Bug 200607

Summary:	[amdgpu] Polaris10 driver crash with DC enabled
Product:	Drivers	Reporter:	Parker Reed (parker.l.reed)
Component:	Video(DRI - non Intel)	Assignee:	drivers_video-dri
Status:	NEW ---
Severity:	normal	CC:	alexdeucher, harry.wentland, nicholas.kazlauskas
Priority:	P1
Hardware:	All
OS:	Linux
Kernel Version:	4.17.6	Subsystem:
Regression:	No	Bisected commit-id:
Attachments:	journalctl log log2 log3

Description Parker Reed 2018-07-19 18:59:16 UTC

Created attachment 277413 [details]
journalctl log

https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/commit/?id=34116bf2fe92c5352663250ca11aa78d8e00c414

I tried the new firmware from there and the amdgpu driver now sporadically crashes. 

Jul 19 14:17:04 e55 kernel: [drm:drm_atomic_helper_wait_for_dependencies [drm_kms_helper]] *ERROR* [PLANE:41:plane-5] flip_done timed out
Jul 19 14:17:04 e55 kernel: [drm:amdgpu_dm_atomic_commit_tail [amdgpu]] *ERROR* amdgpu_dm_commit_planes: acrtc 0, already busy

Hardware is RX 480 8GB 

01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Ellesmere [Radeon RX 470/480/570/570X/580/580X] (rev c7)

Attached log.

Comment 1 Alex Deucher 2018-07-19 19:02:32 UTC

Anything in particular you are doing when it hangs?  Specific app?

Comment 2 Parker Reed 2018-07-19 19:06:53 UTC

I'm streaming via Steam in-home streaming and occasionally a fullscreen application will mess things up. When I reboot, SDDM/Plasma hang because of that driver crash. Leaving the up to date firmware files in there, it comes back after a couple reboots (randomly), and continues to work for a while. For now I have just downgraded back to my distro shipped linux-firmware.

I'm not sure if the fullscreen issue is related but I know it definitely causes SDDM/Plasma init to freak out. So I'd say that's when it hangs the most for me.

Comment 3 Alex Deucher 2018-07-19 19:15:36 UTC

Do you have similar issues with a newer kernel?

Comment 4 Parker Reed 2018-07-19 19:21:10 UTC

I will try from git and report back.

As it turns out the reason I was having to reboot on some fullscreen games freezing was due to this same issue. Both games that were showing that behavior are now working on the older firmware.

Comment 5 Parker Reed 2018-07-19 20:27:16 UTC

Linux e55 4.18.0-rc5-g9ffc59d57228 #1 SMP PREEMPT Thu Jul 19 15:27:18 EDT 2018 x86_64 GNU/Linux

Newly updated linux-firmware (18.20)

log2.txt

Booted fine, went to launch a fullscreen Wine game, threw the error. 

----

Linux e55 4.18.0-rc5-g9ffc59d57228 #1 SMP PREEMPT Thu Jul 19 15:27:18 EDT 2018 x86_64 GNU/Linux

Reverted linux-firmware

log3.txt

This time even after revert SDDM/Plasma failed to load and subsequent driver driver.

I realized that the working state I was in was due to amdgpu.dc=0 and not the version of the firmware. 

Both of those logs are with amdgpu.dc=1 and exhibit the crash no matter the firmware version.

I am currently booted into 18.20 firmware and amdgpu.dc=0 and everything is working.

Comment 6 Parker Reed 2018-07-19 20:27:33 UTC

Created attachment 277421 [details]
log2

Comment 7 Parker Reed 2018-07-19 20:27:48 UTC

Created attachment 277423 [details]
log3

Comment 8 Parker Reed 2018-07-19 20:28:37 UTC

Correction from above:

This time even after revert SDDM/Plasma failed to load and subsequent driver crash.

Comment 9 Michel Dänzer 2018-07-20 07:52:47 UTC

Please change the bug title to reflect that it's related to DC, not firmware.