cpuset: drop cpuset->stack_list and ->parent, take#2

index 7d73905..860ca0f 100644
@@ -558,6 +558,7 @@ static inline struct cgroup* task_cgroup(struct task_struct *task,
 
 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
 					  struct cgroup *cgroup);
+struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos);
 
 /**
  * cgroup_for_each_descendant_pre - pre-order walk of a cgroup's descendants

index 4855892..6643f70 100644
@@ -3017,6 +3017,32 @@ struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
 }
 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
 
+/**
+ * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
+ * @pos: cgroup of interest
+ *
+ * Return the rightmost descendant of @pos.  If there's no descendant,
+ * @pos is returned.  This can be used during pre-order traversal to skip
+ * subtree of @pos.
+ */
+struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
+{
+	struct cgroup *last, *tmp;
+
+	WARN_ON_ONCE(!rcu_read_lock_held());
+
+	do {
+		last = pos;
+		/* ->prev isn't RCU safe, walk ->next till the end */
+		pos = NULL;
+		list_for_each_entry_rcu(tmp, &last->children, sibling)
+			pos = tmp;
+	} while (pos);
+
+	return last;
+}
+EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
+
 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
 {
 	struct cgroup *last;

index 58aa99b..b2f8dad 100644
@@ -103,9 +103,6 @@ struct cpuset {
 	/* for custom sched domain */
 	int relax_domain_level;
 
-	/* used for walking a cpuset hierarchy */
-	struct list_head stack_list;
-
 	struct work_struct hotplug_work;
 };
 

@@ -207,6 +204,20 @@ static struct cpuset top_cpuset = {
 	cgroup_for_each_child((pos_cgrp), (parent_cs)->css.cgroup)	\
 		if (is_cpuset_online(((child_cs) = cgroup_cs((pos_cgrp)))))
 
+/**
+ * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants
+ * @des_cs: loop cursor pointing to the current descendant
+ * @pos_cgrp: used for iteration
+ * @root_cs: target cpuset to walk ancestor of
+ *
+ * Walk @des_cs through the online descendants of @root_cs.  Must be used
+ * with RCU read locked.  The caller may modify @pos_cgrp by calling
+ * cgroup_rightmost_descendant() to skip subtree.
+ */
+#define cpuset_for_each_descendant_pre(des_cs, pos_cgrp, root_cs)	\
+	cgroup_for_each_descendant_pre((pos_cgrp), (root_cs)->css.cgroup) \
+		if (is_cpuset_online(((des_cs) = cgroup_cs((pos_cgrp)))))
+
 /*
  * There are two global mutexes guarding cpuset structures - cpuset_mutex
  * and callback_mutex.  The latter may nest inside the former.  We also

@@ -507,31 +518,24 @@ update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c)
 	return;
 }
 
-static void
-update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c)
+static void update_domain_attr_tree(struct sched_domain_attr *dattr,
+				    struct cpuset *root_cs)
 {
-	LIST_HEAD(q);
-
-	list_add(&c->stack_list, &q);
-	while (!list_empty(&q)) {
-		struct cpuset *cp;
-		struct cgroup *cont;
-		struct cpuset *child;
-
-		cp = list_first_entry(&q, struct cpuset, stack_list);
-		list_del(q.next);
+	struct cpuset *cp;
+	struct cgroup *pos_cgrp;
 
-		if (cpumask_empty(cp->cpus_allowed))
+	rcu_read_lock();
+	cpuset_for_each_descendant_pre(cp, pos_cgrp, root_cs) {
+		/* skip the whole subtree if @cp doesn't have any CPU */
+		if (cpumask_empty(cp->cpus_allowed)) {
+			pos_cgrp = cgroup_rightmost_descendant(pos_cgrp);
 			continue;
+		}
 
 		if (is_sched_load_balance(cp))
 			update_domain_attr(dattr, cp);
-
-		rcu_read_lock();
-		cpuset_for_each_child(child, cont, cp)
-			list_add_tail(&child->stack_list, &q);
-		rcu_read_unlock();
 	}
+	rcu_read_unlock();
 }
 
 /*

@@ -591,7 +595,6 @@ update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c)
 static int generate_sched_domains(cpumask_var_t **domains,
 			struct sched_domain_attr **attributes)
 {
-	LIST_HEAD(q);		/* queue of cpusets to be scanned */
 	struct cpuset *cp;	/* scans q */
 	struct cpuset **csa;	/* array of all cpuset ptrs */
 	int csn;		/* how many cpuset ptrs in csa so far */

@@ -600,6 +603,7 @@ static int generate_sched_domains(cpumask_var_t **domains,
 	struct sched_domain_attr *dattr;  /* attributes for custom domains */
 	int ndoms = 0;		/* number of sched domains in result */
 	int nslot;		/* next empty doms[] struct cpumask slot */
+	struct cgroup *pos_cgrp;
 
 	doms = NULL;
 	dattr = NULL;

@@ -627,33 +631,27 @@ static int generate_sched_domains(cpumask_var_t **domains,
 		goto done;
 	csn = 0;
 
-	list_add(&top_cpuset.stack_list, &q);
-	while (!list_empty(&q)) {
-		struct cgroup *cont;
-		struct cpuset *child;   /* scans child cpusets of cp */
-
-		cp = list_first_entry(&q, struct cpuset, stack_list);
-		list_del(q.next);
-
-		if (cpumask_empty(cp->cpus_allowed))
-			continue;
-
+	rcu_read_lock();
+	cpuset_for_each_descendant_pre(cp, pos_cgrp, &top_cpuset) {
 		/*
-		 * All child cpusets contain a subset of the parent's cpus, so
-		 * just skip them, and then we call update_domain_attr_tree()
-		 * to calc relax_domain_level of the corresponding sched
-		 * domain.
+		 * Continue traversing beyond @cp iff @cp has some CPUs and
+		 * isn't load balancing.  The former is obvious.  The
+		 * latter: All child cpusets contain a subset of the
+		 * parent's cpus, so just skip them, and then we call
+		 * update_domain_attr_tree() to calc relax_domain_level of
+		 * the corresponding sched domain.
 		 */
-		if (is_sched_load_balance(cp)) {
-			csa[csn++] = cp;
+		if (!cpumask_empty(cp->cpus_allowed) &&
+		    !is_sched_load_balance(cp))
 			continue;
-		}
 
-		rcu_read_lock();
-		cpuset_for_each_child(child, cont, cp)
-			list_add_tail(&child->stack_list, &q);
-		rcu_read_unlock();
-  	}
+		if (is_sched_load_balance(cp))
+			csa[csn++] = cp;
+
+		/* skip @cp's subtree */
+		pos_cgrp = cgroup_rightmost_descendant(pos_cgrp);
+	}
+	rcu_read_unlock();
 
 	for (i = 0; i < csn; i++)
 		csa[i]->pn = i;

@@ -2068,31 +2066,6 @@ static void remove_tasks_in_empty_cpuset(struct cpuset *cs)
 	move_member_tasks_to_cpuset(cs, parent);
 }
 
-/*
- * Helper function to traverse cpusets.
- * It can be used to walk the cpuset tree from top to bottom, completing
- * one layer before dropping down to the next (thus always processing a
- * node before any of its children).
- */
-static struct cpuset *cpuset_next(struct list_head *queue)
-{
-	struct cpuset *cp;
-	struct cpuset *child;	/* scans child cpusets of cp */
-	struct cgroup *cont;
-
-	if (list_empty(queue))
-		return NULL;
-
-	cp = list_first_entry(queue, struct cpuset, stack_list);
-	list_del(queue->next);
-	rcu_read_lock();
-	cpuset_for_each_child(child, cont, cp)
-		list_add_tail(&child->stack_list, queue);
-	rcu_read_unlock();
-
-	return cp;
-}
-
 /**
  * cpuset_propagate_hotplug_workfn - propagate CPU/memory hotplug to a cpuset
  * @cs: cpuset in interest

@@ -2229,12 +2202,12 @@ static void cpuset_hotplug_workfn(struct work_struct *work)
 	/* if cpus or mems went down, we need to propagate to descendants */
 	if (cpus_offlined || mems_offlined) {
 		struct cpuset *cs;
-		LIST_HEAD(queue);
+		struct cgroup *pos_cgrp;
 
-		list_add_tail(&top_cpuset.stack_list, &queue);
-		while ((cs = cpuset_next(&queue)))
-			if (cs != &top_cpuset)
-				schedule_cpuset_propagate_hotplug(cs);
+		rcu_read_lock();
+		cpuset_for_each_descendant_pre(cs, pos_cgrp, &top_cpuset)
+			schedule_cpuset_propagate_hotplug(cs);
+		rcu_read_unlock();
 	}
 
 	mutex_unlock(&cpuset_mutex);

index b2f8dad..afadbb99 100644
@@ -87,8 +87,6 @@ struct cpuset {
 	cpumask_var_t cpus_allowed;	/* CPUs allowed to tasks in cpuset */
 	nodemask_t mems_allowed;	/* Memory Nodes allowed to tasks */
 
-	struct cpuset *parent;		/* my parent */
-
 	struct fmeter fmeter;		/* memory_pressure filter */
 
 	/*

@@ -120,6 +118,15 @@ static inline struct cpuset *task_cs(struct task_struct *task)
 			    struct cpuset, css);
 }
 
+static inline struct cpuset *parent_cs(const struct cpuset *cs)
+{
+	struct cgroup *pcgrp = cs->css.cgroup->parent;
+
+	if (pcgrp)
+		return cgroup_cs(pcgrp);
+	return NULL;
+}
+
 #ifdef CONFIG_NUMA
 static inline bool task_has_mempolicy(struct task_struct *task)
 {

@@ -323,7 +330,7 @@ static void guarantee_online_cpus(const struct cpuset *cs,
 				  struct cpumask *pmask)
 {
 	while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
-		cs = cs->parent;
+		cs = parent_cs(cs);
 	if (cs)
 		cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
 	else

@@ -348,7 +355,7 @@ static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
 {
 	while (cs && !nodes_intersects(cs->mems_allowed,
 					node_states[N_MEMORY]))
-		cs = cs->parent;
+		cs = parent_cs(cs);
 	if (cs)
 		nodes_and(*pmask, cs->mems_allowed,
 					node_states[N_MEMORY]);

@@ -461,7 +468,7 @@ static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
 	if (cur == &top_cpuset)
 		goto out;
 
-	par = cur->parent;
+	par = parent_cs(cur);
 
 	/* We must be a subset of our parent cpuset */
 	ret = -EACCES;

@@ -1866,7 +1873,6 @@ static struct cgroup_subsys_state *cpuset_css_alloc(struct cgroup *cont)
 	fmeter_init(&cs->fmeter);
 	INIT_WORK(&cs->hotplug_work, cpuset_propagate_hotplug_workfn);
 	cs->relax_domain_level = -1;
-	cs->parent = cgroup_cs(cont->parent);
 
 	return &cs->css;
 }

@@ -1874,7 +1880,7 @@ static struct cgroup_subsys_state *cpuset_css_alloc(struct cgroup *cont)
 static int cpuset_css_online(struct cgroup *cgrp)
 {
 	struct cpuset *cs = cgroup_cs(cgrp);
-	struct cpuset *parent = cs->parent;
+	struct cpuset *parent = parent_cs(cs);
 	struct cpuset *tmp_cs;
 	struct cgroup *pos_cg;
 

@@ -2058,10 +2064,10 @@ static void remove_tasks_in_empty_cpuset(struct cpuset *cs)
 	 * Find its next-highest non-empty parent, (top cpuset
 	 * has online cpus, so can't be empty).
 	 */
-	parent = cs->parent;
+	parent = parent_cs(cs);
 	while (cpumask_empty(parent->cpus_allowed) ||
 			nodes_empty(parent->mems_allowed))
-		parent = parent->parent;
+		parent = parent_cs(parent);
 
 	move_member_tasks_to_cpuset(cs, parent);
 }

@@ -2373,8 +2379,8 @@ int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask)
  */
 static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs)
 {
-	while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent)
-		cs = cs->parent;
+	while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && parent_cs(cs))
+		cs = parent_cs(cs);
 	return cs;
 }