Can no longer shutdown after drm/radeon: Implement radeon_pci_shutdown

[Alex Deucher]

On Thu, Dec 12, 2013 at 8:56 AM, Markus Trippelsdorf
<markus at trippelsdorf.de> wrote:
> On 2013.12.12 at 03:27 +0000, Deucher, Alexander wrote:
>> > On 2013.12.11 at 23:46 +0000, Deucher, Alexander wrote:
>> > > > -----Original Message-----
>> > > > From: Peter Chubb [mailto:peter.chubb at nicta.com.au]
>> > > > Sent: Wednesday, December 11, 2013 5:11 PM
>> > > > To: Markus Trippelsdorf
>> > > > Cc: Peter Chubb; Deucher, Alexander; airlied at linux.ie; dri-
>> > > > devel at lists.freedesktop.org
>> > > > Subject: Re: Can no longer shutdown after drm/radeon: Implement
>> > > > radeon_pci_shutdown
>> > > >
>> > > > >>>>> "Markus" == Markus Trippelsdorf <markus at trippelsdorf.de>
>> > writes:
>> > > >
>> > > > Markus> On 2013.12.11 at 11:37 +1100, Peter Chubb wrote:
>> > > >
>> > > > Markus> It would be interesting to know where exactly it hangs.  Could
>> > > > Markus> you comment out the *_fini(rdev) calls in
>> > > > Markus> radeon_driver_unload_kms
>> > > > (drivers/gpu/drm/radeon/radeon_kms.c)
>> > > > Markus> one after the other to find out which one is responsible?
>> > > >
>> > > > It's radeon_device_fini() that is the problem.
>> > >
>> > > I think the problem is that the drm subsystem tears down the device
>> > > via drm_driver.unload in drm_dev_unregister(), but now that we have a
>> > > pci_driver.shutdown callback (which is needed for kexec) that gets
>> > > called too so the driver gets torn down twice.
>> >
>> > If that is the case the following patch should fix the issue.
>> > Can you give it a try, Peter?
> (Peter:)
>> Thanks that works.  I tested shutdown, kexec, and s2disk --- all work
>> correctly.
>
>> That may work, but I think it's just papering over a race which may
>> still bite someone else depending on the timing.
>
> This leaves three possibilities:
>
> 1) Revert 846ae41ae99d now and come up with a solution with proper
> locking for 3.14
> 2) Add my simple fix now and implement additional locking if the need
> arises for 3.14.
> 3) Implement a fix with proper locking now.
>
> It's your choice Alex.

I guess I'll revert it for now.  I need to think about it a bit more.
Thanks for your help.

Alex

>
> --
> Markus
> _______________________________________________
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/dri-devel