Re: [RFC PATCH 0/3] drm/xe: Enable ATS.

[Matthew Brost <matthew.brost@intel.com>]

On Mon, Jan 26, 2026 at 01:52:04PM +0100, Maarten Lankhorst wrote:
> RFC since I believe while the patches work, it might need some help to
> validate the semantics, and whether we want to set ATS unconditionally.
> 

I think it would be good to perf numbers of various benchmarks /
workloads to determine the impact of ATS, plus maybe Kconfig / configfs
knob or something to toggle this for quick experiments.

Patches themselve make sense to me.

Matt

> Add the relevant wiring to enable the ATS flag in PASID, and add a minimal
> selftest to validate ATS works and doesn't break on the first access to
> system memory.
> 
> Last patch is untested, hope to see what CI thinks. :-)
> 
> Maarten Lankhorst (3):
>   drm/xe: Force page fault
>   drm/xe: Set the bits in PASID to enable ATS.
>   drm/xe: Add migration selftest for validating job submission and basic
>     features work
> 
>  drivers/gpu/drm/xe/regs/xe_gtt_defs.h   |  1 +
>  drivers/gpu/drm/xe/regs/xe_lrc_layout.h |  1 +
>  drivers/gpu/drm/xe/xe_device.c          |  7 ++++
>  drivers/gpu/drm/xe/xe_lrc.c             |  6 ++++
>  drivers/gpu/drm/xe/xe_migrate.c         | 46 +++++++++++++++++++++++++
>  drivers/gpu/drm/xe/xe_migrate.h         |  1 +
>  drivers/gpu/drm/xe/xe_pt.c              | 36 ++++++++++---------
>  7 files changed, 82 insertions(+), 16 deletions(-)
> 
> -- 
> 2.51.0
>