13:34thum: I have an nvidia NVS810 (GM107) graphics card and linux 4.9.6. During initialization I get output on my screen (passthrough), after getting ""nouveau 0000:03:00.0: DRM: allocated 1920x1920 fb: 0x60000, bo ffff9e7222fda000" the system gets stuck for a couple of seconds and then reboots.
13:34thum: ideas anyone? (whole log: https://pad.riseup.net/p/DXRW2TbVRtN6)
13:36RSpliet: thum: and this reboot is 100% nouveau related? booting with nouveau.modeset=0 does not make your system reboot after 2 minutes?
13:37thum: RSpliet: I would have to try "nouveau.modeset=0", the reboot is related to noueau (most likely), it boots up just fine with the NVS 810 removed.
13:38RSpliet: I've seen many systems with GM107 boot just fine, however this is the first Coreboot system I have come across. Forgive my scepticism ;-) Although it could be that nouveau needs to POST the card because the BIOS/EFI didn't do it for us...
13:39thum: RSpliet: I've been using another (much older) nvidia graphics card (don't have the model handy right now) with that board and boot firmware also just fine :)
13:44RSpliet: there still might be interaction between your GM107 and Coreboot that puts the card in a state that nouveau doesn't expect to find
13:45RSpliet: more debugging information from your kernel would be very helpful (and while we're at it, a 4.10 kernel, random problems could benefit from random fixes :-P)
13:49thum: with "nouveau.modeset=0" the system doesn't crash any more, but the screen is blinking in 0.5 second intervals. I forgot log log serial, so I'll do that again.
13:52pmoreau: If it’s due to Nouveau needing to POST the card, you could try `nouveau.config=NvForcePost=1`
13:54dboyan: karolherbst: I discovered most parameters passed to video encode engines are pointers to buffers of some kind. I don't know what their exact meanings yet, but I can get some hints from access pattern and relation between frames.
13:56dboyan: There are compute invocations between frames in your trace, but not in mine. I guess there's something to do with ffmpeg?
13:57karolherbst: might be
13:57karolherbst: dboyan: you could try to write an apitrace for nvenc
13:57karolherbst: and then you could copy those buffers back or so
13:58dboyan: that's a good idea, never thought about that before
13:58RSpliet: imirkin spotted btw?
13:58RSpliet: Or is he on a well-deserved holiday?
14:03dboyan: karolherbst: btw demmt's IB discovery doesn't work very well with nvenc traces, I just wonder why it doesn't work. It is missing something about GK104_COPY on your traces, and it is missing nearly everything except that on mine.
14:07dboyan: RSpliet: imirkin said he was busy with other things these days
14:50RSpliet: hakzsam: RE last Friday. imirkin proposed I'd increase PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT to 256 on Fermi (nvc0/nvc0_screen.c). I can confirm that this makes the messages in my kernel log disappear
14:50hakzsam: which sort of messages?
14:52RSpliet: [ 83.009944] nouveau 0000:01:00.0: gr: DATA_ERROR 00000004 [INVALID_VALUE] ch 7 [003f893000 Xwayland] subc 0 class 9097 mthd 2704 data 01cad060
14:53RSpliet: triggered by Pidgin. I don't see a visual difference
14:53RSpliet: in both cases (16 and 256) there are misrendered emoticons and icons in Pidgin. They show as black squares
14:54RSpliet: in Kepler emoticons and icons misrender too, but differently (semi-black background, correct icons)
14:58hakzsam: not sure if bumping PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT is the right fix though
14:58hakzsam: could you try recording an apitrace?
14:58hakzsam: mthd 2704 is IMAGE_ADDRESS
14:59hakzsam: maybe something is wrong in nvc0_validate_suf() too :)
15:10RSpliet: hakzsam: On the middle-to-long term I might be able to record you an apitrace
15:11RSpliet: but so far I haven't found a simple way of launching Wayland, XWayland without also starting gnome-shell
15:11RSpliet: ... implying it might become a rather large trace
15:16hakzsam: looks like tricky, yes
16:05dboyan_: RSpliet: You might try weston, I guess.
16:08Tomin: RSpliet: what do you mean by lauching Wayland without gnome-shell? Wayland is a protocol, gnome-shell (or whatever the Wayland component is) is an implementation
16:11dboyan_: Tomin: I guess his meaning is to launch "a wayland compositor that supports xwayland"
16:11hakzsam: RSpliet: ilia just sent a patch
19:37pq: I can't find much at all about playing Minecraft with nouveau. On my G96 it sometimes crashes in a nouveau component when the main menu sould appear, but if it goes past that, it's pretty solid to play.
19:38pq: any idea if that might be related to use of threads with OpenGL? well, perhaps a wrong place to ask, since the question is, would you happen to know if Minecraft uses multiple threads at start.
19:39pq: a cat? :-)
19:39Tom^: sorry, was cleaning the keyboard didnt notice the irc terminal was open
19:40Tom^: or rather focused
19:41pq: seems to usually be a segfault or an assert fail (kref?) in libdrm_nouveau.so
19:41airlied: pq: I have memories of someone saying minecraft was multithreading
19:43Calinou: airlied: Minecraft mods can provide multi-threading, such as OptiFine
19:43Calinou: and I think even vanilla Minecraft has some multi-threading bits
19:44Calinou: (in fact, OptiFine developers may tell you to disable threaded optimizations for OpenGL)
19:44pq: right, as I guessed - should try to find the threading hack-fix to see if it makes the crashes go away
19:45pq: I do have Forge installed
19:46pq: ah, here we go, the assert: libdrm-2.4.75/nouveau/pushbuf.c:727: nouveau_pushbuf_data: Assertion `kref' failed.
19:49pq: it sometimes takes a dozen tries to get started, but then I can play for hours once it gets through
21:32RSpliet: hakzsam: he probably read the logs ;-)
21:35hakzsam: RSpliet: yup, but he is not connected :)
21:39RSpliet:waves and says thanks
21:39hakzsam: I'm still not sure if it's the right fix...
21:39hakzsam: I wonder if images support is buggy in some sort
21:52dashs: Is there a step-by-step tutorial for xorg install on TV thru g98 (old GeForce 8400 )?
22:08Lyude: anyone know what I'd need to patch in nouveau to get a GTX970 with 4GB of ram working? I don't need all of the vram setup, just something that lets me use the card
22:08Lyude: there were some workarounds posted on the bz for that but none of them looked like they had actual patches
22:10skeggsb: Lyude: stick a "size -= 512MiB;" after the loop in nvkm/subdev/fb/ramgf100.c that detects the ram amount
22:11Lyude: skeggsb: cool, thanks