Re: [PATCH] NIU support for skb->rxhash

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

Le jeudi 22 avril 2010 à 04:21 -0700, David Miller a écrit :
> But it turns out using it is largely pointless since the only way to
> get the hash value(s) is through a structure which is prepended to the
> packet data (so we take a cache miss on the packet data anyways)
> instead of being able to fetch it out of the RX descriptors :-/
> 
> If anyone out there is trying to design sane hardware, please put the
> following into your RX descriptors:
> 
> 1) ethernet protocol type (u16)
> 2) a flag bit indicating if the packet destination matched one
>    of the programmed unicast MAC addresses
> 3) a flag bit indicating "multicast"
> 4) a flag bit indicating "broadcast"
> 5) at least 32-bits of the computed flow hash (u32)
> 
> kthx, bye!

Then, our stack also touch all 256 bytes of skb structure itself.

offsetof(struct sk_buff, next)    =0x0
offsetof(struct sk_buff, rxhash)  =0xa8
offsetof(struct sk_buff, dev)     =0x20
offsetof(struct sk_buff, len)     =0x68
offsetof(struct sk_buff, protocol)=0x7e
offsetof(struct sk_buff, network_header)=0xc0
offsetof(struct sk_buff, data)    =0xd8
offsetof(struct sk_buff, head)    =0xd0

Time for a reordering I guess ;)