10:13karolherbst: first run of my MT fixes on nv50
10:13karolherbst: this will be interesting
10:14karolherbst: the changes in the drivers are actually quite small so that might just work without trying it out first :D
10:19karolherbst: nv30 will be interesting, because the driver is quite different in a few areas actually...
10:19karolherbst: and I am sure I broke it
10:27Santurysim: I've almost read the introductory course and I would be grateful if somebody helped me to pick a task to begin with (given I have nve6)
10:44karolherbst: Santurysim: uhm.. any bugs?
10:45karolherbst: usually it's easier to just figure out what you want to have/know and work through it
10:45karolherbst: otherwise GPU load information would be a fun topic, but I've written most of the patches already, just no time to clean it up
11:00mupuf: +1 for using PMU's engine load counters and exposing that
11:00mupuf: you would be the 3rd one to work on it :D
11:01karolherbst: mupuf: I actually have working patches though :D
11:01mupuf: so did I ;)
11:01Santurysim: karolherbst: about bugs: after resume from suspend, kde notification area was messed up. When I tried to make a screenshot, desktop got frozen, and I rebooted via ssh
11:01karolherbst: sounds like a bug
11:02mupuf: And so did I with reclocking on nv50 :D Took 3 days to reclock my nv86, and 3 years later it still wasn't enabled by default :p
11:03Santurysim: karolherbst: I suppose that this bug is marked as hard on trello
11:03karolherbst: might be... the problem is rather in figuring out what's actually wrong
11:03karolherbst: not an easy thing to do as a beginner
11:03karolherbst: mupuf: yeah... reclocking....
11:04karolherbst: but in the future (tm) we might get something
11:04Santurysim: I can try to collect some data... like setting verbosity level to max possible
11:04karolherbst: yeah.. that generally doesn't help
11:05karolherbst: this will be a mesa issue somewhere
11:05karolherbst: or we don't restore data properly on resume
11:05karolherbst: really hard to figure out with just guessing
11:05karolherbst: you might get errors inside dmesg
11:05karolherbst: and those might point into some direction
11:07karolherbst: this nv50_bufctx_fence call inside the vbo kick notify handler is killing things..
11:10mupuf: Santurysim: oh dear, resume bugs are not something I would start with
11:13Santurysim: Hmm... I can clean up code
11:21karolherbst: Santurysim: for that I'd focus on mesa for the moment as the kernel side is a bit in a... sad state I'd say
11:22karolherbst: not code wise, but it's easier to get patches from random people into mesa than into the kernel
11:36Santurysim: karolherbst: got it
12:51Santurysim: Here is what kernel said about freeze: https://pastebin.com/C8pdhJFQ
12:52Santurysim: Now kernel complains about blocked kworker
12:55karolherbst: we have this weird bug on some kepler cards where the context switching firmware hangs
12:55karolherbst: and we don't really know what this is all about
17:39mupuf: karolherbst: that's the last gen we can fix this sort of bugs :s
18:21karolherbst: mupuf: sure.. but at least the files nvidia provides to use are "correct" more or less and issues are most likely caused by us using it wrongly :D although there were bugs and skeggsb_ found those :D
18:22karolherbst: anyway.. we are not alone with those bugs and nvidia helps us
18:23karolherbst: maybe slowly and with some pain, but they usually do
18:28mupuf: Well, good to hear!
18:31Santurysim: Their behaviour is weird, to my mind
18:42karolherbst: Santurysim: it kind of is
18:42karolherbst: they are kind of protective over their IP
19:35karolherbst: nv50 first try
19:35karolherbst: main: [58672/58672] Pass: 54668 Fail: 1988 Skip: 2003 ExpectedFail: 0 UnexpectedPass: 0 Crash: 1 Timeout: 1 Missing: 0 Flake: 11 Duration: 2:20:25 Remaining: 0
19:35karolherbst: mt fixes: [58672/58672] Pass: 54622 Fail: 2022 Skip: 2003 ExpectedFail: 0 UnexpectedPass: 0 Crash: 1 Timeout: 19 Missing: 0 Flake: 5 Duration: 2:57:54 Remaining: 0
19:35karolherbst: honestly.. I expected worse
19:36Santurysim: Low number of flakes
19:42karolherbst: okay.. but overall it looks okay :D
19:42karolherbst: I just need to figure out those timeouts
19:43karolherbst: nv30 will be fun to fix....
20:16mangix: karolherbst: know where your patchset to enable reclocking on maxwell is?
20:17mangix: ah I think I found it
20:30mangix: lol love the comment: https://github.com/karolherbst/nouveau/commit/e3ff0570077f4a7bdea11a736d182eedbfea40fd
20:30karolherbst: yeah .. well :D
20:31karolherbst: it was a wild guess
20:31karolherbst: maybe I ask Nvidia the next time I have a chance :D
20:32karolherbst: but I think it's just for outdoor usage and they don't want the driver or hw to go crazy
20:40mangix: i rebased the patchset for some reason. https://github.com/karolherbst/nouveau/commit/1883fba75c633e84fa9f8ab687af4793f2650d06 seems to not be present in master anymore
20:44karolherbst: mhh weird
20:46mangix: guess something got refactored
20:46karolherbst: yeah.. shouldn't
20:46karolherbst: maybe something strange happened
20:57reinist12: Anyone saw imirkin lately? Inspecting his pics, something rather weird for a jewnamed guy i spotted, typical case of what looks to be big resources of mental kind, i.e conduction in the head :), he should be actually very smart and it to some degree shows out too. At the end of today linux is till just amazing. That a coordination like this hit the nail lovely stuff, many braniacs should have real chanches there and joy involved. I was always braniac
20:57reinist12: my own too. I do not care until my people start to suffer and dropout. Once again, the system is amazing, and the therapy is amazing to be involved in something like that to train your thinking even further.
21:06reinist12: But this is what the world is about, as not all have such invincible resources present intelligence is something that gets a bad feedback quite oftenly, you're looking at 95percent of people without those resources and 5have that in excess, so what chanches to they ever have not to behave like this, cause another man is so much more talented, they also have to do what they do.
21:13reinist12: Human beings body is that flexible that the differences aren't being spotted in random until you start to investigate, that some are rather vastly more special than some others actually, cause the default setting is allready so mighty too.
21:18reinist12: it's a big mgnitude of difference, without any violance involved to another, someone opposite to that is always getting a win, so at the end of the day they are crushed mentally like, ouh i am good, but the othe rins is always touch better which technically cracks them up.
21:18mangix: karolherbst: https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/drivers/gpu/drm/nouveau?h=next-20210701&id=9d350c5e5156792ba7c87b3b1e897656dc2a88e5
21:22karolherbst: mangix: mhhhh
21:23mangix: https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/commit/drivers/gpu/drm/nouveau?h=next-20210701&id=22dcda45a3d1dfe6eeb4ab0a3b9aaa2333cb649d . Joy.
21:24reinist12: So indian court assigns a death pentalty to thei WHO woman, prescribing wrong meds to people who died, and this is only cause of stupidity it was not even conspiration, cause practically she did not understand the composure of things in the mix, and the function of each element there. My case is different, i'd be rather happy to avoid those things but to me everything was assigned on purpose instead to take me down.
21:32karolherbst: mangix: well.. the ban was intentional but I set the channel to private
21:32karolherbst: that person is... well known to put it mildly
21:39orbea: he already hit #dri-devel and #radeon *sigh*
21:44ccr: IIRC on freenode's ircd there was an option to "link" channel bans (and maybe +q modes) from other channel, would've been useful feature (though not used on fdo channels there either)
21:46ccr: no idea if oftc ircd has that feature
21:47mangix: interesting that he uses IRC to spam. Maybe got banned on discord...
21:49ccr: some kind of mental case, I'd venture. sometimes difficult to say if it's a markov-chain or a real human .. but based on how he gets pissed off by certain kinds of interactions, probably a human.
21:49karolherbst: ccr: yeah.. human
21:49karolherbst: and already known for years
21:51ccr: yeah, I've "seen" him around for years too, due to hanging on #intel-gfx since 2017 or so.
21:52ccr: or even longer .. I forget. years roll by. :D
21:52karolherbst: yeah... way longer
21:52karolherbst: it's a sad case but nothing we can do about it
21:55orbea: the healthiest option for all involved parties is to as politely as possible show him the door...
21:55karolherbst: HdkR: I am sure it's before your time.. but is there a stupid reason why on nv84 or so a fence might not get updated or something like that?
21:56karolherbst: ehh mhh
21:56karolherbst: I have an idea
21:59karolherbst: something is funky
22:24karolherbst: that moment when usleep fixes a bug
22:25karolherbst: HdkR: I think Ilia mentioned an issue where on older GPUs we mess up big times when we submit shit too fast :D
22:40karolherbst: okay.. CI pipeline for MRs is merged :3
22:53karolherbst: here as an example: https://gitlab.freedesktop.org/drm/nouveau/-/merge_requests/3