On 27 November 2012 18:56, Steven Rostedt [off-list ref] wrote:

A couple of things. The sched_select_cpu() is not cheap. It has a double
loop of domains/cpus looking for a non idle cpu. If we have 1024 CPUs,
and we are CPU 1023 and all other CPUs happen to be idle, we could be
searching 1023 CPUs before we come up with our own.

Not sure if you missed the first check sched_select_cpu()

+int sched_select_cpu(unsigned int sd_flags)
+{
+       /* If Current cpu isn't idle, don't migrate anything */
+       if (!idle_cpu(cpu))
+               return cpu;

We aren't going to search if we aren't idle.

Also, I really don't like this as a default behavior. It seems that this
solution is for a very special case, and this can become very intrusive
for the normal case.

We tried with an KCONFIG option for it, which Tejun rejected.

To be honest, I'm uncomfortable with this approach. It seems to be
fighting a symptom and not the disease. I'd rather find a way to keep
work from being queued on wrong CPU. If it is a timer, find a way to
move the timer. If it is something else, lets work to fix that. Doing
searches of possibly all CPUs (unlikely, but it is there), just seems
wrong to me.

As Vincent pointed out, on big LITTLE systems we just don't want to
serve works on big cores. That would be wasting too much of power.
Specially if we are going to wake up big cores.

It would be difficult to control the source driver (which queues work) to
little cores. We thought, if somebody wanted to queue work on current
cpu then they must use queue_work_on().

--
viresh