Re: schedule_timeout() called with state TASK_RUNNING from fs/select.c

Re: schedule_timeout() called with state TASK_RUNNING from fs/select.c

[Oleg Nesterov]

Robert P. J. Day wrote:
>
>  as i understand it, schedule_timeout() should always be called with
> a current->state of TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE, no?

No.

Yes, it is pointless to call schedule() if it is known that ->state ==
TASK_RUNNING, but sometimes the task doesn't know its state when it
calls schedule/schedule_timeout, the state could be changed by
try_to_wake_up().

See wait_event_timeout() as an example.

Oleg.

Re: schedule_timeout() called with state TASK_RUNNING from fs/select.c

[Robert P. J. Day]

On Sun, 8 Jul 2007, Oleg Nesterov wrote:

> Robert P. J. Day wrote:
> >
> >  as i understand it, schedule_timeout() should always be called with
> > a current->state of TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE, no?
>
> No.
>
> Yes, it is pointless to call schedule() if it is known that ->state ==
> TASK_RUNNING, but sometimes the task doesn't know its state when it
> calls schedule/schedule_timeout, the state could be changed by
> try_to_wake_up().
>
> See wait_event_timeout() as an example.

fair enough.  i already knew that, technically, it wasn't an error to
call schedule_timeout() with state == TASK_RUNNING.  i was just
following up on something out of love's kernel book, and i got curious
as to whether the above ever happened.  now i know.

rday
--
========================================================================
Robert P. J. Day
Linux Consulting, Training and Annoying Kernel Pedantry
Waterloo, Ontario, CANADA

http://fsdev.net/wiki/index.php?title=Main_Page
========================================================================

schedule_timeout() called with state TASK_RUNNING from fs/select.c

[Robert P. J. Day]

  as i understand it, schedule_timeout() should always be called with
a current->state of TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE, no?
based on a trivial check i added to kernel/timer.c, i can read this in
/var/log/messages:

...
Jul  7 14:48:40 localhost kernel: Bad schedule_timeout state: 0 from Xorg, c015eeb9
Jul  7 14:49:05 localhost kernel: Bad schedule_timeout state: 0 from gnome-terminal, c015ea47
Jul  7 14:50:19 localhost kernel: Bad schedule_timeout state: 0 from Xorg, c015eeb9
Jul  7 14:50:50 localhost kernel: Bad schedule_timeout state: 0 from Xorg, c015eeb9
...

  those last values are __builtin_return_address(0), and here's the
snippet from /proc/kallsyms:

...
c015e7f2 T do_sys_poll
c015eb19 T sys_poll
c015eb50 T do_select
c015ef10 t core_sys_select
...

  i'm guessing those addresses correspond to the calls to
schedule_timeout() in do_sys_poll() and do_select(), if anyone thinks
it's worth looking at.

rday
-- 
========================================================================
Robert P. J. Day
Linux Consulting, Training and Annoying Kernel Pedantry
Waterloo, Ontario, CANADA

http://fsdev.net/wiki/index.php?title=Main_Page
========================================================================