Re: Lockup in tracepoint unregister in sched switch ftrace plugin

[Mathieu Desnoyers]

* Steven Rostedt (rostedt@xxxxxxxxxxx) wrote:
> 
> Mathieu,
> 
> I just downloaded the latest git repo from Linus's tree, and the 
> sched_switch start up test locks up. I traced it down to the first 
> unregister of a trace point. Here's the call path that I see.
> 
> kernel/trace/trace.c: register_tracer
> kernel/trace/trace_selftest.c: trace_selftest_startup_sched_switch
> kernel/trace/trace_sched_switch.c: sched_switch_trace_ctrl_update
>    " "  : stop_sched_trace
>    " "  : tracing_stop_cmdline_record
>    " "  : tracing_stop_sched_switch
>    " "  : tracing_sched_unregister
> 
>   which calls unregister_trace_sched_switch define as macro to:
> 
> kernel/tracepoint.c: tracepoint_probe_unregister
>    " "  : remove_tracepoint
> kernel/rcupdate.c: rcu_barrier_sched
>    " "  : _rcu_barrier
> 
> where it gets stuck at that "wait_for_completion".
> 
> I'm not sure if, because this is a scheduler trace point that we are 
> hitting some kind of race that is preventing the wait_for_completion to 
> finish, or what.
> 
> I'll look more at it tomorrow.
> 

Hi Steven,

Hrm, does this selftest execute early at boot-time ? If yes, and if
classic RCUs are not up yet at that point in bootup, then using
rcu_barrier() will not work well. Another thing to look into is to make
sure tracing_sched_unregister is never called with interrupts or
preemption off.

Mathieu

> -- Steve
> 
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/