[PATCH v3 6/6] powerpc/qspinlock: optimised atomic_try_cmpxchg_lock that adds the lock hint

[Nicholas Piggin <npiggin@gmail.com>]

This brings the behaviour of the uncontended fast path back to
roughly equivalent to simple spinlocks -- a single atomic op with
lock hint.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
---
 arch/powerpc/include/asm/atomic.h    | 28 ++++++++++++++++++++++++++++
 arch/powerpc/include/asm/qspinlock.h |  2 +-
 2 files changed, 29 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/include/asm/atomic.h b/arch/powerpc/include/asm/atomic.h
index 498785ffc25f..f6a3d145ffb7 100644
--- a/arch/powerpc/include/asm/atomic.h
+++ b/arch/powerpc/include/asm/atomic.h
@@ -193,6 +193,34 @@ static __inline__ int atomic_dec_return_relaxed(atomic_t *v)
 #define atomic_xchg(v, new) (xchg(&((v)->counter), new))
 #define atomic_xchg_relaxed(v, new) xchg_relaxed(&((v)->counter), (new))
 
+/*
+ * Don't want to override the generic atomic_try_cmpxchg_acquire, because
+ * we add a lock hint to the lwarx, which may not be wanted for the
+ * _acquire case (and is not used by the other _acquire variants so it
+ * would be a surprise).
+ */
+static __always_inline bool
+atomic_try_cmpxchg_lock(atomic_t *v, int *old, int new)
+{
+	int r, o = *old;
+
+	__asm__ __volatile__ (
+"1:\t"	PPC_LWARX(%0,0,%2,1) "	# atomic_try_cmpxchg_acquire	\n"
+"	cmpw	0,%0,%3							\n"
+"	bne-	2f							\n"
+"	stwcx.	%4,0,%2							\n"
+"	bne-	1b							\n"
+"\t"	PPC_ACQUIRE_BARRIER "						\n"
+"2:									\n"
+	: "=&r" (r), "+m" (v->counter)
+	: "r" (&v->counter), "r" (o), "r" (new)
+	: "cr0", "memory");
+
+	if (unlikely(r != o))
+		*old = r;
+	return likely(r == o);
+}
+
 /**
  * atomic_fetch_add_unless - add unless the number is a given value
  * @v: pointer of type atomic_t
diff --git a/arch/powerpc/include/asm/qspinlock.h b/arch/powerpc/include/asm/qspinlock.h
index f5066f00a08c..b752d34517b3 100644
--- a/arch/powerpc/include/asm/qspinlock.h
+++ b/arch/powerpc/include/asm/qspinlock.h
@@ -37,7 +37,7 @@ static __always_inline void queued_spin_lock(struct qspinlock *lock)
 {
 	u32 val = 0;
 
-	if (likely(atomic_try_cmpxchg_acquire(&lock->val, &val, _Q_LOCKED_VAL)))
+	if (likely(atomic_try_cmpxchg_lock(&lock->val, &val, _Q_LOCKED_VAL)))
 		return;
 
 	queued_spin_lock_slowpath(lock, val);
-- 
2.23.0