00:00 andres: Hm, just got a hung system when doing another xrandr --verbose.
00:02 andres: Also any nouveau debug, or is drm sufficient?
00:02 imirkin_: drm is sufficient
00:05 andres: Updated gist with dmesg from the point I set drm.debug=0x1e. I only subsequently modprobe'd nouveau. And then finally plugged in the external monitor.
00:05 imirkin_: [ 227.420713] [drm:drm_mode_prune_invalid] Not using 2560x1440 mode: CLOCK_HIGH
00:05 imirkin_: W.T.F.
00:06 imirkin_: i disagree.
00:06 imirkin_: silly question, but you're not booting with nouveau.hdmimhz=225 or something equally silly, are you?
00:07 imirkin_: (no, you're not - it prints the params, and it's listed there as default state)
00:07 andres: The above dmesg was from before I did xrandr --setprovideroffloadsource. Added second one with that. Although it does look like the problem is before that...
00:08 imirkin_: doesn't matter.
00:09 imirkin_: https://github.com/skeggsb/nouveau/blob/master/drm/nouveau/nouveau_connector.c#L985
00:09 imirkin_: so ... ideally it's getting detected as an hdmi monitor
00:09 imirkin_: and the chipset >= 0x120
00:09 imirkin_: in which case you end up with 340mhz max
00:12 imirkin_: anyone other than andres have an HDMI monitor plugged in and can send me its edid?
00:13 imirkin_: andres: i think the monitor may not be getting detected as an HDMI monitor =/
00:14 imirkin_: iirc you're supposed to have some like HDMI OUI thing in the extension block
00:15 imirkin_: which this does not seem to hae.
00:15 imirkin_: andres: is this monitor's input HDMI or DVI? i.e. is it plugged into a DVI port with a DVI <-> HDMI adapter?
00:31 imirkin_: andres: i think there's a network situation, not sure you caught my messages earlier
00:31 imirkin_: check logs if you didn't
00:32 imirkin_: basically, the monitor appears to not be reporting that it's hooked up via HDMI, and as a result, you're being limited to 165mhz (as dual-link isn't supported on that port by the nvidia chip)
00:32 andres: I did not get them, just got connectivity back myself. Last I saw was my own "Would it help to printk the computed clock, max_clock and dcb->type?". Done that while not connected to irc. Got "anouveau: MODE_CLOCK_HIGH clock: 241500, max_clock: 165000, type 2"
00:33 imirkin_: right, that's consistent with my analysis
00:33 imirkin_: is the HDMI cable hdmi all the way, i.e. does it connect to an hdmi port on the monitor?
00:34 imirkin_: anyways, if you force drm_detect_hdmi_monitor to return true, i suspect you should get the 2560x1440 mode.
00:37 andres: It does have two hdmi ends, but perhaps there's actually some active component in it (it's one by my employer, and looks a bit odd). I'll try with a plain one. Should've done that earlier...
00:38 imirkin_: basically the monitor is not sending an HDMI EDID
00:38 imirkin_: that means that the monitor doesn't think it's "hdmi". i don't know what decision-making process goes into this.
00:38 imirkin_: afaik it's just like "is this the hdmi port? then send the hdmi edid."
00:38 imirkin_: so the cable itself shouldn't matter
00:38 imirkin_: as long as the port is HDMI in the monitor
00:38 imirkin_: but... who knows
00:39 imirkin_: anyways, i gtg. good luck.
00:39 andres: There's an active looking "knob", which is now making me suspicious.
00:42 andres: imirkin_: Thanks for your help!
01:22 imirkin: andres: as it happens, i'm on an hdmi monitor now, and as you can see, there's a HDMI OUI thing in a vendor-specific block: https://hastebin.com/enuvofanow.sql
01:36 andres: imirkin: Thanks. I'll try with a different cable once home, whatever the chance is that that'll fix anything. Just to make sure I understand correctly: I'll likely still get that hang, because reverse prime somehow requires accelleration?
01:36 andres: (different cable / monitor)
01:36 imirkin: yes, reverse prime requires copy acceleration
01:36 imirkin: which, amusingly, can be provided without the graphics acceleration
01:36 imirkin: however the way nouveau is set up, it's not available
01:39 andres: I'd originally ordered a lenovo p1 gen2 instead of the x1 extreme gen2 I got. With just an intel card. But because they couldn't actually build them at that time for some reson, I relented. I am regretting that now...
01:39 imirkin: =/
01:43 andres: I temporarily installed the binary nvidia tesla crap, just to be able to give a presentation. But brrr.
01:44 imirkin: the only advice i can give you already know yourself, so ... yeah. sad.
01:44 imirkin: (i.e. stay away from nvidia)
01:45 imirkin: depending on the laptop, you may be able to disable it in the bios
01:45 imirkin: if that's the case, you MAY gain that hdmi port back
02:01 andres: That's the first thing I tried, unfortunately.
02:01 andres: There's only discrete *only* or hyprid.
02:29 andres: I guess it's worth testing how well nouveau copes with disabling the *intel* card in bios (no disabling the nvidia). No reverse prime needed anymore, after all.
02:41 imirkin: well, you lose all accel
02:41 imirkin: but the screens should light up
02:42 andres: If external ports work, that'd probably be good enough for presentations etc.
02:42 imirkin: yeah, should be fine
02:42 imirkin: assuming the presentation isn't of the latest 3d game :)
02:45 andres: Heh, no. A bunch of performance graphs would probably be the fanciest bit.
10:36 RSpliet: Whoa! "The patchset has been smoke-tested on NV34 HW by running fbcon and X11."
10:39 RSpliet: (funny enough, these cards run hot enough to actually smoke when they work)
11:13 karolherbst: imirkin: you might want to take a look at this: https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3724
13:53 imirkin: karolherbst: yeah, i saw the patches on list
19:18 andres: imirkin: WRT conversation yesterday: Turns out I don't find my hdmi cable right now, but a 2560x1440 monitor connected via a usbc->hdmi connector is correctly recognized.
19:20 andres: Turning off the intel gpu and booting just via nouveau: Not so much luck. Display is pretty thorughly garbled. I can recognize *something* changes when I type, but that's it.
19:22 andres: "DRM: wndw-0: timeout" and other fun: https://gist.github.com/anarazel/b144f933bc87fcdb3ddfc6053f9bdeea
19:25 karolherbst: andres: mind trying out drm-next? https://cgit.freedesktop.org/drm/drm
19:26 karolherbst: it got a bunch of fixes
19:27 karolherbst: turing is still quite strongly in a WIP phase I'd say so random fails could have been fixed by skeggsb already
19:27 andres: karolherbst: No, don't mind at all - this is with the drm-next pull in 5.6 included already however. Are there additional fixes since?
19:27 karolherbst: ohh
19:29 andres: I'm happy to try out things - I'm aware that turing support is still pretty rough. Luckily I'm now home again, working from my workstation with an AMD GPU in it...
19:30 karolherbst: yeah.. besides that I don't really have any ideas. Just wanted to make sure nobody follows up on an already fixed bug :)
20:12 Lyude: skeggsb: poke, when you get a moment can you take a quick look at https://gitlab.freedesktop.org/lyudess/linux/commit/0dbeaa3ed3c59f67b39ede2c04287cd5f67742f4 to see if this looks like a sensible fix? This is regarding the nvdisplay errors I was showing you before, that commit seems to -mostly- get things working but I'm seeing some other errors sometimes that I think might be unrelated
20:13 Lyude: but i'm also not 100% sure
20:14 Lyude: (I will also get you dmesg output with disp tracing and atomic tracing on, one moment...)
20:17 imirkin_: Lyude: what was you invalid state?
20:17 imirkin_: wasn't 2e, was it?
20:17 Lyude: https://paste.centos.org/view/fcf90ca5 (note that the set/clr bitmasks will differ from mainline nouveau, since we have an extra bit added for CRC)
20:17 imirkin_: coz that's what andres has
20:17 Lyude: imirkin_: yes but I fixed that one
20:17 Lyude: imirkin_: awesome! :)
20:17 Lyude: well, I -think- I did
20:17 imirkin_: what's your fix? the commit above?
20:17 Lyude: yeah, but I'd like to make sure that's not causing the other bug that I just linked to
20:17 imirkin_: andres: --^
20:18 imirkin_: if you feel like trying not-fully-baked stuff
20:18 Lyude: chithead: ^ I think you were asking me about getting help with turing modesetting related stuff as well?
20:18 imirkin_: although looking at the commit, it shouldn't make things _worse_
20:19 Lyude: imirkin_: yeah, I think we might also just need to do that with all of the bits in the clr bitmask since the whole point is to clr/set them as needed per-modeset
20:19 Lyude: probably can't look too further into it today though
20:20 Lyude: (also, new wip crc stuff there to fix more of the issues I found with volta+, still some left to go though but I'm pretty sure I already have an idea of how to fix them)
21:29 skeggsb: andres: i fixed your issue a couple days ago
21:29 skeggsb: https://github.com/skeggsb/nouveau/commit/0764baeeb737771e71b204630fca57f6bd3974e4
21:29 skeggsb: and
21:29 skeggsb: https://github.com/skeggsb/nouveau/commit/e259bf65680933c80e25231b7a09223a68b11029
21:34 andres: skeggsb: Cool, will try that. imirkin_ Lyude, thanks as well! Will also try if skeggsb's commits don't bring it up fully.
21:34 imirkin_: when in doubt, believe skeggsb over anything i say.
21:35 Lyude: trust no one even yourself. except skeggsb
21:35 skeggsb:wouldn't trust me either
21:36 imirkin_: that's ok - i would :)
21:36 imirkin_: ignorance is bliss.
22:14 andres: skeggsb: Nice! Sucessfully boots, internal display not at all garbled. External display is recognized, and shows sensible out (even though it's via usb-c). External doesn't quite work though, hardware cursor and actual position don't seem to match up. Also got errors in dmesg at that point.
22:15 imirkin_: the error messages are hopefully going to contain enough to debug further
22:15 andres: https://gist.github.com/anarazel/c27a2ad9cbfec336a1298096917a0e11
22:16 imirkin_: Feb 06 14:01:21 alap5 kernel: nouveau 0000:01:00.0: disp: chid 73 stat 90001080 reason 1 [PUSHBUFFER_ERR] mthd 0200
22:16 imirkin_: i think that explains the cursor not moing
22:16 karolherbst: imirkin_: in case I get super bored... I've gotten nv4x gpus... :D
22:17 karolherbst: pcie ones
22:17 imirkin_: karolherbst: yay
22:17 imirkin_: which ones?
22:17 karolherbst: nv41 and nv43
22:17 imirkin_: (all nv41+ are PCIe, except NV4A)
22:17 karolherbst: the nv43 is even passivly cooled
22:17 imirkin_: ah nice. i have a nv42, iirc ... Quadro FX 3450
22:18 karolherbst: that's nv41 :p
22:18 imirkin_: nah
22:18 imirkin_: nv42 or 43, definitely
22:18 karolherbst: wiki says nv41.. but maybe that's indeed a nv42 or so
22:18 imirkin_: https://people.freedesktop.org/~imirkin/traces/nv42-vbios.rom
22:18 karolherbst: anyhow, I have the quadro fx 3450 as well
22:18 imirkin_: not passively cooled
22:18 karolherbst: the nv43 is ;)
22:19 imirkin_: surprisingly heavy, considering how light it looks, too
22:19 imirkin_: and has an aux power thing
22:19 imirkin_: and the 3-pin din thing
22:19 karolherbst: huh
22:19 karolherbst: mine is 6 pin
22:19 karolherbst: maybe you got a "upgraded" card :p
22:19 imirkin_: i mean external output
22:19 karolherbst: ahh
22:19 imirkin_: above the DVI ports
22:19 karolherbst: "above"?
22:20 karolherbst: mine is single slot
22:20 andres: imirkin_: The cursor is moving, it's just not where X thinks it is :)
22:20 karolherbst: but yeah, I have this weirdo connector as well
22:20 imirkin_: andres: yeah, some of the cursor updates are failing
22:20 imirkin_: karolherbst: https://www.google.com/search?q=3-pin+DIN
22:21 karolherbst: yeah
22:21 imirkin_: it was for 3d sync, i think
22:22 karolherbst: probably
22:36 imirkin_: karolherbst: lmk if you want a rundown of the issues in the nv30 driver
22:45 Lyude: skeggsb: any thoughts on the patch/kernel log that I showed you btw?