Re: [PATCH v3 01/11] usb: phy: Add APIs for runtime power management

From: Felipe Balbi
Date: Wed Apr 03 2013 - 09:54:43 EST


HI,

On Wed, Apr 03, 2013 at 06:42:48PM +0530, Vivek Gautam wrote:
> >> >> Adding APIs to handle runtime power management on PHY
> >> >> devices. PHY consumers may need to wake-up/suspend PHYs
> >> >> when they work across autosuspend.
> >> >>
> >> >> Signed-off-by: Vivek Gautam <gautam.vivek@xxxxxxxxxxx>
> >> >> ---
> >> >> include/linux/usb/phy.h | 141
> >> >> +++++++++++++++++++++++++++++++++++++++++++++++
> >> >> 1 files changed, 141 insertions(+), 0 deletions(-)
> >> >>
> >> >> diff --git a/include/linux/usb/phy.h b/include/linux/usb/phy.h
> >> >> index 6b5978f..01bf9c1 100644
> >> >> --- a/include/linux/usb/phy.h
> >> >> +++ b/include/linux/usb/phy.h
> >> >> @@ -297,4 +297,145 @@ static inline const char *usb_phy_type_string(enum
> >> >> usb_phy_type type)
> >> >> return "UNKNOWN PHY TYPE";
> >> >> }
> >> >> }
> >> >> +
> >> >> +static inline void usb_phy_autopm_enable(struct usb_phy *x)
> >> >> +{
> >> >> + if (!x || !x->dev) {
> >> >> + dev_err(x->dev, "no PHY or attached device available\n");
> >> >> + return;
> >> >> + }
> >> >> +
> >> >> + pm_runtime_enable(x->dev);
> >> >> +}
> >> >
> >> >
> >> > IMO we need not have wrapper APIs for runtime_enable and runtime_disable
> >> > here. Generally runtime_enable and runtime_disable is done in probe and
> >> > remove of a driver respectively. So it's better to leave the
> >> > runtime_enable/runtime_disable to be done in *phy provider* driver than
> >> > having an API for it to be done by *phy user* driver. Felipe, what do you
> >> > think?
> >>
> >> Thanks!!
> >> That's very true, runtime_enable() and runtime_disable() calls are made by
> >> *phy_provider* only. But a querry here.
> >> Wouldn't in any case a PHY consumer might want to disable runtime_pm on PHY ?
> >> Say, when consumer failed to suspend the PHY properly
> >> (*put_sync(phy->dev)* fails), how much sure is the consumer about the
> >> state of PHY ?
> >
> > no no, wait a minute. We might not want to enable runtime pm for the PHY
> > until the UDC says it can handle runtime pm, no ? I guess this makes a
> > bit of sense (at least in my head :-p).
> >
> > Imagine if PHY is runtime suspended but e.g. DWC3 isn't runtime pm
> > enabled... Does it make sense to leave that control to the USB
> > controller drivers ?
> >
> > I'm open for suggestions
>
> Of course unless the PHY consumer can handle runtime PM for PHY,
> PHY should not ideally be going into runtime_suspend.
>
> Actually trying out few things, here are my observations
>
> Enabling runtime_pm on PHY pushes PHY to go into runtime_suspend state.
> But a device detection wakes up DWC3 controller, and if i don't wake
> up PHY (using get_sync(phy->dev)) here
> in runtime_resume() callback of DWC3, i don't get PHY back in active state.
> So it becomes the duty of DWC3 controller to handle PHY's sleep and wake-up.
> Thereby it becomes logical that DWC3 controller has the right to
> enable runtime_pm
> of PHY.
>
> But there's a catch here. if there are multiple consumers of PHY (like
> USB2 type PHY can
> have DWC3 controller as well as EHCI/OHCI or even HSGadget) then in that case,
> only one of the consumer can enable runtime_pm on PHY. So who decides this.
>
> Aargh!! lot of confusion here :-(


hmmm, maybe add a flag to struct usb_phy and check it on
usb_phy_autopm_enable() ??

How does usbcore handle it ? They request class drivers to pass
supports_autosuspend, but while we should have a similar flag, that's
not enough. We also need a flag to tell us when pm_runtime has already
been enabled.

So how about:

usb_phy_autopm_enable()
{
if (!phy->suports_autosuspend)
return -ENOSYS;

if (phy->autosuspend_enabled)
return 0;

phy->autosuspend_enabled = true;
return pm_runtime_enable(phy->dev);
}

???

--
balbi

Attachment: signature.asc
Description: Digital signature