Re: [RFC][PATCH][2.5] Possible race in wait_task_zombie and finish_task_switch

[Manfred Spraul]

Do we have an idea which reference is miscounted? It seems that there 
are 4 different kinds of references to a task structure:

- the reference for the stack itself, acquired by setting usage to 2, 
dropped by schedule_tail.
- the reference for wait4, acquired by setting usage to 2, dropped by 
wait_task_zombie.
- references for the pid structures, maintained by pid.c
- temporary references for looking at tsk->{fs,mm,files,tty}, used by 
/proc, ptrace, tty.

>kernel BUG at kernel/sched.c:746!
>  
>
Hmm. What is schedule.c:746? There is no BUG in that area in the bk tree.

Zwane, is it easy to reproduce the crash? I could write a patch that 
adds 4 refcounters, then we could find out in which area we must look.

--
    Manfred