when "SYNTIANT_NDP_ERROR_DATA_REREAD" error occurs, zero is
returned in the argument extract size as no bytes are extracted.
However, in the next iteration, zero is passed as extract size,
causing to extract the entire audio tank, hence corrupting the
heap. This exactly matches the crash we observed as the heap
corrupted (around 60KB) matches the size of the audio tank (60KB)

Signed-off-by: Abhishek Akkabathula <[email protected]>

use level based instead of edge based as its a more robust trigger for gpio interrupts

Signed-off-by: Abhishek Akkabathula <[email protected]>

added gaurd code to prevent infinite wait on sample ready interrupts.
In the case that we dont get any sample ready interrupt within given
time, return the apb to upper layer with zero bytes. In future when
async extraction is applied, we can again reuse the same apb. For now,
this case is not handled.

Signed-off-by: Abhishek Akkabathula <[email protected]>

ideally data extraction should happen every 20ms, however, during
testing, some times, it takes upto 60ms to extract a sample. As a
result, the delay keeps building up and resulting in fully filling
the circular audio tank. Hence, extract multiple sample when ever
possible to prevent the audio buildup in tank.

Signed-off-by: Abhishek Akkabathula <[email protected]>

enable barge-in via change of active flowset ID.

Implement host based recovery for NDP120. We simply reset the
device and reload the firmware.

Signed-off-by: Abhishek Akkabathula <[email protected]>