[PATCH v6 0/5] MCPM backend for Exynos5420

[Nicolas Pitre]

On Wed, 14 May 2014, Abhilash Kesavan wrote:

> Hi Nicolas,
> 
> On Tue, May 13, 2014 at 11:25 PM, Nicolas Pitre
> <nicolas.pitre at linaro.org> wrote:
> > On Tue, 13 May 2014, Abhilash Kesavan wrote:
> >
> >> This is v6 of the series adding MCPM backend support for SMP secondary boot
> >> and core switching on Samsung's Exynos5420. The patches are based on the mcpm
> >> support added for Exynos5420 in the Chromium kernel repository here:
> >> https://chromium.googlesource.com/chromiumos/third_party/kernel-next/+/chromeos-3.8
> >>
> >> The patches have been prepared on Kukjin Kim's for-next branch and tested on
> >> SMDK5420 EVT1 as well as an exynos5420 based chromebook (peach-pit) using the
> >> "/dev/b.L_switcher" user interface. Secondary core boot-up has also been tested
> >> on both the boards.
> >
> > OK... Now it is time for real testing.  :-)
> >
> > The /dev/b.L_switcher interface tests the switcher.  Here you really
> > want to hammer the MCPM functionalities and especially your backend code
> > as hard as possible.  I therefore recommend the following test script:
> >
> > ---------- >8
> > #!/bin/bash
> >
> > echo 0 >/sys/kernel/bL_switcher/active
> > sleep 1
> >
> > pids=
> > for cpu in /sys/devices/system/cpu/cpu?/online; do
> >   { cpu_nr=${cpu:27:1}
> >     while true; do
> >       echo 1 > $cpu 2> /dev/null
> >       sleep .00$RANDOM
> >       val1=$(cat $cpu)
> >       echo 0 > $cpu 2> /dev/null
> >       sleep .00$RANDOM
> >       val0=$(cat $cpu)
> >       [ "$val1" = "1" -a "$val0" = "0" ] && echo -n $cpu_nr
> >     done
> >   } &
> >   pids="$pids $!"
> > done
> >
> > trap "kill $pids; echo" 0 15
> > wait $pids
> > ---------- >8
> >
> > Leave this running for a couple hours making sure you see all CPU
> > numbers being printed.  The printing order will be random, but each CPU
> > number should continuously appear.
> I tried this script and I get two errors:
> 
> 1) can't create /sys/devices/system/cpu/cpu//online: nonexistent directory

What do you get if you do:

$ ls -d /sys/devices/system/cpu/cpu?/online

?

> 2) sleep: invalid number '.0026736'

A sufficiently recent coreutils package should have a sleep command that 
accepts fractional values.  Alternatively you may replace it with 
usleep:

	usleep $RANDOM

> For 1) the cpu number is not being appended. if I give a particular
> cpu in the script then hotplug in/out works fine.
> For 2) a constant msleep 10 works.

If you have msleep but not usleep then try:

		msleep $(($RANDOM / 1000))

> Is it OK for me to modify the script to hotplug in/off a randomly chosed core ?

No.  They must *all* be hotplugged simultaneously with some random 
timing.


Nicolas