hpfs: implement prefetch to improve performance

This patch implements prefetch to improve performance.  It helps mostly
when scanning the bitmaps to calculate free space.

Signed-off-by: Mikulas Patocka <mpatocka@artax.karlin.mff.cuni.cz>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

4 files changed, 63 insertions, 5 deletions

diff --git a/fs/hpfs/buffer.c b/fs/hpfs/buffer.c
index f49d149..4d0a1af 100644
--- a/fs/hpfs/buffer.c
+++ b/fs/hpfs/buffer.c
@@ -7,8 +7,37 @@
  */
 #include <linux/sched.h>
 #include <linux/slab.h>
+#include <linux/blkdev.h>
 #include "hpfs_fn.h"
 
+void hpfs_prefetch_sectors(struct super_block *s, unsigned secno, int n)
+{
+	struct buffer_head *bh;
+	struct blk_plug plug;
+
+	if (n <= 0 || unlikely(secno >= hpfs_sb(s)->sb_fs_size))
+		return;
+
+	bh = sb_find_get_block(s, secno);
+	if (bh) {
+		if (buffer_uptodate(bh)) {
+			brelse(bh);
+			return;
+		}
+		brelse(bh);
+	};
+
+	blk_start_plug(&plug);
+	while (n > 0) {
+		if (unlikely(secno >= hpfs_sb(s)->sb_fs_size))
+			break;
+		sb_breadahead(s, secno);
+		secno++;
+		n--;
+	}
+	blk_finish_plug(&plug);
+}
+
 /* Map a sector into a buffer and return pointers to it and to the buffer. */
 
 void *hpfs_map_sector(struct super_block *s, unsigned secno, struct buffer_head **bhp,
@@ -18,6 +47,8 @@ void *hpfs_map_sector(struct super_block *s, unsigned secno, struct buffer_head
 
 	hpfs_lock_assert(s);
 
+	hpfs_prefetch_sectors(s, secno, ahead);
+
 	cond_resched();
 
 	*bhp = bh = sb_bread(s, secno);
@@ -67,6 +98,8 @@ void *hpfs_map_4sectors(struct super_block *s, unsigned secno, struct quad_buffe
 		return NULL;
 	}
 
+	hpfs_prefetch_sectors(s, secno, 4 + ahead);
+
 	qbh->data = data = kmalloc(2048, GFP_NOFS);
 	if (!data) {
 		printk("HPFS: hpfs_map_4sectors: out of memory\n");
diff --git a/fs/hpfs/hpfs_fn.h b/fs/hpfs/hpfs_fn.h
index b7ae286..1b39863 100644
--- a/fs/hpfs/hpfs_fn.h
+++ b/fs/hpfs/hpfs_fn.h
@@ -27,8 +27,9 @@
 #define ALLOC_FWD_MAX	128
 #define ALLOC_M		1
 #define FNODE_RD_AHEAD	16
-#define ANODE_RD_AHEAD	16
-#define DNODE_RD_AHEAD	4
+#define ANODE_RD_AHEAD	0
+#define DNODE_RD_AHEAD	72
+#define COUNT_RD_AHEAD	62
 
 #define FREE_DNODES_ADD	58
 #define FREE_DNODES_DEL	29
@@ -207,6 +208,7 @@ void hpfs_remove_fnode(struct super_block *, fnode_secno fno);
 
 /* buffer.c */
 
+void hpfs_prefetch_sectors(struct super_block *, unsigned, int);
 void *hpfs_map_sector(struct super_block *, unsigned, struct buffer_head **, int);
 void *hpfs_get_sector(struct super_block *, unsigned, struct buffer_head **);
 void *hpfs_map_4sectors(struct super_block *, unsigned, struct quad_buffer_head *, int);
@@ -271,6 +273,7 @@ void hpfs_evict_inode(struct inode *);
 
 __le32 *hpfs_map_dnode_bitmap(struct super_block *, struct quad_buffer_head *);
 __le32 *hpfs_map_bitmap(struct super_block *, unsigned, struct quad_buffer_head *, char *);
+void hpfs_prefetch_bitmap(struct super_block *, unsigned);
 unsigned char *hpfs_load_code_page(struct super_block *, secno);
 __le32 *hpfs_load_bitmap_directory(struct super_block *, secno bmp);
 struct fnode *hpfs_map_fnode(struct super_block *s, ino_t, struct buffer_head **);
diff --git a/fs/hpfs/map.c b/fs/hpfs/map.c
index 803d3da..3aa66ae 100644
--- a/fs/hpfs/map.c
+++ b/fs/hpfs/map.c
@@ -17,6 +17,7 @@ __le32 *hpfs_map_bitmap(struct super_block *s, unsigned bmp_block,
 			 struct quad_buffer_head *qbh, char *id)
 {
 	secno sec;
+	__le32 *ret;
 	unsigned n_bands = (hpfs_sb(s)->sb_fs_size + 0x3fff) >> 14;
 	if (hpfs_sb(s)->sb_chk) if (bmp_block >= n_bands) {
 		hpfs_error(s, "hpfs_map_bitmap called with bad parameter: %08x at %s", bmp_block, id);
@@ -27,7 +28,23 @@ __le32 *hpfs_map_bitmap(struct super_block *s, unsigned bmp_block,
 		hpfs_error(s, "invalid bitmap block pointer %08x -> %08x at %s", bmp_block, sec, id);
 		return NULL;
 	}
-	return hpfs_map_4sectors(s, sec, qbh, 4);
+	ret = hpfs_map_4sectors(s, sec, qbh, 4);
+	if (ret) hpfs_prefetch_bitmap(s, bmp_block + 1);
+	return ret;
+}
+
+void hpfs_prefetch_bitmap(struct super_block *s, unsigned bmp_block)
+{
+	unsigned to_prefetch, next_prefetch;
+	unsigned n_bands = (hpfs_sb(s)->sb_fs_size + 0x3fff) >> 14;
+	if (unlikely(bmp_block >= n_bands))
+		return;
+	to_prefetch = le32_to_cpu(hpfs_sb(s)->sb_bmp_dir[bmp_block]);
+	if (unlikely(bmp_block + 1 >= n_bands))
+		next_prefetch = 0;
+	else
+		next_prefetch = le32_to_cpu(hpfs_sb(s)->sb_bmp_dir[bmp_block + 1]);
+	hpfs_prefetch_sectors(s, to_prefetch, 4 + 4 * (to_prefetch + 4 == next_prefetch));
 }
 
 /*
diff --git a/fs/hpfs/super.c b/fs/hpfs/super.c
index 962e90c..4334cda 100644
--- a/fs/hpfs/super.c
+++ b/fs/hpfs/super.c
@@ -121,7 +121,7 @@ unsigned hpfs_count_one_bitmap(struct super_block *s, secno secno)
 	unsigned long *bits;
 	unsigned count;
 
-	bits = hpfs_map_4sectors(s, secno, &qbh, 4);
+	bits = hpfs_map_4sectors(s, secno, &qbh, 0);
 	if (!bits)
 		return 0;
 	count = bitmap_weight(bits, 2048 * BITS_PER_BYTE);
@@ -134,8 +134,13 @@ static unsigned count_bitmaps(struct super_block *s)
 	unsigned n, count, n_bands;
 	n_bands = (hpfs_sb(s)->sb_fs_size + 0x3fff) >> 14;
 	count = 0;
-	for (n = 0; n < n_bands; n++)
+	for (n = 0; n < COUNT_RD_AHEAD; n++) {
+		hpfs_prefetch_bitmap(s, n);
+	}
+	for (n = 0; n < n_bands; n++) {
+		hpfs_prefetch_bitmap(s, n + COUNT_RD_AHEAD);
 		count += hpfs_count_one_bitmap(s, le32_to_cpu(hpfs_sb(s)->sb_bmp_dir[n]));
+	}
 	return count;
 }