radeon pci ids in driver/xserver...

[Roland Scheidegger]

Benjamin Herrenschmidt wrote:
>> So, does this mean the ids have to be added to both xf86PciInfo.h
>> and atipciids.h now? Please tell me this ain't true. I'm not
>> exactly sure though why the ids need to still be in xf86PciInfo.h
>> too?
> 
> No, only atipciids.h. I didn't remove them from xf86PciInfo.h yet 
> because I intend to do so as a batch once all drivers are converted
> to avoid issue where one driver that i missed may look for some ID
> from a different vendor for whatever reason (like the radeon driver
> looking for some DELL and AMD stuff etc...). Also I didn't have a
> good up to date checkout of the server at hand yesterday and had to
> go :)
Ok, I'm relieved. I was just worried because you didn't remove the ids
there... It just seemed to me that maybe the ids there are needed for 
pcidata/scanpci modules, for whatever they are good for.

> I know and that can certainly be improved upon, right however, my
> goal is only to remove the xf86PciInfo.h to break any xserver
> dependency on building the driver with new IDs. We can surely work on
> something better specific to the radeon driver & DRI module in a
> second step, like a table if IDs with capabilities etc... or
> whatever...
I think maybe dri should use some sort of list from the drm (just 
include at source level), though the current format of this list doesn't 
quite look useful for that (and changing it might not be an option as 
it's used by all drivers). Or ddx should just export an (exact!) family 
id instead of the actual pci id (such a change would definitely kill 
compatibility between old/new ddx/dri drivers though). Some more places 
should die when the ati wrapper driver gets killed, and the list in 
radeon_probe.c could be built automatically from the one in 
radeon_chipset.h easily. That should bring the amount of id lists down 
to a reasonable amount...

Roland