linux-2.6.22.6内核网卡驱动框架分析

网络协议分为很多层，而驱动这层对应于实际的物理网卡部分，这也是最底层的部分，以cs89x0.c这个驱动程序为例来分析下网卡驱动程序框架。

正常开发一个驱动程序时，一般都遵循以下几个步骤：
1.分配某个结构体
2.设置该结构体
3.注册
4.硬件相关操作

首先分析cs89x0.c的入口函数

int __init init_module(void)
{struct net_device *dev = alloc_etherdev(sizeof(struct net_local));struct net_local *lp;int ret = 0;#if defined(CONFIG_ARCH_S3C2410)unsigned int oldval_bwscon;     /* 用来保存BWSCON寄存器的值 */unsigned int oldval_bankcon3;   /* 用来保存S3C2410_BANKCON3寄存器的值 */
#endif#if DEBUGGINGnet_debug = debug;
#elsedebug = 0;
#endifif (!dev)return -ENOMEM;#if defined(CONFIG_ARCH_S3C2410)// 将CS8900A的物理地址转换为虚拟地址，0x300是CS8900A内部的IO空间的偏移地址dev->base_addr = io = (unsigned int)ioremap(S3C24XX_PA_CS8900, SZ_1M) + 0x300;dev->irq = irq = cs8900_irq_map[0]; /* 中断号 *//* 设置默认MAC地址，* MAC地址可以由CS8900A外接的EEPROM设定(有些单板没接EEPROM)，* 或者启动系统后使用ifconfig修改*/dev->dev_addr[0] = 0x08;dev->dev_addr[1] = 0x89;dev->dev_addr[2] = 0x89;dev->dev_addr[3] = 0x89;dev->dev_addr[4] = 0x89;dev->dev_addr[5] = 0x89;/* 设置Bank3: 总线宽度为16, 使能nWAIT, 使能UB/LB。by www.100ask.net */oldval_bwscon = *((volatile unsigned int *)S3C2410_BWSCON);*((volatile unsigned int *)S3C2410_BWSCON) = (oldval_bwscon & ~(3<<12)) \| S3C2410_BWSCON_DW3_16 | S3C2410_BWSCON_WS3 | S3C2410_BWSCON_ST3;/* 设置BANK3的时间参数, by www.100ask.net */oldval_bankcon3 = *((volatile unsigned int *)S3C2410_BANKCON3);*((volatile unsigned int *)S3C2410_BANKCON3) = 0x1f7c;    
#elsedev->irq = irq;dev->base_addr = io;
#endiflp = netdev_priv(dev);#if ALLOW_DMAif (use_dma) {lp->use_dma = use_dma;lp->dma = dma;lp->dmasize = dmasize;}
#endifspin_lock_init(&lp->lock);/* boy, they'd better get these right */if (!strcmp(media, "rj45"))lp->adapter_cnf = A_CNF_MEDIA_10B_T | A_CNF_10B_T;else if (!strcmp(media, "aui"))lp->adapter_cnf = A_CNF_MEDIA_AUI   | A_CNF_AUI;else if (!strcmp(media, "bnc"))lp->adapter_cnf = A_CNF_MEDIA_10B_2 | A_CNF_10B_2;elselp->adapter_cnf = A_CNF_MEDIA_10B_T | A_CNF_10B_T;if (duplex==-1)lp->auto_neg_cnf = AUTO_NEG_ENABLE;if (io == 0) {printk(KERN_ERR "cs89x0.c: Module autoprobing not allowed.\n");printk(KERN_ERR "cs89x0.c: Append io=0xNNN\n");ret = -EPERM;goto out;} else if (io <= 0x1ff) {ret = -ENXIO;goto out;}#if ALLOW_DMAif (use_dma && dmasize != 16 && dmasize != 64) {printk(KERN_ERR "cs89x0.c: dma size must be either 16K or 64K, not %dK\n", dmasize);ret = -EPERM;goto out;}
#endifret = cs89x0_probe1(dev, io, 1);if (ret)goto out;dev_cs89x0 = dev;return 0;
out:
#if defined(CONFIG_ARCH_S3C2410)iounmap(dev->base_addr);/* 恢复寄存器原来的值 */*((volatile unsigned int *)S3C2410_BWSCON) = oldval_bwscon;*((volatile unsigned int *)S3C2410_BANKCON3) = oldval_bankcon3;
#endif    free_netdev(dev);return ret;
}

入口函数里，首先分配了net_device 结构体，然后对该结构体进行进行填充，最后调用cs89x0_probe1进行下一步处理。

cs89x0_probe1(struct net_device *dev, int ioaddr, int modular)
{
.............dev->open		= net_open;dev->stop		= net_close;dev->tx_timeout		= net_timeout;dev->watchdog_timeo	= HZ;**dev->hard_start_xmit 	= net_send_packet;**dev->get_stats		= net_get_stats;dev->set_multicast_list = set_multicast_list;dev->set_mac_address 	= set_mac_address;
.....
.....retval = register_netdev(dev);}

cs89x0_probe1里又进一步对net_device 进行了填充，其中hard_start_xmit 就是发送数据函数，然后通过register_netdev进行注册。

进一步查看net_send_packet

static int net_send_packet(struct sk_buff *skb, struct net_device *dev)
{struct net_local *lp = netdev_priv(dev);if (net_debug > 3) {printk("%s: sent %d byte packet of type %x\n",dev->name, skb->len,(skb->data[ETH_ALEN+ETH_ALEN] << 8) | skb->data[ETH_ALEN+ETH_ALEN+1]);}/* keep the upload from being interrupted, since weask the chip to start transmitting before thewhole packet has been completely uploaded. */spin_lock_irq(&lp->lock);netif_stop_queue(dev);/* initiate a transmit sequence */writeword(dev->base_addr, TX_CMD_PORT, lp->send_cmd);writeword(dev->base_addr, TX_LEN_PORT, skb->len);/* Test to see if the chip has allocated memory for the packet */if ((readreg(dev, PP_BusST) & READY_FOR_TX_NOW) == 0) {/** Gasp!  It hasn't.  But that shouldn't happen since* we're waiting for TxOk, so return 1 and requeue this packet.*/spin_unlock_irq(&lp->lock);if (net_debug) printk("cs89x0: Tx buffer not free!\n");return 1;}/* Write the contents of the packet */writewords(dev->base_addr, TX_FRAME_PORT,skb->data,(skb->len+1) >>1);spin_unlock_irq(&lp->lock);lp->stats.tx_bytes += skb->len;dev->trans_start = jiffies;dev_kfree_skb (skb);/** We DO NOT call netif_wake_queue() here.* We also DO NOT call netif_start_queue().** Either of these would cause another bottom half run through* net_send_packet() before this packet has fully gone out.  That causes* us to hit the "Gasp!" above and the send is rescheduled.  it runs like* a dog.  We just return and wait for the Tx completion interrupt handler* to restart the netdevice layer*/return 0;
}

net_send_packet里用到了sk_buff 这个结构体，sk_buff 就是数据的载体，net_send_packet里通过sk_buff 发送了数据，那数据又是如何接受的呢，其实是通过中断接受数据的，net_interrupt处理如下：

static irqreturn_t net_interrupt(int irq, void *dev_id)
{struct net_device *dev = dev_id;struct net_local *lp;int ioaddr, status;int handled = 0;ioaddr = dev->base_addr;lp = netdev_priv(dev);/* we MUST read all the events out of the ISQ, otherwise we'll neverget interrupted again.  As a consequence, we can't have any limiton the number of times we loop in the interrupt handler.  Thehardware guarantees that eventually we'll run out of events.  Ofcourse, if you're on a slow machine, and packets are arrivingfaster than you can read them off, you're screwed.  Hasta lavista, baby!  */while ((status = readword(dev->base_addr, ISQ_PORT))) {if (net_debug > 4)printk("%s: event=%04x\n", dev->name, status);handled = 1;switch(status & ISQ_EVENT_MASK) {case ISQ_RECEIVER_EVENT:/* Got a packet(s). */net_rx(dev);break;case ISQ_TRANSMITTER_EVENT:lp->stats.tx_packets++;netif_wake_queue(dev);	/* Inform upper layers. */if ((status & (	TX_OK |TX_LOST_CRS |TX_SQE_ERROR |TX_LATE_COL |TX_16_COL)) != TX_OK) {if ((status & TX_OK) == 0) lp->stats.tx_errors++;if (status & TX_LOST_CRS) lp->stats.tx_carrier_errors++;if (status & TX_SQE_ERROR) lp->stats.tx_heartbeat_errors++;if (status & TX_LATE_COL) lp->stats.tx_window_errors++;if (status & TX_16_COL) lp->stats.tx_aborted_errors++;}break;case ISQ_BUFFER_EVENT:if (status & READY_FOR_TX) {/* we tried to transmit a packet earlier,but inexplicably ran out of buffers.That shouldn't happen since we only everload one packet.  Shrug.  Do the rightthing anyway. */netif_wake_queue(dev);	/* Inform upper layers. */}if (status & TX_UNDERRUN) {if (net_debug > 0) printk("%s: transmit underrun\n", dev->name);lp->send_underrun++;if (lp->send_underrun == 3) lp->send_cmd = TX_AFTER_381;else if (lp->send_underrun == 6) lp->send_cmd = TX_AFTER_ALL;/* transmit cycle is done, althoughframe wasn't transmitted - thisavoids having to wait for the upperlayers to timeout on us, in theevent of a tx underrun */netif_wake_queue(dev);	/* Inform upper layers. */}
#if ALLOW_DMAif (lp->use_dma && (status & RX_DMA)) {int count = readreg(dev, PP_DmaFrameCnt);while(count) {if (net_debug > 5)printk("%s: receiving %d DMA frames\n", dev->name, count);if (net_debug > 2 && count >1)printk("%s: receiving %d DMA frames\n", dev->name, count);dma_rx(dev);if (--count == 0)count = readreg(dev, PP_DmaFrameCnt);if (net_debug > 2 && count > 0)printk("%s: continuing with %d DMA frames\n", dev->name, count);}}
#endifbreak;case ISQ_RX_MISS_EVENT:lp->stats.rx_missed_errors += (status >>6);break;case ISQ_TX_COL_EVENT:lp->stats.collisions += (status >>6);break;}}return IRQ_RETVAL(handled);
}

net_interrupt里又调用net_rx(dev);进行处理

net_rx(struct net_device *dev)
{struct net_local *lp = netdev_priv(dev);struct sk_buff *skb;int status, length;int ioaddr = dev->base_addr;status = readword(ioaddr, RX_FRAME_PORT);length = readword(ioaddr, RX_FRAME_PORT);if ((status & RX_OK) == 0) {count_rx_errors(status, lp);return;}/* Malloc up new buffer. */skb = dev_alloc_skb(length + 2);if (skb == NULL) {
#if 0		/* Again, this seems a cruel thing to do */printk(KERN_WARNING "%s: Memory squeeze, dropping packet.\n", dev->name);
#endiflp->stats.rx_dropped++;return;}skb_reserve(skb, 2);	/* longword align L3 header */readwords(ioaddr, RX_FRAME_PORT, skb_put(skb, length), length >> 1);if (length & 1)skb->data[length-1] = readword(ioaddr, RX_FRAME_PORT);if (net_debug > 3) {printk(	"%s: received %d byte packet of type %x\n",dev->name, length,(skb->data[ETH_ALEN+ETH_ALEN] << 8) | skb->data[ETH_ALEN+ETH_ALEN+1]);}skb->protocol=eth_type_trans(skb,dev);netif_rx(skb);dev->last_rx = jiffies;lp->stats.rx_packets++;lp->stats.rx_bytes += length;
}

net_rx里也会构造一个sk_buff 结构体，然后调用netif_rx(skb);进行发包。

总结：发送数据和接受数据是通过hard_start_xmit 和netif_rx完成的，而数据的载体都是sk_buff 结构体。