Date:   Fri, 17 Jan 2020 12:41:29 -0500
From:   Steven Rostedt <rostedt@...dmis.org>
To:     linux-kernel@...r.kernel.org,
        linux-rt-users <linux-rt-users@...r.kernel.org>
Cc:     Thomas Gleixner <tglx@...utronix.de>,
        Carsten Emde <C.Emde@...dl.org>,
        Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
        John Kacur <jkacur@...hat.com>,
        Julia Cartwright <julia@...com>,
        Daniel Wagner <wagi@...om.org>,
        Tom Zanussi <zanussi@...nel.org>
Subject: [PATCH RT 18/32] futex: Make the futex_hash_bucket spinlock_t again and bring back its
 old state

4.19.94-rt39-rc1 stable review patch.
If anyone has any objections, please let me know.

------------------

From: Sebastian Andrzej Siewior <bigeasy@...utronix.de>

[ Upstream commit 954ad80c23edfe71f4e8ce70b961eac884320c3a ]

This is an all-in-one patch that reverts the patches:
   futex: Make the futex_hash_bucket lock raw
   futex: Delay deallocation of pi_state

and adds back the old patches we had:
   futex: workaround migrate_disable/enable in different context
   rtmutex: Handle the various new futex race conditions
   futex: Fix bug on when a requeued RT task times out
   futex: Ensure lock/unlock symetry versus pi_lock and hash bucket lock

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@...utronix.de>
Signed-off-by: Steven Rostedt (VMware) <rostedt@...dmis.org>
---
 kernel/futex.c                  | 230 ++++++++++++++++++--------------
 kernel/locking/rtmutex.c        |  65 ++++++++-
 kernel/locking/rtmutex_common.h |   3 +
 3 files changed, 194 insertions(+), 104 deletions(-)

diff --git a/kernel/futex.c b/kernel/futex.c
index 0b8cff8d9162..e815cf542b82 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -243,7 +243,7 @@ struct futex_q {
 	struct plist_node list;
 
 	struct task_struct *task;
-	raw_spinlock_t *lock_ptr;
+	spinlock_t *lock_ptr;
 	union futex_key key;
 	struct futex_pi_state *pi_state;
 	struct rt_mutex_waiter *rt_waiter;
@@ -264,7 +264,7 @@ static const struct futex_q futex_q_init = {
  */
 struct futex_hash_bucket {
 	atomic_t waiters;
-	raw_spinlock_t lock;
+	spinlock_t lock;
 	struct plist_head chain;
 } ____cacheline_aligned_in_smp;
 
@@ -825,13 +825,13 @@ static void get_pi_state(struct futex_pi_state *pi_state)
  * Drops a reference to the pi_state object and frees or caches it
  * when the last reference is gone.
  */
-static struct futex_pi_state *__put_pi_state(struct futex_pi_state *pi_state)
+static void put_pi_state(struct futex_pi_state *pi_state)
 {
 	if (!pi_state)
-		return NULL;
+		return;
 
 	if (!atomic_dec_and_test(&pi_state->refcount))
-		return NULL;
+		return;
 
 	/*
 	 * If pi_state->owner is NULL, the owner is most probably dying
@@ -851,7 +851,9 @@ static struct futex_pi_state *__put_pi_state(struct futex_pi_state *pi_state)
 		raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
 	}
 
-	if (!current->pi_state_cache) {
+	if (current->pi_state_cache) {
+		kfree(pi_state);
+	} else {
 		/*
 		 * pi_state->list is already empty.
 		 * clear pi_state->owner.
@@ -860,30 +862,6 @@ static struct futex_pi_state *__put_pi_state(struct futex_pi_state *pi_state)
 		pi_state->owner = NULL;
 		atomic_set(&pi_state->refcount, 1);
 		current->pi_state_cache = pi_state;
-		pi_state = NULL;
-	}
-	return pi_state;
-}
-
-static void put_pi_state(struct futex_pi_state *pi_state)
-{
-	kfree(__put_pi_state(pi_state));
-}
-
-static void put_pi_state_atomic(struct futex_pi_state *pi_state,
-				struct list_head *to_free)
-{
-	if (__put_pi_state(pi_state))
-		list_add(&pi_state->list, to_free);
-}
-
-static void free_pi_state_list(struct list_head *to_free)
-{
-	struct futex_pi_state *p, *next;
-
-	list_for_each_entry_safe(p, next, to_free, list) {
-		list_del(&p->list);
-		kfree(p);
 	}
 }
 
@@ -900,7 +878,6 @@ void exit_pi_state_list(struct task_struct *curr)
 	struct futex_pi_state *pi_state;
 	struct futex_hash_bucket *hb;
 	union futex_key key = FUTEX_KEY_INIT;
-	LIST_HEAD(to_free);
 
 	if (!futex_cmpxchg_enabled)
 		return;
@@ -934,7 +911,7 @@ void exit_pi_state_list(struct task_struct *curr)
 		}
 		raw_spin_unlock_irq(&curr->pi_lock);
 
-		raw_spin_lock(&hb->lock);
+		spin_lock(&hb->lock);
 		raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
 		raw_spin_lock(&curr->pi_lock);
 		/*
@@ -944,8 +921,10 @@ void exit_pi_state_list(struct task_struct *curr)
 		if (head->next != next) {
 			/* retain curr->pi_lock for the loop invariant */
 			raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
-			raw_spin_unlock(&hb->lock);
-			put_pi_state_atomic(pi_state, &to_free);
+			raw_spin_unlock_irq(&curr->pi_lock);
+			spin_unlock(&hb->lock);
+			raw_spin_lock_irq(&curr->pi_lock);
+			put_pi_state(pi_state);
 			continue;
 		}
 
@@ -956,7 +935,7 @@ void exit_pi_state_list(struct task_struct *curr)
 
 		raw_spin_unlock(&curr->pi_lock);
 		raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
-		raw_spin_unlock(&hb->lock);
+		spin_unlock(&hb->lock);
 
 		rt_mutex_futex_unlock(&pi_state->pi_mutex);
 		put_pi_state(pi_state);
@@ -964,8 +943,6 @@ void exit_pi_state_list(struct task_struct *curr)
 		raw_spin_lock_irq(&curr->pi_lock);
 	}
 	raw_spin_unlock_irq(&curr->pi_lock);
-
-	free_pi_state_list(&to_free);
 }
 
 #endif
@@ -1452,7 +1429,7 @@ static void __unqueue_futex(struct futex_q *q)
 {
 	struct futex_hash_bucket *hb;
 
-	if (WARN_ON_SMP(!q->lock_ptr || !raw_spin_is_locked(q->lock_ptr))
+	if (WARN_ON_SMP(!q->lock_ptr || !spin_is_locked(q->lock_ptr))
 	    || WARN_ON(plist_node_empty(&q->list)))
 		return;
 
@@ -1580,21 +1557,21 @@ static inline void
 double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
 {
 	if (hb1 <= hb2) {
-		raw_spin_lock(&hb1->lock);
+		spin_lock(&hb1->lock);
 		if (hb1 < hb2)
-			raw_spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
+			spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
 	} else { /* hb1 > hb2 */
-		raw_spin_lock(&hb2->lock);
-		raw_spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
+		spin_lock(&hb2->lock);
+		spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
 	}
 }
 
 static inline void
 double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
 {
-	raw_spin_unlock(&hb1->lock);
+	spin_unlock(&hb1->lock);
 	if (hb1 != hb2)
-		raw_spin_unlock(&hb2->lock);
+		spin_unlock(&hb2->lock);
 }
 
 /*
@@ -1622,7 +1599,7 @@ futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
 	if (!hb_waiters_pending(hb))
 		goto out_put_key;
 
-	raw_spin_lock(&hb->lock);
+	spin_lock(&hb->lock);
 
 	plist_for_each_entry_safe(this, next, &hb->chain, list) {
 		if (match_futex (&this->key, &key)) {
@@ -1641,7 +1618,7 @@ futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
 		}
 	}
 
-	raw_spin_unlock(&hb->lock);
+	spin_unlock(&hb->lock);
 	wake_up_q(&wake_q);
 out_put_key:
 	put_futex_key(&key);
@@ -1948,7 +1925,6 @@ static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
 	struct futex_hash_bucket *hb1, *hb2;
 	struct futex_q *this, *next;
 	DEFINE_WAKE_Q(wake_q);
-	LIST_HEAD(to_free);
 
 	if (nr_wake < 0 || nr_requeue < 0)
 		return -EINVAL;
@@ -2176,6 +2152,16 @@ static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
 				requeue_pi_wake_futex(this, &key2, hb2);
 				drop_count++;
 				continue;
+			} else if (ret == -EAGAIN) {
+				/*
+				 * Waiter was woken by timeout or
+				 * signal and has set pi_blocked_on to
+				 * PI_WAKEUP_INPROGRESS before we
+				 * tried to enqueue it on the rtmutex.
+				 */
+				this->pi_state = NULL;
+				put_pi_state(pi_state);
+				continue;
 			} else if (ret) {
 				/*
 				 * rt_mutex_start_proxy_lock() detected a
@@ -2186,7 +2172,7 @@ static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
 				 * object.
 				 */
 				this->pi_state = NULL;
-				put_pi_state_atomic(pi_state, &to_free);
+				put_pi_state(pi_state);
 				/*
 				 * We stop queueing more waiters and let user
 				 * space deal with the mess.
@@ -2203,7 +2189,7 @@ static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
 	 * in futex_proxy_trylock_atomic() or in lookup_pi_state(). We
 	 * need to drop it here again.
 	 */
-	put_pi_state_atomic(pi_state, &to_free);
+	put_pi_state(pi_state);
 
 out_unlock:
 	double_unlock_hb(hb1, hb2);
@@ -2224,7 +2210,6 @@ static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
 out_put_key1:
 	put_futex_key(&key1);
 out:
-	free_pi_state_list(&to_free);
 	return ret ? ret : task_count;
 }
 
@@ -2248,7 +2233,7 @@ static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
 
 	q->lock_ptr = &hb->lock;
 
-	raw_spin_lock(&hb->lock); /* implies smp_mb(); (A) */
+	spin_lock(&hb->lock); /* implies smp_mb(); (A) */
 	return hb;
 }
 
@@ -2256,7 +2241,7 @@ static inline void
 queue_unlock(struct futex_hash_bucket *hb)
 	__releases(&hb->lock)
 {
-	raw_spin_unlock(&hb->lock);
+	spin_unlock(&hb->lock);
 	hb_waiters_dec(hb);
 }
 
@@ -2295,7 +2280,7 @@ static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
 	__releases(&hb->lock)
 {
 	__queue_me(q, hb);
-	raw_spin_unlock(&hb->lock);
+	spin_unlock(&hb->lock);
 }
 
 /**
@@ -2311,41 +2296,41 @@ static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
  */
 static int unqueue_me(struct futex_q *q)
 {
-	raw_spinlock_t *lock_ptr;
+	spinlock_t *lock_ptr;
 	int ret = 0;
 
 	/* In the common case we don't take the spinlock, which is nice. */
 retry:
 	/*
-	 * q->lock_ptr can change between this read and the following
-	 * raw_spin_lock. Use READ_ONCE to forbid the compiler from reloading
-	 * q->lock_ptr and optimizing lock_ptr out of the logic below.
+	 * q->lock_ptr can change between this read and the following spin_lock.
+	 * Use READ_ONCE to forbid the compiler from reloading q->lock_ptr and
+	 * optimizing lock_ptr out of the logic below.
 	 */
 	lock_ptr = READ_ONCE(q->lock_ptr);
 	if (lock_ptr != NULL) {
-		raw_spin_lock(lock_ptr);
+		spin_lock(lock_ptr);
 		/*
 		 * q->lock_ptr can change between reading it and
-		 * raw_spin_lock(), causing us to take the wrong lock.  This
+		 * spin_lock(), causing us to take the wrong lock.  This
 		 * corrects the race condition.
 		 *
 		 * Reasoning goes like this: if we have the wrong lock,
 		 * q->lock_ptr must have changed (maybe several times)
-		 * between reading it and the raw_spin_lock().  It can
-		 * change again after the raw_spin_lock() but only if it was
-		 * already changed before the raw_spin_lock().  It cannot,
+		 * between reading it and the spin_lock().  It can
+		 * change again after the spin_lock() but only if it was
+		 * already changed before the spin_lock().  It cannot,
 		 * however, change back to the original value.  Therefore
 		 * we can detect whether we acquired the correct lock.
 		 */
 		if (unlikely(lock_ptr != q->lock_ptr)) {
-			raw_spin_unlock(lock_ptr);
+			spin_unlock(lock_ptr);
 			goto retry;
 		}
 		__unqueue_futex(q);
 
 		BUG_ON(q->pi_state);
 
-		raw_spin_unlock(lock_ptr);
+		spin_unlock(lock_ptr);
 		ret = 1;
 	}
 
@@ -2361,16 +2346,13 @@ static int unqueue_me(struct futex_q *q)
 static void unqueue_me_pi(struct futex_q *q)
 	__releases(q->lock_ptr)
 {
-	struct futex_pi_state *ps;
-
 	__unqueue_futex(q);
 
 	BUG_ON(!q->pi_state);
-	ps = __put_pi_state(q->pi_state);
+	put_pi_state(q->pi_state);
 	q->pi_state = NULL;
 
-	raw_spin_unlock(q->lock_ptr);
-	kfree(ps);
+	spin_unlock(q->lock_ptr);
 }
 
 static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
@@ -2503,7 +2485,7 @@ static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
 	 */
 handle_err:
 	raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
-	raw_spin_unlock(q->lock_ptr);
+	spin_unlock(q->lock_ptr);
 
 	switch (err) {
 	case -EFAULT:
@@ -2521,7 +2503,7 @@ static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
 		break;
 	}
 
-	raw_spin_lock(q->lock_ptr);
+	spin_lock(q->lock_ptr);
 	raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
 
 	/*
@@ -2617,7 +2599,7 @@ static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
 	/*
 	 * The task state is guaranteed to be set before another task can
 	 * wake it. set_current_state() is implemented using smp_store_mb() and
-	 * queue_me() calls raw_spin_unlock() upon completion, both serializing
+	 * queue_me() calls spin_unlock() upon completion, both serializing
 	 * access to the hash list and forcing another memory barrier.
 	 */
 	set_current_state(TASK_INTERRUPTIBLE);
@@ -2908,7 +2890,15 @@ static int futex_lock_pi(u32 __user *uaddr, unsigned int flags,
 	 * before __rt_mutex_start_proxy_lock() is done.
 	 */
 	raw_spin_lock_irq(&q.pi_state->pi_mutex.wait_lock);
-	raw_spin_unlock(q.lock_ptr);
+	/*
+	 * the migrate_disable() here disables migration in the in_atomic() fast
+	 * path which is enabled again in the following spin_unlock(). We have
+	 * one migrate_disable() pending in the slow-path which is reversed
+	 * after the raw_spin_unlock_irq() where we leave the atomic context.
+	 */
+	migrate_disable();
+
+	spin_unlock(q.lock_ptr);
 	/*
 	 * __rt_mutex_start_proxy_lock() unconditionally enqueues the @rt_waiter
 	 * such that futex_unlock_pi() is guaranteed to observe the waiter when
@@ -2916,6 +2906,7 @@ static int futex_lock_pi(u32 __user *uaddr, unsigned int flags,
 	 */
 	ret = __rt_mutex_start_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter, current);
 	raw_spin_unlock_irq(&q.pi_state->pi_mutex.wait_lock);
+	migrate_enable();
 
 	if (ret) {
 		if (ret == 1)
@@ -2929,7 +2920,7 @@ static int futex_lock_pi(u32 __user *uaddr, unsigned int flags,
 	ret = rt_mutex_wait_proxy_lock(&q.pi_state->pi_mutex, to, &rt_waiter);
 
 cleanup:
-	raw_spin_lock(q.lock_ptr);
+	spin_lock(q.lock_ptr);
 	/*
 	 * If we failed to acquire the lock (deadlock/signal/timeout), we must
 	 * first acquire the hb->lock before removing the lock from the
@@ -3030,7 +3021,7 @@ static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
 		return ret;
 
 	hb = hash_futex(&key);
-	raw_spin_lock(&hb->lock);
+	spin_lock(&hb->lock);
 
 	/*
 	 * Check waiters first. We do not trust user space values at
@@ -3064,10 +3055,19 @@ static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
 		 * rt_waiter. Also see the WARN in wake_futex_pi().
 		 */
 		raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
-		raw_spin_unlock(&hb->lock);
+		/*
+		 * Magic trickery for now to make the RT migrate disable
+		 * logic happy. The following spin_unlock() happens with
+		 * interrupts disabled so the internal migrate_enable()
+		 * won't undo the migrate_disable() which was issued when
+		 * locking hb->lock.
+		 */
+		migrate_disable();
+		spin_unlock(&hb->lock);
 
 		/* drops pi_state->pi_mutex.wait_lock */
 		ret = wake_futex_pi(uaddr, uval, pi_state);
+		migrate_enable();
 
 		put_pi_state(pi_state);
 
@@ -3103,7 +3103,7 @@ static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
 	 * owner.
 	 */
 	if ((ret = cmpxchg_futex_value_locked(&curval, uaddr, uval, 0))) {
-		raw_spin_unlock(&hb->lock);
+		spin_unlock(&hb->lock);
 		switch (ret) {
 		case -EFAULT:
 			goto pi_faulted;
@@ -3123,7 +3123,7 @@ static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
 	ret = (curval == uval) ? 0 : -EAGAIN;
 
 out_unlock:
-	raw_spin_unlock(&hb->lock);
+	spin_unlock(&hb->lock);
 out_putkey:
 	put_futex_key(&key);
 	return ret;
@@ -3239,7 +3239,7 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 	struct hrtimer_sleeper timeout, *to = NULL;
 	struct futex_pi_state *pi_state = NULL;
 	struct rt_mutex_waiter rt_waiter;
-	struct futex_hash_bucket *hb;
+	struct futex_hash_bucket *hb, *hb2;
 	union futex_key key2 = FUTEX_KEY_INIT;
 	struct futex_q q = futex_q_init;
 	int res, ret;
@@ -3297,20 +3297,55 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 	/* Queue the futex_q, drop the hb lock, wait for wakeup. */
 	futex_wait_queue_me(hb, &q, to);
 
-	raw_spin_lock(&hb->lock);
-	ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
-	raw_spin_unlock(&hb->lock);
-	if (ret)
-		goto out_put_keys;
+	/*
+	 * On RT we must avoid races with requeue and trying to block
+	 * on two mutexes (hb->lock and uaddr2's rtmutex) by
+	 * serializing access to pi_blocked_on with pi_lock.
+	 */
+	raw_spin_lock_irq(&current->pi_lock);
+	if (current->pi_blocked_on) {
+		/*
+		 * We have been requeued or are in the process of
+		 * being requeued.
+		 */
+		raw_spin_unlock_irq(&current->pi_lock);
+	} else {
+		/*
+		 * Setting pi_blocked_on to PI_WAKEUP_INPROGRESS
+		 * prevents a concurrent requeue from moving us to the
+		 * uaddr2 rtmutex. After that we can safely acquire
+		 * (and possibly block on) hb->lock.
+		 */
+		current->pi_blocked_on = PI_WAKEUP_INPROGRESS;
+		raw_spin_unlock_irq(&current->pi_lock);
+
+		spin_lock(&hb->lock);
+
+		/*
+		 * Clean up pi_blocked_on. We might leak it otherwise
+		 * when we succeeded with the hb->lock in the fast
+		 * path.
+		 */
+		raw_spin_lock_irq(&current->pi_lock);
+		current->pi_blocked_on = NULL;
+		raw_spin_unlock_irq(&current->pi_lock);
+
+		ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
+		spin_unlock(&hb->lock);
+		if (ret)
+			goto out_put_keys;
+	}
 
 	/*
-	 * In order for us to be here, we know our q.key == key2, and since
-	 * we took the hb->lock above, we also know that futex_requeue() has
-	 * completed and we no longer have to concern ourselves with a wakeup
-	 * race with the atomic proxy lock acquisition by the requeue code. The
-	 * futex_requeue dropped our key1 reference and incremented our key2
-	 * reference count.
+	 * In order to be here, we have either been requeued, are in
+	 * the process of being requeued, or requeue successfully
+	 * acquired uaddr2 on our behalf.  If pi_blocked_on was
+	 * non-null above, we may be racing with a requeue.  Do not
+	 * rely on q->lock_ptr to be hb2->lock until after blocking on
+	 * hb->lock or hb2->lock. The futex_requeue dropped our key1
+	 * reference and incremented our key2 reference count.
 	 */
+	hb2 = hash_futex(&key2);
 
 	/* Check if the requeue code acquired the second futex for us. */
 	if (!q.rt_waiter) {
@@ -3319,9 +3354,8 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 		 * did a lock-steal - fix up the PI-state in that case.
 		 */
 		if (q.pi_state && (q.pi_state->owner != current)) {
-			struct futex_pi_state *ps_free;
-
-			raw_spin_lock(q.lock_ptr);
+			spin_lock(&hb2->lock);
+			BUG_ON(&hb2->lock != q.lock_ptr);
 			ret = fixup_pi_state_owner(uaddr2, &q, current);
 			if (ret && rt_mutex_owner(&q.pi_state->pi_mutex) == current) {
 				pi_state = q.pi_state;
@@ -3331,9 +3365,8 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 			 * Drop the reference to the pi state which
 			 * the requeue_pi() code acquired for us.
 			 */
-			ps_free = __put_pi_state(q.pi_state);
-			raw_spin_unlock(q.lock_ptr);
-			kfree(ps_free);
+			put_pi_state(q.pi_state);
+			spin_unlock(&hb2->lock);
 		}
 	} else {
 		struct rt_mutex *pi_mutex;
@@ -3347,7 +3380,8 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 		pi_mutex = &q.pi_state->pi_mutex;
 		ret = rt_mutex_wait_proxy_lock(pi_mutex, to, &rt_waiter);
 
-		raw_spin_lock(q.lock_ptr);
+		spin_lock(&hb2->lock);
+		BUG_ON(&hb2->lock != q.lock_ptr);
 		if (ret && !rt_mutex_cleanup_proxy_lock(pi_mutex, &rt_waiter))
 			ret = 0;
 
@@ -4014,7 +4048,7 @@ static int __init futex_init(void)
 	for (i = 0; i < futex_hashsize; i++) {
 		atomic_set(&futex_queues[i].waiters, 0);
 		plist_head_init(&futex_queues[i].chain);
-		raw_spin_lock_init(&futex_queues[i].lock);
+		spin_lock_init(&futex_queues[i].lock);
 	}
 
 	return 0;
diff --git a/kernel/locking/rtmutex.c b/kernel/locking/rtmutex.c
index 44a33057a83a..2a9bf2443acc 100644
--- a/kernel/locking/rtmutex.c
+++ b/kernel/locking/rtmutex.c
@@ -142,6 +142,12 @@ static void fixup_rt_mutex_waiters(struct rt_mutex *lock)
 		WRITE_ONCE(*p, owner & ~RT_MUTEX_HAS_WAITERS);
 }
 
+static int rt_mutex_real_waiter(struct rt_mutex_waiter *waiter)
+{
+	return waiter && waiter != PI_WAKEUP_INPROGRESS &&
+		waiter != PI_REQUEUE_INPROGRESS;
+}
+
 /*
  * We can speed up the acquire/release, if there's no debugging state to be
  * set up.
@@ -415,7 +421,8 @@ int max_lock_depth = 1024;
 
 static inline struct rt_mutex *task_blocked_on_lock(struct task_struct *p)
 {
-	return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL;
+	return rt_mutex_real_waiter(p->pi_blocked_on) ?
+		p->pi_blocked_on->lock : NULL;
 }
 
 /*
@@ -551,7 +558,7 @@ static int rt_mutex_adjust_prio_chain(struct task_struct *task,
 	 * reached or the state of the chain has changed while we
 	 * dropped the locks.
 	 */
-	if (!waiter)
+	if (!rt_mutex_real_waiter(waiter))
 		goto out_unlock_pi;
 
 	/*
@@ -1321,6 +1328,22 @@ static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
 		return -EDEADLK;
 
 	raw_spin_lock(&task->pi_lock);
+	/*
+	 * In the case of futex requeue PI, this will be a proxy
+	 * lock. The task will wake unaware that it is enqueueed on
+	 * this lock. Avoid blocking on two locks and corrupting
+	 * pi_blocked_on via the PI_WAKEUP_INPROGRESS
+	 * flag. futex_wait_requeue_pi() sets this when it wakes up
+	 * before requeue (due to a signal or timeout). Do not enqueue
+	 * the task if PI_WAKEUP_INPROGRESS is set.
+	 */
+	if (task != current && task->pi_blocked_on == PI_WAKEUP_INPROGRESS) {
+		raw_spin_unlock(&task->pi_lock);
+		return -EAGAIN;
+	}
+
+       BUG_ON(rt_mutex_real_waiter(task->pi_blocked_on));
+
 	waiter->task = task;
 	waiter->lock = lock;
 	waiter->prio = task->prio;
@@ -1344,7 +1367,7 @@ static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
 		rt_mutex_enqueue_pi(owner, waiter);
 
 		rt_mutex_adjust_prio(owner);
-		if (owner->pi_blocked_on)
+		if (rt_mutex_real_waiter(owner->pi_blocked_on))
 			chain_walk = 1;
 	} else if (rt_mutex_cond_detect_deadlock(waiter, chwalk)) {
 		chain_walk = 1;
@@ -1444,7 +1467,7 @@ static void remove_waiter(struct rt_mutex *lock,
 {
 	bool is_top_waiter = (waiter == rt_mutex_top_waiter(lock));
 	struct task_struct *owner = rt_mutex_owner(lock);
-	struct rt_mutex *next_lock;
+	struct rt_mutex *next_lock = NULL;
 
 	lockdep_assert_held(&lock->wait_lock);
 
@@ -1470,7 +1493,8 @@ static void remove_waiter(struct rt_mutex *lock,
 	rt_mutex_adjust_prio(owner);
 
 	/* Store the lock on which owner is blocked or NULL */
-	next_lock = task_blocked_on_lock(owner);
+	if (rt_mutex_real_waiter(owner->pi_blocked_on))
+		next_lock = task_blocked_on_lock(owner);
 
 	raw_spin_unlock(&owner->pi_lock);
 
@@ -1506,7 +1530,8 @@ void rt_mutex_adjust_pi(struct task_struct *task)
 	raw_spin_lock_irqsave(&task->pi_lock, flags);
 
 	waiter = task->pi_blocked_on;
-	if (!waiter || rt_mutex_waiter_equal(waiter, task_to_waiter(task))) {
+	if (!rt_mutex_real_waiter(waiter) ||
+	    rt_mutex_waiter_equal(waiter, task_to_waiter(task))) {
 		raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 		return;
 	}
@@ -2325,6 +2350,34 @@ int __rt_mutex_start_proxy_lock(struct rt_mutex *lock,
 	if (try_to_take_rt_mutex(lock, task, NULL))
 		return 1;
 
+#ifdef CONFIG_PREEMPT_RT_FULL
+	/*
+	 * In PREEMPT_RT there's an added race.
+	 * If the task, that we are about to requeue, times out,
+	 * it can set the PI_WAKEUP_INPROGRESS. This tells the requeue
+	 * to skip this task. But right after the task sets
+	 * its pi_blocked_on to PI_WAKEUP_INPROGRESS it can then
+	 * block on the spin_lock(&hb->lock), which in RT is an rtmutex.
+	 * This will replace the PI_WAKEUP_INPROGRESS with the actual
+	 * lock that it blocks on. We *must not* place this task
+	 * on this proxy lock in that case.
+	 *
+	 * To prevent this race, we first take the task's pi_lock
+	 * and check if it has updated its pi_blocked_on. If it has,
+	 * we assume that it woke up and we return -EAGAIN.
+	 * Otherwise, we set the task's pi_blocked_on to
+	 * PI_REQUEUE_INPROGRESS, so that if the task is waking up
+	 * it will know that we are in the process of requeuing it.
+	 */
+	raw_spin_lock(&task->pi_lock);
+	if (task->pi_blocked_on) {
+		raw_spin_unlock(&task->pi_lock);
+		return -EAGAIN;
+	}
+	task->pi_blocked_on = PI_REQUEUE_INPROGRESS;
+	raw_spin_unlock(&task->pi_lock);
+#endif
+
 	/* We enforce deadlock detection for futexes */
 	ret = task_blocks_on_rt_mutex(lock, waiter, task,
 				      RT_MUTEX_FULL_CHAINWALK);
diff --git a/kernel/locking/rtmutex_common.h b/kernel/locking/rtmutex_common.h
index 758dc43872e5..546aaf058b9e 100644
--- a/kernel/locking/rtmutex_common.h
+++ b/kernel/locking/rtmutex_common.h
@@ -132,6 +132,9 @@ enum rtmutex_chainwalk {
 /*
  * PI-futex support (proxy locking functions, etc.):
  */
+#define PI_WAKEUP_INPROGRESS	((struct rt_mutex_waiter *) 1)
+#define PI_REQUEUE_INPROGRESS	((struct rt_mutex_waiter *) 2)
+
 extern struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock);
 extern void rt_mutex_init_proxy_locked(struct rt_mutex *lock,
 				       struct task_struct *proxy_owner);
-- 
2.24.1

Powered by blists - more mailing lists