Re: [PATCH 3/3] mm: oom: show unreclaimable slab info when unreclaimable slabs > user memory

[Yang Shi]

On 10/4/17 7:27 AM, Michal Hocko wrote:
> On Wed 04-10-17 02:06:17, Yang Shi wrote:
> > +static bool is_dump_unreclaim_slabs(void)
> > +{
> > +	unsigned long nr_lru;
> > +
> > +	nr_lru = global_node_page_state(NR_ACTIVE_ANON) +
> > +		 global_node_page_state(NR_INACTIVE_ANON) +
> > +		 global_node_page_state(NR_ACTIVE_FILE) +
> > +		 global_node_page_state(NR_INACTIVE_FILE) +
> > +		 global_node_page_state(NR_ISOLATED_ANON) +
> > +		 global_node_page_state(NR_ISOLATED_FILE) +
> > +		 global_node_page_state(NR_UNEVICTABLE);
> > +
> > +	return (global_node_page_state(NR_SLAB_UNRECLAIMABLE) > nr_lru);
> > +}
>
> I am sorry I haven't pointed this earlier (I was following only half
> way) but this should really be memcg aware. You are checking only global
> counters. I do not think it is an absolute must to provide per-memcg
> data but you should at least check !is_memcg_oom(oc).

BTW, I saw there is already such check in dump_header that looks like 
the below code:

        if (oc->memcg)
                mem_cgroup_print_oom_info(oc->memcg, p);
        else
                show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask);

I'm supposed it'd better to replace "oc->memcg" to "is_memcg_oom(oc)" 
since they do the same check and "is_memcg_oom" interface sounds preferable.

Then I'm going to move unreclaimable slabs dump to the "else" block.

Yang

> [...]
> > +void dump_unreclaimable_slab(void)
> > +{
> > +	struct kmem_cache *s, *s2;
> > +	struct slabinfo sinfo;
> > +
> > +	pr_info("Unreclaimable slab info:\n");
> > +	pr_info("Name                      Used          Total\n");
> > +
> > +	/*
> > +	 * Here acquiring slab_mutex is risky since we don't prefer to get
> > +	 * sleep in oom path. But, without mutex hold, it may introduce a
> > +	 * risk of crash.
> > +	 * Use mutex_trylock to protect the list traverse, dump nothing
> > +	 * without acquiring the mutex.
> > +	 */
> > +	if (!mutex_trylock(&slab_mutex))
> > +		return;
>
> I would move the trylock up so that we do not get empty and confusing
> Unreclaimable slab info: and add a note that we are not dumping anything
> due to lock contention
> 	pr_warn("excessive unreclaimable slab memory but cannot dump stats to give you more details\n");
>
> Other than that this looks sensible to me.
>
> > +	list_for_each_entry_safe(s, s2, &slab_caches, list) {
> > +		if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT))
> > +			continue;
> > +
> > +		memset(&sinfo, 0, sizeof(sinfo));
> > +		get_slabinfo(s, &sinfo);
> > +
> > +		if (sinfo.num_objs > 0)
> > +			pr_info("%-17s %10luKB %10luKB\n", cache_name(s),
> > +				(sinfo.active_objs * s->size) / 1024,
> > +				(sinfo.num_objs * s->size) / 1024);
> > +	}
> > +	mutex_unlock(&slab_mutex);
> > +}
> > +
> >   #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
> >   void *memcg_slab_start(struct seq_file *m, loff_t *pos)
> >   {
> > --
> > 1.8.3.1