Linux内核分析之二——时间片轮转多道程序上下文切换机制之堆栈分析

作者：姚开健

原创作品转载请注明出处

《Linux内核分析》MOOC课程http://mooc.study.163.com/course/USTC-1000029000

在计算机中，能够保证计算机程序运行的一些重要机制包括：存储程序，函数调用堆栈机制，中断机制以及多道程序上下文切换机制。本文要介绍的就是在时间片轮转的情况下，多道程序上下文切换的机制是如何工作的。以一个小小的kernel，运行着几个程序，然后在时间片消耗完之后如何来切换程序，来模拟操作系统的进程切换机制。

首先由三个程序文件：

1、mypcb.h

/*
 *  linux/mykernel/mypcb.h
 *  Kernel internal PCB types

*  Copyright (C) 2013  Mengning
 *
 */
#define MAX_TASK_NUM        4
#define KERNEL_STACK_SIZE   1024*8
/* CPU-specific state of this task */
struct Thread {
    unsigned long		ip;
   unsigned long		sp;
};
typedef struct PCB{
    int pid;
    volatile long state;	/* -1 unrunnable, 0 runnable, >0 stopped */
    char stack[KERNEL_STACK_SIZE];
    /* CPU-specific state of this task */
    struct Thread thread;
    unsigned long	task_entry;
    struct PCB *next;
}tPCB;
void my_schedule(void);

2，mymain.c

#include <linux/types.h>
#include <linux/string.h>
#include <linux/ctype.h>
#include <linux/tty.h>
#include <linux/vmalloc.h>

#include "mypcb.h"
tPCB task[MAX_TASK_NUM];
tPCB * my_current_task = NULL;
volatile int my_need_sched = 0;
void my_process(void);

void __init my_start_kernel(void)
{
    int pid = 0;
    int i;
    /* Initialize process 0*/
    task[pid].pid = pid;
    task[pid].state = 0;/* -1 unrunnable, 0 runnable, >0 stopped */
    task[pid].task_entry = task[pid].thread.ip = (unsigned long)my_process;
    task[pid].thread.sp = (unsigned long)&task[pid].stack[KERNEL_STACK_SIZE-1];
    task[pid].next = &task[pid];
    /*fork more process */
    for(i=1;i<MAX_TASK_NUM;i++)
    {
        memcpy(&task[i],&task[0],sizeof(tPCB));
        task[i].pid = i;
        task[i].state = -1;
        task[i].thread.sp = (unsigned long)&task[i].stack[KERNEL_STACK_SIZE-1];
        task[i].next = task[i-1].next;
        task[i-1].next = &task[i];
    }
    /* start process 0 by task[0] */
    pid = 0;
    my_current_task = &task[pid];
	asm volatile(
    	"movl %1,%%esp\n\t" 	/* set task[pid].thread.sp to esp */
    	"pushl %1\n\t" 	        /* push ebp */
    	"pushl %0\n\t" 	        /* push task[pid].thread.ip */
    	"ret\n\t" 	            /* pop task[pid].thread.ip to eip */
    	"popl %%ebp\n\t"
    	: 
    	: "c" (task[pid].thread.ip),"d" (task[pid].thread.sp)	/* input c or d mean %ecx/%edx*/
	);
}   
void my_process(void)
{
    int i = 0;
    while(1)
    {
        i++;
        if(i%10000000 == 0)
        {
            printk(KERN_NOTICE "this is process %d -\n",my_current_task->pid);
            if(my_need_sched == 1)
            {
                my_need_sched = 0;
        	    my_schedule();
        	}
        	printk(KERN_NOTICE "this is process %d +\n",my_current_task->pid);
        }     
    }
}

3、myinterrupt.c

#include <linux/types.h>
#include <linux/string.h>
#include <linux/ctype.h>
#include <linux/tty.h>
#include <linux/vmalloc.h>

#include "mypcb.h"

extern tPCB task[MAX_TASK_NUM];
extern tPCB * my_current_task;
extern volatile int my_need_sched;
volatile int time_count = 0;

/*
 * Called by timer interrupt.
 * it runs in the name of current running process,
 * so it use kernel stack of current running process
 */
void my_timer_handler(void)
{
#if 1
    if(time_count%1000 == 0 && my_need_sched != 1)
    {
        printk(KERN_NOTICE ">>>my_timer_handler here<<<\n");
        my_need_sched = 1;
    } 
    time_count ++ ;  
#endif
    return;  	
}
void my_schedule(void)
{
    tPCB * next;
    tPCB * prev;
    if(my_current_task == NULL 
        || my_current_task->next == NULL)
    {
    	return;
    }
    printk(KERN_NOTICE ">>>my_schedule<<<\n");
    /* schedule */
    next = my_current_task->next;
    prev = my_current_task;
    if(next->state == 0)/* -1 unrunnable, 0 runnable, >0 stopped */
    {
    	/* switch to next process */
    	asm volatile(	
        	"pushl %%ebp\n\t" 	    /* save ebp */
        	"movl %%esp,%0\n\t" 	/* save esp */
        	"movl %2,%%esp\n\t"     /* restore  esp */
        	"movl $1f,%1\n\t"       /* save eip */	
        	"pushl %3\n\t" 
        	"ret\n\t" 	            /* restore  eip */
        	"1:\t"                  /* next process start here */
        	"popl %%ebp\n\t"
        	: "=m" (prev->thread.sp),"=m" (prev->thread.ip)
        	: "m" (next->thread.sp),"m" (next->thread.ip)
    	); 
    	my_current_task = next; 
    	printk(KERN_NOTICE ">>>switch %d to %d<<<\n",prev->pid,next->pid);   	
    }
    else
    {
        next->state = 0;
        my_current_task = next;
        printk(KERN_NOTICE ">>>switch %d to %d<<<\n",prev->pid,next->pid);
    	/* switch to new process */
    	asm volatile(	
        	"pushl %%ebp\n\t" 	    /* save ebp */
        	"movl %%esp,%0\n\t" 	/* save esp */
        	"movl %2,%%esp\n\t"     /* restore  esp */
        	"movl %2,%%ebp\n\t"     /* restore  ebp */
        	"movl $1f,%1\n\t"       /* save eip */	
        	"pushl %3\n\t" 
        	"ret\n\t" 	            /* restore  eip */
        	: "=m" (prev->thread.sp),"=m" (prev->thread.ip)
        	: "m" (next->thread.sp),"m" (next->thread.ip)
    	);          
    }   
    return;	
}

程序运行过程如：

从上面的文件可以看到，我们定义一个PCB，里面有进程的标号，堆栈指针，堆栈，eip，下一进程指针等等，程序首先在void __init my_start_kernel(void)函数执行，先初始化0号进程，即task[0]，

 /* Initialize process 0*/
    task[pid].pid = pid;
    task[pid].state = 0;/* -1 unrunnable, 0 runnable, >0 stopped */
    task[pid].task_entry = task[pid].thread.ip = (unsigned long)my_process;
    task[pid].thread.sp = (unsigned long)&task[pid].stack[KERNEL_STACK_SIZE-1];
    task[pid].next = &task[pid];

接着初始化1,2,3号进程，

  /*fork more process */
    for(i=1;i<MAX_TASK_NUM;i++)
    {
        memcpy(&task[i],&task[0],sizeof(tPCB));
        task[i].pid = i;
        task[i].state = -1;
        task[i].thread.sp = (unsigned long)&task[i].stack[KERNEL_STACK_SIZE-1];
        task[i].next = task[i-1].next;
        task[i-1].next = &task[i];
    }

接着运行0号进程，这时需要详细说明进程如何开始运行：

 /* start process 0 by task[0] */
    pid = 0;
    my_current_task = &task[pid];
	asm volatile(
    	"movl %1,%%esp\n\t" 	/* set task[pid].thread.sp to esp */
    	"pushl %1\n\t" 	        /* push ebp */
    	"pushl %0\n\t" 	        /* push task[pid].thread.ip */
    	"ret\n\t" 	            /* pop task[pid].thread.ip to eip */
    	"popl %%ebp\n\t"
    	: 
    	: "c" (task[pid].thread.ip),"d" (task[pid].thread.sp)	/* input c or d mean %ecx/%edx*/
	);
}   

0号进程首先运行上面的汇编代码，分别是1，把进程PCB初始化时的栈顶指针送到CPU的esp，2，ebp压栈，因为是0号进程，ebp与esp相同，所以可以用pushl %esp，3，初始化的eip压栈再出栈，此时程序eip为函数void my_process(void)的地址，程序跳转到此函数运行。如果系统的时钟中断函数my_timer_handler时间片运行完，则把可以切换程序标志置为true，所以当my_process函数运行足够时间后，便会访问切换标志，如果为true，则进行程序切换，访问my_schedule(void)函数。

当从0号进程切换至1号进程时，执行

 else
    {
        next->state = 0;
        my_current_task = next;
        printk(KERN_NOTICE ">>>switch %d to %d<<<\n",prev->pid,next->pid);
    	/* switch to new process */
    	asm volatile(	
        	"pushl %%ebp\n\t" 	    /* save ebp */
        	"movl %%esp,%0\n\t" 	/* save esp */
        	"movl %2,%%esp\n\t"     /* restore  esp */
        	"movl %2,%%ebp\n\t"     /* restore  ebp */
        	"movl $1f,%1\n\t"       /* save eip */	
        	"pushl %3\n\t" 
        	"ret\n\t" 	            /* restore  eip */
        	: "=m" (prev->thread.sp),"=m" (prev->thread.ip)
        	: "m" (next->thread.sp),"m" (next->thread.ip)
    	);          
    }   

这是进程切换的关键步骤，首先1，pushl %ebp，pushl %esp，保存刚才运行的进程的ebp和esp，2，接着把新进程原来初始化的esp（包括ebp）送到CPU的esp,ebp，接着保存旧进程的eip（$1f,1的意思是标号1的代码段的地址，f意思是forward，即标号1下面的地址）3，接着把新进程的eip压栈再出栈，程序跳转至my_process函数执行，接着以此类推至3号进程切换至0号进程，执行：

   if(next->state == 0)/* -1 unrunnable, 0 runnable, >0 stopped */
    {
    	/* switch to next process */
    	asm volatile(	
        	"pushl %%ebp\n\t" 	    /* save ebp */
        	"movl %%esp,%0\n\t" 	/* save esp */
        	"movl %2,%%esp\n\t"     /* restore  esp */
        	"movl $1f,%1\n\t"       /* save eip */	
        	"pushl %3\n\t" 
        	"ret\n\t" 	            /* restore  eip */
        	"1:\t"                  /* next process start here */
        	"popl %%ebp\n\t"
        	: "=m" (prev->thread.sp),"=m" (prev->thread.ip)
        	: "m" (next->thread.sp),"m" (next->thread.ip)
    	); 
    	my_current_task = next; 
    	printk(KERN_NOTICE ">>>switch %d to %d<<<\n",prev->pid,next->pid);   	
    }

这时的进程切换也是与之前的切换大同小异，也是先保存旧进程的ebp,esp，接着把新进程保存的esp恢复，然后eip压栈再出栈，跳转至1：处执行，ebp出栈。

通过此程序我们知道，进程的运行开始时需要把栈顶地址（即栈数组的末尾地址）送至CPU的esp，并初始化ebp，接着把进程需要运行的函数地址压栈，再ret（即popl %eip），程序就能进入函数运行，进程启动完毕。

在程序进行切换时，需要保持旧进程的ebp,esp，然后再按照新进程的运行开始过程（如上面所说）一样初始化新进程的相关数据。