00:00 imirkin_: if feral ported hitman, edwin smith may be able to provide some steam keys to interested parties
00:00 karolherbst: feral did port it
00:02 imirkin_: and i ended up sending a patch for nve4_compute with a better description and hakzsam_ pushed it
00:03 karolherbst: :) awesome, thanks
00:03 imirkin_: (i tried to cc you, but your email bounced... probably used the wrong one)
00:04 karolherbst: yeah, currenty please use the mail I use on the ML, which is the gmail one
00:04 karolherbst: I still plan to reactive the old ones
00:04 imirkin_: ah ok
00:06 karolherbst: hum
00:07 karolherbst: MESA_GL_VERSION_OVERRIDE=4.50
00:07 karolherbst: error: context mismatch: expected OpenGL 1.0, but got OpenGL 9.0 core
00:07 karolherbst: I guess that parsing needs a bit of work maybe
00:07 imirkin_: 450
00:07 imirkin_: er
00:07 imirkin_: 4.5
00:08 karolherbst: I know
00:08 imirkin_: and MESA_GLSL_VERSION_OVERRIDE=450
00:08 karolherbst: I know
00:08 karolherbst: and I need to pass --core to apitrace
00:08 karolherbst: but mesa detected OpenGL 9.0 out of it
00:08 karolherbst: or "parsed"
00:08 imirkin_: makes snese
00:08 imirkin_: it parses %d.%d
00:08 imirkin_: and does Version = x*10+y
00:08 karolherbst: I know
00:08 imirkin_: which in the case of 4.50 becomes 40 + 50 :)
00:09 imirkin_: i.e. GL 9.0
00:10 imirkin_: but it won't help for the fact that alignment of various buffers is off
00:10 karolherbst: I guess with those both patches, nouveau should be good enough to run hitman pro :D
00:10 karolherbst: yeah
00:10 karolherbst: but the trace doesn't result in a crash anymore
00:12 imirkin_: coolio
00:13 karolherbst: I wanted to buy the game for myself anyway, but the 50% off sale is gone, so I wait until it's on sale again
04:37 imirkin_: dboyan: so yeah - without you combining the patches somehow, i have no real way of commenting on your impl
04:38 imirkin_: dboyan: i can't comment on random lines in a file on github
04:38 imirkin_: and the current way the patches are arranged don't lend themselves to easily reviewing them at the patch level
10:32 librin: Question on reclocking:
10:32 librin: [600591.781596] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
10:32 librin: [600591.781601] nouveau 0000:01:00.0: clk: error setting pstate 3: -22
10:32 librin: is this harmless or does this mean that either the GPU, the mem or both fail to change states and/or are running with too-low voltages?
10:33 karolherbst: that just means something is wrong, but nothing is broken
10:34 karolherbst: if nouveau can't set the right voltage it skips over setting the frequencies
10:34 karolherbst: librin: it would be good if you create a bug report + vbios
10:49 librin: karolherbst: roger that
10:50 librin: also, one more question:
10:51 librin: on after booting, on 07 pstate sensors show the GPU voltage as 0.85V
10:51 librin: if I push it to either 0a or 0f the voltage goes up to 1.04V
10:51 librin: but what is odd is that when I tune it down to 07 again
10:52 librin: the voltage drops to 0.82V instead of the initial 0.85V
10:52 librin: does this mean Nouveau ends up undervolting the GPU?
10:52 librin: at least, that's how it looks like
11:42 karolherbst: librin: everything is fine. If you don't set a pstate nouveau defaults to the boot values
11:42 karolherbst: those are sometimes not the same as the tables in the vbios tell the driver to use
11:52 librin: ah, okay
13:24 dboyan: imirkin: I agree with you. I'll do that today if I can finish the things on my hands in 2 hours. Otherwise, I'll do that tomorrow, hopefully.
13:43 librin: karolherbst: I seem to be unable to find any hints on how to do a vbios dump anywhere on the nouveau site
13:44 librin: would booting to windows (ewww!) and getting a vbios dump from there be OK for a bug report?
13:47 karolherbst: /sys/kernel/debug/dri/0/vbios.rom
13:49 librin: wow, that was easy
13:49 librin: karolherbst: thanks a bunch! :)
15:43 dboyan: imirkin: I managed to send out the patches before going to bed
15:43 dboyan:hopes he didn't make stupid mistakes while being sleepy
15:52 imirkin_: dboyan_: i saw, thanks. looking at them now.
15:59 imirkin: librin: out of curiousity, which GPU are you having reclocking troubles with?
16:00 karolherbst: librin: and by the way, did you create the bug report with the vbios attached?
16:36 karolherbst: send out my clocking stuff
16:36 karolherbst: *sent
16:36 imirkin: nice
17:48 librin: karolherbst: not yet, I had some business to take care of. I'm back home now, so I will
17:48 librin: imirkin: GK104
17:49 imirkin: sad.
17:49 librin: why so?
17:49 imirkin: because i was hoping that kepler reclocking was all good now
17:49 imirkin: at least on non-esoteric setups, like amazon EC2 :)
17:50 librin: the whole "factory overclocked" and "non-reference board design" might be something to do with it
17:50 librin: s/be/have/
17:52 librin: but it could also be my old kernel
17:54 imirkin: oh, well you should be on 4.10
17:54 imirkin: if you're not, then you should expect to have issues
17:55 imirkin: (unless you were using a branch that had karol's various patches)
17:55 librin: 4.9.9-gentoo
17:56 imirkin: ok, so no.
17:56 imirkin: then you should expect reclocking to be semi-broken
17:56 imirkin: only file the bug if it still doesn't work with 4.10
17:57 librin: okay
17:58 imirkin: (substantial fixes went into 4.10 for kepler reclocking, esp around voltage selection)
18:20 karolherbst: hum odd
18:20 karolherbst: ohh wait, yeah, you wanna test 4.10
19:13 moben: does that also have fixes for nva8 (nv50)? Still had some issues last time I tried on 4.8 or 4.9
19:19 imirkin: moben: i don't think so
19:20 karolherbst: mhh
19:20 karolherbst: it actually depends, but most likely not
19:20 karolherbst: moben: what issues do you have?
19:28 moben: last I tried it crashed when I tried to reclock. :(
19:29 moben: but it took a litte while which is better than it was on earlier kernels where it happened immediately
19:34 moben: ( https://bugs.freedesktop.org/show_bug.cgi?id=88415 )
19:40 karolherbst: huh
19:41 karolherbst: uhhh
19:41 karolherbst: moben: what error do you get while switching between 03 and 07?
19:41 karolherbst: is it by any chance volting related?
19:42 karolherbst: imirkin: "ID 0x3 Voltage entry 85" and guess what, there are 7 entries
19:43 karolherbst: ohh wait, there are 0 entries, cause it's a different table
19:43 karolherbst: 0x40 PM_Mode table version... insane
19:46 karolherbst: mhh, seems like that version is quite old though
19:46 karolherbst: okay, but I think I know what the issue is
19:51 imirkin: karolherbst: well, GT218 - not exactly new ;) makes sense to have an old version of the table.
19:52 moben: karolherbst: I can try again with 4.10 if you want me to
20:18 moben: I get "nouveau: 0000:01:00.0: fb: invalid/missing rammap entry" when going from the boot frequency to the lower one, called "03" (on 4.8.15)
20:19 karolherbst: imirkin: well, the table version is also used for maxwell
20:22 karolherbst: moben: well, that makes sense if "RamCFG 0xff"
20:22 karolherbst: ohh wait
20:22 karolherbst: rammap not ramcfg
20:23 karolherbst: mhh
20:25 moben: karolherbst: managed to dig up a log from a hang: https://paste.pound-python.org/raw/PuwnFcK3ZR1DS9o5J5Zu/
20:25 karolherbst: ok
20:25 karolherbst: memory got screwed up
20:29 imirkin: moben: actually i don't know if that's related to reclocking. that's just an error that lots of teslas get
20:29 imirkin: the push 400040 thing (or 406040)
20:30 imirkin: although the fact that we don't reclock and memory is in a not-fully-great state would definitely explain those errors quite nicely :)
20:32 moben: only happened after reclocking though :)
20:34 moben: as long as I don't try reclocking it's stable. But when I try I have to SysRq sooner or later
20:34 imirkin: =/
20:40 AndrewR: hello. apparently commit e027935a795ecf546f3e4abcc25655766f9615ac in mesa broke nv50 (nv92 more accurately). It started to segfault in glxgears and in wine
20:45 imirkin: i can believe that
20:45 imirkin: backtrace?
20:46 imirkin: i bet that ctx->rast is null or something?
20:46 imirkin: and is being used unchecked coz i figured "there's always a rast" or something
20:54 AndrewR: imirkin, moment, rebuilding mesa (but I wil need another rebuild because this one just with revert, and I disabled -debug in cflags)
20:59 moben: managed to get this error when running glxgears after reclocking: https://paste.pound-python.org/raw/P9My3OItE68QbfBSNz6y/
20:59 moben: still didn't hang yet
21:28 AndrewR: imirkin, http://paste.debian.net/918276/ ? (or I should recompile with O0?)
21:52 imirkin: AndrewR: that's probably enough. will take a look.
21:52 imirkin: as i suspected, nv50->rast->pipe.scissor is what dies :)