1 /* -----------------------------------------------------------------------------
2 * $Id: Storage.c,v 1.53 2001/11/08 12:46:31 simonmar Exp $
4 * (c) The GHC Team, 1998-1999
6 * Storage manager front end
8 * ---------------------------------------------------------------------------*/
10 #include "PosixSource.h"
16 #include "BlockAlloc.h"
24 #include "StoragePriv.h"
26 StgClosure *caf_list = NULL;
28 bdescr *small_alloc_list; /* allocate()d small objects */
29 bdescr *large_alloc_list; /* allocate()d large objects */
30 bdescr *pinned_object_block; /* allocate pinned objects into this block */
31 nat alloc_blocks; /* number of allocate()d blocks since GC */
32 nat alloc_blocks_lim; /* approximate limit on alloc_blocks */
34 StgPtr alloc_Hp = NULL; /* next free byte in small_alloc_list */
35 StgPtr alloc_HpLim = NULL; /* end of block at small_alloc_list */
37 generation *generations; /* all the generations */
38 generation *g0; /* generation 0, for convenience */
39 generation *oldest_gen; /* oldest generation, for convenience */
40 step *g0s0; /* generation 0, step 0, for convenience */
42 lnat total_allocated = 0; /* total memory allocated during run */
45 * Storage manager mutex: protects all the above state from
46 * simultaneous access by two STG threads.
49 pthread_mutex_t sm_mutex = PTHREAD_MUTEX_INITIALIZER;
55 static void *stgAllocForGMP (size_t size_in_bytes);
56 static void *stgReallocForGMP (void *ptr, size_t old_size, size_t new_size);
57 static void stgDeallocForGMP (void *ptr, size_t size);
66 /* If we're doing heap profiling, we want a two-space heap with a
67 * fixed-size allocation area so that we get roughly even-spaced
71 /* As an experiment, try a 2 generation collector
74 #if defined(PROFILING) || defined(DEBUG)
75 if (RtsFlags.ProfFlags.doHeapProfile) {
76 RtsFlags.GcFlags.generations = 1;
77 RtsFlags.GcFlags.steps = 1;
78 RtsFlags.GcFlags.oldGenFactor = 0;
79 RtsFlags.GcFlags.heapSizeSuggestion = 0;
83 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
84 RtsFlags.GcFlags.heapSizeSuggestion >
85 RtsFlags.GcFlags.maxHeapSize) {
86 RtsFlags.GcFlags.maxHeapSize = RtsFlags.GcFlags.heapSizeSuggestion;
89 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
90 RtsFlags.GcFlags.minAllocAreaSize >
91 RtsFlags.GcFlags.maxHeapSize) {
92 prog_belch("maximum heap size (-M) is smaller than minimum alloc area size (-A)");
98 /* allocate generation info array */
99 generations = (generation *)stgMallocBytes(RtsFlags.GcFlags.generations
100 * sizeof(struct _generation),
101 "initStorage: gens");
103 /* Initialise all generations */
104 for(g = 0; g < RtsFlags.GcFlags.generations; g++) {
105 gen = &generations[g];
107 gen->mut_list = END_MUT_LIST;
108 gen->mut_once_list = END_MUT_LIST;
109 gen->collections = 0;
110 gen->failed_promotions = 0;
114 /* A couple of convenience pointers */
115 g0 = &generations[0];
116 oldest_gen = &generations[RtsFlags.GcFlags.generations-1];
118 /* Allocate step structures in each generation */
119 if (RtsFlags.GcFlags.generations > 1) {
120 /* Only for multiple-generations */
122 /* Oldest generation: one step */
123 oldest_gen->n_steps = 1;
125 stgMallocBytes(1 * sizeof(struct _step), "initStorage: last step");
127 /* set up all except the oldest generation with 2 steps */
128 for(g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
129 generations[g].n_steps = RtsFlags.GcFlags.steps;
130 generations[g].steps =
131 stgMallocBytes (RtsFlags.GcFlags.steps * sizeof(struct _step),
132 "initStorage: steps");
136 /* single generation, i.e. a two-space collector */
138 g0->steps = stgMallocBytes (sizeof(struct _step), "initStorage: steps");
141 /* Initialise all steps */
142 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
143 for (s = 0; s < generations[g].n_steps; s++) {
144 stp = &generations[g].steps[s];
148 stp->gen = &generations[g];
155 stp->large_objects = NULL;
156 stp->n_large_blocks = 0;
157 stp->new_large_objects = NULL;
158 stp->scavenged_large_objects = NULL;
159 stp->n_scavenged_large_blocks = 0;
160 stp->is_compacted = 0;
165 /* Set up the destination pointers in each younger gen. step */
166 for (g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
167 for (s = 0; s < generations[g].n_steps-1; s++) {
168 generations[g].steps[s].to = &generations[g].steps[s+1];
170 generations[g].steps[s].to = &generations[g+1].steps[0];
173 /* The oldest generation has one step and it is compacted. */
174 if (RtsFlags.GcFlags.compact) {
175 if (RtsFlags.GcFlags.generations == 1) {
176 belch("WARNING: compaction is incompatible with -G1; disabled");
178 oldest_gen->steps[0].is_compacted = 1;
181 oldest_gen->steps[0].to = &oldest_gen->steps[0];
183 /* generation 0 is special: that's the nursery */
184 generations[0].max_blocks = 0;
186 /* G0S0: the allocation area. Policy: keep the allocation area
187 * small to begin with, even if we have a large suggested heap
188 * size. Reason: we're going to do a major collection first, and we
189 * don't want it to be a big one. This vague idea is borne out by
190 * rigorous experimental evidence.
192 g0s0 = &generations[0].steps[0];
196 weak_ptr_list = NULL;
199 /* initialise the allocate() interface */
200 small_alloc_list = NULL;
201 large_alloc_list = NULL;
203 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
205 /* Tell GNU multi-precision pkg about our custom alloc functions */
206 mp_set_memory_functions(stgAllocForGMP, stgReallocForGMP, stgDeallocForGMP);
209 pthread_mutex_init(&sm_mutex, NULL);
212 IF_DEBUG(gc, statDescribeGens());
218 stat_exit(calcAllocated());
221 /* -----------------------------------------------------------------------------
224 The entry code for every CAF does the following:
226 - builds a CAF_BLACKHOLE in the heap
227 - pushes an update frame pointing to the CAF_BLACKHOLE
228 - invokes UPD_CAF(), which:
229 - calls newCaf, below
230 - updates the CAF with a static indirection to the CAF_BLACKHOLE
232 Why do we build a BLACKHOLE in the heap rather than just updating
233 the thunk directly? It's so that we only need one kind of update
234 frame - otherwise we'd need a static version of the update frame too.
236 newCaf() does the following:
238 - it puts the CAF on the oldest generation's mut-once list.
239 This is so that we can treat the CAF as a root when collecting
242 For GHCI, we have additional requirements when dealing with CAFs:
244 - we must *retain* all dynamically-loaded CAFs ever entered,
245 just in case we need them again.
246 - we must be able to *revert* CAFs that have been evaluated, to
247 their pre-evaluated form.
249 To do this, we use an additional CAF list. When newCaf() is
250 called on a dynamically-loaded CAF, we add it to the CAF list
251 instead of the old-generation mutable list, and save away its
252 old info pointer (in caf->saved_info) for later reversion.
254 To revert all the CAFs, we traverse the CAF list and reset the
255 info pointer to caf->saved_info, then throw away the CAF list.
256 (see GC.c:revertCAFs()).
260 -------------------------------------------------------------------------- */
263 newCAF(StgClosure* caf)
265 /* Put this CAF on the mutable list for the old generation.
266 * This is a HACK - the IND_STATIC closure doesn't really have
267 * a mut_link field, but we pretend it has - in fact we re-use
268 * the STATIC_LINK field for the time being, because when we
269 * come to do a major GC we won't need the mut_link field
270 * any more and can use it as a STATIC_LINK.
272 ACQUIRE_LOCK(&sm_mutex);
274 if (is_dynamically_loaded_rwdata_ptr((StgPtr)caf)) {
275 ((StgIndStatic *)caf)->saved_info = (StgInfoTable *)caf->header.info;
276 ((StgIndStatic *)caf)->static_link = caf_list;
279 ((StgIndStatic *)caf)->saved_info = NULL;
280 ((StgMutClosure *)caf)->mut_link = oldest_gen->mut_once_list;
281 oldest_gen->mut_once_list = (StgMutClosure *)caf;
284 RELEASE_LOCK(&sm_mutex);
287 /* If we are PAR or DIST then we never forget a CAF */
289 //belch("<##> Globalising CAF %08x %s",caf,info_type(caf));
290 newGA=makeGlobal(caf,rtsTrue); /*given full weight*/
296 /* -----------------------------------------------------------------------------
298 -------------------------------------------------------------------------- */
301 allocNurseries( void )
310 for (cap = free_capabilities; cap != NULL; cap = cap->link) {
311 cap->rNursery = allocNursery(NULL, RtsFlags.GcFlags.minAllocAreaSize);
312 cap->rCurrentNursery = cap->rNursery;
313 for (bd = cap->rNursery; bd != NULL; bd = bd->link) {
314 bd->u.back = (bdescr *)cap;
317 /* Set the back links to be equal to the Capability,
318 * so we can do slightly better informed locking.
322 g0s0->blocks = allocNursery(NULL, RtsFlags.GcFlags.minAllocAreaSize);
323 g0s0->n_blocks = RtsFlags.GcFlags.minAllocAreaSize;
324 g0s0->to_blocks = NULL;
325 g0s0->n_to_blocks = 0;
326 MainCapability.r.rNursery = g0s0->blocks;
327 MainCapability.r.rCurrentNursery = g0s0->blocks;
328 /* hp, hpLim, hp_bd, to_space etc. aren't used in G0S0 */
333 resetNurseries( void )
339 /* All tasks must be stopped */
340 ASSERT(n_free_capabilities == RtsFlags.ParFlags.nNodes);
342 for (cap = free_capabilities; cap != NULL; cap = cap->link) {
343 for (bd = cap->rNursery; bd; bd = bd->link) {
344 bd->free = bd->start;
345 ASSERT(bd->gen_no == 0);
346 ASSERT(bd->step == g0s0);
347 IF_DEBUG(sanity,memset(bd->start, 0xaa, BLOCK_SIZE));
349 cap->rCurrentNursery = cap->rNursery;
352 for (bd = g0s0->blocks; bd; bd = bd->link) {
353 bd->free = bd->start;
354 ASSERT(bd->gen_no == 0);
355 ASSERT(bd->step == g0s0);
356 IF_DEBUG(sanity,memset(bd->start, 0xaa, BLOCK_SIZE));
358 MainCapability.r.rNursery = g0s0->blocks;
359 MainCapability.r.rCurrentNursery = g0s0->blocks;
364 allocNursery (bdescr *tail, nat blocks)
369 // Allocate a nursery: we allocate fresh blocks one at a time and
370 // cons them on to the front of the list, not forgetting to update
371 // the back pointer on the tail of the list to point to the new block.
372 for (i=0; i < blocks; i++) {
375 // double-link the nursery: we might need to insert blocks
382 bd->free = bd->start;
390 resizeNursery ( nat blocks )
396 barf("resizeNursery: can't resize in SMP mode");
399 nursery_blocks = g0s0->n_blocks;
400 if (nursery_blocks == blocks) {
404 else if (nursery_blocks < blocks) {
405 IF_DEBUG(gc, fprintf(stderr, "Increasing size of nursery to %d blocks\n",
407 g0s0->blocks = allocNursery(g0s0->blocks, blocks-nursery_blocks);
413 IF_DEBUG(gc, fprintf(stderr, "Decreasing size of nursery to %d blocks\n",
417 while (nursery_blocks > blocks) {
419 next_bd->u.back = NULL;
420 nursery_blocks -= bd->blocks; // might be a large block
425 // might have gone just under, by freeing a large block, so make
426 // up the difference.
427 if (nursery_blocks < blocks) {
428 g0s0->blocks = allocNursery(g0s0->blocks, blocks-nursery_blocks);
432 g0s0->n_blocks = blocks;
433 ASSERT(countBlocks(g0s0->blocks) == g0s0->n_blocks);
436 /* -----------------------------------------------------------------------------
437 The allocate() interface
439 allocate(n) always succeeds, and returns a chunk of memory n words
440 long. n can be larger than the size of a block if necessary, in
441 which case a contiguous block group will be allocated.
442 -------------------------------------------------------------------------- */
450 ACQUIRE_LOCK(&sm_mutex);
452 TICK_ALLOC_HEAP_NOCTR(n);
455 /* big allocation (>LARGE_OBJECT_THRESHOLD) */
456 /* ToDo: allocate directly into generation 1 */
457 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
458 nat req_blocks = (lnat)BLOCK_ROUND_UP(n*sizeof(W_)) / BLOCK_SIZE;
459 bd = allocGroup(req_blocks);
460 dbl_link_onto(bd, &g0s0->large_objects);
463 bd->flags = BF_LARGE;
464 bd->free = bd->start;
465 /* don't add these blocks to alloc_blocks, since we're assuming
466 * that large objects are likely to remain live for quite a while
467 * (eg. running threads), so garbage collecting early won't make
470 alloc_blocks += req_blocks;
471 RELEASE_LOCK(&sm_mutex);
474 /* small allocation (<LARGE_OBJECT_THRESHOLD) */
475 } else if (small_alloc_list == NULL || alloc_Hp + n > alloc_HpLim) {
476 if (small_alloc_list) {
477 small_alloc_list->free = alloc_Hp;
480 bd->link = small_alloc_list;
481 small_alloc_list = bd;
485 alloc_Hp = bd->start;
486 alloc_HpLim = bd->start + BLOCK_SIZE_W;
492 RELEASE_LOCK(&sm_mutex);
497 allocated_bytes( void )
499 return (alloc_blocks * BLOCK_SIZE_W - (alloc_HpLim - alloc_Hp));
502 /* ---------------------------------------------------------------------------
503 Allocate a fixed/pinned object.
505 We allocate small pinned objects into a single block, allocating a
506 new block when the current one overflows. The block is chained
507 onto the large_object_list of generation 0 step 0.
509 NOTE: The GC can't in general handle pinned objects. This
510 interface is only safe to use for ByteArrays, which have no
511 pointers and don't require scavenging. It works because the
512 block's descriptor has the BF_LARGE flag set, so the block is
513 treated as a large object and chained onto various lists, rather
514 than the individual objects being copied. However, when it comes
515 to scavenge the block, the GC will only scavenge the first object.
516 The reason is that the GC can't linearly scan a block of pinned
517 objects at the moment (doing so would require using the
518 mostly-copying techniques). But since we're restricting ourselves
519 to pinned ByteArrays, not scavenging is ok.
521 This function is called by newPinnedByteArray# which immediately
522 fills the allocated memory with a MutableByteArray#.
523 ------------------------------------------------------------------------- */
526 allocatePinned( nat n )
529 bdescr *bd = pinned_object_block;
531 ACQUIRE_LOCK(&sm_mutex);
533 TICK_ALLOC_HEAP_NOCTR(n);
536 // If the request is for a large object, then allocate()
537 // will give us a pinned object anyway.
538 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
539 RELEASE_LOCK(&sm_mutex);
543 // If we don't have a block of pinned objects yet, or the current
544 // one isn't large enough to hold the new object, allocate a new one.
545 if (bd == NULL || (bd->free + n) > (bd->start + BLOCK_SIZE_W)) {
546 pinned_object_block = bd = allocBlock();
547 dbl_link_onto(bd, &g0s0->large_objects);
550 bd->flags = BF_LARGE;
551 bd->free = bd->start;
557 RELEASE_LOCK(&sm_mutex);
561 /* -----------------------------------------------------------------------------
562 Allocation functions for GMP.
564 These all use the allocate() interface - we can't have any garbage
565 collection going on during a gmp operation, so we use allocate()
566 which always succeeds. The gmp operations which might need to
567 allocate will ask the storage manager (via doYouWantToGC()) whether
568 a garbage collection is required, in case we get into a loop doing
569 only allocate() style allocation.
570 -------------------------------------------------------------------------- */
573 stgAllocForGMP (size_t size_in_bytes)
576 nat data_size_in_words, total_size_in_words;
578 /* should be a multiple of sizeof(StgWord) (whole no. of limbs) */
579 ASSERT(size_in_bytes % sizeof(W_) == 0);
581 data_size_in_words = size_in_bytes / sizeof(W_);
582 total_size_in_words = sizeofW(StgArrWords) + data_size_in_words;
584 /* allocate and fill it in. */
585 arr = (StgArrWords *)allocate(total_size_in_words);
586 SET_ARR_HDR(arr, &stg_ARR_WORDS_info, CCCS, data_size_in_words);
588 /* and return a ptr to the goods inside the array */
589 return(BYTE_ARR_CTS(arr));
593 stgReallocForGMP (void *ptr, size_t old_size, size_t new_size)
595 void *new_stuff_ptr = stgAllocForGMP(new_size);
597 char *p = (char *) ptr;
598 char *q = (char *) new_stuff_ptr;
600 for (; i < old_size; i++, p++, q++) {
604 return(new_stuff_ptr);
608 stgDeallocForGMP (void *ptr STG_UNUSED,
609 size_t size STG_UNUSED)
611 /* easy for us: the garbage collector does the dealloc'n */
614 /* -----------------------------------------------------------------------------
616 * -------------------------------------------------------------------------- */
618 /* -----------------------------------------------------------------------------
621 * Approximate how much we've allocated: number of blocks in the
622 * nursery + blocks allocated via allocate() - unused nusery blocks.
623 * This leaves a little slop at the end of each block, and doesn't
624 * take into account large objects (ToDo).
625 * -------------------------------------------------------------------------- */
628 calcAllocated( void )
636 /* All tasks must be stopped. Can't assert that all the
637 capabilities are owned by the scheduler, though: one or more
638 tasks might have been stopped while they were running (non-main)
640 /* ASSERT(n_free_capabilities == RtsFlags.ParFlags.nNodes); */
643 n_free_capabilities * RtsFlags.GcFlags.minAllocAreaSize * BLOCK_SIZE_W
646 for (cap = free_capabilities; cap != NULL; cap = cap->link) {
647 for ( bd = cap->rCurrentNursery->link; bd != NULL; bd = bd->link ) {
648 allocated -= BLOCK_SIZE_W;
650 if (cap->rCurrentNursery->free < cap->rCurrentNursery->start
652 allocated -= (cap->rCurrentNursery->start + BLOCK_SIZE_W)
653 - cap->rCurrentNursery->free;
658 bdescr *current_nursery = MainCapability.r.rCurrentNursery;
660 allocated = (g0s0->n_blocks * BLOCK_SIZE_W) + allocated_bytes();
661 for ( bd = current_nursery->link; bd != NULL; bd = bd->link ) {
662 allocated -= BLOCK_SIZE_W;
664 if (current_nursery->free < current_nursery->start + BLOCK_SIZE_W) {
665 allocated -= (current_nursery->start + BLOCK_SIZE_W)
666 - current_nursery->free;
670 total_allocated += allocated;
674 /* Approximate the amount of live data in the heap. To be called just
675 * after garbage collection (see GarbageCollect()).
684 if (RtsFlags.GcFlags.generations == 1) {
685 live = (g0s0->n_to_blocks - 1) * BLOCK_SIZE_W +
686 ((lnat)g0s0->hp_bd->free - (lnat)g0s0->hp_bd->start) / sizeof(W_);
690 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
691 for (s = 0; s < generations[g].n_steps; s++) {
692 /* approximate amount of live data (doesn't take into account slop
693 * at end of each block).
695 if (g == 0 && s == 0) {
698 stp = &generations[g].steps[s];
699 live += (stp->n_large_blocks + stp->n_blocks - 1) * BLOCK_SIZE_W;
700 if (stp->hp_bd != NULL) {
701 live += ((lnat)stp->hp_bd->free - (lnat)stp->hp_bd->start)
709 /* Approximate the number of blocks that will be needed at the next
710 * garbage collection.
712 * Assume: all data currently live will remain live. Steps that will
713 * be collected next time will therefore need twice as many blocks
714 * since all the data will be copied.
723 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
724 for (s = 0; s < generations[g].n_steps; s++) {
725 if (g == 0 && s == 0) { continue; }
726 stp = &generations[g].steps[s];
727 if (generations[g].steps[0].n_blocks +
728 generations[g].steps[0].n_large_blocks
729 > generations[g].max_blocks
730 && stp->is_compacted == 0) {
731 needed += 2 * stp->n_blocks;
733 needed += stp->n_blocks;
740 /* -----------------------------------------------------------------------------
743 memInventory() checks for memory leaks by counting up all the
744 blocks we know about and comparing that to the number of blocks
745 allegedly floating around in the system.
746 -------------------------------------------------------------------------- */
756 lnat total_blocks = 0, free_blocks = 0;
758 /* count the blocks we current have */
760 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
761 for (s = 0; s < generations[g].n_steps; s++) {
762 stp = &generations[g].steps[s];
763 total_blocks += stp->n_blocks;
764 if (RtsFlags.GcFlags.generations == 1) {
765 /* two-space collector has a to-space too :-) */
766 total_blocks += g0s0->n_to_blocks;
768 for (bd = stp->large_objects; bd; bd = bd->link) {
769 total_blocks += bd->blocks;
770 /* hack for megablock groups: they have an extra block or two in
771 the second and subsequent megablocks where the block
772 descriptors would normally go.
774 if (bd->blocks > BLOCKS_PER_MBLOCK) {
775 total_blocks -= (MBLOCK_SIZE / BLOCK_SIZE - BLOCKS_PER_MBLOCK)
776 * (bd->blocks/(MBLOCK_SIZE/BLOCK_SIZE));
782 /* any blocks held by allocate() */
783 for (bd = small_alloc_list; bd; bd = bd->link) {
784 total_blocks += bd->blocks;
786 for (bd = large_alloc_list; bd; bd = bd->link) {
787 total_blocks += bd->blocks;
790 // count the blocks allocated by the arena allocator
791 total_blocks += arenaBlocks();
793 /* count the blocks on the free list */
794 free_blocks = countFreeList();
796 if (total_blocks + free_blocks != mblocks_allocated *
798 fprintf(stderr, "Blocks: %ld live + %ld free = %ld total (%ld around)\n",
799 total_blocks, free_blocks, total_blocks + free_blocks,
800 mblocks_allocated * BLOCKS_PER_MBLOCK);
803 ASSERT(total_blocks + free_blocks == mblocks_allocated * BLOCKS_PER_MBLOCK);
808 countBlocks(bdescr *bd)
811 for (n=0; bd != NULL; bd=bd->link) {
817 /* Full heap sanity check. */
823 if (RtsFlags.GcFlags.generations == 1) {
824 checkHeap(g0s0->to_blocks);
825 checkChain(g0s0->large_objects);
828 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
829 for (s = 0; s < generations[g].n_steps; s++) {
830 ASSERT(countBlocks(generations[g].steps[s].blocks)
831 == generations[g].steps[s].n_blocks);
832 ASSERT(countBlocks(generations[g].steps[s].large_objects)
833 == generations[g].steps[s].n_large_blocks);
834 if (g == 0 && s == 0) { continue; }
835 checkHeap(generations[g].steps[s].blocks);
836 checkChain(generations[g].steps[s].large_objects);
838 checkMutableList(generations[g].mut_list, g);
839 checkMutOnceList(generations[g].mut_once_list, g);
843 checkFreeListSanity();