[SOLVED] 5.16.12 dkms modules not building

levi · 2022-04-08 23:43:09

I've been running python 3.10 for a couple of weeks now. I don't use blender or many so modern things though, so I can't be sure nothings breaking.

abaumann · 2022-04-09 06:37:03

I pushed python 3.10 yesterday. The few errors I still had on testing justified the move. But, there might be software
(especially software which currently doesn't rebuild), which might have broken Python bindings now (pointing to 3.9).

nateewy · 2022-04-12 21:48:27

The issue is back with kernel 5.17.1 for linux and linux-zen on pentium4. Same errors in the log file with structleak_plugin.so. I am using the 8192eu-dkms-git package in AUR with yay. It worked before with one of the 5.16 kernels.

mocambo · 2022-04-15 17:48:49

abaumann wrote:

It is a challenging "hobby project", I would say. :-)
blender:
libpython3.10.so.1.0 => None
Now I have the same issue there. I might have to move now to Python 3.10 in stable, with all (the possibly nasty) consequences..
I'm always glad to get feedback, but there is no way I can run and test all software myself.

if a Blender has already been mentioned here.
To my knowledge, this package has long since stopped working with older graphics cards and requires OpenGL 3.3 as minimum. Despite it compiles for 32-bit machines, I think it should be removed from Archlinux32 repositories. Any thoughts ?

jghodd · 2022-04-15 19:07:04

@abaumann - aaaand we're back. 5.17.1 has the same issue. having to rebuild it. i see in the config file that gcc is defined as 11.1. you aren;t using gcc-11.1 to build the kernel are you? gcc-11.2 has been in the stable repository for a while now.

haven't seen any issues with python-3.10. yet.

Edit: apparently, it's not just my system going through dkms issues with the "stable" kernels - @nateewy is reporting the same error.

Edit2: it's extremely interesting that i opened this thread the very day my system updated gcc from 11.1 to 11.2 - 03/11/22.
[2022-03-11T19:37:52-0500] [ALPM] upgraded gcc (11.1.0-3.0 -> 11.2.0-3.4)

Last edited by jghodd (2022-04-15 21:38:31)

abaumann · 2022-04-16 07:46:38

Yeah, I think, there is a fundamental problem here: if the kernel depends on a specific gcc version, it effectively
becomes part of the toolchain. The kernels are always built against the gcc version in staging (which tends
to be newer than in stable). The buildmaster doesn't know about this dependency.

So, currently we have to make sure to rebuild gcc and kernels in sync.

Also, we could at least try to update the config setting for gcc with a releng script or so, so that at least the
version don't get out of sync in staging by accident..

IMHO a compiler should expose a stable enough API, so that any specific range of compilers can build the same
kernel. This is not a nice thing if the kernel starts to depend on specific gcc versions..

jghodd · 2022-04-16 18:02:25

@abaumann - i think the problem comes down to the structleak_plugin.so object file. there were 95 bug and regression fixes that differentiate gcc 11.1 and 11.2 which *shouldn't* cause any problems, but my thought is that one of those fixes affected structleak_plugin. i didn't immediately think it could be the compiler version either. it's normal to be able to run software compiled with one version of gcc across a good number of gcc updates, and certainly across any minor updates within a major version, so it's not that i don;t understand where you're coming from. i have a few quickie C programs i've coded up over time as utilities that i haven't ever recompiled, and they still work perfectly, despite the fact that gcc has been updated many times since they were first compiled/linked. however, even the name of the plugin suggests it was created to patch over some kind of odd, leaking structure issue to begin with, and i have to wonder if one (or more) of the 95 fixes fundamentally changed the compiler output for that module specifically (or, possibly, other modules too, but we haven;t come across them yet). so, this might be a one-off issue.

IMHO a compiler should expose a stable enough API, so that any specific range of compilers can build the same kernel. This is not a nice thing if the kernel starts to depend on specific gcc versions.

i concur. i didn't open this thread because i thought someone was doing something wrong, but because it was anomalous behaviour. it took a couple of rebuilds to start me thinking it might be a compiler problem. when i started rebuilding 5.17.1 i noticed the config's GCC setting was 11.1 and, even though the build resets GCC to the version you're actually using, it made me wonder if y'all are using 11.1 to build the stable version of the kernel even though the stable compiler version is 11.2.

i'll take a look at structleak_plugin.cpp (and .h) and see if there's anything obvious, even though i'm not really sure what i'm looking for.

abaumann · 2022-04-16 19:10:20

No worries. :-)
I checked the config files now and synched the gcc and binutils versions.
stable/testing and staging have the same toolchain now.
The differences have two major causes:
1) the toolchain differs in the repos
2) the maintainer (aka me, myself and I) does 'make oldconfig' or so on
a 64-bit host and then checks in those kernel config files.

I hope, the next kernel gets better..

In an ideal world I would add a system test doing: install kernel, checkout software
from the AUR requiring a dkms build, build and install it and then load the module.
Never got to automatized testing.. :-)

levi · 2022-04-16 23:02:28

jghodd wrote:

i have a few quickie C programs i've coded up over time as utilities that i haven't ever recompiled, and they still work perfectly, despite the fact that gcc has been updated many times since they were first compiled/linked. however, even the name of the plugin suggests it was created to patch over some kind of odd, leaking structure issue to begin with, and i have to wonder if one (or more) of the 95 fixes fundamentally changed the compiler output for that module specifically (or, possibly, other modules too, but we haven;t come across them yet). so, this might be a one-off issue.

I assume there's something different between a statically compiled binary and whatever these dkms build are doing. But it's a good thing that statically compiled binaries still work, else our old games and retro software would stop working. I guess the pressure for building dkms software is less long lasting. Of course, if you still had unlinked object files for your old software, I'd be surprised if you could still link them successfully after so many years, but that only really depends on the behaviour of the linker, which normally should work between kernel versions. But statically compiled binaries only really depend on the ABI of the processor, which depending on your processor is somewhat dependable.

jghodd · 2022-04-17 17:42:56

@abaumann - thanks much. that *should* fix things up. i did check structleak_plugin.c and a couple of things caught my eye that follow back to include/linux/compiler.h - which did change between gcc 11.1 and 11.2 (02/02/22). it's not a file that's changed often, but apparently one to keep an eye on.

abaumann · 2022-04-18 08:08:37

I found another early "guinea pig": kmod. This one fails to build on compiler mismatches in the kernel header files..

trujillo-gc · 2022-04-18 11:00:09

Sorry for getting off topic in this thread.
It is to say, that I would like to help, but unfortunately I do not have the knowledge to do so.
What I can say is that you are doing an excellent job. Those of us who enjoy using archlinux 32 appreciate your effort and would never dream of complaining.
Big thanks!
(excuse my english)

nateewy · 2022-04-19 20:03:32

Were the changes made before the zen and linux kernel were made? I get the same problem with kernel 5.17.3.

Last edited by nateewy (2022-04-19 20:04:11)

jghodd · 2022-04-19 20:47:05

am having to do another kernel rebuild, this time with 5.17.3.

Last edited by jghodd (2022-04-19 20:47:30)

jghodd · 2022-04-19 22:17:05

@nateewy - i'm seeing a date of 04/16 on the 5.17.3 kernel from the arch32 core repository.

abaumann · 2022-04-20 06:09:48

So if linux-<variant> and linux-<variant>-headers are both 5.17.3 and the gcc versions in the config (zgrep VERSION /proc/config.gz) match:

CONFIG_GCC_VERSION=110200
CONFIG_CLANG_VERSION=0
CONFIG_AS_VERSION=23800
CONFIG_LD_VERSION=23800
CONFIG_LD_VERSION=23800
CONFIG_LLD_VERSION=0

and gcc and binutils are the same versions as in /proc/config, then it should work.. if not, I don't know how to build a kernel. :-)

For the i486 subarchitecture there is currently a huge backlog, so things are known to fail there..

jghodd · 2022-04-20 16:45:34

@abaumann - yup. you have the right compiler and linker versions. my rebuild works as expected using the same versions.

CONFIG_GCC_VERSION=110200
CONFIG_CLANG_VERSION=0
CONFIG_AS_VERSION=23800
CONFIG_LD_VERSION=23800
CONFIG_LLD_VERSION=0

i'm using the same config (+config.i686) you are. what else could be causing a mismatch?

jghodd · 2022-04-20 17:27:55

@abaumann - what version of glibc are you using? i'm using glibc-2.35-3.0

abaumann · 2022-04-20 17:37:05

glibc 2.35-3.0

jghodd · 2022-04-20 18:17:03

hm. well, wth. what are we missing? the only deviations i use from stock arch32 are go-2:1.17.2-1, kdecoration-5.24.3-1.0, kismet-2022_02_R1-1.1 and mixxx-2.3.2-1.2. i don;t see any of those making any difference to the kernel build...

JerryXiao · 2022-05-05 15:18:28

Same issue with linux 5.17.5.arch1-1.2 subarch i686

All packages are from the arch32 repo and no deviations whatsoever.

abaumann · 2022-05-05 15:43:50

Ok, I can reproduce this with (from FS#63644):

sudo pacman -S linux-headers
echo 'int main() { return 0; }' > test.c
gcc -fplugin=/usr/lib/modules/5.17.5-arch1-1.2/build/scripts/gcc-plugins/structleak_plugin.so test.c
cc1: error: incompatible gcc/plugin versions
cc1: error: failed to initialize plugin /usr/lib/modules/5.17.5-arch1-1.2/build/scripts/gcc-plugins/structleak_plugin.so

So, a gcc helper plugin is part of the kernel header files used for compilation (this is a weird decision, but, oh well).

config.i686 contains:

# Automatically generated file; DO NOT EDIT.
# Linux/x86 5.17.5-arch1 Kernel Configuration
#
CONFIG_CC_VERSION_TEXT="gcc (GCC) 11.2.0"
CONFIG_CC_IS_GCC=y
CONFIG_GCC_VERSION=110200
CONFIG_CLANG_VERSION=0
CONFIG_AS_IS_GNU=y
CONFIG_AS_VERSION=23800

gcc --version
gcc (GCC) 11.2.0
ld --version
GNU ld (GNU Binutils) 2.38

I'm lost. :-)
The toolchain has now been rebuilt and stable since 2 weeks or so, and the kernel has been
compiled yesterday..

Can you check, if you have the same versions?

What I can think of, is, that something is wrong with the plugin itself (some 32-bit issue), I cannot
find a trivial reason (for now)..

JerryXiao · 2022-05-06 02:01:29

@abaumann I have exactly the same versions. In fact, the system was just pacstraped like 30 minutes ago.

abaumann · 2022-05-06 08:21:10

I was digging a little bit into the source code:

scripts/gcc-plugins/structleak_plugin.c
if (!plugin_default_version_check(version, &gcc_version)) {
error(G_("incompatible gcc/plugin versions"));
return 1;
}

I fail to find the definition of plugin_default_version_check?!

abaumann · 2022-05-06 08:29:22

mmh. must be part of gcc..

Arch Linux

#26 2022-04-08 23:43:09

Re: [SOLVED] 5.16.12 dkms modules not building

#27 2022-04-09 06:37:03

Re: [SOLVED] 5.16.12 dkms modules not building

#28 2022-04-12 21:48:27

Re: [SOLVED] 5.16.12 dkms modules not building

#29 2022-04-15 17:48:49

Re: [SOLVED] 5.16.12 dkms modules not building

#30 2022-04-15 19:07:04

Re: [SOLVED] 5.16.12 dkms modules not building

#31 2022-04-16 07:46:38

Re: [SOLVED] 5.16.12 dkms modules not building

#32 2022-04-16 18:02:25

Re: [SOLVED] 5.16.12 dkms modules not building

#33 2022-04-16 19:10:20

Re: [SOLVED] 5.16.12 dkms modules not building

#34 2022-04-16 23:02:28

Re: [SOLVED] 5.16.12 dkms modules not building

#35 2022-04-17 17:42:56

Re: [SOLVED] 5.16.12 dkms modules not building

#36 2022-04-18 08:08:37

Re: [SOLVED] 5.16.12 dkms modules not building

#37 2022-04-18 11:00:09

Re: [SOLVED] 5.16.12 dkms modules not building

#38 2022-04-19 20:03:32

Re: [SOLVED] 5.16.12 dkms modules not building

#39 2022-04-19 20:47:05

Re: [SOLVED] 5.16.12 dkms modules not building

#40 2022-04-19 22:17:05

Re: [SOLVED] 5.16.12 dkms modules not building

#41 2022-04-20 06:09:48

Re: [SOLVED] 5.16.12 dkms modules not building

#42 2022-04-20 16:45:34

Re: [SOLVED] 5.16.12 dkms modules not building

#43 2022-04-20 17:27:55

Re: [SOLVED] 5.16.12 dkms modules not building

#44 2022-04-20 17:37:05

Re: [SOLVED] 5.16.12 dkms modules not building

#45 2022-04-20 18:17:03

Re: [SOLVED] 5.16.12 dkms modules not building

#46 2022-05-05 15:18:28

Re: [SOLVED] 5.16.12 dkms modules not building

#47 2022-05-05 15:43:50

Re: [SOLVED] 5.16.12 dkms modules not building

#48 2022-05-06 02:01:29

Re: [SOLVED] 5.16.12 dkms modules not building

#49 2022-05-06 08:21:10

Re: [SOLVED] 5.16.12 dkms modules not building

#50 2022-05-06 08:29:22

Re: [SOLVED] 5.16.12 dkms modules not building

Board footer