Re: [RFC][PATCH] Reduce size of swap_cgroup by CSS ID v2

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Tue, 10 Feb 2009 13:37:00 +0530
Balbir Singh <balbir@linux.vnet.ibm.com> wrote:

> * KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> [2009-02-09 14:55:57]:
> 
> > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> >  static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page)
> >  {
> > -	struct mem_cgroup *mem;
> > +	unsigned short id;
> > +	struct mem_cgroup *mem = NULL;
> >  	swp_entry_t ent;
> > 
> >  	if (!PageSwapCache(page))
> >  		return NULL;
> > 
> >  	ent.val = page_private(page);
> > -	mem = lookup_swap_cgroup(ent);
> > -	if (!mem)
> > -		return NULL;
> > +	id = lookup_swap_cgroup(ent);
> > +	rcu_read_lock();
> > +	mem = mem_cgroup_lookup(id);
> >  	if (!css_tryget(&mem->css))
> > -		return NULL;
> > +		mem = NULL;
> 
> This part is a bit confusing. If the page got swapped out and the CSS
> it belonged to got swapped out, we set mem to NULL. Is this so that it
> can be charged to root cgroup?
IIUC, this charge will go to "current" process's cgroup.

> If so, could you please add a comment indicating the same.
> 
Ah yes, I'll add some comments.


> > +	rcu_read_unlock();
> >  	return mem;
> >  }
> > 
> > @@ -1275,12 +1296,20 @@ int mem_cgroup_cache_charge(struct page 
> > 
> >  	if (do_swap_account && !ret && PageSwapCache(page)) {
> >  		swp_entry_t ent = {.val = page_private(page)};
> > +		unsigned short id;
> >  		/* avoid double counting */
> > -		mem = swap_cgroup_record(ent, NULL);
> > +		id = swap_cgroup_record(ent, 0);
> > +		rcu_read_lock();
> > +		mem = mem_cgroup_lookup(id);
> >  		if (mem) {
> > +			/*
> > +			 * Recorded ID can be obsolete. We avoid calling
> > +			 * css_tryget()
> > +			 */
> >  			res_counter_uncharge(&mem->memsw, PAGE_SIZE);
> >  			mem_cgroup_put(mem);
> >  		}
> 
> If !mem, do we leak charge?
No, it just means some other thread removed the ID before us.

> BTW, We no longer hold css references if the page is swapped out?
> 
The situation is a bit complicated.
When cgroup is obsolete but its mem_cgroup is alive (because of reference
from swap), css_tryget() always fails. swap_cgroup_record() is atomic
compare-and-exchange, so I think we can trust mem_cgroup_put/get refcnt
management and doesn't need to rely on css's refcnt at swap management.


Thanks,
-Kame

--
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>