Detect mixed-size and mixed-atomicity non-synchronized accesses #3137
Conversation
…if both are reads
|
@bors r=oli-obk |
Detect mixed-size and mixed-atomicity non-synchronized accesses Fixes #2303
| @@ -267,7 +267,7 @@ fn concurrent_wait_wake() { | |||
| FUTEX.store(FREE, Ordering::Relaxed); | |||
| unsafe { | |||
| DATA = 1; | |||
| libc::syscall(libc::SYS_futex, &FUTEX as *const AtomicI32, libc::FUTEX_WAKE, 1); | |||
| libc::syscall(libc::SYS_futex, addr_of!(FUTEX), libc::FUTEX_WAKE, 1); | |||
There was a problem hiding this comment.
I'm confused about the changes to this file. Does this mean that creating a reference to an atomic type is considered a data race?
And if so, is Atomic*::as_ptr not safe as well?
(Or maybe the problem was only in the code creating references to non-atomic integers, and was done for code style consistency?)
There was a problem hiding this comment.
The diff you are commenting on was not necessary, interior mutability blocks the spurious reads.
But other parts of this test use static mut FUTEX: i32, and &FUTEX then induces a non-atomic read that was now causing problems. I figured I'd change everything to use addr_of! consistently while I am at it.
(Or maybe the problem was only in the code creating references to non-atomic integers, and was done for code style consistency?)
Yes, that.
|
💔 Test failed - checks-actions |
|
@bors r=oli-obk |
|
☀️ Test successful - checks-actions |
Fixes #2303