Julien Valroff wrote:
I am pretty sure you are already aware that there is a problem using ATI's drivers with 2.6.10 kernel, [...]
I am aware of the problem. In fact, I'm actually using 2.6.10 myself, but I devised a slightly different solution: I just ripped off the built-in AGP support code, which is where this problem lies (and so did several others before). I'm perfectly happy with the kernel's AGP driver and I believe having a slightly enhanced copy of an older version of the AGP driver within the fglrx module is not a good idea: it will be a source of compatibility problems as long as the kernel evolves, that is, hopefully, forever.
Just in case you are not, here is a link pointing to a patch who worked for me, using Debian kernel-source-2.6.10-3 from sid :http://www.pissflitsche.net/scripts/agpgart_be.patch
This patch works, but only on 2.6.10: you'll lose source compatibility with 2.6.9 and earlier. You'll find a similar patch in the mailing list archives on my site and a better patch from Alexandru Fomin in the "limbo" directory of the source package, but I have issues with this because 1) I don't completely understand the implications of changing from pci_find_class() to pci_get_class() and 2) I think we need to call pci_dev_put() before returning after errors, and only there.
To sum it up, I *don't* want to apply this patch to the code I'm distributing: I'd rather leave users with a module that doesn't build than with one that builds but causes crashes or misbehaves somehow, so I'll wait for an official fix from ATI.
Maybe it is worth posting a piece of news on your web site so that people know you are aware of this problem.
That was the plan but I didn't get around it either. Lack of free time, sorry. I'll do it now if I don't get distracted by work.
-- Ciao, Flavio