05:02 karolherbst: mupuf: like so? https://gist.github.com/karolherbst/29856f6d5e658e4c709a
05:04 mupuf: karolherbst: looks good!
05:04 mupuf: is that something you are also satisfied with?
05:04 karolherbst: yeah
05:04 karolherbst: I've needed to add/change some clk apis though
05:04 mupuf: time for a v2 then?
05:04 karolherbst: but in the end we should remove all those nvkm_clk_*state functions and have only one nvkm_clk_reclock(clk);
05:05 karolherbst: I think...
05:06 karolherbst: the current state selection situation is quite a mess
05:07 karolherbst: mupuf: that's the patch currently: https://github.com/karolherbst/nouveau/commit/0e013134237f4bb77b76f84bfb81e6992875dcac
05:07 karolherbst: mupuf: have to clean it up first though
05:08 mupuf: can't check it out right now :)
05:09 karolherbst: but yeah, I think we need a nvkm_clk_reclock function in the end, it would simplify all the state stuff
05:09 karolherbst: if you want to lower clocks on high temp, modify tstate directly and call nvkm_clk_reclock
05:10 karolherbst: mhh
05:10 karolherbst: maybe the helper make actually sense
08:42 karolherbst: Tom^: wanna do a new benchmark with my newest branch?
09:16 karolherbst: mupuf: v2 incoming
09:21 karolherbst: ... nvidia has now wayland support
09:23 karolherbst: and kms support
09:23 karolherbst: what is happening...
09:23 karolherbst: and prime support?
09:25 karolherbst: ohh wait, that will be most likely the old crappy way
09:31 karolherbst: mhh, is the mailing list a bit broken today?
09:35 dcomp: if the blob has Prime I might actually get use out of this 840M
09:35 karolherbst: it is the crappy prime
09:35 karolherbst: intel modesetting and nvidia renders everything
09:35 karolherbst: so the nvidia gpu is always on
09:36 dcomp: well as long as the perf is better than this haswell gpu
09:37 dcomp: might give better mmiotraces?
09:38 karolherbst: doesn't matter for traces
09:38 karolherbst: bumblebee with primusrun is currently the superior solution, even if it costs you like 5% performance
09:41 dcomp: lpppis there a big difference between primus and virtualgl
09:41 dcomp: because the performance was less that stellar
09:42 dcomp: might be an interaction with Xwayland
09:42 tsermyat: guys, if i do passthugh in kvm to the second card for windows guest is the driver in the guest is used or the driver in the host?
09:44 karolherbst: tsermyat: it highly depends on the setup
09:45 karolherbst: tsermyat: usually only a virtual gpu is passed through and then you get only crappy opengl acceleration
09:45 karolherbst: tsermyat: but with the right hardware one can pass through the entire gpu and use native drivers
09:47 dcomp: tsermyat: if you have vt-d on intel i belive
09:47 tsermyat: karolherbst: well i will passthough entire gpu
09:47 tsermyat: but should i have proprietay nvdiai on host or not?
09:47 karolherbst: dcomp: and then you need a motherboard which also supports it
09:48 karolherbst: tsermyat: never done this myself, I just know that it works if you have the right hardware and I think some qemu patches? No idea
10:34 karolherbst: imirkin: I found a game (wine) where I get some corrupted textures with nouveau (intel works fine) I already created a trace and will create the bug soon
10:35 aaronp: karolherbst, yeah, 364.12 just moved the existing prime support from nvidia.ko to nvidia-drm.ko. Render offload support will require some more work to allow the X driver to load as a GPU screen, and negotiate GLX contexts with Mesa's libglx loaded.
10:38 karolherbst: aaronp: yeah I know :/
10:39 karolherbst: it is just a bit sad that the important PRIME stuff is GPL only. I would usually understand, but this annoys users more in general
10:39 Yoshimo: not a problem if you use nouveau, is it?
10:40 karolherbst: no
10:42 karolherbst: imirkin: here the bug if you want to take a look, the ML seems to be broken again or something https://bugs.freedesktop.org/show_bug.cgi?id=94650
10:48 karolherbst: aaronp: it is nice having kms support though :) no idea how, because I always thought kms is also GPL only
10:53 aaronp: karolherbst, you can read kernel/nvidia-drm/nvidia-drm*.[ch] in the .run package. :)
10:59 karolherbst: uhh, nvidia-drm is MIT
11:03 karolherbst: mhh, well when I got some time and are really bored, maybe I look into that :D
11:11 karolherbst: mupuf: bad luck, the ML seems to be messed up again...
11:53 karolherbst: pmoreau: if you are interessted, there is now vulkan support in wine-staging :D
11:56 tsermyat: karlmag: what is it?
11:56 tsermyat: karolherbst:
11:56 tsermyat: i mean vulkan
11:57 karolherbst: something completly overhyped
11:58 tsermyat: dx11?
11:59 karolherbst: nah I said overhyped. usually with new dx versions most of the people complaint about worse performance somehow... for whatever reasons
12:00 tsermyat: don't understannd the aim of it
12:00 tsermyat: i have nvidia card
14:09 tsermyat: latest nouveau has no glamor support!
14:09 tsermyat: why?
14:12 karolherbst: tsermyat: because it was broken
14:12 karolherbst: either use modesetting or don't use glamor
14:13 tsermyat: what's modesetting?
14:13 karolherbst: a ddx
14:13 tsermyat: i asked guys how to make windows open more quickly they told i should have nouvau with glamor use flag but i see there is no that flag anymore
14:26 imirkin_: tsermyat: glamor = slow. i wouldn't worry about it.
14:27 imirkin_: there's no reason for nouveau to integrate with glamor
14:27 imirkin_: if you want to use glamor, use the modesetting ddx
14:29 tsermyat: imirkin: that modesetting will work even if nouveau is compiled without glamor?
14:29 sarnex: you need xserver compiled with glamor for modesetting
14:30 sarnex: on gentoo that's USE=glamor emerge xorg-server
14:30 imirkin_: tsermyat: nouveau is many things. the nouveau ddx has no support for integrating with glamor. there's no benefit to having this.
14:30 imirkin_: tsermyat: so it doesn't exist. plain and simple.
14:31 imirkin_: tsermyat: if you have information to the contrary, then that information is wrong.
14:31 karolherbst: maybe we should ask him what gpu he has...
14:31 imirkin_: doesn't matter
14:31 imirkin_: having a glamor integration in the nouveau ddx doesn't make sense under any circumstances
14:31 sarnex: he said in #gentoo-wine he has a gt730
14:32 karolherbst: mhh 730 is fermi or kepler
14:32 sarnex: but the one without accel(fermi?
14:32 imirkin_: doubtful - they all have accel
14:32 sarnex: i mean reclocking
14:32 imirkin_: and in any case, none of that affects my comments
14:33 karolherbst: except he has a maxwell, where he should not use the nouveau ddx anymore
14:33 imirkin_: nouveau ddx won't even load in that case
14:33 karolherbst: well the stable one does
14:34 imirkin_: sadly yes. a mistaken and improper attempt at integrating glamor
14:35 imirkin_: by 'stable', you of course mean 'outdated', right? :p
14:35 imirkin_: 1.0.12 is the stable version.
14:36 karolherbst: stable is 1.0.11 in gentoo
14:36 tsermyat: sarnex: xorg is already compiled with glamore
14:36 sarnex: then you only need the conf file i posted in the other channel for modesetting
14:36 imirkin_: sure, on $distro, version $xyz is the latest. but the stable version (aka supported by upstream) is 1.0.12.
14:37 tsermyat: karolherbst: my gou is gt730 fermi
14:38 imirkin_: well, if you want to slow down your experience, you should switch to modesetting + glamor.
14:38 tsermyat: imirkin: i want to speed up!
14:47 karolherbst: imirkin_: that's why I said in gentoo
14:48 tsermyat: imirkin: so i wont do that modesetting
14:48 tsermyat: so i can't get any other 2d accel?
14:48 imirkin_: tsermyat: i def wouldn't recommend it
14:49 imirkin_: tsermyat: the nouveau ddx should provide the best 2d accel
14:49 imirkin_: tsermyat: pastebin your xorg log to make sure everything's ok
14:49 tsermyat: imirkin: ok, then my question today to guys was that i feel like windows in windows os are opening\drawn more quickly. is it placebo or not?
14:51 imirkin_: i dunno... could be related to a lot of things
14:51 imirkin_: like window manager, etc
14:53 tsermyat: imirkin: https://bpaste.net/show/fdf6364ed72e
14:54 imirkin_: tsermyat: that all looks fine
14:54 imirkin_: i also have a GF108, and it feels perfectly fast for opening windows and whatnot...
14:55 tsermyat: thanks
14:55 tsermyat: i wanted to try that awesome wm and can't compile it anyway
14:56 tsermyat: it seems opening fast but for example in xfce when i press start button it takes time for options to appear
14:56 tsermyat: like delay is set
14:56 tsermyat: may be i can decrease that delay?
15:21 karolherbst1: tsermyat: maybe your issue is just slow CPU, small amount of RAM?
15:21 karolherbst1: (or slow HDD)
15:24 tsermyat: karolherbst: well as guys told my mainboard and psu are dying - have electrical issues like not all ram slots work + usb hdd 3.0 but i have 16 g ram now
15:24 tsermyat: karolherbst: is suspect that xfce is the cause
15:25 tsermyat: because only start menu has that delay
15:25 tsermyat: if i load thunar for the first time it takes 3-4 seconds to emerge
15:25 tsermyat: is it long?
15:26 tsermyat: on next click it loads within 1 second or so
15:28 karolherbst: tsermyat: that's disc caching
15:28 karolherbst: the first time you start it, it has to read all kind of stuff from hdd
15:29 karolherbst: the second time, linux cached it inside RAM
15:31 tsermyat: i think hdd is the cause then and due to mainboard ans psu problems
15:33 karolherbst: no, this is normal behaviour
15:33 karolherbst: it is just the way it is
15:34 glennk: sounds like the drive has spun down if it takes a full second to load something off it
15:34 karolherbst: well it a SSD it would be much faster though
15:34 karolherbst: glennk: shared libraries
15:34 glennk: nah, login screen uses most of those so should be warm cache there
15:34 karolherbst: glennk: chromium takes like 10 seconds (+10 loading all the cached/other stuff) here the first time
15:34 karolherbst: glennk: well it depends a lot on what you use, right
15:35 glennk: <tsermyat> it seems opening fast but for example in xfce when i press start button it takes time for options to appear
15:35 karolherbst: but usually computers are sold with crappy discs, so
15:35 glennk: ^^ that use case in particular
15:35 karolherbst: eah
15:35 karolherbst: y
15:35 karolherbst: this is annoying
15:35 karolherbst: I would assume it calculates something, but mhh
15:36 tsermyat: well i wonder how my pc works at all, half of power cords make sata hdds choke and give abrupt sounds
15:36 karolherbst: glennk: dolphin also need slike 4 seconds here for the first time
15:37 glennk: karolherbst, spinning platter?
15:37 karolherbst: glennk: because it has to read the home directory, and maybe there are fancy stuff like thumbnails ;)
15:37 karolherbst: *things
15:37 tsermyat: karolherbst: yes it drives me crazy that exact thing as it's like pc is calculating
15:37 karolherbst: tsermyat: you should ask the xfce guys
15:38 karolherbst: glennk: 2.5" disc ;)
15:38 tsermyat: first i need new pc
15:40 glennk: not sure why people torment themselves with non-ssd system disks
15:41 karolherbst: I lost my trust in ssd, because 2 out of 2 ssds died on me
15:42 tsermyat: never tried and wont' try they are like led lamps expensive but cheating
15:42 karolherbst: why is it cheating?
15:42 karolherbst: they are nice, I just bought the wrong ones :D
15:42 tsermyat: they die soon
15:43 tsermyat: i mean led and spiral lamps
15:43 tsermyat: and bad for health
15:43 tsermyat: especiall spiral ones
15:43 tsermyat: they have very high frequency
15:43 karolherbst: I am sure a HDD is worse for your health than a SSD
15:43 karolherbst: especially if thrown
15:43 glennk: led lamps require proper cooling, mounting them in old fixtures designed to trap heat guarantees they'll fail early
15:44 tsermyat: never used led lamps too expensive but my words hold true for spiral onces
15:44 tsermyat: all dies within a year
15:44 tsermyat: died
15:44 imirkin_: spiral = fluorescent?
15:44 tsermyat: mercury
15:44 tsermyat: or smth like that
15:45 glennk: cfls are horrid, but they are on their way out
15:45 karolherbst: is it still legal to sell them with mercury?
15:46 tsermyat: in some countries yes
16:11 karolherbst: imirkin_: any idea what I could do to debug the issue I have? It looks basically like some uninitialized memory region is used by the application
16:12 imirkin_: not offhand
16:12 imirkin_: i'll try to repro ... at some point
16:13 karolherbst: it does seems to look different with each run
16:13 karolherbst: and pretty clean after the card was off
16:18 karolherbst: at least the copied texture doesn't appear, only some noise..
16:20 imirkin_: is this recent?
16:40 pecisk: yay, GTX 900 support finally in nouveau
16:54 karolherbst: imirkin: no idea
17:01 karolherbst: imirkin: I could check older mesa version if you think something might have broken this
17:46 karolherbst: imirkin: you are right :O it worked before
17:46 karolherbst: I tested 11.0.0
17:46 karolherbst: and with this, it doesn't show the issue
17:47 imirkin: karolherbst: could i request a bisect? :)
17:47 karolherbst: :D
17:47 karolherbst: imirkin: do you know by any chance when DRI3 support was added?
17:47 imirkin: nnnnnot 100% sure
17:47 imirkin: 10.2 or so for mesa, but i think st/mesa got it later
17:48 karolherbst: mhh k
17:48 karolherbst: I even wanted to try out mesa-9, but I couldn't cause of dri3....
17:58 karolherbst: noooooo
17:58 karolherbst: two different issues...
17:59 imirkin: :(
17:59 karolherbst: well
17:59 karolherbst: in the end it is good
17:59 karolherbst: because two issues found
17:59 imirkin: but presumably one was already fixed
18:00 karolherbst: mhh meh, merge base is bad...
18:16 karolherbst: imirkin: this is stupid... 11.0.0 doesn't have any issues and 11-branchpoint has the one eye texture missing, but the book looks okay... (with master both is broken)
18:24 karolherbst: good, 11.1-branchpoint doesn't have either issue
18:50 forevermobile: No screens found on a GTX960 :(
18:51 karolherbst: forevermobile: do you have a xorg.conf file?
18:52 karolherbst: forevermobile: maybe just paste your xorg.log somewhere
18:53 forevermobile: I do, without it I get segfaults. I made it with X -configure.
18:55 forevermobile: Currently don't have a desktop so putting it up on a paste site would be tough.
18:56 karolherbst: well you shouldn't use a xorg.conf file usually
18:56 karolherbst: you should rather figure out why it segfaults without
19:06 forevermobile: Those are hard to Google, last it says is AIGLX: enabled GLX_MESA_copy_sub_buffer and then it's a backtrack.
19:16 forevermobile: Gtx660 has the same issue so there's that.
19:18 Gorka: there is some ETA for hardware accel. on GTX960 NV126 (GM206)?
19:18 karolherbst: Gorka: linux 4.6
19:19 karolherbst: forevermobile: without the entire log we won't be able to help you
19:19 Gorka: nice, thanks karolherbst
19:21 forevermobile: What's doable from cli?
19:21 karolherbst: forevermobile: wgetpaste
19:22 karolherbst: forevermobile: there might be other scripts as well
19:22 karolherbst: I don't know them though
19:31 karolherbst: imirkin: okay, maybe it is no regression, because by accidence it works sometimes
19:36 forevermobile: Https://transfer.sh/9Gl8y/forevermobile
19:41 karolherbst: forevermobile: is nouveau loaded?
19:42 karolherbst: forevermobile: lsmod | grep nouveau
19:42 karolherbst: ohhh
19:42 karolherbst: forevermobile: your kernel is really old
19:43 forevermobile: Its 4.6
19:43 forevermobile: Errr 4.4 apparently
19:43 karolherbst: ohhh right
19:43 karolherbst: I looked at the build operating system
19:44 karolherbst: yeah 4.4 is fine
19:44 karolherbst: k
19:44 karolherbst: then check if the nouveau kernel module is loaded at all
19:44 forevermobile: It is in a few places including video
19:44 karolherbst: lsmod | grep nouveau
19:45 karolherbst: does this command print anything?
19:45 forevermobile: Not after a reboot...
19:46 karolherbst: k, then nouveau isn't loaded
19:46 karolherbst: find /etc/modprobe.d/ -type f -exec grep nouveau {} +
19:46 karolherbst: there might be a entry like "blacklist nouveau" or something
19:48 forevermobile: Nvidia-installer-blacklist-nouveau.conf
19:49 karolherbst: I suppose you removed the nvidia stuff?
19:49 forevermobile: Yea, purges and uninstaller across the board
19:49 karolherbst: k
19:49 karolherbst: then remove the file you found
19:50 karolherbst: you may need to rebuild your initramfs too (not sure though)
19:51 forevermobile: Nouveau is definitely running now (resolution on screen is better) but same segfault
19:52 karolherbst: then the new xorg log please :)
19:52 karolherbst: (I have an idea what is wrong, just want to be sure)
19:55 forevermobile: Https://transfer.sh/15TEuS/forevermobile1
19:56 karolherbst: k, that looks better
19:56 karolherbst: at least the nouveau ddx is loaded now
19:57 karolherbst: forevermobile: after starting X, is nvidia loaded by any chance?
19:58 forevermobile: X doesn't start
19:58 karolherbst: yeah well after it tries to start
19:58 karolherbst: it is still a bit odd
19:59 karolherbst: forevermobile: can you update all alternatives and see if somewhere nvidia was still selected or something like that?
19:59 forevermobile: Lsmod grep nvidia is blank
20:00 forevermobile: Update all alternatives?
20:00 karolherbst: update-alternatives --all
20:01 karolherbst: forevermobile: dmesg doesn't show any errors, right?
20:03 forevermobile: Nothing in a screen full.
20:03 forevermobile: Alternatives never talked about nvidia or nouveau
20:04 karolherbst: can you isntall the xorg-server debug package and see if you get a more meaningfull stacktrace in the log?
20:04 karolherbst: something is not right, but I have no idea what
20:05 karolherbst: forevermobile: debug package of xserver-xorg-core
20:07 forevermobile: Alright, rebooting
20:10 forevermobile: _start+0x29
20:16 karolherbst: likes 5-1 are more important
20:16 karolherbst: *lines
20:21 forevermobile: Is Unknown chipset NV126 important?
20:21 karolherbst: I thought you had a gtx 660?
20:22 forevermobile: I have twin 660s. I threw one in earlier to test. I'd prefer to run my 960 though
20:23 karolherbst: where do you get the unknown chipset error?
20:24 karolherbst: though you have to use the modesetting ddx for the 960 anyway, but the X server should detect it itselfg
20:24 karolherbst: *itself
20:24 forevermobile: In modesetting in both log
20:25 forevermobile: *Xorg
20:25 karolherbst: well as long as only nouveau complaints it shouldn't matter, the modesetting driver driver the card
20:25 karolherbst: but I guess it still crashes?
20:26 forevermobile: Yea I don't get anything interesting on backtrace lines 1-4
20:27 karolherbst: I would guess nvidia bricked those libraries somehow
20:27 karolherbst: try to install xorg-server-core and the mesa packages
20:27 karolherbst: *reinstall
20:28 karolherbst: I think libgl1-mesa-dri is the mesa one
20:28 karolherbst: and libgl1-mesa-glx
20:32 forevermobile: Purge and install and she fires right up
20:34 forevermobile: Do I use arandr with nouveau
20:34 karolherbst: doesn't matter
20:34 karolherbst: you can use what your settings stuff gives your
20:34 karolherbst: or xrandr from cli
20:35 karolherbst: forevermobile: what desktop manager are you using?
20:36 forevermobile: Uhhh a Frankenstein of preference. Openbox with some borrowed elements from lxde
20:36 karolherbst: mhh
20:36 karolherbst: do you use the lxde settings stuff?
20:37 karolherbst: there might be a display module or something like that
20:37 forevermobile: No. I try to avoid anything with major dependencies to any DE.
20:38 karolherbst: then arandr it is
20:38 karolherbst: but it might be that the change isn't persistent this way
20:39 forevermobile: Yea, I usually build the common setups and throw them in my execution pathto when I arandr
20:41 forevermobile: Or in my bash if it's common enough
20:42 karolherbst: yeah well, you shouldn't put that inside your bashrc ;)
20:43 forevermobile: Hmm it freaked when I tried to enable my 4k monitor
20:44 karolherbst: well
20:44 karolherbst: sometimes it helps when you have every display connected at boot
20:45 forevermobile: It goes no signal on both when I do that
20:45 karolherbst: mhh
20:46 karolherbst: and if you only connect the 4k display?
20:46 forevermobile: No signal
20:47 karolherbst: well somebody got 3x 4k displays running with nouveau here, so it should work in theory
20:47 karolherbst: mhh
20:47 karolherbst: maybe dmesg contains something
20:47 karolherbst: forevermobile: can you switch the disaplys to get a working state again?
20:47 forevermobile: Yea, ctl alt f# is dead on the 4k
20:49 forevermobile: Alright it's back on just the WQHD
20:52 karolherbst: forevermobile: you should try to play around a bit with plugging those displays and see if someting in dmesg comes up
20:53 forevermobile: Link training failed
20:53 karolherbst: uhh
20:54 karolherbst: then open a bug for this with dmesg attached, I bet somebody will look at this and reply, because I have to go to bed now, cause it is 5am here and I am still awake
20:55 foreverska: Fair enough. THanks a bunch karolherbst.
20:55 karolherbst: imirkin: so yeah, in the end it just succeeded by accident, you may have to run the trace 10 times to be really sure and you should at least run it once with llvmpipe to see how it should look like, cause somtimes you won't notice anything wrong