1 /* -----------------------------------------------------------------------------
2 * $Id: Storage.c,v 1.52 2001/10/18 14:41:01 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"
27 nat nursery_blocks; /* number of blocks in the nursery */
30 StgClosure *caf_list = NULL;
32 bdescr *small_alloc_list; /* allocate()d small objects */
33 bdescr *large_alloc_list; /* allocate()d large objects */
34 bdescr *pinned_object_block; /* allocate pinned objects into this block */
35 nat alloc_blocks; /* number of allocate()d blocks since GC */
36 nat alloc_blocks_lim; /* approximate limit on alloc_blocks */
38 StgPtr alloc_Hp = NULL; /* next free byte in small_alloc_list */
39 StgPtr alloc_HpLim = NULL; /* end of block at small_alloc_list */
41 generation *generations; /* all the generations */
42 generation *g0; /* generation 0, for convenience */
43 generation *oldest_gen; /* oldest generation, for convenience */
44 step *g0s0; /* generation 0, step 0, for convenience */
46 lnat total_allocated = 0; /* total memory allocated during run */
49 * Storage manager mutex: protects all the above state from
50 * simultaneous access by two STG threads.
53 pthread_mutex_t sm_mutex = PTHREAD_MUTEX_INITIALIZER;
59 static void *stgAllocForGMP (size_t size_in_bytes);
60 static void *stgReallocForGMP (void *ptr, size_t old_size, size_t new_size);
61 static void stgDeallocForGMP (void *ptr, size_t size);
70 /* If we're doing heap profiling, we want a two-space heap with a
71 * fixed-size allocation area so that we get roughly even-spaced
75 /* As an experiment, try a 2 generation collector
78 #if defined(PROFILING) || defined(DEBUG)
79 if (RtsFlags.ProfFlags.doHeapProfile) {
80 RtsFlags.GcFlags.generations = 1;
81 RtsFlags.GcFlags.steps = 1;
82 RtsFlags.GcFlags.oldGenFactor = 0;
83 RtsFlags.GcFlags.heapSizeSuggestion = 0;
87 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
88 RtsFlags.GcFlags.heapSizeSuggestion >
89 RtsFlags.GcFlags.maxHeapSize) {
90 RtsFlags.GcFlags.maxHeapSize = RtsFlags.GcFlags.heapSizeSuggestion;
93 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
94 RtsFlags.GcFlags.minAllocAreaSize >
95 RtsFlags.GcFlags.maxHeapSize) {
96 prog_belch("maximum heap size (-M) is smaller than minimum alloc area size (-A)");
100 initBlockAllocator();
102 /* allocate generation info array */
103 generations = (generation *)stgMallocBytes(RtsFlags.GcFlags.generations
104 * sizeof(struct _generation),
105 "initStorage: gens");
107 /* Initialise all generations */
108 for(g = 0; g < RtsFlags.GcFlags.generations; g++) {
109 gen = &generations[g];
111 gen->mut_list = END_MUT_LIST;
112 gen->mut_once_list = END_MUT_LIST;
113 gen->collections = 0;
114 gen->failed_promotions = 0;
118 /* A couple of convenience pointers */
119 g0 = &generations[0];
120 oldest_gen = &generations[RtsFlags.GcFlags.generations-1];
122 /* Allocate step structures in each generation */
123 if (RtsFlags.GcFlags.generations > 1) {
124 /* Only for multiple-generations */
126 /* Oldest generation: one step */
127 oldest_gen->n_steps = 1;
129 stgMallocBytes(1 * sizeof(struct _step), "initStorage: last step");
131 /* set up all except the oldest generation with 2 steps */
132 for(g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
133 generations[g].n_steps = RtsFlags.GcFlags.steps;
134 generations[g].steps =
135 stgMallocBytes (RtsFlags.GcFlags.steps * sizeof(struct _step),
136 "initStorage: steps");
140 /* single generation, i.e. a two-space collector */
142 g0->steps = stgMallocBytes (sizeof(struct _step), "initStorage: steps");
145 /* Initialise all steps */
146 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
147 for (s = 0; s < generations[g].n_steps; s++) {
148 stp = &generations[g].steps[s];
152 stp->gen = &generations[g];
159 stp->large_objects = NULL;
160 stp->n_large_blocks = 0;
161 stp->new_large_objects = NULL;
162 stp->scavenged_large_objects = NULL;
163 stp->n_scavenged_large_blocks = 0;
164 stp->is_compacted = 0;
169 /* Set up the destination pointers in each younger gen. step */
170 for (g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
171 for (s = 0; s < generations[g].n_steps-1; s++) {
172 generations[g].steps[s].to = &generations[g].steps[s+1];
174 generations[g].steps[s].to = &generations[g+1].steps[0];
177 /* The oldest generation has one step and it is compacted. */
178 if (RtsFlags.GcFlags.compact) {
179 if (RtsFlags.GcFlags.generations == 1) {
180 belch("WARNING: compaction is incompatible with -G1; disabled");
182 oldest_gen->steps[0].is_compacted = 1;
185 oldest_gen->steps[0].to = &oldest_gen->steps[0];
187 /* generation 0 is special: that's the nursery */
188 generations[0].max_blocks = 0;
190 /* G0S0: the allocation area. Policy: keep the allocation area
191 * small to begin with, even if we have a large suggested heap
192 * size. Reason: we're going to do a major collection first, and we
193 * don't want it to be a big one. This vague idea is borne out by
194 * rigorous experimental evidence.
196 g0s0 = &generations[0].steps[0];
200 weak_ptr_list = NULL;
203 /* initialise the allocate() interface */
204 small_alloc_list = NULL;
205 large_alloc_list = NULL;
207 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
209 /* Tell GNU multi-precision pkg about our custom alloc functions */
210 mp_set_memory_functions(stgAllocForGMP, stgReallocForGMP, stgDeallocForGMP);
213 pthread_mutex_init(&sm_mutex, NULL);
216 IF_DEBUG(gc, statDescribeGens());
222 stat_exit(calcAllocated());
225 /* -----------------------------------------------------------------------------
228 The entry code for every CAF does the following:
230 - builds a CAF_BLACKHOLE in the heap
231 - pushes an update frame pointing to the CAF_BLACKHOLE
232 - invokes UPD_CAF(), which:
233 - calls newCaf, below
234 - updates the CAF with a static indirection to the CAF_BLACKHOLE
236 Why do we build a BLACKHOLE in the heap rather than just updating
237 the thunk directly? It's so that we only need one kind of update
238 frame - otherwise we'd need a static version of the update frame too.
240 newCaf() does the following:
242 - it puts the CAF on the oldest generation's mut-once list.
243 This is so that we can treat the CAF as a root when collecting
246 For GHCI, we have additional requirements when dealing with CAFs:
248 - we must *retain* all dynamically-loaded CAFs ever entered,
249 just in case we need them again.
250 - we must be able to *revert* CAFs that have been evaluated, to
251 their pre-evaluated form.
253 To do this, we use an additional CAF list. When newCaf() is
254 called on a dynamically-loaded CAF, we add it to the CAF list
255 instead of the old-generation mutable list, and save away its
256 old info pointer (in caf->saved_info) for later reversion.
258 To revert all the CAFs, we traverse the CAF list and reset the
259 info pointer to caf->saved_info, then throw away the CAF list.
260 (see GC.c:revertCAFs()).
264 -------------------------------------------------------------------------- */
267 newCAF(StgClosure* caf)
269 /* Put this CAF on the mutable list for the old generation.
270 * This is a HACK - the IND_STATIC closure doesn't really have
271 * a mut_link field, but we pretend it has - in fact we re-use
272 * the STATIC_LINK field for the time being, because when we
273 * come to do a major GC we won't need the mut_link field
274 * any more and can use it as a STATIC_LINK.
276 ACQUIRE_LOCK(&sm_mutex);
278 if (is_dynamically_loaded_rwdata_ptr((StgPtr)caf)) {
279 ((StgIndStatic *)caf)->saved_info = (StgInfoTable *)caf->header.info;
280 ((StgIndStatic *)caf)->static_link = caf_list;
283 ((StgIndStatic *)caf)->saved_info = NULL;
284 ((StgMutClosure *)caf)->mut_link = oldest_gen->mut_once_list;
285 oldest_gen->mut_once_list = (StgMutClosure *)caf;
288 RELEASE_LOCK(&sm_mutex);
291 /* If we are PAR or DIST then we never forget a CAF */
293 //belch("<##> Globalising CAF %08x %s",caf,info_type(caf));
294 newGA=makeGlobal(caf,rtsTrue); /*given full weight*/
300 /* -----------------------------------------------------------------------------
302 -------------------------------------------------------------------------- */
305 allocNurseries( void )
314 for (cap = free_capabilities; cap != NULL; cap = cap->link) {
315 cap->rNursery = allocNursery(NULL, RtsFlags.GcFlags.minAllocAreaSize);
316 cap->rCurrentNursery = cap->rNursery;
317 for (bd = cap->rNursery; bd != NULL; bd = bd->link) {
318 bd->u.back = (bdescr *)cap;
321 /* Set the back links to be equal to the Capability,
322 * so we can do slightly better informed locking.
326 nursery_blocks = RtsFlags.GcFlags.minAllocAreaSize;
327 g0s0->blocks = allocNursery(NULL, nursery_blocks);
328 g0s0->n_blocks = nursery_blocks;
329 g0s0->to_blocks = NULL;
330 g0s0->n_to_blocks = 0;
331 MainRegTable.rNursery = g0s0->blocks;
332 MainRegTable.rCurrentNursery = g0s0->blocks;
333 /* hp, hpLim, hp_bd, to_space etc. aren't used in G0S0 */
338 resetNurseries( void )
344 /* All tasks must be stopped */
345 ASSERT(n_free_capabilities == RtsFlags.ParFlags.nNodes);
347 for (cap = free_capabilities; cap != NULL; cap = cap->link) {
348 for (bd = cap->rNursery; bd; bd = bd->link) {
349 bd->free = bd->start;
350 ASSERT(bd->gen_no == 0);
351 ASSERT(bd->step == g0s0);
352 IF_DEBUG(sanity,memset(bd->start, 0xaa, BLOCK_SIZE));
354 cap->rCurrentNursery = cap->rNursery;
357 for (bd = g0s0->blocks; bd; bd = bd->link) {
358 bd->free = bd->start;
359 ASSERT(bd->gen_no == 0);
360 ASSERT(bd->step == g0s0);
361 IF_DEBUG(sanity,memset(bd->start, 0xaa, BLOCK_SIZE));
363 MainRegTable.rNursery = g0s0->blocks;
364 MainRegTable.rCurrentNursery = g0s0->blocks;
369 allocNursery (bdescr *last_bd, nat blocks)
374 /* Allocate a nursery */
375 for (i=0; i < blocks; i++) {
381 bd->free = bd->start;
388 resizeNursery ( nat blocks )
393 barf("resizeNursery: can't resize in SMP mode");
396 if (nursery_blocks == blocks) {
397 ASSERT(g0s0->n_blocks == blocks);
401 else if (nursery_blocks < blocks) {
402 IF_DEBUG(gc, fprintf(stderr, "Increasing size of nursery to %d blocks\n",
404 g0s0->blocks = allocNursery(g0s0->blocks, blocks-nursery_blocks);
410 IF_DEBUG(gc, fprintf(stderr, "Decreasing size of nursery to %d blocks\n",
412 for (bd = g0s0->blocks; nursery_blocks > blocks; nursery_blocks--) {
420 g0s0->n_blocks = nursery_blocks = blocks;
423 /* -----------------------------------------------------------------------------
424 The allocate() interface
426 allocate(n) always succeeds, and returns a chunk of memory n words
427 long. n can be larger than the size of a block if necessary, in
428 which case a contiguous block group will be allocated.
429 -------------------------------------------------------------------------- */
437 ACQUIRE_LOCK(&sm_mutex);
439 TICK_ALLOC_HEAP_NOCTR(n);
442 /* big allocation (>LARGE_OBJECT_THRESHOLD) */
443 /* ToDo: allocate directly into generation 1 */
444 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
445 nat req_blocks = (lnat)BLOCK_ROUND_UP(n*sizeof(W_)) / BLOCK_SIZE;
446 bd = allocGroup(req_blocks);
447 dbl_link_onto(bd, &g0s0->large_objects);
450 bd->flags = BF_LARGE;
451 bd->free = bd->start;
452 /* don't add these blocks to alloc_blocks, since we're assuming
453 * that large objects are likely to remain live for quite a while
454 * (eg. running threads), so garbage collecting early won't make
457 alloc_blocks += req_blocks;
458 RELEASE_LOCK(&sm_mutex);
461 /* small allocation (<LARGE_OBJECT_THRESHOLD) */
462 } else if (small_alloc_list == NULL || alloc_Hp + n > alloc_HpLim) {
463 if (small_alloc_list) {
464 small_alloc_list->free = alloc_Hp;
467 bd->link = small_alloc_list;
468 small_alloc_list = bd;
472 alloc_Hp = bd->start;
473 alloc_HpLim = bd->start + BLOCK_SIZE_W;
479 RELEASE_LOCK(&sm_mutex);
484 allocated_bytes( void )
486 return (alloc_blocks * BLOCK_SIZE_W - (alloc_HpLim - alloc_Hp));
489 /* ---------------------------------------------------------------------------
490 Allocate a fixed/pinned object.
492 We allocate small pinned objects into a single block, allocating a
493 new block when the current one overflows. The block is chained
494 onto the large_object_list of generation 0 step 0.
496 NOTE: The GC can't in general handle pinned objects. This
497 interface is only safe to use for ByteArrays, which have no
498 pointers and don't require scavenging. It works because the
499 block's descriptor has the BF_LARGE flag set, so the block is
500 treated as a large object and chained onto various lists, rather
501 than the individual objects being copied. However, when it comes
502 to scavenge the block, the GC will only scavenge the first object.
503 The reason is that the GC can't linearly scan a block of pinned
504 objects at the moment (doing so would require using the
505 mostly-copying techniques). But since we're restricting ourselves
506 to pinned ByteArrays, not scavenging is ok.
508 This function is called by newPinnedByteArray# which immediately
509 fills the allocated memory with a MutableByteArray#.
510 ------------------------------------------------------------------------- */
513 allocatePinned( nat n )
516 bdescr *bd = pinned_object_block;
518 ACQUIRE_LOCK(&sm_mutex);
520 TICK_ALLOC_HEAP_NOCTR(n);
523 // If the request is for a large object, then allocate()
524 // will give us a pinned object anyway.
525 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
526 RELEASE_LOCK(&sm_mutex);
530 // If we don't have a block of pinned objects yet, or the current
531 // one isn't large enough to hold the new object, allocate a new one.
532 if (bd == NULL || (bd->free + n) > (bd->start + BLOCK_SIZE_W)) {
533 pinned_object_block = bd = allocBlock();
534 dbl_link_onto(bd, &g0s0->large_objects);
537 bd->flags = BF_LARGE;
538 bd->free = bd->start;
544 RELEASE_LOCK(&sm_mutex);
548 /* -----------------------------------------------------------------------------
549 Allocation functions for GMP.
551 These all use the allocate() interface - we can't have any garbage
552 collection going on during a gmp operation, so we use allocate()
553 which always succeeds. The gmp operations which might need to
554 allocate will ask the storage manager (via doYouWantToGC()) whether
555 a garbage collection is required, in case we get into a loop doing
556 only allocate() style allocation.
557 -------------------------------------------------------------------------- */
560 stgAllocForGMP (size_t size_in_bytes)
563 nat data_size_in_words, total_size_in_words;
565 /* should be a multiple of sizeof(StgWord) (whole no. of limbs) */
566 ASSERT(size_in_bytes % sizeof(W_) == 0);
568 data_size_in_words = size_in_bytes / sizeof(W_);
569 total_size_in_words = sizeofW(StgArrWords) + data_size_in_words;
571 /* allocate and fill it in. */
572 arr = (StgArrWords *)allocate(total_size_in_words);
573 SET_ARR_HDR(arr, &stg_ARR_WORDS_info, CCCS, data_size_in_words);
575 /* and return a ptr to the goods inside the array */
576 return(BYTE_ARR_CTS(arr));
580 stgReallocForGMP (void *ptr, size_t old_size, size_t new_size)
582 void *new_stuff_ptr = stgAllocForGMP(new_size);
584 char *p = (char *) ptr;
585 char *q = (char *) new_stuff_ptr;
587 for (; i < old_size; i++, p++, q++) {
591 return(new_stuff_ptr);
595 stgDeallocForGMP (void *ptr STG_UNUSED,
596 size_t size STG_UNUSED)
598 /* easy for us: the garbage collector does the dealloc'n */
601 /* -----------------------------------------------------------------------------
603 * -------------------------------------------------------------------------- */
605 /* -----------------------------------------------------------------------------
608 * Approximate how much we've allocated: number of blocks in the
609 * nursery + blocks allocated via allocate() - unused nusery blocks.
610 * This leaves a little slop at the end of each block, and doesn't
611 * take into account large objects (ToDo).
612 * -------------------------------------------------------------------------- */
615 calcAllocated( void )
623 /* All tasks must be stopped. Can't assert that all the
624 capabilities are owned by the scheduler, though: one or more
625 tasks might have been stopped while they were running (non-main)
627 /* ASSERT(n_free_capabilities == RtsFlags.ParFlags.nNodes); */
630 n_free_capabilities * RtsFlags.GcFlags.minAllocAreaSize * BLOCK_SIZE_W
633 for (cap = free_capabilities; cap != NULL; cap = cap->link) {
634 for ( bd = cap->rCurrentNursery->link; bd != NULL; bd = bd->link ) {
635 allocated -= BLOCK_SIZE_W;
637 if (cap->rCurrentNursery->free < cap->rCurrentNursery->start
639 allocated -= (cap->rCurrentNursery->start + BLOCK_SIZE_W)
640 - cap->rCurrentNursery->free;
645 bdescr *current_nursery = MainRegTable.rCurrentNursery;
647 allocated = (nursery_blocks * BLOCK_SIZE_W) + allocated_bytes();
648 for ( bd = current_nursery->link; bd != NULL; bd = bd->link ) {
649 allocated -= BLOCK_SIZE_W;
651 if (current_nursery->free < current_nursery->start + BLOCK_SIZE_W) {
652 allocated -= (current_nursery->start + BLOCK_SIZE_W)
653 - current_nursery->free;
657 total_allocated += allocated;
661 /* Approximate the amount of live data in the heap. To be called just
662 * after garbage collection (see GarbageCollect()).
671 if (RtsFlags.GcFlags.generations == 1) {
672 live = (g0s0->n_to_blocks - 1) * BLOCK_SIZE_W +
673 ((lnat)g0s0->hp_bd->free - (lnat)g0s0->hp_bd->start) / sizeof(W_);
677 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
678 for (s = 0; s < generations[g].n_steps; s++) {
679 /* approximate amount of live data (doesn't take into account slop
680 * at end of each block).
682 if (g == 0 && s == 0) {
685 stp = &generations[g].steps[s];
686 live += (stp->n_large_blocks + stp->n_blocks - 1) * BLOCK_SIZE_W;
687 if (stp->hp_bd != NULL) {
688 live += ((lnat)stp->hp_bd->free - (lnat)stp->hp_bd->start)
696 /* Approximate the number of blocks that will be needed at the next
697 * garbage collection.
699 * Assume: all data currently live will remain live. Steps that will
700 * be collected next time will therefore need twice as many blocks
701 * since all the data will be copied.
710 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
711 for (s = 0; s < generations[g].n_steps; s++) {
712 if (g == 0 && s == 0) { continue; }
713 stp = &generations[g].steps[s];
714 if (generations[g].steps[0].n_blocks +
715 generations[g].steps[0].n_large_blocks
716 > generations[g].max_blocks
717 && stp->is_compacted == 0) {
718 needed += 2 * stp->n_blocks;
720 needed += stp->n_blocks;
727 /* -----------------------------------------------------------------------------
730 memInventory() checks for memory leaks by counting up all the
731 blocks we know about and comparing that to the number of blocks
732 allegedly floating around in the system.
733 -------------------------------------------------------------------------- */
743 lnat total_blocks = 0, free_blocks = 0;
745 /* count the blocks we current have */
747 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
748 for (s = 0; s < generations[g].n_steps; s++) {
749 stp = &generations[g].steps[s];
750 total_blocks += stp->n_blocks;
751 if (RtsFlags.GcFlags.generations == 1) {
752 /* two-space collector has a to-space too :-) */
753 total_blocks += g0s0->n_to_blocks;
755 for (bd = stp->large_objects; bd; bd = bd->link) {
756 total_blocks += bd->blocks;
757 /* hack for megablock groups: they have an extra block or two in
758 the second and subsequent megablocks where the block
759 descriptors would normally go.
761 if (bd->blocks > BLOCKS_PER_MBLOCK) {
762 total_blocks -= (MBLOCK_SIZE / BLOCK_SIZE - BLOCKS_PER_MBLOCK)
763 * (bd->blocks/(MBLOCK_SIZE/BLOCK_SIZE));
769 /* any blocks held by allocate() */
770 for (bd = small_alloc_list; bd; bd = bd->link) {
771 total_blocks += bd->blocks;
773 for (bd = large_alloc_list; bd; bd = bd->link) {
774 total_blocks += bd->blocks;
777 // count the blocks allocated by the arena allocator
778 total_blocks += arenaBlocks();
780 /* count the blocks on the free list */
781 free_blocks = countFreeList();
783 if (total_blocks + free_blocks != mblocks_allocated *
785 fprintf(stderr, "Blocks: %ld live + %ld free = %ld total (%ld around)\n",
786 total_blocks, free_blocks, total_blocks + free_blocks,
787 mblocks_allocated * BLOCKS_PER_MBLOCK);
790 ASSERT(total_blocks + free_blocks == mblocks_allocated * BLOCKS_PER_MBLOCK);
794 countBlocks(bdescr *bd)
797 for (n=0; bd != NULL; bd=bd->link) {
803 /* Full heap sanity check. */
809 if (RtsFlags.GcFlags.generations == 1) {
810 checkHeap(g0s0->to_blocks);
811 checkChain(g0s0->large_objects);
814 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
815 for (s = 0; s < generations[g].n_steps; s++) {
816 if (g == 0 && s == 0) { continue; }
817 checkHeap(generations[g].steps[s].blocks);
818 checkChain(generations[g].steps[s].large_objects);
819 ASSERT(countBlocks(generations[g].steps[s].blocks)
820 == generations[g].steps[s].n_blocks);
821 ASSERT(countBlocks(generations[g].steps[s].large_objects)
822 == generations[g].steps[s].n_large_blocks);
824 checkMutableList(generations[g].mut_list, g);
825 checkMutOnceList(generations[g].mut_once_list, g);
829 checkFreeListSanity();