在Android中，BufferQueue是Surface实现本地窗口的关键，驻留在SurfaceFlinger进程中进行服务，下面从BufferQueue的结构开始分析，
class BufferQueue : public BnGraphicBufferProducer,
public BnGraphicBufferConsumer,
private IBinder::DeathRecipient {
可见BufferQueue拥有producer和consumer两端。再看看createBufferQueue的实现，这里创建了一个BufferQueueCore，然后以这个为参数依次创建了BufferQueueProducer和BufferQueueConsumer。
void BufferQueue::createBufferQueue(sp<IGraphicBufferProducer>* outProducer,
sp<IGraphicBufferConsumer>* outConsumer,
const sp<IGraphicBufferAlloc>& allocator) {
sp<BufferQueueCore> core(new BufferQueueCore(allocator));
sp<IGraphicBufferProducer> producer(new BufferQueueProducer(core));
sp<IGraphicBufferConsumer> consumer(new BufferQueueConsumer(core));
*outProducer = producer;
*outConsumer = consumer;
}
这里的producer和consumer是要设置的，而allocator是从外面传进来的，如果传NULL，则在BufferQueueCore中会初始化，如下：
BufferQueueCore::BufferQueueCore(const sp<IGraphicBufferAlloc>& allocator) {
if (allocator == NULL) {
sp<ISurfaceComposer> composer(ComposerService::getComposerService());
mAllocator = composer->createGraphicBufferAlloc();
}
}
这里的ISurfaceComposer实现在SurfaceFlinger中，再看createGraphicBufferAlloc的实现：
sp<IGraphicBufferAlloc> SurfaceFlinger::createGraphicBufferAlloc() {
sp<GraphicBufferAlloc> gba(new GraphicBufferAlloc());
return gba;
}
这个GraphicBufferAlloc构造函数是个空壳，看看createGraphicBuffer的实现，
sp<GraphicBuffer> GraphicBufferAlloc::createGraphicBuffer(uint32_t width,
uint32_t height, PixelFormat format, uint32_t usage, status_t* error) {
sp<GraphicBuffer> graphicBuffer(
new GraphicBuffer(width, height, format, usage));
status_t err = graphicBuffer->initCheck();
return graphicBuffer;
}
GraphicBuffer继承自ANativeWindowBuffer，这个定义在window.h中，值得注意的是里面有个buffer_handle_t句柄，用于共享内存映射的文件句柄就保存在里面了。GraphicBuffer实现了Flattenable接口从而可以跨进程传输。GraphicBuffer构造函数中调用initSize开辟内存，
status_t GraphicBuffer::initSize(uint32_t inWidth, uint32_t inHeight,
PixelFormat inFormat, uint32_t inUsage) {
GraphicBufferAllocator& allocator = GraphicBufferAllocator::get();
status_t err = allocator.alloc(inWidth, inHeight, inFormat, inUsage,
&handle, &outStride);
......
return err;
}
这里又冒出来一个GraphicBufferAllocator类，注意和之前的GraphicBufferAlloc是两回事，这里是真正创建内存了。这个allocator是个单例，构造函数中加载Gralloc模块。具体的alloc应该会跟平台相关了。
GraphicBufferAllocator::GraphicBufferAllocator() : mAllocDev(0) {
hw_module_t const* module;
int err = hw_get_module(GRALLOC_HARDWARE_MODULE_ID, &module);
if (err == 0) {
gralloc_open(module, &mAllocDev);
}
}
status_t GraphicBufferAllocator::alloc(uint32_t width, uint32_t height,
PixelFormat format, uint32_t usage, buffer_handle_t* handle,
uint32_t* stride) {
err = mAllocDev->alloc(mAllocDev, static_cast<int>(width),
static_cast<int>(height), format, static_cast<int>(usage), handle,
&outStride);
......
return err;
}
到这里整个BufferQueue的创建就大致清楚了，其核心是GraphicBuffer的Allocator，最终是调用的Gralloc模块来开辟GraphicBuffer。
搜索一下BufferQueue在哪些地方被创建的，有以下几处，
- SurfaceTexture初始化时
- SurfaceFlinger初始化时，为每个显示器创建一个BufferQueue
- Layer创建时的onFirstRef中
我们重点关注Surface，这是应用端和SurfaceFlinger交互的关键了，SurfaceTexture和Layer都和Surface有千丝万缕的联系。
接下来分析BufferQueueProducer，下面是dequeueBuffer函数，首先找空闲的slot，如果buffer需要重新分配则创建GraphicBuffer设置到mSlots中。
status_t BufferQueueProducer::dequeueBuffer(int *outSlot, ...) {
int found = BufferItem::INVALID_BUFFER_SLOT;
while (found == BufferItem::INVALID_BUFFER_SLOT) {
status_t status = waitForFreeSlotThenRelock(FreeSlotCaller::Dequeue, &found);
const sp<GraphicBuffer>& buffer(mSlots[found].mGraphicBuffer);
}
const sp<GraphicBuffer>& buffer(mSlots[found].mGraphicBuffer);
*outSlot = found;
......
if (returnFlags & BUFFER_NEEDS_REALLOCATION) {
sp<GraphicBuffer> graphicBuffer = new GraphicBuffer(...);
......
mSlots[*outSlot].mGraphicBuffer = graphicBuffer;
}
return returnFlags;
}
我们再来看Surface中的dequeueBuffer，Surface作为BufferQueue的client端，其mGraphicBufferProducer必定是IGraphicBufferProducer的Bp端，对应的Bn端在BufferQueue中。这里通过dequeueBuffer后，先判断Buffer是否在BufferQueue端重新分配过了，如果是则需要调用requestBuffer刷新本地的GraphicBuffer。
int Surface::dequeueBuffer(android_native_buffer_t** buffer, int* fenceFd) {
status_t result = mGraphicBufferProducer->dequeueBuffer(&buf, &fence,
reqWidth, reqHeight, reqFormat, reqUsage,
enableFrameTimestamps ? &frameTimestamps : nullptr);
sp<GraphicBuffer>& gbuf(mSlots[buf].buffer);
if ((result & IGraphicBufferProducer::BUFFER_NEEDS_REALLOCATION) || gbuf == nullptr) {
result = mGraphicBufferProducer->requestBuffer(buf, &gbuf);
}
*buffer = gbuf.get();
return OK;
}
这里的问题是Surface和BufferQueue运行于不同的进程，其GraphicBuffer是否指向同一块物理内存。我们分析requestBuffer函数，我们首先看BpGraphicBufferProducer的transact，这里首先IPC调用REQUEST_BUFFER，获取到reply，然后本地先创建一个GraphicBuffer空壳，再从reply中将真正的数据读进来。
// IGraphicBufferProducer.cpp
virtual status_t requestBuffer(int bufferIdx, sp<GraphicBuffer>* buf) {
Parcel data, reply;
data.writeInterfaceToken(IGraphicBufferProducer::getInterfaceDescriptor());
data.writeInt32(bufferIdx);
status_t result =remote()->transact(REQUEST_BUFFER, data, &reply);
bool nonNull = reply.readInt32();
if (nonNull) {
*buf = new GraphicBuffer();
result = reply.read(**buf);
}
result = reply.readInt32();
return result;
}
再来看BnGraphicBufferProducer的onTransact，
case REQUEST_BUFFER: {
CHECK_INTERFACE(IGraphicBufferProducer, data, reply);
int bufferIdx = data.readInt32();
sp<GraphicBuffer> buffer;
int result = requestBuffer(bufferIdx, &buffer);
reply->writeInt32(buffer != 0);
if (buffer != 0) {
reply->write(*buffer);
}
reply->writeInt32(result);
return NO_ERROR;
}
这里会调用GraphicBufferProducer中的requestBuffer，然后将buffer写入reply中。可见GraphicBuffer可以通过Binder传递，
class GraphicBuffer
: public ANativeObjectBase<ANativeWindowBuffer, GraphicBuffer, RefBase>,
public Flattenable<GraphicBuffer>
原来这里实现了Flattenable，实现这个接口的对象可以序列化到buffer中，包括其中的文件描述符。我们看其unflatten实现，这里的fds很可能已经变了，这里关键是registerBuffer，很可能是要开始映射内存了。这个mBufferMapper是GraphicBufferMapper，
status_t GraphicBuffer::unflatten(
void const*& buffer, size_t& size, int const*& fds, size_t& count) {
......
if (handle != 0) {
status_t err = mBufferMapper.registerBuffer(handle);
......
}
......
return NO_ERROR;
}
再来看GraphicBufferMapper的实现，在构造函数中加载了Gralloc模块，看来是平台相关的。到这里有点眼熟，之前提到的GraphicBufferAllocator初始化时也会加载Gralloc模块，那是在SurfaceFlinger中真正创建GraphicBuffer用的，而这里应用端只需将句柄映射到自己的内存空间即可，所以有了这个GraphicBufferMapper。
GraphicBufferMapper::GraphicBufferMapper() : mAllocMod(0) {
hw_module_t const* module;
int err = hw_get_module(GRALLOC_HARDWARE_MODULE_ID, &module);
if (err == 0) {
mAllocMod = (gralloc_module_t const *) module;
}
}
status_t GraphicBufferMapper::registerBuffer(buffer_handle_t handle) {
status_t err;
err = mAllocMod->registerBuffer(mAllocMod, handle);
return err;
}
这里的registerBuffer是和平台相关的，我们看msm8960平台的实现：
gralloc_register_buffer(gralloc_module_t const* module,
buffer_handle_t handle) {
private_handle_t* hnd = (private_handle_t*)handle;
hnd->base = 0;
hnd->base_metadata = 0;
int err = gralloc_map(module, handle);
return 0;
}
这里调用了gralloc_map将handle句柄映射到自己的进程空间，而这块区域与BufferQueue中指向的物理空间是一致的，从而实现了两者跨进程缓冲区共享。
到这里我们大致明白了本地Surface和对端BufferQueue之间的dequeueBuffer时的过程，接下来我们了解一下Surface是怎么和BufferQueue建立联系的。
我们在本文开头提到，在Layer创建时的onFirstRef中会createBufferQueue，而Layer的创建在SurfaceFlinger中的createLayer函数，而这个函数被Client的createSurface调用，这个Client是BnSurfaceComposerClient的子类，因此其中包含createSurface的实现，我们注意到里面会将所有Bp端发起的createSurface请求串行化后再丢给SurfaceFlinger。
再看BpSurfaceComposerClient中的createSurface是被谁调到的，我们发现SurfaceControl中的nativeCreate时会调到，这是SurfaceControl的构造函数中调到的。
// android_view_SurfaceControl.cpp
static jint nativeCreate(JNIEnv* env, jclass clazz, jobject sessionObj,
jstring nameStr, jint w, jint h, jint format, jint flags) {
sp<SurfaceComposerClient> client(android_view_SurfaceSession_getClient(env, sessionObj));
sp<SurfaceControl> surface = client->createSurface(
String8(name.c_str()), w, h, format, flags);
surface->incStrong((void *)nativeCreate);
return int(surface.get());
}
这里createSurface返回的是SurfaceControl，这个其实是SurfaceComposerClient封装了一层，将远端返回的Binder句柄和GraphicBufferProducer都封装起来了，所以通过SurfaceControl可以和SurfaceFlinger通信。再看nativeCreate中通过android_view_SurfaceSession_getClient获取SurfaceComposerClient，看看SurfaceSession是什么，这个是连接到SurfaceFlinger的，Java类中保存了一个long型的mNativeClient，这在Native层对应着SurfaceComposerClient。
public final class SurfaceSession {
// Note: This field is accessed by native code.
private long mNativeClient; // SurfaceComposerClient*
private static native long nativeCreate();
private static native long nativeCreateScoped(long surfacePtr);
private static native void nativeDestroy(long ptr);
private static native void nativeKill(long ptr);
/** Create a new connection with the surface flinger. */
public SurfaceSession() {
mNativeClient = nativeCreate();
}
......
}
再看Native层实现：
static jlong nativeCreate(JNIEnv* env, jclass clazz) {
SurfaceComposerClient* client = new SurfaceComposerClient();
client->incStrong((void*)nativeCreate);
return reinterpret_cast<jlong>(client);
}
这里创建了一个SurfaceComposerClient，并没有看到连到SurfaceFlinger端，看其构造函数也没有，我们想到onFirstRef，
void SurfaceComposerClient::onFirstRef() {
sp<ISurfaceComposer> sm(ComposerService::getComposerService());
if (sm != 0) {
auto rootProducer = mParent.promote();
sp<ISurfaceComposerClient> conn;
conn = sm->createConnection();
if (conn != 0) {
mClient = conn;
mStatus = NO_ERROR;
}
}
}
到这里就清楚了，ISurfaceComposer是个与SurfaceFlinger通信的Binder句柄，通过该句柄的createConnection可以获取ISurfaceComposerClient，而SurfaceComposerClient是在其基础上包装了一层而已。
所以本地应用需要通过SurfaceSession建立和SurfaceFlinger连接，获取SurfaceComposerClient，然后再通过这个句柄调用createSurface，返回GraphicBufferProducer，封装到SurfaceControl中，有了这个GraphicBufferProducer接下来就好办了。