Re: [PATCH memcg] memcg: enable memory accounting in __alloc_pages_bulk

[Shakeel Butt <shakeelb@google.com>]

+Roman

On Fri, Oct 8, 2021 at 2:23 AM Vasily Averin <vvs@virtuozzo.com> wrote:
>
> Enable memory accounting for bulk page allocator.
>
> Fixes: 387ba26fb1cb ("mm/page_alloc: add a bulk page allocator")
> Cc: <stable@vger.kernel.org>
> Signed-off-by: Vasily Averin <vvs@virtuozzo.com>
> ---
>  mm/page_alloc.c | 64 +++++++++++++++++++++++++++++++++++++++++++++++--
>  1 file changed, 62 insertions(+), 2 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index b37435c274cf..602819a232e5 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -5172,6 +5172,55 @@ static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
>         return true;
>  }
>

Please move the following memcg functions to memcontrol.[h|c] files.

> +#ifdef CONFIG_MEMCG_KMEM
> +static bool memcg_bulk_pre_charge_hook(struct obj_cgroup **objcgp, gfp_t gfp,
> +                                       unsigned int nr_pages)
> +{
> +       struct obj_cgroup *objcg = NULL;
> +
> +       if (!memcg_kmem_enabled() || !(gfp & __GFP_ACCOUNT))
> +               return true;
> +
> +       objcg = get_obj_cgroup_from_current();
> +
> +       if (objcg && obj_cgroup_charge(objcg, gfp, nr_pages << PAGE_SHIFT)) {

Please use obj_cgroup_charge_pages() when you move this code to memcontrol.c

> +               obj_cgroup_put(objcg);
> +               return false;
> +       }
> +       obj_cgroup_get_many(objcg, nr_pages);
> +       *objcgp = objcg;
> +       return true;
> +}
> +
> +static void memcg_bulk_charge_hook(struct obj_cgroup *objcg,
> +                                       struct page *page)
> +{
> +       page->memcg_data = (unsigned long)objcg | MEMCG_DATA_KMEM;
> +}
> +
> +static void memcg_bulk_post_charge_hook(struct obj_cgroup *objcg,
> +                                       unsigned int nr_pages)
> +{
> +       obj_cgroup_uncharge(objcg, nr_pages << PAGE_SHIFT);
> +       percpu_ref_put_many(&objcg->refcnt, nr_pages + 1);

Introduce the obj_cgroup_put_many() and you don't need to keep the
extra ref from the pre hook i.e. put the ref in the pre hook.

> +}
> +#else
> +static bool memcg_bulk_pre_charge_hook(struct obj_cgroup **objcgp, gfp_t gfp,
> +                                       unsigned int nr_pages)
> +{
> +       return true;
> +}
> +
> +static void memcg_bulk_charge_hook(struct obj_cgroup *objcgp,
> +                                       struct page *page)
> +{
> +}
> +
> +static void memcg_bulk_post_charge_hook(struct obj_cgroup *objcg,
> +                                       unsigned int nr_pages)
> +{
> +}
> +#endif
>  /*
>   * __alloc_pages_bulk - Allocate a number of order-0 pages to a list or array
>   * @gfp: GFP flags for the allocation
> @@ -5207,6 +5256,8 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
>         gfp_t alloc_gfp;
>         unsigned int alloc_flags = ALLOC_WMARK_LOW;
>         int nr_populated = 0, nr_account = 0;
> +       unsigned int nr_pre_charge = 0;
> +       struct obj_cgroup *objcg = NULL;
>
>         /*
>          * Skip populated array elements to determine if any pages need
> @@ -5275,6 +5326,10 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
>         if (unlikely(!zone))
>                 goto failed;
>
> +       nr_pre_charge = nr_pages - nr_populated;
> +       if (!memcg_bulk_pre_charge_hook(&objcg, gfp, nr_pre_charge))
> +               goto failed;
> +
>         /* Attempt the batch allocation */
>         local_lock_irqsave(&pagesets.lock, flags);
>         pcp = this_cpu_ptr(zone->per_cpu_pageset);
> @@ -5287,9 +5342,9 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
>                         nr_populated++;
>                         continue;
>                 }
> -
>                 page = __rmqueue_pcplist(zone, 0, ac.migratetype, alloc_flags,
>                                                                 pcp, pcp_list);
> +
>                 if (unlikely(!page)) {
>                         /* Try and get at least one page */
>                         if (!nr_populated)
> @@ -5297,6 +5352,8 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
>                         break;
>                 }
>                 nr_account++;
> +               if (objcg)
> +                       memcg_bulk_charge_hook(objcg, page);

Logically this above should be after prep_new_page().

>
>                 prep_new_page(page, 0, gfp, 0);
>                 if (page_list)
> @@ -5310,13 +5367,16 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
>
>         __count_zid_vm_events(PGALLOC, zone_idx(zone), nr_account);
>         zone_statistics(ac.preferred_zoneref->zone, zone, nr_account);
> +       if (objcg)
> +               memcg_bulk_post_charge_hook(objcg, nr_pre_charge - nr_account);
>
>  out:
>         return nr_populated;
>
>  failed_irq:
>         local_unlock_irqrestore(&pagesets.lock, flags);
> -
> +       if (objcg)
> +               memcg_bulk_post_charge_hook(objcg, nr_pre_charge);
>  failed:
>         page = __alloc_pages(gfp, 0, preferred_nid, nodemask);
>         if (page) {
> --
> 2.31.1
>