool)));
930 /* If we still have no work we need to send a FISH to get a spark
933 if (EMPTY_RUN_QUEUE()) {
934 /* =8-[ no local sparks => look for work on other PEs */
936 * We really have absolutely no work. Send out a fish
937 * (there may be some out there already), and wait for
938 * something to arrive. We clearly can't run any threads
939 * until a SCHEDULE or RESUME arrives, and so that's what
940 * we're hoping to see. (Of course, we still have to
941 * respond to other types of messages.)
943 TIME now = msTime() /*CURRENT_TIME*/;
944 IF_PAR_DEBUG(verbose,
945 belch("-- now=%ld", now));
946 IF_PAR_DEBUG(verbose,
947 if (outstandingFishes < RtsFlags.ParFlags.maxFishes &&
948 (last_fish_arrived_at!=0 &&
949 last_fish_arrived_at+RtsFlags.ParFlags.fishDelay > now)) {
950 belch("--$$ delaying FISH until %ld (last fish %ld, delay %ld, now %ld)",
951 last_fish_arrived_at+RtsFlags.ParFlags.fishDelay,
952 last_fish_arrived_at,
953 RtsFlags.ParFlags.fishDelay, now);
956 if (outstandingFishes < RtsFlags.ParFlags.maxFishes &&
957 (last_fish_arrived_at==0 ||
958 (last_fish_arrived_at+RtsFlags.ParFlags.fishDelay <= now))) {
959 /* outstandingFishes is set in sendFish, processFish;
960 avoid flooding system with fishes via delay */
962 sendFish(pe, mytid, NEW_FISH_AGE, NEW_FISH_HISTORY,
965 // Global statistics: count no. of fishes
966 if (RtsFlags.ParFlags.ParStats.Global &&
967 RtsFlags.GcFlags.giveStats > NO_GC_STATS) {
968 globalParStats.tot_fish_mess++;
972 receivedFinish = processMessages();
975 } else if (PacketsWaiting()) { /* Look for incoming messages */
976 receivedFinish = processMessages();
979 /* Now we are sure that we have some work available */
980 ASSERT(run_queue_hd != END_TSO_QUEUE);
982 /* Take a thread from the run queue, if we have work */
983 t = POP_RUN_QUEUE(); // take_off_run_queue(END_TSO_QUEUE);
984 IF_DEBUG(sanity,checkTSO(t));
986 /* ToDo: write something to the log-file
987 if (RTSflags.ParFlags.granSimStats && !sameThread)
988 DumpGranEvent(GR_SCHEDULE, RunnableThreadsHd);
992 /* the spark pool for the current PE */
993 pool = &(MainRegTable.rSparks); // generalise to cap = &MainRegTable
996 belch("--=^ %d threads, %d sparks on [%#x]",
997 run_queue_len(), spark_queue_len(pool), CURRENT_PROC));
1000 if (0 && RtsFlags.ParFlags.ParStats.Full &&
1001 t && LastTSO && t->id != LastTSO->id &&
1002 LastTSO->why_blocked == NotBlocked &&
1003 LastTSO->what_next != ThreadComplete) {
1004 // if previously scheduled TSO not blocked we have to record the context switch
1005 DumpVeryRawGranEvent(TimeOfLastYield, CURRENT_PROC, CURRENT_PROC,
1006 GR_DESCHEDULE, LastTSO, (StgClosure *)NULL, 0, 0);
1009 if (RtsFlags.ParFlags.ParStats.Full &&
1010 (emitSchedule /* forced emit */ ||
1011 (t && LastTSO && t->id != LastTSO->id))) {
1013 we are running a different TSO, so write a schedule event to log file
1014 NB: If we use fair scheduling we also have to write a deschedule
1015 event for LastTSO; with unfair scheduling we know that the
1016 previous tso has blocked whenever we switch to another tso, so
1017 we don't need it in GUM for now
1019 DumpRawGranEvent(CURRENT_PROC, CURRENT_PROC,
1020 GR_SCHEDULE, t, (StgClosure *)NULL, 0, 0);
1021 emitSchedule = rtsFalse;
1025 #else /* !GRAN && !PAR */
1027 /* grab a thread from the run queue
1029 ASSERT(run_queue_hd != END_TSO_QUEUE);
1030 t = POP_RUN_QUEUE();
1031 // Sanity check the thread we're about to run. This can be
1032 // expensive if there is lots of thread switching going on...
1033 IF_DEBUG(sanity,checkTSO(t));
1036 grabCapability(&cap);
1037 cap->r.rCurrentTSO = t;
1039 /* context switches are now initiated by the timer signal, unless
1040 * the user specified "context switch as often as possible", with
1045 RtsFlags.ProfFlags.profileInterval == 0 ||
1047 (RtsFlags.ConcFlags.ctxtSwitchTicks == 0
1048 && (run_queue_hd != END_TSO_QUEUE
1049 || blocked_queue_hd != END_TSO_QUEUE
1050 || sleeping_queue != END_TSO_QUEUE)))
1055 RELEASE_LOCK(&sched_mutex);
1057 IF_DEBUG(scheduler, sched_belch("-->> Running TSO %ld (%p) %s ...",
1058 t->id, t, whatNext_strs[t->what_next]));
1061 startHeapProfTimer();
1064 /* +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ */
1065 /* Run the current thread
1067 switch (cap->r.rCurrentTSO->what_next) {
1069 case ThreadComplete:
1070 /* Thread already finished, return to scheduler. */
1071 ret = ThreadFinished;
1073 case ThreadEnterGHC:
1074 ret = StgRun((StgFunPtr) stg_enterStackTop, &cap->r);
1077 ret = StgRun((StgFunPtr) stg_returnToStackTop, &cap->r);
1079 case ThreadEnterInterp:
1080 ret = interpretBCO(cap);
1083 barf("schedule: invalid what_next field");
1085 /* +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ */
1087 /* Costs for the scheduler are assigned to CCS_SYSTEM */
1089 stopHeapProfTimer();
1093 ACQUIRE_LOCK(&sched_mutex);
1096 IF_DEBUG(scheduler,fprintf(stderr,"scheduler (task %ld): ", osThreadId()););
1097 #elif !defined(GRAN) && !defined(PAR)
1098 IF_DEBUG(scheduler,fprintf(stderr,"scheduler: "););
1100 t = cap->r.rCurrentTSO;
1103 /* HACK 675: if the last thread didn't yield, make sure to print a
1104 SCHEDULE event to the log file when StgRunning the next thread, even
1105 if it is the same one as before */
1107 TimeOfLastYield = CURRENT_TIME;
1113 IF_DEBUG(gran, DumpGranEvent(GR_DESCHEDULE, t));
1114 globalGranStats.tot_heapover++;
1116 globalParStats.tot_heapover++;
1119 // did the task ask for a large block?
1120 if (cap->r.rHpAlloc > BLOCK_SIZE_W) {
1121 // if so, get one and push it on the front of the nursery.
1125 blocks = (nat)BLOCK_ROUND_UP(cap->r.rHpAlloc * sizeof(W_)) / BLOCK_SIZE;
1127 IF_DEBUG(scheduler,belch("--<< thread %ld (%p; %s) stopped: requesting a large block (size %d)",
1129 whatNext_strs[t->what_next], blocks));
1131 // don't do this if it would push us over the
1132 // alloc_blocks_lim limit; we'll GC first.
1133 if (alloc_blocks + blocks < alloc_blocks_lim) {
1135 alloc_blocks += blocks;
1136 bd = allocGroup( blocks );
1138 // link the new group into the list
1139 bd->link = cap->r.rCurrentNursery;
1140 bd->u.back = cap->r.rCurrentNursery->u.back;
1141 if (cap->r.rCurrentNursery->u.back != NULL) {
1142 cap->r.rCurrentNursery->u.back->link = bd;
1144 ASSERT(g0s0->blocks == cap->r.rCurrentNursery &&
1145 g0s0->blocks == cap->r.rNursery);
1146 cap->r.rNursery = g0s0->blocks = bd;
1148 cap->r.rCurrentNursery->u.back = bd;
1150 // initialise it as a nursery block
1154 bd->free = bd->start;
1156 // don't forget to update the block count in g0s0.
1157 g0s0->n_blocks += blocks;
1158 ASSERT(countBlocks(g0s0->blocks) == g0s0->n_blocks);
1160 // now update the nursery to point to the new block
1161 cap->r.rCurrentNursery = bd;
1163 // we might be unlucky and have another thread get on the
1164 // run queue before us and steal the large block, but in that
1165 // case the thread will just end up requesting another large
1167 PUSH_ON_RUN_QUEUE(t);
1172 /* make all the running tasks block on a condition variable,
1173 * maybe set context_switch and wait till they all pile in,
1174 * then have them wait on a GC condition variable.
1176 IF_DEBUG(scheduler,belch("--<< thread %ld (%p; %s) stopped: HeapOverflow",
1177 t->id, t, whatNext_strs[t->what_next]));
1180 ASSERT(!is_on_queue(t,CurrentProc));
1182 /* Currently we emit a DESCHEDULE event before GC in GUM.
1183 ToDo: either add separate event to distinguish SYSTEM time from rest
1184 or just nuke this DESCHEDULE (and the following SCHEDULE) */
1185 if (0 && RtsFlags.ParFlags.ParStats.Full) {
1186 DumpRawGranEvent(CURRENT_PROC, CURRENT_PROC,
1187 GR_DESCHEDULE, t, (StgClosure *)NULL, 0, 0);
1188 emitSchedule = rtsTrue;
1192 ready_to_gc = rtsTrue;
1193 context_switch = 1; /* stop other threads ASAP */
1194 PUSH_ON_RUN_QUEUE(t);
1195 /* actual GC is done at the end of the while loop */
1201 DumpGranEvent(GR_DESCHEDULE, t));
1202 globalGranStats.tot_stackover++;
1205 // DumpGranEvent(GR_DESCHEDULE, t);
1206 globalParStats.tot_stackover++;
1208 IF_DEBUG(scheduler,belch("--<< thread %ld (%p; %s) stopped, StackOverflow",
1209 t->id, t, whatNext_strs[t->what_next]));
1210 /* just adjust the stack for this thread, then pop it back
1216 /* enlarge the stack */
1217 StgTSO *new_t = threadStackOverflow(t);
1219 /* This TSO has moved, so update any pointers to it from the
1220 * main thread stack. It better not be on any other queues...
1221 * (it shouldn't be).
1223 for (m = main_threads; m != NULL; m = m->link) {
1228 threadPaused(new_t);
1229 PUSH_ON_RUN_QUEUE(new_t);
1233 case ThreadYielding:
1236 DumpGranEvent(GR_DESCHEDULE, t));
1237 globalGranStats.tot_yields++;
1240 // DumpGranEvent(GR_DESCHEDULE, t);
1241 globalParStats.tot_yields++;
1243 /* put the thread back on the run queue. Then, if we're ready to
1244 * GC, check whether this is the last task to stop. If so, wake
1245 * up the GC thread. getThread will block during a GC until the
1249 if (t->what_next == ThreadEnterInterp) {
1250 /* ToDo: or maybe a timer expired when we were in Hugs?
1251 * or maybe someone hit ctrl-C
1253 belch("--<< thread %ld (%p; %s) stopped to switch to Hugs",
1254 t->id, t, whatNext_strs[t->what_next]);
1256 belch("--<< thread %ld (%p; %s) stopped, yielding",
1257 t->id, t, whatNext_strs[t->what_next]);
1264 //belch("&& Doing sanity check on yielding TSO %ld.", t->id);
1266 ASSERT(t->link == END_TSO_QUEUE);
1268 ASSERT(!is_on_queue(t,CurrentProc));
1271 //belch("&& Doing sanity check on all ThreadQueues (and their TSOs).");
1272 checkThreadQsSanity(rtsTrue));
1275 if (RtsFlags.ParFlags.doFairScheduling) {
1276 /* this does round-robin scheduling; good for concurrency */
1277 APPEND_TO_RUN_QUEUE(t);
1279 /* this does unfair scheduling; good for parallelism */
1280 PUSH_ON_RUN_QUEUE(t);
1283 /* this does round-robin scheduling; good for concurrency */
1284 APPEND_TO_RUN_QUEUE(t);
1287 /* add a ContinueThread event to actually process the thread */
1288 new_event(CurrentProc, CurrentProc, CurrentTime[CurrentProc],
1290 t, (StgClosure*)NULL, (rtsSpark*)NULL);
1292 belch("GRAN: eventq and runnableq after adding yielded thread to queue again:");
1301 belch("--<< thread %ld (%p; %s) stopped, blocking on node %p [PE %d] with BQ: ",
1302 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)));
1303 if (t->block_info.closure!=(StgClosure*)NULL) print_bq(t->block_info.closure));
1305 // ??? needed; should emit block before
1307 DumpGranEvent(GR_DESCHEDULE, t));
1308 prune_eventq(t, (StgClosure *)NULL); // prune ContinueThreads for t
1311 ASSERT(procStatus[CurrentProc]==Busy ||
1312 ((procStatus[CurrentProc]==Fetching) &&
1313 (t->block_info.closure!=(StgClosure*)NULL)));
1314 if (run_queue_hds[CurrentProc] == END_TSO_QUEUE &&
1315 !(!RtsFlags.GranFlags.DoAsyncFetch &&
1316 procStatus[CurrentProc]==Fetching))
1317 procStatus[CurrentProc] = Idle;
1321 belch("--<< thread %ld (%p; %s) stopped, blocking on node %p with BQ: ",
1322 t->id, t, whatNext_strs[t->what_next], t->block_info.closure));
1325 if (t->block_info.closure!=(StgClosure*)NULL)
1326 print_bq(t->block_info.closure));
1328 /* Send a fetch (if BlockedOnGA) and dump event to log file */
1331 /* whatever we schedule next, we must log that schedule */
1332 emitSchedule = rtsTrue;
1335 /* don't need to do anything. Either the thread is blocked on
1336 * I/O, in which case we'll have called addToBlockedQueue
1337 * previously, or it's blocked on an MVar or Blackhole, in which
1338 * case it'll be on the relevant queue already.
1341 fprintf(stderr, "--<< thread %d (%p) stopped: ", t->id, t);
1342 printThreadBlockage(t);
1343 fprintf(stderr, "\n"));
1345 /* Only for dumping event to log file
1346 ToDo: do I need this in GranSim, too?
1353 case ThreadFinished:
1354 /* Need to check whether this was a main thread, and if so, signal
1355 * the task that started it with the return value. If we have no
1356 * more main threads, we probably need to stop all the tasks until
1359 /* We also end up here if the thread kills itself with an
1360 * uncaught exception, see Exception.hc.
1362 IF_DEBUG(scheduler,belch("--++ thread %d (%p) finished", t->id, t));
1364 endThread(t, CurrentProc); // clean-up the thread
1366 /* For now all are advisory -- HWL */
1367 //if(t->priority==AdvisoryPriority) ??
1368 advisory_thread_count--;
1371 if(t->dist.priority==RevalPriority)
1375 if (RtsFlags.ParFlags.ParStats.Full &&
1376 !RtsFlags.ParFlags.ParStats.Suppressed)
1377 DumpEndEvent(CURRENT_PROC, t, rtsFalse /* not mandatory */);
1382 barf("schedule: invalid thread return code %d", (int)ret);
1386 grabCapability(&cap);
1390 if (RtsFlags.ProfFlags.profileInterval==0 || performHeapProfile) {
1391 GarbageCollect(GetRoots, rtsTrue);
1393 performHeapProfile = rtsFalse;
1394 ready_to_gc = rtsFalse; // we already GC'd
1399 if (ready_to_gc && allFreeCapabilities() )
1404 /* everybody back, start the GC.
1405 * Could do it in this thread, or signal a condition var
1406 * to do it in another thread. Either way, we need to
1407 * broadcast on gc_pending_cond afterward.
1409 #if defined(RTS_SUPPORTS_THREADS)
1410 IF_DEBUG(scheduler,sched_belch("doing GC"));
1412 GarbageCollect(GetRoots,rtsFalse);
1413 ready_to_gc = rtsFalse;
1415 broadcastCondition(&gc_pending_cond);
1418 /* add a ContinueThread event to continue execution of current thread */
1419 new_event(CurrentProc, CurrentProc, CurrentTime[CurrentProc],
1421 t, (StgClosure*)NULL, (rtsSpark*)NULL);
1423 fprintf(stderr, "GRAN: eventq and runnableq after Garbage collection:\n");
1431 IF_GRAN_DEBUG(unused,
1432 print_eventq(EventHd));
1434 event = get_next_event();
1437 /* ToDo: wait for next message to arrive rather than busy wait */
1440 } /* end of while(1) */
1442 IF_PAR_DEBUG(verbose,
1443 belch("== Leaving schedule() after having received Finish"));
1446 /* ---------------------------------------------------------------------------
1447 * deleteAllThreads(): kill all the live threads.
1449 * This is used when we catch a user interrupt (^C), before performing
1450 * any necessary cleanups and running finalizers.
1451 * ------------------------------------------------------------------------- */
1453 void deleteAllThreads ( void )
1456 IF_DEBUG(scheduler,sched_belch("deleting all threads"));
1457 for (t = run_queue_hd; t != END_TSO_QUEUE; t = t->link) {
1460 for (t = blocked_queue_hd; t != END_TSO_QUEUE; t = t->link) {
1463 for (t = sleeping_queue; t != END_TSO_QUEUE; t = t->link) {
1466 run_queue_hd = run_queue_tl = END_TSO_QUEUE;
1467 blocked_queue_hd = blocked_queue_tl = END_TSO_QUEUE;
1468 sleeping_queue = END_TSO_QUEUE;
1471 /* startThread and insertThread are now in GranSim.c -- HWL */
1474 //@node Suspend and Resume, Run queue code, Main scheduling loop, Main scheduling code
1475 //@subsection Suspend and Resume
1477 /* ---------------------------------------------------------------------------
1478 * Suspending & resuming Haskell threads.
1480 * When making a "safe" call to C (aka _ccall_GC), the task gives back
1481 * its capability before calling the C function. This allows another
1482 * task to pick up the capability and carry on running Haskell
1483 * threads. It also means that if the C call blocks, it won't lock
1486 * The Haskell thread making the C call is put to sleep for the
1487 * duration of the call, on the susepended_ccalling_threads queue. We
1488 * give out a token to the task, which it can use to resume the thread
1489 * on return from the C function.
1490 * ------------------------------------------------------------------------- */
1493 suspendThread( StgRegTable *reg )
1498 /* assume that *reg is a pointer to the StgRegTable part
1501 cap = (Capability *)((void *)reg - sizeof(StgFunTable));
1503 ACQUIRE_LOCK(&sched_mutex);
1506 sched_belch("thread %d did a _ccall_gc", cap->r.rCurrentTSO->id));
1508 threadPaused(cap->r.rCurrentTSO);
1509 cap->r.rCurrentTSO->link = suspended_ccalling_threads;
1510 suspended_ccalling_threads = cap->r.rCurrentTSO;
1512 /* Use the thread ID as the token; it should be unique */
1513 tok = cap->r.rCurrentTSO->id;
1515 /* Hand back capability */
1516 releaseCapability(&cap);
1518 #if defined(RTS_SUPPORTS_THREADS) && !defined(SMP)
1519 IF_DEBUG(scheduler, sched_belch("thread %d leaving RTS\n", tok));
1520 startTask(taskStart);
1523 RELEASE_LOCK(&sched_mutex);
1524 RELEASE_LOCK(&rts_mutex);
1529 resumeThread( StgInt tok )
1531 StgTSO *tso, **prev;
1534 IF_DEBUG(scheduler, sched_belch("thread %d returning, waiting for sched. lock.\n", tok));
1535 ACQUIRE_LOCK(&sched_mutex);
1536 ext_threads_waiting++;
1537 IF_DEBUG(scheduler, sched_belch("thread %d returning, ext_thread count: %d.\n", tok, ext_threads_waiting));
1538 RELEASE_LOCK(&sched_mutex);
1540 IF_DEBUG(scheduler, sched_belch("thread %d waiting for RTS lock...\n", tok));
1541 ACQUIRE_LOCK(&rts_mutex);
1542 ext_threads_waiting--;
1543 IF_DEBUG(scheduler, sched_belch("thread %d acquired RTS lock...\n", tok));
1545 #if defined(THREADED_RTS)
1546 /* Free up any RTS-blocked threads. */
1547 broadcastCondition(&thread_ready_cond);
1550 /* Remove the thread off of the suspended list */
1551 prev = &suspended_ccalling_threads;
1552 for (tso = suspended_ccalling_threads;
1553 tso != END_TSO_QUEUE;
1554 prev = &tso->link, tso = tso->link) {
1555 if (tso->id == (StgThreadID)tok) {
1560 if (tso == END_TSO_QUEUE) {
1561 barf("resumeThread: thread not found");
1563 tso->link = END_TSO_QUEUE;
1566 while ( noFreeCapabilities() ) {
1567 IF_DEBUG(scheduler, sched_belch("waiting to resume"));
1568 waitCondition(&thread_ready_cond, &sched_mutex);
1569 IF_DEBUG(scheduler, sched_belch("resuming thread %d", tso->id));
1573 grabCapability(&cap);
1575 cap->r.rCurrentTSO = tso;
1581 /* ---------------------------------------------------------------------------
1583 * ------------------------------------------------------------------------ */
1584 static void unblockThread(StgTSO *tso);
1586 /* ---------------------------------------------------------------------------
1587 * Comparing Thread ids.
1589 * This is used from STG land in the implementation of the
1590 * instances of Eq/Ord for ThreadIds.
1591 * ------------------------------------------------------------------------ */
1593 int cmp_thread(const StgTSO *tso1, const StgTSO *tso2)
1595 StgThreadID id1 = tso1->id;
1596 StgThreadID id2 = tso2->id;
1598 if (id1 < id2) return (-1);
1599 if (id1 > id2) return 1;
1603 /* ---------------------------------------------------------------------------
1604 * Fetching the ThreadID from an StgTSO.
1606 * This is used in the implementation of Show for ThreadIds.
1607 * ------------------------------------------------------------------------ */
1608 int rts_getThreadId(const StgTSO *tso)
1613 /* ---------------------------------------------------------------------------
1614 Create a new thread.
1616 The new thread starts with the given stack size. Before the
1617 scheduler can run, however, this thread needs to have a closure
1618 (and possibly some arguments) pushed on its stack. See
1619 pushClosure() in Schedule.h.
1621 createGenThread() and createIOThread() (in SchedAPI.h) are
1622 convenient packaged versions of this function.
1624 currently pri (priority) is only used in a GRAN setup -- HWL
1625 ------------------------------------------------------------------------ */
1626 //@cindex createThread
1628 /* currently pri (priority) is only used in a GRAN setup -- HWL */
1630 createThread(nat stack_size, StgInt pri)
1632 return createThread_(stack_size, rtsFalse, pri);
1636 createThread_(nat size, rtsBool have_lock, StgInt pri)
1640 createThread(nat stack_size)
1642 return createThread_(stack_size, rtsFalse);
1646 createThread_(nat size, rtsBool have_lock)
1653 /* First check whether we should create a thread at all */
1655 /* check that no more than RtsFlags.ParFlags.maxThreads threads are created */
1656 if (advisory_thread_count >= RtsFlags.ParFlags.maxThreads) {
1658 belch("{createThread}Daq ghuH: refusing to create another thread; no more than %d threads allowed (currently %d)",
1659 RtsFlags.ParFlags.maxThreads, advisory_thread_count);
1660 return END_TSO_QUEUE;
1666 ASSERT(!RtsFlags.GranFlags.Light || CurrentProc==0);
1669 // ToDo: check whether size = stack_size - TSO_STRUCT_SIZEW
1671 /* catch ridiculously small stack sizes */
1672 if (size < MIN_STACK_WORDS + TSO_STRUCT_SIZEW) {
1673 size = MIN_STACK_WORDS + TSO_STRUCT_SIZEW;
1676 stack_size = size - TSO_STRUCT_SIZEW;
1678 tso = (StgTSO *)allocate(size);
1679 TICK_ALLOC_TSO(stack_size, 0);
1681 SET_HDR(tso, &stg_TSO_info, CCS_SYSTEM);
1683 SET_GRAN_HDR(tso, ThisPE);
1685 tso->what_next = ThreadEnterGHC;
1687 /* tso->id needs to be unique. For now we use a heavyweight mutex to
1688 * protect the increment operation on next_thread_id.
1689 * In future, we could use an atomic increment instead.
1691 if (!have_lock) { ACQUIRE_LOCK(&sched_mutex); }
1692 tso->id = next_thread_id++;
1693 if (!have_lock) { RELEASE_LOCK(&sched_mutex); }
1695 tso->why_blocked = NotBlocked;
1696 tso->blocked_exceptions = NULL;
1698 tso->stack_size = stack_size;
1699 tso->max_stack_size = round_to_mblocks(RtsFlags.GcFlags.maxStkSize)
1701 tso->sp = (P_)&(tso->stack) + stack_size;
1704 tso->prof.CCCS = CCS_MAIN;
1707 /* put a stop frame on the stack */
1708 tso->sp -= sizeofW(StgStopFrame);
1709 SET_HDR((StgClosure*)tso->sp,(StgInfoTable *)&stg_stop_thread_info,CCS_SYSTEM);
1710 tso->su = (StgUpdateFrame*)tso->sp;
1714 tso->link = END_TSO_QUEUE;
1715 /* uses more flexible routine in GranSim */
1716 insertThread(tso, CurrentProc);
1718 /* In a non-GranSim setup the pushing of a TSO onto the runq is separated
1724 if (RtsFlags.GranFlags.GranSimStats.Full)
1725 DumpGranEvent(GR_START,tso);
1727 if (RtsFlags.ParFlags.ParStats.Full)
1728 DumpGranEvent(GR_STARTQ,tso);
1729 /* HACk to avoid SCHEDULE
1733 /* Link the new thread on the global thread list.
1735 tso->global_link = all_threads;
1739 tso->dist.priority = MandatoryPriority; //by default that is...
1743 tso->gran.pri = pri;
1745 tso->gran.magic = TSO_MAGIC; // debugging only
1747 tso->gran.sparkname = 0;
1748 tso->gran.startedat = CURRENT_TIME;
1749 tso->gran.exported = 0;
1750 tso->gran.basicblocks = 0;
1751 tso->gran.allocs = 0;
1752 tso->gran.exectime = 0;
1753 tso->gran.fetchtime = 0;
1754 tso->gran.fetchcount = 0;
1755 tso->gran.blocktime = 0;
1756 tso->gran.blockcount = 0;
1757 tso->gran.blockedat = 0;
1758 tso->gran.globalsparks = 0;
1759 tso->gran.localsparks = 0;
1760 if (RtsFlags.GranFlags.Light)
1761 tso->gran.clock = Now; /* local clock */
1763 tso->gran.clock = 0;
1765 IF_DEBUG(gran,printTSO(tso));
1768 tso->par.magic = TSO_MAGIC; // debugging only
1770 tso->par.sparkname = 0;
1771 tso->par.startedat = CURRENT_TIME;
1772 tso->par.exported = 0;
1773 tso->par.basicblocks = 0;
1774 tso->par.allocs = 0;
1775 tso->par.exectime = 0;
1776 tso->par.fetchtime = 0;
1777 tso->par.fetchcount = 0;
1778 tso->par.blocktime = 0;
1779 tso->par.blockcount = 0;
1780 tso->par.blockedat = 0;
1781 tso->par.globalsparks = 0;
1782 tso->par.localsparks = 0;
1786 globalGranStats.tot_threads_created++;
1787 globalGranStats.threads_created_on_PE[CurrentProc]++;
1788 globalGranStats.tot_sq_len += spark_queue_len(CurrentProc);
1789 globalGranStats.tot_sq_probes++;
1791 // collect parallel global statistics (currently done together with GC stats)
1792 if (RtsFlags.ParFlags.ParStats.Global &&
1793 RtsFlags.GcFlags.giveStats > NO_GC_STATS) {
1794 //fprintf(stderr, "Creating thread %d @ %11.2f\n", tso->id, usertime());
1795 globalParStats.tot_threads_created++;
1801 belch("==__ schedule: Created TSO %d (%p);",
1802 CurrentProc, tso, tso->id));
1804 IF_PAR_DEBUG(verbose,
1805 belch("==__ schedule: Created TSO %d (%p); %d threads active",
1806 tso->id, tso, advisory_thread_count));
1808 IF_DEBUG(scheduler,sched_belch("created thread %ld, stack size = %lx words",
1809 tso->id, tso->stack_size));
1816 all parallel thread creation calls should fall through the following routine.
1819 createSparkThread(rtsSpark spark)
1821 ASSERT(spark != (rtsSpark)NULL);
1822 if (advisory_thread_count >= RtsFlags.ParFlags.maxThreads)
1824 barf("{createSparkThread}Daq ghuH: refusing to create another thread; no more than %d threads allowed (currently %d)",
1825 RtsFlags.ParFlags.maxThreads, advisory_thread_count);
1826 return END_TSO_QUEUE;
1830 tso = createThread_(RtsFlags.GcFlags.initialStkSize, rtsTrue);
1831 if (tso==END_TSO_QUEUE)
1832 barf("createSparkThread: Cannot create TSO");
1834 tso->priority = AdvisoryPriority;
1836 pushClosure(tso,spark);
1837 PUSH_ON_RUN_QUEUE(tso);
1838 advisory_thread_count++;
1845 Turn a spark into a thread.
1846 ToDo: fix for SMP (needs to acquire SCHED_MUTEX!)
1849 //@cindex activateSpark
1851 activateSpark (rtsSpark spark)
1855 tso = createSparkThread(spark);
1856 if (RtsFlags.ParFlags.ParStats.Full) {
1857 //ASSERT(run_queue_hd == END_TSO_QUEUE); // I think ...
1858 IF_PAR_DEBUG(verbose,
1859 belch("==^^ activateSpark: turning spark of closure %p (%s) into a thread",
1860 (StgClosure *)spark, info_type((StgClosure *)spark)));
1862 // ToDo: fwd info on local/global spark to thread -- HWL
1863 // tso->gran.exported = spark->exported;
1864 // tso->gran.locked = !spark->global;
1865 // tso->gran.sparkname = spark->name;
1871 /* ---------------------------------------------------------------------------
1874 * scheduleThread puts a thread on the head of the runnable queue.
1875 * This will usually be done immediately after a thread is created.
1876 * The caller of scheduleThread must create the thread using e.g.
1877 * createThread and push an appropriate closure
1878 * on this thread's stack before the scheduler is invoked.
1879 * ------------------------------------------------------------------------ */
1882 scheduleThread(StgTSO *tso)
1884 ACQUIRE_LOCK(&sched_mutex);
1886 /* Put the new thread on the head of the runnable queue. The caller
1887 * better push an appropriate closure on this thread's stack
1888 * beforehand. In the SMP case, the thread may start running as
1889 * soon as we release the scheduler lock below.
1891 PUSH_ON_RUN_QUEUE(tso);
1895 IF_DEBUG(scheduler,printTSO(tso));
1897 RELEASE_LOCK(&sched_mutex);
1900 /* ---------------------------------------------------------------------------
1903 * Initialise the scheduler. This resets all the queues - if the
1904 * queues contained any threads, they'll be garbage collected at the
1907 * ------------------------------------------------------------------------ */
1911 term_handler(int sig STG_UNUSED)
1914 ACQUIRE_LOCK(&term_mutex);
1916 RELEASE_LOCK(&term_mutex);
1927 for (i=0; i<=MAX_PROC; i++) {
1928 run_queue_hds[i] = END_TSO_QUEUE;
1929 run_queue_tls[i] = END_TSO_QUEUE;
1930 blocked_queue_hds[i] = END_TSO_QUEUE;
1931 blocked_queue_tls[i] = END_TSO_QUEUE;
1932 ccalling_threadss[i] = END_TSO_QUEUE;
1933 sleeping_queue = END_TSO_QUEUE;
1936 run_queue_hd = END_TSO_QUEUE;
1937 run_queue_tl = END_TSO_QUEUE;
1938 blocked_queue_hd = END_TSO_QUEUE;
1939 blocked_queue_tl = END_TSO_QUEUE;
1940 sleeping_queue = END_TSO_QUEUE;
1943 suspended_ccalling_threads = END_TSO_QUEUE;
1945 main_threads = NULL;
1946 all_threads = END_TSO_QUEUE;
1951 RtsFlags.ConcFlags.ctxtSwitchTicks =
1952 RtsFlags.ConcFlags.ctxtSwitchTime / TICK_MILLISECS;
1954 #if defined(RTS_SUPPORTS_THREADS)
1955 /* Initialise the mutex and condition variables used by
1957 initMutex(&rts_mutex);
1958 initMutex(&sched_mutex);
1959 initMutex(&term_mutex);
1960 #if defined(THREADED_RTS)
1961 initMutex(&thread_ready_aux_mutex);
1964 initCondition(&thread_ready_cond);
1965 initCondition(&gc_pending_cond);
1968 #if defined(THREADED_RTS)
1970 ACQUIRE_LOCK(&rts_mutex);
1972 sched_belch("worker thread (%d): acquired RTS lock\n", osThreadId()));
1975 /* Install the SIGHUP handler */
1978 struct sigaction action,oact;
1980 action.sa_handler = term_handler;
1981 sigemptyset(&action.sa_mask);
1982 action.sa_flags = 0;
1983 if (sigaction(SIGTERM, &action, &oact) != 0) {
1984 barf("can't install TERM handler");
1989 /* A capability holds the state a native thread needs in
1990 * order to execute STG code. At least one capability is
1991 * floating around (only SMP builds have more than one).
1995 #if defined(RTS_SUPPORTS_THREADS)
1996 /* start our haskell execution tasks */
1998 startTaskManager(RtsFlags.ParFlags.nNodes, taskStart);
2000 startTaskManager(0,taskStart);
2004 #if /* defined(SMP) ||*/ defined(PAR)
2010 exitScheduler( void )
2012 #if defined(RTS_SUPPORTS_THREADS)
2017 /* -----------------------------------------------------------------------------
2018 Managing the per-task allocation areas.
2020 Each capability comes with an allocation area. These are
2021 fixed-length block lists into which allocation can be done.
2023 ToDo: no support for two-space collection at the moment???
2024 -------------------------------------------------------------------------- */
2026 /* -----------------------------------------------------------------------------
2027 * waitThread is the external interface for running a new computation
2028 * and waiting for the result.
2030 * In the non-SMP case, we create a new main thread, push it on the
2031 * main-thread stack, and invoke the scheduler to run it. The
2032 * scheduler will return when the top main thread on the stack has
2033 * completed or died, and fill in the necessary fields of the
2034 * main_thread structure.
2036 * In the SMP case, we create a main thread as before, but we then
2037 * create a new condition variable and sleep on it. When our new
2038 * main thread has completed, we'll be woken up and the status/result
2039 * will be in the main_thread struct.
2040 * -------------------------------------------------------------------------- */
2043 howManyThreadsAvail ( void )
2047 for (q = run_queue_hd; q != END_TSO_QUEUE; q = q->link)
2049 for (q = blocked_queue_hd; q != END_TSO_QUEUE; q = q->link)
2051 for (q = sleeping_queue; q != END_TSO_QUEUE; q = q->link)
2057 finishAllThreads ( void )
2060 while (run_queue_hd != END_TSO_QUEUE) {
2061 waitThread ( run_queue_hd, NULL );
2063 while (blocked_queue_hd != END_TSO_QUEUE) {
2064 waitThread ( blocked_queue_hd, NULL );
2066 while (sleeping_queue != END_TSO_QUEUE) {
2067 waitThread ( blocked_queue_hd, NULL );
2070 (blocked_queue_hd != END_TSO_QUEUE ||
2071 run_queue_hd != END_TSO_QUEUE ||
2072 sleeping_queue != END_TSO_QUEUE);
2076 waitThread(StgTSO *tso, /*out*/StgClosure **ret)
2079 SchedulerStatus stat;
2081 ACQUIRE_LOCK(&sched_mutex);
2083 m = stgMallocBytes(sizeof(StgMainThread), "waitThread");
2088 #if defined(RTS_SUPPORTS_THREADS)
2089 initCondition(&m->wakeup);
2092 m->link = main_threads;
2095 IF_DEBUG(scheduler, fprintf(stderr, "== scheduler: new main thread (%d)\n",
2100 waitCondition(&m->wakeup, &sched_mutex);
2101 } while (m->stat == NoStatus);
2103 /* GranSim specific init */
2104 CurrentTSO = m->tso; // the TSO to run
2105 procStatus[MainProc] = Busy; // status of main PE
2106 CurrentProc = MainProc; // PE to run it on
2110 RELEASE_LOCK(&sched_mutex);
2112 ASSERT(m->stat != NoStatus);
2117 #if defined(RTS_SUPPORTS_THREADS)
2118 closeCondition(&m->wakeup);
2121 IF_DEBUG(scheduler, fprintf(stderr, "== scheduler: main thread (%d) finished\n",
2125 RELEASE_LOCK(&sched_mutex);
2130 //@node Run queue code, Garbage Collextion Routines, Suspend and Resume, Main scheduling code
2131 //@subsection Run queue code
2135 NB: In GranSim we have many run queues; run_queue_hd is actually a macro
2136 unfolding to run_queue_hds[CurrentProc], thus CurrentProc is an
2137 implicit global variable that has to be correct when calling these
2141 /* Put the new thread on the head of the runnable queue.
2142 * The caller of createThread better push an appropriate closure
2143 * on this thread's stack before the scheduler is invoked.
2145 static /* inline */ void
2146 add_to_run_queue(tso)
2149 ASSERT(tso!=run_queue_hd && tso!=run_queue_tl);
2150 tso->link = run_queue_hd;
2152 if (run_queue_tl == END_TSO_QUEUE) {
2157 /* Put the new thread at the end of the runnable queue. */
2158 static /* inline */ void
2159 push_on_run_queue(tso)
2162 ASSERT(get_itbl((StgClosure *)tso)->type == TSO);
2163 ASSERT(run_queue_hd!=NULL && run_queue_tl!=NULL);
2164 ASSERT(tso!=run_queue_hd && tso!=run_queue_tl);
2165 if (run_queue_hd == END_TSO_QUEUE) {
2168 run_queue_tl->link = tso;
2174 Should be inlined because it's used very often in schedule. The tso
2175 argument is actually only needed in GranSim, where we want to have the
2176 possibility to schedule *any* TSO on the run queue, irrespective of the
2177 actual ordering. Therefore, if tso is not the nil TSO then we traverse
2178 the run queue and dequeue the tso, adjusting the links in the queue.
2180 //@cindex take_off_run_queue
2181 static /* inline */ StgTSO*
2182 take_off_run_queue(StgTSO *tso) {
2186 qetlaHbogh Qu' ngaSbogh ghomDaQ {tso} yIteq!
2188 if tso is specified, unlink that tso from the run_queue (doesn't have
2189 to be at the beginning of the queue); GranSim only
2191 if (tso!=END_TSO_QUEUE) {
2192 /* find tso in queue */
2193 for (t=run_queue_hd, prev=END_TSO_QUEUE;
2194 t!=END_TSO_QUEUE && t!=tso;
2198 /* now actually dequeue the tso */
2199 if (prev!=END_TSO_QUEUE) {
2200 ASSERT(run_queue_hd!=t);
2201 prev->link = t->link;
2203 /* t is at beginning of thread queue */
2204 ASSERT(run_queue_hd==t);
2205 run_queue_hd = t->link;
2207 /* t is at end of thread queue */
2208 if (t->link==END_TSO_QUEUE) {
2209 ASSERT(t==run_queue_tl);
2210 run_queue_tl = prev;
2212 ASSERT(run_queue_tl!=t);
2214 t->link = END_TSO_QUEUE;
2216 /* take tso from the beginning of the queue; std concurrent code */
2218 if (t != END_TSO_QUEUE) {
2219 run_queue_hd = t->link;
2220 t->link = END_TSO_QUEUE;
2221 if (run_queue_hd == END_TSO_QUEUE) {
2222 run_queue_tl = END_TSO_QUEUE;
2231 //@node Garbage Collextion Routines, Blocking Queue Routines, Run queue code, Main scheduling code
2232 //@subsection Garbage Collextion Routines
2234 /* ---------------------------------------------------------------------------
2235 Where are the roots that we know about?
2237 - all the threads on the runnable queue
2238 - all the threads on the blocked queue
2239 - all the threads on the sleeping queue
2240 - all the thread currently executing a _ccall_GC
2241 - all the "main threads"
2243 ------------------------------------------------------------------------ */
2245 /* This has to be protected either by the scheduler monitor, or by the
2246 garbage collection monitor (probably the latter).
2251 GetRoots(evac_fn evac)
2258 for (i=0; i<=RtsFlags.GranFlags.proc; i++) {
2259 if ((run_queue_hds[i] != END_TSO_QUEUE) && ((run_queue_hds[i] != NULL)))
2260 evac((StgClosure **)&run_queue_hds[i]);
2261 if ((run_queue_tls[i] != END_TSO_QUEUE) && ((run_queue_tls[i] != NULL)))
2262 evac((StgClosure **)&run_queue_tls[i]);
2264 if ((blocked_queue_hds[i] != END_TSO_QUEUE) && ((blocked_queue_hds[i] != NULL)))
2265 evac((StgClosure **)&blocked_queue_hds[i]);
2266 if ((blocked_queue_tls[i] != END_TSO_QUEUE) && ((blocked_queue_tls[i] != NULL)))
2267 evac((StgClosure **)&blocked_queue_tls[i]);
2268 if ((ccalling_threadss[i] != END_TSO_QUEUE) && ((ccalling_threadss[i] != NULL)))
2269 evac((StgClosure **)&ccalling_threads[i]);
2276 if (run_queue_hd != END_TSO_QUEUE) {
2277 ASSERT(run_queue_tl != END_TSO_QUEUE);
2278 evac((StgClosure **)&run_queue_hd);
2279 evac((StgClosure **)&run_queue_tl);
2282 if (blocked_queue_hd != END_TSO_QUEUE) {
2283 ASSERT(blocked_queue_tl != END_TSO_QUEUE);
2284 evac((StgClosure **)&blocked_queue_hd);
2285 evac((StgClosure **)&blocked_queue_tl);
2288 if (sleeping_queue != END_TSO_QUEUE) {
2289 evac((StgClosure **)&sleeping_queue);
2293 for (m = main_threads; m != NULL; m = m->link) {
2294 evac((StgClosure **)&m->tso);
2296 if (suspended_ccalling_threads != END_TSO_QUEUE) {
2297 evac((StgClosure **)&suspended_ccalling_threads);
2300 #if defined(PAR) || defined(GRAN)
2301 markSparkQueue(evac);
2305 /* -----------------------------------------------------------------------------
2308 This is the interface to the garbage collector from Haskell land.
2309 We provide this so that external C code can allocate and garbage
2310 collect when called from Haskell via _ccall_GC.
2312 It might be useful to provide an interface whereby the programmer
2313 can specify more roots (ToDo).
2315 This needs to be protected by the GC condition variable above. KH.
2316 -------------------------------------------------------------------------- */
2318 void (*extra_roots)(evac_fn);
2323 GarbageCollect(GetRoots,rtsFalse);
2327 performMajorGC(void)
2329 GarbageCollect(GetRoots,rtsTrue);
2333 AllRoots(evac_fn evac)
2335 GetRoots(evac); // the scheduler's roots
2336 extra_roots(evac); // the user's roots
2340 performGCWithRoots(void (*get_roots)(evac_fn))
2342 extra_roots = get_roots;
2343 GarbageCollect(AllRoots,rtsFalse);
2346 /* -----------------------------------------------------------------------------
2349 If the thread has reached its maximum stack size, then raise the
2350 StackOverflow exception in the offending thread. Otherwise
2351 relocate the TSO into a larger chunk of memory and adjust its stack
2353 -------------------------------------------------------------------------- */
2356 threadStackOverflow(StgTSO *tso)
2358 nat new_stack_size, new_tso_size, diff, stack_words;
2362 IF_DEBUG(sanity,checkTSO(tso));
2363 if (tso->stack_size >= tso->max_stack_size) {
2366 belch("@@ threadStackOverflow of TSO %d (%p): stack too large (now %ld; max is %ld",
2367 tso->id, tso, tso->stack_size, tso->max_stack_size);
2368 /* If we're debugging, just print out the top of the stack */
2369 printStackChunk(tso->sp, stg_min(tso->stack+tso->stack_size,
2372 /* Send this thread the StackOverflow exception */
2373 raiseAsync(tso, (StgClosure *)stackOverflow_closure);
2377 /* Try to double the current stack size. If that takes us over the
2378 * maximum stack size for this thread, then use the maximum instead.
2379 * Finally round up so the TSO ends up as a whole number of blocks.
2381 new_stack_size = stg_min(tso->stack_size * 2, tso->max_stack_size);
2382 new_tso_size = (nat)BLOCK_ROUND_UP(new_stack_size * sizeof(W_) +
2383 TSO_STRUCT_SIZE)/sizeof(W_);
2384 new_tso_size = round_to_mblocks(new_tso_size); /* Be MBLOCK-friendly */
2385 new_stack_size = new_tso_size - TSO_STRUCT_SIZEW;
2387 IF_DEBUG(scheduler, fprintf(stderr,"== scheduler: increasing stack size from %d words to %d.\n", tso->stack_size, new_stack_size));
2389 dest = (StgTSO *)allocate(new_tso_size);
2390 TICK_ALLOC_TSO(new_stack_size,0);
2392 /* copy the TSO block and the old stack into the new area */
2393 memcpy(dest,tso,TSO_STRUCT_SIZE);
2394 stack_words = tso->stack + tso->stack_size - tso->sp;
2395 new_sp = (P_)dest + new_tso_size - stack_words;
2396 memcpy(new_sp, tso->sp, stack_words * sizeof(W_));
2398 /* relocate the stack pointers... */
2399 diff = (P_)new_sp - (P_)tso->sp; /* In *words* */
2400 dest->su = (StgUpdateFrame *) ((P_)dest->su + diff);
2402 dest->stack_size = new_stack_size;
2404 /* and relocate the update frame list */
2405 relocate_stack(dest, diff);
2407 /* Mark the old TSO as relocated. We have to check for relocated
2408 * TSOs in the garbage collector and any primops that deal with TSOs.
2410 * It's important to set the sp and su values to just beyond the end
2411 * of the stack, so we don't attempt to scavenge any part of the
2414 tso->what_next = ThreadRelocated;
2416 tso->sp = (P_)&(tso->stack[tso->stack_size]);
2417 tso->su = (StgUpdateFrame *)tso->sp;
2418 tso->why_blocked = NotBlocked;
2419 dest->mut_link = NULL;
2421 IF_PAR_DEBUG(verbose,
2422 belch("@@ threadStackOverflow of TSO %d (now at %p): stack size increased to %ld",
2423 tso->id, tso, tso->stack_size);
2424 /* If we're debugging, just print out the top of the stack */
2425 printStackChunk(tso->sp, stg_min(tso->stack+tso->stack_size,
2428 IF_DEBUG(sanity,checkTSO(tso));
2430 IF_DEBUG(scheduler,printTSO(dest));
2436 //@node Blocking Queue Routines, Exception Handling Routines, Garbage Collextion Routines, Main scheduling code
2437 //@subsection Blocking Queue Routines
2439 /* ---------------------------------------------------------------------------
2440 Wake up a queue that was blocked on some resource.
2441 ------------------------------------------------------------------------ */
2445 unblockCount ( StgBlockingQueueElement *bqe, StgClosure *node )
2450 unblockCount ( StgBlockingQueueElement *bqe, StgClosure *node )
2452 /* write RESUME events to log file and
2453 update blocked and fetch time (depending on type of the orig closure) */
2454 if (RtsFlags.ParFlags.ParStats.Full) {
2455 DumpRawGranEvent(CURRENT_PROC, CURRENT_PROC,
2456 GR_RESUMEQ, ((StgTSO *)bqe), ((StgTSO *)bqe)->block_info.closure,
2457 0, 0 /* spark_queue_len(ADVISORY_POOL) */);
2458 if (EMPTY_RUN_QUEUE())
2459 emitSchedule = rtsTrue;
2461 switch (get_itbl(node)->type) {
2463 ((StgTSO *)bqe)->par.fetchtime += CURRENT_TIME-((StgTSO *)bqe)->par.blockedat;
2468 ((StgTSO *)bqe)->par.blocktime += CURRENT_TIME-((StgTSO *)bqe)->par.blockedat;
2475 barf("{unblockOneLocked}Daq Qagh: unexpected closure in blocking queue");
2482 static StgBlockingQueueElement *
2483 unblockOneLocked(StgBlockingQueueElement *bqe, StgClosure *node)
2486 PEs node_loc, tso_loc;
2488 node_loc = where_is(node); // should be lifted out of loop
2489 tso = (StgTSO *)bqe; // wastes an assignment to get the type right
2490 tso_loc = where_is((StgClosure *)tso);
2491 if (IS_LOCAL_TO(PROCS(node),tso_loc)) { // TSO is local
2492 /* !fake_fetch => TSO is on CurrentProc is same as IS_LOCAL_TO */
2493 ASSERT(CurrentProc!=node_loc || tso_loc==CurrentProc);
2494 CurrentTime[CurrentProc] += RtsFlags.GranFlags.Costs.lunblocktime;
2495 // insertThread(tso, node_loc);
2496 new_event(tso_loc, tso_loc, CurrentTime[CurrentProc],
2498 tso, node, (rtsSpark*)NULL);
2499 tso->link = END_TSO_QUEUE; // overwrite link just to be sure
2502 } else { // TSO is remote (actually should be FMBQ)
2503 CurrentTime[CurrentProc] += RtsFlags.GranFlags.Costs.mpacktime +
2504 RtsFlags.GranFlags.Costs.gunblocktime +
2505 RtsFlags.GranFlags.Costs.latency;
2506 new_event(tso_loc, CurrentProc, CurrentTime[CurrentProc],
2508 tso, node, (rtsSpark*)NULL);
2509 tso->link = END_TSO_QUEUE; // overwrite link just to be sure
2512 /* the thread-queue-overhead is accounted for in either Resume or UnblockThread */
2514 fprintf(stderr," %s TSO %d (%p) [PE %d] (block_info.closure=%p) (next=%p) ,",
2515 (node_loc==tso_loc ? "Local" : "Global"),
2516 tso->id, tso, CurrentProc, tso->block_info.closure, tso->link));
2517 tso->block_info.closure = NULL;
2518 IF_DEBUG(scheduler,belch("-- Waking up thread %ld (%p)",
2522 static StgBlockingQueueElement *
2523 unblockOneLocked(StgBlockingQueueElement *bqe, StgClosure *node)
2525 StgBlockingQueueElement *next;
2527 switch (get_itbl(bqe)->type) {
2529 ASSERT(((StgTSO *)bqe)->why_blocked != NotBlocked);
2530 /* if it's a TSO just push it onto the run_queue */
2532 // ((StgTSO *)bqe)->link = END_TSO_QUEUE; // debugging?
2533 PUSH_ON_RUN_QUEUE((StgTSO *)bqe);
2535 unblockCount(bqe, node);
2536 /* reset blocking status after dumping event */
2537 ((StgTSO *)bqe)->why_blocked = NotBlocked;
2541 /* if it's a BLOCKED_FETCH put it on the PendingFetches list */
2543 bqe->link = (StgBlockingQueueElement *)PendingFetches;
2544 PendingFetches = (StgBlockedFetch *)bqe;
2548 /* can ignore this case in a non-debugging setup;
2549 see comments on RBHSave closures above */
2551 /* check that the closure is an RBHSave closure */
2552 ASSERT(get_itbl((StgClosure *)bqe) == &stg_RBH_Save_0_info ||
2553 get_itbl((StgClosure *)bqe) == &stg_RBH_Save_1_info ||
2554 get_itbl((StgClosure *)bqe) == &stg_RBH_Save_2_info);
2558 barf("{unblockOneLocked}Daq Qagh: Unexpected IP (%#lx; %s) in blocking queue at %#lx\n",
2559 get_itbl((StgClosure *)bqe), info_type((StgClosure *)bqe),
2563 IF_PAR_DEBUG(bq, fprintf(stderr, ", %p (%s)", bqe, info_type((StgClosure*)bqe)));
2567 #else /* !GRAN && !PAR */
2569 unblockOneLocked(StgTSO *tso)
2573 ASSERT(get_itbl(tso)->type == TSO);
2574 ASSERT(tso->why_blocked != NotBlocked);
2575 tso->why_blocked = NotBlocked;
2577 PUSH_ON_RUN_QUEUE(tso);
2579 IF_DEBUG(scheduler,sched_belch("waking up thread %ld", tso->id));
2584 #if defined(GRAN) || defined(PAR)
2585 inline StgBlockingQueueElement *
2586 unblockOne(StgBlockingQueueElement *bqe, StgClosure *node)
2588 ACQUIRE_LOCK(&sched_mutex);
2589 bqe = unblockOneLocked(bqe, node);
2590 RELEASE_LOCK(&sched_mutex);
2595 unblockOne(StgTSO *tso)
2597 ACQUIRE_LOCK(&sched_mutex);
2598 tso = unblockOneLocked(tso);
2599 RELEASE_LOCK(&sched_mutex);
2606 awakenBlockedQueue(StgBlockingQueueElement *q, StgClosure *node)
2608 StgBlockingQueueElement *bqe;
2613 belch("##-_ AwBQ for node %p on PE %d @ %ld by TSO %d (%p): ", \
2614 node, CurrentProc, CurrentTime[CurrentProc],
2615 CurrentTSO->id, CurrentTSO));
2617 node_loc = where_is(node);
2619 ASSERT(q == END_BQ_QUEUE ||
2620 get_itbl(q)->type == TSO || // q is either a TSO or an RBHSave
2621 get_itbl(q)->type == CONSTR); // closure (type constructor)
2622 ASSERT(is_unique(node));
2624 /* FAKE FETCH: magically copy the node to the tso's proc;
2625 no Fetch necessary because in reality the node should not have been
2626 moved to the other PE in the first place
2628 if (CurrentProc!=node_loc) {
2630 belch("## node %p is on PE %d but CurrentProc is %d (TSO %d); assuming fake fetch and adjusting bitmask (old: %#x)",
2631 node, node_loc, CurrentProc, CurrentTSO->id,
2632 // CurrentTSO, where_is(CurrentTSO),
2633 node->header.gran.procs));
2634 node->header.gran.procs = (node->header.gran.procs) | PE_NUMBER(CurrentProc);
2636 belch("## new bitmask of node %p is %#x",
2637 node, node->header.gran.procs));
2638 if (RtsFlags.GranFlags.GranSimStats.Global) {
2639 globalGranStats.tot_fake_fetches++;
2644 // ToDo: check: ASSERT(CurrentProc==node_loc);
2645 while (get_itbl(bqe)->type==TSO) { // q != END_TSO_QUEUE) {
2648 bqe points to the current element in the queue
2649 next points to the next element in the queue
2651 //tso = (StgTSO *)bqe; // wastes an assignment to get the type right
2652 //tso_loc = where_is(tso);
2654 bqe = unblockOneLocked(bqe, node);
2657 /* if this is the BQ of an RBH, we have to put back the info ripped out of
2658 the closure to make room for the anchor of the BQ */
2659 if (bqe!=END_BQ_QUEUE) {
2660 ASSERT(get_itbl(node)->type == RBH && get_itbl(bqe)->type == CONSTR);
2662 ASSERT((info_ptr==&RBH_Save_0_info) ||
2663 (info_ptr==&RBH_Save_1_info) ||
2664 (info_ptr==&RBH_Save_2_info));
2666 /* cf. convertToRBH in RBH.c for writing the RBHSave closure */
2667 ((StgRBH *)node)->blocking_queue = (StgBlockingQueueElement *)((StgRBHSave *)bqe)->payload[0];
2668 ((StgRBH *)node)->mut_link = (StgMutClosure *)((StgRBHSave *)bqe)->payload[1];
2671 belch("## Filled in RBH_Save for %p (%s) at end of AwBQ",
2672 node, info_type(node)));
2675 /* statistics gathering */
2676 if (RtsFlags.GranFlags.GranSimStats.Global) {
2677 // globalGranStats.tot_bq_processing_time += bq_processing_time;
2678 globalGranStats.tot_bq_len += len; // total length of all bqs awakened
2679 // globalGranStats.tot_bq_len_local += len_local; // same for local TSOs only
2680 globalGranStats.tot_awbq++; // total no. of bqs awakened
2683 fprintf(stderr,"## BQ Stats of %p: [%d entries] %s\n",
2684 node, len, (bqe!=END_BQ_QUEUE) ? "RBH" : ""));
2688 awakenBlockedQueue(StgBlockingQueueElement *q, StgClosure *node)
2690 StgBlockingQueueElement *bqe;
2692 ACQUIRE_LOCK(&sched_mutex);
2694 IF_PAR_DEBUG(verbose,
2695 belch("##-_ AwBQ for node %p on [%x]: ",
2699 if(get_itbl(q)->type == CONSTR || q==END_BQ_QUEUE) {
2700 IF_PAR_DEBUG(verbose, belch("## ... nothing to unblock so lets just return. RFP (BUG?)"));
2705 ASSERT(q == END_BQ_QUEUE ||
2706 get_itbl(q)->type == TSO ||
2707 get_itbl(q)->type == BLOCKED_FETCH ||
2708 get_itbl(q)->type == CONSTR);
2711 while (get_itbl(bqe)->type==TSO ||
2712 get_itbl(bqe)->type==BLOCKED_FETCH) {
2713 bqe = unblockOneLocked(bqe, node);
2715 RELEASE_LOCK(&sched_mutex);
2718 #else /* !GRAN && !PAR */
2720 awakenBlockedQueue(StgTSO *tso)
2722 ACQUIRE_LOCK(&sched_mutex);
2723 while (tso != END_TSO_QUEUE) {
2724 tso = unblockOneLocked(tso);
2726 RELEASE_LOCK(&sched_mutex);
2730 //@node Exception Handling Routines, Debugging Routines, Blocking Queue Routines, Main scheduling code
2731 //@subsection Exception Handling Routines
2733 /* ---------------------------------------------------------------------------
2735 - usually called inside a signal handler so it mustn't do anything fancy.
2736 ------------------------------------------------------------------------ */
2739 interruptStgRts(void)
2745 /* -----------------------------------------------------------------------------
2748 This is for use when we raise an exception in another thread, which
2750 This has nothing to do with the UnblockThread event in GranSim. -- HWL
2751 -------------------------------------------------------------------------- */
2753 #if defined(GRAN) || defined(PAR)
2755 NB: only the type of the blocking queue is different in GranSim and GUM
2756 the operations on the queue-elements are the same
2757 long live polymorphism!
2760 unblockThread(StgTSO *tso)
2762 StgBlockingQueueElement *t, **last;
2764 ACQUIRE_LOCK(&sched_mutex);
2765 switch (tso->why_blocked) {
2768 return; /* not blocked */
2771 ASSERT(get_itbl(tso->block_info.closure)->type == MVAR);
2773 StgBlockingQueueElement *last_tso = END_BQ_QUEUE;
2774 StgMVar *mvar = (StgMVar *)(tso->block_info.closure);
2776 last = (StgBlockingQueueElement **)&mvar->head;
2777 for (t = (StgBlockingQueueElement *)mvar->head;
2779 last = &t->link, last_tso = t, t = t->link) {
2780 if (t == (StgBlockingQueueElement *)tso) {
2781 *last = (StgBlockingQueueElement *)tso->link;
2782 if (mvar->tail == tso) {
2783 mvar->tail = (StgTSO *)last_tso;
2788 barf("unblockThread (MVAR): TSO not found");
2791 case BlockedOnBlackHole:
2792 ASSERT(get_itbl(tso->block_info.closure)->type == BLACKHOLE_BQ);
2794 StgBlockingQueue *bq = (StgBlockingQueue *)(tso->block_info.closure);
2796 last = &bq->blocking_queue;
2797 for (t = bq->blocking_queue;
2799 last = &t->link, t = t->link) {
2800 if (t == (StgBlockingQueueElement *)tso) {
2801 *last = (StgBlockingQueueElement *)tso->link;
2805 barf("unblockThread (BLACKHOLE): TSO not found");
2808 case BlockedOnException:
2810 StgTSO *target = tso->block_info.tso;
2812 ASSERT(get_itbl(target)->type == TSO);
2814 if (target->what_next == ThreadRelocated) {
2815 target = target->link;
2816 ASSERT(get_itbl(target)->type == TSO);
2819 ASSERT(target->blocked_exceptions != NULL);
2821 last = (StgBlockingQueueElement **)&target->blocked_exceptions;
2822 for (t = (StgBlockingQueueElement *)target->blocked_exceptions;
2824 last = &t->link, t = t->link) {
2825 ASSERT(get_itbl(t)->type == TSO);
2826 if (t == (StgBlockingQueueElement *)tso) {
2827 *last = (StgBlockingQueueElement *)tso->link;
2831 barf("unblockThread (Exception): TSO not found");
2835 case BlockedOnWrite:
2837 /* take TSO off blocked_queue */
2838 StgBlockingQueueElement *prev = NULL;
2839 for (t = (StgBlockingQueueElement *)blocked_queue_hd; t != END_BQ_QUEUE;
2840 prev = t, t = t->link) {
2841 if (t == (StgBlockingQueueElement *)tso) {
2843 blocked_queue_hd = (StgTSO *)t->link;
2844 if ((StgBlockingQueueElement *)blocked_queue_tl == t) {
2845 blocked_queue_tl = END_TSO_QUEUE;
2848 prev->link = t->link;
2849 if ((StgBlockingQueueElement *)blocked_queue_tl == t) {
2850 blocked_queue_tl = (StgTSO *)prev;
2856 barf("unblockThread (I/O): TSO not found");
2859 case BlockedOnDelay:
2861 /* take TSO off sleeping_queue */
2862 StgBlockingQueueElement *prev = NULL;
2863 for (t = (StgBlockingQueueElement *)sleeping_queue; t != END_BQ_QUEUE;
2864 prev = t, t = t->link) {
2865 if (t == (StgBlockingQueueElement *)tso) {
2867 sleeping_queue = (StgTSO *)t->link;
2869 prev->link = t->link;
2874 barf("unblockThread (I/O): TSO not found");
2878 barf("unblockThread");
2882 tso->link = END_TSO_QUEUE;
2883 tso->why_blocked = NotBlocked;
2884 tso->block_info.closure = NULL;
2885 PUSH_ON_RUN_QUEUE(tso);
2886 RELEASE_LOCK(&sched_mutex);
2890 unblockThread(StgTSO *tso)
2894 ACQUIRE_LOCK(&sched_mutex);
2895 switch (tso->why_blocked) {
2898 return; /* not blocked */
2901 ASSERT(get_itbl(tso->block_info.closure)->type == MVAR);
2903 StgTSO *last_tso = END_TSO_QUEUE;
2904 StgMVar *mvar = (StgMVar *)(tso->block_info.closure);
2907 for (t = mvar->head; t != END_TSO_QUEUE;
2908 last = &t->link, last_tso = t, t = t->link) {
2911 if (mvar->tail == tso) {
2912 mvar->tail = last_tso;
2917 barf("unblockThread (MVAR): TSO not found");
2920 case BlockedOnBlackHole:
2921 ASSERT(get_itbl(tso->block_info.closure)->type == BLACKHOLE_BQ);
2923 StgBlockingQueue *bq = (StgBlockingQueue *)(tso->block_info.closure);
2925 last = &bq->blocking_queue;
2926 for (t = bq->blocking_queue; t != END_TSO_QUEUE;
2927 last = &t->link, t = t->link) {
2933 barf("unblockThread (BLACKHOLE): TSO not found");
2936 case BlockedOnException:
2938 StgTSO *target = tso->block_info.tso;
2940 ASSERT(get_itbl(target)->type == TSO);
2942 while (target->what_next == ThreadRelocated) {
2943 target = target->link;
2944 ASSERT(get_itbl(target)->type == TSO);
2947 ASSERT(target->blocked_exceptions != NULL);
2949 last = &target->blocked_exceptions;
2950 for (t = target->blocked_exceptions; t != END_TSO_QUEUE;
2951 last = &t->link, t = t->link) {
2952 ASSERT(get_itbl(t)->type == TSO);
2958 barf("unblockThread (Exception): TSO not found");
2962 case BlockedOnWrite:
2964 StgTSO *prev = NULL;
2965 for (t = blocked_queue_hd; t != END_TSO_QUEUE;
2966 prev = t, t = t->link) {
2969 blocked_queue_hd = t->link;
2970 if (blocked_queue_tl == t) {
2971 blocked_queue_tl = END_TSO_QUEUE;
2974 prev->link = t->link;
2975 if (blocked_queue_tl == t) {
2976 blocked_queue_tl = prev;
2982 barf("unblockThread (I/O): TSO not found");
2985 case BlockedOnDelay:
2987 StgTSO *prev = NULL;
2988 for (t = sleeping_queue; t != END_TSO_QUEUE;
2989 prev = t, t = t->link) {
2992 sleeping_queue = t->link;
2994 prev->link = t->link;
2999 barf("unblockThread (I/O): TSO not found");
3003 barf("unblockThread");
3007 tso->link = END_TSO_QUEUE;
3008 tso->why_blocked = NotBlocked;
3009 tso->block_info.closure = NULL;
3010 PUSH_ON_RUN_QUEUE(tso);
3011 RELEASE_LOCK(&sched_mutex);
3015 /* -----------------------------------------------------------------------------
3018 * The following function implements the magic for raising an
3019 * asynchronous exception in an existing thread.
3021 * We first remove the thread from any queue on which it might be
3022 * blocked. The possible blockages are MVARs and BLACKHOLE_BQs.
3024 * We strip the stack down to the innermost CATCH_FRAME, building
3025 * thunks in the heap for all the active computations, so they can
3026 * be restarted if necessary. When we reach a CATCH_FRAME, we build
3027 * an application of the handler to the exception, and push it on
3028 * the top of the stack.
3030 * How exactly do we save all the active computations? We create an
3031 * AP_UPD for every UpdateFrame on the stack. Entering one of these
3032 * AP_UPDs pushes everything from the corresponding update frame
3033 * upwards onto the stack. (Actually, it pushes everything up to the
3034 * next update frame plus a pointer to the next AP_UPD object.
3035 * Entering the next AP_UPD object pushes more onto the stack until we
3036 * reach the last AP_UPD object - at which point the stack should look
3037 * exactly as it did when we killed the TSO and we can continue
3038 * execution by entering the closure on top of the stack.
3040 * We can also kill a thread entirely - this happens if either (a) the
3041 * exception passed to raiseAsync is NULL, or (b) there's no
3042 * CATCH_FRAME on the stack. In either case, we strip the entire
3043 * stack and replace the thread with a zombie.
3045 * -------------------------------------------------------------------------- */
3048 deleteThread(StgTSO *tso)
3050 raiseAsync(tso,NULL);
3054 raiseAsync(StgTSO *tso, StgClosure *exception)
3056 StgUpdateFrame* su = tso->su;
3057 StgPtr sp = tso->sp;
3059 /* Thread already dead? */
3060 if (tso->what_next == ThreadComplete || tso->what_next == ThreadKilled) {
3064 IF_DEBUG(scheduler, sched_belch("raising exception in thread %ld.", tso->id));
3066 /* Remove it from any blocking queues */
3069 /* The stack freezing code assumes there's a closure pointer on
3070 * the top of the stack. This isn't always the case with compiled
3071 * code, so we have to push a dummy closure on the top which just
3072 * returns to the next return address on the stack.
3074 if ( LOOKS_LIKE_GHC_INFO((void*)*sp) ) {
3075 *(--sp) = (W_)&stg_dummy_ret_closure;
3079 nat words = ((P_)su - (P_)sp) - 1;
3083 /* If we find a CATCH_FRAME, and we've got an exception to raise,
3084 * then build PAP(handler,exception,realworld#), and leave it on
3085 * top of the stack ready to enter.
3087 if (get_itbl(su)->type == CATCH_FRAME && exception != NULL) {
3088 StgCatchFrame *cf = (StgCatchFrame *)su;
3089 /* we've got an exception to raise, so let's pass it to the
3090 * handler in this frame.
3092 ap = (StgAP_UPD *)allocate(sizeofW(StgPAP) + 2);
3093 TICK_ALLOC_UPD_PAP(3,0);
3094 SET_HDR(ap,&stg_PAP_info,cf->header.prof.ccs);
3097 ap->fun = cf->handler; /* :: Exception -> IO a */
3098 ap->payload[0] = exception;
3099 ap->payload[1] = ARG_TAG(0); /* realworld token */
3101 /* throw away the stack from Sp up to and including the
3104 sp = (P_)su + sizeofW(StgCatchFrame) - 1;
3107 /* Restore the blocked/unblocked state for asynchronous exceptions
3108 * at the CATCH_FRAME.
3110 * If exceptions were unblocked at the catch, arrange that they
3111 * are unblocked again after executing the handler by pushing an
3112 * unblockAsyncExceptions_ret stack frame.
3114 if (!cf->exceptions_blocked) {
3115 *(sp--) = (W_)&stg_unblockAsyncExceptionszh_ret_info;
3118 /* Ensure that async exceptions are blocked when running the handler.
3120 if (tso->blocked_exceptions == NULL) {
3121 tso->blocked_exceptions = END_TSO_QUEUE;
3124 /* Put the newly-built PAP on top of the stack, ready to execute
3125 * when the thread restarts.
3129 tso->what_next = ThreadEnterGHC;
3130 IF_DEBUG(sanity, checkTSO(tso));
3134 /* First build an AP_UPD consisting of the stack chunk above the
3135 * current update frame, with the top word on the stack as the
3138 ap = (StgAP_UPD *)allocate(AP_sizeW(words));
3143 ap->fun = (StgClosure *)sp[0];
3145 for(i=0; i < (nat)words; ++i) {
3146 ap->payload[i] = (StgClosure *)*sp++;
3149 switch (get_itbl(su)->type) {
3153 SET_HDR(ap,&stg_AP_UPD_info,su->header.prof.ccs /* ToDo */);
3154 TICK_ALLOC_UP_THK(words+1,0);
3157 fprintf(stderr, "scheduler: Updating ");
3158 printPtr((P_)su->updatee);
3159 fprintf(stderr, " with ");
3160 printObj((StgClosure *)ap);
3163 /* Replace the updatee with an indirection - happily
3164 * this will also wake up any threads currently
3165 * waiting on the result.
3167 * Warning: if we're in a loop, more than one update frame on
3168 * the stack may point to the same object. Be careful not to
3169 * overwrite an IND_OLDGEN in this case, because we'll screw
3170 * up the mutable lists. To be on the safe side, don't
3171 * overwrite any kind of indirection at all. See also
3172 * threadSqueezeStack in GC.c, where we have to make a similar
3175 if (!closure_IND(su->updatee)) {
3176 UPD_IND_NOLOCK(su->updatee,ap); /* revert the black hole */
3179 sp += sizeofW(StgUpdateFrame) -1;
3180 sp[0] = (W_)ap; /* push onto stack */
3186 StgCatchFrame *cf = (StgCatchFrame *)su;
3189 /* We want a PAP, not an AP_UPD. Fortunately, the
3190 * layout's the same.
3192 SET_HDR(ap,&stg_PAP_info,su->header.prof.ccs /* ToDo */);
3193 TICK_ALLOC_UPD_PAP(words+1,0);
3195 /* now build o = FUN(catch,ap,handler) */
3196 o = (StgClosure *)allocate(sizeofW(StgClosure)+2);
3197 TICK_ALLOC_FUN(2,0);
3198 SET_HDR(o,&stg_catch_info,su->header.prof.ccs /* ToDo */);
3199 o->payload[0] = (StgClosure *)ap;
3200 o->payload[1] = cf->handler;
3203 fprintf(stderr, "scheduler: Built ");
3204 printObj((StgClosure *)o);
3207 /* pop the old handler and put o on the stack */
3209 sp += sizeofW(StgCatchFrame) - 1;
3216 StgSeqFrame *sf = (StgSeqFrame *)su;
3219 SET_HDR(ap,&stg_PAP_info,su->header.prof.ccs /* ToDo */);
3220 TICK_ALLOC_UPD_PAP(words+1,0);
3222 /* now build o = FUN(seq,ap) */
3223 o = (StgClosure *)allocate(sizeofW(StgClosure)+1);
3224 TICK_ALLOC_SE_THK(1,0);
3225 SET_HDR(o,&stg_seq_info,su->header.prof.ccs /* ToDo */);
3226 o->payload[0] = (StgClosure *)ap;
3229 fprintf(stderr, "scheduler: Built ");
3230 printObj((StgClosure *)o);
3233 /* pop the old handler and put o on the stack */
3235 sp += sizeofW(StgSeqFrame) - 1;
3241 /* We've stripped the entire stack, the thread is now dead. */
3242 sp += sizeofW(StgStopFrame) - 1;
3243 sp[0] = (W_)exception; /* save the exception */
3244 tso->what_next = ThreadKilled;
3245 tso->su = (StgUpdateFrame *)(sp+1);
3256 /* -----------------------------------------------------------------------------
3257 resurrectThreads is called after garbage collection on the list of
3258 threads found to be garbage. Each of these threads will be woken
3259 up and sent a signal: BlockedOnDeadMVar if the thread was blocked
3260 on an MVar, or NonTermination if the thread was blocked on a Black
3262 -------------------------------------------------------------------------- */
3265 resurrectThreads( StgTSO *threads )
3269 for (tso = threads; tso != END_TSO_QUEUE; tso = next) {
3270 next = tso->global_link;
3271 tso->global_link = all_threads;
3273 IF_DEBUG(scheduler, sched_belch("resurrecting thread %d", tso->id));
3275 switch (tso->why_blocked) {
3277 case BlockedOnException:
3278 raiseAsync(tso,(StgClosure *)BlockedOnDeadMVar_closure);
3280 case BlockedOnBlackHole:
3281 raiseAsync(tso,(StgClosure *)NonTermination_closure);
3284 /* This might happen if the thread was blocked on a black hole
3285 * belonging to a thread that we've just woken up (raiseAsync
3286 * can wake up threads, remember...).
3290 barf("resurrectThreads: thread blocked in a strange way");
3295 /* -----------------------------------------------------------------------------
3296 * Blackhole detection: if we reach a deadlock, test whether any
3297 * threads are blocked on themselves. Any threads which are found to
3298 * be self-blocked get sent a NonTermination exception.
3300 * This is only done in a deadlock situation in order to avoid
3301 * performance overhead in the normal case.
3302 * -------------------------------------------------------------------------- */
3305 detectBlackHoles( void )
3307 StgTSO *t = all_threads;
3308 StgUpdateFrame *frame;
3309 StgClosure *blocked_on;
3311 for (t = all_threads; t != END_TSO_QUEUE; t = t->global_link) {
3313 while (t->what_next == ThreadRelocated) {
3315 ASSERT(get_itbl(t)->type == TSO);
3318 if (t->why_blocked != BlockedOnBlackHole) {
3322 blocked_on = t->block_info.closure;
3324 for (frame = t->su; ; frame = frame->link) {
3325 switch (get_itbl(frame)->type) {
3328 if (frame->updatee == blocked_on) {
3329 /* We are blocking on one of our own computations, so
3330 * send this thread the NonTermination exception.
3333 sched_belch("thread %d is blocked on itself", t->id));
3334 raiseAsync(t, (StgClosure *)NonTermination_closure);
3355 //@node Debugging Routines, Index, Exception Handling Routines, Main scheduling code
3356 //@subsection Debugging Routines
3358 /* -----------------------------------------------------------------------------
3359 Debugging: why is a thread blocked
3360 -------------------------------------------------------------------------- */
3365 printThreadBlockage(StgTSO *tso)
3367 switch (tso->why_blocked) {
3369 fprintf(stderr,"is blocked on read from fd %d", tso->block_info.fd);
3371 case BlockedOnWrite:
3372 fprintf(stderr,"is blocked on write to fd %d", tso->block_info.fd);
3374 case BlockedOnDelay:
3375 fprintf(stderr,"is blocked until %d", tso->block_info.target);
3378 fprintf(stderr,"is blocked on an MVar");
3380 case BlockedOnException:
3381 fprintf(stderr,"is blocked on delivering an exception to thread %d",
3382 tso->block_info.tso->id);
3384 case BlockedOnBlackHole:
3385 fprintf(stderr,"is blocked on a black hole");
3388 fprintf(stderr,"is not blocked");
3392 fprintf(stderr,"is blocked on global address; local FM_BQ is %p (%s)",
3393 tso->block_info.closure, info_type(tso->block_info.closure));
3395 case BlockedOnGA_NoSend:
3396 fprintf(stderr,"is blocked on global address (no send); local FM_BQ is %p (%s)",
3397 tso->block_info.closure, info_type(tso->block_info.closure));
3401 barf("printThreadBlockage: strange tso->why_blocked: %d for TSO %d (%d)",
3402 tso->why_blocked, tso->id, tso);
3407 printThreadStatus(StgTSO *tso)
3409 switch (tso->what_next) {
3411 fprintf(stderr,"has been killed");
3413 case ThreadComplete:
3414 fprintf(stderr,"has completed");
3417 printThreadBlockage(tso);
3422 printAllThreads(void)
3427 char time_string[TIME_STR_LEN], node_str[NODE_STR_LEN];
3428 ullong_format_string(TIME_ON_PROC(CurrentProc),
3429 time_string, rtsFalse/*no commas!*/);
3431 sched_belch("all threads at [%s]:", time_string);
3433 char time_string[TIME_STR_LEN], node_str[NODE_STR_LEN];
3434 ullong_format_string(CURRENT_TIME,
3435 time_string, rtsFalse/*no commas!*/);
3437 sched_belch("all threads at [%s]:", time_string);
3439 sched_belch("all threads:");
3442 for (t = all_threads; t != END_TSO_QUEUE; t = t->global_link) {
3443 fprintf(stderr, "\tthread %d ", t->id);
3444 printThreadStatus(t);
3445 fprintf(stderr,"\n");
3450 Print a whole blocking queue attached to node (debugging only).
3455 print_bq (StgClosure *node)
3457 StgBlockingQueueElement *bqe;
3461 fprintf(stderr,"## BQ of closure %p (%s): ",
3462 node, info_type(node));
3464 /* should cover all closures that may have a blocking queue */
3465 ASSERT(get_itbl(node)->type == BLACKHOLE_BQ ||
3466 get_itbl(node)->type == FETCH_ME_BQ ||
3467 get_itbl(node)->type == RBH ||
3468 get_itbl(node)->type == MVAR);
3470 ASSERT(node!=(StgClosure*)NULL); // sanity check
3472 print_bqe(((StgBlockingQueue*)node)->blocking_queue);
3476 Print a whole blocking queue starting with the element bqe.
3479 print_bqe (StgBlockingQueueElement *bqe)
3484 NB: In a parallel setup a BQ of an RBH must end with an RBH_Save closure;
3486 for (end = (bqe==END_BQ_QUEUE);
3487 !end; // iterate until bqe points to a CONSTR
3488 end = (get_itbl(bqe)->type == CONSTR) || (bqe->link==END_BQ_QUEUE),
3489 bqe = end ? END_BQ_QUEUE : bqe->link) {
3490 ASSERT(bqe != END_BQ_QUEUE); // sanity check
3491 ASSERT(bqe != (StgBlockingQueueElement *)NULL); // sanity check
3492 /* types of closures that may appear in a blocking queue */
3493 ASSERT(get_itbl(bqe)->type == TSO ||
3494 get_itbl(bqe)->type == BLOCKED_FETCH ||
3495 get_itbl(bqe)->type == CONSTR);
3496 /* only BQs of an RBH end with an RBH_Save closure */
3497 //ASSERT(get_itbl(bqe)->type != CONSTR || get_itbl(node)->type == RBH);
3499 switch (get_itbl(bqe)->type) {
3501 fprintf(stderr," TSO %u (%x),",
3502 ((StgTSO *)bqe)->id, ((StgTSO *)bqe));
3505 fprintf(stderr," BF (node=%p, ga=((%x, %d, %x)),",
3506 ((StgBlockedFetch *)bqe)->node,
3507 ((StgBlockedFetch *)bqe)->ga.payload.gc.gtid,
3508 ((StgBlockedFetch *)bqe)->ga.payload.gc.slot,
3509 ((StgBlockedFetch *)bqe)->ga.weight);
3512 fprintf(stderr," %s (IP %p),",
3513 (get_itbl(bqe) == &stg_RBH_Save_0_info ? "RBH_Save_0" :
3514 get_itbl(bqe) == &stg_RBH_Save_1_info ? "RBH_Save_1" :
3515 get_itbl(bqe) == &stg_RBH_Save_2_info ? "RBH_Save_2" :
3516 "RBH_Save_?"), get_itbl(bqe));
3519 barf("Unexpected closure type %s in blocking queue", // of %p (%s)",
3520 info_type((StgClosure *)bqe)); // , node, info_type(node));
3524 fputc('\n', stderr);
3526 # elif defined(GRAN)
3528 print_bq (StgClosure *node)
3530 StgBlockingQueueElement *bqe;
3531 PEs node_loc, tso_loc;
3534 /* should cover all closures that may have a blocking queue */
3535 ASSERT(get_itbl(node)->type == BLACKHOLE_BQ ||
3536 get_itbl(node)->type == FETCH_ME_BQ ||
3537 get_itbl(node)->type == RBH);
3539 ASSERT(node!=(StgClosure*)NULL); // sanity check
3540 node_loc = where_is(node);
3542 fprintf(stderr,"## BQ of closure %p (%s) on [PE %d]: ",
3543 node, info_type(node), node_loc);
3546 NB: In a parallel setup a BQ of an RBH must end with an RBH_Save closure;
3548 for (bqe = ((StgBlockingQueue*)node)->blocking_queue, end = (bqe==END_BQ_QUEUE);
3549 !end; // iterate until bqe points to a CONSTR
3550 end = (get_itbl(bqe)->type == CONSTR) || (bqe->link==END_BQ_QUEUE), bqe = end ? END_BQ_QUEUE : bqe->link) {
3551 ASSERT(bqe != END_BQ_QUEUE); // sanity check
3552 ASSERT(bqe != (StgBlockingQueueElement *)NULL); // sanity check
3553 /* types of closures that may appear in a blocking queue */
3554 ASSERT(get_itbl(bqe)->type == TSO ||
3555 get_itbl(bqe)->type == CONSTR);
3556 /* only BQs of an RBH end with an RBH_Save closure */
3557 ASSERT(get_itbl(bqe)->type != CONSTR || get_itbl(node)->type == RBH);
3559 tso_loc = where_is((StgClosure *)bqe);
3560 switch (get_itbl(bqe)->type) {
3562 fprintf(stderr," TSO %d (%p) on [PE %d],",
3563 ((StgTSO *)bqe)->id, (StgTSO *)bqe, tso_loc);
3566 fprintf(stderr," %s (IP %p),",
3567 (get_itbl(bqe) == &stg_RBH_Save_0_info ? "RBH_Save_0" :
3568 get_itbl(bqe) == &stg_RBH_Save_1_info ? "RBH_Save_1" :
3569 get_itbl(bqe) == &stg_RBH_Save_2_info ? "RBH_Save_2" :
3570 "RBH_Save_?"), get_itbl(bqe));
3573 barf("Unexpected closure type %s in blocking queue of %p (%s)",
3574 info_type((StgClosure *)bqe), node, info_type(node));
3578 fputc('\n', stderr);
3582 Nice and easy: only TSOs on the blocking queue
3585 print_bq (StgClosure *node)
3589 ASSERT(node!=(StgClosure*)NULL); // sanity check
3590 for (tso = ((StgBlockingQueue*)node)->blocking_queue;
3591 tso != END_TSO_QUEUE;
3593 ASSERT(tso!=NULL && tso!=END_TSO_QUEUE); // sanity check
3594 ASSERT(get_itbl(tso)->type == TSO); // guess what, sanity check
3595 fprintf(stderr," TSO %d (%p),", tso->id, tso);
3597 fputc('\n', stderr);
3608 for (i=0, tso=run_queue_hd;
3609 tso != END_TSO_QUEUE;
3618 sched_belch(char *s, ...)
3623 fprintf(stderr, "scheduler (task %ld): ", osThreadId());
3625 fprintf(stderr, "== ");
3627 fprintf(stderr, "scheduler: ");
3629 vfprintf(stderr, s, ap);
3630 fprintf(stderr, "\n");
3636 //@node Index, , Debugging Routines, Main scheduling code
3640 //* MainRegTable:: @cindex\s-+MainRegTable
3641 //* StgMainThread:: @cindex\s-+StgMainThread
3642 //* awaken_blocked_queue:: @cindex\s-+awaken_blocked_queue
3643 //* blocked_queue_hd:: @cindex\s-+blocked_queue_hd
3644 //* blocked_queue_tl:: @cindex\s-+blocked_queue_tl
3645 //* context_switch:: @cindex\s-+context_switch
3646 //* createThread:: @cindex\s-+createThread
3647 //* gc_pending_cond:: @cindex\s-+gc_pending_cond
3648 //* initScheduler:: @cindex\s-+initScheduler
3649 //* interrupted:: @cindex\s-+interrupted
3650 //* next_thread_id:: @cindex\s-+next_thread_id
3651 //* print_bq:: @cindex\s-+print_bq
3652 //* run_queue_hd:: @cindex\s-+run_queue_hd
3653 //* run_queue_tl:: @cindex\s-+run_queue_tl
3654 //* sched_mutex:: @cindex\s-+sched_mutex
3655 //* schedule:: @cindex\s-+schedule
3656 //* take_off_run_queue:: @cindex\s-+take_off_run_queue
3657 //* term_mutex:: @cindex\s-+term_mutex
3658 //* thread_ready_cond:: @cindex\s-+thread_ready_cond