02:31rhyskidd: can i get a review on these missing SLCG regs for envytools rnndb? https://github.com/envytools/envytools/pull/103
09:34aphirst: Welp. I'm back again with another problem. Sometimes, when I open a video player, X dies booting me back to my login manager. I don't see anything in the Xorg.*.log or journalctl -xe outputs, and it's not /reliably/ reproducible.
09:34aphirst: I'm the one with a nvidia optimus thinkpad T430 using PRIME for dual head
09:34aphirst: who was here the other day talking about the pci magic to activate HDMI audio on the nvidia chip
09:51aphirst: nothing in dmesg either, just a comment that it reinitialised my ethernet connection
09:55aphirst: Of course, I rebooted again and can't reproduce the issue
10:00aphirst: But this is not a perfect world.
11:29karolherbst: aphirst: I expect that Xorg segfaults or something
11:30karolherbst: aphirst: you need to check the old X logs
11:30karolherbst: I think those are called Xorg.0.log.old or something
11:31karolherbst: aphirst: you cann add "Option "NoTrapSignals"" to the Xorg ServerFlags
11:35aphirst: I'm pretty sure there was nothing in the older logs in the folder
11:35aphirst: but I'll see if they're still there
11:38aphirst: hmm, i did some reboots since so it's now hard to see what relates to what
11:38aphirst: I think I'll just delete the old logs entirely (also for lightdm) then reboot and try to trigger the problem again
11:39karolherbst: well, Xorg just overwrites the old log
11:39karolherbst: maybe there is something inside your sys logger, but who knows
11:39aphirst: journalctl showed absolutely nothing suspicious
11:40karolherbst: yeah, you might need to enable that NoTrapSignals thing
11:40aphirst: since I don't have an xorg.conf by default, i just make one with that as the only Option?
11:41aphirst: in a ServerFlags section
11:41aphirst: or should it be a separate file in xorg.conf.d
11:41karolherbst: doesn'T really matter I think
11:44aphirst: would i need an Identifier line?
11:45aphirst: i've made a file 00-debug.conf in xorg.conf.d
11:46aphirst: ok, so i saved the file, restarted x, logged in, and the bug decided to manifest
11:46aphirst: i'm back at my login manager, time to see what got spat out and where
11:46karolherbst: there should be a trace inside the .old file
11:47aphirst: there's also sth in journalctl now regarding systemd-coredump
11:48aphirst: how's about i collect all this info together and pastebin/dropbox it
11:54aphirst: will add the xorg log itself in a second
11:55aphirst: ok, if you refresh that folder, the current and .old logs should be there
11:56aphirst: 01_..._txt is the excerpt from journalctl saying something went wrong
11:56aphirst: 02_..._txt is smething strange i noticed about the subsequent x session, it complains about the HDMI audio
11:56aphirst: i haven't tested it yet but obviously had to scroll past those messages to get to the part about X dying
11:57aphirst: Hum - just opened a really low res video and it's playing about 3 frames per second, with no sound
11:57aphirst: and the HDMI output is "unplugged" and "unavailable" in pavucontrol
11:58aphirst: could this be related to the hacky way I'm using a systemd service to run that pciset command on bootup?
11:58aphirst: race condition esque, if it does it at the wrong time the hw is borked?
12:08ntz: I'd like to ask before buying, is mobile 1050 supposed to work with nouveau ?
12:08ntz: I'm simply NOT going to use nvidia blob
12:11aphirst: seems the 1050 is in the NV130 family
12:12ntz: yeah, I've seen that but I've also read that it doesn't work
12:19imirkin: ntz: you really want to go with AMD if you're interested in open-source support (or intel)
12:20aphirst: i think nouveau is supposed to be OK for older nvidia stuff
12:20imirkin: it's difficult to know ahead of time, but if it's a GP107 then it could work, if it's a GP108 then you won't get acceleration
12:20aphirst: but if you 'care' about the newer nvidia hardware it's probably not what you wan
12:21imirkin: aphirst: i could name a dozen bugs for any particular generation. for some - many more.
12:21aphirst: well yes but I think I could name a handful of bugs for literally everytthing i use
12:21aphirst: In a perfect world, bugs like these would not exist.
12:21aphirst: But this is not a perfect world.
12:22imirkin: serious bugs.
12:22imirkin: anyways - it works well enough for me, and i do try to fix stuff up, but against AMD and Intel it's no contest
12:23imirkin: Intel has a team of probably 30 or more full-time engineers working on this, AMD has at least 3 but I think many more behind the scenes on some of the display stuff
12:23imirkin: both teams have access to internal documentation
12:24imirkin: nouveau has, in sum, probably 1.5 amount of full-time effort -- maybe less. no access to docs, and often hardware.
12:25ntz: to hear this
12:25ntz: 14:19 < imirkin> ntz: you really want to go with AMD if you're interested in open-source support (or intel)
12:25ntz: in #nouveaun channel is kinda surprising for me
12:25aphirst: it's just honesty tbh :P
12:26imirkin: i've been telling people to get amd/intel gpu's for a long time
12:28imirkin: aphirst: isn't your set up using intel as primary? if so any xorg issues are most likely connected to intel.
12:28aphirst: imirkin, it is but i had to juggle about to get the HDMI out to work
12:29aphirst: the HDMI is on the nvidia side
12:29aphirst: then ofc the audio too
12:29imirkin: right... but why does that matter?
12:29aphirst: btw imirkin do you have a chance to look over the logs/coredump i posted
12:29imirkin: or did you disable the IGP so that you get the audio controller?
12:30imirkin: no... you posted a dropbox link, i don't have time to deal with that right now
12:31aphirst: ah well I can post it elsewhere, it was because the coredum was lz4
12:31aphirst: and not just plaintext
12:31imirkin: right - so anything that's not just "click on link to see content" is something that "takes more time" :)
12:32aphirst: i'll repost the plaintext to a gist
12:33imirkin: aphirst: btw, if you're using grub, i think grub might have some ability to mess about with pci configs...
12:34aphirst: it occurs to me i booted this time into linux-ck rather than just plain linux; I can go triple check that the issue can happen in plain linux
12:34aphirst: imirkin, systemd-boot
12:34ntz: these are specs of laptop I'm considering to buy
12:34imirkin: aphirst: that's gummiboot right?
12:35ntz: there is intel, but iirc the hdmi port is wired with nvidia chip only
12:35imirkin: ntz: multi-gpu stuff tends to work poorly across the linux stack. but it does generally function. if you use intel as primary, you don't have to worry about acceleration on the nvidia.
12:36imirkin: ntz: unfortunately for reverse prime to work though, i think we do use "acceleration" to copy the data over...
12:36imirkin: not sure how the modesetting ddx deals with this
12:36aphirst: OK, same issue occurs, am extracting equivalent journalctl/coredump info now
12:37aphirst: on closer inspection it does appear that the issue is in intel_drv
12:38ntz: imirkin: there are literally NOT available laptops with just one chip anymore (if I don't count only intel gfx laptops)
12:38imirkin: that's what i'm suggesting though -- get an intel-only laptop
12:38ntz: intel gfx chip is now built-in all intel cpus
12:38ntz: imirkin: I want laptop with semi-strong gfx
12:39imirkin: then you def don't want anything to do with nouveau on GM20x+.
12:39imirkin: nvidia locked it all down and won't release blobs.
12:40ntz: imirkin: I have business class latitude for $3000 and the built in gfx in i7 kaby lake (i7-7600U) is just still very poor not even comparable with x30m nvidia
12:41ntz: imirkin: nvidia won't release blobs ?
12:41ntz: sorry ?
12:41ntz: imirkin: so in the other words, this laptop ``https://support.hp.com/rs-en/document/c05384994'' is not going to work in linux even with blob ?
12:42imirkin: the chips now check crypto signatures on uploaded firmware
12:42ntz: holy sh***
12:42imirkin: and nvidia won't release the firmware
12:42imirkin: the blob driver is able to properly upload firmware of course
12:43ntz: fucking rapists from nvidia ... I can't believe that ... why for a God sake they've literally removed a possibility to run theirs f***** chips in linux then ?
12:43imirkin: but nvidia has not released all the necessary blobs in a separately redistributable package
12:43imirkin: [have i mentioned that i recommend amd?]
12:43ntz: imirkin: wait, sorry, perhaps I badly understood to it ... there will always be `supported' blob from nvidia for 1050, right ?
12:44imirkin: 'blob' can mean multiple things
12:44ntz: blob is for me proprietary nvidia driver for linux
12:44imirkin: blob driver = piece of code running on your cpu that you downloaded of the internets and are running in ring-0
12:44imirkin: blob firmware = firmware that runs on the gpu that was written (and signed by) nvidia
12:45ntz: just `nvidia.ko' consisting of secret analyzer of what you type and where you connect :P
12:45imirkin: nvidia isn't releasing the blob firmware separately
12:46imirkin: and has either on purpose or by happy coincidence made it much, much, much harder to extract from their driver package
12:46loonycyborg: can you extract it from their driver package somehow?
12:46imirkin: [sufficiently harder that we have yet to succeed at it]
12:46ntz: imirkin: my big apologize but you're confusing me .... please tell me just clearly - ``https://support.hp.com/rs-en/document/c05384994'' ... is nvidia chip in this laptop supposed to work with 1) nouveau, 2) nvidia.ko ?
12:47imirkin: may or may not be able to bring up a screen with nouveau. i can't make specific comments about nvidia's driver, but it might work. plenty of people complaining about their drivers though in these kidns of hybrid setups.
12:48imirkin: [moral of the story: stick to intel/amd]
12:48loonycyborg: I would expect nvidia driver to work
12:49loonycyborg: I might be misparsing it but it doesn't mention having intel video too
12:49loonycyborg: though afaik pretty much all newer laptops are supposed to have hybrid graphics
12:50ntz: guys :D oh my dear guys .... the level of my confusing surprisingly increased since I've entered this chan and naively asked hoping for things to get clearer :D
12:50ntz: loonycyborg: it surely has, it's just in i7 cpu for sure
12:52loonycyborg: well I'm using nvidia card only on desktop atm and it's supported both by nouveau and nvidia-drivers
12:52loonycyborg: I heard on laptops you need to use some thing called "bumblebee" but it should be supported fine too
12:53loonycyborg: or some other dedicated solution to support hybrid graphics
13:00ntz: imirkin: I beg for answer :D, can you just clarify my assumption: nouveau is NOT going to work with new nvidia chips because this rapists company just wants ppl to use nvidia.ko with theirs proprietary code
13:00imirkin: ntz: define 'work'
13:00imirkin: nouveau hobbles along.
13:01imirkin: there's no black and white answer like you're hoping for
13:01ntz:loves imirkin's honesty
13:02imirkin: nouveau will load on all currently released nvidia GPUs, minus the ones where it doesn't :)
13:02ntz: I give up :D ... imirkin, work == I could be able to run X on that and hdmi will work so I'll be able to connect external display
13:02imirkin: given that it loads, you're going to have display, and on all but GP108, very slow acceleration
13:02imirkin: [on GP108, no accel]
13:02ntz: ok, lemme check ... so 1050 is gp108
13:03imirkin: i believe you need accel in order to do reverse prime, which is most likely how you'd be connecting that external monitor.
13:04ntz: https://nouveau.freedesktop.org/wiki/CodeNames/ according to this it's gp107, does it change something :D ??
13:04imirkin: marketing names are tricky.
13:04imirkin: that codenames thing is just a quick reference
13:06ntz: guys, thanks much for your input
13:07ntz: you're great :D, I wish, I'd never entered this chan and asked :D
13:07imirkin: you'd be much better off with an intel-only laptop
13:07imirkin: for that kind of usage scenar
13:07dcomp: So I installed the nvidia kernel driver to get an mmiotrace and figure why I cant suspend on my 840M GM108, But I'm getting "NVRM: RmInitAdapter failed! (0x26:0xffff:1113)" anyone seen that error before from the nvidia driver?
13:08imirkin: also e.g. a bunch of people with GP107's can't get nouveau to load on some dell laptops because it comes up in some weird low-power state
13:09ntz: imirkin: I have several intel laptops, I just wanted to have one laptop with decently strong gpu ... according to this https://www.notebookcheck.net/Mobile-Graphics-Cards-Benchmark-List.844.0.html I am unable to find laptop with comparable amd chip like 1050
13:10ntz: I have my old pavillion with gtx230m which works with nouveau pretty fine
13:10imirkin: i guess it's unclear to me what you're planning on doing with this decently strong gpu
13:11imirkin: if all you want to do is plug in an external monitor whose picture is to be drawn on the intel gpu
13:11ntz: play some game sometimes
13:11imirkin: ok, well that i can guarantee won't end well with nouveau on those chips.
13:11ntz: I just ride on bike a climb a rocks a lot and I have approx 2 months in year something broken and have to stay in bed
13:12imirkin: [and this is why i tell people that sports/exercise is bad for their health]
13:12ntz: bones will recover :P but lack of real life won't :P
13:14loonycyborg: nvidia-drivers will handle games just fine, unless there will be complications from hybrid graphics
13:14loonycyborg: I think better get a desktop :P
13:14imirkin: with amd =]
13:15ntz: imirkin: https://www.notebookcheck.net/Mobile-Graphics-Cards-Benchmark-List.844.0.html ... maybe I look wrong but there's not a viable candidate from amd that can compare to 1050
13:16ntz: and that the laptop is for $1000-1200
13:17loonycyborg: I don't know much about amd. Their proprietiary drivers were shit in the past.
13:18loonycyborg: afaik their opensource drivers are better right now. But perhaps not on all gpu models
13:18loonycyborg: I know very little about amd
13:21ntz: guys, again .... thanks much for all your input .. it's now clearer for me ..
13:22ntz: I will look a bit further and consider a different laptop but that one just satisfies all my reqs at 100%
13:22loonycyborg: in the past nvidia gpu + nvidia-drivers was the best way to get gpu acceleration on linux for games etc
13:22loonycyborg: there's a chance that amd changed it for their newest models but I'm not sure
13:22ntz: well, proprietary driver will work
13:41imirkin: loonycyborg: from what i understand, their situation with hybrid graphics is pretty much universally fail.
13:41imirkin: their "solution" involves you tanking battery on your laptop
13:42loonycyborg: ntz's use case involved constant access to power outlet
13:42loonycyborg: so if tanking battery is the only issue..
13:48karolherbst: imirkin: I am sure it is mostly marketing garbage, and that's "better" to have an nvidia GPU than an intel one
13:48karolherbst: I think intel is still considered the worst crap on earth on any windows machine
13:50karolherbst: I am also quite sure it's still the case, but being behin AMD and Nvidia on windows isn't that bad actually
13:51imirkin: karolherbst: even with a perfect setup, intel is pretty far behind the lowest end nvidia gpu
13:51imirkin: [as far as perf is concerned]
13:53karolherbst: well yeah, but how many user actually care about that
14:05karolherbst: imirkin: what do we want to do regarding that textureGrad patch? Still wait for mwk to respond? We can still fix gm107 in a different patch and I say that understanding why this is important is less important than passing tests (especially the CTS ones). Testing would be kind of important though, but I really don't know what uses textureGrad in a way, so that we can indeed test it properly.
14:17imirkin_: well, i'm sure the patch is *right*
14:17imirkin_: coz, well, tests pass, and nvidia does it that way too.
14:18karolherbst: well we could also wait until all the other patches are finished to pass CTS, because I really don't know if something really uses it or something is really broken due to this
14:19imirkin_: i just see no reason why the original way doesn't work
14:19imirkin_: and that bugs me.
14:20karolherbst: maybe the quadops aren't correct
14:20imirkin_: no, they're correct.
14:20imirkin_: i've stared at them every which way.
14:20imirkin_: and also it *usually* works
14:20imirkin_: there's some edge condition
14:20imirkin_: probably something to do with some things being masked? dunno.
14:20imirkin_: but then what if lane 0 is masked...
14:20karolherbst: maybe some hw internal stuff?
14:21imirkin_: that's a pretty cop-out-way of saying "i have no clue wtf is going on"
14:27karolherbst: well, I won't look at it before fixing all the other issues anyway. And this sounds like a better plan to spend my time anyhow.
14:29karolherbst: and next point on my list is that multisampled blitting.
14:30imirkin_: ideally reator would have a maxwell or pascal plugged in so that we can test some of this stuff (also the fp64 porting)
14:31karolherbst: is there no maxwell or pascal right now?
14:31imirkin_: as i don't presently have a maxwell plugged in locally
14:31imirkin_: i have no idea
14:31imirkin_: i don't even know how to connect to reator anymore
14:31karolherbst: I think I asked mupuf to put in his nve6 and maxwell+
14:31karolherbst: I see
14:31imirkin_: i don't think i've done it since mupuf moved to finland. i.e. not in a while :)
14:32karolherbst: well at least that fp64 works on my kepler afaik
14:32mupuf:does not remember what is in there
14:32mupuf: I booted reator yesterday
14:32karolherbst: I check
14:32mupuf: as I picked up again the fan issue on fermi
14:34karolherbst: what was the right port again?
14:37karolherbst: oh wait, I got it, but I can't access from here anyhow
14:40mupuf: lack of key?
14:40karolherbst: exactly, I am still at work
15:09karolherbst: mupuf: by the way, did you read the paper I told you about?
15:09mupuf: oh, nope :D
15:10mupuf: completely forgot!
16:25Sophira: I've been trying to investigate why nouveau seems to freeze up a lot on my system, and I've discovered that I have an obviously-glitched line in my /proc/mtrr file. Would this be causing issues for nouveau? https://pastebin.ca/3886647 (Gentoo Linux, sys-kernel/gentoo-sources-4.12.12 with no changes, NVidia GeForce GTX 970)
16:26Sophira: (My BIOS does not have a setting for changing MTRR mapping from continuous to discrete, and I have CONFIG_MTRR_SANITIZER=y and CONFIG_MTRR_SANITIZER_ENABLE_DEFAULT=1 set in my kernel with no difference to the contents of /proc/mtrr)
16:31imirkin_:doesn't know the first thing about mtrr... esp how it interacts with PAT
17:21dchotas: Hey guys, I'm on Arch running xf86-video-nouveau on an optimus enables laptop. When I set DRI_PRIME to 1 and run `screenfetch` it returns NV106 altough I have a 920M. Should I be concerned about this?
17:25karolherbst: looks fine
17:25imirkin_: dchotas: what's the thing that's triggering your concern?
17:26imirkin_: that we don't figure out the name the board was marketed with? or something else?
17:26imirkin_: [nice, they're selling GK208's as 920M's...]
17:28dchotas: well, I was just worried that the mislabelling could represent other underlying issues as im not too informed about the matter.
17:30imirkin_: ah no - it's properly labeled. it's NV[chip id], in your case the chip id is 106
17:31imirkin_: which is indicative of a GK208[b] gpu
17:31imirkin_: these are sold under 100 different marketing names
17:36dchotas: ah, I get it, thanks for the heads up!
17:42karolherbst: meh, mmt doesn't work anymore, how to get a working driver now again
17:44karolherbst: nice, the 381 driver still works
17:45karolherbst: imirkin_: maybe we should start saving all those ebuilds, just in case
17:45karolherbst: (or just get those back with git...)
17:49aphirst: btw, ickle in #intel-gfx is convinced that the problem I was having is to do with the intel driver
17:50aphirst: he submitted a patch upstream but the problem still seems to persist
17:50aphirst: prime related, clunky handling of which monitor is 'primary'
17:50aphirst: if i understood right
17:50imirkin_: aphirst: didn't i say the issue was with intel?
17:51aphirst: imirkin_, heh, you did indeed, but I still think that my personal application of bayes' theorem was sound
17:51aphirst: in any case consider my priors well and truly updated
17:52imirkin_: finding a proper prior is always the hardest
17:52aphirst: well i think it's probably best to not worry too much about that
17:53aphirst: asking for a "proper prior" seems to be asking to be able to put the cart before the horse
18:00karolherbst: imirkin_: mhh, nvidia sets GK104_3D.MULTISAMPLE_ENABLE to true, we to false, but I doubt that this alone has any impact of what we do anyhow, and as you already thought, there is no explicit RESOLVE thing set or anything
18:00imirkin_: multisampling affects rasterization
18:01karolherbst: well, I put true there, but no difference ;)
18:01imirkin_: yeah, i assumed as much.
18:01karolherbst: ohh, it could be one of those UNK fields as well
18:02karolherbst: GK104_3D.UNK204 = 0x6
18:06imirkin_: do you understand the problem?
18:11karolherbst: I simply now, that the stencil buffer contains wrong values
18:17karolherbst: uhh, I coul do a trace with samples=1 and samples=2 and compare that
18:18imirkin_: do you understand stencil?
18:18karolherbst: I know what it does, somewhat
18:18imirkin_: if not, that might not be a bad thing to investigate :)
18:20karolherbst: fun. Why does nvidia explicitly disable CSAA when enabling multisampling...
18:20karolherbst: ohh wait, the order is just different
18:21karolherbst: imirkin_: what does "cent" do in interp?
18:22karolherbst: related to multisampling? like getting an avarage value or something like that?
18:23karolherbst: because they set it for samples > 1
18:25imirkin_: highly related to multisampling
18:25imirkin_: meaningless without multisampling ;)
18:25karolherbst: I see
18:26karolherbst: where are our blitter FPs? I only found the VPs
18:37imirkin_: they're generated depending on formats
18:37karolherbst: ohh, I see
18:38imirkin_: it'll do either a 2d- or 3d-blit
18:38imirkin_: 2d if it can, 3d if it can't
18:38karolherbst: yeah, already saw that
18:38imirkin_: and the blitter is generated in nv50_surface.c iirc
18:38imirkin_: the tgsi for it, that is
18:38imirkin_: which is then compiled/etc
18:39karolherbst: is there a problem when I always use the 3d blitter?
18:40karolherbst: or is the 2d blitter simply preferred, because it let us use the 3d engines for "more usefull" things?
19:53karolherbst: imirkin_: what is the 3D.RT stuff?
19:54karolherbst: we don't do any with those, right?
19:54karolherbst: or well
19:54karolherbst: we don't touch RT_HORIZ
19:54karolherbst: or I grep something wrong
22:23karolherbst: mhhh weird, with apitrace i965 and nouveau produce the same results
22:24karolherbst: it complaints about glVertexPointer though
22:26karolherbst: yeah, it makes no sense
22:27karolherbst: I think glReadPixels is simply wrong
22:27karolherbst: not the blit or anything