Data race when allocating a zero sized layout

[bluurryy]

I have noticed that miri reports a datarace when allocating a zero sized layout from multiple threads on a new Bump:

fn create_bump_and_alloc() {
    let bump = bumpalo::Bump::new();
    bump.alloc(());
}

pub fn main() {
    std::thread::spawn(create_bump_and_alloc);
    std::thread::spawn(create_bump_and_alloc);
}

reports:

error: Undefined Behavior: Data race detected between (1) non-atomic write on thread `unnamed-1` and (2) non-atomic read on thread `unnamed-2` at alloc3054+0x20
   --> .../nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/cell.rs:546:18
    |
546 |         unsafe { *self.value.get() }
    |                  ^^^^^^^^^^^^^^^^^ (2) just happened here
    |
help: and (1) occurred earlier here
   --> src/main.rs:3:5
    |
  3 |     bump.alloc(());
    |     ^^^^^^^^^^^^^^

This happens because the EMPTY_CHUNK footer is shared by all newly created Bumps. When doing a zero sized allocation the empty chunk has enough capacity for the allocation so its pointer gets bumped by 0 and then written back. The empty chunk is not actually modified, but its ptr field is written to with the same value it already has.

I don't know if this could cause problems in practice.

A way to fix this could be:

• check for layout.size() == 0 but this will add an additonal branch for slice allocations and allocations by a non statically known layout
• don't use EMPTY_CHUNK but always allocate a chunk on creation