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)
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 */
54 ullong total_allocated = 0; /* total memory allocated during run */
56 nat n_nurseries = 0; /* == RtsFlags.ParFlags.nNodes, convenience */
57 step *nurseries = NULL; /* array of nurseries, >1 only if THREADED_RTS */
61 * Storage manager mutex: protects all the above state from
62 * simultaneous access by two STG threads.
66 * This mutex is used by atomicModifyMutVar# only
68 Mutex atomic_modify_mutvar_mutex;
75 static void *stgAllocForGMP (size_t size_in_bytes);
76 static void *stgReallocForGMP (void *ptr, size_t old_size, size_t new_size);
77 static void stgDeallocForGMP (void *ptr, size_t size);
80 initStep (step *stp, int g, int s)
85 stp->old_blocks = NULL;
86 stp->n_old_blocks = 0;
87 stp->gen = &generations[g];
89 stp->large_objects = NULL;
90 stp->n_large_blocks = 0;
91 stp->scavenged_large_objects = NULL;
92 stp->n_scavenged_large_blocks = 0;
93 stp->is_compacted = 0;
96 initSpinLock(&stp->sync_todo);
97 initSpinLock(&stp->sync_large_objects);
107 if (generations != NULL) {
108 // multi-init protection
114 /* Sanity check to make sure the LOOKS_LIKE_ macros appear to be
115 * doing something reasonable.
117 /* We use the NOT_NULL variant or gcc warns that the test is always true */
118 ASSERT(LOOKS_LIKE_INFO_PTR_NOT_NULL(&stg_BLACKHOLE_info));
119 ASSERT(LOOKS_LIKE_CLOSURE_PTR(&stg_dummy_ret_closure));
120 ASSERT(!HEAP_ALLOCED(&stg_dummy_ret_closure));
122 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
123 RtsFlags.GcFlags.heapSizeSuggestion >
124 RtsFlags.GcFlags.maxHeapSize) {
125 RtsFlags.GcFlags.maxHeapSize = RtsFlags.GcFlags.heapSizeSuggestion;
128 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
129 RtsFlags.GcFlags.minAllocAreaSize >
130 RtsFlags.GcFlags.maxHeapSize) {
131 errorBelch("maximum heap size (-M) is smaller than minimum alloc area size (-A)");
132 RtsFlags.GcFlags.minAllocAreaSize = RtsFlags.GcFlags.maxHeapSize;
135 initBlockAllocator();
137 #if defined(THREADED_RTS)
138 initMutex(&sm_mutex);
139 initMutex(&atomic_modify_mutvar_mutex);
144 /* allocate generation info array */
145 generations = (generation *)stgMallocBytes(RtsFlags.GcFlags.generations
146 * sizeof(struct generation_),
147 "initStorage: gens");
149 /* Initialise all generations */
150 for(g = 0; g < RtsFlags.GcFlags.generations; g++) {
151 gen = &generations[g];
153 gen->mut_list = allocBlock();
154 gen->collections = 0;
155 gen->failed_promotions = 0;
159 /* A couple of convenience pointers */
160 g0 = &generations[0];
161 oldest_gen = &generations[RtsFlags.GcFlags.generations-1];
163 /* Allocate step structures in each generation */
164 if (RtsFlags.GcFlags.generations > 1) {
165 /* Only for multiple-generations */
167 /* Oldest generation: one step */
168 oldest_gen->n_steps = 1;
170 stgMallocBytes(1 * sizeof(struct step_), "initStorage: last step");
172 /* set up all except the oldest generation with 2 steps */
173 for(g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
174 generations[g].n_steps = RtsFlags.GcFlags.steps;
175 generations[g].steps =
176 stgMallocBytes (RtsFlags.GcFlags.steps * sizeof(struct step_),
177 "initStorage: steps");
181 /* single generation, i.e. a two-space collector */
183 g0->steps = stgMallocBytes (sizeof(struct step_), "initStorage: steps");
187 n_nurseries = n_capabilities;
191 nurseries = stgMallocBytes (n_nurseries * sizeof(struct step_),
192 "initStorage: nurseries");
194 /* Initialise all steps */
195 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
196 for (s = 0; s < generations[g].n_steps; s++) {
197 initStep(&generations[g].steps[s], g, s);
201 for (s = 0; s < n_nurseries; s++) {
202 initStep(&nurseries[s], 0, s);
205 /* Set up the destination pointers in each younger gen. step */
206 for (g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
207 for (s = 0; s < generations[g].n_steps-1; s++) {
208 generations[g].steps[s].to = &generations[g].steps[s+1];
210 generations[g].steps[s].to = &generations[g+1].steps[0];
212 oldest_gen->steps[0].to = &oldest_gen->steps[0];
214 for (s = 0; s < n_nurseries; s++) {
215 nurseries[s].to = generations[0].steps[0].to;
218 /* The oldest generation has one step. */
219 if (RtsFlags.GcFlags.compact) {
220 if (RtsFlags.GcFlags.generations == 1) {
221 errorBelch("WARNING: compaction is incompatible with -G1; disabled");
223 oldest_gen->steps[0].is_compacted = 1;
227 generations[0].max_blocks = 0;
228 g0s0 = &generations[0].steps[0];
230 /* The allocation area. Policy: keep the allocation area
231 * small to begin with, even if we have a large suggested heap
232 * size. Reason: we're going to do a major collection first, and we
233 * don't want it to be a big one. This vague idea is borne out by
234 * rigorous experimental evidence.
238 weak_ptr_list = NULL;
240 revertible_caf_list = NULL;
242 /* initialise the allocate() interface */
244 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
246 /* Tell GNU multi-precision pkg about our custom alloc functions */
247 mp_set_memory_functions(stgAllocForGMP, stgReallocForGMP, stgDeallocForGMP);
250 initSpinLock(&gc_alloc_block_sync);
253 IF_DEBUG(gc, statDescribeGens());
261 stat_exit(calcAllocated());
269 for(g = 0; g < RtsFlags.GcFlags.generations; g++)
270 stgFree(generations[g].steps);
271 stgFree(generations);
273 #if defined(THREADED_RTS)
274 closeMutex(&sm_mutex);
275 closeMutex(&atomic_modify_mutvar_mutex);
280 /* -----------------------------------------------------------------------------
283 The entry code for every CAF does the following:
285 - builds a CAF_BLACKHOLE in the heap
286 - pushes an update frame pointing to the CAF_BLACKHOLE
287 - invokes UPD_CAF(), which:
288 - calls newCaf, below
289 - updates the CAF with a static indirection to the CAF_BLACKHOLE
291 Why do we build a BLACKHOLE in the heap rather than just updating
292 the thunk directly? It's so that we only need one kind of update
293 frame - otherwise we'd need a static version of the update frame too.
295 newCaf() does the following:
297 - it puts the CAF on the oldest generation's mut-once list.
298 This is so that we can treat the CAF as a root when collecting
301 For GHCI, we have additional requirements when dealing with CAFs:
303 - we must *retain* all dynamically-loaded CAFs ever entered,
304 just in case we need them again.
305 - we must be able to *revert* CAFs that have been evaluated, to
306 their pre-evaluated form.
308 To do this, we use an additional CAF list. When newCaf() is
309 called on a dynamically-loaded CAF, we add it to the CAF list
310 instead of the old-generation mutable list, and save away its
311 old info pointer (in caf->saved_info) for later reversion.
313 To revert all the CAFs, we traverse the CAF list and reset the
314 info pointer to caf->saved_info, then throw away the CAF list.
315 (see GC.c:revertCAFs()).
319 -------------------------------------------------------------------------- */
322 newCAF(StgClosure* caf)
329 // If we are in GHCi _and_ we are using dynamic libraries,
330 // then we can't redirect newCAF calls to newDynCAF (see below),
331 // so we make newCAF behave almost like newDynCAF.
332 // The dynamic libraries might be used by both the interpreted
333 // program and GHCi itself, so they must not be reverted.
334 // This also means that in GHCi with dynamic libraries, CAFs are not
335 // garbage collected. If this turns out to be a problem, we could
336 // do another hack here and do an address range test on caf to figure
337 // out whether it is from a dynamic library.
338 ((StgIndStatic *)caf)->saved_info = (StgInfoTable *)caf->header.info;
339 ((StgIndStatic *)caf)->static_link = caf_list;
344 /* Put this CAF on the mutable list for the old generation.
345 * This is a HACK - the IND_STATIC closure doesn't really have
346 * a mut_link field, but we pretend it has - in fact we re-use
347 * the STATIC_LINK field for the time being, because when we
348 * come to do a major GC we won't need the mut_link field
349 * any more and can use it as a STATIC_LINK.
351 ((StgIndStatic *)caf)->saved_info = NULL;
352 recordMutableGen(caf, oldest_gen);
358 // An alternate version of newCaf which is used for dynamically loaded
359 // object code in GHCi. In this case we want to retain *all* CAFs in
360 // the object code, because they might be demanded at any time from an
361 // expression evaluated on the command line.
362 // Also, GHCi might want to revert CAFs, so we add these to the
363 // revertible_caf_list.
365 // The linker hackily arranges that references to newCaf from dynamic
366 // code end up pointing to newDynCAF.
368 newDynCAF(StgClosure *caf)
372 ((StgIndStatic *)caf)->saved_info = (StgInfoTable *)caf->header.info;
373 ((StgIndStatic *)caf)->static_link = revertible_caf_list;
374 revertible_caf_list = caf;
379 /* -----------------------------------------------------------------------------
381 -------------------------------------------------------------------------- */
384 allocNursery (step *stp, bdescr *tail, nat blocks)
389 // Allocate a nursery: we allocate fresh blocks one at a time and
390 // cons them on to the front of the list, not forgetting to update
391 // the back pointer on the tail of the list to point to the new block.
392 for (i=0; i < blocks; i++) {
395 processNursery() in LdvProfile.c assumes that every block group in
396 the nursery contains only a single block. So, if a block group is
397 given multiple blocks, change processNursery() accordingly.
401 // double-link the nursery: we might need to insert blocks
408 bd->free = bd->start;
416 assignNurseriesToCapabilities (void)
421 for (i = 0; i < n_nurseries; i++) {
422 capabilities[i].r.rNursery = &nurseries[i];
423 capabilities[i].r.rCurrentNursery = nurseries[i].blocks;
424 capabilities[i].r.rCurrentAlloc = NULL;
426 #else /* THREADED_RTS */
427 MainCapability.r.rNursery = &nurseries[0];
428 MainCapability.r.rCurrentNursery = nurseries[0].blocks;
429 MainCapability.r.rCurrentAlloc = NULL;
434 allocNurseries( void )
438 for (i = 0; i < n_nurseries; i++) {
439 nurseries[i].blocks =
440 allocNursery(&nurseries[i], NULL,
441 RtsFlags.GcFlags.minAllocAreaSize);
442 nurseries[i].n_blocks = RtsFlags.GcFlags.minAllocAreaSize;
443 nurseries[i].old_blocks = NULL;
444 nurseries[i].n_old_blocks = 0;
446 assignNurseriesToCapabilities();
450 resetNurseries( void )
456 for (i = 0; i < n_nurseries; i++) {
458 for (bd = stp->blocks; bd; bd = bd->link) {
459 bd->free = bd->start;
460 ASSERT(bd->gen_no == 0);
461 ASSERT(bd->step == stp);
462 IF_DEBUG(sanity,memset(bd->start, 0xaa, BLOCK_SIZE));
465 assignNurseriesToCapabilities();
469 countNurseryBlocks (void)
474 for (i = 0; i < n_nurseries; i++) {
475 blocks += nurseries[i].n_blocks;
481 resizeNursery ( step *stp, nat blocks )
486 nursery_blocks = stp->n_blocks;
487 if (nursery_blocks == blocks) return;
489 if (nursery_blocks < blocks) {
490 debugTrace(DEBUG_gc, "increasing size of nursery to %d blocks",
492 stp->blocks = allocNursery(stp, stp->blocks, blocks-nursery_blocks);
497 debugTrace(DEBUG_gc, "decreasing size of nursery to %d blocks",
501 while (nursery_blocks > blocks) {
503 next_bd->u.back = NULL;
504 nursery_blocks -= bd->blocks; // might be a large block
509 // might have gone just under, by freeing a large block, so make
510 // up the difference.
511 if (nursery_blocks < blocks) {
512 stp->blocks = allocNursery(stp, stp->blocks, blocks-nursery_blocks);
516 stp->n_blocks = blocks;
517 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
521 // Resize each of the nurseries to the specified size.
524 resizeNurseriesFixed (nat blocks)
527 for (i = 0; i < n_nurseries; i++) {
528 resizeNursery(&nurseries[i], blocks);
533 // Resize the nurseries to the total specified size.
536 resizeNurseries (nat blocks)
538 // If there are multiple nurseries, then we just divide the number
539 // of available blocks between them.
540 resizeNurseriesFixed(blocks / n_nurseries);
543 /* -----------------------------------------------------------------------------
544 The allocate() interface
546 allocateInGen() function allocates memory directly into a specific
547 generation. It always succeeds, and returns a chunk of memory n
548 words long. n can be larger than the size of a block if necessary,
549 in which case a contiguous block group will be allocated.
551 allocate(n) is equivalent to allocateInGen(g0).
552 -------------------------------------------------------------------------- */
555 allocateInGen (generation *g, nat n)
563 TICK_ALLOC_HEAP_NOCTR(n);
568 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_))
570 nat req_blocks = (lnat)BLOCK_ROUND_UP(n*sizeof(W_)) / BLOCK_SIZE;
572 // Attempting to allocate an object larger than maxHeapSize
573 // should definitely be disallowed. (bug #1791)
574 if (RtsFlags.GcFlags.maxHeapSize > 0 &&
575 req_blocks >= RtsFlags.GcFlags.maxHeapSize) {
579 bd = allocGroup(req_blocks);
580 dbl_link_onto(bd, &stp->large_objects);
581 stp->n_large_blocks += bd->blocks; // might be larger than req_blocks
584 bd->flags = BF_LARGE;
585 bd->free = bd->start + n;
590 // small allocation (<LARGE_OBJECT_THRESHOLD) */
592 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
597 bd->link = stp->blocks;
614 return allocateInGen(g0,n);
618 allocatedBytes( void )
622 allocated = alloc_blocks * BLOCK_SIZE_W;
623 if (pinned_object_block != NULL) {
624 allocated -= (pinned_object_block->start + BLOCK_SIZE_W) -
625 pinned_object_block->free;
631 /* -----------------------------------------------------------------------------
634 This allocates memory in the current thread - it is intended for
635 use primarily from STG-land where we have a Capability. It is
636 better than allocate() because it doesn't require taking the
637 sm_mutex lock in the common case.
639 Memory is allocated directly from the nursery if possible (but not
640 from the current nursery block, so as not to interfere with
642 -------------------------------------------------------------------------- */
645 allocateLocal (Capability *cap, nat n)
650 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
651 return allocateInGen(g0,n);
654 /* small allocation (<LARGE_OBJECT_THRESHOLD) */
656 TICK_ALLOC_HEAP_NOCTR(n);
659 bd = cap->r.rCurrentAlloc;
660 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
662 // The CurrentAlloc block is full, we need to find another
663 // one. First, we try taking the next block from the
665 bd = cap->r.rCurrentNursery->link;
667 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
668 // The nursery is empty, or the next block is already
669 // full: allocate a fresh block (we can't fail here).
672 cap->r.rNursery->n_blocks++;
675 bd->step = cap->r.rNursery;
677 // NO: alloc_blocks++;
678 // calcAllocated() uses the size of the nursery, and we've
679 // already bumpted nursery->n_blocks above.
681 // we have a block in the nursery: take it and put
682 // it at the *front* of the nursery list, and use it
683 // to allocate() from.
684 cap->r.rCurrentNursery->link = bd->link;
685 if (bd->link != NULL) {
686 bd->link->u.back = cap->r.rCurrentNursery;
689 dbl_link_onto(bd, &cap->r.rNursery->blocks);
690 cap->r.rCurrentAlloc = bd;
691 IF_DEBUG(sanity, checkNurserySanity(cap->r.rNursery));
698 /* ---------------------------------------------------------------------------
699 Allocate a fixed/pinned object.
701 We allocate small pinned objects into a single block, allocating a
702 new block when the current one overflows. The block is chained
703 onto the large_object_list of generation 0 step 0.
705 NOTE: The GC can't in general handle pinned objects. This
706 interface is only safe to use for ByteArrays, which have no
707 pointers and don't require scavenging. It works because the
708 block's descriptor has the BF_LARGE flag set, so the block is
709 treated as a large object and chained onto various lists, rather
710 than the individual objects being copied. However, when it comes
711 to scavenge the block, the GC will only scavenge the first object.
712 The reason is that the GC can't linearly scan a block of pinned
713 objects at the moment (doing so would require using the
714 mostly-copying techniques). But since we're restricting ourselves
715 to pinned ByteArrays, not scavenging is ok.
717 This function is called by newPinnedByteArray# which immediately
718 fills the allocated memory with a MutableByteArray#.
719 ------------------------------------------------------------------------- */
722 allocatePinned( nat n )
725 bdescr *bd = pinned_object_block;
727 // If the request is for a large object, then allocate()
728 // will give us a pinned object anyway.
729 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
735 TICK_ALLOC_HEAP_NOCTR(n);
738 // we always return 8-byte aligned memory. bd->free must be
739 // 8-byte aligned to begin with, so we just round up n to
740 // the nearest multiple of 8 bytes.
741 if (sizeof(StgWord) == 4) {
745 // If we don't have a block of pinned objects yet, or the current
746 // one isn't large enough to hold the new object, allocate a new one.
747 if (bd == NULL || (bd->free + n) > (bd->start + BLOCK_SIZE_W)) {
748 pinned_object_block = bd = allocBlock();
749 dbl_link_onto(bd, &g0s0->large_objects);
750 g0s0->n_large_blocks++;
753 bd->flags = BF_PINNED | BF_LARGE;
754 bd->free = bd->start;
764 /* -----------------------------------------------------------------------------
766 -------------------------------------------------------------------------- */
769 This is the write barrier for MUT_VARs, a.k.a. IORefs. A
770 MUT_VAR_CLEAN object is not on the mutable list; a MUT_VAR_DIRTY
771 is. When written to, a MUT_VAR_CLEAN turns into a MUT_VAR_DIRTY
772 and is put on the mutable list.
775 dirty_MUT_VAR(StgRegTable *reg, StgClosure *p)
777 Capability *cap = regTableToCapability(reg);
779 if (p->header.info == &stg_MUT_VAR_CLEAN_info) {
780 p->header.info = &stg_MUT_VAR_DIRTY_info;
781 bd = Bdescr((StgPtr)p);
782 if (bd->gen_no > 0) recordMutableCap(p,cap,bd->gen_no);
787 This is the write barrier for MVARs. An MVAR_CLEAN objects is not
788 on the mutable list; a MVAR_DIRTY is. When written to, a
789 MVAR_CLEAN turns into a MVAR_DIRTY and is put on the mutable list.
790 The check for MVAR_CLEAN is inlined at the call site for speed,
791 this really does make a difference on concurrency-heavy benchmarks
792 such as Chaneneos and cheap-concurrency.
795 dirty_MVAR(StgRegTable *reg, StgClosure *p)
797 Capability *cap = regTableToCapability(reg);
799 bd = Bdescr((StgPtr)p);
800 if (bd->gen_no > 0) recordMutableCap(p,cap,bd->gen_no);
803 /* -----------------------------------------------------------------------------
804 Allocation functions for GMP.
806 These all use the allocate() interface - we can't have any garbage
807 collection going on during a gmp operation, so we use allocate()
808 which always succeeds. The gmp operations which might need to
809 allocate will ask the storage manager (via doYouWantToGC()) whether
810 a garbage collection is required, in case we get into a loop doing
811 only allocate() style allocation.
812 -------------------------------------------------------------------------- */
815 stgAllocForGMP (size_t size_in_bytes)
818 nat data_size_in_words, total_size_in_words;
820 /* round up to a whole number of words */
821 data_size_in_words = (size_in_bytes + sizeof(W_) + 1) / sizeof(W_);
822 total_size_in_words = sizeofW(StgArrWords) + data_size_in_words;
824 /* allocate and fill it in. */
825 #if defined(THREADED_RTS)
826 arr = (StgArrWords *)allocateLocal(myTask()->cap, total_size_in_words);
828 arr = (StgArrWords *)allocateLocal(&MainCapability, total_size_in_words);
830 SET_ARR_HDR(arr, &stg_ARR_WORDS_info, CCCS, data_size_in_words);
832 /* and return a ptr to the goods inside the array */
837 stgReallocForGMP (void *ptr, size_t old_size, size_t new_size)
839 void *new_stuff_ptr = stgAllocForGMP(new_size);
841 char *p = (char *) ptr;
842 char *q = (char *) new_stuff_ptr;
844 for (; i < old_size; i++, p++, q++) {
848 return(new_stuff_ptr);
852 stgDeallocForGMP (void *ptr STG_UNUSED,
853 size_t size STG_UNUSED)
855 /* easy for us: the garbage collector does the dealloc'n */
858 /* -----------------------------------------------------------------------------
860 * -------------------------------------------------------------------------- */
862 /* -----------------------------------------------------------------------------
865 * Approximate how much we've allocated: number of blocks in the
866 * nursery + blocks allocated via allocate() - unused nusery blocks.
867 * This leaves a little slop at the end of each block, and doesn't
868 * take into account large objects (ToDo).
869 * -------------------------------------------------------------------------- */
872 calcAllocated( void )
877 allocated = allocatedBytes();
878 allocated += countNurseryBlocks() * BLOCK_SIZE_W;
883 for (i = 0; i < n_nurseries; i++) {
885 for ( bd = capabilities[i].r.rCurrentNursery->link;
886 bd != NULL; bd = bd->link ) {
887 allocated -= BLOCK_SIZE_W;
889 cap = &capabilities[i];
890 if (cap->r.rCurrentNursery->free <
891 cap->r.rCurrentNursery->start + BLOCK_SIZE_W) {
892 allocated -= (cap->r.rCurrentNursery->start + BLOCK_SIZE_W)
893 - cap->r.rCurrentNursery->free;
897 bdescr *current_nursery = MainCapability.r.rCurrentNursery;
899 for ( bd = current_nursery->link; bd != NULL; bd = bd->link ) {
900 allocated -= BLOCK_SIZE_W;
902 if (current_nursery->free < current_nursery->start + BLOCK_SIZE_W) {
903 allocated -= (current_nursery->start + BLOCK_SIZE_W)
904 - current_nursery->free;
909 total_allocated += allocated;
913 /* Approximate the amount of live data in the heap. To be called just
914 * after garbage collection (see GarbageCollect()).
923 if (RtsFlags.GcFlags.generations == 1) {
924 return (g0s0->n_large_blocks + g0s0->n_blocks) * BLOCK_SIZE_W;
927 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
928 for (s = 0; s < generations[g].n_steps; s++) {
929 /* approximate amount of live data (doesn't take into account slop
930 * at end of each block).
932 if (g == 0 && s == 0) {
935 stp = &generations[g].steps[s];
936 live += (stp->n_large_blocks + stp->n_blocks) * BLOCK_SIZE_W;
942 /* Approximate the number of blocks that will be needed at the next
943 * garbage collection.
945 * Assume: all data currently live will remain live. Steps that will
946 * be collected next time will therefore need twice as many blocks
947 * since all the data will be copied.
956 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
957 for (s = 0; s < generations[g].n_steps; s++) {
958 if (g == 0 && s == 0) { continue; }
959 stp = &generations[g].steps[s];
960 if (generations[g].steps[0].n_blocks +
961 generations[g].steps[0].n_large_blocks
962 > generations[g].max_blocks
963 && stp->is_compacted == 0) {
964 needed += 2 * stp->n_blocks;
966 needed += stp->n_blocks;
973 /* ----------------------------------------------------------------------------
976 Executable memory must be managed separately from non-executable
977 memory. Most OSs these days require you to jump through hoops to
978 dynamically allocate executable memory, due to various security
981 Here we provide a small memory allocator for executable memory.
982 Memory is managed with a page granularity; we allocate linearly
983 in the page, and when the page is emptied (all objects on the page
984 are free) we free the page again, not forgetting to make it
987 TODO: The inability to handle objects bigger than BLOCK_SIZE_W means that
988 the linker cannot use allocateExec for loading object code files
989 on Windows. Once allocateExec can handle larger objects, the linker
990 should be modified to use allocateExec instead of VirtualAlloc.
991 ------------------------------------------------------------------------- */
993 static bdescr *exec_block;
995 void *allocateExec (nat bytes)
1002 // round up to words.
1003 n = (bytes + sizeof(W_) + 1) / sizeof(W_);
1005 if (n+1 > BLOCK_SIZE_W) {
1006 barf("allocateExec: can't handle large objects");
1009 if (exec_block == NULL ||
1010 exec_block->free + n + 1 > exec_block->start + BLOCK_SIZE_W) {
1012 lnat pagesize = getPageSize();
1013 bd = allocGroup(stg_max(1, pagesize / BLOCK_SIZE));
1014 debugTrace(DEBUG_gc, "allocate exec block %p", bd->start);
1016 bd->flags = BF_EXEC;
1017 bd->link = exec_block;
1018 if (exec_block != NULL) {
1019 exec_block->u.back = bd;
1022 setExecutable(bd->start, bd->blocks * BLOCK_SIZE, rtsTrue);
1025 *(exec_block->free) = n; // store the size of this chunk
1026 exec_block->gen_no += n; // gen_no stores the number of words allocated
1027 ret = exec_block->free + 1;
1028 exec_block->free += n + 1;
1034 void freeExec (void *addr)
1036 StgPtr p = (StgPtr)addr - 1;
1037 bdescr *bd = Bdescr((StgPtr)p);
1039 if ((bd->flags & BF_EXEC) == 0) {
1040 barf("freeExec: not executable");
1043 if (*(StgPtr)p == 0) {
1044 barf("freeExec: already free?");
1049 bd->gen_no -= *(StgPtr)p;
1052 if (bd->gen_no == 0) {
1053 // Free the block if it is empty, but not if it is the block at
1054 // the head of the queue.
1055 if (bd != exec_block) {
1056 debugTrace(DEBUG_gc, "free exec block %p", bd->start);
1057 dbl_link_remove(bd, &exec_block);
1058 setExecutable(bd->start, bd->blocks * BLOCK_SIZE, rtsFalse);
1061 bd->free = bd->start;
1068 /* -----------------------------------------------------------------------------
1071 memInventory() checks for memory leaks by counting up all the
1072 blocks we know about and comparing that to the number of blocks
1073 allegedly floating around in the system.
1074 -------------------------------------------------------------------------- */
1079 countBlocks(bdescr *bd)
1082 for (n=0; bd != NULL; bd=bd->link) {
1088 // (*1) Just like countBlocks, except that we adjust the count for a
1089 // megablock group so that it doesn't include the extra few blocks
1090 // that would be taken up by block descriptors in the second and
1091 // subsequent megablock. This is so we can tally the count with the
1092 // number of blocks allocated in the system, for memInventory().
1094 countAllocdBlocks(bdescr *bd)
1097 for (n=0; bd != NULL; bd=bd->link) {
1099 // hack for megablock groups: see (*1) above
1100 if (bd->blocks > BLOCKS_PER_MBLOCK) {
1101 n -= (MBLOCK_SIZE / BLOCK_SIZE - BLOCKS_PER_MBLOCK)
1102 * (bd->blocks/(MBLOCK_SIZE/BLOCK_SIZE));
1109 stepBlocks (step *stp)
1111 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
1112 ASSERT(countBlocks(stp->large_objects) == stp->n_large_blocks);
1113 return stp->n_blocks + stp->n_old_blocks +
1114 countAllocdBlocks(stp->large_objects);
1122 lnat gen_blocks[RtsFlags.GcFlags.generations];
1123 lnat nursery_blocks, retainer_blocks,
1124 arena_blocks, exec_blocks;
1125 lnat live_blocks = 0, free_blocks = 0;
1127 // count the blocks we current have
1129 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1131 for (i = 0; i < n_capabilities; i++) {
1132 gen_blocks[g] += countBlocks(capabilities[i].mut_lists[g]);
1134 gen_blocks[g] += countAllocdBlocks(generations[g].mut_list);
1135 for (s = 0; s < generations[g].n_steps; s++) {
1136 stp = &generations[g].steps[s];
1137 gen_blocks[g] += stepBlocks(stp);
1142 for (i = 0; i < n_nurseries; i++) {
1143 nursery_blocks += stepBlocks(&nurseries[i]);
1146 retainer_blocks = 0;
1148 if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_RETAINER) {
1149 retainer_blocks = retainerStackBlocks();
1153 // count the blocks allocated by the arena allocator
1154 arena_blocks = arenaBlocks();
1156 // count the blocks containing executable memory
1157 exec_blocks = countAllocdBlocks(exec_block);
1159 /* count the blocks on the free list */
1160 free_blocks = countFreeList();
1163 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1164 live_blocks += gen_blocks[g];
1166 live_blocks += nursery_blocks +
1167 + retainer_blocks + arena_blocks + exec_blocks;
1169 if (live_blocks + free_blocks != mblocks_allocated * BLOCKS_PER_MBLOCK)
1171 debugBelch("Memory leak detected\n");
1172 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1173 debugBelch(" gen %d blocks : %4lu\n", g, gen_blocks[g]);
1175 debugBelch(" nursery : %4lu\n", nursery_blocks);
1176 debugBelch(" retainer : %4lu\n", retainer_blocks);
1177 debugBelch(" arena blocks : %4lu\n", arena_blocks);
1178 debugBelch(" exec : %4lu\n", exec_blocks);
1179 debugBelch(" free : %4lu\n", free_blocks);
1180 debugBelch(" total : %4lu\n\n", live_blocks + free_blocks);
1181 debugBelch(" in system : %4lu\n", mblocks_allocated * BLOCKS_PER_MBLOCK);
1187 /* Full heap sanity check. */
1193 if (RtsFlags.GcFlags.generations == 1) {
1194 checkHeap(g0s0->blocks);
1195 checkChain(g0s0->large_objects);
1198 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1199 for (s = 0; s < generations[g].n_steps; s++) {
1200 if (g == 0 && s == 0) { continue; }
1201 ASSERT(countBlocks(generations[g].steps[s].blocks)
1202 == generations[g].steps[s].n_blocks);
1203 ASSERT(countBlocks(generations[g].steps[s].large_objects)
1204 == generations[g].steps[s].n_large_blocks);
1205 checkHeap(generations[g].steps[s].blocks);
1206 checkChain(generations[g].steps[s].large_objects);
1208 checkMutableList(generations[g].mut_list, g);
1213 for (s = 0; s < n_nurseries; s++) {
1214 ASSERT(countBlocks(nurseries[s].blocks)
1215 == nurseries[s].n_blocks);
1216 ASSERT(countBlocks(nurseries[s].large_objects)
1217 == nurseries[s].n_large_blocks);
1220 checkFreeListSanity();
1224 /* Nursery sanity check */
1226 checkNurserySanity( step *stp )
1232 for (bd = stp->blocks; bd != NULL; bd = bd->link) {
1233 ASSERT(bd->u.back == prev);
1235 blocks += bd->blocks;
1237 ASSERT(blocks == stp->n_blocks);
1240 // handy function for use in gdb, because Bdescr() is inlined.
1241 extern bdescr *_bdescr( StgPtr p );