It's a good point. It looks like we'll need to refactor the code quite a bit to change the way qh deallocation gets done and prevent it before we end any pending Clear_TT_Buffer requests. Right now qh gets deallocated at a few different places. This means that once in the complete callback, our clearing_tt flag or the like doesn`t exist anymore. Any though on this that might help narrow down the scope of the modifications? François Gervais > -----Original Message----- > From: Alan Stern [mailto:stern@xxxxxxxxxxxxxxxxxxx] > Sent: November 1, 2017 2:48 PM > To: Gervais, Francois <FGervais@xxxxxxxxxxxxxxxxxxxx> > Cc: linux-usb@xxxxxxxxxxxxxxx > Subject: Re: musb: Clear_TT_Buffer implementation > > Please tell your email client to fold lines after 72 columns or so. > > On Wed, 1 Nov 2017, Gervais, Francois wrote: > > > We are running the musb driver on a AM335 cpu paired with a microchip > hub (USB2412). > > > > In one of our use-cases, we have a high chance of having an error on a bulk > split transaction. Since the musb driver does not implement the clear TT > buffer it leaves us with a TT buffer full on the hub which in turns NAK any > new split transactions. > > > > To fix this problem, we added the Clear_TT_Buffer handling in case of a > split transaction error, trying to match the spec as closely as possible. > > > > We came up with the following: > > > > diff --git a/drivers/usb/musb/musb_host.c > > b/drivers/usb/musb/musb_host.c index 26bda88..ac1a959 100755 > > --- a/drivers/usb/musb/musb_host.c > > +++ b/drivers/usb/musb/musb_host.c > > @@ -376,6 +376,11 @@ static void musb_advance_schedule(struct musb > *musb, struct urb *urb, > > break; > > } > > > > + if ((status || urb->unlinked) && > > + (usb_pipecontrol(urb->pipe) || usb_pipebulk(urb->pipe)) && > > + urb->dev->tt) > > + usb_hub_clear_tt_buffer(urb); > > + > > qh->is_ready = 0; > > musb_giveback(musb, urb, status); > > qh->is_ready = ready; > > > > This seems to work in our case but we'd like to get feedback in order to > improve the patch if needed. > > The patch is incomplete. You need to register a completion routine, as the > .clear_tt_buffer_complete member of musb_hc_driver. This routine will be > called after the Clear-TT-Buffer request has completed. Until that happens, > you must not try to use the TT. > > For an example, see ehci_clear_tt_buffer_complete() in > drivers/usb/host/ehci-q.c. > > Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
