13:44imirkin_: Lyude: ping
13:44imirkin_: if you get a chance, have a look at https://bugs.freedesktop.org/show_bug.cgi?id=111044#c13 - seems up your alley
13:46imirkin_: HA! a ha ha HA! i got black rectangles in chrome on intel.
13:58HdkR: imirkin_: Woo, everyone gets black rectangles! :D
13:59imirkin_: time to blacklist intel on chrome... for all mesa versions and all mesa chips
13:59imirkin_: er, all intel chips
14:21HdkR: imirkin_: Look in your chrome flags and see if you have.....out of process rendering? enabled?
14:21HdkR: I think that is what causes it
14:22imirkin_: no, what causes it is nouveua - that's why it got blacklisted from chrome.
14:23imirkin_: Out-of-process Rasterization: Disabled
14:23imirkin_: maybe these black squares were different than other black squares from years past? :)
14:26imirkin_: and of course gmail doesn't work in firefox, so i can't switch
14:27imirkin_: (does anyone else see this? page up/down don't work when editing a reply to an email...)
14:27imirkin_: [in gmail, on firefox]
14:30karolherbst: imirkin_: are you a moderator for the nouveau ML?
14:30imirkin_: i don't think anyone nouveau-related is.
14:31imirkin_: maybe marcheu
14:31orbea: imirkin_: does gmail work in a mail client, like maybe mutt or claws-mail?
14:31imirkin_: orbea: don't know, don't care
14:32imirkin_: gmail has the one true correct way of displaying email, i've never seen any other client implement it.
14:32orbea:shrugs, might of helped avoid chrome :)
14:40karolherbst: imirkin_: mhh, I think we might want to have more moderators then :D
14:42imirkin_: well, there are others, just not directly nouveau-involved
14:42imirkin_: like i think MrCooper might be one?
14:42karolherbst: sure, but it would be cool if some nouveau related people would be moderators as well :)
14:43karolherbst: imirkin_: anyway, the intern was sending his first patches today, and I kind of want others to review it :D
14:43karolherbst: imirkin_: marcheu is indeed the admin of the nouveau list
14:45karolherbst: imirkin_, RSpliet: https://patchwork.freedesktop.org/series/63572/
14:45karolherbst: it was roys trello card :)
14:45karolherbst: ohh wait
14:45karolherbst: mupufs actually
14:51imirkin_: Lyude: it feels liek there are a lot more reports of DP flakiness recently...
14:53imirkin_: karolherbst: is he here?
14:54karolherbst: uhm, not right now. He has no bouncer and he already left the office
14:54imirkin_: (the intern)
14:54karolherbst: mmenzyns would be his nick though
14:54imirkin_: ok. you should impress on him that effort should be invested in writing commit messages
14:54imirkin_: ones that read as though you spent more than 0.007s on them
14:55karolherbst: sure, but I also didn't want him to submit perfect patches in the first round :p
14:55karolherbst: would be to boring
14:57RSpliet: karolherbst: yeah I just spotted the patches on the ML. Is Mark fresh blood on nouveau, or is this a sidejob kinda thing?
14:57karolherbst: but yeah, he will be in the office tomorrow as well, so I can work with him on that. I just kind of wanted him to get used to the community engaging stuff and to learn how to react to reviews adn stuff
14:57karolherbst: RSpliet: well, intern at RH
14:57RSpliet: Awesome. If he keeps going at this pace, get the overlords to offer him a contract too :-P
14:58karolherbst: that's always the idea
14:58RSpliet: Could always do with another pair of hands
14:58karolherbst: anyway, the bigger issue is to find more simplier task to work on
14:59karolherbst: the next biggest thing I was coming up with was the egpu stuff, where it's essentially crawling through tons of backtraces and reorganize the code to suck less :/
14:59karolherbst: still requires like 0 knowledge about GPUs
14:59karolherbst: but quite a lot about the code itself
14:59karolherbst: mhh, I have some G84 GPUs with this reverse fan control thing
14:59karolherbst: this _might_ be good enough
15:00imirkin_: or the GF108 fan thing
15:00karolherbst: but... installing fitting nvidia drivers for reverse engineering is just super pain
15:00imirkin_: someoen from nvidia gave us a pretty detailed explanation
15:00karolherbst: imirkin_: but that requires that I have the hardware
15:00karolherbst: more or less
15:00imirkin_: GF108's are not hard to come by
15:00karolherbst: yeah... maybe
15:00imirkin_: they're the cheapest of them all
15:00imirkin_: probably like $10 on ebay
15:01karolherbst: question is rather if all GF108 are hit
15:01karolherbst: uff.. we have no gf108 in brno
15:01karolherbst: or at all at RH
15:01imirkin_: karolherbst: can you check bin/fbo-stencil copypixels GL_DEPTH32F_STENCIL8 -auto -fbo on mainline?
15:01karolherbst: will do after I got home
15:01imirkin_: mareko fixed acceleration of stencil copies
15:02karolherbst: mhh, I really need to come up with more little things to work on :/
15:03karolherbst: nothing really in the trello I would consider be easy _and_ for the kernel
15:04imirkin_: if you have access to a fancy display (high-bpc, or supporting yuv420), i have some ideas
15:04imirkin_: specifically, someone needs to RE how yuv420 can be done on kepler
15:04imirkin_: (over hdmi)
15:05karolherbst: we have plenty of kepler GPUs with HDMI
15:05imirkin_: that someone is going to be me eventually, but i'd be most happy for someone to beat me to it
15:05karolherbst: no idea ifwe have a high-bpc display
15:05imirkin_: well - don't need high-bpc
15:05imirkin_: just one that supports yuv420
15:05karolherbst: ohh, okay
15:05karolherbst: how do I figure that out?
15:05imirkin_: (for this particular task)
15:05imirkin_: read the edid
15:05karolherbst: ahh, I see
15:05imirkin_: if it's a 4k tv, it definitely supports it
15:06imirkin_: if it's a 4k computer monitor - uncertain
15:06karolherbst: ohh, I see
15:06imirkin_: (ok, maybe not *definitely*, there are few guarantees in life, but let's say "likely")
15:06karolherbst: we have multiple TVs here, but no idea if 4k and no idea if people would get upset if we just unplug those screens from whatever they are doing right now :D
15:07karolherbst: but.. I kind of expect that especially reverse engineering might still take some times, but maybe that display stuff isn't all that complicated
15:08imirkin_: i mean ... no comment. it can be easy, it can be hard. depends on how easy it is ;)
15:09imirkin_: if it's writing some mmio reg, it's a lot easier
15:09imirkin_: if it's some weird evo command, it's harder, since we don't have a tracer for evo
15:10imirkin_: (well, i bet ben knows how to do it, but i don't...)
15:10karolherbst: yeah.. me neither
15:10karolherbst: but that stff is usually documented
15:11karolherbst: no idea if for kepler though
15:11imirkin_: usually :)
15:13imirkin_: yeah, it's documented
15:13imirkin_: the documentation clearly indicates this is impossible
15:13imirkin_: and yet it happens
15:13imirkin_: so ... :)
15:14karolherbst: yeah.. I just doubt we have the hardware here
15:14imirkin_: no worries
15:14imirkin_: i'm sure i'll get it worked out in the next year or so
15:15karolherbst: anyway, will head home now anyway. If anybody of you comes up with nice smaller taks, feel free to add those to our trello board
15:49Lyude: imirkin_: do you know if we recently enabled atomic?
15:49imirkin_: we did not.
15:49imirkin_: it's behind a flag
15:49imirkin_: could be that distros are flipping that flag by default - that'd be a nasty surprise
15:50Lyude: I doubt it
15:50Lyude: There isn't all that much userspace usage of atomic right now for stuff like gnome-shell
15:50Lyude: and the atomic support in X is, flimsy at best
15:53imirkin_: not sure why ben hasn't flipped it on by default ... presumably just fear of the gathering darkness
22:06abooker: Anyone have a few minutes to help me with a blacklisting issue?
22:07abooker: /etc/modprobe.d/blah.conf doesn't seem to have an effect
22:08abooker: And nouveau really seems to have issues with the Intel 620
22:10RSpliet: abooker: could you clarify a couple of things please? Firstly: are you trying to add nouveau to a blacklist or remove? When you say "issues with the Intel 620", what kind of issues are we talking about? Which GPU is it anyway?
22:11RSpliet: Oh... are you talking about the Intel HD Graphics 620? If you have a laptop with hybrid graphics, nouveau just suspends the NVIDIA graphics card until you set it to work. Which never really happens accidentally.
22:12abooker: Happy to. I'm trying to blacklist nouveau. The issue I am seeing is that with Nouveau loaded I get severe CPU stalls
22:12abooker: And resume from sleep never recovers the display
22:12abooker: One sec I'll get details of the nVidia h/w
22:14abooker: Yes, I think you have it right. It's a Lenovo config. hwinfo fails to produce good output but I'm happy to fish any logs that're relevant.
22:17abooker: dmesg is littered with entries similar to:
22:17imirkin_: abooker: boot with nouveau.modeset=0
22:17imirkin_: that will force nouveau to do nothing even if it's loaded
22:18abooker: I will try that again but it wasn't useful previously ... thanks for the help
22:19RSpliet: abooker: on the blacklisting issue: you might have to rebuild your initramfs after blacklisting, in case nouveau was stuck in there by the kernel installation script.
22:19RSpliet: You may have to consult your distro for specifics
22:20abooker: intiramfs is pretty minimal only containing intel microcode
22:20RSpliet: as for logs: dmesg would be nice. Could you stick an unfiltered dmesg (as in: don't just fish out the nouveau messages, leave the whole log intact if you can) onto pastebin/hastebin/fpaste/... and share the url?
22:21abooker: It's shutting down rn and that has proved problematic
22:21abooker: I'll power cycle after giving it a moment and then put up a link
22:27abooker: Ahh man. imirkin_ suggested modeset=0 and I appliedit to the commandline (rather than modprobe options) and that seems to be very helpful.
22:28RSpliet: Ah I'm glad that'll help you get your laptop running! If you could spare a moment or two to fetch the dmesg log anyway when you have nouveau loaded, you might help us find a better fix for the problem than just wholesale disabling nouveau :-)
22:32abooker: Oh yes, I'll have to bounce and turn modestting back on to generate the kernel traces
22:32abooker: Or fish them out of syslog
22:33imirkin_: there are a bunch of known issues around new laptops
22:33abooker: New hardware and new distro release simultaneousl
22:34abooker: Give nme ten and I'll retry modeset=1 to get info, anything else that would be useful?
22:35RSpliet: imirkin_: I think CPUs with Intel HD 620 (Kaby Lake) are approximately 2-3 years old. I'm always a bit confused with your definition of new, so I thought I'd stick with stating facts and letting you be the judge :-P
22:36imirkin_: new = post-_dsm
22:36imirkin_: new = pascal
22:42abooker: :) I meant new as in first time I've worked with the hardware. Your point is well made.
22:43abooker: OTOH if it's been out a while, why am I bumping my head? Should be slick as snot by now. :-D
22:44RSpliet: abooker: we haven't yet put you to the "is he running a brand new kernel" test... And we don't hold kernels to the same standards as GPUs; 5.1 is pretty much as old as we'd tolerate for bug reports ;-)
22:45RSpliet: There's a couple of fixes up for kernel 5.3 that may improve your situation too... if we understand the source of the problem.
22:46RSpliet: Lyude: https://github.com/skeggsb/nouveau/commit/8ab1e59ada466ed929a78f2d37ca8864c9582179
22:46RSpliet: Is this relevant?
22:47imirkin_: that's only for me.
22:47imirkin_: i am probably the only person for whom that plays
22:47imirkin_: you gotta (a) have both a pre- and post-nv50 board plugged in and (b) enable atomic modesetting with nouveau.atomic=1
22:48RSpliet: imirkin_: but not so pre-NV50 that you'll end up with an AGP card. Right, that must be just you :-P
22:48imirkin_: nah - agp doesn't really affect it
22:48imirkin_: also i have a pci nv5, so you can go pretty far back without agp
22:49RSpliet: But finding a motherboard with both a PCI-E slot and AGP is tricky.
22:49imirkin_: i think nv4 was agp-only
22:49RSpliet: Ah yeah
22:49RSpliet: I can still get my hands on a working AGP NV3... but I have no idea what to do with it
22:49imirkin_: give it to mwk :)
22:50imirkin_: he's collecting the remaining specimens in existence, i suspect yours would complete the collection.
22:50RSpliet: Hahaha, is that for the big bonfire of GPUs that don't make sense, ever...?
22:50imirkin_: nv1 was the crazy one, no?
22:50imirkin_: with not-triangles
22:53RSpliet: "Nvidia's final RIVA 128 drivers for Windows 9x include a full OpenGL driver. However, for this driver to function, Windows must be set with a desktop color depth of 16-bit."
22:53imirkin_: sounds reasonable
22:53imirkin_: i'm almost sure i ran in 16-bit color back in those days too
22:53imirkin_: to get 1152x864 on my monitor :)
22:53abooker: Old man is slow, is this what you were meaning by 'clean' dmesg? https://pastebin.com/0ecZ3k9d
22:54RSpliet: abooker: That's the one, thanks!
22:54abooker: wrt kernel rev, I knowyour pain. Mine is an environment considered 'stable' so I can only dream of v5.
22:55abooker: Is there a way for me to chip in a few dollarydoos to show personal appreciation?
22:55aaronp: RSpliet, is that 16-bit mode vs. the normal 8-bit mode? :)
22:55imirkin_: aaronp: now that's just trolling! :p
22:56abooker: I played pong on a machine my BIL brought back from Hong Kong. Monochrome _all_ the way!
22:58RSpliet: abooker: it hangs in nvkm_pmu_reset. There's been a few fixes since 4.19 on that front, some of which indeed await the release of kernel 5.3. Not sure if they have been backported to 4.19LTS
22:59RSpliet: aaronp: no, that's 16 colour mode. My VIC-II doesn't do much more sadly
23:00abooker: I'll try and collect another sample with the resume issue and compare backtraces. It sounds like it's uninteresting but if it's different I'll pipeup in case.
23:01RSpliet: abooker: you may if you find time, but I suspect that until you manage to run a 5.3 kernel (which isn't even out yet. It's crazy!) the info might not be very useful.
23:03abooker: Ok, it's been a while but I'll look into compiling from the tree and see if I can get it upright.