When spaces of swap devices are exhausted, only file pages can be
reclaimed. But there are still some swapcache pages in anon lru list.
This can lead to a premature out-of-memory.
The problem is found with such step:
Firstly, set a 9MB disk swap space, then create a cgroup with 10MB
memory limit, then runs an program to allocates about 15MB memory.
The problem occurs occasionally, which may need about 100 times [1].
Fix it by checking number of swapcache pages in can_reclaim_anon_pages().
If the number is not zero, return true and set swapcache_only to 1.
When scan anon lru list in swapcache_only mode, non-swapcache pages will
be skipped to isolate in order to accelerate reclaim efficiency.
However, in swapcache_only mode, the scan count still increased when scan
non-swapcache pages because there are large number of non-swapcache pages
and rare swapcache pages in swapcache_only mode, and if the non-swapcache
is skipped and do not count, the scan of pages in isolate_lru_folios() can
eventually lead to hung task, just as Sachin reported [2].
By the way, since there are enough times of memory reclaim before OOM, it
is not need to isolate too much swapcache pages in one times.
[1]. https://lore.kernel.org/lkml/CAJD7tkZAfgncV+KbKr36=eDzMnT=9dZOT0dpMWcurHLr6Do+GA@mail.gmail.com/
[2]. https://lore.kernel.org/linux-mm/CAJD7tkafz_2XAuqE8tGLPEcpLngewhUo=5US14PAtSM9tLBUQg@mail.gmail.com/
Signed-off-by: Liu Shixin <[email protected]>
Tested-by: Yosry Ahmed <[email protected]>
Reviewed-by: "Huang, Ying" <[email protected]>
Reviewed-by: Yosry Ahmed <[email protected]>
---
v5->v6: Fix NULL pointing derefence and hung task problem reported by Sachin.
include/linux/swap.h | 6 ++++++
mm/memcontrol.c | 8 ++++++++
mm/vmscan.c | 24 ++++++++++++++++++++++++
3 files changed, 38 insertions(+)
diff --git a/include/linux/swap.h b/include/linux/swap.h
index 493487ed7c38..19f30a29e1f1 100644
--- a/include/linux/swap.h
+++ b/include/linux/swap.h
@@ -658,6 +658,7 @@ static inline void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_p
}
extern long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg);
+extern long mem_cgroup_get_nr_swapcache_pages(struct mem_cgroup *memcg);
extern bool mem_cgroup_swap_full(struct folio *folio);
#else
static inline void mem_cgroup_swapout(struct folio *folio, swp_entry_t entry)
@@ -680,6 +681,11 @@ static inline long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
return get_nr_swap_pages();
}
+static inline long mem_cgroup_get_nr_swapcache_pages(struct mem_cgroup *memcg)
+{
+ return total_swapcache_pages();
+}
+
static inline bool mem_cgroup_swap_full(struct folio *folio)
{
return vm_swap_full();
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 34bae9e63ff2..26ae2e08fd9a 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -7573,6 +7573,14 @@ long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg)
return nr_swap_pages;
}
+long mem_cgroup_get_nr_swapcache_pages(struct mem_cgroup *memcg)
+{
+ if (mem_cgroup_disabled())
+ return total_swapcache_pages();
+
+ return memcg_page_state(memcg, NR_SWAPCACHE);
+}
+
bool mem_cgroup_swap_full(struct folio *folio)
{
struct mem_cgroup *memcg;
diff --git a/mm/vmscan.c b/mm/vmscan.c
index 3df0e2a59052..48c092cc6088 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -136,6 +136,9 @@ struct scan_control {
/* Always discard instead of demoting to lower tier memory */
unsigned int no_demotion:1;
+ /* Swap space is exhausted, only reclaim swapcache for anon LRU */
+ unsigned int swapcache_only:1;
+
/* Allocation order */
s8 order;
@@ -319,10 +322,22 @@ static inline bool can_reclaim_anon_pages(struct mem_cgroup *memcg,
*/
if (get_nr_swap_pages() > 0)
return true;
+ /* Is there any swapcache pages to reclaim? */
+ if (total_swapcache_pages() > 0) {
+ if (sc)
+ sc->swapcache_only = 1;
+ return true;
+ }
} else {
/* Is the memcg below its swap limit? */
if (mem_cgroup_get_nr_swap_pages(memcg) > 0)
return true;
+ /* Is there any swapcache pages in memcg to reclaim? */
+ if (mem_cgroup_get_nr_swapcache_pages(memcg) > 0) {
+ if (sc)
+ sc->swapcache_only = 1;
+ return true;
+ }
}
/*
@@ -1642,6 +1657,15 @@ static unsigned long isolate_lru_folios(unsigned long nr_to_scan,
*/
scan += nr_pages;
+ /*
+ * Count non-swapcache too because the swapcache pages may
+ * be rare and it takes too much times here if not count
+ * the non-swapcache pages.
+ */
+ if (unlikely(sc->swapcache_only && !is_file_lru(lru) &&
+ !folio_test_swapcache(folio)))
+ goto move;
+
if (!folio_test_lru(folio))
goto move;
if (!sc->may_unmap && folio_mapped(folio))
--
2.25.1
On Fri, 15 Sep 2023 16:34:17 +0800 Liu Shixin <[email protected]> wrote:
> When spaces of swap devices are exhausted, only file pages can be
> reclaimed. But there are still some swapcache pages in anon lru list.
> This can lead to a premature out-of-memory.
>
> The problem is found with such step:
>
> Firstly, set a 9MB disk swap space, then create a cgroup with 10MB
> memory limit, then runs an program to allocates about 15MB memory.
>
> The problem occurs occasionally, which may need about 100 times [1].
>
> Fix it by checking number of swapcache pages in can_reclaim_anon_pages().
> If the number is not zero, return true and set swapcache_only to 1.
> When scan anon lru list in swapcache_only mode, non-swapcache pages will
> be skipped to isolate in order to accelerate reclaim efficiency.
>
> However, in swapcache_only mode, the scan count still increased when scan
> non-swapcache pages because there are large number of non-swapcache pages
> and rare swapcache pages in swapcache_only mode, and if the non-swapcache
> is skipped and do not count, the scan of pages in isolate_lru_folios() can
> eventually lead to hung task, just as Sachin reported [2].
>
> By the way, since there are enough times of memory reclaim before OOM, it
> is not need to isolate too much swapcache pages in one times.
>
mhocko earlier suspected this might impact global reclaim. Have you
looked into that further?
On 2023/10/5 0:18, Andrew Morton wrote:
> On Fri, 15 Sep 2023 16:34:17 +0800 Liu Shixin <[email protected]> wrote:
>
>> When spaces of swap devices are exhausted, only file pages can be
>> reclaimed. But there are still some swapcache pages in anon lru list.
>> This can lead to a premature out-of-memory.
>>
>> The problem is found with such step:
>>
>> Firstly, set a 9MB disk swap space, then create a cgroup with 10MB
>> memory limit, then runs an program to allocates about 15MB memory.
>>
>> The problem occurs occasionally, which may need about 100 times [1].
>>
>> Fix it by checking number of swapcache pages in can_reclaim_anon_pages().
>> If the number is not zero, return true and set swapcache_only to 1.
>> When scan anon lru list in swapcache_only mode, non-swapcache pages will
>> be skipped to isolate in order to accelerate reclaim efficiency.
>>
>> However, in swapcache_only mode, the scan count still increased when scan
>> non-swapcache pages because there are large number of non-swapcache pages
>> and rare swapcache pages in swapcache_only mode, and if the non-swapcache
>> is skipped and do not count, the scan of pages in isolate_lru_folios() can
>> eventually lead to hung task, just as Sachin reported [2].
>>
>> By the way, since there are enough times of memory reclaim before OOM, it
>> is not need to isolate too much swapcache pages in one times.
>>
> mhocko earlier suspected this might impact global reclaim. Have you
> looked into that further?
> .
Alreadly test the case of global reclaim using ltp testcases. In previous version,
the reclaim can stall in isolate_lru_folios() because I didn't count non-swapcache
pages into scan so it will waste too much times scanning non-swapcache pages.
In this version, I count non-swapcache pages into scan too and terminate when
scan reaches the threshold. So it will not stall in reclaim now.