07:50 coredump: so, I am getting chipset unknow when loading the driver. During boot I can see dmesg messages of it correctly identifying it as a GP107, but when I try to load the module it just fails
07:50 coredump: are 1050Ms not supported or am I missing a firmware or something?
07:50 coredump: Using Manjaro linux with 4.17 kernl
09:07 crmlt: Hello
09:07 crmlt: I'm unable to start X on Ubuntu 18.10 with nouveau driver.
09:08 crmlt: I always stuck on black screen with mouse cursor.
09:12 crmlt: It doesn't work even with Live CD session.
09:13 gnarface: try it without compositing
09:22 crmlt: gnarface: How?
09:22 gnarface: sorry dunno
09:22 gnarface: i think it's gnome by default?
09:23 crmlt: gnarface: i cant even reach GDM or SDDM
09:23 crmlt: same issue for kde, gnome etc.
09:23 gnarface: tried blackbox?
09:23 gnarface: i think kde and gnome both have it on by default
09:24 crmlt: no i havent
09:25 crmlt: i had't this issue in 17.10 even on wayland session
09:25 gnarface: oh? interesting...
09:26 crmlt: yeah nouveau was always working quite good
09:26 crmlt: there is something strange
09:42 RSpliet: crmlt: unless you've installed the upstream 4.16 kernel and Mesa 18.0.3, you're going to have to talk to the ubuntu people instead. "We" don't do distribution support because it often involves trying to chase bugs that have already been fixed... nouveau is too understaffed and complex for that kind of debugging.
09:43 RSpliet: Luckily, that also means that if you do install the upstream tool chain (which probably involves building yourself, but there might be a magic ubuntu way of installing them if someone else bothered making a package out of it) and problems persist we're more than happy to try and help ;-)
09:44 RSpliet: *building it yourself
14:41 mardikene: maybe i should apalogise infront of imirkin, but only if he is smart enough to understand how percentage calculation works, than one would really know the trick long since, and indeed i am the troll, but basically any gpu can render with 1.5ogl spec as fast as human eye would allready not spot the stuff, pointless to talk about more difficult wavefront or reg fragmenting/partitioning on long latency ops
14:42 mardikene: cause even this is not needed, it will go perfectly very fast without this algorithm too
14:56 karolherbst: imirkin: I've got a gsync display now :)
15:03 imirkin_: well aren't you mr fancy
15:03 imirkin_: maybe you can make hdmi 2.0 work while you're at it?
15:39 karolherbst: imirkin_: dunno if that thing has HDMI 2.0 though
15:40 imirkin_: =/
15:40 karolherbst: ohh, should have
15:40 karolherbst: maxwell2 has HDMI 2.0, right?
15:42 coredump: so, at the risk of annoying everyone with a repeat question: 1050M with kernel 4.17 still gives me 'unkown chipset', is that expected or am I missing some recent change?
15:42 karolherbst: so I guess I could look into hdmi 2.0 first or something
15:42 coredump: mesa 18.0.3
15:43 imirkin_: karolherbst: yes, it does.
15:43 imirkin_: coredump: the gpu isn't powered on
15:44 imirkin_: so we read all 1's
15:44 imirkin_: the unknown chipset is 0xfffffff right?
15:49 coredump: yeah that makes sense
15:49 coredump: let me remove bbswitch and reboot
15:49 coredump: because D3 state yo
15:52 imirkin_: oh yeah, bbswitch + nouveau is not a good combo unless you're very careful
16:03 coredump: good news
16:04 coredump: the driver finds something
16:04 coredump: bad news; nouveau: timeout and CPUX stuck for 22s
16:04 coredump: imirkin_: yeah I should have realized that before (the bbswitch part)
16:05 coredump: Love the ubuntu forums because the post about nouveau locking up is replied by people telling them to install the proprietary drivers.
16:05 coredump: dudes, that's not the answer
16:10 imirkin_: the actual answer is get rid of the nvidia gpu and get an amd one :)
16:10 imirkin_: please fix the forum answers ;)
16:10 imirkin_: anyways, if this is a GP107, then karol had some trouble with bringup as well
16:11 imirkin_: karolherbst: where's your patch to bounce it?
16:11 karolherbst: https://github.com/karolherbst/nouveau/commit/67c57185d716792f806b73f24bf3826040b217a1
16:12 imirkin_: coredump: --^
16:12 coredump: does it make a difference if I am using the driver on early stages or not?
16:12 coredump: like on my initcpio image
16:12 imirkin_: shouldn't
16:13 imirkin_: as long as you don't mess with it first
16:13 imirkin_: with bbswitch or whatnot
16:13 coredump: removed bbswitch so that should be fine.
16:13 karolherbst: imirkin_: I am sure it won't work when the GPU gets suspended
16:13 coredump: I am loading i915 and nouveau on init
16:13 karolherbst: because suspening is just inherently broken for D3cold
16:13 imirkin_: but without nouveau, GPU won't get suspended...
16:14 imirkin_: karolherbst: was your issue a timeout on load?
16:14 karolherbst: the device doesn't get back
16:14 imirkin_: but we time out eventually
16:14 karolherbst: I talk about D3cold
16:14 karolherbst: not D3hot
16:14 karolherbst: with D3hot we timeout, yes
16:14 karolherbst: some secboot screw up
16:15 karolherbst: but if the firmware enables D3cold support, the device just doesn't get back at all
16:15 imirkin_: k
16:15 imirkin_: coredump: let's hope you're in the D3hot case ;)
16:15 karolherbst: doubtful
16:15 karolherbst: all win8+ laptops should do D3cold
16:15 imirkin_: why did it help you then?
16:15 karolherbst: on some hw it seems to work
16:15 karolherbst: uhm
16:16 karolherbst: I remove it from the pci subsystem
16:16 karolherbst: do ACPI calls
16:16 imirkin_: lol ok
16:16 karolherbst: and rescan after enabling the GPU again
16:16 imirkin_: not exactly seamless.
16:16 karolherbst: :)
16:16 karolherbst: not really
16:16 karolherbst: but
16:16 karolherbst: my patch helps here
16:16 karolherbst: and if the GPU doesn't get suspended
16:16 karolherbst: so that at least secboot works "better"
16:18 coredump: I think my secboot option is off on the bios
16:18 coredump: if that makes a diffrence
16:18 imirkin_: different secboot
16:18 coredump: ah
16:18 imirkin_: with something so great, you should have many ;)
16:19 coredump: so before I try ti compile stuff and have hours of fun debugging this: is there any 3D support on the drivers for pascal? As in will I be able to use it for a 3D app or games?
16:19 coredump: because that's the main reason to do this atm
16:21 karolherbst: I have a 1050
16:22 karolherbst: and besides those secboot and suspend/resume issues it works fine otherwise
16:22 karolherbst: more or less
16:22 coredump: https://gist.github.com/coredump/3583bf5baf407ae32cdc2be7a864780f
16:22 coredump: does this look like the issue you solved?
16:22 coredump: mine is a 1050M on a notebook, so that may be a factor too
16:22 karolherbst: coredump: yeah..
16:22 karolherbst: you can't suspend the GPU if you plan to use it
16:22 karolherbst: not with nouveau
16:22 karolherbst: not with bbswitch
16:22 coredump: I wonder what is suspending it
16:22 coredump: because I removed the bbswitch
16:23 karolherbst: nouveau
16:23 karolherbst: or the kernel
16:23 karolherbst: or both
16:23 karolherbst: hard to tell
16:23 karolherbst: nouveau.runpm=0
16:23 coredump: lmc that
16:23 karolherbst: should disable it
16:35 coredump: well, one step closer with that option on the module.
16:35 coredump: loaded fine, was able to see the --listproviders list
16:35 coredump: but DRI_PRIME=1 glxinfo caused it to lock up again
16:36 imirkin_: coredump: support on pascal is quite iffy
16:36 imirkin_: due to lack of reclocking
16:37 coredump: karolherbst: is that patch merged upstream or should I patch it manually to give it a try
16:37 karolherbst: manually
16:47 coredump: ok, let's try that.
16:55 coredump: oof, that's on the kernel side
16:55 coredump: bleeergh
17:59 coredump: welp, this will have to wait
17:59 coredump: will try again in some months
17:59 coredump: maybe it makes it to main
18:01 imirkin_: or maybe you'll get an amd gpu :)
18:04 coredump: bahaha
18:04 coredump: it's on the notebook
18:04 coredump: not easy to replace
18:04 imirkin_: sure it is
18:04 imirkin_: just have to think bigger ;)
18:04 coredump: I would use the proprietary driver but that is a mess specially with my very annoying xrandr config (hidpi noetbook screen, 2k display on one port and 1080p on another)
18:05 imirkin_: even if nouveau accel doesn't work
18:05 imirkin_: you ought to be able to use the outputs
18:05 coredump: everyting works fine on the intel main ting
18:06 coredump: I just wanted steam to work, and apparently it needs a discrete card to do that
18:06 imirkin_: should work with intel
18:06 imirkin_: 1050 should be more powerful
18:07 imirkin_: but all/most steam games should work fine on recent-generation intel
18:07 coredump: it won't start
18:07 coredump: misses something
18:07 imirkin_: report it.
18:07 imirkin_: there's probably a way to make it go
18:07 imirkin_: you'd run into the same issue with nouveau anyways
18:08 coredump: hm
18:08 imirkin_: unless it wants bindless textures
18:08 imirkin_: is this DOW3?
18:08 coredump: X Error of failed request: BadMatch (invalid parameter attributes)
18:08 coredump: Major opcode of failed request: 1 (X_CreateWindow)
18:08 imirkin_: stupid question - are you running X?
18:08 coredump: it may be xmonad being annoying too, but the posts I found onlnie all point to 'I fixed installing drivers for the discrete card'
18:09 coredump: yeah, fuck wayland
18:09 imirkin_: well, X_CreateWindow failing is pretty core
18:09 imirkin_: and entirely unrelated to intel drivers
18:10 imirkin_: that's before glX stuff anyways
19:53 pendingchaos: Is there anywhere where GM107+ instructions are documented (other than envydis/envyas and maxas)?
19:55 pendingchaos: (or src/gallium/drivers/nouveau/codegen/nvc0_ir_emit_gm107.cpp)
20:02 karolherbst: pendingchaos: doubtful
20:02 karolherbst: why?
20:04 pendingchaos: I wanted to know if the behaviour of IADD3.RS and IADD3.LS was known and, if it wasn't, document it somewhere
20:04 pendingchaos: also probably to document XMAD sometime
20:04 imirkin_: right side vs left side?
20:04 imirkin_: no clue
20:04 imirkin_: perhaps where the negative bits apply
20:05 pendingchaos: it seems IADD3.RS -> (src0+src1)>>16 + c
20:05 pendingchaos: LS seems to be a left shift instead
20:05 imirkin_: oh. right shift / left shift.
20:05 imirkin_: makes a lot more sense ;)
20:32 pendingchaos: should I create a envytools pull request creating something like the ISA documentation for fermi and tesla?
20:32 pendingchaos: not really something complete, but for documenting non-obvious things like IADD3 (and currently just IADD3)
20:33 pendingchaos: (in docs/hw/graph/maxwell/)
20:34 HdkR: They must be bored to spam in such a dumb way.
20:35 imirkin_: pendingchaos: that'd be awesome ... should definitely have a good place to document various idiocy
20:36 imirkin_: my all-time favorite is TEX of course :)
20:36 imirkin_: [my pet peeve, really]
20:37 imirkin_: thus far the documentation has been largely in my head, as well as a few other heads, and a handful of comments in envydis and codegen
20:37 imirkin_: hardly ideal.
20:44 HdkR: imirkin_: Ah, envytools doesn't have any way to document instructions in it? :)
20:44 imirkin_: well, envytools is not a single thing
20:44 imirkin_: there are some detailed docs for tesla
20:44 imirkin_: but nothing much after
20:44 imirkin_: and there are hwtests for tesla
20:45 imirkin_: mwk got so far as to RE the sin/cos stuff
20:45 imirkin_: and rcp/rsq
20:46 HdkR: hm
21:00 mwk: ... and I just finished REing rcp/rsq on NV20 :)
21:06 mwk: hmm, fun fact
21:06 mwk: imirkin_: +q on ~a means that commit bots cannot talk
21:10 HdkR: imirkin_: I'm mostly curious if there is a way to document instructions since there are people banging on Maxwell a bit recently. Would be nice if their reversing helps more than just their project :)
21:10 mwk: HdkR: just describe them in text and send a PR
21:14 HdkR: That's definitely one way of doing it :)
21:42 pendingchaos: what are the x and cc flags btw?
21:50 imirkin_: mwk: doh, sorry
21:50 imirkin_: mwk: joss just got too annoying
21:50 imirkin_: pendingchaos: .X = consume carry
21:51 imirkin_: .CC = set carry
21:51 imirkin_: which is a 1-bit on/off flag
22:08 pendingchaos: imirkin_: how does this look so far? https://hastebin.com/iwiroqivud.rst
22:09 imirkin_: mwk: --^
22:09 imirkin_: pendingchaos: mwk has generally been the hwdocs maintainer
22:09 pendingchaos:nods
22:10 imirkin_: might want to run it through a spellcheck eventually :)
22:13 juri_: more docs are good. ;)
22:23 pendingchaos: updated: https://hastebin.com/elonidezon.rst
22:24 pendingchaos: the notes in the introduction should probably be somewhere else
22:51 HdkR: Nice. Just needs some encoding information to be golden :)
23:47 imirkin_: HdkR: encoding info is in envydis
23:49 HdkR: imirkin_: But is it in an easy to read textual document like that markdown example? :D
23:49 imirkin_: easier, once you understand how to read it
23:49 imirkin_: instruction encodings are not an intrinsically simple thing to state...
23:50 HdkR: ...Potentially pulling in the information programmatically from some ground truth :)
23:50 HdkR: Of course
23:51 imirkin_: it'd be cool if we could put together an intel-style manual
23:51 imirkin_: it'd be a TON of work
23:51 imirkin_: with, i feel, zero benefit
23:51 HdkR: Pretty much yes
23:52 HdkR: Lets people who are interested in it have a single location to check out everything though
23:52 HdkR: Way less interesting for people that live and breath the ISA
23:53 imirkin_: do such people exist?
23:53 imirkin_: have you looked at the envydis source? it's quite easy to read
23:53 HdkR: I mean, you could call me one of those people :P
23:54 imirkin_: https://github.com/envytools/envytools/blob/master/envydis/gm107.c
23:54 imirkin_: search for "tabroot"
23:55 HdkR: Not too terrible to parse
23:55 imirkin_: T(x) == tabx
23:56 imirkin_: value/mask -> atoms list
23:56 imirkin_: every so often you can have variants
23:56 imirkin_: { 0x53f8000000000000ull, 0xfff8000000000000ull, OP8B, T(pred), N( "idp"), T(53d8_0), REG_00, REG_08, T(53d8_1), REG_20, REG_39, .fmask = F_SM60 },
23:56 imirkin_: fwiw that's not precise, but ... good enough for my purposes :)
23:57 imirkin_: most of this is just a decoding of what's inside envydis, not properly tested