tmpfs: move swap_state stats update
Hugh Dickins [Tue, 5 Feb 2008 06:28:49 +0000 (22:28 -0800)]
Both unionfs and memcgroups pose challenges to tmpfs and shmem.  To help fix,
it's best to move the swap swizzling functions from swap_state.c to shmem.c.
As a preliminary to that, move swap stats updating down into
__add_to_swap_cache, which will remain internal to swap_state.c.

Well, actually, just move down the incrementation of add_total: remove
noent_race and exist_race completely, they are relics of my 2.4.11 testing.
Alt-SysRq-m users will be thrilled if 2.6.25 is at last free of "race M+N"s.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/swap_state.c

index e787564..18fce36 100644 (file)
@@ -52,16 +52,13 @@ static struct {
        unsigned long del_total;
        unsigned long find_success;
        unsigned long find_total;
-       unsigned long noent_race;
-       unsigned long exist_race;
 } swap_cache_info;
 
 void show_swap_cache_info(void)
 {
-       printk("Swap cache: add %lu, delete %lu, find %lu/%lu, race %lu+%lu\n",
+       printk("Swap cache: add %lu, delete %lu, find %lu/%lu\n",
                swap_cache_info.add_total, swap_cache_info.del_total,
-               swap_cache_info.find_success, swap_cache_info.find_total,
-               swap_cache_info.noent_race, swap_cache_info.exist_race);
+               swap_cache_info.find_success, swap_cache_info.find_total);
        printk("Free swap  = %lukB\n", nr_swap_pages << (PAGE_SHIFT - 10));
        printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
 }
@@ -89,6 +86,7 @@ static int __add_to_swap_cache(struct page *page, swp_entry_t entry,
                        set_page_private(page, entry.val);
                        total_swapcache_pages++;
                        __inc_zone_page_state(page, NR_FILE_PAGES);
+                       INC_CACHE_INFO(add_total);
                }
                write_unlock_irq(&swapper_space.tree_lock);
                radix_tree_preload_end();
@@ -102,10 +100,9 @@ static int add_to_swap_cache(struct page *page, swp_entry_t entry,
        int error;
 
        BUG_ON(PageLocked(page));
-       if (!swap_duplicate(entry)) {
-               INC_CACHE_INFO(noent_race);
+       if (!swap_duplicate(entry))
                return -ENOENT;
-       }
+
        SetPageLocked(page);
        error = __add_to_swap_cache(page, entry, gfp_mask & GFP_KERNEL);
        /*
@@ -114,11 +111,8 @@ static int add_to_swap_cache(struct page *page, swp_entry_t entry,
        if (error) {
                ClearPageLocked(page);
                swap_free(entry);
-               if (error == -EEXIST)
-                       INC_CACHE_INFO(exist_race);
                return error;
        }
-       INC_CACHE_INFO(add_total);
        return 0;
 }
 
@@ -178,11 +172,9 @@ int add_to_swap(struct page * page, gfp_t gfp_mask)
                case 0:                         /* Success */
                        SetPageUptodate(page);
                        SetPageDirty(page);
-                       INC_CACHE_INFO(add_total);
                        return 1;
                case -EEXIST:
                        /* Raced with "speculative" read_swap_cache_async */
-                       INC_CACHE_INFO(exist_race);
                        swap_free(entry);
                        continue;
                default:
@@ -225,9 +217,7 @@ int move_to_swap_cache(struct page *page, swp_entry_t entry)
                if (!swap_duplicate(entry))
                        BUG();
                SetPageDirty(page);
-               INC_CACHE_INFO(add_total);
-       } else if (err == -EEXIST)
-               INC_CACHE_INFO(exist_race);
+       }
        return err;
 }