Re: [PATCH v2 1/2] zswap: make shrinking memcg-aware

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 





On Tue, Oct 17, 2023 at 7:44 PM Ryan Roberts <ryan.roberts@xxxxxxx> wrote:
On 19/09/2023 18:14, Nhat Pham wrote:
> From: Domenico Cerasuolo <cerasuolodomenico@xxxxxxxxx>
>
> Currently, we only have a single global LRU for zswap. This makes it
> impossible to perform worload-specific shrinking - an memcg cannot
> determine which pages in the pool it owns, and often ends up writing
> pages from other memcgs. This issue has been previously observed in
> practice and mitigated by simply disabling memcg-initiated shrinking:
>
> https://lore.kernel.org/all/20230530232435.3097106-1-nphamcs@xxxxxxxxx/T/#u
>
> This patch fully resolves the issue by replacing the global zswap LRU
> with memcg- and NUMA-specific LRUs, and modify the reclaim logic:
>
> a) When a store attempt hits an memcg limit, it now triggers a
>    synchronous reclaim attempt that, if successful, allows the new
>    hotter page to be accepted by zswap.
> b) If the store attempt instead hits the global zswap limit, it will
>    trigger an asynchronous reclaim attempt, in which an memcg is
>    selected for reclaim in a round-robin-like fashion.
>
> Signed-off-by: Domenico Cerasuolo <cerasuolodomenico@xxxxxxxxx>
> Co-developed-by: Nhat Pham <nphamcs@xxxxxxxxx>
> Signed-off-by: Nhat Pham <nphamcs@xxxxxxxxx>
> ---
>  include/linux/list_lru.h   |  39 +++++++
>  include/linux/memcontrol.h |   5 +
>  include/linux/zswap.h      |   9 ++
>  mm/list_lru.c              |  46 ++++++--
>  mm/swap_state.c            |  19 ++++
>  mm/zswap.c                 | 221 +++++++++++++++++++++++++++++--------
>  6 files changed, 287 insertions(+), 52 deletions(-)
>

[...]

> @@ -1199,8 +1272,10 @@ bool zswap_store(struct folio *folio)
>       struct scatterlist input, output;
>       struct crypto_acomp_ctx *acomp_ctx;
>       struct obj_cgroup *objcg = NULL;
> +     struct mem_cgroup *memcg = NULL;
>       struct zswap_pool *pool;
>       struct zpool *zpool;
> +     int lru_alloc_ret;
>       unsigned int dlen = PAGE_SIZE;
>       unsigned long handle, value;
>       char *buf;
> @@ -1218,14 +1293,15 @@ bool zswap_store(struct folio *folio)
>       if (!zswap_enabled || !tree)
>               return false;

> -     /*
> -      * XXX: zswap reclaim does not work with cgroups yet. Without a
> -      * cgroup-aware entry LRU, we will push out entries system-wide based on
> -      * local cgroup limits.
> -      */
>       objcg = get_obj_cgroup_from_folio(folio);
> -     if (objcg && !obj_cgroup_may_zswap(objcg))
> -             goto reject;
> +     if (objcg && !obj_cgroup_may_zswap(objcg)) {
> +             memcg = get_mem_cgroup_from_objcg(objcg);
> +             if (shrink_memcg(memcg)) {
> +                     mem_cgroup_put(memcg);
> +                     goto reject;
> +             }
> +             mem_cgroup_put(memcg);
> +     }

>       /* reclaim space if needed */
>       if (zswap_is_full()) {
> @@ -1240,7 +1316,11 @@ bool zswap_store(struct folio *folio)
>               else
>                       zswap_pool_reached_full = false;
>       }
> -
> +     pool = zswap_pool_current_get();
> +     if (!pool) {
> +             ret = -EINVAL;
> +             goto reject;
> +     }


Hi, I'm working to add support for large folios within zswap, and noticed this
piece of code added by this change. I don't see any corresponding put. Have I
missed some detail or is there a bug here?


>       /* allocate entry */
>       entry = zswap_entry_cache_alloc(GFP_KERNEL);
>       if (!entry) {
> @@ -1256,6 +1336,7 @@ bool zswap_store(struct folio *folio)
>                       entry->length = 0;
>                       entry->value = value;
>                       atomic_inc(&zswap_same_filled_pages);
> +                     zswap_pool_put(pool);

I see you put it in this error path, but after that, there is no further mention.

>                       goto insert_entry;
>               }
>               kunmap_atomic(src);
> @@ -1264,6 +1345,15 @@ bool zswap_store(struct folio *folio)
>       if (!zswap_non_same_filled_pages_enabled)
>               goto freepage;

> +     if (objcg) {
> +             memcg = get_mem_cgroup_from_objcg(objcg);
> +             lru_alloc_ret = memcg_list_lru_alloc(memcg, &pool->list_lru, GFP_KERNEL);
> +             mem_cgroup_put(memcg);
> +
> +             if (lru_alloc_ret)
> +                     goto freepage;
> +     }
> +
>       /* if entry is successfully added, it keeps the reference */
>       entry->pool = zswap_pool_current_get();

The entry takes it's reference to the pool here.

Thanks,
Ryan

Thanks Ryan, I think you're right. Coincidentally, we're about to send a new
version of the series, and will make sure to address this too.

[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux