Re: Boot failure on Arndale with next-20131105

[Jens Axboe]

On 11/05/2013 09:42 AM, Tomasz Figa wrote:
> Hi,
> 
> On Tuesday 05 of November 2013 17:19:00 Tushar Behera wrote:
>> Hi,
>>
>> We are having a boot-time kernel panic on Samsung's Exynos5250-based
>> Arndale board with next-20131105. Bisect points to following commit.
>>
>> <<<
>> commit febca1baea1cfe2d7a0271385d89b03d5fb34f94
>> Author: Chris Mason <chris.mason@xxxxxxxxxxxx>
>> Date:   Thu Oct 31 13:32:42 2013 -0600
>>
>>     block: setup bi_vcnt on clones
>>
>>     commit 9fc6286f347d changed the cloning code to make clones cheaper for
>>     the case where we don't need to clone the iovec array.  But,
>>     the new clone needs the bi_vnct from the original.
>>
>>     Signed-off-by: Chris Mason <chris.mason@xxxxxxxxxxxx>
>>     Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
>>>>>
>>
>> Reverting above commit, Arndale is able to boot again.
> 
> I can confirm exactly the same behavior on Exynos 4210-based Trats board,
> with exactly the same bisection results.

Can either (or both) of you try this?

-- 
Jens Axboe

diff --git a/drivers/md/dm.c b/drivers/md/dm.c
index 8e6174c..a1177e1 100644
--- a/drivers/md/dm.c
+++ b/drivers/md/dm.c
@@ -1123,8 +1123,13 @@ struct clone_info {
 
 static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len)
 {
-	bio->bi_iter.bi_sector = sector;
-	bio->bi_iter.bi_size = to_bytes(len);
+	if (len) {
+		bio->bi_iter.bi_sector = sector;
+		bio->bi_iter.bi_size = to_bytes(len);
+	} else {
+		bio->bi_iter.bi_size = 0;
+		bio->bi_vcnt = 0;
+	}
 }
 
 /*
@@ -1178,8 +1183,7 @@ static void __clone_and_map_simple_bio(struct clone_info *ci,
 	 * and discard, so no need for concern about wasted bvec allocations.
 	 */
 	 __bio_clone(clone, ci->bio);
-	if (len)
-		bio_setup_sector(clone, ci->sector, len);
+	bio_setup_sector(clone, ci->sector, len);
 
 	__map_bio(tio);
 }