locking/rtmutex: Consolidate the fast/slowpath invocation
authorThomas Gleixner <tglx@linutronix.de>
Fri, 26 Mar 2021 15:29:41 +0000 (16:29 +0100)
committerIngo Molnar <mingo@kernel.org>
Mon, 29 Mar 2021 13:57:04 +0000 (15:57 +0200)
The indirection via a function pointer (which is at least optimized into a
tail call by the compiler) is making the code hard to read.

Clean it up and move the futex related trylock functions down to the futex
section.

Move the wake_q wakeup into rt_mutex_slowunlock(). No point in handing it
to the caller. The futex code uses a different function.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20210326153944.247927548@linutronix.de
kernel/locking/rtmutex.c

index 3612821..7d0c168 100644 (file)
@@ -1299,13 +1299,24 @@ static int __sched rt_mutex_slowtrylock(struct rt_mutex *lock)
 }
 
 /*
+ * Performs the wakeup of the top-waiter and re-enables preemption.
+ */
+void __sched rt_mutex_postunlock(struct wake_q_head *wake_q)
+{
+       wake_up_q(wake_q);
+
+       /* Pairs with preempt_disable() in rt_mutex_slowunlock() */
+       preempt_enable();
+}
+
+/*
  * Slow path to release a rt-mutex.
  *
  * Return whether the current task needs to call rt_mutex_postunlock().
  */
-static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock,
-                                       struct wake_q_head *wake_q)
+static void __sched rt_mutex_slowunlock(struct rt_mutex *lock)
 {
+       DEFINE_WAKE_Q(wake_q);
        unsigned long flags;
 
        /* irqsave required to support early boot calls */
@@ -1347,7 +1358,7 @@ static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock,
        while (!rt_mutex_has_waiters(lock)) {
                /* Drops lock->wait_lock ! */
                if (unlock_rt_mutex_safe(lock, flags) == true)
-                       return false;
+                       return;
                /* Relock the rtmutex and try again */
                raw_spin_lock_irqsave(&lock->wait_lock, flags);
        }
@@ -1358,10 +1369,10 @@ static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock,
         *
         * Queue the next waiter for wakeup once we release the wait_lock.
         */
-       mark_wakeup_next_waiter(wake_q, lock);
+       mark_wakeup_next_waiter(&wake_q, lock);
        raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
 
-       return true; /* call rt_mutex_postunlock() */
+       rt_mutex_postunlock(&wake_q);
 }
 
 /*
@@ -1370,60 +1381,21 @@ static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock,
  * The atomic acquire/release ops are compiled away, when either the
  * architecture does not support cmpxchg or when debugging is enabled.
  */
-static __always_inline int
-rt_mutex_fastlock(struct rt_mutex *lock, int state,
-                 int (*slowfn)(struct rt_mutex *lock, int state,
-                               struct hrtimer_sleeper *timeout,
-                               enum rtmutex_chainwalk chwalk))
+static __always_inline int __rt_mutex_lock(struct rt_mutex *lock, long state,
+                                          unsigned int subclass)
 {
-       if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
-               return 0;
+       int ret;
 
-       return slowfn(lock, state, NULL, RT_MUTEX_MIN_CHAINWALK);
-}
+       might_sleep();
+       mutex_acquire(&lock->dep_map, subclass, 0, _RET_IP_);
 
-static __always_inline int
-rt_mutex_fasttrylock(struct rt_mutex *lock,
-                    int (*slowfn)(struct rt_mutex *lock))
-{
        if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
-               return 1;
-
-       return slowfn(lock);
-}
-
-/*
- * Performs the wakeup of the top-waiter and re-enables preemption.
- */
-void __sched rt_mutex_postunlock(struct wake_q_head *wake_q)
-{
-       wake_up_q(wake_q);
-
-       /* Pairs with preempt_disable() in rt_mutex_slowunlock() */
-       preempt_enable();
-}
-
-static __always_inline void
-rt_mutex_fastunlock(struct rt_mutex *lock,
-                   bool (*slowfn)(struct rt_mutex *lock,
-                                  struct wake_q_head *wqh))
-{
-       DEFINE_WAKE_Q(wake_q);
-
-       if (likely(rt_mutex_cmpxchg_release(lock, current, NULL)))
-               return;
-
-       if (slowfn(lock, &wake_q))
-               rt_mutex_postunlock(&wake_q);
-}
-
-static __always_inline void __rt_mutex_lock(struct rt_mutex *lock,
-                                           unsigned int subclass)
-{
-       might_sleep();
+               return 0;
 
-       mutex_acquire(&lock->dep_map, subclass, 0, _RET_IP_);
-       rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, rt_mutex_slowlock);
+       ret = rt_mutex_slowlock(lock, state, NULL, RT_MUTEX_MIN_CHAINWALK);
+       if (ret)
+               mutex_release(&lock->dep_map, _RET_IP_);
+       return ret;
 }
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
@@ -1435,7 +1407,7 @@ static __always_inline void __rt_mutex_lock(struct rt_mutex *lock,
  */
 void __sched rt_mutex_lock_nested(struct rt_mutex *lock, unsigned int subclass)
 {
-       __rt_mutex_lock(lock, subclass);
+       __rt_mutex_lock(lock, TASK_UNINTERRUPTIBLE, subclass);
 }
 EXPORT_SYMBOL_GPL(rt_mutex_lock_nested);
 
@@ -1448,7 +1420,7 @@ EXPORT_SYMBOL_GPL(rt_mutex_lock_nested);
  */
 void __sched rt_mutex_lock(struct rt_mutex *lock)
 {
-       __rt_mutex_lock(lock, 0);
+       __rt_mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0);
 }
 EXPORT_SYMBOL_GPL(rt_mutex_lock);
 #endif
@@ -1464,42 +1436,21 @@ EXPORT_SYMBOL_GPL(rt_mutex_lock);
  */
 int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock)
 {
-       int ret;
-
-       might_sleep();
-
-       mutex_acquire(&lock->dep_map, 0, 0, _RET_IP_);
-       ret = rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE, rt_mutex_slowlock);
-       if (ret)
-               mutex_release(&lock->dep_map, _RET_IP_);
-
-       return ret;
+       return __rt_mutex_lock(lock, TASK_INTERRUPTIBLE, 0);
 }
 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
 
-/*
- * Futex variant, must not use fastpath.
- */
-int __sched rt_mutex_futex_trylock(struct rt_mutex *lock)
-{
-       return rt_mutex_slowtrylock(lock);
-}
-
-int __sched __rt_mutex_futex_trylock(struct rt_mutex *lock)
-{
-       return __rt_mutex_slowtrylock(lock);
-}
-
 /**
  * rt_mutex_trylock - try to lock a rt_mutex
  *
  * @lock:      the rt_mutex to be locked
  *
- * This function can only be called in thread context. It's safe to
- * call it from atomic regions, but not from hard interrupt or soft
- * interrupt context.
+ * This function can only be called in thread context. It's safe to call it
+ * from atomic regions, but not from hard or soft interrupt context.
  *
- * Returns 1 on success and 0 on contention
+ * Returns:
+ *  1 on success
+ *  0 on contention
  */
 int __sched rt_mutex_trylock(struct rt_mutex *lock)
 {
@@ -1508,7 +1459,14 @@ int __sched rt_mutex_trylock(struct rt_mutex *lock)
        if (WARN_ON_ONCE(in_irq() || in_nmi() || in_serving_softirq()))
                return 0;
 
-       ret = rt_mutex_fasttrylock(lock, rt_mutex_slowtrylock);
+       /*
+        * No lockdep annotation required because lockdep disables the fast
+        * path.
+        */
+       if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
+               return 1;
+
+       ret = rt_mutex_slowtrylock(lock);
        if (ret)
                mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
 
@@ -1524,10 +1482,26 @@ EXPORT_SYMBOL_GPL(rt_mutex_trylock);
 void __sched rt_mutex_unlock(struct rt_mutex *lock)
 {
        mutex_release(&lock->dep_map, _RET_IP_);
-       rt_mutex_fastunlock(lock, rt_mutex_slowunlock);
+       if (likely(rt_mutex_cmpxchg_release(lock, current, NULL)))
+               return;
+
+       rt_mutex_slowunlock(lock);
 }
 EXPORT_SYMBOL_GPL(rt_mutex_unlock);
 
+/*
+ * Futex variants, must not use fastpath.
+ */
+int __sched rt_mutex_futex_trylock(struct rt_mutex *lock)
+{
+       return rt_mutex_slowtrylock(lock);
+}
+
+int __sched __rt_mutex_futex_trylock(struct rt_mutex *lock)
+{
+       return __rt_mutex_slowtrylock(lock);
+}
+
 /**
  * __rt_mutex_futex_unlock - Futex variant, that since futex variants
  * do not use the fast-path, can be simple and will not need to retry.