Re: [RFC PATCH 2/2] mm: add zero_user_folio_segments()

[Matthew Wilcox <willy@infradead.org>]

On Fri, Feb 17, 2023 at 12:05:28AM +0800, Yin Fengwei wrote:
> zero_user_folio_segments() has same function as zero_user_segments().
> but take folio as parameter. Update folio_zero_segments(),
> folio_zero_segment() and folio_zero_range() to use it.

If we were going to do something like this, then it should be
folio_zero_segments(), not zero_user_folio_segments().

But I don't understand the advantage to adding this right now.
We're adding a new function that does exactly the same thing
as zero_user_segments() for no real benefit that I can see.

My plan was always to eventually kill off zero_user_segment()
zero_user_segments() and zero_user(), and then move the
implementation of zero_user_segments() to be the implementation
of folio_zero_segments().  But we still have ~100 calls to
those three functions, so we can't do that yet.


If you're looking for something to do, how about batching calls to
page_remove_rmap() in try_to_unmap_one()?  That's a pretty hairy
function, but I think that you'll see similar gains to the ones
you saw with page_add_file_rmap() since it's manipulating the
same counters.