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

[Yasuaki Ishimatsu]

(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?

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
> ï{.nï+ïïïïïïï+%ïïlzwmïïbïëïïrïïzXïïïwïï{ayïÊÚï,jïïfïïïhïïïzïïwïïïïïïj:+vïïïwïjïmïïïïïïïïzZ+ïïïïïÝj"ïï!ïiOïïzïïvï^ïmïïïïnÆïïY&ï


--
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/