09:50emersion: does anyone have a good list of ftrace points for nouveau?
09:50emersion: i'm looking for something like these https://github.com/mikesart/gpuvis/blob/master/sample/trace-cmd-start-tracing.sh#L24
09:52karolherbst: not really. I think some of the ioctls functions would be useful to have though
10:25emersion: i'm getting 20k IO_PAGE_FAULT warnings in my dmesg…
10:25emersion: i wonder what could be the cause
10:36kwizart: emersion, which kernel version for the record ?
10:36emersion: but i'm trying to do things like amd→nouveau dmabuf transfers
10:37emersion: those just stall for seconds
10:38kwizart: interesting, I have page fault on jetson (so nouveau->tegra) reported at https://bugzilla.redhat.com/show_bug.cgi?id=1937129
10:38kwizart: with anything newer than 5.11-rc1
10:39kwizart: (and bissection pointing to 461619f5c3242aaee9ec3f0b7072719bd86ea207] drm/nouveau: switch to new allocator)
10:39kwizart: btw I cannot revert this commit easily on top of current 5.11
10:40emersion: it's not just a page fault you're experiencing, it's a bug when nouveau handles page faults
10:41emersion: i just have the page faults, without the nouveau bug
13:53emersion: hmm, NOUVEAU_GEM_PUSHBUF is taking like 9s
14:01kwizart: emersion, have you tried 5.11-stable-rc ? https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable-rc.git/log/?h=queue/5.11
14:01kwizart: emersion, there are 2 interesting patches (on on nouveau, the other on mm)
14:01kwizart: drm/ttm actually
14:02emersion: this isn't due to a kernel upgrade or anything
14:08emersion: bleh, i don't understand any of this stuff
15:00nzamb1: Hi All, my Linux is constantly freezing because of nouveau driver. I have pasted a log some time ago. I dunno if someone had a chance to have a look. I'm getting following error: https://pastebin.com/a8w9dGpC
15:00nzamb1: I'm running ubuntu 20.04
15:00nzamb1: it's very annoying that the system freezing
15:00imirkin: nzamb1: the top is cut off
15:01nzamb1: I don't understand you
15:01imirkin: the message is longer than what you indicated
15:02imirkin: you didn't include the start of it
15:02nzamb1: ok just a moment
15:06nzamb1: should be better now
15:07imirkin: nzamb1: is the order of the lines in this log inverted somehow?
15:08imirkin: ah yes, it is. that's why the trace looks backwards.
15:08imirkin: that makes a _lot_ more sense now
15:08nzamb1: journalctl -r
15:09nzamb1: yes it's inverted. I can paste without -r if needed
15:09imirkin: unfortunately i have no idea what would cause this
15:09imirkin: can you file a bug here, and make sure to include a "regular" order version of that log? https://gitlab.freedesktop.org/drm/nouveau/-/issues
15:10nzamb1: ok. Thank you. I'll create a bug
15:32nzamb1: imirkin https://gitlab.freedesktop.org/drm/nouveau/-/issues/78
15:32imirkin: nzamb1: please include the output of `lspci -nn -d 10de:`
15:33imirkin: (which should indicate which GPU you have)
15:35imirkin: interesting. 2 GPUs...
15:35imirkin: heh, FX 4800. does it sound like you're sitting on a jet engine?
15:37nzamb1: something like that but freezing GT218 [GeForce 210]
15:38imirkin: are there monitors connected to both? which is set as primary?
15:38nzamb1: monitors connected to booth and I'm running multi seat setup
15:39nzamb1: mykola@k-workstation:~$ loginctl
15:39nzamb1: SESSION UID USER SEAT TTY
15:39nzamb1: c4 1000 mykola seat0
15:39nzamb1: c8 1001 svetlana seat1
15:39imirkin: i'm largely unfamiliar with how multi-seat actually works in practice
15:39imirkin: does each seat get a GPU?
15:39imirkin: or a random collection of outputs?
15:40nzamb1: each seat get a GPU
15:40nzamb1: 1 GPU per 1 seat
15:40imirkin: like could a seat get HDMI-1 from gpu 1, and DVI-D-1 from gpu 2, while another gets DVI-D-2 from gpu1 and HDMI-2 from gpu2?
15:40imirkin: or it's at a "whole card" level?
15:40karolherbst: imirkin: https://gitlab.freedesktop.org/drm/nouveau/-/issues/78 this is fixed, right?
15:40imirkin: (or you just happen to have done it at the whole card level)
15:41nzamb1: it's a "whole card" level
15:41imirkin: karolherbst: i have no clue what the issue is
15:41imirkin: which is why i told him to file a bug
15:41karolherbst: ohhh right
15:41karolherbst: the other vfree thingy was something else.. ehh
15:41imirkin: perhaps it's fixed - i dunno
15:41imirkin: the other thing was kzalloc failing, and for pre-nv50
15:41imirkin: speaking of ... when is skeggsb_ back?
15:42karolherbst: next week
15:42karolherbst: I will talk with him what to do about trivial patches anyway. My plan is to convince him that we can just randomly merge smaller things
15:42nzamb1: if you have a fix and you can give it to me the I can test it
15:43imirkin: karolherbst: kk
15:43imirkin: nzamb1: i think karolherbst got confused with another unrelated issue which i fixed recently
15:44imirkin: we don't know what the problem is, at this point
22:52Lyude: mupuf, skeggsb_, jcline, imirkin - igt changes are on igt-dev/nouveau ml btw, about to send out the kernel side stuff as well