[pull request][net-next 00/17] Mellanox, mlx5e updates 2018-06-01

[pull request][net-next 00/17] Mellanox, mlx5e updates 2018-06-01

[Saeed Mahameed]

Hi dave,

Sorry for the extra 2 patches in this series, but mostly the series
contains small patches and some fixes to previous patches in this
submission window, with one main patch from Tariq to improve legacy
RQ buffer management, for more information please refer to that tag
log below.

Please pull and let me know if there's any problem.

Thanks,
Saeed.

---

The following changes since commit 21ad1173589ef63a93f94e05c879393e2c27588c:

  Merge branch 'hns3-next' (2018-06-01 14:23:59 -0400)

are available in the Git repository at:

  git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux.git tags/mlx5e-updates-2018-06-01

for you to fetch changes up to f65a59ffbcc26135e225058b2e6cd49ab9f9f13f:

  net/mlx5e: TX, Separate cachelines of xmit and completion stats (2018-06-01 16:48:15 -0700)

----------------------------------------------------------------
mlx5e-updates-2018-06-01

1) From Tariq, Two patches to Fix IPoIB issues introduced in
   "net/mlx5e: TX, Use actual WQE size for SQ edge fill"

2) From Eran, Additional improvements to mlx5e statistics reporting

3) From Maor, Increase aRFS flow tables size

4) From Adi, Support MTU change for ethernet representors

5) From Ilan and Adi, Handle QP error events in FPGA

6) From Tariq, last 10 patches mainly deals with RX buffer scheme improvements for legacy RQ
   to use only order-0 pages and fragmented SKBs for large MTUs.

-  Tariq starts with some refactoring and removing HW LRO support from traditional
   (legacy) RQ, since it complicates the buffer scheme and removing it makes it smoother
   to move to cyclic descriptor buffer for traditional RQ.

- Use cyclic WQ in legacy RQ, which has many benefits and paves the way for fragmented SKBs
  for large MTUs.

- Enhance legacy Receive Queue memory scheme, such that only order-0 pages are used.
  Whenever possible, prefer using a linear SKB, and build it wrapping the WQE buffer.
  Otherwise (for example, jumbo frames on x86), use non-linear SKB, with as many frags
  as needed. In this case, multiple WQE scatter entries are used, up to a maximum of 4
  frags and 10KB of MTU.

- TX statistics access improvements.

----------------------------------------------------------------
Adi Nissim (1):
      net/mlx5e: Support configurable MTU for vport representors

Eran Ben Elisha (2):
      net/mlx5e: Present SW stats when state is not opened
      net/mlx5e: Remove redundant active_channels indication

Ilan Tayari (1):
      net/mlx5: FPGA, Handle QP error event

Maor Gottlieb (1):
      net/mlx5e: Increase aRFS flow tables size

Tariq Toukan (12):
      net/mlx5e: IPOIB, Fix overflowing SQ WQE memset
      net/mlx5e: IPOIB, Add a missing skb_pull
      net/mlx5e: TX, Obsolete maintaining local copies of skb->len/data
      net/mlx5e: RX, Generalise name of non-linear SKB head size
      net/mlx5e: RX, Generalise function of SKB frag addition
      net/mlx5e: RX, Dedicate a function for copying SKB header
      net/mlx5e: RX, Remove HW LRO support in legacy RQ
      net/mlx5e: RX, Split WQ objects for different RQ types
      net/mlx5e: RX, Use cyclic WQ in legacy RQ
      net/mlx5e: RX, Enhance legacy Receive Queue memory scheme
      net/mlx5e: RX, Always prefer Linear SKB configuration
      net/mlx5e: TX, Separate cachelines of xmit and completion stats

 drivers/net/ethernet/mellanox/mlx5/core/en.h       |  62 ++-
 drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c  |   2 +-
 .../net/ethernet/mellanox/mlx5/core/en_ethtool.c   |   7 +
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c  | 469 +++++++++++++++------
 drivers/net/ethernet/mellanox/mlx5/core/en_rep.c   |  22 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_rx.c    | 359 ++++++++++------
 .../net/ethernet/mellanox/mlx5/core/en_selftest.c  |   2 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_stats.c |  23 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_stats.h |   9 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_tx.c    |  47 +--
 .../net/ethernet/mellanox/mlx5/core/fpga/core.c    |  28 +-
 drivers/net/ethernet/mellanox/mlx5/core/wq.c       |   1 +
 drivers/net/ethernet/mellanox/mlx5/core/wq.h       |  55 ++-
 13 files changed, 726 insertions(+), 360 deletions(-)

[net-next 01/17] net/mlx5e: IPOIB, Fix overflowing SQ WQE memset

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

IPoIB WQE size is larger than a single WQEBB.  Must not fetch the WQE,
and surely not memset it, until it is guaranteed that there are enough
WQEBBs available before getting to SQ/frag edge.

Fixes: 043dc78ecf07 ("net/mlx5e: TX, Use actual WQE size for SQ edge fill")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en_tx.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
index aafd75257fd0..9829ee02de31 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
@@ -653,8 +653,6 @@ netdev_tx_t mlx5i_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 	int num_dma;
 	__be16 mss;
 
-	mlx5i_sq_fetch_wqe(sq, &wqe, &pi);
-
 	/* Calc ihs and ds cnt, no writes to wqe yet */
 	ds_cnt = sizeof(*wqe) / MLX5_SEND_WQE_DS;
 	if (skb_is_gso(skb)) {
@@ -686,10 +684,12 @@ netdev_tx_t mlx5i_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 	num_wqebbs = DIV_ROUND_UP(ds_cnt, MLX5_SEND_WQEBB_NUM_DS);
 	frag_pi = mlx5_wq_cyc_ctr2fragix(wq, sq->pc);
 	if (unlikely(frag_pi + num_wqebbs > mlx5_wq_cyc_get_frag_size(wq))) {
+		pi = mlx5_wq_cyc_ctr2ix(wq, sq->pc);
 		mlx5e_fill_sq_frag_edge(sq, wq, pi, frag_pi);
-		mlx5i_sq_fetch_wqe(sq, &wqe, &pi);
 	}
 
+	mlx5i_sq_fetch_wqe(sq, &wqe, &pi);
+
 	/* fill wqe */
 	wi       = &sq->db.wqe_info[pi];
 	cseg     = &wqe->ctrl;
-- 
2.17.0

[net-next 02/17] net/mlx5e: IPOIB, Add a missing skb_pull

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

A call to mlx5e_tx_skb_pull_inline was mistakenly dropped
in the cited patch. Get it back.

Fixes: 043dc78ecf07 ("net/mlx5e: TX, Use actual WQE size for SQ edge fill")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en_tx.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
index 9829ee02de31..725c06221e95 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
@@ -705,6 +705,7 @@ netdev_tx_t mlx5i_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 
 	if (ihs) {
 		memcpy(eseg->inline_hdr.start, skb_data, ihs);
+		mlx5e_tx_skb_pull_inline(&skb_data, &skb_len, ihs);
 		eseg->inline_hdr.sz = cpu_to_be16(ihs);
 		dseg += ds_cnt_inl;
 	}
-- 
2.17.0

[net-next 03/17] net/mlx5e: Present SW stats when state is not opened

[Saeed Mahameed]

From: Eran Ben Elisha <eranbe@mellanox.com>

The driver can present all SW stats even when the state not opened.
Fixed get strings, count and stats to support it.

In addition, fix tc2txq to hold a static mapping which doesn't depend on
the amount of open channels, and cannot have the same value on two
different cells  while moving between configurations.
Example:
- OOB 16 channels
- Change to 2 channels, 8 TCs
- tc2txq[15][0] == tc2txq[1][7] == 15
This will cause multiple appearances of the same TX index in statistics
output.

Fixes: 76c3810bade3 ("net/mlx5e: Avoid reset netdev stats on configuration changes")
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 .../net/ethernet/mellanox/mlx5/core/en_main.c | 19 +++++++++++++------
 .../ethernet/mellanox/mlx5/core/en_stats.c    | 10 ----------
 2 files changed, 13 insertions(+), 16 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 9b19863b059d..adc55de6d4f4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -2633,15 +2633,21 @@ static void mlx5e_netdev_set_tcs(struct net_device *netdev)
 		netdev_set_tc_queue(netdev, tc, nch, 0);
 }
 
-static void mlx5e_build_channels_tx_maps(struct mlx5e_priv *priv)
+static void mlx5e_build_tc2txq_maps(struct mlx5e_priv *priv)
 {
-	struct mlx5e_channel *c;
-	struct mlx5e_txqsq *sq;
+	int max_nch = priv->profile->max_nch(priv->mdev);
 	int i, tc;
 
-	for (i = 0; i < priv->profile->max_nch(priv->mdev); i++)
+	for (i = 0; i < max_nch; i++)
 		for (tc = 0; tc < priv->profile->max_tc; tc++)
-			priv->channel_tc2txq[i][tc] = i + tc * priv->channels.num;
+			priv->channel_tc2txq[i][tc] = i + tc * max_nch;
+}
+
+static void mlx5e_build_tx2sq_maps(struct mlx5e_priv *priv)
+{
+	struct mlx5e_channel *c;
+	struct mlx5e_txqsq *sq;
+	int i, tc;
 
 	for (i = 0; i < priv->channels.num; i++) {
 		c = priv->channels.c[i];
@@ -2661,7 +2667,7 @@ void mlx5e_activate_priv_channels(struct mlx5e_priv *priv)
 	netif_set_real_num_tx_queues(netdev, num_txqs);
 	netif_set_real_num_rx_queues(netdev, priv->channels.num);
 
-	mlx5e_build_channels_tx_maps(priv);
+	mlx5e_build_tx2sq_maps(priv);
 	mlx5e_activate_channels(&priv->channels);
 	write_lock(&priv->stats_lock);
 	priv->channels_active = true;
@@ -4446,6 +4452,7 @@ static void mlx5e_nic_init(struct mlx5_core_dev *mdev,
 	if (err)
 		mlx5_core_err(mdev, "TLS initialization failed, %d\n", err);
 	mlx5e_build_nic_netdev(netdev);
+	mlx5e_build_tc2txq_maps(priv);
 	mlx5e_vxlan_init(priv);
 }
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c b/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
index 776b4d68e156..3b2aed43f660 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
@@ -1161,9 +1161,6 @@ static int mlx5e_grp_channels_get_num_stats(struct mlx5e_priv *priv)
 {
 	int max_nch = priv->profile->max_nch(priv->mdev);
 
-	if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
-		return 0;
-
 	return (NUM_RQ_STATS * max_nch) +
 	       (NUM_CH_STATS * max_nch) +
 	       (NUM_SQ_STATS * max_nch * priv->max_opened_tc);
@@ -1175,9 +1172,6 @@ static int mlx5e_grp_channels_fill_strings(struct mlx5e_priv *priv, u8 *data,
 	int max_nch = priv->profile->max_nch(priv->mdev);
 	int i, j, tc;
 
-	if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
-		return idx;
-
 	for (i = 0; i < max_nch; i++)
 		for (j = 0; j < NUM_CH_STATS; j++)
 			sprintf(data + (idx++) * ETH_GSTRING_LEN,
@@ -1187,7 +1181,6 @@ static int mlx5e_grp_channels_fill_strings(struct mlx5e_priv *priv, u8 *data,
 		for (j = 0; j < NUM_RQ_STATS; j++)
 			sprintf(data + (idx++) * ETH_GSTRING_LEN, rq_stats_desc[j].format, i);
 
-	/* priv->channel_tc2txq[i][tc] is valid only when device is open */
 	for (tc = 0; tc < priv->max_opened_tc; tc++)
 		for (i = 0; i < max_nch; i++)
 			for (j = 0; j < NUM_SQ_STATS; j++)
@@ -1204,9 +1197,6 @@ static int mlx5e_grp_channels_fill_stats(struct mlx5e_priv *priv, u64 *data,
 	int max_nch = priv->profile->max_nch(priv->mdev);
 	int i, j, tc;
 
-	if (!test_bit(MLX5E_STATE_OPENED, &priv->state))
-		return idx;
-
 	for (i = 0; i < max_nch; i++)
 		for (j = 0; j < NUM_CH_STATS; j++)
 			data[idx++] =
-- 
2.17.0

[net-next 04/17] net/mlx5e: Remove redundant active_channels indication

[Saeed Mahameed]

From: Eran Ben Elisha <eranbe@mellanox.com>

Now, when all channels stats are saved regardless of the channel's state
{open, closed}, we can safely remove this indication and the stats spin
lock which protects it.

Fixes: 76c3810bade3 ("net/mlx5e: Avoid reset netdev stats on configuration changes")
Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h       | 2 --
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c  | 7 -------
 drivers/net/ethernet/mellanox/mlx5/core/en_rep.c   | 6 ------
 drivers/net/ethernet/mellanox/mlx5/core/en_stats.c | 5 -----
 4 files changed, 20 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 1c04df043e07..372cdf8a496c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -777,8 +777,6 @@ struct mlx5e_priv {
 	struct mutex               state_lock; /* Protects Interface state */
 	struct mlx5e_rq            drop_rq;
 
-	rwlock_t                   stats_lock; /* Protects channels SW stats updates */
-	bool                       channels_active;
 	struct mlx5e_channels      channels;
 	u32                        tisn[MLX5E_MAX_NUM_TC];
 	struct mlx5e_rqt           indir_rqt;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index adc55de6d4f4..dd119bad102d 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -2669,9 +2669,6 @@ void mlx5e_activate_priv_channels(struct mlx5e_priv *priv)
 
 	mlx5e_build_tx2sq_maps(priv);
 	mlx5e_activate_channels(&priv->channels);
-	write_lock(&priv->stats_lock);
-	priv->channels_active = true;
-	write_unlock(&priv->stats_lock);
 	netif_tx_start_all_queues(priv->netdev);
 
 	if (MLX5_VPORT_MANAGER(priv->mdev))
@@ -2693,9 +2690,6 @@ void mlx5e_deactivate_priv_channels(struct mlx5e_priv *priv)
 	 */
 	netif_tx_stop_all_queues(priv->netdev);
 	netif_tx_disable(priv->netdev);
-	write_lock(&priv->stats_lock);
-	priv->channels_active = false;
-	write_unlock(&priv->stats_lock);
 	mlx5e_deactivate_channels(&priv->channels);
 }
 
@@ -4269,7 +4263,6 @@ static void mlx5e_build_nic_netdev_priv(struct mlx5_core_dev *mdev,
 			       profile->max_nch(mdev), netdev->mtu);
 
 	mutex_init(&priv->state_lock);
-	rwlock_init(&priv->stats_lock);
 
 	INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work);
 	INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
index de6364125f0f..1fb4835eac72 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
@@ -130,10 +130,6 @@ static void mlx5e_rep_update_sw_counters(struct mlx5e_priv *priv)
 	struct mlx5e_sq_stats *sq_stats;
 	int i, j;
 
-	read_lock(&priv->stats_lock);
-	if (!priv->channels_active)
-	        goto out;
-
 	memset(s, 0, sizeof(*s));
 	for (i = 0; i < priv->channels.num; i++) {
 		struct mlx5e_channel *c = priv->channels.c[i];
@@ -150,8 +146,6 @@ static void mlx5e_rep_update_sw_counters(struct mlx5e_priv *priv)
 			s->tx_bytes		+= sq_stats->bytes;
 		}
 	}
-out:
-	read_unlock(&priv->stats_lock);
 }
 
 static void mlx5e_rep_get_ethtool_stats(struct net_device *dev,
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c b/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
index 3b2aed43f660..697dc7397ba2 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
@@ -114,9 +114,6 @@ void mlx5e_grp_sw_update_stats(struct mlx5e_priv *priv)
 	int i;
 
 	memset(s, 0, sizeof(*s));
-	read_lock(&priv->stats_lock);
-	if (!priv->channels_active)
-		goto out;
 
 	for (i = 0; i < priv->profile->max_nch(priv->mdev); i++) {
 		struct mlx5e_channel_stats *channel_stats =
@@ -177,8 +174,6 @@ void mlx5e_grp_sw_update_stats(struct mlx5e_priv *priv)
 	}
 
 	memcpy(&priv->stats.sw, s, sizeof(*s));
-out:
-	read_unlock(&priv->stats_lock);
 }
 
 static const struct counter_desc q_stats_desc[] = {
-- 
2.17.0

[net-next 05/17] net/mlx5e: Increase aRFS flow tables size

[Saeed Mahameed]

From: Maor Gottlieb <maorg@mellanox.com>

Increase the aRFS flow table size to 64k so it could contain up to 64k
different streams.

Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c b/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c
index f64b5e78519b..75e4308ba786 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_arfs.c
@@ -213,7 +213,7 @@ static int arfs_add_default_rule(struct mlx5e_priv *priv,
 }
 
 #define MLX5E_ARFS_NUM_GROUPS	2
-#define MLX5E_ARFS_GROUP1_SIZE	BIT(12)
+#define MLX5E_ARFS_GROUP1_SIZE	(BIT(16) - 1)
 #define MLX5E_ARFS_GROUP2_SIZE	BIT(0)
 #define MLX5E_ARFS_TABLE_SIZE	(MLX5E_ARFS_GROUP1_SIZE +\
 				 MLX5E_ARFS_GROUP2_SIZE)
-- 
2.17.0

[net-next 06/17] net/mlx5e: Support configurable MTU for vport representors

[Saeed Mahameed]

From: Adi Nissim <adin@mellanox.com>

The representor MTU was hard coded to 1500 bytes.
Allow setting arbitrary MTU values up to the max supported by the FW.

Signed-off-by: Adi Nissim <adin@mellanox.com>
Reviewed-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h      |  4 ++++
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 14 ++++++++++----
 drivers/net/ethernet/mellanox/mlx5/core/en_rep.c  | 14 ++++++++++++++
 3 files changed, 28 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 372cdf8a496c..a23fa47fbbcc 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -1102,6 +1102,10 @@ void mlx5e_update_stats_work(struct work_struct *work);
 
 int mlx5e_bits_invert(unsigned long a, int size);
 
+typedef int (*change_hw_mtu_cb)(struct mlx5e_priv *priv);
+int mlx5e_change_mtu(struct net_device *netdev, int new_mtu,
+		     change_hw_mtu_cb set_mtu_cb);
+
 /* ethtool helpers */
 void mlx5e_ethtool_get_drvinfo(struct mlx5e_priv *priv,
 			       struct ethtool_drvinfo *drvinfo);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index dd119bad102d..ab7b2a4e6edc 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -3495,7 +3495,8 @@ static netdev_features_t mlx5e_fix_features(struct net_device *netdev,
 	return features;
 }
 
-static int mlx5e_change_mtu(struct net_device *netdev, int new_mtu)
+int mlx5e_change_mtu(struct net_device *netdev, int new_mtu,
+		     change_hw_mtu_cb set_mtu_cb)
 {
 	struct mlx5e_priv *priv = netdev_priv(netdev);
 	struct mlx5e_channels new_channels = {};
@@ -3522,7 +3523,7 @@ static int mlx5e_change_mtu(struct net_device *netdev, int new_mtu)
 
 	if (!reset) {
 		params->sw_mtu = new_mtu;
-		mlx5e_set_dev_port_mtu(priv);
+		set_mtu_cb(priv);
 		netdev->mtu = params->sw_mtu;
 		goto out;
 	}
@@ -3531,7 +3532,7 @@ static int mlx5e_change_mtu(struct net_device *netdev, int new_mtu)
 	if (err)
 		goto out;
 
-	mlx5e_switch_priv_channels(priv, &new_channels, mlx5e_set_dev_port_mtu);
+	mlx5e_switch_priv_channels(priv, &new_channels, set_mtu_cb);
 	netdev->mtu = new_channels.params.sw_mtu;
 
 out:
@@ -3539,6 +3540,11 @@ static int mlx5e_change_mtu(struct net_device *netdev, int new_mtu)
 	return err;
 }
 
+static int mlx5e_change_nic_mtu(struct net_device *netdev, int new_mtu)
+{
+	return mlx5e_change_mtu(netdev, new_mtu, mlx5e_set_dev_port_mtu);
+}
+
 int mlx5e_hwstamp_set(struct mlx5e_priv *priv, struct ifreq *ifr)
 {
 	struct hwtstamp_config config;
@@ -4033,7 +4039,7 @@ static const struct net_device_ops mlx5e_netdev_ops = {
 	.ndo_vlan_rx_kill_vid    = mlx5e_vlan_rx_kill_vid,
 	.ndo_set_features        = mlx5e_set_features,
 	.ndo_fix_features        = mlx5e_fix_features,
-	.ndo_change_mtu          = mlx5e_change_mtu,
+	.ndo_change_mtu          = mlx5e_change_nic_mtu,
 	.ndo_do_ioctl            = mlx5e_ioctl,
 	.ndo_set_tx_maxrate      = mlx5e_set_tx_maxrate,
 	.ndo_udp_tunnel_add      = mlx5e_add_vxlan_port,
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
index 1fb4835eac72..8ab4c96b7f7c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
@@ -900,6 +900,11 @@ static const struct switchdev_ops mlx5e_rep_switchdev_ops = {
 	.switchdev_port_attr_get	= mlx5e_attr_get,
 };
 
+int mlx5e_change_rep_mtu(struct net_device *netdev, int new_mtu)
+{
+	return mlx5e_change_mtu(netdev, new_mtu, NULL);
+}
+
 static const struct net_device_ops mlx5e_netdev_ops_rep = {
 	.ndo_open                = mlx5e_rep_open,
 	.ndo_stop                = mlx5e_rep_close,
@@ -909,6 +914,7 @@ static const struct net_device_ops mlx5e_netdev_ops_rep = {
 	.ndo_get_stats64         = mlx5e_rep_get_stats,
 	.ndo_has_offload_stats	 = mlx5e_has_offload_stats,
 	.ndo_get_offload_stats	 = mlx5e_get_offload_stats,
+	.ndo_change_mtu          = mlx5e_change_rep_mtu,
 };
 
 static void mlx5e_build_rep_params(struct mlx5_core_dev *mdev,
@@ -935,6 +941,10 @@ static void mlx5e_build_rep_params(struct mlx5_core_dev *mdev,
 
 static void mlx5e_build_rep_netdev(struct net_device *netdev)
 {
+	struct mlx5e_priv *priv = netdev_priv(netdev);
+	struct mlx5_core_dev *mdev = priv->mdev;
+	u16 max_mtu;
+
 	netdev->netdev_ops = &mlx5e_netdev_ops_rep;
 
 	netdev->watchdog_timeo    = 15 * HZ;
@@ -947,6 +957,10 @@ static void mlx5e_build_rep_netdev(struct net_device *netdev)
 	netdev->hw_features      |= NETIF_F_HW_TC;
 
 	eth_hw_addr_random(netdev);
+
+	netdev->min_mtu = ETH_MIN_MTU;
+	mlx5_query_port_max_mtu(mdev, &max_mtu, 1);
+	netdev->max_mtu = MLX5E_HW2SW_MTU(&priv->channels.params, max_mtu);
 }
 
 static void mlx5e_init_rep(struct mlx5_core_dev *mdev,
-- 
2.17.0

[net-next 07/17] net/mlx5: FPGA, Handle QP error event

[Saeed Mahameed]

From: Ilan Tayari <ilant@mellanox.com>

Add handlers for this event to perform graceful teardown of the device.

Signed-off-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Adi Nissim <adin@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 .../ethernet/mellanox/mlx5/core/fpga/core.c   | 28 ++++++++++++++++---
 1 file changed, 24 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fpga/core.c b/drivers/net/ethernet/mellanox/mlx5/core/fpga/core.c
index 26caa0475985..436a8136f26f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/fpga/core.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/fpga/core.c
@@ -50,6 +50,11 @@ static const char *const mlx5_fpga_error_strings[] = {
 	"Temperature Critical",
 };
 
+static const char * const mlx5_fpga_qp_error_strings[] = {
+	"Null Syndrome",
+	"Retry Counter Expired",
+	"RNR Expired",
+};
 static struct mlx5_fpga_device *mlx5_fpga_device_alloc(void)
 {
 	struct mlx5_fpga_device *fdev = NULL;
@@ -271,23 +276,38 @@ static const char *mlx5_fpga_syndrome_to_string(u8 syndrome)
 	return "Unknown";
 }
 
+static const char *mlx5_fpga_qp_syndrome_to_string(u8 syndrome)
+{
+	if (syndrome < ARRAY_SIZE(mlx5_fpga_qp_error_strings))
+		return mlx5_fpga_qp_error_strings[syndrome];
+	return "Unknown";
+}
+
 void mlx5_fpga_event(struct mlx5_core_dev *mdev, u8 event, void *data)
 {
 	struct mlx5_fpga_device *fdev = mdev->fpga;
 	const char *event_name;
 	bool teardown = false;
 	unsigned long flags;
+	u32 fpga_qpn;
 	u8 syndrome;
 
-	if (event != MLX5_EVENT_TYPE_FPGA_ERROR) {
+	switch (event) {
+	case MLX5_EVENT_TYPE_FPGA_ERROR:
+		syndrome = MLX5_GET(fpga_error_event, data, syndrome);
+		event_name = mlx5_fpga_syndrome_to_string(syndrome);
+		break;
+	case MLX5_EVENT_TYPE_FPGA_QP_ERROR:
+		syndrome = MLX5_GET(fpga_qp_error_event, data, syndrome);
+		event_name = mlx5_fpga_qp_syndrome_to_string(syndrome);
+		fpga_qpn = MLX5_GET(fpga_qp_error_event, data, fpga_qpn);
+		break;
+	default:
 		mlx5_fpga_warn_ratelimited(fdev, "Unexpected event %u\n",
 					   event);
 		return;
 	}
 
-	syndrome = MLX5_GET(fpga_error_event, data, syndrome);
-	event_name = mlx5_fpga_syndrome_to_string(syndrome);
-
 	spin_lock_irqsave(&fdev->state_lock, flags);
 	switch (fdev->state) {
 	case MLX5_FPGA_STATUS_SUCCESS:
-- 
2.17.0

[net-next 08/17] net/mlx5e: TX, Obsolete maintaining local copies of skb->len/data

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

Instead of maintaining a local copy of skb->len/data and updating
it upon every copy to the WQE inline part, just calculate it once
when needed, using the ihs.

This obsoletes the function mlx5e_tx_skb_pull_inline.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 .../net/ethernet/mellanox/mlx5/core/en_tx.c   | 42 ++++++-------------
 1 file changed, 12 insertions(+), 30 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
index 725c06221e95..f29deb44bf3b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tx.c
@@ -188,28 +188,16 @@ static inline u16 mlx5e_calc_min_inline(enum mlx5_inline_modes mode,
 	return min_t(u16, hlen, skb_headlen(skb));
 }
 
-static inline void mlx5e_tx_skb_pull_inline(unsigned char **skb_data,
-					    unsigned int *skb_len,
-					    unsigned int len)
-{
-	*skb_len -= len;
-	*skb_data += len;
-}
-
-static inline void mlx5e_insert_vlan(void *start, struct sk_buff *skb, u16 ihs,
-				     unsigned char **skb_data,
-				     unsigned int *skb_len)
+static inline void mlx5e_insert_vlan(void *start, struct sk_buff *skb, u16 ihs)
 {
 	struct vlan_ethhdr *vhdr = (struct vlan_ethhdr *)start;
 	int cpy1_sz = 2 * ETH_ALEN;
 	int cpy2_sz = ihs - cpy1_sz;
 
-	memcpy(vhdr, *skb_data, cpy1_sz);
-	mlx5e_tx_skb_pull_inline(skb_data, skb_len, cpy1_sz);
+	memcpy(vhdr, skb->data, cpy1_sz);
 	vhdr->h_vlan_proto = skb->vlan_proto;
 	vhdr->h_vlan_TCI = cpu_to_be16(skb_vlan_tag_get(skb));
-	memcpy(&vhdr->h_vlan_encapsulated_proto, *skb_data, cpy2_sz);
-	mlx5e_tx_skb_pull_inline(skb_data, skb_len, cpy2_sz);
+	memcpy(&vhdr->h_vlan_encapsulated_proto, skb->data + cpy1_sz, cpy2_sz);
 }
 
 static inline void
@@ -357,8 +345,6 @@ netdev_tx_t mlx5e_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 	struct mlx5e_tx_wqe_info *wi;
 
 	struct mlx5e_sq_stats *stats = sq->stats;
-	unsigned char *skb_data = skb->data;
-	unsigned int skb_len = skb->len;
 	u16 ds_cnt, ds_cnt_inl = 0;
 	u16 headlen, ihs, frag_pi;
 	u8 num_wqebbs, opcode;
@@ -385,7 +371,7 @@ netdev_tx_t mlx5e_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 	stats->bytes     += num_bytes;
 	stats->xmit_more += skb->xmit_more;
 
-	headlen = skb_len - ihs - skb->data_len;
+	headlen = skb->len - ihs - skb->data_len;
 	ds_cnt += !!headlen;
 	ds_cnt += skb_shinfo(skb)->nr_frags;
 
@@ -414,15 +400,14 @@ netdev_tx_t mlx5e_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 	eseg->mss = mss;
 
 	if (ihs) {
+		eseg->inline_hdr.sz = cpu_to_be16(ihs);
 		if (skb_vlan_tag_present(skb)) {
-			mlx5e_insert_vlan(eseg->inline_hdr.start, skb,
-					  ihs - VLAN_HLEN, &skb_data, &skb_len);
+			ihs -= VLAN_HLEN;
+			mlx5e_insert_vlan(eseg->inline_hdr.start, skb, ihs);
 			stats->added_vlan_packets++;
 		} else {
-			memcpy(eseg->inline_hdr.start, skb_data, ihs);
-			mlx5e_tx_skb_pull_inline(&skb_data, &skb_len, ihs);
+			memcpy(eseg->inline_hdr.start, skb->data, ihs);
 		}
-		eseg->inline_hdr.sz = cpu_to_be16(ihs);
 		dseg += ds_cnt_inl;
 	} else if (skb_vlan_tag_present(skb)) {
 		eseg->insert.type = cpu_to_be16(MLX5_ETH_WQE_INSERT_VLAN);
@@ -432,7 +417,7 @@ netdev_tx_t mlx5e_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 		stats->added_vlan_packets++;
 	}
 
-	num_dma = mlx5e_txwqe_build_dsegs(sq, skb, skb_data, headlen, dseg);
+	num_dma = mlx5e_txwqe_build_dsegs(sq, skb, skb->data + ihs, headlen, dseg);
 	if (unlikely(num_dma < 0))
 		goto err_drop;
 
@@ -644,8 +629,6 @@ netdev_tx_t mlx5i_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 	struct mlx5e_tx_wqe_info *wi;
 
 	struct mlx5e_sq_stats *stats = sq->stats;
-	unsigned char *skb_data = skb->data;
-	unsigned int skb_len = skb->len;
 	u16 headlen, ihs, pi, frag_pi;
 	u16 ds_cnt, ds_cnt_inl = 0;
 	u8 num_wqebbs, opcode;
@@ -672,7 +655,7 @@ netdev_tx_t mlx5i_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 	stats->bytes     += num_bytes;
 	stats->xmit_more += skb->xmit_more;
 
-	headlen = skb_len - ihs - skb->data_len;
+	headlen = skb->len - ihs - skb->data_len;
 	ds_cnt += !!headlen;
 	ds_cnt += skb_shinfo(skb)->nr_frags;
 
@@ -704,13 +687,12 @@ netdev_tx_t mlx5i_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 	eseg->mss = mss;
 
 	if (ihs) {
-		memcpy(eseg->inline_hdr.start, skb_data, ihs);
-		mlx5e_tx_skb_pull_inline(&skb_data, &skb_len, ihs);
+		memcpy(eseg->inline_hdr.start, skb->data, ihs);
 		eseg->inline_hdr.sz = cpu_to_be16(ihs);
 		dseg += ds_cnt_inl;
 	}
 
-	num_dma = mlx5e_txwqe_build_dsegs(sq, skb, skb_data, headlen, dseg);
+	num_dma = mlx5e_txwqe_build_dsegs(sq, skb, skb->data + ihs, headlen, dseg);
 	if (unlikely(num_dma < 0))
 		goto err_drop;
 
-- 
2.17.0

[net-next 09/17] net/mlx5e: RX, Generalise name of non-linear SKB head size

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

Make name more generic by dropping MPWRQ from it, as it will be
used also in Legacy RQ in a downstream patch.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h          | 2 +-
 drivers/net/ethernet/mellanox/mlx5/core/en_rx.c       | 4 ++--
 drivers/net/ethernet/mellanox/mlx5/core/en_selftest.c | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index a23fa47fbbcc..9b4ed83783e4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -112,7 +112,7 @@ struct page_pool;
 
 #define MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE_MPW            0x2
 
-#define MLX5_MPWRQ_SMALL_PACKET_THRESHOLD	(256)
+#define MLX5E_RX_MAX_HEAD (256)
 
 #define MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ                 (64 * 1024)
 #define MLX5E_DEFAULT_LRO_TIMEOUT                       32
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index 5551e4e22ad0..61a76faa2779 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -1070,7 +1070,7 @@ struct sk_buff *
 mlx5e_skb_from_cqe_mpwrq_nonlinear(struct mlx5e_rq *rq, struct mlx5e_mpw_info *wi,
 				   u16 cqe_bcnt, u32 head_offset, u32 page_idx)
 {
-	u16 headlen = min_t(u16, MLX5_MPWRQ_SMALL_PACKET_THRESHOLD, cqe_bcnt);
+	u16 headlen = min_t(u16, MLX5E_RX_MAX_HEAD, cqe_bcnt);
 	struct mlx5e_dma_info *di = &wi->umr.dma_info[page_idx];
 	u32 frag_offset    = head_offset + headlen;
 	u32 byte_cnt       = cqe_bcnt - headlen;
@@ -1078,7 +1078,7 @@ mlx5e_skb_from_cqe_mpwrq_nonlinear(struct mlx5e_rq *rq, struct mlx5e_mpw_info *w
 	struct sk_buff *skb;
 
 	skb = napi_alloc_skb(rq->cq.napi,
-			     ALIGN(MLX5_MPWRQ_SMALL_PACKET_THRESHOLD, sizeof(long)));
+			     ALIGN(MLX5E_RX_MAX_HEAD, sizeof(long)));
 	if (unlikely(!skb)) {
 		rq->stats->buff_alloc_err++;
 		return NULL;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_selftest.c b/drivers/net/ethernet/mellanox/mlx5/core/en_selftest.c
index 027f54ac1ca2..4d316cc9b008 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_selftest.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_selftest.c
@@ -100,7 +100,7 @@ static int mlx5e_test_link_speed(struct mlx5e_priv *priv)
 
 #ifdef CONFIG_INET
 /* loopback test */
-#define MLX5E_TEST_PKT_SIZE (MLX5_MPWRQ_SMALL_PACKET_THRESHOLD - NET_IP_ALIGN)
+#define MLX5E_TEST_PKT_SIZE (MLX5E_RX_MAX_HEAD - NET_IP_ALIGN)
 static const char mlx5e_test_text[ETH_GSTRING_LEN] = "MLX5E SELF TEST";
 #define MLX5E_TEST_MAGIC 0x5AEED15C001ULL
 
-- 
2.17.0

[net-next 10/17] net/mlx5e: RX, Generalise function of SKB frag addition

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

Rename it and pass truesize as an extra argument, as it will be used also
in Legacy RQ in a downstream patch.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index 61a76faa2779..546e3bc46cd1 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -311,13 +311,11 @@ void mlx5e_dealloc_rx_wqe(struct mlx5e_rq *rq, u16 ix)
 		mlx5e_free_rx_wqe(rq, wi);
 }
 
-static inline void mlx5e_add_skb_frag_mpwqe(struct mlx5e_rq *rq,
-					    struct sk_buff *skb,
-					    struct mlx5e_dma_info *di,
-					    u32 frag_offset, u32 len)
+static inline void
+mlx5e_add_skb_frag(struct mlx5e_rq *rq, struct sk_buff *skb,
+		   struct mlx5e_dma_info *di, u32 frag_offset, u32 len,
+		   unsigned int truesize)
 {
-	unsigned int truesize = ALIGN(len, BIT(rq->mpwqe.log_stride_sz));
-
 	dma_sync_single_for_cpu(rq->pdev,
 				di->addr + frag_offset,
 				len, DMA_FROM_DEVICE);
@@ -1094,9 +1092,11 @@ mlx5e_skb_from_cqe_mpwrq_nonlinear(struct mlx5e_rq *rq, struct mlx5e_mpw_info *w
 	while (byte_cnt) {
 		u32 pg_consumed_bytes =
 			min_t(u32, PAGE_SIZE - frag_offset, byte_cnt);
+		unsigned int truesize =
+			ALIGN(pg_consumed_bytes, BIT(rq->mpwqe.log_stride_sz));
 
-		mlx5e_add_skb_frag_mpwqe(rq, skb, di, frag_offset,
-					 pg_consumed_bytes);
+		mlx5e_add_skb_frag(rq, skb, di, frag_offset,
+				   pg_consumed_bytes, truesize);
 		byte_cnt -= pg_consumed_bytes;
 		frag_offset = 0;
 		di++;
-- 
2.17.0

[net-next 11/17] net/mlx5e: RX, Dedicate a function for copying SKB header

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

Get the logic of copying the packet header into the SKB linear part
into a generic function. Function does copy length alignment
and dma buffer sync.

It is currently called only within the MPWQE flow.
In a downstream patch, it will be called within the legacy RQ flow
as well.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 .../net/ethernet/mellanox/mlx5/core/en_rx.c   | 30 +++++++++++--------
 1 file changed, 17 insertions(+), 13 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index 546e3bc46cd1..634540afdcfc 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -324,6 +324,20 @@ mlx5e_add_skb_frag(struct mlx5e_rq *rq, struct sk_buff *skb,
 			di->page, frag_offset, len, truesize);
 }
 
+static inline void
+mlx5e_copy_skb_header(struct device *pdev, struct sk_buff *skb,
+		      struct mlx5e_dma_info *dma_info,
+		      int offset_from, int offset_to, u32 headlen)
+{
+	const void *from = page_address(dma_info->page) + offset_from;
+	/* Aligning len to sizeof(long) optimizes memcpy performance */
+	unsigned int len = ALIGN(headlen, sizeof(long));
+
+	dma_sync_single_for_cpu(pdev, dma_info->addr + offset_from, len,
+				DMA_FROM_DEVICE);
+	skb_copy_to_linear_data_offset(skb, offset_to, from, len);
+}
+
 static inline void
 mlx5e_copy_skb_header_mpwqe(struct device *pdev,
 			    struct sk_buff *skb,
@@ -331,23 +345,13 @@ mlx5e_copy_skb_header_mpwqe(struct device *pdev,
 			    u32 offset, u32 headlen)
 {
 	u16 headlen_pg = min_t(u32, headlen, PAGE_SIZE - offset);
-	unsigned int len;
 
-	 /* Aligning len to sizeof(long) optimizes memcpy performance */
-	len = ALIGN(headlen_pg, sizeof(long));
-	dma_sync_single_for_cpu(pdev, dma_info->addr + offset, len,
-				DMA_FROM_DEVICE);
-	skb_copy_to_linear_data(skb, page_address(dma_info->page) + offset, len);
+	mlx5e_copy_skb_header(pdev, skb, dma_info, offset, 0, headlen_pg);
 
 	if (unlikely(offset + headlen > PAGE_SIZE)) {
 		dma_info++;
-		headlen_pg = len;
-		len = ALIGN(headlen - headlen_pg, sizeof(long));
-		dma_sync_single_for_cpu(pdev, dma_info->addr, len,
-					DMA_FROM_DEVICE);
-		skb_copy_to_linear_data_offset(skb, headlen_pg,
-					       page_address(dma_info->page),
-					       len);
+		mlx5e_copy_skb_header(pdev, skb, dma_info, 0, headlen_pg,
+				      headlen - headlen_pg);
 	}
 }
 
-- 
2.17.0

[net-next 12/17] net/mlx5e: RX, Remove HW LRO support in legacy RQ

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

Current LRO implementation in Legacy RQ uses high-order pages.
In downstream patches of this series we complete the transition
to using only order-0 pages in RX datapath (which was already done
in Striding RQ).

Unlike the more advanced Striding RQ, Legacy RQ does not make reuse
of any non-consumed buffers of non-full LRO sessions, and combining
it with order-0 pages has many performance drawbacks.

Hence, here we totally remove LRO support in Legacy RQ.
This guarantees having no out-of-order completions, which allows using
a cyclic work queue (instead of a linked-list) in a downstream patch.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 .../ethernet/mellanox/mlx5/core/en_ethtool.c  |  7 ++++
 .../net/ethernet/mellanox/mlx5/core/en_main.c | 33 +++++++++++--------
 2 files changed, 26 insertions(+), 14 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
index 42bd256e680d..fffe514ba855 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c
@@ -1515,6 +1515,9 @@ static int set_pflag_rx_striding_rq(struct net_device *netdev, bool enable)
 			return -EOPNOTSUPP;
 		if (!mlx5e_striding_rq_possible(mdev, &priv->channels.params))
 			return -EINVAL;
+	} else if (priv->channels.params.lro_en) {
+		netdev_warn(netdev, "Can't set legacy RQ with LRO, disable LRO first\n");
+		return -EINVAL;
 	}
 
 	new_channels.params = priv->channels.params;
@@ -1589,6 +1592,10 @@ static int mlx5e_set_priv_flags(struct net_device *netdev, u32 pflags)
 
 out:
 	mutex_unlock(&priv->state_lock);
+
+	/* Need to fix some features.. */
+	netdev_update_features(netdev);
+
 	return err;
 }
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index ab7b2a4e6edc..3f1f0552843c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -182,14 +182,6 @@ void mlx5e_init_rq_type_params(struct mlx5_core_dev *mdev,
 	params->log_rq_mtu_frames = is_kdump_kernel() ?
 		MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE :
 		MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE;
-	switch (params->rq_wq_type) {
-	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
-		break;
-	default: /* MLX5_WQ_TYPE_LINKED_LIST */
-		/* Extra room needed for build_skb */
-		params->lro_wqe_sz -= mlx5e_get_rq_headroom(mdev, params) +
-			SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
-	}
 
 	mlx5_core_info(mdev, "MLX5E: StrdRq(%d) RqSz(%ld) StrdSz(%ld) RxCqeCmprss(%d)\n",
 		       params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ,
@@ -503,14 +495,12 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 			goto err_rq_wq_destroy;
 		}
 
-		byte_count = params->lro_en  ?
-				params->lro_wqe_sz :
-				MLX5E_SW2HW_MTU(params, params->sw_mtu);
+		byte_count = MLX5E_SW2HW_MTU(params, params->sw_mtu);
 #ifdef CONFIG_MLX5_EN_IPSEC
 		if (MLX5_IPSEC_DEV(mdev))
 			byte_count += MLX5E_METADATA_ETHER_LEN;
 #endif
-		rq->wqe.page_reuse = !params->xdp_prog && !params->lro_en;
+		rq->wqe.page_reuse = !params->xdp_prog;
 
 		/* calc the required page order */
 		rq->wqe.frag_sz = MLX5_SKB_FRAG_SZ(rq->buff.headroom + byte_count);
@@ -3311,6 +3301,12 @@ static int set_feature_lro(struct net_device *netdev, bool enable)
 	mutex_lock(&priv->state_lock);
 
 	old_params = &priv->channels.params;
+	if (enable && !MLX5E_GET_PFLAG(old_params, MLX5E_PFLAG_RX_STRIDING_RQ)) {
+		netdev_warn(netdev, "can't set LRO with legacy RQ\n");
+		err = -EINVAL;
+		goto out;
+	}
+
 	reset = test_bit(MLX5E_STATE_OPENED, &priv->state);
 
 	new_channels.params = *old_params;
@@ -3480,16 +3476,24 @@ static netdev_features_t mlx5e_fix_features(struct net_device *netdev,
 					    netdev_features_t features)
 {
 	struct mlx5e_priv *priv = netdev_priv(netdev);
+	struct mlx5e_params *params;
 
 	mutex_lock(&priv->state_lock);
+	params = &priv->channels.params;
 	if (!bitmap_empty(priv->fs.vlan.active_svlans, VLAN_N_VID)) {
 		/* HW strips the outer C-tag header, this is a problem
 		 * for S-tag traffic.
 		 */
 		features &= ~NETIF_F_HW_VLAN_CTAG_RX;
-		if (!priv->channels.params.vlan_strip_disable)
+		if (!params->vlan_strip_disable)
 			netdev_warn(netdev, "Dropping C-tag vlan stripping offload due to S-tag vlan\n");
 	}
+	if (!MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ)) {
+		features &= ~NETIF_F_LRO;
+		if (params->lro_en)
+			netdev_warn(netdev, "Disabling LRO, not supported in legacy RQ\n");
+	}
+
 	mutex_unlock(&priv->state_lock);
 
 	return features;
@@ -4328,7 +4332,8 @@ static void mlx5e_build_nic_netdev(struct net_device *netdev)
 	netdev->hw_enc_features  |= NETIF_F_HW_VLAN_CTAG_TX;
 	netdev->hw_enc_features  |= NETIF_F_HW_VLAN_CTAG_RX;
 
-	if (!!MLX5_CAP_ETH(mdev, lro_cap))
+	if (!!MLX5_CAP_ETH(mdev, lro_cap) &&
+	    mlx5e_check_fragmented_striding_rq_cap(mdev))
 		netdev->vlan_features    |= NETIF_F_LRO;
 
 	netdev->hw_features       = netdev->vlan_features;
-- 
2.17.0

[net-next 13/17] net/mlx5e: RX, Split WQ objects for different RQ types

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

Replace the common RQ WQ object with two separate ones for the
different RQ types.
This is in preparation for switching to using a cyclic WQ type
in Legacy RQ.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h  |   4 +-
 .../net/ethernet/mellanox/mlx5/core/en_main.c | 128 ++++++++++++------
 .../net/ethernet/mellanox/mlx5/core/en_rx.c   |  35 +++--
 3 files changed, 110 insertions(+), 57 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index 9b4ed83783e4..f2f2dcf6b23c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -498,10 +498,9 @@ enum mlx5e_rq_flag {
 
 struct mlx5e_rq {
 	/* data path */
-	struct mlx5_wq_ll      wq;
-
 	union {
 		struct {
+			struct mlx5_wq_ll      wq;
 			struct mlx5e_wqe_frag_info *frag_info;
 			u32 frag_sz;	/* max possible skb frag_sz */
 			union {
@@ -509,6 +508,7 @@ struct mlx5e_rq {
 			};
 		} wqe;
 		struct {
+			struct mlx5_wq_ll      wq;
 			struct mlx5e_umr_wqe   umr_wqe;
 			struct mlx5e_mpw_info *info;
 			mlx5e_fp_skb_from_cqe_mpwrq skb_from_cqe_mpwrq;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 3f1f0552843c..3a007717cba5 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -319,10 +319,30 @@ static inline void mlx5e_build_umr_wqe(struct mlx5e_rq *rq,
 	ucseg->mkey_mask     = cpu_to_be64(MLX5_MKEY_MASK_FREE);
 }
 
+static u32 mlx5e_rqwq_get_size(struct mlx5e_rq *rq)
+{
+	switch (rq->wq_type) {
+	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+		return mlx5_wq_ll_get_size(&rq->mpwqe.wq);
+	default:
+		return mlx5_wq_ll_get_size(&rq->wqe.wq);
+	}
+}
+
+static u32 mlx5e_rqwq_get_cur_sz(struct mlx5e_rq *rq)
+{
+	switch (rq->wq_type) {
+	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+		return rq->mpwqe.wq.cur_sz;
+	default:
+		return rq->wqe.wq.cur_sz;
+	}
+}
+
 static int mlx5e_rq_alloc_mpwqe_info(struct mlx5e_rq *rq,
 				     struct mlx5e_channel *c)
 {
-	int wq_sz = mlx5_wq_ll_get_size(&rq->wq);
+	int wq_sz = mlx5_wq_ll_get_size(&rq->mpwqe.wq);
 
 	rq->mpwqe.info = kzalloc_node(wq_sz * sizeof(*rq->mpwqe.info),
 				      GFP_KERNEL, cpu_to_node(c->cpu));
@@ -370,7 +390,7 @@ static int mlx5e_create_umr_mkey(struct mlx5_core_dev *mdev,
 
 static int mlx5e_create_rq_umr_mkey(struct mlx5_core_dev *mdev, struct mlx5e_rq *rq)
 {
-	u64 num_mtts = MLX5E_REQUIRED_MTTS(mlx5_wq_ll_get_size(&rq->wq));
+	u64 num_mtts = MLX5E_REQUIRED_MTTS(mlx5_wq_ll_get_size(&rq->mpwqe.wq));
 
 	return mlx5e_create_umr_mkey(mdev, num_mtts, PAGE_SHIFT, &rq->umr_mkey);
 }
@@ -397,15 +417,6 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 
 	rqp->wq.db_numa_node = cpu_to_node(c->cpu);
 
-	err = mlx5_wq_ll_create(mdev, &rqp->wq, rqc_wq, &rq->wq,
-				&rq->wq_ctrl);
-	if (err)
-		return err;
-
-	rq->wq.db = &rq->wq.db[MLX5_RCV_DBR];
-
-	wq_sz = mlx5_wq_ll_get_size(&rq->wq);
-
 	rq->wq_type = params->rq_wq_type;
 	rq->pdev    = c->pdev;
 	rq->netdev  = c->netdev;
@@ -434,8 +445,17 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 
 	switch (rq->wq_type) {
 	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+		err = mlx5_wq_ll_create(mdev, &rqp->wq, rqc_wq, &rq->mpwqe.wq,
+					&rq->wq_ctrl);
+		if (err)
+			return err;
+
+		rq->mpwqe.wq.db = &rq->mpwqe.wq.db[MLX5_RCV_DBR];
+
+		wq_sz = mlx5_wq_ll_get_size(&rq->mpwqe.wq);
 
 		pool_size = MLX5_MPWRQ_PAGES_PER_WQE << mlx5e_mpwqe_get_log_rq_size(params);
+
 		rq->post_wqes = mlx5e_post_rx_mpwqes;
 		rq->dealloc_wqe = mlx5e_dealloc_rx_mpwqe;
 
@@ -472,6 +492,15 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 			goto err_destroy_umr_mkey;
 		break;
 	default: /* MLX5_WQ_TYPE_LINKED_LIST */
+		err = mlx5_wq_ll_create(mdev, &rqp->wq, rqc_wq, &rq->wqe.wq,
+					&rq->wq_ctrl);
+		if (err)
+			return err;
+
+		rq->wqe.wq.db = &rq->wqe.wq.db[MLX5_RCV_DBR];
+
+		wq_sz = mlx5_wq_ll_get_size(&rq->wqe.wq);
+
 		rq->wqe.frag_info =
 			kzalloc_node(wq_sz * sizeof(*rq->wqe.frag_info),
 				     GFP_KERNEL, cpu_to_node(c->cpu));
@@ -538,16 +567,21 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 		goto err_rq_wq_destroy;
 
 	for (i = 0; i < wq_sz; i++) {
-		struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i);
-
 		if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
+			struct mlx5e_rx_wqe *wqe =
+				mlx5_wq_ll_get_wqe(&rq->mpwqe.wq, i);
 			u64 dma_offset = mlx5e_get_mpwqe_offset(rq, i);
 
 			wqe->data.addr = cpu_to_be64(dma_offset + rq->buff.headroom);
-		}
+			wqe->data.byte_count = cpu_to_be32(byte_count);
+			wqe->data.lkey = rq->mkey_be;
+		} else {
+			struct mlx5e_rx_wqe *wqe =
+				mlx5_wq_ll_get_wqe(&rq->wqe.wq, i);
 
-		wqe->data.byte_count = cpu_to_be32(byte_count);
-		wqe->data.lkey = rq->mkey_be;
+			wqe->data.byte_count = cpu_to_be32(byte_count);
+			wqe->data.lkey = rq->mkey_be;
+		}
 	}
 
 	INIT_WORK(&rq->dim.work, mlx5e_rx_dim_work);
@@ -744,51 +778,65 @@ static int mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq, int wait_time)
 	unsigned long exp_time = jiffies + msecs_to_jiffies(wait_time);
 	struct mlx5e_channel *c = rq->channel;
 
-	struct mlx5_wq_ll *wq = &rq->wq;
-	u16 min_wqes = mlx5_min_rx_wqes(rq->wq_type, mlx5_wq_ll_get_size(wq));
+	u16 min_wqes = mlx5_min_rx_wqes(rq->wq_type, mlx5e_rqwq_get_size(rq));
 
 	do {
-		if (wq->cur_sz >= min_wqes)
+		if (mlx5e_rqwq_get_cur_sz(rq) >= min_wqes)
 			return 0;
 
 		msleep(20);
 	} while (time_before(jiffies, exp_time));
 
 	netdev_warn(c->netdev, "Failed to get min RX wqes on Channel[%d] RQN[0x%x] wq cur_sz(%d) min_rx_wqes(%d)\n",
-		    c->ix, rq->rqn, wq->cur_sz, min_wqes);
+		    c->ix, rq->rqn, mlx5e_rqwq_get_cur_sz(rq), min_wqes);
 
 	return -ETIMEDOUT;
 }
 
 static void mlx5e_free_rx_descs(struct mlx5e_rq *rq)
 {
-	struct mlx5_wq_ll *wq = &rq->wq;
-	struct mlx5e_rx_wqe *wqe;
 	__be16 wqe_ix_be;
 	u16 wqe_ix;
 
-	/* UMR WQE (if in progress) is always at wq->head */
-	if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ &&
-	    rq->mpwqe.umr_in_progress)
-		mlx5e_free_rx_mpwqe(rq, &rq->mpwqe.info[wq->head]);
-
-	while (!mlx5_wq_ll_is_empty(wq)) {
-		wqe_ix_be = *wq->tail_next;
-		wqe_ix    = be16_to_cpu(wqe_ix_be);
-		wqe       = mlx5_wq_ll_get_wqe(&rq->wq, wqe_ix);
-		rq->dealloc_wqe(rq, wqe_ix);
-		mlx5_wq_ll_pop(&rq->wq, wqe_ix_be,
-			       &wqe->next.next_wqe_index);
-	}
+	if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
+		struct mlx5_wq_ll *wq = &rq->mpwqe.wq;
+
+		if (rq->mpwqe.umr_in_progress)
+			mlx5e_free_rx_mpwqe(rq, &rq->mpwqe.info[wq->head]);
+
+		while (!mlx5_wq_ll_is_empty(wq)) {
+			struct mlx5e_rx_wqe *wqe;
+
+			wqe_ix_be = *wq->tail_next;
+			wqe_ix    = be16_to_cpu(wqe_ix_be);
+			wqe       = mlx5_wq_ll_get_wqe(wq, wqe_ix);
+			rq->dealloc_wqe(rq, wqe_ix);
+			mlx5_wq_ll_pop(wq, wqe_ix_be,
+				       &wqe->next.next_wqe_index);
+		}
+	} else {
+		struct mlx5_wq_ll *wq = &rq->wqe.wq;
+
+		while (!mlx5_wq_ll_is_empty(wq)) {
+			struct mlx5e_rx_wqe *wqe;
+
+			wqe_ix_be = *wq->tail_next;
+			wqe_ix    = be16_to_cpu(wqe_ix_be);
+			wqe       = mlx5_wq_ll_get_wqe(wq, wqe_ix);
+			rq->dealloc_wqe(rq, wqe_ix);
+			mlx5_wq_ll_pop(wq, wqe_ix_be,
+				       &wqe->next.next_wqe_index);
+		}
 
-	if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST && rq->wqe.page_reuse) {
 		/* Clean outstanding pages on handled WQEs that decided to do page-reuse,
 		 * but yet to be re-posted.
 		 */
-		int wq_sz = mlx5_wq_ll_get_size(&rq->wq);
+		if (rq->wqe.page_reuse) {
+			int wq_sz = mlx5_wq_ll_get_size(wq);
 
-		for (wqe_ix = 0; wqe_ix < wq_sz; wqe_ix++)
-			rq->dealloc_wqe(rq, wqe_ix);
+			for (wqe_ix = 0; wqe_ix < wq_sz; wqe_ix++)
+				rq->dealloc_wqe(rq, wqe_ix);
+		}
 	}
 }
 
@@ -2809,7 +2857,7 @@ static int mlx5e_alloc_drop_rq(struct mlx5_core_dev *mdev,
 
 	param->wq.db_numa_node = param->wq.buf_numa_node;
 
-	err = mlx5_wq_ll_create(mdev, &param->wq, rqc_wq, &rq->wq,
+	err = mlx5_wq_ll_create(mdev, &param->wq, rqc_wq, &rq->wqe.wq,
 				&rq->wq_ctrl);
 	if (err)
 		return err;
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index 634540afdcfc..3b12d4de5b98 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -113,7 +113,7 @@ static inline void mlx5e_decompress_cqe(struct mlx5e_rq *rq,
 			mpwrq_get_cqe_consumed_strides(&cq->title);
 	else
 		cq->decmprs_wqe_counter =
-			mlx5_wq_ll_ctr2ix(&rq->wq, cq->decmprs_wqe_counter + 1);
+			mlx5_wq_ll_ctr2ix(&rq->wqe.wq, cq->decmprs_wqe_counter + 1);
 }
 
 static inline void mlx5e_decompress_cqe_no_hash(struct mlx5e_rq *rq,
@@ -369,7 +369,7 @@ void mlx5e_free_rx_mpwqe(struct mlx5e_rq *rq, struct mlx5e_mpw_info *wi)
 
 static void mlx5e_post_rx_mpwqe(struct mlx5e_rq *rq)
 {
-	struct mlx5_wq_ll *wq = &rq->wq;
+	struct mlx5_wq_ll *wq = &rq->mpwqe.wq;
 	struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(wq, wq->head);
 
 	rq->mpwqe.umr_in_progress = false;
@@ -470,7 +470,7 @@ void mlx5e_dealloc_rx_mpwqe(struct mlx5e_rq *rq, u16 ix)
 
 bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq)
 {
-	struct mlx5_wq_ll *wq = &rq->wq;
+	struct mlx5_wq_ll *wq = &rq->wqe.wq;
 	int err;
 
 	if (unlikely(!test_bit(MLX5E_RQ_STATE_ENABLED, &rq->state)))
@@ -546,7 +546,7 @@ static void mlx5e_poll_ico_cq(struct mlx5e_cq *cq, struct mlx5e_rq *rq)
 
 bool mlx5e_post_rx_mpwqes(struct mlx5e_rq *rq)
 {
-	struct mlx5_wq_ll *wq = &rq->wq;
+	struct mlx5_wq_ll *wq = &rq->mpwqe.wq;
 
 	if (unlikely(!test_bit(MLX5E_RQ_STATE_ENABLED, &rq->state)))
 		return false;
@@ -987,6 +987,7 @@ struct sk_buff *skb_from_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe,
 
 void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 {
+	struct mlx5_wq_ll *wq = &rq->wqe.wq;
 	struct mlx5e_wqe_frag_info *wi;
 	struct mlx5e_rx_wqe *wqe;
 	__be16 wqe_counter_be;
@@ -996,7 +997,7 @@ void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 	wqe_counter_be = cqe->wqe_counter;
 	wqe_counter    = be16_to_cpu(wqe_counter_be);
-	wqe            = mlx5_wq_ll_get_wqe(&rq->wq, wqe_counter);
+	wqe            = mlx5_wq_ll_get_wqe(wq, wqe_counter);
 	wi             = &rq->wqe.frag_info[wqe_counter];
 	cqe_bcnt       = be32_to_cpu(cqe->byte_cnt);
 
@@ -1018,7 +1019,7 @@ void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 	mlx5e_free_rx_wqe_reuse(rq, wi);
 wq_ll_pop:
-	mlx5_wq_ll_pop(&rq->wq, wqe_counter_be,
+	mlx5_wq_ll_pop(wq, wqe_counter_be,
 		       &wqe->next.next_wqe_index);
 }
 
@@ -1029,6 +1030,7 @@ void mlx5e_handle_rx_cqe_rep(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	struct mlx5e_priv *priv = netdev_priv(netdev);
 	struct mlx5e_rep_priv *rpriv  = priv->ppriv;
 	struct mlx5_eswitch_rep *rep = rpriv->rep;
+	struct mlx5_wq_ll *wq = &rq->wqe.wq;
 	struct mlx5e_wqe_frag_info *wi;
 	struct mlx5e_rx_wqe *wqe;
 	struct sk_buff *skb;
@@ -1038,7 +1040,7 @@ void mlx5e_handle_rx_cqe_rep(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 	wqe_counter_be = cqe->wqe_counter;
 	wqe_counter    = be16_to_cpu(wqe_counter_be);
-	wqe            = mlx5_wq_ll_get_wqe(&rq->wq, wqe_counter);
+	wqe            = mlx5_wq_ll_get_wqe(wq, wqe_counter);
 	wi             = &rq->wqe.frag_info[wqe_counter];
 	cqe_bcnt       = be32_to_cpu(cqe->byte_cnt);
 
@@ -1063,7 +1065,7 @@ void mlx5e_handle_rx_cqe_rep(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 	mlx5e_free_rx_wqe_reuse(rq, wi);
 wq_ll_pop:
-	mlx5_wq_ll_pop(&rq->wq, wqe_counter_be,
+	mlx5_wq_ll_pop(wq, wqe_counter_be,
 		       &wqe->next.next_wqe_index);
 }
 #endif
@@ -1164,6 +1166,7 @@ void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	u32 head_offset    = wqe_offset & (PAGE_SIZE - 1);
 	u32 page_idx       = wqe_offset >> PAGE_SHIFT;
 	struct mlx5e_rx_wqe *wqe;
+	struct mlx5_wq_ll *wq;
 	struct sk_buff *skb;
 	u16 cqe_bcnt;
 
@@ -1193,9 +1196,10 @@ void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	if (likely(wi->consumed_strides < rq->mpwqe.num_strides))
 		return;
 
-	wqe = mlx5_wq_ll_get_wqe(&rq->wq, wqe_id);
+	wq  = &rq->mpwqe.wq;
+	wqe = mlx5_wq_ll_get_wqe(wq, wqe_id);
 	mlx5e_free_rx_mpwqe(rq, wi);
-	mlx5_wq_ll_pop(&rq->wq, cqe->wqe_id, &wqe->next.next_wqe_index);
+	mlx5_wq_ll_pop(wq, cqe->wqe_id, &wqe->next.next_wqe_index);
 }
 
 int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget)
@@ -1399,6 +1403,7 @@ static inline void mlx5i_complete_rx_cqe(struct mlx5e_rq *rq,
 
 void mlx5i_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 {
+	struct mlx5_wq_ll *wq = &rq->wqe.wq;
 	struct mlx5e_wqe_frag_info *wi;
 	struct mlx5e_rx_wqe *wqe;
 	__be16 wqe_counter_be;
@@ -1408,7 +1413,7 @@ void mlx5i_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 	wqe_counter_be = cqe->wqe_counter;
 	wqe_counter    = be16_to_cpu(wqe_counter_be);
-	wqe            = mlx5_wq_ll_get_wqe(&rq->wq, wqe_counter);
+	wqe            = mlx5_wq_ll_get_wqe(wq, wqe_counter);
 	wi             = &rq->wqe.frag_info[wqe_counter];
 	cqe_bcnt       = be32_to_cpu(cqe->byte_cnt);
 
@@ -1425,7 +1430,7 @@ void mlx5i_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 wq_free_wqe:
 	mlx5e_free_rx_wqe_reuse(rq, wi);
-	mlx5_wq_ll_pop(&rq->wq, wqe_counter_be,
+	mlx5_wq_ll_pop(wq, wqe_counter_be,
 		       &wqe->next.next_wqe_index);
 }
 
@@ -1435,6 +1440,7 @@ void mlx5i_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 void mlx5e_ipsec_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 {
+	struct mlx5_wq_ll *wq = &rq->wqe.wq;
 	struct mlx5e_wqe_frag_info *wi;
 	struct mlx5e_rx_wqe *wqe;
 	__be16 wqe_counter_be;
@@ -1444,7 +1450,7 @@ void mlx5e_ipsec_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 	wqe_counter_be = cqe->wqe_counter;
 	wqe_counter    = be16_to_cpu(wqe_counter_be);
-	wqe            = mlx5_wq_ll_get_wqe(&rq->wq, wqe_counter);
+	wqe            = mlx5_wq_ll_get_wqe(wq, wqe_counter);
 	wi             = &rq->wqe.frag_info[wqe_counter];
 	cqe_bcnt       = be32_to_cpu(cqe->byte_cnt);
 
@@ -1465,8 +1471,7 @@ void mlx5e_ipsec_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 	mlx5e_free_rx_wqe_reuse(rq, wi);
 wq_ll_pop:
-	mlx5_wq_ll_pop(&rq->wq, wqe_counter_be,
-		       &wqe->next.next_wqe_index);
+	mlx5_wq_ll_pop(wq, wqe_counter_be, &wqe->next.next_wqe_index);
 }
 
 #endif /* CONFIG_MLX5_EN_IPSEC */
-- 
2.17.0

[net-next 14/17] net/mlx5e: RX, Use cyclic WQ in legacy RQ

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

Now that LRO is not supported for Legacy RQ, there is no source of
out-of-order completions in the WQ, and we can use a cyclic one.
This has multiple advantages:
- reduces the WQE size (smaller PCI transactions).
- lower overhead in datapath (no handling of 'next' pointers).
- no reserved WQE for the WQ head (was need in linked-list).
- allows using a constant map between frag and dma_info struct, in downstream patch.

Performance tests:
ConnectX-4, single core, single RX ring.
Major gain in packet rate of single ring XDP drop.
Bottleneck is shifted form HW (at 16Mpps) to SW (at 20Mpps).

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h  |  10 +-
 .../net/ethernet/mellanox/mlx5/core/en_main.c |  89 ++++++++------
 .../net/ethernet/mellanox/mlx5/core/en_rep.c  |   2 +-
 .../net/ethernet/mellanox/mlx5/core/en_rx.c   | 115 ++++++++----------
 drivers/net/ethernet/mellanox/mlx5/core/wq.c  |   1 +
 drivers/net/ethernet/mellanox/mlx5/core/wq.h  |  55 ++++++++-
 6 files changed, 161 insertions(+), 111 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index f2f2dcf6b23c..af521dd52993 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -186,9 +186,13 @@ struct mlx5e_tx_wqe {
 	struct mlx5_wqe_data_seg data[0];
 };
 
-struct mlx5e_rx_wqe {
+struct mlx5e_rx_wqe_ll {
 	struct mlx5_wqe_srq_next_seg  next;
-	struct mlx5_wqe_data_seg      data;
+	struct mlx5_wqe_data_seg      data[0];
+};
+
+struct mlx5e_rx_wqe_cyc {
+	struct mlx5_wqe_data_seg      data[0];
 };
 
 struct mlx5e_umr_wqe {
@@ -500,7 +504,7 @@ struct mlx5e_rq {
 	/* data path */
 	union {
 		struct {
-			struct mlx5_wq_ll      wq;
+			struct mlx5_wq_cyc     wq;
 			struct mlx5e_wqe_frag_info *frag_info;
 			u32 frag_sz;	/* max possible skb frag_sz */
 			union {
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 3a007717cba5..7fd2d736fbb1 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -166,7 +166,7 @@ static u16 mlx5e_get_rq_headroom(struct mlx5_core_dev *mdev,
 
 	linear_rq_headroom += NET_IP_ALIGN;
 
-	if (params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST)
+	if (params->rq_wq_type == MLX5_WQ_TYPE_CYCLIC)
 		return linear_rq_headroom;
 
 	if (mlx5e_rx_mpwqe_is_linear_skb(mdev, params))
@@ -205,7 +205,7 @@ void mlx5e_set_rq_type(struct mlx5_core_dev *mdev, struct mlx5e_params *params)
 	params->rq_wq_type = mlx5e_striding_rq_possible(mdev, params) &&
 		MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ) ?
 		MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ :
-		MLX5_WQ_TYPE_LINKED_LIST;
+		MLX5_WQ_TYPE_CYCLIC;
 }
 
 static void mlx5e_update_carrier(struct mlx5e_priv *priv)
@@ -325,7 +325,7 @@ static u32 mlx5e_rqwq_get_size(struct mlx5e_rq *rq)
 	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
 		return mlx5_wq_ll_get_size(&rq->mpwqe.wq);
 	default:
-		return mlx5_wq_ll_get_size(&rq->wqe.wq);
+		return mlx5_wq_cyc_get_size(&rq->wqe.wq);
 	}
 }
 
@@ -491,15 +491,15 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 		if (err)
 			goto err_destroy_umr_mkey;
 		break;
-	default: /* MLX5_WQ_TYPE_LINKED_LIST */
-		err = mlx5_wq_ll_create(mdev, &rqp->wq, rqc_wq, &rq->wqe.wq,
-					&rq->wq_ctrl);
+	default: /* MLX5_WQ_TYPE_CYCLIC */
+		err = mlx5_wq_cyc_create(mdev, &rqp->wq, rqc_wq, &rq->wqe.wq,
+					 &rq->wq_ctrl);
 		if (err)
 			return err;
 
 		rq->wqe.wq.db = &rq->wqe.wq.db[MLX5_RCV_DBR];
 
-		wq_sz = mlx5_wq_ll_get_size(&rq->wqe.wq);
+		wq_sz = mlx5_wq_cyc_get_size(&rq->wqe.wq);
 
 		rq->wqe.frag_info =
 			kzalloc_node(wq_sz * sizeof(*rq->wqe.frag_info),
@@ -568,19 +568,19 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 
 	for (i = 0; i < wq_sz; i++) {
 		if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
-			struct mlx5e_rx_wqe *wqe =
+			struct mlx5e_rx_wqe_ll *wqe =
 				mlx5_wq_ll_get_wqe(&rq->mpwqe.wq, i);
 			u64 dma_offset = mlx5e_get_mpwqe_offset(rq, i);
 
-			wqe->data.addr = cpu_to_be64(dma_offset + rq->buff.headroom);
-			wqe->data.byte_count = cpu_to_be32(byte_count);
-			wqe->data.lkey = rq->mkey_be;
+			wqe->data[0].addr = cpu_to_be64(dma_offset + rq->buff.headroom);
+			wqe->data[0].byte_count = cpu_to_be32(byte_count);
+			wqe->data[0].lkey = rq->mkey_be;
 		} else {
-			struct mlx5e_rx_wqe *wqe =
-				mlx5_wq_ll_get_wqe(&rq->wqe.wq, i);
+			struct mlx5e_rx_wqe_cyc *wqe =
+				mlx5_wq_cyc_get_wqe(&rq->wqe.wq, i);
 
-			wqe->data.byte_count = cpu_to_be32(byte_count);
-			wqe->data.lkey = rq->mkey_be;
+			wqe->data[0].byte_count = cpu_to_be32(byte_count);
+			wqe->data[0].lkey = rq->mkey_be;
 		}
 	}
 
@@ -630,7 +630,7 @@ static void mlx5e_free_rq(struct mlx5e_rq *rq)
 		kfree(rq->mpwqe.info);
 		mlx5_core_destroy_mkey(rq->mdev, &rq->umr_mkey);
 		break;
-	default: /* MLX5_WQ_TYPE_LINKED_LIST */
+	default: /* MLX5_WQ_TYPE_CYCLIC */
 		kfree(rq->wqe.frag_info);
 	}
 
@@ -801,11 +801,12 @@ static void mlx5e_free_rx_descs(struct mlx5e_rq *rq)
 	if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
 		struct mlx5_wq_ll *wq = &rq->mpwqe.wq;
 
+		/* UMR WQE (if in progress) is always at wq->head */
 		if (rq->mpwqe.umr_in_progress)
 			mlx5e_free_rx_mpwqe(rq, &rq->mpwqe.info[wq->head]);
 
 		while (!mlx5_wq_ll_is_empty(wq)) {
-			struct mlx5e_rx_wqe *wqe;
+			struct mlx5e_rx_wqe_ll *wqe;
 
 			wqe_ix_be = *wq->tail_next;
 			wqe_ix    = be16_to_cpu(wqe_ix_be);
@@ -815,24 +816,19 @@ static void mlx5e_free_rx_descs(struct mlx5e_rq *rq)
 				       &wqe->next.next_wqe_index);
 		}
 	} else {
-		struct mlx5_wq_ll *wq = &rq->wqe.wq;
-
-		while (!mlx5_wq_ll_is_empty(wq)) {
-			struct mlx5e_rx_wqe *wqe;
+		struct mlx5_wq_cyc *wq = &rq->wqe.wq;
 
-			wqe_ix_be = *wq->tail_next;
-			wqe_ix    = be16_to_cpu(wqe_ix_be);
-			wqe       = mlx5_wq_ll_get_wqe(wq, wqe_ix);
+		while (!mlx5_wq_cyc_is_empty(wq)) {
+			wqe_ix = mlx5_wq_cyc_get_tail(wq);
 			rq->dealloc_wqe(rq, wqe_ix);
-			mlx5_wq_ll_pop(wq, wqe_ix_be,
-				       &wqe->next.next_wqe_index);
+			mlx5_wq_cyc_pop(wq);
 		}
 
 		/* Clean outstanding pages on handled WQEs that decided to do page-reuse,
 		 * but yet to be re-posted.
 		 */
 		if (rq->wqe.page_reuse) {
-			int wq_sz = mlx5_wq_ll_get_size(wq);
+			int wq_sz = mlx5_wq_cyc_get_size(wq);
 
 			for (wqe_ix = 0; wqe_ix < wq_sz; wqe_ix++)
 				rq->dealloc_wqe(rq, wqe_ix);
@@ -1958,6 +1954,21 @@ static void mlx5e_close_channel(struct mlx5e_channel *c)
 	kfree(c);
 }
 
+static inline u8 mlx5e_get_rqwq_log_stride(u8 wq_type, int ndsegs)
+{
+	int sz = sizeof(struct mlx5_wqe_data_seg) * ndsegs;
+
+	switch (wq_type) {
+	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+		sz += sizeof(struct mlx5e_rx_wqe_ll);
+		break;
+	default: /* MLX5_WQ_TYPE_CYCLIC */
+		sz += sizeof(struct mlx5e_rx_wqe_cyc);
+	}
+
+	return order_base_2(sz);
+}
+
 static void mlx5e_build_rq_param(struct mlx5e_priv *priv,
 				 struct mlx5e_params *params,
 				 struct mlx5e_rq_param *param)
@@ -1965,6 +1976,7 @@ static void mlx5e_build_rq_param(struct mlx5e_priv *priv,
 	struct mlx5_core_dev *mdev = priv->mdev;
 	void *rqc = param->rqc;
 	void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
+	int ndsegs = 1;
 
 	switch (params->rq_wq_type) {
 	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
@@ -1974,16 +1986,16 @@ static void mlx5e_build_rq_param(struct mlx5e_priv *priv,
 		MLX5_SET(wq, wq, log_wqe_stride_size,
 			 mlx5e_mpwqe_get_log_stride_size(mdev, params) -
 			 MLX5_MPWQE_LOG_STRIDE_SZ_BASE);
-		MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ);
 		MLX5_SET(wq, wq, log_wq_sz, mlx5e_mpwqe_get_log_rq_size(params));
 		break;
-	default: /* MLX5_WQ_TYPE_LINKED_LIST */
-		MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
+	default: /* MLX5_WQ_TYPE_CYCLIC */
 		MLX5_SET(wq, wq, log_wq_sz, params->log_rq_mtu_frames);
 	}
 
+	MLX5_SET(wq, wq, wq_type,          params->rq_wq_type);
 	MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN);
-	MLX5_SET(wq, wq, log_wq_stride,    ilog2(sizeof(struct mlx5e_rx_wqe)));
+	MLX5_SET(wq, wq, log_wq_stride,
+		 mlx5e_get_rqwq_log_stride(params->rq_wq_type, ndsegs));
 	MLX5_SET(wq, wq, pd,               mdev->mlx5e_res.pdn);
 	MLX5_SET(rqc, rqc, counter_set_id, priv->q_counter);
 	MLX5_SET(rqc, rqc, vsd,            params->vlan_strip_disable);
@@ -1999,8 +2011,9 @@ static void mlx5e_build_drop_rq_param(struct mlx5e_priv *priv,
 	void *rqc = param->rqc;
 	void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
 
-	MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST);
-	MLX5_SET(wq, wq, log_wq_stride,    ilog2(sizeof(struct mlx5e_rx_wqe)));
+	MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC);
+	MLX5_SET(wq, wq, log_wq_stride,
+		 mlx5e_get_rqwq_log_stride(MLX5_WQ_TYPE_CYCLIC, 1));
 	MLX5_SET(rqc, rqc, counter_set_id, priv->drop_rq_q_counter);
 
 	param->wq.buf_numa_node = dev_to_node(&mdev->pdev->dev);
@@ -2051,7 +2064,7 @@ static void mlx5e_build_rx_cq_param(struct mlx5e_priv *priv,
 		log_cq_size = mlx5e_mpwqe_get_log_rq_size(params) +
 			mlx5e_mpwqe_get_log_num_strides(mdev, params);
 		break;
-	default: /* MLX5_WQ_TYPE_LINKED_LIST */
+	default: /* MLX5_WQ_TYPE_CYCLIC */
 		log_cq_size = params->log_rq_mtu_frames;
 	}
 
@@ -2857,8 +2870,8 @@ static int mlx5e_alloc_drop_rq(struct mlx5_core_dev *mdev,
 
 	param->wq.db_numa_node = param->wq.buf_numa_node;
 
-	err = mlx5_wq_ll_create(mdev, &param->wq, rqc_wq, &rq->wqe.wq,
-				&rq->wq_ctrl);
+	err = mlx5_wq_cyc_create(mdev, &param->wq, rqc_wq, &rq->wqe.wq,
+				 &rq->wq_ctrl);
 	if (err)
 		return err;
 
@@ -3360,7 +3373,7 @@ static int set_feature_lro(struct net_device *netdev, bool enable)
 	new_channels.params = *old_params;
 	new_channels.params.lro_en = enable;
 
-	if (old_params->rq_wq_type != MLX5_WQ_TYPE_LINKED_LIST) {
+	if (old_params->rq_wq_type != MLX5_WQ_TYPE_CYCLIC) {
 		if (mlx5e_rx_mpwqe_is_linear_skb(mdev, old_params) ==
 		    mlx5e_rx_mpwqe_is_linear_skb(mdev, &new_channels.params))
 			reset = false;
@@ -3566,7 +3579,7 @@ int mlx5e_change_mtu(struct net_device *netdev, int new_mtu,
 	new_channels.params = *params;
 	new_channels.params.sw_mtu = new_mtu;
 
-	if (params->rq_wq_type != MLX5_WQ_TYPE_LINKED_LIST) {
+	if (params->rq_wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
 		u8 ppw_old = mlx5e_mpwqe_log_pkts_per_wqe(params);
 		u8 ppw_new = mlx5e_mpwqe_log_pkts_per_wqe(&new_channels.params);
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
index 8ab4c96b7f7c..3857f22b5500 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c
@@ -927,7 +927,7 @@ static void mlx5e_build_rep_params(struct mlx5_core_dev *mdev,
 	params->hard_mtu    = MLX5E_ETH_HARD_MTU;
 	params->sw_mtu      = mtu;
 	params->log_sq_size = MLX5E_REP_PARAMS_LOG_SQ_SIZE;
-	params->rq_wq_type  = MLX5_WQ_TYPE_LINKED_LIST;
+	params->rq_wq_type  = MLX5_WQ_TYPE_CYCLIC;
 	params->log_rq_mtu_frames = MLX5E_REP_PARAMS_LOG_RQ_SIZE;
 
 	params->rx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index 3b12d4de5b98..3cdf2c097356 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -113,7 +113,7 @@ static inline void mlx5e_decompress_cqe(struct mlx5e_rq *rq,
 			mpwrq_get_cqe_consumed_strides(&cq->title);
 	else
 		cq->decmprs_wqe_counter =
-			mlx5_wq_ll_ctr2ix(&rq->wqe.wq, cq->decmprs_wqe_counter + 1);
+			mlx5_wq_cyc_ctr2ix(&rq->wqe.wq, cq->decmprs_wqe_counter + 1);
 }
 
 static inline void mlx5e_decompress_cqe_no_hash(struct mlx5e_rq *rq,
@@ -270,7 +270,12 @@ static inline bool mlx5e_page_reuse(struct mlx5e_rq *rq,
 		!mlx5e_page_is_reserved(wi->di.page);
 }
 
-static int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16 ix)
+static inline struct mlx5e_wqe_frag_info *get_frag(struct mlx5e_rq *rq, u16 ix)
+{
+	return &rq->wqe.frag_info[ix];
+}
+
+static int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe_cyc *wqe, u16 ix)
 {
 	struct mlx5e_wqe_frag_info *wi = &rq->wqe.frag_info[ix];
 
@@ -281,7 +286,7 @@ static int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe *wqe, u16
 		wi->offset = 0;
 	}
 
-	wqe->data.addr = cpu_to_be64(wi->di.addr + wi->offset + rq->buff.headroom);
+	wqe->data[0].addr = cpu_to_be64(wi->di.addr + wi->offset + rq->buff.headroom);
 	return 0;
 }
 
@@ -370,7 +375,7 @@ void mlx5e_free_rx_mpwqe(struct mlx5e_rq *rq, struct mlx5e_mpw_info *wi)
 static void mlx5e_post_rx_mpwqe(struct mlx5e_rq *rq)
 {
 	struct mlx5_wq_ll *wq = &rq->mpwqe.wq;
-	struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(wq, wq->head);
+	struct mlx5e_rx_wqe_ll *wqe = mlx5_wq_ll_get_wqe(wq, wq->head);
 
 	rq->mpwqe.umr_in_progress = false;
 
@@ -470,31 +475,32 @@ void mlx5e_dealloc_rx_mpwqe(struct mlx5e_rq *rq, u16 ix)
 
 bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq)
 {
-	struct mlx5_wq_ll *wq = &rq->wqe.wq;
+	struct mlx5_wq_cyc *wq = &rq->wqe.wq;
 	int err;
 
 	if (unlikely(!test_bit(MLX5E_RQ_STATE_ENABLED, &rq->state)))
 		return false;
 
-	if (mlx5_wq_ll_is_full(wq))
+	if (mlx5_wq_cyc_is_full(wq))
 		return false;
 
 	do {
-		struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(wq, wq->head);
+		u16 head = mlx5_wq_cyc_get_head(wq);
+		struct mlx5e_rx_wqe_cyc *wqe = mlx5_wq_cyc_get_wqe(wq, head);
 
-		err = mlx5e_alloc_rx_wqe(rq, wqe, wq->head);
+		err = mlx5e_alloc_rx_wqe(rq, wqe, head);
 		if (unlikely(err)) {
 			rq->stats->buff_alloc_err++;
 			break;
 		}
 
-		mlx5_wq_ll_push(wq, be16_to_cpu(wqe->next.next_wqe_index));
-	} while (!mlx5_wq_ll_is_full(wq));
+		mlx5_wq_cyc_push(wq);
+	} while (!mlx5_wq_cyc_is_full(wq));
 
 	/* ensure wqes are visible to device before updating doorbell record */
 	dma_wmb();
 
-	mlx5_wq_ll_update_db_record(wq);
+	mlx5_wq_cyc_update_db_record(wq);
 
 	return !!err;
 }
@@ -987,19 +993,15 @@ struct sk_buff *skb_from_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe,
 
 void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 {
-	struct mlx5_wq_ll *wq = &rq->wqe.wq;
+	struct mlx5_wq_cyc *wq = &rq->wqe.wq;
 	struct mlx5e_wqe_frag_info *wi;
-	struct mlx5e_rx_wqe *wqe;
-	__be16 wqe_counter_be;
 	struct sk_buff *skb;
-	u16 wqe_counter;
 	u32 cqe_bcnt;
+	u16 ci;
 
-	wqe_counter_be = cqe->wqe_counter;
-	wqe_counter    = be16_to_cpu(wqe_counter_be);
-	wqe            = mlx5_wq_ll_get_wqe(wq, wqe_counter);
-	wi             = &rq->wqe.frag_info[wqe_counter];
-	cqe_bcnt       = be32_to_cpu(cqe->byte_cnt);
+	ci       = mlx5_wq_cyc_ctr2ix(wq, be16_to_cpu(cqe->wqe_counter));
+	wi       = get_frag(rq, ci);
+	cqe_bcnt = be32_to_cpu(cqe->byte_cnt);
 
 	skb = skb_from_cqe(rq, cqe, wi, cqe_bcnt);
 	if (!skb) {
@@ -1007,20 +1009,19 @@ void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 		if (__test_and_clear_bit(MLX5E_RQ_FLAG_XDP_XMIT, rq->flags)) {
 			wi->di.page = NULL;
 			/* do not return page to cache, it will be returned on XDP_TX completion */
-			goto wq_ll_pop;
+			goto wq_cyc_pop;
 		}
 		/* probably an XDP_DROP, save the page-reuse checks */
 		mlx5e_free_rx_wqe(rq, wi);
-		goto wq_ll_pop;
+		goto wq_cyc_pop;
 	}
 
 	mlx5e_complete_rx_cqe(rq, cqe, cqe_bcnt, skb);
 	napi_gro_receive(rq->cq.napi, skb);
 
 	mlx5e_free_rx_wqe_reuse(rq, wi);
-wq_ll_pop:
-	mlx5_wq_ll_pop(wq, wqe_counter_be,
-		       &wqe->next.next_wqe_index);
+wq_cyc_pop:
+	mlx5_wq_cyc_pop(wq);
 }
 
 #ifdef CONFIG_MLX5_ESWITCH
@@ -1030,30 +1031,26 @@ void mlx5e_handle_rx_cqe_rep(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	struct mlx5e_priv *priv = netdev_priv(netdev);
 	struct mlx5e_rep_priv *rpriv  = priv->ppriv;
 	struct mlx5_eswitch_rep *rep = rpriv->rep;
-	struct mlx5_wq_ll *wq = &rq->wqe.wq;
+	struct mlx5_wq_cyc *wq = &rq->wqe.wq;
 	struct mlx5e_wqe_frag_info *wi;
-	struct mlx5e_rx_wqe *wqe;
 	struct sk_buff *skb;
-	__be16 wqe_counter_be;
-	u16 wqe_counter;
 	u32 cqe_bcnt;
+	u16 ci;
 
-	wqe_counter_be = cqe->wqe_counter;
-	wqe_counter    = be16_to_cpu(wqe_counter_be);
-	wqe            = mlx5_wq_ll_get_wqe(wq, wqe_counter);
-	wi             = &rq->wqe.frag_info[wqe_counter];
-	cqe_bcnt       = be32_to_cpu(cqe->byte_cnt);
+	ci       = mlx5_wq_cyc_ctr2ix(wq, be16_to_cpu(cqe->wqe_counter));
+	wi       = get_frag(rq, ci);
+	cqe_bcnt = be32_to_cpu(cqe->byte_cnt);
 
 	skb = skb_from_cqe(rq, cqe, wi, cqe_bcnt);
 	if (!skb) {
 		if (__test_and_clear_bit(MLX5E_RQ_FLAG_XDP_XMIT, rq->flags)) {
 			wi->di.page = NULL;
 			/* do not return page to cache, it will be returned on XDP_TX completion */
-			goto wq_ll_pop;
+			goto wq_cyc_pop;
 		}
 		/* probably an XDP_DROP, save the page-reuse checks */
 		mlx5e_free_rx_wqe(rq, wi);
-		goto wq_ll_pop;
+		goto wq_cyc_pop;
 	}
 
 	mlx5e_complete_rx_cqe(rq, cqe, cqe_bcnt, skb);
@@ -1064,9 +1061,8 @@ void mlx5e_handle_rx_cqe_rep(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	napi_gro_receive(rq->cq.napi, skb);
 
 	mlx5e_free_rx_wqe_reuse(rq, wi);
-wq_ll_pop:
-	mlx5_wq_ll_pop(wq, wqe_counter_be,
-		       &wqe->next.next_wqe_index);
+wq_cyc_pop:
+	mlx5_wq_cyc_pop(wq);
 }
 #endif
 
@@ -1165,7 +1161,7 @@ void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	u32 wqe_offset     = stride_ix << rq->mpwqe.log_stride_sz;
 	u32 head_offset    = wqe_offset & (PAGE_SIZE - 1);
 	u32 page_idx       = wqe_offset >> PAGE_SHIFT;
-	struct mlx5e_rx_wqe *wqe;
+	struct mlx5e_rx_wqe_ll *wqe;
 	struct mlx5_wq_ll *wq;
 	struct sk_buff *skb;
 	u16 cqe_bcnt;
@@ -1403,19 +1399,15 @@ static inline void mlx5i_complete_rx_cqe(struct mlx5e_rq *rq,
 
 void mlx5i_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 {
-	struct mlx5_wq_ll *wq = &rq->wqe.wq;
+	struct mlx5_wq_cyc *wq = &rq->wqe.wq;
 	struct mlx5e_wqe_frag_info *wi;
-	struct mlx5e_rx_wqe *wqe;
-	__be16 wqe_counter_be;
 	struct sk_buff *skb;
-	u16 wqe_counter;
 	u32 cqe_bcnt;
+	u16 ci;
 
-	wqe_counter_be = cqe->wqe_counter;
-	wqe_counter    = be16_to_cpu(wqe_counter_be);
-	wqe            = mlx5_wq_ll_get_wqe(wq, wqe_counter);
-	wi             = &rq->wqe.frag_info[wqe_counter];
-	cqe_bcnt       = be32_to_cpu(cqe->byte_cnt);
+	ci       = mlx5_wq_cyc_ctr2ix(wq, be16_to_cpu(cqe->wqe_counter));
+	wi       = get_frag(rq, ci);
+	cqe_bcnt = be32_to_cpu(cqe->byte_cnt);
 
 	skb = skb_from_cqe(rq, cqe, wi, cqe_bcnt);
 	if (!skb)
@@ -1430,8 +1422,7 @@ void mlx5i_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 wq_free_wqe:
 	mlx5e_free_rx_wqe_reuse(rq, wi);
-	mlx5_wq_ll_pop(wq, wqe_counter_be,
-		       &wqe->next.next_wqe_index);
+	mlx5_wq_cyc_pop(wq);
 }
 
 #endif /* CONFIG_MLX5_CORE_IPOIB */
@@ -1440,38 +1431,34 @@ void mlx5i_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 void mlx5e_ipsec_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 {
-	struct mlx5_wq_ll *wq = &rq->wqe.wq;
+	struct mlx5_wq_cyc *wq = &rq->wqe.wq;
 	struct mlx5e_wqe_frag_info *wi;
-	struct mlx5e_rx_wqe *wqe;
-	__be16 wqe_counter_be;
 	struct sk_buff *skb;
-	u16 wqe_counter;
 	u32 cqe_bcnt;
+	u16 ci;
 
-	wqe_counter_be = cqe->wqe_counter;
-	wqe_counter    = be16_to_cpu(wqe_counter_be);
-	wqe            = mlx5_wq_ll_get_wqe(wq, wqe_counter);
-	wi             = &rq->wqe.frag_info[wqe_counter];
-	cqe_bcnt       = be32_to_cpu(cqe->byte_cnt);
+	ci       = mlx5_wq_cyc_ctr2ix(wq, be16_to_cpu(cqe->wqe_counter));
+	wi       = get_frag(rq, ci);
+	cqe_bcnt = be32_to_cpu(cqe->byte_cnt);
 
 	skb = skb_from_cqe(rq, cqe, wi, cqe_bcnt);
 	if (unlikely(!skb)) {
 		/* a DROP, save the page-reuse checks */
 		mlx5e_free_rx_wqe(rq, wi);
-		goto wq_ll_pop;
+		goto wq_cyc_pop;
 	}
 	skb = mlx5e_ipsec_handle_rx_skb(rq->netdev, skb);
 	if (unlikely(!skb)) {
 		mlx5e_free_rx_wqe(rq, wi);
-		goto wq_ll_pop;
+		goto wq_cyc_pop;
 	}
 
 	mlx5e_complete_rx_cqe(rq, cqe, cqe_bcnt, skb);
 	napi_gro_receive(rq->cq.napi, skb);
 
 	mlx5e_free_rx_wqe_reuse(rq, wi);
-wq_ll_pop:
-	mlx5_wq_ll_pop(wq, wqe_counter_be, &wqe->next.next_wqe_index);
+wq_cyc_pop:
+	mlx5_wq_cyc_pop(wq);
 }
 
 #endif /* CONFIG_MLX5_EN_IPSEC */
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/wq.c b/drivers/net/ethernet/mellanox/mlx5/core/wq.c
index 5b8b35392025..b97bb72b4db4 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/wq.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/wq.c
@@ -85,6 +85,7 @@ int mlx5_wq_cyc_create(struct mlx5_core_dev *mdev, struct mlx5_wq_param *param,
 	mlx5_fill_fbc(MLX5_GET(wq, wqc, log_wq_stride),
 		      MLX5_GET(wq, wqc, log_wq_sz),
 		      fbc);
+	wq->sz    = wq->fbc.sz_m1 + 1;
 
 	err = mlx5_db_alloc_node(mdev, &wq_ctrl->db, param->db_numa_node);
 	if (err) {
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/wq.h b/drivers/net/ethernet/mellanox/mlx5/core/wq.h
index b9d7c01fc7cb..0b47126815b6 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/wq.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/wq.h
@@ -51,6 +51,9 @@ struct mlx5_wq_ctrl {
 struct mlx5_wq_cyc {
 	struct mlx5_frag_buf_ctrl fbc;
 	__be32			*db;
+	u16			sz;
+	u16			wqe_ctr;
+	u16			cur_sz;
 };
 
 struct mlx5_wq_qp {
@@ -95,6 +98,43 @@ u32 mlx5_wq_ll_get_size(struct mlx5_wq_ll *wq);
 
 void mlx5_wq_destroy(struct mlx5_wq_ctrl *wq_ctrl);
 
+static inline int mlx5_wq_cyc_is_full(struct mlx5_wq_cyc *wq)
+{
+	return wq->cur_sz == wq->sz;
+}
+
+static inline int mlx5_wq_cyc_missing(struct mlx5_wq_cyc *wq)
+{
+	return wq->sz - wq->cur_sz;
+}
+
+static inline int mlx5_wq_cyc_is_empty(struct mlx5_wq_cyc *wq)
+{
+	return !wq->cur_sz;
+}
+
+static inline void mlx5_wq_cyc_push(struct mlx5_wq_cyc *wq)
+{
+	wq->wqe_ctr++;
+	wq->cur_sz++;
+}
+
+static inline void mlx5_wq_cyc_push_n(struct mlx5_wq_cyc *wq, u8 n)
+{
+	wq->wqe_ctr += n;
+	wq->cur_sz += n;
+}
+
+static inline void mlx5_wq_cyc_pop(struct mlx5_wq_cyc *wq)
+{
+	wq->cur_sz--;
+}
+
+static inline void mlx5_wq_cyc_update_db_record(struct mlx5_wq_cyc *wq)
+{
+	*wq->db = cpu_to_be32(wq->wqe_ctr);
+}
+
 static inline u16 mlx5_wq_cyc_ctr2ix(struct mlx5_wq_cyc *wq, u16 ctr)
 {
 	return ctr & wq->fbc.sz_m1;
@@ -105,6 +145,16 @@ static inline u16 mlx5_wq_cyc_ctr2fragix(struct mlx5_wq_cyc *wq, u16 ctr)
 	return ctr & wq->fbc.frag_sz_m1;
 }
 
+static inline u16 mlx5_wq_cyc_get_head(struct mlx5_wq_cyc *wq)
+{
+	return mlx5_wq_cyc_ctr2ix(wq, wq->wqe_ctr);
+}
+
+static inline u16 mlx5_wq_cyc_get_tail(struct mlx5_wq_cyc *wq)
+{
+	return mlx5_wq_cyc_ctr2ix(wq, wq->wqe_ctr - wq->cur_sz);
+}
+
 static inline void *mlx5_wq_cyc_get_wqe(struct mlx5_wq_cyc *wq, u16 ix)
 {
 	return mlx5_frag_buf_get_wqe(&wq->fbc, ix);
@@ -179,11 +229,6 @@ static inline int mlx5_wq_ll_is_empty(struct mlx5_wq_ll *wq)
 	return !wq->cur_sz;
 }
 
-static inline u16 mlx5_wq_ll_ctr2ix(struct mlx5_wq_ll *wq, u16 ctr)
-{
-	return ctr & wq->fbc.sz_m1;
-}
-
 static inline void *mlx5_wq_ll_get_wqe(struct mlx5_wq_ll *wq, u16 ix)
 {
 	return mlx5_frag_buf_get_wqe(&wq->fbc, ix);
-- 
2.17.0

[net-next 15/17] net/mlx5e: RX, Enhance legacy Receive Queue memory scheme

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

Enhance the memory scheme of the legacy RQ, such that
only order-0 pages are used.

Whenever possible, prefer using a linear SKB, and build it
wrapping the WQE buffer.

Otherwise (for example, jumbo frames on x86), use non-linear SKB,
with as many frags as needed. In this case, multiple WQE
scatter entries are used, up to a maximum of 4 frags and 10KB of MTU.

This implied to remove support of HW LRO in legacy RQ, as it would
require large number of page allocations and scatter entries per WQE
on archs with PAGE_SIZE = 4KB, yielding bad performance.

In earlier patches, we guaranteed that all completions are in-order,
and that we use a cyclic WQ.
This creates an oppurtunity for a performance optimization:
The mapping between a "struct mlx5e_dma_info", and the
WQEs (struct mlx5e_wqe_frag_info) pointing to it, is constant
across different cycles of a WQ. This allows initializing
the mapping in the time of RQ creation, and not handle it
in datapath.

A struct mlx5e_dma_info that is shared between different WQEs
is allocated by the first WQE, and freed by the last one.
This implies an important requirement: WQEs that share the same
struct mlx5e_dma_info must be posted within the same NAPI.
Otherwise, upon completion, struct mlx5e_wqe_frag_info would mistakenly
point to the new struct mlx5e_dma_info, not the one that was posted
(and the HW wrote to).
This bulking requirement is actually good also for performance reasons,
hence we extend the bulk beyong the minimal requirement above.

With this memory scheme, the RQs memory footprint is reduce by a
factor of 2 on x86, and by a factor of 32 on PowerPC.
Same factors apply for the number of pages in a GRO session.

Performance tests:
ConnectX-4, single core, single RX ring, default MTU.

x86:
CPU: Intel(R) Xeon(R) CPU E5-2680 v3 @ 2.50GHz

Packet rate (early drop in TC): no degradation
TCP streams: ~5% improvement

PowerPC:
CPU: POWER8 (raw), altivec supported

Packet rate (early drop in TC): 20% gain
TCP streams: 25% gain

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en.h  |  44 +++-
 .../net/ethernet/mellanox/mlx5/core/en_main.c | 238 +++++++++++++-----
 .../net/ethernet/mellanox/mlx5/core/en_rx.c   | 207 ++++++++++-----
 3 files changed, 362 insertions(+), 127 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h
index af521dd52993..eb9eb7aa953a 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h
@@ -101,11 +101,15 @@ struct page_pool;
 	(MLX5E_PARAMS_MAXIMUM_LOG_RQ_SIZE_MPW + \
 	 (MLX5_MPWRQ_LOG_WQE_SZ - MLX5E_ORDER2_MAX_PACKET_MTU))
 
+#define MLX5E_MIN_SKB_FRAG_SZ		(MLX5_SKB_FRAG_SZ(MLX5_RX_HEADROOM))
+#define MLX5E_LOG_MAX_RX_WQE_BULK	\
+	(ilog2(PAGE_SIZE / roundup_pow_of_two(MLX5E_MIN_SKB_FRAG_SZ)))
+
 #define MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE                0x6
 #define MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE                0xa
 #define MLX5E_PARAMS_MAXIMUM_LOG_SQ_SIZE                0xd
 
-#define MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE                0x1
+#define MLX5E_PARAMS_MINIMUM_LOG_RQ_SIZE (1 + MLX5E_LOG_MAX_RX_WQE_BULK)
 #define MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE                0xa
 #define MLX5E_PARAMS_MAXIMUM_LOG_RQ_SIZE min_t(u8, 0xd,	\
 					       MLX5E_LOG_MAX_RQ_NUM_PACKETS_MPW)
@@ -462,8 +466,9 @@ struct mlx5e_dma_info {
 };
 
 struct mlx5e_wqe_frag_info {
-	struct mlx5e_dma_info di;
+	struct mlx5e_dma_info *di;
 	u32 offset;
+	bool last_in_page;
 };
 
 struct mlx5e_umr_dma_info {
@@ -476,6 +481,8 @@ struct mlx5e_mpw_info {
 	DECLARE_BITMAP(xdp_xmit_bitmap, MLX5_MPWRQ_PAGES_PER_WQE);
 };
 
+#define MLX5E_MAX_RX_FRAGS 4
+
 /* a single cache unit is capable to serve one napi call (for non-striding rq)
  * or a MPWQE (for striding rq).
  */
@@ -493,6 +500,9 @@ typedef void (*mlx5e_fp_handle_rx_cqe)(struct mlx5e_rq*, struct mlx5_cqe64*);
 typedef struct sk_buff *
 (*mlx5e_fp_skb_from_cqe_mpwrq)(struct mlx5e_rq *rq, struct mlx5e_mpw_info *wi,
 			       u16 cqe_bcnt, u32 head_offset, u32 page_idx);
+typedef struct sk_buff *
+(*mlx5e_fp_skb_from_cqe)(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe,
+			 struct mlx5e_wqe_frag_info *wi, u32 cqe_bcnt);
 typedef bool (*mlx5e_fp_post_rx_wqes)(struct mlx5e_rq *rq);
 typedef void (*mlx5e_fp_dealloc_wqe)(struct mlx5e_rq*, u16);
 
@@ -500,16 +510,27 @@ enum mlx5e_rq_flag {
 	MLX5E_RQ_FLAG_XDP_XMIT = BIT(0),
 };
 
+struct mlx5e_rq_frag_info {
+	int frag_size;
+	int frag_stride;
+};
+
+struct mlx5e_rq_frags_info {
+	struct mlx5e_rq_frag_info arr[MLX5E_MAX_RX_FRAGS];
+	u8 num_frags;
+	u8 log_num_frags;
+	u8 wqe_bulk;
+};
+
 struct mlx5e_rq {
 	/* data path */
 	union {
 		struct {
-			struct mlx5_wq_cyc     wq;
-			struct mlx5e_wqe_frag_info *frag_info;
-			u32 frag_sz;	/* max possible skb frag_sz */
-			union {
-				bool page_reuse;
-			};
+			struct mlx5_wq_cyc          wq;
+			struct mlx5e_wqe_frag_info *frags;
+			struct mlx5e_dma_info      *di;
+			struct mlx5e_rq_frags_info  info;
+			mlx5e_fp_skb_from_cqe       skb_from_cqe;
 		} wqe;
 		struct {
 			struct mlx5_wq_ll      wq;
@@ -523,7 +544,6 @@ struct mlx5e_rq {
 	};
 	struct {
 		u16            headroom;
-		u8             page_order;
 		u8             map_dir;   /* dma map direction */
 	} buff;
 
@@ -879,6 +899,12 @@ mlx5e_skb_from_cqe_mpwrq_linear(struct mlx5e_rq *rq, struct mlx5e_mpw_info *wi,
 struct sk_buff *
 mlx5e_skb_from_cqe_mpwrq_nonlinear(struct mlx5e_rq *rq, struct mlx5e_mpw_info *wi,
 				   u16 cqe_bcnt, u32 head_offset, u32 page_idx);
+struct sk_buff *
+mlx5e_skb_from_cqe_linear(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe,
+			  struct mlx5e_wqe_frag_info *wi, u32 cqe_bcnt);
+struct sk_buff *
+mlx5e_skb_from_cqe_nonlinear(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe,
+			     struct mlx5e_wqe_frag_info *wi, u32 cqe_bcnt);
 
 void mlx5e_update_stats(struct mlx5e_priv *priv);
 
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 7fd2d736fbb1..2c634e50d051 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -51,6 +51,7 @@
 struct mlx5e_rq_param {
 	u32			rqc[MLX5_ST_SZ_DW(rqc)];
 	struct mlx5_wq_param	wq;
+	struct mlx5e_rq_frags_info frags_info;
 };
 
 struct mlx5e_sq_param {
@@ -93,7 +94,7 @@ bool mlx5e_check_fragmented_striding_rq_cap(struct mlx5_core_dev *mdev)
 	return true;
 }
 
-static u32 mlx5e_mpwqe_get_linear_frag_sz(struct mlx5e_params *params)
+static u32 mlx5e_rx_get_linear_frag_sz(struct mlx5e_params *params)
 {
 	if (!params->xdp_prog) {
 		u16 hw_mtu = MLX5E_SW2HW_MTU(params, params->sw_mtu);
@@ -107,19 +108,27 @@ static u32 mlx5e_mpwqe_get_linear_frag_sz(struct mlx5e_params *params)
 
 static u8 mlx5e_mpwqe_log_pkts_per_wqe(struct mlx5e_params *params)
 {
-	u32 linear_frag_sz = mlx5e_mpwqe_get_linear_frag_sz(params);
+	u32 linear_frag_sz = mlx5e_rx_get_linear_frag_sz(params);
 
 	return MLX5_MPWRQ_LOG_WQE_SZ - order_base_2(linear_frag_sz);
 }
 
+static bool mlx5e_rx_is_linear_skb(struct mlx5_core_dev *mdev,
+				   struct mlx5e_params *params)
+{
+	u32 frag_sz = mlx5e_rx_get_linear_frag_sz(params);
+
+	return !params->lro_en && frag_sz <= PAGE_SIZE;
+}
+
 static bool mlx5e_rx_mpwqe_is_linear_skb(struct mlx5_core_dev *mdev,
 					 struct mlx5e_params *params)
 {
-	u32 frag_sz = mlx5e_mpwqe_get_linear_frag_sz(params);
+	u32 frag_sz = mlx5e_rx_get_linear_frag_sz(params);
 	s8 signed_log_num_strides_param;
 	u8 log_num_strides;
 
-	if (params->lro_en || frag_sz > PAGE_SIZE)
+	if (!mlx5e_rx_is_linear_skb(mdev, params))
 		return false;
 
 	if (MLX5_CAP_GEN(mdev, ext_stride_num_range))
@@ -145,7 +154,7 @@ static u8 mlx5e_mpwqe_get_log_stride_size(struct mlx5_core_dev *mdev,
 					  struct mlx5e_params *params)
 {
 	if (mlx5e_rx_mpwqe_is_linear_skb(mdev, params))
-		return order_base_2(mlx5e_mpwqe_get_linear_frag_sz(params));
+		return order_base_2(mlx5e_rx_get_linear_frag_sz(params));
 
 	return MLX5E_MPWQE_STRIDE_SZ(mdev,
 		MLX5E_GET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS));
@@ -163,16 +172,15 @@ static u16 mlx5e_get_rq_headroom(struct mlx5_core_dev *mdev,
 {
 	u16 linear_rq_headroom = params->xdp_prog ?
 		XDP_PACKET_HEADROOM : MLX5_RX_HEADROOM;
+	bool is_linear_skb;
 
 	linear_rq_headroom += NET_IP_ALIGN;
 
-	if (params->rq_wq_type == MLX5_WQ_TYPE_CYCLIC)
-		return linear_rq_headroom;
-
-	if (mlx5e_rx_mpwqe_is_linear_skb(mdev, params))
-		return linear_rq_headroom;
+	is_linear_skb = (params->rq_wq_type == MLX5_WQ_TYPE_CYCLIC) ?
+		mlx5e_rx_is_linear_skb(mdev, params) :
+		mlx5e_rx_mpwqe_is_linear_skb(mdev, params);
 
-	return 0;
+	return is_linear_skb ? linear_rq_headroom : 0;
 }
 
 void mlx5e_init_rq_type_params(struct mlx5_core_dev *mdev,
@@ -400,6 +408,61 @@ static inline u64 mlx5e_get_mpwqe_offset(struct mlx5e_rq *rq, u16 wqe_ix)
 	return (wqe_ix << MLX5E_LOG_ALIGNED_MPWQE_PPW) << PAGE_SHIFT;
 }
 
+static void mlx5e_init_frags_partition(struct mlx5e_rq *rq)
+{
+	struct mlx5e_wqe_frag_info next_frag, *prev;
+	int i;
+
+	next_frag.di = &rq->wqe.di[0];
+	next_frag.offset = 0;
+	prev = NULL;
+
+	for (i = 0; i < mlx5_wq_cyc_get_size(&rq->wqe.wq); i++) {
+		struct mlx5e_rq_frag_info *frag_info = &rq->wqe.info.arr[0];
+		struct mlx5e_wqe_frag_info *frag =
+			&rq->wqe.frags[i << rq->wqe.info.log_num_frags];
+		int f;
+
+		for (f = 0; f < rq->wqe.info.num_frags; f++, frag++) {
+			if (next_frag.offset + frag_info[f].frag_stride > PAGE_SIZE) {
+				next_frag.di++;
+				next_frag.offset = 0;
+				if (prev)
+					prev->last_in_page = true;
+			}
+			*frag = next_frag;
+
+			/* prepare next */
+			next_frag.offset += frag_info[f].frag_stride;
+			prev = frag;
+		}
+	}
+
+	if (prev)
+		prev->last_in_page = true;
+}
+
+static int mlx5e_init_di_list(struct mlx5e_rq *rq,
+			      struct mlx5e_params *params,
+			      int wq_sz, int cpu)
+{
+	int len = wq_sz << rq->wqe.info.log_num_frags;
+
+	rq->wqe.di = kvzalloc_node(len * sizeof(*rq->wqe.di),
+				   GFP_KERNEL, cpu_to_node(cpu));
+	if (!rq->wqe.di)
+		return -ENOMEM;
+
+	mlx5e_init_frags_partition(rq);
+
+	return 0;
+}
+
+static void mlx5e_free_di_list(struct mlx5e_rq *rq)
+{
+	kvfree(rq->wqe.di);
+}
+
 static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 			  struct mlx5e_params *params,
 			  struct mlx5e_rq_param *rqp,
@@ -409,8 +472,7 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 	struct mlx5_core_dev *mdev = c->mdev;
 	void *rqc = rqp->rqc;
 	void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq);
-	u32 byte_count, pool_size;
-	int npages;
+	u32 pool_size;
 	int wq_sz;
 	int err;
 	int i;
@@ -480,8 +542,6 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 		rq->mpwqe.log_stride_sz = mlx5e_mpwqe_get_log_stride_size(mdev, params);
 		rq->mpwqe.num_strides = BIT(mlx5e_mpwqe_get_log_num_strides(mdev, params));
 
-		byte_count = rq->mpwqe.num_strides << rq->mpwqe.log_stride_sz;
-
 		err = mlx5e_create_rq_umr_mkey(mdev, rq);
 		if (err)
 			goto err_rq_wq_destroy;
@@ -489,7 +549,7 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 
 		err = mlx5e_rq_alloc_mpwqe_info(rq, c);
 		if (err)
-			goto err_destroy_umr_mkey;
+			goto err_free;
 		break;
 	default: /* MLX5_WQ_TYPE_CYCLIC */
 		err = mlx5_wq_cyc_create(mdev, &rqp->wq, rqc_wq, &rq->wqe.wq,
@@ -501,13 +561,17 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 
 		wq_sz = mlx5_wq_cyc_get_size(&rq->wqe.wq);
 
-		rq->wqe.frag_info =
-			kzalloc_node(wq_sz * sizeof(*rq->wqe.frag_info),
-				     GFP_KERNEL, cpu_to_node(c->cpu));
-		if (!rq->wqe.frag_info) {
-			err = -ENOMEM;
-			goto err_rq_wq_destroy;
-		}
+		rq->wqe.info = rqp->frags_info;
+		rq->wqe.frags =
+			kvzalloc_node((wq_sz << rq->wqe.info.log_num_frags) *
+				      sizeof(*rq->wqe.frags),
+				      GFP_KERNEL, cpu_to_node(c->cpu));
+		if (!rq->wqe.frags)
+			goto err_free;
+
+		err = mlx5e_init_di_list(rq, params, wq_sz, c->cpu);
+		if (err)
+			goto err_free;
 		rq->post_wqes = mlx5e_post_rx_wqes;
 		rq->dealloc_wqe = mlx5e_dealloc_rx_wqe;
 
@@ -518,30 +582,19 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 #endif
 			rq->handle_rx_cqe = c->priv->profile->rx_handlers.handle_rx_cqe;
 		if (!rq->handle_rx_cqe) {
-			kfree(rq->wqe.frag_info);
 			err = -EINVAL;
 			netdev_err(c->netdev, "RX handler of RQ is not set, err %d\n", err);
-			goto err_rq_wq_destroy;
+			goto err_free;
 		}
 
-		byte_count = MLX5E_SW2HW_MTU(params, params->sw_mtu);
-#ifdef CONFIG_MLX5_EN_IPSEC
-		if (MLX5_IPSEC_DEV(mdev))
-			byte_count += MLX5E_METADATA_ETHER_LEN;
-#endif
-		rq->wqe.page_reuse = !params->xdp_prog;
-
-		/* calc the required page order */
-		rq->wqe.frag_sz = MLX5_SKB_FRAG_SZ(rq->buff.headroom + byte_count);
-		npages = DIV_ROUND_UP(rq->wqe.frag_sz, PAGE_SIZE);
-		rq->buff.page_order = order_base_2(npages);
-
-		byte_count |= MLX5_HW_START_PADDING;
+		rq->wqe.skb_from_cqe = mlx5e_rx_is_linear_skb(mdev, params) ?
+			mlx5e_skb_from_cqe_linear :
+			mlx5e_skb_from_cqe_nonlinear;
 		rq->mkey_be = c->mkey_be;
 	}
 
 	/* Create a page_pool and register it with rxq */
-	pp_params.order     = rq->buff.page_order;
+	pp_params.order     = 0;
 	pp_params.flags     = 0; /* No-internal DMA mapping in page_pool */
 	pp_params.pool_size = pool_size;
 	pp_params.nid       = cpu_to_node(c->cpu);
@@ -555,21 +608,21 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 	 */
 	rq->page_pool = page_pool_create(&pp_params);
 	if (IS_ERR(rq->page_pool)) {
-		if (rq->wq_type != MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ)
-			kfree(rq->wqe.frag_info);
 		err = PTR_ERR(rq->page_pool);
 		rq->page_pool = NULL;
-		goto err_rq_wq_destroy;
+		goto err_free;
 	}
 	err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq,
 					 MEM_TYPE_PAGE_POOL, rq->page_pool);
 	if (err)
-		goto err_rq_wq_destroy;
+		goto err_free;
 
 	for (i = 0; i < wq_sz; i++) {
 		if (rq->wq_type == MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ) {
 			struct mlx5e_rx_wqe_ll *wqe =
 				mlx5_wq_ll_get_wqe(&rq->mpwqe.wq, i);
+			u32 byte_count =
+				rq->mpwqe.num_strides << rq->mpwqe.log_stride_sz;
 			u64 dma_offset = mlx5e_get_mpwqe_offset(rq, i);
 
 			wqe->data[0].addr = cpu_to_be64(dma_offset + rq->buff.headroom);
@@ -578,9 +631,21 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 		} else {
 			struct mlx5e_rx_wqe_cyc *wqe =
 				mlx5_wq_cyc_get_wqe(&rq->wqe.wq, i);
+			int f;
 
-			wqe->data[0].byte_count = cpu_to_be32(byte_count);
-			wqe->data[0].lkey = rq->mkey_be;
+			for (f = 0; f < rq->wqe.info.num_frags; f++) {
+				u32 frag_size = rq->wqe.info.arr[f].frag_size |
+					MLX5_HW_START_PADDING;
+
+				wqe->data[f].byte_count = cpu_to_be32(frag_size);
+				wqe->data[f].lkey = rq->mkey_be;
+			}
+			/* check if num_frags is not a pow of two */
+			if (rq->wqe.info.num_frags < (1 << rq->wqe.info.log_num_frags)) {
+				wqe->data[f].byte_count = 0;
+				wqe->data[f].lkey = cpu_to_be32(MLX5_INVALID_LKEY);
+				wqe->data[f].addr = 0;
+			}
 		}
 	}
 
@@ -600,8 +665,16 @@ static int mlx5e_alloc_rq(struct mlx5e_channel *c,
 
 	return 0;
 
-err_destroy_umr_mkey:
-	mlx5_core_destroy_mkey(mdev, &rq->umr_mkey);
+err_free:
+	switch (rq->wq_type) {
+	case MLX5_WQ_TYPE_LINKED_LIST_STRIDING_RQ:
+		kfree(rq->mpwqe.info);
+		mlx5_core_destroy_mkey(mdev, &rq->umr_mkey);
+		break;
+	default: /* MLX5_WQ_TYPE_CYCLIC */
+		kvfree(rq->wqe.frags);
+		mlx5e_free_di_list(rq);
+	}
 
 err_rq_wq_destroy:
 	if (rq->xdp_prog)
@@ -631,7 +704,8 @@ static void mlx5e_free_rq(struct mlx5e_rq *rq)
 		mlx5_core_destroy_mkey(rq->mdev, &rq->umr_mkey);
 		break;
 	default: /* MLX5_WQ_TYPE_CYCLIC */
-		kfree(rq->wqe.frag_info);
+		kvfree(rq->wqe.frags);
+		mlx5e_free_di_list(rq);
 	}
 
 	for (i = rq->page_cache.head; i != rq->page_cache.tail;
@@ -823,17 +897,8 @@ static void mlx5e_free_rx_descs(struct mlx5e_rq *rq)
 			rq->dealloc_wqe(rq, wqe_ix);
 			mlx5_wq_cyc_pop(wq);
 		}
-
-		/* Clean outstanding pages on handled WQEs that decided to do page-reuse,
-		 * but yet to be re-posted.
-		 */
-		if (rq->wqe.page_reuse) {
-			int wq_sz = mlx5_wq_cyc_get_size(wq);
-
-			for (wqe_ix = 0; wqe_ix < wq_sz; wqe_ix++)
-				rq->dealloc_wqe(rq, wqe_ix);
-		}
 	}
+
 }
 
 static int mlx5e_open_rq(struct mlx5e_channel *c,
@@ -1954,6 +2019,61 @@ static void mlx5e_close_channel(struct mlx5e_channel *c)
 	kfree(c);
 }
 
+#define DEFAULT_FRAG_SIZE (2048)
+
+static void mlx5e_build_rq_frags_info(struct mlx5_core_dev *mdev,
+				      struct mlx5e_params *params,
+				      struct mlx5e_rq_frags_info *info)
+{
+	u32 byte_count = MLX5E_SW2HW_MTU(params, params->sw_mtu);
+	int frag_size_max = DEFAULT_FRAG_SIZE;
+	u32 buf_size = 0;
+	int i;
+
+#ifdef CONFIG_MLX5_EN_IPSEC
+	if (MLX5_IPSEC_DEV(mdev))
+		byte_count += MLX5E_METADATA_ETHER_LEN;
+#endif
+
+	if (mlx5e_rx_is_linear_skb(mdev, params)) {
+		int frag_stride;
+
+		frag_stride = mlx5e_rx_get_linear_frag_sz(params);
+		frag_stride = roundup_pow_of_two(frag_stride);
+
+		info->arr[0].frag_size = byte_count;
+		info->arr[0].frag_stride = frag_stride;
+		info->num_frags = 1;
+		info->wqe_bulk = PAGE_SIZE / frag_stride;
+		goto out;
+	}
+
+	if (byte_count > PAGE_SIZE +
+	    (MLX5E_MAX_RX_FRAGS - 1) * frag_size_max)
+		frag_size_max = PAGE_SIZE;
+
+	i = 0;
+	while (buf_size < byte_count) {
+		int frag_size = byte_count - buf_size;
+
+		if (i < MLX5E_MAX_RX_FRAGS - 1)
+			frag_size = min(frag_size, frag_size_max);
+
+		info->arr[i].frag_size = frag_size;
+		info->arr[i].frag_stride = roundup_pow_of_two(frag_size);
+
+		buf_size += frag_size;
+		i++;
+	}
+	info->num_frags = i;
+	/* number of different wqes sharing a page */
+	info->wqe_bulk = 1 + (info->num_frags % 2);
+
+out:
+	info->wqe_bulk = max_t(u8, info->wqe_bulk, 8);
+	info->log_num_frags = order_base_2(info->num_frags);
+}
+
 static inline u8 mlx5e_get_rqwq_log_stride(u8 wq_type, int ndsegs)
 {
 	int sz = sizeof(struct mlx5_wqe_data_seg) * ndsegs;
@@ -1990,6 +2110,8 @@ static void mlx5e_build_rq_param(struct mlx5e_priv *priv,
 		break;
 	default: /* MLX5_WQ_TYPE_CYCLIC */
 		MLX5_SET(wq, wq, log_wq_sz, params->log_rq_mtu_frames);
+		mlx5e_build_rq_frags_info(mdev, params, &param->frags_info);
+		ndsegs = param->frags_info.num_frags;
 	}
 
 	MLX5_SET(wq, wq, wq_type,          params->rq_wq_type);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index 3cdf2c097356..d3a1dd20e41d 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -164,8 +164,6 @@ static inline u32 mlx5e_decompress_cqes_start(struct mlx5e_rq *rq,
 	return mlx5e_decompress_cqes_cont(rq, cq, 1, budget_rem) - 1;
 }
 
-#define RQ_PAGE_SIZE(rq) ((1 << rq->buff.page_order) << PAGE_SHIFT)
-
 static inline bool mlx5e_page_is_reserved(struct page *page)
 {
 	return page_is_pfmemalloc(page) || page_to_nid(page) != numa_mem_id();
@@ -214,7 +212,7 @@ static inline bool mlx5e_rx_cache_get(struct mlx5e_rq *rq,
 	stats->cache_reuse++;
 
 	dma_sync_single_for_device(rq->pdev, dma_info->addr,
-				   RQ_PAGE_SIZE(rq),
+				   PAGE_SIZE,
 				   DMA_FROM_DEVICE);
 	return true;
 }
@@ -230,7 +228,7 @@ static inline int mlx5e_page_alloc_mapped(struct mlx5e_rq *rq,
 		return -ENOMEM;
 
 	dma_info->addr = dma_map_page(rq->pdev, dma_info->page, 0,
-				      RQ_PAGE_SIZE(rq), rq->buff.map_dir);
+				      PAGE_SIZE, rq->buff.map_dir);
 	if (unlikely(dma_mapping_error(rq->pdev, dma_info->addr))) {
 		put_page(dma_info->page);
 		dma_info->page = NULL;
@@ -243,8 +241,7 @@ static inline int mlx5e_page_alloc_mapped(struct mlx5e_rq *rq,
 static void mlx5e_page_dma_unmap(struct mlx5e_rq *rq,
 					struct mlx5e_dma_info *dma_info)
 {
-	dma_unmap_page(rq->pdev, dma_info->addr, RQ_PAGE_SIZE(rq),
-		       rq->buff.map_dir);
+	dma_unmap_page(rq->pdev, dma_info->addr, PAGE_SIZE, rq->buff.map_dir);
 }
 
 void mlx5e_page_release(struct mlx5e_rq *rq, struct mlx5e_dma_info *dma_info,
@@ -262,58 +259,96 @@ void mlx5e_page_release(struct mlx5e_rq *rq, struct mlx5e_dma_info *dma_info,
 	}
 }
 
-static inline bool mlx5e_page_reuse(struct mlx5e_rq *rq,
-				    struct mlx5e_wqe_frag_info *wi)
+static inline int mlx5e_get_rx_frag(struct mlx5e_rq *rq,
+				    struct mlx5e_wqe_frag_info *frag)
+{
+	int err = 0;
+
+	if (!frag->offset)
+		/* On first frag (offset == 0), replenish page (dma_info actually).
+		 * Other frags that point to the same dma_info (with a different
+		 * offset) should just use the new one without replenishing again
+		 * by themselves.
+		 */
+		err = mlx5e_page_alloc_mapped(rq, frag->di);
+
+	return err;
+}
+
+static inline void mlx5e_put_rx_frag(struct mlx5e_rq *rq,
+				     struct mlx5e_wqe_frag_info *frag)
 {
-	return rq->wqe.page_reuse && wi->di.page &&
-		(wi->offset + rq->wqe.frag_sz <= RQ_PAGE_SIZE(rq)) &&
-		!mlx5e_page_is_reserved(wi->di.page);
+	if (frag->last_in_page)
+		mlx5e_page_release(rq, frag->di, true);
 }
 
 static inline struct mlx5e_wqe_frag_info *get_frag(struct mlx5e_rq *rq, u16 ix)
 {
-	return &rq->wqe.frag_info[ix];
+	return &rq->wqe.frags[ix << rq->wqe.info.log_num_frags];
 }
 
-static int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe_cyc *wqe, u16 ix)
+static int mlx5e_alloc_rx_wqe(struct mlx5e_rq *rq, struct mlx5e_rx_wqe_cyc *wqe,
+			      u16 ix)
 {
-	struct mlx5e_wqe_frag_info *wi = &rq->wqe.frag_info[ix];
+	struct mlx5e_wqe_frag_info *frag = get_frag(rq, ix);
+	int err;
+	int i;
 
-	/* check if page exists, hence can be reused */
-	if (!wi->di.page) {
-		if (unlikely(mlx5e_page_alloc_mapped(rq, &wi->di)))
-			return -ENOMEM;
-		wi->offset = 0;
+	for (i = 0; i < rq->wqe.info.num_frags; i++, frag++) {
+		err = mlx5e_get_rx_frag(rq, frag);
+		if (unlikely(err))
+			goto free_frags;
+
+		wqe->data[i].addr = cpu_to_be64(frag->di->addr +
+						frag->offset + rq->buff.headroom);
 	}
 
-	wqe->data[0].addr = cpu_to_be64(wi->di.addr + wi->offset + rq->buff.headroom);
 	return 0;
+
+free_frags:
+	while (--i >= 0)
+		mlx5e_put_rx_frag(rq, --frag);
+
+	return err;
 }
 
 static inline void mlx5e_free_rx_wqe(struct mlx5e_rq *rq,
 				     struct mlx5e_wqe_frag_info *wi)
 {
-	mlx5e_page_release(rq, &wi->di, true);
-	wi->di.page = NULL;
+	int i;
+
+	for (i = 0; i < rq->wqe.info.num_frags; i++, wi++)
+		mlx5e_put_rx_frag(rq, wi);
 }
 
-static inline void mlx5e_free_rx_wqe_reuse(struct mlx5e_rq *rq,
-					   struct mlx5e_wqe_frag_info *wi)
+void mlx5e_dealloc_rx_wqe(struct mlx5e_rq *rq, u16 ix)
 {
-	if (mlx5e_page_reuse(rq, wi)) {
-		rq->stats->page_reuse++;
-		return;
-	}
+	struct mlx5e_wqe_frag_info *wi = get_frag(rq, ix);
 
 	mlx5e_free_rx_wqe(rq, wi);
 }
 
-void mlx5e_dealloc_rx_wqe(struct mlx5e_rq *rq, u16 ix)
+static int mlx5e_alloc_rx_wqes(struct mlx5e_rq *rq, u16 ix, u8 wqe_bulk)
 {
-	struct mlx5e_wqe_frag_info *wi = &rq->wqe.frag_info[ix];
+	struct mlx5_wq_cyc *wq = &rq->wqe.wq;
+	int err;
+	int i;
 
-	if (wi->di.page)
-		mlx5e_free_rx_wqe(rq, wi);
+	for (i = 0; i < wqe_bulk; i++) {
+		struct mlx5e_rx_wqe_cyc *wqe = mlx5_wq_cyc_get_wqe(wq, ix + i);
+
+		err = mlx5e_alloc_rx_wqe(rq, wqe, ix + i);
+		if (unlikely(err))
+			goto free_wqes;
+	}
+
+	return 0;
+
+free_wqes:
+	while (--i >= 0)
+		mlx5e_dealloc_rx_wqe(rq, ix + i);
+
+	return err;
 }
 
 static inline void
@@ -476,26 +511,28 @@ void mlx5e_dealloc_rx_mpwqe(struct mlx5e_rq *rq, u16 ix)
 bool mlx5e_post_rx_wqes(struct mlx5e_rq *rq)
 {
 	struct mlx5_wq_cyc *wq = &rq->wqe.wq;
+	u8 wqe_bulk;
 	int err;
 
 	if (unlikely(!test_bit(MLX5E_RQ_STATE_ENABLED, &rq->state)))
 		return false;
 
-	if (mlx5_wq_cyc_is_full(wq))
+	wqe_bulk = rq->wqe.info.wqe_bulk;
+
+	if (mlx5_wq_cyc_missing(wq) < wqe_bulk)
 		return false;
 
 	do {
 		u16 head = mlx5_wq_cyc_get_head(wq);
-		struct mlx5e_rx_wqe_cyc *wqe = mlx5_wq_cyc_get_wqe(wq, head);
 
-		err = mlx5e_alloc_rx_wqe(rq, wqe, head);
+		err = mlx5e_alloc_rx_wqes(rq, head, wqe_bulk);
 		if (unlikely(err)) {
 			rq->stats->buff_alloc_err++;
 			break;
 		}
 
-		mlx5_wq_cyc_push(wq);
-	} while (!mlx5_wq_cyc_is_full(wq));
+		mlx5_wq_cyc_push_n(wq, wqe_bulk);
+	} while (mlx5_wq_cyc_missing(wq) >= wqe_bulk);
 
 	/* ensure wqes are visible to device before updating doorbell record */
 	dma_wmb();
@@ -949,11 +986,11 @@ struct sk_buff *mlx5e_build_linear_skb(struct mlx5e_rq *rq, void *va,
 	return skb;
 }
 
-static inline
-struct sk_buff *skb_from_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe,
-			     struct mlx5e_wqe_frag_info *wi, u32 cqe_bcnt)
+struct sk_buff *
+mlx5e_skb_from_cqe_linear(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe,
+			  struct mlx5e_wqe_frag_info *wi, u32 cqe_bcnt)
 {
-	struct mlx5e_dma_info *di = &wi->di;
+	struct mlx5e_dma_info *di = wi->di;
 	u16 rx_headroom = rq->buff.headroom;
 	struct sk_buff *skb;
 	void *va, *data;
@@ -968,7 +1005,6 @@ struct sk_buff *skb_from_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe,
 				      frag_size, DMA_FROM_DEVICE);
 	prefetchw(va); /* xdp_frame data area */
 	prefetch(data);
-	wi->offset += frag_size;
 
 	if (unlikely((cqe->op_own >> 4) != MLX5_CQE_RESP_SEND)) {
 		rq->stats->wqe_err++;
@@ -991,6 +1027,56 @@ struct sk_buff *skb_from_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe,
 	return skb;
 }
 
+struct sk_buff *
+mlx5e_skb_from_cqe_nonlinear(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe,
+			     struct mlx5e_wqe_frag_info *wi, u32 cqe_bcnt)
+{
+	struct mlx5e_rq_frag_info *frag_info = &rq->wqe.info.arr[0];
+	struct mlx5e_wqe_frag_info *head_wi = wi;
+	u16 headlen      = min_t(u32, MLX5E_RX_MAX_HEAD, cqe_bcnt);
+	u16 frag_headlen = headlen;
+	u16 byte_cnt     = cqe_bcnt - headlen;
+	struct sk_buff *skb;
+
+	if (unlikely((cqe->op_own >> 4) != MLX5_CQE_RESP_SEND)) {
+		rq->stats->wqe_err++;
+		return NULL;
+	}
+
+	/* XDP is not supported in this configuration, as incoming packets
+	 * might spread among multiple pages.
+	 */
+	skb = napi_alloc_skb(rq->cq.napi,
+			     ALIGN(MLX5E_RX_MAX_HEAD, sizeof(long)));
+	if (unlikely(!skb)) {
+		rq->stats->buff_alloc_err++;
+		return NULL;
+	}
+
+	prefetchw(skb->data);
+
+	while (byte_cnt) {
+		u16 frag_consumed_bytes =
+			min_t(u16, frag_info->frag_size - frag_headlen, byte_cnt);
+
+		mlx5e_add_skb_frag(rq, skb, wi->di, wi->offset + frag_headlen,
+				   frag_consumed_bytes, frag_info->frag_stride);
+		byte_cnt -= frag_consumed_bytes;
+		frag_headlen = 0;
+		frag_info++;
+		wi++;
+	}
+
+	/* copy header */
+	mlx5e_copy_skb_header(rq->pdev, skb, head_wi->di, head_wi->offset,
+			      0, headlen);
+	/* skb linear part was allocated with headlen and aligned to long */
+	skb->tail += headlen;
+	skb->len  += headlen;
+
+	return skb;
+}
+
 void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 {
 	struct mlx5_wq_cyc *wq = &rq->wqe.wq;
@@ -1003,23 +1089,23 @@ void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	wi       = get_frag(rq, ci);
 	cqe_bcnt = be32_to_cpu(cqe->byte_cnt);
 
-	skb = skb_from_cqe(rq, cqe, wi, cqe_bcnt);
+	skb = rq->wqe.skb_from_cqe(rq, cqe, wi, cqe_bcnt);
 	if (!skb) {
 		/* probably for XDP */
 		if (__test_and_clear_bit(MLX5E_RQ_FLAG_XDP_XMIT, rq->flags)) {
-			wi->di.page = NULL;
-			/* do not return page to cache, it will be returned on XDP_TX completion */
+			/* do not return page to cache,
+			 * it will be returned on XDP_TX completion.
+			 */
 			goto wq_cyc_pop;
 		}
-		/* probably an XDP_DROP, save the page-reuse checks */
-		mlx5e_free_rx_wqe(rq, wi);
-		goto wq_cyc_pop;
+		goto free_wqe;
 	}
 
 	mlx5e_complete_rx_cqe(rq, cqe, cqe_bcnt, skb);
 	napi_gro_receive(rq->cq.napi, skb);
 
-	mlx5e_free_rx_wqe_reuse(rq, wi);
+free_wqe:
+	mlx5e_free_rx_wqe(rq, wi);
 wq_cyc_pop:
 	mlx5_wq_cyc_pop(wq);
 }
@@ -1041,16 +1127,16 @@ void mlx5e_handle_rx_cqe_rep(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	wi       = get_frag(rq, ci);
 	cqe_bcnt = be32_to_cpu(cqe->byte_cnt);
 
-	skb = skb_from_cqe(rq, cqe, wi, cqe_bcnt);
+	skb = rq->wqe.skb_from_cqe(rq, cqe, wi, cqe_bcnt);
 	if (!skb) {
+		/* probably for XDP */
 		if (__test_and_clear_bit(MLX5E_RQ_FLAG_XDP_XMIT, rq->flags)) {
-			wi->di.page = NULL;
-			/* do not return page to cache, it will be returned on XDP_TX completion */
+			/* do not return page to cache,
+			 * it will be returned on XDP_TX completion.
+			 */
 			goto wq_cyc_pop;
 		}
-		/* probably an XDP_DROP, save the page-reuse checks */
-		mlx5e_free_rx_wqe(rq, wi);
-		goto wq_cyc_pop;
+		goto free_wqe;
 	}
 
 	mlx5e_complete_rx_cqe(rq, cqe, cqe_bcnt, skb);
@@ -1060,7 +1146,8 @@ void mlx5e_handle_rx_cqe_rep(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 
 	napi_gro_receive(rq->cq.napi, skb);
 
-	mlx5e_free_rx_wqe_reuse(rq, wi);
+free_wqe:
+	mlx5e_free_rx_wqe(rq, wi);
 wq_cyc_pop:
 	mlx5_wq_cyc_pop(wq);
 }
@@ -1409,7 +1496,7 @@ void mlx5i_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	wi       = get_frag(rq, ci);
 	cqe_bcnt = be32_to_cpu(cqe->byte_cnt);
 
-	skb = skb_from_cqe(rq, cqe, wi, cqe_bcnt);
+	skb = rq->wqe.skb_from_cqe(rq, cqe, wi, cqe_bcnt);
 	if (!skb)
 		goto wq_free_wqe;
 
@@ -1421,7 +1508,7 @@ void mlx5i_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	napi_gro_receive(rq->cq.napi, skb);
 
 wq_free_wqe:
-	mlx5e_free_rx_wqe_reuse(rq, wi);
+	mlx5e_free_rx_wqe(rq, wi);
 	mlx5_wq_cyc_pop(wq);
 }
 
@@ -1441,7 +1528,7 @@ void mlx5e_ipsec_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	wi       = get_frag(rq, ci);
 	cqe_bcnt = be32_to_cpu(cqe->byte_cnt);
 
-	skb = skb_from_cqe(rq, cqe, wi, cqe_bcnt);
+	skb = rq->wqe.skb_from_cqe(rq, cqe, wi, cqe_bcnt);
 	if (unlikely(!skb)) {
 		/* a DROP, save the page-reuse checks */
 		mlx5e_free_rx_wqe(rq, wi);
@@ -1456,7 +1543,7 @@ void mlx5e_ipsec_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe)
 	mlx5e_complete_rx_cqe(rq, cqe, cqe_bcnt, skb);
 	napi_gro_receive(rq->cq.napi, skb);
 
-	mlx5e_free_rx_wqe_reuse(rq, wi);
+	mlx5e_free_rx_wqe(rq, wi);
 wq_cyc_pop:
 	mlx5_wq_cyc_pop(wq);
 }
-- 
2.17.0

[net-next 16/17] net/mlx5e: RX, Always prefer Linear SKB configuration

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

Prefer the linear SKB configuration of Legacy RQ over the
non-linear one of Striding RQ.

This implies that ConnectX-4 LX now uses legacy RQ by default,
as it does not support the linear configuration of Striding RQ.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index 2c634e50d051..333d4ed52b94 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -4405,9 +4405,16 @@ void mlx5e_build_nic_params(struct mlx5_core_dev *mdev,
 	MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_CQE_COMPRESS, params->rx_cqe_compress_def);
 
 	/* RQ */
-	if (mlx5e_striding_rq_possible(mdev, params))
-		MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ,
-				!slow_pci_heuristic(mdev));
+	/* Prefer Striding RQ, unless any of the following holds:
+	 * - Striding RQ configuration is not possible/supported.
+	 * - Slow PCI heuristic.
+	 * - Legacy RQ would use linear SKB while Striding RQ would use non-linear.
+	 */
+	if (!slow_pci_heuristic(mdev) &&
+	    mlx5e_striding_rq_possible(mdev, params) &&
+	    (mlx5e_rx_mpwqe_is_linear_skb(mdev, params) ||
+	     !mlx5e_rx_is_linear_skb(mdev, params)))
+		MLX5E_SET_PFLAG(params, MLX5E_PFLAG_RX_STRIDING_RQ, true);
 	mlx5e_set_rq_type(mdev, params);
 	mlx5e_init_rq_type_params(mdev, params);
 
-- 
2.17.0

[net-next 17/17] net/mlx5e: TX, Separate cachelines of xmit and completion stats

[Saeed Mahameed]

From: Tariq Toukan <tariqt@mellanox.com>

Avoid false sharing of cachelines by separating the cachelines of
TX stats that are dertied in xmit flow and in completion flow.

Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
---
 drivers/net/ethernet/mellanox/mlx5/core/en_stats.c | 8 ++++----
 drivers/net/ethernet/mellanox/mlx5/core/en_stats.h | 9 +++++----
 2 files changed, 9 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c b/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
index 697dc7397ba2..1646859974ce 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_stats.c
@@ -64,11 +64,11 @@ static const struct counter_desc sw_stats_desc[] = {
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_csum_partial) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_csum_partial_inner) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_queue_stopped) },
-	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_queue_wake) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_queue_dropped) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_xmit_more) },
-	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_cqe_err) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_recover) },
+	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_queue_wake) },
+	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, tx_cqe_err) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, rx_wqe_err) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, rx_mpwqe_filler) },
 	{ MLX5E_DECLARE_STAT(struct mlx5e_sw_stats, rx_buff_alloc_err) },
@@ -1137,11 +1137,11 @@ static const struct counter_desc sq_stats_desc[] = {
 	{ MLX5E_DECLARE_TX_STAT(struct mlx5e_sq_stats, nop) },
 	{ MLX5E_DECLARE_TX_STAT(struct mlx5e_sq_stats, csum_none) },
 	{ MLX5E_DECLARE_TX_STAT(struct mlx5e_sq_stats, stopped) },
-	{ MLX5E_DECLARE_TX_STAT(struct mlx5e_sq_stats, wake) },
 	{ MLX5E_DECLARE_TX_STAT(struct mlx5e_sq_stats, dropped) },
 	{ MLX5E_DECLARE_TX_STAT(struct mlx5e_sq_stats, xmit_more) },
-	{ MLX5E_DECLARE_TX_STAT(struct mlx5e_sq_stats, cqe_err) },
 	{ MLX5E_DECLARE_TX_STAT(struct mlx5e_sq_stats, recover) },
+	{ MLX5E_DECLARE_TX_STAT(struct mlx5e_sq_stats, wake) },
+	{ MLX5E_DECLARE_TX_STAT(struct mlx5e_sq_stats, cqe_err) },
 };
 
 static const struct counter_desc ch_stats_desc[] = {
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_stats.h b/drivers/net/ethernet/mellanox/mlx5/core/en_stats.h
index 390c7afa5188..643153bb3607 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_stats.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_stats.h
@@ -75,11 +75,11 @@ struct mlx5e_sw_stats {
 	u64 tx_csum_partial;
 	u64 tx_csum_partial_inner;
 	u64 tx_queue_stopped;
-	u64 tx_queue_wake;
 	u64 tx_queue_dropped;
 	u64 tx_xmit_more;
-	u64 tx_cqe_err;
 	u64 tx_recover;
+	u64 tx_queue_wake;
+	u64 tx_cqe_err;
 	u64 rx_wqe_err;
 	u64 rx_mpwqe_filler;
 	u64 rx_buff_alloc_err;
@@ -203,10 +203,11 @@ struct mlx5e_sq_stats {
 	/* less likely accessed in data path */
 	u64 csum_none;
 	u64 stopped;
-	u64 wake;
 	u64 dropped;
-	u64 cqe_err;
 	u64 recover;
+	/* dirtied @completion */
+	u64 wake ____cacheline_aligned_in_smp;
+	u64 cqe_err;
 };
 
 struct mlx5e_ch_stats {
-- 
2.17.0

Re: [pull request][net-next 00/17] Mellanox, mlx5e updates 2018-06-01

[David Miller]

From: Saeed Mahameed <saeedm@mellanox.com>
Date: Fri,  1 Jun 2018 17:05:27 -0700

> Sorry for the extra 2 patches in this series, but mostly the series
> contains small patches and some fixes to previous patches in this
> submission window, with one main patch from Tariq to improve legacy
> RQ buffer management, for more information please refer to that tag
> log below.
> 
> Please pull and let me know if there's any problem.

Pulled, thanks Saeed.