01:12imirkin: skeggsb: looks like more DP fail :)
01:13imirkin: might be related to the OR refactor
01:41mangix: wayland fails in f26. not good.
02:03mangix: getting this: https://paste.fedoraproject.org/paste/I-udg5iCrtJMDnheZWa2jw
02:03mangix: last line keeps getting spammed
02:03mangix: journald can't keep up with the spam
02:06imirkin: mangix: update your kernel... feels like the 64-bit issue
02:07imirkin: [with secboot]
02:07mangix: what 64-bit issue?
02:07imirkin: there was some issue with secboot and >= 4GB vram where the pointer would get messed up
02:08imirkin: or something like that
02:08mangix: i have those patches
02:08imirkin: i'm wrong then. all hope is lost.
02:17mangix: hrm that's a weird address
02:17mangix:had an epiphany
02:19mangix: GART: 1048576 MiB
02:19mangix: what is this GART?
02:20imirkin: 1TB - 40bit address space.
02:37mwk: mangix: GART is system memory that is mapped to GPU's virtual memory system
03:03rhyskidd: mwk: No problem by the way with your comment on my envytools PR :)
03:03mwk: rhyskidd: ?
03:03rhyskidd: is mwk == koriakin?
03:04rhyskidd: it was the nv04_pgraph_is_nv25p() based clean up that I've now dropped
03:05mwk: ah, I thought it was about the ELPG_ENABLE thing
03:06rhyskidd: I'm thinking how best to verify those ELPG_ENABLE bits, given they aren't *used* much in the pre-Pascal mmiotraces I have access to
03:07mwk: meh about that
03:07mwk: just fix the variants and we're good to go
03:07mwk: it's quite obvious from the placement that they're the same bits as in ENABLE
03:07mwk: my complaint is that you have eg. GP100- variants for PGRAPH
03:08mwk: while it obviously exists since GF104
05:39airlied: skeggsb: you're git pull url is wrong btw, git://github.com/skeggsb/git/linux has an extra git
05:39skeggsb: oh, oops, want me to re-send?
05:40airlied: nope it's fine, I just edited out
05:40skeggsb: i'm currently playing "downgrade random shit to find what regressed wayland in f26"
05:40skeggsb: no luck so far on obvious candidates
05:41airlied: skeggsb: what happens? fails to start?
05:42skeggsb: it appears to be limited to gm20x (maybe gp10x i guess, but not sure), but the gpu just goes psycho for unknown reasons
05:42skeggsb: spews dmesg then hangs
05:44airlied: skeggsb: which bit dies, graphics or display?
05:51airlied: skeggsb: does just running mutter against bare X break it?
05:51skeggsb: the gnome X session is fine, just wayland
05:51airlied: oh doh
06:02airlied: skeggsb: does the gdm session work?
06:02skeggsb: yeah, gdm works for some reason
06:02airlied: a g86 bug with PRAMIN exhaustion just went past me
06:02airlied: with wayland
06:03skeggsb: i did see something about that... i *am* noticing that when i do try to login with wayland, we're using a *fucktonne* of channels
06:03skeggsb: which would probably also be why g86 is running out of space there i guess.. that shouldn't be fatal on its own, we fall back to a slower method of touching those structures if we run out of PRAMIN BAR space
06:07skeggsb: does something in f26 cause more "normal" apps/widgets/whatever to be accelerated by chance?
06:07skeggsb: (ie. why so many channels being used?)
06:09airlied: skeggsb: maybe something with gtk
06:10airlied: but I didn't think much changed, I should update something to f26 at some point
06:17skeggsb: airlied: it's related alright
06:17skeggsb: the minute we go over 16MiB of PRAMIN usage on gm20x, we start getting the weird GPU faults people are reporting
06:17skeggsb: the BAR is 32MiB, so not sure wtf is up there
06:18skeggsb: i force-limited the size to 16MiB here, and can login OK (with the PRAMIN exhausted messages, but, as i said, that shouldn't actually matter)
06:20skeggsb: i should plug in a g86 now and make sure it works as expected there
06:20skeggsb: and probably add some kind of LRU to instmem mappings....
06:38skeggsb: airlied: so, the g86 works, it hits the exhaustion thing too.. the slowdown is probably explained by any new page tables or channel structures etc having to be written to by an awfully slow method instead
06:38skeggsb: i don't see the level of slowness the reporter mentions though
07:07mangix: skeggsb: both wayland and xwayland have similar issues
07:07mangix: earlier paste; https://paste.fedoraproject.org/paste/I-udg5iCrtJMDnheZWa2jw
08:11mangix: btw this is some kind of change in wayland causing nouveau to lock up like this. f25 with the same kernel was fine
08:31mangix: sometimes it crashes on logind, other times on xorg. weird.
10:10tjaalton: i've got a bug report saying that mesa 12.0.x -> 17.0 upgrade broke blur on 7300GT, and is still broken on git. asked him to file it upstream
11:31imirkin: tjaalton: correct thing to do :)
11:35imirkin: tjaalton: of course "blur" isn't a thing - it's just some shader thing that broke which happens to hit the shader that does blur on some game...
11:36imirkin: tjaalton: if you could have them provide an apitrace, which when replayed, reproduces the issue, that'd be super.
12:52tjaalton: imirkin: it shows with unity, dunno how to reproduce with other means
12:53karolherbst: tjaalton: apitrace
13:12tarragon: does nouveau have this?? --> feral games
13:13tarragon: oops I meant this --> http://phoronix.com/scan.php?page=news_item&px=Kabylake-Intel-GVT-g
13:21mwk: tarragon: that's Intel-specific technology, so no
13:21mwk: no way
13:23karolherbst: well there is always Vt-d
13:24mwk: that's not exactly virtualization...
13:24karolherbst: I know
13:24karolherbst: but that's the farest you can come to it
13:26karolherbst: uhm.. *clostest
13:27karolherbst: who is comming to XDC by the way?
13:30mupuf: me me me!
13:32karolherbst: I added myself to the list, I think the best would be to share a place to stay like we did last year?
13:34mupuf: well, I let's at least be in the same hotel so we can help shuttle people with the car we rented
13:39karolherbst: ohh, you rented a car?
13:46RSpliet: mupuf: I'll pass (like with SHA2017 :-( ), think I'll be best off investing that kind of money into a driver's license :-P
13:46karolherbst: but yeah, staying at least in the same hotel sounds like a good idea to me. I guess we can figure out details later. I just need to start doing those bureaucratic things on my end, cause I have no (valid) travel pass....
13:47mupuf: RSpliet: sure, except if you had a talk proposal, you could get funded to go
13:47RSpliet: mupuf: I don't atm. And PhD is a bit more pressing than XDC unfortunately
13:47mupuf: this is indeed the hard cold reality, yes :s
13:48RSpliet: Was hoping to get some fermi, reg alloc, insn sched stuff done perhaps, but I can only spend my time once :-)
13:48RSpliet: So if things go well, perhaps next year. Otherwise things get easier if/when I move into decently paid work
13:49mupuf: and we could maybe meet at fosdem!
13:49karolherbst: sounds like a plan
13:49RSpliet: Who knows... That's about when I need to start stressing about write-ups and publications :-D
13:50imirkin_: skeggsb: https://github.com/skeggsb/nouveau/commit/4f425acdeb471e7e7b1804cebd9ca606b94c9223 -- why didn't that get included in your pull?
13:50imirkin_: it's affecting people with otherwise-working setups
13:54karolherbst: imirkin_, pmoreau, hakzsam: XDC status on your side?
13:55hakzsam: no idea
14:11imirkin_: karolherbst: huh?
14:11imirkin_: what status?
14:12mupuf: imirkin_: I guess he is asking wether you plan on attending or not
14:42pmoreau: karolherbst: Not coming as I have too much stuff to do between labs and papers submissions.
14:45karolherbst: pmoreau: ack
15:14rhyskidd_: I'm considering the trip to XDC this time, given closer for me than recent years
16:32tjaalton: karolherbst: apitrace of the session?
16:32tjaalton: basically compiz
16:39karolherbst: no, of the application
16:40karolherbst: or does compiz produce this issue?
16:40karolherbst: usually it's the application itself doing OpenGL stuff
16:41imirkin_: tjaalton: well, so here's the deal - i need to reproduce the issue. you can be sure i'm not about to install compiz. so apitrace works nicely.
16:41imirkin_: (if it's compiz that's failing - if it's a GL application then i'd need that GL application)
16:44tjaalton: karolherbst: it shows with transparent backgrounds when blur effect is enabled, no other reproducer so far
16:49tjaalton: imirkin_: yeah we'll see..
16:50tjaalton: is 8600gt close enough to 7300gt?
16:51tjaalton: because i have the former somewhere
16:57imirkin_: tjaalton: nope
16:57imirkin_: 7300gt is nv4x, while 8600gt is g8x/g9x
16:57imirkin_: there was a *major* architectural change between those
16:58imirkin_: and they use different gallium driver backends
16:58imirkin_: (nv30 vs nv50 backends)
16:58imirkin_: tjaalton: perhaps also worth mentioning that the nv30 backend sucks
16:58imirkin_: that doesn't mean we don't try, but ... limited resources and all
17:02tjaalton: imirkin_: right, okay
17:02imirkin_: tjaalton: you can take the apitrace on any gpu though...
17:03imirkin_: HOWEVER you'd have to limit mesa to the features available on nv4x
17:03imirkin_: which may not be an extremely easy thing to do =/
17:04tjaalton: ah, well i'm not sure where my card is, so :)
17:05imirkin_: basically 7300gt is a DX9 part, while 8600gt is a DX10 part
17:06imirkin_: obviously if the user can do a bisect and find the responsible change, that would also go a long ways to figuring out the issue
17:07imirkin_: [probably a dozen bisection steps]
19:54pmoreau: After using the Intel chip as main card on this laptop, I had forgotten how easily Nouveau would lockup with KDE; it could be due to something else but the lockup occured during a window resizing effect.
19:55pmoreau: And I had also forgotten how NVIDIA would overscale the interface on HiDPI… :-/
20:00imirkin_: moral of the story - use nouveau ^ kde :)
20:01karolherbst: pmoreau: :D
20:02pmoreau: Well, it was working fine as long as I was on the integrated chip. But today, probably after updating to 4.12(.3), it stopped working…
20:02karolherbst: pmoreau: guess why I always suggest using intel even on a desktop if somebody mentions they have an intel gpu
20:03pmoreau: On the same hand, I lost keyboard support, I have no idea how as I kept the same kernel config as usual
20:03karolherbst: pmoreau: you still need to _press_ the keys, even using intel
23:47skeggsb: mangix: yeah, i know *what* the issue is, i'm trying to find the root cause so i can fix it properly. it's not caused by a driver change, but something in f26 is using the gpu a bit differently and triggering a bug
23:47skeggsb: imirkin_: oversight, i'll fix that
23:48imirkin: ok cool. just wanted to make sure you weren't excluding it on purpose.
23:48imirkin: the other thing with not creating heads is wtvr, only known affected chip is GP108 which isn't supported officially yet
23:48skeggsb: nope.. just single-mindedly trying to fix all these horrific issues that have popped up and pulling me away from other stuff :P
23:49imirkin: i assume you saw the new DP errors?
23:49imirkin: fun, right? :)
23:49skeggsb:needs more time in the day, or a clone
23:49imirkin: everything seems to be crumbling at the same time... not sure what happened
23:50skeggsb: definitely don't get paid enough for this shit hehe ;)
23:50imirkin: well - having more people full time on nouveau would sure be great
23:50imirkin: us volunteers try to help, but it's hard to keep up
23:50imirkin: so we stick to our little areas of the world
23:51airlied: skeggsb: I don't think we can fix whatever changed in f26, I assume it was deliberate change
23:51skeggsb: airlied: i can fix the driver
23:52airlied: yup just probably not worroying so much about root causing :)
23:52skeggsb: oh, by root cause i mean: why instmem mappings >16MiB don't work
23:52airlied: oh sorry, yeah that is a real problem :)
23:52imirkin: 32MB bar advertised but only first 16MB work?
23:52imirkin: that's nice.
23:52airlied: can you make the binary use > 16MB?
23:53skeggsb: imirkin: yep
23:53imirkin: someone was off-by-one in the pci config
23:53skeggsb: airlied: presumably, i will try that at some point, but still working through ruling out some stupid driver bug in mmu mapping or bar setup etc