[LLVMbugs] [Bug 7117] New: regparm attribute combined with K&R declarations causes illegal instruction

bugzilla-daemon at llvm.org bugzilla-daemon at llvm.org
Tue May 11 13:23:17 PDT 2010


http://llvm.org/bugs/show_bug.cgi?id=7117

           Summary: regparm attribute combined with K&R declarations
                    causes illegal instruction
           Product: clang
           Version: trunk
          Platform: PC
        OS/Version: FreeBSD
            Status: NEW
          Severity: normal
          Priority: P
         Component: -New Bugs
        AssignedTo: unassignedclangbugs at nondot.org
        ReportedBy: dimitry at andric.com
                CC: llvmbugs at cs.uiuc.edu


While building and testing the FreeBSD version of GNU libregex with
clang, I consistently got 'illegal instruction' from it at runtime.

It turned out clang had inserted instruction "ud2" at many places inside
the object code.  This seems to have been caused by quite a number of
functions inside the regex compiler having this attribute tacked on:

__attribute ((regparm (3), stdcall))

While this in itself should probably cause no harm, it apparently does
become problematic, when used in combination with K&R-style function
declarations.

I reduced this phenomenon to the following testcase:


int re_string_construct(int) __attribute__((regparm(3), stdcall));
int re_string_construct(dfa)
{
}
int re_compile_internal(void)
{
  init_dfa();
  re_string_construct(0);
}


Compile with "clang -cc1 -O2 -emit-llvm" and the last function becomes:

define i32 @re_compile_internal() noreturn nounwind {
entry:
  %call = tail call i32 (...)* @init_dfa() nounwind ; <i32> [#uses=0]
  tail call void @llvm.trap()
  unreachable
}

which is obviously incorrect.  Compiling with -O0, or removing the
attribute altogether produces correct output.

(Yes, I realize K&R is really really deprecated now, but unfortunately a
lot of code out there still uses it, especially older GNU stuff... :)

-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.



More information about the llvm-bugs mailing list