Re: [PATCH v2 1/2] usb: dwc3: gadget: Bail out in pullup if soft reset timeout happens

[Thinh Nguyen]

On Thu, Mar 30, 2023, Krishna Kurapati PSSNV wrote:
> 
> 
> On 3/30/2023 5:40 AM, Thinh Nguyen wrote:
> > On Wed, Mar 29, 2023, Krishna Kurapati PSSNV wrote:
> > > 
> > > 
> > > On 3/29/2023 2:50 AM, Thinh Nguyen wrote:
> > > > Hi,
> > > > 
> > > > On Tue, Mar 28, 2023, Krishna Kurapati wrote:
> > > > > If the core soft reset timeout happens, avoid setting up event
> > > > > buffers and starting gadget as the writes to these registers
> > > > > may not reflect when in reset and setting the run stop bit
> > > > > can lead the controller to access wrong event buffer address
> > > > > resulting in a crash.
> > > > > 
> > > > > Signed-off-by: Krishna Kurapati <quic_kriskura@xxxxxxxxxxx>
> > > > > ---
> > > > >    drivers/usb/dwc3/gadget.c | 5 ++++-
> > > > >    1 file changed, 4 insertions(+), 1 deletion(-)
> > > > > 
> > > > > diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c
> > > > > index 3c63fa97a680..f0472801d9a5 100644
> > > > > --- a/drivers/usb/dwc3/gadget.c
> > > > > +++ b/drivers/usb/dwc3/gadget.c
> > > > > @@ -2620,13 +2620,16 @@ static int dwc3_gadget_pullup(struct usb_gadget *g, int is_on)
> > > > >    		 * device-initiated disconnect requires a core soft reset
> > > > >    		 * (DCTL.CSftRst) before enabling the run/stop bit.
> > > > >    		 */
> > > > > -		dwc3_core_soft_reset(dwc);
> > > > > +		ret = dwc3_core_soft_reset(dwc);
> > > > > +		if (ret)
> > > > > +			goto done;
> > > > >    		dwc3_event_buffers_setup(dwc);
> > > > >    		__dwc3_gadget_start(dwc);
> > > > >    		ret = dwc3_gadget_run_stop(dwc, true, false);
> > > > >    	}
> > > > > +done:
> > > > >    	pm_runtime_put(dwc->dev);
> > > > >    	return ret;
> > > > > -- 
> > > > > 2.40.0
> > > > > 
> > > > 
> > > > I think there's one more place that may needs this check. Can you also
> > > > add this check in __dwc3_set_mode()?
> > > 
> > > Hi Thinh,
> > > 
> > >    Sure. Will do it.
> > > Will the below be good enough ? Or would it be good to add an error/warn log
> > > there>
> > 
> > There's already a warning message in dwc3_core_soft_reset() if it fails.
> > 
> > > 
> > > kriskura@hu-kriskura-hyd:/local/mnt/workspace/krishna/skales2/skales/kernel$
> > > git diff drivers/usb/
> > > diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c
> > > index 476b63618511..8d1d213d1dcd 100644
> > > --- a/drivers/usb/dwc3/core.c
> > > +++ b/drivers/usb/dwc3/core.c
> > > @@ -210,7 +210,9 @@ static void __dwc3_set_mode(struct work_struct *work)
> > >                  }
> > >                  break;
> > >          case DWC3_GCTL_PRTCAP_DEVICE:
> > > -               dwc3_core_soft_reset(dwc);
> > > +               ret = dwc3_core_soft_reset(dwc);
> > > +               if (ret)
> > > +                       goto out;
> > > 
> > >                  dwc3_event_buffers_setup(dwc);
> > > 
> > 
> > If soft-reset failed, the controller is in a bad state. We should not
> > perform any further operation until the next hard reset. We should flag
> > the controller as dead. I don't think we have the equivalent of the
> > host's HCD_FLAG_DEAD. It may require some work in the UDC core. Perhaps
> > we can flag within dwc3 for now and prevent any further operation for a
> > simpler fix.
> > 
> Hi Thinh,
> 
>  Are you referring that if __dwc3_set_mode failed with core soft reset
> timing out, the caller i.e., dwc3_set_mode who queues the work need to know
> that the operation actually failed. So we can add a flag to indicate that
> gadget is dead and the caller of dwc3_set_mode can check the flag to see if
> the operation is successful or not.
> 
> Or am I misunderstanding your comment ?
> 

Not just in __dwc3_set_mode(). I mean any time dwc3_core_soft_reset
fails, then we set this flag. So that it can prevent the user calling
any gadget ops causing more crashes/invalid behavior. The
dwc->softconnect is already wrong on pullup() on failure.

So that we can have a check in different gadget ops. For pullup():

static int dwc3_gadget_pullup() {
	if (dwc->udc_is_dead) {
		dev_err(dev, "reset me. x_x \n");
		return;
	}

	abc();
}

Perhaps the effort is probably the same if we enhance the UDC core for
this? In any case, I'm fine either way.

Thanks,
Thinh