patch from njk: make x86_64 __uniclone branchless.

src/thread/x86_64/clone.s

@@ -4,6 +4,7 @@
 .type   __uniclone,%function
 /* rdi = child_stack, rsi = start, rdx = pthread_struct */
 __uniclone:
+        subq    $8,%rsp         /* pad parent stack to prevent branch later */
         subq    $16,%rdi        /* grow child_stack */
         mov     %rsi,8(%rdi)    /* push start onto child_stack as return ptr */
         mov     %rdx,0(%rdi)    /* push pthread_struct onto child_stack */
@@ -15,8 +16,7 @@ __uniclone:
         mov     %r10,%rdx       /* rdx = parent_id */
         syscall                 /* clone(flags, child_stack, parent_id,
                                  *       child_id, tls) */
-	test	%rax,%rax
-	jnz	1f              /* if we're in the parent -> goto 1f */
-        pop     %rdi            /* restore pthread_struct from child stack */
-1:      ret
+        pop     %rdi            /* child stack: restore pthread_struct
+                                 * parent stack: undo rsp displacement */
+        ret
 .size __uniclone,.-__uniclone