net/mlx5e: TX latency optimization to save DMA reads

A regular TX WQE execution involves two or more DMA reads -
one to fetch the WQE, and another one per WQE gather entry.

These DMA reads obviously increase the TX latency.
There are two mlx5 mechanisms to bypass these DMA reads:
1) Inline WQE
2) Blue Flame (BF)

An inline WQE contains a whole packet, thus saves the DMA read/s
of the regular WQE gather entry/s. Inline WQE support was already
added in the previous commit.

A BF WQE is written directly to the device I/O mapped memory, thus
enables saving the DMA read that fetches the WQE.

The BF WQE I/O write must be in cache line granularity, thus uses
the CPU write combining mechanism.
A BF WQE I/O write acts also as a TX doorbell for notifying the
device of new TX WQEs.
A BF WQE is written to the same I/O mapped address as the regular TX
doorbell, thus this address is being mapped twice - once by ioremap()
and once by io_mapping_map_wc().

While both mechanisms reduce the TX latency, they both consume more CPU
cycles than a regular WQE:
- A BF WQE must still be written to host memory, in addition to being
  written directly to the device I/O mapped memory.
- An inline WQE involves copying the SKB data into it.

To handle this tradeoff, we introduce here a heuristic algorithm that
strives to avoid using these two mechanisms in case the TX queue is
being back-pressured by the device, and limit their usage rate otherwise.

An inline WQE will always be "Blue Flamed" (written directly to the
device I/O mapped memory) while a BF WQE may not be inlined (may contain
gather entries).

Preliminary testing using netperf UDP_RR shows that the latency goes down
from 17.5us to 16.9us, while the message rate (tested with pktgen) stays
the same.

Signed-off-by: Achiad Shochat <achiad@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 21 insertions, 5 deletions

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
index 351ac6982e22..64380bc0cd6a 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
@@ -57,7 +57,7 @@ void mlx5e_send_nop(struct mlx5e_sq *sq, bool notify_hw)
 
 	if (notify_hw) {
 		cseg->fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE;
-		mlx5e_tx_notify_hw(sq, wqe);
+		mlx5e_tx_notify_hw(sq, wqe, 0);
 	}
 }
 
@@ -110,7 +110,7 @@ u16 mlx5e_select_queue(struct net_device *dev, struct sk_buff *skb,
 }
 
 static inline u16 mlx5e_get_inline_hdr_size(struct mlx5e_sq *sq,
-					    struct sk_buff *skb)
+					    struct sk_buff *skb, bool bf)
 {
 	/* Some NIC TX decisions, e.g loopback, are based on the packet
 	 * headers and occur before the data gather.
@@ -118,7 +118,7 @@ static inline u16 mlx5e_get_inline_hdr_size(struct mlx5e_sq *sq,
 	 */
 #define MLX5E_MIN_INLINE (ETH_HLEN + 2/*vlan tag*/)
 
-	if (skb_headlen(skb) <= sq->max_inline)
+	if (bf && (skb_headlen(skb) <= sq->max_inline))
 		return skb_headlen(skb);
 
 	return MLX5E_MIN_INLINE;
@@ -137,6 +137,7 @@ static netdev_tx_t mlx5e_sq_xmit(struct mlx5e_sq *sq, struct sk_buff *skb)
 
 	u8  opcode = MLX5_OPCODE_SEND;
 	dma_addr_t dma_addr = 0;
+	bool bf = false;
 	u16 headlen;
 	u16 ds_cnt;
 	u16 ihs;
@@ -149,6 +150,11 @@ static netdev_tx_t mlx5e_sq_xmit(struct mlx5e_sq *sq, struct sk_buff *skb)
 	else
 		sq->stats.csum_offload_none++;
 
+	if (sq->cc != sq->prev_cc) {
+		sq->prev_cc = sq->cc;
+		sq->bf_budget = (sq->cc == sq->pc) ? MLX5E_SQ_BF_BUDGET : 0;
+	}
+
 	if (skb_is_gso(skb)) {
 		u32 payload_len;
 
@@ -161,7 +167,10 @@ static netdev_tx_t mlx5e_sq_xmit(struct mlx5e_sq *sq, struct sk_buff *skb)
 		sq->stats.tso_packets++;
 		sq->stats.tso_bytes += payload_len;
 	} else {
-		ihs = mlx5e_get_inline_hdr_size(sq, skb);
+		bf = sq->bf_budget &&
+		     !skb->xmit_more &&
+		     !skb_shinfo(skb)->nr_frags;
+		ihs = mlx5e_get_inline_hdr_size(sq, skb, bf);
 		MLX5E_TX_SKB_CB(skb)->num_bytes = max_t(unsigned int, skb->len,
 							ETH_ZLEN);
 	}
@@ -233,14 +242,21 @@ static netdev_tx_t mlx5e_sq_xmit(struct mlx5e_sq *sq, struct sk_buff *skb)
 	}
 
 	if (!skb->xmit_more || netif_xmit_stopped(sq->txq)) {
+		int bf_sz = 0;
+
+		if (bf && sq->uar_bf_map)
+			bf_sz = MLX5E_TX_SKB_CB(skb)->num_wqebbs << 3;
+
 		cseg->fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE;
-		mlx5e_tx_notify_hw(sq, wqe);
+		mlx5e_tx_notify_hw(sq, wqe, bf_sz);
 	}
 
 	/* fill sq edge with nops to avoid wqe wrap around */
 	while ((sq->pc & wq->sz_m1) > sq->edge)
 		mlx5e_send_nop(sq, false);
 
+	sq->bf_budget = bf ? sq->bf_budget - 1 : 0;
+
 	sq->stats.packets++;
 	return NETDEV_TX_OK;