On 2021-02-11 6:14 p.m., Helge Deller wrote: > Yes, happens with latest version too. > >> It looks as if we might not need the ptl lock in set_pte_at() and probably elsewhere. > Just removing the locks from set_pte_at() didn't solved it. > After removing the others too, it works. > Patch is attached below. > I added a memory-barrier to set_pte() too. You are ahead of me. Should the memory barrier be a smp_mb()? Dave -- John David Anglin dave.anglin@xxxxxxxx
