RE: [PATCH v2 08/22] usb: chipidea: Remove locking in ci_udc_start()

[Peter Chen]

>Quoting Peter Chen (2016-08-06 00:54:35)
>> On Fri, Aug 05, 2016 at 02:53:56PM -0700, Stephen Boyd wrote:
>> > Quoting Peter Chen (2016-07-08 02:45:28)
>> > > On Thu, Jul 07, 2016 at 03:20:59PM -0700, Stephen Boyd wrote:
>> > > > We don't call hw_device_reset() with the ci->lock held, so it
>> > > > doesn't seem like this lock here is protecting anything. Let's
>> > > > just remove it. This allows us to call sleeping functions like
>> > > > phy_init() from within the CI_HDRC_CONTROLLER_RESET_EVENT hook.
>> > > >
>> > > > Cc: Peter Chen <peter.chen@xxxxxxx>
>> > > > Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
>> > > > Signed-off-by: Stephen Boyd <stephen.boyd@xxxxxxxxxx>
>> > > > ---
>> > > >  drivers/usb/chipidea/udc.c | 3 ---
>> > > >  1 file changed, 3 deletions(-)
>> > > >
>> > > > diff --git a/drivers/usb/chipidea/udc.c
>> > > > b/drivers/usb/chipidea/udc.c index 065f5d97aa67..f16be4710cdb
>> > > > 100644
>> > > > --- a/drivers/usb/chipidea/udc.c
>> > > > +++ b/drivers/usb/chipidea/udc.c
>> > > > @@ -1719,7 +1719,6 @@ static int ci_udc_start(struct usb_gadget *gadget,
>> > > >                        struct usb_gadget_driver *driver)  {
>> > > >       struct ci_hdrc *ci = container_of(gadget, struct ci_hdrc, gadget);
>> > > > -     unsigned long flags;
>> > > >       int retval = -ENOMEM;
>> > > >
>> > > >       if (driver->disconnect == NULL) @@ -1746,7 +1745,6 @@
>> > > > static int ci_udc_start(struct usb_gadget *gadget,
>> > > >
>> > > >       pm_runtime_get_sync(&ci->gadget.dev);
>> > > >       if (ci->vbus_active) {
>> > > > -             spin_lock_irqsave(&ci->lock, flags);
>> > > >               hw_device_reset(ci);
>> > > >       } else {
>> > > >               usb_udc_vbus_handler(&ci->gadget, false); @@
>> > > > -1755,7 +1753,6 @@ static int ci_udc_start(struct usb_gadget *gadget,
>> > > >       }
>> > > >
>> > > >       retval = hw_device_state(ci, ci->ep0out->qh.dma);
>> > > > -     spin_unlock_irqrestore(&ci->lock, flags);
>> > > >       if (retval)
>> > > >               pm_runtime_put_sync(&ci->gadget.dev);
>> > > >
>> > >
>> > > The main purpose for this is disabling interrupt when reset
>> > > controller, in case the unexpected interrupts occur.
>> > >
>> > > You can move this between hw_device_state.
>> > >
>> >
>> > Ok, but we don't hold the ci->lock in ci_udc_vbus_session(). Is that
>> > a bug as well?
>>
>> I agree with your patch. In fact, during the reset controller, the
>> interrupt has not been enabled, it should no unexpected interrupt.
>>
>
>So then we can leave this patch as is? It still isn't clear to me what sequence of
>events that would cause a problem if we don't hold the
>ci->lock around hw_device_state().

Yes. I am ok with this patch.  After thinking more, it is safe to be without lock.
And we have this unlock code in nxp internal tree, and without any problems
during 1-2 years.

Acked-by: Peter Chen <peter.chen@xxxxxxx>

Peter