On Fri, 7 Nov 2014, Sebastian Andrzej Siewior wrote:
> > Ah. In usb_resume(), what happened after usb_resume_both() returned?
> > The code says this:
> >
> > status = usb_resume_both(udev, msg);
> > if (status == 0) {
> > pm_runtime_disable(dev);
> > pm_runtime_set_active(dev);
> > pm_runtime_enable(dev);
> > unbind_marked_interfaces(udev);
> > }
> >
> > So if usb_resume_both() returns 0, the runtime PM status is supposed to
> > get changed to RPM_ACTIVE by the pm_runtime_set_active() call.
>
> so it returns 0 but pm_runtime_set_active() return with -EBUSY:
> __pm_runtime_set_status() {
> …
> if (parent) {
> if (… && parent->power.runtime_status != RPM_ACTIVE) {
> error = -EBUSY;
> …
> if (error)
> goto out;
> …
>
> The parent is musb-hdrc.1.auto. The sort order of resume is:
>
> musb-dsps 47401c00.usb: bus resume
> => dsps_resume() (glue layer)
> musb-hdrc musb-hdrc.1.auto: bus resume
> => musb_resume(2328)
> and after musb_resume() completes I see usb_resume() being invoked (and
> the first hub resume).
> Later, once khubd kicks in, I see musb_runtime_resume() probably via
> usb_autopm_get_interface(hub).
>
> That means the musb-hcd is not RPM_ACTIVE after resume and this is what
> USB expects.
Exactly.
> If I put pm_runtime_get_sync() + put in musb_resume() then the problem
> is gone. I don't see ehci-hcd doing this in resume, in fact I don't see
> ehci doing pm_runtime_* at all. So for ehci the device is probably
> always RPM_ACTIVE.
No, it doesn't always have to be RPM_ACTIVE.
Remember, the most common implementation of EHCI is the PCI version.
In drivers/pci/pci-driver.c, the pci_pm_suspend() routine calls
pm_runtime_resume(). Therefore a PCI EHCI controller is always in the
RPM_ACTIVE state _during a system sleep transition_. Maybe that's
what you meant.
EHCI controllers on other bus types might not behave so well. That
probably should be fixed in ehci_resume().
> And musb does it only in probe and nowhere else. I would expect it in
> the ISR as well but it is what it is.
> On the the other hand: musb_resume() invokes usb_hcd_resume_root_hub()
> what probably brings usb_resume() for the hub into the game. Shouldn't
> it pm_get_sync() the HCD?
No, usb_hcd_resume_root_hub() uses a freezable work queue. If it gets
called during system resume, nothing will happen until after the resume
is finished and the work queue starts running again.
> And still, if the HCD has a short suspend
> delay it might go into suspend before the khubd is invoked (but then it
> probably kills the status URB).
We need the musb device to go to the RPM_ACTIVE state at some point
during the system suspend/resume, and to remain that way until after
the root hub has been resumed. Putting the
pm_runtime_disable(dev);
pm_runtime_set_active(dev);
pm_runtime_enable(dev);
calls in musb_resume() is one way to accomplish this.
Alan Stern
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
