GPU hang & mmu fault on GC7000
Simon Zeni
simon.zeni at collabora.com
Thu May 2 01:15:35 UTC 2024
Hi,
I'm experiencing GPU hang with on a i.MX8 Quad Max (GC7000 rev 6009) with
linux-fslc [1] on the 6.1-2.2.x-imx branch. I've reproduced the issue on the
5.15-2.2.x-imx branch and I have not yet sorted out the 6.6.x+fslc with this
device.
Both issues are reproducible with `glmark-es2-drm` using the benchmarks
`terrain` and the `desktop:effect=shadow:windows=4`. The latter has to
be run
with `--run-forever` to produce the hang.
> [ 774.318157] etnaviv-gpu 53100000.gpu: GPU not yet idle, mask:
0x7fffbf76
> [ 774.830151] etnaviv-gpu 53100000.gpu: GPU not yet idle, mask:
0x7fffbf76
> [ 775.342025] etnaviv-gpu 53100000.gpu: recover hung GPU!
The MMU fault only occurs with the `terrain` benchmark`.
> [ 775.354701] etnaviv-gpu 53100000.gpu: MMU fault status 0x00000002
> [ 775.360796] etnaviv-gpu 53100000.gpu: MMU 0 fault addr 0x00000000
I have collected devcoredumps from both benchmarks, they both share the same
idle state, dma debug state and dma address:
> 00000004 = 7fffbf76 Idle: FE- DE+ PE+ SH- PA+ SE+ RA+ TX- VG+ IM+ FP+ TS+
> 00000660 = 00000001 Cmd: [dec DMA: idle Fetch: idle] Req idle Cal idle
> 00000664 = 00000905 Command DMA address
Both of them also have the line `Checking MMU entries... failed` before
listing
the buffer offsets, is that expected?
I have searched the IRC logs (thanks for keeping this channel logged),
and if I
get this correctly this address is an offset in the mmu.bin file
generated by
`viv-unpack` [2]. The result in both dumps seems inconclusive.
> 00000900 02 00 00 00 02 00 00 00 02 00 00 00 02 00 00 00
|................|
We figured out recently that the DMA address might be corrupted, dumping it
first [3] gives a different result. Am I going in the right direction to
understand, debug and fix this (those?) issues? What more can I do, and
where
should I look?
Thanks,
Simon
[1]: https://github.com/Freescale/linux-fslc
[2]: https://github.com/etnaviv/etna-gpu-tools/blob/master/dump/viv-unpack.c
[3]: https://lists.freedesktop.org/archives/etnaviv/2024-May/004744.html
More information about the etnaviv
mailing list