Bug 109242 - [RADV] The Witcher 3 system freeze
Summary: [RADV] The Witcher 3 system freeze
Status: NEW
Alias: None
Product: Mesa
Classification: Unclassified
Component: Drivers/Vulkan/radeon (show other bugs)
Version: 18.3
Hardware: x86-64 (AMD64) Linux (All)
: medium normal
Assignee: mesa-dev
QA Contact: mesa-dev
Depends on:
Reported: 2019-01-07 15:57 UTC by Dmitry
Modified: 2019-01-13 19:04 UTC (History)
0 users

See Also:
i915 platform:
i915 features:

dmesg via ssh (73.91 KB, text/plain)
2019-01-09 02:21 UTC, Roman Elshin
game save (696.58 KB, application/x-7z-compressed)
2019-01-09 02:22 UTC, Roman Elshin
r600_debug_info (2.52 KB, text/plain)
2019-01-09 10:54 UTC, Dmitry
VGT param hack (596 bytes, patch)
2019-01-09 13:54 UTC, Samuel Pitoiset
Details | Splinter Review
Possible fix partial_vs_wave (2.23 KB, patch)
2019-01-12 20:26 UTC, Bas Nieuwenhuizen
Details | Splinter Review

Note You need to log in before you can comment on or make changes to this bug.
Description Dmitry 2019-01-07 15:57:09 UTC
The Witcher 3 freezes the system when using the RADV driver. 
Mesa: 18.3.1
Kernel driver: amdgpu
OS: Arch
GPU: RX460

There is no such problem with AMDVLK.
Comment 1 Samuel Pitoiset 2019-01-07 16:07:16 UTC
Does this is a recent regression for you?
Comment 2 Dmitry 2019-01-08 01:56:31 UTC
Before that, there were no heavy loads, and hangs during active combat.
Comment 3 Samuel Pitoiset 2019-01-08 08:11:32 UTC
Can you explain how to reproduce? Can you upload a savegame? Is there anything in dmesg when it hangs?
Comment 4 Dmitry 2019-01-08 13:37:54 UTC
Any heavy scene, for example fight with the first Griffin on a plot.

Yes I can. It hangs in the fight.

dmesg is not done because I do a physical reboot.
Comment 5 Samuel Pitoiset 2019-01-08 16:50:35 UTC
Where is the savegame? Also what kernel and llvm versions are you using? The ones provided by Arch Linux?

I have just tried to reproduce GPU hangs on my side by playing around few minutes, no success.
Comment 6 Dmitry 2019-01-08 18:00:27 UTC
Saving the game is not important, I started the game first and realized that hangs in any battles. Problem new.

LLVM 7.0.1-1

I understand this is a problem with the RX460. On bug tracker has several reports about The Witcher 3. And even with my RX460
Comment 7 Roman Elshin 2019-01-09 02:21:23 UTC
Created attachment 143024 [details]
dmesg via ssh
Comment 8 Roman Elshin 2019-01-09 02:22:54 UTC
Created attachment 143025 [details]
game save
Comment 9 Roman Elshin 2019-01-09 02:23:41 UTC
It looks like i have a same issue with rx560.
Comment 10 Samuel Pitoiset 2019-01-09 09:26:28 UTC
Just installed your savegame, when do you have a GPU hang? When fighting with that creature?
Comment 11 Roman Elshin 2019-01-09 09:49:00 UTC
Yes, usually  hangs happens after just the first strike to it.
Comment 12 Samuel Pitoiset 2019-01-09 10:05:23 UTC
I tried multiple times, no hangs so far. What graphics settings are you using?

I tested on my RX480 with LLVM 7.0.1 and Mesa 18.3.1.
Comment 13 Dmitry 2019-01-09 10:36:37 UTC
Medium preset, 1080p, HairWorks off
Comment 14 Samuel Pitoiset 2019-01-09 10:41:13 UTC
Can you attach the output of "R600_DEBUG=info glxgears" ?
Comment 15 Dmitry 2019-01-09 10:54:57 UTC
Created attachment 143033 [details]
Comment 16 Samuel Pitoiset 2019-01-09 11:21:50 UTC
Is the GPU hang 100% reproducible for you guys? Maybe, it only affects Polaris11 and Polaris12. (I only have Polaris10)
Comment 17 Dmitry 2019-01-09 12:00:43 UTC
Yes, always.
On AMDVLK hangs too, but then returns to normal operation.
Comment 18 Samuel Pitoiset 2019-01-09 12:41:53 UTC
In your first comment you said "There is no such problem with AMDVLK."... So, it also hangs with AMDVLK but recovery happens while with RADV the whole machine freezes and you need a hard reboot?
Comment 19 Dmitry 2019-01-09 12:45:28 UTC
C AMDVLK hangs only the game client, and with RADV the whole system. But it is possible and the system hangs for seconds, have not checked Alt + Tab.
Comment 20 Samuel Pitoiset 2019-01-09 13:54:26 UTC
Created attachment 143035 [details] [review]
VGT param hack

Can you try the attached patch? I really doubt it will fix the problem but it's worth trying.
Comment 21 Roman Elshin 2019-01-09 15:38:25 UTC
It looks like it fixes problem, thanks!
Comment 22 Dmitry 2019-01-09 16:01:05 UTC
I don't know how to use it.
Comment 23 Samuel Pitoiset 2019-01-09 16:05:16 UTC
(In reply to Dmitry from comment #22)
> I don't know how to use it.

You need to build mesa from git [1]


That said, it's actually unexpected that this hack fixes the problem.
Comment 24 Dmitry 2019-01-09 17:00:56 UTC
Mesa-git I can take from AUR. But I still don't understand how to use hook.
Comment 25 Roman Elshin 2019-01-09 17:11:16 UTC
But it definitely works here (just tested several other places with fights - no hangs any more), are there any side effects expected?

I  think must be a way to add applying this patch in PKGBUILD (I am not familiar with arch)
Comment 26 Bas Nieuwenhuizen 2019-01-12 11:52:39 UTC
Setting those options when they are not needed might slow things down.

I think this is the proper fix:

Comment 27 Roman Elshin 2019-01-12 14:05:56 UTC
>I think this is the proper fix:

It looks like this not fix a problem (but makes it less reproducible).
Comment 28 Bas Nieuwenhuizen 2019-01-12 20:26:33 UTC
Created attachment 143082 [details] [review]
Possible fix partial_vs_wave

So it seems, looks like it is random for me :(

I tried the attached patch 10 times without hanging on the boss from the savefile, could you test whether this fixes it for you?
Comment 29 Roman Elshin 2019-01-13 05:37:01 UTC
>Created attachment 143082 [details] [review] [review]
>Possible fix partial_vs_wave
>So it seems, looks like it is random for me :(

Yes, these hangs looks very random now (or it other hangs which also hides with  "VGT param hack").
Comment 30 Roman Elshin 2019-01-13 10:22:42 UTC
>Created attachment 143082 [details] [review] [review]

Are both patches must be applied? (it was tested only first or second, not together).
Comment 31 Roman Elshin 2019-01-13 19:04:55 UTC
>Are both patches must be applied? (it was tested only first or second, not >together).

With both patches applied i can't reproduce this issue, thanks!

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.