Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 20 Jun 2018 09:30:11 -0700
From:   "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
To:     Steven Rostedt <rostedt@goodmis.org>
Cc:     Byungchul Park <byungchul.park@lge.com>, jiangshanlai@gmail.com,
        josh@joshtriplett.org, mathieu.desnoyers@efficios.com,
        linux-kernel@vger.kernel.org, kernel-team@lge.com,
        joel@joelfernandes.org
Subject: Re: [RFC 1/2] rcu: Do prepare and cleanup idle depending on in_nmi()
Reply-To: paulmck@linux.vnet.ibm.com
References: <1529484440-20634-1-git-send-email-byungchul.park@lge.com>
 <20180620145058.GP3593@linux.vnet.ibm.com>
 <20180620114335.7a314642@gandalf.local.home>
 <20180620155658.GU3593@linux.vnet.ibm.com>
 <20180620121156.4da98db2@gandalf.local.home>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20180620121156.4da98db2@gandalf.local.home>
User-Agent: Mutt/1.5.21 (2010-09-15)
Message-Id: <20180620163011.GV3593@linux.vnet.ibm.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, Jun 20, 2018 at 12:11:56PM -0400, Steven Rostedt wrote:
> On Wed, 20 Jun 2018 08:56:58 -0700
> "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> wrote:
> 
> > OK, so in theory this change is safe from a tracing perspective.  But
> > it does add conditionals to a fastpath.
> 
> Does it?
> 
> I see it replacing two conditions from both rcu_irq_enter/exit() with a
> single one in rcu_nmi_enter/exit(). Sure it adds one to rcu_nmi_enter()
> but that's a far less fast path than rcu_irq_enter(), which this patch
> removes a conditional from.

Fair point.  But I am still a bit nervous about the in_nmi().  That
could be avoided by an argument to rcu_nmi_enter() and rcu_nmi_exit()
(or common-code functions derived from these), and the usual inlining
should eliminate both the argument and the check from generated code.

However, it is also the case that the original invokes
rcu_dynticks_task_exit() before the rcu_dynticks_eqs_exit() and
rcu_cleanup_after_idle() afterwards, and the new code executes them
both afterwards.  Why is this transformation safe?

							Thanx, Paul