Re: [PATCH] 2.4.18 scheduler bugs

Ingo Molnar (mingo@elte.hu)
Fri, 15 Mar 2002 21:42:33 +0100 (CET)


On Fri, 15 Mar 2002, Alan Cox wrote:

> > > moment work for them becomes available. I see no reason why an idle cpu
> > > should be forced to remain idle until the next tick, nor why fixing that
> > > should be considered `broken'.
> >
> > performance. IPIs are expensive.
>
> On a PIII I can see this being the case, especially as they dont power
> save on hlt nowdays.

it's an option, and the default is to use the hlt instruction. The main
reason is to let Linux save power - and those who need that final
performance edge (and it's measurable), can enable it. HTL still uses less
power than the tight idle loop.

> [...] But on the Athlon the IPI isnt going down a little side channel
> between cpus.

but even in the Athlon case an IPI is still an IRQ entry, which will add
at least 200 cycles or more to the idle wakeup latency.

Ingo

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/