On Tue, Jan 8, 2019 at 8:37 AM Daniel Thompson daniel.thompson@linaro.org wrote:
On Tue, Jan 08, 2019 at 11:25:56AM +0100, Ard Biesheuvel wrote:
On Tue, 8 Jan 2019 at 10:36, Renato Golin renato.golin@linaro.org wrote:
Hi Ard,
I don't have the whole context, but my read of that email is:
- The claim is that "some people had some issues with indeterminate
hardware and indeterminate versions of mesa", however... 2. He "did run the WebGL CTS suite, but that resulted in some hangs from the the max-texture-size-equivalent test, and some browser-level weirdness after some tests where later tests all fail"
I don't think those two statements are compatible. He can reproduce lots of failures on his own machine, probably just didn't have time to investigate all of them in detail.
Furthermore, he claims the failures are "due to what [he has] to assume is a browser bug" without any evidence to support it, and later on claims the driver is fine because "accelerated WebGL [...] in practice worked just fine (at least in my usage of it)".
To me, it smells like someone complaining that a broken piece of software is black-listed and shouldn't have because everyone know it's broken anyway, but it kinda works, so it's fine.
I think one of the complaints is that there is a double standard here.
I was curious enough about this to at least run the test suite recommended in the bug tracker.
My Intel/x86_64/Intel/F29 laptop and my Socionext/AArch64/nVidia/buster Developerbox work pretty much the same. Which is to say that the test suite causes the graphics stacks of both machines to lock up entirely (including rejecting any attempt to switch virtual terminals).
Regrettably that makes it difficult to access and share the test results prior to failure (I don't *think* the two machines fail at the same point although I did neglect to photograph the display so I can't be sure).
If you hadn't already, this tends to be the sort of thing where having another machine to ssh from can be useful..
BR, -R
Daniel.
If that's a fair reading, I personally support blacklisting, and I second Chromium's suggestion to make the driver a first-class citizen as a way to remove it from the blacklist.
Who would do such work is a question that, to me, has no easy answer...
- Linaro has no GPU working group and NVidia is not a member, so
working on their drivers, even if open source, if we had the expertise, would be free lunch. 2. NVidia doesn't care about OSS drivers (much) because they already have their own proprietary ones on the platforms they care about. 3. Arm can't work on OSS NVidia drivers as that would compete with Mali.
Someone could hire community developers to do that work, or at least to validate it on Arm and create a list of bugs that need to be fixed, with more details than just "works for me". Linaro could do the validation matrix but would have to do it for both Arm and x86, and then hope the nouveau community would pick up the tab and fix them all. We'd also have to provide access to hardware for them to test, etc.
An alternative crappy solution would be to IFDEF the inclusion in the blacklist *exclusively* for Arm, given even we still don't care much about bugs in NVidia+Arm. But that's gotta lose some kudos from whomever proposes it and will be met with fierce refusal from the Chromium community.
The bottom line is: not many people care about nouveau on Arm, given the only platform that actually uses it today is the Synquacer.
Nouveau is also used on non-PCI NVidia ARM SoCs with integrated graphics.
I may be wrong, there may be a thriving community for NVidia on Arm out there. If there is, MHO is that we should talk to them instead of putting pressure on Chrimum to lift the ban. If not, there's no pressure to be put in the first place.
If, however, you propose we put pressure on nouveau specifically, for both Arm and x86, then I think it should come from the other side (x86) first, and Arm's not a first-class citizen on nouveau anyway. All in all, we're at the very bottom of the priority stack, there's no pressure we can put on anything, but Linaro could do the heavy lifting of validation matrix and help the nouveau community to identify and validate their fixes.
Thanks Renato