It's common for the rx_queue to be written to in ISR context (by the
HCI driver). This means that if there's lots of data coming in from
the driver the rx_queue might get empty very rarely. With the current
code this means that the rx fiber might end up not yielding to other
higher priority runnable fibers for long durations (as long as there's
data in the rx_queue).
To solve the issue, call fiber_yield() explicitly after each processed
buffer from the rx_queue. This way we give other fibers a chance to
run even if there's a heavy flow of data from the Bluetooth
controller.
Change-Id: Ib3dbe6536a62360ad5153ed12eee8489645e4109
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
b76426de6f