Re: ZONE_PADDING wastes 4 bytes of the new cacheline

[Jon Masters <jonmasters@gmail.com>]

On Thu, 21 Oct 2004 21:21:55 +0100, DaMouse <damouse@gmail.com> wrote:
> On Thu, 21 Oct 2004 13:54:41 -0500 (CDT), Adam Heath <doogie@debian.org> wrote:
> > On Thu, 21 Oct 2004, Andrea Arcangeli wrote:
> >
> > > On Thu, Oct 21, 2004 at 12:51:18PM +0200, Mikael Pettersson wrote:
> > > > Have you verified that? GCCs up to and including 2.95.3 and
> > > > early versions of 2.96 miscompiled the kernel when spinlocks
> > > > where empty structs on UP. I.e., you might not get a compile-time
> > > > error but runtime corruption instead.
> > >
> > > peraphs we should add a check on the compiler and force people to use
> > > gcc >= 3?
> > >
> > > Otherwise adding an #ifdef will fix 2.95, just like the spinlock does in
> > > UP.
> > >
> > > btw, the only machine where I still have gcc 2.95.3 is not uptodate
> > > enough to run 2.6 regardless of the fact 2.6 could compile on such
> > > machine or not.
> >
> > So compile a 2.6 kernel on the machine with 2.95.3 for another machine.
> >
> 
> I think what he was referring to was that most machines with 2.95.x
> have older kernels anyway.

That's probably mostly true even for embedded folks, but I don't think
it's a good idea to completely throw away 2.95 users just yet. Better
to use ifdefs or somesuch for now.

Jon.