Re: Low overhead patches for the memory cgroup controller (v3)

[Balbir Singh <balbir@linux.vnet.ibm.com>]

* KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> [2009-06-05 14:51:41]:

> On Fri, 5 Jun 2009 13:31:07 +0800
> Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> 
> > Here is the new version of the patch with the RFC dropped. Andrew,
> > Kame, could you please take a look. I am just about to fly out to get
> > back home tomorrow, so there might be some silence, unless I get to
> > the next WiFi enabled airport.
> > 
> > 
> > From: Balbir Singh <balbir@linux.vnet.ibm.com>
> > 
> > Changelog v3 -> v2
> > 
> > 1. Rebase to mmotm 2nd June 2009
> > 2. Test with some of the test cases recommended by Daisuke-San
> > 
> > Changelog v2 -> v1
> > 1. Fix and implement review comments.
> > 
> > Feature: Remove the overhead associated with the root cgroup
> > 
> > This patch changes the memory cgroup and removes the overhead associated
> > with accounting all pages in the root cgroup. As a side-effect, we can
> > no longer set a memory hard limit in the root cgroup.
> > 
> > A new flag is used to track page_cgroup associated with the root cgroup
> > pages. A new flag to track whether the page has been accounted or not
> > has been added as well. Flags are now set atomically for page_cgroup,
> > pcg_default_flags is now obsolete, but I've not removed it yet. It
> > provides some readability to help the code.
> > 
> > Tests Results:
> > 
> > Obtained by
> > 
> > 1. Using tmpfs for mounting filesystem
> > 2. Changing sync to be /bin/true (so that sync is not the bottleneck)
> > 3. Used -s #cpus*40 -e #cpus*40
> > 
> > Reaim
> > 		withoutpatch	patch
> > AIM9		9532.48		9807.59
> > dbase		19344.60	19285.71
> > new_dbase	20101.65	20163.13
> > shared		11827.77	11886.65
> > compute		17317.38	17420.05
> > 
> 
> A few comments.
> 
> 
> > Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
> > ---
> > 
> >  include/linux/page_cgroup.h |   12 ++++++++++++
> >  mm/memcontrol.c             |   42 ++++++++++++++++++++++++++++++++++++++----
> >  mm/page_cgroup.c            |    1 -
> >  3 files changed, 50 insertions(+), 5 deletions(-)
> > 
> > 
> > diff --git a/include/linux/page_cgroup.h b/include/linux/page_cgroup.h
> > index 7339c7b..41cc16c 100644
> > --- a/include/linux/page_cgroup.h
> > +++ b/include/linux/page_cgroup.h
> > @@ -26,6 +26,8 @@ enum {
> >  	PCG_LOCK,  /* page cgroup is locked */
> >  	PCG_CACHE, /* charged as cache */
> >  	PCG_USED, /* this object is in use. */
> > +	PCG_ROOT, /* page belongs to root cgroup */
> > +	PCG_ACCT_LRU, /* page has been accounted for */
> >  };
> >  
> >  #define TESTPCGFLAG(uname, lname)			\
> > @@ -42,9 +44,19 @@ static inline void ClearPageCgroup##uname(struct page_cgroup *pc)	\
> >  
> >  /* Cache flag is set only once (at allocation) */
> >  TESTPCGFLAG(Cache, CACHE)
> > +SETPCGFLAG(Cache, CACHE)
> >  
> >  TESTPCGFLAG(Used, USED)
> >  CLEARPCGFLAG(Used, USED)
> > +SETPCGFLAG(Used, USED)
> > +
> > +SETPCGFLAG(Root, ROOT)
> > +CLEARPCGFLAG(Root, ROOT)
> > +TESTPCGFLAG(Root, ROOT)
> > +
> > +SETPCGFLAG(AcctLru, ACCT_LRU)
> > +CLEARPCGFLAG(AcctLru, ACCT_LRU)
> > +TESTPCGFLAG(AcctLru, ACCT_LRU)
> >  
> I prefer AcctLRU rather than AcctLru. LRU is LRU or lru and not Lru through
> the kernel.

OK, I'll make that change. I agree LRU is better.

> 
> >  static inline int page_cgroup_nid(struct page_cgroup *pc)
> >  {
> > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > index a83e039..9561d10 100644
> > --- a/mm/memcontrol.c
> > +++ b/mm/memcontrol.c
> > @@ -43,6 +43,7 @@
> >  
> >  struct cgroup_subsys mem_cgroup_subsys __read_mostly;
> >  #define MEM_CGROUP_RECLAIM_RETRIES	5
> > +struct mem_cgroup *root_mem_cgroup __read_mostly;
> >  
> >  #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
> >  /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
> > @@ -197,6 +198,10 @@ enum charge_type {
> >  #define PCGF_CACHE	(1UL << PCG_CACHE)
> >  #define PCGF_USED	(1UL << PCG_USED)
> >  #define PCGF_LOCK	(1UL << PCG_LOCK)
> > +/* Not used, but added here for completeness */
> > +#define PCGF_ROOT	(1UL << PCG_ROOT)
> > +#define PCGF_ACCT	(1UL << PCG_ACCT)
> > +
> >  static const unsigned long
> >  pcg_default_flags[NR_CHARGE_TYPE] = {
> >  	PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */
> 
> Could you delete this default_flags ? This is of no use after this patch.
>

Yes, I mentioned in the comment that they are for readability of the
code. I can remove them if required.
 
> 
> > @@ -375,7 +380,7 @@ void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
> >  		return;
> >  	pc = lookup_page_cgroup(page);
> >  	/* can happen while we handle swapcache. */
> > -	if (list_empty(&pc->lru) || !pc->mem_cgroup)
> > +	if ((!PageCgroupAcctLru(pc) && list_empty(&pc->lru)) || !pc->mem_cgroup)
> >  		return;
> I wonder this condition is valid one or not..
> 
> IMHO, all check here should be
> 
> ==
> 	if (!PageCgroupAcctLru(pc) || !pc->mem_cgroup)
> 		return;
> 	mz = page_cgroup_zoneinfo(pc);
> 	mem = pc->mem_cgroup;
> 	MEM_CGROUP_ZSTAT(mz, lru) -= 1;
> 	ClearPageCgroupAcctLru(pc);
> 	if (PageCgroupRoot(pc))
> 		return;
> 	VM_BUGON(list_empty(&pc->lru);
> 	list_del_init(&pc->lru);
> 	return;

We needed this check because

1. After PageCgroupRoot(), list_empty() will always return true for
root cgroup
2. For non root, it won't

The check is enhanced to say, don't go by list_empty(), look to see if
this is root.

I think we can change the condition and stop relying on list_empty()
for the check. I agree.


> ==
> 
> I'm sorry if there is a case
>    (PageCgroupAcctLru(pc) && !PageCgroupRoot(pc) && list_empty(&pc->lru))
>

Should not be, I think the list_empty() was used to indicated already
unaccounted, so explicit flags should work fine.
 
> 
> >  	/*
> >  	 * We don't check PCG_USED bit. It's cleared when the "page" is finally
> > @@ -384,6 +389,9 @@ void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru)
> >  	mz = page_cgroup_zoneinfo(pc);
> >  	mem = pc->mem_cgroup;
> >  	MEM_CGROUP_ZSTAT(mz, lru) -= 1;
> > +	ClearPageCgroupAcctLru(pc);
> > +	if (PageCgroupRoot(pc))
> > +		return;
> >  	list_del_init(&pc->lru);
> >  	return;
> >  }
> > @@ -407,8 +415,8 @@ void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru)
> >  	 * For making pc->mem_cgroup visible, insert smp_rmb() here.
> >  	 */
> >  	smp_rmb();
> > -	/* unused page is not rotated. */
> > -	if (!PageCgroupUsed(pc))
> > +	/* unused or root page is not rotated. */
> > +	if (!PageCgroupUsed(pc) || PageCgroupRoot(pc))
> >  		return;
> >  	mz = page_cgroup_zoneinfo(pc);
> >  	list_move(&pc->lru, &mz->lists[lru]);
> > @@ -432,6 +440,9 @@ void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
> >  
> >  	mz = page_cgroup_zoneinfo(pc);
> >  	MEM_CGROUP_ZSTAT(mz, lru) += 1;
> > +	SetPageCgroupAcctLru(pc);
> > +	if (PageCgroupRoot(pc))
> > +		return;
> >  	list_add(&pc->lru, &mz->lists[lru]);
> >  }
> >  
> > @@ -1107,9 +1118,24 @@ static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
> >  		css_put(&mem->css);
> >  		return;
> >  	}
> > +
> >  	pc->mem_cgroup = mem;
> >  	smp_wmb();
> > -	pc->flags = pcg_default_flags[ctype];
> > +	switch (ctype) {
> > +	case MEM_CGROUP_CHARGE_TYPE_CACHE:
> > +	case MEM_CGROUP_CHARGE_TYPE_SHMEM:
> > +		SetPageCgroupCache(pc);
> > +		SetPageCgroupUsed(pc);
> > +		break;
> > +	case MEM_CGROUP_CHARGE_TYPE_MAPPED:
> > +		SetPageCgroupUsed(pc);
> > +		break;
> > +	default:
> > +		break;
> > +	}
> > +
> > +	if (mem == root_mem_cgroup)
> > +		SetPageCgroupRoot(pc);
> >  
> >  	mem_cgroup_charge_statistics(mem, pc, true);
> >  
> My concern here is there will be a racy moment that pc->flag shows
>   PageCgroupUsed(pc) && !PageCgroupRoot(pc) even if pc->mem_cgroup == root_mem_cgroup.
> 
> Then, The order of code here should be
> ==
> 	if (mem == root_mem_cgroup)
> 		SetPageCgroupRoot(pc);
> 	pc->mem_cgroup == mem;;
> 	smp_wmb();
> 	switch(type) {
> 	case....
> 	}
> 	// Used bit is set at last.
> ==
> 
> But I wonder it's better to use
> ==
> static inline int page_cgroup_is_under_root(pc)
> {
> 	pc->mem_cgroup == root_mem_cgroup;
> }
> ==
> I'm not sure why PageCgroupRoot() "bit" is necessary.
> Could you clarify the benefit of Root flag ?

The Root flags was used for accounting, but I think we can start
removing it now.

> 
> 
> 
> > @@ -1515,6 +1541,8 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
> >  	mem_cgroup_charge_statistics(mem, pc, false);
> >  
> >  	ClearPageCgroupUsed(pc);
> > +	if (mem == root_mem_cgroup)
> > +		ClearPageCgroupRoot(pc);
> >  	/*
> >  	 * pc->mem_cgroup is not cleared here. It will be accessed when it's
> >  	 * freed from LRU. This is safe because uncharged page is expected not
> > @@ -2036,6 +2064,10 @@ static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
> >  	name = MEMFILE_ATTR(cft->private);
> >  	switch (name) {
> >  	case RES_LIMIT:
> > +		if (memcg == root_mem_cgroup) { /* Can't set limit on root */
> > +			ret = -EINVAL;
> > +			break;
> > +		}
> >  		/* This function does all necessary parse...reuse it */
> >  		ret = res_counter_memparse_write_strategy(buffer, &val);
> >  		if (ret)
> > @@ -2502,6 +2534,7 @@ mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
> >  	if (cont->parent == NULL) {
> >  		enable_swap_cgroup();
> >  		parent = NULL;
> > +		root_mem_cgroup = mem;
> >  	} else {
> >  		parent = mem_cgroup_from_cont(cont->parent);
> >  		mem->use_hierarchy = parent->use_hierarchy;
> > @@ -2530,6 +2563,7 @@ mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
> >  	return &mem->css;
> >  free_out:
> >  	__mem_cgroup_free(mem);
> > +	root_mem_cgroup = NULL;
> >  	return ERR_PTR(error);
> >  }
> >  
> > diff --git a/mm/page_cgroup.c b/mm/page_cgroup.c
> > index ecc3918..4406a9c 100644
> > --- a/mm/page_cgroup.c
> > +++ b/mm/page_cgroup.c
> > @@ -276,7 +276,6 @@ void __meminit pgdat_page_cgroup_init(struct pglist_data *pgdat)
> >  
> >  #endif
> >  
> > -
> >  #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
> >  
> >  static DEFINE_MUTEX(swap_cgroup_mutex);
> > 
> Unnecessary diff here.
>

Yes, I'll add back the space.

Thanks for the review 

-- 
	Balbir

--
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>