[Bug] 关于smp环境下内核中信号处理部分锁相关问题

RT-Thread Version

master

Hardware Type/Architectures

all

Develop Toolchain

Other

Describe the bug

1. _signal_deliver 部分代码

  //获取信号自旋锁
   level = rt_spin_lock_irqsave(&_thread_signal_lock);

    /* thread is not interested in pended signals */
    if (!(tid->sig_pending & tid->sig_mask))
    {
        rt_spin_unlock_irqrestore(&_thread_signal_lock, level);
        return;
    }

 //没有获取调度器自旋锁 _mp_scheduler_lock的情况下读取调度相关状态位
// 其他核心可能同时修改这个字段

    if ((RT_SCHED_CTX(tid).stat & RT_THREAD_SUSPEND_MASK) == RT_THREAD_SUSPEND_MASK)
    {
        /* resume thread to handle signal */
#ifdef RT_USING_SMART
        rt_thread_wakeup(tid);
#else
        rt_thread_resume(tid);
#endif
        /* add signal state */
        RT_SCHED_CTX(tid).stat |= (RT_THREAD_STAT_SIGNAL | RT_THREAD_STAT_SIGNAL_PENDING);

        rt_spin_unlock_irqrestore(&_thread_signal_lock, level);

        /* re-schedule */
        rt_schedule();
    }

对于上面的代码，我存在两点疑惑，如果有大佬能解答，十分感激：

1. 能通过两个不同自旋锁保护状态字的不同位域吗？如使用 _mp_scheduler_lock保护调度相关的状态位域，通过 _thread_signal_lock保护信号相关位域。

2. 为什么上面代码在读取调度相关位域时没有加 _mp_scheduler_lock 进行保护，而是直接读取判断 if ((RT_SCHED_CTX(tid).stat & RT_THREAD_SUSPEND_MASK) == RT_THREAD_SUSPEND_MASK)，我注意到另外一个函数的实现是要求持有 _mp_scheduler_lock 的：

rt_uint8_t rt_sched_thread_is_suspended(struct rt_thread *thread)
{
    RT_SCHED_DEBUG_IS_LOCKED;// 检测是否持有  _mp_scheduler_lock
    return (RT_SCHED_CTX(thread).stat & RT_THREAD_SUSPEND_MASK) == RT_THREAD_SUSPEND_MASK;
}

Other additional context

No response

Start by reading _signal_deliver and rt_sched_thread_is_suspended, then trace the ownership rules for _thread_signal_lock and _mp_scheduler_lock in the SMP scheduler code. Compare which fields of RT_SCHED_CTX(tid).stat each lock protects and inspect related call sites. Done means the locking question is resolved and any required correction is covered by an appropriate regression test.