00:21t-ask: Mmm, maybe that pixel shift relates to my triple head setup where one display has a higher resolution than the others
00:22t-ask: I may try it another day
00:27imirkin: thoroughly unlikely
00:45imirkin: hakzsam: hm interesting. i wonder why NV_vdpau_interop disappeared on the NV96
00:50imirkin: hakzsam: actually it's gone from all of your things... i wonder if you just don't have the firmware
00:50imirkin: or perhaps you built without vdpau
04:40gnurou: dmj_s76: I am maintaining a (working) backport of 4.7's (and soon 4.8) DRM to 4.4 if you are interested
04:40gnurou: ah, he left...
06:30s0be: I'm experiencing rcu stalls on linux next on a gm107. The regression is somewhere between 20160617 and now. Anyone know any changes that have gone in that might cause this right off?
06:30s0be: the stall is in drm_open during pmops
06:31s0be: looking at the git history for drivers/gpu/drm/nouveau, I'd guess that it's not actually directly related to a change in this driver, but a more general pci pm change :-/
07:05hakzsam: imirkin, I build without vdpau
08:57pecisk: karolherbst_work: any help from Nvidia publishing those firmwares? WIll be able to reverse re-clocking for Pascal? ;)
09:01Calinou: it'd be good if Pascal firmware landed this year
09:01Calinou: but I'm not sure about that :/
09:02pecisk: Calinou: what are those? :) http://phoronix.com/scan.php?page=news_item&px=NVIDIA-GP100-Firmware
09:04Calinou: but only GP100?
09:04Calinou: so, are GTX 1080/1070 impacted by this?
09:10pmoreau: Calinou: No, 1080 is GP104, and 1070 is not GP100 for sure (probably a GP104 as well, not sure).
09:15pecisk: which is GP100? :)
09:16Calinou: a Tesla card
09:16Calinou: so, not a consumer card
09:27RSpliet: pmoreau: iirc 1070 is GP104, 1060 will be GP106
09:28pmoreau: Seems reasonnable, I wasn’t sure
10:21karolherbst_work: mupuf_: yay, got comments for every patch now :p
10:21karolherbst_work: mupuf_: I hope we can finish my reclocking stuff this or next week
10:45mupuf_: karolherbst_work: is it public?
11:01karolherbst_work: mupuf_: nope
11:01karolherbst_work: mupuf_: but I guess I fix that stuff and send an v5 out
11:03karolherbst_work: mupuf_: but I could forward it to you if you want
11:03mupuf_: no, I can wait for the v5, just document your changes and who asked for it
11:07karolherbst_work: skeggsb_ wants to keep the mclk requiernments idea though :/
11:07karolherbst_work: and I am like pretty sure we won't need it that much actually
11:08karolherbst_work: and that we keep nvkm_clk_tstate so that therm can set the current temp for clk
11:09karolherbst_work: well, I need to go through that stuff over the week
11:10mupuf_: karolherbst_work: great! Don't hesitate to talk to yourself here, I will see if I have some feedback for you!
11:11karolherbst_work: but now I have to work first :p (debugging a silly sqlite db issue)
11:14mupuf_: karolherbst_work: have fun!
11:15karolherbst_work: well that task won't be fun, stupid db thought it may be a good idea to go into read only mode for whatever reason
12:52pmoreau: karolherbst_work: As you have been playing with PCIe-speeds & co., does this error "pci: failed to adjust cap speed" ring a bell? (from https://bugzilla.kernel.org/show_bug.cgi?id=119831)
15:16dmj_s76: skeggsb_: The latest commit allows nouveau to bail on Pascal (and other future devices), correct?
15:17imirkin_: he's asleep now, but yes, it should.
15:17imirkin_: and it should be moderately easy to backport to earlier kernels
17:15imirkin_: mogorva: did force_glsl_version=130 help with Hoard?
18:23dmj_s76: imirkin_: Just tested the latest commit, and it seems to be an improvement, but still some issues
18:24imirkin_: did you test with JUST the latest commit, or the whole branch?
18:24Yoshimo: what could cause my monitor to report some kind of undervolting issue when switching from the textbased console to the graphic login?
18:25imirkin_: Yoshimo: what kind of monitor?
18:25imirkin_: and what exactly does it report?
18:25dmj_s76: Before the screen would blank before boot completed, whereas now it will boot to where you can access VTs but no functional graphical session.
18:25imirkin_: dmj_s76: pastebin dmesg + xorg log
18:25Yoshimo: it is a AOC flatscreen connected by a displayport i think
18:25dmj_s76: Just the latest commit applied to the Ubuntu kernel
18:26imirkin_: Yoshimo: dunno. if it were a CRT, could be a transformer inside (where switching modes actually requires work to be done)
18:27imirkin_: dmj_s76: perhaps you don't have xf86-video-fbdev installed?
18:27imirkin_: (the logs will make the issue clear though)
18:28Yoshimo: let me reboot and take a picture
18:34karolherbst: sooo. patch fixing time
18:54karolherbst: are changes in the commit messages worth mentioning in the changed commit message as a vN: ... message?
18:55dmj_s76: imirkin_: dmesg: http://paste.ubuntu.com/19200779/
18:56dmj_s76: imirkin_: Xorg.0.log: http://paste.ubuntu.com/19200823/
18:56karolherbst: I think I will put in the cover letter as "v5: reworded some commits, blah blah blah"
18:56imirkin_: dmj_s76: what's that second gpu in there? i wonder if it's upsetting matters
18:57dmj_s76: imirkin_: There's only one gpu
18:57imirkin_: oh, the ast output
18:58imirkin_: nah, there's the aspeed thing in there. probably on the motherboard.
18:58imirkin_: for the remote kvm, i'd guess
18:58imirkin_: soooo ... not easily disabled.
18:58dmj_s76: dual cpu workstation mobo
18:59imirkin_: can you put a quick xorg.conf together that forces it to only use the nvidia chip?
19:00imirkin_: basically looks like xorg crashes for some reason
19:00imirkin_: somewhere in randr-land... i wouldn't be surprised if it were as a result of that "gpu" screen driven off the AST
19:00imirkin_: or actually, you can just add Option "AutoAddGPU" "off"
19:01imirkin_: into ... serverlayout?
19:07imirkin_: i wonder if it'd be reasonable to create an efi kms driver... dunno what all efi lets you do, graphically
19:09imirkin_: airlied: have you ever thought about doing that? looks like pageflip would be a pretty expensive operation, but ... doable
19:10imirkin_: (since it would have to blit the whole screen)
19:18Yoshimo: so the text that moves in blue over the screen is "N. unterst. Eing" which i would assume means low undervolted Input" or something along that line
19:23dmj_s76: imirkin_: I just moved the card and ssd over to another computer (different motherboard) and it seems to work with the patched kernel
19:24imirkin_: dmj_s76: ok. so i'd GUESS that the issue lies somewhere between xf86-video-fbdev, xserver, and xf86-video-modesetting
19:24imirkin_: my guess is that this is not an oft-tested combination
19:24dmj_s76: oh, and using the nomodeset option works with the workstation board
19:25imirkin_: right, since that nukes the ast driver
19:26imirkin_: in any case, the issue has nothing to do with nouveau :)
19:40airlied: imirkin_: efi doesn't really let you do anything once you exit boot services
19:40airlied: can't even change the mode
19:41dmj_s76: imirkin_: For that motherboard there's actually a jumper to enable/disable the ast2400
19:42dmj_s76: boots fine with the patch and the jumper set to disable.
20:02karolherbst: skeggsb: your comments about the line length are odd, because (at least here) I wasn't above 80 in any case (7 first patches checked)
20:06karolherbst: mupuf_: :O ben NAKed my awesome volt compare tool...
20:14dmj_s76: imirkin_: Thanks for your help!
21:11inglor: karolherbst: thanks for the answer (2 days later)
21:13inglor: aparently my 760W psu can't handle the 4-way SLI (in windows that is!) and the PC turned off once :P
21:13inglor: So I removed one of the 690
21:13inglor: so now sitting next to me. don't think I will use it
21:13inglor: (hint hint)
21:15karolherbst: well, if you find people who would have a greater use of that card, feel free to ask them if they might want to free you from the card (which I think, may is blocking your way to your door, or jumps randomly at you :p )
21:15inglor: I have some in mind - they are working in open source drivers for nvidia :D
21:16karolherbst: I know that mupuf_ is quite _the_ nvidia card collector and is paying especially good for rare models
21:16inglor: I thought everyone in here were into nvidia cards (I mean what else nouveau is for? :D)
21:17karolherbst: I only have like 1 nvidia gpu
21:18karolherbst: from the "hardwareWanted" page: "Any NV3X/4X with a PCI/PCIE port (save NV4B and NV43), NVA with GDDR3 or 5 memory, NVAA/AC/AF, NVC3/D7, NVE4/F0/F1/106 or NV118/120/124/12B, I'm willing to pay a good price." :p
21:18inglor: Also I haven't been able to use it for compute capabilities as 1) I lost interest in cuda about 1 year ago, 2) h264x encoding is not _really_ better.
21:18karolherbst: that's from mupuf_
21:19inglor: mupuf_: ping
21:20karolherbst: inglor: I bet your 690 also have fancy LEDs?
21:20inglor: that's a NVE4 right ?
21:20inglor: some not so fancy.. but yeay
21:20karolherbst: 2x nve4
21:20inglor: it's green and says GEFORCE GTX
21:21karolherbst: no shit
21:21karolherbst: that's like a super rare model!
21:21inglor: needs 2x 4 pins!
21:21inglor: these are hungry! :(
21:21karolherbst: it has LEDs
21:21inglor: oh yeah
21:21inglor: I'm sure all the power goes to LEDs
21:23karolherbst: if you pay like 1k for a gpu, the power bill doesn't matter that much anyway anymore
21:23inglor: TDP 300W
21:23inglor: yeah good thing I got them for free :D
21:23inglor: I'm not that impressed
21:24inglor: my Asus 670 was way quiter...
21:24inglor: and not so bad in performance
21:24inglor: this one.. boy it's Hot! (and loud)
21:24inglor: so back to the branch you mentioned
21:24inglor: stable kepler 5
21:24inglor: how can I use the branch?
21:25inglor: I'll probably need to recompile kernel
21:25inglor: right? :D
21:26karolherbst: just run make inside drm
21:26karolherbst: and you will get a drm/nouveau/nouveau.ko file
21:26inglor: sound easy..
21:27karolherbst: but I am currently a bit messing with that branch, but will keep it clean (I hope)
21:27karolherbst: well not yet anymore
21:27karolherbst: but over the week
21:28inglor: naive question. .ko modules are a bunch of linked .o files ?
21:28karolherbst: ko files are kernel modules
21:29inglor: ok cool I have a nouveau.ko file
21:30inglor: now I need to properly switch to nouveau - copy over the new file and reboot ? :D
21:30inglor: I'm using nvida now
21:30karolherbst: you have a nouveau.ko(.xz) file inside /lib/modules/$kernel_version/...
21:31karolherbst: this one has to be removed and the compiled one has to be copied there
21:31inglor: most probably no (not yet) as I'm compiling my own kernel with only some modules
21:32karolherbst: you would need to build nouveau again for the new kernel
21:32inglor: makes sense
21:33inglor: It's ok I've done it couple of times
21:33inglor: install mesa-libgl (this removes nvidia). download and recompile kernel make sure nouveau is checked. and reboot.
21:33inglor: then get into modules and replace the .ko file - another reboot
21:35inglor: mupuf_ is asleep..
21:35karolherbst: I doubt that, because the sun is still up for him :p
21:35inglor: ah US?
21:36karolherbst: the sun was up for me 1.5h ago too
21:36inglor: ok the other side of atlantic..
21:36karolherbst: just a bit in the north
21:36inglor: It's going to be hard to post the card.. from uk
21:36karolherbst: sunlight until 10:30pm is something :p
21:37karolherbst: you could send it to RSpliet
21:37inglor: yeah london is similar (till 9.00)
21:37karolherbst: which should be the best idea actually
21:37inglor: I've only been able to catch you online! :D
21:38karolherbst: try to catch RSpliet
21:38inglor: I only logging late to my VPS that's why
21:38inglor: got it! ./4.6.4-1-ck/kernel/drivers/gpu/drm/nouveau/nouveau.ko.gz
21:39inglor: gz not xz (kernel is still gunzipping stuff)
21:40karolherbst: anyway, RSpliet is in the UK too until XDC, so everything is good
21:40karolherbst: I think
21:41inglor: RSpliet: ping!
21:41inglor: there I pinged
21:41inglor: I need emails! :(
21:41karolherbst: the mailing list is full of them :p
21:41inglor: ah mailing list
21:42inglor: why didn't I think of that earlier..
21:42karolherbst: https://lists.freedesktop.org/archives/nouveau/2016-June/025286.html :p
21:43karolherbst: and now I will head to bed :p
22:11imirkin_: airlied: oh well. didn't realize the boot services distinction.
22:12imirkin_: dmj_s76: seems like asking people to flip a jumper before installing also seems suboptimal. ideally someone would track down what the fbdev + modesetting badness is...
22:17dmj_s76: imirkin_: So far there's a limited number of customers with the jumper set that way who have a pascal card.
22:19dmj_s76: Potentially just flip the jumper before shipping systems in future though. I suspect very few customers will use the onboard ast2400.
22:20imirkin_: well presumably it's there for the kvm? or is it meant for the primary graphics?
22:23dmj_s76: imirkin_: It's an internal motherboard port which goes to an optional vga bracket
22:23imirkin_: ah ok. i assumed there was a drac-equivalent on there
22:24imirkin_: RAS? i forget what these things are called in non-dell land
22:24dmj_s76: In theory you could not have to install a GPU, but this particular machine is commonly used for Deep Learning GPU computing and the like.
22:25imirkin_: well, presumably some of those people will get the crazy tesla gpu's
22:25imirkin_: which tend not to have video outputs
22:26dmj_s76: Some, yes. The Titans and high-end GeForce cards are also popular.
23:16mupuf_: inglor: nope, not asleep :D
23:16mupuf_: just away