xsk: i40e: ice: ixgbe: mlx5: Test for dma_need_sync earlier for better performance

Test for dma_need_sync earlier to increase
performance. xsk_buff_dma_sync_for_cpu() takes an xdp_buff as
parameter and from that the xsk_buff_pool reference is dug out. Perf
shows that this dereference causes a lot of cache misses. But as the
buffer pool is now sent down to the driver at zero-copy initialization
time, we might as well use this pointer directly, instead of going via
the xsk_buff and we can do so already in xsk_buff_dma_sync_for_cpu()
instead of in xp_dma_sync_for_cpu. This gets rid of these cache
misses.

Throughput increases with 3% for the xdpsock l2fwd sample application
on my machine.
Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Björn Töpel <bjorn.topel@intel.com>
Link: https://lore.kernel.org/bpf/1598603189-32145-11-git-send-email-magnus.karlsson@intel.com

drivers/net/ethernet/intel/i40e/i40e_xsk.c

@@ -314,7 +314,7 @@ int i40e_clean_rx_irq_zc(struct i40e_ring *rx_ring, int budget)
 
 		bi = i40e_rx_bi(rx_ring, rx_ring->next_to_clean);
 		(*bi)->data_end = (*bi)->data + size;
-		xsk_buff_dma_sync_for_cpu(*bi);
+		xsk_buff_dma_sync_for_cpu(*bi, rx_ring->xsk_pool);
 
 		xdp_res = i40e_run_xdp_zc(rx_ring, *bi);
 		if (xdp_res) {

drivers/net/ethernet/intel/ice/ice_xsk.c

@@ -595,7 +595,7 @@ int ice_clean_rx_irq_zc(struct ice_ring *rx_ring, int budget)
 
 		rx_buf = &rx_ring->rx_buf[rx_ring->next_to_clean];
 		rx_buf->xdp->data_end = rx_buf->xdp->data + size;
-		xsk_buff_dma_sync_for_cpu(rx_buf->xdp);
+		xsk_buff_dma_sync_for_cpu(rx_buf->xdp, rx_ring->xsk_pool);
 
 		xdp_res = ice_run_xdp_zc(rx_ring, rx_buf->xdp);
 		if (xdp_res) {

drivers/net/ethernet/intel/ixgbe/ixgbe_xsk.c

@@ -287,7 +287,7 @@ int ixgbe_clean_rx_irq_zc(struct ixgbe_q_vector *q_vector,
 		}
 
 		bi->xdp->data_end = bi->xdp->data + size;
-		xsk_buff_dma_sync_for_cpu(bi->xdp);
+		xsk_buff_dma_sync_for_cpu(bi->xdp, rx_ring->xsk_pool);
 		xdp_res = ixgbe_run_xdp_zc(adapter, rx_ring, bi->xdp);
 
 		if (xdp_res) {

drivers/net/ethernet/mellanox/mlx5/core/en/xsk/rx.c

@@ -48,7 +48,7 @@ struct sk_buff *mlx5e_xsk_skb_from_cqe_mpwrq_linear(struct mlx5e_rq *rq,
 
 	xdp->data_end = xdp->data + cqe_bcnt32;
 	xdp_set_data_meta_invalid(xdp);
-	xsk_buff_dma_sync_for_cpu(xdp);
+	xsk_buff_dma_sync_for_cpu(xdp, rq->xsk_pool);
 	prefetch(xdp->data);
 
 	rcu_read_lock();
@@ -99,7 +99,7 @@ struct sk_buff *mlx5e_xsk_skb_from_cqe_linear(struct mlx5e_rq *rq,
 
 	xdp->data_end = xdp->data + cqe_bcnt;
 	xdp_set_data_meta_invalid(xdp);
-	xsk_buff_dma_sync_for_cpu(xdp);
+	xsk_buff_dma_sync_for_cpu(xdp, rq->xsk_pool);
 	prefetch(xdp->data);
 
 	if (unlikely(get_cqe_opcode(cqe) != MLX5_CQE_RESP_SEND)) {

include/net/xdp_sock_drv.h

@@ -99,10 +99,13 @@ static inline void *xsk_buff_raw_get_data(struct xsk_buff_pool *pool, u64 addr)
 	return xp_raw_get_data(pool, addr);
 }
 
-static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp)
+static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool)
 {
 	struct xdp_buff_xsk *xskb = container_of(xdp, struct xdp_buff_xsk, xdp);
 
+	if (!pool->dma_need_sync)
+		return;
+
 	xp_dma_sync_for_cpu(xskb);
 }
 
@@ -222,7 +225,7 @@ static inline void *xsk_buff_raw_get_data(struct xsk_buff_pool *pool, u64 addr)
 	return NULL;
 }
 
-static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp)
+static inline void xsk_buff_dma_sync_for_cpu(struct xdp_buff *xdp, struct xsk_buff_pool *pool)
 {
 }
 

include/net/xsk_buff_pool.h

@@ -114,9 +114,6 @@ static inline dma_addr_t xp_get_frame_dma(struct xdp_buff_xsk *xskb)
 void xp_dma_sync_for_cpu_slow(struct xdp_buff_xsk *xskb);
 static inline void xp_dma_sync_for_cpu(struct xdp_buff_xsk *xskb)
 {
-	if (!xskb->pool->dma_need_sync)
-		return;
-
 	xp_dma_sync_for_cpu_slow(xskb);
 }