85      阿里云  技术社区[云栖]

时间子系统17_hard lockup机制

//	使能hard lockup探测
//	调用路径：watchdog_enable->watchdog_nmi_enable
//	函数任务：
//		1.初始化hard lockup检测事件
//			2.hard lockup阈值为10s
//		2.向performance monitoring子系统注册hard lockup检测事件
//		3.使能hard lockup检测事件
//	注：
//		performance monitoring，x86中的硬件设备，当cpu clock经过了指定个周期后发出一个NMI中断。
1.1 static int watchdog_nmi_enable(unsigned int cpu)
{
	//hard lockup事件
	struct perf_event_attr *wd_attr;
	struct perf_event *event = per_cpu(watchdog_ev, cpu);
	....
	wd_attr = &wd_hw_attr;
	//hard lockup检测周期，10s
	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
	//向performance monitoring注册hard lockup检测事件
	event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL);
	....
	//使能hard lockup的检测
	per_cpu(watchdog_ev, cpu) = event;
	perf_event_enable(per_cpu(watchdog_ev, cpu));
	return 0;
}

//	换算hard lockup检测周期到cpu频率
1.2 u64 hw_nmi_get_sample_period(int watchdog_thresh)
{
	return (u64)(cpu_khz) * 1000 * watchdog_thresh;
}

//	hard lockup检测事件发生时的nmi回调函数
//	函数任务：
//		1.判断是否发生了hard lockup
//			1.1 dump hard lockup信息
1.3 static void watchdog_overflow_callback(struct perf_event *event,
		 struct perf_sample_data *data,
		 struct pt_regs *regs)
{
	//判断是否发生hard lockup
	if (is_hardlockup()) {
		int this_cpu = smp_processor_id();

		//打印hard lockup信息
		if (hardlockup_panic)
			panic("Watchdog detected hard LOCKUP on cpu %d", this_cpu);
		else
			WARN(1, "Watchdog detected hard LOCKUP on cpu %d", this_cpu);

		return;
	}
	return;
}

//	判断是否发生hard lockup
//	注：
//		如果时钟中断在指定阈值范围内为运行，核心认为可屏蔽中断被屏蔽时间过长
1.4 static int is_hardlockup(void)
{
	//获取watchdog timer的运行次数
	unsigned long hrint = __this_cpu_read(hrtimer_interrupts);
	//在一个hard lockup检测时间阈值内，如果watchdog timer未运行，说明cpu中断被屏蔽时间超过阈值
	if (__this_cpu_read(hrtimer_interrupts_saved) == hrint)
		return 1;
	//记录watchdog timer运行的次数
	__this_cpu_write(hrtimer_interrupts_saved, hrint);
	return 0;
}

//	关闭hard lockup检测机制
//	函数任务：
//		1.向performance monitoring子系统注销hard lockup检测控制块
//		2.清空per-cpu hard lockup检测控制块
//		3.释放hard lock检测控制块
2.1 static void watchdog_nmi_disable(unsigned int cpu)
{
	struct perf_event *event = per_cpu(watchdog_ev, cpu);
	if (event) {
		//向performance monitoring子系统注销hard lockup检测控制块
		perf_event_disable(event);
		//清空per-cpu hard lockup检测控制块
		per_cpu(watchdog_ev, cpu) = NULL;
		//释放hard lock检测控制块
		perf_event_release_kernel(event);
	}
	return;
}

最后更新：2017-04-03 14:54:38

  上一篇： leetcode难度及面试频率

  下一篇： 时间子系统16_soft lockup机制