Patchwork [V3] ext4: speed-up releasing blocks on commit

login
register
mail settings
Submitter Andrey Sidorov
Date March 28, 2013, 7:26 p.m.
Message ID <1364498818-17909-1-git-send-email-qrxd43@motorola.com>
Download mbox | patch
Permalink /patch/232172/
State Accepted
Headers show

Comments

Andrey Sidorov - March 28, 2013, 7:26 p.m.
Improve mb_free_blocks speed by clearing entire range at once instead of
iterating over each bit. Freeing block-by-block also makes buddy bitmap
subtree flip twice making most of the work a no-op. Very few bits in buddy
bitmap require change, e.g. freeing entire group is a 1 bit flip only.
As a result, releasing blocks of 60G file now takes 5ms instead of 2.7s.
This is especially good for non-preemptive kernels as there is no
rescheduling during release.

Signed-off-by: Andrey Sidorov <qrxd43@motorola.com>
---
 fs/ext4/mballoc.c |  233 +++++++++++++++++++++++++++++++++++++++--------------
 1 file changed, 172 insertions(+), 61 deletions(-)
Theodore Ts'o - April 9, 2013, 4:36 p.m.
On Thu, Mar 28, 2013 at 11:26:58PM +0400, Andrey Sidorov wrote:
> Improve mb_free_blocks speed by clearing entire range at once instead of
> iterating over each bit. Freeing block-by-block also makes buddy bitmap
> subtree flip twice making most of the work a no-op. Very few bits in buddy
> bitmap require change, e.g. freeing entire group is a 1 bit flip only.
> As a result, releasing blocks of 60G file now takes 5ms instead of 2.7s.
> This is especially good for non-preemptive kernels as there is no
> rescheduling during release.
> 
> Signed-off-by: Andrey Sidorov <qrxd43@motorola.com>

Thanks, added to the dev branch for testing.

					- Ted
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index 580aada..f92e05a 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -405,6 +405,12 @@  static inline void mb_clear_bit(int bit, void *addr)
 	ext4_clear_bit(bit, addr);
 }
 
+static inline int mb_test_and_clear_bit(int bit, void *addr)
+{
+	addr = mb_correct_addr_and_bit(&bit, addr);
+	return ext4_test_and_clear_bit(bit, addr);
+}
+
 static inline int mb_find_next_zero_bit(void *addr, int max, int start)
 {
 	int fix = 0, ret, tmpmax;
@@ -764,6 +770,24 @@  void ext4_mb_generate_buddy(struct super_block *sb,
 	spin_unlock(&EXT4_SB(sb)->s_bal_lock);
 }
 
+static void mb_regenerate_buddy(struct ext4_buddy *e4b)
+{
+	int count;
+	int order = 1;
+	void *buddy;
+
+	while ((buddy = mb_find_buddy(e4b, order++, &count))) {
+		ext4_set_bits(buddy, 0, count);
+	}
+	e4b->bd_info->bb_fragments = 0;
+	memset(e4b->bd_info->bb_counters, 0,
+		sizeof(*e4b->bd_info->bb_counters) *
+		(e4b->bd_sb->s_blocksize_bits + 2));
+
+	ext4_mb_generate_buddy(e4b->bd_sb, e4b->bd_buddy,
+		e4b->bd_bitmap, e4b->bd_group);
+}
+
 /* The buddy information is attached the buddy cache inode
  * for convenience. The information regarding each group
  * is loaded via ext4_mb_load_buddy. The information involve
@@ -1246,6 +1270,33 @@  static void mb_clear_bits(void *bm, int cur, int len)
 	}
 }
 
+/* clear bits in given range
+ * will return first found zero bit if any, -1 otherwise
+ */
+static int mb_test_and_clear_bits(void *bm, int cur, int len)
+{
+	__u32 *addr;
+	int zero_bit = -1;
+
+	len = cur + len;
+	while (cur < len) {
+		if ((cur & 31) == 0 && (len - cur) >= 32) {
+			/* fast path: clear whole word at once */
+			addr = bm + (cur >> 3);
+			if (*addr != (__u32)(-1) && zero_bit == -1)
+				zero_bit = cur + mb_find_next_zero_bit(addr, 32, 0);
+			*addr = 0;
+			cur += 32;
+			continue;
+		}
+		if (!mb_test_and_clear_bit(cur, bm) && zero_bit == -1)
+			zero_bit = cur;
+		cur++;
+	}
+
+	return zero_bit;
+}
+
 void ext4_set_bits(void *bm, int cur, int len)
 {
 	__u32 *addr;
@@ -1264,17 +1315,90 @@  void ext4_set_bits(void *bm, int cur, int len)
 	}
 }
 
+/*
+ * _________________________________________________________________ */
+
+static inline int mb_buddy_adjust_border(int* bit, void* bitmap, int side)
+{
+	if (mb_test_bit(*bit + side, bitmap)) {
+		mb_clear_bit(*bit, bitmap);
+		(*bit) -= side;
+		return 1;
+	}
+	else {
+		(*bit) += side;
+		mb_set_bit(*bit, bitmap);
+		return -1;
+	}
+}
+
+static void mb_buddy_mark_free(struct ext4_buddy *e4b, int first, int last)
+{
+	int max;
+	int order = 1;
+	void *buddy = mb_find_buddy(e4b, order, &max);
+
+	while (buddy) {
+		void *buddy2;
+
+		/* Bits in range [first; last] are known to be set since
+		 * corresponding blocks were allocated. Bits in range
+		 * (first; last) will stay set because they form buddies on
+		 * upper layer. We just deal with borders if they don't
+		 * align with upper layer and then go up.
+		 * Releasing entire group is all about clearing
+		 * single bit of highest order buddy.
+		 */
+
+		/* Example:
+		 * ---------------------------------
+		 * |   1   |   1   |   1   |   1   |
+		 * ---------------------------------
+		 * | 0 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
+		 * ---------------------------------
+		 *   0   1   2   3   4   5   6   7
+		 *      \_____________________/
+		 *
+		 * Neither [1] nor [6] is aligned to above layer.
+		 * Left neighbour [0] is free, so mark it busy,
+		 * decrease bb_counters and extend range to
+		 * [0; 6]
+		 * Right neighbour [7] is busy. It can't be coaleasced with [6], so
+		 * mark [6] free, increase bb_counters and shrink range to
+		 * [0; 5].
+		 * Then shift range to [0; 2], go up and do the same.
+		 */
+
+
+		if (first & 1)
+			e4b->bd_info->bb_counters[order] += mb_buddy_adjust_border(&first, buddy, -1);
+		if (!(last & 1))
+			e4b->bd_info->bb_counters[order] += mb_buddy_adjust_border(&last, buddy, 1);
+		if (first > last)
+			break;
+		order++;
+
+		if (first == last || !(buddy2 = mb_find_buddy(e4b, order, &max))) {
+			mb_clear_bits(buddy, first, last - first + 1);
+			e4b->bd_info->bb_counters[order - 1] += last - first + 1;
+			break;
+		}
+		first >>= 1;
+		last >>= 1;
+		buddy = buddy2;
+	}
+}
+
 static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
-			  int first, int count)
+			   int first, int count)
 {
-	int block = 0;
-	int max = 0;
-	int order;
-	void *buddy;
-	void *buddy2;
+	int left_is_free = 0;
+	int right_is_free = 0;
+	int block;
+	int last = first + count - 1;
 	struct super_block *sb = e4b->bd_sb;
 
-	BUG_ON(first + count > (sb->s_blocksize << 3));
+	BUG_ON(last >= (sb->s_blocksize << 3));
 	assert_spin_locked(ext4_group_lock_ptr(sb, e4b->bd_group));
 	mb_check_buddy(e4b);
 	mb_free_blocks_double(inode, e4b, first, count);
@@ -1283,67 +1407,54 @@  static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
 	if (first < e4b->bd_info->bb_first_free)
 		e4b->bd_info->bb_first_free = first;
 
-	/* let's maintain fragments counter */
+	/* access memory sequentially: check left neighbour,
+	 * clear range and then check right neighbour
+	 */
 	if (first != 0)
-		block = !mb_test_bit(first - 1, e4b->bd_bitmap);
-	if (first + count < EXT4_SB(sb)->s_mb_maxs[0])
-		max = !mb_test_bit(first + count, e4b->bd_bitmap);
-	if (block && max)
-		e4b->bd_info->bb_fragments--;
-	else if (!block && !max)
-		e4b->bd_info->bb_fragments++;
-
-	/* let's maintain buddy itself */
-	while (count-- > 0) {
-		block = first++;
-		order = 0;
-
-		if (!mb_test_bit(block, e4b->bd_bitmap)) {
-			ext4_fsblk_t blocknr;
-
-			blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
-			blocknr += EXT4_C2B(EXT4_SB(sb), block);
-			ext4_grp_locked_error(sb, e4b->bd_group,
-					      inode ? inode->i_ino : 0,
-					      blocknr,
-					      "freeing already freed block "
-					      "(bit %u)", block);
-		}
-		mb_clear_bit(block, e4b->bd_bitmap);
-		e4b->bd_info->bb_counters[order]++;
-
-		/* start of the buddy */
-		buddy = mb_find_buddy(e4b, order, &max);
+		left_is_free = !mb_test_bit(first - 1, e4b->bd_bitmap);
+	block = mb_test_and_clear_bits(e4b->bd_bitmap, first, count);
+	if (last + 1 < EXT4_SB(sb)->s_mb_maxs[0])
+		right_is_free = !mb_test_bit(last + 1, e4b->bd_bitmap);
 
-		do {
-			block &= ~1UL;
-			if (mb_test_bit(block, buddy) ||
-					mb_test_bit(block + 1, buddy))
-				break;
+	if (unlikely(block != -1)) {
+		ext4_fsblk_t blocknr;
 
-			/* both the buddies are free, try to coalesce them */
-			buddy2 = mb_find_buddy(e4b, order + 1, &max);
+		blocknr = ext4_group_first_block_no(sb, e4b->bd_group);
+		blocknr += EXT4_C2B(EXT4_SB(sb), block);
+		ext4_grp_locked_error(sb, e4b->bd_group,
+				      inode ? inode->i_ino : 0,
+				      blocknr,
+				      "freeing already freed block "
+				      "(bit %u)", block);
+		mb_regenerate_buddy(e4b);
+		goto done;
+	}
 
-			if (!buddy2)
-				break;
+	/* let's maintain fragments counter */
+	if (left_is_free && right_is_free)
+		e4b->bd_info->bb_fragments--;
+	else if (!left_is_free && !right_is_free)
+		e4b->bd_info->bb_fragments++;
 
-			if (order > 0) {
-				/* for special purposes, we don't set
-				 * free bits in bitmap */
-				mb_set_bit(block, buddy);
-				mb_set_bit(block + 1, buddy);
-			}
-			e4b->bd_info->bb_counters[order]--;
-			e4b->bd_info->bb_counters[order]--;
+	/* buddy[0] == bd_bitmap is a special case, so handle
+	 * it right away and let mb_buddy_mark_free stay free of
+	 * zero order checks.
+	 * Check if neighbours are to be coaleasced,
+	 * adjust bitmap bb_counters and borders appropriately.
+	 */
+	if (first & 1) {
+		first += !left_is_free;
+		e4b->bd_info->bb_counters[0] += left_is_free ? -1 : 1;
+	}
+	if (!(last & 1)) {
+		last -= !right_is_free;
+		e4b->bd_info->bb_counters[0] += right_is_free ? -1 : 1;
+	}
 
-			block = block >> 1;
-			order++;
-			e4b->bd_info->bb_counters[order]++;
+	if (first <= last)
+		mb_buddy_mark_free(e4b, first >> 1, last >> 1);
 
-			mb_clear_bit(block, buddy2);
-			buddy = buddy2;
-		} while (1);
-	}
+done:
 	mb_set_largest_free_order(sb, e4b->bd_info);
 	mb_check_buddy(e4b);
 }