> On Monday 25 June 2001  3:13 am, Jason R Thorpe wrote:
> > On Mon, Jun 25, 2001 at 12:29:33AM +0100, Chris Gilbert wrote:
> >  > Currently it actually looks like I should look into reducing the number
> >  > of calls to splx, we call splvm a hell of a lot in the pmap, I might
> >  > look at the locking down in the i386 version see if we can replace the
> >  > splvm's with it. Another major gain would be to sort out pmap_release so
> >  > it doesn't have to walk the whole of a the L1 table looking for items to
> >  > free off (we should do that in pmap_remove)
> >
> > You should need almost no calls to splvm()/splx() in the pmap module,
> > assuming you have a "real" pmap_kenter_pa()/pmap_kremove().
> 
> Currently we don't, they both call pmap_enter or pmap_remove.  Checking the 
> pmap.c history we just had wrappers inserted at some point in the past so 
> that the core kernel could do pmap_kenter and kremove with out checking for 
> their existence.  I'll take a look at sorting that out (I think I've got them 
> in a diff from richard).

Chris,

The patches I sent you did exactly what Jason suggests (and thus removed 
most of the calls to splvm/splx).

R.