]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
cpuset: introduce cpuset_for_each_child()
authorTejun Heo <tj@kernel.org>
Mon, 7 Jan 2013 16:51:07 +0000 (08:51 -0800)
committerTejun Heo <tj@kernel.org>
Mon, 7 Jan 2013 16:51:07 +0000 (08:51 -0800)
Instead of iterating cgroup->children directly, introduce and use
cpuset_for_each_child() which wraps cgroup_for_each_child() and
performs online check.  As it uses the generic iterator, it requires
RCU read locking too.

As cpuset is currently protected by cgroup_mutex, non-online cpusets
aren't visible to all the iterations and this patch currently doesn't
make any functional difference.  This will be used to de-couple cpuset
locking from cgroup core.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
kernel/cpuset.c

index e857887bd2469efc88f5b82b6f7221196da6c6f3..4b054b9faf3db48a9bbbf47f17438fae515a395a 100644 (file)
@@ -200,6 +200,19 @@ static struct cpuset top_cpuset = {
                  (1 << CS_MEM_EXCLUSIVE)),
 };
 
+/**
+ * cpuset_for_each_child - traverse online children of a cpuset
+ * @child_cs: loop cursor pointing to the current child
+ * @pos_cgrp: used for iteration
+ * @parent_cs: target cpuset to walk children of
+ *
+ * Walk @child_cs through the online children of @parent_cs.  Must be used
+ * with RCU read locked.
+ */
+#define cpuset_for_each_child(child_cs, pos_cgrp, parent_cs)           \
+       cgroup_for_each_child((pos_cgrp), (parent_cs)->css.cgroup)      \
+               if (is_cpuset_online(((child_cs) = cgroup_cs((pos_cgrp)))))
+
 /*
  * There are two global mutexes guarding cpuset structures.  The first
  * is the main control groups cgroup_mutex, accessed via
@@ -419,48 +432,55 @@ static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
 {
        struct cgroup *cont;
        struct cpuset *c, *par;
+       int ret;
+
+       rcu_read_lock();
 
        /* Each of our child cpusets must be a subset of us */
-       list_for_each_entry(cont, &cur->css.cgroup->children, sibling) {
-               if (!is_cpuset_subset(cgroup_cs(cont), trial))
-                       return -EBUSY;
-       }
+       ret = -EBUSY;
+       cpuset_for_each_child(c, cont, cur)
+               if (!is_cpuset_subset(c, trial))
+                       goto out;
 
        /* Remaining checks don't apply to root cpuset */
+       ret = 0;
        if (cur == &top_cpuset)
-               return 0;
+               goto out;
 
        par = cur->parent;
 
        /* We must be a subset of our parent cpuset */
+       ret = -EACCES;
        if (!is_cpuset_subset(trial, par))
-               return -EACCES;
+               goto out;
 
        /*
         * If either I or some sibling (!= me) is exclusive, we can't
         * overlap
         */
-       list_for_each_entry(cont, &par->css.cgroup->children, sibling) {
-               c = cgroup_cs(cont);
+       ret = -EINVAL;
+       cpuset_for_each_child(c, cont, par) {
                if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
                    c != cur &&
                    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
-                       return -EINVAL;
+                       goto out;
                if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
                    c != cur &&
                    nodes_intersects(trial->mems_allowed, c->mems_allowed))
-                       return -EINVAL;
+                       goto out;
        }
 
        /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */
-       if (cgroup_task_count(cur->css.cgroup)) {
-               if (cpumask_empty(trial->cpus_allowed) ||
-                   nodes_empty(trial->mems_allowed)) {
-                       return -ENOSPC;
-               }
-       }
+       ret = -ENOSPC;
+       if (cgroup_task_count(cur->css.cgroup) &&
+           (cpumask_empty(trial->cpus_allowed) ||
+            nodes_empty(trial->mems_allowed)))
+               goto out;
 
-       return 0;
+       ret = 0;
+out:
+       rcu_read_unlock();
+       return ret;
 }
 
 #ifdef CONFIG_SMP
@@ -501,10 +521,10 @@ update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c)
                if (is_sched_load_balance(cp))
                        update_domain_attr(dattr, cp);
 
-               list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
-                       child = cgroup_cs(cont);
+               rcu_read_lock();
+               cpuset_for_each_child(child, cont, cp)
                        list_add_tail(&child->stack_list, &q);
-               }
+               rcu_read_unlock();
        }
 }
 
@@ -623,10 +643,10 @@ static int generate_sched_domains(cpumask_var_t **domains,
                        continue;
                }
 
-               list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
-                       child = cgroup_cs(cont);
+               rcu_read_lock();
+               cpuset_for_each_child(child, cont, cp)
                        list_add_tail(&child->stack_list, &q);
-               }
+               rcu_read_unlock();
        }
 
        for (i = 0; i < csn; i++)
@@ -1824,7 +1844,8 @@ static int cpuset_css_online(struct cgroup *cgrp)
 {
        struct cpuset *cs = cgroup_cs(cgrp);
        struct cpuset *parent = cs->parent;
-       struct cgroup *tmp_cg;
+       struct cpuset *tmp_cs;
+       struct cgroup *pos_cg;
 
        if (!parent)
                return 0;
@@ -1853,12 +1874,14 @@ static int cpuset_css_online(struct cgroup *cgrp)
         * changed to grant parent->cpus_allowed-sibling_cpus_exclusive
         * (and likewise for mems) to the new cgroup.
         */
-       list_for_each_entry(tmp_cg, &cgrp->parent->children, sibling) {
-               struct cpuset *tmp_cs = cgroup_cs(tmp_cg);
-
-               if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs))
+       rcu_read_lock();
+       cpuset_for_each_child(tmp_cs, pos_cg, parent) {
+               if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs)) {
+                       rcu_read_unlock();
                        return 0;
+               }
        }
+       rcu_read_unlock();
 
        mutex_lock(&callback_mutex);
        cs->mems_allowed = parent->mems_allowed;
@@ -2027,10 +2050,10 @@ static struct cpuset *cpuset_next(struct list_head *queue)
 
        cp = list_first_entry(queue, struct cpuset, stack_list);
        list_del(queue->next);
-       list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
-               child = cgroup_cs(cont);
+       rcu_read_lock();
+       cpuset_for_each_child(child, cont, cp)
                list_add_tail(&child->stack_list, queue);
-       }
+       rcu_read_unlock();
 
        return cp;
 }