Re: [PATCH] iio: common: st_sensors: fix channel data parsing

[Lorenzo Bianconi <lorenzo.bianconi83@gmail.com>]

On Oct 01, Jonathan Cameron wrote:
> On 01/10/16 15:18, Lorenzo Bianconi wrote:
> >> On 28/09/16 21:06, Lorenzo Bianconi wrote:
> >>> Using realbits as i2c/spi read len, when that value is not byte aligned
> >>> (e.g 12 bits), lead to skip msb part of out data registers.
> >>> Fix this using storagebits as read length
> >>>
> >>> Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@st.com>
> >> Hi Lorenzo,
> > 
> > Hi Jonathan,
> > 
> >>
> >> This would ideally have had a fixes tag.
> >>
> >> I think it was
> >> Fixes: e7385de5 ("iio:st_sensors: align on storagebits boundaries")
> >> Doing that would also have lead to you to the logic that lead to
> >> this buggy change in the first place. Would also have shown
> >> you that there is another place that probably suffers from the
> >> same sort of issue.
> >>
> > 
> > Right. I missed the same issue in st_sensors_read_axis_data()
> > 
> >> Gregor can you take a look at this please.
> >>
> >> If I recall the issue that lead to the original patch it was
> >> that we were miss handling 24 bit values on pressure sensors and
> >> this was intended to pad them?
> >>
> >> I think the 'right' fix will be something along the lines of:
> >> unsigned int bytes_to_read = (channel->scan_type.realbits + 7) >> 3;
> >>
> > 
> > This should not be correct if you consider 8bit accel like LIS331DL.
> 8 + 7 = 15
> 15 >> 3 = 1 byte.

Right sorry :)

> 
> Would get interest if the data was shifted to run across two bytes, but
> it's not.  Could handle that case as well by doing
> (channel->scan_type.realbits + channel->scan_type.shift + 7) >> 3
> 

Ack. I will send a v2 afer testing on multiple devices (8, 12, 24 realbits)

> > Taking a look to lps22hb code I think storagebits should be 24 and not
> > 32 for pressure channel. What do you think?
> No. Storage bytes must always be a power of 2. It's assumed at various
> points in the buffer handling code in the core and userspace code.
> This is why Gregor's fix was needed.
> 
> These 24 bit packed readings are a pain, but it was much worse to try
> and remove the aligned data assumptions that need them to be powers
> of 2.  Just think about 3 bytes floating in a little endian integer
> field vs the big endian version. It's a real pain to unwind.

Right :)

Regards,
Lorenzo

> > 
> >> Should give 2 bytes for a 12 bit and still the 3 bytes needed for a
> >> 24 bit read.
> >>
> >> Thanks,
> >>
> > 
> > Thanks,
> > Lorenzo
> > 
> >> Jonathan
> >>> ---
> >>>  drivers/iio/common/st_sensors/st_sensors_buffer.c | 5 ++---
> >>>  1 file changed, 2 insertions(+), 3 deletions(-)
> >>>
> >>> diff --git a/drivers/iio/common/st_sensors/st_sensors_buffer.c b/drivers/iio/common/st_sensors/st_sensors_buffer.c
> >>> index fe7775b..d01aa34 100644
> >>> --- a/drivers/iio/common/st_sensors/st_sensors_buffer.c
> >>> +++ b/drivers/iio/common/st_sensors/st_sensors_buffer.c
> >>> @@ -30,16 +30,15 @@ static int st_sensors_get_buffer_element(struct iio_dev *indio_dev, u8 *buf)
> >>>
> >>>       for_each_set_bit(i, indio_dev->active_scan_mask, num_data_channels) {q
> >>>               const struct iio_chan_spec *channel = &indio_dev->channels[i];
> >>> -             unsigned int bytes_to_read = channel->scan_type.realbits >> 3;
> >>>               unsigned int storage_bytes =
> >>>                       channel->scan_type.storagebits >> 3;
> >>>
> >>>               buf = PTR_ALIGN(buf, storage_bytes);
> >>>               if (sdata->tf->read_multiple_byte(&sdata->tb, sdata->dev,
> >>>                                                 channel->address,
> >>> -                                               bytes_to_read, buf,
> >>> +                                               storage_bytes, buf,
> >>>                                                 sdata->multiread_bit) <
> >>> -                 bytes_to_read)
> >>> +                 storage_bytes)
> >>>                       return -EIO;
> >>>
> >>>               /* Advance the buffer pointer */
> >>>
> >>
> > 
> > 
> > 
> 

-- 
UNIX is Sexy: who | grep -i blonde | talk; cd ~; wine; talk; touch; unzip; touch; strip; gasp; finger; gasp; mount; fsck; more; yes; gasp; umount; make clean; sleep