Re: [PATCH v4 12/12] sched,signal,ptrace: Rework TASK_TRACED, TASK_STOPPED state

[Steven Rostedt <rostedt@xxxxxxxxxxx>]

On Tue, 21 Jun 2022 17:15:47 +0200
Alexander Gordeev <agordeev@xxxxxxxxxxxxx> wrote:

> So I assume (checked actually) the return 0 below from kernel/sched/core.c:
> wait_task_inactive() is where it bails out:
> 
> 3303                 while (task_running(rq, p)) {
> 3304                         if (match_state && unlikely(READ_ONCE(p->__state) != match_state))
> 3305                                 return 0;
> 3306                         cpu_relax();
> 3307                 }
> 
> Yet, the child task is always found in __TASK_TRACED state (as seen
> in crash dumps):
> 
> > 101447  11342  13      ce3a8100      RU   0.0   10040   4412  strace  
>   101450  101447   0      bb04b200      TR   0.0    2272   1136  kill_child
>   108261  101447   2      d0b10100      TR   0.0    2272    532  kill_child
> crash> task bb04b200 __state  
> PID: 101450  TASK: bb04b200          CPU: 0   COMMAND: "kill_child"
>   __state = 8,
> 
> crash> task d0b10100 __state  
> PID: 108261  TASK: d0b10100          CPU: 2   COMMAND: "kill_child"
>   __state = 8,

If you are using crash, can you enable all trace events?

Then you should be able to extract the ftrace ring buffer from crash using
the trace.so extend (https://github.com/fujitsu/crash-trace)

I guess it should still work with s390.

Then you can see the events that lead up to the crash.

-- Steve