之前把rtems上支持的所有的调度器介绍了一遍，但是一直都没有说上下文是如何切换的。其实上下文切换的逻辑就是保存线程TCB和通用寄存器，然后根据新的线程TCB信息进行运行。本文从上下文切换的角度来讲清楚RTEMS中是怎么完成上下文切换的

从schedule函数角度看

对于任何的任务，调用schedule就是根据当前任务的情况执行调度，那么主动调用如下函数


( *scheduler->Operations.schedule )( scheduler, the_thread );

这里不同的调度算法实现的回调不一样，为了方便介绍，这里以优先级调度为例，那么最后调用函数如下


static inline void _Scheduler_uniprocessor_Update_heir(
  Thread_Control *heir,
  Thread_Control *new_heir
)
{
  _Assert( heir != new_heir );
#if defined(RTEMS_SMP)
  /*
   * We need this state only for _Thread_Get_CPU_time_used_locked().  Cannot
   * use _Scheduler_Thread_change_state() since THREAD_SCHEDULER_BLOCKED to
   * THREAD_SCHEDULER_BLOCKED state changes are illegal for the real SMP
   * schedulers.
   */
  heir->Scheduler.state = THREAD_SCHEDULER_BLOCKED;
  new_heir->Scheduler.state = THREAD_SCHEDULER_SCHEDULED;
#endif
  _Thread_Update_CPU_time_used( heir, _Thread_Get_CPU( heir ) );
  _Thread_Heir = new_heir;
  _Thread_Dispatch_necessary = true;
}

再设置完一些state之后，我们重点关注如下两个值的设置


  _Thread_Heir = new_heir;
  _Thread_Dispatch_necessary = true;

这里的new_heir是即将要运行的任务，他是Thread_Control * 指针。而_Thread_Dispatch_necessary是Per_CPU_Control结构体的一个成员dispatch_necessary，我们通过dispatch_necessary来判断当前任务是否要scheduler。

从_Thread_Do_dispatch函数角度看

我们当任务在合适的实际，我们可以设置dispatch enable，或者直接调用_Thread_Do_dispatch函数，这样线程就直接开始开始调度的实际行为。我们看起核心代码如下


  do {
    Thread_Control                     *heir;
    heir = _Thread_Get_heir_and_make_it_executing( cpu_self );

    if ( heir == executing ) {
      break;
    }

    _ISR_Local_enable( level );

    _Thread_Save_fp( executing );
    _Context_Switch( &executing->Registers, &heir->Registers );
    _Thread_Restore_fp( executing );
    _User_extensions_Thread_switch( NULL, executing );

    cpu_self = _Per_CPU_Get();

    _ISR_Local_disable( level );
  } while ( cpu_self->dispatch_necessary );

可以看到，只要dispatch_necessary设置为true，那么任务就会执行调度。执行调度的函数为_Context_Switch，它会根据每个架构的实现来运行上下文切换，我们基于armv8架构来介绍

从_CPU_Context_switch函数来看

这里主要是Arm64的实现，先看源码


DEFINE_FUNCTION_AARCH64(_CPU_Context_switch)
	.globl	_CPU_Context_switch_no_return
	.set	_CPU_Context_switch_no_return, _CPU_Context_switch
#ifdef AARCH64_MULTILIB_ARCH_V8_ILP32
/* Sanitize inputs for ILP32 ABI */
	mov w0, w0
	mov w1, w1
  #ifdef RTEMS_SMP
    #define reg_2 x2
  #else
    #define reg_2 w2
  #endif
#else
#define reg_2 x2
#endif

/* Start saving context */
	GET_SELF_CPU_CONTROL	reg_2
	ldr	w3, [x2, #PER_CPU_ISR_DISPATCH_DISABLE]  // 读取isr_dispatch_disable

	stp x19, x20, [x0]               // 保存aapcs中规定的callee寄存器 也就是x19-x28
	stp x21, x22, [x0, #0x10]
	stp x23, x24, [x0, #0x20]
	stp x25, x26, [x0, #0x30]
	stp x27, x28, [x0, #0x40]
	stp fp,  lr,  [x0, #0x50]       // 保存fp和lr
	mov x4,  sp                     // 保存sp寄存器
	str x4,  [x0, #0x60]

#ifdef AARCH64_MULTILIB_VFP
	add	x5, x0, #AARCH64_CONTEXT_CONTROL_D8_OFFSET
	stp d8,  d9,  [x5]              // 保存浮点寄存器
	stp d10, d11, [x5, #0x10]
	stp d12, d13, [x5, #0x20]
	stp d14, d15, [x5, #0x30]
#endif

	str	x3, [x0, #AARCH64_CONTEXT_CONTROL_ISR_DISPATCH_DISABLE] //将本线程的isr_dispatch_disable设置到x0

#ifdef RTEMS_SMP
	/*
	 * The executing thread no longer executes on this processor.  Switch
	 * the stack to the temporary interrupt stack of this processor.  Mark
	 * the context of the executing thread as not executing.
	 */
	dmb	SY          //所有数据dmb
	add	sp, x2, #(PER_CPU_INTERRUPT_FRAME_AREA + CPU_INTERRUPT_FRAME_SIZE) //分配一个栈
	mov	x3, #0  //将0写到is_executing中
	strb	w3, [x0, #AARCH64_CONTEXT_CONTROL_IS_EXECUTING_OFFSET]

.L_check_is_executing:

	/* Check the is executing indicator of the heir context */
	add	x3, x1, #AARCH64_CONTEXT_CONTROL_IS_EXECUTING_OFFSET //获取is_executing的值
	ldaxrb	w4, [x3] //原子读取x3的内容到w4
	cmp	x4, #0 
	bne	.L_get_potential_new_heir  //如果已经其他cpu执行则跳到其他线程tcb上重新判断

	/* Try to update the is executing indicator of the heir context */
	mov	x4, #1
	stlxrb	w5, w4, [x3]   // 将is_executing设置回去，结果保存在w5
	cmp	x5, #0  
	bne	.L_get_potential_new_heir // 如果设置失败，则跳到其他tcb上重新判断
	dmb	SY  // dmb所有数据
#endif

/* Start restoring context */
.L_restore:
#if !defined(RTEMS_SMP) && defined(AARCH64_MULTILIB_HAS_LOAD_STORE_EXCLUSIVE)
	clrex  // 清空独占监视器
#endif

	ldr	x3, [x1, #AARCH64_CONTEXT_CONTROL_THREAD_ID_OFFSET]  // 加载下一个线程的thread_id

	ldr	x4, [x1, #AARCH64_CONTEXT_CONTROL_ISR_DISPATCH_DISABLE] // 加载下一个线程的isr_dispatch_disable

#ifdef AARCH64_MULTILIB_VFP
	add	x5, x1, #AARCH64_CONTEXT_CONTROL_D8_OFFSET // 加载下一个线程的浮点寄存器
	ldp d8,  d9,  [x5]
	ldp d10, d11, [x5, #0x10]
	ldp d12, d13, [x5, #0x20]
	ldp d14, d15, [x5, #0x30]
#endif

	msr	TPIDR_EL0, x3   // 将x3设置到TPIDR_EL0

	str	w4, [x2, #PER_CPU_ISR_DISPATCH_DISABLE]  // 更新isr_dispatch_disable

	ldp x19, x20, [x1]  // 恢复下一个线程的callee寄存器
	ldp x21, x22, [x1, #0x10]
	ldp x23, x24, [x1, #0x20]
	ldp x25, x26, [x1, #0x30]
	ldp x27, x28, [x1, #0x40]
	ldp fp,  lr,  [x1, #0x50]   // 恢复fp和lr
	ldr x4,  [x1, #0x60]     
	mov sp,  x4       //恢复sp
	ret

上面代码尽可能给出了注释，有些补充解释如下

x0和x1

我们知道_CPU_Context_switch的原型是：


_CPU_Context_switch( _executing, _heir )

所以x0 是当前线程tcb，x1是下一个线程的tcb

Context_Control结构体

我们知道需要保存x19-x30 sp等寄存器，那么保存的地方在TPIDR_EL0上，TPIDR_EL0上存在的是Context_Control的指针，所以Context_Control结构体如下。


typedef struct {
  uint64_t register_x19;
  uint64_t register_x20;
  uint64_t register_x21;
  uint64_t register_x22;
  uint64_t register_x23;
  uint64_t register_x24;
  uint64_t register_x25;
  uint64_t register_x26;
  uint64_t register_x27;
  uint64_t register_x28;
  uint64_t register_fp;
  uint64_t register_lr;
  uint64_t register_sp;
  uint64_t isr_dispatch_disable;
  uint64_t thread_id;
#ifdef AARCH64_MULTILIB_VFP
  uint64_t register_d8;
  uint64_t register_d9;
  uint64_t register_d10;
  uint64_t register_d11;
  uint64_t register_d12;
  uint64_t register_d13;
  uint64_t register_d14;
  uint64_t register_d15;
#endif
#ifdef RTEMS_SMP
  volatile bool is_executing;
#endif
} Context_Control;