move raid0 chunk size to 4K*n granularity.
motivation for this patch is to have a better access to raid550. if a raid 5 is 3M
stripe (4-1),and you have two of these raids 5's, and on top of you have a raid0,
it is better to access raid550 with a 3MB buffers and not 1M ( no raid5 write penalty).
Andre, Patch is applied on top of your last post. now it is your turn to merge :)
md.c | 24 ++++++++++-----
raid0.c | 102 ++++++++++++++++++++++++++++++----------------------------------
2 files changed, 65 insertions(+), 61 deletions(-)
Signed-Off-by:raziebe@xxxxxxxxx
diff --git a/drivers/md/md.c b/drivers/md/md.c
index ed5727c..5eab782 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -440,12 +440,14 @@ static inline sector_t calc_dev_sboffset(struct block_device *bdev)
return MD_NEW_SIZE_SECTORS(num_sectors);
}
+
static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
{
sector_t num_sectors = rdev->sb_start;
-
- if (chunk_size)
- num_sectors &= ~((sector_t)chunk_size/512 - 1);
+ if (chunk_size) {
+ int chunk_sects = chunk_size>>9;
+ num_sectors = (num_sectors/chunk_sects)*chunk_sects;
+ }
return num_sectors;
}
@@ -3512,7 +3514,7 @@ min_sync_store(mddev_t *mddev, const char *buf, size_t len)
/* Must be a multiple of chunk_size */
if (mddev->chunk_size) {
- if (min & (sector_t)((mddev->chunk_size>>9)-1))
+ if (min % (sector_t)(mddev->chunk_size>>9))
return -EINVAL;
}
mddev->resync_min = min;
@@ -3549,7 +3551,7 @@ max_sync_store(mddev_t *mddev, const char *buf, size_t len)
/* Must be a multiple of chunk_size */
if (mddev->chunk_size) {
- if (max & (sector_t)((mddev->chunk_size>>9)-1))
+ if (max % (sector_t)((mddev->chunk_size>>9)))
return -EINVAL;
}
mddev->resync_max = max;
@@ -3993,11 +3995,19 @@ static int do_md_run(mddev_t * mddev)
/*
* chunk-size has to be a power of 2
*/
- if ( (1 << ffz(~chunk_size)) != chunk_size) {
+ if ((1 << ffz(~chunk_size)) != chunk_size &&
+ mddev->level != 0) {
printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
return -EINVAL;
}
-
+ /*
+ * raid0 chunk size has to divide by a page
+ */
+ if (mddev->level == 0 && (chunk_size % 4096)) {
+ printk(KERN_ERR "chunk_size of %d not valid\n",
+ chunk_size);
+ return -EINVAL;
+ }
/* devices must have minimum size of one chunk */
list_for_each_entry(rdev, &mddev->disks, same_set) {
if (test_bit(Faulty, &rdev->flags))
diff --git a/drivers/md/raid0.c b/drivers/md/raid0.c
index 36b747a..9865316 100644
--- a/drivers/md/raid0.c
+++ b/drivers/md/raid0.c
@@ -53,7 +53,7 @@ static int raid0_congested(void *data, int bits)
}
-static int create_strip_zones (mddev_t *mddev)
+static int raid0_create_strip_zones(mddev_t *mddev)
{
int i, c, j;
sector_t current_start, curr_zone_start;
@@ -237,7 +237,7 @@ static int raid0_mergeable_bvec(struct request_queue *q,
unsigned int chunk_sectors = mddev->chunk_size >> 9;
unsigned int bio_sectors = bvm->bi_size >> 9;
- max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
+ max = (chunk_sectors - ((sector % chunk_sectors) + bio_sectors)) << 9;
if (max < 0) max = 0; /* bio_add cannot handle a negative return */
if (max <= biovec->bv_len && bio_sectors == 0)
return biovec->bv_len;
@@ -259,26 +259,37 @@ static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
return array_sectors;
}
+static int raid0_is_power2_chunk(mddev_t *mddev)
+{
+ if ((1 << ffz(~mddev->chunk_size)) == mddev->chunk_size)
+ return 1;
+ return 0;
+}
+
+
static int raid0_run(mddev_t *mddev)
{
int ret;
+ int segment_boundary = (mddev->chunk_size>>1)-1;
if (mddev->chunk_size == 0) {
printk(KERN_ERR "md/raid0: non-zero chunk size required.\n");
return -EINVAL;
}
- printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
- mdname(mddev),
- mddev->chunk_size >> 9,
- (mddev->chunk_size>>1)-1);
blk_queue_max_sectors(mddev->queue, mddev->chunk_size >> 9);
- blk_queue_segment_boundary(mddev->queue, (mddev->chunk_size>>1) - 1);
+ if (!raid0_is_power2_chunk(mddev))
+ segment_boundary = ~(ffz(~mddev->chunk_size))>>1;
+ printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
+ mdname(mddev),
+ mddev->chunk_size >> 9,
+ segment_boundary);
+ blk_queue_segment_boundary(mddev->queue, segment_boundary);
mddev->queue->queue_lock = &mddev->queue->__queue_lock;
mddev->private = kmalloc(sizeof(raid0_conf_t), GFP_KERNEL);
if (!mddev->private)
return -ENOMEM;
- ret = create_strip_zones(mddev);
+ ret = raid0_create_strip_zones(mddev);
if (ret < 0) {
kfree(mddev->private);
mddev->private = NULL;
@@ -322,31 +333,35 @@ static int raid0_stop (mddev_t *mddev)
return 0;
}
-/* Find the zone which holds a particular offset */
-static struct strip_zone *find_zone(struct raid0_private_data *conf,
- sector_t sector)
+static int raid0_position_bio(mddev_t *mddev, struct bio *bio, sector_t sector)
{
- int i;
-
- for (i = 0; i < conf->nr_strip_zones; i++) {
- struct strip_zone *z = conf->strip_zone + i;
-
- if (sector < z->zone_start + z->sectors)
- return z;
- }
- BUG();
- return NULL;
+ sector_t sect_in_chunk;
+ mdk_rdev_t *tmp_dev;
+ sector_t chunk_in_dev;
+ sector_t rsect;
+ sector_t x;
+ raid0_conf_t *conf = mddev_to_conf(mddev);
+ sector_t chunk_sects = mddev->chunk_size >> 9;
+ struct strip_zone *zone = &conf->strip_zone[0];
+
+ while (sector >= zone->zone_start + zone->sectors)
+ zone++;
+ sect_in_chunk = sector % chunk_sects;
+ x = (sector - zone->zone_start) / chunk_sects;
+ sector_div(x, zone->nb_dev);
+ chunk_in_dev = x;
+ x = sector / chunk_sects;
+ tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
+ rsect = (chunk_in_dev * chunk_sects) + zone->dev_start + sect_in_chunk;
+ bio->bi_bdev = tmp_dev->bdev;
+ bio->bi_sector = rsect + tmp_dev->data_offset;
+ return 0;
}
-static int raid0_make_request (struct request_queue *q, struct bio *bio)
+static int raid0_make_request(struct request_queue *q, struct bio *bio)
{
mddev_t *mddev = q->queuedata;
- unsigned int sect_in_chunk, chunksect_bits, chunk_sects;
- raid0_conf_t *conf = mddev_to_conf(mddev);
- struct strip_zone *zone;
- mdk_rdev_t *tmp_dev;
- sector_t chunk;
- sector_t sector, rsect;
+ unsigned int chunk_sects;
const int rw = bio_data_dir(bio);
int cpu;
@@ -362,10 +377,9 @@ static int raid0_make_request (struct request_queue *q, struct bio *bio)
part_stat_unlock();
chunk_sects = mddev->chunk_size >> 9;
- chunksect_bits = ffz(~chunk_sects);
- sector = bio->bi_sector;
- if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
+ if (unlikely(chunk_sects < ((bio->bi_sector % chunk_sects)
+ + (bio->bi_size >> 9)))) {
struct bio_pair *bp;
/* Sanity check -- queue functions should prevent this happening */
if (bio->bi_vcnt != 1 ||
@@ -374,7 +388,8 @@ static int raid0_make_request (struct request_queue *q, struct bio *bio)
/* This is a one page bio that upper layers
* refuse to split for us, so we need to split it.
*/
- bp = bio_split(bio, chunk_sects - (bio->bi_sector & (chunk_sects - 1)));
+ bp = bio_split(bio, chunk_sects -
+ (bio->bi_sector % chunk_sects));
if (raid0_make_request(q, &bp->bio1))
generic_make_request(&bp->bio1);
if (raid0_make_request(q, &bp->bio2))
@@ -383,29 +398,8 @@ static int raid0_make_request (struct request_queue *q, struct bio *bio)
bio_pair_release(bp);
return 0;
}
- zone = find_zone(conf, sector);
- if (!zone)
+ if (!raid0_position_bio(mddev, bio, bio->bi_sector))
return 1;
- sect_in_chunk = bio->bi_sector & (chunk_sects - 1);
- {
- sector_t x = (sector - zone->zone_start) >> chunksect_bits;
-
- sector_div(x, zone->nb_dev);
- chunk = x;
-
- x = sector >> chunksect_bits;
- tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
- }
- rsect = (chunk << chunksect_bits) + zone->dev_start + sect_in_chunk;
-
- bio->bi_bdev = tmp_dev->bdev;
- bio->bi_sector = rsect + tmp_dev->data_offset;
-
- /*
- * Let the main block layer submit the IO and resolve recursion:
- */
- return 1;
-
bad_map:
printk("raid0_make_request bug: can't convert block across chunks"
" or bigger than %dk %llu %d\n", chunk_sects / 2,
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
