PCI Subsystem Rework for X.org 7.1 Proposal

[Jesse Barnes]

On Friday, February 17, 2006 6:37 pm, Ian Romanick wrote:
> Jesse Barnes wrote:
> >>ISSUE: Should special routines for reading / writing MMIO regions
> >> ala xf86WriteMmio8 be added?
> >
> > Not unless we find a platform that requires that, IMO.  Memory
> > mapped should mean memory mapped.
> >
> > Also, see above question about how to handle mapping with
> > particular attributes.
>
> I was going back through this thread, and I got a bit confused about
> this comment.  We platforms now that need these in some cases.  Take
> a look at various flavors of xf86(Read|Write)Mmio(8|16|24)(Be|Le) in
> xserver/xorg/hw/xfree86/common/compiler.h.  Or, are these not for
> what I think they're for?

I'd have to see examples.  It seems like, for the most part, they 
shouldn't be necessary in userspace, since the kernel will provide us 
with directly dereferenceable pointers to MMIO space.  If we end up 
having to write a bunch of support ourselves for a different back end 
though, then they might be necessary.  IOW, whether or not they should 
be implemented depends on whether or not we care about the reason why 
they exist anymore, which I'm not sure of.

Jesse