On 08/06/16 15:42, Peter Chen wrote:

On Wed, Jun 08, 2016 at 01:12:10PM +0300, Roger Quadros wrote:

quoted

Hi,

On 08/06/16 12:53, Peter Chen wrote:

quoted

On Wed, Jun 08, 2016 at 12:03:40PM +0300, Roger Quadros wrote:

quoted

+int usb_otg_unregister(struct device *dev)
+{
+	struct usb_otg *otg;
+
+	mutex_lock(&otg_list_mutex);
+	otg = usb_otg_get_data(dev);
+	if (!otg) {
+		dev_err(dev, "otg: %s: device not in otg list\n",
+			__func__);
+		mutex_unlock(&otg_list_mutex);
+		return -EINVAL;
+	}
+
+	/* prevent unregister till both host & gadget have unregistered */
+	if (otg->host || otg->gadget) {
+		dev_err(dev, "otg: %s: host/gadget still registered\n",
+			__func__);

You need to call mutex_unlock here

Indeed. good catch.

quoted

quoted

+
+int usb_otg_gadget_ready(struct usb_gadget *gadget, bool ready)
+{

What this API is for? Why need it in this version?

we moved gadget to otg registration from udc_bind_to_driver() to usb_add_gadget_udc_release().
This means there is a window when gadget function driver (e.g. g_zero) is not loaded.
We don't want to start the gadget controller in that window.

usb_otg_gadget_ready() is used by gadget core to notify the otg core when the function driver
is ready or not-ready.

Why you need to move this from gadget's probe to udc's probe? Currently,
the sequence of gadget and udc's probe is random, but udc_bind_to_driver
is called when udc is ready.

Because we need to support -EPROBE_DEFER at udc probe.
At udc_bind_to_driver there is no scope for -EPROBE_DEFER.

Another issue is that we need to register with otg and have otg->caps available
before driver->bind() is called in udc_bind_to_driver(). Else gadget function
will not have the correct OTG descriptor.

cheers,
-roger