On Sat, 2016-12-03 at 23:07 -0800, Eric Dumazet wrote:
> From: Eric Dumazet <[email protected]>
>
> 1) Old code was hard to maintain, due to complex lock chains.
> (We probably will be able to remove some kfree_rcu() in callers)
>
> 2) Using a single timer to update all estimators does not scale.
>
> 3) Code was buggy on 32bit kernel (WRITE_ONCE() on 64bit quantity
> is not supposed to work well)
> diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
> index
> c7adcb57654ea57d1ba6702c91743cb7d2c74d28..859b60bfa86712031186fffc09c65bc43aa065dd
> 100644
> --- a/net/ipv4/tcp_output.c
> +++ b/net/ipv4/tcp_output.c
> @@ -2081,6 +2081,14 @@ static bool tcp_small_queue_check(struct sock *sk,
> const struct sk_buff *skb,
> limit <<= factor;
>
> if (atomic_read(&sk->sk_wmem_alloc) > limit) {
> + /* Special case where TX completion is delayed too much :
> + * If the skb we try to send is the first skb in write queue,
> + * then send it !
> + * No need to wait for TX completion to call us back.
> + */
> + if (skb == sk->sk_write_queue.next)
> + return false;
> +
Oups, this has nothing to do here. I will send a v2, sorry.
