[RFC 2/4] memcg: high-low watermark

[Li Zefan]

KAMEZAWA Hiroyuki wrote:
> Add high/low watermarks to res_counter.
> *This patch itself has no behavior changes to memory resource controller.
> 
> Changelog: very old one -> this one (v1)
>  - watarmark_state is removed and all state check is done under lock.
>  - changed res_counter_charge() interface. The only user is memory
>    resource controller. Anyway, returning -ENOMEM here is a bit starnge.
>  - Added watermark enable/disable flag for someone don't want watermarks.
>  - Restarted against 2.6.25-mm1.
>  - some subsystem which doesn't want high-low watermark can work withou it.
> 
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu at jp.fujitsu.com>
> From: YAMAMOTO Takashi <yamamoto at valinux.co.jp>
> 
> ---
>  include/linux/res_counter.h |   41 ++++++++++++++++++++++++---
>  kernel/res_counter.c        |   66 ++++++++++++++++++++++++++++++++++++++++----
>  mm/memcontrol.c             |    2 -
>  3 files changed, 99 insertions(+), 10 deletions(-)
> 
> Index: mm-2.6.26-rc2-mm1/include/linux/res_counter.h
> ===================================================================
> --- mm-2.6.26-rc2-mm1.orig/include/linux/res_counter.h
> +++ mm-2.6.26-rc2-mm1/include/linux/res_counter.h
> @@ -16,6 +16,16 @@
>  #include <linux/cgroup.h>
>  
>  /*
> + * status of resource coutner's usage.
> + */
> +enum res_state {
> +	RES_BELOW_LOW,	/* usage < lwmark */

It seems it's 'usage <= lwmark'

> +	RES_BELOW_HIGH,	/* lwmark < usage < hwmark */

and 'lwmark < usage <= hwmark'

> +	RES_BELOW_LIMIT,	/* hwmark < usage < limit. */

and 'hwmark < usage <= limit'