[OE-core] Kernel check of PV versus kernel source

[Nathan Rossi]

On Thu, Sep 29, 2016 at 10:10 AM, Philip Balister <philip at balister.org> wrote:
> This check seems to have gone in recently and doesn't appear to work
> very well.
>
> ERROR: linux-xlnx-3.14-xilinx+gitAUTOINC+2b48a8aeea-r0
> do_kernel_version_sanity_check: Package Version
> (3.14-xilinx+gitAUTOINC+2b48a8aeea) does not match of kernel being built
> (4.6). Please update the PV variable to match the kernel source.
>
> I hunt around and the kernel source appears to be 3.14.2. Sometimes the
> check passes, sometimes it asks for 3.14.2 actual PV is something like
> 3.14-xilinx+gitAUTOINC+2b48a8aeea-r0
>
> Clearly, something isn't deterministic here.

It turns out "do_kernel_checkout" is not sufficiently late enough such
that the kernel source is static. This is because the
do_kernel_metadata also might change the tree (or checkout a different
branch). This is a quirk that linux-xlnx uses see paragraph below.

However kernel_version_sanity_check should avoid racing with
kernel_metadata, I have sent out a patch to resolve this and have CC'd
you Philip.

------

There is some strange behaviour that linux-xlnx was working with due
to how the some of the linux-yocto code/metadata stuff works and
issues with branchless checkouts (I don't remember specifics, it has
been a while since I implemented that). So it was setup to trick it
into checking out master of the repo, then in the kernel_metadata task
resetting master to the SRCREV. It appears that trick is no longer
needed though (KBRANCH = "" now works fine), so I will fix up
linux-xlnx, which will resolve the need to do the trick.

Note: linux-xlnx source tree does tags for releases and does not have
any requirement that the tag is on a branch.

Regards,
Nathan

>
> Philip
> --
> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core at lists.openembedded.org
> http://lists.openembedded.org/mailman/listinfo/openembedded-core