[PATCH 2/2] drm/xe/uapi: Restore uapi for access counting

[Dandamudi, Priyanka]

> -----Original Message-----
> From: Intel-xe <intel-xe-bounces at lists.freedesktop.org> On Behalf Of Lucas
> De Marchi
> Sent: Wednesday, October 2, 2024 11:13 PM
> To: Brost, Matthew <matthew.brost at intel.com>
> Cc: Singh, Apoorva <apoorva.singh at intel.com>; intel-
> xe at lists.freedesktop.org; Zeng, Oak <oak.zeng at intel.com>; Brian Welty
> <brian.welty at intel.com>
> Subject: Re: [PATCH 2/2] drm/xe/uapi: Restore uapi for access counting
> 
> On Wed, Oct 02, 2024 at 03:03:46PM GMT, Matthew Brost wrote:
> >On Tue, Oct 01, 2024 at 04:02:55PM -0500, Lucas De Marchi wrote:
> >> On Tue, Oct 01, 2024 at 07:24:32PM GMT, apoorva.singh at intel.com wrote:
> >> > From: Brian Welty <brian.welty at intel.com>
> >> >
> >> > In order to enable access counters, this must be enabled in LRC for
> >> > the exec_queue. Add basic uAPI to set configuration of access
> >> > counter trigger threshold and granularity.
> >> >
> >> > When access counters are enabled and threshold is hit, the access
> >> > counter handler in xe_gt_pagefault.c will migrate the buffer to
> >> > that to that GT's local VRAM.
> >> >
> >> > Signed-off-by: Brian Welty <brian.welty at intel.com>
> >> > Signed-off-by: Apoorva Singh <apoorva.singh at intel.com>
> >>
> >>
> >> What do you mean by **Restore**? I'd expect a minimum amount of
> >> documentation explaining what this is and how a real userspace can be
> >> used it for implementing X, Y and Z.  Link to userspace is also
> >> missing, and is required for uapi.
> >>
> >
> >I was going to say do we have user space for this yet? I also suspect
> >getting this implementation right is going to part of the SVM work. I
> >personally rather leave this disabled until we get the SVM
> >implementation for this all worked out so we can freely change this if
> >required.
> 
> Agreed. If we want to merge parts that are required and we are still not ready
> with the entire feature, then it'd be ok... provided it's properly explained and
> documented. But touching the UAPI is not. This should come only after we
> have everything in place and a userspace consumer is ready.
> 
> Lucas De Marchi
> 
> >
> >I can't imagine any user space is really demanding this to be enabled
> >yet.
> >
> >Matt
> >
> >> Lucas De Marchi

Earlier tried to bring access counter tests along with kernel here but dropped the idea as the existing code related to access counter was also removed as there are no customers then. I think its better to add into internal first.