locking/mutex: Optimize mutex_trylock() fast-path

A while back Viro posted a number of 'interesting' mutex_is_locked()
users on IRC, one of those was RCU.

RCU seems to use mutex_is_locked() to avoid doing mutex_trylock(), the
regular load before modify pattern.

While the use isn't wrong per se, its curious in that its needed at all,
mutex_trylock() should be good enough on its own to avoid the pointless
cacheline bounces.

So fix those and remove the mutex_is_locked() (ab)use from RCU.

Reported-by: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Paul McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Davidlohr Bueso <dave@stgolabs.net>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <Waiman.Long@hpe.com>
Link: http://lkml.kernel.org/r/20160601185815.GW3190@twins.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>

arch/ia64/include/asm/mutex.h

@@ -82,7 +82,7 @@ __mutex_fastpath_unlock(atomic_t *count, void (*fail_fn)(atomic_t *))
 static inline int
 __mutex_fastpath_trylock(atomic_t *count, int (*fail_fn)(atomic_t *))
 {
-	if (cmpxchg_acq(count, 1, 0) == 1)
+	if (atomic_read(count) == 1 && cmpxchg_acq(count, 1, 0) == 1)
 		return 1;
 	return 0;
 }

arch/powerpc/include/asm/mutex.h

@@ -124,7 +124,7 @@ __mutex_fastpath_unlock(atomic_t *count, void (*fail_fn)(atomic_t *))
 static inline int
 __mutex_fastpath_trylock(atomic_t *count, int (*fail_fn)(atomic_t *))
 {
-	if (likely(__mutex_cmpxchg_lock(count, 1, 0) == 1))
+	if (likely(atomic_read(count) == 1 && __mutex_cmpxchg_lock(count, 1, 0) == 1))
 		return 1;
 	return 0;
 }

arch/x86/include/asm/mutex_32.h

@@ -101,7 +101,7 @@ static inline int __mutex_fastpath_trylock(atomic_t *count,
 					   int (*fail_fn)(atomic_t *))
 {
 	/* cmpxchg because it never induces a false contention state. */
-	if (likely(atomic_cmpxchg(count, 1, 0) == 1))
+	if (likely(atomic_read(count) == 1 && atomic_cmpxchg(count, 1, 0) == 1))
 		return 1;
 
 	return 0;

arch/x86/include/asm/mutex_64.h

@@ -118,10 +118,10 @@ do {								\
 static inline int __mutex_fastpath_trylock(atomic_t *count,
 					   int (*fail_fn)(atomic_t *))
 {
-	if (likely(atomic_cmpxchg(count, 1, 0) == 1))
+	if (likely(atomic_read(count) == 1 && atomic_cmpxchg(count, 1, 0) == 1))
 		return 1;
-	else
-		return 0;
+
+	return 0;
 }
 
 #endif /* _ASM_X86_MUTEX_64_H */

include/asm-generic/mutex-dec.h

@@ -80,7 +80,7 @@ __mutex_fastpath_unlock(atomic_t *count, void (*fail_fn)(atomic_t *))
 static inline int
 __mutex_fastpath_trylock(atomic_t *count, int (*fail_fn)(atomic_t *))
 {
-	if (likely(atomic_cmpxchg_acquire(count, 1, 0) == 1))
+	if (likely(atomic_read(count) == 1 && atomic_cmpxchg_acquire(count, 1, 0) == 1))
 		return 1;
 	return 0;
 }

include/asm-generic/mutex-xchg.h

@@ -91,8 +91,12 @@ __mutex_fastpath_unlock(atomic_t *count, void (*fail_fn)(atomic_t *))
 static inline int
 __mutex_fastpath_trylock(atomic_t *count, int (*fail_fn)(atomic_t *))
 {
-	int prev = atomic_xchg_acquire(count, 0);
+	int prev;
 
+	if (atomic_read(count) != 1)
+		return 0;
+
+	prev = atomic_xchg_acquire(count, 0);
 	if (unlikely(prev < 0)) {
 		/*
 		 * The lock was marked contended so we must restore that

kernel/rcu/tree.c

@@ -3681,7 +3681,6 @@ static bool exp_funnel_lock(struct rcu_state *rsp, unsigned long s)
 	if (ULONG_CMP_LT(READ_ONCE(rnp->exp_seq_rq), s) &&
 	    (rnp == rnp_root ||
 	     ULONG_CMP_LT(READ_ONCE(rnp_root->exp_seq_rq), s)) &&
-	    !mutex_is_locked(&rsp->exp_mutex) &&
 	    mutex_trylock(&rsp->exp_mutex))
 		goto fastpath;