Configuring cgroups

[DK]

Hi,

I have trouble configuring cgroups, please help! When I issue the
following command I get the error "mount: Device or resource busy"

# mount -t cgroup -ocpu none /dev/cpuctl

These are the configurations options that I have enabled.

CONFIG_GROUP_SCHED=y
CONFIG_FAIR_GROUP_SCHED=y
CONFIG_RT_GROUP_SCHED=y
# CONFIG_USER_SCHED is not set
CONFIG_CGROUP_SCHED=y
CONFIG_CGROUPS=y
CONFIG_CGROUP_DEBUG=y
CONFIG_CGROUP_NS=y
CONFIG_CGROUP_FREEZER=y
CONFIG_CGROUP_DEVICE=y
CONFIG_CGROUP_CPUACCT=y
CONFIG_RESOURCE_COUNTERS=y
CONFIG_CGROUP_MEM_RES_CTLR=y
CONFIG_CGROUP_MEM_RES_CTLR_SWAP=y
CONFIG_MM_OWNER=y

Why is mount saying device busy?

note: I was following the steps in the documentation(sched-design-
CFS.txt)

Regards,
Kanishka

[Kanishka Ariyapala]

after some googling came to know that android framework is it self using cgroups. Is it the reason for the device or resource busy error?
Dose it mean that I won't be able to use cgroups?

--
unsubscribe: android-kerne...@googlegroups.com
website: http://groups.google.com/group/android-kernel

[Ken Chen]

Currently, Android dalvik VM leverages cpu cgroup to determine whether to use kernel scheduler policies, /dev/cpuctl is mounted in init.rc by default. You can create a sub-cpu cgroup under /dev/cpuctl, and control the priorization of processes there as long as your process is granted with CAP_SYS_NICE capability or running as root.

Ken

--

-- Ken Chen

[Kanishka Ariyapala]

Thank you Ken for the reply. I created a cgroup under /dev/cpuctl and added a process to the tasks file. using echo PID > tasks. however it wont get registered there, of course the PID changes every time when the application is re-started. In my case I was testing with browser and the music player. Once you exit and come back the PID is gone.

Since NAME of the app is constant(ex - com.android.browser) I will have to get the PID from the task_struct I guess, and assign the PID. How ever what is the mechanism to make a process change between two cgroups? In a programmatical manner with out issuing echo PID. (can it be hard coded in the source..)

I don't know if this is a stupid Q anyways, normal applications have the user name as something like app_XX is there any way of changing that to a root? through capabilities may be?
CAP_SYS_NICE is defined in capability.h how do I know if it is granted to my process? if not granting it. (in the kernel space as well as inside the app)

[Ken Chen]

On Wed, May 30, 2012 at 3:56 AM, Kanishka Ariyapala <kanis...@gmail.com> wrote:

Thank you Ken for the reply. I created a cgroup under /dev/cpuctl and added a process to the tasks file. using echo PID > tasks. however it wont get registered there, of course the PID changes every time when the application is re-started. In my case I was testing with browser and the music player. Once you exit and come back the PID is gone.

<Ken>That's expected behavior of cgroup.

Since NAME of the app is constant(ex - com.android.browser) I will have to get the PID from the task_struct I guess, and assign the PID. How ever what is the mechanism to make a process change between two cgroups? In a programmatical manner with out issuing echo PID. (can it be hard coded in the source..)

<Ken> Simply echo PID to new cgroup, it will be moved out from the previous one automatically.

 

I don't know if this is a stupid Q anyways, normal applications have the user name as something like app_XX is there any way of changing that to a root? through capabilities may be?
CAP_SYS_NICE is defined in capability.h how do I know if it is granted to my process? if not granting it. (in the kernel space as well as inside the app)

<Ken> You can not changing app_xx to root, but root can grant CAP_SYS_NICE to app_xx.  

--

-- Ken Chen

[San Mehat]

Make sure you have http://git.android-x86.org/?p=kernel/common.git;a=commit;h=5038d42d2e60c855cf5b097197d95adaa1269d32

The code-change also should make it somewhat clearer on how the checks end up happening.

Cheers,

-san

San Mehat | Staff Software Engineer | s...@google.com | 415-366-6172

[Kanishka Ariyapala]

Thank you Ken & San for the replies. I want to make a process move between two cgroups depending on a value from a sensor(GPS) So I need to automate it, with out user issuing echo. How to get around this? kernel module or hard coding it in sched.c or cgroup.c..

[DK]

On May 31, 10:47 am, Ken Chen <ken.chen....@gmail.com> wrote:
> On Wed, May 30, 2012 at 3:56 AM, Kanishka Ariyapala

> <kanishka...@gmail.com>wrote:

>
> > Thank you Ken for the reply. I created a cgroup under /dev/cpuctl and
> > added a process to the tasks file. using echo PID > tasks. however it wont
> > get registered there, of course the PID changes every time when the
> > application is re-started. In my case I was testing with browser and the
> > music player. Once you exit and come back the PID is gone.
>
> > <Ken>That's expected behavior of cgroup.
>
> Since NAME of the app is constant(ex - com.android.browser) I will have to> get the PID from the task_struct I guess, and assign the PID. How ever what
> > is the mechanism to make a process change between two cgroups? In a
> > programmatical manner with out issuing echo PID. (can it be hard coded in
> > the source..)
>
> > <Ken> Simply echo PID to new cgroup, it will be moved out from the
>
> previous one automatically.
>
> > I don't know if this is a stupid Q anyways, normal applications have the
> > user name as something like app_XX is there any way of changing that to a
> > root? through capabilities may be?
> > CAP_SYS_NICE is defined in capability.h how do I know if it is granted to
> > my process? if not granting it. (in the kernel space as well as inside the
> > app)
>
> <Ken> You can not changing app_xx to root, but root can grant CAP_SYS_NICE
> to app_xx.
>

> <DK> How can this be done? in the kernel source or in side the application

>
>
>
> > On Tue, May 29, 2012 at 4:17 PM, Ken Chen <ken.chen....@gmail.com> wrote:
>
> >> Currently, Android dalvik VM leverages cpu cgroup to determine whether to
> >> use kernel scheduler policies, /dev/cpuctl is mounted in init.rc by
> >> default. You can create a sub-cpu cgroup under /dev/cpuctl, and control the
> >> priorization of processes there as long as your process is granted with
> >> CAP_SYS_NICE capability or running as root.
>
> >> Ken
>
> >> On Mon, May 28, 2012 at 6:55 AM, Kanishka Ariyapala <
> >> kanishka...@gmail.com> wrote:
>
> >>> after some googling came to know that android framework is it self using
> >>> cgroups. Is it the reason for the device or resource busy error?
> >>> Dose it mean that I won't be able to use cgroups?
>