前言

• Camera sensor获取自然图像（图像代表着image data buffer）通过ISP处理之后到显示器显示，这个过程代表着buffer从camera到display的传输，也是camera module中一个重要的流程，在此博客中和大家分享。
• 为弄清楚snapdragon camera APP preview功能下image buffer是怎么从camera module传输到display module进行显示，本文详细介绍总体流程其中包括buffer flow和各部分组件功能。
• 主要涉及四个部分：1.snapdragon camera APP：发送image buffer request，设置surfaceview；2.camera framework and camera service：处理request回传result；3.surfaceflinger：处理buffe queue；4.surface view：建立显示surface，触发preview buffer。
• 图片title使用斜体表示

Camera preview 总体概括

Fig1是Google提供的camera 2API调用流程，主要关注画红线camera preview buffer flow流程，下面分为几个部分介绍：

Fig1，camera preview buffer flow

Establish snapdragon surfaceview surface

首先，Snapdragon camera preview功能通过surfaceview创建surface，通过其显示image buffer。camera device createCaptureSession把创建成功surface传输下去。下面分析具体代码流程。
首先初始化创建surfaceview，并为其设置状态回调callback。

// display the view
mSurfaceView = (AutoFitSurfaceView) mRootView.findViewById(R.id.mdp_preview_content);
mSurfaceHolder = mSurfaceView.getHolder();
mSurfaceHolder.addCallback(callback);
mSurfaceView.addOnLayoutChangeListener(new View.OnLayoutChangeListener() {@Overridepublic void onLayoutChange(View v, int left, int top, int right,int bottom, int oldLeft, int oldTop, int oldRight,int oldBottom) {int width = right - left;int height = bottom - top;if (mFaceView != null) {mFaceView.onSurfaceTextureSizeChanged(width, height);}if (mStatsNNFocusRenderer != null) {mStatsNNFocusRenderer.onSurfaceTextureSizeChanged(width, height);}if (mT2TFocusRenderer != null) {mT2TFocusRenderer.onSurfaceTextureSizeChanged(width, height);}}
});

当surfaceview创建成功时回调surfaceCreated，通过调用previewUIReady()去通知CaptureModule开启预览的操作，

private SurfaceHolder.Callback callback = new SurfaceHolder.Callback() {// SurfaceHolder callbacks@Overridepublic void surfaceCreated(SurfaceHolder holder) {Log.v(TAG, "surfaceCreated");mSurfaceHolder = holder;previewUIReady();if(mTrackingFocusRenderer != null && mTrackingFocusRenderer.isVisible()) {mTrackingFocusRenderer.setSurfaceDim(mSurfaceView.getLeft(), mSurfaceView.getTop(), mSurfaceView.getRight(), mSurfaceView.getBottom());}if(mT2TFocusRenderer != null && mT2TFocusRenderer.isShown()) {mT2TFocusRenderer.setSurfaceDim(mSurfaceView.getLeft(), mSurfaceView.getTop(),mSurfaceView.getRight(), mSurfaceView.getBottom());}if(mStatsNNFocusRenderer != null && mStatsNNFocusRenderer.isShown()) {mStatsNNFocusRenderer.setSurfaceDim(mSurfaceView.getLeft(), mSurfaceView.getTop(),mSurfaceView.getRight(), mSurfaceView.getBottom());}}
};

判断当前时刻CaptureUI的预览surface view是否创建成功并可用，获取此预览surface。

Surface surface = null;
try {waitForPreviewSurfaceReady();
} catch (RuntimeException e) {Log.v(TAG,"createSession: normal status occur Time out waiting for surface ");
}
surface = getPreviewSurfaceForSession(id);

开始准备预览流显示surface通过create session配置。

List<Surface> surfaces = mFrameProcessor.getInputSurfaces();
for(Surface surs : surfaces) {mPreviewRequestBuilder[id].addTarget(surs);list.add(surs);
}

Camera Device Create Preview Session

• 回到Fig1，随后Camera device创建capture session之后，通过setRepeatingRequest不断的发送request请求Show preview stream(想象stream是image buffer在不断的循环)，request在发送的过程中包含在queue的结构（repeating request list -> pending request queue -> in-progress queue），保证每时每刻camera HAL都能够有足够request buffer处理。
• Camera Device hardware对image buffer指向地址填充完毕之后，通过output stream（configured outputs）传输到surfaceview 创建的surface显示，自此整个体系形成一个生产-消费-归还的循环。

Camera preview 详细流程

Fig2展现camera preview显示图像总体流程，并且表达出各部分之间的关系，一供分为三个层次结构。

• 顶部JAVA layer包含Java APP、Java framework，包括Snapdragon camera APP（未画出），surface view and display区域。
• 中部Native layer，包括camera service process and surfaceflinger process。
• 底部HAL layer，HIDL只是作为中间连接层，包含camera provider process，其中包含Camx module。
  
  Fig2 camera preview显示图像总体流程

Camera Buffer Flow

由于Fig2复杂性，首先介绍camera buffer flow勾勒出camera sensor获取image，处理之后传输到display显示的主线。之后在对其进行补充。
camera的buffer在使用功能上主要分为两类：output stream buffer and input stream buffer。Output and input主语是camera hardware，如果从camera HAL输出就是output stream。

• output buffer是empty buffer，request携带output buffer到HAL的过程中里面没有数据，HAL会将生成的buffer地址填入到output buffer，最后通过callback返回到framework。
• input buffer是有实际内容的buffer，通过reprocess流程将其送到HAL Layer处理。reprocess的使用情况举例，如使用HDR algorithm：native layer算法需要使用raw image进行处理合成，会通过output stream获取。处理完之后再通过input stream把raw image送给HAL layer，对其raw image转为YUV，再对其进行后续ISP algorithm process。Native layer algorithm use raw image处理会有更高的精度和无细节损失，但是Native layer并不能直接把raw image传输给Java layer使用，需要使用ISP硬件转换处理，所以reprocess再次传输到HAL。
• 再说一个input buffer和output buffer代码层面根本的区别，input stream作为comsumer是从buffer queue中acquire graphic buffer（comsumer拿去真正消费掉）。output stream从buffer queue中dequeue graphic buffer，虽然代码中都是把comsumer 作为alias，但是并不是真正的消费掉，而是传输到HAL进行sensor image data的填充，是一种对buffer内容的produce（producer在生产内容）。

Fig3显示一个生产消费者模型，buffer在其中是一个循环的流程（具体参考：BufferQueue 学习总结）。

Fig3 生产消费者模型

应用到camera service 、 surfaceflinger and surfaceview之间关系的情境下，figure转化为Fig4。
真正的producer是surfaceview创建的native 对象ANativeWindow， nativewindow创建出preview stream buffer，surface and camera service之间有一个dequeue、queue循环buffer过程。camera service output stream作为中间bridge，通过binder调用到surfaceflinger，surfaceflinger使用buffer queue结构管理preview graphic buffer。

Fig4 camera service 、 surfaceflinger and surfaceview生产消费者模型

本文只介绍camera捕获image到display显示，所以只分析Fig4从surface、camera service到surfaceflinger单向流程。

Camera Service dequeue buffer

函数调用逻辑如图Fig5显示，Camera3Device中threadloop函数一旦run之后就一直循环，不断下发camera preview request。Camera3OutputStream直接使用Anativewindow dequeue buffer，之后传递buffer handle让camera provider填充image data到申请好的buffer地址中。

status_t Camera3OutputStream::getBufferLockedCommon(ANativeWindowBuffer** anb, int* fenceFd) {ATRACE_HFR_CALL();status_t res;if ((res = getBufferPreconditionCheckLocked()) != OK) {return res;}bool gotBufferFromManager = false;if (mUseBufferManager) {sp<GraphicBuffer> gb;res = mBufferManager->getBufferForStream(getId(), getStreamSetId(), &gb, fenceFd);if (res == OK) {// Attach this buffer to the bufferQueue: the buffer will be in dequeue state after a// successful return.*anb = gb.get();res = mConsumer->attachBuffer(*anb);if (shouldLogError(res, mState)) {ALOGE("%s: Stream %d: Can't attach the output buffer to this surface: %s (%d)",__FUNCTION__, mId, strerror(-res), res);}if (res != OK) {checkRetAndSetAbandonedLocked(res);return res;}gotBufferFromManager = true;ALOGV("Stream %d: Attached new buffer", getId());} else if (res == ALREADY_EXISTS) {// Have sufficient free buffers already attached, can just// dequeue from buffer queueALOGV("Stream %d: Reusing attached buffer", getId());gotBufferFromManager = false;} else if (res != OK) {ALOGE("%s: Stream %d: Can't get next output buffer from buffer manager: %s (%d)",__FUNCTION__, mId, strerror(-res), res);return res;}}if (!gotBufferFromManager) {sp<ANativeWindow> currentConsumer = mConsumer;mLock.unlock();nsecs_t dequeueStart = systemTime(SYSTEM_TIME_MONOTONIC);res = currentConsumer->dequeueBuffer(currentConsumer.get(), anb, fenceFd);nsecs_t dequeueEnd = systemTime(SYSTEM_TIME_MONOTONIC);mDequeueBufferLatency.add(dequeueStart, dequeueEnd);mLock.lock();if (mUseBufferManager && res == TIMED_OUT) {checkRemovedBuffersLocked();sp<GraphicBuffer> gb;res = mBufferManager->getBufferForStream(getId(), getStreamSetId(), &gb, fenceFd, /*noFreeBuffer*/true);if (res == OK) {// Attach this buffer to the bufferQueue: the buffer will be in dequeue state after// a successful return.*anb = gb.get();res = mConsumer->attachBuffer(*anb);gotBufferFromManager = true;ALOGV("Stream %d: Attached new buffer", getId());if (res != OK) {if (shouldLogError(res, mState)) {ALOGE("%s: Stream %d: Can't attach the output buffer to this surface:"" %s (%d)", __FUNCTION__, mId, strerror(-res), res);}checkRetAndSetAbandonedLocked(res);return res;}} else {ALOGE("%s: Stream %d: Can't get next output buffer from buffer manager:"" %s (%d)", __FUNCTION__, mId, strerror(-res), res);return res;}} else if (res != OK) {if (shouldLogError(res, mState)) {ALOGE("%s: Stream %d: Can't dequeue next output buffer: %s (%d)",__FUNCTION__, mId, strerror(-res), res);}checkRetAndSetAbandonedLocked(res);return res;}}if (res == OK) {checkRemovedBuffersLocked();}return res;
}

getBufferLockedCommon function本文情况下运行，mUseBufferManager为FALSE，因为OutputBuffer的申请有Camera3BufferManager 申请和Anativewindow 申请两种情况，选择在snapdragon APP中设置：

• snapdragon APP中调用OutputConfiguration(@NonNull Surface surface)函数，surfaceGroupId默认赋值SURFACE_GROUP_ID_NONE(-1)，mUseBufferManager等于FALSE不使用buffer manager分配buffer。
• 直接使用ANativeWindow分配preview buffer，运行 if (!gotBufferFromManager) 逻辑
  
  Fig5 Camera Service dequeue buffer

Graphic buffer is transmit by binder to surfaceflinger process

Camera service获取到camera provider填充完成的image buffer之后要传递给display module显示经过Fig6、Fig7函数调用过程。
在cameraserver process中由surface module发起（后面会解释为什么是surface作为发起人），由hook_queueBuffer调用到surface，通过binder transact and ontransact结构，跨进程把从ANativewindow中申请的graphic buffer传输给surfaceflinger process，surfaceflinger再把buffer 传输到BufferQueue结构中进行管理，等待surfaceflinger process acquire use。

• 补充说明：mGraphicBufferProducer 是在 Surface 构造器中初始化的。它实际指向一个 BpGraphicBufferProducer 对象。BnGraphicBufferProducer::onTransact 调用的queuebuffer，是class BnGraphicBufferProducer的子类class bufferqueueproducer中的queuebuffer function。实际在运行中都是子类实例化调用父类的函数。

Fig6 Graphic buffer is transmit by binder to surfaceflinger process

SurfaceView建立native surface为真正的producer，surfaceflinger通过layer建立的BufferQueueLayer为真正的consumer，BufferQueueLayer调用updatetexImage从BufferQueue中acquire buffer，bindTextureImageLocked function Bind the new buffer to the GL texture（添加纹理之后才送到GPU进行渲染）。

Fig 7 BufferQueueLayer consumer buffer

Display module 显示data image

以上内容介绍完cameraserver process到surfaceflinger process preview buffer 传输过程，最后介绍surfaceflinger 怎么把申请graphic buffer显示到display。
Preview UI的刷新是由VSYNC机制来驱动，VSYNC作为整个机制的心脏不停的刷新着preview UI，如Fig 8所示。

• 先看CPU渲染路径，在APP render use CPU，surface把绘画完成之后，通过lock function向BufferQueue申请AnativeWindow buffer，再通过unlockAndPost function放入（queuebuffer）BufferQueue结构中，VSYNC发送trigger 信号触发surfaceflinger process来consume。
• 再看复杂的GPU渲染路径，VSYNC core通过event message调用到BitTube，BitTube发送socket mSendFd message通知监听对象，surfaceflinger会通过mReceiveFd监听mSendFd message。surfaceflinger作为consumer从BufferQueue中acquire graphic buffer，对buffer添加texture纹理之后送到GPU（ondraw function process）进行渲染。结束渲染之后做两件事情，
  - Frist：Surface作为producer把buffer传输到BufferQueue结构中进行管理；
  - Second：GPU渲染完成之后会调用onFrameAvailable通知surfaceflinger有可用buffer，SurfaceFlinger再通过内部MessageQueue调 用requestNextVsync请求接收下一个VSYNC用于合成，下一个VSYNC到了之后回调MessageQueue的handleMessage函数，实际调到SurfaceFlinger的onMessageReceived函数，处理REFRESH message，最后FrameBufferSurface作为consumer从BufferQueue结构中acquire graphic buffer。

不管是CPU or GPU渲染，最后都是传输到HW Composer进行layer合成，GPU渲染是通过onFrameTarget传输到FrameBufferTarget layer，CPU渲染是传输到Overlay layer。HWC完成之后就set到display module显示。

• HWC是Android中进行窗口layer合成和显示的HAL层模块。HWC通常由显示设备制造商（OEM）完成，为surfaceflinger服务提供硬件支持。

Fig8 display module显示graphic buffer flow

SurfaceView module

以上基本上介绍完Fig2中主干流程，这一部分再介绍SurfaceView module如何与display module and surfaceflinger process建立联系。

-SurfaceView是一种特殊的View，可在子线程进行UI绘制，它具有独立于应用程序之外的surface，主要用来处理复杂，耗时的UI绘制，如视频播放，camera preview。

SurfaceView create createBufferQueueLayer

SurfaceView module通过updateSurface function创建出surfacecontrol。传递到WindowManagerService进程中对其进行具体的初始化。

protected void updateSurface() {
if (creating) {viewRoot.createBoundsSurface(mSubLayer);mSurfaceSession = new SurfaceSession();mDeferredDestroySurfaceControl = mSurfaceControl;updateOpaqueFlag();final String name = "SurfaceView - " + viewRoot.getTitle().toString();mSurfaceControl = new SurfaceControl.Builder(mSurfaceSession).setName(name).setOpaque((mSurfaceFlags & SurfaceControl.OPAQUE) != 0).setBufferSize(mSurfaceWidth, mSurfaceHeight).setFormat(mFormat).setParent(viewRoot.getSurfaceControl()).setFlags(mSurfaceFlags).build();mBackgroundControl = new SurfaceControl.Builder(mSurfaceSession).setName("Background for -" + name).setOpaque(true).setColorLayer().setParent(mSurfaceControl).build();} else if (mSurfaceControl == null) {return;}

通过SurfaceControl的内部Builder类来创建SurfaceControl对象的，并且将SurfaceControl的Name，BufferSize，Format，Flags等等一并保存到内部Builder中

public SurfaceControl build() {if (mWidth < 0 || mHeight < 0) {throw new IllegalStateException("width and height must be positive or unset");}if ((mWidth > 0 || mHeight > 0) && (isEffectLayer() || isContainerLayer())) {throw new IllegalStateException("Only buffer layers can set a valid buffer size.");}return new SurfaceControl(mSession, mName, mWidth, mHeight, mFormat, mFlags, mParent, mMetadata,mLocalOwnerView, mCallsite);
}

SurfaceControl构造函数中调用SurfaceComposerClient::mClient的createSurface函数，mClient是ISurfaceComposerClient的Binder Bp端，Bn端则是SurfaceFlinger进程中的Client对象，createSurface具体实现是在surfaceflinger。

SurfaceComposerClient::createSurfaceCheckederr = mClient->createSurface(name, w, h, format, flags, parentHandle, std::move(metadata),&handle, &gbp, &transformHint);*outSurface = new SurfaceControl(this, handle, gbp, transformHint);status_t Client::createSurface(const String8& name, uint32_t w, uint32_t h, PixelFormat format,uint32_t flags, const sp<IBinder>& parentHandle,LayerMetadata metadata, sp<IBinder>* handle,sp<IGraphicBufferProducer>* gbp, uint32_t* outTransformHint) {// We rely on createLayer to check permissions.return mFlinger->createLayer(name, this, w, h, format, flags, std::move(metadata), handle, gbp,parentHandle, nullptr, outTransformHint);}SurfaceFlinger::createLayerresult = createBufferQueueLayer(client, std::move(uniqueName), w, h, flags, std::move(metadata), format, handle, gbp, &layer);

Create native surface

上面SurfaceFlinger进程的Layer创建好之后，接着通过new创建Surface，参数handle持有创建的Layer的引用。

static jlong nativeGetFromSurfaceControl(JNIEnv* env, jclass clazz,jlong nativeObject,jlong surfaceControlNativeObj) {Surface* self(reinterpret_cast<Surface *>(nativeObject));sp<SurfaceControl> ctrl(reinterpret_cast<SurfaceControl *>(surfaceControlNativeObj));sp<Surface> surface(ctrl->getSurface());if (surface != NULL) {surface->incStrong(&sRefBaseOwner);}return reinterpret_cast<jlong>(surface.get());
}

首次创建Surface，调用SurfaceControl的getSurface函数创建Surface。

sp<Surface> SurfaceControl::getSurface() const
{Mutex::Autolock _l(mLock);if (mSurfaceData == nullptr) {return generateSurfaceLocked();}return mSurfaceData;
}
sp<Surface> SurfaceControl::generateSurfaceLocked() const
{// This surface is always consumed by SurfaceFlinger, so the// producerControlledByApp value doesn't matter; using false.mSurfaceData = new Surface(mGraphicBufferProducer, false);return mSurfaceData;
}

native层SurfaceControl创建好了之后就可以通过此对象创建native层的Surface对象，最后将native层Surface指针保存到java层Surface。

Set preview UI Window

Surfaceview onAttachedToWindow function中先调用父类view onAttachedToWindow。因为surfaceview surface window默认在GUI window下层，如果需要能够在display展示给用户观看，需要设置透明区域（The surface is Z ordered so that it is behind the window holding its SurfaceView; the SurfaceView punches a hole in its window to allow its surface to be displayed.）使用requestTransparentRegion function实现。

  protected void onAttachedToWindow() {super.onAttachedToWindow();getViewRootImpl().addSurfaceChangedCallback(this);mWindowStopped = false;mViewVisibility = getVisibility() == VISIBLE;updateRequestedVisibility();mAttachedToWindow = true;mParent.requestTransparentRegion(SurfaceView.this);if (!mGlobalListenersAdded) {ViewTreeObserver observer = getViewTreeObserver();observer.addOnScrollChangedListener(mScrollChangedListener);observer.addOnPreDrawListener(mDrawListener);mGlobalListenersAdded = true;}}

Complement

• Fig4流程逻辑上很好理解，但是会存在一些疑问，为什么camera service and surface会有一个dequeue、queue buffer的逻辑？buffer circulation结构中acquire、release接口没有使用吗？Surface 中的buffer哪里来的？
• 完整buffer流程如Fig 9，Surfaceflinger create BufferQueueLayer，BufferQueueLayer create BufferQueue construction，BufferQueue再通过binder传输buffer给Surface，因为Surface and camera service都在cameraservice process，所以可以直接传输buffer，camera service获取到buffer之后会通过HIDL interface 让camera HAL填充image data，最后还给BufferQueue。
• 生产消费者模型只是BufferQueue自身的circulation，让surfaceflinger更好的管理buffer对象。
• 整个流程图仅限surfaceview使用流程，其他view很可能不同
  
  Fig 9，buffer完整流动模型

总结

• 这篇博客主要详细讲述从camera获取image到display显示的整个code flow，主要是也是想了解端到端的过程，对自己以后想做的双摄显示项目做一个基础。后来和头聊天了解到如果想要把camera module做好还需要了解display module and gallery module，东西真的挺多的，最近在做一个算法上移的工作，希望以后有机会和大家分享。
• 如果有什么错误的地方也请大家指出留言，觉得好点个赞，互粉一下，相互交流，谢谢大家！

参考文档

SnapdragonCamera源码分析（三）createSessions & startPreview
深入浅出CameraServer的Buffer管理机制
Android 源码 Camera2 预览流程分析二
AndroidQ 图形系统（11）UI刷新，SurfaceFlinger，Vsync机制总结
AndroidQ 图形系统（1）Surface与SurfaceControl创建分析