Re: [PATCH 2/2] mv643xx_eth: fix race in trasmit path.

[Maxime Bizon <mbizon@freebox.fr>]

On Mon, 2011-06-20 at 09:48 +0200, Richard Cochran wrote:

Hi Richard,

> Because the socket buffer is freed in the completion interrupt, it is not
> safe to access it after submitting it to the hardware.

I don't see why.

skb is freed from txq_reclaim() which grabs the tx queue lock before,
(hence the lockless __skb_queue_xxx() in both functions)

What am I missing ?

> Cc: stable@kernel.org
> Cc: Lennert Buytenhek <buytenh@wantstofly.org>
> Signed-off-by: Richard Cochran <richard.cochran@omicron.at>
> ---
>  drivers/net/mv643xx_eth.c |    6 ++++--
>  1 files changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/net/mv643xx_eth.c b/drivers/net/mv643xx_eth.c
> index a5d9b1c..1b7d2c1 100644
> --- a/drivers/net/mv643xx_eth.c
> +++ b/drivers/net/mv643xx_eth.c
> @@ -859,7 +859,7 @@ no_csum:
>  static netdev_tx_t mv643xx_eth_xmit(struct sk_buff *skb, struct net_device *dev)
>  {
>  	struct mv643xx_eth_private *mp = netdev_priv(dev);
> -	int queue;
> +	int length, queue;
>  	struct tx_queue *txq;
>  	struct netdev_queue *nq;
>  
> @@ -881,10 +881,12 @@ static netdev_tx_t mv643xx_eth_xmit(struct sk_buff *skb, struct net_device *dev)
>  		return NETDEV_TX_OK;
>  	}
>  
> +	length = skb->len;
> +
>  	if (!txq_submit_skb(txq, skb)) {
>  		int entries_left;
>  
> -		txq->tx_bytes += skb->len;
> +		txq->tx_bytes += length;
>  		txq->tx_packets++;
>  
>  		entries_left = txq->tx_ring_size - txq->tx_desc_count;

-- 
Maxime