From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756752Ab3BURtJ (ORCPT ); Thu, 21 Feb 2013 12:49:09 -0500 Received: from relay2.sgi.com ([192.48.179.30]:39573 "EHLO relay.sgi.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1754598Ab3BURtG (ORCPT ); Thu, 21 Feb 2013 12:49:06 -0500 Message-ID: <51265E0F.6090209@sgi.com> Date: Thu, 21 Feb 2013 11:49:03 -0600 From: Nathan Zimmer User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/20130106 Thunderbird/17.0.2 MIME-Version: 1.0 To: Viresh Kumar CC: , , , Subject: Re: [PATCH v3 2/2] cpufreq: Convert the cpufreq_driver_lock to use the rcu References: <1360116290-23849-1-git-send-email-nzimmer@sgi.com> <1361404583-5557-1-git-send-email-nzimmer@sgi.com> <1361404583-5557-3-git-send-email-nzimmer@sgi.com> In-Reply-To: Content-Type: text/plain; charset="ISO-8859-1"; format=flowed Content-Transfer-Encoding: 7bit X-Originating-IP: [128.162.233.180] Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 02/20/2013 11:50 PM, Viresh Kumar wrote: > On 21 February 2013 05:26, Nathan Zimmer wrote: >> In general rwlocks are discourged so we are moving it to use the rcu instead. >> This does require a bit of care since the cpufreq_driver_lock protects both >> the cpufreq_driver and the cpufreq_cpu_data array. >> Also since many of the function pointers on cpufreq_driver may sleep when >> called we have to grab them under the rcu_read_lock but call them after >> rcu_read_unlock(); > Even i have started reading rcu documentation now :) > >> Cc: Viresh Kumar >> Cc: "Rafael J. Wysocki" >> Signed-off-by: Nathan Zimmer >> --- >> drivers/cpufreq/cpufreq.c | 312 +++++++++++++++++++++++++++++++++------------- >> 1 file changed, 224 insertions(+), 88 deletions(-) >> >> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c >> @@ -255,20 +258,21 @@ static inline void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci) >> void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state) >> { >> struct cpufreq_policy *policy; >> - unsigned long flags; >> + u8 flags; > I think you can get rid of flags. > >> BUG_ON(irqs_disabled()); >> >> if (cpufreq_disabled()) >> return; >> >> - freqs->flags = cpufreq_driver->flags; >> pr_debug("notification %u of frequency transition to %u kHz\n", >> state, freqs->new); >> >> - read_lock_irqsave(&cpufreq_driver_lock, flags); >> + rcu_read_lock(); >> + flags = rcu_dereference(cpufreq_driver)->flags; > use freq->flags here ... > >> policy = per_cpu(cpufreq_cpu_data, freqs->cpu); >> - read_unlock_irqrestore(&cpufreq_driver_lock, flags); >> + rcu_read_unlock(); >> + freqs->flags = flags; >> >> switch (state) { >> >> @@ -277,7 +281,7 @@ void cpufreq_notify_transition(struct cpufreq_freqs *freqs, unsigned int state) >> * which is not equal to what the cpufreq core thinks is >> * "old frequency". >> */ >> - if (!(cpufreq_driver->flags & CPUFREQ_CONST_LOOPS)) { >> + if (!(flags & CPUFREQ_CONST_LOOPS)) { > and here. Of course. >> if ((policy) && (policy->cpu == freqs->cpu) && >> (policy->cur) && (policy->cur != freqs->old)) { >> pr_debug("Warning: CPU frequency is" > >> @@ -742,35 +773,39 @@ static int cpufreq_add_dev_interface(unsigned int cpu, >> - write_lock_irqsave(&cpufreq_driver_lock, flags); >> + spin_lock_irqsave(&cpufreq_driver_lock, flags); >> for_each_cpu(j, policy->cpus) { >> per_cpu(cpufreq_cpu_data, j) = policy; >> per_cpu(cpufreq_policy_cpu, j) = policy->cpu; >> } >> - write_unlock_irqrestore(&cpufreq_driver_lock, flags); >> + spin_unlock_irqrestore(&cpufreq_driver_lock, flags); >> + synchronize_rcu(); > I don't think (but i can be wrong too :) ), that we need a synchronize_rcu() > here. We need it only at places where we have updated the cpufreq_driver > pointer. > > As we aren't doing any rcu specific read/update for cpufreq_cpu_data. Good point. I placed a similar sycnronize_rcu in cpufreq_add_policy_cpu and cpufreq_add_dev. I will remove them also. Thanks, I will respin. Nate