[PATCH 0/2] target/ppc: Fix PMU instruction counting

[Nicholas Piggin]

The crux of the problem being that dynamic exits from a TB would
not count instructions previously executed in the TB. I don't
know how important it is for PMU to count instructions exactly,
however for instruction replay this can lead to different counts
for the same execution (e.g., because TBs can be different sized)
and that blows up reverse debugging.

I posted something on this out before, but missed a few things
(most notably faulting memory access). And found that forcing 1
insn per TB seems to be the only feasible way to do this.

Sorry to ping you on this again Richard, it's not urgent but
you're the guru with this stuff and I'm hesitant to change it
without a better opinion ... Simple band aid for the meanwhile
could be leave it as is but just disable counting if
record/replay is in use.

Thanks,
Nick

Nicholas Piggin (2):
  target/ppc: Fix PMC5 instruction counting
  target/ppc: Tidy pmu_count_insns implementation

 target/ppc/translate.c | 163 +++++++++++++++++++++--------------------
 1 file changed, 83 insertions(+), 80 deletions(-)

-- 
2.43.0