14:28peterius: I'm getting these 10 second mini freezes on my gtx 1060, anyone know anything about this? the mouse cursor still moves but everything else freezes
14:29peterius: I have these feeling it's trying to switch graphics cards on me or something and failing without any real errors...
14:29karolherbst: peterius: nothing inside dmesg?
14:30peterius: well, there's some stuff...[ 15.545428] ACPI Warning: \_SB.PCI0.GFX0._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20170303/nsarguments-95)
14:31peterius: there was something else, about some graphics card thing not being recognized...
14:31peterius: [ 15.537461] i915 0000:00:02.0: Direct firmware load for i915/kbl_dmc_ver1_01.bin failed with error -2
14:31peterius: i tried removing i915 but it won't give me a console before X11 so... and X11 wouldn't start either I think
14:32peterius: [ 15.845133] nouveau 0000:01:00.0: Direct firmware load for nvidia/gp106/gr/sw_nonctx.bin failed with error -2
14:32peterius: this, I actually thought I fixed this...
14:32peterius: there was a symlink and I replaced it with the real file, apparently it still didn't find it
14:32karolherbst: a symlink is fine
14:32karolherbst: what kernel are you on?
14:33peterius: I've tried 4.12.5 and now 4.12.12
14:34karolherbst: I think those aren't included in your initramfs
14:34karolherbst: the firmware files for nouveau and i915
14:35karolherbst: you have those files under /lib/firmware/ ?
14:35peterius: well, I think my initramfs is just for the uuid... I have all the modules compiled in... actually
14:35peterius: I just remembered, I set it to compile all the firmware into the kernel because it couldn't find the wireless firmware
14:35peterius: but now it finds it so...
14:35karolherbst: then you need to compile all firmware files into your kernel as well
14:35karolherbst: I see
14:35karolherbst: well for me it looks like that something related to this just fails
14:35karolherbst: as if not all firmware files are inside your kernel or anything like that
14:36peterius: yeah I thought it just put every piece of firmware in /lib/firmware into the kernel, it was this huge list... maybe I have to explicitly mention the nouveau and i915 firmware
14:36karolherbst: otherwise your kernel would get super big
14:36karolherbst: I would only add the files you actually need
14:37peterius: well, it's weird, I tried to, but then it seemed to add everything
14:37gnarface: don't you have to re-run `update-initramfs -u` after manually adding firmware to /lib?
14:38gnarface: on debian at least, with the default initramfs behavior, if you have one, that firmware probabably needs to be in it, and if you just copied it unpackaged to /lib/firmware, you need to take extra steps to make sure that it is
14:38peterius: well, I thnk gentoo put all the firmware into lib... I didn't download any... even for the wireless thing, it happened to have the new board file that I downloaded for it already, it just wasn't finding it
14:38gnarface: see if gentoo also has an update-initramfs script. if it does, run it
14:38gnarface: just as snake oil
14:38gnarface: it may work
14:39peterius: so tso there's "include firmare blobs"... there's external firmware blobs, and then theres blob root directory
14:39imirkin_: peterius: are you using the nvidia chip or intel chip for display?
14:40peterius: I don't know
14:40imirkin_: grep . /sys/class/drm/card*-*/status
14:41peterius: card0-eDP-1 connected?
14:41imirkin_: ok, and card0 is ...
14:42imirkin_: cat /sys/class/drm/card0/device/label
14:43peterius: it must be the i915
14:43peterius: let me try explicitly naming the firmware files and see if that helps
14:49karolherbst: peterius: well, in the compile you should get rows like nvidia/gp106/gr/sw_nonctx.bin.S or something
14:50karolherbst: gnarface: not needed if you include the firmware files into the kernel directly
14:56imirkin_: peterius: ok, so then your issues are most likely intel-related rather than nouveau-related?
14:56peterius: yeah... I mean unless it's a mode switching thing
14:58peterius: I'll fix these driver things and then try some stuff and see what happens, thanks again
14:58gnarface: karolherbst: yea, but i was assuming that's still an if
15:00karolherbst: well, I didn't say you have to have them inside the kernel, initramfs should be fine as well though
15:01gnarface: i just know it's a problem i've run into repeatedly on debian; doing weird stuff then needing to manually flush the updates into initramfs
18:25peterius: yeah, I guess if these missing firmware things had been an issue, it wouldn't have started up
18:25peterius: I also tried installing the various modules Xorg complained about, xf86-video-intel, and xf86-video-nv, which I guess is different from nouveau?
18:25peterius: but I don't think it should be hanging like this
18:26gnarface: nv is some deprecated old one
18:27peterius: oh... also fbdev, and vesa... it said those modules weren't there... should I uinstall nv?"
18:27gnarface: i think xorg auto-detect just tries stuff in a predetermined order
18:28gnarface: if you don't want to use any of them, you should just manually configure it to the driver you actually want to use
18:28gnarface: then it won't matter whether the other drivers are installed or not
18:28gnarface: i doubt you'd actually want it to choose vesa unless there were no other options
18:28peterius: ugh... I thought the days of Xorg.conf were over...
18:29peterius: https://pastebin.com/b3pbJTTT this one also has some enqueue errors from i915
18:29gnarface: you still need one to do some stuff, but the good news is that you only need PART of one, if you're doing simple stuff
18:29gnarface: (on debian derivatives, look in /usr/share/X11/xorg.conf.d/)
18:30gnarface: ignoring those errors is an option too, probably
18:30gnarface: as long as it eventually settles on nouveau
18:30gnarface: or, you know, whichever one you actually want to use
18:31gnarface: nouveau or intel i guess?
18:31imirkin_: peterius: your display is attached to the intel
18:31imirkin_: and acceleration is running off the intel
18:32imirkin_: so ... you probably want to look to intel for any issues you're having.
18:42karolherbst: imirkin_: we have two arb_framebuffer_srgb regressions in 17.3
18:42karolherbst: blit renderbuffer/texture linear_to_srgb upsample enabled
18:42imirkin_: that sounds familiar for some reason
18:42karolherbst: will bisect it
18:43imirkin_: iirc it was random
18:44karolherbst: doesn't seem like it
18:44karolherbst: 20 passes in a row
18:44imirkin_: or you haven't captured the randomness properly ;)
18:44karolherbst: it always failed with master/cts branch
18:44imirkin_: e.g. run this, then this, then that, then it starts passing
18:45karolherbst: well, I will bisect it for now and then I might know more
18:45imirkin_: i could be misremembering of course.
18:45imirkin_: btw i pushed the fix for int64 shifts
18:46karolherbst: yeah, I saw it
18:46karolherbst: currently I want to track down regressions, because I saw some comparing piglit results from begining of this year
18:46karolherbst: and actually plan to do it more automated
18:46imirkin_: yeah, people have been "fixing" stuff all year long without me paying too much attention
18:46imirkin_: which probably means a bunch of stuff is broken on nouveau
18:47karolherbst: well, I will change that paying attention part :p
18:47imirkin_: i do remember seeing nha's change back in the day for that shift thing
18:47imirkin_: which is how i was able to track this down so quickly
18:48imirkin_: but i think i thought at the time "nah, it'll be fine" :)
18:49karolherbst: yeah, happens
18:51karolherbst: well I plan to setup ezbench for CTS/Piglit at least, not quite sure about which GPUs, depends on how many resources I can get for that
19:01airlied: fixing parallel piglit is the main blocker on having a CI
19:01imirkin_: parallel piglit seems to work for everyone except me
19:03karolherbst: airlied: well it works for me on nouveau if I exclude textureGather
19:03karolherbst: "./piglit-run.py -x glx -x streaming-texture-leak -x max-texture-size -x textureGather tests/gpu.py" is what I run
19:03karolherbst: no issues so far on nouveau with that
19:06karolherbst: airlied: but I would really like to setup some CI for that for nouveau for at least every chipset since tesla
19:06karolherbst: well chipset family
20:15imirkin_: airlied: reverse prime = tearing, right?
20:18imirkin_: airlied: this is in connection with https://bugs.freedesktop.org/show_bug.cgi?id=103202
20:21karolherbst: imirkin_: ddc32537d6db69198e88ef0dfe19770bf9daa536 broke it
20:21karolherbst: hakzsam: :p shame on you
20:22imirkin_: are we reporting wrong viewport width/height?
20:23karolherbst: no clue, have to go home now anyhow, will investigate after getting there
20:23imirkin_: i doubt this change directly broke anything, but i guess i dunno
20:25imirkin_: oh wait, lol
20:25imirkin_: + clamp_viewport(ctx, (GLfloat *)&x, (GLfloat *)&y,
20:25imirkin_: + (GLfloat *)&width, (GLfloat *)&height);
20:25imirkin_: that's totally wrong ;)
20:25imirkin_: hakzsam: tsk tsk tsk
20:26hakzsam: af45b8159cf5b9c15000e4221af6dc4cb85bd05e :)
20:26imirkin_: yeah that's probably better :)
20:55airlied: imirkin_: yeah tearing i think is normal
20:55imirkin_: airlied: and no compositor will save you?
20:56imirkin_: or atomic or whatever?
20:59Peterrr: Hi all, I am using the 4.13 kernel and I have a GTX 970. The nvidia proprietary driver is not installed and when I type lsmod I can see there is nouveau loaded. But when going in display configuration in the mate desktop I can only see one monitor so I cant configure the others, is that normal? the 970 is fully supported?
21:00imirkin_: pastebin xorg log + dmesg
21:03airlied: imirkin_: dont think so its just X rendering
21:03Peterrr: wait, I restarted on 4.9 to see something, but i was on 4.13 and it was not working, I reboot on 4.13 and give you that information
21:04imirkin_: airlied: ok. the bug talks about a GL application
21:04karolherbst: hakzsam: okay, this explains why the result looks different on master than on the broken commit
21:04imirkin_: airlied: like does glXSwapBuffers do anything useful for an app rendered to a reverse prime screen?
21:04imirkin_: [i imagine not]
21:04hakzsam: karolherbst: right
21:05karolherbst: but still broken for nouveau and it was not before
21:06karolherbst: but maybe on your fixed commit it is fixed again
21:06karolherbst: will check that
21:08karolherbst: hakzsam: ohh, it passes on your commit
21:08karolherbst: so another commit broke it later again
21:11airlied: imirkin_: the app has no idea
21:14imirkin_: airlied: yeah that's what i figured
21:14imirkin_: airlied: and i assume that having 2 screens with diff refresh rates yields a similarly happy scenario?
21:14imirkin_: [on the same gpu]
21:20airlied: imirkin_: yes even with same refresh can get it, since we don't genlock
21:21imirkin_: right ok
21:21imirkin_: and there's no way to say "sync to THIS connector" or whatever
21:22airlied: imirkin_: I think compositor sync to whoever has bigger crtc
21:23imirkin_: hehe ok
21:23imirkin_: oh, ickle reminds us that there's the TearFree option
21:23imirkin_: which might work
21:53karolherbst: imirkin_: 6839d3369905eb02151334ea7b4cd39ddcfa6770 broke it the second time
21:54imirkin_: ok, that was another clusterfuck i decided not to get involved in...
21:56karolherbst: in the commit message is actually a reference to the failing test
21:59karolherbst: imirkin_: so my guess would be that nouveau doesn't handle samples==1 "correctly"?
21:59karolherbst: this is weird
21:59imirkin_: no, it handles it correctly
21:59imirkin_: however the definition of 'correct' changed
21:59karolherbst: yeah, that's why I put it in quotes
22:00imirkin_: this is another one of those changes that seemed like it should be fine
22:03karolherbst: mhh, mhhh
22:03imirkin_: but i definitely saw potential for massive fail
22:03karolherbst: actually I could guess, that samples=1 still reaches nouveau, but now it is actually treated as samples==2 in gallium, not like 0 as before
22:05imirkin_: and/or our shit is broken for samples==2 but was working by accident before
22:06karolherbst: okay, I see
22:07karolherbst: guess we need to fix those stuff for the CTS anyhow somehow, generally
22:21karolherbst: nice, more CTS fixes upstreamed :)
22:45karolherbst: we have more regressions since 17.1.10 :/
22:45karolherbst: spec.!opengl 1.5.draw-elements
22:51john_cephalopoda: I get some system freezes with chromium, that seem to be related to nouveau.
22:52john_cephalopoda: Here is the excerpt from the kernel ring buffer, with two subsequent crashes: https://bpaste.net/raw/3f263aa6556c
22:52karolherbst: john_cephalopoda: kernel?
22:54john_cephalopoda: Mesa version is 17.2.2
22:55john_cephalopoda: It has happened for a long time, so no regression.
22:55john_cephalopoda: It actually got better, now I don't have to reboot to recover, it is enough to send SIGTERM to all non-init processes.
22:56john_cephalopoda: (But that could also be due to xorg-server changing or similar)
22:56karolherbst: well, nouveau tries to recover if somebody _really_ bad happens
22:57karolherbst: john_cephalopoda: are you experienced with installing nouveau as a module system wide?
22:59john_cephalopoda: You mean making nouveau a kernel module instead of compiling it into the kernel?
22:59karolherbst: well, also to install it from git basically
23:02john_cephalopoda: I am using a source-based distro, so patching the kernel and/or mesa is possible.
23:02karolherbst: well, I don't mean to patch the kernel nor mesa
23:02karolherbst: just getting a tree via git and install that module system wide
23:03john_cephalopoda: A kernel module or an xorg module?
23:07john_cephalopoda: I currently compile nouveau right into the kernel, so I would probably have to remove it before I can install it as a module.
23:09Lekensteyn: karolherbst: regarding your earlier comment, atm I am not working on the HDMI audio reset logic, don't know a clean way to do it
23:09karolherbst: Lekensteyn: worst case: there ain't a clean way
23:09karolherbst: john_cephalopoda: well you would need to compile as a module anyhow, because otherwise ttm wouldn't get selected
23:10karolherbst: you would just need to overwrite the nouveau module from the kernel
23:10karolherbst: john_cephalopoda: and you could grab a copy of the current nouveau code from here: https://github.com/karolherbst/nouveau
23:10karolherbst: it's basically nouveau master rebased for 4.13
23:10karolherbst: hence the branch name: master_4.13
23:11karolherbst: not sure if this changes anything, but mhh
23:12karolherbst: we have _some_ memory related fixes there
23:12karolherbst: and if chromium gets more stable out of the sudden, we did something right
23:12john_cephalopoda: The issue takes some time until it is triggered, and triggering it is pretty inconsistent.
23:12karolherbst: how much time?
23:14john_cephalopoda: It depends on various factors. Playing youtube videos in parallel seems to trigger it quicker. Then it can take ~ 10 minutes to appear. In other situations, it can take hours or doesn't appear at all.
23:15karolherbst: oh well
23:16peterrr: sorry I'm back, I do not remember what you ask me to do, pastebin xorg log file and what else?
23:17john_cephalopoda: I'll take a look at it tomorrow. Night!
23:22peterrr: So I just installed debian 9 and then upgrade to unstable, I have a 4.13 kernel, nouveau seems loaded in lsmod, but I have 4 monitors and only one is showing something, If i go in mate display option i see only one screen called unknown default
23:22peterrr: https://pastebin.com/V93hV8E1 here is the xorg log
23:24karolherbst: peterrr: well nouveau isn't used
23:24karolherbst: is nouveau even loaded?
23:24karolherbst: ohh wait
23:25peterrr: i see nouveau in lsmod
23:25karolherbst: sudo cat /sys/module/nouveau/parameters/modeset
23:26karolherbst: sudo cat /sys/module/nouveau/parameters/noaccel
23:26karolherbst: sudo cat /sys/module/nouveau/parameters/nofbaccel
23:26peterrr: 0 and 0
23:27robclark: [ 16.649] (EE) open /dev/dri/card0: No such file or directory
23:28robclark: looks a bit funny
23:28robclark: yeah, dmesg.. missing firmware or something?
23:29robclark: [ 15.348186] nouveau 0000:01:00.0: firmware: failed to load nvidia/gm204/gr/sw_nonctx.bin (-2)
23:29karolherbst: well mhh
23:29karolherbst: but modesetting should still work afaik
23:29karolherbst: at least the modesetting ddx should have been picked up
23:29karolherbst: I might be wrong
23:29karolherbst: ohh, seems like nouveau totally gives up "nouveau: probe of 0000:01:00.0 failed with error -2"
23:30peterrr: i did absolutely no modification at all, installed deb 9, upgrade to unstable thats all, if this may help
23:30karolherbst: peterrr: do you have linux-firmware installed?
23:30robclark: yeah, it didn't gte far enough to make card0 dev file..
23:30karolherbst: peterrr: or whatever that package is called on debian
23:31peterrr: its firmware-linux im installing it now
23:31karolherbst: good :)
23:31peterrr: is that why
23:31karolherbst: most likely
23:31peterrr: thanks a lot
23:31peterrr: brb installing and reboot
23:31karolherbst: you might need to regenerate the initramfs, but I hope the package manager takes care of this
23:32peterrr: I reboot! brb
23:33karolherbst: mhh, maybe nouveau should still try to get at least modesetting to work if we fail to init the gr or something
23:33karolherbst: better than having like nothing
23:33peterrr: works! thanks a lot
23:34karolherbst: those distros
23:34peterrr: last question, to set up the screen in the great order Its more useful to use xorg.conf? than using xrandr and load it automatically at the same time as my desktop environment?
23:34karolherbst: use xrandr
23:35skeggsb: um, it's *supposed* to support that already... wonder how that broke this time
23:35karolherbst: yeah, that's why I was confused because I thought it would
23:44peterrr: so i setup the multi monitor orders with mate control center, now I would like to use this with awesome wm, is there a way with xrandr to retrieve the full line that i need to type to set this order? xrandr -q give too much information
23:45karolherbst: peterrr: mhh, usually your desktop environment handles that, not the window manager (weird, isn't it?)
23:45karolherbst: or at least it is this way on plasma, maybe other DEs are weird here and let the window manager handle this
23:45karolherbst: but why should they
23:47karolherbst: imirkin_: 330d0607ed60fd3edca192e54b4246310f06652f broke spec@!opengl 1.5@draw-elements
23:49peterrr: karolherbst: Im gonna use awesome wm without any dm, i just found that the GUI program arandr is able to return a one line command to put the order you choose so I guess i just have to exec it when the wm start
23:52karolherbst: the Ubyte indices - offset: 2/3 (glMultiDraw) subtests are broken
23:52peterrr: well thanks a lot karolherbst for solving my issue