04:22mlankhorst: -12 = ENOMEM, something allocating a negative number? :P
04:23mlankhorst: or maybe just enable more nouveau debug spam
06:23imirkin: mlankhorst: you mean the bug i mentioned? the vbios stuff was just a distraction.
06:24imirkin: mlankhorst: the main issue is that PDISP hangs
12:36Smilex: I get an unhandled syscall 319 from a call to Ioctl, using the git sources. Does anyone here know where nouveau or drm set such handlers?
12:40imirkin_: Smilex: where do you see this message?
12:42imirkin_: tbh i'm not sure what it means
12:46Smilex: imirkin_: when calling valgrind on my code
12:46Smilex: also it won't create the gl context
12:46imirkin_: Smilex: oh, that makes more sense. i assume you just need a newer valgrind...
12:47imirkin_: let me see if i can find the syscall map...
12:48Smilex: imirkin_: I have valgrind 3.10
12:49imirkin_: hm, same version i have
12:50imirkin_: aha, found 'em!
12:50imirkin_: 319 common memfd_create sys_memfd_create
12:50imirkin_: (assuming x86_64)
12:50imirkin_: iirc systemd uses memfd for something
12:51Smilex: imirkin_: I don't have systemd
12:52Smilex: or I shouldn't have
12:52mlankhorst: newer glibc uses memfd_create too
12:52imirkin_: mlankhorst: for what?
12:52imirkin_: [tbh i dunno what memfd is... same thing as pipe but fancier?]
12:53mlankhorst: for shm_create probably
12:53mlankhorst: shm_open I mean
12:53Smilex: but the stack trace just mentions nouveau and libdrm
12:53imirkin_: which probably calls shm_open?
12:54mlankhorst: implement it by returning -ENOSYS :-)
12:55mlankhorst: easy to add syscalls, newer valgrind might already have it
12:55Smilex: imirkin_: grep didn't find shm_open in xf86-video-nouveau or drm source
12:55Smilex: mlankhorst: is that something I should add in my code?
12:56mlankhorst: you can add a break on memfd_create or sys_mem_fd_create with gdb
12:56mlankhorst: see what shows up
12:57imirkin_: or just look at the backtrace, it should tell you exactly what's invoking that sysclal
12:58mlankhorst: i was assuming it didnt get a backtrace, but yeah I guess because its valgrind it ought..
12:58imirkin_: btw, if it's a 32-bit app, then syscall 319 is: 319 i386 epoll_pwait sys_epoll_pwait
12:58imirkin_: which is much less new
13:05Smilex: imirkin_: I have multilib off
13:08Smilex: imirkin_: the backtrace is garbled
13:08Smilex: mlankhorst: I did "break memfd_create" and nothing
13:14Smilex: trying to compile with -ggdb to maybe ungarble the stack trace
13:14imirkin_: Smilex: in any case, whatever it ends up being, the answer will be "update valgrind"
13:14imirkin_: or "downgrade library X"
13:16Smilex: imirkin_: but then it isn't why it's failing to create a GL context?
13:17imirkin_: no. it's just valgrind ailing
13:17imirkin_: or you have a *wildly* miscompiled system
13:17imirkin_: assuming that's not the case, pastebin dmesg, xorg log, and glxinfo output
13:18Smilex: imirkin_: there are linker errors in glxinfo.c
13:18Smilex: I guess I could stop with the git addiction
13:18imirkin_: ... depending on the exact errors/etc, they're probably expected and not an issue
13:20Smilex: imirkin_: no they are unresolved symbols
13:20imirkin_: again... probably not an issue ;)
13:20imirkin_: anyways, if you want help, pastebin those things.
13:21Smilex: imirkin_: well it won't compile, meaning I don't have the program
13:21imirkin_: i thought you meant at runtime it was complaining about missing __driSomethingExtensions_bla
13:21imirkin_: skip the glxinfo for now... pastebin the other things
13:22imirkin_: let's make sure that nouveau is all happy
13:24Smilex: http://sprunge.us/WNjI <- dmesg
13:25Smilex: http://sprunge.us/YWSB <- Xorg.0.log
13:26Smilex: imirkin_: also isn't nouveau happy when it can render a browser? Or am I wrong in assuming such things
13:27imirkin_: sure, but that might also work when you're not using nouveua at all
13:28imirkin_: alrighty, 2x G92, xorg seems happy
13:29tobijk: i dont see a problem in these logs...
13:29imirkin_: glx seems happy too
13:29Smilex: yeah. I heard Mesa has no support for SLI, but I have been toying with the idea of getting it to work myself
13:29Smilex: so I just installed it
13:29imirkin_: Smilex: don't want to overly discourage you, but that would be quite an undertaking
13:30Smilex: imirkin_: Don't worry. I am already too discouraged to try. But at least now I don't have the hardware excuse
13:30Smilex: but back to the issue. I guess I should try some other software
13:30Smilex: maybe not based on SDL2
13:30imirkin_: so... you mentioned you failed at compiling glxinfo?
13:31imirkin_: what was the failure?
13:31glennk: a better project would be using dri/GL on several devices for some project
13:31imirkin_: glennk: ?
13:31glennk: ie app driven SLI rather than trying to do it in the driver, this is the direction DX12 and vulkan are going in anyway
13:31imirkin_: oh, yeah.
13:31imirkin_: that's why SLI is tough -- it's unclear wtf to do :)
13:32Smilex: imirkin_: http://sprunge.us/YOGO
13:32glennk: yeah, especially since for the blob drivers sli support is something thats implemented per category of app
13:32imirkin_: Smilex: comes up empty...
13:33imirkin_: the hw has funny sli support though... there's a concept of slave and master
13:33imirkin_: and you can send commands in the fifo which only execute on slave or only on master (special header bit)
13:33Smilex: imirkin_: http://sprunge.us/YeNa
13:33glennk: thats basically where the nv conditional rendering extension came from btw
13:34Smilex: glennk: but then I'l have to wait for Vulkan, right?
13:34imirkin_: Smilex: use the non-9999 version of mesa-progs maybe?
13:35glennk: you can use GL, the bits are (mostly) there to do it with dri/mesa from the app level
13:35imirkin_: glennk: conditional rendering is a little different, and there are different bits to enable that
13:35Smilex: glennk: Oh so directly using mesa somehow? So it will be driver specific
13:35imirkin_: (not at all in the commands themselves)
13:35Smilex: imirkin_: you don't happen to know how to disable the -9999 mask on gentoo?
13:36imirkin_: it should be masked by default ;)
13:36glennk: dri specific, but could work on intel/nouveau/radeon
13:36imirkin_: you can stick it into packages.mask
13:36imirkin_: e.g. =x11-apps/mesa-progs-9999
13:36imirkin_: and let your unmask and mask things fight ;)
13:36glennk: basically open separate drm devices and use the PRIME mechanisms for sharing buffers
13:39Smilex: imirkin_: http://sprunge.us/ZJUP <- glxinfo
13:40imirkin_: Smilex: build mesa with -bindist
13:40imirkin_: Smilex: and before you run into it, also install libtxc_dxtn
13:42Smilex: imirkin_: what does -bindist do?
13:42imirkin_: equery u mesa
13:42imirkin_: basically enables GL3+
13:42Smilex: imirkin_: so now I get GL3+?
13:43imirkin_: when you rebuild, yeah
13:44Smilex: good, I use that
13:46imirkin_: Smilex: btw, why are you doing cpu=k8??
13:46imirkin_: you appear to have a haswell
13:46Smilex: imirkin_: I just made a wild guess
13:46Smilex: imirkin_: what should I be using?
13:48Smilex: same for -mtune
13:48imirkin_: kill that param entirely
13:48imirkin_: -march implies mtune
13:50imirkin_: gnurou: any idea if my shader header question is likely to get answered?
13:51Smilex: imirkin_: it worked!
13:51Smilex: you guys rock
13:52imirkin_: oh, just a little heads-up -- a bunch of people have reported that VDPAU hangs their G92's
13:52Smilex: also should I now rebuild the entire system since I changed -march?
13:52imirkin_: i doubt there's that much of a difference
13:53Smilex: is this something recent?
13:54imirkin_: but you need the blob video decoding firmware for vdpau to work at all in the first place
13:54Smilex: well I haven't had the issue
13:54Smilex: I only use that when I decide to go on a gaming binge
13:55imirkin_: blob video decoding firmware?
13:55imirkin_: it's for watching movies...
13:55imirkin_: (or other video content)
13:57Smilex: I mean I only use the blob when going on a gaming binge
13:57Smilex: reading up on vdpau I saw that VLC uses it, so I might encounter the issue at some point
14:09imirkin_: Smilex: not if you don't have the extracted blob video decoding firmware
14:25tetracon: hi I've encountered a problem with my graphics card. As it seems it contains an unknown Maxwell chipset
14:25imirkin_: tetracon: 0x118?
14:25imirkin_: yep, 0x118. aka GM108
14:26tetracon: yeah exactly :)
14:26imirkin_: tetracon: https://bugs.freedesktop.org/show_bug.cgi?id=89558
14:26tetracon: ah I see thnx
18:18neonrooster: I have two video cards, one Intel HD 4600 and one GeForce 7300 LE. I plugged one DVI monitor into the Intel and the other DVI monitor into the Nvidia. I have xf86-video-intel and xf86-video-nouveau installed. However, the Nvidia display won't appear!
18:18neonrooster: any idea how to make it work?
18:20imirkin: neonrooster: take a look at the "using outputs on discrete GPU" section of http://nouveau.freedesktop.org/wiki/Optimus/
19:51nutron: Hi guys, I'm not sure if this is an X problem or a nouveau one. I am attempting to rotate 3 screens, and upon invoking the xrandr command to do so. It segfaults. Is this a FAQ or do I need to update things somehow?
19:52imirkin: nutron: are you the guy who filed a bug about it earlier today?
19:52imirkin: nutron: either way, are you using glamor? if so, don't
19:52nutron: No, that wasn't me.
19:52nutron: Don't even know what glamor is.
19:52imirkin: pastebin your xorg log
19:54nutron: imirkin: http://paste.debian.net/171475/
19:56imirkin: i see.
19:56imirkin: 2x GF116
19:56imirkin: and the third screen is in the 2nd one?
19:56nutron: omg... what does this do on the commandline? I'm sure I thought it was a good idea =) nouveau.config=PCE1=0
19:56nutron: imirkin: correct.
19:56nutron: if I don't rotate... everything is hunky dory.
19:56imirkin: nutron: that disables the 2nd copy engine, which GF116 has. recent kernels should know about that, but i think that 3.16 didn't
19:56nutron: has been working fine this way for a long time.
19:57imirkin: er. i meant to say "which GF116 doesn't have"
19:57imirkin: ooh, fun! the crash is in generic Xorg code
19:58imirkin: airlied: can reverse prime outputs be rotated?
19:58imirkin: coz i got a backtrace right here that seems to indicate that they can't ;)
19:58nutron: gotcha. I use xrandr --setprovideroutputsource 1 0 then fire it up with this command.... oh a bug then =)
19:59imirkin: nutron: this isn't a new issue, right? (i mean, it's not something that worked "before" and now doesn't?)
20:00nutron: no, I've worked this config in landscape mode for many months. Now I got the bright idea to rotate all screens and here I am.
20:00nutron: Same command except I changed all --rotation normal to --rotation left
20:00imirkin: there's no reason in theory why it shouldn't work
20:01nutron: works with 2 screens on the primary. As soon as I fire up the third. I see the segfault.
20:01imirkin: and yet here we are :)
20:01nutron: Heh. Mysterious. Even magical. I love bugs like this....
20:02airlied: imirkin: no
20:03imirkin: airlied: any good reason, or just lots of bad ones?
20:03nutron: I did find a forum post about this back from 2013 though... https://bbs.archlinux.org/viewtopic.php?pid=1359971#p1359971
20:03imirkin: nutron: sounds like zaphodheads + xinerama might be the only way for you to do it for now =/
20:03imirkin: nutron: that would mean you'd lose direct rendering, in case that's important to you
20:04nutron: Omg... I forget how to do that heh it's been oh so long....
20:04imirkin: nutron: http://nouveau.freedesktop.org/wiki/MultiMonitorDesktop/
20:04nutron: Is this X or nouveau being a bitch? I'd like to provide as much info about the bug as I can before I walk away from this config.
20:05imirkin: sounds like Xorg :(
20:05nutron: Where can I look for a bug report if it's X related?
20:05imirkin: that will only find bugs people filed
20:06imirkin: not ones people know about ;)
20:06nutron: heh, well I figure it's old and I can't be the only one of two peeps running into this....
20:06imirkin: but yeah, bugs.fd.o would be the bugzilla of record
20:06airlied: imirkin: tried it once, failed, never got back to it
20:06nutron: I'll go poke around.
20:06nutron: imirkin: thanks for your help.
20:07airlied: same as fixing sw cursor
20:07imirkin: airlied: i'm surprised you haven't had "enterprise" customers asking you guys for it that would merit you or someone else spending a few days on it
20:08airlied: it also likely need an X server abi break, which makes things more annoying
20:10imirkin: basically everything i ever ask for seems to require an X server ABI break
20:11imirkin: otoh it seems like the x server ABI is broken all the time, so...
20:11airlied: neither the cursor nor the rotation code are the easiest to understand
20:11imirkin: (at least i have to rebuild drivers on every X upgrade)
20:12airlied: it breaks, but for RHEL customers it takes a lot longer to cycle out the breaks, so I usually have really bad timing on when to fix things
20:13nutron: imirkin: I see airlied is talking about the cursor.... would this be related at all?
20:13nutron: err... I mean this: https://bugs.freedesktop.org/show_bug.cgi?id=26108
20:14imirkin: oooh, is swcursor and rotate fail realted? that'd be awesome
20:14imirkin: no, that was a diff crash
20:14imirkin: and certainly fixed by now
20:15imirkin: note that it's from 2010
20:15imirkin: and i courtesy-closed it in 2013 :)
20:15imirkin: and it's now ...
20:15nutron: Indeed heh... though like I said. That forum post was 2013 =)
20:16imirkin: either way, that bug looked like a regular nouveau driver bug
20:16imirkin: from the ancient 0.16 days
20:17nutron: Oh Looks like I found it.
20:18imirkin: yep, that looks like it
20:20imirkin: or rather, it looks like it got hijacked by someone
20:20imirkin: originally it was crashing when just using reverse prime at all
20:21nutron: imirkin: think I could get xrandr to start a seperate screen on the third monitor as a stop-gap?
20:21imirkin: not how xrandr works
20:21nutron: although I have no idea how that'd work since xrandr doesn't list the provider until I render from first to second...
20:22nutron: Ok, I think I'm out of my league on this...
20:31imirkin: nutron: just add 2 device/2 screen sections in your xorg.conf and you're good to go
20:31imirkin: note that most wm's have lost the ability to deal with multiple screens though