sched: Unify load_balance{,_newidle}()
authorPeter Zijlstra <a.p.zijlstra@chello.nl>
Wed, 23 Dec 2009 14:10:31 +0000 (15:10 +0100)
committerIngo Molnar <mingo@elte.hu>
Thu, 21 Jan 2010 12:40:13 +0000 (13:40 +0100)
load_balance() and load_balance_newidle() look remarkably similar, one
key point they differ in is the condition on when to active balance.

So split out that logic into a separate function.

One side effect is that previously load_balance_newidle() used to fail
and return -1 under these conditions, whereas now it doesn't. I've not
yet fully figured out the whole -1 return case for either
load_balance{,_newidle}().

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

index 65d0820..1040832 100644 (file)
@@ -2816,6 +2816,39 @@ find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle,
 /* Working cpumask for load_balance and load_balance_newidle. */
 static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask);
 
+static int need_active_balance(struct sched_domain *sd, int sd_idle, int idle)
+{
+       if (idle == CPU_NEWLY_IDLE) {
+               /*
+                * The only task running in a non-idle cpu can be moved to this
+                * cpu in an attempt to completely freeup the other CPU
+                * package.
+                *
+                * The package power saving logic comes from
+                * find_busiest_group(). If there are no imbalance, then
+                * f_b_g() will return NULL. However when sched_mc={1,2} then
+                * f_b_g() will select a group from which a running task may be
+                * pulled to this cpu in order to make the other package idle.
+                * If there is no opportunity to make a package idle and if
+                * there are no imbalance, then f_b_g() will return NULL and no
+                * action will be taken in load_balance_newidle().
+                *
+                * Under normal task pull operation due to imbalance, there
+                * will be more than one task in the source run queue and
+                * move_tasks() will succeed.  ld_moved will be true and this
+                * active balance code will not be triggered.
+                */
+               if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
+                   !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
+                       return 0;
+
+               if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP)
+                       return 0;
+       }
+
+       return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
+}
+
 /*
  * Check this_cpu to ensure it is balanced within domain. Attempt to move
  * tasks if there is an imbalance.
@@ -2902,8 +2935,7 @@ redo:
                schedstat_inc(sd, lb_failed[idle]);
                sd->nr_balance_failed++;
 
-               if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) {
-
+               if (need_active_balance(sd, sd_idle, idle)) {
                        raw_spin_lock_irqsave(&busiest->lock, flags);
 
                        /* don't kick the migration_thread, if the curr
@@ -3049,66 +3081,37 @@ redo:
                int active_balance = 0;
 
                schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]);
-               if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
-                   !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
-                       return -1;
-
-               if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP)
-                       return -1;
+               sd->nr_balance_failed++;
 
-               if (sd->nr_balance_failed++ < 2)
-                       return -1;
+               if (need_active_balance(sd, sd_idle, CPU_NEWLY_IDLE)) {
+                       double_lock_balance(this_rq, busiest);
 
-               /*
-                * The only task running in a non-idle cpu can be moved to this
-                * cpu in an attempt to completely freeup the other CPU
-                * package. The same method used to move task in load_balance()
-                * have been extended for load_balance_newidle() to speedup
-                * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2)
-                *
-                * The package power saving logic comes from
-                * find_busiest_group().  If there are no imbalance, then
-                * f_b_g() will return NULL.  However when sched_mc={1,2} then
-                * f_b_g() will select a group from which a running task may be
-                * pulled to this cpu in order to make the other package idle.
-                * If there is no opportunity to make a package idle and if
-                * there are no imbalance, then f_b_g() will return NULL and no
-                * action will be taken in load_balance_newidle().
-                *
-                * Under normal task pull operation due to imbalance, there
-                * will be more than one task in the source run queue and
-                * move_tasks() will succeed.  ld_moved will be true and this
-                * active balance code will not be triggered.
-                */
+                       /*
+                        * don't kick the migration_thread, if the curr
+                        * task on busiest cpu can't be moved to this_cpu
+                        */
+                       if (!cpumask_test_cpu(this_cpu,
+                                             &busiest->curr->cpus_allowed)) {
+                               double_unlock_balance(this_rq, busiest);
+                               all_pinned = 1;
+                               return ld_moved;
+                       }
 
-               /* Lock busiest in correct order while this_rq is held */
-               double_lock_balance(this_rq, busiest);
+                       if (!busiest->active_balance) {
+                               busiest->active_balance = 1;
+                               busiest->push_cpu = this_cpu;
+                               active_balance = 1;
+                       }
 
-               /*
-                * don't kick the migration_thread, if the curr
-                * task on busiest cpu can't be moved to this_cpu
-                */
-               if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) {
                        double_unlock_balance(this_rq, busiest);
-                       all_pinned = 1;
-                       return ld_moved;
-               }
-
-               if (!busiest->active_balance) {
-                       busiest->active_balance = 1;
-                       busiest->push_cpu = this_cpu;
-                       active_balance = 1;
+                       /*
+                        * Should not call ttwu while holding a rq->lock
+                        */
+                       raw_spin_unlock(&this_rq->lock);
+                       if (active_balance)
+                               wake_up_process(busiest->migration_thread);
+                       raw_spin_lock(&this_rq->lock);
                }
-
-               double_unlock_balance(this_rq, busiest);
-               /*
-                * Should not call ttwu while holding a rq->lock
-                */
-               raw_spin_unlock(&this_rq->lock);
-               if (active_balance)
-                       wake_up_process(busiest->migration_thread);
-               raw_spin_lock(&this_rq->lock);
-
        } else
                sd->nr_balance_failed = 0;