Provide a function to generically release pages that were isolated back
to the LRU. The function supports mixing zones etc.
Signed-off-by: Christoph Lameter <[email protected]>
---
mm/vmscan.c | 72 ++++++++++++++++++++++++++++++++++--------------------------
1 file changed, 41 insertions(+), 31 deletions(-)
Index: linux-2.6/mm/vmscan.c
===================================================================
--- linux-2.6.orig/mm/vmscan.c 2007-08-19 23:12:43.000000000 -0700
+++ linux-2.6/mm/vmscan.c 2007-08-19 23:13:24.000000000 -0700
@@ -581,6 +581,42 @@ keep:
return nr_reclaimed;
}
+/*
+ * Put back any unfreeable pages.
+ */
+void release_lru_pages(struct list_head *page_list)
+{
+ struct page *page;
+ struct pagevec pvec;
+ struct zone *zone = NULL;
+
+ pagevec_init(&pvec, 1);
+ while (!list_empty(page_list)) {
+ page = lru_to_page(page_list);
+ VM_BUG_ON(PageLRU(page));
+ if (zone != page_zone(page)) {
+ if (zone)
+ spin_unlock_irq(&zone->lru_lock);
+ zone = page_zone(page);
+ spin_lock_irq(&zone->lru_lock);
+ }
+ SetPageLRU(page);
+ list_del(&page->lru);
+ if (PageActive(page))
+ add_page_to_active_list(zone, page);
+ else
+ add_page_to_inactive_list(zone, page);
+ if (!pagevec_add(&pvec, page)) {
+ spin_unlock_irq(&zone->lru_lock);
+ __pagevec_release(&pvec);
+ spin_lock_irq(&zone->lru_lock);
+ }
+ }
+ if (zone)
+ spin_unlock_irq(&zone->lru_lock);
+ pagevec_release(&pvec);
+}
+
/* LRU Isolation modes. */
#define ISOLATE_INACTIVE 0 /* Isolate inactive pages. */
#define ISOLATE_ACTIVE 1 /* Isolate active pages. */
@@ -756,21 +792,17 @@ static unsigned long shrink_inactive_lis
struct zone *zone, struct scan_control *sc)
{
LIST_HEAD(page_list);
- struct pagevec pvec;
unsigned long nr_scanned = 0;
unsigned long nr_reclaimed = 0;
- pagevec_init(&pvec, 1);
-
lru_add_drain();
- spin_lock_irq(&zone->lru_lock);
do {
- struct page *page;
unsigned long nr_taken;
unsigned long nr_scan;
unsigned long nr_freed;
unsigned long nr_active;
+ spin_lock_irq(&zone->lru_lock);
nr_taken = isolate_lru_pages(sc->swap_cluster_max,
&zone->inactive_list,
&page_list, &nr_scan, sc->order,
@@ -794,34 +826,12 @@ static unsigned long shrink_inactive_lis
} else
__count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scan);
__count_zone_vm_events(PGSTEAL, zone, nr_freed);
+ local_irq_enable();
+ release_lru_pages(&page_list);
- if (nr_taken == 0)
- goto done;
-
- spin_lock(&zone->lru_lock);
- /*
- * Put back any unfreeable pages.
- */
- while (!list_empty(&page_list)) {
- page = lru_to_page(&page_list);
- VM_BUG_ON(PageLRU(page));
- SetPageLRU(page);
- list_del(&page->lru);
- if (PageActive(page))
- add_page_to_active_list(zone, page);
- else
- add_page_to_inactive_list(zone, page);
- if (!pagevec_add(&pvec, page)) {
- spin_unlock_irq(&zone->lru_lock);
- __pagevec_release(&pvec);
- spin_lock_irq(&zone->lru_lock);
- }
- }
+ if (!nr_taken)
+ break;
} while (nr_scanned < max_scan);
- spin_unlock(&zone->lru_lock);
-done:
- local_irq_enable();
- pagevec_release(&pvec);
return nr_reclaimed;
}
--
On (20/08/07 14:50), Christoph Lameter didst pronounce:
> Provide a function to generically release pages that were isolated back
> to the LRU. The function supports mixing zones etc.
>
> Signed-off-by: Christoph Lameter <[email protected]>
>
> ---
> mm/vmscan.c | 72 ++++++++++++++++++++++++++++++++++--------------------------
> 1 file changed, 41 insertions(+), 31 deletions(-)
>
> Index: linux-2.6/mm/vmscan.c
> ===================================================================
> --- linux-2.6.orig/mm/vmscan.c 2007-08-19 23:12:43.000000000 -0700
> +++ linux-2.6/mm/vmscan.c 2007-08-19 23:13:24.000000000 -0700
> @@ -581,6 +581,42 @@ keep:
> return nr_reclaimed;
> }
>
> +/*
> + * Put back any unfreeable pages.
> + */
> +void release_lru_pages(struct list_head *page_list)
> +{
Can the migrate.c#putback_lru_pages() be replaced with this?
> + struct page *page;
> + struct pagevec pvec;
> + struct zone *zone = NULL;
> +
> + pagevec_init(&pvec, 1);
> + while (!list_empty(page_list)) {
> + page = lru_to_page(page_list);
> + VM_BUG_ON(PageLRU(page));
> + if (zone != page_zone(page)) {
> + if (zone)
> + spin_unlock_irq(&zone->lru_lock);
> + zone = page_zone(page);
> + spin_lock_irq(&zone->lru_lock);
Is this really necessary? Why situation would occur that would have a
list of pages in multiple zones?
Also, it may be worth commenting here that __pagevec_release() is able to
handle lists of pages in multiple zones.
> + }
> + SetPageLRU(page);
> + list_del(&page->lru);
> + if (PageActive(page))
> + add_page_to_active_list(zone, page);
> + else
> + add_page_to_inactive_list(zone, page);
> + if (!pagevec_add(&pvec, page)) {
> + spin_unlock_irq(&zone->lru_lock);
> + __pagevec_release(&pvec);
> + spin_lock_irq(&zone->lru_lock);
> + }
> + }
> + if (zone)
> + spin_unlock_irq(&zone->lru_lock);
> + pagevec_release(&pvec);
> +}
> +
> /* LRU Isolation modes. */
> #define ISOLATE_INACTIVE 0 /* Isolate inactive pages. */
> #define ISOLATE_ACTIVE 1 /* Isolate active pages. */
> @@ -756,21 +792,17 @@ static unsigned long shrink_inactive_lis
> struct zone *zone, struct scan_control *sc)
> {
> LIST_HEAD(page_list);
> - struct pagevec pvec;
> unsigned long nr_scanned = 0;
> unsigned long nr_reclaimed = 0;
>
> - pagevec_init(&pvec, 1);
> -
> lru_add_drain();
> - spin_lock_irq(&zone->lru_lock);
> do {
> - struct page *page;
> unsigned long nr_taken;
> unsigned long nr_scan;
> unsigned long nr_freed;
> unsigned long nr_active;
>
> + spin_lock_irq(&zone->lru_lock);
> nr_taken = isolate_lru_pages(sc->swap_cluster_max,
> &zone->inactive_list,
> &page_list, &nr_scan, sc->order,
> @@ -794,34 +826,12 @@ static unsigned long shrink_inactive_lis
> } else
> __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scan);
> __count_zone_vm_events(PGSTEAL, zone, nr_freed);
> + local_irq_enable();
> + release_lru_pages(&page_list);
>
Separate these apart by a line. I thought the local_irq_enable() was related
to the call to release_lru_pages(&page_list) while reading the patch
which isn't the case at all.
> - if (nr_taken == 0)
> - goto done;
> -
> - spin_lock(&zone->lru_lock);
> - /*
> - * Put back any unfreeable pages.
> - */
> - while (!list_empty(&page_list)) {
> - page = lru_to_page(&page_list);
> - VM_BUG_ON(PageLRU(page));
> - SetPageLRU(page);
> - list_del(&page->lru);
> - if (PageActive(page))
> - add_page_to_active_list(zone, page);
> - else
> - add_page_to_inactive_list(zone, page);
> - if (!pagevec_add(&pvec, page)) {
> - spin_unlock_irq(&zone->lru_lock);
> - __pagevec_release(&pvec);
> - spin_lock_irq(&zone->lru_lock);
> - }
> - }
> + if (!nr_taken)
> + break;
> } while (nr_scanned < max_scan);
> - spin_unlock(&zone->lru_lock);
> -done:
> - local_irq_enable();
> - pagevec_release(&pvec);
> return nr_reclaimed;
> }
>
--
Mel Gorman
Part-time Phd Student Linux Technology Center
University of Limerick IBM Dublin Software Lab
On Tue, 21 Aug 2007, Mel Gorman wrote:
> > + */
> > +void release_lru_pages(struct list_head *page_list)
> > +{
>
> Can the migrate.c#putback_lru_pages() be replaced with this?
Correct. We can get rid of the putback_lru_pages in migrate.c
with this.
> > + struct page *page;
> > + struct pagevec pvec;
> > + struct zone *zone = NULL;
> > +
> > + pagevec_init(&pvec, 1);
> > + while (!list_empty(page_list)) {
> > + page = lru_to_page(page_list);
> > + VM_BUG_ON(PageLRU(page));
> > + if (zone != page_zone(page)) {
> > + if (zone)
> > + spin_unlock_irq(&zone->lru_lock);
> > + zone = page_zone(page);
> > + spin_lock_irq(&zone->lru_lock);
>
> Is this really necessary? Why situation would occur that would have a
> list of pages in multiple zones?
Because we reclaim from multiple zones and gather laundry from different
zones.
> Also, it may be worth commenting here that __pagevec_release() is able to
> handle lists of pages in multiple zones.
Ok.
> > __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scan);
> > __count_zone_vm_events(PGSTEAL, zone, nr_freed);
> > + local_irq_enable();
> > + release_lru_pages(&page_list);
> >
>
> Separate these apart by a line. I thought the local_irq_enable() was related
> to the call to release_lru_pages(&page_list) while reading the patch
> which isn't the case at all.
Will do.