01:31pabs3: karolherbst: nothing in dmesg, just the gnome-shell messages saying "kernel rejected pushbuf: No such device"
01:40pabs3: karolherbst: last night I also got a nouveau related vlc crash, here are the related systemd journal entries and gdb backtrace https://people.debian.org/~pabs/tmp/crash.txt
01:40pabs3: hope it is useful :)
09:46karolherbst: pabs3: yeah.. not quite sure yet
09:48karolherbst: pabs3: tried using the propriatary firmware? could be the firmware bug we have
11:33pabs3: karolherbst: /lib/firmware/nvidia/ is installed, does it get autoloaded or is there something to tweak?
11:34karolherbst: pabs3: I meant the NvGrUseFW=1 thing
11:34karolherbst: normaly there is some package to extract the firmware
11:34karolherbst: and needs to be enable with nouveau.config=NvGrUseFW=1
11:35karolherbst: booting with it should print an error about firmware not being found if it's not extracted first
11:35pabs3: is there a way to tell if the firmware for my GPU is in /lib/firmware/nvidia/? I have the Debian package of linux-firmware.git installed
11:36pabs3: I'll try it next time I reboot
11:41karolherbst: pabs3: there should be a special nouveau firmware one
11:46pabs3: karolherbst: hmm, I don't see them in Debian's firmware-nonfree source package nor corresponding binary package, what are the names?
11:46karolherbst: no clue
11:46karolherbst: nouveau-firmware would be the obvious one
11:46karolherbst: besides that? no idea
11:48karolherbst: maybe debian has no package for that
11:48karolherbst: pabs3: we use this script to extract it, but that's a bit annoying if you don't have a distribution package doing it for you https://github.com/envytools/firmware/blob/master/extract_firmware.py
11:49pabs3: the driver in linux.git, seems to just look at nvidia paths: /lib/firmware/nvidia/gp108/acr/bl.bin
11:49karolherbst: that's not it
11:52karolherbst: pabs3: for ubuntu there is this package: https://launchpad.net/ubuntu/+source/nouveau-firmware
11:53pabs3: that isn't in Debian unfortunately
11:53karolherbst: yeah.. although you can probably just install the deb file..
11:54pabs3: will take a look
11:55karolherbst: or maybe that's a different thing? uff... this is just annoying, ubuntu has a terrible interface
11:56pabs3: huh, the package says it is based on this old tarball from 2009 https://people.freedesktop.org/~pq/nouveau-drm/
11:56karolherbst: yeah, then it's something else
11:56pabs3: it looks like it corresponds to the paths in the NvGrUseFW=1 code
11:57karolherbst: arch has a perfectly fine package here: https://aur.archlinux.org/packages/nouveau-fw
11:57karolherbst: at least the nvidia driver is there to download
11:57karolherbst: and the pkgbuild shows what needs to be done
12:02pabs3: hmm, no, nouveau-firmware doesn't appear to be used by Linux
12:03pabs3: perhaps building it from the nvidia-graphics-drivers* source packages is the way to go
12:03pabs3: karolherbst: is envytools the extract_firmware.py to use or re-vp2?
12:04karolherbst: why would you need envytools?
12:05karolherbst: just download the nvidia driver there and the script
12:05karolherbst: and run it lik the pkgbuild
12:06karolherbst: I always think there was a better page on the wiki to describe all of that
12:06karolherbst: but I can't find it
12:06pabs3: well the arch PKGBUILD links to a different extract_firmware.py than the one you linked to above
12:06karolherbst: yep, we fixed it for python3 and stuff
12:06karolherbst: you can use the one I linked to
12:06karolherbst: the result should be the same
12:34pabs3: ok, done using NVIDIA-Linux-x86-340.108.run
12:51karolherbst: pabs3: seems like we have no explicit debug message for that :/ uff.. but I hope you rebooted and got a firmware loading issue before extracting it or will you trust the code and hope the new firmare gets used?
12:51karolherbst: could potentially fix your issue
12:52karolherbst: but.. we have no idea what is broken in ours soo...
12:54pabs3: oh, there is no debug message for which firmware nouveau.config=NvGrUseFW=1 loads? I see, will mv the dir out the way so I can get the error containing the right files
13:17imirkin: pabs3: you must use 320.25
13:17imirkin: or 325.25 or whatever
13:18karolherbst: ohh, missed that part
13:18imirkin: pabs3: use the one from envytools/firmware, not the one from imirkin/re-vp2
13:18karolherbst: pabs3: yeah, just use the nvidia package from the arch page
13:18karolherbst: it's the correct one
13:20pabs3: which version? the envytools script has 325.15 as the only version in ARCHIVE_ORDERS
13:21pabs3: imirkin, karolherbst: ^
13:21karolherbst: pabs3: yep
13:21karolherbst: but you said you use 340.108
13:21pabs3: I did, just downloaded it from Debian. will get 325.15
13:21imirkin: pabs3: yes, 325.15 is th eone to use
13:22imirkin: all available from nvidia
13:22imirkin: pabs3: you're the one with the crashy 550 Ti, right?
13:22pabs3: GK107 [GeForce GT 740]
13:23karolherbst: it's still a crashy GPU :p
13:23imirkin: yeah, but i was going to say that there was some dude in here ages ago complaining about a crashy 550 Ti
13:23imirkin: like ... 5 years ago or more
13:23pabs3: here most recently Xwayland is crashing (seems to be caused by GPU hangs?), before that it was GPU hangs
13:23imirkin: and i don't think i ever found a way to help
13:24karolherbst: pabs3: yeah, the gpu context gets killed
13:24karolherbst: and it seems like.. for no reason
13:24karolherbst: "CTXSW_TIMEOUT" usually means context switching fails, which.. well.. is annoying to debug
13:26pabs3: I got a couple of warnings from the script https://paste.debian.net/hidden/5a1b2a6e/
13:26imirkin: pabs3: that's fine
13:26imirkin: it basically looks for gzip-looking things for the archives
13:26imirkin: and more gzip-looking things exist, but contain the wrong data
13:26imirkin: (or are just random bytes that happen to start with the gzip magic string)
13:28pabs3: k, installed and will check the filename before enabling it
13:34pabs3: karolherbst, imirkin: so once I've booted with the right firmware loaded, I should just wait and see what happens? anything else?
13:34karolherbst: nope, that's all
13:35pabs3: k :)
14:42deltasquared: just a curiosity about the lack of power management control on nvidia cards, do they tend to be locked to their *lowest* frequencies as a result of this? mostly hypothetical, but wondering if a cheap GT1030 would be a decent "spare" card for an otherwise headless system only used for CPU tasks.
14:43deltasquared: if they were locked to lowest frequencies that'd actually suit the above use case just fine, heh
14:43imirkin: deltasquared: why do you need a spare GPU for a headless system that only does CPU tasks?
14:43deltasquared: imirkin: UEFI menus.
14:43imirkin: ah, no serial?
14:43deltasquared: I was hoping to use threadripper, so likely no... :P
14:44imirkin: anyways, the UEFI menus would be run by the GOP which is on the ROM of the GPU, so nouveau wouldn't be involved
14:44imirkin: i'm on a GT1030 right now, as it happens
14:44imirkin: using it with 2x 1920x1200 screens, but an otherwise "old" setup -- no new shit
14:44deltasquared: I figured the UEFI part would be fine. it'd be nice to have at least some extremely basic acceleration if I ever did need to open a UI on the thing if things broke though.
14:44deltasquared: I mean, I'm sure it'd handle fbcon just fine for starters ;)
14:45imirkin: yeah, there's basic acceleration
14:45deltasquared: I just didn't want it sitting there using more power than necessary I guess.
14:45karolherbst: deltasquared: well... mesa and nouveau _could_ be ported over to UEFI, but I guess nobody wants to do that
14:45karolherbst: deltasquared: but not being able to do PM also means you waste power
14:45imirkin: GT1030 should be more than sufficient for a UEFI GPU
14:45karolherbst: even on the lowest perf level
14:46karolherbst: it's more than just setting clocks or voltages
14:46deltasquared: karolherbst: right, because they can't go "idle"?
14:46karolherbst: more or less
14:46imirkin: you can get fanless ones, i think
14:46karolherbst: turing has an absurd low idle power consumption
14:46deltasquared: imirkin: it'd still be sitting there acting as a small heating element. :P
14:46karolherbst: even 250W can draw like 2W when fully idle
14:46imirkin: deltasquared: absolutely.
14:46karolherbst: but not with nouveau
14:47imirkin: you could remove it after UEFI is configured.
14:47deltasquared: imirkin: that's my fallback plan, yes.
14:47imirkin: that's the lowest amount of power it will draw -- when it's not plugged in :)
14:49deltasquared: I have to admit though, "no idling available" and generally no power management aside, I did take a peek at the feature matrix. it was more complete than I was expecting to be honest
14:49deltasquared: not a criticism or anything! just pleasant surprise
14:49imirkin: almost everything sorta-works
14:49imirkin: GT1030 won't have video decoding accel -- too new
14:50deltasquared: yeah, I spotted that
14:50imirkin: and a handful of the more advanced nvidia-specific GL extensions aren't implemented
14:50imirkin: and no vulkan
14:51deltasquared: just a thought, I noticed the older generations required nvidia's firmware. if that has their signature, could you theoretically not talk to that to get power management? (inb4 moar reverse engineering)
14:52deltasquared: I'm honestly considering getting one just to experiment. wouldn't have much other use for it atm lol
14:52imirkin: newer generations require nvidia's firmware
14:52imirkin: for graphics engine context switching and other stuff
14:52imirkin: the firmware which they've made available in linux-firmware does not support these things
14:53deltasquared: oh, so that's the distinction with the bits marked EXTFW
14:53imirkin: getting their driver's firmware is technically complicated (although i guess not anymore), and would not be redistributable anyways
14:53imirkin: EXTFW is for video decoding acceleration
14:57deltasquared: huh, considering how to help with the dev efforts... now I realise one reason why nvidia are douches with running their windows drivers in a VM. it would hamper stuff like this project spying on it.
15:17karolherbst: deltasquared: nope, it's mainly because of money
15:17karolherbst: need to buy expensive quadro/grid cards to do that :p
15:17deltasquared: that too
15:17karolherbst: well, that's their main reason :p
15:18deltasquared: "we see you're trying to do a professional usecase on our consumer cards, why not pay $$$$ for an upgrade?"
15:18karolherbst: if money is not included, it's not their reason :D
15:18deltasquared: it does make me curious what that "announcement" nvidia was gonna make was about before the event got cancelled due to a certain human malware.
15:18karolherbst: uhm... can't talk about it, but I know
15:19deltasquared: can't even say if it's a Good Thing™?
15:19deltasquared: I'm probably digging my own grave now lol
15:20deltasquared: augh, now I'm itching to now, celestia damnit I shouldn't have asked xD
15:20ericonr: curiosity did kill the cat
15:21deltasquared: ok, here's a question that shouldn't be in violation of anything NDA-like, just possibly something I've missed... did nvidia make any alternate arrangments for when that info was going public?
15:21deltasquared: like, did they say *publicly*
15:21deltasquared: publicly when $secret was going public... words are hard
15:22deltasquared: just because it's quite possible I've been living under a rock about such things for the past few months
15:22karolherbst: deltasquared: dunno, but I could ask
15:22deltasquared: well, somebody has connections. :P
15:26deltasquared: ngl, the possibility (because you haven't really said as such ;) ) that such a connection between the nouveau project and nvidia could even exist is quite frankly amazing, given the latter's track record to date. dare I say it makes me feel optimistically hopeful? :>
15:26deltasquared: ok, I'll stop prying now xD
15:26karolherbst: well.. comes with the job :p more or less
15:27karolherbst: just annoying not being able to talk about some of it :/
15:27deltasquared: an annoyance I feel on the other side of the fence, I can assure you
15:30deltasquared: I'd say something about wondering what nvidia cards would be like on (improved, not to dismiss nouveau's efforts though) open drivers, but again nothing of the sort has been said.
15:32deltasquared: augh, I'm feeling the annoyance of the situation and I'm not the one bound to silence here! :P
15:32deltasquared: someone ought to buy you a beer at least when it does come to light :P
15:33deltasquared:makes a mental note
15:35karolherbst: let's say the stuff getting announced would be to some degree helpful :p
15:35imirkin: what's the question? when will this stuff be released?
15:36imirkin: i assume never, despite insinuations to the contrary
15:36deltasquared: karolherbst: which is successfully giving nothing away, because I would imagine "helpful" to some capacity is almost certainly implied. well done ;)
15:36deltasquared: imirkin: I would assume and expect never, but would be cautiously hopeful. that way if anything good happens it's a pleasant surprise, and if not we can all continue like nothing happened.
15:44deltasquared: karolherbst: anyway, whatever you can enquire about without getting in trouble for discussing it with others not bound by $nda_or_other_agreement ... I'm sure a lot of people out there would be interested, not just me.
15:47deltasquared: right, now I'm definitely done prying, I think I've used up my allowable budget for curiosity on the matter for a while
20:43TimurTabi: I'm working on adding a "get version" RPC with a fixed unit/function number. If it gets approved, I'll update the driver patch.
20:43TimurTabi: oops, wrong window