Embedded system driver primary [4] - under the basis of character device driver _ concurrency control

In order to achieve multiple processes using our character device at the same time,我们引入“并发控制”这一概念.

一、上下文和并发场合

执行流：有开始有结束总体顺序执行的一段代码  context

应用编程：任务上下文

内核编程：

1. 任务上下文：五状态 可阻塞

    a. 应用进程或线程运行在用户空间

    b. 应用进程或线程运行在内核空间（通过调用syscall来间接使用内核空间）

    c. 内核线程始终在内核空间

2. 异常上下文：不可阻塞

    中断上下文

竞态：多任务并行执行时,If the same resource is operated at the same time,会引起资源的错乱,这种错乱情形被称为竞态

共享资源：可能会被多个任务同时使用的资源

临界区：操作共享资源的代码段

为了解决竞态,需要提供一种控制机制,来避免在同一时刻使用共享资源,This mechanism is called a concurrency control mechanism

并发控制机制分类：

1. 原子操作类

2. 忙等待类

3. 阻塞类

 

通用并发控制机制的一般使用套路：

/*互斥问题：*/

并发控制机制初始化为可用

P操作

临界区

V操作

/*同步问题：*/

//并发控制机制初始化为不可用

//先行方：

.....

V操作

   

//后行方：

P操作

.....

二、中断屏蔽（了解）

一种同步机制的辅助手段

禁止本cpu中断                使能本cpu中断

local_irq_disable();        local_irq_enable();        

local_irq_save(flags);      local_irq_restore(flags);   与cpu的中断位相关

local_bh_disable();         local_bh_enable();          与中断低半部有关,关闭、打开软中断

禁止中断

临界区    //临界区代码不能占用太长时间,需要很快完成

打开中断

 

适用场合：中断上下文与某任务共享资源时,或多个不同优先级的中断上下文间共享资源时

 

三、原子变量（掌握）

原子变量：存取不可被打断的特殊整型变量

a.设置原子量的值

void atomic_set(atomic_t *v,int i); //设置原子量的值为i

atomic_t v = ATOMIC_INIT(0);    //定义原子变量v并初始化为0

v = 10;//错误

b.获取原子量的值

atomic_read(atomic_t *v);       //返回原子量的值

c.原子变量加减

void atomic_add(int i,atomic_t *v);//原子变量增加i

void atomic_sub(int i,atomic_t *v);//原子变量减少i

d.原子变量自增自减

void atomic_inc(atomic_t *v);//原子变量增加1

void atomic_dec(atomic_t *v);//原子变量减少1

e.操作并测试：运算后结果为0则返回真,否则返回假

int atomic_inc_and_test(atomic_t *v);

int atomic_dec_and_test(atomic_t *v);

int atomic_sub_and_test(int i,atomic_t *v);

原子位操作方法：

a.设置位

void set_bit(nr, void *addr);       //设置addr的第nr位为1

b.清除位

void clear_bit(nr , void *addr);    //清除addr的第nr位为0

c.改变位

void change_bit(nr , void *addr);   //改变addr的第nr位为1

d.测试位

void test_bit(nr , void *addr);     //测试addr的第nr位是否为1

适用场合：共享资源为单个整型变量的互斥场合

示例代码：atomic.c

#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/fs.h>
#include <linux/cdev.h>
#include <asm/atomic.h>

struct openonce_dev {
	struct cdev mydev;
	atomic_t open_flag;
};

int major = 11;
int minor = 0;
int num = 1;

struct openonce_dev gmydev;

int myopen (struct inode *pnode, struct file *pfile)
{
	struct openonce_dev *pgmydev = NULL;
	pfile->private_data = (void *)container_of(pnode->i_cdev,struct openonce_dev,mydev);
	pgmydev = (struct openonce_dev *)pfile->private_data;
	if(atomic_dec_and_test(&pgmydev->open_flag))
	{
		return 0;
	}
	else
	{
		atomic_inc(&pgmydev->open_flag);
		printk("The device is open already\n");
		return -1;
	}
	

	return 0;
}
 
int myclose (struct inode *pnode, struct file *pfile)
{
	struct openonce_dev *pgmydev = (struct openonce_dev *)pfile->private_data;
	atomic_set(&gmydev.open_flag,1);

	return 0;
}
 
struct file_operations myops = {
	.owner = THIS_MODULE,
	.open = myopen,
	.release = myclose,
};


int __init atomic_init(void)
{
	int ret = 0;

	dev_t devno = MKDEV(major,minor);
	ret = register_chrdev_region(devno,num,"openonce");
 
	if(ret)
	{
		ret = alloc_chrdev_region(&devno,minor,num,"openonce");
		if(ret)
		{
			printk("devno failed.\n");
			return -1;
		}
		major = MAJOR(devno);
		minor = MINOR(devno);
	}
 
	/*给mydev指定操作函数集*/
	cdev_init(&gmydev.mydev,&myops);

	/*将mydev添加到内核对应的数据结构里*/
	gmydev.mydev.owner = THIS_MODULE;
	cdev_add(&gmydev.mydev,devno,num);

	atomic_set(&gmydev.open_flag,1);

	return 0;
}

void __exit atomic_exit(void)
{
	dev_t devno = MKDEV(major,minor);
 
	cdev_del(&gmydev.mydev);
 
	unregister_chrdev_region(devno, num);
}


MODULE_LICENSE("GPL");
MODULE_AUTHOR("imysy_22");
MODULE_DESCRIPTION("This is a script explained by the author who's name is imysy_22.");
MODULE_ALIAS("HI");

module_init(atomic_init);
module_exit(atomic_exit);

openonce_app.c

#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <fcntl.h>

int main(int argc,char *argv[])
{

	if(argc < 2)
	{
		printf("The argument is too few.\n");
		return -1;
	}

	int fd = -1;

	fd = open(argv[1],O_RDONLY);
	if(fd < 0)
	{
		printf("open failed\n");
		return -1;
	}

	while(1)
	{

	}

	close(fd);
	fd = -1;

	return 0;
}

Makefile:

ifeq ($(KERNELRELEASE),)

ifeq ($(ARCH),arm)
KERNELDIR ?= /home/linux/fs4412/linux-3.14
ROOTFS ?= /opt/4412/rootfs
else
KERNELDIR ?= /lib/modules/$(shell uname -r)/build
endif
PWD := $(shell pwd)

modules:
	$(MAKE) -C $(KERNELDIR) M=$(PWD) modules

modules_install:
	$(MAKE) -C $(KERNELDIR) M=$(PWD) modules INSTALL_MOD_PATH=$(ROOTFS) modules_install

clean:
	rm -rf  *.o  *.ko  .*.cmd  *.mod.*  modules.order  Module.symvers   .tmp_versions

else
obj-m += atomic.o


endif

 There is only one application at a timeappThis driver can be opened.

四、自旋锁：基于忙等待的并发控制机制

a.定义自旋锁

spinlock_t  lock;

b.初始化自旋锁

spin_lock_init(spinlock_t *);

c.获得自旋锁

spin_lock(spinlock_t *);    //成功获得自旋锁立即返回,否则自旋在那里直到该自旋锁的保持者释放

spin_trylock(spinlock_t *); //成功获得自旋锁立即返回真,否则返回假,而不是像上一个那样"在原地打转”

d.释放自旋锁

spin_unlock(spinlock_t *);

 

```

#include <linux/spinlock.h>

定义spinlock_t类型的变量lock

spin_lock_init(&lock)后才能正常使用spinlock

 

spin_lock(&lock);

临界区

spin_unlock(&lock);

```

 

适用场合：

1. 异常上下文之间或异常上下文与任务上下文之间共享资源时

2. 任务上下文之间且临界区执行时间很短时

3. 互斥问题

示例代码：spinlock.c

#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/fs.h>
#include <linux/cdev.h>
#include <asm/spinlock.h>

struct openonce_dev {
	struct cdev mydev;
	int open_flag;
	spinlock_t lock;
};

int major = 11;
int minor = 0;
int num = 1;

struct openonce_dev gmydev;

int myopen (struct inode *pnode, struct file *pfile)
{
	struct openonce_dev *pgmydev = NULL;
	pfile->private_data = (void *)container_of(pnode->i_cdev,struct openonce_dev,mydev);
	pgmydev = (struct openonce_dev *)pfile->private_data;

	spin_lock(&pgmydev->lock);
	if(pgmydev->open_flag)
	{
		spin_unlock(&pgmydev->lock);
		pgmydev->open_flag = 0;
		return 0;
	}
	else
	{
		spin_unlock(&pgmydev->lock);
		printk("The device is open already\n");
		return -1;
	}
	

	return 0;
}
 
int myclose (struct inode *pnode, struct file *pfile)
{
	struct openonce_dev *pgmydev = (struct openonce_dev *)pfile->private_data;
	spin_lock(&pgmydev->lock);
	pgmydev->open_flag = 1;
	spin_unlock(&pgmydev->lock);
	return 0;
}
 
struct file_operations myops = {
	.owner = THIS_MODULE,
	.open = myopen,
	.release = myclose,
};


int __init spinlock_init(void)
{
	int ret = 0;

	dev_t devno = MKDEV(major,minor);
	ret = register_chrdev_region(devno,num,"openonce");
 
	if(ret)
	{
		ret = alloc_chrdev_region(&devno,minor,num,"openonce");
		if(ret)
		{
			printk("devno failed.\n");
			return -1;
		}
		major = MAJOR(devno);
		minor = MINOR(devno);
	}
 
	/*给mydev指定操作函数集*/
	cdev_init(&gmydev.mydev,&myops);

	/*将mydev添加到内核对应的数据结构里*/
	gmydev.mydev.owner = THIS_MODULE;
	cdev_add(&gmydev.mydev,devno,num);

	gmydev.open_flag = 1;
	spin_lock_init(&gmydev.lock);

	return 0;
}

void __exit spinlock_exit(void)
{
	dev_t devno = MKDEV(major,minor);
 
	cdev_del(&gmydev.mydev);
 
	unregister_chrdev_region(devno, num);
}


MODULE_LICENSE("GPL");
MODULE_AUTHOR("imysy_22");
MODULE_DESCRIPTION("This is a script explained by the author who's name is imysy_22.");
MODULE_ALIAS("HI");

module_init(spinlock_init);
module_exit(spinlock_exit);

Others are exactly the same as atomic variables

五、信号量：基于阻塞的并发控制机制

a.定义信号量

struct semaphore sem;

b.初始化信号量

void sema_init(struct semaphore *sem, int val);

c.获得信号量P

int down(struct semaphore *sem);//深度睡眠

int down_interruptible(struct semaphore *sem);//浅度睡眠

d.释放信号量V

void up(struct semaphore *sem);

 

```

#include <linux/semaphore.h>

```

适用场合：任务上下文之间且临界区执行时间较长时的互斥或同步问题

六、互斥锁：基于阻塞的互斥机制

a.初始化

struct mutex  my_mutex;

mutex_init(&my_mutex);

b.获取互斥体

void  mutex_lock(struct mutex *lock);

c.释放互斥体

void mutex_unlock(struct mutex *lock);

1. 定义对应类型的变量

2. 初始化对应变量

P/加锁

临界区

V/解锁

```

#include <linux/mutex.h>

```

适用场合：任务上下文之间且临界区执行时间较长时的互斥问题

七、选择并发控制机制的原则

1. Contexts that do not allow sleep require a busy-wait class,A sleepable context can take a blocking class.Competing resources accessed in an exception context must use the busy-wait class.

2. The blocking class is recommended for applications with long critical section operations,It is recommended to use the busy-wait class for operations with very short critical sections.

3. 中断屏蔽仅在有与中断上下文共享资源时使用.

4. Use atomic variables when the shared resource is just a simple integer