Upgrades trim free request sizes before inserting them into to free map,

making range consolidation much more effective particularly for small
deletes.

This reduces memory used by the free map as well as reducing the number
of bio requests down to geom required to process all deletes.

In tests this achieved a factor of 10 reduction of trim ranges / geom
call downs.

While I'm here correct the description of zio_vdev_io_start.

PR:		kern/173254
Submitted by:	Steven Hartland
Approved by:	pjd (mentor)

1 files changed, 13 insertions, 2 deletions

diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/trim_map.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/trim_map.c
index 5de5f2a..4bf0734 100644
--- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/trim_map.c
+++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/trim_map.c
@@ -28,6 +28,17 @@
 #include <sys/vdev_impl.h>
 #include <sys/trim_map.h>
 
+/*
+ * Calculate the zio end, upgrading based on ashift which would be
+ * done by zio_vdev_io_start.
+ *
+ * This makes free range consolidation much more effective
+ * than it would otherwise be as well as ensuring that entire
+ * blocks are invalidated by writes.
+ */
+#define	TRIM_ZIO_END(zio)	((zio)->io_offset +		\
+ 	P2ROUNDUP((zio)->io_size, 1ULL << (zio)->io_vd->vdev_top->vdev_ashift))
+
 typedef struct trim_map {
 	list_t		tm_head;		/* List of segments sorted by txg. */
 	avl_tree_t	tm_queued_frees;	/* AVL tree of segments waiting for TRIM. */
@@ -270,7 +281,7 @@ trim_map_free(zio_t *zio)
 		return;
 
 	mutex_enter(&tm->tm_lock);
-	trim_map_free_locked(tm, zio->io_offset, zio->io_offset + zio->io_size,
+	trim_map_free_locked(tm, zio->io_offset, TRIM_ZIO_END(zio),
 	    vd->vdev_spa->spa_syncing_txg);
 	mutex_exit(&tm->tm_lock);
 }
@@ -288,7 +299,7 @@ trim_map_write_start(zio_t *zio)
 		return (B_TRUE);
 
 	start = zio->io_offset;
-	end = start + zio->io_size;
+	end = TRIM_ZIO_END(zio);
 	tsearch.ts_start = start;
 	tsearch.ts_end = end;