04:10rhyskidd: any reviewers for TU106 / RTX 2070 envytools support? https://github.com/envytools/envytools/pull/173
04:11rhyskidd: the travis-ci warning is incorrect, the ci was having backend problems yesterday
04:14rhyskidd: imirkin: thanks
04:17rhyskidd: i've got a few other small tweaks and doc fixes coming, cleaning up patches from code written on the flight
04:17rhyskidd: little nits that had been annoying for a while. some dedicated time let them be worked on
04:29imirkin: skeggsb: any idea about https://bugs.freedesktop.org/show_bug.cgi?id=108500 ? very odd...
04:29imirkin: looks like MCP89 uses the same mmu as g84
04:48imirkin: skeggsb: looks like nv50_vmm_valid will reject compressed allocations in stolen vram? i don't think that's right...
04:52imirkin: skeggsb: also i don't see anywhere where mmu->func->vmm.global might get set
07:19rhyskidd: karolherbst,skeggsb: any comments on this Pascal, Volta and Turing compute wip series? https://github.com/envytools/envytools/compare/master...Echelon9:feature/compute-improvements
07:20rhyskidd: i'm also going back and adding some Maxwell and earlier compute methods, which will be a better base for the volta and turing changes (eg some undocumented methods are now deprecated)
07:28masterboy: Hi guys, could you help me - the driver hangs my system after minimal use. I am using NV4C here is the dmesg https://paste.gnome.org/pqgx9isye
07:28masterboy: i set aside 256mb memory for it
07:28masterboy: hi gnarface :) yes
07:28masterboy: ubuntu shows me that it is nv4c :)
07:28masterboy: the driver was unusable with 64mb of shared ram but 256 makes it work for a while...
07:29masterboy: i updated the bios to the latest 2006 one :)
07:31masterboy: tell me if any other logs should i provide :)
07:33masterboy: it is a amd cpu with nvidia chipset-integrated gpu
07:33masterboy: athlon II 2.9ghz ;)
07:35masterboy: btw i am using this system now :) It seems if i load more programs the gpu will crash
07:41masterboy: any idea what those errors mean?
07:43gnarface: you mean this? nouveau 0000:00:0d.0: bus: MMIO write of 010a0001 FAULT at 00b010
07:43masterboy: yes :)
07:43gnarface: got a picture of the screen corruption?
07:44gnarface: i don't know if it's bad software or bad hardware, but you can sometimes tell by the type of screen corruption
07:44masterboy: I can photo it but I would call it snow. I can draw it too :)
07:45gnarface: certain types of screen corruption are indicative of bad ram. does it look like letter glyphs are corrupted, even on the bios or POST boot screen?
07:45gnarface: if it's a problem with the driver, it won't be able to affect the bios screen or the early boot screens
07:45masterboy: ok i will hang the system and reboot and show the picture. no everything is fine, even now. until i load some more programs
07:46gnarface: bad ram would also fit that pattern
07:46gnarface: it won't all go bad at once
07:46gnarface: just a little bit
07:46gnarface: so the quickest way to find the bad spot is to use it all
07:46masterboy: i could try memtest in grub ubuntu?
07:47gnarface: that would test the system dram but i don't know if it's complaining about video ram here or not
07:47gnarface: and memtest won't test the video ram
07:47gnarface: so it'd be useless for that
07:47gnarface: still, couldn't hurt
07:47masterboy: gnarface, it is an integrated gpu it has only ddr2 ram
07:47gnarface: ooooh, right
07:48gnarface: intel integrated gpus do that, but i'm not actually sure the nvidia ones all do
07:48gnarface: it might have it's own ram
07:48gnarface: either way, it won't hurt to run memtest on it unless it's already damaged
07:49masterboy: it does not - i can see the total ram is 256 less :)
07:49masterboy: thanks i will try
07:49masterboy: and get back to you :)
08:42masterboy: hi again i put the pictures here https://drive.google.com/open?id=1tdmM-WKD69umDdvdqTTAUI03zQSVsJL_
08:42masterboy: i hope the link works gnarface :)
08:42masterboy: the ram is fine according to memtest
08:44gnarface: are you sure it doesn't have like maybe 1 or 2MB dedicated ram?
08:45masterboy: i will google it now. What about the pictures?
08:45gnarface: i just realized, memtest probably wouldn't test the part of the ram that is dedicated to the video card in the bios
08:46gnarface: you'd have a better chance of memtest finding a bad spot if you set the gpu to use less ram i think...
08:46gnarface: because the first picture definitely looks like bad ram
08:46gnarface: second one does too really
08:47gnarface: but i think in theory even the compositor or window manager could corrupt the screen this way
08:47gnarface: i don't really know how to know for sure without seeing a bios screen also corrupted
08:47masterboy: bios screen is fine
08:47masterboy: i set 32 video ram and tested then it is ok
08:47masterboy: [ 1.906130] nouveau 0000:00:0d.0: NVIDIA C61 (04c000a2)
08:47masterboy: [ 1.915641] nouveau 0000:00:0d.0: bios: version 05.61.32.19.04
08:47masterboy: [ 1.916739] nouveau 0000:00:0d.0: fb: 256 MiB of unknown memory type
08:47masterboy: [ 1.965913] nouveau 0000:00:0d.0: DRM: VRAM: 253 MiB
08:47masterboy: [ 1.965915] nouveau 0000:00:0d.0: DRM: GART: 512 MiB
08:47gnarface: did you set the optional tests too?
08:47masterboy: maybe dmesg shows if it has integrated ram?
08:48gnarface: i don't really know sorry
08:48gnarface: that looks like it's saying 256MB
08:49gnarface: or 253?
08:49masterboy: gnarface, no i don't know what optional tests to choose i id the basic test for 15 minutes and the output was no errors
08:49gnarface: maybe it's 3...
08:49masterboy: yes i set it in bios to 256
08:49gnarface: i forget where the other tests are
08:50masterboy: so the error messages in dmesg show that it is a memory error?
08:50gnarface: there's like 6 or 7 normal tests that will all run in a series by default, and there's a couple extra extended tests that catch slipperier failures
08:50gnarface: i don't know for sure if those dmesg errors are memory errors, i just think they probably are
08:51masterboy: okay thanks :)
08:51gnarface: combined with this behavior, which i've seen before, it seems likely
08:52gnarface: when this happens, can you still get a network connection to the machine?
08:52gnarface: will it still respond to ping or ssh?
08:53masterboy: gnarface, i have to set it up to be able to ssh to it? never done this before on a just installed ubuntu
08:53gnarface: i don't know for sure
08:54gnarface: i think ssh may or may not be enabled by default depending on how you installed?
08:54gnarface: i don't use ubuntu but with debian it depends on your installation options
08:54masterboy: ok thanks, i will try to harvest crash logs
08:55gnarface: well, to be clear, you can easily add the ssh service
08:55gnarface: you don't have to reinstall or anything
08:55gnarface: just install openssh-server
08:58gnarface: you can always uninstall it or disable it later too
09:00masterboy: ok it works
09:00masterboy: now i have to hang the system
09:00masterboy: see you in a bit
09:01gnarface: see if you can hang it with a ssh client still connected
09:01gnarface: then see if the ssh client hangs too
09:01gnarface: or disconnects
09:01masterboy: aye :)
09:01masterboy: i got some more errors
09:01masterboy: [ 1522.558100] nouveau 0000:00:0d.0: bus: MMIO write of 03440001 FAULT at 00b020
09:01masterboy: [ 1522.560850] nouveau 0000:00:0d.0: bus: MMIO write of 03440001 FAULT at 00b020
09:01masterboy: [ 1529.825541] nouveau 0000:00:0d.0: bus: MMIO write of 0db70001 FAULT at 00b030
09:01masterboy: [ 1532.795493] nouveau 0000:00:0d.0: bus: MMIO write of 00000000 FAULT at 00b030
09:03gnarface: write fault
09:03gnarface: seems bad
09:04gnarface: seems like it failed to write to ram or something
09:04gnarface: seems like it could be having a problem with the wrong addresses too
09:04gnarface: might be software or hardware, i'm not a driver developer, i should be clear about that
09:05gnarface: does it freeze up instantly then?
09:05gnarface: or at least corrupt visibly?
09:05gnarface: oh i guess that's a yes
11:16masterboy: gnarface, now i see that C61 is not supported by nouveau
11:16masterboy: only c51
11:17masterboy: i installed the nvidia drivers and now it works fine
11:17masterboy: i guess the problem is there is no support for c61 in the first place
11:17gnarface: masterboy: oh, bummer. well that's unfortunate. i was going to suggest trying the ram sticks individually with memtest, but it sounds like that would be pointless now
11:21masterboy: i guess there is no chance to get GeForce 6150SE nForce 430/integrated/SSE2/3DNOW! in nouveau? :P
11:21masterboy: i tried ssh but the machine was unusable
11:23masterboy: the errors i get are FAULT at 00b000 FAULT at 00b010 FAULT at 00b020 FAULT at 00b030
11:23masterboy: before the hang...
11:27gnarface: i'm the wrong guy to ask about it
11:27gnarface: try imirkin maybe
11:28masterboy: GeForce 6150SE
11:28masterboy: Quadro NVS 210s / nForce 430 mhm it is supported.. i am getting confused. :)
11:28masterboy: gnarface, thanks for all the help :)
11:28gnarface: i don't have any way to know if they've even tested that particular board
11:28gnarface: or how recently if they have
11:29masterboy: thnaks for support anyway
11:34gnarface: no problem, sorry i couldn't be of more help
11:48masterboy: it seems the nvidia 304 driver is eol and does not support ubuntu 16 lts 4.15 kernel heh and 4.4 kernel lags...
11:49masterboy: now i need to install a different kernel...
11:50masterboy: the ubuntu guys missed the nvidia 304 driver support in their lts :D
12:08masterboy: The Linux 304.* legacy driver series is the last to support the NV4x and G7x GPUs and motherboard chipsets based on them. Support for new Linux kernels and X servers, as well as fixes for critical bugs, will be included in 304.* legacy releases through the end of 2017.
12:08masterboy: so i have to find a lts kernel or get nouveau working :)
12:09gnarface: i think Steam requires 340
12:09gnarface: so if you're doing this to play on Steam, it'll have to be nouveau
12:10gnarface: hmm, i wonder if it's something simple like you just need to specify the right amount of ram
12:11gnarface: if you can use the legacy driver package installer with a newer kernel i think that would be better
12:12masterboy: the nvidia drivers show i am using 151 MB (59% so i need to set 256 mb for sure :)
12:12masterboy: i will try now to install lts kernels and see if i can get 4.14 :P
12:13masterboy: it seems debian has a patch to make it work on 4.15 and up kernels... have to install debian then...
12:14masterboy: Building the kernel module has been tested up to Linux 4.15. Debian
12:16gnarface: in theory you could add the patch to your own kernel too
12:17gnarface: that might be more trouble than just patching and rebuilding though
12:28imirkin: masterboy: those 0b000 faults are nothing to worry about
12:29masterboy: hi imirkin well something hangs the nvidia c61 :(
12:29imirkin: <masterboy> [ 1.916739] nouveau 0000:00:0d.0: fb: 256 MiB of unknown memory type
12:29imirkin: that's surprising.
12:30imirkin: skeggsb: --^
12:30masterboy: imirkin, it is ddr2 800mhz :)
12:31imirkin: well, the point is that it should say "stolen memory"
12:31imirkin: having "unknown memory type" is likely to lead to other issues
12:33masterboy: thanks imirkin for the clarification. :) How can I help?
12:33imirkin: i'd try an older kernel ... something in the 4.0 vintage, pre-4.3
12:34gnarface: my bad
12:34imirkin: if that works better, then we can figure out which of the VMM rewrites screwed it up
12:34imirkin: if it doesn't, then it's likely a userspace issue
12:35masterboy: imirkin i will try now and see if nvidia drivers work on 4.15 with the latest patches to 403 and then i will try an older kernel thanks. :)
12:36imirkin: nvidia drivers will work MUCH better on a nv4x
12:36imirkin: nouveau has a number of known issues with that hw
12:36masterboy: i have time on my hands now so i check :)
12:37masterboy: but it seems lag is fixed only in kernel 4.5
12:38imirkin: the 6150SE is particularly vexing as many of them have a DVI port that we just can't drive.
12:38masterboy: Your issue is probably related to this. You need to enable PRIME Synchronization but that requires Kernel 4.5+ and X.org 1.19 (ABI 23) which are not available in Ubuntu 16.04 by default. Either use some ppa which include those versions or switch to another distro which comes with newer packages..
12:38imirkin: PRIME won't work with nv4x
12:38masterboy: oh i see
12:39masterboy: imirkin, my mobo does not have a dvi, only vga
12:39imirkin: ah ok
12:39imirkin: some had a DVI port sitting on an external encoder
12:39imirkin: we never figured out how to drive that external encoder
12:40masterboy: imirkin, maybe there is a problem because of amd declaring the memory as unganged?
12:41masterboy: it is a amd thing
12:41masterboy: okay i will check older kernels :)
12:41imirkin: at this point i don't even know what your issue is...
12:41imirkin: does glxgears work?
12:41imirkin: and just heavier stuff dies eventually?
12:41masterboy: imirkin, everything works until it hangs
12:42masterboy: see my photos
12:43imirkin: that could just be nouveau being buggy
12:43masterboy: 256mb pc hang
12:43imirkin: rather than anything more sinister
12:43masterboy: imirkin, ah and the unity dash is not transparent on nouveau
12:43masterboy: but on nvidia it is
12:44masterboy: maybe nouveau cant handle the 3d bling of unity mhmhm
12:45imirkin: that's highly likely
12:45imirkin: even likelier that unity detects nouveau and stops doing pointless things
12:46masterboy: as if gnome would be better :D
13:35masterboy: imirkin, it seems nouveau does not crash on ubuntu kernel 4.4 but does crash on 4.15
13:36masterboy: 4.4 and 4.15 being the only official options
13:36masterboy: i get the same errors but it does not crash at least :)
13:38masterboy: 4.4 log https://paste.gnome.org/p3expuxzd
13:39masterboy: [ 1.888423] nouveau 0000:00:0d.0: fb: 256 MiB of unknown memory type
13:39masterboy: [ 1.947981] nouveau 0000:00:0d.0: DRM: VRAM: 253 MiB
13:39masterboy: [ 1.947983] nouveau 0000:00:0d.0: DRM: GART: 512 MiB
13:39karolherbst: imirkin: regarding that MCP89 mmu thing. We have this issue for months now and I am under the impression that skeggsb told me that he is sure that the set mmu class is correct. I have the same issue on one of my machines and using the mcp77 one just works though
13:40masterboy: vram should be 256 not 253 maybe? :P
13:47masterboy: i will try debian 4.9 now and see if it is a userspace thing
14:16masterboy: okay debian 9 with kernel 4.9 works fine too. so it is a ubuntu thing with 4.15 kernel...
15:13crmlt: Does this Option "AccelMethod" "glamor"
20:46selk: Hello, is the gt 1050ti well-supported by nouveau?
20:49RSpliet: selk: What do you consider well-supported?
20:49RSpliet: I'd say no, but it's a bit subjective
20:50selk: if it is full supported..
20:50RSpliet: oh hah, no that definitely not
20:51selk: which is the best card for nouveau?
20:51selk: in terms of support, compatibility
20:51RSpliet: 780 Ti I think
20:51RSpliet: or Titan
20:51selk: oh, I will check
20:52RSpliet: Anything newer than that (on the high end at least) is not going to let you bring clocks to the cards full potential - you'll just have a crippled experience
20:52HdkR: Kepler woo
20:53RSpliet: Also unsure whether we can do HW video decoding on anything beyond kepler (and the first gen maxwell, but that's all low-end cards)
20:53RSpliet: Equally, we can't change clocks on the Fermi gen cards, so don't get stuff older than say a GTX 680 or 690
21:07HdkR: RSpliet: I assume Maxwell/Pascal reclocking isn't implemented even if you snag the firmwares from the blob?
21:15pmoreau: karolherbst: I remember there was a discussion somewhere about that (either SPIRV-LLVM-Translator or SPIRV-Tools), and I think the answer was: yes. But I’d need to double-check the spec.
21:16selk: 1060 and 1070 are supported?
21:16selk: according to https://nouveau.freedesktop.org/wiki/CodeNames/ is supported
21:17selk: but on the bottom, i have the doubt.
21:17HdkR: selk: Supported isn't the same as your original ask of "fully supported"
21:17HdkR: Things will run, but perf will be garbage and isn't bound to change anytime soon
21:17selk: well.. i am going to bought a card, basically i want to know if it will work with nouveau
21:18HdkR: Card will run at idle clocks, which is <100mhz on those cards. Probably hit other issues that aren't just perf related too...
21:20pmoreau: HdkR: Reclocking works for GM20x but isn’t enabled as fan speed cannot be changed. Pascal doesn’t work as you need to be in secure mode to change the voltage. (IIRC)
21:21pmoreau: Since no one has been able to find the PMU firmwares, no clue if that would be enough to get reclocking working or if more RE is needed.
21:21karolherbst: HdkR: idle clocks aren't that low :D
21:22karolherbst: HdkR: GPUs tend to boot on max clocks from the lowest perf level, which is round about 300-500 MHz or something
21:23karolherbst: pmoreau: well, we would need to re the PMU interface
21:23pmoreau: True, there is that
21:23karolherbst: but I think this is more of a setup the PMU and let it do its work kind of thing
21:24HdkR: karolherbst: Ah, maybe I'm too used to laptop idle clocks being super low
21:25HdkR:can't remember what idled at 78Mhz
21:26karolherbst: nvidia does odd things on idle
21:27karolherbst: practically going below the lowest cstate
21:27karolherbst: which... more or less doesnt change much as you are stuck with the same voltage
21:28HdkR: Also just noticed this GPU I was using was idling in P0 state. No wonder idle amount for this PC was insane
21:28HdkR: idle power consumption*
22:26selk: HdkR: can I ask what card are you using for nouveau?
22:35HdkR: selk: I'm not using Nouveau
22:36HdkR: I need to get karolherbst supporting cuda first
22:36karolherbst: be careful what you wish for :D
22:37HdkR: Alright, and need Turing support :P
22:51pmoreau: HdkR: I would ask for RTX support, personally. ;-)
22:53pmoreau: Getting VK_NV_raytracing on my Pascal would be nice.
22:53pmoreau: And with Nouveau, I could then play with improving the BVH quality and build speed. Lots of fun in perspective!
22:58HdkR: pmoreau: Sure, implementing the extension would be a fun time consuming task
22:58HdkR: Somthing as a common codebase in mesa would be neat
23:57imirkin: selk: buy amd
23:57imirkin: *way* better supported
23:58imirkin: not even close.
23:58imirkin: not even comparable, that's how much better.
23:59imirkin: karolherbst: mcp89 using g84_mmu seems fine. the issue is that it doesn't let you do compression, and fails nouveau_bo_new. that's not cool.