[PATCH] net: core: improve the tx_hash calculating

[PATCH] net: core: improve the tx_hash calculating

[Tonghao Zhang]

Use the % instead of while, and it may simple code and improve
the calculating. The real_num_tx_queues has been checked when
allocating and setting it.

Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
---
 net/core/dev.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 1844d9b..edc5b75 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2617,15 +2617,13 @@ void netif_device_attach(struct net_device *dev)
  */
 static u16 skb_tx_hash(const struct net_device *dev, struct sk_buff *skb)
 {
-	u32 hash;
 	u16 qoffset = 0;
 	u16 qcount = dev->real_num_tx_queues;
 
 	if (skb_rx_queue_recorded(skb)) {
-		hash = skb_get_rx_queue(skb);
-		while (unlikely(hash >= qcount))
-			hash -= qcount;
-		return hash;
+		/* When setting the real_num_tx_queues, we make sure
+		 * real_num_tx_queues != 0. */
+		return skb_get_rx_queue(skb) % qcount;
 	}
 
 	if (dev->num_tc) {
-- 
1.8.3.1

Re: [PATCH] net: core: improve the tx_hash calculating

[Sergei Shtylyov]

Hello!

On 5/31/2018 1:14 PM, Tonghao Zhang wrote:

> Use the % instead of while, and it may simple code and improve
> the calculating. The real_num_tx_queues has been checked when
> allocating and setting it.
> 
> Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
> ---
>   net/core/dev.c | 8 +++-----
>   1 file changed, 3 insertions(+), 5 deletions(-)
> 
> diff --git a/net/core/dev.c b/net/core/dev.c
> index 1844d9b..edc5b75 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -2617,15 +2617,13 @@ void netif_device_attach(struct net_device *dev)
>    */
>   static u16 skb_tx_hash(const struct net_device *dev, struct sk_buff *skb)
>   {
> -	u32 hash;
>   	u16 qoffset = 0;
>   	u16 qcount = dev->real_num_tx_queues;
>   
>   	if (skb_rx_queue_recorded(skb)) {
> -		hash = skb_get_rx_queue(skb);
> -		while (unlikely(hash >= qcount))
> -			hash -= qcount;
> -		return hash;
> +		/* When setting the real_num_tx_queues, we make sure
> +		 * real_num_tx_queues != 0. */

    In the networking code, multiline comments should look like:

/* bla
  * bla
  */

> +		return skb_get_rx_queue(skb) % qcount;
>   	}
>   
>   	if (dev->num_tc) {

MBR, Sergei

Re: [PATCH] net: core: improve the tx_hash calculating

[Eric Dumazet]

On 05/31/2018 06:14 AM, Tonghao Zhang wrote:
> Use the % instead of while, and it may simple code and improve
> the calculating. The real_num_tx_queues has been checked when
> allocating and setting it.
> 
> Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>
> ---
>  net/core/dev.c | 8 +++-----
>  1 file changed, 3 insertions(+), 5 deletions(-)
> 
> diff --git a/net/core/dev.c b/net/core/dev.c
> index 1844d9b..edc5b75 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -2617,15 +2617,13 @@ void netif_device_attach(struct net_device *dev)
>   */
>  static u16 skb_tx_hash(const struct net_device *dev, struct sk_buff *skb)
>  {
> -	u32 hash;
>  	u16 qoffset = 0;
>  	u16 qcount = dev->real_num_tx_queues;
>  
>  	if (skb_rx_queue_recorded(skb)) {
> -		hash = skb_get_rx_queue(skb);
> -		while (unlikely(hash >= qcount))
> -			hash -= qcount;
> -		return hash;
> +		/* When setting the real_num_tx_queues, we make sure
> +		 * real_num_tx_queues != 0. */
> +		return skb_get_rx_queue(skb) % qcount;
>  	}
>  
>  	if (dev->num_tc) {
> 


This patch is adding a divide. We do not need it most of the time.

Re: [PATCH] net: core: improve the tx_hash calculating

[David Miller]

From: Tonghao Zhang <xiangxia.m.yue@gmail.com>
Date: Thu, 31 May 2018 03:14:01 -0700

> Use the % instead of while, and it may simple code and improve
> the calculating. The real_num_tx_queues has been checked when
> allocating and setting it.
> 
> Signed-off-by: Tonghao Zhang <xiangxia.m.yue@gmail.com>

The loop is there to avoid the expensive modulus operation,
not to avoid the real_num_tx_queues == 0 case.