Input源码解读——从"Show tabs"开始
本文基于Android T版本源码，梳理当用户在开发者选项中开启Show tabs功能后显示第点按操作的视觉反馈的原理，来进一步了解Android Input系统
Settings 写入设置
首先是设置应用(Settings)提供的开发者选项画面响应点击，将Show taps选项对应的设置Key SHOW_TOUCHES的 ON 值通过android.provder.Settings接口写入到保存系统设置数据的SettingsProvier中。
// packages/apps/Settings/src/com/android/settings/development/ShowTapsPreferenceController.java public class ShowTapsPreferenceController extends DeveloperOptionsPreferenceController ... { ... @Override public boolean onPreferenceChange(Preference preference, Object newValue) { final boolean isEnabled = (Boolean) newValue; Settings.System.putInt(mContext.getContentResolver(), Settings.System.SHOW_TOUCHES, isEnabled ? SETTING_VALUE_ON : SETTING_VALUE_OFF); return true; } ... } InputManagerService监听设置
负责管理输入的系统服务InputManagerService在启动之际，会监听设置中的 SHOW_TOUCHES字段的变化，在设置产生变化的时候调用native侧的代码进行处理。
// frameworks/base/services/core/java/com/android/server/input/InputManagerService.java public class InputManagerService extends IInputManager.Stub... { ... public void start() { ... registerShowTouchesSettingObserver(); ... } private void registerShowTouchesSettingObserver() { mContext.getContentResolver().registerContentObserver( Settings.System.getUriFor(Settings.System.SHOW_TOUCHES), true, new ContentObserver(mHandler) { @Override public void onChange(boolean selfChange) { updateShowTouchesFromSettings(); } }, UserHandle.USER_ALL); } private void updateShowTouchesFromSettings() { int setting = getShowTouchesSetting(0); mNative.setShowTouches(setting != 0); } ... // frameworks/base/services/core/java/com/android/server/input/NativeInputManagerService.java public interface NativeInputManagerService { ... void setShowTouches(boolean enabled); ... } // frameworks/base/services/core/jni/com_android_server_input_InputManagerService.cpp class NativeInputManager : public virtual RefBase, ...{ ... void setShowTouches(bool enabled); ... } void NativeInputManager::setShowTouches(bool enabled) { { // acquire lock AutoMutex _l(mLock); if (mLocked.showTouches == enabled) { return; } ALOGI("Setting show touches feature to %s.", enabled ? "enabled" : "disabled"); mLocked.showTouches = enabled; } // release lock mInputManager->getReader().requestRefreshConfiguration( InputReaderConfiguration::CHANGE_SHOW_TOUCHES); } 这里的mInputManager是InputManagerInterface对象实例，InputManager是InputManagerInterface和子类，所以通过mInputManager可以连接NativeInputManager和InputReader。
这里向负责读取事件的InputReader发出更新配置的请求，配置变更的Type为 CHANGE_SHOW_TOUCHES。
通过 InputReader 请求刷新配置
InputReader接收到配置变化的Type之后，会根据记录待刷新配置的变量 mConfigurationChangesToRefresh判断当前是否已经在刷新过程中。
如果尚未处于刷新中，则更新mConfigurationChangesToRefresh的值，并唤醒EventHub进行配置刷新。
// frameworks/native/services/inputflinger/reader/InputReader.cpp void InputReader::requestRefreshConfiguration(uint32_t changes) { std::scoped_lock _l(mLock); if (changes) { bool needWake = !mConfigurationChangesToRefresh; mConfigurationChangesToRefresh |= changes; if (needWake) { mEventHub->wake(); } } } EventHub 唤醒 InputReader 线程
InputManagerService过来的刷新请求最终需要InputReader线程来处理。
可是 InputReader 线程处在从EventHub中读取事件和没有事件时便调用epoll_wait进入等待状态的循环当中。
所以为了让其立即处理配置变化，需要EventHub的手动唤醒。
// frameworks/native/services/inputflinger/reader/EventHub.cpp void EventHub::wake() { ALOGV("wake() called"); ssize_t nWrite; do { nWrite = write(mWakeWritePipeFd, "W", 1); } while (nWrite == -1 && errno == EINTR); if (nWrite != 1 && errno != EAGAIN) { ALOGW("Could not write wake signal: %s", strerror(errno)); } } size_t EventHub::getEvents(int timeoutMillis, RawEvent* buffer, size_t bufferSize) { ... for (;;) { ... int pollResult = epoll_wait(mEpollFd, mPendingEventItems, EPOLL_MAX_EVENTS, timeoutMillis); ... } ... } InputReader线程刷新配置
EventHub唤醒后处于等待状态的getEvents会结束，之后InputReader线程会进入下次循环即loopOnce。
其首先将检查是否存在待刷新的配置变化changes，存在的话调用refreshConfigurationLocked让InputDevice去重新适配变化。
// frameworks/native/services/inputflinger/reader/InputReader.cpp void InputReader::loopOnce() { ... std::vector<InputDeviceInfo> inputDevices; { // acquire lock ... uint32_t changes = mConfigurationChangesToRefresh; if (changes) { mConfigurationChangesToRefresh = 0; timeoutMillis = 0; refreshConfigurationLocked(changes); } else if (mNextTimeout != LLONG_MAX) { nsecs_t now = systemTime(SYSTEM_TIME_MONOTONIC); timeoutMillis = toMillisecondTimeoutDelay(now, mNextTimeout); } } // release lock size_t count = mEventHub->getEvents(timeoutMillis, mEventBuffer, EVENT_BUFFER_SIZE); ... } 需要留意，refreshConfigurationLocked在调用InputDevice进一步处理之前需要先获取配置的变化放入mConfig中。
// frameworks/native/services/inputflinger/reader/InputReader.cpp void InputReader::refreshConfigurationLocked(uint32_t changes) { mPolicy->getReaderConfiguration(&mConfig); ... if (changes & InputReaderConfiguration::CHANGE_MUST_REOPEN) { mEventHub->requestReopenDevices(); } else { for (auto& devicePair : mDevices) { std::shared_ptr<InputDevice>& device = devicePair.second; device->configure(now, &mConfig, changes); } } ... } InputDevice配置变化
InputDevice的configure需要处理很多配置变化，比如键盘布局、麦克风等。对于Show taps的变化关注调用 InputMapper的congfigure即可。
// frameworks/native/services/inputflinger/reader/InputDevice.cpp void InputDevice::configure(nsecs_t when, const InputReaderConfiguration* config, uint32_t changes) { ... if (!isIgnored()) { ... for_each_mapper([this, when, config, changes](InputMapper& mapper) { mapper.configure(when, config, changes); mSources |= mapper.getSources(); }); ... } } TouchInputMapper 进一步处理
众多输入事件的物理数据需要对应的InputMapper来转化为上层能识别的事件类型。比如识别键盘输入的 KeyboardInputMapper、识别震动的VibratorInputMapper等等。
现在的触摸屏都支持多点触控，所以是MultiTouchInputMapper来处理的。可MultiTouchInputMapper没有复写 configure()，而是沿用由父类TouchInputMapper的共通处理。
// frameworks/native/services/inputflinger/reader/mapper/TouchInputMapper.cpp void TouchInputMapper::configure(nsecs_t when, const InputReaderConfiguration* config, uint32_t changes) { ... bool resetNeeded = false; if (!changes || (changes & (InputReaderConfiguration::CHANGE_DISPLAY_INFO | InputReaderConfiguration::CHANGE_POINTER_CAPTURE | InputReaderConfiguration::CHANGE_POINTER_GESTURE_ENABLEMENT | InputReaderConfiguration::CHANGE_SHOW_TOUCHES | InputReaderConfiguration::CHANGE_EXTERNAL_STYLUS_PRESENCE))) { // Configure device sources, display dimensions, orientation and // scaling factors. configureInputDevice(when, &resetNeeded); } ... } TouchInputMapper会依据changes的类型进行对应处理，对于SHOW_TOUCHES的变化需要调用configureInputDevice进一步处理。
创建和初始化 PointerController
configureInputDevice进行多个参数的测量和配置，其中和Show taps相关的是PointerController的创建，该类是 Mouse、Taps、Pointer location 等系统 Touch 显示的专用类。
// frameworks/native/services/inputflinger/reader/mapper/TouchInputMapper.cpp void TouchInputMapper::configureInputDevice(nsecs_t when, bool* outResetNeeded) { ... // Create pointer controller if needed, and keep it around if Pointer Capture is enabled to // preserve the cursor position. if (mDeviceMode == DeviceMode::POINTER || (mDeviceMode == DeviceMode::DIRECT && mConfig.showTouches) || (mParameters.deviceType == Parameters::DeviceType::POINTER && mConfig.pointerCaptureRequest.enable)) { if (mPointerController == nullptr) { mPointerController = getContext()->getPointerController(getDeviceId()); } if (mConfig.pointerCaptureRequest.enable) { mPointerController->fade(PointerControllerInterface::Transition::IMMEDIATE); } } else { mPointerController.reset(); } ... } 这里调用InputReaderContext#getPointerController，InputReader::ContextImpl是InputReaderContext的子类，所以会回调到InputReader开启PointerController的创建和初始化。
// frameworks/native/services/inputflinger/reader/InputReader.cpp std::shared_ptr<PointerControllerInterface> InputReader::ContextImpl::getPointerController( int32_t deviceId) { // lock is already held by the input loop return mReader->getPointerControllerLocked(deviceId); } std::shared_ptr<PointerControllerInterface> InputReader::getPointerControllerLocked( int32_t deviceId) { std::shared_ptr<PointerControllerInterface> controller = mPointerController.lock(); if (controller == nullptr) { controller = mPolicy->obtainPointerController(deviceId); mPointerController = controller; updatePointerDisplayLocked(); } return controller; } 这里调用InputReaderPolicyInterface#obtainPointerController，而NativeInputManager是InputReaderPolicyInterface的子类。
// frameworks/base/services/core/jni/com_android_server_input_InputManagerService.cpp std::shared_ptr<PointerControllerInterface> NativeInputManager::obtainPointerController( int32_t /* deviceId */) { ... std::shared_ptr<PointerController> controller = mLocked.pointerController.lock(); if (controller == nullptr) { ensureSpriteControllerLocked(); controller = PointerController::create(this, mLooper, mLocked.spriteController); mLocked.pointerController = controller; updateInactivityTimeoutLocked(); } return controller; } PointerController 构建的同时需要构建持有的 MouseCursorController。
// frameworks/base/libs/input/PointerController.cpp std::shared_ptr<PointerController> PointerController::create( ... ) { std::shared_ptr<PointerController> controller = std::shared_ptr<PointerController>( new PointerController(policy, looper, spriteController)); ... return controller; } PointerController::PointerController( ... ) : mContext(policy, looper, spriteController, *this), mCursorController(mContext) { std::scoped_lock lock(mLock); mLocked.presentation = Presentation::SPOT; ... } obtainPointerController执行完之后调用updatePointerDisplayLocked执行PointerController的初始化。
初始化 PointerController
调用PointerController的setDisplayViewport传入显示用的DisplayViewPort。
// frameworks/native/services/inputflinger/reader/InputReader.cpp void InputReader::updatePointerDisplayLocked() { ... std::optional<DisplayViewport> viewport = mConfig.getDisplayViewportById(mConfig.defaultPointerDisplayId); if (!viewport) { ... viewport = mConfig.getDisplayViewportById(ADISPLAY_ID_DEFAULT); } ... controller->setDisplayViewport(*viewport); } setDisplayViewport需要持有的MouseCursorController进一步初始化。
// frameworks/base/libs/input/PointerController.cpp void PointerController::setDisplayViewport(const DisplayViewport& viewport) { ... mCursorController.setDisplayViewport(viewport, getAdditionalMouseResources); } MouseCursorController需要获取Display相关的参数，并执行两个重要步骤：loadResourcesLocked/updatePointerLocked
// frameworks/base/libs/input/MouseCursorController.cpp void MouseCursorController::setDisplayViewport(const DisplayViewport& viewport, bool getAdditionalMouseResources) { ... // Reset cursor position to center if size or display changed. if (oldViewport.displayId != viewport.displayId || oldDisplayWidth != newDisplayWidth || oldDisplayHeight != newDisplayHeight) { float minX, minY, maxX, maxY; if (getBoundsLocked(&minX, &minY, &maxX, &maxY)) { mLocked.pointerX = (minX + maxX) * 0.5f; mLocked.pointerY = (minY + maxY) * 0.5f; // Reload icon resources for density may be changed. loadResourcesLocked(getAdditionalMouseResources); ... } } else if (oldViewport.orientation != viewport.orientation) { ... } updatePointerLocked(); } 加载 Pointer 相关资源
// frameworks/base/libs/input/MouseCursorController.cpp void MouseCursorController::loadResourcesLocked(bool getAdditionalMouseResources) REQUIRES(mLock) { ... policy->loadPointerResources(&mResources, mLocked.viewport.displayId); policy->loadPointerIcon(&mLocked.pointerIcon, mLocked.viewport.displayId); ... } 省略诸多细节，loadPointerResources将通过InputManagerService的JNI端以及PointerIcon的JNI端创建PointerIcon实例，并读取显示的资源。
getSystemIcon则是负责的函数，其将读取系统资源里名为Pointer的Style，并读取指针对应的资源 ID。
// frameworks/base/core/java/android/view/PointerIcon.java public static PointerIcon getSystemIcon(@NonNull Context context, int type) { ... int typeIndex = getSystemIconTypeIndex(type); if (typeIndex == 0) { typeIndex = getSystemIconTypeIndex(TYPE_DEFAULT); } int defStyle = sUseLargeIcons ? com.android.internal.R.style.LargePointer : com.android.internal.R.style.Pointer; TypedArray a = context.obtainStyledAttributes(null, com.android.internal.R.styleable.Pointer, 0, defStyle); int resourceId = a.getResourceId(typeIndex, -1); ... icon = new PointerIcon(type); if ((resourceId & 0xff000000) == 0x01000000) { icon.mSystemIconResourceId = resourceId; } else { icon.loadResource(context, context.getResources(), resourceId); } systemIcons.append(type, icon); return icon; } private static int getSystemIconTypeIndex(int type) { switch (type) { ... case TYPE_SPOT_TOUCH: return com.android.internal.R.styleable.Pointer_pointerIconSpotTouch; ... default: return 0; } } 资源 ID 为 pointer_spot_touch_icon。
<!-- frameworks/base/core/res/res/drawable/pointer_spot_touch_icon.xml --> <?xml version="1.0" encoding="utf-8"?> <pointer-icon xmlns:android="http://schemas.android.com/apk/res/android" android:bitmap="@drawable/pointer_spot_touch" android:hotSpotX="16dp" android:hotSpotY="16dp" /> 其指向的图片就是如下熟悉的 Spot png：pointer_spot_touch.png。之后的loadPointerIcon阶段会将该图片解析成 Bitmap 并被管理在SpriteIcon中。
而SpriteIcon在updatePointerLocked阶段会被存放到SpriteController中，等待显示的调度。
// frameworks/base/libs/input/MouseCursorController.cpp void MouseCursorController::updatePointerLocked() REQUIRES(mLock) { if (!mLocked.viewport.isValid()) { return; } sp<SpriteController> spriteController = mContext.getSpriteController(); spriteController->openTransaction(); ... if (mLocked.updatePointerIcon) { if (mLocked.requestedPointerType == mContext.getPolicy()->getDefaultPointerIconId()) { mLocked.pointerSprite->setIcon(mLocked.pointerIcon); ... } mLocked.updatePointerIcon = false; } spriteController->closeTransaction(); } 显示Tap
点击的时候EventHub#getEvents会产生事件，InputReader#loopOnce会调用processEventsLocked处理事件。
// frameworks/native/services/inputflinger/reader/InputReader.cpp void InputReader::loopOnce() { ... size_t count = mEventHub->getEvents(timeoutMillis, mEventBuffer, EVENT_BUFFER_SIZE); { // acquire lock ... if (count) { processEventsLocked(mEventBuffer, count); } .... } // release lock ... } 之后调用InputMapper开始加工事件，并在TouchInputMapper#cookAndDispatch的时候调用updateTouchSpots更新 PointerController的一些参数。
// frameworks/native/services/inputflinger/reader/mapper/TouchInputMapper.cpp void TouchInputMapper::updateTouchSpots() { ... mPointerController->setPresentation(PointerControllerInterface::Presentation::SPOT); mPointerController->fade(PointerControllerInterface::Transition::GRADUAL); mPointerController->setButtonState(mCurrentRawState.buttonState); setTouchSpots(mCurrentCookedState.cookedPointerData.pointerCoords, mCurrentCookedState.cookedPointerData.idToIndex, mCurrentCookedState.cookedPointerData.touchingIdBits, mViewport.displayId); } 其中比较关键的setTouchSpots是显示Taps的关键步骤，准备 x、y 坐标和压力值。
在 Reader 而不是 Dispatch、更不是 ViewRootImpl 的时候处理的原因在于：Read 到事件即显示可以更早地响，同时不用占用 App 进程。
// frameworks/native/services/inputflinger/reader/mapper/TouchInputMapper.cpp void TouchInputMapper::setTouchSpots(const PointerCoords* spotCoords, const uint32_t* spotIdToIndex, BitSet32 spotIdBits, int32_t displayId) { std::array<PointerCoords, MAX_POINTERS> outSpotCoords{}; for (BitSet32 idBits(spotIdBits); !idBits.isEmpty();) { const uint32_t index = spotIdToIndex[idBits.clearFirstMarkedBit()]; float x = spotCoords[index].getX(); float y = spotCoords[index].getY(); float pressure = spotCoords[index].getAxisValue(AMOTION_EVENT_AXIS_PRESSURE); ... } mPointerController->setSpots(outSpotCoords.data(), spotIdToIndex, spotIdBits, displayId); } 其后PointerController会通过TouchSpotController创建Spot实例向其发送updateSprite请求。最后回调 SpriteController调用setIcon处理。
// frameworks/base/libs/input/TouchSpotController.cpp void TouchSpotController::Spot::updateSprite(const SpriteIcon* icon, float x, float y, int32_t displayId) { sprite->setLayer(Sprite::BASE_LAYER_SPOT + id); ... if (icon != mLastIcon) { mLastIcon = icon; if (icon) { sprite->setIcon(*icon); sprite->setVisible(true); } else { sprite->setVisible(false); } } } // frameworks/base/libs/input/SpriteController.cpp void SpriteController::SpriteImpl::setIcon(const SpriteIcon& icon) { AutoMutex _l(mController->mLock); ... invalidateLocked(dirty); } void SpriteController::SpriteImpl::invalidateLocked(uint32_t dirty) { ... if (!wasDirty) { mController->invalidateSpriteLocked(this); } } void SpriteController::invalidateSpriteLocked(const sp<SpriteImpl>& sprite) { bool wasEmpty = mLocked.invalidatedSprites.isEmpty(); mLocked.invalidatedSprites.push(sprite); if (wasEmpty) { if (mLocked.transactionNestingCount != 0) { mLocked.deferredSpriteUpdate = true; } else { mLooper->sendMessage(mHandler, Message(MSG_UPDATE_SPRITES)); } } } MSG_UPDATE_SPRITES经过 Handler 回调doUpdateSprites，将取出封装在SpriteUpdate中的SpriteIcon并执行 draw。
// frameworks/base/libs/input/SpriteController.cpp void SpriteController::doUpdateSprites() { ... for (size_t i = 0; i < numSprites; i++) { SpriteUpdate& update = updates.editItemAt(i); if ((update.state.dirty & DIRTY_BITMAP) && update.state.surfaceDrawn) { update.state.surfaceDrawn = false; update.surfaceChanged = surfaceChanged = true; } if (update.state.surfaceControl != NULL && !update.state.surfaceDrawn && update.state.wantSurfaceVisible()) { sp<Surface> surface = update.state.surfaceControl->getSurface(); if (update.state.icon.draw(surface)) { update.state.surfaceDrawn = true; update.surfaceChanged = surfaceChanged = true; } } } ... updates.clear(); } 最后，SpriteIcon将取出Bitmap描画到Surface的Canvas上去。
// frameworks/base/libs/input/SpriteIcon.cpp bool SpriteIcon::draw(sp<Surface> surface) const { ... graphics::Paint paint; paint.setBlendMode(ABLEND_MODE_SRC); graphics::Canvas canvas(outBuffer, (int32_t)surface->getBuffersDataSpace()); canvas.drawBitmap(bitmap, 0, 0, &paint); ... status = surface->unlockAndPost(); if (status) { ALOGE("Error %d unlocking and posting sprite surface after drawing.", status); } return !status; } 总体流程
通过一个框图简单回顾一下整个流程。
可以看到，简简单单的 Show taps 功能，从设置、配置、刷新再到显示，经历了多个进程、多个模块的协力。
涉及的Input核心逻辑框图
