On Thu, 9 May 2019 at 20:08, Nick Desaulniers ndesaulniers@google.com wrote:
From: Nathan Chancellor natechancellor@gmail.com Date: Thu, May 9, 2019 at 11:38 AM To: Nick Desaulniers Cc: Matt Hart, Mark Brown, Arnd Bergmann, Anmar Oueja, Kernel Build Reports Mailman List, Tri Vo, kernelci.org bot, Rasmus Villemoes
(sorry, gmail is not formatting the above response headers correctly, hopefully a bug in today's build)
On Thu, May 09, 2019 at 11:26:39AM -0700, Nick Desaulniers wrote:
From the line:
Full Build Summary: https://staging.kernelci.org/build/next-clang/branch/master/kernel/next-2019...
it seems like there's lots of builds being done (oh man, lots of red, we'll need to dive deeper into arm32). As long as we have the above link (and I remember to look there), I think we should be fine. It
I'll start looking at it later tonight. Looks like a large majority of them are assembler related.
/usr/bin/as: unrecognized option '-EL'
Something funny is going on with assembler selection, /usr/bin/as is probably the x86 assembler, which won't work for arm.
I thought we had armv7_defconfigs green in our TravisCI runs? https://travis-ci.com/ClangBuiltLinux/continuous-integration/builds/11116730...
I wonder if the build machines maybe don't have arm-linux-gnueabihf-as (or arm-linux-gnueabihf-binutils package) installed on them? Matt, can you confirm?
The build container is built from this: https://github.com/kernelci/kernelci-core/blob/master/jenkins/dockerfiles/cl...
So yes, looks like it only has binutils-aarch64-linux-gnu installed, and we need binutils-arm-linux-gnueabihf from debian as well.
I'll just check before I put up a PR, that installing the extra package wont break arm64 builds.
I should send Masahiro a patch to Kbuild that checks that these actually exist, and fail with an actionable error message saying something like "yo, I couldn't find the assembler you told me to use, what are you doing?" rather than "oh I didn't recognize this flag (cause it's for arm and I'm x86)." -- Thanks, ~Nick Desaulniers