> On Saturday 17 February 2001  9:31 pm, Jason R Thorpe wrote:
> > On Sat, Feb 17, 2001 at 07:20:23PM +0000, Ben Harris wrote:
> >  > > Well, 32M is a pretty small address space -- it's going to affect
> >  > > where you put the stack, is certainly going to affect shared
> >  > > libraries, etc., and may constitute a new ABI.
> >  >
> >  > Heh.  It's still bigger than the arm26 user address space.
> >  >
> >  > Hmm.  Could you dynamically enable it for processes whose hard memory
> >  > limits were low enough?  That'd be cute.
> >
> > I guess you could run-time switch, sure...
> 
> Oddly enough the only thing I can think of that pushes the 32MB limit would 
> be compiling, mozilla, and some of the window managers, eg kde2, and even 
> then couldn't you just allocate the processes 2 sections apart until you hit 
> 128 processes (gives you 128 processes of 64MB), then fill in the gaps if you 
> can (ok won't work in all cases but would cover most desktop users, certainly 
> wouldn't work for servers)

I don't think you can do that.  All small procs must live in the 0-32Mb 
virtual memory region  (above that address and the cpu doesn't map in the 
process-specific tag when doing VA translations, so the process needs its 
own tlb tables.

> 
> (of course I could be speaking rubbish, I've only glanced over the fcs stuff 
> in the arm arm)

So could I... ;-)