On Tue, Jul 7, 2015 at 5:05 PM, Ben Skeggs <skeggsb at gmail.com> wrote:> On 8 July 2015 at 06:06, Ilia Mirkin <imirkin at alum.mit.edu> wrote: >> Ben, >> >> Looks like the reality is that glamor is just not hooked up properly >> in the nouveau DDX. Mainly it's missing DRI2, which in turn means no >> core GL contexts, and probably lots of other issues. While this could >> probably be fixed somehow, I doubt there's any advantage to using the >> nouveau DDX over something like modesetting nowadays. >> >> How would you feel about dropping glamor support from the nouveau ddx >> and failing to load for GPUs that don't have EXA support (unless >> AccelMode = none is forced for them). That way it'll fall back to >> loading modesetting which should be properly set up for DRI2 and so >> on. > I have no objections to this. In fact, in Fedora at least (I floated > the idea in #nouveau a while back too), in the near future I plan on > having the DDX fail to load on all GPUs where modesetting+glamor can > be used (unless overridden by a config option).IMHO that's a little strong (I assume you mean nv50+ here?). In fact I'm planning to complete my Maxwell EXA impl. The current reality is that modesetting+glamor doesn't render correctly at least on maxwell, but possibly others as well. The EXA paths are very well tested and are stable. I think not relying on mesa in the DDX is a nice advantage too. -ilia
On 8 July 2015 at 07:09, Ilia Mirkin <imirkin at alum.mit.edu> wrote:> On Tue, Jul 7, 2015 at 5:05 PM, Ben Skeggs <skeggsb at gmail.com> wrote: >> On 8 July 2015 at 06:06, Ilia Mirkin <imirkin at alum.mit.edu> wrote: >>> Ben, >>> >>> Looks like the reality is that glamor is just not hooked up properly >>> in the nouveau DDX. Mainly it's missing DRI2, which in turn means no >>> core GL contexts, and probably lots of other issues. While this could >>> probably be fixed somehow, I doubt there's any advantage to using the >>> nouveau DDX over something like modesetting nowadays. >>> >>> How would you feel about dropping glamor support from the nouveau ddx >>> and failing to load for GPUs that don't have EXA support (unless >>> AccelMode = none is forced for them). That way it'll fall back to >>> loading modesetting which should be properly set up for DRI2 and so >>> on. >> I have no objections to this. In fact, in Fedora at least (I floated >> the idea in #nouveau a while back too), in the near future I plan on >> having the DDX fail to load on all GPUs where modesetting+glamor can >> be used (unless overridden by a config option). > > IMHO that's a little strong (I assume you mean nv50+ here?). In fact > I'm planning to complete my Maxwell EXA impl. The current reality is > that modesetting+glamor doesn't render correctly at least on maxwell, > but possibly others as well. The EXA paths are very well tested and > are stable. I think not relying on mesa in the DDX is a nice advantage > too.The reality is that this is what people will be using when the big switch to wayland by default happens, and the idea is to limit the number of codepaths we have to care about and maintain. I'm aware that there are a few rendering issues left in glamor (I'm not sure if it's glamor, or our 3D driver having some bugs), but we should probably just fix those :) I believe Dave plans on doing something similar in Intel/Radeon too. Ben.> > -ilia
On Tue, Jul 7, 2015 at 5:16 PM, Ben Skeggs <skeggsb at gmail.com> wrote:> On 8 July 2015 at 07:09, Ilia Mirkin <imirkin at alum.mit.edu> wrote: >> On Tue, Jul 7, 2015 at 5:05 PM, Ben Skeggs <skeggsb at gmail.com> wrote: >>> On 8 July 2015 at 06:06, Ilia Mirkin <imirkin at alum.mit.edu> wrote: >>>> Ben, >>>> >>>> Looks like the reality is that glamor is just not hooked up properly >>>> in the nouveau DDX. Mainly it's missing DRI2, which in turn means no >>>> core GL contexts, and probably lots of other issues. While this could >>>> probably be fixed somehow, I doubt there's any advantage to using the >>>> nouveau DDX over something like modesetting nowadays. >>>> >>>> How would you feel about dropping glamor support from the nouveau ddx >>>> and failing to load for GPUs that don't have EXA support (unless >>>> AccelMode = none is forced for them). That way it'll fall back to >>>> loading modesetting which should be properly set up for DRI2 and so >>>> on. >>> I have no objections to this. In fact, in Fedora at least (I floated >>> the idea in #nouveau a while back too), in the near future I plan on >>> having the DDX fail to load on all GPUs where modesetting+glamor can >>> be used (unless overridden by a config option). >> >> IMHO that's a little strong (I assume you mean nv50+ here?). In fact >> I'm planning to complete my Maxwell EXA impl. The current reality is >> that modesetting+glamor doesn't render correctly at least on maxwell, >> but possibly others as well. The EXA paths are very well tested and >> are stable. I think not relying on mesa in the DDX is a nice advantage >> too. > The reality is that this is what people will be using when the big > switch to wayland by default happens, and the idea is to limit the > number of codepaths we have to care about and maintain.Yeah, in 100 years when it actually works :p> I'm aware > that there are a few rendering issues left in glamor (I'm not sure if > it's glamor, or our 3D driver having some bugs), but we should > probably just fix those :)I'm all for fixing bugs in the 3D drivers. But until that happens, no reason that users should suffer. [I also definitely know some people specially don't have mesa installed so that they get 2d accel but no 3d accel since nouveau can be so hang-y.]> > I believe Dave plans on doing something similar in Intel/Radeon too.I think those kernel drivers tend to be a lot more stable. The nouveau kernel driver hangs left and right though under any actual 3d load. And yeah, the "2d" accel is all 3d anyways, but it's a lot less code and it's been way more tested. -ilia