04:43 jam3s2002: hi. I'm not sure this is the right place to ask, but I'm trying to figure out if my card is compatible. I guess first question is whether or not quadro cards work, and 2nd question is whether or not support for volta has been added... Trying to get a quadro gv100 working in gentoo, but having a difficult time finding documentation
04:45 HdkR: Quadro works. Volta+ doesn't
04:45 HdkR: Volta and Turing support is being worked on but it is no where near the state of being able to be used
04:50 jam3s2002: thanks for the assistance!
10:27 mupuf: imirkin, abderrahim[m]: the metrics collection about the busyness of engines can be queried from the PMU
10:27 mupuf:has written code for this, and karol worked with it too
10:28 mupuf: but it somewhat got stalled when the PMU became more locked down :s
10:30 mupuf:is trying to find the code
10:31 mupuf: https://cgit.freedesktop.org/~mperes/nouveau/log/?h=ppwr_rework
10:31 mupuf: some of the code has landed, IIRC
13:42 abderrahim[m]: Any tricks for analysing mmiotrace output? any ready-made scripts?
13:42 abderrahim[m]: or is awk my friend?
13:43 abderrahim[m]: (I already used demmio)
14:01 abderrahim[m]: So here is what the blob does https://paste.gnome.org/pvxqyzxzw
14:01 abderrahim[m]: (I took everything between two consecutive SLEEP lines)
14:04 mupuf: abderrahim[m]: what are you looking at?
14:04 mupuf: demmio is the goto tool to analyze mmiotraces, no idea what else you would want from it
14:05 abderrahim[m]: mupuf: I was looking at how to compare two traces to isolate the difference. I managed to do that now
14:05 mupuf: ah, there are scripts for this IIRC
14:06 mupuf: well, just get rid of the time, and you should have diffable traces
14:06 mupuf: I just had a way to diff peeks
14:07 mupuf: it's in the vbios repository
14:07 abderrahim[m]: now, I'm trying to figure out the difference between the trace I pasted above and nvidia_backlight from the kernel
14:07 abderrahim[m]: mupuf: thanks
14:09 mupuf: [0] 406.164207 MMIO32 W 0x61c084 0xc0002e9e PDISPLAY.SOR[0].PWM_CTL <= { DUTY = 0x2e9e | CLOCK_SELECT = CRYSTAL | TRIGGER } --> that's the line oyu are looking for
14:10 mupuf: Do you know what PWM is?
14:10 mupuf: the trickiest part is setting the clock tree properly to compute the right div, and then the right duty
14:11 mupuf: in this case, yo uare lucky, it is using the crystal as input (27MHz)
14:12 abderrahim[m]: I found that line, but honestly I don't know what PWM is
14:12 mupuf: it looks like the value set for DIV is so that oyu would get a 1kHz frequency
14:12 mupuf: 27MHz/0x679a ~= 1 kHz
14:12 mupuf: https://www.arduino.cc/en/Tutorial/PWM
14:12 RSpliet: abderrahim[m]: Pulse Width Modulation. It's a fairly simple concept in the sense that even I grasp it :-P
14:13 abderrahim[m]: RSpliet: :)
14:13 mupuf: as for the sequence needed to change the PWM values, you first need to write to the DIV register
14:14 mupuf: then to the duty, but you add this TRIGGER bit which commits the changes you wrote to the actual HW (the register's values are double buffered)
14:15 RSpliet: Generally to create those block waves as you see in mupuf's link, you use a clock that is way higher than the target frequency of the block wave, then you count clock cycles in the fast clock for on and off
14:15 mupuf: I suggest you try using nvapoke to experiment with this :)
14:16 mupuf: RSpliet: oh yeah, thanks for stating this
14:21 sigod: anyone know if support for video acceleration on a gtx660 has improved ie without the need for nvidia firmware?
14:51 gnarface: sigod: afaik it still doesn't work at all
14:52 gnarface: oh wait, i'm thinking of a different card. it might work, nevermind.
15:07 mupuf: sigod: without nvidia's firmwares,m I think you might be dreaming ;)
17:40 imirkin: HdkR: fyi, volta + turing modesetting should be fine, 3d backend for volta is in progress afaik, dunno about turing.
17:41 imirkin: sigod: if you're talking about context switching for the problematic GTX660's, i don't think any work has been done
18:27 abderrahim[m]: It seems the code in nvidia_backlight is correct. So it is probably not getting called
18:27 abderrahim[m]:has set the brightness to a comfortable value using nvapoke
18:28 abderrahim[m]: any pointer on how to debug it? is there a kind of printf I can use in the kernel?
18:29 imirkin: iirc there's a lot of confusion about these things
18:29 imirkin: since there's acpi backlight
18:29 imirkin: and it's not always clear which one to use
18:29 imirkin: mind pastebinning your dmesg?
18:30 imirkin: [also i assume you're talking about nouveau_backlight.c?]
18:31 abderrahim[m]: yeah, nouveau_backlight
18:32 imirkin: in the dmesg it should indicate which backlight provider is used iirc
18:34 abderrahim[m]: there is no mention of "backlight" in dmesg
18:34 imirkin: how about acpi_video
18:35 imirkin: ls /sys/class/backlight/
18:36 abderrahim[m]: acpi_video0 nv_backlight
18:36 abderrahim[m]: nothing in the dmesg though
18:36 abderrahim[m]: https://paste.gnome.org/pnxcquhfg
18:36 imirkin: go into the nv_backlight dir
18:36 imirkin: there should be some obvious brightness control file in there
18:38 imirkin: see if writing to it has any effect
18:38 abderrahim[m]: it works indeed
18:38 imirkin: ok, hold on
18:38 imirkin: try booting with
18:39 imirkin: acpi_backlight=vendor
18:39 abderrahim[m]: ok
18:44 imirkin: if that helps, file a kernel bug with your system specifics
18:44 abderrahim[m]: Yes, it works
18:44 imirkin: there's various logic to detect which one to use
18:45 imirkin: based on a variety of heuristics
18:46 abderrahim[m]: bbl
20:25 imirkin: abderrahim[m]: there are a bunch of quirks that push the logic one way or another -- which laptop do you have?
20:38 abderrahim[m]: A sony vaio (model number VPCEH3U1E)
21:17 imirkin: no quirks about those...
21:18 imirkin: abderrahim[m]: have a look at this comment by hansg: https://bugzilla.redhat.com/show_bug.cgi?id=1163574#c14
21:24 abderrahim[m]: Will try to do it tomorrow
21:24 abderrahim[m]: Good night everyone