On Wed, Jan 2, 2019 at 2:43 PM Hugh Dickins <hughd@xxxxxxxxxx> wrote: > > Wrong. Without heavier locking that would add unwelcome overhead to > common paths, we shall "always" need the retry logic. It does not > come into play very often, but here are two examples of why it's > needed (if I thought longer, I might find more). And in practice, > yes, I sometimes saw 1 retry needed. > Understood. Sorry, I missed these corner cases. > I don't use frontswap myself, and haven't paid any attention to the > frontswap partial swapoff case (though notice now that shmem_unuse() > lacks the plumbing needed for it - that needs fixing); but doubt it > would be a good idea to refactor it out as a separate case. > I shall rework the shmem side to take care of the frontswap and retain the retry logic in a simplified manner. Thanks again for all the comments and insights.. ~Vineeth
