[PATCH] HID: Allow drivers without listeners to remain on the bus

[PATCH] HID: Allow drivers without listeners to remain on the bus

[David Herrmann]

hid-picolcd and hid-wiimote do not allow any of hidinput, hiddev or hidraw
to claim the device but still want to remain on the bus. Hence, if a
driver uses the raw_event callback but no listener claimed the device, we
still leave it on the bus as the driver probably handles everything by
itself.

Under some circumstances (eg., hidinput_connect() fails and raw_event set)
a device may not be left on the bus even though it has no listeners. But
then if hidinput_connect() fails there are bigger issues than a device
that is left unhandled. So we can safely use this heuristic to avoid
adding another flag for special devices like hid-picolcd and hid-wiimote.

This also removes the ugly hack from hid-picolcd as this is no longer
required.

Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
---
Hi Jiri

I posted another patchset that introduced HID_CLAIMED_OTHER to avoid this
heuristic but Henrik convinced me that this trivial check is the cleaner and
less intrusive implementation. If you're interested, the old patchset was here:
  http://thread.gmane.org/gmane.linux.kernel.input/25945

Henrik, can you add your signed-off-by as this was your idea?

Thanks
David

 drivers/hid/hid-core.c    | 8 ++++++--
 drivers/hid/hid-picolcd.c | 4 ----
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/hid/hid-core.c b/drivers/hid/hid-core.c
index 9cdc74e..83ca3ba 100644
--- a/drivers/hid/hid-core.c
+++ b/drivers/hid/hid-core.c
@@ -1373,8 +1373,12 @@ int hid_connect(struct hid_device *hdev, unsigned int connect_mask)
 	if ((connect_mask & HID_CONNECT_HIDRAW) && !hidraw_connect(hdev))
 		hdev->claimed |= HID_CLAIMED_HIDRAW;
 
-	if (!hdev->claimed) {
-		hid_err(hdev, "claimed by neither input, hiddev nor hidraw\n");
+	/* There are device drivers that do not allow hidinput, hiddev or
+	 * hidraw to be connected but still want to remain on the bus. We
+	 * require that these devices use the raw_event callback. Otherwise, we
+	 * simply remove them from the bus if no listener claimed them. */
+	if (!hdev->claimed && !hdev->driver->raw_event) {
+		hid_err(hdev, "device has no listeners, quitting\n");
 		return -ENODEV;
 	}
 
diff --git a/drivers/hid/hid-picolcd.c b/drivers/hid/hid-picolcd.c
index 45c3433..74c388d 100644
--- a/drivers/hid/hid-picolcd.c
+++ b/drivers/hid/hid-picolcd.c
@@ -2613,11 +2613,7 @@ static int picolcd_probe(struct hid_device *hdev,
 		goto err_cleanup_data;
 	}
 
-	/* We don't use hidinput but hid_hw_start() fails if nothing is
-	 * claimed. So spoof claimed input. */
-	hdev->claimed = HID_CLAIMED_INPUT;
 	error = hid_hw_start(hdev, 0);
-	hdev->claimed = 0;
 	if (error) {
 		hid_err(hdev, "hardware start failed\n");
 		goto err_cleanup_data;
-- 
1.7.11.2

Re: [PATCH] HID: Allow drivers without listeners to remain on the bus

[Henrik Rydberg]

Hi David,

> hid-picolcd and hid-wiimote do not allow any of hidinput, hiddev or hidraw
> to claim the device but still want to remain on the bus. Hence, if a
> driver uses the raw_event callback but no listener claimed the device, we
> still leave it on the bus as the driver probably handles everything by
> itself.
> 
> Under some circumstances (eg., hidinput_connect() fails and raw_event set)
> a device may not be left on the bus even though it has no listeners. But
> then if hidinput_connect() fails there are bigger issues than a device
> that is left unhandled. So we can safely use this heuristic to avoid
> adding another flag for special devices like hid-picolcd and hid-wiimote.
> 
> This also removes the ugly hack from hid-picolcd as this is no longer
> required.
> 
> Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
> ---
> Hi Jiri
> 
> I posted another patchset that introduced HID_CLAIMED_OTHER to avoid this
> heuristic but Henrik convinced me that this trivial check is the cleaner and
> less intrusive implementation. If you're interested, the old patchset was here:
>   http://thread.gmane.org/gmane.linux.kernel.input/25945
> 
> Henrik, can you add your signed-off-by as this was your idea?

Gladly, but I would like to change the wording slightly, since a driver
with raw_event set is a listener, and we are still bailing out if
there are no listeners at all. The title is thus misleading.

> 
> Thanks
> David
> 
>  drivers/hid/hid-core.c    | 8 ++++++--
>  drivers/hid/hid-picolcd.c | 4 ----
>  2 files changed, 6 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/hid/hid-core.c b/drivers/hid/hid-core.c
> index 9cdc74e..83ca3ba 100644
> --- a/drivers/hid/hid-core.c
> +++ b/drivers/hid/hid-core.c
> @@ -1373,8 +1373,12 @@ int hid_connect(struct hid_device *hdev, unsigned int connect_mask)
>  	if ((connect_mask & HID_CONNECT_HIDRAW) && !hidraw_connect(hdev))
>  		hdev->claimed |= HID_CLAIMED_HIDRAW;
>  
> -	if (!hdev->claimed) {
> -		hid_err(hdev, "claimed by neither input, hiddev nor hidraw\n");
> +	/* There are device drivers that do not allow hidinput, hiddev or
> +	 * hidraw to be connected but still want to remain on the bus. We
> +	 * require that these devices use the raw_event callback. Otherwise, we
> +	 * simply remove them from the bus if no listener claimed them. */

How about "Drivers using raw_event() are not required to connect to anything else"?

> +	if (!hdev->claimed && !hdev->driver->raw_event) {
> +		hid_err(hdev, "device has no listeners, quitting\n");
>  		return -ENODEV;
>  	}
>  
> diff --git a/drivers/hid/hid-picolcd.c b/drivers/hid/hid-picolcd.c
> index 45c3433..74c388d 100644
> --- a/drivers/hid/hid-picolcd.c
> +++ b/drivers/hid/hid-picolcd.c
> @@ -2613,11 +2613,7 @@ static int picolcd_probe(struct hid_device *hdev,
>  		goto err_cleanup_data;
>  	}
>  
> -	/* We don't use hidinput but hid_hw_start() fails if nothing is
> -	 * claimed. So spoof claimed input. */
> -	hdev->claimed = HID_CLAIMED_INPUT;
>  	error = hid_hw_start(hdev, 0);
> -	hdev->claimed = 0;
>  	if (error) {
>  		hid_err(hdev, "hardware start failed\n");
>  		goto err_cleanup_data;
> -- 
> 1.7.11.2
> 

Thanks,
Henrik

Re: [PATCH] HID: Allow drivers without listeners to remain on the bus

[Jiri Kosina]

On Tue, 17 Jul 2012, Henrik Rydberg wrote:

> > hid-picolcd and hid-wiimote do not allow any of hidinput, hiddev or hidraw
> > to claim the device but still want to remain on the bus. Hence, if a
> > driver uses the raw_event callback but no listener claimed the device, we
> > still leave it on the bus as the driver probably handles everything by
> > itself.
> > 
> > Under some circumstances (eg., hidinput_connect() fails and raw_event set)
> > a device may not be left on the bus even though it has no listeners. But
> > then if hidinput_connect() fails there are bigger issues than a device
> > that is left unhandled. So we can safely use this heuristic to avoid
> > adding another flag for special devices like hid-picolcd and hid-wiimote.
> > 
> > This also removes the ugly hack from hid-picolcd as this is no longer
> > required.
> > 
> > Signed-off-by: David Herrmann <dh.herrmann@googlemail.com>
> > ---
> > Hi Jiri
> > 
> > I posted another patchset that introduced HID_CLAIMED_OTHER to avoid this
> > heuristic but Henrik convinced me that this trivial check is the cleaner and
> > less intrusive implementation. If you're interested, the old patchset was here:
> >   http://thread.gmane.org/gmane.linux.kernel.input/25945
> > 
> > Henrik, can you add your signed-off-by as this was your idea?
> 
> Gladly, but I would like to change the wording slightly, since a driver
> with raw_event set is a listener, and we are still bailing out if
> there are no listeners at all. The title is thus misleading.

So could you guys please come up with a revised version of the patch? I 
like the aproach much more than the previous one, so would like to queue 
it for 3.6 merge window.

Thanks,

-- 
Jiri Kosina
SUSE Labs