DMARC-Filter: OpenDMARC Filter v1.3.2 mx1.redhat.com 67785C03BD7C
Message-ID: <1490793272.28917.5.camel@redhat.com>
Subject: Re: [BUG nohz]: wrong user and system time accounting
From: Rik van Riel <riel@redhat.com>
To: Frederic Weisbecker <fweisbec@gmail.com>,
        Luiz Capitulino <lcapitulino@redhat.com>
Cc: linux-kernel@vger.kernel.org, linux-rt-users@vger.kernel.org,
        Wanpeng Li <kernellwp@gmail.com>
Date: Wed, 29 Mar 2017 09:14:32 -0400
In-Reply-To: <20170329130447.GB8306@lerouge>
References: <20170323165512.60945ac6@redhat.com>
         <20170329130447.GB8306@lerouge>
Organization: Red Hat, Inc.
Content-Type: text/plain; charset="UTF-8"
Mime-Version: 1.0
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1381
Lines: 41

On Wed, 2017-03-29 at 15:04 +0200, Frederic Weisbecker wrote:
> On Thu, Mar 23, 2017 at 04:55:12PM -0400, Luiz Capitulino wrote:
> > 
> > When there are two or more tasks executing in user-space and
> > taking 100% of a nohz_full CPU, top reports 70% system time
> > and 30% user time utilization. Sometimes I'm even able to get
> > 100% system time and 0% user time.
> > 
> > This was reproduced with latest Linus tree (093b995), but I
> > don't believe it's a regression (at least not a recent one)
> > as I can reproduce it with older kernels. Also, I have
> > CONFIG_IRQ_TIME_ACCOUNTING=y and haven't tried to reproduce
> > without it yet.
> > 
> > Below you'll find the steps to reproduce and some initial
> > analysis.
> > 
> > Steps to reproduce
> > ------------------
> > 
> > 1. Set up a CPU for nohz_full with isolcpus= nohz_full=
> > 
> > 2. Pin two tasks that hog the CPU 100% of the time to that CPU
> 
> I failed to reproduce with your config. I'm still getting 99%
> userspace
> cputime. So I'm wondering if the hogging style plays a role.
> 
> I run pure user loops:
> 
>     int main(int argc, char **argv)
>     {
>         for (;;);
>         return 0
>     }
> 
> Does your user program perform syscalls or IOs of some sort?

Luiz's program makes a syscall every millisecond,
if started with the arguments he gave as his
reproducer.