gecko-dev/dom/media/MemoryBlockCache.cpp

252 строки
8.4 KiB
C++
Исходник Обычный вид История

/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim:set ts=2 sw=2 sts=2 et cindent: */
/* 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 "MemoryBlockCache.h"
#include "MediaPrefs.h"
#include "mozilla/Atomics.h"
#include "mozilla/Logging.h"
#include "mozilla/Telemetry.h"
#include "prsystem.h"
namespace mozilla {
#undef LOG
LazyLogModule gMemoryBlockCacheLog("MemoryBlockCache");
#define LOG(x, ...) \
MOZ_LOG(gMemoryBlockCacheLog, LogLevel::Debug, ("%p " x, this, ##__VA_ARGS__))
// Combined sizes of all MemoryBlockCache buffers.
// Initialized to 0 by non-local static initialization.
// Increases when a buffer grows (during initialization or unexpected OOB
// writes), decreases when a MemoryBlockCache (with its buffer) is destroyed.
static Atomic<size_t> mCombinedSizes;
enum MemoryBlockCacheTelemetryErrors
{
// Don't change order/numbers! Add new values at the end and update
// MEMORYBLOCKCACHE_ERRORS description in Histograms.json.
InitUnderuse = 0,
InitAllocation = 1,
ReadOverrun = 2,
WriteBlockOverflow = 3,
WriteBlockCannotGrow = 4,
MoveBlockSourceOverrun = 5,
MoveBlockDestOverflow = 6,
MoveBlockCannotGrow = 7,
};
MemoryBlockCache::MemoryBlockCache(int64_t aContentLength)
// Buffer whole blocks.
: mInitialContentLength((aContentLength >= 0) ? size_t(aContentLength) : 0)
, mMutex("MemoryBlockCache")
, mHasGrown(false)
{
if (aContentLength <= 0) {
LOG("MemoryBlockCache() MEMORYBLOCKCACHE_ERRORS='InitUnderuse'");
Telemetry::Accumulate(Telemetry::HistogramID::MEMORYBLOCKCACHE_ERRORS,
InitUnderuse);
}
}
MemoryBlockCache::~MemoryBlockCache()
{
size_t sizes = static_cast<size_t>(mCombinedSizes -= mBuffer.Length());
LOG("~MemoryBlockCache() - destroying buffer of size %zu; combined sizes now "
"%zu",
mBuffer.Length(),
sizes);
}
bool
MemoryBlockCache::EnsureBufferCanContain(size_t aContentLength)
{
mMutex.AssertCurrentThreadOwns();
if (aContentLength == 0) {
return true;
}
const size_t initialLength = mBuffer.Length();
const size_t desiredLength =
((aContentLength - 1) / BLOCK_SIZE + 1) * BLOCK_SIZE;
if (initialLength >= desiredLength) {
// Already large enough.
return true;
}
// Need larger buffer. If we are allowed more memory, attempt to re-allocate.
const size_t extra = desiredLength - initialLength;
// Note: There is a small race between testing `atomic + extra > limit` and
// committing to it with `atomic += extra` below; but this is acceptable, as
// in the worst case it may allow a small number of buffers to go past the
// limit.
// The alternative would have been to reserve the space first with
// `atomic += extra` and then undo it with `atomic -= extra` in case of
// failure; but this would have meant potentially preventing other (small but
// successful) allocations.
static const size_t sysmem =
std::max<size_t>(PR_GetPhysicalMemorySize(), 32 * 1024 * 1024);
const size_t limit = std::min(
size_t(MediaPrefs::MediaMemoryCachesCombinedLimitKb()) * 1024,
sysmem * MediaPrefs::MediaMemoryCachesCombinedLimitPcSysmem() / 100);
const size_t currentSizes = static_cast<size_t>(mCombinedSizes);
if (currentSizes + extra > limit) {
LOG("EnsureBufferCanContain(%zu) - buffer size %zu, wanted + %zu = %zu;"
" combined sizes %zu + %zu > limit %zu",
aContentLength,
initialLength,
extra,
desiredLength,
currentSizes,
extra,
limit);
return false;
}
if (!mBuffer.SetLength(desiredLength, mozilla::fallible)) {
LOG("EnsureBufferCanContain(%zu) - buffer size %zu, wanted + %zu = %zu, "
"allocation failed",
aContentLength,
initialLength,
extra,
desiredLength);
return false;
}
MOZ_ASSERT(mBuffer.Length() == desiredLength);
const size_t capacity = mBuffer.Capacity();
const size_t extraCapacity = capacity - desiredLength;
if (extraCapacity != 0) {
// Our buffer was given a larger capacity than the requested length, we may
// as well claim that extra capacity, both for our accounting, and to
// possibly bypass some future growths that would fit in this new capacity.
mBuffer.SetLength(capacity);
}
size_t newSizes =
static_cast<size_t>(mCombinedSizes += (extra + extraCapacity));
LOG("EnsureBufferCanContain(%zu) - buffer size %zu + requested %zu + bonus "
"%zu = %zu; combined "
"sizes %zu",
aContentLength,
initialLength,
extra,
extraCapacity,
capacity,
newSizes);
mHasGrown = true;
return true;
}
nsresult
MemoryBlockCache::Init()
{
MutexAutoLock lock(mMutex);
if (mBuffer.IsEmpty()) {
LOG("Init()");
// Attempt to pre-allocate buffer for expected content length.
if (!EnsureBufferCanContain(mInitialContentLength)) {
LOG("Init() MEMORYBLOCKCACHE_ERRORS='InitAllocation'");
Telemetry::Accumulate(Telemetry::HistogramID::MEMORYBLOCKCACHE_ERRORS,
InitAllocation);
return NS_ERROR_FAILURE;
}
} else {
LOG("Init() again");
// Re-initialization - Just erase data.
MOZ_ASSERT(mBuffer.Length() >= mInitialContentLength);
memset(mBuffer.Elements(), 0, mBuffer.Length());
}
// Ignore initial growth.
mHasGrown = false;
return NS_OK;
}
nsresult
MemoryBlockCache::WriteBlock(uint32_t aBlockIndex,
Span<const uint8_t> aData1,
Span<const uint8_t> aData2)
{
MutexAutoLock lock(mMutex);
size_t offset = BlockIndexToOffset(aBlockIndex);
if (offset + aData1.Length() + aData2.Length() > mBuffer.Length() &&
!mHasGrown) {
LOG("WriteBlock() MEMORYBLOCKCACHE_ERRORS='WriteBlockOverflow'");
Telemetry::Accumulate(Telemetry::HistogramID::MEMORYBLOCKCACHE_ERRORS,
WriteBlockOverflow);
}
if (!EnsureBufferCanContain(offset + aData1.Length() + aData2.Length())) {
LOG("WriteBlock() MEMORYBLOCKCACHE_ERRORS='WriteBlockCannotGrow'");
Telemetry::Accumulate(Telemetry::HistogramID::MEMORYBLOCKCACHE_ERRORS,
WriteBlockCannotGrow);
return NS_ERROR_FAILURE;
}
memcpy(mBuffer.Elements() + offset, aData1.Elements(), aData1.Length());
if (aData2.Length() > 0) {
memcpy(mBuffer.Elements() + offset + aData1.Length(),
aData2.Elements(),
aData2.Length());
}
return NS_OK;
}
nsresult
MemoryBlockCache::Read(int64_t aOffset,
uint8_t* aData,
int32_t aLength,
int32_t* aBytes)
{
MutexAutoLock lock(mMutex);
MOZ_ASSERT(aOffset >= 0);
if (aOffset + aLength > int64_t(mBuffer.Length())) {
LOG("Read() MEMORYBLOCKCACHE_ERRORS='ReadOverrun'");
Telemetry::Accumulate(Telemetry::HistogramID::MEMORYBLOCKCACHE_ERRORS,
ReadOverrun);
return NS_ERROR_FAILURE;
}
memcpy(aData, mBuffer.Elements() + aOffset, aLength);
*aBytes = aLength;
return NS_OK;
}
nsresult
MemoryBlockCache::MoveBlock(int32_t aSourceBlockIndex, int32_t aDestBlockIndex)
{
MutexAutoLock lock(mMutex);
size_t sourceOffset = BlockIndexToOffset(aSourceBlockIndex);
size_t destOffset = BlockIndexToOffset(aDestBlockIndex);
if (sourceOffset + BLOCK_SIZE > mBuffer.Length()) {
LOG("MoveBlock() MEMORYBLOCKCACHE_ERRORS='MoveBlockSourceOverrun'");
Telemetry::Accumulate(Telemetry::HistogramID::MEMORYBLOCKCACHE_ERRORS,
MoveBlockSourceOverrun);
return NS_ERROR_FAILURE;
}
if (destOffset + BLOCK_SIZE > mBuffer.Length() && !mHasGrown) {
LOG("MoveBlock() MEMORYBLOCKCACHE_ERRORS='MoveBlockDestOverflow'");
Telemetry::Accumulate(Telemetry::HistogramID::MEMORYBLOCKCACHE_ERRORS,
MoveBlockDestOverflow);
}
if (!EnsureBufferCanContain(destOffset + BLOCK_SIZE)) {
LOG("MoveBlock() MEMORYBLOCKCACHE_ERRORS='MoveBlockCannotGrow'");
Telemetry::Accumulate(Telemetry::HistogramID::MEMORYBLOCKCACHE_ERRORS,
MoveBlockCannotGrow);
return NS_ERROR_FAILURE;
}
memcpy(mBuffer.Elements() + destOffset,
mBuffer.Elements() + sourceOffset,
BLOCK_SIZE);
return NS_OK;
}
} // End namespace mozilla.
// avoid redefined macro in unified build
#undef LOG