[PATCH 0/5] Use MI_MATH to apply RMW WA in LRC

[Michal Wajdeczko]

It was noticed that on VF we skip applying LRC WAs since we don't
process any RTP rules in xe_rtp_process_to_sr() and also because
during preparation of the WA job driver may read some registers
to perform RMW operation. By converting WA job to use MI_MATH we
can run this job also on VF. Inspired by i915 series [1].

Cc: Matt Roper <matthew.d.roper at intel.com>
Cc: Michal Winiarski <michal.winiarski at intel.com>

[1] https://patchwork.freedesktop.org/series/67276/

Michal Wajdeczko (5):
  drm/xe: Add MI_LOAD_REGISTER_REG command definition
  drm/xe: Add MI_MATH and ALU instruction definitions
  drm/xe: Avoid reading RMW registers in emit_wa_job
  drm/xe/vf: Stop applying save-restore MMIOs if VF
  drm/xe/vf: Unblock xe_rtp_process_to_sr for VFs

 .../gpu/drm/xe/instructions/xe_alu_commands.h | 72 ++++++++++++++++
 .../gpu/drm/xe/instructions/xe_mi_commands.h  |  5 ++
 drivers/gpu/drm/xe/regs/xe_engine_regs.h      |  4 +
 drivers/gpu/drm/xe/xe_gt.c                    | 84 ++++++++++++++-----
 drivers/gpu/drm/xe/xe_reg_sr.c                |  3 +
 drivers/gpu/drm/xe/xe_rtp.c                   |  3 -
 6 files changed, 147 insertions(+), 24 deletions(-)
 create mode 100644 drivers/gpu/drm/xe/instructions/xe_alu_commands.h

-- 
2.47.1