00:40imirkin: anyone with a fermi or kepler1 or tesla? mind running 'rendercheck -t triangles -o clear -f a8r8g8b8' and letting me know if anything passes?
04:54steve555: Hi imirkin, I saw your request and here are my results: http://pastebin.com/a6g60TET
04:55imirkin: steve555: thanks for checking. that's on a fermi, right?
04:55steve555: I can't stop long as I have to go to work v shortly.
04:55steve555: Yes it is.
04:56imirkin: steve555: thanks for checking!
04:56imirkin: i think something that the triangles test does is causing fail in the ddx... =/ i guess i'll track it down eventually
04:56steve555: Here is my card: 01:00.0 VGA compatible controller: NVIDIA Corporation GF114 [GeForce GTX 560] (rev a1) (prog-if 00 [VGA controller])
04:57steve555: And no worries :)
04:57imirkin: on the bright side, it fails across every generation
04:57imirkin: so i think it's some high-level thing nouveau gets wrong
04:58steve555: That is good to know, I'm sure you'll find a way.
04:58steve555: I'll have to go now, happy hunting!
05:00imirkin: see ya
05:27ShimmerFairy: imirkin: for the record, I saw that request of yours in the logs as well (GT 610), there were zero tests passed of six with kernel 4.8.1. I can share the detailed output if you need it.
05:28imirkin: ShimmerFairy: thanks. that's what i get too - all fail.
05:28imirkin: (and also what steve got)
05:29ShimmerFairy: I have to say, the idea of a RENDER compliance test coming back "all fail" and yet me still using X11 sufficiently is a bit funny, as someone new to this area of development :)
05:30imirkin: well, it's a very specific test
05:30imirkin: appears to be some sort of failure in the Triangles & co calls of the render ext. i suspect they're never used.
05:31imirkin: you'll notice that if you do e.g. -t composite, it all passes
05:32ShimmerFairy: Yeah, it makes more sense now seeing that the compositor is a separate component from the triangle stuff (if I understand correctly).
05:33imirkin: also note that the compositor probably uses the COMPOSITE extension rather than the composite stuff in the RENDER extension.
05:36ShimmerFairy: right, I was thinking of the extension when I said that. Like I said, new to this :)
05:36imirkin: yeah, lots of little details.
05:37ShimmerFairy: I'd like to help develop nouveau, specifically because I'd like something as good as nvidia performance-wise, but without the inexplicable breaking of the kernel nvidia seems prone to causing (it's very weird).
05:41imirkin: well, nouveau stability is ... lacking
05:41imirkin: any and all help welcome, of course.
05:42ShimmerFairy: I think I'll keep trying to poke at the wiki, even though a lot of it looks outdated (but I'm guessing the information doesn't age that fast?), to try and get my bearings on the project.
05:44imirkin: a lot of it is indeed outdated
05:48imirkin: [as i just mentioned in #xorg-devel, i think i see what's up... we fail PrepareAccess, but the exa logic isn't ready for that function to fail. the solution is to create a transfer thing that returns a staging area]
05:51ShimmerFairy: The thing that seems most interesting to me is how you do this kind of development where you'd have to restart at least xorg depending on what you've changed exactly.
05:53imirkin: X -config /etc/X11/xorg-test.conf -sharevts -novtswitch :1
06:08imirkin: [ideally you have a second machine where you can dick around with all this stuff, but i do all my dev on my main box]
06:25ShimmerFairy: I have just the one machine as well :)
06:33cousin_luigi: My gtx660 is still unusable (it freezes after minutes and sysrq is no help). What do I need to set up a serial console to get more information about this?
06:36imirkin: cousin_luigi: GK106 right?
06:36imirkin: some people have had success by using the blob ctxsw fw with it
06:37cousin_luigi: imirkin: yes and what is cxtsw?
06:37cousin_luigi: imirkin: But I'd like to find out more so that I can at least contribute with more information.
06:38cousin_luigi: imirkin: I mean, can I tail -f /var/log/kernel.log (or systemd equivalent) from a serial console or do I need something lower level?
06:39imirkin: have a look at https://bugs.freedesktop.org/show_bug.cgi?id=72180 and https://bugs.freedesktop.org/show_bug.cgi?id=93629#c10
06:45cousin_luigi: imirkin: I think I tried installing the firmware. Using that script of yours.
06:45cousin_luigi: imirkin: No big improvement, actually it seemed to crash more IIRC.
06:45imirkin: ah, fun
06:46imirkin: sorry - i don't have any great debugging advice =/
06:47cousin_luigi: imirkin: by the way I packaged it for opensuse
06:49imirkin: oh, are you using the mesa that comes with opensuse? they decided to patch mesa with some half-baked patches. try upstream and see if that helps.
06:52cousin_luigi: imirkin: hmm
06:53cousin_luigi: imirkin: but would that account for such hard crashes?
06:53imirkin: probably not....
06:53imirkin: it should only cause deadlocks in userspace
07:15cousin_luigi: imirkin: then I'm back to square one:/
07:16cousin_luigi: From the wiki I see I should just follow the serial console guide from kernel.org, but I'm for oral tradition:)
07:17imirkin: yeah, serial console is great if you can set it up
07:17imirkin: otherwise netconsole could be useful
07:25cousin_luigi: imirkin: I only have to look for the de-9 header to connect to my motherboard.
07:25cousin_luigi: I must have it somewhere...
07:25imirkin: hakzsam: looks like heaven shows some neat artifacts on GM107 set to Ultra with MSAA x4. not quite sure how to describe it... not major, but definitely noticeable.
07:27imirkin: looks fine with MSAA turned off... probably something subtle :( ugh
07:27imirkin: yeah, it's definitely a MSAA artifact.
08:25cousin_luigi: ok, serial is set up. If anyone knows what to do once having the two machines connected via null-modem, please tell me
11:14pmoreau: skeggsb: Ping regarding https://bugzilla.kernel.org/show_bug.cgi?id=153911#c3 Shouldn’t NV_INFO & co respect the CONFIG_NOUVEAU_DEBUG_DEFAULT variable? Should those macro even be kept, or switched to the nvkm ones instead?
12:44hakzsam: imirkin: I have never seen blob using 128-bits const load
12:44hakzsam: and yeah, I know for heaven
12:44hakzsam: never looked into this though
12:45cousin_luigi: Any suggestion about the use of the serial console for debugging will be appreciated.
12:46cousin_luigi: just the kernel option mentioned in the document linked in the FAQ or is there more to it?
14:18karolherbst: imirkin: k, the game will switch simply over to use RGBA8
14:18cousin_luigi: http://pastebin.com/raw/4tXSYcXf <- serial console set up, what do you make of this?
14:18karolherbst: imirkin: but maybe it would be nice to have a driconf option for disabling formats you can't render to for texturing or something like that
14:22cousin_luigi: this while trying to log into plasma5
18:04imirkin: gr. having trouble with the new tic format :(
18:16karolherbst: well at least it works with maxwell1, so progess!
18:16karolherbst: does it indeed work smoother than modesetting?
18:18imirkin: no clue, and not the point
18:18imirkin: the point is to provide acceleration with well-tested and error-handling code
18:21zoominee: Hi! I just bought a GeForce 730 and it works but freezes sometimes. Had the same problem with GeForce 640 before. So I ran piglit - where can I send the output?
18:22zoominee: Also, does anyone want my old GeForce 640? The fan is noisy, apart from that it works. I don't mind paying the postage (in Europe).
18:24karolherbst: imirkin: meh, perf >= stability, everybody knows that
18:24imirkin: karolherbst: well the solution i'm handing end-users who want e.g. kde is to remove nouveau_dri.so ... so modesetting isn't so great for that ;)
18:26karolherbst: maybe kde would just work with the ddx
18:28imirkin: but ultimately, what i want in X is stability
18:30imirkin: w00t, that fixed it. apparently if you tell the sampler that you have a 1x1 texture and to take the border from the texture ... nothing good comes of that.
18:31karolherbst: can't see why
18:31cousin_luigi: imirkin: do you have an opinion on the log I showed earlier?
18:31imirkin: cousin_luigi: my opinion is that it sucks :)
18:31cousin_luigi: imirkin: anything else I can do?
18:31imirkin: cousin_luigi: lots of people are having trouble with KDE
18:32imirkin: i've been telling people to either not use KDE or not use nouveau_dri.so for KDE stuff, at their preference
18:32karolherbst: well, we need to fix it at some point
18:32imirkin: perhaps someone who cares about KDE && nouveau will show up and figure out what's up
18:32imirkin: but so far it's one or the other :)
18:32karolherbst: it's no kde related issue though
18:33pmoreau: What is the issue with KDE? It has been running perfectly on my laptop.
18:33karolherbst: pmoreau: plasma5
18:33imirkin: everyone's reporting hangs
18:33pmoreau: karolherbst: Plasma 5 here as well
18:33karolherbst: no clue then
18:34karolherbst: pmoreau: on your desktop, right? laptop doesn't count for obvious reasons
18:34pmoreau: zoominee: Is that 640 a Kepler?
18:34cousin_luigi: pmoreau: mine is a kepler and I think it was sent by the devil himself
18:35pmoreau: karolherbst: Laptop, but I had the NVIDIA as main since the firmware disables the Intel unless told not to.
18:35imirkin: nv50 vs nvc0? who knows.
18:36karolherbst: pmoreau: ohh right
18:36pmoreau: cousin_luigi: :-D That's the one for which you needed some serial console setup?
18:36karolherbst: the heck? :D
18:36cousin_luigi: pmoreau: precisely
18:36cousin_luigi: it hangs, hard
18:37cousin_luigi: perhaps I should try a livecd with bleeding edge everything
18:37pmoreau: That’s no fun…
18:37cousin_luigi: pmoreau: which one?
18:38pmoreau: Most likely won’t help, some GK106/107 have been misbehaving for quite some time now.
18:38pmoreau: "which one?" ?
18:38cousin_luigi: pmoreau: having these problems or testing bleeding edge stuff?
18:39cousin_luigi: Is there a semi-recent nvidia card that works well with nouveau? Because I'm willing to trade.
18:39pmoreau: having those problems
18:39cousin_luigi: pmoreau: But I bet you're having worse problems on your island, aren't you?;)
18:40pmoreau: I think Tom^ is quite happy with his 780 Ti on Nouveau, IIRC, especially with Karol's reclocking patches
18:40cousin_luigi: pmoreau: does he use plasma5?
18:40cousin_luigi: Tom^: ^
18:40zoominee: pmoreau: according to the list at nouveau, yes.
18:41pmoreau: I had worse problem on my old laptop, where none of the two NVIDIA GPUs would let me boot with Nouveau. Took some time to fix them
18:41Tom^: it does infact run about almost all the things the blob does at a bit of a performance cost
18:41Tom^: the only issue im having is that once in a while it randomly freezes upon gaming
18:41zoominee: pmoreau, lspci output for that GeForce 640 see https://paste.pound-python.org/show/bnnU4Tm1aCu8cPUcTQy3/
18:42Tom^: no matter the clocks, or pstates. so its somewhat related to something else. and it always happends when i dont even have time to try to ssh in to see if it errors :P
18:42Tom^: silly cs:go competetive games.. :D
18:43Tom^: cousin_luigi: and no i dont use plasma5
18:43cousin_luigi:would content himself with desktop use
18:43cousin_luigi: right now it crashes after minutes after logging in, assuming it does.
18:44pmoreau: zoominee: Oh, a GK208 and not a GK107, cool! I would be interested since I don’t have any Kepler, outside of the GK107 in my laptop.
18:44zoominee: pmoreau, I can send it to you, what's your postal address?
18:48cousin_luigi: Is there anything I can do about my problem? More debugging or anything?
18:54pmoreau: I don’t know if making an API trace right before the hang would help…
19:05cousin_luigi: pmoreau: I'll do it if you tell me how.
19:17mezo: hey, any way to fix this? http://pastebin.com/4jTmCafy
19:17karolherbst: mezo: update your kernel
19:18imirkin: mezo: revert the relevant drm-next commit from the nouveau repo
19:18karolherbst: current master needs something newer than 4.8
19:18karolherbst: the drm-next kernel actually
19:18mezo: i see
19:18karolherbst: I have a 4.8 compatible tree though
19:19mezo: perfect ;) thanks
19:19imirkin: anyone with a desktop GM20x - https://github.com/imirkin/xf86-video-nouveau - give that a shot and see if the DDX works OK (instead of using modesetting).
19:47karolherbst: does red hat have access to the cts?
19:47karolherbst: I find that hard to believe to be honest
19:47imirkin: people at RH have access to CTS, yes.
19:47karolherbst: but not RH itself, right?
19:47karolherbst: aka ben
19:47imirkin: RH is a legal entity
19:48imirkin: ben has access to CTS, but i don't think he's availed himself of that access
19:48karolherbst: couldn't he run nouveau through that and the opengl 4.4, 4.5 situation would be solved this way?
19:48imirkin: dave has. he's run it against nouveau a couple times too.
19:48karolherbst: I see
19:48imirkin: but it was a pile o' fail. a lot of which was CTS, and some of which, i'm sure, was nouveau.
19:48imirkin: sadly with just the test names, it was a bit hard for me to tell
19:49karolherbst: but again
19:49imirkin: (well, he gave me the full output, but that output was a lot less useful than the dEQP output tends to be)
19:49karolherbst: if ben would run the cts on nouveau and it would kind of work, wouldn't that be enough for exposing 4.5?
19:50imirkin: we can expose 4.5 today, i'm only holding it back coz the AMD guys don't expose it
19:50imirkin: we can expose 4.6 for all i care...
19:50karolherbst: we hold back due to legal issues, don't we?
19:50imirkin: not to my knowledge.
19:51karolherbst: well, we discuessed it here
19:51imirkin: are you a lawyer?
19:51karolherbst: and it seems like you can't just say you implement opengl 4.5
19:51airlied: bumping nouveau to 4.4/4.5 would be a disservice to downstreams
19:51karolherbst: uhh right
19:51RSpliet: you can't claim conformance...
19:51RSpliet: nor use the trademarks
19:51karolherbst: downstream needs to have access to the cts, right?
19:51airlied: who'd have to patch it out to avoid trademark issues
19:51imirkin: nor do we do that anywhere.
19:51airlied: karolherbst: we can work out access to CTS for nouveau devs at some point
19:52karolherbst: airlied: right
19:52imirkin: (claim conformance or use trademarks)
19:52karolherbst: airlied: I am asking, cause michael just went ahaed....
19:52airlied: fixiing CTS is the problem
19:52imirkin: if it's just the version string, i'd be down for s/OpenGL/Libre3D/g all over
19:52airlied: imirkin: displaying the version string is claiming conformance in theory, in practice I'm not sure we care
19:53imirkin: airlied: right, my main point is it's not like we're putting out marketing material saying that we have their blessing
19:53airlied: at least for radv I display a large "I'm not conformant warning"
19:53imirkin: airlied: also, given recent changes in userspace programs, i think downstream would be best off not shipping nouveau_dri at all
19:53airlied: imirkin: if the commit says this driver isn't conformant it probably is okay
19:54karolherbst: airlied: I am sure, if it would come to a law suite, you could get away with fair use
19:54airlied: I just try not to ship KDE :-P
19:54karolherbst: airlied: :D harsh
19:54airlied: more likely phoronix is the one in danger of lawsuits, since he runs around saying conformant in his articles
19:54karolherbst: I doubt that
19:55karolherbst: cause he actually asked khronos
19:55imirkin: maybe we can create a PIPE_CAP_BLOODMONEY and then append (Not Conformant) to the end of all version strings if it's not set
19:55karolherbst: and he said, we should ask khronos and they will give us access, basically)
19:55imirkin: to make it clearer
19:55karolherbst: airlied: saw this? http://www.phoronix.com/scan.php?page=news_item&px=Khronos-OSS-Conformance-Clarify
19:56airlied: well it's not really blood money, they had to do something about the shitty situation
19:56airlied: and charging money to fund CTS developement was the best solution
19:56imirkin: it's "if you don't pay us, we'll sue" money. i call that blood money.
19:56airlied: imirkin: they'd call it trademark violoation
19:57karolherbst: but that aside, I find it annoying, that michael just do things like that out of the sudden
19:57imirkin: airlied: lawyers and i tend to use different terms. same thing though :)
19:57karolherbst: airlied: I am like 70% sure, it would fall under fair use
19:57karolherbst: 30% because I don't have the money for court
19:57karolherbst: basically we just reimplement the API
19:57imirkin: karolherbst: just ignore that echo chamber entirely.
19:57karolherbst: and if the API say: return "OpenGL" here, it's their fault
19:58karolherbst: imirkin: nonono, you don't understand
19:58karolherbst: he actually talked with khronos guys
19:58imirkin: so what?
19:58karolherbst: they said, we should talk to them
19:58karolherbst: now it is public, if we do nothing, it seems like we are uninterested or something like that, anyway, silly situation, and stupid
19:58imirkin: anything coming from phoronix has approximately 0 legitimacy in my mind
19:58karolherbst: well, sure
19:59imirkin: so... stop listening to him, and the troll will lose his power :)
19:59airlied: so if you had access to CTS, would you actually fix the bugs is the question, or just advertise 4.5 and move on?
19:59karolherbst: I think we would do it with radeonsi, no crashes
19:59karolherbst: and then expose it :p
19:59imirkin: airlied: in all likelihood, i'd fix the bugs.
20:00imirkin: depends on the bugs i suppose? can't you just comment out all the failing tests and submit that for conformance? :)
20:00airlied: imirkin: if someone reviews it and passses it yes
20:00airlied: if someone reviews it and reject it, no
20:01imirkin: so you can't comment out EVERY test... doh :)
20:02karolherbst: airlied: but basically you are saying that everybody builds a distributed package of mesa, would need to pay the CTS thing?
20:02karolherbst: with the current situation
20:03karolherbst: well one interested link was in that article: https://www.khronos.org/members/ip-framework
20:03karolherbst: no idea what that means specificly though
20:03airlied: karolherbst: no I'm saying they'd have to avoid shipping a non-conformant implementation
20:03airlied: I don't think they money really matters
20:03karolherbst: ohh, I see
20:03airlied: the whole point of conformance is you don't have people shipping things that don't conform
20:03karolherbst: they want to be on the safe side I think anyway
20:04airlied: and you can enforce that with trademarks
20:04karolherbst: but I doubt it works like that for non commercial projects
20:04airlied: so you don't get shitty half implemented GL4.5 and people saying GL sucks
20:04karolherbst: as long as they don't violate the trademark
20:04airlied: like try shipping a half done DX11 driver
20:04airlied: and watch what happens
20:04karolherbst: windows crashes?
20:05airlied: that and you have a lawyer outside
20:05karolherbst: well, I did it for non commercial use cases
20:05karolherbst: I claim fair use
20:06airlied: karolherbst: I think as long as downstreams don't go marketing things at GL4.5 conformant then it probably won't matter
20:06karolherbst: most likely, yes
20:07imirkin: airlied: anyways, i certainly don't want to actively do things that'll mess up distros
20:08imirkin: airlied: at the same time, the amount of time i can sink into nouveau is not infinite
20:08airlied: imirkin: leave it at GL4.3
20:08airlied: let someone else deal with it
20:08imirkin: i can do that.
22:52imirkin: skeggsb: you probably saw, but i've sent patches for EXA support for GM10x and GM20x... GP10x should be about as easy as adding GM20x, esp if we're willing to call it "NV_MAXWELL".
22:53imirkin: skeggsb: i'll push once i get some tested-by's from people with the HW, unless i hear something to the contrary from you
22:55skeggsb: feel free to push them if you wish, i still have zero plans to continue updating it beyond bug fixes going forward... ie. i'm not porting the stuff necessary for mst to the nouveau ddx either, when -modesetting is perfectly capable and supports everything already
22:55imirkin: k, no worries.
22:59imirkin: tbh i'm a bit surprised you're behind the whole modesetting thing, given how bad our stability issues are, but ... i guess we all have diff perspectives.
23:09skeggsb: it decreases the surface area of things to maintain, which we already struggle with
23:12imirkin: from what i'm seeing, it's the one thing we have that works well. the GL bits cause hangs and crashes (both at the app level, and the kernel level).
23:12imirkin: a reasonable solution to such situations is to just tell people to nuke nouveau_dri. but if they're using modesetting, then it kills X accel as well.
23:14imirkin: anyways, if you don't count my aside into the stupid triangles thing (which is broken for all hw), getting GM10x + GM20x up and running took me a total of like 4h. imho time well spent.