Pre-allocate vdev I/O buffers

The vdev queue layer may require a small number of buffers
when attempting to create aggregate I/O requests.  Rather than
attempting to allocate them from the global zio buffers, which
is slow under memory pressure, it makes sense to pre-allocate
them because...

1) These buffers are short lived.  They are only required for
the life of a single I/O at which point they can be used by
the next I/O.

2) The maximum number of concurrent buffers needed by a vdev is
small.  It's roughly limited by the zfs_vdev_max_pending tunable
which defaults to 10.

By keeping a small list of these buffer per-vdev we can ensure
one is always available when we need it.  This significantly
reduces contention on the vq->vq_lock, because we no longer
need to perform a slow allocation under this lock.  This is
particularly important when memory is already low on the system.

It would probably be wise to extend the use of these buffers beyond
aggregate I/O and in to the raidz implementation.  The inability
to quickly allocate buffer for the parity stripes could result in
similiar problems.

Signed-off-by: Brian Behlendorf <[email protected]>

1 files changed, 22 insertions, 0 deletions

diff --git a/module/zfs/zio.c b/module/zfs/zio.c
index 206ed9a93..fe2bdc867 100644
--- a/module/zfs/zio.c
+++ b/module/zfs/zio.c
@@ -73,6 +73,7 @@ char *zio_type_name[ZIO_TYPES] = {
  */
 kmem_cache_t *zio_cache;
 kmem_cache_t *zio_link_cache;
+kmem_cache_t *zio_vdev_cache;
 kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
 kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
 int zio_bulk_flags = 0;
@@ -141,6 +142,8 @@ zio_init(void)
 	    zio_cons, zio_dest, NULL, NULL, NULL, KMC_KMEM);
 	zio_link_cache = kmem_cache_create("zio_link_cache",
 	    sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, KMC_KMEM);
+	zio_vdev_cache = kmem_cache_create("zio_vdev_cache", sizeof(vdev_io_t),
+	    PAGESIZE, NULL, NULL, NULL, NULL, NULL, KMC_VMEM);
 
 	/*
 	 * For small buffers, we want a cache for each multiple of
@@ -230,6 +233,7 @@ zio_fini(void)
 		zio_data_buf_cache[c] = NULL;
 	}
 
+	kmem_cache_destroy(zio_vdev_cache);
 	kmem_cache_destroy(zio_link_cache);
 	kmem_cache_destroy(zio_cache);
 
@@ -295,6 +299,24 @@ zio_data_buf_free(void *buf, size_t size)
 }
 
 /*
+ * Dedicated I/O buffers to ensure that memory fragmentation never prevents
+ * or significantly delays the issuing of a zio.   These buffers are used
+ * to aggregate I/O and could be used for raidz stripes.
+ */
+void *
+zio_vdev_alloc(void)
+{
+	return (kmem_cache_alloc(zio_vdev_cache, KM_PUSHPAGE));
+}
+
+void
+zio_vdev_free(void *buf)
+{
+	kmem_cache_free(zio_vdev_cache, buf);
+
+}
+
+/*
  * ==========================================================================
  * Push and pop I/O transform buffers
  * ==========================================================================