CI: fakeroot segfaults in SDK

[aparcar]

The CI uses SDK snapshots to build changed packages for multiple architectures. It currently fails for arm_cortex-a9_vfpv3-d16
with the following error message:

/functions.sh"; echo "default_prerm \$0 \$@"; ) > prerm; chmod 0755 postinst prerm;  )
install -d -m0755 /home/build/openwrt/bin/targets/mvebu/cortexa9/packages
/home/build/openwrt/staging_dir/host/bin/fakeroot /home/build/openwrt/scripts/ipkg-build -m "" /home/build/openwrt/build_dir/target-arm_cortex-a9+vfpv3-d16_musl_eabi/toolchain/ipkg-arm_cortex-a9_vfpv3-d16/libgcc /home/build/openwrt/bin/targets/mvebu/cortexa9/packages
/home/build/openwrt/staging_dir/host/bin/fakeroot: line 185:  8682 Segmentation fault      (core dumped) FAKEROOTKEY=$FAKEROOTKEY LD_LIBRARY_PATH="$PATHS" LD_PRELOAD="$LIB" "$@"
make[2]: Leaving directory '/home/build/openwrt/package/toolchain'
make[2]: *** [Makefile:764: /home/build/openwrt/bin/targets/mvebu/cortexa9/packages/libgcc1_8.4.0-2_arm_cortex-a9_vfpv3-d16.ipk] Error 139
make[1]: *** [package/Makefile:114: package/toolchain/compile] Error 2

It used to work for a while but now crashes. As it only fails for the single architecture I'm not sure if it's an issue with the two recent fakeroot commits over at openwrt.org:

openwrt/openwrt@0fdbd2e
openwrt/openwrt@f93cb5c

[EricLuehrsen]

cc: @jow- @dangowrt

[neheb]

Does this happen with only the SDK?

[danvd]

Minor addition: Upon submitting any PR the test "Test Build / Test arm_cortex-a9_vfpv3-d16 (pull_request)" is also failing due to this segfault.

/home/build/openwrt/staging_dir/host/bin/fakeroot: line 185: 31187 Segmentation fault      (core dumped) FAKEROOTKEY=$FAKEROOTKEY LD_LIBRARY_PATH="$PATHS" LD_PRELOAD="$LIB" "$@"

[aparcar]

I couldn't reproduce the issue locally...

[aparcar]

Looking at the docker CI builds, it's failing since the October 30. Which happens to be the same date as the above mentioned patches were added. In the Docker CI the x86/64 target fails, so it is not a problem of mvebu/cortex-a9 but all targets.

[aparcar]

Failing Docker CI: https://gitlab.com/openwrt/docker/-/pipelines
Failing Docker CI test: https://gitlab.com/openwrt/docker/-/jobs/837338425

[jefferyto]

This is not just a CI issue - I am seeing this error locally with the armvirt-32, armvirt-64, malta-be, x86-generic and x86-64 SDKs (8 Nov) on Ubuntu 20.10.

I can't update any packages without a working SDK 🙁

Edit: Manually calling $(STAGING_DIR_HOST)/bin/fakeroot (with STAGING_DIR_HOST set) results in the segfault.

[jefferyto]

Passing the -l $(FAKEROOT_SO) option with the absolute path to libfakeroot.so (removed in openwrt/openwrt@0fdbd2e) works around the segfault.

It appears it isn't finding the right shared library for some reason.

[dangowrt]

@jow- Any idea why this could be happening? It's quite weird because it seems to depend on which of the buildhosts made the SDK for it to have that bug or not...

[ynezz]

Should be fixed via https://git.openwrt.org/d4521fb132266fd6c94796e19d3d1e925847338d