> On Thu, Oct 18, 2001 at 07:14:31PM +0100, Richard Earnshaw wrote:
> 
>  > But that doesn't solve the critical problem that for a given pmap we can't 
>  > tell if the pmap is physically shared with another process (to know this, 
>  > you need to know the proc it is part of); if it is, and we remove the 
>  > page0 entry, then we loose the whole system.
> 
> Well, there is a reference count mechanism in the pmap subsystem...
> pmap_reference() and pmap_destroy() frob it.

Doesn't help, since the whole uvmspace is shared for a vfork.  So 
pmap_reference is never called.  The relevant refcount is p_vmspae->
vm_refcnt, which as far as I can tell can't be found via a pmap handle.

R.