[regression] [bisected] Radeon GPU crashes at SteamVR launch

System information

• OS: Arch Linux
• GPU: 0b:00.0 VGA compatible controller [0300]: Advanced Micro Devices, Inc. [AMD/ATI] Navi 21 [Radeon RX 6800/6800 XT / 6900 XT] [1002:73bf] (rev c1)
• Kernel version: 5.17.1
• Mesa version: 22.0.1
• Xserver version (if applicable): 1.21.1.3
• Desktop manager and compositor: i3
• SteamVR: 1.22.6

Describe the issue

Launching SteamVR consistently crashes the GPU: Valve Index screen turns on before the main display turns black for a second during the GPU reset. Xserver then becomes unstable and system must be rebooted.

The following trace is logged in dmesg:

[drm:dm_vblank_get_counter [amdgpu]] ERROR dc_stream_state is NULL for crtc '1'!
[drm:dm_crtc_get_scanoutpos [amdgpu]] ERROR dc_stream_state is NULL for crtc '1'!
[drm:dm_vblank_get_counter [amdgpu]] ERROR dc_stream_state is NULL for crtc '1'!
------------[ cut here ]------------
amdgpu 0000:0b:00.0: drm_WARN_ON_ONCE(drm_drv_uses_atomic_modeset(dev))
WARNING: CPU: 3 PID: 2263 at drivers/gpu/drm/drm_vblank.c:728 drm_crtc_vblank_helper_get_vblank_timestamp_internal+0x369/0x380
Modules linked in: nf_tables nfnetlink snd_seq_dummy snd_hrtimer snd_seq cfg80211 8021q garp mrp stp llc nct6775 hwmon_vid eeepc_wmi intel_ra>
crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel ccp sr_mod xhci_pci crypto_simd cryptd rng_core cdrom xhci_pci_re>
CPU: 3 PID: 2263 Comm: VulkanVblankThr Not tainted 5.17.1-arch1-1 #1 0ea933cb6bfe82a8dc16ab834a4bccdd297f98b7
Hardware name: ASUS System Product Name/ROG CROSSHAIR VIII DARK HERO, BIOS 3601 05/26/2021
RIP: 0010:drm_crtc_vblank_helper_get_vblank_timestamp_internal+0x369/0x380
Code: 4c 8b 6f 50 4d 85 ed 75 03 4c 8b 2f e8 f0 6b 01 00 48 c7 c1 40 f3 d1 ad 4c 89 ea 48 c7 c7 c2 5d d1 ad 48 89 c6 e8 a3 43 3d 00 <0f> 0b e>
RSP: 0018:ffff9beb86303b20 EFLAGS: 00010082
RAX: 0000000000000000 RBX: ffffffffc0b7e840 RCX: 0000000000000027
RDX: ffff8dca0eae1728 RSI: 0000000000000001 RDI: ffff8dca0eae1720
RBP: ffff9beb86303b90 R08: 0000000000000000 R09: ffff9beb86303950
R10: ffff9beb86303948 R11: ffff8dca2f2a9b28 R12: 0000000000000000
R13: ffff8dc3023dae30 R14: 0000000000000000 R15: ffff8dc3376b21d8
FS: 00007fa79444b640(0000) GS:ffff8dca0eac0000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007fa6ec001278 CR3: 00000001e6984000 CR4: 0000000000750ee0
PKRU: 55555554
Call Trace:
<TASK>
drm_get_last_vbltimestamp+0xb2/0xc0
drm_update_vblank_count+0x91/0x3d0
drm_vblank_enable+0x14b/0x180
drm_vblank_get+0x95/0xe0
drm_crtc_queue_sequence_ioctl+0xfd/0x2d0
? __check_object_size+0x46/0x140
? drm_crtc_get_sequence_ioctl+0x1a0/0x1a0
drm_ioctl_kernel+0xb8/0x140
drm_ioctl+0x22a/0x3d0
? drm_crtc_get_sequence_ioctl+0x1a0/0x1a0
amdgpu_drm_ioctl+0x49/0x80 [amdgpu 08a70cd20fdf14582ce9165e3698aeaecdd8c8f8]
__x64_sys_ioctl+0x82/0xb0
do_syscall_64+0x5c/0x80
? do_user_addr_fault+0x1d7/0x690
? do_syscall_64+0x69/0x80
? exc_page_fault+0x72/0x170
entry_SYSCALL_64_after_hwframe+0x44/0xae
RIP: 0033:0x7fa7ac2a7e6f
Code: 00 48 89 44 24 18 31 c0 48 8d 44 24 60 c7 04 24 10 00 00 00 48 89 44 24 08 48 8d 44 24 20 48 89 44 24 10 b8 10 00 00 00 0f 05 <41> 89 c>
RSP: 002b:00007fa79444ab00 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
RAX: ffffffffffffffda RBX: 00007fa79444ab90 RCX: 00007fa7ac2a7e6f
RDX: 00007fa79444ab90 RSI: 00000000c018643c RDI: 000000000000004a
RBP: 00000000c018643c R08: 0000000000000000 R09: 00007fa6ec000be0
R10: 000000000000006e R11: 0000000000000246 R12: 000055d760b895b8
R13: 000000000000004a R14: 000055d760c41d00 R15: 0000000000000000
</TASK>
---[ end trace 0000000000000000 ]---
[drm:dm_vblank_get_counter [amdgpu]] ERROR dc_stream_state is NULL for crtc '1'!
[drm:dm_crtc_get_scanoutpos [amdgpu]] ERROR dc_stream_state is NULL for crtc '1'!
[drm:dm_vblank_get_counter [amdgpu]] ERROR dc_stream_state is NULL for crtc '1'!

Regression

Downgrading Mesa to 21.3.7 allows SteamVR to start as expected.

Extra information

The bug has been reported on Linux's Bugzilla too.