Looking at the following excerpt from FreeRTOSConfig.h
if DATA_MODEL == DATAMODELSMALL
#define configMINIMAL_STACK_SIZE ( ( unsigned short ) 110 )
#define configMINIMAL_STACK_SIZE ( ( unsigned short ) 80 )
I cannot understand why the stack size requirement should be bigger in the small data model than in the larger data models. I'd expect exactly the opposite. Any clue?
You are right, looking at the port layer code, it is not clear why it is done that way around. There are a few possibilities:
1) It is a mistake, but that is doubtful as the stack sizes are set to allow the demo applications to squeeze in as much as possible and there would have been problems if it was the wrong way around.
2) It might be that the demo project does something different depending on the memory model, and that the larger demo requires a larger stack (configMINIMALSTACKSIZE is used by the kernel is just one place, and that is to dimension the size of the stack used by the idle task. The demo projects use it all over the place though as a convenient way of dimensioning stacks of tasks that are run on all the RTOS ports).
3) The libraries or function call stack used in the small model take up more space, perhaps as the registers hold smaller values more data has to be pushed onto the stack. I don't know that this is the case, just speculating.