15:45 Wolf480pl: Hello. I have a GK104 in an Optimus setup, kernel 4.8.6. Got nouveau working like a year ago with config=War00C800_0=1 option. Both provideroutputsource and provideroffloadsink used to work. Today I tried using it again, offloadsink still works, but outputsource crashes Xorg.
15:45 Wolf480pl: No errors in dmesg or in xorg log
15:46 Wolf480pl: except some ACPI warnings during boot in dmesg
15:46 imirkin_: Wolf480pl: hmm weird. are you sure there's nothing in xorg on exit?
15:46 imirkin_: and then you don't really need to do RA at all
15:47 imirkin_: hakzsam_: --^ that last was meant for you
15:47 imirkin_: [wrong window]
15:50 Wolf480pl: the last thing in Xorg.0.log is "[Xi] Too man valuators reported for device 'Virtual core pointer'. Ignoring event"
15:51 imirkin_: Wolf480pl: that's in the crashed log?
15:51 imirkin_: Wolf480pl: are you sure X doesn't restart and you're looking at the new log?
15:51 Wolf480pl: yes, I movet all old logs elsewhere
15:51 Wolf480pl: but
15:51 imirkin_: check Xorg.0.log.old
15:51 Wolf480pl: on the tty where the X was running
15:51 Wolf480pl: there's a piece of sth like
15:52 Wolf480pl: "DetachUnboundGPU: Assertion slave->iSGPU failed.xinit: connection to X server lost"
15:52 Wolf480pl: but no newlines
15:52 Wolf480pl: and it's covering previous messages
15:52 imirkin_: interesting.
15:52 Wolf480pl: maybe I'll try to redirect stdout and stderr somewhere
15:53 imirkin_: can you pastebin xorg log?
15:53 Wolf480pl: (also, there's no Xorg.0.log.old)
15:54 Wolf480pl: I'll pastebin the log in a sec
15:59 Wolf480pl: Xorg.0.log: https://gist.github.com/anonymous/5b9db81f2294a72f4a445d7b693d7e48
16:00 imirkin_: odd. dunno.
16:00 imirkin_: try downgrading to 1.0.12?
16:00 imirkin_: there were some reverse-prime-related changes in 1.0.13
16:01 Wolf480pl: you mean downgrading xf86-video-nouveau to 1.0.12
16:01 imirkin_: yes
16:02 Wolf480pl: should I reboot after that, or is a fresh xorg start enough?
16:02 imirkin_: fresh xorg start is enough
16:05 Wolf480pl: still crashes
16:08 Wolf480pl: log: https://gist.github.com/anonymous/c08f89b0f3c0d1bd88daf5a33e124fe2
16:08 imirkin_: very odd
16:09 imirkin_: you could run X inside gdb. but that would require you start it from a remote machine.
16:09 imirkin_: and also some gdb expertise
16:09 imirkin_: as X uses signals a lot
16:09 imirkin_: you have to ignore a bunch
16:10 Wolf480pl: btw. in the console: "6EnableIOPorts: failed to set IOPL for I/O (Operation not permitted)Xorg: dispatch.c:3961: DetachUnboundGPU: Assertion `slave->isGPU' failed.xinit: connection to X server lost
16:11 Wolf480pl: btw. I tried redirecting startx's stderr to catch it, but Xorg refuses to start with redirected stderr (can't open console or something)
16:11 Wolf480pl: so I guess I'll have to try gdb
16:11 imirkin_: sounds like a core xorg bug =/
16:11 Lekensteyn: Wolf480pl: what xrandr command are you using? I managed to crash xrandr with xorg 1.18.x (or 1.17.x?) by swapping the provider and source for --setprovideroutputsource (or by using "0" for one of them)
16:11 Lekensteyn: that was a few months ago
16:12 Wolf480pl: --setprovideroutputsource Intel nouveau
16:12 imirkin_: that's backwards
16:12 Wolf480pl: either I can't read, or the manpage is wrong...
16:12 imirkin_: https://nouveau.freedesktop.org/wiki/Optimus/
16:12 imirkin_: manpage is written to maximize confusion
16:12 imirkin_: sounds like mission accomplished
16:12 Lekensteyn: :P
16:12 Lekensteyn: patches welcome?
16:13 imirkin_: i just remember it as "it's the opposite of the thing that makes sense"
16:13 imirkin_: and everything works :)
16:13 Lekensteyn: whenever I look at the --setprovider... text, I have to read it several times before it makes sense
16:13 imirkin_: it's basically the ol' at&t vs intel syntax issue
16:23 yasu: what is a GEM object and DRM object? I assume that GEM object manages GPU memories. DRM object manages one CUDA context such as some GEM instances. Is that correct? What exactly do they mean?
16:29 karolherbst: yasu: there should be docs about that stuff within the kernel
16:32 yasu: I see I'll check them thank you
19:21 NanoSector: Lekensteyn: I think your patch is released with linux 4.8.7 now, I just wanted to say thank you so much for the help fixing this :)
20:09 karolherbst: skeggsb: mind adding support for the display/FB stuff to rnndb? I guess I could try to extract all that stuff out of the source files, but I think it would be faster if you do it
20:48 karolherbst: hakzsam_: two questions: what is plugged in, and how long are you gonna need it today?
20:49 hakzsam_: karolherbst: I forgot to shutdown reator
20:49 hakzsam_: gf108 and gm107
20:49 hakzsam_: feel free to rebot
20:50 hakzsam_: *reboot
21:05 karolherbst: mhh, on reator nouveau always needs so long to load, that I think I just caused a null pointer access
21:06 karolherbst: and sensors is silly "ERROR: Can't get value of subfeature in0_min: Can't read"
21:07 karolherbst: uhh, what crap
21:07 imirkin_: he needs to get a tray loader for gpu's :)
21:07 karolherbst: yes
21:08 imirkin_: i wonder if thunderbolt lets you do something clever here
21:08 karolherbst: mhhh
21:08 imirkin_: like create little powered thunderbolt enclosures
21:08 imirkin_: which house 1 GPU
21:08 karolherbst: if you don't care about having just 4x slot
21:08 imirkin_: and then have a switch or something :)
21:09 karolherbst: *4x pcie width
21:09 karolherbst: yeah
21:09 karolherbst: but I guess that isn't exactly cheap
21:09 imirkin_: yeah. probably not the *definition* of cheap...
21:09 imirkin_: also aren't the latest thunderbolt things like 40gbps?
21:09 imirkin_: and 1x pci = 1gbps? or is it 1GB/s?
21:09 karolherbst: and?
21:10 karolherbst: pcie is quite fast
21:10 imirkin_: yes, but measurably fast
21:10 karolherbst: ~ 8gbps per lane
21:10 karolherbst: for 3.0
21:11 imirkin_: with PCIev3, yeah
21:11 imirkin_: looks like
21:11 imirkin_: 2gbps for v1
21:11 imirkin_: i guess more v = more better :)
21:11 karolherbst: yeah
21:11 imirkin_: allegedly there's a v4 coming out
21:11 karolherbst: 4.0 doubles speed
21:12 karolherbst: anyway, you won't get more out of thunderbolt than a 4x pcie slot
21:12 karolherbst: well
21:12 karolherbst: you could daisy chain
21:12 karolherbst: no idea how much GPU will like that
21:13 imirkin_: PCIe is a network architecture
21:13 imirkin_: you can have switches, etc
21:17 karolherbst: ... nouveau 0000:01:00.0: volt: min: 0uv max: 1080000uv
21:17 karolherbst: makes sense?
21:18 karolherbst: looks like an invalid volt entry
21:18 karolherbst: oh well
21:22 imirkin_: 1.08V? doesn't seem crazy...
21:24 karolherbst: I meant the min value
21:24 imirkin_: that seems a little low
21:25 karolherbst: yeah
21:25 imirkin_: at least it's not negative!
21:25 karolherbst: and the entry in the vbios is 0x00 0x00 0xff 0xff
21:25 imirkin_: heh ok
21:26 karolherbst: wut
21:26 karolherbst: "Sorry, user karol is not allowed to execute '/bin/bash' as root " the heck
21:27 karolherbst: the heck
21:27 karolherbst: what is wrong with portage
21:27 karolherbst: it just overwrites all my files
21:27 karolherbst: in etc
21:27 karolherbst: ....
21:40 Lekensteyn: NanoSector: you're welcome, the patch is an easy workaround. Lenovo IdeaPad Z510 owners are more unlucky: https://github.com/Bumblebee-Project/bbswitch/issues/142
21:42 imirkin_: Lekensteyn: reading throught that bug... seems like a comedy of errors
21:42 karolherbst: uhh well, normal
21:48 NanoSector: Lekensteyn: yeah I read your email
21:50 NanoSector: going to boot with 4.8.7 now without pcie_port_pm=off parameter to see if it's really fixed :)
21:52 NanoSector: yep it is, thanks again :)
22:11 Lekensteyn: imirkin_: well, it is Lenovo, I heard you complain about Clevo, but Lenovo is completely nuts :P
22:11 Lekensteyn: sorry, karolherbst was calling out on Clevo
22:12 karolherbst: well yeah, but everyondy firmware is nuts one way or the other
22:20 Lyude: they changed the way their laptops report tablet mode for literally no reason the other day
22:20 Lyude: in ACPI I mean
22:22 Lekensteyn: perhaps new windows 10 requirements?
22:22 Lekensteyn: probably for the same reason you now get Win+P instead of an ACPI notification when you press the display switch Fn key..
22:25 imirkin_: Lekensteyn: always a reason... just often not a good one, and often not apparent to you.