[Mesa-dev] [PATCH 0/2] RadeonSI: Use UMR for hang debugging

Edward O'Callaghan funfunctor at folklore1984.net
Sun Feb 5 03:39:40 UTC 2017


Thanks for getting this out to the wider community !

Patch 1 isn't the most ideal way to invocate the umr binary but I can't
think of an immediately better solution. So, Patches 1 & 2 are,
Reviewed-by: Edward O'Callaghan <funfunctor at folklore1984.net>

On 02/05/2017 09:44 AM, Marek Olšák wrote:
> Hi,
> 
> These are for our new UMR debugging tool. (see the amd-gfx list)
> 
> The idea is that you install umr, do "chmod +s" on it, and when you enable
> GALLIUM_DDEBUG in the unpipelined mode and encounter a GPU hang, the hang
> report will contain the shader asm annotated with where the GPU waves are
> stuck, and it will also print the EXEC mask and the current instruction
> in hexadecimal.
> 
> There are 4 kinds of GPU hangs we can differentiate between now:
> 1) A wave is stuck on a shader instruction and doesn't make progress.
>    A corrupted descriptor or a resource allocation deadlock are possible
>    causes. If a wave is stuck on s_waitcnt, the descriptor might be
>    corrupted.
> 2) Infinite loop. You can tell that from EXEC and also if the PC (program
>    counter) is changing if you run UMR multiple times manually.
> 3) Shader binary corruption. The instruction being executed doesn't match
>    the driver's copy of the shader binary.
> 4) Not a shader hang. The GPU is stuck, but there are no active waves.
> 
> Since these have been reviewed by Nicolai already, I'll push them some
> time next week if there are no review comments.
> 
> Marek
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: OpenPGP digital signature
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170205/13d7ed6a/attachment.sig>


More information about the mesa-dev mailing list