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

[Neil Zhang]

> -----Original Message-----
> From: Yasuaki Ishimatsu [mailto:isimatu.yasuaki@xxxxxxxxxxxxxx]
> Sent: 2014å10æ21æ 11:27
> To: Neil Zhang; Dan Streetman
> Cc: Greg KH; linux-kernel@xxxxxxxxxxxxxxx
> 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@xxxxxxxxxxxxxx]
> >> Sent: 2014å10æ21æ 10:57
> >> To: Dan Streetman; Neil Zhang
> >> Cc: Greg KH; linux-kernel@xxxxxxxxxxxxxxx
> >> 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@xxxxxxxxxxx> wrote:
> >>>> Greg,
> >>>>
> >>>>
> >>>> -----Original Message-----
> >>>> From: Greg KH [mailto:gregkh@xxxxxxxxxxxxxxxxxxx]
> >>>> Sent: 2014å10æ20æ 14:48
> >>>> To: Neil Zhang
> >>>> Cc: linux-kernel@xxxxxxxxxxxxxxx
> >>>> 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.

> 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@xxxxxxxx>
> >>>
> >>>>
> >>>>> 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@xxxxxxxxxxxxxxx
> >>> 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
> >  {.n +       +%  lzwm  b ë  r  zX   w  {ay ÊÚ ,j
> >   f   h   z  w
> 
>    j:+v   w j m         zZ+     Ýj"  ! iO  z  v ^ m
> nÆ  Y&
> >
> 


Best Regards,
Neil Zhang