linux: complete `siginfo_t` definition by dybucc · Pull Request #5345 · rust-lang/libc

dybucc · 2026-07-31T15:31:04Z

Description

Completes the siginfo_t definition under Linux targets. Closes #716.

See the patch messages for details on each of the GNU targets, uClibc
targets and musl targets.

Checklist

Relevant tests in libc-test/semver have been updated
Commit messages permalink to headers for added or changed API
Placeholder or unstable values like *LAST or *MAX have the
standard doc comment
Tested locally (cargo test -p libc-test --target mytarget);
especially relevant for platforms that may not be checked in CI

@rustbot label +stable-nominated

tgross35 · 2026-08-04T03:06:36Z

Would you mind using this as an opportunity to move signifo_t to src/new? At least on glibc, that should make things a lot easier with less duplication. No need to be perfect with siginfo-arch.h since that's tricky to replicate in Rust, cfg is fine.

Also fyi, @xtqqczze found https://github.com/sailfishos-mirror/glibc as a mirror for glibc that's much easier to search and less flaky, that's fine to link (as is sourceware, of course).

tgross35 · 2026-08-04T04:13:39Z

There will be quite a bit less to review after that, so will hold off taking a closer look. Let me know if you need help figuring out the module structure.

@rustbot author

rustbot · 2026-08-04T04:13:43Z

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

dybucc · 2026-08-06T06:31:18Z

I'm not sure what, but I had something else to do in this patchset. Hopefully
the review will surface it.

@rustbot ready

tgross35

Sorry this has been on my todo list for a while, finally got through it

View changes since this review

tgross35 · 2026-08-31T07:06:28Z

+        si_utime: crate::clock_t,
+        si_stime: crate::clock_t,


On x32, this needs some kind of alignment adjustment https://github.com/sailfishos-mirror/glibc/blob/4a07bb292f921c10e71fbf48c4a7f44391feb06c/sysdeps/unix/sysv/linux/x86/bits/siginfo-arch.h#L12-L14, maybe a struct __SI_CLOCK_T(crate::clock_t) that gets an alignment attribute on specific platforms.

tgross35 · 2026-08-31T07:10:08Z

+};
+
+s_no_extra_traits! {
+    pub struct siginfo_t {


Similarly it needs an alignment attribute on x32 for __SI_ALIGNMENT. I guess these were preexisting but may as well be fixed here.

tgross35 · 2026-08-31T07:24:25Z

+
+    #[cfg(target_pointer_width = "64")]
+    #[inline]
+    pub unsafe fn si_band(&self) -> c_int {


This can be a type __SI_BAND_TYPE = ... to match the header

tgross35 · 2026-08-31T07:28:18Z

Could you merge the siginfo_arch files into siginfo_t? Since there are just a few changed fields we may as well reduce the duplication, and it's easier to see the differences (like the original source).

You could keep the files separate for __SI_BAND_TYPE and __SI_CLOCK_T but I don't think that's worth it for 1-2 cfg_ifs. Just make a note that this one file represents multiple headers.

tgross35 · 2026-08-31T07:59:29Z

+        #[cfg(not(any(target_arch = "mips", target_arch = "mips64")))]
+        pub si_errno: c_int,
+        pub si_code: c_int,
+        #[cfg(any(target_arch = "mips", target_arch = "mips64"))]


You'll also need mips32r6 and mips64r6 here, and everywhere mips is configured. Somewhat annoyingly #t-compiler > mipsr6 targets `target_arch` confusion.

tgross35 · 2026-08-31T08:11:53Z

 // FIXME(pthread): eventually all platforms should use this module
 #[cfg(target_os = "linux")]
-pub(crate) use sysdeps::nptl::*;
+#[allow(unused)]
+pub(crate) use sysdeps::nptl::{
+    bits as nptl_bits,
+    pthread,
+};
 #[cfg(target_os = "linux")]
-pub(crate) use sysdeps::unix::linux::*;
+pub(crate) use sysdeps::unix::linux::{
+    bits as linux_bits,
+    net,
+};


If bits isn't needed, can only pthread be reexported?

It might even be possible to just drop the pub(crate) from the bits modules, since they're not really needed at this level.

tgross35 · 2026-08-31T08:16:15Z

+        #[cfg(target_env = "uclibc")]
+        pub use sysdeps::linux::common::bits::siginfo::*;
+
        #[cfg(target_env = "gnu")]
-        pub use net::route::*;
-        #[cfg(target_env = "gnu")]
-        pub use signal::*;
+        pub use self::{
+            linux_bits::types::siginfo_t::*,
+            net::route::*,
+            signal::*,
+        };


The loose idea is that the modules reexported here should correspond to headers that are used. That is, #include <linux/tls.h> maps to pub use linux::tls::*;, and #include <signal.h> maps to pub use signal::*. So these probably need a mod signal somewhere that reexports from bits, since users don't import sysdeps/linux/common/bits/signinfo.h directly.

Replace references to `siginfo_t` in the regular repository layout with a crate-relative path that uses the reexport from the `new` module. This allows the next few patches to all work as individual revisions.

Replace alignment/padding fields with a one-to-one definition as that used upstream. These have been incorporated as definitions in the `new` module to avoid huge repetition across target triples without overrides. The glibc definitions are mostly the same across target architectures except for MIPS, SPARC and x86. x86 is confusing because there's only an override when running under x86_64 and compiling for x32. I'm not sure how Rust handles this, so the `#[path]`-redirected files for that architecture are the same as the generic ones. See [^1] for the generic file and the following list for the architecture-specific overrides. - `mips`: <https://github.com/sailfishos-mirror/glibc/blob/4a07bb292f921c10e71fbf48c4a7f44391feb06c/sysdeps/unix/sysv/linux/mips/bits/siginfo-arch.h> - `sparc`: <https://github.com/sailfishos-mirror/glibc/blob/4a07bb292f921c10e71fbf48c4a7f44391feb06c/sysdeps/unix/sysv/linux/sparc/bits/siginfo-arch.h> - `x86`: <https://github.com/sailfishos-mirror/glibc/blob/4a07bb292f921c10e71fbf48c4a7f44391feb06c/sysdeps/unix/sysv/linux/x86/bits/siginfo-arch.h> [^1]: <https://github.com/sailfishos-mirror/glibc/blob/4a07bb292f921c10e71fbf48c4a7f44391feb06c/sysdeps/unix/sysv/linux/bits/types/siginfo_t.h>

Remove older bindings to `siginfo_t` using dummy fields instead of the upstream `union`. This follows from the patch providing a full definition for glibc bindings. Add reexports for `siginfo_t` under the `new` module. This has required renaming the reexport for the `nptl` module as it also contains a `bits` module of the same name as the `bits::types::siginfo_t` module path under `sysdeps`.

Replace alignment/padding fields with a mirrored definition of the upstream `union`. See [^1] for details. [^1]: <https://github.com/kraj/musl/blob/a42e9dee266f398026a33d0793c66225c7997755/include/signal.h>

Remove older bindings to `siginfo_t` using dummy fields instead of the upstream `union`. This follows from the patch providing a full definition for musl bindings. Add modified bindings under `new`.

Replace alignment/padding dummy fields with a mirrored definition fitting that of upstream's `union`. uClibc supports targets for which we have no support in Rust. No architecture-specific definition has been provided for those. See [^1] for details on the generic definition and [^2] for details on the MIPS definition. [^1]: <https://github.com/wbx-github/uclibc-ng/blob/60d8e8c0cb9be8a241f6f2645daba260c8aec33c/libc/sysdeps/linux/common/bits/siginfo.h> [^2]: <https://github.com/wbx-github/uclibc-ng/blob/60d8e8c0cb9be8a241f6f2645daba260c8aec33c/libc/sysdeps/linux/mips/bits/siginfo.h>

Remove older bindings to `siginfo_t` using dummy fields instead of the upstream `union`. This follows from the patch providing a full definition for uClibc bindings. Add modified bindings to `new` module.

rustbot · 2026-08-31T15:32:55Z

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.

rustbot added O-arm O-gnu O-linux O-linux-like O-mips O-musl O-powerpc O-riscv O-sparc O-unix O-x86 S-waiting-on-review stable-nominated This PR should be considered for cherry-pick to libc's stable release branch labels Jul 31, 2026

dybucc changed the title ~~Linux: complete siginfo_t definition~~ linux: complete siginfo_t definition Jul 31, 2026

This comment was marked as outdated.

Sign in to view

rustbot added S-waiting-on-author and removed S-waiting-on-review labels Aug 4, 2026