On Thu, Apr 25, 2019 at 11:13:23AM +0000, David Laight wrote: > I'm not at all sure that 'compare and exchange' operations > are atomic on all cpus if the data is misaligned and crosses > a page boundary and either (or both) pages need faulting in > (or hit a TLB miss). Most sane architectures already trap if you attempt that.
