On Wed, 2002-07-31 at 06:41, Keith Owens wrote:
> On Tue, 30 Jul 2002 15:10:30 -0700 (PDT),
> Ravi Wijayaratne <ravi_wija@xxxxxxxxx> wrote:
> >We are experiencing an XFS stack overflow. Not a
> >recurrent issue, but happens once in a while.
> >
> >Here is the oops message spewed out by our stack
> >depth checker. It calls BUG() as soon as it sees
> >some thing has overwritten the task_struct
> >
> >The oops was deliberately triggered. So please dont
> >worry about the oops. My worry is the call depth 114.
> >Here is the data about the system.
>
> The stack backtrace code in the kernel prints anything on stack that
> looks like a kernel address. In this case a lot of the entries are
> spurious, unfortunately it is not clear what the real stack trace is.
> Does this kernel have preempt turned on? At first glance it looks like
> XFS was interrupted by some TCP/IP code, that should not happen.
>
> Can you compile the kernel with frame pointers (CONFIG_FRAME_POINTER=y)
> and kdb (CONFIG_KDB=y) then drop into kdb when the stack overflows.
> The kdb backtrace (command 'bt') using frame pointers is far more
> accurate and will show what is really happening.
You probably need to use kgcc to do this, kdb has some issues with some
more recent compilers and tends to skip large chunks of stack frames.
Steve
--
Steve Lord voice: +1-651-683-3511
Principal Engineer, Filesystem Software email: lord@xxxxxxx
|