inline llsc atomics when building for sh4a

If we're building for sh4a, the compiler is already free to use
instructions only available on sh4a, so we can do the same and inline the
llsc atomics. If we're building for an older processor, we still do the
same runtime atomics selection as before.

arch/sh/atomic.h

@@ -22,6 +22,88 @@ static inline int a_ctz_64(uint64_t x)
 	return a_ctz_l(y);
 }
 
+#define LLSC_CLOBBERS "r0", "t", "memory"
+#define LLSC_START(mem) "synco\n"  \
+	"0:	movli.l @" mem ", r0\n"
+#define LLSC_END(mem)              \
+	"1:	movco.l r0, @" mem "\n"    \
+	"	bf 0b\n"                   \
+	"	synco\n"
+
+static inline int __sh_cas_llsc(volatile int *p, int t, int s)
+{
+	int old;
+	__asm__ __volatile__(
+		LLSC_START("%1")
+		"	mov r0, %0\n"
+		"	cmp/eq %0, %2\n"
+		"	bf 1f\n"
+		"	mov %3, r0\n"
+		LLSC_END("%1")
+		: "=&r"(old) : "r"(p), "r"(t), "r"(s) : LLSC_CLOBBERS);
+	return old;
+}
+
+static inline int __sh_swap_llsc(volatile int *x, int v)
+{
+	int old;
+	__asm__ __volatile__(
+		LLSC_START("%1")
+		"	mov r0, %0\n"
+		"	mov %2, r0\n"
+		LLSC_END("%1")
+		: "=&r"(old) : "r"(x), "r"(v) : LLSC_CLOBBERS);
+	return old;
+}
+
+static inline int __sh_fetch_add_llsc(volatile int *x, int v)
+{
+	int old;
+	__asm__ __volatile__(
+		LLSC_START("%1")
+		"	mov r0, %0\n"
+		"	add %2, r0\n"
+		LLSC_END("%1")
+		: "=&r"(old) : "r"(x), "r"(v) : LLSC_CLOBBERS);
+	return old;
+}
+
+static inline void __sh_store_llsc(volatile int *p, int x)
+{
+	__asm__ __volatile__(
+		"	synco\n"
+		"	mov.l %1, @%0\n"
+		"	synco\n"
+		: : "r"(p), "r"(x) : "memory");
+}
+
+static inline void __sh_and_llsc(volatile int *x, int v)
+{
+	__asm__ __volatile__(
+		LLSC_START("%0")
+		"	and %1, r0\n"
+		LLSC_END("%0")
+		: : "r"(x), "r"(v) : LLSC_CLOBBERS);
+}
+
+static inline void __sh_or_llsc(volatile int *x, int v)
+{
+	__asm__ __volatile__(
+		LLSC_START("%0")
+		"	or %1, r0\n"
+		LLSC_END("%0")
+		: : "r"(x), "r"(v) : LLSC_CLOBBERS);
+}
+
+#ifdef __SH4A__
+#define a_cas(p,t,s)     __sh_cas_llsc(p,t,s)
+#define a_swap(x,v)      __sh_swap_llsc(x,v)
+#define a_fetch_add(x,v) __sh_fetch_add_llsc(x, v)
+#define a_store(x,v)     __sh_store_llsc(x, v)
+#define a_and(x,v)       __sh_and_llsc(x, v)
+#define a_or(x,v)        __sh_or_llsc(x, v)
+#else
+
 int  __sh_cas(volatile int *, int, int);
 int  __sh_swap(volatile int *, int);
 int  __sh_fetch_add(volatile int *, int);
@@ -35,6 +117,7 @@ void __sh_or(volatile int *, int);
 #define a_store(x,v)     __sh_store(x, v)
 #define a_and(x,v)       __sh_and(x, v)
 #define a_or(x,v)        __sh_or(x, v)
+#endif
 
 static inline void *a_cas_p(volatile void *p, void *t, void *s)
 {

arch/sh/src/atomic.c

@@ -1,12 +1,7 @@
-#include "libc.h"
+#ifndef __SH4A__
 
-#define LLSC_CLOBBERS   "r0", "t", "memory"
-#define LLSC_START(mem) "synco\n"  \
-	"0:	movli.l @" mem ", r0\n"
-#define LLSC_END(mem)              \
-	"1:	movco.l r0, @" mem "\n"    \
-	"	bf 0b\n"                   \
-	"	synco\n"
+#include "atomic.h"
+#include "libc.h"
 
 /* gusa is a hack in the kernel which lets you create a sequence of instructions
  * which will be restarted if the process is preempted in the middle of the
@@ -34,114 +29,74 @@
 
 int __sh_cas(volatile int *p, int t, int s)
 {
+	if (__hwcap & CPU_HAS_LLSC) return __sh_cas_llsc(p, t, s);
+
 	int old;
-	if (__hwcap & CPU_HAS_LLSC) {
-		__asm__ __volatile__(
-			LLSC_START("%1")
-			"	mov r0, %0\n"
-			"	cmp/eq %0, %2\n"
-			"	bf 1f\n"
-			"	mov %3, r0\n"
-			LLSC_END("%1")
-			: "=&r"(old) : "r"(p), "r"(t), "r"(s) : LLSC_CLOBBERS);
-	} else {
-		__asm__ __volatile__(
-			GUSA_START_EVEN("%1", "%0")
-			"	cmp/eq %0, %2\n"
-			"	bf 1f\n"
-			GUSA_END("%1", "%3")
-			: "=&r"(old) : "r"(p), "r"(t), "r"(s) : GUSA_CLOBBERS, "t");
-	}
+	__asm__ __volatile__(
+		GUSA_START_EVEN("%1", "%0")
+		"	cmp/eq %0, %2\n"
+		"	bf 1f\n"
+		GUSA_END("%1", "%3")
+		: "=&r"(old) : "r"(p), "r"(t), "r"(s) : GUSA_CLOBBERS, "t");
 	return old;
 }
 
 int __sh_swap(volatile int *x, int v)
 {
+	if (__hwcap & CPU_HAS_LLSC) return __sh_swap_llsc(x, v);
+
 	int old;
-	if (__hwcap & CPU_HAS_LLSC) {
-		__asm__ __volatile__(
-			LLSC_START("%1")
-			"	mov r0, %0\n"
-			"	mov %2, r0\n"
-			LLSC_END("%1")
-			: "=&r"(old) : "r"(x), "r"(v) : LLSC_CLOBBERS);
-	} else {
-		__asm__ __volatile__(
-			GUSA_START_EVEN("%1", "%0")
-			GUSA_END("%1", "%2")
-			: "=&r"(old) : "r"(x), "r"(v) : GUSA_CLOBBERS);
-	}
+	__asm__ __volatile__(
+		GUSA_START_EVEN("%1", "%0")
+		GUSA_END("%1", "%2")
+		: "=&r"(old) : "r"(x), "r"(v) : GUSA_CLOBBERS);
 	return old;
 }
 
 int __sh_fetch_add(volatile int *x, int v)
 {
+	if (__hwcap & CPU_HAS_LLSC) return __sh_fetch_add_llsc(x, v);
+
 	int old, dummy;
-	if (__hwcap & CPU_HAS_LLSC) {
-		__asm__ __volatile__(
-			LLSC_START("%1")
-			"	mov r0, %0\n"
-			"	add %2, r0\n"
-			LLSC_END("%1")
-			: "=&r"(old) : "r"(x), "r"(v) : LLSC_CLOBBERS);
-	} else {
-		__asm__ __volatile__(
-			GUSA_START_EVEN("%2", "%0")
-			"	mov %0, %1\n"
-			"	add %3, %1\n"
-			GUSA_END("%2", "%1")
-			: "=&r"(old), "=&r"(dummy) : "r"(x), "r"(v) : GUSA_CLOBBERS);
-	}
+	__asm__ __volatile__(
+		GUSA_START_EVEN("%2", "%0")
+		"	mov %0, %1\n"
+		"	add %3, %1\n"
+		GUSA_END("%2", "%1")
+		: "=&r"(old), "=&r"(dummy) : "r"(x), "r"(v) : GUSA_CLOBBERS);
 	return old;
 }
 
 void __sh_store(volatile int *p, int x)
 {
-	if (__hwcap & CPU_HAS_LLSC) {
-		__asm__ __volatile__(
-			"	synco\n"
-			"	mov.l %1, @%0\n"
-			"	synco\n"
-			: : "r"(p), "r"(x) : "memory");
-	} else {
-		__asm__ __volatile__(
-			"	mov.l %1, @%0\n"
-			: : "r"(p), "r"(x) : "memory");
-	}
+	if (__hwcap & CPU_HAS_LLSC) return __sh_store_llsc(p, x);
+	__asm__ __volatile__(
+		"	mov.l %1, @%0\n"
+		: : "r"(p), "r"(x) : "memory");
 }
 
 void __sh_and(volatile int *x, int v)
 {
+	if (__hwcap & CPU_HAS_LLSC) return __sh_and_llsc(x, v);
+
 	int dummy;
-	if (__hwcap & CPU_HAS_LLSC) {
-		__asm__ __volatile__(
-			LLSC_START("%0")
-			"	and %1, r0\n"
-			LLSC_END("%0")
-			: : "r"(x), "r"(v) : LLSC_CLOBBERS);
-	} else {
-		__asm__ __volatile__(
-			GUSA_START_ODD("%1", "%0")
-			"	and %2, %0\n"
-			GUSA_END("%1", "%0")
-			: "=&r"(dummy) : "r"(x), "r"(v) : GUSA_CLOBBERS);
-	}
+	__asm__ __volatile__(
+		GUSA_START_ODD("%1", "%0")
+		"	and %2, %0\n"
+		GUSA_END("%1", "%0")
+		: "=&r"(dummy) : "r"(x), "r"(v) : GUSA_CLOBBERS);
 }
 
 void __sh_or(volatile int *x, int v)
 {
+	if (__hwcap & CPU_HAS_LLSC) return __sh_or_llsc(x, v);
+
 	int dummy;
-	if (__hwcap & CPU_HAS_LLSC) {
-		__asm__ __volatile__(
-			LLSC_START("%0")
-			"	or %1, r0\n"
-			LLSC_END("%0")
-			: : "r"(x), "r"(v) : LLSC_CLOBBERS);
-	} else {
-		__asm__ __volatile__(
-			GUSA_START_ODD("%1", "%0")
-			"	or %2, %0\n"
-			GUSA_END("%1", "%0")
-			: "=&r"(dummy) : "r"(x), "r"(v) : GUSA_CLOBBERS);
-	}
+	__asm__ __volatile__(
+		GUSA_START_ODD("%1", "%0")
+		"	or %2, %0\n"
+		GUSA_END("%1", "%0")
+		: "=&r"(dummy) : "r"(x), "r"(v) : GUSA_CLOBBERS);
 }
+
+#endif