02:37 hrw: hello
02:38 hrw: got another card for my aarch64 experiments: old asus geforce gts250 one (loud as hell)
02:38 hrw: 01:00.0 VGA compatible controller [0300]: NVIDIA Corporation G92 [GeForce GTS 250] [10de:0615] (rev a2) (prog-if 00 [VGA controller])
02:38 hrw: Subsystem: ASUSTeK Computer Inc. Device [1043:8303]
02:39 hrw: works in desktop but once put in aarch64 machine nouveau module does not load
02:39 hrw: 11:36 root@pinkiepie-rawhide:hrw$ modprobe nouveau
02:39 hrw: modprobe: ERROR: could not insert 'nouveau': No such device
02:40 hrw: any ideas?
02:40 hrw: Linux pinkiepie 3.18.0-1.fc22.aarch64 #1 SMP Thu Dec 11 11:47:50 UTC 2014 aarch64 aarch64 aarch64 GNU/Linux
02:53 hrw: hm. maybe power issue
03:02 hrw: nope. connected 550W psu
03:02 hrw: anyway nothing with driver - probably pcie stuff
03:14 hrw: hmm...
03:14 hrw: [ 99.585085] nouveau E[ DEVICE][0000:01:00.0] unknown chipset, 0x140001e0
03:14 hrw: [ 99.591843] nouveau E[ DRM] failed to create 0x00000080, -22
03:14 hrw: [ 99.598073] nouveau: probe of 0000:01:00.0 failed with error -22
03:16 hrw: on x86-64 it was working:
03:16 hrw: sty 07 11:10:47 puchatek kernel: nouveau [ DEVICE][0000:01:00.0] BOOT0 : 0x092a80a2
03:16 hrw: sty 07 11:10:47 puchatek kernel: nouveau [ DEVICE][0000:01:00.0] Chipset: G92 (NV92)
03:16 hrw: sty 07 11:10:47 puchatek kernel: nouveau [ DEVICE][0000:01:00.0] Family : NV50
03:21 skeggsb_: hrw: my immediate guess would be that something else in the kernel is fucking up, can you post your full dmesg output?
03:24 hrw: skeggsb_: http://malenstwo.juszkiewicz.com.pl/~hrw/test/dmesg-nouveau-pinkiepie.txt
03:24 hrw: skeggsb_: I suspect pcie support
03:24 skeggsb_: [ 0.077381] pci 0000:01:00.0: BAR 0: no space for [mem size 0x01000000]
03:24 skeggsb_: [ 0.077384] pci 0000:01:00.0: BAR 0: failed to assign [mem size 0x01000000
03:24 skeggsb_: yes
03:24 skeggsb_: that's what i was expecting
03:27 hrw: cost of unfinished architecture ;d
03:27 skeggsb_: oh, i didn't scroll enough to see aarch64
03:28 skeggsb_: i had a bug filed against nouveau on the rh bugzilla for the same thing
03:29 hrw: will search for it
03:50 mlankhorst: yeah haven't had much luck with aarch64 either :P
03:50 mlankhorst: i tried..
03:50 hrw: mlankhorst: I am starting to harvesting pcie gfx cards from friends of friends and their coworkers ;D
03:51 hrw: mlankhorst: so far have only radeon hd5450 (oops on radeon load), radeon r7 240 (fails), geforce gt720 (not supported, sold), geforce gts250 (fails)
03:51 mlankhorst: you can get slightly further by adding the arch to ttm_io_prot..
03:52 mlankhorst: but haven't found a way yet to prevent a horrible death..
03:53 hrw: mlankhorst: will check 3.19-rc
03:53 mlankhorst: there was some other thing you would need to patch too, ,if you still get an error i can probably remember which..
03:53 mlankhorst: hrw: it's not part of 3.19-rc, need to add that ifdef manually
03:53 mlankhorst: but maybe 3.19 works
03:54 mlankhorst: I ended up with some external cpu abort error or something
03:55 hrw: mlankhorst: anyway fresher than 3.18 it would be
07:05 imirkin_: RSpliet: btw, looks like nv50 mad _does_ support the various neg flags in all instruction variants... should probably update that as well.
07:05 RSpliet: imirkin_: yeah, that was one of the updates I wanted to do before sending it to ML :)
07:05 RSpliet: the register compare function is wrong to, I think I'd rather inline a comparison
07:06 RSpliet: *too
07:06 imirkin_: not sure what you're talking about, but great :)
07:07 imirkin_: split stuff out though... so that when it breaks and people bisect, they bisect it to the exact issue
07:07 RSpliet: sure
07:07 imirkin_: (not that you'd _ever_ have bugs in your code...)
07:07 RSpliet: oh stop it!
07:08 imirkin_: but sometimes hw does unexpected things
08:23 tobijk: hey, somebody knows if it is allowed to mmess the stream layout with arb_gpu_shader5 like this: layout(stream = 1) out float stream1_0_out; layout(stream = 2) out float stream2_0_out; layout(stream = 1) out vec4 stream1_1_out; ?
08:24 tobijk: or do they need to be in order?
08:54 hrw: skeggsb_: updated board firmware and now gts250 gets memory assigned and nouveau sort-of loads: http://pastebin.com/UVrUrD1b
09:42 linas: So... with the latest kernel, 3.18.1, I get a gpu lockup during boot. I'm willing to debug that ... any recommendation on how to best do that?
09:47 buhman: serial console?
09:47 buhman: late kms?
09:47 linas: system comes up ... cnosole is usable
09:48 buhman: oh
09:48 linas: X gets hung up.
09:48 linas: I mean I can stick printfs in the kernel, but I'm not sure where to drill
09:48 linas: s/printf/printk/
09:49 buhman: there's no fun stuff in the X log or dmesg?
09:49 linas: welll . 3 seconds itno boot, I get "GPU lockup - switching to software fbcon"
09:50 linas: for X11, I have a dual-card setup. X tries to come up, displays on the card that did not lockup, stays blanked on the card that did lock up.
09:50 linas: and amazingly/confusingly, no errors in Xorg.0.log
09:51 linas: booting X a second time gives "E[Xorg[2920]] failed to idle channel 0xcccc0002"
09:51 buhman: did you tell X to use xinerama?
09:51 linas: yes!
09:52 linas: its a two-card, 3-monitor setup, on two old NV40 cards
09:52 buhman: the same card?
09:52 linas: xrandr is utterly borked by this setup, so xinerama is the only thing that works
09:52 buhman: xrandr doesn't support multiple gpus last I checked
09:52 linas: no, two different cards, NV46 and NV44A
09:53 buhman: do each of those cards work individually?
09:53 linas: the setup, and the cards, used to work great for years, on old ubuntu 12.04 (kernel 3.2.0)
09:53 buhman: I'd be curious if the bios initialized one or the other if that changes what happens
09:54 linas: Its sort-of-ish works on ubuntu 14.04 (kernel 3.13) but gets TTM out of space erors after a few days.
09:54 buhman: dmesg should probably say somewhere what nouveau did
09:54 linas: I was trying 3.18.1 to see if the TTM errors go away, but instead, get the GPU lockup
09:55 buhman: I think you should bisect
09:55 buhman: and file a bug report
09:55 linas: Yeah, I filed umpteen bug reports
09:55 buhman: with a bisect?
09:55 linas: no one pays attention to them
09:56 linas: I'm not sure II have time to mess with bisect
09:56 buhman: not sure the developers have time to mess with your bug report ;p
09:56 mlankhorst: we care, especially about regressions
09:56 linas: well, exactly
09:56 linas: reports. rplural
09:56 linas: multiple issues
09:56 linas: so I thought I'd debug myself,
09:56 buhman: linas: you asked: you should bisect.
09:56 linas: which is why I'm here :-)
09:58 linas: OK, I'll try bisecting, but this is my main production desktop, and I work on deadlines, .. may take a little while.
09:59 linas: after bisecting, file a bug report at freedeskop.org, and chat here, then?
13:47 yurikoles: imirkin: are you here?
14:47 Eliasvan: @pmoreau: Hi, I've got some very fresh piglit tests for GTX760, but I can't contact @imirkin, is it possible for you to pass my results to imirkin (I don't have his email)
14:54 tobijk: Eliasvan: just wait a bit and he'll be here chatting :)
14:55 Eliasvan: ok ;)
14:57 tobijk: btw, how is the overall performance? crashes/fails/passes?
14:58 Eliasvan: Summary: 20412/20644
14:58 tobijk: oh thats fine :)
14:59 Eliasvan: pass 20637 skip 4546 fail 209 dmesg-warn 23 warn 6 dmesg-fail 4 crash 3 timeout 0 total 25428
15:00 tobijk: name me the crashes pls, maybe i'l ltry to fix them :D
15:01 Eliasvan: arb_gpu_shader5-xfb-streams-without-invocations; eglTerminate then unbind context; EGL_MESA_configless_context/basic
15:03 tobijk: can you run the last one and see if it sometimes passes?
15:04 tobijk: -> ./bin/egl-configless-context -auto -fbo
15:07 Eliasvan: ok, it will take a while, it's on another PC, but maybe I'll first say what the errors are:
15:07 Eliasvan: All crash info on "EGL_MESA_configless_context/basic":
15:07 Eliasvan: Returncode: -6
15:07 Eliasvan: warning: couldn't find a 16-bit config with no depth buffer
15:07 Eliasvan: egl-configless-context: /home/archie/archlive-nouveau/pkg/piglit-git/src/piglit/tests/egl/egl-configless-context.c:278: run_tests: Assertion `value == 0x0405' failed.
15:07 Eliasvan: |
15:07 Eliasvan: All crash info on "arb_gpu_shader5-xfb-streams-without-invocations":
15:07 Eliasvan: Returncode: -11
15:07 Eliasvan: |
15:07 Eliasvan: All crash info on "eglTerminate then unbind context":
15:08 Eliasvan: Returncode: -11
15:08 Eliasvan: |
15:08 Eliasvan: Done.
15:08 tobijk: ke :)
15:08 Eliasvan: so should I go and retest that test a couple times?
15:09 tobijk: no actuall if you hit that assert, its useless to test
15:09 Eliasvan: ah, k :)
15:10 Eliasvan: is that assertion normal to occur?
15:10 tobijk: it should not
15:10 tobijk: that one is passing on my machine
15:11 Eliasvan: because it also fails on "http://people.freedesktop.org/~imirkin/nvc0-comparison/nvcf-2014-11-23-mupuf-2/spec/EGL_MESA_configless_context/basic.html"
15:11 Eliasvan: which is from 2014-11-23
15:11 tobijk: mh maybe its ust fixed with latest mesa
15:12 Eliasvan: Ive got latest mesa, up to a couple hours ago
15:12 tobijk: ah allright
15:13 Eliasvan: mesa-git-10.5.0_devel.67177
15:14 Eliasvan: mesa/refs/heads/master = 426fd535d9e2812c52dd709f8e5532bae501b3f6
15:15 tobijk: yeah thats recent enough, dont worry
15:27 Eliasvan: tobijk: do you mind if I give you the link to my tests? Because I've got to sleep right now ;) Or should I contact him another time?
15:28 tobijk: no give em to me
15:28 Eliasvan: Here's the link, if someone's interested: http://www.megafileupload.com/en/file/595981/piglit-results-all-tar-xz.html
15:28 tobijk: i'll forward them to imirkin
15:28 Eliasvan: Thanks tobijk!
15:28 tobijk: np :)
15:29 Eliasvan: Oh yeah, I had also a question:
15:29 Eliasvan: are piglit OpenCL test-results also welcome?
15:29 tobijk: well they are a bit useless right now with nouveau (i guess all will fail :/)
15:29 Eliasvan: really?
15:29 tobijk: but any testing is always welcome
15:30 Eliasvan: someone I know said that he was able to run Blender cycles with nouveau
15:30 Eliasvan: is that possible?
15:30 tobijk: dont know what blender is using, sorry
15:31 Eliasvan: it uses OpenCL for raytracing
15:31 tobijk: maybe it has a fallback cpu backend
15:31 tobijk: dunno
15:31 Eliasvan: yes, it has, but he said it was using the GPU
15:32 tobijk: oh well, maybe i got old info :)
15:32 Eliasvan: weird, because with Beignet I'm not able to use Blender Cycles (OpenCL)
15:32 Eliasvan: (that's on Intel)
15:34 Eliasvan: ok, thanks for all help, see you!
15:34 tobijk: bye!
19:01 linas: blender is the primary linux 3d authoring platform. Animators use it.