1 files changed, 200 insertions, 49 deletions

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 39eb6011bc3..d5594a4268d 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
@@ -2517,25 +2516,32 @@ 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 +2553,38 @@ 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 +5014,7 @@ 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 +5556,20 @@ 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 +5609,12 @@ 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 +5662,7 @@ 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 +5683,10 @@ 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)
@@ -5962,6 +6008,44 @@ 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 +6064,12 @@ 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 +6077,6 @@ 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 +6085,24 @@ 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);
 
-		if (cpumask_test_cpu(cpu, sg_span))
+		/*
+		 * 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);
+
+		/*
+		 * 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 +6176,7 @@ 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 +6218,7 @@ 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 +6268,8 @@ 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 +6379,13 @@ 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 +6443,42 @@ 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 +6486,6 @@ 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 +6496,41 @@ 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 +6555,7 @@ 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 +6642,7 @@ 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 +6695,6 @@ 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 +6702,7 @@ 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 +6843,6 @@ 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();