Subject: Re: looking for devices on PCI bus
To: Matthias Drochner <>
From: Manuel Bouyer <>
List: tech-kern
Date: 05/02/2001 18:04:50
On Wed, May 02, 2001 at 05:53:32PM +0200, Matthias Drochner wrote:
> said:
> > Well, this is another problem entirely, as the X servers are in
> > userland. We would need a sysctl tree for this, but I'm not sure how
> > to make this fit in the sysctl tree. 
> While this is not relevant for the original question, I'd throw
> in my opinion on the right way to do this:
> The X server should get all the information it needs from the wsdisplay
> device. This could be the pci config register contents, or it could
> even provide the mmap() interface for all memory space ranges.
> What the i386 X servers do currently - scanning the PCI bus from
> userland - is dangerous because there are stateful transactions
> which could be interrupted by a context switch.
> Unfortunately, the structure of the xfree86 code makes this quite hard.

Yes, this would be an elegant way of doing things. Maybe this could even
avoid access to /dev/mem by the X server ?

But I'd still like to have the device tree (not only PCI) exposed to userland
in some easily parsable form (dmesg isn't good because we may have several
boot here, or only one partial). I can see several uses of it for some local
scripts I have here :)

Manuel Bouyer, LIP6, Universite Paris VI.