[PATCH] account for slab reclaim in try_to_free_pages()

try_to_free_pages() currently fails to notice that it successfully freed slab
pages via shrink_slab().  So it can keep looping and eventually call
out_of_memory(), even though there's a lot of memory now free.

And even if it doesn't do that, it can free too much memory.

The patch changes try_to_free_pages() so that it will notice freed slab pages
and will return when enough memory has been freed via shrink_slab().

Many options were considered, but must of them were unacceptably inaccurate,
intrusive or sleazy.  I ended up putting the accounting into a stack-local
structure which is pointed to by current->reclaim_state.

One reason for this is that we can cleanly resurrect the current->local_pages
pool by putting it into struct reclaim_state.

(current->local_pages was removed because the per-cpu page pools in the page
allocator largely duplicate its function.  But it is still possible for
interrupt-time allocations to steal just-freed pages, so we might want to put
it back some time.)

include/linux/sched.h

@@ -294,6 +294,7 @@ extern struct user_struct root_user;
 
 typedef struct prio_array prio_array_t;
 struct backing_dev_info;
+struct reclaim_state;
 
 /* POSIX.1b interval timer structure. */
 struct k_itimer {
@@ -433,6 +434,10 @@ struct task_struct {
 
 /* journalling filesystem info */
 	void *journal_info;
+
+/* VM state */
+	struct reclaim_state *reclaim_state;
+
 	struct dentry *proc_dentry;
 	struct backing_dev_info *backing_dev_info;
 

include/linux/swap.h

@@ -66,6 +66,14 @@ typedef struct {
 	unsigned long val;
 } swp_entry_t;
 
+/*
+ * current->reclaim_state points to one of these when a task is running
+ * memory reclaim
+ */
+struct reclaim_state {
+	unsigned long reclaimed_slab;
+};
+
 #ifdef __KERNEL__
 
 struct address_space;

mm/page_alloc.c

@@ -537,6 +537,7 @@ __alloc_pages(unsigned int gfp_mask, unsigned int order,
 	int i;
 	int cold;
 	int do_retry;
+	struct reclaim_state reclaim_state;
 
 	if (wait)
 		might_sleep();
@@ -608,7 +609,12 @@ __alloc_pages(unsigned int gfp_mask, unsigned int order,
 		goto nopage;
 
 	current->flags |= PF_MEMALLOC;
+	reclaim_state.reclaimed_slab = 0;
+	current->reclaim_state = &reclaim_state;
+
 	try_to_free_pages(classzone, gfp_mask, order);
+
+	current->reclaim_state = NULL;
 	current->flags &= ~PF_MEMALLOC;
 
 	/* go through the zonelist yet one more time */

mm/slab.c

@@ -76,6 +76,7 @@
 #include	<linux/config.h>
 #include	<linux/slab.h>
 #include	<linux/mm.h>
+#include	<linux/swap.h>
 #include	<linux/cache.h>
 #include	<linux/interrupt.h>
 #include	<linux/init.h>
@@ -734,6 +735,7 @@ static inline void kmem_freepages (kmem_cache_t *cachep, void *addr)
 {
 	unsigned long i = (1<<cachep->gfporder);
 	struct page *page = virt_to_page(addr);
+	const unsigned long nr_freed = i;
 
 	/* free_pages() does not clear the type bit - we do that.
 	 * The pages have been unlinked from their cache-slab,
@@ -742,9 +744,11 @@ static inline void kmem_freepages (kmem_cache_t *cachep, void *addr)
 	 */
 	while (i--) {
 		ClearPageSlab(page);
-		dec_page_state(nr_slab);
 		page++;
 	}
+	sub_page_state(nr_slab, nr_freed);
+	if (current->reclaim_state)
+		current->reclaim_state->reclaimed_slab += nr_freed;
 	free_pages((unsigned long)addr, cachep->gfporder);
 }
 

mm/vmscan.c

@@ -817,12 +817,14 @@ shrink_caches(struct zone *classzone, int priority, int *total_scanned,
  * excessive rotation of the inactive list, which is _supposed_ to be an LRU,
  * yes?
  */
-int try_to_free_pages(struct zone *classzone,
+int try_to_free_pages(struct zone *cz,
 		unsigned int gfp_mask, unsigned int order)
 {
 	int priority;
+	int ret = 0;
 	const int nr_pages = SWAP_CLUSTER_MAX;
 	int nr_reclaimed = 0;
+	struct reclaim_state *reclaim_state = current->reclaim_state;
 
 	inc_page_state(allocstall);
 
@@ -831,11 +833,12 @@ int try_to_free_pages(struct zone *classzone,
 		struct page_state ps;
 
 		get_page_state(&ps);
-		nr_reclaimed += shrink_caches(classzone, priority,
-					&total_scanned, gfp_mask,
-					nr_pages, &ps);
-		if (nr_reclaimed >= nr_pages)
-			return 1;
+		nr_reclaimed += shrink_caches(cz, priority, &total_scanned,
+						gfp_mask, nr_pages, &ps);
+		if (nr_reclaimed >= nr_pages) {
+			ret = 1;
+			goto out;
+		}
 		if (!(gfp_mask & __GFP_FS))
 			break;		/* Let the caller handle it */
 		/*
@@ -847,12 +850,18 @@ int try_to_free_pages(struct zone *classzone,
 
 		/* Take a nap, wait for some writeback to complete */
 		blk_congestion_wait(WRITE, HZ/10);
-		if (classzone - classzone->zone_pgdat->node_zones < ZONE_HIGHMEM)
+		if (cz - cz->zone_pgdat->node_zones < ZONE_HIGHMEM) {
 			shrink_slab(total_scanned, gfp_mask);
+			if (reclaim_state) {
+				nr_reclaimed += reclaim_state->reclaimed_slab;
+				reclaim_state->reclaimed_slab = 0;
+			}
+		}
 	}
 	if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY))
 		out_of_memory();
-	return 0;
+out:
+	return ret;
 }
 
 /*
@@ -878,6 +887,7 @@ static int balance_pgdat(pg_data_t *pgdat, int nr_pages, struct page_state *ps)
 	int to_free = nr_pages;
 	int priority;
 	int i;
+	struct reclaim_state *reclaim_state = current->reclaim_state;
 
 	inc_page_state(pageoutrun);
 
@@ -908,8 +918,11 @@ static int balance_pgdat(pg_data_t *pgdat, int nr_pages, struct page_state *ps)
 				max_scan = SWAP_CLUSTER_MAX;
 			to_free -= shrink_zone(zone, max_scan, GFP_KERNEL,
 					to_reclaim, &nr_mapped, ps, priority);
-			if (i < ZONE_HIGHMEM)
+			if (i < ZONE_HIGHMEM) {
+				reclaim_state->reclaimed_slab = 0;
 				shrink_slab(max_scan + nr_mapped, GFP_KERNEL);
+				to_free += reclaim_state->reclaimed_slab;
+			}
 			if (zone->all_unreclaimable)
 				continue;
 			if (zone->pages_scanned > zone->present_pages * 2)
@@ -940,10 +953,14 @@ int kswapd(void *p)
 	pg_data_t *pgdat = (pg_data_t*)p;
 	struct task_struct *tsk = current;
 	DEFINE_WAIT(wait);
+	struct reclaim_state reclaim_state = {
+		.reclaimed_slab = 0,
+	};
 
 	daemonize("kswapd%d", pgdat->node_id);
 	set_cpus_allowed(tsk, node_to_cpumask(pgdat->node_id));
-	
+	current->reclaim_state = &reclaim_state;
+
 	/*
 	 * Tell the memory management that we're a "memory allocator",
 	 * and that if we need more memory we should get access to it