amd64/126543: [amd64] __builtin_frame_address does not return 0 when reaching the top of the stack

[Douglas William Thrift]

>Number:         126543
>Category:       amd64
>Synopsis:       [amd64] __builtin_frame_address does not return 0 when reaching the top of the stack
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    freebsd-amd64
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Fri Aug 15 01:50:03 UTC 2008
>Closed-Date:
>Last-Modified:
>Originator:     Douglas William Thrift
>Release:        FreeBSD 7.0-STABLE-jc2 amd64
>Organization:
>Environment:
System: FreeBSD slowhand.douglasthrift.net 7.0-STABLE-jc2 FreeBSD 7.0-STABLE-jc2 #0: Thu Jun 26 21:33:03 PDT 2008 user at jail8.johncompanies.com:/usr/obj/usr/src/sys/jail8 amd64
>Description:
On the amd64 architecture the gcc builtin function __builtin_frame_address does not return 0 when it is called for the level that should be the top of the stack. Instead it seems to continue returning addresses and then causes a bus error.

GCC documentation: http://gcc.gnu.org/onlinedocs/gcc-4.2.4/gcc/Return-Address.html

I've seen this occur on this system and also on two systems running FreeBSD 7.0-RELEASE-p3.

This causes devel/libexecinfo to be completely useless and crashy.
>How-To-Repeat:
Compile the following C program:

#include <stdio.h>

int main(int argc, char *argv[])
{
    void *address;
    int index;

    for (index = 0; address != NULL && index != 5; ++index)
    {
        switch (index)
        {
        case 0: address = __builtin_frame_address(0); break;
        case 1: address = __builtin_frame_address(1); break;
        case 2: address = __builtin_frame_address(2); break;
        case 3: address = __builtin_frame_address(3); break;
        case 4: address = __builtin_frame_address(4); break;
        }

        printf("0x%x\n", address);
    }

    return 0;
}

Run on an i386 machine its output is fine:
[douglas at backhome:~]$ ./test
0xbfbfec68
0xbfbfec9c
0x0

Run on an amd64 machine is not so fine:
[douglas at justonenight:~]$ ./test
0xffffeaf0
0xffffeb40
0xffffecef
0x4c454853
Bus error: 10 (core dumped)

The last hex value printed seems to consistently be a string from the environment variables. Judging by the two stack frame addresses printed on i386, perhaps the third address printed on amd64 is the one that should be a zero.
>Fix:
>Release-Note:
>Audit-Trail:
>Unformatted: