Michel D=C3=A4nzer wrote:
> On Tue, 2003-07-29 at 19:31, Ian Romanick wrote:
>>Michel D=C3=A4nzer wrote:
>>>On Tue, 2003-07-29 at 09:06, Chris wrote:
>>>>This works perfectly, only using the DRM and the GL driver, (and addi=
>>>>ForcePCIMode the the config),=20
>>>This will remain necessary until PCI GART is deemed stable with AGP
>>>cards or there's a way to tell what kind of card it is.
>>Where would be the right place to do this? In the kernel? In the DDX=20
>>driver? In the client-side DRI driver? If it's in either of the first=
>>two places, we *can* do it, it would be just a major PITA. We can look=
>>at the bus where the card is and work backwards up the tree. If we hit=
>>an AGP controller, the card must be AGP.
> The DDX driver needs to know to decide whether to enable the DRI with
> PCI GART or disable it, but the kernel might be in a better position to
> find out (though some XFree86 hackers might disagree :)...
Since the DDX driver has root permissions, it can make the same querries=20
that could be made in the kernel to walk the PCI device tree. I feel=20
like this is a decision that, if it can be, should be made in=20
user-space. The thought of writing the code to walk the PCI device tree=20
looking for an AGP controller is not a pleasant one, but I may be over=20
complicating things in my mind.
>>With a lot of the cards you can tell just be looking at the device (or=20
>>subsystem) ID, but I don't think that's the case with the Radeons.=20
>>Don't they use the same PCI device ID for both AGP and PCI versions?=20
> They do for the chips, don't know about the subsystems. Hui?
If the decision can be made solely based on the PCI ID (or the subsystem=20
ID), then it is trivial to do in the DDX driver.