13:42glisse: anybody seen acr loading failure on 1080 ?
14:01karolherbst: glisse: donnu, it doesn't load relible on my 1050
14:01karolherbst: glisse: it's the gr init time out thing you see, right?
14:01karolherbst: glisse: that patch might help, mind trying it out? https://github.com/karolherbst/linux/commit/ecb1d072ebdb5538e5b4fd5210c4173dd7f33719
16:05glisse: karolherbst: just someone at lsf that has that problem, it always work for me so i was kind of surprise for pascal
16:05glisse: karolherbst: i will point out your patch
19:10jamincollins: I believe I've followed the nouveau Optimus documentation, but I can't seem to get DRI_PRIME=1 working for either DRI2 or DRI3: https://gist.github.com/jamincollins/d80bbf5ddae1074c47051ba9e63caf1d
19:11jamincollins: also, there are long (multi-second) delays when the X session initially starts up
19:12gnarface: did you check the xorg log to see if it's even loading the nouveau driver first? i thought that long delay comes from it cycling through framebuffer drivers
19:13jamincollins: if it wasn't loading the driver I don't believe I'd have it listed in the `xrandr --listproviders` output
19:13jamincollins: [ 1126.320] (--) NOUVEAU(G0): Chipset: "NVIDIA NV137"
19:13gnarface: maybe so, i dunno
19:14gnarface: though the delay could be caused by that even if it settles on the right one eventually
19:14jamincollins: that was my experience before getting to this point, in order to even get this far, I had to create an xorg config stub telling it to load the nouveau driver
19:15karolherbst: jamincollins: with dri3 you won't have to do any of the xrandr calls
19:15jamincollins: that would be great, but neither of them work =)
19:15karolherbst: is nouveau loaded at all?
19:16karolherbst: "nvc0_screen_create:1007 - Error allocating PGRAPH context for M2MF: -16"
19:16karolherbst: okay, so something fails hard
19:16karolherbst: dmesg should contain more information
19:16karolherbst: I think I know what it is, but I rather want to check to be sure
19:18karolherbst: jamincollins: you might need to set nouveau.runpm=0 _and_ need this kernel patch: https://github.com/karolherbst/linux/commit/ecb1d072ebdb5538e5b4fd5210c4173dd7f33719
19:18karolherbst: maybe only one of them is required
19:18karolherbst: dunno if runpm works on your machine
19:18karolherbst: dunno if secboot fails
19:18karolherbst: dmesg should tell us
19:18jamincollins: [ 2652.137590] nouveau 0000:01:00.0: timeout
19:18jamincollins: [ 2652.137638] WARNING: CPU: 8 PID: 4474 at drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgf100.c:207 gf100_vmm_flush_+0x17b/0x190 [nouveau]
19:19karolherbst: mind uploading your full dmesg somewhere?
19:19jamincollins: working on it, it's HUGE
19:23karolherbst: "dmesg | xclip -selection clipboard" helps :)
19:23jamincollins: $ dmesg | pastebinit
19:23karolherbst: or that
19:23karolherbst: anyway, I am more interested in what errors we have before the stacks
19:24jamincollins: I'll reboot and brb with a fresh state
19:27jamincollins: that's as fresh as I can get it
19:28jamincollins: still seems to be missing 12 seconds
19:28karolherbst: mhhh .. :/
19:29jamincollins: looking in journalctl now
19:29karolherbst: jamincollins: journalctl --boot 0 --dmesg
19:29karolherbst: ahh.. yeah
19:30karolherbst: cool, that looks better
19:30jamincollins: Apr 30 13:25:03 odin kernel: nouveau 0000:01:00.0: fifo: fault 00 [READ] at 0000000000000000 engine 1f [PHYSICAL] client 07 [GPC0/T1_2]>
19:31karolherbst: jamincollins: okay.. it's the secboot issue
19:31karolherbst: jamincollins: mind compiling a kernel with the patch applied I linked above?
19:31jamincollins: it's been a long while, but I can give it a shot
19:31karolherbst: I think it shouldn't be too hard with arch
19:32jamincollins: and you'll need to relink me, rebooted between then and now
19:33karolherbst: jamincollins: you might want to test that with nouveau.runpm=0 set as well to rule out suspending the GPU messes up things later :/
19:33karolherbst: kind of working on getting that fixed as well
19:34karolherbst: this gf100_gr_init failing issue kind of only happens on a few random GPUs... no idea what's wrong there as nvidia literally prevents us from debuggin this on a hardware level
19:34jamincollins: why make it easy?
19:34karolherbst: well, signed firmware and such
19:35karolherbst: we essentially boot some engines with signed firmware and while the engine is booting, we can't look into the state
19:35karolherbst: essentially it's a black box :)
19:35karolherbst: and then it stops because $something failed
19:35karolherbst: and of course, because it stopped, we can't look into the state as well
19:35karolherbst: it's super annoying
19:35karolherbst: anyway, that patch fixes this on my machine as well
19:36jamincollins: attempting a kernel build
19:36gnarface: i think you guys should try hiring a psychic to help guess what's going on in those states
19:36gnarface: nothing else works, might as well try everything, right?
19:36karolherbst: gnarface: wouldn't come up with worse guesses at least
19:37gnarface: you never know. sometimes psychics are good at guessing stuff :)
19:37karolherbst: that's like 99% of their skills, yes
19:37karolherbst: good guessing :p
19:37gnarface: it would be pretty funny if that worked
19:38karolherbst: I'd rather try AI enforced reverse engineering :p
19:38gnarface: "we couldn't reverse engineer anything past this point, so we hired some professional guessers..."
19:38gnarface: i can picture the documentary now
19:38karolherbst: too cheap
19:39karolherbst: but actually, AI should be able to come up with a working GL driver...
19:40karolherbst: apitrace in, commands out
19:40karolherbst: just.. that none of the current systems would be fast enough for it
19:40karolherbst: would probably take millions of years
19:44jamincollins: so, karolherbst, which line in the dmesg output cinched it for you?
19:45karolherbst: jamincollins: nouveau 0000:01:00.0: gr: init failed, -16
19:46karolherbst: and the absence of other issues
19:47jamincollins: the wait for the X session to actually start is long enough I gave up on it the first several times
19:48jamincollins: got lucky (I guess) and was distracted while waiting for one of the attempted boots
20:25jamincollins: karolherbst: installing now
20:29jamincollins: looks better in terms of boot time
20:30jamincollins: and DRI works
20:31jamincollins: *and* my external display works =)
20:32jamincollins: I'm in a little slice of heaven right now
20:33jamincollins: karolherbst: THANK YOU!
20:40karolherbst: jamincollins: we might be able to merge that commit at some point, the nouveau maintainer asked me to get more user testing on that one
20:40karolherbst: or rather proof it helps
20:52jamincollins: karolherbst: I'm happy to note such on the report
20:53jamincollins: and I updated the github gist with a link to the patch
21:02jamincollins: am I right in thinking that with nouveau and Optimus that the nvidia chip _should_ power down at some point?