Re: [RFC][BUGFIX][PATCH] memcg: fix underflow of mapped_file stat

[Daisuke Nishimura]

> > > @@ -2517,65 +2519,70 @@ int mem_cgroup_prepare_migration(struct 
> > >  		css_get(&mem->css);
> > >  	}
> > >  	unlock_page_cgroup(pc);
> > > -
> > > -	if (mem) {
> > > -		ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
> > > -		css_put(&mem->css);
> > > -	}
> > > -	*ptr = mem;
> > > +	/*
> > > +	 * If the page is uncharged before migration (removed from radix-tree)
> > > +	 * we return here.
> > > +	 */
> > > +	if (!mem)
> > > +		return 0;
> > > +	ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false);
> > > +	css_put(&mem->css); /* drop extra refcnt */
> > it should be:
> > 
> > 	*ptr = mem;
> > 	ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false);
> > 	css_put(&mem->css);
> > 
> > as Andrea has fixed already.
> > 
> Ah, yes. I'll rebase this onto Andrea's fix.
> 
> 
> 
> > > +	if (ret)
We should check "if (ret || !*ptr)" not to do commit in !*ptr case.

> > > + 	 * Considering ANON pages, we can't depend on lock_page.
> > > + 	 * If a page may be unmapped before it's remapped, new page's
> > > + 	 * mapcount will not increase. (case that mapcount 0->1 never occur.)
> > > + 	 * PageCgroupUsed() and SwapCache checks will be done.
> > > + 	 *
> > > + 	 * Once mapcount goes to 1, our hook to page_remove_rmap will do
> > > + 	 * enough jobs.
> > > + 	 */
> > > +	if (PageAnon(used) && !page_mapped(used))
> > > +		mem_cgroup_uncharge_page(used);
> > mem_cgroup_uncharge_page() does the same check :)
> > 
> Ok. I'll fix.
> 
Considering more, we'd better to check PageAnon() at least not to call
mem_cgroup_uncharge_page() for cache page.


Thanks,
Daisuke Nishimura.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/