全部博文（146）
分类： LINUX
2008-11-11 16:36:57
一数据接收
为了了解网卡数据接收的过程。有必要先讨论DMA的具体过程。
DMA传输数据可以分为以下几个步骤：
首先：CPU向DMA送命令,如DMA方式，主存地址，传送的字数等，之后CPU执行原来的程序.
然后DMA 控制在 I/O 设备与主存间交换数据。接收数据完后, 向CPU发DMA请求,取得总线控制权，进行数据传送，修改卡上主存地址，修改字数计数器内且检查其值是否为零,不为零则继续传送，若已为零，则向 CPU发中断请求.。
也就是说，网卡收到包时，将它放入当前skb->data中。再来一个包时。DMA会修改卡上主存地址，转到skb->next，将数据放入其中。这也就是，一个skb->data存储一个数据包的原因。
好了，现在就可以来看具体的代码实现了。
当网络数据到络，网卡将其放到DMA内存，然后DMA向CPU报告中断，CPU根据中断向量，找到中断处理例程，也就是我们前面注册的e100_intr()进行处理。
static irqreturn_t e100_intr(int irq, void *dev_id, struct pt_regs *regs)
{
struct net_device *netdev = dev_id;
struct nic *nic = netdev_priv(netdev);
u8 stat_ack = readb(&nic->csr->scb.stat_ack);
DPRINTK(INTR, DEBUG, "stat_ack = 0x%02X\n", stat_ack);
if(stat_ack == stat_ack_not_ours || /* Not our interrupt */
stat_ack == stat_ack_not_present) /* Hardware is ejected */
return IRQ_NONE;
/* Ack interrupt(s) */
//发送中断ACK。Cpu向设备发送ACK。表示此中断已经处理
writeb(stat_ack, &nic->csr->scb.stat_ack);
/* We hit Receive No Resource (RNR); restart RU after cleaning */
if(stat_ack & stat_ack_rnr)
nic->ru_running = 0;
//禁用中断
e100_disable_irq(nic);
//CPU开始调度此设备。转而会运行netdev->poll
netif_rx_schedule(netdev);
return IRQ_HANDLED;
}
netif_rx_schedule(netdev)后，cpu开始调度此设备，轮询设备是否有数据要处理。转后调用netdev->poll函数，即：e100_poll（）
static int e100_poll(struct net_device *netdev, int *budget)
{
struct nic *nic = netdev_priv(netdev);
unsigned int work_to_do = min(netdev->quota, *budget);
unsigned int work_done = 0;
int tx_cleaned;
//开始对nic中，DMA数据的处理
e100_rx_clean(nic, &work_done, work_to_do);
tx_cleaned = e100_tx_clean(nic);
/* If no Rx and Tx cleanup work was done, exit polling mode. */
if((!tx_cleaned && (work_done == 0)) || !netif_running(netdev)) {
netif_rx_complete(netdev);
e100_enable_irq(nic);
return 0;
}
*budget -= work_done;
netdev->quota -= work_done;
return 1;
}
跟踪进e100_rx_clean():
static inline void e100_rx_clean(struct nic *nic, unsigned int *work_done,
unsigned int work_to_do)
{
struct rx *rx;
/* Indicate newly arrived packets */
//遍历环形DMA中的数据，调用e100_rx_indicate()进行处理
for(rx = nic->rx_to_clean; rx->skb; rx = nic->rx_to_clean = rx->next) {
if(e100_rx_indicate(nic, rx, work_done, work_to_do))
break; /* No more to clean */
}
/* Alloc new skbs to refill list */
for(rx = nic->rx_to_use; !rx->skb; rx = nic->rx_to_use = rx->next) {
if(unlikely(e100_rx_alloc_skb(nic, rx)))
break; /* Better luck next time (see watchdog) */
}
e100_start_receiver(nic);
}
在这里，它会遍历环形DMA中的数据，即从nic->rx_to_clean开始的数据，直至数据全部处理完
进入处理函数：e100_rx_indicate（）
static inline int e100_rx_indicate(struct nic *nic, struct rx *rx,
unsigned int *work_done, unsigned int work_to_do)
{
struct sk_buff *skb = rx->skb;
//从这里取得rfd.其中包括了一些接收信息，但不是链路传过来的有效数据
struct rfd *rfd = (struct rfd *)skb->data;
u16 rfd_status, actual_size;
if(unlikely(work_done && *work_done >= work_to_do))
return -EAGAIN;
//同步DMA缓存
pci_dma_sync_single_for_cpu(nic->pdev, rx->dma_addr,
sizeof(struct rfd), PCI_DMA_FROMDEVICE);
//取得接收状态
rfd_status = le16_to_cpu(rfd->status);
DPRINTK(RX_STATUS, DEBUG, "status=0x%04X\n", rfd_status);
/* If data isn't ready, nothing to indicate */
//没有接收完全，返回
if(unlikely(!(rfd_status & cb_complete)))
return -EAGAIN;
//取得接收数据的长度
actual_size = le16_to_cpu(rfd->actual_size) & 0x3FFF;
if(unlikely(actual_size > RFD_BUF_LEN - sizeof(struct rfd)))
actual_size = RFD_BUF_LEN - sizeof(struct rfd);
//取消DMA缓存映射
pci_unmap_single(nic->pdev, rx->dma_addr,
RFD_BUF_LEN, PCI_DMA_FROMDEVICE);
//由于RFD不是链路传入的数据，清除
skb_reserve(skb, sizeof(struct rfd));
//调整skb中的tail指针，与len更新
skb_put(skb, actual_size);
//取得链路层协议
skb->protocol = eth_type_trans(skb, nic->netdev);
//接收失败
if(unlikely(!(rfd_status & cb_ok))) {
/* Don't indicate if hardware indicates errors */
nic->net_stats.rx_dropped++;
dev_kfree_skb_any(skb);
}
//数据超长。Drop it
else if(actual_size > nic->netdev->mtu + VLAN_ETH_HLEN) {
/* Don't indicate oversized frames */
nic->rx_over_length_errors++;
nic->net_stats.rx_dropped++;
dev_kfree_skb_any(skb);
} else {
//成功的接收了，更新统计计数
nic->net_stats.rx_packets++;
nic->net_stats.rx_bytes += actual_size;
nic->netdev->last_rx = jiffies;
//送至上次协议处理
netif_receive_skb(skb);
if(work_done)
(*work_done)++;
}
rx->skb = NULL;
return 0;
}
上面代码中要去判断接收是否完全，为什么要去判断呢？根据DMA机制，是网卡把数据放入DMA之后。DMA再向CPU发中断的嘛？呵呵。在这里进行接收完全判断是因为：
1：由其它原因造成的中断
2：在处理中断时候。数据又到达了。网卡依然会把它放至下一个skb。而在代码处理中是遍历处理的，也就是说处理下一个skb的时候，可能网卡正在传数据。
好了，运行到netif_receive_skb()之后，数据包被送到上层。
二数据的发送
在进入到发送函数之前，我们先来看e100_up()->e100_alloc_cbs函数：
static int e100_alloc_cbs(struct nic *nic)
{
struct cb *cb;
unsigned int i, count = nic->params.cbs.count;
nic->cuc_cmd = cuc_start;
nic->cb_to_use = nic->cb_to_send = nic->cb_to_clean = NULL;
nic->cbs_avail = 0;
//线性DMA映射，这里返回的是虚拟地址，供CPU使用的
nic->cbs = pci_alloc_consistent(nic->pdev,
sizeof(struct cb) * count, &nic->cbs_dma_addr);
if(!nic->cbs)
return -ENOMEM;
//建立环形的发送缓冲区
for(cb = nic->cbs, i = 0; i < count; cb++, i++) {
cb->next = (i + 1 < count) ? cb + 1 : nic->cbs;
cb->prev = (i == 0) ? nic->cbs + count - 1 : cb - 1;
cb->dma_addr = nic->cbs_dma_addr + i * sizeof(struct cb);
cb->link = cpu_to_le32(nic->cbs_dma_addr +
((i+1) % count) * sizeof(struct cb));
cb->skb = NULL;
}
//初始化各指针，使其指向缓冲初始位置
nic->cb_to_use = nic->cb_to_send = nic->cb_to_clean = nic->cbs;
nic->cbs_avail = count;
return 0;
}
在这一段代码里，完成了发送的准备工作，建立了发送环形缓存。在发送数剧时，只要把数据送入缓存即可
数据最终会调用dev-> hard_start_xmit函数。在e100代码里，也就是e100_xmit_frame(). 进入里面看下：
static int e100_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
{
struct nic *nic = netdev_priv(netdev);
int err;
if(nic->flags & ich_10h_workaround) {
e100_exec_cmd(nic, cuc_nop, 0);
udelay(1);
}
err = e100_exec_cb(nic, skb, e100_xmit_prepare);
switch(err) {
case -ENOSPC:
/* We queued the skb, but now we're out of space. */
netif_stop_queue(netdev);
break;
case -ENOMEM:
/* This is a hard error - log it. */
DPRINTK(TX_ERR, DEBUG, "Out of Tx resources, returning skb\n");
netif_stop_queue(netdev);
return 1;
}
netdev->trans_start = jiffies;
return 0;
}
继续跟踪进 e100_exec_cb(nic, skb, e100_xmit_prepare);
static inline int e100_exec_cb(struct nic *nic, struct sk_buff *skb,
void (*cb_prepare)(struct nic *, struct cb *, struct sk_buff *))
{
struct cb *cb;
unsigned long flags;
int err = 0;
spin_lock_irqsave(&nic->cb_lock, flags);
if(unlikely(!nic->cbs_avail)) {
err = -ENOMEM;
goto err_unlock;
}
//将skb 推入环形发送缓冲
//cb_to_use：发送缓冲当前的使用位置
cb = nic->cb_to_use;
nic->cb_to_use = cb->next;
nic->cbs_avail--;
cb->skb = skb;
if(unlikely(!nic->cbs_avail))
err = -ENOSPC;
cb_prepare(nic, cb, skb);
/* Order is important otherwise we'll be in a race with h/w:
* set S-bit in current first, then clear S-bit in previous. */
cb->command |= cpu_to_le16(cb_s);
wmb();
cb->prev->command &= cpu_to_le16(~cb_s);
//当发送数据不为空。将余下数剧全部发送
while(nic->cb_to_send != nic->cb_to_use) {
if(unlikely(e100_exec_cmd(nic, nic->cuc_cmd,
nic->cb_to_send->dma_addr))) {
/* Ok, here's where things get sticky. It's
* possible that we can't schedule the command
* because the controller is too busy, so
* let's just queue the command and try again
* when another command is scheduled. */
break;
} else {
nic->cuc_cmd = cuc_resume;
nic->cb_to_send = nic->cb_to_send->next;
}
}
err_unlock:
spin_unlock_irqrestore(&nic->cb_lock, flags);
return err;
}
在这里我们看到，发送数据过程主要由e100_exec_cmd完成。跟踪进去
static inline int e100_exec_cmd(struct nic *nic, u8 cmd, dma_addr_t dma_addr)
{
unsigned long flags;
unsigned int i;
int err = 0;
spin_lock_irqsave(&nic->cmd_lock, flags);
/* Previous command is accepted when SCB clears */
for(i = 0; i < E100_WAIT_SCB_TIMEOUT; i++) {
if(likely(!readb(&nic->csr->scb.cmd_lo)))
break;
cpu_relax();
if(unlikely(i > (E100_WAIT_SCB_TIMEOUT >> 1)))
udelay(5);
}
if(unlikely(i == E100_WAIT_SCB_TIMEOUT)) {
err = -EAGAIN;
goto err_unlock;
}
if(unlikely(cmd != cuc_resume))
//将数据的存放地址放入对应寄存器
writel(dma_addr, &nic->csr->scb.gen_ptr);
//将发送操作写入控制寄存器
writeb(cmd, &nic->csr->scb.cmd_lo);
err_unlock:
spin_unlock_irqrestore(&nic->cmd_lock, flags);
return err;
}
从此可以看到。Intel
