On Mon, Apr 18, 2022 at 08:33:06AM +0000, Visa Hankala wrote: > I think the sanest solution is to add the release and acquire barriers > in refcnt_rele().
Getting memory barriers right is too complicated for developers doing MP stuff. The existing locking and refcount primitives have to implement that functionality. I am on visa@'s side and would prefer a memory barrier in refcount API instead of searching for races in MP code. Better waste some CPU cycles in some cases than having strange behavior due to missing barries in other cases. bluhm
