shmem: convert shmem_replace_page() to use folios throughout
authorMatthew Wilcox (Oracle) <willy@infradead.org>
Fri, 2 Sep 2022 19:46:04 +0000 (20:46 +0100)
committerAndrew Morton <akpm@linux-foundation.org>
Mon, 3 Oct 2022 21:02:45 +0000 (14:02 -0700)
Introduce folio_set_swap_entry() to abstract how both folio->private and
swp_entry_t work.  Use swap_address_space() directly instead of
indirecting through folio_mapping().  Include an assertion that the old
folio is not large as we only allocate a single-page folio to replace it.
Use folio_put_refs() instead of calling folio_put() twice.

Link: https://lkml.kernel.org/r/20220902194653.1739778-9-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
include/linux/swap.h
mm/shmem.c

index 2ede1e3..61e13d1 100644 (file)
@@ -355,6 +355,11 @@ static inline swp_entry_t folio_swap_entry(struct folio *folio)
        return entry;
 }
 
+static inline void folio_set_swap_entry(struct folio *folio, swp_entry_t entry)
+{
+       folio->private = (void *)entry.val;
+}
+
 /* linux/mm/workingset.c */
 void workingset_age_nonresident(struct lruvec *lruvec, unsigned long nr_pages);
 void *workingset_eviction(struct folio *folio, struct mem_cgroup *target_memcg);
index 9e851fc..4113f1b 100644 (file)
@@ -1560,12 +1560,6 @@ static struct folio *shmem_alloc_folio(gfp_t gfp,
        return folio;
 }
 
-static struct page *shmem_alloc_page(gfp_t gfp,
-                       struct shmem_inode_info *info, pgoff_t index)
-{
-       return &shmem_alloc_folio(gfp, info, index)->page;
-}
-
 static struct folio *shmem_alloc_and_acct_folio(gfp_t gfp, struct inode *inode,
                pgoff_t index, bool huge)
 {
@@ -1617,51 +1611,49 @@ static bool shmem_should_replace_folio(struct folio *folio, gfp_t gfp)
 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
                                struct shmem_inode_info *info, pgoff_t index)
 {
-       struct page *oldpage, *newpage;
        struct folio *old, *new;
        struct address_space *swap_mapping;
        swp_entry_t entry;
        pgoff_t swap_index;
        int error;
 
-       oldpage = *pagep;
-       entry.val = page_private(oldpage);
+       old = page_folio(*pagep);
+       entry = folio_swap_entry(old);
        swap_index = swp_offset(entry);
-       swap_mapping = page_mapping(oldpage);
+       swap_mapping = swap_address_space(entry);
 
        /*
         * We have arrived here because our zones are constrained, so don't
         * limit chance of success by further cpuset and node constraints.
         */
        gfp &= ~GFP_CONSTRAINT_MASK;
-       newpage = shmem_alloc_page(gfp, info, index);
-       if (!newpage)
+       VM_BUG_ON_FOLIO(folio_test_large(old), old);
+       new = shmem_alloc_folio(gfp, info, index);
+       if (!new)
                return -ENOMEM;
 
-       get_page(newpage);
-       copy_highpage(newpage, oldpage);
-       flush_dcache_page(newpage);
+       folio_get(new);
+       folio_copy(new, old);
+       flush_dcache_folio(new);
 
-       __SetPageLocked(newpage);
-       __SetPageSwapBacked(newpage);
-       SetPageUptodate(newpage);
-       set_page_private(newpage, entry.val);
-       SetPageSwapCache(newpage);
+       __folio_set_locked(new);
+       __folio_set_swapbacked(new);
+       folio_mark_uptodate(new);
+       folio_set_swap_entry(new, entry);
+       folio_set_swapcache(new);
 
        /*
         * Our caller will very soon move newpage out of swapcache, but it's
         * a nice clean interface for us to replace oldpage by newpage there.
         */
        xa_lock_irq(&swap_mapping->i_pages);
-       error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
+       error = shmem_replace_entry(swap_mapping, swap_index, old, new);
        if (!error) {
-               old = page_folio(oldpage);
-               new = page_folio(newpage);
                mem_cgroup_migrate(old, new);
-               __inc_lruvec_page_state(newpage, NR_FILE_PAGES);
-               __inc_lruvec_page_state(newpage, NR_SHMEM);
-               __dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
-               __dec_lruvec_page_state(oldpage, NR_SHMEM);
+               __lruvec_stat_mod_folio(new, NR_FILE_PAGES, 1);
+               __lruvec_stat_mod_folio(new, NR_SHMEM, 1);
+               __lruvec_stat_mod_folio(old, NR_FILE_PAGES, -1);
+               __lruvec_stat_mod_folio(old, NR_SHMEM, -1);
        }
        xa_unlock_irq(&swap_mapping->i_pages);
 
@@ -1671,18 +1663,17 @@ static int shmem_replace_page(struct page **pagep, gfp_t gfp,
                 * both PageSwapCache and page_private after getting page lock;
                 * but be defensive.  Reverse old to newpage for clear and free.
                 */
-               oldpage = newpage;
+               old = new;
        } else {
-               lru_cache_add(newpage);
-               *pagep = newpage;
+               folio_add_lru(new);
+               *pagep = &new->page;
        }
 
-       ClearPageSwapCache(oldpage);
-       set_page_private(oldpage, 0);
+       folio_clear_swapcache(old);
+       old->private = NULL;
 
-       unlock_page(oldpage);
-       put_page(oldpage);
-       put_page(oldpage);
+       folio_unlock(old);
+       folio_put_refs(old, 2);
        return error;
 }
 
@@ -2383,6 +2374,12 @@ static struct inode *shmem_get_inode(struct super_block *sb, struct inode *dir,
 }
 
 #ifdef CONFIG_USERFAULTFD
+static struct page *shmem_alloc_page(gfp_t gfp,
+                       struct shmem_inode_info *info, pgoff_t index)
+{
+       return &shmem_alloc_folio(gfp, info, index)->page;
+}
+
 int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
                           pmd_t *dst_pmd,
                           struct vm_area_struct *dst_vma,