[OE-core] Summary of kernel-devsrc package issues (Sumo)

[Bas Mevissen]

On 2018-06-04 14:32, Bruce Ashfield wrote:
> On Mon, Jun 4, 2018 at 5:04 AM, Bas Mevissen <abuse at basmevissen.nl>
> wrote:
> 
>> Hi all,
>> 
>> I ran into several issues with kernel-devsrc when building Poky Sumu
>> with the meta-freescale layer for an Freescale i.mx6sx ARM board. In
>> this mail(thread) I would like to collect them all and discuss the
>> possible options to resolve them.
>> 
>> * Install dependency issues (both during image creation and manual
>> install on the board):
>> 
>> The kernel-devsrc spec file defines a number of auto-generated
>> run-time dependencies. These include
>> (1) /bin/awk
>> (2) /usr/bin/python
>> 
>> (1) is due to scripts/ver_linux being an awk script using /bin/awk
>> in its shebang in various (vendor) kernels, even recent ones.
>> Instead of fixing all those kernel sources, I would suggest:
>> 1a) adding a link to /bin/awk in the awk or the kernel-devsrc
>> package
>> 1b) automatically fix the kernel source included in the
>> kernel-devsrc package
>> 1c) having a more helpful error message?
>> 
>> (2) is normally fine, but on my build, there is only
>> /usr/bin/python3. A manually created symlink fixed the issue. I'm
>> wondering what a more permanent solution would be:
>> 2a) Fix all Python script in the installed kernel source like 1b)?
>> (the python3-native package does create such a symlink, but not the
>> target one!)
>> 2b) See whether there is a package that already creates that symlink
>> and make kernel-devsrc depend on it?
>> 
>> * External module build issue:
>> 
>> When building an external module (rtl8812au driver) on the board
>> using kernel-devsrc, I ran into the following issues:
>> 
>> (3) No symlink from /lib/modules/`uname -r`/build to the kernel
>> source. Solution:
>> 3a) Add this symlink (and optionally "source") to kernel-devsrc,
>> comparable to Fedora kernel-headers package
>> 
>> (4) The scripts where not build, resulting in build error. Solution:
>> 4b) Pre-build scripts ("make scripts") in kernel-devsrc. Should be
>> ok, as the package is already arch dependent.
>> 
>> (remarkably, there is mention of both in the kernel-devsrc.bb [1]
>> file, but I think the current solutions are wrong as the package is
>> not ready-to-use currently)
>> 
>> * Multiple kernel support:
>> 
>> When developing, multiple kernel versions might be installed.
>> Current version of the package installs in /usr/src/kernel without
>> the kver in the patch.
>> 
>> (5) So you can have only 1 set of kernel sources installed.
>> Proposal:
>> (5a) Install in /usr/src/kernels/`uname r` following the way Fedora
>> does it.
>> 
>> * Huge size of package
>> 
>> (6) The package contains an awful lot of files and is huge. Not all
>> of them are needed to compile external kernel modules
>> 6a) Remove the sources and only provide what is needed to build
>> external kernel modules (like fedora)
>> 6b) Create the package suggested at 6a) (or add it to kernel-dev)
>> and keep the devsrc as the full-source alternative.
>> 
>> Your thoughts? I'm more than happy to fix the issues and provide
>> patches. However, I need input on what direction to go on the issues
>> I ran into until now.
> 
> There's already a completely re-written kernel-devsrc that I posted in
> the last
> dev cycle, which will be re-sent in the next few days. The only reason
> it didn't
> merge was due to a multlib issue, which should be sorted out now.
> 

I wasn't aware of that. I looked up the one you posted in March and that 
already addresses a lot of the issues I listed. Nice!

> So please, don't do any patches on top of the existing devsrc
> structure, since
> it is essentially obsolete.
> 

Yes, definitely. I'm looking forward to see your updated version.


> Cheers,
> 
> Bruce
> 
>> Thanks,
>> 
>> Bas.
>> 

Cheers,

Bas.

>> --
>> _______________________________________________
>> Openembedded-core mailing list
>> Openembedded-core at lists.openembedded.org
>> http://lists.openembedded.org/mailman/listinfo/openembedded-core [2]
> 
> --
> "Thou shalt not follow the NULL pointer, for chaos and madness await
> thee at its end"
> 
> Links:
> ------
> [1] http://kernel-devsrc.bb
> [2] http://lists.openembedded.org/mailman/listinfo/openembedded-core