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 nat 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);
150 static void shutdown_gc_threads (nat n_threads);
152 #if 0 && defined(DEBUG)
153 static void gcCAFs (void);
156 /* -----------------------------------------------------------------------------
157 The mark bitmap & stack.
158 -------------------------------------------------------------------------- */
160 #define MARK_STACK_BLOCKS 4
162 bdescr *mark_stack_bdescr;
167 // Flag and pointers used for falling back to a linear scan when the
168 // mark stack overflows.
169 rtsBool mark_stack_overflowed;
170 bdescr *oldgen_scan_bd;
173 /* -----------------------------------------------------------------------------
174 GarbageCollect: the main entry point to the garbage collector.
176 Locks held: all capabilities are held throughout GarbageCollect().
177 -------------------------------------------------------------------------- */
180 GarbageCollect ( rtsBool force_major_gc )
184 lnat live, allocated;
185 lnat oldgen_saved_blocks = 0;
186 gc_thread *saved_gct;
189 // necessary if we stole a callee-saves register for gct:
193 CostCentreStack *prev_CCS;
198 debugTrace(DEBUG_gc, "starting GC");
200 #if defined(RTS_USER_SIGNALS)
201 if (RtsFlags.MiscFlags.install_signal_handlers) {
207 // tell the stats department that we've started a GC
210 // tell the STM to discard any cached closures it's hoping to re-use
219 // attribute any costs to CCS_GC
225 /* Approximate how much we allocated.
226 * Todo: only when generating stats?
228 allocated = calcAllocated();
230 /* Figure out which generation to collect
232 n = initialise_N(force_major_gc);
234 /* Allocate + initialise the gc_thread structures.
238 /* Start threads, so they can be spinning up while we finish initialisation.
242 /* How many threads will be participating in this GC?
243 * We don't try to parallelise minor GC.
245 #if defined(THREADED_RTS)
246 if (n < (4*1024*1024 / BLOCK_SIZE)) {
249 n_gc_threads = RtsFlags.ParFlags.gcThreads;
251 trace(TRACE_gc|DEBUG_gc, "GC (gen %d): %dKB to collect, using %d thread(s)",
252 N, n * (BLOCK_SIZE / 1024), n_gc_threads);
257 #ifdef RTS_GTK_FRONTPANEL
258 if (RtsFlags.GcFlags.frontpanel) {
259 updateFrontPanelBeforeGC(N);
264 // check for memory leaks if DEBUG is on
265 memInventory(traceClass(DEBUG_gc));
268 // check stack sanity *before* GC (ToDo: check all threads)
269 IF_DEBUG(sanity, checkFreeListSanity());
271 // Initialise all our gc_thread structures
272 for (t = 0; t < n_gc_threads; t++) {
273 init_gc_thread(gc_threads[t]);
276 // Initialise all the generations/steps that we're collecting.
277 for (g = 0; g <= N; g++) {
278 init_collected_gen(g,n_gc_threads);
281 // Initialise all the generations/steps that we're *not* collecting.
282 for (g = N+1; g < RtsFlags.GcFlags.generations; g++) {
283 init_uncollected_gen(g,n_gc_threads);
286 /* Allocate a mark stack if we're doing a major collection.
289 mark_stack_bdescr = allocGroup(MARK_STACK_BLOCKS);
290 mark_stack = (StgPtr *)mark_stack_bdescr->start;
291 mark_sp = mark_stack;
292 mark_splim = mark_stack + (MARK_STACK_BLOCKS * BLOCK_SIZE_W);
294 mark_stack_bdescr = NULL;
297 // this is the main thread
300 /* -----------------------------------------------------------------------
301 * follow all the roots that we know about:
302 * - mutable lists from each generation > N
303 * we want to *scavenge* these roots, not evacuate them: they're not
304 * going to move in this GC.
305 * Also do them in reverse generation order, for the usual reason:
306 * namely to reduce the likelihood of spurious old->new pointers.
308 for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
309 generations[g].saved_mut_list = generations[g].mut_list;
310 generations[g].mut_list = allocBlock();
311 // mut_list always has at least one block.
314 // the main thread is running: this prevents any other threads from
315 // exiting prematurely, so we can start them now.
316 // NB. do this after the mutable lists have been saved above, otherwise
317 // the other GC threads will be writing into the old mutable lists.
319 wakeup_gc_threads(n_gc_threads);
321 for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
322 scavenge_mutable_list(&generations[g]);
325 // follow roots from the CAF list (used by GHCi)
329 // follow all the roots that the application knows about.
333 #if defined(RTS_USER_SIGNALS)
334 // mark the signal handlers (signals should be already blocked)
335 markSignalHandlers(mark_root);
338 // Mark the weak pointer list, and prepare to detect dead weak pointers.
342 // Mark the stable pointer table.
343 markStablePtrTable(mark_root);
345 /* -------------------------------------------------------------------------
346 * Repeatedly scavenge all the areas we know about until there's no
347 * more scavenging to be done.
352 // The other threads are now stopped. We might recurse back to
353 // here, but from now on this is the only thread.
355 // if any blackholes are alive, make the threads that wait on
357 if (traverseBlackholeQueue()) {
362 // must be last... invariant is that everything is fully
363 // scavenged at this point.
364 if (traverseWeakPtrList()) { // returns rtsTrue if evaced something
369 // If we get to here, there's really nothing left to do.
373 shutdown_gc_threads(n_gc_threads);
375 // Update pointers from the Task list
378 // Now see which stable names are still alive.
382 // We call processHeapClosureForDead() on every closure destroyed during
383 // the current garbage collection, so we invoke LdvCensusForDead().
384 if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_LDV
385 || RtsFlags.ProfFlags.bioSelector != NULL)
389 // NO MORE EVACUATION AFTER THIS POINT!
390 // Finally: compaction of the oldest generation.
391 if (major_gc && oldest_gen->steps[0].is_compacted) {
392 // save number of blocks for stats
393 oldgen_saved_blocks = oldest_gen->steps[0].n_old_blocks;
397 IF_DEBUG(sanity, checkGlobalTSOList(rtsFalse));
399 // Two-space collector: free the old to-space.
400 // g0s0->old_blocks is the old nursery
401 // g0s0->blocks is to-space from the previous GC
402 if (RtsFlags.GcFlags.generations == 1) {
403 if (g0s0->blocks != NULL) {
404 freeChain(g0s0->blocks);
409 // For each workspace, in each thread:
410 // * clear the BF_EVACUATED flag from each copied block
411 // * move the copied blocks to the step
417 for (t = 0; t < n_gc_threads; t++) {
421 for (s = 1; s < total_steps; s++) {
424 // ASSERT( ws->scan_bd == ws->todo_bd );
425 ASSERT( ws->scan_bd ? ws->scan_bd->u.scan == ws->scan_bd->free : 1 );
427 // Push the final block
428 if (ws->scan_bd) { push_scanned_block(ws->scan_bd, ws); }
430 ASSERT(countBlocks(ws->scavd_list) == ws->n_scavd_blocks);
432 prev = ws->part_list;
433 for (bd = ws->part_list; bd != NULL; bd = bd->link) {
434 bd->flags &= ~BF_EVACUATED; // now from-space
438 prev->link = ws->scavd_list;
440 for (bd = ws->scavd_list; bd != NULL; bd = bd->link) {
441 bd->flags &= ~BF_EVACUATED; // now from-space
444 prev->link = ws->step->blocks;
445 if (ws->part_list != NULL) {
446 ws->step->blocks = ws->part_list;
448 ws->step->blocks = ws->scavd_list;
450 ws->step->n_blocks += ws->n_part_blocks;
451 ws->step->n_blocks += ws->n_scavd_blocks;
452 ASSERT(countBlocks(ws->step->blocks) == ws->step->n_blocks);
457 // Two-space collector: swap the semi-spaces around.
458 // Currently: g0s0->old_blocks is the old nursery
459 // g0s0->blocks is to-space from this GC
460 // We want these the other way around.
461 if (RtsFlags.GcFlags.generations == 1) {
462 bdescr *nursery_blocks = g0s0->old_blocks;
463 nat n_nursery_blocks = g0s0->n_old_blocks;
464 g0s0->old_blocks = g0s0->blocks;
465 g0s0->n_old_blocks = g0s0->n_blocks;
466 g0s0->blocks = nursery_blocks;
467 g0s0->n_blocks = n_nursery_blocks;
470 /* run through all the generations/steps and tidy up
475 for (i=0; i < n_gc_threads; i++) {
476 if (n_gc_threads > 1) {
477 trace(TRACE_gc,"thread %d:", i);
478 trace(TRACE_gc," copied %ld", gc_threads[i]->copied * sizeof(W_));
479 trace(TRACE_gc," any_work %ld", gc_threads[i]->any_work);
480 trace(TRACE_gc," no_work %ld", gc_threads[i]->no_work);
481 trace(TRACE_gc," scav_find_work %ld", gc_threads[i]->scav_find_work);
483 copied += gc_threads[i]->copied;
487 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
490 generations[g].collections++; // for stats
493 // Count the mutable list as bytes "copied" for the purposes of
494 // stats. Every mutable list is copied during every GC.
496 nat mut_list_size = 0;
497 for (bd = generations[g].mut_list; bd != NULL; bd = bd->link) {
498 mut_list_size += bd->free - bd->start;
500 copied += mut_list_size;
503 "mut_list_size: %lu (%d vars, %d arrays, %d MVARs, %d others)",
504 (unsigned long)(mut_list_size * sizeof(W_)),
505 mutlist_MUTVARS, mutlist_MUTARRS, mutlist_MVARS, mutlist_OTHERS);
508 for (s = 0; s < generations[g].n_steps; s++) {
510 stp = &generations[g].steps[s];
512 // for generations we collected...
515 /* free old memory and shift to-space into from-space for all
516 * the collected steps (except the allocation area). These
517 * freed blocks will probaby be quickly recycled.
519 if (!(g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1)) {
520 if (stp->is_compacted)
522 // for a compacted step, just shift the new to-space
523 // onto the front of the now-compacted existing blocks.
524 for (bd = stp->blocks; bd != NULL; bd = bd->link) {
525 bd->flags &= ~BF_EVACUATED; // now from-space
527 // tack the new blocks on the end of the existing blocks
528 if (stp->old_blocks != NULL) {
529 for (bd = stp->old_blocks; bd != NULL; bd = next) {
530 // NB. this step might not be compacted next
531 // time, so reset the BF_COMPACTED flags.
532 // They are set before GC if we're going to
533 // compact. (search for BF_COMPACTED above).
534 bd->flags &= ~BF_COMPACTED;
537 bd->link = stp->blocks;
540 stp->blocks = stp->old_blocks;
542 // add the new blocks to the block tally
543 stp->n_blocks += stp->n_old_blocks;
544 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
548 freeChain(stp->old_blocks);
550 stp->old_blocks = NULL;
551 stp->n_old_blocks = 0;
554 /* LARGE OBJECTS. The current live large objects are chained on
555 * scavenged_large, having been moved during garbage
556 * collection from large_objects. Any objects left on
557 * large_objects list are therefore dead, so we free them here.
559 for (bd = stp->large_objects; bd != NULL; bd = next) {
565 // update the count of blocks used by large objects
566 for (bd = stp->scavenged_large_objects; bd != NULL; bd = bd->link) {
567 bd->flags &= ~BF_EVACUATED;
569 stp->large_objects = stp->scavenged_large_objects;
570 stp->n_large_blocks = stp->n_scavenged_large_blocks;
573 else // for older generations...
575 /* For older generations, we need to append the
576 * scavenged_large_object list (i.e. large objects that have been
577 * promoted during this GC) to the large_object list for that step.
579 for (bd = stp->scavenged_large_objects; bd; bd = next) {
581 bd->flags &= ~BF_EVACUATED;
582 dbl_link_onto(bd, &stp->large_objects);
585 // add the new blocks we promoted during this GC
586 stp->n_large_blocks += stp->n_scavenged_large_blocks;
591 // update the max size of older generations after a major GC
592 resize_generations();
594 // Guess the amount of live data for stats.
595 live = calcLiveBlocks() * BLOCK_SIZE_W;
596 debugTrace(DEBUG_gc, "Slop: %ldKB",
597 (live - calcLiveWords()) / (1024/sizeof(W_)));
599 // Free the small objects allocated via allocate(), since this will
600 // all have been copied into G0S1 now.
601 if (RtsFlags.GcFlags.generations > 1) {
602 if (g0s0->blocks != NULL) {
603 freeChain(g0s0->blocks);
609 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
611 // Start a new pinned_object_block
612 pinned_object_block = NULL;
614 // Free the mark stack.
615 if (mark_stack_bdescr != NULL) {
616 freeGroup(mark_stack_bdescr);
620 for (g = 0; g <= N; g++) {
621 for (s = 0; s < generations[g].n_steps; s++) {
622 stp = &generations[g].steps[s];
623 if (stp->bitmap != NULL) {
624 freeGroup(stp->bitmap);
632 // mark the garbage collected CAFs as dead
633 #if 0 && defined(DEBUG) // doesn't work at the moment
634 if (major_gc) { gcCAFs(); }
638 // resetStaticObjectForRetainerProfiling() must be called before
640 if (n_gc_threads > 1) {
641 barf("profiling is currently broken with multi-threaded GC");
642 // ToDo: fix the gct->scavenged_static_objects below
644 resetStaticObjectForRetainerProfiling(gct->scavenged_static_objects);
647 // zero the scavenged static object list
650 for (i = 0; i < n_gc_threads; i++) {
651 zero_static_object_list(gc_threads[i]->scavenged_static_objects);
658 // start any pending finalizers
660 scheduleFinalizers(last_free_capability, old_weak_ptr_list);
663 // send exceptions to any threads which were about to die
665 resurrectThreads(resurrected_threads);
668 // Update the stable pointer hash table.
669 updateStablePtrTable(major_gc);
671 // check sanity after GC
672 IF_DEBUG(sanity, checkSanity());
674 // extra GC trace info
675 if (traceClass(TRACE_gc|DEBUG_gc)) statDescribeGens();
678 // symbol-table based profiling
679 /* heapCensus(to_blocks); */ /* ToDo */
682 // restore enclosing cost centre
688 // check for memory leaks if DEBUG is on
689 memInventory(traceClass(DEBUG_gc));
692 #ifdef RTS_GTK_FRONTPANEL
693 if (RtsFlags.GcFlags.frontpanel) {
694 updateFrontPanelAfterGC( N, live );
698 // ok, GC over: tell the stats department what happened.
699 stat_endGC(allocated, live, copied, N);
701 #if defined(RTS_USER_SIGNALS)
702 if (RtsFlags.MiscFlags.install_signal_handlers) {
703 // unblock signals again
704 unblockUserSignals();
713 /* -----------------------------------------------------------------------------
714 * Mark all nodes pointed to by sparks in the spark queues (for GC) Does an
715 * implicit slide i.e. after marking all sparks are at the beginning of the
716 * spark pool and the spark pool only contains sparkable closures
717 * -------------------------------------------------------------------------- */
721 markSparkQueue (evac_fn evac, Capability *cap)
723 StgClosure **sparkp, **to_sparkp;
724 nat n, pruned_sparks; // stats only
727 PAR_TICKY_MARK_SPARK_QUEUE_START();
732 pool = &(cap->r.rSparks);
734 ASSERT_SPARK_POOL_INVARIANTS(pool);
736 #if defined(PARALLEL_HASKELL)
743 to_sparkp = pool->hd;
744 while (sparkp != pool->tl) {
745 ASSERT(*sparkp!=NULL);
746 ASSERT(LOOKS_LIKE_CLOSURE_PTR(((StgClosure *)*sparkp)));
747 // ToDo?: statistics gathering here (also for GUM!)
748 if (closure_SHOULD_SPARK(*sparkp)) {
750 *to_sparkp++ = *sparkp;
751 if (to_sparkp == pool->lim) {
752 to_sparkp = pool->base;
759 if (sparkp == pool->lim) {
763 pool->tl = to_sparkp;
765 PAR_TICKY_MARK_SPARK_QUEUE_END(n);
767 #if defined(PARALLEL_HASKELL)
768 debugTrace(DEBUG_sched,
769 "marked %d sparks and pruned %d sparks on [%x]",
770 n, pruned_sparks, mytid);
772 debugTrace(DEBUG_sched,
773 "marked %d sparks and pruned %d sparks",
777 debugTrace(DEBUG_sched,
778 "new spark queue len=%d; (hd=%p; tl=%p)\n",
779 sparkPoolSize(pool), pool->hd, pool->tl);
783 /* ---------------------------------------------------------------------------
784 Where are the roots that we know about?
786 - all the threads on the runnable queue
787 - all the threads on the blocked queue
788 - all the threads on the sleeping queue
789 - all the thread currently executing a _ccall_GC
790 - all the "main threads"
792 ------------------------------------------------------------------------ */
795 GetRoots( evac_fn evac )
801 // Each GC thread is responsible for following roots from the
802 // Capability of the same number. There will usually be the same
803 // or fewer Capabilities as GC threads, but just in case there
804 // are more, we mark every Capability whose number is the GC
805 // thread's index plus a multiple of the number of GC threads.
806 for (i = gct->thread_index; i < n_capabilities; i += n_gc_threads) {
807 cap = &capabilities[i];
808 evac((StgClosure **)(void *)&cap->run_queue_hd);
809 evac((StgClosure **)(void *)&cap->run_queue_tl);
810 #if defined(THREADED_RTS)
811 evac((StgClosure **)(void *)&cap->wakeup_queue_hd);
812 evac((StgClosure **)(void *)&cap->wakeup_queue_tl);
814 for (task = cap->suspended_ccalling_tasks; task != NULL;
816 debugTrace(DEBUG_sched,
817 "evac'ing suspended TSO %lu", (unsigned long)task->suspended_tso->id);
818 evac((StgClosure **)(void *)&task->suspended_tso);
821 #if defined(THREADED_RTS)
822 markSparkQueue(evac,cap);
826 #if !defined(THREADED_RTS)
827 evac((StgClosure **)(void *)&blocked_queue_hd);
828 evac((StgClosure **)(void *)&blocked_queue_tl);
829 evac((StgClosure **)(void *)&sleeping_queue);
833 /* -----------------------------------------------------------------------------
834 isAlive determines whether the given closure is still alive (after
835 a garbage collection) or not. It returns the new address of the
836 closure if it is alive, or NULL otherwise.
838 NOTE: Use it before compaction only!
839 It untags and (if needed) retags pointers to closures.
840 -------------------------------------------------------------------------- */
844 isAlive(StgClosure *p)
846 const StgInfoTable *info;
852 /* The tag and the pointer are split, to be merged later when needed. */
853 tag = GET_CLOSURE_TAG(p);
854 q = UNTAG_CLOSURE(p);
856 ASSERT(LOOKS_LIKE_CLOSURE_PTR(q));
859 // ignore static closures
861 // ToDo: for static closures, check the static link field.
862 // Problem here is that we sometimes don't set the link field, eg.
863 // for static closures with an empty SRT or CONSTR_STATIC_NOCAFs.
865 if (!HEAP_ALLOCED(q)) {
869 // ignore closures in generations that we're not collecting.
871 if (bd->gen_no > N) {
875 // if it's a pointer into to-space, then we're done
876 if (bd->flags & BF_EVACUATED) {
880 // large objects use the evacuated flag
881 if (bd->flags & BF_LARGE) {
885 // check the mark bit for compacted steps
886 if ((bd->flags & BF_COMPACTED) && is_marked((P_)q,bd)) {
890 switch (info->type) {
895 case IND_OLDGEN: // rely on compatible layout with StgInd
896 case IND_OLDGEN_PERM:
897 // follow indirections
898 p = ((StgInd *)q)->indirectee;
903 return ((StgEvacuated *)q)->evacuee;
906 if (((StgTSO *)q)->what_next == ThreadRelocated) {
907 p = (StgClosure *)((StgTSO *)q)->link;
919 /* -----------------------------------------------------------------------------
920 Figure out which generation to collect, initialise N and major_gc.
922 Also returns the total number of blocks in generations that will be
924 -------------------------------------------------------------------------- */
927 initialise_N (rtsBool force_major_gc)
930 nat s, blocks, blocks_total;
935 if (force_major_gc) {
936 N = RtsFlags.GcFlags.generations - 1;
941 for (g = RtsFlags.GcFlags.generations - 1; g >= 0; g--) {
943 for (s = 0; s < generations[g].n_steps; s++) {
944 blocks += generations[g].steps[s].n_blocks;
945 blocks += generations[g].steps[s].n_large_blocks;
947 if (blocks >= generations[g].max_blocks) {
951 blocks_total += blocks;
955 blocks_total += countNurseryBlocks();
957 major_gc = (N == RtsFlags.GcFlags.generations-1);
961 /* -----------------------------------------------------------------------------
962 Initialise the gc_thread structures.
963 -------------------------------------------------------------------------- */
966 alloc_gc_thread (int n)
972 t = stgMallocBytes(sizeof(gc_thread) + total_steps * sizeof(step_workspace),
977 initCondition(&t->wake_cond);
978 initMutex(&t->wake_mutex);
979 t->wakeup = rtsFalse;
984 t->free_blocks = NULL;
993 for (s = 0; s < total_steps; s++)
996 ws->step = &all_steps[s];
997 ASSERT(s == ws->step->abs_no);
1003 ws->buffer_todo_bd = NULL;
1005 ws->part_list = NULL;
1006 ws->n_part_blocks = 0;
1008 ws->scavd_list = NULL;
1009 ws->n_scavd_blocks = 0;
1017 alloc_gc_threads (void)
1019 if (gc_threads == NULL) {
1020 #if defined(THREADED_RTS)
1022 gc_threads = stgMallocBytes (RtsFlags.ParFlags.gcThreads *
1024 "alloc_gc_threads");
1026 for (i = 0; i < RtsFlags.ParFlags.gcThreads; i++) {
1027 gc_threads[i] = alloc_gc_thread(i);
1030 gc_threads = stgMallocBytes (sizeof(gc_thread*),
1031 "alloc_gc_threads");
1033 gc_threads[0] = alloc_gc_thread(0);
1038 /* ----------------------------------------------------------------------------
1040 ------------------------------------------------------------------------- */
1042 static nat gc_running_threads;
1044 #if defined(THREADED_RTS)
1045 static Mutex gc_running_mutex;
1052 ACQUIRE_LOCK(&gc_running_mutex);
1053 n_running = ++gc_running_threads;
1054 RELEASE_LOCK(&gc_running_mutex);
1055 ASSERT(n_running <= n_gc_threads);
1063 ACQUIRE_LOCK(&gc_running_mutex);
1064 ASSERT(n_gc_threads != 0);
1065 n_running = --gc_running_threads;
1066 RELEASE_LOCK(&gc_running_mutex);
1071 // gc_thread_work(): Scavenge until there's no work left to do and all
1072 // the running threads are idle.
1075 gc_thread_work (void)
1079 debugTrace(DEBUG_gc, "GC thread %d working", gct->thread_index);
1081 // gc_running_threads has already been incremented for us; either
1082 // this is the main thread and we incremented it inside
1083 // GarbageCollect(), or this is a worker thread and the main
1084 // thread bumped gc_running_threads before waking us up.
1086 // Every thread evacuates some roots.
1088 GetRoots(mark_root);
1092 // scavenge_loop() only exits when there's no work to do
1095 debugTrace(DEBUG_gc, "GC thread %d idle (%d still running)",
1096 gct->thread_index, r);
1098 while (gc_running_threads != 0) {
1104 // any_work() does not remove the work from the queue, it
1105 // just checks for the presence of work. If we find any,
1106 // then we increment gc_running_threads and go back to
1107 // scavenge_loop() to perform any pending work.
1110 // All threads are now stopped
1111 debugTrace(DEBUG_gc, "GC thread %d finished.", gct->thread_index);
1115 #if defined(THREADED_RTS)
1117 gc_thread_mainloop (void)
1119 while (!gct->exit) {
1121 // Wait until we're told to wake up
1122 ACQUIRE_LOCK(&gct->wake_mutex);
1123 gct->wakeup = rtsFalse;
1124 while (!gct->wakeup) {
1125 debugTrace(DEBUG_gc, "GC thread %d standing by...",
1127 waitCondition(&gct->wake_cond, &gct->wake_mutex);
1129 RELEASE_LOCK(&gct->wake_mutex);
1130 if (gct->exit) break;
1133 // start performance counters in this thread...
1134 if (gct->papi_events == -1) {
1135 papi_init_eventset(&gct->papi_events);
1137 papi_thread_start_gc1_count(gct->papi_events);
1143 // count events in this thread towards the GC totals
1144 papi_thread_stop_gc1_count(gct->papi_events);
1150 #if defined(THREADED_RTS)
1152 gc_thread_entry (gc_thread *my_gct)
1155 debugTrace(DEBUG_gc, "GC thread %d starting...", gct->thread_index);
1156 gct->id = osThreadId();
1157 gc_thread_mainloop();
1162 start_gc_threads (void)
1164 #if defined(THREADED_RTS)
1167 static rtsBool done = rtsFalse;
1169 gc_running_threads = 0;
1170 initMutex(&gc_running_mutex);
1173 // Start from 1: the main thread is 0
1174 for (i = 1; i < RtsFlags.ParFlags.gcThreads; i++) {
1175 createOSThread(&id, (OSThreadProc*)&gc_thread_entry,
1184 wakeup_gc_threads (nat n_threads USED_IF_THREADS)
1186 #if defined(THREADED_RTS)
1188 for (i=1; i < n_threads; i++) {
1190 ACQUIRE_LOCK(&gc_threads[i]->wake_mutex);
1191 gc_threads[i]->wakeup = rtsTrue;
1192 signalCondition(&gc_threads[i]->wake_cond);
1193 RELEASE_LOCK(&gc_threads[i]->wake_mutex);
1198 // After GC is complete, we must wait for all GC threads to enter the
1199 // standby state, otherwise they may still be executing inside
1200 // any_work(), and may even remain awake until the next GC starts.
1202 shutdown_gc_threads (nat n_threads USED_IF_THREADS)
1204 #if defined(THREADED_RTS)
1207 for (i=1; i < n_threads; i++) {
1209 ACQUIRE_LOCK(&gc_threads[i]->wake_mutex);
1210 wakeup = gc_threads[i]->wakeup;
1211 // wakeup is false while the thread is waiting
1212 RELEASE_LOCK(&gc_threads[i]->wake_mutex);
1218 /* ----------------------------------------------------------------------------
1219 Initialise a generation that is to be collected
1220 ------------------------------------------------------------------------- */
1223 init_collected_gen (nat g, nat n_threads)
1230 // Throw away the current mutable list. Invariant: the mutable
1231 // list always has at least one block; this means we can avoid a
1232 // check for NULL in recordMutable().
1234 freeChain(generations[g].mut_list);
1235 generations[g].mut_list = allocBlock();
1236 for (i = 0; i < n_capabilities; i++) {
1237 freeChain(capabilities[i].mut_lists[g]);
1238 capabilities[i].mut_lists[g] = allocBlock();
1242 for (s = 0; s < generations[g].n_steps; s++) {
1244 // generation 0, step 0 doesn't need to-space
1245 if (g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1) {
1249 stp = &generations[g].steps[s];
1250 ASSERT(stp->gen_no == g);
1252 // deprecate the existing blocks
1253 stp->old_blocks = stp->blocks;
1254 stp->n_old_blocks = stp->n_blocks;
1258 // we don't have any to-be-scavenged blocks yet
1260 stp->todos_last = NULL;
1263 // initialise the large object queues.
1264 stp->scavenged_large_objects = NULL;
1265 stp->n_scavenged_large_blocks = 0;
1267 // mark the large objects as not evacuated yet
1268 for (bd = stp->large_objects; bd; bd = bd->link) {
1269 bd->flags &= ~BF_EVACUATED;
1272 // for a compacted step, we need to allocate the bitmap
1273 if (stp->is_compacted) {
1274 nat bitmap_size; // in bytes
1275 bdescr *bitmap_bdescr;
1278 bitmap_size = stp->n_old_blocks * BLOCK_SIZE / (sizeof(W_)*BITS_PER_BYTE);
1280 if (bitmap_size > 0) {
1281 bitmap_bdescr = allocGroup((lnat)BLOCK_ROUND_UP(bitmap_size)
1283 stp->bitmap = bitmap_bdescr;
1284 bitmap = bitmap_bdescr->start;
1286 debugTrace(DEBUG_gc, "bitmap_size: %d, bitmap: %p",
1287 bitmap_size, bitmap);
1289 // don't forget to fill it with zeros!
1290 memset(bitmap, 0, bitmap_size);
1292 // For each block in this step, point to its bitmap from the
1293 // block descriptor.
1294 for (bd=stp->old_blocks; bd != NULL; bd = bd->link) {
1295 bd->u.bitmap = bitmap;
1296 bitmap += BLOCK_SIZE_W / (sizeof(W_)*BITS_PER_BYTE);
1298 // Also at this point we set the BF_COMPACTED flag
1299 // for this block. The invariant is that
1300 // BF_COMPACTED is always unset, except during GC
1301 // when it is set on those blocks which will be
1303 bd->flags |= BF_COMPACTED;
1309 // For each GC thread, for each step, allocate a "todo" block to
1310 // store evacuated objects to be scavenged, and a block to store
1311 // evacuated objects that do not need to be scavenged.
1312 for (t = 0; t < n_threads; t++) {
1313 for (s = 0; s < generations[g].n_steps; s++) {
1315 // we don't copy objects into g0s0, unless -G0
1316 if (g==0 && s==0 && RtsFlags.GcFlags.generations > 1) continue;
1318 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1322 ws->todo_large_objects = NULL;
1324 ws->part_list = NULL;
1325 ws->n_part_blocks = 0;
1327 // allocate the first to-space block; extra blocks will be
1328 // chained on as necessary.
1330 ws->buffer_todo_bd = NULL;
1331 alloc_todo_block(ws,0);
1333 ws->scavd_list = NULL;
1334 ws->n_scavd_blocks = 0;
1340 /* ----------------------------------------------------------------------------
1341 Initialise a generation that is *not* to be collected
1342 ------------------------------------------------------------------------- */
1345 init_uncollected_gen (nat g, nat threads)
1352 for (s = 0; s < generations[g].n_steps; s++) {
1353 stp = &generations[g].steps[s];
1354 stp->scavenged_large_objects = NULL;
1355 stp->n_scavenged_large_blocks = 0;
1358 for (t = 0; t < threads; t++) {
1359 for (s = 0; s < generations[g].n_steps; s++) {
1361 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1364 ws->buffer_todo_bd = NULL;
1365 ws->todo_large_objects = NULL;
1367 ws->part_list = NULL;
1368 ws->n_part_blocks = 0;
1370 ws->scavd_list = NULL;
1371 ws->n_scavd_blocks = 0;
1373 // If the block at the head of the list in this generation
1374 // is less than 3/4 full, then use it as a todo block.
1375 if (stp->blocks && isPartiallyFull(stp->blocks))
1377 ws->todo_bd = stp->blocks;
1378 ws->todo_free = ws->todo_bd->free;
1379 ws->todo_lim = ws->todo_bd->start + BLOCK_SIZE_W;
1380 stp->blocks = stp->blocks->link;
1382 ws->todo_bd->link = NULL;
1384 // this block is also the scan block; we must scan
1385 // from the current end point.
1386 ws->scan_bd = ws->todo_bd;
1387 ws->scan_bd->u.scan = ws->scan_bd->free;
1389 // subtract the contents of this block from the stats,
1390 // because we'll count the whole block later.
1391 copied -= ws->scan_bd->free - ws->scan_bd->start;
1397 alloc_todo_block(ws,0);
1402 // Move the private mutable lists from each capability onto the
1403 // main mutable list for the generation.
1404 for (i = 0; i < n_capabilities; i++) {
1405 for (bd = capabilities[i].mut_lists[g];
1406 bd->link != NULL; bd = bd->link) {
1409 bd->link = generations[g].mut_list;
1410 generations[g].mut_list = capabilities[i].mut_lists[g];
1411 capabilities[i].mut_lists[g] = allocBlock();
1415 /* -----------------------------------------------------------------------------
1416 Initialise a gc_thread before GC
1417 -------------------------------------------------------------------------- */
1420 init_gc_thread (gc_thread *t)
1422 t->static_objects = END_OF_STATIC_LIST;
1423 t->scavenged_static_objects = END_OF_STATIC_LIST;
1425 t->failed_to_evac = rtsFalse;
1426 t->eager_promotion = rtsTrue;
1427 t->thunk_selector_depth = 0;
1431 t->scav_find_work = 0;
1435 /* -----------------------------------------------------------------------------
1436 Function we pass to GetRoots to evacuate roots.
1437 -------------------------------------------------------------------------- */
1440 mark_root(StgClosure **root)
1445 /* -----------------------------------------------------------------------------
1446 Initialising the static object & mutable lists
1447 -------------------------------------------------------------------------- */
1450 zero_static_object_list(StgClosure* first_static)
1454 const StgInfoTable *info;
1456 for (p = first_static; p != END_OF_STATIC_LIST; p = link) {
1458 link = *STATIC_LINK(info, p);
1459 *STATIC_LINK(info,p) = NULL;
1463 /* -----------------------------------------------------------------------------
1465 -------------------------------------------------------------------------- */
1472 for (c = (StgIndStatic *)revertible_caf_list; c != NULL;
1473 c = (StgIndStatic *)c->static_link)
1475 SET_INFO(c, c->saved_info);
1476 c->saved_info = NULL;
1477 // could, but not necessary: c->static_link = NULL;
1479 revertible_caf_list = NULL;
1483 markCAFs( evac_fn evac )
1487 for (c = (StgIndStatic *)caf_list; c != NULL;
1488 c = (StgIndStatic *)c->static_link)
1490 evac(&c->indirectee);
1492 for (c = (StgIndStatic *)revertible_caf_list; c != NULL;
1493 c = (StgIndStatic *)c->static_link)
1495 evac(&c->indirectee);
1499 /* ----------------------------------------------------------------------------
1500 Update the pointers from the task list
1502 These are treated as weak pointers because we want to allow a main
1503 thread to get a BlockedOnDeadMVar exception in the same way as any
1504 other thread. Note that the threads should all have been retained
1505 by GC by virtue of being on the all_threads list, we're just
1506 updating pointers here.
1507 ------------------------------------------------------------------------- */
1510 update_task_list (void)
1514 for (task = all_tasks; task != NULL; task = task->all_link) {
1515 if (!task->stopped && task->tso) {
1516 ASSERT(task->tso->bound == task);
1517 tso = (StgTSO *) isAlive((StgClosure *)task->tso);
1519 barf("task %p: main thread %d has been GC'd",
1532 /* ----------------------------------------------------------------------------
1533 Reset the sizes of the older generations when we do a major
1536 CURRENT STRATEGY: make all generations except zero the same size.
1537 We have to stay within the maximum heap size, and leave a certain
1538 percentage of the maximum heap size available to allocate into.
1539 ------------------------------------------------------------------------- */
1542 resize_generations (void)
1546 if (major_gc && RtsFlags.GcFlags.generations > 1) {
1547 nat live, size, min_alloc;
1548 nat max = RtsFlags.GcFlags.maxHeapSize;
1549 nat gens = RtsFlags.GcFlags.generations;
1551 // live in the oldest generations
1552 live = oldest_gen->steps[0].n_blocks +
1553 oldest_gen->steps[0].n_large_blocks;
1555 // default max size for all generations except zero
1556 size = stg_max(live * RtsFlags.GcFlags.oldGenFactor,
1557 RtsFlags.GcFlags.minOldGenSize);
1559 // minimum size for generation zero
1560 min_alloc = stg_max((RtsFlags.GcFlags.pcFreeHeap * max) / 200,
1561 RtsFlags.GcFlags.minAllocAreaSize);
1563 // Auto-enable compaction when the residency reaches a
1564 // certain percentage of the maximum heap size (default: 30%).
1565 if (RtsFlags.GcFlags.generations > 1 &&
1566 (RtsFlags.GcFlags.compact ||
1568 oldest_gen->steps[0].n_blocks >
1569 (RtsFlags.GcFlags.compactThreshold * max) / 100))) {
1570 oldest_gen->steps[0].is_compacted = 1;
1571 // debugBelch("compaction: on\n", live);
1573 oldest_gen->steps[0].is_compacted = 0;
1574 // debugBelch("compaction: off\n", live);
1577 // if we're going to go over the maximum heap size, reduce the
1578 // size of the generations accordingly. The calculation is
1579 // different if compaction is turned on, because we don't need
1580 // to double the space required to collect the old generation.
1583 // this test is necessary to ensure that the calculations
1584 // below don't have any negative results - we're working
1585 // with unsigned values here.
1586 if (max < min_alloc) {
1590 if (oldest_gen->steps[0].is_compacted) {
1591 if ( (size + (size - 1) * (gens - 2) * 2) + min_alloc > max ) {
1592 size = (max - min_alloc) / ((gens - 1) * 2 - 1);
1595 if ( (size * (gens - 1) * 2) + min_alloc > max ) {
1596 size = (max - min_alloc) / ((gens - 1) * 2);
1606 debugBelch("live: %d, min_alloc: %d, size : %d, max = %d\n", live,
1607 min_alloc, size, max);
1610 for (g = 0; g < gens; g++) {
1611 generations[g].max_blocks = size;
1616 /* -----------------------------------------------------------------------------
1617 Calculate the new size of the nursery, and resize it.
1618 -------------------------------------------------------------------------- */
1621 resize_nursery (void)
1623 if (RtsFlags.GcFlags.generations == 1)
1624 { // Two-space collector:
1627 /* set up a new nursery. Allocate a nursery size based on a
1628 * function of the amount of live data (by default a factor of 2)
1629 * Use the blocks from the old nursery if possible, freeing up any
1632 * If we get near the maximum heap size, then adjust our nursery
1633 * size accordingly. If the nursery is the same size as the live
1634 * data (L), then we need 3L bytes. We can reduce the size of the
1635 * nursery to bring the required memory down near 2L bytes.
1637 * A normal 2-space collector would need 4L bytes to give the same
1638 * performance we get from 3L bytes, reducing to the same
1639 * performance at 2L bytes.
1641 blocks = g0s0->n_old_blocks;
1643 if ( RtsFlags.GcFlags.maxHeapSize != 0 &&
1644 blocks * RtsFlags.GcFlags.oldGenFactor * 2 >
1645 RtsFlags.GcFlags.maxHeapSize )
1647 long adjusted_blocks; // signed on purpose
1650 adjusted_blocks = (RtsFlags.GcFlags.maxHeapSize - 2 * blocks);
1652 debugTrace(DEBUG_gc, "near maximum heap size of 0x%x blocks, blocks = %d, adjusted to %ld",
1653 RtsFlags.GcFlags.maxHeapSize, blocks, adjusted_blocks);
1655 pc_free = adjusted_blocks * 100 / RtsFlags.GcFlags.maxHeapSize;
1656 if (pc_free < RtsFlags.GcFlags.pcFreeHeap) /* might even * be < 0 */
1660 blocks = adjusted_blocks;
1664 blocks *= RtsFlags.GcFlags.oldGenFactor;
1665 if (blocks < RtsFlags.GcFlags.minAllocAreaSize)
1667 blocks = RtsFlags.GcFlags.minAllocAreaSize;
1670 resizeNurseries(blocks);
1672 else // Generational collector
1675 * If the user has given us a suggested heap size, adjust our
1676 * allocation area to make best use of the memory available.
1678 if (RtsFlags.GcFlags.heapSizeSuggestion)
1681 nat needed = calcNeeded(); // approx blocks needed at next GC
1683 /* Guess how much will be live in generation 0 step 0 next time.
1684 * A good approximation is obtained by finding the
1685 * percentage of g0s0 that was live at the last minor GC.
1687 * We have an accurate figure for the amount of copied data in
1688 * 'copied', but we must convert this to a number of blocks, with
1689 * a small adjustment for estimated slop at the end of a block
1694 g0s0_pcnt_kept = ((copied / (BLOCK_SIZE_W - 10)) * 100)
1695 / countNurseryBlocks();
1698 /* Estimate a size for the allocation area based on the
1699 * information available. We might end up going slightly under
1700 * or over the suggested heap size, but we should be pretty
1703 * Formula: suggested - needed
1704 * ----------------------------
1705 * 1 + g0s0_pcnt_kept/100
1707 * where 'needed' is the amount of memory needed at the next
1708 * collection for collecting all steps except g0s0.
1711 (((long)RtsFlags.GcFlags.heapSizeSuggestion - (long)needed) * 100) /
1712 (100 + (long)g0s0_pcnt_kept);
1714 if (blocks < (long)RtsFlags.GcFlags.minAllocAreaSize) {
1715 blocks = RtsFlags.GcFlags.minAllocAreaSize;
1718 resizeNurseries((nat)blocks);
1722 // we might have added extra large blocks to the nursery, so
1723 // resize back to minAllocAreaSize again.
1724 resizeNurseriesFixed(RtsFlags.GcFlags.minAllocAreaSize);
1729 /* -----------------------------------------------------------------------------
1730 Sanity code for CAF garbage collection.
1732 With DEBUG turned on, we manage a CAF list in addition to the SRT
1733 mechanism. After GC, we run down the CAF list and blackhole any
1734 CAFs which have been garbage collected. This means we get an error
1735 whenever the program tries to enter a garbage collected CAF.
1737 Any garbage collected CAFs are taken off the CAF list at the same
1739 -------------------------------------------------------------------------- */
1741 #if 0 && defined(DEBUG)
1748 const StgInfoTable *info;
1759 ASSERT(info->type == IND_STATIC);
1761 if (STATIC_LINK(info,p) == NULL) {
1762 debugTrace(DEBUG_gccafs, "CAF gc'd at 0x%04lx", (long)p);
1764 SET_INFO(p,&stg_BLACKHOLE_info);
1765 p = STATIC_LINK2(info,p);
1769 pp = &STATIC_LINK2(info,p);
1776 debugTrace(DEBUG_gccafs, "%d CAFs live", i);