01:02karolherbst: mupuf, skeggsb: do you think it will be okay to use the coefficients I REd on my gpu for the voltage table for all gpus for now? The worst what could happen is, that we use the max voltage stated in the entries. I am sure that we need this so called "speedo" value which also get used in the gk20a code
01:02karolherbst: but the speedo value is given by the tegra code to nouveau :/
01:02karolherbst: and I have no idea where to find it
01:10karolherbst: Tom^: are you there?
01:36Jayhost``: Well Saturday was a bust and then MS keyboard crapped out. Will try again.
01:39Tom^: karolherbst: sort of.
01:39Tom^: karolherbst: irl catching up, dont think ive spent more then an hour infront of computer in a month now :P
01:40karolherbst: I see
01:41Tom^: karolherbst: bought one of these too http://i.imgur.com/PIDUlNZ.jpg . been destroying the earth slowly with it, one liter of gasoline at a time.
01:42karolherbst: Tom^: do you still have the fan fix from mupuf?
01:42Tom^: hm i think i could dig that up yes
01:42karolherbst: ahh no, it is there
01:42karolherbst: if you got time, could you test the stable_reclocking_kepler_v2 branch?
01:43karolherbst: I think it has everything to fix volting in an okayish way
01:43karolherbst: it's based upon linux 4.4
01:43karolherbst: it doesn't revolt on increasing/decreasing temperature yet though
01:44karolherbst: there is also a tool you could run with the blob (it compares blob voltage with the one nouveau would set)
01:45Tom^: in envytools?
01:45karolherbst: we are able to turn nouveau into a library and write tools based on it
01:45karolherbst: and just run it alongside the blob
01:46karolherbst: you also get a little bit more perf in heaven, because the clocks should be higher now
01:48Tom^: but moar flickers because of dynamic reclocking :P
01:50karolherbst: there is no dyn reclocking in that branch
01:50karolherbst: I want to upstream most of it, so it can't be there yet
01:50Tom^: best of two worlds hehe
01:51karolherbst: do you have currently nouveau or the blob running?
01:51karolherbst: then we should run that tool first
01:51karolherbst: just run make inside the top level directory of nouveau
01:51karolherbst: then as root "LD_LIBRARY_PATH=lib bin/nv_cmp_volt"
01:53Tom^: error while loading shared libraries: libnvif.so: cannot open shared object file: No such file
01:53karolherbst: if you use sudo
01:53karolherbst: sudo LD_LIBRARY_PATH=lib bin/nv_cmp_volt
01:56Tom^: karolherbst: https://gist.github.com/anonymous/7fc04ac397de965ca040
01:56Tom^: seems to be off ~-24481 uV
01:56karolherbst: looks okayish
01:56karolherbst: let me check something
01:56karolherbst: step size is 12500 uV
01:57karolherbst: so we are off by one step basically
01:57karolherbst: because I round up
01:57karolherbst: should be fine
01:57karolherbst: 0.0125V undervolting may be stable enough
01:57karolherbst: put some load to it
01:57karolherbst: maybe the value gets better
01:58karolherbst: either way, I used my gpu to get the coefficients to calculate teh voltage, and I am sure there are chipset/gpu specific factors I don't found yet
01:58karolherbst: but all in all it is better than stock nouveau
01:58karolherbst: stock nouveau was like 10% off for me
01:58Tom^: karolherbst: https://gist.github.com/anonymous/1693e4f31dd5d149a429
01:59karolherbst: now something looks odd
01:59karolherbst: can you open nvidia-settings and tell me which clocks nvidia clocked to?
02:00Tom^: core 1097 and mem 7000
02:01karolherbst: mhh something is wrong
02:02karolherbst: nouveau voltage shouldn't be below 950000 in your last paste
02:02Tom^: no matter the temp or load i never go past 1097 tho seems to be some kind of limit.
02:02karolherbst: voltage limit
02:06karolherbst: it sets the pstate voltage
02:09karolherbst: now I see it
02:09karolherbst: mode 2
02:09karolherbst: yeah, odd
02:10karolherbst: okay, but now why is the wrong cstate selected
02:11karolherbst: Tom^: if nvidia clocks to 1097, is the second last coloumn 35?
02:12Tom^: its either 32 or 28
02:12Tom^: it keeps jumping a bit between 1083~ and 1097 , not sure why tho
02:12karolherbst: ohh maybe nvf1 uses something differently
02:13karolherbst: maybe the readout is just messy
02:14karolherbst: anyway, can you fetch my repository and reset --hard the branch?
02:14karolherbst: git fetch origin
02:14karolherbst: git reset --hard origin/stable_kepler_reclocking_v2
02:14karolherbst: and then make in the top directory
02:15Tom^: you got to long branch names, even you get them wrong ;)
02:15karolherbst: no way
02:15karolherbst: ahh right
02:16karolherbst: it's stable_reclocking_kepler_v2
02:16karolherbst: my bad
02:18karolherbst: it will be still less, but only because that tool can't find the right cstate :/
02:18karolherbst: the blob doesn't tell us so we have to readout the clocks and stuff
02:18Tom^: clocking down when idle, https://gist.github.com/anonymous/ffc161e5d2d008dbc2ca and clocking up on load https://gist.github.com/anonymous/c30cb078beaf3dedc623
02:18karolherbst: ahhh better
02:19karolherbst: but it seems too high in general
02:19karolherbst: no problem though
02:19karolherbst: I may just misinterpret the mode 2 entries
02:19karolherbst: or wrong clock calculated
02:19karolherbst: who knows
02:20karolherbst: anyway, if you compile inside drm again you can try out the branch
02:23karolherbst: Tom^: the jumping may be because of power consumption or something :/
02:26karolherbst: Tom^: anyway, the pstate file will behave a bit odd now. It will still list the highest possible clock, but it won't clock to it if you set the pstate
02:26karolherbst: the AC line will contain something lower
02:28karolherbst: okay, and I will RE the missing bit for your voltage table then...
02:43karolherbst: Tom^: okay, I pushed my branch with my changes, it should be good now
02:44karolherbst: well the tool should be still broken, but nouveau should be better now
02:45karolherbst: this volting stuff is so damn annoying :/
02:48karolherbst: mupuf: there is no chance nvidia tells us where to get this "speedo" value, right?
03:32karolherbst: Tom^: so do you want to create a new heaven benchmark? :D
13:13karolherbst: skeggsb: if you got time, could you have a look at this branch? https://github.com/karolherbst/nouveau/commits/stable_reclocking_kepler_to_upstream
13:13karolherbst: it's not quite ready yet, but I think I cleaned most of the stuff up already and there are also some really trivial commits
14:05Jayhost: karolherbst does this line mean anything? Workqueue: events gk104_fifo_recover_work [nouveau]
14:06karolherbst: well it means something went wrong
14:08Jayhost: I think imirkin said I might be able to use valgrind to track down this line. GPC0/TPC1/MP trap: global 00000004 [MULTIPLE_WARP_ERRORS] warp 000d [GPR_OUT_OF_BOUNDS]
14:09karolherbst: the envytools mmt valgrind
14:10karolherbst: Jayhost: https://nouveau.freedesktop.org/wiki/Valgrind-mmt/
14:10Jayhost: Okay. I've got Doom 3 and valgrind-mmt built
15:22steve555: Hi all, I've compiled the latest mesa from git, I have the latest gentoo. I have put an alias in my .bashrc file in my home folder, but 'nvgl' isn't working, I've tried various things to get it to work to no avail.
15:25steve555: Here is what I have so far in my bashrc:http://pastebin.com/qMRBDa7t
15:33pmoreau: steve555: `alias nvgl="LD_LIBRARY_PATH="/home/steve/mesa/lib64" LIBGL_DRIVERS_PATH="/home/steve/mesa/lib64/gallium/"` should be better (do not put `nvgl=` inside the double quotes)
15:35steve555: pmoreau: well I re-did that line now, and now it reads: alias nvgl='LD_LIBRARY_PATH="/home/steve/mesa/lib64/" LIBGL_DRIVERS_PATH="/home/steve/mesa/lib64/gallium/"'
15:36steve555: I'll log in and out now, and see if that fixes it.
15:39pmoreau: steve555: You could simply do `source ~/.bashrc` to re-parse it for the current terminal.
15:39steve555: pmoreau: I did a 'ldconfig' before I logged out, and I've just tested it and 'nvgl works fine now thanks.
15:39pmoreau: you're welcome
15:40steve555: but that is a useful command to remember for nextime :)
15:50Mittttens: .tell a
15:52Mittttens: java: pushbuf.c:727: nouveau_pushbuf_data: Assertion `kref' failed.
15:53Mittttens: this is happening on nve4
15:53Mittttens: with the latest drivers and mesa
19:27orbea: I'm getting segmentation faults with xorg when making a video fullscreen with mplayer and mpv, it doesn't happen every time and will either freeze the video player forcing me to kill it or crash x with a segmentation fault. It possibly happens only or just more often after resuming from pm-suspend. I'm using linux-4.4.3 and here is a paste with the xorg log and some relevent version numbers -
19:28orbea: could this be nouveau related?
19:29orbea: Or could you suggest a better place to report this?
21:55imirkin: Mittttens: file a bug, please include a stacktrace, and most ideally a way to reproduce.
21:56imirkin: orbea: try to see if this looks familiar to anyone in #xorg-devel -- this *seems* like potentially a core X bug, but perhaps nouveau is just feeding it crap
21:56imirkin: i'd specifically try pinging ajax
21:58Mittttens: imirkin i dont have the energy to do that tbh
21:58imirkin: Mittttens: ok, perhaps someone else will come across the issue and be able to debug it further
21:58Mittttens: if you're keen to reproduce it i just ran minecraft with minecraftforge
21:58Mittttens: if not then w/e
21:59imirkin: can you provide me sufficient pointers on how to retrieve the various software?
21:59Mittttens: well i used to multimc5 launcher but i'm not sure that has much to do with it
21:59imirkin: i have little to no clue what minecraft is, or minecraftforge. i have some vague recollection that microsoft bought something related to minecraft, but that's about it.
21:59Mittttens: if you use multimc5 just create an instance using version 1.8.9
21:59imirkin: you're like 30 steps ahead of me
21:59Mittttens: and then hit the install forge button under the instance settings
21:59imirkin: what software do i need to install?
22:00Mittttens: this here
22:00Mittttens: is all you need to install
22:00imirkin: ok, and then it should be apparent how to get minecraft started?
22:00Mittttens: if you're using arch it's on the AUR but i doubt it's in the repos for anything
22:00Mittttens: but afaik it's pretty easy to build
22:00Mittttens: then you just run multimc5
22:00imirkin: building things is no trouble for me
22:00imirkin: i just have no idea what to build ;)
22:00Mittttens: right click on the backgroudn and hit new instance
22:00Mittttens: build multimc5
22:00Mittttens: that's it
22:00imirkin: ok, and then i should see the crash?
22:01Mittttens: no, hit new instance, make sure you're on version 1.8.9
22:01imirkin: or do i have to do things in-game?
22:01Mittttens: then hit under instance settings "install forge"
22:01Mittttens: install the latest version
22:01Mittttens: meaning the one at the very top
22:01Mittttens: and then hit play
22:01Mittttens: pretty ez
22:01imirkin: and the crash happens....
22:02Mittttens: yeah a little while into the loading
22:02Mittttens: the crash should happen
22:02imirkin: ok great
22:02Mittttens: previously it would crash nouveau
22:02Mittttens: but when i did it with another desktop open it didn't
22:02Mittttens: it just crashed the game throwing a nouveau error
22:02Mittttens: the one memo'd you
22:03Mittttens: https://0x0.st/8yu.png and the instance stuff should look like this
22:03imirkin: i see. well i'm much too tired to act on this now
22:03imirkin: but perhaps when i have some time i'll check it all out
22:03Mittttens: i know how you feel ;)
22:04imirkin: i don't have a kepler, but hopefully it crashes the same way on my shitty shitty fermi