[Nouveau] VGA resume & thaw (wake up from S3 & S4) broken - kernel(nouveau) exclusively

[Ilia Mirkin]

On Sat, Sep 13, 2014 at 5:25 PM, poma <pomidorabelisima at gmail.com> wrote:
> On 13.09.2014 22:58, Ilia Mirkin wrote:
>> On Sat, Sep 13, 2014 at 4:52 PM, poma <pomidorabelisima at gmail.com> wrote:
>>> On 13.09.2014 07:02, poma wrote:
>>>> On 13.09.2014 06:57, poma wrote:
>>>>>
>>>>> Actually I have nothing to show cause logs are all OK.
>>>>> Haha, it seems to me that the bugs become intelligent.
>>>>>
>>>>> 3.15.10-201.fc20.x86_64
>>>>> 3.16.2-200.fc20.x86_64
>>>>> 3.17.0-0.rc4.git3.2.fc22.1.x86_64
>>>>>  nouveau  [     DRM] suspending display...
>>>>>  nouveau  [     DRM] unpinning framebuffer(s)...
>>>>>  nouveau  [     DRM] evicting buffers...
>>>>>  nouveau  [     DRM] waiting for kernel channels to go idle...
>>>>>  nouveau  [     DRM] suspending client object trees...
>>>>>  nouveau  [     DRM] suspending kernel object tree...
>>>>> ...
>>>>>  nouveau  [     DRM] re-enabling device...
>>>>>  nouveau  [     DRM] resuming kernel object tree...
>>>>>  nouveau  [   VBIOS][0000:02:00.0] running init tables
>>>>>  nouveau  [  PTHERM][0000:02:00.0] fan management: automatic
>>>>>  nouveau  [     CLK][0000:02:00.0] --: core 566 MHz shader 1400 MHz memory 399 MHz
>>>>>  nouveau  [     DRM] resuming client object trees...
>>>>>  nouveau  [     DRM] resuming display...
>>>>>  nouveau 0000:02:00.0: no hotplug settings from platform
>>>>>  nouveau 0000:02:00.0: no hotplug settings from platform
>>>>>
>>>>> Logs(dmesg) are literally identical.
>>>>>
>>>>> 3.15.10-201.fc20.x86_64 - nouveau(fb) resume & thaw PASSED, and that's all what works.
>>>>>         Kernels >= 3.16 - nouveau(fb) resume & thaw BROKEN
>>>>>             ALL Kernels - vesa(fb)    resume & thaw BROKEN.
>>>>>
>>>>>
>>>>
>>>> Excusez-moi,
>>>> BROKEN == The display remains OFF.
>>>>
>>>
>>> More precisely stated it looks like this:
>>>
>>> - Last kernel with working resume/thaw
>>>   http://koji.fedoraproject.org/koji/buildinfo?buildID=538208
>>>   kernel-3.16.0-0.rc0.git9.1.fc21
>>>   2014-06-13
>>>
>>> - First kernel with broken resume/thaw
>>>   kernel-3.16.0-0.rc0.git10.1.fc21
>>>   http://koji.fedoraproject.org/koji/buildinfo?buildID=538244
>>>   http://pkgs.fedoraproject.org/repo/pkgs/kernel/patch-3.15-git10.xz/f98fb42e79c966f8e139e27b61d933e0/patch-3.15-git10.xz
>>>   2014-06-13
>>>
>>> The only difference in dmesg between working and broken kernel module with drm.debug=14 is
>>>  [drm:drm_helper_hpd_irq_event] [CONNECTOR:18:DVI-I-1] status updated from connected to connected
>>>  [drm:drm_helper_hpd_irq_event] [CONNECTOR:18:DVI-I-1] status updated from connected to connected
>>>
>>> - git commit probably introduced breakage
>>>   drm/nouveau/disp: add internal representaion of output paths and connectors
>>>   https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/drivers/gpu/drm?id=7a014a
>>>   https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/commit/drivers/gpu/drm/nouveau?id=7a014a
>>>   2014-06-11
>>
>> Do you have any special reason to believe this to be the culprit
>> (besides the fact that it has something to do with hpd)? Can you do an
>> actual bisect to confirm?
>>
>> Thanks,
>>
>>   -ilia
>
> "hpd" what?

hpd = hotplug detect

> First Fedora Kernel with broken video resume/thaw aka FFKWBVRT i.e. kernel-3.16.0-0.rc0.git10.1.fc21
> comes with 'patch-3.15-git10.xz' which closely resembles "git/commit/drivers/gpu/drm/nouveau?id=7a014a".
> Simple as that.

I see. So no reason to believe that it's not e.g. 20014cb or 377b1f1
or any one of the other patches pulled in by merge commit bc1dfff04a?

> Why are you asking me that bisect formality, man is it not enough that I tested two dozen kernels.

Doing a bisect would involve half as many kernels... Knowing the exact
commit that breaks things is a fairly useful debug tactic, and
drastically increases the chances that a problem gets fixed. Not sure
why you're referring to it as a formality.

git bisect start v3.16-rc1 v3.15 -- drivers/gpu/drm/nouveau

(Pro tip: use a config tailored to your machine rather than a distro
config if you want to spend 5 min/compile instead of 1 hour/compile)

Or you can wait and hope that someone else will have the same problem
and works out the commit that causes it.

Or you can see if it has already been fixed in the latest kernels. Or
try the experimental repo at
http://cgit.freedesktop.org/~darktama/nouveau/ (a bit of a pain to
build, unfortunately, you need some unknown kernel version to build
against, usually ~latest works).

Good luck,

  -ilia