Skip to contentSkip to conversation

Change a argument of disk_map_sector_rcu() to hd_part->start_sect from req->__sector

PATCH In review Yasuaki Ishimatsu 1 participant 1 message Dec 6, 2010 Block layer

lore
Share & tools
lore

Yasuaki Ishimatsu

author
#0 lore raw

PATCH 1/2 Change a argument of disk_map_sector_rcu() to hd_part->start_sect from req->__sector

From: Yasuaki Ishimatsu <[email protected]>

PROBLEM:

The previous patch does not fix the problem completely, since the problem occurs on following case.

When sdX region is between sdX1 partition and sdX2 partition, a sdX region's I/O is merged sdX1 partition's I/O or sdX2 partition's I/O.

1. sdX region is between sdX1 partition and sdX2 partition

sdX1 partition sdX region sdX1 partition
|----------------|------------|----------------|

2. issue a sdX region's I/O and increment a in_flight of sdX

sdX1 partition sdX region sdX1 partition

  |----------------|------------|----------------|
                    <-   I/O  ->

3. issue a sdX1 partition's I/O

sdX1 partition sdX region sdX1 partition

  |----------------|------------|----------------|
       <-   I/O  -> <-   I/O  ->

4. merge a sdX1 partition's I/O and a sdX region's I/O, since the previous patch cannot care this case.

sdX1 partition sdX region sdX1 partition

  |----------------|------------|----------------|
       <-         I/O         ->

5. finish a I/O and decrement a in_flight of sdX1 even though it was not incremented

HOW TO FIX:

The patch fixes avobe case by changing a argument of disk_map_sector_rcu().

The problem also is caused by merging different partition's I/Os. But, it is not easy to fix, when I/O is sdX region. So the patch allows step 4, when I/O is only sdX region. But, it does not change a result of disk_map_sector_rcu() by using partition's start sector ot original I/O , even if the I/O is merged.

Signed-off-by Yasuaki Ishimatsu <[email protected]>
---
 block/blk-core.c  |   10 ++++++----
 block/blk-merge.c |    2 +-
 2 files changed, 7 insertions(+), 5 deletions(-)
block/blk-core.c +6 -4
Index: linux-2.6.37-rc3/block/blk-core.c
===================================================================
@@ -64,11 +64,13 @@ static void drive_stat_acct(struct reque
 		return;
 
 	cpu = part_stat_lock();
-	part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
 
-	if (!new_io)
+	if (!new_io) {
+		part = disk_map_sector_rcu(rq->rq_disk, blk_rq_part_sector(rq));
 		part_stat_inc(cpu, part, merges[rw]);
+	}
 	else {
+		part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
 		part_round_stats(cpu, part);
 		part_inc_in_flight(part, rw);
 		rq->__part_start_sect = part->start_sect;

@@ -1778,7 +1780,7 @@ static void blk_account_io_completion(st
 		int cpu;
 
 		cpu = part_stat_lock();
-		part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
+		part = disk_map_sector_rcu(req->rq_disk, blk_rq_part_sector(req));
 		part_stat_add(cpu, part, sectors[rw], bytes >> 9);
 		part_stat_unlock();
 	}

@@ -1798,7 +1800,7 @@ static void blk_account_io_done(struct r
 		int cpu;
 
 		cpu = part_stat_lock();
-		part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
+		part = disk_map_sector_rcu(req->rq_disk, blk_rq_part_sector(req));
 
 		part_stat_inc(cpu, part, ios[rw]);
 		part_stat_add(cpu, part, ticks[rw], duration);
block/blk-merge.c +1 -1
Index: linux-2.6.37-rc3/block/blk-merge.c
===================================================================
@@ -356,7 +356,7 @@ static void blk_account_io_merge(struct
 		int cpu;
 
 		cpu = part_stat_lock();
-		part = disk_map_sector_rcu(req->rq_disk, blk_rq_pos(req));
+		part = disk_map_sector_rcu(req->rq_disk, blk_rq_part_sector(req));
 
 		part_round_stats(cpu, part);
 		part_dec_in_flight(part, rq_data_dir(req));