[openib-general] Re: [PATCH 03/13] [RFC] ipath copy routines

[Andrew Morton]

Robert Walsh <rjwalsh at pathscale.com> wrote:
>
> > > +	movl %edx,%ecx
> > > +	shrl $1,%ecx
> > > +	andl $1,%edx	
> > > +	cld
> > > +	rep 
> > > +	movsq 
> > > +	movl %edx,%ecx
> > > +	rep
> > > +	movsd
> > > +	ret
> > 
> > err, we have a portability problem.
> 
> Any chance we could get these moved into the x86_64 arch directory,
> then?

That would make sense.  Give it a non-ipath-related name and require that
all architectures which wish to run this driver must implement that
(documented) function.

And, in Kconfig, make sure that architectures which don't implement that
library function do not attempt to build this driver.  To avoid breaking
`make allmodconfig'.

>  We have to do double-word copies, or our chip gets unhappy.

In what form is this chip available?  As a standard PCI/PCIX card which
people will want to plug into power4/ia64/x86 machines?  Or is it in some
way exclusively tied to x86_64?