Re: [PATCH v3 0/7] Fix MCR inconsistencies in RTP tables

[Gustavo Sousa <gustavo.sousa@intel.com>]

Gustavo Sousa <gustavo.sousa@intel.com> writes:

> The Xe driver uses struct xe_reg as base type to represent a register
> definition, and defines struct xe_reg_mcr as a sort of subtype to
> represent multicast replicated (MCR) registers.  It uses this subtyping
> to "force" the use of the correct MCR-based API for MMIO operations on
> MCR registers.
>
> In regular driver code, usage of those MCR registers usually need some
> reasoning about whether the access will need to be steered or a
> multicast; and the correct MCR API will be selected.
>
> For RTP tables, on the other hand, that's not the case.  The developer
> when defining an action the developer can either select a regular or a
> MCR register and the register save/restore logic will make the selection
> between regular vs MCR MMIO access depending on the register instance
> defined in the RTP action.
>
> That allows some mistakes to go unnoticed: a non-MCR register instance
> could be used for an RTP action against a platform where such a register
> is actually MCR; and vice-versa.
>
> This series fixes those mistakes.
>
> Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com>

Pushed to drm-xe-next.

Thanks Matt and Michal for the reviews!

--
Gustavo Sousa

> ---
> Changes in v3:
> - Dropped patch "drm/xe/kunit: Add xe_kunit_helper_is_live_test()",
>   which became unnecessary in v2.
> - This version is being sent mainly to get fresh CI results before
>   applying, just in case any new inconsistencies were introduced in the
>   meantime.
> - Link to v2: https://patch.msgid.link/20260508-rtp-mcr-check-v2-0-9897b147a5d2@intel.com
>
> Changes in v2:
> - Incorporated review feedback. Please see individual patches for the
>   respective changelogs.
> - Dropped inclusion of hw_engines in xe_wa test.  We will implement a more
>   focused test to check for MCR inconsistencies in the future, which will
>   iterate over all applicable RTP tables.
> - As a result of the above, dropped MMIO interception patches, as they became
>   unnecessary.
> - Link to v1: https://patch.msgid.link/20260116-rtp-mcr-check-v1-0-d420b9c1a327@intel.com
>
> ---
> Gustavo Sousa (7):
>       drm/xe: Define CACHE_MODE_1 as MCR register
>       drm/xe: Define and use MCR version of COMMON_SLICE_CHICKEN1
>       drm/xe: Define and use MCR version of COMMON_SLICE_CHICKEN4
>       drm/xe: Extract xe_hw_engine_setup_reg_lrc()
>       drm/xe/kunit: Use KUNIT_EXPECT_EQ() in xe_wa_gt()
>       drm/xe/mcr: Extract reg_in_steering_type_ranges()
>       drm/xe/reg_sr: Do sanity check for MCR vs non-MCR
>
>  drivers/gpu/drm/xe/regs/xe_gt_regs.h   |  4 +-
>  drivers/gpu/drm/xe/tests/xe_rtp_test.c | 71 ++++++++++++++++++++++++++++++----
>  drivers/gpu/drm/xe/tests/xe_wa_test.c  | 14 ++++++-
>  drivers/gpu/drm/xe/xe_gt.c             |  8 ++--
>  drivers/gpu/drm/xe/xe_gt_mcr.c         | 67 +++++++++++++++++++++++---------
>  drivers/gpu/drm/xe/xe_gt_mcr.h         |  1 +
>  drivers/gpu/drm/xe/xe_hw_engine.c      | 15 ++++++-
>  drivers/gpu/drm/xe/xe_hw_engine.h      |  2 +-
>  drivers/gpu/drm/xe/xe_reg_sr.c         | 36 +++++++++++++++++
>  drivers/gpu/drm/xe/xe_tuning.c         |  2 +-
>  drivers/gpu/drm/xe/xe_wa.c             |  6 +--
>  11 files changed, 186 insertions(+), 40 deletions(-)
> ---
> base-commit: 2edbd77a2045d12d30c95dbb5842b831e3da0035
> change-id: 20260112-rtp-mcr-check-f976cb1adf94
>
> Best regards,
> --  
> Gustavo Sousa <gustavo.sousa@intel.com>