10:58 dcomp: so I'm running piglit on my gm108 ... And getting quite a few SCHED_ERROR 0d ... Is that normal?
10:59 hakzsam: it is not
10:59 hakzsam: piglit runs fine on gm107
10:59 hakzsam: which test?
10:59 hakzsam: which version of mesa as well?
11:01 dcomp: Piglit running at the moment. But within the first 5000 tests
11:01 dcomp: mesa version is: 12.1.0_devel.83540.2060f19-1
11:02 dcomp: compiled on arch
11:02 hakzsam: compiled from today?
11:03 dcomp: compiled yesterday.
11:04 dcomp: last patch: vbo: Fix handling of POS/GENERIC0 attributes.
11:04 skeggsb: no, that's pretty "normal" (ie. it's known to happen, but probably shouldn't) on maxwell piglit
11:04 hakzsam: okay, just before GL4.1 on maxwell :)
11:04 hakzsam: skeggsb, on gm108?
11:04 hakzsam: because it rusn quite fine on gm107
11:05 skeggsb: oh, it runs fine, but those sched errors happen
11:05 hakzsam: I have never seen those sched errors with a full piglit run on gm107 though
11:06 skeggsb: weird, i'm pretty sure i have
11:06 skeggsb: i've had them on *some* board
11:06 skeggsb: you run without -1?
11:06 hakzsam: anyways, I guess writing a sched code calculator should probably help?
11:06 hakzsam: yes
11:06 hakzsam: err, with -1 always
11:07 skeggsb: i don't think it'll help *that* issue, but yes, it'd help a lot!
11:07 skeggsb: ah, it might just happen without
11:07 skeggsb: takes too long with -1 :P
11:07 hakzsam: sure
11:07 hakzsam: dcomp, did you use -1 as well?
11:07 hakzsam: bbl
11:07 dcomp: -1? is a a piglit option?
11:08 dcomp: DRI_PRIME=1 piglit run all nouveau-mesa
11:08 dcomp: nothing special on the command line
11:09 skeggsb: it forces piglit to run only 1 test at a time
11:11 skeggsb: hakzsam: i have a few ideas in a branch somewhere that helped with the "sched error 0d" thing, but haven't (naturally) gotten around to making that mergable yet
11:12 skeggsb: iirc it was locking around the chan kick stuff (and maybe a few other things), but i'd have to dig it up
11:16 karolherbst_work: I think I also get some random stuff on my kepler with piglit, also some sched errors, but the gpu never hangs because of that
11:16 skeggsb: yeah they weren't fatal
11:17 karolherbst_work: I also get some LB errors usually
11:17 karolherbst_work: and stuff like that
11:17 skeggsb: some of the dmesg errors are normal (the MEM_OUT_OF_BOUNDS etc) for some tests that are deliberately triggering those conditions
11:17 skeggsb: those ones aren't normal
11:19 dcomp: should I run piglit in a higher pstate ... Or should it not make a difference
11:19 skeggsb: doubt it'd make a difference
11:19 skeggsb: or, not much of one, you'll be limited by context switch and channel creation speed more than anything
11:20 karolherbst_work: right, I get "LB_ERROR"
11:20 karolherbst_work: and FB_FLUSH_TIMEOUT
11:21 karolherbst_work: skeggsb: I could try to look into those, but they didn't cause anything severe here and I have no clue how they appear at all anyway or what triggers those. They just happen randomly for now
11:42 hakzsam: skeggsb, okay, cool
11:42 hakzsam: dcomp, please try to run piglit with -1
11:44 dcomp: hakzsam: I'll do that after this run. only 10000 more tests
11:44 hakzsam: okay
11:44 hakzsam: and send me the results file please
11:45 hakzsam: results.json.bz2 when the run is done
11:45 hakzsam: I could compare with gm107
11:46 hakzsam: dcomp, and the next time you will update mesa, you will get GL4.1 on your maxwell, because currently it's GL3.3
11:48 karolherbst_work: I plan to do some more work this weekend on nouveau, but I really don't know a good little task I could do :/
11:48 karolherbst_work: oh well, I could cleanup my mesa patches I think are safe
11:50 dcomp: is runpm expected to work with config=NvClkMode?
11:51 karolherbst_work: dcomp: I am pretty sure anything runpm related and reclocking is totally broken now, but skeggsb thinks otherwise :p
11:51 karolherbst_work: dcomp: I am like super sture, that it ain't fun to reclock when the gpu also suspends in between
11:52 dcomp: Is it meant to reclock on resume>
11:53 dcomp: or switch back to whatever the vbios sets it at. (which in my case is not working)
11:56 karolherbst_work: I am sure it switches back
11:57 karolherbst_work: but it is just totally broken anyway
11:57 karolherbst_work: even if you reclock by hand, it resets to defaults
11:57 karolherbst_work: after resume
11:57 karolherbst_work: I fixed that stuff on my branch though
11:57 karolherbst_work: the only thing missing there is reclock why suspended
11:57 karolherbst_work: *while
11:58 dcomp: is this the stable_reclocking branch>
11:58 karolherbst_work: yeah
11:58 karolherbst_work: it also has support for memory reclocking on gm10x :p
11:59 dcomp: I'll try running without runpm=0 then. I need NvClkMode=7 to boot the card
11:59 karolherbst_work: mhhh
12:00 karolherbst_work: so if the card isn't reclocked nouveau will print out errors?
12:00 karolherbst_work: mhh
12:00 dcomp: yeah.
12:01 dcomp: It'll actually end up spewing SCHED ERROR 20 and not let me unload the module
12:07 karolherbst_work: well
12:07 karolherbst_work: you could try out my branch either way to check if memory reclocking works without issues
12:08 karolherbst_work: it isn't really that much tested besides a "yeah, seems to work" and some small tests on my side, though some other users tested it and didn't really reported anything bad back
12:09 dcomp: I am running your branch
12:09 dcomp: just with runpm=0
12:10 dcomp: and NvClkMode*
12:20 karolherbst_work: I see
12:20 karolherbst_work: no, with my branch it should work
12:25 karolherbst_work: dcomp: I actually I didn't test it _that_ much, but it should be much better than now. And if you find any issues while runpm is enabled, you can always tell them to me
13:11 dcomp: hakzsam: I'm running with -1 now. https://home.dcomp.co.uk/piglit/html/problems.html
13:12 dcomp: ^ is the last run
13:53 shrdlu68: Hi, I have an ageding iMac G5 with a GeForce FX Go5200. I haven't had any luck with nouveau thus far, booting into a blank screen if I don't boot with `nomodeset video=ofonly`
13:56 shrdlu68: I saw some information onllne about incompatibility with 64k PAGE_SIZE, so I set mine to 4k. No MSI. All, this was a while back, though, I gather.
14:01 RSpliet: wow, Samsung has the most annoyingly unreadable DRAM packages in the world
14:01 RSpliet: ultra-shiny to obscure the black-on-black letters under a microscope
14:01 RSpliet: but nice to know the Jetson TX1 has two K4F6E304HB-MGCH :-P
14:02 karolherbst_work: :D
14:04 shrdlu68: Says at nouveau.freedesktop.org that 2d and 3d acceleration is supported on all cards.
14:05 shrdlu68: I knew it! It's a family curse, that's why mine won't work...
14:06 RSpliet: shrdlu68: there's only a handful of uses with that particular GPU. 2D 3D accel used to work on GeForce FX mostly, not sure about the status on Mac, but I recall another user struggling with such a combination
14:07 shrdlu68: RSpliet: Know anyone with a working GeForce FX?
14:10 karolherbst_work: those OF macs are double painful. imirkin once tried or actually fixed nouveau for those
14:11 shrdlu68: imirkin_: I'm calling you out.
14:12 imirkin: should work...
14:12 imirkin: kinda-sorta
14:12 imirkin: however that's for the 3D stuff. 2d items should come up without issue
14:13 imirkin: iirc i fixed up kernel 4.3... or maybe it was 4.4
14:13 shrdlu68: No 2D for me.
14:13 imirkin: which ran fine on my G5 + agp NV34
14:13 shrdlu68: Perhaps I should try with PAGE_SIZE 64k?
14:13 imirkin: definitely not
14:13 shrdlu68: Or with MSI?
14:14 imirkin: also, unless you're using a HPC load, 4k pages are preferable in any case
14:14 imirkin: NV3x didn't support MSI at all
14:14 shrdlu68: Well then, it's an old family curse.
14:14 imirkin: if you don't want it to remain that way, you should share more information
14:15 imirkin: like ... what issue you're running into
14:15 shrdlu68: Just boots into a blank screen. I'm on kernel 4.4.6.
14:16 imirkin: contents of dmesg may be interesting
14:16 shrdlu68: Blank black screen, to be specific.
14:16 imirkin: when in the boot process does the screen turn black?
14:17 shrdlu68: Well sometimes instead of turning blank it hangs at "switching to nouveaufb..."
14:17 karolherbst_work: imirkin: sure it was 4.4?
14:18 imirkin: well, it was definitely the initial 4.3 merge that broke it
14:18 imirkin: which happened to coincide with when i got the G5
14:18 imirkin: i don't remember if all the fixes went into 4.3 or if some went into 4.4...
14:18 imirkin: let's see...
14:19 imirkin: yep, v4.3 should have all the fixes
14:20 imirkin: specifically commits 15ee005893c5ac8aabb66c2ad0033ef31dbdcfa3 and 25d295882a1adfcdaaad85369289677b87c7c8f0
14:21 karolherbst_work: shrdlu68: try to get dmesg via ssh after the screen goes black/freezes
14:22 shrdlu68: Ok. Lemme reboot...
14:22 orbea: ideally, programs should not use the OpenGL compat profile?
14:23 orbea: and rather use the core profile?
14:23 imirkin: that would be ideal, yes.
14:23 karolherbst_work: orbea: there is no compat profile on mesa afaik anyway (or at least no useful one)
14:23 orbea: yea, its 3.0
14:23 imirkin: although there are some features that hw still implements which are only available in compat
14:23 imirkin: like alpha tests and quads
14:24 karolherbst_work: orbea: on mac os x it is even 2.1
14:24 orbea: heh...
14:25 orbea: thanks, I will pass this along to the dev of the program I am having issues with (oneof the libretro n64 cores)
14:26 karolherbst_work: yeah, people should stop using compat profiles or stop using GL3+ features :p
14:32 shrdlu68_: Here's the complete dmesg output: http://vhbin.net/46768s6jhh6e
14:43 shrdlu68_: The paste was truncated, dunno why. Anyway, here: http://vhbin.net/46768s6jhh6e
14:43 RSpliet: shrdlu68_: is the paste truncated or the log?
14:44 RSpliet:waves at effractur
14:44 RSpliet: hello other PowerPC GeForce FX user!
14:44 shrdlu68_: The paste was truncated, dunno why. Anyway, here: https://virtualhacker.net/zerobin/?39ca4e796a7011c6#v4ggmbJWGImd6wqH6q5MiT4WRCMEi9YwT8V2voBRsHY=
14:44 shrdlu68_: *wave violently
14:44 RSpliet: there's no mention of nouveau in that dmesg
14:46 shrdlu68_: RSpliet: Yeah, there's rivafb, I'm confused.
14:49 shrdlu68_: This is strange, lspci says driver in use is rivafb. Lemme remove it and recompile...
14:51 imirkin: it's likely that nouveau doens't play well with rivafb/nvidiafb
14:51 imirkin: orbea: well, it depends what hw you want to target
14:51 shrdlu68_: Yeah, seems to be the case according to this person: http://www.linuxforums.org/forum/red-hat-fedora-linux/21588-what-rivafb-how-do-i-disable-fedora-core-3-a.html
14:52 imirkin: orbea: if you want to target DX10+ hw, core context is probably best. however if you also want to include DX9 hw, then you have to limit yourself to GL 2.1
14:55 shrdlu68_: CONFIG_RIVA_FB says "This driver supports graphics boards with the nVidia Riva/Gefo"
14:55 shrdlu68_: -rce chips.
14:55 imirkin: it's a fbdev driver
14:55 shrdlu68_: Safe to disable?
14:55 imirkin: definitely
14:55 imirkin: as is CONFIG_NVIDIA_FB
14:56 imirkin: you could keep them as modules and blacklist them if you wanted
14:56 shrdlu68_: Oh, so what exactly is the "framebuffer"?
14:57 imirkin: fbdev... old thing before kms
14:57 shrdlu68_: Should I disable the Openfirmware fb too?
14:57 imirkin: no, leave that one
14:57 imirkin: otherwise you don't get a console
14:57 shrdlu68_: Ok.
14:58 imirkin: [while booting... nouveau will kick out offb when it loads)
15:01 shrdlu68_: imirkin: Libations for everybody! It works now.
15:01 imirkin: ok... don't go too crazy with the 3d stuff btw -- it (a) doesn't even work that great on x86, and (b) works even worse on ppc
15:02 shrdlu68_: No armagetronad? :(
15:03 imirkin: well, feel free to try stuff. just don't be too surprised when it doesn't work
15:03 imirkin: but hey, at least the colors show up properly now
15:03 shrdlu68_: Yeah! I can see tux!
15:04 imirkin: used to get swizzled wrong iirc
15:05 shrdlu68_: It's perfect.
15:06 imirkin: and of course there's llvmpipe if you really need to run some 3d application that doesn't work
15:06 imirkin: (you can force-enable it with LIBGL_ALWAYS_SOFTWARE=1)
15:06 imirkin: gtg, enjoy!
16:40 dcomp: hakzsam: https://home.dcomp.co.uk/piglit/mesa-git.83540.2060f19/problems.html
16:41 dcomp: https://home.dcomp.co.uk/piglit/mesa-git.83540.2060f19/results.json.bz2
16:45 dcomp: I had a watcher script which killed any test which ran for longer than 2 minutes
16:45 dcomp: to ensure it made progress
20:15 hakzsam: dcomp, I don't see any unexpected fails, but you should use: ./piglit-run.py -x glx -x streaming-texture-leak -1 --dmesg tests/gpu.py
20:27 Lekensteyn: skeggsb: I see a "Divide error" in gm107_ltc_oneinit (full demsg at http://pastebin.com/raw/3fasFxqE for some pre-v4.7 version). Could it be fixed by "drm/nouveau/ltc/gm107-: fix typo in the address of NV_PLTCG_LTC0_LTS0_INTR" or is it some other issue?
20:33 Lekensteyn: skeggsb: oh, it is actually the final v4.7 release
20:34 imirkin_: [ 29.092509] nouveau 0000:04:00.0: fb: 0 MiB DDR3
20:34 imirkin_: that's probably a bad sign of things to come
21:40 kloofy: Lekensteyn: but, all my things in lifes fight has been majorly against the flow, i am really saying due to this, it almost seems that spamming is the most i can do, scheduler is also easy, i can't gather myself due to problems, i know everything needed how things work though
21:50 kloofy: i just mean you have proven when you have enough of nice background that you can handle such over internet work as tough as the gpu world is, and i have proven that my backrground is such a crap, that i can't in that form, so basically so it is and so it will be, i am now listed as annoyins spammer ontop of my old very rough troubles
21:55 kloofy: so infact indeed i am out of this stuff for good, it's just that limits how much i can take was crossed since i was ten, and miles over , and i still have those troubles, i am aware that i am not functioning naturally anymore
21:55 kloofy: hence not suitable for such work, bye.
23:17 Lekensteyn: imirkin_: heh, I didn't pay attention/notice that part, that looks strange. Could it be a bad reading due to the bad link speed or is the hw/sw defect? (the report was from https://github.com/Bumblebee-Project/Bumblebee/issues/780)
23:17 imirkin_:blames bumblebee
23:23 Lekensteyn: Bumblebee has somehow become a point for questions related to hybrid graphics, but in this case it has nothing to do with the issue
23:23 imirkin_: he's using it
23:23 Lekensteyn: the error is reported when nouveau loads, and only much later bbswitch is loaded
23:24 imirkin_: ah
23:24 imirkin_: but i mean ... it's not turned off via bumblebee when nouveau loads?
23:24 imirkin_: and hasn't been?
23:24 imirkin_: ah indeed.
23:24 Lekensteyn: nope, bumblebee is no special magic, it relies on a driver for offloading and on bbswitch for pm
23:24 imirkin_: so there's another example of a GM108 user for whom vram was not properly initialized by the vbios
23:24 Lekensteyn: bumblebee itself cannot turn off the gfx
23:25 imirkin_: the recommendation i made for him was to use karol's branch and boot with NvClkMode=7
23:25 imirkin_: which forced nouveau to perform the reclocking sequence on start
23:25 imirkin_: which helped
23:25 imirkin_: however his issues appeared differently
23:26 imirkin_: vram=0 seems to indicate that we're getting a lot of 0's back
23:26 Lekensteyn: anything I could suggest to the reporter to gather more details that could help debugging this?
23:26 imirkin_: sorry, not sure
23:26 Lekensteyn: maybe I'll ask karol tomorrow