Re: [Dri-devel] __MUST_HAVE_AGP

[F. <jrf...@tu...>]

On Wed, Jun 11, 2003 at 12:04:22PM +0200, Guido Landra wrote:
> Hi,
> 
> I have not clear idea about the dependencies between drm module and 
> agpgart module.
> 
> Only "mga", "i810" and "i830" have __MUST_HAVE_AGP set to 1.
> So they can be loaded only after agpgart.o have been loaded.
> Other drivers, such as "radeon", can be loaded also when agpgart.o is 
> not loaded.
>
> I get failure on DRI startup, before mapping agp memory blocks:
> [dri] AGP return 0! Disabling DRI.
> It is ok only with PCI cards, because this part is skipped.
> 
> __MUST_HAVE_AGP is equivalent to something like not(__HAVE_PCI_CARD)?

That's about it. It means that it requires the AGP to be present and
functional.

> Are there some radeon AGP cards able to work without agpgart?

Which drivers are you using?  PCI support in Radeon was busted some time
ago, which meant that AGP had to be present.


> I have an nvidia NFORCE1-based motherboard.
> Agpgart module does not support my north bridge: does this mean that I 
> can't accelerate any AGP card (radeon 7200 for example)?

That's a difficult question. I don't know whether the Radeon hardware
allows to be run in PCI mode. The issue is that the Radeon driver for
PCI cards relies on the existence of a PCI-GART (it does the same as the
AGP GART, but in the PCI bus) which I don't know if it's present on AGP
cards.

> Do you have any idea to solve this problem?

Not myself. But perhaps someone more familiar with the Radeon design can
give a more accurate answer. Also, isn't there any AGP support for the
NFORCE1 chipset planned?

José Fonseca