[PATCH v3 04/11] riscv: Improve zacas fully-ordered cmpxchg()

[Alexandre Ghiti]

The current fully-ordered cmpxchgXX() implementation results in:

  amocas.X.rl     a5,a4,(s1)
  fence           rw,rw

This provides enough sync but we can actually use the following better
mapping instead:

  amocas.X.aqrl   a5,a4,(s1)

Suggested-by: Andrea Parri <andrea at rivosinc.com>
Signed-off-by: Alexandre Ghiti <alexghiti at rivosinc.com>
---
 arch/riscv/include/asm/cmpxchg.h | 71 ++++++++++++++++++++------------
 1 file changed, 44 insertions(+), 27 deletions(-)

diff --git a/arch/riscv/include/asm/cmpxchg.h b/arch/riscv/include/asm/cmpxchg.h
index c86722a101d0..97b24da38897 100644
--- a/arch/riscv/include/asm/cmpxchg.h
+++ b/arch/riscv/include/asm/cmpxchg.h
@@ -105,7 +105,10 @@
  * indicated by comparing RETURN with OLD.
  */
 
-#define __arch_cmpxchg_masked(sc_sfx, cas_sfx, prepend, append, r, p, o, n)	\
+#define __arch_cmpxchg_masked(sc_sfx, cas_sfx,				\
+			      sc_prepend, sc_append,			\
+			      cas_prepend, cas_append,			\
+			      r, p, o, n)				\
 ({									\
 	__label__ no_zabha_zacas, end;					\
 									\
@@ -119,9 +122,9 @@
 			 : : : : no_zabha_zacas);			\
 									\
 		__asm__ __volatile__ (					\
-			prepend						\
+			cas_prepend					\
 			"	amocas" cas_sfx " %0, %z2, %1\n"	\
-			append						\
+			cas_append					\
 			: "+&r" (r), "+A" (*(p))			\
 			: "rJ" (n)					\
 			: "memory");					\
@@ -139,7 +142,7 @@ no_zabha_zacas:;							\
 	ulong __rc;							\
 									\
 	__asm__ __volatile__ (						\
-		prepend							\
+		sc_prepend						\
 		"0:	lr.w %0, %2\n"					\
 		"	and  %1, %0, %z5\n"				\
 		"	bne  %1, %z3, 1f\n"				\
@@ -147,7 +150,7 @@ no_zabha_zacas:;							\
 		"	or   %1, %1, %z4\n"				\
 		"	sc.w" sc_sfx " %1, %1, %2\n"			\
 		"	bnez %1, 0b\n"					\
-		append							\
+		sc_append						\
 		"1:\n"							\
 		: "=&r" (__retx), "=&r" (__rc), "+A" (*(__ptr32b))	\
 		: "rJ" ((long)__oldx), "rJ" (__newx),			\
@@ -159,7 +162,10 @@ no_zabha_zacas:;							\
 end:;									\
 })
 
-#define __arch_cmpxchg(lr_sfx, sc_cas_sfx, prepend, append, r, p, co, o, n)	\
+#define __arch_cmpxchg(lr_sfx, sc_sfx, cas_sfx,				\
+		       sc_prepend, sc_append,				\
+		       cas_prepend, cas_append,				\
+		       r, p, co, o, n)					\
 ({									\
 	__label__ no_zacas, end;					\
 	register unsigned int __rc;					\
@@ -170,9 +176,9 @@ end:;									\
 			 : : : : no_zacas);				\
 									\
 		__asm__ __volatile__ (					\
-			prepend						\
-			"	amocas" sc_cas_sfx " %0, %z2, %1\n"	\
-			append						\
+			cas_prepend					\
+			"	amocas" cas_sfx " %0, %z2, %1\n"	\
+			cas_append					\
 			: "+&r" (r), "+A" (*(p))			\
 			: "rJ" (n)					\
 			: "memory");					\
@@ -181,12 +187,12 @@ end:;									\
 									\
 no_zacas:								\
 	__asm__ __volatile__ (						\
-		prepend							\
+		sc_prepend						\
 		"0:	lr" lr_sfx " %0, %2\n"				\
 		"	bne  %0, %z3, 1f\n"				\
-		"	sc" sc_cas_sfx " %1, %z4, %2\n"			\
+		"	sc" sc_sfx " %1, %z4, %2\n"			\
 		"	bnez %1, 0b\n"					\
-		append							\
+		sc_append						\
 		"1:\n"							\
 		: "=&r" (r), "=&r" (__rc), "+A" (*(p))			\
 		: "rJ" (co o), "rJ" (n)					\
@@ -195,7 +201,9 @@ no_zacas:								\
 end:;									\
 })
 
-#define _arch_cmpxchg(ptr, old, new, sc_sfx, prepend, append)		\
+#define _arch_cmpxchg(ptr, old, new, sc_sfx, cas_sfx,			\
+		      sc_prepend, sc_append,				\
+		      cas_prepend, cas_append)				\
 ({									\
 	__typeof__(ptr) __ptr = (ptr);					\
 	__typeof__(*(__ptr)) __old = (old);				\
@@ -204,22 +212,28 @@ end:;									\
 									\
 	switch (sizeof(*__ptr)) {					\
 	case 1:								\
-		__arch_cmpxchg_masked(sc_sfx, ".b" sc_sfx,		\
-					prepend, append,		\
-					__ret, __ptr, __old, __new);    \
+		__arch_cmpxchg_masked(sc_sfx, ".b" cas_sfx,		\
+				      sc_prepend, sc_append,		\
+				      cas_prepend, cas_append,		\
+				      __ret, __ptr, __old, __new);	\
 		break;							\
 	case 2:								\
-		__arch_cmpxchg_masked(sc_sfx, ".h" sc_sfx,		\
-					prepend, append,		\
-					__ret, __ptr, __old, __new);	\
+		__arch_cmpxchg_masked(sc_sfx, ".h" cas_sfx,		\
+				      sc_prepend, sc_append,		\
+				      cas_prepend, cas_append,		\
+				      __ret, __ptr, __old, __new);	\
 		break;							\
 	case 4:								\
-		__arch_cmpxchg(".w", ".w" sc_sfx, prepend, append,	\
-				__ret, __ptr, (long), __old, __new);	\
+		__arch_cmpxchg(".w", ".w" sc_sfx, ".w" cas_sfx,		\
+			       sc_prepend, sc_append,			\
+			       cas_prepend, cas_append,			\
+			       __ret, __ptr, (long), __old, __new);	\
 		break;							\
 	case 8:								\
-		__arch_cmpxchg(".d", ".d" sc_sfx, prepend, append,	\
-				__ret, __ptr, /**/, __old, __new);	\
+		__arch_cmpxchg(".d", ".d" sc_sfx, ".d" cas_sfx,		\
+			       sc_prepend, sc_append,			\
+			       cas_prepend, cas_append,			\
+			       __ret, __ptr, /**/, __old, __new);	\
 		break;							\
 	default:							\
 		BUILD_BUG();						\
@@ -228,16 +242,19 @@ end:;									\
 })
 
 #define arch_cmpxchg_relaxed(ptr, o, n)					\
-	_arch_cmpxchg((ptr), (o), (n), "", "", "")
+	_arch_cmpxchg((ptr), (o), (n), "", "", "", "", "", "")
 
 #define arch_cmpxchg_acquire(ptr, o, n)					\
-	_arch_cmpxchg((ptr), (o), (n), "", "", RISCV_ACQUIRE_BARRIER)
+	_arch_cmpxchg((ptr), (o), (n), "", "",				\
+		      "", RISCV_ACQUIRE_BARRIER, "", RISCV_ACQUIRE_BARRIER)
 
 #define arch_cmpxchg_release(ptr, o, n)					\
-	_arch_cmpxchg((ptr), (o), (n), "", RISCV_RELEASE_BARRIER, "")
+	_arch_cmpxchg((ptr), (o), (n), "", "",				\
+		      RISCV_RELEASE_BARRIER, "", RISCV_RELEASE_BARRIER, "")
 
 #define arch_cmpxchg(ptr, o, n)						\
-	_arch_cmpxchg((ptr), (o), (n), ".rl", "", "	fence rw, rw\n")
+	_arch_cmpxchg((ptr), (o), (n), ".rl", ".aqrl",			\
+		      "", RISCV_FULL_BARRIER, "", "")
 
 #define arch_cmpxchg_local(ptr, o, n)					\
 	arch_cmpxchg_relaxed((ptr), (o), (n))
-- 
2.39.2