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;
MIME-Version: 1.0
From:   Anatol Pomozov <anatol.pomozov@gmail.com>
Date:   Tue, 8 Jan 2019 11:33:39 -0800
Message-ID: <CAOMFOmU=t9YbE68YpoQn0Ug0XN1F+0iuJp9J=htCRRrbdaPSVA@mail.gmail.com>
Subject: seqcount usage in xt_replace_table()
To:     fw@strlen.de, Dmitry Vyukov <dvyukov@google.com>,
        paulmck@linux.ibm.com, LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hello folks,

A bit of context what I am doing. I am trying to port KTSAN (Kernel
Thread Sanitizer) tool to v4.20. That tool tracks shared data usage
and makes sure it is accessed in a thread-safe manner.

seqlock is a synchronization primitive used by Linux kernel. KTSAN
annotates read_seqbegin()/read_seqretry() and tracks what data been
accessed in its critical section.

During KTSAN port I found and interesting seqcount usage introduced in
commit 80055dab5de0c8677bc148c4717ddfc753a9148e

If I read this commit correctly xt_replace_table() does not use
seqlock in a canonical way to specify a critical section. Instead the
code reads the counter and waits until it gets to a specific value.

Now I want KTSAN to play with this code nicely. I need to tell KTSAN
something like "this raw_read_seqcount() does not start a critical
section, just ignore it". So temporary I introduced
raw_read_seqcount_nocritical() function that is ignored by KTSAN. Is
it a good solution?

Or maybe xt_replace_table() can be enhanced? When I hear that
something waits until an event happens on all CPUs I think about
wait_event() function. Would it be better for xt_replace_table() to
introduce an atomic counter that is decremented by CPUs, and the main
CPU waits until the counter gets zero?

WDYT?