Re: 2.6.16-rc5 pppd oops on disconnects

[Paul Fulghum <paulkf@microgate.com>]

On Fri, 2006-03-10 at 13:48 -0500, Bob Copeland wrote:
> > > Call Trace:
> > >  [<c017592e>] sysfs_hash_and_remove+0x34/0x10a
> > >  [<c01e756e>] class_device_del+0xa0/0x11c
> > >  [<c01e75f5>] class_device_unregister+0xb/0x16
> > >  [<d01f81f3>] acm_tty_unregister+0x1d/0x63 [cdc_acm]
> >
> > This looks more like
> > http://bugzilla.kernel.org/show_bug.cgi?id=5876
> 
> Hmm... it looks different from that bug - in that case the root cause
> was sysfs_make_dirent failing, presumably when the sysfs node for the
> device was being set up, by unplugging and re-plugging the device a
> lot.  Here it's oopsing when the node is being removed, after it's
> been in use a while and unplugged only once.  But yes ppp may not have
> anything to do with it.  I'll try it on an older kernel to see if I
> can reproduce there...

The i_sem to i_mutex change started in the 2.6.16 series.
Running against 2.6.15 would be interesting. Being able
to repeat every time is a plus. I'm not that familiar
with the sysfs stuff, but the slab poisoning is pretty
damning. The dentry was released and then accessed.

I looked at cdc_acm for disconnect and close and
did not see any problems (such as trying to call
tty_unregister_device twice for a device).

-- 
Paul Fulghum
Microgate Systems, Ltd