Re: [PATCH v2 1/5] usb: gadget: Add remote wakeup capable flag

[Thinh Nguyen]

On Mon, Jan 23, 2023, Elson Serrao wrote:
> 
> 
> On 1/23/2023 11:33 AM, Thinh Nguyen wrote:
> > On Sat, Jan 21, 2023, Thinh Nguyen wrote:
> > > On Fri, Jan 20, 2023, Alan Stern wrote:
> > > > On Sat, Jan 21, 2023 at 02:02:36AM +0000, Thinh Nguyen wrote:
> > > > > On Fri, Jan 20, 2023, Alan Stern wrote:
> > > > > > A UDC design might have multiple versions, some supporting remote wakeup
> > > > > > and others not.  But drivers generally use a single static
> > > > > > usb_gadget_ops structure, and they don't modify it at runtime to account
> > > > > > for hardware differences.  So if a single driver controls those multiple
> > > > > > versions, you can't rely on the presence of gadget->ops->wakeup to
> > > > > > indicate whether there actually is hardware remote wakeup support.
> > > > > > 
> > > > > > Ideally, the usb_gadget structure should have a wakeup_capable flag
> > > > > > which the UDC driver would set appropriately (probably during its probe
> > > > > > routine).
> > > > > > 
> > > > > 
> > > > > I was thinking that it can be handled by the
> > > > > usb_gadget_enable_remote_wakeup() so we can do away with the
> > > > > wakeup_capable flag.
> > > > 
> > > > usb_gadget_enable_remote_wakeup() gets called when the gadget or
> > > > function is suspended, right?  But a gadget driver may want to know long
> > > > before that whether the UDC supports remote wakeup, in order to set up
> > > > its config descriptor correctly.
> > > > 
> > > 
> > > No, this is to be called during set configuration. If the configuration
> > > doesn't support remote wakeup, the device should not be able to send
> > > remote wakeup.
> > > 
> > 
> > On second thought, you're right about the descriptor. It's better to
> > warn and prevent the remote wakeup bit from being set in the descriptor
> > if the UDC doesn't support remote wakeup. Warning the user at set
> > configuration is too late.
> > 
> > So, we need both rw_capable flag and usb_gadget_enable_remote_wakeup().
> > 
> > Thanks,
> > Thinh
> 
> Do we need usb_gadget_enable_remote_wakeup() gadget-op ?
> As per the discussion, we can have rw_capable flag in usb_gadget struct and
> set it during gagdet init/probe if the UDC supports resume signalling OR
> wants the remote wakeup feature to be enabled.
> This flag now represents UDC capability to initiate resume signalling.
> 
> During enumeration phase, when preparing the config descriptor we can use
> gadget->rw_capable flag to rightly modify the remote wakeup
> bit. Based on this, host will decide whether to arm the device for remote
> wakeup or not.
> 

If the configuration doesn't allow remote wakeup, the device should not
be able to send signal to wake up the host regardless whether the host
armed the device for remote wake or not.

The rw_capable flag will inform the composite layer whether the UDC is
capable of remote wakeup. The composite layer needs to tell the UDC
whether the configuration allows for remote wakeup.

Whether it's usb_gadget_enable_remote_wakeup() or the remote wakeup bit
in the bmAttribute, the composite layer needs to communicate that to the
controller driver. But we should not expect the UDC driver to parse for
that bit. The prepared control transfer from the the composite layer
should be abstracted from the UDC driver.

> For gadget->ops->wakeup callback support we already have explicit checks
> when invoking this gadget op and device wont be able to send remote wakeup
> if callback support doesn't exist.
> Please let me know if I am missing something.

As mentioned previously, we should also warn the user if the UDC doesn't
support remote wakeup and prevent the remote wakeup bit being set in the
descriptor.

Let me know if it makes sense.

Thanks,
Thinh