> On Thu, 12 Jul 2001, Richard Earnshaw wrote:
> 
> > > USRSTACK, VM_MIN_ADDRESS and VM_MAXUSER_ADDRESS
> >
> > These shouldn't be hard-coded into user-land programs.  Instead we should
> > have a mechanism for getting the numbers from the kernel.  Then machines
> > can set these as appropriate and user-land binaries will continue to work
> > correctly.  This information could easily be made available via the
> > sysctl(3) interface.
> 
> This was discussed on tech-kern in December/January 2000/2001.
> 
> <URL:http://mail-index.netbsd.org/tech-kern/2000/12/26/0001.html>
> <URL:http://mail-index.netbsd.org/tech-kern/2001/01/02/0002.html>
> <URL:http://mail-index.netbsd.org/tech-kern/2001/01/03/0033.html>
> 

Hmm, so it looks like there wasn't a conclusion...

It seems we have two possible situations.

1) We leave it as it is, which means that all builds must use the same 
values of these parameters, or we must have different binaries for each 
possible kernel variation of this value.

2) We make it dynamic, which means that binaries would work on all native 
builds, but cross dumps, or dumps made before the native kernel values 
change, would not work.

I know which sounds the most preferable to me.

R.