Re: [PATCH v2 0/8] Return head pages from find_*_entry

[Hugh Dickins]

On Thu, 10 Sep 2020, Matthew Wilcox (Oracle) wrote:

> This patch series started out as part of the THP patch set, but it has
> some nice effects along the way and it seems worth splitting it out and
> submitting separately.
> 
> Currently find_get_entry() and find_lock_entry() return the page
> corresponding to the requested index, but the first thing most callers do
> is find the head page, which we just threw away.  As part of auditing
> all the callers, I found some misuses of the APIs and some plain
> inefficiencies that I've fixed.
> 
> The diffstat is unflattering, but I added more kernel-doc and a new wrapper.
> 
> v2:
>  - Rework how shmem_getpage_gfp() handles getting a head page back from
>    find_lock_entry()
>  - Renamed find_get_swap_page() to find_get_incore_page()
>  - Make sure find_get_incore_page() doesn't return a head page
>  - Fix the missing include of linux/shmem_fs.h
>  - Move find_get_entry and find_lock_entry prototypes to mm/internal.h
>  - Rename thp_valid_index() to thp_contains()
>  - Fix thp_contains() for hugetlbfs and swapcache
>  - Add find_lock_head() wrapper around pagecache_get_page()
> 
> Matthew Wilcox (Oracle) (8):
>   mm: Factor find_get_incore_page out of mincore_page
>   mm: Use find_get_incore_page in memcontrol
>   mm: Optimise madvise WILLNEED
>   proc: Optimise smaps for shmem entries
>   i915: Use find_lock_page instead of find_lock_entry
>   mm: Convert find_get_entry to return the head page
>   mm/shmem: Return head page from find_lock_entry
>   mm: Add find_lock_head
> 
>  drivers/gpu/drm/i915/gem/i915_gem_shmem.c |  4 +--
>  fs/proc/task_mmu.c                        |  8 +----
>  include/linux/pagemap.h                   | 43 +++++++++++++++++-----
>  include/linux/swap.h                      |  7 ++++
>  mm/filemap.c                              | 44 +++++++++++------------
>  mm/internal.h                             |  3 ++
>  mm/madvise.c                              | 21 ++++++-----
>  mm/memcontrol.c                           | 24 ++-----------
>  mm/mincore.c                              | 28 ++-------------
>  mm/shmem.c                                | 20 +++++------
>  mm/swap_state.c                           | 32 +++++++++++++++++
>  11 files changed, 127 insertions(+), 107 deletions(-)
> 
> -- 
> 2.28.0

I was testing mmotm today (plus the shmem.c and swap_state.c
fixes that you posted, but I did not try the madvise.c one) -
my usual tmpfs swapping loads (plus hyperactive khugepaged to
maximize the THPs).  It behaved well, no problems found.

But I probably won't get to try your series of 12 for a few days.

Hugh