5176 lock contention on godfather zio

Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Alex Reece <alex.reece@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Richard Elling <richard.elling@gmail.com>
Reviewed by: Bayard Bell <Bayard.Bell@nexenta.com>
Approved by: Garrett D'Amore <garrett@damore.org>

4 files changed, 26 insertions, 11 deletions

diff --git a/usr/src/cmd/zdb/zdb.c b/usr/src/cmd/zdb/zdb.c
index 2def6241b7..c886a9dc72 100644
--- a/usr/src/cmd/zdb/zdb.c
+++ b/usr/src/cmd/zdb/zdb.c
@@ -2565,10 +2565,12 @@ dump_block_stats(spa_t *spa)
 	 * all async I/Os to complete.
 	 */
 	if (dump_opt['c']) {
-		(void) zio_wait(spa->spa_async_zio_root);
-		spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
-		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
-		    ZIO_FLAG_GODFATHER);
+		for (int i = 0; i < max_ncpus; i++) {
+			(void) zio_wait(spa->spa_async_zio_root[i]);
+			spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
+			    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
+			    ZIO_FLAG_GODFATHER);
+		}
 	}
 
 	if (zcb.zcb_haderrors) {
diff --git a/usr/src/uts/common/fs/zfs/spa.c b/usr/src/uts/common/fs/zfs/spa.c
index 72870b5898..be308e2f87 100644
--- a/usr/src/uts/common/fs/zfs/spa.c
+++ b/usr/src/uts/common/fs/zfs/spa.c
@@ -1222,7 +1222,9 @@ spa_unload(spa_t *spa)
 	 * Wait for any outstanding async I/O to complete.
 	 */
 	if (spa->spa_async_zio_root != NULL) {
-		(void) zio_wait(spa->spa_async_zio_root);
+		for (int i = 0; i < max_ncpus; i++)
+			(void) zio_wait(spa->spa_async_zio_root[i]);
+		kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
 		spa->spa_async_zio_root = NULL;
 	}
 
@@ -2141,8 +2143,13 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
 	/*
 	 * Create "The Godfather" zio to hold all async IOs
 	 */
-	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
-	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
+	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
+	    KM_SLEEP);
+	for (int i = 0; i < max_ncpus; i++) {
+		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
+		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
+		    ZIO_FLAG_GODFATHER);
+	}
 
 	/*
 	 * Parse the configuration into a vdev tree.  We explicitly set the
@@ -3481,8 +3488,13 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 	/*
 	 * Create "The Godfather" zio to hold all async IOs
 	 */
-	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
-	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
+	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
+	    KM_SLEEP);
+	for (int i = 0; i < max_ncpus; i++) {
+		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
+		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
+		    ZIO_FLAG_GODFATHER);
+	}
 
 	/*
 	 * Create the root vdev.
diff --git a/usr/src/uts/common/fs/zfs/sys/spa_impl.h b/usr/src/uts/common/fs/zfs/sys/spa_impl.h
index 739b95988c..48b28eb5a1 100644
--- a/usr/src/uts/common/fs/zfs/sys/spa_impl.h
+++ b/usr/src/uts/common/fs/zfs/sys/spa_impl.h
@@ -204,7 +204,8 @@ struct spa {
 	uint64_t	spa_failmode;		/* failure mode for the pool */
 	uint64_t	spa_delegation;		/* delegation on/off */
 	list_t		spa_config_list;	/* previous cache file(s) */
-	zio_t		*spa_async_zio_root;	/* root of all async I/O */
+	/* per-CPU array of root of async I/O: */
+	zio_t		**spa_async_zio_root;
 	zio_t		*spa_suspend_zio_root;	/* root of all suspended I/O */
 	kmutex_t	spa_suspend_lock;	/* protects suspend_zio_root */
 	kcondvar_t	spa_suspend_cv;		/* notification of resume */
diff --git a/usr/src/uts/common/fs/zfs/zio.c b/usr/src/uts/common/fs/zfs/zio.c
index d8833f6707..b154cb639d 100644
--- a/usr/src/uts/common/fs/zfs/zio.c
+++ b/usr/src/uts/common/fs/zfs/zio.c
@@ -1367,7 +1367,7 @@ zio_nowait(zio_t *zio)
 		 */
 		spa_t *spa = zio->io_spa;
 
-		zio_add_child(spa->spa_async_zio_root, zio);
+		zio_add_child(spa->spa_async_zio_root[CPU_SEQID], zio);
 	}
 
 	zio_execute(zio);