No need to atomically set VO bit in post_alloc?

[qinsoon]

For allocators that have its allocation buffer, it seems unnecesary to set VO bit atomically.

mmtk-core/src/policy/immix/immixspace.rs

Lines 139 to 142 in 3be73b8

	fn initialize_object_metadata(&self, _object: ObjectReference, _alloc: bool) {
	#[cfg(feature = "vo_bit")]
	crate::util::metadata::vo_bit::set_vo_bit::<VM>(_object);
	}

mmtk-core/src/util/metadata/vo_bit/mod.rs

Lines 64 to 71 in 3be73b8

	pub fn set_vo_bit<VM: VMBinding>(object: ObjectReference) {
	debug_assert!(
	!is_vo_bit_set::<VM>(object),
	"{:x}: VO bit already set",
	object
	);
	VO_BIT_SIDE_METADATA_SPEC.store_atomic::<u8>(object.to_address::<VM>(), 1, Ordering::SeqCst);
	}

[wks]

That's right. It should include all bump-pointer allocators and the native MarkSweep, too (because each thread allocates in its own block).