A portable pre‐emptive microkernel with message passing

                          Martin Åberg

                       martin@fripost.org

                              2012


This  kernel  is  designed with the aims of being easy to use and
understand, containing suitable inter process  communication  fa‐
cilities,  having  a  straight  forward  implementation and being
highly portable.

Each execution unit, task, has its own stack and context.  Multi‐
tasking  is  pre‐emptive  and priority based, with equal priority
tasks being given access to the processor in a round robin  fash‐
ion.

Inter‐process communication facilities include signaling and mes‐
sage passing. The kernel signal primitive allows a task or inter‐
rupt  service  routine to signal any other task asynchronously. A
task can issue a blocking wait kernel call to wait for a specific
signal  to  arrive.  There  is support in the kernel for handling
multiple signal channels per task, and this facility is used  for
the  message passing sub system. The signals don’t carry any pay‐
load data themselves though.

Both asynchronous and synchronous message passing  is  supported.
Each task has its own message queue in which messages are put and
delivered in FIFO order.

Time spent in the kernel is easily bounded when the maximum  num‐
ber  of tasks is known. Actually, the there is only one mechanism
in the kernel whose time complexity does depend  on  a  variable.
That  is the enquing in priority order of a task into the list or
ready tasks. No dynamic memory is allocated when the multitasking
kernel  is  up and running. These properties make the microkernel
suitable for real time operation.

One design goal was to make the kernel interface and  implementa‐
tion  portable. This goal is achieved by dividing the kernel code
into a platform independent part and a platform  dependent  part.
When porting the kernel to a new platform, only the interface for
the platform dependent part has to be  implemented.  Current  the
only implementation targets Atmel AVR.