Re: [patch]vmscan: protect exectuable page from inactive list scan

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

> On Wed, 2010-09-29 at 18:17 +0800, Johannes Weiner wrote:
> > On Wed, Sep 29, 2010 at 10:57:40AM +0800, Shaohua Li wrote:
> > > With commit 645747462435, pte referenced file page isn't activated in inactive
> > > list scan. For VM_EXEC page, if it can't get a chance to active list, the
> > > executable page protect loses its effect. We protect such page in inactive scan
> > > here, now such page will be guaranteed cached in a full scan of active and
> > > inactive list, which restores previous behavior.
> > 
> > This change was in the back of my head since the used-once detection
> > was merged but there were never any regressions reported that would
> > indicate a requirement for it.
> The executable page protect is to improve responsibility. I would expect
> it's hard for user to report such regression. 

Seems strange. 8cab4754d24a0f was introduced for fixing real world problem.
So, I wonder why current people can't feel the same lag if it is.


> > Does this patch fix a problem you observed?
> No, I haven't done test where Fengguang does in commit 8cab4754d24a0f.

But, I am usually not against a number. If you will finished to test them I'm happy :)


> 
> > > --- a/mm/vmscan.c
> > > +++ b/mm/vmscan.c
> > > @@ -608,8 +608,15 @@ static enum page_references page_check_references(struct page *page,
> > >  		 * quickly recovered.
> > >  		 */
> > >  		SetPageReferenced(page);
> > > -
> > > -		if (referenced_page)
> > > +		/*
> > > +		 * Identify pte referenced and file-backed pages and give them
> > > +		 * one trip around the active list. So that executable code get
> > > +		 * better chances to stay in memory under moderate memory
> > > +		 * pressure. JVM can create lots of anon VM_EXEC pages, so we
> > > +		 * ignore them here.
> > 
> > PTE-referenced PageAnon() pages are activated unconditionally a few
> > lines further up, so the page_is_file_cache() check filters only shmem
> > pages.  I doubt this was your intention...?
> This is intented. the executable page protect is just to protect
> executable file pages. please see 8cab4754d24a0f.

8cab4754d24a0f was using !PageAnon() but your one are using page_is_file_cache.
8cab4754d24a0f doesn't tell us the reason of the change, no?



--
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>