Hi Paul, good to get a mail from you :-)
On Tue, Sep 15, 2015 at 04:31:03PM -0700, pauldzim . wrote:
> It's dangerous to use Update Transfer when the DWC3_TRB_CTRL_LST bit
> might be set in one of the TRBs. The reason is, there is a window
> between checking that the transfer has not completed
On Wed, Sep 16, 2015 at 09:17:20AM -0500, Felipe Balbi wrote:
> Hi Paul, good to get a mail from you :-)
>
> On Tue, Sep 15, 2015 at 04:31:03PM -0700, pauldzim . wrote:
> > It's dangerous to use Update Transfer when the DWC3_TRB_CTRL_LST bit
> > might be set in one of the TRBs. The reason is,
Hi Felipe,
It's dangerous to use Update Transfer when the DWC3_TRB_CTRL_LST bit
might be set in one of the TRBs. The reason is, there is a window
between checking that the transfer has not completed yet and issuing
the Update Transfer command. If the hardware happens to complete the
transfer and
We can infer Update Transfer by the fact that
req_queue is empty and DWC3_EP_BUSY isn't set.
This let's us a) rely on Update Transfer more often
(should be good for deeper queue lengths) and b) remove
the extra start_new parameter (done on a follow-up
patch)
Signed-off-by: Felipe Balbi