On 7/14/09 9:08 AM, Thierry Reding wrote:

* Dmitry Torokhov wrote:

quoted

Hi Richard,

On Tue, Jun 23, 2009 at 01:54:48PM +0200, Richard Röjfors wrote:

[...]

quoted

+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);
+	}
+}

[...]

quoted

-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