Re: [-mm] Add an owner to the mm_struct (v5)

From: Balbir Singh
Date: Thu Apr 03 2008 - 02:43:48 EST


KAMEZAWA Hiroyuki wrote:
> just nitpicks ;)
>
> On Thu, 03 Apr 2008 11:29:01 +0530
> Balbir Singh <balbir@xxxxxxxxxxxxxxxxxx> wrote:
>
>> #define mm_match_cgroup(mm, cgroup) \
>> - ((cgroup) == rcu_dereference((mm)->mem_cgroup))
>> + ((cgroup) == mem_cgroup_from_task((mm)->owner))
>>
>
> After this patch, above should be
> ==
> #define mm_match_cgroup_css(mm, css, subsys_id)
> ((css) == task_subsys_state((mm)->owner, subsys_id)
> ==
> This generic macro may be good for your purpose.
>

When we call mm_match_cgroup_*, we don't want to dereference mem_cont->css to
get the css, hence we abstract it away. This is called from mm/rmap.c

>
>> #endif
>> diff -puN init/Kconfig~memory-controller-add-mm-owner init/Kconfig
>> --- linux-2.6.25-rc8/init/Kconfig~memory-controller-add-mm-owner 2008-04-03 10:08:23.000000000 +0530
>> +++ linux-2.6.25-rc8-balbir/init/Kconfig 2008-04-03 10:08:23.000000000 +0530
>> @@ -371,9 +371,21 @@ config RESOURCE_COUNTERS
>> infrastructure that works with cgroups
>> depends on CGROUPS
>>
>> +config MM_OWNER
>> + bool "Enable ownership of mm structure"
>> + help
>> + This option enables mm_struct's to have an owner. The advantage
>> + of this approach is that it allows for several independent memory
>> + based cgorup controllers to co-exist independently without too
>> + much space overhead
> Above is an explanation for this patch.
> More simple text is better... How about
> ==
> This is necessary for some cgroup subsystem related to memory management.
> ==

Yes, but several other developers have also asked for it. revoke*, swap
namespaces, etc will use it. I wanted to have a common definition.

>> +
>> + This feature adds fork/exit overhead. So enable this only if
>> + you need resource controllers
>> +
>
>
>> config CGROUP_MEM_RES_CTLR
>> bool "Memory Resource Controller for Control Groups"
>> depends on CGROUPS && RESOURCE_COUNTERS
>> + select MM_OWNER
>
> I don't like "select"....this should be
> depends on CGROUPS && RESOURCE_COUNTERS && MM_OWNER
>

I discussed this will Paul and I think select is better. The user might ignore
to enable MM_OWNER and wonder why memory controller or other features are not
getting enabled.

> Thanks,
> -Kame

Thanks for the review

--
Warm Regards,
Balbir Singh
Linux Technology Center
IBM, ISTL
--
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/