Re: [RESEND PATCH 2/2] PM/runtime: handle ->runtime_suspend failure correctly

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Sunday, October 09, 2011, tom.leiming@gmail.com wrote:
> From: Ming Lei <ming.lei@canonical.com>
> 
> If ->runtime_suspend returns -EAGAIN or -EBUSY, the device should
> still be in ACTIVE state, so it is not needed to send idle notification
> to its parent; if ->runtime_suspend returns other fatal failure, it
> doesn't make sense to send idle notification to its parent.
> 
> So skip these when failure is returned from ->runtime_suspend, also add
> comments for this handling in rpm_suspend.
> 
> This patch also updates comments for rpm_suspend:
> 
> - 'Cancel a pending idle notification' should be put before, also
> should be changed as 'Cancel a pending idle notification or
> autosuspend/suspend'

That should be a different patch I think?

> - idle notification for suspend failure has been removed, so update
> comments for it
> 
> Signed-off-by: Ming Lei <ming.lei@canonical.com>
> ---
> v1: some minor change on Alan's suggestion
> ---
>  drivers/base/power/runtime.c |   34 +++++++++++++++++++---------------
>  1 files changed, 19 insertions(+), 15 deletions(-)
> 
> diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
> index 441b5a3..e3c6a8f 100644
> --- a/drivers/base/power/runtime.c
> +++ b/drivers/base/power/runtime.c
> @@ -284,14 +284,17 @@ static int rpm_callback(int (*cb)(struct device *), struct device *dev)
>   * @dev: Device to suspend.
>   * @rpmflags: Flag bits.
>   *
> - * Check if the device's runtime PM status allows it to be suspended.  If
> - * another suspend has been started earlier, either return immediately or wait
> - * for it to finish, depending on the RPM_NOWAIT and RPM_ASYNC flags.  Cancel a
> - * pending idle notification.  If the RPM_ASYNC flag is set then queue a
> - * suspend request; otherwise run the ->runtime_suspend() callback directly.
> - * If a deferred resume was requested while the callback was running then carry
> - * it out; otherwise send an idle notification for the device (if the suspend
> - * failed) or for its parent (if the suspend succeeded).
> + * Check if the device's runtime PM status allows it to be suspended. Cancel
> + * a pending idle notification or autosuspend/suspend. If another suspend has
> + * been started earlier, either return immediately or wait for it to finish,
> + * depending on the RPM_NOWAIT and RPM_ASYNC flags. If the RPM_ASYNC flag is
> + * set then queue a suspend request; otherwise run the ->runtime_suspend()
> + * callback directly. If ->runtime_suspend returns failure, just cancel
> + * pending request and wake up waited tasks, then return immediatelly.
> + * After ->runtime_suspend succeeded, if a deferred resume was requested
> + * while the callback was running then carry it out; otherwise send an idle
> + * notification for its parent (if both ignore_children and irq_safe
> + * are not set).
>   *
>   * This function must be called under dev->power.lock with interrupts disabled.
>   */
> @@ -410,15 +413,16 @@ static int rpm_suspend(struct device *dev, int rpmflags)
>  			dev->power.runtime_error = 0;
>  		else
>  			pm_runtime_cancel_pending(dev);
> -	} else {
> +		wake_up_all(&dev->power.wait_queue);
> +		goto out;
> +	}
>   no_callback:

I don't think the change above is correct.  The code below
no_callback only should be executed if retval is zero.

To achieve the goal (i.e. avoid notifying the parent if -EAGAIN or
-EBUSY is returned by the callbacks) it would be sufficient to
do parent = NULL along with resetting power.runtime_error.

> -		__update_runtime_status(dev, RPM_SUSPENDED);
> -		pm_runtime_deactivate_timer(dev);
> +	__update_runtime_status(dev, RPM_SUSPENDED);
> +	pm_runtime_deactivate_timer(dev);
>  
> -		if (dev->parent) {
> -			parent = dev->parent;
> -			atomic_add_unless(&parent->power.child_count, -1, 0);
> -		}
> +	if (dev->parent) {
> +		parent = dev->parent;
> +		atomic_add_unless(&parent->power.child_count, -1, 0);
>  	}
>  	wake_up_all(&dev->power.wait_queue);

Thanks,
Rafael