On Tue, Feb 27, 2018 at 09:56:13AM +0800, Aaron Lu wrote:
> On Mon, Feb 26, 2018 at 01:48:14PM -0800, David Rientjes wrote:
> > On Mon, 26 Feb 2018, Aaron Lu wrote:
> >
> > > Matthew Wilcox found that all callers of free_pcppages_bulk() currently
> > > update pcp->count immediately after so it's natural to do it inside
> > > free_pcppages_bulk().
> > >
> > > No functionality or performance change is expected from this patch.
> > >
> > > Suggested-by: Matthew Wilcox <wi...@infradead.org>
> > > Signed-off-by: Aaron Lu <aaron...@intel.com>
> > > ---
> > > mm/page_alloc.c | 10 +++-------
> > > 1 file changed, 3 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > > index cb416723538f..3154859cccd6 100644
> > > --- a/mm/page_alloc.c
> > > +++ b/mm/page_alloc.c
> > > @@ -1117,6 +1117,7 @@ static void free_pcppages_bulk(struct zone *zone,
> > > int count,
> > > int batch_free = 0;
> > > bool isolated_pageblocks;
> > >
> > > + pcp->count -= count;
> > > spin_lock(&zone->lock);
> > > isolated_pageblocks = has_isolate_pageblock(zone);
> > >
> >
> > Why modify pcp->count before the pages have actually been freed?
>
> When count is still count and not zero after pages have actually been
> freed :-)
>
> >
> > I doubt that it matters too much, but at least /proc/zoneinfo uses
> > zone->lock. I think it should be done after the lock is dropped.
>
> Agree that it looks a bit weird to do it beforehand and I just want to
> avoid adding one more local variable here.
>
> pcp->count is not protected by zone->lock though so even we do it after
> dropping the lock, it could still happen that zoneinfo shows a wrong
> value of pcp->count while it should be zero(this isn't a problem since
> zoneinfo doesn't need to be precise).
>
> Anyway, I'll follow your suggestion here to avoid confusion.
What about this: update pcp->count when page is dropped off pcp list.
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index cb416723538f..faa33eac1635 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -1148,6 +1148,7 @@ static void free_pcppages_bulk(struct zone *zone, int
count,
page = list_last_entry(list, struct page, lru);
/* must delete as __free_one_page list manipulates */
list_del(&page->lru);
+ pcp->count--;
mt = get_pcppage_migratetype(page);
/* MIGRATE_ISOLATE page should not go to pcplists */
@@ -2416,10 +2417,8 @@ void drain_zone_pages(struct zone *zone, struct
per_cpu_pages *pcp)
local_irq_save(flags);
batch = READ_ONCE(pcp->batch);
to_drain = min(pcp->count, batch);
- if (to_drain > 0) {
+ if (to_drain > 0)
free_pcppages_bulk(zone, to_drain, pcp);
- pcp->count -= to_drain;
- }
local_irq_restore(flags);
}
#endif
@@ -2441,10 +2440,8 @@ static void drain_pages_zone(unsigned int cpu, struct
zone *zone)
pset = per_cpu_ptr(zone->pageset, cpu);
pcp = &pset->pcp;
- if (pcp->count) {
+ if (pcp->count)
free_pcppages_bulk(zone, pcp->count, pcp);
- pcp->count = 0;
- }
local_irq_restore(flags);
}
@@ -2668,7 +2665,6 @@ static void free_unref_page_commit(struct page *page,
unsigned long pfn)
if (pcp->count >= pcp->high) {
unsigned long batch = READ_ONCE(pcp->batch);
free_pcppages_bulk(zone, batch, pcp);
- pcp->count -= batch;
}
}
