内核源码分析之软中断（基于3.16-rc4）

1.和软中断相关的数据结构：

softing_vec数组（kernel/softirq.c）

1 static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp;

NR_SOFTIRQS值为10，说明内核支持10个软中断函数。

 

softirq_action结构体（include/linux/interrupt.h）

1 struct softirq_action

2 {

3     void    (*action)(struct softirq_action *);

4 };

action是函数指针变量，指向了某个软中断函数。

 

irq_cpustat_t结构体（arch/x86/include/asm/hardirq.h）

 1 typedef struct {

 2     unsigned int __softirq_pending;

 3     unsigned int __nmi_count;    /* arch dependent */

 4 #ifdef CONFIG_X86_LOCAL_APIC

 5     unsigned int apic_timer_irqs;    /* arch dependent */

 6     unsigned int irq_spurious_count;

 7     unsigned int icr_read_retry_count;

 8 #endif

 9 #ifdef CONFIG_HAVE_KVM

10     unsigned int kvm_posted_intr_ipis;

11 #endif

12     unsigned int x86_platform_ipis;    /* arch dependent */

13     unsigned int apic_perf_irqs;

14     unsigned int apic_irq_work_irqs;

15 #ifdef CONFIG_SMP

16     unsigned int irq_resched_count;

17     unsigned int irq_call_count;

18     /*

19      * irq_tlb_count is double-counted in irq_call_count, so it must be

20      * subtracted from irq_call_count when displaying irq_call_count

21      */

22     unsigned int irq_tlb_count;

23 #endif

24 #ifdef CONFIG_X86_THERMAL_VECTOR

25     unsigned int irq_thermal_count;

26 #endif

27 #ifdef CONFIG_X86_MCE_THRESHOLD

28     unsigned int irq_threshold_count;

29 #endif

30 #if IS_ENABLED(CONFIG_HYPERV) || defined(CONFIG_XEN)

31     unsigned int irq_hv_callback_count;

32 #endif

33 } ____cacheline_aligned irq_cpustat_t;

每个cpu都有一个这样的结构体变量，在软中断中，我们要使用的是第2行的成员，32位的软中断掩码。当有一个软中断被挂起（将要被执行）的时候，会设置该掩码中的相应位。

 

2.软中断的执行过程

首先使用open_softirq()函数注册软中断函数，代码如下（kernel/softirq.c）：

1 void open_softirq(int nr, void (*action)(struct softirq_action *))

2 {

3     softirq_vec[nr].action = action;

4 }

将软中断函数指针action存入softirq_vec数组的对应元素中。

接着，使用raise_softirq()激活软中断，代码如下（kernel/softirq.c）：

1 void raise_softirq(unsigned int nr)

2 {

3     unsigned long flags;

4 

5     local_irq_save(flags);

6     raise_softirq_irqoff(nr);

7     local_irq_restore(flags);

8 }

第5行关闭本地中断，第7行恢复中断。第6行激活nr所对应的软中断函数。接着分析 raise_softirq_irqoff()，代码如下（kernel/softirq.c）：

 1 inline void raise_softirq_irqoff(unsigned int nr)

 2 {

 3     __raise_softirq_irqoff(nr);

 4 

 5     /*

 6      * If we're in an interrupt or softirq, we're done

 7      * (this also catches softirq-disabled code). We will

 8      * actually run the softirq once we return from

 9      * the irq or softirq.

10      *

11      * Otherwise we wake up ksoftirqd to make sure we

12      * schedule the softirq soon.

13      */

14     if (!in_interrupt())

15         wakeup_softirqd();

16 }

第3行__raise_softirq_irqoff函数设置了软中断掩码的相应位，代码如下（kernel/softirq.c）。然后第14行判断软中断是否已经激活或者被禁用，如果没有，那么在15行激活内核线程ksoftirq，去执行软中断。

1 void (unsigned int nr)

2 {

3     trace_softirq_raise(nr);

4     or_softirq_pending(1UL << nr);

5 }

具体而言，在第4行的函数中设置掩码位。

 

3.下面分析下在哪些地方都可以进入软中断。

第一个地方，当然就是上边所提到的，内核线程ksoftirq被激活的时候。下面看看ksoftirq线程（kernel/softirq.c）。

1 DEFINE_PER_CPU(struct task_struct *, ksoftirqd);

给每个cpu都定义一个指向struct task_struct类型的结构体变量，很显然，该变量存的是ksoftirq线程的进程描述符。（由此也说明，linux的线程和进程是一个东西）

接着，我们要看看ksoftirq线程要执行的函数（kernel/softirq.c）。

 1 static void run_ksoftirqd(unsigned int cpu)

 2 {

 3     local_irq_disable();

 4     if (local_softirq_pending()) {

 5         /*

 6          * We can safely run softirq on inline stack, as we are not deep

 7          * in the task stack here.

 8          */

 9         __do_softirq();

10         rcu_note_context_switch(cpu);

11         local_irq_enable();

12         cond_resched();

13         return;

14     }

15     local_irq_enable();

16 }

该函数就是ksoftirq线程的线程体。第9行__do_softirq()中调用所有软中断函数。

我们回过头来再分析下wakeup_softirqd()，看看ksoftirq线程怎样被唤醒（kernel/softirq.c）。

1 static void wakeup_softirqd(void)

2 {

3     /* Interrupts are disabled: no need to stop preemption */

4     struct task_struct *tsk = __this_cpu_read(ksoftirqd);

5 

6     if (tsk && tsk->state != TASK_RUNNING)

7         wake_up_process(tsk);

8 }

第4行把本地cpu的ksoftirq线程的描述符读到tsk变量中，第6行中判断ksoftirq线程如果没有运行的话，第7行唤醒该线程。

第二个地方，中断处理程序do_IRQ完成处理或者调用irq_exit函数时。下面看看irq_exit代码（kernel/softirq.c）。

 1 void irq_exit(void)

 2 {

 3 #ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED

 4     local_irq_disable();

 5 #else

 6     WARN_ON_ONCE(!irqs_disabled());

 7 #endif

 8 

 9     account_irq_exit_time(current);

10     preempt_count_sub(HARDIRQ_OFFSET);

11     if (!in_interrupt() && local_softirq_pending())

12         invoke_softirq();

13 

14     tick_irq_exit();

15     rcu_irq_exit();

16     trace_hardirq_exit(); /* must be last! */

17 }

不用我说了吧，我觉得你一眼就能瞄见了第12行。看下invoke_softirq函数（kernel/softirq.c）。

 1 static inline void invoke_softirq(void)

 2 {

 3     if (!force_irqthreads) {

 4 #ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK

 5         /*

 6          * We can safely execute softirq on the current stack if

 7          * it is the irq stack, because it should be near empty

 8          * at this stage.

 9          */

10         __do_softirq();

11 #else

12         /*

13          * Otherwise, irq_exit() is called on the task stack that can

14          * be potentially deep already. So call softirq in its own stack

15          * to prevent from any overrun.

16          */

17         do_softirq_own_stack();

18 #endif

19     } else {

20         wakeup_softirqd();

21     }

22 }

第3行force_irqthreads值为0，所以该函数也调用了__do_softirq()来执行软中断。

还有几处地方暂时不分析了，以后有空补上。

 

4.下面来看下__do_softirq()函数（kernel/softirq.c）。

 1 asmlinkage __visible void __do_softirq(void)

 2 {

 3     unsigned long end = jiffies + MAX_SOFTIRQ_TIME;

 4     unsigned long old_flags = current->flags;

 5     int max_restart = MAX_SOFTIRQ_RESTART;

 6     struct softirq_action *h;

 7     bool in_hardirq;

 8     __u32 pending;

 9     int softirq_bit;

10 

11     /*

12      * Mask out PF_MEMALLOC s current task context is borrowed for the

13      * softirq. A softirq handled such as network RX might set PF_MEMALLOC

14      * again if the socket is related to swap

15      */

16     current->flags &= ~PF_MEMALLOC;

17 

18     pending = local_softirq_pending();

19     account_irq_enter_time(current);

20 

21     __local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET);

22     in_hardirq = lockdep_softirq_start();

23 

24 restart:

25     /* Reset the pending bitmask before enabling irqs */

26     set_softirq_pending(0);

27 

28     local_irq_enable();

29 

30     h = softirq_vec;

31 

32     while ((softirq_bit = ffs(pending))) {

33         unsigned int vec_nr;

34         int prev_count;

35 

36         h += softirq_bit - 1;

37 

38         vec_nr = h - softirq_vec;

39         prev_count = preempt_count();

40 

41         kstat_incr_softirqs_this_cpu(vec_nr);

42 

43         trace_softirq_entry(vec_nr);

44         h->action(h);

45         trace_softirq_exit(vec_nr);

46         if (unlikely(prev_count != preempt_count())) {

47             pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n",

48                    vec_nr, softirq_to_name[vec_nr], h->action,

49                    prev_count, preempt_count());

50             preempt_count_set(prev_count);

51         }

52         h++;

53         pending >>= softirq_bit;

54     }

55 

56     rcu_bh_qs(smp_processor_id());

57     local_irq_disable();

58 

59     pending = local_softirq_pending();

60     if (pending) {

61         if (time_before(jiffies, end) && !need_resched() &&

62             --max_restart)

63             goto restart;

64 

65         wakeup_softirqd();

66     }

67 

68     lockdep_softirq_end(in_hardirq);

69     account_irq_exit_time(current);

70     __local_bh_enable(SOFTIRQ_OFFSET);

71     WARN_ON_ONCE(in_interrupt());

72     tsk_restore_flags(current, old_flags, PF_MEMALLOC);

73 } 

在该函数中循环调用的所有的被激活的软中断函数。第5行MAX_SOFTIRQ_RESTART值为10，表示最多循环10次（不能让其他进程等太久），第32行获取pending表中第一被设置的比特位，第44行开始执行设置过的软中断函数。第53行对pending进行右移运算，然后进入下次循环。直到将本轮所有已设置的软中断函数全部执行完，退出循环。第59行重新获得本地cpu的软中断掩码，第61行如果时间没有超出end而且没有出现更高优先级的进程并且10次寻环未用完，那么跳回restart，重新for循环。否则，第65行唤醒softirqd内核线程。然后退出本函数。

 

至此，软中断的处理过程就分析完了。