Re: [PATCH 10/10] dm-zoned: Drive-managed zoned block device target

[Bart Van Assche <Bart.VanAssche@sandisk.com>]

On Fri, 2017-04-21 at 12:55 +0900, damien.lemoal@wdc.com wrote:
> +static void dmz_shrink_mblock_cache(struct dmz_target *dmz, bool idle)
> +{
> +       struct dmz_mblock *mblk;
> +       unsigned int nr_mblks;
> +
> +       if (!dmz->max_nr_mblks)
> +               return;
> +
> +       if (idle)
> +               nr_mblks = dmz->min_nr_mblks;
> +       else
> +               nr_mblks = dmz->max_nr_mblks;
> +
> +       while (atomic_read(&dmz->nr_mblks) > nr_mblks &&
> +              !list_empty(&dmz->mblk_lru_list)) {
> +               mblk = list_first_entry(&dmz->mblk_lru_list,
> +                                       struct dmz_mblock, link);
> +               list_del_init(&mblk->link);
> +               rb_erase(&mblk->node, &dmz->mblk_rbtree);
> +               dmz_free_mblock(dmz, mblk);
> +       }
> +}

(off-list)

Hello Damien,

Is mblk_lru_list perhaps a cache that should be freed under memory pressure?
If so, if you repost this patch series please add a shrinker (struct shrinker
+ register_shrinker()) such that this memory can be freed if memory pressure
becomes too high.

Thanks,

Bart.