Android输入事件从读取到分发二：谁在循环监听事件的到来

通过上一节初步阅读代码，已经找到了读写/dev/input/设备文件节点的位置。但是最后，我觉得应该有一个线程，一直循环监听这些输入设备，有事件的时候就去处理，没有事件的时候就睡眠，等待事件的到来。那么，这部分的代码是怎么样的呢？

上一节只是为了定位android系统在什么地方监听输入设备，所以很多地方没有仔细分析，这一节，带着文章开头提出的问题，再一次分析源码，而我们的入口，任然是系统启动后，注册服务的SystemServer.java文件。

            Slog.i(TAG, "Input Manager");
            inputManager = new InputManagerService(context);

            Slog.i(TAG, "Window Manager");
            wm = WindowManagerService.main(context, inputManager,
                    mFactoryTestMode != FactoryTest.FACTORY_TEST_LOW_LEVEL,
                    !mFirstBoot, mOnlyCore);
            ServiceManager.addService(Context.WINDOW_SERVICE, wm);
            ServiceManager.addService(Context.INPUT_SERVICE, inputManager);

            mActivityManagerService.setWindowManager(wm);

            inputManager.setWindowManagerCallbacks(wm.getInputMonitor());
            inputManager.start();

还是注册InputManagerService的这部分代码，上一节我们只是分析了InputManagerService的构造函数，没有分析最后调用的这个start方法，那这一次，就从这个start方法入手，再一次追踪android输入系统在系统开机后的一系列动作，从而明白输入系统是怎么一步步运行起来的。start函数的代码如下：

   public void start() {
        Slog.i(TAG, "Starting input manager");
        nativeStart(mPtr);

        // Add ourself to the Watchdog monitors.
        Watchdog.getInstance().addMonitor(this);

        registerPointerSpeedSettingObserver();
        registerShowTouchesSettingObserver();

        mContext.registerReceiver(new BroadcastReceiver() {
            @Override
            public void onReceive(Context context, Intent intent) {
                updatePointerSpeedFromSettings();
                updateShowTouchesFromSettings();
            }
        }, new IntentFilter(Intent.ACTION_USER_SWITCHED), null, mHandler);

        updatePointerSpeedFromSettings();
        updateShowTouchesFromSettings();
    }

start方法一开始就调用了nativeStart方法，记得上一节分析InputServiceManager的构造函数时，它调用了nativeInit()方法，从而展开了一系列输入事件管理的初始化动作。那么这里的nativeStart是不是就是在构造函数中调用nativeInit后，正式启动输入事件的管理框架的呢？感觉好像是的。nativeStart应该和nativeInit在同一个文件中： base/services/core/jni/com_android_server_input_InputManagerService.cpp，源码如下：

static void nativeStart(JNIEnv* env, jclass clazz, jlong ptr) {
    NativeInputManager* im = reinterpret_cast(ptr);

    status_t result = im->getInputManager()->start();
    if (result) {
        jniThrowRuntimeException(env, "Input manager could not be started.");
    }
}

这里调用了NativeInputManager的getInputManager方法，这个方法很简单：

inline sp getInputManager() const { return mInputManager; }

它是一个内联方法，只是简单返回mInputManager变量，mInputManager是什么呢？

mInputManager = new InputManager(eventHub, this, this);

可以看到它是一个InputManager的实例。所以，在nativeStart方法中，调用的start()是InputManager的start方法，这个方法源码如下：

status_t InputManager::start() {
    status_t result = mDispatcherThread->run("InputDispatcher", PRIORITY_URGENT_DISPLAY);
    if (result) {
        ALOGE("Could not start InputDispatcher thread due to error %d.", result);
        return result;
    }

    result = mReaderThread->run("InputReader", PRIORITY_URGENT_DISPLAY);
    if (result) {
        ALOGE("Could not start InputReader thread due to error %d.", result);

        mDispatcherThread->requestExit();
        return result;
    }

    return OK;
}

这里只做了两件事情，分别启动一个线程,先看第一个线程：

mDispatcherThread = new InputDispatcherThread(mDispatcher);

从名字上看，是事件的分发线程，目前我关注的不是分发，而是事件的读取，所以这里先做个标记，之后研究分发事件的时候，再认真看这里，目前，这里略过。那么第二个线程是什么呢？

mReaderThread = new InputReaderThread(mReader);

从名字上看，它是一个读取事件的线程，好像关键点已经要到来了。nativeStart方法中调用了这个线程的run方法，这会导致thread_loop方法杯调用。InputReaderThread类定义在InputReader.h中，实现在InputRead.cpp中，源码如下：

// --- InputReaderThread ---

InputReaderThread::InputReaderThread(const sp& reader) :
        Thread(/*canCallJava*/ true), mReader(reader) {
}

InputReaderThread::~InputReaderThread() {
}

bool InputReaderThread::threadLoop() {
    mReader->loopOnce();
    return true;
}

InputReaderThreader的构造函数和析构函数都为空，而thread_loop中也只是调用了mReader->loopOnce()方法。注意，这里是C++的线程定义方式，C++线程内建的函数在thread_loop函数返回true后，会不断重复调用thread_loop函数。所以这里可以看做一个死循环。但这里并不是轮询机制，因为loopOnce函数可能会休眠。这里很可能就是我们找的事件监听的循环线程，它不断重复的调用LoogOnce来监听输入事件。所以，现在看一下loopOnce()函数，这是个mReader指向的函数，mReader是一个 InputReader的实例，所以这里就看一下InputReader类中的loopOnce方法：

void InputReader::loopOnce() {
    int32_t oldGeneration;
    int32_t timeoutMillis;
    bool inputDevicesChanged = false;
    Vector inputDevices;
    { // acquire lock
        AutoMutex _l(mLock);

        oldGeneration = mGeneration;
        timeoutMillis = -1;

        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);

    { // acquire lock
        AutoMutex _l(mLock);
        mReaderIsAliveCondition.broadcast();

        if (count) {
            processEventsLocked(mEventBuffer, count);
        }

        if (mNextTimeout != LLONG_MAX) {
            nsecs_t now = systemTime(SYSTEM_TIME_MONOTONIC);
            if (now >= mNextTimeout) {
#if DEBUG_RAW_EVENTS
                ALOGD("Timeout expired, latency=%0.3fms", (now - mNextTimeout) * 0.000001f);
#endif
                mNextTimeout = LLONG_MAX;
                timeoutExpiredLocked(now);
            }
        }

        if (oldGeneration != mGeneration) {
            inputDevicesChanged = true;
            getInputDevicesLocked(inputDevices);
        }
    } // release lock

    // Send out a message that the describes the changed input devices.
    if (inputDevicesChanged) {
        mPolicy->notifyInputDevicesChanged(inputDevices);
    }

    // Flush queued events out to the listener.
    // This must happen outside of the lock because the listener could potentially call
    // back into the InputReader's methods, such as getScanCodeState, or become blocked
    // on another thread similarly waiting to acquire the InputReader lock thereby
    // resulting in a deadlock.  This situation is actually quite plausible because the
    // listener is actually the input dispatcher, which calls into the window manager,
    // which occasionally calls into the input reader.
    mQueuedListener->flush();
}

这个函数就比较长了，主要也就两个代码块，中间夹了一个函数。这个函数有点特别，因为它是EventHub下的一个函数，而EventHub类，就是上节分析出来的，真正直接监听/dev/input/设备节点的类。这个函数就是： mEventHub->getEvents，源码当然在EventHub中：

size_t EventHub::getEvents(int timeoutMillis, RawEvent* buffer, size_t bufferSize) {
    ALOG_ASSERT(bufferSize >= 1);

    AutoMutex _l(mLock);

    struct input_event readBuffer[bufferSize];

    RawEvent* event = buffer;
    size_t capacity = bufferSize;
    bool awoken = false;
    for (;;) {
        nsecs_t now = systemTime(SYSTEM_TIME_MONOTONIC);

        // Reopen input devices if needed.
        if (mNeedToReopenDevices) {
            mNeedToReopenDevices = false;

            ALOGI("Reopening all input devices due to a configuration change.");

            closeAllDevicesLocked();
            mNeedToScanDevices = true;
            break; // return to the caller before we actually rescan
        }

        // Report any devices that had last been added/removed.
        while (mClosingDevices) {
            Device* device = mClosingDevices;
            ALOGV("Reporting device closed: id=%d, name=%s\n",
                 device->id, device->path.string());
            mClosingDevices = device->next;
            event->when = now;
            event->deviceId = device->id == mBuiltInKeyboardId ? BUILT_IN_KEYBOARD_ID : device->id;
            event->type = DEVICE_REMOVED;
            event += 1;
            delete device;
            mNeedToSendFinishedDeviceScan = true;
            if (--capacity == 0) {
                break;
            }
        }

        if (mNeedToScanDevices) {
            mNeedToScanDevices = false;
            scanDevicesLocked();
            mNeedToSendFinishedDeviceScan = true;
        }

        while (mOpeningDevices != NULL) {
            Device* device = mOpeningDevices;
            ALOGV("Reporting device opened: id=%d, name=%s\n",
                 device->id, device->path.string());
            mOpeningDevices = device->next;
            event->when = now;
            event->deviceId = device->id == mBuiltInKeyboardId ? 0 : device->id;
            event->type = DEVICE_ADDED;
            event += 1;
            mNeedToSendFinishedDeviceScan = true;
            if (--capacity == 0) {
                break;
            }
        }

        if (mNeedToSendFinishedDeviceScan) {
            mNeedToSendFinishedDeviceScan = false;
            event->when = now;
            event->type = FINISHED_DEVICE_SCAN;
            event += 1;
            if (--capacity == 0) {
                break;
            }
        }

        // Grab the next input event.
        bool deviceChanged = false;
        while (mPendingEventIndex < mPendingEventCount) {
            const struct epoll_event& eventItem = mPendingEventItems[mPendingEventIndex++];
            if (eventItem.data.u32 == EPOLL_ID_INOTIFY) {
                if (eventItem.events & EPOLLIN) {
                    mPendingINotify = true;
                } else {
                    ALOGW("Received unexpected epoll event 0x%08x for INotify.", eventItem.events);
                }
                continue;
            }

            if (eventItem.data.u32 == EPOLL_ID_WAKE) {
                if (eventItem.events & EPOLLIN) {
                    ALOGV("awoken after wake()");
                    awoken = true;
                    char buffer[16];
                    ssize_t nRead;
                    do {
                        nRead = read(mWakeReadPipeFd, buffer, sizeof(buffer));
                    } while ((nRead == -1 && errno == EINTR) || nRead == sizeof(buffer));
                } else {
                    ALOGW("Received unexpected epoll event 0x%08x for wake read pipe.",
                            eventItem.events);
                }
                continue;
            }

            ssize_t deviceIndex = mDevices.indexOfKey(eventItem.data.u32);
            if (deviceIndex < 0) {
                ALOGW("Received unexpected epoll event 0x%08x for unknown device id %d.",
                        eventItem.events, eventItem.data.u32);
                continue;
            }

            Device* device = mDevices.valueAt(deviceIndex);
            if (eventItem.events & EPOLLIN) {
                int32_t readSize = read(device->fd, readBuffer,
                        sizeof(struct input_event) * capacity);
                if (readSize == 0 || (readSize < 0 && errno == ENODEV)) {
                    // Device was removed before INotify noticed.
                    ALOGW("could not get event, removed? (fd: %d size: %" PRId32
                            " bufferSize: %zu capacity: %zu errno: %d)\n",
                            device->fd, readSize, bufferSize, capacity, errno);
                    deviceChanged = true;
                    closeDeviceLocked(device);
                } else if (readSize < 0) {
                    if (errno != EAGAIN && errno != EINTR) {
                        ALOGW("could not get event (errno=%d)", errno);
                    }
                } else if ((readSize % sizeof(struct input_event)) != 0) {
                    ALOGE("could not get event (wrong size: %d)", readSize);
                } else {
                    int32_t deviceId = device->id == mBuiltInKeyboardId ? 0 : device->id;

                    size_t count = size_t(readSize) / sizeof(struct input_event);
                    for (size_t i = 0; i < count; i++) {
                        struct input_event& iev = readBuffer[i];
                        ALOGV("%s got: time=%d.%06d, type=%d, code=%d, value=%d",
                                device->path.string(),
                                (int) iev.time.tv_sec, (int) iev.time.tv_usec,
                                iev.type, iev.code, iev.value);

                        // Some input devices may have a better concept of the time
                        // when an input event was actually generated than the kernel
                        // which simply timestamps all events on entry to evdev.
                        // This is a custom Android extension of the input protocol
                        // mainly intended for use with uinput based device drivers.
                        if (iev.type == EV_MSC) {
                            if (iev.code == MSC_ANDROID_TIME_SEC) {
                                device->timestampOverrideSec = iev.value;
                                continue;
                            } else if (iev.code == MSC_ANDROID_TIME_USEC) {
                                device->timestampOverrideUsec = iev.value;
                                continue;
                            }
                        }
                        if (device->timestampOverrideSec || device->timestampOverrideUsec) {
                            iev.time.tv_sec = device->timestampOverrideSec;
                            iev.time.tv_usec = device->timestampOverrideUsec;
                            if (iev.type == EV_SYN && iev.code == SYN_REPORT) {
                                device->timestampOverrideSec = 0;
                                device->timestampOverrideUsec = 0;
                            }
                            ALOGV("applied override time %d.%06d",
                                    int(iev.time.tv_sec), int(iev.time.tv_usec));
                        }

#ifdef HAVE_POSIX_CLOCKS
                        // Use the time specified in the event instead of the current time
                        // so that downstream code can get more accurate estimates of
                        // event dispatch latency from the time the event is enqueued onto
                        // the evdev client buffer.
                        //
                        // The event's timestamp fortuitously uses the same monotonic clock
                        // time base as the rest of Android.  The kernel event device driver
                        // (drivers/input/evdev.c) obtains timestamps using ktime_get_ts().
                        // The systemTime(SYSTEM_TIME_MONOTONIC) function we use everywhere
                        // calls clock_gettime(CLOCK_MONOTONIC) which is implemented as a
                        // system call that also queries ktime_get_ts().
                        event->when = nsecs_t(iev.time.tv_sec) * 1000000000LL
                                + nsecs_t(iev.time.tv_usec) * 1000LL;
                        ALOGV("event time %" PRId64 ", now %" PRId64, event->when, now);

                        // Bug 7291243: Add a guard in case the kernel generates timestamps
                        // that appear to be far into the future because they were generated
                        // using the wrong clock source.
                        //
                        // This can happen because when the input device is initially opened
                        // it has a default clock source of CLOCK_REALTIME.  Any input events
                        // enqueued right after the device is opened will have timestamps
                        // generated using CLOCK_REALTIME.  We later set the clock source
                        // to CLOCK_MONOTONIC but it is already too late.
                        //
                        // Invalid input event timestamps can result in ANRs, crashes and
                        // and other issues that are hard to track down.  We must not let them
                        // propagate through the system.
                        //
                        // Log a warning so that we notice the problem and recover gracefully.
                        if (event->when >= now + 10 * 1000000000LL) {
                            // Double-check.  Time may have moved on.
                            nsecs_t time = systemTime(SYSTEM_TIME_MONOTONIC);
                            if (event->when > time) {
                                ALOGW("An input event from %s has a timestamp that appears to "
                                        "have been generated using the wrong clock source "
                                        "(expected CLOCK_MONOTONIC): "
                                        "event time %" PRId64 ", current time %" PRId64
                                        ", call time %" PRId64 ".  "
                                        "Using current time instead.",
                                        device->path.string(), event->when, time, now);
                                event->when = time;
                            } else {
                                ALOGV("Event time is ok but failed the fast path and required "
                                        "an extra call to systemTime: "
                                        "event time %" PRId64 ", current time %" PRId64
                                        ", call time %" PRId64 ".",
                                        event->when, time, now);
                            }
                        }
#else
                        event->when = now;
#endif
                        event->deviceId = deviceId;
                        event->type = iev.type;
                        event->code = iev.code;
                        event->value = iev.value;
                        event += 1;
                        capacity -= 1;
                    }
                    if (capacity == 0) {
                        // The result buffer is full.  Reset the pending event index
                        // so we will try to read the device again on the next iteration.
                        mPendingEventIndex -= 1;
                        break;
                    }
                }
            } else if (eventItem.events & EPOLLHUP) {
                ALOGI("Removing device %s due to epoll hang-up event.",
                        device->identifier.name.string());
                deviceChanged = true;
                closeDeviceLocked(device);
            } else {
                ALOGW("Received unexpected epoll event 0x%08x for device %s.",
                        eventItem.events, device->identifier.name.string());
            }
        }

        // readNotify() will modify the list of devices so this must be done after
        // processing all other events to ensure that we read all remaining events
        // before closing the devices.
        if (mPendingINotify && mPendingEventIndex >= mPendingEventCount) {
            mPendingINotify = false;
            readNotifyLocked();
            deviceChanged = true;
        }

        // Report added or removed devices immediately.
        if (deviceChanged) {
            continue;
        }

        // Return now if we have collected any events or if we were explicitly awoken.
        if (event != buffer || awoken) {
            break;
        }

        // Poll for events.  Mind the wake lock dance!
        // We hold a wake lock at all times except during epoll_wait().  This works due to some
        // subtle choreography.  When a device driver has pending (unread) events, it acquires
        // a kernel wake lock.  However, once the last pending event has been read, the device
        // driver will release the kernel wake lock.  To prevent the system from going to sleep
        // when this happens, the EventHub holds onto its own user wake lock while the client
        // is processing events.  Thus the system can only sleep if there are no events
        // pending or currently being processed.
        //
        // The timeout is advisory only.  If the device is asleep, it will not wake just to
        // service the timeout.
        mPendingEventIndex = 0;

        mLock.unlock(); // release lock before poll, must be before release_wake_lock
        release_wake_lock(WAKE_LOCK_ID);

        int pollResult = epoll_wait(mEpollFd, mPendingEventItems, EPOLL_MAX_EVENTS, timeoutMillis);

        acquire_wake_lock(PARTIAL_WAKE_LOCK, WAKE_LOCK_ID);
        mLock.lock(); // reacquire lock after poll, must be after acquire_wake_lock

        if (pollResult == 0) {
            // Timed out.
            mPendingEventCount = 0;
            break;
        }

        if (pollResult < 0) {
            // An error occurred.
            mPendingEventCount = 0;

            // Sleep after errors to avoid locking up the system.
            // Hopefully the error is transient.
            if (errno != EINTR) {
                ALOGW("poll failed (errno=%d)\n", errno);
                usleep(100000);
            }
        } else {
            // Some events occurred.
            mPendingEventCount = size_t(pollResult);
        }
    }

    // All done, return the number of events we read.
    return event - buffer;
}

这个函数很大，很长，挺复杂的，但它的确就是直接读写/dev/input/设备文件节点的函数：

int pollResult = epoll_wait(mEpollFd, mPendingEventItems, EPOLL_MAX_EVENTS, timeoutMillis);

这里监听有没有事件到来，没有就休眠，有的话就返回......

至此，文章一开始提出的问题就得到了解决，问题是：”哪个线程一直监听着事件的到来，是怎么监听的？“

答案显然是：InputReaderThread线程一直监听着事件的到来，最终使用epoll_wait监听这些/dev/input/下的文件节点的文件描述符。

得到了这个答案后，再结合上一节得到的结论，事件的输入监听部分基本框架弄明白了，那下一步，就是继续分析事件的分发，毕竟事件最终要分发道view或着activity这些ui组件上。