1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team 1998-2008
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"
25 #include "BlockAlloc.h"
29 #include "RtsSignals.h"
31 #if defined(RTS_GTK_FRONTPANEL)
32 #include "FrontPanel.h"
35 #include "RetainerProfile.h"
36 #include "LdvProfile.h"
37 #include "RaiseAsync.h"
47 #include "MarkStack.h"
52 #include <string.h> // for memset()
55 /* -----------------------------------------------------------------------------
57 -------------------------------------------------------------------------- */
59 /* STATIC OBJECT LIST.
62 * We maintain a linked list of static objects that are still live.
63 * The requirements for this list are:
65 * - we need to scan the list while adding to it, in order to
66 * scavenge all the static objects (in the same way that
67 * breadth-first scavenging works for dynamic objects).
69 * - we need to be able to tell whether an object is already on
70 * the list, to break loops.
72 * Each static object has a "static link field", which we use for
73 * linking objects on to the list. We use a stack-type list, consing
74 * objects on the front as they are added (this means that the
75 * scavenge phase is depth-first, not breadth-first, but that
78 * A separate list is kept for objects that have been scavenged
79 * already - this is so that we can zero all the marks afterwards.
81 * An object is on the list if its static link field is non-zero; this
82 * means that we have to mark the end of the list with '1', not NULL.
84 * Extra notes for generational GC:
86 * Each generation has a static object list associated with it. When
87 * collecting generations up to N, we treat the static object lists
88 * from generations > N as roots.
90 * We build up a static object list while collecting generations 0..N,
91 * which is then appended to the static object list of generation N+1.
94 /* N is the oldest generation being collected, where the generations
95 * are numbered starting at 0. A major GC (indicated by the major_gc
96 * flag) is when we're collecting all generations. We only attempt to
97 * deal with static objects and GC CAFs when doing a major GC.
102 /* Data used for allocation area sizing.
104 static lnat g0s0_pcnt_kept = 30; // percentage of g0s0 live at last minor GC
114 /* Thread-local data for each GC thread
116 gc_thread **gc_threads = NULL;
118 #if !defined(THREADED_RTS)
119 StgWord8 the_gc_thread[sizeof(gc_thread) + 64 * sizeof(step_workspace)];
122 // Number of threads running in *this* GC. Affects how many
123 // step->todos[] lists we have to look in to find work.
127 long copied; // *words* copied & scavenged during this GC
129 rtsBool work_stealing;
133 /* -----------------------------------------------------------------------------
134 Static function declarations
135 -------------------------------------------------------------------------- */
137 static void mark_root (void *user, StgClosure **root);
138 static void zero_static_object_list (StgClosure* first_static);
139 static nat initialise_N (rtsBool force_major_gc);
140 static void init_collected_gen (nat g, nat threads);
141 static void init_uncollected_gen (nat g, nat threads);
142 static void init_gc_thread (gc_thread *t);
143 static void update_task_list (void);
144 static void resize_generations (void);
145 static void resize_nursery (void);
146 static void start_gc_threads (void);
147 static void scavenge_until_all_done (void);
148 static StgWord inc_running (void);
149 static StgWord dec_running (void);
150 static void wakeup_gc_threads (nat n_threads, nat me);
151 static void shutdown_gc_threads (nat n_threads, nat me);
153 #if 0 && defined(DEBUG)
154 static void gcCAFs (void);
157 /* -----------------------------------------------------------------------------
159 -------------------------------------------------------------------------- */
161 bdescr *mark_stack_top_bd; // topmost block in the mark stack
162 bdescr *mark_stack_bd; // current block in the mark stack
163 StgPtr mark_sp; // pointer to the next unallocated mark stack entry
165 /* -----------------------------------------------------------------------------
166 GarbageCollect: the main entry point to the garbage collector.
168 Locks held: all capabilities are held throughout GarbageCollect().
169 -------------------------------------------------------------------------- */
172 GarbageCollect (rtsBool force_major_gc,
173 nat gc_type USED_IF_THREADS,
178 lnat live, allocated, max_copied, avg_copied, slop;
179 gc_thread *saved_gct;
182 // necessary if we stole a callee-saves register for gct:
186 CostCentreStack *prev_CCS;
191 #if defined(RTS_USER_SIGNALS)
192 if (RtsFlags.MiscFlags.install_signal_handlers) {
198 ASSERT(sizeof(step_workspace) == 16 * sizeof(StgWord));
199 // otherwise adjust the padding in step_workspace.
201 // tell the stats department that we've started a GC
204 // tell the STM to discard any cached closures it's hoping to re-use
207 // lock the StablePtr table
216 // attribute any costs to CCS_GC
222 /* Approximate how much we allocated.
223 * Todo: only when generating stats?
225 allocated = calcAllocated();
227 /* Figure out which generation to collect
229 n = initialise_N(force_major_gc);
231 #if defined(THREADED_RTS)
232 work_stealing = RtsFlags.ParFlags.parGcLoadBalancingEnabled &&
233 N >= RtsFlags.ParFlags.parGcLoadBalancingGen;
234 // It's not always a good idea to do load balancing in parallel
235 // GC. In particular, for a parallel program we don't want to
236 // lose locality by moving cached data into another CPU's cache
237 // (this effect can be quite significant).
239 // We could have a more complex way to deterimine whether to do
240 // work stealing or not, e.g. it might be a good idea to do it
241 // if the heap is big. For now, we just turn it on or off with
245 /* Start threads, so they can be spinning up while we finish initialisation.
249 #if defined(THREADED_RTS)
250 /* How many threads will be participating in this GC?
251 * We don't try to parallelise minor GCs (unless the user asks for
252 * it with +RTS -gn0), or mark/compact/sweep GC.
254 if (gc_type == PENDING_GC_PAR) {
255 n_gc_threads = RtsFlags.ParFlags.nNodes;
263 debugTrace(DEBUG_gc, "GC (gen %d): %d KB to collect, %ld MB in use, using %d thread(s)",
264 N, n * (BLOCK_SIZE / 1024), mblocks_allocated, n_gc_threads);
266 #ifdef RTS_GTK_FRONTPANEL
267 if (RtsFlags.GcFlags.frontpanel) {
268 updateFrontPanelBeforeGC(N);
273 // check for memory leaks if DEBUG is on
274 memInventory(DEBUG_gc);
277 // check stack sanity *before* GC
278 IF_DEBUG(sanity, checkFreeListSanity());
279 IF_DEBUG(sanity, checkMutableLists(rtsTrue));
281 // Initialise all our gc_thread structures
282 for (t = 0; t < n_gc_threads; t++) {
283 init_gc_thread(gc_threads[t]);
286 // Initialise all the generations/steps that we're collecting.
287 for (g = 0; g <= N; g++) {
288 init_collected_gen(g,n_gc_threads);
291 // Initialise all the generations/steps that we're *not* collecting.
292 for (g = N+1; g < RtsFlags.GcFlags.generations; g++) {
293 init_uncollected_gen(g,n_gc_threads);
296 /* Allocate a mark stack if we're doing a major collection.
298 if (major_gc && oldest_gen->steps[0].mark) {
299 mark_stack_bd = allocBlock();
300 mark_stack_top_bd = mark_stack_bd;
301 mark_stack_bd->link = NULL;
302 mark_stack_bd->u.back = NULL;
303 mark_sp = mark_stack_bd->start;
305 mark_stack_bd = NULL;
306 mark_stack_top_bd = NULL;
310 // this is the main thread
312 if (n_gc_threads == 1) {
313 SET_GCT(gc_threads[0]);
315 SET_GCT(gc_threads[cap->no]);
318 SET_GCT(gc_threads[0]);
321 /* -----------------------------------------------------------------------
322 * follow all the roots that we know about:
325 // the main thread is running: this prevents any other threads from
326 // exiting prematurely, so we can start them now.
327 // NB. do this after the mutable lists have been saved above, otherwise
328 // the other GC threads will be writing into the old mutable lists.
330 wakeup_gc_threads(n_gc_threads, gct->thread_index);
332 // Mutable lists from each generation > N
333 // we want to *scavenge* these roots, not evacuate them: they're not
334 // going to move in this GC.
335 // Also do them in reverse generation order, for the usual reason:
336 // namely to reduce the likelihood of spurious old->new pointers.
338 for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
339 scavenge_mutable_list(generations[g].saved_mut_list, &generations[g]);
340 freeChain_sync(generations[g].saved_mut_list);
341 generations[g].saved_mut_list = NULL;
345 // scavenge the capability-private mutable lists. This isn't part
346 // of markSomeCapabilities() because markSomeCapabilities() can only
347 // call back into the GC via mark_root() (due to the gct register
349 if (n_gc_threads == 1) {
350 for (n = 0; n < n_capabilities; n++) {
351 scavenge_capability_mut_lists(&capabilities[n]);
354 scavenge_capability_mut_lists(&capabilities[gct->thread_index]);
357 // follow roots from the CAF list (used by GHCi)
359 markCAFs(mark_root, gct);
361 // follow all the roots that the application knows about.
363 markSomeCapabilities(mark_root, gct, gct->thread_index, n_gc_threads,
364 rtsTrue/*prune sparks*/);
366 #if defined(RTS_USER_SIGNALS)
367 // mark the signal handlers (signals should be already blocked)
368 markSignalHandlers(mark_root, gct);
371 // Mark the weak pointer list, and prepare to detect dead weak pointers.
375 // Mark the stable pointer table.
376 markStablePtrTable(mark_root, gct);
378 /* -------------------------------------------------------------------------
379 * Repeatedly scavenge all the areas we know about until there's no
380 * more scavenging to be done.
384 scavenge_until_all_done();
385 // The other threads are now stopped. We might recurse back to
386 // here, but from now on this is the only thread.
388 // if any blackholes are alive, make the threads that wait on
390 if (traverseBlackholeQueue()) {
395 // must be last... invariant is that everything is fully
396 // scavenged at this point.
397 if (traverseWeakPtrList()) { // returns rtsTrue if evaced something
402 // If we get to here, there's really nothing left to do.
406 shutdown_gc_threads(n_gc_threads, gct->thread_index);
408 // Update pointers from the Task list
411 // Now see which stable names are still alive.
415 // We call processHeapClosureForDead() on every closure destroyed during
416 // the current garbage collection, so we invoke LdvCensusForDead().
417 if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_LDV
418 || RtsFlags.ProfFlags.bioSelector != NULL)
422 // NO MORE EVACUATION AFTER THIS POINT!
424 // Two-space collector: free the old to-space.
425 // g0s0->old_blocks is the old nursery
426 // g0s0->blocks is to-space from the previous GC
427 if (RtsFlags.GcFlags.generations == 1) {
428 if (g0->steps[0].blocks != NULL) {
429 freeChain(g0->steps[0].blocks);
430 g0->steps[0].blocks = NULL;
434 // For each workspace, in each thread, move the copied blocks to the step
440 for (t = 0; t < n_gc_threads; t++) {
444 if (RtsFlags.GcFlags.generations == 1) {
449 for (; s < total_steps; s++) {
452 // Push the final block
454 push_scanned_block(ws->todo_bd, ws);
457 ASSERT(gct->scan_bd == NULL);
458 ASSERT(countBlocks(ws->scavd_list) == ws->n_scavd_blocks);
461 for (bd = ws->scavd_list; bd != NULL; bd = bd->link) {
462 ws->step->n_words += bd->free - bd->start;
466 prev->link = ws->step->blocks;
467 ws->step->blocks = ws->scavd_list;
469 ws->step->n_blocks += ws->n_scavd_blocks;
473 // Add all the partial blocks *after* we've added all the full
474 // blocks. This is so that we can grab the partial blocks back
475 // again and try to fill them up in the next GC.
476 for (t = 0; t < n_gc_threads; t++) {
480 if (RtsFlags.GcFlags.generations == 1) {
485 for (; s < total_steps; s++) {
489 for (bd = ws->part_list; bd != NULL; bd = next) {
491 if (bd->free == bd->start) {
493 ws->part_list = next;
500 ws->step->n_words += bd->free - bd->start;
505 prev->link = ws->step->blocks;
506 ws->step->blocks = ws->part_list;
508 ws->step->n_blocks += ws->n_part_blocks;
510 ASSERT(countBlocks(ws->step->blocks) == ws->step->n_blocks);
511 ASSERT(countOccupied(ws->step->blocks) == ws->step->n_words);
516 // Finally: compact or sweep the oldest generation.
517 if (major_gc && oldest_gen->steps[0].mark) {
518 if (oldest_gen->steps[0].compact)
519 compact(gct->scavenged_static_objects);
521 sweep(&oldest_gen->steps[0]);
524 /* run through all the generations/steps and tidy up
531 for (i=0; i < n_gc_threads; i++) {
532 if (n_gc_threads > 1) {
533 debugTrace(DEBUG_gc,"thread %d:", i);
534 debugTrace(DEBUG_gc," copied %ld", gc_threads[i]->copied * sizeof(W_));
535 debugTrace(DEBUG_gc," scanned %ld", gc_threads[i]->scanned * sizeof(W_));
536 debugTrace(DEBUG_gc," any_work %ld", gc_threads[i]->any_work);
537 debugTrace(DEBUG_gc," no_work %ld", gc_threads[i]->no_work);
538 debugTrace(DEBUG_gc," scav_find_work %ld", gc_threads[i]->scav_find_work);
540 copied += gc_threads[i]->copied;
541 max_copied = stg_max(gc_threads[i]->copied, max_copied);
543 if (n_gc_threads == 1) {
551 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
554 generations[g].collections++; // for stats
555 if (n_gc_threads > 1) generations[g].par_collections++;
558 // Count the mutable list as bytes "copied" for the purposes of
559 // stats. Every mutable list is copied during every GC.
561 nat mut_list_size = 0;
562 for (bd = generations[g].mut_list; bd != NULL; bd = bd->link) {
563 mut_list_size += bd->free - bd->start;
565 for (n = 0; n < n_capabilities; n++) {
566 for (bd = capabilities[n].mut_lists[g];
567 bd != NULL; bd = bd->link) {
568 mut_list_size += bd->free - bd->start;
571 copied += mut_list_size;
574 "mut_list_size: %lu (%d vars, %d arrays, %d MVARs, %d others)",
575 (unsigned long)(mut_list_size * sizeof(W_)),
576 mutlist_MUTVARS, mutlist_MUTARRS, mutlist_MVARS, mutlist_OTHERS);
579 for (s = 0; s < generations[g].n_steps; s++) {
581 stp = &generations[g].steps[s];
583 // for generations we collected...
586 /* free old memory and shift to-space into from-space for all
587 * the collected steps (except the allocation area). These
588 * freed blocks will probaby be quickly recycled.
590 if (!(g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1)) {
593 // tack the new blocks on the end of the existing blocks
594 if (stp->old_blocks != NULL) {
597 for (bd = stp->old_blocks; bd != NULL; bd = next) {
601 if (!(bd->flags & BF_MARKED))
604 stp->old_blocks = next;
613 stp->n_words += bd->free - bd->start;
615 // NB. this step might not be compacted next
616 // time, so reset the BF_MARKED flags.
617 // They are set before GC if we're going to
618 // compact. (search for BF_MARKED above).
619 bd->flags &= ~BF_MARKED;
621 // between GCs, all blocks in the heap except
622 // for the nursery have the BF_EVACUATED flag set.
623 bd->flags |= BF_EVACUATED;
630 prev->link = stp->blocks;
631 stp->blocks = stp->old_blocks;
634 // add the new blocks to the block tally
635 stp->n_blocks += stp->n_old_blocks;
636 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
637 ASSERT(countOccupied(stp->blocks) == stp->n_words);
641 freeChain(stp->old_blocks);
643 stp->old_blocks = NULL;
644 stp->n_old_blocks = 0;
647 /* LARGE OBJECTS. The current live large objects are chained on
648 * scavenged_large, having been moved during garbage
649 * collection from large_objects. Any objects left on the
650 * large_objects list are therefore dead, so we free them here.
652 freeChain(stp->large_objects);
653 stp->large_objects = stp->scavenged_large_objects;
654 stp->n_large_blocks = stp->n_scavenged_large_blocks;
655 ASSERT(countBlocks(stp->large_objects) == stp->n_large_blocks);
657 else // for older generations...
659 /* For older generations, we need to append the
660 * scavenged_large_object list (i.e. large objects that have been
661 * promoted during this GC) to the large_object list for that step.
663 for (bd = stp->scavenged_large_objects; bd; bd = next) {
665 dbl_link_onto(bd, &stp->large_objects);
668 // add the new blocks we promoted during this GC
669 stp->n_large_blocks += stp->n_scavenged_large_blocks;
670 ASSERT(countBlocks(stp->large_objects) == stp->n_large_blocks);
675 // update the max size of older generations after a major GC
676 resize_generations();
678 // Calculate the amount of live data for stats.
679 live = calcLiveWords();
681 // Free the small objects allocated via allocate(), since this will
682 // all have been copied into G0S1 now.
683 if (RtsFlags.GcFlags.generations > 1) {
684 if (g0->steps[0].blocks != NULL) {
685 freeChain(g0->steps[0].blocks);
686 g0->steps[0].blocks = NULL;
688 g0->steps[0].n_blocks = 0;
689 g0->steps[0].n_words = 0;
691 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
693 // Start a new pinned_object_block
694 for (n = 0; n < n_capabilities; n++) {
695 capabilities[n].pinned_object_block = NULL;
698 // Free the mark stack.
699 if (mark_stack_top_bd != NULL) {
700 debugTrace(DEBUG_gc, "mark stack: %d blocks",
701 countBlocks(mark_stack_top_bd));
702 freeChain(mark_stack_top_bd);
706 for (g = 0; g <= N; g++) {
707 for (s = 0; s < generations[g].n_steps; s++) {
708 stp = &generations[g].steps[s];
709 if (stp->bitmap != NULL) {
710 freeGroup(stp->bitmap);
718 // mark the garbage collected CAFs as dead
719 #if 0 && defined(DEBUG) // doesn't work at the moment
720 if (major_gc) { gcCAFs(); }
724 // resetStaticObjectForRetainerProfiling() must be called before
726 if (n_gc_threads > 1) {
727 barf("profiling is currently broken with multi-threaded GC");
728 // ToDo: fix the gct->scavenged_static_objects below
730 resetStaticObjectForRetainerProfiling(gct->scavenged_static_objects);
733 // zero the scavenged static object list
736 for (i = 0; i < n_gc_threads; i++) {
737 zero_static_object_list(gc_threads[i]->scavenged_static_objects);
744 // start any pending finalizers
746 scheduleFinalizers(cap, old_weak_ptr_list);
749 // send exceptions to any threads which were about to die
751 resurrectThreads(resurrected_threads);
752 performPendingThrowTos(exception_threads);
755 // Update the stable pointer hash table.
756 updateStablePtrTable(major_gc);
758 // check sanity after GC
759 IF_DEBUG(sanity, checkSanity());
761 // extra GC trace info
762 IF_DEBUG(gc, statDescribeGens());
765 // symbol-table based profiling
766 /* heapCensus(to_blocks); */ /* ToDo */
769 // restore enclosing cost centre
775 // check for memory leaks if DEBUG is on
776 memInventory(DEBUG_gc);
779 #ifdef RTS_GTK_FRONTPANEL
780 if (RtsFlags.GcFlags.frontpanel) {
781 updateFrontPanelAfterGC( N, live );
785 // ok, GC over: tell the stats department what happened.
786 slop = calcLiveBlocks() * BLOCK_SIZE_W - live;
787 stat_endGC(allocated, live, copied, N, max_copied, avg_copied, slop);
789 // unlock the StablePtr table
792 // Guess which generation we'll collect *next* time
793 initialise_N(force_major_gc);
795 #if defined(RTS_USER_SIGNALS)
796 if (RtsFlags.MiscFlags.install_signal_handlers) {
797 // unblock signals again
798 unblockUserSignals();
807 /* -----------------------------------------------------------------------------
808 Figure out which generation to collect, initialise N and major_gc.
810 Also returns the total number of blocks in generations that will be
812 -------------------------------------------------------------------------- */
815 initialise_N (rtsBool force_major_gc)
818 nat s, blocks, blocks_total;
823 if (force_major_gc) {
824 N = RtsFlags.GcFlags.generations - 1;
829 for (g = RtsFlags.GcFlags.generations - 1; g >= 0; g--) {
831 for (s = 0; s < generations[g].n_steps; s++) {
832 blocks += generations[g].steps[s].n_words / BLOCK_SIZE_W;
833 blocks += generations[g].steps[s].n_large_blocks;
835 if (blocks >= generations[g].max_blocks) {
839 blocks_total += blocks;
843 blocks_total += countNurseryBlocks();
845 major_gc = (N == RtsFlags.GcFlags.generations-1);
849 /* -----------------------------------------------------------------------------
850 Initialise the gc_thread structures.
851 -------------------------------------------------------------------------- */
853 #define GC_THREAD_INACTIVE 0
854 #define GC_THREAD_STANDING_BY 1
855 #define GC_THREAD_RUNNING 2
856 #define GC_THREAD_WAITING_TO_CONTINUE 3
859 new_gc_thread (nat n, gc_thread *t)
866 initSpinLock(&t->gc_spin);
867 initSpinLock(&t->mut_spin);
868 ACQUIRE_SPIN_LOCK(&t->gc_spin);
869 t->wakeup = GC_THREAD_INACTIVE; // starts true, so we can wait for the
870 // thread to start up, see wakeup_gc_threads
874 t->free_blocks = NULL;
883 for (s = 0; s < total_steps; s++)
886 ws->step = &all_steps[s];
887 ASSERT(s == ws->step->abs_no);
891 ws->todo_q = newWSDeque(128);
892 ws->todo_overflow = NULL;
893 ws->n_todo_overflow = 0;
895 ws->part_list = NULL;
896 ws->n_part_blocks = 0;
898 ws->scavd_list = NULL;
899 ws->n_scavd_blocks = 0;
907 if (gc_threads == NULL) {
908 #if defined(THREADED_RTS)
910 gc_threads = stgMallocBytes (RtsFlags.ParFlags.nNodes *
914 for (i = 0; i < RtsFlags.ParFlags.nNodes; i++) {
916 stgMallocBytes(sizeof(gc_thread) + total_steps * sizeof(step_workspace),
919 new_gc_thread(i, gc_threads[i]);
922 gc_threads = stgMallocBytes (sizeof(gc_thread*),"alloc_gc_threads");
924 new_gc_thread(0,gc_threads[0]);
933 if (gc_threads != NULL) {
934 #if defined(THREADED_RTS)
936 for (i = 0; i < n_capabilities; i++) {
937 for (s = 0; s < total_steps; s++)
939 freeWSDeque(gc_threads[i]->steps[s].todo_q);
941 stgFree (gc_threads[i]);
943 stgFree (gc_threads);
945 for (s = 0; s < total_steps; s++)
947 freeWSDeque(gc_threads[0]->steps[s].todo_q);
949 stgFree (gc_threads);
955 /* ----------------------------------------------------------------------------
957 ------------------------------------------------------------------------- */
959 static volatile StgWord gc_running_threads;
965 new = atomic_inc(&gc_running_threads);
966 ASSERT(new <= n_gc_threads);
973 ASSERT(gc_running_threads != 0);
974 return atomic_dec(&gc_running_threads);
987 // scavenge objects in compacted generation
988 if (mark_stack_bd != NULL && !mark_stack_empty()) {
992 // Check for global work in any step. We don't need to check for
993 // local work, because we have already exited scavenge_loop(),
994 // which means there is no local work for this thread.
995 for (s = total_steps-1; s >= 0; s--) {
996 if (s == 0 && RtsFlags.GcFlags.generations > 1) {
1000 if (ws->todo_large_objects) return rtsTrue;
1001 if (!looksEmptyWSDeque(ws->todo_q)) return rtsTrue;
1002 if (ws->todo_overflow) return rtsTrue;
1005 #if defined(THREADED_RTS)
1006 if (work_stealing) {
1008 // look for work to steal
1009 for (n = 0; n < n_gc_threads; n++) {
1010 if (n == gct->thread_index) continue;
1011 for (s = total_steps-1; s >= 0; s--) {
1012 ws = &gc_threads[n]->steps[s];
1013 if (!looksEmptyWSDeque(ws->todo_q)) return rtsTrue;
1025 scavenge_until_all_done (void)
1031 traceEvent(&capabilities[gct->thread_index], EVENT_GC_WORK);
1033 #if defined(THREADED_RTS)
1034 if (n_gc_threads > 1) {
1043 // scavenge_loop() only exits when there's no work to do
1046 traceEvent(&capabilities[gct->thread_index], EVENT_GC_IDLE);
1048 debugTrace(DEBUG_gc, "%d GC threads still running", r);
1050 while (gc_running_threads != 0) {
1056 // any_work() does not remove the work from the queue, it
1057 // just checks for the presence of work. If we find any,
1058 // then we increment gc_running_threads and go back to
1059 // scavenge_loop() to perform any pending work.
1062 traceEvent(&capabilities[gct->thread_index], EVENT_GC_DONE);
1065 #if defined(THREADED_RTS)
1068 gcWorkerThread (Capability *cap)
1070 gc_thread *saved_gct;
1072 // necessary if we stole a callee-saves register for gct:
1075 cap->in_gc = rtsTrue;
1077 gct = gc_threads[cap->no];
1078 gct->id = osThreadId();
1080 // Wait until we're told to wake up
1081 RELEASE_SPIN_LOCK(&gct->mut_spin);
1082 gct->wakeup = GC_THREAD_STANDING_BY;
1083 debugTrace(DEBUG_gc, "GC thread %d standing by...", gct->thread_index);
1084 ACQUIRE_SPIN_LOCK(&gct->gc_spin);
1087 // start performance counters in this thread...
1088 if (gct->papi_events == -1) {
1089 papi_init_eventset(&gct->papi_events);
1091 papi_thread_start_gc1_count(gct->papi_events);
1094 // Every thread evacuates some roots.
1096 markSomeCapabilities(mark_root, gct, gct->thread_index, n_gc_threads,
1097 rtsTrue/*prune sparks*/);
1098 scavenge_capability_mut_lists(&capabilities[gct->thread_index]);
1100 scavenge_until_all_done();
1103 // count events in this thread towards the GC totals
1104 papi_thread_stop_gc1_count(gct->papi_events);
1107 // Wait until we're told to continue
1108 RELEASE_SPIN_LOCK(&gct->gc_spin);
1109 gct->wakeup = GC_THREAD_WAITING_TO_CONTINUE;
1110 debugTrace(DEBUG_gc, "GC thread %d waiting to continue...",
1112 ACQUIRE_SPIN_LOCK(&gct->mut_spin);
1113 debugTrace(DEBUG_gc, "GC thread %d on my way...", gct->thread_index);
1120 #if defined(THREADED_RTS)
1123 waitForGcThreads (Capability *cap USED_IF_THREADS)
1125 nat n_threads = RtsFlags.ParFlags.nNodes;
1128 rtsBool retry = rtsTrue;
1131 for (i=0; i < n_threads; i++) {
1132 if (i == me) continue;
1133 if (gc_threads[i]->wakeup != GC_THREAD_STANDING_BY) {
1134 prodCapability(&capabilities[i], cap->running_task);
1137 for (j=0; j < 10; j++) {
1139 for (i=0; i < n_threads; i++) {
1140 if (i == me) continue;
1142 setContextSwitches();
1143 if (gc_threads[i]->wakeup != GC_THREAD_STANDING_BY) {
1153 #endif // THREADED_RTS
1156 start_gc_threads (void)
1158 #if defined(THREADED_RTS)
1159 gc_running_threads = 0;
1164 wakeup_gc_threads (nat n_threads USED_IF_THREADS, nat me USED_IF_THREADS)
1166 #if defined(THREADED_RTS)
1168 for (i=0; i < n_threads; i++) {
1169 if (i == me) continue;
1171 debugTrace(DEBUG_gc, "waking up gc thread %d", i);
1172 if (gc_threads[i]->wakeup != GC_THREAD_STANDING_BY) barf("wakeup_gc_threads");
1174 gc_threads[i]->wakeup = GC_THREAD_RUNNING;
1175 ACQUIRE_SPIN_LOCK(&gc_threads[i]->mut_spin);
1176 RELEASE_SPIN_LOCK(&gc_threads[i]->gc_spin);
1181 // After GC is complete, we must wait for all GC threads to enter the
1182 // standby state, otherwise they may still be executing inside
1183 // any_work(), and may even remain awake until the next GC starts.
1185 shutdown_gc_threads (nat n_threads USED_IF_THREADS, nat me USED_IF_THREADS)
1187 #if defined(THREADED_RTS)
1189 for (i=0; i < n_threads; i++) {
1190 if (i == me) continue;
1191 while (gc_threads[i]->wakeup != GC_THREAD_WAITING_TO_CONTINUE) { write_barrier(); }
1196 #if defined(THREADED_RTS)
1198 releaseGCThreads (Capability *cap USED_IF_THREADS)
1200 nat n_threads = RtsFlags.ParFlags.nNodes;
1203 for (i=0; i < n_threads; i++) {
1204 if (i == me) continue;
1205 if (gc_threads[i]->wakeup != GC_THREAD_WAITING_TO_CONTINUE)
1206 barf("releaseGCThreads");
1208 gc_threads[i]->wakeup = GC_THREAD_INACTIVE;
1209 ACQUIRE_SPIN_LOCK(&gc_threads[i]->gc_spin);
1210 RELEASE_SPIN_LOCK(&gc_threads[i]->mut_spin);
1215 /* ----------------------------------------------------------------------------
1216 Initialise a generation that is to be collected
1217 ------------------------------------------------------------------------- */
1220 init_collected_gen (nat g, nat n_threads)
1227 // Throw away the current mutable list. Invariant: the mutable
1228 // list always has at least one block; this means we can avoid a
1229 // check for NULL in recordMutable().
1231 freeChain(generations[g].mut_list);
1232 generations[g].mut_list = allocBlock();
1233 for (i = 0; i < n_capabilities; i++) {
1234 freeChain(capabilities[i].mut_lists[g]);
1235 capabilities[i].mut_lists[g] = allocBlock();
1240 for (i = 0; i < n_capabilities; i++) {
1241 stp = &nurseries[i];
1242 stp->old_threads = stp->threads;
1243 stp->threads = END_TSO_QUEUE;
1247 for (s = 0; s < generations[g].n_steps; s++) {
1249 // generation 0, step 0 doesn't need to-space, unless -G1
1250 if (g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1) {
1254 stp = &generations[g].steps[s];
1255 ASSERT(stp->gen_no == g);
1257 // we'll construct a new list of threads in this step
1258 // during GC, throw away the current list.
1259 stp->old_threads = stp->threads;
1260 stp->threads = END_TSO_QUEUE;
1262 // deprecate the existing blocks
1263 stp->old_blocks = stp->blocks;
1264 stp->n_old_blocks = stp->n_blocks;
1268 stp->live_estimate = 0;
1270 // initialise the large object queues.
1271 stp->scavenged_large_objects = NULL;
1272 stp->n_scavenged_large_blocks = 0;
1274 // mark the small objects as from-space
1275 for (bd = stp->old_blocks; bd; bd = bd->link) {
1276 bd->flags &= ~BF_EVACUATED;
1279 // mark the large objects as from-space
1280 for (bd = stp->large_objects; bd; bd = bd->link) {
1281 bd->flags &= ~BF_EVACUATED;
1284 // for a compacted step, we need to allocate the bitmap
1286 nat bitmap_size; // in bytes
1287 bdescr *bitmap_bdescr;
1290 bitmap_size = stp->n_old_blocks * BLOCK_SIZE / (sizeof(W_)*BITS_PER_BYTE);
1292 if (bitmap_size > 0) {
1293 bitmap_bdescr = allocGroup((lnat)BLOCK_ROUND_UP(bitmap_size)
1295 stp->bitmap = bitmap_bdescr;
1296 bitmap = bitmap_bdescr->start;
1298 debugTrace(DEBUG_gc, "bitmap_size: %d, bitmap: %p",
1299 bitmap_size, bitmap);
1301 // don't forget to fill it with zeros!
1302 memset(bitmap, 0, bitmap_size);
1304 // For each block in this step, point to its bitmap from the
1305 // block descriptor.
1306 for (bd=stp->old_blocks; bd != NULL; bd = bd->link) {
1307 bd->u.bitmap = bitmap;
1308 bitmap += BLOCK_SIZE_W / (sizeof(W_)*BITS_PER_BYTE);
1310 // Also at this point we set the BF_MARKED flag
1311 // for this block. The invariant is that
1312 // BF_MARKED is always unset, except during GC
1313 // when it is set on those blocks which will be
1315 if (!(bd->flags & BF_FRAGMENTED)) {
1316 bd->flags |= BF_MARKED;
1323 // For each GC thread, for each step, allocate a "todo" block to
1324 // store evacuated objects to be scavenged, and a block to store
1325 // evacuated objects that do not need to be scavenged.
1326 for (t = 0; t < n_threads; t++) {
1327 for (s = 0; s < generations[g].n_steps; s++) {
1329 // we don't copy objects into g0s0, unless -G0
1330 if (g==0 && s==0 && RtsFlags.GcFlags.generations > 1) continue;
1332 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1334 ws->todo_large_objects = NULL;
1336 ws->part_list = NULL;
1337 ws->n_part_blocks = 0;
1339 // allocate the first to-space block; extra blocks will be
1340 // chained on as necessary.
1342 ASSERT(looksEmptyWSDeque(ws->todo_q));
1343 alloc_todo_block(ws,0);
1345 ws->todo_overflow = NULL;
1346 ws->n_todo_overflow = 0;
1348 ws->scavd_list = NULL;
1349 ws->n_scavd_blocks = 0;
1355 /* ----------------------------------------------------------------------------
1356 Initialise a generation that is *not* to be collected
1357 ------------------------------------------------------------------------- */
1360 init_uncollected_gen (nat g, nat threads)
1367 // save the current mutable lists for this generation, and
1368 // allocate a fresh block for each one. We'll traverse these
1369 // mutable lists as roots early on in the GC.
1370 generations[g].saved_mut_list = generations[g].mut_list;
1371 generations[g].mut_list = allocBlock();
1372 for (n = 0; n < n_capabilities; n++) {
1373 capabilities[n].saved_mut_lists[g] = capabilities[n].mut_lists[g];
1374 capabilities[n].mut_lists[g] = allocBlock();
1377 for (s = 0; s < generations[g].n_steps; s++) {
1378 stp = &generations[g].steps[s];
1379 stp->scavenged_large_objects = NULL;
1380 stp->n_scavenged_large_blocks = 0;
1383 for (s = 0; s < generations[g].n_steps; s++) {
1385 stp = &generations[g].steps[s];
1387 for (t = 0; t < threads; t++) {
1388 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1390 ASSERT(looksEmptyWSDeque(ws->todo_q));
1391 ws->todo_large_objects = NULL;
1393 ws->part_list = NULL;
1394 ws->n_part_blocks = 0;
1396 ws->scavd_list = NULL;
1397 ws->n_scavd_blocks = 0;
1399 // If the block at the head of the list in this generation
1400 // is less than 3/4 full, then use it as a todo block.
1401 if (stp->blocks && isPartiallyFull(stp->blocks))
1403 ws->todo_bd = stp->blocks;
1404 ws->todo_free = ws->todo_bd->free;
1405 ws->todo_lim = ws->todo_bd->start + BLOCK_SIZE_W;
1406 stp->blocks = stp->blocks->link;
1408 stp->n_words -= ws->todo_bd->free - ws->todo_bd->start;
1409 ws->todo_bd->link = NULL;
1410 // we must scan from the current end point.
1411 ws->todo_bd->u.scan = ws->todo_bd->free;
1416 alloc_todo_block(ws,0);
1420 // deal out any more partial blocks to the threads' part_lists
1422 while (stp->blocks && isPartiallyFull(stp->blocks))
1425 stp->blocks = bd->link;
1426 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1427 bd->link = ws->part_list;
1429 ws->n_part_blocks += 1;
1430 bd->u.scan = bd->free;
1432 stp->n_words -= bd->free - bd->start;
1434 if (t == n_gc_threads) t = 0;
1439 /* -----------------------------------------------------------------------------
1440 Initialise a gc_thread before GC
1441 -------------------------------------------------------------------------- */
1444 init_gc_thread (gc_thread *t)
1446 t->static_objects = END_OF_STATIC_LIST;
1447 t->scavenged_static_objects = END_OF_STATIC_LIST;
1449 t->mut_lists = capabilities[t->thread_index].mut_lists;
1451 t->failed_to_evac = rtsFalse;
1452 t->eager_promotion = rtsTrue;
1453 t->thunk_selector_depth = 0;
1458 t->scav_find_work = 0;
1461 /* -----------------------------------------------------------------------------
1462 Function we pass to evacuate roots.
1463 -------------------------------------------------------------------------- */
1466 mark_root(void *user USED_IF_THREADS, StgClosure **root)
1468 // we stole a register for gct, but this function is called from
1469 // *outside* the GC where the register variable is not in effect,
1470 // so we need to save and restore it here. NB. only call
1471 // mark_root() from the main GC thread, otherwise gct will be
1473 gc_thread *saved_gct;
1482 /* -----------------------------------------------------------------------------
1483 Initialising the static object & mutable lists
1484 -------------------------------------------------------------------------- */
1487 zero_static_object_list(StgClosure* first_static)
1491 const StgInfoTable *info;
1493 for (p = first_static; p != END_OF_STATIC_LIST; p = link) {
1495 link = *STATIC_LINK(info, p);
1496 *STATIC_LINK(info,p) = NULL;
1500 /* ----------------------------------------------------------------------------
1501 Update the pointers from the task list
1503 These are treated as weak pointers because we want to allow a main
1504 thread to get a BlockedOnDeadMVar exception in the same way as any
1505 other thread. Note that the threads should all have been retained
1506 by GC by virtue of being on the all_threads list, we're just
1507 updating pointers here.
1508 ------------------------------------------------------------------------- */
1511 update_task_list (void)
1515 for (task = all_tasks; task != NULL; task = task->all_link) {
1516 if (!task->stopped && task->tso) {
1517 ASSERT(task->tso->bound == task);
1518 tso = (StgTSO *) isAlive((StgClosure *)task->tso);
1520 barf("task %p: main thread %d has been GC'd",
1533 /* ----------------------------------------------------------------------------
1534 Reset the sizes of the older generations when we do a major
1537 CURRENT STRATEGY: make all generations except zero the same size.
1538 We have to stay within the maximum heap size, and leave a certain
1539 percentage of the maximum heap size available to allocate into.
1540 ------------------------------------------------------------------------- */
1543 resize_generations (void)
1547 if (major_gc && RtsFlags.GcFlags.generations > 1) {
1548 nat live, size, min_alloc, words;
1549 nat max = RtsFlags.GcFlags.maxHeapSize;
1550 nat gens = RtsFlags.GcFlags.generations;
1552 // live in the oldest generations
1553 if (oldest_gen->steps[0].live_estimate != 0) {
1554 words = oldest_gen->steps[0].live_estimate;
1556 words = oldest_gen->steps[0].n_words;
1558 live = (words + BLOCK_SIZE_W - 1) / BLOCK_SIZE_W +
1559 oldest_gen->steps[0].n_large_blocks;
1561 // default max size for all generations except zero
1562 size = stg_max(live * RtsFlags.GcFlags.oldGenFactor,
1563 RtsFlags.GcFlags.minOldGenSize);
1565 if (RtsFlags.GcFlags.heapSizeSuggestionAuto) {
1566 RtsFlags.GcFlags.heapSizeSuggestion = size;
1569 // minimum size for generation zero
1570 min_alloc = stg_max((RtsFlags.GcFlags.pcFreeHeap * max) / 200,
1571 RtsFlags.GcFlags.minAllocAreaSize);
1573 // Auto-enable compaction when the residency reaches a
1574 // certain percentage of the maximum heap size (default: 30%).
1575 if (RtsFlags.GcFlags.generations > 1 &&
1576 (RtsFlags.GcFlags.compact ||
1578 oldest_gen->steps[0].n_blocks >
1579 (RtsFlags.GcFlags.compactThreshold * max) / 100))) {
1580 oldest_gen->steps[0].mark = 1;
1581 oldest_gen->steps[0].compact = 1;
1582 // debugBelch("compaction: on\n", live);
1584 oldest_gen->steps[0].mark = 0;
1585 oldest_gen->steps[0].compact = 0;
1586 // debugBelch("compaction: off\n", live);
1589 if (RtsFlags.GcFlags.sweep) {
1590 oldest_gen->steps[0].mark = 1;
1593 // if we're going to go over the maximum heap size, reduce the
1594 // size of the generations accordingly. The calculation is
1595 // different if compaction is turned on, because we don't need
1596 // to double the space required to collect the old generation.
1599 // this test is necessary to ensure that the calculations
1600 // below don't have any negative results - we're working
1601 // with unsigned values here.
1602 if (max < min_alloc) {
1606 if (oldest_gen->steps[0].compact) {
1607 if ( (size + (size - 1) * (gens - 2) * 2) + min_alloc > max ) {
1608 size = (max - min_alloc) / ((gens - 1) * 2 - 1);
1611 if ( (size * (gens - 1) * 2) + min_alloc > max ) {
1612 size = (max - min_alloc) / ((gens - 1) * 2);
1622 debugBelch("live: %d, min_alloc: %d, size : %d, max = %d\n", live,
1623 min_alloc, size, max);
1626 for (g = 0; g < gens; g++) {
1627 generations[g].max_blocks = size;
1632 /* -----------------------------------------------------------------------------
1633 Calculate the new size of the nursery, and resize it.
1634 -------------------------------------------------------------------------- */
1637 resize_nursery (void)
1639 lnat min_nursery = RtsFlags.GcFlags.minAllocAreaSize * n_capabilities;
1641 if (RtsFlags.GcFlags.generations == 1)
1642 { // Two-space collector:
1645 /* set up a new nursery. Allocate a nursery size based on a
1646 * function of the amount of live data (by default a factor of 2)
1647 * Use the blocks from the old nursery if possible, freeing up any
1650 * If we get near the maximum heap size, then adjust our nursery
1651 * size accordingly. If the nursery is the same size as the live
1652 * data (L), then we need 3L bytes. We can reduce the size of the
1653 * nursery to bring the required memory down near 2L bytes.
1655 * A normal 2-space collector would need 4L bytes to give the same
1656 * performance we get from 3L bytes, reducing to the same
1657 * performance at 2L bytes.
1659 blocks = generations[0].steps[0].n_blocks;
1661 if ( RtsFlags.GcFlags.maxHeapSize != 0 &&
1662 blocks * RtsFlags.GcFlags.oldGenFactor * 2 >
1663 RtsFlags.GcFlags.maxHeapSize )
1665 long adjusted_blocks; // signed on purpose
1668 adjusted_blocks = (RtsFlags.GcFlags.maxHeapSize - 2 * blocks);
1670 debugTrace(DEBUG_gc, "near maximum heap size of 0x%x blocks, blocks = %d, adjusted to %ld",
1671 RtsFlags.GcFlags.maxHeapSize, blocks, adjusted_blocks);
1673 pc_free = adjusted_blocks * 100 / RtsFlags.GcFlags.maxHeapSize;
1674 if (pc_free < RtsFlags.GcFlags.pcFreeHeap) /* might even * be < 0 */
1678 blocks = adjusted_blocks;
1682 blocks *= RtsFlags.GcFlags.oldGenFactor;
1683 if (blocks < min_nursery)
1685 blocks = min_nursery;
1688 resizeNurseries(blocks);
1690 else // Generational collector
1693 * If the user has given us a suggested heap size, adjust our
1694 * allocation area to make best use of the memory available.
1696 if (RtsFlags.GcFlags.heapSizeSuggestion)
1699 nat needed = calcNeeded(); // approx blocks needed at next GC
1701 /* Guess how much will be live in generation 0 step 0 next time.
1702 * A good approximation is obtained by finding the
1703 * percentage of g0s0 that was live at the last minor GC.
1705 * We have an accurate figure for the amount of copied data in
1706 * 'copied', but we must convert this to a number of blocks, with
1707 * a small adjustment for estimated slop at the end of a block
1712 g0s0_pcnt_kept = ((copied / (BLOCK_SIZE_W - 10)) * 100)
1713 / countNurseryBlocks();
1716 /* Estimate a size for the allocation area based on the
1717 * information available. We might end up going slightly under
1718 * or over the suggested heap size, but we should be pretty
1721 * Formula: suggested - needed
1722 * ----------------------------
1723 * 1 + g0s0_pcnt_kept/100
1725 * where 'needed' is the amount of memory needed at the next
1726 * collection for collecting all steps except g0s0.
1729 (((long)RtsFlags.GcFlags.heapSizeSuggestion - (long)needed) * 100) /
1730 (100 + (long)g0s0_pcnt_kept);
1732 if (blocks < (long)min_nursery) {
1733 blocks = min_nursery;
1736 resizeNurseries((nat)blocks);
1740 // we might have added extra large blocks to the nursery, so
1741 // resize back to minAllocAreaSize again.
1742 resizeNurseriesFixed(RtsFlags.GcFlags.minAllocAreaSize);
1747 /* -----------------------------------------------------------------------------
1748 Sanity code for CAF garbage collection.
1750 With DEBUG turned on, we manage a CAF list in addition to the SRT
1751 mechanism. After GC, we run down the CAF list and blackhole any
1752 CAFs which have been garbage collected. This means we get an error
1753 whenever the program tries to enter a garbage collected CAF.
1755 Any garbage collected CAFs are taken off the CAF list at the same
1757 -------------------------------------------------------------------------- */
1759 #if 0 && defined(DEBUG)
1766 const StgInfoTable *info;
1777 ASSERT(info->type == IND_STATIC);
1779 if (STATIC_LINK(info,p) == NULL) {
1780 debugTrace(DEBUG_gccafs, "CAF gc'd at 0x%04lx", (long)p);
1782 SET_INFO(p,&stg_BLACKHOLE_info);
1783 p = STATIC_LINK2(info,p);
1787 pp = &STATIC_LINK2(info,p);
1794 debugTrace(DEBUG_gccafs, "%d CAFs live", i);