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"
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"
54 #include <string.h> // for memset()
57 /* -----------------------------------------------------------------------------
59 -------------------------------------------------------------------------- */
61 /* STATIC OBJECT LIST.
64 * We maintain a linked list of static objects that are still live.
65 * The requirements for this list are:
67 * - we need to scan the list while adding to it, in order to
68 * scavenge all the static objects (in the same way that
69 * breadth-first scavenging works for dynamic objects).
71 * - we need to be able to tell whether an object is already on
72 * the list, to break loops.
74 * Each static object has a "static link field", which we use for
75 * linking objects on to the list. We use a stack-type list, consing
76 * objects on the front as they are added (this means that the
77 * scavenge phase is depth-first, not breadth-first, but that
80 * A separate list is kept for objects that have been scavenged
81 * already - this is so that we can zero all the marks afterwards.
83 * An object is on the list if its static link field is non-zero; this
84 * means that we have to mark the end of the list with '1', not NULL.
86 * Extra notes for generational GC:
88 * Each generation has a static object list associated with it. When
89 * collecting generations up to N, we treat the static object lists
90 * from generations > N as roots.
92 * We build up a static object list while collecting generations 0..N,
93 * which is then appended to the static object list of generation N+1.
96 /* N is the oldest generation being collected, where the generations
97 * are numbered starting at 0. A major GC (indicated by the major_gc
98 * flag) is when we're collecting all generations. We only attempt to
99 * deal with static objects and GC CAFs when doing a major GC.
104 /* Data used for allocation area sizing.
106 static lnat g0s0_pcnt_kept = 30; // percentage of g0s0 live at last minor GC
116 /* Thread-local data for each GC thread
118 gc_thread **gc_threads = NULL;
119 // gc_thread *gct = NULL; // this thread's gct TODO: make thread-local
121 // Number of threads running in *this* GC. Affects how many
122 // step->todos[] lists we have to look in to find work.
126 long copied; // *words* copied & scavenged during this GC
130 /* -----------------------------------------------------------------------------
131 Static function declarations
132 -------------------------------------------------------------------------- */
134 static void mark_root (void *user, StgClosure **root);
135 static void zero_static_object_list (StgClosure* first_static);
136 static nat initialise_N (rtsBool force_major_gc);
137 static void init_collected_gen (nat g, nat threads);
138 static void init_uncollected_gen (nat g, nat threads);
139 static void init_gc_thread (gc_thread *t);
140 static void update_task_list (void);
141 static void resize_generations (void);
142 static void resize_nursery (void);
143 static void start_gc_threads (void);
144 static void scavenge_until_all_done (void);
145 static nat inc_running (void);
146 static nat dec_running (void);
147 static void wakeup_gc_threads (nat n_threads, nat me);
148 static void shutdown_gc_threads (nat n_threads, nat me);
150 #if 0 && defined(DEBUG)
151 static void gcCAFs (void);
154 /* -----------------------------------------------------------------------------
155 The mark bitmap & stack.
156 -------------------------------------------------------------------------- */
158 #define MARK_STACK_BLOCKS 4
160 bdescr *mark_stack_bdescr;
165 // Flag and pointers used for falling back to a linear scan when the
166 // mark stack overflows.
167 rtsBool mark_stack_overflowed;
168 bdescr *oldgen_scan_bd;
171 /* -----------------------------------------------------------------------------
172 GarbageCollect: the main entry point to the garbage collector.
174 Locks held: all capabilities are held throughout GarbageCollect().
175 -------------------------------------------------------------------------- */
178 GarbageCollect (rtsBool force_major_gc,
179 nat gc_type USED_IF_THREADS,
180 Capability *cap USED_IF_THREADS)
184 lnat live, allocated, max_copied, avg_copied, slop;
185 gc_thread *saved_gct;
188 // necessary if we stole a callee-saves register for gct:
192 CostCentreStack *prev_CCS;
197 #if defined(RTS_USER_SIGNALS)
198 if (RtsFlags.MiscFlags.install_signal_handlers) {
204 ASSERT(sizeof(step_workspace) == 16 * sizeof(StgWord));
205 // otherwise adjust the padding in step_workspace.
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 /* Start threads, so they can be spinning up while we finish initialisation.
238 #if defined(THREADED_RTS)
239 /* How many threads will be participating in this GC?
240 * We don't try to parallelise minor GCs (unless the user asks for
241 * it with +RTS -gn0), or mark/compact/sweep GC.
243 if (gc_type == PENDING_GC_PAR) {
244 n_gc_threads = RtsFlags.ParFlags.nNodes;
252 trace(TRACE_gc|DEBUG_gc, "GC (gen %d): %d KB to collect, %ld MB in use, using %d thread(s)",
253 N, n * (BLOCK_SIZE / 1024), mblocks_allocated, n_gc_threads);
255 #ifdef RTS_GTK_FRONTPANEL
256 if (RtsFlags.GcFlags.frontpanel) {
257 updateFrontPanelBeforeGC(N);
262 // check for memory leaks if DEBUG is on
263 memInventory(traceClass(DEBUG_gc));
266 // check stack sanity *before* GC
267 IF_DEBUG(sanity, checkFreeListSanity());
268 IF_DEBUG(sanity, checkMutableLists(rtsTrue));
270 // Initialise all our gc_thread structures
271 for (t = 0; t < n_gc_threads; t++) {
272 init_gc_thread(gc_threads[t]);
275 // Initialise all the generations/steps that we're collecting.
276 for (g = 0; g <= N; g++) {
277 init_collected_gen(g,n_gc_threads);
280 // Initialise all the generations/steps that we're *not* collecting.
281 for (g = N+1; g < RtsFlags.GcFlags.generations; g++) {
282 init_uncollected_gen(g,n_gc_threads);
285 /* Allocate a mark stack if we're doing a major collection.
287 if (major_gc && oldest_gen->steps[0].mark) {
288 nat mark_stack_blocks;
289 mark_stack_blocks = stg_max(MARK_STACK_BLOCKS,
290 oldest_gen->steps[0].n_old_blocks / 100);
291 mark_stack_bdescr = allocGroup(mark_stack_blocks);
292 mark_stack = (StgPtr *)mark_stack_bdescr->start;
293 mark_sp = mark_stack;
294 mark_splim = mark_stack + (mark_stack_blocks * BLOCK_SIZE_W);
296 mark_stack_bdescr = NULL;
299 // this is the main thread
301 if (n_gc_threads == 1) {
304 gct = gc_threads[cap->no];
310 /* -----------------------------------------------------------------------
311 * follow all the roots that we know about:
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, gct->thread_index);
321 // Mutable lists from each generation > N
322 // we want to *scavenge* these roots, not evacuate them: they're not
323 // going to move in this GC.
324 // Also do them in reverse generation order, for the usual reason:
325 // namely to reduce the likelihood of spurious old->new pointers.
327 for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
328 scavenge_mutable_list(generations[g].saved_mut_list, &generations[g]);
329 freeChain_sync(generations[g].saved_mut_list);
330 generations[g].saved_mut_list = NULL;
334 // scavenge the capability-private mutable lists. This isn't part
335 // of markSomeCapabilities() because markSomeCapabilities() can only
336 // call back into the GC via mark_root() (due to the gct register
338 if (n_gc_threads == 1) {
339 for (n = 0; n < n_capabilities; n++) {
340 scavenge_capability_mut_lists(&capabilities[n]);
343 scavenge_capability_mut_lists(&capabilities[gct->thread_index]);
346 // follow roots from the CAF list (used by GHCi)
348 markCAFs(mark_root, gct);
350 // follow all the roots that the application knows about.
352 markSomeCapabilities(mark_root, gct, gct->thread_index, n_gc_threads,
353 rtsTrue/*prune sparks*/);
355 #if defined(RTS_USER_SIGNALS)
356 // mark the signal handlers (signals should be already blocked)
357 markSignalHandlers(mark_root, gct);
360 // Mark the weak pointer list, and prepare to detect dead weak pointers.
364 // Mark the stable pointer table.
365 markStablePtrTable(mark_root, gct);
367 /* -------------------------------------------------------------------------
368 * Repeatedly scavenge all the areas we know about until there's no
369 * more scavenging to be done.
373 scavenge_until_all_done();
374 // The other threads are now stopped. We might recurse back to
375 // here, but from now on this is the only thread.
377 // if any blackholes are alive, make the threads that wait on
379 if (traverseBlackholeQueue()) {
384 // must be last... invariant is that everything is fully
385 // scavenged at this point.
386 if (traverseWeakPtrList()) { // returns rtsTrue if evaced something
391 // If we get to here, there's really nothing left to do.
395 shutdown_gc_threads(n_gc_threads, gct->thread_index);
397 // Update pointers from the Task list
400 // Now see which stable names are still alive.
404 // We call processHeapClosureForDead() on every closure destroyed during
405 // the current garbage collection, so we invoke LdvCensusForDead().
406 if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_LDV
407 || RtsFlags.ProfFlags.bioSelector != NULL)
411 // NO MORE EVACUATION AFTER THIS POINT!
413 // Two-space collector: free the old to-space.
414 // g0s0->old_blocks is the old nursery
415 // g0s0->blocks is to-space from the previous GC
416 if (RtsFlags.GcFlags.generations == 1) {
417 if (g0s0->blocks != NULL) {
418 freeChain(g0s0->blocks);
423 // For each workspace, in each thread, move the copied blocks to the step
429 for (t = 0; t < n_gc_threads; t++) {
433 if (RtsFlags.GcFlags.generations == 1) {
438 for (; s < total_steps; s++) {
441 // Push the final block
443 push_scanned_block(ws->todo_bd, ws);
446 ASSERT(gct->scan_bd == NULL);
447 ASSERT(countBlocks(ws->scavd_list) == ws->n_scavd_blocks);
450 for (bd = ws->scavd_list; bd != NULL; bd = bd->link) {
451 ws->step->n_words += bd->free - bd->start;
455 prev->link = ws->step->blocks;
456 ws->step->blocks = ws->scavd_list;
458 ws->step->n_blocks += ws->n_scavd_blocks;
462 // Add all the partial blocks *after* we've added all the full
463 // blocks. This is so that we can grab the partial blocks back
464 // again and try to fill them up in the next GC.
465 for (t = 0; t < n_gc_threads; t++) {
469 if (RtsFlags.GcFlags.generations == 1) {
474 for (; s < total_steps; s++) {
478 for (bd = ws->part_list; bd != NULL; bd = next) {
480 if (bd->free == bd->start) {
482 ws->part_list = next;
489 ws->step->n_words += bd->free - bd->start;
494 prev->link = ws->step->blocks;
495 ws->step->blocks = ws->part_list;
497 ws->step->n_blocks += ws->n_part_blocks;
499 ASSERT(countBlocks(ws->step->blocks) == ws->step->n_blocks);
500 ASSERT(countOccupied(ws->step->blocks) == ws->step->n_words);
505 // Finally: compact or sweep the oldest generation.
506 if (major_gc && oldest_gen->steps[0].mark) {
507 if (oldest_gen->steps[0].compact)
508 compact(gct->scavenged_static_objects);
510 sweep(&oldest_gen->steps[0]);
513 /* run through all the generations/steps and tidy up
520 for (i=0; i < n_gc_threads; i++) {
521 if (n_gc_threads > 1) {
522 trace(TRACE_gc,"thread %d:", i);
523 trace(TRACE_gc," copied %ld", gc_threads[i]->copied * sizeof(W_));
524 trace(TRACE_gc," scanned %ld", gc_threads[i]->scanned * sizeof(W_));
525 trace(TRACE_gc," any_work %ld", gc_threads[i]->any_work);
526 trace(TRACE_gc," no_work %ld", gc_threads[i]->no_work);
527 trace(TRACE_gc," scav_find_work %ld", gc_threads[i]->scav_find_work);
529 copied += gc_threads[i]->copied;
530 max_copied = stg_max(gc_threads[i]->copied, max_copied);
532 if (n_gc_threads == 1) {
540 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
543 generations[g].collections++; // for stats
544 if (n_gc_threads > 1) generations[g].par_collections++;
547 // Count the mutable list as bytes "copied" for the purposes of
548 // stats. Every mutable list is copied during every GC.
550 nat mut_list_size = 0;
551 for (bd = generations[g].mut_list; bd != NULL; bd = bd->link) {
552 mut_list_size += bd->free - bd->start;
554 for (n = 0; n < n_capabilities; n++) {
555 for (bd = capabilities[n].mut_lists[g];
556 bd != NULL; bd = bd->link) {
557 mut_list_size += bd->free - bd->start;
560 copied += mut_list_size;
563 "mut_list_size: %lu (%d vars, %d arrays, %d MVARs, %d others)",
564 (unsigned long)(mut_list_size * sizeof(W_)),
565 mutlist_MUTVARS, mutlist_MUTARRS, mutlist_MVARS, mutlist_OTHERS);
568 for (s = 0; s < generations[g].n_steps; s++) {
570 stp = &generations[g].steps[s];
572 // for generations we collected...
575 /* free old memory and shift to-space into from-space for all
576 * the collected steps (except the allocation area). These
577 * freed blocks will probaby be quickly recycled.
579 if (!(g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1)) {
582 // tack the new blocks on the end of the existing blocks
583 if (stp->old_blocks != NULL) {
586 for (bd = stp->old_blocks; bd != NULL; bd = next) {
590 if (!(bd->flags & BF_MARKED))
593 stp->old_blocks = next;
602 stp->n_words += bd->free - bd->start;
604 // NB. this step might not be compacted next
605 // time, so reset the BF_MARKED flags.
606 // They are set before GC if we're going to
607 // compact. (search for BF_MARKED above).
608 bd->flags &= ~BF_MARKED;
610 // between GCs, all blocks in the heap except
611 // for the nursery have the BF_EVACUATED flag set.
612 bd->flags |= BF_EVACUATED;
619 prev->link = stp->blocks;
620 stp->blocks = stp->old_blocks;
623 // add the new blocks to the block tally
624 stp->n_blocks += stp->n_old_blocks;
625 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
626 ASSERT(countOccupied(stp->blocks) == stp->n_words);
630 freeChain(stp->old_blocks);
632 stp->old_blocks = NULL;
633 stp->n_old_blocks = 0;
636 /* LARGE OBJECTS. The current live large objects are chained on
637 * scavenged_large, having been moved during garbage
638 * collection from large_objects. Any objects left on
639 * large_objects list are therefore dead, so we free them here.
641 for (bd = stp->large_objects; bd != NULL; bd = next) {
647 stp->large_objects = stp->scavenged_large_objects;
648 stp->n_large_blocks = stp->n_scavenged_large_blocks;
651 else // for older generations...
653 /* For older generations, we need to append the
654 * scavenged_large_object list (i.e. large objects that have been
655 * promoted during this GC) to the large_object list for that step.
657 for (bd = stp->scavenged_large_objects; bd; bd = next) {
659 dbl_link_onto(bd, &stp->large_objects);
662 // add the new blocks we promoted during this GC
663 stp->n_large_blocks += stp->n_scavenged_large_blocks;
668 // update the max size of older generations after a major GC
669 resize_generations();
671 // Calculate the amount of live data for stats.
672 live = calcLiveWords();
674 // Free the small objects allocated via allocate(), since this will
675 // all have been copied into G0S1 now.
676 if (RtsFlags.GcFlags.generations > 1) {
677 if (g0s0->blocks != NULL) {
678 freeChain(g0s0->blocks);
685 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
687 // Start a new pinned_object_block
688 pinned_object_block = NULL;
690 // Free the mark stack.
691 if (mark_stack_bdescr != NULL) {
692 freeGroup(mark_stack_bdescr);
696 for (g = 0; g <= N; g++) {
697 for (s = 0; s < generations[g].n_steps; s++) {
698 stp = &generations[g].steps[s];
699 if (stp->bitmap != NULL) {
700 freeGroup(stp->bitmap);
708 // mark the garbage collected CAFs as dead
709 #if 0 && defined(DEBUG) // doesn't work at the moment
710 if (major_gc) { gcCAFs(); }
714 // resetStaticObjectForRetainerProfiling() must be called before
716 if (n_gc_threads > 1) {
717 barf("profiling is currently broken with multi-threaded GC");
718 // ToDo: fix the gct->scavenged_static_objects below
720 resetStaticObjectForRetainerProfiling(gct->scavenged_static_objects);
723 // zero the scavenged static object list
726 for (i = 0; i < n_gc_threads; i++) {
727 zero_static_object_list(gc_threads[i]->scavenged_static_objects);
734 // start any pending finalizers
736 scheduleFinalizers(last_free_capability, old_weak_ptr_list);
739 // send exceptions to any threads which were about to die
741 resurrectThreads(resurrected_threads);
742 performPendingThrowTos(exception_threads);
745 // Update the stable pointer hash table.
746 updateStablePtrTable(major_gc);
748 // check sanity after GC
749 IF_DEBUG(sanity, checkSanity());
751 // extra GC trace info
752 if (traceClass(TRACE_gc|DEBUG_gc)) statDescribeGens();
755 // symbol-table based profiling
756 /* heapCensus(to_blocks); */ /* ToDo */
759 // restore enclosing cost centre
765 // check for memory leaks if DEBUG is on
766 memInventory(traceClass(DEBUG_gc));
769 #ifdef RTS_GTK_FRONTPANEL
770 if (RtsFlags.GcFlags.frontpanel) {
771 updateFrontPanelAfterGC( N, live );
775 // ok, GC over: tell the stats department what happened.
776 slop = calcLiveBlocks() * BLOCK_SIZE_W - live;
777 stat_endGC(allocated, live, copied, N, max_copied, avg_copied, slop);
779 // Guess which generation we'll collect *next* time
780 initialise_N(force_major_gc);
782 #if defined(RTS_USER_SIGNALS)
783 if (RtsFlags.MiscFlags.install_signal_handlers) {
784 // unblock signals again
785 unblockUserSignals();
794 /* -----------------------------------------------------------------------------
795 Figure out which generation to collect, initialise N and major_gc.
797 Also returns the total number of blocks in generations that will be
799 -------------------------------------------------------------------------- */
802 initialise_N (rtsBool force_major_gc)
805 nat s, blocks, blocks_total;
810 if (force_major_gc) {
811 N = RtsFlags.GcFlags.generations - 1;
816 for (g = RtsFlags.GcFlags.generations - 1; g >= 0; g--) {
818 for (s = 0; s < generations[g].n_steps; s++) {
819 blocks += generations[g].steps[s].n_words / BLOCK_SIZE_W;
820 blocks += generations[g].steps[s].n_large_blocks;
822 if (blocks >= generations[g].max_blocks) {
826 blocks_total += blocks;
830 blocks_total += countNurseryBlocks();
832 major_gc = (N == RtsFlags.GcFlags.generations-1);
836 /* -----------------------------------------------------------------------------
837 Initialise the gc_thread structures.
838 -------------------------------------------------------------------------- */
840 #define GC_THREAD_INACTIVE 0
841 #define GC_THREAD_STANDING_BY 1
842 #define GC_THREAD_RUNNING 2
843 #define GC_THREAD_WAITING_TO_CONTINUE 3
846 alloc_gc_thread (int n)
852 t = stgMallocBytes(sizeof(gc_thread) + total_steps * sizeof(step_workspace),
857 initSpinLock(&t->gc_spin);
858 initSpinLock(&t->mut_spin);
859 ACQUIRE_SPIN_LOCK(&t->gc_spin);
860 t->wakeup = GC_THREAD_INACTIVE; // starts true, so we can wait for the
861 // thread to start up, see wakeup_gc_threads
865 t->free_blocks = NULL;
874 for (s = 0; s < total_steps; s++)
877 ws->step = &all_steps[s];
878 ASSERT(s == ws->step->abs_no);
882 ws->buffer_todo_bd = NULL;
884 ws->part_list = NULL;
885 ws->n_part_blocks = 0;
887 ws->scavd_list = NULL;
888 ws->n_scavd_blocks = 0;
898 if (gc_threads == NULL) {
899 #if defined(THREADED_RTS)
901 gc_threads = stgMallocBytes (RtsFlags.ParFlags.nNodes *
905 for (i = 0; i < RtsFlags.ParFlags.nNodes; i++) {
906 gc_threads[i] = alloc_gc_thread(i);
909 gc_threads = stgMallocBytes (sizeof(gc_thread*),
912 gc_threads[0] = alloc_gc_thread(0);
917 /* ----------------------------------------------------------------------------
919 ------------------------------------------------------------------------- */
921 static nat gc_running_threads;
923 #if defined(THREADED_RTS)
924 static Mutex gc_running_mutex;
931 ACQUIRE_LOCK(&gc_running_mutex);
932 n_running = ++gc_running_threads;
933 RELEASE_LOCK(&gc_running_mutex);
934 ASSERT(n_running <= n_gc_threads);
942 ACQUIRE_LOCK(&gc_running_mutex);
943 ASSERT(n_gc_threads != 0);
944 n_running = --gc_running_threads;
945 RELEASE_LOCK(&gc_running_mutex);
959 // scavenge objects in compacted generation
960 if (mark_stack_overflowed || oldgen_scan_bd != NULL ||
961 (mark_stack_bdescr != NULL && !mark_stack_empty())) {
965 // Check for global work in any step. We don't need to check for
966 // local work, because we have already exited scavenge_loop(),
967 // which means there is no local work for this thread.
968 for (s = total_steps-1; s >= 0; s--) {
969 if (s == 0 && RtsFlags.GcFlags.generations > 1) {
973 if (ws->todo_large_objects) return rtsTrue;
974 if (ws->step->todos) return rtsTrue;
983 scavenge_until_all_done (void)
987 debugTrace(DEBUG_gc, "GC thread %d working", gct->thread_index);
990 #if defined(THREADED_RTS)
991 if (n_gc_threads > 1) {
1000 // scavenge_loop() only exits when there's no work to do
1003 debugTrace(DEBUG_gc, "GC thread %d idle (%d still running)",
1004 gct->thread_index, r);
1006 while (gc_running_threads != 0) {
1012 // any_work() does not remove the work from the queue, it
1013 // just checks for the presence of work. If we find any,
1014 // then we increment gc_running_threads and go back to
1015 // scavenge_loop() to perform any pending work.
1018 // All threads are now stopped
1019 debugTrace(DEBUG_gc, "GC thread %d finished.", gct->thread_index);
1022 #if defined(THREADED_RTS)
1025 gcWorkerThread (Capability *cap)
1027 cap->in_gc = rtsTrue;
1029 gct = gc_threads[cap->no];
1030 gct->id = osThreadId();
1032 // Wait until we're told to wake up
1033 RELEASE_SPIN_LOCK(&gct->mut_spin);
1034 gct->wakeup = GC_THREAD_STANDING_BY;
1035 debugTrace(DEBUG_gc, "GC thread %d standing by...", gct->thread_index);
1036 ACQUIRE_SPIN_LOCK(&gct->gc_spin);
1039 // start performance counters in this thread...
1040 if (gct->papi_events == -1) {
1041 papi_init_eventset(&gct->papi_events);
1043 papi_thread_start_gc1_count(gct->papi_events);
1046 // Every thread evacuates some roots.
1048 markSomeCapabilities(mark_root, gct, gct->thread_index, n_gc_threads,
1049 rtsTrue/*prune sparks*/);
1050 scavenge_capability_mut_lists(&capabilities[gct->thread_index]);
1052 scavenge_until_all_done();
1055 // count events in this thread towards the GC totals
1056 papi_thread_stop_gc1_count(gct->papi_events);
1059 // Wait until we're told to continue
1060 RELEASE_SPIN_LOCK(&gct->gc_spin);
1061 gct->wakeup = GC_THREAD_WAITING_TO_CONTINUE;
1062 debugTrace(DEBUG_gc, "GC thread %d waiting to continue...",
1064 ACQUIRE_SPIN_LOCK(&gct->mut_spin);
1065 debugTrace(DEBUG_gc, "GC thread %d on my way...", gct->thread_index);
1071 waitForGcThreads (Capability *cap USED_IF_THREADS)
1073 #if defined(THREADED_RTS)
1074 nat n_threads = RtsFlags.ParFlags.nNodes;
1077 rtsBool retry = rtsTrue;
1080 for (i=0; i < n_threads; i++) {
1081 if (i == me) continue;
1082 if (gc_threads[i]->wakeup != GC_THREAD_STANDING_BY) {
1083 prodCapability(&capabilities[i], cap->running_task);
1086 for (j=0; j < 10000000; j++) {
1088 for (i=0; i < n_threads; i++) {
1089 if (i == me) continue;
1091 setContextSwitches();
1092 if (gc_threads[i]->wakeup != GC_THREAD_STANDING_BY) {
1103 start_gc_threads (void)
1105 #if defined(THREADED_RTS)
1106 gc_running_threads = 0;
1107 initMutex(&gc_running_mutex);
1112 wakeup_gc_threads (nat n_threads USED_IF_THREADS, nat me USED_IF_THREADS)
1114 #if defined(THREADED_RTS)
1116 for (i=0; i < n_threads; i++) {
1117 if (i == me) continue;
1119 debugTrace(DEBUG_gc, "waking up gc thread %d", i);
1120 if (gc_threads[i]->wakeup != GC_THREAD_STANDING_BY) barf("wakeup_gc_threads");
1122 gc_threads[i]->wakeup = GC_THREAD_RUNNING;
1123 ACQUIRE_SPIN_LOCK(&gc_threads[i]->mut_spin);
1124 RELEASE_SPIN_LOCK(&gc_threads[i]->gc_spin);
1129 // After GC is complete, we must wait for all GC threads to enter the
1130 // standby state, otherwise they may still be executing inside
1131 // any_work(), and may even remain awake until the next GC starts.
1133 shutdown_gc_threads (nat n_threads USED_IF_THREADS, nat me USED_IF_THREADS)
1135 #if defined(THREADED_RTS)
1137 for (i=0; i < n_threads; i++) {
1138 if (i == me) continue;
1139 while (gc_threads[i]->wakeup != GC_THREAD_WAITING_TO_CONTINUE) { write_barrier(); }
1145 releaseGCThreads (Capability *cap USED_IF_THREADS)
1147 #if defined(THREADED_RTS)
1148 nat n_threads = RtsFlags.ParFlags.nNodes;
1151 for (i=0; i < n_threads; i++) {
1152 if (i == me) continue;
1153 if (gc_threads[i]->wakeup != GC_THREAD_WAITING_TO_CONTINUE)
1154 barf("releaseGCThreads");
1156 gc_threads[i]->wakeup = GC_THREAD_INACTIVE;
1157 ACQUIRE_SPIN_LOCK(&gc_threads[i]->gc_spin);
1158 RELEASE_SPIN_LOCK(&gc_threads[i]->mut_spin);
1163 /* ----------------------------------------------------------------------------
1164 Initialise a generation that is to be collected
1165 ------------------------------------------------------------------------- */
1168 init_collected_gen (nat g, nat n_threads)
1175 // Throw away the current mutable list. Invariant: the mutable
1176 // list always has at least one block; this means we can avoid a
1177 // check for NULL in recordMutable().
1179 freeChain(generations[g].mut_list);
1180 generations[g].mut_list = allocBlock();
1181 for (i = 0; i < n_capabilities; i++) {
1182 freeChain(capabilities[i].mut_lists[g]);
1183 capabilities[i].mut_lists[g] = allocBlock();
1187 for (s = 0; s < generations[g].n_steps; s++) {
1189 stp = &generations[g].steps[s];
1190 ASSERT(stp->gen_no == g);
1192 // we'll construct a new list of threads in this step
1193 // during GC, throw away the current list.
1194 stp->old_threads = stp->threads;
1195 stp->threads = END_TSO_QUEUE;
1197 // generation 0, step 0 doesn't need to-space
1198 if (g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1) {
1202 // deprecate the existing blocks
1203 stp->old_blocks = stp->blocks;
1204 stp->n_old_blocks = stp->n_blocks;
1208 stp->live_estimate = 0;
1210 // we don't have any to-be-scavenged blocks yet
1212 stp->todos_last = NULL;
1215 // initialise the large object queues.
1216 stp->scavenged_large_objects = NULL;
1217 stp->n_scavenged_large_blocks = 0;
1219 // mark the small objects as from-space
1220 for (bd = stp->old_blocks; bd; bd = bd->link) {
1221 bd->flags &= ~BF_EVACUATED;
1224 // mark the large objects as from-space
1225 for (bd = stp->large_objects; bd; bd = bd->link) {
1226 bd->flags &= ~BF_EVACUATED;
1229 // for a compacted step, we need to allocate the bitmap
1231 nat bitmap_size; // in bytes
1232 bdescr *bitmap_bdescr;
1235 bitmap_size = stp->n_old_blocks * BLOCK_SIZE / (sizeof(W_)*BITS_PER_BYTE);
1237 if (bitmap_size > 0) {
1238 bitmap_bdescr = allocGroup((lnat)BLOCK_ROUND_UP(bitmap_size)
1240 stp->bitmap = bitmap_bdescr;
1241 bitmap = bitmap_bdescr->start;
1243 debugTrace(DEBUG_gc, "bitmap_size: %d, bitmap: %p",
1244 bitmap_size, bitmap);
1246 // don't forget to fill it with zeros!
1247 memset(bitmap, 0, bitmap_size);
1249 // For each block in this step, point to its bitmap from the
1250 // block descriptor.
1251 for (bd=stp->old_blocks; bd != NULL; bd = bd->link) {
1252 bd->u.bitmap = bitmap;
1253 bitmap += BLOCK_SIZE_W / (sizeof(W_)*BITS_PER_BYTE);
1255 // Also at this point we set the BF_MARKED flag
1256 // for this block. The invariant is that
1257 // BF_MARKED is always unset, except during GC
1258 // when it is set on those blocks which will be
1260 if (!(bd->flags & BF_FRAGMENTED)) {
1261 bd->flags |= BF_MARKED;
1268 // For each GC thread, for each step, allocate a "todo" block to
1269 // store evacuated objects to be scavenged, and a block to store
1270 // evacuated objects that do not need to be scavenged.
1271 for (t = 0; t < n_threads; t++) {
1272 for (s = 0; s < generations[g].n_steps; s++) {
1274 // we don't copy objects into g0s0, unless -G0
1275 if (g==0 && s==0 && RtsFlags.GcFlags.generations > 1) continue;
1277 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1279 ws->todo_large_objects = NULL;
1281 ws->part_list = NULL;
1282 ws->n_part_blocks = 0;
1284 // allocate the first to-space block; extra blocks will be
1285 // chained on as necessary.
1287 ws->buffer_todo_bd = NULL;
1288 alloc_todo_block(ws,0);
1290 ws->scavd_list = NULL;
1291 ws->n_scavd_blocks = 0;
1297 /* ----------------------------------------------------------------------------
1298 Initialise a generation that is *not* to be collected
1299 ------------------------------------------------------------------------- */
1302 init_uncollected_gen (nat g, nat threads)
1309 // save the current mutable lists for this generation, and
1310 // allocate a fresh block for each one. We'll traverse these
1311 // mutable lists as roots early on in the GC.
1312 generations[g].saved_mut_list = generations[g].mut_list;
1313 generations[g].mut_list = allocBlock();
1314 for (n = 0; n < n_capabilities; n++) {
1315 capabilities[n].saved_mut_lists[g] = capabilities[n].mut_lists[g];
1316 capabilities[n].mut_lists[g] = allocBlock();
1319 for (s = 0; s < generations[g].n_steps; s++) {
1320 stp = &generations[g].steps[s];
1321 stp->scavenged_large_objects = NULL;
1322 stp->n_scavenged_large_blocks = 0;
1325 for (s = 0; s < generations[g].n_steps; s++) {
1327 stp = &generations[g].steps[s];
1329 for (t = 0; t < threads; t++) {
1330 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1332 ws->buffer_todo_bd = NULL;
1333 ws->todo_large_objects = NULL;
1335 ws->part_list = NULL;
1336 ws->n_part_blocks = 0;
1338 ws->scavd_list = NULL;
1339 ws->n_scavd_blocks = 0;
1341 // If the block at the head of the list in this generation
1342 // is less than 3/4 full, then use it as a todo block.
1343 if (stp->blocks && isPartiallyFull(stp->blocks))
1345 ws->todo_bd = stp->blocks;
1346 ws->todo_free = ws->todo_bd->free;
1347 ws->todo_lim = ws->todo_bd->start + BLOCK_SIZE_W;
1348 stp->blocks = stp->blocks->link;
1350 stp->n_words -= ws->todo_bd->free - ws->todo_bd->start;
1351 ws->todo_bd->link = NULL;
1352 // we must scan from the current end point.
1353 ws->todo_bd->u.scan = ws->todo_bd->free;
1358 alloc_todo_block(ws,0);
1362 // deal out any more partial blocks to the threads' part_lists
1364 while (stp->blocks && isPartiallyFull(stp->blocks))
1367 stp->blocks = bd->link;
1368 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1369 bd->link = ws->part_list;
1371 ws->n_part_blocks += 1;
1372 bd->u.scan = bd->free;
1374 stp->n_words -= bd->free - bd->start;
1376 if (t == n_gc_threads) t = 0;
1381 /* -----------------------------------------------------------------------------
1382 Initialise a gc_thread before GC
1383 -------------------------------------------------------------------------- */
1386 init_gc_thread (gc_thread *t)
1388 t->static_objects = END_OF_STATIC_LIST;
1389 t->scavenged_static_objects = END_OF_STATIC_LIST;
1391 t->mut_lists = capabilities[t->thread_index].mut_lists;
1393 t->failed_to_evac = rtsFalse;
1394 t->eager_promotion = rtsTrue;
1395 t->thunk_selector_depth = 0;
1400 t->scav_find_work = 0;
1403 /* -----------------------------------------------------------------------------
1404 Function we pass to evacuate roots.
1405 -------------------------------------------------------------------------- */
1408 mark_root(void *user, StgClosure **root)
1410 // we stole a register for gct, but this function is called from
1411 // *outside* the GC where the register variable is not in effect,
1412 // so we need to save and restore it here. NB. only call
1413 // mark_root() from the main GC thread, otherwise gct will be
1415 gc_thread *saved_gct;
1424 /* -----------------------------------------------------------------------------
1425 Initialising the static object & mutable lists
1426 -------------------------------------------------------------------------- */
1429 zero_static_object_list(StgClosure* first_static)
1433 const StgInfoTable *info;
1435 for (p = first_static; p != END_OF_STATIC_LIST; p = link) {
1437 link = *STATIC_LINK(info, p);
1438 *STATIC_LINK(info,p) = NULL;
1442 /* ----------------------------------------------------------------------------
1443 Update the pointers from the task list
1445 These are treated as weak pointers because we want to allow a main
1446 thread to get a BlockedOnDeadMVar exception in the same way as any
1447 other thread. Note that the threads should all have been retained
1448 by GC by virtue of being on the all_threads list, we're just
1449 updating pointers here.
1450 ------------------------------------------------------------------------- */
1453 update_task_list (void)
1457 for (task = all_tasks; task != NULL; task = task->all_link) {
1458 if (!task->stopped && task->tso) {
1459 ASSERT(task->tso->bound == task);
1460 tso = (StgTSO *) isAlive((StgClosure *)task->tso);
1462 barf("task %p: main thread %d has been GC'd",
1475 /* ----------------------------------------------------------------------------
1476 Reset the sizes of the older generations when we do a major
1479 CURRENT STRATEGY: make all generations except zero the same size.
1480 We have to stay within the maximum heap size, and leave a certain
1481 percentage of the maximum heap size available to allocate into.
1482 ------------------------------------------------------------------------- */
1485 resize_generations (void)
1489 if (major_gc && RtsFlags.GcFlags.generations > 1) {
1490 nat live, size, min_alloc, words;
1491 nat max = RtsFlags.GcFlags.maxHeapSize;
1492 nat gens = RtsFlags.GcFlags.generations;
1494 // live in the oldest generations
1495 if (oldest_gen->steps[0].live_estimate != 0) {
1496 words = oldest_gen->steps[0].live_estimate;
1498 words = oldest_gen->steps[0].n_words;
1500 live = (words + BLOCK_SIZE_W - 1) / BLOCK_SIZE_W +
1501 oldest_gen->steps[0].n_large_blocks;
1503 // default max size for all generations except zero
1504 size = stg_max(live * RtsFlags.GcFlags.oldGenFactor,
1505 RtsFlags.GcFlags.minOldGenSize);
1507 // minimum size for generation zero
1508 min_alloc = stg_max((RtsFlags.GcFlags.pcFreeHeap * max) / 200,
1509 RtsFlags.GcFlags.minAllocAreaSize);
1511 // Auto-enable compaction when the residency reaches a
1512 // certain percentage of the maximum heap size (default: 30%).
1513 if (RtsFlags.GcFlags.generations > 1 &&
1514 (RtsFlags.GcFlags.compact ||
1516 oldest_gen->steps[0].n_blocks >
1517 (RtsFlags.GcFlags.compactThreshold * max) / 100))) {
1518 oldest_gen->steps[0].mark = 1;
1519 oldest_gen->steps[0].compact = 1;
1520 // debugBelch("compaction: on\n", live);
1522 oldest_gen->steps[0].mark = 0;
1523 oldest_gen->steps[0].compact = 0;
1524 // debugBelch("compaction: off\n", live);
1527 if (RtsFlags.GcFlags.sweep) {
1528 oldest_gen->steps[0].mark = 1;
1531 // if we're going to go over the maximum heap size, reduce the
1532 // size of the generations accordingly. The calculation is
1533 // different if compaction is turned on, because we don't need
1534 // to double the space required to collect the old generation.
1537 // this test is necessary to ensure that the calculations
1538 // below don't have any negative results - we're working
1539 // with unsigned values here.
1540 if (max < min_alloc) {
1544 if (oldest_gen->steps[0].compact) {
1545 if ( (size + (size - 1) * (gens - 2) * 2) + min_alloc > max ) {
1546 size = (max - min_alloc) / ((gens - 1) * 2 - 1);
1549 if ( (size * (gens - 1) * 2) + min_alloc > max ) {
1550 size = (max - min_alloc) / ((gens - 1) * 2);
1560 debugBelch("live: %d, min_alloc: %d, size : %d, max = %d\n", live,
1561 min_alloc, size, max);
1564 for (g = 0; g < gens; g++) {
1565 generations[g].max_blocks = size;
1570 /* -----------------------------------------------------------------------------
1571 Calculate the new size of the nursery, and resize it.
1572 -------------------------------------------------------------------------- */
1575 resize_nursery (void)
1577 if (RtsFlags.GcFlags.generations == 1)
1578 { // Two-space collector:
1581 /* set up a new nursery. Allocate a nursery size based on a
1582 * function of the amount of live data (by default a factor of 2)
1583 * Use the blocks from the old nursery if possible, freeing up any
1586 * If we get near the maximum heap size, then adjust our nursery
1587 * size accordingly. If the nursery is the same size as the live
1588 * data (L), then we need 3L bytes. We can reduce the size of the
1589 * nursery to bring the required memory down near 2L bytes.
1591 * A normal 2-space collector would need 4L bytes to give the same
1592 * performance we get from 3L bytes, reducing to the same
1593 * performance at 2L bytes.
1595 blocks = g0s0->n_blocks;
1597 if ( RtsFlags.GcFlags.maxHeapSize != 0 &&
1598 blocks * RtsFlags.GcFlags.oldGenFactor * 2 >
1599 RtsFlags.GcFlags.maxHeapSize )
1601 long adjusted_blocks; // signed on purpose
1604 adjusted_blocks = (RtsFlags.GcFlags.maxHeapSize - 2 * blocks);
1606 debugTrace(DEBUG_gc, "near maximum heap size of 0x%x blocks, blocks = %d, adjusted to %ld",
1607 RtsFlags.GcFlags.maxHeapSize, blocks, adjusted_blocks);
1609 pc_free = adjusted_blocks * 100 / RtsFlags.GcFlags.maxHeapSize;
1610 if (pc_free < RtsFlags.GcFlags.pcFreeHeap) /* might even * be < 0 */
1614 blocks = adjusted_blocks;
1618 blocks *= RtsFlags.GcFlags.oldGenFactor;
1619 if (blocks < RtsFlags.GcFlags.minAllocAreaSize)
1621 blocks = RtsFlags.GcFlags.minAllocAreaSize;
1624 resizeNurseries(blocks);
1626 else // Generational collector
1629 * If the user has given us a suggested heap size, adjust our
1630 * allocation area to make best use of the memory available.
1632 if (RtsFlags.GcFlags.heapSizeSuggestion)
1635 nat needed = calcNeeded(); // approx blocks needed at next GC
1637 /* Guess how much will be live in generation 0 step 0 next time.
1638 * A good approximation is obtained by finding the
1639 * percentage of g0s0 that was live at the last minor GC.
1641 * We have an accurate figure for the amount of copied data in
1642 * 'copied', but we must convert this to a number of blocks, with
1643 * a small adjustment for estimated slop at the end of a block
1648 g0s0_pcnt_kept = ((copied / (BLOCK_SIZE_W - 10)) * 100)
1649 / countNurseryBlocks();
1652 /* Estimate a size for the allocation area based on the
1653 * information available. We might end up going slightly under
1654 * or over the suggested heap size, but we should be pretty
1657 * Formula: suggested - needed
1658 * ----------------------------
1659 * 1 + g0s0_pcnt_kept/100
1661 * where 'needed' is the amount of memory needed at the next
1662 * collection for collecting all steps except g0s0.
1665 (((long)RtsFlags.GcFlags.heapSizeSuggestion - (long)needed) * 100) /
1666 (100 + (long)g0s0_pcnt_kept);
1668 if (blocks < (long)RtsFlags.GcFlags.minAllocAreaSize) {
1669 blocks = RtsFlags.GcFlags.minAllocAreaSize;
1672 resizeNurseries((nat)blocks);
1676 // we might have added extra large blocks to the nursery, so
1677 // resize back to minAllocAreaSize again.
1678 resizeNurseriesFixed(RtsFlags.GcFlags.minAllocAreaSize);
1683 /* -----------------------------------------------------------------------------
1684 Sanity code for CAF garbage collection.
1686 With DEBUG turned on, we manage a CAF list in addition to the SRT
1687 mechanism. After GC, we run down the CAF list and blackhole any
1688 CAFs which have been garbage collected. This means we get an error
1689 whenever the program tries to enter a garbage collected CAF.
1691 Any garbage collected CAFs are taken off the CAF list at the same
1693 -------------------------------------------------------------------------- */
1695 #if 0 && defined(DEBUG)
1702 const StgInfoTable *info;
1713 ASSERT(info->type == IND_STATIC);
1715 if (STATIC_LINK(info,p) == NULL) {
1716 debugTrace(DEBUG_gccafs, "CAF gc'd at 0x%04lx", (long)p);
1718 SET_INFO(p,&stg_BLACKHOLE_info);
1719 p = STATIC_LINK2(info,p);
1723 pp = &STATIC_LINK2(info,p);
1730 debugTrace(DEBUG_gccafs, "%d CAFs live", i);