From: Thomas Gleixner <tglx@linutronix.de>
Date: Fri, 10 Jun 2011 11:04:15 +0200
Subject: rtmutex: Handle the various new futex race conditions
Git-repo: git://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-rt-devel.git
Git-commit: d9b3f7595591db358cbe7af490f1a8718957d0de
Patch-mainline: Queued in subsystem maintainer repository
References: SLE Realtime Extension
RT opens a few new interesting race conditions in the rtmutex/futex
combo due to futex hash bucket lock being a 'sleeping' spinlock and
therefor not disabling preemption.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Mike Galbraith <mgalbraith@suse.de>
---
kernel/futex.c | 77 ++++++++++++++++++++++++++++++++--------
kernel/locking/rtmutex.c | 37 ++++++++++++++++---
kernel/locking/rtmutex_common.h | 2 +
3 files changed, 95 insertions(+), 21 deletions(-)
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -2147,6 +2147,16 @@ retry_private:
*/
requeue_pi_wake_futex(this, &key2, hb2);
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
@@ -3202,7 +3212,7 @@ static int futex_wait_requeue_pi(u32 __u
{
struct hrtimer_sleeper timeout, *to = 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;
@@ -3258,20 +3268,55 @@ static int futex_wait_requeue_pi(u32 __u
/* Queue the futex_q, drop the hb lock, wait for wakeup. */
futex_wait_queue_me(hb, &q, to);
- spin_lock(&hb->lock);
- ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
- 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(¤t->pi_lock);
+ if (current->pi_blocked_on) {
+ /*
+ * We have been requeued or are in the process of
+ * being requeued.
+ */
+ raw_spin_unlock_irq(¤t->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(¤t->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(¤t->pi_lock);
+ current->pi_blocked_on = NULL;
+ raw_spin_unlock_irq(¤t->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) {
@@ -3280,14 +3325,15 @@ static int futex_wait_requeue_pi(u32 __u
* did a lock-steal - fix up the PI-state in that case.
*/
if (q.pi_state && (q.pi_state->owner != current)) {
- spin_lock(q.lock_ptr);
+ spin_lock(&hb2->lock);
+ BUG_ON(&hb2->lock != q.lock_ptr);
ret = fixup_pi_state_owner(uaddr2, &q, current);
/*
* Drop the reference to the pi state which
* the requeue_pi() code acquired for us.
*/
put_pi_state(q.pi_state);
- spin_unlock(q.lock_ptr);
+ spin_unlock(&hb2->lock);
/*
* Adjust the return value. It's either -EFAULT or
* success (1) but the caller expects 0 for success.
@@ -3306,7 +3352,8 @@ static int futex_wait_requeue_pi(u32 __u
pi_mutex = &q.pi_state->pi_mutex;
ret = rt_mutex_wait_proxy_lock(pi_mutex, to, &rt_waiter);
- 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;
--- a/kernel/locking/rtmutex.c
+++ b/kernel/locking/rtmutex.c
@@ -135,6 +135,11 @@ static void fixup_rt_mutex_waiters(struc
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;
+}
+
/*
* We can speed up the acquire/release, if there's no debugging state to be
* set up.
@@ -379,7 +384,8 @@ 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;
}
/*
@@ -515,7 +521,7 @@ static int rt_mutex_adjust_prio_chain(st
* 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;
/*
@@ -951,6 +957,23 @@ static int task_blocks_on_rt_mutex(struc
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;
@@ -974,7 +997,7 @@ static int task_blocks_on_rt_mutex(struc
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;
@@ -1070,7 +1093,7 @@ static void remove_waiter(struct rt_mute
{
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);
@@ -1096,7 +1119,8 @@ static void remove_waiter(struct rt_mute
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);
@@ -1132,7 +1156,8 @@ void rt_mutex_adjust_pi(struct task_stru
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;
}
--- a/kernel/locking/rtmutex_common.h
+++ b/kernel/locking/rtmutex_common.h
@@ -99,6 +99,8 @@ enum rtmutex_chainwalk {
/*
* PI-futex support (proxy locking functions, etc.):
*/
+#define PI_WAKEUP_INPROGRESS ((struct rt_mutex_waiter *) 1)
+
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);