Re: [PATCH 0/4] CPUFreq Fixes for 3.9

[Artem Savkov <artem.savkov@gmail.com>]

On Fri, Feb 08, 2013 at 10:39:13AM +0530, Viresh Kumar wrote:
> On 8 February 2013 04:37, Rafael J. Wysocki <rjw@sisk.pl> wrote:
> > BTW, there still are locking problems in linux-next.  Why do we need
> > to take cpufreq_driver_lock() around driver->init() in cpufreq_add_dev(),
> > in particular?
> I thought a bit more and realized there is no such limitation on
> cpufreq_driver->ops about calling routines which can sleep. And thus
> we shoudln't
> have locks around any of these. I have got a patch for it, that i
> would fold-back into
> the original patch that introduced locking fixes (attached too for testing):
Tested this patch on top of linux-pm.git/bleeding-edge
Now everything seems to be alright.

> From: Viresh Kumar <viresh.kumar@linaro.org>
> Date: Fri, 8 Feb 2013 10:35:31 +0530
> Subject: [PATCH] cpufreq: Remove unnecessary locking
> 
> I have placed some locks intentionally around calls to driver->ops (init/exit),
> which look to be wrong as these calls can call routines that potentially sleep.
> 
> Lets remove these locks.
> 
> Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
> ---
>  drivers/cpufreq/cpufreq.c | 7 -------
>  1 file changed, 7 deletions(-)
> 
> diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
> index 5d8a422..04aab05 100644
> --- a/drivers/cpufreq/cpufreq.c
> +++ b/drivers/cpufreq/cpufreq.c
> @@ -795,10 +795,8 @@ static int cpufreq_add_dev_interface(unsigned int cpu,
> 
>  	if (ret) {
>  		pr_debug("setting policy failed\n");
> -		spin_lock_irqsave(&cpufreq_driver_lock, flags);
>  		if (driver->exit)
>  			driver->exit(policy);
> -		spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
>  	}
>  	return ret;
> 
> @@ -920,17 +918,14 @@ static int cpufreq_add_dev(struct device *dev,
> struct subsys_interface *sif)
>  	init_completion(&policy->kobj_unregister);
>  	INIT_WORK(&policy->update, handle_update);
> 
> -	spin_lock_irqsave(&cpufreq_driver_lock, flags);
>  	/* call driver. From then on the cpufreq must be able
>  	 * to accept all calls to ->verify and ->setpolicy for this CPU
>  	 */
>  	ret = driver->init(policy);
>  	if (ret) {
>  		pr_debug("initialization failed\n");
> -		spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
>  		goto err_set_policy_cpu;
>  	}
> -	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
> 
>  	/* related cpus should atleast have policy->cpus */
>  	cpumask_or(policy->related_cpus, policy->related_cpus, policy->cpus);
> @@ -1100,10 +1095,8 @@ static int __cpufreq_remove_dev(struct device
> *dev, struct subsys_interface *sif
>  		wait_for_completion(cmp);
>  		pr_debug("wait complete\n");
> 
> -		spin_lock_irqsave(&cpufreq_driver_lock, flags);
>  		if (driver->exit)
>  			driver->exit(data);
> -		spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
> 
>  		free_cpumask_var(data->related_cpus);
>  		free_cpumask_var(data->cpus);

Tested-by: Artem Savkov <artem.savkov@gmail.com>

-- 
Kind regards,
Artem