Re: [PATCH 3/3] mm/mmap: Regression fix for unmapped_area{_topdown}

["Liam R. Howlett" <Liam.Howlett@Oracle.com>]

* Liam R. Howlett <Liam.Howlett@Oracle.com> [230414 13:26]:
> * Edgecombe, Rick P <rick.p.edgecombe@intel.com> [230414 12:27]:
> > On Fri, 2023-04-14 at 10:57 -0400, Liam R. Howlett wrote:<br>
> > > +       tmp = mas_next(&mas, ULONG_MAX);
> > > +       if (tmp && (tmp->vm_flags & VM_GROWSDOWN)) {
> > 
> > Why also check VM_GROWSDOWN here (and VM_GROWSUP below)?
> > vm_start/end_gap() already have checks inside.
> 
> An artifact of a plan that was later abandoned.
> 
> > 
> > > +               if (vm_start_gap(tmp) < gap + length - 1) {
> > > +                       low_limit = tmp->vm_end;
> > > +                       mas_reset(&mas);
> > > +                       goto retry;
> > > +               }
> > > +       } else {
> > > +               tmp = mas_prev(&mas, 0);
> > > +               if (tmp && (tmp->vm_flags & VM_GROWSUP) &&
> > > +                   vm_end_gap(tmp) > gap) {
> > > +                       low_limit = vm_end_gap(tmp); 
> > > +                       mas_reset(&mas);
> > > +                       goto retry; 
> > > +               }
> > > +       } 
> > > +
> > 
> > Could it be like this?
> 
> Yes, I'll make this change.  Thanks for the suggestion.


Wait, I like how it is.

In my version, if there is a stack that is VM_GROWSDOWN there, but does
not intercept the gap, then I won't check the prev.. in yours, we will
never avoid checking prev.

> 
> > 
> > tmp = mas_next(&mas, ULONG_MAX);
> > if (tmp && vm_start_gap(tmp) < gap + length - 1) {
> > 		low_limit = tmp->vm_end;
> > 		mas_reset(&mas);
> > 		goto retry;
> > 	}
> > } else {
> > 	tmp = mas_prev(&mas, 0);
> > 	if (tmp && vm_end_gap(tmp) > gap) {
> > 		low_limit = vm_end_gap(tmp);
> > 		mas_reset(&mas);
> > 		goto retry;
> > 	}
> > }
> >