Re: [PATCH] memcg: use ID in page_cgroup

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

On Tue, 24 Aug 2010 00:47:50 -0700
Greg Thelen <gthelen@google.com> wrote:

> KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> writes:
> 
> > I have an idea to remove page_cgroup->page pointer, 8bytes reduction per page.
> > But it will be after this work.
> > ==
> > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> >
> > Now, addresses of memory cgroup can be calculated by their ID without complex.
> > This patch relplaces pc->mem_cgroup from a pointer to a unsigned short.
> > On 64bit architecture, this offers us more 6bytes room per page_cgroup.
> > Use 2bytes for blkio-cgroup's page tracking. More 4bytes will be used for
> > some light-weight concurrent access.
> >
> > We may able to move this id onto flags field but ...go step by step.
> >
> > Changelog: 20100811
> >  - using new rcu APIs, as rcu_dereference_check() etc.
> > Changelog: 20100804
> >  - added comments to page_cgroup.h
> > Changelog: 20100730
> >  - fixed some garbage added by debug code in early stage
> >
> > Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> > ---
> >  include/linux/page_cgroup.h |    6 ++++-
> >  mm/memcontrol.c             |   52 +++++++++++++++++++++++++++-----------------
> >  mm/page_cgroup.c            |    2 -
> >  3 files changed, 38 insertions(+), 22 deletions(-)
> >
> > Index: mmotm-0811/include/linux/page_cgroup.h
> > ===================================================================
> > --- mmotm-0811.orig/include/linux/page_cgroup.h
> > +++ mmotm-0811/include/linux/page_cgroup.h
> > @@ -9,10 +9,14 @@
> >   * page_cgroup helps us identify information about the cgroup
> >   * All page cgroups are allocated at boot or memory hotplug event,
> >   * then the page cgroup for pfn always exists.
> > + *
> > + * TODO: It seems ID for cgroup can be packed into "flags". But there will
> > + * be race between assigning ID <-> set/clear flags. Please be careful.
> >   */
> >  struct page_cgroup {
> >  	unsigned long flags;
> > -	struct mem_cgroup *mem_cgroup;
> > +	unsigned short mem_cgroup;	/* ID of assigned memory cgroup */
> > +	unsigned short blk_cgroup;	/* Not Used..but will be. */
> >  	struct page *page;
> >  	struct list_head lru;		/* per cgroup LRU list */
> >  };
> > Index: mmotm-0811/mm/page_cgroup.c
> > ===================================================================
> > --- mmotm-0811.orig/mm/page_cgroup.c
> > +++ mmotm-0811/mm/page_cgroup.c
> > @@ -15,7 +15,7 @@ static void __meminit
> >  __init_page_cgroup(struct page_cgroup *pc, unsigned long pfn)
> >  {
> >  	pc->flags = 0;
> > -	pc->mem_cgroup = NULL;
> > +	pc->mem_cgroup = 0;
> >  	pc->page = pfn_to_page(pfn);
> >  	INIT_LIST_HEAD(&pc->lru);
> >  }
> > Index: mmotm-0811/mm/memcontrol.c
> > ===================================================================
> > --- mmotm-0811.orig/mm/memcontrol.c
> > +++ mmotm-0811/mm/memcontrol.c
> > @@ -300,12 +300,13 @@ static atomic_t mem_cgroup_num;
> >  #define NR_MEMCG_GROUPS (CONFIG_MEM_CGROUP_MAX_GROUPS + 1)
> >  static struct mem_cgroup *mem_cgroups[NR_MEMCG_GROUPS] __read_mostly;
> >  
> > -/* Must be called under rcu_read_lock */
> > -static struct mem_cgroup *id_to_memcg(unsigned short id)
> > +/* Must be called under rcu_read_lock, set safe==true if under lock */
> > +static struct mem_cgroup *id_to_memcg(unsigned short id, bool safe)
> >  {
> >  	struct mem_cgroup *ret;
> >  	/* see mem_cgroup_free() */
> > -	ret = rcu_dereference_check(mem_cgroups[id], rch_read_lock_held());
> > +	ret = rcu_dereference_check(mem_cgroups[id],
> > +				rch_read_lock_held() || safe);
> >  	if (likely(ret && ret->valid))
> >  		return ret;
> >  	return NULL;
> > @@ -381,7 +382,12 @@ struct cgroup_subsys_state *mem_cgroup_c
> >  static struct mem_cgroup_per_zone *
> >  page_cgroup_zoneinfo(struct page_cgroup *pc)
> >  {
> > -	struct mem_cgroup *mem = pc->mem_cgroup;
> > +	/*
> > +	 * The caller should guarantee this "pc" is under lock. In typical
> > +	 * case, this function is called by lru function with zone->lru_lock.
> > +	 * It is a safe access.
> > +	 */
> > +	struct mem_cgroup *mem = id_to_memcg(pc->mem_cgroup, true);
> >  	int nid = page_cgroup_nid(pc);
> >  	int zid = page_cgroup_zid(pc);
> >  
> > @@ -723,6 +729,11 @@ static inline bool mem_cgroup_is_root(st
> >  	return (mem == root_mem_cgroup);
> >  }
> >  
> > +static inline bool mem_cgroup_is_rootid(unsigned short id)
> > +{
> > +	return (id == 1);
> > +}
> > +
> >  /*
> >   * Following LRU functions are allowed to be used without PCG_LOCK.
> >   * Operations are called by routine of global LRU independently from memcg.
> > @@ -755,7 +766,7 @@ void mem_cgroup_del_lru_list(struct page
> >  	 */
> >  	mz = page_cgroup_zoneinfo(pc);
> >  	MEM_CGROUP_ZSTAT(mz, lru) -= 1;
> > -	if (mem_cgroup_is_root(pc->mem_cgroup))
> > +	if (mem_cgroup_is_rootid(pc->mem_cgroup))
> >  		return;
> >  	VM_BUG_ON(list_empty(&pc->lru));
> >  	list_del_init(&pc->lru);
> > @@ -782,7 +793,7 @@ void mem_cgroup_rotate_lru_list(struct p
> >  	 */
> >  	smp_rmb();
> >  	/* unused or root page is not rotated. */
> > -	if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup))
> > +	if (!PageCgroupUsed(pc) || mem_cgroup_is_rootid(pc->mem_cgroup))
> >  		return;
> >  	mz = page_cgroup_zoneinfo(pc);
> >  	list_move(&pc->lru, &mz->lists[lru]);
> > @@ -808,7 +819,7 @@ void mem_cgroup_add_lru_list(struct page
> >  	mz = page_cgroup_zoneinfo(pc);
> >  	MEM_CGROUP_ZSTAT(mz, lru) += 1;
> >  	SetPageCgroupAcctLRU(pc);
> > -	if (mem_cgroup_is_root(pc->mem_cgroup))
> > +	if (mem_cgroup_is_rootid(pc->mem_cgroup))
> >  		return;
> >  	list_add(&pc->lru, &mz->lists[lru]);
> >  }
> > @@ -1497,7 +1508,7 @@ void mem_cgroup_update_file_mapped(struc
> >  		return;
> >  
> >  	lock_page_cgroup(pc);
> > -	mem = pc->mem_cgroup;
> > +	mem = id_to_memcg(pc->mem_cgroup, true);
> >  	if (!mem || !PageCgroupUsed(pc))
> >  		goto done;
> >  
> > @@ -1885,14 +1896,14 @@ struct mem_cgroup *try_get_mem_cgroup_fr
> >  	pc = lookup_page_cgroup(page);
> >  	lock_page_cgroup(pc);
> >  	if (PageCgroupUsed(pc)) {
> > -		mem = pc->mem_cgroup;
> > +		mem = id_to_memcg(pc->mem_cgroup, true);
> >  		if (mem && !css_tryget(&mem->css))
> >  			mem = NULL;
> >  	} else if (PageSwapCache(page)) {
> >  		ent.val = page_private(page);
> >  		id = lookup_swap_cgroup(ent);
> >  		rcu_read_lock();
> > -		mem = id_to_memcg(id);
> > +		mem = id_to_memcg(id, false);
> >  		if (mem && !css_tryget(&mem->css))
> >  			mem = NULL;
> >  		rcu_read_unlock();
> > @@ -1921,7 +1932,7 @@ static void __mem_cgroup_commit_charge(s
> >  		return;
> >  	}
> >  
> > -	pc->mem_cgroup = mem;
> > +	pc->mem_cgroup = css_id(&mem->css);
> >  	/*
> >  	 * We access a page_cgroup asynchronously without lock_page_cgroup().
> >  	 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
> > @@ -1979,7 +1990,7 @@ static void __mem_cgroup_move_account(st
> >  	VM_BUG_ON(PageLRU(pc->page));
> >  	VM_BUG_ON(!PageCgroupLocked(pc));
> 
> Should this be VM_BUG_ON(!rcu_read_lock_held())?  I suspect that
> mem_cgroup_move_account() should grab rcu read lock (see my comment below).
> 

No. (see below)


> >  	VM_BUG_ON(!PageCgroupUsed(pc));
> > -	VM_BUG_ON(pc->mem_cgroup != from);
> > +	VM_BUG_ON(id_to_memcg(pc->mem_cgroup, true) != from);
> >  
> >  	if (PageCgroupFileMapped(pc)) {
> >  		/* Update mapped_file data for mem_cgroup */
> > @@ -1994,7 +2005,7 @@ static void __mem_cgroup_move_account(st
> >  		mem_cgroup_cancel_charge(from);
> >  
> >  	/* caller should have done css_get */
> > -	pc->mem_cgroup = to;
> > +	pc->mem_cgroup = css_id(&to->css);
> >  	mem_cgroup_charge_statistics(to, pc, true);
> >  	/*
> >  	 * We charges against "to" which may not have any tasks. Then, "to"
> > @@ -2014,11 +2025,11 @@ static int mem_cgroup_move_account(struc
> >  {
> >  	int ret = -EINVAL;
> >  	lock_page_cgroup(pc);
> 
> Should this be rcu_read_lock() instead of lock_page_cgroup()?
> 

No. 
 - lock_page_cgroup() is for keeping page_cgroup's status stable.
 - rcu_read_lock() is for delaying to discard mem_cgroup object.

rcu_read_lock() is just for delaying to discard object, not for avoiding
racy updates. All _updates_ requires proper lock or speculative logic as
atomic_inc_not_zero() etc... Basically, RCU is for avoiding use-after-free.

Thanks,
-Kame