Re: [PATCH RFC 00/14] shrink skb cb to 44 bytes

[Florian Westphal <fw@strlen.de>]

David Miller <davem@davemloft.net> wrote:
> From: Eric Dumazet <eric.dumazet@gmail.com>
> Date: Mon, 02 Mar 2015 20:02:05 -0800
> 
> > About the GRO layout change, reason why 'struct sk_buff *last;' is at
> > the end of struct napi_gro_cb is that this field is not used in fast
> > path.
> 
> Understood.

Moved it back to end, thanks!

> While reviewing this I noticed that the jiffies timestamp in GRO cb
> could really be u32 if we want instead of full "unsigned long".

Made it an u16 at the moment -- then __packed is no longer needed
and it fits in 40 bytes.

> > Note : We could try to use one bit in skb to advertise zero shinfo(skb).
> > 
> > Many skbs have a zeroed shinfo() (but shinfo->dataref == 1) , and
> > dereferencing skb_shinfo adds a cache line miss. 
> > 
> > -> We could avoid memset(shinfo, 0, offsetof(struct skb_shared_info,
> > dataref)) & atomic_set(&shinfo->dataref, 1); 
> > 
> >  in alloc_skb() and friends completely.
> > 
> > Unfortunately this kind of change would be quite invasive...
>
> Right, all these kinds of things touch everything.

Indeed, but thanks for the hint Eric -- I'll investigate.