On Mon, May 11, 2020 at 11:22:00AM +0200, Peter Zijlstra wrote:
> (_completely_ untested)
>
> ---
> arch/arm64/kernel/perf_event.c | 27 ++++++++++++++++++---------
> include/linux/sched_clock.h | 28 ++++++++++++++++++++++++++++
> kernel/time/sched_clock.c | 41 +++++++++++++----------------------------
> 3 files changed, 59 insertions(+), 37 deletions(-)
>
> diff --git a/arch/arm64/kernel/perf_event.c b/arch/arm64/kernel/perf_event.c
> index 4d7879484cec..81a49a916660 100644
> --- a/arch/arm64/kernel/perf_event.c
> +++ b/arch/arm64/kernel/perf_event.c
> @@ -1165,28 +1165,37 @@ device_initcall(armv8_pmu_driver_init)
> void arch_perf_update_userpage(struct perf_event *event,
> struct perf_event_mmap_page *userpg, u64 now)
> {
> - u32 freq;
> - u32 shift;
> + struct clock_read_data *rd;
> + unsigned int seq;
>
> /*
> * Internal timekeeping for enabled/running/stopped times
> * is always computed with the sched_clock.
> */
> - freq = arch_timer_get_rate();
> userpg->cap_user_time = 1;
> + userpg->cap_user_time_zero = 1;
> +
> + do {
> + rd = sched_clock_read_begin(&seq);
> +
> + userpg->time_mult = rd->mult;
> + userpg->time_shift = rd->shift;
> + userpg->time_offset = rd->epoch_ns;
^^^^^^^ wants to be time_zero
> +
> + userpg->time_zero -= (rd->epoch_cyc * rd->shift) >> rd->shift;
> +
> + } while (sched_clock_read_retry(seq));
> +
> + userpg->time_offset = userpf->time_zero - now;
>
> - clocks_calc_mult_shift(&userpg->time_mult, &shift, freq,
> - NSEC_PER_SEC, 0);
And that ^^^ was complete crap.
> /*
> * time_shift is not expected to be greater than 31 due to
> * the original published conversion algorithm shifting a
> * 32-bit value (now specifies a 64-bit value) - refer
> * perf_event_mmap_page documentation in perf_event.h.
> */
> - if (shift == 32) {
> - shift = 31;
> + if (userpg->time_shift == 32) {
> + userpg->time_shift = 31;
> userpg->time_mult >>= 1;
> }
> - userpg->time_shift = (u16)shift;
> - userpg->time_offset = -now;
> }
