va_list q

[Peter Wemm]

On Wednesday 21 April 2004 10:01 pm, Valentin Nechayev wrote:
> Hi,
> I was notified that maintained port (net/pppd23) can't be built on
> amd64. Failure message is:
>
> cc -O -pipe   -g -I../include -DHAVE_PATHS_H -D_BITYPES  -c utils.c
> utils.c: In function `vslprintf':
> utils.c:253: error: invalid use of non-lvalue array
> *** Error code 1
>
> The code in question is:
>
> int
> vslprintf(buf, buflen, fmt, args)
> [...]
> #ifndef __powerpc__
>
> >            n = vslprintf(buf, buflen + 1, f, va_arg(args,
> > va_list));
>
> #else
>             /* On the powerpc, a va_list is an array of 1 structure
> */ n = vslprintf(buf, buflen + 1, f, va_arg(args, void *)); #endif
>
> (I selected failed line with '>')
>
> va_arg() uses translation to pointer to specified type; this
> translation fails with message shown above. (Translation is used
> AFAIS on all platforms, but only amd64 causes error.) I have no
> freebsd/amd64 box now. Can anybody say how can it be fixed, and
> whether shown powerpc hack is applicable to amd64 if it's last-resort
> solution?

amd64 needs to use the same code that is in the #ifdef __powerpc__.  Its 
what we use in src/usr.sbin/pppd FWIW.

On powerpc and amd64, the argument passing ABI is so complicated that 
the sequence counters can't be fit in the spare bits in a pointer like 
on the other platforms.  So on those two (and some other) platforms, 
gcc implements va_list as a pointer to an external structure.

As an aside, this breaks code that assums it can copy va_lists by 
assignment.  On powerpc and amd64, you *must* use va_copy(), or you 
simply copy the pointer, not the actual argument passing state.
-- 
Peter Wemm - peter at wemm.org; peter at FreeBSD.org; peter at yahoo-inc.com
"All of this is for nothing if we don't go to the stars" - JMS/B5