On Mon, Dec 26, 2005 at 19:32:26 -0800, Matt Thomas wrote:

> > On Sun, Dec 25, 2005 at 13:54:13 -0500, Charles M. Hannum wrote:
> > 
> > > What prevents you from just moving errno to a different register
> > > before executing the PLT?  Surely you have another callee-saved
> > > register you can destroy.  You can probably even put the move
> > > instruction in one of the branch delay slots, so it doesn't cost
> > > you anything.
> > 
> > Right, but that kinda reinforces my other point.  If the interface
> > between syscall code and cerror is private, why do we need to
> > bother with exporting cerror at all?  We can just make it .hidden
> > and make the call to go directly to cerror, instead of going via
> > PLT.
> 
> cerror is also used by libposix.  So either posix will need its own
> or able to use libc's.

I know.  As I said in my original mail:

| libposix will need a copy of cerror to compensate for that change.
| Another possible solution is to change libposix to (tail)call into
| libc instead of duplicating the actual syscall code in libposix.

and I think calling libc is a cleaner option, with code like the
following:

    #include <sys/cdefs.h>

    extern int __posix_rename(const char *, const char *);

    #ifdef __weak_alias
    #define rename _rename
    __weak_alias(rename,_rename)
    #endif

    int
    rename(const char *from, const char *to)
    {
        return __posix_rename(from, to);
    }

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