Re: [Buildroot] Upgrading buildroot and rootfs/kernel compatibility

[Thomas Petazzoni via buildroot <buildroot@buildroot.org>]

Hello Grant,

On Fri, 13 Sep 2024 19:29:52 -0000 (UTC)
Grant Edwards <grant.b.edwards@gmail.com> wrote:

> I started with a buildroot 2020.02.7 configuration provided by a
> silicon vendor that used a linaro-6.3.1-2017.02 toolchain.  That
> worked OK after I manually upgraded a couple packages that were too
> old to build using my host system's gcc 13.3.
> 
> Now I'm trying to upgrade buildroot to 2024.02.6.  First I tried using
> the same toolchain as above.  That failed. AFAICT, it was because
> Fortran support was enabled in the toolchain, and BF 2024 can't
> tolerate that unless Fortran support is also enabled somewhere else
> (it wasn't clear where).

Do you have more details on what happened? Your explanation "BF 2024
can't tolerate that unless Fortran support is also enabled somewhere
else" is very vague, and I believe using your old toolchain should
normally still work, so I would consider the fact that BR 2024.02
doesn't work with your older toolchain to potentially be a bug.

> Then I tried the "standard" external toolchains listed in the
> 2024.02.6 menuconfig.
> 
> ARM
> 
>    With arm-gnu-toolchain-13.2.rel1 the rootfs built cleanly, but the
>    target panicked as soon as the init process was started. I assume
>    that's some sort of Linux system-call incompatibility between the
>    glibc included in the ARM toolchain and my 4.19.142 kernel (which I
>    think was built using buildroot 2020 -- probably using the Linaro
>    6.3.1-2017.02 toolchain).

This toolchain is using 4.20.x kernel headers, so technically it's a
bit too new for your kernel, even though glibc has some compatibility
stuff. But I believe the problem is probably not here. Maybe the issue
is that your toolchain is an EABI toolchain, while the new one is
definitely EABIhf, and maybe you don't have CONFIG_VFP=y in your
kernel, or something like this? Could you provide your kernel
configuration?

> Linaro
> 
>    With the linaro-7.3.1-2018.05 toolchain, the rootfs builds cleanly
>    and the resulting image appears to work, though it's about 20%
>    larger than before with the same set of packages selected.  Is that
>    sort of size increase typical? (I know, it depends on the
>    packages.)

A 20% increase obviously seems like a lot. Is this only with this
toolchain, or also with the ARM 13.2 toolchain above?

Please note that Buildroot switched from -Os by default to -O2 by
default, so images are generally slightly bigger now, but offer better
performance. However, a 20% increase seems like a lot, and I wouldn't
expect the switch from -Os to -O2 to have caused such a significant
increase in the image size. Is this 20% increase all over the place
(like all binaries are bigger), or can you track it down to some
specific package? Maybe a specific package is now installing a lot more
stuff than it used to. You can track this after doing a full build, by
running "make graph-size", which will collect some nice statistics
information about how much stuff has been installed by each package.

Best regards,

Thomas
-- 
Thomas Petazzoni, co-owner and CEO, Bootlin
Embedded Linux and Kernel engineering and training
https://bootlin.com
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot