[dpdk-dev] [PATCH 07/11] net/mlx5: allow LRO in regular Rx queue

[Matan Azrad]

LRO support was only for MPRQ, hence mprq Rx burst was selected when
LRO was configured in the port.

The current support for MPRQ is suffering from bad memory utilization
since an external mempool is allocated by the PMD for the packets data
in addition to the user mempool, besides that, the user may get packet
data addresses which were not configured by him.

Even though MPRQ has the best performance for packet receiving in the
most cases and because of the above facts it is better to remove the
automatic MPRQ select when LRO is configured.

Move MPRQ to be selected only when the user force it by the PMD
arguments including LRO case.

Allow LRO offload using the regular RQ with the regular Rx burst
function.

Signed-off-by: Matan Azrad <matan at mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo at mellanox.com>
---
 drivers/net/mlx5/mlx5.c          |  4 +---
 drivers/net/mlx5/mlx5_ethdev.c   |  6 ------
 drivers/net/mlx5/mlx5_prm.h      |  3 +++
 drivers/net/mlx5/mlx5_rxq.c      | 27 ++++++++++++++-------------
 drivers/net/mlx5/mlx5_rxtx.h     |  4 ++--
 drivers/net/mlx5/mlx5_rxtx_vec.c |  2 ++
 6 files changed, 22 insertions(+), 24 deletions(-)

diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c
index ad0883d..a490bf2 100644
--- a/drivers/net/mlx5/mlx5.c
+++ b/drivers/net/mlx5/mlx5.c
@@ -1856,7 +1856,7 @@ struct mlx5_dev_spawn_data {
 		if (priv->counter_fallback)
 			DRV_LOG(INFO, "Use fall-back DV counter management\n");
 		/* Check for LRO support. */
-		if (config.dest_tir && mprq && config.hca_attr.lro_cap) {
+		if (config.dest_tir && config.hca_attr.lro_cap) {
 			/* TBD check tunnel lro caps. */
 			config.lro.supported = config.hca_attr.lro_cap;
 			DRV_LOG(DEBUG, "Device supports LRO");
@@ -1869,8 +1869,6 @@ struct mlx5_dev_spawn_data {
 				config.hca_attr.lro_timer_supported_periods[0];
 			DRV_LOG(DEBUG, "LRO session timeout set to %d usec",
 				config.lro.timeout);
-			config.mprq.enabled = 1;
-			DRV_LOG(DEBUG, "Enable MPRQ for LRO use");
 		}
 	}
 	if (config.mprq.enabled && mprq) {
diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
index e627909..9d11831 100644
--- a/drivers/net/mlx5/mlx5_ethdev.c
+++ b/drivers/net/mlx5/mlx5_ethdev.c
@@ -433,12 +433,6 @@ struct ethtool_link_settings {
 			dev->data->port_id, priv->rxqs_n, rxqs_n);
 		priv->rxqs_n = rxqs_n;
 		/*
-		 * WHen using LRO, MPRQ is implicitly enabled.
-		 * Adjust threshold value to ensure MPRQ can be enabled.
-		 */
-		if (lro_on && priv->config.mprq.min_rxqs_num > priv->rxqs_n)
-			priv->config.mprq.min_rxqs_num = priv->rxqs_n;
-		/*
 		 * If the requested number of RX queues is not a power of two,
 		 * use the maximum indirection table size for better balancing.
 		 * The result is always rounded to the next power of two.
diff --git a/drivers/net/mlx5/mlx5_prm.h b/drivers/net/mlx5/mlx5_prm.h
index 0716bbd..6ea6345 100644
--- a/drivers/net/mlx5/mlx5_prm.h
+++ b/drivers/net/mlx5/mlx5_prm.h
@@ -237,6 +237,9 @@
 /* Amount of data bytes after eth data segment. */
 #define MLX5_ESEG_EXTRA_DATA_SIZE 32u
 
+/* The maximum log value of segments per RQ WQE. */
+#define MLX5_MAX_LOG_RQ_SEGS 5u
+
 /* Completion mode. */
 enum mlx5_completion_mode {
 	MLX5_COMP_ONLY_ERR = 0x0,
diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
index 5e54156..ad5b0a9 100644
--- a/drivers/net/mlx5/mlx5_rxq.c
+++ b/drivers/net/mlx5/mlx5_rxq.c
@@ -93,7 +93,6 @@
 
 /**
  * Check whether Multi-Packet RQ is enabled for the device.
- * MPRQ can be enabled explicitly, or implicitly by enabling LRO.
  *
  * @param dev
  *   Pointer to Ethernet device.
@@ -1607,6 +1606,7 @@ struct mlx5_rxq_ctrl *
 	unsigned int max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
 	unsigned int non_scatter_min_mbuf_size = max_rx_pkt_len +
 							RTE_PKTMBUF_HEADROOM;
+	unsigned int max_lro_size = 0;
 
 	if (non_scatter_min_mbuf_size > mb_len && !(offloads &
 						    DEV_RX_OFFLOAD_SCATTER)) {
@@ -1672,8 +1672,9 @@ struct mlx5_rxq_ctrl *
 		tmpl->rxq.strd_headroom_en = strd_headroom_en;
 		tmpl->rxq.mprq_max_memcpy_len = RTE_MIN(mb_len -
 			    RTE_PKTMBUF_HEADROOM, config->mprq.max_memcpy_len);
-		mlx5_max_lro_msg_size_adjust(dev, RTE_MIN(max_rx_pkt_len,
-		   (1u << tmpl->rxq.strd_num_n) * (1u << tmpl->rxq.strd_sz_n)));
+		max_lro_size = RTE_MIN(max_rx_pkt_len,
+				       (1u << tmpl->rxq.strd_num_n) *
+				       (1u << tmpl->rxq.strd_sz_n));
 		DRV_LOG(DEBUG,
 			"port %u Rx queue %u: Multi-Packet RQ is enabled"
 			" strd_num_n = %u, strd_sz_n = %u",
@@ -1681,6 +1682,7 @@ struct mlx5_rxq_ctrl *
 			tmpl->rxq.strd_num_n, tmpl->rxq.strd_sz_n);
 	} else if (max_rx_pkt_len <= (mb_len - RTE_PKTMBUF_HEADROOM)) {
 		tmpl->rxq.sges_n = 0;
+		max_lro_size = max_rx_pkt_len;
 	} else if (offloads & DEV_RX_OFFLOAD_SCATTER) {
 		unsigned int size = non_scatter_min_mbuf_size;
 		unsigned int sges_n;
@@ -1690,20 +1692,18 @@ struct mlx5_rxq_ctrl *
 		 * and round it to the next power of two.
 		 */
 		sges_n = log2above((size / mb_len) + !!(size % mb_len));
-		tmpl->rxq.sges_n = sges_n;
-		/* Make sure rxq.sges_n did not overflow. */
-		size = mb_len * (1 << tmpl->rxq.sges_n);
-		size -= RTE_PKTMBUF_HEADROOM;
-		if (size < max_rx_pkt_len) {
+		if (sges_n > MLX5_MAX_LOG_RQ_SEGS) {
 			DRV_LOG(ERR,
 				"port %u too many SGEs (%u) needed to handle"
-				" requested maximum packet size %u",
-				dev->data->port_id,
-				1 << sges_n,
-				max_rx_pkt_len);
-			rte_errno = EOVERFLOW;
+				" requested maximum packet size %u, the maximum"
+				" supported are %u", dev->data->port_id,
+				1 << sges_n, max_rx_pkt_len,
+				1u << MLX5_MAX_LOG_RQ_SEGS);
+			rte_errno = ENOTSUP;
 			goto error;
 		}
+		tmpl->rxq.sges_n = sges_n;
+		max_lro_size = max_rx_pkt_len;
 	}
 	if (mprq_en && !mlx5_rxq_mprq_enabled(&tmpl->rxq))
 		DRV_LOG(WARNING,
@@ -1725,6 +1725,7 @@ struct mlx5_rxq_ctrl *
 		rte_errno = EINVAL;
 		goto error;
 	}
+	mlx5_max_lro_msg_size_adjust(dev, max_lro_size);
 	/* Toggle RX checksum offload if hardware supports it. */
 	tmpl->rxq.csum = !!(offloads & DEV_RX_OFFLOAD_CHECKSUM);
 	tmpl->rxq.hw_timestamp = !!(offloads & DEV_RX_OFFLOAD_TIMESTAMP);
diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
index 60d871c..5704d0a 100644
--- a/drivers/net/mlx5/mlx5_rxtx.h
+++ b/drivers/net/mlx5/mlx5_rxtx.h
@@ -105,7 +105,7 @@ struct mlx5_rxq_data {
 	unsigned int hw_timestamp:1; /* Enable HW timestamp. */
 	unsigned int vlan_strip:1; /* Enable VLAN stripping. */
 	unsigned int crc_present:1; /* CRC must be subtracted. */
-	unsigned int sges_n:2; /* Log 2 of SGEs (max buffers per packet). */
+	unsigned int sges_n:3; /* Log 2 of SGEs (max buffers per packet). */
 	unsigned int cqe_n:4; /* Log 2 of CQ elements. */
 	unsigned int elts_n:4; /* Log 2 of Mbufs. */
 	unsigned int rss_hash:1; /* RSS hash result is enabled. */
@@ -115,7 +115,7 @@ struct mlx5_rxq_data {
 	unsigned int strd_shift_en:1; /* Enable 2bytes shift on a stride. */
 	unsigned int err_state:2; /* enum mlx5_rxq_err_state. */
 	unsigned int strd_headroom_en:1; /* Enable mbuf headroom in MPRQ. */
-	unsigned int :3; /* Remaining bits. */
+	unsigned int :2; /* Remaining bits. */
 	volatile uint32_t *rq_db;
 	volatile uint32_t *cq_db;
 	uint16_t port_id;
diff --git a/drivers/net/mlx5/mlx5_rxtx_vec.c b/drivers/net/mlx5/mlx5_rxtx_vec.c
index f6ec828..3815ff6 100644
--- a/drivers/net/mlx5/mlx5_rxtx_vec.c
+++ b/drivers/net/mlx5/mlx5_rxtx_vec.c
@@ -151,6 +151,8 @@ int __attribute__((cold))
 		return -ENOTSUP;
 	if (mlx5_mprq_enabled(dev))
 		return -ENOTSUP;
+	if (mlx5_lro_on(dev))
+		return -ENOTSUP;
 	/* All the configured queues should support. */
 	for (i = 0; i < priv->rxqs_n; ++i) {
 		struct mlx5_rxq_data *rxq = (*priv->rxqs)[i];
-- 
1.8.3.1