Hi, On 2/28/23 11:05, Heikki Krogerus wrote: > On Tue, Feb 28, 2023 at 10:03:03AM +0100, Hans de Goede wrote: >> When ucsi_init() fails, ucsi->connector is NULL, yet in case of >> ucsi_acpi we may still get events which cause the ucs_acpi code to call >> ucsi_connector_change(), which then derefs the NULL ucsi->connector >> pointer. >> >> Fix this by adding a check for ucsi->connector being NULL, as is >> already done in ucsi_resume() for similar reasons. >> >> Fixes: bdc62f2bae8f ("usb: typec: ucsi: Simplified registration and I/O API") >> Cc: stable@xxxxxxxxxxxxxxx >> Signed-off-by: Hans de Goede <hdegoede@xxxxxxxxxx> >> --- >> drivers/usb/typec/ucsi/ucsi.c | 8 +++++++- >> 1 file changed, 7 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/usb/typec/ucsi/ucsi.c b/drivers/usb/typec/ucsi/ucsi.c >> index 1cf8947c6d66..e762897cb25a 100644 >> --- a/drivers/usb/typec/ucsi/ucsi.c >> +++ b/drivers/usb/typec/ucsi/ucsi.c >> @@ -842,7 +842,13 @@ static void ucsi_handle_connector_change(struct work_struct *work) >> */ >> void ucsi_connector_change(struct ucsi *ucsi, u8 num) >> { >> - struct ucsi_connector *con = &ucsi->connector[num - 1]; >> + struct ucsi_connector *con; >> + >> + /* Check for ucsi_init() failure */ >> + if (!ucsi->connector) >> + return; >> + >> + con = &ucsi->connector[num - 1]; >> >> if (!(ucsi->ntfy & UCSI_ENABLE_NTFY_CONNECTOR_CHANGE)) { >> dev_dbg(ucsi->dev, "Bogus connector change event\n"); > > I think we should try to rely on that ucsi->ntfy. Would this work: > > diff --git a/drivers/usb/typec/ucsi/ucsi.c b/drivers/usb/typec/ucsi/ucsi.c > index fe1963e328378..0da1e9c66971a 100644 > --- a/drivers/usb/typec/ucsi/ucsi.c > +++ b/drivers/usb/typec/ucsi/ucsi.c > @@ -928,15 +928,13 @@ static void ucsi_handle_connector_change(struct work_struct *work) > */ > void ucsi_connector_change(struct ucsi *ucsi, u8 num) > { > - struct ucsi_connector *con = &ucsi->connector[num - 1]; > - > if (!(ucsi->ntfy & UCSI_ENABLE_NTFY_CONNECTOR_CHANGE)) { > dev_dbg(ucsi->dev, "Bogus connector change event\n"); > return; > } > > if (!test_and_set_bit(EVENT_PENDING, &ucsi->flags)) > - schedule_work(&con->work); > + schedule_work(&ucsi->connector[num - 1].work); > } > EXPORT_SYMBOL_GPL(ucsi_connector_change); > This hunk is not necessary, the con pointer pointing to lala land is not an issue as long as we don't deref it. The &ucsi->connector[num - 1]; does not deref ucsi->connector it it simply adds an offset to it and stores that in con (the backtrace I got pointed to the schedule_work call). But I guess your way does make it more obvious that we don't deref ucsi->connector. > @@ -1404,6 +1402,7 @@ static int ucsi_init(struct ucsi *ucsi) > ucsi->connector = NULL; > > err_reset: > + ucsi->ntfy = 0; > memset(&ucsi->cap, 0, sizeof(ucsi->cap)); > ucsi_reset_ppm(ucsi); > err: In would expect this to fix things, but I only have access to the monitor triggering this on Mondays, so I can only 100% confirm next Monday. Note this does open the race I try to fix in patch 2/3 again. So what should be done here is to make ntfy a local variable and only store it in ucsi->ntfy on success. Regards, Hans