Android7.0 PowerManagerService(2) WakeLock的使用及流程

作为移动终端，电量是一种稀缺资源，需要尽可能的节省。于是，Android系统在空闲时，会主动进入到休眠状态。

我们知道整个Android系统中运行着很多个进程，因此必须有一种机制能够知道每个进程是否正在进行重要的工作，只有这样Android系统才能对整个终端当前的状态做出判断。

显然我们不能启动一个进程，去主动监管其它所有进程的工作状态，这样CPU开销太大，反而加剧了电量的消耗。为此Android引入了基于WakeLock的电量管理机制，而PMS就是专门负责管理WakeLock的进程。

个人觉得WakeLock机制的思想，有点类似于早期通信领域局域网中的令牌环机制。当局域网中有设备需要发送数据时，需要申请令牌(Token)，申请到令牌才能发送数据；设备发送完数据后，再释放掉令牌。

与此相似，Android设备中运行的进程需要使用电量资源时，也需要向PMS申请一个WakeLock；当工作完成后，就释放掉申请的WakeLock。PMS通过判断当前是否还有进程持有WakeLock，就能得出系统是否空闲的结论。

从这里也可以看出，当我们写一个永不停止工作的线程，但不申请WakeLock时，系统仍然可以休眠。在休眠时，CPU不会再耗费资源去调度该线程，于是“永不停止工作”被打上了引号。

接下来，我们就来看看PMS中的WakeLock。

一、创建WakeLock

我们以RIL.java为例，看看一般情况下，PMS以外的其它进程如何使用WakeLock。

在RIL.java的构造函数中：

public RIL(Context context, int preferredNetworkType,
int cdmaSubscription, Integer instanceId) {
.............
PowerManager pm = (PowerManager)context.getSystemService(Context.POWER_SERVICE);
//获取WakeLock，第一个参数决定了WakeLock的等级和flag
mWakeLock = pm.newWakeLock(PowerManager.PARTIAL_WAKE_LOCK, RILJ_LOG_TAG);

//默认WakeLocked会ReferenceCounted，即一次申请对应一次释放
//设为false后，一次释放就可以对应所有的申请
mWakeLock.setReferenceCounted(false);
...........
//RIL.java中自己维护了WakeLockCount
mWakeLockCount = 0;
...........
}

从上面的代码，可以看出调用PowerManager的newWakeLock函数，可以创建出WakeLock。

我们看看newWakeLock函数定义：

public WakeLock newWakeLock(int levelAndFlags, String tag) {
//检查参数有效性，即levelAndFlags必须对应于PowerManager中定义的WakeLock级别和flag，tag不能为空
validateWakeLockParameters(levelAndFlags, tag);

//此WakeLock为PowerManager定义的内部类
return new WakeLock(levelAndFlags, tag, mContext.getOpPackageName());
}

1、WakeLock Level

newWakeLock中的第一个参数对应于WakeLock的级别和标志位构成的位图。

目前，在PowerManger中一共为WakeLock定义了7种level。

/**
* Wake lock level: Ensures that the CPU is running; the screen and keyboard
* backlight will be allowed to go off.
*
* If the user presses the power button, then the screen will be turned off
* but the CPU will be kept on until all partial wake locks have been released.
* /
public static final int PARTIAL_WAKE_LOCK = 0x00000001;

/**
* Wake lock level: Ensures that the screen is on (but may be dimmed);
* the keyboard backlight will be allowed to go off.
*
* If the user presses the power button, then the SCREEN_DIM_WAKE_LOCK will be
* implicitly released by the system, causing both the screen and the CPU to be turned off.
*/
@Deprecated
public static final int SCREEN_DIM_WAKE_LOCK = 0x00000006;

/**
* Wake lock level: Ensures that the screen is on at full brightness;
* the keyboard backlight will be allowed to go off.
*
*If the user presses the power button, then the SCREEN_BRIGHT_WAKE_LOCK will be
* implicitly released by the system, causing both the screen and the CPU to be turned off.
*/
@Deprecated
public static final int SCREEN_BRIGHT_WAKE_LOCK = 0x0000000a;

/**
* Wake lock level: Ensures that the screen and keyboard backlight are on at
* full brightness.
*
*If the user presses the power button, then the FULL_WAKE_LOCK will be
* implicitly released by the system, causing both the screen and the CPU to be turned off.
*/
@Deprecated
public static final int FULL_WAKE_LOCK = 0x0000001a;

/**
* Wake lock level: Turns the screen off when the proximity sensor activates.
* If the proximity sensor detects that an object is nearby, the screen turns off
* immediately.  Shortly after the object moves away, the screen turns on again.
*
* A proximity wake lock does not prevent the device from falling asleep
* unlike link FULL_WAKE_LOCK, SCREEN_BRIGHT_WAKE_LOCK and SCREEN_DIM_WAKE_LOCK.
* If there is no user activity and no other wake locks are held, then the device will fall asleep (and lock) as usual.
* However, the device will not fall asleep while the screen has been turned off
* by the proximity sensor because it effectively counts as ongoing user activity.
*
* Cannot be used with ACQUIRE_CAUSES_WAKEUP (WakeLock的flag).
*/
//例如拨号，打通后接听电话，屏幕变黑
public static final int PROXIMITY_SCREEN_OFF_WAKE_LOCK = 0x00000020;

/**
* Wake lock level: Put the screen in a low power state and allow the CPU to suspend
* if no other wake locks are held.
*
* This is used by the dream manager to implement doze mode.  It currently
* has no effect unless the power manager is in the dozing state.
* /
public static final int DOZE_WAKE_LOCK = 0x00000040;

/**
* Wake lock level: Keep the device awake enough to allow drawing to occur.
*
* This is used by the window manager to allow applications to draw while the
* system is dozing.  It currently has no effect unless the power manager is in
* the dozing state.
* /
public static final int DRAW_WAKE_LOCK = 0x00000080;

从上面的代码注释可以看出，WakeLock主要用于控制CPU、屏幕和键盘三大部分（当然，现在的Anroid中基本没有键盘了）。

对于PARTIAL_WAKE_LOCK、SCREEN_DIM_WAKE_LOCK、SCREEN_BRIGHT_WAKE_LOCK和FULL_WAKE_LOCK而言，不考虑Power键的话，随着等级的提高，权限也相应增大，即持有高等级的锁，能够激活的部分越多；如果考虑Power键的话，PARTIAL_WAKE_LOCK可以保证CPU不休眠，反而是权限最大的。

PROXIMITY_SCREEN_OFF_WAKE_LOCK、DOZE_WAKE_LOCK和DRAW_WAKE_LOCK都是和具体场景相关的锁。

2、WakeLock Flag

PowerManager定义的WakeLock Flag很多，无法一一列举，就看一下比较常用的：

/**
* Wake lock flag: Turn the screen on when the wake lock is acquired.
*
* Normally wake locks don't actually wake the device, they just cause
* the screen to remain on once it's already on.  Think of the video player
* application as the normal behavior.  Notifications that pop up and want
* the device to be on are the exception; use this flag to be like them.
*
* Cannot be used with PARTIAL_WAKE_LOCK.
* /
public static final int ACQUIRE_CAUSES_WAKEUP = 0x10000000;

/**
* Wake lock flag: When this wake lock is released, poke the user activity timer
* so the screen stays on for a little longer.
*
* Will not turn the screen on if it is not already on.
*
* Cannot be used with PARTIAL_WAKE_LOCK.
* /
public static final int ON_AFTER_RELEASE = 0x20000000;

..................

总结一下上面的内容，如下所示：



WakeLock Flag一般与WakeLock Level组合使用，使用的时候参照一下注释即可。

3、WakeLock的构造函数

最后，我们来看看WakeLock的构造函数：

WakeLock(int flags, String tag, String packageName) {
//level and flag
mFlags = flags;

//创建类对应的打印Tag
mTag = tag;

//创建类的类名
mPackageName = packageName;

//创建一个Binder对象
//PMS将作为该Binder的客户端监听对应进程是否死亡
mToken = new Binder();
mTraceName = "WakeLock (" + mTag + ")";
}

WakeLock的构造函数中需要注意的地方是，创建了一个Binder对象。

回忆一下RIL.java中创建WakeLock的过程，我们就知道这个Binder对象应该是创建在RIL.java所在的Phone进程中。

二、Acquire WakeLock

从上面的分析，我们知道一个进程创建的WakeLock，实际上表明了该进程执行某个工作时对电量的需求，例如声明该工作需要保持屏幕处于点亮状态，或该工作需要CPU处于唤醒态等。

因此，进程创建了WakeLock后，需要将WakeLock发送到PMS中，让PMS明白该进程的需求。

这种将WakeLock通知到PMS的过程，就被称为acquire WakeLock。

同样，我们还是以RIL.java中的使用过程举例：

private void send(RILRequest rr) {
Message msg;

if (mSocket == null) {
rr.onError(RADIO_NOT_AVAILABLE, null);
rr.release();
return;
}

msg = mSender.obtainMessage(EVENT_SEND, rr);
//重点在这里
acquireWakeLock(rr, FOR_WAKELOCK);
msg.sendToTarget();
}

当AP侧向modem发送请求时，将要调用RIL.java的send函数。send函数将会发送消息给RILSender，后者利用socket将消息发送给rild进程。

从上面的代码可以看出，在发送消息给RILSender之前，调用了acquireWakeLock函数：

private void acquireWakeLock(RILRequest rr, int wakeLockType) {
synchronized(rr) {
.............
switch(wakeLockType) {
case FOR_WAKELOCK:
synchronized (mWakeLock) {
//调用acquire函数
mWakeLock.acquire();
mWakeLockCount++;
mWlSequenceNum++;
............
}
break;
.........
}
rr.mWakeLockType = wakeLockType;
}
}

我们跟进一下PowerManager中WakeLock的acquire函数：

public void acquire() {
synchronized (mToken) {
acquireLocked();
}
}

private void acquireLocked() {
//前面已经提过，RIL.java中已经将mRefCounted置为false
//如果不将mRefCounted置为false，意味着acquire和release必须一一对应
//那么每个WakeLock只能acquire一次
if (!mRefCounted || mCount++ == 0) {
........
try {
mService.acquireWakeLock(mToken, mFlags, mTag, mPackageName, mWorkSource,
mHistoryTag);
} catch (RemoteException e) {
throw e.rethrowFromSystemServer();
}
mHeld = true;
}
}

容易看出实际的工作流程将通过Binder通信进入到PMS中：

public void acquireWakeLock(IBinder lock, int flags, String tag, String packageName,
WorkSource ws, String historyTag) {
//参数和权限检查
............
final int uid = Binder.getCallingUid();
final int pid = Binder.getCallingPid();
final long ident = Binder.clearCallingIdentity();
try {
acquireWakeLockInternal(lock, flags, tag, packageName, ws, historyTag, uid, pid);
} finally {
Binder.restoreCallingIdentity(ident);
}
}

private void acquireWakeLockInternal(IBinder lock, int flags, String tag, String packageName,
WorkSource ws, String historyTag, int uid, int pid) {
synchronized (mLock) {
...........
//PMS中也定义了WakeLock内部类
WakeLock wakeLock;

//PMS中维持了一个ArrayList，记录当前已申请的WakeLock
//findWakeLockIndexLocked查找ArrayList，判断参数对应的WakeLock，是否在之前被申请过
int index = findWakeLockIndexLocked(lock);
boolean notifyAcquire;
if (index >= 0) {
//如果index大于0，说明此时Acquire的是一个旧的WakeLock
//例如RIL会多次调用send函数，于是除第一次外，都会进入这个分支
wakeLock = mWakeLocks.get(index);

//这是判断WakeLock对应的成员变量是否发生改变
if (!wakeLock.hasSameProperties(flags, tag, ws, uid, pid)) {
// Update existing wake lock.  This shouldn't happen but is harmless.
notifyWakeLockChangingLocked(wakeLock, flags, tag, packageName,
uid, pid, ws, historyTag);
//若wakelock属性发生了变化，更新该属性
wakeLock.updateProperties(flags, tag, packageName, ws, historyTag, uid, pid);
}
notifyAcquire = false;
} else {
//创建一个新的WakeLock，例如RIL第一次调用send就会进入该分支
wakeLock = new WakeLock(lock, flags, tag, packageName, ws, historyTag, uid, pid);
try {
//1、监控申请WakeLock的进程是否死亡
lock.linkToDeath(wakeLock, 0);
} catch (RemoteException ex) {
throw new IllegalArgumentException("Wake lock is already dead.");
}
//添加到wakelock列表
mWakeLocks.add(wakeLock);

//2、特殊处理PARTIAL_WAKE_LOCK
//实际上，根据Doze模式的白名单更新wakelock的disabled变量
setWakeLockDisabledStateLocked(wakeLock);
notifyAcquire = true;
}

//3、处理WakeLock对应的Flag
//实际上判断WakeLock是否有ACQUIRE_CAUSES_WAKEUP，在必要时唤醒屏幕
applyWakeLockFlagsOnAcquireLocked(wakeLock, uid);
mDirty |= DIRTY_WAKE_LOCKS;

//更新电源状态，以后单独分析
updatePowerStateLocked();
if (notifyAcquire) {
// This needs to be done last so we are sure we have acquired the
// kernel wake lock.  Otherwise we have a race where the system may
// go to sleep between the time we start the accounting in battery
// stats and when we actually get around to telling the kernel to
// stay awake.
//通知wakeLock发生变化
//电量统计服务做相关统计
notifyWakeLockAcquiredLocked(wakeLock);
}
}
}

如上代码中标注的注释，acquireWakeLockInternal中有几处比较重要的地方，我们一起来分析一下。

1、监听客户端进程死亡

上面的代码中，第一次创建WakeLock后，调用了：

.........
lock.linkToDeath(wakeLock, 0);
.........

我们将acquire WakeLock的进程定义为PMS的客户端进程，那么上面代码的lock，就是客户端进程中创建的Binder对象的代理。对于RIL而言，就是存在于Phone进程中的Binder的代理。

PMS调用Binder代理的linkToDeath，实际上使得PMS成为了对应进程Binder的客户端。于是，当对应进程死亡后，将通知PMS。

linkToDeath传入的必须是继承IBinder.DeathRecipient的对象，作为进程死亡的”讣告”接收者。

我们看看PMS中WakeLock与此相关的定义：

private final class WakeLock implements IBinder.DeathRecipient {
...........
@Override
public void binderDied() {
//发现客户端进程死亡后，调用PMS的handleWakeLockDeath进行处理，传入的参数为WakeLock自己
PowerManagerService.this.handleWakeLockDeath(this);
}
.......
}

我们看看PMS的handleWakeLockDeath函数：

private void handleWakeLockDeath(WakeLock wakeLock) {
synchronized (mLock) {
..........
int index = mWakeLocks.indexOf(wakeLock);
if (index < 0) {
return;
}

removeWakeLockLocked(wakeLock, index);
}
}

跟进removeWakeLockLocked函数：

private void removeWakeLockLocked(WakeLock wakeLock, int index) {
mWakeLocks.remove(index);
//通知到BatteryStatsService
notifyWakeLockReleasedLocked(wakeLock);

//处理WakeLock对应的flag，与后文applyWakeLockFlagsOnAcquireLocked一起分析
//实际上是判断是否需要立即息屏
applyWakeLockFlagsOnReleaseLocked(wakeLock);
mDirty |= DIRTY_WAKE_LOCKS;
//锁移除后，还是利用updatePowerStateLocked更新电源状态
updatePowerStateLocked();
}

2、特殊处理PARTIAL_WAKE_LOCK

PMS处理第一次创建的WakeLock时，还会调用setWakeLockDisabledStateLocked函数进行处理：

private boolean setWakeLockDisabledStateLocked(WakeLock wakeLock) {
//仅会特殊处理PARTIAL_WAKE_LOCK，毕竟PARTIAL_WAKE_LOCK要求按Power键后CPU依然可以工作
if ((wakeLock.mFlags & PowerManager.WAKE_LOCK_LEVEL_MASK)
== PowerManager.PARTIAL_WAKE_LOCK) {
boolean disabled = false;

//设备处于Doze定义的device idle模式时
if (mDeviceIdleMode) {
final int appid = UserHandle.getAppId(wakeLock.mOwnerUid);

// If we are in idle mode, we will ignore all partial wake locks that are
// for application uids that are not whitelisted.

//判断是否为非系统应用
if (appid >= Process.FIRST_APPLICATION_UID &&
//白名单search
Arrays.binarySearch(mDeviceIdleWhitelist, appid) < 0 &&
Arrays.binarySearch(mDeviceIdleTempWhitelist, appid) < 0 &&
//判断进程的类型
//ActivityManager中定义的数字最小的为：常驻的操作UI的系统进程
//因此大概可理解为：数字越大，对处理事件的时效性要求越低
mUidState.get(wakeLock.mOwnerUid,
ActivityManager.PROCESS_STATE_CACHED_EMPTY)
> ActivityManager.PROCESS_STATE_FOREGROUND_SERVICE) {
disabled = true;
}
}
if (wakeLock.mDisabled != disabled) {
wakeLock.mDisabled = disabled;
return true;
}
}
return false;
}

上面代码大致的含义就是：

在Android Doze模式下，当终端处于device Idle Mode时，

对于一个非系统应用而言，如果该应用不在系统定义的白名单中，

并且该应用所在进程的类型表明，该进程对事件处理的时效性要求不高，

那么即使该应用申请了PARTIAL_WAKE_LOCK，也不能阻止系统进入休眠状态。

有些设备商，为了优化系统的功耗，就修改了这个地方。

例如，有些系统应用其实也很耗电，因此可以去掉该函数中对非系统应用的限制，对系统应用也进行管控。

3、处理WakeLock对应的Flag

前面的代码已经提到，当acquire WakeLock时，将调用applyWakeLockFlagsOnAcquireLocked处理WakeLock对应的flag；

当由于进程死亡，释放WakeLock时，会调用applyWakeLockFlagsOnReleaseLocked处理WakeLock对应的flag。

从函数命名来看，这两个函数应该有相似的地方。

3.1 applyWakeLockFlagsOnAcquireLocked

我们先看看applyWakeLockFlagsOnAcquireLocked：

private void applyWakeLockFlagsOnAcquireLocked(WakeLock wakeLock, int uid) {
//仅处理ACQUIRE_CAUSES_WAKEUP flag，同时要求WakeLock的level是与screen有关的，
//即FULL_WAKE_LOCK、SCREEN_BRIGHT_WAKE_LOCK和SCREEN_DIM_WAKE_LOCK
if ((wakeLock.mFlags & PowerManager.ACQUIRE_CAUSES_WAKEUP) != 0
&& isScreenLock(wakeLock)) {
..............
wakeUpNoUpdateLocked(SystemClock.uptimeMillis(), wakeLock.mTag, opUid,
opPackageName, opUid);
}
}

private boolean wakeUpNoUpdateLocked(long eventTime, String reason, int reasonUid,
String opPackageName, int opUid) {
............
//不满足以下条件，没有唤醒屏幕的必要
if (eventTime < mLastSleepTime || mWakefulness == WAKEFULNESS_AWAKE
|| !mBootCompleted || !mSystemReady) {
return false;
}

try {
mLastWakeTime = eventTime;
//修改PMS的一些成员变量，并进行通知
//其中主要的是将mDirty变量的DIRTY_WAKEFULNESS位置为了1
//PMS根据mDirty的位信息管理电源状态，同时唤醒屏幕
setWakefulnessLocked(WAKEFULNESS_AWAKE, 0);

//通知给电源统计服务
mNotifier.onWakeUp(reason, reasonUid, opPackageName, opUid);

//调用userActivityNoUpdateLocked函数
userActivityNoUpdateLocked(
eventTime, PowerManager.USER_ACTIVITY_EVENT_OTHER, 0, reasonUid);
} .....
return true;
}

3.1.1 setWakefulnessLocked

我们看看唤醒屏幕相关的操作：

private void setWakefulnessLocked(int wakefulness, int reason) {
if (mWakefulness != wakefulness) {
mWakefulness = wakefulness;
mWakefulnessChanging = true;
mDirty |= DIRTY_WAKEFULNESS;
//定义于frameworks/base/services/core/java/com/android/server/power/Notifier.java中
mNotifier.onWakefulnessChangeStarted(wakefulness, reason);
}
}

public void onWakefulnessChangeStarted(final int wakefulness, int reason) {
final boolean interactive = PowerManagerInternal.isInteractive(wakefulness);
.......
// Tell the activity manager about changes in wakefulness, not just interactivity.
mHandler.post(new Runnable() {
@Override
public void run() {
mActivityManagerInternal.onWakefulnessChanged(wakefulness);
}
});

// Handle any early interactive state changes.
// Finish pending incomplete ones from a previous cycle.
if (mInteractive != interactive) {
// Finish up late behaviors if needed.
if (mInteractiveChanging) {
handleLateInteractiveChange();
}

// Start input as soon as we start waking up or going to sleep.
mInputManagerInternal.setInteractive(interactive);
mInputMethodManagerInternal.setInteractive(interactive);

// Notify battery stats.
try {
mBatteryStats.noteInteractive(interactive);
} catch (RemoteException ex) { }

// Handle early behaviors.
mInteractive = interactive;
mInteractiveChangeReason = reason;
mInteractiveChanging = true;
//重点在这个位置
handleEarlyInteractiveChange();
}
}

/**
* Handle early interactive state changes such as getting applications or the lock
* screen running and ready for the user to see (such as when turning on the screen).
*/
private void handleEarlyInteractiveChange() {
synchronized (mLock) {
if (mInteractive) {
// Waking up...
mHandler.post(new Runnable() {
@Override
public void run() {
EventLog.writeEvent(EventLogTags.POWER_SCREEN_STATE, 1, 0, 0, 0);
//mPolicy对应于PhoneWindowManager
mPolicy.startedWakingUp();
}
});

// Send interactive broadcast.
mPendingInteractiveState = INTERACTIVE_STATE_AWAKE;
mPendingWakeUpBroadcast = true;
updatePendingBroadcastLocked();
} else {
// Going to sleep...
// Tell the policy that we started going to sleep.
final int why = translateOffReason(mInteractiveChangeReason);
mHandler.post(new Runnable() {
@Override
public void run() {
mPolicy.startedGoingToSleep(why);
}
});
}
}
}

从上面的代码来看，应该是PhoneWindowManager完成亮屏前的初始化工作，然后回调到PowerManager的wakeUp函数。

整个过程还是比较复杂的，需要单独进行分析，此处不做进一步说明。

3.2 applyWakeLockFlagsOnReleaseLocked

现在我们再看看applyWakeLockFlagsOnReleaseLocked函数：

private void applyWakeLockFlagsOnReleaseLocked(WakeLock wakeLock) {
//仅处理ON_AFTER_RELEASE，同样要求WakeLock的level是与screen有关的
//ON_AFTER_RELEASE并不会立即息屏
if ((wakeLock.mFlags & PowerManager.ON_AFTER_RELEASE) != 0
&& isScreenLock(wakeLock)) {
userActivityNoUpdateLocked(SystemClock.uptimeMillis(),
PowerManager.USER_ACTIVITY_EVENT_OTHER,
PowerManager.USER_ACTIVITY_FLAG_NO_CHANGE_LIGHTS,
wakeLock.mOwnerUid);
}
}

可以看出applyWakeLockFlagsOnAcquireLocked和applyWakeLockFlagsOnReleaseLocked最后均会调用userActivityNoUpdateLocked函数，只是参数不同。

3.3 userActivityNoUpdateLocked

我们一起来看一下userActivityNoUpdateLocked：

private boolean userActivityNoUpdateLocked(long eventTime, int event, int flags, int uid) {
.............
//过时的事件不需要处理
if (eventTime < mLastSleepTime || eventTime < mLastWakeTime
|| !mBootCompleted || !mSystemReady) {
return false;
}

...........
try {
if (eventTime > mLastInteractivePowerHintTime) {
//调用native加载的动态库的powerHint函数，具体意义不是很清楚
powerHintInternal(POWER_HINT_INTERACTION, 0);
mLastInteractivePowerHintTime = eventTime;
}

//调用BatteryStatsService的noteUserActivity函数，看代码好像是做一些记录
mNotifier.onUserActivity(event, uid);

//根据参数信息修改mDirty的一些变量
.............
} finally {
........
}
}

从以上代码来看，acquire WakeLock将申请信息递交给PMS统一进行处理。

PMS根据WakeLock的level和flag，完成修改一些变量、通知BatteryStatsService等工作后，

最终还是依赖于updatePowerStateLocked函数来进行实际的电源状态更新操作。

PMS类中有很多***NoUpdateLocked()方法，这些方法都有一些共性，就是仅更新状态，不负责具体的执行。因为PMS中具体的执行逻辑都是在updatePowerStateLocked方法中。

上述acquire WakeLock主要的工作大致可以总结为下图：



三、释放WakeLock

当进程完成工作后，需要释放之前申请的WakeLock。我们同样以RIL.java中的操作为例：

private void processResponse (Parcel p) {
int type;

type = p.readInt();

if (type == RESPONSE_UNSOLICITED || type == RESPONSE_UNSOLICITED_ACK_EXP) {
...........
} else if (type == RESPONSE_SOLICITED || type == RESPONSE_SOLICITED_ACK_EXP) {
//处理请求对应的回复信息
RILRequest rr = processSolicited (p, type);

if (rr != null) {
if (type == RESPONSE_SOLICITED) {
//重点在这里
decrementWakeLock(rr);
}
rr.release();
return;
}
} else if (type == RESPONSE_SOLICITED_ACK) {
...........
}
}

我们跟进decrementWakeLock函数：

private void decrementWakeLock(RILRequest rr) {
synchronized(rr) {
switch(rr.mWakeLockType) {
case FOR_WAKELOCK:
synchronized (mWakeLock) {
//前面已经提到过，RIL.java多个请求复用同一个WakeLock
//并且利用mWakeLockCount记录复用的次数
//这么设计的目的是：RIL发送请求的数量非常多，复用WakeLock可以避免多次构造释放
//同时减少与PMS之间Binder通信的次数
if (mWakeLockCount > 1) {
mWakeLockCount--;
} else {
mWakeLockCount = 0;
//所有请求均得到了处理，调用PowerManager中WakeLock的release函数
mWakeLock.release();
}
}
break;
........
}
}
........
}

现在我们跟进PowerManager中WakeLock定义的release函数：

/**
* Releases the wake lock with flags to modify the release behavior.
*
* This method releases your claim to the CPU or screen being on.
* The screen may turn off shortly after you release the wake lock, or it may
* not if there are other wake locks still held.
*
*/
public void release(int flags) {
synchronized (mToken) {
if (!mRefCounted || --mCount == 0) {
mHandler.removeCallbacks(mReleaser);
if (mHeld) {
.......
try {
//还是会调用到PMS中的函数
mService.releaseWakeLock(mToken, flags);
} catch (RemoteException e) {
throw e.rethrowFromSystemServer();
}
mHeld = false;
}

}
....
}
}

最后一起来看看PMS中释放WakeLock的函数：

public void releaseWakeLock(IBinder lock, int flags) {
//参数和权限检查
.............
final long ident = Binder.clearCallingIdentity();
try {
releaseWakeLockInternal(lock, flags);
} finally {
Binder.restoreCallingIdentity(ident);
}
}

private void releaseWakeLockInternal(IBinder lock, int flags) {
synchronized (mLock) {
//根据Binder代理，从存储的ArrayList中找到对应WakeLock的序号
int index = findWakeLockIndexLocked(lock);
...........
WakeLock wakeLock = mWakeLocks.get(index);
...........
//RELEASE_FLAG_WAIT_FOR_NO_PROXIMITY，表示当sensor判断终端离物体较远时，
//才真正释放PROXIMITY_SCREEN_OFF_WAKE_LOCK等级的WakeLock
if ((flags & PowerManager.RELEASE_FLAG_WAIT_FOR_NO_PROXIMITY) != 0) {
mRequestWaitForNegativeProximity = true;
}

//PMS不再关注客户端进程是否死亡
wakeLock.mLock.unlinkToDeath(wakeLock, 0);
removeWakeLockLocked(wakeLock, index);
}
}

private void removeWakeLockLocked(WakeLock wakeLock, int index) {
mWakeLocks.remove(index);
//通知BatteryStatsService
notifyWakeLockReleasedLocked(wakeLock);

//之前分析过，会做一些记录信息等
applyWakeLockFlagsOnReleaseLocked(wakeLock);
mDirty |= DIRTY_WAKE_LOCKS;
//依然靠updatePowerStateLocked函数更新终端的电源状态
updatePowerStateLocked();
}

整个release的过程大致可以总结为下图：



四、总结

通过前面的分析，我们知道了向PMS申请电量的基本用法类似于：

........
//1、创建
PowerManager pm = (PowerManager)context.getSystemService(Context.POWER_SERVICE);
mWakeLock = pm.newWakeLock(PowerManager.PARTIAL_WAKE_LOCK, RILJ_LOG_TAG);
......
//2、acquire
mWakeLock.acquire();
.........
//3、release
mWakeLock.release();
...........

当申请发送到PMS后，PMS将针对WakeLock的level和flag信息进行一些处理。

无论是acquire还是release WakeLock，PMS最终将利用updatePowerStateLocked函数对终端的电源状态进行调整。

我们将单独分析一下PMS核心的updatePowerStateLocked函数。