13:41 <init_x13s> Getting a lot of instability with firefox on latest x13s ironrobin stack. (6.4-rc1 / mesa 3.1.0) Am I the only one?

13:46 <init_x13s> proably related to video acceleration I guess.

13:49 <jhovold> init_x13s: video acceleration should not be enabled by default

13:50 <jhovold> seeing more frequent crashes with firefox during video playback too with gpu support v2 and 6.3.3 and 6.4-rc3, but could be some change in userspace too (e.g. new firefox)

13:50 <jhovold> crashed farily regularly with some video sources also before the update so probably not a new issue

13:51 <jhovold> mesa 23.1.0

13:52 <init_x13s> good to know, il build 23.1.1 for fun. Chromium seems more stable

14:06 <jhovold> init_x13s: possibly more stable with 23.1.1 bases on a quick test, let's hope that was it

14:22 <init_x13s> completely deactivating hardware acceleration in firefox solves it for now.

14:37 <robclark> crash as in segfault or gpu crash? In the former case, a backtrace would be useful.. in the latter a gpu devcore dump (https://docs.mesa3d.org/drivers/freedreno.html#gpu-devcoredump)

14:38 <jhovold> robclark: I see the following message before firefox exits:

14:38 <jhovold> MESA: error: fd_bo_dmabuf:534: failed to get dmabuf fd: -1

14:43 <robclark> does adding this to fd_bo_dmabuf() trigger the extra error msg:

14:43 <robclark> https://www.irccloud.com/pastebin/kSOSzdZL/

14:46 <init_x13s> I have numerous stacktraces from firefox,

14:48 <init_x13s> https://pastebin.com/X9FAYmgQ

14:49 <jhovold> robclark: i'll try rebuild with and see if i can find a good reproducer

14:50 <jhovold> only happens with some sites. crashes with swedish public tv appear to be gone with mesa 23.1.1, which is were I thought I noticed something had changed

14:51 <robclark> init_x13s: about all I can tell from that is it is an assert from libxul.so

14:51 <init_x13s> yeah not super helpful indeed

14:51 <init_x13s> ill check how to enable devcoredump as well

14:53 <robclark> devcoredump should most likely already be enabled in kernel.. there won't be anything interesting unless you see msgs in dmesg about gpu faults or hangs

14:55 <init_x13s> understood

16:24 <clover[m]> Mesa bumped to 23.1.1 in ironrobin, i guess i will maintain it until upstream takes the freedreno driver

16:36 <jhovold> robclark: still no good reproducer but I was able to hit the additional assert you asked me to add

16:40 <clover[m]> jhovold: i don't think firefox will support hardware video accel until libva-v4l2-driver is ready for x13s

16:40 <robclark> hmm, ok.. well I can suggest a workaround, but I think it is probably a firefox bug.. basically somehow we are getting asked to export something that wasn't allocated with the PIPE_BIND_SHARED flag

16:42 <jhovold> clover[m]: yeah, that was my understanding as well

16:42 <robclark> in fd_device_new(), make use_heap be false, that will disable suballocation

16:43 <jhovold> robclark: thanks, i'll give it a try chromium seems more stable so this may very well be a firefox bug

16:43 <robclark> btw, I did order an x13s this morning.. maybe it can replace my ~4yr old xps13 as my daily driver non-corp laptop

16:43 <clover[m]> put arch on it!

16:44 <jhovold> nice!

16:44 <robclark> heheh, I think I'll stick with fedora

16:44 <robclark> I did try arch on on of the chromebooks for a bit but went back to fedora ;-)

16:44 <clover[m]> i haven't seen anyone using fedora yet. good luck ;)

16:45 <robclark> yeah, hopefully I don't have to mess too much with grub.. once I get past that I should be able to handle the rest

16:47 <clover[m]> im using systemd-boot. much more straightforward set up imo

16:53 <robclark> hmm, the interwebs seems to have instructions for systemd-boot.. maybe I'll give that a try

16:55 <robclark> does also appear that there is some fedora work on x13s.. https://fedoraproject.org/wiki/Thinkpad_X13s

16:57 <clover[m]> Libre____: ^

16:58 <Libre____> oho

16:58 <clover[m]> needs some TLC, it still says no GPU

17:04 <clover[m]> google meet is working since upgrade mesa. was broken in vivaldi yesterday. not sure if there is a connection

17:08 <robclark> well, I mean.. it *is* a wiki page after all.. that means by definition it is out of date :-P

17:11 <jhovold> robclark: that use_heap hack made all the difference, no issues with firefox anymore. thanks!

17:12 <robclark> ofc.. now the question is _what_ is it that it is trying to export and why don't we get the hint at allocation time.. or how can we gracefully cope..

17:13 <jhovold> yeah, i wouldn't know who to poke about that... you still think it's a firefox rather than mesa bug?

17:14 <robclark> I'd have to poke around a bit w/ gdb and figure out what it is that it is trying to export

17:15 <steev> robclark: poke ajhalaney[m], probably

17:16 <steev> re: fedora

17:16 <konradybcio> robclark you daily driving the x13s almost sounds like more a690 massaging ;)

17:19 <robclark> heheh.. well it would be convenient to have an a690 device.. but I think it could be fast enough to be my primary (non-corp) laptop

17:20 <robclark> jhovold: looks like radeonsi does handle the case of copying/migrating to non-suballocated buffer.. so I guess maybe freedreno needs to do a similar thing

17:20 <konradybcio> it's no match for the m1 but i find it sufficient if you compile your kernel somewhere else..

17:21 <konradybcio> ..or incremental build most of the time

17:22 <robclark> non-incremental kernel build is already not great on my xps13.. but I spend a lot more time doing incrmental builds

17:23 <jhovold> robclark: ah, that sounds promising

17:32 <robclark> jhovold: if you don't mind, file an issue on gitlab.. and, if you do manage to figure out a good way to repro that would be useful as well

17:52 <ajhalaney[m]> robclark: I do run fedora on my x13s but it's a real monster at this point. I ought to start clean but I'm kind of married to it since I use it for work. Custom kernel (need to figure out how close the fedora kernel is and enable whatever configs are missing), never could get grub to work and had to use systemd-boot, some more custom packages that I need to try and figure out how to properly package, etc

17:52 <ajhalaney[m]> https://copr.fedorainfracloud.org/coprs/ahalaney/pd-mapper/

17:53 <ajhalaney[m]> I sort of went outside the package manager for dealing with mesa and audio and regret it cuz now it's not really clean. Probably should have tried something like silverblue or just be good about actually making custom packages

17:54 <ajhalaney[m]> It's doable tho just gotta be willing to tinker with it. Like I said I use it for work, but not for calls.

17:57 <robclark> ajhalaney[m]: I don't mind running my own kernel, I'd probably be doing that anyways

17:58 <ajhalaney[m]> I figured with you and DRM stuff :) but sounds like others are also interested so just highlighting. The grub thing was annoying. Debian grub worked ok, idk what happened with fedoras grub. So many patches. Kind of like systemd-boot after trying it though

17:58 <robclark> I guess by now fedora mesa should have a690 support.. hopefully.. usually what I do is install my own mesa builds to an alt path and use LD_LIBRARY_PATH

17:58 <robclark> ahh, good point

18:04 <steev> ajhalaney[m]: fwiw, the alsa ucm configs are in upstream

18:08 <ajhalaney[m]> Nice! Hopefully dnf just blew over my local modifications here and there but who knows. Didn't take the wisest approach. Don't hire me as a sysadmin lol

18:31 <clover[m]> steev: audio does work with upstream alsa configs but the sound is every quieter i can barely hear the speakers at 50%

18:32 <clover[m]> any ideas why?

18:32 <clover[m]> s/every/even/

18:41 <steev> ajhalaney[m]: i do the same thing... before i build the package

18:41 <steev> clover[m]: probably needs some tweaking still

18:42 <clover[m]> tweaking the speaker PA values "fixed" it

19:11 <clover[m]> are y'all seeing this similar issue in alsa-restore.service? Operation not permitted...

19:11 <clover[m]> https://pastebin.com/BCH6aVXb

19:16 <jhovold> robclark: sure, i'll do so next week. thanks for your help so far!

19:17 <jhovold> robclark: btw, does the LD_LIBRARY_PATH setup still work? The way I use to run in broke with commit 1026d29344192 ("dri: Introduce internal Mesa DRI driver loader extension.")

19:18 <jhovold> apparently I had been mixing components from the arch installation and my custom build without realising

19:38 <bamse> jhovold: even if you set LIBGL_DRIVERS_PATH as well?

19:42 <Libre____> ajhalaney[m]: I see, but after your mods you're happy? GPU works fine? suspend and battery life?

19:49 <ajhalaney[m]> Yup but I use it plugged in mostly and don't suspend ever lol Libre____. But it works well. I mostly want a aarch64 build machine and then sort of feel good about very very infrequently contributing upstream with the laptop

19:49 <Libre____> aha

19:49 <Libre____> I'd get it partly because of the alleged long battery life

19:56 <robclark> jhovold: it should.. not sure if it is dependent on glvnd or not.. I use something like:

19:56 <robclark> https://www.irccloud.com/pastebin/aZTRpCv1/

20:03 <bamse> Libre____: we're at about 10 hours battery life with linux right now...and are working on various improvements

20:03 <Libre____> good good

20:03 <bamse> would be interesting to see some independent battery benchmarks in windows though

20:04 <bamse> Libre____: what doesn't work as it shoudl though, is suspend

20:05 <bamse> Libre____: we can suspend and resume, but you don't get more than a few extra hours compared to idle...

20:05 <bamse> this is also being worked on...probably more than the idle numbers...

20:06 <Libre____> 10 hours is pretty amazing as is

20:06 <Libre____> compared to the mess that is my thinkpad x395 :P

20:06 <bamse> certainly

20:07 <bamse> you can spend all day hacking in your hammock...on lte/5g...

20:08 <bamse> but unfortunatley you need to return it to the charging station over night..so far

21:58 <clover[m]> or coding in a canoe

22:01 <bamse> clover[m]: i'm always worried that water and gravity will be colluding...