Add a serial port flush operation which can be used to discard either
serial port data received but not read, or serial port written but not
transmitted, or both types of data.
Signed-off-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name>
Add an interface and implementation for a port scheduler that round-
robins between ports the scheduler is attached to, serializing command
execution among one or more MMPortSerial instances.
Theory of operation:
Sources (e.g. MMPort subclasses) register themselves with the scheduler.
Each source notifies the scheduler whenever its command queue depth changes,
for example when new commands are submitted, when commands are completed,
or when commands are canceled.
The scheduler will round-robin between all sources with pending commands,
sleeping when there are no pending commands from any source.
For each source with a pending command the scheduler will emit the
'send-command' signal with that source's ID. The given source should
send the next command in its queue to the modem.
When that command is finished (either successfully or with an error/timeout)
the source must call mm_port_scheduler_notify_command_done() to notify the
scheduler that it may advance to the next source with a pending command, if
any. If the 'send-command' signal and the notify_command_done() call are not
balanced the scheduler may stall.
Signed-off-by: Dan Williams <dan@ioncontrol.co>
By default all the commands we were sending through the serial port
were added at the tail of the pending queue, but we may want to queue
them at the head in very specific cases (e.g. while sending an SMS).
Response parsing was being done in different places for AT and QCDM subclasses;
in the case of AT it was being done early, before returning the byte array in
the mm_serial_port_command_finish() response. In the case of QCDM, it was being
done after mm_serial_port_command_finish(), and that was forcing every caller to
cleanup the response buffer once the response was processed.
With the new logic in this patch, the response is always parsed (i.e. looked for
a valid response or an error detected) before mm_serial_port_command_finish()
returns, and actually this method now returns a totally different GByteArray,
not the internal response buffer GByteArray, so there's no longer any need for
the caller to explicitly clean it up. The one doing the cleanup is the parser
method itself in every case.
This change also allows us to return serial port responses in idle, but that's
not changed for now as there's no immediate need.
Maciej S. Szmigiero
Dan Williams
Aleksander Morgado
Alfonso Sánchez-Beato