Re: [PATCH v3 0/4] Add debugfs facility to catch RTP mistakes

[Daniel Charles <daniel.charles@intel.com>]

On 2/12/2026 4:09 PM, Matt Roper wrote:
> Unlike i915, Xe doesn't currently do any readback verification of
> workaround/tuning settings that it tries to program.  If an entry gets
> placed on the wrong table, or applied to the wrong platform/IP, the
> programming may be silently ineffective without anyone noticing.
>
> i915's solution of doing verification at application time was
> problematic because raising warnings during probe would be treated by CI
> as a fatal problem and would prevent CI from even attempting any other
> testing until the issue was dealt with.  Let's avoid that with Xe and
> instead perform the verification on-demand when a debugfs entry is read.
> This makes it very easy to do verification at various times during
> manual debugging.  It should also be simply to wrap a dedicated IGT test
> around the debugfs entry such that any unexpected failures are reported
> against that one test rather than bringing down the entire CI system.
>
> Note that verification of LRC programming is a bit more complicated than
> verification of GT/engine programming since we can't just blindly read
> the register values from the CPU and expect the RTP values to be in
> effect.  Applying LRC progamming ensures that every GPU context created
> should have the updated values in the LRC, and those values will get
> loaded into the register when the hardware context switches to it.
> However when no context is running on the hardware, the register values
> may revert to their hardware defaults which do not have the programming
> applied.  Instead, we should verify LRC programming by parsing the
> engine's 'default LRC' image and making sure the desired programming
> appears somewhere in one of the MI_LOAD_REGISTER_IMM state emission
> commands.
>
> Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
> Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
>
> v2:
>   - Avoid processing RTP -> SR for gt/hwe lists in VF to ensure the
>     debugfs entry is safe to use in an SR-IOV VF.  (Ashutosh)
>   - Add verification of LRC adjustments by searching engine type's
>     default LRC.
>
> v3:
>   - Add missing checks to avoid SR-IOV VF processing of RTP -> SR in
>     hw_engine_setup_default_state() and
>     xe_reg_whitelist_process_engine().
>
> Matt Roper (4):
>    drm/xe/reg_sr: Don't process gt/hwe lists in VF
>    drm/xe/reg_sr: Add debugfs to verify status of reg_sr programming
>    drm/xe: Add facility to lookup the value of a register in a default
>      LRC
>    drm/xe/reg_sr: Allow register_save_restore_check debugfs to verify LRC
>      values

I reviewed the series and I have no further comments.

Reviewed-by: Daniel Charles <daniel.charles@intel.com>


>
>   drivers/gpu/drm/xe/xe_gt_debugfs.c    | 26 ++++++++
>   drivers/gpu/drm/xe/xe_hw_engine.c     |  3 +
>   drivers/gpu/drm/xe/xe_lrc.c           | 96 +++++++++++++++++++++++++++
>   drivers/gpu/drm/xe/xe_lrc.h           |  4 ++
>   drivers/gpu/drm/xe/xe_reg_sr.c        | 71 +++++++++++++++++++-
>   drivers/gpu/drm/xe/xe_reg_sr.h        |  7 ++
>   drivers/gpu/drm/xe/xe_reg_whitelist.c |  3 +
>   drivers/gpu/drm/xe/xe_tuning.c        |  7 ++
>   drivers/gpu/drm/xe/xe_wa.c            |  6 ++
>   9 files changed, 221 insertions(+), 2 deletions(-)
>