Hi, On Wed, Mar 27, 2024 at 12:39:04PM +0000, Diogo Ivo wrote: > The UCSI specification mentions that when the OPM is notified by the > PPM of an asynchronous event it should first send all the commands it > needs to get the details of the event and only after acknowledge it by > sending ACK_CC_CI with the Connector Change bit set, while the current > code sends this ack immediately after scheduling all the partner_tasks. > Move this ACK_CC_CI command to the end of all partner_tasks. > > This fixes a problem with some LG Gram laptops where the PPM sometimes > notifies the OPM with the Connector Change Indicator field set in the > CCI after an ACK_CC_CI command is sent, causing the UCSI stack to check > the connector status indefinitely since the EVENT_PENDING bit is already > cleared. This causes an interrupt storm and an artificial high load on > these platforms. > > It would also be interesting to see if we could take this approach and > implement the discussion in [1] regarding sending an ACK_CC_CI command > that acks both the command completion and the connector change. > > Signed-off-by: Diogo Ivo <diogo.ivo@xxxxxxxxxxxxxxxxxx> > > [1]: https://lore.kernel.org/all/20240320073927.1641788-1-lk@xxxxxxx/ > --- > drivers/usb/typec/ucsi/ucsi.c | 27 ++++++++++++++++++++------- > 1 file changed, 20 insertions(+), 7 deletions(-) > > diff --git a/drivers/usb/typec/ucsi/ucsi.c b/drivers/usb/typec/ucsi/ucsi.c > index 0c8f3b3a99d6..b8b39e43aba8 100644 > --- a/drivers/usb/typec/ucsi/ucsi.c > +++ b/drivers/usb/typec/ucsi/ucsi.c > @@ -69,6 +69,23 @@ static int ucsi_acknowledge_connector_change(struct ucsi *ucsi) > return ucsi->ops->sync_write(ucsi, UCSI_CONTROL, &ctrl, sizeof(ctrl)); > } > > +static void ucsi_handle_ack_connector_change(struct ucsi_connector *con) > +{ > + struct ucsi *ucsi = con->ucsi; > + int ret; > + > + if (list_empty(&con->partner_tasks)) { > + mutex_lock(&ucsi->ppm_lock); > + ret = ucsi_acknowledge_connector_change(ucsi); > + mutex_unlock(&ucsi->ppm_lock); > + > + if (ret) > + dev_err(ucsi->dev, "%s: ACK failed (%d)", __func__, ret); What if a real async connector change event happens here? It can because you just cleared the connector change condition. But it will be ignored because EVENT_PENDING is still set. In practive the new event might even be reported in the CCI along with the completion of the ACK command above (without an additional async event). So the PPM now thinks it reported the event to us and will not send an async event again. But we will never check for it because we ignored it and the UCSI will be stuck forever. What UCSI backend (ACPI, CCG, ...) is this? > + > + clear_bit(EVENT_PENDING, &ucsi->flags); > + } > +} > + > static int ucsi_exec_command(struct ucsi *ucsi, u64 command); regards Christian
