Re: [PATCH] mm: Simplify and batch working set shadow pages LRU isolation locking

[Johannes Weiner <hannes@cmpxchg.org>]

On Thu, Oct 26, 2017 at 04:48:54PM -0700, Andi Kleen wrote:
>  static unsigned long scan_shadow_nodes(struct shrinker *shrinker,
>  				       struct shrink_control *sc)
>  {
> +	struct list_head *tmp, *pos;
>  	unsigned long ret;
> +	LIST_HEAD(nodes);
> +	spinlock_t *lock = NULL;
>  
> -	/* list_lru lock nests inside IRQ-safe mapping->tree_lock */
> +	ret = list_lru_shrink_walk(&shadow_nodes, sc, shadow_lru_isolate, &nodes);
>  	local_irq_disable();
> -	ret = list_lru_shrink_walk(&shadow_nodes, sc, shadow_lru_isolate, NULL);
> +	list_for_each_safe (pos, tmp, &nodes)
> +		free_shadow_node(pos, &lock);

The nlru->lock in list_lru_shrink_walk() is the only thing that keeps
truncation blocked on workingset_update_node() -> list_lru_del() and
so ultimately keeping it from freeing the radix tree node.

It's not safe to access the nodes on the private list after that.

Batching mapping->tree_lock is possible, but you have to keep the
lock-handoff scheme. Pass a &mapping to list_lru_shrink_walk() and
only unlock and spin_trylock(&mapping->tree_lock) if it changes?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>