Re: [PATCH v10 5/6] usb:cdns3 Add Cadence USB3 DRD Driver

[Chunfeng Yun]

On Tue, 2019-08-13 at 10:48 +0300, Roger Quadros wrote:
> 
> On 13/08/2019 10:30, Chunfeng Yun wrote:
> > On Mon, 2019-08-12 at 16:04 +0300, Roger Quadros wrote:
> >>
> >> On 12/08/2019 15:46, Felipe Balbi wrote:
> >>>
> >>> Hi,
> >>>
> >>> Roger Quadros <rogerq@xxxxxx> writes:
> >>>>> The sysfs file we expose from the class for the role switches is
> >>>>> primarily meant for supporting proprietary protocols that require us
> >>>>> to basically override the connector USB data role. The default role
> >>>>> should always be selected in the drivers.
> >>>>
> >>>> OK. Let's take this example
> >>>> - Port is dual-role port micro AB.
> >>>> - microAB to type-A adapter is connected which pulls ID low. port transitions
> >>>> to "host" role by the controller driver.
> >>>> - proprietary protocol want to switch role to device role so writes "device" to
> >>>> mode switch sysfs. port transitions to "device" role.
> >>>>
> >>>> Now, how does controller driver know to fall back to HW based role switching?
> >>>
> >>> Use a 'disconnect' or 'suspend' event to go reset it? But that should,
> >>> probably, be done at kernel space, no?
> >>>
> >>
> >> Yes that could be one option.
> >> So after a disconnect, sysfs role should reflect actual hardware role. correct?
> > 
> > Maybe it's difficult to support both HW based role switch and SW based
> > role switch by sysfs at the same if the HW's FSM rely on, such as, the
> > state of Vbus pin or ID pin. Likes the upper example, when user writes
> > "device" to mode switch sysfs, the driver should skip the HW state of ID
> > pin, due to it's state is Low, or force it as High.
> > 
> 
> We do need a clear way of indicating that SW wants to override so HW
> state is ignored.
> 
> > Another option way is that introduces a property in DTS to indicate the
> > way the driver want to use (HW based or SW based, usb_role_switch
> > doesn't provide this information for the controller driver), but is not
> > flexible enough.
> 
> That is not good enough for us. We need both HW and SW based role switching.
> 
> Can we introduce a new state (e.g. "auto") in usb_role_switch. This would
> explicitly indicate the driver to do HW based switching.
But "auto" is not a role?

How about introducing a new attribute in usb_role_switch?

> 
> This way we don't need to depend on connect/disconnect events and can
> do role switch tests even without cable/device connected.
>