Re: [PATCH net-next v2] net: bonding: add bond_is_icmpv6_nd() helper

[Jay Vosburgh <jv@jvosburgh.net>]

Paolo Abeni <pabeni@redhat.com> wrote:

>On 7/14/25 2:53 PM, Tonghao Zhang wrote:
>>> 2025年7月14日 15:13，Hangbin Liu <liuhangbin@gmail.com> 写道：
>>> Hmm, I don’t see much improvement with this patch compared to without it.
>>> So I don’t think this update is necessary.
>> 
>> This patch use the skb_header_pointer instead of pskb_network_may_pull. The skb_header_pointer is more efficient than pskb_network_may_pull.
>>  And use the comm helper can consolidate some duplicate code.
>
>I think the eventual cleanup here is very subjective, especially
>compared to the diffstat. Any eventual performance improvement should be
>supported by some figures, in relevant tests.
>
>In this specific case I don't think you will be able to measure any
>relevant gain; pskb_network_may_pull() could be slower than
>skb_header_pointer() only when the headers are not in the liner part,
>and that in turns could happen only if we are already on some kind of
>slow path.

	Does the ICMP6 header of a received packet end up in the linear
part, or in a frag?  That's what's being inspected by this code, and I
presumed that the linear area would be just the protocol headers
although thinking about it now, I suppose ICMPv6 is a protocol, but I
don't know how often the pull of the ICMPv6 message data would have to
copy to the linear area.

	-J

---
	-Jay Vosburgh, jv@jvosburgh.net