Hi Dmitry,

On Tue, 29 Jan 2013 17:16:29 -0800
Dmitry Torokhov [off-list ref] wrote:
...

On Tue, Jan 29, 2013 at 01:25:24PM +0100, Anatolij Gustschin wrote:

quoted

On hardware with limited gpios one column select gpio can select
two different rows when using some additional hardware logic:
high value selects one row, low value selects another row. Add
support for such matrix keyboards and document device tree
bindings used to describe them.

Since half of the columns is always not selected, interrupts won't be
generated for press events on these columns. To generate interrupts
for not selected columns we need to periodicaly switch to these columns
in order to catch the potential press events. This is done by additional
work function.

So this effectively switches the driver from interrupt driven to polled
model, right? The device does not really need an interrupt anymore...

Yes, partially. I do not need to scan the state of the rows right
after column switching. The interrupts for press events will only be
generated if a key in the affected row was actually pressed. So the
device is still interrupt driven and the row state read-out happens
in the work scheduled from the interrupt handler.

Thanks,

Anatolij