patch feedback

*** 47,59 ****
  /*
   * The original i386 ABI said that the stack pointer need be only 4-byte
   * aligned before a function call (STACK_ALIGN == 4).  The ABI supplement
   * version 1.0 changed the required alignment to 16-byte for the benefit of
   * floating point code compiled using sse2.  The compiler assumes this
!  * alignment and maintains it for calls made from that function.  If the
!  * stack is initially properly aligned, it will continue to be so aligned.
!  * If it is not initially so aligned, it will never become so aligned.
   *
   * One slightly confusing detail to keep in mind is that the 16-byte
   * alignment (%esp & 0xf == 0) is true just *before* the call instruction.
   * The call instruction will then push a return value, decrementing %esp by
   * 4.  Therefore, if one dumps %esp at the at the very first instruction in
--- 47,59 ----
  /*
   * The original i386 ABI said that the stack pointer need be only 4-byte
   * aligned before a function call (STACK_ALIGN == 4).  The ABI supplement
   * version 1.0 changed the required alignment to 16-byte for the benefit of
   * floating point code compiled using sse2.  The compiler assumes this
!  * alignment and maintains it for calls it generates.  If the stack is
!  * initially properly aligned, it will continue to be so aligned.  If it is
!  * not initially so aligned, it will never become so aligned.
   *
   * One slightly confusing detail to keep in mind is that the 16-byte
   * alignment (%esp & 0xf == 0) is true just *before* the call instruction.
   * The call instruction will then push a return value, decrementing %esp by
   * 4.  Therefore, if one dumps %esp at the at the very first instruction in
*** 85,107 ****
          size = sizeof (long) * (argc + 1);
  
          tsp = (long *)(((uintptr_t)ucp->uc_stack.ss_sp +
              ucp->uc_stack.ss_size - size) & ~(STACK_ALIGN - 1));
  
          sp = tsp - 1;
  
-         va_start(ap, argc);
- 
-         while (argc-- > 0) {
-                 *tsp++ = va_arg(ap, long);
-         }
- 
-         va_end(ap);
- 
          *sp = (long)resumecontext;              /* return address */
  
          ucp->uc_mcontext.gregs[UESP] = (greg_t)sp;
  }
  
  
  static void
  resumecontext(void)
--- 85,112 ----
          size = sizeof (long) * (argc + 1);
  
          tsp = (long *)(((uintptr_t)ucp->uc_stack.ss_sp +
              ucp->uc_stack.ss_size - size) & ~(STACK_ALIGN - 1));
  
+         /*
+          * Since we're emulating the call instruction, we must push the
+          * return address (which involves adjusting the stack pointer to
+          * have the proper 4-byte bias).
+          */
          sp = tsp - 1;
  
          *sp = (long)resumecontext;              /* return address */
  
          ucp->uc_mcontext.gregs[UESP] = (greg_t)sp;
+ 
+         /*
+          * "push" all the arguments
+          */
+         va_start(ap, argc);
+         while (argc-- > 0)
+                 *tsp++ = va_arg(ap, long);
+         va_end(ap);
  }
  
  
  static void
  resumecontext(void)