[v4,net-next,3/9] ice: respect metadata in legacy-rx/ice_construct_skb()

Commit Message

Alexander Lobakin Dec. 8, 2021, 2:06 p.m. UTC

In "legacy-rx" mode represented by ice_construct_skb(), we can
still use XDP (and XDP metadata), but after XDP_PASS the metadata
will be lost as it doesn't get copied to the skb.
Copy it along with the frame headers. Account its size on skb
allocation, and when copying just treat it as a part of the frame
and do a pull after to "move" it to the "reserved" zone.
Point net_prefetch() to xdp->data_meta instead of data. This won't
change anything when the meta is not here, but will save some cache
misses otherwise.

Suggested-by: Jesper Dangaard Brouer <[e-mail]>
Suggested-by: Maciej Fijalkowski <[e-mail]>
Signed-off-by: Alexander Lobakin <[e-mail]>
Reviewed-by: Michal Swiatkowski <[e-mail]>
---
 drivers/net/ethernet/intel/ice/ice_txrx.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

Comments

Bhandare, KiranX Jan. 10, 2022, 10:16 a.m. UTC | #1

> -----Original Message-----
> From: Intel-wired-lan <[e-mail]> On Behalf Of
> Alexander Lobakin
> Sent: Wednesday, December 8, 2021 7:37 PM
> To: [e-mail]
> Cc: Song Liu <[e-mail]>; Alexei Starovoitov <[e-mail]>;
> Andrii Nakryiko <[e-mail]>; Daniel Borkmann
> <[e-mail]>; John Fastabend <[e-mail]>;
> Jesper Dangaard Brouer <[e-mail]>; Yonghong Song
> <[e-mail]>; Jesper Dangaard Brouer <[e-mail]>; KP Singh
> <[e-mail]>; Jakub Kicinski <[e-mail]>;
> [e-mail]; [e-mail]; David S. Miller
> <[e-mail]>; Björn Töpel <[e-mail]>;
> [e-mail]; Martin KaFai Lau <[e-mail]>
> Subject: [Intel-wired-lan] [PATCH v4 net-next 3/9] ice: respect metadata in
> legacy-rx/ice_construct_skb()
> 
> In "legacy-rx" mode represented by ice_construct_skb(), we can still use XDP
> (and XDP metadata), but after XDP_PASS the metadata will be lost as it
> doesn't get copied to the skb.
> Copy it along with the frame headers. Account its size on skb allocation, and
> when copying just treat it as a part of the frame and do a pull after to "move"
> it to the "reserved" zone.
> Point net_prefetch() to xdp->data_meta instead of data. This won't change
> anything when the meta is not here, but will save some cache misses
> otherwise.
> 
> Suggested-by: Jesper Dangaard Brouer <[e-mail]>
> Suggested-by: Maciej Fijalkowski <[e-mail]>
> Signed-off-by: Alexander Lobakin <[e-mail]>
> Reviewed-by: Michal Swiatkowski <[e-mail]>
> ---
>  drivers/net/ethernet/intel/ice/ice_txrx.c | 15 +++++++++++----
>  1 file changed, 11 insertions(+), 4 deletions(-)
> 

Tested-by: Kiran Bhandare <[e-mail]>  A Contingent Worker at Intel

Patch

diff --git a/drivers/net/ethernet/intel/ice/ice_txrx.c b/drivers/net/ethernet/intel/ice/ice_txrx.c
index bc3ba19dc88f..d724b6376c43 100644
--- a/drivers/net/ethernet/intel/ice/ice_txrx.c
+++ b/drivers/net/ethernet/intel/ice/ice_txrx.c
@@ -968,15 +968,17 @@  static struct sk_buff *
 ice_construct_skb(struct ice_rx_ring *rx_ring, struct ice_rx_buf *rx_buf,
 		  struct xdp_buff *xdp)
 {
+	unsigned int metasize = xdp->data - xdp->data_meta;
 	unsigned int size = xdp->data_end - xdp->data;
 	unsigned int headlen;
 	struct sk_buff *skb;
 
 	/* prefetch first cache line of first page */
-	net_prefetch(xdp->data);
+	net_prefetch(xdp->data_meta);
 
 	/* allocate a skb to store the frags */
-	skb = __napi_alloc_skb(&rx_ring->q_vector->napi, ICE_RX_HDR_SIZE,
+	skb = __napi_alloc_skb(&rx_ring->q_vector->napi,
+			       ICE_RX_HDR_SIZE + metasize,
 			       GFP_ATOMIC | __GFP_NOWARN);
 	if (unlikely(!skb))
 		return NULL;
@@ -988,8 +990,13 @@  ice_construct_skb(struct ice_rx_ring *rx_ring, struct ice_rx_buf *rx_buf,
 		headlen = eth_get_headlen(skb->dev, xdp->data, ICE_RX_HDR_SIZE);
 
 	/* align pull length to size of long to optimize memcpy performance */
-	memcpy(__skb_put(skb, headlen), xdp->data, ALIGN(headlen,
-							 sizeof(long)));
+	memcpy(__skb_put(skb, headlen + metasize), xdp->data_meta,
+	       ALIGN(headlen + metasize, sizeof(long)));
+
+	if (metasize) {
+		skb_metadata_set(skb, metasize);
+		__skb_pull(skb, metasize);
+	}
 
 	/* if we exhaust the linear part then add what is left as a frag */
 	size -= headlen;