On Wed, May 10, 2017 at 12:57:01PM -0700, Shaohua Li wrote:

quoted

---
 drivers/md/raid1.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

v2:
  updated the title not to include internal defect number :)

diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c
index a1f3fbe..e7ab3bb 100644
--- a/drivers/md/raid1.c
+++ b/drivers/md/raid1.c

@@ -628,8 +628,11 @@ static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sect
 					break;
 			}
 			continue;
-		} else
+		} else {
+			if ((sectors > best_good_sectors) && (best_disk >= 0))
+				best_disk = disk;

best_disk = -1 seems better here to me. If this is the only disk, we will
choose in below 'if (best_disk == -1)' check. If there are more disks which
cover the whole range, we will choose the best disk with minimum seek/pending.

Agreed, it's a neater way to fix it. I'll send the patch.

Tomek