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;
134 /* -----------------------------------------------------------------------------
135 Static function declarations
136 -------------------------------------------------------------------------- */
138 static void mark_root (void *user, StgClosure **root);
139 static void zero_static_object_list (StgClosure* first_static);
140 static nat initialise_N (rtsBool force_major_gc);
141 static void init_collected_gen (nat g, nat threads);
142 static void init_uncollected_gen (nat g, nat threads);
143 static void init_gc_thread (gc_thread *t);
144 static void update_task_list (void);
145 static void resize_generations (void);
146 static void resize_nursery (void);
147 static void start_gc_threads (void);
148 static void scavenge_until_all_done (void);
149 static nat inc_running (void);
150 static nat dec_running (void);
151 static void wakeup_gc_threads (nat n_threads, nat me);
152 static void shutdown_gc_threads (nat n_threads, nat me);
153 static void continue_gc_threads (nat n_threads, nat me);
155 #if 0 && defined(DEBUG)
156 static void gcCAFs (void);
159 /* -----------------------------------------------------------------------------
160 The mark bitmap & stack.
161 -------------------------------------------------------------------------- */
163 #define MARK_STACK_BLOCKS 4
165 bdescr *mark_stack_bdescr;
170 // Flag and pointers used for falling back to a linear scan when the
171 // mark stack overflows.
172 rtsBool mark_stack_overflowed;
173 bdescr *oldgen_scan_bd;
176 /* -----------------------------------------------------------------------------
177 GarbageCollect: the main entry point to the garbage collector.
179 Locks held: all capabilities are held throughout GarbageCollect().
180 -------------------------------------------------------------------------- */
183 GarbageCollect (rtsBool force_major_gc,
184 nat gc_type USED_IF_THREADS,
185 Capability *cap USED_IF_THREADS)
189 lnat live, allocated, max_copied, avg_copied, slop;
190 gc_thread *saved_gct;
193 // necessary if we stole a callee-saves register for gct:
197 CostCentreStack *prev_CCS;
202 #if defined(RTS_USER_SIGNALS)
203 if (RtsFlags.MiscFlags.install_signal_handlers) {
209 ASSERT(sizeof(step_workspace) == 16 * sizeof(StgWord));
210 // otherwise adjust the padding in step_workspace.
212 // tell the stats department that we've started a GC
215 // tell the STM to discard any cached closures it's hoping to re-use
224 // attribute any costs to CCS_GC
230 /* Approximate how much we allocated.
231 * Todo: only when generating stats?
233 allocated = calcAllocated();
235 /* Figure out which generation to collect
237 n = initialise_N(force_major_gc);
239 /* Start threads, so they can be spinning up while we finish initialisation.
243 #if defined(THREADED_RTS)
244 /* How many threads will be participating in this GC?
245 * We don't try to parallelise minor GCs (unless the user asks for
246 * it with +RTS -gn0), or mark/compact/sweep GC.
248 if (gc_type == PENDING_GC_PAR) {
249 n_gc_threads = RtsFlags.ParFlags.nNodes;
257 trace(TRACE_gc|DEBUG_gc, "GC (gen %d): %d KB to collect, %ld MB in use, using %d thread(s)",
258 N, n * (BLOCK_SIZE / 1024), mblocks_allocated, n_gc_threads);
260 #ifdef RTS_GTK_FRONTPANEL
261 if (RtsFlags.GcFlags.frontpanel) {
262 updateFrontPanelBeforeGC(N);
267 // check for memory leaks if DEBUG is on
268 memInventory(traceClass(DEBUG_gc));
271 // check stack sanity *before* GC
272 IF_DEBUG(sanity, checkFreeListSanity());
273 IF_DEBUG(sanity, checkMutableLists());
275 // Initialise all our gc_thread structures
276 for (t = 0; t < n_gc_threads; t++) {
277 init_gc_thread(gc_threads[t]);
280 // Initialise all the generations/steps that we're collecting.
281 for (g = 0; g <= N; g++) {
282 init_collected_gen(g,n_gc_threads);
285 // Initialise all the generations/steps that we're *not* collecting.
286 for (g = N+1; g < RtsFlags.GcFlags.generations; g++) {
287 init_uncollected_gen(g,n_gc_threads);
290 /* Allocate a mark stack if we're doing a major collection.
292 if (major_gc && oldest_gen->steps[0].mark) {
293 nat mark_stack_blocks;
294 mark_stack_blocks = stg_max(MARK_STACK_BLOCKS,
295 oldest_gen->steps[0].n_old_blocks / 100);
296 mark_stack_bdescr = allocGroup(mark_stack_blocks);
297 mark_stack = (StgPtr *)mark_stack_bdescr->start;
298 mark_sp = mark_stack;
299 mark_splim = mark_stack + (mark_stack_blocks * BLOCK_SIZE_W);
301 mark_stack_bdescr = NULL;
304 // this is the main thread
306 if (n_gc_threads == 1) {
309 gct = gc_threads[cap->no];
315 /* -----------------------------------------------------------------------
316 * follow all the roots that we know about:
317 * - mutable lists from each generation > N
318 * we want to *scavenge* these roots, not evacuate them: they're not
319 * going to move in this GC.
320 * Also do them in reverse generation order, for the usual reason:
321 * namely to reduce the likelihood of spurious old->new pointers.
323 for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
324 generations[g].saved_mut_list = generations[g].mut_list;
325 generations[g].mut_list = allocBlock();
326 // mut_list always has at least one block.
329 // the main thread is running: this prevents any other threads from
330 // exiting prematurely, so we can start them now.
331 // NB. do this after the mutable lists have been saved above, otherwise
332 // the other GC threads will be writing into the old mutable lists.
334 wakeup_gc_threads(n_gc_threads, gct->thread_index);
336 for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
337 scavenge_mutable_list(&generations[g]);
340 // follow roots from the CAF list (used by GHCi)
342 markCAFs(mark_root, gct);
344 // follow all the roots that the application knows about.
346 markSomeCapabilities(mark_root, gct, gct->thread_index, n_gc_threads,
347 rtsTrue/*prune sparks*/);
349 #if defined(RTS_USER_SIGNALS)
350 // mark the signal handlers (signals should be already blocked)
351 markSignalHandlers(mark_root, gct);
354 // Mark the weak pointer list, and prepare to detect dead weak pointers.
358 // Mark the stable pointer table.
359 markStablePtrTable(mark_root, gct);
361 /* -------------------------------------------------------------------------
362 * Repeatedly scavenge all the areas we know about until there's no
363 * more scavenging to be done.
367 scavenge_until_all_done();
368 // The other threads are now stopped. We might recurse back to
369 // here, but from now on this is the only thread.
371 // if any blackholes are alive, make the threads that wait on
373 if (traverseBlackholeQueue()) {
378 // must be last... invariant is that everything is fully
379 // scavenged at this point.
380 if (traverseWeakPtrList()) { // returns rtsTrue if evaced something
385 // If we get to here, there's really nothing left to do.
389 shutdown_gc_threads(n_gc_threads, gct->thread_index);
391 // Update pointers from the Task list
394 // Now see which stable names are still alive.
398 // We call processHeapClosureForDead() on every closure destroyed during
399 // the current garbage collection, so we invoke LdvCensusForDead().
400 if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_LDV
401 || RtsFlags.ProfFlags.bioSelector != NULL)
405 // NO MORE EVACUATION AFTER THIS POINT!
407 // Two-space collector: free the old to-space.
408 // g0s0->old_blocks is the old nursery
409 // g0s0->blocks is to-space from the previous GC
410 if (RtsFlags.GcFlags.generations == 1) {
411 if (g0s0->blocks != NULL) {
412 freeChain(g0s0->blocks);
417 // For each workspace, in each thread, move the copied blocks to the step
423 for (t = 0; t < n_gc_threads; t++) {
427 if (RtsFlags.GcFlags.generations == 1) {
432 for (; s < total_steps; s++) {
435 // Push the final block
437 push_scanned_block(ws->todo_bd, ws);
440 ASSERT(gct->scan_bd == NULL);
441 ASSERT(countBlocks(ws->scavd_list) == ws->n_scavd_blocks);
444 for (bd = ws->scavd_list; bd != NULL; bd = bd->link) {
445 ws->step->n_words += bd->free - bd->start;
449 prev->link = ws->step->blocks;
450 ws->step->blocks = ws->scavd_list;
452 ws->step->n_blocks += ws->n_scavd_blocks;
456 // Add all the partial blocks *after* we've added all the full
457 // blocks. This is so that we can grab the partial blocks back
458 // again and try to fill them up in the next GC.
459 for (t = 0; t < n_gc_threads; t++) {
463 if (RtsFlags.GcFlags.generations == 1) {
468 for (; s < total_steps; s++) {
472 for (bd = ws->part_list; bd != NULL; bd = next) {
474 if (bd->free == bd->start) {
476 ws->part_list = next;
483 ws->step->n_words += bd->free - bd->start;
488 prev->link = ws->step->blocks;
489 ws->step->blocks = ws->part_list;
491 ws->step->n_blocks += ws->n_part_blocks;
493 ASSERT(countBlocks(ws->step->blocks) == ws->step->n_blocks);
494 ASSERT(countOccupied(ws->step->blocks) == ws->step->n_words);
499 // Finally: compact or sweep the oldest generation.
500 if (major_gc && oldest_gen->steps[0].mark) {
501 if (oldest_gen->steps[0].compact)
502 compact(gct->scavenged_static_objects);
504 sweep(&oldest_gen->steps[0]);
507 /* run through all the generations/steps and tidy up
514 for (i=0; i < n_gc_threads; i++) {
515 if (n_gc_threads > 1) {
516 trace(TRACE_gc,"thread %d:", i);
517 trace(TRACE_gc," copied %ld", gc_threads[i]->copied * sizeof(W_));
518 trace(TRACE_gc," scanned %ld", gc_threads[i]->scanned * sizeof(W_));
519 trace(TRACE_gc," any_work %ld", gc_threads[i]->any_work);
520 trace(TRACE_gc," no_work %ld", gc_threads[i]->no_work);
521 trace(TRACE_gc," scav_find_work %ld", gc_threads[i]->scav_find_work);
523 copied += gc_threads[i]->copied;
524 max_copied = stg_max(gc_threads[i]->copied, max_copied);
526 if (n_gc_threads == 1) {
534 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
537 generations[g].collections++; // for stats
538 if (n_gc_threads > 1) generations[g].par_collections++;
541 // Count the mutable list as bytes "copied" for the purposes of
542 // stats. Every mutable list is copied during every GC.
544 nat mut_list_size = 0;
545 for (bd = generations[g].mut_list; bd != NULL; bd = bd->link) {
546 mut_list_size += bd->free - bd->start;
548 copied += mut_list_size;
551 "mut_list_size: %lu (%d vars, %d arrays, %d MVARs, %d others)",
552 (unsigned long)(mut_list_size * sizeof(W_)),
553 mutlist_MUTVARS, mutlist_MUTARRS, mutlist_MVARS, mutlist_OTHERS);
556 for (s = 0; s < generations[g].n_steps; s++) {
558 stp = &generations[g].steps[s];
560 // for generations we collected...
563 /* free old memory and shift to-space into from-space for all
564 * the collected steps (except the allocation area). These
565 * freed blocks will probaby be quickly recycled.
567 if (!(g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1)) {
570 // tack the new blocks on the end of the existing blocks
571 if (stp->old_blocks != NULL) {
574 for (bd = stp->old_blocks; bd != NULL; bd = next) {
578 if (!(bd->flags & BF_MARKED))
581 stp->old_blocks = next;
590 stp->n_words += bd->free - bd->start;
592 // NB. this step might not be compacted next
593 // time, so reset the BF_MARKED flags.
594 // They are set before GC if we're going to
595 // compact. (search for BF_MARKED above).
596 bd->flags &= ~BF_MARKED;
598 // between GCs, all blocks in the heap except
599 // for the nursery have the BF_EVACUATED flag set.
600 bd->flags |= BF_EVACUATED;
607 prev->link = stp->blocks;
608 stp->blocks = stp->old_blocks;
611 // add the new blocks to the block tally
612 stp->n_blocks += stp->n_old_blocks;
613 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
614 ASSERT(countOccupied(stp->blocks) == stp->n_words);
618 freeChain(stp->old_blocks);
620 stp->old_blocks = NULL;
621 stp->n_old_blocks = 0;
624 /* LARGE OBJECTS. The current live large objects are chained on
625 * scavenged_large, having been moved during garbage
626 * collection from large_objects. Any objects left on
627 * large_objects list are therefore dead, so we free them here.
629 for (bd = stp->large_objects; bd != NULL; bd = next) {
635 stp->large_objects = stp->scavenged_large_objects;
636 stp->n_large_blocks = stp->n_scavenged_large_blocks;
639 else // for older generations...
641 /* For older generations, we need to append the
642 * scavenged_large_object list (i.e. large objects that have been
643 * promoted during this GC) to the large_object list for that step.
645 for (bd = stp->scavenged_large_objects; bd; bd = next) {
647 dbl_link_onto(bd, &stp->large_objects);
650 // add the new blocks we promoted during this GC
651 stp->n_large_blocks += stp->n_scavenged_large_blocks;
656 // update the max size of older generations after a major GC
657 resize_generations();
659 // Calculate the amount of live data for stats.
660 live = calcLiveWords();
662 // Free the small objects allocated via allocate(), since this will
663 // all have been copied into G0S1 now.
664 if (RtsFlags.GcFlags.generations > 1) {
665 if (g0s0->blocks != NULL) {
666 freeChain(g0s0->blocks);
673 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
675 // Start a new pinned_object_block
676 pinned_object_block = NULL;
678 // Free the mark stack.
679 if (mark_stack_bdescr != NULL) {
680 freeGroup(mark_stack_bdescr);
684 for (g = 0; g <= N; g++) {
685 for (s = 0; s < generations[g].n_steps; s++) {
686 stp = &generations[g].steps[s];
687 if (stp->bitmap != NULL) {
688 freeGroup(stp->bitmap);
696 // mark the garbage collected CAFs as dead
697 #if 0 && defined(DEBUG) // doesn't work at the moment
698 if (major_gc) { gcCAFs(); }
702 // resetStaticObjectForRetainerProfiling() must be called before
704 if (n_gc_threads > 1) {
705 barf("profiling is currently broken with multi-threaded GC");
706 // ToDo: fix the gct->scavenged_static_objects below
708 resetStaticObjectForRetainerProfiling(gct->scavenged_static_objects);
711 // zero the scavenged static object list
714 for (i = 0; i < n_gc_threads; i++) {
715 zero_static_object_list(gc_threads[i]->scavenged_static_objects);
722 // start any pending finalizers
724 scheduleFinalizers(last_free_capability, old_weak_ptr_list);
727 // send exceptions to any threads which were about to die
729 resurrectThreads(resurrected_threads);
730 performPendingThrowTos(exception_threads);
733 // Update the stable pointer hash table.
734 updateStablePtrTable(major_gc);
736 // check sanity after GC
737 IF_DEBUG(sanity, checkSanity());
739 // extra GC trace info
740 if (traceClass(TRACE_gc|DEBUG_gc)) statDescribeGens();
743 // symbol-table based profiling
744 /* heapCensus(to_blocks); */ /* ToDo */
747 // restore enclosing cost centre
753 // check for memory leaks if DEBUG is on
754 memInventory(traceClass(DEBUG_gc));
757 #ifdef RTS_GTK_FRONTPANEL
758 if (RtsFlags.GcFlags.frontpanel) {
759 updateFrontPanelAfterGC( N, live );
763 // ok, GC over: tell the stats department what happened.
764 slop = calcLiveBlocks() * BLOCK_SIZE_W - live;
765 stat_endGC(allocated, live, copied, N, max_copied, avg_copied, slop);
767 // Guess which generation we'll collect *next* time
768 initialise_N(force_major_gc);
770 #if defined(RTS_USER_SIGNALS)
771 if (RtsFlags.MiscFlags.install_signal_handlers) {
772 // unblock signals again
773 unblockUserSignals();
777 continue_gc_threads(n_gc_threads, gct->thread_index);
784 /* -----------------------------------------------------------------------------
785 Figure out which generation to collect, initialise N and major_gc.
787 Also returns the total number of blocks in generations that will be
789 -------------------------------------------------------------------------- */
792 initialise_N (rtsBool force_major_gc)
795 nat s, blocks, blocks_total;
800 if (force_major_gc) {
801 N = RtsFlags.GcFlags.generations - 1;
806 for (g = RtsFlags.GcFlags.generations - 1; g >= 0; g--) {
808 for (s = 0; s < generations[g].n_steps; s++) {
809 blocks += generations[g].steps[s].n_words / BLOCK_SIZE_W;
810 blocks += generations[g].steps[s].n_large_blocks;
812 if (blocks >= generations[g].max_blocks) {
816 blocks_total += blocks;
820 blocks_total += countNurseryBlocks();
822 major_gc = (N == RtsFlags.GcFlags.generations-1);
826 /* -----------------------------------------------------------------------------
827 Initialise the gc_thread structures.
828 -------------------------------------------------------------------------- */
830 #define GC_THREAD_INACTIVE 0
831 #define GC_THREAD_STANDING_BY 1
832 #define GC_THREAD_RUNNING 2
833 #define GC_THREAD_WAITING_TO_CONTINUE 3
836 alloc_gc_thread (int n)
842 t = stgMallocBytes(sizeof(gc_thread) + total_steps * sizeof(step_workspace),
847 initSpinLock(&t->gc_spin);
848 initSpinLock(&t->mut_spin);
849 ACQUIRE_SPIN_LOCK(&t->gc_spin);
850 t->wakeup = GC_THREAD_INACTIVE; // starts true, so we can wait for the
851 // thread to start up, see wakeup_gc_threads
855 t->free_blocks = NULL;
864 for (s = 0; s < total_steps; s++)
867 ws->step = &all_steps[s];
868 ASSERT(s == ws->step->abs_no);
872 ws->buffer_todo_bd = NULL;
874 ws->part_list = NULL;
875 ws->n_part_blocks = 0;
877 ws->scavd_list = NULL;
878 ws->n_scavd_blocks = 0;
888 if (gc_threads == NULL) {
889 #if defined(THREADED_RTS)
891 gc_threads = stgMallocBytes (RtsFlags.ParFlags.nNodes *
895 for (i = 0; i < RtsFlags.ParFlags.nNodes; i++) {
896 gc_threads[i] = alloc_gc_thread(i);
899 gc_threads = stgMallocBytes (sizeof(gc_thread*),
902 gc_threads[0] = alloc_gc_thread(0);
907 /* ----------------------------------------------------------------------------
909 ------------------------------------------------------------------------- */
911 static nat gc_running_threads;
913 #if defined(THREADED_RTS)
914 static Mutex gc_running_mutex;
921 ACQUIRE_LOCK(&gc_running_mutex);
922 n_running = ++gc_running_threads;
923 RELEASE_LOCK(&gc_running_mutex);
924 ASSERT(n_running <= n_gc_threads);
932 ACQUIRE_LOCK(&gc_running_mutex);
933 ASSERT(n_gc_threads != 0);
934 n_running = --gc_running_threads;
935 RELEASE_LOCK(&gc_running_mutex);
949 // scavenge objects in compacted generation
950 if (mark_stack_overflowed || oldgen_scan_bd != NULL ||
951 (mark_stack_bdescr != NULL && !mark_stack_empty())) {
955 // Check for global work in any step. We don't need to check for
956 // local work, because we have already exited scavenge_loop(),
957 // which means there is no local work for this thread.
958 for (s = total_steps-1; s >= 0; s--) {
959 if (s == 0 && RtsFlags.GcFlags.generations > 1) {
963 if (ws->todo_large_objects) return rtsTrue;
964 if (ws->step->todos) return rtsTrue;
973 scavenge_until_all_done (void)
977 debugTrace(DEBUG_gc, "GC thread %d working", gct->thread_index);
980 #if defined(THREADED_RTS)
981 if (n_gc_threads > 1) {
990 // scavenge_loop() only exits when there's no work to do
993 debugTrace(DEBUG_gc, "GC thread %d idle (%d still running)",
994 gct->thread_index, r);
996 while (gc_running_threads != 0) {
1002 // any_work() does not remove the work from the queue, it
1003 // just checks for the presence of work. If we find any,
1004 // then we increment gc_running_threads and go back to
1005 // scavenge_loop() to perform any pending work.
1008 // All threads are now stopped
1009 debugTrace(DEBUG_gc, "GC thread %d finished.", gct->thread_index);
1012 #if defined(THREADED_RTS)
1015 gcWorkerThread (Capability *cap)
1017 cap->in_gc = rtsTrue;
1019 gct = gc_threads[cap->no];
1020 gct->id = osThreadId();
1022 // Wait until we're told to wake up
1023 RELEASE_SPIN_LOCK(&gct->mut_spin);
1024 gct->wakeup = GC_THREAD_STANDING_BY;
1025 debugTrace(DEBUG_gc, "GC thread %d standing by...", gct->thread_index);
1026 ACQUIRE_SPIN_LOCK(&gct->gc_spin);
1029 // start performance counters in this thread...
1030 if (gct->papi_events == -1) {
1031 papi_init_eventset(&gct->papi_events);
1033 papi_thread_start_gc1_count(gct->papi_events);
1036 // Every thread evacuates some roots.
1038 markSomeCapabilities(mark_root, gct, gct->thread_index, n_gc_threads,
1039 rtsTrue/*prune sparks*/);
1041 scavenge_until_all_done();
1044 // count events in this thread towards the GC totals
1045 papi_thread_stop_gc1_count(gct->papi_events);
1048 // Wait until we're told to continue
1049 RELEASE_SPIN_LOCK(&gct->gc_spin);
1050 gct->wakeup = GC_THREAD_WAITING_TO_CONTINUE;
1051 debugTrace(DEBUG_gc, "GC thread %d waiting to continue...",
1053 ACQUIRE_SPIN_LOCK(&gct->mut_spin);
1054 debugTrace(DEBUG_gc, "GC thread %d on my way...", gct->thread_index);
1060 waitForGcThreads (Capability *cap USED_IF_THREADS)
1062 #if defined(THREADED_RTS)
1063 nat n_threads = RtsFlags.ParFlags.nNodes;
1066 rtsBool retry = rtsTrue;
1069 for (i=0; i < n_threads; i++) {
1070 if (i == me) continue;
1071 if (gc_threads[i]->wakeup != GC_THREAD_STANDING_BY) {
1072 prodCapability(&capabilities[i], cap->running_task);
1075 for (j=0; j < 10000000; j++) {
1077 for (i=0; i < n_threads; i++) {
1078 if (i == me) continue;
1080 setContextSwitches();
1081 if (gc_threads[i]->wakeup != GC_THREAD_STANDING_BY) {
1092 start_gc_threads (void)
1094 #if defined(THREADED_RTS)
1095 gc_running_threads = 0;
1096 initMutex(&gc_running_mutex);
1101 wakeup_gc_threads (nat n_threads USED_IF_THREADS, nat me USED_IF_THREADS)
1103 #if defined(THREADED_RTS)
1105 for (i=0; i < n_threads; i++) {
1106 if (i == me) continue;
1108 debugTrace(DEBUG_gc, "waking up gc thread %d", i);
1109 if (gc_threads[i]->wakeup != GC_THREAD_STANDING_BY) barf("wakeup_gc_threads");
1111 gc_threads[i]->wakeup = GC_THREAD_RUNNING;
1112 ACQUIRE_SPIN_LOCK(&gc_threads[i]->mut_spin);
1113 RELEASE_SPIN_LOCK(&gc_threads[i]->gc_spin);
1118 // After GC is complete, we must wait for all GC threads to enter the
1119 // standby state, otherwise they may still be executing inside
1120 // any_work(), and may even remain awake until the next GC starts.
1122 shutdown_gc_threads (nat n_threads USED_IF_THREADS, nat me USED_IF_THREADS)
1124 #if defined(THREADED_RTS)
1126 for (i=0; i < n_threads; i++) {
1127 if (i == me) continue;
1128 while (gc_threads[i]->wakeup != GC_THREAD_WAITING_TO_CONTINUE) { write_barrier(); }
1134 continue_gc_threads (nat n_threads USED_IF_THREADS, nat me USED_IF_THREADS)
1136 #if defined(THREADED_RTS)
1138 for (i=0; i < n_threads; i++) {
1139 if (i == me) continue;
1140 if (gc_threads[i]->wakeup != GC_THREAD_WAITING_TO_CONTINUE) barf("continue_gc_threads");
1142 gc_threads[i]->wakeup = GC_THREAD_INACTIVE;
1143 ACQUIRE_SPIN_LOCK(&gc_threads[i]->gc_spin);
1144 RELEASE_SPIN_LOCK(&gc_threads[i]->mut_spin);
1149 /* ----------------------------------------------------------------------------
1150 Initialise a generation that is to be collected
1151 ------------------------------------------------------------------------- */
1154 init_collected_gen (nat g, nat n_threads)
1161 // Throw away the current mutable list. Invariant: the mutable
1162 // list always has at least one block; this means we can avoid a
1163 // check for NULL in recordMutable().
1165 freeChain(generations[g].mut_list);
1166 generations[g].mut_list = allocBlock();
1167 for (i = 0; i < n_capabilities; i++) {
1168 freeChain(capabilities[i].mut_lists[g]);
1169 capabilities[i].mut_lists[g] = allocBlock();
1173 for (s = 0; s < generations[g].n_steps; s++) {
1175 stp = &generations[g].steps[s];
1176 ASSERT(stp->gen_no == g);
1178 // we'll construct a new list of threads in this step
1179 // during GC, throw away the current list.
1180 stp->old_threads = stp->threads;
1181 stp->threads = END_TSO_QUEUE;
1183 // generation 0, step 0 doesn't need to-space
1184 if (g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1) {
1188 // deprecate the existing blocks
1189 stp->old_blocks = stp->blocks;
1190 stp->n_old_blocks = stp->n_blocks;
1194 stp->live_estimate = 0;
1196 // we don't have any to-be-scavenged blocks yet
1198 stp->todos_last = NULL;
1201 // initialise the large object queues.
1202 stp->scavenged_large_objects = NULL;
1203 stp->n_scavenged_large_blocks = 0;
1205 // mark the small objects as from-space
1206 for (bd = stp->old_blocks; bd; bd = bd->link) {
1207 bd->flags &= ~BF_EVACUATED;
1210 // mark the large objects as from-space
1211 for (bd = stp->large_objects; bd; bd = bd->link) {
1212 bd->flags &= ~BF_EVACUATED;
1215 // for a compacted step, we need to allocate the bitmap
1217 nat bitmap_size; // in bytes
1218 bdescr *bitmap_bdescr;
1221 bitmap_size = stp->n_old_blocks * BLOCK_SIZE / (sizeof(W_)*BITS_PER_BYTE);
1223 if (bitmap_size > 0) {
1224 bitmap_bdescr = allocGroup((lnat)BLOCK_ROUND_UP(bitmap_size)
1226 stp->bitmap = bitmap_bdescr;
1227 bitmap = bitmap_bdescr->start;
1229 debugTrace(DEBUG_gc, "bitmap_size: %d, bitmap: %p",
1230 bitmap_size, bitmap);
1232 // don't forget to fill it with zeros!
1233 memset(bitmap, 0, bitmap_size);
1235 // For each block in this step, point to its bitmap from the
1236 // block descriptor.
1237 for (bd=stp->old_blocks; bd != NULL; bd = bd->link) {
1238 bd->u.bitmap = bitmap;
1239 bitmap += BLOCK_SIZE_W / (sizeof(W_)*BITS_PER_BYTE);
1241 // Also at this point we set the BF_MARKED flag
1242 // for this block. The invariant is that
1243 // BF_MARKED is always unset, except during GC
1244 // when it is set on those blocks which will be
1246 if (!(bd->flags & BF_FRAGMENTED)) {
1247 bd->flags |= BF_MARKED;
1254 // For each GC thread, for each step, allocate a "todo" block to
1255 // store evacuated objects to be scavenged, and a block to store
1256 // evacuated objects that do not need to be scavenged.
1257 for (t = 0; t < n_threads; t++) {
1258 for (s = 0; s < generations[g].n_steps; s++) {
1260 // we don't copy objects into g0s0, unless -G0
1261 if (g==0 && s==0 && RtsFlags.GcFlags.generations > 1) continue;
1263 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1265 ws->todo_large_objects = NULL;
1267 ws->part_list = NULL;
1268 ws->n_part_blocks = 0;
1270 // allocate the first to-space block; extra blocks will be
1271 // chained on as necessary.
1273 ws->buffer_todo_bd = NULL;
1274 alloc_todo_block(ws,0);
1276 ws->scavd_list = NULL;
1277 ws->n_scavd_blocks = 0;
1283 /* ----------------------------------------------------------------------------
1284 Initialise a generation that is *not* to be collected
1285 ------------------------------------------------------------------------- */
1288 init_uncollected_gen (nat g, nat threads)
1295 for (s = 0; s < generations[g].n_steps; s++) {
1296 stp = &generations[g].steps[s];
1297 stp->scavenged_large_objects = NULL;
1298 stp->n_scavenged_large_blocks = 0;
1301 for (s = 0; s < generations[g].n_steps; s++) {
1303 stp = &generations[g].steps[s];
1305 for (t = 0; t < threads; t++) {
1306 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1308 ws->buffer_todo_bd = NULL;
1309 ws->todo_large_objects = NULL;
1311 ws->part_list = NULL;
1312 ws->n_part_blocks = 0;
1314 ws->scavd_list = NULL;
1315 ws->n_scavd_blocks = 0;
1317 // If the block at the head of the list in this generation
1318 // is less than 3/4 full, then use it as a todo block.
1319 if (stp->blocks && isPartiallyFull(stp->blocks))
1321 ws->todo_bd = stp->blocks;
1322 ws->todo_free = ws->todo_bd->free;
1323 ws->todo_lim = ws->todo_bd->start + BLOCK_SIZE_W;
1324 stp->blocks = stp->blocks->link;
1326 stp->n_words -= ws->todo_bd->free - ws->todo_bd->start;
1327 ws->todo_bd->link = NULL;
1328 // we must scan from the current end point.
1329 ws->todo_bd->u.scan = ws->todo_bd->free;
1334 alloc_todo_block(ws,0);
1338 // deal out any more partial blocks to the threads' part_lists
1340 while (stp->blocks && isPartiallyFull(stp->blocks))
1343 stp->blocks = bd->link;
1344 ws = &gc_threads[t]->steps[g * RtsFlags.GcFlags.steps + s];
1345 bd->link = ws->part_list;
1347 ws->n_part_blocks += 1;
1348 bd->u.scan = bd->free;
1350 stp->n_words -= bd->free - bd->start;
1352 if (t == n_gc_threads) t = 0;
1357 // Move the private mutable lists from each capability onto the
1358 // main mutable list for the generation.
1359 for (i = 0; i < n_capabilities; i++) {
1360 for (bd = capabilities[i].mut_lists[g];
1361 bd->link != NULL; bd = bd->link) {
1364 bd->link = generations[g].mut_list;
1365 generations[g].mut_list = capabilities[i].mut_lists[g];
1366 capabilities[i].mut_lists[g] = allocBlock();
1370 /* -----------------------------------------------------------------------------
1371 Initialise a gc_thread before GC
1372 -------------------------------------------------------------------------- */
1375 init_gc_thread (gc_thread *t)
1377 t->static_objects = END_OF_STATIC_LIST;
1378 t->scavenged_static_objects = END_OF_STATIC_LIST;
1381 t->failed_to_evac = rtsFalse;
1382 t->eager_promotion = rtsTrue;
1383 t->thunk_selector_depth = 0;
1388 t->scav_find_work = 0;
1391 /* -----------------------------------------------------------------------------
1392 Function we pass to evacuate roots.
1393 -------------------------------------------------------------------------- */
1396 mark_root(void *user, StgClosure **root)
1398 // we stole a register for gct, but this function is called from
1399 // *outside* the GC where the register variable is not in effect,
1400 // so we need to save and restore it here. NB. only call
1401 // mark_root() from the main GC thread, otherwise gct will be
1403 gc_thread *saved_gct;
1412 /* -----------------------------------------------------------------------------
1413 Initialising the static object & mutable lists
1414 -------------------------------------------------------------------------- */
1417 zero_static_object_list(StgClosure* first_static)
1421 const StgInfoTable *info;
1423 for (p = first_static; p != END_OF_STATIC_LIST; p = link) {
1425 link = *STATIC_LINK(info, p);
1426 *STATIC_LINK(info,p) = NULL;
1430 /* ----------------------------------------------------------------------------
1431 Update the pointers from the task list
1433 These are treated as weak pointers because we want to allow a main
1434 thread to get a BlockedOnDeadMVar exception in the same way as any
1435 other thread. Note that the threads should all have been retained
1436 by GC by virtue of being on the all_threads list, we're just
1437 updating pointers here.
1438 ------------------------------------------------------------------------- */
1441 update_task_list (void)
1445 for (task = all_tasks; task != NULL; task = task->all_link) {
1446 if (!task->stopped && task->tso) {
1447 ASSERT(task->tso->bound == task);
1448 tso = (StgTSO *) isAlive((StgClosure *)task->tso);
1450 barf("task %p: main thread %d has been GC'd",
1463 /* ----------------------------------------------------------------------------
1464 Reset the sizes of the older generations when we do a major
1467 CURRENT STRATEGY: make all generations except zero the same size.
1468 We have to stay within the maximum heap size, and leave a certain
1469 percentage of the maximum heap size available to allocate into.
1470 ------------------------------------------------------------------------- */
1473 resize_generations (void)
1477 if (major_gc && RtsFlags.GcFlags.generations > 1) {
1478 nat live, size, min_alloc, words;
1479 nat max = RtsFlags.GcFlags.maxHeapSize;
1480 nat gens = RtsFlags.GcFlags.generations;
1482 // live in the oldest generations
1483 if (oldest_gen->steps[0].live_estimate != 0) {
1484 words = oldest_gen->steps[0].live_estimate;
1486 words = oldest_gen->steps[0].n_words;
1488 live = (words + BLOCK_SIZE_W - 1) / BLOCK_SIZE_W +
1489 oldest_gen->steps[0].n_large_blocks;
1491 // default max size for all generations except zero
1492 size = stg_max(live * RtsFlags.GcFlags.oldGenFactor,
1493 RtsFlags.GcFlags.minOldGenSize);
1495 // minimum size for generation zero
1496 min_alloc = stg_max((RtsFlags.GcFlags.pcFreeHeap * max) / 200,
1497 RtsFlags.GcFlags.minAllocAreaSize);
1499 // Auto-enable compaction when the residency reaches a
1500 // certain percentage of the maximum heap size (default: 30%).
1501 if (RtsFlags.GcFlags.generations > 1 &&
1502 (RtsFlags.GcFlags.compact ||
1504 oldest_gen->steps[0].n_blocks >
1505 (RtsFlags.GcFlags.compactThreshold * max) / 100))) {
1506 oldest_gen->steps[0].mark = 1;
1507 oldest_gen->steps[0].compact = 1;
1508 // debugBelch("compaction: on\n", live);
1510 oldest_gen->steps[0].mark = 0;
1511 oldest_gen->steps[0].compact = 0;
1512 // debugBelch("compaction: off\n", live);
1515 if (RtsFlags.GcFlags.sweep) {
1516 oldest_gen->steps[0].mark = 1;
1519 // if we're going to go over the maximum heap size, reduce the
1520 // size of the generations accordingly. The calculation is
1521 // different if compaction is turned on, because we don't need
1522 // to double the space required to collect the old generation.
1525 // this test is necessary to ensure that the calculations
1526 // below don't have any negative results - we're working
1527 // with unsigned values here.
1528 if (max < min_alloc) {
1532 if (oldest_gen->steps[0].compact) {
1533 if ( (size + (size - 1) * (gens - 2) * 2) + min_alloc > max ) {
1534 size = (max - min_alloc) / ((gens - 1) * 2 - 1);
1537 if ( (size * (gens - 1) * 2) + min_alloc > max ) {
1538 size = (max - min_alloc) / ((gens - 1) * 2);
1548 debugBelch("live: %d, min_alloc: %d, size : %d, max = %d\n", live,
1549 min_alloc, size, max);
1552 for (g = 0; g < gens; g++) {
1553 generations[g].max_blocks = size;
1558 /* -----------------------------------------------------------------------------
1559 Calculate the new size of the nursery, and resize it.
1560 -------------------------------------------------------------------------- */
1563 resize_nursery (void)
1565 if (RtsFlags.GcFlags.generations == 1)
1566 { // Two-space collector:
1569 /* set up a new nursery. Allocate a nursery size based on a
1570 * function of the amount of live data (by default a factor of 2)
1571 * Use the blocks from the old nursery if possible, freeing up any
1574 * If we get near the maximum heap size, then adjust our nursery
1575 * size accordingly. If the nursery is the same size as the live
1576 * data (L), then we need 3L bytes. We can reduce the size of the
1577 * nursery to bring the required memory down near 2L bytes.
1579 * A normal 2-space collector would need 4L bytes to give the same
1580 * performance we get from 3L bytes, reducing to the same
1581 * performance at 2L bytes.
1583 blocks = g0s0->n_blocks;
1585 if ( RtsFlags.GcFlags.maxHeapSize != 0 &&
1586 blocks * RtsFlags.GcFlags.oldGenFactor * 2 >
1587 RtsFlags.GcFlags.maxHeapSize )
1589 long adjusted_blocks; // signed on purpose
1592 adjusted_blocks = (RtsFlags.GcFlags.maxHeapSize - 2 * blocks);
1594 debugTrace(DEBUG_gc, "near maximum heap size of 0x%x blocks, blocks = %d, adjusted to %ld",
1595 RtsFlags.GcFlags.maxHeapSize, blocks, adjusted_blocks);
1597 pc_free = adjusted_blocks * 100 / RtsFlags.GcFlags.maxHeapSize;
1598 if (pc_free < RtsFlags.GcFlags.pcFreeHeap) /* might even * be < 0 */
1602 blocks = adjusted_blocks;
1606 blocks *= RtsFlags.GcFlags.oldGenFactor;
1607 if (blocks < RtsFlags.GcFlags.minAllocAreaSize)
1609 blocks = RtsFlags.GcFlags.minAllocAreaSize;
1612 resizeNurseries(blocks);
1614 else // Generational collector
1617 * If the user has given us a suggested heap size, adjust our
1618 * allocation area to make best use of the memory available.
1620 if (RtsFlags.GcFlags.heapSizeSuggestion)
1623 nat needed = calcNeeded(); // approx blocks needed at next GC
1625 /* Guess how much will be live in generation 0 step 0 next time.
1626 * A good approximation is obtained by finding the
1627 * percentage of g0s0 that was live at the last minor GC.
1629 * We have an accurate figure for the amount of copied data in
1630 * 'copied', but we must convert this to a number of blocks, with
1631 * a small adjustment for estimated slop at the end of a block
1636 g0s0_pcnt_kept = ((copied / (BLOCK_SIZE_W - 10)) * 100)
1637 / countNurseryBlocks();
1640 /* Estimate a size for the allocation area based on the
1641 * information available. We might end up going slightly under
1642 * or over the suggested heap size, but we should be pretty
1645 * Formula: suggested - needed
1646 * ----------------------------
1647 * 1 + g0s0_pcnt_kept/100
1649 * where 'needed' is the amount of memory needed at the next
1650 * collection for collecting all steps except g0s0.
1653 (((long)RtsFlags.GcFlags.heapSizeSuggestion - (long)needed) * 100) /
1654 (100 + (long)g0s0_pcnt_kept);
1656 if (blocks < (long)RtsFlags.GcFlags.minAllocAreaSize) {
1657 blocks = RtsFlags.GcFlags.minAllocAreaSize;
1660 resizeNurseries((nat)blocks);
1664 // we might have added extra large blocks to the nursery, so
1665 // resize back to minAllocAreaSize again.
1666 resizeNurseriesFixed(RtsFlags.GcFlags.minAllocAreaSize);
1671 /* -----------------------------------------------------------------------------
1672 Sanity code for CAF garbage collection.
1674 With DEBUG turned on, we manage a CAF list in addition to the SRT
1675 mechanism. After GC, we run down the CAF list and blackhole any
1676 CAFs which have been garbage collected. This means we get an error
1677 whenever the program tries to enter a garbage collected CAF.
1679 Any garbage collected CAFs are taken off the CAF list at the same
1681 -------------------------------------------------------------------------- */
1683 #if 0 && defined(DEBUG)
1690 const StgInfoTable *info;
1701 ASSERT(info->type == IND_STATIC);
1703 if (STATIC_LINK(info,p) == NULL) {
1704 debugTrace(DEBUG_gccafs, "CAF gc'd at 0x%04lx", (long)p);
1706 SET_INFO(p,&stg_BLACKHOLE_info);
1707 p = STATIC_LINK2(info,p);
1711 pp = &STATIC_LINK2(info,p);
1718 debugTrace(DEBUG_gccafs, "%d CAFs live", i);