Add shims for `gettid`-esque functions #4397

tgross35 · 2025-06-13T00:49:40Z

Various platforms provide a function to return the current OS thread ID, but they all use a slightly different name. Add shims for these functions for Apple, FreeBSD, and Windows, with tests to account for a few more platforms that are not yet supported by Miri.

These should be useful in general but should also help support printing the OS thread ID in panic messages 1.

tgross35 · 2025-06-13T01:56:36Z

src/shims/env.rs

+    /// Get an "OS" thread ID for any thread.
+    fn get_tid(&self, thread: ThreadId) -> u32 {
+        let this = self.eval_context_ref();
+        let index = thread.to_u32();
+
+        // todo: is this also true on Windows?
+        // Compute a TID for this thread, ensuring that the main thread has PID == TID.
+        this.get_pid().strict_add(index)
+    }


@ChrisDenton is there any relationship between the PID and TIDs on Windows?

This (preexisting) logic works for Unix because the first TID for a process is the same as the PID, but if that logic isn't true on Windows then I'll need to update this. If it is true on Windows, then I think the CreateThread shim needs to get updated to return the same TID logic in lpThreadId (hence the failing test).

On Windows thread IDs and process IDs are unrelated.

But it doesn't hurt to use the Unix scheme on Windows as well I guess?

Great, that makes things easier. Thanks.

Oops, I replied before seeing your message Ralf. Apparently it seems like a Linux thing more specifically than a Unix thing, I can't find documentation for other platforms about a relationship between PID and TID.

I just updated to only offset by PID on Linux, but do you prefer I put it back on all platforms and update the ID saved by CreateThread instead?

Just checked a handful of different platforms on cfarm:

Linux

main tid 2212321 0x21c1e1 main pid 2212321 0x21c1e1

FreeBSD:

main tid 0 0 main pid 23750 0x5cc6

OpenBSD:

main tid 137368 0x21898 main pid 93069 0x16b8d

NetBSD:

main tid 5057 0x13c1 main pid 5057 0x13c1

NetBSD checking a spawned thread

in spawn: pid 1474 0x5c2 tid 13912 0x3658 in main: pid 1474 0x5c2 tid 1474 0x5c2

MacOS:

main tid 7763176 0x7674e8 main pid 51507 0xc933

Based on that nonscientific experiment I'll add NetBSD to the list of platforms that says tid is an offset from pid. For the others it's probably actually better that they don't line up, so users don't rely on that non-cross-platform behavior.

tgross35 · 2025-06-18T21:13:29Z

Updated to include the SYS_gettid syscall

src/shims/unix/linux_like/syscall.rs

tgross35 · 2025-06-24T08:17:29Z

Rebased because of the test failures but that didn't help; I don't know why this seems to have an effect on alloc-access-tracking.rs.

RalfJung · 2025-06-28T11:06:23Z

src/shims/extern_static.rs

@@ -66,7 +66,7 @@ impl<'tcx> MiriMachine<'tcx> {
                    ecx,
                    &["__cxa_thread_atexit_impl", "__clock_gettime64"],
                )?;
-                Self::weak_symbol_extern_statics(ecx, &["getrandom", "statx"])?;
+                Self::weak_symbol_extern_statics(ecx, &["getrandom", "gettid", "statx"])?;


You're adding a new global variable here, that's why alloc-access-tracking changes. You'll have to increment the ID there by 1.

It seems I needed to increment by 2, not sure if that is expected

RalfJung

Lots of little comments :)
@rustbot author

src/shims/unix/env.rs

src/shims/env.rs

src/shims/unix/env.rs

src/shims/unix/linux_like/syscall.rs

RalfJung · 2025-06-28T11:16:59Z

tests/pass-dep/shims/gettid.rs

+
+        // Spin until the main thread has a chance to read this thread's ID
+        while !CAN_JOIN.load(Ordering::Relaxed) {
+            thread::sleep(sleep_duration);


Just use yield_now here, so need to have arbitrary sleep durations.

RalfJung · 2025-06-28T11:17:28Z

tests/pass-dep/shims/gettid.rs

+/// Specific platforms can query the tid of arbitrary threads; test that here.
+#[cfg(any(target_vendor = "apple", windows))]
+mod queried {


Why does this use the low-level OS APIs for thread creation, instead of the Rust std APIs?

Windows and apple have a way to get the thread ID of threads other than the current one, which is being tested here. But to do this you need the pthread_t or the HANDLE, which as far as I know we can't get from std.

I'll comment this a bit better

tests/pass-dep/shims/gettid.rs

rustbot · 2025-06-28T11:19:44Z

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

Various platforms provide a function to return the current OS thread ID, but they all use a slightly different name. Add shims for these functions for Apple, FreeBSD, and Windows, with tests to account for those and a few more platforms that are not yet supported by Miri. The syscall and extern symbol is included as well on Linux. These should be useful in general but should also help support printing the OS thread ID in panic messages [1]. [1]: rust-lang/rust#115746

tgross35 · 2025-06-28T21:22:32Z

I believe I addressed all of the concerns. I also updated the Windows and Mac implementations (and tests) since they return gracefully rather than crashing with null or (on Windows) invalid pointers.

@rustbot ready

tgross35 force-pushed the gettid-shims branch 4 times, most recently from 4ca7e19 to 82b76b0 Compare June 13, 2025 01:42

tgross35 commented Jun 13, 2025

View reviewed changes

tgross35 changed the title ~~Add shims for gettid and similar functions~~ Add shims for gettid-esque functions Jun 13, 2025

tgross35 force-pushed the gettid-shims branch 5 times, most recently from 3add687 to eda5348 Compare June 13, 2025 20:08

tgross35 mentioned this pull request Jun 13, 2025

Print thread ID in panic message rust-lang/rust#115746

Open

tgross35 force-pushed the gettid-shims branch 2 times, most recently from 14a3826 to 72a99da Compare June 18, 2025 21:00

tgross35 force-pushed the gettid-shims branch 3 times, most recently from 7f29fbb to 2a46fbe Compare June 18, 2025 21:35

tgross35 commented Jun 18, 2025

View reviewed changes

src/shims/unix/linux_like/syscall.rs Outdated Show resolved Hide resolved

tgross35 force-pushed the gettid-shims branch 2 times, most recently from a683afe to 68601cd Compare June 24, 2025 07:51

RalfJung reviewed Jun 28, 2025

View reviewed changes

rustbot added the S-waiting-on-author Status: Waiting for the PR author to address review comments label Jun 28, 2025

tgross35 force-pushed the gettid-shims branch from 50fd6a5 to 8c05f42 Compare June 28, 2025 19:14

tgross35 added 3 commits June 28, 2025 14:15

fixup! 68601cd drop extra syscall arg checks

5f12c71

fixup! 68601cd remove an unneeded preexisting comment

1d57f0a

tgross35 added 5 commits June 28, 2025 14:15

fixup! 68601cd switch to yield_now

75a94a0

fixup! 68601cd comments and test updates

4a0a2f0

fixup! 68601cd Apple actually returns EINVAL on null pointer

c5770b9

fixup! 68601cd assert OS rather than vendor

c4fc629

fixup! 68601cd this binding, comment update

de427c1

tgross35 force-pushed the gettid-shims branch from 8c05f42 to de427c1 Compare June 28, 2025 19:15

tgross35 added 3 commits June 28, 2025 14:25

fixup! 68601cd drop the assertion table, and we don't need a Symbol

90695e4

fixup! 68601cd increment alloc id by two...?

6f960f3

fixup! 68601cd fix windows behavior

64f2447

rustbot added S-waiting-on-review Status: Waiting for a review to complete and removed S-waiting-on-author Status: Waiting for the PR author to address review comments labels Jun 28, 2025

Add shims for gettid-esque functions #4397

Are you sure you want to change the base?

Add shims for gettid-esque functions #4397

Conversation

tgross35 commented Jun 13, 2025

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 Jun 13, 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!

tgross35 Jun 13, 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!

tgross35 commented Jun 18, 2025

Uh oh!

Uh oh!

tgross35 commented Jun 24, 2025

Uh oh!

RalfJung Jun 28, 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!

RalfJung left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

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!

Uh oh!

rustbot commented Jun 28, 2025

Uh oh!

tgross35 commented Jun 28, 2025

Uh oh!

Uh oh!

Add shims for `gettid`-esque functions #4397

Add shims for `gettid`-esque functions #4397

tgross35 Jun 13, 2025 •

edited

Loading

tgross35 Jun 13, 2025 •

edited

Loading

RalfJung Jun 28, 2025 •

edited

Loading