Re: [PATCH] iio: imu: st_lsm6dsx: irq not handled unless data pushed to buffers

[Lorenzo Bianconi <lorenzo.bianconi@redhat.com>]

> On 07/11/2018 05:26 PM, Lorenzo Bianconi wrote:
> > On Jul 11, Jorge Ramirez-Ortiz wrote:
> > > On 07/11/2018 02:29 PM, Lorenzo Bianconi wrote:
> > > > > Currently IRQ_NONE is returned only when there is no data on the fifo.
> > > > > 
> > > > > When there is no data on the fifo the driver can not push to the
> > > > > buffers and therefore user space readers polling for data available
> > > > > will not be awoken and continue to wait.
> > > > > 
> > > > > This commit just extends the same semantics to fifo read errors.
> > > > Hi Jorge,
> > > > 
> > > > IRQ_NONE is used to indicate this interrupt is not intended for this driver
> > > > (this could happen if the irq line is in open-drain). If the interrupt is for
> > > > st_lsm6dsx I would prefer to return IRQ_HANDLED even in case of error.
> > > yes I understand.
> > > 
> > > This was just a trivial attempt (I guess a really bad idea) to get some
> > > debug info (via /proc/irq/.../spurious) any time the driver read (spi/i2c)
> > > fails when processing the data ready irq.
> > > do you think it would make sense to add a dev_err to
> > > st_lsm6dsx_i2c_read/st_lsm6dsx_spi_read? at the moment the driver would fail
> > > silently
> > do you mean something like (just compiled, not tested):
> > 
> > --- a/drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c
> > +++ b/drivers/iio/imu/st_lsm6dsx/st_lsm6dsx_buffer.c
> > @@ -298,8 +298,11 @@ static int st_lsm6dsx_read_fifo(struct st_lsm6dsx_hw *hw)
> >   	err = regmap_bulk_read(hw->regmap,
> >   			       hw->settings->fifo_ops.fifo_diff.addr,
> >   			       &fifo_status, sizeof(fifo_status));
> > -	if (err < 0)
> > +	if (err < 0) {
> > +		dev_err(hw->dev, "failed to read fifo status reg (err=%d)\n",
> > +			err);
> >   		return err;
> > +	}
> >   	if (fifo_status & cpu_to_le16(ST_LSM6DSX_FIFO_EMPTY_MASK))
> >   		return 0;
> > @@ -313,8 +316,12 @@ static int st_lsm6dsx_read_fifo(struct st_lsm6dsx_hw *hw)
> >   	for (read_len = 0; read_len < fifo_len; read_len += pattern_len) {
> >   		err = st_lsm6dsx_read_block(hw, hw->buff, pattern_len);
> > -		if (err < 0)
> > +		if (err < 0) {
> > +			dev_err(hw->dev,
> > +				"failed to read pattern from fifo (err=%d)\n",
> > +				err);
> >   			return err;
> > +		}
> >   		/*
> >   		 * Data are written to the FIFO with a specific pattern
> > @@ -385,8 +392,11 @@ static int st_lsm6dsx_read_fifo(struct st_lsm6dsx_hw *hw)
> >   	if (unlikely(reset_ts)) {
> >   		err = st_lsm6dsx_reset_hw_ts(hw);
> > -		if (err < 0)
> > +		if (err < 0) {
> > +			dev_err(hw->dev, "failed to reset hw ts (err=%d)\n",
> > +				err);
> >   			return err;
> > +		}
> >   	}
> >   	return read_len;
> >   }
> > 
> > Regards,
> > Lorenzo
> 
> 
> yes, you beat me to it but yes, that is what I was thinking about.
> 

Ops, it was not intended as a race :). Feel free to send your patch since you
proposed the idea.

Regards,
Lorenzo

>