Hi Prashanth, On Fri, Sep 15, 2023 at 07:10:25PM +0530, Prashanth K wrote: > On 15-09-23 06:02 pm, Heikki Krogerus wrote: > > On Tue, Sep 12, 2023 at 04:37:47PM +0530, Prashanth K wrote: > > > > > > > > > On 11-09-23 06:19 pm, Heikki Krogerus wrote: > > > > On Mon, Sep 11, 2023 at 02:34:15PM +0530, Prashanth K wrote: > > > > > Currently if ucsi_send_command() fails, then we bail out without > > > > > clearing EVENT_PENDING flag. So when the next connector change > > > > > event comes, ucsi_connector_change() won't queue the con->work, > > > > > because of which none of the new events will be processed. > > > > > > > > > > Fix this by clearing EVENT_PENDING flag if ucsi_send_command() > > > > > fails. > > > > > > > > > > Cc: <stable@xxxxxxxxxxxxxxx> # 5.16 > > > > > Fixes: 512df95b9432 ("usb: typec: ucsi: Better fix for missing unplug events issue") > > > > > Signed-off-by: Prashanth K <quic_prashk@xxxxxxxxxxx> > > > > > --- > > > > > drivers/usb/typec/ucsi/ucsi.c | 1 + > > > > > 1 file changed, 1 insertion(+) > > > > > > > > > > diff --git a/drivers/usb/typec/ucsi/ucsi.c b/drivers/usb/typec/ucsi/ucsi.c > > > > > index c6dfe3d..509c67c 100644 > > > > > --- a/drivers/usb/typec/ucsi/ucsi.c > > > > > +++ b/drivers/usb/typec/ucsi/ucsi.c > > > > > @@ -884,6 +884,7 @@ static void ucsi_handle_connector_change(struct work_struct *work) > > > > > if (ret < 0) { > > > > > dev_err(ucsi->dev, "%s: GET_CONNECTOR_STATUS failed (%d)\n", > > > > > __func__, ret); > > > > > + clear_bit(EVENT_PENDING, &con->ucsi->flags); > > > > > goto out_unlock; > > > > > } > > > > > > > > I think it would be better to just move that label (out_unlock) above > > > > the point where clear_bit() is already called instead of separately > > > > calling it like that. That way the Connector Change Event will > > > > also get acknowledged. > > > Do we really need to ACK in this case since we didn't process the current > > > connector change event > > > > You won't get the next event before the first one was ACK'd, right? > > > > The spec says that we need to ACK if we received AND processed a CCI > > "4.5.4 Acknowledge Command Completion and/or Change Indication (R) > This command is used to acknowledge to the PPM that the OPM received and > processed a Command Completion and/or a Connector Change Indication." > > And here in this case, we have received, but not processed the event. > So I'm not really sure what to do here in this case. If we don't send an > ACK, then would the PPM think that OPM is not responding and reset it? > OR would it resend the previous event again since we didn't ACK? Every PPM behaves differently. Did you actually see that happening - GET_CONNECTOR_STATUS failed? Can you reproduce it? thanks, -- heikki