Merge tag 'mlx5-updates-2020-09-21' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux

Saeed Mahameed says:

====================
mlx5-updates-2020-09-21

Multi packet TX descriptor support for SKBs.

This series introduces some refactoring of the regular TX data path in
mlx5 and adds the Enhanced TX MPWQE feature support. MPWQE stands for
multi-packet work queue element, and it can serve multiple packets,
reducing the PCI bandwidth spent on control traffic. It should improve
performance in scenarios where PCI is the bottleneck, and xmit_more is
signaled by the kernel. The refactoring done in this series also
improves the packet rate on its own.

MPWQE is already implemented in the XDP tx path, this series adds the
support of MPWQE for regular kernel SKB tx path.

MPWQE is supported from ConnectX-5 and onward, for legacy devices we need
to keep backward compatibility for regular (Single packet) WQE descriptor.

MPWQE is not compatible with certain offloads and features, such as TLS
offload, TSO, nonlinear SKBs. If such incompatible features are in use,
the driver gracefully falls back to non-MPWQE per SKB.

Prior to the final patch "net/mlx5e: Enhanced TX MPWQE for SKBs" that adds
the actual support, Maxim did some refactoring to the tx data path to
split it into stages and smaller helper functions that can be utilized and
reused for both legacy and new MPWQE feature.

Performance testing:

UDP performance is improved in a single stream pktgen test:
  Packet rate: 16.86 Mpps (±0.15 Mpps) -> 20.94 Mpps (±0.33 Mpps)
  Instructions per packet: 434 -> 329
  Cycles per packet: 158 -> 123
  Instructions per cycle: 2.75 -> 2.67

TCP and XDP_TX single stream tests show no performance difference.

MPWQE can reduce PCI bandwidth:
  PCI Gen2, pktgen at fixed rate of 36864000 pps on 24 CPU cores:
    Inbound PCI utilization with MPWQE off: 80.3%
    Inbound PCI utilization with MPWQE on: 59.0%
  PCI Gen3, pktgen at fixed rate of 56064000 pps on 24 CPU cores:
    Inbound PCI utilization with MPWQE off: 65.4%
    Inbound PCI utilization with MPWQE on: 49.3%

MPWQE can also reduce CPU load, increasing the packet rate in case of
CPU bottleneck:
  PCI Gen2, pktgen at full rate on 24 CPU cores:
    Packet rate with MPWQE off: 37.5 Mpps
    Packet rate with MPWQE on: 49.0 Mpps
  PCI Gen3, pktgen at full rate on 24 CPU cores:
    Packet rate with MPWQE off: 57.0 Mpps
    Packet rate with MPWQE on: 66.8 Mpps

Burst size in all pktgen tests is 32.

CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz (x86_64)
NIC: Mellanox ConnectX-6 Dx
GCC 10.2.0
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 41 insertions, 42 deletions

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h b/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
index 24336c60123a..07ee1d236ab3 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h
@@ -7,6 +7,21 @@
 #include "en.h"
 #include <linux/indirect_call_wrapper.h>
 
+#define MLX5E_TX_WQE_EMPTY_DS_COUNT (sizeof(struct mlx5e_tx_wqe) / MLX5_SEND_WQE_DS)
+
+/* The mult of MLX5_SEND_WQE_MAX_WQEBBS * MLX5_SEND_WQEBB_NUM_DS
+ * (16 * 4 == 64) does not fit in the 6-bit DS field of Ctrl Segment.
+ * We use a bound lower that MLX5_SEND_WQE_MAX_WQEBBS to let a
+ * full-session WQE be cache-aligned.
+ */
+#if L1_CACHE_BYTES < 128
+#define MLX5E_TX_MPW_MAX_WQEBBS (MLX5_SEND_WQE_MAX_WQEBBS - 1)
+#else
+#define MLX5E_TX_MPW_MAX_WQEBBS (MLX5_SEND_WQE_MAX_WQEBBS - 2)
+#endif
+
+#define MLX5E_TX_MPW_MAX_NUM_DS (MLX5E_TX_MPW_MAX_WQEBBS * MLX5_SEND_WQEBB_NUM_DS)
+
 #define INL_HDR_START_SZ (sizeof(((struct mlx5_wqe_eth_seg *)NULL)->inline_hdr.start))
 
 enum mlx5e_icosq_wqe_type {
@@ -46,8 +61,6 @@ void mlx5e_free_rx_in_progress_descs(struct mlx5e_rq *rq);
 u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 		       struct net_device *sb_dev);
 netdev_tx_t mlx5e_xmit(struct sk_buff *skb, struct net_device *dev);
-void mlx5e_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
-		   struct mlx5e_tx_wqe *wqe, u16 pi, bool xmit_more);
 bool mlx5e_poll_tx_cq(struct mlx5e_cq *cq, int napi_budget);
 void mlx5e_free_txqsq_descs(struct mlx5e_txqsq *sq);
 
@@ -110,6 +123,7 @@ struct mlx5e_tx_wqe_info {
 	u32 num_bytes;
 	u8 num_wqebbs;
 	u8 num_dma;
+	u8 num_fifo_pkts;
 #ifdef CONFIG_MLX5_EN_TLS
 	struct page *resync_dump_frag_page;
 #endif
@@ -194,23 +208,6 @@ static inline u16 mlx5e_icosq_get_next_pi(struct mlx5e_icosq *sq, u16 size)
 }
 
 static inline void
-mlx5e_fill_sq_frag_edge(struct mlx5e_txqsq *sq, struct mlx5_wq_cyc *wq,
-			u16 pi, u16 nnops)
-{
-	struct mlx5e_tx_wqe_info *edge_wi, *wi = &sq->db.wqe_info[pi];
-
-	edge_wi = wi + nnops;
-
-	/* fill sq frag edge with nops to avoid wqe wrapping two pages */
-	for (; wi < edge_wi; wi++) {
-		memset(wi, 0, sizeof(*wi));
-		wi->num_wqebbs = 1;
-		mlx5e_post_nop(wq, sq->sqn, &sq->pc);
-	}
-	sq->stats->nop += nnops;
-}
-
-static inline void
 mlx5e_notify_hw(struct mlx5_wq_cyc *wq, u16 pc, void __iomem *uar_map,
 		struct mlx5_wqe_ctrl_seg *ctrl)
 {
@@ -228,29 +225,6 @@ mlx5e_notify_hw(struct mlx5_wq_cyc *wq, u16 pc, void __iomem *uar_map,
 	mlx5_write64((__be32 *)ctrl, uar_map);
 }
 
-static inline bool mlx5e_transport_inline_tx_wqe(struct mlx5_wqe_ctrl_seg *cseg)
-{
-	return cseg && !!cseg->tis_tir_num;
-}
-
-static inline u8
-mlx5e_tx_wqe_inline_mode(struct mlx5e_txqsq *sq, struct mlx5_wqe_ctrl_seg *cseg,
-			 struct sk_buff *skb)
-{
-	u8 mode;
-
-	if (mlx5e_transport_inline_tx_wqe(cseg))
-		return MLX5_INLINE_MODE_TCP_UDP;
-
-	mode = sq->min_inline_mode;
-
-	if (skb_vlan_tag_present(skb) &&
-	    test_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state))
-		mode = max_t(u8, MLX5_INLINE_MODE_L2, mode);
-
-	return mode;
-}
-
 static inline void mlx5e_cq_arm(struct mlx5e_cq *cq)
 {
 	struct mlx5_core_cq *mcq;
@@ -276,6 +250,23 @@ mlx5e_dma_push(struct mlx5e_txqsq *sq, dma_addr_t addr, u32 size,
 	dma->type = map_type;
 }
 
+static inline struct sk_buff **mlx5e_skb_fifo_get(struct mlx5e_txqsq *sq, u16 i)
+{
+	return &sq->db.skb_fifo[i & sq->skb_fifo_mask];
+}
+
+static inline void mlx5e_skb_fifo_push(struct mlx5e_txqsq *sq, struct sk_buff *skb)
+{
+	struct sk_buff **skb_item = mlx5e_skb_fifo_get(sq, sq->skb_fifo_pc++);
+
+	*skb_item = skb;
+}
+
+static inline struct sk_buff *mlx5e_skb_fifo_pop(struct mlx5e_txqsq *sq)
+{
+	return *mlx5e_skb_fifo_get(sq, sq->skb_fifo_cc++);
+}
+
 static inline void
 mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma)
 {
@@ -291,6 +282,14 @@ mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma)
 	}
 }
 
+void mlx5e_sq_xmit_simple(struct mlx5e_txqsq *sq, struct sk_buff *skb, bool xmit_more);
+void mlx5e_tx_mpwqe_ensure_complete(struct mlx5e_txqsq *sq);
+
+static inline bool mlx5e_tx_mpwqe_is_full(struct mlx5e_tx_mpwqe *session)
+{
+	return session->ds_count == MLX5E_TX_MPW_MAX_NUM_DS;
+}
+
 static inline void mlx5e_rqwq_reset(struct mlx5e_rq *rq)
 {
 	if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {