Re: [PATCH 6.1 000/127] 6.1.45-rc1 review

From: Nathan Chancellor
Date: Fri Aug 11 2023 - 00:13:49 EST


On Thu, Aug 10, 2023 at 08:41:42PM -0700, Guenter Roeck wrote:
> On 8/10/23 20:22, Naresh Kamboju wrote:
> > Build errors:
> > -----
> > ld.lld: error: ./arch/x86/kernel/vmlinux.lds:193: at least one side of
> > the expression must be absolute
> > make[2]: *** [scripts/Makefile.vmlinux:34: vmlinux] Error 1
> >
>
> We see that with v5.10.y, v5.15.y, and v6.1.y when building ChromeOS images with
> clang/lld. There are additional problems with LTO and the built-in assembler. See
> https://www.linuxquestions.org/questions/slackware-14/error-building-kernel-6-1-44-on-current-with-clang-4175727865/
> for a summary.

Yup, we have issues open for all of those:

https://github.com/ClangBuiltLinux/linux/issues/1907
https://github.com/ClangBuiltLinux/linux/issues/1909
https://github.com/ClangBuiltLinux/linux/issues/1911

1907 is fixed in -tip and I am sure it will make -rc6 [1].

1909 is fixed with [2] but it is sitting in x86/core (i.e., slated for
next merge window). I am guessing at the time it was picked up, it was
not fixing a noticeable issue, which is obviously not the case now. Nick
reached out to the -tip folks on IRC to inquire about getting that
applied to a branch that is going to Linus soon, as it is more of a
process issue since it has conflicts with SRSO and an separate issue
that was pointed out post-acceptance (which I addressed and pushed to
[3] for testing). I never saw a response there (which is understandable,
it is a busy time...) so looping the -tip folks in now, just to make
sure it does not get lost (apologies if this is noise).

1911 is still being investigated (some additional eyes on it would not
hurt).

[1]: https://git.kernel.org/tip/cbe8ded48b939b9d55d2c5589ab56caa7b530709
[2]: https://git.kernel.org/tip/973ab2d61f33dc85212c486e624af348c4eeb5c9
[3]: https://github.com/ClangBuiltLinux/linux/commit/150c42407f87463c27a2459e06845965291d9973

> As far as I can see none of those problems has been fixed in the upstream kernel.

Indeed, embargos are fun... :)

Cheers,
Nathan