> > I actually see no reason for this patch, please see also my remarks > > /question on patch #4. > > You mean that we can't improve the fully-ordered version here? Well, I mean, this patch doesn't AFAICT. You've recently fixed these LR/SC mappings in 1d84afaf0252 ("riscv: Fix fully ordered LR/SC xchg[8|16]() implementations"), and the remaining look just fine to me. Andrea