This series adds support for Graviton5's customised CMN-S3 which has
zeroed discovery registers.

Robin, I understand moving driver state into the core perf header isn't
ideal, but I couldn't find another way to grow the struct. The v2
submission failed the kernel test robot build on i386 (COMPILE_TEST) due
to arm_cmn_hw_event exceeding the static_assert against the 'target'
field offset when CMN_MAX_DIMENSION is increased beyond 12.

Patch 1 moves struct arm_cmn_hw_event into the hw_perf_event union to
resolve this. I'd appreciate your feedback on this approach, or any
alternative you'd suggest.

Patch 2 adds the Graviton5 workarounds themselves (unchanged from v2
except for the DTC logid fix below).

Changes since v2:
 - Revert DTC logical ID assignment back to xp->logid (per Robin's
   review)
 - Add patch 1/2 to move arm_cmn_hw_event into hw_perf_event union
   to resolve 32-bit build failure

Aviv Bakal (2):
  perf/arm-cmn: Move struct arm_cmn_hw_event into struct hw_perf_event
  perf/arm-cmn: Add workarounds for CMN-S3 on Graviton5

 drivers/perf/arm-cmn.c     | 55 +++++++++++++++++++++-----------------
 include/linux/perf_event.h | 22 +++++++++++++++
 2 files changed, 52 insertions(+), 25 deletions(-)

-- 
2.47.3