02:40 <robclark> there is perfetto that has panfrost support.. not sure if there is a standalone thing like fdperf/radeontop/etc

08:08 <anarsoul> alyssa: have you looked into why dEQP-GLES2.functional.multisampled_render_to_texture.readpixels fails on t720?

08:08 <anarsoul> on lima it fails with "Mesa: User error: GL_INVALID_ENUM in glGetIntegerv(pname=GL_MAX_SAMPLES)", so I doubt it's a lima bug

12:35 <alyssa> anarsoul: prbably a mesa/st bug?

12:35 <alyssa> (or a deqp-gles2 one)

12:36 <alyssa> t720 is the only panfrost device we cap to gles2

12:36 <alyssa> and a lot of tests magically pass when faking gles3

13:08 <robmur01> oh, maybe I should try that for my "X won't start on T620/T720" conundrum...

13:15 <alyssa> robmur01: that one is probably unrelated ..

13:15 <alyssa> then again glamor 🤷

13:16 <macc24> robmur01: i think they might mean random programs...

13:51 <robmur01> indeed, turns out even hacking in PAN_DBG_DEQP doesn't help :(

13:53 <alyssa> there's not a debug option for it i don't think

15:52 <robmur01> weird... looking at the Xorg log the difference seems to be at the point where it loads the kms driver - with a T830 present, it's fine, but T620 leads to "Falling back to old probe method for modesetting" then "No devices detected". Why the GPU should make any difference at all for the display bemuses me :/

15:54 <alyssa> I assume you've patched out the GPU allowlist check

15:54 <alyssa> ?

15:55 <daniels> it shouldn't even be that - modesetting will pick up the KMS device, and only later figure out how to (ac|de)celerate its rendering

15:55 <daniels> I'm assuming they are not identical KMS drivers ... you aren't trying to run on fbdev or something are you :P

15:59 <robmur01> yup, the hack are just in panfrost_create_screen() and panfrost_model_name() to allow anything, and adding an hdlcd entrypoint to mesa

15:59 <robmur01> daniels: it's literally the same machine, I'm just unbinding/binding panfrost to/from either GPU and retrying

16:01 <robmur01> having both bound at once seems too confuse most things a bit too much :D

16:06 <macc24> robmur01: what if you try with this config https://github.com/Maccraft123/Cadmium/blob/master/board/duet/xorg.conf

16:07 <macc24> would be funniest shit ever if it worked witht his

16:23 <robmur01> macc24: the default config works just fine though - card0 is T830, card1 is HDLCD, all is well. Close X, poke at /sys/bus/drivers/panfrost such that card0 is now T620, suddenly X is somehow unable to probe card1 properly. Swap GPUs back again, happy once more.

16:23 <macc24> eeh>

16:23 <macc24> ?

16:24 <robmur01> can't really be bothered to reflash a T720 onto the FPGA right now (getting that one to work is fiddly), but AFAIR there was no visible difference in the log between that and T620 last time I tried

16:24 <macc24> that's... super odd

16:24 <robmur01> it's only now that I've looked closely at the difference between the working and non-working conditions

17:04 <robmur01> well, flip. Seems the relative order of paths under /sys/devices/platform/ matters... urgh :(

18:33 <chewitt> robmur01 poking T620?

18:35 <macc24> hm i can try poking t628 if there's a cheap exynos chromebook locally that i can afford...

18:37 <macc24> ooh there's auction for one locally, not paying more than $10 for it lmao

18:37 <chewitt> Odroid XU4 is T628 also

19:21 <macc24> chewitt: i prefer laptop form factor

19:21 <macc24> i never really liked sbc computers... they are pain to use as dev machines

19:22 <HdkR> They are easier for me since I just SSH in ;)

19:23 <macc24> well if something's only a sbc/devkit then i can deal with it

19:23 <macc24> HdkR: that implies ethernet jack/wifi works

19:23 <macc24> that's... not always the case

19:24 <HdkR> It's a pretty good assumption, since most people bring up USB at least

19:25 <macc24> well yes

19:26 <macc24> but everything likes to break and if i don't test some arm chromebooks before kernel release, who does?

19:26 <HdkR> Such is the life of a niche device maintainer :D

19:27 <alyssa> My fleet is a mix of SBCs, laptops, and phones

19:27 <alyssa> Phone is by far the worst of the 3

19:27 <macc24> honestly i quite like it, it sends me into every corner of kernel world and i learn stuff quickly so i have basic understanding of most kernel driver types

19:27 <alyssa> SBCs and laptops can go both ways depending on the use case

19:28 <alyssa> oh and a desktop of course

19:28 <HdkR> alyssa: How soon until you add server to that list?

19:29 <macc24> alyssa: honestly... i prefer phones with good debug interfaces than sbc computers

19:29 <HdkR> Get like an Ampere AltraMax with FPGA cards plugged in to it? :)

19:29 <macc24> i don't think there's a device form factor that i *don't* have in some form...

22:10 <daniels> robmur01: hmm, modesetting should probe the card* devices until it finds one which is KMS-capable

22:10 <daniels> robmur01: competing window systems do that ;)

22:36 <robclark> sbcs/laptops/tablets/etc.. I'd just be happy if more of them supported ccd/suzyq since that has been such a lifesaver on so many -rc1's ;-)