Xchg on x86 implies LOCK semantics and therefore is slow. Approximate it
using this_cpu_cmpxchg().
Signed-off-by: Christoph Lameter <[email protected]>
---
arch/x86/include/asm/percpu.h | 21 +++++++++++++++++++++
1 file changed, 21 insertions(+)
Index: linux-2.6/arch/x86/include/asm/percpu.h
===================================================================
--- linux-2.6.orig/arch/x86/include/asm/percpu.h 2009-12-18 15:49:09.000000000 -0600
+++ linux-2.6/arch/x86/include/asm/percpu.h 2009-12-18 15:50:12.000000000 -0600
@@ -156,6 +156,15 @@ do { \
} \
})
+#define this_cpu_xchg_x86(var, new) \
+({ __typeof(*var) __tmp_var,__tmp_old; \
+ do { \
+ __tmp_old = __this_cpu_read(var); \
+ __tmp_var = __this_cpu_cmpxchg(var, __tmp_old, new); \
+ } while (__tmp_var != __tmp_old); \
+ __tmp_old; \
+})
+
/*
* percpu_read() makes gcc load the percpu variable every time it is
* accessed while percpu_read_stable() allows the value to be cached.
@@ -204,6 +213,9 @@ do { \
#define __this_cpu_cmpxchg_1(pcp, old,new) cmpxchg_local(__this_cpu_ptr(pcp), old, new)
#define __this_cpu_cmpxchg_2(pcp, old,new) cmpxchg_local(__this_cpu_ptr(pcp), old, new)
#define __this_cpu_cmpxchg_4(pcp, old,new) cmpxchg_local(__this_cpu_ptr(pcp), old, new)
+#define __this_cpu_xchg_1(pcp, new) this_cpu_xchg_x86((pcp), new)
+#define __this_cpu_xchg_2(pcp, new) this_cpu_xchg_x86((pcp), new)
+#define __this_cpu_xchg_4(pcp, new) this_cpu_xchg_x86((pcp), new)
#define this_cpu_read_1(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
#define this_cpu_read_2(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
@@ -232,6 +244,9 @@ do { \
#define this_cpu_cmpxchg_1(pcp, old,new) cmpxchg_local(this_cpu_ptr(pcp), old, new)
#define this_cpu_cmpxchg_2(pcp, old,new) cmpxchg_local(this_cpu_ptr(pcp), old, new)
#define this_cpu_cmpxchg_4(pcp, old,new) cmpxchg_local(this_cpu_ptr(pcp), old, new)
+#define this_cpu_xchg_1(pcp, new) this_cpu_xchg_x86((pcp), new)
+#define this_cpu_xchg_2(pcp, new) this_cpu_xchg_x86((pcp), new)
+#define this_cpu_xchg_4(pcp, new) this_cpu_xchg_x86((pcp), new)
#define irqsafe_cpu_add_1(pcp, val) percpu_to_op("add", (pcp), val)
#define irqsafe_cpu_add_2(pcp, val) percpu_to_op("add", (pcp), val)
@@ -254,6 +269,9 @@ do { \
#define irqsafe_cpu_cmpxchg_1(pcp, old,new) cmpxchg_local(this_cpu_ptr(pcp), old, new)
#define irqsafe_cpu_cmpxchg_2(pcp, old,new) cmpxchg_local(this_cpu_ptr(pcp), old, new)
#define irqsafe_cpu_cmpxchg_4(pcp, old,new) cmpxchg_local(this_cpu_ptr(pcp), old, new)
+#define irqsafe_cpu_xchg_1(pcp, new) this_cpu_xchg_x86((pcp), new)
+#define irqsafe_cpu_xchg_2(pcp, new) this_cpu_xchg_x86((pcp), new)
+#define irqsafe_cpu_xchg_4(pcp, new) this_cpu_xchg_x86((pcp), new)
/*
* Per cpu atomic 64 bit operations are only available under 64 bit.
@@ -269,6 +287,7 @@ do { \
#define __this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
#define __this_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val)
#define __this_cpu_cmpxchg_8(pcp, old,new) cmpxchg_local(__this_cpu_ptr(pcp), old, new)
+#define __this_cpu_xchg_8(pcp, new) this_cpu_xchg_x86((pcp), new)
#define this_cpu_read_8(pcp) percpu_from_op("mov", (pcp), "m"(pcp))
#define this_cpu_write_8(pcp, val) percpu_to_op("mov", (pcp), val)
@@ -279,6 +298,7 @@ do { \
#define this_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
#define this_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val)
#define this_cpu_cmpxchg_8(pcp, old,new) cmpxchg_local(this_cpu_ptr(pcp), old, new)
+#define this_cpu_xchg_8(pcp, new) this_cpu_xchg_x86((pcp), new)
#define irqsafe_cpu_add_8(pcp, val) percpu_to_op("add", (pcp), val)
#define irqsafe_cpu_inc_8(pcp) percpu_var_op("inc", (pcp))
@@ -287,6 +307,7 @@ do { \
#define irqsafe_cpu_or_8(pcp, val) percpu_to_op("or", (pcp), val)
#define irqsafe_cpu_xor_8(pcp, val) percpu_to_op("xor", (pcp), val)
#define irqsafe_cpu_cmpxchg_8(pcp, old,new) cmpxchg_local(this_cpu_ptr(pcp), old, new)
+#define irqsafe_cpu_xchg_8(pcp, new) this_cpu_xchg_x86((pcp), new)
#endif
--