00:04skeggsb: yeah, i can reprod
02:44whompy: skeggsb: FYI, just booted a fresh drm-next including your atomic changes on an NVA5. Seems to be behaving fine thus far. Launched Starcraft II and clicked around for a few to test too.
02:44whompy: Anything in particular that should be tested?
03:43skeggsb: whompy: thanks for letting me know
03:43skeggsb: anything you can think of really, mode changes, suspend, hotplugging stuff etc etc
04:57mark4: well i cant use fedora anyway and now i cant use gentoo either
04:57mark4: fedora does not allow you to set the UID of a mounted partiton in fstab
04:58mark4: and gentoo no longer has an ebuild for kernel 4.5 and i will NOT use any kernel that does not have an ebuild. nor will i fuck up my system by creating one
04:58mark4: so im lost as to what to do
04:58mark4: linuxmint wont even boot on this system
04:59mark4: the most recent kernel i can use in gentoo is 4.4.30
04:59mark4: fucking depressing
05:06Lyude: weren't you just complaining in #fedora
05:09mark4: i was asking how to mount a partiton with a specific uid
05:09mark4: they said you cant
05:09mark4: i left
05:09mark4: mount -o uid=xxxxx does not work in fedora
05:09mark4: and that means my user cannot cd into ~
05:09mark4: permission denied
05:10mark4: so i will reinstall gentoo with kernel 4.4.30 which is the most recent kernel that will work
05:10skeggsb: um.. isn't that only a thing for filesystems that don't support proper permissions...
05:10Lyude: Okay, but this is #nouveau, a channel about gpu drivers
05:10mark4: because gentoo does not support 4.5 or 4.6 kernels
05:10skeggsb: [ebuild N ] sys-kernel/gentoo-sources-4.8.6:4.8.6::gentoo USE="-build -experimental -symlink" 89,949 KiB
12:26chillfan: Hey, how is power management for the gtx 780 ti going? Last I checked there was some experimental stuff in someones tree.. just wondered if that's in one of the main/testing branches yet
12:27pmoreau: The reclocking patches from Karol have been merged to drm-next, and should be found in Linux 4.10.
12:27chillfan: voltage for evga sc is what I'm thinking of
12:27chillfan: ah nice one, I'll check out 4.10 later then
12:28pmoreau: s/should/will, since the merge window hasn’t even opened up yet I think
12:29chillfan: good to know, I'll be sure to test it when 4.10 goes main line
13:20karolherbst: chillfan: skeggsb has a linux tree with those patches: https://github.com/skeggsb/linux/tree/linux-4.10
13:21karolherbst: branch: linux-4.10
13:21karolherbst: despite that name, it is a 4.9-rc2 tree though
13:45chillfan: ah right
15:57Lekensteyn: karolherbst: here you have nvidia-smi output for a Quadro K610M http://sprunge.us/CeET
16:04karolherbst: Lekensteyn: well, those low end quadros are usually too low end and those tools still produce crap, but let me take a look
16:12Lekensteyn: I can get one for a newer quadro card, but my friend is quite busy typing so I will not poke him for now
16:13Celmor: any way to find out the 'register address to read temperature' from my GPU?
16:14imirkin_: Celmor: should just work ... run 'sensors'
16:15Celmor: imirkin_, that won't work without an appriopriate driver, I have vfio-pci loaded for it and got suggested in #vfio-users to ask here if anyone would know that address so I could still read the temperature
16:15Celmor: I am passing through that GPU to a VM machine
16:17imirkin_: Celmor: ah, gotcha
16:18Celmor: and in my case the fan on the GPUs radiator (water cooling) is controlled by the motherboard
16:18imirkin_: which gpu?
16:19Celmor: GTX 980 Ti
16:19imirkin_: nvapeek 20400
16:19Celmor: what's that?
16:20imirkin_: command to run to get the temp
16:20Celmor: where do I get that binary from?
16:21Celmor: ah, AUR has it
16:22karolherbst: Celmor: there are several registers
16:22karolherbst: but yeah, that one is fine already
16:22imirkin_: if that doesn't work, we can look at something fancier
16:23karolherbst: I think the mmio protection has to be disabled as well
16:23karolherbst: Celmor: it might be, that you have to boot with iomem=relaxed
16:24karolherbst: but… what fun would it be to reverse engineer the windows driver this way :D
16:24Celmor: no problem, just give me a minute to try the above command
16:25Celmor: hmm https://pastebin.com/raw/L32gnhiK
16:25imirkin_: 27 deg C
16:25imirkin_: seems reasonable.
16:25Celmor: I could boot up VM and check
16:25imirkin_: a little cool, but you said you have a watercooling thingie
16:25Celmor: but how do I translate it?
16:25karolherbst: there is 2044c and 20450 as well
16:25imirkin_: 0x1b == 27 decimal
16:25Celmor: so just hex -> dec
16:26karolherbst: 20450 is an offset you have to substract from the 2044c value though
16:26karolherbst: so, more complicated
16:26Celmor: 0000005c 00000040
16:26karolherbst: but it seems to be able to read below 0°C
16:26imirkin_: karolherbst: er ... where do you see that?
16:26karolherbst: 0x5c - 0x40 = 0x1c
16:27karolherbst: imirkin_: what do you mean?
16:27imirkin_: i'm looking at therm/g84.c
16:27karolherbst: imirkin_: https://github.com/envytools/envytools/commit/d71cd71e4f0207e2ca6d33737f217010f4cc9314
16:27Celmor: these temperatures seem a lot higher though
16:27karolherbst: Celmor: 0x5c - 0x40 = 0x1c ;)
16:27Celmor: so they have different offsets?
16:27karolherbst: the second one is the offset
16:27imirkin_: karolherbst: ok wtvr. but 20400 still works. so who cares.
16:27karolherbst: so the sensor is able to report starting from -64°C
16:28Celmor: but the offset is fixed, right?
16:28karolherbst: well, who knows
16:28karolherbst: it's in the reg, so who cares
16:28karolherbst: anyway, until you live at the north pole and really cool your GPU, I guess 20400 is fine
16:28Celmor: how would I best call this 'nvapeek 20400' in C ?
16:29karolherbst: link against libnva
16:29Celmor: what does it provide?
16:29karolherbst: check out this file: https://github.com/karolherbst/envytools/blob/master/nva/nvapeek.c
16:29karolherbst: you need to do something like that
16:29karolherbst: just simplier ;9
16:30Celmor: 'link against linva' hmm, I know how to include stuff, but 'link against'?
16:30karolherbst: checkout nva_rd
16:30karolherbst: gcc -lnva
16:31Celmor: not libnva?
16:31karolherbst: the lib is omitted
16:31imirkin_: echo $((0x`nvapeek 20400 | tail -1`))
16:31imirkin_: Celmor: --^
16:31karolherbst: well, he asked for C
16:31karolherbst: uhh, system()
16:32imirkin_: oh, in C as in ... from a C program. not in degrees C
16:32karolherbst: maybe you are right
16:32karolherbst: no clue
16:32Celmor: the less calls the better, if possible I'd like to incorporate that into my i3status (bar)
16:32karolherbst: Celmor: can't you include shell scripts?
16:33imirkin_: er actually, i guess i misremembered how nvapeek outputs
16:33imirkin_: should be more like
16:33imirkin_: echo $((0x`nvapeek 20400 | cut -d ' ' -f 2`))
16:34Celmor: https://pastebin.com/ffZt8UMA vs. https://pastebin.com/PcD1d393
16:34Celmor: karolherbst, ^
16:34Celmor: well, the first one does more
16:35Celmor: I still need to read stdin in the second one
16:35imirkin_: either way, this thing needs root
16:35imirkin_: so you probably want to make it a binary
16:35imirkin_: so you can suid it
16:38Celmor: where do I find nva.h?
16:39Celmor: so I need the git clone and not jost the compiled thing
16:39karolherbst: doesn't the package install those files?
16:39imirkin_: ideally it would have been installed
16:39karolherbst: might be nva/nva.h
16:39Celmor: where though?
16:39imirkin_: but you may be the first user in the history of envytools to want to link it via an external program
16:39imirkin_: so that use-case may not have been fully thought-out
16:39karolherbst: uhh, I did it already
16:39karolherbst: and mupuf even before me
16:39Celmor: I merely tried to compile https://github.com/karolherbst/envytools/blob/master/nva/nvapeek.c
16:40imirkin_: karolherbst: via a properly installed version of the package?
16:40karolherbst: Celmor: yeah, that won't work
16:40karolherbst: imirkin_: sure
16:40imirkin_: karolherbst: or as part of envytools source tree?
16:40Celmor: how do I get it to work?
16:41karolherbst: imirkin_: https://github.com/mupuf/pdaemon_trace/blob/master/pwr_read/pwr_read.c
16:41karolherbst: ahhh #include <envytools/nva.h>
16:41karolherbst: there we go
16:42karolherbst: uhh, what compiler version do you have?
16:43Celmor: gcc version 6.2.1 20160830 (GCC)
16:43karolherbst: I guess copy paste mess up
16:43karolherbst: downlaod the raw file
16:45Celmor: I guess I would need these functions too? https://pastebin.com/raw/GExQcuDg
16:45Celmor: or 'symbols'
16:46Celmor: additionally to -lnva ?
16:46Celmor: same output
16:46karolherbst: let me thing
16:47karolherbst: doesn't have the -l come after the source?
16:47karolherbst: I know that gcc is sometimes picky
16:49Celmor: more errors
16:50karolherbst: -lpiceaccess maybe
16:50karolherbst: no clue, have to go anywa
16:51Celmor: /usr/bin/ld: cannot find -lpiceacces
16:51Celmor: thanks for the help so far though
16:52imirkin_: correct spelling required
16:55imirkin_: did you do what karol suggested
16:55imirkin_: and move the nvapeek.c thing to the front of the gcc command?
16:55Celmor: what do you mean?
16:55imirkin_: gcc foo.c -lfoo -lbar
16:56Celmor: I had it at the beginning from the start but moved it to the end after he said 'doesn't have the -l come after the source?'
16:56Celmor: putting it at the start does something
16:56Celmor: no errors
16:57Celmor: it works
16:58Celmor: imirkin_, thanks for your help
17:00urmet: if you link static libraries, the linker discards everything that is not used. and it does it in command line order - if the libraries come first, there are no refences to them and they get discarded
17:01Celmor: makes sense
17:02urmet: nowdays, not much
17:05urmet: memory and cpu power is cheap enough to get rid of that kind of ordering dependency
17:39Celmor: the temparature sensor has only a resolution of 1 degree?
17:57karolherbst: Celmor: yes
17:57karolherbst: Celmor: maybe there is somewhere a better one, but nvidia also only reports with 1°C accuracy
17:57karolherbst: maybe on windows it is different?
18:08mupuf: Celmor: the hw actually has a .5°C accuracy
18:08mupuf: but we do not read it
19:44Jayhost: imirkin_ any idea why I can't get mesa to recognize libdrm drmDevicePTR. I set export LIBDRM_LIBS="-I/home/ian/libdrm/install/lib/"
19:45imirkin_: you want to hand it the extra PKG_CONFIG_PATH
19:45imirkin_: you're mixing libs and includes
19:50Jayhost: Ahh. Thanks
19:54Lekensteyn: karolherbst: is a Quadro M1000M of any use? Here is the nvidia-smi log from Linux (the Performance lines are still N/A, was Xorg needed?) http://sprunge.us/UMTb
19:54karolherbst: Lekensteyn: I think we need a Tesla for sure for more info
19:55Lekensteyn: example models?
19:55karolherbst: maybe a fermi would do the trick as well
19:55karolherbst: something like the C2050
19:55Lekensteyn: what detail are you looking for?
19:56karolherbst: everything where currently is N/A
19:57karolherbst: but I highly doubt we can get anything from it anymore
19:58karolherbst: but today I will install my new SSD and my new 1TB storage :)
19:58karolherbst: cu later then
20:03hakzsam_: does someone have already extracted the gm107 gr firmwares and stored them somewhere, by any chances?
20:10mupuf: well, that was random
20:10mupuf: (talking about karol's message)
20:15karolherbst: yo, seems fast enough "Timing buffered disk reads: 1438 MB in 3.00 seconds = 478.89 MB/sec" :)
20:18Lekensteyn: if that is a PCIe card, then it is quite disappointing
20:19karolherbst: it is mSata actually
20:22Lekensteyn: anyway, fast enough so whether it is 500MB/s or 1GB/s, you would not notice :p
20:23karolherbst: uhh, I don't want to put boot on my ssd... this will be messy
20:27karolherbst: crap, I have no space left to actually move stuff....
21:33karolherbst: yo, 8sec boot time
21:33karolherbst: silly kernel needs 6 seconds alone
21:37karolherbst: and the longest userspace service is some kind of bs: 1.369s laptop-mode.service
21:38karolherbst: second place block device: 663ms dev-sdc1.device
21:38karolherbst: crappy firmware and kernel, seriously: Startup finished in 3.859s (firmware) + 5.407s (loader) + 5.915s (kernel) + 2.342s (userspace) = 17.525s
21:39karolherbst: well to be fair, the firmware also includes the firmware password input
21:41mupuf: see ;)
21:41mupuf: ok, done with the foundation work ... took way longer than expected
21:42mupuf: back to this fan insanity
22:01karolherbst: I hope it will make sense... at some point
22:02mupuf: there is no way we won't make sense of it
22:02mupuf: right? :D
22:03karolherbst: usually there is always some kind of pattern afaik, even if it doesn't make sense, it still has some kind of integrity usually
22:03mupuf: anyway, by the end of the month, I will ask for more information about this table if I really cannot make sense of it
22:03karolherbst: did you try to get the same result on another GPU?
22:07mupuf: not yet
22:07mupuf: but I went through the entire range though
22:07mupuf: until 7k, I tested every single data point
22:07mupuf: after this, it was every 50
22:12karolherbst: I guess the pattern somewhat continued?
22:14mupuf: I think these monotically increasing thing is just a bug in the sampling
22:15mupuf: how can DUTY be bigger than 32 bits?
22:15mupuf: and I cannot reproduce now
22:15mupuf: with the same values
22:15karolherbst: maybe you need to wait longer for the result
22:15mupuf: while I reproduce the normal side
22:16karolherbst: now the question is what is going on about the lower and the upper curve
22:16karolherbst: maybe also a waiting issue? and if you wait for like 10 seconds, it would all look fine and nice?
22:18mupuf: testing with 2 seconds delay now
22:18karolherbst: you had 1 second before?
22:18mupuf: no wait at all
22:18karolherbst: ohh I see
22:18mupuf: same behaviour
22:19mupuf: but if I do not have this insane shit going on
22:19mupuf: then I can check if the lower bits mean something different
22:19karolherbst: mhhh, that would make sense if the value blocks would be constant in size
22:20mupuf: arg, true, it is sometimes 3, sometimes 4
22:22mupuf: AAAAHHHH I understand
22:22mupuf: it was just libreoffice being an idiot
22:22karolherbst: ohhhhh wait
22:22mupuf: 1e3, 1e4, 1e5
22:22mupuf: and it was interpretating it :D
22:23karolherbst: okay, that explains the one thing
22:41mupuf: so, now, let's check why it was going back to 0
22:41mupuf: I am sure it is again libreoffice being stupid
22:41karolherbst: most likely
22:41karolherbst: by any chance, do you have numbers starting with a 0?
22:44mupuf: e0 -> e9
22:45mupuf: same bug
22:45karolherbst: conclusion: never use libreoffice again :p
22:47mupuf: yeah, I tried using gnuplot like I used to do ... but I could not get it to output anything sensible after doing the hex -> dec
22:47mupuf: anyway, now it looks sane-ish
22:50mupuf: now, let's try to understand if any parameter influences this ascending square waveform
22:50mupuf: first one, frequency of the PWM
22:52barteks2x: os there some overlay that has this? https://github.com/imirkin/mesa/commits/locking
22:52karolherbst: why not use /etc/portage/patches ?
22:52barteks2x: wrong channel
22:52barteks2x: I wanted to ask on gentoo channel
22:53barteks2x: anyway, now that I got reply, I can use that
22:56mupuf: karolherbst: seems to depend heavily on the PWM frequency
22:57mupuf: lol, I set it to 1Hz
22:57karolherbst: constant value?
22:57mupuf: nope, it seems to grow linearly without hickups now
22:57karolherbst: that would have been my second guess
22:57karolherbst: middle value?
22:58mupuf: well, I would have to scale to know
22:58karolherbst: is the first value more like 0, 100 or 200?
22:59mupuf:want to check that I simply did not affect the period at which the thing grows that I would miss entirely the square pattern
22:59mupuf: 0x0, 0x10, 0x20, ...
23:00mupuf: well, it looks funky
23:01karolherbst: I guess
23:01mupuf:is outputing to da_funk.csv
23:02mupuf: will take some time to see how it performs
23:12karolherbst: huh, that one disc came with SMART disabled
23:12karolherbst: the heck
23:13mupuf: ah ah ah
23:13mupuf: this is ... hmm hmm
23:13karolherbst: 135 MB/s, quite fast for a 2,5" HDD
23:14mupuf: http://fs.mupuf.org/mupuf/nvidia/6b13_sawtooth-4.png <-- that's with the frequency set to 0
23:14karolherbst: let me guess
23:14karolherbst: every 4096 it goes back to 0?
23:15karolherbst: not quite
23:15karolherbst: the heck
23:31mupuf: it just overflows, that's it
23:31karolherbst: ohhh, makes sense
23:36karolherbst: at some point I need to unmess my fstab file :(
23:51karolherbst: mupuf: :O theory
23:51karolherbst: the old curve is also just an overflow, but the oferflow happens before divisions or crap like that
23:51karolherbst: and then there is a post addition to the value, which is defined somewhere
23:52karolherbst: mupuf: does a pwm freq of 2 produce halfed values?
23:52karolherbst: and doubled teeth?
23:53mupuf: I can't set it as low as this
23:53karolherbst: I thought you set it to 1
23:53karolherbst: or 0