android surfaceflinger研究----Surface机制
转自： http://blog.csdn.net/windskier/article/details/7041610
前一篇文章介绍了android的显示系统，这篇文章中，我们把视角往上层移动一下，研究一下framework是如何与surfaceflinger进行业务交互的。
1）如何创建surface
2）如何显示窗口等等
所有的这一切都是通过系统服务WindowManagerService与surfaceflinger来进行的。
Android中的Surface机制这一块代码写的比较难理解，光叫Surface的类就有3个，因此本篇文章从两部分来分析。
首先，想要理解Surface机制，还是需要首先理清各个类之间的关系。
其次，在理解了整个Surface机制的类关系之后，到时我们再结合前一篇文章中对显示系统的介绍，研究一下一个Surface是如何和显示系统建立起联系来的，这个联系主要是指Surface的显示buffer的存储管理。
在下篇文章中，再分析SurfaceFlinger是如何将已经存储了窗口图形数据的Surface Buffer显示到显示系统中。
1. Surface机制的静态关系
将这一部分叫做Surface机制，是有别于SurfaceFlinger而言的，android的图形系统中，作为C/S模型两端的WMS和SurfaceFlinger是图形系统业务的核心，但是不把WMS和SurfaceFlinger中间的这层联系搞清楚的话，是很难理解整个图形系统的，在本文中我将两者之间的这个联系关系称之为Surface机制，它的主要任务就是创建一个Surface，ViewRoot在这个Surface上描绘当前的窗口，SurfaceFlinger将这个Surface flinger(扔)给显示系统将其呈现在硬件设备上。其实这里这个Surface在不同的模块中是以不同的形态存在的，唯一不变的就是其对应的显示Buffer。
1.1 ViewRoot和WMS共享Surface
我们知道每个Activity都会有一个ViewRootImpl作为Activity Window与WMS交互的接口，ViewRootImpl会绘制整个Activity的窗口View到Surface上，因此我们在ViewRootImpl中就有了创建Surface的需求。看一下代码中的Surface的定义：
ViewRootImpl.java
// These can be accessed by any thread, must be protected with a lock.
// Surface can never be reassigned or cleared (use Surface.clear()).
private final Surface mSurface = new Surface();
Surface(SurfaceTexture surfaceTexture)@Surface.java
/**
* Create Surface from a {@link SurfaceTexture}.
*
* Images drawn to the Surface will be made available to the {@link
* SurfaceTexture}, which can attach them an OpenGL ES texture via {@link
* SurfaceTexture#updateTexImage}.
*
* @param surfaceTexture The {@link SurfaceTexture} that is updated by this
* Surface.
*/
public Surface(SurfaceTexture surfaceTexture) {
if (DEBUG_RELEASE) {
mCreationStack = new Exception();
}
mCanvas = new CompatibleCanvas();
initFromSurfaceTexture(surfaceTexture);
}
由上面可以看出在ViewRootImpl中定义的Surface只是一个空壳，那么真正的Surface是在哪里被初始化的呢？大管家WMS中！当ViewRootImpl请求WMS relayout时，会将ViewSurface中的Surface交给WMS初始化。在WMS中，对应每个WindowState对象，在relayout窗口时，同样会创建一个Surface，wms中的这个Surface会真正的初始化，然后再将这个WMS Surface复制给ViewRootImpl中的Surface。这么实现的目的就是保证ViewRootImpl和WMS共享同一个Surface。ViewRootImpl对Surface进行绘制，WMS对这个Surface进行初始化及管理。很和谐！
private int relayoutWindow(WindowManager.LayoutParams params, int viewVisibility,
boolean insetsPending) throws RemoteException {
...
int relayoutResult = sWindowSession.relayout(
mWindow, mSeq, params,
(int) (mView.getMeasuredWidth() * appScale + 0.5f),
(int) (mView.getMeasuredHeight() * appScale + 0.5f),
viewVisibility, insetsPending, mWinFrame,
mPendingContentInsets, mPendingVisibleInsets,
mPendingConfiguration, mSurface);
//Log.d(TAG, "<<<<<< BACK FROM relayout");
if (restore) {
params.restore();
}
if (mTranslator != null) {
mTranslator.translateRectInScreenToAppWinFrame(mWinFrame);
mTranslator.translateRectInScreenToAppWindow(mPendingContentInsets);
mTranslator.translateRectInScreenToAppWindow(mPendingVisibleInsets);
}
return relayoutResult;
}
relayoutWindow()@WindowManagerService.java
public int relayoutWindow(Session session, IWindow client, int seq,
WindowManager.LayoutParams attrs, int requestedWidth,
int requestedHeight, int viewVisibility, boolean insetsPending,
Rect outFrame, Rect outContentInsets, Rect outVisibleInsets,
Configuration outConfig, Surface outSurface) {
boolean displayed = false;
boolean inTouchMode;
boolean configChanged;
// if they don't have this permission, mask out the status bar bits
synchronized(mWindowMap) {
WindowState win = windowForClientLocked(session, client, false);
if (viewVisibility == View.VISIBLE &&
(win.mAppToken == null || !win.mAppToken.clientHidden)) {
displayed = !win.isVisibleLw();
...
try {
Surface surface = win.createSurfaceLocked();
if (surface != null) {
outSurface.copyFrom(surface);
win.mReportDestroySurface = false;
win.mSurfacePendingDestroy = false;
if (SHOW_TRANSACTIONS) Slog.i(TAG,
" OUT SURFACE " + outSurface + ": copied");
} else {
// For some reason there isn't a surface. Clear the
// caller's object so they see the same state.
outSurface.release();
}
} catch (Exception e) {
mInputMonitor.updateInputWindowsLw(true /*force*/);
Slog.w(TAG, "Exception thrown when creating surface for client "
+ client + " (" + win.mAttrs.getTitle() + ")",
e);
Binder.restoreCallingIdentity(origId);
return 0;
}
...
}
}
1.2 SurfaceSession
SurfaceSession可以认为是创建Surface过程中，WMS和SurfaceFlinger之间的会话层，通过这个SurfaceSession实现了Surface的创建。
一个SurfaceSession表示一个到SurfaceFlinger的连接，通过它可以创建一个或多个Surface实例。
SurfaceSession是JAVA层的概念，@SurfaceSession.java。它对应的native实体是一个SurfaceComposerClient对象。
SurfaceComposerClient通过ComposerService类来获得SurfaceFlinger的IBinder接口ISurfaceComposer，但是光获得SurfaceFlinger的IBinder接口是不够的，要想请求SurfaceFlinger创建一个Surface，还需要向SurfaceFlinger获得一个IBinder接口ISurfaceComposerClient，通过这个ISurfaceComposerClient来请求SurfaceFlinger创建一个Surface，为什么这么绕呢，为什么不直接让SurfaceFlinger创建Surface呢？
站在SurfaceFlinger的角度来考虑，对于SurfaceFlinger来说，可能有多个Client来请求SurfaceFlinger的业务，每个Client可能会请求SurfaceFlinger创建多个Surface，那么SurfaceFlinger本地需要提供一套机制来保存每个client请求创建的Surface，SurfaceFlinger通过为每个client创建一个Client对象实现这个机制，并将这个Client的IBinder接口ISurfaceComposerClient返给SurfaceComposerClient对象。SurfaceComposerClient对象在通过ISurfaceComposerClient去请求创建Surface（ISurfaceComposerClient的功能只有创建和销毁Surface）。
@SurfaceFlinger.h
class Client : public BnSurfaceComposerClient
class SurfaceFlinger :
public BinderService<SurfaceFlinger>,
public BnSurfaceComposer,
public IBinder::DeathRecipient,
protected Thread
@SurfaceComposerClient.cpp
void SurfaceComposerClient::onFirstRef() {
sp<ISurfaceComposer> sm(getComposerService());
if (sm != 0) {
sp<ISurfaceComposerClient> conn = sm->createConnection();
if (conn != 0) {
mClient = conn;
mStatus = NO_ERROR;
}
}
}
在SurfaceFlinger中对应的执行代码为：
sp<ISurfaceComposerClient> SurfaceFlinger::createConnection()
{
sp<ISurfaceComposerClient> bclient;
sp<Client> client(new Client(this));
status_t err = client->initCheck();
if (err == NO_ERROR) {
bclient = client;
}
return bclient;
}
下图描述了整个SurfaceSession的内部结构与工作流程。
蓝色箭头是SurfaceComposerClient通过ComposerService获得SurfaceFlinger的IBinder接口ISurfaceComposer过程；
红色箭头表示SurfaceComposerClient通过IPC请求SurfaceFlinger创建Client的过程，并获得Client的IBinder接口ISurfaceComposerClient；
绿色箭头表示SurfaceComposerClient通过IPC请求Client创建Surface。
1.3 Surface的形态
上一节我们分析了SurfaceSession的静态结构，得知Surface的创建过程是通过SurfaceSession这个中间会话层去请求SurfaceFlinger去创建的，并且这篇文章中，我们说了半天Surface了，那么究竟我们要创建的Surface究竟是什么样的一个东西呢，它的具体形态是什么呢？这一小节我们就来分析以下Surface的形态。
1.3.1 client端Surface的形态
首先，我们看一下Surface在WMS中定义的代码
createSurfaceLocked()@WindowState.java
Surface createSurfaceLocked() {
if (mSurface == null) {
...
try {
final boolean isHwAccelerated = (mAttrs.flags &
WindowManager.LayoutParams.FLAG_HARDWARE_ACCELERATED) != 0;
final int format = isHwAccelerated ? PixelFormat.TRANSLUCENT : mAttrs.format;
if (!PixelFormat.formatHasAlpha(mAttrs.format)) {
flags |= Surface.OPAQUE;
}
mSurface = new Surface(
mSession.mSurfaceSession, mSession.mPid,
mAttrs.getTitle().toString(),
0, w, h, format, flags);
if (SHOW_TRANSACTIONS || SHOW_SURFACE_ALLOC) Slog.i(WindowManagerService.TAG,
" CREATE SURFACE "
+ mSurface + " IN SESSION "
+ mSession.mSurfaceSession
+ ": pid=" + mSession.mPid + " format="
+ mAttrs.format + " flags=0x"
+ Integer.toHexString(flags)
+ " / " + this);
}
...
}
return mSurface;
}
我们可以看到，它将SurfaceSession对象当作参数传递给了Surface的构造函数。往下看Surface的构造函数。
@Surface.java
/** create a surface with a name @hide */
public Surface(SurfaceSession s,
int pid, String name, int display, int w, int h, int format, int flags)
throws OutOfResourcesException {
if (DEBUG_RELEASE) {
mCreationStack = new Exception();
}
mCanvas = new CompatibleCanvas();
init(s,pid,name,display,w,h,format,flags);
mName = name;
}
这个构造函数，不同于我们在ViewRootImpl中看到的Surface的构造函数，这个构造函数并不是一个空壳，它做了本地实体的初始化工作，因此这个Surface才是一个真正的Suface。
Native函数init(Surface_init@android_view_Surface.cpp)会调到SurfaceComposerClient::createSurface，再向下的过程在上一节的图中描述很清楚了，在此不作介绍了。同时，先不管SurfaceFlinger为SurfaceComposerClient创建的Surface到底是一个什么东西，我们先看看SurfaceComposerClient为WMS创建的是一个什么东西？
@SurfaceComposerClient.cpp
sp<SurfaceControl> SurfaceComposerClient::createSurface(
const String8& name,
DisplayID display,
uint32_t w,
uint32_t h,
PixelFormat format,
uint32_t flags)
{
sp<SurfaceControl> result;
if (mStatus == NO_ERROR) {
ISurfaceComposerClient::surface_data_t data;
sp<ISurface> surface = mClient->createSurface(&data, name, //mClient为BpSurfaceComposerClient
display, w, h, format, flags);
if (surface != 0) {
result = new SurfaceControl(this, surface, data);
}
}
return result;
}
从上面的代码我们可以看出，SurfaceComposerClient为WMS返回的是一个SurfaceControl对象，这个SurfaceControl对象包含了surfaceFlinger为SurfaceComposerClient创建的surface,这个surfaceFlinge创建的Surface在Client端的形态为ISurface。这个过程下面分析SurfaceFlinger端的Surface形态时会看到。
SurfaceControl类中还有一个非常重要的成员mSurfaceData<class Surface : public SurfaceTextureClient>，它的类型也叫做Surface,定义在frameworks/base/include/surfaceflinger/surface.h。这个Surface提供了显示Buffer的管理。在文章的后面再介绍。
@frameworks/base/libs/gui/Surface.cpp
sp<Surface> SurfaceControl::getSurface() const
{
Mutex::Autolock _l(mLock);
if (mSurfaceData == 0) {
sp<SurfaceControl> surface_control(const_cast<SurfaceControl*>(this));
mSurfaceData = new Surface(surface_control);
}
return mSurfaceData;
}
1.3.2 SurfaceFlinger端Surface形态
sp<ISurface> SurfaceFlinger::createSurface(
ISurfaceComposerClient::surface_data_t* params,
const String8& name,
const sp<Client>& client,
DisplayID d, uint32_t w, uint32_t h, PixelFormat format,
uint32_t flags)
{
sp<LayerBaseClient> layer;
sp<ISurface> surfaceHandle;
if (int32_t(w|h) < 0) {
LOGE("createSurface() failed, w or h is negative (w=%d, h=%d)",
int(w), int(h));
return surfaceHandle;
}
//LOGD("createSurface for pid %d (%d x %d)", pid, w, h);
sp<Layer> normalLayer;
switch (flags & eFXSurfaceMask) {
case eFXSurfaceNormal:
normalLayer = createNormalSurface(client, d, w, h, flags, format);
layer = normalLayer;
break;
case eFXSurfaceBlur:
// for now we treat Blur as Dim, until we can implement it
// efficiently.
case eFXSurfaceDim:
layer = createDimSurface(client, d, w, h, flags);
break;
case eFXSurfaceScreenshot:
layer = createScreenshotSurface(client, d, w, h, flags);
break;
}
if (layer != 0) {
layer->initStates(w, h, flags);
layer->setName(name);
ssize_t token = addClientLayer(client, layer);
surfaceHandle = layer->getSurface();
if (surfaceHandle != 0) {
params->token = token;
params->identity = layer->getIdentity();
if (normalLayer != 0) {
Mutex::Autolock _l(mStateLock);
mLayerMap.add(layer->getSurfaceBinder(), normalLayer);
}
}
setTransactionFlags(eTransactionNeeded);
}
return surfaceHandle;
}
当client请求SurfaceFlinger创建Surface时，SurfaceFlinger首先根据WMS提供的窗口的属性来一个命名为Layer概念的对象，然后再根据Layer创建它的子类对象LayerBaseClient::BSurface。此时第三个名为Surface(BSurface)类出现了，下一节我们来介绍一下这个Layer的概念。
1.4 Layer
1.4.1 Layer的分类
目前，android4.0中有3中Layer类型，如上图所示。
1. Layer, 普通的Layer，它为每个Client端请求的Surface创建显示Buffer。
2. LayerDim，这种Layer不会创建显示Buffer，它只是将通过这个Layer将原来FrameBuffer上的数据进行暗淡处理；
3. LayerScreenshot，这种Layer不会创建显示Buffer，它只是将通过这个Layer将原来FrameBuffer上的数据进行抓取；
从这些Layer看出，我们分析的重点就是第一种Layer，下面我们着重分析一下普通的Layer。Layer的具体业务我们在下一篇文章中分析
1.4.2 Layer的管理
上文我们在分析SurfaceSession的时候，也分析过，一个Client可能会创建多个Surface，也就是要创建多个Layer，那么SurfaceFlinger端如何管理多个Layer呢？SurfaceFlinger维护了3个Vector来管理Layer<Client: DefaultKeyedVector< size_t, wp<LayerBaseClient> > mLayers;SurfaceFlinger:LayerVector layersSortedByZ;>。
第一种方式，我们知道SurfaceFlinger会为每个SurfaceSession创建一个Client对象，这第一种方式就是将所有为某一个SurfacSession创建的Layer保存在它对应的Client对象中。
SurfaceFlinger::createSurface()@SurfaceFlinger.cpp，调用路径如下：
SurfaceFlinger::addClientLayer->
client->attachLayer(lbc)->
mLayers.add(name, layer);
第二种方式，在SurfaceFlinger中以排序的方式保存下来。其调用路径如下：
SurfaceFlinger::addClientLayer->
SurfaceFlinger::addLayer_l->
mCurrentState.layersSortedByZ.add(layer)
第三种方式，将所有的创建的普通的Layer保存起来，以便Client Surface在请求实现Buffer时能够辨识Client Surface对应的Layer。
SurfaceFlinger::createSurface()@SurfaceFlinger.cpp
surfaceHandle = layer->getSurface();
if (surfaceHandle != 0) {
params->token = token;
params->identity = layer->getIdentity();
if (normalLayer != 0) {
Mutex::Autolock _l(mStateLock);
mLayerMap.add(layer->getSurfaceBinder(), normalLayer);
}
}
2. Surface 显示Buffer的存储管理
在前文介绍Client端的Surface形态的内容时，我们提到SurfaceControl中还会维护一个名为Surface对象，它定义在 frameworks/base/libs/surfaceflinger/Surface.h中，它负责向LayerBaseClient::BSurface请求显示Buffer，同时将显示Buffer交给JAVA Surface的Canvas去绘制窗口，我们称这个Surface为Client Surface。
2.1 窗口绘制
draw()@ViewRoot.java
Canvas canvas;
try {
int left = dirty.left;
int top = dirty.top;
int right = dirty.right;
int bottom = dirty.bottom;
final long lockCanvasStartTime;
if (ViewDebug.DEBUG_LATENCY) {
lockCanvasStartTime = System.nanoTime();
}
canvas = surface.lockCanvas(dirty);
if (ViewDebug.DEBUG_LATENCY) {
long now = System.nanoTime();
Log.d(TAG, "Latency: Spent "
+ ((now - lockCanvasStartTime) * 0.000001f)
+ "ms waiting for surface.lockCanvas()");
}
if (left != dirty.left || top != dirty.top || right != dirty.right ||
bottom != dirty.bottom) {
mAttachInfo.mIgnoreDirtyState = true;
}
// TODO: Do this in native
canvas.setDensity(mDensity);
上面的代码显示，JAVA Surface 会lock canvas。而Client Surface的创建就在这个过程中，即下面代码中的第一行getSurface().我们先不管Client Surface的创建，先看看Canvas是如何与Client Surface的显示Buffer关联的。
Surface_lockCanvas@android_view_Surface.cpp
static jobject Surface_lockCanvas(JNIEnv* env, jobject clazz, jobject dirtyRect)
{
const sp<Surface>& surface(getSurface(env, clazz));
if (!Surface::isValid(surface)) {
doThrowIAE(env);
return 0;
}
// get dirty region
Region dirtyRegion;
if (dirtyRect) {
Rect dirty;
dirty.left = env->GetIntField(dirtyRect, ro.l);
dirty.top = env->GetIntField(dirtyRect, ro.t);
dirty.right = env->GetIntField(dirtyRect, ro.r);
dirty.bottom= env->GetIntField(dirtyRect, ro.b);
if (!dirty.isEmpty()) {
dirtyRegion.set(dirty);
}
} else {
dirtyRegion.set(Rect(0x3FFF,0x3FFF));
}
Surface::SurfaceInfo info;
status_t err = surface->lock(&info, &dirtyRegion);
if (err < 0) {
const char* const exception = (err == NO_MEMORY) ?
OutOfResourcesException :
"java/lang/IllegalArgumentException";
jniThrowException(env, exception, NULL);
return 0;
}
// Associate a SkCanvas object to this surface
jobject canvas = env->GetObjectField(clazz, so.canvas);
env->SetIntField(canvas, co.surfaceFormat, info.format);
SkCanvas* nativeCanvas = (SkCanvas*)env->GetIntField(canvas, no.native_canvas);
SkBitmap bitmap;
ssize_t bpr = info.s * bytesPerPixel(info.format);
bitmap.setConfig(convertPixelFormat(info.format), info.w, info.h, bpr);
if (info.format == PIXEL_FORMAT_RGBX_8888) {
bitmap.setIsOpaque(true);
}
if (info.w > 0 && info.h > 0) {
bitmap.setPixels(info.bits);
} else {
// be safe with an empty bitmap.
bitmap.setPixels(NULL);
}
nativeCanvas->setBitmapDevice(bitmap);
SkRegion clipReg;
if (dirtyRegion.isRect()) { // very common case
const Rect b(dirtyRegion.getBounds());
clipReg.setRect(b.left, b.top, b.right, b.bottom);
} else {
size_t count;
Rect const* r = dirtyRegion.getArray(&count);
while (count) {
clipReg.op(r->left, r->top, r->right, r->bottom, SkRegion::kUnion_Op);
r++, count--;
}
}
nativeCanvas->clipRegion(clipReg);
int saveCount = nativeCanvas->save();
env->SetIntField(clazz, so.saveCount, saveCount);
if (dirtyRect) {
const Rect& bounds(dirtyRegion.getBounds());
env->SetIntField(dirtyRect, ro.l, bounds.left);
env->SetIntField(dirtyRect, ro.t, bounds.top);
env->SetIntField(dirtyRect, ro.r, bounds.right);
env->SetIntField(dirtyRect, ro.b, bounds.bottom);
}
return canvas;
}
上面的代码，我们可以看出，Canvas的Bitmap设备设置了Client Surface的显示Buffer为其Bitmap pixel存储空间。
bitmap.setPixels(info.bits);
这样Canvas的绘制空间就有了。下一步就该绘制窗口了。
draw()@ViewRootImpl.java
try {
canvas.translate(0, -yoff);
if (mTranslator != null) {
mTranslator.translateCanvas(canvas);
}
canvas.setScreenDensity(scalingRequired
? DisplayMetrics.DENSITY_DEVICE : 0);
mAttachInfo.mSetIgnoreDirtyState = false;
mView.draw(canvas);
}
其中ViewRootImpl中的mView为整个窗口的View。它将Render自己和所有它的子View。
2.2 Client Surface的初始化(SurfaceControl中的Surface)
Client Surface的创建是从ViewRootImpl首次Lock canvas时进行的，这么做的目的可能也是为了节约空间，减少不必要的开支。
Client Surface的初始化和显示Buffer的管理过程比较复杂，下图给出了这一部分的一个静态结构图，有些东西从图上表现不出来，下面我简单的介绍一下。
