| Message ID | 20230208072141.1103738-1-kai.heng.feng@canonical.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | [v2] iio: light: cm32181: Unregister second I2C client if present | expand |
Hi, Thank you for working on this, some remarks inline. On 2/8/23 08:21, Kai-Heng Feng wrote: > If a second client that talks to the actual I2C address was created in > probe(), there should be a corresponding cleanup in remove() to avoid > leakage. > > So if the "client" is not the same one used by I2C core, unregister it > accordingly. > > Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=2152281 > Fixes: c1e62062ff54 ("iio: light: cm32181: Handle CM3218 ACPI devices with 2 I2C resources") > Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com> > --- > v2: > - Use devm_add_action_or_reset() instead of remove() callback to avoid > race. > > drivers/iio/light/cm32181.c | 16 ++++++++++++++++ > 1 file changed, 16 insertions(+) > > diff --git a/drivers/iio/light/cm32181.c b/drivers/iio/light/cm32181.c > index b1674a5bfa368..a3e5f56101c9f 100644 > --- a/drivers/iio/light/cm32181.c > +++ b/drivers/iio/light/cm32181.c > @@ -429,6 +429,16 @@ static const struct iio_info cm32181_info = { > .attrs = &cm32181_attribute_group, > }; > > +static void cm32181_disable(void *data) > +{ > + struct i2c_client *client = data; > + struct cm32181_chip *cm32181 = iio_priv(i2c_get_clientdata(client)); > + > + /* Unregister the dummy client */ > + if (cm32181->client != client) > + i2c_unregister_device(cm32181->client); > +} > + > static int cm32181_probe(struct i2c_client *client) > { > struct device *dev = &client->dev; > @@ -479,6 +489,12 @@ static int cm32181_probe(struct i2c_client *client) > return ret; > } > > + ret = devm_add_action_or_reset(dev, cm32181_disable, client); > + if (ret) { > + dev_err(dev, "%s: add devres action failed\n", __func__); > + return ret; > + } > + This is too late, we will still exit without unregistering the client if the cm32181_reg_init() call fails. It would be best to do this directly after the i2c_acpi_new_device() call, so inside the "if (ACPI_HANDLE(dev) && client->addr == SMBUS_ALERT_RESPONSE_ADDRESS) {" block. This way you can also remove the "if (cm32181->client != client)" check from cm32181_disable() since it now only runs when the client was registered in the first place. Also please rename cm32181_disable() to cm32181_unregister_dummy_client() so that the name actually matches what it does. Regards, Hans > ret = devm_iio_device_register(dev, indio_dev); > if (ret) { > dev_err(dev, "%s: regist device failed\n", __func__);
Hi Hans, On Mon, Feb 13, 2023 at 7:17 PM Hans de Goede <hdegoede@redhat.com> wrote: > > Hi, > > Thank you for working on this, some remarks inline. > > On 2/8/23 08:21, Kai-Heng Feng wrote: > > If a second client that talks to the actual I2C address was created in > > probe(), there should be a corresponding cleanup in remove() to avoid > > leakage. > > > > So if the "client" is not the same one used by I2C core, unregister it > > accordingly. > > > > Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=2152281 > > Fixes: c1e62062ff54 ("iio: light: cm32181: Handle CM3218 ACPI devices with 2 I2C resources") > > Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com> > > --- > > v2: > > - Use devm_add_action_or_reset() instead of remove() callback to avoid > > race. > > > > drivers/iio/light/cm32181.c | 16 ++++++++++++++++ > > 1 file changed, 16 insertions(+) > > > > diff --git a/drivers/iio/light/cm32181.c b/drivers/iio/light/cm32181.c > > index b1674a5bfa368..a3e5f56101c9f 100644 > > --- a/drivers/iio/light/cm32181.c > > +++ b/drivers/iio/light/cm32181.c > > @@ -429,6 +429,16 @@ static const struct iio_info cm32181_info = { > > .attrs = &cm32181_attribute_group, > > }; > > > > +static void cm32181_disable(void *data) > > +{ > > + struct i2c_client *client = data; > > + struct cm32181_chip *cm32181 = iio_priv(i2c_get_clientdata(client)); > > + > > + /* Unregister the dummy client */ > > + if (cm32181->client != client) > > + i2c_unregister_device(cm32181->client); > > +} > > + > > static int cm32181_probe(struct i2c_client *client) > > { > > struct device *dev = &client->dev; > > @@ -479,6 +489,12 @@ static int cm32181_probe(struct i2c_client *client) > > return ret; > > } > > > > + ret = devm_add_action_or_reset(dev, cm32181_disable, client); > > + if (ret) { > > + dev_err(dev, "%s: add devres action failed\n", __func__); > > + return ret; > > + } > > + > > This is too late, we will still exit without unregistering the client if > the cm32181_reg_init() call fails. > > It would be best to do this directly after the i2c_acpi_new_device() > call, so inside the "if (ACPI_HANDLE(dev) && client->addr == SMBUS_ALERT_RESPONSE_ADDRESS) {" > block. You are right, not sure what I was thinking but this is much better. > > This way you can also remove the "if (cm32181->client != client)" > check from cm32181_disable() since it now only runs when the client > was registered in the first place. > > Also please rename cm32181_disable() to cm32181_unregister_dummy_client() > so that the name actually matches what it does. Sure, will change that in next version. > > Regards, > > Hans > > > > > > > ret = devm_iio_device_register(dev, indio_dev); > > if (ret) { > > dev_err(dev, "%s: regist device failed\n", __func__); >
diff --git a/drivers/iio/light/cm32181.c b/drivers/iio/light/cm32181.c index b1674a5bfa368..a3e5f56101c9f 100644 --- a/drivers/iio/light/cm32181.c +++ b/drivers/iio/light/cm32181.c @@ -429,6 +429,16 @@ static const struct iio_info cm32181_info = { .attrs = &cm32181_attribute_group, }; +static void cm32181_disable(void *data) +{ + struct i2c_client *client = data; + struct cm32181_chip *cm32181 = iio_priv(i2c_get_clientdata(client)); + + /* Unregister the dummy client */ + if (cm32181->client != client) + i2c_unregister_device(cm32181->client); +} + static int cm32181_probe(struct i2c_client *client) { struct device *dev = &client->dev; @@ -479,6 +489,12 @@ static int cm32181_probe(struct i2c_client *client) return ret; } + ret = devm_add_action_or_reset(dev, cm32181_disable, client); + if (ret) { + dev_err(dev, "%s: add devres action failed\n", __func__); + return ret; + } + ret = devm_iio_device_register(dev, indio_dev); if (ret) { dev_err(dev, "%s: regist device failed\n", __func__);
If a second client that talks to the actual I2C address was created in probe(), there should be a corresponding cleanup in remove() to avoid leakage. So if the "client" is not the same one used by I2C core, unregister it accordingly. Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=2152281 Fixes: c1e62062ff54 ("iio: light: cm32181: Handle CM3218 ACPI devices with 2 I2C resources") Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com> --- v2: - Use devm_add_action_or_reset() instead of remove() callback to avoid race. drivers/iio/light/cm32181.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+)