Re: [PATCH] PM: Enable asynchronous noirq resume threads to save the resuming time

["Rafael J. Wysocki" <rjw@rjwysocki.net>]

On Tuesday, January 14, 2014 03:18:08 PM Chuansheng Liu wrote:
> 
> Currently, the dpm_resume_noirq() is done synchronously, and for PCI devices
> pci_pm_resume_noirq():
> 
> pci_pm_resume_noirq()
>  pci_pm_default_resume_early()
>   pci_power_up()
>    pci_raw_set_power_state()
> Which set the device from D3hot to D0 mostly, for every device, there will
> be one 10ms(pci_pm_d3_delay) to wait.
> 
> Hence normally dpm_resume_noirq() will cost > 100ms, which is bigger for mobile
> platform.

pci_pm_d3_delay usually is not 10 ms.  What is 10 ms is dev->d3_delay, but
that also is not on every platform now.

That said the approach here makes sense, but I have two questions:

1. On what systems has it been tested?
2. What about suspend_late/resume_early?  If the _noirq things are to be executed
   asynchronously, those should be too.

Thanks!

> Here implementing it with asynchronous way which will reduce much.
> 
> For example below, The 80% time is saved.
> With synchronous way:
> [ 1411.272218] PM: noirq resume of devices complete after 92.223 msecs
> With asynchronous way:
> [  110.616735] PM: noirq resume of devices complete after 10.544 msecs
> 
> Signed-off-by: Liu, Chuansheng <chuansheng.liu@intel.com>
> ---
>  drivers/base/power/main.c |   42 +++++++++++++++++++++++++++++++++---------
>  1 file changed, 33 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/base/power/main.c b/drivers/base/power/main.c
> index 1b41fca..1b9d774 100644
> --- a/drivers/base/power/main.c
> +++ b/drivers/base/power/main.c
> @@ -505,6 +505,19 @@ static int device_resume_noirq(struct device *dev, pm_message_t state)
>  	return error;
>  }
>  
> +static bool is_async(struct device *dev);
> +
> +static void async_resume_noirq(void *data, async_cookie_t cookie)
> +{
> +	struct device *dev = (struct device *)data;
> +	int error;
> +
> +	error = device_resume_noirq(dev, pm_transition);
> +	if (error)
> +		pm_dev_err(dev, pm_transition, " noirq", error);
> +	put_device(dev);
> +}
> +
>  /**
>   * dpm_resume_noirq - Execute "noirq resume" callbacks for all devices.
>   * @state: PM transition of the system being carried out.
> @@ -514,29 +527,40 @@ static int device_resume_noirq(struct device *dev, pm_message_t state)
>   */
>  static void dpm_resume_noirq(pm_message_t state)
>  {
> +	struct device *dev;
>  	ktime_t starttime = ktime_get();
> +	pm_transition = state;
> +
> +	list_for_each_entry(dev, &dpm_noirq_list, power.entry) {
> +		if (is_async(dev)) {
> +			get_device(dev);
> +			async_schedule(async_resume_noirq, dev);
> +		}
> +	}
>  
>  	mutex_lock(&dpm_list_mtx);
>  	while (!list_empty(&dpm_noirq_list)) {
> -		struct device *dev = to_device(dpm_noirq_list.next);
> -		int error;
> +		dev = to_device(dpm_noirq_list.next);
>  
>  		get_device(dev);
>  		list_move_tail(&dev->power.entry, &dpm_late_early_list);
>  		mutex_unlock(&dpm_list_mtx);
>  
> -		error = device_resume_noirq(dev, state);
> -		if (error) {
> -			suspend_stats.failed_resume_noirq++;
> -			dpm_save_failed_step(SUSPEND_RESUME_NOIRQ);
> -			dpm_save_failed_dev(dev_name(dev));
> -			pm_dev_err(dev, state, " noirq", error);
> +		if (!is_async(dev)) {
> +			int error;
> +			error = device_resume_noirq(dev, state);
> +			if (error) {
> +				suspend_stats.failed_resume_noirq++;
> +				dpm_save_failed_step(SUSPEND_RESUME_NOIRQ);
> +				dpm_save_failed_dev(dev_name(dev));
> +				pm_dev_err(dev, state, " noirq", error);
> +			}
>  		}
> -
>  		mutex_lock(&dpm_list_mtx);
>  		put_device(dev);
>  	}
>  	mutex_unlock(&dpm_list_mtx);
> +	async_synchronize_full();
>  	dpm_show_time(starttime, state, "noirq");
>  	resume_device_irqs();
>  	cpuidle_resume();
> 

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.