[Devel] Re: [PATCH 2/2] scheduler: cgroups cpuaccouting: Make cpuusage atomic

[KAMEZAWA Hiroyuki]

On Wed, 19 May 2010 20:58:59 +0200
Thomas Renninger <trenn at suse.de> wrote:

> and avoid locking on 32 bit.
> This resolves an ugly dependency in cgroups_cpuaccount.c
> to per_cpu runqueues lock variable.
> 
> Signed-off-by: Thomas Renninger <trenn at suse.de>
> CC: linux-kernel at vger.kernel.org
> CC: mike at android.com
> CC: menage at google.com
> CC: lizf at cn.fujitsu.com
> CC: containers at lists.linux-foundation.org
> CC: mingo at elte.hu
> CC: peterz at infradead.org
> ---
>  kernel/cgroup_cpuaccount.c |   39 +++++++++------------------------------
>  kernel/sched.c             |   11 -----------
>  kernel/sched.h             |    7 -------
>  3 files changed, 9 insertions(+), 48 deletions(-)
>  delete mode 100644 kernel/sched.h
> 
> diff --git a/kernel/cgroup_cpuaccount.c b/kernel/cgroup_cpuaccount.c
> index 0ad356a..0a53487 100644
> --- a/kernel/cgroup_cpuaccount.c
> +++ b/kernel/cgroup_cpuaccount.c
> @@ -10,8 +10,6 @@
>  
>  #include <asm/cputime.h>
>  
> -#include "sched.h"
> -
>  /*
>   * CPU accounting code for task groups.
>   *
> @@ -23,7 +21,7 @@
>  struct cpuacct {
>  	struct cgroup_subsys_state css;
>  	/* cpuusage holds pointer to a u64-type object on every cpu */
> -	u64 __percpu *cpuusage;
> +	atomic64_t __percpu *cpuusage;

Hmmm...how about adding seq_counter for 32bit rather than atomic64_t ?

Thanks,
-Kame

_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers