Re: Potential timer bug

[Tipp Moseley <tipp.moseley@gmail.com>]

Err.  I'm stupid.  This isn't a bug.




On Fri, 11 Feb 2005 02:09:10 -0700, Tipp Moseley <tipp.moseley@gmail.com> wrote:
> Hello,
> 
> I am running on a uniprocessor x86 with CONFIG_SMP disabled and
> CONFIG_PREEMPT enabled.
> 
> The problem I have encountered is when using a timer in a module.  The
> timer is set to execute every 3 ticks, and does nothing but increment
> a counter, and that works fine.  However, when the module is unloaded
> sometimes the system hangs on module exit and barfs something like:
> 
> Unable to handle kernel paging request at virtual address e18861bc
>  printing eip:
> c0122a88
> *pde = 015e5067
> *pte = 00000000
> Oops: 0002 [#1]
> PREEMPT
> Modules linked in: yenta socket, rsrc_nonstatic sonypi
> CPU: 0
> ...
> Call Trace:
>  __do_softirq+0x76/0x90
>  do_softirq+0x41/0x50
> ...
> <0>Kernel panic - not syncing: Fatal exception in interrupt
> 
> I am using del_timer_sync to delete the timer in the module_exit
> routine, and sometimes it works correctly.  My theory is that since
> with CONFIG_SMP disabled, del_timer_sync is the same as del_timer.
> This allows the timer to potentially execute after the module has
> unloaded, causing the invalid paging request.
> 
> My solution to the problem (which works, but is probably not optimal)
> is to change '#ifdef CONFIG_SMP' to '#if defined(CONFIG_SMP) ||
> defined(CONFIG_PREEMPT)' around the code defining timer_del_sync.  A
> patch is attached.  Let me know if there's any more information that I
> can provide.
> 
> Thanks,
> 
> Tipp Moseley
> 
> 
>