On Mon, Sep 29, 2008 at 09:39:31AM +0400, Evgeniy Polyakov wrote:

Hi Lennert.

Hi Evgeniy,

quoted

Doesn't it?  It calls skb_release_head_state() (new function), which
should do that.

Argh, you are right, I missed this hunk:
-/* Free everything but the sk_buff shell. */
-static void skb_release_all(struct sk_buff *skb)
+static void skb_release_head_state(struct sk_buff *skb) 

Patches look good.

Cool, thanks for having a look.

Could you (or anyone else) try this patch on one of your boxes and
see what kind of effect it has on performance?


cheers,
Lennert