| Message ID | 20211208140702.642741-7-alexandr.lobakin@intel.com (mailing list archive) |
|---|---|
| State | Awaiting Upstream |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | net: intel: napi_alloc_skb() vs metadata | expand |
On 12/8/2021 16:06, Alexander Lobakin wrote: > {__,}napi_alloc_skb() allocates and reserves additional NET_SKB_PAD > + NET_IP_ALIGN for any skb. > OTOH, igc_construct_skb_zc() currently allocates and reserves > additional `xdp->data_meta - xdp->data_hard_start`, which is about > XDP_PACKET_HEADROOM for XSK frames. > There's no need for that at all as the frame is post-XDP and will > go only to the networking stack core. > Pass the size of the actual data only (+ meta) to > __napi_alloc_skb() and don't reserve anything. This will give > enough headroom for stack processing. > Also, net_prefetch() xdp->data_meta and align the copy size to > speed-up memcpy() a little and better match igc_costruct_skb(). > > Fixes: fc9df2a0b520 ("igc: Enable RX via AF_XDP zero-copy") > Signed-off-by: Alexander Lobakin <alexandr.lobakin@intel.com> > Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com> > --- > drivers/net/ethernet/intel/igc/igc_main.c | 13 +++++++------ > 1 file changed, 7 insertions(+), 6 deletions(-) > Tested-by: Nechama Kraus <nechamax.kraus@linux.intel.com>
diff --git a/drivers/net/ethernet/intel/igc/igc_main.c b/drivers/net/ethernet/intel/igc/igc_main.c index 142c57b7a451..a2e8d43be3a1 100644 --- a/drivers/net/ethernet/intel/igc/igc_main.c +++ b/drivers/net/ethernet/intel/igc/igc_main.c @@ -2446,19 +2446,20 @@ static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget) static struct sk_buff *igc_construct_skb_zc(struct igc_ring *ring, struct xdp_buff *xdp) { + unsigned int totalsize = xdp->data_end - xdp->data_meta; unsigned int metasize = xdp->data - xdp->data_meta; - unsigned int datasize = xdp->data_end - xdp->data; - unsigned int totalsize = metasize + datasize; struct sk_buff *skb; - skb = __napi_alloc_skb(&ring->q_vector->napi, - xdp->data_end - xdp->data_hard_start, + net_prefetch(xdp->data_meta); + + skb = __napi_alloc_skb(&ring->q_vector->napi, totalsize, GFP_ATOMIC | __GFP_NOWARN); if (unlikely(!skb)) return NULL; - skb_reserve(skb, xdp->data_meta - xdp->data_hard_start); - memcpy(__skb_put(skb, totalsize), xdp->data_meta, totalsize); + memcpy(__skb_put(skb, totalsize), xdp->data_meta, + ALIGN(totalsize, sizeof(long))); + if (metasize) { skb_metadata_set(skb, metasize); __skb_pull(skb, metasize);
{__,}napi_alloc_skb() allocates and reserves additional NET_SKB_PAD + NET_IP_ALIGN for any skb. OTOH, igc_construct_skb_zc() currently allocates and reserves additional `xdp->data_meta - xdp->data_hard_start`, which is about XDP_PACKET_HEADROOM for XSK frames. There's no need for that at all as the frame is post-XDP and will go only to the networking stack core. Pass the size of the actual data only (+ meta) to __napi_alloc_skb() and don't reserve anything. This will give enough headroom for stack processing. Also, net_prefetch() xdp->data_meta and align the copy size to speed-up memcpy() a little and better match igc_costruct_skb(). Fixes: fc9df2a0b520 ("igc: Enable RX via AF_XDP zero-copy") Signed-off-by: Alexander Lobakin <alexandr.lobakin@intel.com> Reviewed-by: Michal Swiatkowski <michal.swiatkowski@linux.intel.com> --- drivers/net/ethernet/intel/igc/igc_main.c | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-)