[Bug 66963] Rv6xx dpm problems

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=66963

Zetok <zetok at openmailbox.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |zetok at openmailbox.org

--- Comment #288 from Zetok <zetok at openmailbox.org> ---
Created attachment 126164
  --> https://bugs.freedesktop.org/attachment.cgi?id=126164&action=edit
fixes GPU freeze by reverting 02376d8282b88f07d0716da6155094c8760b1a13 on
4.6.3, tested with r9 290

(In reply to Kajzer from comment #269)
> Done :
> 
> git bisect start '--' 'drivers/gpu/drm/radeon'
> # good: [19583ca584d6f574384e17fe7613dfaeadcdc4a6] Linux 3.16
> git bisect good 19583ca584d6f574384e17fe7613dfaeadcdc4a6
> # bad: [bfe01a5ba2490f299e1d2d5508cbbbadd897bbe9] Linux 3.17
> git bisect bad bfe01a5ba2490f299e1d2d5508cbbbadd897bbe9
> # bad: [03f62abd112d5150b6ce8957fa85d4f6e85e357f] drm/radeon: split PT setup
> in more functions
> git bisect bad 03f62abd112d5150b6ce8957fa85d4f6e85e357f
> # bad: [391bfec33cd4e103274f197924d41ef648b849de] drm/radeon: remove visible
> vram size limit on bo allocation (v4)
> git bisect bad 391bfec33cd4e103274f197924d41ef648b849de
> # good: [da9976206c15178eeae1b4445c9266125bf35b0a] drm/radeon: enable
> display scaling on all connectors (v2)
> git bisect good da9976206c15178eeae1b4445c9266125bf35b0a
> # good: [380670aebfca998bb67b9cf05fc7f28ebeac4b18] drm/radeon: Demote 'BO
> allocation size too large' message to debug only
> git bisect good 380670aebfca998bb67b9cf05fc7f28ebeac4b18
> # bad: [02376d8282b88f07d0716da6155094c8760b1a13] drm/radeon: Allow
> write-combined CPU mappings of BOs in GTT (v2)
> git bisect bad 02376d8282b88f07d0716da6155094c8760b1a13
> # good: [77497f2735ad6e29c55475e15e9790dbfa2c2ef8] drm/radeon: Pass GART
> page flags to radeon_gart_set_page() explicitly
> git bisect good 77497f2735ad6e29c55475e15e9790dbfa2c2ef8
> # first bad commit: [02376d8282b88f07d0716da6155094c8760b1a13] drm/radeon:
> Allow write-combined CPU mappings of BOs in GTT (v2)
> 
> commit 02376d8282b88f07d0716da6155094c8760b1a13
> Author: Michel Dänzer <michel.daenzer at amd.com>
> Date:   Thu Jul 17 19:01:08 2014 +0900
> 
>     drm/radeon: Allow write-combined CPU mappings of BOs in GTT (v2)
>     
>     v2: fix rebase onto drm-fixes
>     
>     Signed-off-by: Michel Dänzer <michel.daenzer at amd.com>
>     Reviewed-by: Christian König <christian.koenig at amd.com>
>     Signed-off-by: Alex Deucher <alexander.deucher at amd.com>

Thank you for the great work with bisecting!

My box has been having ~constant hangs while playing games on all the kernels
I've used with my new GPU, R9 290. The behavior seems to have gotten worse over
newer mesa/kernel versions, to the point where playing just a few minutes could
result in a "hard" system hang – black screen, no response to `reisub`. Or it
could "just" result in radeon driver crash… Well, from that it was at least
possible to reboot with `reisub`. Not that it was a nice thing.

Anyway, the kernel on which it was reproducible usually within <30min of play
(quite often <10 min of play) was 4.6.3.

I've reverted 02376d8282b88f07d0716da6155094c8760b1a13 on checked out 4.6.3,
copied my Gentoo kernel config, and to my surprise, my resolving of revert
conflicts not only compiled, but booted, and after a few (>5) hours of playing,
I can say that I'm fairly sure that revert makes hangs disappear. Of course
I'll be further "testing", but that's it for me today.

Slight note regarding performance – I've noticed that when playing with patch
reverted there sometimes occur slight microstutters – i.e. tops 0.2-0.3s long,
barely noticeable, and not a problem, given that they were happening rarely,
and box finally was not freezing into oblivion when playing games.


Anyway; there are 2 problems I have with GPU – one is that resuming DPM fails
which causes slight (10s) freeze, and the other that the reverted patch
introduced, is freeze/crashing once resuming DPM fails. Crashing/freeze is gone
after revert, but the resuming DPM still fails. Given that freeze/crash is gone
I don't really care though.

I'm attaching the patch that reverts 02376d8282b88f07d0716da6155094c8760b1a13
for 4.6.3.

I'll also attach the dmsg output without commit reverted, where it crashed(?),
and dmsg output with commit reverted where resuming DPM still fails, but the
freeze/crash is gone.

Note that I don't know C, and I have no idea what reverting the patch actually
does (aside from fixing stuff for me).



Btw,

(In reply to bugs.freedesktop.org from comment #287)
> The following commands from comment #239 seems to have corrected the issue
> on my Raden HD3650 Mobility:
> 
> >> echo high > /sys/class/drm/card0/device/power_dpm_force_performance_level
> >> echo performance > /sys/class/drm/card0/device/power_dpm_state
> 
> This seems to be equivalent to the patch in comment #244. Any chance of
> getting this patch into the kernel?
> 
> If anyone is willing to write a fix, as opposed to a workaround, for this
> issue, I would be happy to test it on my device.

Yeah, no, that didn't help a bit.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/dri-devel/attachments/20160901/f53debc3/attachment.html>