[SOLVED] mpv-0.37.0 crashes radeon driver (gpu reset)

Message

sdauth · Post by **sdauth** » Fri Dec 15, 2023 7:06 pm

Hi,
I updated mpv from 0.36 to 0.37 and it crashes (and triggers gpu reset) consistently when going fullscreen (tried with different files) :

Code: Select all

[  310.629977] radeon 0000:00:01.0: ring 0 stalled for more than 10470msec
[  310.629987] radeon 0000:00:01.0: GPU lockup (current fence id 0x00000000000003e6 last fence id 0x00000000000003f5 on ring 0)
[  310.672137] radeon 0000:00:01.0: Saved 458 dwords of commands on ring 0.
[  310.672208] radeon 0000:00:01.0: GPU softreset: 0x00000009
[  310.672212] radeon 0000:00:01.0:   GRBM_STATUS               = 0xA04E3828
[  310.672216] radeon 0000:00:01.0:   GRBM_STATUS_SE0           = 0x08800007
[  310.672219] radeon 0000:00:01.0:   GRBM_STATUS_SE1           = 0x00000007
[  310.672223] radeon 0000:00:01.0:   SRBM_STATUS               = 0x20000040
[  310.672280] radeon 0000:00:01.0:   SRBM_STATUS2              = 0x00000000
[  310.672283] radeon 0000:00:01.0:   R_008674_CP_STALLED_STAT1 = 0x00000000
[  310.672286] radeon 0000:00:01.0:   R_008678_CP_STALLED_STAT2 = 0x00018000
[  310.672289] radeon 0000:00:01.0:   R_00867C_CP_BUSY_STAT     = 0x00010002
[  310.672292] radeon 0000:00:01.0:   R_008680_CP_STAT          = 0x80030243
[  310.672295] radeon 0000:00:01.0:   R_00D034_DMA_STATUS_REG   = 0x44C83D57
[  310.672298] radeon 0000:00:01.0:   R_00D834_DMA_STATUS_REG   = 0x44C83D57
[  310.672301] radeon 0000:00:01.0:   VM_CONTEXT0_PROTECTION_FAULT_ADDR   0x00000000
[  310.672304] radeon 0000:00:01.0:   VM_CONTEXT0_PROTECTION_FAULT_STATUS 0x00000000
[  310.672307] radeon 0000:00:01.0:   VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x00000000
[  310.672310] radeon 0000:00:01.0:   VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x00000000
[  310.682831] radeon 0000:00:01.0: GRBM_SOFT_RESET=0x0000DF7B
[  310.682904] radeon 0000:00:01.0: SRBM_SOFT_RESET=0x00010100
[  310.684062] radeon 0000:00:01.0:   GRBM_STATUS               = 0x00003828
[  310.684068] radeon 0000:00:01.0:   GRBM_STATUS_SE0           = 0x00000007
[  310.684073] radeon 0000:00:01.0:   GRBM_STATUS_SE1           = 0x00000007
[  310.684077] radeon 0000:00:01.0:   SRBM_STATUS               = 0x20000040
[  310.684135] radeon 0000:00:01.0:   SRBM_STATUS2              = 0x00000000
[  310.684138] radeon 0000:00:01.0:   R_008674_CP_STALLED_STAT1 = 0x00000000
[  310.684142] radeon 0000:00:01.0:   R_008678_CP_STALLED_STAT2 = 0x00000000
[  310.684150] radeon 0000:00:01.0:   R_00867C_CP_BUSY_STAT     = 0x00000000
[  310.684153] radeon 0000:00:01.0:   R_008680_CP_STAT          = 0x00000000
[  310.684156] radeon 0000:00:01.0:   R_00D034_DMA_STATUS_REG   = 0x44C83D57
[  310.684160] radeon 0000:00:01.0:   R_00D834_DMA_STATUS_REG   = 0x44C83D57
[  310.684223] radeon 0000:00:01.0: GPU reset succeeded, trying to resume
[  310.707992] [drm] PCIE GART of 1024M enabled (table at 0x00000000001D6000).
[  310.708113] radeon 0000:00:01.0: WB enabled
[  310.708119] radeon 0000:00:01.0: fence driver on ring 0 use gpu addr 0x0000000020000c00
[  310.708499] radeon 0000:00:01.0: fence driver on ring 5 use gpu addr 0x0000000000075a18
[  310.718688] radeon 0000:00:01.0: failed VCE resume (-22).
[  310.718693] radeon 0000:00:01.0: fence driver on ring 1 use gpu addr 0x0000000020000c04
[  310.718697] radeon 0000:00:01.0: fence driver on ring 2 use gpu addr 0x0000000020000c08
[  310.718700] radeon 0000:00:01.0: fence driver on ring 3 use gpu addr 0x0000000020000c0c
[  310.718704] radeon 0000:00:01.0: fence driver on ring 4 use gpu addr 0x0000000020000c10
[  310.737009] [drm] ring test on 0 succeeded in 2 usecs
[  310.737019] [drm] ring test on 3 succeeded in 3 usecs
[  310.737026] [drm] ring test on 4 succeeded in 3 usecs
[  310.783071] [drm] ring test on 5 succeeded in 2 usecs
[  310.803097] [drm] UVD initialized successfully.
[  311.901908] [drm:r600_ib_test] *ERROR* radeon: fence wait timed out.
[  311.901927] [drm:radeon_ib_ring_tests] *ERROR* radeon: failed testing IB on GFX ring (-110).
[  311.944524] radeon 0000:00:01.0: GPU softreset: 0x00000009
[  311.944528] radeon 0000:00:01.0:   GRBM_STATUS               = 0xA04E3828
[  311.944530] radeon 0000:00:01.0:   GRBM_STATUS_SE0           = 0x08800007
[  311.944532] radeon 0000:00:01.0:   GRBM_STATUS_SE1           = 0x00000007
[  311.944534] radeon 0000:00:01.0:   SRBM_STATUS               = 0x20000040
[  311.944589] radeon 0000:00:01.0:   SRBM_STATUS2              = 0x00000000
[  311.944591] radeon 0000:00:01.0:   R_008674_CP_STALLED_STAT1 = 0x00000000
[  311.944593] radeon 0000:00:01.0:   R_008678_CP_STALLED_STAT2 = 0x00018000
[  311.944595] radeon 0000:00:01.0:   R_00867C_CP_BUSY_STAT     = 0x00000002
[  311.944596] radeon 0000:00:01.0:   R_008680_CP_STAT          = 0x80030243
[  311.944598] radeon 0000:00:01.0:   R_00D034_DMA_STATUS_REG   = 0x44C83D57
[  311.944600] radeon 0000:00:01.0:   R_00D834_DMA_STATUS_REG   = 0x44C83D57
[  311.944602] radeon 0000:00:01.0:   VM_CONTEXT0_PROTECTION_FAULT_ADDR   0x00000000
[  311.944604] radeon 0000:00:01.0:   VM_CONTEXT0_PROTECTION_FAULT_STATUS 0x00000000
[  311.944605] radeon 0000:00:01.0:   VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x00000000
[  311.944607] radeon 0000:00:01.0:   VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x00000000
[  311.947515] radeon 0000:00:01.0: GRBM_SOFT_RESET=0x0000DF7B
[  311.947568] radeon 0000:00:01.0: SRBM_SOFT_RESET=0x00010100
[  311.948723] radeon 0000:00:01.0:   GRBM_STATUS               = 0x00003828
[  311.948725] radeon 0000:00:01.0:   GRBM_STATUS_SE0           = 0x00000007
[  311.948727] radeon 0000:00:01.0:   GRBM_STATUS_SE1           = 0x00000007
[  311.948729] radeon 0000:00:01.0:   SRBM_STATUS               = 0x20000040
[  311.948785] radeon 0000:00:01.0:   SRBM_STATUS2              = 0x00000000
[  311.948786] radeon 0000:00:01.0:   R_008674_CP_STALLED_STAT1 = 0x00000000
[  311.948788] radeon 0000:00:01.0:   R_008678_CP_STALLED_STAT2 = 0x00000000
[  311.948790] radeon 0000:00:01.0:   R_00867C_CP_BUSY_STAT     = 0x00000000
[  311.948791] radeon 0000:00:01.0:   R_008680_CP_STAT          = 0x00000000
[  311.948793] radeon 0000:00:01.0:   R_00D034_DMA_STATUS_REG   = 0x44C83D57
[  311.948795] radeon 0000:00:01.0:   R_00D834_DMA_STATUS_REG   = 0x44C83D57
[  311.948860] radeon 0000:00:01.0: GPU reset succeeded, trying to resume

If I downgrade to 0.36, no problem, no more crash.
What can I do to find the cause of this ?

mpv config :

Code: Select all

vo=gpu-next
profile=gpu-hq

kernel : 6.1.67

edit : On another machine with different gpu (intel gpu), 0.37.0 works perfectly (windowed & fullscreen), same kernel, same mpv config except profile=gpu instead of profile=gpu-hq.
Thanks.

sdauth · Post by **sdauth** » Fri Dec 15, 2023 8:21 pm

After trial and error test with mpv config, it turns out some things have changed related to profile=gpu-hq settings with mpv 0.37.0 (https://github.com/mpv-player/mpv/issues/12564)
The new gpu-hq profile settings are much more ressource intensive apparently.
Nonetheless, it doesn't seem right that it triggers a gpu reset... If you know / have any more information, please share.

Anyway, commenting out "profile=gpu-hq" in mpv.config solves the issue. (it uses default settings instead)

Anon-E-moose · Post by **Anon-E-moose** » Fri Dec 15, 2023 8:38 pm

any difference between versions "mpv --show-profile=gpu-hq"

sdauth · Post by **sdauth** » Fri Dec 15, 2023 9:40 pm

Anon-E-moose wrote:any difference between versions "mpv --show-profile=gpu-hq"

0.37.0 :

Code: Select all

Profile gpu-hq: 
 profile=high-quality
  scale=ewa_lanczossharp
  hdr-peak-percentile=99.995
  hdr-contrast-recovery=0.30
  deband=yes

0.36.0 :

Code: Select all

Profile gpu-hq: 
 scale=spline36
 cscale=spline36
 dscale=mitchell
 dither-depth=auto
 hdr-contrast-recovery=0.30
 correct-downscaling=yes
 linear-downscaling=yes
 sigmoid-upscaling=yes
 deband=yes

iirc, scale=ewa_lanczossharp is quite heavy and my gpu is just too old to handle it so this is what probably was causing the issue, although once again, it shouldn't cause a gpu reset. I'll open a bug later on github.

Anon-E-moose · Post by **Anon-E-moose** » Fri Dec 15, 2023 10:48 pm

You should be able to create a profile based on the older one, call it gpu-hq-1 or whatever

Post by **Ionen** » Sat Dec 16, 2023 9:45 am

Yes, high-quality (formerly gpu-hq, which is still recognized as an alias) became much more demanding in 0.37. Albeit it should in theory only result in dropped frames, unless something else going on with your card (like not getting enough power).

The new default profile is actually higher quality than 0.36's defaults (similar to old gpu-hq, so no harm in using it), while profile=fast is more like the old defaults.

See https://github.com/mpv-player/mpv/commit/703f158880 that's linked in the issue.

May also want to try vo=gpu-next (this will become the default sooner or later, possibly mpv 0.38 albeit change hasn't been done in master yet) and also gpu-api=vulkan (preferred with esp. with gpu-next). Maybe it'll work out better with your card.

sdauth · Post by **sdauth** » Sat Dec 16, 2023 10:35 pm

Anon-E-moose wrote:You should be able to create a profile based on the older one, call it gpu-hq-1 or whatever

Thanks. This works perfectly (and doesn't cause any crash when going fullscreen) so I'll just use those "old" settings.

Code: Select all

~/.config/mpv/mpv.conf

Code: Select all

vo=gpu-next
profile=gpu-hq-2

[gpu-hq-2]
scale=spline36
cscale=spline36
dscale=mitchell
dither-depth=auto
hdr-contrast-recovery=0.30
correct-downscaling=yes
linear-downscaling=yes
sigmoid-upscaling=yes
deband=yes

sdauth · Post by **sdauth** » Sat Dec 16, 2023 11:10 pm

@Ionen :
I was wondering if the issue was related to mesa so I compiled mesa-23.3.1 (not in stable yet) then restarted X and tried again with profile=high-quality
This time, it works perfectly. No more crashes when going fullscreen although I still notice a very fast glitch on the window when going fullscreen (this was also experienced with previous profile) :

Code: Select all

[...]
[   2.354][d][cplayer] Run command: cycle, flags=73, args=[name="fullscreen", value="1.000000"]
[   2.355][v][cplayer] Set property: fullscreen -> 1
[...]
[   2.490][v][vo/gpu-next/libplacebo] Spent 16.544 ms compiling shader
[...]

Ionen wrote:Yes, high-quality (formerly gpu-hq, which is still recognized as an alias) became much more demanding in 0.37. Albeit it should in theory only result in dropped frames.

Indeed, while some small files play just fine with high-quality profile, I observe almost 100% gpu utlilization (with radeontop) on every big files (movies) and consequently, framedrop. So although it doesn't cause any crash anymore, I will continue to use previous settings. I can't add a dedicated gpu (I'm using an APU currently) sadly as I don't have any available pcie