Re: [PATCH v3 14/21] mm: introduce lruvec locking primitives

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

On Thu, 23 Feb 2012 17:52:47 +0400
Konstantin Khlebnikov <khlebnikov@openvz.org> wrote:

> This is initial preparation for lru_lock splitting.
> 
> This locking primites designed to hide splitted nature of lru_lock
> and to avoid overhead for non-splitted lru_lock in non-memcg case.
> 
> * Lock via lruvec reference
> 
> lock_lruvec(lruvec, flags)
> lock_lruvec_irq(lruvec)
> 
> * Lock via page reference
> 
> lock_page_lruvec(page, flags)
> lock_page_lruvec_irq(page)
> relock_page_lruvec(lruvec, page, flags)
> relock_page_lruvec_irq(lruvec, page)
> __relock_page_lruvec(lruvec, page) ( lruvec != NULL, page in same zone )
> 
> They always returns pointer to some locked lruvec, page anyway can be
> not in lru, PageLRU() sign is stable while we hold returned lruvec lock.
> Caller must guarantee page to lruvec reference validity.
> 
> * Lock via page, without stable page reference
> 
> __lock_page_lruvec_irq(&lruvec, page)
> 
> It returns true of lruvec succesfully locked and PageLRU is set.
> Initial lruvec can be NULL. Consequent calls must be in the same zone.
> 
> * Unlock
> 
> unlock_lruvec(lruvec, flags)
> unlock_lruvec_irq(lruvec)
> 
> * Wait
> 
> wait_lruvec_unlock(lruvec)
> Wait for lruvec unlock, caller must have stable reference to lruvec.
> 
> __wait_lruvec_unlock(lruvec)
> Wait for lruvec unlock before locking other lrulock for same page,
> nothing if there only one possible lruvec per page.
> Used at page-to-lruvec reference switching to stabilize PageLRU sign.
> 
> Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org>

O.K. I like this.

Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

Hmm....Could you add a comment in memcg part ? (see below)



> ---
>  mm/huge_memory.c |    8 +-
>  mm/internal.h    |  176 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
>  mm/memcontrol.c  |   14 ++--
>  mm/swap.c        |   58 ++++++------------
>  mm/vmscan.c      |   77 ++++++++++--------------
>  5 files changed, 237 insertions(+), 96 deletions(-)
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 09e7069..74996b8 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -1228,13 +1228,11 @@ static int __split_huge_page_splitting(struct page *page,
>  static void __split_huge_page_refcount(struct page *page)
>  {
>  	int i;
> -	struct zone *zone = page_zone(page);
>  	struct lruvec *lruvec;
>  	int tail_count = 0;
>  
>  	/* prevent PageLRU to go away from under us, and freeze lru stats */
> -	spin_lock_irq(&zone->lru_lock);
> -	lruvec = page_lruvec(page);
> +	lruvec = lock_page_lruvec_irq(page);
>  	compound_lock(page);
>  	/* complete memcg works before add pages to LRU */
>  	mem_cgroup_split_huge_fixup(page);
> @@ -1316,11 +1314,11 @@ static void __split_huge_page_refcount(struct page *page)
>  	BUG_ON(atomic_read(&page->_count) <= 0);
>  
>  	__dec_zone_page_state(page, NR_ANON_TRANSPARENT_HUGEPAGES);
> -	__mod_zone_page_state(zone, NR_ANON_PAGES, HPAGE_PMD_NR);
> +	__mod_zone_page_state(lruvec_zone(lruvec), NR_ANON_PAGES, HPAGE_PMD_NR);
>  
>  	ClearPageCompound(page);
>  	compound_unlock(page);
> -	spin_unlock_irq(&zone->lru_lock);
> +	unlock_lruvec_irq(lruvec);
>  
>  	for (i = 1; i < HPAGE_PMD_NR; i++) {
>  		struct page *page_tail = page + i;
> diff --git a/mm/internal.h b/mm/internal.h
> index ef49dbf..9454752 100644
> --- a/mm/internal.h
> +++ b/mm/internal.h
> @@ -13,6 +13,182 @@
>  
>  #include <linux/mm.h>
>  
> +static inline void lock_lruvec(struct lruvec *lruvec, unsigned long *flags)
> +{
> +	spin_lock_irqsave(&lruvec_zone(lruvec)->lru_lock, *flags);
> +}
> +
> +static inline void lock_lruvec_irq(struct lruvec *lruvec)
> +{
> +	spin_lock_irq(&lruvec_zone(lruvec)->lru_lock);
> +}
> +
> +static inline void unlock_lruvec(struct lruvec *lruvec, unsigned long *flags)
> +{
> +	spin_unlock_irqrestore(&lruvec_zone(lruvec)->lru_lock, *flags);
> +}
> +
> +static inline void unlock_lruvec_irq(struct lruvec *lruvec)
> +{
> +	spin_unlock_irq(&lruvec_zone(lruvec)->lru_lock);
> +}
> +
> +static inline void wait_lruvec_unlock(struct lruvec *lruvec)
> +{
> +	spin_unlock_wait(&lruvec_zone(lruvec)->lru_lock);
> +}
> +
> +#ifdef CONFIG_CGROUP_MEM_RES_CTLR
> +
> +/* Dynamic page to lruvec mapping */
> +
> +/* Lock other lruvec for other page in the same zone */
> +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);
> +}
> +
> +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);
> +	}
> +	return page_lruvec(page);
> +}

Could you add comments/caution to the caller 

 - !PageLRU(page) case ?
 - Can the caller assume page_lruvec(page) == lruvec ? If no, which lru_vec is locked ?

etc...


> +
> +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);
> +	}
> +	return page_lruvec(page);
> +}


Same here.

Thanks,
-Kame

--
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>