[Intel-gfx] [PATCH 0/7] Per client engine busyness

[Tvrtko Ursulin]

Hi,

On 15/05/2021 11:40, Maxime Schmitt wrote:
> Hi,
> 
> Nice to see something like this being worked on.
> 
> I wrote a top-like tool some time back (nvtop).
> I targeted NVIDIA, because it was the GPU I had at the time. Also,
> their driver provides a nice library to retrieve the information from
> (NVML).
> 
> Seeing this thread I think it would be nice to support more vendors now
> that the information is available.
> 
> I took a look at the DRM documentation, but I am only finding the in-
> kernel functions and not what is being exposed to user space. Maybe I
> am searching at the wrong place.
> Is there some documentation, from the user space point of view, on the
> way to discover the GPUs and the metrics that are exposed?
There isn't a common framework yet, its under discussion in this thread.

AMD has some stuff under /proc/<pid>/fdinfo/<fd>, and i915 at the moment 
has only global GPU stats (export as a perf/PMU device, see "perf list | 
grep i915.*/") used by current version of intel_gpu_top.

Regards,

Tvrtko

P.S. I suggest you use reply-to-all when replying on mailing list 
threads so it's easier to spot your message.