Date: Tue, 24 Nov 2015 21:06:14 -0200
From: Marcelo Tosatti <mtosatti@redhat.com>
To: Chao Peng <chao.p.peng@linux.intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>, LKML <linux-kernel@vger.kernel.org>,
        Peter Zijlstra <peterz@infradead.org>, x86@kernel.org,
        Luiz Capitulino <lcapitulino@redhat.com>,
        Vikas Shivappa <vikas.shivappa@intel.com>, Tejun Heo <tj@kernel.org>,
        Yu Fenghua <fenghua.yu@intel.com>
Subject: Re: [RFD] CAT user space interface revisited
Message-ID: <20151124230613.GA16984@amt.cnet>
References: <alpine.DEB.2.11.1511181534450.3761@nanos>
 <20151124073124.GA17000@pengc-linux.bj.intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20151124073124.GA17000@pengc-linux.bj.intel.com>
User-Agent: Mutt/1.5.21 (2010-09-15)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2650
Lines: 69

On Tue, Nov 24, 2015 at 03:31:24PM +0800, Chao Peng wrote:
> On Wed, Nov 18, 2015 at 07:25:03PM +0100, Thomas Gleixner wrote:
> > 
> > Let's look at partitioning itself. We have two options:
> > 
> >    1) Per task partitioning
> > 
> >    2) Per CPU partitioning
> > 
> > So far we only talked about #1, but I think that #2 has a value as
> > well. Let me give you a simple example.
> 
> I would second this. In practice per CPU partitioning is useful for
> realtime as well. And I can see three possible solutions:
> 
>      1) What you suggested below, to address both problems in one
>         framework. But I wonder if it would end with too complex.
> 
>      2) Achieve per CPU partitioning with per task partitioning. For
>         example, if current CAT patch can solve the kernel threads
> 	problem, together with CPU pinning, we then can set a same CBM
> 	for all the tasks/kernel threads run on an isolated CPU. 

As for the kernel threads problem, it seems its a silly limitation of
the code which handles writes to cgroups:

diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index f89d929..0603652 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -2466,16 +2466,6 @@ static ssize_t __cgroup_procs_write(struct
kernfs_open_file *of, char *buf,
        if (threadgroup)
                tsk = tsk->group_leader;

-       /*
-        * Workqueue threads may acquire PF_NO_SETAFFINITY and become
-        * trapped in a cpuset, or RT worker may be born in a cgroup
-        * with no rt_runtime allocated.  Just say no.
-        */
-       if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
-               ret = -EINVAL;
-               goto out_unlock_rcu;
-       }
-
        get_task_struct(tsk);
        rcu_read_unlock();

For a cgroup hierarchy with no cpusets (such as CAT only) this
limitation makes no sense (looking for a place where to move this to).

Any ETA on per-socket bitmasks? 

> 
>      3) I wonder if it feasible to separate the two requirements? For
>         example, divides the work into three components: rdt-base,
> 	per task interface (current cgroup interface/IOCTL or something)
> 	and per CPU interface. The two interfaces are exclusive and
> 	selected at build time. One thing to reject this option would be
> 	even with per CPU partitioning, we still need per task partitioning,
> 	in that case we will go to option 1) again.
> 
> Thanks,
> Chao
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/