s390/spinlock: add niai spinlock hints

The z14 machine introduces new mode of the next-instruction-access-intent
NIAI instruction. With NIAI-8 it is possible to pin a cache-line on a
CPU for a small amount of time, NIAI-7 releases the cache-line again.
Finally NIAI-4 can be used to prevent the CPU to speculatively access
memory beyond the compare-and-swap instruction to get the lock.

Use these instruction in the spinlock code.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/spinlock.h

@@ -92,10 +92,11 @@ static inline void arch_spin_unlock(arch_spinlock_t *lp)
 {
 	typecheck(int, lp->lock);
 	asm volatile(
-		"st	%1,%0\n"
-		: "+Q" (lp->lock)
-		: "d" (0)
-		: "cc", "memory");
+#ifdef CONFIG_HAVE_MARCH_ZEC12_FEATURES
+		"	.long	0xb2fa0070\n"	/* NIAI 7 */
+#endif
+		"	st	%1,%0\n"
+		: "=Q" (lp->lock) : "d" (0) : "cc", "memory");
 }
 
 static inline void arch_spin_unlock_wait(arch_spinlock_t *lock)

arch/s390/lib/spinlock.c

@@ -32,42 +32,63 @@ static int __init spin_retry_setup(char *str)
 }
 __setup("spin_retry=", spin_retry_setup);
 
+static inline int arch_load_niai4(int *lock)
+{
+	int owner;
+
+	asm volatile(
+#ifdef CONFIG_HAVE_MARCH_ZEC12_FEATURES
+		"	.long	0xb2fa0040\n"	/* NIAI 4 */
+#endif
+		"	l	%0,%1\n"
+		: "=d" (owner) : "Q" (*lock) : "memory");
+       return owner;
+}
+
+static inline int arch_cmpxchg_niai8(int *lock, int old, int new)
+{
+	int expected = old;
+
+	asm volatile(
+#ifdef CONFIG_HAVE_MARCH_ZEC12_FEATURES
+		"	.long	0xb2fa0080\n"	/* NIAI 8 */
+#endif
+		"	cs	%0,%3,%1\n"
+		: "=d" (old), "=Q" (*lock)
+		: "0" (old), "d" (new), "Q" (*lock)
+		: "cc", "memory");
+	return expected == old;
+}
+
 void arch_spin_lock_wait(arch_spinlock_t *lp)
 {
 	int cpu = SPINLOCK_LOCKVAL;
-	int owner, count, first_diag;
+	int owner, count;
+
+	/* Pass the virtual CPU to the lock holder if it is not running */
+	owner = arch_load_niai4(&lp->lock);
+	if (owner && arch_vcpu_is_preempted(~owner))
+		smp_yield_cpu(~owner);
 
-	first_diag = 1;
+	count = spin_retry;
 	while (1) {
-		owner = ACCESS_ONCE(lp->lock);
+		owner = arch_load_niai4(&lp->lock);
 		/* Try to get the lock if it is free. */
 		if (!owner) {
-			if (__atomic_cmpxchg_bool(&lp->lock, 0, cpu))
+			if (arch_cmpxchg_niai8(&lp->lock, 0, cpu))
 				return;
 			continue;
 		}
-		/* First iteration: check if the lock owner is running. */
-		if (first_diag && arch_vcpu_is_preempted(~owner)) {
-			smp_yield_cpu(~owner);
-			first_diag = 0;
+		if (count-- >= 0)
 			continue;
-		}
-		/* Loop for a while on the lock value. */
 		count = spin_retry;
-		do {
-			owner = ACCESS_ONCE(lp->lock);
-		} while (owner && count-- > 0);
-		if (!owner)
-			continue;
 		/*
 		 * For multiple layers of hypervisors, e.g. z/VM + LPAR
 		 * yield the CPU unconditionally. For LPAR rely on the
 		 * sense running status.
 		 */
-		if (!MACHINE_IS_LPAR || arch_vcpu_is_preempted(~owner)) {
+		if (!MACHINE_IS_LPAR || arch_vcpu_is_preempted(~owner))
 			smp_yield_cpu(~owner);
-			first_diag = 0;
-		}
 	}
 }
 EXPORT_SYMBOL(arch_spin_lock_wait);
@@ -75,42 +96,36 @@ EXPORT_SYMBOL(arch_spin_lock_wait);
 void arch_spin_lock_wait_flags(arch_spinlock_t *lp, unsigned long flags)
 {
 	int cpu = SPINLOCK_LOCKVAL;
-	int owner, count, first_diag;
+	int owner, count;
 
 	local_irq_restore(flags);
-	first_diag = 1;
+
+	/* Pass the virtual CPU to the lock holder if it is not running */
+	owner = arch_load_niai4(&lp->lock);
+	if (owner && arch_vcpu_is_preempted(~owner))
+		smp_yield_cpu(~owner);
+
+	count = spin_retry;
 	while (1) {
-		owner = ACCESS_ONCE(lp->lock);
+		owner = arch_load_niai4(&lp->lock);
 		/* Try to get the lock if it is free. */
 		if (!owner) {
 			local_irq_disable();
-			if (__atomic_cmpxchg_bool(&lp->lock, 0, cpu))
+			if (arch_cmpxchg_niai8(&lp->lock, 0, cpu))
 				return;
 			local_irq_restore(flags);
 			continue;
 		}
-		/* Check if the lock owner is running. */
-		if (first_diag && arch_vcpu_is_preempted(~owner)) {
-			smp_yield_cpu(~owner);
-			first_diag = 0;
+		if (count-- >= 0)
 			continue;
-		}
-		/* Loop for a while on the lock value. */
 		count = spin_retry;
-		do {
-			owner = ACCESS_ONCE(lp->lock);
-		} while (owner && count-- > 0);
-		if (!owner)
-			continue;
 		/*
 		 * For multiple layers of hypervisors, e.g. z/VM + LPAR
 		 * yield the CPU unconditionally. For LPAR rely on the
 		 * sense running status.
 		 */
-		if (!MACHINE_IS_LPAR || arch_vcpu_is_preempted(~owner)) {
+		if (!MACHINE_IS_LPAR || arch_vcpu_is_preempted(~owner))
 			smp_yield_cpu(~owner);
-			first_diag = 0;
-		}
 	}
 }
 EXPORT_SYMBOL(arch_spin_lock_wait_flags);