Re: [OE-core] kernel-module dependency mystery

[Max <max.oss.09@gmail.com>]

Hi Mike

Am Mittwoch, dem 23.04.2025 um 14:34 +0200 schrieb Mike Looijmans via lists.openembedded.org:
> I built the same image for 5 different machines. Only one of them fails 
> with the logging below. Note that all machines install the 
> "kernel-module-dyplo" package, and on 4 machines that succeeds just fine.
> 
> 
>   * Solver encountered 2 problem(s):
>   * Problem 1/2:
>   *   - conflicting requests
>   *   - package 
> kernel-module-dyplo-1.1+154+bb94f350+bb94f3599b-r0.tdpzu9 requires 
> kernel-module-dyplo-pcie-6.6.40-xilinx-g6a2cb0f14db5, but none of the 
> providers can be installed
>   *   - nothing provides kernel-6.6.40-xilinx-g6a2cb0f14db5 needed by 
> kernel-module-dyplo-6.6.40-xilinx-g6a2cb0f14db5-1.1+154+bb94f350+bb94f3599b-r0.tdpzu9
>   *   - nothing provides kernel-6.6.40-xilinx-g6a2cb0f14db5 needed by 
> kernel-module-dyplo-pcie-6.6.40-xilinx-g6a2cb0f14db5-1.1+154+bb94f350+bb94f3599b-r0.tdpzu9
>   *
>   * Solution 1:
>   *   - do not ask to install a package providing kernel-module-dyplo
> 
>   * Problem 2/2:
>   *   - conflicting requests
>   *   - package 

> kernel-module-dyplo-1.1+154+bb94f350+bb94f3599b-r0.tdpzu9 requires 
> kernel-module-dyplo-pcie-6.6.40-xilinx-g6a2cb0f14db5, but none of the 
> providers can be installed
>   *   - nothing provides kernel-6.6.40-xilinx-g6a2cb0f14db5 needed by 
> kernel-module-dyplo-6.6.40-xilinx-g6a2cb0f14db5-1.1+154+bb94f350+bb94f3599b-r0.tdpzu9
>   *   - nothing provides kernel-6.6.40-xilinx-g6a2cb0f14db5 needed by 
> kernel-module-dyplo-pcie-6.6.40-xilinx-g6a2cb0f14db5-1.1+154+bb94f350+bb94f3599b-r0.tdpzu9
>   *
>   * Solution 1:
>   *   - do not ask to install a package providing kernel-module-dyplo
> 

We had an issue with the exact same symptom and had a hard time
to find a fix. It turned out that:

- The kernel is built with a patch applied from SRC_URI
AND
- We had LOCALVERSION_AUTO=y in the kernel config
AND
- The kernel is in sstate
AND
- The kernel sources no longer exist under tmp/work-shared
AND
- The external kernel module is not taken from sstate

If patches are applied in a kernel recipe they are applied
with 'git am' and a new hash value will be present with
each run of do_patch.

If LOCALVERSION_AUTO=y is in the kernel config that new
hash becomes part of the kernel and external kernel module
packages.

As the kernel is from an older do_patch and the module
from a newer the hashes used in building differ and you
get the above OE build error. Even if it wouldn't the
different hashes become part of the VERMAGIC and the
kernel would refuse to load the module.

Our solution was to no longer build with LOCALVERSION_AUTO
and since then we no longer see the issue.

Regards
Max

> I'm on "scarthgap" branch.
> 
> The "kernel-module-dyplo" recipe can be found here:
> 
> https://github.com/topic-embedded-products/meta-dyplo/blob/master/recipes-dyplo/dyplo-core/kernel-module-dyplo.bb
> 
> The isue at first glance appears to be that 
> "kernel-6.6.40-xilinx-g6a2cb0f14db5" mixes the git version hash from the 
> module itself and the git hash of the kernel. There's no such package, 
> it should have been "kernel-6.6.40-xilinx-gd7b52381656b"
> 
> 
> -=-=-=-=-=-=-=-=-=-=-=-
> Links: You receive all messages sent to this group.
> View/Reply Online (#215287): https://lists.openembedded.org/g/openembedded-core/message/215287
> Mute This Topic: https://lists.openembedded.org/mt/112412917/3617484
> Group Owner: openembedded-core+owner@lists.openembedded.org
> Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [max.oss.09@gmail.com]
> -=-=-=-=-=-=-=-=-=-=-=-
>