Date:	Fri, 22 Jul 2011 15:24:10 +0200
From:	Eric Andersson <eric.andersson@...xphere.com>
To:	Jonathan Cameron <jic23@....ac.uk>
Cc:	dmitry.torokhov@...il.com, linux-input@...r.kernel.org,
	linux-kernel@...r.kernel.org, zhengguang.guo@...ch-sensortec.com,
	stefan.nilsson@...xphere.com, alan@...rguk.ukuu.org.uk,
	Albert Zhang <xu.zhang@...ch-sensortec.com>
Subject: Re: [PATCH v4 1/1] input: add driver for Bosch Sensortec's BMA150
 accelerometer

> Mostly looks fine.  Couple of nitpicks and one query about the interrupt
> handling (more general than this driver).
> 
> Can I confirm I've understood this correctly. When one receives any of the
> interrupts, the driver just kicks out a reading?  Given there is a delay
> before this read, imagine both high and low thresholds are set.  It's entirely
> possible for both to occur before the thread gets there and the value that is
> read to be somewhere in between... We'll get one interrupt and have no idea
> what happened...
> 
> Is there any way of working out what has occured and pushing that out?
In our case we are enabling three conditional interrupts where each
one of them have the chance of asserting the interrupt pin when
their particular interrupt criterias are fulfilled. This means that
one interrupt, for instance any motion, can assert the irq pin and
then at the same time have a high- and/or low-g interrupt consecutively
activated as well (but pin asserted by any motion already) where the
pin will be kept asserted until all the criterias for each asserted
interrupt are exited (i.e. not fullfilled anymore). So even if there
is a delay from the pin assertion and the irq thread read-out it
doesn't really matter if consecutive interrupts are activated in
between. At least one of the criterias will be met and there will be a motion
detection to read out. Of course, the interrupt types could be masked
out from the status register (addr. 0x09 bits 0 & 1 for low-/high-g
int), but I really don't see the purpose of this.

A comment regarding latched vs. non-latched irqs - we are intentionally not
using latched interrupts with the current setup since we currently don't
see the need for it. A latched setup would typically be needed in a high
inflow interrupt feature, such as new data interrupt, where the interrupt
buffer can be flooded due to high event frequency.
An alternative could be to activate the latched interrupt feature and
explicitly clear the interrupts from the driver, but this would just fire
off a new event directly until any of the interrupt criterias are fullfilled.
If you have any preferences or requirements when it comes to this, please share
your thoughts and we'll act upon this for the next version.

> > +	if (bma150->client->irq > 0) {
> > +		ret = input_register_device(idev);
> > +		if (ret < 0) {
> > +			input_free_device(idev);
> > +			return ret;
> > +		}
> > +		ret = request_threaded_irq(bma150->client->irq, NULL, bma150_irq_thread,
> > +			IRQF_TRIGGER_RISING, BMA150_DRIVER, bma150);
> Does this want to be oneshot? I don't think you have the latch enabled, so strobing
> across a threshold could result in an interrupt storm.
Sure, it could be worth having it sequential to the read-outs from the thread
instead of buffering events while reading. I'll fix.

As for your nitpicks, I will fix them for the next version.

Thanks for reviewing!

-- 
Best regards,
 Eric

 http://www.unixphere.com
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists