Re: [PATCH v3 10/12] mm/collapse: work out the orders a VMA allows once per VMA

[Kiryl Shutsemau <kirill@shutemov.name>]

On Wed, Sep 23, 2026 at 03:19:36PM +0200, David Hildenbrand (Arm) wrote:
 [...]
> 
> > +		/* One mask for the whole VMA */
> > +		orders = collapse_possible_orders(vma, vma->vm_flags,
> > +						  cc->policy.tva_type);
> 
> cc->policy.tva_type is always sattic here, no?

Yes, TVA_KHUGEPAGED.

> > @@ -3252,9 +3254,11 @@ int madvise_collapse(struct vm_area_struct *vma, unsigned long start,
> >  			}
> >  			vma = found;
> >  			hend = min(hend, vma->vm_end & HPAGE_PMD_MASK);
> > +			orders = collapse_possible_orders(vma, vma->vm_flags,
> > +							  cc->policy.tva_type);
> 
> 
> That's always TVA_FORCED_COLLAPSE, no?

Yep.

> It's a shame we cannot get rid of cc->policy.tva_type because we need it for
> hugepage_vma_revalidate to calculate orders. Which sucks a bit.

It stays for the engine, not the callers.

The fixup I will fold in v4 is below.

diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index 2a10a30055c6..8fad934befe2 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -2905,7 +2905,7 @@ static void collapse_scan_mm_slot(unsigned int progress_max,
 		}
 		/* One mask for the whole VMA */
 		orders = collapse_possible_orders(vma, vma->vm_flags,
-						  cc->policy.tva_type);
+						  TVA_KHUGEPAGED);
 		if (!orders) {
 			cc->progress++;
 			continue;
diff --git a/mm/madvise.c b/mm/madvise.c
index 33bcd390ce43..3087640ddaad 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -1010,7 +1010,7 @@ static int madvise_collapse(struct madvise_behavior *madv_behavior)
 			vma = found;
 			hend = min(hend, vma->vm_end & HPAGE_PMD_MASK);
 			orders = collapse_possible_orders(vma, vma->vm_flags,
-							  cc->policy.tva_type);
+							  TVA_FORCED_COLLAPSE);
 		}
 
 		result = collapse_scan_pmd(vma, addr, cc, orders);

-- 
  Kiryl Shutsemau / Kirill A. Shutemov