Re: [PATCH v3 2/3] mm/free_pcppages_bulk: do not hold lock when picking pages to free

From: Aaron Lu
Date: Mon Feb 26 2018 - 22:16:14 EST


On Tue, Feb 27, 2018 at 10:00:58AM +0800, Aaron Lu wrote:
> On Mon, Feb 26, 2018 at 01:53:10PM -0800, David Rientjes wrote:
> > On Mon, 26 Feb 2018, Aaron Lu wrote:
> >
> > > @@ -1144,26 +1142,31 @@ static void free_pcppages_bulk(struct zone *zone, int count,
> > > batch_free = count;
> > >
> > > do {
> > > - int mt; /* migratetype of the to-be-freed page */
> > > -
> > > page = list_last_entry(list, struct page, lru);
> > > /* must delete as __free_one_page list manipulates */
> >
> > Looks good in general, but I'm not sure how I reconcile this comment with
> > the new implementation that later links page->lru again.
>
> Thanks for pointing this out.
>
> I think the comment is useless now since there is a list_add_tail right
> below so it's obvious we need to take the page off its original list.
> I'll remove the comment in an update.

Thinking again, I think I'll change the comment to:

- /* must delete as __free_one_page list manipulates */
+ /* must delete to avoid corrupting pcp list */
list_del(&page->lru);
pcp->count--;

Meanwhile, I'll add one more comment about why list_for_each_entry_safe
is used:

+ /*
+ * Use safe version since after __free_one_page(),
+ * page->lru.next will not point to original list.
+ */
+ list_for_each_entry_safe(page, tmp, &head, lru) {
+ int mt = get_pcppage_migratetype(page);
+ /* MIGRATE_ISOLATE page should not go to pcplists */
+ VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
+ /* Pageblock could have been isolated meanwhile */
+ if (unlikely(isolated_pageblocks))
+ mt = get_pageblock_migratetype(page);
+
+ __free_one_page(page, page_to_pfn(page), zone, 0, mt);
+ trace_mm_page_pcpu_drain(page, 0, mt);
+ }
spin_unlock(&zone->lock);
}