Re: [PATCH v2 2/4] Defer skb allocation -- new skb_set calls & chain pages in virtio_net

[Rusty Russell <rusty@rustcorp.com.au>]

On Fri, 11 Dec 2009 11:13:02 pm Shirley Ma wrote:
> Signed-off-by: Shirley Ma <xma@us.ibm.com>

I don't think there's a good way of splitting this change across multiple
patches.  And I don't think this patch will compile; I don't think we can
get rid of trim_pages yet.

We *could* first split the receive paths into multiple parts as you have
(eg. add_recvbuf_big etc), then actually rewrite them, but that's too much
work to refactor the code twice.  

So just roll all the driver changes into one patch; so you will have two
patches: one which creates the destroy_bufs API for virtio, and one which
uses it in the virtio_net driver.

> +static struct sk_buff *skb_goodcopy(struct virtnet_info *vi, struct page **page,
> +				    unsigned int *len)

This actually allocates an skb, and transfers the first page to it (via copy
and possibly the first fragment).  skb_from_page() perhaps?

> +		hdr_len = sizeof(hdr->hdr);
> +		offset = sizeof(struct padded_vnet_hdr);
> +	}
> +
> +	memcpy(hdr, p, hdr_len);
> +
> +	*len -= hdr_len;

Perhaps you should return NULL here as an error if *len was < hdr_len?

Thanks,
Rusty.