02:21 danielmassey: you die out, #llvm trashbags similarly modified their channel as moderated, this does not stop me from killing off all the trash that does not belong to this planet, you are not going to get another chance to carry out your gene and dick research. you did dug your own grave literally. I will suck all your blood out with my allies just like it was done to me, I take all of your assets and
02:21 danielmassey: money away from you.
02:21 danielmassey: and discord and matrix is also easy to fool.
02:24 dwfreed: karolherbst: you may want to do the same here; I can auto-voice the discord bridge at least (because that's easy to match)
02:25 HdkR: 🎉Party!
03:19 airlied[d]: _lyude[d]: so for hdmi2.1 if a mode needs FRL but the HDMI is in TMDS or vice-versa, you have to shut down the whole head, and bring it back up, I'm just not sure how to do that with atomic modesetting
03:22 _lyude[d]: airlied[d]: are you looking for `nv50_atomic_state->flush_disable`? I might not have the name of the struct totally correct, but it's the global atomic state struct thing nouveau has
03:24 _lyude[d]: That at least makes it so that disabling resources is done in a separate flush then enabling them, which I assume would do what you're asking for
03:24 airlied[d]: oh maybe I can figure out how to set that, I'll start there
03:24 _lyude[d]: We use it in dispnv50/crc.c I think
03:25 _lyude[d]: MST also needs this so it would make sense
03:30 _lyude[d]: (also - make sure you set the DRM state flag for requiring a modest. but I assume you probably knew that part at least)
03:46 airlied[d]: hmm doesn't seem to work at my first attempt, but I'll play around a bit more
04:04 hiroshimakito: so I removed the Me region and disabled hap, and my work computer is stable now, in cambodia they started that shit, same people were extorting me there and still do but I never will forward any money to them, I make my modifications now, indrek raud fraud computer was surveillanced by ukranian and Russian Armenian computers, he never realized that what vpro is, its meant for remote boot
04:04 hiroshimakito: and it has 486 die and fw for rndis.
04:04 hiroshimakito: but I have several intel and amd computers and tablets
04:07 airlied[d]: _lyude[d]: okay got it to kinda work, can switch from 120 hz to 60hz mode and back
04:29 bradcooley: that Cambodian streets are empty same for beaches is entirely logical when they do not kick the criminals out it’s there to stay so. and the porno slut appealed for rape where they got the old crocodile show going on, bull that tries to take over businesses was called full of steroids and many kill offs happened, but realistical success they never had with their bikers gang except the
04:29 bradcooley: opposite the bitches drawer was just wiped out of teeth after they assaulted another injury to me, if they show their faces one more time around me all of them are handled more brutally this time.
04:37 airlied[d]: the sequencing on this is messy, have to work out the ordering deps, also if we boot in FRL mode, it hangs going back to TMDS
05:04 airlied[d]: oh turns out booting with HDMI plugged on on this laptop causes errors when nouveau first sets a mode, even without FRL hacks
11:50 raphaelsutos: so they used a project named deathmetal to intrude into employers vpro machine where as modules of that project that germans did were named after the workers. the certificate that was signed to handle indrek’s computer was from Perth Australia , and tranny tornado also arranged this mindless buffolo and notified legal powers and scam negro anal stuffers about rape as well as wank cases,
11:50 raphaelsutos: this is mentally ill diagnosed slut , with crocodile Gloria they harass every business with their takeover events gangster wannabes, every Khmer bum or scrub screws them, if they start to do those events again near our areas we eliminate them, their crank gangsters, brothers and all their family altogether. at the moment none visits those disgusting trashes and Khmers are out of tourism at
11:50 raphaelsutos: least they can stuff their crocodiles duo.
11:51 raphaelsutos: scrubs such as exporters play the game always alike, you have to know their sick moves
11:51 raphaelsutos: to understand how ill they are
12:08 frontalgary: so you see Nathan having one eye missing, laura frontal teeth, jura and Martin dead, you wanna collide with this gang, it will happen so that Fecalherbst’s balls are removed which he will eat, later they remove all his guts by torture, samewise fuck off with your German enterpreneurs charity e-mails as well as pegasus extortion efforts you are soon dead I tell you so, you have no clue what
12:08 frontalgary: people you are messing with. so in near future we slaughter jack with my forces added.
12:47 sunrise_sky[d]: Hi guys. Is it possible to use and test NVK on a 'normal' kernel or do we need something specific?
12:48 karolherbst[d]: just use a normal kernel
12:48 karolherbst[d]: just needs to be relatively new
12:48 marysaka[d]: normal kernel + GSP firmware
12:48 karolherbst[d]: and with relatively I mean like.. a year?
12:48 karolherbst[d]: or two?
13:03 sunrise_sky[d]: so, if I install, let's say Fedora (without the closed drivers), it will already have everything I need?
13:04 marysaka[d]: yes
13:04 sunrise_sky[d]: awesome!
13:05 sunrise_sky[d]: can NVK coexist with the proprietary nvidia driver?
13:06 marysaka[d]: sunrise_sky[d]: it should be able to just need to make sure nouveau is loaded if you want to run with NVK
13:06 marysaka[d]: I know that device select layer seems to break since 25.2 with NVIDIA blobs on Fedora at least so use NODEVICE_SELECT=1 if you get a weird stackoverflow
13:07 loanselot[d]: hey, my pascal gpu reports vk 1.4 hostImageCopy to be available but there's a comment stating its not supported on nvk_get_vk_version before turing
13:07 loanselot[d]: is it not supported by nvk or my gpu?
13:07 sunrise_sky[d]: marysaka[d]: can nouveau be loaded alongside the proprietary driver, or is that something that you'd need to reboot for?
13:08 marysaka[d]: sunrise_sky[d]: not really I think it's a bit tricky... I do switch between the proprietary driver and nouveau at runtime but I do not use a desktop environement when that happen (remoting to my test bench)
13:09 soreau: I don't think it's as versatile as amdgpu wrt the foss and proprietary drivers.. different kernel module, which takes ownership of the hw, right?
13:09 jannau: marysaka[d]: the device select isssue with mesa-25.2.7 in fedora should be fixed
13:10 marysaka[d]: ah nice I should probably update then
13:10 sunrise_sky[d]: marysaka[d]: I see. So, easiest to just not even install the proprietary one
13:11 karolherbst[d]: loanselot[d]: sure it's a pascal one?
13:11 loanselot[d]: karolherbst[d]: yea i have 1050 ti
13:11 marysaka[d]: sunrise_sky[d]: Probably, it's just quite handy in my case to do some command stream and shaders capture
13:11 karolherbst[d]: loanselot[d]: is the vulkan version overridden then?
13:12 karolherbst[d]: like `MESA_VK_VERSION_OVERRIDE` being set
13:12 marysaka[d]: marysaka[d]: I do plan to finish up a testing backend to run NVK on top of the proprietary and "open" kernel driver but still need to get back to it
13:13 loanselot[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1445402509161009245/image.png?ex=6930377c&is=692ee5fc&hm=77dd1700097652efb09edf9721f09f2fffc8bc276497ef877553262dc7b6b379&
13:13 loanselot[d]: karolherbst[d]: nope i didnt override it, i followed the breakpoint it falls back to 1.3 because of this
13:13 karolherbst[d]: and yet it still reports 1.4?
13:13 loanselot[d]: no it reports 1.3
13:14 karolherbst[d]: ohh but it reports `hostImageCopy`?
13:15 loanselot[d]: no no im just trying to figure out if `hostImageCopy` is not implemented yet by nvk, but can be added later because the closed source driver supports it
13:15 karolherbst[d]: ahh
13:16 karolherbst[d]: yeah it's available on turing+, but I guess nvidia supports it on pascal as well?
13:16 x512[m]: marysaka[d]: Did you published your NVK on NVRM work somewhere?
13:17 marysaka[d]: x512[m]: Not yet but will probably get back to it this week or next week, will push when that happen
13:17 loanselot[d]: hold on i remembered something
13:17 loanselot[d]: the nvidia driver says `hostImageCopy` but if you try to create an image with `hostImageCopy`'s flags it will throw error 😛
13:17 karolherbst[d]: nice
13:20 x512[m]: Interesting to know how much better it than my work. I need some extra abstractions because NVRM access on Haiku is a bit different from Linux and there are no plans to port Nvidia DRM KMS driver, so no syncobj stuff etc..
13:20 x512[m]: Haiku do not support mmap for devfs files for example.
13:51 mohamexiety[d]: Host copy with pascal is implemented. I don’t think we disabled it afterwards
13:51 mohamexiety[d]: There was just a kernel side bug where the kernel was doing some remapping/swizzling on vram stuff
13:52 mohamexiety[d]: But what we did end up doing is just put everything in GART similar to Kepler iirc
13:59 karolherbst[d]: well it's still disabled tho
14:47 _lyude[d]: airlied[d]: Interesting
18:29 airlied[d]: _lyude[d]: I suspect it's moving EDP to head 1 from head might be a problem, but not sure yet. To get EDP working I had to disable inherit from boot state for it
18:30 _lyude[d]: airlied[d]: Do you have any logs from it?
18:31 airlied[d]: Not yet, the EDP problem should be reproducible on any laptop since ampere I think, just switch the bios to discrete graphics
18:32 _lyude[d]: right - I remember this bug
18:33 airlied[d]: We inherit the outputs and the first edp modest does a disable then enable and never lights up
18:33 _lyude[d]: But it trains properly if we don't inherit?
18:33 airlied[d]: Stopping the inherit just does an enable and works okay
18:33 _lyude[d]: ooooooo heck yeah. Ok, that gives me a very good idea of things that might be going wrong here then
18:33 airlied[d]: It's why it works on module reload
18:34 airlied[d]: Because there is no inherit
18:34 airlied[d]: I'm not 100% sure it's link train vs stream activation
18:35 airlied[d]: The DP stream active bit never gets set
18:35 airlied[d]: So link seems to train but no data gets sent
18:36 airlied[d]: If I have HDMI plugged in I get a core channel notified timeout
19:17 violet_purple_red[d]: new kernel dropped 🤩
19:18 _lyude[d]: airlied[d]: btw - I'll try to take a look at this sometime soon, want to see if I can get some new versions of patches out today first
19:18 _lyude[d]: also - JFYI, IRQ spinlocks might land in 7.0 🙂
19:32 airlied[d]: _lyude[d]: why did you decide frts didn't need the same treatment as sb?
19:32 _lyude[d]: airlied[d]: It could be that I made a mistake but as far as I can tell I think we only load frts in the tu102_gsp_oneinit
19:33 _lyude[d]: sb seems to be the only one we use past the initial driver load
19:33 _lyude[d]: (I think we use sb every time we shut down the GPU)
19:39 airlied[d]: I think we use sb on shutdown and frts on startup
19:40 airlied[d]: but maybe we only need to do frts once on takeover from bios
19:58 _lyude[d]: I believe so
19:58 _lyude[d]: Since that's the only place I found us actually loading frts
20:37 airlied[d]: _lyude[d]: https://people.freedesktop.org/~airlied/scratch/nouveau-hang-msgs.gz
20:38 airlied[d]: the core channel timeout after the first modeset
20:46 _lyude[d]: gotcha, unfortunately I only just remembered we don't really have any debugging output for the inherit code. But, this gives me a pretty strong place to start
20:47 _lyude[d]: since I'm fairly certain I remember having some machines here that had eDP link training issues as well
20:47 _lyude[d]: but I think I had to put down troubleshooting them when the nova work started ramping up
20:55 airlied[d]: Ben fixed a bunch of stuff with lttpr changes
20:55 airlied[d]: I'm pretty sure the problem is more likely a sequencing problem with what bits of the pipeline need to be turned on when
20:55 _lyude[d]: that is quite likely, yeah
20:56 _lyude[d]: especially since iirc, there were some big reworks he wanted to do to the order we do modeset stuff in
20:58 _lyude[d]: "I don't really have any "plans" as such. However, a lot of the tracking that was done for delaying SOR routing until supervisor (all the .arm/.asy transitions etc) can be torn out and simplified now that it's all done immediately" i think this was it
20:58 _lyude[d]: (from chat history with ben in the past)
21:09 airlied[d]: I'd just like to make sure we have some understanding of the corner cases pre-nova, though I'd like to root cause the eDP panel one
21:18 _lyude[d]: mhm, true true