[patch V5 67/72] rtmutex: Prevent lockdep false positive with PI futexes

From: Thomas Gleixner
Date: Sun Aug 15 2021 - 17:32:39 EST


From: Thomas Gleixner <tglx@xxxxxxxxxxxxx>

On PREEMPT_RT the futex hashbucket spinlock becomes 'sleeping' and rtmutex
based. That causes a lockdep false positive because some of the futex
functions invoke spin_unlock(&hb->lock) with the wait_lock of the rtmutex
associated to the pi_futex held. spin_unlock() in turn takes wait_lock of
the rtmutex on which the spinlock is based which makes lockdep notice a
lock recursion.

Give the futex/rtmutex wait_lock a separate key.

Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
---
kernel/locking/rtmutex_api.c | 12 ++++++++++++
1 file changed, 12 insertions(+)
---
--- a/kernel/locking/rtmutex_api.c
+++ b/kernel/locking/rtmutex_api.c
@@ -214,7 +214,19 @@ EXPORT_SYMBOL_GPL(__rt_mutex_init);
void __sched rt_mutex_init_proxy_locked(struct rt_mutex_base *lock,
struct task_struct *proxy_owner)
{
+ static struct lock_class_key pi_futex_key;
+
__rt_mutex_base_init(lock);
+ /*
+ * On PREEMPT_RT the futex hashbucket spinlock becomes 'sleeping'
+ * and rtmutex based. That causes a lockdep false positive because
+ * some of the futex functions invoke spin_unlock(&hb->lock) with
+ * the wait_lock of the rtmutex associated to the pi_futex held.
+ * spin_unlock() in turn takes wait_lock of the rtmutex on which
+ * the spinlock is based which makes lockdep notice a lock
+ * recursion. Give the futex/rtmutex wait_lock a separate key.
+ */
+ lockdep_set_class(&lock->wait_lock, &pi_futex_key);
rt_mutex_set_owner(lock, proxy_owner);
}