PERFORCE change 33663 for review

[Peter Wemm]

Eric Anholt wrote:
> On Thu, 2003-06-26 at 10:57, John Baldwin wrote:
> > On 25-Jun-2003 Peter Wemm wrote:
> > > http://perforce.freebsd.org/chv.cgi?CH=33663
> > > 
> > > Change 33663 by peter at peter_hammer on 2003/06/25 15:05:09
> > > 
> > >       Port sym to amd64
> > 
> > This is possibly not correct.  Do all hammer's support the P3+
> > SFENCE and related instructions?  Even i386 should probably be
> > using what bus_space_barrier() uses.  Heck, sym should probably
> > just be using bus_space_barrier anyways.
> 
> It would sure be nice to have an MI call for the bus_space_barrier()
> calls that don't need a bus_space_tag.  The DRM unfortunately doesn't
> (and won't ever, I think) do bus_space, so we have to have
> platform-specific ifdefs for read, write, and read/write barriers.

Yes, it has the *FENCE instructions, but it still has the same memory
ordering semantics.  *FENCE has most effect on the SSE/SSE2 instructions,
not so much the regular x86* instruction stream, unless MTRR/PAT are in
the middle of things as well.  This is usually not the case for things
like disk IO..

Cheers,
-Peter
--
Peter Wemm - peter at wemm.org; peter at FreeBSD.org; peter at yahoo-inc.com
"All of this is for nothing if we don't go to the stars" - JMS/B5