00:23imirkin: mrnuke: first off, i'd recommend switching to modesetting over nouveau. alternatively (or additionally) might want to give Option "Accel" "None" a shot [for nouveau] -- might be faster.
02:38bozhan: karolherbst: hi, yesterday you tell me that there is something strange with some GPIO.... can i do something to improve situation?
02:39karolherbst: bozhan: basically get voltage used by blob at highest perf level + your vbios
02:39karolherbst: on kepler+ that is
02:39pecisk: karolherbst, I got logs from that tool, but it seems that column was stuck in 16
02:40karolherbst: pecisk: yeah, I think we need more gpios, wait a sec
02:40karolherbst: but if it was always 16, its bascially a good sign
02:40karolherbst: because the VIDs used should be over 16
02:41karolherbst: and the GPIOs are used as a bitmask
02:41karolherbst: so we just need those GPIOs for the higher bits
02:41karolherbst: or lower that is
02:41karolherbst: mhh you got some GPIOs we don't have in envytools, I will add them I guess
02:42pecisk: karolherbst, it was 16, yes
02:48bozhan: do i have to do something special before start vbtracetool because my card i hybrid optimus... in QUICKSTART is suggested to run commad in VT not in X but...
02:51karolherbst: pecisk: in pwr_read.c
02:51karolherbst: line 467
02:51karolherbst: add vid |= nvd0_gpio_sense(4); vid <<= 1; after the vid = 0;
02:52karolherbst: and then try that out
02:53karolherbst: bozhan: "QUICKSTART" ?
02:54karolherbst: and what is vbtracetool all about?
02:54pecisk: karolherbst, will do when I get to my PC
02:54pecisk: btw, for some reason envytools are in rawhide, but not in F23 repos
02:54pecisk: got to compile them by hand
02:56bozhan: forget what i've said:)) pls would you tell me how to get voltage used by blob at highest perf level + your vbios .... ot where to read ..
02:57karolherbst: bozhan: reag gpios out :D
02:58karolherbst: bozhan: what card do you have?
02:59bozhan: NVIDIA GK107 (0e70d0a2)
03:00karolherbst: pecisk: you have 5 GPIOs for the VID, which means 5 bits
03:00bozhan: [GeForce GT 755M] (rev a1)
03:00karolherbst: mhh which is strange
03:01pecisk: karolherbst, btw, there's my log of TF2 session http://fpaste.org/267851/42397667/
03:02karolherbst: its a bit odd, that there are so many values missing
03:02Rarruga: Hi ppl. Could somebody help me. I have two cards and three monitors and trying to utilize RandR. My system automatically can see two monitors (attached to the first card) but does not see the third monitor/card. What to do?
03:02karolherbst: only clocks, but well
03:02RSpliet: Rarruga: multi-card set-ups have been extensively covered on the wiki
03:03Rarruga: RSpliet, hm
03:03Rarruga: RSpliet, I've found nothing
03:03Rarruga: RSpliet, will try again
03:04Rarruga: but it is mostly for xinerama, which is shit
03:04Rarruga: with nouveau, at least
03:05RSpliet: well, you get to choose between shit and not working at all
03:05Rarruga: RandR 1.2 is just mentioned
03:06Rarruga: somebody here suggested me not to use xinerama (I had problems with software GL) and move to RandR ...
03:06RSpliet: sorry man, this is the only known working solution, and yes, you will lose accelerated GL
03:06Rarruga: oh, ok
03:06RSpliet: if you want 3 monitors and acceleration, you can also get a Kepler card
03:06RSpliet: they can drive 3 monitors from one card
03:07Rarruga: no, i have a better solution. Move back to nvidia driver
03:07RSpliet: or that
03:07Rarruga: but anyway, thanks.
03:19karolherbst: I am thinking, shouldn't DRI PRIME be able to handle such cofigurations?
03:20karolherbst: or doesn't it work when the same driver drives both cards?
03:20pecisk: btw, people, what's the status with Optimus? Is it even possible to enable it or use it somehow? Or I just have to force enable one card and stick with it?
03:20karolherbst: pecisk: yes
03:20karolherbst: pecisk: http://nouveau.freedesktop.org/wiki/Optimus/
03:21karolherbst: I do it all the time with nouveau
03:21karolherbst: basically you run the application where you want to use the dedicated card with DRI_PRIME=1
03:22karolherbst: pecisk: if DRI3 works for you, I would suggesting using this, but somehow DRI3 seems to be disabled for a lot of people with intel cards
03:22pecisk: karolherbst, I will guess - binary drivers doesn't have such functionality right?
03:22karolherbst: well kind of
03:22karolherbst: you can run the intel card with modesetting driver
03:22karolherbst: and render everything on the nvidia card
03:22karolherbst: which is just stupid for laptops :D
03:23karolherbst: and then there is bumblebee which kind of works pretty good with the blob
03:23pecisk: karolherbst, DRI_PRIME looks cool :)
03:24karolherbst: there is a note missing though
03:24karolherbst: you have to enable vsync on both sides
03:24karolherbst: compositor and application
03:24karolherbst: otherwise you get strange tearing
03:24pecisk: karolherbst, btw it gives me Intel on both cases with glxinfo and grep
03:25karolherbst: I think you have only DRI2 with intel
03:25karolherbst: then you have to follow the instructions for DRI2
03:26karolherbst: for nouveau development you only want to use DRI3 anyway
03:26pecisk: xrandr --listproviders gives me just Intel
03:26karolherbst: so maybe it's better to push the intel card to DRI3 :D
03:26karolherbst: you need the nouveau kernel module loaded
03:26pecisk: nouveau driver is loaded and used
03:26karolherbst: and the nouveau X driver isntalled
03:26karolherbst: then X log please :D
03:27karolherbst: I think I know the issue, but
03:27karolherbst: I bet X loads the modesetting driver for your nouveua card
03:27karolherbst: but we will see
03:33karolherbst: pecisk: what about the log?
03:36pecisk: karolherbst, https://drive.google.com/file/d/0B31QWsnd2TuWUFhoNDBOSDcwUVU/view?usp=sharing
03:36pecisk: sorry, my Fedora install seems to be broken, journalctl takes looooong time get things out of
03:40karolherbst: reduce the max size
03:40karolherbst: isn't there a /var/log/Xorg.0.log anymore?
03:40pecisk: karolherbst, not in journald world
03:41karolherbst: I mean I also use journald, but still got the X logs
03:41karolherbst: but journald is really bad designed :D so I really don't care
03:41pecisk: afaik Fedora has disabled most of them
03:41karolherbst: I hope you never get a 20GB journal
03:42pecisk: I don't think it's badly designed, I use it on servers and it is veryhandy
03:42pecisk: karolherbst, well, it should rotate/roll over
03:42karolherbst: I had one, and journald just took 30 seconds on each boot to be loaded
03:42pecisk: but it seems it's broken for my install :)
03:42karolherbst: and it did rotate
03:42pecisk: it rotates on my Debian 8
03:42karolherbst: I think it verifies the entire journal on each boot or something like that
03:42pecisk: it has quirks, bu I kinda like it, it will be ok
03:42karolherbst: not sure though
03:42pecisk: anyway, any good info on that log?
03:44karolherbst: ohh seems like you got DRI3 on intel
03:45karolherbst: pecisk: could you do "LIBGL_DEBUG=verbose glxinfo >/dev/null" and tell me which DRI version is getting used?
03:46karolherbst: there is a line missing?
03:46karolherbst: I also get something like "libGL: Using DRI3 for screen 0"
03:47karolherbst: pecisk: mesa version?
03:49pecisk: rest of lines are this
03:49pecisk: libGL: Can't open configuration file /root/.drirc: No such file or directory
03:49karolherbst: pecisk: are you in the video group?
03:50pecisk: karolherbst, nope
03:50pecisk: karolherbst, but I run this line n root too
03:50karolherbst: you should be
03:50karolherbst: for various reasons
03:50pecisk: karolherbst, I think it is a bit broken...I will try live usb stick with Fedora 23 and see what it does report
03:51pecisk: it might be not even to debug atm
03:51karolherbst: pecisk: jsut add you to the video group
03:51pecisk: still the same issue
03:51karolherbst: and the you need to logout/login
03:54karolherbst: then DRI_PRIME=1 should work, if not, something else is odd
03:56pecisk: karolherbst, ok, I have stuff to finish, but I will try relogin and will report results
04:40mlankhorst: hm, wonder why wikipedia has 'see also nouveau' on the vbe page..
04:45fling: mrnuke: no crashes anymore
04:45fling: A single error -> [14550.619781] nouveau E[ PFIFO][0000:01:00.0] FB_FLUSH_TIMEOUT https://bpaste.net/show/d3e85ce58f9b
05:13pecisk: karolherbst, adding to video group didn't help
05:14pecisk: karlmag, I will reinstall my laptop later this week and will get it another try
05:21pecisk: sorry karlmag wrong highlight
05:24karolherbst: pecisk: usually "reinstalling" never helps, because it's not what it does ;)
05:25karolherbst: reinstalling is more like "makeing a possibile fix to an error situation impossible"
05:26karolherbst: pecisk: I bet you don't have the nouveau ddx installed
05:26pecisk: karolherbst, ddx?
05:26karolherbst: with that installed you would at least get DRI2 offloading right. But maybe your xorg.conf is also messed up
05:26karolherbst: ddx = xorg driver
05:26pecisk: karolherbst, well, I have 21 here, so it is 1 year old
05:26pecisk: I will try 23 so it might change things a bit
05:26pecisk: and yes, my xorg is definitely messed up
05:26karolherbst: doesn't really matter, you could upgrade instead of reinstalling. It's not windows ;)
05:26pecisk: as I tried to start it before
05:27pecisk: karolherbst, I love to do reinstall because it's cleaner...i like to know my starting point isn't affected by some left overs
05:27karolherbst: mhh you could post your xorg.conf and custom files inside xorg.conf.d
05:28karolherbst: pecisk: as I said: you remove the possibility to fix some bugs then, but its your choice :)
05:28pecisk: it seems I have disabled xorg.conf
05:28karolherbst: and how did you do that?
05:28pecisk: karolherbst, xorg.conf.bak
05:28pecisk: renamed :)
05:28pecisk: xorg.conf.d only has keyboard conf
05:29karolherbst: ohh okay
05:29karolherbst: then that should be fine
05:30karolherbst: you should install the nouveua ddx then
05:30karolherbst: I bet somewhere in the stack dri3 is missing, but well
05:30karolherbst: doesn't really matter
05:30karolherbst: xorg-x11-drv-nouveau is the package
05:32karolherbst: sadly you would need to restart X after that and have the nouveua kernel module loaded before X starts, but then Prime offloading should work through the DRI2 method
05:32karolherbst: imirkin: any idea how to debug application which possibly mess up my kernel? :/
05:43pecisk: karolherbst, I have such package already
05:44karolherbst: it should get loaded by X
05:44karolherbst: is nouveau loaded at boot?
05:44karolherbst: or is it blacklisted
05:46pecisk: karolherbst, it is loaded as is intel driver
05:47karolherbst: maybe updating the system is the best idea in that situation then
05:47karolherbst: it's your choice if you want to reinstall or upgrade, I would choose the latter though
05:47karolherbst: I think something is missing in your software stack but I don't see what exactly :/
05:59karolherbst: imirkin: okay, the issue is worse on lower core clocks :/
06:06karolherbst: ohh high kernel load on low fps
06:17karolherbst: mhh got these errors: "[thread 0000000d][I]: Recommended max number of vertices addressed in one glDrawRangeElements() call: 3000
06:17karolherbst: [thread 0000000d][I]: Recommended max number of indices addressed in one glDrawRangeElements() call: 3000" maybe this is a clue?
06:18karolherbst: ohh and this " nouveau 0000:01:00.0: witcher2: multiple instances of buffer 89 on validation list"
06:53pecisk: karolherbst: what was that line again to see DRI version
06:53pecisk:from Fedora 23 live usb
07:02pecisk: karolherbst, wow. this happened.
07:02pecisk: launched with fedora 23 live usb
07:03pecisk: tried xrandr command, it showed both cards
07:03pecisk: tried DRI3 method, no luck
07:03pecisk: but after waiting for your command computer just froze
07:04pecisk: when trying to reload to live usb no desktop or anything but console spammed with SCHED_ERR UNKNO6 errors
07:04pecisk: not letting distribution to load from live usb
07:04pecisk: returning to installed fedora 21 fixed it
07:21pecisk: this reminds me something....something like I got same freezes when installing Fedora 21 which I got around somehow
08:27karolherbst: imirkin: the witcher2 issue you fixed was for the gog version?
08:29karolherbst: ohh the bug doesn't say so
08:39mrnuke: imirkin: I had no idea I could do any of the things you recommended to do :)
08:42karolherbst: mwk: I also found one with that "48 = Generic Initialized: This GPIO is used, but does not have a specific function assigned to it or has a function defined elsewhere. System software should initialize this GPIO using the _INIT values for the chip. This function should be specified when a GPIO needs to be set statically during initialization. This is different than function 25, which implies that the GPIO is not used by NVIDIA software."
08:42karolherbst: I have no clue what it does
08:43karolherbst: thought it may be better to leave that out until somebody cares enough to RE that
08:44mupuf: karolherbst: FRAMELOCK_DAUGHTER_CARD_INTERRIPT --> Interript? :D
08:44karolherbst: ohh shit
08:45karolherbst: stupid keyboard :D
08:45mupuf: you can push this kind of patches straight away now
08:45mupuf: this is trivial-enough
08:45karolherbst: somehow I didn't get anything though :/ and I still can't do that
08:47karolherbst: found it
08:47mupuf: karolherbst: visit https://github.com/envytools
08:47mupuf: I just sent you another invitation
08:47karolherbst: just did that now
08:47karolherbst: ohh its okay now
08:48karolherbst: I am in
08:48mupuf: congrats :D
08:49karolherbst: these laptop cards have tons of nice gpios
08:50karolherbst: and a lot on each card
08:50karolherbst: ohh wait
08:50karolherbst: this is a desktop card I am looking at
08:52karolherbst: mupuf: anyway, should nouveau be able to mess up my intel driver?
08:57karolherbst: mupuf: would it be possible for you to try some kepler cards out on a pcie v1 only board? I want to check a bit in a reg
08:57karolherbst: bit 2 in 0x8c1c0
09:11pecisk: karolherbst, ping?
09:12karolherbst: pecisk: hi
09:12karolherbst: pecisk: by the way, to you have a pcie v1 only board?
09:19pecisk: karolherbst, I remembered why it might be problematic to turn on Nouveau on my laptop
09:19pecisk: karolherbst, because I have GTX 850M
09:20pecisk: it's Maxwell
09:20karolherbst: no problem
09:21karolherbst: but you need a newer module
09:21karolherbst: GM107 actually
09:21pecisk: how much newer? :)
09:21karolherbst: I would use newest
09:21pecisk: yeah 107
09:21karolherbst: it should work, though no accell I think
09:21mrnuke: what about 007 ?
09:21pecisk: so console install and then manual upgrade of Nouveau driver
09:21pecisk: yeah, no acceleration
09:21karolherbst: pecisk: depends on the kernel
09:22karolherbst: everythong newer than 4.1 should be fine too I guess
09:22karolherbst: you can then use an external tree. Mine is 4.1 compatible
09:22karolherbst: the official tree is based on 4.2 or 4.3
09:22pecisk: well, afaik F23 has 4.2, and it literary crashed my laptop that it didn't even recover after reboot
09:22karolherbst: should be fine
09:22karolherbst: but the crash is messy
09:23karolherbst: what does F22 have?
09:23pecisk: I ran it, I browsed internet a bit on live usb env and it froze on me in 4 mins
09:23karolherbst: could be nouveau related though
09:24pecisk: F22 is 4.0
09:25karolherbst: what you could is, boot F23 into console
09:25karolherbst: blacklist nouveau or something
09:25karolherbst: and then try again
09:25karolherbst: or enable ssh and then just ssh into the machine when it freezes
09:25pecisk: boot it into vesa?
09:25karolherbst: no, just don't start X
09:25karolherbst: or switch to tty and stop X
09:26pecisk: I will poke around
09:27pecisk: do devs need some traces or tests ran for Maxwell? Or everyone just waits for microcode
09:28karolherbst: GM107 doesn't need that
09:28karolherbst: I think if accell works, then these maxwells are good to go?
09:28karolherbst: even clocking should work (tm)
09:29imirkin: RSpliet: ftr, reverse prime is the "new" way of doing multi-gpu cross-monitor stuff
09:29karolherbst: haha, funny, my EC got messed up again :(
09:29karolherbst: stupid thing
09:30pecisk: karolherbst, well it is kinda described as WIP on feature matrix, but I will give it a go when I get home
09:31karolherbst: I don't care the slightiest about this matrix to be honest
09:31karolherbst: if something works on my card, it works, if not, it doesn't
09:31karolherbst: so I rather test stuff out then to check some kind of matrix
09:31imirkin: pecisk: GM20x is very different from GM10x
09:32karolherbst: is GM10x that different from kepler actually or just some "minor" stuff?
09:32pecisk: imirkin, it's described as one family but if it should work, it's awesome, I will do testing :)
09:33imirkin: pecisk: where?
09:33karolherbst: mhh GM10x has a smaller memory bus, but more L2 cache
09:34imirkin: karolherbst: quite different on an internal level. the 3d engine is quite similar... lost immediate vertex submission, that's about it
09:34imirkin: oh, and a whole new ISA
09:34karolherbst: so different opcodes
09:35imirkin: no, same opcodes
09:35imirkin: just a different encoding
09:35karolherbst: I could imagine, that kepler was a mess at first
09:35pecisk: imirkin, http://nouveau.freedesktop.org/wiki/CodeNames/
09:35pecisk: but as I said, I know nothing :)
09:35karolherbst: pecisk: "NV110 family (Maxwell)"
09:35karolherbst: check at the bottom
09:35imirkin: along with rejiggered texture argument order, since, you know, you can.
09:36imirkin: pecisk: well, there's only so much one can say concisely. GM20x is unsupported. GM10x is sorta-supported.
09:36pecisk: I see
09:36pecisk: thanks for clarification
09:36karolherbst: hey, at least I didn't crash that maxwell card while pushing it to 0f pstate
09:36pecisk: hehehe ;)
09:36pecisk: well I want to get it stable first :)
09:37karolherbst: yeah, core clock stuff still messy on kepler
09:37karolherbst: not for everybody though
09:37karolherbst: but for some
09:37karolherbst: my card is pretty perfect now on a kernel level, but I don't need that much
09:37imirkin: karolherbst: it's messy for people who look at the code. not messy at all for people who just echo pstate levels into the file ;)
09:37pecisk: yeah, it looks pretty awesome to me :)
09:38pecisk: (don't show me code) :D
09:38karolherbst: like I don't care about video acceleration
09:38karolherbst: still I am searching for kepler owners with a pcie v1 only board :D
09:38karolherbst: I really need those
09:39karolherbst: this is like the corner case I can't inspect enough for the pcie work
09:40pecisk: errr, pcie v1 isn't like very very old?
09:40karolherbst: and most of the kepler owners have a pcie v3 board, very few v2
09:40karolherbst: and nearly nobody v1
09:41karolherbst: nouveau should put up a page with "we search for those system configuration, please join IRC"
09:41pecisk: that's actually what you have to do
09:41pecisk: or even better put it on linux_gaming and phoronix
09:42karolherbst: ohh, somehow I don't trust anybody on the phoronix forums
09:42pecisk: errrr yeah, they are a bit......trollllish
09:42karolherbst: a bit
09:46karolherbst: pecisk: nobody did test my branch though ;)
09:47pecisk: karolherbst, I suspect many are afraid to do so :)
09:47karolherbst: :D don't understand this
09:48karolherbst: you should have compared 07/0a/0f
09:48karolherbst: for me 0f has around 3x-4x more perf than 07
09:49karolherbst: how can I start X on the maxwell card on reator? Then I would just check if 0f really works there
09:50imirkin: same way you start X on any other box?
09:50imirkin: (hint: 'X' is a good start)
09:50karolherbst: mhhh, I maybe started X two or three times myself directly
09:50karolherbst: yeah, but I want to have only the maxwell card
09:50imirkin: then make a config that points to it
09:51karolherbst: ohh okay
09:51imirkin: take a look at /home/imirkin/xorg-nv.conf (iirc)
09:51imirkin: er, probably xorg-nouveau.conf actually
09:56karolherbst: X complains about no screen found
09:57karolherbst: that*s my config: https://gist.github.com/karolherbst/d5a20c8c43d53604dd63
09:58karolherbst: modesetting driver I guess
09:59karolherbst: okay, I have to load the nouveau module first :D
10:01karolherbst: wuut, accel works on that card?
10:04karolherbst: I thought it doesn't work at all
10:04imirkin: GM20x, no
10:04imirkin: GM10x, like i've said over and over and over and over again, is mostly fine
10:04karolherbst: ohh okay
10:04karolherbst: then I was wrong
10:05imirkin: although it does have a ton of issues
10:05karolherbst: anyway, the core just messed up at 0f
10:12joi: karolherbst: if you know how to reproduce those read faults you can mmt the offending app and search for fault address in demmt output - if you are lucky enough demmt will detect which mthd triggered it
10:13karolherbst: joi: load a game in witcher 2 :D
10:13karolherbst: this sometimes triggers it
10:15joi: it's not really playable here
10:16joi: 5 fps is quite annoying
10:17karolherbst: joi: gog or steam version?
10:17karolherbst: ohh maxwell doesn't do memory reclocking anyway
10:18karolherbst: I have this nice issue: https://bugs.freedesktop.org/attachment.cgi?id=117061
10:18imirkin: joi: did you try reducing the resolution?
10:18karolherbst: joi: do you have a gddr5 card? :D
10:19joi: imirkin: I don't remember exact numbers, but it didn't the game playable
10:19joi: karolherbst: yes
10:19karolherbst: joi: wanna have some gddr5 perf?
10:19joi: I tried your gddr5 branch
10:19karolherbst: ohh and?
10:19imirkin: joi: too bad
10:20karolherbst: withcer 2 needs a really fast card
10:20joi: 0f was like 5% faster than 0a
10:20karolherbst: that's strange
10:20karolherbst: but it depends on the case
10:20karolherbst: what did you try out?
10:21joi: I also tested nexuiz - same numbers
10:21joi: (or rather % improvement)
10:22karolherbst: mhh, 0f won't do much on gpu core bottlenecked stuff anyway
10:22joi: I posted results on pastebin, let me find it
10:22karolherbst: for witcher it was like 350fps 07, 600fps 0a and 1000fps 0f 800x600 on low settings
10:23karolherbst: for me
10:24karolherbst: ohh my gddr5 patch doesn't change performance
10:25karolherbst: it just makes gddr5 0f pstate more stable
10:26karolherbst: mhh, strange
10:26karolherbst: I could test shadow warrior
10:26karolherbst: ohh you did a retrace
10:26karolherbst: okay, well
10:27karolherbst: the pcie speed increase is quite nice though
10:28karolherbst: 2.5% on a desktop system
10:31imirkin: skeggsb: "all copies or substantial busions of the Software" -- one s/port/bus/ too many? :)
10:32imirkin: in the future i highly recommend using \b -- avoids these idiocies quite nicely
10:33joi: I retested witcher2 now, in game: 07: 5 fps, 0a: 10 fps, 0f: 13 fps
10:34imirkin: joi: and you have gddr5?
10:34imirkin: surprising that 0f doesn't have a bigger jump
10:34imirkin: perhaps it's limited by pcie?
10:35joi: nouveau 0000:01:00.0: fb: 2048 MiB GDDR5
10:35karolherbst: imirkin: some games are gpu core limited
10:35imirkin: ah i guess
10:36karolherbst: his gpu core clock doesn't change at 0f, so
10:36karolherbst: same for me
10:36karolherbst: joi: do you have antichamber?
10:36Yoshimo: mupuf: https://phoronix.com/scan.php?page=news_item&px=Nouveau-Volt-GM107-Too says you don't have a 9xx card to confirm that it works the same , is that something one could help you with remotely?
10:36joi: karolherbst: yes
10:36karolherbst: joi: test it with it
10:37karolherbst: I got a 4x speed bump from 07 to 0f
10:37karolherbst: but disable vsync first :D
10:37karolherbst: I easily hit the 60 fps border there
10:37karolherbst: already at 0a
10:39karolherbst: Yoshimo: check the regs I would say
10:39karolherbst: 0x20340 and 0x20344
10:39karolherbst: the former may be always 0x60
10:39joi: 07: 30, 0a: 63, 0f: 106
10:40karolherbst: this game seems to be heavily gpu memory bottlenecked
10:40karolherbst: for whatever reasons
10:40karolherbst: more tha 50% pefr just by reclocking memory
10:40karolherbst: this is insane
10:43mupuf: Yoshimo: hey
10:44karolherbst: joi: I don't think that pcie speed will effect antichamber, but feel free to try that, too :D
10:44mupuf: karolherbst: if you are not memory-limited, you are doing it wrong
10:44karolherbst: tell that the nexuiz guys :D
10:45mupuf: karolherbst: the driver is also to blame
10:45karolherbst: ohh okay
10:45joi: karolherbst: I have it applied
10:46karolherbst: and then there are strange games like talos principle, which get a 25% perf boost by raising pcie speed
10:46joi: ... or not
10:47joi: I swear I applied it along with gddr5 patch
10:47karolherbst: on kepler its piece of cake to do it by hand anyway
10:48karolherbst: joi: but was 0f pstate unstable for you without my patch?
10:48karolherbst: I do't care about the "more perf" as much as about the stability tbh
10:49karolherbst: ohh okay, nice
10:50karolherbst: from what I know, there is only this isohub thingy left, which somehow only effect one so far? and then GPIO based gpu core voltage issues
10:50karolherbst: mupuf: anything else you know of?
10:53Yoshimo: mupuf: i can offer to run tests on my 980 in case that helps you in any way
10:58joi: with pcie speed patch in antichamber I get 0a: 67, 0f: 120
10:58karolherbst: actually quite nice
10:58joi: vs previously 07: 30, 0a: 63, 0f: 106
10:58karolherbst: around 10% for 8.0
10:58karolherbst: desktop or laptop?
10:58karolherbst: okay, then this is normal
10:59joi: I'll try witcher2 now
10:59karolherbst: the more pixels you need to transfer over pcie, the more the pcie speed matters
10:59karolherbst: don't bother
10:59karolherbst: if you are low on fps, pcie speed won't change anything
11:00karolherbst: did some withcer testing today already, so
11:00karolherbst: pcie really doesn't matter there
11:00karolherbst: though there seems to be a stable +5% improvement across everything
11:00mupuf: Yoshimo: well, as karolherbst said, it would be nice to see if you can read some regs
11:00karolherbst: on latpops
11:00mupuf: and write to them
11:00karolherbst: 0x20340 and 0x20344 ;)
11:00mupuf: if you can then everything should be good
11:01karolherbst: 0x20340 should be 0x60
11:01karolherbst: and 0x20344 something between 0x20 and 0x50
11:01karolherbst: mupuf: did you encounter anything else?
11:01mupuf: mine is lower than 0x20
11:01karolherbst: well with your aptch I also get to 0x1f
11:01mupuf: but I want to see the values in any cas
11:01joi: karolherbst: so when will you submit your patches?
11:02karolherbst: joi: gddr5 is already pending
11:02Yoshimo: sorry, i need that on a beginner level, how do i do that?
11:02karolherbst: skeggsb just want to get an answer from nvidia
11:02karolherbst: but he wanted to push the gddr5 patch for 4.4 anyway
11:02karolherbst: Yoshimo: nvapeek
11:02karolherbst: from envytools
11:03imirkin: or a magnet
11:03imirkin: and a _very_ steady hand
11:03karolherbst: and a needle
11:06karolherbst: Yoshimo: I hope there is a package for envytools for you
11:07Yoshimo: i think i alrady used one of its tools to make a trace more readable
11:07karolherbst: ohh okay
11:07karolherbst: there should be a tool called nvapeek somehere
11:07karolherbst: then just do nvapeek 0x20340
11:07karolherbst: and nvapeek 0x20344
11:08karolherbst: joi: the pcie stuff is not finished, and I won't make it before 4.5 actually
11:08karolherbst: there is ton of stuff I have to consider, just to cover most of the corner cases
11:09Yoshimo: karolherbst: does it require nouveau to be the active screen driver?
11:09karolherbst: Yoshimo: no
11:09karolherbst: the gpu just need to be on
11:09karolherbst: so, nexuiz testing :D
11:11karolherbst: two new bugs for nexuiz I guess
11:12karolherbst: I mean, I know that some games prevent alt+tab while in fullscreen mode
11:12karolherbst: but necuiz even does that in window mode
11:12karolherbst: or did I do anything wrong?
11:15karolherbst: somehow xonotic looks much nicier managed anyway?
11:24karolherbst: joi: okay, you did something wrong
11:24karolherbst: I get like 80 fps on maxed out settings on 0f
11:24karolherbst: and only around 50 on 0a
11:26joi: heh, general protection fault from nouveau_fence_new, anyone interested?
11:27karolherbst: go away with those :p we don't need them
11:28joi: it was while I was testing witcher2
11:29karolherbst: yeah, I get those too
11:29karolherbst: or different stuff
11:29karolherbst: witcher 2 messes up my entire drm stack
11:29karolherbst: even my intel card goes down
11:31Yoshimo: mupuf karolherbst , results from nvapeek https://pastee.org/9h8ee
11:31karolherbst: ohh wow
11:31karolherbst: 0x6c in 0x20340
11:32karolherbst: never saw that, but mupuf should know more
11:32karolherbst: seems alright though
11:32karolherbst: Yoshimo: if you want have some fun, reduce the second one until your card crashes :)
11:33mupuf: Yoshimo: can I get your vbios?
11:34mupuf: looks like the PWM controller should receive a frequency of 256 kHz instead of 280 kHz for ours
11:35Yoshimo: mupuf: sure, which method would be the best? nvagetbios wants an extraction method
11:35mupuf: but hey, that looks good, now ... let's try to increase the voltage a little
11:35mupuf: nvagetbios -s prom > vbios.rom
11:35mupuf: nvapoke 20334 80000027
11:36mupuf: and then nvapeek 20334
11:36mupuf: karolherbst: no need to add the 0x
11:37mupuf: it is parsed as hex
11:37karolherbst: I didn't mean that
11:37karolherbst: okay, then agian: *20344
11:37Yoshimo: so 1mb file with a second bios saved
11:38mupuf: Yoshimo: ok, looks good
11:40Yoshimo: nvapeek says 00020344: 00000047
11:42Yoshimo: so mupuf where do you want to have the vbios file?
11:42karolherbst: Yoshimo: what exactly did you poke?
11:42mupuf: Yoshimo: are you running nouveau or the blob?
11:42Yoshimo: karolherbst: nvapoke 20334 80000027
11:43Yoshimo: currently blob to be able to play games
11:43mupuf: Yoshimo: ok, then you need to disable scaling
11:43mupuf: otherwise, you will crash your card
11:43mupuf: nvidia-settings will allow you to always use the full speed mode
11:43mupuf: no scaling
11:44mupuf: when you have done this, read back the value of 20334 and rewrite + 1 and with the 8 at the begining
11:44mupuf: and read back
11:44mupuf: then send us the log of what you did
11:44Yoshimo: so where is scaling, nvidia x settings?
11:44karolherbst: mupuf: I thought the reg is 20344?
11:45mupuf: oh, you must e right
11:45mupuf: I may misremember
11:45karolherbst: 20334 is nothing here ;)
11:50karolherbst: mupuf: allthough setting to max performance, doesn't mean that the blob will set to highest clock though :/
11:50karolherbst: it just means, the driver prefers it, and on my system it clocks at 705 MHz while idle, and to something higher while at load
11:50karolherbst: even the voltage is different
11:56Yoshimo: 20334 is indeed "..." karolherbst
11:57karolherbst: yeah, 20344 is the important one
12:00Yoshimo: https://pastee.org/whdvg , like that?
12:01karolherbst: seems fine so far
12:02karolherbst: and if you want to have some fun, you can decreasethat value until your gpu crahses ;)
12:02karolherbst: but I think that's enough
12:02karolherbst: Yoshimo: we still need your vbios
12:02karolherbst: could you upload it somewhere and give us the link?
12:02Yoshimo: let me find a hoster
12:05pecisk: karolherbst, you got some attention http://www.phoronix.com/scan.php?page=news_item&px=Nouveau-GDDR5-Branch
12:06karolherbst: ... now it happened
12:07joi: heh, I reproduced the read fault
12:07joi: fifo: read fault at 000850a000 engine 1b [CE2] client 18 [GR_CE] reason 02 [PTE] on channel 5 [007f856000 witcher2]
12:07karolherbst: then its not just me
12:09pecisk: karolherbst, what kernel parameter I had to use to get cstate file?
12:10karolherbst: pecisk: none
12:10karolherbst: its just something I did for myself currently
12:10imirkin: pecisk: looking at that reddit thread, please avoid saying things like "fermi and maxwell don't work"
12:10karolherbst: I just did it for fun back then, but it came in handy for some cards
12:10imirkin: pecisk: fermi works just fine, just no reclocking support. GM10x maxwell mostly semi-works, but again, no reclocking.
12:11pecisk: imirkin, ok, then I will put that there
12:11imirkin: pecisk: i think people are understanding your comments to mean NO support
12:11pecisk: sorry about that
12:11imirkin: [i don't have a reddit account and am not looking to get into that whole madness]
12:11karolherbst: imirkin: wise move
12:12karolherbst: ohh I already have 9 post on phoronix? how that happened
12:13pecisk: linux_gaming is only place I visit in reddit..there's some nice crowd there
12:19mupuf: Yoshimo: oopsie, your vbios crashes nvbios :o
12:20pecisk: imirkin, I specified my posts to your liking about Nouveau support for Maxwell and Fermi
12:20Yoshimo: hopefully not my fault mupuf
12:20imirkin: pecisk: cool thanks
12:20pecisk: ok, time to burn/test my laptop with Nouveau testing
12:20mupuf: Yoshimo: possibly not
12:23karolherbst: mupuf: the boost table looks odd
12:23karolherbst: but somehow
12:27Yoshimo: odd but right, how can that be?
12:31Yoshimo: nvbios says it doesn't know version 4.1 and crashes parsing it anyway
12:31Yoshimo: ftp://download.nvidia.com/open-gpu-doc/DCB/2/DCB-4.x-Specification.html says they changed the format a little
12:32imirkin: fyi that doc is just about the DCB
12:32imirkin: not the whole vbios
12:32imirkin: (well, it actually also has info on a few more tables, but hardly everything)
12:32Yoshimo: yea but the crash happens in the parsing of that table
12:34mupuf: Yoshimo: -- Mode PWM, acceptable range [600000, 1275000] µV, frequency 256 kHz, base voltage 600000 µV (unk = 16), range 675000 µV --
12:34mupuf: so, this is all good
12:34mupuf: we can enable support for your gpu then!
12:34mupuf: what is your card?
12:34mupuf: nvapeek 0?
12:35Yoshimo: 00000000: 124000a1 980 from msi
12:37mupuf: ok, well, I will enable it for all the maxwells but the GM20B which is likely to use another one
12:37mupuf: another way*
12:37mupuf: likely the way used by GK20A
12:39Yoshimo: anything else i can poke and peek for you guys?
12:44mupuf: one last thing
12:44mupuf: nvapeek 101000
12:44mupuf: and your email address (in private message)
12:46mupuf: Yoshimo: you are credited in a very small patch I just sent
12:46mupuf: skeggsb: looks like we are all good for the maxwells for voltage management!
12:46skeggsb: i'm surprised about that tbh, but, cool
12:46mupuf: and they use the same constant for the clock
12:46imirkin: mupuf: just need to figure out memory reclock scripts :)
12:47imirkin: should be a piece o' cake
12:47mupuf: skeggsb: yes, agreed
12:47Yoshimo: 00101000: 8040008e
12:47skeggsb:wonders how they managed to allow the host to control voltage but not fan speed
12:51karolherbst: the article hit
12:51karolherbst: from 8 uique visitors to 40 now
12:51Yoshimo: looks like too many people still buy nvidia cards karol
12:52karolherbst: skeggsb: I can't control fan either here :), but that's on a laptop with dedicated gpu fan
12:52mupuf: karolherbst: yeah, it is something totally different :p
12:53skeggsb: karolherbst: on discrete gm2xx boards, the register interface is read-only to the host
12:53skeggsb: (for the fan regs)
12:53karolherbst: ohh I see
12:53karolherbst: allthough nouveau tries to st my fan through pwm
12:53karolherbst: but nothing happens really
12:54karolherbst: I think my EC controls the gpu fan somehow
12:54karolherbst: not sure how exactly
12:57imirkin: skeggsb: btw, did you see the nva0 fan fail bug? i'm guessing something went wrong in the anx conversion
12:57skeggsb: hm, no not yet, i have one of those boards though and didn't notice that
12:57imirkin: i tried to make sense of what you were doing, but there were way too many changes without any hint of what was being done in the commit log =/
12:57skeggsb: i'll look at it when i'm back in the office
12:57imirkin: well, it was on resume from suspend
12:58skeggsb: that particular board needs work in general (it was somewhat broken even before), so maybe i should do that at the same time :P
12:59imirkin: fine by me. but it seems like there was just a straight-up regression there
12:59skeggsb: yeah, it's definitely plausible
13:01pmoreau: mwk: Ping :-) Did you had some time to review the latest patches I added to both pull requests?
13:13karolherbst: joi: did you plan to work on that witcher 2 bug? I have no real clue how to do that despite doing some mmts :/
13:13pecisk: karolherbst, this is what I get in console when booting with regular Fedora 23 live usb
13:13pecisk: karolherbst, https://drive.google.com/file/d/0B31QWsnd2TuWQlRXdmthY2VRcTQ/view?usp=docslist_api
13:14pmoreau: imirkin: Given that the other rd/wr functions are present in the header, I should add the mask one as well. Or change rd/wr to static. For consistency
13:14karolherbst: pecisk: ohhhh that's kind of messy
13:14karolherbst: I think the live usb thingy tries to do something really stupid
13:14karolherbst: okay, try that: boot into console mode
13:15pecisk: karolherbst, there's basic graphics mode
13:16karolherbst: no console mode is what we need
13:17karolherbst: and then we start from that
13:17pecisk: there's no console mode, it seems so
13:17karolherbst: ohh messy
13:17karolherbst: is there anything else?
13:18pecisk: started in vesa (basic mode) and ctrl+alt+F1?
13:18pecisk: i have tty open now
13:19karolherbst: maybe that works
13:19karolherbst: lsmod | grep -e i915 -e nouveau
13:19karolherbst: is nouveau or i915 loaded?
13:20pecisk: yes, both
13:23karolherbst: is nouveau used?
13:23pecisk: no, both i915 and nouveau shows 0
13:24karolherbst: the nouveau card shouldn't be used in X in the first place (or at least not to display anything), the sched errors are just strange though
13:24karolherbst: okay, mhhh
13:24karolherbst: pecisk: what login manager does fedora use?
13:24pecisk: I *suspect* this is live usb bug
13:25karolherbst: yeah it will be
13:25pecisk: in wayland mode
13:25karolherbst: systemctl start gdm
13:25karolherbst: then switch to a new tty and check for any errors in dmesg
13:25karolherbst: also the Xorg.0.log may be interessting
13:30pmoreau: mupuf: What happened to the Nouveau logo btw? Some comments were made and then no more news. Is it considered finished and ready to use or not?
13:46pecisk: karolherbst, I think trick is gdm is integrated into gnome shell and if you have gnome shell loaded, you have gdm as well
13:48karolherbst: yeah, and?
13:48karolherbst: ahh okay
13:48karolherbst: well were you able to start gdm?
13:48karolherbst: I don't the name of the service though
13:49pecisk: I did, but it just eat it up like "I am already up"
13:49pecisk: (desktop is already loaded)
13:50mupuf: pmoreau: yes, it is considered done: http://fs.mupuf.org/mupuf/nvidia/logo/
13:51imirkin: i think there were questions about the license
13:52karolherbst: pecisk: ohh, no sched error now?
13:53karolherbst: mupuf: :D looks kind of nice
13:54karolherbst: I just hope nvidia won't ever complain
13:55pmoreau: mupuf: Ok, cool! :-)
13:55pecisk: there's no direct resemblance of their logo
13:56pecisk: also name and intent is clear
13:56karolherbst: pecisk: I just wished all lawers would think like that ;)
13:56pecisk: karolherbst, you can't know what lawyers will do
13:56karolherbst: but I think it may be fine
13:56karolherbst: it seems to be different enough
13:57pecisk: I like it, although I would like to see monochrome small version
13:57karolherbst: yeah it looks nice
14:01karolherbst: pecisk: so are you able to do anything on the desktop on that live usb?
14:02pecisk: now it works with VESA driver
14:02pecisk: I will try to install it and then play around
14:02karolherbst: yeah well :/
14:02karolherbst: we may be able to fix every X configuration issue then
14:02karolherbst: might get messy
14:02pecisk: if it's usb live error I will be able to reproduce it anyway
14:03karolherbst: pecisk: though I am kind of dissappinted by the reaction. no clone :D
14:03pecisk: if it's not, well, I will get broken config on laptop :D
14:03karolherbst: ohh wait, 3 today :/
14:03mupuf: karolherbst: I sent them the file a while ago, no answer
14:03karolherbst: mupuf: :D
14:03karolherbst: clever step
14:04karolherbst: skeggsb: I bet you still got no answers?
14:11hakzsam_: oh, I like this logo too :)
14:33pecisk: karolherbst, I installed Fedora 23 and its setup Xorg with safe mopde for now
14:35pecisk: have to remove nomodeset from grub
14:36pecisk: easy enough
14:36pecisk: now I know how to recover it from that error so it's cool
14:47karolherbst: pecisk: so does it work now?
14:50pecisk: karolherbst, got to login screen aaaand it froze
14:50karolherbst: pecisk: nouveau.runpm=0
14:50karolherbst: try that
14:53pecisk: karolherbst, got shed_error :)
14:53karolherbst: uhhh mhh
14:54karolherbst: but does X still works?
14:54pecisk: it's when I am loading
14:54karolherbst: then xorg.conf and anything stupid from xorg.conf.d
14:54pecisk: it happens right at the begining
14:54pecisk: no xorg.conf and just keyboard stuff in .d
14:54karolherbst: X log
14:55pecisk: I will try to get previous boot from journald
14:55pecisk: so schema is like this
14:55karolherbst: I am pretty sure its someting really stupid, like X tries to use the nvidia card as primary device or something
14:55pecisk: I launch it, it runs for 3 mins, and then it freezes
14:55pecisk: and in next boot I get sched_error
14:55karolherbst: ohh mhh
14:56karolherbst: blacklist nouveau
14:56pecisk: sched_error is too early for xorg
14:56pecisk: I just use nomodeset, it ignores nouveau
14:56karolherbst: yeah, blacklist nouveau
14:56karolherbst: never use nomodeset
14:56karolherbst: it also pains the intel card
14:57karolherbst: you use nomodeset if you give up entirely and don't care about your gpus anymore :)
14:57pecisk: where do you black list modules these days?
14:57pecisk: yeah, it's ultimate safe mode
14:59pecisk: ./etc/modprobe.d/blacklist ?
15:00karolherbst: will I be killed for something like this? "return gk104_pci2_rd32(pci, 0x1c0) & 0x4 == 0x4 ? 2 : 1"
15:00karolherbst: pecisk: doesn't matter which file there, just add a blacklist entry there
15:01joi: heh, I can crash the xserver with 100% "success" rate by passing "--sb" option to glretrace
15:02joi: if anyone is interested, xorg.log: http://paste.ubuntu.com/12430845/
15:03pecisk: karolherbst, ok, blacklisting still gave me sched_error, but it didn't halt loading
15:04pecisk: I guess it's initrd then who does that?
15:04karolherbst: ohhhh mhhh
15:04karolherbst: I never used any fancy initrd :D
15:05karolherbst: don't know how to add a blacklist entry then
15:05pecisk: that sched_error
15:05pecisk: it's getting spewed by intervals
15:06pecisk: hmmm, blacklist didn't work
15:06pecisk: nouveau still loaded
15:06pecisk: but at least I have console and it didn't hang
15:06pecisk: hey it's a start
15:07karolherbst: usually you get these sched_errors when reclocking does something terrible wrong
15:07karolherbst: but I guess there are other reasons, too
15:08pecisk: last error for that spewage is 'runlist 4 update timeout'
15:08karolherbst: pecisk: I think you simply need to rebuild your initrd now
15:09karolherbst: pecisk: sudo dracut --omit-drivers nouveau /boot/initramfs-$(uname -r).img $(uname -r)
15:09karolherbst: does that work?
15:10pecisk: one sec
15:10pecisk: everytime I log into console, or do sudo
15:13pecisk: needed --force
15:13pecisk: buliding new initramfs
15:13pecisk: still doesn't explain why module itself got loaded though
15:15pecisk: karolherbst, ok, that worked
15:15pecisk: no nouveau
15:15karolherbst: if you have hardware the kernel tries to search for fitting modules
15:15karolherbst: that's just the way it goes
15:15karolherbst: pecisk: okay, at least something
15:16karolherbst: now we can start with the ugly parts :D
15:16pecisk: karolherbst, ok, but that's tomorrow :)
15:16pecisk: got to sleep
15:16karolherbst: okay, no problem
15:16karolherbst: at least intel should work now?
15:17pecisk: this time it's my working laptop so I will be able to play around during my free time
15:17pecisk: intel works
15:17pecisk: no sched_error or anything
15:17karolherbst: at least something
15:17pecisk: k, will continue tomorrow :)
15:17pecisk: thanks and good luck
15:29karolherbst: what's the best error code when something unexpected happend? like trying to get the card into v2 mode fails allthough it's supported?
15:30karolherbst: EINVAL or something else?
16:10mwk: karolherbst: there's always good old -EIO
16:10karolherbst: mhh was thinking about that, yes
16:10mwk: "I was just reading that file when my hdd disappeared"
16:11karolherbst: but with -EIO I understand more like: something was send to the card, but something different was received
16:11karolherbst: mwk: yeah, for me -EIO seems a bit drastic :/
16:11karolherbst: there are valid reasons why changing pcie version doesn't work
16:12karolherbst: I just don't know them all currently
16:19karolherbst: mwk: any suggestions? https://github.com/karolherbst/nouveau/commit/f9183b999218c296340acf9a512c81d08211f6cb
19:41marcosps: hi guys :)
19:45marcosps: imirkin: around?
20:04marcosps: imirkin: so, some ideas: When calling getImmediate from insnCanLoad, the insn (got from src->value->getUniqueInsn()) is always a ld, instead of a OP_MERGE :)
20:04imirkin: marcosps: you seem to be having a lot of trouble with all this...
20:04imirkin: i'm not sure how much more i can help without actually just doing it myself
20:05marcosps: imirkin: Sorry. But, I thought when should achieve a OP_MERGE in getImmediate when calling it from insnCanLoad, but right now, I'm just seen a ld there =?
20:06imirkin: look, this stuff isn't magic -- trace through what's getting called how and why
20:06imirkin: but given the amount of trouble you've been having, perhaps the compiler isn't the best place for you to have started
20:07marcosps: imirkin: maybe, but I'll try to finish this task before start a new one....
20:07imirkin: sounds good
20:08marcosps: I was confused about the src == OP_MERGE and insn == OP_MERGE hehehehe
20:08marcosps: you said to me that insn needs to be an OP_MERGE in getImmediate, so it's not happening here :P
20:09marcosps: But, I'll try to make it happen, and then, talk to you again. I feel frustrated about bothering you every single day without any concrete ideas...
20:09imirkin: well, you can ignore the things i say
20:09imirkin: at the end of the day, it just has to work
20:09imirkin: the things i say could be slightly off, or i could be expressing them poorly
20:10marcosps: imirkin: Sorry, I didn't wanted to be rude. You helped me a lot until I reach this point and now I need to make some work myself :)
20:12imirkin: first you want to make it so that the immediate gets propagated directly into the instruction that uses it
20:12imirkin: and then ... that's it :) fix up the emitter and you're good to go.
20:13marcosps: imirkin: so, I'm almost there :)
20:14imirkin: well.... you seem to be having a LOT of trouble with the initial step
20:14imirkin: even combining 2 values to make 1 64-bit one
20:14imirkin: which is probably a bad sign =/
20:15marcosps: imirkin: Yes, sadly I never did something like it before, even in university. But, I'm learning a lot doing it...
20:15imirkin: you had mentioned you were familiar with compilers
20:17marcosps: imirkin: but we used java there, and it was a simple compiler to read integers and strings, and then write a .net assembly...
20:17marcosps: we just needed to do something like "push int32 $a" and something related
20:20imirkin: no worries
20:20imirkin: it's tricky stuff
20:23marcosps: imirkin: Unfortunately, I was wrong about my knowledge about compilers. Don't hate me :)
23:41UBR47K: I am unable to go past the linux booting process (tons of text being spewed before being introduced to the login console) with a computer with 2x TITANX if I use nouveau driver because of unrecognized chipset, disabling nouveau modesetting will allow me to go past this unknown chipset error/warning
23:42UBR47K: I am currently running with the proprietary driver, is there any tests that are desired/needed for this hardware?
23:42UBR47K: to benefit the nouveau project