mm: new shrinker API
authorDave Chinner <dchinner@redhat.com>
Wed, 28 Aug 2013 00:17:56 +0000 (10:17 +1000)
committerAl Viro <viro@zeniv.linux.org.uk>
Tue, 10 Sep 2013 22:56:30 +0000 (18:56 -0400)
The current shrinker callout API uses an a single shrinker call for
multiple functions.  To determine the function, a special magical value is
passed in a parameter to change the behaviour.  This complicates the
implementation and return value specification for the different
behaviours.

Separate the two different behaviours into separate operations, one to
return a count of freeable objects in the cache, and another to scan a
certain number of objects in the cache for freeing.  In defining these new
operations, ensure the return values and resultant behaviours are clearly
defined and documented.

Modify shrink_slab() to use the new API and implement the callouts for all
the existing shrinkers.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Glauber Costa <glommer@parallels.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Cc: Arve Hjønnevåg <arve@android.com>
Cc: Carlos Maiolino <cmaiolino@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: David Rientjes <rientjes@google.com>
Cc: Gleb Natapov <gleb@redhat.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: J. Bruce Fields <bfields@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: John Stultz <john.stultz@linaro.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Kent Overstreet <koverstreet@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Cc: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
include/linux/shrinker.h
mm/vmscan.c

index ac6b8ee07825ee6ce99dff9faf9901a224a97b5e..884e76222e1baf66d3a7649a3f07dcf7f54f52fe 100644 (file)
@@ -4,6 +4,12 @@
 /*
  * This struct is used to pass information from page reclaim to the shrinkers.
  * We consolidate the values for easier extention later.
+ *
+ * The 'gfpmask' refers to the allocation we are currently trying to
+ * fulfil.
+ *
+ * Note that 'shrink' will be passed nr_to_scan == 0 when the VM is
+ * querying the cache size, so a fastpath for that case is appropriate.
  */
 struct shrink_control {
        gfp_t gfp_mask;
@@ -12,23 +18,37 @@ struct shrink_control {
        unsigned long nr_to_scan;
 };
 
+#define SHRINK_STOP (~0UL)
 /*
  * A callback you can register to apply pressure to ageable caches.
  *
- * 'sc' is passed shrink_control which includes a count 'nr_to_scan'
- * and a 'gfpmask'.  It should look through the least-recently-used
- * 'nr_to_scan' entries and attempt to free them up.  It should return
- * the number of objects which remain in the cache.  If it returns -1, it means
- * it cannot do any scanning at this time (eg. there is a risk of deadlock).
+ * @shrink() should look through the least-recently-used 'nr_to_scan' entries
+ * and attempt to free them up.  It should return the number of objects which
+ * remain in the cache.  If it returns -1, it means it cannot do any scanning at
+ * this time (eg. there is a risk of deadlock).
  *
- * The 'gfpmask' refers to the allocation we are currently trying to
- * fulfil.
+ * @count_objects should return the number of freeable items in the cache. If
+ * there are no objects to free or the number of freeable items cannot be
+ * determined, it should return 0. No deadlock checks should be done during the
+ * count callback - the shrinker relies on aggregating scan counts that couldn't
+ * be executed due to potential deadlocks to be run at a later call when the
+ * deadlock condition is no longer pending.
  *
- * Note that 'shrink' will be passed nr_to_scan == 0 when the VM is
- * querying the cache size, so a fastpath for that case is appropriate.
+ * @scan_objects will only be called if @count_objects returned a non-zero
+ * value for the number of freeable objects. The callout should scan the cache
+ * and attempt to free items from the cache. It should then return the number
+ * of objects freed during the scan, or SHRINK_STOP if progress cannot be made
+ * due to potential deadlocks. If SHRINK_STOP is returned, then no further
+ * attempts to call the @scan_objects will be made from the current reclaim
+ * context.
  */
 struct shrinker {
        int (*shrink)(struct shrinker *, struct shrink_control *sc);
+       unsigned long (*count_objects)(struct shrinker *,
+                                      struct shrink_control *sc);
+       unsigned long (*scan_objects)(struct shrinker *,
+                                     struct shrink_control *sc);
+
        int seeks;      /* seeks to recreate an obj */
        long batch;     /* reclaim batch size, 0 = default */
 
index 2cff0d491c6dca84391edd100e1726696c1475d5..4d4e859b4b9c7e59853eafbeb4348afb4a701a5b 100644 (file)
@@ -205,19 +205,24 @@ static inline int do_shrinker_shrink(struct shrinker *shrinker,
  *
  * Returns the number of slab objects which we shrunk.
  */
-unsigned long shrink_slab(struct shrink_control *shrink,
+unsigned long shrink_slab(struct shrink_control *shrinkctl,
                          unsigned long nr_pages_scanned,
                          unsigned long lru_pages)
 {
        struct shrinker *shrinker;
-       unsigned long ret = 0;
+       unsigned long freed = 0;
 
        if (nr_pages_scanned == 0)
                nr_pages_scanned = SWAP_CLUSTER_MAX;
 
        if (!down_read_trylock(&shrinker_rwsem)) {
-               /* Assume we'll be able to shrink next time */
-               ret = 1;
+               /*
+                * If we would return 0, our callers would understand that we
+                * have nothing else to shrink and give up trying. By returning
+                * 1 we keep it going and assume we'll be able to shrink next
+                * time.
+                */
+               freed = 1;
                goto out;
        }
 
@@ -225,14 +230,16 @@ unsigned long shrink_slab(struct shrink_control *shrink,
                unsigned long long delta;
                long total_scan;
                long max_pass;
-               int shrink_ret = 0;
                long nr;
                long new_nr;
                long batch_size = shrinker->batch ? shrinker->batch
                                                  : SHRINK_BATCH;
 
-               max_pass = do_shrinker_shrink(shrinker, shrink, 0);
-               if (max_pass <= 0)
+               if (shrinker->count_objects)
+                       max_pass = shrinker->count_objects(shrinker, shrinkctl);
+               else
+                       max_pass = do_shrinker_shrink(shrinker, shrinkctl, 0);
+               if (max_pass == 0)
                        continue;
 
                /*
@@ -248,8 +255,8 @@ unsigned long shrink_slab(struct shrink_control *shrink,
                do_div(delta, lru_pages + 1);
                total_scan += delta;
                if (total_scan < 0) {
-                       printk(KERN_ERR "shrink_slab: %pF negative objects to "
-                              "delete nr=%ld\n",
+                       printk(KERN_ERR
+                       "shrink_slab: %pF negative objects to delete nr=%ld\n",
                               shrinker->shrink, total_scan);
                        total_scan = max_pass;
                }
@@ -277,20 +284,33 @@ unsigned long shrink_slab(struct shrink_control *shrink,
                if (total_scan > max_pass * 2)
                        total_scan = max_pass * 2;
 
-               trace_mm_shrink_slab_start(shrinker, shrink, nr,
+               trace_mm_shrink_slab_start(shrinker, shrinkctl, nr,
                                        nr_pages_scanned, lru_pages,
                                        max_pass, delta, total_scan);
 
                while (total_scan >= batch_size) {
-                       int nr_before;
 
-                       nr_before = do_shrinker_shrink(shrinker, shrink, 0);
-                       shrink_ret = do_shrinker_shrink(shrinker, shrink,
-                                                       batch_size);
-                       if (shrink_ret == -1)
-                               break;
-                       if (shrink_ret < nr_before)
-                               ret += nr_before - shrink_ret;
+                       if (shrinker->scan_objects) {
+                               unsigned long ret;
+                               shrinkctl->nr_to_scan = batch_size;
+                               ret = shrinker->scan_objects(shrinker, shrinkctl);
+
+                               if (ret == SHRINK_STOP)
+                                       break;
+                               freed += ret;
+                       } else {
+                               int nr_before;
+                               long ret;
+
+                               nr_before = do_shrinker_shrink(shrinker, shrinkctl, 0);
+                               ret = do_shrinker_shrink(shrinker, shrinkctl,
+                                                               batch_size);
+                               if (ret == -1)
+                                       break;
+                               if (ret < nr_before)
+                                       freed += nr_before - ret;
+                       }
+
                        count_vm_events(SLABS_SCANNED, batch_size);
                        total_scan -= batch_size;
 
@@ -308,12 +328,12 @@ unsigned long shrink_slab(struct shrink_control *shrink,
                else
                        new_nr = atomic_long_read(&shrinker->nr_in_batch);
 
-               trace_mm_shrink_slab_end(shrinker, shrink_ret, nr, new_nr);
+               trace_mm_shrink_slab_end(shrinker, freed, nr, new_nr);
        }
        up_read(&shrinker_rwsem);
 out:
        cond_resched();
-       return ret;
+       return freed;
 }
 
 static inline int is_page_cache_freeable(struct page *page)