xfs: reduce the number of pagb_lock roundtrips in xfs_alloc_clear_busy

Instead of finding the per-ag and then taking and releasing the pagb_lock
for every single busy extent completed sort the list of busy extents and
only switch betweens AGs where nessecary.  This becomes especially important
with the online discard support which will hit this lock more often.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>

fs/xfs/linux-2.6/xfs_buf.c

@@ -33,7 +33,6 @@
 #include <linux/migrate.h>
 #include <linux/backing-dev.h>
 #include <linux/freezer.h>
-#include <linux/list_sort.h>
 
 #include "xfs_sb.h"
 #include "xfs_inum.h"

fs/xfs/linux-2.6/xfs_linux.h

@@ -70,6 +70,7 @@
 #include <linux/ctype.h>
 #include <linux/writeback.h>
 #include <linux/capability.h>
+#include <linux/list_sort.h>
 
 #include <asm/page.h>
 #include <asm/div64.h>

fs/xfs/xfs_alloc.c

@@ -2964,24 +2964,60 @@ fail:
 	*rlen = 0;
 }
 
-void
-xfs_alloc_busy_clear(
+static void
+xfs_alloc_busy_clear_one(
 	struct xfs_mount	*mp,
+	struct xfs_perag	*pag,
 	struct xfs_busy_extent	*busyp)
 {
-	struct xfs_perag	*pag;
-
-	list_del_init(&busyp->list);
-
-	pag = xfs_perag_get(mp, busyp->agno);
-	spin_lock(&pag->pagb_lock);
 	if (busyp->length) {
 		trace_xfs_alloc_busy_clear(mp, busyp->agno, busyp->bno,
 						busyp->length);
 		rb_erase(&busyp->rb_node, &pag->pagb_tree);
 	}
-	spin_unlock(&pag->pagb_lock);
-	xfs_perag_put(pag);
 
+	list_del_init(&busyp->list);
 	kmem_free(busyp);
 }
+
+void
+xfs_alloc_busy_clear(
+	struct xfs_mount	*mp,
+	struct list_head	*list)
+{
+	struct xfs_busy_extent	*busyp, *n;
+	struct xfs_perag	*pag = NULL;
+	xfs_agnumber_t		agno = NULLAGNUMBER;
+
+	list_for_each_entry_safe(busyp, n, list, list) {
+		if (busyp->agno != agno) {
+			if (pag) {
+				spin_unlock(&pag->pagb_lock);
+				xfs_perag_put(pag);
+			}
+			pag = xfs_perag_get(mp, busyp->agno);
+			spin_lock(&pag->pagb_lock);
+			agno = busyp->agno;
+		}
+
+		xfs_alloc_busy_clear_one(mp, pag, busyp);
+	}
+
+	if (pag) {
+		spin_unlock(&pag->pagb_lock);
+		xfs_perag_put(pag);
+	}
+}
+
+/*
+ * Callback for list_sort to sort busy extents by the AG they reside in.
+ */
+int
+xfs_busy_extent_ag_cmp(
+	void			*priv,
+	struct list_head	*a,
+	struct list_head	*b)
+{
+	return container_of(a, struct xfs_busy_extent, list)->agno -
+		container_of(b, struct xfs_busy_extent, list)->agno;
+}

fs/xfs/xfs_alloc.h

@@ -140,7 +140,7 @@ xfs_alloc_busy_insert(struct xfs_trans *tp, xfs_agnumber_t agno,
 	xfs_agblock_t bno, xfs_extlen_t len);
 
 void
-xfs_alloc_busy_clear(struct xfs_mount *mp, struct xfs_busy_extent *busyp);
+xfs_alloc_busy_clear(struct xfs_mount *mp, struct list_head *list);
 
 int
 xfs_alloc_busy_search(struct xfs_mount *mp, xfs_agnumber_t agno,
@@ -149,6 +149,15 @@ xfs_alloc_busy_search(struct xfs_mount *mp, xfs_agnumber_t agno,
 void
 xfs_alloc_busy_reuse(struct xfs_mount *mp, xfs_agnumber_t agno,
 	xfs_agblock_t fbno, xfs_extlen_t flen, bool userdata);
+
+int
+xfs_busy_extent_ag_cmp(void *priv, struct list_head *a, struct list_head *b);
+
+static inline void xfs_alloc_busy_sort(struct list_head *list)
+{
+	list_sort(NULL, list, xfs_busy_extent_ag_cmp);
+}
+
 #endif	/* __KERNEL__ */
 
 /*

fs/xfs/xfs_log_cil.c

@@ -361,13 +361,12 @@ xlog_cil_committed(
 	int	abort)
 {
 	struct xfs_cil_ctx	*ctx = args;
-	struct xfs_busy_extent	*busyp, *n;
 
 	xfs_trans_committed_bulk(ctx->cil->xc_log->l_ailp, ctx->lv_chain,
 					ctx->start_lsn, abort);
 
-	list_for_each_entry_safe(busyp, n, &ctx->busy_extents, list)
-		xfs_alloc_busy_clear(ctx->cil->xc_log->l_mp, busyp);
+	xfs_alloc_busy_sort(&ctx->busy_extents);
+	xfs_alloc_busy_clear(ctx->cil->xc_log->l_mp, &ctx->busy_extents);
 
 	spin_lock(&ctx->cil->xc_cil_lock);
 	list_del(&ctx->committing);

fs/xfs/xfs_trans.c

@@ -608,10 +608,8 @@ STATIC void
 xfs_trans_free(
 	struct xfs_trans	*tp)
 {
-	struct xfs_busy_extent	*busyp, *n;
-
-	list_for_each_entry_safe(busyp, n, &tp->t_busy, list)
-		xfs_alloc_busy_clear(tp->t_mountp, busyp);
+	xfs_alloc_busy_sort(&tp->t_busy);
+	xfs_alloc_busy_clear(tp->t_mountp, &tp->t_busy);
 
 	atomic_dec(&tp->t_mountp->m_active_trans);
 	xfs_trans_free_dqinfo(tp);