Re: [PATCH, i386]: Fix PR 41900, call *%esp shouldn't be generated because of CPU errata

[Uros Bizjak <ubizjak at gmail dot com>]

On 11/03/2009 07:47 PM, Uros Bizjak wrote:
> > > > > works OK on Fedora 11 and CentOS 5.3.
> > > > The extra volatile doesn't help at all, it still fails here (Fedora 
> > > > 11,
> > > > x86_64 kernel, SELinux enforcing).
> > > Hm, indeed. Is there a solution by somehow enabling execution (like
> > > trampolines)?
> > You can assemble with -Wa,--execstack or link with -Wl,-z,execstack.
> > But it helps only in some cases, some versions of the policy won't allow
> > unconfined processes to have exec stack, you'd need to chcon -t 
> > java_exec_t
> > or similar.  This is highly Linux specific though.
> No, this test failed due to missing "volatile" (fixed now). AFAICS, it
>
> I think I'll just make the test compile-only and scan for call without 
> *%esp argument, as you suggested in your previous mails.

Like this:

2009-11-03  Uros Bizjak <ubizjak@gmail.com>

    * gcc.target/i386/pr41900.c: Make test compile only.  Scan assembler
    dump to not include "call *%esp".

Tested on F11, -m32.

Uros.

Index: gcc.target/i386/pr41900.c
===================================================================
--- gcc.target/i386/pr41900.c    (revision 153863)
+++ gcc.target/i386/pr41900.c    (working copy)
@@ -1,4 +1,4 @@
-/* { dg-do run } */
+/* { dg-do compile } */
 /* { dg-require-effective-target ilp32 } */
 /* { dg-options "-O2 -fomit-frame-pointer 
-mpreferred-stack-boundary=2" } */

@@ -9,3 +9,5 @@
   ((void (*)(void)) &code) ();
   return 0;
 }
+
+/* { dg-final { scan-assembler-not "call\[ \\t\]+\\*%esp" } } */