07:19RoverGT: Hi all!
07:27RoverGT: Hi, I'm getting a problem with my gt710, first nouveau didn't probe because my hardware is able to work only with 16K pages. Then we found thread with fix for SPARC with 8K page and did the similar thing: changed pageshift in nouveau_bo.c file from 12 to 14. But colors look inverted(black is green etc). Kernel is 4.4.24 version. Any Ideas how to fix this? Thanks
10:50sbstn: I'm having a hang with nouveau (cursor moves, display stuck otherwise, keyboard stuck), seems to be always with firefox as well (google maps seems to trigger it every time)
10:52sbstn: I don't really know what to make of the kernel logs due different kind of error messages
10:54sbstn: for example: https://pastebin.com/e0uDjmNr https://pastebin.com/ATijb3GX (waited a bit before resetting, overall got over 1M lines of logs mostly similar messages from past few hours)
10:55sbstn: 4.9.0 kernel and gtx 670
12:11imirkin: RoverGT: what hardware is this, out of curiousity?
12:11imirkin: RoverGT: also, is it BE by any chance?
12:20imirkin: skeggsb_: https://github.com/skeggsb/nouveau/blob/master/drm/nouveau/nouveau_ttm.c#L240 -- should that be reg->num_pages << PAGE_SHIFT instead of 12?
12:31RoverGT: imirkin: MIPS devboard
12:31imirkin: ah. is it LE or BE?
12:32imirkin: at least there's that ;)
12:33RoverGT: But it works....exept colors
12:35RoverGT: we used 14 instead of 12
12:36imirkin: where precisely?
12:38RoverGT: imirkin: in nouveau_bo.c
12:38imirkin: right ... where in nouveau_bo.c?
12:41imirkin: RoverGT: btw, was it this shade of green perchance? http://www.hexcolortool.com/#008500
12:41RoverGT: nvbo->page_shift = 14; somewhere near string 220
12:42imirkin: ah yeah, that won't work...
12:42RoverGT: imirkin: this color you've sent is purple
12:42imirkin: oh grrr
12:42imirkin: that thing doesn't reload properly
12:42imirkin: can you copy #008500 into the "enter a color" thing, and look at that?
12:43imirkin: actually should be #008700 but wtvr.
12:43RoverGT: looks like it
12:44imirkin: ok, so that happens to be the RGB color that YUV 0 comes out to
12:44imirkin: no clue why you'd have a YUV conversion enabled, but that's the only way i can think of you'd be seeing green
12:45RoverGT: we changed some other 12 to 14 and image looked differently, but still readable
12:45imirkin: (and it's a darker hue of green... pretty specific)
12:45imirkin: skeggsb_: https://github.com/skeggsb/nouveau/blob/master/drm/nouveau/nouveau_bo.c#L225
12:45imirkin: should that be >> nvbo->page_shift?
12:46imirkin: ah no. it shouldn't be.
12:46imirkin: RoverGT: well, there's obviously some confusion in the code about GPU pages vs system pages
12:46imirkin: RoverGT: unfortunately it's tricky to fix because you have to keep the concepts very separate
12:47RoverGT: Yup, I read about problems with pages when got -12 probe error
12:48RoverGT: Someone fixed this on SPARC with 8K pages
12:49imirkin: yeah, dunno if that's a valid fix
12:50imirkin: perhaps we should allocate VRAM in granularity of system pages...
12:50imirkin: that would make handling pagefaults for mapped vram a lot simpler
12:50imirkin: otoh mapping a random 4 pages in shouldn't be a big issue. dunno.
12:57RoverGT: Maybe it is possible to do something with colors, just add some offset to make black, black again.
13:12imirkin: yeah dunno. i forget how the LUT is uploaded exactly
13:12imirkin: could be that it's pointing at a bogus page
13:14orbea: so one of newer dolphin-emu commits started segfaulting with nouveau with either the hybrid or exclusive shaders. Is this something dolphin is doing wrong, or did they hit a nouveau limitation? gdb log: https://pastebin.com/CavSKjwZ first bad dolphin commit: https://github.com/stenzek/dolphin/commit/bcaf00c60ddb3d5465c6cfbeafb9e48500059a0e and the ubershaders PR:
13:14orbea: *ubershader commits
13:18imirkin: nvc0_program_translate:610 - shader translation failed: -4
13:18imirkin: that's really bad
13:18imirkin: make an apitrace
13:19orbea: that was there before the segfault btw
13:19imirkin: OGL: Issue draw call after compiling on worker thread
13:19imirkin: that won't work with nouveau.
13:19imirkin: drawing from multiple contexts at the same time = crash on nouveau
13:20orbea: shame :\
13:20orbea: here is an older trace when it still worked http://ks392457.kimsufi.com/orbea/stuff/trace/dolphin-emu_ubershaders.trace.xz
13:20imirkin: does that also have the shader translation failed stuff?
13:20orbea: when the 'shader translation failed: -4' was still present
13:20imirkin: can you file a bug?
13:20orbea: for nouveau?
13:21imirkin: coz i'll forget in 5 minuts
13:21imirkin: and i don't have time to investigate right this second
13:21orbea: understandable :)
13:26imirkin: i was going to make a comment on that issue, but seems like you beat me to it
13:29orbea:waits for bugzilla to send me an e-mail resetting my pw :\
13:40orbea: made a bug https://bugs.freedesktop.org/show_bug.cgi?id=101708
16:00Lyude: mupuf: btw, did we have any plans for doing https://trello.com/c/Rm03teQP/48-test-the-i2c-stack in userspace? or are we also looking for a hw solution for this (e.g. chamelium)
18:11sbstn: I don't really know what to make of the kernel logs due different kind of error messages. Kernel 4.9.0 and gtx 670
18:12sbstn: Some of the logs: https://pastebin.com/e0uDjmNr https://pastebin.com/ATijb3GX (waited a bit before resetting, overall got over 1M lines of logs mostly similar messages from past few hours)
18:12imirkin_: sbstn: i saw those logs
18:12imirkin_: unfortunately i don't really know what to make of them either =/
18:12imirkin_: i could say "try kernel 4.12", although i can't earnestly say that i expect it to improve the situation
18:13sbstn: it is kind of interesting due how I can trigger those hangs by simply visiting google maps
18:13sbstn: its also very annoying :\
18:13imirkin_: yeah, google maps used to hang stuff all the time
18:13imirkin_: on i965 too
18:13imirkin_: they made lots of use of webgl
18:13sbstn: oh, didn't realize that
18:14imirkin_: but i thought things had been better
18:14imirkin_: you might be able to configure the browser to disable some stuff
18:14sbstn: yeah, I think I'll do just that
18:14sbstn: thought it might have been the drivers, but if it also happens on other hardware I imagine the issue might be elsewhere as well
18:15imirkin_: no, unrelated issues
18:15imirkin_: it was just a heavy GL user
18:17sbstn: thanks, think I tuning browser will be good enough solution for now
18:18imirkin_: (i965's all better now btw... but they have a real development team with documentation behind the driver, so it's like cheating...)
18:21sbstn: (yeah, I wouldn't mind using i965 if I had such cpu, but decided to go save a 20€ instead by buying a xeon model without, only later got to question that decision)
18:22imirkin_: amd also has a paid team supporting open-source graphics
18:22imirkin_: i'd say it ends up being less well supported than intel, but still pretty good
18:22imirkin_: (there are a number of reasons for that, of course... like there are a ton more different parts for amd to support)
18:26sbstn: I never was a huge fan of nvidia's proprietary ones; I definitely appriciate the work done for nouveau drivers :)
18:27imirkin_: btw, starting with 4.10 you should be able to mostly-reliably reclock your gpu
18:29sbstn: I recall trying it when it was new feature: trying out the highest clock setting had some issues ;D
18:29imirkin_: yeah, a lot of that's been addressed now
18:29imirkin_: karolherbst spent a ton of time working out the kinks
18:30karolherbst: yeah, skeggsb_ should merge those patches :p
18:30karolherbst: ohh wait
18:30karolherbst: you talk about the 4.10 ones
18:30karolherbst: still he should merge the other fixes as well
20:44mupuf: Lyude: skeggsb_ has implemented the i2c stack in pdaemon already
20:44Lyude: the testing for it as well?
20:44mupuf: the point was to talk to the power sensor
20:45mupuf: the testing can be done with a tool that I wrote
20:45Lyude: Oh. whoops, i mixed that up with a different unrelated issue then
20:45mupuf: it polls on the sda/scl lines and decodes the i2c transactions
21:40karolherbst: 's like total chaos here right now
21:41imirkin_: https://www.washingtonpost.com/world/europe/hamburg-braces-for-mass-protests-asg-20-makes-the-city-a-siteof-global-discord/2017/07/05/afafae46-6044-11e7-80a2-8c226031ac3f_story.html ?
21:42karolherbst: imirkin_: https://www.pscp.tv/w/1lDxLjlrmyyGm
21:42karolherbst: I am literally there, but safe in a big building
21:42imirkin_: sounds fun.
21:42karolherbst: kind of
21:43karolherbst: like the permament sound of helicopter and explosions
21:43karolherbst: still need to get home somehow
21:43imirkin_: leave that to tomorrow...
21:43karolherbst: it's safe though
21:43karolherbst: only if the police is in the near it gets loud
21:43karolherbst: but everywhere where the police is not it's calm
21:44karolherbst: the police also respect some rules like not punching bystanders
21:59Riastradh: Heh. `We don't punch bystanders' is quite the low bar.
22:01karolherbst: they ran into the demonstration after one bottle was thrown
22:01karolherbst: and totally broke that 12k people demonstration apart
22:03Riastradh: Obvious lesson: Don't like a 12k-person protest? Mingle into it and chuck a single bottle at the police!