unit test failures on s390x machines

[alexanderkjall]

When running the testsuite on an s390x machine I get failures like these:

111s thread 'tests::xeon_gold_6252::vendor_check' panicked at src/tests/xeon_gold_6252.rs:605:5:
111s assertion `left == right` failed
111s   left: "uneGIeniletn"
111s  right: "GenuineIntel"
111s stack backtrace:
111s    0: __rustc::rust_begin_unwind
111s              at /usr/src/rustc-1.90.0/library/std/src/panicking.rs:697:5
111s    1: core::panicking::panic_fmt
111s              at /usr/src/rustc-1.90.0/library/core/src/panicking.rs:75:14
111s    2: core::panicking::assert_failed_inner
111s              at /usr/src/rustc-1.90.0/library/core/src/panicking.rs:448:17
111s    3: core::panicking::assert_failed
111s              at /usr/src/rustc-1.90.0/library/core/src/panicking.rs:403:5
111s    4: raw_cpuid::tests::xeon_gold_6252::vendor_check
111s              at ./src/tests/xeon_gold_6252.rs:605:5
111s    5: raw_cpuid::tests::xeon_gold_6252::vendor_check::{{closure}}
111s              at ./src/tests/xeon_gold_6252.rs:602:18
111s    6: core::ops::function::FnOnce::call_once
111s              at /usr/src/rustc-1.90.0/library/core/src/ops/function.rs:253:5
111s    7: core::ops::function::FnOnce::call_once
111s              at /usr/src/rustc-1.90.0/library/core/src/ops/function.rs:253:5
111s note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
111s 
111s 
111s failures:
111s     tests::i7_12700k::brand_string
111s     tests::i7_12700k::vendor_check
111s     tests::ryzen_matisse::brand_string
111s     tests::ryzen_matisse::cache_info
111s     tests::ryzen_matisse::direct_cache_access
111s     tests::ryzen_matisse::extended_processor_and_feature_identifiers
111s     tests::ryzen_matisse::l1_tlb_cache
111s     tests::ryzen_matisse::perfmon_info
111s     tests::ryzen_matisse::processor_serial
111s     tests::ryzen_matisse::svm
111s     tests::ryzen_matisse::vendor_check
111s     tests::xeon_gold_6252::brand_string
111s     tests::xeon_gold_6252::vendor_check

Full test output here: https://ci.debian.net/data/autopkgtest/testing/s390x/r/rust-raw-cpuid/67009698/log.gz

It seems like the unsafe code in VendorInfo as_str doesn't work on BigEndian machines (understandable, as intel is small endian).

Would you like a pull request that gate's this functionality?

[gz]

hi @alexanderkjall thanks for the report,thats interesting :) yes we can add support for this. Is there maybe something in cpuid info that indicates the endianness (I am not aware of anything, aside for checking if the cpu is a s390x)

[alexanderkjall]

You can gate by big/little endian like this #[cfg(target_endian = "big")].

But I don't think that is the correct way to go about this, my guess (without having read all code here) is that the test suite doesn't execute the actual CPUID instruction (as that would fail on non-x86), and just mocks the output of the instruction call. That output is then mis-constructed into a string because there is a assumption about little endian, that fails on the s390x.

Just so that we are talking about the same thing, s390x is also called the IBM Z, it's one of these systems: https://www.ibm.com/products/z17 it's not an x86 cpu :)

I can send a pull request that adds a #[cfg(any(target_arch = "x86_64", target_arch = "x86"))] to those tests, if you agree that it's a good solution :)