Diff below makes sure we properly align the stack for executing SSE
instructions on i386 like we already do on amd64. Our GCC already
generates code in main() to do this, but other compilers (including
certain versions of GCC as pointed out to me by Jeorg Sonnenberger)
might not.
ok?
Index: crt0.c
===================================================================
RCS file: /cvs/src/lib/csu/i386/crt0.c,v
retrieving revision 1.13
diff -u -p -r1.13 crt0.c
--- crt0.c 4 Aug 2005 16:33:05 -0000 1.13
+++ crt0.c 26 Jul 2011 09:53:29 -0000
@@ -56,15 +56,20 @@ __asm(".text\n"
" .globl _start\n"
"_start:\n"
"__start:\n"
-" pushl %ebx #ps_strings\n"
+" movl %esp,%ebp\n"
+" subl $12,%esp # align stack\n"
+" andl $~15,%esp\n"
+" addl $12,%esp\n"
+" pushl %ebx # ps_strings\n"
" pushl %ecx # obj\n"
" pushl %edx # cleanup\n"
-" movl 12(%esp),%eax\n"
-" leal 20(%esp,%eax,4),%ecx\n"
-" leal 16(%esp),%edx\n"
+" movl 0(%ebp),%eax\n"
+" leal 8(%ebp,%eax,4),%ecx\n"
+" leal 4(%ebp),%edx\n"
" pushl %ecx\n"
" pushl %edx\n"
" pushl %eax\n"
+" xorl %ebp,%ebp # mark deepest stack frame\n"
" call ___start ");
void
