01:52 mlankhorst: imirkin: not really? shared fences are remembered
01:52 mlankhorst: exclusive fences imply syncing to the previous exclusive and al previous shared fences
02:41 Psy-Q: do you still need mmiotrace dumps for GK106?
02:42 mupuf: Psy-Q: more data is never bad!
02:42 mupuf: we have a very linited amount of them in our private repo
02:42 Psy-Q: mupuf: alright :) i have GK106 and GF114
02:42 mupuf: both would be welcomed!
02:43 mupuf:will need to send a call of mmiotraces again soon
02:43 mupuf: we have almost no traces for maxwell
02:43 Psy-Q: awesome! i'll get on it when the next hobby-spare-time-woohoo time window rolls on
02:43 Psy-Q: or i might kick this old machine and make it compile mesa during lunch break
02:44 Psy-Q: maxwell, hmm, no. my laptop has GK104
02:44 Psy-Q: don't know how to capture that (it's optimus)
02:44 Psy-Q: i was thinking of getting a GM204 for home, though! as soon as i do, i can trace things to bits :)
02:44 mupuf: the normal instructions still apply
02:45 mupuf: but you need to use the proprietary driver
02:45 mupuf: hehe, nice
02:45 Psy-Q: mupuf: oh good! so i would just "primusrun glxgears" with nvidia-current (or nvidia, whatever) in bumblebee's config and it will trace the right thing?
02:46 Psy-Q: lunch break now, argh, no time to set up a mesa compile. but soon!
03:07 pmoreau: Hum… I could mmiotrace the Titan X. Just need to remember to do it before going home.
03:11 mupuf: Psy-Q: yes, that should owrk
03:11 mupuf: pmoreau: yeah, would be nice!
03:12 pmoreau: mupuf: And I have one VBIOS a user sent me that I still need to add to the repo… --"
03:12 mupuf: along with the strap peek?
03:13 pmoreau: I have an extract of the dmesg, which displays BOOT0 :-)
03:36 mupuf: pmoreau: that's not the interesting part :D
03:36 mupuf: we need the reg 101000
03:37 pmoreau: Oh no! BOOT0 is reg 0…
03:37 pmoreau: Mixed both
04:48 Psy-Q: hm, the debian default kernel doesn't support mmiotrace?
04:48 Psy-Q: or do i just need debugfs? hm.
04:49 Psy-Q: hm, maybe the kernel was too old
04:52 Psy-Q: gonna stick testing on it since i don't want a frankendebian
05:23 karolherbst: pmoreau: why not ordering gm200 right in the xml?
05:23 karolherbst: ohh
05:23 karolherbst: my mistake :D
05:23 karolherbst: sorry for that
05:31 pmoreau: No problem :-)
05:35 Psy-Q: for the pciid when submitting an mmiotrace, is 10de:11c6 required or is 11c6 enough?
05:39 pmoreau: 10de is the vendor prefix for NVIDIA IIRC, so 11c6 should be enough.
05:39 pmoreau: Let me check
05:39 Psy-Q: great, thanks
05:40 pmoreau: Yep, 10de is NVIDIA's vendor prefix
05:40 Psy-Q: cool, thanks :D gonna tarball this stuff and email it in
05:41 pmoreau: Great, thanks!
05:48 karolherbst: pmoreau: I will check some clock gating stuff today for tesla cards, wanna try something out later?
05:48 pmoreau: karolherbst: For sure I want!! :-)
05:48 pmoreau: As long as it doesn't mean staying up until 5am, it should be fine ;D
05:49 karolherbst: :D
05:49 karolherbst: first I have to figure out how that works there
05:50 pmoreau: There should be some PG/CG stuff in the traces I uploaded, IIRC
05:51 karolherbst: I hope it is as easy as on fermi+ :D
05:51 karolherbst: and with that I mean, that there is one poke needed to reduce the power consumption a tiny bit already
05:52 Psy-Q: mail sent
05:52 Psy-Q: if there are any particular things you want me to run on those cards (it's 10 seconds of glxgears now), i can do that. i don't know crap about driver development, but i can at least send a log :)
05:53 karolherbst: pmoreau: how stable can you messure your power consumption?
05:54 karolherbst: Psy-Q: it is a gddr5 kepler card, right?
05:54 Psy-Q: karolherbst: yes, but let me double check if it's GDDR5 or 3
05:54 karolherbst: everything above 640 is gddr5
05:55 karolherbst: or at least should be
05:55 Psy-Q: yes, the particular model here is GDDR5
05:55 Psy-Q: so i guess all the 650 ti are
05:55 karolherbst: every played around with the pstates?
05:55 Psy-Q: i *think* i have a fermi GDDR3 at home
05:55 karolherbst: GDDR3 or DDR3? :D
05:55 Psy-Q: karolherbst: oh crap. let me check
05:55 karolherbst: I don't think there are any GDDR3 fermis
05:56 karolherbst: ohh wait
05:56 karolherbst: there is :D
05:56 pmoreau: karolherbst: As stable as powertop can I would say… not exactly sure about what you have in mind
05:56 karolherbst: pmoreau: I mean how much does the value change?
05:56 karolherbst: across reads
05:56 Psy-Q: wow, strange, it's also a GDDR5 at least on nvidia.com's specs
05:56 karolherbst: without a dedicated power sensor on the gpu, it is a bit messy to do assumptions from the battery power draw
05:57 pmoreau: If I let it stabilise, I would say ~0.2W
05:57 karolherbst: Psy-Q: not strange, depends on the card
05:57 karolherbst: Psy-Q: fermi has a lot of gddr5 models
05:57 karolherbst: pmoreau: okay, that helps
05:57 karolherbst: pmoreau: how high?
05:58 Psy-Q: karolherbst: haven't played with pstates yet because i couldn't get everything running yet :(
05:58 Psy-Q: but maybe if i take an hour off tonight i could give it a try
05:58 karolherbst: Psy-Q: ohhh
05:58 pmoreau: karolherbst: Let's see, it stabilised around ~14-15W on idle IIRC
05:58 karolherbst: Psy-Q: well I develop nouveau on my kepler card and pretty much everything works
05:58 karolherbst: pmoreau: nvapeek 1588
05:58 pmoreau: (I'm compiling some stuff right now, so can't check :-) )
05:58 karolherbst: ohh
05:58 karolherbst: bother :D
05:58 Psy-Q: so far all i did was put a fresh 64-bit debian on it grabbed the libdrm, mesa source
05:58 pmoreau: :p
05:58 karolherbst: :O
05:58 karolherbst: Psy-Q: usually _not_ a good idea
05:59 karolherbst: to grab debian for nouveau
05:59 Psy-Q: karolherbst: it's my home turf! :D
05:59 karolherbst: depends a lot on what debian though, but generally you should get something more bleeding edge
05:59 pmoreau: karolherbst: ...
05:59 Psy-Q: karolherbst: ah, it's debian testing
05:59 karolherbst: kernel version?
05:59 Psy-Q: 4.2 something
05:59 karolherbst: ahh okay
05:59 karolherbst: then its fine
05:59 Psy-Q: whew
06:00 karolherbst: though I would still recommend some ubuntu derivate like k or x
06:00 Psy-Q: i can also slap my own kernel in there if need be
06:00 karolherbst: there is a nice ppa for mesa and co
06:00 karolherbst: which updates the graphics stack to current master builds
06:00 Psy-Q: the oibaf one?
06:00 karolherbst: no
06:00 karolherbst: xorg-edgers
06:01 karolherbst: isn't oibaf bloated?
06:01 karolherbst: ohh it is the nine stuff
06:01 karolherbst: meh
06:01 Psy-Q: it might be, i only saw mesa on there once (when 11 wasn't anywhere else)
06:01 karolherbst: it is just mesa
06:01 karolherbst: no, xorg-edgers is the usefull one
06:01 Psy-Q: ok
06:02 Psy-Q: might slap a buntu on there at some point. i will see how much pain it is to roll my own mesa, if it's more useful to have a buntu for testing, that'll be that
06:02 karolherbst: it is pain
06:02 Psy-Q: i'm a bit terrified of compiling some of that stuff for i386 in case i want to test a 32-bit only game
06:02 karolherbst: in every regard
06:02 Psy-Q: ok :)
06:02 karolherbst: especially if you need the blob and nouveau co installed
06:03 karolherbst: Psy-Q: yes, that's even more painfull then
06:03 Psy-Q: awesome, i'm looking forward to that :D
06:03 karolherbst: :D
06:03 karolherbst: pmoreau: clock gating on tesla looks like a lot of fun
06:04 pmoreau: ;-)
06:05 karolherbst: what the heck is PBUS.DEBUG_6 ?
06:05 karolherbst: pmoreau: is [0] your embedded card?
06:06 pmoreau: I don't remember
06:06 pmoreau: Have a look at the value of reg 0 for each
06:06 pmoreau: The 96 is the discrete, and ac the integrated
06:07 karolherbst: 0 is g96 :/
06:07 pmoreau: (I pasted the value of 1588 for the MCP79, not the G96)
06:07 karolherbst: okay, the mcß79 one is more fun anyway
06:08 karolherbst: where?
06:08 pmoreau: 14:59:24 pmoreau | karolherbst: ...
06:08 karolherbst: ohh 0
06:08 karolherbst: yeah figures
06:08 karolherbst: okay, so it is disabled
06:08 karolherbst: nice
06:08 pmoreau: That's the MCP79
06:08 karolherbst: we can try stuff out then
06:08 karolherbst: maybe we can cut of 3W
06:08 pmoreau: Do you want it for the G96?
06:08 karolherbst: later
06:08 karolherbst: the card is off anyway
06:08 karolherbst: => not important
06:09 karolherbst: and ugly to deal
06:09 pmoreau: 3W out of 14W? That's already nice! :-)
06:09 karolherbst: maybe only 1.5W
06:09 karolherbst: don't know
06:09 karolherbst: there should be a difference though
06:09 karolherbst: how hot is your gpu?
06:10 pmoreau: Right now, 77 degrees C
06:10 karolherbst: uhh
06:10 karolherbst: I guess because of compiling?
06:10 pmoreau: Yeah
06:10 karolherbst: :/
06:10 pmoreau: The CPU is around 90
06:10 karolherbst: yeah, that's normal
06:10 karolherbst: mine is around 98 :D
06:10 pmoreau: :D
06:10 karolherbst: compiling libreoffice
06:11 karolherbst: though one core is only at 92 :/
06:11 karolherbst: lazy core
06:11 pmoreau: So your GPU should be hot as well
06:11 karolherbst: nope
06:11 pmoreau: Or, the integrated one at least
06:11 karolherbst: I don't have a crappy design
06:11 karolherbst: yeah well
06:11 karolherbst: the intel gpu is inside the cpu :D
06:11 Psy-Q: sorry if the question is dumb, but the pstate stuff is on kernel level (while loading nouveau.ko), so i would need a newer (git) kernel anyway to get the latest pstate code?
06:12 karolherbst: Psy-Q: nope
06:12 Psy-Q: (or an out-of-tree build?)
06:12 karolherbst: Psy-Q: 4.2 is fine
06:12 Psy-Q: karolherbst: ok
06:12 Psy-Q: thanks
06:12 Psy-Q: building mesa right now. hahaha
06:12 karolherbst: Psy-Q: you will need to compile your own kernel if you want it to be stable on your gpu :p
06:12 Psy-Q: i wonder!
06:12 karolherbst: nah, mesa is unimportant for that
06:12 karolherbst: pstate is only on a kernel level
06:12 Psy-Q: karolherbst: stability, who needs that!
06:12 karolherbst: :p
06:13 karolherbst: sadly ben didn't merge my gddr5 patch :/
06:13 karolherbst: and now he is like away for some days
06:14 karolherbst: pmoreau: I need 0x001530 and 0x001098 too
06:14 karolherbst: and 0x001590
06:16 pmoreau: 1530: 800412ff
06:17 pmoreau: 1098: 214a0004
06:17 pmoreau: 1590: ...
06:17 karolherbst: okay
06:17 karolherbst: tell me when you are done compiling then :D
06:17 karolherbst: I think this may be easy too
06:18 pmoreau: Might take a while: I'm compiling Unreal Engine 4 on a Core2Duo, and I started to compile it on Sunday's afternoon. :D
06:18 karolherbst: :D
06:18 karolherbst: ctrl+z
06:18 karolherbst: :p
06:19 karolherbst: and continue with fg
06:19 pmoreau: I'll probably do that ;)
06:19 karolherbst: :D
06:19 pmoreau: Discharge rate of 44W \o/
06:20 karolherbst: yay
06:20 karolherbst: well ctrl+z only pauses make
06:20 karolherbst: not the gcc invocations
06:20 pmoreau: I'll grab some water and let it converge
06:20 pmoreau: Well, the gcc invocations disappeared :-)
06:20 uxv: Hi. I'm having an issue with nouveau driver, here's my topic on a forum which describes the problem https://bbs.archlinux.org/viewtopic.php?id=203765 Thank you.
06:21 pmoreau: 44W was with the invocations
06:22 pmoreau: uxv: Is it something that appeared after an upgrade?
06:23 uxv: pmoreau: Well, afair errors appear after I updated to mesa 11.
06:24 pmoreau: Could you downgrade to a previous version of Mesa to check that?
06:25 uxv: pmoreau: But I'm not sure when it start, as I don't frequently use webgl sites.
06:25 pmoreau: And also, if you could give a link to a full dmesg (you can `cat /dev/kmsg > somefile`) for example
06:25 uxv: pmoreau: I don't know how to downgrade. Can you you help me with that?
06:25 uxv: pmoreau: ok...
06:25 pmoreau: If you still have a Mesa 10.x package in pacman's cache, it should be easy
06:26 pmoreau: Check in /var/cache/pacman/pkg/
06:26 uxv: pmoreau: Oh, no. I hae cleaned it with pacman -Scc :(
06:26 pmoreau: :D
06:26 pmoreau: To bad…
06:26 pmoreau: s/to/too
06:27 Psy-Q: hrmf, maybe xorg-edgers is more productive
06:28 Psy-Q: can't find nouveau_dri.so, i'm sure that has something to do with the amd64 diversions that multiarch does
06:28 pmoreau: uxv: The easiest then, is probably to compile your own Mesa using the PKGBUILD from the mesa package
06:28 uxv: pmoreau: Here's the today's boot dmesg: http://pastebin.com/yn5QdeaS
06:29 Psy-Q: i think i'll put in a second hard drive
06:29 pmoreau: uxv: Thanks! If you could get one with the error messages, it would be great
06:29 Psy-Q: one for nvidia-related stuff and gaming, one for nouveau experiments
06:31 uxv: pmoreau: Well, if I go to webgl site the messages added will be the same as I linked in the forum's topic.
06:32 uxv: pmoreau: Here's the example: http://pastebin.com/8jj6y8DD
06:33 pmoreau: karolherbst: It should have stabilised enough for some testing :-)
06:33 karolherbst: :)
06:34 pmoreau: uxv: Right, except you are missing the first error messages I think, which could shed some more light on the problem
06:34 karolherbst: so first we just try enable those
06:34 pmoreau: uxv: Anything in xorg.log.0 BTW?
06:34 karolherbst: pmoreau: poke 0x001590 0x00001500
06:35 karolherbst: and 0x001588 0x00000011
06:35 karolherbst: peek them to check if the values get set
06:35 karolherbst: then if they got set, check if the pwoer consumption changed
06:35 karolherbst: (I doubt it, but who knows)
06:35 uxv: pmoreau: Well, I can check the xorg log, but then I should to reboot :)
06:36 pmoreau: karolherbst: They are set
06:36 pmoreau: karolherbst: And no changes
06:36 uxv: pmoreau: If you'll wait I'll do that. ok?
06:36 karolherbst: pmoreau: poke 0x001098 214a0024
06:37 pmoreau: uxv: I will be there for the next 8 hours or so :-)
06:37 uxv: pmoreau: :) ok, now I'm quit for a while...
06:38 karolherbst: pmoreau: any change now?
06:38 pmoreau: karolherbst: Hard to say, it is oscillating between 14.7W and 15.2W.
06:39 karolherbst: mhh did the highest and the lowest value change
06:39 pmoreau: Ok, it just jumped to 16.3W… --"
06:39 pmoreau: And down to 14.8W
06:39 karolherbst: :/
06:39 karolherbst: without a dedicated power sensor on the gpu this is quite hard to tell
06:40 karolherbst: but if it's getting lower than usually
06:40 karolherbst: mhh
06:40 karolherbst: pmoreau: you could also try out poke 0x001530 0x800412fa
06:41 pmoreau: I killed X and the Wifi, hopefully it will be more stable
06:42 karolherbst: :D
06:42 karolherbst: then undo the reg pokes :p
06:42 karolherbst: 0x001588 and 0x001590 should get 0x0
06:42 karolherbst: 0x001530 0x800412ff
06:42 karolherbst: and 0x001098 0x214a0004
06:44 pmoreau: I'll check how it changes during the next five minutes and report back if it is stable enough
06:45 karolherbst: k :)
06:46 pmoreau: (I was going to say: it seems quite stable so far, only changing by 0.1W and then it jumped by 1.5W… --")
06:46 karolherbst: yeah :/
06:46 karolherbst: but it is also good to know these max and min values
06:46 karolherbst: this is usually enough
06:51 pmoreau: Ok, let's try again :-)
06:51 karolherbst: pmoreau: nvapoke 0x001530 0x800412fa; nvapoke0x001098 0x214a0024; nvapoke 0x001604 0x000a8a04; nvapoke 0x001604 0x000a8804; nvapoke 0x001588 0x00000001
06:51 karolherbst: and check if something changes
06:52 pmoreau: No poke of 1590?
06:53 karolherbst: not yet
06:53 pmoreau: Or 1530?
06:53 pmoreau: Ok
06:53 karolherbst: 1590 is mainly for the video engines
06:53 karolherbst: there is a 1530
06:53 karolherbst: :p
06:53 karolherbst: the first one
06:53 pmoreau: Oops :D
06:54 karolherbst: peek 0x00251c
06:55 pmoreau: 251c: 03f
06:55 karolherbst: k
06:55 karolherbst: any change in power consumption=ä
06:55 karolherbst: ?
06:55 pmoreau: Absolutely no changes so far
06:55 pmoreau: Or it is under 0.1W
06:57 karolherbst: there is so much in the trace related to those :/
06:57 karolherbst: painful
06:57 pmoreau: :/
06:57 karolherbst: poke 0x001590 0x00001500 and poke 0x001588 0x00000011
06:58 karolherbst: I also doubt this changes anything :/
06:59 karolherbst: there are some writes into SCRATCH3 for several engines :/
07:00 mupuf: there is a bug report you need to read to lower the power usage on tesla GPUs
07:01 mupuf: https://bugs.freedesktop.org/show_bug.cgi?id=37922
07:01 pmoreau: Oh right, that one
07:02 pmoreau: karolherbst: No changes with the additional pokes
07:03 karolherbst: mupuf: this sounds all messy and stuff
07:04 mupuf: comment 24 is what you are looking for
07:04 karolherbst: yeah, saw it already
07:04 mupuf: if you do not like it, then you do not like the hw but still have to deal with it :D
07:06 karolherbst: the thing is, I don't see the blob doing all those fancy bits on pmoreaus card
07:06 karolherbst: ohh wiat, maybe it does
07:06 karolherbst: mhh
07:06 karolherbst: strange
07:07 karolherbst: pmoreau: poke 0x001098 0x214a0064
07:07 mupuf:likes to poke it poke it
07:07 pmoreau: mupuf: :p
07:08 pmoreau: karolherbst: Slightly better, ~0.1–0.2W better
07:08 karolherbst: like in stable better?
07:09 pmoreau: Min and max
07:09 karolherbst: k
07:09 karolherbst: at least something :/
07:09 pmoreau: With the laptop being at min most of the time
07:09 karolherbst: mhhh
07:10 karolherbst: that's like 1% less power consumption :/ meh
07:10 karolherbst: mupuf: any tesla cards with power sensors?
07:17 mupuf: karolherbst: one
07:17 mupuf: my old old old old old laptop
07:17 karolherbst: :D
07:18 mupuf: the nva0 may have a voltage regulator with power output
07:18 mupuf: I need to scan it
07:18 mupuf:is a tad smarter than 3 years ago
07:19 karlmag: having more experience != smarter (per se... :-P )
07:20 mupuf: karlmag: I beg to disagree
07:20 mupuf: intelligence and knowledge go hand in hand
07:21 mupuf: pure cpu power do not get you far without the proper data
07:21 karlmag: I'm not saying that's a general observation btw, just that it doesn't have to be.
07:21 karolherbst: :D
07:21 mupuf: oh, true
07:21 mupuf: having data that you cannot handle is not helpful
07:21 karolherbst: mupuf: well we are not in the state, where cpus develop their own algorihtms yet :p
07:21 karolherbst: generally
07:21 mupuf: hehe
07:22 karlmag: hehe... and I think I'm happy it's that way, really :-P
07:22 karolherbst: karlmag: ohhh if you knew :D
07:22 karlmag: *shrugs*
07:22 karolherbst: karlmag: read about neural networks
07:22 karolherbst: and check where they can be used already
07:23 karlmag: I know a little bit about neural networks, though haven't revisited the subject for a few years so...
07:23 karolherbst: like music compising, writing articels, stuff :p
07:23 karlmag: garbage in, garbage out.. much like you get with lots of people, really.
07:23 mupuf: nothing new here though :D
07:23 karlmag: *shrugs*
07:23 karolherbst: :D
07:24 karolherbst: right
07:24 karlmag: mupuf: probably not, just bigger netorks
07:24 mupuf: right
07:24 karlmag: networks even
07:24 mupuf: and more training data
07:24 mupuf: I do not think human are any better
07:24 karolherbst: ohh also stock market news are usually already auto generated
07:24 mupuf: just that the network is bigger
07:24 karolherbst: so other computers can read them more easily :D
07:24 mupuf: and we have fake conscience
07:24 uxv: pmoreau: Hello again, my friend :) Here's the kmsg: http://pastebin.com/Gx3ijWw0 and here is the xorg.log: http://pastebin.com/3fx8SyGt
07:25 karlmag: mupuf: in the end we'll figure out it's just a part of the matrix :-P
07:25 karolherbst: karlmag: can we if nobody already broke out?
07:25 mupuf: if I can fly and avoid spending time in transport, I'm all for it :D
07:25 karolherbst: :D
07:26 karlmag: hehe
07:27 pmoreau: uxv: Thanks!
07:27 uxv: pmoreau: What's the diagnosis?
07:27 uxv: pmoreau: You wellcome :)
07:28 pmoreau: I have no idea what PBDMA0 means
07:28 pmoreau: And there seems to be something wrong with the bo
07:28 uxv: Huh :)
07:28 uxv: With what? What is 'bo'?
07:29 pmoreau: But there should have been some fixes recently regarding bos
07:29 pmoreau: Buffer Object would be my guess
07:29 pmoreau: imirkin: -^
07:29 uxv: I see.
07:32 pmoreau: Maybe this one is somewhat related? https://bugs.freedesktop.org/show_bug.cgi?id=89985
07:33 uxv: pmoreau: Should I downgrade to previous mesa version? Or I can wait for next official Arch build. I don't frequently use webgl.
07:35 pmoreau: uxv: Does OpenGL programs work BTW?
07:36 uxv: Yes, OpenGL programs work fine. I use Blender almost every day.
07:38 pmoreau: uxv: You should open a bug report at bugs.freedesktop.org, under product=Mesa, component=Drivers/DRI/Nouveau, and attach the kmsg and xorg.log files to it (attach the files directly, do not put a link to pastebin).
07:38 uxv: It's somewhat connected to browser/webgl only, I think. I ran OpenGL apps in QEMU recently and there were no errors or hangs.
07:39 pmoreau: I have no idea what the status is regarding WebGL on Nouveau.
07:39 uxv: pmoreau: ok, I will open the report :)
07:40 pmoreau: uxv: Thanks! I do not know much about that part of Nouveau.
07:40 pmoreau: Will be easier for the other developers to get back in touch with you.
07:40 uxv: pmoreau: Thank you very much for your patience :)
07:40 pmoreau: You're welcome :-)
08:02 karolherbst: mupuf: do you still want to look over my pcie patches?
08:02 mupuf: sure!
08:02 mupuf: something that we should be able to land :D
08:03 mupuf: for a change :D
08:03 karolherbst: :D
08:03 karolherbst: ben is now gone for a week :/
08:03 mupuf: not tonight though, I have a joe satriani concert
08:04 karolherbst: k
08:04 karolherbst: who is in charge to decide if the pstate file should be moved to debugfs?
08:05 imirkin: ultimately ben, but the current pstate file is not appropriate for sysfs
08:05 karolherbst: imirkin: yeah, but I think he doesn't care that much about that
08:05 karolherbst: I mean where this file is
08:06 karolherbst: imirkin: if you want it to be moved inside debugfs, then you could take a look at http://lists.freedesktop.org/archives/nouveau/2015-October/022641.html if you have time ;)
08:07 benwaffle: Hello. I switched from nvidia's driver to nouveau, and now gnome freezes. I booted a live CD (kernel 4.2.3) and dmesg shows: GPU lockup - switching to software fbcon
08:08 mupuf: karolherbst: : no-one does, but the rules are there...
08:10 karolherbst: imirkin: we could leave the empty sysfs infrastructure though :/
08:11 imirkin: benwaffle: if you're interested in debugging, provide logs. otherwise switching back to blob might be your best bet.
08:15 benwaffle: imirkin: i would but now i can't switch to a tty :/
08:17 imirkin: benwaffle: ssh can be your friend
08:17 imirkin: benwaffle: what gpu do you have btw?
08:19 benwaffle: GTX 460v2
08:19 imirkin: hm, weird. not aware of any problems that'd have with recent kernels :(
08:28 pmoreau: uxv: Could you please attach each file separately as a text attachment, rather than having an archive?
08:29 uxv: pmoreau: :) ok.
08:29 pmoreau: That way anyone can read them from the browser, without having to download + extract the archive
08:33 uxv: pmoreau: Ok, I've done. Let's hope that the issue soon will be resolved.
08:33 uxv: Da link: https://bugs.freedesktop.org/show_bug.cgi?id=92556
08:33 imirkin: unlikely. i haven't the faintest clue what those messages mean or what could cause that.
08:35 pmoreau: :/
08:36 uxv: :(
08:37 uxv: ok, guys, thank you and hope we'll meet again.
08:46 benwaffle: imirkin: i will get a full log later, but idk if this helps: http://pastebin.com/rWAUWGdE
08:46 benwaffle: it gets printed over and over
08:47 imirkin: still have no idea what that means
08:48 imirkin: it sounds like there's something quite nasty going on
08:48 imirkin: you didn't happen to switch to nouveau because nvidia blob was all crashing did you?
08:49 benwaffle: on
08:49 benwaffle: no
08:49 benwaffle: nvidia worked just fine
08:49 benwaffle: i wanted wayland though
08:50 imirkin: ah
09:32 karolherbst: mupuf: it seems like that these automatic clock gates are usually not disabled at runtime. Only for PCIPHER :/
09:34 imirkin_: karolherbst: PCIPHER is part of a mega-engine that handles that and all the video decoding stuff. just fyi.
09:34 karolherbst: sounds like it
09:34 karolherbst: like for dvd drm stuff?
09:36 karolherbst: ohh found one where PGRAPH is also set to non auto :/
09:36 imirkin_: the cipher stuff? not dvd, hdcp i think
09:37 karolherbst: ohh okay
09:37 karolherbst: thats weird now
09:37 imirkin_: not 10000% sure
09:37 karolherbst: found a trace where the blob changes these clock gates reg every second
09:37 karolherbst: for about 1000 seconds
09:37 karolherbst: even longer
09:38 mwk: PCIPHER is AES actually
09:38 karolherbst: even faster
09:38 mwk: and also somewhat involved in HDCP
09:38 karolherbst: mhhh
09:38 imirkin_: "also"? is it involved in anything else?
09:38 karolherbst: I don't think it is actually required to disable those automatic clock gates
09:38 karolherbst: it may be a performance thing
09:38 mwk: imirkin_: AES
09:38 karolherbst: I noticed a perf drop after enabling those on my card
09:39 imirkin_: mwk: sure, but... where is that used other than HDCP in a graphics context?
09:39 karolherbst: and perf drop means like 0,1%
09:39 mwk: video decoding, I think
09:39 imirkin_: mwk: really? video decoding not in conjunction with hdcp?
09:40 imirkin_: i thought it was to be able to pass encrypted videos through to hdcp or... something
09:40 imirkin_: i guess i don't really know how that flow works
09:40 mwk: sw decrypts dvd through css or whatever, reencrypts with aes
09:40 karolherbst: mwk: I thought the css stuff is on top of something else? don't know
09:40 karolherbst: why should it be reencrypted?
09:40 mwk: because GPU doesn't know css
09:41 karolherbst: :D
09:41 karolherbst: how broken is css by the way? :D
09:41 mwk: and content needs to be encrypted between cpu and gpu due to bullshit
09:41 mwk: karolherbst: thoroughly
09:41 karolherbst: bullshit like in contracts?
09:41 mwk: yep
09:41 karolherbst: :D
09:41 karolherbst: my godness
09:41 karolherbst: I don't wonder anymore, why cracking css is legal
09:42 mwk: gpu in turn decrypts aes, decodes stuff, encrypts through hdcp, sends to monitor
09:42 mwk: karolherbst: css cracking is likely not legal
09:42 karolherbst: it is
09:42 karolherbst: at least here
09:42 mwk: it's just that nobody cares anymore
09:42 karolherbst: if crackins something is childsplay it is legal
09:42 karolherbst: no, by law it is legal
09:42 mwk: depends on the location, I suppose
09:45 karolherbst: mwk: I am pretty sure it is legal in entire europe at least
10:00 hannson: what's an easy way to shift the screen a bit to the left? there's like a little bar of a few pixels that's just pink and doesn't do anything...
10:07 imirkin_: hannson: what gpu and what kernel?
10:08 imirkin_: hannson: that's an indication that some hdmi settings are mis-set
10:08 imirkin_: [i assume the screen is connected over hdmi...]
10:08 hannson: gtx 770 and 4.2.3-1
10:08 imirkin_: hmmmmmmm
10:09 imirkin_: that should work fine :(
10:09 hannson: imirkin_: it's connected over display port
10:09 imirkin_: oh goodie. i wonder if we're missing something for DP
10:09 hannson: imirkin_: sorry typo: gtx970
10:10 imirkin_: oh
10:10 imirkin_: that's quite a typo
10:10 imirkin_: there were a number of DP-related fixes for maxwell recently
10:11 imirkin_: doesn't look as though they went into 4.2 -- try the 4.3-rc's
10:11 hannson: imirkin_: thanks, will do!
10:12 imirkin_: although those might have only been for GM10x... dunno
10:13 imirkin_: it might need a new eld thingie
10:13 imirkin_: are you plugging into a TV or otherwise audio-enabled device?
10:14 hannson: imirkin_: no
10:14 imirkin_: hm.
10:14 imirkin_: perhaps that's the audio signal and your device isn't interpreting it? dunno
10:16 imirkin_: yeah, i bet we need a new hdagm204.c
13:51 RSpliet: gosh darnit...
13:51 RSpliet: I don't know what about you guys, but to me
13:51 RSpliet: http://hastebin.com/ekiwaboyik.coffee
13:51 RSpliet: looks like a test pattern
13:51 imirkin_: i know this pattern
13:51 imirkin_: comes from vbios
13:51 imirkin_: during link training
13:52 RSpliet: yeah, although I take it it's "upload once"?
13:52 imirkin_: not sure what you mean
13:52 RSpliet: if you look at gt215, you see there's an init sequence where a test pattern is uploaded
13:52 RSpliet: then link is trained (hmm, again once come to think of it)
13:53 RSpliet: and never again until reboot
13:53 imirkin_: http://cgit.freedesktop.org/~darktama/nouveau/tree/drm/nouveau/nvkm/subdev/fb/ramgk104.c#n1271
13:53 imirkin_: look familiar?
13:53 RSpliet: yeah
13:53 RSpliet: I guess we'll have to rename that gf100_ram_train_init_0
13:57 imirkin_: making some more progress?
13:58 RSpliet: random observation really
13:58 RSpliet: I need to focus on GDDR5 MR generation though
13:58 imirkin_: ah
13:58 RSpliet: could benefit your GT215 as well ;-)
13:58 imirkin_: iirc i already did it for my GT215
13:58 karolherbst: wow
13:58 RSpliet: all of it?
13:58 karolherbst: my found for today: GDDR3 kepler card
13:58 karolherbst: GK107
13:59 RSpliet: none of that is upstream apparently
13:59 imirkin_: RSpliet: https://github.com/imirkin/nouveau/commit/a41f62639b14bab1f306241445398f525f8e47b7
13:59 karolherbst: NVS 510
13:59 imirkin_: probably not all of oit
13:59 imirkin_: but some of it :)
13:59 karolherbst: or is it only DDR3?
13:59 RSpliet: imirkin_: yeah, that's as far as I got :-P
13:59 karolherbst: mhh, wiki is wrong again
14:00 imirkin_: ?
14:00 RSpliet: there's a couple more bits that require touching apparently
15:34 mupuf: never heard of PCIPHER before :D
15:34 mupuf: I guess it is the ancester of PCRYPT
15:34 mupuf: which got introduced in G98 IIRC
15:34 mupuf: first fuc engine too
15:35 imirkin_: mupuf: it *is* PCRYPT, just renamed
15:35 mupuf: ack
15:35 mupuf: but it is not the same hw implementation anyway
15:36 imirkin_: ben literally s/PCRYPT/PCIPHER :)
15:36 mupuf: oh!!!
15:37 mupuf: I prefer our old name :D
15:37 mupuf: it is one letter shorter :D
15:37 mupuf: but it also decrypts... so .... meh
15:37 imirkin_: and copies
15:38 mupuf: true too
15:38 mupuf: does it handle other memory formats that linear?
15:39 imirkin_: mmmmaybe
15:39 imirkin_: i think so actually
15:40 imirkin_: i guess not
15:40 imirkin_: but it's used to move entire bo's, so tiling doesn't matter
15:46 mupuf: well, it must be to copy super small buffers then :D
15:46 imirkin_: why?
15:47 mupuf: if tiling does not matter, it means that you copy buffer that are the size of the width of the first level of tiling :D
17:54 imirkin: sweet, the binary patching really fixes all the interpolation tests
17:54 imirkin: just need to figure out how to do the same thing on nv50... hm
19:28 phire: nvidia does binary patching?
23:35 imirkin: phire: not 100% sure... it definitely messes up on some of those piglit tests though :)