03:15imirkin: skeggsb: https://people.freedesktop.org/~imirkin/patches/0001-drmmode-update-logic-for-dynamic-connectors-paths-an.patch
03:15imirkin: skeggsb: this is my porting of MST-related items
03:16imirkin: skeggsb: you said you'd be willing to give that sort of patch a testing
03:17imirkin: skeggsb: it goes without saying, but if you test it out and get any crashes, ideally you'd give me the full backtraces for those :)
03:17imirkin: [and obviously feel free to fix it up yourself, although i suspect you will not be so-inclined]
17:34AndrewR: imirkin, yor patch for disabling (hanging) h264 decoding showed up in 4.14 .. at least vdpau no more lists h264 as "supported". Otherwise nouveau works (after I fixed my .config for fbcon change - menuconfig from 4.12 unsetted fbcon from M to N - so it was surprize at reboot (no consoles))
17:35imirkin: AndrewR: cool
17:35imirkin: i have a G92 now btw (even plugged in)
17:35imirkin: but haven't been able to get it to do much
17:38AndrewR: imirkin, you also have ppc mac, yes? One with AGP? Someone had PCIE version of it ...with nvidia card , and few radeon cards ...from my remote observer perspective it sounds like much more exotic case than my ordinary x86 machine
17:38imirkin: one with AGP, yes
17:39imirkin: but the AGP is broken on macs (basically) so it's effectively PCI
17:41AndrewR: imirkin, I was wondering what kind of cache control those macs have? Like on x86 it was MTRR and/or PAT ..on macs (ppc) they should have some equivalent? Or not? There aparently was bug in kvm for amd processors,where not setting those caching modes correctly resulted in low performance and even hangs ...
17:42imirkin: sorry, no clue about POWER things
17:43AndrewR: (moment, here is my dog wanting to go out ..I will try to find something on this question after 15-20 mins)
18:07AndrewR: http://mirror.informatimago.com/next/developer.apple.com/technotes/tn/tn1008.html - but this one about old plain pci
18:08Llmiseyhaa: AndrewR: Thank you for linking that! The patch should help my wife and I. <3
18:08Llmiseyhaa: I see it's not in 4.14, and I don't think 4.13 came out after that was posted, so I'll be adding that patch to our local spins.
18:11Llmiseyhaa: (Yeah I subject her to running a Linux host with Windows guest because, I guess, we like pain. Also we're paranoid about Windows telemetry.)
18:12AndrewR: Llmiseyhaa, I have amd processor, but not yet tested kvm/qume on this kernel...will try to test both 4.14 and 4.14 + patch. Right now I have strange compilation failure for cross-compiling and want to report it ...
18:20Llmiseyhaa: Well, I had a stepping 0 Ryzen 7 and that had the segfault bug in heavy workloads (like big compiles) that would trigger in bash of all things
18:20Llmiseyhaa: But only bash being used in the compile scripts o..O Weirdness.
18:20Llmiseyhaa: But they got me a stepping 1 Ryzen 7 and that fixed me right up
18:20Llmiseyhaa: So there's a different 4.14 patch?
18:20Llmiseyhaa: I only saw that one patch that at https://patchwork.kernel.org/patch/10027525/
18:23Llmiseyhaa: Or is it just that 4.14 made changes to that file so it's got to be offset to fit? I mean it's simple so I could add it by hand if I had to (=
18:41AndrewR: Llmiseyhaa, https://lkml.org/lkml/2017/10/26/402 - patch from this thread at least applicable to 4.14
18:42Llmiseyhaa: Looks like the one I got on patchworks... yeah
18:42AndrewR: but I have different problem I just reported to LKML (cross-compilation fail)...
18:44Llmiseyhaa: Aaaah, well; can you link me to an HTTP posting of that? I can see if it's something that jumps out at me. (=
18:44Llmiseyhaa: I'm no kernel dev, but a decade long practice as a sysadmin tends to leave one with some skills on such (=
18:55AndrewR: (also, just tested qemu + old 2.4 based iso - it works, but for some reason after qemu exit kvm's device permissions reset to root-only access, so, next qemu launch fail for user - may be in new distros udev handles this, but I was just using unsecure chmod at /dev/kvm from rc.local)
19:00AndrewR: also, my qemu still doesn't have those ... https://lists.gnu.org/archive/html/qemu-devel/2017-11/msg02098.html (i use sdl2 video output). from discussion it looks like some bugs were actually on SDL2 side, so, may be I'll update this too ...self-rolling distro is fun..at times, at least i have something to distract me from depression)
19:11AndrewR: also, more on nouveau + seamonkey topic ..I have hw compositing enbled, iand it works, but setting 2d accel result in crash ... https://forums.freebsd.org/threads/59089/ - you need to add gfx.canvas.azure.accelerated = true from mozilla's about:config interface (new boolean with this name and set it to true. test on empty profile... I used emptied root's instance, for not interrupting my main browser)
19:12Llmiseyhaa: Oh wow you're more hardcore than I am.
19:13Llmiseyhaa: ok yeah no experience with the ORC unwinder yet. Sorry )=
19:13AndrewR: Llmiseyhaa, but not hardcore enough for fixing those ...
19:14AndrewR: Llmiseyhaa, kernel's help text promised better performance at expence of increased mem usage (+2-4 mb - not an issue for 12 Gb machine)
19:14Llmiseyhaa: I guess that's the computer equivalent of me beating my face against a 7M-GTE Toyota engine. Damn thing doesn't have a proper diagnostic output. So it's back to old school style carb debugging, but without the mechanical feedback of a carb. Just bought an oscilloscope so I can start watching the signal outputs. >..<
19:14Llmiseyhaa: (Signal outputs on sensors, I mean.)
19:15Llmiseyhaa: My primary hobby has switched over to the mechanical side of late, what with IT being my day job. Damn, man, wish I could help more.
19:19AndrewR: Llmiseyhaa, I tend to be much worse on mechanical matters ...even if I know (outdated) theory it will not help me with even simplest problems ...
19:20Llmiseyhaa: Practice, practice, practice. Also, having not learned an appropriate amount of fear. So I'll throw myself at stuff and figure it out as I go because I'm that woman that never learned that she should be afraid to do things she doesn't know. (=
19:23Llmiseyhaa: Man... what a zoo of computers at home... a two socket Opteron (Bulldozer v.2) system, an FX system (Bulldozer v.2 again), and a Ryzen 7 1700. All probably need the PAT fix; though the Opteron doesn't do any hardware passthrough so maybe not.
20:13AndrewR: on kvm/qemu issue: https://bugzilla.redhat.com/show_bug.cgi?id=1431876
21:37AndrewR: https://pastebin.ca/3934542 - those seems harmless ...
23:17Lyude: looks like that the clockgate settings between multiple cards of the same gen varies more then i thought it did, although I can't say I'm surprisd
23:18Lyude: e.g. like, nvf0 has different cg settings then nvf1
23:29Lyude: RSpliet: how plausable does the idea of different vbioses having different stock clockgating settings on POST soudn?
23:33RSpliet: Lyude: I can't exclude the idea, but easier to see if you can find two NVF1s with different settings
23:33RSpliet: if so,-> VBIOS (either script or tables), if not -> per-card values
23:33RSpliet: per-chip values I mean obvz
23:33Lyude: yep, that's what i've been doing so far and I think I figured out the anamoly
23:34Lyude: i was missing half of the programming sequence for CG_CTRL :)
23:34Lyude: and, the nvidia driver writing to each CG_CTRL register twice is not coincidental
23:35Lyude: looks like you program bits 0-7 first, then 8-31
23:35RSpliet: ah, although nvgpu doesn't do that presumably?
23:35RSpliet: Does NVIDIA use broadcast regs for per-GPC or per-SM regs?
23:36Lyude: what does GPC/SM stand for?
23:36RSpliet: Graphics Processor Cluster, Streaming Multiprocessor
23:37Lyude: ah, I'm not sure as I haven't looked that closely at that, I've just been doing tons of text scans through their codebase
23:37Lyude: that being said though: I wonder if there is also more I'm missing here, because I keep crashing this kepler2 card trying to enable clockgating
23:38RSpliet: clock-gating or powergating?
23:38Lyude: clock, and not even at BLCG yet
23:38Lyude: just plain CG_CTRL
23:40Lyude: RSpliet: whicb brings me to ask, I -think- they're different but I haven't confirmed yet, do we know if CG_CTRL and BLCG rely on eachother to be programmed properly
23:40Lyude: I don't believe so, but I'm just throwing possible reasons why this might be crashing to see if any of them stick
23:45RSpliet: are there dividers programmed for clock gating? Or is it "delays to completely cut"?
23:46Lyude: I think it's just delays, I haven't programmed any dividers. I may have just done something silly here though
23:46Lyude: it should be noted the entirity of enabling this level of gating is just writing register values from nvidia in the right order