fix riscv64 atomic asm constraints

most egregious problem was the lack of memory clobber and lack of
volatile asm; this made the atomics memory barriers but not compiler
barriers. use of "+r" rather than "=r" for a clobbered temp was also
wrong, since the initial value is indeterminate.

arch/riscv64/atomic_arch.h

@@ -8,13 +8,15 @@ static inline void a_barrier()
 static inline int a_cas(volatile int *p, int t, int s)
 {
 	int old, tmp;
-	__asm__("\n1:	lr.w.aqrl %0, %2\n"
+	__asm__ __volatile__ (
+		"\n1:	lr.w.aqrl %0, %2\n"
 		"	bne %0, %3, 1f\n"
 		"	sc.w.aqrl %1, %4, %2\n"
 		"	bnez %1, 1b\n"
 		"1:"
-		: "=&r"(old), "+r"(tmp), "+A"(*p)
-		: "r"(t), "r"(s));
+		: "=&r"(old), "=r"(tmp), "+A"(*p)
+		: "r"(t), "r"(s)
+		: "memory");
 	return old;
 }
 
@@ -23,12 +25,14 @@ static inline void *a_cas_p(volatile void *p, void *t, void *s)
 {
 	void *old;
 	int tmp;
-	__asm__("\n1:	lr.d.aqrl %0, %2\n"
+	__asm__ __volatile__ (
+		"\n1:	lr.d.aqrl %0, %2\n"
 		"	bne %0, %3, 1f\n"
 		"	sc.d.aqrl %1, %4, %2\n"
 		"	bnez %1, 1b\n"
 		"1:"
-		: "=&r"(old), "+r"(tmp), "+A"(*(long *)p)
-		: "r"(t), "r"(s));
+		: "=&r"(old), "=r"(tmp), "+A"(*(long *)p)
+		: "r"(t), "r"(s)
+		: "memory");
 	return old;
 }