On Fri, 5 Jul 2019, Benjamin Herrenschmidt wrote: > On Fri, 2019-07-05 at 07:44 +1000, Benjamin Herrenschmidt wrote: > > > > - At some point I had code to reject EP queue() if the device > > > > is > > > > suspended with -ESHUTDOWN. The end result was bad ... > > > > f_mass_storage > > > > goes into an infinite loop of trying to queue the same stuff in > > > > start_out_transfer() when that happens. It looks like it's not > > > > really > > > > handling errors from queue() in a particularily useful way. > > > > > > Don't reject EP queue requests. Accept them as you would at any > > > time; > > > they will complete after the port is resumed. > > > > Except the suspend on a bus reset clears the port enable. You can't > > resume from that, only reset the port no ? Or am I missing something > > ? > > Talking of which... do we need this ? > > --- a/drivers/usb/gadget/composite.c > +++ b/drivers/usb/gadget/composite.c > @@ -1976,6 +1976,7 @@ void composite_disconnect(struct usb_gadget *gadget) > * disconnect callbacks? > */ > spin_lock_irqsave(&cdev->lock, flags); > + cdev->suspended = 0; > if (cdev->config) > reset_config(cdev); > if (cdev->driver->disconnect) > > Otherwise with my vhub or with dummy_hcd, a suspend followed by a reset > will keep that stale suspended flag to 1 (which has no effect at the moment > but still...) > > If yes, I'll submit a patch accordingly...
According to the USB spec, a host is not supposed to reset a suspended port (it's supposed to resume the port and then do the reset). But of course this can happen anyway, so we should handle it properly. Alan Stern
