compiler_fence documentation: emphasize synchronization, not reordering · patricklam/verify-rust-std@e00784f (original) (raw)
`@@ -3570,10 +3570,9 @@ unsafe fn atomic_umin<T: Copy>(dst: *mut T, val: T, order: Ordering) -> T {
`
3570
3570
``
3571
3571
`/// An atomic fence.
`
3572
3572
`///
`
3573
``
`-
/// Depending on the specified order, a fence prevents the compiler and CPU from
`
3574
``
`-
/// reordering certain types of memory operations around it.
`
3575
``
`-
/// That creates synchronizes-with relationships between it and atomic operations
`
3576
``
`-
/// or fences in other threads.
`
``
3573
`+
/// Fences create synchronization between themselves and atomic operations or fences in other
`
``
3574
`+
/// threads. To achieve this, a fence prevents the compiler and CPU from reordering certain types of
`
``
3575
`+
/// memory operations around it.
`
3577
3576
`///
`
3578
3577
`` /// A fence 'A' which has (at least) [Release
] ordering semantics, synchronizes
``
3579
3578
`` /// with a fence 'B' with (at least) [Acquire
] semantics, if and only if there
``
`@@ -3594,6 +3593,12 @@ unsafe fn atomic_umin<T: Copy>(dst: *mut T, val: T, order: Ordering) -> T {
`
3594
3593
`/// }
`
3595
3594
```` /// ```
`3596`
`3595`
`///
`
``
`3596`
`` +
/// Note that in the example above, it is crucial that the accesses to `x` are atomic. Fences cannot
``
``
`3597`
`+
/// be used to establish synchronization among non-atomic accesses in different threads. However,
`
``
`3598`
`+
/// thanks to the happens-before relationship between A and B, any non-atomic accesses that
`
``
`3599`
`+
/// happen-before A are now also properly synchronized with any non-atomic accesses that
`
``
`3600`
`+
/// happen-after B.
`
``
`3601`
`+
///
`
`3597`
`3602`
`` /// Atomic operations with [`Release`] or [`Acquire`] semantics can also synchronize
``
`3598`
`3603`
`/// with a fence.
`
`3599`
`3604`
`///
`
`@@ -3659,33 +3664,30 @@ pub fn fence(order: Ordering) {
`
`3659`
`3664`
`}
`
`3660`
`3665`
`}
`
`3661`
`3666`
``
`3662`
``
`-
/// A compiler memory fence.
`
``
`3667`
`+
/// A "compiler-only" atomic fence.
`
`3663`
`3668`
`///
`
`3664`
``
`` -
/// `compiler_fence` does not emit any machine code, but restricts the kinds
``
`3665`
``
`-
/// of memory re-ordering the compiler is allowed to do. Specifically, depending on
`
`3666`
``
`` -
/// the given [`Ordering`] semantics, the compiler may be disallowed from moving reads
``
`3667`
``
`-
/// or writes from before or after the call to the other side of the call to
`
`3668`
``
`` -
/// `compiler_fence`. Note that it does **not** prevent the *hardware*
``
`3669`
``
`-
/// from doing such re-ordering. This is not a problem in a single-threaded,
`
`3670`
``
`-
/// execution context, but when other threads may modify memory at the same
`
`3671`
``
`` -
/// time, stronger synchronization primitives such as [`fence`] are required.
``
``
`3669`
`` +
/// Like [`fence`], this function establishes synchronization with other atomic operations and
``
``
`3670`
`` +
/// fences. However, unlike [`fence`], `compiler_fence` only establishes synchronization with
``
``
`3671`
`+
/// operations *in the same thread*. This may at first sound rather useless, since code within a
`
``
`3672`
`+
/// thread is typically already totally ordered and does not need any further synchronization.
`
``
`3673`
`+
/// However, there are cases where code can run on the same thread without being ordered:
`
``
`3674`
`+
/// - The most common case is that of a *signal handler*: a signal handler runs in the same thread
`
``
`3675`
`` +
/// as the code it interrupted, but it is not ordered with respect to that code. `compiler_fence`
``
``
`3676`
`+
/// can be used to establish synchronization between a thread and its signal handler, the same way
`
``
`3677`
`` +
/// that `fence` can be used to establish synchronization across threads.
``
``
`3678`
`+
/// - Similar situations can arise in embedded programming with interrupt handlers, or in custom
`
``
`3679`
`` +
/// implementations of preemptive green threads. In general, `compiler_fence` can establish
``
``
`3680`
`+
/// synchronization with code that is guaranteed to run on the same hardware CPU.
`
`3672`
`3681`
`///
`
`3673`
``
`-
/// The re-ordering prevented by the different ordering semantics are:
`
``
`3682`
`` +
/// See [`fence`] for how a fence can be used to achieve synchronization. Note that just like
``
``
`3683`
`` +
/// [`fence`], synchronization still requires atomic operations to be used in both threads -- it is
``
``
`3684`
`+
/// not possible to perform synchronization entirely with fences and non-atomic operations.
`
`3674`
`3685`
`///
`
`3675`
``
`` -
/// - with [`SeqCst`], no re-ordering of reads and writes across this point is allowed.
``
`3676`
``
`` -
/// - with [`Release`], preceding reads and writes cannot be moved past subsequent writes.
``
`3677`
``
`` -
/// - with [`Acquire`], subsequent reads and writes cannot be moved ahead of preceding reads.
``
`3678`
``
`` -
/// - with [`AcqRel`], both of the above rules are enforced.
``
``
`3686`
`` +
/// `compiler_fence` does not emit any machine code, but restricts the kinds of memory re-ordering
``
``
`3687`
`` +
/// the compiler is allowed to do. `compiler_fence` corresponds to [`atomic_signal_fence`] in C and
``
``
`3688`
`+
/// C++.
`
`3679`
`3689`
`///
`
`3680`
``
`` -
/// `compiler_fence` is generally only useful for preventing a thread from
``
`3681`
``
`-
/// racing *with itself*. That is, if a given thread is executing one piece
`
`3682`
``
`-
/// of code, and is then interrupted, and starts executing code elsewhere
`
`3683`
``
`-
/// (while still in the same thread, and conceptually still on the same
`
`3684`
``
`-
/// core). In traditional programs, this can only occur when a signal
`
`3685`
``
`-
/// handler is registered. In more low-level code, such situations can also
`
`3686`
``
`-
/// arise when handling interrupts, when implementing green threads with
`
`3687`
``
`-
/// pre-emption, etc. Curious readers are encouraged to read the Linux kernel's
`
`3688`
``
`-
/// discussion of [memory barriers].
`
``
`3690`
`` +
/// [`atomic_signal_fence`]: https://en.cppreference.com/w/cpp/atomic/atomic_signal_fence
``
`3689`
`3691`
`///
`
`3690`
`3692`
`/// # Panics
`
`3691`
`3693`
`///
`
`@@ -3723,8 +3725,6 @@ pub fn fence(order: Ordering) {
`
`3723`
`3725`
`/// }
`
`3724`
`3726`
`/// }
`
`3725`
`3727`
```` /// ```
3726
``
`-
///
`
3727
``
`-
/// [memory barriers]: https://www.kernel.org/doc/Documentation/memory-barriers.txt
`
3728
3728
`#[inline]
`
3729
3729
`#[stable(feature = "compiler_fences", since = "1.21.0")]
`
3730
3730
`#[rustc_diagnostic_item = "compiler_fence"]
`