Re: [PATCH 2/4] mm: migrate: convert numamigrate_isolate_page() to numamigrate_isolate_folio()

[Kefeng Wang]

On 2023/8/2 20:30, Matthew Wilcox wrote:
> On Wed, Aug 02, 2023 at 05:53:44PM +0800, Kefeng Wang wrote:
> > -static int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page)
> > +static int numamigrate_isolate_folio(pg_data_t *pgdat, struct folio *folio)
> >   {
> > -	int nr_pages = thp_nr_pages(page);
> > -	int order = compound_order(page);
> > +	int nr_pages = folio_nr_pages(folio);
> > +	int order = folio_order(folio);
> >   
> > -	VM_BUG_ON_PAGE(order && !PageTransHuge(page), page);
> > +	VM_BUG_ON_FOLIO(order && !folio_test_pmd_mappable(folio), folio);
>
> I don't know why we have this assertion.  I would be inclined to delete
> it as part of generalising the migration code to handle arbitrary sizes
> of folio, rather than assert that we only support PMD size folios.

Ok, will drop it.
> >   	/* Do not migrate THP mapped by multiple processes */
> > -	if (PageTransHuge(page) && total_mapcount(page) > 1)
> > +	if (folio_test_pmd_mappable(folio) && folio_estimated_sharers(folio) > 1)
> >   		return 0;
>
> I don't know if this is the right logic.  We've willing to move folios
> mapped by multiple processes, as long as they're smaller than PMD size,
> but once they get to PMD size they're magical and can't be moved?

It seems that the logical is introduced by commit 04fa5d6a6547 ("mm:
migrate: check page_count of THP before migrating") and refactor by
340ef3902cf2 ("mm: numa: cleanup flow of transhuge page migration"),


  "Hugh Dickins pointed out that migrate_misplaced_transhuge_page() does
  not check page_count before migrating like base page migration and
  khugepage.  He could not see why this was safe and he is right."

For now, there is no migrate_misplaced_transhuge_page() and base/thp
page migrate's path is unified, there is a check(for old/new kernel) in 
migrate_misplaced_page(),

 "Don't migrate file pages that are mapped in multiple processes
 with execute permissions as they are probably shared libraries."

We could drop the above check in numamigrate_isolate_page(), but
according to 04fa5d6a6547, maybe disable migrate page shared by
multi-process during numa balance for both base/thp page.


>