On Wed, 4 Dec 2019, Erkka Talvitie wrote:
> > > So if CERR == EHCI_TUNE_CERR and the QTD_PID != 1 (not IN) then we
> > > should return -EPIPE, as the existing code does. But if QTD_PID == 1
> > > then the code should continue, as your patch does -- with one
> > > difference: Put the additional check for EHCI_TUNE_CERR between the
> > > tests for DBE and XACT instead of after XACT (because XACT would
> > > decrement CERR whereas DBE wouldn't).
> >
> > Good point, I agree.
> >
> > >
> > > Can you make that change and test it?
> >
> > Sure, I have made the change and test it as soon as possible.
>
> I am not actually totally sure if I understood you correctly, but I tested a change where the first stall check is like this (PID_CODE_IN is defined as 1):
>
> - } else if (QTD_CERR(token)) {
> + } else if (QTD_CERR(token) && (QTD_PID (token) != PID_CODE_IN)) {
> status = -EPIPE;
>
> And the second stall check (now between DBE and XACT):
> + } else if (QTD_CERR(token)) {
> + status = -EPIPE;
>
> Is this what you meant? Please correct me if I am wrong.
Actually, what I meant for the first part was:
} else if (QTD_CERR(token) &&
(QTD_CERR(token) < EHCI_TUNE_CERR ||
QTD_PID(token) != PID_CODE_IN)) {
And of course there should be a comment before this line, explaining
what it does. By the way, the accepted format for multi-line comments
is:
/*
* Blah blah blah
* Blah blah blah
*/
The second part of the patch looks okay (but again, with a comment).
> Anyways with these changes the issue does not reproduce.
Very good.
Alan Stern
