From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932127AbdDLUVP (ORCPT ); Wed, 12 Apr 2017 16:21:15 -0400 Received: from Galois.linutronix.de ([146.0.238.70]:35495 "EHLO Galois.linutronix.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932080AbdDLUVI (ORCPT ); Wed, 12 Apr 2017 16:21:08 -0400 Message-Id: <20170412201042.958216363@linutronix.de> User-Agent: quilt/0.63-1 Date: Wed, 12 Apr 2017 22:07:36 +0200 From: Thomas Gleixner To: LKML Cc: Peter Zijlstra , Ingo Molnar , Sebastian Siewior , Benjamin Herrenschmidt , "David S. Miller" , Fenghua Yu , Herbert Xu , Lai Jiangshan , Len Brown , Michael Ellerman , "Rafael J. Wysocki" , Tejun Heo , Tony Luck , Viresh Kumar , linux-pm@vger.kernel.org Subject: [patch 10/13] cpufreq/sh: Replace racy task affinity logic References: <20170412200726.941336635@linutronix.de> MIME-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-15 Content-Disposition: inline; filename=cpufreq-sh--Use-work_on_cpu--.patch Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org The target() callback must run on the affected cpu. This is achieved by temporarily setting the affinity of the calling thread to the requested CPU and reset it to the original affinity afterwards. That's racy vs. concurrent affinity settings for that thread resulting in code executing on the wrong CPU. Replace it by work_on_cpu(). All call pathes which invoke the callbacks are already protected against CPU hotplug. Signed-off-by: Thomas Gleixner Cc: "Rafael J. Wysocki" Cc: Viresh Kumar Cc: linux-pm@vger.kernel.org --- drivers/cpufreq/sh-cpufreq.c | 45 +++++++++++++++++++++++++------------------ 1 file changed, 27 insertions(+), 18 deletions(-) --- a/drivers/cpufreq/sh-cpufreq.c +++ b/drivers/cpufreq/sh-cpufreq.c @@ -30,54 +30,63 @@ static DEFINE_PER_CPU(struct clk, sh_cpuclk); +struct cpufreq_target { + struct cpufreq_policy *policy; + unsigned int freq; +}; + static unsigned int sh_cpufreq_get(unsigned int cpu) { return (clk_get_rate(&per_cpu(sh_cpuclk, cpu)) + 500) / 1000; } -/* - * Here we notify other drivers of the proposed change and the final change. - */ -static int sh_cpufreq_target(struct cpufreq_policy *policy, - unsigned int target_freq, - unsigned int relation) +static long __sh_cpufreq_target(void *arg) { - unsigned int cpu = policy->cpu; + struct cpufreq_target *target = arg; + struct cpufreq_policy *policy = target->policy; + int cpu = policy->cpu; struct clk *cpuclk = &per_cpu(sh_cpuclk, cpu); - cpumask_t cpus_allowed; struct cpufreq_freqs freqs; struct device *dev; long freq; - cpus_allowed = current->cpus_allowed; - set_cpus_allowed_ptr(current, cpumask_of(cpu)); - - BUG_ON(smp_processor_id() != cpu); + if (smp_processor_id() != cpu) + return -ENODEV; dev = get_cpu_device(cpu); /* Convert target_freq from kHz to Hz */ - freq = clk_round_rate(cpuclk, target_freq * 1000); + freq = clk_round_rate(cpuclk, target->freq * 1000); if (freq < (policy->min * 1000) || freq > (policy->max * 1000)) return -EINVAL; - dev_dbg(dev, "requested frequency %u Hz\n", target_freq * 1000); + dev_dbg(dev, "requested frequency %u Hz\n", target->freq * 1000); freqs.old = sh_cpufreq_get(cpu); freqs.new = (freq + 500) / 1000; freqs.flags = 0; - cpufreq_freq_transition_begin(policy, &freqs); - set_cpus_allowed_ptr(current, &cpus_allowed); + cpufreq_freq_transition_begin(target->policy, &freqs); clk_set_rate(cpuclk, freq); - cpufreq_freq_transition_end(policy, &freqs, 0); + cpufreq_freq_transition_end(target->policy, &freqs, 0); dev_dbg(dev, "set frequency %lu Hz\n", freq); - return 0; } +/* + * Here we notify other drivers of the proposed change and the final change. + */ +static int sh_cpufreq_target(struct cpufreq_policy *policy, + unsigned int target_freq, + unsigned int relation) +{ + struct cpufreq_target data = { .policy = policy, .freq = target_freq }; + + return work_on_cpu(policy->cpu, __sh_cpufreq_target, &data); +} + static int sh_cpufreq_verify(struct cpufreq_policy *policy) { struct clk *cpuclk = &per_cpu(sh_cpuclk, policy->cpu);