Re: [PATCH] usb: echi-hcd: Add ehci_setup check before echi_shutdown

From: Alan Stern
Date: Fri May 20 2016 - 11:57:54 EST


On Fri, 20 May 2016, Andy Gross wrote:

> On 20 May 2016 at 09:31, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
> > On Thu, 19 May 2016, Andy Gross wrote:
> >
> >> On 19 May 2016 at 05:12, Srinivas Kandagatla
> >> <srinivas.kandagatla@xxxxxxxxxx> wrote:
> >>
> >> <snip>
> >>
> >> > +++ b/drivers/usb/host/ehci-hcd.c
> >> > @@ -368,6 +368,15 @@ static void ehci_shutdown(struct usb_hcd *hcd)
> >> > {
> >> > struct ehci_hcd *ehci = hcd_to_ehci(hcd);
> >> >
> >> > + /**
> >> > + * Protect the system from crashing at system shutdown in cases where
> >> > + * usb host is not added yet from OTG controller driver.
> >> > + * As ehci_setup() not done yet, so stop accessing registers or
> >> > + * variables initialized in ehci_setup()
> >> > + */
> >> > + if (!ehci->sbrn)
> >> > + return;
> >> > +
> >> > spin_lock_irq(&ehci->lock);
> >> > ehci->shutdown = true;
> >> > ehci->rh_state = EHCI_RH_STOPPING;
> >>
> >>
> >> Should we also not check this in ehci_suspend and ehci_resume? If I
> >> do suspend/resume, it crashes in a similar manner. I know this goes
> >> beyond the problem scope of the patch. Perhaps I should send in an
> >> additional patch with the code?
> >>
> >> Are there any other places in the ehci-hcd where this needs to be checked?
> >
> > Really, this is an indication that the problem lies not in ehci-hcd but
> > rather in the platform glue code. The platform code allows itself to
> > be registered as the driver of the controller device but does not
> > initialize the ehci-hcd parts.
> >
> > Neither ehci_suspend nor ehci_resume is called directly by the driver
> > core; the calls all have to pass through the glue code. So maybe the
> > glue layer needs to be careful about invoking these routines in
> > ehci-hcd before ehci-hcd has been initialized.
>
> That's a good point. Perhaps all of this should be in the glue,
> including the shutdown case.

Many of the platform glue files set usb_hcd_platform_shutdown() as
their shutdown callback, which makes it impossible for them to include
a controller-specific test.

Alan Stern