PowerPC agpmode issues

[Mike]

Certainly 750~800 fps in glxgears vs 3000+ in debian squeeze, i cant bring
myself to say that it's an acceptable situation no matter how tired i am of
the problem knowing how well the setup could do. It's clear that the
implementation is broken for everything but x86, if that's a acceptable
situation to then perhaps a recommendation to use a BSD derived system on
everything other then x86 ought be the recommended not more hacky crippling
workarounds, this is turning into a hackjob resembling the 90's already,
just to get to anything remotely functional. Might as well deprecate the
entire AGP system

On 8 February 2016 at 12:41, Boris Reinhard <reinhard.boris at googlemail.com>
wrote:

> Definitely would have made sense for years, but could someone possibly
> look into a proper solution?
>
> Michel Dänzer <michel at daenzer.net> schrieb am Mo., 8. Feb. 2016 11:00:
>
>> On 05.02.2016 11:47, Mike wrote:
>> > Hi.
>> > Managed to get the Radeon R300 running on mesa 11.1.1 with an old 2013
>> > patch from Michel Dànzer, next problem is of course enabling agpmode,
>> > running with pci-mode with radeon.agpmode=-1 works, but is of course
>> > slow, and seems to load the cpu a lot.
>> >
>> > Upon initial investigation i could not initially believe agp could be
>> > this this broken for this long, until i found this.
>> >  "committed with Ben Skeggs on Feb 26, 2013"
>> >
>> https://github.com/DespairFactor/bullhead/commit/650e1203c11354ba84d69ba445abc0efcfe3890a
>> >
>> http://lxr.free-electrons.com/source/drivers/gpu/drm/nouveau/nouveau_agp.c?v=4.2
>> > #ifdef __powerpc__
>> > /* Disable AGP by default on all PowerPC machines for
>> > * now -- At least some UniNorth-2 AGP bridges are
>> > * known to be broken: DMA from the host to the card
>> > * works just fine, but writeback from the card to the
>> > * host goes straight to memory untranslated bypassing
>> > * the GATT somehow, making them quite painful to deal
>> > * with...
>> > */
>> > if (nouveau_agpmode == -1)
>> > return false;
>> > #endif
>> >
>> >  and now later this:
>> >
>> https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/nouveau/nvkm/subdev/pci/agp.c
>> > #ifdef __powerpc__
>> > /* Disable AGP by default on all PowerPC machines for now -- At
>> > * least some UniNorth-2 AGP bridges are known to be broken:
>> > * DMA from the host to the card works just fine, but writeback
>> > * from the card to the host goes straight to memory
>> > * untranslated bypassing that GATT somehow, making them quite
>> > * painful to deal with...
>> > */
>> > mode = 0;
>> > #endif
>> >
>> > All seems to point to serious issues had around the time of change to
>> > ums to kms and a serious regression hitting the linux kernel? No?
>>
>> Not really. UMS was just using a static GART mapping, whereas KMS is
>> dynamically binding pages into the GART as needed. Apparently, older
>> revisions of UniNorth have trouble with the latter. I think it would
>> make sense to change radeon.agpmode to default to -1 on powerpc as well.
>> Any takers?
>>
>>
>> --
>> Earthling Michel Dänzer               |               http://www.amd.com
>> Libre software enthusiast             |             Mesa and X developer
>> _______________________________________________
>> Linuxppc-dev mailing list
>> Linuxppc-dev at lists.ozlabs.org
>> https://lists.ozlabs.org/listinfo/linuxppc-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ozlabs.org/pipermail/linuxppc-dev/attachments/20160208/a411bdb1/attachment-0001.html>