_locore_start needs to set up an 8-byte aligned stack
Sadly, during removal of special early-boot stacks
(027eb01d0ff8f66be55208b58ecd7cb2b7b27714) the stack passed into mlsetup
lost its 8-byte aligned-ness.

@@ -121,17 +121,22 @@
         orr     r0, #0x1000
         orr     r0, #0x400000
         mcr     p15, 0, r0, c1, c0, 0
 
         /*
-         * mlsetup() takes the struct regs as an argument. main doesn't take any
-         * and should never return. After the push below, we should have a
-         * 8-byte aligned stack pointer. This is why we subtracted four earlier
-         * on if we were 8-byte aligned.
+         * mlsetup() takes the struct regs as an argument. main doesn't take
+         * any and should never return. Currently, we have an 8-byte aligned
+         * stack.  We want to push a zero frame pointer to terminate any
+         * stack walking, but that would cause us to end up with only a
+         * 4-byte aligned stack.  So, to keep things nice and correct, we
+         * push a zero value twice - it's similar to a typical function
+         * entry:
+         *      push { r9, lr }
          */
         mov     r9,#0
-        push    { r9 }
+        push    { r9 }          /* link register */
+        push    { r9 }          /* frame pointer */
         mov     r0, sp
         bl      mlsetup
         bl      main
         /* NOTREACHED */
         ldr     r0,=__return_from_main