Remove sched_elect in serialchar_callback

[mlouielu]

DO NOT USE sched_elect inside the irq.

If a context switch is done inside the irq,
it will make v7m-head.S pop {pc} to a undefined place,
or not able to pop {pc} out.

cc. @yenwu.

[iankuan]

@mlouielu , let me double-check it. You mean that triggering scheduler in irq context would violate flow of exception handler due to exception return ?

If so, you might be right. How about the result of test suites and behavior of minishell?

[mlouielu]

Although I think this is a correct fix, please hold off a second.

I still trying to figure out why we can not do a context switch inside a regular ISR....it is weird.

[mlouielu]

Confirm that this have some bug inside, reproduce step:

• type in some words in minishell
• sleep will failed.

[mlouielu]

Currently, I'm working on #36 to ensure that this problem isn't coming from the scheduler.

[jserv]

The use of sched_elect(0) SHOULD be deprecated. Let's introduce new kernel function schedule().

[jserv]

+1
I defer to @yenwu.

[iankuan]

@mlouielu , I already confirmed that that's the reason for serial halted. Please mention what's wrong with that in perspective of exception flow and the benefits to following exception service convention. Unexpected exception return increases the probability of occurrence of faults, such as INVPC, INVSTATE. In this situation, we can't make sure that it wouldn't cause faults or invalid interrupt raising due to unfinished interrupt handling.

[iankuan]

Moreover, PendSV handler is involved with sched_elect(). It's in irq context and works well due to the lowest priority of pendsv. It's always at the bottom of interrupt handle stack and only exception return to Thread mode. It fits the requirements of the scheduler.