On Tue, 13 May 2014, Jianyu Zhan wrote:
> >> This means they guarantee that even they are preemted the vm
> >> counter won't be modified incorrectly. Because the counter is page-related
> >> (e.g., a new anon page added), and they are exclusively hold the pte lock.
> >
> >But there are multiple pte locks for numerous page. Another process could
> >modify the counter because the pte lock for a different page was
> >available which would cause counter corruption.
> >
> >
> >> So, as you concludes in the other mail that __modd_zone_page_stat
> >> couldn't be used.
> >> in mlocked_vma_newpage, then what qualifies other call sites for using
> >> it, in the same situation?
>
> Thanks, now everything is clear.
>
> I've renewed the patch, would you please review it? Thanks!
>
> ---<8---
> mm: use the light version __mod_zone_page_state in mlocked_vma_newpage()
>
> mlocked_vma_newpage() is called with pte lock held(a spinlock), which
> implies preemtion disabled, and the vm stat counter is not modified from
> interrupt context, so we need not use an irq-safe mod_zone_page_state() here,
> using a light-weight version __mod_zone_page_state() would be OK.
>
> This patch also documents __mod_zone_page_state() and some of its
> callsites. The comment above __mod_zone_page_state() is from Hugh
> Dickins, and acked by Christoph.
>
> Most credits to Hugh and Christoph for the clarification on the usage of
> the __mod_zone_page_state().
>
> Suggested-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Signed-off-by: Hugh Dickins <hughd@xxxxxxxxxx>
No, I didn't sign it off, just offered some text (which you already
acknowledged graciously). Andrew, please delete that line, but...
> Signed-off-by: Jianyu Zhan <nasa4836@xxxxxxxxx>
Acked-by: Hugh Dickins <hughd@xxxxxxxxxx>
Yes, this is good - and even better is the version which Andrew
has slipped into mmotm, removing the duplicated comment line from
page_remove_rmap(), and reformatting the comments to fit better.
It is a little random, in that we can easily see other
__mod_zone_page_states to which the same comment could be applied;
but that's okay, please leave it as is, you've made the point enough
to help other people get it, and it would get boring to repeat it more.
Hugh
> ---
> mm/internal.h | 7 ++++++-
> mm/rmap.c | 10 ++++++++++
> mm/vmstat.c | 4 +++-
> 3 files changed, 19 insertions(+), 2 deletions(-)
>
> diff --git a/mm/internal.h b/mm/internal.h
> index 07b6736..53d439e 100644
> --- a/mm/internal.h
> +++ b/mm/internal.h
> @@ -196,7 +196,12 @@ static inline int mlocked_vma_newpage(struct vm_area_struct *vma,
> return 0;
>
> if (!TestSetPageMlocked(page)) {
> - mod_zone_page_state(page_zone(page), NR_MLOCK,
> + /*
> + * We use the irq-unsafe __mod_zone_page_stat because
> + * this counter is not modified from interrupt context, and the
> + * pte lock is held(spinlock), which implies preemtion disabled.
> + */
> + __mod_zone_page_state(page_zone(page), NR_MLOCK,
> hpage_nr_pages(page));
> count_vm_event(UNEVICTABLE_PGMLOCKED);
> }
> diff --git a/mm/rmap.c b/mm/rmap.c
> index 9c3e773..2fa4375 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -986,6 +986,11 @@ void do_page_add_anon_rmap(struct page *page,
> {
> int first = atomic_inc_and_test(&page->_mapcount);
> if (first) {
> + /*
> + * We use the irq-unsafe __{inc|mod}_zone_page_stat because
> + * these counters are not modified in interrupt context, and
> + * pte lock(a spinlock) is held, which implies preemtion disabled.
> + */
> if (PageTransHuge(page))
> __inc_zone_page_state(page,
> NR_ANON_TRANSPARENT_HUGEPAGES);
> @@ -1077,6 +1082,11 @@ void page_remove_rmap(struct page *page)
> /*
> * Hugepages are not counted in NR_ANON_PAGES nor NR_FILE_MAPPED
> * and not charged by memcg for now.
> + *
> + * We use the irq-unsafe __{inc|mod}_zone_page_stat because
> + * these counters are not modified in interrupt context, and
> + * these counters are not modified in interrupt context, and
> + * pte lock(a spinlock) is held, which implies preemtion disabled.
> */
> if (unlikely(PageHuge(page)))
> goto out;
> diff --git a/mm/vmstat.c b/mm/vmstat.c
> index 302dd07..704928e 100644
> --- a/mm/vmstat.c
> +++ b/mm/vmstat.c
> @@ -207,7 +207,9 @@ void set_pgdat_percpu_threshold(pg_data_t *pgdat,
> }
>
> /*
> - * For use when we know that interrupts are disabled.
> + * For use when we know that interrupts are disabled,
> + * or when we know that preemption is disabled and that
> + * particular counter cannot be updated from interrupt context.
> */
> void __mod_zone_page_state(struct zone *zone, enum zone_stat_item item,
> int delta)
> --
> 2.0.0-rc1
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>