[RFC 0/3] perf/x86: Rework RDPMC access handling

This series is preparation for supporting perf user counter access on arm64.

Originally, the arm64 implementation was just a copy of the x86 version, but
Will did not like the mm_context state tracking nor the IPIs (mm_cpumask
doesn't work for arm64). The hook into switch_mm and the IPIs feel like
working around limitations in the perf core (aka the platform problem).
So this series aims to solve that such that all the state for RDPMC is
tracked and controlled via the perf core.

Alternatively, I could avoid all the x86 changes here with a new PMU
callback (.set_user_access()?) and plumb that into the perf core context
and mmap code. However, it's better in the long run if there's a common
implementation.

So far, I've only tested the perf core changes with the arm64 version of
the code which is similar to the x86 version here. I'm hoping for some
quick feedback on the direction here and whether I've missed some usecase
that isn't handled.

Rob


Rob Herring (3):
  x86: perf: Move RDPMC event flag to a common definition
  perf/x86: Control RDPMC access from .enable() hook
  perf/x86: Call mmap event callbacks on event's CPU

 arch/x86/events/core.c             | 113 +++++++++++++++--------------
 arch/x86/events/perf_event.h       |   2 +-
 arch/x86/include/asm/mmu.h         |   1 -
 arch/x86/include/asm/mmu_context.h |   6 --
 arch/x86/include/asm/perf_event.h  |   1 -
 arch/x86/mm/tlb.c                  |  29 +-------
 include/linux/perf_event.h         |   9 ++-
 kernel/events/core.c               |  56 +++++++++++---
 8 files changed, 113 insertions(+), 104 deletions(-)

--
2.27.0