1. 前言

Wakeup count是Wakeup events framework的组成部分，用于解决“system suspend和system wakeup events之间的同步问题”。本文将结合“Linux电源管理(6)_Generic PM之Suspend功能”和“Linux电源管理(7)_Wakeup events framework”两篇文章，分析wakeup count的功能、实现逻辑、背后的思考，同时也是对这两篇文章的复习和总结。

2. wakeup count在电源管理中的位置

wakeup count的实现位于wakeup events framework中（drivers/base/power/wakeup.c），主要为两个模块提供接口：通过PM core向用户空间提供sysfs接口；直接向autosleep（请参考下一篇文章）提供接口。

3. wakeup count的功能

wakeup count的功能是suspend同步，实现思路是这样的： 1）任何想发起电源状态切换的实体（可以是用户空间电源管理进程，也可以是内核线程，简称C），在发起状态切换前，读取系统的wakeup counts（该值记录了当前的wakeup event总数），并将读取的counts告知wakeup events framework。 2）wakeup events framework记录该counts到一个全局变量中（saved_count）。 3）随后C发起电源状态切换（如STR），执行suspend过程。 4）在suspend的过程中，wakeup events framework照旧工作（直到系统中断被关闭），上报wakeup events，增加wakeup events counts。 5）suspend执行的一些时间点（可参考“Linux电源管理(6)_Generic PM之Suspend功能”），会调用wakeup  events framework提供的接口（pm_wakeup_pending），检查是否有wakeup没有处理。 6）检查逻辑很简单，就是比较当前的wakeup counts和saved wakeup counts（C发起电源状态切换时的counts），如果不同，就要终止suspend过程。

4. wakeup count的实现逻辑

4.1 一个例子

在进行代码分析之前，我们先用伪代码的形式，写出一个利用wakeup count进行suspend操作的例子，然后基于该例子，分析相关的实现。 1: do { 2: ret = read(&cnt, "/sys/power/wakeup_count"); 3: if (ret) { 4: ret = write(cnt, "/sys/power/wakeup_count"); 5: } else { 6: countine; 7: } 8: } while (!ret); 9:  10: write("mem", "/sys/power/state"); 11:  12: /* goto here after wakeup */

例子很简单： a）读取wakeup count值，如果成功，将读取的值回写。否则说明有正在处理的wakeup events，continue。 b）回写后，判断返回值是否成功，如果不成功（说明读、写的过程中产生了wakeup events），继续读、写，直到成功。成功后，可以触发电源状态切换。

4.2 /sys/power/wakeup_count

wakeup_count文件是在kernel/power/main.c中，利用power_attr注册的，如下（大家可以仔细研读一下那一大段注释，内核很多注释写的非常好，而好的注释，就是软件功力的体现）： 1: /* 2: * The 'wakeup_count' attribute, along with the functions defined in 3: * drivers/base/power/wakeup.c, provides a means by which wakeup events can be 4: * handled in a non-racy way. 5: * 6: * If a wakeup event occurs when the system is in a sleep state, it simply is 7: * woken up. In turn, if an event that would wake the system up from a sleep 8: * state occurs when it is undergoing a transition to that sleep state, the 9: * transition should be aborted. Moreover, if such an event occurs when the 10: * system is in the working state, an attempt to start a transition to the 11: * given sleep state should fail during certain period after the detection of 12: * the event. Using the 'state' attribute alone is not sufficient to satisfy 13: * these requirements, because a wakeup event may occur exactly when 'state' 14: * is being written to and may be delivered to user space right before it is 15: * frozen, so the event will remain only partially processed until the system is 16: * woken up by another event. In particular, it won't cause the transition to 17: * a sleep state to be aborted. 18: * 19: * This difficulty may be overcome if user space uses 'wakeup_count' before 20: * writing to 'state'. It first should read from 'wakeup_count' and store 21: * the read value. Then, after carrying out its own preparations for the system 22: * transition to a sleep state, it should write the stored value to 23: * 'wakeup_count'. If that fails, at least one wakeup event has occurred since 24: * 'wakeup_count' was read and 'state' should not be written to. Otherwise, it 25: * is allowed to write to 'state', but the transition will be aborted if there 26: * are any wakeup events detected after 'wakeup_count' was written to. 27: */ 28:  29: static ssize_t wakeup_count_show(struct kobject *kobj, 30: struct kobj_attribute *attr, 31: char *buf) 32: { 33: unsigned int val; 34:  35: return pm_get_wakeup_count(&val, true) ? 36: sprintf(buf, "%u ", val) : -EINTR; 37: } 38:  39: static ssize_t wakeup_count_store(struct kobject *kobj, 40: struct kobj_attribute *attr, 41: const char *buf, size_t n) 42: { 43: unsigned int val; 44: int error; 45:  46: error = pm_autosleep_lock(); 47: if (error) 48: return error; 49:  50: if (pm_autosleep_state() > PM_SUSPEND_ON) { 51: error = -EBUSY; 52: goto out; 53: } 54:  55: error = -EINVAL; 56: if (sscanf(buf, "%u", &val) == 1) { 57: if (pm_save_wakeup_count(val)) 58: error = n; 59: } 60:  61: out: 62: pm_autosleep_unlock(); 63: return error; 64: } 65:  66: tr(wakeup_count); 实现很简单：read时，直接调用pm_get_wakeup_count（注意第2个参数）；write时，直接调用pm_save_wakeup_count（注意用户空间的wakeup count功能和auto sleep互斥，会在下篇文章解释原因）。这两个接口均是wakeup events framework提供的接口，跟着代码往下看吧。

4.3 pm_get_wakeup_count

pm_get_wakeup_count的实现如下： 1: bool pm_get_wakeup_count(unsigned int *count, bool block) 2: { 3: unsigned int cnt, inpr; 4:  5: if (block) { 6: DEFINE_WAIT(wait); 7:  8: for (;;) { 9: prepare_to_wait(&wakeup_count_wait_queue, &wait, 10: TASK_INTERRUPTIBLE); 11: split_counters(&cnt, &inpr); 12: if (inpr == 0 || signal_pending(current)) 13: break; 14:  15: schedule(); 16: } 17: finish_wait(&wakeup_count_wait_queue, &wait); 18: } 19:  20: split_counters(&cnt, &inpr); 21: *count = cnt; 22: return !inpr; 23: }

该接口有两个参数，一个是保存返回的count值得指针，另一个指示是否block，具体请参考代码逻辑： a）如果block为false，直接读取registered wakeup events和wakeup events in progress两个counter值，将registered wakeup events交给第一个参数，并返回wakeup events in progress的状态（若返回false，说明当前有wakeup events正在处理，不适合suspend）。 b）如果block为true，定义一个等待队列，等待wakeup events in progress为0，再返回counter。   注1：由4.2小节可知，sysfs发起的read动作，block为true，所以如果有正在处理的wakeup events，read进程会阻塞。其它模块（如auto sleep）发起的read，则可能不需要阻塞。

4.4 pm_save_wakeup_count

pm_save_wakeup_count的实现如下： 1: bool pm_save_wakeup_count(unsigned int count) 2: { 3: unsigned int cnt, inpr; 4: unsigned long flags; 5:  6: events_check_enabled = false; 7: spin_lock_irqsave(&events_lock, flags); 8: split_counters(&cnt, &inpr); 9: if (cnt == count && inpr == 0) { 10: saved_count = count; 11: events_check_enabled = true; 12: } 13: spin_unlock_irqrestore(&events_lock, flags); 14: return events_check_enabled; 15: }

1）注意这个变量，events_check_enabled，如果它不为真，pm_wakeup_pending接口直接返回false，意味着如果不利用wakeup count功能，suspend过程中不会做任何wakeup events检查，也就不会进行任何的同步。 2）解除当前的registered wakeup events、wakeup events in progress，保存在变量cnt和inpr中。 3）如果写入的值和cnt不同（说明读、写的过程中产生events），或者inpr不为零（说明有events正在被处理），返回false（说明此时不宜suspend）。 4）否则，events_check_enabled置位（后续的pm_wakeup_pending才会干活），返回true（可以suspend），并将当前的wakeup count保存在saved count变量中。

4.5 /sys/power/state

再回忆一下“Linux电源管理(6)_Generic PM之Suspend功能”中suspend的流程，在suspend_enter接口中，suspend前的最后一刻，会调用pm_wakeup_pending接口，代码如下： 1: static int suspend_enter(suspend_state_t state, bool *wakeup) 2: { 3: ... 4: error = syscore_suspend(); 5: if (!error) { 6: *wakeup = pm_wakeup_pending(); 7: if (!(suspend_test(TEST_CORE) || *wakeup)) { 8: error = suspend_ops->enter(state); 9: events_check_enabled = false; 10: } 11: syscore_resume(); 12: } 13: ... 14: } 在write wakeup_count到调用pm_wakeup_pending这一段时间内，wakeup events framework会照常产生wakeup events，因此如果pending返回true，则不能“enter”，终止suspend吧！ 注2：wakeup后，会清除events_check_enabled标记。 “Linux电源管理(7)_Wakeup events framework”中已经介绍过pm_wakeup_pending了，让我们再看一遍吧： 1: bool pm_wakeup_pending(void) 2: { 3: unsigned long flags; 4: bool ret = false; 5:  6: spin_lock_irqsave(&events_lock, flags); 7: if (events_check_enabled) { 8: unsigned int cnt, inpr; 9:  10: split_counters(&cnt, &inpr); 11: ret = (cnt != saved_count || inpr > 0); 12: events_check_enabled = !ret; 13: } 14: spin_unlock_irqrestore(&events_lock, flags); 15:  16: if (ret) 17: print_active_wakeup_sources(); 18:  19: return ret; 20: }

a）首先会判断events_check_enabled是否有效，无效直接返回false。有效的话： b）获得cnt和inpr，如果cnt不等于saved_count（说明这段时间内有events产生），或者inpr不为0（说明有events正在被处理），返回true（告诉调用者，放弃吧，时机不到）。同时清除events_check_enabled的状态。 c）否则，返回false（放心睡吧），同时保持events_check_enabled的置位状态（以免pm_wakeup_pending再次调用）。

Okay，结束了，等待wakeup吧~~~~