On Wed, Jun 30, 2021 at 10:32:02AM +0200, Michal Hocko wrote:

On Wed 30-06-21 05:00:29, Matthew Wilcox wrote:

quoted

These are the folio equivalents of lock_page_memcg() and
unlock_page_memcg().  Reimplement them as wrappers.

Is there any reason why you haven't followed the same approach as for
the previous patches. I mean callers can call page_folio and then
lock_page_memcg wrapper shouldn't be really needed.

At this point in the patch series there are ~20 places which call
lock_page_memcg().  I think it makes more sense to leave the wrapper
in place, and then we can remove the wrapper once all/most of these
places are converted to use folios.  There are another 5 conversions
already in the patch series, eg here:

https://git.infradead.org/users/willy/pagecache.git/commitdiff/a41c942c8e4b41df30be128ef6998ff1849fa36a

I do not really want to be annoying here but I have to say that I like
the conversion by previous patches much better than this wrapper
approach as mentioned during the previous review already. If you have
some reasons to stick with this approach for this particular case then
make it explicit in the changelog.

OK, I can point to the number of callers as a reason to keep the
wrappers in place.  I intended to just do the conversion here, but
seeing the number of callers made me reconsider.