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)
41 * All these globals require sm_mutex to access in THREADED_RTS mode.
43 StgClosure *caf_list = NULL;
44 StgClosure *revertible_caf_list = NULL;
47 bdescr *pinned_object_block; /* allocate pinned objects into this block */
48 nat alloc_blocks; /* number of allocate()d blocks since GC */
49 nat alloc_blocks_lim; /* approximate limit on alloc_blocks */
51 static bdescr *exec_block;
53 generation *generations = NULL; /* all the generations */
54 generation *g0 = NULL; /* generation 0, for convenience */
55 generation *oldest_gen = NULL; /* oldest generation, for convenience */
56 step *g0s0 = NULL; /* generation 0, step 0, for convenience */
59 step *all_steps = NULL; /* single array of steps */
61 ullong total_allocated = 0; /* total memory allocated during run */
63 nat n_nurseries = 0; /* == RtsFlags.ParFlags.nNodes, convenience */
64 step *nurseries = NULL; /* array of nurseries, >1 only if THREADED_RTS */
68 * Storage manager mutex: protects all the above state from
69 * simultaneous access by two STG threads.
73 * This mutex is used by atomicModifyMutVar# only
75 Mutex atomic_modify_mutvar_mutex;
82 static void *stgAllocForGMP (size_t size_in_bytes);
83 static void *stgReallocForGMP (void *ptr, size_t old_size, size_t new_size);
84 static void stgDeallocForGMP (void *ptr, size_t size);
87 initStep (step *stp, int g, int s)
90 stp->abs_no = RtsFlags.GcFlags.steps * g + s;
94 stp->live_estimate = 0;
95 stp->old_blocks = NULL;
96 stp->n_old_blocks = 0;
97 stp->gen = &generations[g];
99 stp->large_objects = NULL;
100 stp->n_large_blocks = 0;
101 stp->scavenged_large_objects = NULL;
102 stp->n_scavenged_large_blocks = 0;
107 initSpinLock(&stp->sync_todo);
108 initSpinLock(&stp->sync_large_objects);
110 stp->threads = END_TSO_QUEUE;
111 stp->old_threads = END_TSO_QUEUE;
120 if (generations != NULL) {
121 // multi-init protection
127 /* Sanity check to make sure the LOOKS_LIKE_ macros appear to be
128 * doing something reasonable.
130 /* We use the NOT_NULL variant or gcc warns that the test is always true */
131 ASSERT(LOOKS_LIKE_INFO_PTR_NOT_NULL((StgWord)&stg_BLACKHOLE_info));
132 ASSERT(LOOKS_LIKE_CLOSURE_PTR(&stg_dummy_ret_closure));
133 ASSERT(!HEAP_ALLOCED(&stg_dummy_ret_closure));
135 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
136 RtsFlags.GcFlags.heapSizeSuggestion >
137 RtsFlags.GcFlags.maxHeapSize) {
138 RtsFlags.GcFlags.maxHeapSize = RtsFlags.GcFlags.heapSizeSuggestion;
141 if (RtsFlags.GcFlags.maxHeapSize != 0 &&
142 RtsFlags.GcFlags.minAllocAreaSize >
143 RtsFlags.GcFlags.maxHeapSize) {
144 errorBelch("maximum heap size (-M) is smaller than minimum alloc area size (-A)");
145 RtsFlags.GcFlags.minAllocAreaSize = RtsFlags.GcFlags.maxHeapSize;
148 initBlockAllocator();
150 #if defined(THREADED_RTS)
151 initMutex(&sm_mutex);
152 initMutex(&atomic_modify_mutvar_mutex);
157 /* allocate generation info array */
158 generations = (generation *)stgMallocBytes(RtsFlags.GcFlags.generations
159 * sizeof(struct generation_),
160 "initStorage: gens");
162 /* allocate all the steps into an array. It is important that we do
163 it this way, because we need the invariant that two step pointers
164 can be directly compared to see which is the oldest.
165 Remember that the last generation has only one step. */
166 total_steps = 1 + (RtsFlags.GcFlags.generations - 1) * RtsFlags.GcFlags.steps;
167 all_steps = stgMallocBytes(total_steps * sizeof(struct step_),
168 "initStorage: steps");
170 /* Initialise all generations */
171 for(g = 0; g < RtsFlags.GcFlags.generations; g++) {
172 gen = &generations[g];
174 gen->mut_list = allocBlock();
175 gen->collections = 0;
176 gen->par_collections = 0;
177 gen->failed_promotions = 0;
181 /* A couple of convenience pointers */
182 g0 = &generations[0];
183 oldest_gen = &generations[RtsFlags.GcFlags.generations-1];
185 /* Allocate step structures in each generation */
186 if (RtsFlags.GcFlags.generations > 1) {
187 /* Only for multiple-generations */
189 /* Oldest generation: one step */
190 oldest_gen->n_steps = 1;
191 oldest_gen->steps = all_steps + (RtsFlags.GcFlags.generations - 1)
192 * RtsFlags.GcFlags.steps;
194 /* set up all except the oldest generation with 2 steps */
195 for(g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
196 generations[g].n_steps = RtsFlags.GcFlags.steps;
197 generations[g].steps = all_steps + g * RtsFlags.GcFlags.steps;
201 /* single generation, i.e. a two-space collector */
203 g0->steps = all_steps;
207 n_nurseries = n_capabilities;
211 nurseries = stgMallocBytes (n_nurseries * sizeof(struct step_),
212 "initStorage: nurseries");
214 /* Initialise all steps */
215 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
216 for (s = 0; s < generations[g].n_steps; s++) {
217 initStep(&generations[g].steps[s], g, s);
221 for (s = 0; s < n_nurseries; s++) {
222 initStep(&nurseries[s], 0, s);
225 /* Set up the destination pointers in each younger gen. step */
226 for (g = 0; g < RtsFlags.GcFlags.generations-1; g++) {
227 for (s = 0; s < generations[g].n_steps-1; s++) {
228 generations[g].steps[s].to = &generations[g].steps[s+1];
230 generations[g].steps[s].to = &generations[g+1].steps[0];
232 oldest_gen->steps[0].to = &oldest_gen->steps[0];
234 for (s = 0; s < n_nurseries; s++) {
235 nurseries[s].to = generations[0].steps[0].to;
238 /* The oldest generation has one step. */
239 if (RtsFlags.GcFlags.compact || RtsFlags.GcFlags.sweep) {
240 if (RtsFlags.GcFlags.generations == 1) {
241 errorBelch("WARNING: compact/sweep is incompatible with -G1; disabled");
243 oldest_gen->steps[0].mark = 1;
244 if (RtsFlags.GcFlags.compact)
245 oldest_gen->steps[0].compact = 1;
249 generations[0].max_blocks = 0;
250 g0s0 = &generations[0].steps[0];
252 /* The allocation area. Policy: keep the allocation area
253 * small to begin with, even if we have a large suggested heap
254 * size. Reason: we're going to do a major collection first, and we
255 * don't want it to be a big one. This vague idea is borne out by
256 * rigorous experimental evidence.
260 weak_ptr_list = NULL;
262 revertible_caf_list = NULL;
264 /* initialise the allocate() interface */
266 alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
270 /* Tell GNU multi-precision pkg about our custom alloc functions */
271 mp_set_memory_functions(stgAllocForGMP, stgReallocForGMP, stgDeallocForGMP);
274 initSpinLock(&gc_alloc_block_sync);
282 IF_DEBUG(gc, statDescribeGens());
290 stat_exit(calcAllocated());
296 stgFree(g0s0); // frees all the steps
297 stgFree(generations);
299 #if defined(THREADED_RTS)
300 closeMutex(&sm_mutex);
301 closeMutex(&atomic_modify_mutvar_mutex);
306 /* -----------------------------------------------------------------------------
309 The entry code for every CAF does the following:
311 - builds a CAF_BLACKHOLE in the heap
312 - pushes an update frame pointing to the CAF_BLACKHOLE
313 - invokes UPD_CAF(), which:
314 - calls newCaf, below
315 - updates the CAF with a static indirection to the CAF_BLACKHOLE
317 Why do we build a BLACKHOLE in the heap rather than just updating
318 the thunk directly? It's so that we only need one kind of update
319 frame - otherwise we'd need a static version of the update frame too.
321 newCaf() does the following:
323 - it puts the CAF on the oldest generation's mut-once list.
324 This is so that we can treat the CAF as a root when collecting
327 For GHCI, we have additional requirements when dealing with CAFs:
329 - we must *retain* all dynamically-loaded CAFs ever entered,
330 just in case we need them again.
331 - we must be able to *revert* CAFs that have been evaluated, to
332 their pre-evaluated form.
334 To do this, we use an additional CAF list. When newCaf() is
335 called on a dynamically-loaded CAF, we add it to the CAF list
336 instead of the old-generation mutable list, and save away its
337 old info pointer (in caf->saved_info) for later reversion.
339 To revert all the CAFs, we traverse the CAF list and reset the
340 info pointer to caf->saved_info, then throw away the CAF list.
341 (see GC.c:revertCAFs()).
345 -------------------------------------------------------------------------- */
348 newCAF(StgClosure* caf)
355 // If we are in GHCi _and_ we are using dynamic libraries,
356 // then we can't redirect newCAF calls to newDynCAF (see below),
357 // so we make newCAF behave almost like newDynCAF.
358 // The dynamic libraries might be used by both the interpreted
359 // program and GHCi itself, so they must not be reverted.
360 // This also means that in GHCi with dynamic libraries, CAFs are not
361 // garbage collected. If this turns out to be a problem, we could
362 // do another hack here and do an address range test on caf to figure
363 // out whether it is from a dynamic library.
364 ((StgIndStatic *)caf)->saved_info = (StgInfoTable *)caf->header.info;
365 ((StgIndStatic *)caf)->static_link = caf_list;
370 /* Put this CAF on the mutable list for the old generation.
371 * This is a HACK - the IND_STATIC closure doesn't really have
372 * a mut_link field, but we pretend it has - in fact we re-use
373 * the STATIC_LINK field for the time being, because when we
374 * come to do a major GC we won't need the mut_link field
375 * any more and can use it as a STATIC_LINK.
377 ((StgIndStatic *)caf)->saved_info = NULL;
378 recordMutableGen(caf, oldest_gen->no);
384 // An alternate version of newCaf which is used for dynamically loaded
385 // object code in GHCi. In this case we want to retain *all* CAFs in
386 // the object code, because they might be demanded at any time from an
387 // expression evaluated on the command line.
388 // Also, GHCi might want to revert CAFs, so we add these to the
389 // revertible_caf_list.
391 // The linker hackily arranges that references to newCaf from dynamic
392 // code end up pointing to newDynCAF.
394 newDynCAF(StgClosure *caf)
398 ((StgIndStatic *)caf)->saved_info = (StgInfoTable *)caf->header.info;
399 ((StgIndStatic *)caf)->static_link = revertible_caf_list;
400 revertible_caf_list = caf;
405 /* -----------------------------------------------------------------------------
407 -------------------------------------------------------------------------- */
410 allocNursery (step *stp, bdescr *tail, nat blocks)
415 // Allocate a nursery: we allocate fresh blocks one at a time and
416 // cons them on to the front of the list, not forgetting to update
417 // the back pointer on the tail of the list to point to the new block.
418 for (i=0; i < blocks; i++) {
421 processNursery() in LdvProfile.c assumes that every block group in
422 the nursery contains only a single block. So, if a block group is
423 given multiple blocks, change processNursery() accordingly.
427 // double-link the nursery: we might need to insert blocks
434 bd->free = bd->start;
442 assignNurseriesToCapabilities (void)
447 for (i = 0; i < n_nurseries; i++) {
448 capabilities[i].r.rNursery = &nurseries[i];
449 capabilities[i].r.rCurrentNursery = nurseries[i].blocks;
450 capabilities[i].r.rCurrentAlloc = NULL;
452 #else /* THREADED_RTS */
453 MainCapability.r.rNursery = &nurseries[0];
454 MainCapability.r.rCurrentNursery = nurseries[0].blocks;
455 MainCapability.r.rCurrentAlloc = NULL;
460 allocNurseries( void )
464 for (i = 0; i < n_nurseries; i++) {
465 nurseries[i].blocks =
466 allocNursery(&nurseries[i], NULL,
467 RtsFlags.GcFlags.minAllocAreaSize);
468 nurseries[i].n_blocks = RtsFlags.GcFlags.minAllocAreaSize;
469 nurseries[i].old_blocks = NULL;
470 nurseries[i].n_old_blocks = 0;
472 assignNurseriesToCapabilities();
476 resetNurseries( void )
482 for (i = 0; i < n_nurseries; i++) {
484 for (bd = stp->blocks; bd; bd = bd->link) {
485 bd->free = bd->start;
486 ASSERT(bd->gen_no == 0);
487 ASSERT(bd->step == stp);
488 IF_DEBUG(sanity,memset(bd->start, 0xaa, BLOCK_SIZE));
491 assignNurseriesToCapabilities();
495 countNurseryBlocks (void)
500 for (i = 0; i < n_nurseries; i++) {
501 blocks += nurseries[i].n_blocks;
507 resizeNursery ( step *stp, nat blocks )
512 nursery_blocks = stp->n_blocks;
513 if (nursery_blocks == blocks) return;
515 if (nursery_blocks < blocks) {
516 debugTrace(DEBUG_gc, "increasing size of nursery to %d blocks",
518 stp->blocks = allocNursery(stp, stp->blocks, blocks-nursery_blocks);
523 debugTrace(DEBUG_gc, "decreasing size of nursery to %d blocks",
527 while (nursery_blocks > blocks) {
529 next_bd->u.back = NULL;
530 nursery_blocks -= bd->blocks; // might be a large block
535 // might have gone just under, by freeing a large block, so make
536 // up the difference.
537 if (nursery_blocks < blocks) {
538 stp->blocks = allocNursery(stp, stp->blocks, blocks-nursery_blocks);
542 stp->n_blocks = blocks;
543 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
547 // Resize each of the nurseries to the specified size.
550 resizeNurseriesFixed (nat blocks)
553 for (i = 0; i < n_nurseries; i++) {
554 resizeNursery(&nurseries[i], blocks);
559 // Resize the nurseries to the total specified size.
562 resizeNurseries (nat blocks)
564 // If there are multiple nurseries, then we just divide the number
565 // of available blocks between them.
566 resizeNurseriesFixed(blocks / n_nurseries);
570 /* -----------------------------------------------------------------------------
571 move_TSO is called to update the TSO structure after it has been
572 moved from one place to another.
573 -------------------------------------------------------------------------- */
576 move_TSO (StgTSO *src, StgTSO *dest)
580 // relocate the stack pointer...
581 diff = (StgPtr)dest - (StgPtr)src; // In *words*
582 dest->sp = (StgPtr)dest->sp + diff;
585 /* -----------------------------------------------------------------------------
586 The allocate() interface
588 allocateInGen() function allocates memory directly into a specific
589 generation. It always succeeds, and returns a chunk of memory n
590 words long. n can be larger than the size of a block if necessary,
591 in which case a contiguous block group will be allocated.
593 allocate(n) is equivalent to allocateInGen(g0).
594 -------------------------------------------------------------------------- */
597 allocateInGen (generation *g, lnat n)
605 TICK_ALLOC_HEAP_NOCTR(n);
610 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_))
612 lnat req_blocks = (lnat)BLOCK_ROUND_UP(n*sizeof(W_)) / BLOCK_SIZE;
614 // Attempting to allocate an object larger than maxHeapSize
615 // should definitely be disallowed. (bug #1791)
616 if (RtsFlags.GcFlags.maxHeapSize > 0 &&
617 req_blocks >= RtsFlags.GcFlags.maxHeapSize) {
619 // heapOverflow() doesn't exit (see #2592), but we aren't
620 // in a position to do a clean shutdown here: we
621 // either have to allocate the memory or exit now.
622 // Allocating the memory would be bad, because the user
623 // has requested that we not exceed maxHeapSize, so we
625 stg_exit(EXIT_HEAPOVERFLOW);
628 bd = allocGroup(req_blocks);
629 dbl_link_onto(bd, &stp->large_objects);
630 stp->n_large_blocks += bd->blocks; // might be larger than req_blocks
631 alloc_blocks += bd->blocks;
634 bd->flags = BF_LARGE;
635 bd->free = bd->start + n;
640 // small allocation (<LARGE_OBJECT_THRESHOLD) */
642 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
647 bd->link = stp->blocks;
664 return allocateInGen(g0,n);
668 allocatedBytes( void )
672 allocated = alloc_blocks * BLOCK_SIZE_W;
673 if (pinned_object_block != NULL) {
674 allocated -= (pinned_object_block->start + BLOCK_SIZE_W) -
675 pinned_object_block->free;
681 // split N blocks off the front of the given bdescr, returning the
682 // new block group. We treat the remainder as if it
683 // had been freshly allocated in generation 0.
685 splitLargeBlock (bdescr *bd, nat blocks)
689 // subtract the original number of blocks from the counter first
690 bd->step->n_large_blocks -= bd->blocks;
692 new_bd = splitBlockGroup (bd, blocks);
694 dbl_link_onto(new_bd, &g0s0->large_objects);
695 g0s0->n_large_blocks += new_bd->blocks;
696 new_bd->gen_no = g0s0->no;
698 new_bd->flags = BF_LARGE;
699 new_bd->free = bd->free;
700 ASSERT(new_bd->free <= new_bd->start + new_bd->blocks * BLOCK_SIZE_W);
702 // add the new number of blocks to the counter. Due to the gaps
703 // for block descriptor, new_bd->blocks + bd->blocks might not be
704 // equal to the original bd->blocks, which is why we do it this way.
705 bd->step->n_large_blocks += bd->blocks;
710 /* -----------------------------------------------------------------------------
713 This allocates memory in the current thread - it is intended for
714 use primarily from STG-land where we have a Capability. It is
715 better than allocate() because it doesn't require taking the
716 sm_mutex lock in the common case.
718 Memory is allocated directly from the nursery if possible (but not
719 from the current nursery block, so as not to interfere with
721 -------------------------------------------------------------------------- */
724 allocateLocal (Capability *cap, lnat n)
729 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
730 return allocateInGen(g0,n);
733 /* small allocation (<LARGE_OBJECT_THRESHOLD) */
735 TICK_ALLOC_HEAP_NOCTR(n);
738 bd = cap->r.rCurrentAlloc;
739 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
741 // The CurrentAlloc block is full, we need to find another
742 // one. First, we try taking the next block from the
744 bd = cap->r.rCurrentNursery->link;
746 if (bd == NULL || bd->free + n > bd->start + BLOCK_SIZE_W) {
747 // The nursery is empty, or the next block is already
748 // full: allocate a fresh block (we can't fail here).
751 cap->r.rNursery->n_blocks++;
754 bd->step = cap->r.rNursery;
756 // NO: alloc_blocks++;
757 // calcAllocated() uses the size of the nursery, and we've
758 // already bumpted nursery->n_blocks above. We'll GC
759 // pretty quickly now anyway, because MAYBE_GC() will
760 // notice that CurrentNursery->link is NULL.
762 // we have a block in the nursery: take it and put
763 // it at the *front* of the nursery list, and use it
764 // to allocate() from.
765 cap->r.rCurrentNursery->link = bd->link;
766 if (bd->link != NULL) {
767 bd->link->u.back = cap->r.rCurrentNursery;
770 dbl_link_onto(bd, &cap->r.rNursery->blocks);
771 cap->r.rCurrentAlloc = bd;
772 IF_DEBUG(sanity, checkNurserySanity(cap->r.rNursery));
779 /* ---------------------------------------------------------------------------
780 Allocate a fixed/pinned object.
782 We allocate small pinned objects into a single block, allocating a
783 new block when the current one overflows. The block is chained
784 onto the large_object_list of generation 0 step 0.
786 NOTE: The GC can't in general handle pinned objects. This
787 interface is only safe to use for ByteArrays, which have no
788 pointers and don't require scavenging. It works because the
789 block's descriptor has the BF_LARGE flag set, so the block is
790 treated as a large object and chained onto various lists, rather
791 than the individual objects being copied. However, when it comes
792 to scavenge the block, the GC will only scavenge the first object.
793 The reason is that the GC can't linearly scan a block of pinned
794 objects at the moment (doing so would require using the
795 mostly-copying techniques). But since we're restricting ourselves
796 to pinned ByteArrays, not scavenging is ok.
798 This function is called by newPinnedByteArray# which immediately
799 fills the allocated memory with a MutableByteArray#.
800 ------------------------------------------------------------------------- */
803 allocatePinned( lnat n )
806 bdescr *bd = pinned_object_block;
808 // If the request is for a large object, then allocate()
809 // will give us a pinned object anyway.
810 if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
816 TICK_ALLOC_HEAP_NOCTR(n);
819 // we always return 8-byte aligned memory. bd->free must be
820 // 8-byte aligned to begin with, so we just round up n to
821 // the nearest multiple of 8 bytes.
822 if (sizeof(StgWord) == 4) {
826 // If we don't have a block of pinned objects yet, or the current
827 // one isn't large enough to hold the new object, allocate a new one.
828 if (bd == NULL || (bd->free + n) > (bd->start + BLOCK_SIZE_W)) {
829 pinned_object_block = bd = allocBlock();
830 dbl_link_onto(bd, &g0s0->large_objects);
831 g0s0->n_large_blocks++;
834 bd->flags = BF_PINNED | BF_LARGE;
835 bd->free = bd->start;
845 /* -----------------------------------------------------------------------------
847 -------------------------------------------------------------------------- */
850 This is the write barrier for MUT_VARs, a.k.a. IORefs. A
851 MUT_VAR_CLEAN object is not on the mutable list; a MUT_VAR_DIRTY
852 is. When written to, a MUT_VAR_CLEAN turns into a MUT_VAR_DIRTY
853 and is put on the mutable list.
856 dirty_MUT_VAR(StgRegTable *reg, StgClosure *p)
858 Capability *cap = regTableToCapability(reg);
860 if (p->header.info == &stg_MUT_VAR_CLEAN_info) {
861 p->header.info = &stg_MUT_VAR_DIRTY_info;
862 bd = Bdescr((StgPtr)p);
863 if (bd->gen_no > 0) recordMutableCap(p,cap,bd->gen_no);
867 // Setting a TSO's link field with a write barrier.
868 // It is *not* necessary to call this function when
869 // * setting the link field to END_TSO_QUEUE
870 // * putting a TSO on the blackhole_queue
871 // * setting the link field of the currently running TSO, as it
872 // will already be dirty.
874 setTSOLink (Capability *cap, StgTSO *tso, StgTSO *target)
877 if ((tso->flags & (TSO_DIRTY|TSO_LINK_DIRTY)) == 0) {
878 tso->flags |= TSO_LINK_DIRTY;
879 bd = Bdescr((StgPtr)tso);
880 if (bd->gen_no > 0) recordMutableCap((StgClosure*)tso,cap,bd->gen_no);
886 dirty_TSO (Capability *cap, StgTSO *tso)
889 if ((tso->flags & (TSO_DIRTY|TSO_LINK_DIRTY)) == 0) {
890 bd = Bdescr((StgPtr)tso);
891 if (bd->gen_no > 0) recordMutableCap((StgClosure*)tso,cap,bd->gen_no);
893 tso->flags |= TSO_DIRTY;
897 This is the write barrier for MVARs. An MVAR_CLEAN objects is not
898 on the mutable list; a MVAR_DIRTY is. When written to, a
899 MVAR_CLEAN turns into a MVAR_DIRTY and is put on the mutable list.
900 The check for MVAR_CLEAN is inlined at the call site for speed,
901 this really does make a difference on concurrency-heavy benchmarks
902 such as Chaneneos and cheap-concurrency.
905 dirty_MVAR(StgRegTable *reg, StgClosure *p)
907 Capability *cap = regTableToCapability(reg);
909 bd = Bdescr((StgPtr)p);
910 if (bd->gen_no > 0) recordMutableCap(p,cap,bd->gen_no);
913 /* -----------------------------------------------------------------------------
914 Allocation functions for GMP.
916 These all use the allocate() interface - we can't have any garbage
917 collection going on during a gmp operation, so we use allocate()
918 which always succeeds. The gmp operations which might need to
919 allocate will ask the storage manager (via doYouWantToGC()) whether
920 a garbage collection is required, in case we get into a loop doing
921 only allocate() style allocation.
922 -------------------------------------------------------------------------- */
925 stgAllocForGMP (size_t size_in_bytes)
928 nat data_size_in_words, total_size_in_words;
930 /* round up to a whole number of words */
931 data_size_in_words = (size_in_bytes + sizeof(W_) + 1) / sizeof(W_);
932 total_size_in_words = sizeofW(StgArrWords) + data_size_in_words;
934 /* allocate and fill it in. */
935 #if defined(THREADED_RTS)
936 arr = (StgArrWords *)allocateLocal(myTask()->cap, total_size_in_words);
938 arr = (StgArrWords *)allocateLocal(&MainCapability, total_size_in_words);
940 SET_ARR_HDR(arr, &stg_ARR_WORDS_info, CCCS, data_size_in_words);
942 /* and return a ptr to the goods inside the array */
947 stgReallocForGMP (void *ptr, size_t old_size, size_t new_size)
950 void *new_stuff_ptr = stgAllocForGMP(new_size);
952 char *p = (char *) ptr;
953 char *q = (char *) new_stuff_ptr;
955 min_size = old_size < new_size ? old_size : new_size;
956 for (; i < min_size; i++, p++, q++) {
960 return(new_stuff_ptr);
964 stgDeallocForGMP (void *ptr STG_UNUSED,
965 size_t size STG_UNUSED)
967 /* easy for us: the garbage collector does the dealloc'n */
970 /* -----------------------------------------------------------------------------
972 * -------------------------------------------------------------------------- */
974 /* -----------------------------------------------------------------------------
977 * Approximate how much we've allocated: number of blocks in the
978 * nursery + blocks allocated via allocate() - unused nusery blocks.
979 * This leaves a little slop at the end of each block, and doesn't
980 * take into account large objects (ToDo).
981 * -------------------------------------------------------------------------- */
984 calcAllocated( void )
989 allocated = allocatedBytes();
990 allocated += countNurseryBlocks() * BLOCK_SIZE_W;
995 for (i = 0; i < n_nurseries; i++) {
997 for ( bd = capabilities[i].r.rCurrentNursery->link;
998 bd != NULL; bd = bd->link ) {
999 allocated -= BLOCK_SIZE_W;
1001 cap = &capabilities[i];
1002 if (cap->r.rCurrentNursery->free <
1003 cap->r.rCurrentNursery->start + BLOCK_SIZE_W) {
1004 allocated -= (cap->r.rCurrentNursery->start + BLOCK_SIZE_W)
1005 - cap->r.rCurrentNursery->free;
1009 bdescr *current_nursery = MainCapability.r.rCurrentNursery;
1011 for ( bd = current_nursery->link; bd != NULL; bd = bd->link ) {
1012 allocated -= BLOCK_SIZE_W;
1014 if (current_nursery->free < current_nursery->start + BLOCK_SIZE_W) {
1015 allocated -= (current_nursery->start + BLOCK_SIZE_W)
1016 - current_nursery->free;
1021 total_allocated += allocated;
1025 /* Approximate the amount of live data in the heap. To be called just
1026 * after garbage collection (see GarbageCollect()).
1029 calcLiveBlocks(void)
1035 if (RtsFlags.GcFlags.generations == 1) {
1036 return g0s0->n_large_blocks + g0s0->n_blocks;
1039 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1040 for (s = 0; s < generations[g].n_steps; s++) {
1041 /* approximate amount of live data (doesn't take into account slop
1042 * at end of each block).
1044 if (g == 0 && s == 0) {
1047 stp = &generations[g].steps[s];
1048 live += stp->n_large_blocks + stp->n_blocks;
1055 countOccupied(bdescr *bd)
1060 for (; bd != NULL; bd = bd->link) {
1061 ASSERT(bd->free <= bd->start + bd->blocks * BLOCK_SIZE_W);
1062 words += bd->free - bd->start;
1067 // Return an accurate count of the live data in the heap, excluding
1076 if (RtsFlags.GcFlags.generations == 1) {
1077 return g0s0->n_words + countOccupied(g0s0->large_objects);
1081 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1082 for (s = 0; s < generations[g].n_steps; s++) {
1083 if (g == 0 && s == 0) continue;
1084 stp = &generations[g].steps[s];
1085 live += stp->n_words + countOccupied(stp->large_objects);
1091 /* Approximate the number of blocks that will be needed at the next
1092 * garbage collection.
1094 * Assume: all data currently live will remain live. Steps that will
1095 * be collected next time will therefore need twice as many blocks
1096 * since all the data will be copied.
1105 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1106 for (s = 0; s < generations[g].n_steps; s++) {
1107 if (g == 0 && s == 0) { continue; }
1108 stp = &generations[g].steps[s];
1110 // we need at least this much space
1111 needed += stp->n_blocks + stp->n_large_blocks;
1113 // any additional space needed to collect this gen next time?
1114 if (g == 0 || // always collect gen 0
1115 (generations[g].steps[0].n_blocks +
1116 generations[g].steps[0].n_large_blocks
1117 > generations[g].max_blocks)) {
1118 // we will collect this gen next time
1121 needed += stp->n_blocks / BITS_IN(W_);
1123 needed += stp->n_blocks / 100;
1126 continue; // no additional space needed for compaction
1128 needed += stp->n_blocks;
1136 /* ----------------------------------------------------------------------------
1139 Executable memory must be managed separately from non-executable
1140 memory. Most OSs these days require you to jump through hoops to
1141 dynamically allocate executable memory, due to various security
1144 Here we provide a small memory allocator for executable memory.
1145 Memory is managed with a page granularity; we allocate linearly
1146 in the page, and when the page is emptied (all objects on the page
1147 are free) we free the page again, not forgetting to make it
1150 TODO: The inability to handle objects bigger than BLOCK_SIZE_W means that
1151 the linker cannot use allocateExec for loading object code files
1152 on Windows. Once allocateExec can handle larger objects, the linker
1153 should be modified to use allocateExec instead of VirtualAlloc.
1154 ------------------------------------------------------------------------- */
1156 #if defined(linux_HOST_OS)
1158 // On Linux we need to use libffi for allocating executable memory,
1159 // because it knows how to work around the restrictions put in place
1162 void *allocateExec (nat bytes, void **exec_ret)
1166 ret = ffi_closure_alloc (sizeof(void *) + (size_t)bytes, (void**)&exec);
1168 if (ret == NULL) return ret;
1169 *ret = ret; // save the address of the writable mapping, for freeExec().
1170 *exec_ret = exec + 1;
1174 // freeExec gets passed the executable address, not the writable address.
1175 void freeExec (void *addr)
1178 writable = *((void**)addr - 1);
1180 ffi_closure_free (writable);
1186 void *allocateExec (nat bytes, void **exec_ret)
1193 // round up to words.
1194 n = (bytes + sizeof(W_) + 1) / sizeof(W_);
1196 if (n+1 > BLOCK_SIZE_W) {
1197 barf("allocateExec: can't handle large objects");
1200 if (exec_block == NULL ||
1201 exec_block->free + n + 1 > exec_block->start + BLOCK_SIZE_W) {
1203 lnat pagesize = getPageSize();
1204 bd = allocGroup(stg_max(1, pagesize / BLOCK_SIZE));
1205 debugTrace(DEBUG_gc, "allocate exec block %p", bd->start);
1207 bd->flags = BF_EXEC;
1208 bd->link = exec_block;
1209 if (exec_block != NULL) {
1210 exec_block->u.back = bd;
1213 setExecutable(bd->start, bd->blocks * BLOCK_SIZE, rtsTrue);
1216 *(exec_block->free) = n; // store the size of this chunk
1217 exec_block->gen_no += n; // gen_no stores the number of words allocated
1218 ret = exec_block->free + 1;
1219 exec_block->free += n + 1;
1226 void freeExec (void *addr)
1228 StgPtr p = (StgPtr)addr - 1;
1229 bdescr *bd = Bdescr((StgPtr)p);
1231 if ((bd->flags & BF_EXEC) == 0) {
1232 barf("freeExec: not executable");
1235 if (*(StgPtr)p == 0) {
1236 barf("freeExec: already free?");
1241 bd->gen_no -= *(StgPtr)p;
1244 if (bd->gen_no == 0) {
1245 // Free the block if it is empty, but not if it is the block at
1246 // the head of the queue.
1247 if (bd != exec_block) {
1248 debugTrace(DEBUG_gc, "free exec block %p", bd->start);
1249 dbl_link_remove(bd, &exec_block);
1250 setExecutable(bd->start, bd->blocks * BLOCK_SIZE, rtsFalse);
1253 bd->free = bd->start;
1260 #endif /* mingw32_HOST_OS */
1262 /* -----------------------------------------------------------------------------
1265 memInventory() checks for memory leaks by counting up all the
1266 blocks we know about and comparing that to the number of blocks
1267 allegedly floating around in the system.
1268 -------------------------------------------------------------------------- */
1272 // Useful for finding partially full blocks in gdb
1273 void findSlop(bdescr *bd);
1274 void findSlop(bdescr *bd)
1278 for (; bd != NULL; bd = bd->link) {
1279 slop = (bd->blocks * BLOCK_SIZE_W) - (bd->free - bd->start);
1280 if (slop > (1024/sizeof(W_))) {
1281 debugBelch("block at %p (bdescr %p) has %ldKB slop\n",
1282 bd->start, bd, slop / (1024/sizeof(W_)));
1288 countBlocks(bdescr *bd)
1291 for (n=0; bd != NULL; bd=bd->link) {
1297 // (*1) Just like countBlocks, except that we adjust the count for a
1298 // megablock group so that it doesn't include the extra few blocks
1299 // that would be taken up by block descriptors in the second and
1300 // subsequent megablock. This is so we can tally the count with the
1301 // number of blocks allocated in the system, for memInventory().
1303 countAllocdBlocks(bdescr *bd)
1306 for (n=0; bd != NULL; bd=bd->link) {
1308 // hack for megablock groups: see (*1) above
1309 if (bd->blocks > BLOCKS_PER_MBLOCK) {
1310 n -= (MBLOCK_SIZE / BLOCK_SIZE - BLOCKS_PER_MBLOCK)
1311 * (bd->blocks/(MBLOCK_SIZE/BLOCK_SIZE));
1318 stepBlocks (step *stp)
1320 ASSERT(countBlocks(stp->blocks) == stp->n_blocks);
1321 ASSERT(countBlocks(stp->large_objects) == stp->n_large_blocks);
1322 return stp->n_blocks + stp->n_old_blocks +
1323 countAllocdBlocks(stp->large_objects);
1326 // If memInventory() calculates that we have a memory leak, this
1327 // function will try to find the block(s) that are leaking by marking
1328 // all the ones that we know about, and search through memory to find
1329 // blocks that are not marked. In the debugger this can help to give
1330 // us a clue about what kind of block leaked. In the future we might
1331 // annotate blocks with their allocation site to give more helpful
1334 findMemoryLeak (void)
1337 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1338 for (i = 0; i < n_capabilities; i++) {
1339 markBlocks(capabilities[i].mut_lists[g]);
1341 markBlocks(generations[g].mut_list);
1342 for (s = 0; s < generations[g].n_steps; s++) {
1343 markBlocks(generations[g].steps[s].blocks);
1344 markBlocks(generations[g].steps[s].large_objects);
1348 for (i = 0; i < n_nurseries; i++) {
1349 markBlocks(nurseries[i].blocks);
1350 markBlocks(nurseries[i].large_objects);
1355 // if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_RETAINER) {
1356 // markRetainerBlocks();
1360 // count the blocks allocated by the arena allocator
1362 // markArenaBlocks();
1364 // count the blocks containing executable memory
1365 markBlocks(exec_block);
1367 reportUnmarkedBlocks();
1372 memInventory (rtsBool show)
1376 lnat gen_blocks[RtsFlags.GcFlags.generations];
1377 lnat nursery_blocks, retainer_blocks,
1378 arena_blocks, exec_blocks;
1379 lnat live_blocks = 0, free_blocks = 0;
1382 // count the blocks we current have
1384 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1386 for (i = 0; i < n_capabilities; i++) {
1387 gen_blocks[g] += countBlocks(capabilities[i].mut_lists[g]);
1389 gen_blocks[g] += countAllocdBlocks(generations[g].mut_list);
1390 for (s = 0; s < generations[g].n_steps; s++) {
1391 stp = &generations[g].steps[s];
1392 gen_blocks[g] += stepBlocks(stp);
1397 for (i = 0; i < n_nurseries; i++) {
1398 nursery_blocks += stepBlocks(&nurseries[i]);
1401 retainer_blocks = 0;
1403 if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_RETAINER) {
1404 retainer_blocks = retainerStackBlocks();
1408 // count the blocks allocated by the arena allocator
1409 arena_blocks = arenaBlocks();
1411 // count the blocks containing executable memory
1412 exec_blocks = countAllocdBlocks(exec_block);
1414 /* count the blocks on the free list */
1415 free_blocks = countFreeList();
1418 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1419 live_blocks += gen_blocks[g];
1421 live_blocks += nursery_blocks +
1422 + retainer_blocks + arena_blocks + exec_blocks;
1424 #define MB(n) (((n) * BLOCK_SIZE_W) / ((1024*1024)/sizeof(W_)))
1426 leak = live_blocks + free_blocks != mblocks_allocated * BLOCKS_PER_MBLOCK;
1431 debugBelch("Memory leak detected:\n");
1433 debugBelch("Memory inventory:\n");
1435 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1436 debugBelch(" gen %d blocks : %5lu blocks (%lu MB)\n", g,
1437 gen_blocks[g], MB(gen_blocks[g]));
1439 debugBelch(" nursery : %5lu blocks (%lu MB)\n",
1440 nursery_blocks, MB(nursery_blocks));
1441 debugBelch(" retainer : %5lu blocks (%lu MB)\n",
1442 retainer_blocks, MB(retainer_blocks));
1443 debugBelch(" arena blocks : %5lu blocks (%lu MB)\n",
1444 arena_blocks, MB(arena_blocks));
1445 debugBelch(" exec : %5lu blocks (%lu MB)\n",
1446 exec_blocks, MB(exec_blocks));
1447 debugBelch(" free : %5lu blocks (%lu MB)\n",
1448 free_blocks, MB(free_blocks));
1449 debugBelch(" total : %5lu blocks (%lu MB)\n",
1450 live_blocks + free_blocks, MB(live_blocks+free_blocks));
1452 debugBelch("\n in system : %5lu blocks (%lu MB)\n",
1453 mblocks_allocated * BLOCKS_PER_MBLOCK, mblocks_allocated);
1461 ASSERT(n_alloc_blocks == live_blocks);
1466 /* Full heap sanity check. */
1472 if (RtsFlags.GcFlags.generations == 1) {
1473 checkHeap(g0s0->blocks);
1474 checkChain(g0s0->large_objects);
1477 for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
1478 for (s = 0; s < generations[g].n_steps; s++) {
1479 if (g == 0 && s == 0) { continue; }
1480 ASSERT(countBlocks(generations[g].steps[s].blocks)
1481 == generations[g].steps[s].n_blocks);
1482 ASSERT(countBlocks(generations[g].steps[s].large_objects)
1483 == generations[g].steps[s].n_large_blocks);
1484 checkHeap(generations[g].steps[s].blocks);
1485 checkChain(generations[g].steps[s].large_objects);
1489 for (s = 0; s < n_nurseries; s++) {
1490 ASSERT(countBlocks(nurseries[s].blocks)
1491 == nurseries[s].n_blocks);
1492 ASSERT(countBlocks(nurseries[s].large_objects)
1493 == nurseries[s].n_large_blocks);
1496 checkFreeListSanity();
1499 #if defined(THREADED_RTS)
1500 // check the stacks too in threaded mode, because we don't do a
1501 // full heap sanity check in this case (see checkHeap())
1502 checkMutableLists(rtsTrue);
1504 checkMutableLists(rtsFalse);
1508 /* Nursery sanity check */
1510 checkNurserySanity( step *stp )
1516 for (bd = stp->blocks; bd != NULL; bd = bd->link) {
1517 ASSERT(bd->u.back == prev);
1519 blocks += bd->blocks;
1521 ASSERT(blocks == stp->n_blocks);
1524 // handy function for use in gdb, because Bdescr() is inlined.
1525 extern bdescr *_bdescr( StgPtr p );