Separate L4Re from Linux code, add aarch64 and enable tests #4479

farao · 2025-06-09T17:11:05Z

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

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 · 2025-06-09T17:11:12Z

Some changes occurred in the Android module

cc @maurer

farao · 2025-06-13T08:50:10Z

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.

tgross35

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.

tgross35 · 2025-06-26T02:39:52Z

ci/run-docker.sh

I assume this isn't directly related to l4re, mind splitting it to a separate commit? (same PR is fine of course)

tgross35 · 2025-06-26T02:42:08Z

libc-test/build.rs

@@ -3990,6 +4007,10 @@ fn test_linux(target: &str) {
    });

    cfg.skip_struct(move |ty| {
+        if ty.starts_with("l4_") {


Are these anonymous on l4re?

Actually turned out to be not necessary. I removed this and similar lines.

tgross35 · 2025-06-26T02:42:22Z

libc-test/build.rs

@@ -4169,6 +4190,12 @@ fn test_linux(target: &str) {
    });

    cfg.skip_const(move |name| {
+        // L4Re requires a min stack size of 64k; that isn't defined in uClibc, but
+        // somewhere in the core libraries. uClibc wants 16k, but that's not enough.
+        if name == "PTHREAD_STACK_MIN" {


This should be l4re && ... right?

tgross35 · 2025-06-26T02:44:02Z

libc-test/build.rs

@@ -3716,11 +3728,12 @@ fn test_linux(target: &str) {
               "libgen.h",
               "limits.h",
               "link.h",
-               "linux/sysctl.h",
+               [uclibc]: "linux/if_ether.h",
+               [linux]: "linux/sysctl.h",


I think it is probably safer to flip the logic here. That is, rather than [linux], make these [!l4re] so any other OS that happens to reuse some of this function gets the same defaults.

True, that's better! I changed it.

tgross35 · 2025-06-26T02:44:32Z

libc-test/build.rs

@@ -4920,7 +4949,9 @@ fn test_linux(target: &str) {

    cfg.generate(src_hotfix_dir().join("lib.rs"), "main.rs");

-    test_linux_like_apis(target);
+    if linux {


Similarly, if !l4re here

tgross35 · 2025-06-26T02:52:16Z

src/unix/linux_like/l4re/uclibc/mod.rs

+        (*(self as *const siginfo_t as *const siginfo_sigfault)).si_addr
+    }
+
+    pub unsafe fn si_value(&self) -> crate::sigval {
+        #[repr(C)]
+        struct siginfo_si_value {
+            _si_signo: c_int,
+            _si_errno: c_int,
+            _si_code: c_int,
+            _si_timerid: c_int,
+            _si_overrun: c_int,
+            si_value: crate::sigval,
+        }
+        (*(self as *const siginfo_t as *const siginfo_si_value)).si_value


I know this is preexisting, but mind updating to the slightly safer newer casting APIs? So ptr::from_ref(self).cast::<siginfo_sigfault>()

tgross35 · 2025-06-26T02:57:52Z

src/unix/linux_like/l4re/uclibc/x86_64/mod.rs

+    // FIXME(1.0): this is actually a union
+    pub struct sem_t {


ulibc was broken before this right? If so, might as well make it correct here.

Yes, that was broken before. I fixed it now.

tgross35 · 2025-06-26T03:03:24Z

src/unix/linux_like/shared.rs

@@ -0,0 +1,2127 @@
+//! Shared definitions between Linux and L4Re
+
+// TODO: check if there is more overlap with emscripten and android


FIXME(linux) if this comment is meant to be left in the codebase

I used the FIXME. There are actually a lot of cases where the same struct that is shared between l4re and linux is also shared with one of the other targets (android or emscripten) and one might think about adding all these cases to linux_like/mod.rs instead (with the exception of the differing target as #[cfg..].

tgross35 · 2025-07-07T17:20:28Z

@rustbot author, for the above review and a rebase

rustbot · 2025-07-07T17:20:33Z

Reminder, once the PR becomes ready for a review, use @rustbot ready.

farao · 2025-07-07T18:55:24Z

Thanks for the comments! Unfortunately, I'll only be able to get to them beginning of next week but I'll make it a priority then!

farao · 2025-07-14T15:08:01Z

@tgross35 Should be all fixed now. CI runs through (the failing freebsd checks are again not due to my change). Anything else?

farao · 2025-07-20T11:24:32Z

@tgross35 any chance that we can get this into the next release? :)

tgross35

A handful more requests here then I think this should be good to go.

If there is a bit more commit splitting that can be done here, that would be helpful for the sake of better history and slightly easier cherry picks. E.g. move emscripten+android items to linux_like in one, add the new l4re module in another, create shared in a third.

tgross35 · 2025-07-28T09:33:50Z

src/unix/linux_like/l4re/mod.rs

+#[allow(deprecated)]
+#[deprecated(since = "0.2.64", note = "Not stable across OS versions")]
+pub const RLIMIT_NLIMITS: crate::__rlimit_resource_t = RLIM_NLIMITS;


Since this is l4re-only, it can be removed

True. I removed the deprecation warning instead, since L4Re's uclibc configuration has the define and I did not find a deprecation warning there.

tgross35 · 2025-07-28T09:50:21Z

src/unix/linux_like/mod.rs

+    #[cfg_attr(
+        any(
+            target_env = "musl",
+            target_env = "ohos",
+            target_env = "uclibc",
+            target_pointer_width = "32"
+        ),
+        repr(align(4))
+    )]
+    #[cfg_attr(
+        all(
+            not(target_env = "musl"),
+            not(target_env = "ohos"),
+            not(target_env = "uclibc"),
+            target_pointer_width = "64"
+        ),
+        repr(align(8))
+    )]
+    #[cfg(not(target_os = "android"))]
+    pub struct pthread_rwlockattr_t {


emscripten also needs 4-alignment right? Same for the other structs in this hunk

Oh, I missed that, thanks. I added it.

tgross35 · 2025-07-28T09:56:14Z

src/unix/linux_like/mod.rs

@@ -349,7 +595,7 @@ cfg_if! {
                u64.hash(state);
            }
        }
-
+    } else if #[cfg(feature = "extra_traits")] {


This isn't equivalent right? These used to be emitted unconditionally, but now are only if the all(feature = "extra_traits", not(target_os = "l4re")) condition above did not match.

You're right. I split it up into two blocks.

tgross35 · 2025-07-28T09:57:57Z

src/unix/linux_like/mod.rs

+#[cfg(not(target_os = "android"))]
+pub const PTHREAD_MUTEX_INITIALIZER: pthread_mutex_t = pthread_mutex_t {
+    size: [0; crate::__SIZEOF_PTHREAD_MUTEX_T],
+};
+#[cfg(not(target_os = "android"))]
+pub const PTHREAD_COND_INITIALIZER: pthread_cond_t = pthread_cond_t {
+    size: [0; crate::__SIZEOF_PTHREAD_COND_T],
+};
+#[cfg(not(any(target_os = "android", target_env = "uclibc")))]
+pub const PTHREAD_RWLOCK_INITIALIZER: pthread_rwlock_t = pthread_rwlock_t {
+    size: [0; crate::__SIZEOF_PTHREAD_RWLOCK_T],
+};


Weren't these emitted on Android before? Also, looks like they moved up in the file (used to be around 2717)

They are present on android (in b32/mod.rs and b64/mod.rs) but implemented differently, therefore I left them out here. Before, they were in another file (emscripten/mod.rs and linux/mod.rs) afaics.

tgross35 · 2025-07-28T10:11:32Z

src/unix/linux_like/shared.rs

@@ -0,0 +1,2127 @@
+//! Shared definitions between Linux and L4Re


Maybe call this file linux_l4re_shared to match the description?

tgross35 · 2025-07-28T10:11:45Z

src/unix/linux_like/shared.rs

+
+// FIXME(linux): check if there is more overlap with emscripten and android
+
+use core::mem::size_of;


Should already come via the prelude

tgross35 · 2025-07-28T10:17:44Z

src/unix/linux_like/shared.rs

+    pub struct __c_anonymous_ifru_map {
+        pub mem_start: c_ulong,
+        pub mem_end: c_ulong,
+        pub base_addr: c_ushort,
+        pub irq: c_uchar,
+        pub dma: c_uchar,
+        pub port: c_uchar,
+    }


This was in s! rather than s_no_extra_traits before right?

True, fixed.

tgross35 · 2025-07-28T10:21:40Z

src/unix/linux_like/mod.rs

+        impl PartialEq for dirent {
+            fn eq(&self, other: &dirent) -> bool {
+                self.d_ino == other.d_ino
+                    && self.d_off == other.d_off
+                    && self.d_reclen == other.d_reclen
+                    && self.d_type == other.d_type
+                    && self
+                        .d_name
+                        .iter()
+                        .zip(other.d_name.iter())
+                        .all(|(a, b)| a == b)
+            }
+        }


Why is this trait impl (and the others below it) in a separate module from the structs they go with?

This is because the types of the fields of dirent are different by os (but still they can share the trait implementation). But maybe that's overshooting the mark for code deduplication? Let me know if I shall put them together with their struct implementations again.

tgross35 · 2025-07-28T10:24:24Z

src/unix/linux_like/shared.rs

+pub const ABDAY_1: crate::nl_item = base;
+pub const ABDAY_2: crate::nl_item = base + 0x1;
+pub const ABDAY_3: crate::nl_item = base + 0x2;
+pub const ABDAY_4: crate::nl_item = base + 0x3;
+pub const ABDAY_5: crate::nl_item = base + 0x4;
+pub const ABDAY_6: crate::nl_item = base + 0x5;
+pub const ABDAY_7: crate::nl_item = base + 0x6;


These were gated on gnu || musl || ohos before. Is that not needed anymore?

yes, that's not needed anymore. uclibc has them too, just starting at a different number. So I introduced the base variable which is still different between uclibc and not uclibc.

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).

rustbot added ctest Issues relating to the ctest crate O-android O-gnu O-linux O-linux-like O-musl O-unix O-x86 S-waiting-on-review labels Jun 9, 2025

farao force-pushed the revert branch 11 times, most recently from 92284e5 to 40a082a Compare June 12, 2025 16:33

farao mentioned this pull request Jun 13, 2025

Separate L4Re code, add aarch64 and enable tests #4383

Draft

3 tasks

farao force-pushed the revert branch 2 times, most recently from 26e9993 to 5b60e70 Compare June 23, 2025 09:15

tgross35 reviewed Jun 26, 2025

View reviewed changes

rustbot added S-waiting-on-author and removed S-waiting-on-review labels Jul 7, 2025

farao force-pushed the revert branch 5 times, most recently from c2490f3 to 804f6f9 Compare July 14, 2025 14:48

farao force-pushed the revert branch from 804f6f9 to bde9453 Compare July 25, 2025 13:56

tgross35 mentioned this pull request Jul 28, 2025

ci: fix docker mount volumes SE-Linux enabled systems #4587

Closed

tgross35 requested changes Jul 28, 2025

View reviewed changes

ci: fix docker mount volumes SE-Linux enabled systems

9da7d4d

farao force-pushed the revert branch 2 times, most recently from 80cafff to 67e08ed Compare July 28, 2025 16:21

farao force-pushed the revert branch from 67e08ed to 5715560 Compare July 28, 2025 16:32

farao requested a review from tgross35 July 28, 2025 17:03

		@@ -0,0 +1,2127 @@
		//! Shared definitions between Linux and L4Re

		// TODO: check if there is more overlap with emscripten and android


		// FIXME(linux): check if there is more overlap with emscripten and android

		use core::mem::size_of;

Separate L4Re from Linux code, add aarch64 and enable tests #4479

Are you sure you want to change the base?

Separate L4Re from Linux code, add aarch64 and enable tests #4479

Uh oh!

Conversation

farao commented Jun 9, 2025

Sources

Checklist

Uh oh!

rustbot commented Jun 9, 2025

Uh oh!

farao commented Jun 13, 2025

Uh oh!

tgross35 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

farao Jul 14, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tgross35 commented Jul 7, 2025

Uh oh!

rustbot commented Jul 7, 2025

Uh oh!

farao commented Jul 7, 2025

Uh oh!

farao commented Jul 14, 2025

Uh oh!

farao commented Jul 20, 2025

Uh oh!

tgross35 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

farao Jul 14, 2025 •

edited

Loading