BUG: scheduling while atomic 分析【转】

本文转载自：https://blog.csdn.net/cfy_phonex/article/details/12090943

遇到一个典型的ｓｃｈｅｄｕｌｅ问题。

1. < 3>[26578.636839] C1 [ swapper/1] BUG: scheduling while atomic: swapper/1/0/0x00000002
2. < 6>[26578.636869] C0 [ kworker/u:1] CPU1 is up
3. < 4>[26578.636900] C1 [ swapper/1] Modules linked in: bcm15500_i2c_ts
4. < 4>[26578.636961] C1 [ swapper/1] [] (unwind_backtrace+0x0/0x11c) from [] (__schedule+0x70/0x6e0)
5. < 4>[26578.636991] C1 [ swapper/1] [] (__schedule+0x70/0x6e0) from [] (schedule_preempt_disabled+0x14/0x20)
6. < 4>[26578.637052] C1 [ swapper/1] [] (schedule_preempt_disabled+0x14/0x20) from [] (cpu_idle+0xf0/0x104)
7. < 4>[26578.637083] C1 [ swapper/1] [] (cpu_idle+0xf0/0x104) from [] (cpu_die+0x2c/0x5c)
8. < 3>[26578.637510] C1 [ swapper/1] BUG: scheduling while atomic: swapper/1/0/0x00000002
9. < 4>[26578.637510] C1 [ swapper/1] Modules linked in: bcm15500_i2c_ts
10. < 4>[26578.637602] C1 [ swapper/1] [] (unwind_backtrace+0x0/0x11c) from [] (__schedule+0x70/0x6e0)
11. < 4>[26578.637663] C1 [ swapper/1] [] (__schedule+0x70/0x6e0) from [] (schedule_preempt_disabled+0x14/0x20)
12. < 4>[26578.637724] C1 [ swapper/1] [] (schedule_preempt_disabled+0x14/0x20) from [] (cpu_idle+0xf0/0x104)
13. < 4>[26578.637754] C1 [ swapper/1] [] (cpu_idle+0xf0/0x104) from [] (cpu_die+0x2c/0x5c)
14. < 3>[26578.648069] C1 [ swapper/1] BUG: scheduling while atomic: swapper/1/0/0x00000002

查看源代码

1. /*
2. * __schedule() is the main scheduler function.
3. */
4. static void __sched __schedule(void)
5. {
6. struct task_struct *prev, *next;
7. unsigned long *switch_count;
8. struct rq *rq;
9. int cpu;
11. need_resched:
12. preempt_disable();
13. cpu = smp_processor_id();
14. rq = cpu_rq(cpu);
15. rcu_note_context_switch(cpu);
16. prev = rq->curr;
18. schedule_debug(prev);
19. 　　　　....
20. ｝

1. /*
2. * Print scheduling while atomic bug:
3. */
4. static noinline void __schedule_bug(struct task_struct *prev)
5. {
6. if (oops_in_progress)
7. return;
9. printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
10. prev->comm, prev->pid, preempt_count());
12. debug_show_held_locks(prev);
13. print_modules();
14. if (irqs_disabled())
15. print_irqtrace_events(prev);
17. dump_stack();
18. }
20. /*
21. * Various schedule()-time debugging checks and statistics:
22. */
23. static inline void schedule_debug(struct task_struct *prev)
24. {
25. /*
26. * Test if we are atomic. Since do_exit() needs to call into
27. * schedule() atomically, we ignore that path for now.
28. * Otherwise, whine if we are scheduling when we should not be.
29. */
30. if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
31. __schedule_bug(prev);
32. rcu_sleep_check();
34. profile_hit(SCHED_PROFILING, __builtin_return_address( 0));
36. schedstat_inc(this_rq(), sched_count);
37. }

可以看出，　满足如下条件将会打印该出错信息

unlikely(in_atomic_preempt_off() && !prev->exit_state

为0表示TASK_RUNNING状态,当前进程在运行;　并且处于原子状态,,那么就不能切换给其它的进程

1. Linux/include/linux/sched.h
3. /*
4. * Task state bitmask. NOTE! These bits are also
5. * encoded in fs/proc/array.c: get_task_state().
6. *
7. * We have two separate sets of flags: task->state
8. * is about runnability, while task->exit_state are
9. * about the task exiting. Confusing, but this way
10. * modifying one set can't modify the other one by
11. * mistake.
12. */
13. #define TASK_RUNNING 0
14. #define TASK_INTERRUPTIBLE 1
15. #define TASK_UNINTERRUPTIBLE 2
16. #define __TASK_STOPPED 4
17. #define __TASK_TRACED 8
18. /* in tsk->exit_state */
19. #define EXIT_ZOMBIE 16
20. #define EXIT_DEAD 32
21. /* in tsk->state again */
22. #define TASK_DEAD 64
23. #define TASK_WAKEKILL 128
24. #define TASK_WAKING 256
25. #define TASK_STATE_MAX 512

1. kernel/include/linux/hardirq.h
3. #if defined(CONFIG_PREEMPT_COUNT)
4. # define PREEMPT_CHECK_OFFSET 1
5. #else
6. # define PREEMPT_CHECK_OFFSET 0
7. #endif
9. /*
10. * Are we running in atomic context? WARNING: this macro cannot
11. * always detect atomic context; in particular, it cannot know about
12. * held spinlocks in non-preemptible kernels. Thus it should not be
13. * used in the general case to determine whether sleeping is possible.
14. * Do not use in_atomic() in driver code.
15. */
16. #define in_atomic() ((preempt_count() & ~PREEMPT_ACTIVE) != 0)
18. /*
19. * Check whether we were atomic before we did preempt_disable():
20. * (used by the scheduler, *after* releasing the kernel lock)
21. */
22. #define in_atomic_preempt_off() \
23. ((preempt_count() & ~PREEMPT_ACTIVE) != PREEMPT_CHECK_OFFSET)

结论整理

linux内核打印"BUG: scheduling while atomic"和"bad: scheduling from the idle thread"错误的时候，

通常是在中断处理函数中调用了可以休眠的函数，如semaphore,mutex,sleep之类的可休眠的函数，

而linux内核要求在中断处理的时候，不允许系统调度，不允许抢占，要等到中断处理完成才能做其他事情。

因此，要充分考虑中断处理的时间，一定不能太久。
另外一个能产生此问题的是在idle进程里面，做了不该做的事情。现在Linux用于很多手持式设备，为了降低功耗，

通常的作法是在idle进程里面降低CPU或RAM的频率、关闭一些设备等等。要保证这些动作的原子性才能确保

不发生"bad: scheduling from the idle thread"这样的错误！

禁止内核抢占是指内核不会主动的抢占你的process，但是现在是你在自己的程序中主动call schedule()，

kernel并不能阻止你这么作。

Scheduling while atomic" means that a thread has called schedule() during an operation which is supposed to be atomic (ie uninterrupted).

1. 190 NOTE: ***** WARNING *****
2. 191 NEVER SLEEP IN A COMPLETION HANDLER. These are normally called
3. 192 during hardware interrupt processing. If you can, defer substantial
4. 193 work to a tasklet (bottom half) to keep system latencies low. You'll
5. 194 probably need to use spinlocks to protect data structures you manipulate
6. 195 in completion handlers.

1. 1. GFP_ATOMIC is used when
   2. (a) you are inside a completion handler, an interrupt, bottom half, tasklet or timer, or
   3. (b) you are holding a spinlock or rwlock (does not apply to semaphores), or
   4. (c) current->state != TASK_RUNNING, this is the case only after you've changed it.