On Sat, Sep 11, 2021 at 12:56:37PM +0100, Pierre Langlois wrote:
> I just tried out 5.14.2 on my rockpro64 (rk3399 chip) and ethernet isn't
> working :-/. It probably also affects the pinebook pro.
> 
> Doing a bit of searching, I see it was reported upstream and this patch
> introduced the issue https://patchwork.kernel.org/project/linux-arm-kernel/patch/20210823143754.14294-1-michael.riesch@wolfvision.net/
> 
> I see Manjaro has patched their kernel to revert the patch as well:
> https://gitlab.manjaro.org/manjaro-arm/packages/core/linux/-/commit/5e87e6f3b764d64bbb5b183e379a5fe64c3646ad

Thanks for the report!

> I suppose we could also apply that patch, although I'm sure it'll get
> fixed upstream in a point release. We probably don't want to delay the 5.14
> switch too much, maybe we could keep the linux-libre-5.13 variable
> around for a little longer though? If not, I'm sure I could just use
> 5.10 though.

When they are two stable kernels [0], I typically keep the previous
stable kernel as the Guix "default" until it stops being supported
upstream, and then the latest stable kernel is made the default, and the
older one is removed from Guix.

I see in the upstream report that Sasha Levin is aware of the problem,
and he is one of the stable kernel maintainers. So 5.13 should continue
to work, and 5.14 should be fixed soon.

We don't usually patch our kernels except for really critical problems;
We use the stable and long-term support releases directly from upstream
in order to avoid the old-school "custom distro kernel" thing.

How about we keep things as they are now, and then reconsider if 5.13 is
due for removal and the bug still exists in 5.14? I think that most
rk3399 users in Guix should be okay for now, since the default kernel,
5.13, still works for them, right?

[0] Kernel releases are classified as "stable" here:
https://www.kernel.org/