LXR linux/Documentation/networking/driver.txt

   1Document about softnet driver issues
   2
   3Transmit path guidelines:
   4
   51) The ndo_start_xmit method must not return NETDEV_TX_BUSY under
   6   any normal circumstances.  It is considered a hard error unless
   7   there is no way your device can tell ahead of time when it's
   8   transmit function will become busy.
   9
  10   Instead it must maintain the queue properly.  For example,
  11   for a driver implementing scatter-gather this means:
  12
  13        static netdev_tx_t drv_hard_start_xmit(struct sk_buff *skb,
  14                                               struct net_device *dev)
  15        {
  16                struct drv *dp = netdev_priv(dev);
  17
  18                lock_tx(dp);
  19                ...
  20                /* This is a hard error log it. */
  21                if (TX_BUFFS_AVAIL(dp) <= (skb_shinfo(skb)->nr_frags + 1)) {
  22                        netif_stop_queue(dev);
  23                        unlock_tx(dp);
  24                        printk(KERN_ERR PFX "%s: BUG! Tx Ring full when queue awake!\n",
  25                               dev->name);
  26                        return NETDEV_TX_BUSY;
  27                }
  28
  29                ... queue packet to card ...
  30                ... update tx consumer index ...
  31
  32                if (TX_BUFFS_AVAIL(dp) <= (MAX_SKB_FRAGS + 1))
  33                        netif_stop_queue(dev);
  34
  35                ...
  36                unlock_tx(dp);
  37                ...
  38                return NETDEV_TX_OK;
  39        }
  40
  41   And then at the end of your TX reclamation event handling:
  42
  43        if (netif_queue_stopped(dp->dev) &&
  44            TX_BUFFS_AVAIL(dp) > (MAX_SKB_FRAGS + 1))
  45                netif_wake_queue(dp->dev);
  46
  47   For a non-scatter-gather supporting card, the three tests simply become:
  48
  49                /* This is a hard error log it. */
  50                if (TX_BUFFS_AVAIL(dp) <= 0)
  51
  52   and:
  53
  54                if (TX_BUFFS_AVAIL(dp) == 0)
  55
  56   and:
  57
  58        if (netif_queue_stopped(dp->dev) &&
  59            TX_BUFFS_AVAIL(dp) > 0)
  60                netif_wake_queue(dp->dev);
  61
  622) An ndo_start_xmit method must not modify the shared parts of a
  63   cloned SKB.
  64
  653) Do not forget that once you return NETDEV_TX_OK from your
  66   ndo_start_xmit method, it is your driver's responsibility to free
  67   up the SKB and in some finite amount of time.
  68
  69   For example, this means that it is not allowed for your TX
  70   mitigation scheme to let TX packets "hang out" in the TX
  71   ring unreclaimed forever if no new TX packets are sent.
  72   This error can deadlock sockets waiting for send buffer room
  73   to be freed up.
  74
  75   If you return NETDEV_TX_BUSY from the ndo_start_xmit method, you
  76   must not keep any reference to that SKB and you must not attempt
  77   to free it up.
  78
  79Probing guidelines:
  80
  811) Any hardware layer address you obtain for your device should
  82   be verified.  For example, for ethernet check it with
  83   linux/etherdevice.h:is_valid_ether_addr()
  84
  85Close/stop guidelines:
  86
  871) After the ndo_stop routine has been called, the hardware must
  88   not receive or transmit any data.  All in flight packets must
  89   be aborted. If necessary, poll or wait for completion of 
  90   any reset commands.
  91
  922) The ndo_stop routine will be called by unregister_netdevice
  93   if device is still UP.
  94