cpufreq: Call __cpufreq_governor() with correct policy->cpus mask
authorViresh Kumar <viresh.kumar@linaro.org>
Thu, 11 Apr 2013 08:04:53 +0000 (08:04 +0000)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Thu, 11 Apr 2013 20:50:09 +0000 (22:50 +0200)
__cpufreq_governor() must be called with a correct policy->cpus mask.
In __cpufreq_remove_dev() we initially clear policy->cpus with
cpumask_clear_cpu() and then call
__cpufreq_governor(policy, CPUFREQ_GOV_POLICY_EXIT). If the governor
is doing some per-cpu stuff in EXIT callback, this can create
uncertain behavior.

Generic governors in drivers/cpufreq/ doesn't do any per-cpu stuff
in EXIT callback and so we don't face any issues currently. But its
better to keep the code clean, so we don't face any issues in future.

Now, we call cpumask_clear_cpu() only when multiple cpus are managed
by policy.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
drivers/cpufreq/cpufreq.c

index fd97a62..3564947 100644 (file)
@@ -1105,7 +1105,9 @@ static int __cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif
 
        WARN_ON(lock_policy_rwsem_write(cpu));
        cpus = cpumask_weight(data->cpus);
-       cpumask_clear_cpu(cpu, data->cpus);
+
+       if (cpus > 1)
+               cpumask_clear_cpu(cpu, data->cpus);
        unlock_policy_rwsem_write(cpu);
 
        if (cpu != data->cpu) {