Re: [PATCH 2/2] gadget: Support for the usb charger framework

[Greg KH <gregkh@linuxfoundation.org>]

On Fri, Aug 07, 2015 at 05:22:47PM +0800, Baolin Wang wrote:
> On 7 August 2015 at 17:07, Peter Chen <Peter.Chen@freescale.com> wrote:
> >
> >> >>  /**
> >> >>   * struct usb_udc - describes one usb device controller @@ -127,12
> >> >> +128,45 @@ void usb_gadget_giveback_request(struct usb_ep *ep,  }
> >> >> EXPORT_SYMBOL_GPL(usb_gadget_giveback_request);
> >> >>
> >> >> +int usb_gadget_register_notify(struct usb_gadget *gadget,
> >> >> +                            struct notifier_block *nb) {
> >> >> +     unsigned long flags;
> >> >> +     int ret;
> >> >> +
> >> >> +     spin_lock_irqsave(&gadget->lock, flags);
> >> >
> >> > I find you use so many spin_lock_irqsave, any reasons for that?
> >> > Why mutex_lock can't be used?
> >> >
> >>
> >> The spin_lock_irqsave() can make it as a atomic notifier, that can make sure the
> >> gadget state event can be quickly reported to the user who register a notifier
> >> on the gadget device. Is it OK?
> >>
> >
> > I don't think it is a good reason, spin_lock_irqsave is usually used for protecting
> > data which is accessed at atomic environment.
> >
> 
> Yes, we want the notify process is a atomic environment which do not
> want to be interrupted by irq or other things to make the notice can
> be quickly reported to the user.

No, this is a "slow" event, you don't need to notify anyone under atomic
context, that's crazy.

> Also i think the notify process is less cost, thus i use the spinlock. Thanks.

No, use a mutex please, that's the safe thing.  This is not
time-critical code at all.

thanks,

greg k-h