Better explanation of jump-to-CFS optimization removal

GSN-EDF and friends rely on being called even if there is currently
no runnable real-time task on the runqueue for (at least) two reasons:
1) To initiate migrations. LITMUS^RT pull tasks for migrations; this requires
     plugins to be called even if no task is currently present.
2) To maintain invariants when jobs block.

1 files changed, 10 insertions, 7 deletions

diff --git a/kernel/sched.c b/kernel/sched.c
index 9ad41979c0b2..1701eaebb79c 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -487,6 +487,7 @@ struct rt_rq {
 
 /* Litmus related fields in a runqueue */
 struct litmus_rq {
+        unsigned long nr_running;
         struct task_struct *prev;
 };
 
@@ -5420,13 +5421,15 @@ pick_next_task(struct rq *rq)
         /*
          * Optimization: we know that if all tasks are in
          * the fair class we can call that function directly:
-         */
-        /*
-         * LITMUS_TODO: can we move processes out of fair class?
-         * i.e., create a litmus_rq
-         */
-        /* Don't do this for LITMUS
-        if (likely(rq->nr_running == rq->cfs.nr_running)) {
+
+         * NOT IN LITMUS^RT!
+
+         * This breaks many assumptions in the plugins.
+         * Do not uncomment without thinking long and hard
+         * about how this affects global plugins such as GSN-EDF.
+
+        if (rq->nr_running == rq->cfs.nr_running) {
+                TRACE("taking shortcut in pick_next_task()\n");
                 p = fair_sched_class.pick_next_task(rq);
                 if (likely(p))
                         return p;