Manish Jaggi noticed that running LTP oom01/oom02 ltp tests with high core
count causes random kernel panics when an OOM victim which consumed memory
in a way the OOM reaper does not help was selected by the OOM killer [1].
Since commit 696453e66630ad45 ("mm, oom: task_will_free_mem should skip
oom_reaped tasks") changed task_will_free_mem(current) in out_of_memory()
to return false as soon as MMF_OOM_SKIP is set, many threads sharing the
victim's mm were not able to try allocation from memory reserves after the
OOM reaper gave up reclaiming memory.
Therefore, this patch allows OOM victims to use ALLOC_OOM watermark for
last second allocation attempt.
[1] http://lkml.kernel.org/r/e6c83a26-1d59-4afd-55cf-04e58bdde188@xxxxxxxxxxxxxxxxxx
Fixes: 696453e66630ad45 ("mm, oom: task_will_free_mem should skip oom_reaped tasks")
Signed-off-by: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>
Reported-by: Manish Jaggi <mjaggi@xxxxxxxxxxxxxxxxxx>
Acked-by: Michal Hocko <mhocko@xxxxxxxx>
Cc: Michal Hocko <mhocko@xxxxxxxx>
Cc: Oleg Nesterov <oleg@xxxxxxxxxx>
Cc: Vladimir Davydov <vdavydov.dev@xxxxxxxxx>
Cc: David Rientjes <rientjes@xxxxxxxxxx>
Cc: Andrea Arcangeli <aarcange@xxxxxxxxxx>
Cc: Johannes Weiner <hannes@xxxxxxxxxxx>
---
mm/page_alloc.c | 39 +++++++++++++++++++++++++++++----------
1 file changed, 29 insertions(+), 10 deletions(-)
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 73f5d45..5d054a4 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -3309,6 +3309,10 @@ void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
return page;
}
+static struct page *alloc_pages_before_oomkill(gfp_t gfp_mask,
+ unsigned int order,
+ const struct alloc_context *ac);
+
static inline struct page *
__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
const struct alloc_context *ac, unsigned long *did_some_progress)
@@ -3334,16 +3338,7 @@ void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
return NULL;
}
- /*
- * Go through the zonelist yet one more time, keep very high watermark
- * here, this is only to catch a parallel oom killing, we must fail if
- * we're still under heavy pressure. But make sure that this reclaim
- * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY
- * allocation which will never fail due to oom_lock already held.
- */
- page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
- ~__GFP_DIRECT_RECLAIM, order,
- ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
+ page = alloc_pages_before_oomkill(gfp_mask, order, ac);
if (page)
goto out;
@@ -3755,6 +3750,30 @@ bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
return !!__gfp_pfmemalloc_flags(gfp_mask);
}
+static struct page *alloc_pages_before_oomkill(gfp_t gfp_mask,
+ unsigned int order,
+ const struct alloc_context *ac)
+{
+ /*
+ * Go through the zonelist yet one more time, keep very high watermark
+ * here, this is only to catch a parallel oom killing, we must fail if
+ * we're still under heavy pressure. But make sure that this reclaim
+ * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY
+ * allocation which will never fail due to oom_lock already held.
+ * Also, make sure that OOM victims can try ALLOC_OOM watermark
+ * in case they haven't tried ALLOC_OOM watermark.
+ */
+ int alloc_flags = ALLOC_CPUSET | ALLOC_WMARK_HIGH;
+ int reserve_flags;
+
+ gfp_mask |= __GFP_HARDWALL;
+ gfp_mask &= ~__GFP_DIRECT_RECLAIM;
+ reserve_flags = __gfp_pfmemalloc_flags(gfp_mask);
+ if (reserve_flags)
+ alloc_flags = reserve_flags;
+ return get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
+}
+
/*
* Checks whether it makes sense to retry the reclaim to make a forward progress
* for the given allocation request.
--
1.8.3.1
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
