On Wed 30-05-18 13:42:56, Andrew Morton wrote:

On Mon, 21 May 2018 03:39:46 +0100 ufo19890607 [off-list ref] wrote:

quoted

From: yuzhoujian <redacted>

The dump_header does not print the memcg's name when the system
oom happened. So users cannot locate the certain container which
contains the task that has been killed by the oom killer.

System oom report will print the memcg's name after this patch,
so users can get the memcg's path from the oom report and check
the certain container more quickly.

lkp-robot is reporting an oops.

quoted

--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c

@@ -433,6 +433,7 @@ static void dump_header(struct oom_control *oc, struct task_struct *p)
 	if (is_memcg_oom(oc))
 		mem_cgroup_print_oom_info(oc->memcg, p);
 	else {
+		mem_cgroup_print_oom_memcg_name(oc->memcg, p);
 		show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask);
 		if (is_dump_unreclaim_slabs())
 			dump_unreclaimable_slab();

static inline bool is_memcg_oom(struct oom_control *oc)
{
	return oc->memcg != NULL;
}

So in the mem_cgroup_print_oom_memcg_name() call which this patch adds,
oc->memcg is known to be NULL.  How can this possibly work?  

This version is broken. The current version [1] seems to be doing the
right thing in that regards AFAICS. It has some other issues though.
Can we drop the current code from the mmotm tree and start over?

[1] http://lkml.kernel.org/r/1527413551-5982-1-git-send-email-ufo19890607@gmail.com
-- 
Michal Hocko
SUSE Labs