Stream: git-wasmtime

Topic: wasmtime / issue #7445 mpk: some GitHub CI runners have M...

Wasmtime GitHub notifications bot (Nov 01 2023 at 16:45):

abrown opened issue #7445.

Wasmtime GitHub notifications bot (Nov 01 2023 at 16:50):

abrown edited issue #7445:

GitHub CI runners are showing some strange behavior: on certain runners (unknown which ones), the CPUID bits claim that MPK is supported, but running any MPK code (e.g., RDPKRU) causes a SIGILL crash. #7446 disables MPK until this is resolved.

Some instances of this failure:

• https://github.com/bytecodealliance/wasmtime/actions/runs/6721387888/job/18267026371#step:15:3283
• https://github.com/bytecodealliance/wasmtime/actions/runs/6714337681/job/18247418366?pr=7335#step:15:3280

Wasmtime GitHub notifications bot (Nov 01 2023 at 17:51):

abrown commented on issue #7445:

Finally got a reproduction that includes the lscpu output: https://github.com/bytecodealliance/wasmtime/actions/runs/6722788583/job/18272718138?pr=7448#step:4:20. Something must be different about how that AMD EPYC 7763 sets up its CPUID bits.

Wasmtime GitHub notifications bot (Nov 01 2023 at 17:53):

alexcrichton commented on issue #7445:

Nice! Seems like this should be easy enough to manage by updating the supported check to require Intel CPUs and filter out AMD ones then?

Wasmtime GitHub notifications bot (Nov 01 2023 at 18:03):

abrown commented on issue #7445:

That's an option; I was also thinking about looking up how the Linux kernel sets the pku flag. They probably have to deal with this exact issue there. There might even more special cases there to be aware of...

Wasmtime GitHub notifications bot (Nov 01 2023 at 18:10):

abrown commented on issue #7445:

Yeah, it appears that Intel and AMD differ on what is presented in that CPUID leaf:

• https://github.com/torvalds/linux/blob/8bc9e6515183935fa0cccaf67455c439afe4982b/arch/x86/include/asm/cpufeatures.h#L380-L383
• https://github.com/torvalds/linux/blob/8bc9e6515183935fa0cccaf67455c439afe4982b/tools/testing/selftests/mm/pkey-x86.h#L65C1-L66
• https://github.com/torvalds/linux/blob/8bc9e6515183935fa0cccaf67455c439afe4982b/tools/testing/selftests/kvm/include/x86_64/processor.h#L122-L155

Wasmtime GitHub notifications bot (Nov 07 2023 at 12:10):

nagisa commented on issue #7445:

I can reproduce this issue locally reliably on my 2nd generation EPYC and can test things out it helps anyway.

Wasmtime GitHub notifications bot (Nov 07 2023 at 12:14):

nagisa edited a comment on issue #7445:

I can reproduce this issue (the SIGILL) locally & reliably on my 2nd generation EPYC and can test things out it helps anyway.

Wasmtime GitHub notifications bot (Nov 09 2023 at 19:42):

abrown closed issue #7445:

GitHub CI runners are showing some strange behavior: on certain runners (unknown which ones), the CPUID bits claim that MPK is supported, but running any MPK code (e.g., RDPKRU) causes a SIGILL crash. #7446 disables MPK until this is resolved.

Some instances of this failure:

• https://github.com/bytecodealliance/wasmtime/actions/runs/6721387888/job/18267026371#step:15:3283
• https://github.com/bytecodealliance/wasmtime/actions/runs/6714337681/job/18247418366?pr=7335#step:15:3280

Last updated: Nov 22 2024 at 17:03 UTC