Re: [PATCH v4 1/3] mm: pagewalk: simplify hugepage boundary

[Lance Yang <ioworker0@gmail.com>]

From: Lance Yang <lance.yang@linux.dev>


On Wed, 24 Dec 2025 15:50:34 +0800, Baolin Wang wrote:
> Hi Andrew,
> 
> On 2025/11/28 15:01, Sahil Chandna wrote:
> > Replace hugepage boundary computation with ALIGN() helper instead of
> > an open coded expression. This helps to improves code readability.
> > 
> > This was flagged by Coccinelle (misc/minmax.cocci) as an opportunity
> > to use min(), after which the boundary computation was updated following
> > review suggestions.
> > 
> > Found by: make coccicheck MODE=report M=mm/
> > No functional change intended.
> > 
> > Acked-by: David Hildenbrand (Red Hat) <david@kernel.org>
> > Suggested-by: David Hildenbrand (Red Hat) <david@kernel.org>
> > Suggested-by: Matthew Wilcox <willy@infradead.org>
> > Suggested-by: David Laight <david.laight.linux@gmail.com>
> > Signed-off-by: Sahil Chandna <chandna.sahil@gmail.com>
> > ---
> >   mm/pagewalk.c | 3 +--
> >   1 file changed, 1 insertion(+), 2 deletions(-)
> > 
> > diff --git a/mm/pagewalk.c b/mm/pagewalk.c
> > index 9f91cf85a5be..9fd59d517f37 100644
> > --- a/mm/pagewalk.c
> > +++ b/mm/pagewalk.c
> > @@ -312,8 +312,7 @@ static int walk_pgd_range(unsigned long addr, unsigned long end,
> >   static unsigned long hugetlb_entry_end(struct hstate *h, unsigned long addr,
> >   				       unsigned long end)
> >   {
> > -	unsigned long boundary = (addr & huge_page_mask(h)) + huge_page_size(h);
> > -	return boundary < end ? boundary : end;
> > +	return min(ALIGN(addr, huge_page_size(h)), end);
> >   }
> 
> Please drop this patch from the mm-new branch, as it causes 
> 'run_vmtests.sh' to hang. Specifically, it leads to the system hanging 
> when executing hugepage-vmemmap test, because the program falls into an 
> infinite loop in walk_hugetlb_range() and cannot break out.

Good catch! The problem is that ALIGN() returns addr itself when already
aligned, causing the infinite loop ...

> 
> This patch does introduce functional changes and makes an incorrect 
> assumption that the 'end' must be aligned to the hugepage size. However, 

Yep. This patch is not equivalent to the original code when addr is
already aligned :)

> this is not necessarily the case. For example, see how pagemap_read() 
> calculates the 'end':
> 
> "
> end = start_vaddr + ((count / PM_ENTRY_BYTES) << PAGE_SHIFT);
> "
> 
> Revert this patch, mm selftests work well.