Re: [PATCH 14/16] iio: adc: max1027: Consolidate the end of conversion helper

[Jonathan Cameron]

On Mon, 30 Aug 2021 12:44:48 +0000
"Sa, Nuno" <Nuno.Sa@xxxxxxxxxx> wrote:

> > -----Original Message-----
> > From: Jonathan Cameron <jic23@xxxxxxxxxx>
> > Sent: Monday, August 30, 2021 12:37 PM
> > To: Miquel Raynal <miquel.raynal@xxxxxxxxxxx>
> > Cc: Lars-Peter Clausen <lars@xxxxxxxxxx>; Thomas Petazzoni
> > <thomas.petazzoni@xxxxxxxxxxx>; linux-iio@xxxxxxxxxxxxxxx; linux-
> > kernel@xxxxxxxxxxxxxxx
> > Subject: Re: [PATCH 14/16] iio: adc: max1027: Consolidate the end of
> > conversion helper
> > 
> > On Wed, 18 Aug 2021 13:11:37 +0200
> > Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote:
> >   
> > > Now that we have a dedicated handler for End Of Conversion  
> > interrupts,  
> > > let's create a second path:
> > > - Situation 1: we are using the external hardware trigger, a  
> > conversion  
> > >   has been triggered and the ADC pushed the data to its FIFO, we  
> > need to  
> > >   retrieve the data and push it to the IIO buffers.
> > > - Situation 2: we are not using the external hardware trigger, hence  
> > we  
> > >   are likely waiting in a blocked thread waiting for this interrupt to
> > >   happen: in this case we just wake up the waiting thread.
> > >
> > > Signed-off-by: Miquel Raynal <miquel.raynal@xxxxxxxxxxx>
> > > ---
> > >  drivers/iio/adc/max1027.c | 20 +++++++++++++++++---
> > >  1 file changed, 17 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/iio/adc/max1027.c b/drivers/iio/adc/max1027.c
> > > index 8d86e77fb5db..8c5995ae59f2 100644
> > > --- a/drivers/iio/adc/max1027.c
> > > +++ b/drivers/iio/adc/max1027.c
> > > @@ -235,6 +235,7 @@ struct max1027_state {
> > >  	struct iio_trigger		*trig;
> > >  	__be16				*buffer;
> > >  	struct mutex			lock;
> > > +	bool				data_rdy;
> > >  	bool				cnvst_trigger;
> > >  	u8				reg ____cacheline_aligned;
> > >  };
> > > @@ -243,12 +244,22 @@ static  
> > DECLARE_WAIT_QUEUE_HEAD(max1027_queue);  
> > >
> > >  static int max1027_wait_eoc(struct iio_dev *indio_dev)
> > >  {
> > > +	struct max1027_state *st = iio_priv(indio_dev);
> > >  	unsigned int conversion_time =  
> > MAX1027_CONVERSION_UDELAY;  
> > > +	int ret;
> > >
> > > -	if (indio_dev->active_scan_mask)
> > > -		conversion_time *= hweight32(*indio_dev-
> > >active_scan_mask);
> > > +	if (st->spi->irq) {
> > > +		ret =  
> > wait_event_interruptible_timeout(max1027_queue,  
> > > +						       st->data_rdy, HZ /  
> > 1000);  
> > > +		st->data_rdy = false;
> > > +		if (ret == -ERESTARTSYS)
> > > +			return ret;
> > > +	} else {
> > > +		if (indio_dev->active_scan_mask)
> > > +			conversion_time *= hweight32(*indio_dev-
> > >active_scan_mask);
> > >
> > > -	usleep_range(conversion_time, conversion_time * 2);
> > > +		usleep_range(conversion_time, conversion_time * 2);
> > > +	}
> > >
> > >  	return 0;
> > >  }
> > > @@ -481,6 +492,9 @@ static irqreturn_t  
> > max1027_eoc_irq_handler(int irq, void *private)  
> > >  	if (st->cnvst_trigger) {
> > >  		ret = max1027_read_scan(indio_dev);
> > >  		iio_trigger_notify_done(indio_dev->trig);
> > > +	} else {
> > > +		st->data_rdy = true;
> > > +		wake_up(&max1027_queue);  
> > 
> > I can't see why a queue is appropriate for this.  Use a completion and
> > have
> > one per instance of the device.  No need for the flag etc in that case as
> > complete() means we have had an interrupt.
> >   
> 
> In the case that 'st-> cnvst_trigger' is not set but the spi IRQ
> is present, we will wait until we get 'wake_up()' called from here. I wonder if
> that is a good idea as the device own trigger is not being used. FWIW, I think this
> sync logic is a bit confusing... I would still use the normal trigger infrastructure
> ('iio_trigger_generic_data_rdy_poll()') and use the 'cnvst_trigger' flag in the
> trigger handler to manually start conversions + wait till eoc. But I might be missing
> something though.
> 
> Regarding this handler, I just realized that this is the hard IRQ handler which
> might end up calling 'max1027_read_scan()' which in turn calls 'spi_read()'. Am I
> missing something here?
Good point.  This will need to be the threaded handler if done like this.

J

> 
> - Nuno Sá