Android 显示系统:SurfaceFlinger详解
一、Android系统启动
Android设备从按下开机键到桌面显示画面,大致过程如下图流程:
开机显示桌面、从桌面点击 App 图标到 Activity显示在屏幕上的过程又是怎样的呢?下面介绍Android系统中的“画家” - SurfaceFlinger.
SurfaceFlinger 启动过程:
二、SurfaceFlinger代码剖析[Android 11]
代码路径:/frameworks/native/services/surfaceflinger/
SurfaceFlinger二进制分成surfaceflinger可执行文件(main入口)和libsurfaceflinger.so库文件(功能实现),由main_surfaceflinger.cpp文件编译而成,Android.bp代码模块编译配置如下:
1.【执行文件-surfaceflinger】
...
cc_binary { name: "surfaceflinger", defaults: ["surfaceflinger_defaults"], init_rc: ["surfaceflinger.rc"], srcs: ["main_surfaceflinger.cpp"], whole_static_libs: [ "libsigchain", ], shared_libs: [ "android.frameworks.displayservice@1.0", "android.hardware.configstore-utils", "android.hardware.configstore@1.0", "android.hardware.graphics.allocator@2.0", "libbinder", "libcutils", "libdisplayservicehidl", "libhidlbase", "libhidltransport", "liblayers_proto", "liblog", "libsurfaceflinger", "libtimestats_proto", "libutils", ], static_libs: [ "libserviceutils", "libtrace_proto", ], ldflags: ["-Wl,--export-dynamic"], // TODO(b/71715793): These version-scripts are required due to the use of // whole_static_libs to pull in libsigchain. To work, the files had to be // locally duplicated from their original location // $ANDROID_ROOT/art/sigchainlib/ multilib: { lib32: { version_script: "version-script32.txt", }, lib64: { version_script: "version-script64.txt", }, }, }
...
SurfaceFlinger可执行二进制文件surfaceflinger由main_surfaceflinger.cpp文件独立编译而成,主要负责搭建进程启动环境:
int main(int, char**) { signal(SIGPIPE, SIG_IGN); // 从8.0开始,Android提供了hidl机制,将原先直接由JNI->Native->HAL的接口调用形式,统一规范成hidl service/client交互形式。 // 该方式从一方面规范和统一了Android Framework和HAL的调用机制,但实际从项目维度,这种调用方式对性能上开销,将比直接调用的方式要花费更多的时间。 hardware::configureRpcThreadpool(1 /* maxThreads */, false /* callerWillJoin */); startGraphicsAllocatorService(); // When SF is launched in its own process, limit the number of // binder threads to 4. ProcessState::self()->setThreadPoolMaxThreadCount(4); // start the thread pool sp<ProcessState> ps(ProcessState::self()); ps->startThreadPool(); // 创建SurfaceFlinger对象,由强指针指向。 // SurfaceFlinger继承RefBase类,所以此处一旦new出对象赋给sp指针后,将立刻出发SurfaceFlinger类的onFirstRef方法的调用。 // instantiate surfaceflinger sp<SurfaceFlinger> flinger = surfaceflinger::createSurfaceFlinger(); setpriority(PRIO_PROCESS, 0, PRIORITY_URGENT_DISPLAY); set_sched_policy(0, SP_FOREGROUND); // Put most SurfaceFlinger threads in the system-background cpuset // Keeps us from unnecessarily using big cores // Do this after the binder thread pool init if (cpusets_enabled()) set_cpuset_policy(0, SP_SYSTEM); // SurfaceFlinger类正式初始化 // initialize before clients can connect flinger->init(); // SurfaceFlinger向ServiceManager注册Binder服务, // 这样在其他进程中可以通过getService+SERVICE_NAME来获取SurfaceFlinger服务,继而可以和SurfaceFlinger类进行Binder通信。 // publish surface flinger sp<IServiceManager> sm(defaultServiceManager()); sm->addService(String16(SurfaceFlinger::getServiceName()), flinger, false, IServiceManager::DUMP_FLAG_PRIORITY_CRITICAL | IServiceManager::DUMP_FLAG_PROTO); //里面的new DisplayService()方法调用HIDL定义接口 Return<sp<IDisplayEventReceiver >> getEventReceiver() override; startDisplayService(); // dependency on SF getting registered above if (SurfaceFlinger::setSchedFifo(true) != NO_ERROR) { ALOGW("Couldn't set to SCHED_FIFO: %s", strerror(errno)); } // SurfaceFlinger类进入主循环(此处注意SurfaceFlinger类未继承Threads类,不遵循Threads类的接口执行顺序) // run surface flinger in this thread flinger->run(); return 0; }
HIDL接口介绍可以参考:https://source.android.google.cn/reference/hidl/
2.【动态库-libsurfaceflinger.so】
Android.bp代码模块编译配置如下:
...
cc_library_shared { name: "libsurfaceflinger", defaults: ["libsurfaceflinger_defaults"], cflags: [ "-fvisibility=hidden", "-Werror=format", "-DREDUCE_VIDEO_WORKLOAD", "-DUSE_AML_HW_ACTIVE_MODE", ], srcs: [ ":libsurfaceflinger_sources", ], logtags: ["EventLog/EventLogTags.logtags"], include_dirs: [ "frameworks/native/vulkan/vkjson", "frameworks/native/vulkan/include", "hardware/amlogic/gralloc/amlogic", "hardware/amlogic/hwcomposer/tvp", "hardware/amlogic/gralloc", ], static_libs: [ "libomxutils_static@2", "libamgralloc_ext_static@2", ], cppflags: [ "-fwhole-program-vtables", // requires ThinLTO ], lto: { thin: true, }, }
...
上面提到的createSurfaceFlinger()中会调用new SurfaceFlinger(),然后会执行到:onFirstRef():
void SurfaceFlinger::onFirstRef() { mEventQueue.init(this); }
onFirstRef() 中会创建 Handler 并初始化: /frameworks/native/services/surfaceflinger/Scheduler/MessageQueue.cpp
//MessageQueue.cpp void MessageQueue::init(const sp<SurfaceFlinger>& flinger) { mFlinger = flinger; mLooper = new Looper(true); mHandler = new Handler(*this); }
然后会执行到 SurfaceFlinger::init(),该方法主要功能是:
- 初始化 EGL
- 创建 HWComposer
- 初始化非虚拟显示屏
- 启动 EventThread 线程
- 启动开机动画
// Do not call property_set on main thread which will be blocked by init // Use StartPropertySetThread instead. void SurfaceFlinger::init() { ALOGI( "SurfaceFlinger's main thread ready to run. " "Initializing graphics H/W..."); Mutex::Autolock _l(mStateLock);
// 对于CompositionEngine 属性进行设置, 创建RenderEngine对象 // Get a RenderEngine for the given display / config (can't fail) // TODO(b/77156734): We need to stop casting and use HAL types when possible. // Sending maxFrameBufferAcquiredBuffers as the cache size is tightly tuned to single-display. mCompositionEngine->setRenderEngine(renderengine::RenderEngine::create( renderengine::RenderEngineCreationArgs::Builder() .setPixelFormat(static_cast<int32_t>(defaultCompositionPixelFormat)) .setImageCacheSize(maxFrameBufferAcquiredBuffers) .setUseColorManagerment(useColorManagement) .setEnableProtectedContext(enable_protected_contents(false)) .setPrecacheToneMapperShaderOnly(false) .setSupportsBackgroundBlur(mSupportsBlur) .setContextPriority(useContextPriority ? renderengine::RenderEngine::ContextPriority::HIGH : renderengine::RenderEngine::ContextPriority::MEDIUM) .build())); mCompositionEngine->setTimeStats(mTimeStats); LOG_ALWAYS_FATAL_IF(mVrFlingerRequestsDisplay, "Starting with vr flinger active is not currently supported.");
//创建HWComposer对象并传入一个name属性,再通过mCompositionEngine->setHwComposer设置对象属性。 mCompositionEngine->setHwComposer(getFactory().createHWComposer(getBE().mHwcServiceName)); mCompositionEngine->getHwComposer().setConfiguration(this, getBE().mComposerSequenceId);
//processDisplayHotplugEventsLocked(); 处理 任何初始热插拔和显示更改的结果
//在此方法中主要有调用 initScheduler(displayId); // Process any initial hotplug and resulting display changes. processDisplayHotplugEventsLocked(); const auto display = getDefaultDisplayDeviceLocked(); LOG_ALWAYS_FATAL_IF(!display, "Missing internal display after registering composer callback."); LOG_ALWAYS_FATAL_IF(!getHwComposer().isConnected(*display->getId()), "Internal display is disconnected."); if (useVrFlinger) { auto vrFlingerRequestDisplayCallback = [this](bool requestDisplay) { // This callback is called from the vr flinger dispatch thread. We // need to call signalTransaction(), which requires holding // mStateLock when we're not on the main thread. Acquiring // mStateLock from the vr flinger dispatch thread might trigger a // deadlock in surface flinger (see b/66916578), so post a message // to be handled on the main thread instead. static_cast<void>(schedule([=] { ALOGI("VR request display mode: requestDisplay=%d", requestDisplay); mVrFlingerRequestsDisplay = requestDisplay; signalTransaction(); })); }; mVrFlinger = dvr::VrFlinger::Create(getHwComposer().getComposer(), getHwComposer() .fromPhysicalDisplayId(*display->getId()) .value_or(0), vrFlingerRequestDisplayCallback); if (!mVrFlinger) { ALOGE("Failed to start vrflinger"); } } // initialize our drawing state mDrawingState = mCurrentState; // set initial conditions (e.g. unblank default device) initializeDisplays(); char primeShaderCache[PROPERTY_VALUE_MAX]; property_get("service.sf.prime_shader_cache", primeShaderCache, "1"); if (atoi(primeShaderCache)) { getRenderEngine().primeCache(); } // Inform native graphics APIs whether the present timestamp is supported: const bool presentFenceReliable = !getHwComposer().hasCapability(hal::Capability::PRESENT_FENCE_IS_NOT_RELIABLE); mStartPropertySetThread = getFactory().createStartPropertySetThread(presentFenceReliable); if (mStartPropertySetThread->Start() != NO_ERROR) { ALOGE("Run StartPropertySetThread failed!"); } ALOGV("Done initializing"); }
首先看下如何创建 HWComposer:frameworks/native/services/surfaceflinger/SurfaceFlingerDefaultFactory.cpp
//make_unique 相当于 new,(能够取代new 而且无需 delete pointer,有助于代码管理)。 std::unique_ptr<HWComposer> DefaultFactory::createHWComposer(const std::string& serviceName) { return std::make_unique<android::impl::HWComposer>(serviceName); }
对于CompositionEngine进行初始化:
std::unique_ptr<compositionengine::CompositionEngine> DefaultFactory::createCompositionEngine() { return compositionengine::impl::createCompositionEngine(); }
同样是通过make_unique创建了 CompositionEngine对象:
std::unique_ptr<compositionengine::CompositionEngine> createCompositionEngine() { return std::make_unique<CompositionEngine>(); }
再回到flinger->init()中 processDisplayHotplugEventsLocked(); 处理任何初始热插拔和显示更改的结果,在此方法中主要有调用 initScheduler(displayId):
void SurfaceFlinger::initScheduler(DisplayId primaryDisplayId) { if (mScheduler) { // In practice it's not allowed to hotplug in/out the primary display once it's been // connected during startup, but some tests do it, so just warn and return. ALOGW("Can't re-init scheduler"); return; } auto currentConfig = HwcConfigIndexType(getHwComposer().getActiveConfigIndex(primaryDisplayId)); mRefreshRateConfigs = std::make_unique<scheduler::RefreshRateConfigs>(getHwComposer().getConfigs( primaryDisplayId), currentConfig); mRefreshRateStats = std::make_unique<scheduler::RefreshRateStats>(*mRefreshRateConfigs, *mTimeStats, currentConfig, hal::PowerMode::OFF); mRefreshRateStats->setConfigMode(currentConfig); mPhaseConfiguration = getFactory().createPhaseConfiguration(*mRefreshRateConfigs);
// 处创建Scheduler对象 // start the EventThread mScheduler = getFactory().createScheduler([this](bool enabled) { setPrimaryVsyncEnabled(enabled); }, *mRefreshRateConfigs, *this); //创建app链接 和 sf链接
mAppConnectionHandle = mScheduler->createConnection("app", mPhaseConfiguration->getCurrentOffsets().late.app, impl::EventThread::InterceptVSyncsCallback()); mSfConnectionHandle = mScheduler->createConnection("sf", mPhaseConfiguration->getCurrentOffsets().late.sf, [this](nsecs_t timestamp) { mInterceptor->saveVSyncEvent(timestamp); }); mEventQueue->setEventConnection(mScheduler->getEventConnection(mSfConnectionHandle)); mVSyncModulator.emplace(*mScheduler, mAppConnectionHandle, mSfConnectionHandle, mPhaseConfiguration->getCurrentOffsets()); mRegionSamplingThread = new RegionSamplingThread(*this, *mScheduler, RegionSamplingThread::EnvironmentTimingTunables()); // Dispatch a config change request for the primary display on scheduler // initialization, so that the EventThreads always contain a reference to a // prior configuration. // // This is a bit hacky, but this avoids a back-pointer into the main SF // classes from EventThread, and there should be no run-time binder cost // anyway since there are no connected apps at this point. const nsecs_t vsyncPeriod = mRefreshRateConfigs->getRefreshRateFromConfigId(currentConfig).getVsyncPeriod(); mScheduler->onPrimaryDisplayConfigChanged(mAppConnectionHandle, primaryDisplayId.value, currentConfig, vsyncPeriod); }
详细看下app、sf的链接:
Scheduler::ConnectionHandle Scheduler::createConnection( const char* connectionName, nsecs_t phaseOffsetNs, impl::EventThread::InterceptVSyncsCallback interceptCallback) { auto vsyncSource = makePrimaryDispSyncSource(connectionName, phaseOffsetNs); auto eventThread = std::make_unique<impl::EventThread>(std::move(vsyncSource), std::move(interceptCallback)); return createConnection(std::move(eventThread)); }
可以看到创建了DispSyncSource对象,且构造方法传入了四个值,dispSync对象,phaseOffset偏移量,traceVsync为true,name就是 app或 sf
DispSyncSource::DispSyncSource(DispSync* dispSync, nsecs_t phaseOffset, bool traceVsync, const char* name) : mName(name), mValue(base::StringPrintf("VSYNC-%s", name), 0), //对mValue进行了赋值,systrace上我们看到的 VSYNC-app VSYNC-sf 标签就是它 mTraceVsync(traceVsync), //mTraceVsync为true,在onDispSyncEvent方法中 mVsyncOnLabel(base::StringPrintf("VsyncOn-%s", name)), mDispSync(dispSync), mPhaseOffset(base::StringPrintf("VsyncOffset-%s", name), phaseOffset) //对mPhaseOffset进行初始化 vsync信号到来时候,sf、app的偏移量
所以我们在systrace上面看到的 VSYNC-app/VSYNC-sf 驼峰 0 1变化,来源于这个。
创建EventThread对象,传入sf 或 app 相关联的vsyncSource对象:
auto eventThread = std::make_unique<impl::EventThread>(std::move(vsyncSource), std::move(interceptCallback));
说明:
1)每个 ConnectionHandle 对象里有个 id,作为 Scheduler 对象中 mConnections 属性(map<id, Connection>)的键值,Connection 对象中又包含 ConnectionHandle、EventThreadConnection、EventThread 3个属性。
2)mScheduler->getEventConnection(mSfConnectionHandle) 中,以 mSfConnectionHandle 的 id 为键值,在 Scheduler 的 mConnections(unordered_map<int64_t, Connection>)中找到对应的Connection,并返回其 EventThreadConnection 成员属性。
3)getHwComposer().registerCallback() 中,依次调用 HwComposer、Device 的 registerCallback() 方法,并在 Device 中 将 SurfaceFlinger 对象封装到 ComposerCallbackBridge 中;对于封装后的对象,依次调用 Composer、IComposerClient 的 registerCallback() 方法,注入到 IComposerClient 的实现类中。
相关问题:
① 屏幕刷新速率比系统帧速率快:
此时,在前缓冲区内容全部映射到屏幕上之后,后缓冲区尚未准备好下一帧,屏幕将无法读取下一帧,所以只能继续显示当前一帧的图形,造成一帧显示多次,也就是卡顿。
② 系统帧速率比屏幕刷新率快
此时,屏幕未完全把前缓冲区的一帧映射到屏幕,而系统已经在后缓冲区准备好了下一帧,并要求读取下一帧到屏幕,将会导致屏幕上半部分是上一帧的图形,而下半部分是下一帧的图形,造成屏幕上显示多帧,也就是屏幕撕裂。
为了解决上述问题,Android显示系统一般会有多级缓冲,即在屏幕刷新的同时在另外一个buffer准备下一帧数据,以此提高性能:
前缓冲区:用来显示内容到屏幕的帧缓冲区
后缓冲区:用于后台合成下一帧图形的帧缓冲区
垂直同步(VSync):当屏幕从缓冲区扫描完一帧到屏幕上之后,开始扫描下一帧之前,发出的一个同步信号,该信号用来切换前缓冲区和后缓冲区。
屏幕刷新率(HZ):代表屏幕在一秒内刷新屏幕的次数,Android手机一般为60HZ(也就是1秒刷新60帧,大约16.67毫秒刷新1帧)
系统帧速率(FPS):代表了系统在一秒内合成的帧数,该值的大小由系统算法和硬件决定。
3. 服务启动配置文件:/frameworks/native/services/surfaceflinger/surfaceflinger.rc
上面发现服务配置文件也在Android.mk中被加载:LOCAL_INIT_RC := surfaceflinger.rc
service surfaceflinger /system/bin/surfaceflinger
class core animation
user system
group graphics drmrpc readproc
onrestart restart zygote
writepid /dev/stune/foreground/tasks
socket pdx/system/vr/display/client stream 0666 system graphics u:object_r:pdx_display_client_endpoint_socket:s0
socket pdx/system/vr/display/manager stream 0666 system graphics u:object_r:pdx_display_manager_endpoint_socket:s0
socket pdx/system/vr/display/vsync stream 0666 system graphics u:object_r:pdx_display_vsync_endpoint_socket:s0
4. Surface 创建过程
Surface 创建的过程就是 Activity 显示的过程,在 ActivityThread.handleResumeActivity() 中调用了 Activity.makeVisible()具体实现:
void makeVisible() { if (!mWindowAdded) { ViewManager wm = getWindowManager();//此处 getWindowManager 获取的是 WindowManagerImpl 对象 wm.addView(mDecor, getWindow().getAttributes()); mWindowAdded = true; } mDecor.setVisibility(View.VISIBLE); }
WindowManagerImpl.java:
public void addView(@NonNull View view, @NonNull ViewGroup.LayoutParams params) { applyDefaultToken(params); mGlobal.addView(view, params, mDisplay, mParentWindow); }
WindowManagerGlobal.java:
public void addView(View view, ViewGroup.LayoutParams params, Display display, Window parentWindow) { ... final WindowManager.LayoutParams wparams = (WindowManager.LayoutParams) params; //创建 ViewRootImpl ViewRootImpl root = new ViewRootImpl(view.getContext(), display); view.setLayoutParams(wparams); mViews.add(view); mRoots.add(root); mParams.add(wparams); //设置 View root.setView(view, wparams, panelParentView); ... }
创建 ViewRootImpl:
public ViewRootImpl(Context context, Display display) { //获取 IWindowSession的代理类 this(context, display, WindowManagerGlobal.getWindowSession(), false /* useSfChoreographer */); }
WindowManagerGlobal.java:
@UnsupportedAppUsage public static IWindowSession getWindowSession() { synchronized (WindowManagerGlobal.class) { if (sWindowSession == null) { try { // Emulate the legacy behavior. The global instance of InputMethodManager // was instantiated here. // TODO(b/116157766): Remove this hack after cleaning up @UnsupportedAppUsage //获取 IMS 的代理类 InputMethodManager.ensureDefaultInstanceForDefaultDisplayIfNecessary(); IWindowManager windowManager = getWindowManagerService(); //经过 Binder 调用,最终调用 WMS sWindowSession = windowManager.openSession( new IWindowSessionCallback.Stub() { @Override public void onAnimatorScaleChanged(float scale) { ValueAnimator.setDurationScale(scale); } }); } catch (RemoteException e) { throw e.rethrowFromSystemServer(); } } return sWindowSession; } }
WindowManagerService.openSession:
// ------------------------------------------------------------- // IWindowManager API // ------------------------------------------------------------- @Override public IWindowSession openSession(IWindowSessionCallback callback) { //创建session对象 return new Session(this, callback); }
再次经过 Binder 将数据写回 app 进程,则获取的便是 Session 的代理对象 IWindowSession。
创建完 ViewRootImpl 对象后,接下来调用该对象的 setView() 方法:
ViewRootImpl:
public void setView(View view, WindowManager.LayoutParams attrs, View panelParentView) { synchronized (this) { requestLayout(); //详见下面分析 ... //通过 Binder调用,进入 system 进程的 Session
res = mWindowSession.addToDisplayAsUser(mWindow, mSeq, mWindowAttributes,
getHostVisibility(), mDisplay.getDisplayId(), userId, mTmpFrame,
mAttachInfo.mContentInsets, mAttachInfo.mStableInsets,
mAttachInfo.mDisplayCutout, inputChannel,
mTempInsets, mTempControls);
...
}
}
Session.java
@Override public int addToDisplayAsUser(IWindow window, int seq, WindowManager.LayoutParams attrs, int viewVisibility, int displayId, int userId, Rect outFrame, Rect outContentInsets, Rect outStableInsets, DisplayCutout.ParcelableWrapper outDisplayCutout, InputChannel outInputChannel, InsetsState outInsetsState, InsetsSourceControl[] outActiveControls) { //调用WMS的addWindow方法 return mService.addWindow(this, window, seq, attrs, viewVisibility, displayId, outFrame, outContentInsets, outStableInsets, outDisplayCutout, outInputChannel, outInsetsState, outActiveControls, userId); }
WindowManagerService.java:
public int addWindow(Session session, IWindow client, int seq, LayoutParams attrs, int viewVisibility, int displayId, Rect outFrame, Rect outContentInsets, Rect outStableInsets, DisplayCutout.ParcelableWrapper outDisplayCutout, InputChannel outInputChannel, InsetsState outInsetsState, InsetsSourceControl[] outActiveControls, int requestUserId) { Arrays.fill(outActiveControls, null); int[] appOp = new int[1]; final boolean isRoundedCornerOverlay = (attrs.privateFlags & PRIVATE_FLAG_IS_ROUNDED_CORNERS_OVERLAY) != 0; int res = mPolicy.checkAddPermission(attrs.type, isRoundedCornerOverlay, attrs.packageName, appOp); if (res != WindowManagerGlobal.ADD_OKAY) { return res; } WindowState parentWindow = null; final int callingUid = Binder.getCallingUid(); final int callingPid = Binder.getCallingPid(); final long origId = Binder.clearCallingIdentity(); final int type = attrs.type; synchronized (mGlobalLock) { if (!mDisplayReady) { throw new IllegalStateException("Display has not been initialialized"); } ... //创建 WindowState final WindowState win = new WindowState(this, session, client, token, parentWindow, appOp[0], seq, attrs, viewVisibility, session.mUid, userId, session.mCanAddInternalSystemWindow); if (win.mDeathRecipient == null) { // Client has apparently died, so there is no reason to // continue. ProtoLog.w(WM_ERROR, "Adding window client %s" + " that is dead, aborting.", client.asBinder()); return WindowManagerGlobal.ADD_APP_EXITING; } if (win.getDisplayContent() == null) { ProtoLog.w(WM_ERROR, "Adding window to Display that has been removed."); return WindowManagerGlobal.ADD_INVALID_DISPLAY; } // 调整 WindowManager的LayoutParams 参数 final DisplayPolicy displayPolicy = displayContent.getDisplayPolicy(); displayPolicy.adjustWindowParamsLw(win, win.mAttrs, callingPid, callingUid); res = displayPolicy.validateAddingWindowLw(attrs, callingPid, callingUid); if (res != WindowManagerGlobal.ADD_OKAY) { return res; } // 打开输入通道 final boolean openInputChannels = (outInputChannel != null && (attrs.inputFeatures & INPUT_FEATURE_NO_INPUT_CHANNEL) == 0); if (openInputChannels) { win.openInputChannel(outInputChannel); } ... displayContent.getInputMonitor().setUpdateInputWindowsNeededLw(); boolean focusChanged = false; //当该窗口能接收按键事件,则更新聚焦窗口 if (win.canReceiveKeys()) { focusChanged = updateFocusedWindowLocked(UPDATE_FOCUS_WILL_ASSIGN_LAYERS, false /*updateInputWindows*/); if (focusChanged) { imMayMove = false; } } if (imMayMove) { displayContent.computeImeTarget(true /* updateImeTarget */); } ... } Binder.restoreCallingIdentity(origId); return res; }
创建 SurfaceSession 对象,并将当前 Session 添加到 WMS.mSessions 成员变量。
Session.java:
void windowAddedLocked(String packageName) { mPackageName = packageName; mRelayoutTag = "relayoutWindow: " + mPackageName; if (mSurfaceSession == null) { if (DEBUG) { Slog.v(TAG_WM, "First window added to " + this + ", creating SurfaceSession"); } mSurfaceSession = new SurfaceSession(); ProtoLog.i(WM_SHOW_TRANSACTIONS, " NEW SURFACE SESSION %s", mSurfaceSession); mService.mSessions.add(this); if (mLastReportedAnimatorScale != mService.getCurrentAnimatorScale()) { mService.dispatchNewAnimatorScaleLocked(this); } } mNumWindow++; }
SurfaceSession 的创建会调用 JNI,在 JNI 调用 nativeCreate()。
android_view_SurfaceSession.cpp:
static jlong nativeCreate(JNIEnv* env, jclass clazz) { SurfaceComposerClient* client = new SurfaceComposerClient(); client->incStrong((void*)nativeCreate); return reinterpret_cast<jlong>(client); }
static jlong nativeCreate(JNIEnv* env, jclass clazz, jobject sessionObj, jstring nameStr, jint w, jint h, jint format, jint flags, jlong parentObject, jobject metadataParcel) { ScopedUtfChars name(env, nameStr); sp<SurfaceComposerClient> client; if (sessionObj != NULL) { client = android_view_SurfaceSession_getClient(env, sessionObj); } else { client = SurfaceComposerClient::getDefault(); } SurfaceControl *parent = reinterpret_cast<SurfaceControl*>(parentObject); sp<SurfaceControl> surface; LayerMetadata metadata; Parcel* parcel = parcelForJavaObject(env, metadataParcel); if (parcel && !parcel->objectsCount()) { status_t err = metadata.readFromParcel(parcel); if (err != NO_ERROR) { jniThrowException(env, "java/lang/IllegalArgumentException", "Metadata parcel has wrong format"); } } status_t err = client->createSurfaceChecked( String8(name.c_str()), w, h, format, &surface, flags, parent, std::move(metadata)); if (err == NAME_NOT_FOUND) { jniThrowException(env, "java/lang/IllegalArgumentException", NULL); return 0; } else if (err != NO_ERROR) { jniThrowException(env, OutOfResourcesException, NULL); return 0; } surface->incStrong((void *)nativeCreate); return reinterpret_cast<jlong>(surface.get()); }
通过以上JNI接口获取SurfaceComposerClient 对象,作为跟 SurfaceFlinger 通信的代理对象。
void SurfaceComposerClient::onFirstRef() { //getComposerService() 将返回 SF 的 Binder 代理端的 BpSurfaceFlinger 对象 sp<ISurfaceComposer> sf(ComposerService::getComposerService()); if (sf != nullptr && mStatus == NO_INIT) { sp<ISurfaceComposerClient> conn; //调用 SF 的 createConnection() conn = sf->createConnection(); if (conn != nullptr) { mClient = conn; mStatus = NO_ERROR; } } }
比如截屏接口就会通过SurfaceControl调用到其中的 capture 接口:
status_t ScreenshotClient::capture(const sp<IBinder>& display, ui::Dataspace reqDataSpace, ui::PixelFormat reqPixelFormat, const Rect& sourceCrop, uint32_t reqWidth, uint32_t reqHeight, bool useIdentityTransform, ui::Rotation rotation, bool captureSecureLayers, sp<GraphicBuffer>* outBuffer, bool& outCapturedSecureLayers) { sp<ISurfaceComposer> s(ComposerService::getComposerService()); if (s == nullptr) return NO_INIT; status_t ret = s->captureScreen(display, outBuffer, outCapturedSecureLayers, reqDataSpace, reqPixelFormat, sourceCrop, reqWidth, reqHeight, useIdentityTransform, rotation, captureSecureLayers); if (ret != NO_ERROR) { return ret; } return ret; }
然后具体看下核心的SurfaceFlinger实现:
SurfaceFlinger.cpp:
sp<ISurfaceComposerClient> SurfaceFlinger::createConnection() { //创建一个Client const sp<Client> client = new Client(this); return client->initCheck() == NO_ERROR ? client : nullptr; }
回到之前,创建完 ViewRootImpl 对象后,接下来调用该对象的 setView() 方法。在 setView() 中调用了 requestLayout() 方法,现在具体来看下这个方法调用流程:
@Override public void requestLayout() { if (!mHandlingLayoutInLayoutRequest) { checkThread(); mLayoutRequested = true; scheduleTraversals(); } }
@UnsupportedAppUsage void scheduleTraversals() { if (!mTraversalScheduled) { mTraversalScheduled = true; mTraversalBarrier = mHandler.getLooper().getQueue().postSyncBarrier();
//启动TraversalRunnable mChoreographer.postCallback( Choreographer.CALLBACK_TRAVERSAL, mTraversalRunnable, null); notifyRendererOfFramePending(); pokeDrawLockIfNeeded(); } }
final class TraversalRunnable implements Runnable { @Override public void run() { doTraversal(); } }
void doTraversal() { if (mTraversalScheduled) { mTraversalScheduled = false; mHandler.getLooper().getQueue().removeSyncBarrier(mTraversalBarrier); if (mProfile) { Debug.startMethodTracing("ViewAncestor"); }
//调用performTraversals
performTraversals();
if (mProfile) { Debug.stopMethodTracing(); mProfile = false; } } }
private void performTraversals() { // cache mView since it is used so much below... final View host = mView; //它就是 DecorView ... if (mFirst || windowShouldResize || viewVisibilityChanged || cutoutChanged || params != null || mForceNextWindowRelayout) { mForceNextWindowRelayout = false; if (isViewVisible) { // If this window is giving internal insets to the window // manager, and it is being added or changing its visibility, // then we want to first give the window manager "fake" // insets to cause it to effectively ignore the content of // the window during layout. This avoids it briefly causing // other windows to resize/move based on the raw frame of the // window, waiting until we can finish laying out this window // and get back to the window manager with the ultimately // computed insets. insetsPending = computesInternalInsets && (mFirst || viewVisibilityChanged); } ... try { if (DEBUG_LAYOUT) { Log.i(mTag, "host=w:" + host.getMeasuredWidth() + ", h:" + host.getMeasuredHeight() + ", params=" + params); } if (mAttachInfo.mThreadedRenderer != null) { // relayoutWindow may decide to destroy mSurface. As that decision // happens in WindowManager service, we need to be defensive here // and stop using the surface in case it gets destroyed. if (mAttachInfo.mThreadedRenderer.pause()) { // Animations were running so we need to push a frame // to resume them mDirty.set(0, 0, mWidth, mHeight); } mChoreographer.mFrameInfo.addFlags(FrameInfo.FLAG_WINDOW_LAYOUT_CHANGED); } // 关键函数relayoutWindow relayoutResult = relayoutWindow(params, viewVisibility, insetsPending); if (DEBUG_LAYOUT) Log.v(mTag, "relayout: frame=" + frame.toShortString() + " cutout=" + mPendingDisplayCutout.get().toString() + " surface=" + mSurface); // If the pending {@link MergedConfiguration} handed back from // {@link #relayoutWindow} does not match the one last reported, // WindowManagerService has reported back a frame from a configuration not yet // handled by the client. In this case, we need to accept the configuration so we // do not lay out and draw with the wrong configuration. if (!mPendingMergedConfiguration.equals(mLastReportedMergedConfiguration)) { if (DEBUG_CONFIGURATION) Log.v(mTag, "Visible with new config: " + mPendingMergedConfiguration.getMergedConfiguration()); performConfigurationChange(mPendingMergedConfiguration, !mFirst, INVALID_DISPLAY /* same display */); updatedConfiguration = true; } } catch (RemoteException e) { } ... } boolean cancelDraw = mAttachInfo.mTreeObserver.dispatchOnPreDraw() || !isViewVisible; if (!cancelDraw) { if (mPendingTransitions != null && mPendingTransitions.size() > 0) { for (int i = 0; i < mPendingTransitions.size(); ++i) { mPendingTransitions.get(i).startChangingAnimations(); } mPendingTransitions.clear(); } //开始绘制,其中调用了draw(fullRedrawNeeded); performDraw(); } else { if (isViewVisible) { // Try again scheduleTraversals(); } else if (mPendingTransitions != null && mPendingTransitions.size() > 0) { for (int i = 0; i < mPendingTransitions.size(); ++i) { mPendingTransitions.get(i).endChangingAnimations(); } mPendingTransitions.clear(); } } if (mAttachInfo.mContentCaptureEvents != null) { notifyContentCatpureEvents(); } mIsInTraversal = false; }
再看下relayoutWindow的实现:
...
if (mSurfaceControl.isValid()) { if (!useBLAST()) { //先创建一个本地Surface,然后调用copyFrom 将SurfaceControl信息拷贝到Surface中 mSurface.copyFrom(mSurfaceControl); } else { final Surface blastSurface = getOrCreateBLASTSurface(mSurfaceSize.x, mSurfaceSize.y); // If blastSurface == null that means it hasn't changed since the last time we // called. In this situation, avoid calling transferFrom as we would then // inc the generation ID and cause EGL resources to be recreated. if (blastSurface != null) { mSurface.transferFrom(blastSurface); } } } else { destroySurface(); }
....
SurfaceControl 类可以看作是一个 wrapper 类,最后会执行 copyFrom() 将其返回给 App 客户端:
@UnsupportedAppUsage public void copyFrom(SurfaceControl other) { if (other == null) { throw new IllegalArgumentException("other must not be null"); } long surfaceControlPtr = other.mNativeObject; if (surfaceControlPtr == 0) { throw new NullPointerException( "null SurfaceControl native object. Are you using a released SurfaceControl?"); } //通过JNI获取源SurfaceControl long newNativeObject = nativeGetFromSurfaceControl(mNativeObject, surfaceControlPtr); synchronized (mLock) { if (newNativeObject == mNativeObject) { return; } if (mNativeObject != 0) { nativeRelease(mNativeObject); } //保存到全局mNativeObject用于外部调用 setNativeObjectLocked(newNativeObject); } }
Surface 显示过程总结:
在 App 进程中创建 PhoneWindow 后会创建 ViewRoot。ViewRoot 的创建会创建一个 Surface,这个 Surface 其实是空的,通过与 WindowManagerService 通信 copyFrom() 一个NativeSurface 与 SurfaceFlinger 通信时。
关于Native Window:
Native Window是OpenGL与本地窗口系统之间搭建了桥梁。整个GUI系统至少需要两种本地窗口:
(1)面向管理者(SurfaceFlinger)
SurfaceFlinger是系统中所有UI界面的管理者,需要直接或间接的持有“本地窗口”,此本地窗口是FramebufferNativeWindow。
(2)面向应用程序
这类本地窗口是Surface。
正常情况按照SDK向导生成APK应用程序,是采用Skia等第三方图形库,而对于希望使用OpenGL ES来完成复杂界面渲染的应用开发者来说,Android也提供封装的GLSurfaceView(或其他方式)来实现图形显示。
①FramebufferNativeWindow
EGL需要通过本地窗口来为OpenGL/OpenGL ES创建环境。由于OpenGL/ES对多平台支持,考虑到兼容性和移植性。不同平台的本地窗口EGLNativeWindowType数据类型不同。
Android平台的数据类型是ANativeWindow,像是一份“协议”,规定了一个本地窗口的形态和功能。ANativeWindow是FramebufferNativeWindow的父类。
Android中,由于多缓冲技术,EGLNativeWindowType所管理的缓冲区最少2个,最大3个。
FramebufferNativeWindow初始化需要Golloc支持,步骤如下:
-
- 加载GRALLOC_HARDWARE_MODULE_ID模块,参见上节。
- 分别打开fb和gralloc设备,打开后的设备由全局变量fbDev和grDev管理。
- 根据设备的属性来给FramebufferNativeWindow赋初值。
- 根据FramebufferNativeWindow的实现来填充ANativeWindow中的“协议”
- 其他一些必要的初始化
②应用程序的本地窗口 - Surface
Surface也继承了ANativeWindow
class Surface : public ANativeObjectBase<ANativeWindow, Surface, RefBase>
Surface是面向Android系统中所有UI应用程序的,即它承担着应用进程中的UI显示需求。
需要面向上层实现(主要是Java层)提供绘制图像的画板。SurfaceFlinger需要收集系统中所有应用程序绘制的图像数据,然后集中显示到物理屏幕上。Surface需要扮演相应角色,本质上还是由SurfaceFlinger服务统一管理的,涉及到很多跨进程的通信细节。
③Surface的创建
Surface将通过mGraphicBufferProducer来获取buffer,这些缓冲区会被记录在mSlots中数据中。mGraphicBufferProducer这一核心成员的初始化流程如下:
ViewRootImpl持有一个Java层的Surface对象(mSurface)。
ViewRootImpl向WindowManagerService发起relayout请求,此时mSurface被赋予真正的有效值,将辗转生成的SurfaceControl通过Surface.copyFrom()函数复制到mSurface中。
由此,Surface由SurfaceControl管理,SurfaceControl由SurfaceComposerClient创建。SurfaceComposerClient获得的匿名Binder是ISurfaceComposer,其服务端实现是SurfaceFlinger。而Surface依赖的IGraphicBufferProducer对象在Service端的实现是BufferQueue。
class SurfaceFlinger : public BinderService<SurfaceFlinger>, //在ServiceManager中注册为SurfaceFlinger public BnSurfaceComposer,//实现的接口却叫ISurfaceComposer
④SurfaceFlinger服务框架:
Buffer,Consumer,Producer是“生产者-消费者”模型中的3个参与对象,如何协调好它们的工作是应用程序能否正常显示UI的关键。
Buffer是BufferQueue,Producer是应用程序,Consumer是SurfaceFlinger。
Surface内部提供一个BufferQueue,与上层和SurfaceFlinger形成一个生产者消费者模型,上层对应Producer,SurfaceFlinger对应Consumer。三者通过Buffer产生联系,每个Buffer都有四种状态:
- Free:可被上层使用;
- Dequeued:出列,正在被上层使用;
- Queued:入列,已完成上层绘制,等待SurfaceFlinger合成;
- Acquired:被获取,SurfaceFlinger正持有该Buffer进行合成;
如此循环,形成一个Buffer被循环使用的过程(FREE-> DEQUEUED->QUEUED->ACQUIRED->FREE)。
BufferQueue中的mSlots数组用于管理期内的缓冲区,最大容器是32。数据缓冲区的空间是动态分配的,应用程序与SurfaceFlinger都是使用OpenGL ES来完成UI显示。Layer类在SurfaceFlinger中表示“层”,通俗地讲就是代表了一个“画面”,最终物理屏幕上的显示结果就是通过对系统中同时存在的所有“画面”进行处理叠加而成。
posted on 2019-07-30 21:00 sheldon_blogs 阅读(51833) 评论(5) 编辑 收藏 举报