On Sun, Mar 22, 2015 at 12:00:55PM +0100, Hans de Goede wrote:

Hi,

On 22-03-15 05:03, Dmitry Torokhov wrote:

quoted

Hi Hans,

On Tue, Mar 10, 2015 at 06:05:33PM +0100, Hans de Goede wrote:

quoted

+	error = devm_request_threaded_irq(dev, client->irq, NULL, icn8318_irq,
+					  IRQF_TRIGGER_FALLING | IRQF_ONESHOT,

Shouldn't we let DT data tell us what trigger to use? I.e. just leave
IRQF_ONESHOT here?

That is an interesting question, that new data is available is signalled by
the irq pin of the chip going low is a property of the chip, not the board
layout, so I believe it is best to leave this as is.

My concern is that even if pin behavior is property of chip maybe on
some boards we want to use level-triggered interrupts instead of edge?
And if we indeed want to hard-code the trigger then shouldn't the
binding document use onecell mapping (so that users do not attempt to
configure triggers from DT)?

Also note that if we want to get this from devicetree, that simply leaving out the
flag is not enough, we must specifically get the data from devicetree and pass
it into request_irq AFAICT. So the above would change to:

	irqflags = irqd_get_trigger_type(irq_get_irq_data(client->irq)) | IRQF_ONESHOT,
	error = devm_request_threaded_irq(dev, client->irq, NULL, icn8318_irq, irqflags,

No, of_irq_get() that i2c core calls before probing driver should
already set the trigger type for us. There is no need for the individual
drivers to do that.

Thanks.

-- 
Dmitry