Re: [PATCH v4 4/5] sched/core: Prevent race condition between cpuset and __sched_setscheduler()

[Juri Lelli <juri.lelli@redhat.com>]

On 15/06/18 09:01, Juri Lelli wrote:

[...]

> I'll try harder to find alternatives, but suggestions are welcome! :-)

I wonder if something like the following might actually work. IIUC
cpuset.c comment [1], callback_lock is the one to actually take if one
needs to only query cpusets.

[1] https://elixir.bootlin.com/linux/latest/source/kernel/cgroup/cpuset.c#L266

--->8---
 include/linux/cpuset.h |  4 +--
 kernel/cgroup/cpuset.c | 72 +++++++++++++++++++++++++-------------------------
 kernel/sched/core.c    | 24 +++++++----------
 3 files changed, 48 insertions(+), 52 deletions(-)

diff --git a/include/linux/cpuset.h b/include/linux/cpuset.h
index a1970862ab8e..4bbb3f5a3020 100644
--- a/include/linux/cpuset.h
+++ b/include/linux/cpuset.h
@@ -55,7 +55,7 @@ extern void cpuset_init_smp(void);
 extern void cpuset_force_rebuild(void);
 extern void cpuset_update_active_cpus(void);
 extern void cpuset_wait_for_hotplug(void);
-extern int cpuset_lock(void);
+extern void cpuset_lock(void);
 extern void cpuset_unlock(void);
 extern void cpuset_cpus_allowed(struct task_struct *p, struct cpumask *mask);
 extern void cpuset_cpus_allowed_fallback(struct task_struct *p);
@@ -178,7 +178,7 @@ static inline void cpuset_update_active_cpus(void)
 
 static inline void cpuset_wait_for_hotplug(void) { }
 
-static inline int cpuset_lock(void) { return 1; }
+static inline void cpuset_lock(void) { }
 
 static inline void cpuset_unlock(void) { }
 
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index d26fd4795aa3..d5a0b4ec31af 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -288,7 +288,7 @@ static struct cpuset top_cpuset = {
  */
 
 static DEFINE_MUTEX(cpuset_mutex);
-static DEFINE_SPINLOCK(callback_lock);
+static DEFINE_RAW_SPINLOCK(callback_lock);
 
 static struct workqueue_struct *cpuset_migrate_mm_wq;
 
@@ -921,9 +921,9 @@ static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus)
 			continue;
 		rcu_read_unlock();
 
-		spin_lock_irq(&callback_lock);
+		raw_spin_lock_irq(&callback_lock);
 		cpumask_copy(cp->effective_cpus, new_cpus);
-		spin_unlock_irq(&callback_lock);
+		raw_spin_unlock_irq(&callback_lock);
 
 		WARN_ON(!is_in_v2_mode() &&
 			!cpumask_equal(cp->cpus_allowed, cp->effective_cpus));
@@ -988,9 +988,9 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	if (retval < 0)
 		return retval;
 
-	spin_lock_irq(&callback_lock);
+	raw_spin_lock_irq(&callback_lock);
 	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
-	spin_unlock_irq(&callback_lock);
+	raw_spin_unlock_irq(&callback_lock);
 
 	/* use trialcs->cpus_allowed as a temp variable */
 	update_cpumasks_hier(cs, trialcs->cpus_allowed);
@@ -1174,9 +1174,9 @@ static void update_nodemasks_hier(struct cpuset *cs, nodemask_t *new_mems)
 			continue;
 		rcu_read_unlock();
 
-		spin_lock_irq(&callback_lock);
+		raw_spin_lock_irq(&callback_lock);
 		cp->effective_mems = *new_mems;
-		spin_unlock_irq(&callback_lock);
+		raw_spin_unlock_irq(&callback_lock);
 
 		WARN_ON(!is_in_v2_mode() &&
 			!nodes_equal(cp->mems_allowed, cp->effective_mems));
@@ -1244,9 +1244,9 @@ static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
 	if (retval < 0)
 		goto done;
 
-	spin_lock_irq(&callback_lock);
+	raw_spin_lock_irq(&callback_lock);
 	cs->mems_allowed = trialcs->mems_allowed;
-	spin_unlock_irq(&callback_lock);
+	raw_spin_unlock_irq(&callback_lock);
 
 	/* use trialcs->mems_allowed as a temp variable */
 	update_nodemasks_hier(cs, &trialcs->mems_allowed);
@@ -1337,9 +1337,9 @@ static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
 	spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs))
 			|| (is_spread_page(cs) != is_spread_page(trialcs)));
 
-	spin_lock_irq(&callback_lock);
+	raw_spin_lock_irq(&callback_lock);
 	cs->flags = trialcs->flags;
-	spin_unlock_irq(&callback_lock);
+	raw_spin_unlock_irq(&callback_lock);
 
 	if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed)
 		rebuild_sched_domains_locked();
@@ -1754,7 +1754,7 @@ static int cpuset_common_seq_show(struct seq_file *sf, void *v)
 	cpuset_filetype_t type = seq_cft(sf)->private;
 	int ret = 0;
 
-	spin_lock_irq(&callback_lock);
+	raw_spin_lock_irq(&callback_lock);
 
 	switch (type) {
 	case FILE_CPULIST:
@@ -1773,7 +1773,7 @@ static int cpuset_common_seq_show(struct seq_file *sf, void *v)
 		ret = -EINVAL;
 	}
 
-	spin_unlock_irq(&callback_lock);
+	raw_spin_unlock_irq(&callback_lock);
 	return ret;
 }
 
@@ -1988,12 +1988,12 @@ static int cpuset_css_online(struct cgroup_subsys_state *css)
 
 	cpuset_inc();
 
-	spin_lock_irq(&callback_lock);
+	raw_spin_lock_irq(&callback_lock);
 	if (is_in_v2_mode()) {
 		cpumask_copy(cs->effective_cpus, parent->effective_cpus);
 		cs->effective_mems = parent->effective_mems;
 	}
-	spin_unlock_irq(&callback_lock);
+	raw_spin_unlock_irq(&callback_lock);
 
 	if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags))
 		goto out_unlock;
@@ -2020,12 +2020,12 @@ static int cpuset_css_online(struct cgroup_subsys_state *css)
 	}
 	rcu_read_unlock();
 
-	spin_lock_irq(&callback_lock);
+	raw_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);
+	raw_spin_unlock_irq(&callback_lock);
 out_unlock:
 	mutex_unlock(&cpuset_mutex);
 	return 0;
@@ -2064,7 +2064,7 @@ static void cpuset_css_free(struct cgroup_subsys_state *css)
 static void cpuset_bind(struct cgroup_subsys_state *root_css)
 {
 	mutex_lock(&cpuset_mutex);
-	spin_lock_irq(&callback_lock);
+	raw_spin_lock_irq(&callback_lock);
 
 	if (is_in_v2_mode()) {
 		cpumask_copy(top_cpuset.cpus_allowed, cpu_possible_mask);
@@ -2075,7 +2075,7 @@ static void cpuset_bind(struct cgroup_subsys_state *root_css)
 		top_cpuset.mems_allowed = top_cpuset.effective_mems;
 	}
 
-	spin_unlock_irq(&callback_lock);
+	raw_spin_unlock_irq(&callback_lock);
 	mutex_unlock(&cpuset_mutex);
 }
 
@@ -2173,12 +2173,12 @@ hotplug_update_tasks_legacy(struct cpuset *cs,
 {
 	bool is_empty;
 
-	spin_lock_irq(&callback_lock);
+	raw_spin_lock_irq(&callback_lock);
 	cpumask_copy(cs->cpus_allowed, new_cpus);
 	cpumask_copy(cs->effective_cpus, new_cpus);
 	cs->mems_allowed = *new_mems;
 	cs->effective_mems = *new_mems;
-	spin_unlock_irq(&callback_lock);
+	raw_spin_unlock_irq(&callback_lock);
 
 	/*
 	 * Don't call update_tasks_cpumask() if the cpuset becomes empty,
@@ -2215,10 +2215,10 @@ hotplug_update_tasks(struct cpuset *cs,
 	if (nodes_empty(*new_mems))
 		*new_mems = parent_cs(cs)->effective_mems;
 
-	spin_lock_irq(&callback_lock);
+	raw_spin_lock_irq(&callback_lock);
 	cpumask_copy(cs->effective_cpus, new_cpus);
 	cs->effective_mems = *new_mems;
-	spin_unlock_irq(&callback_lock);
+	raw_spin_unlock_irq(&callback_lock);
 
 	if (cpus_updated)
 		update_tasks_cpumask(cs);
@@ -2311,21 +2311,21 @@ static void cpuset_hotplug_workfn(struct work_struct *work)
 
 	/* synchronize cpus_allowed to cpu_active_mask */
 	if (cpus_updated) {
-		spin_lock_irq(&callback_lock);
+		raw_spin_lock_irq(&callback_lock);
 		if (!on_dfl)
 			cpumask_copy(top_cpuset.cpus_allowed, &new_cpus);
 		cpumask_copy(top_cpuset.effective_cpus, &new_cpus);
-		spin_unlock_irq(&callback_lock);
+		raw_spin_unlock_irq(&callback_lock);
 		/* we don't mess with cpumasks of tasks in top_cpuset */
 	}
 
 	/* synchronize mems_allowed to N_MEMORY */
 	if (mems_updated) {
-		spin_lock_irq(&callback_lock);
+		raw_spin_lock_irq(&callback_lock);
 		if (!on_dfl)
 			top_cpuset.mems_allowed = new_mems;
 		top_cpuset.effective_mems = new_mems;
-		spin_unlock_irq(&callback_lock);
+		raw_spin_unlock_irq(&callback_lock);
 		update_tasks_nodemask(&top_cpuset);
 	}
 
@@ -2412,9 +2412,9 @@ void __init cpuset_init_smp(void)
 /**
  * cpuset_lock - Grab the cpuset_mutex from another subsysytem
  */
-int cpuset_lock(void)
+void cpuset_lock(void)
 {
-	return mutex_trylock(&cpuset_mutex);
+	raw_spin_lock(&callback_lock);
 }
 
 /**
@@ -2422,7 +2422,7 @@ int cpuset_lock(void)
  */
 void cpuset_unlock(void)
 {
-	mutex_unlock(&cpuset_mutex);
+	raw_spin_unlock(&callback_lock);
 }
 
 /**
@@ -2440,11 +2440,11 @@ void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask)
 {
 	unsigned long flags;
 
-	spin_lock_irqsave(&callback_lock, flags);
+	raw_spin_lock_irqsave(&callback_lock, flags);
 	rcu_read_lock();
 	guarantee_online_cpus(task_cs(tsk), pmask);
 	rcu_read_unlock();
-	spin_unlock_irqrestore(&callback_lock, flags);
+	raw_spin_unlock_irqrestore(&callback_lock, flags);
 }
 
 void cpuset_cpus_allowed_fallback(struct task_struct *tsk)
@@ -2492,11 +2492,11 @@ nodemask_t cpuset_mems_allowed(struct task_struct *tsk)
 	nodemask_t mask;
 	unsigned long flags;
 
-	spin_lock_irqsave(&callback_lock, flags);
+	raw_spin_lock_irqsave(&callback_lock, flags);
 	rcu_read_lock();
 	guarantee_online_mems(task_cs(tsk), &mask);
 	rcu_read_unlock();
-	spin_unlock_irqrestore(&callback_lock, flags);
+	raw_spin_unlock_irqrestore(&callback_lock, flags);
 
 	return mask;
 }
@@ -2588,14 +2588,14 @@ bool __cpuset_node_allowed(int node, gfp_t gfp_mask)
 		return true;
 
 	/* Not hardwall and node outside mems_allowed: scan up cpusets */
-	spin_lock_irqsave(&callback_lock, flags);
+	raw_spin_lock_irqsave(&callback_lock, flags);
 
 	rcu_read_lock();
 	cs = nearest_hardwall_ancestor(task_cs(current));
 	allowed = node_isset(node, cs->mems_allowed);
 	rcu_read_unlock();
 
-	spin_unlock_irqrestore(&callback_lock, flags);
+	raw_spin_unlock_irqrestore(&callback_lock, flags);
 	return allowed;
 }
 
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index a5b0c6c25b44..9c5285cc082c 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -4218,14 +4218,6 @@ static int __sched_setscheduler(struct task_struct *p,
 		if (attr->sched_flags & SCHED_FLAG_SUGOV)
 			return -EINVAL;
 
-		/*
-		 * Make sure we don't race with the cpuset subsystem where root
-		 * domains can be rebuilt or modified while operations like DL
-		 * admission checks are carried out.
-		 */
-		if (!cpuset_lock())
-			return -EBUSY;
-
 		retval = security_task_setscheduler(p);
 		if (retval)
 			return retval;
@@ -4241,6 +4233,13 @@ static int __sched_setscheduler(struct task_struct *p,
 	rq = task_rq_lock(p, &rf);
 	update_rq_clock(rq);
 
+	/*
+	 * Make sure we don't race with the cpuset subsystem where root
+	 * domains can be rebuilt or modified while operations like DL
+	 * admission checks are carried out.
+	 */
+	cpuset_lock();
+
 	/*
 	 * Changing the policy of the stop threads its a very bad idea:
 	 */
@@ -4302,9 +4301,8 @@ static int __sched_setscheduler(struct task_struct *p,
 	/* Re-check policy now with rq lock held: */
 	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
 		policy = oldpolicy = -1;
+		cpuset_unlock();
 		task_rq_unlock(rq, p, &rf);
-		if (user)
-			cpuset_unlock();
 		goto recheck;
 	}
 
@@ -4361,9 +4359,8 @@ static int __sched_setscheduler(struct task_struct *p,
 
 	/* Avoid rq from going away on us: */
 	preempt_disable();
+	cpuset_unlock();
 	task_rq_unlock(rq, p, &rf);
-	if (user)
-		cpuset_unlock();
 
 	if (pi)
 		rt_mutex_adjust_pi(p);
@@ -4375,9 +4372,8 @@ static int __sched_setscheduler(struct task_struct *p,
 	return 0;
 
 unlock:
+	cpuset_unlock();
 	task_rq_unlock(rq, p, &rf);
-	if (user)
-		cpuset_unlock();
 	return retval;
 }