Re: [PATCH] drivers: base: update cpu offline info when do hotplug

[Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>]

(2014/10/21 12:36), Neil Zhang wrote:
>
>
>> -----Original Message-----
>> From: Yasuaki Ishimatsu [mailto:isimatu.yasuaki@jp.fujitsu.com]
>> Sent: 2014年10月21日 11:27
>> To: Neil Zhang; Dan Streetman
>> Cc: Greg KH; linux-kernel@vger.kernel.org
>> Subject: Re: [PATCH] drivers: base: update cpu offline info when do hotplug
>>
>> (2014/10/21 12:18), Neil Zhang wrote:
>>> Yasuaki,
>>>
>>>> -----Original Message-----
>>>> From: Yasuaki Ishimatsu [mailto:isimatu.yasuaki@jp.fujitsu.com]
>>>> Sent: 2014年10月21日 10:57
>>>> To: Dan Streetman; Neil Zhang
>>>> Cc: Greg KH; linux-kernel@vger.kernel.org
>>>> Subject: Re: [PATCH] drivers: base: update cpu offline info when do
>>>> hotplug
>>>>
>>>> Hi Neil and Dan,
>>>>
>>>> (2014/10/21 2:02), Dan Streetman wrote:
>>>>> On Mon, Oct 20, 2014 at 3:40 AM, Neil Zhang <zhangwm@marvell.com> wrote:
>>>>>> Greg,
>>>>>>
>>>>>>
>>>>>> -----Original Message-----
>>>>>> From: Greg KH [mailto:gregkh@linuxfoundation.org]
>>>>>> Sent: 2014年10月20日 14:48
>>>>>> To: Neil Zhang
>>>>>> Cc: linux-kernel@vger.kernel.org
>>>>>> Subject: Re: [PATCH] drivers: base: update cpu offline info when do
>>>>>> hotplug
>>>>>>
>>>>>> On Sun, Oct 19, 2014 at 11:39:23PM -0700, Neil Zhang wrote:
>>>>>>>> How much noise is this going to cause on a big/little system that
>>>>>>>> constantly hot unplug/plugs processors all of the time?
>>>>>>>
>>>>>>> Can you explain more what kind of noise will be introduced on a
>>>>>>> big/little
>>>> system?
>>>>>>
>>>>>> Have you tested this on such a machine?
>>>>>>
>>>>>> I didn't have such kind of machine on hand.
>>>>>> Can anyone has such machine to verify it?
>>>>>> Thanks!
>>>>>
>>>>> I tested this on a ppc PowerVM system, using dlpar operations to
>>>>> remove/add cpus.
>>>>>
>>>>> Without this patch the cpu online nodes get out of sync with the
>>>>> main online node (and the actual state of the cpus), because they
>>>>> aren't updated as the cpus are brought up/down:
>>>>>
>>>>
>>>>> [root@br10p02 cpu]$ pwd
>>>>> /sys/devices/system/cpu
>>>>> [root@br10p02 cpu]$ cat online
>>>>> 0-39
>>>>> [root@br10p02 cpu]$ for n in {0..47} ; do test $( cat cpu$n/online )
>>>>> -eq 1 && echo -n "$n " ; done ; echo ""
>>>>> 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
>>>>> 26
>>>>> 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47
>>>>
>>>> How is the issue reproduced.
>>>>
>>>> Here is a result on my x86 box with linux-3.18-rc1.
>>>>
>>>> - before offline CPU
>>>> # cd /sys/devices/system/cpu/
>>>> # cat online
>>>> 0-59
>>>> # for n in {0..59} ; do test $( cat cpu$n/online ) -eq 1 && echo -n
>>>> "$n " ; done ; echo ""
>>>> 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
>>>> 26 27 28
>>>> 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51
>>>> 52 53 54
>>>> 55
>>>> 56 57 58 59
>>>>
>>>> - after offline CPU{1..59}
>>>> # for n in {1..59} ; do echo 0 > cpu$n/online; done # cat online
>>>> 0
>>>> # for n in {0..59} ; do test $( cat cpu$n/online ) -eq 1 && echo -n
>>>> "$n " ; done ; echo ""
>>>> 0
>>>>
>>>> It seems that dev->offline is set to correct valute.
>>>>
>>>
>>
>>> Please use an in kernel governor to up / down a core instead of sysfs
>> interface.
>>
>> Thank you for the information. But I don't know the in kernel governor?
>> Could you point a documentation of it?
>>
>

> Simply means that you call cpu_down / cpu_up directly in kernel base on the profiler.

I understood it.
When using sysfs, device_online/offline() changes dev->offline.
Therefore, this problem did not occur in my previous test.

Thanks,
Yasuaki Ishimatsu


>
>> Thanks,
>> Yasuaki Ishimatsu
>>
>>>
>>>> Thanks,
>>>> Yasuaki Ishimatsu
>>>>
>>>>>
>>>>>
>>>>> While with the patch, the cpu online nodes are kept up to date as
>>>>> the cpus are brought up/down:
>>>>>
>>>>> [root@br10p02 cpu]$ pwd
>>>>> /sys/devices/system/cpu
>>>>> [root@br10p02 cpu]$ cat online
>>>>> 0-39
>>>>> [root@br10p02 cpu]$ for n in {0..47} ; do test $( cat cpu$n/online )
>>>>> -eq 1 && echo -n "$n " ; done ; echo ""
>>>>> 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
>>>>> 26
>>>>> 27 28 29 30 31 32 33 34 35 36 37 38 39
>>>>>
>>>>>
>>>>> Feel free to add
>>>>>
>>>>> Tested-by: Dan Streetman <ddstreet@ieee.org>
>>>>>
>>>>>>
>>>>>>> As I know IKS on arm will use cpu_suspend way to power down a core.
>>>>>>
>>>>>> Are you sure that it also doesn't use that same functionality to
>>>>>> drop a
>>>> processor to save power?
>>>>>>
>>>>>> As I know it use cpu_suspend to switch out a processor in IKS and
>>>>>> there is
>>>> no cpu hotplug notifier in this procedure.
>>>>>>
>>>>>>
>>>>>> Why do you need/want this notification?  What are you going to do
>>>>>> with this
>>>> information that you don't already have?
>>>>>>
>>>>>> The offline won't be updated if an in kernel hotplug governor plug
>>>>>> in / out
>>>> a core which cause the sysfs interface report a wrong status.
>>>>>>
>>>>>>
>>>>>> thanks,
>>>>>>
>>>>>> greg k-h
>>>>>>
>>>>>>
>>>>>> Best Regards,
>>>>>> Neil Zhang
>>>>> --
>>>>> To unsubscribe from this list: send the line "unsubscribe
>>>>> linux-kernel" in the body of a message to majordomo@vger.kernel.org
>>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>>> Please read the FAQ at  http://www.tux.org/lkml/
>>>>>
>>>>
>>>
>>>
>>> Best Regards,
>>> Neil Zhang
>>> \x04 {.n +       +%  lzwm  b 맲  r  zX  \x19\x1e w  {ay \x1dʇڙ ,j
>>>    f   h   z \x1e w
>>
>>     j:+v   w j m         zZ+     ݢj"  ! iO  z  v ^\x14\x04\x1a ^[m
>> nƊ  Y&
>>>
>>
>
>
> Best Regards,
> Neil Zhang
> \x13��칻\x1c�&�~�&�\x18��+-��ݶ\x17��w��˛���m�b��dz�ޖ)���w*\x1fjg���\x1e�����ݢj/���z�ޖ��2�ޙ���&�)ߡ�a��\x7f��\x1e�G���h�\x0f�j:+v���w�٥>W����i�axP\x10j�m����\f-�\x1a+��d�_
>