01:45blipk: any info on getting Pascall cards to display 4K over displayport?
01:45imirkin: should Just Work (tm)
01:45blipk: get 'configure crtc 0 failed'
01:45blipk: no outputs in xrandr
01:45imirkin: are you interested in working out what's going on?
01:45imirkin: pastebin dmesg and xorg logs
01:46blipk: okay, well I've installed the nvidia drivers now
01:46blipk: I need to reinstall OS again so I'll do it then
01:46imirkin: or just keep using nvidia and remember this experience when deciding which vendor's hw to purchase next time
01:50blipk: I'll probably be stuck with this card for a while
01:50blipk: so I'd like to get it working with nouveau
01:50imirkin: ok, well we can debug it then. you'll need to have it running in order to figure out what's up
01:50imirkin: make sure you're using an up-to-date kernel
01:50imirkin: 4.19 is good.
01:56blipk: I'm on 4.4
01:57imirkin: that kernel came out in ... many years before pascal did
01:57imirkin: so no way that's going to work
01:57imirkin: you need something released in 2018 :)
01:58imirkin: 4.4 came out Jan 2016
02:01blipk: 4.4 > 4.19
02:01blipk: or maybe not..
02:02blipk: so why are 2018 distro releases still using 4.4?
02:02blipk: is there a reason?
02:07imirkin: apparently as things get older, they become more stable
02:07imirkin: like cheese and wine, distros like to age kernels
02:08imirkin: [but then they realize they're missing out on useful stuff, so they "backport" logic from newer kernels, and create these frankenkernel monstrosities...]
02:11blipk: makes sense
02:11blipk: how can I disable the nvidia driver and reboot into nouveau?
02:13imirkin: depends on distro -- should ask in a distro-specific support chan
02:16blipk: right, thanks
02:16blipk: think I'm just going to switch distro
02:58HdkR: imirkin: There is only so long you can wait before cheese gets moldy :P
10:24RSpliet: HdkR: the French love their cheese mouldy. I imagine the same is true for your average Debian or RHEL user.
10:25HdkR: lol, I guess
10:27karolherbst: RSpliet: allthough to be fair, the rhel kernel contains usually recent drm ;)
10:27karolherbst: it's RHs pain to do that, true
10:28RSpliet: by which you mean, Ben and your pain :-P
10:28karolherbst: you'd be surprised
10:30karolherbst: anyway, afaik nothing gets backported for debian
10:30karolherbst: which is painful for users
14:22joepublic: Debian has kernel "4.18+98~bpo9+1" in backports if you don't mind getting a whole new kernel with your changes
16:14jorp: is it possible the a bios update to a desktop could stop nouveau drivers from working properly?
17:04Scotty_Trees: i doubt it jorp
17:33bbear: hello everyone
17:34bbear: I filled a length bug report on bugzilla.kernel.org : https://bugzilla.kernel.org/show_bug.cgi?id=201867
17:34bbear: If anybody want to investigate with more details, I'm ready to help in live.
17:34bbear: I don't know if this class of bugs is well known to you, nouveau developers, as this seems tied to ACPI problems with the discrete GPU.
17:35bbear: Also, it seems to be a recurrent problem for high end GPU it seems.
17:35bbear: (and recent ones)
17:35karolherbst: bbear: I am sure that nouveau.runpm=0 fixes your issues
17:35bbear: what is it doing ?
17:35karolherbst: disable suspending the GPU
17:36karolherbst: the problem is that we need to do something in order to not mess it up with recent hardware
17:36karolherbst: thing is, neither does the nvidia driver
17:36bbear: yes but will the fan continue to rotate at full speed 100% of the time ?
17:36karolherbst: it depends on your EC firmware
17:36karolherbst: if it thinks it is a good idea to do that if the GPU is on, then it won't change that
17:37bbear: karolherbst: so basically you suggest doing modprobe nouveau runpm=0 in the command line ?
17:37bbear: or having that in modprobe.conf.d maybe
17:38karolherbst: which will at least let you use your GPU
17:39karolherbst: it's a long outstanding issue and I try to get proper documentation on this to fix it relibly
17:39karolherbst: I have some more or less working fixes, but it wasn't good enough
17:48bbear: karolherbst: sorry I just rebooted.
17:49bbear: I did modprobe nouveau modeset=1 runpm=0. Power consumption is now ~35W instead of ~16W.
17:50karolherbst: yeah, that's the unfortunate part
17:50bbear: There is other noise source around so I can't tell if the fan really goes wilder than without the driver.
17:51bbear: but that fixes that "lspci freezes your laptop" thing
17:52karolherbst: and you should be able to use the GPU, for GL and external displays and stuff
17:52bbear: yes, so far it is good, the only thing that annoy me is the fan setting
17:52bbear: I wonder, can you do runpm=1 and use an external display nevertheless ?
17:53bbear: can you even do runpm=1 and modeset=1 ?
17:53karolherbst: well, the gPU would be on anyway then
17:53karolherbst: so doens't really matter if you enable runpm or not
17:53karolherbst: runpm matters if you don't use the gPU
17:53karolherbst: but you should have modeset enabled
17:54bbear: so you are telling me that basically even if I use the nvidia GPU as a sink for the integrated GPU, the fan will rotate full speed ?
17:54karolherbst: or did you add a modeset=0 somewhere?
17:54bbear: what would be the use of doing modeset=0
17:54karolherbst: bbear: well, if your EC firmware things it's smart to rotate the fans at full speed if the GPU is on there isn't anything we can do about that
17:54karolherbst: even with runpm enabled that would be the case then
17:55karolherbst: we could do some stuff to lower power consumption though
17:55karolherbst: Lyude was/is working on that
17:55karolherbst: but sadly not for pascal
17:55karolherbst: we have no idea if we are able to do anything there without getting a signed firmware image from nvidia for the PMU
17:55karolherbst: the situation is becoming unfortunate unfortunately
17:56bbear: karolherbst: you can't extract it from the GPU ?
17:56Lyude: (help is welcome btw)
17:56karolherbst: would have to be done from the binary driver
17:56HdkR: I'd say the situation is dire
17:56karolherbst: and we can
17:56karolherbst: thing is
17:56karolherbst: the API is unstable
17:56karolherbst: and we will need to port to a newer API basically every time a new chipset is out there or gets updated
17:56karolherbst: because things can always break
17:56bbear: so how does the proprietary driver ?
17:57karolherbst: well, they have more developers
17:57bbear: they can't maintain a firmware in the driver for each card issued, could they ?
17:57karolherbst: if they update their PMU image, they just update the uses
17:57karolherbst: bbear: :D sure
17:57karolherbst: they even change it every release
17:57karolherbst: well chipset
17:57karolherbst: not card
17:58karolherbst: but newer cards, even with already known chipsets, could require updated firmwares
17:58karolherbst: that already happened with the released ones we got
17:58karolherbst: it would be an endless catching up game
17:58karolherbst: we are working (tm) with nvidia on having a proper solution for that though
17:58karolherbst: just.... it takes more time than anticipated
18:00karolherbst: Lyude: do you know if clockgating is secured on pascal btw?
18:00karolherbst: I would expect it is
18:00karolherbst: but... you never know
18:00karolherbst: bbear: if the osi thing works for you, you could use that instead of runpm=0
18:01loonycyborg: Is there any chance you could make fan control work with nouveau too? :P
18:01karolherbst: it just forces the firmware to use an older way of turning the GPU off
18:01karolherbst: which doesn't trigger the issues
18:01karolherbst: loonycyborg: well, you need signed firmware for that
18:01karolherbst: but on the laptop it's not controlled by the GPU
18:01karolherbst: but the EC
18:02loonycyborg: well I got a desktop videocard without fan too
18:02loonycyborg: but it's really cheap one
18:03loonycyborg: does nvidia plan to share a to control fan somehow in the future? or mounting a cryptographic attack is the only way?
18:04karolherbst: well, that's the "we are working on it" part
18:04karolherbst: but if somebody finds a way to execute random code, so be it
18:04karolherbst: we can execute random code, that's not the issue
18:04karolherbst: but we aren't able to access everything
18:07bbear: why are they verifying their microcode update
18:07bbear: is this only to prevent reverse engineering attempts or is there a real technical interest in doing so ?
18:08bbear: what is the EC firmware ? energy control firmware ?
18:08karolherbst: embedded controller
18:08karolherbst: it's part of your UEFI
18:08karolherbst: or rather ACPI
18:09bbear: I think the temp is nouveau-pci-0100
18:09karolherbst: shouldn't be higher than 50 on idle
18:09bbear: when I type `sensors` it tells me 46°C
18:09bbear: that's a high for an idling GPU
18:09karolherbst: it's a laptop
18:09karolherbst: 90 is high
18:10bbear: I mean, it's high with the fan full speed
18:10karolherbst: 100 is where you should start to worry
18:10bbear: I mean I really don't have any need for this power
18:10karolherbst: you want to use external displays, no?
18:10bbear: I'm sorry I bought this laptop
18:10karolherbst: some laptops use the integrated GPU for external displlays
18:10karolherbst: they are smart
18:10bbear: If I could disable the fan while using the secondary display I would have a run for my money but now...
18:11karolherbst: maybe you can change something in the firmware?
18:11karolherbst: bbear: by any chance, is that a clevo based on?
18:11bbear: it is
18:11karolherbst: yeah... figured as much
18:11karolherbst: those are stupid in that regard
18:11karolherbst: sometimes the fan spins down after a while
18:11bbear: clevo P955ER
18:11karolherbst: but the EC firmware is super jumpy
18:11karolherbst: and thinks that 60 is too hot
18:12karolherbst: even for the CPU
18:12karolherbst: it's just stupid
18:12karolherbst: sometimes you find modded firmwares around which give you more control over all that, but... you could also brick your system
18:13bbear: karolherbst: I wouldn't know where to start to change temperature trigger. The BIOS is even more basic than a 15 years old PC.
18:13karolherbst: the EC has some regs to do manual fan control
18:13bbear: I was wondering if some fancy UEFI interface could unlock the thing if I installed windows or if I had a disk reader.
18:13karolherbst: there is some kernel module for that somewhere around
18:13karolherbst: your model might not be supported though
18:14karolherbst: bbear: there is a clevo_xsm_wmi thing, but I don't know if it allows you to control the fans
18:14karolherbst: but you should be able to find something somewhere
18:15karolherbst: it's all a bit whacky though
18:15karolherbst: and you might upset the EC
18:18bbear: life's not easy.
18:18karolherbst: nope, it ain't
18:20bbear: basically the proprietary driver doesn't give me any advantage over nouveau here, given the fact that I would like just to decrease fan speed and use the dGPU for coding in dual screen mode ?
18:20bbear: I think the answer is no, I tried almost everything I could and it was really not working.
18:21bbear: Maybe I'll install Windows and see if that makes a difference. They probably will have tools to setup backlight in the keyboard and that kind of things.
18:21karolherbst: not really. the prop driver should be able to reduce power consumption and there are some hacky ways to use the external displays
18:22karolherbst: there is some intel virtual display thing which could be used
18:22karolherbst: but you wouldn't be able to power off the GPU as well
18:22bbear: karolherbst: yes, so basically what I am looking for is a drop in power consumption. Do you think the nouveau driver will have that soon ?
18:23bbear: (I'm not even sure, 20W for this GPU seems the lowest consumption it reaches.
18:23karolherbst: soon? most likely not
18:23bbear: is there possibilities that nvidia will opensource the proprietary driver ?
18:23bbear: how much money should be thrown at them ?
18:24bbear: I would like to help, not sure if I can gain the knowledge needed for that though.
18:25HdkR: There are much better things than open sourcing that nightmare that could be done
18:27karolherbst: it wouldn't be accepted in mainline anyway
18:27karolherbst: there are some bits open source in the nvidia module though
18:27joepublic: "The kernel hacker's guide to making hacks look respectable and merge-able"
18:27karolherbst: so there is at least hope
18:27karolherbst: joepublic: ;)
18:28karolherbst: dev spending more time making hacks mergable than writing those hacks would give you the impression that those hacks aren't hacks anymore :p
18:28bbear: Signal processing controller: Intel Corporation Cannon Lake PCH Thermal Controller <- this is the EC you were talking about ?
18:28karolherbst: EC is an ACPI thing
18:29karolherbst: it doesn't really show up as a device
18:29bbear: I have this: pch_cannonlake-virtual-0, temp1 +64°C
18:29karolherbst: right, but that's not the ec
18:29karolherbst: the EC is a firmware thing
18:29karolherbst: there are some ACPI methods
18:29karolherbst: and you can map the EC memory and read/write it
18:29karolherbst: but that's pretty low level
18:30bbear: I could fire my laptop ?
18:30karolherbst: pcie devices are more high level than ECs :p
18:31bbear: by the way I noticed that, using nouveau and runpm=0, the ACPI interface temperature was 511°C
18:31karolherbst: which is pretty much -1
18:31bbear: ha ok
18:31karolherbst: hex(511) == 0x1ff
18:31karolherbst: 9 bit field, weird
18:32karolherbst: but why is it 511...
18:32karolherbst: I don't get those crappy clevo firmwares anyway, I think they are just broken, clevo knows it, but they just won't fix it
18:33karolherbst: clevos are the ones with the most firmware issues, honestly. Lyude would disagree, but that's just because Lyude doesn't know better :p
18:39Lyude: i mean, i've seen some /real/ bad machines
18:39karolherbst: Lyude: clevo ones?
18:40Lyude: karolherbst: no I mean in general
18:40karolherbst: ahh, right
18:40karolherbst: Clevos are the worst though
18:40Lyude: UEFIs that just randomly brick themselves are a thing
18:40karolherbst: issues on literally every machine
18:40karolherbst: :D crap
19:33bbear: I dunnow if some people can be interested but here there is some specs about the Hardware plugs for this model : http://repo.palkeo.com/clevo-mirror/P95XEP6_ER/P955ER_SM_20180522.zip
19:35karolherbst: bbear: yeah.. those are always around somewhere for the clevos
19:36karolherbst: thing is, those aren't that useful
19:36karolherbst: nice if you want to do hardware modding, but that's basically it
20:34bbear: I found a key that set up fan control mode (max, adaptative and slow)
20:35bbear: would you be interested if capturing those events maybe ?
20:35bbear: IDK how I would do it though, but it seems to make my PC quiet a lot more.
20:35karolherbst: bbear: ohh, right, those magic Fn + F1 key?
20:35karolherbst: might be F2
20:35karolherbst: afaik those are EC internal
20:35bbear: it was Fn + 1
20:35karolherbst: the OS doesn't know you even hit it
20:35bbear: There is this so called Q-mode
20:36bbear: karolherbst: no way to capture then ?
20:36karolherbst: the keyboard is directly wired through the EC
20:36karolherbst: you should see it inside the schematics
20:36karolherbst: but that also means it doesn't work on external keyboards
20:36karolherbst: _maybe_ there is some ACPI notification we could listen on
20:36karolherbst: but that's all way beyond me anyway
20:37karolherbst: bbear: which mode is it on by default? adaptive?
20:37bbear: what's approximately cool with this CLEVO laptop is that you can unscrew everything.
20:37karolherbst: older models only had fast/adatpvie afaik
20:37bbear: karolherbst: no I think it is agressive
20:37karolherbst: yeah, the clevo ones are pretty hackable
20:38karolherbst: my first clevo had 3 HDDs + 1 SSD in the end :D
20:38bbear: wait i'm selling you crap
20:38bbear: actually there is a maximum mode, an adaptative mode and a max-Q mode
20:39karolherbst: it should be safe to use the slowest mode
20:39bbear: max-Q must be related to power management of the card.
20:39karolherbst: we can't put the GPU in higher perf states anyhow
20:39karolherbst: so the GPU won't generate that much heat to begin with
20:39bbear: so that would be max-Q ?
20:39karolherbst: maybe Q is for quiet
20:39karolherbst: and tries to be as queit as possible?
20:39karolherbst: no idea
20:39karolherbst: bbear: what gets exposed through sensors through nouveau?
20:39karolherbst: only temp?
20:40karolherbst: or more?
20:40bbear: only temp
20:40bbear: but then in max mode the power consumption is 50W
20:40bbear: I doubt the fan can takes 12W by himself
20:40karolherbst: ? the fan itself draws 12W?
20:41karolherbst: probably not
20:41bbear: but the first mode is ok
20:41karolherbst: maybe some weirdo power capping going on
20:41karolherbst: not the first time I saw that
20:54bbear: that's the lshw part related to the motherboard : https://paste.ubuntu.com/p/crGStCHH8r/