* Ingo Molnar <[EMAIL PROTECTED]> wrote: > thanks. It shows the anomaly in action:
> So all the time references we have show that (no surprise here) 1 > second passed between the two samples. But sched_clock() shows a > _large_ jump: > > .clock : 125652924079659272 > .clock : 125653018059166371 > > also reflected in .clock_max_delta: > > .clock_max_delta : 92976502936 > > that's a 93 seconds jump (!) in a single 1-second sample. [...] we also had a similar jump in exec_clock, which suggests that the sched_clock() "jump" occured while there was a task running. (i.e. not during ACPI-C2/C3) Could you try the patch below? It should catch such large jumps. The 'clock_overflows' counter in /proc/sched_debug will show whether this new protection triggered on your box. Ingo Index: linux/kernel/sched.c =================================================================== --- linux.orig/kernel/sched.c +++ linux/kernel/sched.c @@ -149,12 +149,12 @@ struct rq { u64 clock, prev_clock_raw; s64 clock_max_delta; - u64 fair_clock, prev_fair_clock; - u64 exec_clock, prev_exec_clock; + u64 fair_clock, prev_fair_clock; + u64 exec_clock, prev_exec_clock; s64 wait_runtime; unsigned long wait_runtime_overruns, wait_runtime_underruns; - unsigned int clock_warps; + unsigned int clock_warps, clock_overflows; unsigned int clock_unstable_events; struct sched_class *load_balance_class; @@ -245,9 +245,17 @@ static inline unsigned long long __rq_cl clock++; rq->clock_warps++; } else { - if (unlikely(delta > rq->clock_max_delta)) - rq->clock_max_delta = delta; - clock += delta; + /* + * Catch too large forward jumps too: + */ + if (delta > 2*TICK_NSEC) { + clock++; + rq->clock_overflows++; + } else { + if (unlikely(delta > rq->clock_max_delta)) + rq->clock_max_delta = delta; + clock += delta; + } } rq->prev_clock_raw = now; Index: linux/kernel/sched_debug.c =================================================================== --- linux.orig/kernel/sched_debug.c +++ linux/kernel/sched_debug.c @@ -117,13 +119,13 @@ static void print_cpu(struct seq_file *m P(clock); P(prev_clock_raw); P(clock_warps); + P(clock_overflows); P(clock_unstable_events); P(clock_max_delta); - rq->clock_max_delta = 0; P(fair_clock); - P(prev_fair_clock); + P(prev_fair_clock); P(exec_clock); - P(prev_exec_clock); + P(prev_exec_clock); P(wait_runtime); P(wait_runtime_overruns); P(wait_runtime_underruns); - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/