libpciaccess x86 backend

[Samuel Thibault]

Dave Airlie, le Wed 20 Jan 2010 06:42:30 +1000, a écrit :
> On Mon, Jan 18, 2010 at 12:37 AM, Samuel Thibault
> <samuel.thibault at ens-lyon.org> wrote:
> > This adds support on x86 for OSes that do not have a PCI interface, tinkering
> > with I/O ports, and makes use of it on GNU/Hurd.
> >
> 
> Now this might be a dumb question but how to hurd drivers interact
> with the PCI layer, they all just bang on it directly?

There is no Hurd PCI driver (yet).

The only PCI code being used is a temporary glue into GNU Mach.

Yes, an interface between that and userspace should be designed, etc.
But since it's a temporary glue, it's not really worth doing it right
now. Still people would like to already try Xorg.

> It just seems lazy to add a direct to hw code again in userspace when
> clearly the kernel should be able for it,

Depends on the kernel.  A kernel could very well not deal with PCI at
all and let userspace agree on using a "PCI server", that could for
instance use libpciaccess.

> not that I really mind, I'm just hoping other OSes don't get this
> lazy, which marking this code as generic x86 might make them.

I understand your concern. I still believe it is more helpful to provide
this in the interim than to let yet another barrier from Xorg support
for exotic kernels.

Samuel