Re: [net-next 0/6] move CAN skb headroom content to skb extensions

[Florian Westphal <fw@strlen.de>]

Oliver Hartkopp <socketcan@hartkopp.net> wrote:
> Ok. But then I don't see any real pressure to do the extension right 
> now. There doesn't seem so much changes adding new skb extensions users. 
>   And we still have a free slot even if all users would have been 
> enabled (which is not the case due to the mutually exclusive options).

Yes.

> The good thing is that skb extensions are very efficient. Which leads to 
> the bad thing that we only can detect the problems at build time with
> 
> BUILD_BUG_ON(SKB_EXT_NUM > 8);
> BUILD_BUG_ON(skb_ext_total_length() > 255);
> 
> For (SKB_EXT_NUM > 8) the upgrade of active_extensions to u16 should 
> simply make it. Probably with some #ifdef magic only.
> 
> But thinking about BUILD_BUG_ON(skb_ext_total_length() > 255):
> 
> Shouldn't this be SKB_EXT_CHUNKSIZEOF(struct skb_ext) + sum of the skb 
> ext user data which can be up to 255 *  SKB_EXT_ALIGN_VALUE (= 2040) ???

Why?  Its not about the size in bytes, its there to make sure
ext->offset[] (u8) won't overflow for any of the extensions.

Maybe a comment would help.