X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=rts%2Fsm%2FStorage.c;h=f0506cd77cd8037a8774b8c723e2153f770f98f6;hb=ce9a12321f228ab68934e3031c32ab7f9a2173fc;hp=e10304c5a10127f1598db919f52fa165f9b70787;hpb=fdd7a41eec615cf3d77a95896a6183326e60c2ca;p=ghc-hetmet.git diff --git a/rts/sm/Storage.c b/rts/sm/Storage.c index e10304c..f0506cd 100644 --- a/rts/sm/Storage.c +++ b/rts/sm/Storage.c @@ -13,18 +13,15 @@ #include "PosixSource.h" #include "Rts.h" + +#include "Storage.h" #include "RtsUtils.h" -#include "RtsFlags.h" #include "Stats.h" -#include "Hooks.h" #include "BlockAlloc.h" -#include "MBlock.h" #include "Weak.h" #include "Sanity.h" #include "Arena.h" -#include "OSThreads.h" #include "Capability.h" -#include "Storage.h" #include "Schedule.h" #include "RetainerProfile.h" // for counting memory blocks (memInventory) #include "OSMem.h" @@ -32,9 +29,10 @@ #include "GC.h" #include "Evac.h" -#include #include +#include "ffi.h" + /* * All these globals require sm_mutex to access in THREADED_RTS mode. */ @@ -46,6 +44,8 @@ bdescr *pinned_object_block; /* allocate pinned objects into this block */ nat alloc_blocks; /* number of allocate()d blocks since GC */ nat alloc_blocks_lim; /* approximate limit on alloc_blocks */ +static bdescr *exec_block; + generation *generations = NULL; /* all the generations */ generation *g0 = NULL; /* generation 0, for convenience */ generation *oldest_gen = NULL; /* oldest generation, for convenience */ @@ -65,19 +65,9 @@ step *nurseries = NULL; /* array of nurseries, >1 only if THREADED_RTS * * simultaneous access by two STG threads. */ Mutex sm_mutex; -/* - * This mutex is used by atomicModifyMutVar# only - */ -Mutex atomic_modify_mutvar_mutex; #endif - -/* - * Forward references - */ -static void *stgAllocForGMP (size_t size_in_bytes); -static void *stgReallocForGMP (void *ptr, size_t old_size, size_t new_size); -static void stgDeallocForGMP (void *ptr, size_t size); +static void allocNurseries ( void ); static void initStep (step *stp, int g, int s) @@ -100,7 +90,6 @@ initStep (step *stp, int g, int s) stp->compact = 0; stp->bitmap = NULL; #ifdef THREADED_RTS - initSpinLock(&stp->sync_todo); initSpinLock(&stp->sync_large_objects); #endif stp->threads = END_TSO_QUEUE; @@ -145,7 +134,6 @@ initStorage( void ) #if defined(THREADED_RTS) initMutex(&sm_mutex); - initMutex(&atomic_modify_mutvar_mutex); #endif ACQUIRE_SM_LOCK; @@ -261,15 +249,17 @@ initStorage( void ) alloc_blocks = 0; alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize; - /* Tell GNU multi-precision pkg about our custom alloc functions */ - mp_set_memory_functions(stgAllocForGMP, stgReallocForGMP, stgDeallocForGMP); + exec_block = NULL; #ifdef THREADED_RTS initSpinLock(&gc_alloc_block_sync); - initSpinLock(&recordMutableGen_sync); whitehole_spin = 0; #endif + N = 0; + + initGcThreads(); + IF_DEBUG(gc, statDescribeGens()); RELEASE_SM_LOCK; @@ -289,9 +279,9 @@ freeStorage (void) freeAllMBlocks(); #if defined(THREADED_RTS) closeMutex(&sm_mutex); - closeMutex(&atomic_modify_mutvar_mutex); #endif stgFree(nurseries); + freeGcThreads(); } /* ----------------------------------------------------------------------------- @@ -340,6 +330,7 @@ newCAF(StgClosure* caf) { ACQUIRE_SM_LOCK; +#ifdef DYNAMIC if(keepCAFs) { // HACK: @@ -357,6 +348,7 @@ newCAF(StgClosure* caf) caf_list = caf; } else +#endif { /* Put this CAF on the mutable list for the old generation. * This is a HACK - the IND_STATIC closure doesn't really have @@ -366,7 +358,7 @@ newCAF(StgClosure* caf) * any more and can use it as a STATIC_LINK. */ ((StgIndStatic *)caf)->saved_info = NULL; - recordMutableGen(caf, oldest_gen); + recordMutableGen(caf, oldest_gen->no); } RELEASE_SM_LOCK; @@ -447,7 +439,7 @@ assignNurseriesToCapabilities (void) #endif } -void +static void allocNurseries( void ) { nat i; @@ -607,11 +599,19 @@ allocateInGen (generation *g, lnat n) if (RtsFlags.GcFlags.maxHeapSize > 0 && req_blocks >= RtsFlags.GcFlags.maxHeapSize) { heapOverflow(); + // heapOverflow() doesn't exit (see #2592), but we aren't + // in a position to do a clean shutdown here: we + // either have to allocate the memory or exit now. + // Allocating the memory would be bad, because the user + // has requested that we not exceed maxHeapSize, so we + // just exit. + stg_exit(EXIT_HEAPOVERFLOW); } bd = allocGroup(req_blocks); dbl_link_onto(bd, &stp->large_objects); stp->n_large_blocks += bd->blocks; // might be larger than req_blocks + alloc_blocks += bd->blocks; bd->gen_no = g->no; bd->step = stp; bd->flags = BF_LARGE; @@ -738,7 +738,9 @@ allocateLocal (Capability *cap, lnat n) bd->flags = 0; // NO: alloc_blocks++; // calcAllocated() uses the size of the nursery, and we've - // already bumpted nursery->n_blocks above. + // already bumpted nursery->n_blocks above. We'll GC + // pretty quickly now anyway, because MAYBE_GC() will + // notice that CurrentNursery->link is NULL. } else { // we have a block in the nursery: take it and put // it at the *front* of the nursery list, and use it @@ -789,7 +791,9 @@ allocatePinned( lnat n ) // If the request is for a large object, then allocate() // will give us a pinned object anyway. if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) { - return allocate(n); + p = allocate(n); + Bdescr(p)->flags |= BF_PINNED; + return p; } ACQUIRE_SM_LOCK; @@ -797,13 +801,6 @@ allocatePinned( lnat n ) TICK_ALLOC_HEAP_NOCTR(n); CCS_ALLOC(CCCS,n); - // we always return 8-byte aligned memory. bd->free must be - // 8-byte aligned to begin with, so we just round up n to - // the nearest multiple of 8 bytes. - if (sizeof(StgWord) == 4) { - n = (n+1) & ~1; - } - // If we don't have a block of pinned objects yet, or the current // one isn't large enough to hold the new object, allocate a new one. if (bd == NULL || (bd->free + n) > (bd->start + BLOCK_SIZE_W)) { @@ -855,7 +852,7 @@ void setTSOLink (Capability *cap, StgTSO *tso, StgTSO *target) { bdescr *bd; - if ((tso->flags & (TSO_DIRTY|TSO_LINK_DIRTY)) == 0) { + if (tso->dirty == 0 && (tso->flags & TSO_LINK_DIRTY) == 0) { tso->flags |= TSO_LINK_DIRTY; bd = Bdescr((StgPtr)tso); if (bd->gen_no > 0) recordMutableCap((StgClosure*)tso,cap,bd->gen_no); @@ -867,11 +864,11 @@ void dirty_TSO (Capability *cap, StgTSO *tso) { bdescr *bd; - if ((tso->flags & (TSO_DIRTY|TSO_LINK_DIRTY)) == 0) { + if (tso->dirty == 0 && (tso->flags & TSO_LINK_DIRTY) == 0) { bd = Bdescr((StgPtr)tso); if (bd->gen_no > 0) recordMutableCap((StgClosure*)tso,cap,bd->gen_no); } - tso->flags |= TSO_DIRTY; + tso->dirty = 1; } /* @@ -892,63 +889,6 @@ dirty_MVAR(StgRegTable *reg, StgClosure *p) } /* ----------------------------------------------------------------------------- - Allocation functions for GMP. - - These all use the allocate() interface - we can't have any garbage - collection going on during a gmp operation, so we use allocate() - which always succeeds. The gmp operations which might need to - allocate will ask the storage manager (via doYouWantToGC()) whether - a garbage collection is required, in case we get into a loop doing - only allocate() style allocation. - -------------------------------------------------------------------------- */ - -static void * -stgAllocForGMP (size_t size_in_bytes) -{ - StgArrWords* arr; - nat data_size_in_words, total_size_in_words; - - /* round up to a whole number of words */ - data_size_in_words = (size_in_bytes + sizeof(W_) + 1) / sizeof(W_); - total_size_in_words = sizeofW(StgArrWords) + data_size_in_words; - - /* allocate and fill it in. */ -#if defined(THREADED_RTS) - arr = (StgArrWords *)allocateLocal(myTask()->cap, total_size_in_words); -#else - arr = (StgArrWords *)allocateLocal(&MainCapability, total_size_in_words); -#endif - SET_ARR_HDR(arr, &stg_ARR_WORDS_info, CCCS, data_size_in_words); - - /* and return a ptr to the goods inside the array */ - return arr->payload; -} - -static void * -stgReallocForGMP (void *ptr, size_t old_size, size_t new_size) -{ - size_t min_size; - void *new_stuff_ptr = stgAllocForGMP(new_size); - nat i = 0; - char *p = (char *) ptr; - char *q = (char *) new_stuff_ptr; - - min_size = old_size < new_size ? old_size : new_size; - for (; i < min_size; i++, p++, q++) { - *q = *p; - } - - return(new_stuff_ptr); -} - -static void -stgDeallocForGMP (void *ptr STG_UNUSED, - size_t size STG_UNUSED) -{ - /* easy for us: the garbage collector does the dealloc'n */ -} - -/* ----------------------------------------------------------------------------- * Stats and stuff * -------------------------------------------------------------------------- */ @@ -1134,9 +1074,37 @@ calcNeeded(void) should be modified to use allocateExec instead of VirtualAlloc. ------------------------------------------------------------------------- */ -static bdescr *exec_block; +#if defined(linux_HOST_OS) + +// On Linux we need to use libffi for allocating executable memory, +// because it knows how to work around the restrictions put in place +// by SELinux. -void *allocateExec (nat bytes) +void *allocateExec (nat bytes, void **exec_ret) +{ + void **ret, **exec; + ACQUIRE_SM_LOCK; + ret = ffi_closure_alloc (sizeof(void *) + (size_t)bytes, (void**)&exec); + RELEASE_SM_LOCK; + if (ret == NULL) return ret; + *ret = ret; // save the address of the writable mapping, for freeExec(). + *exec_ret = exec + 1; + return (ret + 1); +} + +// freeExec gets passed the executable address, not the writable address. +void freeExec (void *addr) +{ + void *writable; + writable = *((void**)addr - 1); + ACQUIRE_SM_LOCK; + ffi_closure_free (writable); + RELEASE_SM_LOCK +} + +#else + +void *allocateExec (nat bytes, void **exec_ret) { void *ret; nat n; @@ -1172,6 +1140,7 @@ void *allocateExec (nat bytes) exec_block->free += n + 1; RELEASE_SM_LOCK + *exec_ret = ret; return ret; } @@ -1209,6 +1178,8 @@ void freeExec (void *addr) RELEASE_SM_LOCK } +#endif /* mingw32_HOST_OS */ + /* ----------------------------------------------------------------------------- Debugging @@ -1421,7 +1392,7 @@ checkSanity( void ) if (RtsFlags.GcFlags.generations == 1) { checkHeap(g0s0->blocks); - checkChain(g0s0->large_objects); + checkLargeObjects(g0s0->large_objects); } else { for (g = 0; g < RtsFlags.GcFlags.generations; g++) { @@ -1432,10 +1403,7 @@ checkSanity( void ) ASSERT(countBlocks(generations[g].steps[s].large_objects) == generations[g].steps[s].n_large_blocks); checkHeap(generations[g].steps[s].blocks); - checkChain(generations[g].steps[s].large_objects); - if (g > 0) { - checkMutableList(generations[g].mut_list, g); - } + checkLargeObjects(generations[g].steps[s].large_objects); } } @@ -1452,9 +1420,9 @@ checkSanity( void ) #if defined(THREADED_RTS) // check the stacks too in threaded mode, because we don't do a // full heap sanity check in this case (see checkHeap()) - checkGlobalTSOList(rtsTrue); + checkMutableLists(rtsTrue); #else - checkGlobalTSOList(rtsFalse); + checkMutableLists(rtsFalse); #endif }