Re: [PATCH v2 3/9] rcu,tracing: Create trace_rcu_{enter,exit}()

From: Steven Rostedt <rostedt@goodmis.org>
To: Joel Fernandes <joel@joelfernandes.org>
Cc: "Paul E. McKenney" <paulmck@kernel.org>,
	Peter Zijlstra <peterz@infradead.org>,
	linux-kernel@vger.kernel.org, linux-arch@vger.kernel.org,
	mingo@kernel.org, gregkh@linuxfoundation.org,
	gustavo@embeddedor.com, tglx@linutronix.de,
	josh@joshtriplett.org, mathieu.desnoyers@efficios.com,
	jiangshanlai@gmail.com, Masami Hiramatsu <mhiramat@kernel.org>
Subject: Re: [PATCH v2 3/9] rcu,tracing: Create trace_rcu_{enter,exit}()
Date: Thu, 13 Feb 2020 16:38:25 -0500	[thread overview]
Message-ID: <20200213163800.5c51a5f1@gandalf.local.home> (raw)
In-Reply-To: <20200213211930.GG170680@google.com>

[ Added Masami ]

On Thu, 13 Feb 2020 16:19:30 -0500
Joel Fernandes <joel@joelfernandes.org> wrote:

> On Thu, Feb 13, 2020 at 12:54:42PM -0800, Paul E. McKenney wrote:
> > On Thu, Feb 13, 2020 at 03:44:44PM -0500, Joel Fernandes wrote:  
> > > On Thu, Feb 13, 2020 at 10:56:12AM -0800, Paul E. McKenney wrote:
> > > [...]   
> > > > > > It might well be that I could make these functions be NMI-safe, but
> > > > > > rcu_prepare_for_idle() in particular would be a bit ugly at best.
> > > > > > So, before looking into that, I have a question.  Given these proposed
> > > > > > changes, will rcu_nmi_exit_common() and rcu_nmi_enter_common() be able
> > > > > > to just use in_nmi()?  
> > > > > 
> > > > > That _should_ already be the case today. That is, if we end up in a
> > > > > tracer and in_nmi() is unreliable we're already screwed anyway.  
> > > > 
> > > > So something like this, then?  This is untested, probably doesn't even
> > > > build, and could use some careful review from both Peter and Steve,
> > > > at least.  As in the below is the second version of the patch, the first
> > > > having been missing a couple of important "!" characters.  
> > > 
> > > I removed the static from rcu_nmi_enter()/exit() as it is called from
> > > outside, that makes it build now. Updated below is Paul's diff. I also added
> > > NOKPROBE_SYMBOL() to rcu_nmi_exit() to match rcu_nmi_enter() since it seemed
> > > asymmetric.  
> > 
> > My compiler complained about the static and the __always_inline, so I
> > fixed those.  But please help me out on adding the NOKPROBE_SYMBOL()
> > to rcu_nmi_exit().  What bad thing happens if we leave this on only
> > rcu_nmi_enter()?  
> 
> It seemed odd to me we were not allowing kprobe on the rcu_nmi_enter() but
> allowing it on exit (from a code reading standpoint) so my reaction was to
> add it to both, but we could probably keep that as a separate
> patch/discussion since it is slightly unrelated to the patch.. Sorry to
> confuse the topic.
>

rcu_nmi_enter() was marked NOKPROBE or other reasons. See commit
c13324a505c77 ("x86/kprobes: Prohibit probing on functions before
kprobe_int3_handler()")

The issue was that we must not allow anything in do_int3() call kprobe
code before kprobe_int3_handler() is called. Because ist_enter() (in
do_int3()) calls rcu_nmi_enter() it had to be marked NOKPROBE. It had
nothing to do with it being RCU nor NMI, but because it was simply
called in do_int3().

Thus, there's no reason to make rcu_nmi_exit() NOKPROBE. But a commont
to why rcu_nmi_enter() would probably be useful, like below:

-- Steve

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 1694a6b57ad8..e2c9e3e2f480 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -846,6 +846,12 @@ void rcu_nmi_enter(void)
 {
 	rcu_nmi_enter_common(false);
 }
+/*
+ * On x86, All functions in do_int3() must be marked NOKPROBE before
+ * kprobe_int3_handler() is called. ist_enter() which is called in do_int3()
+ * before kprobe_int3_handle() happens to call rcu_nmi_enter() in which case
+ * rcu_nmi_enter() must be marked NOKRPOBE.
+ */
 NOKPROBE_SYMBOL(rcu_nmi_enter);
 
 /**
