[Bug-gdb] Re: Suspected bug in GDB 5.0 -- SIGSEV sent for unaligned access instead of SIGBUS

[Peter.Schauer]

Older versions of GDB had this in procfs.c:

#endif /* not FAULTED_USE_SIGINFO */
            default:
              /* Use the signal which the kernel assigns.  This is better than
                 trying to second-guess it from the fault.  In fact, I suspect
                 that FLTACCESS can be either SIGSEGV or SIGBUS.  */
#ifdef UNIXWARE
              statval = ((pi->prstatus.pr_lwp.pr_info.si_signo) << 8) | 0177;
#else
              statval = ((pi->prstatus.pr_info.si_signo) << 8) | 0177;
#endif
              break;

which has been left out during a procfs.c overhaul and causes the behaviour
you are observing.

The new code has some FIXMEs though:

                switch (what) { /* FIXME: FAULTED_USE_SIGINFO */
                  /* FIXME: use si_signo where possible. */
                default:         /* FIXME: use si_signo if possible for fault */

so it seems that the problem is known.

I do not know enough about other procfs implementations (UNIXWARE, IRIX etc.)
to come up with a proper fix, sorry.

> Dear gdb maintainer,
> 
> Versions of gdb prior to 5.0 delivered a SIGBUS
> for an unaligned access.  The shell has this behavior,
> too.  However, gdb 5.0 interprets unaligned accesses
> as SIGSEGV.  This is a problem when debugging programs
> that use SIGBUS.
> 
> Here is an example of a program with an unaligned
> access running in gdb 4.17 and gdb 5.0:
> 
> /mnt3/jlp> gcc -g signal.c
> /mnt3/jlp> a.out
> Bus Error
> /mnt3/jlp> cat signal.c
> int main()
> {
>   union {
>     char c[4];
>     long *lP;
>   } u;
> 
>   *u.lP = malloc(5);
>   u.c[3]++;
> 
>   *u.lP = 42;  
>   return 0;
> }
> /mnt3/jlp> 
> 
> /mnt3/jlp> /usr/local/bin/gdb-4.17 a.out
> GNU gdb 4.17
> Copyright 1998 Free Software Foundation, Inc.
> GDB is free software, covered by the GNU General Public License, and you are
> welcome to change it and/or distribute copies of it under certain conditions.
> Type "show copying" to see the conditions.
> There is absolutely no warranty for GDB.  Type "show warranty" for details.
> This GDB was configured as "sparc-sun-solaris2.6"...
> (gdb) b main
> Breakpoint 1 at 0x10564: file signal.c, line 10.
> (gdb) r
> Starting program: /mnt3/jlp/a.out 
> in gdb.  .cshrc was not sourced
> 
> Breakpoint 1, main () at signal.c:10
> 10      *u.lP = malloc(4);
> (gdb) n
> 11      u.c[3]++;
> (gdb) 
> 13      *u.lP = 42;  
> (gdb) 
> 
> Program terminated with signal SIGBUS, Bus error.
> The program no longer exists.
> (gdb) quit
> /mnt3/jlp> gdb a.out
> GNU gdb 5.0 (UI_OUT)
> Copyright 2000 Free Software Foundation, Inc.
> GDB is free software, covered by the GNU General Public License, and you are
> welcome to change it and/or distribute copies of it under certain conditions.
> Type "show copying" to see the conditions.
> There is absolutely no warranty for GDB.  Type "show warranty" for details.
> This GDB was configured as "sparc-sun-solaris2.7"...
> (gdb) b main
> Breakpoint 1 at 0x10564: file signal.c, line 10.
> (gdb) r
> Starting program: /mnt3/jlp/a.out 
> in gdb.  .cshrc was not sourced
> 
> Breakpoint 1, main () at signal.c:10
> 10      *u.lP = malloc(4);
> (gdb) n
> 11      u.c[3]++;
> (gdb) 
> 13      *u.lP = 42;  
> (gdb) 
> 
> Program received signal SIGSEGV, Segmentation fault.
> 0x00010590 in main () at signal.c:13
> 13      *u.lP = 42;  
> (gdb) quit
> The program is running.  Exit anyway? (y or n) y
> /mnt3/jlp> 
> 
> 
> 


-- 
Peter Schauer                   address@hidden