Skip to contentSkip to conversation

HID: hid-sensor-hub: Clear HID_CLAIMED_INPUT bit earlier

PATCH In review Axel Lin 2 participants 6 messages Sep 13, 2012

Subsystems (2)
lore
Share & tools
lore

Axel Lin

author
#0 lore raw

PATCH 1/2 HID: hid-sensor-hub: Clear HID_CLAIMED_INPUT bit earlier

Clear HID_CLAIMED_INPUT bit of hdev->claimed, this prevents calling hidinput_disconnect() in hid_disconnect(), which is called by hid_hw_stop().

We need to clear HID_CLAIMED_INPUT bit earlier rather than in sensor_hub_remove() because we also call hid_hw_stop() in sensor_hub_probe error handling.

Also adds comments for why we set hdev->claimed = HID_CLAIMED_INPUT.

---
Hi srinivas,
I don't have this hardware, I'd appreciate if you can test these patches. Axel

 drivers/hid/hid-sensor-hub.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/hid/hid-sensor-hub.c +3 -1
index 4ac759c..50bc8c7 100644
@@ -539,8 +539,11 @@ static int sensor_hub_probe(struct hid_device *hdev,
 	}
 	INIT_LIST_HEAD(&hdev->inputs);
 
+	/* We don't use hidinput but hid_hw_start() fails if nothing is
+	 * claimed. So spoof claimed input. */
 	hdev->claimed = HID_CLAIMED_INPUT;
 	ret = hid_hw_start(hdev, 0);
+	hdev->claimed &= ~HID_CLAIMED_INPUT;
 	if (ret) {
 		hid_err(hdev, "hw start failed\n");
 		goto err_free;
@@ -627,7 +630,6 @@ static void sensor_hub_remove(struct hid_device *hdev)
 	int i;
 
 	hid_dbg(hdev, " hardware removed\n");
-	hdev->claimed &= ~HID_CLAIMED_INPUT;
 	hid_hw_stop(hdev);
 	hid_hw_close(hdev);
 	spin_lock_irqsave(&data->lock, flags);
Signature · 1 line
1.7.9.5
author
#1 lore raw

PATCH 2/2 HID: hid-sensor-hub: Fix sensor_hub_probe error handling

Fix below issues in sensor_hub_probe error handling:
1. In the case of goto err_close, hid_hw_stop(hdev) is called twice. Fix it.
2. If fails to allocate MFD device name, we also need to free all successfully allocated names in previous iterations.

---
 drivers/hid/hid-sensor-hub.c |    4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/hid/hid-sensor-hub.c +1 -3
index 50bc8c7..8bd7620 100644
@@ -584,7 +584,7 @@ static int sensor_hub_probe(struct hid_device *hdev,
 				hid_err(hdev,
 					"Failed MFD device name\n");
 					ret = -ENOMEM;
-					goto err_free_cells;
+					goto err_free_names;
 			}
 			sd->hid_sensor_hub_client_devs[
 				sd->hid_sensor_client_cnt].name = name;
@@ -608,10 +608,8 @@ static int sensor_hub_probe(struct hid_device *hdev,
 err_free_names:
 	for (i = 0; i < sd->hid_sensor_client_cnt ; ++i)
 		kfree(sd->hid_sensor_hub_client_devs[i].name);
-err_free_cells:
 	kfree(sd->hid_sensor_hub_client_devs);
 err_close:
-	hid_hw_stop(hdev);
 	hid_hw_close(hdev);
 err_stop_hw:
 	hid_hw_stop(hdev);
Signature · 1 line
1.7.9.5
#2 lore raw

RE: [PATCH 2/2] HID: hid-sensor-hub: Fix sensor_hub_probe error handling

Agreed. It works.

Thanks,
Srinivas

-----Original Message-----
From: Axel Lin [mailto:[email protected]]
Sent: Wednesday, September 12, 2012 6:40 PM
To: Jiri Kosina
Cc: Pandruvada, Srinivas; Jonathan Cameron; [email protected]; [email protected]
Subject: [PATCH 2/2] HID: hid-sensor-hub: Fix sensor_hub_probe error handling

Fix below issues in sensor_hub_probe error handling:
1. In the case of goto err_close, hid_hw_stop(hdev) is called twice. Fix it.
2. If fails to allocate MFD device name, we also need to free all successfully allocated names in previous iterations.

---
 drivers/hid/hid-sensor-hub.c |    4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/hid/hid-sensor-hub.c +1 -1
@@ -584,7 +584,7 @@ static int sensor_hub_probe(struct hid_device *hdev,
 				hid_err(hdev,
 					"Failed MFD device name\n");
 					ret = -ENOMEM;
-					goto err_free_cells;
+					goto err_free_names;
 			}
 			sd->hid_sensor_hub_client_devs[
 				sd->hid_sensor_client_cnt].name = name; @@ -608,10 +608,8 @@ static int sensor_hub_probe(struct hid_device *hdev,

err_free_names:
for (i = 0; i < sd->hid_sensor_client_cnt ; ++i)
kfree(sd->hid_sensor_hub_client_devs[i].name);
-err_free_cells:
kfree(sd->hid_sensor_hub_client_devs);
err_close:
- hid_hw_stop(hdev);
hid_hw_close(hdev);
err_stop_hw:
hid_hw_stop(hdev);
--
1.7.9.5

author
#3 lore raw
In reply to Axel Lin

2012/9/13 Axel Lin <[email protected]>:

Quoted 23 lines

Clear HID_CLAIMED_INPUT bit of hdev->claimed, this prevents calling hidinput_disconnect() in hid_disconnect(), which is called by hid_hw_stop().

We need to clear HID_CLAIMED_INPUT bit earlier rather than in sensor_hub_remove() because we also call hid_hw_stop() in sensor_hub_probe error handling.

Also adds comments for why we set hdev->claimed = HID_CLAIMED_INPUT.

---
Hi srinivas,
I don't have this hardware, I'd appreciate if you can test these patches. Axel

 drivers/hid/hid-sensor-hub.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/hid/hid-sensor-hub.c +0 -0
index 4ac759c..50bc8c7 100644

drivers/hid/hid-sensor-hub.c static int sensor_hub_probe(struct hid_device *hdev, · line 545

Quoted 7 lines
drivers/hid/hid-sensor-hub.c +2 -0
@@ -539,8 +539,11 @@ static int sensor_hub_probe(struct hid_device *hdev,
        }
        INIT_LIST_HEAD(&hdev->inputs);
 
+       /* We don't use hidinput but hid_hw_start() fails if nothing is
+        * claimed. So spoof claimed input. */
        hdev->claimed = HID_CLAIMED_INPUT;
        ret = hid_hw_start(hdev, 0);

I check the implementation of hid_hw_start again, it looks hid_hw_start allows connect_mask to be 0.
So we don't need setting "hdev->claimed = HID_CLAIMED_INPUT;" here. Comments?

Axel

#4 lore raw

Hi Axel,

We don't need this anymore as when connect_mask is 0, the hid_connect will not be called. So we don't need hdev->claimed = HID_CLAIMED_INPUT anymore.

Thanks for the review.

-Srinivas

-----Original Message-----
From: Axel Lin [mailto:[email protected]]
Sent: Thursday, September 13, 2012 6:25 AM
To: Jiri Kosina
Cc: Pandruvada, Srinivas; Jonathan Cameron; [email protected]; [email protected]
Subject: Re: [PATCH 1/2] HID: hid-sensor-hub: Clear HID_CLAIMED_INPUT bit earlier

2012/9/13 Axel Lin <[email protected]>:

Quoted 23 lines

Clear HID_CLAIMED_INPUT bit of hdev->claimed, this prevents calling hidinput_disconnect() in hid_disconnect(), which is called by hid_hw_stop().

We need to clear HID_CLAIMED_INPUT bit earlier rather than in sensor_hub_remove() because we also call hid_hw_stop() in sensor_hub_probe error handling.

Also adds comments for why we set hdev->claimed = HID_CLAIMED_INPUT.

---
Hi srinivas,
I don't have this hardware, I'd appreciate if you can test these patches. Axel

 drivers/hid/hid-sensor-hub.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
(quoted diff) +0 -0

b/drivers/hid/hid-sensor-hub.c index 4ac759c..50bc8c7 100644

drivers/hid/hid-sensor-hub.c static int sensor_hub_probe(struct hid_device *hdev, · line 545

Quoted 7 lines
drivers/hid/hid-sensor-hub.c +2 -0
@@ -539,8 +539,11 @@ static int sensor_hub_probe(struct hid_device *hdev,
        }
        INIT_LIST_HEAD(&hdev->inputs);
 
+       /* We don't use hidinput but hid_hw_start() fails if nothing is
+        * claimed. So spoof claimed input. */
        hdev->claimed = HID_CLAIMED_INPUT;
        ret = hid_hw_start(hdev, 0);

I check the implementation of hid_hw_start again, it looks hid_hw_start allows connect_mask to be 0. So we don't need setting "hdev->claimed = HID_CLAIMED_INPUT;" here.
Comments?

Axel

author
#5 lore raw

2012/9/14 Pandruvada, Srinivas <[email protected]>:

Quoted 5 lines

Hi Axel,

We don't need this anymore as when connect_mask is 0, the hid_connect will not be called. So we don't need hdev->claimed = HID_CLAIMED_INPUT anymore.

Okay. I'm going to resend the patch for this serial. Axel