Received-SPF: pass (google.com: domain of linux-kernel+bounces-195082-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Message-ID: <a9f559ac6da9e44cdcffe1cb80a4468096efb1f8.camel@codeconstruct.com.au>
Subject: Re: [PATCH] mctp i2c: Add rx trace
From: Jeremy Kerr <jk@codeconstruct.com.au>
To: Tal Yacobi <talycb8@gmail.com>
Cc: matt@codeconstruct.com.au, rostedt@goodmis.org, mhiramat@kernel.org, 
	mathieu.desnoyers@efficios.com, netdev@vger.kernel.org, 
	linux-kernel@vger.kernel.org, linux-trace-kernel@vger.kernel.org
Date: Thu, 30 May 2024 16:42:12 +0800
In-Reply-To: <CAN7-cG5w9rjhJnDJJSOTiopxkj75mOJEC9eKZqmJaqbeMudPQQ@mail.gmail.com>
References: <20240528143420.742611-1-talycb8@gmail.com>
	 <520cf8db945cf8dce4afdaddb59ceda65463a406.camel@codeconstruct.com.au>
	 <CAN7-cG5w9rjhJnDJJSOTiopxkj75mOJEC9eKZqmJaqbeMudPQQ@mail.gmail.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.4-2 
Precedence: bulk
MIME-Version: 1.0

Hi Tal,

> > > mctp-i2c rx implementation doesn't call
> > > __i2c_transfer which calls the i2c reply trace function.
> >=20
> > No, but we can trace the i2c rx path through the trace_i2c_slave
> > tracepoint. It is a little messier than tracing trace_i2c_write,
> > but
> > has been sufficient with the debugging I've needed in the past.
>=20
> Oh, I missed that.
> I had to test it with an older kernel without i2c_slave tracing
> so I looked only at the regular i2c and mctp trace paths.

OK! That tracepoint was (coincidentally) added in 5.18, same as the
MCTP-over-i2c transport. So we should have coverage for both features
on upstream kernels, at least.

> > > Add an mctp_reply trace function that will be used instead.
> >=20
> > Can you elaborate a little on what you were/are looking to inspect
> > here? (mainly: which packet fields are you interested in?) That
> > will
> > help to determine the best approach here.
>=20
> Sure, I basically wanted to trace the i2c packet buffer in a simple
> way.

OK - did you specifically need the i2c transport headers? Since the
MCTP interfaces are regular net devices, the easiest way to trace
generic MCTP transfers is generally via a packet capture (tcpdump,
wireshark, etc).

Cheers,


Jeremy