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
129 SpinLock recordMutableGen_sync;
132 /* -----------------------------------------------------------------------------
133 Static function declarations
134 -------------------------------------------------------------------------- */
136 static void mark_root (void *user, StgClosure **root);
137 static void zero_static_object_list (StgClosure* first_static);
138 static nat initialise_N (rtsBool force_major_gc);
139 static void alloc_gc_threads (void);
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 nat inc_running (void);
149 static nat dec_running (void);
150 static void wakeup_gc_threads (nat n_threads);
151 static void shutdown_gc_threads (nat n_threads);
153 #if 0 && defined(DEBUG)
154 static void gcCAFs (void);
157 /* -----------------------------------------------------------------------------
158 The mark bitmap & stack.
159 -------------------------------------------------------------------------- */
161 #define MARK_STACK_BLOCKS 4
163 bdescr *mark_stack_bdescr;
168 // Flag and pointers used for falling back to a linear scan when the
169 // mark stack overflows.
170 rtsBool mark_stack_overflowed;
171 bdescr *oldgen_scan_bd;
174 /* -----------------------------------------------------------------------------
175 GarbageCollect: the main entry point to the garbage collector.
177 Locks held: all capabilities are held throughout GarbageCollect().
178 -------------------------------------------------------------------------- */
181 GarbageCollect ( rtsBool force_major_gc )
185 lnat live, allocated, max_copied, avg_copied, slop;
186 lnat oldgen_saved_blocks = 0;
187 gc_thread *saved_gct;
190 // necessary if we stole a callee-saves register for gct:
194 CostCentreStack *prev_CCS;
199 #if defined(RTS_USER_SIGNALS)
200 if (RtsFlags.MiscFlags.install_signal_handlers) {
206 ASSERT(sizeof(step_workspace) == 16 * sizeof(StgWord));
207 // otherwise adjust the padding in step_workspace.
209 // tell the stats department that we've started a GC
212 // tell the STM to discard any cached closures it's hoping to re-use
221 // attribute any costs to CCS_GC
227 /* Approximate how much we allocated.
228 * Todo: only when generating stats?
230 allocated = calcAllocated();
232 /* Figure out which generation to collect
234 n = initialise_N(force_major_gc);
236 /* Allocate + initialise the gc_thread structures.
240 /* Start threads, so they can be spinning up while we finish initialisation.
244 /* How many threads will be participating in this GC?
245 * We don't try to parallelise minor GC.
247 #if defined(THREADED_RTS)
248 if (n < (4*1024*1024 / BLOCK_SIZE)) {
251 n_gc_threads = RtsFlags.ParFlags.gcThreads;
256 trace(TRACE_gc|DEBUG_gc, "GC (gen %d): %d KB to collect, %ld MB in use, using %d thread(s)",
257 N, n * (BLOCK_SIZE / 1024), mblocks_allocated, n_gc_threads);
259 #ifdef RTS_GTK_FRONTPANEL
260 if (RtsFlags.GcFlags.frontpanel) {
261 updateFrontPanelBeforeGC(N);
266 // check for memory leaks if DEBUG is on
267 memInventory(traceClass(DEBUG_gc));
270 // check stack sanity *before* GC (ToDo: check all threads)
271 IF_DEBUG(sanity, checkFreeListSanity());
273 // Initialise all our gc_thread structures
274 for (t = 0; t < n_gc_threads; t++) {
275 init_gc_thread(gc_threads[t]);
278 // Initialise all the generations/steps that we're collecting.
279 for (g = 0; g <= N; g++) {
280 init_collected_gen(g,n_gc_threads);
283 // Initialise all the generations/steps that we're *not* collecting.
284 for (g = N+1; g < RtsFlags.GcFlags.generations; g++) {
285 init_uncollected_gen(g,n_gc_threads);
288 /* Allocate a mark stack if we're doing a major collection.
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
302 /* -----------------------------------------------------------------------
303 * follow all the roots that we know about:
304 * - mutable lists from each generation > N
305 * we want to *scavenge* these roots, not evacuate them: they're not
306 * going to move in this GC.
307 * Also do them in reverse generation order, for the usual reason:
308 * namely to reduce the likelihood of spurious old->new pointers.
310 for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
311 generations[g].saved_mut_list = generations[g].mut_list;
312 generations[g].mut_list = allocBlock();
313 // mut_list always has at least one block.
316 // the main thread is running: this prevents any other threads from
317 // exiting prematurely, so we can start them now.
318 // NB. do this after the mutable lists have been saved above, otherwise
319 // the other GC threads will be writing into the old mutable lists.
321 wakeup_gc_threads(n_gc_threads);
323 for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
324 scavenge_mutable_list(&generations[g]);
327 // follow roots from the CAF list (used by GHCi)
329 markCAFs(mark_root, gct);
331 // follow all the roots that the application knows about.
333 markSomeCapabilities(mark_root, gct, gct->thread_index, n_gc_threads);
335 #if defined(RTS_USER_SIGNALS)
336 // mark the signal handlers (signals should be already blocked)
337 markSignalHandlers(mark_root, gct);
340 // Mark the weak pointer list, and prepare to detect dead weak pointers.
344 // Mark the stable pointer table.
345 markStablePtrTable(mark_root, gct);
347 /* -------------------------------------------------------------------------
348 * Repeatedly scavenge all the areas we know about until there's no
349 * more scavenging to be done.
353 scavenge_until_all_done();
354 // The other threads are now stopped. We might recurse back to
355 // here, but from now on this is the only thread.
357 // if any blackholes are alive, make the threads that wait on
359 if (traverseBlackholeQueue()) {
364 // must be last... invariant is that everything is fully
365 // scavenged at this point.
366 if (traverseWeakPtrList()) { // returns rtsTrue if evaced something
371 // If we get to here, there's really nothing left to do.
375 shutdown_gc_threads(n_gc_threads);
377 // Update pointers from the Task list
380 // Now see which stable names are still alive.
384 // We call processHeapClosureForDead() on every closure destroyed during
385 // the current garbage collection, so we invoke LdvCensusForDead().
386 if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_LDV
387 || RtsFlags.ProfFlags.bioSelector != NULL)
391 // NO MORE EVACUATION AFTER THIS POINT!
392 // Finally: compaction of the oldest generation.
393 if (major_gc && oldest_gen->steps[0].is_compacted) {
394 // save number of blocks for stats
395 oldgen_saved_blocks = oldest_gen->steps[0].n_old_blocks;
396 compact(gct->scavenged_static_objects);
399 IF_DEBUG(sanity, checkGlobalTSOList(rtsFalse));
401 // Two-space collector: free the old to-space.
402 // g0s0->old_blocks is the old nursery
403 // g0s0->blocks is to-space from the previous GC
404 if (RtsFlags.GcFlags.generations == 1) {
405 if (g0s0->blocks != NULL) {
406 freeChain(g0s0->blocks);
411 // For each workspace, in each thread, move the copied blocks to the step
417 for (t = 0; t < n_gc_threads; t++) {
421 if (RtsFlags.GcFlags.generations == 1) {
426 for (; s < total_steps; s++) {
429 // Push the final block
431 push_scanned_block(ws->todo_bd, ws);
434 ASSERT(gct->scan_bd == NULL);
435 ASSERT(countBlocks(ws->scavd_list) == ws->n_scavd_blocks);
438 for (bd = ws->scavd_list; bd != NULL; bd = bd->link) {
439 ws->step->n_words += bd->free - bd->start;
443 prev->link = ws->step->blocks;
444 ws->step->blocks = ws->scavd_list;
446 ws->step->n_blocks += ws->n_scavd_blocks;
450 // Add all the partial blocks *after* we've added all the full
451 // blocks. This is so that we can grab the partial blocks back
452 // again and try to fill them up in the next GC.
453 for (t = 0; t < n_gc_threads; t++) {
457 if (RtsFlags.GcFlags.generations == 1) {
462 for (; s < total_steps; s++) {
466 for (bd = ws->part_list; bd != NULL; bd = next) {
468 if (bd->free == bd->start) {
470 ws->part_list = next;
477 ws->step->n_words += bd->free - bd->start;
482 prev->link = ws->step->blocks;
483 ws->step->blocks = ws->part_list;
485 ws->step->n_blocks += ws->n_part_blocks;
487 ASSERT(countBlocks(ws->step->blocks) == ws->step->n_blocks);
488 ASSERT(countOccupied(ws->step->blocks) == ws->step->n_words);
493 /* run through all the generations/steps and tidy up
500 for (i=0; i < n_gc_threads; i++) {
501 if (n_gc_threads > 1) {
502 trace(TRACE_gc,"thread %d:", i);
503 trace(TRACE_gc," copied %ld", gc_threads[i]->copied * sizeof(W_));
504 trace(TRACE_gc," scanned %ld", gc_threads[i]->scanned * sizeof(W_));
505 trace(TRACE_gc," any_work %ld", gc_threads[i]->any_work);
506 trace(TRACE_gc," no_work %ld", gc_threads[i]->no_work);
507 trace(TRACE_gc," scav_find_work %ld", gc_threads[i]->scav_find_work);
509 copied += gc_threads[i]->copied;
510 max_copied = stg_max(gc_threads[i]->copied, max_copied);
512 if (n_gc_threads == 1) {
520 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
523 generations[g].collections++; // for stats
524 if (n_gc_threads > 1) generations[g].par_collections++;
527 // Count the mutable list as bytes "copied" for the purposes of
528 // stats. Every mutable list is copied during every GC.
530 nat mut_list_size = 0;
531 for (bd = generations[g].mut_list; bd != NULL; bd = bd->link) {
532 mut_list_size += bd->free - bd->start;
534 copied += mut_list_size;
537 "mut_list_size: %lu (%d vars, %d arrays, %d MVARs, %d others)",
538 (unsigned long)(mut_list_size * sizeof(W_)),
539 mutlist_MUTVARS, mutlist_MUTARRS, mutlist_MVARS, mutlist_OTHERS);
542 for (s = 0; s < generations[g].n_steps; s++) {
544 stp = &generations[g].steps[s];
546 // for generations we collected...
549 /* free old memory and shift to-space into from-space for all
550 * the collected steps (except the allocation area). These
551 * freed blocks will probaby be quickly recycled.
553 if (!(g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1)) {
554 if (stp->is_compacted)
556 // for a compacted step, just shift the new to-space
557 // onto the front of the now-compacted existing blocks.
558 for (bd = stp->blocks; bd != NULL; bd = bd->link) {
559 stp->n_words += bd->free - bd->start;
561 // tack the new blocks on the end of the existing blocks
562 if (stp->old_blocks != NULL) {
563 for (bd = stp->old_blocks; bd != NULL; bd = next) {
564 // NB. this step might not be compacted next
565 // time, so reset the BF_COMPACTED flags.
566 // They are set before GC if we're going to
567 // compact. (search for BF_COMPACTED above).
568 bd->flags &= ~BF_COMPACTED;
571 bd->link = stp->blocks;
574 stp->blocks = stp->old_blocks;
576 // add the new blocks to the block tally
577 stp->n_blocks += stp->n_old_blocks;
578 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
579 ASSERT(countOccupied(stp->blocks) == stp->n_words);
583 freeChain(stp->old_blocks);
585 stp->old_blocks = NULL;
586 stp->n_old_blocks = 0;
589 /* LARGE OBJECTS. The current live large objects are chained on
590 * scavenged_large, having been moved during garbage
591 * collection from large_objects. Any objects left on
592 * large_objects list are therefore dead, so we free them here.
594 for (bd = stp->large_objects; bd != NULL; bd = next) {
600 stp->large_objects = stp->scavenged_large_objects;
601 stp->n_large_blocks = stp->n_scavenged_large_blocks;
604 else // for older generations...
606 /* For older generations, we need to append the
607 * scavenged_large_object list (i.e. large objects that have been
608 * promoted during this GC) to the large_object list for that step.
610 for (bd = stp->scavenged_large_objects; bd; bd = next) {
612 dbl_link_onto(bd, &stp->large_objects);
615 // add the new blocks we promoted during this GC
616 stp->n_large_blocks += stp->n_scavenged_large_blocks;
621 // update the max size of older generations after a major GC
622 resize_generations();
624 // Calculate the amount of live data for stats.
625 live = calcLiveWords();
627 // Free the small objects allocated via allocate(), since this will
628 // all have been copied into G0S1 now.
629 if (RtsFlags.GcFlags.generations > 1) {
630 if (g0s0->blocks != NULL) {
631 freeChain(g0s0->blocks);
638 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
640 // Start a new pinned_object_block
641 pinned_object_block = NULL;
643 // Free the mark stack.
644 if (mark_stack_bdescr != NULL) {
645 freeGroup(mark_stack_bdescr);
649 for (g = 0; g <= N; g++) {
650 for (s = 0; s < generations[g].n_steps; s++) {
651 stp = &generations[g].steps[s];
652 if (stp->bitmap != NULL) {
653 freeGroup(stp->bitmap);
661 // mark the garbage collected CAFs as dead
662 #if 0 && defined(DEBUG) // doesn't work at the moment
663 if (major_gc) { gcCAFs(); }
667 // resetStaticObjectForRetainerProfiling() must be called before
669 if (n_gc_threads > 1) {
670 barf("profiling is currently broken with multi-threaded GC");
671 // ToDo: fix the gct->scavenged_static_objects below
673 resetStaticObjectForRetainerProfiling(gct->scavenged_static_objects);
676 // zero the scavenged static object list
679 for (i = 0; i < n_gc_threads; i++) {
680 zero_static_object_list(gc_threads[i]->scavenged_static_objects);
687 // start any pending finalizers
689 scheduleFinalizers(last_free_capability, old_weak_ptr_list);
692 // send exceptions to any threads which were about to die
694 resurrectThreads(resurrected_threads);
695 performPendingThrowTos(exception_threads);
698 // Update the stable pointer hash table.
699 updateStablePtrTable(major_gc);
701 // check sanity after GC
702 IF_DEBUG(sanity, checkSanity());
704 // extra GC trace info
705 if (traceClass(TRACE_gc|DEBUG_gc)) statDescribeGens();
708 // symbol-table based profiling
709 /* heapCensus(to_blocks); */ /* ToDo */
712 // restore enclosing cost centre
718 // check for memory leaks if DEBUG is on
719 memInventory(traceClass(DEBUG_gc));
722 #ifdef RTS_GTK_FRONTPANEL
723 if (RtsFlags.GcFlags.frontpanel) {
724 updateFrontPanelAfterGC( N, live );
728 // ok, GC over: tell the stats department what happened.
729 slop = calcLiveBlocks() * BLOCK_SIZE_W - live;
730 stat_endGC(allocated, live, copied, N, max_copied, avg_copied, slop);
732 #if defined(RTS_USER_SIGNALS)
733 if (RtsFlags.MiscFlags.install_signal_handlers) {
734 // unblock signals again
735 unblockUserSignals();
744 /* -----------------------------------------------------------------------------
745 Figure out which generation to collect, initialise N and major_gc.
747 Also returns the total number of blocks in generations that will be
749 -------------------------------------------------------------------------- */
752 initialise_N (rtsBool force_major_gc)
755 nat s, blocks, blocks_total;
760 if (force_major_gc) {
761 N = RtsFlags.GcFlags.generations - 1;
766 for (g = RtsFlags.GcFlags.generations - 1; g >= 0; g--) {
768 for (s = 0; s < generations[g].n_steps; s++) {
769 blocks += generations[g].steps[s].n_words / BLOCK_SIZE_W;
770 blocks += generations[g].steps[s].n_large_blocks;
772 if (blocks >= generations[g].max_blocks) {
776 blocks_total += blocks;
780 blocks_total += countNurseryBlocks();
782 major_gc = (N == RtsFlags.GcFlags.generations-1);
786 /* -----------------------------------------------------------------------------
787 Initialise the gc_thread structures.
788 -------------------------------------------------------------------------- */
791 alloc_gc_thread (int n)
797 t = stgMallocBytes(sizeof(gc_thread) + total_steps * sizeof(step_workspace),
802 initCondition(&t->wake_cond);
803 initMutex(&t->wake_mutex);
804 t->wakeup = rtsTrue; // starts true, so we can wait for the
805 // thread to start up, see wakeup_gc_threads
810 t->free_blocks = NULL;
819 for (s = 0; s < total_steps; s++)
822 ws->step = &all_steps[s];
823 ASSERT(s == ws->step->abs_no);
827 ws->buffer_todo_bd = NULL;
829 ws->part_list = NULL;
830 ws->n_part_blocks = 0;
832 ws->scavd_list = NULL;
833 ws->n_scavd_blocks = 0;
841 alloc_gc_threads (void)
843 if (gc_threads == NULL) {
844 #if defined(THREADED_RTS)
846 gc_threads = stgMallocBytes (RtsFlags.ParFlags.gcThreads *
850 for (i = 0; i < RtsFlags.ParFlags.gcThreads; i++) {
851 gc_threads[i] = alloc_gc_thread(i);
854 gc_threads = stgMallocBytes (sizeof(gc_thread*),
857 gc_threads[0] = alloc_gc_thread(0);
862 /* ----------------------------------------------------------------------------
864 ------------------------------------------------------------------------- */
866 static nat gc_running_threads;
868 #if defined(THREADED_RTS)
869 static Mutex gc_running_mutex;
876 ACQUIRE_LOCK(&gc_running_mutex);
877 n_running = ++gc_running_threads;
878 RELEASE_LOCK(&gc_running_mutex);
879 ASSERT(n_running <= n_gc_threads);
887 ACQUIRE_LOCK(&gc_running_mutex);
888 ASSERT(n_gc_threads != 0);
889 n_running = --gc_running_threads;
890 RELEASE_LOCK(&gc_running_mutex);
895 scavenge_until_all_done (void)
899 debugTrace(DEBUG_gc, "GC thread %d working", gct->thread_index);
903 // scavenge_loop() only exits when there's no work to do
906 debugTrace(DEBUG_gc, "GC thread %d idle (%d still running)",
907 gct->thread_index, r);
909 while (gc_running_threads != 0) {
915 // any_work() does not remove the work from the queue, it
916 // just checks for the presence of work. If we find any,
917 // then we increment gc_running_threads and go back to
918 // scavenge_loop() to perform any pending work.
921 // All threads are now stopped
922 debugTrace(DEBUG_gc, "GC thread %d finished.", gct->thread_index);
925 #if defined(THREADED_RTS)
927 // gc_thread_work(): Scavenge until there's no work left to do and all
928 // the running threads are idle.
931 gc_thread_work (void)
933 // gc_running_threads has already been incremented for us; this is
934 // a worker thread and the main thread bumped gc_running_threads
935 // before waking us up.
937 // Every thread evacuates some roots.
939 markSomeCapabilities(mark_root, gct, gct->thread_index, n_gc_threads);
941 scavenge_until_all_done();
946 gc_thread_mainloop (void)
950 // Wait until we're told to wake up
951 ACQUIRE_LOCK(&gct->wake_mutex);
952 gct->wakeup = rtsFalse;
953 while (!gct->wakeup) {
954 debugTrace(DEBUG_gc, "GC thread %d standing by...",
956 waitCondition(&gct->wake_cond, &gct->wake_mutex);
958 RELEASE_LOCK(&gct->wake_mutex);
959 if (gct->exit) break;
962 // start performance counters in this thread...
963 if (gct->papi_events == -1) {
964 papi_init_eventset(&gct->papi_events);
966 papi_thread_start_gc1_count(gct->papi_events);
972 // count events in this thread towards the GC totals
973 papi_thread_stop_gc1_count(gct->papi_events);
979 #if defined(THREADED_RTS)
981 gc_thread_entry (gc_thread *my_gct)
984 debugTrace(DEBUG_gc, "GC thread %d starting...", gct->thread_index);
985 gct->id = osThreadId();
986 gc_thread_mainloop();
991 start_gc_threads (void)
993 #if defined(THREADED_RTS)
996 static rtsBool done = rtsFalse;
998 gc_running_threads = 0;
999 initMutex(&gc_running_mutex);
1002 // Start from 1: the main thread is 0
1003 for (i = 1; i < RtsFlags.ParFlags.gcThreads; i++) {
1004 createOSThread(&id, (OSThreadProc*)&gc_thread_entry,
1013 wakeup_gc_threads (nat n_threads USED_IF_THREADS)
1015 #if defined(THREADED_RTS)
1017 for (i=1; i < n_threads; i++) {
1019 debugTrace(DEBUG_gc, "waking up gc thread %d", i);
1021 ACQUIRE_LOCK(&gc_threads[i]->wake_mutex);
1022 if (gc_threads[i]->wakeup) {
1023 RELEASE_LOCK(&gc_threads[i]->wake_mutex);
1029 gc_threads[i]->wakeup = rtsTrue;
1030 signalCondition(&gc_threads[i]->wake_cond);
1031 RELEASE_LOCK(&gc_threads[i]->wake_mutex);
1036 // After GC is complete, we must wait for all GC threads to enter the
1037 // standby state, otherwise they may still be executing inside
1038 // any_work(), and may even remain awake until the next GC starts.
1040 shutdown_gc_threads (nat n_threads USED_IF_THREADS)
1042 #if defined(THREADED_RTS)
1045 for (i=1; i < n_threads; i++) {
1047 ACQUIRE_LOCK(&gc_threads[i]->wake_mutex);
1048 wakeup = gc_threads[i]->wakeup;
1049 // wakeup is false while the thread is waiting
1050 RELEASE_LOCK(&gc_threads[i]->wake_mutex);
1056 /* ----------------------------------------------------------------------------
1057 Initialise a generation that is to be collected
1058 ------------------------------------------------------------------------- */
1061 init_collected_gen (nat g, nat n_threads)
1068 // Throw away the current mutable list. Invariant: the mutable
1069 // list always has at least one block; this means we can avoid a
1070 // check for NULL in recordMutable().
1072 freeChain(generations[g].mut_list);
1073 generations[g].mut_list = allocBlock();
1074 for (i = 0; i < n_capabilities; i++) {
1075 freeChain(capabilities[i].mut_lists[g]);
1076 capabilities[i].mut_lists[g] = allocBlock();
1080 for (s = 0; s < generations[g].n_steps; s++) {
1082 stp = &generations[g].steps[s];
1083 ASSERT(stp->gen_no == g);
1085 // we'll construct a new list of threads in this step
1086 // during GC, throw away the current list.
1087 stp->old_threads = stp->threads;
1088 stp->threads = END_TSO_QUEUE;
1090 // generation 0, step 0 doesn't need to-space
1091 if (g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1) {
1095 // deprecate the existing blocks
1096 stp->old_blocks = stp->blocks;
1097 stp->n_old_blocks = stp->n_blocks;
1102 // we don't have any to-be-scavenged blocks yet
1104 stp->todos_last = NULL;
1107 // initialise the large object queues.
1108 stp->scavenged_large_objects = NULL;
1109 stp->n_scavenged_large_blocks = 0;
1111 // mark the small objects as from-space
1112 for (bd = stp->old_blocks; bd; bd = bd->link) {
1113 bd->flags &= ~BF_EVACUATED;
1116 // mark the large objects as from-space
1117 for (bd = stp->large_objects; bd; bd = bd->link) {
1118 bd->flags &= ~BF_EVACUATED;
1121 // for a compacted step, we need to allocate the bitmap
1122 if (stp->is_compacted) {
1123 nat bitmap_size; // in bytes
1124 bdescr *bitmap_bdescr;
1127 bitmap_size = stp->n_old_blocks * BLOCK_SIZE / (sizeof(W_)*BITS_PER_BYTE);
1129 if (bitmap_size > 0) {
1130 bitmap_bdescr = allocGroup((lnat)BLOCK_ROUND_UP(bitmap_size)
1132 stp->bitmap = bitmap_bdescr;
1133 bitmap = bitmap_bdescr->start;
1135 debugTrace(DEBUG_gc, "bitmap_size: %d, bitmap: %p",
1136 bitmap_size, bitmap);
1138 // don't forget to fill it with zeros!
1139 memset(bitmap, 0, bitmap_size);
1141 // For each block in this step, point to its bitmap from the
1142 // block descriptor.
1143 for (bd=stp->old_blocks; bd != NULL; bd = bd->link) {
1144 bd->u.bitmap = bitmap;
1145 bitmap += BLOCK_SIZE_W / (sizeof(W_)*BITS_PER_BYTE);
1147 // Also at this point we set the BF_COMPACTED flag
1148 // for this block. The invariant is that
1149 // BF_COMPACTED is always unset, except during GC
1150 // when it is set on those blocks which will be
1152 bd->flags |= BF_COMPACTED;
1158 // For each GC thread, for each step, allocate a "todo" block to
1159 // store evacuated objects to be scavenged, and a block to store
1160 // evacuated objects that do not need to be scavenged.
1161 for (t = 0; t < n_threads; t++) {
1162 for (s = 0; s < generations[g].n_steps; s++) {
1164 // we don't copy objects into g0s0, unless -G0
1165 if (g==0 && s==0 && RtsFlags.GcFlags.generations > 1) continue;
1167 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1169 ws->todo_large_objects = NULL;
1171 ws->part_list = NULL;
1172 ws->n_part_blocks = 0;
1174 // allocate the first to-space block; extra blocks will be
1175 // chained on as necessary.
1177 ws->buffer_todo_bd = NULL;
1178 alloc_todo_block(ws,0);
1180 ws->scavd_list = NULL;
1181 ws->n_scavd_blocks = 0;
1187 /* ----------------------------------------------------------------------------
1188 Initialise a generation that is *not* to be collected
1189 ------------------------------------------------------------------------- */
1192 init_uncollected_gen (nat g, nat threads)
1199 for (s = 0; s < generations[g].n_steps; s++) {
1200 stp = &generations[g].steps[s];
1201 stp->scavenged_large_objects = NULL;
1202 stp->n_scavenged_large_blocks = 0;
1205 for (s = 0; s < generations[g].n_steps; s++) {
1207 stp = &generations[g].steps[s];
1209 for (t = 0; t < threads; t++) {
1210 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1212 ws->buffer_todo_bd = NULL;
1213 ws->todo_large_objects = NULL;
1215 ws->part_list = NULL;
1216 ws->n_part_blocks = 0;
1218 ws->scavd_list = NULL;
1219 ws->n_scavd_blocks = 0;
1221 // If the block at the head of the list in this generation
1222 // is less than 3/4 full, then use it as a todo block.
1223 if (stp->blocks && isPartiallyFull(stp->blocks))
1225 ws->todo_bd = stp->blocks;
1226 ws->todo_free = ws->todo_bd->free;
1227 ws->todo_lim = ws->todo_bd->start + BLOCK_SIZE_W;
1228 stp->blocks = stp->blocks->link;
1230 stp->n_words -= ws->todo_bd->free - ws->todo_bd->start;
1231 ws->todo_bd->link = NULL;
1232 // we must scan from the current end point.
1233 ws->todo_bd->u.scan = ws->todo_bd->free;
1238 alloc_todo_block(ws,0);
1242 // deal out any more partial blocks to the threads' part_lists
1244 while (stp->blocks && isPartiallyFull(stp->blocks))
1247 stp->blocks = bd->link;
1248 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1249 bd->link = ws->part_list;
1251 ws->n_part_blocks += 1;
1252 bd->u.scan = bd->free;
1254 stp->n_words -= bd->free - bd->start;
1256 if (t == n_gc_threads) t = 0;
1261 // Move the private mutable lists from each capability onto the
1262 // main mutable list for the generation.
1263 for (i = 0; i < n_capabilities; i++) {
1264 for (bd = capabilities[i].mut_lists[g];
1265 bd->link != NULL; bd = bd->link) {
1268 bd->link = generations[g].mut_list;
1269 generations[g].mut_list = capabilities[i].mut_lists[g];
1270 capabilities[i].mut_lists[g] = allocBlock();
1274 /* -----------------------------------------------------------------------------
1275 Initialise a gc_thread before GC
1276 -------------------------------------------------------------------------- */
1279 init_gc_thread (gc_thread *t)
1281 t->static_objects = END_OF_STATIC_LIST;
1282 t->scavenged_static_objects = END_OF_STATIC_LIST;
1285 t->failed_to_evac = rtsFalse;
1286 t->eager_promotion = rtsTrue;
1287 t->thunk_selector_depth = 0;
1292 t->scav_find_work = 0;
1295 /* -----------------------------------------------------------------------------
1296 Function we pass to evacuate roots.
1297 -------------------------------------------------------------------------- */
1300 mark_root(void *user, StgClosure **root)
1302 // we stole a register for gct, but this function is called from
1303 // *outside* the GC where the register variable is not in effect,
1304 // so we need to save and restore it here. NB. only call
1305 // mark_root() from the main GC thread, otherwise gct will be
1307 gc_thread *saved_gct;
1316 /* -----------------------------------------------------------------------------
1317 Initialising the static object & mutable lists
1318 -------------------------------------------------------------------------- */
1321 zero_static_object_list(StgClosure* first_static)
1325 const StgInfoTable *info;
1327 for (p = first_static; p != END_OF_STATIC_LIST; p = link) {
1329 link = *STATIC_LINK(info, p);
1330 *STATIC_LINK(info,p) = NULL;
1334 /* ----------------------------------------------------------------------------
1335 Update the pointers from the task list
1337 These are treated as weak pointers because we want to allow a main
1338 thread to get a BlockedOnDeadMVar exception in the same way as any
1339 other thread. Note that the threads should all have been retained
1340 by GC by virtue of being on the all_threads list, we're just
1341 updating pointers here.
1342 ------------------------------------------------------------------------- */
1345 update_task_list (void)
1349 for (task = all_tasks; task != NULL; task = task->all_link) {
1350 if (!task->stopped && task->tso) {
1351 ASSERT(task->tso->bound == task);
1352 tso = (StgTSO *) isAlive((StgClosure *)task->tso);
1354 barf("task %p: main thread %d has been GC'd",
1367 /* ----------------------------------------------------------------------------
1368 Reset the sizes of the older generations when we do a major
1371 CURRENT STRATEGY: make all generations except zero the same size.
1372 We have to stay within the maximum heap size, and leave a certain
1373 percentage of the maximum heap size available to allocate into.
1374 ------------------------------------------------------------------------- */
1377 resize_generations (void)
1381 if (major_gc && RtsFlags.GcFlags.generations > 1) {
1382 nat live, size, min_alloc;
1383 nat max = RtsFlags.GcFlags.maxHeapSize;
1384 nat gens = RtsFlags.GcFlags.generations;
1386 // live in the oldest generations
1387 live = (oldest_gen->steps[0].n_words + BLOCK_SIZE_W - 1) / BLOCK_SIZE_W+
1388 oldest_gen->steps[0].n_large_blocks;
1390 // default max size for all generations except zero
1391 size = stg_max(live * RtsFlags.GcFlags.oldGenFactor,
1392 RtsFlags.GcFlags.minOldGenSize);
1394 // minimum size for generation zero
1395 min_alloc = stg_max((RtsFlags.GcFlags.pcFreeHeap * max) / 200,
1396 RtsFlags.GcFlags.minAllocAreaSize);
1398 // Auto-enable compaction when the residency reaches a
1399 // certain percentage of the maximum heap size (default: 30%).
1400 if (RtsFlags.GcFlags.generations > 1 &&
1401 (RtsFlags.GcFlags.compact ||
1403 oldest_gen->steps[0].n_blocks >
1404 (RtsFlags.GcFlags.compactThreshold * max) / 100))) {
1405 oldest_gen->steps[0].is_compacted = 1;
1406 // debugBelch("compaction: on\n", live);
1408 oldest_gen->steps[0].is_compacted = 0;
1409 // debugBelch("compaction: off\n", live);
1412 // if we're going to go over the maximum heap size, reduce the
1413 // size of the generations accordingly. The calculation is
1414 // different if compaction is turned on, because we don't need
1415 // to double the space required to collect the old generation.
1418 // this test is necessary to ensure that the calculations
1419 // below don't have any negative results - we're working
1420 // with unsigned values here.
1421 if (max < min_alloc) {
1425 if (oldest_gen->steps[0].is_compacted) {
1426 if ( (size + (size - 1) * (gens - 2) * 2) + min_alloc > max ) {
1427 size = (max - min_alloc) / ((gens - 1) * 2 - 1);
1430 if ( (size * (gens - 1) * 2) + min_alloc > max ) {
1431 size = (max - min_alloc) / ((gens - 1) * 2);
1441 debugBelch("live: %d, min_alloc: %d, size : %d, max = %d\n", live,
1442 min_alloc, size, max);
1445 for (g = 0; g < gens; g++) {
1446 generations[g].max_blocks = size;
1451 /* -----------------------------------------------------------------------------
1452 Calculate the new size of the nursery, and resize it.
1453 -------------------------------------------------------------------------- */
1456 resize_nursery (void)
1458 if (RtsFlags.GcFlags.generations == 1)
1459 { // Two-space collector:
1462 /* set up a new nursery. Allocate a nursery size based on a
1463 * function of the amount of live data (by default a factor of 2)
1464 * Use the blocks from the old nursery if possible, freeing up any
1467 * If we get near the maximum heap size, then adjust our nursery
1468 * size accordingly. If the nursery is the same size as the live
1469 * data (L), then we need 3L bytes. We can reduce the size of the
1470 * nursery to bring the required memory down near 2L bytes.
1472 * A normal 2-space collector would need 4L bytes to give the same
1473 * performance we get from 3L bytes, reducing to the same
1474 * performance at 2L bytes.
1476 blocks = g0s0->n_blocks;
1478 if ( RtsFlags.GcFlags.maxHeapSize != 0 &&
1479 blocks * RtsFlags.GcFlags.oldGenFactor * 2 >
1480 RtsFlags.GcFlags.maxHeapSize )
1482 long adjusted_blocks; // signed on purpose
1485 adjusted_blocks = (RtsFlags.GcFlags.maxHeapSize - 2 * blocks);
1487 debugTrace(DEBUG_gc, "near maximum heap size of 0x%x blocks, blocks = %d, adjusted to %ld",
1488 RtsFlags.GcFlags.maxHeapSize, blocks, adjusted_blocks);
1490 pc_free = adjusted_blocks * 100 / RtsFlags.GcFlags.maxHeapSize;
1491 if (pc_free < RtsFlags.GcFlags.pcFreeHeap) /* might even * be < 0 */
1495 blocks = adjusted_blocks;
1499 blocks *= RtsFlags.GcFlags.oldGenFactor;
1500 if (blocks < RtsFlags.GcFlags.minAllocAreaSize)
1502 blocks = RtsFlags.GcFlags.minAllocAreaSize;
1505 resizeNurseries(blocks);
1507 else // Generational collector
1510 * If the user has given us a suggested heap size, adjust our
1511 * allocation area to make best use of the memory available.
1513 if (RtsFlags.GcFlags.heapSizeSuggestion)
1516 nat needed = calcNeeded(); // approx blocks needed at next GC
1518 /* Guess how much will be live in generation 0 step 0 next time.
1519 * A good approximation is obtained by finding the
1520 * percentage of g0s0 that was live at the last minor GC.
1522 * We have an accurate figure for the amount of copied data in
1523 * 'copied', but we must convert this to a number of blocks, with
1524 * a small adjustment for estimated slop at the end of a block
1529 g0s0_pcnt_kept = ((copied / (BLOCK_SIZE_W - 10)) * 100)
1530 / countNurseryBlocks();
1533 /* Estimate a size for the allocation area based on the
1534 * information available. We might end up going slightly under
1535 * or over the suggested heap size, but we should be pretty
1538 * Formula: suggested - needed
1539 * ----------------------------
1540 * 1 + g0s0_pcnt_kept/100
1542 * where 'needed' is the amount of memory needed at the next
1543 * collection for collecting all steps except g0s0.
1546 (((long)RtsFlags.GcFlags.heapSizeSuggestion - (long)needed) * 100) /
1547 (100 + (long)g0s0_pcnt_kept);
1549 if (blocks < (long)RtsFlags.GcFlags.minAllocAreaSize) {
1550 blocks = RtsFlags.GcFlags.minAllocAreaSize;
1553 resizeNurseries((nat)blocks);
1557 // we might have added extra large blocks to the nursery, so
1558 // resize back to minAllocAreaSize again.
1559 resizeNurseriesFixed(RtsFlags.GcFlags.minAllocAreaSize);
1564 /* -----------------------------------------------------------------------------
1565 Sanity code for CAF garbage collection.
1567 With DEBUG turned on, we manage a CAF list in addition to the SRT
1568 mechanism. After GC, we run down the CAF list and blackhole any
1569 CAFs which have been garbage collected. This means we get an error
1570 whenever the program tries to enter a garbage collected CAF.
1572 Any garbage collected CAFs are taken off the CAF list at the same
1574 -------------------------------------------------------------------------- */
1576 #if 0 && defined(DEBUG)
1583 const StgInfoTable *info;
1594 ASSERT(info->type == IND_STATIC);
1596 if (STATIC_LINK(info,p) == NULL) {
1597 debugTrace(DEBUG_gccafs, "CAF gc'd at 0x%04lx", (long)p);
1599 SET_INFO(p,&stg_BLACKHOLE_info);
1600 p = STATIC_LINK2(info,p);
1604 pp = &STATIC_LINK2(info,p);
1611 debugTrace(DEBUG_gccafs, "%d CAFs live", i);