From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751158AbbEEAlI (ORCPT ); Mon, 4 May 2015 20:41:08 -0400 Received: from mail-pa0-f50.google.com ([209.85.220.50]:33810 "EHLO mail-pa0-f50.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751264AbbEEAk7 (ORCPT ); Mon, 4 May 2015 20:40:59 -0400 Message-ID: <55481197.7070905@m4x.org> Date: Tue, 05 May 2015 08:40:55 +0800 From: Nicolas Iooss User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.6.0 MIME-Version: 1.0 To: paulmck@linux.vnet.ibm.com, Josh Triplett CC: Steven Rostedt , Mathieu Desnoyers , Lai Jiangshan , linux-kernel@vger.kernel.org Subject: Re: [PATCH] rcu: declare rcu_data variables in the section they are defined in References: <1430647073-19753-1-git-send-email-nicolas.iooss_linux@m4x.org> <20150503192701.GB2206@x> <20150504203325.GE5381@linux.vnet.ibm.com> In-Reply-To: <20150504203325.GE5381@linux.vnet.ibm.com> Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 05/05/2015 04:33 AM, Paul E. McKenney wrote: > On Sun, May 03, 2015 at 12:27:02PM -0700, Josh Triplett wrote: >> On Sun, May 03, 2015 at 05:57:53PM +0800, Nicolas Iooss wrote: >>> Commit 11bbb235c26f ("rcu: Use DEFINE_PER_CPU_SHARED_ALIGNED for >>> rcu_data") replaced DEFINE_PER_CPU by DEFINE_PER_CPU_SHARED_ALIGNED in >>> the definition of rcu_sched and rcu_bh without updating >>> kernel/rcu/tree.h. >>> >>> This makes clang report a section mismatch (-Wsection warning) when >>> building LLVMLinux because the variables are declared in .data..percpu >>> but defined in .data..percpu..shared_aligned. >>> >>> Signed-off-by: Nicolas Iooss >> >> Good catch. >> Reviewed-by: Josh Triplett > > Agreed, good catch! But don't we also need to worry about > rcu_preempt_data? Yes, I missed it because I didn't know that allyesconfig/allmodconfig does not select CONFIG_PREEMPT (it selects CONFIG_PREEMPT_NONE instead). > Also, given that tree_trace.c now uses iterators > rather than direct access via the per-CPU variables, wouldn't the > following be more appropriate? (-Very- lightly tested.) This doesn't work with CONFIG_TREE_RCU_TRACE, because kernel/rcu/tree_trace.c uses rcu_preempt_state (in v4.1-rc2). I've successfully built an allmodconfig+CONFIG_PREEMPT kernel for x86_64 with the following patch (I have not tested the result). Anyway, thanks for your quick replies. Nicolas --- a/kernel/rcu/tree.c +++ b/kernel/rcu/tree.c @@ -91,7 +91,7 @@ static const char *tp_##sname##_varname __used __tracepoint_string = sname##_var #define RCU_STATE_INITIALIZER(sname, sabbr, cr) \ DEFINE_RCU_TPS(sname) \ -DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, sname##_data); \ +static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, sname##_data); \ struct rcu_state sname##_state = { \ .level = { &sname##_state.node[0] }, \ .rda = &sname##_data, \ --- a/kernel/rcu/tree.h +++ b/kernel/rcu/tree.h @@ -519,14 +519,11 @@ extern struct list_head rcu_struct_flavors; * RCU implementation internal declarations: */ extern struct rcu_state rcu_sched_state; -DECLARE_PER_CPU(struct rcu_data, rcu_sched_data); extern struct rcu_state rcu_bh_state; -DECLARE_PER_CPU(struct rcu_data, rcu_bh_data); #ifdef CONFIG_PREEMPT_RCU extern struct rcu_state rcu_preempt_state; -DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data); #endif /* #ifdef CONFIG_PREEMPT_RCU */ #ifdef CONFIG_RCU_BOOST