From mboxrd@z Thu Jan 1 00:00:00 1970 From: Kulikov Vasiliy Date: Tue, 07 Sep 2010 19:21:55 +0000 Subject: Re: [PATCH] md: do not use ++ in rcu_dereference() argument Message-Id: <20100907192154.GA14214@albatros> List-Id: References: <1283711539-7123-1-git-send-email-segooon@gmail.com> <201009062210.56824.arnd@arndb.de> In-Reply-To: <201009062210.56824.arnd@arndb.de> MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 8bit To: Arnd Bergmann Cc: "Paul E. McKenney" , kernel-janitors@vger.kernel.org, Neil Brown , Jens Axboe , linux-raid@vger.kernel.org, linux-kernel@vger.kernel.org On Mon, Sep 06, 2010 at 22:10 +0200, Arnd Bergmann wrote: > On Sunday 05 September 2010 20:32:18 Kulikov Vasiliy wrote: > > From: Vasiliy Kulikov > > > > rcu_dereference() is macro, so it might use its argument twice. > > Argument must not has side effects. > > > > It was found by compiler warning: > > drivers/md/raid1.c: In function ‘read_balance’: > > drivers/md/raid1.c:445: warning: operation on ‘new_disk’ may be undefined > > I think the rcu_dereference macro should really not evaluate its argument > twice, and I don't see where it does. > As a general rule, we try to write macros in Linux such that they behave > like functions and don't have surprising side-effects. > > Which kernel and gcc version do you see the warning with? > > Arnd gcc version 4.4.3 (Ubuntu 4.4.3-4ubuntu5), linux-next. #define __rcu_dereference_check(p, c, space) \ ({ \ typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \ ^ rcu_lockdep_assert(c); \ (void) (((typeof (*p) space *)p) = p); \ ^ ^ smp_read_barrier_depends(); \ ((typeof(*p) __force __kernel *)(_________p1)); \ }) If I understand this, it is evaluated three times, right? -- Vasiliy