On Fri, Sep 6, 2024 at 4:05 PM Shakeel Butt <shakeel.butt@xxxxxxxxx> wrote:
>
> The tracing of invalidation and truncation operations on large files
> showed that xa_get_order() is among the top functions where kernel
> spends a lot of CPUs. xa_get_order() needs to traverse the tree to reach
> the right node for a given index and then extract the order of the
> entry. However it seems like at many places it is being called within an
> already happening tree traversal where there is no need to do another
> traversal. Just use xas_get_order() at those places.
>
> Signed-off-by: Shakeel Butt <shakeel.butt@xxxxxxxxx>
> ---
> mm/filemap.c | 6 +++---
> mm/shmem.c | 2 +-
> 2 files changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/mm/filemap.c b/mm/filemap.c
> index 070dee9791a9..7e3412941a8d 100644
> --- a/mm/filemap.c
> +++ b/mm/filemap.c
> @@ -2112,7 +2112,7 @@ unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
> VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
> folio);
> } else {
> - nr = 1 << xa_get_order(&mapping->i_pages, xas.xa_index);
> + nr = 1 << xas_get_order(&xas);
> base = xas.xa_index & ~(nr - 1);
> /* Omit order>0 value which begins before the start */
> if (base < *start)
> @@ -3001,7 +3001,7 @@ static inline loff_t folio_seek_hole_data(struct xa_state *xas,
> static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
> {
> if (xa_is_value(folio))
> - return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
> + return PAGE_SIZE << xas_get_order(xas);
> return folio_size(folio);
> }
>
> @@ -4297,7 +4297,7 @@ static void filemap_cachestat(struct address_space *mapping,
> if (xas_retry(&xas, folio))
> continue;
>
> - order = xa_get_order(xas.xa, xas.xa_index);
> + order = xas_get_order(&xas);
Yikesy that's my bad. This is late, but FWIW:
Reviewed-by: Nhat Pham <nphamcs@xxxxxxxxx>
