mm: free idle swap cache page after COW
authorHuang Ying <ying.huang@intel.com>
Tue, 29 Jun 2021 02:37:12 +0000 (19:37 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 29 Jun 2021 17:53:49 +0000 (10:53 -0700)
With commit 09854ba94c6a ("mm: do_wp_page() simplification"), after COW,
the idle swap cache page (neither the page nor the corresponding swap
entry is mapped by any process) will be left in the LRU list, even if it's
in the active list or the head of the inactive list.  So, the page
reclaimer may take quite some overhead to reclaim these actually unused
pages.

To help the page reclaiming, in this patch, after COW, the idle swap cache
page will be tried to be freed.  To avoid to introduce much overhead to
the hot COW code path,

a) there's almost zero overhead for non-swap case via checking
   PageSwapCache() firstly.

b) the page lock is acquired via trylock only.

To test the patch, we used pmbench memory accessing benchmark with
working-set larger than available memory on a 2-socket Intel server with a
NVMe SSD as swap device.  Test results shows that the pmbench score
increases up to 23.8% with the decreased size of swap cache and swapin
throughput.

Link: https://lkml.kernel.org/r/20210601053143.1380078-1-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Suggested-by: Johannes Weiner <hannes@cmpxchg.org> [use free_swap_cache()]
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Tim Chen <tim.c.chen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/swap.h
mm/memory.c
mm/swap_state.c

index 46d51d058d05e1a64e8e35005d558f61c4312b96..49b1dd2c100b6594340613ace045771dfacacea3 100644 (file)
@@ -446,6 +446,7 @@ extern void __delete_from_swap_cache(struct page *page,
 extern void delete_from_swap_cache(struct page *);
 extern void clear_shadow_from_swap_cache(int type, unsigned long begin,
                                unsigned long end);
+extern void free_swap_cache(struct page *);
 extern void free_page_and_swap_cache(struct page *);
 extern void free_pages_and_swap_cache(struct page **, int);
 extern struct page *lookup_swap_cache(swp_entry_t entry,
@@ -551,6 +552,10 @@ static inline void put_swap_device(struct swap_info_struct *si)
 #define free_pages_and_swap_cache(pages, nr) \
        release_pages((pages), (nr));
 
+static inline void free_swap_cache(struct page *page)
+{
+}
+
 static inline void show_swap_cache_info(void)
 {
 }
index b15367c285bde4df06dea6a269af51ee583c3f9f..a4d82a6de0003605d7832653321957fc86bd6f8b 100644 (file)
@@ -3023,6 +3023,8 @@ static vm_fault_t wp_page_copy(struct vm_fault *vmf)
                                munlock_vma_page(old_page);
                        unlock_page(old_page);
                }
+               if (page_copied)
+                       free_swap_cache(old_page);
                put_page(old_page);
        }
        return page_copied ? VM_FAULT_WRITE : 0;
index 1a2ba4056f3760bd41805a37ff9f215b8e078220..4f8a912ff69255c86405359f5b8176350c8297db 100644 (file)
@@ -286,7 +286,7 @@ void clear_shadow_from_swap_cache(int type, unsigned long begin,
  * try_to_free_swap() _with_ the lock.
  *                                     - Marcelo
  */
-static inline void free_swap_cache(struct page *page)
+void free_swap_cache(struct page *page)
 {
        if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
                try_to_free_swap(page);