12:57 pmoreau: Grrr… Dunno what changed in the latest releases of the nvidia-340xx package, but it's as if it added a nouveau.modeset=0 or nomodeset somewhere. Sooooo annoying!
12:59 imirkin: pmoreau: grep -r modeset /etc/modprobe*
13:00 pmoreau: imirkin: Already looked there :-)
13:00 pmoreau: Only a harmless blacklist
13:01 pmoreau: "+ cat "${pkgdir}"/usr/src/nvidia-${pkgver}/uvm/dkms.conf.fragment >> "${pkgdir}"/usr/src/nvidia-${pkgver}/dkms.conf"
13:01 pmoreau: Hmm… I wonder what's in that file
13:01 imirkin: there's also a nvidia_modeset thing now
13:01 pmoreau: Right
13:05 pmoreau: imirkin: You didn't experience anything similar?
13:06 imirkin: i never install bs like that
13:06 pmoreau: You get others to do traces for you of the blob? :D
13:09 imirkin: i just never install things like that. i run them out of controlled environments
13:09 imirkin: except with gentoo, those scripts are pretty reliable
13:10 pmoreau: ok
13:11 karolherbst: pmoreau: maybe grub directly?
13:12 pmoreau: Not using grub
13:12 pmoreau: But I thought about the commandline
13:13 pmoreau: But no modeset/nomodeset on it
13:13 karolherbst: mhh
13:14 karolherbst: pmoreau: why would anybody disable modeset anyway
13:14 karolherbst: pmoreau: if you install through pkgbuild, open a bug: _never_ disable modeset on recent kernels
13:20 karolherbst: what is /usr/lib/modprobe.d/ by the way?
13:20 pmoreau: There's nothing about changing modeset in the PKGBUILD, nor in the install script
13:21 pmoreau: Same as /etc/modprobe.d
13:22 karolherbst: ohh there is a echo 'blacklist nouveau' > "${pkgdir}/usr/lib/modprobe.d/nvidia.conf"
13:22 karolherbst: mhh
13:22 pmoreau: Except /etc/modprobe.d is for the user to add changes, whereas /usr/lib should be left untouched
13:22 karolherbst: ahh right
13:22 karolherbst: yeah, I don't see anything either
13:23 pmoreau: I do have my own blacklist nouveau as well
13:23 pmoreau: (and blacklist nvidia)
13:23 pmoreau: Don't want either to start without being asked
13:24 karolherbst: I figured that already
13:24 pmoreau: karolherbst: Do you know what the dkms version of the driver is? Is it the one handling kms?
13:24 karolherbst: no
13:25 karolherbst: dkms is just one module for multiple kernel versions
13:25 karolherbst: you build against some dkms magic API and then it works with multiple kernel versions
13:25 karolherbst: don't know the internals though
13:25 karolherbst: it is nvidia unrelated and can be used for like everything
13:26 pmoreau: Oh!! I should install that one instead :-)
13:26 karolherbst: the modeset module comes with 358
13:26 pmoreau: Makes sense
13:26 karolherbst: though it does not thing except to annoy
13:26 pmoreau: :D
13:27 karolherbst: you need dev branch of bumblebee
13:27 karolherbst: because rmmod can't remove nvidia anymore
13:27 karolherbst: because nvidia-modset uses it
15:26 pmoreau: imirkin: Hey, I'm not too far from the blob: "10000801 4400c780 10108005 40490ff d00e0005 a0c00781" vs "10000801 4400c780 10108005 40490ff d0000005 a0c00780 30000003 780" (blob vs Nouveau)
15:32 imirkin: the commands in the pushbuf might matter too
15:33 imirkin: the g14 thing worries me a bit
15:33 imirkin: i've seen a bunch of references to a c14 in the code, i bet they're related
15:34 imirkin: also you might need an exit, not a ret
15:34 pmoreau: What is g14? I mean, why 14?
15:35 pmoreau: Yeah, I had an exit at some point, and… it exited the code for some reason
15:35 pmoreau: Need to add it back
15:35 pmoreau: What code were you referring to BTW?
15:54 pmoreau: Oh, nevermind for the g14, think I understood. :-)
16:07 pmoreau: Ohhhhh! Nouveau defines a CB at address 202b0000 of size 4096 (0x1000), and passes as param 0x202b1000
16:08 pmoreau: Looks like this CB is the memory where I want to write, and I correctly get the pointer to it
16:09 pmoreau: But it's not going to work because Nouveau emits a CB but I read from a GLOBAL… --"
17:45 norris: just curious: has anyone seen recent issues with NVCF (550 Ti) locking up and switching to fbcon? with messages like:
17:45 norris: nouveau E[ DRM] GPU lockup - switching to software fbcon
17:46 norris: I don't see much else in the Xorg or kernel logs about it, so I don't have much to go on
18:54 imirkin: norris: anything logged before that? what kernel?
19:24 norris: imirkin: kernel 4.2
19:25 norris: using the ubuntu 15.10 kernel, but I think I got similar results last time I tried my mainline 4.2 build
19:25 imirkin: anything in the logs before the lockup?
19:26 norris: nope
19:26 imirkin: any observations about some commonalities wrt what you're doing prior to the lockup?
19:26 norris: just web browsing (firefox)
19:27 norris: I had gotten other lockups on ubuntu 15.04 with 3.9 kernel, so I tried 15.04 w/ mainline 4.2 (similar problems), then upgraded to 15.10 when I saw the release (and am seeing similar lockups)
19:27 imirkin: sorry, no great ideas from me =/
19:28 imirkin: you could try a 4.3-rc kernel... if you wait for rc7 that will have a fix for some hangs people have been experiencing on resume, but i highly doubt it's the same issue
19:29 norris: there were some more verbose errors on 15.04 (with a 3.19 kernel), but I'm not sure if that's a good datapoint. I still have the logs though...
19:29 norris: yes I could try 4.3-rc, I suppose
19:31 imirkin: nouveau got a bit of a rewrite in 4.3, perhaps something magically got fixed
19:31 norris: fwiw, this lockup was without any suspend/resume
19:31 norris: oh
19:31 norris: yes, rewrites can do that :)
19:32 norris: but often the reverse too :)
19:33 imirkin: well, low vram quantities could also trigger the issue
19:34 imirkin: but i doubt your NVCF has 256mb of vram
19:42 norris: imirkin: referring to this? 2a6c521bb41c drm/nouveau/gem: return only valid domain when there's only one
19:42 imirkin: norris: yes
19:47 norris: the box for the GPU says 1024MB vmem, but I'm not actually sure how to check that... lspci lists ranges of 32MB, 128MB, and 64MB
19:48 imirkin: dmesg should say it when loading nouveau
19:48 norris: ah right, confirms 1024MB
19:53 norris: imirkin: ok, well I'll see if 4.3-rc does anything for me, or if I can get more useful debug info. thanks for the help!
19:59 norris: ugh, just locked up again (cursor can move but graphics dead); this time Xorg.0.log contains a bit (nothing in dmesg yet):
19:59 norris: snippet: http://pastebin.com/8mHwzpVz
20:01 norris: I'll file a bug if I see anything more
20:11 imirkin: that's just another way of saying "gpu is hung"
20:11 imirkin: (X queues events, and when that queue gets too long it prints that error)
20:16 norris: ah makes sense