Re: [PATCH] use per cpu data for single cpu ipi calls

From: Ingo Molnar
Date: Thu Jan 29 2009 - 08:43:10 EST

* Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:

> On Thu, 2009-01-29 at 12:13 +0100, Ingo Molnar wrote:
> > * Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> >
> > > > Can generic_smp_call_function_single_interrupt() ever see
> > > > CSD_FLAG_ALLOC set now? If not, that kfree can go away.
> > >
> > > Like said above, removing that kmalloc will hurt people.
> >
> > the whole promise of generic-IPI seems to dwindle and we'll get back to
> > roughly where we started out from.
> I'd not go that far, the per-cpu csd is a nice fallback, but removing
> that kmalloc is just silly.
> And I see no reason why arch specific could do any better (aside from
> sparc which has fancy ipis).

well if the hw can embedd at least 64 bit of dataload in the IPI itself,
we could get rid of a lot of complications. But since we cannot (on x86
and most other architectures), we've got to demultiplex from generic data
which brings up serialization issues and adds overhead .

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at