02:14 hakzsam: orbea, did you see errors in dmesg?
03:14 qwertz_: hello! i'm trying to setup PRIME with nouveau and i915. both modules are loaded and i have been able to perform a switch. however, xrandr --listproviders only shows me one provider.
03:15 qwertz_: is this related to the discrete card being "DynOff" as available through vgaswitcheroo?
03:15 karolherbst: qwertz_: do you want to use a display on the nvidia gpu?
03:16 karolherbst: if not, then just setup dri3 and you don't have to mess with xrandr anymore
03:16 karolherbst: it could be that you don't have the nouveau ddx or something else went bad
03:16 karolherbst: so
03:16 karolherbst: dmesg, xorg.log would be nice
03:18 qwertz_: karolherbst, my digital outputs are hardwired to the NVIDIA card. i want to be able to switch from an energy saving traveller setup (LCD + Intel) to a docking station home setup (external screen + NVIDIA) without rebooting and changing cards in the BIOS.
03:18 qwertz_: so yes, there will be a display on the NVIDIA card.
03:18 qwertz_: logs will follow.
03:19 karolherbst: ohh okay
03:19 karolherbst: then you need dri2 to work
03:21 qwertz_: dmesg: http://pastebin.com/ppAu0yb5; xorg.0.log: http://pastebin.com/xsM9jDkd
03:22 qwertz_: karolherbst, thanks for your assistance!
03:22 qwertz_: this whole graphics setup is a bit out of my comfort zone...
03:25 qwertz_: systemd says something this: Jan 06 12:01:49 AERechenknecht kernel: nouveau E[ PBUS][0000:01:00.0] MMIO read of 0x00000000 FAULT at 0x002140 [ !ENGINE ]
03:26 karolherbst: doesn't really matter
03:26 karolherbst: the kernel part looks fine
03:26 karolherbst: okay the xorg server part looks bad
03:26 qwertz_: a common cause of missing providers seems to bad xorg.confs. however, it seems my xorg starts without config file.
03:26 karolherbst: qwertz_: you need to install the nouveau xorg server driver package
03:27 karolherbst: like you have an intel one too
03:27 karolherbst: xserver-xorg-video-nouveau for ubuntu I think
03:28 qwertz_: karolherbst, "xserver-xorg-video-nouveau is already the newest version."
03:28 qwertz_: they're both there!
03:28 karolherbst: mhh doesn't make sense
03:28 qwertz_: karolherbst, that's what i understand so far as well...
03:29 karolherbst: qwertz_: content of your /usr/lib/xorg/modules/drivers/ directory
03:29 qwertz_: there may be something related: i was having severe problems booting my laptop only on the NVIDIA card. i guess something related to ACPI. i can only boot if i also have the intel card enabled.
03:30 qwertz_: karolherbst, contains this: -rw-r--r-- 1 root root 215K Jun 9 2015 nouveau_drv.so
03:31 karolherbst: okay
03:31 karolherbst: then your xorg.conf file if you have any
03:31 qwertz_: karolherbst, here's the full dir: http://pastebin.com/MqhYz7yc
03:31 qwertz_: karolherbst, i have no xorg.conf
03:32 qwertz_: at least not in /etc/X11, which is the right place to look, isn't it?
03:33 karolherbst: yeah
03:33 karolherbst: mhhh strange
03:35 qwertz_: as mentioned before, i had severe problems with the NVIDIA card... is there any chance that this might be hardware-related?
03:35 karolherbst: don't think so
03:35 karolherbst: the xserver sees the nvidia card
03:35 karolherbst: but it simply doesn't load the nouveau driver for it
03:37 qwertz_: so i need to distinguish kernel and xserver here? the kernel apparently loads the module, but then X can't make any use of that?
03:38 karolherbst: qwertz_: do you have libdrm-nouveau2 installed?
03:38 karolherbst: qwertz_: yeah something like that
03:38 qwertz_: karolherbst, yes, is installed.
03:39 karolherbst: strange, very strange
03:39 qwertz_: could this be a timing issue?
03:39 karolherbst: no
03:40 karolherbst: maybe you have any leftovers in /usr/share/X11/xorg.conf.d or /usr/share/X11/xorg.conf or /etc/X11/xorg.conf.d
03:40 qwertz_: does it make any sense to try upstart instead of systemd once?
03:40 karolherbst: nope
03:40 qwertz_: karolherbst, i'll check those directories. i have fiddled around a lot.
03:41 qwertz_: /usr/share/X11/xorg.conf.d contains no file related to nouveau. grep finds no mentions of [nN]ouveau.
03:42 qwertz_: /usr/share/X11 has no xorg.conf
03:42 karolherbst: well there doesn't need to be a reference to nouveau directly
03:42 karolherbst: could be nvidia as well
03:42 karolherbst: or just the pci id
03:44 qwertz_: hm, a first inspection tells me the contens of /usr/share/X11/xorg.conf.d look innocent too
03:45 karolherbst: mupuf: something against installing some dependencies for ffmpeg on reator blob side?
03:45 karolherbst: I want to test out some nvenc stuff there
03:47 qwertz_: karolherbst, no mentions of the PCI ID in either /etc/X11 or /usr/share/X11
03:48 karolherbst: meh :/
03:49 karolherbst: mupuf: ohh seems like disabling yasm was enough
03:49 karolherbst: and uploading the nvEncodeApi.h file :D
03:53 qwertz_: karolherbst, could it be beneficial to manually create a xorg.conf? is this the file that tells X which modules to load?
03:53 karolherbst: the xorg-server should load the stuff automatically
03:53 karolherbst: airlied: any ideas?
04:23 khazakar: Hi all!
04:24 khazakar: I have a unexpected (imo) trouble with my GTX 760 running under nouveau -> yesterday engine reclocking worked like a charm (with choose 0e/0f pstate),but today only memory is reclocked,not "engine"
04:26 khazakar: Engine is not reclocked even I run gpu stress.
04:38 duelle: I would like to set up a three screen setup on my Lenovo T530 (1x internal screen, 2x external screens [DVI]) using reverse PRIME as the notebook has a Intel HD4000 and Nvidia NVS5400 built in. The DVI ports are connected to the Nvidia card. The xrandr command for revPRIME works fine, but when arranging the screens I get an error message: http://pastebin.com/q2hXjfdd
04:45 khazakar: Hm...edid-decode?
04:47 duelle: khazakar: Doesn't that have to do with modes of the screens? At first sight they seemed to be ok (resolution, rate).
04:48 duelle: Additionally I can add each screen separately, but not both at the same time.
04:48 duelle: So I can't enable them to be active at the same time.
04:49 khazakar: Did you tried lxrandr? to detach 2 screens to the same source? You want to see the same on both screens,right?
04:50 khazakar: Or enhance your view? You can do it with lxrandr imo
04:50 duelle: khazakar: tbh I never heard of lxrandr. My intended screen setup is | DVI1 | LVDS1 | DVI2 | all next to each other with the notebook screen in the middle. So no - I do not want the same on both external screens.
04:51 duelle: I tried arandr before - but a guy here told me that it seems to mess with position attributes. So I tried to do it manually using xrandr
04:52 khazakar: IMO lxrandr would do the job. I can connect to my Dell E650 with Quadro NVS160M my FHD monitor using VGA and I can place the screen like want to do with simple click
04:53 qwertz_: duelle, out of interest: does your xrandr --listproviders give you two providers? we have the same graphics setup (T430 here) and i'm having problems with the NVIDIA.
04:53 duelle: khazakar: I guess then arandr is a similar tool.
04:54 duelle: qwertz_: Currently I have two providers: Intel and nouveau.
04:55 qwertz_: duelle, thanks! then your thinkpad does something mine fails to do...
04:55 duelle: khazakar: But using the arandr tool it leads to the same error message as in the paste above.
04:55 duelle: qwertz_: I had that issue only showing one provider when explicitly setting the devices in the xorg.conf
04:56 qwertz_: duelle, i don't use xorg.conf...
04:56 qwertz_: duelle, which distro are you using?
04:56 duelle: qwertz_: Arch Linux. I don't have xorg.conf stuff in place now. That was just another try to get my setting to run. But in that case I only had one provider listed.
04:58 duelle: Might be that I have to relogin .. gonna give lxrandr a try.. brb
05:00 khazakar: Hm...I have a problem with engine reclock on my GTX 760 - yesterday worked like a charm,today not.
05:00 duelle: lxrandr seems not to activate the second external screen at all. If I set it the way I want it to be, it just mirrors the main screen on the left one and the right one stays black. No error messages though.
05:01 duelle: I have to correct: Same error message
05:01 duelle: "xrandr: Configure crtc 5 failed"
05:02 khazakar: Hm...
05:06 khazakar: Have you DisplayPort?
05:07 khazakar: DisplayPort output*
05:07 duelle: khazakar: Kind of - each of the DVI ports has a DP port in parallel.
05:08 khazakar: "For triple-monitor support you need to be aware that most non-DisplayPort graphics cards only support two simultaneous outputs, because they only have two PLLs, and hence can generate two pixel clocks."
05:08 duelle: khazakar: From the hardware side this should be no problem. Before reinstalling my notebook I had a working setup using bumblebee. So the hardware limitations should be no problem.
05:09 duelle: ..hopefully
05:09 khazakar: So It's only software-related? Are you using nouveau or blob?
05:09 duelle: nouveau
05:11 duelle: I guess that the driver (or whatever) can't really address both screens at once - or something similar. As I can operate each at a time - but not both at the same time. Problably some setting/tweak..
05:11 khazakar: Hm...this? http://nouveau.freedesktop.org/wiki/MultiMonitorDesktop/
05:13 khazakar: BTW you can use bumblebee with nouveau :P
05:13 duelle: The hardware itself worked before in that way using nouveau and bumblebee. Now I read that it would be much easier to use rev PRIME ( https://wiki.archlinux.org/index.php/PRIME#Reverse_PRIME )
05:14 duelle: khazakar: Yeah I did. But I read that the usage of bumblebee would be somewhat deprecated and PRIME would be the new way to something like this.
05:16 duelle: And at least in my case it was quite annoying with bumblebee - as it restarted x multiple times when scanning for screens and enabling them. So PRIME seemed quite appealing to me.
05:16 khazakar: PRIME as the new way?When? If something works with bumblebee but not with reversePRIME I think is better to use bumblebee and wait to enhance ReversePRIME. I think is better to have working something with some bug than don't work at all :P
05:18 duelle: khazakar: The deprecation is stated on the Bumblebee ArchWiki page regarding (Bumblebee with nouveau): "Note: This method is deprecated and will most likely not work anymore. Use the nvidia module instead. If you want nouveau, use PRIME."
05:19 duelle: You're right regarding the working vs. not working thing. But I thought there might be something to tweak or some setting to fix this issue. I am not that deep into the X/xrandr/graphics stuff yet.
05:19 duelle: That's why I got here ;)
05:21 khazakar: Like me - that's how I got there with engine reclock problem on my GTX760 :P
05:23 duelle: khazakar: The interesting part in my case is, that it has to have something to do with the nvidia card or the settings. If I have only one screen attached to the DVI port and the other external screen to the VGA port (attached to the Intel card) it works like a charm..
05:24 duelle: That's what made me hope that it might work eventually with both DVI ports too.
05:24 khazakar: So..Can you run something,using intel or nvidia card without bumblebee? Mayby you're running only on Intel card,which don't have triple-monitor capability,hm? :>
05:27 imirkin: duelle: i suspect your issue is with the intel ddx
05:27 imirkin: duelle: try disabling LVDS1 and see if both of the DP-N monitors come up
05:28 imirkin: duelle: if that works, then that means that the intel ddx is trying to allocate crtc's for the "remote" outputs even though it doesn't need to
05:29 imirkin: as for bumblebee with nouveau, i don't recommend it. bumblebee does things behind the driver's back which will most likely mess up nouveau unless you're careful.
05:29 duelle: khazakar: Good point - though I think I read that three screens are possible even with the Intel card.
05:29 imirkin: qwertz_: you have an xorg config issue. pastebin the output of 'grep -r . /etc/X11/xorg.conf*'
05:30 duelle: imirkin: Hello again - thanks for the hint. I will give it a try.
05:30 imirkin: duelle: intel only supports 3 CRTC's under very specific conditions... i suspect you don't meet them.
05:31 jarnos_: imirkin, I still wonder why decoding mpeg2 properly does not work with vdpau using my hardware. I mean isn't proper software decoder unavailable for linux?
05:32 imirkin: jarnos_: vdpau = hardware decoding
05:32 imirkin: as provided by mesa, in your case
05:32 duelle: imirkin: Right now it gives the same error as before ("configure crtc 5 failed.") and keeping the second external screen black and the notebook screen on. I will restart X and try it from scratch. brb
05:33 imirkin: jarnos_: your hw can't take an mpeg2 stream, so mesa has to do VLD decoding on it
05:33 khazakar: imirkin,can you help me with unexpectly don't working reclocking my GFX card?Yesterday it worked like a charm
05:33 imirkin: jarnos_: i suspect about 100 man hours went into the mesa mpeg2 VLD decoding logic, compared to maybe 10000 going into whatever's in most video players
05:33 imirkin: jarnos_: we can't reuse an existing library due to licensing issues (specifically libmpeg2)
05:34 imirkin: jarnos_: and libavcodec is a giant mess
05:34 imirkin: khazakar: figure out what you changed since yesterday? :)
05:35 jarnos_: imirkin, :(
05:35 khazakar: Only changed from manual to auto control PWM and change minimum from 20 to 65%. Only that.
05:36 imirkin: khazakar: perhaps you're loading a diff nouveau module than you were before?
05:37 khazakar: I don't think so,because I don't changed kernel. Still 4.4 rc6 from experimental debian repo.
05:37 imirkin: khazakar: is dmesg showing an indication of what's going wrong
05:38 imirkin: duelle_: rebooting is almost never necessary
05:38 khazakar: Only spam with evbug in dmesg... :P
05:38 duelle_: imirkin: I did not reboot - just restarted lightdm/X. The issue persists (with the same error message) even if I disable the main screen (notebook internal)
05:39 duelle_: I first enabled one of the external screens and disabled the "main" screen. Afterwards I tried to add the second external screen. Same error message.
05:39 imirkin: duelle_: let's try this:
05:40 imirkin: xrandr --output DP-1-1 --right-of LVDS1 --auto
05:40 imirkin: does one screen light up?
05:40 duelle_: Just a second - have to enable rev PRIME again
05:41 duelle_: imirkin: $ xrandr --output DP-1-2 --right-of LVDS1 --auto ## Yes it does.
05:41 imirkin: great. now do
05:41 imirkin: xrandr --output LVDS1 --off
05:41 imirkin: does that leave just the DP-1-2 screen on?
05:42 duelle_: Check
05:42 imirkin: now do
05:42 imirkin: xrandr --output DP-1-1 --right-of DP-1-2 --auto
05:43 khazakar: What's more - only memory reclocks,but no engine. yesterday engine reclock worked like a charm.NVidia,why? :P
05:43 duelle_: imirkin: $ xrandr --output DP-1-3 --right-of DP-1-2 --auto # => xrandr: Configure crtc 5 failed
05:43 imirkin: khazakar: have you tried rebooting? (or at least reloading nouveau)?
05:43 duelle_: The external screen that was enabled is still active. The other two are black.
05:43 khazakar: Sure,reboot don't change antything
05:43 imirkin: duelle_: hmmmm very interesting.
05:44 imirkin: i wonder if there's a way to find out wtf crtc 5 is
05:44 imirkin: i still want to blame intel
05:44 imirkin: since, you know, blame the other guy ;)
05:44 khazakar: yup,google knows only crct 0 :P
05:44 khazakar: s/crct 0/crtc 5
05:46 duelle_: imirkin: When I first ran into this issue I tried to do this: https://wiki.archlinux.org/index.php/PRIME#Discrete_Card_as_Primary_GPU
05:46 duelle_: But then the last command failed that intel is not defined or could not be found..
05:47 duelle_: So after I ran into this issue here, I tried to do the DiscreteCard as Primary stuff. But that didn't work either. Because the Intel card is not listed at all.
05:47 khazakar: Mine check the pstate (BTW. yesterday core reclocked only to 966 MHz,not max 1215 MHz w/boost) => http://pastebin.com/9Xxu7Fxn
05:48 imirkin: khazakar: upstream doesn't do that. you must have been using a different module from karol's tree.
05:48 khazakar: I use gallium-nine and 4.4 rc6 did that.
05:49 imirkin: khazakar: gallium-nine is entirely unrelated to this dicussion.
05:49 imirkin: discussion*
05:49 khazakar: Vanillia 4.4rc6 from experimental Debian repository
05:49 khazakar: DRM-next from ubuntu mainline kernel ppa didn't help
05:50 imirkin: khazakar: ok, well the only tree that has the "respect base clocks" logic or any sort of "don't use the absolute max frequency" logic is karol's tree
05:52 imirkin: duelle_: can you install x11trace
05:52 khazakar: Where I can find his PPA/repository?
05:52 imirkin: duelle_: and run 'x11trace xrandr'
05:52 karolherbst: qwertz_: sorry that I can't help you though :/
05:52 imirkin: khazakar: https://github.com/karolherbst/nouveau
05:52 karolherbst: imirkin: know any reasons why the X server doesn't load the nouveau ddx allthough the nouveau module is loaded?
05:53 karolherbst: imirkin: <qwertz_> dmesg: http://pastebin.com/ppAu0yb5; xorg.0.log: http://pastebin.com/xsM9jDkd
05:53 imirkin: karolherbst: plenty. that's why i asked him to show me all the xorg config stuff.
05:53 karolherbst: ahhhh right
05:54 karolherbst: khazakar: did you upload your vbios somewhere?
05:55 khazakar: No,but I can if needed. Where I can locate vbios to upload?
05:55 duelle_: imirkin: installing
05:55 imirkin: duelle_: actually run x11trace <the xrandr command that fails>
05:55 imirkin: and then pastebin the full output of that
05:55 karolherbst: khazakar: /sys/kernel/debug/dri/0/vbios.rom
05:55 khazakar: Thanks,now I'm copying it :P
05:56 imirkin: qwertz_: [ 16.491] (**) intel(0): Option "AccelMethod" "sna"
05:56 imirkin: qwertz_: most likely that's what's killing it. drop the intel device section and all will be well.
05:56 imirkin: sna is the default anyways
05:57 khazakar: Ok,uploaded here => https://www.sendspace.com/file/7q7las
05:57 duelle_: imirkin: This is the output of "$ x11trace xrandr --output DP-1-3 --right-of DP-1-2 --auto | tee -a /tmp/x11trace.log": http://pastebin.com/zq28PVJh
05:57 imirkin: duelle_: great, give me a few to process :)
05:59 karolherbst: khazakar: k, I know that issue :D
05:59 khazakar: yesterday It worked like a charm on vanillia 4.4 rc6 kernel,today - not. :D
06:00 karolherbst: khazakar: I bet if you clock to 0a first you get a higher core clock with 0e 0f already
06:00 karolherbst: that's why it kind of worked yesterday
06:00 karolherbst: but your card should be able to reach even higher clocks
06:01 khazakar: I know,but idk why. Even to 1215 MHz on engine. It's GTX 760 Zotac version with referntial cooling
06:02 khazakar: Sure,it worked with 0a,then 0f.Now I'm looking for change to 0a,then 0f during boot process :P
06:02 karolherbst: well I know why it doesn't work
06:02 karolherbst: dmesg will show you an voltage error
06:03 imirkin: duelle_: something is going horribly wrong *somewhere*.... unfortunately i'm mostly unfamiliar with all this xrandr junk =/
06:03 karolherbst: khazakar: stable_reclocking_kepler branch of my repository has the fixed to somehow fix that
06:03 imirkin: duelle_: perhaps airlied will know... but he's asleep right now. stick around.
06:04 khazakar: My dmesg goes crazy and I see only evbug with input type xD
06:04 duelle_: imirkin: Just one thing I just saw: In the beginning of the output there was this, which was not forwarded to the log file: "No display name to create specified, trying :9 Got connection from unknown(local)"
06:04 imirkin: duelle_: that's fine
06:05 karolherbst: khazakar: yeah anyway, if you want proper reclocking you need my branch and maybe a voltage hack later
06:05 imirkin: duelle_: basically the way x11trace works is it creates a new X screen and tells the application to talk to that... basically man-in-the-middles the X protocol
06:05 khazakar: Sure,I want a better reclock,because I want to reach my "OpenSource Gaming Rig" with my PC :P
06:05 khazakar: To play games using nouveau,gallium-nine etc
06:07 duelle_: Ok, just wanted to tell you as it was not added to the log output. Thanks a lot for your help! I will reset my setup now and will stay here for some time.
06:07 duelle_: brb
06:08 khazakar: BTW - It's possible to render 3D etc using my Quadro NVS 160M in my notebook? :P
06:08 imirkin: duelle_: i also noticed you're on a slightly old X version... there's a 1.17.4, and there's also a 1.18
06:08 imirkin: khazakar: should be fine. although i think that's one of the ones some people have had problems with.
06:09 duelle__: re
06:09 imirkin: duelle__: i also noticed you're on a slightly old X version... there's a 1.17.4, and there's also a 1.18
06:09 khazakar: BTW - the nouveau can be related to my Ambient Light Sensor problem in my Dell Latitude E6500?
06:10 khazakar: because I don't know where I can find a help with this...
06:10 imirkin: sorry dunno
06:11 duelle__: imirkin: That's odd.. I (at least thought) that arch uses quite recent versions.. I will check.
06:11 imirkin: duelle__: maybe i'm mixing up your logs with someone else's
06:11 duelle__: imirkin: According to my package manager it is: xorg-server 1.18.0-3
06:12 imirkin: duelle__: ah ok. then i must have gotten confused.
06:13 peb`: I met some troubles yesterday with nouveau and linux 4.3.0-1-amd64 under debian
06:13 khazakar: And...what problems with rendering on nouveau?
06:14 peb`: here is a (2 lines) excerpt
06:14 peb`: Jan 5 15:28:24 dawaj kernel: [ 38.053652] BUG: unable to handle kernel NULL pointer dereference at (null)
06:14 peb`: Jan 5 15:28:24 dawaj kernel: [ 38.053677] IP: [<ffffffffa0c100a5>] nvkm_pmu_pgob+0x5/0x20 [nouveau]
06:14 peb`: I have a full trace in my syslog
06:15 peb`: practically, symptoms are that I can't load my WM
06:15 imirkin: peb`: pastebin full dmesg
06:15 peb`: GDM3 succeeds authentication then all freezes
06:15 imirkin: peb`: this sounds oddly familiar actually
06:15 imirkin: peb`: but i can't quite remember what the deal was... we fixed it though...
06:16 peb`: I don't have the exact dmesg as my laptop mostly froze
06:16 imirkin: sec
06:16 peb`: but systemd caught some logs I can give you :)
06:16 peb`: it only happens under my 4.3 kernel
06:16 peb`: under my 4.2 everything works
06:16 peb`: (that seems weird)
06:16 imirkin: peb`: this patch should fix it: 1c4b0d12f487a874928ad6efd73a8939c6d23be4
06:16 imirkin: peb`: i assume you have a GF117?
06:17 peb`: good guess
06:17 peb`: 03:00.0 3D controller: NVIDIA Corporation GF117M
06:17 imirkin: peb`: er, wrong tree... that hash won't help
06:17 imirkin: peb`: commit 579b7c58215329803ce184704463de09f0f310ac upstream
06:17 peb`: imirkin: your assertion is helpful as I can expect this bug to be fixed soon in debian sid
06:17 peb`: will have a look
06:17 peb`: (I'm curious)
06:18 imirkin: the _real_ issue is that for some idiotic reason the pmu isn't hooked up on gf117
06:18 imirkin: it doesn't really do anything useful at this point so it doesn't really matter
06:18 imirkin: i'm going to request a stable backport for that commit
06:19 peb`: ew
06:19 peb`: okay, I understand
06:19 peb`: so this has to be patched in the kernel
06:19 peb`: (doesn't it?)
06:20 khazakar: karolherbst It is possible to include your patches to 4.4 rc6 kernel in debian experimental,it will work?
06:20 karolherbst: khazakar: well you have to compile a new nouveau package or apply those patches to the debian kernel package
06:22 imirkin: peb`: indeed.
06:22 khazakar: Now I'm downloading 4.4 rc8 from experimental debian repository and I will add this to main kernel repository
06:22 khazakar: kernel package *
06:23 imirkin: khazakar: won't have the stuff in karol's tree
06:23 peb`: imirkin: that NULL pointer who refuses to be dereferenced, he's not really nice. :P
06:23 peb`: thanks for your answer!
06:24 imirkin: peb`: you can work around this by adding "nouveau.modeset=0" to your kernel parameters
06:24 peb`: it avoids me to annoy everybody doing another bug report :p
06:24 peb`: okay
06:24 khazakar: imirkin but I will apply a nouveau patch to mainline source ;)
06:24 imirkin: peb`: this will disable nouveau entirely
06:25 peb`: I'll think about what fits best my requirements
06:25 peb`: I still can use 4.2 for now
06:25 peb`: it works correctly
06:25 peb`: anyway, thanks a lot!
06:25 imirkin: np
06:28 karolherbst: khazakar: ? what do you mean by adding to main kernel repository?
06:30 khazakar: Main kernel package*,my mistake
06:30 karolherbst: ahh k
06:32 khazakar: Now download 4.4 rc8,then add your patches,compile,make package and voila :D
06:34 karolherbst: khazakar: https://github.com/karolherbst/nouveau/compare/master_4.3...stable_reclocking_kepler.patch
06:34 khazakar: It will work for 4.4rc8?
06:35 karolherbst: yeah, but you have to apply it inside drivers/gpu/drm/nouveau
06:36 khazakar: No problem.I had a more difficult things like compile kernel for my Xperia E4g with Mediatek SoC... :D
06:38 khazakar: Difficult,because the source was mixed,screwed up etc to be unusable and unable to compile ;)
06:38 khazakar: brb
07:01 jarnos_: For some reason displaying EPG in mythtv does not work, when it should start as I start watching live-TV. It works, if I start EGP manually separately, though. It happens, if I use an OpenGL playback profile in settings.
07:01 jarnos_: ^EPG
07:02 imirkin: CPG = consumer products & goods... what's EPG?
07:05 karolherbst: yay, pmu messaging stress testing :/
07:05 karolherbst: 33k requests and 4 IRQs lost :/
07:06 jarnos_: imirkin, Electronic program guide or something.
07:07 imirkin: karolherbst: but... it recovered?
07:07 karolherbst: I have a workaround patch
07:07 karolherbst: https://github.com/karolherbst/nouveau/commit/f159e2910b44c52b89b6512d1213be5d02bdafd9
07:07 karolherbst: the reply is there, nouveau just doesn*t get the IRQ to read it
07:11 karolherbst: imirkin: and no, nouveau can't recover from that
07:11 karolherbst: wait_event can't be interrupted or terminated
07:11 imirkin: karolherbst: wait_event_interruptible? :)
07:11 karolherbst: wait_event_timeout
07:12 karolherbst: is I think the best solution here
07:12 jarnos_: It displays TV-playback small frame, but does not display EPG, or displays it only partly after some time. And it disappears, if I switch to another application and bach to mythtv.
07:12 karolherbst: if the pmu needs more than 1 second to reply then something is fishy already
07:15 karolherbst: imirkin: 0.008% is the chance of lossing an IRQ :/ but I want to generate 1M requests and see if with my patch it don't mess up real hard, so I have to wait now :D
07:20 karolherbst: we still have no idea why that happens though, only theories like IRQ coalescing, but this happens with sequential requests, where there is only one wait for one reply :/
07:22 karolherbst: imirkin: https://lwn.net/Articles/392136/ maybe this helps :D
07:24 illlmnec: Hello, I am having display corruptions and X freezes with my Quadro FX 380 card in a dual monitor configuration and I would like some help to debug these issues. I am under Ubuntu 15.10, and the problems appeared since I upgrade to 14.04-14.10.
07:25 imirkin: illlmnec: anything in dmesg?
07:27 illlmnec: Yes, I do have info in dmesg: nouveau E[ PFB][0000:02:00.0] trapped read at 0x002043b000 on channel 0x0000fb2a [Xorg[4201]] PGRAPH/TEXTURE/00 reason: PAGE_NOT_PRESENT
07:28 imirkin: gr.
07:28 qwertz_: imirkin, thanks for the hint earlier
07:29 imirkin: illlmnec: i hate that error. unfortunately i also haven't the faintest clue how it can happen :(
07:29 imirkin: and yet reality appears to differ
07:29 qwertz_: imirkin, does "drop the intel device section" refer to xorg.conf?
07:29 imirkin: qwertz_: yes.
07:30 qwertz_: imirkin, my x server runs without a config file!
07:30 illlmnec: Is there any other logs I could enable/send to help debug this issue?
07:30 imirkin: skeggsb: how would you feel about adding a thing which dumps the current context's execution... stuff... when getting a trapped read/write notification?
07:30 imirkin: illlmnec: sadly no.
07:30 imirkin: qwertz_: that's a lie.
07:31 imirkin: qwertz_: ** means "value comes from config file"
07:31 imirkin: qwertz_: do the grep i told you to do.
07:32 qwertz_: imirkin, i'll check the scrollback to find the grep
07:32 qwertz_: imirkin, i'm really puzzled. there's no xorg.conf in /etc/X11
07:32 imirkin: qwertz_: grep -r . /etc/X11/xorg.conf*
07:35 qwertz_: imirkin, grep: /etc/X11/xorg.conf*: No such file or directory
07:35 karolherbst: qwertz_: check /usr/share/X11/xorg.conf*
07:35 karolherbst: maybe the package did something insane
07:35 karolherbst: *packager
07:35 imirkin: qwertz_: locate xorg.conf.d
07:36 imirkin: qwertz_: /usr/share/X11/xorg.conf.d
07:36 qwertz_: imirkin, /usr/share/X11/xorg.conf.d
07:36 imirkin: qwertz_: (from your xorg log)
07:36 imirkin: repeat the grep there
07:37 qwertz_: 20-intel.conf!
07:37 qwertz_: this looks like a xorg.conf
07:37 imirkin: all those files are appended together, effectively
07:38 qwertz_: imirkin, ah, i see!
07:38 qwertz_: okay, so why does X look in /usr/share and not only in /etc/X11
07:38 qwertz_: ?
07:38 karolherbst: because /usr/share is what your distribution is shipping
07:38 imirkin: qwertz_: your distro configured it to do so?
07:38 qwertz_: imirkin, i see
07:39 qwertz_: okay, i'll try to get rid of those files and restart X
07:39 imirkin: perhaps it does so by default if there's nothing in /etc/X11 -- dunno
07:39 qwertz_: imirkin, thanks a lot!
07:41 qwertz_: imirkin, it works! thank you so much! now i have three providers for xrandr!
07:41 imirkin: cool. that's a harmless bug btw
07:41 imirkin: i.e. that one provider is listed twice
07:41 qwertz_: imirkin, yep, they have the same ID
07:41 imirkin: i forget where the bug is, but it doesn't matter in practice.
07:43 qwertz_: so now the fun part can start, i.e. figuring out which switcheroo scripts i'll need in practice
07:43 imirkin: qwertz_: well, the gpu should auto power on/off as it is used/not-used
07:43 imirkin: no need to mess with switcheroo
07:44 illlmnec: imirkin: I have some other messages, such as this: [306640.051586] nouveau E[ PGRAPH][0000:02:00.0] DATA_ERROR INVALID_VALUE [306640.051595] nouveau E[ PGRAPH][0000:02:00.0] ch 2 [0x000fb2a000 Xorg[4201]] subc 2 class 0x502d mthd 0x08dc data 0x00020000
07:44 imirkin: illlmnec: can you pastebin your full dmesg?
07:44 imirkin: perhaps i'll see something that'll spark a theory
07:44 khazakar: I'm on. :P
07:44 karolherbst: khazakar: dmesg | grep nouveau | grep base
07:46 khazakar: I was taking a bath atm,now I will patch,compile and make a package ;)
07:46 karolherbst: :D
07:46 qwertz_: oh my god! i just put my laptop in the docking station and XFCE auto-configured the NVIDIA-card for me!
07:47 khazakar: Where I need to put your fixes? // What laptop qwertz_? :>
07:47 qwertz_: khazakar, Thinkpad T430
07:47 karolherbst: khazakar: drivers/gpu/drm/nouveau
07:47 karolherbst: yeah window manager can be smart sometimes
07:47 illlmnec: imirkin: http://pastebin.com/eFEvP4Sj
07:48 qwertz_: hm, the screen looks a bit more blurry than usual. i think the intel card still renders the image
07:48 imirkin: illlmnec: ohhhh... the plasmashell thing
07:48 imirkin: illlmnec: update your kernel.
07:48 qwertz_: yep, glxinfo says it's still all intel
07:48 imirkin: illlmnec: something should have been backported for stable...
07:49 imirkin: qwertz_: indeed, that's how reverse-prime works
07:49 karolherbst: qwertz_: yeah, that's normal
07:49 illlmnec: imirkin: ok. will update to latest ubuntu kernel (4.2.0-23). I will also try to get the dmesg for the freeze, when it happens.
07:49 imirkin: qwertz_: you can also do rendering offloading, but it's just as likely that the nvidia gpu won't be any faster with nouveau -- GF108 is crap in the first place, and we can't reclock fermi's
07:49 imirkin: illlmnec: hopefully that one has my patch
07:50 qwertz_: anyway, it's all great! i'll guess i'll never have to reboot into BIOS to enable/disable any card ever again
07:50 khazakar: I need to redownload it,because I cannot replace some folders,yh
07:50 illlmnec: imirkin: I am running a vanilla 4.2.1 right now because I was trying to find a kernel that would work :)
07:51 imirkin: illlmnec: https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/?id=03cf3c9337feb72ef10b85a2fdba27b358bffa5b
07:51 imirkin: illlmnec: should be in 4.2.6
07:52 imirkin: the bug talks about suspend/resume, but that's not necessarily the only way to hit the issue
07:52 imirkin: it's just a really good one :)
07:54 khazakar: And again download ~30 min,because experimental repo have low bandwidth
07:55 qwertz_: karolherbst, imirkin, thank you very much again! you made my day, guys!
07:55 karolherbst: so it works? :D
07:55 karolherbst: ohh right, it does :D
07:55 imirkin: qwertz_: see http://nouveau.freedesktop.org/wiki/Optimus/ for instructions on how to offload 3d to nouveau
07:56 illlmnec: imirkin: ok, I found that I can get 4.2.8 for Ubuntu 15.10, I'll use that instead.
07:56 imirkin: qwertz_: the only reason right now to do so is if you want GL 4.1 -- intel is missing fp64 for the moment (and only gained tess in git, not in any released version)
07:56 qwertz_: imirkin, already tried that with glxgears :) > 2400 fps
07:56 karolherbst: yeah well
07:56 karolherbst: that's kind of nothing :D
07:57 qwertz_: karolherbst, i'm really not doing this for performance, but rather for digital output without rebooting &BIOS action :)
07:57 karolherbst: yeah I know
07:57 karolherbst: ohh wait, 4.5 might be getting great for you then :)
07:57 qwertz_: well, i'll curiously check if all that survives a reboot
07:57 karolherbst: I hope my pcie patches make it in time for 4.5
07:58 qwertz_: karolherbst, what to expect?
07:58 karolherbst: more bandwith for prime setups
07:58 imirkin: qwertz_: if you really want to kill it, suspend, then undock, then resume :)
07:58 karolherbst: which basically means more performance overall
07:58 qwertz_: imirkin, that's what i'll do!
07:58 karolherbst: anything qt5 related will crash anyway
07:58 qwertz_: karolherbst, eager to see those patches in!
07:58 karolherbst: :D
07:58 illlmnec: imirkin: Thank you for the help.
07:58 imirkin: preferably timed with the actual undocking... i.e. undock *while* it's resuming :)
07:59 imirkin: illlmnec: don't thank me 'till it works :) iirc some people still had more issues after that patch too
07:59 karolherbst: qwertz_: yeah well I don't know if nouveau will enable reclocking for fermi just for the pcie stuff though
08:00 khazakar: karolherbst,do you know any tweaks for quadro mobile cards like NVS 160M with nouveau? BTW even without your patches,with limited reclocking to 966 on core I have ~11k fps on small,and ~2,1k fps on fullscreen on my GTX760.
08:00 qwertz_: imirkin, well, that was at most partially successful. it didn't crash, but it also didn't show me a screen
08:00 karolherbst: khazakar: you should get more with the patches
08:01 karolherbst: something above 1050MHz
08:01 karolherbst: khazakar: when you are lucky you may also get close to 1200MHz without crashing
08:02 karolherbst: I doubt that though
08:02 khazakar: I can have max 1215 MHz :P. So more viable for my mini project "Open Source Gaming Rig"? :P
08:02 karolherbst: :D
08:02 karolherbst: yeah well
08:02 karolherbst: does the binary driver _ever_ clock to 1215MHz?
08:03 khazakar: I think yes,I don't used it some time,but I think - yes.
08:03 karolherbst: mhhh
08:03 karolherbst: I would expect that it can hit the 1202 MHz clock
08:03 karolherbst: but 1215 shouldn't be possible
08:03 khazakar: And something for my Quadro NVS160M are viable? I want to make "OpenSource Engineering Notebook" from my Dell... :D
08:04 karolherbst: imirkin: after 1M PMU reqeusts, I have 22 lost IRQs, doesn't sound too bad actually
08:04 imirkin: khazakar: what are the actual issues you have with it?
08:05 khazakar: Just to get max what is possible from that Quadro using nouveau. Driconf dropped HyperZ,so I don't know more options to have faster e.x rendering or making 3D models/making 3D games(in the future)
08:06 karolherbst: imirkin: it is much worse while reclocking
08:06 karolherbst: 35k requests and 32 lost IRQs :/
08:09 imirkin: orbea: dmesg would be more relevant -- that xorg log just says "gpu hung. gpu hung! GPU HUNG!"
08:10 imirkin: duelle: btw, dunno if it's easy, but you might consider updating to xorg 1.17.4 or maybe even xorg 1.18 -- problems like this get fixed :)
08:11 khazakar: imirkin He showed that he have 1.18 on his ArchLinux :P
08:11 imirkin: gr, was i looking at the wrong log *again*
08:16 khazakar: It's normal :P
08:24 karolherbst: yay
08:24 karolherbst: the pmu fifo is full :D
08:25 khazakar: karolherbst what is the fsrm_kepler on your github? :P
08:25 duelle: imirkin: Didn't we already find out that I am on 1.18.0-3 already? ;)
08:25 karolherbst: some broing stuff :p
08:25 orbea: imirkin: I think syslog is what shows it in slackware, dmesg doesn't show anything from before my last boot I think... http://dpaste.com/1A8H73S
08:26 orbea:hopes that is helpful
08:27 imirkin: orbea: ok yeah. ugh.
08:27 imirkin: orbea: yes and no. i can't really do anything useful with that info =/
08:27 imirkin: duelle: yeah. sorry :)
08:29 orbea: I can try to get more useful info later
08:29 imirkin: orbea: not your fault
08:29 imirkin: orbea: we just suck at printing useful info
08:29 orbea: ah
08:49 masta: How goes with Nouvaeu on Tegra124 ?
08:50 imirkin: masta: should work. but my Jetson TK1 board is supremely unstable, unrelatedly to nouveau
08:50 imirkin: couldn't get it to stay up for more than a few minutes
08:51 masta: imirkin: oh that sucks! on my Acer chromebook, it's using mode setting... so I was just wondering if anybody was working on accelerated
08:52 masta: (the chromebook is using mainline and fedora userland)
08:52 imirkin: modesetting is done through tegra
08:52 imirkin: nouveau is driving a separate gpu device
08:53 karlmag:lols at himself.. I managed to read that as "gnu" and really wondered what you where talking about :-P
08:53 imirkin: karlmag: yes... why would people be talking about gnus in #nouveau...
08:54 masta: the elinux wiki claims u-boot needs to be just right, so maybe I'll chainload a more recent u-boot or see about flahsing the onboard
08:54 imirkin: i'm actually suspecting it has to do with nfsroot
08:54 imirkin: my ppc g5 had similar stability issues
08:55 imirkin: although my qualcomm boards seem to do just fine
08:55 imirkin: the hang always happens on nfs activity, like when i tab-complete or something
08:55 masta: imirkin: aww, that sucks
08:55 imirkin: but i don't have serial/etc
08:55 imirkin: and nothing shows up over netconsole
08:56 masta: always wanted to try netconsole
08:56 khazakar: Ok,compiling with changed some options like removing SELinux support... :D
08:57 karlmag: imirkin: people might have their reasons? *grin*
08:58 masta: imirkin: howcome no serial ?
08:58 imirkin: masta: no null-modem cable
08:59 masta: bummer, you should for sure get one of those eventually imirkin
09:01 imirkin: eventually.
09:01 karlmag: "serial cable" can be many things these days..
09:02 karlmag: it may include USB for example..
09:02 imirkin: karlmag: DB9
09:02 karlmag: If you have machines with that it does make it easier for sure
09:03 imirkin: karlmag: i have machines with it. just no cable.
09:04 karlmag: I could probably have hacked up one for you, but I seem to remember you live on the other side of the globe..
09:05 imirkin: i could also purchase one. it's just not high up on my priority list.
09:05 karlmag: there is (always) that..
09:06 karolherbst: this damedn pmu :/
09:06 RSpliet: ^ that damedn keyboard
09:07 khazakar: Always you can buy cheap USB=>RS232 :P
09:07 karolherbst: yeah well after around 20k pstate reclocks the pmu is totally messed up
09:07 imirkin: khazakar: i have one.
09:07 RSpliet: karolherbst: messed up in what way?
09:07 karlmag: khazakar: they actually sometimes doesn't work.
09:07 karolherbst: RSpliet: well the message fifo is full :D
09:07 imirkin: karlmag: nah, they work as long as you're using it for serial
09:07 karolherbst: and I can't send any new requests anymore
09:07 karolherbst: but I also don't get any replies
09:08 karolherbst: totally weird
09:08 karolherbst: there is some data in the fifo though
09:08 karlmag: imirkin: usually, but I've had (and had others tell me) that they don't work in some circumstanses.
09:08 RSpliet: karolherbst: the PMU software is available all-assembly
09:08 karolherbst: but POST equals GET
09:08 RSpliet: did you finish your fuc emulator?
09:08 karolherbst: :D no
09:08 karolherbst: did I say I want to write one? :/
09:08 RSpliet: :D no
09:08 karolherbst: k :D
09:09 karolherbst: anyway
09:09 karolherbst: as long as I only make a simple requests
09:09 karolherbst: like reading out the pdaemon counters through the pmu and send it to the host
09:09 karolherbst: I can easily do 1M requests
09:09 karolherbst: where I like loose 20 IRQs I can recover though
09:09 karolherbst: and I also always get the reply from the pmu
09:09 khazakar: That's why I bought docking station for my notebook to perform operations,which uses RS232 :P
09:09 RSpliet: can you observe similar behaviour on GT21x?
09:09 karolherbst: but not so much if I change pstates
09:09 masta: ok, since I'm new to nouveau I'll go find your online documents, read... and then try to figure out how I can help the tegra124 use case.
09:10 karolherbst: RSpliet: didn't test it yet
09:10 imirkin: masta: it should Just Work (tm) for you
09:10 imirkin: masta: once you get the firmware
09:10 RSpliet: could be worth it, the assembly we upload differs quite a bit
09:10 imirkin: masta: i was going to try to see if the nouveau firmware will just work for it too, but given my uptime situation... i didn't have the patience for it.
09:10 karolherbst: RSpliet: that's my code to repair the lost IRQ: https://github.com/karolherbst/nouveau/commit/f159e2910b44c52b89b6512d1213be5d02bdafd9
09:10 masta: imirkin: ok, I'll give it a whirl
09:11 RSpliet: karolherbst: do you lose IRQs if you boot with a single CPU?
09:11 karolherbst: RSpliet: actually with this code is is _much_ better, because without it, 1000 requests can mess up the PMU already
09:11 karolherbst: RSpliet: ohhh I could try disabling the other cores
09:11 karolherbst: or would I have to really boot with only one core?
09:11 RSpliet: booting with one core only seems easier, there's a kernel param for that
09:12 karolherbst: mhh
09:12 karolherbst: I can also just disbale cores at runtime
09:12 karolherbst: which is faster
09:12 RSpliet: you mean after SMP might have already ruined the FIFO? :-P
09:12 karolherbst: done :D
09:12 karolherbst: well
09:12 karolherbst: it is a laptop
09:12 karolherbst: I can just power off the gpu
09:12 karolherbst: :D
09:14 karolherbst: RSpliet: yes, but not so often
09:14 karolherbst: and I always recover the reply
09:14 RSpliet: anyway, if you lose IRQs it might also just be a good idea to figure out why, I mean from the OS pov. Could they coalesce (eg. a second IRQ fired before the first one was handled), does nouveau prematurely clear the bit? forget to re-enable the IRQ in time?
09:15 RSpliet: or does PMU actually not send an IRQ from time to time?
09:15 karolherbst: no
09:15 karolherbst: I have only one request at any time
09:15 karolherbst: nevermore
09:15 RSpliet: and one request always results in exactly one reply
09:16 karolherbst: mhh does the memory reclocking stuff expects any result at all?
09:16 karolherbst: I am not quite sure about that now :/
09:16 RSpliet: yes, it returns the time of execution
09:16 karolherbst: ahh oaky
09:16 karolherbst: then I only do this
09:17 RSpliet: but: no strange combination of preemption could occur that triggers two separate requests (one reclocking while one reading out sensors)?
09:17 RSpliet: are you absolutely 100% sure? ;-)
09:17 karolherbst: that's the code I am running: i=0; while true; do echo $((i=$i+2)) && echo 0f > pstate && echo $((i=$i+2)) && echo 07 > pstate; done
09:17 karolherbst: and the pstate echo blocks until it's done
09:18 RSpliet: fb/ramxxxx is not the only PMU consumer, is it?
09:19 karolherbst: yeah well... I think in a laptop there won't be anything else besides that
09:19 karolherbst: I mean the gpu isn't doing anything usually
09:19 karolherbst: anyway
09:19 karolherbst: the pmu send code is synchronous
09:20 karolherbst: ohh wait,
09:20 karolherbst: when the requests wants a reply it has to wait until other requests are done
09:22 RSpliet: there's MEMX, KERN and I2C_ messages
09:22 RSpliet: oh and PERF
09:22 karolherbst: yeah
09:22 karolherbst: but I only do reclocking
09:22 karolherbst: currently
09:22 RSpliet: are you sure nouveau doesn't do anything else? ;-)
09:23 karolherbst: yes
09:23 karolherbst: what should it do otherwise=
09:24 RSpliet: idk what's on the I2C lines, but... something periodically?
09:24 karolherbst: mhh
09:24 karolherbst: currently not
09:25 RSpliet: anyway, lots of factors to consider, lots of fun to be had w/ debugging ;-)
09:25 RSpliet: bbl
09:25 karolherbst: mhhh
09:25 karolherbst: by the way
09:25 karolherbst: if I reclock and read the pdaemon counters out alternately
09:25 karolherbst: I really loose the replies :O
09:26 karolherbst: otherwise not
09:26 karolherbst: if I only do reclocking I can recover those
09:26 karolherbst: or if I only read the counters
09:26 karolherbst: mhh interessting
09:30 khazakar: Eh,making kernel package using checkinstall aren't so easy,eh. :D
09:31 towo`: make deb-pkg
09:32 towo`: that's the way to go, checkinstall will never work for a proper kernel-package
09:39 khazakar: K,thanks. :P
10:37 masta: anybody here running PS3 with Linux (old firmware) for RSX acceleration?
12:04 imirkin: skeggsb: is it just me or is bufctx totally broken? it doesn't seem to free any of its internal structures... ever.
12:09 phomes: pstate on one of my G92 cards actually works
12:09 imirkin: skeggsb: ping me when you're around. i'd like to discuss bufctx with you. it totally makes no sense to me.
12:11 phomes: or, at least core and shader changes. Memory stays at 399 while the two options on this card are 100 and 800
12:11 imirkin: phomes: memory is kinda important
12:11 imirkin: phomes: i guess the reason that card works is that the memory reclock totally fails
12:12 imirkin: phomes: you can disable memory reclocking on the other cards too by doing nouveau.config=NvMemExec=0
12:13 phomes: I think you are right. The journal says: nouveau 0000:0f:00.0: fb: invalid/missing rammap entry
12:15 imirkin: but engine reclocking isn't that useful without memory...
12:15 imirkin: memory bandwidth is the usual bottleneck
12:16 waressearcher2: is anypony here ?
12:17 towo`: no
12:17 towo`: only 205 bots
12:17 towo`: what a useless start in a channel
12:27 phomes: imirkin: the max setting (800 MHz) for memory matches the specs for the card. So that part is correct at least. Is there anything more I can check or try out with the card?
12:27 imirkin: phomes: catch RSpliet when he's around... he's the tesla reclocking master
12:34 thereh: is it possible to overclock GPU ? should I tweak nouveau source or there are options I can set when loading the driver or can I do it on the fly ?
12:37 imirkin: thereh: what GPU?
12:38 thereh: radeon x300
12:38 imirkin: wrong channel?
12:38 thereh: what ?
12:38 imirkin: nouveau is about nvidia gpu's
12:39 imirkin: i think you're looking for #radeon
12:41 thereh: I can't sent to #radeon channel
12:42 imirkin: probably need to register your nick
12:42 hakzsam: anybody here running the blob with a fermi?
12:55 hakzsam: imirkin, well, arb_pipeline_statistics_query-comp doesn't even work on the blob
12:55 hakzsam: it fails at linking the shader :/
12:57 imirkin: hakzsam: i doubt that one's gotten too much testing
12:57 imirkin: hakzsam: piglit could be wrong
12:59 hakzsam: okay, I'll double check the test
12:59 hakzsam: imirkin, it uses GL_LINK_STATUS with getProgramiv()
12:59 hakzsam: but the return value is zero
13:07 hakzsam: imirkin, http://hastebin.com/ocamusocay.avrasm did you already get this with valgrind-mmt?
13:07 hakzsam: this is on reator but I got the same thing on my computer..
13:09 imirkin: hakzsam: nope
13:09 imirkin: your piglit crashed...
13:09 imirkin: LOG: MSG: ==18052== by 0x532E4B3: piglit_report_result (piglit-util.c:269)
13:09 imirkin: LOG: MSG: ==18052== by 0x4017E6: piglit_init (indirect-compute.c:87)
13:10 imirkin: or at least called exit()
13:11 hakzsam: it only crashes with valgrind-mmt...
13:14 hakzsam: that's crazy, I don't see any things related to COMPUTE in the mmt trace
13:14 hakzsam: I have a serious problem :)
13:15 imirkin: share the mmt?
13:16 imirkin: demmt could be missing it
13:16 imirkin: hakzsam: are you trying to figure out how to do indirect compute?
13:16 imirkin: it should be pretty straightforward...
13:16 imirkin: look at how indirect draw is done
13:16 hakzsam: the mmt trace is probably corrupted
13:17 hakzsam: nope, I try to figure out GL_COMPUTE_SHADER_INVOCATIONS_ARB
13:18 imirkin: is that the pipeline statistics thing?
13:18 hakzsam: yes
13:18 imirkin: oh right. we don't already pull it anyways...
13:22 hakzsam: yes, but I need it for bin/arb_pipeline_statistics_query-comp
13:22 hakzsam: this is not that hard but... I need to find which value to use for QUERY_GET ;)
13:23 hakzsam: imirkin, like you did for tesselation shaders
13:23 khazakar: Hi all. Now I'm using kernel with patches from karol github,but still I have "choke" to 966 MHz on core. Only fan control better itself
13:33 khazakar: imirkin,you know hot to make voltage hack on my gtx 760?
13:34 khazakar: s/hot/how
13:39 imirkin: khazakar: pastebin your dmesg
13:41 khazakar: spam with evbug,nothing else. dmesg works fine only on my notebook
13:41 skeggsb: imirkin: i barely recall how all that stuff works now, but no, it doesn't free stuff until the bufctx is destroyed, it shoves unused refs onto a freelist and reuses them
13:42 imirkin: skeggsb: joi is seeing that list just grow infinitely
13:43 skeggsb: that's... unexpected
13:44 imirkin: skeggsb: where by "that list" i mean both the one in the bin, and the one in bufctx->current
13:46 khazakar: So...what I need to pastebin,when dmesg on my PC works crazy
13:46 thereh: /var/log/dmesg
13:47 khazakar: Clear
13:47 imirkin: skeggsb: urgh, there's also a mix of singlely and doubly linked lists
13:48 khazakar: It's siduction,debian unstable-based distro :P
13:48 khazakar: The same works on my notebook and dmesg works like a charm :D
14:07 khazakar: BTW - How I can help with developing a nouveau driver,even I know only a very little about C and programming generally?
14:37 imirkin: hey guys, i'd really like to get a bunch of testing done on this patch: http://patchwork.freedesktop.org/patch/69589/
14:37 imirkin: ideally it shouldn't break things, but i'm sure it will
14:38 imirkin: Tom^: --^
14:38 hakzsam: imirkin, I bet you did not run piglit? ;)
14:39 imirkin: hakzsam: safe bet
14:39 imirkin: hakzsam: but piglit is *very* unlikely to be affected
14:39 imirkin: hakzsam: basically this is going to affect big and semi-big shaders
14:39 imirkin: esp ones which do integer division
14:39 imirkin: how many piglits do that? :)
14:40 hakzsam: I don't know :)
15:48 imirkin: skeggsb: looks like this bufctx thing might have all been a case of PEBKAC.
15:53 skeggsb: imirkin: how so?
15:54 joi: not up to date mesa, missing http://cgit.freedesktop.org/mesa/mesa/commit/?id=109c348284843054f708f4403260739b7db18275
15:54 imirkin: one could argue between which keyboard and which chair the problem lay
15:54 skeggsb: ;)
15:54 imirkin: was it joi for not having the latest mesa, or was it me for not explicitly checking he had my fix even when it sounded like that exact problem
17:31 orbea: imirkin: i tried metro 2030 again, this time without clocking up, it ofc happened again, but I was able to reboot from a ssh. Here is dmesg - http://dpaste.com/3328EBZ syslog - http://dpaste.com/0AXRSJ5 xorg log - http://dpaste.com/2CX5M9Q
17:32 orbea: the entire screen went red and black
17:33 imirkin: orbea: i assume no chance you got the game process's output?
17:33 imirkin: skeggsb: thoughts on improving our error reporting?
17:33 orbea: i tried to apitrace it, but it didn't work
17:34 imirkin: i really don't like that xorg is the process that hits an error first
17:35 orbea: I might of needed the 32 bit apitrace...
17:35 skeggsb: imirkin: not sure what else you want aside from passing on the errors the gpu tells us...
17:35 imirkin: that smells like ttm messup
17:35 skeggsb: dumping the entire gpu state is unfeasible..
17:35 imirkin: skeggsb: i want the reason those errors happened
17:35 imirkin: skeggsb: ok, but we could dump some of it
17:35 imirkin: maybe even into a debugfs file like the intel guys do it
17:35 skeggsb: "someone fucked up and told the gpu to read from an unmapped address|?
17:35 imirkin: specifically the commands leading up to this
17:35 imirkin: the list of buffers that were actively mapped
17:35 imirkin: etc
17:35 skeggsb: what else is there to that kind of error?
17:36 imirkin: i want to know what code to look at
17:36 imirkin: it's not retarded monkeys submitting commands... it's a handful of very fixed pieces of code
17:36 imirkin: so you can generally map the commands back to the specific code that was using it
17:36 imirkin: then you can see where that buffer was being used
17:36 imirkin: etc
17:37 skeggsb: the info on the location in the push buffer that triggered a given fault is long gone before gr even reports the error
17:37 imirkin: sure, but it doesn't have to be gone
17:37 imirkin: we could keep it around for a bit
17:37 imirkin: i don't care if it's some special debug mode
17:37 imirkin: but we're getting errors like this
17:37 imirkin: and there is NOTHING i can do to debug them
17:38 imirkin: basically having a thing which is like "here is the batch that led up to this error" imho would go a long way
17:39 imirkin: where batch == pushbuf contents, buffer lists + their va's, maybe some other things that are easy that i'm not thinking of
17:39 skeggsb: well, you could make userspace no recycle/destroy push buffers until it's guaranteed they've completed - and dump them when a channel has hung or something
17:39 skeggsb: not*
17:39 imirkin: skeggsb: ok, but it's some steam game, who knows where its output goes...
17:40 imirkin: skeggsb: i guess i could write it to /tmp/something
17:40 skeggsb:is really wondering why the sudden influx of ce2 issues too
17:40 imirkin: skeggsb: but i'd also want to correlate that with the info from dmesg
17:40 imirkin: i.e. it is *this* buffer that wasn't there
17:41 imirkin: the ce2 stuff is weird... i can't help but feel it's something i did in one of my resource management rejiggers
17:41 skeggsb:needs to push the patch that lets the kernel recover from that
17:41 imirkin: CE2 = p2mf right?
17:42 imirkin: i've also *never* seen it myself... but i don't use that GK208 for much
17:42 skeggsb: no, i believe "p2mf" is InlineToMemory (i think that's what nvidia call it), ce2 is DmaCopy
17:42 skeggsb: class 0x*b5
17:43 skeggsb: subchannel 4 on gr channels
17:43 imirkin: hm ok... let's see what this maps to in mesa
17:43 imirkin: ah ok. it's the thing i thought it was
17:43 imirkin: SUBC_COPY
17:43 skeggsb: sounds legit
17:44 imirkin: i thought that was p2mf :)
17:44 imirkin: anyways.... i thought i was pretty careful about making sure that bo's weren't released before being used
17:44 imirkin: but.... it's all subtle
17:44 skeggsb: i thought p2mf was what we called class 0x*40, but i could be wrong there too ;)
17:44 imirkin: p2mf is the m2mf replacement
17:45 imirkin: #define NVE4_P2MF_CLASS 0x0000a040
17:45 imirkin: ok you're right :)
17:46 imirkin: if (screen->m2mf->oclass == NVE4_P2MF_CLASS) {
17:46 imirkin: BEGIN_NVC0(push, SUBC_COPY(NV01_SUBCHAN_OBJECT), 1);
17:46 imirkin: PUSH_DATA (push, 0xa0b5);
17:46 imirkin: i think i can be forgiven for being confused
17:46 skeggsb: yes, that tripped me up when i seen it too
17:46 skeggsb: somewhat misleading!
18:05 imirkin: skeggsb: should also make a ttm option that evicts everything not currently being used
20:16 Tom^: imirkin: will do.
20:17 Tom^: imirkin: where is the patch gonna be most noticeable?
20:17 Tom^: if it regresses that is :P