[PATCH] cfq-iosched: Kill O(N) runtime of cfq_resort_rr_list()
authorJens Axboe <axboe@suse.de>
Fri, 28 Jul 2006 07:48:51 +0000 (09:48 +0200)
committerJens Axboe <axboe@nelson.home.kernel.dk>
Sat, 30 Sep 2006 18:29:39 +0000 (20:29 +0200)
Currently it scales with number of processes in that priority group,
which is potentially not very nice as it's called quite often.
Basically we always need to do tail inserts, except for the case of a
new process. So just mark/detect a queue as such.

Signed-off-by: Jens Axboe <axboe@suse.de>
block/cfq-iosched.c

index 0452108..c6e649f 100644 (file)
@@ -152,7 +152,6 @@ struct cfq_queue {
        unsigned long slice_start;
        unsigned long slice_end;
        unsigned long slice_left;
-       unsigned long service_last;
 
        /* number of requests that are on the dispatch list */
        int on_dispatch[2];
@@ -174,6 +173,7 @@ enum cfqq_state_flags {
        CFQ_CFQQ_FLAG_fifo_expire,
        CFQ_CFQQ_FLAG_idle_window,
        CFQ_CFQQ_FLAG_prio_changed,
+       CFQ_CFQQ_FLAG_queue_new,
 };
 
 #define CFQ_CFQQ_FNS(name)                                             \
@@ -198,6 +198,7 @@ CFQ_CFQQ_FNS(must_dispatch);
 CFQ_CFQQ_FNS(fifo_expire);
 CFQ_CFQQ_FNS(idle_window);
 CFQ_CFQQ_FNS(prio_changed);
+CFQ_CFQQ_FNS(queue_new);
 #undef CFQ_CFQQ_FNS
 
 static struct cfq_queue *cfq_find_cfq_hash(struct cfq_data *, unsigned int, unsigned short);
@@ -350,7 +351,7 @@ cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted)
 {
        struct cfq_data *cfqd = cfqq->cfqd;
-       struct list_head *list, *entry;
+       struct list_head *list;
 
        BUG_ON(!cfq_cfqq_on_rr(cfqq));
 
@@ -375,31 +376,26 @@ static void cfq_resort_rr_list(struct cfq_queue *cfqq, int preempted)
        }
 
        /*
-        * if queue was preempted, just add to front to be fair. busy_rr
-        * isn't sorted, but insert at the back for fairness.
+        * If this queue was preempted or is new (never been serviced), let
+        * it be added first for fairness but beind other new queues.
+        * Otherwise, just add to the back  of the list.
         */
-       if (preempted || list == &cfqd->busy_rr) {
-               if (preempted)
-                       list = list->prev;
+       if (preempted || cfq_cfqq_queue_new(cfqq)) {
+               struct list_head *n = list;
+               struct cfq_queue *__cfqq;
 
-               list_add_tail(&cfqq->cfq_list, list);
-               return;
-       }
+               while (n->next != list) {
+                       __cfqq = list_entry_cfqq(n->next);
+                       if (!cfq_cfqq_queue_new(__cfqq))
+                               break;
 
-       /*
-        * sort by when queue was last serviced
-        */
-       entry = list;
-       while ((entry = entry->prev) != list) {
-               struct cfq_queue *__cfqq = list_entry_cfqq(entry);
+                       n = n->next;
+               }
 
-               if (!__cfqq->service_last)
-                       break;
-               if (time_before(__cfqq->service_last, cfqq->service_last))
-                       break;
+               list = n;
        }
 
-       list_add(&cfqq->cfq_list, entry);
+       list_add_tail(&cfqq->cfq_list, list);
 }
 
 /*
@@ -591,13 +587,12 @@ __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq,
        if (cfq_cfqq_wait_request(cfqq))
                del_timer(&cfqd->idle_slice_timer);
 
-       if (!preempted && !cfq_cfqq_dispatched(cfqq)) {
-               cfqq->service_last = now;
+       if (!preempted && !cfq_cfqq_dispatched(cfqq))
                cfq_schedule_dispatch(cfqd);
-       }
 
        cfq_clear_cfqq_must_dispatch(cfqq);
        cfq_clear_cfqq_wait_request(cfqq);
+       cfq_clear_cfqq_queue_new(cfqq);
 
        /*
         * store what was left of this slice, if the queue idled out
@@ -1297,13 +1292,13 @@ retry:
                hlist_add_head(&cfqq->cfq_hash, &cfqd->cfq_hash[hashval]);
                atomic_set(&cfqq->ref, 0);
                cfqq->cfqd = cfqd;
-               cfqq->service_last = 0;
                /*
                 * set ->slice_left to allow preemption for a new process
                 */
                cfqq->slice_left = 2 * cfqd->cfq_slice_idle;
                cfq_mark_cfqq_idle_window(cfqq);
                cfq_mark_cfqq_prio_changed(cfqq);
+               cfq_mark_cfqq_queue_new(cfqq);
                cfq_init_prio_data(cfqq);
        }
 
@@ -1672,12 +1667,8 @@ static void cfq_completed_request(request_queue_t *q, struct request *rq)
        if (!cfq_class_idle(cfqq))
                cfqd->last_end_request = now;
 
-       if (!cfq_cfqq_dispatched(cfqq)) {
-               if (cfq_cfqq_on_rr(cfqq)) {
-                       cfqq->service_last = now;
-                       cfq_resort_rr_list(cfqq, 0);
-               }
-       }
+       if (!cfq_cfqq_dispatched(cfqq) && cfq_cfqq_on_rr(cfqq))
+               cfq_resort_rr_list(cfqq, 0);
 
        if (sync)
                RQ_CIC(rq)->last_end_request = now;