arm裸机下读写寄存器很容易,各个寄存器和内存的地址是单一地址空间,他们是用相同的指令进行读写操作的.而在linux下就要复杂很多,因为linux支持多个体系架构的CPU。比如arm和x86就不一样，具体的差别我暂时也说不上来，这个涉及到CPU体系的设计。目前我只关心：linux为了支持多个硬件体系，在IO访问上做了自己的接口。可以通过IO内存和IO端口这两种方式进行IO访问。在LED的例子上给出这两种方式的具体实现：
1.利用IO Port的方式：
#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/init.h>
#include <linux/kernel.h> /* printk() */
#include <linux/slab.h> /* kmalloc() */
#include <linux/fs.h> /* everything... */
#include <linux/errno.h> /* error codes */
#include <linux/types.h> /* size_t */
#include <linux/proc_fs.h>
#include <linux/fcntl.h> /* O_ACCMODE */
#include <linux/seq_file.h>
#include <linux/cdev.h>
#include <linux/ioport.h>
#include <mach/regs-gpio.h>
#include <asm/system.h> /* cli(), *_flags */
#include <asm/uaccess.h> /* copy_*_user */
#include <asm/io.h>
#define LED_NUM 4
struct led_dev
{
struct cdev dev;
unsigned port;
unsigned long offset;
};
struct led_dev led[4];
dev_t dev = 0;
static struct resource *led_resource;
int led_open(struct inode *inode, struct file *filp)
{
struct led_dev *led; /* device information */
led = container_of(inode->i_cdev, struct led_dev, dev);
filp->private_data = led; /* for other methods */
return 0; /* success */
}
int led_release(struct inode *inode, struct file *filp)
{
return 0;
}
ssize_t led_read(struct file *filp, char __user *buf, size_t count, loff_t *f_pos)
{
return 0;
}
ssize_t led_write(struct file *filp, const char __user *buf, size_t count, loff_t *f_pos)
{
char data;
struct led_dev *led;
u32 value;
printk(KERN_INFO "debug by baikal: led dev write\n");
led = (struct led_dev *)filp->private_data;
copy_from_user(&data,buf,count);
if(data == '0')
{
printk(KERN_INFO "debug by baikal: led off\n");
value = inl((unsigned)(S3C2410_GPBDAT));
outl(value | 1<<led->offset,(unsigned)(S3C2410_GPBDAT));
//value = ioread32(led->base);
//iowrite32( value | 1<<led->offset, led->base);
}
else
{
printk(KERN_INFO "debug by baikal: led on\n");
value = inl((unsigned)(S3C2410_GPBDAT));
outl(value & ~(1<<led->offset),(unsigned)(S3C2410_GPBDAT));
//value = ioread32(led->base);
//iowrite32( value & ~(1<<led->offset), led->base);
}
}
struct file_operations led_fops = {
.owner = THIS_MODULE,
.read = led_read,
.write = led_write,
//.ioctl = led_ioctl,
.open = led_open,
.release = led_release,
};
static int led_init(void)
{
int result, i;
result = alloc_chrdev_region(&dev, 0, LED_NUM,"LED");
if (result < 0) {
printk(KERN_WARNING "LED: can't get major %d\n", MAJOR(dev));
return result;
}
led_resource = request_region(0x56000014,0x4,"led");
if(led_resource == NULL)
{
printk(KERN_ERR " Unable to register LED I/O addresses\n");
return -1;
}
for(i = 0; i < LED_NUM; i++)
{
cdev_init( &led[i].dev, &led_fops);
//led[i].port = ioport_map(0x56000014,0x4);
//led[i].base = ioremap(0x56000014,0x4);
led[i].offset = i + 5; //leds GPB5\6\7\8
led[i].dev.owner = THIS_MODULE;
led[i].dev.ops = &led_fops;
result = cdev_add(&led[i].dev,MKDEV(MAJOR(dev),i),1);
if(result < 0)
{
printk(KERN_ERR "LED: can't add led%d\n",i);
return result;
}
}
return 0;
}
static void led_exit(void)
{
int i;
release_region(0x56000014,0x4);
for( i = 0; i < LED_NUM; i++)
{
//iounmap(led[i].base);
cdev_del(&led[i].dev);
}
unregister_chrdev_region(dev, LED_NUM);
}
module_init(led_init);
module_exit(led_exit);
MODULE_AUTHOR("Baikal");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("Simple LED Driver");
2.利用IO Mem的方式：
#include <linux/module.h>
#include <linux/moduleparam.h>
#include <linux/init.h>
#include <linux/kernel.h> /* printk() */
#include <linux/slab.h> /* kmalloc() */
#include <linux/fs.h> /* everything... */
#include <linux/errno.h> /* error codes */
#include <linux/types.h> /* size_t */
#include <linux/proc_fs.h>
#include <linux/fcntl.h> /* O_ACCMODE */
#include <linux/seq_file.h>
#include <linux/cdev.h>
#include <linux/ioport.h>
#include <asm/system.h> /* cli(), *_flags */
#include <asm/uaccess.h> /* copy_*_user */
#include <asm/io.h>
#define LED_NUM 4
struct led_dev
{
struct cdev dev;
void __iomem *base;
unsigned long offset;
};
struct led_dev led[4];
dev_t dev = 0;
int led_open(struct inode *inode, struct file *filp)
{
struct led_dev *led; /* device information */
led = container_of(inode->i_cdev, struct led_dev, dev);
filp->private_data = led; /* for other methods */
return 0; /* success */
}
int led_release(struct inode *inode, struct file *filp)
{
return 0;
}
ssize_t led_read(struct file *filp, char __user *buf, size_t count, loff_t *f_pos)
{
return 0;
}
ssize_t led_write(struct file *filp, const char __user *buf, size_t count, loff_t *f_pos)
{
char data;
struct led_dev *led;
u32 value;
printk(KERN_INFO "debug by baikal: led dev write\n");
led = (struct led_dev *)filp->private_data;
copy_from_user(&data,buf,count);
if(data == '0')
{
printk(KERN_INFO "debug by baikal: led off\n");
value = ioread32(led->base);
iowrite32( value | 1<<led->offset, led->base);
}
else
{
printk(KERN_INFO "debug by baikal: led on\n");
value = ioread32(led->base);
iowrite32( value & ~(1<<led->offset), led->base);
}
}
struct file_operations led_fops = {
.owner = THIS_MODULE,
.read = led_read,
.write = led_write,
//.ioctl = led_ioctl,
.open = led_open,
.release = led_release,
};
static int led_init(void)
{
int result, i;
result = alloc_chrdev_region(&dev, 0, LED_NUM,"LED");
if (result < 0) {
printk(KERN_WARNING "LED: can't get major %d\n", MAJOR(dev));
return result;
}
for(i = 0; i < LED_NUM; i++)
{
cdev_init( &led[i].dev, &led_fops);
request_mem_region(0x56000014,0x4,"led");
led[i].base = ioremap(0x56000014,0x4);
led[i].offset = i + 5; //leds GPB5\6\7\8
led[i].dev.owner = THIS_MODULE;
led[i].dev.ops = &led_fops;
result = cdev_add(&led[i].dev,MKDEV(MAJOR(dev),i),1);
if(result < 0)
{
printk(KERN_ERR "LED: can't add led%d\n",i);
return result;
}
}
return 0;
}
static void led_exit(void)
{
int i;
release_mem_region(0x56000014,0x4);
for( i = 0; i < LED_NUM; i++)
{
iounmap(led[i].base);
cdev_del(&led[i].dev);
}
unregister_chrdev_region(dev, LED_NUM);
}
module_init(led_init);
module_exit(led_exit);
MODULE_AUTHOR("Baikal");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("Simple LED Driver");
目前，对于具体体系上的linux在移植过程中如何实现这两种方式的方法还不清楚，现在只是会用。等以后有机会了再慢慢理清楚。
