Skip to contentSkip to conversation

Subject : PATCH [001:001]: raid0 devices size should be aligned to chunk size

PATCH raz ben yehuda 2 participants 2 messages Jun 1, 2009 – Jun 4, 2009 MD RAID

lore
Share & tools
lore
#0 lore raw

raid devices should be aligned to a chunk size (when this feature will be enabled) because zones will become un-aligned to chunks and we will traverse between zones in a single IO access.

 raid0.c |   22 +++++++++++++++++++++-
 1 file changed, 21 insertions(+), 1 deletion(-)
Signed-off-by [email protected]

---

drivers/md/raid0.c +21 -1
index 0fcd9b5..987a926 100644
@@ -302,6 +302,26 @@ static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
 	return array_sectors;
 }
 
+/*
+ *	Align devices to a chunk size, else zones will not be aligned
+*/
+static void align_devices_to_chunk_size(struct list_head *disks,
+					int chunk_sects)
+{
+	sector_t sectors;
+	mdk_rdev_t *rdev;
+
+	list_for_each_entry(rdev, disks, same_set) {
+		sectors = rdev->sectors;
+		sector_div(sectors, chunk_sects);
+		sectors = sectors * chunk_sects;
+		printk(KERN_INFO "raid0: reduce device size: %lld --> %lld\n",
+				(unsigned long long)rdev->sectors,
+				(unsigned long long)sectors);
+		rdev->sectors = sectors;
+	}
+}
+
 static int raid0_run(mddev_t *mddev)
 {
 	int ret;
@@ -313,7 +333,7 @@ static int raid0_run(mddev_t *mddev)
 	}
 	blk_queue_max_sectors(mddev->queue, mddev->chunk_sectors);
 	mddev->queue->queue_lock = &mddev->queue->__queue_lock;
-
+	align_devices_to_chunk_size(&mddev->disks, mddev->chunk_sectors);
 	ret = create_strip_zones(mddev);
 	if (ret < 0)
 		return ret;
#1 lore raw
Quoted 4 lines · On Monday June 1, [email protected] wrote:

raid devices should be aligned to a chunk size (when this feature will be enabled) because zones will become un-aligned to chunks and we will traverse between zones in a single IO access.

I'm surprised that you need this as the common code in md is supposed to round everything th chunk_size...
However I found a bug that it was still assuming power-of-2 for v1.x metadata.
So I fixed that.
Then I decide that it is best to do the rounding in the personalities
- most of them do already.
So I added rounding code to linear and raid0

http://neil.brown.name/git?p=md;a=commitdiff;h=c9aaceea6015213e571761c0f0899614863c4868

and then removed all the related code from md.c

It is all in my for-next.

thanks,
NeilBrown

Quoted 46 lines
 raid0.c |   22 +++++++++++++++++++++-
 1 file changed, 21 insertions(+), 1 deletion(-)
Signed-off-by [email protected]

---

drivers/md/raid0.c +21 -1
index 0fcd9b5..987a926 100644
@@ -302,6 +302,26 @@ static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
 	return array_sectors;
 }
 
+/*
+ *	Align devices to a chunk size, else zones will not be aligned
+*/
+static void align_devices_to_chunk_size(struct list_head *disks,
+					int chunk_sects)
+{
+	sector_t sectors;
+	mdk_rdev_t *rdev;
+
+	list_for_each_entry(rdev, disks, same_set) {
+		sectors = rdev->sectors;
+		sector_div(sectors, chunk_sects);
+		sectors = sectors * chunk_sects;
+		printk(KERN_INFO "raid0: reduce device size: %lld --> %lld\n",
+				(unsigned long long)rdev->sectors,
+				(unsigned long long)sectors);
+		rdev->sectors = sectors;
+	}
+}
+
 static int raid0_run(mddev_t *mddev)
 {
 	int ret;
@@ -313,7 +333,7 @@ static int raid0_run(mddev_t *mddev)
 	}
 	blk_queue_max_sectors(mddev->queue, mddev->chunk_sectors);
 	mddev->queue->queue_lock = &mddev->queue->__queue_lock;
-
+	align_devices_to_chunk_size(&mddev->disks, mddev->chunk_sectors);
 	ret = create_strip_zones(mddev);
 	if (ret < 0)
 		return ret;