Re: [PATCH v4 3/8] foundations of per-cgroup memory pressure controlling.

From: Glauber Costa
Date: Tue Oct 04 2011 - 03:14:31 EST

On 10/04/2011 04:57 AM, KAMEZAWA Hiroyuki wrote:
On Mon, 3 Oct 2011 14:18:38 +0400
Glauber Costa<glommer@xxxxxxxxxxxxx> wrote:

This patch converts struct sock fields memory_pressure,
memory_allocated, sockets_allocated, and sysctl_mem (now prot_mem)
to function pointers, receiving a struct mem_cgroup parameter.

enter_memory_pressure is kept the same, since all its callers
have socket a context, and the kmem_cgroup can be derived from
the socket itself.

To keep things working, the patch convert all users of those fields
to use acessor functions.

In my benchmarks I didn't see a significant performance difference
with this patch applied compared to a baseline (around 1 % diff, thus
inside error margin).

Signed-off-by: Glauber Costa<glommer@xxxxxxxxxxxxx>
CC: David S. Miller<davem@xxxxxxxxxxxxx>
CC: Hiroyouki Kamezawa<kamezawa.hiroyu@xxxxxxxxxxxxxx>
CC: Eric W. Biederman<ebiederm@xxxxxxxxxxxx>

A nitpick.

struct sock;
+struct proto;
void sock_update_memcg(struct sock *sk);
void sock_release_memcg(struct sock *sk);
+void memcg_sock_mem_alloc(struct mem_cgroup *mem, struct proto *prot,
+ int amt, int *parent_failure);
+void memcg_sock_mem_free(struct mem_cgroup *mem, struct proto *prot, int amt);
+void memcg_sockets_allocated_dec(struct mem_cgroup *mem, struct proto *prot);
+void memcg_sockets_allocated_inc(struct mem_cgroup *mem, struct proto *prot);
+/* memcontrol includes sockets.h, that includes memcontrol.h ... */
+static inline void memcg_sock_mem_alloc(struct mem_cgroup *mem,
+ struct proto *prot, int amt,
+ int *parent_failure)

In these days, at naming memory cgroup pointers, we use "memcg" instead of
"mem". So, could you use "memcg" for represeinting memory cgroup ?

+void memcg_sock_mem_alloc(struct mem_cgroup *mem, struct proto *prot,
+ int amt, int *parent_failure)
+ mem = parent_mem_cgroup(mem);
+ for (; mem != NULL; mem = parent_mem_cgroup(mem)) {
+ long alloc;
+ long *prot_mem = prot->prot_mem(mem);
+ /*
+ * Large nestings are not the common case, and stopping in the
+ * middle would be complicated enough, that we bill it all the
+ * way through the root, and if needed, unbill everything later
+ */
+ alloc = atomic_long_add_return(amt,
+ prot->memory_allocated(mem));
+ *parent_failure |= (alloc> prot_mem[2]);
+ }

Hmm. why not using res_counter ? for reusing 'unbill' code ?


Well, besides the cost, we'd have atomic_t for !cgroups, and res_counter for cgroups. I think there is value in keeping them the same.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at