16:07 <robmur01> chewitt: is that with all the Baikal hacks on top?

16:08 <robmur01> forcing memattr to outer non-cacheable makes me wonder if they actually have ACE-lite coherency wired up

16:08 <chewitt> robmur01 Yes, with their changes on top

16:08 <chewitt> kernel branch I'm using is here: https://github.com/chewitt/linux/commits/samsung-5.14.y

16:11 <robmur01> I have a vague memory of seeing the suggestion somewhere that the GPU on Exynos 542x *can* be coherent, but that the interconnect isn't configured that way by default

16:14 <kinkinkijkin> exynos 5422 in the xu4 is still giving a hard time?

16:14 <robmur01> it never stopped :)

16:14 <kinkinkijkin> yeah I missed a long chunk and was hoping it did, lol

16:18 <daniels> I'm afraid there has not been a community spontaneously form around it and get it fixed

16:18 <kinkinkijkin> I lost my board unfortunately so I can't be of much help with it anymore

16:19 <daniels> most of the 5422/5800 hardware Collabora had has died of old age over the past couple of years

16:19 <kinkinkijkin> not very many more years until the xu4 reaches EOL too...

16:20 <daniels> that's still being sold? :o

16:21 <daniels> I think we still have a couple of working XU3s, but on the other hand they haven't been switched on for a year or two now ...

16:21 <kinkinkijkin> yes, but stores aren't ordering it

16:21 <kinkinkijkin> it's still being manufactured though by hardkernel

16:21 <chewitt> It's been on 'special' discount for a while to run down stocks .. when they're gone it's EOL/EOS

16:21 <kinkinkijkin> oh?

16:21 <kinkinkijkin> that's a change

16:23 <chewitt> or maybe I misread the sale stickers on the website

16:23 <kinkinkijkin> last i had read they were going to continue manufacturing until 2021

16:23 <kinkinkijkin> oh

16:23 <kinkinkijkin> it's 2021

16:23 <chewitt> https://forum.odroid.com/viewtopic.php?p=327649#p327649 implies it's still being made

16:24 <chewitt> the Baikal folks have an interest in seeing it work on their new shiny .. so maybe that sparks some interest in getting T628 to a slightly better place

16:38 <robmur01> FWIW my quest to get the CTS running on T624 has now caught up to where it was before, but I'm stumped by cts-runner dereferencing NULL somewhere to do with EglGetDisplay (from memory, may be inaccurate) :(

16:58 <chewitt> CTS = ?

16:59 <robmur01> Conformance Test Suite - https://github.com/KhronosGroup/VK-GL-CTS

16:59 <chewitt> my petrolhead brain can't get past something to do with Cadillac's .. I'm still jetlagged too..

16:59 <chewitt> ahh

17:00 <chewitt> I should figure out how to build that into my image

17:00 <robmur01> I suppose I could try building KODI if it's not too hard (Arch doesn't package a generic one)

17:02 <robmur01> (or maybe even Kodi... caps fail)

17:03 <chewitt> it has a fair list of dependencies but they are internally packaged and it's fairly simple to build

17:04 <robmur01> hmm, it's gone 6, let's stop pretending to work and fire up the Juno :D

17:06 <chewitt> 9pm here and I will turn into a zombie if I don't get some Zzz so I might abandon ship

17:06 <chewitt> there's a number of official Kodi lurkers in here tho

18:54 <daniels> robmur01: which winsys are you trying to use for it? we gave up and just use Xvfb for CTS because the tests don't really run with anything else

18:58 <robmur01> daniels: I've been using the "null" target since the board doesn't have X

18:59 <robmur01> although I think "default" ends up behaving much the same

19:02 <robmur01> I was under the naive assumption that things might work with plain GBM/DRM...

19:02 <daniels> lololololololololololololololololololol

19:02 <daniels> at a stretch, surfaceless does more or less work for some things

19:02 <daniels> but it's not because we love X11 that we run under Xvfb in CI :P

19:18 <robmur01> sigh, guess it's time to cave and install xorg then

19:23 <kinkinkijkin> xorg isn't so bad, sometimes, occasionally, given that it works and isn't being slow and you're on a fast device and xorg

19:23 <kinkinkijkin> s/and xorg/

19:23 <daniels> robmur01: Xvfb

19:25 <robmur01> eh, I have a physical display, might as well see what's happening

19:26 <daniels> it does make it drastically slower however ...

19:26 <daniels> like, factor of 10

19:26 <robmur01> noted

19:31 <robmur01> (and FWIW the xorg meta-package does include xorg-server-xvfb anyway)

19:53 <robclark> daniels: I thought for deqp-gl* (and probably vk?) we use surfaceless, and we only use x11 for deqp-egl?

19:54 <daniels> I haven't looked in the latest round of changes, but last I looked it was still surfaceful

19:55 <robclark> looks like we are using `export EGL_PLATFORM=surfaceless`

20:35 <robmur01> hmm, there's a try-catch around getConfigListEGL() 5 frames up from the segfault... wut?

20:37 <robmur01> it's been a long time since I C++ed in anger, but I'm pretty sure that's not how exceptions are supposed to work :/

21:02 <HdkR> What an interesting thing to rely on try-catch for

22:03 <robmur01> well, the "default" target has in fact built successfully and runs without crashing, but moans about not finding any configs and the default one not being suitable. Under xvfb-run it tries to use swrast (and fails since I configured that out) rather than Panfrost, and a proper X server also refuses to start, so I utterly give up for now.

22:07 <robmur01> back to what I *can* test, glmark2 -bshadow is crashing again: at "cfg.msaa = mali_sampling_mode(fb->rts[0].view);" in pan_emit_sfbd(), fb->rts is NULL

23:16 <robmur01> hmm, X not starting also appears somehow related to SFBD GPUs... replace the T720 on the FPGA with a T820 and unbind the T620, and lo, it works