On Thu, Jul 02, 2015 at 03:07:01PM +0200, Peter Zijlstra wrote:
> @@ -606,22 +600,46 @@ static void cputime_adjust(struct task_c
>
> if (utime == 0) {
> stime = rtime;
> - } else if (stime == 0) {
> - utime = rtime;
> - } else {
> - cputime_t total = stime + utime;
> + goto update;
> + }
>
> - stime = scale_stime((__force u64)stime,
> - (__force u64)rtime, (__force u64)total);
> - utime = rtime - stime;
> + if (stime == 0) {
> + utime = rtime;
> + goto update;
> }
>
> - cputime_advance(&prev->stime, stime);
> - cputime_advance(&prev->utime, utime);
> + stime = scale_stime((__force u64)stime, (__force u64)rtime,
> + (__force u64)(stime + utime));
> +
> + /*
> + * Make sure stime doesn't go backwards; this preserves monotonicity
> + * for utime because rtime is monotonic.
> + *
> + * utime_i+1 = rtime_i+1 - stime_i
I'm not sure what is meant by _i+1.
I guess stime_i means prev->stime. stime_i+1 the new update of prev->stime
But then what is rtime_i and rtime_i+1 since we have no scaled rtime?
> + * = rtime_i+1 - (rtime_i - stime_i)
> + * = (rtime_i+1 - rtime_i) + stime_i
> + * >= stime_i
> + */
> + if (stime < prev->stime)
> + stime = prev->stime;
> + utime = rtime - stime;
> +
> + /*
> + * Make sure utime doesn't go backwards; this still preserves
> + * monotonicity for stime, analogous argument to above.
> + */
> + if (utime < prev->utime) {
> + utime = prev->utime;
> + stime = rtime - utime;
I see, so we are guaranteed that this final stime won't get below
prev->stime because older prev->stime + prev->utime <= newest rtime. I
guess that's more or less what's in the comments above :-)
> + }
>
> +update:
> + prev->stime = stime;
> + prev->utime = utime;
> out:
> *ut = prev->utime;
> *st = prev->stime;
> + raw_spin_unlock(&prev->lock);
> }
>
> void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t
> *st)
Ok I scratched my head a lot on this patch and the issues behind and it looks
good to me. I worried about introducing a spinlock but we had two cmpxchg before
that. The overhead is close.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
