rustc SIGSEGV on RISC-V when compiling serde and other crates (stack overflow on 4 GB stack)

[planetryan]

I tried compiling this Rust project on native riscv hardware. The project uses deps like serde, arrayvec, and others that involve macros and traits.

I expected to see this happen:
build using cargo build --release

Instead this happened:

rustc crashes with a segmentation fault (SIGSEGV) inside libc.so.6. This occurs on both stable and nightly Rust, even after setting high stack limits (RUST_MIN_STACK=4294967296, i.e. 4 GB) and with 8 GiB of swap active.

rustc --version --verbose:

rayan@k1:~$ rustc --version --verbose
rustc 1.88.0 (6b00bc3 2025-06-23)
binary: rustc
commit-hash: 6b00bc3
commit-date: 2025-06-23
host: riscv64gc-unknown-linux-gnu
release: 1.88.0
LLVM version: 20.1.5

Repository:
https://github.com/planetryan/qoa

To reproduce, clone the repository and run on riscv cpu

export RUST_MIN_STACK=4294967296

ulimit -s unlimited

cargo build --release

Even compiling deps like serde crashs the compiler with a SIGSEGV on native RISC-V hardware.

[workingjubilee]

https://github.com/planetryan/qoa/blob/723822226a8ce1fea2226fa37f9cb6b1b06f686f/Cargo.toml#L55C1-L55C23

@planetryan This uses LTO.

Does this crash still occur if you remove that?

[planetryan]

I removed LTO crate, it compiled with no issue 👍

[saethlin]

Why was this closed? Does the same problem exist on x86_64?

[planetryan]

Why was this closed? Does the same problem exist on x86_64?

To my knowledge, only on riscv, and the issue seems to have been fixed

[saethlin]

LTO is a resource hog, but it is not supposed to use more than 4 GB of stack. So I'm not sure what you mean by "fixed".

[workingjubilee]

agreed, usually it's heap-heavy IME

[nikic]

How did you conclude that this is a stack overflow? You said that there is a segmentation fault, but did not provide a stack trace.