2022-04-29 15:29:40

by Hugh Dickins

[permalink] [raw]
Subject: Re: [PATCH AUTOSEL 13/14] mm/thp: ClearPageDoubleMap in first page_add_file_rmap()

On Thu, 28 Apr 2022, Greg Kroah-Hartman wrote:

> From: Hugh Dickins <[email protected]>
>
> commit bd55b0c2d64e84a75575f548a33a3dfecc135b65 upstream.
>
> PageDoubleMap is maintained differently for anon and for shmem+file: the
> shmem+file one was never cleared, because a safe place to do so could
> not be found; so it would blight future use of the cached hugepage until
> evicted.
>
> See https://lore.kernel.org/lkml/[email protected]/
>
> But page_add_file_rmap() does provide a safe place to do so (though later
> than one might wish): allowing testing to return to an initial state
> without a damaging drop_caches.
>
> Link: https://lkml.kernel.org/r/[email protected]
> Fixes: 9a73f61bdb8a ("thp, mlock: do not mlock PTE-mapped file huge pages")
> Signed-off-by: Hugh Dickins <[email protected]>
> Reviewed-by: Yang Shi <[email protected]>
> Cc: "Kirill A. Shutemov" <[email protected]>
> Signed-off-by: Andrew Morton <[email protected]>
> Signed-off-by: Linus Torvalds <[email protected]>
> Signed-off-by: Greg Kroah-Hartman <[email protected]>

NAK.

I thought we had a long-standing agreement that AUTOSEL does not try
to add patches from akpm's tree which had not been marked for stable.

(Whereas, if a developer asks for such a patch to be added to stable
later, and verifies the result, that's of course a different matter.)

I've chosen to answer to this patch of my 3 in your 14 AUTOSELs,
because this one is just an improvement, not at all a bugfix needed
for stable (maybe AUTOSEL noticed "racy" or "safely" in the comments,
and misunderstood). The "Fixes" was intended to help any humans who
wanted to backport into their trees.

I do recall that this 13/14, and 14/14, are mods to mm/rmap.c
which followed other (mm/munlock) mods to mm/rmap.c in 5.18-rc1,
which affected the out path of the function involved, and somehow
made 14/14 a little cleaner. I'm sorry, but I just don't rate it
worth my time at the moment, to verify whether 14/14 happens to
have ended up as a correct patch or not.

And nobody can verify them without these AUTOSELs saying to which
tree they are targeted - 5.17 I suppose.

Hugh

> ---
> mm/rmap.c | 11 +++++++++++
> 1 file changed, 11 insertions(+)
>
> diff --git a/mm/rmap.c b/mm/rmap.c
> index 9e27f9f038d3..444d0d958aff 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1252,6 +1252,17 @@ void page_add_file_rmap(struct page *page, bool compound)
> }
> if (!atomic_inc_and_test(compound_mapcount_ptr(page)))
> goto out;
> +
> + /*
> + * It is racy to ClearPageDoubleMap in page_remove_file_rmap();
> + * but page lock is held by all page_add_file_rmap() compound
> + * callers, and SetPageDoubleMap below warns if !PageLocked:
> + * so here is a place that DoubleMap can be safely cleared.
> + */
> + VM_WARN_ON_ONCE(!PageLocked(page));
> + if (nr == nr_pages && PageDoubleMap(page))
> + ClearPageDoubleMap(page);
> +
> if (PageSwapBacked(page))
> __mod_lruvec_page_state(page, NR_SHMEM_PMDMAPPED,
> nr_pages);
> --
> 2.36.0