Re: [PATCH v6 1/4] mfd: add support for Diolan DLN-2 devices

[Johan Hovold <johan@kernel.org>]

On Wed, Oct 08, 2014 at 01:54:07PM +0300, Octavian Purdila wrote:
> On Wed, Oct 8, 2014 at 12:23 PM, Johan Hovold <johan@kernel.org> wrote:
> > On Thu, Sep 25, 2014 at 07:07:31PM +0300, Octavian Purdila wrote:
> >
> >> +static void dln2_rx_transfer(struct dln2_dev *dln2, struct urb *urb,
> >> +                          u16 handle, u16 rx_slot)
> >> +{
> >> +     struct dln2_mod_rx_slots *rxs = &dln2->mod_rx_slots[handle];
> >> +     struct dln2_rx_context *rxc;
> >> +     struct device *dev = &dln2->interface->dev;
> >> +
> >> +     spin_lock(&rxs->lock);
> >
> > You must use spin_lock_irqsave here as you call it from the completion
> > handler.
> 
> Why? AFAICS the completion handler gets called from the HCD irq handler:

The completion handler is currently called with local interrupts
disabled but that is about to change once all drivers have been updated: 

	http://marc.info/?l=linux-usb&m=137353360511003&w=2

In this case you could probably get away with not disabling interrupts
anyway, but using the irqsave versions would make it obvious.

> >> +static void dln2_disconnect(struct usb_interface *interface)
> >> +{
> >> +     struct dln2_dev *dln2 = usb_get_intfdata(interface);
> >> +     int i, j;
> >> +
> >> +     /* don't allow starting new transfers */
> >> +     spin_lock(&dln2->disconnect_lock);
> >> +     dln2->disconnect = true;
> >> +     spin_unlock(&dln2->disconnect_lock);
> >> +
> >> +     /* cancel in progress transfers */
> >> +     for (i = 0; i < DLN2_HANDLES; i++) {
> >> +             struct dln2_mod_rx_slots *rxs = &dln2->mod_rx_slots[i];
> >> +             unsigned long flags;
> >> +
> >> +             spin_lock_irqsave(&rxs->lock, flags);
> >
> > Just stick to spin_lock in this function.
> 
> AFAICS disconnect is called from a kernel thread. Are there guarantees
> that we can't get a call to the completion routine while we are
> running it?

Brain fart, nevermind.

Johan