Illegal instruction (core dumped) on RISC-V

[Denis Ovsienko]

$ clang --version
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0. Program arguments: clang --version
1. Compilation construction
Stack dump without symbol names (ensure you have llvm-symbolizer in your PATH or set the environment var `LLVM_SYMBOLIZER_PATH` to point to it):
/lib/riscv64-linux-gnu/libLLVM-14.so.1(_ZN4llvm3sys15PrintStackTraceERNS_11raw_ostreamEi+0x38)[0x3f8bbaf6ee]
Illegal instruction (core dumped)

ii clang-14 1:14.0.0-1ubuntu1 riscv64 C, C++ and Objective-C compiler
ii libllvm14:riscv64 1:14.0.0-1ubuntu1 riscv64 Modular compiler and toolchain technologies, runtime library

$ uname -a
Linux linux-riscv64 5.17.0-1007-allwinner #7-Ubuntu SMP Fri Oct 14 15:43:39 UTC 2022 riscv64 riscv64 riscv64 GNU/Linux

Steps to reproduce:
1. Install Ubuntu 22.04.1 RISC-V from https://cdimage.ubuntu.com/releases/22.04.1/release/ubuntu-22.04.1-preinstalled-server-riscv64+nezha.img.xz on a Nezha D1 board (1GB RAM).
2. Install all updates.
3. apt-get install clang
4. Run clang.

Let me know in case you need additional diagnostics. SSH access can be provided on request to project developers.

[Alexandre Ghiti (alexghiti)]

Thanks for all the steps, the issue is easily reproducible.

The illegal instruction trap is caused by a 'fence.tso' instruction that the C906 core in the D1 SoC fails to decode: it is known issue and fixed in opensbi 1.1 by emulating a fence instead.

I filled an SRU here https://bugs.launchpad.net/ubuntu/+source/llvm-defaults/+bug/1995080 and I'm testing the fix in my ppa:alexghiti/riscv.

Thanks again for the report,

Alex

[Denis Ovsienko (dovsienko)]

For what it's worth, the current version does not immediately fail after start:

$ uname -r
5.17.0-1010-allwinner

$ clang --version
Ubuntu clang version 14.0.0-1ubuntu1
Target: riscv64-unknown-linux-gnu
Thread model: posix
InstalledDir: /usr/bin

ii clang 1:14.0-55~exp2 riscv64
ii nezha-boot0 20220228+g0ad88bf-0ubuntu2~22.04.1

If this problem is considered fixed, I can add Clang into the same workload that GCC does on this board and stress-test Clang more.

[Heinrich Schuchardt (xypron)]

OpenSBI 1.1 is in Jammy since 2022-11-29. U-Boot for Jammy has been updated in 2022-12-06 using that OpenSBI. So the issue can be considered fixed since then.

[Denis Ovsienko (dovsienko)]

Thank you for confirming, I confirm in my test Clang can compile the full project again, so this bug looks resolved.