Miri detects a deadlock?

[saethlin]

I'm trying to do some hacking locally, and Miri is detecting a deadlock. So I'm putting up this PR just to see if this reproduces in your CI, in which case this is probably related to new Miri behavior.

Yup, deadlock:

  error: deadlock: the evaluated program deadlocked
     --> /home/runner/.rustup/toolchains/nightly-2022-06-10-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/sys/unix/futex.rs:69:21
      |
  69  |                     )
      |                     ^ the evaluated program deadlocked
      |
      = note: inside `std::sys::unix::futex::futex_wait` at /home/runner/.rustup/toolchains/nightly-2022-06-10-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/sys/unix/futex.rs:69:21
      = note: inside `std::sys_common::thread_parker::futex::Parker::park` at /home/runner/.rustup/toolchains/nightly-2022-06-10-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/sys_common/thread_parker/futex.rs:52:13
      = note: inside `std::thread::park` at /home/runner/.rustup/toolchains/nightly-2022-06-10-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/thread/mod.rs:929:9
      = note: inside `std::sync::mpsc::blocking::WaitToken::wait` at /home/runner/.rustup/toolchains/nightly-2022-06-10-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/sync/mpsc/blocking.rs:67:13
      = note: inside `std::sync::mpsc::oneshot::Packet::<()>::recv` at /home/runner/.rustup/toolchains/nightly-2022-06-10-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/sync/mpsc/oneshot.rs:140:21
      = note: inside `std::sync::mpsc::Receiver::<()>::recv` at /home/runner/.rustup/toolchains/nightly-2022-06-10-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/sync/mpsc/mod.rs:1161:49
  note: inside closure at src/imp_std.rs:357:17
     --> src/imp_std.rs:357:17
      |
  357 |                 rx2.recv().unwrap();
      |                 ^^^^^^^^^^
  note: inside closure at src/imp_std.rs:269:54
     --> src/imp_std.rs:269:54
      |
  269 |             let _ = self.initialize(|| Ok::<T, Void>(f()));
      |                                                      ^^^
  note: inside closure at src/imp_std.rs:85:23
     --> src/imp_std.rs:85:23
      |
  85  |                 match f() {
      |                       ^^^
      = note: inside `std::ops::function::impls::<impl std::ops::FnMut<()> for &mut dyn std::ops::FnMut() -> bool>::call_mut` at /home/runner/.rustup/toolchains/nightly-2022-06-10-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ops/function.rs:290:13
  note: inside `imp::initialize_or_wait` at src/imp_std.rs:213:20
     --> src/imp_std.rs:213:20
      |
  213 |                 if init() {
      |                    ^^^^^^
  note: inside `imp::OnceCell::<()>::initialize::<[closure@src/imp_std.rs:269:37: 269:58], imp::tests::<impl imp::OnceCell<T>>::init::Void>` at src/imp_std.rs:81:9
     --> src/imp_std.rs:81:9
      |
  81  | /         initialize_or_wait(
  82  | |             &self.queue,
  83  | |             Some(&mut || {
  84  | |                 let f = unsafe { take_unchecked(&mut f) };
  ...   |
  95  | |             }),
  96  | |         );
      | |_________^
  note: inside `imp::tests::<impl imp::OnceCell<()>>::init::<[closure@src/imp_std.rs:355:20: 358:14]>` at src/imp_std.rs:269:21
     --> src/imp_std.rs:269:21
      |
  269 |             let _ = self.initialize(|| Ok::<T, Void>(f()));
      |                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  note: inside closure at src/imp_std.rs:355:13
     --> src/imp_std.rs:355:13
      |
  355 | /             O.init(|| {
  356 | |                 tx1.send(()).unwrap();
  357 | |                 rx2.recv().unwrap();
  358 | |             });
      | |______________^

  note: some details are omitted, run with `MIRIFLAGS=-Zmiri-backtrace=full` for a verbose backtrace

  error: aborting due to previous error

  error: test failed, to rerun pass '--lib'
  TEST_MIRI: 58.98s
  error: command `cargo miri test --features unstable` failed, exit status: 1
  test imp::tests::wait_for_force_to_finish ... ::endgroup::

[cbeuw]

Some hacky investigation pointed out two outdated values were read in this test (I'll add proper logging to Miri weak memory at some point so people can debug with MIRI_LOG alone)

     = note: inside `std::sync::atomic::atomic_load::<u32>` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/core/src/sync/atomic.rs:2586:24
     = note: inside `std::sync::atomic::AtomicU32::load` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/core/src/sync/atomic.rs:1712:26
     = note: inside `std::sys::unix::futex::futex_wait` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/sys/unix/futex.rs:33:12
     = note: inside `std::sys_common::thread_parker::futex::Parker::park` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/sys_common/thread_parker/futex.rs:52:13
     = note: inside `std::thread::park` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/thread/mod.rs:929:9
     = note: inside `std::sync::mpsc::blocking::WaitToken::wait` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/sync/mpsc/blocking.rs:67:13
     = note: inside `std::sync::mpsc::oneshot::Packet::<()>::recv` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/sync/mpsc/oneshot.rs:140:21
     = note: inside `std::sync::mpsc::Receiver::<()>::recv` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/sync/mpsc/mod.rs:1161:49
note: inside `imp::tests::wait_for_force_to_finish` at src/imp_std.rs:361:9

  19:     0x55a515f9c64a - miri::shims::unix::linux::sync::futex::hda7315750bbb03b7
                               at /home/cbeuw/rust/miri/src/shims/unix/linux/sync.rs:141:29
...
    = note: inside `std::sys::unix::futex::futex_wait` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/sys/unix/futex.rs:61:21
    = note: inside `std::sys_common::thread_parker::futex::Parker::park` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/sys_common/thread_parker/futex.rs:52:13
    = note: inside `std::thread::park` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/thread/mod.rs:929:9
    = note: inside `std::sync::mpsc::blocking::WaitToken::wait` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/sync/mpsc/blocking.rs:67:13
    = note: inside `std::sync::mpsc::oneshot::Packet::<()>::recv` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/sync/mpsc/oneshot.rs:140:21
    = note: inside `std::sync::mpsc::Receiver::<()>::recv` at /home/cbeuw/.rustup/toolchains/miri/lib/rustlib/src/rust/library/std/src/sync/mpsc/mod.rs:1161:49
note: inside `imp::tests::wait_for_force_to_finish` at src/imp_std.rs:361:9

I'm not familiar with once_cell, don't know if these pointers are helpful to you? I will look into this further

[saethlin]

I don't know once_cell or atomics very well actually, I'm just here because I wanted to patch the int-to-ptr casts out of it then I ran the tests and they reported a deadlock.

In any case, what I'm curious about here is whether this is a bug in once_cell or Miri. If you're confident this looks like a once_cell issue, I'll report it.

I just remembered to check with ThreadSanitizer and it reports a data race here so I'm more confident in reporting this as a bug but it is odd that we get a deadlock, not a data race reported. Operator error, I forgot you need to pass -Zbuild-std or TSan misbehaves.

[Noratrieb]

Looking into the tests a bit closer, the deadlock appears to happen on the lines 357 and 361, though inserting a panic! before line 357 makes it deadlock still (without any spans though, which is fun), but commenting out the recv after the panic! stops it, so that's quite weird.

[cbeuw]

Can confirm this is not a once_cell issue. This panics deadlocks too.

use std::{sync::mpsc::channel, thread};

fn test() {
    let (tx1, rx1) = channel();
    let (tx2, rx2) = channel();
    let t1 = thread::spawn(move || {
        tx1.send(()).unwrap();
        rx2.recv().unwrap();
    });

    rx1.recv().unwrap();

    tx2.send(()).unwrap();

    assert!(t1.join().is_ok());
}

pub fn main() {
    for i in 0..1000 {
        println!("iter {i}");
        test();
    }
}

$ cargo miri --version
miri 0.1.0 (4d6eca1 2022-06-08)

[saethlin]

Closing this because it has served its purpose (see linked Miri issue above)