simplify epoll_event packed check

[rmehri01]

Description

• Simplify the check for whether we should use repr(packed) on epoll_event to be compliant with the below sources
  • It looks like musl, android, and uapi all only use packed on x86_64, but glibc only uses packed on x86 so we can do any(target_arch = "x86_64", all(target_arch = "x86", target_env = "gnu"))

Closes #4628

Sources

• musl: kraj/musl@08514d0
• android: ttps://cs.android.com/android/platform/superproject/main/+/main:bionic/libc/include/bits/epoll_event.h;l=52-54;drc=dd6763a6c677687ad06727af037ed9a7b018c773
• uapi: https://github.com/torvalds/linux/blob/8f5ae30d69d7543eee0d70083daf4de8fe15d585/include/uapi/linux/eventpoll.h#L71-L86
• glibc: https://github.com/bminor/glibc/blob/8543577b04ded6d979ffcc5a818930e4d74d0645/sysdeps/unix/sysv/linux/sys/epoll.h#L85-L89

Checklist

• Relevant tests in libc-test/semver have been updated
• No placeholder or unstable values like *LAST or *MAX are
  included (see #3131)
• Tested locally (cd libc-test && cargo test --target mytarget);
  especially relevant for platforms that may not be checked in CI

@rustbot label +stable-nominated

[rmehri01]

Ah this seems to fail on i686 which is 32-bit so maybe this check is already correct, though maybe making the check #[cfg_attr(target_arch = "x86_64", repr(packed))] instead would be more reflective of https://github.com/bminor/glibc/blob/8543577b04ded6d979ffcc5a818930e4d74d0645/sysdeps/unix/sysv/linux/sys/epoll.h#L85-L89

[tgross35]

Looking around:

• Musl packs only on x86-64 (your link)
• Android does the same ttps://cs.android.com/android/platform/superproject/main/+/main:bionic/libc/include/bits/epoll_event.h;l=52-54;drc=dd6763a6c677687ad06727af037ed9a7b018c773
• UAPI does the same https://github.com/torvalds/linux/blob/8f5ae30d69d7543eee0d70083daf4de8fe15d585/include/uapi/linux/eventpoll.h#L71-L86
• ...but glibc seems to pack it unconditionally https://github.com/bminor/glibc/blob/8543577b04ded6d979ffcc5a818930e4d74d0645/sysdeps/unix/sysv/linux/sys/epoll.h#L85-L89

So I think it would be most accurate to update this to cfg(any(target_arch = "x86_64", target_env = "gnu"))

[rmehri01]

• ...but glibc seems to pack it unconditionally https://github.com/bminor/glibc/blob/8543577b04ded6d979ffcc5a818930e4d74d0645/sysdeps/unix/sysv/linux/sys/epoll.h#L85-L89

Oh interesting. Maybe I have it the wrong way around but for glibc, I think that __EPOLL_PACKED is empty by default for any targets except x86 based on https://github.com/bminor/glibc/blob/8543577b04ded6d979ffcc5a818930e4d74d0645/sysdeps/unix/sysv/linux/sys/epoll.h#L31-L33 and https://github.com/bminor/glibc/blob/8543577b04ded6d979ffcc5a818930e4d74d0645/sysdeps/unix/sysv/linux/x86/bits/epoll.h#L29?

[tgross35]

Oh yup you're right, I should have checked the __EPOLL_PACKED def (I was just looking at that yesterday too). The change here should be any(target_arch = "x86_64", all(target_arch = "x86", target_env = "gnu")) then (not enabled on all gnu targets)

[tgross35]

Great, thank you!