Re: [PATCH] usb: core: Add runtime resume checking

[Peter Chen]

On Thu, Aug 11, 2016 at 04:07:21PM +0800, Baolin Wang wrote:
> Hi Peter,
> 
> On 11 August 2016 at 14:54, Peter Chen <hzpeterchen@xxxxxxxxx> wrote:
> > On Thu, Aug 11, 2016 at 11:08:52AM +0800, Baolin Wang wrote:
> >> Hi Alan,
> >>
> >> On 10 August 2016 at 22:25, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
> >> > On Wed, 10 Aug 2016, Baolin Wang wrote:
> >> >
> >> >> >> >> For example: No slave attached----> usb interface runtime suspend
> >> >> >> >> ----> usb device runtime suspend -----> xhci suspend -----> power off
> >> >> >> >> usb controller. After that if the system wants to enter suspend state,
> >> >> >> >> then it also will issue usb_dev_suspend(), then the
> >> >> >> >> pm_runtime_resume() function (issued in choose_wakeup() function) will
> >> >> >> >> return -ESHUTDOWN due to xhci has been suspend and hardware is not
> >> >> >> >> accessible.
> >> >> >> >
> >> >> >> > Why the controller does not be resumed when the root hub has issued
> >> >> >> > runtime resume?
> >> >> >>
> >> >> >> Because now there is no slave attached, we will not power on usb
> >> >> >> controller and resume xhci.
> >> >> >>
> >> >> >
> >> >> > I don't know why you set policy like this. Even the controller is
> >> >> > resumed, it will still be suspended by system suspend. What's more,
> >> >> > if you disallow it, how can you change your wakeup setting?
> >> >> > Eg, at runtime suspend, we enable wakeup by default. But for system
> >> >> > sleep, we disable wakeup by default.
> >> >> >
> >> >> > At runtime, if there is no device on the port, even we resumes the
> >> >> > controller and roohub, it still will be suspended soon.
> >> >>
> >> >> For saving power, if there is no slave attached, why we need to power
> >> >> on usb controller and resume xhci? Especially for mobile device.
> >> >
> >> > You need to power-on the USB controller in order to change the wakeup
> >> > setting.
> >>
> >> pm_runtime_resume() function issued in choose_wakeup() can just resume
> >> usb device, which can not power-on the USB controller and resume xHCI.
> >> Cause the user won't know you need to power-on usb controller now,
> >> then how to power-on the USB controller when changing wakeup setting?
> >>
> >> >
> >> >> Moreover the usb device runtime suspend/resume is separate with xhci
> >> >> suspend/resume, when xhci entered suspend state, only slave attaching
> >> >> can resume xhci.
> >> >
> >> > That's right.  Suppose the user wants the system to stay asleep when a
> >> > slave attaches.  But when you suspended the xHCI controller, it was set
> >> > to wake up when a new slave attaches.  So when a slave is attached, the
> >> > controller will wake up the system.  That's bad.
> >> >
> >> > If the wakeup setting needs to be changed when the system suspend
> >> > begins, then you have to power-on the controller to make that change.
> >> >
> >> > Given a choice between using a little power or behaving incorrectly,
> >> > you should choose to use the power.
> >>
> >> OK. But that is a real problem. It will pm_runtime_resume() falied
> >> (issued in choose_wakeup()), cause usb controller has powered-off and
> >> xHCI controller has suspended and we have no method to notify the user
> >> to power-on USB controller. Any good suggestion to solve this, Alan
> >> and Peter? Thanks.
> >>
> >
> > Maybe you can show us the call stack why pm_runtime_resume has failed
> > at your environment.
> 
> OK. I try to explain it at below:
> 
> For example: No slave attached----> usb interface runtime suspend
> ----> usb device runtime suspend (routine is:
> usb_suspend_device()--->generic_suspend() --> hcd_bus_suspend()-->
> xhci_bus_suspend()) -----> xhci_ suspend() -----> power off usb
> controller. After that if the system wants to enter suspend state,
> then it also will issue usb_dev_suspend(), then the
> pm_runtime_resume() function (issued in choose_wakeup() function) will
> return -ESHUTDOWN due to xhci has been suspend and hardware is not accessible.
> 
> We issue the pm_runtime_resume() function routine: usb_resume_device()
> ----> generic_resume() ----> hcd_bus_resume() ---> xhci_bus_resume(),
> but now xHCI is not accessible due to xhci_suspend() is issued and USB
> controller is power off when no slave attached. That is why
> pm_runtime_resume failed.
> 

What host controller driver you are using? Assume you are using
xhci-plat.c. The problem for you is this driver does not implement
runtime pm operations, so the flag HCD_FLAG_HW_ACCESSIBLE is not set.
Maybe Robert's patch is a good start for you [1]

[1] http://www.spinics.net/lists/linux-usb/msg144602.html

-- 

Best Regards,
Peter Chen