> Hi, I noticed that Ingo released a low-latency patch for the 2.2.13
> kernel.
>
> 2.2.10-lowlatency was not optimized for SMP low-latency ( latencytest on
> /proc stress gives high latencies).
>
> Does anyone (Ingo in particular) know if the 2.2.13 is SMP optimized ?
> (spinlock changes)
no, and i dont think it will ever be. The only and worst offender is the
'big kernel lock', which is much less of a problem in 2.3. We cannot
generally reschedule at points that do a lock_kernel(). Maybe we could add
a new lock_kernel_reschedule() function, but possibly you'll have to add
this to gazillion places.
there is a workaround: in reschedule_idle() whenever we wake up a RT
process we could mark all running processes as 'reschedule ASAP' (set the
need_resched flag). This guarantees that _someone_ will notice and
reschedule eventually. (an SMP kernel is never worse than a UP kernel,
latency-wise) This is only for RT tasks though. Do something like this at
the beginning of reschedule_idle():
if (policy != SCHED_OTHER)
for (i = 0; i < smp_num_cpus; i++) {
cpu = cpu_logical_map(i);
tsk = cpu_curr(cpu);
tsk->need_resched = 1;
}
/* continue doing the normal idle reschedule part */
(this is of course an ugly hack, but if it makes a difference we can see
how this could be done cleanly. It only affects RT tasks.)
-- mingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/