Re: Xorg, libpciaccess and /dev/mem

[Samuel Thibault]

olafBuddenhagen@gmx.net, le Fri 12 Mar 2010 00:32:44 +0100, a écrit :
> On Thu, Mar 11, 2010 at 05:35:13PM +0100, Samuel Thibault wrote:
> > Using the iopl device is a bit of a hack, a more proper solution needs
> > to be done.
> 
> I don't consider it a hack. What would be a more "proper" solution in
> your opinion?...

You need to dig back the discussions on IRC or the list (can't
remember), but things like proper memory attribution: avoid letting two
processes access the same area, possible delegation, etc. In a word:
translators :)

> Of course, one day hopefully we will have a lovely PCI bus server, which
> allows clients to map individual regions according to the device
> configuration...

Which would be such kind of delegation and translators.

> But that will still need a method for obtaining a
> memory object for the physical address space from the kernel.

Right.

> > Maybe in the meanwhile we should alias mem and iopl in the gnumach
> > kernel, so already installed hurds can run Xorg (and the name "iopl"
> > doesn't sound so good to me to replace mem :)).
> 
> Yeah, I actually mentioned on IRC that probably the best fix is renaming
> the kernel device...

And break existing code that uses it?

Samuel