/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ /* vim: set ts=8 sts=2 et sw=2 tw=80: */ /* This Source Code Form is subject to the terms of the Mozilla Public * License, v. 2.0. If a copy of the MPL was not distributed with this * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ #include #include #include #include #include "PLDHashTable.h" #include "mozilla/HashFunctions.h" #include "mozilla/MathAlgorithms.h" #include "nsAlgorithm.h" #include "mozilla/Likely.h" #include "mozilla/MemoryReporting.h" #include "mozilla/ChaosMode.h" using namespace mozilla; #ifdef DEBUG class AutoReadOp { Checker& mChk; public: explicit AutoReadOp(Checker& aChk) : mChk(aChk) { mChk.StartReadOp(); } ~AutoReadOp() { mChk.EndReadOp(); } }; class AutoWriteOp { Checker& mChk; public: explicit AutoWriteOp(Checker& aChk) : mChk(aChk) { mChk.StartWriteOp(); } ~AutoWriteOp() { mChk.EndWriteOp(); } }; class AutoIteratorRemovalOp { Checker& mChk; public: explicit AutoIteratorRemovalOp(Checker& aChk) : mChk(aChk) { mChk.StartIteratorRemovalOp(); } ~AutoIteratorRemovalOp() { mChk.EndIteratorRemovalOp(); } }; class AutoDestructorOp { Checker& mChk; public: explicit AutoDestructorOp(Checker& aChk) : mChk(aChk) { mChk.StartDestructorOp(); } ~AutoDestructorOp() { mChk.EndDestructorOp(); } }; #endif /* static */ PLDHashNumber PLDHashTable::HashStringKey(PLDHashTable* aTable, const void* aKey) { return HashString(static_cast(aKey)); } /* static */ PLDHashNumber PLDHashTable::HashVoidPtrKeyStub(PLDHashTable* aTable, const void* aKey) { return (PLDHashNumber)(ptrdiff_t)aKey >> 2; } /* static */ bool PLDHashTable::MatchEntryStub(PLDHashTable* aTable, const PLDHashEntryHdr* aEntry, const void* aKey) { const PLDHashEntryStub* stub = (const PLDHashEntryStub*)aEntry; return stub->key == aKey; } /* static */ bool PLDHashTable::MatchStringKey(PLDHashTable* aTable, const PLDHashEntryHdr* aEntry, const void* aKey) { const PLDHashEntryStub* stub = (const PLDHashEntryStub*)aEntry; // XXX tolerate null keys on account of sloppy Mozilla callers. return stub->key == aKey || (stub->key && aKey && strcmp((const char*)stub->key, (const char*)aKey) == 0); } /* static */ void PLDHashTable::MoveEntryStub(PLDHashTable* aTable, const PLDHashEntryHdr* aFrom, PLDHashEntryHdr* aTo) { memcpy(aTo, aFrom, aTable->mEntrySize); } /* static */ void PLDHashTable::ClearEntryStub(PLDHashTable* aTable, PLDHashEntryHdr* aEntry) { memset(aEntry, 0, aTable->mEntrySize); } static const PLDHashTableOps gStubOps = { PLDHashTable::HashVoidPtrKeyStub, PLDHashTable::MatchEntryStub, PLDHashTable::MoveEntryStub, PLDHashTable::ClearEntryStub, nullptr }; /* static */ const PLDHashTableOps* PLDHashTable::StubOps() { return &gStubOps; } static bool SizeOfEntryStore(uint32_t aCapacity, uint32_t aEntrySize, uint32_t* aNbytes) { uint64_t nbytes64 = uint64_t(aCapacity) * uint64_t(aEntrySize); *aNbytes = aCapacity * aEntrySize; return uint64_t(*aNbytes) == nbytes64; // returns false on overflow } // Compute max and min load numbers (entry counts). We have a secondary max // that allows us to overload a table reasonably if it cannot be grown further // (i.e. if ChangeTable() fails). The table slows down drastically if the // secondary max is too close to 1, but 0.96875 gives only a slight slowdown // while allowing 1.3x more elements. static inline uint32_t MaxLoad(uint32_t aCapacity) { return aCapacity - (aCapacity >> 2); // == aCapacity * 0.75 } static inline uint32_t MaxLoadOnGrowthFailure(uint32_t aCapacity) { return aCapacity - (aCapacity >> 5); // == aCapacity * 0.96875 } static inline uint32_t MinLoad(uint32_t aCapacity) { return aCapacity >> 2; // == aCapacity * 0.25 } // Compute the minimum capacity (and the Log2 of that capacity) for a table // containing |aLength| elements while respecting the following contraints: // - table must be at most 75% full; // - capacity must be a power of two; // - capacity cannot be too small. static inline void BestCapacity(uint32_t aLength, uint32_t* aCapacityOut, uint32_t* aLog2CapacityOut) { // Compute the smallest capacity allowing |aLength| elements to be inserted // without rehashing. uint32_t capacity = (aLength * 4 + (3 - 1)) / 3; // == ceil(aLength * 4 / 3) if (capacity < PLDHashTable::kMinCapacity) { capacity = PLDHashTable::kMinCapacity; } // Round up capacity to next power-of-two. uint32_t log2 = CeilingLog2(capacity); capacity = 1u << log2; MOZ_ASSERT(capacity <= PLDHashTable::kMaxCapacity); *aCapacityOut = capacity; *aLog2CapacityOut = log2; } /* static */ MOZ_ALWAYS_INLINE uint32_t PLDHashTable::HashShift(uint32_t aEntrySize, uint32_t aLength) { if (aLength > kMaxInitialLength) { MOZ_CRASH("Initial length is too large"); } uint32_t capacity, log2; BestCapacity(aLength, &capacity, &log2); uint32_t nbytes; if (!SizeOfEntryStore(capacity, aEntrySize, &nbytes)) { MOZ_CRASH("Initial entry store size is too large"); } // Compute the hashShift value. return kHashBits - log2; } PLDHashTable::PLDHashTable(const PLDHashTableOps* aOps, uint32_t aEntrySize, uint32_t aLength) : mOps(aOps) , mHashShift(HashShift(aEntrySize, aLength)) , mEntrySize(aEntrySize) , mEntryCount(0) , mRemovedCount(0) , mEntryStore() #ifdef DEBUG , mChecker() #endif { } PLDHashTable& PLDHashTable::operator=(PLDHashTable&& aOther) { if (this == &aOther) { return *this; } // Destruct |this|. this->~PLDHashTable(); // |mOps| and |mEntrySize| are const so we can't assign them. Instead, we // require that they are equal. The justification for this is that they're // conceptually part of the type -- indeed, if PLDHashTable was a templated // type like nsTHashtable, they *would* be part of the type -- so it only // makes sense to assign in cases where they match. MOZ_RELEASE_ASSERT(mOps == aOther.mOps); MOZ_RELEASE_ASSERT(mEntrySize == aOther.mEntrySize); // Move non-const pieces over. mHashShift = Move(aOther.mHashShift); mEntryCount = Move(aOther.mEntryCount); mRemovedCount = Move(aOther.mRemovedCount); mEntryStore = Move(aOther.mEntryStore); #ifdef DEBUG mChecker = Move(aOther.mChecker); #endif // Clear up |aOther| so its destruction will be a no-op. { #ifdef DEBUG AutoDestructorOp op(mChecker); #endif aOther.mEntryStore.Set(nullptr); } return *this; } PLDHashNumber PLDHashTable::Hash1(PLDHashNumber aHash0) { return aHash0 >> mHashShift; } // Double hashing needs the second hash code to be relatively prime to table // size, so we simply make hash2 odd. void PLDHashTable::Hash2(PLDHashNumber aHash, uint32_t& aHash2Out, uint32_t& aSizeMaskOut) { uint32_t sizeLog2 = kHashBits - mHashShift; aHash2Out = ((aHash << sizeLog2) >> mHashShift) | 1; aSizeMaskOut = (PLDHashNumber(1) << sizeLog2) - 1; } // Reserve mKeyHash 0 for free entries and 1 for removed-entry sentinels. Note // that a removed-entry sentinel need be stored only if the removed entry had // a colliding entry added after it. Therefore we can use 1 as the collision // flag in addition to the removed-entry sentinel value. Multiplicative hash // uses the high order bits of mKeyHash, so this least-significant reservation // should not hurt the hash function's effectiveness much. /* static */ MOZ_ALWAYS_INLINE bool PLDHashTable::EntryIsFree(PLDHashEntryHdr* aEntry) { return aEntry->mKeyHash == 0; } /* static */ MOZ_ALWAYS_INLINE bool PLDHashTable::EntryIsRemoved(PLDHashEntryHdr* aEntry) { return aEntry->mKeyHash == 1; } /* static */ MOZ_ALWAYS_INLINE bool PLDHashTable::EntryIsLive(PLDHashEntryHdr* aEntry) { return aEntry->mKeyHash >= 2; } /* static */ MOZ_ALWAYS_INLINE void PLDHashTable::MarkEntryFree(PLDHashEntryHdr* aEntry) { aEntry->mKeyHash = 0; } /* static */ MOZ_ALWAYS_INLINE void PLDHashTable::MarkEntryRemoved(PLDHashEntryHdr* aEntry) { aEntry->mKeyHash = 1; } // Match an entry's mKeyHash against an unstored one computed from a key. /* static */ bool PLDHashTable::MatchEntryKeyhash(PLDHashEntryHdr* aEntry, PLDHashNumber aKeyHash) { return (aEntry->mKeyHash & ~kCollisionFlag) == aKeyHash; } // Compute the address of the indexed entry in table. PLDHashEntryHdr* PLDHashTable::AddressEntry(uint32_t aIndex) { return reinterpret_cast( mEntryStore.Get() + aIndex * mEntrySize); } PLDHashTable::~PLDHashTable() { #ifdef DEBUG AutoDestructorOp op(mChecker); #endif if (!mEntryStore.Get()) { return; } // Clear any remaining live entries. char* entryAddr = mEntryStore.Get(); char* entryLimit = entryAddr + Capacity() * mEntrySize; while (entryAddr < entryLimit) { PLDHashEntryHdr* entry = (PLDHashEntryHdr*)entryAddr; if (EntryIsLive(entry)) { mOps->clearEntry(this, entry); } entryAddr += mEntrySize; } // Entry storage is freed last, by ~EntryStore(). } void PLDHashTable::ClearAndPrepareForLength(uint32_t aLength) { // Get these values before the destructor clobbers them. const PLDHashTableOps* ops = mOps; uint32_t entrySize = mEntrySize; this->~PLDHashTable(); new (this) PLDHashTable(ops, entrySize, aLength); } void PLDHashTable::Clear() { ClearAndPrepareForLength(kDefaultInitialLength); } // If |IsAdd| is true, the return value is always non-null and it may be a // previously-removed entry. If |IsAdd| is false, the return value is null on a // miss, and will never be a previously-removed entry on a hit. This // distinction is a bit grotty but this function is hot enough that these // differences are worthwhile. template PLDHashEntryHdr* PL_DHASH_FASTCALL PLDHashTable::SearchTable(const void* aKey, PLDHashNumber aKeyHash) { MOZ_ASSERT(mEntryStore.Get()); NS_ASSERTION(!(aKeyHash & kCollisionFlag), "!(aKeyHash & kCollisionFlag)"); // Compute the primary hash address. PLDHashNumber hash1 = Hash1(aKeyHash); PLDHashEntryHdr* entry = AddressEntry(hash1); // Miss: return space for a new entry. if (EntryIsFree(entry)) { return (Reason == ForAdd) ? entry : nullptr; } // Hit: return entry. PLDHashMatchEntry matchEntry = mOps->matchEntry; if (MatchEntryKeyhash(entry, aKeyHash) && matchEntry(this, entry, aKey)) { return entry; } // Collision: double hash. PLDHashNumber hash2; uint32_t sizeMask; Hash2(aKeyHash, hash2, sizeMask); // Save the first removed entry pointer so Add() can recycle it. (Only used // if Reason==ForAdd.) PLDHashEntryHdr* firstRemoved = nullptr; for (;;) { if (Reason == ForAdd) { if (MOZ_UNLIKELY(EntryIsRemoved(entry))) { if (!firstRemoved) { firstRemoved = entry; } } else { entry->mKeyHash |= kCollisionFlag; } } hash1 -= hash2; hash1 &= sizeMask; entry = AddressEntry(hash1); if (EntryIsFree(entry)) { return (Reason == ForAdd) ? (firstRemoved ? firstRemoved : entry) : nullptr; } if (MatchEntryKeyhash(entry, aKeyHash) && matchEntry(this, entry, aKey)) { return entry; } } // NOTREACHED return nullptr; } // This is a copy of SearchTable(), used by ChangeTable(), hardcoded to // 1. assume |aIsAdd| is true, // 2. assume that |aKey| will never match an existing entry, and // 3. assume that no entries have been removed from the current table // structure. // Avoiding the need for |aKey| means we can avoid needing a way to map entries // to keys, which means callers can use complex key types more easily. PLDHashEntryHdr* PL_DHASH_FASTCALL PLDHashTable::FindFreeEntry(PLDHashNumber aKeyHash) { MOZ_ASSERT(mEntryStore.Get()); NS_ASSERTION(!(aKeyHash & kCollisionFlag), "!(aKeyHash & kCollisionFlag)"); // Compute the primary hash address. PLDHashNumber hash1 = Hash1(aKeyHash); PLDHashEntryHdr* entry = AddressEntry(hash1); // Miss: return space for a new entry. if (EntryIsFree(entry)) { return entry; } // Collision: double hash. PLDHashNumber hash2; uint32_t sizeMask; Hash2(aKeyHash, hash2, sizeMask); for (;;) { NS_ASSERTION(!EntryIsRemoved(entry), "!EntryIsRemoved(entry)"); entry->mKeyHash |= kCollisionFlag; hash1 -= hash2; hash1 &= sizeMask; entry = AddressEntry(hash1); if (EntryIsFree(entry)) { return entry; } } // NOTREACHED } bool PLDHashTable::ChangeTable(int32_t aDeltaLog2) { MOZ_ASSERT(mEntryStore.Get()); // Look, but don't touch, until we succeed in getting new entry store. int32_t oldLog2 = kHashBits - mHashShift; int32_t newLog2 = oldLog2 + aDeltaLog2; uint32_t newCapacity = 1u << newLog2; if (newCapacity > kMaxCapacity) { return false; } uint32_t nbytes; if (!SizeOfEntryStore(newCapacity, mEntrySize, &nbytes)) { return false; // overflowed } char* newEntryStore = (char*)malloc(nbytes); if (!newEntryStore) { return false; } // We can't fail from here on, so update table parameters. mHashShift = kHashBits - newLog2; mRemovedCount = 0; // Assign the new entry store to table. memset(newEntryStore, 0, nbytes); char* oldEntryStore; char* oldEntryAddr; oldEntryAddr = oldEntryStore = mEntryStore.Get(); mEntryStore.Set(newEntryStore); PLDHashMoveEntry moveEntry = mOps->moveEntry; // Copy only live entries, leaving removed ones behind. uint32_t oldCapacity = 1u << oldLog2; for (uint32_t i = 0; i < oldCapacity; ++i) { PLDHashEntryHdr* oldEntry = (PLDHashEntryHdr*)oldEntryAddr; if (EntryIsLive(oldEntry)) { oldEntry->mKeyHash &= ~kCollisionFlag; PLDHashEntryHdr* newEntry = FindFreeEntry(oldEntry->mKeyHash); NS_ASSERTION(EntryIsFree(newEntry), "EntryIsFree(newEntry)"); moveEntry(this, oldEntry, newEntry); newEntry->mKeyHash = oldEntry->mKeyHash; } oldEntryAddr += mEntrySize; } free(oldEntryStore); return true; } MOZ_ALWAYS_INLINE PLDHashNumber PLDHashTable::ComputeKeyHash(const void* aKey) { MOZ_ASSERT(mEntryStore.Get()); PLDHashNumber keyHash = mOps->hashKey(this, aKey); keyHash *= kGoldenRatio; // Avoid 0 and 1 hash codes, they indicate free and removed entries. if (keyHash < 2) { keyHash -= 2; } keyHash &= ~kCollisionFlag; return keyHash; } PLDHashEntryHdr* PLDHashTable::Search(const void* aKey) { #ifdef DEBUG AutoReadOp op(mChecker); #endif PLDHashEntryHdr* entry = mEntryStore.Get() ? SearchTable(aKey, ComputeKeyHash(aKey)) : nullptr; return entry; } PLDHashEntryHdr* PLDHashTable::Add(const void* aKey, const mozilla::fallible_t&) { #ifdef DEBUG AutoWriteOp op(mChecker); #endif // Allocate the entry storage if it hasn't already been allocated. if (!mEntryStore.Get()) { uint32_t nbytes; // We already checked this in the constructor, so it must still be true. MOZ_RELEASE_ASSERT(SizeOfEntryStore(CapacityFromHashShift(), mEntrySize, &nbytes)); mEntryStore.Set((char*)malloc(nbytes)); if (!mEntryStore.Get()) { return nullptr; } memset(mEntryStore.Get(), 0, nbytes); } // If alpha is >= .75, grow or compress the table. If aKey is already in the // table, we may grow once more than necessary, but only if we are on the // edge of being overloaded. uint32_t capacity = Capacity(); if (mEntryCount + mRemovedCount >= MaxLoad(capacity)) { // Compress if a quarter or more of all entries are removed. int deltaLog2; if (mRemovedCount >= capacity >> 2) { deltaLog2 = 0; } else { deltaLog2 = 1; } // Grow or compress the table. If ChangeTable() fails, allow overloading up // to the secondary max. Once we hit the secondary max, return null. if (!ChangeTable(deltaLog2) && mEntryCount + mRemovedCount >= MaxLoadOnGrowthFailure(capacity)) { return nullptr; } } // Look for entry after possibly growing, so we don't have to add it, // then skip it while growing the table and re-add it after. PLDHashNumber keyHash = ComputeKeyHash(aKey); PLDHashEntryHdr* entry = SearchTable(aKey, keyHash); if (!EntryIsLive(entry)) { // Initialize the entry, indicating that it's no longer free. if (EntryIsRemoved(entry)) { mRemovedCount--; keyHash |= kCollisionFlag; } if (mOps->initEntry) { mOps->initEntry(entry, aKey); } entry->mKeyHash = keyHash; mEntryCount++; } return entry; } PLDHashEntryHdr* PLDHashTable::Add(const void* aKey) { PLDHashEntryHdr* entry = Add(aKey, fallible); if (!entry) { if (!mEntryStore.Get()) { // We OOM'd while allocating the initial entry storage. uint32_t nbytes; (void) SizeOfEntryStore(CapacityFromHashShift(), mEntrySize, &nbytes); NS_ABORT_OOM(nbytes); } else { // We failed to resize the existing entry storage, either due to OOM or // because we exceeded the maximum table capacity or size; report it as // an OOM. The multiplication by 2 gets us the size we tried to allocate, // which is double the current size. NS_ABORT_OOM(2 * EntrySize() * EntryCount()); } } return entry; } void PLDHashTable::Remove(const void* aKey) { #ifdef DEBUG AutoWriteOp op(mChecker); #endif PLDHashEntryHdr* entry = mEntryStore.Get() ? SearchTable(aKey, ComputeKeyHash(aKey)) : nullptr; if (entry) { RawRemove(entry); ShrinkIfAppropriate(); } } void PLDHashTable::RemoveEntry(PLDHashEntryHdr* aEntry) { #ifdef DEBUG AutoWriteOp op(mChecker); #endif RawRemove(aEntry); ShrinkIfAppropriate(); } void PLDHashTable::RawRemove(PLDHashEntryHdr* aEntry) { // Unfortunately, we can only do weak checking here. That's because // RawRemove() can be called legitimately while an Enumerate() call is // active, which doesn't fit well into how Checker's mState variable works. MOZ_ASSERT(mChecker.IsWritable()); MOZ_ASSERT(mEntryStore.Get()); MOZ_ASSERT(EntryIsLive(aEntry), "EntryIsLive(aEntry)"); // Load keyHash first in case clearEntry() goofs it. PLDHashNumber keyHash = aEntry->mKeyHash; mOps->clearEntry(this, aEntry); if (keyHash & kCollisionFlag) { MarkEntryRemoved(aEntry); mRemovedCount++; } else { MarkEntryFree(aEntry); } mEntryCount--; } // Shrink or compress if a quarter or more of all entries are removed, or if the // table is underloaded according to the minimum alpha, and is not minimal-size // already. void PLDHashTable::ShrinkIfAppropriate() { uint32_t capacity = Capacity(); if (mRemovedCount >= capacity >> 2 || (capacity > kMinCapacity && mEntryCount <= MinLoad(capacity))) { uint32_t log2; BestCapacity(mEntryCount, &capacity, &log2); int32_t deltaLog2 = log2 - (kHashBits - mHashShift); MOZ_ASSERT(deltaLog2 <= 0); (void) ChangeTable(deltaLog2); } } size_t PLDHashTable::ShallowSizeOfExcludingThis(MallocSizeOf aMallocSizeOf) const { #ifdef DEBUG AutoReadOp op(mChecker); #endif return aMallocSizeOf(mEntryStore.Get()); } size_t PLDHashTable::ShallowSizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const { return aMallocSizeOf(this) + ShallowSizeOfExcludingThis(aMallocSizeOf); } PLDHashTable::Iterator::Iterator(Iterator&& aOther) : mTable(aOther.mTable) , mStart(aOther.mStart) , mLimit(aOther.mLimit) , mCurrent(aOther.mCurrent) , mNexts(aOther.mNexts) , mNextsLimit(aOther.mNextsLimit) , mHaveRemoved(aOther.mHaveRemoved) { // No need to change |mChecker| here. aOther.mTable = nullptr; aOther.mStart = nullptr; aOther.mLimit = nullptr; aOther.mCurrent = nullptr; aOther.mNexts = 0; aOther.mNextsLimit = 0; aOther.mHaveRemoved = false; } PLDHashTable::Iterator::Iterator(PLDHashTable* aTable) : mTable(aTable) , mStart(mTable->mEntryStore.Get()) , mLimit(mTable->mEntryStore.Get() + mTable->Capacity() * mTable->mEntrySize) , mCurrent(mTable->mEntryStore.Get()) , mNexts(0) , mNextsLimit(mTable->EntryCount()) , mHaveRemoved(false) { #ifdef DEBUG mTable->mChecker.StartReadOp(); #endif if (ChaosMode::isActive(ChaosFeature::HashTableIteration) && mTable->Capacity() > 0) { // Start iterating at a random entry. It would be even more chaotic to // iterate in fully random order, but that's harder. mCurrent += ChaosMode::randomUint32LessThan(mTable->Capacity()) * mTable->mEntrySize; } // Advance to the first live entry, if there is one. if (!Done()) { while (IsOnNonLiveEntry()) { MoveToNextEntry(); } } } PLDHashTable::Iterator::~Iterator() { if (mTable) { if (mHaveRemoved) { mTable->ShrinkIfAppropriate(); } #ifdef DEBUG mTable->mChecker.EndReadOp(); #endif } } bool PLDHashTable::Iterator::Done() const { return mNexts == mNextsLimit; } MOZ_ALWAYS_INLINE bool PLDHashTable::Iterator::IsOnNonLiveEntry() const { MOZ_ASSERT(!Done()); return !EntryIsLive(reinterpret_cast(mCurrent)); } MOZ_ALWAYS_INLINE void PLDHashTable::Iterator::MoveToNextEntry() { mCurrent += mTable->mEntrySize; if (mCurrent == mLimit) { mCurrent = mStart; // Wrap-around. Possible due to Chaos Mode. } } PLDHashEntryHdr* PLDHashTable::Iterator::Get() const { MOZ_ASSERT(!Done()); PLDHashEntryHdr* entry = reinterpret_cast(mCurrent); MOZ_ASSERT(EntryIsLive(entry)); return entry; } void PLDHashTable::Iterator::Next() { MOZ_ASSERT(!Done()); mNexts++; // Advance to the next live entry, if there is one. if (!Done()) { do { MoveToNextEntry(); } while (IsOnNonLiveEntry()); } } void PLDHashTable::Iterator::Remove() { // This cast is needed for the same reason as the one in the destructor. mTable->RawRemove(Get()); mHaveRemoved = true; } #ifdef DEBUG void PLDHashTable::MarkImmutable() { mChecker.SetNonWritable(); } #endif