Stackful and stackless task unified

[joey-shao]

I have push async branch, the duty of this branch is to unified the stackful and stackless task in abstraction. To finish the whole work, we should create a real "async os". Stackful task is same to original task, needs kernel stack ,execution context and enable preempt; while stackless task don't need the explicitly execution context, kernel stack and disable kernel preempt, the executor self containing the whole execution condition.

Two major work I think,

1. should we redesign some task api, suck as park, unpark
2. remove block_on and spread 'async'

I am not sure my implementation here is follow the original purpose, please check and tell me.

[greatbridf]

Ditto.

My opinion is that we should rework the tasking system. In the current implementation, we use stackful coroutine as the fundamental tasking method, which has introduced tons of problems...

The new proposal is that we should run a stackless runtime on each core. So we can have only one stack for each cores. If the task should be stackful, we allocate a stack, call captured_trap_return to go into the environment, capture the needed exceptions (timer interrupt of course in order to get preempted and other needed) and leave other unrelated interrupts untouched (such as IRQs and unrelated page faults).

This would reduce the mass complexity in the current tasking code and reduce extra overheads.

I have also some commits on my local repo and have done some redesigning and work. It's expected to be pushed up soon...

[greatbridf]

Ditto.

My opinion is that we should rework the tasking system. In the current implementation, we use stackful coroutine as the fundamental tasking method, which has introduced tons of problems...

The new proposal is that we should run a stackless runtime on each core. So we can have only one stack for each cores. If the task should be stackful, we allocate a stack, call captured_trap_return to go into the environment, capture the needed exceptions (timer interrupt of course in order to get preempted and other needed) and leave other unrelated interrupts untouched (such as IRQs and unrelated page faults).

This would reduce the mass complexity in the current tasking code and reduce extra overheads.

I have also some commits on my local repo and have done some redesigning and work. It's expected to be pushed up soon...

In the new approach, the global stackful and stackless tasks are unified and have no difference in the view of the scheduler. The difference of stackfulness will be contained in the actual future object inside of the task.

We can also reuse the interfaces provided in the HAL's trap module.