RFC: AGP + nouveau + quirks

[Alex Deucher]

On Fri, Oct 11, 2013 at 12:33 AM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
> I was just looking at
> https://bugs.freedesktop.org/show_bug.cgi?id=20341 (see last comment).
> Basically he's able to to use the card with agpmode=2 but not the
> auto-detected agpmode=4. I also saw that the radeon driver has a large
> quirks list, and it seems silly to reproduce it in nouveau.
>
> However the quirks in the radeon driver are a bit odd -- they are per
> agp chipset/card chipset pair, rather than a blanket max speed for the
> whole agp chipset. I find this surprising, but I suppose I don't know
> that much about all this stuff. I was thinking for nouveau, it would
> just be at the agp chipset level.
>
> Do you guys think that it's reasonable to make the quirks list
> dependent on the agp chipset only? In that case, I can refactor and
> create The One True List (tm). Otherwise I'll just hack something into
> nouveau inspired by the radeon list. [Also for the single list, it
> might be better for the agpgart driver itself to know about the
> restrictions, but that's probably more refactoring than I care for.]
>
> Comments/suggestions/concerns appreciated.

It's very much dependent on combinations of gpus and chipsets in my
experience.  One could probably distill down the actual quirks to some
general chipset or GPU specific ones, but unfortunately AGP is getting
so old now that errata lists for old chipsets and GPUs are probably
hard to come by.  I don't foresee any new AGP systems any time soon,
so I'm not sure it's worth putting too much effort into it.

Alex