Re: [PATCH 1/2] mm/zswap: global lru and shrinker shared by all zswap_pools

[Chengming Zhou <zhouchengming@bytedance.com>]

On 2024/2/13 20:57, Yosry Ahmed wrote:
> On Sun, Feb 11, 2024 at 01:57:04PM +0000, Chengming Zhou wrote:
>> Dynamic zswap_pool creation may create/reuse to have multiple
>> zswap_pools in a list, only the first will be current used.
>>
>> Each zswap_pool has its own lru and shrinker, which is not
>> necessary and has its problem:
>>
>> 1. When memory has pressure, all shrinker of zswap_pools will
>>    try to shrink its own lru, there is no order between them.
>>
>> 2. When zswap limit hit, only the last zswap_pool's shrink_work
>>    will try to shrink its lru, which is inefficient.
>>
>> Anyway, having a global lru and shrinker shared by all zswap_pools
>> is better and efficient.
> 
> It is also a great simplification.
> 
>>
>> Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
>> ---
>>  mm/zswap.c | 153 ++++++++++++++++++++++---------------------------------------
>>  1 file changed, 55 insertions(+), 98 deletions(-)
>>
>> diff --git a/mm/zswap.c b/mm/zswap.c
>> index 62fe307521c9..7668db8c10e3 100644
>> --- a/mm/zswap.c
>> +++ b/mm/zswap.c
>> @@ -176,14 +176,17 @@ struct zswap_pool {
>>  	struct kref kref;
>>  	struct list_head list;
>>  	struct work_struct release_work;
>> -	struct work_struct shrink_work;
>>  	struct hlist_node node;
>>  	char tfm_name[CRYPTO_MAX_ALG_NAME];
>> +};
>> +
>> +struct {
> 
> static?

Ah, right, will add static.

> 
>>  	struct list_lru list_lru;
>> -	struct mem_cgroup *next_shrink;
>> -	struct shrinker *shrinker;
> 
> Just curious, any reason to change the relative ordering of members
> here? It produces a couple more lines of diff :)

The list_lru and nr_stored atomic variable are used in zswap_store/load
hotpath, the other shrinker related sound like cold path. I thought it's
normal and clearer to put them according to their usages.

> 
>>  	atomic_t nr_stored;
>> -};
>> +	struct shrinker *shrinker;
>> +	struct work_struct shrink_work;
>> +	struct mem_cgroup *next_shrink;
>> +} zswap;
>>  
>>  /*
>>   * struct zswap_entry
>> @@ -301,9 +304,6 @@ static void zswap_update_total_size(void)
>>  * pool functions
>>  **********************************/
>>  
>> -static void zswap_alloc_shrinker(struct zswap_pool *pool);
>> -static void shrink_worker(struct work_struct *w);
>> -
>>  static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
>>  {
>>  	int i;
>> @@ -353,30 +353,16 @@ static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
>>  	if (ret)
>>  		goto error;
>>  
>> -	zswap_alloc_shrinker(pool);
>> -	if (!pool->shrinker)
>> -		goto error;
>> -
>> -	pr_debug("using %s compressor\n", pool->tfm_name);
>> -
> 
> Why are we removing this debug print?

Oh, I just noticed it's only necessary to print dmesg when "create" success,
the below "zswap_pool_debug()" will print its compressor too.

> 
>>  	/* being the current pool takes 1 ref; this func expects the
>>  	 * caller to always add the new pool as the current pool
>>  	 */
>>  	kref_init(&pool->kref);
>>  	INIT_LIST_HEAD(&pool->list);
>> -	if (list_lru_init_memcg(&pool->list_lru, pool->shrinker))
>> -		goto lru_fail;
>> -	shrinker_register(pool->shrinker);
>> -	INIT_WORK(&pool->shrink_work, shrink_worker);
>> -	atomic_set(&pool->nr_stored, 0);
>>  
>>  	zswap_pool_debug("created", pool);
>>  
>>  	return pool;
>>  
>> -lru_fail:
>> -	list_lru_destroy(&pool->list_lru);
>> -	shrinker_free(pool->shrinker);
>>  error:
>>  	if (pool->acomp_ctx)
>>  		free_percpu(pool->acomp_ctx);
> [..]
>> @@ -816,14 +777,10 @@ void zswap_folio_swapin(struct folio *folio)
>>  
>>  void zswap_memcg_offline_cleanup(struct mem_cgroup *memcg)
>>  {
>> -	struct zswap_pool *pool;
>> -
>> -	/* lock out zswap pools list modification */
>> +	/* lock out zswap shrinker walking memcg tree */
>>  	spin_lock(&zswap_pools_lock);
>> -	list_for_each_entry(pool, &zswap_pools, list) {
>> -		if (pool->next_shrink == memcg)
>> -			pool->next_shrink = mem_cgroup_iter(NULL, pool->next_shrink, NULL);
>> -	}
>> +	if (zswap.next_shrink == memcg)
>> +		zswap.next_shrink = mem_cgroup_iter(NULL, zswap.next_shrink, NULL);
> 
> Now that next_shrink has nothing to do with zswap pools, it feels weird
> that we are using zswap_pools_lock for its synchronization. Does it make
> sense to have a separate lock for it just for semantic purposes?

Agree, I think so, it's clearer to have another lock.

> 
>>  	spin_unlock(&zswap_pools_lock);
>>  }
>>  
> [..]
>> @@ -1328,7 +1284,6 @@ static unsigned long zswap_shrinker_scan(struct shrinker *shrinker,
>>  static unsigned long zswap_shrinker_count(struct shrinker *shrinker,
>>  		struct shrink_control *sc)
>>  {
>> -	struct zswap_pool *pool = shrinker->private_data;
>>  	struct mem_cgroup *memcg = sc->memcg;
>>  	struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(sc->nid));
>>  	unsigned long nr_backing, nr_stored, nr_freeable, nr_protected;
>> @@ -1343,7 +1298,7 @@ static unsigned long zswap_shrinker_count(struct shrinker *shrinker,
>>  #else
>>  	/* use pool stats instead of memcg stats */
>>  	nr_backing = get_zswap_pool_size(pool) >> PAGE_SHIFT;
> 
> "pool" is still being used here.

Oops, should be changed to zswap_pool_total_size here.

> 
>> -	nr_stored = atomic_read(&pool->nr_stored);
>> +	nr_stored = atomic_read(&zswap.nr_stored);
>>  #endif
>>  
>>  	if (!nr_stored)
> [..]  
>> @@ -1804,6 +1749,21 @@ static int zswap_setup(void)
>>  	if (ret)
>>  		goto hp_fail;
>>  
>> +	shrink_wq = alloc_workqueue("zswap-shrink",
>> +			WQ_UNBOUND|WQ_MEM_RECLAIM, 1);
>> +	if (!shrink_wq)
>> +		goto hp_fail;
> 
> I think we need a new label here to call cpuhp_remove_multi_state(), but
> apparently this is missing from the current code for some reason.

You are right! This should use a new label to "cpuhp_remove_multi_state()",
will fix it.

> 
>> +
>> +	zswap.shrinker = zswap_alloc_shrinker();
>> +	if (!zswap.shrinker)
>> +		goto shrinker_fail;
>> +	if (list_lru_init_memcg(&zswap.list_lru, zswap.shrinker))
>> +		goto lru_fail;
>> +	shrinker_register(zswap.shrinker);
>> +
>> +	INIT_WORK(&zswap.shrink_work, shrink_worker);
>> +	atomic_set(&zswap.nr_stored, 0);
>> +
>>  	pool = __zswap_pool_create_fallback();
>>  	if (pool) {
>>  		pr_info("loaded using pool %s/%s\n", pool->tfm_name,
>> @@ -1815,19 +1775,16 @@ static int zswap_setup(void)
>>  		zswap_enabled = false;
>>  	}
>>  
>> -	shrink_wq = alloc_workqueue("zswap-shrink",
>> -			WQ_UNBOUND|WQ_MEM_RECLAIM, 1);
>> -	if (!shrink_wq)
>> -		goto fallback_fail;
>> -
>>  	if (zswap_debugfs_init())
>>  		pr_warn("debugfs initialization failed\n");
>>  	zswap_init_state = ZSWAP_INIT_SUCCEED;
>>  	return 0;
>>  
>> -fallback_fail:
>> -	if (pool)
>> -		zswap_pool_destroy(pool);
>> +lru_fail:
>> +	list_lru_destroy(&zswap.list_lru);
> 
> Do we need to call list_lru_destroy() here? I know it is currently being
> called if list_lru_init_memcg() fails, but I fail to understand why. It
> seems like list_lru_destroy() will do nothing anyway.

Right, it's not needed to call list_lru_destroy() here, it should do nothing,
will delete it.

Thanks!

> 
>> +	shrinker_free(zswap.shrinker);
>> +shrinker_fail:
>> +	destroy_workqueue(shrink_wq);
>>  hp_fail:
>>  	kmem_cache_destroy(zswap_entry_cache);
>>  cache_fail:
>>
>> -- 
>> b4 0.10.1