On Tue, Jan 26, 2016 at 09:04:20AM -0800, Stefan Agner wrote:

On 2016-01-25 23:58, Michael Hennerich wrote:

quoted

On 01/26/2016 04:04 AM, Stefan Agner wrote:

quoted

The measurements read from the controller which are temporary stored
in conversion_data, are interpreted wrong. The first measurement X+
contains the Y position, and the second measurement Y+ the X position
(see also Table 11 Register Table in the data sheet).

The problem is already known and a swap option has been introduced:
commit 6680884a4420 ("Input: ad7879 - add option to correct xy axis")

However, with that the meaning of the new boolean is inverted since
the underlying values are already swapped. With this change, a true
in swap_xy actually swaps the two axis.

Signed-off-by: Stefan Agner <stefan@agner.ch>
---
Hi Michael,

It seems that swap_xy is not used in any board which is in mainline,
hence swap_xy is always false. Therefore, up until now all boards
actually used swapped axis. However, I doubt that the blackfin boards
really have those axis swapped, it is probably more likely that the
userspace calibration took care of it.

However, if they are really swapped, we should set the swap_xy flag
to 1 for those board...

Do you happen to now what is the case with those boards?

Hi Stefan,

I would be hesitant to invert the default behaviour of the driver.
Too many people in the field already using it as it is.

Afaik, we should be able to change in-kernel API's (especially if they
are wrong) since we do not guarantee any API...

quoted

A XY swap can have multiple reasons.

Lot's of small VGA/QVGA TFTs have the option to switch the scan
direction from Landscape to Portrait. In addition you can also rotate
and flip or mirror using VDMA options. So it really depends on the use
case, how the touch panel is mounted to the screen or how it is wired.

Ok, I see the reason for that functionality.

I am mainly concerned about the new DT bindings. The touchscreen binding
documents specify touchscreen-swapped-x-y, see:
https://www.kernel.org/doc/Documentation/devicetree/bindings/input/touchscreen/touchscreen.txt

I would like to make sure that this property is really swapping axis
(and not necessary if the hardware is implemented according to the
datasheet...)

We could also implement a workaround to keep the platform data behavior
as is (invert the swap_xy flag)...

That is probably the best option.

Thanks.

-- 
Dmitry