bcachefs: btree key cache pcpu freedlist

Originally, the btree key cache code would always allocate new entries
by reusing from the recently-freed list, if that list wasn't empty. But
that behaviour was dropped, for lock contention reasons.

But it seems that entries stranded on the freed list have been
contributing to some of our oom issues, because long running btree
transactions will prevent them from being freed.

This patch re-adds allocating from the freed list, but it also adds
percpu buffers to solve the lock contention issues - and the new percpu
freed lists will improve the evict paths, too.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

1 files changed, 101 insertions, 20 deletions

diff --git a/fs/bcachefs/btree_key_cache.c b/fs/bcachefs/btree_key_cache.c
index bc0c8386e403..97c72f3917ec 100644
--- a/fs/bcachefs/btree_key_cache.c
+++ b/fs/bcachefs/btree_key_cache.c
@@ -84,7 +84,7 @@ static void bkey_cached_free(struct btree_key_cache *bc,
 		start_poll_synchronize_srcu(&c->btree_trans_barrier);
 
 	list_move_tail(&ck->list, &bc->freed);
-	bc->nr_freed++;
+	atomic_long_inc(&bc->nr_freed);
 
 	kfree(ck->k);
 	ck->k		= NULL;
@@ -94,10 +94,88 @@ static void bkey_cached_free(struct btree_key_cache *bc,
 	six_unlock_intent(&ck->c.lock);
 }
 
+static void bkey_cached_free_fast(struct btree_key_cache *bc,
+				  struct bkey_cached *ck)
+{
+	struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
+	struct btree_key_cache_freelist *f;
+	bool freed = false;
+
+	BUG_ON(test_bit(BKEY_CACHED_DIRTY, &ck->flags));
+
+	ck->btree_trans_barrier_seq =
+		start_poll_synchronize_srcu(&c->btree_trans_barrier);
+
+	list_del_init(&ck->list);
+	atomic_long_inc(&bc->nr_freed);
+
+	kfree(ck->k);
+	ck->k		= NULL;
+	ck->u64s	= 0;
+
+	preempt_disable();
+	f = this_cpu_ptr(bc->pcpu_freed);
+
+	if (f->nr < ARRAY_SIZE(f->objs)) {
+		f->objs[f->nr++] = ck;
+		freed = true;
+	}
+	preempt_enable();
+
+	if (!freed) {
+		mutex_lock(&bc->lock);
+		preempt_disable();
+		f = this_cpu_ptr(bc->pcpu_freed);
+
+		while (f->nr > ARRAY_SIZE(f->objs) / 2) {
+			struct bkey_cached *ck2 = f->objs[--f->nr];
+
+			list_move_tail(&ck2->list, &bc->freed);
+		}
+		preempt_enable();
+
+		list_move_tail(&ck->list, &bc->freed);
+		mutex_unlock(&bc->lock);
+	}
+
+	six_unlock_write(&ck->c.lock);
+	six_unlock_intent(&ck->c.lock);
+}
+
 static struct bkey_cached *
 bkey_cached_alloc(struct btree_key_cache *c)
 {
-	struct bkey_cached *ck;
+	struct bkey_cached *ck = NULL;
+	struct btree_key_cache_freelist *f;
+
+	preempt_disable();
+	f = this_cpu_ptr(c->pcpu_freed);
+	if (f->nr)
+		ck = f->objs[--f->nr];
+	preempt_enable();
+
+	if (!ck) {
+		mutex_lock(&c->lock);
+		preempt_disable();
+		f = this_cpu_ptr(c->pcpu_freed);
+
+		while (!list_empty(&c->freed) &&
+		       f->nr < ARRAY_SIZE(f->objs) / 2) {
+			ck = list_last_entry(&c->freed, struct bkey_cached, list);
+			list_del_init(&ck->list);
+			f->objs[f->nr++] = ck;
+		}
+
+		ck = f->nr ? f->objs[--f->nr] : NULL;
+		preempt_enable();
+		mutex_unlock(&c->lock);
+	}
+
+	if (ck) {
+		six_lock_intent(&ck->c.lock, NULL, NULL);
+		six_lock_write(&ck->c.lock, NULL, NULL);
+		return ck;
+	}
 
 	ck = kmem_cache_alloc(bch2_key_cache, GFP_NOFS|__GFP_ZERO);
 	if (likely(ck)) {
@@ -120,16 +198,6 @@ bkey_cached_reuse(struct btree_key_cache *c)
 	struct bkey_cached *ck;
 	unsigned i;
 
-	mutex_lock(&c->lock);
-	list_for_each_entry_reverse(ck, &c->freed, list)
-		if (bkey_cached_lock_for_evict(ck)) {
-			c->nr_freed--;
-			list_del(&ck->list);
-			mutex_unlock(&c->lock);
-			return ck;
-		}
-	mutex_unlock(&c->lock);
-
 	rcu_read_lock();
 	tbl = rht_dereference_rcu(c->table.tbl, &c->table);
 	for (i = 0; i < tbl->size; i++)
@@ -190,9 +258,7 @@ btree_key_cache_create(struct bch_fs *c,
 			six_unlock_intent(&ck->c.lock);
 			kfree(ck);
 		} else {
-			mutex_lock(&bc->lock);
-			bkey_cached_free(bc, ck);
-			mutex_unlock(&bc->lock);
+			bkey_cached_free_fast(bc, ck);
 		}
 
 		return NULL;
@@ -465,9 +531,7 @@ evict:
 
 		bkey_cached_evict(&c->btree_key_cache, ck);
 
-		mutex_lock(&c->btree_key_cache.lock);
-		bkey_cached_free(&c->btree_key_cache, ck);
-		mutex_unlock(&c->btree_key_cache.lock);
+		bkey_cached_free_fast(&c->btree_key_cache, ck);
 	}
 out:
 	bch2_trans_iter_exit(trans, &b_iter);
@@ -612,7 +676,7 @@ static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink,
 
 		list_del(&ck->list);
 		kmem_cache_free(bch2_key_cache, ck);
-		bc->nr_freed--;
+		atomic_long_dec(&bc->nr_freed);
 		scanned++;
 		freed++;
 	}
@@ -685,6 +749,7 @@ void bch2_fs_btree_key_cache_exit(struct btree_key_cache *bc)
 	struct bkey_cached *ck, *n;
 	struct rhash_head *pos;
 	unsigned i;
+	int cpu;
 
 	if (bc->shrink.list.next)
 		unregister_shrinker(&bc->shrink);
@@ -701,6 +766,16 @@ void bch2_fs_btree_key_cache_exit(struct btree_key_cache *bc)
 			}
 	rcu_read_unlock();
 
+	for_each_possible_cpu(cpu) {
+		struct btree_key_cache_freelist *f =
+			per_cpu_ptr(bc->pcpu_freed, cpu);
+
+		for (i = 0; i < f->nr; i++) {
+			ck = f->objs[i];
+			list_add(&ck->list, &bc->freed);
+		}
+	}
+
 	list_for_each_entry_safe(ck, n, &bc->freed, list) {
 		cond_resched();
 
@@ -721,6 +796,8 @@ void bch2_fs_btree_key_cache_exit(struct btree_key_cache *bc)
 
 	if (bc->table_init_done)
 		rhashtable_destroy(&bc->table);
+
+	free_percpu(bc->pcpu_freed);
 }
 
 void bch2_fs_btree_key_cache_init_early(struct btree_key_cache *c)
@@ -734,6 +811,10 @@ int bch2_fs_btree_key_cache_init(struct btree_key_cache *bc)
 	struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
 	int ret;
 
+	bc->pcpu_freed = alloc_percpu(struct btree_key_cache_freelist);
+	if (!bc->pcpu_freed)
+		return -ENOMEM;
+
 	ret = rhashtable_init(&bc->table, &bch2_btree_key_cache_params);
 	if (ret)
 		return ret;
@@ -748,7 +829,7 @@ int bch2_fs_btree_key_cache_init(struct btree_key_cache *bc)
 
 void bch2_btree_key_cache_to_text(struct printbuf *out, struct btree_key_cache *c)
 {
-	prt_printf(out, "nr_freed:\t%zu\n",	c->nr_freed);
+	prt_printf(out, "nr_freed:\t%zu\n",	atomic_long_read(&c->nr_freed));
 	prt_printf(out, "nr_keys:\t%lu\n",	atomic_long_read(&c->nr_keys));
 	prt_printf(out, "nr_dirty:\t%lu\n",	atomic_long_read(&c->nr_dirty));
 }