| 1 |
1275 |
phoenix |
Documents about softnet driver issues in general can be found
|
| 2 |
|
|
at:
|
| 3 |
|
|
|
| 4 |
|
|
http://www.firstfloor.org/~andi/softnet/
|
| 5 |
|
|
|
| 6 |
|
|
Transmit path guidelines:
|
| 7 |
|
|
|
| 8 |
|
|
1) The hard_start_xmit method must never return '1' under any
|
| 9 |
|
|
normal circumstances. It is considered a hard error unless
|
| 10 |
|
|
there is no way your device can tell ahead of time when it's
|
| 11 |
|
|
transmit function will become busy.
|
| 12 |
|
|
|
| 13 |
|
|
Instead it must maintain the queue properly. For example,
|
| 14 |
|
|
for a driver implementing scatter-gather this means:
|
| 15 |
|
|
|
| 16 |
|
|
static int drv_hard_start_xmit(struct sk_buff *skb,
|
| 17 |
|
|
struct net_device *dev)
|
| 18 |
|
|
{
|
| 19 |
|
|
struct drv *dp = dev->priv;
|
| 20 |
|
|
|
| 21 |
|
|
lock_tx(dp);
|
| 22 |
|
|
...
|
| 23 |
|
|
/* This is a hard error log it. */
|
| 24 |
|
|
if (TX_BUFFS_AVAIL(dp) <= (skb_shinfo(skb)->nr_frags + 1)) {
|
| 25 |
|
|
netif_stop_queue(dev);
|
| 26 |
|
|
unlock_tx(dp);
|
| 27 |
|
|
printk(KERN_ERR PFX "%s: BUG! Tx Ring full when queue awake!\n",
|
| 28 |
|
|
dev->name);
|
| 29 |
|
|
return 1;
|
| 30 |
|
|
}
|
| 31 |
|
|
|
| 32 |
|
|
... queue packet to card ...
|
| 33 |
|
|
... update tx consumer index ...
|
| 34 |
|
|
|
| 35 |
|
|
if (TX_BUFFS_AVAIL(dp) <= (MAX_SKB_FRAGS + 1))
|
| 36 |
|
|
netif_stop_queue(dev);
|
| 37 |
|
|
|
| 38 |
|
|
...
|
| 39 |
|
|
unlock_tx(dp);
|
| 40 |
|
|
...
|
| 41 |
|
|
}
|
| 42 |
|
|
|
| 43 |
|
|
And then at the end of your TX reclaimation event handling:
|
| 44 |
|
|
|
| 45 |
|
|
if (netif_queue_stopped(dp->dev) &&
|
| 46 |
|
|
TX_BUFFS_AVAIL(dp) > (MAX_SKB_FRAGS + 1))
|
| 47 |
|
|
netif_wake_queue(dp->dev);
|
| 48 |
|
|
|
| 49 |
|
|
For a non-scatter-gather supporting card, the three tests simply become:
|
| 50 |
|
|
|
| 51 |
|
|
/* This is a hard error log it. */
|
| 52 |
|
|
if (TX_BUFFS_AVAIL(dp) <= 0)
|
| 53 |
|
|
|
| 54 |
|
|
and:
|
| 55 |
|
|
|
| 56 |
|
|
if (TX_BUFFS_AVAIL(dp) == 0)
|
| 57 |
|
|
|
| 58 |
|
|
and:
|
| 59 |
|
|
|
| 60 |
|
|
if (netif_queue_stopped(dp->dev) &&
|
| 61 |
|
|
TX_BUFFS_AVAIL(dp) > 0)
|
| 62 |
|
|
netif_wake_queue(dp->dev);
|
| 63 |
|
|
|
| 64 |
|
|
2) Do not forget to update netdev->trans_start to jiffies after
|
| 65 |
|
|
each new tx packet is given to the hardware.
|
| 66 |
|
|
|
| 67 |
|
|
3) Do not forget that once you return 0 from your hard_start_xmit
|
| 68 |
|
|
method, it is your driver's responsibility to free up the SKB
|
| 69 |
|
|
and in some finite amount of time.
|
| 70 |
|
|
|
| 71 |
|
|
For example, this means that it is not allowed for your TX
|
| 72 |
|
|
mitigation scheme to let TX packets "hang out" in the TX
|
| 73 |
|
|
ring unreclaimed forever if no new TX packets are sent.
|
| 74 |
|
|
This error can deadlock sockets waiting for send buffer room
|
| 75 |
|
|
to be freed up.
|
| 76 |
|
|
|
| 77 |
|
|
If you return 1 from the hard_start_xmit method, you must not keep
|
| 78 |
|
|
any reference to that SKB and you must not attempt to free it up.
|
| 79 |
|
|
|
| 80 |
|
|
Probing guidelines:
|
| 81 |
|
|
|
| 82 |
|
|
1) Any hardware layer address you obtain for your device should
|
| 83 |
|
|
be verified. For example, for ethernet check it with
|
| 84 |
|
|
linux/etherdevice.h:is_valid_ether_addr()
|
