[Xen-devel] [PATCH v2 03/17] xen: arm32: ensure cmpxchg has full barrier semantics

[Ian Campbell <ian.campbell@xxxxxxxxxx>]

Unrelated reads/writes should not pass the xchg.

Provide cmpxchg_local for parity with arm64, although it appears to be unused.
It also helps make the reason for the separation of __cmpxchg_mb more
apparent.

With this our cmpxchg is in sync with Linux v3.14-rc7.

Signed-off-by: Ian Campbell <ian.campbell@xxxxxxxxxx>
Acked-by: Julien Grall <julien.grall@xxxxxxxxxx>
---

We got our cmpxchg implementation from Linux which AFAICS has always had these
additional barriers. I don't recall us having decided that Xen barriers should
not have this property as well, and if we did we were remiss in not adding a
comment etc... If my memory is faulty then I am happy to replace thispatch
with one which adds a comment instead.
---
 xen/include/asm-arm/arm32/system.h |   26 +++++++++++++++++++++++---
 1 file changed, 23 insertions(+), 3 deletions(-)

diff --git a/xen/include/asm-arm/arm32/system.h 
b/xen/include/asm-arm/arm32/system.h
index 9f233fe..dfaa3b6 100644
--- a/xen/include/asm-arm/arm32/system.h
+++ b/xen/include/asm-arm/arm32/system.h
@@ -113,9 +113,29 @@ static always_inline unsigned long __cmpxchg(
     return oldval;
 }
 
-#define cmpxchg(ptr,o,n)                                                \
-    ((__typeof__(*(ptr)))__cmpxchg((ptr),(unsigned long)(o),            \
-                                   (unsigned long)(n),sizeof(*(ptr))))
+static inline unsigned long __cmpxchg_mb(volatile void *ptr, unsigned long old,
+                                        unsigned long new, int size)
+{
+       unsigned long ret;
+
+       smp_mb();
+       ret = __cmpxchg(ptr, old, new, size);
+       smp_mb();
+
+       return ret;
+}
+
+#define cmpxchg(ptr,o,n)                                               \
+       ((__typeof__(*(ptr)))__cmpxchg_mb((ptr),                        \
+                                         (unsigned long)(o),           \
+                                         (unsigned long)(n),           \
+                                         sizeof(*(ptr))))
+
+#define cmpxchg_local(ptr,o,n)                                         \
+       ((__typeof__(*(ptr)))__cmpxchg((ptr),                           \
+                                      (unsigned long)(o),              \
+                                      (unsigned long)(n),              \
+                                      sizeof(*(ptr))))
 
 #define local_irq_disable() asm volatile ( "cpsid i @ local_irq_disable\n" : : 
: "cc" )
 #define local_irq_enable()  asm volatile ( "cpsie i @ local_irq_enable\n" : : 
: "cc" )
-- 
1.7.10.4


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel