https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=281749
--- Comment #54 from Mateusz Guzik <[email protected]> --- (In reply to Konstantin Belousov from comment #52) A release fence followed by an acquire fence does not prevent the CPU from reordering things so that the store happens *after*. As in: vp->v_seqc++; atomic_thread_fence_rel(); atomic_thread_fence_acq(); if (vp->v_usecount) ... can still load v_usecount prior to publishing v_seqc -- You are receiving this mail because: You are the assignee for the bug.
