GPU hang when Plex attempts HW HDR tone mapping

[pkoci]

Hi,

 

I'm using UNRAID 6.11.5 and asrock J4105 motherboard (gemini lake) and I've been having a problem with the latest versions of PMS for a long time. Specifically if it attempts HW HDR tone mapping it fails every time and then attempts software transcoding. At this point the following appears in the unraid system logs: 

 

Jun 17 15:23:34 UNRAID kernel: i915 0000:00:02.0: [drm] Resetting rcs0 for preemption time out
Jun 17 15:23:34 UNRAID kernel: i915 0000:00:02.0: [drm] Plex Transcoder[23298] context reset due to GPU hang
Jun 17 15:23:34 UNRAID kernel: i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:e757fefe, in Plex Transcoder [23298]

 

HW transcoding works seamlessly if HDR tone mapping is turned off. Is it possible that this is caused by buggy Intel drivers that Plex needs for HW HDR tone mapping? Or is the bug on the UNRAID side? Plex staff confirmed that HW HDR tone mapping works on systems with the same or similar CPU (gemini lake).

 

Any ideas?

[squiddles88]

I have the exact same problem with a J3455 CPU, tested on 6.12 too. Only when trying to tone map. Works fine otherwise.

 

Jun 19 16:55:31 Apollo kernel: i915 0000:00:02.0: [drm] Plex Transcoder[8336] context reset due to GPU hang

Jun 19 16:55:31 Apollo kernel: i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:00280000, in Plex Transcoder [8336]

Edited June 19, 2023 by squiddles88

[squiddles88]

I've also tested Jellyfin and get the same GPU hangs when attempting to tone map:

 

Jun 19 17:28:00 Apollo kernel: i915 0000:00:02.0: [drm] Resetting rcs0 for CS error
Jun 19 17:28:00 Apollo kernel: i915 0000:00:02.0: [drm] ffmpeg[18618] context reset due to GPU hang
Jun 19 17:28:00 Apollo kernel: i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:8ed1fff2, in ffmpeg [18618]
Jun 19 17:28:00 Apollo kernel: ffmpeg[18618]: segfault at 178 ip 0000152962cb2ed8 sp 00007ffff5a79908 error 4 in libOpenCL.so.1.0.0[152962caf000+6000] likely on CPU 0 (core 0, socket 0)
Jun 19 17:28:00 Apollo kernel: Code: 66 90 ff a0 78 01 00 00 b8 dc ff ff ff c3 0f 1f 40 00 f3 0f 1e fa 48 8b 05 35 77 00 00 48 85 c0 75 10 48 85 ff 74 11 48 8b 07 <ff> a0 78 01 00 00 66 90 ff a0 80 01 00 00 b8 dc ff ff ff c3 0f 1f
Jun 19 17:28:04 Apollo kernel: i915 0000:00:02.0: [drm] Resetting rcs0 for CS error
Jun 19 17:28:04 Apollo kernel: i915 0000:00:02.0: [drm] ffmpeg[18868] context reset due to GPU hang
Jun 19 17:28:04 Apollo kernel: i915 0000:00:02.0: [drm] GPU HANG: ecode 9:1:8c17bc37, in ffmpeg [18868]
Jun 19 17:28:05 Apollo kernel: ffmpeg[18868]: segfault at 178 ip 00001465ac0b2ed8 sp 00007ffe40db6c78 error 4 in libOpenCL.so.1.0.0[1465ac0af000+6000] likely on CPU 1 (core 1, socket 0)

 

[pkoci]

there's one more thing. Every time the system is rebooted, this shows up in the system log:

Jun 19 19:15:40 UNRAID kernel: i915 0000:00:02.0: [drm] failed to retrieve link info, disabling eDP

 

Any ideas?

[pkoci]

Any idea? See the diagnostics

unraid-diagnostics-20230807-2045.zip

 

Edited August 7, 2023 by pkoci

[squiddles88]

Can someone please help with this? It's been going on for ages.