* [PATCH net-next 2/4] net: use skb_get_hash in tx queue selection
@ 2014-03-21 0:09 Tom Herbert
2014-03-21 1:31 ` Eric Dumazet
0 siblings, 1 reply; 4+ messages in thread
From: Tom Herbert @ 2014-03-21 0:09 UTC (permalink / raw)
To: davem, netdev
__skb_get_hash tries to get the skb->hash from a connected socket
associated with an skb. __skb_tx_hash and get_xps_queue just call
skb_get_hash().
Signed-off-by: Tom Herbert <therbert@google.com>
---
include/linux/netdevice.h | 2 +-
include/linux/skbuff.h | 2 +-
net/core/flow_dissector.c | 22 +++++++++-------------
3 files changed, 11 insertions(+), 15 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 4b6d12c..a1fa2e4 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -2481,7 +2481,7 @@ static inline int netif_set_xps_queue(struct net_device *dev,
* as a distribution range limit for the returned value.
*/
static inline u16 skb_tx_hash(const struct net_device *dev,
- const struct sk_buff *skb)
+ struct sk_buff *skb)
{
return __skb_tx_hash(dev, skb, dev->real_num_tx_queues);
}
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index aa2c22c..f392b74 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -2863,7 +2863,7 @@ static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
return skb->queue_mapping != 0;
}
-u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
+u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
unsigned int num_tx_queues);
static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
diff --git a/net/core/flow_dissector.c b/net/core/flow_dissector.c
index 107ed12..432d35a 100644
--- a/net/core/flow_dissector.c
+++ b/net/core/flow_dissector.c
@@ -212,6 +212,12 @@ void __skb_get_hash(struct sk_buff *skb)
struct flow_keys keys;
u32 hash;
+ if (skb->sk && skb->sk->sk_hash) {
+ skb->hash = skb->sk->sk_hash;
+ skb->l4_hash = 1;
+ return;
+ }
+
if (!skb_flow_dissect(skb, &keys))
return;
@@ -240,7 +246,7 @@ EXPORT_SYMBOL(__skb_get_hash);
* Returns a Tx hash based on the given packet descriptor a Tx queues' number
* to be used as a distribution range.
*/
-u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
+u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
unsigned int num_tx_queues)
{
u32 hash;
@@ -260,11 +266,7 @@ u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
qcount = dev->tc_to_txq[tc].count;
}
- if (skb->sk && skb->sk->sk_hash)
- hash = skb->sk->sk_hash;
- else
- hash = (__force u16) skb->protocol;
- hash = __flow_hash_1word(hash);
+ hash = __flow_hash_1word(skb_get_hash(skb));
return (u16) (((u64) hash * qcount) >> 32) + qoffset;
}
@@ -339,13 +341,7 @@ static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
if (map->len == 1)
queue_index = map->queues[0];
else {
- u32 hash;
- if (skb->sk && skb->sk->sk_hash)
- hash = skb->sk->sk_hash;
- else
- hash = (__force u16) skb->protocol ^
- skb->hash;
- hash = __flow_hash_1word(hash);
+ u32 hash = __flow_hash_1word(skb_get_hash(skb));
queue_index = map->queues[
((u64)hash * map->len) >> 32];
}
--
1.9.1.423.g4596e3a
^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [PATCH net-next 2/4] net: use skb_get_hash in tx queue selection
2014-03-21 0:09 [PATCH net-next 2/4] net: use skb_get_hash in tx queue selection Tom Herbert
@ 2014-03-21 1:31 ` Eric Dumazet
2014-03-21 1:37 ` Eric Dumazet
2014-03-21 19:24 ` Tom Herbert
0 siblings, 2 replies; 4+ messages in thread
From: Eric Dumazet @ 2014-03-21 1:31 UTC (permalink / raw)
To: Tom Herbert; +Cc: davem, netdev
On Thu, 2014-03-20 at 17:09 -0700, Tom Herbert wrote:
> __skb_get_hash tries to get the skb->hash from a connected socket
> associated with an skb. __skb_tx_hash and get_xps_queue just call
> skb_get_hash().
>
> Signed-off-by: Tom Herbert <therbert@google.com>
> ---
> include/linux/netdevice.h | 2 +-
> include/linux/skbuff.h | 2 +-
> net/core/flow_dissector.c | 22 +++++++++-------------
> 3 files changed, 11 insertions(+), 15 deletions(-)
>
> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> index 4b6d12c..a1fa2e4 100644
> --- a/include/linux/netdevice.h
> +++ b/include/linux/netdevice.h
> @@ -2481,7 +2481,7 @@ static inline int netif_set_xps_queue(struct net_device *dev,
> * as a distribution range limit for the returned value.
> */
> static inline u16 skb_tx_hash(const struct net_device *dev,
> - const struct sk_buff *skb)
> + struct sk_buff *skb)
> {
> return __skb_tx_hash(dev, skb, dev->real_num_tx_queues);
> }
> diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
> index aa2c22c..f392b74 100644
> --- a/include/linux/skbuff.h
> +++ b/include/linux/skbuff.h
> @@ -2863,7 +2863,7 @@ static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
> return skb->queue_mapping != 0;
> }
>
> -u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
> +u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
> unsigned int num_tx_queues);
>
> static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
> diff --git a/net/core/flow_dissector.c b/net/core/flow_dissector.c
> index 107ed12..432d35a 100644
> --- a/net/core/flow_dissector.c
> +++ b/net/core/flow_dissector.c
> @@ -212,6 +212,12 @@ void __skb_get_hash(struct sk_buff *skb)
> struct flow_keys keys;
> u32 hash;
>
> + if (skb->sk && skb->sk->sk_hash) {
> + skb->hash = skb->sk->sk_hash;
> + skb->l4_hash = 1;
> + return;
> + }
Wouldn't it be faster to simply set skb->hash in providers (TCP, UDP),
to that we do not add this test here ?
After packet has been a while in a qdisc, access to skb->sk_hash will
likely trigger a cache line miss.
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH net-next 2/4] net: use skb_get_hash in tx queue selection
2014-03-21 1:31 ` Eric Dumazet
@ 2014-03-21 1:37 ` Eric Dumazet
2014-03-21 19:24 ` Tom Herbert
1 sibling, 0 replies; 4+ messages in thread
From: Eric Dumazet @ 2014-03-21 1:37 UTC (permalink / raw)
To: Tom Herbert; +Cc: davem, netdev
On Thu, 2014-03-20 at 18:31 -0700, Eric Dumazet wrote:
> Wouldn't it be faster to simply set skb->hash in providers (TCP, UDP),
> to that we do not add this test here ?
>
> After packet has been a while in a qdisc, access to skb->sk_hash will
> likely trigger a cache line miss.
I meant access to skb->sk->sk_hash will likely trigger a cache line
miss.
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH net-next 2/4] net: use skb_get_hash in tx queue selection
2014-03-21 1:31 ` Eric Dumazet
2014-03-21 1:37 ` Eric Dumazet
@ 2014-03-21 19:24 ` Tom Herbert
1 sibling, 0 replies; 4+ messages in thread
From: Tom Herbert @ 2014-03-21 19:24 UTC (permalink / raw)
To: Eric Dumazet; +Cc: David Miller, Linux Netdev List
On Thu, Mar 20, 2014 at 6:31 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> On Thu, 2014-03-20 at 17:09 -0700, Tom Herbert wrote:
>> __skb_get_hash tries to get the skb->hash from a connected socket
>> associated with an skb. __skb_tx_hash and get_xps_queue just call
>> skb_get_hash().
>>
>> Signed-off-by: Tom Herbert <therbert@google.com>
>> ---
>> include/linux/netdevice.h | 2 +-
>> include/linux/skbuff.h | 2 +-
>> net/core/flow_dissector.c | 22 +++++++++-------------
>> 3 files changed, 11 insertions(+), 15 deletions(-)
>>
>> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
>> index 4b6d12c..a1fa2e4 100644
>> --- a/include/linux/netdevice.h
>> +++ b/include/linux/netdevice.h
>> @@ -2481,7 +2481,7 @@ static inline int netif_set_xps_queue(struct net_device *dev,
>> * as a distribution range limit for the returned value.
>> */
>> static inline u16 skb_tx_hash(const struct net_device *dev,
>> - const struct sk_buff *skb)
>> + struct sk_buff *skb)
>> {
>> return __skb_tx_hash(dev, skb, dev->real_num_tx_queues);
>> }
>> diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
>> index aa2c22c..f392b74 100644
>> --- a/include/linux/skbuff.h
>> +++ b/include/linux/skbuff.h
>> @@ -2863,7 +2863,7 @@ static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
>> return skb->queue_mapping != 0;
>> }
>>
>> -u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
>> +u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
>> unsigned int num_tx_queues);
>>
>> static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
>> diff --git a/net/core/flow_dissector.c b/net/core/flow_dissector.c
>> index 107ed12..432d35a 100644
>> --- a/net/core/flow_dissector.c
>> +++ b/net/core/flow_dissector.c
>> @@ -212,6 +212,12 @@ void __skb_get_hash(struct sk_buff *skb)
>> struct flow_keys keys;
>> u32 hash;
>>
>> + if (skb->sk && skb->sk->sk_hash) {
>> + skb->hash = skb->sk->sk_hash;
>> + skb->l4_hash = 1;
>> + return;
>> + }
>
> Wouldn't it be faster to simply set skb->hash in providers (TCP, UDP),
> to that we do not add this test here ?
>
Excellent idea. We should be able to eliminate all occurrences of "if
(skb->sk && skb->sk->sk_hash)" in the stack!
> After packet has been a while in a qdisc, access to skb->sk_hash will
> likely trigger a cache line miss.
>
>
>
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2014-03-21 19:24 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-03-21 0:09 [PATCH net-next 2/4] net: use skb_get_hash in tx queue selection Tom Herbert
2014-03-21 1:31 ` Eric Dumazet
2014-03-21 1:37 ` Eric Dumazet
2014-03-21 19:24 ` Tom Herbert
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.