linux-input.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH 1/2] tsc2007: remove HR timer
@ 2009-06-23 11:54 Richard Röjfors
  2009-06-25 21:20 ` Andrew Morton
  2009-07-14  4:59 ` Dmitry Torokhov
  0 siblings, 2 replies; 10+ messages in thread
From: Richard Röjfors @ 2009-06-23 11:54 UTC (permalink / raw)
  To: linux-input
  Cc: Linux Kernel Mailing List, kwangwoo.lee, Thierry Reding,
	Trilok Soni, Andrew Morton

This patch removes the HR timer, since it's bad to do synchronous I2C
in the HR timer callback context. The new implementation makes use
of the global workqueue. The work is scheduled every 5ms when polling
rather than 5 us.

Signed-off-by: Richard Röjfors <richard.rojfors.ext@mocean-labs.com>
---
Index: linux-2.6.30/drivers/input/touchscreen/tsc2007.c
===================================================================
--- linux-2.6.30/drivers/input/touchscreen/tsc2007.c	(revision 932)
+++ linux-2.6.30/drivers/input/touchscreen/tsc2007.c	(revision 943)
@@ -21,15 +21,13 @@
  */

 #include <linux/module.h>
-#include <linux/hrtimer.h>
 #include <linux/slab.h>
 #include <linux/input.h>
 #include <linux/interrupt.h>
 #include <linux/i2c.h>
 #include <linux/i2c/tsc2007.h>

-#define TS_POLL_DELAY	(10 * 1000)	/* ns delay before the first sample */
-#define TS_POLL_PERIOD	(5 * 1000)	/* ns delay between samples */
+#define TS_POLL_PERIOD	msecs_to_jiffies(5) /* ms delay between samples */

 #define TSC2007_MEASURE_TEMP0		(0x0 << 4)
 #define TSC2007_MEASURE_AUX		(0x2 << 4)
@@ -70,13 +68,11 @@
 struct tsc2007 {
 	struct input_dev	*input;
 	char			phys[32];
-	struct hrtimer		timer;
+	struct delayed_work	work;
 	struct ts_event		tc;

 	struct i2c_client	*client;

-	spinlock_t		lock;
-
 	u16			model;
 	u16			x_plate_ohms;

@@ -142,8 +138,7 @@
 	if (rt > MAX_12BIT) {
 		dev_dbg(&ts->client->dev, "ignored pressure %d\n", rt);

-		hrtimer_start(&ts->timer, ktime_set(0, TS_POLL_PERIOD),
-			      HRTIMER_MODE_REL);
+		schedule_delayed_work(&ts->work, TS_POLL_PERIOD);
 		return;
 	}

@@ -153,7 +148,7 @@
 	 * in some cases may not even settle at the expected value.
 	 *
 	 * The only safe way to check for the pen up condition is in the
-	 * timer by reading the pen signal state (it's a GPIO _and_ IRQ).
+	 * work function by reading the pen signal state (it's a GPIO and IRQ).
 	 */
 	if (rt) {
 		struct input_dev *input = ts->input;
@@ -175,8 +170,7 @@
 			x, y, rt);
 	}

-	hrtimer_start(&ts->timer, ktime_set(0, TS_POLL_PERIOD),
-			HRTIMER_MODE_REL);
+	schedule_delayed_work(&ts->work, TS_POLL_PERIOD);
 }

 static int tsc2007_read_values(struct tsc2007 *tsc)
@@ -197,13 +191,10 @@
 	return 0;
 }

-static enum hrtimer_restart tsc2007_timer(struct hrtimer *handle)
+static void tsc2007_work(struct work_struct *work)
 {
-	struct tsc2007 *ts = container_of(handle, struct tsc2007, timer);
-	unsigned long flags;
-
-	spin_lock_irqsave(&ts->lock, flags);
-
+	struct tsc2007 *ts =
+		container_of(to_delayed_work(work), struct tsc2007, work);
 	if (unlikely(!ts->get_pendown_state() && ts->pendown)) {
 		struct input_dev *input = ts->input;

@@ -222,30 +213,20 @@
 		tsc2007_read_values(ts);
 		tsc2007_send_event(ts);
 	}
-
-	spin_unlock_irqrestore(&ts->lock, flags);
-
-	return HRTIMER_NORESTART;
 }

 static irqreturn_t tsc2007_irq(int irq, void *handle)
 {
 	struct tsc2007 *ts = handle;
-	unsigned long flags;

-	spin_lock_irqsave(&ts->lock, flags);
-
 	if (likely(ts->get_pendown_state())) {
 		disable_irq_nosync(ts->irq);
-		hrtimer_start(&ts->timer, ktime_set(0, TS_POLL_DELAY),
-					HRTIMER_MODE_REL);
+		schedule_delayed_work(&ts->work, 0);
 	}

 	if (ts->clear_penirq)
 		ts->clear_penirq();

-	spin_unlock_irqrestore(&ts->lock, flags);
-
 	return IRQ_HANDLED;
 }

@@ -278,11 +259,6 @@

 	ts->input = input_dev;

-	hrtimer_init(&ts->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
-	ts->timer.function = tsc2007_timer;
-
-	spin_lock_init(&ts->lock);
-
 	ts->model             = pdata->model;
 	ts->x_plate_ohms      = pdata->x_plate_ohms;
 	ts->get_pendown_state = pdata->get_pendown_state;
@@ -308,6 +284,8 @@

 	ts->irq = client->irq;

+	INIT_DELAYED_WORK(&ts->work, tsc2007_work);
+
 	err = request_irq(ts->irq, tsc2007_irq, 0,
 			client->dev.driver->name, ts);
 	if (err < 0) {
@@ -325,7 +303,6 @@

  err_free_irq:
 	free_irq(ts->irq, ts);
-	hrtimer_cancel(&ts->timer);
  err_free_mem:
 	input_free_device(input_dev);
 	kfree(ts);
@@ -337,11 +314,13 @@
 	struct tsc2007	*ts = i2c_get_clientdata(client);
 	struct tsc2007_platform_data *pdata;

+	/* cancel any work */
+	cancel_delayed_work(&ts->work);
+
 	pdata = client->dev.platform_data;
 	pdata->exit_platform_hw();

 	free_irq(ts->irq, ts);
-	hrtimer_cancel(&ts->timer);
 	input_unregister_device(ts->input);
 	kfree(ts);

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH 1/2] tsc2007: remove HR timer
  2009-06-23 11:54 [PATCH 1/2] tsc2007: remove HR timer Richard Röjfors
@ 2009-06-25 21:20 ` Andrew Morton
  2009-07-09 16:51   ` Richard Röjfors
  2009-07-14  4:59 ` Dmitry Torokhov
  1 sibling, 1 reply; 10+ messages in thread
From: Andrew Morton @ 2009-06-25 21:20 UTC (permalink / raw)
  To: Richard Röjfors
  Cc: linux-input, linux-kernel, kwangwoo.lee, thierry.reding,
	soni.trilok

On Tue, 23 Jun 2009 13:54:48 +0200
Richard R__jfors <richard.rojfors.ext@mocean-labs.com> wrote:

> This patch removes the HR timer, since it's bad to do synchronous I2C
> in the HR timer callback context. The new implementation makes use
> of the global workqueue. The work is scheduled every 5ms when polling
> rather than 5 us.
> 

"it's bad" isn't a very good description of the problem which the patch
fixes.

This matters.  People wish to make decisions about whether this patch
is needed in 2.6.29.x, 2.6.30.x, 2.6.31, 2.6.32, etc.  Without knowing
the effects of the problem which the patch fixes, we cannot make that
decision!

> 
> +	/* cancel any work */
> +	cancel_delayed_work(&ts->work);
> +

Should this have been cancel_delayed_work_sync()?


/*
 * Kill off a pending schedule_delayed_work().  Note that the work callback
 * function may still be running on return from cancel_delayed_work(), unless
 * it returns 1 and the work doesn't re-arm itself. Run flush_workqueue() or
 * cancel_work_sync() to wait on it.
 */
static inline int cancel_delayed_work(struct delayed_work *work)




^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH 1/2] tsc2007: remove HR timer
  2009-06-25 21:20 ` Andrew Morton
@ 2009-07-09 16:51   ` Richard Röjfors
  0 siblings, 0 replies; 10+ messages in thread
From: Richard Röjfors @ 2009-07-09 16:51 UTC (permalink / raw)
  To: Andrew Morton
  Cc: linux-input, linux-kernel, kwangwoo.lee, thierry.reding,
	soni.trilok

On 6/25/09 11:20 PM, Andrew Morton wrote:
> On Tue, 23 Jun 2009 13:54:48 +0200
> Richard R__jfors<richard.rojfors.ext@mocean-labs.com>  wrote:
>
>> This patch removes the HR timer, since it's bad to do synchronous I2C
>> in the HR timer callback context. The new implementation makes use
>> of the global workqueue. The work is scheduled every 5ms when polling
>> rather than 5 us.
>>
>
> "it's bad" isn't a very good description of the problem which the patch
> fixes.

The problem is that the synchronous I2C calls in HR-timer callback 
context might end up in a deadlock.

>
> This matters.  People wish to make decisions about whether this patch
> is needed in 2.6.29.x, 2.6.30.x, 2.6.31, 2.6.32, etc.  Without knowing
> the effects of the problem which the patch fixes, we cannot make that
> decision!
>
>> +	/* cancel any work */
>> +	cancel_delayed_work(&ts->work);
>> +
>
> Should this have been cancel_delayed_work_sync()?

It should, I will post a new patch shortly which should be applied after 
these two which fixes this.

--Richard

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH 1/2] tsc2007: remove HR timer
  2009-06-23 11:54 [PATCH 1/2] tsc2007: remove HR timer Richard Röjfors
  2009-06-25 21:20 ` Andrew Morton
@ 2009-07-14  4:59 ` Dmitry Torokhov
  2009-07-14  7:08   ` Thierry Reding
                     ` (2 more replies)
  1 sibling, 3 replies; 10+ messages in thread
From: Dmitry Torokhov @ 2009-07-14  4:59 UTC (permalink / raw)
  To: Richard Röjfors
  Cc: linux-input, Linux Kernel Mailing List, kwangwoo.lee,
	Thierry Reding, Trilok Soni, Andrew Morton

Hi Richard,

On Tue, Jun 23, 2009 at 01:54:48PM +0200, Richard Röjfors wrote:
> This patch removes the HR timer, since it's bad to do synchronous I2C
> in the HR timer callback context. The new implementation makes use
> of the global workqueue. The work is scheduled every 5ms when polling
> rather than 5 us.
> 
> Signed-off-by: Richard Röjfors <richard.rojfors.ext@mocean-labs.com>
> ---
> Index: linux-2.6.30/drivers/input/touchscreen/tsc2007.c
> ===================================================================
> --- linux-2.6.30/drivers/input/touchscreen/tsc2007.c	(revision 932)
> +++ linux-2.6.30/drivers/input/touchscreen/tsc2007.c	(revision 943)
> @@ -21,15 +21,13 @@
>   */
> 
>  #include <linux/module.h>
> -#include <linux/hrtimer.h>
>  #include <linux/slab.h>
>  #include <linux/input.h>
>  #include <linux/interrupt.h>
>  #include <linux/i2c.h>
>  #include <linux/i2c/tsc2007.h>
> 
> -#define TS_POLL_DELAY	(10 * 1000)	/* ns delay before the first sample */
> -#define TS_POLL_PERIOD	(5 * 1000)	/* ns delay between samples */
> +#define TS_POLL_PERIOD	msecs_to_jiffies(5) /* ms delay between samples */
> 
>  #define TSC2007_MEASURE_TEMP0		(0x0 << 4)
>  #define TSC2007_MEASURE_AUX		(0x2 << 4)
> @@ -70,13 +68,11 @@
>  struct tsc2007 {
>  	struct input_dev	*input;
>  	char			phys[32];
> -	struct hrtimer		timer;
> +	struct delayed_work	work;
>  	struct ts_event		tc;
> 
>  	struct i2c_client	*client;
> 
> -	spinlock_t		lock;
> -
>  	u16			model;
>  	u16			x_plate_ohms;
> 
> @@ -142,8 +138,7 @@
>  	if (rt > MAX_12BIT) {
>  		dev_dbg(&ts->client->dev, "ignored pressure %d\n", rt);
> 
> -		hrtimer_start(&ts->timer, ktime_set(0, TS_POLL_PERIOD),
> -			      HRTIMER_MODE_REL);
> +		schedule_delayed_work(&ts->work, TS_POLL_PERIOD);
>  		return;
>  	}
> 
> @@ -153,7 +148,7 @@
>  	 * in some cases may not even settle at the expected value.
>  	 *
>  	 * The only safe way to check for the pen up condition is in the
> -	 * timer by reading the pen signal state (it's a GPIO _and_ IRQ).
> +	 * work function by reading the pen signal state (it's a GPIO and IRQ).
>  	 */
>  	if (rt) {
>  		struct input_dev *input = ts->input;
> @@ -175,8 +170,7 @@
>  			x, y, rt);
>  	}
> 
> -	hrtimer_start(&ts->timer, ktime_set(0, TS_POLL_PERIOD),
> -			HRTIMER_MODE_REL);
> +	schedule_delayed_work(&ts->work, TS_POLL_PERIOD);
>  }
> 
>  static int tsc2007_read_values(struct tsc2007 *tsc)
> @@ -197,13 +191,10 @@
>  	return 0;
>  }
> 
> -static enum hrtimer_restart tsc2007_timer(struct hrtimer *handle)
> +static void tsc2007_work(struct work_struct *work)
>  {
> -	struct tsc2007 *ts = container_of(handle, struct tsc2007, timer);
> -	unsigned long flags;
> -
> -	spin_lock_irqsave(&ts->lock, flags);
> -
> +	struct tsc2007 *ts =
> +		container_of(to_delayed_work(work), struct tsc2007, work);
>  	if (unlikely(!ts->get_pendown_state() && ts->pendown)) {
>  		struct input_dev *input = ts->input;
> 
> @@ -222,30 +213,20 @@
>  		tsc2007_read_values(ts);
>  		tsc2007_send_event(ts);
>  	}
> -
> -	spin_unlock_irqrestore(&ts->lock, flags);
> -
> -	return HRTIMER_NORESTART;
>  }
> 
>  static irqreturn_t tsc2007_irq(int irq, void *handle)
>  {
>  	struct tsc2007 *ts = handle;
> -	unsigned long flags;
> 
> -	spin_lock_irqsave(&ts->lock, flags);
> -
>  	if (likely(ts->get_pendown_state())) {
>  		disable_irq_nosync(ts->irq);
> -		hrtimer_start(&ts->timer, ktime_set(0, TS_POLL_DELAY),
> -					HRTIMER_MODE_REL);
> +		schedule_delayed_work(&ts->work, 0);

Originally we scheduled reading with timy delay to let the device
settle, why don't we schedule with 1ms delay?

>  	}
> 
>  	if (ts->clear_penirq)
>  		ts->clear_penirq();
> 
> -	spin_unlock_irqrestore(&ts->lock, flags);
> -
>  	return IRQ_HANDLED;
>  }
> 
> @@ -278,11 +259,6 @@
> 
>  	ts->input = input_dev;
> 
> -	hrtimer_init(&ts->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
> -	ts->timer.function = tsc2007_timer;
> -
> -	spin_lock_init(&ts->lock);
> -
>  	ts->model             = pdata->model;
>  	ts->x_plate_ohms      = pdata->x_plate_ohms;
>  	ts->get_pendown_state = pdata->get_pendown_state;
> @@ -308,6 +284,8 @@
> 
>  	ts->irq = client->irq;
> 
> +	INIT_DELAYED_WORK(&ts->work, tsc2007_work);
> +
>  	err = request_irq(ts->irq, tsc2007_irq, 0,
>  			client->dev.driver->name, ts);
>  	if (err < 0) {
> @@ -325,7 +303,6 @@
> 
>   err_free_irq:
>  	free_irq(ts->irq, ts);
> -	hrtimer_cancel(&ts->timer);

Why don't we cancel work here?

>   err_free_mem:
>  	input_free_device(input_dev);
>  	kfree(ts);
> @@ -337,11 +314,13 @@
>  	struct tsc2007	*ts = i2c_get_clientdata(client);
>  	struct tsc2007_platform_data *pdata;
> 
> +	/* cancel any work */
> +	cancel_delayed_work(&ts->work);
> +
>  	pdata = client->dev.platform_data;
>  	pdata->exit_platform_hw();
> 

So what happens if IRQ is raised here and work is scheduled again?

>  	free_irq(ts->irq, ts);
> -	hrtimer_cancel(&ts->timer);
>  	input_unregister_device(ts->input);
>  	kfree(ts);
> 

Could you please try the patch below and see if the touchscreen still
works for you? It should be applied on top of your patch.

Thanks!

-- 
Dmitry

Input: tsc2007 - properly shut off interrupts/delayed work

From: Dmitry Torokhov <dmitry.torokhov@gmail.com>

Properly shut off interrupts/delayed work by free-ing IRQ first
and then ensuring that enable/disable is balanced. Also add
__devinit/__devexit markings, restore poll delay/period scheduling
logic, make sure we call exit_platform_hw() method when probe
fails.

Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
---

 drivers/input/touchscreen/tsc2007.c |   76 ++++++++++++++++++++++-------------
 1 files changed, 48 insertions(+), 28 deletions(-)


diff --git a/drivers/input/touchscreen/tsc2007.c b/drivers/input/touchscreen/tsc2007.c
index b512697..3c71e85 100644
--- a/drivers/input/touchscreen/tsc2007.c
+++ b/drivers/input/touchscreen/tsc2007.c
@@ -27,7 +27,8 @@
 #include <linux/i2c.h>
 #include <linux/i2c/tsc2007.h>
 
-#define TS_POLL_PERIOD	msecs_to_jiffies(1) /* ms delay between samples */
+#define TS_POLL_DELAY			1 /* ms delay between samples */
+#define TS_POLL_PERIOD			1 /* ms delay between samples */
 
 #define TSC2007_MEASURE_TEMP0		(0x0 << 4)
 #define TSC2007_MEASURE_AUX		(0x2 << 4)
@@ -76,7 +77,7 @@ struct tsc2007 {
 	u16			model;
 	u16			x_plate_ohms;
 
-	unsigned		pendown;
+	bool			pendown;
 	int			irq;
 
 	int			(*get_pendown_state)(void);
@@ -131,18 +132,18 @@ static void tsc2007_send_event(void *tsc)
 	} else
 		rt = 0;
 
-	/* Sample found inconsistent by debouncing or pressure is beyond
+	/*
+	 * Sample found inconsistent by debouncing or pressure is beyond
 	 * the maximum. Don't report it to user space, repeat at least
 	 * once more the measurement
 	 */
 	if (rt > MAX_12BIT) {
 		dev_dbg(&ts->client->dev, "ignored pressure %d\n", rt);
-
-		schedule_delayed_work(&ts->work, TS_POLL_PERIOD);
 		return;
 	}
 
-	/* NOTE: We can't rely on the pressure to determine the pen down
+	/*
+	 * NOTE: We can't rely on the pressure to determine the pen down
 	 * state, even this controller has a pressure sensor.  The pressure
 	 * value can fluctuate for quite a while after lifting the pen and
 	 * in some cases may not even settle at the expected value.
@@ -157,7 +158,7 @@ static void tsc2007_send_event(void *tsc)
 			dev_dbg(&ts->client->dev, "DOWN\n");
 
 			input_report_key(input, BTN_TOUCH, 1);
-			ts->pendown = 1;
+			ts->pendown = true;
 		}
 
 		input_report_abs(input, ABS_X, x);
@@ -169,8 +170,6 @@ static void tsc2007_send_event(void *tsc)
 		dev_dbg(&ts->client->dev, "point(%4d,%4d), pressure (%4u)\n",
 			x, y, rt);
 	}
-
-	schedule_delayed_work(&ts->work, TS_POLL_PERIOD);
 }
 
 static int tsc2007_read_values(struct tsc2007 *tsc)
@@ -195,6 +194,7 @@ static void tsc2007_work(struct work_struct *work)
 {
 	struct tsc2007 *ts =
 		container_of(to_delayed_work(work), struct tsc2007, work);
+
 	if (unlikely(!ts->get_pendown_state() && ts->pendown)) {
 		struct input_dev *input = ts->input;
 
@@ -204,7 +204,7 @@ static void tsc2007_work(struct work_struct *work)
 		input_report_abs(input, ABS_PRESSURE, 0);
 		input_sync(input);
 
-		ts->pendown = 0;
+		ts->pendown = false;
 		enable_irq(ts->irq);
 	} else {
 		/* pen is still down, continue with the measurement */
@@ -212,6 +212,9 @@ static void tsc2007_work(struct work_struct *work)
 
 		tsc2007_read_values(ts);
 		tsc2007_send_event(ts);
+
+		schedule_delayed_work(&ts->work,
+				      msecs_to_jiffies(TS_POLL_PERIOD));
 	}
 }
 
@@ -221,7 +224,8 @@ static irqreturn_t tsc2007_irq(int irq, void *handle)
 
 	if (likely(ts->get_pendown_state())) {
 		disable_irq_nosync(ts->irq);
-		schedule_delayed_work(&ts->work, 0);
+		schedule_delayed_work(&ts->work,
+				      msecs_to_jiffies(TS_POLL_DELAY));
 	}
 
 	if (ts->clear_penirq)
@@ -230,8 +234,21 @@ static irqreturn_t tsc2007_irq(int irq, void *handle)
 	return IRQ_HANDLED;
 }
 
-static int tsc2007_probe(struct i2c_client *client,
-			const struct i2c_device_id *id)
+static void tsc2007_free_irq(struct tsc2007 *ts)
+{
+	free_irq(ts->irq, ts);
+	if (cancel_delayed_work_sync(&ts->work)) {
+		/*
+		 * Work was pending, therefore we need to enable
+		 * IRQ here to balance the disable_irq() done in the
+		 * interrupt handler.
+		 */
+		enable_irq(ts->irq);
+	}
+}
+
+static int __devinit tsc2007_probe(struct i2c_client *client,
+				   const struct i2c_device_id *id)
 {
 	struct tsc2007 *ts;
 	struct tsc2007_platform_data *pdata = pdata = client->dev.platform_data;
@@ -255,17 +272,15 @@ static int tsc2007_probe(struct i2c_client *client,
 	}
 
 	ts->client = client;
-	i2c_set_clientdata(client, ts);
-
+	ts->irq = client->irq;
 	ts->input = input_dev;
+	INIT_DELAYED_WORK(&ts->work, tsc2007_work);
 
 	ts->model             = pdata->model;
 	ts->x_plate_ohms      = pdata->x_plate_ohms;
 	ts->get_pendown_state = pdata->get_pendown_state;
 	ts->clear_penirq      = pdata->clear_penirq;
 
-	pdata->init_platform_hw();
-
 	snprintf(ts->phys, sizeof(ts->phys),
 		 "%s/input0", dev_name(&client->dev));
 
@@ -280,12 +295,9 @@ static int tsc2007_probe(struct i2c_client *client,
 	input_set_abs_params(input_dev, ABS_Y, 0, MAX_12BIT, 0, 0);
 	input_set_abs_params(input_dev, ABS_PRESSURE, 0, MAX_12BIT, 0, 0);
 
+	pdata->init_platform_hw();
 	tsc2007_read_values(ts);
 
-	ts->irq = client->irq;
-
-	INIT_DELAYED_WORK(&ts->work, tsc2007_work);
-
 	err = request_irq(ts->irq, tsc2007_irq, 0,
 			client->dev.driver->name, ts);
 	if (err < 0) {
@@ -297,29 +309,37 @@ static int tsc2007_probe(struct i2c_client *client,
 	if (err)
 		goto err_free_irq;
 
+	i2c_set_clientdata(client, ts);
 	dev_info(&client->dev, "registered with irq (%d)\n", ts->irq);
 
 	return 0;
 
  err_free_irq:
-	free_irq(ts->irq, ts);
+	tsc2007_free_irq(ts);
+	pdata->exit_platform_hw();
  err_free_mem:
 	input_free_device(input_dev);
 	kfree(ts);
 	return err;
 }
 
-static int tsc2007_remove(struct i2c_client *client)
+static int __devexit tsc2007_remove(struct i2c_client *client)
 {
 	struct tsc2007	*ts = i2c_get_clientdata(client);
-	struct tsc2007_platform_data *pdata;
+	struct tsc2007_platform_data *pdata = client->dev.platform_data;
 
-	cancel_delayed_work_sync(&ts->work);
+	free_irq(ts->irq, ts);
+	if (cancel_delayed_work_sync(&ts->work)) {
+		/*
+		 * Work was pending, therefore we need to enable
+		 * IRQ here to balance the disabel done in the
+		 * interrupt handler.
+		 */
+		enable_irq(ts->irq);
+	}
 
-	pdata = client->dev.platform_data;
 	pdata->exit_platform_hw();
 
-	free_irq(ts->irq, ts);
 	input_unregister_device(ts->input);
 	kfree(ts);
 
@@ -340,7 +360,7 @@ static struct i2c_driver tsc2007_driver = {
 	},
 	.id_table	= tsc2007_idtable,
 	.probe		= tsc2007_probe,
-	.remove		= tsc2007_remove,
+	.remove		= __devexit_p(tsc2007_remove),
 };
 
 static int __init tsc2007_init(void)
--
To unsubscribe from this list: send the line "unsubscribe linux-input" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply related	[flat|nested] 10+ messages in thread

* Re: [PATCH 1/2] tsc2007: remove HR timer
  2009-07-14  4:59 ` Dmitry Torokhov
@ 2009-07-14  7:08   ` Thierry Reding
  2009-07-14  8:00     ` Richard Röjfors
  2009-07-14  8:21     ` Dmitry Torokhov
  2009-07-14  8:47   ` Richard Röjfors
  2009-07-24 14:39   ` Richard Röjfors
  2 siblings, 2 replies; 10+ messages in thread
From: Thierry Reding @ 2009-07-14  7:08 UTC (permalink / raw)
  To: Dmitry Torokhov
  Cc: Richard Röjfors, linux-input, Linux Kernel Mailing List,
	kwangwoo.lee, Trilok Soni, Andrew Morton

* Dmitry Torokhov wrote:
> Hi Richard,
> 
> On Tue, Jun 23, 2009 at 01:54:48PM +0200, Richard Röjfors wrote:
[...]
> +static void tsc2007_free_irq(struct tsc2007 *ts)
> +{
> +	free_irq(ts->irq, ts);
> +	if (cancel_delayed_work_sync(&ts->work)) {
> +		/*
> +		 * Work was pending, therefore we need to enable
> +		 * IRQ here to balance the disable_irq() done in the
> +		 * interrupt handler.
> +		 */
> +		enable_irq(ts->irq);
> +	}
> +}
[...]
> -static int tsc2007_remove(struct i2c_client *client)
> +static int __devexit tsc2007_remove(struct i2c_client *client)
>  {
>  	struct tsc2007	*ts = i2c_get_clientdata(client);
> -	struct tsc2007_platform_data *pdata;
> +	struct tsc2007_platform_data *pdata = client->dev.platform_data;
>  
> -	cancel_delayed_work_sync(&ts->work);
> +	free_irq(ts->irq, ts);
> +	if (cancel_delayed_work_sync(&ts->work)) {
> +		/*
> +		 * Work was pending, therefore we need to enable
> +		 * IRQ here to balance the disabel done in the
> +		 * interrupt handler.
> +		 */
> +		enable_irq(ts->irq);
> +	}

Shouldn't this be tsc2007_free_irq(ts) as well?

Thierry
--
To unsubscribe from this list: send the line "unsubscribe linux-input" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH 1/2] tsc2007: remove HR timer
  2009-07-14  7:08   ` Thierry Reding
@ 2009-07-14  8:00     ` Richard Röjfors
  2009-07-14  8:21       ` Dmitry Torokhov
  2009-07-14  8:21     ` Dmitry Torokhov
  1 sibling, 1 reply; 10+ messages in thread
From: Richard Röjfors @ 2009-07-14  8:00 UTC (permalink / raw)
  To: Thierry Reding
  Cc: Dmitry Torokhov, linux-input, Linux Kernel Mailing List,
	kwangwoo.lee, Trilok Soni, Andrew Morton

On 7/14/09 9:08 AM, Thierry Reding wrote:
> * Dmitry Torokhov wrote:
>> Hi Richard,
>>
>> On Tue, Jun 23, 2009 at 01:54:48PM +0200, Richard Röjfors wrote:
> [...]
>> +static void tsc2007_free_irq(struct tsc2007 *ts)
>> +{
>> +	free_irq(ts->irq, ts);
>> +	if (cancel_delayed_work_sync(&ts->work)) {
>> +		/*
>> +		 * Work was pending, therefore we need to enable
>> +		 * IRQ here to balance the disable_irq() done in the
>> +		 * interrupt handler.
>> +		 */
>> +		enable_irq(ts->irq);
>> +	}
>> +}
> [...]
>> -static int tsc2007_remove(struct i2c_client *client)
>> +static int __devexit tsc2007_remove(struct i2c_client *client)
>>   {
>>   	struct tsc2007	*ts = i2c_get_clientdata(client);
>> -	struct tsc2007_platform_data *pdata;
>> +	struct tsc2007_platform_data *pdata = client->dev.platform_data;
>>
>> -	cancel_delayed_work_sync(&ts->work);
>> +	free_irq(ts->irq, ts);
>> +	if (cancel_delayed_work_sync(&ts->work)) {
>> +		/*
>> +		 * Work was pending, therefore we need to enable
>> +		 * IRQ here to balance the disabel done in the
>> +		 * interrupt handler.
>> +		 */
>> +		enable_irq(ts->irq);
>> +	}
>
> Shouldn't this be tsc2007_free_irq(ts) as well?

Is this really good enough? The work function might re-enable the IRQ.

Isn't it better to look at penstate, we know that the IRQ is enabled if 
the state is UP, else the IRQ is disabled.

--Richard

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH 1/2] tsc2007: remove HR timer
  2009-07-14  8:00     ` Richard Röjfors
@ 2009-07-14  8:21       ` Dmitry Torokhov
  0 siblings, 0 replies; 10+ messages in thread
From: Dmitry Torokhov @ 2009-07-14  8:21 UTC (permalink / raw)
  To: Richard Röjfors
  Cc: Thierry Reding, linux-input, Linux Kernel Mailing List,
	kwangwoo.lee, Trilok Soni, Andrew Morton

On Tue, Jul 14, 2009 at 10:00:06AM +0200, Richard Röjfors wrote:
> On 7/14/09 9:08 AM, Thierry Reding wrote:
>> * Dmitry Torokhov wrote:
>>> Hi Richard,
>>>
>>> On Tue, Jun 23, 2009 at 01:54:48PM +0200, Richard Röjfors wrote:
>> [...]
>>> +static void tsc2007_free_irq(struct tsc2007 *ts)
>>> +{
>>> +	free_irq(ts->irq, ts);
>>> +	if (cancel_delayed_work_sync(&ts->work)) {
>>> +		/*
>>> +		 * Work was pending, therefore we need to enable
>>> +		 * IRQ here to balance the disable_irq() done in the
>>> +		 * interrupt handler.
>>> +		 */
>>> +		enable_irq(ts->irq);
>>> +	}
>>> +}
>> [...]
>>> -static int tsc2007_remove(struct i2c_client *client)
>>> +static int __devexit tsc2007_remove(struct i2c_client *client)
>>>   {
>>>   	struct tsc2007	*ts = i2c_get_clientdata(client);
>>> -	struct tsc2007_platform_data *pdata;
>>> +	struct tsc2007_platform_data *pdata = client->dev.platform_data;
>>>
>>> -	cancel_delayed_work_sync(&ts->work);
>>> +	free_irq(ts->irq, ts);
>>> +	if (cancel_delayed_work_sync(&ts->work)) {
>>> +		/*
>>> +		 * Work was pending, therefore we need to enable
>>> +		 * IRQ here to balance the disabel done in the
>>> +		 * interrupt handler.
>>> +		 */
>>> +		enable_irq(ts->irq);
>>> +	}
>>
>> Shouldn't this be tsc2007_free_irq(ts) as well?
>
> Is this really good enough? The work function might re-enable the IRQ.

I am pretty sure it is. cancel_delayed_work_sync() will terurn true if
there was work scheduled but not yet executed. In this case we know that
IRQ was disabled by interrupt handler and will re-enable it.

-- 
Dmitry
--
To unsubscribe from this list: send the line "unsubscribe linux-input" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH 1/2] tsc2007: remove HR timer
  2009-07-14  7:08   ` Thierry Reding
  2009-07-14  8:00     ` Richard Röjfors
@ 2009-07-14  8:21     ` Dmitry Torokhov
  1 sibling, 0 replies; 10+ messages in thread
From: Dmitry Torokhov @ 2009-07-14  8:21 UTC (permalink / raw)
  To: Thierry Reding
  Cc: Richard Röjfors, linux-input, Linux Kernel Mailing List,
	kwangwoo.lee, Trilok Soni, Andrew Morton

On Tue, Jul 14, 2009 at 09:08:06AM +0200, Thierry Reding wrote:
> * Dmitry Torokhov wrote:
> > Hi Richard,
> > 
> > On Tue, Jun 23, 2009 at 01:54:48PM +0200, Richard Röjfors wrote:
> [...]
> > +static void tsc2007_free_irq(struct tsc2007 *ts)
> > +{
> > +	free_irq(ts->irq, ts);
> > +	if (cancel_delayed_work_sync(&ts->work)) {
> > +		/*
> > +		 * Work was pending, therefore we need to enable
> > +		 * IRQ here to balance the disable_irq() done in the
> > +		 * interrupt handler.
> > +		 */
> > +		enable_irq(ts->irq);
> > +	}
> > +}
> [...]
> > -static int tsc2007_remove(struct i2c_client *client)
> > +static int __devexit tsc2007_remove(struct i2c_client *client)
> >  {
> >  	struct tsc2007	*ts = i2c_get_clientdata(client);
> > -	struct tsc2007_platform_data *pdata;
> > +	struct tsc2007_platform_data *pdata = client->dev.platform_data;
> >  
> > -	cancel_delayed_work_sync(&ts->work);
> > +	free_irq(ts->irq, ts);
> > +	if (cancel_delayed_work_sync(&ts->work)) {
> > +		/*
> > +		 * Work was pending, therefore we need to enable
> > +		 * IRQ here to balance the disabel done in the
> > +		 * interrupt handler.
> > +		 */
> > +		enable_irq(ts->irq);
> > +	}
> 
> Shouldn't this be tsc2007_free_irq(ts) as well?
> 

Yep, it should, thanks.

-- 
Dmitry
--
To unsubscribe from this list: send the line "unsubscribe linux-input" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH 1/2] tsc2007: remove HR timer
  2009-07-14  4:59 ` Dmitry Torokhov
  2009-07-14  7:08   ` Thierry Reding
@ 2009-07-14  8:47   ` Richard Röjfors
  2009-07-24 14:39   ` Richard Röjfors
  2 siblings, 0 replies; 10+ messages in thread
From: Richard Röjfors @ 2009-07-14  8:47 UTC (permalink / raw)
  To: Dmitry Torokhov
  Cc: linux-input, Linux Kernel Mailing List, kwangwoo.lee,
	Thierry Reding, Trilok Soni, Andrew Morton

On 7/14/09 6:59 AM, Dmitry Torokhov wrote:
> Hi Richard,
>
> On Tue, Jun 23, 2009 at 01:54:48PM +0200, Richard Röjfors wrote:
>> This patch removes the HR timer, since it's bad to do synchronous I2C
>> in the HR timer callback context. The new implementation makes use
>> of the global workqueue. The work is scheduled every 5ms when polling
>> rather than 5 us.
>>
>> Signed-off-by: Richard Röjfors<richard.rojfors.ext@mocean-labs.com>
>> ---
>> Index: linux-2.6.30/drivers/input/touchscreen/tsc2007.c
>> ===================================================================
>> --- linux-2.6.30/drivers/input/touchscreen/tsc2007.c	(revision 932)
>> +++ linux-2.6.30/drivers/input/touchscreen/tsc2007.c	(revision 943)
>> @@ -21,15 +21,13 @@
>>    */
>>
>>   #include<linux/module.h>
>> -#include<linux/hrtimer.h>
>>   #include<linux/slab.h>
>>   #include<linux/input.h>
>>   #include<linux/interrupt.h>
>>   #include<linux/i2c.h>
>>   #include<linux/i2c/tsc2007.h>
>>
>> -#define TS_POLL_DELAY	(10 * 1000)	/* ns delay before the first sample */
>> -#define TS_POLL_PERIOD	(5 * 1000)	/* ns delay between samples */
>> +#define TS_POLL_PERIOD	msecs_to_jiffies(5) /* ms delay between samples */
>>
>>   #define TSC2007_MEASURE_TEMP0		(0x0<<  4)
>>   #define TSC2007_MEASURE_AUX		(0x2<<  4)
>> @@ -70,13 +68,11 @@
>>   struct tsc2007 {
>>   	struct input_dev	*input;
>>   	char			phys[32];
>> -	struct hrtimer		timer;
>> +	struct delayed_work	work;
>>   	struct ts_event		tc;
>>
>>   	struct i2c_client	*client;
>>
>> -	spinlock_t		lock;
>> -
>>   	u16			model;
>>   	u16			x_plate_ohms;
>>
>> @@ -142,8 +138,7 @@
>>   	if (rt>  MAX_12BIT) {
>>   		dev_dbg(&ts->client->dev, "ignored pressure %d\n", rt);
>>
>> -		hrtimer_start(&ts->timer, ktime_set(0, TS_POLL_PERIOD),
>> -			      HRTIMER_MODE_REL);
>> +		schedule_delayed_work(&ts->work, TS_POLL_PERIOD);
>>   		return;
>>   	}
>>
>> @@ -153,7 +148,7 @@
>>   	 * in some cases may not even settle at the expected value.
>>   	 *
>>   	 * The only safe way to check for the pen up condition is in the
>> -	 * timer by reading the pen signal state (it's a GPIO _and_ IRQ).
>> +	 * work function by reading the pen signal state (it's a GPIO and IRQ).
>>   	 */
>>   	if (rt) {
>>   		struct input_dev *input = ts->input;
>> @@ -175,8 +170,7 @@
>>   			x, y, rt);
>>   	}
>>
>> -	hrtimer_start(&ts->timer, ktime_set(0, TS_POLL_PERIOD),
>> -			HRTIMER_MODE_REL);
>> +	schedule_delayed_work(&ts->work, TS_POLL_PERIOD);
>>   }
>>
>>   static int tsc2007_read_values(struct tsc2007 *tsc)
>> @@ -197,13 +191,10 @@
>>   	return 0;
>>   }
>>
>> -static enum hrtimer_restart tsc2007_timer(struct hrtimer *handle)
>> +static void tsc2007_work(struct work_struct *work)
>>   {
>> -	struct tsc2007 *ts = container_of(handle, struct tsc2007, timer);
>> -	unsigned long flags;
>> -
>> -	spin_lock_irqsave(&ts->lock, flags);
>> -
>> +	struct tsc2007 *ts =
>> +		container_of(to_delayed_work(work), struct tsc2007, work);
>>   	if (unlikely(!ts->get_pendown_state()&&  ts->pendown)) {
>>   		struct input_dev *input = ts->input;
>>
>> @@ -222,30 +213,20 @@
>>   		tsc2007_read_values(ts);
>>   		tsc2007_send_event(ts);
>>   	}
>> -
>> -	spin_unlock_irqrestore(&ts->lock, flags);
>> -
>> -	return HRTIMER_NORESTART;
>>   }
>>
>>   static irqreturn_t tsc2007_irq(int irq, void *handle)
>>   {
>>   	struct tsc2007 *ts = handle;
>> -	unsigned long flags;
>>
>> -	spin_lock_irqsave(&ts->lock, flags);
>> -
>>   	if (likely(ts->get_pendown_state())) {
>>   		disable_irq_nosync(ts->irq);
>> -		hrtimer_start(&ts->timer, ktime_set(0, TS_POLL_DELAY),
>> -					HRTIMER_MODE_REL);
>> +		schedule_delayed_work(&ts->work, 0);
>
> Originally we scheduled reading with timy delay to let the device
> settle, why don't we schedule with 1ms delay?

Good point, scheduling of the work queue takes some time, but of course 
that varies.

>
>>   	}
>>
>>   	if (ts->clear_penirq)
>>   		ts->clear_penirq();
>>
>> -	spin_unlock_irqrestore(&ts->lock, flags);
>> -
>>   	return IRQ_HANDLED;
>>   }
>>
>> @@ -278,11 +259,6 @@
>>
>>   	ts->input = input_dev;
>>
>> -	hrtimer_init(&ts->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
>> -	ts->timer.function = tsc2007_timer;
>> -
>> -	spin_lock_init(&ts->lock);
>> -
>>   	ts->model             = pdata->model;
>>   	ts->x_plate_ohms      = pdata->x_plate_ohms;
>>   	ts->get_pendown_state = pdata->get_pendown_state;
>> @@ -308,6 +284,8 @@
>>
>>   	ts->irq = client->irq;
>>
>> +	INIT_DELAYED_WORK(&ts->work, tsc2007_work);
>> +
>>   	err = request_irq(ts->irq, tsc2007_irq, 0,
>>   			client->dev.driver->name, ts);
>>   	if (err<  0) {
>> @@ -325,7 +303,6 @@
>>
>>    err_free_irq:
>>   	free_irq(ts->irq, ts);
>> -	hrtimer_cancel(&ts->timer);
>
> Why don't we cancel work here?

We should.

>
>>    err_free_mem:
>>   	input_free_device(input_dev);
>>   	kfree(ts);
>> @@ -337,11 +314,13 @@
>>   	struct tsc2007	*ts = i2c_get_clientdata(client);
>>   	struct tsc2007_platform_data *pdata;
>>
>> +	/* cancel any work */
>> +	cancel_delayed_work(&ts->work);
>> +
>>   	pdata = client->dev.platform_data;
>>   	pdata->exit_platform_hw();
>>
>
> So what happens if IRQ is raised here and work is scheduled again?
>
>>   	free_irq(ts->irq, ts);
>> -	hrtimer_cancel(&ts->timer);
>>   	input_unregister_device(ts->input);
>>   	kfree(ts);
>>
>
> Could you please try the patch below and see if the touchscreen still
> works for you? It should be applied on top of your patch.

I have modified my code accordingly to most of your changes. I don't 
have any platform callbacks in my platform, so make sure those are 
optional. Will try it out!

--Richard
--
To unsubscribe from this list: send the line "unsubscribe linux-input" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: [PATCH 1/2] tsc2007: remove HR timer
  2009-07-14  4:59 ` Dmitry Torokhov
  2009-07-14  7:08   ` Thierry Reding
  2009-07-14  8:47   ` Richard Röjfors
@ 2009-07-24 14:39   ` Richard Röjfors
  2 siblings, 0 replies; 10+ messages in thread
From: Richard Röjfors @ 2009-07-24 14:39 UTC (permalink / raw)
  To: Dmitry Torokhov
  Cc: linux-input, Linux Kernel Mailing List, kwangwoo.lee,
	Thierry Reding, Trilok Soni, Andrew Morton

Dmitry Torokhov wrote:
> 
> Could you please try the patch below and see if the touchscreen still
> works for you? It should be applied on top of your patch.

Sorry for coming back late, we switched I2C IP so I had to write a new 
driver for it before trying this out again.

The irq balancing seem to work as in you patch. I didn't run exactly
you patch, I took some of the changes into my code, since we can not
provide the callbacks.

I have also done some other small changes, to saves us one of five I2C 
transactions in polling mode.

I Will post this incremental patch in a couple of minutes.

Thanks
--Richard

^ permalink raw reply	[flat|nested] 10+ messages in thread

end of thread, other threads:[~2009-07-24 14:39 UTC | newest]

Thread overview: 10+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2009-06-23 11:54 [PATCH 1/2] tsc2007: remove HR timer Richard Röjfors
2009-06-25 21:20 ` Andrew Morton
2009-07-09 16:51   ` Richard Röjfors
2009-07-14  4:59 ` Dmitry Torokhov
2009-07-14  7:08   ` Thierry Reding
2009-07-14  8:00     ` Richard Röjfors
2009-07-14  8:21       ` Dmitry Torokhov
2009-07-14  8:21     ` Dmitry Torokhov
2009-07-14  8:47   ` Richard Röjfors
2009-07-24 14:39   ` Richard Röjfors

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).