Re: [PATCH v2 0/2] extcon: Inform the state of both ID and VBUS pin for USB

[Peter Chen <peter.chen@freescale.com>]

On Thu, May 28, 2015 at 05:23:57PM +0300, Roger Quadros wrote:
> +Peter & Li,
> 
> Ivan,
> 
> On 28/05/15 11:45, Ivan T. Ivanov wrote:
> >
> >Hi Chanwoo,
> >
> >On Wed, 2015-05-27 at 21:15 +0900, Chanwoo Choi wrote:
> >>Previously, I discussed how to inform the changed state of both ID
> >>and VBUS pin for USB connector on patch-set[1].
> >>[1] https://lkml.org/lkml/2015/4/2/310
> >>
> >>So, this patch adds the extcon_set_cable_line_state() function to inform
> >>the additional state of external connectors without additional register/
> >>unregister functions. This function uses the existing notifier chain
> >>which is registered by extcon_register_notifier() / extcon_register_interest().
> >>
> >>The extcon_set_cable_line_state() can inform the new state of both
> >>ID and VBUS pin state through extcon_set_cable_line_state().
> >>
> >>For exmaple:
> >>- On extcon-usb-gpio.c as extcon provider driver as following:
> >>         static void usb_extcon_detect_cable(struct work_struct *work)
> >>         {
> >>                 ...
> >>                 /* check ID and update cable state */
> >>                 id = gpiod_get_value_cansleep(info->id_gpiod);
> >>                 if (id) {
> >>                         extcon_set_cable_state_(info->edev, EXTCON_USB_HOST, false);
> >>                         extcon_set_cable_state_(info->edev, EXTCON_USB, true);
> >>
> >>                         extcon_set_cable_line_state(info->edev, EXTCON_USB,
> >>                                                         EXTCON_USB_ID_HIGH);
> >
> >I am getting more and more confused :-). Why EXTCON_USB is now used for ID notifications?
> >It should be EXTCON_USB_HOST, no? Why we need another function, framework already have
> >required information from the function one line above, do I miss something?
> 
> This is because the existing EXTCON_USB_HOST and EXTCON_USB do not capture all
> the 4 states of ID and VBUS pins that we need for a real USB driver to work.
> 
> It looks like it was designed from user space users perspective where they are
> only interested in USB role. i.e. host or peripheral.
> 
> Right now we are mixing both ID/VBUS and HOST/Peripheral states.
> This will break when we consider OTG role switching.
> With role switching, the USB device might start as a peripheral but switch role to host
> on the fly and the existing setup (including these patches) can't cater to that
> if user space is relying on EXTCON_USB_HOST/EXTCON_USB events.
> Because they are hard-wired to the ID pin state which doesn't change during
> role switch without cable switch.
> 
> The USB driver doesn't care about EXTCON_USB_HOST/peripheral states.
> It just needs ID/VBUS states and should decide the Host/Peripheral state from
> that and other inputs (like HNP/user request/etc).
> 
> The flow could be like this
> 
> (extcon-usb-driver) -> [ID/VBUS states] -> (USB driver) -> [HOST/Peripheral states]

Agree. Chanwoo, USB driver knows better than extcon driver about USB
role (host/peripheral), so the app should use USB interface to know it,
in fact, I don't be aware any use case needs to know USB role?
Are there any users for EXTCON_USB and EXTCON_USB_HOST currently?

> 
> If that is not going to happen then we will have to switch to this
> 
> (usb phy driver) -> [ID/VBUS states] -> (USB driver) -> (extcon f/w) -> [Host/Peripheral states]
> 
> 
> Felipe, Peter, Li,
> 
> what do you guys suggest?
> 
> cheers,
> -roger
> 
> >
> >>                 } else {
> >>                         extcon_set_cable_state_(info->edev, EXTCON_USB, false);
> >>                         extcon_set_cable_state_(info->edev, EXTCON_USB_HOST, true);
> >>
> >>                         extcon_set_cable_line_state(info->edev, EXTCON_USB,
> >>                                                         EXTCON_USB_ID_LOW);
> >>                 }
> >>         }
> >>
> >>- On specific extcon consumder driver as following:
> >>         static int xxx_probe(struct platform_device *pdev)
> >>         {
> >>                 struct notifier_chain nh;
> >>
> >>                 nb.notifier_call = extcon_usb_notifier;
> >>                 ret = extcon_register_notifier(edev, EXTCON_USB, &nb);
> >
> >This is bit misleading. 'nb' could not be in stack.
> >
> >Regards,
> >Ivan
> >

-- 

Best Regards,
Peter Chen