Auto merge of rust-lang#2370 - RalfJung:weak-mem-test, r=RalfJung

add another weak mem consistency test

`@cbeuw` your code seems to do this correctly? :D

Author: bors

tests/pass/0weak_memory_consistency.rs

@@ -20,8 +20,8 @@
 // "Mathematizing C++ concurrency", ACM SIGPLAN Notices, vol. 46, no. 1, pp. 55-66, 2011.
 // Available: https://ss265.host.cs.st-andrews.ac.uk/papers/n3132.pdf.
 
-use std::sync::atomic::AtomicUsize;
 use std::sync::atomic::Ordering::*;
+use std::sync::atomic::{fence, AtomicBool, AtomicI32};
 use std::thread::spawn;
 
 #[derive(Copy, Clone)]
@@ -32,13 +32,19 @@ unsafe impl<T> Sync for EvilSend<T> {}
 
 // We can't create static items because we need to run each test
 // multiple times
-fn static_atomic(val: usize) -> &'static AtomicUsize {
-    let ret = Box::leak(Box::new(AtomicUsize::new(val)));
+fn static_atomic(val: i32) -> &'static AtomicI32 {
+    let ret = Box::leak(Box::new(AtomicI32::new(val)));
+    ret.store(val, Relaxed); // work around https://github.com/rust-lang/miri/issues/2164
+    ret
+}
+fn static_atomic_bool(val: bool) -> &'static AtomicBool {
+    let ret = Box::leak(Box::new(AtomicBool::new(val)));
+    ret.store(val, Relaxed); // work around https://github.com/rust-lang/miri/issues/2164
     ret
 }
 
 // Spins until it acquires a pre-determined value.
-fn acquires_value(loc: &AtomicUsize, val: usize) -> usize {
+fn acquires_value(loc: &AtomicI32, val: i32) -> i32 {
     while loc.load(Acquire) != val {
         std::hint::spin_loop();
     }
@@ -207,7 +213,7 @@ fn test_sc_store_buffering() {
 }
 
 fn test_single_thread() {
-    let x = AtomicUsize::new(42);
+    let x = AtomicI32::new(42);
 
     assert_eq!(x.load(Relaxed), 42);
 
@@ -216,6 +222,42 @@ fn test_single_thread() {
     assert_eq!(x.load(Relaxed), 43);
 }
 
+fn test_sync_through_rmw_and_fences() {
+    // Example from https://github.com/llvm/llvm-project/issues/56450#issuecomment-1183695905
+    #[no_mangle]
+    pub fn rdmw(storing: &AtomicI32, sync: &AtomicI32, loading: &AtomicI32) -> i32 {
+        storing.store(1, Relaxed);
+        fence(Release);
+        sync.fetch_add(0, Relaxed);
+        fence(Acquire);
+        loading.load(Relaxed)
+    }
+
+    let x = static_atomic(0);
+    let y = static_atomic(0);
+    let z = static_atomic(0);
+
+    // Since each thread is so short, we need to make sure that they truely run at the same time
+    // Otherwise t1 will finish before t2 even starts
+    let go = static_atomic_bool(false);
+
+    let t1 = spawn(move || {
+        while !go.load(Relaxed) {}
+        rdmw(y, x, z)
+    });
+
+    let t2 = spawn(move || {
+        while !go.load(Relaxed) {}
+        rdmw(z, x, y)
+    });
+
+    go.store(true, Relaxed);
+
+    let a = t1.join().unwrap();
+    let b = t2.join().unwrap();
+    assert_ne!((a, b), (0, 0));
+}
+
 pub fn main() {
     for _ in 0..50 {
         test_single_thread();
@@ -225,5 +267,6 @@ pub fn main() {
         test_wrc();
         test_corr();
         test_sc_store_buffering();
+        test_sync_through_rmw_and_fences();
     }
 }

tests/pass/weak_memory/weak.rs

@@ -8,8 +8,8 @@
 // Spurious failure is possible, if you are really unlucky with
 // the RNG and always read the latest value from the store buffer.
 
-use std::sync::atomic::AtomicUsize;
 use std::sync::atomic::Ordering::*;
+use std::sync::atomic::{fence, AtomicUsize};
 use std::thread::spawn;
 
 #[derive(Copy, Clone)]
@@ -70,9 +70,9 @@ fn seq_cst() -> bool {
     r3 == 1
 }
 
-fn initialization_write() -> bool {
+fn initialization_write(add_fence: bool) -> bool {
     let x = static_atomic(11);
-    assert_eq!(x.load(Relaxed), 11);
+    assert_eq!(x.load(Relaxed), 11); // work around https://github.com/rust-lang/miri/issues/2164
 
     let wait = static_atomic(0);
 
@@ -85,6 +85,9 @@ fn initialization_write() -> bool {
 
     let j2 = spawn(move || {
         reads_value(wait, 1);
+        if add_fence {
+            fence(AcqRel);
+        }
         x.load(Relaxed)
     });
 
@@ -94,15 +97,55 @@ fn initialization_write() -> bool {
     r2 == 11
 }
 
-// Asserts that the function returns true at least once in 100 runs
-macro_rules! assert_once {
-    ($f:ident) => {
-        assert!(std::iter::repeat_with(|| $f()).take(100).any(|x| x));
-    };
+fn faa_replaced_by_load() -> bool {
+    // Example from https://github.com/llvm/llvm-project/issues/56450#issuecomment-1183695905
+    #[no_mangle]
+    pub fn rdmw(storing: &AtomicUsize, sync: &AtomicUsize, loading: &AtomicUsize) -> usize {
+        storing.store(1, Relaxed);
+        fence(Release);
+        // sync.fetch_add(0, Relaxed);
+        sync.load(Relaxed);
+        fence(Acquire);
+        loading.load(Relaxed)
+    }
+
+    let x = static_atomic(0);
+    assert_eq!(x.load(Relaxed), 0); // work around https://github.com/rust-lang/miri/issues/2164
+    let y = static_atomic(0);
+    assert_eq!(y.load(Relaxed), 0); // work around https://github.com/rust-lang/miri/issues/2164
+    let z = static_atomic(0);
+    assert_eq!(z.load(Relaxed), 0); // work around https://github.com/rust-lang/miri/issues/2164
+
+    // Since each thread is so short, we need to make sure that they truely run at the same time
+    // Otherwise t1 will finish before t2 even starts
+    let go = static_atomic(0);
+
+    let t1 = spawn(move || {
+        while go.load(Relaxed) == 0 {}
+        rdmw(y, x, z)
+    });
+
+    let t2 = spawn(move || {
+        while go.load(Relaxed) == 0 {}
+        rdmw(z, x, y)
+    });
+
+    go.store(1, Relaxed);
+
+    let a = t1.join().unwrap();
+    let b = t2.join().unwrap();
+    (a, b) == (0, 0)
+}
+
+/// Asserts that the function returns true at least once in 100 runs
+fn assert_once(f: fn() -> bool) {
+    assert!(std::iter::repeat_with(|| f()).take(100).any(|x| x));
 }
 
 pub fn main() {
-    assert_once!(relaxed);
-    assert_once!(seq_cst);
-    assert_once!(initialization_write);
+    assert_once(relaxed);
+    assert_once(seq_cst);
+    assert_once(|| initialization_write(false));
+    assert_once(|| initialization_write(true));
+    assert_once(faa_replaced_by_load);
 }