小编点评

## Soft Interrupts in Linux Kernel This article describes the soft interrupt handling process in the Linux kernel, highlighting the two stages: **upper half** and **lower half**. **Upper Half:** * This stage focuses on handling interrupts quickly. * It operates in an **interrupt disabled mode**, handling tasks like I/O, file operations, and miscellaneous low-priority work. * It directly handles hardware requests, performing low-level tasks. **Lower Half:** * This stage is responsible for delaying the processing of tasks in the upper half. * It runs in an **inter-kernel thread** with a lower priority than the upper half. * It handles long-running tasks and submits them to the lower half. **Soft Interrupts vs. Hard Interrupts:** * Soft interrupts are triggered by **kernel events**, while hard interrupts are triggered by **hardware events**. * Examples of kernel events include network activity, file system operations, and device activity. * Examples of hardware events include interrupts from peripherals, such as the hard disk. **Monitoring Soft Interrupts:** * You can use the `proc/softirqs` file to monitor the current state of soft interrupts. * It provides information such as the type, number of running threads, and recent events. * Use the `sar` command to monitor network and system statistics. **Observing Network Traffic:** * Use `sar` with the `-n` option to capture network statistics. * The relevant fields include **rxpck/s** and **txpck/s** for received and transmitted bytes per second. * Check for high values in **rxkB/s** and **txkB/s** to identify potential bottlenecks. **Additional Information:** * Each CPU has its own dedicated soft interrupt thread. * The `ps` command shows the number of running threads and the name of the thread for each soft interrupt. * Use `tcpdump` to capture and analyze network traffic. By understanding the soft interrupt handling process, you can identify and diagnose potential issues related to network performance and system responsiveness.

正文

 

软中断：为了处理中断处理程序执行过长和中断丢失的问题，linux将中断处理过程分成两个阶段，也就是上半部和下半部：

• 上半部：用来快速处理中断，它在中断禁止模式下运行，主要处理跟硬件紧密相关的或跟时间敏感的工作，直接处理硬件请求，也就是我们常说的硬中断，特点是快速执行
• 下半部:用来延迟处理上半部未完成的工作，通常以内核线程的方式运行。由内核触发，也就是我们常说的软中断。

当然软中断不只包括了硬件设备中断处理程序的下半部，一些内核自定义的事件也属于软中断，比如内核调度和RCU锁等

查看软中断和内核线程

proc文件系统是一种内核空间和用户空间进行通信的机制，可以用来查看内核的数据结构，或者用来动态修改内核的配置，其中：
/proc/softirqs 提供了软中断的运行情况
/proc/interrupts 提供了硬中断的运行情况

watch -d cat /proc/softirqs

• 第一列显示软中断类别，总共有10个类别，分贝对应不同的工作类型；
• 要注意同一种软中断在不同的cpu上的分布情况，正常情况下，同一种中断在不同的cpu上的累计次数应该差不多

ps aux |grep softirq
软中断实际上是以内核线程的方式运行的，每个cpu都对应一个软中断内核线程，就叫做ksoftirqd/1,
查看其运行状况：

top
定位软中断问题，如下图：

 

• 关注%cpu行的si，当>0时，就有软中断产生
• 进程列表中如圈出的这种就是软中断内核进程

sar
使用watch -d cat /proc/softirqs看到变化速率最快的是NET_RX（网络接收），那么我们就可以使用sar来观察网络的接收情况，不仅可以观察网络收发的吞吐量（BPS，每秒收发的字节数），还可以观察网络收发的PPS，即每秒收发的网络帧数。

• 第一列：表示报告的时间。
• 第二列：IFACE 表示网卡。
• 第三、四列：rxpck/s 和 txpck/s 分别表示每秒接收、发送的网络帧数，也就是 PPS。
• 第五、六列：rxkB/s 和 txkB/s 分别表示每秒接收、发送的千字节数，也就是 BPS

tcpdump
怀疑是网络接收中断的问题，若上面sar的结果eno16777736的接收网络帧和字节数异常，我们使用tcpdump来抓包查看