Merge branch 'rds-more-fixes'

Santosh Shilimkar says:

====================
RDS: Few more fixes

As indicated in the earlier series [1], this is a follow-up series which
addresses few issues around the RDS FMR code. With [1] and the subject
series, now I can run many parallel threads with multiple sockets with
N x N traffic. The stress tests has survived overnight runs.

[1] https://lkml.org/lkml/2015/8/22/127
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

5 files changed, 50 insertions, 17 deletions

diff --git a/net/rds/ib.c b/net/rds/ib.c
index 13814227b3b2..d020fade312c 100644
--- a/net/rds/ib.c
+++ b/net/rds/ib.c
@@ -366,6 +366,7 @@ void rds_ib_exit(void)
 	rds_ib_sysctl_exit();
 	rds_ib_recv_exit();
 	rds_trans_unregister(&rds_ib_transport);
+	rds_ib_fmr_exit();
 }
 
 struct rds_transport rds_ib_transport = {
@@ -401,10 +402,14 @@ int rds_ib_init(void)
 
 	INIT_LIST_HEAD(&rds_ib_devices);
 
-	ret = ib_register_client(&rds_ib_client);
+	ret = rds_ib_fmr_init();
 	if (ret)
 		goto out;
 
+	ret = ib_register_client(&rds_ib_client);
+	if (ret)
+		goto out_fmr_exit;
+
 	ret = rds_ib_sysctl_init();
 	if (ret)
 		goto out_ibreg;
@@ -427,6 +432,8 @@ out_sysctl:
 	rds_ib_sysctl_exit();
 out_ibreg:
 	rds_ib_unregister_client();
+out_fmr_exit:
+	rds_ib_fmr_exit();
 out:
 	return ret;
 }
diff --git a/net/rds/ib.h b/net/rds/ib.h
index 6422c52682e5..9fc95e38659a 100644
--- a/net/rds/ib.h
+++ b/net/rds/ib.h
@@ -313,6 +313,8 @@ void *rds_ib_get_mr(struct scatterlist *sg, unsigned long nents,
 void rds_ib_sync_mr(void *trans_private, int dir);
 void rds_ib_free_mr(void *trans_private, int invalidate);
 void rds_ib_flush_mrs(void);
+int rds_ib_fmr_init(void);
+void rds_ib_fmr_exit(void);
 
 /* ib_recv.c */
 int rds_ib_recv_init(void);
diff --git a/net/rds/ib_rdma.c b/net/rds/ib_rdma.c
index 7b7aac8cdb56..251d1ce0b7c7 100644
--- a/net/rds/ib_rdma.c
+++ b/net/rds/ib_rdma.c
@@ -83,6 +83,25 @@ struct rds_ib_mr_pool {
 	struct ib_fmr_attr	fmr_attr;
 };
 
+struct workqueue_struct *rds_ib_fmr_wq;
+
+int rds_ib_fmr_init(void)
+{
+	rds_ib_fmr_wq = create_workqueue("rds_fmr_flushd");
+	if (!rds_ib_fmr_wq)
+		return -ENOMEM;
+	return 0;
+}
+
+/* By the time this is called all the IB devices should have been torn down and
+ * had their pools freed.  As each pool is freed its work struct is waited on,
+ * so the pool flushing work queue should be idle by the time we get here.
+ */
+void rds_ib_fmr_exit(void)
+{
+	destroy_workqueue(rds_ib_fmr_wq);
+}
+
 static int rds_ib_flush_mr_pool(struct rds_ib_mr_pool *pool, int free_all, struct rds_ib_mr **);
 static void rds_ib_teardown_mr(struct rds_ib_mr *ibmr);
 static void rds_ib_mr_pool_flush_worker(struct work_struct *work);
@@ -341,8 +360,6 @@ static struct rds_ib_mr *rds_ib_alloc_fmr(struct rds_ib_device *rds_ibdev)
 		goto out_no_cigar;
 	}
 
-	memset(ibmr, 0, sizeof(*ibmr));
-
 	ibmr->fmr = ib_alloc_fmr(rds_ibdev->pd,
 			(IB_ACCESS_LOCAL_WRITE |
 			 IB_ACCESS_REMOTE_READ |
@@ -528,11 +545,13 @@ static inline unsigned int rds_ib_flush_goal(struct rds_ib_mr_pool *pool, int fr
 /*
  * given an llist of mrs, put them all into the list_head for more processing
  */
-static void llist_append_to_list(struct llist_head *llist, struct list_head *list)
+static unsigned int llist_append_to_list(struct llist_head *llist,
+					 struct list_head *list)
 {
 	struct rds_ib_mr *ibmr;
 	struct llist_node *node;
 	struct llist_node *next;
+	unsigned int count = 0;
 
 	node = llist_del_all(llist);
 	while (node) {
@@ -540,7 +559,9 @@ static void llist_append_to_list(struct llist_head *llist, struct list_head *lis
 		ibmr = llist_entry(node, struct rds_ib_mr, llnode);
 		list_add_tail(&ibmr->unmap_list, list);
 		node = next;
+		count++;
 	}
+	return count;
 }
 
 /*
@@ -581,7 +602,7 @@ static int rds_ib_flush_mr_pool(struct rds_ib_mr_pool *pool,
 	LIST_HEAD(unmap_list);
 	LIST_HEAD(fmr_list);
 	unsigned long unpinned = 0;
-	unsigned int nfreed = 0, ncleaned = 0, free_goal;
+	unsigned int nfreed = 0, dirty_to_clean = 0, free_goal;
 	int ret = 0;
 
 	rds_ib_stats_inc(s_ib_rdma_mr_pool_flush);
@@ -623,8 +644,8 @@ static int rds_ib_flush_mr_pool(struct rds_ib_mr_pool *pool,
 	/* Get the list of all MRs to be dropped. Ordering matters -
 	 * we want to put drop_list ahead of free_list.
 	 */
-	llist_append_to_list(&pool->drop_list, &unmap_list);
-	llist_append_to_list(&pool->free_list, &unmap_list);
+	dirty_to_clean = llist_append_to_list(&pool->drop_list, &unmap_list);
+	dirty_to_clean += llist_append_to_list(&pool->free_list, &unmap_list);
 	if (free_all)
 		llist_append_to_list(&pool->clean_list, &unmap_list);
 
@@ -652,7 +673,6 @@ static int rds_ib_flush_mr_pool(struct rds_ib_mr_pool *pool,
 			kfree(ibmr);
 			nfreed++;
 		}
-		ncleaned++;
 	}
 
 	if (!list_empty(&unmap_list)) {
@@ -678,7 +698,7 @@ static int rds_ib_flush_mr_pool(struct rds_ib_mr_pool *pool,
 	}
 
 	atomic_sub(unpinned, &pool->free_pinned);
-	atomic_sub(ncleaned, &pool->dirty_count);
+	atomic_sub(dirty_to_clean, &pool->dirty_count);
 	atomic_sub(nfreed, &pool->item_count);
 
 out:
@@ -715,16 +735,18 @@ void rds_ib_free_mr(void *trans_private, int invalidate)
 
 	/* If we've pinned too many pages, request a flush */
 	if (atomic_read(&pool->free_pinned) >= pool->max_free_pinned ||
-	    atomic_read(&pool->dirty_count) >= pool->max_items / 10)
-		schedule_delayed_work(&pool->flush_worker, 10);
+	    atomic_read(&pool->dirty_count) >= pool->max_items / 5)
+		queue_delayed_work(rds_ib_fmr_wq, &pool->flush_worker, 10);
 
 	if (invalidate) {
 		if (likely(!in_interrupt())) {
 			rds_ib_flush_mr_pool(pool, 0, NULL);
 		} else {
 			/* We get here if the user created a MR marked
-			 * as use_once and invalidate at the same time. */
-			schedule_delayed_work(&pool->flush_worker, 10);
+			 * as use_once and invalidate at the same time.
+			 */
+			queue_delayed_work(rds_ib_fmr_wq,
+					   &pool->flush_worker, 10);
 		}
 	}
 
diff --git a/net/rds/ib_recv.c b/net/rds/ib_recv.c
index ed9b41e3b277..6bbe62060060 100644
--- a/net/rds/ib_recv.c
+++ b/net/rds/ib_recv.c
@@ -1102,9 +1102,10 @@ int rds_ib_recv_init(void)
 	rds_ib_frag_slab = kmem_cache_create("rds_ib_frag",
 					sizeof(struct rds_page_frag),
 					0, SLAB_HWCACHE_ALIGN, NULL);
-	if (!rds_ib_frag_slab)
+	if (!rds_ib_frag_slab) {
 		kmem_cache_destroy(rds_ib_incoming_slab);
-	else
+		rds_ib_incoming_slab = NULL;
+	} else
 		ret = 0;
 out:
 	return ret;
diff --git a/net/rds/rdma.c b/net/rds/rdma.c
index c1df9b1cf3b2..4c93badeabf2 100644
--- a/net/rds/rdma.c
+++ b/net/rds/rdma.c
@@ -435,9 +435,10 @@ void rds_rdma_unuse(struct rds_sock *rs, u32 r_key, int force)
 
 	/* If the MR was marked as invalidate, this will
 	 * trigger an async flush. */
-	if (zot_me)
+	if (zot_me) {
 		rds_destroy_mr(mr);
-	rds_mr_put(mr);
+		rds_mr_put(mr);
+	}
 }
 
 void rds_rdma_free_op(struct rm_rdma_op *ro)