[PATCH v8 06/19] cpuset: Don't use the cpu_possible_mask as a last resort for cgroup v1

[Will Deacon]

On Fri, Jun 04, 2021 at 06:11:03PM +0100, Valentin Schneider wrote:
> On 02/06/21 17:47, Will Deacon wrote:
> > @@ -3322,9 +3322,13 @@ void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask)
> >
> >  void cpuset_cpus_allowed_fallback(struct task_struct *tsk)
> >  {
> > +	const struct cpumask *cs_mask;
> > +	const struct cpumask *possible_mask = task_cpu_possible_mask(tsk);
> > +
> >       rcu_read_lock();
> > -	do_set_cpus_allowed(tsk, is_in_v2_mode() ?
> > -		task_cs(tsk)->cpus_allowed : cpu_possible_mask);
> > +	cs_mask = task_cs(tsk)->cpus_allowed;
> > +	if (is_in_v2_mode() && cpumask_subset(cs_mask, possible_mask))
> > +		do_set_cpus_allowed(tsk, cs_mask);
> 
> Since the task will still go through the is_cpu_allowed() loop in
> select_fallback_rq() after this, is the subset check actually required
> here?

Yes, I think it's needed. do_set_cpus_allowed() doesn't do any checking
against the task_cpu_possible_mask, so if we returned to
select_fallback_rq() with a mask containing a mixture of 32-bit-capable and
64-bit-only CPUs then we'd end up setting an affinity mask for a 32-bit
task which contains 64-bit-only cores.

> It would have more merit if cpuset_cpus_allowed_fallback() returned whether
> it actually changed the allowed mask or not, in which case we could branch
> either to the is_cpu_allowed() loop (as we do unconditionally now), or to
> the 'state == possible' switch case.

I think this is a cleanup, so I can include it as a separate patch (see
below).

Will

--->8

diff --git a/include/linux/cpuset.h b/include/linux/cpuset.h
index 414a8e694413..d2b9c41c8edf 100644
--- a/include/linux/cpuset.h
+++ b/include/linux/cpuset.h
@@ -59,7 +59,7 @@ extern void cpuset_wait_for_hotplug(void);
 extern void cpuset_read_lock(void);
 extern void cpuset_read_unlock(void);
 extern void cpuset_cpus_allowed(struct task_struct *p, struct cpumask *mask);
-extern void cpuset_cpus_allowed_fallback(struct task_struct *p);
+extern bool cpuset_cpus_allowed_fallback(struct task_struct *p);
 extern nodemask_t cpuset_mems_allowed(struct task_struct *p);
 #define cpuset_current_mems_allowed (current->mems_allowed)
 void cpuset_init_current_mems_allowed(void);
@@ -188,8 +188,9 @@ static inline void cpuset_cpus_allowed(struct task_struct *p,
        cpumask_copy(mask, task_cpu_possible_mask(p));
 }
 
-static inline void cpuset_cpus_allowed_fallback(struct task_struct *p)
+static inline bool cpuset_cpus_allowed_fallback(struct task_struct *p)
 {
+       return false;
 }
 
 static inline nodemask_t cpuset_mems_allowed(struct task_struct *p)
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 4e7c271e3800..a6bab2259f98 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -3327,17 +3327,22 @@ void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask)
  * which will not contain a sane cpumask during cases such as cpu hotplugging.
  * This is the absolute last resort for the scheduler and it is only used if
  * _every_ other avenue has been traveled.
+ *
+ * Returns true if the affinity of @tsk was changed, false otherwise.
  **/
 
-void cpuset_cpus_allowed_fallback(struct task_struct *tsk)
+bool cpuset_cpus_allowed_fallback(struct task_struct *tsk)
 {
        const struct cpumask *cs_mask;
+       bool changed = false;
        const struct cpumask *possible_mask = task_cpu_possible_mask(tsk);
 
        rcu_read_lock();
        cs_mask = task_cs(tsk)->cpus_allowed;
-       if (is_in_v2_mode() && cpumask_subset(cs_mask, possible_mask))
+       if (is_in_v2_mode() && cpumask_subset(cs_mask, possible_mask)) {
                do_set_cpus_allowed(tsk, cs_mask);
+               changed = true;
+       }
        rcu_read_unlock();
 
        /*
@@ -3357,6 +3362,7 @@ void cpuset_cpus_allowed_fallback(struct task_struct *tsk)
         * select_fallback_rq() will fix things ups and set cpu_possible_mask
         * if required.
         */
+       return changed;
 }
 
 void __init cpuset_init_current_mems_allowed(void)
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index fc7de4f955cf..9d7a74a07632 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2951,8 +2951,7 @@ static int select_fallback_rq(int cpu, struct task_struct *p)
                /* No more Mr. Nice Guy. */
                switch (state) {
                case cpuset:
-                       if (IS_ENABLED(CONFIG_CPUSETS)) {
-                               cpuset_cpus_allowed_fallback(p);
+                       if (cpuset_cpus_allowed_fallback(p)) {
                                state = possible;
                                break;
                        }