sched: Move mmdrop to RCU on RT

Takes sleeping locks and calls into the memory allocator, so nothing
we want to do in task switch and oder atomic contexts.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

4 files changed, 45 insertions, 2 deletions

diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
index f8d1492a114f4..b238ebfbb4d63 100644
--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -11,6 +11,7 @@
 #include <linux/completion.h>
 #include <linux/cpumask.h>
 #include <linux/uprobes.h>
+#include <linux/rcupdate.h>
 #include <linux/page-flags-layout.h>
 #include <asm/page.h>
 #include <asm/mmu.h>
@@ -504,6 +505,9 @@ struct mm_struct {
 	bool tlb_flush_pending;
 #endif
 	struct uprobes_state uprobes_state;
+#ifdef CONFIG_PREEMPT_RT_BASE
+	struct rcu_head delayed_drop;
+#endif
 #ifdef CONFIG_X86_INTEL_MPX
 	/* address of the bounds directory */
 	void __user *bd_addr;
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 391c483813eec..5f6dc5bba0ca3 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -2606,12 +2606,24 @@ extern struct mm_struct * mm_alloc(void);
 
 /* mmdrop drops the mm and the page tables */
 extern void __mmdrop(struct mm_struct *);
+
 static inline void mmdrop(struct mm_struct * mm)
 {
 	if (unlikely(atomic_dec_and_test(&mm->mm_count)))
 		__mmdrop(mm);
 }
 
+#ifdef CONFIG_PREEMPT_RT_BASE
+extern void __mmdrop_delayed(struct rcu_head *rhp);
+static inline void mmdrop_delayed(struct mm_struct *mm)
+{
+	if (atomic_dec_and_test(&mm->mm_count))
+		call_rcu(&mm->delayed_drop, __mmdrop_delayed);
+}
+#else
+# define mmdrop_delayed(mm)	mmdrop(mm)
+#endif
+
 /* mmput gets rid of the mappings and all user-space */
 extern void mmput(struct mm_struct *);
 /* Grab a reference to a task's mm, if it is not already going away */
diff --git a/kernel/fork.c b/kernel/fork.c
index 2154273c3cf53..6ad981b218668 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -702,6 +702,19 @@ void __mmdrop(struct mm_struct *mm)
 }
 EXPORT_SYMBOL_GPL(__mmdrop);
 
+#ifdef CONFIG_PREEMPT_RT_BASE
+/*
+ * RCU callback for delayed mm drop. Not strictly rcu, but we don't
+ * want another facility to make this work.
+ */
+void __mmdrop_delayed(struct rcu_head *rhp)
+{
+	struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
+
+	__mmdrop(mm);
+}
+#endif
+
 /*
  * Decrement the use count and release all resources for an mm.
  */
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 915a7e3bacd95..b09fefa8ea16e 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2593,8 +2593,12 @@ static struct rq *finish_task_switch(struct task_struct *prev)
 	finish_arch_post_lock_switch();
 
 	fire_sched_in_preempt_notifiers(current);
+	/*
+	 * We use mmdrop_delayed() here so we don't have to do the
+	 * full __mmdrop() when we are the last user.
+	 */
 	if (mm)
-		mmdrop(mm);
+		mmdrop_delayed(mm);
 	if (unlikely(prev_state == TASK_DEAD)) {
 		if (prev->sched_class->task_dead)
 			prev->sched_class->task_dead(prev);
@@ -5255,6 +5259,8 @@ void sched_setnuma(struct task_struct *p, int nid)
 #endif /* CONFIG_NUMA_BALANCING */
 
 #ifdef CONFIG_HOTPLUG_CPU
+static DEFINE_PER_CPU(struct mm_struct *, idle_last_mm);
+
 /*
  * Ensures that the idle task is using init_mm right before its cpu goes
  * offline.
@@ -5269,7 +5275,11 @@ void idle_task_exit(void)
 		switch_mm(mm, &init_mm, current);
 		finish_arch_post_lock_switch();
 	}
-	mmdrop(mm);
+	/*
+	 * Defer the cleanup to an alive cpu. On RT we can neither
+	 * call mmdrop() nor mmdrop_delayed() from here.
+	 */
+	per_cpu(idle_last_mm, smp_processor_id()) = mm;
 }
 
 /*
@@ -5641,6 +5651,10 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
 
 	case CPU_DEAD:
 		calc_load_migrate(rq);
+		if (per_cpu(idle_last_mm, cpu)) {
+			mmdrop(per_cpu(idle_last_mm, cpu));
+			per_cpu(idle_last_mm, cpu) = NULL;
+		}
 		break;
 #endif
 	}