[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <56935DB5.1030502@gmail.com>
Date: Mon, 11 Jan 2016 15:45:57 +0800
From: Zang MingJie <zealot0630@...il.com>
To: netdev@...r.kernel.org
Cc: dev@...nvswitch.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] net: preserve IP control block during GSO segmentation
On 01/08/2016 08:13 PM, David Laight wrote:
> You could set SKB_SGO_CB_OFFSET to sizeof ((skb)->cb) - sizeof (struct skb_gso_cb)
> so that the end of 'cb' is always used.
> (Assuming the former is a multiple of 4.)
>
> It might be worth using an on-stack structure passed through as a separate
> parameter - it doesn't look as though it has to be queued with the skb.
> (Clearly a bigger change.)
I would definitely prefer the stack structure.
As a kernel developer, sometime I can hardly figure out which struct
current cb is without debug it, and the worst they are not documented
anywhere. I can hardly know the life time of the cb types.
If using a stack, things can be much easier. only an extra var to store
the stack top:
int cb_top;
and several macro to manage the stack:
SKB_CB_PUSH(skb, type)
SKB_CB_POP(skb)
SKB_CB_TOP(skb, type)
and maybe a debug variable to store current cb type.
All current cb macro can be replaced by SKB_CB_TOP.
Although it is a big change, I think it worths, for both performance and
maintainability
Powered by blists - more mailing lists