Re: [PATCH v14 2/7] PM / wakeup: Add device_children_wakeup_capable()

[Matthias Kaehlcke <mka@chromium.org>]

Hi Pavan,

On Sat, Apr 30, 2022 at 08:41:30AM +0530, Pavan Kondeti wrote:
> Hi Matthias,
> 
> On Fri, Apr 29, 2022 at 12:19:22PM -0700, Matthias Kaehlcke wrote:
> > Hi Pavan,
> > 
> > On Fri, Apr 29, 2022 at 06:29:56PM +0530, Pavan Kondeti wrote:
> > > Hi Matthias,
> > > 
> > > On Mon, Apr 25, 2022 at 06:33:03PM +0530, Pavan Kondeti wrote:
> > > > Hi Matthias,
> > > > 
> > > > On Fri, Apr 22, 2022 at 11:44:36AM -0700, Matthias Kaehlcke wrote:
> > > > > On Fri, Apr 22, 2022 at 01:57:17PM +0200, Rafael J. Wysocki wrote:
> > > > > > On Tue, Apr 19, 2022 at 9:11 PM Sandeep Maheswaram
> > > > > > <quic_c_sanm@quicinc.com> wrote:
> > > > > > >
> > > > > > > From: Matthias Kaehlcke <mka@chromium.org>
> > > > > > >
> > > > > > > Add device_children_wakeup_capable() which checks whether the device itself
> > > > > > > or one if its descendants is wakeup capable.
> > > > > > 
> > > > > > device_wakeup_path() exists for a very similar purpose.
> > > > > > 
> > > > > > Is it not usable for whatever you need the new function introduced here?
> > > > > 
> > > > > I wasn't aware of it's function, there are no doc comments and the
> > > > > name isn't really self explanatory.
> > > > > 
> > > > > In a quick test device_wakeup_path() returned inconsistent values for the
> > > > > root hub, sometimes true, others false when a wakeup capable USB device was
> > > > > connected.
> > > > 
> > > > We will also test the same to double confirm the behavior of
> > > > device_wakeup_path(). I am assuming that you checked device_wakeup_path()
> > > > only during system suspend path.
> > > > 
> > > > Here is what I understood by looking at __device_suspend(). Please share
> > > > your thoughts on this.
> > > > 
> > > > power.wakeup_path is set to true for the parent *after* a wakeup capable
> > > > device is suspended. This means when the root hub(s) is suspended, it is
> > > > propagated to xhci-plat and when xhci-plat is suspended, it is propagated
> > > > to dwc3. bottom up propgation during system suspend.
> > > > 
> > > > I believe we can directly check something like this in the dwc3 driver
> > > > instead of having another wrapper like device_children_wakeup_capable().
> > > > 
> > > > diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c
> > > > index 1170b80..a783257 100644
> > > > --- a/drivers/usb/dwc3/core.c
> > > > +++ b/drivers/usb/dwc3/core.c
> > > > @@ -1878,8 +1878,14 @@ static int dwc3_suspend_common(struct dwc3 *dwc, pm_message_t msg)
> > > >  		break;
> > > >  	case DWC3_GCTL_PRTCAP_HOST:
> > > >  		if (!PMSG_IS_AUTO(msg)) {
> > > > +			/*
> > > > +			 * Don't kill the host when dwc3 is wakeup capable and
> > > > +			 * its children needs wakeup.
> > > > +			 */
> > > > +			if (device_may_wakeup(dwc->dev) && device_wakeup_path(dwc->dev))
> > > > +				handle_it();
> > > > +		} else {
> > > >  			dwc3_core_exit(dwc);
> > > > -			break;
> > > >  		}
> > > >  
> > > >  		/* Let controller to suspend HSPHY before PHY driver suspends */
> > > > 
> > > 
> > > device_wakeup_path(dwc->dev) is returning true all the time irrespective of
> > > the wakeup capability (and enabled status) of the connected USB devices. That
> > > is because xhci-plat device is configured to wakeup all the time. Since the
> > > child is wakeup capable, its parent i.e dwc3 has device_wakeup_path() set.
> > > device_children_wakeup_capable() will also suffer the problem. However,
> > > 
> > > device_children_wakeup_capable(&hcd->self.root_hub->dev) is what Sandeep's
> > > patch is using. That is not correct. we have two root hubs (HS and SS) associated
> > > with a USB3 controller and calling it on one root hub is incorrect. 
> > > device_children_wakeup_capable() must be called on xhci-plat so that it covers
> > > both HS and SS root hubs
> > 
> > Thanks for pointing that out!
> > 
> > > I am thinking of dynamically enabling/disabling xhci-plat wakeup capability so
> > > that the wakeup path is correctly propagated to dwc3. something like below.
> > > Does it make sense to you?
> > > 
> > > diff --git a/drivers/usb/host/xhci-plat.c b/drivers/usb/host/xhci-plat.c
> > > index 649ffd8..be0c55b 100644
> > > --- a/drivers/usb/host/xhci-plat.c
> > > +++ b/drivers/usb/host/xhci-plat.c
> > > @@ -412,6 +412,9 @@ static int __maybe_unused xhci_plat_suspend(struct device *dev)
> > >  	struct xhci_hcd	*xhci = hcd_to_xhci(hcd);
> > >  	int ret;
> > >  
> > > +	if (!device_wakeup_path(dev))
> > > +		device_wakeup_disable(dev);
> > > +
> > >  	if (pm_runtime_suspended(dev))
> > >  		pm_runtime_resume(dev);
> > >  
> > > @@ -443,6 +446,8 @@ static int __maybe_unused xhci_plat_resume(struct device *dev)
> > >  	pm_runtime_set_active(dev);
> > >  	pm_runtime_enable(dev);
> > >  
> > > +	device_wakeup_enable(dev);
> > 
> > I think this also needs to be done conditionally, otherwise it would
> > create a new wake source on every resume when wakeup is already
> > enabled.
> > 
> Right, this needs to be done conditionally. However, there is a silent
> warning inside device_wakeup_enable() if it is called during system
> transition. Not sure if we really need to worry about that or not.

I guess it's up to the maintainers. Removing and adding the wakeup source on
suspend/resume is a bit of a hack, but it might be acceptable if it addresses
the issue and doesn't have negative side effects.