Re: [PATCH] mark_page_accessed() in do_swap_page() move latter than memcg charge

[KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>]

> (CC to Rik and Nick)
> 
> Hi
> 
> Thank you reviewing.
> 
> > > mark_page_accessed() update reclaim_stat statics.
> > > but currently, memcg charge is called after mark_page_accessed().
> > > 
> > > then, mark_page_accessed() don't update memcg statics correctly.
> > 
> > Statics?  "Stats" is a good abbreviation for statistics,
> > but statics are something else.
> 
> Doh! your are definitly right. thanks.
> 
> > > ---
> > >  mm/memory.c |    4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > Index: b/mm/memory.c
> > > ===================================================================
> > > --- a/mm/memory.c
> > > +++ b/mm/memory.c
> > > @@ -2426,8 +2426,6 @@ static int do_swap_page(struct mm_struct
> > >  		count_vm_event(PGMAJFAULT);
> > >  	}
> > >  
> > > -	mark_page_accessed(page);
> > > -
> > >  	lock_page(page);
> > >  	delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
> > >  
> > > @@ -2480,6 +2478,8 @@ static int do_swap_page(struct mm_struct
> > >  		try_to_free_swap(page);
> > >  	unlock_page(page);
> > >  
> > > +	mark_page_accessed(page);
> > > +
> > >  	if (write_access) {
> > >  		ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
> > >  		if (ret & VM_FAULT_ERROR)
> > 
> > This catches my eye, because I'd discussed with Nick and was going to
> > send in a patch which entirely _removes_ this mark_page_accessed call
> > from do_swap_page (and replaces follow_page's mark_page_accessed call
> > by a pte_mkyoung): they seem inconsistent to me, in the light of
> > bf3f3bc5e734706730c12a323f9b2068052aa1f0 mm: don't mark_page_accessed
> > in fault path.
> 
> Actually, bf3f3bc5e734706730c12a323f9b2068052aa1f0 only remove 
> the mark_page_accessed() in filemap_fault().
> current mmotm's do_swap_page() still have mark_page_accessed().
> 
> but your suggestion is very worth.
> ok, I'm thinking and sorting out again.
> 
> Rik's commit 9ff473b9a72942c5ac0ad35607cae28d8d59ed7a vmscan: evict streaming IO first
> does "reclaim stastics don't only update at reclaim, but also at fault and read/write.
> it makes proper anon/file reclaim balancing stastics value before starting actual reclaim".
> and it depend on fault path calling mark_page_accessed(). 
> 
> Then, we need following change. I think.
> 
>   - Remove calling mark_page_accessed() in do_swap_page().
>     it makes consistency against filemap_fault().
>   - Add calling update_page_reclaim_stat() into do_swap_page() and 
>     filemap_fault().
> 
> Am I overlooking something?

Doh! please ignore last mail's patch. I forgot grab zone->lru_lock.
it's perfectly buggy.

I'll make it again tommorow.


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>