lkmc_futex: use inline assembly directly.

Makes it more predictable for gem5 LDXR experiments, since otherwise a
dynamically linked syscall has LLSC instructions of its own.

lkmc/futex.h

@@ -12,9 +12,22 @@ static int
 lkmc_futex(int *uaddr, int futex_op, int val,
         const struct timespec *timeout, int *uaddr2, int val3)
 {
-    (void)uaddr2;
+    register uint64_t x0 __asm__ ("x0") = (uint64_t)uaddr;
-    return syscall(SYS_futex, uaddr, futex_op, val,
+    register uint64_t x1 __asm__ ("x1") = futex_op;
-                    timeout, uaddr, val3);
+    register uint64_t x2 __asm__ ("x2") = val;
+    register const struct timespec *x3 __asm__ ("x3") = timeout;
+    register int *x4 __asm__ ("x4") = uaddr2;
+    register uint64_t x5 __asm__ ("x5") = val3;
+    register uint64_t x8 __asm__ ("x8") = SYS_futex; /* syscall number */
+    __asm__ __volatile__ (
+        "svc 0;"
+        : "+r" (x0)
+        : "r" (x1), "r" (x2), "r" (x3), "r" (x4), "r" (x5), "r" (x8)
+        : "memory"
+    );
+    return x0;
+    //return syscall(SYS_futex, uaddr, futex_op, val,
+    //                timeout, uaddr2, val3);
 }
 
 #endif