If clone_children is enabled, effective masks won't be initialized
due to the bug:
# mount -t cgroup -o cpuset xxx /mnt
# echo 1 > cgroup.clone_children
# mkdir /mnt/tmp
# cat /mnt/tmp/
# cat cpuset.effective_cpus
# cat cpuset.cpus
0-15
And then this cpuset won't constrain the tasks in it.
Either the bug or the fix has no effect on unified hierarchy, as
there's no clone_chidren flag there any more.
Reported-by: Christian Brauner <[email protected]>
Reported-by: Serge Hallyn <[email protected]>
Cc: <[email protected]> # 3.17+
Signed-off-by: Zefan Li <[email protected]>
---
kernel/cpuset.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/kernel/cpuset.c b/kernel/cpuset.c
index 64b257f..7e9d711 100644
--- a/kernel/cpuset.c
+++ b/kernel/cpuset.c
@@ -1992,7 +1992,9 @@ static int cpuset_css_online(struct cgroup_subsys_state *css)
spin_lock_irq(&callback_lock);
cs->mems_allowed = parent->mems_allowed;
+ cs->effective_mems = parent->mems_allowed;
cpumask_copy(cs->cpus_allowed, parent->cpus_allowed);
+ cpumask_copy(cs->effective_cpus, parent->cpus_allowed);
spin_unlock_irq(&callback_lock);
out_unlock:
mutex_unlock(&cpuset_mutex);
--
1.8.0.2
When we clear cpuset.cpus, cpuset.effective_cpus won't be cleared:
# mount -t cgroup -o cpuset xxx /mnt
# mkdir /mnt/tmp
# echo 0 > /mnt/tmp/cpuset.cpus
# echo > /mnt/tmp/cpuset.cpus
# cat cpuset.cpus
# cat cpuset.effective_cpus
0-15
And a kernel warning in update_cpumasks_hier() is triggered:
------------[ cut here ]------------
WARNING: CPU: 0 PID: 4028 at kernel/cpuset.c:894 update_cpumasks_hier+0x471/0x650()
Cc: <[email protected]> # 3.17+
Signed-off-by: Zefan Li <[email protected]>
---
kernel/cpuset.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/kernel/cpuset.c b/kernel/cpuset.c
index 7e9d711..29463c2 100644
--- a/kernel/cpuset.c
+++ b/kernel/cpuset.c
@@ -873,7 +873,7 @@ static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus)
* If it becomes empty, inherit the effective mask of the
* parent, which is guaranteed to have some CPUs.
*/
- if (cpumask_empty(new_cpus))
+ if (cgroup_on_dfl(cp->css.cgroup) && cpumask_empty(new_cpus))
cpumask_copy(new_cpus, parent->effective_cpus);
/* Skip the whole subtree if the cpumask remains the same. */
@@ -1129,7 +1129,7 @@ static void update_nodemasks_hier(struct cpuset *cs, nodemask_t *new_mems)
* If it becomes empty, inherit the effective mask of the
* parent, which is guaranteed to have some MEMs.
*/
- if (nodes_empty(*new_mems))
+ if (cgroup_on_dfl(cp->css.cgroup) && nodes_empty(*new_mems))
*new_mems = parent->effective_mems;
/* Skip the whole subtree if the nodemask remains the same. */
--
1.8.0.2
From: Jason Low <[email protected]>
The cpuset.sched_relax_domain_level can control how far we do
immediate load balancing on a system. However, it was found on recent
kernels that echo'ing a value into cpuset.sched_relax_domain_level
did not reduce any immediate load balancing.
The reason this occurred was because the update_domain_attr_tree() traversal
did not update for the "top_cpuset". This resulted in nothing being changed
when modifying the sched_relax_domain_level parameter.
This patch is able to address that problem by having update_domain_attr_tree()
allow updates for the root in the cpuset traversal.
Fixes: fc560a26acce ("cpuset: replace cpuset->stack_list with cpuset_for_each_descendant_pre()")
Cc: <[email protected]> # 3.9+
Signed-off-by: Jason Low <[email protected]>
Signed-off-by: Zefan Li <[email protected]>
---
kernel/cpuset.c | 3 ---
1 file changed, 3 deletions(-)
diff --git a/kernel/cpuset.c b/kernel/cpuset.c
index 29463c2..9e25599 100644
--- a/kernel/cpuset.c
+++ b/kernel/cpuset.c
@@ -548,9 +548,6 @@ static void update_domain_attr_tree(struct sched_domain_attr *dattr,
rcu_read_lock();
cpuset_for_each_descendant_pre(cp, pos_css, root_cs) {
- if (cp == root_cs)
- continue;
-
/* skip the whole subtree if @cp doesn't have any CPU */
if (cpumask_empty(cp->cpus_allowed)) {
pos_css = css_rightmost_descendant(pos_css);
--
1.8.0.2
From: Jason Low <[email protected]>
The cpuset.sched_relax_domain_level can control how far we do
immediate load balancing on a system. However, it was found on recent
kernels that echo'ing a value into cpuset.sched_relax_domain_level
did not reduce any immediate load balancing.
The reason this occurred was because the update_domain_attr_tree() traversal
did not update for the "top_cpuset". This resulted in nothing being changed
when modifying the sched_relax_domain_level parameter.
This patch is able to address that problem by having update_domain_attr_tree()
allow updates for the root in the cpuset traversal.
Fixes: fc560a26acce ("cpuset: replace cpuset->stack_list with cpuset_for_each_descendant_pre()")
Cc: <[email protected]> # 3.9+
Signed-off-by: Jason Low <[email protected]>
Signed-off-by: Zefan Li <[email protected]>
---
This is a resend. I forgot to edit the subject when sending this patch...
---
kernel/cpuset.c | 3 ---
1 file changed, 3 deletions(-)
diff --git a/kernel/cpuset.c b/kernel/cpuset.c
index 29463c2..9e25599 100644
--- a/kernel/cpuset.c
+++ b/kernel/cpuset.c
@@ -548,9 +548,6 @@ static void update_domain_attr_tree(struct sched_domain_attr *dattr,
rcu_read_lock();
cpuset_for_each_descendant_pre(cp, pos_css, root_cs) {
- if (cp == root_cs)
- continue;
-
/* skip the whole subtree if @cp doesn't have any CPU */
if (cpumask_empty(cp->cpus_allowed)) {
pos_css = css_rightmost_descendant(pos_css);
-- 1.8.0.2
Quoting Zefan Li ([email protected]):
> If clone_children is enabled, effective masks won't be initialized
> due to the bug:
>
> # mount -t cgroup -o cpuset xxx /mnt
> # echo 1 > cgroup.clone_children
> # mkdir /mnt/tmp
> # cat /mnt/tmp/
> # cat cpuset.effective_cpus
>
> # cat cpuset.cpus
> 0-15
>
> And then this cpuset won't constrain the tasks in it.
>
> Either the bug or the fix has no effect on unified hierarchy, as
> there's no clone_chidren flag there any more.
>
> Reported-by: Christian Brauner <[email protected]>
> Reported-by: Serge Hallyn <[email protected]>
Thanks - this give sme the correct output in /proc/self/status and
cpuest.cpus. (I didn't do a stress test but that seems unlikely to
be broken)
Tested-by: Serge Hallyn <[email protected]>
> Cc: <[email protected]> # 3.17+
> Signed-off-by: Zefan Li <[email protected]>
> ---
> kernel/cpuset.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/kernel/cpuset.c b/kernel/cpuset.c
> index 64b257f..7e9d711 100644
> --- a/kernel/cpuset.c
> +++ b/kernel/cpuset.c
> @@ -1992,7 +1992,9 @@ static int cpuset_css_online(struct cgroup_subsys_state *css)
>
> spin_lock_irq(&callback_lock);
> cs->mems_allowed = parent->mems_allowed;
> + cs->effective_mems = parent->mems_allowed;
> cpumask_copy(cs->cpus_allowed, parent->cpus_allowed);
> + cpumask_copy(cs->effective_cpus, parent->cpus_allowed);
> spin_unlock_irq(&callback_lock);
> out_unlock:
> mutex_unlock(&cpuset_mutex);
> --
> 1.8.0.2
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/