Use Linux atomic primitives by default.

Previously Solaris style atomic primitives were implemented simply by
wrapping the desired operation in a global spinlock.  This was easy to
implement at the time when I wasn't 100% sure I could safely layer the
Solaris atomic primatives on the Linux counterparts.  It however was
likely not good for performance.

After more investigation however it does appear the Solaris primitives
can be layered on Linux's fairly safely.  The Linux atomic_t type really
just wraps a long so we can simply cast the Solaris unsigned value to
either a atomic_t or atomic64_t.  The only lingering problem for both
implementations is that Solaris provides no atomic read function.  This
means reading a 64-bit value on a 32-bit arch can (and will) result in
word breaking.  I was very concerned about this initially, but upon
further reflection it is a limitation of the Solaris API.  So really
we are just being bug-for-bug compatible here.

With this change the default implementation is layered on top of Linux
atomic types.  However, because we're assuming a lot about the internal
implementation of those types I've made it easy to fall-back to the
generic approach.  Simply build with --enable-atomic_spinlocks if
issues are encountered with the new implementation.

1 files changed, 97 insertions, 49 deletions

diff --git a/include/sys/atomic.h b/include/sys/atomic.h
index cd0eb3b0a..4f4a1e058 100644
--- a/include/sys/atomic.h
+++ b/include/sys/atomic.h
@@ -27,31 +27,94 @@
 #ifndef _SPL_ATOMIC_H
 #define _SPL_ATOMIC_H
 
-#ifdef  __cplusplus
-extern "C" {
-#endif
-
 #include <linux/module.h>
 #include <linux/spinlock.h>
-#include <sys/isa_defs.h>
+#include <sys/types.h>
 
-/* XXX: Serialize everything through global locks.  This is
- * going to be bad for performance, but for now it's the easiest
- * way to ensure correct behavior.  I don't like it at all.
- * It would be nicer to make these function to the atomic linux
- * functions, but the normal uint64_t type complicates this.
+/*
+ * Two approaches to atomic operations are implemented each with its
+ * own benefits are drawbacks imposed by the Solaris API.  Neither
+ * approach handles the issue of word breaking when using a 64-bit
+ * atomic variable on a 32-bit arch.  The Solaris API would need to
+ * add an atomic read call to correctly support this.
+ *
+ * When ATOMIC_SPINLOCK is defined all atomic operations will be
+ * serialized through global spin locks.  This is bad for performance
+ * but it does allow a simple generic implementation.
+ *
+ * When ATOMIC_SPINLOCK is not defined the Linux atomic operations
+ * are used.  This is safe as long as the core Linux implementation
+ * doesn't change because we are relying on the fact that an atomic
+ * type is really just a uint32 or uint64.  If this changes at some
+ * point in the future we need to fall-back to the spin approach.
  */
-extern spinlock_t atomic64_lock;
+#ifdef ATOMIC_SPINLOCK
 extern spinlock_t atomic32_lock;
+extern spinlock_t atomic64_lock;
 
-static __inline__ uint32_t
+static __inline__ void
+atomic_inc_32(volatile uint32_t *target)
+{
+	spin_lock(&atomic32_lock);
+	(*target)++;
+	spin_unlock(&atomic32_lock);
+}
+
+static __inline__ void
+atomic_dec_32(volatile uint32_t *target)
+{
+	spin_lock(&atomic32_lock);
+	(*target)--;
+	spin_unlock(&atomic32_lock);
+}
+
+static __inline__ void
 atomic_add_32(volatile uint32_t *target, int32_t delta)
 {
+	spin_lock(&atomic32_lock);
+	*target += delta;
+	spin_unlock(&atomic32_lock);
+}
+
+static __inline__ void
+atomic_sub_32(volatile uint32_t *target, int32_t delta)
+{
+	spin_lock(&atomic32_lock);
+	*target -= delta;
+	spin_unlock(&atomic32_lock);
+}
+
+static __inline__ uint32_t
+atomic_add_32_nv(volatile uint32_t *target, uint32_t delta)
+{
+	spin_lock(&atomic32_lock);
+	*target += delta;
+	spin_unlock(&atomic32_lock);
+
+	return *target;
+}
+
+static __inline__ uint32_t
+atomic_sub_32_nv(volatile uint32_t *target, uint32_t delta)
+{
+	spin_lock(&atomic32_lock);
+	*target -= delta;
+	spin_unlock(&atomic32_lock);
+
+	return *target;
+}
+
+static __inline__ uint32_t
+atomic_cas_32(volatile uint32_t *target,  uint32_t cmp,
+              uint32_t newval)
+{
 	uint32_t rc;
 
 	spin_lock(&atomic32_lock);
 	rc = *target;
-	*target += delta;
+	if (*target == cmp)
+		*target = newval;
+
 	spin_unlock(&atomic32_lock);
 
 	return rc;
@@ -73,30 +136,20 @@ atomic_dec_64(volatile uint64_t *target)
 	spin_unlock(&atomic64_lock);
 }
 
-static __inline__ uint64_t
+static __inline__ void
 atomic_add_64(volatile uint64_t *target, uint64_t delta)
 {
-	uint64_t rc;
-
 	spin_lock(&atomic64_lock);
-	rc = *target;
 	*target += delta;
 	spin_unlock(&atomic64_lock);
-
-	return rc;
 }
 
-static __inline__ uint64_t
+static __inline__ void
 atomic_sub_64(volatile uint64_t *target, uint64_t delta)
 {
-	uint64_t rc;
-
 	spin_lock(&atomic64_lock);
-	rc = *target;
 	*target -= delta;
 	spin_unlock(&atomic64_lock);
-
-	return rc;
 }
 
 static __inline__ uint64_t
@@ -121,7 +174,7 @@ atomic_sub_64_nv(volatile uint64_t *target, uint64_t delta)
 
 static __inline__ uint64_t
 atomic_cas_64(volatile uint64_t *target,  uint64_t cmp,
-               uint64_t newval)
+              uint64_t newval)
 {
 	uint64_t rc;
 
@@ -134,45 +187,40 @@ atomic_cas_64(volatile uint64_t *target,  uint64_t cmp,
 	return rc;
 }
 
-static __inline__ uint32_t
-atomic_cas_32(volatile uint32_t *target,  uint32_t cmp,
-               uint32_t newval)
-{
-	uint32_t rc;
 
-	spin_lock(&atomic32_lock);
-	rc = *target;
-	if (*target == cmp)
-		*target = newval;
+#else /* ATOMIC_SPINLOCK */
 
-	spin_unlock(&atomic32_lock);
+#define atomic_inc_32(v)	atomic_inc((atomic_t *)(v))
+#define atomic_dec_32(v)	atomic_dec((atomic_t *)(v))
+#define atomic_add_32(v, i)	atomic_add((i), (atomic_t *)(v))
+#define atomic_sub_32(v, i)	atomic_sub((i), (atomic_t *)(v))
+#define atomic_add_32_nv(v, i)	atomic_add_return((i), (atomic_t *)(v))
+#define atomic_sub_32_nv(v, i)	atomic_sub_return((i), (atomic_t *)(v))
+#define atomic_cas_32(v, x, y)	atomic_cmpxchg((atomic_t *)(v), x, y)
+#define atomic_inc_64(v)	atomic64_inc((atomic64_t *)(v))
+#define atomic_dec_64(v)	atomic64_dec((atomic64_t *)(v))
+#define atomic_add_64(v, i)	atomic64_add((i), (atomic64_t *)(v))
+#define atomic_sub_64(v, i)	atomic64_sub((i), (atomic64_t *)(v))
+#define atomic_add_64_nv(v, i)	atomic64_add_return((i), (atomic64_t *)(v))
+#define atomic_sub_64_nv(v, i)	atomic64_sub_return((i), (atomic64_t *)(v))
+#define atomic_cas_64(v, x, y)	atomic64_cmpxchg((atomic64_t *)(v), x, y)
 
-	return rc;
-}
+#endif /* ATOMIC_SPINLOCK */
 
 #ifdef _LP64
-/* XXX: Implement atomic_cas_ptr() in terms of uint64'ts.  This
- * is of course only safe and correct for 64 bit arches...  but
- * for now I'm OK with that.
- */
 static __inline__ void *
 atomic_cas_ptr(volatile void *target,  void *cmp, void *newval)
 {
 	return (void *)atomic_cas_64((volatile uint64_t *)target,
 	                             (uint64_t)cmp, (uint64_t)newval);
 }
-#else
+#else /* _LP64 */
 static __inline__ void *
 atomic_cas_ptr(volatile void *target,  void *cmp, void *newval)
 {
 	return (void *)atomic_cas_32((volatile uint32_t *)target,
 	                             (uint32_t)cmp, (uint32_t)newval);
 }
-#endif
-
-#ifdef  __cplusplus
-}
-#endif
+#endif /* _LP64 */
 
 #endif  /* _SPL_ATOMIC_H */
-