Regression on nightly about sancov_cntrs

[catenacyber]

Code

Suricata is a project with both C and Rust code.
Its fuzz targets get compiled by oss-fuzz

To reproduce the bug, you can clone github.com/google/oss-fuzz
and run python infra/helper.py build_fuzzers suricata

I expected to see this happen: Compilation succeeds

Instead, this happened:
Compilation failed with multiple similar errors

/usr/bin/ld: __sancov_cntrs has both ordered [`__sancov_cntrs[ROHashInitFinalize]' in libsuricata_c.a(util-rohash.o)] and unordered [`__sancov_cntrs[_ZN100_$LT$hashbrown..raw..RawIterHashInner$LT$A$GT$$u20$as$u20$core..iter..traits..iterator..Iterator$GT$4next17hce20d46d046f29e5E]' in ../rust/target/x86_64-unknown-linux-gnu/release/libsuricata_rust.a(suricata_rust-6f68cb102e8bc119.suricata_rust.7485b842-cgu.0.rcgu.o)] sections

A workaround is to use ld.gold instead of ld

Version it worked on

It most recently worked on: rustc 1.56.0-nightly (a0035916e 2021-08-20

Version with regression

rustc 1.56.0-nightly (d3e2578c3 2021-08-21) was the first to fail

Still failing on
rustc --version --verbose:

rustc 1.56.0-nightly (af140757b 2021-08-22)
binary: rustc
commit-hash: af140757b4cb1a60d107c690720311ba8e06e7de
commit-date: 2021-08-22
host: x86_64-unknown-linux-gnu
release: 1.56.0-nightly
LLVM version: 13.0.0

@rustbot modify labels: +regression-from-nightly-to-nightly -regression-untriaged

[nikic]

Very likely related to LLVM 13 upgrade. I'm not very familiar with this functionality, is there possibly a requirement that the Rust and C code are compiled with the same LLVM version? That is, you need to use clang 13 for the C code?

[catenacyber]

C code is compiled with
clang version 12.0.0 (https://github.com/llvm/llvm-project.git 6de4865545da73687dd6d28d153cd345ed5e7918)

is there possibly a requirement that the Rust and C code are compiled with the same LLVM version?

The workaround of using ld.gold seems to show this is not a requirement, right ?

[hkratz]

This change in LLVM 13 might be related: https://maskray.me/blog/2021-01-31-metadata-sections-comdat-and-shf-link-order

[catenacyber]

It looks like
export RUSTFLAGS="$RUSTFLAGS -Zsanitizer=address -Cpasses=sancov -Cllvm-args=-sanitizer-coverage-level=4 -Cllvm-args=-sanitizer-coverage-trace-compares -Cllvm-args=-sanitizer-coverage-inline-8bit-counters -Cllvm-args=-sanitizer-coverage-trace-geps -Cllvm-args=-sanitizer-coverage-prune-blocks=0 -Cllvm-args=-sanitizer-coverage-pc-table -Clink-dead-code -Cllvm-args=-sanitizer-coverage-stack-depth" is responsible

[catenacyber]

So, I rather think this is a bug in the linker, so closing here