Re: + mm-migratec-rework-migration_entry_wait-to-not-take-a-pageref.patch added to -mm tree

[Alistair Popple <apopple@nvidia.com>]

On Saturday, 11 December 2021 3:34:36 PM AEDT Matthew Wilcox wrote:
> On Thu, Nov 18, 2021 at 01:54:27PM -0800, akpm@linux-foundation.org wrote:
> > +++ a/mm/migrate.c
> > @@ -305,15 +305,7 @@ void __migration_entry_wait(struct mm_st
> >  	page = pfn_swap_entry_to_page(entry);
> >  	page = compound_head(page);
> >  
> > -	/*
> > -	 * Once page cache replacement of page migration started, page_count
> > -	 * is zero; but we must not call put_and_wait_on_page_locked() without
> > -	 * a ref. Use get_page_unless_zero(), and just fault again if it fails.
> > -	 */
> > -	if (!get_page_unless_zero(page))
> > -		goto out;
> > -	pte_unmap_unlock(ptep, ptl);
> > -	put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE);
> > +	migration_entry_wait_on_locked(page_folio(page), ptep, ptl);
> 
> This is clearly bogus.  The 'page = compound_head(page)' line should
> be deleted.
> 
> But I think we should go further and turn this into:
> 
> 	migration_entry_wait_on_locked(entry, ptep, ptl);
> 
> Neither caller has anything useful to do with the page any more.

Thanks for spotting that, I have posted a new version which does that.