10:17pmoreau: karolherbst: I’m hoping to go to FOSDEM.
10:18karolherbst: "hoping" is not good enough
10:18pmoreau: I haven’t checked with my supervisor, but it should be good. Unless I have some teaching to do, which I need to check first.
10:27karolherbst: do it today then :p
10:28karolherbst: (I haven't checked as well, but... I am sure I am not the only one here going there and I just assume it will be fine)
10:29karolherbst: just a matter of if I need to take PTO days or not :O, well I'll ask
11:46RSpliet: I'm not sure if I'm still welcome in Brussles, flying/training in from Britain
11:47karolherbst: RSpliet: you got it wrong: everybody (except those anti-EU politicians) is welcomed. And even if it means that in the near future only 10k people will live in the UK
11:48karolherbst: flee while you still can!
11:48RSpliet: Heh... think you'll start spelling England with the T from "tax haven" soon
11:49karolherbst: oh well, but jokes aside, there won't be a brexit anyhow... just wait a little and then everything will be soon over
11:49karolherbst: I am not joking
11:49RSpliet: 'd love to see that happen, but with the political climate here wouldn't hold my breath for it
11:50karolherbst: ;) well didn't get it better after the last vote? Or did I somehow missunderstood something
11:50karolherbst: currently they are in full "not face loosing
11:50karolherbst: " mode
11:51RSpliet: Ehh yeah, to their own population, not so much to the rest of the world
11:51RSpliet: the face of BoJo is one they'd rather lose more than anything (I hope)
11:52karolherbst: I was already joking about the non-brexit prior the vote, not joking anymore after that :p
14:29MichaelP: arch linux xf86-video-nouveau system freezes only mouse works. Go back to nvidia driver everything runs normal. What could be causing this ?
14:31imirkin: a bug in nouveau somewhere, most likely
14:33MichaelP: imirkin: if that the problem... looking on google i see samething since 2015
14:33imirkin: what did you search for, "random freezes with nouveau"?
14:34MichaelP: arch linux nouveau freezes
14:34imirkin: something must be wrong with google
14:34imirkin: if it's only back to 2015
14:34imirkin: more like since the dawn of time
14:35MichaelP: i see one that says Firefox + nouveau on fresh -current install freezes system 2013
14:35imirkin: i can't tell if you're being serious or not, but you do realize that it's not like *1* bug that causes freezes, right?
14:36MichaelP: It's mainly when running google chrome
14:37imirkin: if you're looking for a stable experience, you're going to be best off using nvidia blob drivers, or getting an AMD gpu if you're interested in supporting open-source.
14:39MichaelP: laptop has amd in it.... desktop i have nvidia gforce gt 730
14:40imirkin: well, that's basically the gpu i have plugged in right now
14:40imirkin: 02:00.0 VGA compatible controller : NVIDIA Corporation GK208 [GeForce GT 730] [10de:1287] (rev a1)
14:40MichaelP: the 730
14:40imirkin: no problems with crashes/etc
14:40imirkin: but i also don't do crazy things
14:41imirkin: like touching gnome/kde, compositors, etc
14:41imirkin: i do use chrome, which is a little crazy, but it's hard to avoid *all* the insanity in one's life
14:42MichaelP: I don't do crazy things either... music video's vlc mpv.... internet browseing google-chrome-beta... google chrome is when things freeze
14:44MichaelP: What distro you running ?
14:46imirkin: gentoo, but that's unlikely to matter
14:49MichaelP: I think sometimes a distro does... Like last few installs of opensuse tumbleweed the last few weeks. Install bvidia reboot to black screen....
14:49MichaelP: opps nvidia not bvidia
14:56kherbst: the .... I found that silly mmiotrace bug with 0x200 big mappings
14:56kherbst: the hell
15:01karolherbst: like you call ioremap(resource, size) and get something like 0xffffb01dc1b11200 back, but because that all makes sense, but then we get a page aligned address into mmiotrace_iounmap, which is 0xffffb01dc1b11000
15:06karolherbst: well, I should just page allign on register and unregister.... this should make our life much much easier
15:14karolherbst: nice, fixed
15:16karolherbst: allthough not super sure if we can indeed get several arms on the same page with mappings smaller than pages but meh
15:16karolherbst: skeggsb: try this instead of your hacky nvidia patch
15:16karolherbst: only the register_kmmio_probe and unregister_kmmio_probe part should be important
15:27tagr: skeggsb: do you happen to have a public kernel tree with the patches for v4.15-rc1? I'm trying to track down what exactly is causing the regressions on Tegra but the non-linearity of linux-next makes that somewhat difficult
15:28karolherbst: tagr: you can use the out of tree module
15:28imirkin_: tagr: https://github.com/skeggsb/linux/commits/linux-4.15
15:28karolherbst: tagr: https://github.com/karolherbst/nouveau/commits/master_4.14
15:28karolherbst: or that
15:28imirkin_: but if it's in nouveau only, the out-of-tree thing may be faster
15:29karolherbst: out of tree is sometimes painful because you need to use the exact drm-next thing and stuff like that
15:29karolherbst: but I usually rebase those things
15:29tagr: I don't exactly know where it's coming from, I've been getting many false positives and a couple of different symptoms, so I'm just trying to narrow things down at this point
15:30karolherbst: tagr: that's normal...
15:30karolherbst: I don't like to bistect linus tree as well
15:31tagr: I haven't run into many issues with linus' tree, but I've found linux-next to be a major pain unless you end up getting lucky
15:31imirkin_: linux-next isn't meant to be bisected
15:32tagr: often a plain bisect will jump between different merge bases and then you'll sometimes end up on -rc1, other times on -rc7...
15:32imirkin_: it's pretty funny when peopel try to base patches on it
15:32karolherbst: I don't think any of those kernel trees are actually meant to be bisected....
15:32karolherbst: at least I think if they would be, linux wouldn't allow such a merging pain
15:32karolherbst: or somebody else
15:32karolherbst: and yes, the pain is merging indeed
15:33karolherbst: with every merge you do
15:33tagr: well, depending on your area of interest it's difficult to bisect anything other than linux-next
15:33karolherbst: best part is, where you bisect the 4.14 rc kernel and end up in 4.12-rc* for some reason
15:33imirkin_: yes, embedded development does suck
15:34karolherbst: bisecting linux even sucks on a non embedded system and not just linux-next
15:34imirkin_: [blame the hw engineers who can't stick to a single thing and improve it rather than redoing everything from scratch every time]
15:34tagr: for Tegra we've had regressions in the strangest of places, sometimes causing no output from a device to appear at all, so a bisect on linux-next will often give you at least a rough idea of where to start looking
15:34tagr: hehe =)
15:35karolherbst: imirkin_: throw more hw engineers at the problem
15:35imirkin_: (and of course the hw engineers blame the sw people for not being able to keep up)
15:35karolherbst: allthough I sincerly believe the world needs less hw engineers rather than more....
15:37imirkin_: it's a bit frightening how little i know about the hw development process
15:39karolherbst: well, it isn't that much different than sw development I think, at least for the small stuff...
15:40karolherbst: it isn't like they draw everything by hand still
15:40imirkin_: i can imagine what it might be
15:40imirkin_: but i have no idea what it actually is.
15:40karolherbst: switch jobs and find out
15:40imirkin_: "Looking for a job: guy who knows nothing about the field."
15:40imirkin_: (perhaps i should switch to marketing?)
15:41karolherbst: you just tell them you worked a lot with nvidia hw
15:41karolherbst: should be enough
15:41tagr: having more SW engineers switch into a HW engineering position might actually be a good thing
15:41karolherbst: you don't have to strictly tell them you never done hw engineering
15:41karolherbst: just tell them what you do know
15:42karolherbst: imirkin_: maybe we should just start to produce our own GPUs, which are compatible with the Nouveau driver :D
15:43tagr: imirkin_, karolherbst: thanks for the pointers, exactly what I was looking for
15:43karolherbst: and those are completly blob free and the vbios is provided as a source file as well
15:43imirkin_: [this took too long to look up, but, check the top quote from dave barry at http://allthetropes.wikia.com/wiki/Adjacent_to_This_Complete_Breakfast]
15:43imirkin_: in re me saying that i worked with nvidia hw.
15:45karolherbst: does any HR person care?
15:45karolherbst: well some do
15:45karolherbst: but you know
15:48tagr: ugh... looks like we have at least two separate regressions, one of them I can't see on skeggsb's tree
15:49imirkin_: you don't give him enough credit
15:49imirkin_: he's _really_ good...
15:50tagr: sorry if it came across like that, I didn't want to be criticizing his work at all
15:51imirkin_:was being facetious
15:51imirkin_: as if the ability to secretly cause regressions in other people's code was a positive thing
15:51imirkin_: obviously i failed.
15:51tagr: or I did
15:51imirkin_: let's just compromise - we *both* failed
15:52tagr: I can live with that =)
16:04kherbst: imirkin_: ... I was like thinking: I get me a laptop for work with a pascal GPU to debug all those fancy MST dock features and stuff like that. Guess what is the problem here?
16:06karolherbst: I don't know why, but for some reasons all the laptops I picked, had all their connectors wired to the intel gpu
16:08tagr: karolherbst: I guess it makes sense, though, most of the point of Optimus is to have the Intel GPU display everything and only power up the NVIDIA one for rendering, right?
16:08karolherbst: tagr: most laptops use the dedicated GPU for this
16:08karolherbst: like most
16:08karolherbst: well the intel can't handle 2x 4K screens under linux as it seems
16:09karolherbst: so it kind of makes sense?
16:09tagr: heh, yeah, that's not so useful
16:10tagr: karolherbst: no way of muxing the the connectors either?
16:11karolherbst: but I think only apple uses hw muxes
16:12pmoreau: Didn’t even Apple dropped hw muxes in favour of sw muxes some time ago?
16:12karolherbst: did they?
16:14pmoreau: I’m not sure. I thought they did, but Lekensteyn might remember better than I do.
16:15tobijk: karolherbst: "acer aspire vn7-593g" has wired shit to the nvidia gpu
16:15karolherbst: tobijk: it is easer to llist which laptop doesn't
16:15karolherbst: I know two laptops now
16:16tobijk: karolherbst: so in this generations they all started wiring up the nvidia gpu again? ~_~
16:17Lekensteyn: pmoreau: Lukas Wunner knows more about Apple hw than I do
16:19pmoreau: Lekensteyn: Doesn’t mean you know nothing about it ;-) He doesn’t seem to be logged in, therefore I didn’t tag him.
16:24Lekensteyn: if by "harware mux" you mean a phsyical switch, then I don't know of any modern laptops that have them
16:24Lekensteyn: hardware mux*
16:24karolherbst: Lekensteyn: of course not
16:24karolherbst: but I am quite sure they still have a mux builtin they can control through software
16:25Lekensteyn: heh, my reply was too disappointing to karol :p
16:26Lekensteyn: thinking about it, yes it was a stupid answer
16:26Lekensteyn: my laptop from 2 (?) years ago has a hardware mux
16:27Lekensteyn: it is in the schematics and has a firmware setup option to switch between hybrid/discrete mode
16:28Lekensteyn: but on the other hand, I was surprised to see that a Dell XPS 9560 has its external outputs wired to the intel GPU, didn't know they were still doing this
16:29pmoreau: how does a software mux work: do you move the frame to the CPU and then upload it to the other GPU? That would be super slow and inefficient, wouldn’t it?
16:31kherbst: Lekensteyn: well, that's the exact laptop I have :D
16:33Lekensteyn: karolherbst: and so do my siblings, that's how I learned about it :)
16:33karolherbst: well there aren't many where it is wired to the intel one
16:33Lekensteyn: pmoreau: that's basically what PRIME is doing, nvidia is used as rendering device and intel as display
16:36pmoreau: Lekensteyn: OK, thanks
16:37karolherbst: mhhhh something is super odd
16:37karolherbst: I get all the mappings into the mmiotrace log file, but no memory accesses
16:39karolherbst: uhhh. it uses a 1G page
17:16imirkin_: karolherbst: sometimes the dock connectors go to the dGPU
17:16imirkin_: but yeah, not always
17:28karolherbst: so, it should kind of work now: https://github.com/karolherbst/linux/commits/mmiotrace_fix
17:28karolherbst: there seems to be still an issue while iounmapping everything "mmiotrace: purging non-iounmapped trace", but at least the machine doesn't crash
17:29karolherbst: still, other vmm issues, but that's no big issue right now
17:29karolherbst: skeggsb: ^
17:30karolherbst: and that's not even mmiotrace fault really... but the mm subsytem giving different pointers into mmiotrace for ioremap and iounmap
17:30karolherbst: no wonder I never found the issue until I looked at the code surounding it
20:22pmoreau: Yes, figured out how to enable the OpenCL extensions.
20:24imirkin_: fuck, forgot to look at your spirv stuff over the weekend
20:27pmoreau: Don’t worry: it might be better not to look at the code too closely. :-D
20:29tobijk: imirkin_: second thought about the patches you created lately? :D
20:29pmoreau: I really need to solve the liveSet computation, to be able to use loops.
20:33imirkin_: tobijk: i was playing with a4xx on freedreno
20:33imirkin_: i.e. no
20:33imirkin_: oh, does one of you guys have a fermi handy?
20:35pmoreau: I should be able to find one rather quickly and plug it in.
20:36imirkin_: build modetest with that and then test the C8 thing and see if you get a white screen or not
20:36imirkin_: (not = you get the SMPTE pattern)
20:36imirkin_: note that i'm looking for pre-GF119
20:36imirkin_: it fails on G92
20:36pmoreau: I should have a GF100
20:36imirkin_: (and, presumably based on the bug, MCP79)
20:37pmoreau: I could try that on my MCP79
20:37pmoreau: (as well)
20:37imirkin_: but it works on GK208
20:37pmoreau: And I should have a G92 somewhere
20:37imirkin_: no need to test on G92 - i already have ;)
20:37pmoreau: Ah :-)
20:40tobijk: imirkin_: only does it have to be a fermi? i have an old tesla around
20:41imirkin_: tobijk: because i've confirmed it's broken on tesla
20:42imirkin_: same code is used for G84:GF119
20:42imirkin_: stands to reason that fermi will have the same issues
20:42imirkin_: i was just hoping someone could confirm
20:43pmoreau: GF100 plugged in, going to give the patch a try
20:44imirkin_: pmoreau: thanks! just buidl modetest and then run like ... modetest -s DVI-I-1:1920x1200@C8 or something
20:45imirkin_: from the console
20:45imirkin_: hit enter to exit, it will kill the current console, but just flipping back and forth between vt's should get it back
20:45imirkin_: you should see a SMPTE test pattern
20:45pmoreau: Without X running I guess?
20:45imirkin_: doesn't matter - just run it from a vt
20:45imirkin_: X will drop master
20:46pmoreau: And where do I get modetest from?
20:46pmoreau: OK, thanks
20:46imirkin_: [which is what the patch applies to]
20:50imirkin_: karolherbst: have you ever used the userspace out-of-tree stuff to play with modesetting?
20:50imirkin_: i assume it doesn't work if the code i want to test is in nv50_display.c
20:52pmoreau: imirkin_: How do you force it to open nouveau rather than i915? The screen is connected to the NVIDIA card, but it still is opening the Intel device :-/
20:52imirkin_: -D PCI:1:0:0
20:53karolherbst: imirkin_: I can't, literally... except the uefi allows me to switch my main GPU
20:54imirkin_: karolherbst: k
20:54karolherbst: have to wait until my desktop arrives
20:54pmoreau: imirkin_: I get a white screen
20:54imirkin_: pmoreau: ok, that's what i get too
20:54imirkin_: thanks for confirming!
20:54pmoreau: You’re welcome :-)
20:54imirkin_: [and GF119 and later should all work fine]
20:55imirkin_: [and i need to double-check what the deal with dispnv04 is]
20:55karolherbst: what are you testing?
20:55karolherbst: C8 format?
20:55karolherbst: (whatever that is)
20:56imirkin_: C8 - indexed 256-color palette
20:56pmoreau: (Going to try on the GM206)
20:57pmoreau: (since it is already plugged in)
21:03pmoreau: (I get the SMPTE pattern on GM206)
21:03imirkin_: not surprising.
21:03imirkin_: but still nice.
23:26imirkin_: skeggsb: so ... any good ideas for why G84:GF119 C8 broke? looks like we used to set lores for that before.
23:26imirkin_: (airlied / karolherbst -- know if he's still on vacation?)
23:28karolherbst: imirkin_: yeah
23:28karolherbst: until next week
23:28imirkin_: karolherbst: and are you doing nouveau stuff for work now? or still just for fun?
23:33karolherbst: imirkin_: 100%
23:33karolherbst: imirkin_: but not all graphics
23:34karolherbst: currently more like fixing pascal issues now
23:34karolherbst: on my laptop runpm doesn't work and starting the gr falcons neither
23:34imirkin_: yeah - a lot of people have that problem
23:35karolherbst: well I should focus more on compute stuff when I come to it and skeggsb should stay with graphics, that's roughly the idea
23:35karolherbst: imirkin_: well, loading nvidia first and then nouveau works
23:35karolherbst: but I think you already know
23:35imirkin_: i wonder which RH customer gives a shit
23:35imirkin_: (not that i'm opposed ...)
23:35karolherbst: can't tell you :p
23:36imirkin_: doesn't stop me from wondering!
23:36karolherbst: to be honest, I would really _love_ to tell you, because the story is quite funny actually. Maybe at some point you will know :p
23:36imirkin_: my best guess is "nvidia"