GPU hangs while gaming

Erotaku · 2021-04-19 18:12:18

seth wrote:

I tested everything else
Incl. the amdgpu driver on xorg (no wayland/xwayland)?

I only tested the performance side of things. But I did go and test the stability side on xorg too after reading your reply.
Ah... So, no GPU hanging using the opensource drivers after playing for 1+ hours. But for completeness sake, I went back to sway/wayland and tested opensource drivers. And after another 1+ hours.. There is no GPU hanging here as well?
Okay, what the hell... how? I didn't change a thing from yesterday, and it was clearly crashing right before I went to sleep after a max of a 15 minute session.

I... just don't know anymore. I am already quite sick today, and I feel like my head just can not process things right now.

Did you install the amdgpu-pro-libgl & lib32-amdgpu-pro-libgl packages and what does vulkaninfo tell you (idk. whether the pro vulkan libs work w/o the GL ones or xf86-video-amdgpu but it's kinda suspicious that your performance dies and the stbility improves w/ that driver which makes me wonder whether you're actually running the sw emulation. Perhaps game™ allows you to query the used renderer?)

I only installed vulkan-amdgpu-pro aur package and nothing more.
vulkaninfo: https://pastebin.com/4AkEQ2Hy

Lone_Wolf wrote:

For clarity : are all games that hang using vulkan ?

Natively vulkan? Definitely not. One uses dx12 (I switched to dx11 a few times there though. But I went back to dx12 because of the extra smoothness) and the other is dx11.
DXVK kind of vulkan? That's a yes.

Last edited by Erotaku (2021-04-19 18:17:49)

Lone_Wolf · 2021-04-20 09:44:33

The vulkaninfo output kinda suggests you have more then just the vulkan amdgpu-pro drivers, possibly because of mesa-git or mesa-minimal-git presence .

post output of

$ ls /usr/share/vulkan/icd.d/
$ pacman -Qs vulkan

Last edited by Lone_Wolf (2021-04-20 10:15:52)

Erotaku · 2021-04-20 13:50:46

Lone_Wolf wrote:

The vulkaninfo output kinda suggests you have more then just the vulkan amdgpu-pro drivers, possibly because of mesa-git or mesa-minimal-git presence .

post output of
$ ls /usr/share/vulkan/icd.d/
$ pacman -Qs vulkan

Oh yea, I did notice I have a few icd files that I do not recognise. Could that have any effect? Anyways, here you go:

amd_pro_icd64.json
intel_icd.i686.json
intel_icd.x86_64.json
lvp_icd.x86_64.json
radeon_icd.i686.json
radeon_icd.x86_64.json
virtio_icd.x86_64.json

 local/lib32-mesa-git 21.2.0_devel.138334.c74d93cf018-1
    an open-source implementation of the OpenGL specification, git version
local/lib32-vulkan-icd-loader 1.2.174-1
    Vulkan Installable Client Driver (ICD) Loader (32-bit)
local/mesa-git 21.2.0_devel.138325.a1653854f50-1
    an open-source implementation of the OpenGL specification, git version
local/spirv-tools 2020.7-1 (vulkan-devel)
    API and commands for processing SPIR-V modules
local/vulkan-amdgpu-pro 20.45_1188099-5 (Radeon_Software_for_Linux)
    AMDGPU Pro Vulkan driver
local/vulkan-icd-loader 1.2.174-2
    Vulkan Installable Client Driver (ICD) Loader
local/vulkan-tools 1.2.172-1 (vulkan-devel)
    Vulkan Utilities and Tools

Lone_Wolf · 2021-04-21 10:22:13

intel_icd.i686.json & radeon_icd.i686.json comes from lib32-mesa-git and are only used by 32-bit applications .

amd_pro_icd64.json comes from vulkan-amdgpu-pro .
intel_icd.x86_64.json, lvp_icd.x86_64.json, radeon_icd.x86_64.json & virtio_icd.x86_64.json all come from mesa-git .

Assuming the 'hanging' games are 64-bit the vulkan drivers that matter to you are amd_pro_icd64.json and radeon_icd.x86_64.json

I suggest you start the 'hanging' games with

$ VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/radeon_icd.x86_64.json start-command-for-game

$ VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/amd_pro_icd64.json start-command-for-game

That should help to figure out which driver works best for you and narrow down causes of hanging.

Erotaku · 2021-04-21 22:22:52

Lone_Wolf wrote:

intel_icd.i686.json & radeon_icd.i686.json comes from lib32-mesa-git and are only used by 32-bit applications .
amd_pro_icd64.json comes from vulkan-amdgpu-pro .
intel_icd.x86_64.json, lvp_icd.x86_64.json, radeon_icd.x86_64.json & virtio_icd.x86_64.json all come from mesa-git .
Assuming the 'hanging' games are 64-bit the vulkan drivers that matter to you are amd_pro_icd64.json and radeon_icd.x86_64.json
I suggest you start the 'hanging' games with
That should help to figure out which driver works best for you and narrow down causes of hanging.

That is pretty much what I have been doing through lutris for a few days now, indeed. The radeon_icd.86_64 was the one that caused the crashes, and the amd_pro_icd64 was the one that worked well (albeit with a performance hit that did get better over time, but not 100%)
At least that was case until two days ago. While I haven't had as much time to play extensively as I had wanted, but after at least 4 hour of playing, I experienced no GPU hang whatsoever. Which is driving me crazy, to be honest.

From my experence in the Linux world, nothing happens without a reason. Every change and command you execute has an effect, and nothing happens out of thin air. Why did this stop? maybe it will happen again on the long run? If it's really fixed, then what exactly was the cause? Absolutely no idea.

I am thinking of starting to rollback one by one change that I have done over the course of the 10 days of tinkering I did. I think the main ones I really want to be free from are the git llvm-minimal and lib32 packages (you are awesome for maintaining them), and replace it with the upstream versions. I am comfortable with the mesa-git package though. Then test from that point forward.

Would that be safe? Or would it break anything?

Lone_Wolf · 2021-04-23 10:45:09

https://schlockmercenary.fandom.com/wiki/The_Seventy_Maxims_of_Maximally_Effective_Mercenaries wrote:

70. Failure is not an option - it is mandatory. The option is whether or not to let failure be the last thing you do

That mainly depends on how well you understand pacman / makepkg .

Are you booting to console or graphical environment ?
If the 2nd check https://wiki.archlinux.org/index.php/Sy … _boot_into so you know how to boot to console if needed.

If you're building mesa-git in your normal environment, make sure only llvm-libs-** is present, no other llvm or clang packages .

mesa-git and lib32-mesa-git needs to be same versions (or very close) and built against same llvm variant .
If building on live systems, this can cause problems on installing mesa-git with an older / using different llvm lib32-mesa-git .
Removing lib32-mesa-git will likely fail but can be circumenvented by using a special pacman option .
Check man pacman for nodeps description .

Building in a clean chroot avoids those issues but introduces other ones[1].

[1]
- I never figured out how to pass the value of a a user-set environment variable to extra-x86_64-build.
my workaround is to temporarily hardcode the desired value in the PKGBUILD

- using devtools to build against non-repo packages requires using makechrootpkg -I option with absolute paths .
It's very sensitive for typos.

Erotaku · 2021-04-27 19:51:12

Lone_Wolf wrote:

https://schlockmercenary.fandom.com/wiki/The_Seventy_Maxims_of_Maximally_Effective_Mercenaries wrote:
70. Failure is not an option - it is mandatory. The option is whether or not to let failure be the last thing you do

Thank you for the inspiration quote. I loved it.

I would like to apologize for replying with an update this late. Life has taken an oath to frustrate me as much as possible lately. So, I was busy with other issues I had to attend to.

Anyway, I have successfully reverted back to the stable branch (except mesa-git) for the packages installed in question. All I did was remove with a pacman -Rndd each package along its lib32 variant at at a time. Not that it made any difference with this approach, but just in the off chance this prevented something bad...
Sooo, yeah... Everything is still working fine. I can not get any game to crash or cause any GPU hangs. Except one game that does just that the moment I launch it. It's just an instant GPU hang for me. But I don't think it's anything related, because that only happens when I apply a dx9 to dx12 mod to the game. Removing the mod gets the game to work perfectly.

I am definitely glad that the problem is solved... I think. But still, a big part of me wants to know what exactly happened so I can arm myself with the needed knowledge for whatever the future may bring in my linux adventure. I think the last plausible theory would be like what one guy told me on discord. He basically said it's a corrupt shader, and clearing cache alongside installing a different llvm fixes it. That could very well be it. I don't see anything else being the culprit... But what do I know at this point.

I will also revert back to linux-firmware from the git version. And that would be the last git-variant package I have changed in an effort to fix this problem.

So... for now, thank you all very much once again. I really hope this is over once and for all. And I hope I don't bump this thread again any time soon!

Arch Linux

#26 2021-04-19 18:12:18

Re: GPU hangs while gaming

#27 2021-04-20 09:44:33

Re: GPU hangs while gaming

#28 2021-04-20 13:50:46

Re: GPU hangs while gaming

#29 2021-04-21 10:22:13

Re: GPU hangs while gaming

#30 2021-04-21 22:22:52

Re: GPU hangs while gaming

#31 2021-04-23 10:45:09

Re: GPU hangs while gaming

#32 2021-04-27 19:51:12

Re: GPU hangs while gaming

Board footer