On 20-11-30 02:37:54, Jun Li wrote: > Hi Peter, > > -----Original Message----- > > From: Peter Chen <peter.chen@xxxxxxxxxx> > > Sent: Friday, November 27, 2020 2:28 PM > > To: heikki.krogerus@xxxxxxxxxxxxxxx > > Cc: linux-usb@xxxxxxxxxxxxxxx; dl-linux-imx <linux-imx@xxxxxxx>; Peter > > Chen <peter.chen@xxxxxxx>; Jun Li <jun.li@xxxxxxx> > > Subject: [PATCH 1/1] usb: roles: reference controller's parent device if > > existed > > > > From: Peter Chen <peter.chen@xxxxxxx> > > > > For some DRD IP drivers (eg, dwc3/cdns3/chipidea), the core device is created > > and deleted by glue layer device. So, if role switch user (eg, tcpci), core > > device, and glue layer device are all built as module, and glue layer device > > is removed first, the core device's driver ->remove function will be called, > > and its device's driver pointer will be NULL, and cause below oops. > > > > To fix it, if there is a parent for controller device (role switch device's > > parent), it references to parent too. > > This may not enough, the trouble is role switch class isn't aware of > the role switch is unregistered by its provider while typec class is > using it, this can happen by different ways, if we do role switch provider > driver unbind, we still suffer this problem: > > echo usb_controller_dev > unbind > Yes, to fix this issue totally, we need to change the way role switch class used by controller and type-c driver, until now, we still have not a good way. We could fix module unload issue first, unbind two devices may not a common use case. > I am not sure if we should prevent this happening at USB controller > drivers(provider side), or this is allowed and we should enhance > role switch class to be aware of this situation and properly handle it. > -- Thanks, Peter Chen
