1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 1998-2006
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)
37 * All these globals require sm_mutex to access in THREADED_RTS mode.
39 StgClosure *caf_list = NULL;
40 StgClosure *revertible_caf_list = NULL;
43 bdescr *pinned_object_block; /* allocate pinned objects into this block */
44 nat alloc_blocks; /* number of allocate()d blocks since GC */
45 nat alloc_blocks_lim; /* approximate limit on alloc_blocks */
47 generation *generations = NULL; /* all the generations */
48 generation *g0 = NULL; /* generation 0, for convenience */
49 generation *oldest_gen = NULL; /* oldest generation, for convenience */
50 step *g0s0 = NULL; /* generation 0, step 0, for convenience */
52 ullong total_allocated = 0; /* total memory allocated during run */
54 nat n_nurseries = 0; /* == RtsFlags.ParFlags.nNodes, convenience */
55 step *nurseries = NULL; /* array of nurseries, >1 only if THREADED_RTS */
59 * Storage manager mutex: protects all the above state from
60 * simultaneous access by two STG threads.
64 * This mutex is used by atomicModifyMutVar# only
66 Mutex atomic_modify_mutvar_mutex;
73 static void *stgAllocForGMP (size_t size_in_bytes);
74 static void *stgReallocForGMP (void *ptr, size_t old_size, size_t new_size);
75 static void stgDeallocForGMP (void *ptr, size_t size);
78 initStep (step *stp, int g, int s)
83 stp->old_blocks = NULL;
84 stp->n_old_blocks = 0;
85 stp->gen = &generations[g];
91 stp->scavd_hpLim = NULL;
94 stp->large_objects = NULL;
95 stp->n_large_blocks = 0;
96 stp->new_large_objects = NULL;
97 stp->scavenged_large_objects = NULL;
98 stp->n_scavenged_large_blocks = 0;
99 stp->is_compacted = 0;
109 if (generations != NULL) {
110 // multi-init protection
116 /* Sanity check to make sure the LOOKS_LIKE_ macros appear to be
117 * doing something reasonable.
119 ASSERT(LOOKS_LIKE_INFO_PTR(&stg_BLACKHOLE_info));
120 ASSERT(LOOKS_LIKE_CLOSURE_PTR(&stg_dummy_ret_closure));
121 ASSERT(!HEAP_ALLOCED(&stg_dummy_ret_closure));
123 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
124 RtsFlags.GcFlags.heapSizeSuggestion >
125 RtsFlags.GcFlags.maxHeapSize) {
126 RtsFlags.GcFlags.maxHeapSize = RtsFlags.GcFlags.heapSizeSuggestion;
129 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
130 RtsFlags.GcFlags.minAllocAreaSize >
131 RtsFlags.GcFlags.maxHeapSize) {
132 errorBelch("maximum heap size (-M) is smaller than minimum alloc area size (-A)");
133 RtsFlags.GcFlags.minAllocAreaSize = RtsFlags.GcFlags.maxHeapSize;
136 initBlockAllocator();
138 #if defined(THREADED_RTS)
139 initMutex(&sm_mutex);
140 initMutex(&atomic_modify_mutvar_mutex);
145 /* allocate generation info array */
146 generations = (generation *)stgMallocBytes(RtsFlags.GcFlags.generations
147 * sizeof(struct generation_),
148 "initStorage: gens");
150 /* Initialise all generations */
151 for(g = 0; g < RtsFlags.GcFlags.generations; g++) {
152 gen = &generations[g];
154 gen->mut_list = allocBlock();
155 gen->collections = 0;
156 gen->failed_promotions = 0;
160 /* A couple of convenience pointers */
161 g0 = &generations[0];
162 oldest_gen = &generations[RtsFlags.GcFlags.generations-1];
164 /* Allocate step structures in each generation */
165 if (RtsFlags.GcFlags.generations > 1) {
166 /* Only for multiple-generations */
168 /* Oldest generation: one step */
169 oldest_gen->n_steps = 1;
171 stgMallocBytes(1 * sizeof(struct step_), "initStorage: last step");
173 /* set up all except the oldest generation with 2 steps */
174 for(g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
175 generations[g].n_steps = RtsFlags.GcFlags.steps;
176 generations[g].steps =
177 stgMallocBytes (RtsFlags.GcFlags.steps * sizeof(struct step_),
178 "initStorage: steps");
182 /* single generation, i.e. a two-space collector */
184 g0->steps = stgMallocBytes (sizeof(struct step_), "initStorage: steps");
188 n_nurseries = n_capabilities;
192 nurseries = stgMallocBytes (n_nurseries * sizeof(struct step_),
193 "initStorage: nurseries");
195 /* Initialise all steps */
196 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
197 for (s = 0; s < generations[g].n_steps; s++) {
198 initStep(&generations[g].steps[s], g, s);
202 for (s = 0; s < n_nurseries; s++) {
203 initStep(&nurseries[s], 0, s);
206 /* Set up the destination pointers in each younger gen. step */
207 for (g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
208 for (s = 0; s < generations[g].n_steps-1; s++) {
209 generations[g].steps[s].to = &generations[g].steps[s+1];
211 generations[g].steps[s].to = &generations[g+1].steps[0];
213 oldest_gen->steps[0].to = &oldest_gen->steps[0];
215 for (s = 0; s < n_nurseries; s++) {
216 nurseries[s].to = generations[0].steps[0].to;
219 /* The oldest generation has one step. */
220 if (RtsFlags.GcFlags.compact) {
221 if (RtsFlags.GcFlags.generations == 1) {
222 errorBelch("WARNING: compaction is incompatible with -G1; disabled");
224 oldest_gen->steps[0].is_compacted = 1;
228 generations[0].max_blocks = 0;
229 g0s0 = &generations[0].steps[0];
231 /* The allocation area. Policy: keep the allocation area
232 * small to begin with, even if we have a large suggested heap
233 * size. Reason: we're going to do a major collection first, and we
234 * don't want it to be a big one. This vague idea is borne out by
235 * rigorous experimental evidence.
239 weak_ptr_list = NULL;
241 revertible_caf_list = NULL;
243 /* initialise the allocate() interface */
245 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
247 /* Tell GNU multi-precision pkg about our custom alloc functions */
248 mp_set_memory_functions(stgAllocForGMP, stgReallocForGMP, stgDeallocForGMP);
250 IF_DEBUG(gc, statDescribeGens());
258 stat_exit(calcAllocated());
266 for(g = 0; g < RtsFlags.GcFlags.generations; g++)
267 stgFree(generations[g].steps);
268 stgFree(generations);
270 #if defined(THREADED_RTS)
271 closeMutex(&sm_mutex);
272 closeMutex(&atomic_modify_mutvar_mutex);
277 /* -----------------------------------------------------------------------------
280 The entry code for every CAF does the following:
282 - builds a CAF_BLACKHOLE in the heap
283 - pushes an update frame pointing to the CAF_BLACKHOLE
284 - invokes UPD_CAF(), which:
285 - calls newCaf, below
286 - updates the CAF with a static indirection to the CAF_BLACKHOLE
288 Why do we build a BLACKHOLE in the heap rather than just updating
289 the thunk directly? It's so that we only need one kind of update
290 frame - otherwise we'd need a static version of the update frame too.
292 newCaf() does the following:
294 - it puts the CAF on the oldest generation's mut-once list.
295 This is so that we can treat the CAF as a root when collecting
298 For GHCI, we have additional requirements when dealing with CAFs:
300 - we must *retain* all dynamically-loaded CAFs ever entered,
301 just in case we need them again.
302 - we must be able to *revert* CAFs that have been evaluated, to
303 their pre-evaluated form.
305 To do this, we use an additional CAF list. When newCaf() is
306 called on a dynamically-loaded CAF, we add it to the CAF list
307 instead of the old-generation mutable list, and save away its
308 old info pointer (in caf->saved_info) for later reversion.
310 To revert all the CAFs, we traverse the CAF list and reset the
311 info pointer to caf->saved_info, then throw away the CAF list.
312 (see GC.c:revertCAFs()).
316 -------------------------------------------------------------------------- */
319 newCAF(StgClosure* caf)
326 // If we are in GHCi _and_ we are using dynamic libraries,
327 // then we can't redirect newCAF calls to newDynCAF (see below),
328 // so we make newCAF behave almost like newDynCAF.
329 // The dynamic libraries might be used by both the interpreted
330 // program and GHCi itself, so they must not be reverted.
331 // This also means that in GHCi with dynamic libraries, CAFs are not
332 // garbage collected. If this turns out to be a problem, we could
333 // do another hack here and do an address range test on caf to figure
334 // out whether it is from a dynamic library.
335 ((StgIndStatic *)caf)->saved_info = (StgInfoTable *)caf->header.info;
336 ((StgIndStatic *)caf)->static_link = caf_list;
341 /* Put this CAF on the mutable list for the old generation.
342 * This is a HACK - the IND_STATIC closure doesn't really have
343 * a mut_link field, but we pretend it has - in fact we re-use
344 * the STATIC_LINK field for the time being, because when we
345 * come to do a major GC we won't need the mut_link field
346 * any more and can use it as a STATIC_LINK.
348 ((StgIndStatic *)caf)->saved_info = NULL;
349 recordMutableGen(caf, oldest_gen);
355 // An alternate version of newCaf which is used for dynamically loaded
356 // object code in GHCi. In this case we want to retain *all* CAFs in
357 // the object code, because they might be demanded at any time from an
358 // expression evaluated on the command line.
359 // Also, GHCi might want to revert CAFs, so we add these to the
360 // revertible_caf_list.
362 // The linker hackily arranges that references to newCaf from dynamic
363 // code end up pointing to newDynCAF.
365 newDynCAF(StgClosure *caf)
369 ((StgIndStatic *)caf)->saved_info = (StgInfoTable *)caf->header.info;
370 ((StgIndStatic *)caf)->static_link = revertible_caf_list;
371 revertible_caf_list = caf;
376 /* -----------------------------------------------------------------------------
378 -------------------------------------------------------------------------- */
381 allocNursery (step *stp, bdescr *tail, nat blocks)
386 // Allocate a nursery: we allocate fresh blocks one at a time and
387 // cons them on to the front of the list, not forgetting to update
388 // the back pointer on the tail of the list to point to the new block.
389 for (i=0; i < blocks; i++) {
392 processNursery() in LdvProfile.c assumes that every block group in
393 the nursery contains only a single block. So, if a block group is
394 given multiple blocks, change processNursery() accordingly.
398 // double-link the nursery: we might need to insert blocks
405 bd->free = bd->start;
413 assignNurseriesToCapabilities (void)
418 for (i = 0; i < n_nurseries; i++) {
419 capabilities[i].r.rNursery = &nurseries[i];
420 capabilities[i].r.rCurrentNursery = nurseries[i].blocks;
421 capabilities[i].r.rCurrentAlloc = NULL;
423 #else /* THREADED_RTS */
424 MainCapability.r.rNursery = &nurseries[0];
425 MainCapability.r.rCurrentNursery = nurseries[0].blocks;
426 MainCapability.r.rCurrentAlloc = NULL;
431 allocNurseries( void )
435 for (i = 0; i < n_nurseries; i++) {
436 nurseries[i].blocks =
437 allocNursery(&nurseries[i], NULL,
438 RtsFlags.GcFlags.minAllocAreaSize);
439 nurseries[i].n_blocks = RtsFlags.GcFlags.minAllocAreaSize;
440 nurseries[i].old_blocks = NULL;
441 nurseries[i].n_old_blocks = 0;
443 assignNurseriesToCapabilities();
447 resetNurseries( void )
453 for (i = 0; i < n_nurseries; i++) {
455 for (bd = stp->blocks; bd; bd = bd->link) {
456 bd->free = bd->start;
457 ASSERT(bd->gen_no == 0);
458 ASSERT(bd->step == stp);
459 IF_DEBUG(sanity,memset(bd->start, 0xaa, BLOCK_SIZE));
462 assignNurseriesToCapabilities();
466 countNurseryBlocks (void)
471 for (i = 0; i < n_nurseries; i++) {
472 blocks += nurseries[i].n_blocks;
478 resizeNursery ( step *stp, nat blocks )
483 nursery_blocks = stp->n_blocks;
484 if (nursery_blocks == blocks) return;
486 if (nursery_blocks < blocks) {
487 debugTrace(DEBUG_gc, "increasing size of nursery to %d blocks",
489 stp->blocks = allocNursery(stp, stp->blocks, blocks-nursery_blocks);
494 debugTrace(DEBUG_gc, "decreasing size of nursery to %d blocks",
498 while (nursery_blocks > blocks) {
500 next_bd->u.back = NULL;
501 nursery_blocks -= bd->blocks; // might be a large block
506 // might have gone just under, by freeing a large block, so make
507 // up the difference.
508 if (nursery_blocks < blocks) {
509 stp->blocks = allocNursery(stp, stp->blocks, blocks-nursery_blocks);
513 stp->n_blocks = blocks;
514 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
518 // Resize each of the nurseries to the specified size.
521 resizeNurseriesFixed (nat blocks)
524 for (i = 0; i < n_nurseries; i++) {
525 resizeNursery(&nurseries[i], blocks);
530 // Resize the nurseries to the total specified size.
533 resizeNurseries (nat blocks)
535 // If there are multiple nurseries, then we just divide the number
536 // of available blocks between them.
537 resizeNurseriesFixed(blocks / n_nurseries);
540 /* -----------------------------------------------------------------------------
541 The allocate() interface
543 allocateInGen() function allocates memory directly into a specific
544 generation. It always succeeds, and returns a chunk of memory n
545 words long. n can be larger than the size of a block if necessary,
546 in which case a contiguous block group will be allocated.
548 allocate(n) is equivalent to allocateInGen(g0).
549 -------------------------------------------------------------------------- */
552 allocateInGen (generation *g, nat n)
560 TICK_ALLOC_HEAP_NOCTR(n);
565 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_))
567 nat req_blocks = (lnat)BLOCK_ROUND_UP(n*sizeof(W_)) / BLOCK_SIZE;
569 // Attempting to allocate an object larger than maxHeapSize
570 // should definitely be disallowed. (bug #1791)
571 if (RtsFlags.GcFlags.maxHeapSize > 0 &&
572 req_blocks >= RtsFlags.GcFlags.maxHeapSize) {
576 bd = allocGroup(req_blocks);
577 dbl_link_onto(bd, &stp->large_objects);
578 stp->n_large_blocks += bd->blocks; // might be larger than req_blocks
581 bd->flags = BF_LARGE;
582 bd->free = bd->start + n;
587 // small allocation (<LARGE_OBJECT_THRESHOLD) */
589 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
594 bd->link = stp->blocks;
611 return allocateInGen(g0,n);
615 allocatedBytes( void )
619 allocated = alloc_blocks * BLOCK_SIZE_W;
620 if (pinned_object_block != NULL) {
621 allocated -= (pinned_object_block->start + BLOCK_SIZE_W) -
622 pinned_object_block->free;
628 /* -----------------------------------------------------------------------------
631 This allocates memory in the current thread - it is intended for
632 use primarily from STG-land where we have a Capability. It is
633 better than allocate() because it doesn't require taking the
634 sm_mutex lock in the common case.
636 Memory is allocated directly from the nursery if possible (but not
637 from the current nursery block, so as not to interfere with
639 -------------------------------------------------------------------------- */
642 allocateLocal (Capability *cap, nat n)
647 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
648 return allocateInGen(g0,n);
651 /* small allocation (<LARGE_OBJECT_THRESHOLD) */
653 TICK_ALLOC_HEAP_NOCTR(n);
656 bd = cap->r.rCurrentAlloc;
657 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
659 // The CurrentAlloc block is full, we need to find another
660 // one. First, we try taking the next block from the
662 bd = cap->r.rCurrentNursery->link;
664 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
665 // The nursery is empty, or the next block is already
666 // full: allocate a fresh block (we can't fail here).
669 cap->r.rNursery->n_blocks++;
672 bd->step = cap->r.rNursery;
674 // NO: alloc_blocks++;
675 // calcAllocated() uses the size of the nursery, and we've
676 // already bumpted nursery->n_blocks above.
678 // we have a block in the nursery: take it and put
679 // it at the *front* of the nursery list, and use it
680 // to allocate() from.
681 cap->r.rCurrentNursery->link = bd->link;
682 if (bd->link != NULL) {
683 bd->link->u.back = cap->r.rCurrentNursery;
686 dbl_link_onto(bd, &cap->r.rNursery->blocks);
687 cap->r.rCurrentAlloc = bd;
688 IF_DEBUG(sanity, checkNurserySanity(cap->r.rNursery));
695 /* ---------------------------------------------------------------------------
696 Allocate a fixed/pinned object.
698 We allocate small pinned objects into a single block, allocating a
699 new block when the current one overflows. The block is chained
700 onto the large_object_list of generation 0 step 0.
702 NOTE: The GC can't in general handle pinned objects. This
703 interface is only safe to use for ByteArrays, which have no
704 pointers and don't require scavenging. It works because the
705 block's descriptor has the BF_LARGE flag set, so the block is
706 treated as a large object and chained onto various lists, rather
707 than the individual objects being copied. However, when it comes
708 to scavenge the block, the GC will only scavenge the first object.
709 The reason is that the GC can't linearly scan a block of pinned
710 objects at the moment (doing so would require using the
711 mostly-copying techniques). But since we're restricting ourselves
712 to pinned ByteArrays, not scavenging is ok.
714 This function is called by newPinnedByteArray# which immediately
715 fills the allocated memory with a MutableByteArray#.
716 ------------------------------------------------------------------------- */
719 allocatePinned( nat n )
722 bdescr *bd = pinned_object_block;
724 // If the request is for a large object, then allocate()
725 // will give us a pinned object anyway.
726 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
732 TICK_ALLOC_HEAP_NOCTR(n);
735 // we always return 8-byte aligned memory. bd->free must be
736 // 8-byte aligned to begin with, so we just round up n to
737 // the nearest multiple of 8 bytes.
738 if (sizeof(StgWord) == 4) {
742 // If we don't have a block of pinned objects yet, or the current
743 // one isn't large enough to hold the new object, allocate a new one.
744 if (bd == NULL || (bd->free + n) > (bd->start + BLOCK_SIZE_W)) {
745 pinned_object_block = bd = allocBlock();
746 dbl_link_onto(bd, &g0s0->large_objects);
747 g0s0->n_large_blocks++;
750 bd->flags = BF_PINNED | BF_LARGE;
751 bd->free = bd->start;
761 /* -----------------------------------------------------------------------------
763 -------------------------------------------------------------------------- */
766 This is the write barrier for MUT_VARs, a.k.a. IORefs. A
767 MUT_VAR_CLEAN object is not on the mutable list; a MUT_VAR_DIRTY
768 is. When written to, a MUT_VAR_CLEAN turns into a MUT_VAR_DIRTY
769 and is put on the mutable list.
772 dirty_MUT_VAR(StgRegTable *reg, StgClosure *p)
774 Capability *cap = regTableToCapability(reg);
776 if (p->header.info == &stg_MUT_VAR_CLEAN_info) {
777 p->header.info = &stg_MUT_VAR_DIRTY_info;
778 bd = Bdescr((StgPtr)p);
779 if (bd->gen_no > 0) recordMutableCap(p,cap,bd->gen_no);
784 This is the write barrier for MVARs. An MVAR_CLEAN objects is not
785 on the mutable list; a MVAR_DIRTY is. When written to, a
786 MVAR_CLEAN turns into a MVAR_DIRTY and is put on the mutable list.
787 The check for MVAR_CLEAN is inlined at the call site for speed,
788 this really does make a difference on concurrency-heavy benchmarks
789 such as Chaneneos and cheap-concurrency.
792 dirty_MVAR(StgRegTable *reg, StgClosure *p)
794 Capability *cap = regTableToCapability(reg);
796 bd = Bdescr((StgPtr)p);
797 if (bd->gen_no > 0) recordMutableCap(p,cap,bd->gen_no);
800 /* -----------------------------------------------------------------------------
801 Allocation functions for GMP.
803 These all use the allocate() interface - we can't have any garbage
804 collection going on during a gmp operation, so we use allocate()
805 which always succeeds. The gmp operations which might need to
806 allocate will ask the storage manager (via doYouWantToGC()) whether
807 a garbage collection is required, in case we get into a loop doing
808 only allocate() style allocation.
809 -------------------------------------------------------------------------- */
812 stgAllocForGMP (size_t size_in_bytes)
815 nat data_size_in_words, total_size_in_words;
817 /* round up to a whole number of words */
818 data_size_in_words = (size_in_bytes + sizeof(W_) + 1) / sizeof(W_);
819 total_size_in_words = sizeofW(StgArrWords) + data_size_in_words;
821 /* allocate and fill it in. */
822 #if defined(THREADED_RTS)
823 arr = (StgArrWords *)allocateLocal(myTask()->cap, total_size_in_words);
825 arr = (StgArrWords *)allocateLocal(&MainCapability, total_size_in_words);
827 SET_ARR_HDR(arr, &stg_ARR_WORDS_info, CCCS, data_size_in_words);
829 /* and return a ptr to the goods inside the array */
834 stgReallocForGMP (void *ptr, size_t old_size, size_t new_size)
836 void *new_stuff_ptr = stgAllocForGMP(new_size);
838 char *p = (char *) ptr;
839 char *q = (char *) new_stuff_ptr;
841 for (; i < old_size; i++, p++, q++) {
845 return(new_stuff_ptr);
849 stgDeallocForGMP (void *ptr STG_UNUSED,
850 size_t size STG_UNUSED)
852 /* easy for us: the garbage collector does the dealloc'n */
855 /* -----------------------------------------------------------------------------
857 * -------------------------------------------------------------------------- */
859 /* -----------------------------------------------------------------------------
862 * Approximate how much we've allocated: number of blocks in the
863 * nursery + blocks allocated via allocate() - unused nusery blocks.
864 * This leaves a little slop at the end of each block, and doesn't
865 * take into account large objects (ToDo).
866 * -------------------------------------------------------------------------- */
869 calcAllocated( void )
874 allocated = allocatedBytes();
875 allocated += countNurseryBlocks() * BLOCK_SIZE_W;
880 for (i = 0; i < n_nurseries; i++) {
882 for ( bd = capabilities[i].r.rCurrentNursery->link;
883 bd != NULL; bd = bd->link ) {
884 allocated -= BLOCK_SIZE_W;
886 cap = &capabilities[i];
887 if (cap->r.rCurrentNursery->free <
888 cap->r.rCurrentNursery->start + BLOCK_SIZE_W) {
889 allocated -= (cap->r.rCurrentNursery->start + BLOCK_SIZE_W)
890 - cap->r.rCurrentNursery->free;
894 bdescr *current_nursery = MainCapability.r.rCurrentNursery;
896 for ( bd = current_nursery->link; bd != NULL; bd = bd->link ) {
897 allocated -= BLOCK_SIZE_W;
899 if (current_nursery->free < current_nursery->start + BLOCK_SIZE_W) {
900 allocated -= (current_nursery->start + BLOCK_SIZE_W)
901 - current_nursery->free;
906 total_allocated += allocated;
910 /* Approximate the amount of live data in the heap. To be called just
911 * after garbage collection (see GarbageCollect()).
920 if (RtsFlags.GcFlags.generations == 1) {
921 return (g0s0->n_large_blocks + g0s0->n_blocks) * BLOCK_SIZE_W;
924 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
925 for (s = 0; s < generations[g].n_steps; s++) {
926 /* approximate amount of live data (doesn't take into account slop
927 * at end of each block).
929 if (g == 0 && s == 0) {
932 stp = &generations[g].steps[s];
933 live += (stp->n_large_blocks + stp->n_blocks) * BLOCK_SIZE_W;
939 /* Approximate the number of blocks that will be needed at the next
940 * garbage collection.
942 * Assume: all data currently live will remain live. Steps that will
943 * be collected next time will therefore need twice as many blocks
944 * since all the data will be copied.
953 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
954 for (s = 0; s < generations[g].n_steps; s++) {
955 if (g == 0 && s == 0) { continue; }
956 stp = &generations[g].steps[s];
957 if (generations[g].steps[0].n_blocks +
958 generations[g].steps[0].n_large_blocks
959 > generations[g].max_blocks
960 && stp->is_compacted == 0) {
961 needed += 2 * stp->n_blocks;
963 needed += stp->n_blocks;
970 /* ----------------------------------------------------------------------------
973 Executable memory must be managed separately from non-executable
974 memory. Most OSs these days require you to jump through hoops to
975 dynamically allocate executable memory, due to various security
978 Here we provide a small memory allocator for executable memory.
979 Memory is managed with a page granularity; we allocate linearly
980 in the page, and when the page is emptied (all objects on the page
981 are free) we free the page again, not forgetting to make it
984 TODO: The inability to handle objects bigger than BLOCK_SIZE_W means that
985 the linker cannot use allocateExec for loading object code files
986 on Windows. Once allocateExec can handle larger objects, the linker
987 should be modified to use allocateExec instead of VirtualAlloc.
988 ------------------------------------------------------------------------- */
990 static bdescr *exec_block;
992 void *allocateExec (nat bytes)
999 // round up to words.
1000 n = (bytes + sizeof(W_) + 1) / sizeof(W_);
1002 if (n+1 > BLOCK_SIZE_W) {
1003 barf("allocateExec: can't handle large objects");
1006 if (exec_block == NULL ||
1007 exec_block->free + n + 1 > exec_block->start + BLOCK_SIZE_W) {
1009 lnat pagesize = getPageSize();
1010 bd = allocGroup(stg_max(1, pagesize / BLOCK_SIZE));
1011 debugTrace(DEBUG_gc, "allocate exec block %p", bd->start);
1013 bd->flags = BF_EXEC;
1014 bd->link = exec_block;
1015 if (exec_block != NULL) {
1016 exec_block->u.back = bd;
1019 setExecutable(bd->start, bd->blocks * BLOCK_SIZE, rtsTrue);
1022 *(exec_block->free) = n; // store the size of this chunk
1023 exec_block->gen_no += n; // gen_no stores the number of words allocated
1024 ret = exec_block->free + 1;
1025 exec_block->free += n + 1;
1031 void freeExec (void *addr)
1033 StgPtr p = (StgPtr)addr - 1;
1034 bdescr *bd = Bdescr((StgPtr)p);
1036 if ((bd->flags & BF_EXEC) == 0) {
1037 barf("freeExec: not executable");
1040 if (*(StgPtr)p == 0) {
1041 barf("freeExec: already free?");
1046 bd->gen_no -= *(StgPtr)p;
1049 if (bd->gen_no == 0) {
1050 // Free the block if it is empty, but not if it is the block at
1051 // the head of the queue.
1052 if (bd != exec_block) {
1053 debugTrace(DEBUG_gc, "free exec block %p", bd->start);
1054 dbl_link_remove(bd, &exec_block);
1055 setExecutable(bd->start, bd->blocks * BLOCK_SIZE, rtsFalse);
1058 bd->free = bd->start;
1065 /* -----------------------------------------------------------------------------
1068 memInventory() checks for memory leaks by counting up all the
1069 blocks we know about and comparing that to the number of blocks
1070 allegedly floating around in the system.
1071 -------------------------------------------------------------------------- */
1076 countBlocks(bdescr *bd)
1079 for (n=0; bd != NULL; bd=bd->link) {
1085 // (*1) Just like countBlocks, except that we adjust the count for a
1086 // megablock group so that it doesn't include the extra few blocks
1087 // that would be taken up by block descriptors in the second and
1088 // subsequent megablock. This is so we can tally the count with the
1089 // number of blocks allocated in the system, for memInventory().
1091 countAllocdBlocks(bdescr *bd)
1094 for (n=0; bd != NULL; bd=bd->link) {
1096 // hack for megablock groups: see (*1) above
1097 if (bd->blocks > BLOCKS_PER_MBLOCK) {
1098 n -= (MBLOCK_SIZE / BLOCK_SIZE - BLOCKS_PER_MBLOCK)
1099 * (bd->blocks/(MBLOCK_SIZE/BLOCK_SIZE));
1106 stepBlocks (step *stp)
1108 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
1109 ASSERT(countBlocks(stp->large_objects) == stp->n_large_blocks);
1110 return stp->n_blocks + stp->n_old_blocks +
1111 countAllocdBlocks(stp->large_objects);
1119 lnat gen_blocks[RtsFlags.GcFlags.generations];
1120 lnat nursery_blocks, retainer_blocks,
1121 arena_blocks, exec_blocks;
1122 lnat live_blocks = 0, free_blocks = 0;
1124 // count the blocks we current have
1126 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1128 for (i = 0; i < n_capabilities; i++) {
1129 gen_blocks[g] += countBlocks(capabilities[i].mut_lists[g]);
1131 gen_blocks[g] += countAllocdBlocks(generations[g].mut_list);
1132 for (s = 0; s < generations[g].n_steps; s++) {
1133 stp = &generations[g].steps[s];
1134 gen_blocks[g] += stepBlocks(stp);
1139 for (i = 0; i < n_nurseries; i++) {
1140 nursery_blocks += stepBlocks(&nurseries[i]);
1143 retainer_blocks = 0;
1145 if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_RETAINER) {
1146 retainer_blocks = retainerStackBlocks();
1150 // count the blocks allocated by the arena allocator
1151 arena_blocks = arenaBlocks();
1153 // count the blocks containing executable memory
1154 exec_blocks = countAllocdBlocks(exec_block);
1156 /* count the blocks on the free list */
1157 free_blocks = countFreeList();
1160 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1161 live_blocks += gen_blocks[g];
1163 live_blocks += nursery_blocks +
1164 + retainer_blocks + arena_blocks + exec_blocks;
1166 if (live_blocks + free_blocks != mblocks_allocated * BLOCKS_PER_MBLOCK)
1168 debugBelch("Memory leak detected\n");
1169 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1170 debugBelch(" gen %d blocks : %4lu\n", g, gen_blocks[g]);
1172 debugBelch(" nursery : %4lu\n", nursery_blocks);
1173 debugBelch(" retainer : %4lu\n", retainer_blocks);
1174 debugBelch(" arena blocks : %4lu\n", arena_blocks);
1175 debugBelch(" exec : %4lu\n", exec_blocks);
1176 debugBelch(" free : %4lu\n", free_blocks);
1177 debugBelch(" total : %4lu\n\n", live_blocks + free_blocks);
1178 debugBelch(" in system : %4lu\n", mblocks_allocated * BLOCKS_PER_MBLOCK);
1184 /* Full heap sanity check. */
1190 if (RtsFlags.GcFlags.generations == 1) {
1191 checkHeap(g0s0->blocks);
1192 checkChain(g0s0->large_objects);
1195 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1196 for (s = 0; s < generations[g].n_steps; s++) {
1197 if (g == 0 && s == 0) { continue; }
1198 ASSERT(countBlocks(generations[g].steps[s].blocks)
1199 == generations[g].steps[s].n_blocks);
1200 ASSERT(countBlocks(generations[g].steps[s].large_objects)
1201 == generations[g].steps[s].n_large_blocks);
1202 checkHeap(generations[g].steps[s].blocks);
1203 checkChain(generations[g].steps[s].large_objects);
1205 checkMutableList(generations[g].mut_list, g);
1210 for (s = 0; s < n_nurseries; s++) {
1211 ASSERT(countBlocks(nurseries[s].blocks)
1212 == nurseries[s].n_blocks);
1213 ASSERT(countBlocks(nurseries[s].large_objects)
1214 == nurseries[s].n_large_blocks);
1217 checkFreeListSanity();
1221 /* Nursery sanity check */
1223 checkNurserySanity( step *stp )
1229 for (bd = stp->blocks; bd != NULL; bd = bd->link) {
1230 ASSERT(bd->u.back == prev);
1232 blocks += bd->blocks;
1234 ASSERT(blocks == stp->n_blocks);
1237 // handy function for use in gdb, because Bdescr() is inlined.
1238 extern bdescr *_bdescr( StgPtr p );