// Copyright 2009 The Go Authors. All rights reserved. // Use of this source code is governed by a BSD-style // license that can be found in the LICENSE file. // See malloc.h for overview. // // TODO(rsc): double-check stats. package runtime #include #include #include #include "go-alloc.h" #include "runtime.h" #include "arch.h" #include "malloc.h" #include "interface.h" #include "go-type.h" #include "race.h" // Map gccgo field names to gc field names. // Eface aka __go_empty_interface. #define type __type_descriptor // Type aka __go_type_descriptor #define kind __code #define string __reflection #define KindPtr GO_PTR #define KindNoPointers GO_NO_POINTERS // Mark mheap as 'no pointers', it does not contain interesting pointers but occupies ~45K. MHeap runtime_mheap; int32 runtime_checking; extern MStats mstats; // defined in zruntime_def_$GOOS_$GOARCH.go extern volatile intgo runtime_MemProfileRate __asm__ (GOSYM_PREFIX "runtime.MemProfileRate"); // Allocate an object of at least size bytes. // Small objects are allocated from the per-thread cache's free lists. // Large objects (> 32 kB) are allocated straight from the heap. // If the block will be freed with runtime_free(), typ must be 0. void* runtime_mallocgc(uintptr size, uintptr typ, uint32 flag) { M *m; G *g; int32 sizeclass; intgo rate; MCache *c; MCacheList *l; uintptr npages; MSpan *s; MLink *v; bool incallback; if(size == 0) { // All 0-length allocations use this pointer. // The language does not require the allocations to // have distinct values. return &runtime_zerobase; } m = runtime_m(); g = runtime_g(); incallback = false; if(m->mcache == nil && g->ncgo > 0) { // For gccgo this case can occur when a cgo or SWIG function // has an interface return type and the function // returns a non-pointer, so memory allocation occurs // after syscall.Cgocall but before syscall.CgocallDone. // We treat it as a callback. runtime_exitsyscall(); m = runtime_m(); incallback = true; flag |= FlagNoInvokeGC; } if(runtime_gcwaiting() && g != m->g0 && m->locks == 0 && !(flag & FlagNoInvokeGC)) { runtime_gosched(); m = runtime_m(); } if(m->mallocing) runtime_throw("malloc/free - deadlock"); // Disable preemption during settype_flush. // We can not use m->mallocing for this, because settype_flush calls mallocgc. m->locks++; m->mallocing = 1; if(DebugTypeAtBlockEnd) size += sizeof(uintptr); c = m->mcache; if(size <= MaxSmallSize) { // Allocate from mcache free lists. // Inlined version of SizeToClass(). if(size <= 1024-8) sizeclass = runtime_size_to_class8[(size+7)>>3]; else sizeclass = runtime_size_to_class128[(size-1024+127) >> 7]; size = runtime_class_to_size[sizeclass]; l = &c->list[sizeclass]; if(l->list == nil) runtime_MCache_Refill(c, sizeclass); v = l->list; l->list = v->next; l->nlist--; if(!(flag & FlagNoZero)) { v->next = nil; // block is zeroed iff second word is zero ... if(size > sizeof(uintptr) && ((uintptr*)v)[1] != 0) runtime_memclr((byte*)v, size); } c->local_cachealloc += size; } else { // TODO(rsc): Report tracebacks for very large allocations. // Allocate directly from heap. npages = size >> PageShift; if((size & PageMask) != 0) npages++; s = runtime_MHeap_Alloc(&runtime_mheap, npages, 0, 1, !(flag & FlagNoZero)); if(s == nil) runtime_throw("out of memory"); s->limit = (byte*)(s->start<start << PageShift); // setup for mark sweep runtime_markspan(v, 0, 0, true); } if(!(flag & FlagNoGC)) runtime_markallocated(v, size, (flag&FlagNoScan) != 0); if(DebugTypeAtBlockEnd) *(uintptr*)((uintptr)v+size-sizeof(uintptr)) = typ; // TODO: save type even if FlagNoScan? Potentially expensive but might help // heap profiling/tracing. if(UseSpanType && !(flag & FlagNoScan) && typ != 0) { uintptr *buf, i; buf = m->settype_buf; i = m->settype_bufsize; buf[i++] = (uintptr)v; buf[i++] = typ; m->settype_bufsize = i; } m->mallocing = 0; if(UseSpanType && !(flag & FlagNoScan) && typ != 0 && m->settype_bufsize == nelem(m->settype_buf)) runtime_settype_flush(m); m->locks--; if(!(flag & FlagNoProfiling) && (rate = runtime_MemProfileRate) > 0) { if(size >= (uint32) rate) goto profile; if((uint32) m->mcache->next_sample > size) m->mcache->next_sample -= size; else { // pick next profile time // If you change this, also change allocmcache. if(rate > 0x3fffffff) // make 2*rate not overflow rate = 0x3fffffff; m->mcache->next_sample = runtime_fastrand1() % (2*rate); profile: runtime_setblockspecial(v, true); runtime_MProf_Malloc(v, size); } } if(!(flag & FlagNoInvokeGC) && mstats.heap_alloc >= mstats.next_gc) runtime_gc(0); if(raceenabled) runtime_racemalloc(v, size); if(incallback) runtime_entersyscall(); return v; } void* __go_alloc(uintptr size) { return runtime_mallocgc(size, 0, FlagNoInvokeGC); } // Free the object whose base pointer is v. void __go_free(void *v) { M *m; int32 sizeclass; MSpan *s; MCache *c; uint32 prof; uintptr size; if(v == nil) return; // If you change this also change mgc0.c:/^sweep, // which has a copy of the guts of free. m = runtime_m(); if(m->mallocing) runtime_throw("malloc/free - deadlock"); m->mallocing = 1; if(!runtime_mlookup(v, nil, nil, &s)) { runtime_printf("free %p: not an allocated block\n", v); runtime_throw("free runtime_mlookup"); } prof = runtime_blockspecial(v); if(raceenabled) runtime_racefree(v); // Find size class for v. sizeclass = s->sizeclass; c = m->mcache; if(sizeclass == 0) { // Large object. size = s->npages<start<local_nlargefree++; c->local_largefree += size; } else { // Small object. size = runtime_class_to_size[sizeclass]; if(size > sizeof(uintptr)) ((uintptr*)v)[1] = (uintptr)0xfeedfeedfeedfeedll; // mark as "needs to be zeroed" // Must mark v freed before calling MCache_Free: // it might coalesce v and other blocks into a bigger span // and change the bitmap further. runtime_markfreed(v, size); c->local_nsmallfree[sizeclass]++; runtime_MCache_Free(c, v, sizeclass, size); } if(prof) runtime_MProf_Free(v, size); m->mallocing = 0; } int32 runtime_mlookup(void *v, byte **base, uintptr *size, MSpan **sp) { M *m; uintptr n, i; byte *p; MSpan *s; m = runtime_m(); m->mcache->local_nlookup++; if (sizeof(void*) == 4 && m->mcache->local_nlookup >= (1<<30)) { // purge cache stats to prevent overflow runtime_lock(&runtime_mheap); runtime_purgecachedstats(m->mcache); runtime_unlock(&runtime_mheap); } s = runtime_MHeap_LookupMaybe(&runtime_mheap, v); if(sp) *sp = s; if(s == nil) { runtime_checkfreed(v, 1); if(base) *base = nil; if(size) *size = 0; return 0; } p = (byte*)((uintptr)s->start<sizeclass == 0) { // Large object. if(base) *base = p; if(size) *size = s->npages<elemsize; if(base) { i = ((byte*)v - p)/n; *base = p + i*n; } if(size) *size = n; return 1; } MCache* runtime_allocmcache(void) { intgo rate; MCache *c; runtime_lock(&runtime_mheap); c = runtime_FixAlloc_Alloc(&runtime_mheap.cachealloc); runtime_unlock(&runtime_mheap); runtime_memclr((byte*)c, sizeof(*c)); // Set first allocation sample size. rate = runtime_MemProfileRate; if(rate > 0x3fffffff) // make 2*rate not overflow rate = 0x3fffffff; if(rate != 0) c->next_sample = runtime_fastrand1() % (2*rate); return c; } void runtime_freemcache(MCache *c) { runtime_MCache_ReleaseAll(c); runtime_lock(&runtime_mheap); runtime_purgecachedstats(c); runtime_FixAlloc_Free(&runtime_mheap.cachealloc, c); runtime_unlock(&runtime_mheap); } void runtime_purgecachedstats(MCache *c) { MHeap *h; int32 i; // Protected by either heap or GC lock. h = &runtime_mheap; mstats.heap_alloc += c->local_cachealloc; c->local_cachealloc = 0; mstats.nlookup += c->local_nlookup; c->local_nlookup = 0; h->largefree += c->local_largefree; c->local_largefree = 0; h->nlargefree += c->local_nlargefree; c->local_nlargefree = 0; for(i=0; i<(int32)nelem(c->local_nsmallfree); i++) { h->nsmallfree[i] += c->local_nsmallfree[i]; c->local_nsmallfree[i] = 0; } } extern uintptr runtime_sizeof_C_MStats __asm__ (GOSYM_PREFIX "runtime.Sizeof_C_MStats"); #define MaxArena32 (2U<<30) void runtime_mallocinit(void) { byte *p; uintptr arena_size, bitmap_size, spans_size; extern byte _end[]; byte *want; uintptr limit; uint64 i; runtime_sizeof_C_MStats = sizeof(MStats); p = nil; arena_size = 0; bitmap_size = 0; spans_size = 0; // for 64-bit build USED(p); USED(arena_size); USED(bitmap_size); USED(spans_size); runtime_InitSizes(); // limit = runtime_memlimit(); // See https://code.google.com/p/go/issues/detail?id=5049 // TODO(rsc): Fix after 1.1. limit = 0; // Set up the allocation arena, a contiguous area of memory where // allocated data will be found. The arena begins with a bitmap large // enough to hold 4 bits per allocated word. if(sizeof(void*) == 8 && (limit == 0 || limit > (1<<30))) { // On a 64-bit machine, allocate from a single contiguous reservation. // 128 GB (MaxMem) should be big enough for now. // // The code will work with the reservation at any address, but ask // SysReserve to use 0x0000XXc000000000 if possible (XX=00...7f). // Allocating a 128 GB region takes away 37 bits, and the amd64 // doesn't let us choose the top 17 bits, so that leaves the 11 bits // in the middle of 0x00c0 for us to choose. Choosing 0x00c0 means // that the valid memory addresses will begin 0x00c0, 0x00c1, ..., 0x00df. // In little-endian, that's c0 00, c1 00, ..., df 00. None of those are valid // UTF-8 sequences, and they are otherwise as far away from // ff (likely a common byte) as possible. If that fails, we try other 0xXXc0 // addresses. An earlier attempt to use 0x11f8 caused out of memory errors // on OS X during thread allocations. 0x00c0 causes conflicts with // AddressSanitizer which reserves all memory up to 0x0100. // These choices are both for debuggability and to reduce the // odds of the conservative garbage collector not collecting memory // because some non-pointer block of memory had a bit pattern // that matched a memory address. // // Actually we reserve 136 GB (because the bitmap ends up being 8 GB) // but it hardly matters: e0 00 is not valid UTF-8 either. // // If this fails we fall back to the 32 bit memory mechanism arena_size = MaxMem; bitmap_size = arena_size / (sizeof(void*)*8/4); spans_size = arena_size / PageSize * sizeof(runtime_mheap.spans[0]); spans_size = ROUND(spans_size, PageSize); for(i = 0; i <= 0x7f; i++) { p = (void*)(uintptr)(i<<40 | 0x00c0ULL<<32); p = runtime_SysReserve(p, bitmap_size + spans_size + arena_size); if(p != nil) break; } } if (p == nil) { // On a 32-bit machine, we can't typically get away // with a giant virtual address space reservation. // Instead we map the memory information bitmap // immediately after the data segment, large enough // to handle another 2GB of mappings (256 MB), // along with a reservation for another 512 MB of memory. // When that gets used up, we'll start asking the kernel // for any memory anywhere and hope it's in the 2GB // following the bitmap (presumably the executable begins // near the bottom of memory, so we'll have to use up // most of memory before the kernel resorts to giving out // memory before the beginning of the text segment). // // Alternatively we could reserve 512 MB bitmap, enough // for 4GB of mappings, and then accept any memory the // kernel threw at us, but normally that's a waste of 512 MB // of address space, which is probably too much in a 32-bit world. bitmap_size = MaxArena32 / (sizeof(void*)*8/4); arena_size = 512<<20; spans_size = MaxArena32 / PageSize * sizeof(runtime_mheap.spans[0]); if(limit > 0 && arena_size+bitmap_size+spans_size > limit) { bitmap_size = (limit / 9) & ~((1<mcache = runtime_allocmcache(); // See if it works. runtime_free(runtime_malloc(1)); } void* runtime_MHeap_SysAlloc(MHeap *h, uintptr n) { byte *p; if(n > (uintptr)(h->arena_end - h->arena_used)) { // We are in 32-bit mode, maybe we didn't use all possible address space yet. // Reserve some more space. byte *new_end; uintptr needed; needed = (uintptr)h->arena_used + n - (uintptr)h->arena_end; needed = ROUND(needed, 256<<20); new_end = h->arena_end + needed; if(new_end <= h->arena_start + MaxArena32) { p = runtime_SysReserve(h->arena_end, new_end - h->arena_end); if(p == h->arena_end) h->arena_end = new_end; } } if(n <= (uintptr)(h->arena_end - h->arena_used)) { // Keep taking from our reservation. p = h->arena_used; runtime_SysMap(p, n, &mstats.heap_sys); h->arena_used += n; runtime_MHeap_MapBits(h); runtime_MHeap_MapSpans(h); if(raceenabled) runtime_racemapshadow(p, n); return p; } // If using 64-bit, our reservation is all we have. if(sizeof(void*) == 8 && (uintptr)h->bitmap >= 0xffffffffU) return nil; // On 32-bit, once the reservation is gone we can // try to get memory at a location chosen by the OS // and hope that it is in the range we allocated bitmap for. p = runtime_SysAlloc(n, &mstats.heap_sys); if(p == nil) return nil; if(p < h->arena_start || (uintptr)(p+n - h->arena_start) >= MaxArena32) { runtime_printf("runtime: memory allocated by OS (%p) not in usable range [%p,%p)\n", p, h->arena_start, h->arena_start+MaxArena32); runtime_SysFree(p, n, &mstats.heap_sys); return nil; } if(p+n > h->arena_used) { h->arena_used = p+n; if(h->arena_used > h->arena_end) h->arena_end = h->arena_used; runtime_MHeap_MapBits(h); runtime_MHeap_MapSpans(h); if(raceenabled) runtime_racemapshadow(p, n); } return p; } static struct { Lock; byte* pos; byte* end; } persistent; enum { PersistentAllocChunk = 256<<10, PersistentAllocMaxBlock = 64<<10, // VM reservation granularity is 64K on windows }; // Wrapper around SysAlloc that can allocate small chunks. // There is no associated free operation. // Intended for things like function/type/debug-related persistent data. // If align is 0, uses default align (currently 8). void* runtime_persistentalloc(uintptr size, uintptr align, uint64 *stat) { byte *p; if(align != 0) { if(align&(align-1)) runtime_throw("persistentalloc: align is now a power of 2"); if(align > PageSize) runtime_throw("persistentalloc: align is too large"); } else align = 8; if(size >= PersistentAllocMaxBlock) return runtime_SysAlloc(size, stat); runtime_lock(&persistent); persistent.pos = (byte*)ROUND((uintptr)persistent.pos, align); if(persistent.pos + size > persistent.end) { persistent.pos = runtime_SysAlloc(PersistentAllocChunk, &mstats.other_sys); if(persistent.pos == nil) { runtime_unlock(&persistent); runtime_throw("runtime: cannot allocate memory"); } persistent.end = persistent.pos + PersistentAllocChunk; } p = persistent.pos; persistent.pos += size; runtime_unlock(&persistent); if(stat != &mstats.other_sys) { // reaccount the allocation against provided stat runtime_xadd64(stat, size); runtime_xadd64(&mstats.other_sys, -(uint64)size); } return p; } static Lock settype_lock; void runtime_settype_flush(M *mp) { uintptr *buf, *endbuf; uintptr size, ofs, j, t; uintptr ntypes, nbytes2, nbytes3; uintptr *data2; byte *data3; void *v; uintptr typ, p; MSpan *s; buf = mp->settype_buf; endbuf = buf + mp->settype_bufsize; runtime_lock(&settype_lock); while(buf < endbuf) { v = (void*)*buf; *buf = 0; buf++; typ = *buf; buf++; // (Manually inlined copy of runtime_MHeap_Lookup) p = (uintptr)v>>PageShift; p -= (uintptr)runtime_mheap.arena_start >> PageShift; s = runtime_mheap.spans[p]; if(s->sizeclass == 0) { s->types.compression = MTypes_Single; s->types.data = typ; continue; } size = s->elemsize; ofs = ((uintptr)v - (s->start<types.compression) { case MTypes_Empty: ntypes = (s->npages << PageShift) / size; nbytes3 = 8*sizeof(uintptr) + 1*ntypes; data3 = runtime_mallocgc(nbytes3, 0, FlagNoProfiling|FlagNoScan|FlagNoInvokeGC); s->types.compression = MTypes_Bytes; s->types.data = (uintptr)data3; ((uintptr*)data3)[1] = typ; data3[8*sizeof(uintptr) + ofs] = 1; break; case MTypes_Words: ((uintptr*)s->types.data)[ofs] = typ; break; case MTypes_Bytes: data3 = (byte*)s->types.data; for(j=1; j<8; j++) { if(((uintptr*)data3)[j] == typ) { break; } if(((uintptr*)data3)[j] == 0) { ((uintptr*)data3)[j] = typ; break; } } if(j < 8) { data3[8*sizeof(uintptr) + ofs] = j; } else { ntypes = (s->npages << PageShift) / size; nbytes2 = ntypes * sizeof(uintptr); data2 = runtime_mallocgc(nbytes2, 0, FlagNoProfiling|FlagNoScan|FlagNoInvokeGC); s->types.compression = MTypes_Words; s->types.data = (uintptr)data2; // Move the contents of data3 to data2. Then deallocate data3. for(j=0; jsettype_bufsize = 0; } uintptr runtime_gettype(void *v) { MSpan *s; uintptr t, ofs; byte *data; s = runtime_MHeap_LookupMaybe(&runtime_mheap, v); if(s != nil) { t = 0; switch(s->types.compression) { case MTypes_Empty: break; case MTypes_Single: t = s->types.data; break; case MTypes_Words: ofs = (uintptr)v - (s->start<types.data)[ofs/s->elemsize]; break; case MTypes_Bytes: ofs = (uintptr)v - (s->start<types.data; t = data[8*sizeof(uintptr) + ofs/s->elemsize]; t = ((uintptr*)data)[t]; break; default: runtime_throw("runtime_gettype: invalid compression kind"); } if(0) { runtime_lock(&settype_lock); runtime_printf("%p -> %d,%X\n", v, (int32)s->types.compression, (int64)t); runtime_unlock(&settype_lock); } return t; } return 0; } // Runtime stubs. void* runtime_mal(uintptr n) { return runtime_mallocgc(n, 0, 0); } void * runtime_new(const Type *typ) { return runtime_mallocgc(typ->__size, (uintptr)typ | TypeInfo_SingleObject, typ->kind&KindNoPointers ? FlagNoScan : 0); } static void* cnew(const Type *typ, intgo n, int32 objtyp) { if((objtyp&(PtrSize-1)) != objtyp) runtime_throw("runtime: invalid objtyp"); if(n < 0 || (typ->__size > 0 && (uintptr)n > (MaxMem/typ->__size))) runtime_panicstring("runtime: allocation size out of range"); return runtime_mallocgc(typ->__size*n, (uintptr)typ | objtyp, typ->kind&KindNoPointers ? FlagNoScan : 0); } // same as runtime_new, but callable from C void* runtime_cnew(const Type *typ) { return cnew(typ, 1, TypeInfo_SingleObject); } void* runtime_cnewarray(const Type *typ, intgo n) { return cnew(typ, n, TypeInfo_Array); } func GC() { runtime_gc(1); } func SetFinalizer(obj Eface, finalizer Eface) { byte *base; uintptr size; const FuncType *ft; const Type *fint; const PtrType *ot; if(obj.__type_descriptor == nil) { runtime_printf("runtime.SetFinalizer: first argument is nil interface\n"); goto throw; } if(obj.__type_descriptor->__code != GO_PTR) { runtime_printf("runtime.SetFinalizer: first argument is %S, not pointer\n", *obj.__type_descriptor->__reflection); goto throw; } if(!runtime_mlookup(obj.__object, &base, &size, nil) || obj.__object != base) { runtime_printf("runtime.SetFinalizer: pointer not at beginning of allocated block\n"); goto throw; } ft = nil; ot = (const PtrType*)obj.__type_descriptor; fint = nil; if(finalizer.__type_descriptor != nil) { if(finalizer.__type_descriptor->__code != GO_FUNC) goto badfunc; ft = (const FuncType*)finalizer.__type_descriptor; if(ft->__dotdotdot || ft->__in.__count != 1) goto badfunc; fint = *(Type**)ft->__in.__values; if(__go_type_descriptors_equal(fint, obj.__type_descriptor)) { // ok - same type } else if(fint->__code == GO_PTR && (fint->__uncommon == nil || fint->__uncommon->__name == nil || obj.type->__uncommon == nil || obj.type->__uncommon->__name == nil) && __go_type_descriptors_equal(((const PtrType*)fint)->__element_type, ((const PtrType*)obj.type)->__element_type)) { // ok - not same type, but both pointers, // one or the other is unnamed, and same element type, so assignable. } else if(fint->kind == GO_INTERFACE && ((const InterfaceType*)fint)->__methods.__count == 0) { // ok - satisfies empty interface } else if(fint->kind == GO_INTERFACE && __go_convert_interface_2(fint, obj.__type_descriptor, 1) != nil) { // ok - satisfies non-empty interface } else goto badfunc; } if(!runtime_addfinalizer(obj.__object, finalizer.__type_descriptor != nil ? *(void**)finalizer.__object : nil, ft, ot)) { runtime_printf("runtime.SetFinalizer: finalizer already set\n"); goto throw; } return; badfunc: runtime_printf("runtime.SetFinalizer: cannot pass %S to finalizer %S\n", *obj.__type_descriptor->__reflection, *finalizer.__type_descriptor->__reflection); throw: runtime_throw("runtime.SetFinalizer"); }