Re: lockdep complaints in slab allocator

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Mon, Nov 23, 2009 at 10:01:15PM +0200, Pekka Enberg wrote:
> On Mon, 2009-11-23 at 21:50 +0200, Pekka Enberg wrote:
> > On Mon, 23 Nov 2009, Pekka Enberg wrote:
> > > > That turns out to be _very_ hard. How about something like the following
> > > > untested patch which delays slab_destroy() while we're under nc->lock.
> > 
> > On Mon, 2009-11-23 at 13:30 -0600, Christoph Lameter wrote:
> > > Code changes to deal with a diagnostic issue?
> > 
> > OK, fair enough. If I suffer permanent brain damage from staring at the
> > SLAB code for too long, I hope you and Matt will chip in to pay for my
> > medication.
> > 
> > I think I was looking at the wrong thing here. The problem is in
> > cache_free_alien() so the comment in slab_destroy() isn't relevant.
> > Looking at init_lock_keys() we already do special lockdep annotations
> > but there's a catch (as explained in a comment on top of
> > on_slab_alc_key):
> > 
> >  * We set lock class for alien array caches which are up during init.
> >  * The lock annotation will be lost if all cpus of a node goes down and
> >  * then comes back up during hotplug
> > 
> > Paul said he was running CPU hotplug so maybe that explains the problem?
> 
> Maybe something like this untested patch fixes the issue...

I will give it a go!

							Thanx, Paul

> 			Pekka
> 
> diff --git a/mm/slab.c b/mm/slab.c
> index 7dfa481..84de47e 100644
> --- a/mm/slab.c
> +++ b/mm/slab.c
> @@ -604,6 +604,26 @@ static struct kmem_cache cache_cache = {
> 
>  #define BAD_ALIEN_MAGIC 0x01020304ul
> 
> +/*
> + * chicken and egg problem: delay the per-cpu array allocation
> + * until the general caches are up.
> + */
> +static enum {
> +	NONE,
> +	PARTIAL_AC,
> +	PARTIAL_L3,
> +	EARLY,
> +	FULL
> +} g_cpucache_up;
> +
> +/*
> + * used by boot code to determine if it can use slab based allocator
> + */
> +int slab_is_available(void)
> +{
> +	return g_cpucache_up >= EARLY;
> +}
> +
>  #ifdef CONFIG_LOCKDEP
> 
>  /*
> @@ -620,40 +640,52 @@ static struct kmem_cache cache_cache = {
>  static struct lock_class_key on_slab_l3_key;
>  static struct lock_class_key on_slab_alc_key;
> 
> -static inline void init_lock_keys(void)
> -
> +static void init_node_lock_keys(int q)
>  {
> -	int q;
>  	struct cache_sizes *s = malloc_sizes;
> 
> -	while (s->cs_size != ULONG_MAX) {
> -		for_each_node(q) {
> -			struct array_cache **alc;
> -			int r;
> -			struct kmem_list3 *l3 = s->cs_cachep->nodelists[q];
> -			if (!l3 || OFF_SLAB(s->cs_cachep))
> -				continue;
> -			lockdep_set_class(&l3->list_lock, &on_slab_l3_key);
> -			alc = l3->alien;
> -			/*
> -			 * FIXME: This check for BAD_ALIEN_MAGIC
> -			 * should go away when common slab code is taught to
> -			 * work even without alien caches.
> -			 * Currently, non NUMA code returns BAD_ALIEN_MAGIC
> -			 * for alloc_alien_cache,
> -			 */
> -			if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC)
> -				continue;
> -			for_each_node(r) {
> -				if (alc[r])
> -					lockdep_set_class(&alc[r]->lock,
> -					     &on_slab_alc_key);
> -			}
> +	if (g_cpucache_up != FULL)
> +		return;
> +
> +	for (s = malloc_sizes; s->cs_size != ULONG_MAX; s++) {
> +		struct array_cache **alc;
> +		struct kmem_list3 *l3;
> +		int r;
> +
> +		l3 = s->cs_cachep->nodelists[q];
> +		if (!l3 || OFF_SLAB(s->cs_cachep))
> +			return;
> +		lockdep_set_class(&l3->list_lock, &on_slab_l3_key);
> +		alc = l3->alien;
> +		/*
> +		 * FIXME: This check for BAD_ALIEN_MAGIC
> +		 * should go away when common slab code is taught to
> +		 * work even without alien caches.
> +		 * Currently, non NUMA code returns BAD_ALIEN_MAGIC
> +		 * for alloc_alien_cache,
> +		 */
> +		if (!alc || (unsigned long)alc == BAD_ALIEN_MAGIC)
> +			return;
> +		for_each_node(r) {
> +			if (alc[r])
> +				lockdep_set_class(&alc[r]->lock,
> +					&on_slab_alc_key);
>  		}
> -		s++;
>  	}
>  }
> +
> +static inline void init_lock_keys(void)
> +{
> +	int node;
> +
> +	for_each_node(node)
> +		init_node_lock_keys(node);
> +}
>  #else
> +static void init_node_lock_keys(int q)
> +{
> +}
> +
>  static inline void init_lock_keys(void)
>  {
>  }
> @@ -665,26 +697,6 @@ static inline void init_lock_keys(void)
>  static DEFINE_MUTEX(cache_chain_mutex);
>  static struct list_head cache_chain;
> 
> -/*
> - * chicken and egg problem: delay the per-cpu array allocation
> - * until the general caches are up.
> - */
> -static enum {
> -	NONE,
> -	PARTIAL_AC,
> -	PARTIAL_L3,
> -	EARLY,
> -	FULL
> -} g_cpucache_up;
> -
> -/*
> - * used by boot code to determine if it can use slab based allocator
> - */
> -int slab_is_available(void)
> -{
> -	return g_cpucache_up >= EARLY;
> -}
> -
>  static DEFINE_PER_CPU(struct delayed_work, reap_work);
> 
>  static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
> @@ -1254,6 +1266,8 @@ static int __cpuinit cpuup_prepare(long cpu)
>  		kfree(shared);
>  		free_alien_cache(alien);
>  	}
> +	init_node_lock_keys(node);
> +
>  	return 0;
>  bad:
>  	cpuup_canceled(cpu);
> 
>