Re: [net-next-2.6 PATCH] net: netif_set_real_num_rx_queues may cap num_rx_queues at init time

[Eric Dumazet <eric.dumazet@gmail.com>]

Le lundi 04 octobre 2010 à 15:00 -0700, John Fastabend a écrit :
> The logic for netif_set_real_num_rx_queues is the following,
> 
> netif_set_real_num_rx_queues(dev, rxq)
> {
> 	...
> 	if (dev->reg_state == NETREG_REGISTERED) {
> 		...
> 	} else {
> 		dev->num_rx_queues = rxq;
> 	}
> 
> 	dev->real_num_rx_queues = rxq;
> 	return 0;
> }
> 
> Some drivers init path looks like the following,
> 
> alloc_etherdev_mq(priv_sz, max_num_queues_ever);
> ...
> netif_set_real_num_rx_queues(dev, queues_to_use_now);
> ...
> register_netdev(dev);
> ...
> 
> Because netif_set_real_num_rx_queues sets num_rx_queues if the
> reg state is not NETREG_REGISTERED we end up with the incorrect
> max number of rx queues. This patch proposes to remove the else
> clause above so this does not occur.  Also just reading the
> function set_real_num it seems a bit unexpected that num_rx_queues
> gets set.
> 

You dont tell why its "incorrect".

Why should we keep num_rx_queues > real_num_rx_queues ?

It creates /sys files, and Qdisc stuff for nothing...



> CC: Ben Hutchings <bhutchings@solarflare.com>
> 
> Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
> ---
> 
>  net/core/dev.c |    2 --
>  1 files changed, 0 insertions(+), 2 deletions(-)
> 
> diff --git a/net/core/dev.c b/net/core/dev.c
> index a313bab..f78d996 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -1592,8 +1592,6 @@ int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
>  						  rxq);
>  		if (rc)
>  			return rc;
> -	} else {
> -		dev->num_rx_queues = rxq;
>  	}
>  
>  	dev->real_num_rx_queues = rxq;
> 
> --