HID: thingm: fix workqueue race on remove

thingm_remove_rgb() needs to flush the workqueue after all the LED classes
have been unregistered, otherwise the removal might race with another LED
event coming, causing thingm_led_set() to schedule additional work after
thingm_remove_rgb() has flushed it. This obviously causes oops later, as
the scheduled work has been freed in the meantime.

In addition to that, move the hid_hw_stop() to an earlier place, so that
dmesg is not polluted by failure messages about not being able to write
the LED while the device is being shut down.

Reported-and-tested-by: Dylan Alex Simon <dylan-kernel@dylex.net>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>

diff --git a/drivers/hid/hid-thingm.c b/drivers/hid/hid-thingm.c
index 134be89..f206398 100644 (file)
--- a/drivers/hid/hid-thingm.c
+++ b/drivers/hid/hid-thingm.c
@@ -208,10 +208,10 @@ unregister_red:
 
 static void thingm_remove_rgb(struct thingm_rgb *rgb)
 {
-       flush_work(&rgb->work);
        led_classdev_unregister(&rgb->red.ldev);
        led_classdev_unregister(&rgb->green.ldev);
        led_classdev_unregister(&rgb->blue.ldev);
+       flush_work(&rgb->work);
 }
 
 static int thingm_probe(struct hid_device *hdev, const struct hid_device_id *id)
@@ -286,10 +286,10 @@ static void thingm_remove(struct hid_device *hdev)
        struct thingm_device *tdev = hid_get_drvdata(hdev);
        int i;
 
+       hid_hw_stop(hdev);
+
        for (i = 0; i < tdev->fwinfo->numrgb; ++i)
                thingm_remove_rgb(tdev->rgb + i);
-
-       hid_hw_stop(hdev);
 }
 
 static const struct hid_device_id thingm_table[] = {