[PATCH v3 4/6] mm/vmalloc: make set_area_direct_map HUGE_VMAP friendly

[Mike Rapoport]

On Tue, Sep 08, 2026 at 12:41:48PM +0200, David Hildenbrand (Arm) wrote:
> On 9/3/26 11:28, Mike Rapoport (Microsoft) wrote:
> > set_area_direct_map() always updates direct map alias permissions in
> > single page increments.
> > 
> > For HUGE_VMAP areas it's suboptimal. Not only the loop in
> > set_area_direct_map() needlessly has more iterations (e.g times 512 on
> > x86), but it also causes fragmentation of the direct map that could be
> > avoided for the HUGE_VMAP areas populated with large pages.
> > 
> > All pages in an area are always of the same order: either same-order
> > large pages when VM_ALLOW_HUGE_VMAP is set and all huge pages were
> > successfully allocated, or order-0 page when VM_ALLOW_HUGE_VMAP is
> > cleared or when huge pages allocation fails and fallback path is taken.
> > 
> > Instead of updating the direct map permissions for every order-0 page in
> > an area, use the area's page_order as the loop increment and update the
> > large pages in one call to set_direct_map_{invalid,default}_noflush().
> > 
> > Signed-off-by: Mike Rapoport (Microsoft) <rppt at kernel.org>
> > ---
> >  mm/vmalloc.c | 13 ++++++++-----
> >  1 file changed, 8 insertions(+), 5 deletions(-)
> > 
> > diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> > index 5506b180f5c27..6ed6c160abed7 100644
> > --- a/mm/vmalloc.c
> > +++ b/mm/vmalloc.c
> > @@ -3367,12 +3367,15 @@ static inline void set_area_direct_map(const struct vm_struct *area,
> >  				       int (*set_direct_map)(struct page *page,
> >  							     unsigned int nr))
> >  {
> > -	unsigned long i;
> > +	unsigned int nr = (1U << vm_area_page_order(area));
> > +
> > +	for (unsigned long i = 0; i < area->nr_pages; i += nr) {
> > +		if (page_address(area->pages[i])) {
> > +			int err = set_direct_map(area->pages[i], nr);
> 
> Do we really expect some pages to have a directmap and others not?

Every page in the area can be allocated separately and
__vmalloc_area_node() has

	if (!(gfp_mask & (GFP_DMA | GFP_DMA32)))
		gfp_mask |= __GFP_HIGHMEM;

so generally speaking yes :)

> -- 
> Cheers,
> 
> David

-- 
Sincerely yours,
Mike.