On Wed, 26 Jul 2006 15:43:39 +0200
"Nicolas Boulay" <[EMAIL PROTECTED]> wrote:
> > Not true. You map only the memory regions of the card into
> > your adress space. Not the whole adress space of the kernel.
> > Of course, this way you can always trash your card which in
> > turn can trash your system, but no generic driver can protect
> > against this anyways.
>
> The "generic" drivers could be tailord for the OGD PCI bridge. When
> you say that you don't need register, i beleived that you want to open
> /dev/mem and write inside it.
Oh, that explains the missunderstanding. No, i might be sick,
but i'm not insane ;)
My idea was rather to ask the kernel to map adresspace
of pci card xy into the adresspace of the current process.
(think of an mmap of a file)
> > Register are nothing more than memory locations in pci terms.
> > Please note that registers on a card are from the host point of
> > view something very different than register in a CPU.
> >
>
> :) I'm in hardware engenering since 5 years, thanks...
Oh.. you should have said that earlier ;)
> register in memory are often sensible to latency rather than bandwith.
> And out of order writing are often a very bad idea. So there is a
> difference, even if it's always memory adress.
Good point.
Attila Kinali
--
心をこめて聞け心をこめて話せ
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
