03:42 kreyren: > Mar 12 04:26:03 dreamon kernel: [ 7.344433] nouveau 0000:01:00.0: pmu: firmware unavailable
03:42 kreyren: What kind of firmware does it expect? x.x
03:42 kreyren: PMU = Power Management Unit?
03:42 HdkR: aye
03:42 HdkR: Controls reclocking
03:42 kreyren: So that's the PMU on the GPU right
03:43 HdkR: yes
03:43 imirkin: also fans, etc
03:44 kreyren:is searching the web for the PMU chip as the GPU is painful to disassemble
03:44 imirkin: it's a falcon cpu
03:44 imirkin: which is an ISA that nvidia created
03:44 imirkin: although i guess more recent ones are ARM or RISC-V or something
03:45 HdkR: riscv
03:45 kreyren: So it's not solvable? This is a Maxwell GPU
03:45 imirkin: maxwell would have been falcon
03:45 kreyren:is confused
03:45 imirkin: not sure what 'solvable' means
03:45 kreyren: being able to provide the firmware for nouveau to use it?
03:45 imirkin: of course it's solvable
03:46 kreyren: how O.o
03:46 imirkin: just call up nvidia and tell them to either sign our firmware, or to make full-featured firmware with signatures redistributable.
03:46 imirkin: piece o' cake.
03:46 imirkin: 1-800-NVIDIA :)
03:46 imirkin: (yes, i know it's one digit short)
03:47 kreyren: o.o
03:47 kreyren: Can't i just make it to use unsigned
03:47 imirkin: sure
03:47 imirkin: but then it can't change fan speeds
03:47 kreyren: x.x
03:47 kreyren: so what should i do
03:48 kreyren: other then reverse engineering the VBIOS
03:48 HdkR: reverse engineering doesn't solve the signing problem :)
03:48 kreyren: O.o
03:48 imirkin: how would that help?
03:48 kreyren: seems to work for BIOS based on my experience with coreboot O.o
03:49 imirkin: you could figure out how to extract the full-featured firmware from the blob drivers
03:49 imirkin: and feed it to nouveau
03:49 imirkin: and then adapt nouveau code to implement whatever ABI such firmware implements
03:49 kreyren: and what are the implications of it not having the firmware?
03:50 imirkin: none at all? i think you don't get accel
03:50 kreyren: i have accel
03:50 imirkin: if you're using the nvidia-supplied firmware from linux-firmware, it runs all this stuff internally without any OS control
03:50 imirkin: and doesn't support reclocking
03:50 imirkin: they made shitty firmware just for nouveau's benefit
03:50 kreyren: but it reclocks using nouveau.NvBoost=1 O.o
03:51 imirkin: which GPU do you have?
03:51 kreyren: or at least it seems that the performance has improved after that
03:51 kreyren: GTX970
03:51 imirkin: you're just dreaming
03:51 kreyren: O.o
03:51 imirkin: among other things, nouveau.NvBoost isn't a thing
03:51 imirkin: nouveau.config=NvBoost=1 is a thing
03:51 kreyren: ah i see x.x
03:51 imirkin: which enables slightly altered reclocking logic for kepler
03:52 imirkin: based on what voltage levels it allows itself to use
03:52 imirkin: now here's the funny part
03:52 imirkin: GM20x can reclock just fine using the logic and firmware we already have
03:52 imirkin: the ONLY thing our firmware can't do is control the fans.
03:53 kreyren: so if i connect the fans to the mobo power as CPU fan
03:53 imirkin: because that's only accessible from the high-secure modes
03:53 imirkin: there's some set of instructions somewhere for how to make this go
03:53 kreyren: or like i can do jumper wiring
03:53 imirkin: karolherbst would be the person to talk to
03:53 imirkin: pretty sure at least one person took off the "official" fans, and put on their own
03:53 imirkin: and it worked fine
03:54 kreyren: like the fan unit itself?
03:54 imirkin: yea
03:54 kreyren: the fans on the GPU have firmware on them?
03:54 imirkin: attached to mobo or whatever
03:54 imirkin: no
03:54 kreyren: o.o
03:54 kreyren: ah i see
03:54 imirkin: but their speed is controlled by the PMU firmware
03:54 imirkin: and the PMU firmware can only do that if it's signed
03:54 kreyren: well the mobo has high airflow and it so far doesn't overheat though
03:54 imirkin: otherwise you get PERMISSION DENIED
03:56 imirkin: (actually i forget precisely what happens ... but end result is that we can't control the fans)
03:56 kreyren: i see O.o
03:57 imirkin: kherbst: --^
03:57 imirkin: i think he has a kernel branch for this somewhere
03:57 imirkin: let me see if i can track it down..
03:58 kreyren: o.o
03:59 imirkin: oh wait. did we upstream this?
03:59 imirkin: ah no. we did not.
04:00 imirkin: kreyren: https://github.com/karolherbst/nouveau/commits/clk_to_upstream
04:00 imirkin: this is what i found
04:00 imirkin: it adds a NvFanless setting, which if you set will enable some reclocking goodness
04:00 imirkin: this looks like a newer one: https://github.com/karolherbst/nouveau/commits/clk_cleanup_for_real
04:01 kreyren:started tracking on https://github.com/Kreyren/kreyren/issues/74
04:03 kreyren: imirkin: will research thanks ^-^
04:09 imirkin: (note that this only applies to GM20x ... pascal has a different memory controller, so all bets are off)
04:14 kreyren: <imirkin "(note that this only applies to "> noted
04:14 kreyren: i also have issues with this https://github.com/Kreyren/kreyren/issues/75
04:14 kreyren: which crashes the system
04:17 kreyren: or at least seems to cause a kernel panic as i get thrown back to TTY with this all over the screen
04:30 damo22: blow air? EPERM
04:31 damo22: :P
04:34 damo22: kreyren: are you using libre kernel? i thought linux-firmware has firmware blob for PMU ?
04:35 kreyren: <damo22 "Krey: are you using libre kernel"> debian provided kernel with linux-firmware
04:35 kreyren: note this being devuan though
04:35 damo22: i might be wrong, im just guessing
04:35 damo22: but i have stuff in /lib/firmware/nvidia
04:36 kreyren: < https://matrix.org/_matrix/media/r0/download/matrix.org/LQjPBoDJPaSrBGUCOAqsWFCq/message.txt >
04:37 damo22: yeah that stuff
04:37 damo22: theres a bunch of blobs in each folder per model
04:38 damo22: does that allow the pmu to set the fan speed?
04:38 kreyren: no idea x.x
04:40 damo22: maybe someone who knows will chime in
04:41 kreyren: or like how could i check
05:46 imirkin: damo22: it manages the fan, but does not allow the OS to set it explicitly
09:31 kherbst: imirkin: soo.. the RS232 console just works :)
14:22 imirkin: karolherbst: woohoo!
18:33 imirkin: pmoreau: hey, were you planning on reviewing the remainder of that compute MR? (mostly just emission changes left)
18:33 imirkin: pmoreau: also i updated the nv50_compute branch with my latest fixes and cleanups
18:35 nzamb1: Hi All, I have an issue with nouveau driver on Ubuntu 20.04. Is this a right place to ask for a help?
18:35 imirkin: nzamb1: yes.
18:36 nzamb1: my problem is that Xorg is failing with core every day. ANd it looks like because of the driver. Here is the error message from syslog:
18:36 nzamb1: Mar 12 13:49:49 k-workstation kernel: [18218.835290] nouveau 0000:28:00.0: Direct firmware load for nouveau/nv84_xuc00f failed with error -2
18:36 nzamb1: Mar 12 13:49:49 k-workstation kernel: [18218.835295] nouveau 0000:28:00.0: vp: unable to load firmware nouveau/nv84_xuc00f
18:36 nzamb1: Mar 12 13:49:49 k-workstation kernel: [18218.835296] nouveau 0000:28:00.0: vp: init failed, -2
18:36 nzamb1: Mar 12 13:49:49 k-workstation kernel: [18218.835322] nouveau 0000:28:00.0: Direct firmware load for nouveau/nv84_xuc103 failed with error -2
18:36 nzamb1: Mar 12 13:49:49 k-workstation kernel: [18218.835323] nouveau 0000:28:00.0: bsp: unable to load firmware nouveau/nv84_xuc103
18:36 nzamb1: Mar 12 13:49:49 k-workstation kernel: [18218.835324] nouveau 0000:28:00.0: bsp: init failed, -2
18:37 imirkin: that's not a problem
18:37 imirkin: also, please use pastebin for larger pastes
18:37 imirkin: (more than a couple lines)