*/
 unsigned int __read_mostly sysctl_sched_shares_window = 10000000UL;
 
+#ifdef CONFIG_CFS_BANDWIDTH
+/*
+ * Amount of runtime to allocate from global (tg) to local (per-cfs_rq) pool
+ * each time a cfs_rq requests quota.
+ *
+ * Note: in the case that the slice exceeds the runtime remaining (either due
+ * to consumption or the quota being specified to be smaller than the slice)
+ * we will always only issue the remaining available time.
+ *
+ * default: 5 msec, units: microseconds
+  */
+unsigned int sysctl_sched_cfs_bandwidth_slice = 5000UL;
+#endif
+
 static const struct sched_class fair_sched_class;
 
 /**************************************************************
 
 #endif /* CONFIG_FAIR_GROUP_SCHED */
 
+static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq,
+                                  unsigned long delta_exec);
 
 /**************************************************************
  * Scheduling class tree data structure manipulation methods:
                cpuacct_charge(curtask, delta_exec);
                account_group_exec_runtime(curtask, delta_exec);
        }
+
+       account_cfs_rq_runtime(cfs_rq, delta_exec);
 }
 
 static inline void
 {
        return 100000000ULL;
 }
+
+static inline u64 sched_cfs_bandwidth_slice(void)
+{
+       return (u64)sysctl_sched_cfs_bandwidth_slice * NSEC_PER_USEC;
+}
+
+static void assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
+{
+       struct task_group *tg = cfs_rq->tg;
+       struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
+       u64 amount = 0, min_amount;
+
+       /* note: this is a positive sum as runtime_remaining <= 0 */
+       min_amount = sched_cfs_bandwidth_slice() - cfs_rq->runtime_remaining;
+
+       raw_spin_lock(&cfs_b->lock);
+       if (cfs_b->quota == RUNTIME_INF)
+               amount = min_amount;
+       else if (cfs_b->runtime > 0) {
+               amount = min(cfs_b->runtime, min_amount);
+               cfs_b->runtime -= amount;
+       }
+       raw_spin_unlock(&cfs_b->lock);
+
+       cfs_rq->runtime_remaining += amount;
+}
+
+static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq,
+                                    unsigned long delta_exec)
+{
+       if (!cfs_rq->runtime_enabled)
+               return;
+
+       cfs_rq->runtime_remaining -= delta_exec;
+       if (cfs_rq->runtime_remaining > 0)
+               return;
+
+       assign_cfs_rq_runtime(cfs_rq);
+}
+
+static __always_inline void account_cfs_rq_runtime(struct cfs_rq *cfs_rq,
+                                                  unsigned long delta_exec)
+{
+       if (!cfs_rq->runtime_enabled)
+               return;
+
+       __account_cfs_rq_runtime(cfs_rq, delta_exec);
+}
+
+#else
+static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq,
+                                    unsigned long delta_exec) {}
 #endif
 
 /**************************************************
 {
        struct sched_entity *se = &rq->curr->se;
 
-       for_each_sched_entity(se)
-               set_next_entity(cfs_rq_of(se), se);
+       for_each_sched_entity(se) {
+               struct cfs_rq *cfs_rq = cfs_rq_of(se);
+
+               set_next_entity(cfs_rq, se);
+               /* ensure bandwidth has been allocated on our new cfs_rq */
+               account_cfs_rq_runtime(cfs_rq, 0);
+       }
 }
 
 #ifdef CONFIG_FAIR_GROUP_SCHED