Re: [PATCH 2/2] mm: oom: show unreclaimable slab info when kernel panic

[David Rientjes <rientjes@google.com>]

On Wed, 20 Sep 2017, Yang Shi wrote:

> > > --- a/mm/slab_common.c
> > > +++ b/mm/slab_common.c
> > > @@ -35,6 +35,8 @@
> > >   static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
> > >   		    slab_caches_to_rcu_destroy_workfn);
> > >   +#define K(x) ((x)/1024)
> > > +
> > >   /*
> > >    * Set of flags that will prevent slab merging
> > >    */
> > > @@ -1272,6 +1274,34 @@ static int slab_show(struct seq_file *m, void *p)
> > >   	return 0;
> > >   }
> > >   +void show_unreclaimable_slab()
> > > +{
> > > +	struct kmem_cache *s = NULL;
> > > +	struct slabinfo sinfo;
> > > +
> > > +	memset(&sinfo, 0, sizeof(sinfo));
> > > +
> > > +	printk("Unreclaimable slabs:\n");
> > > +
> > > +	/*
> > > +	 * Here acquiring slab_mutex is unnecessary since we don't prefer to
> > > +	 * get sleep in oom path right before kernel panic, and avoid race
> > > condition.
> > > +	 * Since it is already oom, so there should be not any big allocation
> > > +	 * which could change the statistics significantly.
> > > +	 */
> > > +	list_for_each_entry(s, &slab_caches, list) {
> > > +		if (!is_root_cache(s))
> > > +			continue;
> > > +
> > > +		get_slabinfo(s, &sinfo);
> > > +
> > > +		if (!is_reclaimable(s) && sinfo.num_objs > 0)
> > > +			printk("%-17s %luKB\n", cache_name(s),
> > > K(sinfo.num_objs * s->size));
> > > +	}
> > 
> > I like this, but could we be even more helpful by giving the user more
> > information from sinfo beyond just the total size of objects allocated?
> 
> Sure, we definitely can. But, the question is what info is helpful to users to
> diagnose oom other than the size.
> 
> I think of the below:
> 	- the number of active objs, the number of total objs, the percentage
> of active objs per cache
> 	- the number of active slabs, the number of total slabs, the
> percentage of active slabs per cache
> 
> Anything else?
> 

Right now it's a useful tool to find out what unreclaimable slab is 
sitting around that is causing the system to run out of memory.  If we 
knew how much of this slab is actually in use vs free, it can determine if 
its stranding or if there's a bug in the slab allocator itself.

We wouldn't need percentages, we can calculate that directly from the 
data if necessary.