HI, On Fri, Sep 07, 2012 at 10:03:45AM +0530, Pratyush Anand wrote: > We are clearing resource_index when we issue END TRANSFER command. which is just fine, actually... > Consider a situation, when we have executed END TRANSFER, core has > generated xfernotready, but gadget is too slow and have not queued > any request. So we set PENDING flag. Now application queues another > request. Software does not have notion for current microframe to submit > isoc TRBs. So, it will issue END TRANSFER again. If we have forced > resource_index to zero during first END TRANSFER, then second END > TRANSFER will be issued with wrong resource_index (0), which is not > correct and confusing for core. I have observed that core does generate > xfernotready even if END TRANSFER is issued with resource_index 0, but > then it is not able to start transfer again. then you should just clear PENDING flag, right ? > Moreover I have observed that START TRANSFER returns a new > resource_index only if SET XFER RESOURCE is issued again. > > Signed-off-by: Pratyush Anand <pratyush.anand@xxxxxx> > --- > drivers/usb/dwc3/gadget.c | 5 +---- > 1 files changed, 1 insertions(+), 4 deletions(-) > > diff --git a/drivers/usb/dwc3/gadget.c b/drivers/usb/dwc3/gadget.c > index f13daa2..0f6874c 100644 > --- a/drivers/usb/dwc3/gadget.c > +++ b/drivers/usb/dwc3/gadget.c > @@ -530,6 +530,7 @@ static int __dwc3_gadget_ep_enable(struct dwc3_ep *dep, > if (ret) > return ret; > > + dep->resource_index = 0; > dep->endpoint.desc = desc; > dep->comp_desc = comp_desc; > dep->type = usb_endpoint_type(desc); > @@ -1919,16 +1920,12 @@ static void dwc3_stop_active_transfer(struct dwc3 *dwc, u32 epnum) > > dep = dwc->eps[epnum]; > > - if (!dep->resource_index) > - return; the thing is that this will let us fall into a situation where we issue END_TRANSFER even when resource_index is zero, which isn't good at all. -- balbi
Attachment:
signature.asc
Description: Digital signature
