Re: [PATCH v12 6/6] usb: gadget: f_ecm: Add suspend/resume and remote wakeup support

[Thinh Nguyen]

On Fri, Mar 17, 2023, Elson Serrao wrote:
> 
> 
> On 3/17/2023 4:20 PM, Thinh Nguyen wrote:
> > On Fri, Mar 17, 2023, Thinh Nguyen wrote:
> > > On Fri, Mar 17, 2023, Elson Serrao wrote:
> > > > 
> > > > 
> > > > On 3/16/2023 5:11 PM, Thinh Nguyen wrote:
> > > > > On Thu, Mar 16, 2023, Elson Roy Serrao wrote:
> > > > > > When host sends a suspend notification to the device, handle
> > > > > > the suspend callbacks in the function driver. Enhanced super
> > > > > > speed devices can support function suspend feature to put the
> > > > > > function in suspend state. Handle function suspend callback.
> > > > > > 
> > > > > > Depending on the remote wakeup capability the device can either
> > > > > > trigger a remote wakeup or wait for the host initiated resume to
> > > > > > start data transfer again.
> > > > > > 
> > > > > > Signed-off-by: Elson Roy Serrao <quic_eserrao@xxxxxxxxxxx>
> > > > > > ---
> > > > > >    drivers/usb/gadget/function/f_ecm.c   | 71 +++++++++++++++++++++++++++++++++++
> > > > > >    drivers/usb/gadget/function/u_ether.c | 63 +++++++++++++++++++++++++++++++
> > > > > >    drivers/usb/gadget/function/u_ether.h |  4 ++
> > > > > >    3 files changed, 138 insertions(+)
> > > > > > 
> > > > > > diff --git a/drivers/usb/gadget/function/f_ecm.c b/drivers/usb/gadget/function/f_ecm.c
> > > > > > index a7ab30e..c43cd557 100644
> > > > > > --- a/drivers/usb/gadget/function/f_ecm.c
> > > > > > +++ b/drivers/usb/gadget/function/f_ecm.c
> > > > > > @@ -633,6 +633,8 @@ static void ecm_disable(struct usb_function *f)
> > > > > >    	usb_ep_disable(ecm->notify);
> > > > > >    	ecm->notify->desc = NULL;
> > > > > > +	f->func_suspended = false;
> > > > > > +	f->func_wakeup_armed = false;
> > > > > >    }
> > > > > >    /*-------------------------------------------------------------------------*/
> > > > > > @@ -885,6 +887,71 @@ static struct usb_function_instance *ecm_alloc_inst(void)
> > > > > >    	return &opts->func_inst;
> > > > > >    }
> > > > > > +static void ecm_suspend(struct usb_function *f)
> > > > > > +{
> > > > > > +	struct f_ecm *ecm = func_to_ecm(f);
> > > > > > +	struct usb_composite_dev *cdev = ecm->port.func.config->cdev;
> > > > > > +
> > > > > > +	if (f->func_suspended) {
> > > > > > +		DBG(cdev, "Function already suspended\n");
> > > > > > +		return;
> > > > > > +	}
> > > > > > +
> > > > > > +	DBG(cdev, "ECM Suspend\n");
> > > > > > +
> > > > > > +	gether_suspend(&ecm->port);
> > > > > > +}
> > > > > > +
> > > > > > +static void ecm_resume(struct usb_function *f)
> > > > > > +{
> > > > > > +	struct f_ecm *ecm = func_to_ecm(f);
> > > > > > +	struct usb_composite_dev *cdev = ecm->port.func.config->cdev;
> > > > > > +
> > > > > > +	/*
> > > > > > +	 * If the function is in USB3 Function Suspend state, resume is
> > > > > > +	 * canceled. In this case resume is done by a Function Resume request.
> > > > > > +	 */
> > > > > > +	if (f->func_suspended)
> > > > > > +		return;
> > > > > > +
> > > > > > +	DBG(cdev, "ECM Resume\n");
> > > > > > +
> > > > > > +	gether_resume(&ecm->port);
> > > > > > +}
> > > > > > +
> > > > > > +static int ecm_get_status(struct usb_function *f)
> > > > > > +{
> > > > > > +	struct usb_configuration *c = f->config;
> > > > > > +
> > > > > > +	/* D0 and D1 bit set to 0 if device is not wakeup capable */
> > > > > > +	if (!(USB_CONFIG_ATT_WAKEUP & c->bmAttributes))
> > > > > > +		return 0;
> > > > > > +
> > > > > > +	return (f->func_wakeup_armed ? USB_INTRF_STAT_FUNC_RW : 0) |
> > > > > > +		USB_INTRF_STAT_FUNC_RW_CAP;
> > > > > > +}
> > > > > 
> > > > > Why do we need to implement ecm_get_status if it's already handled in
> > > > > composite.c now?
> > > > > 
> > > > 
> > > > Yes this can be removed now. Will modify accordingly.
> > > > > > +
> > > > > > +static int ecm_func_suspend(struct usb_function *f, u8 options)
> > > > > > +{
> > > > > > +	struct usb_composite_dev *cdev = f->config->cdev;
> > > > > > +
> > > > > > +	DBG(cdev, "func susp %u cmd\n", options);
> > > > > > +
> > > > > > +	if (options & (USB_INTRF_FUNC_SUSPEND_LP >> 8)) {
> > > > > 
> > > > > This feature selector doesn't indicate whether it's SetFeature or
> > > > > ClearFeature request. ecm_func_suspend is supposed to be for
> > > > > SetFeature(suspend) only. Perhaps we may have to define func_resume()
> > > > > for ClearFeature(suspend)?
> > > > > 
> > > 
> > > > Host uses the same feature selector FUNCTION_SUSPEND for function suspend
> > > > and function resume and func_suspend() callback can be used to
> > > > handle both the cases ? The distinction comes whether it is a
> > > 
> > > How do you plan to handle that? Pass this info in some unused/reserved
> > > bit of the "options" argument? Introduce a new parameter to the
> > > func_suspend()?
> > > 
> > > If that's the case, then you need to update the document on
> > > func_suspend() to also support ClearFeature(suspend). Right now it's
> > > documented for SetFeature only. Also, make sure that other existing
> > > function drivers will not break because of the change of the
> > > func_suspend behavior.
> > > 
> > > > SetFeature(FUNCTION_SUSPEND) or ClearFeature(FUNCTION_SUSPEND) which can be
> > > > easily done in the func_suspend callback itself. We can add another callback
> > > > func_resume specific to ClearFeature(FUNCTION_SUSPEND) but wont that be
> > > > redundant and more callback handling on function driver/composite side as
> > > > well? Please let me know your opinion.
> > > > 
> > > 
> > > We actually didn't properly define func_suspend and its counter part. It
> > > seems cleaner to me to introduce func_resume as it seems more intuitive
> > > and easier to read. Let me know how you plan to use func_suspend() for
> > > both cases.
> > > 
> > 
> > How about we handle function suspend resume in composite also? I mean
> > something like this:
> > 
> > diff --git a/drivers/usb/gadget/composite.c b/drivers/usb/gadget/composite.c
> > index 36add1879ed2..79dc055eb5f7 100644
> > --- a/drivers/usb/gadget/composite.c
> > +++ b/drivers/usb/gadget/composite.c
> > @@ -1948,9 +1948,18 @@ composite_setup(struct usb_gadget *gadget, const struct usb_ctrlrequest *ctrl)
> >   		f = cdev->config->interface[intf];
> >   		if (!f)
> >   			break;
> > -		status = f->get_status ? f->get_status(f) : 0;
> > -		if (status < 0)
> > -			break;
> > +
> > +		if (f->get_status) {
> > +			status = f->get_status(f);
> > +			if (status < 0)
> > +				break;
> > +		} else {
> > +			if (f->config->bmAttributes & USB_CONFIG_ATT_WAKEUP) {
> > +				status |= USB_INTRF_STAT_FUNC_RW_CAP;
> > +				if (f->func_wakeup_armed)
> > +					status |= USB_INTRF_STAT_FUNC_RW;
> > +			}
> > +		}
> >   		put_unaligned_le16(status & 0x0000ffff, req->buf);
> >   		break;
> >   	/*
> > @@ -1971,9 +1980,28 @@ composite_setup(struct usb_gadget *gadget, const struct usb_ctrlrequest *ctrl)
> >   			f = cdev->config->interface[intf];
> >   			if (!f)
> >   				break;
> > +			if (w_index & USB_INTRF_FUNC_SUSPEND_RW) {
> > +				if (!(f->config->bmAttributes & USB_CONFIG_ATT_WAKEUP))
> > +					break;
> > +
> > +				f->func_wakeup_armed = (ctrl->bRequest == USB_REQ_SET_FEATURE);
> > +			}
> > +
> >   			value = 0;
> > -			if (f->func_suspend)
> > +			if (f->func_suspend) {
> >   				value = f->func_suspend(f, w_index >> 8);
> > +			} else if (w_index & USB_INTRF_FUNC_SUSPEND_LP) {
> > +				if (f->suspend && && !f->func_suspended &&
> > +				    ctrl->bRequest == USB_REQ_SET_FEATURE)) {
> > +					f->suspend(f);
> > +					f->func_suspended = true;
> > +				} else if (f->resume && f->func_suspended &&
> > +					   ctrl->bRequest == USB_REQ_CLEAR_FEATURE_FEATURE)) {
> > +					f->resume(f);
> > +					f->func_suspended = false;
> > +				}
> > +			}
> > +
> >   			if (value < 0) {
> >   				ERROR(cdev,
> >   				      "func_suspend() returned error %d\n",
> > 
> At individual function driver level there is no need to differentiate
> between suspend() and func_suspend() APIs, as both are intended to put the
> function in suspend state. So your idea/implementation above makes it much
> more clearer. Let composite also handle this and call either f->suspend() or
> f->resume() callback based on the setup packet received. Thank you for this
> suggestion.
> 
> > 
> > Also, do we need the f->func_suspended flag? we'd need the remote wakeup
> > flag for the status, but when do we need f->func_suspended? It seems
> > like it can be handled within the function driver's scope.
> 
> f->func_suspended flag I had added for below purposes
> 
> 1.) Function drivers should know the right wakeup() op to be called.
> That is if they are in FUNC_SUSPEND then call usb_func_wakeup() and if they
> are in device suspend then call usb_gadget_wakeup(). (we can use
> f->func_wakeup_armed flag for this purpose as well)
> 
> 2.) If a function is in USB3 FUNCTION_SUSPEND state then it shouldn't allow
> f->resume() called through composite_resume() as the exit from
> FUNCTION_SUSPEND state is via ClearFeature(FUNCTION_SUSPEND).
> 
> So we need a way to tell function drivers if they are in USB3
> FUNCTION_SUSPEND state OR device suspend.

Ok. So do you plan to update composite_resume to check that? Perhaps
document this expected behavior also?

> 
> Please let me know if you see any alternative or better approach here.
> 

If we have a use for it in composite.c, then it should be made
available. Otherwise it should only exist in the function driver (ie. in
f_ecm structure).

Thanks,
Thinh