[PATCH 1/2] HID: wacom: Fix error path of power-supply initialization

power_supply_unregister() must not be called if power_supply_register() failed.
The wdata->psy.dev pointer may point to invalid memory after a failed
power_supply_register() and hence wacom_remove() will fail while calling
power_supply_unregister().

This changes the wacom_probe function to fail if it cannot register the
power_supply devices. If we would want to keep the previous behaviour we had to
keep some flag about the power_supply state and check it on wacom_remove, but
this seems inappropriate here. Hence, we simply fail, too, if
power_supply_register fails.

Signed-off-by: David Herrmann <redacted>
---
 drivers/hid/hid-wacom.c |   21 +++++++++------------
 1 files changed, 9 insertions(+), 12 deletions(-)

diff --git a/drivers/hid/hid-wacom.c b/drivers/hid/hid-wacom.c
index 0688832..f66a597 100644
--- a/drivers/hid/hid-wacom.c
+++ b/drivers/hid/hid-wacom.c

@@ -353,11 +353,7 @@ static int wacom_probe(struct hid_device *hdev,
 	if (ret) {
 		hid_warn(hdev, "can't create sysfs battery attribute, err: %d\n",
 			 ret);
-		/*
-		 * battery attribute is not critical for the tablet, but if it
-		 * failed then there is no need to create ac attribute
-		 */
-		goto move_on;
+		goto err_battery;
 	}
 
 	wdata->ac.properties = wacom_ac_props;

@@ -371,14 +367,8 @@ static int wacom_probe(struct hid_device *hdev,
 	if (ret) {
 		hid_warn(hdev,
 			 "can't create ac battery attribute, err: %d\n", ret);
-		/*
-		 * ac attribute is not critical for the tablet, but if it
-		 * failed then we don't want to battery attribute to exist
-		 */
-		power_supply_unregister(&wdata->battery);
+		goto err_ac;
 	}
-
-move_on:
 #endif
 	hidinput = list_entry(hdev->inputs.next, struct hid_input, list);
 	input = hidinput->input;

@@ -416,6 +406,13 @@ move_on:
 
 	return 0;
 
+#ifdef CONFIG_HID_WACOM_POWER_SUPPLY
+err_ac:
+	power_supply_unregister(&wdata->battery);
+err_battery:
+	device_remove_file(&hdev->dev, &dev_attr_speed);
+	hid_hw_stop(hdev);
+#endif
 err_free:
 	kfree(wdata);
 	return ret;

-- 
1.7.6.1

HID devices can be hotplugged so we should unregister all sysfs attributes when
removing a driver. Otherwise, manually unloading the wacom-driver will not
remove the sysfs attributes. Only when the device is disconnected, they are
removed, eventually.

Signed-off-by: David Herrmann <redacted>
---
 drivers/hid/hid-wacom.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/drivers/hid/hid-wacom.c b/drivers/hid/hid-wacom.c
index f66a597..a597039 100644
--- a/drivers/hid/hid-wacom.c
+++ b/drivers/hid/hid-wacom.c

@@ -423,6 +423,7 @@ static void wacom_remove(struct hid_device *hdev)
 #ifdef CONFIG_HID_WACOM_POWER_SUPPLY
 	struct wacom_data *wdata = hid_get_drvdata(hdev);
 #endif
+	device_remove_file(&hdev->dev, &dev_attr_speed);
 	hid_hw_stop(hdev);
 
 #ifdef CONFIG_HID_WACOM_POWER_SUPPLY

-- 
1.7.6.1

On Mon, 5 Sep 2011, David Herrmann wrote:

power_supply_unregister() must not be called if power_supply_register() failed.
The wdata->psy.dev pointer may point to invalid memory after a failed
power_supply_register() and hence wacom_remove() will fail while calling
power_supply_unregister().

This changes the wacom_probe function to fail if it cannot register the
power_supply devices. If we would want to keep the previous behaviour we had to
keep some flag about the power_supply state and check it on wacom_remove, but
this seems inappropriate here. Hence, we simply fail, too, if
power_supply_register fails.

Signed-off-by: David Herrmann <redacted>

Applied, thanks.

-- 
Jiri Kosina
SUSE Labs

On Mon, 5 Sep 2011, David Herrmann wrote:

HID devices can be hotplugged so we should unregister all sysfs attributes when
removing a driver. Otherwise, manually unloading the wacom-driver will not
remove the sysfs attributes. Only when the device is disconnected, they are
removed, eventually.

Signed-off-by: David Herrmann <redacted>

Applied, thanks David.

-- 
Jiri Kosina
SUSE Labs