1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 1998-2008
5 * Storage manager front end
7 * Documentation on the architecture of the Storage Manager can be
8 * found in the online commentary:
10 * http://hackage.haskell.org/trac/ghc/wiki/Commentary/Rts/Storage
12 * ---------------------------------------------------------------------------*/
14 #include "PosixSource.h"
20 #include "BlockAlloc.h"
25 #include "OSThreads.h"
26 #include "Capability.h"
29 #include "RetainerProfile.h" // for counting memory blocks (memInventory)
39 * All these globals require sm_mutex to access in THREADED_RTS mode.
41 StgClosure *caf_list = NULL;
42 StgClosure *revertible_caf_list = NULL;
45 bdescr *pinned_object_block; /* allocate pinned objects into this block */
46 nat alloc_blocks; /* number of allocate()d blocks since GC */
47 nat alloc_blocks_lim; /* approximate limit on alloc_blocks */
49 generation *generations = NULL; /* all the generations */
50 generation *g0 = NULL; /* generation 0, for convenience */
51 generation *oldest_gen = NULL; /* oldest generation, for convenience */
52 step *g0s0 = NULL; /* generation 0, step 0, for convenience */
55 step *all_steps = NULL; /* single array of steps */
57 ullong total_allocated = 0; /* total memory allocated during run */
59 nat n_nurseries = 0; /* == RtsFlags.ParFlags.nNodes, convenience */
60 step *nurseries = NULL; /* array of nurseries, >1 only if THREADED_RTS */
64 * Storage manager mutex: protects all the above state from
65 * simultaneous access by two STG threads.
69 * This mutex is used by atomicModifyMutVar# only
71 Mutex atomic_modify_mutvar_mutex;
78 static void *stgAllocForGMP (size_t size_in_bytes);
79 static void *stgReallocForGMP (void *ptr, size_t old_size, size_t new_size);
80 static void stgDeallocForGMP (void *ptr, size_t size);
83 initStep (step *stp, int g, int s)
86 stp->abs_no = RtsFlags.GcFlags.steps * g + s;
90 stp->old_blocks = NULL;
91 stp->n_old_blocks = 0;
92 stp->gen = &generations[g];
94 stp->large_objects = NULL;
95 stp->n_large_blocks = 0;
96 stp->scavenged_large_objects = NULL;
97 stp->n_scavenged_large_blocks = 0;
98 stp->is_compacted = 0;
101 initSpinLock(&stp->sync_todo);
102 initSpinLock(&stp->sync_large_objects);
112 if (generations != NULL) {
113 // multi-init protection
119 /* Sanity check to make sure the LOOKS_LIKE_ macros appear to be
120 * doing something reasonable.
122 /* We use the NOT_NULL variant or gcc warns that the test is always true */
123 ASSERT(LOOKS_LIKE_INFO_PTR_NOT_NULL(&stg_BLACKHOLE_info));
124 ASSERT(LOOKS_LIKE_CLOSURE_PTR(&stg_dummy_ret_closure));
125 ASSERT(!HEAP_ALLOCED(&stg_dummy_ret_closure));
127 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
128 RtsFlags.GcFlags.heapSizeSuggestion >
129 RtsFlags.GcFlags.maxHeapSize) {
130 RtsFlags.GcFlags.maxHeapSize = RtsFlags.GcFlags.heapSizeSuggestion;
133 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
134 RtsFlags.GcFlags.minAllocAreaSize >
135 RtsFlags.GcFlags.maxHeapSize) {
136 errorBelch("maximum heap size (-M) is smaller than minimum alloc area size (-A)");
137 RtsFlags.GcFlags.minAllocAreaSize = RtsFlags.GcFlags.maxHeapSize;
140 initBlockAllocator();
142 #if defined(THREADED_RTS)
143 initMutex(&sm_mutex);
144 initMutex(&atomic_modify_mutvar_mutex);
149 /* allocate generation info array */
150 generations = (generation *)stgMallocBytes(RtsFlags.GcFlags.generations
151 * sizeof(struct generation_),
152 "initStorage: gens");
154 /* allocate all the steps into an array. It is important that we do
155 it this way, because we need the invariant that two step pointers
156 can be directly compared to see which is the oldest.
157 Remember that the last generation has only one step. */
158 total_steps = 1 + (RtsFlags.GcFlags.generations - 1) * RtsFlags.GcFlags.steps;
159 all_steps = stgMallocBytes(total_steps * sizeof(struct step_),
160 "initStorage: steps");
162 /* Initialise all generations */
163 for(g = 0; g < RtsFlags.GcFlags.generations; g++) {
164 gen = &generations[g];
166 gen->mut_list = allocBlock();
167 gen->collections = 0;
168 gen->par_collections = 0;
169 gen->failed_promotions = 0;
173 /* A couple of convenience pointers */
174 g0 = &generations[0];
175 oldest_gen = &generations[RtsFlags.GcFlags.generations-1];
177 /* Allocate step structures in each generation */
178 if (RtsFlags.GcFlags.generations > 1) {
179 /* Only for multiple-generations */
181 /* Oldest generation: one step */
182 oldest_gen->n_steps = 1;
183 oldest_gen->steps = all_steps + (RtsFlags.GcFlags.generations - 1)
184 * RtsFlags.GcFlags.steps;
186 /* set up all except the oldest generation with 2 steps */
187 for(g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
188 generations[g].n_steps = RtsFlags.GcFlags.steps;
189 generations[g].steps = all_steps + g * RtsFlags.GcFlags.steps;
193 /* single generation, i.e. a two-space collector */
195 g0->steps = all_steps;
199 n_nurseries = n_capabilities;
203 nurseries = stgMallocBytes (n_nurseries * sizeof(struct step_),
204 "initStorage: nurseries");
206 /* Initialise all steps */
207 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
208 for (s = 0; s < generations[g].n_steps; s++) {
209 initStep(&generations[g].steps[s], g, s);
213 for (s = 0; s < n_nurseries; s++) {
214 initStep(&nurseries[s], 0, s);
217 /* Set up the destination pointers in each younger gen. step */
218 for (g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
219 for (s = 0; s < generations[g].n_steps-1; s++) {
220 generations[g].steps[s].to = &generations[g].steps[s+1];
222 generations[g].steps[s].to = &generations[g+1].steps[0];
224 oldest_gen->steps[0].to = &oldest_gen->steps[0];
226 for (s = 0; s < n_nurseries; s++) {
227 nurseries[s].to = generations[0].steps[0].to;
230 /* The oldest generation has one step. */
231 if (RtsFlags.GcFlags.compact) {
232 if (RtsFlags.GcFlags.generations == 1) {
233 errorBelch("WARNING: compaction is incompatible with -G1; disabled");
235 oldest_gen->steps[0].is_compacted = 1;
239 generations[0].max_blocks = 0;
240 g0s0 = &generations[0].steps[0];
242 /* The allocation area. Policy: keep the allocation area
243 * small to begin with, even if we have a large suggested heap
244 * size. Reason: we're going to do a major collection first, and we
245 * don't want it to be a big one. This vague idea is borne out by
246 * rigorous experimental evidence.
250 weak_ptr_list = NULL;
252 revertible_caf_list = NULL;
254 /* initialise the allocate() interface */
256 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
258 /* Tell GNU multi-precision pkg about our custom alloc functions */
259 mp_set_memory_functions(stgAllocForGMP, stgReallocForGMP, stgDeallocForGMP);
262 initSpinLock(&gc_alloc_block_sync);
263 initSpinLock(&recordMutableGen_sync);
267 IF_DEBUG(gc, statDescribeGens());
275 stat_exit(calcAllocated());
281 stgFree(g0s0); // frees all the steps
282 stgFree(generations);
284 #if defined(THREADED_RTS)
285 closeMutex(&sm_mutex);
286 closeMutex(&atomic_modify_mutvar_mutex);
291 /* -----------------------------------------------------------------------------
294 The entry code for every CAF does the following:
296 - builds a CAF_BLACKHOLE in the heap
297 - pushes an update frame pointing to the CAF_BLACKHOLE
298 - invokes UPD_CAF(), which:
299 - calls newCaf, below
300 - updates the CAF with a static indirection to the CAF_BLACKHOLE
302 Why do we build a BLACKHOLE in the heap rather than just updating
303 the thunk directly? It's so that we only need one kind of update
304 frame - otherwise we'd need a static version of the update frame too.
306 newCaf() does the following:
308 - it puts the CAF on the oldest generation's mut-once list.
309 This is so that we can treat the CAF as a root when collecting
312 For GHCI, we have additional requirements when dealing with CAFs:
314 - we must *retain* all dynamically-loaded CAFs ever entered,
315 just in case we need them again.
316 - we must be able to *revert* CAFs that have been evaluated, to
317 their pre-evaluated form.
319 To do this, we use an additional CAF list. When newCaf() is
320 called on a dynamically-loaded CAF, we add it to the CAF list
321 instead of the old-generation mutable list, and save away its
322 old info pointer (in caf->saved_info) for later reversion.
324 To revert all the CAFs, we traverse the CAF list and reset the
325 info pointer to caf->saved_info, then throw away the CAF list.
326 (see GC.c:revertCAFs()).
330 -------------------------------------------------------------------------- */
333 newCAF(StgClosure* caf)
340 // If we are in GHCi _and_ we are using dynamic libraries,
341 // then we can't redirect newCAF calls to newDynCAF (see below),
342 // so we make newCAF behave almost like newDynCAF.
343 // The dynamic libraries might be used by both the interpreted
344 // program and GHCi itself, so they must not be reverted.
345 // This also means that in GHCi with dynamic libraries, CAFs are not
346 // garbage collected. If this turns out to be a problem, we could
347 // do another hack here and do an address range test on caf to figure
348 // out whether it is from a dynamic library.
349 ((StgIndStatic *)caf)->saved_info = (StgInfoTable *)caf->header.info;
350 ((StgIndStatic *)caf)->static_link = caf_list;
355 /* Put this CAF on the mutable list for the old generation.
356 * This is a HACK - the IND_STATIC closure doesn't really have
357 * a mut_link field, but we pretend it has - in fact we re-use
358 * the STATIC_LINK field for the time being, because when we
359 * come to do a major GC we won't need the mut_link field
360 * any more and can use it as a STATIC_LINK.
362 ((StgIndStatic *)caf)->saved_info = NULL;
363 recordMutableGen(caf, oldest_gen);
369 // An alternate version of newCaf which is used for dynamically loaded
370 // object code in GHCi. In this case we want to retain *all* CAFs in
371 // the object code, because they might be demanded at any time from an
372 // expression evaluated on the command line.
373 // Also, GHCi might want to revert CAFs, so we add these to the
374 // revertible_caf_list.
376 // The linker hackily arranges that references to newCaf from dynamic
377 // code end up pointing to newDynCAF.
379 newDynCAF(StgClosure *caf)
383 ((StgIndStatic *)caf)->saved_info = (StgInfoTable *)caf->header.info;
384 ((StgIndStatic *)caf)->static_link = revertible_caf_list;
385 revertible_caf_list = caf;
390 /* -----------------------------------------------------------------------------
392 -------------------------------------------------------------------------- */
395 allocNursery (step *stp, bdescr *tail, nat blocks)
400 // Allocate a nursery: we allocate fresh blocks one at a time and
401 // cons them on to the front of the list, not forgetting to update
402 // the back pointer on the tail of the list to point to the new block.
403 for (i=0; i < blocks; i++) {
406 processNursery() in LdvProfile.c assumes that every block group in
407 the nursery contains only a single block. So, if a block group is
408 given multiple blocks, change processNursery() accordingly.
412 // double-link the nursery: we might need to insert blocks
419 bd->free = bd->start;
427 assignNurseriesToCapabilities (void)
432 for (i = 0; i < n_nurseries; i++) {
433 capabilities[i].r.rNursery = &nurseries[i];
434 capabilities[i].r.rCurrentNursery = nurseries[i].blocks;
435 capabilities[i].r.rCurrentAlloc = NULL;
437 #else /* THREADED_RTS */
438 MainCapability.r.rNursery = &nurseries[0];
439 MainCapability.r.rCurrentNursery = nurseries[0].blocks;
440 MainCapability.r.rCurrentAlloc = NULL;
445 allocNurseries( void )
449 for (i = 0; i < n_nurseries; i++) {
450 nurseries[i].blocks =
451 allocNursery(&nurseries[i], NULL,
452 RtsFlags.GcFlags.minAllocAreaSize);
453 nurseries[i].n_blocks = RtsFlags.GcFlags.minAllocAreaSize;
454 nurseries[i].old_blocks = NULL;
455 nurseries[i].n_old_blocks = 0;
457 assignNurseriesToCapabilities();
461 resetNurseries( void )
467 for (i = 0; i < n_nurseries; i++) {
469 for (bd = stp->blocks; bd; bd = bd->link) {
470 bd->free = bd->start;
471 ASSERT(bd->gen_no == 0);
472 ASSERT(bd->step == stp);
473 IF_DEBUG(sanity,memset(bd->start, 0xaa, BLOCK_SIZE));
476 assignNurseriesToCapabilities();
480 countNurseryBlocks (void)
485 for (i = 0; i < n_nurseries; i++) {
486 blocks += nurseries[i].n_blocks;
492 resizeNursery ( step *stp, nat blocks )
497 nursery_blocks = stp->n_blocks;
498 if (nursery_blocks == blocks) return;
500 if (nursery_blocks < blocks) {
501 debugTrace(DEBUG_gc, "increasing size of nursery to %d blocks",
503 stp->blocks = allocNursery(stp, stp->blocks, blocks-nursery_blocks);
508 debugTrace(DEBUG_gc, "decreasing size of nursery to %d blocks",
512 while (nursery_blocks > blocks) {
514 next_bd->u.back = NULL;
515 nursery_blocks -= bd->blocks; // might be a large block
520 // might have gone just under, by freeing a large block, so make
521 // up the difference.
522 if (nursery_blocks < blocks) {
523 stp->blocks = allocNursery(stp, stp->blocks, blocks-nursery_blocks);
527 stp->n_blocks = blocks;
528 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
532 // Resize each of the nurseries to the specified size.
535 resizeNurseriesFixed (nat blocks)
538 for (i = 0; i < n_nurseries; i++) {
539 resizeNursery(&nurseries[i], blocks);
544 // Resize the nurseries to the total specified size.
547 resizeNurseries (nat blocks)
549 // If there are multiple nurseries, then we just divide the number
550 // of available blocks between them.
551 resizeNurseriesFixed(blocks / n_nurseries);
554 /* -----------------------------------------------------------------------------
555 The allocate() interface
557 allocateInGen() function allocates memory directly into a specific
558 generation. It always succeeds, and returns a chunk of memory n
559 words long. n can be larger than the size of a block if necessary,
560 in which case a contiguous block group will be allocated.
562 allocate(n) is equivalent to allocateInGen(g0).
563 -------------------------------------------------------------------------- */
566 allocateInGen (generation *g, nat n)
574 TICK_ALLOC_HEAP_NOCTR(n);
579 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_))
581 nat req_blocks = (lnat)BLOCK_ROUND_UP(n*sizeof(W_)) / BLOCK_SIZE;
583 // Attempting to allocate an object larger than maxHeapSize
584 // should definitely be disallowed. (bug #1791)
585 if (RtsFlags.GcFlags.maxHeapSize > 0 &&
586 req_blocks >= RtsFlags.GcFlags.maxHeapSize) {
590 bd = allocGroup(req_blocks);
591 dbl_link_onto(bd, &stp->large_objects);
592 stp->n_large_blocks += bd->blocks; // might be larger than req_blocks
595 bd->flags = BF_LARGE;
596 bd->free = bd->start + n;
601 // small allocation (<LARGE_OBJECT_THRESHOLD) */
603 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
608 bd->link = stp->blocks;
625 return allocateInGen(g0,n);
629 allocatedBytes( void )
633 allocated = alloc_blocks * BLOCK_SIZE_W;
634 if (pinned_object_block != NULL) {
635 allocated -= (pinned_object_block->start + BLOCK_SIZE_W) -
636 pinned_object_block->free;
642 // split N blocks off the start of the given bdescr, returning the
643 // remainder as a new block group. We treat the remainder as if it
644 // had been freshly allocated in generation 0.
646 splitLargeBlock (bdescr *bd, nat blocks)
650 // subtract the original number of blocks from the counter first
651 bd->step->n_large_blocks -= bd->blocks;
653 new_bd = splitBlockGroup (bd, blocks);
655 dbl_link_onto(new_bd, &g0s0->large_objects);
656 g0s0->n_large_blocks += new_bd->blocks;
657 new_bd->gen_no = g0s0->no;
659 new_bd->flags = BF_LARGE;
660 new_bd->free = bd->free;
662 // add the new number of blocks to the counter. Due to the gaps
663 // for block descriptor, new_bd->blocks + bd->blocks might not be
664 // equal to the original bd->blocks, which is why we do it this way.
665 bd->step->n_large_blocks += bd->blocks;
670 /* -----------------------------------------------------------------------------
673 This allocates memory in the current thread - it is intended for
674 use primarily from STG-land where we have a Capability. It is
675 better than allocate() because it doesn't require taking the
676 sm_mutex lock in the common case.
678 Memory is allocated directly from the nursery if possible (but not
679 from the current nursery block, so as not to interfere with
681 -------------------------------------------------------------------------- */
684 allocateLocal (Capability *cap, nat n)
689 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
690 return allocateInGen(g0,n);
693 /* small allocation (<LARGE_OBJECT_THRESHOLD) */
695 TICK_ALLOC_HEAP_NOCTR(n);
698 bd = cap->r.rCurrentAlloc;
699 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
701 // The CurrentAlloc block is full, we need to find another
702 // one. First, we try taking the next block from the
704 bd = cap->r.rCurrentNursery->link;
706 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
707 // The nursery is empty, or the next block is already
708 // full: allocate a fresh block (we can't fail here).
711 cap->r.rNursery->n_blocks++;
714 bd->step = cap->r.rNursery;
716 // NO: alloc_blocks++;
717 // calcAllocated() uses the size of the nursery, and we've
718 // already bumpted nursery->n_blocks above.
720 // we have a block in the nursery: take it and put
721 // it at the *front* of the nursery list, and use it
722 // to allocate() from.
723 cap->r.rCurrentNursery->link = bd->link;
724 if (bd->link != NULL) {
725 bd->link->u.back = cap->r.rCurrentNursery;
728 dbl_link_onto(bd, &cap->r.rNursery->blocks);
729 cap->r.rCurrentAlloc = bd;
730 IF_DEBUG(sanity, checkNurserySanity(cap->r.rNursery));
737 /* ---------------------------------------------------------------------------
738 Allocate a fixed/pinned object.
740 We allocate small pinned objects into a single block, allocating a
741 new block when the current one overflows. The block is chained
742 onto the large_object_list of generation 0 step 0.
744 NOTE: The GC can't in general handle pinned objects. This
745 interface is only safe to use for ByteArrays, which have no
746 pointers and don't require scavenging. It works because the
747 block's descriptor has the BF_LARGE flag set, so the block is
748 treated as a large object and chained onto various lists, rather
749 than the individual objects being copied. However, when it comes
750 to scavenge the block, the GC will only scavenge the first object.
751 The reason is that the GC can't linearly scan a block of pinned
752 objects at the moment (doing so would require using the
753 mostly-copying techniques). But since we're restricting ourselves
754 to pinned ByteArrays, not scavenging is ok.
756 This function is called by newPinnedByteArray# which immediately
757 fills the allocated memory with a MutableByteArray#.
758 ------------------------------------------------------------------------- */
761 allocatePinned( nat n )
764 bdescr *bd = pinned_object_block;
766 // If the request is for a large object, then allocate()
767 // will give us a pinned object anyway.
768 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
774 TICK_ALLOC_HEAP_NOCTR(n);
777 // we always return 8-byte aligned memory. bd->free must be
778 // 8-byte aligned to begin with, so we just round up n to
779 // the nearest multiple of 8 bytes.
780 if (sizeof(StgWord) == 4) {
784 // If we don't have a block of pinned objects yet, or the current
785 // one isn't large enough to hold the new object, allocate a new one.
786 if (bd == NULL || (bd->free + n) > (bd->start + BLOCK_SIZE_W)) {
787 pinned_object_block = bd = allocBlock();
788 dbl_link_onto(bd, &g0s0->large_objects);
789 g0s0->n_large_blocks++;
792 bd->flags = BF_PINNED | BF_LARGE;
793 bd->free = bd->start;
803 /* -----------------------------------------------------------------------------
805 -------------------------------------------------------------------------- */
808 This is the write barrier for MUT_VARs, a.k.a. IORefs. A
809 MUT_VAR_CLEAN object is not on the mutable list; a MUT_VAR_DIRTY
810 is. When written to, a MUT_VAR_CLEAN turns into a MUT_VAR_DIRTY
811 and is put on the mutable list.
814 dirty_MUT_VAR(StgRegTable *reg, StgClosure *p)
816 Capability *cap = regTableToCapability(reg);
818 if (p->header.info == &stg_MUT_VAR_CLEAN_info) {
819 p->header.info = &stg_MUT_VAR_DIRTY_info;
820 bd = Bdescr((StgPtr)p);
821 if (bd->gen_no > 0) recordMutableCap(p,cap,bd->gen_no);
825 // Setting a TSO's link field with a write barrier.
826 // It is *not* necessary to call this function when
827 // * setting the link field to END_TSO_QUEUE
828 // * putting a TSO on the blackhole_queue
829 // * setting the link field of the currently running TSO, as it
830 // will already be dirty.
832 setTSOLink (Capability *cap, StgTSO *tso, StgTSO *target)
835 if ((tso->flags & (TSO_DIRTY|TSO_LINK_DIRTY)) == 0) {
836 tso->flags |= TSO_LINK_DIRTY;
837 bd = Bdescr((StgPtr)tso);
838 if (bd->gen_no > 0) recordMutableCap((StgClosure*)tso,cap,bd->gen_no);
844 dirty_TSO (Capability *cap, StgTSO *tso)
847 if ((tso->flags & TSO_DIRTY) == 0) {
848 tso->flags |= TSO_DIRTY;
849 bd = Bdescr((StgPtr)tso);
850 if (bd->gen_no > 0) recordMutableCap((StgClosure*)tso,cap,bd->gen_no);
855 This is the write barrier for MVARs. An MVAR_CLEAN objects is not
856 on the mutable list; a MVAR_DIRTY is. When written to, a
857 MVAR_CLEAN turns into a MVAR_DIRTY and is put on the mutable list.
858 The check for MVAR_CLEAN is inlined at the call site for speed,
859 this really does make a difference on concurrency-heavy benchmarks
860 such as Chaneneos and cheap-concurrency.
863 dirty_MVAR(StgRegTable *reg, StgClosure *p)
865 Capability *cap = regTableToCapability(reg);
867 bd = Bdescr((StgPtr)p);
868 if (bd->gen_no > 0) recordMutableCap(p,cap,bd->gen_no);
871 /* -----------------------------------------------------------------------------
872 Allocation functions for GMP.
874 These all use the allocate() interface - we can't have any garbage
875 collection going on during a gmp operation, so we use allocate()
876 which always succeeds. The gmp operations which might need to
877 allocate will ask the storage manager (via doYouWantToGC()) whether
878 a garbage collection is required, in case we get into a loop doing
879 only allocate() style allocation.
880 -------------------------------------------------------------------------- */
883 stgAllocForGMP (size_t size_in_bytes)
886 nat data_size_in_words, total_size_in_words;
888 /* round up to a whole number of words */
889 data_size_in_words = (size_in_bytes + sizeof(W_) + 1) / sizeof(W_);
890 total_size_in_words = sizeofW(StgArrWords) + data_size_in_words;
892 /* allocate and fill it in. */
893 #if defined(THREADED_RTS)
894 arr = (StgArrWords *)allocateLocal(myTask()->cap, total_size_in_words);
896 arr = (StgArrWords *)allocateLocal(&MainCapability, total_size_in_words);
898 SET_ARR_HDR(arr, &stg_ARR_WORDS_info, CCCS, data_size_in_words);
900 /* and return a ptr to the goods inside the array */
905 stgReallocForGMP (void *ptr, size_t old_size, size_t new_size)
907 void *new_stuff_ptr = stgAllocForGMP(new_size);
909 char *p = (char *) ptr;
910 char *q = (char *) new_stuff_ptr;
912 for (; i < old_size; i++, p++, q++) {
916 return(new_stuff_ptr);
920 stgDeallocForGMP (void *ptr STG_UNUSED,
921 size_t size STG_UNUSED)
923 /* easy for us: the garbage collector does the dealloc'n */
926 /* -----------------------------------------------------------------------------
928 * -------------------------------------------------------------------------- */
930 /* -----------------------------------------------------------------------------
933 * Approximate how much we've allocated: number of blocks in the
934 * nursery + blocks allocated via allocate() - unused nusery blocks.
935 * This leaves a little slop at the end of each block, and doesn't
936 * take into account large objects (ToDo).
937 * -------------------------------------------------------------------------- */
940 calcAllocated( void )
945 allocated = allocatedBytes();
946 allocated += countNurseryBlocks() * BLOCK_SIZE_W;
951 for (i = 0; i < n_nurseries; i++) {
953 for ( bd = capabilities[i].r.rCurrentNursery->link;
954 bd != NULL; bd = bd->link ) {
955 allocated -= BLOCK_SIZE_W;
957 cap = &capabilities[i];
958 if (cap->r.rCurrentNursery->free <
959 cap->r.rCurrentNursery->start + BLOCK_SIZE_W) {
960 allocated -= (cap->r.rCurrentNursery->start + BLOCK_SIZE_W)
961 - cap->r.rCurrentNursery->free;
965 bdescr *current_nursery = MainCapability.r.rCurrentNursery;
967 for ( bd = current_nursery->link; bd != NULL; bd = bd->link ) {
968 allocated -= BLOCK_SIZE_W;
970 if (current_nursery->free < current_nursery->start + BLOCK_SIZE_W) {
971 allocated -= (current_nursery->start + BLOCK_SIZE_W)
972 - current_nursery->free;
977 total_allocated += allocated;
981 /* Approximate the amount of live data in the heap. To be called just
982 * after garbage collection (see GarbageCollect()).
991 if (RtsFlags.GcFlags.generations == 1) {
992 return g0s0->n_large_blocks + g0s0->n_blocks;
995 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
996 for (s = 0; s < generations[g].n_steps; s++) {
997 /* approximate amount of live data (doesn't take into account slop
998 * at end of each block).
1000 if (g == 0 && s == 0) {
1003 stp = &generations[g].steps[s];
1004 live += stp->n_large_blocks + stp->n_blocks;
1011 countOccupied(bdescr *bd)
1016 for (; bd != NULL; bd = bd->link) {
1017 words += bd->free - bd->start;
1022 // Return an accurate count of the live data in the heap, excluding
1031 if (RtsFlags.GcFlags.generations == 1) {
1032 return g0s0->n_words + countOccupied(g0s0->large_objects);
1036 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1037 for (s = 0; s < generations[g].n_steps; s++) {
1038 if (g == 0 && s == 0) continue;
1039 stp = &generations[g].steps[s];
1040 live += stp->n_words + countOccupied(stp->large_objects);
1046 /* Approximate the number of blocks that will be needed at the next
1047 * garbage collection.
1049 * Assume: all data currently live will remain live. Steps that will
1050 * be collected next time will therefore need twice as many blocks
1051 * since all the data will be copied.
1060 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1061 for (s = 0; s < generations[g].n_steps; s++) {
1062 if (g == 0 && s == 0) { continue; }
1063 stp = &generations[g].steps[s];
1064 if (g == 0 || // always collect gen 0
1065 (generations[g].steps[0].n_blocks +
1066 generations[g].steps[0].n_large_blocks
1067 > generations[g].max_blocks
1068 && stp->is_compacted == 0)) {
1069 needed += 2 * stp->n_blocks + stp->n_large_blocks;
1071 needed += stp->n_blocks + stp->n_large_blocks;
1078 /* ----------------------------------------------------------------------------
1081 Executable memory must be managed separately from non-executable
1082 memory. Most OSs these days require you to jump through hoops to
1083 dynamically allocate executable memory, due to various security
1086 Here we provide a small memory allocator for executable memory.
1087 Memory is managed with a page granularity; we allocate linearly
1088 in the page, and when the page is emptied (all objects on the page
1089 are free) we free the page again, not forgetting to make it
1092 TODO: The inability to handle objects bigger than BLOCK_SIZE_W means that
1093 the linker cannot use allocateExec for loading object code files
1094 on Windows. Once allocateExec can handle larger objects, the linker
1095 should be modified to use allocateExec instead of VirtualAlloc.
1096 ------------------------------------------------------------------------- */
1098 static bdescr *exec_block;
1100 void *allocateExec (nat bytes)
1107 // round up to words.
1108 n = (bytes + sizeof(W_) + 1) / sizeof(W_);
1110 if (n+1 > BLOCK_SIZE_W) {
1111 barf("allocateExec: can't handle large objects");
1114 if (exec_block == NULL ||
1115 exec_block->free + n + 1 > exec_block->start + BLOCK_SIZE_W) {
1117 lnat pagesize = getPageSize();
1118 bd = allocGroup(stg_max(1, pagesize / BLOCK_SIZE));
1119 debugTrace(DEBUG_gc, "allocate exec block %p", bd->start);
1121 bd->flags = BF_EXEC;
1122 bd->link = exec_block;
1123 if (exec_block != NULL) {
1124 exec_block->u.back = bd;
1127 setExecutable(bd->start, bd->blocks * BLOCK_SIZE, rtsTrue);
1130 *(exec_block->free) = n; // store the size of this chunk
1131 exec_block->gen_no += n; // gen_no stores the number of words allocated
1132 ret = exec_block->free + 1;
1133 exec_block->free += n + 1;
1139 void freeExec (void *addr)
1141 StgPtr p = (StgPtr)addr - 1;
1142 bdescr *bd = Bdescr((StgPtr)p);
1144 if ((bd->flags & BF_EXEC) == 0) {
1145 barf("freeExec: not executable");
1148 if (*(StgPtr)p == 0) {
1149 barf("freeExec: already free?");
1154 bd->gen_no -= *(StgPtr)p;
1157 if (bd->gen_no == 0) {
1158 // Free the block if it is empty, but not if it is the block at
1159 // the head of the queue.
1160 if (bd != exec_block) {
1161 debugTrace(DEBUG_gc, "free exec block %p", bd->start);
1162 dbl_link_remove(bd, &exec_block);
1163 setExecutable(bd->start, bd->blocks * BLOCK_SIZE, rtsFalse);
1166 bd->free = bd->start;
1173 /* -----------------------------------------------------------------------------
1176 memInventory() checks for memory leaks by counting up all the
1177 blocks we know about and comparing that to the number of blocks
1178 allegedly floating around in the system.
1179 -------------------------------------------------------------------------- */
1183 // Useful for finding partially full blocks in gdb
1184 void findSlop(bdescr *bd);
1185 void findSlop(bdescr *bd)
1189 for (; bd != NULL; bd = bd->link) {
1190 slop = (bd->blocks * BLOCK_SIZE_W) - (bd->free - bd->start);
1191 if (slop > (1024/sizeof(W_))) {
1192 debugBelch("block at %p (bdescr %p) has %ldKB slop\n",
1193 bd->start, bd, slop / (1024/sizeof(W_)));
1199 countBlocks(bdescr *bd)
1202 for (n=0; bd != NULL; bd=bd->link) {
1208 // (*1) Just like countBlocks, except that we adjust the count for a
1209 // megablock group so that it doesn't include the extra few blocks
1210 // that would be taken up by block descriptors in the second and
1211 // subsequent megablock. This is so we can tally the count with the
1212 // number of blocks allocated in the system, for memInventory().
1214 countAllocdBlocks(bdescr *bd)
1217 for (n=0; bd != NULL; bd=bd->link) {
1219 // hack for megablock groups: see (*1) above
1220 if (bd->blocks > BLOCKS_PER_MBLOCK) {
1221 n -= (MBLOCK_SIZE / BLOCK_SIZE - BLOCKS_PER_MBLOCK)
1222 * (bd->blocks/(MBLOCK_SIZE/BLOCK_SIZE));
1229 stepBlocks (step *stp)
1231 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
1232 ASSERT(countBlocks(stp->large_objects) == stp->n_large_blocks);
1233 return stp->n_blocks + stp->n_old_blocks +
1234 countAllocdBlocks(stp->large_objects);
1238 memInventory (rtsBool show)
1242 lnat gen_blocks[RtsFlags.GcFlags.generations];
1243 lnat nursery_blocks, retainer_blocks,
1244 arena_blocks, exec_blocks;
1245 lnat live_blocks = 0, free_blocks = 0;
1248 // count the blocks we current have
1250 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1252 for (i = 0; i < n_capabilities; i++) {
1253 gen_blocks[g] += countBlocks(capabilities[i].mut_lists[g]);
1255 gen_blocks[g] += countAllocdBlocks(generations[g].mut_list);
1256 for (s = 0; s < generations[g].n_steps; s++) {
1257 stp = &generations[g].steps[s];
1258 gen_blocks[g] += stepBlocks(stp);
1263 for (i = 0; i < n_nurseries; i++) {
1264 nursery_blocks += stepBlocks(&nurseries[i]);
1267 retainer_blocks = 0;
1269 if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_RETAINER) {
1270 retainer_blocks = retainerStackBlocks();
1274 // count the blocks allocated by the arena allocator
1275 arena_blocks = arenaBlocks();
1277 // count the blocks containing executable memory
1278 exec_blocks = countAllocdBlocks(exec_block);
1280 /* count the blocks on the free list */
1281 free_blocks = countFreeList();
1284 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1285 live_blocks += gen_blocks[g];
1287 live_blocks += nursery_blocks +
1288 + retainer_blocks + arena_blocks + exec_blocks;
1290 #define MB(n) (((n) * BLOCK_SIZE_W) / ((1024*1024)/sizeof(W_)))
1292 leak = live_blocks + free_blocks != mblocks_allocated * BLOCKS_PER_MBLOCK;
1296 debugBelch("Memory leak detected:\n");
1298 debugBelch("Memory inventory:\n");
1300 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1301 debugBelch(" gen %d blocks : %5lu blocks (%lu MB)\n", g,
1302 gen_blocks[g], MB(gen_blocks[g]));
1304 debugBelch(" nursery : %5lu blocks (%lu MB)\n",
1305 nursery_blocks, MB(nursery_blocks));
1306 debugBelch(" retainer : %5lu blocks (%lu MB)\n",
1307 retainer_blocks, MB(retainer_blocks));
1308 debugBelch(" arena blocks : %5lu blocks (%lu MB)\n",
1309 arena_blocks, MB(arena_blocks));
1310 debugBelch(" exec : %5lu blocks (%lu MB)\n",
1311 exec_blocks, MB(exec_blocks));
1312 debugBelch(" free : %5lu blocks (%lu MB)\n",
1313 free_blocks, MB(free_blocks));
1314 debugBelch(" total : %5lu blocks (%lu MB)\n",
1315 live_blocks + free_blocks, MB(live_blocks+free_blocks));
1317 debugBelch("\n in system : %5lu blocks (%lu MB)\n",
1318 mblocks_allocated * BLOCKS_PER_MBLOCK, mblocks_allocated);
1324 /* Full heap sanity check. */
1330 if (RtsFlags.GcFlags.generations == 1) {
1331 checkHeap(g0s0->blocks);
1332 checkChain(g0s0->large_objects);
1335 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1336 for (s = 0; s < generations[g].n_steps; s++) {
1337 if (g == 0 && s == 0) { continue; }
1338 ASSERT(countBlocks(generations[g].steps[s].blocks)
1339 == generations[g].steps[s].n_blocks);
1340 ASSERT(countBlocks(generations[g].steps[s].large_objects)
1341 == generations[g].steps[s].n_large_blocks);
1342 checkHeap(generations[g].steps[s].blocks);
1343 checkChain(generations[g].steps[s].large_objects);
1345 checkMutableList(generations[g].mut_list, g);
1350 for (s = 0; s < n_nurseries; s++) {
1351 ASSERT(countBlocks(nurseries[s].blocks)
1352 == nurseries[s].n_blocks);
1353 ASSERT(countBlocks(nurseries[s].large_objects)
1354 == nurseries[s].n_large_blocks);
1357 checkFreeListSanity();
1361 /* Nursery sanity check */
1363 checkNurserySanity( step *stp )
1369 for (bd = stp->blocks; bd != NULL; bd = bd->link) {
1370 ASSERT(bd->u.back == prev);
1372 blocks += bd->blocks;
1374 ASSERT(blocks == stp->n_blocks);
1377 // handy function for use in gdb, because Bdescr() is inlined.
1378 extern bdescr *_bdescr( StgPtr p );