please note that normally a PCI device that does represent
a PCI to PCI bus bridge should carry the AGP-caps registers
if it does support AGP transfer protocoll.
please further note that thare are already lots of systems
out there where the bus bridge is on device XX whilst the
the AGP-caps registers do reside on a different pci function
of the pci device or even on a different pci device at all
(meaning that its still the same chip die but not the same pci unit).
please finally note that there were even systems designed
that do not have any PCI-PCI (AGP) bus bridge device at all
but do have some random device with AGP-caps registers
and the AGP master device (=graphics) is found on bus #0
in the system, and bus #0 is the one and only bus in the system.
let me call the first case "sane" and the other two cases "insane".
and now think of an autodetection sheme that can fiddle out
all those three and a half cases automagically.
if there is auto detection on a per device base
then i expect the code to undergo some learning curve
until all odd box setups are identified and addressed.
another thing on the state of oddness with AGP devices:
my expirience showed me cases where the AGP master and slave devices
had multiple AGP-caps register sets, but only one of them was allowed
to be programmed for making it operable. for the future there might
pop up systems that do expose multiple master devices that all might
need programming so that there is no general way of doing it right.
-Alex.
PS: never understood how intel urged all the world for making GART paging
an integral component of the AGP standard whilst paging setup
got not really standarized up to AGP v2.0 - further paging is
anyways a rather general concept, found in CPUs, PCI/AGP graphics
and of course mainboard chipsets.
> -----Original Message-----
> From: Linus Torvalds [mailto:[EMAIL PROTECTED]]
> Sent: Saturday, January 11, 2003 19:29
> To: Jeff Hartmann
> Cc: Dieter N�tzel; [EMAIL PROTECTED]
> Subject: RE: [Dri-devel] [ANNOUNCE] AGPGART Version 2.0
>
>
>
> On Sat, 11 Jan 2003, Linus Torvalds wrote:
> >
> > Please don't add some silly bus scanning. We have perfectly good
> > mechanisms for registering drivers, and if the AGP bridge
> shows up as a
> > PCI device, it should just use the "register_pci_driver()"
> thing, and then
> > the PCI layer will do the right thing.
>
> Note that this is even more important since the AGP driver
> may not show up
> as a PCI device AT ALL. There's no way that the AGP layer can
> sanely scan
> some bus to find anything, because the only thing that knows
> where the AGP
> device may be and how to detect it is the low-level AGP driver itself.
>
> Linus
>
>
>
> -------------------------------------------------------
> This SF.NET email is sponsored by:
> SourceForge Enterprise Edition + IBM + LinuxWorld = Something 2 See!
> http://www.vasoftware.com
> _______________________________________________
> Dri-devel mailing list
> [EMAIL PROTECTED]
> https://lists.sourceforge.net/lists/listinfo/dri-devel
>
>
