On Thu, Jul 01, 1999 at 03:32:22PM +0200, ks@ub.uni-mainz.de wrote:
> yes, I got this, too, on my ss20 (it's not converted to elf yet) and
> there is a line OBJECT_FMT=AOUT in /etc/mk.conf (if this matters).

So this is not a ELF problem. Hum ...

> tying to build a new userland failed due to the new malloc.c not
> compiling (at least) on sparcs (no problem on an i386). the error
> is as follows: (I'v reported this on 06/28/99, too)

Just ran into this as well. For now I put back the 1.4 malloc.c in place,
and send-pr'd the thing.

> 
> [...]
> 
> pgshift depends on SUN4 or (SUN4C or SUN4M) and is set in locore.s.
> maybe the magic to get pgshift set right should be done in some header
> file malloc includes. but then malloc compiled on an ss5 or IPC
> doesn't do TRT on an 4/300.
> is it possible to expose the kernel variable 'pgshift' to userland? If
> not, I can see no (easy?) way to compile malloc on sparc architecture
> independend. hm, how did the old malloc handle this??? time to find
> some space for an 1.4-release tree...

We don't have pgshift, but we have the hw.pagesize sysctl. I'm not sure this
is the rigth value to use on all hardware through (there may be ports where
the page size used by the kernel is a multiple of the hw page size).

A new sysctl, kern.pagesize should probably be created ...

--
Manuel Bouyer, LIP6, Universite Paris VI.           Manuel.Bouyer@lip6.fr
--