cvs commit: src/sys/i386/i386 apic_vector.s local_apic.c mp_machdep.c src/sys/i386/include apicvar.h smp.h src/sys/i386/isa clock.c

[John Baldwin]

On Tuesday 15 February 2005 01:21 am, Alan Cox wrote:
> On Mon, Feb 14, 2005 at 04:30:56PM -0500, John Baldwin wrote:
> > On Monday 14 February 2005 01:42 pm, Coleman Kane wrote:
> > > I've been testing these since they were posted. All is well thus far!
> >
> > I need perf numbers though.  They should work fine, but I really don't
> > know if they will help or hurt performance.  Also, the perf
> > characteristics are likely different on UP vs SMP.
>
> In regards to the changes to atomic.h, I'm curious why you didn't make
> the output of lfence and sfence conditional on SMP, like the output of
> the lock prefix is?

I wasn't sure if it would matter.  Probably amd64 first needs to merge over 
the changes from i386 to not use fences for atomic_store_rel() for UP kernels 
as well first, and then use the fence instructions for just SMP kernels for 
now.

-- 
John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org