Re: [RFC][PATCH 2/5] memcg : handle swap cache

From: KAMEZAWA Hiroyuki
Date: Tue Nov 04 2008 - 19:08:37 EST


On Tue, 4 Nov 2008 19:28:22 +0900
Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx> wrote:

> On Tue, 4 Nov 2008 18:04:29 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> > On Tue, 4 Nov 2008 17:42:01 +0900
> > Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx> wrote:
> >
> > > > +#ifdef CONFIG_SWAP
> > > > +int mem_cgroup_cache_charge_swapin(struct page *page,
> > > > + struct mm_struct *mm, gfp_t mask)
> > > > +{
> > > > + int ret = 0;
> > > > +
> > > > + if (mem_cgroup_subsys.disabled)
> > > > + return 0;
> > > > + if (unlikely(!mm))
> > > > + mm = &init_mm;
> > > > +
> > > > + ret = mem_cgroup_charge_common(page, mm, mask,
> > > > + MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL);
> > > > + /*
> > > > + * The page may be dropped from SwapCache because we don't have
> > > > + * lock_page().This may cause charge-after-uncharge trouble.
> > > > + * Fix it up here. (the caller have refcnt to this page and
> > > > + * page itself is guaranteed not to be freed.)
> > > > + */
> > > > + if (ret && !PageSwapCache(page))
> > > > + mem_cgroup_uncharge_swapcache(page);
> > > > +
> > > Hmm.. after [5/5], mem_cgroup_cache_charge_swapin has 'locked' parameter,
> > > calls lock_page(if !locked), and checks PageSwapCache under page lock.
> > >
> > > Why not doing it in this patch?
> > >
> >
> > My intention is to guard swap_cgroup by lock_page() against SwapCache.
> > In Mem+Swap controller. we get "memcg" from information in page->private.
> > I think we need lock_page(), there.
> >
> > But here, we don't refer page->private information.
> > I think we don't need lock_page() because there is no inofrmation we depends on.
> >
> I just thought it would be simpler to check PageSwapCache after holding
> page lock rather than to handle the case that the page might be removed from
> swap cache.
>
> And to be honest, I can't understand the "charge-after-uncharge trouble".
> Could you explain more?
>
Maybe typical case is following.
__delete_from_swapcache can happen while the page is unlocked.
==
some other thread.
page = shmem_swapin()
swapin_readahead();
# page is SwapCache here.
# but this page is not locked.
___delete_from_swapcache(page)
# This is not SwapCache. => uncharge swapcache.
mem_cgroup_charge_cache_swapin();
{
charge(); # charged this page but we don't know this is still swapcache.
if (!PageSwapCache(page)) {
# Oh we should unroll this.
}
}
=

Thanks,
-Kame















--
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/