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

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

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.

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/
>