Re: [PATCH v3] ksm: use range-walk function to jump over holes in scan_get_next_rmap_item

[David Hildenbrand]

On 21.10.25 05:00, Pedro Demarchi Gomes wrote:
> On 10/17/25 19:23, David Hildenbrand wrote:
>
> > This patch does to much in a single patch which makes it
> > rather hard to review.
> >
> > As a first step, we should focus on leaving most of
> > scan_get_next_rmap_item() alone and only focus on replacing
> > folio_walk by walk_page_range_vma().
> >
> > Follow-up cleanups could try cleaning up scan_get_next_rmap_item()
> > -- and boy oh boy, does that function scream for quite some cleanups.
> >
> > This is something minimal based on your v3. I applied plenty of more
> > cleanups and I wish we could further shrink the pmd_entry function,
> > but I have to give up for today (well, it's already tomorrow :) ).
>
> Should I send a v4 to be applied on top of your minimal patch? This
> v4 would eliminate the need of the for_each_vma using the test_walk
> callback like the previous versions.

It would be good if you could test the rework I sent and see if you want 
to do any tweaks to it. It was a rather quick rework on my side.


Then resend that as v4, which is then minimal and we can reasonable add 
Fixes: + Cc: stable.

Right from that start we used follow_page() on each individual address.

So likely best to add

	Fixes: 31dbd01f3143 ("ksm: Kernel SamePage Merging")

Once that fix is in you can send further cleanups that are independent 
of the fix itself, like removing the for_each_vma() etc.

--
Cheers

David / dhildenb