On Thu, Jun 18, 2020 at 04:49:23PM -0700, Palmer Dabbelt wrote: > AFAICT this changes the behavior of the perf event: it used to count any > fault, > whereas now it only counts those that succeed successfully. If everyone else > is doing it that way then I'm happy to change us over, but this definately > isn't just avoiding retries.
Right, I'm preparing v2 to keep the old behavior of PERF_COUNT_SW_PAGE_FAULTS, while with a nicer approach. Thanks for looking! -- Peter Xu