Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 26 Oct 2020 09:20:10 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     Joel Fernandes <joel@joelfernandes.org>
Cc:     Michael Jeanson <mjeanson@efficios.com>,
        linux-kernel@vger.kernel.org, mathieu.desnoyers@efficios.com,
        Steven Rostedt <rostedt@goodmis.org>,
        Alexei Starovoitov <ast@kernel.org>,
        Yonghong Song <yhs@fb.com>,
        "Paul E . McKenney" <paulmck@kernel.org>,
        Ingo Molnar <mingo@redhat.com>,
        Arnaldo Carvalho de Melo <acme@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@redhat.com>,
        Namhyung Kim <namhyung@kernel.org>, bpf@vger.kernel.org
Subject: Re: [RFC PATCH 6/6] tracing: use sched-RCU instead of SRCU for
 rcuidle tracepoints
Message-ID: <20201026082010.GC2628@hirez.programming.kicks-ass.net>
References: <20201023195352.26269-1-mjeanson@efficios.com>
 <20201023195352.26269-7-mjeanson@efficios.com>
 <20201023211359.GC3563800@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20201023211359.GC3563800@google.com>
Precedence: bulk

On Fri, Oct 23, 2020 at 05:13:59PM -0400, Joel Fernandes wrote:
> On Fri, Oct 23, 2020 at 03:53:52PM -0400, Michael Jeanson wrote:
> > From: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
> > 
> > Considering that tracer callbacks expect RCU to be watching (for
> > instance, perf uses rcu_read_lock), we need rcuidle tracepoints to issue
> > rcu_irq_{enter,exit}_irqson around calls to the callbacks. So there is
> > no point in using SRCU anymore given that rcuidle tracepoints need to
> > ensure RCU is watching. Therefore, simply use sched-RCU like normal
> > tracepoints for rcuidle tracepoints.
> 
> High level question:
> 
> IIRC, doing this increases overhead for general tracing that does not use
> perf, for 'rcuidle' tracepoints such as the preempt/irq enable/disable
> tracepoints. I remember adding SRCU because of this reason.
> 
> Can the 'rcuidle' information not be pushed down further, such that perf does
> it because it requires RCU to be watching, so that it does not effect, say,
> trace events?

There's very few trace_.*_rcuidle() users left. We should eradicate them
and remove the option. It's bugs to begin with.