Alex Crichton (Jun 01 2023 at 22:47): Alex Crichton (Jun 01 2023 at 22:47):@Ulrich Weigand you might be interested in that the 1.70.0 release of Rust, which notably contains an LLVM 15->16 upgrade, is causing issues on CI for s390x. CI itself showed segfaults in QEMU but locally I can't even get that far as it's SIGILL-ing in rustc itself (somewhere buried in LLVM). My guess/hunch is that there maybe s390x-specific bugs in LLVM to cause rustc to SIGILL
Ulrich Weigand (Jun 02 2023 at 07:14):Hi @Alex Crichton , at this link I only see a riscv64 crash, not a s390x one - did you intend to post a different link? I've just run a build and test of current mainline using rust 1.70 natively on s390x, and everything looks fine to me.
Alex Crichton (Jun 02 2023 at 14:59):Ah sorry I think I got confused looking at logs, but looking through things again this is the crash I think which tipped me off about s390x. I can reproduce that locally and my previous characterization was mistaken, this is a SIGILL in rustc itself during cross-compilation.
My guess is that if you can't reproduce that locally it
Alex Crichton (Jun 02 2023 at 15:00):may be related to cross-compilation? An LLVM with asserts is probably the best way to debug further since it's probably something internal in LLVM failing to trigger the SIGILL
Alex Crichton (Jun 02 2023 at 18:39):Ok I can reproduce this locally with a cross-compile for rustc 1.70 but I can't reproduce with 1.71 beta. Between those rustc versions there were some LLVM changes, looks like minor bumps to LLVM (e.g. 16.0.0 to 16.0.4 or something like that). There is indeed a SystemZ-specific change there too so may be that we don't need to take action here
Last updated: Nov 22 2024 at 16:03 UTC