Re: [Bugme-new] [Bug 38312] New: Oops in kmem_cache_alloc

[James Bottomley <James.Bottomley@HansenPartnership.com>]

On Mon, 2011-06-27 at 23:04 +0200, Steinar H. Gunderson wrote:
> On Mon, Jun 27, 2011 at 03:41:40PM -0500, James Bottomley wrote:
> > Possibly ... if it's a refcounting bug on the host structure (which
> > would cause shost->pool to have bogus data).  However, in that case,
> > there should be some reference to freeing the host in the logs above the
> > oops (or some event that triggered it).   For just a running system, we
> > don't ever free the host structure until all the devices are gone.
> 
> I checked the serial port log (I log the serial console from another machine,
> to be sure to get these kinds of bugs even if they hit the network and/or
> SCSI subsystems), and the only thing is that cron seems to have segfaulted a
> time. This is unusual, but I take it it shouldn't crash the kernel in itself
> (and it might be due to the result of some glibc up- and downgrading around
> that time).

That does make it pretty unlikely to be a bogus pointer caused by reuse
of a freed host structure.  At this point, I'm afraid, I don't have any
other ideas.

James