sched/deadline: Add bandwidth management for SCHED_DEADLINE tasks

In order of deadline scheduling to be effective and useful, it is
important that some method of having the allocation of the available
CPU bandwidth to tasks and task groups under control.
This is usually called "admission control" and if it is not performed
at all, no guarantee can be given on the actual scheduling of the
-deadline tasks.

Since when RT-throttling has been introduced each task group have a
bandwidth associated to itself, calculated as a certain amount of
runtime over a period. Moreover, to make it possible to manipulate
such bandwidth, readable/writable controls have been added to both
procfs (for system wide settings) and cgroupfs (for per-group
settings).

Therefore, the same interface is being used for controlling the
bandwidth distrubution to -deadline tasks and task groups, i.e.,
new controls but with similar names, equivalent meaning and with
the same usage paradigm are added.

However, more discussion is needed in order to figure out how
we want to manage SCHED_DEADLINE bandwidth at the task group level.
Therefore, this patch adds a less sophisticated, but actually
very sensible, mechanism to ensure that a certain utilization
cap is not overcome per each root_domain (the single rq for !SMP
configurations).

Another main difference between deadline bandwidth management and
RT-throttling is that -deadline tasks have bandwidth on their own
(while -rt ones doesn't!), and thus we don't need an higher level
throttling mechanism to enforce the desired bandwidth.

This patch, therefore:

 - adds system wide deadline bandwidth management by means of:
    * /proc/sys/kernel/sched_dl_runtime_us,
    * /proc/sys/kernel/sched_dl_period_us,
   that determine (i.e., runtime / period) the total bandwidth
   available on each CPU of each root_domain for -deadline tasks;

 - couples the RT and deadline bandwidth management, i.e., enforces
   that the sum of how much bandwidth is being devoted to -rt
   -deadline tasks to stay below 100%.

This means that, for a root_domain comprising M CPUs, -deadline tasks
can be created until the sum of their bandwidths stay below:

    M * (sched_dl_runtime_us / sched_dl_period_us)

It is also possible to disable this bandwidth management logic, and
be thus free of oversubscribing the system up to any arbitrary level.
Signed-off-by: Dario Faggioli <raistlin@linux.it>
Signed-off-by: Juri Lelli <juri.lelli@gmail.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1383831828-15501-12-git-send-email-juri.lelli@gmail.comSigned-off-by: Ingo Molnar <mingo@kernel.org>

include/linux/sched.h

@@ -1104,6 +1104,7 @@ struct sched_dl_entity {
 	u64 dl_runtime;		/* maximum runtime for each instance	*/
 	u64 dl_deadline;	/* relative deadline of each instance	*/
 	u64 dl_period;		/* separation of two instances (period) */
+	u64 dl_bw;		/* dl_runtime / dl_deadline		*/
 
 	/*
 	 * Actual scheduling parameters. Initialized with the values above,

include/linux/sched/sysctl.h

@@ -81,6 +81,15 @@ static inline unsigned int get_sysctl_timer_migration(void)
 extern unsigned int sysctl_sched_rt_period;
 extern int sysctl_sched_rt_runtime;
 
+/*
+ *  control SCHED_DEADLINE reservations:
+ *
+ *  /proc/sys/kernel/sched_dl_period_us
+ *  /proc/sys/kernel/sched_dl_runtime_us
+ */
+extern unsigned int sysctl_sched_dl_period;
+extern int sysctl_sched_dl_runtime;
+
 #ifdef CONFIG_CFS_BANDWIDTH
 extern unsigned int sysctl_sched_cfs_bandwidth_slice;
 #endif
@@ -99,4 +108,8 @@ extern int sched_rt_handler(struct ctl_table *table, int write,
 		void __user *buffer, size_t *lenp,
 		loff_t *ppos);
 
+int sched_dl_handler(struct ctl_table *table, int write,
+		void __user *buffer, size_t *lenp,
+		loff_t *ppos);
+
 #endif /* _SCHED_SYSCTL_H */

kernel/sched/deadline.c

@@ -16,6 +16,8 @@
  */
 #include "sched.h"
 
+struct dl_bandwidth def_dl_bandwidth;
+
 static inline struct task_struct *dl_task_of(struct sched_dl_entity *dl_se)
 {
 	return container_of(dl_se, struct task_struct, dl);
@@ -46,6 +48,27 @@ static inline int is_leftmost(struct task_struct *p, struct dl_rq *dl_rq)
 	return dl_rq->rb_leftmost == &dl_se->rb_node;
 }
 
+void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime)
+{
+	raw_spin_lock_init(&dl_b->dl_runtime_lock);
+	dl_b->dl_period = period;
+	dl_b->dl_runtime = runtime;
+}
+
+extern unsigned long to_ratio(u64 period, u64 runtime);
+
+void init_dl_bw(struct dl_bw *dl_b)
+{
+	raw_spin_lock_init(&dl_b->lock);
+	raw_spin_lock(&def_dl_bandwidth.dl_runtime_lock);
+	if (global_dl_runtime() == RUNTIME_INF)
+		dl_b->bw = -1;
+	else
+		dl_b->bw = to_ratio(global_dl_period(), global_dl_runtime());
+	raw_spin_unlock(&def_dl_bandwidth.dl_runtime_lock);
+	dl_b->total_bw = 0;
+}
+
 void init_dl_rq(struct dl_rq *dl_rq, struct rq *rq)
 {
 	dl_rq->rb_root = RB_ROOT;
@@ -57,6 +80,8 @@ void init_dl_rq(struct dl_rq *dl_rq, struct rq *rq)
 	dl_rq->dl_nr_migratory = 0;
 	dl_rq->overloaded = 0;
 	dl_rq->pushable_dl_tasks_root = RB_ROOT;
+#else
+	init_dl_bw(&dl_rq->dl_bw);
 #endif
 }
 
@@ -359,8 +384,9 @@ static bool dl_entity_overflow(struct sched_dl_entity *dl_se,
 	 * of anything below microseconds resolution is actually fiction
 	 * (but still we want to give the user that illusion >;).
 	 */
-	left = (pi_se->dl_period >> 10) * (dl_se->runtime >> 10);
-	right = ((dl_se->deadline - t) >> 10) * (pi_se->dl_runtime >> 10);
+	left = (pi_se->dl_period >> DL_SCALE) * (dl_se->runtime >> DL_SCALE);
+	right = ((dl_se->deadline - t) >> DL_SCALE) *
+		(pi_se->dl_runtime >> DL_SCALE);
 
 	return dl_time_before(right, left);
 }
@@ -911,8 +937,8 @@ static void check_preempt_curr_dl(struct rq *rq, struct task_struct *p,
 	 * In the unlikely case current and p have the same deadline
 	 * let us try to decide what's the best thing to do...
 	 */
-	if ((s64)(p->dl.deadline - rq->curr->dl.deadline) == 0 &&
-	    !need_resched())
+	if ((p->dl.deadline == rq->curr->dl.deadline) &&
+	    !test_tsk_need_resched(rq->curr))
 		check_preempt_equal_dl(rq, p);
 #endif /* CONFIG_SMP */
 }
@@ -1000,6 +1026,14 @@ static void task_fork_dl(struct task_struct *p)
 static void task_dead_dl(struct task_struct *p)
 {
 	struct hrtimer *timer = &p->dl.dl_timer;
+	struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
+
+	/*
+	 * Since we are TASK_DEAD we won't slip out of the domain!
+	 */
+	raw_spin_lock_irq(&dl_b->lock);
+	dl_b->total_bw -= p->dl.dl_bw;
+	raw_spin_unlock_irq(&dl_b->lock);
 
 	hrtimer_cancel(timer);
 }
@@ -1226,7 +1260,7 @@ static struct task_struct *pick_next_pushable_dl_task(struct rq *rq)
 	BUG_ON(task_current(rq, p));
 	BUG_ON(p->nr_cpus_allowed <= 1);
 
-	BUG_ON(!p->se.on_rq);
+	BUG_ON(!p->on_rq);
 	BUG_ON(!dl_task(p));
 
 	return p;
@@ -1373,7 +1407,7 @@ static int pull_dl_task(struct rq *this_rq)
 		     dl_time_before(p->dl.deadline,
 				    this_rq->dl.earliest_dl.curr))) {
 			WARN_ON(p == src_rq->curr);
-			WARN_ON(!p->se.on_rq);
+			WARN_ON(!p->on_rq);
 
 			/*
 			 * Then we pull iff p has actually an earlier

kernel/sched/sched.h

@@ -73,6 +73,13 @@ extern void update_cpu_load_active(struct rq *this_rq);
 #define NICE_0_LOAD		SCHED_LOAD_SCALE
 #define NICE_0_SHIFT		SCHED_LOAD_SHIFT
 
+/*
+ * Single value that decides SCHED_DEADLINE internal math precision.
+ * 10 -> just above 1us
+ * 9  -> just above 0.5us
+ */
+#define DL_SCALE (10)
+
 /*
  * These are the 'tuning knobs' of the scheduler:
  */
@@ -107,7 +114,7 @@ static inline int task_has_dl_policy(struct task_struct *p)
 	return dl_policy(p->policy);
 }
 
-static inline int dl_time_before(u64 a, u64 b)
+static inline bool dl_time_before(u64 a, u64 b)
 {
 	return (s64)(a - b) < 0;
 }
@@ -115,8 +122,8 @@ static inline int dl_time_before(u64 a, u64 b)
 /*
  * Tells if entity @a should preempt entity @b.
  */
-static inline
-int dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
+static inline bool
+dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b)
 {
 	return dl_time_before(a->deadline, b->deadline);
 }
@@ -136,6 +143,50 @@ struct rt_bandwidth {
 	u64			rt_runtime;
 	struct hrtimer		rt_period_timer;
 };
+/*
+ * To keep the bandwidth of -deadline tasks and groups under control
+ * we need some place where:
+ *  - store the maximum -deadline bandwidth of the system (the group);
+ *  - cache the fraction of that bandwidth that is currently allocated.
+ *
+ * This is all done in the data structure below. It is similar to the
+ * one used for RT-throttling (rt_bandwidth), with the main difference
+ * that, since here we are only interested in admission control, we
+ * do not decrease any runtime while the group "executes", neither we
+ * need a timer to replenish it.
+ *
+ * With respect to SMP, the bandwidth is given on a per-CPU basis,
+ * meaning that:
+ *  - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU;
+ *  - dl_total_bw array contains, in the i-eth element, the currently
+ *    allocated bandwidth on the i-eth CPU.
+ * Moreover, groups consume bandwidth on each CPU, while tasks only
+ * consume bandwidth on the CPU they're running on.
+ * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw
+ * that will be shown the next time the proc or cgroup controls will
+ * be red. It on its turn can be changed by writing on its own
+ * control.
+ */
+struct dl_bandwidth {
+	raw_spinlock_t dl_runtime_lock;
+	u64 dl_runtime;
+	u64 dl_period;
+};
+
+static inline int dl_bandwidth_enabled(void)
+{
+	return sysctl_sched_dl_runtime >= 0;
+}
+
+extern struct dl_bw *dl_bw_of(int i);
+
+struct dl_bw {
+	raw_spinlock_t lock;
+	u64 bw, total_bw;
+};
+
+static inline u64 global_dl_period(void);
+static inline u64 global_dl_runtime(void);
 
 extern struct mutex sched_domains_mutex;
 
@@ -423,6 +474,8 @@ struct dl_rq {
 	 */
 	struct rb_root pushable_dl_tasks_root;
 	struct rb_node *pushable_dl_tasks_leftmost;
+#else
+	struct dl_bw dl_bw;
 #endif
 };
 
@@ -449,6 +502,7 @@ struct root_domain {
 	 */
 	cpumask_var_t dlo_mask;
 	atomic_t dlo_count;
+	struct dl_bw dl_bw;
 
 	/*
 	 * The "RT overload" flag: it gets set if a CPU has more than
@@ -897,7 +951,18 @@ static inline u64 global_rt_runtime(void)
 	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
 }
 
+static inline u64 global_dl_period(void)
+{
+	return (u64)sysctl_sched_dl_period * NSEC_PER_USEC;
+}
+
+static inline u64 global_dl_runtime(void)
+{
+	if (sysctl_sched_dl_runtime < 0)
+		return RUNTIME_INF;
 
+	return (u64)sysctl_sched_dl_runtime * NSEC_PER_USEC;
+}
 
 static inline int task_current(struct rq *rq, struct task_struct *p)
 {
@@ -1145,6 +1210,7 @@ extern void update_max_interval(void);
 extern void init_sched_dl_class(void);
 extern void init_sched_rt_class(void);
 extern void init_sched_fair_class(void);
+extern void init_sched_dl_class(void);
 
 extern void resched_task(struct task_struct *p);
 extern void resched_cpu(int cpu);
@@ -1152,8 +1218,12 @@ extern void resched_cpu(int cpu);
 extern struct rt_bandwidth def_rt_bandwidth;
 extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime);
 
+extern struct dl_bandwidth def_dl_bandwidth;
+extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime);
 extern void init_dl_task_timer(struct sched_dl_entity *dl_se);
 
+unsigned long to_ratio(u64 period, u64 runtime);
+
 extern void update_idle_cpu_load(struct rq *this_rq);
 
 extern void init_task_runnable_average(struct task_struct *p);

kernel/sysctl.c

@@ -414,6 +414,20 @@ static struct ctl_table kern_table[] = {
 		.mode		= 0644,
 		.proc_handler	= sched_rr_handler,
 	},
+	{
+		.procname	= "sched_dl_period_us",
+		.data		= &sysctl_sched_dl_period,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= sched_dl_handler,
+	},
+	{
+		.procname	= "sched_dl_runtime_us",
+		.data		= &sysctl_sched_dl_runtime,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= sched_dl_handler,
+	},
 #ifdef CONFIG_SCHED_AUTOGROUP
 	{
 		.procname	= "sched_autogroup_enabled",