sched/fair: Unify switched_{from,to}_fair() and task_move_group_fair()
authorByungchul Park <byungchul.park@lge.com>
Thu, 20 Aug 2015 11:22:00 +0000 (20:22 +0900)
committerIngo Molnar <mingo@kernel.org>
Sun, 13 Sep 2015 07:52:49 +0000 (09:52 +0200)
By observing that switched_from_fair() detaches from a runqueue, and
switched_to_fair() attaches to a runqueue, we can see that
task_move_group_fair() is one followed by the other with flipping the
runqueue in between.

Therefore extract all the common bits and implement all three
functions in terms of them.

This should fix a few corner cases wrt. vruntime normalization; where,
when we take a task off of a runqueue we convert to an approximation
of lag by subtracting min_vruntime, and when placing a task on the a
runqueue to the reverse.

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Byungchul Park <byungchul.park@lge.com>
[peterz: Changelog]
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: yuyang.du@intel.com
Link: http://lkml.kernel.org/r/1440069720-27038-6-git-send-email-byungchul.park@lge.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/sched/fair.c

index 5cd7054aac85fc465861c6b7242e082f1397fd63..b96d8dd65331b5cfee5b404deaf6966347884187 100644 (file)
@@ -7924,21 +7924,39 @@ prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
                check_preempt_curr(rq, p, 0);
 }
 
-static void switched_from_fair(struct rq *rq, struct task_struct *p)
+static inline bool vruntime_normalized(struct task_struct *p)
 {
        struct sched_entity *se = &p->se;
-       struct cfs_rq *cfs_rq = cfs_rq_of(se);
 
        /*
-        * Ensure the task's vruntime is normalized, so that when it's
-        * switched back to the fair class the enqueue_entity(.flags=0) will
-        * do the right thing.
+        * In both the TASK_ON_RQ_QUEUED and TASK_ON_RQ_MIGRATING cases,
+        * the dequeue_entity(.flags=0) will already have normalized the
+        * vruntime.
+        */
+       if (p->on_rq)
+               return true;
+
+       /*
+        * When !on_rq, vruntime of the task has usually NOT been normalized.
+        * But there are some cases where it has already been normalized:
         *
-        * If it's queued, then the dequeue_entity(.flags=0) will already
-        * have normalized the vruntime, if it's !queued, then only when
-        * the task is sleeping will it still have non-normalized vruntime.
+        * - A forked child which is waiting for being woken up by
+        *   wake_up_new_task().
+        * - A task which has been woken up by try_to_wake_up() and
+        *   waiting for actually being woken up by sched_ttwu_pending().
         */
-       if (!task_on_rq_queued(p) && p->state != TASK_RUNNING) {
+       if (!se->sum_exec_runtime || p->state == TASK_WAKING)
+               return true;
+
+       return false;
+}
+
+static void detach_task_cfs_rq(struct task_struct *p)
+{
+       struct sched_entity *se = &p->se;
+       struct cfs_rq *cfs_rq = cfs_rq_of(se);
+
+       if (!vruntime_normalized(p)) {
                /*
                 * Fix up our vruntime so that the current sleep doesn't
                 * cause 'unlimited' sleep bonus.
@@ -7951,9 +7969,10 @@ static void switched_from_fair(struct rq *rq, struct task_struct *p)
        detach_entity_load_avg(cfs_rq, se);
 }
 
-static void switched_to_fair(struct rq *rq, struct task_struct *p)
+static void attach_task_cfs_rq(struct task_struct *p)
 {
        struct sched_entity *se = &p->se;
+       struct cfs_rq *cfs_rq = cfs_rq_of(se);
 
 #ifdef CONFIG_FAIR_GROUP_SCHED
        /*
@@ -7964,33 +7983,32 @@ static void switched_to_fair(struct rq *rq, struct task_struct *p)
 #endif
 
        /* Synchronize task with its cfs_rq */
-       attach_entity_load_avg(cfs_rq_of(&p->se), &p->se);
+       attach_entity_load_avg(cfs_rq, se);
+
+       if (!vruntime_normalized(p))
+               se->vruntime += cfs_rq->min_vruntime;
+}
 
-       if (!task_on_rq_queued(p)) {
+static void switched_from_fair(struct rq *rq, struct task_struct *p)
+{
+       detach_task_cfs_rq(p);
+}
+
+static void switched_to_fair(struct rq *rq, struct task_struct *p)
+{
+       attach_task_cfs_rq(p);
 
+       if (task_on_rq_queued(p)) {
                /*
-                * Ensure the task has a non-normalized vruntime when it is switched
-                * back to the fair class with !queued, so that enqueue_entity() at
-                * wake-up time will do the right thing.
-                *
-                * If it's queued, then the enqueue_entity(.flags=0) makes the task
-                * has non-normalized vruntime, if it's !queued, then it still has
-                * normalized vruntime.
+                * We were most likely switched from sched_rt, so
+                * kick off the schedule if running, otherwise just see
+                * if we can still preempt the current task.
                 */
-               if (p->state != TASK_RUNNING)
-                       se->vruntime += cfs_rq_of(se)->min_vruntime;
-               return;
+               if (rq->curr == p)
+                       resched_curr(rq);
+               else
+                       check_preempt_curr(rq, p, 0);
        }
-
-       /*
-        * We were most likely switched from sched_rt, so
-        * kick off the schedule if running, otherwise just see
-        * if we can still preempt the current task.
-        */
-       if (rq->curr == p)
-               resched_curr(rq);
-       else
-               check_preempt_curr(rq, p, 0);
 }
 
 /* Account for a task changing its policy or group.
@@ -8027,57 +8045,14 @@ void init_cfs_rq(struct cfs_rq *cfs_rq)
 #ifdef CONFIG_FAIR_GROUP_SCHED
 static void task_move_group_fair(struct task_struct *p, int queued)
 {
-       struct sched_entity *se = &p->se;
-       struct cfs_rq *cfs_rq;
-
-       /*
-        * If the task was not on the rq at the time of this cgroup movement
-        * it must have been asleep, sleeping tasks keep their ->vruntime
-        * absolute on their old rq until wakeup (needed for the fair sleeper
-        * bonus in place_entity()).
-        *
-        * If it was on the rq, we've just 'preempted' it, which does convert
-        * ->vruntime to a relative base.
-        *
-        * Make sure both cases convert their relative position when migrating
-        * to another cgroup's rq. This does somewhat interfere with the
-        * fair sleeper stuff for the first placement, but who cares.
-        */
-       /*
-        * When !queued, vruntime of the task has usually NOT been normalized.
-        * But there are some cases where it has already been normalized:
-        *
-        * - Moving a forked child which is waiting for being woken up by
-        *   wake_up_new_task().
-        * - Moving a task which has been woken up by try_to_wake_up() and
-        *   waiting for actually being woken up by sched_ttwu_pending().
-        *
-        * To prevent boost or penalty in the new cfs_rq caused by delta
-        * min_vruntime between the two cfs_rqs, we skip vruntime adjustment.
-        */
-       if (!queued && (!se->sum_exec_runtime || p->state == TASK_WAKING))
-               queued = 1;
-
-       cfs_rq = cfs_rq_of(se);
-       if (!queued)
-               se->vruntime -= cfs_rq->min_vruntime;
-
-       /* Synchronize task with its prev cfs_rq */
-       detach_entity_load_avg(cfs_rq, se);
+       detach_task_cfs_rq(p);
        set_task_rq(p, task_cpu(p));
 
 #ifdef CONFIG_SMP
        /* Tell se's cfs_rq has been changed -- migrated */
        p->se.avg.last_update_time = 0;
 #endif
-
-       se->depth = se->parent ? se->parent->depth + 1 : 0;
-       cfs_rq = cfs_rq_of(se);
-       if (!queued)
-               se->vruntime += cfs_rq->min_vruntime;
-
-       /* Virtually synchronize task with its new cfs_rq */
-       attach_entity_load_avg(cfs_rq, se);
+       attach_task_cfs_rq(p);
 }
 
 void free_fair_sched_group(struct task_group *tg)
This page took 0.036493 seconds and 5 git commands to generate.