00:53karolherbst: Tom^: any any results?
01:51hakzsam: skeggsb_, btw, I'll try to have a look at the nvif series this weekend (esp the mesa one)
03:04Tom^: karolherbst: nope didnt test, i had a few beers in me. was tired and the sound of maybe running on 1.22V didnt sound healthy :p
03:19karolherbst: it is just +0.05V above what it runs on "normally"
03:21karolherbst: but you can always decrease the value by a multiple of 0.0125
03:33Tom^: (info.max - 0.0125) ?
03:39karolherbst: or 0.0125 * 4
03:39karolherbst: currently the voltage needs to be a value inside the voltage table
03:39karolherbst: and yours goes like this: -- Mode GPIO, Base voltage 825000 µV, voltage step -12500 µV, acceptable range [825000, 1212500] µV --
03:40karolherbst: don't know why the step is sometimes negative, but nouveau does the right thing
03:59Tom^: hm SSE register return with SSE disabled , im not allowed todo float operations?
04:07karolherbst: Tom^: not in the kernel
04:07karolherbst: it is possible, but messy
04:08karolherbst: Tom^: what did you try to do?
04:09Tom^: i need breakfast before anything else, and mostly coffe.
04:10karolherbst: Tom^: well the voltage is usually stored as uv
04:21Tom^: oh right so info.max is essentially 1212500
04:37RSpliet: no, no floats in kernel, you don't need them even if you think you do
04:53snacks: hey, apparently someone got Nouveau working with freesync, but I can't find a source or whether it actually happened.
04:53snacks: did it actually happen?
04:53snacks: if so, does anyone have a source?
05:11agpu: hi. im trying to get nouveau to work with my 860m gpu (GK104), but it doesnt even seem to start correctly: https://paste.fedoraproject.org/295362/14487162/
05:12agpu: is there something i can do about it?
05:12karolherbst: agpu: yes
05:12karolherbst: known issue
05:12agpu: that was fast lol
05:12karolherbst: there is a quirk only some (or maybe all?) mobile kepler cards need
05:13agpu: which is?
05:13karolherbst: agpu: try it out with nouveau.config=War00C800_0=1
05:18chillfan: hey, i'm using the nouveau driver and getting a black screen at boot (the xserver starts without issue) and can no longer drop to a console, any instructions on how to fix this? I guess it has to do with the framebuffer or modesetting, just wondering if there's some link for explicit instructions about different modes or disabling that
05:18karolherbst: chillfan: dmesg and xlog please
05:18chillfan: one moment :)
05:21RSpliet: snacks: doesn't ring a bell, sorry
05:21chillfan: is there some place to pastebin that which will accept files as an input?
05:22agpu: karolherbst: i can just put "options nouveau config=War00C800_0=1" into a file in /etc/modprobe.d/nouveau.conf right?
05:22karolherbst: agpu: no, you forgot the dot
05:22chillfan: nvm i'll copy/paste just a sec
05:22karolherbst: but even then you need to update your initramfs
05:22Tom^: and include that file into it
05:23chillfan: this is my dmesg http://pastebin.com/xpXY3kfR
05:23Tom^: which isnt done by default :p
05:23chillfan: will just put the xorg.log in there too
05:23karolherbst: agpu: ohhh no
05:23karolherbst: it is fine
05:23karolherbst: agpu: my fault
05:23karolherbst: but update initramfs ,)
05:24karolherbst: chillfan: dmesg looks fine
05:24chillfan: and the xorg log http://pastebin.com/csCHQMsL
05:24chillfan: yeah it's just a black screen for the terminal, xorg is working okay just not the console
05:24chillfan: card is nvidia gtx 780 ti
05:25karolherbst: chillfan: you mean you get a blakc screen after login?
05:25karolherbst: chillfan: are you using plasma5 by any chance?
05:25chillfan: karolherbst: around the time that modules start to load everything goes black, can't see the console coming up anymore, if i hit ctrl + alt + f1 i'll not see a console
05:25karolherbst: chillfan: ohh okay
05:25chillfan: no, nothing special, openbox and xorg from debian jessie (well devuan, but more or less same)
05:26chillfan: if i try to drop down to console, i just see a frozen image of the xserver
05:27chillfan: kernel is 4.1.13
05:27Tom^: sounds like something that is fixed already because i dont see any such problems on my 780ti :o
05:27karolherbst: chillfan: 780 ti?
05:27karolherbst: might be because of old kernel
05:27chillfan: vendor is evga sc
05:27karolherbst: and something fixed it later
05:27chillfan: oh i see,
05:27chillfan: Tom^: which kernel do you use? maybe I'll try and report back :)
05:28karolherbst: chillfan: there wasn't mach chance to test on 780 ti cards
05:28karolherbst: because for some reasons, most of the 780 ti owner just used the nvidia driver :D
05:28chillfan:hides a guilty look
05:29karolherbst: the 780 ti card is a bit different compared to other keplers
05:29karolherbst: or maybe not
05:29karolherbst: but it is something which isn't well tested
05:29Tom^: as long as i get the card to 0f with proper volt i see no usage of the blob.
05:29chillfan: yeah i thought it might be, as it didn't seem to have the same code name
05:29karolherbst: Tom^: well
05:29karolherbst: Tom^: you can only expect around 70% perf
05:30chillfan: 70% might be enough for me depending what i test
05:30chillfan: or use
05:30Tom^: karolherbst: well now, nouveau on 0a ran games better in wine-nine then nvida blob did on same perf levels.
05:30Tom^: karolherbst: so thats left to be tested. :p
05:31chillfan: 0a, is this something to do with reclocking?
05:31karolherbst: Tom^: yeah, but this is no fair comparison ;)
05:31karolherbst: chillfan: yeah
05:31Tom^: chillfan: mm
05:31karolherbst: chillfan: mid perf level
05:31chillfan: karolherbst: interesting, i'll have to check the docs for that
05:31karolherbst: which docs?
05:31Tom^: chillfan: its a bit work in progress on the 780ti
05:32chillfan: like er, enabling instructions
05:32chillfan: Tom^: which is okay if it does something for now, if i get enough fps in just a couple of games i'm happy
05:33karolherbst: chillfan: well on the highest pstate you can expect 70% performance compared to the blob
05:33chillfan: glad to see the xserver start for my card, thought it was a no go when i got black screen
05:33karolherbst: but for this, you really need the 4.4 kernel or self compiled ;)
05:34Tom^: and it wont quite yet run on highest pstate because of being to low volt :p
05:34karolherbst: chillfan: ohh so it works with a newer kernel?
05:34chillfan: karolherbst: the xserver starts
05:34chillfan: in 4.1 kernel
05:34chillfan: haven't built 4.3 yet
05:35chillfan: guessing i should be on the edge to get the proper experience
05:36agpu: it works! :D
05:37karolherbst: agpu: nice
05:37chillfan: honestly though, i expect console issues with anything to do with nvidia, it has always hated the frame buffer
05:37agpu: thanks karolherbst
05:37chillfan: for all cards i've tried
05:37karolherbst: agpu: we need some info no
05:37karolherbst: agpu: so we can enable that quirk upstream
05:37karolherbst: chillfan: lspci -s 01:00.0 -vv
05:37agpu: ok np what do you need
05:37karolherbst: I hope the gpu is at 01:00.0?
05:38karolherbst: ohh wrong
05:38chillfan: wrong name?
05:38chillfan: my gpu is on that
05:38karolherbst: lspci -s 01:00.0 -nnv
05:38karolherbst: second line
05:38karolherbst: there is something like [number:number]
05:38karolherbst: this is what I need
05:39chillfan: from me or agpu?
05:39chillfan: k got confused :D
05:39karolherbst: yeah, my mistake
05:39karolherbst: agpu: lspci -s 01:00.0 -nnv
05:39chillfan: no worries
05:40chillfan: should my kernel maybe contain anything other than the standard options nouveau enables?
05:40chillfan: i generally turn off all other framebuffer stuff
05:40Tom^: nouveau.pstate=1 :p
05:40chillfan: just to check i'm not misconfiguring it i mean
05:40karolherbst: chillfan: well you need the nouveaudrmfb ;)
05:40Tom^: oh you mean the kern configs
05:41karolherbst: chillfan: but I doubt you can disable it
05:41karolherbst: chillfan: anyway, dmesg output looks fine
05:41karolherbst: chillfan: I just assume it is something messy
05:42karolherbst: "(II) NOUVEAU(0): Setting screen physical size to 508 x 285" is this normal?
05:42chillfan: hm no
05:43karolherbst: yeah well I don't know if these are dots or inch or whatever :D
05:43chillfan: xrandr says display size is 1920x1080
05:43karolherbst: chillfan: end of "connected primary" line
05:43karolherbst: I think this size is in mm
05:43karolherbst: so it shall be fine
05:43agpu: karolherbst: https://paste.fedoraproject.org/295365/87182151/
05:43karolherbst: chillfan: right, maybe give me your xrandr output
05:44karolherbst: agpu: ahh also a clevo
05:44karolherbst: damn a 870m :/
05:44agpu: why? ^^
05:44karolherbst: agpu: no, it is just nearly the same I have most likely
05:44chillfan: i think it should be using 144Hz too as per the xorg.conf
05:45karolherbst: chillfan: ohhhh
05:45chillfan: but i guess one step at a time
05:45karolherbst: you have a xorg.conf?
05:45karolherbst: please paste that too
05:45chillfan: hah will do
05:47karolherbst: agpu: clevo P157sm?
05:48karolherbst: chillfan: there seems to be nothing important in that xorg.conf file
05:48karolherbst: chillfan: remove it then :p
05:48karolherbst: I guess you have the nvidia stuff removed or disabled or something?
05:48agpu: karolherbst: no, P150SM-A
05:48karolherbst: agpu: ohh this one
05:49chillfan: probably should, yeah nvidia driver made the file
05:49chillfan: i just haven't tidied it up yet
05:49chillfan: or removed it as i might be able to, was leaving it in case it sets my refresh for me
05:49agpu: karolherbst: its a rebrand actually. but i think it doesnt make a difference
05:49chillfan: anyway just built 4.3 so might boot to that in a sec
05:50karolherbst: agpu: did you build the kernel or nouveau yourself?
05:50karolherbst: agpu: no,not really
05:51agpu: karolherbst: i installed the kernel + mesa from fedora rawhide Linux fedora 4.4.0-0.rc2.git2.1.fc24.x86_64
05:51karolherbst: then it doesn't matter
05:52agpu: i thought it would help to update because there was a lot of news around nouveau currently
05:52karolherbst: ohh me
05:52karolherbst: one line and I really did a mistake
05:53chillfan: ok I can see a new option in linux 4.3
05:53chillfan: CONFIG_DRM_FBDEV_EMULATION for xfree86 dri
05:53agpu: karolherbst: should i test reclocking?
05:54karolherbst: agpu: yes
05:54chillfan: is that maybe something to do with it? I guess I want this disabled since I don't care about it, but maybe that's the feature i was missing?
05:54karolherbst: agpu: http://lists.freedesktop.org/archives/nouveau/2015-November/023397.html
05:54karolherbst: agpu: when this lands, you don't need the option anymore
05:54karolherbst: chillfan: no clue
05:55chillfan: well only one way to find out i guess, brb then :)
05:56agpu: hehe nice
05:56chillfan: well, same issue for now, will tinker with some options see what changes
05:57karolherbst: chillfan: did you remove the xorg.conf?
05:57chillfan: let me try that brb
05:59agpu: i will reboot with reclocking enabled brb
06:00chillfan: ok, with that removed still the same
06:00chillfan: bear in mind, xserver totally working
06:00karolherbst: so just the dispaly is black?
06:00chillfan: no display isn't black, i talk to you from xserver now
06:01karolherbst: ohh okay, so what is then the issue? :/
06:01chillfan: but you know when you boot and you have the console, lines with [ ok ] or [ done ] etc
06:01chillfan: that is what's black
06:01chillfan: so if i stopped the xserver, i'd lose access to the display
06:02karolherbst: not even a blinking cursor in the top left corner?
06:02karolherbst: does the display go off or is there just nothing on it?
06:02chillfan: on the 3.18 kernel i was using previously, the monitor reported no signal
06:02chillfan: but on 4.1 monitor is clearly getting a signal but no display
06:03chillfan: and same currently on 4.3
06:03Tom^: are you using hdmi or displayport?
06:03chillfan: i am using dvi-d
06:03Tom^: then *shrug* because i use dvi too :<
06:03Tom^: heck even two monitors and they both get mirrored as fast as kernel initramfs is loaded
06:04chillfan: well my usual suspect for things like this is modesetting or fb driver
06:04karolherbst: chillfan: you don't have any kernel command line or something?
06:05Tom^: or does fedora try to load some bootsplash that isnt showing?
06:05chillfan: karolherbst: what i see is the kernel output right up until the point the nouveau module loads
06:05chillfan: Tom^: i'm using devuan (debian with sysvinit)
06:05chillfan: basically yes
06:06Tom^: no i meant *debian to my fedora
06:06chillfan: oh hah, no there's no graphical splashy thing
06:06agpu: is AC the current power level?
06:07chillfan: the AC?
06:07karolherbst: agpu: current clock
06:07chillfan: oh sorry my mistake, thought that was directed at me heh
06:07karolherbst: agpu: but this line also display whether you are on battery or not
06:08chillfan: ok just made some changes brb :)
06:08karolherbst: when it is 0, the gpu is off
06:08agpu: lol makes sense haah
06:09agpu: AC: core 324 MHz memory 648 MHz (by default)
06:09chillfan: ok so monitor reports no signal after nouveau driver loads, but signal comes back when xserver starts
06:10karolherbst: chillfan: yeah, then there is something odd with where the kernel wants to display the stuff
06:10karolherbst: chillfan: what cpu do you have?
06:10chillfan: intel i5 4690k
06:10chillfan: with the intel graphics disabled in bios
06:11karolherbst: the board doens't have dvi-d?
06:11chillfan: the graphics board is dvi-d
06:11karolherbst: nono, the motherboard I mean
06:11karolherbst: not the nvidia gpu
06:11chillfan: let me check
06:12karolherbst: mhh, but when the intel gpu is disabled in the bios, strange
06:13karolherbst: chillfan: usually for video accell and stuff you are far more better of with the intel gpu for general usage
06:13agpu: looks like it worked but im not 100% sure
06:13agpu: 0f: core 405-967 MHz memory 5000 MHz AC DC *
06:13karolherbst: chillfan: prime offloading should also work on a desktop machine
06:14karolherbst: chillfan: maybe the kernel does try to put the signal somewhere else
06:14karolherbst: chillfan: you might want to just plug the monitor into different ports or something
06:14chillfan: hm the motherboard supports dvi-d, the card is dual head, perhaps.. it mistakes?
06:15Tom^: agpu: the line under that reports current. that * is just not to be trusted :p
06:15agpu: AC: core 966 MHz memory 5000 MHz
06:15agpu: so it did work
06:15agpu: i guess
06:16chillfan: well just checked now, no signal at all on the other dvi-d
06:16Tom^: indeed, now it just depends on if you got high enough voltage for that clock or not. for me i didnt have it. :p
06:16karolherbst: Tom^: ohh he will for sure
06:16karolherbst: Tom^: on PWM you just get max voltage if you go beyond what is supported
06:17karolherbst: but those cards hardly ever use above 1V
06:17chillfan: ok i'll try to work around it by disable modeset
06:17karolherbst: chillfan: no
06:17agpu: why doesnt my fan go to full rotation though?
06:17Tom^: karolherbst: pwm?
06:17karolherbst: chillfan: disabling modeset is not a fix, nor a workaround
06:17karolherbst: it is: disable accell on my gpu
06:17karolherbst: so no opengl at all ;)
06:18karolherbst: I have no idea what could causes that, maybe imirkin knows more about this
06:19karolherbst: Tom^: yeah well, later gpus aren't volting through GPIOs anymore, but a pwm
06:19Agiofws: image link: https://i.imgur.com/7KuT4oa.png
06:19Tom^: karolherbst: ah ok.
06:19karolherbst: Tom^: some keplers and all maxwells
06:20Tom^: chillfan: you dont have an optiom like this that sets primary GPU ? https://i.ytimg.com/vi/BtDkczrf5fI/maxresdefault.jpg
06:20chillfan: Tom^: no i just disable the intel stuff and all works
06:24karolherbst: chillfan: you should know that you only get very poort video acceleration on your card
06:24karolherbst: with nouveau
06:24Tom^: karolherbst: this is what im thinking https://gist.github.com/anonymous/4d727ec1e5bed7158254
06:25chillfan: alright, i'll have a try and see what happens anyway
06:25chillfan: more of a test to see what i can get out nouveau atm anyway
06:27chillfan: https://wiki.freedesktop.org/nouveau/KernelModeSetting/ this seems to suggest i have a misconfigured kernel
06:28chillfan: can i possibly do without fb?
06:28chillfan: ah nvm i'll tinker with it, starting with fb on
06:33chillfan: nouveau.pstate=1 should be in lilo/grub?
06:34chillfan: bbs :)
06:37agpu: i somehow managed to break it
06:38agpu: no it looks like DRI_PRIME doesnt work anymore ^^
06:41karolherbst: agpu: yeah well, that happens sometimes
06:41karolherbst: reclocking isn't rock stable currently
06:42karolherbst: but this looks like voltage too low
06:42karolherbst: agpu: may I have your vbios?
07:16karolherbst: Tom^: did you try it out with a higher voltage?
07:20sdfgdgd: someone just gave me a nv34 device but i tried using it and its very slow and i saw in the feature matrix that all power mangement its missing. Is there anything that could be done to help with it?
07:21karolherbst: mupuf_: what do you think about a NvChipQual config options, which accepts values for 0-100 and determines which voltage nouveau will use. Default will be something like 25, which means nouveau will use (volt.max * 75 + volt.min * 25) / 100 as the voltage
07:21pmoreau: mwk: Ping for #11, #15 (and why not #27 as well, but preferably the first two ones O:-) )
07:57mupuf_: karolherbst: you want to add a workaround instead of spending the time to study voltage management?
08:04chillfan: so i think i found what is wrong
08:04chillfan: as it's now working
08:05chillfan: i think the nouveafb driver expects an fbdriver already there and loaded
08:05chillfan: is it not designed to work with just 'vga text console'?
08:06chillfan: nvm working, i'll have a tinker with more options later :)
08:09karolherbst: mupuf_: not really
08:09karolherbst: mupuf_: I was thinking that if a gpu has a higher quality, then you need a lower voltage for a specific clock
08:09karolherbst: it would be an option which would enable higher cstates
08:10karolherbst: if you spend some time yourself figuring out how far you can go
08:15karolherbst: mupuf_: by the way, do you think these unk0/unk1 values might be power consumption related? https://gist.github.com/karolherbst/28fcfc36013873249077
08:15karolherbst: like in 0.01W
08:15karolherbst: the dellm6800 has a power budget of 100W
08:15karolherbst: mine has 80W
08:17karolherbst: sadly these are the only two vbios I found with values there :/
08:18mupuf_: karolherbst: yes, that would be a nice knob, but who would like to do the validation?
08:19karolherbst: I have no idea, I have to work out why nvafakebios doesn't work for me :/
08:19karolherbst: second: the blob always clocks to max for me (way beyond the turbo clock"
08:19karolherbst: base 705, turbo 797, max 862 and blob always choose max
08:20karolherbst: would be perfect to test with faking vbios without risking anything :/
08:20karolherbst: but well
08:21mupuf_: I would say try to debug it with nouveau
08:22karolherbst: mupuf_: when I nvafakebios
08:22karolherbst: should nvagetbios return the faked one?
08:23mupuf_: if you use pramin
08:25karolherbst: mupuf_: "Invalid checksum. Broken vbios or broken retrieval method?"
08:25karolherbst: when trying pramin after faking
08:25mupuf_: what the heck!
08:25karolherbst: "Invalid signature(0x55aa). You may want to try another retrieval method."
08:26karolherbst: but there seems to be something
08:27karolherbst: and my changes are thre too
08:29karolherbst: mupuf_: after 0x10000 the uploaded vbios and the read out vbios differ
08:29mupuf_: wait a sec, we may manage to make it work
08:30karolherbst: and with differ I mean, they differ for real :D
08:30mwk: pmoreau: I'll do that now; remind me, is there an ordering to these PRs?
08:30mwk: as in, should I do 11 before 15 or the other way around?
08:33mupuf_: karolherbst: something is writing on the vbios
08:33mupuf_: we need to change the vga rom base address
08:33mupuf_: oh my, the 2d acceleration on maxwell is funky
08:34mupuf_: glamor does not seem perfect at all :D
08:34karolherbst: mupuf_: okay, so what does that then mean?
08:34imirkin: gnurou: ah ok. well it would save me some time if you could document the new descriptors... otherwise i'll have to find time to RE it
08:34karolherbst: mupuf_: it doesn't look like random data though
08:35karolherbst: mupuf_: it looks like this: https://gist.github.com/karolherbst/ed5051e63a3116456113
08:35mupuf_: karolherbst: http://pastebin.com/4S5FbZ2F
08:35mupuf_: this is on my nve6
08:35imirkin: gnurou: which probably won't be until mupuf gets a GM20x into reator (or i happen to get one)
08:35mupuf_: see, the blob reads from the ROM_WINDOW
08:35mupuf_: in my case, it reads beef
08:35mupuf_: and then it gives up
08:36mupuf_: and tries reading from PROM instead ... and reads 55aa
08:36mupuf_: or aa55
08:36karolherbst: I know that 55aa
08:36karolherbst: thats a checksum
08:36imirkin: no, it's the pci rom header
08:36mupuf_: nope, that's not a checksum
08:36karolherbst: no? mhh
08:37karolherbst: mhh okay
08:37karolherbst: I thought I already saw that number, oh well
08:37gnurou: imirkin: my goal is to make things easy for you guys, so I'll try to document as well. no promises though as it is not my decision to do so.
08:38mupuf_: so, I would like you to mmiotrace the blob after nvafakebios to make sure it picks up the right bios and that the checksum indeed fails
08:38mupuf_: and then, we can be crazy and try to pick up another linear address to upload the bios to
08:38imirkin: gnurou: ok, well as you can imagine, my priority is based around real use-cases, and for now, GM20x is not a real use-case for nouveau. when it becomes one, i'll spend some time on it :)
08:38karolherbst: I hope mmiotracing works now again
08:38mupuf_: karolherbst: does that sound crazy-enough for you?
08:39karolherbst: no, I have no clue how that pramin stuff works or what it is, so everything sounds like a good idea to me anyway
08:41Yoshimo: imirkin: there are many cards that need work, you can't do them all
08:41imirkin: Yoshimo: meh, it's not so bad -- they tend to be in (large) groups
08:42imirkin: Yoshimo: it's just that there's no way to get GM20x working with nouveau, so i don't need to worry about it
08:42Yoshimo: i am always in favor of doing one group of cards after the other
08:43gnurou: imirkin: I understand and apologize for the inconveniences we are introducing. I also hope you understand that there are several nvidians fighting for your side. ;)
08:43imirkin: gnurou: thanks :) and i'm in no way blaming you... merely explaining my position.
08:44gnurou: imirkin: fully understand that. we have the same goal.
08:45karolherbst: mupuf_: thing is, my laptop went a bit crazy after I tried to fake the vbios, maybe it was bbswtich related or something, because it always happens after I power on my gpu
08:45pmoreau: mwk: Sorry, hadn't seen you had pinged me. The renaming should be done first, as the EVO relies on it (IIRC).
08:45mwk: pmoreau: whoops, I already started going through the EVO ones
08:45mwk: I'll review these commit-by-commit FWIW
08:45gnurou: imirkin: and we probably share more frustrations than you think ;)
08:46mupuf_: karolherbst: that's what happens when you write over memory of another process :p
08:46pmoreau: mwk: Yeah, I saw your comment: this is what made me realise I should check IRC. :-D
08:46pmoreau: mwk: I thought I had removed all those names using a '.' in them, but looks like some survived. :-/
08:49karolherbst: mupuf_: I don't know, somehting really bad happens
08:49karolherbst: mupuf_: display turns black, networks seems to be down and sysrq keys doesn't work anymore
08:49karolherbst: but there is also no crash in pstore
08:50mupuf_: karolherbst: do you understand what is happening?
08:50mupuf_: imagine you are writing over some memory that is not yours
08:50karolherbst: i see
08:50mupuf_: now imagine if it was kernel memory
08:50mupuf_: what could go right, right?
08:51karolherbst: so my kernel just goes into a state worse than the worst
08:54karolherbst: ohh yay, filesystem corruptions
08:55mwk: pmoreau: I have a few minor comments to the EVO patchset, it's good to merge once you fix them
08:56pmoreau: mwk: Cool! I'm working on fixing them
08:58karolherbst: mupuf_: so, what could I try out?
09:03mupuf_: karolherbst: well, you would have to allocate memory for it and get its linear address
09:03mupuf_: which means you may have to write a kernel module that does a kmalloc
09:03mupuf_: and then ask for its linear address
09:15Tom^: karolherbst: nope was streaming hockey, SHL. but im gonna do it now.
09:16pmoreau: mwk: 0x100 doesn't exist in the documentation of the EVO base channel, for none on the cards.
09:16Tom^: karolherbst: so unless you find any issues with this change https://gist.github.com/anonymous/044bb05ead2a3bd1bd75 im going for it! i hope it wont burn up :p
09:18mwk: pmoreau: then the documentation is incomplete
09:18pmoreau: mwk: But there is one on the Overlay channel: `SET_COMPOSITION_CONTROL`
09:19pmoreau: for all cards
09:19mwk: I've scanned all the EVO channels
09:19mwk: 0x100 is clearly something, it doesn't give the ILLEGAL_METHOD error like all the others
09:19mwk: and there's an earlier comment saying that blob pokes something into it
09:20mwk: it might be a method that nv doesn't want to document, eg. something dealing with HDCP
09:20pmoreau: Could be
09:20mwk: so leave it in, maybe with a comment that it's missing from nv docs
09:21pmoreau: gnurou, tagr: Any idea what could be method 0x100 of the EVO base channel? It is not present in the header files you released earlier this year.
09:21pmoreau: mwk: I'll do that
09:21mwk: we're talking G80, for reference
09:24Tom^: karolherbst: yea that didnt work, clk failed to raise voltage -50000
09:25karolherbst: Tom^: ohh right
09:25karolherbst: 0 - 50000
09:26Tom^: you mean info.max can be 0 at times?
09:32Tom^: karolherbst: easily fixed, if(info.max <= 0) return info.max; :p
09:32Tom^: now im at 1.18 V
09:32Tom^: time to see if it borks
09:35Tom^: karolherbst: hm however it put me on 1307mhz now, so not sure i dare max it out. :p
09:40chillfan: alright, looks like all working and just bad configured kernel before heh
09:42pmoreau: mwk: :-( My push removed all your comments… For which method did you wanted me to add an 'r' access attribute? The `FREE` one? (In https://github.com/pierremoreau/envytools/commit/c7f61231db1b28237204d101ef922d58b94483ce)
09:44chillfan: so erm, how can i set the pstate, check temps, change fan control, etc?
09:50pmoreau: mwk: Thankfully, I can still see your comment by accessing the Github notification. \o/ Going to fix that one and repush
09:50Tom^: karolherbst: hm any idea why the lowest cstate available on 0f is 40 which still is a bit to high mhz.
09:57karolherbst: Tom^: yes, because you messed with the voltage calculation ;)
09:57pmoreau: mwk: I should have fixed all of your comments for #11. :-)
09:57pmoreau: Thank you for your review!
09:59karolherbst: Tom^: but it should work fine
09:59Tom^: karolherbst: well unigine heaven ran at stable 60 - 80 fps at high settings on 1920x1080 before freezing. CTSX_TIMEOUT or similiar
10:00Tom^: so before i consider it a bug somewhere i rather be at max clocks i know works. :p
10:00mwk: pmoreau: strictly speaking FREE is *not* a method, since it's read-only, as opposed to write-only :)
10:00mwk: pmoreau: good job, merge it
10:01mwk: I don't see any dependency on 15, so let's get it over with
10:01pmoreau: mwk: Oh! Great!
10:01Tom^: karolherbst: does the driver control the fan or is that internal? because i didnt notice it speedup any before it froze. perhaps it simply just overheated?
10:01pmoreau: mwk: Well, there is an implicit dependency since it uses the BLOCKLINEAR notation from #15. ;-)
10:01karolherbst: Tom^: mhh I guess the fan turns up, because the gpu crashed
10:02Tom^: well no it didnt speedup at all
10:02mwk: pmoreau: meh :)
10:02Tom^: is what i meant
10:02karolherbst: Tom^: I doubt it overheated though
10:02Tom^: yea well perhaps misguided term, but more of to hot for these volts and mhz.
10:02Tom^: at 100% usage :P
10:03karolherbst: you could monitor it through hwmon, but I doubt that it was heat related
10:03Tom^: because it ran for quite a bit, compared to when i didnt have volts raised where it simply froze instantly
10:05karolherbst: it could be that something is still missing somewhere
10:05karolherbst: Tom^: could you try to run the gpu at the same clock as under windows?
10:05karolherbst: or the blob
10:05Tom^: well i cant, because lowest cstate is 40 which in turn is 1164 mhz
10:05Tom^: ~80mhz more then blob/stock at highest boost state
10:06karolherbst: why is 40 the lowest ?
10:06chillfan: hm does nouveau use it87?
10:06Tom^: karolherbst: you said it yourself, because i broke my volt calculations. xD
10:07mwk: finally :)
10:07karolherbst: Tom^: yeah, then unbreak it :p
10:07Tom^: karolherbst: not sure, https://gist.github.com/anonymous/592e0117a8b34370c74c
10:07mwk: now, #15...
10:07Tom^: but then im back on to low volt, and it freezes because of that
10:07pmoreau: mwk: Yup! I was wondering whether it was going to celebrate its first birthday or not. :-D
10:08karolherbst: Tom^: you could just remove the volt adjustement
10:08karolherbst: or check if the result is 0 before doing so
10:08Tom^: which i am
10:09karolherbst: something seems wrong then
10:09Tom^: karolherbst: https://gist.github.com/anonymous/da379fc0f9b49879f7ad
10:10Tom^: heh seems you use tabs meanwhile i expand tabs to spaces :>
10:10karolherbst: Tom^: well kernel stuff says 8 spaces is a tab
10:11karolherbst: something seems to be really odd
10:13karolherbst: ohh maybe I did something wrong :o
10:13Tom^: \o/ if so im glad my experiments made you find it.
10:19karolherbst: Tom^: would you like to boot with config=volt=debug and give me the dmesg output?
10:20Tom^: with my quirk?
10:20karolherbst: doesn't matter
10:23karolherbst: Tom^: I think I know what is wrong
10:23karolherbst: but I want to confirm this with your dmesg
10:23Tom^: uhm wasnt it nouveau.config=volt=debug
10:24Tom^: because i dont see any changes in dmesg https://gist.github.com/anonymous/94da39948e75503b54ae besides my cmdline
10:24karolherbst: ohhh right
10:24karolherbst: I am dumb
10:26Tom^: now :p https://gist.github.com/anonymous/683066d2b455ce69e0da
10:29karolherbst: this looks totally fine though: min: 825000uv max: 1212500uv
10:35karolherbst: Tom^: could you open drm/nouveau/nvkm/subdev/clk/base.c
10:35karolherbst: there should be a line with "if (volt && (voltage > volt->max_voltage || voltage < volt->min_voltage))"
10:36karolherbst: remove that if for now and reboot with those changes
10:36karolherbst: also the return after that if has to be removed
10:38Tom^: i hope this wont put me in some insane cstate when i got to pstate 0f :p
10:42Tom^: karolherbst: sure that gave me the cstates back just that it complains about failing to raise voltage -37500 and i suspect its because of my quirk. it cant reduce the voltage :p
10:43karolherbst: then there is still something fishy then
10:44Tom^: i mean im going to 0f which puts me in a high cstate with higher voltage, then i try to put it to cstate 35, which should reduce it a bit. but my quirk foobars the calculations
10:44karolherbst: yeah, but your change messes something up
10:44mwk: pmoreau: I've gone through #15, there are also a few fixes, mostly related to the "unit" word
10:44mwk: which should be block
10:45karolherbst: Tom^: ohhh I know
10:45pmoreau: mwk: Thanks a lot! I'll have a look at your comments after dinner. :-)
10:46karolherbst: Tom^: instead of if(info.max <= 0), do if(info.max <= 875000)
10:46karolherbst: and then revert this change in clk/base.c
10:47Tom^: or perhaps even info.max <= info.min ?
10:51Tom^: karolherbst: sort of worked just that im on 1.21V now. you ruined my quirk ;_; =D
10:52karolherbst: select a proper cstate
10:52karolherbst: Tom^: well this function calculates the voltage used for a voltage id, it doesn't effect the max voltage used at all
10:52Tom^: yea i did, which is what put me on 1.21 too :p
10:52karolherbst: select a lower one then :p
10:53Tom^: but my mhz :<
10:54Tom^: oh well there we go, 1088mhz and 1.18v
10:54Tom^: time to see if it freezes or not
10:56Tom^: karolherbst: http://i.imgur.com/hAG4ILA.jpg
10:57karolherbst: mhhh 77
10:57karolherbst: what settings?
10:57Tom^: however im gonna have to see what temps this card usually is on because i aborted at 87C when it still didnt had increased fan speed
10:57karolherbst: yeah well
10:58karolherbst: that doesn't sound right
10:58Tom^: http://i.imgur.com/iUnHBZE.png settings
10:59Tom^: so awesome fps tbh, :p
11:00karolherbst: I only get 21 fps there
11:00Tom^: "Maximum temperature for a 780Ti regardless of model is 95c." so yea, fan control is borked. :p
11:03Tom^:achievment earned: world first nouveau 80+ fps.
11:03Tom^: on unigine heaven.
11:04Tom^: no idea where lm_sensors get the "high, crit, emerg" temps from but those are all way , way to high.
11:04karolherbst: yeah I know
11:04karolherbst: but the more important issue is, is that nouveau can't control your fan
11:04Tom^: is nouveau perhaps just doing the same that it doesnt find my gpu hot so it doesnt increase the fan?
11:05karolherbst: Tom^: no idea
11:05karolherbst: I have to go now anyway, maybe mupuf_ wants to help you figure this one out
11:07Tom^: yea temp_max etc are all way to high in /sys/devices/pci0000\:00/0000\:00\:01.0/0000\:01\:00.0/hwmon/hwmon2/t
11:07karolherbst: yeah but that doesn't matter much
11:07karolherbst: can you contrl the fan manually there?
11:07Tom^: sure if i set pwm1_enabled to 0 and pwm1 to the percent it should run at
11:09Tom^: hm nop, Invalid argument when trying to change pwm1
11:11karolherbst: well we can look into this tomorrow maybe
11:12karolherbst: at least I haven't touched anything fan related for now
13:02pmoreau: mwk: I pushed two new commits taking care of your comments, to make it easier to review the additional changes than if I had rebased everything.
13:03pmoreau: mwk: I'll squash them into the commits they fix (and rebase on master) before merging.
13:08koz_: If my card will only reclock stably to 0a and no higher, is there anything much I can do about it?
13:08RSpliet: koz: first thing you want to try is the freshest kernel you can possibly imagine
13:08koz_: RSpliet: So 4.4rc-whatever?
13:09RSpliet: I think that would be a good start yes
13:09pmoreau: koz_: If it's a Fermi, complain to / ping RSpliet, if it's a Kepler, complain to / ping karolherbst. :-D
13:10koz_: pmoreau: I guess karolherbst is getting a complaint when he next appears. :P
13:10pmoreau:searches for popcorn
13:10koz_: But even on 0a, the card is pretty solid (680).
13:10Yoshimo: that guy is working his ass off, cut him some slack
13:10RSpliet: Yoshimo: pmoreau wasn't entirely serious ;-)
13:11RSpliet: pmoreau_: you can complain about Fermi all you want, that's not going to make a difference :-D
13:11RSpliet: koz_: which kernel are you using then?
13:11pmoreau: Yoshimo: I know how much Karol is involved!
13:12RSpliet:throws some popcorn in pmoreau's clevage
13:12pmoreau: RSpliet: :-/ It seemed to help to get reclocking working on my G96 ;-)
13:12Yoshimo: i will join the phoronix club and complain about missing maxwell firmware then ;)
13:12koz_: RSpliet: 4.2.something IIRC. Let me check.
13:12RSpliet: hahaha, well, you get that privelege when you send me a card :-D
13:13RSpliet: koz_: then fire up a 4.4-rcX kernel
13:13koz_: RSpliet: Well, *compile*, then fire up.
13:13RSpliet: koz_: well, if you're on Fedora you could grab one from Koji :-)
13:13pmoreau: Yoshimo: Give gnurou and the other NVIDIA guys some rest. ;-)
13:13koz_: RSpliet: Arch. :P
13:14koz_: I guess we've heard nothing about the firmware recently, then?
13:14RSpliet: fair enough, anyway, 4.4 has Karol's fix in, gets it going for a lot of cards
13:14Yoshimo: eventually it will come, i have a gtx 280 and a 560ti to messwith until then
13:14RSpliet: if not, you have a weird voltage regulator and some work is somewhere in the pipeline
13:14pmoreau: RSpliet: Meh… I only have one Fermi, that you already have though. I should buy a few and send some to you, or set up a Reator. :-)
13:15koz_: RSpliet: Will I need to update Mesa and libDRM as well to match?
13:15RSpliet: pmoreau: I'm playing with my Fermi for now, once it's up and running I know there's a 85% solution
13:15RSpliet: from then, I can look at other cards :-)
13:15RSpliet: koz_: don't think so no
13:15koz_: RSpliet: OK, that's good then.
13:16pmoreau: koz_: There were some comments this morning hinting at some internal progress, but it'll probably take some more time before we see it.
13:16koz_: pmoreau: Ah. Well, I'm merely curious.
13:16koz_: I don't have any great personal investment in getting Maxwells up and going - kepler4lyf.
13:17koz_:flashes gang sign.
13:17pmoreau: I hope to make use of my 960 at some point, and of the Pascal I'll buy when they come out.
13:18RSpliet:goes and gets some drink to help me RE that mem reclocking sequence
13:18koz_: pmoreau: I have a Maxwell in the laptop I'm on, but I don't really need the power behind it.
13:18koz_: My desktop is powered by an EVGA GeForce GTX 680.
13:19pmoreau:is hoping to win in his fight against LLVM to get clover up and running again
13:21Tom^: koz_: what happends when you try to run at 0f?
13:22koz_: Tom^: It either straight-up doesn't run my monitor, or displays a hideous pixellated display with some kind of PGRAPH error.
13:22Tom^: ok yes, get 4.4
13:23koz_: OK, I'll try.
13:23koz_: My main loss is the memory clock actually.
13:23koz_: 0a gives me like, two-thirds max core clock.
13:23koz_: But like, a *quarter* max memory clock.
13:25Tom^: my only problem is the fan isnt ramping up so i cant run at 0f :<
13:25Tom^: otherwise i had 77fps on unigine heaven on high at 1920x1080
13:26koz_: Tom^: Nice! I'm trying to get Minetest on max settings, max view distance and 512 textures at 1920 x 1080 to give a consistent 60 frames.
13:27koz_: Currently, my card can *sorta* manage consistent 30.
13:53imirkin: Tom^: fan should scale up on its own, although perhaps you don't like the way it ramps up
13:53imirkin: Tom^: if you flip nouveau into manual control mode, then you can set the percentages yourself
13:53Tom^: imirkin: it doesnt, :/
13:53Tom^: imirkin: i was on 87C when i aborted unigine heaven. and still only 1050rpm
13:54imirkin: Tom^: hmmm... odd. 87C isn't so bad btw...
13:54Tom^: 90ish or so is the topmost this card runs at 95 is when things go bad.
13:54imirkin: Tom^: but i would have expected it to have spun up before then... perhaps fan speed adjustment just plain doesn't work? that'd def explain it...
13:55Tom^: and yea i tried setting pwm1_enable to 0 , however changing pwm1 just gives me invalid argument
13:55imirkin: mmmm... iirc you have to echo 1 into it? there's a wiki page.
13:55imirkin: or rather... docs
13:57Tom^: the thing is tho is that temp1_max , temp1_crit and temp1_emergency is way above what it should be
13:57Tom^: so wouldnt surprise me if thats the only issue and it then thinks the card is cool enough while it isnt.
13:59Tom^: but then again the only values i find is the ones in static void nvkm_therm_fan_set_defaults in nvkm/subdev/therm/fan.c and those seems reasonable.
14:02imirkin: those are all read from vbios
14:02imirkin: perhaps incorrectly
14:05imirkin: gnurou: i know at least one frustration that i have and you don't -- docs :)
14:06imirkin: gnurou: and perhaps even more importantly, notes on what's changed generation to generation
14:07imirkin: mupuf_: i assume you don't have any GM20x?
14:08Tom^: ok i left it running now, and once it hit 90C the fans went mad and did go 100% for a second or two until it dropped down. to ~84C
14:09imirkin: there's supposed to be a ramp
14:09imirkin: it's probably misunderstanding what's going on
14:10Tom^: yea atleast it can control the fan. heh
14:10imirkin: this is mupuf_'s favourite area... make the vbios available to him somewhere
14:10Tom^: "[ 941.444992] nouveau 0000:01:00.0: therm: temperature (90 C) hit the 'fanboost' threshold"
14:10imirkin: yeah, that's one of the emergency measures
14:10imirkin: but normally fan speed is supposed to scale so you don't hit that
14:10imirkin: at least... i think
14:11Tom^: what timezone does he live in?
14:12Tom^: so i know when to be online heh
14:15imirkin: GMT+2 or so... helsinki
14:18imirkin: koz_: there's an important patch in 4.4-rc1 which should improve your reclocking experience greatly
14:19Tom^: imirkin: btw http://i.imgur.com/h8wN7Nv.jpg =D =D those fps.
14:21imirkin: Tom^: that's 76 more than i get...
14:21imirkin: btw, if you want to see an awesome effect, enable 8x msaa
14:21Tom^: pstates make one hella difference, on 0a im only getting like 25
14:21imirkin: yeah, your gpu also cost about 20x more than mine
14:22Tom^: woah its like being on drugs
14:23imirkin: whoa wait, that's 77 fps at 1920x1080? i was talking about the fps i get with 640x480 :)
14:25Tom^: i think its time to give in, the temperature table from the vbios aint telling me much.
14:35koz_: imirkin: It did.
14:35imirkin: koz_: great :)
14:35koz_: I can now get 30fps consistently even when I turn everything up to 11 in Minetest.
14:35koz_: Well, ~30fps.
14:35imirkin: i guess that's good? i don't think i've ever run minetest
14:36koz_: imirkin: It's pretty good.
14:36koz_: Given that I have a 680, which is probably the most powerful card I can run that reclocks well.
14:36Tom^: if only i get the fans working the 780ti is the one. :p
14:37koz_: Tom^: Is that the thing stopping its reclock being good?
14:37Tom^: anyways minetest looks like minecraft, how can it even run at 30fps?
14:37Tom^: its like square blocks with 12x12 textures
14:37koz_: Tom^: This is with 512 textures, max view distance, and all setting at max.
14:38imirkin: i guess there are multiple square blocks?
14:39koz_: Also, do any of you folks know what environment variable I have to set to show Open GL debug info when running (e.g.) Minetest?
14:39koz_: I remember there being an application-agnostic way...
14:39imirkin: what sort of debug info were you hoping for?
14:40koz_: imirkin: Calls. I know Irrlicht has some inefficient calling issues, so I wanna patch mine.
14:40koz_: I remember someone telling me this one, but I forgot.
14:40imirkin: mmmmmm there is something
14:40imirkin: MESA_DEBUG=api maybe?
14:40koz_: It was something like that, yes.
14:40koz_: I'll try it.
14:42imirkin: only on debug builds though
14:49koz_: imirkin: GL_DEBUG=verbose was what I sought. Thanks for directing me to it, though.
17:14karolherbst: :O complaints?
17:15karolherbst: koz_: if you want to help, vbios pls :p
17:26karolherbst: koz_: ohh it works, nice :)