blk-mq: avoid excessive boot delays with large lun counts
authorJeff Moyer <jmoyer@redhat.com>
Tue, 3 Nov 2015 15:40:06 +0000 (10:40 -0500)
committerJens Axboe <axboe@fb.com>
Tue, 3 Nov 2015 15:42:19 +0000 (08:42 -0700)
commit2404e607a9ee36db361bebe32787dafa1f7d6c00
tree38f0de79314cdf010e331897684f7097a28dbe51
parentcdea01b2bf98affb7e9c44530108a4a28535eee8
blk-mq: avoid excessive boot delays with large lun counts

Hi,

Zhangqing Luo reported long boot times on a system with thousands of
LUNs when scsi-mq was enabled.  He narrowed the problem down to
blk_mq_add_queue_tag_set, where every queue is frozen in order to set
the BLK_MQ_F_TAG_SHARED flag.  Each added device will freeze all queues
added before it in sequence, which involves waiting for an RCU grace
period for each one.  We don't need to do this.  After the second queue
is added, only new queues need to be initialized with the shared tag.
We can do that by percolating the flag up to the blk_mq_tag_set, and
updating the newly added queue's hctxs if the flag is set.

This problem was introduced by commit 0d2602ca30e41 (blk-mq: improve
support for shared tags maps).

Reported-and-tested-by: Jason Luo <zhangqing.luo@oracle.com>
Reviewed-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
block/blk-mq.c