2007-08-20 21:54:57

by Christoph Lameter

[permalink] [raw]
Subject: [RFC 5/7] Laundry handling for direct reclaim

Direct reclaim collects a global laundry list in try_to_free_pages().

Pages are only written back after a reclaim pass is complete.

Signed-off-by: Christoph Lameter <[email protected]>

---
mm/vmscan.c | 12 +++++++++---
1 file changed, 9 insertions(+), 3 deletions(-)

Index: linux-2.6/mm/vmscan.c
===================================================================
--- linux-2.6.orig/mm/vmscan.c 2007-08-19 23:30:15.000000000 -0700
+++ linux-2.6/mm/vmscan.c 2007-08-19 23:53:43.000000000 -0700
@@ -1099,7 +1099,7 @@ static unsigned long shrink_zone(int pri
* scan then give up on it.
*/
static unsigned long shrink_zones(int priority, struct zone **zones,
- struct scan_control *sc)
+ struct scan_control *sc, struct list_head *laundry)
{
unsigned long nr_reclaimed = 0;
int i;
@@ -1121,7 +1121,7 @@ static unsigned long shrink_zones(int pr

sc->all_unreclaimable = 0;

- nr_reclaimed += shrink_zone(priority, zone, sc, NULL);
+ nr_reclaimed += shrink_zone(priority, zone, sc, laundry);
}
return nr_reclaimed;
}
@@ -1156,6 +1156,7 @@ unsigned long try_to_free_pages(struct z
.swappiness = vm_swappiness,
.order = order,
};
+ LIST_HEAD(laundry);

count_vm_event(ALLOCSTALL);

@@ -1170,16 +1171,19 @@ unsigned long try_to_free_pages(struct z
}

for (priority = DEF_PRIORITY; priority >= 0; priority--) {
+
sc.nr_scanned = 0;
if (!priority)
disable_swap_token();
- nr_reclaimed += shrink_zones(priority, zones, &sc);
+ nr_reclaimed += shrink_zones(priority, zones, &sc, &laundry);
shrink_slab(sc.nr_scanned, gfp_mask, lru_pages);
if (reclaim_state) {
nr_reclaimed += reclaim_state->reclaimed_slab;
reclaim_state->reclaimed_slab = 0;
}
+
total_scanned += sc.nr_scanned;
+
if (nr_reclaimed >= sc.swap_cluster_max) {
ret = 1;
goto out;
@@ -1223,6 +1227,8 @@ out:

zone->prev_priority = priority;
}
+ nr_reclaimed += shrink_page_list(&laundry, &sc, NULL);
+ release_lru_pages(&laundry);
return ret;
}


--


2007-08-21 15:07:06

by mel

[permalink] [raw]
Subject: Re: [RFC 5/7] Laundry handling for direct reclaim

On (20/08/07 14:50), Christoph Lameter didst pronounce:
> Direct reclaim collects a global laundry list in try_to_free_pages().
>
> Pages are only written back after a reclaim pass is complete.
>
> Signed-off-by: Christoph Lameter <[email protected]>
>
> ---
> mm/vmscan.c | 12 +++++++++---
> 1 file changed, 9 insertions(+), 3 deletions(-)
>
> Index: linux-2.6/mm/vmscan.c
> ===================================================================
> --- linux-2.6.orig/mm/vmscan.c 2007-08-19 23:30:15.000000000 -0700
> +++ linux-2.6/mm/vmscan.c 2007-08-19 23:53:43.000000000 -0700
> @@ -1099,7 +1099,7 @@ static unsigned long shrink_zone(int pri
> * scan then give up on it.
> */
> static unsigned long shrink_zones(int priority, struct zone **zones,
> - struct scan_control *sc)
> + struct scan_control *sc, struct list_head *laundry)
> {
> unsigned long nr_reclaimed = 0;
> int i;
> @@ -1121,7 +1121,7 @@ static unsigned long shrink_zones(int pr
>
> sc->all_unreclaimable = 0;
>
> - nr_reclaimed += shrink_zone(priority, zone, sc, NULL);
> + nr_reclaimed += shrink_zone(priority, zone, sc, laundry);
> }
> return nr_reclaimed;
> }
> @@ -1156,6 +1156,7 @@ unsigned long try_to_free_pages(struct z
> .swappiness = vm_swappiness,
> .order = order,
> };
> + LIST_HEAD(laundry);

Why is the laundry not made part of the scan_control?

>
> count_vm_event(ALLOCSTALL);
>
> @@ -1170,16 +1171,19 @@ unsigned long try_to_free_pages(struct z
> }
>
> for (priority = DEF_PRIORITY; priority >= 0; priority--) {
> +
> sc.nr_scanned = 0;
> if (!priority)
> disable_swap_token();
> - nr_reclaimed += shrink_zones(priority, zones, &sc);
> + nr_reclaimed += shrink_zones(priority, zones, &sc, &laundry);
> shrink_slab(sc.nr_scanned, gfp_mask, lru_pages);
> if (reclaim_state) {
> nr_reclaimed += reclaim_state->reclaimed_slab;
> reclaim_state->reclaimed_slab = 0;
> }
> +
> total_scanned += sc.nr_scanned;
> +
> if (nr_reclaimed >= sc.swap_cluster_max) {
> ret = 1;
> goto out;
> @@ -1223,6 +1227,8 @@ out:
>
> zone->prev_priority = priority;
> }
> + nr_reclaimed += shrink_page_list(&laundry, &sc, NULL);
> + release_lru_pages(&laundry);
> return ret;
> }
>
>
> --
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to [email protected]. For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

--
--
Mel Gorman
Part-time Phd Student Linux Technology Center
University of Limerick IBM Dublin Software Lab

2007-08-21 15:19:30

by mel

[permalink] [raw]
Subject: Re: [RFC 5/7] Laundry handling for direct reclaim

On (20/08/07 14:50), Christoph Lameter didst pronounce:
> Direct reclaim collects a global laundry list in try_to_free_pages().
>
> Pages are only written back after a reclaim pass is complete.
>
> Signed-off-by: Christoph Lameter <[email protected]>
>
> ---
> mm/vmscan.c | 12 +++++++++---
> 1 file changed, 9 insertions(+), 3 deletions(-)
>
> Index: linux-2.6/mm/vmscan.c
> ===================================================================
> --- linux-2.6.orig/mm/vmscan.c 2007-08-19 23:30:15.000000000 -0700
> +++ linux-2.6/mm/vmscan.c 2007-08-19 23:53:43.000000000 -0700
> @@ -1099,7 +1099,7 @@ static unsigned long shrink_zone(int pri
> * scan then give up on it.
> */
> static unsigned long shrink_zones(int priority, struct zone **zones,
> - struct scan_control *sc)
> + struct scan_control *sc, struct list_head *laundry)
> {
> unsigned long nr_reclaimed = 0;
> int i;
> @@ -1121,7 +1121,7 @@ static unsigned long shrink_zones(int pr
>
> sc->all_unreclaimable = 0;
>
> - nr_reclaimed += shrink_zone(priority, zone, sc, NULL);
> + nr_reclaimed += shrink_zone(priority, zone, sc, laundry);
> }
> return nr_reclaimed;
> }
> @@ -1156,6 +1156,7 @@ unsigned long try_to_free_pages(struct z
> .swappiness = vm_swappiness,
> .order = order,
> };
> + LIST_HEAD(laundry);
>
> count_vm_event(ALLOCSTALL);
>
> @@ -1170,16 +1171,19 @@ unsigned long try_to_free_pages(struct z
> }
>
> for (priority = DEF_PRIORITY; priority >= 0; priority--) {
> +
> sc.nr_scanned = 0;
> if (!priority)
> disable_swap_token();
> - nr_reclaimed += shrink_zones(priority, zones, &sc);
> + nr_reclaimed += shrink_zones(priority, zones, &sc, &laundry);
> shrink_slab(sc.nr_scanned, gfp_mask, lru_pages);
> if (reclaim_state) {
> nr_reclaimed += reclaim_state->reclaimed_slab;
> reclaim_state->reclaimed_slab = 0;
> }
> +
> total_scanned += sc.nr_scanned;
> +

Could this not isolate a load of dirty pages on the laundry list and then
shortly later go to sleep in congestion_wait() ? It would appear that with
writeout deferred that the going to sleep is going to do nothing to help
the situation.

> if (nr_reclaimed >= sc.swap_cluster_max) {
> ret = 1;
> goto out;
> @@ -1223,6 +1227,8 @@ out:
>
> zone->prev_priority = priority;
> }
> + nr_reclaimed += shrink_page_list(&laundry, &sc, NULL);
> + release_lru_pages(&laundry);
> return ret;
> }
>

--
Mel Gorman
Part-time Phd Student Linux Technology Center
University of Limerick IBM Dublin Software Lab

2007-08-21 20:55:42

by Christoph Lameter

[permalink] [raw]
Subject: Re: [RFC 5/7] Laundry handling for direct reclaim

On Tue, 21 Aug 2007, Mel Gorman wrote:

> > @@ -1156,6 +1156,7 @@ unsigned long try_to_free_pages(struct z
> > .swappiness = vm_swappiness,
> > .order = order,
> > };
> > + LIST_HEAD(laundry);
>
> Why is the laundry not made part of the scan_control?

That is one possibility. The other is to treat laundry as a lru type list
under zone->lru_lock. This would allow the writeback process (whichever
that is) to be independent of the producer of the laundry. Dirty pages
could be isolated from an atomic context.

2007-08-21 21:00:40

by Christoph Lameter

[permalink] [raw]
Subject: Re: [RFC 5/7] Laundry handling for direct reclaim

On Tue, 21 Aug 2007, Mel Gorman wrote:

> > + nr_reclaimed += shrink_zones(priority, zones, &sc, &laundry);
> > shrink_slab(sc.nr_scanned, gfp_mask, lru_pages);
> > if (reclaim_state) {
> > nr_reclaimed += reclaim_state->reclaimed_slab;
> > reclaim_state->reclaimed_slab = 0;
> > }
> > +
> > total_scanned += sc.nr_scanned;
> > +
>
> Could this not isolate a load of dirty pages on the laundry list and then
> shortly later go to sleep in congestion_wait() ? It would appear that with
> writeout deferred that the going to sleep is going to do nothing to help
> the situation.

Yep that seems to be the problem that Peter saw. We need to throttle
later.