1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team 1998-2006
5 * Generational garbage collector
7 * Documentation on the architecture of the Garbage Collector can be
8 * found in the online commentary:
10 * http://hackage.haskell.org/trac/ghc/wiki/Commentary/Rts/Storage/GC
12 * ---------------------------------------------------------------------------*/
14 // #include "PosixSource.h"
19 #include "OSThreads.h"
20 #include "LdvProfile.h"
25 #include "BlockAlloc.h"
31 #include "ParTicky.h" // ToDo: move into Rts.h
32 #include "RtsSignals.h"
36 #if defined(RTS_GTK_FRONTPANEL)
37 #include "FrontPanel.h"
40 #include "RetainerProfile.h"
41 #include "RaiseAsync.h"
53 #include <string.h> // for memset()
56 /* -----------------------------------------------------------------------------
58 -------------------------------------------------------------------------- */
60 /* STATIC OBJECT LIST.
63 * We maintain a linked list of static objects that are still live.
64 * The requirements for this list are:
66 * - we need to scan the list while adding to it, in order to
67 * scavenge all the static objects (in the same way that
68 * breadth-first scavenging works for dynamic objects).
70 * - we need to be able to tell whether an object is already on
71 * the list, to break loops.
73 * Each static object has a "static link field", which we use for
74 * linking objects on to the list. We use a stack-type list, consing
75 * objects on the front as they are added (this means that the
76 * scavenge phase is depth-first, not breadth-first, but that
79 * A separate list is kept for objects that have been scavenged
80 * already - this is so that we can zero all the marks afterwards.
82 * An object is on the list if its static link field is non-zero; this
83 * means that we have to mark the end of the list with '1', not NULL.
85 * Extra notes for generational GC:
87 * Each generation has a static object list associated with it. When
88 * collecting generations up to N, we treat the static object lists
89 * from generations > N as roots.
91 * We build up a static object list while collecting generations 0..N,
92 * which is then appended to the static object list of generation N+1.
95 /* N is the oldest generation being collected, where the generations
96 * are numbered starting at 0. A major GC (indicated by the major_gc
97 * flag) is when we're collecting all generations. We only attempt to
98 * deal with static objects and GC CAFs when doing a major GC.
103 /* Data used for allocation area sizing.
105 static lnat g0s0_pcnt_kept = 30; // percentage of g0s0 live at last minor GC
115 /* Thread-local data for each GC thread
117 gc_thread **gc_threads = NULL;
118 // gc_thread *gct = NULL; // this thread's gct TODO: make thread-local
120 // Number of threads running in *this* GC. Affects how many
121 // step->todos[] lists we have to look in to find work.
125 long copied; // *words* copied & scavenged during this GC
128 SpinLock recordMutableGen_sync;
131 /* -----------------------------------------------------------------------------
132 Static function declarations
133 -------------------------------------------------------------------------- */
135 static void mark_root (StgClosure **root);
136 static void zero_static_object_list (StgClosure* first_static);
137 static void initialise_N (rtsBool force_major_gc);
138 static void alloc_gc_threads (void);
139 static void init_collected_gen (nat g, nat threads);
140 static void init_uncollected_gen (nat g, nat threads);
141 static void init_gc_thread (gc_thread *t);
142 static void update_task_list (void);
143 static void resize_generations (void);
144 static void resize_nursery (void);
145 static void start_gc_threads (void);
146 static void gc_thread_work (void);
147 static nat inc_running (void);
148 static nat dec_running (void);
149 static void wakeup_gc_threads (nat n_threads);
151 #if 0 && defined(DEBUG)
152 static void gcCAFs (void);
155 /* -----------------------------------------------------------------------------
156 The mark bitmap & stack.
157 -------------------------------------------------------------------------- */
159 #define MARK_STACK_BLOCKS 4
161 bdescr *mark_stack_bdescr;
166 // Flag and pointers used for falling back to a linear scan when the
167 // mark stack overflows.
168 rtsBool mark_stack_overflowed;
169 bdescr *oldgen_scan_bd;
172 /* -----------------------------------------------------------------------------
173 GarbageCollect: the main entry point to the garbage collector.
175 Locks held: all capabilities are held throughout GarbageCollect().
176 -------------------------------------------------------------------------- */
179 GarbageCollect ( rtsBool force_major_gc )
183 lnat live, allocated;
184 lnat oldgen_saved_blocks = 0;
185 gc_thread *saved_gct;
188 // necessary if we stole a callee-saves register for gct:
192 CostCentreStack *prev_CCS;
197 debugTrace(DEBUG_gc, "starting GC");
199 #if defined(RTS_USER_SIGNALS)
200 if (RtsFlags.MiscFlags.install_signal_handlers) {
206 // tell the stats department that we've started a GC
209 // tell the STM to discard any cached closures it's hoping to re-use
218 // attribute any costs to CCS_GC
224 /* Approximate how much we allocated.
225 * Todo: only when generating stats?
227 allocated = calcAllocated();
229 /* Figure out which generation to collect
231 initialise_N(force_major_gc);
233 /* Allocate + initialise the gc_thread structures.
237 /* Start threads, so they can be spinning up while we finish initialisation.
241 /* How many threads will be participating in this GC?
242 * We don't try to parallelise minor GC.
244 #if defined(THREADED_RTS)
248 n_gc_threads = RtsFlags.ParFlags.gcThreads;
254 #ifdef RTS_GTK_FRONTPANEL
255 if (RtsFlags.GcFlags.frontpanel) {
256 updateFrontPanelBeforeGC(N);
261 // check for memory leaks if DEBUG is on
262 memInventory(traceClass(DEBUG_gc));
265 // check stack sanity *before* GC (ToDo: check all threads)
266 IF_DEBUG(sanity, checkFreeListSanity());
268 // Initialise all the generations/steps that we're collecting.
269 for (g = 0; g <= N; g++) {
270 init_collected_gen(g,n_gc_threads);
273 // Initialise all the generations/steps that we're *not* collecting.
274 for (g = N+1; g < RtsFlags.GcFlags.generations; g++) {
275 init_uncollected_gen(g,n_gc_threads);
278 /* Allocate a mark stack if we're doing a major collection.
281 mark_stack_bdescr = allocGroup(MARK_STACK_BLOCKS);
282 mark_stack = (StgPtr *)mark_stack_bdescr->start;
283 mark_sp = mark_stack;
284 mark_splim = mark_stack + (MARK_STACK_BLOCKS * BLOCK_SIZE_W);
286 mark_stack_bdescr = NULL;
289 // Initialise all our gc_thread structures
290 for (t = 0; t < n_gc_threads; t++) {
291 init_gc_thread(gc_threads[t]);
294 // the main thread is running: this prevents any other threads from
295 // exiting prematurely, so we can start them now.
297 wakeup_gc_threads(n_gc_threads);
299 // this is the main thread
302 /* -----------------------------------------------------------------------
303 * follow all the roots that we know about:
304 * - mutable lists from each generation > N
305 * we want to *scavenge* these roots, not evacuate them: they're not
306 * going to move in this GC.
307 * Also do them in reverse generation order, for the usual reason:
308 * namely to reduce the likelihood of spurious old->new pointers.
311 for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
312 generations[g].saved_mut_list = generations[g].mut_list;
313 generations[g].mut_list = allocBlock();
314 // mut_list always has at least one block.
316 for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
317 scavenge_mutable_list(&generations[g]);
321 // follow roots from the CAF list (used by GHCi)
325 // follow all the roots that the application knows about.
329 #if defined(RTS_USER_SIGNALS)
330 // mark the signal handlers (signals should be already blocked)
331 markSignalHandlers(mark_root);
334 // Mark the weak pointer list, and prepare to detect dead weak pointers.
338 // Mark the stable pointer table.
339 markStablePtrTable(mark_root);
341 /* -------------------------------------------------------------------------
342 * Repeatedly scavenge all the areas we know about until there's no
343 * more scavenging to be done.
348 // The other threads are now stopped. We might recurse back to
349 // here, but from now on this is the only thread.
351 // if any blackholes are alive, make the threads that wait on
353 if (traverseBlackholeQueue()) {
358 // must be last... invariant is that everything is fully
359 // scavenged at this point.
360 if (traverseWeakPtrList()) { // returns rtsTrue if evaced something
365 // If we get to here, there's really nothing left to do.
369 // Update pointers from the Task list
372 // Now see which stable names are still alive.
376 // We call processHeapClosureForDead() on every closure destroyed during
377 // the current garbage collection, so we invoke LdvCensusForDead().
378 if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_LDV
379 || RtsFlags.ProfFlags.bioSelector != NULL)
383 // NO MORE EVACUATION AFTER THIS POINT!
384 // Finally: compaction of the oldest generation.
385 if (major_gc && oldest_gen->steps[0].is_compacted) {
386 // save number of blocks for stats
387 oldgen_saved_blocks = oldest_gen->steps[0].n_old_blocks;
391 IF_DEBUG(sanity, checkGlobalTSOList(rtsFalse));
393 // Two-space collector: free the old to-space.
394 // g0s0->old_blocks is the old nursery
395 // g0s0->blocks is to-space from the previous GC
396 if (RtsFlags.GcFlags.generations == 1) {
397 if (g0s0->blocks != NULL) {
398 freeChain(g0s0->blocks);
403 // For each workspace, in each thread:
404 // * clear the BF_EVACUATED flag from each copied block
405 // * move the copied blocks to the step
411 for (t = 0; t < n_gc_threads; t++) {
415 for (s = 1; s < total_steps; s++) {
418 // ASSERT( ws->scan_bd == ws->todo_bd );
419 ASSERT( ws->scan_bd ? ws->scan == ws->scan_bd->free : 1 );
421 // Push the final block
422 if (ws->scan_bd) { push_scan_block(ws->scan_bd, ws); }
424 ASSERT(countBlocks(ws->scavd_list) == ws->n_scavd_blocks);
426 prev = ws->scavd_list;
427 for (bd = ws->scavd_list; bd != NULL; bd = bd->link) {
428 bd->flags &= ~BF_EVACUATED; // now from-space
431 prev->link = ws->stp->blocks;
432 ws->stp->blocks = ws->scavd_list;
433 ws->stp->n_blocks += ws->n_scavd_blocks;
434 ASSERT(countBlocks(ws->stp->blocks) == ws->stp->n_blocks);
439 // Two-space collector: swap the semi-spaces around.
440 // Currently: g0s0->old_blocks is the old nursery
441 // g0s0->blocks is to-space from this GC
442 // We want these the other way around.
443 if (RtsFlags.GcFlags.generations == 1) {
444 bdescr *nursery_blocks = g0s0->old_blocks;
445 nat n_nursery_blocks = g0s0->n_old_blocks;
446 g0s0->old_blocks = g0s0->blocks;
447 g0s0->n_old_blocks = g0s0->n_blocks;
448 g0s0->blocks = nursery_blocks;
449 g0s0->n_blocks = n_nursery_blocks;
452 /* run through all the generations/steps and tidy up
457 for (i=0; i < n_gc_threads; i++) {
459 trace(TRACE_gc,"thread %d:", i);
460 trace(TRACE_gc," copied %ld", gc_threads[i]->copied * sizeof(W_));
461 trace(TRACE_gc," any_work %ld", gc_threads[i]->any_work);
462 trace(TRACE_gc," no_work %ld", gc_threads[i]->no_work);
463 trace(TRACE_gc," scav_global_work %ld", gc_threads[i]->scav_global_work);
464 trace(TRACE_gc," scav_local_work %ld", gc_threads[i]->scav_local_work);
466 copied += gc_threads[i]->copied;
470 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
473 generations[g].collections++; // for stats
476 // Count the mutable list as bytes "copied" for the purposes of
477 // stats. Every mutable list is copied during every GC.
479 nat mut_list_size = 0;
480 for (bd = generations[g].mut_list; bd != NULL; bd = bd->link) {
481 mut_list_size += bd->free - bd->start;
483 copied += mut_list_size;
486 "mut_list_size: %lu (%d vars, %d arrays, %d MVARs, %d others)",
487 (unsigned long)(mut_list_size * sizeof(W_)),
488 mutlist_MUTVARS, mutlist_MUTARRS, mutlist_MVARS, mutlist_OTHERS);
491 for (s = 0; s < generations[g].n_steps; s++) {
493 stp = &generations[g].steps[s];
495 // for generations we collected...
498 /* free old memory and shift to-space into from-space for all
499 * the collected steps (except the allocation area). These
500 * freed blocks will probaby be quickly recycled.
502 if (!(g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1)) {
503 if (stp->is_compacted)
505 // for a compacted step, just shift the new to-space
506 // onto the front of the now-compacted existing blocks.
507 for (bd = stp->blocks; bd != NULL; bd = bd->link) {
508 bd->flags &= ~BF_EVACUATED; // now from-space
510 // tack the new blocks on the end of the existing blocks
511 if (stp->old_blocks != NULL) {
512 for (bd = stp->old_blocks; bd != NULL; bd = next) {
513 // NB. this step might not be compacted next
514 // time, so reset the BF_COMPACTED flags.
515 // They are set before GC if we're going to
516 // compact. (search for BF_COMPACTED above).
517 bd->flags &= ~BF_COMPACTED;
520 bd->link = stp->blocks;
523 stp->blocks = stp->old_blocks;
525 // add the new blocks to the block tally
526 stp->n_blocks += stp->n_old_blocks;
527 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
531 freeChain(stp->old_blocks);
533 stp->old_blocks = NULL;
534 stp->n_old_blocks = 0;
537 /* LARGE OBJECTS. The current live large objects are chained on
538 * scavenged_large, having been moved during garbage
539 * collection from large_objects. Any objects left on
540 * large_objects list are therefore dead, so we free them here.
542 for (bd = stp->large_objects; bd != NULL; bd = next) {
548 // update the count of blocks used by large objects
549 for (bd = stp->scavenged_large_objects; bd != NULL; bd = bd->link) {
550 bd->flags &= ~BF_EVACUATED;
552 stp->large_objects = stp->scavenged_large_objects;
553 stp->n_large_blocks = stp->n_scavenged_large_blocks;
556 else // for older generations...
558 /* For older generations, we need to append the
559 * scavenged_large_object list (i.e. large objects that have been
560 * promoted during this GC) to the large_object list for that step.
562 for (bd = stp->scavenged_large_objects; bd; bd = next) {
564 bd->flags &= ~BF_EVACUATED;
565 dbl_link_onto(bd, &stp->large_objects);
568 // add the new blocks we promoted during this GC
569 stp->n_large_blocks += stp->n_scavenged_large_blocks;
574 // update the max size of older generations after a major GC
575 resize_generations();
577 // Guess the amount of live data for stats.
578 live = calcLiveBlocks() * BLOCK_SIZE_W;
579 debugTrace(DEBUG_gc, "Slop: %ldKB",
580 (live - calcLiveWords()) / (1024/sizeof(W_)));
582 // Free the small objects allocated via allocate(), since this will
583 // all have been copied into G0S1 now.
584 if (RtsFlags.GcFlags.generations > 1) {
585 if (g0s0->blocks != NULL) {
586 freeChain(g0s0->blocks);
592 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
594 // Start a new pinned_object_block
595 pinned_object_block = NULL;
597 // Free the mark stack.
598 if (mark_stack_bdescr != NULL) {
599 freeGroup(mark_stack_bdescr);
603 for (g = 0; g <= N; g++) {
604 for (s = 0; s < generations[g].n_steps; s++) {
605 stp = &generations[g].steps[s];
606 if (stp->bitmap != NULL) {
607 freeGroup(stp->bitmap);
615 // mark the garbage collected CAFs as dead
616 #if 0 && defined(DEBUG) // doesn't work at the moment
617 if (major_gc) { gcCAFs(); }
621 // resetStaticObjectForRetainerProfiling() must be called before
623 if (n_gc_threads > 1) {
624 barf("profiling is currently broken with multi-threaded GC");
625 // ToDo: fix the gct->scavenged_static_objects below
627 resetStaticObjectForRetainerProfiling(gct->scavenged_static_objects);
630 // zero the scavenged static object list
633 for (i = 0; i < n_gc_threads; i++) {
634 zero_static_object_list(gc_threads[i]->scavenged_static_objects);
641 // start any pending finalizers
643 scheduleFinalizers(last_free_capability, old_weak_ptr_list);
646 // send exceptions to any threads which were about to die
648 resurrectThreads(resurrected_threads);
651 // Update the stable pointer hash table.
652 updateStablePtrTable(major_gc);
654 // check sanity after GC
655 IF_DEBUG(sanity, checkSanity());
657 // extra GC trace info
658 if (traceClass(TRACE_gc)) statDescribeGens();
661 // symbol-table based profiling
662 /* heapCensus(to_blocks); */ /* ToDo */
665 // restore enclosing cost centre
671 // check for memory leaks if DEBUG is on
672 memInventory(traceClass(DEBUG_gc));
675 #ifdef RTS_GTK_FRONTPANEL
676 if (RtsFlags.GcFlags.frontpanel) {
677 updateFrontPanelAfterGC( N, live );
681 // ok, GC over: tell the stats department what happened.
682 stat_endGC(allocated, live, copied, N);
684 #if defined(RTS_USER_SIGNALS)
685 if (RtsFlags.MiscFlags.install_signal_handlers) {
686 // unblock signals again
687 unblockUserSignals();
696 /* -----------------------------------------------------------------------------
697 * Mark all nodes pointed to by sparks in the spark queues (for GC) Does an
698 * implicit slide i.e. after marking all sparks are at the beginning of the
699 * spark pool and the spark pool only contains sparkable closures
700 * -------------------------------------------------------------------------- */
704 markSparkQueue (evac_fn evac, Capability *cap)
706 StgClosure **sparkp, **to_sparkp;
707 nat n, pruned_sparks; // stats only
710 PAR_TICKY_MARK_SPARK_QUEUE_START();
715 pool = &(cap->r.rSparks);
717 ASSERT_SPARK_POOL_INVARIANTS(pool);
719 #if defined(PARALLEL_HASKELL)
726 to_sparkp = pool->hd;
727 while (sparkp != pool->tl) {
728 ASSERT(*sparkp!=NULL);
729 ASSERT(LOOKS_LIKE_CLOSURE_PTR(((StgClosure *)*sparkp)));
730 // ToDo?: statistics gathering here (also for GUM!)
731 if (closure_SHOULD_SPARK(*sparkp)) {
733 *to_sparkp++ = *sparkp;
734 if (to_sparkp == pool->lim) {
735 to_sparkp = pool->base;
742 if (sparkp == pool->lim) {
746 pool->tl = to_sparkp;
748 PAR_TICKY_MARK_SPARK_QUEUE_END(n);
750 #if defined(PARALLEL_HASKELL)
751 debugTrace(DEBUG_sched,
752 "marked %d sparks and pruned %d sparks on [%x]",
753 n, pruned_sparks, mytid);
755 debugTrace(DEBUG_sched,
756 "marked %d sparks and pruned %d sparks",
760 debugTrace(DEBUG_sched,
761 "new spark queue len=%d; (hd=%p; tl=%p)\n",
762 sparkPoolSize(pool), pool->hd, pool->tl);
766 /* ---------------------------------------------------------------------------
767 Where are the roots that we know about?
769 - all the threads on the runnable queue
770 - all the threads on the blocked queue
771 - all the threads on the sleeping queue
772 - all the thread currently executing a _ccall_GC
773 - all the "main threads"
775 ------------------------------------------------------------------------ */
778 GetRoots( evac_fn evac )
784 // Each GC thread is responsible for following roots from the
785 // Capability of the same number. There will usually be the same
786 // or fewer Capabilities as GC threads, but just in case there
787 // are more, we mark every Capability whose number is the GC
788 // thread's index plus a multiple of the number of GC threads.
789 for (i = gct->thread_index; i < n_capabilities; i += n_gc_threads) {
790 cap = &capabilities[i];
791 evac((StgClosure **)(void *)&cap->run_queue_hd);
792 evac((StgClosure **)(void *)&cap->run_queue_tl);
793 #if defined(THREADED_RTS)
794 evac((StgClosure **)(void *)&cap->wakeup_queue_hd);
795 evac((StgClosure **)(void *)&cap->wakeup_queue_tl);
797 for (task = cap->suspended_ccalling_tasks; task != NULL;
799 debugTrace(DEBUG_sched,
800 "evac'ing suspended TSO %lu", (unsigned long)task->suspended_tso->id);
801 evac((StgClosure **)(void *)&task->suspended_tso);
804 #if defined(THREADED_RTS)
805 markSparkQueue(evac,cap);
809 #if !defined(THREADED_RTS)
810 evac((StgClosure **)(void *)&blocked_queue_hd);
811 evac((StgClosure **)(void *)&blocked_queue_tl);
812 evac((StgClosure **)(void *)&sleeping_queue);
816 /* -----------------------------------------------------------------------------
817 isAlive determines whether the given closure is still alive (after
818 a garbage collection) or not. It returns the new address of the
819 closure if it is alive, or NULL otherwise.
821 NOTE: Use it before compaction only!
822 It untags and (if needed) retags pointers to closures.
823 -------------------------------------------------------------------------- */
827 isAlive(StgClosure *p)
829 const StgInfoTable *info;
835 /* The tag and the pointer are split, to be merged later when needed. */
836 tag = GET_CLOSURE_TAG(p);
837 q = UNTAG_CLOSURE(p);
839 ASSERT(LOOKS_LIKE_CLOSURE_PTR(q));
842 // ignore static closures
844 // ToDo: for static closures, check the static link field.
845 // Problem here is that we sometimes don't set the link field, eg.
846 // for static closures with an empty SRT or CONSTR_STATIC_NOCAFs.
848 if (!HEAP_ALLOCED(q)) {
852 // ignore closures in generations that we're not collecting.
854 if (bd->gen_no > N) {
858 // if it's a pointer into to-space, then we're done
859 if (bd->flags & BF_EVACUATED) {
863 // large objects use the evacuated flag
864 if (bd->flags & BF_LARGE) {
868 // check the mark bit for compacted steps
869 if ((bd->flags & BF_COMPACTED) && is_marked((P_)q,bd)) {
873 switch (info->type) {
878 case IND_OLDGEN: // rely on compatible layout with StgInd
879 case IND_OLDGEN_PERM:
880 // follow indirections
881 p = ((StgInd *)q)->indirectee;
886 return ((StgEvacuated *)q)->evacuee;
889 if (((StgTSO *)q)->what_next == ThreadRelocated) {
890 p = (StgClosure *)((StgTSO *)q)->link;
902 /* -----------------------------------------------------------------------------
903 Figure out which generation to collect, initialise N and major_gc.
904 -------------------------------------------------------------------------- */
907 initialise_N (rtsBool force_major_gc)
911 if (force_major_gc) {
912 N = RtsFlags.GcFlags.generations - 1;
916 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
917 if (generations[g].steps[0].n_blocks +
918 generations[g].steps[0].n_large_blocks
919 >= generations[g].max_blocks) {
923 major_gc = (N == RtsFlags.GcFlags.generations-1);
927 /* -----------------------------------------------------------------------------
928 Initialise the gc_thread structures.
929 -------------------------------------------------------------------------- */
932 alloc_gc_thread (int n)
938 t = stgMallocBytes(sizeof(gc_thread) + total_steps * sizeof(step_workspace),
943 initCondition(&t->wake_cond);
944 initMutex(&t->wake_mutex);
945 t->wakeup = rtsFalse;
950 t->free_blocks = NULL;
959 for (s = 0; s < total_steps; s++)
962 ws->stp = &all_steps[s];
963 ASSERT(s == ws->stp->abs_no);
970 ws->buffer_todo_bd = NULL;
972 ws->scavd_list = NULL;
973 ws->n_scavd_blocks = 0;
981 alloc_gc_threads (void)
983 if (gc_threads == NULL) {
984 #if defined(THREADED_RTS)
986 gc_threads = stgMallocBytes (RtsFlags.ParFlags.gcThreads *
990 for (i = 0; i < RtsFlags.ParFlags.gcThreads; i++) {
991 gc_threads[i] = alloc_gc_thread(i);
994 gc_threads = stgMallocBytes (sizeof(gc_thread*),
997 gc_threads[0] = alloc_gc_thread(0);
1002 /* ----------------------------------------------------------------------------
1004 ------------------------------------------------------------------------- */
1006 static nat gc_running_threads;
1008 #if defined(THREADED_RTS)
1009 static Mutex gc_running_mutex;
1016 ACQUIRE_LOCK(&gc_running_mutex);
1017 n_running = ++gc_running_threads;
1018 RELEASE_LOCK(&gc_running_mutex);
1026 ACQUIRE_LOCK(&gc_running_mutex);
1027 n_running = --gc_running_threads;
1028 RELEASE_LOCK(&gc_running_mutex);
1033 // gc_thread_work(): Scavenge until there's no work left to do and all
1034 // the running threads are idle.
1037 gc_thread_work (void)
1041 debugTrace(DEBUG_gc, "GC thread %d working", gct->thread_index);
1043 // gc_running_threads has already been incremented for us; either
1044 // this is the main thread and we incremented it inside
1045 // GarbageCollect(), or this is a worker thread and the main
1046 // thread bumped gc_running_threads before waking us up.
1048 // Every thread evacuates some roots.
1050 GetRoots(mark_root);
1054 // scavenge_loop() only exits when there's no work to do
1057 debugTrace(DEBUG_gc, "GC thread %d idle (%d still running)",
1058 gct->thread_index, r);
1060 while (gc_running_threads != 0) {
1066 // any_work() does not remove the work from the queue, it
1067 // just checks for the presence of work. If we find any,
1068 // then we increment gc_running_threads and go back to
1069 // scavenge_loop() to perform any pending work.
1072 // All threads are now stopped
1073 debugTrace(DEBUG_gc, "GC thread %d finished.", gct->thread_index);
1077 #if defined(THREADED_RTS)
1079 gc_thread_mainloop (void)
1081 while (!gct->exit) {
1083 // Wait until we're told to wake up
1084 ACQUIRE_LOCK(&gct->wake_mutex);
1085 while (!gct->wakeup) {
1086 debugTrace(DEBUG_gc, "GC thread %d standing by...",
1088 waitCondition(&gct->wake_cond, &gct->wake_mutex);
1090 RELEASE_LOCK(&gct->wake_mutex);
1091 gct->wakeup = rtsFalse;
1092 if (gct->exit) break;
1095 // start performance counters in this thread...
1096 if (gct->papi_events == -1) {
1097 papi_init_eventset(&gct->papi_events);
1099 papi_thread_start_gc1_count(gct->papi_events);
1105 // count events in this thread towards the GC totals
1106 papi_thread_stop_gc1_count(gct->papi_events);
1112 #if defined(THREADED_RTS)
1114 gc_thread_entry (gc_thread *my_gct)
1117 debugTrace(DEBUG_gc, "GC thread %d starting...", gct->thread_index);
1118 gct->id = osThreadId();
1119 gc_thread_mainloop();
1124 start_gc_threads (void)
1126 #if defined(THREADED_RTS)
1129 static rtsBool done = rtsFalse;
1131 gc_running_threads = 0;
1132 initMutex(&gc_running_mutex);
1135 // Start from 1: the main thread is 0
1136 for (i = 1; i < RtsFlags.ParFlags.gcThreads; i++) {
1137 createOSThread(&id, (OSThreadProc*)&gc_thread_entry,
1146 wakeup_gc_threads (nat n_threads USED_IF_THREADS)
1148 #if defined(THREADED_RTS)
1150 for (i=1; i < n_threads; i++) {
1152 ACQUIRE_LOCK(&gc_threads[i]->wake_mutex);
1153 gc_threads[i]->wakeup = rtsTrue;
1154 signalCondition(&gc_threads[i]->wake_cond);
1155 RELEASE_LOCK(&gc_threads[i]->wake_mutex);
1160 /* ----------------------------------------------------------------------------
1161 Initialise a generation that is to be collected
1162 ------------------------------------------------------------------------- */
1165 init_collected_gen (nat g, nat n_threads)
1172 // Throw away the current mutable list. Invariant: the mutable
1173 // list always has at least one block; this means we can avoid a
1174 // check for NULL in recordMutable().
1176 freeChain(generations[g].mut_list);
1177 generations[g].mut_list = allocBlock();
1178 for (i = 0; i < n_capabilities; i++) {
1179 freeChain(capabilities[i].mut_lists[g]);
1180 capabilities[i].mut_lists[g] = allocBlock();
1184 for (s = 0; s < generations[g].n_steps; s++) {
1186 // generation 0, step 0 doesn't need to-space
1187 if (g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1) {
1191 stp = &generations[g].steps[s];
1192 ASSERT(stp->gen_no == g);
1194 // deprecate the existing blocks
1195 stp->old_blocks = stp->blocks;
1196 stp->n_old_blocks = stp->n_blocks;
1200 // we don't have any to-be-scavenged blocks yet
1202 stp->todos_last = NULL;
1205 // initialise the large object queues.
1206 stp->scavenged_large_objects = NULL;
1207 stp->n_scavenged_large_blocks = 0;
1209 // mark the large objects as not evacuated yet
1210 for (bd = stp->large_objects; bd; bd = bd->link) {
1211 bd->flags &= ~BF_EVACUATED;
1214 // for a compacted step, we need to allocate the bitmap
1215 if (stp->is_compacted) {
1216 nat bitmap_size; // in bytes
1217 bdescr *bitmap_bdescr;
1220 bitmap_size = stp->n_old_blocks * BLOCK_SIZE / (sizeof(W_)*BITS_PER_BYTE);
1222 if (bitmap_size > 0) {
1223 bitmap_bdescr = allocGroup((lnat)BLOCK_ROUND_UP(bitmap_size)
1225 stp->bitmap = bitmap_bdescr;
1226 bitmap = bitmap_bdescr->start;
1228 debugTrace(DEBUG_gc, "bitmap_size: %d, bitmap: %p",
1229 bitmap_size, bitmap);
1231 // don't forget to fill it with zeros!
1232 memset(bitmap, 0, bitmap_size);
1234 // For each block in this step, point to its bitmap from the
1235 // block descriptor.
1236 for (bd=stp->old_blocks; bd != NULL; bd = bd->link) {
1237 bd->u.bitmap = bitmap;
1238 bitmap += BLOCK_SIZE_W / (sizeof(W_)*BITS_PER_BYTE);
1240 // Also at this point we set the BF_COMPACTED flag
1241 // for this block. The invariant is that
1242 // BF_COMPACTED is always unset, except during GC
1243 // when it is set on those blocks which will be
1245 bd->flags |= BF_COMPACTED;
1251 // For each GC thread, for each step, allocate a "todo" block to
1252 // store evacuated objects to be scavenged, and a block to store
1253 // evacuated objects that do not need to be scavenged.
1254 for (t = 0; t < n_threads; t++) {
1255 for (s = 0; s < generations[g].n_steps; s++) {
1257 // we don't copy objects into g0s0, unless -G0
1258 if (g==0 && s==0 && RtsFlags.GcFlags.generations > 1) continue;
1260 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1265 ws->todo_large_objects = NULL;
1267 // allocate the first to-space block; extra blocks will be
1268 // chained on as necessary.
1270 ws->buffer_todo_bd = NULL;
1271 gc_alloc_todo_block(ws);
1273 ws->scavd_list = NULL;
1274 ws->n_scavd_blocks = 0;
1280 /* ----------------------------------------------------------------------------
1281 Initialise a generation that is *not* to be collected
1282 ------------------------------------------------------------------------- */
1285 init_uncollected_gen (nat g, nat threads)
1292 for (s = 0; s < generations[g].n_steps; s++) {
1293 stp = &generations[g].steps[s];
1294 stp->scavenged_large_objects = NULL;
1295 stp->n_scavenged_large_blocks = 0;
1298 for (t = 0; t < threads; t++) {
1299 for (s = 0; s < generations[g].n_steps; s++) {
1301 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1304 ws->buffer_todo_bd = NULL;
1305 ws->todo_large_objects = NULL;
1307 ws->scavd_list = NULL;
1308 ws->n_scavd_blocks = 0;
1310 // If the block at the head of the list in this generation
1311 // is less than 3/4 full, then use it as a todo block.
1312 if (stp->blocks && isPartiallyFull(stp->blocks))
1314 ws->todo_bd = stp->blocks;
1315 ws->todo_free = ws->todo_bd->free;
1316 ws->todo_lim = ws->todo_bd->start + BLOCK_SIZE_W;
1317 stp->blocks = stp->blocks->link;
1319 ws->todo_bd->link = NULL;
1321 // this block is also the scan block; we must scan
1322 // from the current end point.
1323 ws->scan_bd = ws->todo_bd;
1324 ws->scan = ws->scan_bd->free;
1326 // subtract the contents of this block from the stats,
1327 // because we'll count the whole block later.
1328 copied -= ws->scan_bd->free - ws->scan_bd->start;
1335 gc_alloc_todo_block(ws);
1340 // Move the private mutable lists from each capability onto the
1341 // main mutable list for the generation.
1342 for (i = 0; i < n_capabilities; i++) {
1343 for (bd = capabilities[i].mut_lists[g];
1344 bd->link != NULL; bd = bd->link) {
1347 bd->link = generations[g].mut_list;
1348 generations[g].mut_list = capabilities[i].mut_lists[g];
1349 capabilities[i].mut_lists[g] = allocBlock();
1353 /* -----------------------------------------------------------------------------
1354 Initialise a gc_thread before GC
1355 -------------------------------------------------------------------------- */
1358 init_gc_thread (gc_thread *t)
1360 t->static_objects = END_OF_STATIC_LIST;
1361 t->scavenged_static_objects = END_OF_STATIC_LIST;
1363 t->failed_to_evac = rtsFalse;
1364 t->eager_promotion = rtsTrue;
1365 t->thunk_selector_depth = 0;
1369 t->scav_global_work = 0;
1370 t->scav_local_work = 0;
1374 /* -----------------------------------------------------------------------------
1375 Function we pass to GetRoots to evacuate roots.
1376 -------------------------------------------------------------------------- */
1379 mark_root(StgClosure **root)
1384 /* -----------------------------------------------------------------------------
1385 Initialising the static object & mutable lists
1386 -------------------------------------------------------------------------- */
1389 zero_static_object_list(StgClosure* first_static)
1393 const StgInfoTable *info;
1395 for (p = first_static; p != END_OF_STATIC_LIST; p = link) {
1397 link = *STATIC_LINK(info, p);
1398 *STATIC_LINK(info,p) = NULL;
1402 /* -----------------------------------------------------------------------------
1404 -------------------------------------------------------------------------- */
1411 for (c = (StgIndStatic *)revertible_caf_list; c != NULL;
1412 c = (StgIndStatic *)c->static_link)
1414 SET_INFO(c, c->saved_info);
1415 c->saved_info = NULL;
1416 // could, but not necessary: c->static_link = NULL;
1418 revertible_caf_list = NULL;
1422 markCAFs( evac_fn evac )
1426 for (c = (StgIndStatic *)caf_list; c != NULL;
1427 c = (StgIndStatic *)c->static_link)
1429 evac(&c->indirectee);
1431 for (c = (StgIndStatic *)revertible_caf_list; c != NULL;
1432 c = (StgIndStatic *)c->static_link)
1434 evac(&c->indirectee);
1438 /* ----------------------------------------------------------------------------
1439 Update the pointers from the task list
1441 These are treated as weak pointers because we want to allow a main
1442 thread to get a BlockedOnDeadMVar exception in the same way as any
1443 other thread. Note that the threads should all have been retained
1444 by GC by virtue of being on the all_threads list, we're just
1445 updating pointers here.
1446 ------------------------------------------------------------------------- */
1449 update_task_list (void)
1453 for (task = all_tasks; task != NULL; task = task->all_link) {
1454 if (!task->stopped && task->tso) {
1455 ASSERT(task->tso->bound == task);
1456 tso = (StgTSO *) isAlive((StgClosure *)task->tso);
1458 barf("task %p: main thread %d has been GC'd",
1471 /* ----------------------------------------------------------------------------
1472 Reset the sizes of the older generations when we do a major
1475 CURRENT STRATEGY: make all generations except zero the same size.
1476 We have to stay within the maximum heap size, and leave a certain
1477 percentage of the maximum heap size available to allocate into.
1478 ------------------------------------------------------------------------- */
1481 resize_generations (void)
1485 if (major_gc && RtsFlags.GcFlags.generations > 1) {
1486 nat live, size, min_alloc;
1487 nat max = RtsFlags.GcFlags.maxHeapSize;
1488 nat gens = RtsFlags.GcFlags.generations;
1490 // live in the oldest generations
1491 live = oldest_gen->steps[0].n_blocks +
1492 oldest_gen->steps[0].n_large_blocks;
1494 // default max size for all generations except zero
1495 size = stg_max(live * RtsFlags.GcFlags.oldGenFactor,
1496 RtsFlags.GcFlags.minOldGenSize);
1498 // minimum size for generation zero
1499 min_alloc = stg_max((RtsFlags.GcFlags.pcFreeHeap * max) / 200,
1500 RtsFlags.GcFlags.minAllocAreaSize);
1502 // Auto-enable compaction when the residency reaches a
1503 // certain percentage of the maximum heap size (default: 30%).
1504 if (RtsFlags.GcFlags.generations > 1 &&
1505 (RtsFlags.GcFlags.compact ||
1507 oldest_gen->steps[0].n_blocks >
1508 (RtsFlags.GcFlags.compactThreshold * max) / 100))) {
1509 oldest_gen->steps[0].is_compacted = 1;
1510 // debugBelch("compaction: on\n", live);
1512 oldest_gen->steps[0].is_compacted = 0;
1513 // debugBelch("compaction: off\n", live);
1516 // if we're going to go over the maximum heap size, reduce the
1517 // size of the generations accordingly. The calculation is
1518 // different if compaction is turned on, because we don't need
1519 // to double the space required to collect the old generation.
1522 // this test is necessary to ensure that the calculations
1523 // below don't have any negative results - we're working
1524 // with unsigned values here.
1525 if (max < min_alloc) {
1529 if (oldest_gen->steps[0].is_compacted) {
1530 if ( (size + (size - 1) * (gens - 2) * 2) + min_alloc > max ) {
1531 size = (max - min_alloc) / ((gens - 1) * 2 - 1);
1534 if ( (size * (gens - 1) * 2) + min_alloc > max ) {
1535 size = (max - min_alloc) / ((gens - 1) * 2);
1545 debugBelch("live: %d, min_alloc: %d, size : %d, max = %d\n", live,
1546 min_alloc, size, max);
1549 for (g = 0; g < gens; g++) {
1550 generations[g].max_blocks = size;
1555 /* -----------------------------------------------------------------------------
1556 Calculate the new size of the nursery, and resize it.
1557 -------------------------------------------------------------------------- */
1560 resize_nursery (void)
1562 if (RtsFlags.GcFlags.generations == 1)
1563 { // Two-space collector:
1566 /* set up a new nursery. Allocate a nursery size based on a
1567 * function of the amount of live data (by default a factor of 2)
1568 * Use the blocks from the old nursery if possible, freeing up any
1571 * If we get near the maximum heap size, then adjust our nursery
1572 * size accordingly. If the nursery is the same size as the live
1573 * data (L), then we need 3L bytes. We can reduce the size of the
1574 * nursery to bring the required memory down near 2L bytes.
1576 * A normal 2-space collector would need 4L bytes to give the same
1577 * performance we get from 3L bytes, reducing to the same
1578 * performance at 2L bytes.
1580 blocks = g0s0->n_old_blocks;
1582 if ( RtsFlags.GcFlags.maxHeapSize != 0 &&
1583 blocks * RtsFlags.GcFlags.oldGenFactor * 2 >
1584 RtsFlags.GcFlags.maxHeapSize )
1586 long adjusted_blocks; // signed on purpose
1589 adjusted_blocks = (RtsFlags.GcFlags.maxHeapSize - 2 * blocks);
1591 debugTrace(DEBUG_gc, "near maximum heap size of 0x%x blocks, blocks = %d, adjusted to %ld",
1592 RtsFlags.GcFlags.maxHeapSize, blocks, adjusted_blocks);
1594 pc_free = adjusted_blocks * 100 / RtsFlags.GcFlags.maxHeapSize;
1595 if (pc_free < RtsFlags.GcFlags.pcFreeHeap) /* might even * be < 0 */
1599 blocks = adjusted_blocks;
1603 blocks *= RtsFlags.GcFlags.oldGenFactor;
1604 if (blocks < RtsFlags.GcFlags.minAllocAreaSize)
1606 blocks = RtsFlags.GcFlags.minAllocAreaSize;
1609 resizeNurseries(blocks);
1611 else // Generational collector
1614 * If the user has given us a suggested heap size, adjust our
1615 * allocation area to make best use of the memory available.
1617 if (RtsFlags.GcFlags.heapSizeSuggestion)
1620 nat needed = calcNeeded(); // approx blocks needed at next GC
1622 /* Guess how much will be live in generation 0 step 0 next time.
1623 * A good approximation is obtained by finding the
1624 * percentage of g0s0 that was live at the last minor GC.
1626 * We have an accurate figure for the amount of copied data in
1627 * 'copied', but we must convert this to a number of blocks, with
1628 * a small adjustment for estimated slop at the end of a block
1633 g0s0_pcnt_kept = ((copied / (BLOCK_SIZE_W - 10)) * 100)
1634 / countNurseryBlocks();
1637 /* Estimate a size for the allocation area based on the
1638 * information available. We might end up going slightly under
1639 * or over the suggested heap size, but we should be pretty
1642 * Formula: suggested - needed
1643 * ----------------------------
1644 * 1 + g0s0_pcnt_kept/100
1646 * where 'needed' is the amount of memory needed at the next
1647 * collection for collecting all steps except g0s0.
1650 (((long)RtsFlags.GcFlags.heapSizeSuggestion - (long)needed) * 100) /
1651 (100 + (long)g0s0_pcnt_kept);
1653 if (blocks < (long)RtsFlags.GcFlags.minAllocAreaSize) {
1654 blocks = RtsFlags.GcFlags.minAllocAreaSize;
1657 resizeNurseries((nat)blocks);
1661 // we might have added extra large blocks to the nursery, so
1662 // resize back to minAllocAreaSize again.
1663 resizeNurseriesFixed(RtsFlags.GcFlags.minAllocAreaSize);
1668 /* -----------------------------------------------------------------------------
1669 Sanity code for CAF garbage collection.
1671 With DEBUG turned on, we manage a CAF list in addition to the SRT
1672 mechanism. After GC, we run down the CAF list and blackhole any
1673 CAFs which have been garbage collected. This means we get an error
1674 whenever the program tries to enter a garbage collected CAF.
1676 Any garbage collected CAFs are taken off the CAF list at the same
1678 -------------------------------------------------------------------------- */
1680 #if 0 && defined(DEBUG)
1687 const StgInfoTable *info;
1698 ASSERT(info->type == IND_STATIC);
1700 if (STATIC_LINK(info,p) == NULL) {
1701 debugTrace(DEBUG_gccafs, "CAF gc'd at 0x%04lx", (long)p);
1703 SET_INFO(p,&stg_BLACKHOLE_info);
1704 p = STATIC_LINK2(info,p);
1708 pp = &STATIC_LINK2(info,p);
1715 debugTrace(DEBUG_gccafs, "%d CAFs live", i);