Testcase is extracted from Wine. /home/marcus/projects/gcc/BIN/bin/gcc -m32 -fPIC -O2 -o reg reg.i ./reg Segmentation fault (I did not try the i586 version of the compiler.) Looking at the disassembly the stdcall function is compiled as: foo: pushl %ebp movl %esp, %ebp pushl %ebx call __i686.get_pc_thunk.bx addl $_GLOBAL_OFFSET_TABLE_, %ebx subl $4, %esp movl $0, (%esp) call bar@PLT addl $4, %esp popl %ebx popl %ebp ret It does adjust the stackframe after the call to bar wrongly, either it should not do it, or just emit a "leave".
Created attachment 13389 [details] reg.i gcc -fPIC -O2 -m32 reg.i
i bisected gcc to find the regression and it was apparently introduced by revision 123146. 2007-03-22 Richard Henderson <rth@redhat.com> * config/i386/i386.c (ix86_function_regparm): Early exit for 64-bit; don't increase local_regparm with force_align_arg_pointer check. (ix86_function_sseregparm): Assert 32-bit. (type_has_variadic_args_p): New. [...]
the interesting part is that the declaration is just slightly incompatible with the actual definition. If I remove the prototype the problem goes away.
Confirmed.
Reproducing the problem requires -maccumulate-outgoing-args, as implied by i686 tuning.
Subject: Bug 31628 Author: rth Date: Sat Apr 21 00:53:37 2007 New Revision: 124014 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=124014 Log: PR target/31628 * config/i386/i386.c (type_has_variadic_args_p): Look for any TREE_LIST with a void_type_node value, not void_list_node exactly. Modified: trunk/gcc/ChangeLog trunk/gcc/config/i386/i386.c
Fixed.
Subject: Bug 31628 Author: hjl Date: Fri May 18 19:29:45 2007 New Revision: 124831 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=124831 Log: 2007-05-18 H.J. Lu <hongjiu.lu@intel.com> PR target/31628 * gcc.target/i386/pr31628.c: New. Added: trunk/gcc/testsuite/gcc.target/i386/pr31628.c Modified: trunk/gcc/testsuite/ChangeLog