Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:3 as permitted sender) client-ip=2620:137:e000::3:3;
Date:   Fri, 15 Sep 2023 15:32:12 +0300
From:   Heikki Krogerus <heikki.krogerus@linux.intel.com>
To:     Prashanth K <quic_prashk@quicinc.com>
Cc:     Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        linux-usb@vger.kernel.org, linux-kernel@vger.kernel.org,
        "# 5 . 16" <stable@vger.kernel.org>
Subject: Re: usb: typec: ucsi: Clear EVENT_PENDING bit if ucsi_send_command
 fails
Message-ID: <ZQROzNqr7fbmJC87@kuha.fi.intel.com>
References: <1694423055-8440-1-git-send-email-quic_prashk@quicinc.com>
 <ZP8M6zqgsLTK25PI@kuha.fi.intel.com>
 <21d247d3-83be-ba53-c982-2ab0e2e4ffb3@quicinc.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <21d247d3-83be-ba53-c982-2ab0e2e4ffb3@quicinc.com>
Precedence: bulk

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@vger.kernel.org> # 5.16
> > > Fixes: 512df95b9432 ("usb: typec: ucsi: Better fix for missing unplug events issue")
> > > Signed-off-by: Prashanth K <quic_prashk@quicinc.com>
> > > ---
> > >   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?

> > 
> > If this really can happen, then I think it would be good to also
> > schedule a task for ucsi_check_connection():
> > 
> >          if (ret < 0) {
> >                  dev_err(ucsi->dev, "%s: GET_CONNECTOR_STATUS failed (%d)\n",
> >                          __func__, ret);
> > +               ucsi_partner_task(con, ucsi_check_connection, 1, HZ);
> >                  goto out_unlock;
> >          }
> > 
> > thanks,
> > 
> Retrying is a good idea, but ucsi_check_connection() doesn't have the full
> functionality compared to handle_connector_change. I guess
> ucsi_check_connection() will send a set_role, but won't handle the
> connector_change scenarios happening due to PR/DR swap, which will lead to
> deadlocks (due to wait_for_completion). This is just an example. So its
> better to bail out and process the next events, because the failure here is
> from the glink layer.

Fair enough.

-- 
heikki