[SPARC64]: Implement atomic backoff.

When the cpu count is high and contention hits an atomic object, the
processors can synchronize such that some cpus continually get knocked
out and cannot complete the atomic update.

So implement an exponential backoff when SMP.

Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/arch/sparc64/lib/atomic.S b/arch/sparc64/lib/atomic.S
index 9633750..70ac418 100644
--- a/arch/sparc64/lib/atomic.S
+++ b/arch/sparc64/lib/atomic.S
@@ -1,10 +1,10 @@
-/* $Id: atomic.S,v 1.4 2001/11/18 00:12:56 davem Exp $
- * atomic.S: These things are too big to do inline.
+/* atomic.S: These things are too big to do inline.
  *
- * Copyright (C) 1999 David S. Miller (davem@redhat.com)
+ * Copyright (C) 1999, 2007 David S. Miller (davem@davemloft.net)
  */
 
 #include <asm/asi.h>
+#include <asm/backoff.h>
 
 	.text
 
@@ -16,27 +16,31 @@
 	.globl	atomic_add
 	.type	atomic_add,#function
 atomic_add: /* %o0 = increment, %o1 = atomic_ptr */
+	BACKOFF_SETUP(%o2)
 1:	lduw	[%o1], %g1
 	add	%g1, %o0, %g7
 	cas	[%o1], %g1, %g7
 	cmp	%g1, %g7
-	bne,pn	%icc, 1b
+	bne,pn	%icc, 2f
 	 nop
 	retl
 	 nop
+2:	BACKOFF_SPIN(%o2, %o3, 1b)
 	.size	atomic_add, .-atomic_add
 
 	.globl	atomic_sub
 	.type	atomic_sub,#function
 atomic_sub: /* %o0 = decrement, %o1 = atomic_ptr */
+	BACKOFF_SETUP(%o2)
 1:	lduw	[%o1], %g1
 	sub	%g1, %o0, %g7
 	cas	[%o1], %g1, %g7
 	cmp	%g1, %g7
-	bne,pn	%icc, 1b
+	bne,pn	%icc, 2f
 	 nop
 	retl
 	 nop
+2:	BACKOFF_SPIN(%o2, %o3, 1b)
 	.size	atomic_sub, .-atomic_sub
 
 	/* On SMP we need to use memory barriers to ensure
@@ -60,89 +64,101 @@
 	.globl	atomic_add_ret
 	.type	atomic_add_ret,#function
 atomic_add_ret: /* %o0 = increment, %o1 = atomic_ptr */
+	BACKOFF_SETUP(%o2)
 	ATOMIC_PRE_BARRIER
 1:	lduw	[%o1], %g1
 	add	%g1, %o0, %g7
 	cas	[%o1], %g1, %g7
 	cmp	%g1, %g7
-	bne,pn	%icc, 1b
+	bne,pn	%icc, 2f
 	 add	%g7, %o0, %g7
 	sra	%g7, 0, %o0
 	ATOMIC_POST_BARRIER
 	retl
 	 nop
+2:	BACKOFF_SPIN(%o2, %o3, 1b)
 	.size	atomic_add_ret, .-atomic_add_ret
 
 	.globl	atomic_sub_ret
 	.type	atomic_sub_ret,#function
 atomic_sub_ret: /* %o0 = decrement, %o1 = atomic_ptr */
+	BACKOFF_SETUP(%o2)
 	ATOMIC_PRE_BARRIER
 1:	lduw	[%o1], %g1
 	sub	%g1, %o0, %g7
 	cas	[%o1], %g1, %g7
 	cmp	%g1, %g7
-	bne,pn	%icc, 1b
+	bne,pn	%icc, 2f
 	 sub	%g7, %o0, %g7
 	sra	%g7, 0, %o0
 	ATOMIC_POST_BARRIER
 	retl
 	 nop
+2:	BACKOFF_SPIN(%o2, %o3, 1b)
 	.size	atomic_sub_ret, .-atomic_sub_ret
 
 	.globl	atomic64_add
 	.type	atomic64_add,#function
 atomic64_add: /* %o0 = increment, %o1 = atomic_ptr */
+	BACKOFF_SETUP(%o2)
 1:	ldx	[%o1], %g1
 	add	%g1, %o0, %g7
 	casx	[%o1], %g1, %g7
 	cmp	%g1, %g7
-	bne,pn	%xcc, 1b
+	bne,pn	%xcc, 2f
 	 nop
 	retl
 	 nop
+2:	BACKOFF_SPIN(%o2, %o3, 1b)
 	.size	atomic64_add, .-atomic64_add
 
 	.globl	atomic64_sub
 	.type	atomic64_sub,#function
 atomic64_sub: /* %o0 = decrement, %o1 = atomic_ptr */
+	BACKOFF_SETUP(%o2)
 1:	ldx	[%o1], %g1
 	sub	%g1, %o0, %g7
 	casx	[%o1], %g1, %g7
 	cmp	%g1, %g7
-	bne,pn	%xcc, 1b
+	bne,pn	%xcc, 2f
 	 nop
 	retl
 	 nop
+2:	BACKOFF_SPIN(%o2, %o3, 1b)
 	.size	atomic64_sub, .-atomic64_sub
 
 	.globl	atomic64_add_ret
 	.type	atomic64_add_ret,#function
 atomic64_add_ret: /* %o0 = increment, %o1 = atomic_ptr */
+	BACKOFF_SETUP(%o2)
 	ATOMIC_PRE_BARRIER
 1:	ldx	[%o1], %g1
 	add	%g1, %o0, %g7
 	casx	[%o1], %g1, %g7
 	cmp	%g1, %g7
-	bne,pn	%xcc, 1b
+	bne,pn	%xcc, 2f
 	 add	%g7, %o0, %g7
 	mov	%g7, %o0
 	ATOMIC_POST_BARRIER
 	retl
 	 nop
+2:	BACKOFF_SPIN(%o2, %o3, 1b)
 	.size	atomic64_add_ret, .-atomic64_add_ret
 
 	.globl	atomic64_sub_ret
 	.type	atomic64_sub_ret,#function
 atomic64_sub_ret: /* %o0 = decrement, %o1 = atomic_ptr */
+	BACKOFF_SETUP(%o2)
 	ATOMIC_PRE_BARRIER
 1:	ldx	[%o1], %g1
 	sub	%g1, %o0, %g7
 	casx	[%o1], %g1, %g7
 	cmp	%g1, %g7
-	bne,pn	%xcc, 1b
+	bne,pn	%xcc, 2f
 	 sub	%g7, %o0, %g7
 	mov	%g7, %o0
 	ATOMIC_POST_BARRIER
 	retl
 	 nop
+2:	BACKOFF_SPIN(%o2, %o3, 1b)
 	.size	atomic64_sub_ret, .-atomic64_sub_ret