I'm sorry for my very late reply.
I've been working on the stale swap cache problem for a long time as you know :)
On Sun, 17 May 2009 12:15:43 +0800, Balbir Singh <[email protected]> wrote:
> * KAMEZAWA Hiroyuki <[email protected]> [2009-05-16 02:45:03]:
>
> > I think set/clear flag here adds race condtion....because pc->flags is
> > modfied by
> > pc->flags = pcg_dafault_flags[ctype] in commit_charge()
> > you have to modify above lines to be
> >
> > SetPageCgroupCache(pc) or some..
> > ...
> > SetPageCgroupUsed(pc)
> >
> > Then, you can use set_bit() without lock_page_cgroup().
> > (Currently, pc->flags is modified only under lock_page_cgroup(), so,
> > non atomic code is used.)
> >
>
> Here is the next version of the patch
>
>
> Feature: Remove the overhead associated with the root cgroup
>
> From: Balbir Singh <[email protected]>
>
> 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.
>
I agree to this idea itself.
> 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:
> 1. Tested lightly, previous versions showed good performance improvement 10%.
>
You should test current version :)
And I think you should test this patch under global memory pressure too
to check whether it doesn't cause bug or under/over flow of something, etc.
memcg's LRU handling about SwapCache is different from usual one.
> NOTE:
> I haven't got the time right now to run oprofile and get detailed test results,
> since I am in the middle of travel.
>
> Please review the code for functional correctness and if you can test
> it even better. I would like to push this in, especially if the %
> performance difference I am seeing is reproducible elsewhere as well.
>
> Signed-off-by: Balbir Singh <[email protected]>
> ---
>
> 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..ebdae9a 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, /* page has been accounted for */
> };
>
Those new flags are protected by zone->lru_lock, right ?
If so, please add some comments.
And I'm not sure why you need 2 flags. Isn't PCG_ROOT enough for you ?
> #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(Acct, ACCT)
> +CLEARPCGFLAG(Acct, ACCT)
> +TESTPCGFLAG(Acct, ACCT)
>
> static inline int page_cgroup_nid(struct page_cgroup *pc)
> {
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index 9712ef7..35415fc 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 = 0 */
> @@ -196,6 +197,10 @@ 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 */
> @@ -420,7 +425,7 @@ 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 ((!PageCgroupAcct(pc) && list_empty(&pc->lru)) || !pc->mem_cgroup)
> return;
> /*
> * We don't check PCG_USED bit. It's cleared when the "page" is finally
> @@ -429,6 +434,9 @@ 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;
> + ClearPageCgroupAcct(pc);
> + if (PageCgroupRoot(pc))
> + return;
> list_del_init(&pc->lru);
> return;
> }
> @@ -452,8 +460,8 @@ 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]);
> @@ -477,6 +485,9 @@ void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru)
>
> mz = page_cgroup_zoneinfo(pc);
> MEM_CGROUP_ZSTAT(mz, lru) += 1;
> + SetPageCgroupAcct(pc);
> + if (PageCgroupRoot(pc))
> + return;
> list_add(&pc->lru, &mz->lists[lru]);
> }
>
> @@ -1114,9 +1125,24 @@ 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);
>
Shouldn't we set PCG_LOCK ?
unlock_page_cgroup() will be called after this.
Moreover, IIUC, pc->flags is not cleared at page free/alloc, so if a page
is reused, pc->flags has the old value.
PCG_CACHE flag, at least, is used by the decision in mem_cgroup_charge_statistics().
> @@ -1521,6 +1547,8 @@ __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
> @@ -2038,6 +2066,10 @@ 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)
It's a nitpick, I prefer not to show *.limit_in_bytes if we cannot write to them.
Thanks,
Daisuke Nishimura.
> @@ -2504,6 +2536,7 @@ 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;
> @@ -2532,6 +2565,7 @@ 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 09b73c5..6145ff6 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);
>
>
> --
> Balbir
> > @@ -1114,9 +1125,24 @@ 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);
> >
> Shouldn't we set PCG_LOCK ?
> unlock_page_cgroup() will be called after this.
>
Ah, lock_page_cgroup() has already set it.
please ignore this comment.
Sorry for noise.
Daisuke Nishimura.
* [email protected] <[email protected]> [2009-06-01 13:25:05]:
> I'm sorry for my very late reply.
>
> I've been working on the stale swap cache problem for a long time as you know :)
>
> On Sun, 17 May 2009 12:15:43 +0800, Balbir Singh <[email protected]> wrote:
> > * KAMEZAWA Hiroyuki <[email protected]> [2009-05-16 02:45:03]:
> >
> > > I think set/clear flag here adds race condtion....because pc->flags is
> > > modfied by
> > > pc->flags = pcg_dafault_flags[ctype] in commit_charge()
> > > you have to modify above lines to be
> > >
> > > SetPageCgroupCache(pc) or some..
> > > ...
> > > SetPageCgroupUsed(pc)
> > >
> > > Then, you can use set_bit() without lock_page_cgroup().
> > > (Currently, pc->flags is modified only under lock_page_cgroup(), so,
> > > non atomic code is used.)
> > >
> >
> > Here is the next version of the patch
> >
> >
> > Feature: Remove the overhead associated with the root cgroup
> >
> > From: Balbir Singh <[email protected]>
> >
> > 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.
> >
> I agree to this idea itself.
>
Thanks!
> > 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:
> > 1. Tested lightly, previous versions showed good performance improvement 10%.
> >
> You should test current version :)
> And I think you should test this patch under global memory pressure too
> to check whether it doesn't cause bug or under/over flow of something, etc.
> memcg's LRU handling about SwapCache is different from usual one.
>
OK, I've tested it using my stress tool, but I'll modify to add some
of the things you've pointed out.
> > NOTE:
> > I haven't got the time right now to run oprofile and get detailed test results,
> > since I am in the middle of travel.
> >
> > Please review the code for functional correctness and if you can test
> > it even better. I would like to push this in, especially if the %
> > performance difference I am seeing is reproducible elsewhere as well.
> >
> > Signed-off-by: Balbir Singh <[email protected]>
> > ---
> >
> > 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..ebdae9a 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, /* page has been accounted for */
> > };
> >
> Those new flags are protected by zone->lru_lock, right ?
> If so, please add some comments.
> And I'm not sure why you need 2 flags. Isn't PCG_ROOT enough for you ?
>
Nope.. the accounting is independent of charge/uncharge.
--
Balbir