e100的RFD是什么意思?

chishanmingshen

e100_rx_alloc_skb()中的RFD到底是什么意思???根el有何关系?求助...

1.         /* Link the RFD to end of RFA by linking previous RFD to
   
2.          * this one.  We are safe to touch the previous RFD because
   
3.          * it is protected by the before last buffer's el bit being set */
   
4.         if (rx->prev->skb) {
   
5.                 struct rfd *prev_rfd = (struct rfd *)rx->prev->skb->data;
   
6.                 put_unaligned(cpu_to_le32(rx->dma_addr), &prev_rfd->link);<------为何把当前rx的dma_addr保存在上一个rx的rfd中???
   
7.         }
   

复制代码

frogsu

RFD应该是接收描述符的简称(Receive Frame Descriptor),至于为啥要把当前rx的dma_addr保存在上一个rx的rfd中,是因为网卡的接收描述符的DMA缓冲块是环形缓冲区,这样就建立了一个环形链。

chishanmingshen

那这个跟el有啥关系?请指点...回复 2# frogsu


   

frogsu

首先你贴出来的代码里是关于建立环形缓冲区的，提到了el bit位设置问题。"We are safe to touch the previous RFD because it is protected by the before last buffer's el bit being set." 从这句话字面意思是说在el bit 被设置之前，可以安全使用其RFD。猜测el bit应该是关于e100的接收包操作相关的操作，估计表达接收中、或者接收完成之类的。我之前接触e1000e驱动，感觉比较像。呵呵，具体的你还是参考参考e100的datasheet。

chishanmingshen

*
*        In order to keep updates to the RFD link field from colliding with
*        hardware writes to mark packets complete, we use the feature that
*        hardware will not write to a size 0 descriptor and mark the previous
*        packet as end-of-list (EL).   After updating the link, we remove EL
*        and only then restore the size such that hardware may use the
*        previous-to-end RFD.
*


e100.c中的注释, 我没看不明白...

frogsu

      查阅了网上的资料:http://blog.csdn.net/bekars/article/details/1066360
      知道了原来e100使用了共享内存方式来接收数据，定义了一个RFD。所以，在接收数据时，只需要把接收数据的skb->data（第一个RFD）的物理地址写入共享控制模块的指针里，网卡就能自动根据这个地址进行DMA，并且根据RFD中link域，找下个RFD。
     至于这句话的意思，我的理解如下：
     为了防止RFD的link域更新与硬件写包完成相冲突，可利用硬件不会写长度为0或EL置位的RFD的特征。所以，在更新RFD的link域后，需恢复RFD的长度，这样硬件就可以重新利用整块RFD空间了。(不知道是否准确？！)

chishanmingshen

有道理

1. 
   
2. 
   
3. static void e100_rx_clean(struct nic *nic, unsigned int *work_done,
   
4.         unsigned int work_to_do)
   
5. {
   
6.         struct rx *rx;
   
7.         int restart_required = 0, err = 0;
   
8.         struct rx *old_before_last_rx, *new_before_last_rx;
   
9.         struct rfd *old_before_last_rfd, *new_before_last_rfd;
   
10. 
    
11.         /* Indicate newly arrived packets */
    
12.         for(rx = nic->rx_to_clean; rx->skb; rx = nic->rx_to_clean = rx->next) {
    
13.                 err = e100_rx_indicate(nic, rx, work_done, work_to_do);
    
14.                 /* Hit quota or no more to clean */
    
15.                 if (-EAGAIN == err || -ENODATA == err)
    
16.                         break;
    
17.         }
    
18. 
    
19. 
    
20.         /* On EAGAIN, hit quota so have more work to do, restart once
    
21.          * cleanup is complete.
    
22.          * Else, are we already rnr? then pay attention!!! this ensures that
    
23.          * the state machine progression never allows a start with a
    
24.          * partially cleaned list, avoiding a race between hardware
    
25.          * and rx_to_clean when in NAPI mode */
    
26.         if (-EAGAIN != err && RU_SUSPENDED == nic->ru_running)
    
27.                 restart_required = 1;
    
28. 
    
29.         old_before_last_rx = nic->rx_to_use->prev->prev;
    
30.         old_before_last_rfd = (struct rfd *)old_before_last_rx->skb->data;
    
31. 
    
32.         /* Alloc new skbs to refill list */
    
33.         for(rx = nic->rx_to_use; !rx->skb; rx = nic->rx_to_use = rx->next) {
    
34.                 if(unlikely(e100_rx_alloc_skb(nic, rx)))
    
35.                         break; /* Better luck next time (see watchdog) */
    
36.         }
    
37. 
    
38.         new_before_last_rx = nic->rx_to_use->prev->prev;
    
39.         if (new_before_last_rx != old_before_last_rx) {
    
40.                 /* Set the el-bit on the buffer that is before the last buffer.
    
41.                  * This lets us update the next pointer on the last buffer
    
42.                  * without worrying about hardware touching it.
    
43.                  * We set the size to 0 to prevent hardware from touching this
    
44.                  * buffer.
    
45.                  * When the hardware hits the before last buffer with el-bit
    
46.                  * and size of 0, it will RNR interrupt, the RUS will go into
    
47.                  * the No Resources state.  It will not complete nor write to
    
48.                  * this buffer. */
    
49.                 new_before_last_rfd =
    
50.                         (struct rfd *)new_before_last_rx->skb->data;
    
51.                 new_before_last_rfd->size = 0;
    
52.                 new_before_last_rfd->command |= cpu_to_le16(cb_el);
    
53.                 pci_dma_sync_single_for_device(nic->pdev,
    
54.                         new_before_last_rx->dma_addr, sizeof(struct rfd),
    
55.                         PCI_DMA_TODEVICE);
    
56. 
    
57.                 /* Now that we have a new stopping point, we can clear the old
    
58.                  * stopping point.  We must sync twice to get the proper
    
59.                  * ordering on the hardware side of things. */
    
60.                 old_before_last_rfd->command &= ~cpu_to_le16(cb_el);
    
61.                 pci_dma_sync_single_for_device(nic->pdev,
    
62.                         old_before_last_rx->dma_addr, sizeof(struct rfd),
    
63.                         PCI_DMA_TODEVICE);
    
64.                 old_before_last_rfd->size = cpu_to_le16(VLAN_ETH_FRAME_LEN);
    
65.                 pci_dma_sync_single_for_device(nic->pdev,
    
66.                         old_before_last_rx->dma_addr, sizeof(struct rfd),
    
67.                         PCI_DMA_TODEVICE);
    
68.         }
    
69. 
    
70.         if(restart_required) {
    
71.                 // ack the rnr?
    
72.                 iowrite8(stat_ack_rnr, &nic->csr->scb.stat_ack);
    
73.                 e100_start_receiver(nic, nic->rx_to_clean);
    
74.                 if(work_done)
    
75.                         (*work_done)++;
    
76.         }
    
77. }
    
78. 
    

复制代码

old_before_last_rx和new_before_last_rx是什么关系?

chishanmingshen

自己搞明白了，呵呵。

就是为了防止硬件读错最后一个元素，将该链表的倒数第二个元素做特殊处理：设置el标志。

结贴！

btw: 建议后来人不要看e100或8139的了，太老了，缺点多。。。


回复 7# chishanmingshen


   

chishanmingshen

非常感谢！

发现这个帖子不是悬赏的，给不了你分数。。。。呵呵

回复 6# frogsu


   

灌水菜鸟

回复 8# chishanmingshen


不看老网卡看啥？上来就多对列等各种网络加速和负载减免的NB网卡驱动么？那一定更迷茫，呵呵。