On Fri, Sep 07, 2018 at 12:16:34PM +0200, Peter Zijlstra wrote:
> This does the whole seqcount thing 6x, which is a bit of a waste.
>
> struct snapshot {
> u32 times[NR_PSI_STATES];
> };
>
> static inline struct snapshot get_times_snapshot(struct psi_group *pg, int cpu)
> {
> struct pci_group_cpu *pgc = per_cpu_ptr(pg->pcpu, cpu);
> struct snapshot s;
> unsigned int seq;
> u32 delta;
> int i;
>
> do {
> seq = read_seqcount_begin(&pgc->seq);
>
> delta = cpu_clock(cpu) - pgc->state_start;
> for (i = 0; i < NR_PSI_STATES; i++) {
> s.times[i] = gpc->times[i];
> if (test_state(pgc->tasks, i))
> s.times[i] += delta;
> }
>
> } while (read_seqcount_retry(&pgc->seq, seq);
Sorry, I forgot the whole times_prev thing:
for (i = 0; i < NR_PSI_STATES; i++) {
tmp = s.times[i];
s.times[i] -= pgc->times_prev[i];
pgc->times_prev[i] = tmp;
}
> return s;
> }
>
>
> for_each_possible_cpu(cpu) {
> struct snapshot s = get_times_snapshot(pg, cpu);
>
> nonidle = nsecs_to_jiffies(s.times[PSI_NONIDLE]);
> nonidle_total += nonidle;
>
> for (i = 0; i < PSI_NONIDLE; i++)
> deltas[s] += (u64)s.times[i] * nonidle;
>
> /* ... */
>
> }
>
>
> It's a bit cumbersome, but that's because of C.
