Re: [Xen-devel] [PATCH] xen/netfront: handle compound page fragments on transmit

[Eric Dumazet <eric.dumazet@xxxxxxxxx>]

On Tue, 2012-11-20 at 11:40 +0000, Ian Campbell wrote:
> An SKB paged fragment can consist of a compound page with order > 0.
> However the netchannel protocol deals only in PAGE_SIZE frames.
> 
> Handle this in xennet_make_frags by iterating over the frames which
> make up the page.
> 
> This is the netfront equivalent to 6a8ed462f16b for netback.
> 
> Signed-off-by: Ian Campbell <ian.campbell@xxxxxxxxxx>
> Cc: netdev@xxxxxxxxxxxxxxx
> Cc: xen-devel@xxxxxxxxxxxxx
> Cc: Eric Dumazet <edumazet@xxxxxxxxxx>
> Cc: Konrad Rzeszutek Wilk <konrad@xxxxxxxxxx>
> Cc: ANNIE LI <annie.li@xxxxxxxxxx>
> Cc: Sander Eikelenboom <linux@xxxxxxxxxxxxxx>
> Cc: Stefan Bader <stefan.bader@xxxxxxxxxxxxx>
> ---
>  drivers/net/xen-netfront.c |   58 +++++++++++++++++++++++++++++++++----------
>  1 files changed, 44 insertions(+), 14 deletions(-)
> 
> diff --git a/drivers/net/xen-netfront.c b/drivers/net/xen-netfront.c
> index caa0110..a12b99a 100644
> --- a/drivers/net/xen-netfront.c
> +++ b/drivers/net/xen-netfront.c
> @@ -452,24 +452,54 @@ static void xennet_make_frags(struct sk_buff *skb, 
> struct net_device *dev,
>       /* Grant backend access to each skb fragment page. */
>       for (i = 0; i < frags; i++) {
>               skb_frag_t *frag = skb_shinfo(skb)->frags + i;
> +             struct page *page = skb_frag_page(frag);
> +             unsigned long size = skb_frag_size(frag);
> +             unsigned long offset = frag->page_offset;
>  
> -             tx->flags |= XEN_NETTXF_more_data;
> +             /* Data must not cross a page boundary. */
> +             BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
>  
> -             id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
> -             np->tx_skbs[id].skb = skb_get(skb);
> -             tx = RING_GET_REQUEST(&np->tx, prod++);
> -             tx->id = id;
> -             ref = gnttab_claim_grant_reference(&np->gref_tx_head);
> -             BUG_ON((signed short)ref < 0);
> +             /* Skip unused frames from start of page */

'frame' in the comment means an order-0 page ?

> +             page += offset >> PAGE_SHIFT;
> +             offset &= ~PAGE_MASK;
>  
> -             mfn = pfn_to_mfn(page_to_pfn(skb_frag_page(frag)));
> -             gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
> -                                             mfn, GNTMAP_readonly);
> +             while (size > 0) {
> +                     unsigned long bytes;
>  
> -             tx->gref = np->grant_tx_ref[id] = ref;
> -             tx->offset = frag->page_offset;
> -             tx->size = skb_frag_size(frag);
> -             tx->flags = 0;
> +                     BUG_ON(offset >= PAGE_SIZE);
> +
> +                     bytes = PAGE_SIZE - offset;
> +                     if (bytes > size)
> +                             bytes = size;
> +
> +                     tx->flags |= XEN_NETTXF_more_data;
> +
> +                     id = get_id_from_freelist(&np->tx_skb_freelist, 
> np->tx_skbs);
> +                     np->tx_skbs[id].skb = skb_get(skb);

BTW this skb_get() means extra atomic operations for every 4096 bytes
unit, and an extra atomic op (and test for final 0) at TX completion.
This could be avoided, by setting np->tx_skbs[id].skb = skb only for the
very last unit.

> +                     tx = RING_GET_REQUEST(&np->tx, prod++);
> +                     tx->id = id;
> +                     ref = gnttab_claim_grant_reference(&np->gref_tx_head);
> +                     BUG_ON((signed short)ref < 0);
> +
> +                     mfn = pfn_to_mfn(page_to_pfn(page));
> +                     gnttab_grant_foreign_access_ref(ref, 
> np->xbdev->otherend_id,
> +                                                     mfn, GNTMAP_readonly);
> +
> +                     tx->gref = np->grant_tx_ref[id] = ref;
> +                     tx->offset = offset;
> +                     tx->size = bytes;
> +                     tx->flags = 0;
> +
> +                     offset += bytes;
> +                     size -= bytes;
> +
> +                     /* Next frame */
> +                     if (offset == PAGE_SIZE && size) {
> +                             BUG_ON(!PageCompound(page));
> +                             page++;
> +                             offset = 0;
> +                     }
> +             }
>       }
>  
>       np->tx.req_prod_pvt = prod;

Acked-by: Eric Dumazet <edumazet@xxxxxxxxxx>

Thanks !



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel