Arch Linux User Repository

For testing users, contributors: check out https://github.com/archlinux-jerry/nvidia-340xx
Warning: force push may happen to match the aur tree

hello again . first arch linux do everything you want . BUT you must do everything by yourself . when arch linux drop 32bit support , some people make they own 32bit arch linux !

at the moment i use strace + "programm". works very good. no big cpu/mem impact .

i think bui ld-id is a dead end , but for my and you learning

i tried to set the build id option in /etc/makepkgconf as explained here --->https://interrupt.memfault.com/blog/gnu-build-id-for-firmware

add to -> CFLAGS=" -Wl,--build-id . for a quick test try "yay sfml2" . it quickly crash compile with error "unknown build otion" if you make something wrong. to check build id , readelf -n "file"

readelf -n /opt/sfml2/lib/libsfml-graphics.so or readelf -n /bin/"yoursoftware/" ! i dont know if this is 100% correct ! know i try to compile nvidia-340

That's what I was talking about some time ago: you'll face more and more problems with this driver, unfortunately. And one day, everything will stop working. We all have to accept it...

Fortunately, there are still LTS kernels. Even binary ones at Chaotic AUR repo for Garuda Linux (btw I'm using 5.15.x and 6.6.x branches from there and everything works OK). Also, you can always compile any kernel you like yourself (though it takes time and system resources).

Yes @r3n4n. After some researches, because I'm not a C developer, here it is what I think I have found. A build-id is just a unique identifier "for debugging purposes". This information is stored in a property, a "section", named ".note.gnu.build-id" in artifacts built by GCC. When GCC is built with --enable-linker-build-id, gcc generates the .note.gnu.build-id section by default or you have to set it at build time with -Wl,--build-id option. Visibly Linux kernel doesn't accept no more to run an artifact, as a shared library, that has not this information embedded in it. I don't know if I understand right but it seems that Nvidia's shared libraries (.so), which are a bit old, were written without this build identifier. I don't know if it's possible to find a work-around. Is it possible to insert this property with a fake build-id during the makepkg process? Or set a Linux kernel env var to accept running artifacts without a build-id? I don't know yet at all. But, quite good news, by themselves, applications can always run.

Probabily is because the proprietary blob does not follow the requirements of current kernels, regarding support for critical security features.

Same issue here, with kitty that I installed just to test: segfault. I use keepassxc and it doesn't start no more. Segfault too. As you, @c0mmand0x72, when tested with strace -eopen keepassxc and coredumpctl gdb, Keepassxc is OK. Coredump, as you, show the same issue, modules without build-id. Thanks for those tools and their usage to debug! I'm gonna try to pass --build-id to build the package.

Module libnvidia-glcore.so.340.108 without build-id.
Module libnvidia-tls.so.340.108 without build-id.
Module libGL.so.1 without build-id.

BTW: the current package, release 39, compile against Linux 6.16.0-arch2-1.

im sorry but i miss sa line. so when i use gdb with coredump and bt , it shows a problem with libnvidia-tls.so and with ?libthreads? .

hello again . first thx for the patch :) BUT! i has problems too with some apps from the AUR that crash with segvaults. BUT! SURPRICE ! as example "attract-git" from AUR yay attarct-git -> compile + installs with no errors when run attract -> segfault ! BUT! strace -eopen attract -> all ok ! gdb attract , run -> all ok ! LOL

the coredumps show first in gdb Module libnvidia-glcore.so.340.108 without build-id. Module libnvidia-tls.so.340.108 without build-id. Module libGL.so.1 without build-id

quick possible solution ?? "Tells GCC to pass --build-id option to the linker for all final links" source > https://github.com/conan-io/conan-center-index/issues/18809 https://github.com/conan-io/conan-docker-tools/pull/531 https://gcc.gnu.org/install/configure.html

it only affects my AUR stuff like guake,nm-applet,caffeine etc but when startet at the desktop they run fine

for learning , coredump type coredumpctl pick the PID of a dump then type coredumpctl gdb PID

again , i have only time at weekend (: so pls give some feedback to save time

@lllllllllllll Just tested with kitty. Yes, the same thing: [1.079] The output buffer does not support sRGB color encoding, colors will be incorrect. zsh: segmentation fault (core dumped) kitty.

@lllllllllllll I'm having the issue with Goldendict. I get Segmentation fault (core dumped) when running the application.

kitty terminal not working for me. "The output buffer does not support sRGB color encoding, colors will be incorrect. Segmentation fault (core dumped) kitty" anyone else having this issue?