[Devel] Re: [PATCH v5 00/10] per-cgroup tcp memory pressure

KAMEZAWA Hiroyuki kamezawa.hiroyu at jp.fujitsu.com
Mon Nov 21 18:07:07 PST 2011


On Fri, 18 Nov 2011 17:39:03 -0200
Glauber Costa <glommer at parallels.com> wrote:

> On 11/17/2011 07:35 PM, David Miller wrote:
> > TCP specific stuff in mm/memcontrol.c, at best that's not nice at all.
> 
> How crucial is that? Thing is that as far as I am concerned, all the 
> memcg people really want the inner layout of struct mem_cgroup to be 
> private to memcontrol.c 

This is just because memcg is just related to memory management and I don't
want it be wide spreaded, 'struct mem_cgroup' has been changed often.

But I don't like to have TCP code in memcgroup.c.

New idea is welcome.

> This means that at some point, we need to have
> at least a wrapper in memcontrol.c that is able to calculate the offset
> of the tcp structure, and since most functions are actually quite 
> simple, that would just make us do more function calls.
> 
> Well, an alternative to that would be to use a void pointer in the newly 
> added struct cg_proto to an already parsed memcg-related field
> (in this case tcp_memcontrol), that would be passed to the functions
> instead of the whole memcg structure. Do you think this would be 
> preferable ?
> 
like this ?

struct mem_cgroup_sub_controls {
	struct mem_cgroup *mem;
	union {
		struct tcp_mem_control tcp;
	} data;
};
/* for loosely coupled controls for memcg */
struct memcg_sub_controls_function
{
	struct memcg_sub_controls	(*create)(struct mem_cgroup *);
	struct memcg_sub_controls	(*destroy)(struct mem_cgroup *);
}

int register_memcg_sub_controls(char *name, 
		struct memcg_sub_controls_function *abis);


struct mem_cgroup {
	.....
	.....
	/* Root memcg will have no sub_controls! */
	struct memcg_sub_controls	*sub_controls[NR_MEMCG_SUB_CONTROLS];
}


Maybe some functions should be exported. 

Thanks,
-Kame





More information about the Devel mailing list