[BUG] Round robin scheduling when SMP is enabled causes crashes

[pussuw]

Description / Steps to reproduce the issue

The round robin scheduling logic causes crashes via:

nuttx/sched/sched/sched_removereadytorun.c

Lines 287 to 294 in db31e56

	bool nxsched_remove_readytorun(FAR struct tcb_s *tcb)
	{
	if (tcb->task_state == TSTATE_TASK_RUNNING)
	{
	DEBUGASSERT(tcb->cpu == this_cpu());
	nxsched_remove_running(tcb);
	return true;
	}

The crash occurs when the task being pre-empted is running on a different CPU / core. The call tree is as follows:

_assert() at assert.c:906 0xa001f690	
__assert() at lib_assert.c:38 0xa0005d4a	
nxsched_remove_readytorun() at sched_removereadytorun.c:291 0xa0027f2c	
nxsched_reprioritize_rtr() at sched_reprioritizertr.c:67 0xa00214f2	
nxsched_process_roundrobin() at sched_roundrobin.c:141 0xa003c800	
nxsched_cpu_scheduler() at sched_processtimer.c:77 0xa003c424	
nxsched_process_scheduler() at sched_processtimer.c:134 0xa003c424	
nxsched_process_timer() at sched_processtimer.c:189 0xa003c424	

This happens when the system tick advances. An arbitrary CPU will handle the timer interrupt and advance the system ticker. It will then try to run the round robin scheduling logic for every CPU.

nuttx/sched/sched/sched_processtimer.c

Lines 128 to 138 in db31e56

	flags = enter_critical_section();
	/* Perform scheduler operations on all CPUs */
	for (i = 0; i < CONFIG_SMP_NCPUS; i++)
	{
	nxsched_cpu_scheduler(i);
	}
	leave_critical_section(flags);
	}

Pre-empting a task on another CPU directly will obviously not work -> crash.

What do you think @hujun260 should this be handled by your SMP call logic, or something else ?

On which OS does this issue occur?

[OS: Linux]

What is the version of your OS?

Ubuntu

NuttX Version

master

Issue Architecture

[Arch: all]

Issue Area

[Area: Kernel]

Verification

• I have verified before submitting the report.

[hujun260]

Please verify again with the latest code（apache/master）

[pussuw]

Oh, I did not notice you had a patch for this already. I'll try it on Monday but I'm sure this bug report is not valid anymore.

[pussuw]

For reference: fixed by #14611

[pussuw]

The issue is verified to be fixed by #14611