Arch Linux

Mechanicus

Member

Registered: 2025-01-13

Posts: 121

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

@Amunn thank you for the update!

NuSkool

Member

Registered: 2015-03-23

Posts: 265

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

linux-amdgpu-testing-6.13.2.arch1-3 is ready for testing.

EDIT: on the 6.13.12 1-3 test kernel

@Amunn, Just verifying, are you talking the same (w/ typo) or different kernels here?

6.13.2.arch1-3
6.13.12 1-3

Earlier today I read this as the same kernel. Now later I've noticed the diff.
I've got some testing time available but don't want to waste time on a dead kernel...

---

Scripts I Use :  https://github.com/Cody-Learner
$ grep -m1 'model name' /proc/cpuinfo :  AMD Ryzen 5 PRO 2400GE w/ Radeon Vega Graphics
$ glxinfo | grep Device                               :  Device: AMD Radeon Vega 11 Graphics (radeonsi, raven, ACO, DRM 3.61, 6.13.9-rc1) (0x15dd)
$ sudo dmesg | awk '/drm/ && /gfx/'      :  [    6.427009] [drm] add ip block number 6 <gfx_v9_0>

pacoandres

Member

Registered: 2020-03-05

Posts: 56

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

I have been testing the @Mechanicus latest patches individually, the conclusions are at https://github.com/pacoandres/laikm/issues/23

Amunn

Member

Registered: 2025-02-09

Posts: 17

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

@NuSkool Yes I tested 6.13.2.arch.1-3, was tired and wrote it out wrong.
I tested with the current packages from the repo every time.

Last edited by Amunn (2025-02-11 10:39:38)

Mechanicus

Member

Registered: 2025-01-13

Posts: 121

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

AMDGPU testing

glibc 2.41+r6+gcf88351b685d-1

Build:linux-amdgpu-testing-6.13.2.arch1-4 - freeze confirmed.
Included patches:
- Simplify GFXOFF handling
- Optimize mutex protected blocks in amdgpu_vm_flush
- Make amdgpu_vmid::current_gpu_reset_count atomic_t type

Full pipeline cleanup: all extra mutexes/workarounds removed
Build: linux-amdgpu-testing-6.13.2.arch1-5 - freezes.
Included patches:
- Optimize mutex protected blocks in amdgpu_vm_flush
- Make amdgpu_vmid::current_gpu_reset_count atomic_t type
- Make amdgpu_fence_driver::sync_seq atomic type
- Simplify GFXOFF handling v2
- Remove workaround for TLB seq race
- Remove jpeg1&vcn1 hardware bug workaround

GitHub

What to check:
- stability
If stable:
- temperature
- average package power (sudo turbostat -s PkgWatt)
- performance - for example WebGL Aquarium fps for different amount of fish

Kernel option to keep during testing period: fsck.mode=force

laikm from pacoandres for reporting issues

Last edited by Mechanicus (2025-02-13 13:48:52)

OJaksch

Member

From: Germany

Registered: 2016-02-23

Posts: 40

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

Just thought i would say that after 6 days i had no crashes with the latest Mesa from @Lone_Wolf in this post https://bbs.archlinux.org/viewtopic.php … 0#p2223890, which if i got it right, it will be the regular Mesa 25.0 when it's released, and i'm constantly spawning new windows and i use mpv a lot. 

I know it's not the end of it as people are doing a lot of work with these kernel builds but at least i think i have a stable system now with no more crashes so i would like to thank everyone for that.

Installed this today - and nothing else - and have worked the whole day w/o any issues, so day #1 successfully.

Amunn

Member

Registered: 2025-02-09

Posts: 17

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

Currently testing 6.13.2.arch1-4, no crashes for around 3 h of use and 5h since boot.

I feel this is too early to conclude since I ran previous tests for up to 10h before crashing.

This seems like a performance upgrade over the mainline kernel, and almost as good as previous tests (might be just a feeling, as I have no concrete proof to back it up).
The fullscreen video thing (Steam store) I mentioned earlier is back, but it's better than it used to be.

Fish test:
fps: 60
canvas width: 1024
canvas height: 1024
idle: 2.5W
10k: 60 fps 11.4W
15k: 50 fps 12.8W
20k: 38 fps 12.8W
30k: 27 fps 12.5W
Highest temp reached = 61C
Idle = 42C (running stock cooler slow to reduce noise)

EDIT: 4 more hours of light use, no crash yet. Will report again tomorrow.

Last edited by Amunn (2025-02-12 00:04:00)

pacoandres

Member

Registered: 2020-03-05

Posts: 56

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

linux-amdgpu-testing-6.13.2.arch1-4 got a freeze in 5 min while reading documentation with firefox. No other apps opened.

The only log line is

feb 12 12:02:35 monelle kernel: amdgpu 0000:09:00.0: amdgpu: Dumping IP State

Amunn

Member

Registered: 2025-02-09

Posts: 17

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

So the linux-amdgpu-testing-6.13.2.arch1-4 kernel crashed after around 10h of light use over three boots.

I looked at the logs, and saw something possibly related 30min before the crash:

Feb 12 15:59:10 user fuzzel[12901]: png: libpng: iCCP: known incorrect sRGB profile
Feb 12 15:59:48 user fuzzel[13080]: png: libpng: iCCP: known incorrect sRGB profile
Feb 12 15:59:48 user fuzzel[13080]: png: libpng: iCCP: known incorrect sRGB profile
Feb 12 15:59:48 user fuzzel[13080]: png: libpng: iCCP: known incorrect sRGB profile
Feb 12 15:59:56 user fuzzel[13252]: png: libpng: iCCP: known incorrect sRGB profile
Feb 12 15:59:56 user fuzzel[13252]: png: libpng: iCCP: known incorrect sRGB profile
Feb 12 15:59:56 user fuzzel[13252]: png: libpng: iCCP: known incorrect sRGB profile
Feb 12 16:00:01 user fuzzel[13280]: png: libpng: iCCP: known incorrect sRGB profile
Feb 12 16:00:01 user fuzzel[13280]: png: libpng: iCCP: known incorrect sRGB profile
Feb 12 16:00:01 user fuzzel[13280]: png: libpng: iCCP: known incorrect sRGB profile
Feb 12 16:00:50 user systemd-coredump[14016]: Process 14010 (spotify) of user 1000 terminated ab>
Feb 12 16:00:50 user systemd[1]: Created slice Slice /system/systemd-coredump.
Feb 12 16:00:50 user systemd[1]: Started Process Core Dump (PID 14016/UID 0).
Feb 12 16:00:50 user systemd-coredump[14025]: Process 14017 (spotify) of user 1000 terminated ab>
Feb 12 16:00:50 user systemd[1]: Started Process Core Dump (PID 14025/UID 0).
Feb 12 16:00:50 user (sd-parse-elf)[14030]: Could not parse number of program headers from core >
Feb 12 16:00:50 user systemd-coredump[14027]: [?] Process 14017 (spotify) of user 1000 dumped co>
                                                     
                                                     Module [dso] without build-id.
                                                     Module [dso]
                                                     Stack trace of thread 14017:
                                                     #0  0x0000765e2a45a624 n/a (n/a + 0x0)
                                                     #1  0x0000765e2a400ba0 n/a (n/a + 0x0)
                                                     #2  0x0000765e2a3e8582 n/a (n/a + 0x0)
                                                     #3  0x0000765e2a3e93bf n/a (n/a + 0x0)
                                                     #4  0x0000765e2a4e8419 n/a (n/a + 0x0)
                                                     #5  0x0000765e2a4e9714 n/a (n/a + 0x0)
                                                     #6  0x00006502f224cd70 n/a (n/a + 0x0)
                                                     ELF object binary architecture: AMD x86-64
Feb 12 16:00:50 user systemd-coredump[14024]: [?] Process 14010 (spotify) of user 1000 dumped co>
                                                     
                                                     Stack trace of thread 14010:
                                                     #0  0x0000765e2a45a624 n/a (n/a + 0x0)
                                                     #1  0x0000765e2a400ba0 n/a (n/a + 0x0)
                                                     #2  0x0000765e2a3e8582 n/a (n/a + 0x0)
                                                     #3  0x0000765e2a3e93bf n/a (n/a + 0x0)
                                                     #4  0x0000765e2a4e8419 n/a (n/a + 0x0)
                                                     #5  0x0000765e2a4e9714 n/a (n/a + 0x0)
                                                     #6  0x00006502f224cd70 n/a (n/a + 0x0)
                                                     ELF object binary architecture: AMD x86-64
Feb 12 16:00:50 user systemd[1]: systemd-coredump@1-14025-0.service: Deactivated successfully.
Feb 12 16:00:50 user systemd[1]: systemd-coredump@1-14025-0.service: Consumed 312ms CPU time, 13>
Feb 12 16:00:50 user systemd[1]: systemd-coredump@0-14016-0.service: Deactivated successfully.
Feb 12 16:00:50 user systemd[1]: systemd-coredump@0-14016-0.service: Consumed 311ms CPU time, 13>
Feb 12 16:01:50 user systemd-coredump[14050]: Process 14043 (spotify) of user 1000 terminated ab>
Feb 12 16:01:50 user systemd[1]: Started Process Core Dump (PID 14050/UID 0).
Feb 12 16:01:50 user systemd-coredump[14051]: [?] Process 14043 (spotify) of user 1000 dumped co>
                                                     
                                                     Stack trace of thread 14043:
                                                     #0  0x00007cf7ab85a624 n/a (n/a + 0x0)
                                                     #1  0x00007cf7ab800ba0 n/a (n/a + 0x0)
                                                     #2  0x00007cf7ab7e8582 n/a (n/a + 0x0)
                                                     #3  0x00007cf7ab7e93bf n/a (n/a + 0x0)
                                                     #4  0x00007cf7ab8e8419 n/a (n/a + 0x0)
                                                     #5  0x00007cf7ab8e9714 n/a (n/a + 0x0)
                                                     #6  0x000059a796c28d70 n/a (n/a + 0x0)
                                                     ELF object binary architecture: AMD x86-64
Feb 12 16:01:50 user systemd[1]: systemd-coredump@2-14050-0.service: Deactivated successfully.
Feb 12 16:01:50 user systemd[1]: systemd-coredump@2-14050-0.service: Consumed 301ms CPU time, 14>
Feb 12 16:23:05 user kernel: amdgpu 0000:0a:00.0: amdgpu: Dumping IP State
Feb 12 16:23:26 user pipewire[842]: spa.alsa: front:2p: (0 suppressed) snd_pcm_avail after recov>
Feb 12 16:23:26 user pipewire[842]: spa.alsa: front:2p: snd_pcm_mmap_commit error: Broken pipe
Feb 12 16:23:33 user pipewire[842]: spa.alsa: front:2p: snd_pcm_mmap_commit error: Broken pipe
Feb 12 16:23:37 user pipewire[842]: spa.alsa: front:2p: (10 suppressed) snd_pcm_avail after reco>
Feb 12 16:23:42 user kernel: amdgpu 0000:0a:00.0: amdgpu: failed to write reg 28b4 wait reg 28c6
Feb 12 16:23:59 user kernel: amdgpu 0000:0a:00.0: amdgpu: failed to write reg 1a6f4 wait reg 1a7>

Some irrelevant stuff removed for clarity.

OJaksch

Member

From: Germany

Registered: 2016-02-23

Posts: 40

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

Just thought i would say that after 6 days i had no crashes with the latest Mesa from @Lone_Wolf in this post https://bbs.archlinux.org/viewtopic.php … 0#p2223890, which if i got it right, it will be the regular Mesa 25.0 when it's released, and i'm constantly spawning new windows and i use mpv a lot. 

I know it's not the end of it as people are doing a lot of work with these kernel builds but at least i think i have a stable system now with no more crashes so i would like to thank everyone for that.

Installed this today - and nothing else - and have worked the whole day w/o any issues, so day #1 successfully.

2nd day w/o any issues! Still nothing else installed except @Lone_Wolf 's "25.0 binary", no special kernel parameters, no nothing

NuSkool

Member

Registered: 2015-03-23

Posts: 265

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

Report: linux-amdgpu-testing 6.13.2.arch1-4

Ran for ~4HR before freezing.

Start date time   : Feb 11, 2025  12:10PM
Finish date time : Feb 11, 2025 ~04:00PM

pacman -Q linux-amdgpu-testing : linux-amdgpu-testing 6.13.2.arch1-4
uname -rs                                          : Linux 6.13.2-arch1-4-amdgpu-testing
pacman -Q glibc                               : glibc 2.41+r6+gcf88351b685d-1
pacman -Q mesa                              : mesa 1:24.3.4-1
cat /proc/cmdline                             : ... rw loglevel=3 sysrq_always_enabled=1 amd_pstate=passive fsck.mode=force

When the freeze occurred, I had just installed 'fbterm' and trying to get 'qterminal' running in it, in different tty than my desktop session.
Core dump:

systemd-coredump[5040]: Process 5038 (qterminal) of user 1000 terminated abnormally with signal 6/ABRT, processing...
systemd[1]: Created slice Slice /system/systemd-coredump.
systemd[1]: Started Process Core Dump (PID 5040/UID 0).
(sd-parse-elf)[5043]: Could not parse number of program headers from core file: invalid `Elf' handle
systemd-coredump[5041]: [?] Process 5038 (qterminal) of user 1000 dumped core.
                                                   
                                                   Module [dso] without build-id.
                                                   Module [dso] without build-id.
                                                   Module [dso] without build-id.
                                                   Module [dso] without build-id.
                                                   Module [dso]
                                                   Module [dso] without build-id.
                                                   Stack trace of thread 5038:
                                                   #0  0x00007d9897aa5624 n/a (n/a + 0x0)
                                                   #1  0x00007d9897a4bba0 n/a (n/a + 0x0)
                                                   #2  0x00007d9897a33582 n/a (n/a + 0x0)
                                                   #3  0x00007d98974905c2 n/a ([dso] + 0x6db5c2)
                                                   #4  0x00007d989749154b n/a ([dso] + 0x6dc54b)
                                                   #5  0x00007d9897cdef5d n/a (n/a + 0x0)
                                                   #6  0x00007d9897d8acd8 n/a (n/a + 0x0)
                                                   #7  0x00007d9897558a7d n/a ([dso] + 0x7a3a7d)
                                                   #8  0x00007d9897d8ad6e n/a (n/a + 0x0)
                                                   #9  0x00007d98986f9ae6 n/a (n/a + 0x0)
                                                   #10 0x000059ec425662d3 n/a (n/a + 0x0)
                                                   #11 0x00007d9897a35488 n/a (n/a + 0x0)
                                                   #12 0x00007d9897a3554c n/a (n/a + 0x0)
                                                   #13 0x000059ec42569655 n/a (n/a + 0x0)
                                                   ELF object binary architecture: AMD x86-64

Last edited by NuSkool (2025-02-12 18:18:45)

---

Scripts I Use :  https://github.com/Cody-Learner
$ grep -m1 'model name' /proc/cpuinfo :  AMD Ryzen 5 PRO 2400GE w/ Radeon Vega Graphics
$ glxinfo | grep Device                               :  Device: AMD Radeon Vega 11 Graphics (radeonsi, raven, ACO, DRM 3.61, 6.13.9-rc1) (0x15dd)
$ sudo dmesg | awk '/drm/ && /gfx/'      :  [    6.427009] [drm] add ip block number 6 <gfx_v9_0>

Mechanicus

Member

Registered: 2025-01-13

Posts: 121

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

linux-amdgpu-testing-6.13.2.arch1-5 is the last thing I can do for now without power-gating state change workaround. In this build 3 mutex iocks/unlocks removed from the pipeline, GFXOFF switch logic refactored and simplified. There is no delay and no extra operations anymore.

Amunn

Member

Registered: 2025-02-09

Posts: 17

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

I'm on 1-5 test now, 1h since boot. Testing more tomorrow.
So far it seems fast somehow, I have instant video playback, haven't experienced this before on my current system.
I tried going through the thumbnails on the steam store with video on, they load and start right when I hover over.

Thanks for the effort btw!

Amunn

Member

Registered: 2025-02-09

Posts: 17

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

I got a crash after about 30min today (1h without a crash yesterday).
edit: on the 6.13.2.arch1-5 kernel
2200G cpu

Feb 13 08:43:56 user kernel: amdgpu 0000:0a:00.0: amdgpu: Dumping IP State
Feb 13 08:44:39 user kernel: amdgpu 0000:0a:00.0: amdgpu: failed to write reg 28b4 wait reg 28c6
Feb 13 08:44:43 user pipewire[847]: pw.node: (alsa_output.pci-0000_0a_00.6.analog-stereo-58) graph xrun not-triggered (1 suppressed)
Feb 13 08:44:43 user pipewire[847]: pw.node: (alsa_output.pci-0000_0a_00.6.analog-stereo-58) xrun state:0x7ee7649ab008 pending:1/1 s:3083772418428 a:3083772425592 f:3083772466481 waiting:7164 process:40>
Feb 13 08:44:45 user pipewire[847]: spa.alsa: front:2p: (0 suppressed) snd_pcm_avail after recover: Broken pipe
Feb 13 08:44:53 user kernel: amdgpu 0000:0a:00.0: amdgpu: failed to write reg 1a6f4 wait reg 1a706

Last edited by Amunn (2025-02-13 07:50:23)

Lone_Wolf

Administrator

From: Netherlands, Europe

Registered: 2005-10-04

Posts: 13,540

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

mesa 25.0.0 rc3 has been released with many fixes for radeonsi.

Although none of the fixes appear to be for raven / raven2 / computing there may be people that prefer a binary closer to the next stable release .

I've uploaded a new mesa 25.0.0-rc3  binary .

---

Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.

clean chroot building not flexible enough ?
Try clean chroot manager by graysky

orbit-oc

Member

Registered: 2024-12-15

Posts: 85

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

Thanks to @Lone_Wolf for the updated package mesa-test-git-25.0.0_rc3.201182.3a8abfa39b7-1-x86_64.

I still ask myself why mesa 24.3.5 which should be released on 05.02.2025 is still a long time coming.
Considering the severity of this bug, a separate version 24.3.4.1 could and should have been released long ago.
In the meantime further distributions are affected...

Mechanicus

Member

Registered: 2025-01-13

Posts: 121

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

AMDGPU testing

glibc 2.41+r6+gcf88351b685d-1

This build contains all optimization patches and powergating workaround. Let's see if it can work any better than the regular kernel.
Build: linux-amdgpu-testing-6.13.2.arch1-6 - stable.
Included patches:
- Optimize mutex protected blocks in amdgpu_vm_flush
- Make amdgpu_vmid::current_gpu_reset_count atomic_t type
- Make amdgpu_fence_driver::sync_seq atomic type
- Simplify GFXOFF handling v2
- Remove workaround for TLB seq race
- Remove jpeg1&vcn1 hardware bug workaround
- Deny GFXOFF in amdgpu_ring_commit using PG state change

GitHub

What to check:
- stability
If stable:
- temperature
- average package power (sudo turbostat -s PkgWatt)
- performance - for example WebGL Aquarium fps for different amount of fish

Kernel option to keep during testing period: fsck.mode=force

laikm from pacoandres for reporting issues

Last edited by Mechanicus (2025-03-01 13:49:55)

Amunn

Member

Registered: 2025-02-09

Posts: 17

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

@Mechanicus Ok, on 1-6 now.
I installed mesa-test-git earlier, am I doing the test correctly if I removed it and installed mesa (24.3.4-1) and vulkan-radeon(same version number)?

pacoandres

Member

Registered: 2020-03-05

Posts: 56

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

I've been testing linux-amdgpu-testing-6.13.2.arch1-6 for about two hours with no freeze. I couldn't do any benchmark yet, maybe tomorrow with a long and heavy test.

@Amunn: if you downgrade mesa and vulkan-radeon to 24.3.4 the test conditions are the same as before you upgraded to git version.

OJaksch

Member

From: Germany

Registered: 2016-02-23

Posts: 40

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

Just thought i would say that after 6 days i had no crashes with the latest Mesa from @Lone_Wolf in this post https://bbs.archlinux.org/viewtopic.php … 0#p2223890, which if i got it right, it will be the regular Mesa 25.0 when it's released, and i'm constantly spawning new windows and i use mpv a lot. 

I know it's not the end of it as people are doing a lot of work with these kernel builds but at least i think i have a stable system now with no more crashes so i would like to thank everyone for that.

Installed this today - and nothing else - and have worked the whole day w/o any issues, so day #1 successfully.

2nd day w/o any issues! Still nothing else installed except @Lone_Wolf 's "25.0 binary", no special kernel parameters, no nothing

3rd day w/o any issues. Installing mesa-test-git-25.0.0_rc3 now and continuing my test series...

Aivar

Member

Registered: 2025-02-07

Posts: 3

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

6.13.2-arch1-6-amdgpu-testing
I couldn't freeze it with anything (except cat /sys/kernel/debug/dri/1/amdgpu_gpu_recover).
Overall, the system is working briskly.

---

Blame the google-translator for all language errors.

Amunn

Member

Registered: 2025-02-09

Posts: 17

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

I still have no crashes, and the aquarium test performance is increased by around 5-10%.
Power consumption is up by 10% as well, but at 14w it's still not much. Will keep testing, light use with many electron based programs open. Thanks for the effort in development of the driver!

pacoandres

Member

Registered: 2020-03-05

Posts: 56

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

I've been working all day with linux-amdgpu-testing-6.13.2.arch1-6 and have no freezes.

Performance, consumption and temperature are sightly worse than with other testing kernels, but as is less than 5% this could be due to other causes.

Thanks for the effort.

OJaksch

Member

From: Germany

Registered: 2016-02-23

Posts: 40

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

3rd day w/o any issues. Installing mesa-test-git-25.0.0_rc3 now and continuing my test series...

4th day w/o any issues - 1st day with mesa-test-git-25.0.0_rc3, but I''ll tend it calling "fixed it for me".

gfx_v9
AMD Ryzen 5 PRO 3400G with Radeon Vega Graphics
Advanced Micro Devices, Inc. [AMD/ATI] Picasso/Raven 2 [Radeon Vega Series / Radeon Vega Mobile Series] [1002:15d8]

Mechanicus

Member

Registered: 2025-01-13

Posts: 121

Re: Issues with Mesa 24.3.x and amdgpu Vega graphics

AMDGPU testing

This build aims to double check the place where the powergating state change workaround should be applied. The crash is also a valid result.
Build: linux-amdgpu-testing-6.13.2.arch1-7 - crashed and confirmed that the workaround is placed correctly in 1-6 version.
Included patches:
- Optimize mutex protected blocks in amdgpu_vm_flush
- Simplify GFXOFF handling
- Add powergating state change workaround for GFX9 v3
- Remove workaround for TLB seq race

Last edited by Mechanicus (2025-02-15 08:17:31)