mirror of https://github.com/hsoft/collapseos.git
44 lines
1.7 KiB
Plaintext
44 lines
1.7 KiB
Plaintext
# Asynchronous Communications Interface Adapters
|
|
|
|
Machines talking to each other is generally useful and they
|
|
often use ACIA devices to do so. Collapse OS has drivers for
|
|
a few chips of this type and they all have a similar approach:
|
|
unbuffered communication using RTS/CTS handshaking as flow con-
|
|
trol.
|
|
|
|
The reason for being unbuffered is simplicity and RAM. The logic
|
|
to implement input buffering is non-trivial and, alone, doesn't
|
|
buy us much in terms of reliability: you still have to signal
|
|
the other side when your buffer is nearly full.
|
|
|
|
Because we don't really need speed, we adopt a one-byte-at-once
|
|
approach: The RTS flag is always high (signalling that it's not
|
|
ready for communication) *except* when calling the ACIA driver's
|
|
"read" word, which is blocking.
|
|
|
|
That "read" word will pull RTS low, wait for a byte, then pull
|
|
it high again.
|
|
|
|
This slows down communication, but it's simple and reliable.
|
|
|
|
Note that this doesn't help making communications with modern
|
|
systems (which are much faster than a typical Collapse OS
|
|
machine and have their buffer output faster than the RTS flag
|
|
can be raised) very much. We have to take extra care, when
|
|
communicating from modern system, not to send too much data too
|
|
fast. But for COS-to-COS communication, this simple system
|
|
works.
|
|
|
|
# Broken hardware
|
|
|
|
Some designs are broken with this scheme. For example, the
|
|
RS2014 SIO module hard-wires CTS to GND because the FTDI
|
|
connector doesn't have such a pin (modern computers can always
|
|
handle the load).
|
|
|
|
In these cases, a solution would be to use Break signals as a
|
|
workaround, but I prefer avoiding complexity for now. So when
|
|
you deal with broken design, you'll have to sidestep it either
|
|
by implementing your own Break handling or by lowering com-
|
|
munication speed.
|