kern/59629: amdgpu(4): accelerated code may free Xorg, killing it causes crash

[vezhlys%gmail.com@localhost]

>Number:         59629
>Category:       kern
>Synopsis:       amdgpu(4): accelerated code may free Xorg, killing it causes crash
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Tue Sep 02 22:20:00 +0000 2025
>Originator:     Andrius V
>Release:        NetBSD 11.0_BETA
>Organization:
>Environment:
NetBSD agraphic-pc 11.0_BETA NetBSD 11.0_BETA (GENERIC) #0: Mon Sep  1 00:18:27 EEST 2025  andriusv@agraphic-pc:/home/andriusv/workspace/obj/sys/arch/amd64/compile/GENERIC amd64
>Description:
I have two amdgpu(4) "capable" graphics cards: RX460 and RX5700XT. I was mainly using the latter with the NetBSD 10/10.1, since it worked relatively well. However, Xorg was reporting that 2D/3D acceleration was disabled no matter if I would enable in configuration or not.

With upgrade to NetBSD 11 Beta, Xorg started to report that 2D/3D acceleration is enabled (once configured). Unfortunately, as soon as accelerated code is being executed (glxgears, firefox with hardware acceleration, scummvm, supertuxkart, etc), it causes Xorg to freeze and it flashes once with most of the screen turning into artifacts.
 
RX460 has this problem in a lesser extent, but still more often than not it would cause screen to turn into artifacts (stripes all over the screen, with sometimes mouse movement still visible, but nothing can be done, since all screen is just in "colorful stripes").

If accelerated code works sometimes, it does seem to work pretty well (e.g. glxgears shows very high FPS, scummVM games using opengl acceleration work smoothly, etc). Small artifacts still may appear over the place (black rectangles, broken text in browser), but typically doesn't hinder the usage that much.

Once the freeze happens, the system itself doesn't crash. But I can't switched to terminal using Alt+F<1-4> buttons. I can connect to the system using ssh.

In addition to all the problems, the mouse movement became very sluggish by default on amdgpu(4) (with the other problems described in kern/59628, but need to verify if it is amdgpu only).

Attempt to kill X session typically ends up in crash:

[   114.187571] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   114.407569] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   115.167564] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   115.197563] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   115.927558] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   116.207556] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   116.687553] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   117.217549] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   117.447547] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   118.207541] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   118.227541] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   118.937536] {drm:netbsd:amdgpu_job_timedout+0xaa} *ERROR* ring gfx timeout, signaled seq=3316, emitted seq=3316
[   118.937536] {drm:netbsd:amdgpu_job_timedout+0xc7} *ERROR* Process information: process firefox pid 1866 thread firefox:cs0 pid 2805
[   118.937536] amdgpu0: GPU reset begin!
[   118.967536] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   119.217534] amdgpu0: GPU BACO reset
[   119.237534] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   119.577531] warning: [drm] schedsdma0 is not ready, skipping
[   119.577531] warning: [drm] schedsdma1 is not ready, skipping
[   119.577531] amdgpu0: autoconfiguration error: error: failed to clear page tables on GEM object close (-2)
[   119.577531] uvm_fault(0xffff81fd82aa4c80, 0x0, 1) -> e
[   119.577531] fatal page fault in supervisor mode
[   119.577531] trap type 6 code 0 rip 0xffffffff809bc396 cs 0x8 rflags 0x10296 cr2 0x8 ilevel 0 rsp 0xffff88926e222800
[   119.577531] curlwp 0xffff81fd90aa5000 pid 2464.2518 lowest kstack 0xffff88926e21e2c0
[   119.577531] panic: trap
[   119.577531] cpu2: Begin traceback...
[   119.577531] vpanic() at netbsd:vpanic+0x171
[   119.577531] panic() at netbsd:panic+0x3c
[   119.577531] trap() at netbsd:trap+0xb43
[   119.577531] --- trap (number 6) ---
[   119.577531] amdgpu_vm_sdma_commit() at netbsd:amdgpu_vm_sdma_commit+0x3c
[   119.577531] amdgpu_vm_bo_update_mapping() at netbsd:amdgpu_vm_bo_update_mapping+0x6ce
[   119.577531] amdgpu_vm_clear_freed() at netbsd:amdgpu_vm_clear_freed+0xe9
[   119.577531] {drm:netbsd:amdgpu_cs_ioctl+0xe1f} *ERROR* Failed to initialize parser -87!
[   119.577531] amdgpu_gem_object_close() at netbsd:amdgpu_gem_object_close+0x161
[   119.577531] drm_gem_object_release_handle() at netbsd:drm_gem_object_release_handle+0x37
[   119.577531] linux_idr_for_each() at netbsd:linux_idr_for_each+0x52
[   119.577531] drm_gem_release() at netbsd:drm_gem_release+0x20
[   119.577531] drm_file_free() at netbsd:drm_file_free+0x2b5
[   119.587532] drm_close() at netbsd:drm_close+0x5a
[   119.587532] closef() at netbsd:closef+0x9a
[   119.587532] fd_free() at netbsd:fd_free+0xfa
[   119.587532] exit1() at netbsd:exit1+0x119
[   119.587532] sigexit() at netbsd:sigexit+0x1c2
[   119.587532] postsig() at netbsd:postsig+0x35e
[   119.587532] lwp_userret() at netbsd:lwp_userret+0x198
[   119.587532] mi_userret() at ffffffff805b9b8e
[   119.587532] syscall() at netbsd:syscall+0xdc
[   119.587532] --- syscall (number 4) ---
[   119.587532] netbsd:syscall+0xdc:
[   119.587532] cpu2: End traceback...


>How-To-Repeat:
Start xorg with glamor and DRI "3" accelaration on amdgpu(4) card (see in Xorg.0.log that acceleration is enabled).
Start hardware accelerated application (firefox, glxgears, scummvm).
Observe freeze/artifacts.
Try to kill Xorg (over ssh) or restart system - observe crash (typically).
>Fix: