Re: [PATCH -V4 04/10] memcg: Add HugeTLB extension

From: Michal Hocko
Date: Wed Mar 28 2012 - 07:33:09 EST


On Fri 16-03-12 23:09:24, Aneesh Kumar K.V wrote:
> From: "Aneesh Kumar K.V" <aneesh.kumar@xxxxxxxxxxxxxxxxxx>
>
> This patch implements a memcg extension that allows us to control
> HugeTLB allocations via memory controller.

And the infrastructure is not used at this stage (you forgot to
mention).
The changelog should be much more descriptive.

>
> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@xxxxxxxxxxxxxxxxxx>
> ---
> include/linux/hugetlb.h | 1 +
> include/linux/memcontrol.h | 42 +++++++++++++
> init/Kconfig | 8 +++
> mm/hugetlb.c | 2 +-
> mm/memcontrol.c | 138 ++++++++++++++++++++++++++++++++++++++++++++
> 5 files changed, 190 insertions(+), 1 deletions(-)
>
[...]
> diff --git a/init/Kconfig b/init/Kconfig
> index 3f42cd6..f0eb8aa 100644
> --- a/init/Kconfig
> +++ b/init/Kconfig
> @@ -725,6 +725,14 @@ config CGROUP_PERF
>
> Say N if unsure.
>
> +config MEM_RES_CTLR_HUGETLB
> + bool "Memory Resource Controller HugeTLB Extension (EXPERIMENTAL)"
> + depends on CGROUP_MEM_RES_CTLR && HUGETLB_PAGE && EXPERIMENTAL
> + default n
> + help
> + Add HugeTLB management to memory resource controller. When you
> + enable this, you can put a per cgroup limit on HugeTLB usage.

How does it interact with the hard/soft limists etc...

[...]
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index 6728a7a..4b36c5e 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -235,6 +235,10 @@ struct mem_cgroup {
> */
> struct res_counter memsw;
> /*
> + * the counter to account for hugepages from hugetlb.
> + */
> + struct res_counter hugepage[HUGE_MAX_HSTATE];
> + /*
> * Per cgroup active and inactive list, similar to the
> * per zone LRU lists.
> */
> @@ -3156,6 +3160,128 @@ static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
> }
> #endif
>
> +#ifdef CONFIG_MEM_RES_CTLR_HUGETLB
> +static bool mem_cgroup_have_hugetlb_usage(struct mem_cgroup *memcg)
> +{
> + int idx;
> + for (idx = 0; idx < hugetlb_max_hstate; idx++) {

Maybe we should expose for_each_hstate as well...

> + if (memcg->hugepage[idx].usage > 0)
> + return 1;
> + }
> + return 0;
> +}
> +
> +int mem_cgroup_hugetlb_charge_page(int idx, unsigned long nr_pages,
> + struct mem_cgroup **ptr)
> +{
> + int ret = 0;
> + struct mem_cgroup *memcg;
> + struct res_counter *fail_res;
> + unsigned long csize = nr_pages * PAGE_SIZE;
> +
> + if (mem_cgroup_disabled())
> + return 0;
> +again:
> + rcu_read_lock();
> + memcg = mem_cgroup_from_task(current);
> + if (!memcg)
> + memcg = root_mem_cgroup;
> + if (mem_cgroup_is_root(memcg)) {
> + rcu_read_unlock();
> + goto done;
> + }
> + if (!css_tryget(&memcg->css)) {
> + rcu_read_unlock();
> + goto again;
> + }
> + rcu_read_unlock();
> +
> + ret = res_counter_charge(&memcg->hugepage[idx], csize, &fail_res);
> + css_put(&memcg->css);
> +done:
> + *ptr = memcg;

Why do we set ptr even for the failure case after we dropped a
reference?

> + return ret;
> +}
> +
> +void mem_cgroup_hugetlb_commit_charge(int idx, unsigned long nr_pages,
> + struct mem_cgroup *memcg,
> + struct page *page)
> +{
> + struct page_cgroup *pc;
> +
> + if (mem_cgroup_disabled())
> + return;
> +
> + pc = lookup_page_cgroup(page);
> + lock_page_cgroup(pc);
> + if (unlikely(PageCgroupUsed(pc))) {
> + unlock_page_cgroup(pc);
> + mem_cgroup_hugetlb_uncharge_memcg(idx, nr_pages, memcg);
> + return;
> + }
> + pc->mem_cgroup = memcg;
> + /*
> + * We access a page_cgroup asynchronously without lock_page_cgroup().
> + * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
> + * is accessed after testing USED bit. To make pc->mem_cgroup visible
> + * before USED bit, we need memory barrier here.
> + * See mem_cgroup_add_lru_list(), etc.
> + */
> + smp_wmb();

Is this really necessary for hugetlb pages as well?

> + SetPageCgroupUsed(pc);
> +
> + unlock_page_cgroup(pc);
> + return;
> +}
> +
[...]
> @@ -4887,6 +5013,7 @@ err_cleanup:
> static struct cgroup_subsys_state * __ref
> mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
> {
> + int idx;
> struct mem_cgroup *memcg, *parent;
> long error = -ENOMEM;
> int node;
> @@ -4929,9 +5056,14 @@ mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
> * mem_cgroup(see mem_cgroup_put).
> */
> mem_cgroup_get(parent);
> + for (idx = 0; idx < HUGE_MAX_HSTATE; idx++)

Do we have to init all hstates or is hugetlb_max_hstate enough?

> + res_counter_init(&memcg->hugepage[idx],
> + &parent->hugepage[idx]);
> } else {
> res_counter_init(&memcg->res, NULL);
> res_counter_init(&memcg->memsw, NULL);
> + for (idx = 0; idx < HUGE_MAX_HSTATE; idx++)
> + res_counter_init(&memcg->hugepage[idx], NULL);

Same here
--
Michal Hocko
SUSE Labs
SUSE LINUX s.r.o.
Lihovarska 1060/12
190 00 Praha 9
Czech Republic
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/