[PATCH v4 0/3] Improve gpu_scheduler trace events

Daniel Vetter daniel at ffwll.ch
Mon Jun 10 16:31:46 UTC 2024


On Mon, Jun 10, 2024 at 03:26:53PM +0200, Pierre-Eric Pelloux-Prayer wrote:
> v3: https://lists.freedesktop.org/archives/dri-devel/2024-June/456792.html
> 
> Changes since v3:
> * trace device name instead of drm_device primary index
> * no pointer deref in the TP_printk anymore. Instead the fence context/seqno
> are saved in TP_fast_assign

Some high-level comments:

- Quick summary of the what, why and how in the cover letter would be
  great.

- Link to the userspace, once you have that. At least last time we chatted
  that was still wip.

- Maybe most important to make this actually work, work well, and work
  long-term: I think we should clearly commit to these tracepoints being
  stable uapi, and document that by adding a stable tracepoint section in
  the drm uapi book.

  And then get acks from a pile of driver maintainers that they really
  think this is a good idea and has a future. Should also help with
  getting good review on the tracepoints themselves.

  Otherwise I fear we'll miss the mark again and still force userspace to
  hand-roll tracing for every driver, or maybe worse, even specific kernel
  versions.

Cheers, Sima

> 
> Pierre-Eric Pelloux-Prayer (3):
>   drm/sched: add device name to the drm_sched_process_job event
>   drm/sched: cleanup gpu_scheduler trace events
>   drm/sched: trace dependencies for gpu jobs
> 
>  .../gpu/drm/scheduler/gpu_scheduler_trace.h   | 97 +++++++++++++++----
>  drivers/gpu/drm/scheduler/sched_entity.c      |  8 +-
>  drivers/gpu/drm/scheduler/sched_main.c        |  2 +-
>  3 files changed, 84 insertions(+), 23 deletions(-)
> 
> -- 
> 2.40.1
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch


More information about the dri-devel mailing list