11:34bk1603[m]: Are the tasks tagged `[Juior]` on the OldTodo page still relevant? And are they the best way for new contributors to start contributing to nouveau?
11:36karolherbst: bk1603[m]: it seems some or still valid, just not sure how useful. Normally I suggest people to work on stuff they hit themselves or find very annoying
11:36karolherbst: or features/bugs they care about
11:36karolherbst: vram testing sounds interesting actually..
11:44bk1603[m]: I haven't hit any bug on my use of nouveau so far. The only thing that I find missing is the fact that in the proprietary drivers I could entirely turn off the external gpu and only use the intel card. (Did wonders for backup life, I was hitting close to 15 hours back then.)
11:44bk1603[m]: Is that possible on nouveau?
11:44bk1603[m]: (My bios doesn't allow me to do that.)
11:45bk1603[m]: * I haven't hit any bug on my use of nouveau so far. The only thing that I find missing is the fact that in the proprietary drivers I could entirely turn off the external gpu and only use the intel card. (Did wonders for battery life, I was hitting close to 15 hours back then.)
11:45karolherbst: bk1603[m]: there is no reason why you'd want to do that
11:45karolherbst: nouveau will turn of the GPU if not used
11:45karolherbst: regardless of what is configured
11:45karolherbst: or normally it should
11:46karolherbst: if it's not the case maybe something is wrong
11:46bk1603[m]: Oh I see, and rendering X via nouveau would depend on my configurations right? Maybe that's why this is happening?
11:46karolherbst: with "grep . /sys/bus/pci/devices/*/power/runtime_status" you can check the runtime pm status of all pci devices
11:46karolherbst: and the GPU (and all subdevices) should be "suspended"
11:46karolherbst: the pci controller the gpu is on as well
11:47karolherbst: you can sheck with "lspci -tvv" where the GPU is located in the tree
11:47karolherbst: for me it's "+-01.0-----00.0 NVIDIA Corporation GP107M [GeForce GTX 1050 Mobile]"
11:48karolherbst: which means, it's on 0000:00.01.0 and this controller controls the 0000:01 bus
11:48karolherbst: bk1603[m]: by default it should use the integrated GPU
11:48karolherbst: so yeah, maybe your xorg config is screwed up and the best idea is to just remove it
11:48karolherbst: just leave the distribution specific files in place
11:49karolherbst: you can check users of the GPUs with "grep . /sys/kernel/debug/dri/*/clients"
11:49karolherbst: _usually_ 0/128 means integrated and 1/129 discrete
11:50karolherbst: but on some system that stuff could be different
11:50karolherbst: you can check "grep . /sys/kernel/debug/dri/*/name" to see what device it corresponds to
11:50bk1603[m]: I see, I'll check them right away :)
11:54bk1603[m]: Oh ha! It is suspended, wonder what's causing the battery drain then? Anyways thanks a lot for clearing this out!
11:55karolherbst: bk1603[m]: well, you should check if the audio device is also suspended
11:55bk1603[m]: Oh let me take a look
11:55karolherbst: "cat /sys/kernel/debug/vgaswitcheroo/switch" is usually a quick way of checking that as well
11:55karolherbst: less verbose
11:55karolherbst: bk1603[m]: what GPU do you have btw? but it does sound like something new if you get 15 hours
11:57bk1603[m]: I'm using gtx 1650 ti mobile version. With the graphics card I can get anywhere between 6~8 hours, and without it I practically only charged it once ever 2~3 days. That was when I was on ubuntu with the proprietary drivers and prime switch installed. But then I moved to nixOS. And now the battery life I am getting is not as spectacular
11:58bk1603[m]: Not that it's unusable, but now that I know I could get as much, it makes me greedy :P
11:59karolherbst: and you should get above 10 hours easily
11:59karolherbst: I just suspect something is messed up
12:00karolherbst: either because of missing ACPI support or because there are sub devices not suspending, etc...
12:00karolherbst: the audio one is something common some distributions just mess up
12:00karolherbst: you might even see a USB and USCI controller on the GPU depending on the laptop
12:02bk1603[m]: I'll check the sub devices right away. But I don't think it's audio cause I haven't had audio working on this laptop ever since I installed ubuntu. (It needs the newer `sof-soundwire` and `alsa-ucm-conf` and haven't really gotten around to setting them up.)
12:03karolherbst: why would you need to do something special for the audio
12:03karolherbst: it should just work
12:03karolherbst: if not, your distribution is broken
12:06bk1603[m]: Umm, the thing is the newer intel audio drivers weren't supported upto kernel 5.4, they were only recently added. Even ubuntu doesn't have the audio working out of box on my laptop
12:06bk1603[m]: I think there was a linux-oem release but that didn't exactly work. I was able to get the audio working on arch once, by compiling everything from source, and getting the bleeding edge drivers
12:07bk1603[m]: Compiling everything requires kernel 5.7, I have upgraded to that, but I just haven't gotten around to compiling them again, partly because I forgot one or two components
12:09bk1603[m]: There are lots of issues with new hardware. I had to choose between kernel 5.4 (which supported deep sleep) and 5.7 (which had working sound) on ubuntu, and I went with 5.4 cause I mostly use earphones anyways.
12:09karolherbst: ahh, well..
12:09karolherbst: some distributions are annoying and keep using older kernels
12:10karolherbst: but ubuntu has upstream kernel builds, they are just painful to use (on purpose)
12:10karolherbst: but yeah...
12:10bk1603[m]: haha yep that's one way of going about it.
12:10karolherbst: I suspect without supported audio, things might not work as well?
12:10karolherbst: guess the packaging could have cheated a little
12:12karolherbst: bk1603[m]: I am not quite sure how special the GPUs audio device is and if a newer driver for that would even be required, but that could lead to issues
12:12karolherbst: but I suspect more the USB/UCSI if you have those
12:12karolherbst: anyway.. on those newer machines I think 5.4 is just too old and in newer releases there were patches to fix it all up
12:13bk1603[m]: Hmm, I think I'll get around to fixing my audio probably next weekend. I can then verify if that's what was causing it.
12:13bk1603[m]: Yep that makes sense
12:13bk1603[m]: Also as for USB/UCSI, I could only find USB, and that isn't active
12:14karolherbst: I don't know how the UCSI stuff is named
12:14karolherbst: but I suspect you have three other devices on the GPU?
12:14karolherbst: or just two?
12:14karolherbst: and the audio one is on active?
12:15bk1603[m]: Nope, no devices under the GPU node
12:15karolherbst: if there are no devices, then it should just work
12:15bk1603[m]: There is only the GPU and nothing else as a child to that node.
12:16karolherbst: well, those subdevices are not realy children of the GPU
12:16karolherbst: they are all directly under the pcie controller, but they share the same PCIe bus
12:16bk1603[m]: Oh I see
12:16karolherbst: like if the GPU is 01:00.0, the audio device is usually 01:00.1
12:16karolherbst: and the USB/UCSI ones are 01:00.2 and 01:00.3
12:17karolherbst: and then for me the pcie controller is at 00:01.0
12:17karolherbst: and all of those need to be suspended
12:18bk1603[m]:sent a long message: < https://matrix.org/_matrix/media/r0/download/matrix.org/lTvtPvPwVMCyvDmwcGTFtqdK/message.txt >
12:18bk1603[m]: There are no other devices that start with 01
12:18bk1603[m]: the gpu has 01:00.0
12:18karolherbst: ohh indeed
12:18karolherbst: it's just the GPU
12:19karolherbst: can you check if 00:01.0 is suspended?
12:19bk1603[m]: Yep that's suspended
12:20karolherbst: I am inclined to say it's not the GPU causing the high power consumption then...
12:20karolherbst: maybe "dmesg" shows something, but normally at this point using the nvidia driver doesn't make a difference really
12:20karolherbst: unless there is some ACPI stuff going on
12:21karolherbst: also checking with powertop or turbostat might make sense to see if the CPU is entering deeper C and PC states
12:21bk1603[m]: Hmm, weird, also `powertop` reports a power consumption of 6+ W for now
12:21bk1603[m]: Which is fine
12:21bk1603[m]: But sometimes it goes 10+, and doesn't list any processes that could be causing it
12:21karolherbst: yeah.. that sounds alright
12:22karolherbst: probably some daemon or so
12:22bk1603[m]: Which led me to believe that it must be the gpu, the kernel module just isn't loaded and so powertop can't identify it or something
12:22karolherbst: does something for a short time and then goes back
12:22bk1603[m]: Hmm now that you mention it, that could be the case too
12:22karolherbst: sometimes using the wrong desktop can lead to this
12:22karolherbst: some desktops are just terrible if it comes to power consumption
12:22bk1603[m]: (I remembered this happened when I was using the proprietary drivers, and the solution was to switch the gpu off using bbswitch)
12:23bk1603[m]: I'm only using dwm :P
12:23bk1603[m]: I don't think that would draw as much power or resources
12:23karolherbst: ohh, it's more the opposite
12:23bk1603[m]: Oh really? TIL
12:23karolherbst: they usually do things to reduce power consumption
12:23karolherbst: like a few years ago when I was checking, xfce4 caused higher power ocnsumption than let's say KDE
12:24karolherbst: bunch of toogles desktops have to flip to make it good
12:24bk1603[m]: Damn, that's a surprise. But then it makes sense too, since they have power management tools
12:24karolherbst: random stuff like HDD/SSD idling or diming the screen and stuff
12:25karolherbst: and if you are on OLED you want everything to be black. etc,,,
12:25bk1603[m]: Yeah I almost forgot that
12:25bk1603[m]: Also no OLED
12:26bk1603[m]: I'll find some power manager for the laptop then.
12:26karolherbst: yeah,, stuff like tuned,tlp or laptop-mode-tools usually help a bit
12:26bk1603[m]: Also umm a few links on the `IntroductoryCourse` webpage are broken. The one that I really wanted to read was the nvidia hardware introduction
12:26bk1603[m]: but that leads to a 404 on github
12:27bk1603[m]: pathscape have moved and I couldn't find the wiki on the new repo
12:27bk1603[m]: any alternatives?
12:27karolherbst: let's see...
12:27karolherbst: yeah.. external links
12:27karolherbst: some webpages just don't care
12:28bk1603[m]: yeah that's true :]
12:28karolherbst: we just recently moved the wiki over to gitlab pages and I wanted to make sure no links break :D
12:29karolherbst: but also our wiki needs a lot of work to get cleaned up and stuff :/
12:29bk1603[m]: That's awesome! Would that mean we'd have alternative links to read up the stuff that we can't for now?
12:30karolherbst: well, worst case somebody would have to write the text first
12:30bk1603[m]: Oh yep the wiki is a bit old, and the blogs that it refers to are even older. Almost all the links from the blogs are dead now.
12:30karolherbst: but maybe there are other websites giving good explenations
12:30karolherbst: at least now it's on gitlab and random people can actually change things :D
12:30karolherbst: that wasn't possible before
12:31karolherbst: just hit edit and do the gitlab stuff
12:31bk1603[m]: xD I'd love to help, but I don't know as much about the nvidia hardware to write an introduction about it. In fact a lot of what I know about graphics is from the Introductory course wiki
12:32bk1603[m]: I can help with the cleanup probably, and help with removing the dead links. That shouldn't take long. But we'd still be missing some awesome articles
12:33bk1603[m]: (that legit gave me a laugh lol xD)
12:34karolherbst: there is one thing though
12:34karolherbst: I'd like to have a list of all nvidia whitepapers
12:34karolherbst: they usually publish stuff for each generation of hardware
12:34karolherbst: like that one
12:35karolherbst: those usually have some detailed explenation on some things
12:35karolherbst: and they come from nvidia, so they usually just leave things out on pupose but don't throw in wrong information
12:35bk1603[m]: I see, I'll try and find as many of them as I can, and consider this as my beginner task for contributing to nouveau :P
12:36karolherbst: that would be awesome
12:36bk1603[m]: Also just to get a rough idea, what's the oldest architecture we want them for?
12:36bk1603[m]: Or is it as many as I can find?
12:36karolherbst: all of them :p
12:37karolherbst: would be cool to begin with the oldest and see how the GPU developed
12:37karolherbst: but not sure how long nvidia keeps those pdfs
12:38bk1603[m]: hehe okay, I'll try and find as many of them as I can. (I'll probably read a few too if I find enough time, they seem to be interesting, at least the one you linked is.)
12:38karolherbst: they usually are quite interesting :)
12:39bk1603[m]: Cool! Off I go then :D
15:38bk1603[m]: karolherbst: I've found everything till fermi, (maybe even tesla, one of the papers have both tesla and volta ig,) but finding anything older than that seems to be taking a lot of effort.
15:38bk1603[m]: I think I'll read a few of them before searching further
15:38karolherbst: cool :)
15:39karolherbst: feel free to create an MR to the wiki and link them.. or maybe just move the pdfs in? Not sure if distribution the pdfs would violate copyright law or not
15:39bk1603[m]: I'll send what I found for now, just a minute
15:43bk1603[m]:sent a long message: < https://matrix.org/_matrix/media/r0/download/matrix.org/aFqvJitfMILjQuOVTYZowWxh/message.txt >
15:43bk1603[m]: Oh I just saw your message
15:44bk1603[m]: Hmm, I think linking them would be safer. The download pages mention nothing about redistribution.
15:46bk1603[m]: The legal notice at the end too doesn't say anything in that regard.
15:49bk1603[m]: The fermi one is under creative commons
15:50karolherbst: thanks for the list though!
15:50karolherbst: I'd like to not have too many external links as those things just disappear with time
15:50karolherbst: but mhh
15:50karolherbst: I'll ask nvidia if there is anything they could do
15:51bk1603[m]: Yep that'll be the best way of going about it.
15:51bk1603[m]: Also you're welcome! :)
15:51bk1603[m]: And I'm gonna go read some of this stuff now then
15:52karolherbst: mhh, I think there is more actually..
15:52karolherbst: ampere: https://www.nvidia.com/content/dam/en-zz/Solutions/geforce/ampere/pdf/NVIDIA-ampere-GA102-GPU-Architecture-Whitepaper-V1.pdf
15:53bk1603[m]: Oh ha I must have missed it, yep that's the last one that I could find today. I think I missed it because I was creating a list from the freedesktop page for codenames. (https://wiki.freedesktop.org/nouveau/CodeNames/)
15:53bk1603[m]: And that doesn't contain ampere
15:53karolherbst: GTX200 stuff: https://www.nvidia.com/docs/IO/55506/GeForce_GTX_200_GPU_Technical_Brief.pdf
15:54bk1603[m]: Oh I couldn't find that. I guess searching by the codenames isn't the best practice then
15:54karolherbst: yeah... it's messy
15:55karolherbst: and there are multipe ones per chipset and stuff..
15:55bk1603[m]: I see, I'll look for more stuff then, and see if I can find any others.
15:55karolherbst: heh.. there is also this: https://research.nvidia.com/publications
15:55karolherbst: but that's more research stuff
15:55RSpliet: Oh they got a separate page for research. Nice
15:56bk1603[m]: I tried searching for things on that page, and I couldn't find anything so I just skipped it :P
15:57bk1603[m]: Particularly I searched for white paper and architecture names
15:57karolherbst: bk1603[m]: yeah.. well, I think we should keep track of the interesting and helpful bits somewhere though :) But yeah, I mean you want to mainly read them :p I think we want to have a link list, so maybe we should create a dedicated md page just with links for stuff with a short comment
15:57karolherbst: but this depublication thing bothers me...
15:57bk1603[m]: Oh well than I have another thing to share
15:58bk1603[m]: For volta
15:58karolherbst: right, but that's research
15:58karolherbst: but it references some docs
15:58karolherbst: which is nice
15:59bk1603[m]: I'm sorry but I couldn't understand that, can we not link to research papers? Or is it not very credible?
15:59karolherbst: bk1603[m]: the thing is, we don't know if it's credible or not
15:59karolherbst: so there are two things: 1. what nvidia provides 2. what others found out
16:00karolherbst: 1. is usually trustworthy, but they keep some things secret
16:00karolherbst: 2. is useful for the things nvidia doesn't talk about or where they contradict nvidia while having proof
16:00karolherbst: ahh.. that gv100 thing is quite useless :D
16:00karolherbst: the nvidia one I mean
16:01karolherbst: but you already found the volta whitepaper
16:01bk1603[m]: Oh I see, so we are looking for as many things as we can find from nvidia directly
16:01karolherbst: yeah, mainly that
16:02karolherbst: it's also mainly to get a rough overview
16:02karolherbst: and the nvidia ones are quite nice for that
16:02karolherbst: although they also contain a bit of marketing and pr bits..
16:02RSpliet: Research papers are deliberately as thin on the details as they can be to explain necessary concepts
16:02RSpliet: Often oversimplified
16:02karolherbst: or talk about stuff they do in their software stack and not in hardware directly
16:02RSpliet: But sometimes very revealing
16:02bk1603[m]: Ok, I'll try and search by cards in various architectures too. Maybe I find something? (Like you did for 200 series)
16:02bk1603[m]: RSpliet: I see, I did not know that :)
16:03karolherbst: bk1603[m]: I know that there are some multiple ones for kepler.. let's see
16:03bk1603[m]: That and both maxwell and volta ones have tesla p100 and tesla 100 respectively in their headings
16:03karolherbst: or maybe not...
16:03karolherbst: nvidia focuses on their high end models most of the time
16:04bk1603[m]: Oh got it
16:06karolherbst: ehh screw it "filetype:pdf site:nvidia.com" :D maybe that shows something nice as well
16:08karolherbst: another fermi one: https://www.nvidia.com/content/PDF/fermi_white_papers/NVIDIA_Fermi_Compute_Architecture_Whitepaper.pdf
16:09bk1603[m]: Just added to the list. Umm there was a specific page where nvidia provided the links for the white papers, maybe webarchive would help with old papers?
16:14bk1603[m]: Aha just found one for curie. GeForce 6 specifically
16:16bk1603[m]: One question, if I already have a whitepaper on an architecture, do I also include to the list the whitepaper for specific cards under that architecture?
16:17bk1603[m]: I think I should probably. More information on the cards should equal more help, but I'm not super sure
16:17karolherbst: well, it doesn't hurt
16:24bk1603[m]: well so far I've found gtx 750 ti, gf 100 series and gtx 1080 after the fermi one that you sent
16:33bk1603[m]: Oh so the ampere series is the rtx 30 one, I see now why it wasn't on the codename page on freedesktop.
17:10bk1603[m]:sent a long message: < https://matrix.org/_matrix/media/r0/download/matrix.org/KReqWNxbweJGuXUBwwQtXolj/message.txt >
22:44joey: hello, i built a 5.8.3 kernel using only linus tree and option 'nouveau=m', journalctl says it loads, detects my card, and then a message it doesnt have a video bios. anything i overlooked? 'general' distro's like arch work fine
22:45joey: i onyl care about console, not x
22:51imirkin: then don't bother loading nouveau?
22:51imirkin: nouveau must have a video bios to work
22:52joey: yes but i'd imagine it does some magic and load it from somewhere, maybe even the one on the card itself..
22:52imirkin: and it can access it on the other distros
22:52imirkin: which leads me to believe you compiled your kernel funny
22:52imirkin: missing some required feature
22:52joey: yus is why i aske here, to see ifi overlookes somethuing obvious
22:53karolherbst: joey: some GPUs require ACPI support
22:53imirkin: but ultiamtely it doesn't matter - if you don't need graphics mode or accel
22:53karolherbst: make sure you have ACPI up and running
22:53imirkin: then just don't load nouveua at all, and you'll be much happier
22:53joey: i did a make defconfig and tweaked device drivers to include nouveau=m
22:53karolherbst: joey: do you have ACPI enabled?
22:54karolherbst: although that only matters for laptops
22:54joey: well i'm happy if my primary display shows my console without using nomodeset
22:54karolherbst: but yeah.. can't really help with broken kernel config :/ surprising that defconfig won't work though
22:54joey: im buildign a 'allmodconfig' kernel now, to see if i made a mistake
22:55joey: and then a localmodconfig should kill the crap ;)
22:56karolherbst: joey: you might want to ensure that efifb is set to yes
22:56karolherbst: not that you tracking something down which is unrelated to nouveau
22:56karolherbst: but yeah... that's all surprising
22:56karolherbst: it's a normal desktop, isn't it?
22:57joey: it is, with 3 monitors
22:57joey: guess i made a silly switch somewhere in the config.. will try the allmod and see how it goes
22:57karolherbst: mhh, would be weird though :/
22:58karolherbst: there isn't really much you can do wrong if it comes down to vbios loading
22:59karolherbst: joey: are you 100% sure it's about the vbios?
22:59karolherbst: newer GPUs require firmware files from the linux-firmware git (which is usually packaged in distributions)
23:00joey: i remember setting 'mark vga/efi as generic system framebuffer', maybe that wwas it
23:00karolherbst: so not sure about your set up, but there are a few things which can go wrong
23:00joey: as i didnt have any display at all
23:00karolherbst: kernel logs might help
23:00joey: not from i915 not from nvidia
23:00joey: kernel is recently built, i'll give it anotehr go ;)
23:07joey: "newer GPU's requiere firmware files", i think that could be it.. i've a GTX960... and didnt take any firmware whatsover yet.. thought it would use the onboard BIOS
23:14imirkin: need it for accel
23:14imirkin: and some kernels were broken in that they failed to fall back without firmware
23:15joey: well i'm building from absolute scratch (Linux from scratch) lol, but never thought of the firmwarw.. thought nouveau was "just" a driver that interacts with onboard bios
23:16joey: and yeah, in arch it's kinda the first thing you install apart from linux-base...
23:27joey: ok so unpacking to /lib/firmware should solve it :) THAT's what i overlooked haha
23:27karolherbst: but could be something else
23:27karolherbst: you'll see
23:27joey: yus ;) gonna boot it in 30 mins
23:27joey: fun times.. console only on your main PC xd