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:   Sun, 9 Feb 2020 18:20:08 -0800
From:   Andrew Morton <akpm@linux-foundation.org>
To:     Qian Cai <cai@lca.pw>
Cc:     Matthew Wilcox <willy@infradead.org>,
        Marco Elver <elver@google.com>, jhubbard@nvidia.com,
        ira.weiny@intel.com, dan.j.williams@intel.com, jack@suse.cz,
        linux-mm@kvack.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH -next v2] mm: mark an intentional data race in
 page_zonenum
Message-Id: <20200209182008.008c06f1cf4347a95f9de0a5@linux-foundation.org>
In-Reply-To: <95A69596-D5F9-41EB-84A0-AE32D17FE320@lca.pw>
References: <20200206183000.913-1-cai@lca.pw>
        <20200206202005.GY8731@bombadil.infradead.org>
        <95A69596-D5F9-41EB-84A0-AE32D17FE320@lca.pw>
Mime-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, 6 Feb 2020 15:25:41 -0500 Qian Cai <cai@lca.pw> wrote:

> 
> 
> > On Feb 6, 2020, at 3:20 PM, Matthew Wilcox <willy@infradead.org> wrote:
> > 
> > On Thu, Feb 06, 2020 at 01:30:00PM -0500, Qian Cai wrote:
> >> Both the read and write are done only with the non-exclusive mmap_sem
> >> held. Since the read only check for a specific bit range (up to 3 bits)
> >> in the flag but the write here never change those 3 bits, so load
> >> tearing would be harmless here. Thus, just mark it as an intentional
> >> data races using the data_race() macro which is designed for those
> >> situations [1].
> > 
> > This changelog makes me think you don't really understand the situation.
> > 
> > A page never changes its zone number.  The zone number happens to be
> > stored in the same word as other bits which are modified, but the zone
> > number bits will never be modified by any other write.  So we can accept
> > a reload of the zone bits after an intervening write and we don't need
> > to use READ_ONCE().
> 
> Maybe your explanation is better, but I did try to explain the same thing.
> I’ll let Andrew to decide if he would like to update the commit log a bit
> with your wording.

Using data_race() here seems misleading - there is no race, but we're
using data_race() to suppress a false positive warning from KCSAN, yes?

That's a bit hacky.  If this happens rarely then perhaps adding a
suitable comment in page_zonenum() which explains this will suffice. 
But if we keep on abusing data_race() in this fashion then it would be
better to add a new macro for this purpose.