swap: make each swap partition have one address_space

When I use several fast SSD to do swap, swapper_space.tree_lock is
heavily contended.  This makes each swap partition have one
address_space to reduce the lock contention.  There is an array of
address_space for swap.  The swap entry type is the index to the array.

In my test with 3 SSD, this increases the swapout throughput 20%.

[akpm@linux-foundation.org: revert unneeded change to  __add_to_swap_cache]
Signed-off-by: Shaohua Li <shli@fusionio.com>
Cc: Hugh Dickins <hughd@google.com>
Acked-by: Rik van Riel <riel@redhat.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Change-Id: I8503ace83342398bf7be3d2216616868cca65311

fs/proc/meminfo.c

@@ -40,7 +40,7 @@ static int meminfo_proc_show(struct seq_file *m, void *v)
 		* sysctl_overcommit_ratio / 100) + total_swap_pages;
 
 	cached = global_page_state(NR_FILE_PAGES) -
-			total_swapcache_pages - i.bufferram;
+			total_swapcache_pages() - i.bufferram;
 	if (cached < 0)
 		cached = 0;
 
@@ -109,7 +109,7 @@ static int meminfo_proc_show(struct seq_file *m, void *v)
 		K(i.freeram),
 		K(i.bufferram),
 		K(cached),
-		K(total_swapcache_pages),
+		K(total_swapcache_pages()),
 		K(pages[LRU_ACTIVE_ANON]   + pages[LRU_ACTIVE_FILE]),
 		K(pages[LRU_INACTIVE_ANON] + pages[LRU_INACTIVE_FILE]),
 		K(pages[LRU_ACTIVE_ANON]),

include/linux/swap.h

@@ -8,7 +8,7 @@
 #include <linux/memcontrol.h>
 #include <linux/sched.h>
 #include <linux/node.h>
-
+#include <linux/fs.h>
 #include <linux/atomic.h>
 #include <asm/page.h>
 
@@ -319,8 +319,9 @@ extern int swap_writepage(struct page *page, struct writeback_control *wbc);
 extern void end_swap_bio_read(struct bio *bio, int err);
 
 /* linux/mm/swap_state.c */
-extern struct address_space swapper_space;
+extern struct address_space swapper_spaces[];
-#define total_swapcache_pages  swapper_space.nrpages
+#define swap_address_space(entry) (&swapper_spaces[swp_type(entry)])
+extern unsigned long total_swapcache_pages(void);
 extern void show_swap_cache_info(void);
 extern int add_to_swap(struct page *);
 extern int add_to_swap_cache(struct page *, swp_entry_t, gfp_t);
@@ -388,7 +389,7 @@ mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
 
 #define nr_swap_pages				0L
 #define total_swap_pages			0L
-#define total_swapcache_pages			0UL
+#define total_swapcache_pages()			0UL
 
 #define si_swapinfo(val) \
 	do { (val)->freeswap = (val)->totalswap = 0; } while (0)

mm/memcontrol.c

@@ -5186,7 +5186,7 @@ static struct page *mc_handle_swap_pte(struct vm_area_struct *vma,
 	 * Because lookup_swap_cache() updates some statistics counter,
 	 * we call find_get_page() with swapper_space directly.
 	 */
-	page = find_get_page(&swapper_space, ent.val);
+	page = find_get_page(swap_address_space(ent), ent.val);
 	if (do_swap_account)
 		entry->val = ent.val;
 
@@ -5229,7 +5229,7 @@ static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
 		swp_entry_t swap = radix_to_swp_entry(page);
 		if (do_swap_account)
 			*entry = swap;
-		page = find_get_page(&swapper_space, swap.val);
+		page = find_get_page(swap_address_space(swap), swap.val);
 	}
 #endif
 	return page;

mm/mincore.c

@@ -75,7 +75,7 @@ static unsigned char mincore_page(struct address_space *mapping, pgoff_t pgoff)
 	/* shmem/tmpfs may return swap: account for swapcache page too. */
 	if (radix_tree_exceptional_entry(page)) {
 		swp_entry_t swap = radix_to_swp_entry(page);
-		page = find_get_page(&swapper_space, swap.val);
+		page = find_get_page(swap_address_space(swap), swap.val);
 	}
 #endif
 	if (page) {
@@ -135,7 +135,8 @@ static void mincore_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 			} else {
 #ifdef CONFIG_SWAP
 				pgoff = entry.val;
-				*vec = mincore_page(&swapper_space, pgoff);
+				*vec = mincore_page(swap_address_space(entry),
+					pgoff);
 #else
 				WARN_ON(1);
 				*vec = 1;

mm/swap.c

@@ -843,9 +843,14 @@ EXPORT_SYMBOL(pagevec_lookup_tag);
 void __init swap_setup(void)
 {
 	unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT);
-
 #ifdef CONFIG_SWAP
-	bdi_init(swapper_space.backing_dev_info);
+	int i;
+
+	bdi_init(swapper_spaces[0].backing_dev_info);
+	for (i = 0; i < MAX_SWAPFILES; i++) {
+		spin_lock_init(&swapper_spaces[i].tree_lock);
+		INIT_LIST_HEAD(&swapper_spaces[i].i_mmap_nonlinear);
+	}
 #endif
 
 	/* Use a smaller cluster for small-memory machines */

mm/swap_state.c

@@ -35,12 +35,12 @@ static struct backing_dev_info swap_backing_dev_info = {
 	.capabilities	= BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
 };
 
-struct address_space swapper_space = {
+struct address_space swapper_spaces[MAX_SWAPFILES] = {
-	.page_tree	= RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
+	[0 ... MAX_SWAPFILES - 1] = {
-	.tree_lock	= __SPIN_LOCK_UNLOCKED(swapper_space.tree_lock),
+		.page_tree	= RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
-	.a_ops		= &swap_aops,
+		.a_ops		= &swap_aops,
-	.i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear),
+		.backing_dev_info = &swap_backing_dev_info,
-	.backing_dev_info = &swap_backing_dev_info,
+	}
 };
 
 #define INC_CACHE_INFO(x)	do { swap_cache_info.x++; } while (0)
@@ -52,9 +52,19 @@ static struct {
 	unsigned long find_total;
 } swap_cache_info;
 
+unsigned long total_swapcache_pages(void)
+{
+	int i;
+	unsigned long ret = 0;
+
+	for (i = 0; i < MAX_SWAPFILES; i++)
+		ret += swapper_spaces[i].nrpages;
+	return ret;
+}
+
 void show_swap_cache_info(void)
 {
-	printk("%lu pages in swap cache\n", total_swapcache_pages);
+	printk("%lu pages in swap cache\n", total_swapcache_pages());
 	printk("Swap cache stats: 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);
@@ -69,6 +79,7 @@ void show_swap_cache_info(void)
 static int __add_to_swap_cache(struct page *page, swp_entry_t entry)
 {
 	int error;
+	struct address_space *address_space;
 
 	VM_BUG_ON(!PageLocked(page));
 	VM_BUG_ON(PageSwapCache(page));
@@ -78,14 +89,16 @@ static int __add_to_swap_cache(struct page *page, swp_entry_t entry)
 	SetPageSwapCache(page);
 	set_page_private(page, entry.val);
 
-	spin_lock_irq(&swapper_space.tree_lock);
+	address_space = swap_address_space(entry);
-	error = radix_tree_insert(&swapper_space.page_tree, entry.val, page);
+	spin_lock_irq(&address_space->tree_lock);
+	error = radix_tree_insert(&address_space->page_tree,
+					entry.val, page);
 	if (likely(!error)) {
-		total_swapcache_pages++;
+		address_space->nrpages++;
 		__inc_zone_page_state(page, NR_FILE_PAGES);
 		INC_CACHE_INFO(add_total);
 	}
-	spin_unlock_irq(&swapper_space.tree_lock);
+	spin_unlock_irq(&address_space->tree_lock);
 
 	if (unlikely(error)) {
 		/*
@@ -121,14 +134,19 @@ int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
  */
 void __delete_from_swap_cache(struct page *page)
 {
+	swp_entry_t entry;
+	struct address_space *address_space;
+
 	VM_BUG_ON(!PageLocked(page));
 	VM_BUG_ON(!PageSwapCache(page));
 	VM_BUG_ON(PageWriteback(page));
 
-	radix_tree_delete(&swapper_space.page_tree, page_private(page));
+	entry.val = page_private(page);
+	address_space = swap_address_space(entry);
+	radix_tree_delete(&address_space->page_tree, page_private(page));
 	set_page_private(page, 0);
 	ClearPageSwapCache(page);
-	total_swapcache_pages--;
+	address_space->nrpages--;
 	__dec_zone_page_state(page, NR_FILE_PAGES);
 	INC_CACHE_INFO(del_total);
 }
@@ -194,12 +212,14 @@ int add_to_swap(struct page *page)
 void delete_from_swap_cache(struct page *page)
 {
 	swp_entry_t entry;
+	struct address_space *address_space;
 
 	entry.val = page_private(page);
 
-	spin_lock_irq(&swapper_space.tree_lock);
+	address_space = swap_address_space(entry);
+	spin_lock_irq(&address_space->tree_lock);
 	__delete_from_swap_cache(page);
-	spin_unlock_irq(&swapper_space.tree_lock);
+	spin_unlock_irq(&address_space->tree_lock);
 
 	swapcache_free(entry, page);
 	page_cache_release(page);
@@ -262,7 +282,7 @@ struct page * lookup_swap_cache(swp_entry_t entry)
 {
 	struct page *page;
 
-	page = find_get_page(&swapper_space, entry.val);
+	page = find_get_page(swap_address_space(entry), entry.val);
 
 	if (page)
 		INC_CACHE_INFO(find_success);
@@ -289,7 +309,8 @@ struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
 		 * called after lookup_swap_cache() failed, re-calling
 		 * that would confuse statistics.
 		 */
-		found_page = find_get_page(&swapper_space, entry.val);
+		found_page = find_get_page(swap_address_space(entry),
+					entry.val);
 		if (found_page)
 			break;
 

mm/swapfile.c

@@ -76,7 +76,7 @@ __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
 	struct page *page;
 	int ret = 0;
 
-	page = find_get_page(&swapper_space, entry.val);
+	page = find_get_page(swap_address_space(entry), entry.val);
 	if (!page)
 		return 0;
 	/*
@@ -735,7 +735,8 @@ int free_swap_and_cache(swp_entry_t entry)
 	p = swap_info_get(entry);
 	if (p) {
 		if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
-			page = find_get_page(&swapper_space, entry.val);
+			page = find_get_page(swap_address_space(entry),
+						entry.val);
 			if (page && !trylock_page(page)) {
 				page_cache_release(page);
 				page = NULL;

mm/util.c

@@ -360,9 +360,12 @@ struct address_space *page_mapping(struct page *page)
 
 	VM_BUG_ON(PageSlab(page));
 #ifdef CONFIG_SWAP
-	if (unlikely(PageSwapCache(page)))
+	if (unlikely(PageSwapCache(page))) {
-		mapping = &swapper_space;
+		swp_entry_t entry;
-	else
+
+		entry.val = page_private(page);
+		mapping = swap_address_space(entry);
+	} else
 #endif
 	if ((unsigned long)mapping & PAGE_MAPPING_ANON)
 		mapping = NULL;