On Mon, Oct 24, 2016 at 11:47 AM, Johannes Weiner <han...@cmpxchg.org> wrote:
>
> How about this instead: given that we already mark the shadow entries
> exceptional, and the exceptional bit is part of the radix tree API,
> can we just introduce a node->exceptional counter for those entries
> and have the radix tree code assist us with that instead? It adds the
> counting for non-shadow exceptional entries as well (shmem swap slots,
> and DAX non-page entries), unfortunately, but this is way cleaner. It
> also makes mapping->nrexceptional and node->exceptional consistent in
> DAX (Jan, could you please double check the accounting there?)
>
> What do you think? Lightly tested patch below.
This certainly looks way better to me. I didn't *test* it, but it
doesn't make me scratch my head the way your previous patch did.
Linus
