]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/commitdiff
sched: add some comments to the bandwidth code
authorPeter Zijlstra <a.p.zijlstra@chello.nl>
Tue, 23 Sep 2008 13:33:43 +0000 (15:33 +0200)
committerIngo Molnar <mingo@elte.hu>
Tue, 23 Sep 2008 14:23:16 +0000 (16:23 +0200)
Hopefully clarify some of this code a little.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched_rt.c

index 2e228bd5395e36cd49101411bca1e796fbd8fef1..d570a8cc4fcd268658966f665b643e65db0b508e 100644 (file)
@@ -231,6 +231,9 @@ static inline struct rt_bandwidth *sched_rt_bandwidth(struct rt_rq *rt_rq)
 #endif /* CONFIG_RT_GROUP_SCHED */
 
 #ifdef CONFIG_SMP
+/*
+ * We ran out of runtime, see if we can borrow some from our neighbours.
+ */
 static int do_balance_runtime(struct rt_rq *rt_rq)
 {
        struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
@@ -250,9 +253,18 @@ static int do_balance_runtime(struct rt_rq *rt_rq)
                        continue;
 
                spin_lock(&iter->rt_runtime_lock);
+               /*
+                * Either all rqs have inf runtime and there's nothing to steal
+                * or __disable_runtime() below sets a specific rq to inf to
+                * indicate its been disabled and disalow stealing.
+                */
                if (iter->rt_runtime == RUNTIME_INF)
                        goto next;
 
+               /*
+                * From runqueues with spare time, take 1/n part of their
+                * spare time, but no more than our period.
+                */
                diff = iter->rt_runtime - iter->rt_time;
                if (diff > 0) {
                        diff = div_u64((u64)diff, weight);
@@ -274,6 +286,9 @@ next:
        return more;
 }
 
+/*
+ * Ensure this RQ takes back all the runtime it lend to its neighbours.
+ */
 static void __disable_runtime(struct rq *rq)
 {
        struct root_domain *rd = rq->rd;
@@ -289,17 +304,33 @@ static void __disable_runtime(struct rq *rq)
 
                spin_lock(&rt_b->rt_runtime_lock);
                spin_lock(&rt_rq->rt_runtime_lock);
+               /*
+                * Either we're all inf and nobody needs to borrow, or we're
+                * already disabled and thus have nothing to do, or we have
+                * exactly the right amount of runtime to take out.
+                */
                if (rt_rq->rt_runtime == RUNTIME_INF ||
                                rt_rq->rt_runtime == rt_b->rt_runtime)
                        goto balanced;
                spin_unlock(&rt_rq->rt_runtime_lock);
 
+               /*
+                * Calculate the difference between what we started out with
+                * and what we current have, that's the amount of runtime
+                * we lend and now have to reclaim.
+                */
                want = rt_b->rt_runtime - rt_rq->rt_runtime;
 
+               /*
+                * Greedy reclaim, take back as much as we can.
+                */
                for_each_cpu_mask(i, rd->span) {
                        struct rt_rq *iter = sched_rt_period_rt_rq(rt_b, i);
                        s64 diff;
 
+                       /*
+                        * Can't reclaim from ourselves or disabled runqueues.
+                        */
                        if (iter == rt_rq || iter->rt_runtime == RUNTIME_INF)
                                continue;
 
@@ -319,8 +350,16 @@ static void __disable_runtime(struct rq *rq)
                }
 
                spin_lock(&rt_rq->rt_runtime_lock);
+               /*
+                * We cannot be left wanting - that would mean some runtime
+                * leaked out of the system.
+                */
                BUG_ON(want);
 balanced:
+               /*
+                * Disable all the borrow logic by pretending we have inf
+                * runtime - in which case borrowing doesn't make sense.
+                */
                rt_rq->rt_runtime = RUNTIME_INF;
                spin_unlock(&rt_rq->rt_runtime_lock);
                spin_unlock(&rt_b->rt_runtime_lock);
@@ -343,6 +382,9 @@ static void __enable_runtime(struct rq *rq)
        if (unlikely(!scheduler_running))
                return;
 
+       /*
+        * Reset each runqueue's bandwidth settings
+        */
        for_each_leaf_rt_rq(rt_rq, rq) {
                struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);