Re: [RFC][PATCH 1/3] usb: add a hcd notify entry in hc_driver

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Mon, May 04, 2015 at 11:15:30AM +0800, Lu Baolu wrote:
> This patch adds a new entry pointer in hc_driver. With this new entry,
> USB core can notify host driver when something happens and host driver
> is willing to be notified. One use case of this interface comes from
> xHCI compatible host controller driver.
> 
> The xHCI spec is designed to allow an xHC implementation to cache the
> endpoint state. Caching endpoint state allows an xHC to reduce latency
> when handling ERDYs and other USB asynchronous events. However holding
> this state in xHC consumes resources and power. The xHCI spec designs
> some methods through which host controller driver can hint xHC about
> how to optimize its operation, e.g. to determine when it holds state
> internally or pushes it out to memory, when to power down logic, etc.
> 
> When a USB device is going to suspend, states of all endpoints cached
> in the xHC should be pushed out to memory to save power and resources.
> Vice versa, when a USB device resumes, those states should be brought
> back to the cache. USB core suspends or resumes a USB device by sending
> set/clear port feature requests to the parent hub where the USB device
> is connected. Unfortunately, these operations are transparent to xHCI
> driver unless the USB device is plugged in a root port. This patch
> utilizes the notify interface to notify xHCI driver whenever a USB
> device's power state is changed.
> 
> It is harmless if a USB devices under USB 3.0 host controller suspends
> or resumes without a notification to hcd driver. However there may be
> less opportunities for power savings and there may be increased latency
> for restarting an endpoint. The precise impact will be different for
> each xHC implementation. It all depends on what an implementation does
> with the hints.
> 
> Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
> ---
>  drivers/usb/core/generic.c | 10 ++++++++--
>  drivers/usb/core/hcd.c     | 23 +++++++++++++++++++++++
>  include/linux/usb/hcd.h    | 11 ++++++++++-
>  3 files changed, 41 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/usb/core/generic.c b/drivers/usb/core/generic.c
> index 358ca8d..92bee33 100644
> --- a/drivers/usb/core/generic.c
> +++ b/drivers/usb/core/generic.c
> @@ -211,8 +211,12 @@ static int generic_suspend(struct usb_device *udev, pm_message_t msg)
>  	/* Non-root devices don't need to do anything for FREEZE or PRETHAW */
>  	else if (msg.event == PM_EVENT_FREEZE || msg.event == PM_EVENT_PRETHAW)
>  		rc = 0;
> -	else
> +	else {
> +		hcd_notify(udev, HCD_MSG_DEV_SUSPEND, &msg);
>  		rc = usb_port_suspend(udev, msg);
> +		if (rc)
> +			hcd_notify(udev, HCD_MSG_DEV_RESUME, &msg);
> +	}
>  
>  	return rc;
>  }
> @@ -228,8 +232,10 @@ static int generic_resume(struct usb_device *udev, pm_message_t msg)
>  	 */
>  	if (!udev->parent)
>  		rc = hcd_bus_resume(udev, msg);
> -	else
> +	else {
>  		rc = usb_port_resume(udev, msg);
> +		hcd_notify(udev, HCD_MSG_DEV_RESUME, &msg);
> +	}
>  	return rc;
>  }
>  
> diff --git a/drivers/usb/core/hcd.c b/drivers/usb/core/hcd.c
> index 45a915c..725d611 100644
> --- a/drivers/usb/core/hcd.c
> +++ b/drivers/usb/core/hcd.c
> @@ -2289,6 +2289,29 @@ void usb_hcd_resume_root_hub (struct usb_hcd *hcd)
>  }
>  EXPORT_SYMBOL_GPL(usb_hcd_resume_root_hub);
>  
> +/**
> + * hcd_notify - notify hcd driver with a message
> + * @udev: USB device
> + * @type: message type of this notification
> + * @data: message type specific data
> + *
> + * Call back to hcd driver to notify an event.
> + */
> +void hcd_notify(struct usb_device *udev,
> +		enum hcd_notification_type type, void *data)
> +{
> +	struct usb_hcd *hcd;
> +
> +	if (!udev)
> +		return;
> +
> +	hcd = bus_to_hcd(udev->bus);
> +
> +	if (hcd->driver->notify)
> +		hcd->driver->notify(hcd, udev, type, data);
> +}
> +EXPORT_SYMBOL_GPL(hcd_notify);

Why does this have to be exported?