Re: [PATCH -V3] mm: Add PageLayzyFree() helper functions for MADV_FREE

[Huang\, Ying]

Michal Hocko <mhocko@xxxxxxxxxx> writes:

> On Mon 09-03-20 09:55:38, David Hildenbrand wrote:
>> On 09.03.20 03:17, Huang, Ying wrote:
> [...]
>> > @@ -1235,7 +1234,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
>> >  		 * Try to allocate it some swap space here.
>> >  		 * Lazyfree page could be freed directly
>> >  		 */
>> > -		if (PageAnon(page) && PageSwapBacked(page)) {
>> > +		if (PageAnon(page) && !__PageLazyFree(page)) {
>> >  			if (!PageSwapCache(page)) {
>> >  				if (!(sc->gfp_mask & __GFP_IO))
>> >  					goto keep_locked;
>> > @@ -1411,7 +1410,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
>> >  			}
>> >  		}
>> >  
>> > -		if (PageAnon(page) && !PageSwapBacked(page)) {
>> > +		if (PageLazyFree(page)) {
>> >  			/* follow __remove_mapping for reference */
>> >  			if (!page_ref_freeze(page, 1))
>> >  				goto keep_locked;
>> > 
>> 
>> I still prefer something like
>> 
>> diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
>> index fd6d4670ccc3..7538501230bd 100644
>> --- a/include/linux/page-flags.h
>> +++ b/include/linux/page-flags.h
>> @@ -63,6 +63,10 @@
>>   * page_waitqueue(page) is a wait queue of all tasks waiting for the page
>>   * to become unlocked.
>>   *
>> + * PG_swapbacked used with anonymous pages (PageAnon()) indicates that a
>> + * page is backed by swap. Anonymous pages without PG_swapbacked are
>> + * pages that can be lazily freed (e.g., MADV_FREE) on demand.
>> + *
>>   * PG_uptodate tells whether the page's contents is valid.  When a read
>>   * completes, the page becomes uptodate, unless a disk I/O error happened.
>>   *
>> 
>> and really don't like the use of !__PageLazyFree() instead of PageSwapBacked().
>
> I have to say that I do not have a strong opinion about helper
> functions. In general I tend to be against adding them unless there is a
> very good reason for them. This particular patch is in a gray zone a bit.
>
> There are few places which are easier to follow but others sound like,
> we have a hammer let's use it. E.g. shrink_page_list path above.

I can remove all these places.  Only keep the helpful places.

> There is a clear comment explaining PageAnon && PageSwapBacked check
> being LazyFree related but do I have to know that this is LazyFree
> path? I believe that seeing PageSwapBacked has a more meaning to me
> because it tells me that anonymous pages without a backing store
> doesn't really need swap entry.  This happens to be Lazy free related
> today but with a heavy overloading of our flags this might differ in
> the future. You have effectively made a more generic description more
> specific without a very good reason.

Yes.  The following piece isn't lazy free specific.  We can keep use PageSwapBacked().

 @@ -1235,7 +1234,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
  		 * Try to allocate it some swap space here.
  		 * Lazyfree page could be freed directly
  		 */
 -		if (PageAnon(page) && PageSwapBacked(page)) {
 +		if (PageAnon(page) && !__PageLazyFree(page)) {
  			if (!PageSwapCache(page)) {
  				if (!(sc->gfp_mask & __GFP_IO))
  					goto keep_locked;

And the following piece is lazy free specific.  I think it helps.

 @@ -1411,7 +1410,7 @@ static unsigned long shrink_page_list(struct list_head *page_list,
  			}
  		}
  
 -		if (PageAnon(page) && !PageSwapBacked(page)) {
 +		if (PageLazyFree(page)) {
  			/* follow __remove_mapping for reference */
  			if (!page_ref_freeze(page, 1))
  				goto keep_locked;
 
> On the other hand having PG_swapbacked description in page-flags.h above
> gives a very useful information which was previously hidden at the
> definition so this is a clear improvement.

Yes.  I think it's good to add document for PG_swapbacked definition.

> That being said I think that the patch is not helpful enough. I would
> much rather see a simply documentation update.