"Couldn't open mtdev" crashing X

[Thomas Bushnell, BSG]

I've given up and returned the hardware. It would have been possible to
debug it a week ago, but now I no longer possess it.

Thomas


On Fri, Jun 3, 2011 at 12:56 PM, Chase Douglas
<chase.douglas at canonical.com>wrote:

> On 05/27/2011 12:47 AM, Thomas Bushnell, BSG wrote:
> > I just noticed that the mtdev source I've been looking at is Debian
> > unstable, which is a tad newer than natty, and in particular, adds an
> > additional safety check in the front of mtdev_open. If indeed a null
> > pointer is passed in to mtdev_open's first arg, it would quickly
> > segfault (though I can't see why the fault would be in mtdev_open and
> > not mtdev_init).
> >
> > I wonder if this might not be the immediate cause of the crash. I don't
> > know how to quickly get the natty source for evdev (with the patches to
> > call mtdev from Chase Douglas); the Debian unstable version of evdev
> > does not yet include that patch. From the one I see
> > (http://lists.x.org/archives/xorg-devel/2011-January/018405.html) this
> > shouldn't happen:
> >
> > The dev could only be bad if EvdevOpenDevice were to have pEvdev->device
> > set, but pEvdev->mtdev still unset. That can't happen afaict, though I
> > would prefer the patch initialized mtdev in a separate branch to make
> > that clearer.
> >
> > Still, I'm interested to know if the patch to check the args to
> > mtdev_open was prompted by actual problems or just general cleanup.
> >
> > It is also odd to me that the crash is reported to be at mtdev_open+22,
> > but according to gdb, that's in the middle of an instruction, and if the
> > PC got pointed there by mistake, its a register-only instruction anyhow,
> > so it can't fault.
>
> Hi Thomas,
>
> I've read all your emails, but I'm preplexed as well as to why X is
> crashing. I don't believe any new mtdev code was added for any specific
> reason other than general cleanups, and I've not heard of anyone else
> encountering such a bug.
>
> If you can attach to it through gdb, you could step through the
> mtdev_open call to see which line is causing a segfault. You may find it
> helpful to build a test package of xserver-xorg-input-evdev without
> optimizations by setting the env var DEB_BUILD_OPTIONS="noopt".
>
> Please file a bug for this in Launchpad, and subscribe the utouch-bugs
> team.
>
> Thanks!
>
> -- Chase
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/ubuntu-installer/attachments/20110603/74705a36/attachment.html>