From: Konstantin Khlebnikov <khlebnikov@openvz.org>
To: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hugh Dickins <hughd@google.com>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
"linux-mm@kvack.org" <linux-mm@kvack.org>,
Johannes Weiner <hannes@cmpxchg.org>,
Andrew Morton <akpm@linux-foundation.org>,
Andi Kleen <andi@firstfloor.org>
Subject: Re: [PATCH v3 20/21] mm: split zone->lru_lock
Date: Tue, 28 Feb 2012 10:39:45 +0400 [thread overview]
Message-ID: <4F4C76B1.8000503@openvz.org> (raw)
In-Reply-To: <20120228104913.0fc13bfd.kamezawa.hiroyu@jp.fujitsu.com>
KAMEZAWA Hiroyuki wrote:
> On Thu, 23 Feb 2012 17:53:23 +0400
> Konstantin Khlebnikov<khlebnikov@openvz.org> wrote:
>
>> Looks like all ready for splitting zone->lru_lock into per-lruvec pieces.
>>
>> lruvec locking loop protected with rcu, actually there is irq-disabling instead
>> of rcu_read_lock(). Memory controller already releases its lru-vectors after
>> syncronize_rcu() in cgroup_diput(). Probably it should be replaced with synchronize_sched()
>>
>> Signed-off-by: Konstantin Khlebnikov<khlebnikov@openvz.org>
>
> Do we need rcu_read_lock() even if we check isolated pages at pre_destroy() ?
> If pre_destroy() ends, pages under a memcg being destroyed were moved to other
> cgroup while it's usolated.
>
> So,
> - PageLRU(page) guarantees lruvec is valid.
> - if !PageLRU(page), the caller of lru_lock should know what it does.
> Once isolated, pre_destroy() never ends and page_lruvec(page) is always stable.
This will be racy, because lruvec can be changed and released between checking PageLRU() and page_lruvec() or
between page_lruvec() and spin_lock(), so we need to protect whole operation with single locking interval.
>
> Thanks,
> -Kame
>
>> ---
>> include/linux/mmzone.h | 3 +-
>> mm/compaction.c | 2 +
>> mm/internal.h | 66 +++++++++++++++++++++++++-----------------------
>> mm/page_alloc.c | 2 +
>> mm/swap.c | 2 +
>> 5 files changed, 40 insertions(+), 35 deletions(-)
>>
>> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
>> index 2e3a298..9880150 100644
>> --- a/include/linux/mmzone.h
>> +++ b/include/linux/mmzone.h
>> @@ -304,6 +304,8 @@ struct zone_reclaim_stat {
>> };
>>
>> struct lruvec {
>> + spinlock_t lru_lock;
>> +
>> struct list_head pages_lru[NR_LRU_LISTS];
>> unsigned long pages_count[NR_LRU_COUNTERS];
>>
>> @@ -386,7 +388,6 @@ struct zone {
>> ZONE_PADDING(_pad1_)
>>
>> /* Fields commonly accessed by the page reclaim scanner */
>> - spinlock_t lru_lock;
>> struct lruvec lruvec;
>>
>> unsigned long pages_scanned; /* since last reclaim */
>> diff --git a/mm/compaction.c b/mm/compaction.c
>> index fa74cbe..8661bb58 100644
>> --- a/mm/compaction.c
>> +++ b/mm/compaction.c
>> @@ -306,7 +306,7 @@ static isolate_migrate_t isolate_migratepages(struct zone *zone,
>> lruvec = NULL;
>> }
>> if (need_resched() ||
>> - (lruvec&& spin_is_contended(&zone->lru_lock))) {
>> + (lruvec&& spin_is_contended(&lruvec->lru_lock))) {
>> if (lruvec)
>> unlock_lruvec_irq(lruvec);
>> lruvec = NULL;
>> diff --git a/mm/internal.h b/mm/internal.h
>> index 6dd2e70..9a9fd53 100644
>> --- a/mm/internal.h
>> +++ b/mm/internal.h
>> @@ -15,27 +15,27 @@
>>
>> static inline void lock_lruvec(struct lruvec *lruvec, unsigned long *flags)
>> {
>> - spin_lock_irqsave(&lruvec_zone(lruvec)->lru_lock, *flags);
>> + spin_lock_irqsave(&lruvec->lru_lock, *flags);
>> }
>>
>> static inline void lock_lruvec_irq(struct lruvec *lruvec)
>> {
>> - spin_lock_irq(&lruvec_zone(lruvec)->lru_lock);
>> + spin_lock_irq(&lruvec->lru_lock);
>> }
>>
>> static inline void unlock_lruvec(struct lruvec *lruvec, unsigned long *flags)
>> {
>> - spin_unlock_irqrestore(&lruvec_zone(lruvec)->lru_lock, *flags);
>> + spin_unlock_irqrestore(&lruvec->lru_lock, *flags);
>> }
>>
>> static inline void unlock_lruvec_irq(struct lruvec *lruvec)
>> {
>> - spin_unlock_irq(&lruvec_zone(lruvec)->lru_lock);
>> + spin_unlock_irq(&lruvec->lru_lock);
>> }
>>
>> static inline void wait_lruvec_unlock(struct lruvec *lruvec)
>> {
>> - spin_unlock_wait(&lruvec_zone(lruvec)->lru_lock);
>> + spin_unlock_wait(&lruvec->lru_lock);
>> }
>>
>> #ifdef CONFIG_CGROUP_MEM_RES_CTLR
>> @@ -46,37 +46,39 @@ static inline void wait_lruvec_unlock(struct lruvec *lruvec)
>> static inline struct lruvec *__relock_page_lruvec(struct lruvec *locked_lruvec,
>> struct page *page)
>> {
>> - /* Currenyly only one lru_lock per-zone */
>> - return page_lruvec(page);
>> + struct lruvec *lruvec;
>> +
>> + do {
>> + lruvec = page_lruvec(page);
>> + if (likely(lruvec == locked_lruvec))
>> + return lruvec;
>> + spin_unlock(&locked_lruvec->lru_lock);
>> + spin_lock(&lruvec->lru_lock);
>> + locked_lruvec = lruvec;
>> + } while (1);
>> }
>>
>> static inline struct lruvec *relock_page_lruvec_irq(struct lruvec *lruvec,
>> struct page *page)
>> {
>> - struct zone *zone = page_zone(page);
>> -
>> if (!lruvec) {
>> - spin_lock_irq(&zone->lru_lock);
>> - } else if (zone != lruvec_zone(lruvec)) {
>> - unlock_lruvec_irq(lruvec);
>> - spin_lock_irq(&zone->lru_lock);
>> + local_irq_disable();
>> + lruvec = page_lruvec(page);
>> + spin_lock(&lruvec->lru_lock);
>> }
>> - return page_lruvec(page);
>> + return __relock_page_lruvec(lruvec, page);
>> }
>>
>> static inline struct lruvec *relock_page_lruvec(struct lruvec *lruvec,
>> struct page *page,
>> unsigned long *flags)
>> {
>> - struct zone *zone = page_zone(page);
>> -
>> if (!lruvec) {
>> - spin_lock_irqsave(&zone->lru_lock, *flags);
>> - } else if (zone != lruvec_zone(lruvec)) {
>> - unlock_lruvec(lruvec, flags);
>> - spin_lock_irqsave(&zone->lru_lock, *flags);
>> + local_irq_save(*flags);
>> + lruvec = page_lruvec(page);
>> + spin_lock(&lruvec->lru_lock);
>> }
>> - return page_lruvec(page);
>> + return __relock_page_lruvec(lruvec, page);
>> }
>>
>> /*
>> @@ -87,22 +89,24 @@ static inline struct lruvec *relock_page_lruvec(struct lruvec *lruvec,
>> static inline bool __lock_page_lruvec_irq(struct lruvec **lruvec,
>> struct page *page)
>> {
>> - struct zone *zone;
>> bool ret = false;
>>
>> + rcu_read_lock();
>> + /*
>> + * If we see there PageLRU(), it means page has valid lruvec link.
>> + * We need protect whole operation with single rcu-interval, otherwise
>> + * lruvec which hold this LRU sign can run out before we secure it.
>> + */
>> if (PageLRU(page)) {
>> if (!*lruvec) {
>> - zone = page_zone(page);
>> - spin_lock_irq(&zone->lru_lock);
>> - } else
>> - zone = lruvec_zone(*lruvec);
>> -
>> - if (PageLRU(page)) {
>> *lruvec = page_lruvec(page);
>> + lock_lruvec_irq(*lruvec);
>> + }
>> + *lruvec = __relock_page_lruvec(*lruvec, page);
>> + if (PageLRU(page))
>> ret = true;
>> - } else
>> - *lruvec =&zone->lruvec;
>> }
>> + rcu_read_unlock();
>>
>> return ret;
>> }
>> @@ -110,7 +114,7 @@ static inline bool __lock_page_lruvec_irq(struct lruvec **lruvec,
>> /* Wait for lruvec unlock before locking other lruvec for the same page */
>> static inline void __wait_lruvec_unlock(struct lruvec *lruvec)
>> {
>> - /* Currently only one lru_lock per-zone */
>> + wait_lruvec_unlock(lruvec);
>> }
>>
>> #else /* CONFIG_CGROUP_MEM_RES_CTLR */
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index ab42446..beadcc9 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -4294,6 +4294,7 @@ void init_zone_lruvec(struct zone *zone, struct lruvec *lruvec)
>> enum lru_list lru;
>>
>> memset(lruvec, 0, sizeof(struct lruvec));
>> + spin_lock_init(&lruvec->lru_lock);
>> for_each_lru(lru)
>> INIT_LIST_HEAD(&lruvec->pages_lru[lru]);
>> #ifdef CONFIG_CGROUP_MEM_RES_CTLR
>> @@ -4369,7 +4370,6 @@ static void __paginginit free_area_init_core(struct pglist_data *pgdat,
>> #endif
>> zone->name = zone_names[j];
>> spin_lock_init(&zone->lock);
>> - spin_lock_init(&zone->lru_lock);
>> zone_seqlock_init(zone);
>> zone->zone_pgdat = pgdat;
>>
>> diff --git a/mm/swap.c b/mm/swap.c
>> index 998c71c..8156181 100644
>> --- a/mm/swap.c
>> +++ b/mm/swap.c
>> @@ -700,7 +700,7 @@ void lru_add_page_tail(struct lruvec *lruvec,
>> VM_BUG_ON(!PageHead(page));
>> VM_BUG_ON(PageCompound(page_tail));
>> VM_BUG_ON(PageLRU(page_tail));
>> - VM_BUG_ON(NR_CPUS != 1&& !spin_is_locked(&lruvec_zone(lruvec)->lru_lock));
>> + VM_BUG_ON(NR_CPUS != 1&& !spin_is_locked(&lruvec->lru_lock));
>>
>> SetPageLRU(page_tail);
>>
>>
>>
>
--
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/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
next prev parent reply other threads:[~2012-02-28 6:39 UTC|newest]
Thread overview: 65+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-02-23 13:51 [PATCH v3 00/21] mm: lru_lock splitting Konstantin Khlebnikov
2012-02-23 13:51 ` [PATCH v3 01/21] memcg: unify inactive_ratio calculation Konstantin Khlebnikov
2012-02-28 0:05 ` KAMEZAWA Hiroyuki
2012-02-23 13:51 ` [PATCH v3 02/21] memcg: make mm_match_cgroup() hirarchical Konstantin Khlebnikov
2012-02-23 18:03 ` Johannes Weiner
2012-02-23 19:46 ` Konstantin Khlebnikov
2012-02-23 22:06 ` Johannes Weiner
2012-02-28 0:11 ` KAMEZAWA Hiroyuki
2012-02-28 6:31 ` Konstantin Khlebnikov
2012-02-23 13:51 ` [PATCH v3 03/21] memcg: fix page_referencies cgroup filter on global reclaim Konstantin Khlebnikov
2012-02-28 0:13 ` KAMEZAWA Hiroyuki
2012-02-23 13:51 ` [PATCH v3 04/21] memcg: use vm_swappiness from target memory cgroup Konstantin Khlebnikov
2012-02-28 0:15 ` KAMEZAWA Hiroyuki
2012-02-23 13:52 ` [PATCH v3 05/21] mm: rename lruvec->lists into lruvec->pages_lru Konstantin Khlebnikov
2012-02-28 0:20 ` KAMEZAWA Hiroyuki
2012-02-28 6:04 ` Konstantin Khlebnikov
2012-02-23 13:52 ` [PATCH v3 06/21] mm: lruvec linking functions Konstantin Khlebnikov
2012-02-28 0:27 ` KAMEZAWA Hiroyuki
2012-02-28 6:09 ` Konstantin Khlebnikov
2012-02-23 13:52 ` [PATCH v3 07/21] mm: add lruvec->pages_count Konstantin Khlebnikov
2012-02-28 0:35 ` KAMEZAWA Hiroyuki
2012-02-28 6:16 ` Konstantin Khlebnikov
2012-02-23 13:52 ` [PATCH v3 08/21] mm: unify inactive_list_is_low() Konstantin Khlebnikov
2012-02-28 0:36 ` KAMEZAWA Hiroyuki
2012-02-23 13:52 ` [PATCH v3 09/21] mm: add lruvec->reclaim_stat Konstantin Khlebnikov
2012-02-28 0:38 ` KAMEZAWA Hiroyuki
2012-02-23 13:52 ` [PATCH v3 10/21] mm: kill struct mem_cgroup_zone Konstantin Khlebnikov
2012-02-28 0:41 ` KAMEZAWA Hiroyuki
2012-02-23 13:52 ` [PATCH v3 11/21] mm: move page-to-lruvec translation upper Konstantin Khlebnikov
2012-02-28 0:42 ` KAMEZAWA Hiroyuki
2012-02-23 13:52 ` [PATCH v3 12/21] mm: push lruvec into update_page_reclaim_stat() Konstantin Khlebnikov
2012-02-28 0:44 ` KAMEZAWA Hiroyuki
2012-02-23 13:52 ` [PATCH v3 13/21] mm: push lruvecs from pagevec_lru_move_fn() to iterator Konstantin Khlebnikov
2012-02-28 0:45 ` KAMEZAWA Hiroyuki
2012-02-23 13:52 ` [PATCH v3 14/21] mm: introduce lruvec locking primitives Konstantin Khlebnikov
2012-02-28 0:56 ` KAMEZAWA Hiroyuki
2012-02-28 6:23 ` Konstantin Khlebnikov
2012-02-23 13:52 ` [PATCH v3 15/21] mm: handle lruvec relocks on lumpy reclaim Konstantin Khlebnikov
2012-02-28 1:01 ` KAMEZAWA Hiroyuki
2012-02-28 6:25 ` Konstantin Khlebnikov
2012-02-23 13:52 ` [PATCH v3 16/21] mm: handle lruvec relocks in compaction Konstantin Khlebnikov
2012-02-28 1:13 ` KAMEZAWA Hiroyuki
2012-02-28 6:31 ` Konstantin Khlebnikov
2012-02-23 13:53 ` [PATCH v3 17/21] mm: handle lruvec relock in memory controller Konstantin Khlebnikov
2012-02-28 1:22 ` KAMEZAWA Hiroyuki
2012-02-23 13:53 ` [PATCH v3 18/21] mm: add to lruvec isolated pages counters Konstantin Khlebnikov
2012-02-24 5:32 ` Konstantin Khlebnikov
2012-02-28 1:38 ` KAMEZAWA Hiroyuki
2012-02-23 13:53 ` [PATCH v3 19/21] memcg: check lru vectors emptiness in pre-destroy Konstantin Khlebnikov
2012-02-28 1:43 ` KAMEZAWA Hiroyuki
2012-02-23 13:53 ` [PATCH v3 20/21] mm: split zone->lru_lock Konstantin Khlebnikov
2012-02-28 1:49 ` KAMEZAWA Hiroyuki
2012-02-28 6:39 ` Konstantin Khlebnikov [this message]
2012-02-23 13:53 ` [PATCH v3 21/21] mm: zone lru vectors interleaving Konstantin Khlebnikov
2012-02-23 14:44 ` Hillf Danton
2012-02-23 16:21 ` Andi Kleen
2012-02-23 18:48 ` [PATCH 1/2] mm: configure lruvec split by boot options Konstantin Khlebnikov
2012-02-23 18:48 ` [PATCH 2/2] mm: show zone lruvec state in /proc/zoneinfo Konstantin Khlebnikov
2012-02-25 0:05 ` [PATCH v3 00/21] mm: lru_lock splitting Tim Chen
2012-02-25 5:34 ` Konstantin Khlebnikov
2012-02-25 2:15 ` KAMEZAWA Hiroyuki
2012-02-25 5:31 ` Konstantin Khlebnikov
2012-02-26 23:54 ` KAMEZAWA Hiroyuki
2012-02-28 1:52 ` KAMEZAWA Hiroyuki
2012-02-28 6:49 ` Konstantin Khlebnikov
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=4F4C76B1.8000503@openvz.org \
--to=khlebnikov@openvz.org \
--cc=akpm@linux-foundation.org \
--cc=andi@firstfloor.org \
--cc=hannes@cmpxchg.org \
--cc=hughd@google.com \
--cc=kamezawa.hiroyu@jp.fujitsu.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox