[Intel-gfx] [git pull] drm fixes

[Josh Boyer]

On Tue, Mar 24, 2015 at 12:49 PM, Daniel Vetter <daniel at ffwll.ch> wrote:
> On Tue, Mar 24, 2015 at 05:48:31PM +0100, Daniel Vetter wrote:
>> On Tue, Mar 24, 2015 at 12:10:28PM -0400, Josh Boyer wrote:
>> > On Tue, Mar 24, 2015 at 10:46 AM, Josh Boyer <jwboyer at fedoraproject.org> wrote:
>> > > On Tue, Mar 24, 2015 at 10:34 AM, Daniel Vetter <daniel at ffwll.ch> wrote:
>> > >> On Tue, Mar 24, 2015 at 10:22:30AM -0400, Josh Boyer wrote:
>> > >>> On Tue, Mar 24, 2015 at 9:57 AM, Josh Boyer <jwboyer at fedoraproject.org> wrote:
>> > >>> > On Tue, Mar 24, 2015 at 9:40 AM, Daniel Vetter <daniel at ffwll.ch> wrote:
>> > >>> >> On Tue, Mar 24, 2015 at 09:15:32AM -0400, Josh Boyer wrote:
>> > >>> >>> On Tue, Mar 24, 2015 at 3:32 AM, Daniel Vetter <daniel at ffwll.ch> wrote:
>> > >>> >>> > On Mon, Mar 23, 2015 at 02:34:27PM -0400, Josh Boyer wrote:
>> > >>> >>> >> On Mon, Mar 23, 2015 at 11:33 AM, Josh Boyer <jwboyer at fedoraproject.org> wrote:
>> > >>> >>> >>
>> > >>> >>> >> <snip>
>> > >>> >>> >>
>> > >>> >>> >> >> Xi Ruoyao (1):
>> > >>> >>> >> >>       drm/i915: Ensure plane->state->fb stays in sync with plane->fb
>> > >>> >>> >>
>> > >>> >>> >> Turns out to be that commit.
>> > >>> >>> >>
>> > >>> >>> >> git bisect start 'drivers/gpu/drm/i915/'
>> > >>> >>> >> # good: [b314acaccd7e0d55314d96be4a33b5f50d0b3344] Merge branch
>> > >>> >>> >> 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
>> > >>> >>> >> git bisect good b314acaccd7e0d55314d96be4a33b5f50d0b3344
>> > >>> >>> >> # bad: [bc465aa9d045feb0e13b4a8f32cc33c1943f62d6] Linux 4.0-rc5
>> > >>> >>> >> git bisect bad bc465aa9d045feb0e13b4a8f32cc33c1943f62d6
>> > >>> >>> >> # bad: [319c1d420a0b62d9dbb88104afebaabc968cdbfa] drm/i915: Ensure
>> > >>> >>> >> plane->state->fb stays in sync with plane->fb
>> > >>> >>> >> git bisect bad 319c1d420a0b62d9dbb88104afebaabc968cdbfa
>> > >>> >>> >> # first bad commit: [319c1d420a0b62d9dbb88104afebaabc968cdbfa]
>> > >>> >>> >> drm/i915: Ensure plane->state->fb stays in sync with plane->fb
>> > >>> >>> >>
>> > >>> >>> >> Doing a straight revert on top of 4.0-rc5 makes things work again,
>> > >>> >>> >> albeit with the WARN_ON(obj->frontbuffer_bits) splat still being
>> > >>> >>> >> there.
>> > >>> >>> >
>> > >>> >>> > Can you please test the tip of drm-fixes:
>> > >>> >>> >
>> > >>> >>> > commit 8218c3f4df3bb1c637c17552405039a6dd3c1ee1
>> > >>> >>> > Author: Daniel Vetter <daniel.vetter at ffwll.ch>
>> > >>> >>> > Date:   Fri Feb 27 12:58:13 2015 +0100
>> > >>> >>> >
>> > >>> >>> >     drm: Fixup racy refcounting in plane_force_disable
>> > >>> >>> >
>> > >>> >>> > http://cgit.freedesktop.org/~airlied/linux/commit/?h=drm-fixes&id=8218c3f4df3bb1c637c17552405039a6dd3c1ee1
>> > >>> >>> >
>> > >>> >>> > Because fumble that patch didn't make it to drm-fixes a while ago and
>> > >>> >>> > instead landed in drm-next.
>> > >>> >>>
>> > >>> >>> That seems to have helped with totally different issues a macbook I
>> > >>> >>> have was seeing.  However, it still doesn't fix the issue with the
>> > >>> >>> Celeron based NUC machine.
>> > >>> >>>
>> > >>> >>> I built a kernel based on Linus' latest tree as of this morning,
>> > >>> >>> without reverting 319c1d4 and adding the commit you pointed to.  The
>> > >>> >>> NUC still won't boot without HDMI connected.  With HDMI connected I
>> > >>> >>> still see the trace below.  If I do the blacklist and then insmod
>> > >>> >>> dance with HDMI unplugged it shows the same spew I reported yesterday
>> > >>> >>> which starts with the same backtrace.
>> > >>> >>>
>> > >>> >>> I'll try building a kernel with 319c1d4 reverted + your patch.  I
>> > >>> >>> suspect things will work fine with that combination because the two
>> > >>> >>> issues are unrelated.
>> > >>> >>
>> > >>> >> Can you please boot with drm.debug=0xff for the below case and grab
>> > >>> >> complete dmesg? There'll be a lot of crap in the logs, you might need to
>> > >>> >> blow up the logbuf size massively. But that log should contain everything
>> > >>> >> I need to figure out where that framebuffer we're blowing up on is going.
>> > >>> >
>> > >>> > I provided both with HDMI attached and without (via insmod).  If you
>> > >>> > want them emailed directly let me know, but they were large.
>> > >>> >
>> > >>> > Boot with drm.debug=0xff and HDMI connected:
>> > >>> >
>> > >>> > https://jwboyer.fedorapeople.org/pub/drm-ff-dmesg.txt
>> > >>> >
>> > >>> > Boot with drm.debug=0xff without HDMI connected and i915 loaded via
>> > >>> > manual insmod after boot:
>> > >>> >
>> > >>> > https://jwboyer.fedorapeople.org/pub/drm-ff-no-hdmi-insmod.txt
>> > >>>
>> > >>> Here's one more from the macbook I mentioned.  It's showing the same
>> > >>> kref.h splat:
>> > >>>
>> > >>> https://jwboyer.fedorapeople.org/pub/drm-ff-macbook.txt
>> > >>
>> > >> Ok there's at least one fixup for which we've failed to apply when porting
>> > >> the fb refcounting fix from -next. Can you please cherry-pick
>> > >>
>> > >> commit f55548b5af87ebfc586ca75748947f1c1b1a4a52
>> > >> Author: Damien Lespiau <damien.lespiau at intel.com>
>> > >> Date:   Thu Feb 5 18:30:20 2015 +0000
>> > >>
>> > >>     drm/i915: Don't try to reference the fb in get_initial_plane_config()
>> > >>
>> > >> From linux-next?
>> > >
>> > > Yes, building now.  Will let you know as soon as I test it on both machines.
>> >
>> > OK, with that commit applied I no longer get the kref.h splat and the
>> > NUC machine boots headless.  I still see the backtrace below on both
>> > the NUC and the macbook.  I have a copy of it with drm.debug=0xff from
>> > the NUC here:
>> >
>> > https://jwboyer.fedorapeople.org/pub/nuc-drm-debug-ff-with-fixes.txt
>> >
>> > Getting better at least :).
>>
>> Ok thanks for testing. I'll look at that one tomorrow, wasted too much
>> time with trying to resurrect a few machines that should have matched the
>> common parts of what goes wrong here.
>>
>> Jani, can you please cherry-pick the above commit to -fixes?
>
> Actually add Jani this time around ...
> -Daniel
>
>>
>> One more question: Is the frontbuffer_bits splat now also gone? That was
>> the one I have no clue about, but since somewhere around 4.0-rc it started
>> poppping up in a few places ... Thus far it was always the canary for some
>> other bug though.

As far as I can tell, it's gone.  I don't see it on any of my i915
machines running the kernel with those two patches.  I'll keep an eye
out for it as we work through 4.0-rcX.

josh