blk-mq hangs easily with LLVM+clang test suite

[David Zarzycki <dave@xxxxxx>]

Hello,

I think I’ve narrowed down a kernel hang to the multi-queue block layer when running the LLVM+clang test suite. Here is what I’ve gathered:

1) The hang does *not* reproduce on tmpfs or a SATA SSD drive.
2) The hang does reproduce on blk-mq devices (NVMe or loopback to a file in tmpfs).
3) The hang does reproduce with ext4 or xfs, so the filesystem doesn’t seem to matter.

How can I help debug this further?

Thanks,
Dave

For reference, reproduction steps:

cd /to/a/blk-mq/partition
git clone https://git.llvm.org/git/llvm.git
cd llvm/tools
git clone https://git.llvm.org/git/clang.git
cd ..
mkdir t
cd t
cmake -G Ninja \
	-DCMAKE_BUILD_TYPE=Release \
	-DLLVM_ENABLE_ASSERTIONS=TRUE \
	-DLLVM_TARGETS_TO_BUILD=X86 ..
# It only takes a few runs of the test suite on a 48-core system
while true
do
	ninja check-all
done