On 7/1/19 1:18 PM, Shakeel Butt wrote:
On production we have noticed hard lockups on large machines running
large jobs due to kswaps hoarding lru lock within isolate_lru_pages when
sc->reclaim_idx is 0 which is a small zone. The lru was couple hundred
GiBs and the condition (page_zonenum(page) > sc->reclaim_idx) in
isolate_lru_pages was basically skipping GiBs of pages while holding the
LRU spinlock with interrupt disabled.
On further inspection, it seems like there are two issues:
1) If the kswapd on the return from balance_pgdat() could not sleep
(i.e. node is still unbalanced), the classzone_idx is unintentionally
set to 0 and the whole reclaim cycle of kswapd will try to reclaim
only the lowest and smallest zone while traversing the whole memory.
2) Fundamentally isolate_lru_pages() is really bad when the allocation
has woken kswapd for a smaller zone on a very large machine running very
large jobs. It can hoard the LRU spinlock while skipping over 100s of
GiBs of pages.
This patch only fixes the (1). The (2) needs a more fundamental solution.
To fix (1), in the kswapd context, if pgdat->kswapd_classzone_idx is
invalid use the classzone_idx of the previous kswapd loop otherwise use
the one the waker has requested.
Fixes: e716f2eb24de ("mm, vmscan: prevent kswapd sleeping prematurely
due to mismatched classzone_idx")
Signed-off-by: Shakeel Butt <shakeelb@xxxxxxxxxx>
---
Changelog since v1:
- fixed the patch based on Yang Shi's comment.
mm/vmscan.c | 27 +++++++++++++++------------
1 file changed, 15 insertions(+), 12 deletions(-)
diff --git a/mm/vmscan.c b/mm/vmscan.c
index 9e3292ee5c7c..eacf87f07afe 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -3760,19 +3760,18 @@ static int balance_pgdat(pg_data_t *pgdat, int order, int classzone_idx)
}
/*
- * pgdat->kswapd_classzone_idx is the highest zone index that a recent
- * allocation request woke kswapd for. When kswapd has not woken recently,
- * the value is MAX_NR_ZONES which is not a valid index. This compares a
- * given classzone and returns it or the highest classzone index kswapd
- * was recently woke for.
+ * The pgdat->kswapd_classzone_idx is used to pass the highest zone index to be
+ * reclaimed by kswapd from the waker. If the value is MAX_NR_ZONES which is not
+ * a valid index then either kswapd runs for first time or kswapd couldn't sleep
+ * after previous reclaim attempt (node is still unbalanced). In that case
+ * return the zone index of the previous kswapd reclaim cycle.
*/
static enum zone_type kswapd_classzone_idx(pg_data_t *pgdat,
- enum zone_type classzone_idx)
+ enum zone_type prev_classzone_idx)
{
if (pgdat->kswapd_classzone_idx == MAX_NR_ZONES)
- return classzone_idx;
-
- return max(pgdat->kswapd_classzone_idx, classzone_idx);
+ return prev_classzone_idx;
+ return pgdat->kswapd_classzone_idx;
}
static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_order,
@@ -3908,7 +3907,7 @@ static int kswapd(void *p)
/* Read the new order and classzone_idx */
alloc_order = reclaim_order = pgdat->kswapd_order;
- classzone_idx = kswapd_classzone_idx(pgdat, 0);
+ classzone_idx = kswapd_classzone_idx(pgdat, classzone_idx);
pgdat->kswapd_order = 0;
pgdat->kswapd_classzone_idx = MAX_NR_ZONES;
@@ -3961,8 +3960,12 @@ void wakeup_kswapd(struct zone *zone, gfp_t gfp_flags, int order,
if (!cpuset_zone_allowed(zone, gfp_flags))
return;
pgdat = zone->zone_pgdat;
- pgdat->kswapd_classzone_idx = kswapd_classzone_idx(pgdat,
- classzone_idx);
+
+ if (pgdat->kswapd_classzone_idx == MAX_NR_ZONES)
+ pgdat->kswapd_classzone_idx = classzone_idx;
+ else
+ pgdat->kswapd_classzone_idx = max(pgdat->kswapd_classzone_idx,
+ classzone_idx);
pgdat->kswapd_order = max(pgdat->kswapd_order, order);
if (!waitqueue_active(&pgdat->kswapd_wait))
return;
I agree the manipulation to classzone_idx looks convoluted. This version
looks correct to me. You could add: Reviewed-by: Yang Shi
<yang.shi@xxxxxxxxxxxxxxxxx>
