Re: memcg: cat: memory.memsw.* : Operation not supported

[Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

(2012/06/28 5:24), Tejun Heo wrote:
> Hello,
>
> On Wed, Jun 27, 2012 at 01:21:27PM -0700, David Rientjes wrote:
>> Well it also has a prerequisite that memcg doesn't have: CONFIG_SWAP, so
>
> Right.
>
>> even if CONFIG_CGROUP_MEM_RES_CTLR_SWAP is folded into
>> CONFIG_CGROUP_MEM_RES_CTLR, then these should still depend on CONFIG_SWAP
>> since configuring them would imply there is some limit to be enforced.
>>
>> But to answer your question:
>>
>>     text	   data	    bss	    dec	    hex	filename
>>    25777	   3644	   4128	  33549	   830d	memcontrol.o.swap_disabled
>>    27294	   4476	   4128	  35898	   8c3a	memcontrol.o.swap_enabled
>
> I still wish it's folded into CONFIG_MEMCG and conditionalized just on
> CONFIG_SWAP tho.
>

In old days, memsw controller was not very stable. So, we devided the config.
And, it makes size of memory for swap-device double (adds 2bytes per swapent.)
That is the problem.

>> Is it really too painful to not create these files when
>> CONFIG_CGROUP_MEM_RES_CTLR_SWAP is disabled?  If so, can we at least allow
>> them to be opened but return -EINVAL if memory.memsw.limit_in_bytes is
>> written?
>
> Not at all, that was the first version anyway, which (IIRC) KAME
> didn't like and suggested always creating those files.  KAME, what do
> you think?
>

IIRC...at that time, we made decision, cgroup has no feature to
'create files dynamically'. Then, we made it in static, decision was done
at compile time and ignores "do_swap_account".

Now, IIUC, we have the feature. So, it's may be a time to create the file
with regard to "do_swap_account", making decision at boot time.


Thanks,
-Kame