04:02imirkin: RSpliet: isn't that logic used for kepler too?
04:02imirkin: or is that sddr?
11:40RSpliet: Fermi+ is GDDR5 and SDDR3 indeed
11:41RSpliet: (or... "DDR3" as most people call it)
11:59sgt-hartman: Hi, first time i come here searching for some help about nouveau drivers.
12:01sgt-hartman: Please could you tell me before if i can ask, i'm not really sure if it is an IRC chat for help or if it developper focused (don't want to bother you with my issue!)
12:03RSpliet: sgt-hartman: you're welcome here for help with nouveau, but bear in mind that we are not "distro support". The first question you're likely being asked is whether you are running the latest upstream kernel (4.20) and mesa (19.0 I think?)
12:04RSpliet: And... on Sundays it might be quiet ;-)
12:08sgt-hartman: yes, but its the perfect day for hobby unfortunatelly
12:08diogenes_: Hello guys, i've got some artefacts when playing one particular game with wine 4.0, all othe games works fine, kernel 4.19, Mesa 18.2.6
12:08diogenes_: is there any trick to enable/disable to get rid of artefacts?
12:26sgt-hartman: So, like a bottle in the sea i expose my issue. I can't make vsync to work in my setup (kernel 4.20, mesa 18.2, fedora 29). When i launch glxgears, i have an ultra fast framerate (700 fps). GLXVblank is activated in xorg.conf. I use an crt old screen (no edid) with custom 15khz modelines
12:28sgt-hartman: This configuration works some years ago. But can't say exactly since when it does not work anymore
12:30gnarface: does it need vblank_mode=1?
12:32sgt-hartman: I tried vlank_mode=1 (and 2) without changes
12:32gnarface: dual-screen setup?
12:33sgt-hartman: no, but multiseat
12:34sgt-hartman: i have two cards, one radeon and one nvidia
12:34sgt-hartman: its a workstation with pci express cards. when i invert the setup, radeon card do vsync correctly
12:38gnarface: hmm, so it's gotta be something in that one user's config then, right?
12:40sgt-hartman: i just run another try, glxgears output this when i use "vblank_mode": Attention, default value of option vblank_mode overriden by environment. Running synchronized to the vertical refresh. the framerate should be approximatively the same as monitor refresh rate
12:41sgt-hartman: hmm not sure. But i can test with only one seat to make sure it has nothing to do with that.
12:52sgt-hartman: What is strange, is when i try another monitor (a standard ASUS flat panel), vsync works correctly
12:56gnarface: hmm, i wonder if they need to match
12:57sgt-hartman: the card and the screen ?
12:59gnarface: no i meant the two screens
13:00gnarface: but earlier, before you logged off to check with only one seat, what i meant for you to check was a completely new user, like maybe the card difference is incidental, and there's some persistent local user config that's messing with vsync for that one home directory
13:00sgt-hartman: They are not connected at the same time. I just connected this standard flat screen instead of the other for testing purposes
13:00gnarface: oh hmmm
13:00gnarface: well that's strange
13:00gnarface: you're using auto-detect for horizontal and vertical refreshes though, i assume?
13:01sgt-hartman: the difference between theses two screen, is the one needs custom modelines in xorg to work, it has no edid
13:01gnarface: is that the one that won't vsync?
13:01sgt-hartman: and i force it with "video=VGA-2:e" in kernel command line
13:02sgt-hartman: but it used to works by the time
13:02gnarface: well the thing about it complaining about it being overridden by the environment is suspect
13:02sgt-hartman: i made this setup a lot of time, on ubuntu by the time, and in fedora 27 if my memory serves me right
13:02gnarface: maybe a change to the game itself caused this?
13:03sgt-hartman: the game ?
13:03gnarface: er, glxgears in this case i guess
13:04gnarface: wait did you check anything other than glxgears?
13:04gnarface: it's track record isn't actually blemish-free
13:04sgt-hartman: i'm not sure because the main issue i have is when running the emulator mame. It run way too fast. so i decided to test with glxgears to make sure i didn't have a vsync issue
13:05sgt-hartman: Mame is configured to "waitvsync". And in fact when i switch video cards, with the radeon on it run correctly. And glxgears reports a ~60fps framerate which is correct
13:06sgt-hartman: I just switch cards, and set "driver" to "radeon" in xorg
13:07sgt-hartman: (by the way, thanks for help me ^^"
13:08sgt-hartman: another info, i tried with two distincts nvidia cards
13:08sgt-hartman: with same results
13:10gnarface: does xrandr do anything?
13:10gnarface: i mean, does xrandr show the card changing settings or something? maybe it's resyncing out of vsync range somehow...
13:11gnarface: i'm sorry, i can't be very much help here
13:11sgt-hartman: yeah no worries, thank again.
13:11sgt-hartman: xrandr seems to work correctly
13:11sgt-hartman: i see the custom modeline set in xorg.conf
13:12gnarface: no complaints about it in the Xorg log?
13:12gnarface: i wonder if the syntax of some config line changed pehraps
13:13gnarface: glxgears clearly thinks it's the user environment somehow though, that's probably an important clue
13:13gnarface: usually that is a problem disabling it though, not enabling it
13:13sgt-hartman: I hoped "GLXVBlank" would have help, but in any case it seem activated by default
13:14gnarface: it could even interfere, i don't know for sure. but yea, i recall that it is expected vblank_mode to be enabled by default these days
13:14gnarface: that might have changed recently
13:15sgt-hartman: yes, by the time there were an issue of GLXVblank not activated by default, but it seems to be fixed since a long time
13:15sgt-hartman: maybe nouveau doesn't like custom modelines ?
13:15gnarface: no, that's not it, i'm using them
13:16gnarface: do you know where your Xorg log is?
13:16sgt-hartman: maybe because it's a modeline for 15khz monitors .
13:16sgt-hartman: I read xorg log via "journalctl"
13:16gnarface: oh, right, that systemd thing
13:16gnarface: well look through it for lines starting with "(EE)"
13:17gnarface: they are often clues
13:17sgt-hartman: Ok. I setup my system, reboot, then look at xorg log
13:32sgt-hartman: so, i just took a look in Xorg.log, there is no errors but theses lines repeted 10 or 20 times:
13:32sgt-hartman: (WW) NOUVEAU(0): nouveau_dri2_flip_event_handler: Pageflip has impossible msc 44843 < target_msc 44844
13:33gnarface: that looks relevant
13:33sgt-hartman: and a warning: "Warning: Unsupported high keycode 372 for name <I372> ignored" but seems irrelevant here
13:39sgt-hartman: seems a bug so.
13:40sgt-hartman: I can fill a bug ticket. What is the best place for that? fedora bug tracker ? nouveau mailing list ?
13:40gnarface: uh, the nouveau.freedesktop.org one i think
13:41gnarface: i dunno exactly but yea the nouveau one
13:41sgt-hartman: okay. Thanks for your help
13:44gnarface: sgt-hartman: wait, i'm actaully not sure maybe the fedora one actually
13:44gnarface: sgt-hartman: is there a fedora irc room?
13:48gnarface: it might be a good idea to ask them what they think
13:49gnarface: these are all the stock versions of stuff you have installed, right?
13:49sgt-hartman: yes, no custom package
13:50gnarface: yea maybe ask them about it first
13:50sgt-hartman: i'll do
13:51sgt-hartman: i'm curious, you work on nouveau drivers ?
13:52gnarface: no i don't know anything about the innards, sorry
13:52sgt-hartman: no worrie it was just out of curiosity. Thanks for help man
13:52gnarface: they're all asleep right now or something like that, if you wait around long enough they'll be back
13:53sgt-hartman: haha, and its sunday! theses guys need to rest too!
13:54sgt-hartman: need to reboot my system, i come back soon
16:20imirkin: sgt-hartman: still having the same issue?
16:20imirkin: diogenes_: make an apitrace of the issues, double-check that replaying the trace shows those artifacts, then upload the trace somewhere and let me know
16:22diogenes_: imirkin, thank you for your reply but i'm not that advanced into this so how to make and apitrace?
16:23imirkin: diogenes_: grab apitrace at https://github.com/apitrace/apitrace
16:23imirkin: and then run
16:23imirkin: apitrace trace wine fooapp.exe
16:23diogenes_: oh ok thanks, let look into iy
16:41sgt-hartman: hi imirkin, yes i have not found a solution yet
16:41diogenes_: imirkin, is it ok to use the apitracer from my package manager or i have to compile it from github?
16:49orbea: diogenes_: whichever one works
17:02diogenes_: imirkin, orbea this is the command i run: DRI_PRIME=1 apitrace trace /opt/wine/bin/wine JustCause2.exe
17:03diogenes_: this is the output i received
17:03diogenes_: i cut only to the point where is just starts repeating one and the same line
17:06imirkin: sgt-hartman: pastebin xorg log and dmesg
17:06imirkin: diogenes_: right ... you need a 32-bit apitrace
17:06imirkin: since it's a 32-bit wine application
17:07diogenes_: oh ok i'll install it
17:09imirkin: sgt-hartman: fwiw i've never heard of vsync not working so completely in modern times... i could imagine it with a Riva TNT since it was a bit funky, but you're talking about PCIe so that's out
17:15sgt-hartman: dmesg here: https://pastebin.com/nq28grWg and xorg.log there: https://pastebin.com/ae9ZpdiS
17:16imirkin: ok. GT218 on kernel 4.20. why do you do VGA-2:e?
17:17imirkin: that seems like the sort of thing that woulc cause your troubles
17:17sgt-hartman: i need it because my monitor is an old school 15Khz screen without edid
17:17imirkin: shouldn't matter
17:17imirkin: but it matters for console?
17:18sgt-hartman: i know its kind of exotic setup! but it used to work
17:18imirkin: do you use custom modelines, or default-ish ones?
17:18sgt-hartman: yes i use it with groovymame, a patched mame (emulator) that change resolutions on the fly
17:18sgt-hartman: i have setup a custom modeline right
17:18sgt-hartman: i copy-paste my xorg.conf..
17:19imirkin: ok, but that's much later.... when the kernel is booting, do you see console output after nouveau loads?
17:20imirkin: also, you mention this used to work ... what kernel did this work on?
17:20sgt-hartman: i can't tell which kernel worked the last time
17:20sgt-hartman: i made this setup last time maybe 1 year ago
17:21sgt-hartman: it was on fedora 27 , and worked well on ubuntu too
17:21imirkin: so ... 1-year-ago kernel worked ok?
17:21sgt-hartman: maybe yes, my memory is not really precise
17:21sgt-hartman: yes i see kernel boot logs, but my screen is not able to display them correctly (31khz)
17:22imirkin: right ok
17:22imirkin: HorizSync 15-50
17:22sgt-hartman: video=VGA-2:e is required, without that the system believe there is no monitor connected
17:22imirkin: so that's actually more like 15-16 or whatever
17:23sgt-hartman: yeah, its an old monitor, like a tv
17:23sgt-hartman: CRT TV set
17:24sgt-hartman: the same setup with a radeon card works well
17:24imirkin: do you have to force-enable there too?
17:24sgt-hartman: (i just change "driver: radeon" in xorg.conf
17:24sgt-hartman: with the radeon no. I don't know why but the connector is activated by default
17:24imirkin: i suspect the two are related.
17:25sgt-hartman: possible yes
17:25imirkin: here would be my recommendation of something to try
17:25imirkin: compose an edid that is appropriate for your monitor, i.e. contains the right modeline
17:25imirkin: and tell your kernel to load it
17:25imirkin: for that VGA-2 connector
17:25imirkin: instead of VGA-2:e
17:26sgt-hartman: unfortunatelly, i tried that but i can't make the system to load my edid firmware
17:26imirkin: how did you try to do it?
17:27sgt-hartman: i opened a bug in fedora tracker about that:
17:27sgt-hartman: basically i use this kernel argument: drm.edid_firmware=VGA-1:edid/edid_generic15.bin
17:27imirkin: not VGA-2?
17:27diogenes_: imirkin, here: http://dpaste.com/3JBPGGG
17:28sgt-hartman: the setup was not exactly the same (just copy pasted from my bug entry)
17:28imirkin: diogenes_: great, now try replaying wine-preloader.1.trace (glretrace) and see if the issue persists
17:28sgt-hartman: i thinked it was unrelated
17:28sgt-hartman: two distinct issues..
17:28imirkin: and was the edid file available at the time the drivers loaded?
17:29imirkin: i.e. most likely in the initrd for silly systems like fedora?
17:29diogenes_: imirkin, you eman instead of wine to use wine-preloader?
17:29imirkin: diogenes_: no
17:29imirkin: look in the log
17:29imirkin: it says where it dumped the trace
17:29imirkin: try replaying that trace with glretrace
17:29diogenes_: oh ok
17:30sgt-hartman: imirkin: the edid firmware works well with the radeon, but using nouveau it seems ignored
17:31imirkin: i think you're doing something wrong
17:31imirkin: and radeon does something different enough that it makes up for the wrongness
17:31imirkin: look at it -- platform VGA-1: Direct firmware load for edid/edid_generic15.bin failed with error -2
17:31imirkin: that means the firwmare ain't there
17:32imirkin: i suspect radeon tries again later, while nouveau doesn't
17:32sgt-hartman: yeah i noticed that, but despite this it works with radeon. Boot entries a correctly displayed and i see the correct modeline with xrandr
17:32sgt-hartman: oh ok
17:32imirkin: so ... let me ask again ... where is this edid_generic15.bin placed?
17:32sgt-hartman: in /lib/firmware/edid
17:32imirkin: and did you run the thing to rebuild the initrd after you did that?
17:33diogenes_: imirkin, great, the glitch is not present with glretrace
17:33sgt-hartman: hmm, i don't think so. The only thing i do, when i update /etc/default/grub is to launch "grub2-mkconfig -o /boot/grub2/grub.cfg"
17:34imirkin: sgt-hartman: yeah, i think there's something else you need to run
17:34imirkin: gen-initramfs? i don't know. i don't use those distros ... too complex for me.
17:34sgt-hartman: yeah, it recall me something..
17:34sgt-hartman: haha. Ok i'll check for fedora the good command
17:34imirkin: i stick to gentoo where everything's nice and simple.
17:35diogenes_: imirkin, so what's next? why the artifacts are not present when i tun glretrace? is there a fix?
17:36imirkin: diogenes_: there's always a fix, but this means that you can't communicate to me (yet) what causes the issues
17:36imirkin: if replaying the trace reproduced the problems, then you'd just send me the trace for debugging
17:36sgt-hartman: i think its : "dracut --regenerate-all"
17:36imirkin: sure, why not
17:36imirkin: that's pretty intuitive.
17:37sgt-hartman: i setup the edid and reboot, and come back
17:37sgt-hartman: thanks for your help :)
17:38diogenes_: imirkin, thank you for guiding me through, you're great!
17:38imirkin: don't thank me 'til the issue's fixed...
17:38imirkin: diogenes_: which GPU do you have?
17:39diogenes_: NVIDIA GK107M [GeForce GT 650M]
17:40imirkin: so much for my "let's blame maxwell" strategy...
17:41diogenes_: hehe, it's not that huge of a deal, at least i can play the game, it flickers from time to time but i think i can handle it, if there's nothing to do about it, in any case i'm not going to nvidia proprietary
17:42imirkin: you coudl try messing with clocks
17:42imirkin: are you reclocking already?
17:42imirkin: if not, you probably should anwyays
17:42diogenes_: i use nouveau.config=NvClkMode=15
17:42diogenes_: is that it?
17:43diogenes_: it has very cool performance though
17:43diogenes_: just like with proprietary nvidia
17:43imirkin: surprising. but i'll take it.
17:43imirkin: normally we're 20-40% behind
17:43diogenes_: i make a furmark and difference is only in 4 fps comparing to nvidia proprietary
17:43diogenes_: of course with nouveau.config=NvClkMode=15
17:44imirkin: out of? 8? :)
17:44diogenes_: hehe no, out of 800+
17:44imirkin: perhaps karol has access to Just Cause 2
17:45imirkin: do you know who publishes it? if it's Feral or something, i might be able to get it free
17:45diogenes_: it's swuare enix
17:46imirkin: k, never heard of them, probably no policy for mesa developer access
17:46diogenes_: so you guys are doing a great job with nouveau, fantastic, is there any plans to teach nouveau how to switch through power states 5, 10, 15 automatically? depending on the workload?
17:46imirkin: (and it's not a linux port, so nevermind)
17:46imirkin: plans? sure.
17:46imirkin: by the year 2525... :)
17:47imirkin: afaik no one is really working on it. karol had made some progress towards it, but it stalled
17:47diogenes_: imirkin, actually this particular game didn't work on wine till 4.0, i was strugling to make it work for 4 years :) and this game was the only reason i kept windows around but not anymore
17:47imirkin: one thing is that if we're 99.9% reliable, that's pretty good when you switch once a day
17:47imirkin: no so much if you switch once a minute
17:48imirkin: iirc we still get flicker on switch
17:48imirkin: not relevant if you don't have any displays hanging off the gpu, of course
17:49diogenes_: imirkin, or if not automatically then maybe on-the-fly manually but with some shortcuts like DRI_PRIME=1 -o 15 instead of appending it to grub?
17:49imirkin: perhaps we could flip it on automatically if you have no displays, and put the pedal to the metal if you do? dunno.
17:49imirkin: you can do it on-the-fly manually now already
17:49imirkin: echo stuff to /sys/kernel/debug/dri/1/pstate
17:50diogenes_: no displays, only the notebook one
17:50imirkin: (only while a game is running tho)
17:50imirkin: karol has patches for that too :)
17:50imirkin: (what *doesn't* he have patches for...)
17:50diogenes_: so i start the game and i run: echo stuff to /sys/kernel/debug/dri/1/pstate
17:51diogenes_: or i can make a custom shortcut Exec = command && echo 15 > /sys/kernel/debug/dri/1/pstate
17:51imirkin: echo 0f
17:51imirkin: but yeah'
17:52sgt-hartman: no luck :(
17:52diogenes_: hmm i'll try that, thank you, dinner time :) have agreat day/evening ahead!
17:52imirkin: sgt-hartman: ok. unfortunately i have to do some other stuff now, so i can't look at it. but keep hanging out in here. might be able to have another look on monday. or someone more competent could come along.
17:52sgt-hartman: i tried one time with "drm.edid_firmware=VGA-2:edid/edid_generic15.bin". The screen stay black
17:53sgt-hartman: Ok, thank you so much for your help. I'll try to find more informations about initramfs regeneration.
17:54sgt-hartman: maybe the issue is around
20:47rhyskidd: skeggsb: do you have any docs on INIT_GENERIC_CONDITION condition codes 0x08, 0x09 and 0x0a?
20:47rhyskidd: i've seen each with at least pascal+
20:49rhyskidd: maybe just a name that we could read-the-tea-leaves from?
20:58RSpliet: sgt-hartman: I think it's simply "dracut --force"
20:59RSpliet: But you might want to look into which files it picks to include in your initramfs. It might simply not selecting your custom EDID file
22:39rhyskidd: skeggsb: i sent a patch fix for '2642e0b5 gr/gf100-: expose fecs methods for pausing ctxsw' from your linux-5.1 branch to the mailing list
22:44rhyskidd: i'm likely to have spotty access to internet for about a week, but that was the only comment i had from going through the pending patches for linux-5.1