block: optimizations in blk_rq_timed_out_timer()
malahal@us.ibm.com [Thu, 30 Oct 2008 07:51:58 +0000 (08:51 +0100)]
Now the rq->deadline can't be zero if the request is in the
timeout_list, so there is no need to have next_set. There is no need to
access a request's deadline field if blk_rq_timed_out is called on it.

Signed-off-by: Malahal Naineni <malahal@us.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

block/blk-timeout.c

index 69185ea..116bbf3 100644 (file)
@@ -111,7 +111,7 @@ static void blk_rq_timed_out(struct request *req)
 void blk_rq_timed_out_timer(unsigned long data)
 {
        struct request_queue *q = (struct request_queue *) data;
-       unsigned long flags, uninitialized_var(next), next_set = 0;
+       unsigned long flags, next = 0;
        struct request *rq, *tmp;
 
        spin_lock_irqsave(q->queue_lock, flags);
@@ -126,12 +126,10 @@ void blk_rq_timed_out_timer(unsigned long data)
                        if (blk_mark_rq_complete(rq))
                                continue;
                        blk_rq_timed_out(rq);
+               } else {
+                       if (!next || time_after(next, rq->deadline))
+                               next = rq->deadline;
                }
-               if (!next_set) {
-                       next = rq->deadline;
-                       next_set = 1;
-               } else if (time_after(next, rq->deadline))
-                       next = rq->deadline;
        }
 
        if (next_set && !list_empty(&q->timeout_list))