diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index dc9301d31f12..9ebeb3cfb492 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -709,6 +709,8 @@ long vhost_dev_set_owner(struct vhost_dev *dev)
 
 	/* No owner, become one */
 	dev->mm = get_task_mm(current);
+	debug_mm = dev->mm;
+	printk("assign debug_mm to %p\n", dev->mm);
 	worker = kthread_create(vhost_worker, dev, "vhost-%d", current->pid);
 	if (IS_ERR(worker)) {
 		err = PTR_ERR(worker);
@@ -962,7 +964,10 @@ void vhost_dev_cleanup(struct vhost_dev *dev)
 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 		mmu_notifier_unregister(&dev->mmu_notifier, dev->mm);
 #endif
+		printk("dev->mm %p == debug_mm %p %d\n", dev->mm,
+			debug_mm, dev->mm == debug_mm);
 		mmput(dev->mm);
+		printk("debug_mm is %p\n", debug_mm);
 	}
 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 	for (i = 0; i < dev->nvqs; i++)
diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
index 8ec38b11b361..3126d0c23763 100644
--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -515,6 +515,7 @@ struct mm_struct {
 };
 
 extern struct mm_struct init_mm;
+extern struct mm_struct *debug_mm;
 
 /* Pointer magic because the dynamic array size confuses some compilers. */
 static inline void mm_init_cpumask(struct mm_struct *mm)
diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h
index a3fda9f024c3..dd76a8e0ad58 100644
--- a/include/linux/sched/mm.h
+++ b/include/linux/sched/mm.h
@@ -9,6 +9,7 @@
 #include <linux/gfp.h>
 #include <linux/sync_core.h>
 
+
 /*
  * Routines for handling mm_structs
  */
@@ -33,6 +34,10 @@ extern struct mm_struct *mm_alloc(void);
  */
 static inline void mmgrab(struct mm_struct *mm)
 {
+	if (debug_mm == mm) {
+		printk("mmgrab %p\n", mm);
+		dump_stack();
+	}
 	atomic_inc(&mm->mm_count);
 }
 
@@ -45,6 +50,10 @@ static inline void mmdrop(struct mm_struct *mm)
 	 * required by the membarrier system call before returning to
 	 * user-space, after storing to rq->curr.
 	 */
+	if (debug_mm == mm) {
+		printk("mmdrop %p before %d\n", mm, atomic_read(&mm->mm_count));
+		dump_stack();
+	}
 	if (unlikely(atomic_dec_and_test(&mm->mm_count)))
 		__mmdrop(mm);
 }
@@ -88,11 +97,19 @@ static inline bool mmget_still_valid(struct mm_struct *mm)
  */
 static inline void mmget(struct mm_struct *mm)
 {
+	if (debug_mm == mm) {
+		printk("mget %p\n", mm);
+		dump_stack();
+	}
 	atomic_inc(&mm->mm_users);
 }
 
 static inline bool mmget_not_zero(struct mm_struct *mm)
 {
+	if (debug_mm == mm) {
+		printk("mmget_not_zero %p\n", mm);
+		dump_stack();
+	}
 	return atomic_inc_not_zero(&mm->mm_users);
 }
 
diff --git a/kernel/fork.c b/kernel/fork.c
index b2b87d450b80..9a4e5d69e9bf 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1076,6 +1076,10 @@ void mmput(struct mm_struct *mm)
 {
 	might_sleep();
 
+	if (debug_mm == mm) {
+		printk("mmput mm %p before %d\n", mm, atomic_read(&mm->mm_users));
+		dump_stack();
+	}
 	if (atomic_dec_and_test(&mm->mm_users))
 		__mmput(mm);
 }
@@ -1092,6 +1096,10 @@ static void mmput_async_fn(struct work_struct *work)
 
 void mmput_async(struct mm_struct *mm)
 {
+	if (debug_mm == mm) {
+		printk("mmput mm %p before %d\n", mm, atomic_read(&mm->mm_users));
+		dump_stack();
+	}
 	if (atomic_dec_and_test(&mm->mm_users)) {
 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
 		schedule_work(&mm->async_put_work);
diff --git a/mm/init-mm.c b/mm/init-mm.c
index a787a319211e..f6253a53a891 100644
--- a/mm/init-mm.c
+++ b/mm/init-mm.c
@@ -38,3 +38,6 @@ struct mm_struct init_mm = {
 	.cpu_bitmap	= { [BITS_TO_LONGS(NR_CPUS)] = 0},
 	INIT_MM_CONTEXT(init_mm)
 };
+
+struct mm_struct *debug_mm = NULL;
+EXPORT_SYMBOL_GPL(debug_mm);