Re: [PATCH 6/7] mm: Pass pvec directly to find_get_entries

[Jan Kara]

On Tue 25-08-20 14:28:14, Matthew Wilcox wrote:
> On Tue, Aug 25, 2020 at 02:33:24PM +0200, Jan Kara wrote:
> > On Mon 24-08-20 18:36:39, Matthew Wilcox wrote:
> > > We already have functions in filemap which take a pagevec, eg
> > > page_cache_delete_batch() and delete_from_page_cache_batch().
> > 
> > Right but those are really pretty internal helper functions so I don't
> > think they form or strong precedence.
> 
> To be honest, I saw that as being the way forward for the page cache APIs.
> If we're going to use a batching mechanism, it should be pagevecs, and
> it should be built into the page cache interfaces rather than hanging
> out off on the side.
> 
> > > So if we're going to merge the two functions, it seems more natural to
> > > have it in filemap.c and called find_get_entries(), but I'm definitely
> > > open to persuasion on this!
> > 
> > I agree that having non-trivial xarray code in mm/swap.c isn't attractive
> > either. Dunno, I dislike the inconsistency between find_get_pages() and
> > find_get_entries() you create but they aren't completely consistent anyway
> > so I can live with that. Or we can just leave the pagevec_lookup_entries()
> > wrapper and the API will stay consistent...
> 
> I was thinking about this some more [1] [2].  I think we can get to the
> point where find_get_pages(), find_get_entries() and find_get_pages_tag()
> (and all their variants) end up taking a pagevec as their last argument.
> 
> Also, I was thinking that all these names are wrong.  Really, they're
> mapping_get_pages(), mapping_get_entries() and mapping_get_marked_pages().
> So maybe I should move in that direction.

Well, as I wrote to you in one of the replies. IMO pagevec unnecessarily
complicate matters and we should rather have for_each_mapping_page() and
for_each_mapping_entry() magic macros that hide pagevecs inside. Most of
users process returned pages/entries one by one so these macros would
simplify them. So it would seem better to me to go more into this direction
than to spread pagevecs...

> [1] https://lore.kernel.org/lkml/20200824214841.17132-1-willy@xxxxxxxxxxxxx/
> [2] https://lore.kernel.org/lkml/20200824183424.4222-1-willy@xxxxxxxxxxxxx/

								Honza
-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR