Re: BUGed to death

[Duncan Sands <baldrick@wanadoo.fr>]

> BUG_ON is already marked unlikely.
> See include/linux/kernel.h
>
> The costs here are doing the actual checks, nothing to do with
> the branch prediction.

Some places don't seem to know about BUG_ON, for example
(from include/linux/skbuff.h):

static inline char *__skb_pull(struct sk_buff *skb, unsigned int len)
{
        skb->len -= len;
        if (skb->len < skb->data_len)
                BUG();
        return skb->data += len;
}