1 files changed, 487 insertions, 130 deletions
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 39eb6011bc3..5d011ef4c0d 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -142,9 +142,8 @@ const_debug unsigned int sysctl_sched_features =
 #define SCHED_FEAT(name, enabled)	\
 	#name ,
 
-static __read_mostly char *sched_feat_names[] = {
+static const char * const sched_feat_names[] = {
 #include "features.h"
-	NULL
 };
 
 #undef SCHED_FEAT
@@ -1097,7 +1096,7 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
 	 * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks.
 	 *
 	 * sched_move_task() holds both and thus holding either pins the cgroup,
-	 * see set_task_rq().
+	 * see task_group().
 	 *
 	 * Furthermore, all task_rq users should acquire both locks, see
 	 * task_rq_lock().
@@ -2082,7 +2081,6 @@ context_switch(struct rq *rq, struct task_struct *prev,
 #endif
 
 	/* Here we just switch the register state and the stack. */
-	rcu_switch_from(prev);
 	switch_to(prev, next, prev);
 
 	barrier();
@@ -2162,11 +2160,73 @@ unsigned long this_cpu_load(void)
 }
 
 
+/*
+ * Global load-average calculations
+ *
+ * We take a distributed and async approach to calculating the global load-avg
+ * in order to minimize overhead.
+ *
+ * The global load average is an exponentially decaying average of nr_running +
+ * nr_uninterruptible.
+ *
+ * Once every LOAD_FREQ:
+ *
+ *   nr_active = 0;
+ *   for_each_possible_cpu(cpu)
+ *   	nr_active += cpu_of(cpu)->nr_running + cpu_of(cpu)->nr_uninterruptible;
+ *
+ *   avenrun[n] = avenrun[0] * exp_n + nr_active * (1 - exp_n)
+ *
+ * Due to a number of reasons the above turns in the mess below:
+ *
+ *  - for_each_possible_cpu() is prohibitively expensive on machines with
+ *    serious number of cpus, therefore we need to take a distributed approach
+ *    to calculating nr_active.
+ *
+ *        \Sum_i x_i(t) = \Sum_i x_i(t) - x_i(t_0) | x_i(t_0) := 0
+ *                      = \Sum_i { \Sum_j=1 x_i(t_j) - x_i(t_j-1) }
+ *
+ *    So assuming nr_active := 0 when we start out -- true per definition, we
+ *    can simply take per-cpu deltas and fold those into a global accumulate
+ *    to obtain the same result. See calc_load_fold_active().
+ *
+ *    Furthermore, in order to avoid synchronizing all per-cpu delta folding
+ *    across the machine, we assume 10 ticks is sufficient time for every
+ *    cpu to have completed this task.
+ *
+ *    This places an upper-bound on the IRQ-off latency of the machine. Then
+ *    again, being late doesn't loose the delta, just wrecks the sample.
+ *
+ *  - cpu_rq()->nr_uninterruptible isn't accurately tracked per-cpu because
+ *    this would add another cross-cpu cacheline miss and atomic operation
+ *    to the wakeup path. Instead we increment on whatever cpu the task ran
+ *    when it went into uninterruptible state and decrement on whatever cpu
+ *    did the wakeup. This means that only the sum of nr_uninterruptible over
+ *    all cpus yields the correct result.
+ *
+ *  This covers the NO_HZ=n code, for extra head-aches, see the comment below.
+ */
+
 /* Variables and functions for calc_load */
 static atomic_long_t calc_load_tasks;
 static unsigned long calc_load_update;
 unsigned long avenrun[3];
-EXPORT_SYMBOL(avenrun);
+EXPORT_SYMBOL(avenrun); /* should be removed */
+
+/**
+ * get_avenrun - get the load average array
+ * @loads:	pointer to dest load array
+ * @offset:	offset to add
+ * @shift:	shift count to shift the result left
+ *
+ * These values are estimates at best, so no need for locking.
+ */
+void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
+{
+	loads[0] = (avenrun[0] + offset) << shift;
+	loads[1] = (avenrun[1] + offset) << shift;
+	loads[2] = (avenrun[2] + offset) << shift;
+}
 
 static long calc_load_fold_active(struct rq *this_rq)
 {
@@ -2183,6 +2243,9 @@ static long calc_load_fold_active(struct rq *this_rq)
 	return delta;
 }
 
+/*
+ * a1 = a0 * e + a * (1 - e)
+ */
 static unsigned long
 calc_load(unsigned long load, unsigned long exp, unsigned long active)
 {
@@ -2194,30 +2257,118 @@ calc_load(unsigned long load, unsigned long exp, unsigned long active)
 
 #ifdef CONFIG_NO_HZ
 /*
- * For NO_HZ we delay the active fold to the next LOAD_FREQ update.
+ * Handle NO_HZ for the global load-average.
+ *
+ * Since the above described distributed algorithm to compute the global
+ * load-average relies on per-cpu sampling from the tick, it is affected by
+ * NO_HZ.
+ *
+ * The basic idea is to fold the nr_active delta into a global idle-delta upon
+ * entering NO_HZ state such that we can include this as an 'extra' cpu delta
+ * when we read the global state.
+ *
+ * Obviously reality has to ruin such a delightfully simple scheme:
+ *
+ *  - When we go NO_HZ idle during the window, we can negate our sample
+ *    contribution, causing under-accounting.
+ *
+ *    We avoid this by keeping two idle-delta counters and flipping them
+ *    when the window starts, thus separating old and new NO_HZ load.
+ *
+ *    The only trick is the slight shift in index flip for read vs write.
+ *
+ *        0s            5s            10s           15s
+ *          +10           +10           +10           +10
+ *        |-|-----------|-|-----------|-|-----------|-|
+ *    r:0 0 1           1 0           0 1           1 0
+ *    w:0 1 1           0 0           1 1           0 0
+ *
+ *    This ensures we'll fold the old idle contribution in this window while
+ *    accumlating the new one.
+ *
+ *  - When we wake up from NO_HZ idle during the window, we push up our
+ *    contribution, since we effectively move our sample point to a known
+ *    busy state.
+ *
+ *    This is solved by pushing the window forward, and thus skipping the
+ *    sample, for this cpu (effectively using the idle-delta for this cpu which
+ *    was in effect at the time the window opened). This also solves the issue
+ *    of having to deal with a cpu having been in NOHZ idle for multiple
+ *    LOAD_FREQ intervals.
  *
  * When making the ILB scale, we should try to pull this in as well.
  */
-static atomic_long_t calc_load_tasks_idle;
+static atomic_long_t calc_load_idle[2];
+static int calc_load_idx;
+
+static inline int calc_load_write_idx(void)
+{
+	int idx = calc_load_idx;
+
+	/*
+	 * See calc_global_nohz(), if we observe the new index, we also
+	 * need to observe the new update time.
+	 */
+	smp_rmb();
+
+	/*
+	 * If the folding window started, make sure we start writing in the
+	 * next idle-delta.
+	 */
+	if (!time_before(jiffies, calc_load_update))
+		idx++;
+
+	return idx & 1;
+}
+
+static inline int calc_load_read_idx(void)
+{
+	return calc_load_idx & 1;
+}
 
-void calc_load_account_idle(struct rq *this_rq)
+void calc_load_enter_idle(void)
 {
+	struct rq *this_rq = this_rq();
 	long delta;
 
+	/*
+	 * We're going into NOHZ mode, if there's any pending delta, fold it
+	 * into the pending idle delta.
+	 */
 	delta = calc_load_fold_active(this_rq);
-	if (delta)
-		atomic_long_add(delta, &calc_load_tasks_idle);
+	if (delta) {
+		int idx = calc_load_write_idx();
+		atomic_long_add(delta, &calc_load_idle[idx]);
+	}
 }
 
-static long calc_load_fold_idle(void)
+void calc_load_exit_idle(void)
 {
-	long delta = 0;
+	struct rq *this_rq = this_rq();
+
+	/*
+	 * If we're still before the sample window, we're done.
+	 */
+	if (time_before(jiffies, this_rq->calc_load_update))
+		return;
 
 	/*
-	 * Its got a race, we don't care...
+	 * We woke inside or after the sample window, this means we're already
+	 * accounted through the nohz accounting, so skip the entire deal and
+	 * sync up for the next window.
 	 */
-	if (atomic_long_read(&calc_load_tasks_idle))
-		delta = atomic_long_xchg(&calc_load_tasks_idle, 0);
+	this_rq->calc_load_update = calc_load_update;
+	if (time_before(jiffies, this_rq->calc_load_update + 10))
+		this_rq->calc_load_update += LOAD_FREQ;
+}
+
+static long calc_load_fold_idle(void)
+{
+	int idx = calc_load_read_idx();
+	long delta = 0;
+
+	if (atomic_long_read(&calc_load_idle[idx]))
+		delta = atomic_long_xchg(&calc_load_idle[idx], 0);
 
 	return delta;
 }
@@ -2303,66 +2454,39 @@ static void calc_global_nohz(void)
 {
 	long delta, active, n;
 
-	/*
-	 * If we crossed a calc_load_update boundary, make sure to fold
-	 * any pending idle changes, the respective CPUs might have
-	 * missed the tick driven calc_load_account_active() update
-	 * due to NO_HZ.
-	 */
-	delta = calc_load_fold_idle();
-	if (delta)
-		atomic_long_add(delta, &calc_load_tasks);
-
-	/*
-	 * It could be the one fold was all it took, we done!
-	 */
-	if (time_before(jiffies, calc_load_update + 10))
-		return;
-
-	/*
-	 * Catch-up, fold however many we are behind still
-	 */
-	delta = jiffies - calc_load_update - 10;
-	n = 1 + (delta / LOAD_FREQ);
+	if (!time_before(jiffies, calc_load_update + 10)) {
+		/*
+		 * Catch-up, fold however many we are behind still
+		 */
+		delta = jiffies - calc_load_update - 10;
+		n = 1 + (delta / LOAD_FREQ);
 
-	active = atomic_long_read(&calc_load_tasks);
-	active = active > 0 ? active * FIXED_1 : 0;
+		active = atomic_long_read(&calc_load_tasks);
+		active = active > 0 ? active * FIXED_1 : 0;
 
-	avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n);
-	avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n);
-	avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n);
+		avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n);
+		avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n);
+		avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n);
 
-	calc_load_update += n * LOAD_FREQ;
-}
-#else
-void calc_load_account_idle(struct rq *this_rq)
-{
-}
+		calc_load_update += n * LOAD_FREQ;
+	}
 
-static inline long calc_load_fold_idle(void)
-{
-	return 0;
+	/*
+	 * Flip the idle index...
+	 *
+	 * Make sure we first write the new time then flip the index, so that
+	 * calc_load_write_idx() will see the new time when it reads the new
+	 * index, this avoids a double flip messing things up.
+	 */
+	smp_wmb();
+	calc_load_idx++;
 }
+#else /* !CONFIG_NO_HZ */
 
-static void calc_global_nohz(void)
-{
-}
-#endif
+static inline long calc_load_fold_idle(void) { return 0; }
+static inline void calc_global_nohz(void) { }
 
-/**
- * get_avenrun - get the load average array
- * @loads:	pointer to dest load array
- * @offset:	offset to add
- * @shift:	shift count to shift the result left
- *
- * These values are estimates at best, so no need for locking.
- */
-void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
-{
-	loads[0] = (avenrun[0] + offset) << shift;
-	loads[1] = (avenrun[1] + offset) << shift;
-	loads[2] = (avenrun[2] + offset) << shift;
-}
+#endif /* CONFIG_NO_HZ */
 
 /*
  * calc_load - update the avenrun load estimates 10 ticks after the
@@ -2370,11 +2494,18 @@ void get_avenrun(unsigned long *loads, unsigned long offset, int shift)
  */
 void calc_global_load(unsigned long ticks)
 {
-	long active;
+	long active, delta;
 
 	if (time_before(jiffies, calc_load_update + 10))
 		return;
 
+	/*
+	 * Fold the 'old' idle-delta to include all NO_HZ cpus.
+	 */
+	delta = calc_load_fold_idle();
+	if (delta)
+		atomic_long_add(delta, &calc_load_tasks);
+
 	active = atomic_long_read(&calc_load_tasks);
 	active = active > 0 ? active * FIXED_1 : 0;
 
@@ -2385,12 +2516,7 @@ void calc_global_load(unsigned long ticks)
 	calc_load_update += LOAD_FREQ;
 
 	/*
-	 * Account one period with whatever state we found before
-	 * folding in the nohz state and ageing the entire idle period.
-	 *
-	 * This avoids loosing a sample when we go idle between 
-	 * calc_load_account_active() (10 ticks ago) and now and thus
-	 * under-accounting.
+	 * In case we idled for multiple LOAD_FREQ intervals, catch up in bulk.
 	 */
 	calc_global_nohz();
 }
@@ -2407,7 +2533,6 @@ static void calc_load_account_active(struct rq *this_rq)
 		return;
 
 	delta  = calc_load_fold_active(this_rq);
-	delta += calc_load_fold_idle();
 	if (delta)
 		atomic_long_add(delta, &calc_load_tasks);
 
@@ -2415,6 +2540,10 @@ static void calc_load_account_active(struct rq *this_rq)
 }
 
 /*
+ * End of global load-average stuff
+ */
+
+/*
  * The exact cpuload at various idx values, calculated at every tick would be
  * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load
  *
@@ -2517,25 +2646,32 @@ static void __update_cpu_load(struct rq *this_rq, unsigned long this_load,
 	sched_avg_update(this_rq);
 }
 
+#ifdef CONFIG_NO_HZ
+/*
+ * There is no sane way to deal with nohz on smp when using jiffies because the
+ * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading
+ * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}.
+ *
+ * Therefore we cannot use the delta approach from the regular tick since that
+ * would seriously skew the load calculation. However we'll make do for those
+ * updates happening while idle (nohz_idle_balance) or coming out of idle
+ * (tick_nohz_idle_exit).
+ *
+ * This means we might still be one tick off for nohz periods.
+ */
+
 /*
  * Called from nohz_idle_balance() to update the load ratings before doing the
  * idle balance.
  */
 void update_idle_cpu_load(struct rq *this_rq)
 {
-	unsigned long curr_jiffies = jiffies;
+	unsigned long curr_jiffies = ACCESS_ONCE(jiffies);
 	unsigned long load = this_rq->load.weight;
 	unsigned long pending_updates;
 
 	/*
-	 * Bloody broken means of dealing with nohz, but better than nothing..
-	 * jiffies is updated by one cpu, another cpu can drift wrt the jiffy
-	 * update and see 0 difference the one time and 2 the next, even though
-	 * we ticked at roughtly the same rate.
-	 *
-	 * Hence we only use this from nohz_idle_balance() and skip this
-	 * nonsense when called from the scheduler_tick() since that's
-	 * guaranteed a stable rate.
+	 * bail if there's load or we're actually up-to-date.
 	 */
 	if (load || curr_jiffies == this_rq->last_load_update_tick)
 		return;
@@ -2547,12 +2683,38 @@ void update_idle_cpu_load(struct rq *this_rq)
 }
 
 /*
+ * Called from tick_nohz_idle_exit() -- try and fix up the ticks we missed.
+ */
+void update_cpu_load_nohz(void)
+{
+	struct rq *this_rq = this_rq();
+	unsigned long curr_jiffies = ACCESS_ONCE(jiffies);
+	unsigned long pending_updates;
+
+	if (curr_jiffies == this_rq->last_load_update_tick)
+		return;
+
+	raw_spin_lock(&this_rq->lock);
+	pending_updates = curr_jiffies - this_rq->last_load_update_tick;
+	if (pending_updates) {
+		this_rq->last_load_update_tick = curr_jiffies;
+		/*
+		 * We were idle, this means load 0, the current load might be
+		 * !0 due to remote wakeups and the sort.
+		 */
+		__update_cpu_load(this_rq, 0, pending_updates);
+	}
+	raw_spin_unlock(&this_rq->lock);
+}
+#endif /* CONFIG_NO_HZ */
+
+/*
  * Called from scheduler_tick()
  */
 static void update_cpu_load_active(struct rq *this_rq)
 {
 	/*
-	 * See the mess in update_idle_cpu_load().
+	 * See the mess around update_idle_cpu_load() / update_cpu_load_nohz().
 	 */
 	this_rq->last_load_update_tick = jiffies;
 	__update_cpu_load(this_rq, this_rq->load.weight, 1);
@@ -4982,7 +5144,7 @@ void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
 		p->sched_class->set_cpus_allowed(p, new_mask);
 
 	cpumask_copy(&p->cpus_allowed, new_mask);
-	p->rt.nr_cpus_allowed = cpumask_weight(new_mask);
+	p->nr_cpus_allowed = cpumask_weight(new_mask);
 }
 
 /*
@@ -5524,15 +5686,20 @@ static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */
 
 #ifdef CONFIG_SCHED_DEBUG
 
-static __read_mostly int sched_domain_debug_enabled;
+static __read_mostly int sched_debug_enabled;
 
-static int __init sched_domain_debug_setup(char *str)
+static int __init sched_debug_setup(char *str)
 {
-	sched_domain_debug_enabled = 1;
+	sched_debug_enabled = 1;
 
 	return 0;
 }
-early_param("sched_debug", sched_domain_debug_setup);
+early_param("sched_debug", sched_debug_setup);
+
+static inline bool sched_debug(void)
+{
+	return sched_debug_enabled;
+}
 
 static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 				  struct cpumask *groupmask)
@@ -5572,7 +5739,12 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 			break;
 		}
 
-		if (!group->sgp->power) {
+		/*
+		 * Even though we initialize ->power to something semi-sane,
+		 * we leave power_orig unset. This allows us to detect if
+		 * domain iteration is still funny without causing /0 traps.
+		 */
+		if (!group->sgp->power_orig) {
 			printk(KERN_CONT "\n");
 			printk(KERN_ERR "ERROR: domain->cpu_power not "
 					"set\n");
@@ -5620,7 +5792,7 @@ static void sched_domain_debug(struct sched_domain *sd, int cpu)
 {
 	int level = 0;
 
-	if (!sched_domain_debug_enabled)
+	if (!sched_debug_enabled)
 		return;
 
 	if (!sd) {
@@ -5641,6 +5813,10 @@ static void sched_domain_debug(struct sched_domain *sd, int cpu)
 }
 #else /* !CONFIG_SCHED_DEBUG */
 # define sched_domain_debug(sd, cpu) do { } while (0)
+static inline bool sched_debug(void)
+{
+	return false;
+}
 #endif /* CONFIG_SCHED_DEBUG */
 
 static int sd_degenerate(struct sched_domain *sd)
@@ -5848,6 +6024,11 @@ static void destroy_sched_domains(struct sched_domain *sd, int cpu)
  * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this
  * allows us to avoid some pointer chasing select_idle_sibling().
  *
+ * Iterate domains and sched_groups downward, assigning CPUs to be
+ * select_idle_sibling() hw buddy.  Cross-wiring hw makes bouncing
+ * due to random perturbation self canceling, ie sw buddies pull
+ * their counterpart to their CPU's hw counterpart.
+ *
  * Also keep a unique ID per domain (we use the first cpu number in
  * the cpumask of the domain), this allows us to quickly tell if
  * two cpus are in the same cache domain, see cpus_share_cache().
@@ -5861,8 +6042,40 @@ static void update_top_cache_domain(int cpu)
 	int id = cpu;
 
 	sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES);
-	if (sd)
+	if (sd) {
+		struct sched_domain *tmp = sd;
+		struct sched_group *sg, *prev;
+		bool right;
+
+		/*
+		 * Traverse to first CPU in group, and count hops
+		 * to cpu from there, switching direction on each
+		 * hop, never ever pointing the last CPU rightward.
+		 */
+		do {
+			id = cpumask_first(sched_domain_span(tmp));
+			prev = sg = tmp->groups;
+			right = 1;
+
+			while (cpumask_first(sched_group_cpus(sg)) != id)
+				sg = sg->next;
+
+			while (!cpumask_test_cpu(cpu, sched_group_cpus(sg))) {
+				prev = sg;
+				sg = sg->next;
+				right = !right;
+			}
+
+			/* A CPU went down, never point back to domain start. */
+			if (right && cpumask_first(sched_group_cpus(sg->next)) == id)
+				right = false;
+
+			sg = right ? sg->next : prev;
+			tmp->idle_buddy = cpumask_first(sched_group_cpus(sg));
+		} while ((tmp = tmp->child));
+
 		id = cpumask_first(sched_domain_span(sd));
+	}
 
 	rcu_assign_pointer(per_cpu(sd_llc, cpu), sd);
 	per_cpu(sd_llc_id, cpu) = id;
@@ -5962,6 +6175,44 @@ struct sched_domain_topology_level {
 	struct sd_data      data;
 };
 
+/*
+ * Build an iteration mask that can exclude certain CPUs from the upwards
+ * domain traversal.
+ *
+ * Asymmetric node setups can result in situations where the domain tree is of
+ * unequal depth, make sure to skip domains that already cover the entire
+ * range.
+ *
+ * In that case build_sched_domains() will have terminated the iteration early
+ * and our sibling sd spans will be empty. Domains should always include the
+ * cpu they're built on, so check that.
+ *
+ */
+static void build_group_mask(struct sched_domain *sd, struct sched_group *sg)
+{
+	const struct cpumask *span = sched_domain_span(sd);
+	struct sd_data *sdd = sd->private;
+	struct sched_domain *sibling;
+	int i;
+
+	for_each_cpu(i, span) {
+		sibling = *per_cpu_ptr(sdd->sd, i);
+		if (!cpumask_test_cpu(i, sched_domain_span(sibling)))
+			continue;
+
+		cpumask_set_cpu(i, sched_group_mask(sg));
+	}
+}
+
+/*
+ * Return the canonical balance cpu for this group, this is the first cpu
+ * of this group that's also in the iteration mask.
+ */
+int group_balance_cpu(struct sched_group *sg)
+{
+	return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg));
+}
+
 static int
 build_overlap_sched_groups(struct sched_domain *sd, int cpu)
 {
@@ -5980,6 +6231,12 @@ build_overlap_sched_groups(struct sched_domain *sd, int cpu)
 		if (cpumask_test_cpu(i, covered))
 			continue;
 
+		child = *per_cpu_ptr(sdd->sd, i);
+
+		/* See the comment near build_group_mask(). */
+		if (!cpumask_test_cpu(i, sched_domain_span(child)))
+			continue;
+
 		sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
 				GFP_KERNEL, cpu_to_node(cpu));
 
@@ -5987,8 +6244,6 @@ build_overlap_sched_groups(struct sched_domain *sd, int cpu)
 			goto fail;
 
 		sg_span = sched_group_cpus(sg);
-
-		child = *per_cpu_ptr(sdd->sd, i);
 		if (child->child) {
 			child = child->child;
 			cpumask_copy(sg_span, sched_domain_span(child));
@@ -5997,10 +6252,24 @@ build_overlap_sched_groups(struct sched_domain *sd, int cpu)
 
 		cpumask_or(covered, covered, sg_span);
 
-		sg->sgp = *per_cpu_ptr(sdd->sgp, cpumask_first(sg_span));
-		atomic_inc(&sg->sgp->ref);
+		sg->sgp = *per_cpu_ptr(sdd->sgp, i);
+		if (atomic_inc_return(&sg->sgp->ref) == 1)
+			build_group_mask(sd, sg);
+
+		/*
+		 * Initialize sgp->power such that even if we mess up the
+		 * domains and no possible iteration will get us here, we won't
+		 * die on a /0 trap.
+		 */
+		sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span);
 
-		if (cpumask_test_cpu(cpu, sg_span))
+		/*
+		 * Make sure the first group of this domain contains the
+		 * canonical balance cpu. Otherwise the sched_domain iteration
+		 * breaks. See update_sg_lb_stats().
+		 */
+		if ((!groups && cpumask_test_cpu(cpu, sg_span)) ||
+		    group_balance_cpu(sg) == cpu)
 			groups = sg;
 
 		if (!first)
@@ -6074,6 +6343,7 @@ build_sched_groups(struct sched_domain *sd, int cpu)
 
 		cpumask_clear(sched_group_cpus(sg));
 		sg->sgp->power = 0;
+		cpumask_setall(sched_group_mask(sg));
 
 		for_each_cpu(j, span) {
 			if (get_group(j, sdd, NULL) != group)
@@ -6115,7 +6385,7 @@ static void init_sched_groups_power(int cpu, struct sched_domain *sd)
 		sg = sg->next;
 	} while (sg != sd->groups);
 
-	if (cpu != group_first_cpu(sg))
+	if (cpu != group_balance_cpu(sg))
 		return;
 
 	update_group_power(sd, cpu);
@@ -6165,11 +6435,8 @@ int sched_domain_level_max;
 
 static int __init setup_relax_domain_level(char *str)
 {
-	unsigned long val;
-
-	val = simple_strtoul(str, NULL, 0);
-	if (val < sched_domain_level_max)
-		default_relax_domain_level = val;
+	if (kstrtoint(str, 0, &default_relax_domain_level))
+		pr_warn("Unable to set relax_domain_level\n");
 
 	return 1;
 }
@@ -6279,14 +6546,13 @@ static struct sched_domain_topology_level *sched_domain_topology = default_topol
 #ifdef CONFIG_NUMA
 
 static int sched_domains_numa_levels;
-static int sched_domains_numa_scale;
 static int *sched_domains_numa_distance;
 static struct cpumask ***sched_domains_numa_masks;
 static int sched_domains_curr_level;
 
 static inline int sd_local_flags(int level)
 {
-	if (sched_domains_numa_distance[level] > REMOTE_DISTANCE)
+	if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE)
 		return 0;
 
 	return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE;
@@ -6344,6 +6610,42 @@ static const struct cpumask *sd_numa_mask(int cpu)
 	return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)];
 }
 
+static void sched_numa_warn(const char *str)
+{
+	static int done = false;
+	int i,j;
+
+	if (done)
+		return;
+
+	done = true;
+
+	printk(KERN_WARNING "ERROR: %s\n\n", str);
+
+	for (i = 0; i < nr_node_ids; i++) {
+		printk(KERN_WARNING "  ");
+		for (j = 0; j < nr_node_ids; j++)
+			printk(KERN_CONT "%02d ", node_distance(i,j));
+		printk(KERN_CONT "\n");
+	}
+	printk(KERN_WARNING "\n");
+}
+
+static bool find_numa_distance(int distance)
+{
+	int i;
+
+	if (distance == node_distance(0, 0))
+		return true;
+
+	for (i = 0; i < sched_domains_numa_levels; i++) {
+		if (sched_domains_numa_distance[i] == distance)
+			return true;
+	}
+
+	return false;
+}
+
 static void sched_init_numa(void)
 {
 	int next_distance, curr_distance = node_distance(0, 0);
@@ -6351,7 +6653,6 @@ static void sched_init_numa(void)
 	int level = 0;
 	int i, j, k;
 
-	sched_domains_numa_scale = curr_distance;
 	sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL);
 	if (!sched_domains_numa_distance)
 		return;
@@ -6362,23 +6663,41 @@ static void sched_init_numa(void)
 	 *
 	 * Assumes node_distance(0,j) includes all distances in
 	 * node_distance(i,j) in order to avoid cubic time.
-	 *
-	 * XXX: could be optimized to O(n log n) by using sort()
 	 */
 	next_distance = curr_distance;
 	for (i = 0; i < nr_node_ids; i++) {
 		for (j = 0; j < nr_node_ids; j++) {
-			int distance = node_distance(0, j);
-			if (distance > curr_distance &&
-					(distance < next_distance ||
-					 next_distance == curr_distance))
-				next_distance = distance;
+			for (k = 0; k < nr_node_ids; k++) {
+				int distance = node_distance(i, k);
+
+				if (distance > curr_distance &&
+				    (distance < next_distance ||
+				     next_distance == curr_distance))
+					next_distance = distance;
+
+				/*
+				 * While not a strong assumption it would be nice to know
+				 * about cases where if node A is connected to B, B is not
+				 * equally connected to A.
+				 */
+				if (sched_debug() && node_distance(k, i) != distance)
+					sched_numa_warn("Node-distance not symmetric");
+
+				if (sched_debug() && i && !find_numa_distance(distance))
+					sched_numa_warn("Node-0 not representative");
+			}
+			if (next_distance != curr_distance) {
+				sched_domains_numa_distance[level++] = next_distance;
+				sched_domains_numa_levels = level;
+				curr_distance = next_distance;
+			} else break;
 		}
-		if (next_distance != curr_distance) {
-			sched_domains_numa_distance[level++] = next_distance;
-			sched_domains_numa_levels = level;
-			curr_distance = next_distance;
-		} else break;
+
+		/*
+		 * In case of sched_debug() we verify the above assumption.
+		 */
+		if (!sched_debug())
+			break;
 	}
 	/*
 	 * 'level' contains the number of unique distances, excluding the
@@ -6403,7 +6722,7 @@ static void sched_init_numa(void)
 			return;
 
 		for (j = 0; j < nr_node_ids; j++) {
-			struct cpumask *mask = kzalloc_node(cpumask_size(), GFP_KERNEL, j);
+			struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL);
 			if (!mask)
 				return;
 
@@ -6490,7 +6809,7 @@ static int __sdt_alloc(const struct cpumask *cpu_map)
 
 			*per_cpu_ptr(sdd->sg, j) = sg;
 
-			sgp = kzalloc_node(sizeof(struct sched_group_power),
+			sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(),
 					GFP_KERNEL, cpu_to_node(j));
 			if (!sgp)
 				return -ENOMEM;
@@ -6543,7 +6862,6 @@ struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl,
 	if (!sd)
 		return child;
 
-	set_domain_attribute(sd, attr);
 	cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu));
 	if (child) {
 		sd->level = child->level + 1;
@@ -6551,6 +6869,7 @@ struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl,
 		child->parent = sd;
 	}
 	sd->child = child;
+	set_domain_attribute(sd, attr);
 
 	return sd;
 }
@@ -6691,7 +7010,6 @@ static int init_sched_domains(const struct cpumask *cpu_map)
 	if (!doms_cur)
 		doms_cur = &fallback_doms;
 	cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
-	dattr_cur = NULL;
 	err = build_sched_domains(doms_cur[0], NULL);
 	register_sched_domain_sysctl();
 
@@ -6816,34 +7134,66 @@ match2:
 	mutex_unlock(&sched_domains_mutex);
 }
 
+static int num_cpus_frozen;	/* used to mark begin/end of suspend/resume */
+
 /*
  * Update cpusets according to cpu_active mask.  If cpusets are
  * disabled, cpuset_update_active_cpus() becomes a simple wrapper
  * around partition_sched_domains().
+ *
+ * If we come here as part of a suspend/resume, don't touch cpusets because we
+ * want to restore it back to its original state upon resume anyway.
  */
 static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action,
 			     void *hcpu)
 {
-	switch (action & ~CPU_TASKS_FROZEN) {
+	switch (action) {
+	case CPU_ONLINE_FROZEN:
+	case CPU_DOWN_FAILED_FROZEN:
+
+		/*
+		 * num_cpus_frozen tracks how many CPUs are involved in suspend
+		 * resume sequence. As long as this is not the last online
+		 * operation in the resume sequence, just build a single sched
+		 * domain, ignoring cpusets.
+		 */
+		num_cpus_frozen--;
+		if (likely(num_cpus_frozen)) {
+			partition_sched_domains(1, NULL, NULL);
+			break;
+		}
+
+		/*
+		 * This is the last CPU online operation. So fall through and
+		 * restore the original sched domains by considering the
+		 * cpuset configurations.
+		 */
+
 	case CPU_ONLINE:
 	case CPU_DOWN_FAILED:
-		cpuset_update_active_cpus();
-		return NOTIFY_OK;
+		cpuset_update_active_cpus(true);
+		break;
 	default:
 		return NOTIFY_DONE;
 	}
+	return NOTIFY_OK;
 }
 
 static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action,
 			       void *hcpu)
 {
-	switch (action & ~CPU_TASKS_FROZEN) {
+	switch (action) {
 	case CPU_DOWN_PREPARE:
-		cpuset_update_active_cpus();
-		return NOTIFY_OK;
+		cpuset_update_active_cpus(false);
+		break;
+	case CPU_DOWN_PREPARE_FROZEN:
+		num_cpus_frozen++;
+		partition_sched_domains(1, NULL, NULL);
+		break;
 	default:
 		return NOTIFY_DONE;
 	}
+	return NOTIFY_OK;
 }
 
 void __init sched_init_smp(void)
@@ -7308,6 +7658,7 @@ void sched_destroy_group(struct task_group *tg)
  */
 void sched_move_task(struct task_struct *tsk)
 {
+	struct task_group *tg;
 	int on_rq, running;
 	unsigned long flags;
 	struct rq *rq;
@@ -7322,6 +7673,12 @@ void sched_move_task(struct task_struct *tsk)
 	if (unlikely(running))
 		tsk->sched_class->put_prev_task(rq, tsk);
 
+	tg = container_of(task_subsys_state_check(tsk, cpu_cgroup_subsys_id,
+				lockdep_is_held(&tsk->sighand->siglock)),
+			  struct task_group, css);
+	tg = autogroup_task_group(tsk, tg);
+	tsk->sched_task_group = tg;
+
 #ifdef CONFIG_FAIR_GROUP_SCHED
 	if (tsk->sched_class->task_move_group)
 		tsk->sched_class->task_move_group(tsk, on_rq);