Re: [RFC PATCH 2/2] usb: dwc3: Support 'snps,gadget-keep-connect-sys-sleep' feature

[Thinh Nguyen <Thinh.Nguyen@synopsys.com>]

On Tue, Mar 21, 2023, Thinh Nguyen wrote:
> On Tue, Mar 21, 2023, Roger Quadros wrote:
> > Hi Thinh,
> > 
> > On 20/03/2023 20:52, Thinh Nguyen wrote:
> > > Hi,
> > > 
> > > On Mon, Mar 20, 2023, Roger Quadros wrote:
> > >> Implement 'snps,gadget-keep-connect-sys-sleep' property.
> > >>
> > >> Do not stop the gadget controller and disconnect if this
> > >> property is present and we are connected to a USB Host.
> > >>
> > >> Prevent System sleep if Gadget is not in USB suspend.
> > >>
> > >> Signed-off-by: Roger Quadros <rogerq@kernel.org>
> > >> ---
> > >>  drivers/usb/dwc3/core.c   | 25 +++++++++++++++++++------
> > >>  drivers/usb/dwc3/core.h   |  2 ++
> > >>  drivers/usb/dwc3/gadget.c | 25 +++++++++++++++++++++++--
> > >>  3 files changed, 44 insertions(+), 8 deletions(-)
> > >>
> > >> diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c
> > >> index 476b63618511..a47bbaa27302 100644
> > >> --- a/drivers/usb/dwc3/core.c
> > >> +++ b/drivers/usb/dwc3/core.c
> > >> @@ -1575,6 +1575,9 @@ static void dwc3_get_properties(struct dwc3 *dwc)
> > >>  	dwc->dis_split_quirk = device_property_read_bool(dev,
> > >>  				"snps,dis-split-quirk");
> > >>  
> > >> +	dwc->gadget_keep_connect_sys_sleep = device_property_read_bool(dev,
> > >> +				"snps,gadget-keep-connect-sys-sleep");
> > >> +
> > >>  	dwc->lpm_nyet_threshold = lpm_nyet_threshold;
> > >>  	dwc->tx_de_emphasis = tx_de_emphasis;
> > >>  
> > >> @@ -2027,14 +2030,20 @@ static int dwc3_suspend_common(struct dwc3 *dwc, pm_message_t msg)
> > >>  {
> > >>  	unsigned long	flags;
> > >>  	u32 reg;
> > >> +	int ret;
> > >>  
> > >>  	switch (dwc->current_dr_role) {
> > >>  	case DWC3_GCTL_PRTCAP_DEVICE:
> > >>  		if (pm_runtime_suspended(dwc->dev))
> > >>  			break;
> > >> -		dwc3_gadget_suspend(dwc);
> > >> +		ret = dwc3_gadget_suspend(dwc);
> > >> +		if (ret) {
> > >> +			dev_err(dwc->dev, "gadget not suspended: %d\n", ret);
> > >> +			return ret;
> > >> +		}
> > >>  		synchronize_irq(dwc->irq_gadget);
> > >> -		dwc3_core_exit(dwc);
> > >> +		if(!dwc->gadget_keep_connect_sys_sleep)
> > >> +			dwc3_core_exit(dwc);
> > >>  		break;
> > >>  	case DWC3_GCTL_PRTCAP_HOST:
> > >>  		if (!PMSG_IS_AUTO(msg) && !device_may_wakeup(dwc->dev)) {
> > >> @@ -2088,11 +2097,15 @@ static int dwc3_resume_common(struct dwc3 *dwc, pm_message_t msg)
> > >>  
> > >>  	switch (dwc->current_dr_role) {
> > >>  	case DWC3_GCTL_PRTCAP_DEVICE:
> > >> -		ret = dwc3_core_init_for_resume(dwc);
> > >> -		if (ret)
> > >> -			return ret;
> > >> +		if (!dwc->gadget_keep_connect_sys_sleep)
> > >> +		{
> > >> +			ret = dwc3_core_init_for_resume(dwc);
> > >> +			if (ret)
> > >> +				return ret;
> > >> +
> > >> +			dwc3_set_prtcap(dwc, DWC3_GCTL_PRTCAP_DEVICE);
> > >> +		}
> > >>  
> > >> -		dwc3_set_prtcap(dwc, DWC3_GCTL_PRTCAP_DEVICE);
> > >>  		dwc3_gadget_resume(dwc);
> > >>  		break;
> > >>  	case DWC3_GCTL_PRTCAP_HOST:
> > >> diff --git a/drivers/usb/dwc3/core.h b/drivers/usb/dwc3/core.h
> > >> index 582ebd9cf9c2..f84bac815bed 100644
> > >> --- a/drivers/usb/dwc3/core.h
> > >> +++ b/drivers/usb/dwc3/core.h
> > >> @@ -1328,6 +1328,8 @@ struct dwc3 {
> > >>  	unsigned		dis_split_quirk:1;
> > >>  	unsigned		async_callbacks:1;
> > >>  
> > >> +	unsigned		gadget_keep_connect_sys_sleep:1;
> > >> +
> > >>  	u16			imod_interval;
> > >>  
> > >>  	int			max_cfg_eps;
> > >> diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
> > >> index 3c63fa97a680..8062e44f63f6 100644
> > >> --- a/drivers/usb/dwc3/gadget.c
> > >> +++ b/drivers/usb/dwc3/gadget.c
> > >> @@ -4572,12 +4572,23 @@ void dwc3_gadget_exit(struct dwc3 *dwc)
> > >>  int dwc3_gadget_suspend(struct dwc3 *dwc)
> > >>  {
> > >>  	unsigned long flags;
> > >> +	int link_state;
> > >>  
> > >>  	if (!dwc->gadget_driver)
> > >>  		return 0;
> > >>  
> > >> -	dwc3_gadget_run_stop(dwc, false, false);
> > >> +	if (dwc->gadget_keep_connect_sys_sleep && dwc->connected) {
> > >> +		link_state = dwc3_gadget_get_link_state(dwc);
> > >> +		/* Prevent PM Sleep if not in U3/L2 */
> > >> +		if (link_state != DWC3_LINK_STATE_U3)
> > >> +			return -EBUSY;
> > >> +
> > >> +		/* don't stop/disconnect */
> > >> +		dwc3_gadget_disable_irq(dwc);
> > > 
> > > We shouldn't disable event interrupt here. What will happen if the
> > 
> > Due to some reason, if I don't disable the event interrupts here then
> > after USB resume the USB controller is malfunctioning.
> > It no longer responds to any requests from Host.
> 
> You should look into this. These events are important as they can tell
> whether the host initiates resume.
> 
> > 
> > > device is disconnected and reconnect to the host while the device is
> > > still in system suspend? The host would not be able to communicate with
> > > the device then.
> > 
> > In the TI platform, The system is woken up on any VBUS/linestate change
> > and in dwc3_gadget_resume we enable the events again and check for pending
> > events. Is it pointless to check for pending events there?
> > 
> 
> It seems fragile for the implementation to be dependent on platform
> specific feature right?
> 
> Also, what will happen in a typical case when the host puts the device
> in suspend and initiates resume while the device is in system suspend
> (and stay in suspend over a period of time)? There is no VBUS change.
> There will be problem if host detects no response from device in time.
> 
> Don't we need these events to wakeup the device?
> 

We may not be able to suspend everything in system suspend for this
case. I'm thinking of treating these events as if they are PME to wakeup
the device, but they are not the same. It may not be simple to handle
this. The lower layers may need to stay awake for the dwc3 to handle
these events. Hm... it gets a bit complicated.

Thanks,
Thinh