Re: [PATCH] Software receive packet steering

[Tom Herbert <therbert@google.com>]

On Wed, Apr 8, 2009 at 5:36 PM, David Miller <davem@davemloft.net> wrote:
> From: Tom Herbert <therbert@google.com>
> Date: Wed, 8 Apr 2009 15:48:12 -0700
>
>> +#ifdef CONFIG_NET_SOFTRPS
>> +     int                     rps_cpu;
>> +     struct list_head        rps_poll_list;
>> +     spinlock_t              rps_poll_list_lock;
>> +     struct call_single_data rps_csd;
>> +     unsigned long           rps_flags;
>> +#define RPS_SOFTIRQ_PENDING  0x1
>> +#define RPS_SOFTIRQ_COMPLETING       0x2
>> +#endif
>
> Have you seen my patch that does this with remote softirqs?
> Then you don't need a lock for the list, it can be lockless
> since only the local processor ever accesses the list.
>
> In fact, it kills the per-cpu backlog completely, everything
> takes the netif_receive_skb() path and shoots the packet to
> the remote cpu.
>
> Here is that patch, for reference.  It won't apply to anything
> recent but all the ideas are there.
>

Thanks for the reference to that patch, are you planning to apply it?