hid: core: add managed version of hid_hw_start

Most calls to hid_hw_stop are boilerplate code. The function comment
states: "This is usually called from remove function or from probe when something failed and hid_hw_start was called already."

A managed version of hid_hw_start allows to avoid this boilerplate code in most cases (similar to the other devm_ functions). hid_hw_stop is called automatically before the device is deleted.

---
 drivers/hid/hid-core.c | 22 ++++++++++++++++++++++
 include/linux/hid.h    |  3 +++
 2 files changed, 25 insertions(+)

index e95f23f..61c6f63 100644
@@ -2752,6 +2752,28 @@ int hid_check_keys_pressed(struct hid_device *hid)
 
 EXPORT_SYMBOL_GPL(hid_check_keys_pressed);
 
+static void __hid_hw_stop(void *data)
+{
+	hid_hw_stop((struct hid_device *)data);
+}
+
+int __must_check devm_hid_hw_start(struct hid_device *hdev,
+				   unsigned int connect_mask)
+{
+	int ret;
+
+	ret = hid_hw_start(hdev, connect_mask);
+	if (ret)
+		return ret;
+
+	ret = devm_add_action(&hdev->dev, __hid_hw_stop, hdev);
+	if (ret)
+		hid_hw_stop(hdev);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(devm_hid_hw_start);
+
 static int __init hid_init(void)
 {
 	int ret;

index 75b66ec..2d85b3e 100644
@@ -774,6 +774,9 @@ extern bool hid_ignore(struct hid_device *);
 extern int hid_add_device(struct hid_device *);
 extern void hid_destroy_device(struct hid_device *);
 
+extern int __must_check devm_hid_hw_start(struct hid_device *hdev,
+					  unsigned int connect_mask);
+
 extern int __must_check __hid_register_driver(struct hid_driver *,
 		struct module *, const char *mod_name);
 

2.6.4

Generally I agree with the idea, but I'll be merging this only together with patches that are converting drivers to actually use this API.

Thanks,

Jiri Kosina
SUSE Labs

Am 08.01.2016 um 10:51 schrieb Jiri Kosina:

OK. When working on it I saw that hid_hw_stop is called by the core if the driver doesn't define a remove callback.
This means if we use devm_hid_hw_start in a driver with a probe but no remove callback then hid_hw_stop will be called twice.
To prevent this I'd switch all drivers with probe but no remove to the new call and remove the fallback call to hid_hw_stop in the core. At the same time I'd switch the fallback call to hid_hw_start in the core (in case the driver doesn't define a probe callback) to the new call. This way the change is transparent to drivers which define neither probe nor remove.
Fine with you?

Regards, Heiner