[Intel-gfx] [PATCH v3 00/14] Explicit MCR handling and MTL steering
Matt Roper
matthew.d.roper at intel.com
Fri Oct 14 23:02:25 UTC 2022
Steering of multicast/replicated registers becomes a bit more
complicated on Meteor Lake. Whereas previously the control register we
used to manage the steering was only used by our driver[*], software's
control of steering has now been consolidated with the controls for
various other hardware/firmware agents into a single register. We can
no longer utilize pre-programmed implicit steering since other firmware
agents may change the steering target and not restore it afterward;
we'll need to explicitly steer all types of MCR registers (including the
GSLICE/COMPUTE/DSS ranges that have been handled implicitly in the
past). Furthermore, since multiple agents will now be sharing a single
steering control register, races are possible. To address this, the
hardware adds a new MCR semaphore register which is supposed to be used
to temporarily lock the steering while performing MCR operations.
It's going to become important for us to handle accesses of multicast
registers very explicitly going forward. This series provides some prep
work for that by updating our register definitions to clearly define
registers as either MCR or non-MCR and ensure that we're using the
intel_gt_mcr_*() functions rather than intel_uncore_*() when operating
on MCR registers. By declaring MCR registers as a new C type (i.e., not
an i915_reg_t) we can have the compiler help us find any mistakes where
non-MCR functions are used on MCR registers and vice-versa.
This series also includes the general MCR steering tables and logic, but
does not yet introduce the support for the semaphore register that will
be used to coordinate steering updates with other agents; we'll provide
that support in a separate series once this preparation work has landed.
[*] This is a bit of an oversimplification; there are some hardware and
software debug tools that use the same MCR_SELECTOR register that i915
does and which could potentially re-steer MCR accesses behind our back.
E.g., simply using IGT's "intel_reg" tool to write the MCR_SELECTOR
register at the wrong time could interfere with driver operation. But
given that these debug facilities require root privileges to run and are
only used by people intentionally debugging the driver or hardware, we
can ignore such races for real-world usage.
v2:
- Add MCR handling of TLB invalidation registers (since they're in the
GAM range, their read behavior behaves a bit more like unicast
registers, but they should still be treated as MCR for writes).
- Add MCR rmw and 'wait_for_reg' operations; these are only typically
needed for GAM registers.
- Switched a few more instances of MISCCPCTL, HALF_SLICE_CHICKEN3, and
the fault registers over to MCR handling where appropriate. (Bala)
v3:
- Several checkpatch fixes
- Update intel_gt_mcr_multicast_rmw() to the register's original value,
aligning with the recent change to intel_uncore_rmw.
- Move transition to MCR functions in mmio_invalidate_full() to the
earlier patch that is dedicated to such changes. (Bala)
- Enforce stricter type checking on i915_mmio_reg_offset. (Jani)
- Drop several unnecessary GVT changes.
- Use IP naming (Xe_LPG and Xe_LPM+) more consistently.
- Update debug dumping of steering for MTL and beyond.
- Handle the MULTICAST bit in the steering control register more
explicitly on MTL and beyond where we no longer have implicit
steering. (Bala)
Matt Roper (14):
drm/i915/gen8: Create separate reg definitions for new MCR registers
drm/i915/xehp: Create separate reg definitions for new MCR registers
drm/i915/gt: Drop a few unused register definitions
drm/i915/gt: Correct prefix on a few registers
drm/i915/gt: Add intel_gt_mcr_multicast_rmw() operation
drm/i915/xehp: Check for faults on primary GAM
drm/i915/gt: Add intel_gt_mcr_wait_for_reg_fw()
drm/i915: Define MCR registers explicitly
drm/i915/gt: Always use MCR functions on multicast registers
drm/i915/guc: Handle save/restore of MCR registers explicitly
drm/i915/gt: Add MCR-specific workaround initializers
drm/i915: Define multicast registers as a new type
drm/i915/xelpg: Add multicast steering
drm/i915/xelpmp: Add multicast steering for media GT
drivers/gpu/drm/i915/gt/intel_engine_cs.c | 4 +-
drivers/gpu/drm/i915/gt/intel_ggtt.c | 4 +-
drivers/gpu/drm/i915/gt/intel_gt.c | 126 ++++-
drivers/gpu/drm/i915/gt/intel_gt_mcr.c | 279 ++++++++--
drivers/gpu/drm/i915/gt/intel_gt_mcr.h | 24 +-
drivers/gpu/drm/i915/gt/intel_gt_regs.h | 162 +++---
drivers/gpu/drm/i915/gt/intel_gt_types.h | 2 +
drivers/gpu/drm/i915/gt/intel_gtt.c | 22 +-
drivers/gpu/drm/i915/gt/intel_gtt.h | 2 +-
drivers/gpu/drm/i915/gt/intel_mocs.c | 12 +-
drivers/gpu/drm/i915/gt/intel_workarounds.c | 491 +++++++++++-------
.../gpu/drm/i915/gt/intel_workarounds_types.h | 9 +-
.../gpu/drm/i915/gt/selftest_workarounds.c | 2 +-
drivers/gpu/drm/i915/gt/uc/intel_guc_ads.c | 60 ++-
.../gpu/drm/i915/gt/uc/intel_guc_capture.c | 8 +-
drivers/gpu/drm/i915/gt/uc/intel_guc_fw.c | 12 +-
drivers/gpu/drm/i915/gvt/handlers.c | 4 +-
drivers/gpu/drm/i915/gvt/mmio_context.c | 14 +-
drivers/gpu/drm/i915/i915_gpu_error.c | 12 +-
drivers/gpu/drm/i915/i915_pci.c | 1 +
drivers/gpu/drm/i915/i915_reg_defs.h | 27 +-
drivers/gpu/drm/i915/intel_gvt_mmio_table.c | 2 +-
drivers/gpu/drm/i915/intel_pm.c | 18 +-
23 files changed, 902 insertions(+), 395 deletions(-)
--
2.37.3
More information about the Intel-gfx
mailing list