00:06imirkin: skeggsb: so do you want me to do anything else on those hdmi2 patches, or are they good to go?
00:06imirkin: not sure when i'll get to the other hdmi2 features, they're somewhat orthogonal to the speed thing. although they will affect how the clock is calculated (more dp-like)
00:09imirkin: in other news, i've secured a GTX 960 with DP, so i'll be able to play with DP-MST at some point for xf86-video-nouveau ... finally.
17:21Loeb: I seem to be having issues with my desktop completely hanging about ~30s after login. Error didn't happen when I was on the nvidia driver stack, but I also couldn't get it to use my second card.
17:21Loeb: Running two k620s. Desktop use, tried on fedora xfce and xubuntu
17:22Loeb: When it hangs I can still move my cursor but can't switch TTY screens
17:24karolherbst: Loeb: mhh, maybe you could ssh into the machine and see if something is inside the log?
17:24karolherbst: or dmesg?
17:25karolherbst: or maybe even some dead processes running
17:26Loeb: No ssh, trying to get into tty1 before the hang
17:27karolherbst: Loeb: maybe you could delay a shutdown by 2-3 minutes and wait for the machine to shutdown
17:27karolherbst: and then just check the log from last boot?
17:27karolherbst: "journalctl --dmesg --boot -1" is one way to fetch those after rebooting
17:28Loeb: on a "successful" boot where I got into the tty there's some errors from nouvaeu in here, one for "bus" on each card, one for "gr" on each card and some fifo errors on the first card
17:29Loeb: I think the times that it hung, the last dmesg errors were the "gr" errors, but I believe those were before the machine actually hung. Let me get it to error again.
17:29karolherbst: so it seems like some engines start to hang for whatever reasons
17:29karolherbst: maybe some process names?
17:29karolherbst: or the usual Xorg?
17:30Loeb: Process names?
17:30karolherbst: Loeb: what you can do is to verify that the nouveau DDX is getting used instead of modesetting
17:30karolherbst: check the Xorg log
17:30Loeb: This is just a vanilla xubuntu install right now, in terms of things being ran
17:30karolherbst: no idea what runs there
17:31karolherbst: we know that we have some issues with glamor, so best is not to use glamor
17:31Loeb: no clue what that is
17:31karolherbst: and this is usually done by using the nouveau DDX
17:31karolherbst: yeah, doesn't matter that much. Just a way how to do 2D acceleration by using OpenGL
17:31karolherbst: nouveau ain't that great at this at the moment if you do too much opengl at once and in multiple threads
17:31karolherbst: might be the cause here, maybe isn't
17:32Loeb: Not sure if xfce uses openGL or not.
17:32karolherbst: doesn't matter
17:32karolherbst: it is X itself
17:32karolherbst: what version of ubuntu?
17:32Loeb: So looking back at a crash there were some fifo errors on the primary card but those were at least 15s before the system hung
17:32karolherbst: installing "xserver-xorg-video-nouveau" might help if it isn't
17:33karolherbst: but then again, I have no idea what kind of preference is set in ubuntu and if they always load modesetting
17:33Loeb: that package is already installed
17:33Loeb: Where would the modsetting be?
17:33karolherbst: best is to fetch the /var/log/Xorg.0.log file
17:33karolherbst: and check there or paste it on some site
17:33karolherbst: modesetting is installed alongside the xorg server
17:33karolherbst: so no way to remove it
17:34karolherbst: anyway, the /var/log/Xorg.0.log file might help here
17:34Loeb: That log or the .old from previous boot?
17:34karolherbst: should be the same
17:35karolherbst: well - some errors, but the normal content should be the same
17:35Loeb: the previous boot ends on some modelines
17:35Loeb: again before system hung
17:35Loeb: current boot has some more stuff
17:35Loeb: I'm looking at it from tty right now
17:35karolherbst: is there more of NOUVEAU: or more of modeset:?
17:37karolherbst: anyway, I kind of plan to finally work on those issues in the near future. It just isn't trivial as it is more of a core issue we have
17:37karolherbst: currently I want to properly handle those hangs and freezes and let the applications know so they can abort or repair the state or whatever
17:38karolherbst: but yeah, won't help you today
17:38Loeb: haha I jumped back into the desktop from tty and it hung :(
17:38Loeb: I just want desktop outputs on both cards, the nvidia driver wasn't using the second card so I was hoping jumping back to nouveau would be easier
17:38karolherbst: mhh, second card as in two nvidia GPUs?
17:38Loeb: I haven't tried a DE other than XFCE yet, originally thought it was a fedora issue.
17:39Loeb: Yes, two K620 cards
17:39Loeb: System needs to be able to drive 6 displays
17:39karolherbst: mhh, there is some special thing you can use, but that has some other drawbacks on nvidia :/
17:39Loeb: currently have 4 hooked up
17:39Loeb: Based on what I looked up, nvidia wants you to use their xorgconfig tool to add settings to the xorg.conf automatically, but doing that broke lots of things
17:40karolherbst: yeah, something like that
17:40karolherbst: with nouveau the situation isn't that much better as you start pushing the pcie bandwidth limits, especially with 6 screens
17:41Loeb: They're only 1080p screens, I wouldn't think that would be a huge issue
17:42karolherbst: do the math, it kind of is
17:42Loeb: PCIe 16x... at least 2.0, maybe 3.0. Not sure what the cards support
17:43karolherbst: 1.4 GB/s are needed for the raw data itself
17:44karolherbst: which, with disabled reclocking, 2.5 PCIe gives you 4.0 GB/s
17:44karolherbst: - all the other overhead we have
17:44karolherbst: we currently have some issues here, and doing a 4k screen is sometimes too much already
17:44karolherbst: not nouveau itself
17:44karolherbst: but just how we do that inside linux
17:45Loeb: yeah 4k should be roughly 4x the bandwidth of 1080p
17:45karolherbst: anyway, people reported that higher clocks/pcie bandwidth makes such setups more smooth
17:45Loeb: but I'm pushing 3 of those per card/slot
17:45karolherbst: but you have 3x 1080p ;)
17:45karolherbst: nothing really tested
17:45karolherbst: and something we usually don't get any reports
17:45karolherbst: but basically everytime somebody tries it, it gets quite slow
17:46karolherbst: (I tested a 4k screen in such a setup once, 4k connected through nouveau, main GPU was intel)
17:46karolherbst: was a bit painful
17:46karolherbst: but at least on your GPUs you would be able to set the clock levels/pcie link speed, so that might just work
17:46karolherbst: anyway, coming back to the crash, it might be triggered by glamor, which is used with the modesetting X driver
17:47Loeb: are they running at a lower speed by default for some reason?
17:47karolherbst: yeah, reclocking isn't that stable
17:47karolherbst: it could just break on some machines
17:47karolherbst: as it can be different on every GPU
17:47karolherbst: so users have to change clocks manually
17:47Loeb: Er, is this core GPU speed or PCIe bus speed?
17:47karolherbst: it is kind of the same thing in the end or changed within the same process
17:47karolherbst: nvidia GPUs tend to boot at the 2.5 PCIe speeds
17:48Loeb: Sure enough it's not crashing with 1 display connected -.-
17:48karolherbst: or are getting initliaized by some vbios scripts
17:48karolherbst: mhh, fun
17:49karolherbst: Loeb: anyway, if you are able to fetch some meaningful logs, might be worth to explain the situation and create a bug report on the freedesktop bugzilla
17:49karolherbst: or link them here
17:51Loeb: looks like with one monitor the fifo errors didn't happen
17:51Loeb: don't have a good way of getting the logs off of here right now
17:52Loeb: gonna poke around and see if I can make the proprietary driver work any better
17:53karolherbst: yeah, good luck with that
18:48Loeb: Got it sorted, nvidia's tools for making modifications to xorg.conf are awful
21:55pendingchaos: imirkin: how would you recommend I test xf86-video-nouveau? specifically it's Maxwell shaders
22:46Lyude: imirkin: still need me to test taht stuff you were talking about over the weekend?