Separate L4Re from Linux code, add aarch64 and enable tests #4479
Conversation
rustbot
added
ctest
|
Some changes occurred in the Android module cc @maurer |
farao
force-pushed
the
revert
branch
11 times, most recently
from
92284e5 to
40a082a
Compare
|
Hi @tgross35, this is the refactoring of the recent L4Re PR: #4383 (which is kept around just in case). I think the failure of the freebsd nightly checks is not my fault. The rest succeeds now. I did most of what you requested in your comment on the old PR, I just kept the linux/mod.rs file around since there is a massive part of code that is linux-only and is not supported by L4Re and it did not seem to make sense to put it all in shared.rs. Also, I put some more code that's shared from the sub modules (emscripten, android, linux, l4re) up in linux_like/mod.rs. In theory, I think there's potential to do that with more code, that's just the one that I would have put in shared.rs but realized that it could go into linux_like/mod.rs instead. |
farao
force-pushed
the
revert
branch
2 times, most recently
from
26e9993 to
5b60e70
Compare
tgross35
left a comment
tgross35
left a comment
There was a problem hiding this comment.
Sorry this has taken a while to get to, but thank you for all the changes here! The shape of this one looks much better. I have a handful of small comments but will need to take a deeper look at the big refactor portions again.
|
@rustbot author, for the above review and a rebase |
|
Reminder, once the PR becomes ready for a review, use |
tgross35
left a comment
tgross35
left a comment
There was a problem hiding this comment.
Still some leftovers in the pthread definitions, and I would like the uclibc pthread defs to match source rather than using the array. But after this, LGTM!
| #[cfg_attr(any(target_pointer_width = "32",), repr(align(4)))] | ||
| #[cfg_attr(not(any(target_pointer_width = "32",)), repr(align(8)))] | ||
| pub struct pthread_mutexattr_t { | ||
| #[doc(hidden)] | ||
| size: [u8; crate::__SIZEOF_PTHREAD_MUTEXATTR_T], | ||
| } | ||
|
|
||
| #[cfg_attr(any(target_pointer_width = "32"), repr(align(4)))] | ||
| #[cfg_attr(all(target_pointer_width = "64"), repr(align(8)))] | ||
| pub struct pthread_rwlockattr_t { | ||
| #[doc(hidden)] | ||
| size: [u8; crate::__SIZEOF_PTHREAD_RWLOCKATTR_T], | ||
| } | ||
|
|
||
| #[repr(align(4))] | ||
| pub struct pthread_condattr_t { | ||
| #[doc(hidden)] | ||
| size: [u8; crate::__SIZEOF_PTHREAD_CONDATTR_T], | ||
| } | ||
|
|
||
| #[repr(align(4))] | ||
| pub struct pthread_barrierattr_t { | ||
| #[doc(hidden)] | ||
| size: [u8; crate::__SIZEOF_PTHREAD_BARRIERATTR_T], | ||
| } |
There was a problem hiding this comment.
Oh no it wasn't a deviant decision, that's actually how they're defined in the linux libs https://github.com/bminor/glibc/blob/f9e61cd446d45016e20b6fe85ab87364ebdbec1b/sysdeps/nptl/bits/pthreadtypes.h#L67-L72. I would indeed prefer to match the struct definitions (with all fields private). It's much easier to make sure the fields line up than to add manually compute the size/align, and I don't know that the sizes are likely to always stay the same across platforms anyway.
src/unix/linux_like/linux/mod.rs
Outdated
| size: [u8; crate::__SIZEOF_PTHREAD_COND_T], | ||
| pub size: [u8; crate::__SIZEOF_PTHREAD_COND_T], |
| #[cfg_attr( | ||
| all( | ||
| any(target_env = "musl", target_env = "ohos"), | ||
| any(target_env = "musl", target_env = "ohos", target_os = "emscripten"), | ||
| target_pointer_width = "32" | ||
| ), | ||
| repr(align(4)) | ||
| )] | ||
| #[cfg_attr( | ||
| all( | ||
| any(target_env = "musl", target_env = "ohos"), | ||
| any(target_env = "musl", target_env = "ohos", target_os = "emscripten"), | ||
| target_pointer_width = "64" | ||
| ), | ||
| repr(align(8)) | ||
| )] | ||
| #[cfg_attr( | ||
| all( | ||
| not(any(target_env = "musl", target_env = "ohos")), | ||
| not(any(target_env = "musl", target_env = "ohos", target_os = "emscripten")), | ||
| target_arch = "x86" | ||
| ), | ||
| repr(align(4)) | ||
| )] | ||
| #[cfg_attr( | ||
| all( | ||
| not(any(target_env = "musl", target_env = "ohos")), | ||
| not(any(target_env = "musl", target_env = "ohos", target_os = "emscripten")), | ||
| not(target_arch = "x86") | ||
| ), | ||
| repr(align(8)) | ||
| )] | ||
| pub struct pthread_cond_t { |
There was a problem hiding this comment.
Could you please make sure to self-review these pthread diffs? Again this has some changes for emscripten that aren't relevant.
| target_arch = "powerpc", | ||
| target_arch = "sparc", | ||
| target_arch = "x86_64", | ||
| target_arch = "x86" | ||
| target_arch = "x86", | ||
| target_os = "emscripten", | ||
| ) | ||
| ), | ||
| repr(align(4)) |
| target_arch = "powerpc", | ||
| target_arch = "sparc", | ||
| target_arch = "x86_64", | ||
| target_arch = "x86" | ||
| target_arch = "x86", | ||
| target_os = "emscripten", | ||
| )) | ||
| ), | ||
| repr(align(8)) | ||
| )] | ||
| pub struct pthread_mutex_t { | ||
| #[doc(hidden)] | ||
| size: [u8; crate::__SIZEOF_PTHREAD_MUTEX_T], | ||
| pub size: [c_char; crate::__SIZEOF_PTHREAD_MUTEX_T], | ||
| } | ||
|
|
||
| #[cfg_attr( | ||
| all( | ||
| target_pointer_width = "32", | ||
| any( | ||
| target_os = "emscripten", | ||
| target_arch = "mips", | ||
| target_arch = "mips32r6", | ||
| target_arch = "arm", |
| any( | ||
| target_pointer_width = "64", | ||
| not(any( | ||
| target_os = "emscripten", | ||
| target_arch = "mips", | ||
| target_arch = "mips32r6", | ||
| target_arch = "arm", |
| repr(align(8)) | ||
| )] | ||
| pub struct pthread_rwlock_t { | ||
| size: [u8; crate::__SIZEOF_PTHREAD_RWLOCK_T], | ||
| pub size: [u8; crate::__SIZEOF_PTHREAD_RWLOCK_T], | ||
| } | ||
|
|
||
| #[cfg_attr( |
This is imperfect but should get things closer to what is available on l4re's uclibc fork. Based on work by Marius Melzer [rust-lang#4479]. [rust-lang#4479]: rust-lang#4479 (backport <rust-lang#4836>) (cherry picked from commit 34d8ce7)
The L4Re code was previously attached to the Linux code which was not correct in many ways. This commit separates the L4Re code and enables the libc-tests and includes the fixes for the failing tests.
|
This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed. Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers. |
|
I replaced the pthread_* definitions with the uclibc (linuxthreads) ones in use by l4re. I removed the remaining emscripten lines after the last refactoring, sorry for missing those. @rustbot ready |
| pub struct _pthread_fastlock { | ||
| pub __status: c_long, | ||
| pub __spinlock: c_int, | ||
| } | ||
|
|
||
| pub struct pthread_cond_t { | ||
| pub __c_lock: _pthread_fastlock, | ||
| pub __c_waiting: _pthread_descr, | ||
| pub __padding: [u8; PTHREAD_COND_PADDING_SIZE], | ||
| pub __align: __pthread_cond_align_t, | ||
| } | ||
|
|
||
| pub struct pthread_condattr_t { | ||
| pub __dummy: c_int, | ||
| } | ||
|
|
||
| pub struct pthread_mutex_t { | ||
| pub __m_reserved: c_int, | ||
| pub __m_count: c_int, | ||
| pub __m_owner: _pthread_descr, | ||
| pub __m_kind: c_int, | ||
| pub __m_lock: _pthread_fastlock, | ||
| } | ||
|
|
||
| pub struct pthread_mutexattr_t { | ||
| pub __mutexkind: c_int, | ||
| } | ||
|
|
||
| pub struct pthread_rwlock_t { | ||
| pub __rw_lock: _pthread_fastlock, | ||
| pub __rw_readers: c_int, | ||
| pub __rw_writer: _pthread_descr, | ||
| pub __rw_read_waiting: _pthread_descr, | ||
| pub __rw_write_waiting: _pthread_descr, | ||
| pub __rw_kind: c_int, | ||
| pub __rw_pshared: c_int, | ||
| } | ||
|
|
||
| pub struct pthread_rwlockattr_t { | ||
| pub __lockkind: c_int, | ||
| pub __pshared: c_int, | ||
| } | ||
|
|
||
| pub struct pthread_barrier_t { | ||
| pub __ba_lock: _pthread_fastlock, | ||
| pub __ba_required: c_int, | ||
| pub __ba_present: c_int, | ||
| pub __ba_waiting: _pthread_descr, | ||
| } | ||
|
|
||
| pub struct pthread_barrierattr_t { | ||
| pub __pshared: c_int, | ||
| } |
There was a problem hiding this comment.
The types are intended to be opaque, so please make the fields private. (that's generally the rule for __-prefixed types/fields anyway).
Last thing, after that I'm happy to merge!
tgross35
left a comment
tgross35
left a comment
There was a problem hiding this comment.
You know what, please just send a followup making those fields private. Let's get this in.
I'm sorry this needed such a huge diff and a difficult process, but thank you for the persistence here! I'll be doing a new release in the next couple of days and will include this.
tgross35
added
the
stable-nominated
github-merge-queue
bot
removed this pull request from the merge queue due to failed status checks
|
Nice, thanks a lot for baring with me for this large change(s)! |
4 participants
The L4Re code was previously attached to the Linux code which was not correct in many ways. This commit separates the L4Re code and enables the libc-tests and includes the fixes for the failing tests. Aarch64 is added as a second supported architecture (more to come).
Sources
L4Re-adapted version of uclibc: https://github.com/kernkonzept/l4re-core/tree/master/uclibc/lib.
Checklist
libc-test/semverhave been updated*LASTor*MAXareincluded (see #3131)
cd libc-test && cargo test --target mytarget);especially relevant for platforms that may not be checked in CI