On Fri, Aug 31, 2001 at 10:23:44 +1000, Simon Burge wrote:

> > > > get at the faulting address in its SIGSEGV handler.
> > > 
> > > sc_eip should give you what you want.
> > 
> > My understanding is that it needs the address of the access that
> > caused the fault (e.g SFVA on sun4m), not the address of the
> > instruction that tried to perform the access.
> 
> A quick glance at a few <machine/signal.h>'s show that sc_pc *seems*
> to be present on all architectures.

sc_pc is %pc to restore.  What's needed is the address that causes the
fault.  E.g. for

    int *bad = (int *)0xdeadbee0;
    *bad = 0;  /* handler is interested in 'bad' not the %pc */

On sparc the handler will be called with code=0xdeadbee0 (the fault
address).  And because of a clever trick in sigcode all the cpu
regiters at the time of the signal are also available with a little
stack scavenge (what would be in mcontext_t on sysv).

SY, Uwe
-- 
uwe@ptc.spbu.ru                         |       Zu Grunde kommen
http://www.ptc.spbu.ru/~uwe/            |       Ist zu Grunde gehen