To make libertm thread-safe, calls to ertm_proto_cycle must be completed
in an atomic manner, ensuring that the serial link to the wrpc is used
in an exclusive manner by each client thread.

There is only one exception to this rule, and a crucial one: there are
separate operations for sending config values to the wrpc, and for
committing the sent values. The doublet of the two calls to
set_board_config and commit_board_config, which are called only in one
place and always in sequence, must be atomic in its entirety, which is
most simply achieved by defining a larger critical section in
update_board_config. This is ugly, but the shortest path to fully atomic
operations from the way the code was originally organized.

This refactoring allows to simplify the locking: this way, the lock file
descriptor is accessible to ertm_proto_cycle, which is (nearly) the one
and only critical section.

ertm bootloader interprets
which is
as a very destructive command, which has consequences when cmds are
accidentally sent while the board is in bootloader mode.

This takes care to avoid this kind of collision and the ensuing
deletion of firmware/bricking of the board.
Signed-off-by: Juan David Gonzalez Cobas <dcobas@cern.ch>
Reported-by: Tomasz Wlostowski <twlostow@cern.ch>

This has the purpose of making some definitions
public without having to pull in the entire contents of board-state.h
and all the myriad .h files this would #include

ertm14: ensure correct settings of DDS FTW/Level Adjust when the NCO reset is subscribed (see commets)

dev: indicate force-triggered channels as armed so that fine_pulse_gen_is_triggered() can take forced pulses into consideration too

ertm_get_channel_enable returned the logical NOT of the enable flag,
because the out_state byte has an enum of possible values of which ON is
0 and OFF is 2... This patch makes justice to lo/ref channels and gets
the right value in return

It was mysteriously overlooked when implemented

Add API export of this function

This branch implements a separate function to obtain the enable/disable
status of channels.
Reported-by: Saul Novel <saul.novel.gonzalez@cern.ch>

although frequency and sync status are channelno-independent for LO/REF,
the enable/disable status is not. Thus, the channel number validation
must be refactored to filter out bad LO/REF channel numbers *only* in
ertm_get_output_state

Somehow this went overlooked in the design of the API

Greatly increases performance of the Linux control API. More explanation in the comments of uart_link_recv().