Skip to contentSkip to conversation

net/mlx5: misc changes 2025-07-09

PATCH Applied Tariq Toukan 4 participants 13 messages Jul 8, 2025 – Jul 10, 2025

Subsystems (5)
lore
Share & tools
lore

Tariq Toukan

author maintainer
#0 lore raw

cover net/mlx5: misc changes 2025-07-09

Hi,

This series contains misc enhancements to the mlx5 driver.

Regards,
Tariq

Carolina Jubran (1):
net/mlx5e: Remove unused VLAN insertion logic in TX path

Cosmin Ratiu (1):
net/mlx5e: CT: extract a memcmp from a spinlock section

Gal Pressman (1):
net/mlx5e: Replace recursive VLAN push handling with an iterative loop

Maor Gottlieb (1):
net/mlx5: Warn when write combining is not supported

Tariq Toukan (1):
net/mlx5e: RX, Remove unnecessary RQT redirects

 drivers/net/ethernet/mellanox/mlx5/core/en.h  |  1 -
 .../net/ethernet/mellanox/mlx5/core/en/ptp.c  |  2 -
 .../mellanox/mlx5/core/en/reporter_tx.c       |  1 -
 .../ethernet/mellanox/mlx5/core/en/rx_res.c   |  4 +-
 .../mellanox/mlx5/core/en/tc/act/vlan.c       | 43 ++++++++++---------
 .../ethernet/mellanox/mlx5/core/en/tc_ct.c    |  5 ++-
 .../net/ethernet/mellanox/mlx5/core/en_main.c |  2 -
 .../net/ethernet/mellanox/mlx5/core/en_tx.c   |  9 +---
 drivers/net/ethernet/mellanox/mlx5/core/wc.c  |  3 ++
 9 files changed, 31 insertions(+), 39 deletions(-)

base-commit: 19c066f940666bf6c0982635e4441100ca8d75bc

Signature · 1 line
2.31.1
author maintainer
#1 lore raw

PATCH 1/5 net/mlx5e: Remove unused VLAN insertion logic in TX path

From: Carolina Jubran <[email protected]>

The VLAN insertion capability (wqe_vlan_insert) was never enabled on all mlx5 devices. When VLAN TX offload is advertised but this capability is not supported, the driver uses inline headers to insert the VLAN tag.

To support this, the driver used to set the MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE bit to enforce L2 inline mode when wqe_vlan_insert was not supported. Since the capability is disabled on all devices, this logic was always active, and the SQ flag has become redundant. L2 inline is enforced unconditionally for VLAN-tagged packets.

The skb_vlan_tag_present() check in the else-if block of mlx5e_sq_xmit_wqe() is never true by this point in the TX flow, as the VLAN tag has already been inserted by the driver using inline headers. As a result, this code is never executed.

Remove the redundant SQ state, dead VLAN insertion code block, and related logic.

---
 drivers/net/ethernet/mellanox/mlx5/core/en.h             | 1 -
 drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c         | 2 --
 drivers/net/ethernet/mellanox/mlx5/core/en/reporter_tx.c | 1 -
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c        | 2 --
 drivers/net/ethernet/mellanox/mlx5/core/en_tx.c          | 9 +--------
 5 files changed, 1 insertion(+), 14 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en.h +0 -1
index 65a73913b9a2..64e69e616b1f 100644
@@ -383,7 +383,6 @@ enum {
 	MLX5E_SQ_STATE_RECOVERING,
 	MLX5E_SQ_STATE_IPSEC,
 	MLX5E_SQ_STATE_DIM,
-	MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE,
 	MLX5E_SQ_STATE_PENDING_XSK_TX,
 	MLX5E_SQ_STATE_PENDING_TLS_RX_RESYNC,
 	MLX5E_NUM_SQ_STATES, /* Must be kept last */
drivers/net/ethernet/mellanox/mlx5/core/en/ptp.c +0 -2
index 5d0014129a7e..391b4e9c9dc4 100644
@@ -340,8 +340,6 @@ static int mlx5e_ptp_alloc_txqsq(struct mlx5e_ptp *c, int txq_ix,
 	sq->stats     = &c->priv->ptp_stats.sq[tc];
 	sq->ptpsq     = ptpsq;
 	INIT_WORK(&sq->recover_work, mlx5e_tx_err_cqe_work);
-	if (!MLX5_CAP_ETH(mdev, wqe_vlan_insert))
-		set_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state);
 	sq->stop_room = param->stop_room;
 	sq->ptp_cyc2time = mlx5_sq_ts_translator(mdev);
 
drivers/net/ethernet/mellanox/mlx5/core/en/reporter_tx.c +0 -1
index c3bda4612fa9..bd96988e102c 100644
@@ -13,7 +13,6 @@ static const char * const sq_sw_state_type_name[] = {
 	[MLX5E_SQ_STATE_RECOVERING] = "recovering",
 	[MLX5E_SQ_STATE_IPSEC] = "ipsec",
 	[MLX5E_SQ_STATE_DIM] = "dim",
-	[MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE] = "vlan_need_l2_inline",
 	[MLX5E_SQ_STATE_PENDING_XSK_TX] = "pending_xsk_tx",
 	[MLX5E_SQ_STATE_PENDING_TLS_RX_RESYNC] = "pending_tls_rx_resync",
 };
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +0 -2
index e8e5b347f9b2..fee323ade522 100644
@@ -1677,8 +1677,6 @@ static int mlx5e_alloc_txqsq(struct mlx5e_channel *c,
 	sq->hw_mtu    = MLX5E_SW2HW_MTU(params, params->sw_mtu);
 	sq->max_sq_mpw_wqebbs = mlx5e_get_max_sq_aligned_wqebbs(mdev);
 	INIT_WORK(&sq->recover_work, mlx5e_tx_err_cqe_work);
-	if (!MLX5_CAP_ETH(mdev, wqe_vlan_insert))
-		set_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state);
 	if (mlx5_ipsec_device_caps(c->priv->mdev))
 		set_bit(MLX5E_SQ_STATE_IPSEC, &sq->state);
 	if (param->is_mpw)
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +1 -8
index 55a8629f0792..e6a301ba3254 100644
@@ -256,8 +256,7 @@ mlx5e_tx_wqe_inline_mode(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 
 	mode = sq->min_inline_mode;
 
-	if (skb_vlan_tag_present(skb) &&
-	    test_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state))
+	if (skb_vlan_tag_present(skb))
 		mode = max_t(u8, MLX5_INLINE_MODE_L2, mode);
 
 	return mode;
@@ -483,12 +482,6 @@ mlx5e_sq_xmit_wqe(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 		}
 		eseg->inline_hdr.sz |= cpu_to_be16(ihs);
 		dseg += wqe_attr->ds_cnt_inl;
-	} else if (skb_vlan_tag_present(skb)) {
-		eseg->insert.type = cpu_to_be16(MLX5_ETH_WQE_INSERT_VLAN);
-		if (skb->vlan_proto == cpu_to_be16(ETH_P_8021AD))
-			eseg->insert.type |= cpu_to_be16(MLX5_ETH_WQE_SVLAN);
-		eseg->insert.vlan_tci = cpu_to_be16(skb_vlan_tag_get(skb));
-		stats->added_vlan_packets++;
 	}
 
 	dseg += wqe_attr->ds_cnt_ids;
Signature · 1 line
2.31.1
reviewer Reviewed-by on v1 1/5
#2 lore raw

Re: [PATCH net-next 1/5] net/mlx5e: Remove unused VLAN insertion logic in TX path

Quoted 19 lines · Tariq Toukan

From: Carolina Jubran <[email protected]>

The VLAN insertion capability (wqe_vlan_insert) was never enabled on all mlx5 devices. When VLAN TX offload is advertised but this capability is not supported, the driver uses inline headers to insert the VLAN tag.

To support this, the driver used to set the MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE bit to enforce L2 inline mode when wqe_vlan_insert was not supported. Since the capability is disabled on all devices, this logic was always active, and the SQ flag has become redundant. L2 inline is enforced unconditionally for VLAN-tagged packets.

The skb_vlan_tag_present() check in the else-if block of mlx5e_sq_xmit_wqe() is never true by this point in the TX flow, as the VLAN tag has already been inserted by the driver using inline headers. As a result, this code is never executed.

Quoted 5 lines

Remove the redundant SQ state, dead VLAN insertion code block, and related logic.

author maintainer
#3 lore raw

PATCH 2/5 net/mlx5e: CT: extract a memcmp from a spinlock section

In reply to Tariq Toukan

From: Cosmin Ratiu <[email protected]>

This reduces the time the lock is held and reduces contention.

---
 drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/net/ethernet/mellanox/mlx5/core/en/tc_ct.c +4 -1
index 81332cd4a582..870d12364f99 100644
@@ -1195,6 +1195,7 @@ mlx5_tc_ct_block_flow_offload_add(struct mlx5_ct_ft *ft,
 	struct flow_action_entry *meta_action;
 	unsigned long cookie = flow->cookie;
 	struct mlx5_ct_entry *entry;
+	bool has_nat;
 	int err;
 
 	meta_action = mlx5_tc_ct_get_ct_metadata_action(flow_rule);
@@ -1236,6 +1237,8 @@ mlx5_tc_ct_block_flow_offload_add(struct mlx5_ct_ft *ft,
 	err = mlx5_tc_ct_rule_to_tuple_nat(&entry->tuple_nat, flow_rule);
 	if (err)
 		goto err_set;
+	has_nat = memcmp(&entry->tuple, &entry->tuple_nat,
+			 sizeof(entry->tuple));
 
 	spin_lock_bh(&ct_priv->ht_lock);
 
@@ -1244,7 +1247,7 @@ mlx5_tc_ct_block_flow_offload_add(struct mlx5_ct_ft *ft,
 	if (err)
 		goto err_entries;
 
-	if (memcmp(&entry->tuple, &entry->tuple_nat, sizeof(entry->tuple))) {
+	if (has_nat) {
 		err = rhashtable_lookup_insert_fast(&ct_priv->ct_tuples_nat_ht,
 						    &entry->tuple_nat_node,
 						    tuples_nat_ht_params);
Signature · 1 line
2.31.1
author maintainer
#5 lore raw

PATCH 3/5 net/mlx5e: Replace recursive VLAN push handling with an iterative loop

In reply to Tariq Toukan

From: Gal Pressman <[email protected]>

mlx5e_tc_act_vlan_add_push_action() uses tail-recursion to walk through a stack of VLAN devices.

There is no need for a complicated recursion with unnecessary stack consumption and less obvious code flow, rewrite the function so that it uses a do while loop instead.

---
 .../mellanox/mlx5/core/en/tc/act/vlan.c       | 43 ++++++++++---------
 1 file changed, 22 insertions(+), 21 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/vlan.c +22 -21
index a13c5e707b83..9bdb5820c553 100644
@@ -94,29 +94,30 @@ mlx5e_tc_act_vlan_add_push_action(struct mlx5e_priv *priv,
 				  struct net_device **out_dev,
 				  struct netlink_ext_ack *extack)
 {
-	struct net_device *vlan_dev = *out_dev;
-	struct flow_action_entry vlan_act = {
-		.id = FLOW_ACTION_VLAN_PUSH,
-		.vlan.vid = vlan_dev_vlan_id(vlan_dev),
-		.vlan.proto = vlan_dev_vlan_proto(vlan_dev),
-		.vlan.prio = 0,
-	};
-	int err;
-
-	err = parse_tc_vlan_action(priv, &vlan_act, attr->esw_attr, &attr->action, extack, NULL);
-	if (err)
-		return err;
-
-	rcu_read_lock();
-	*out_dev = dev_get_by_index_rcu(dev_net(vlan_dev), dev_get_iflink(vlan_dev));
-	rcu_read_unlock();
-	if (!*out_dev)
-		return -ENODEV;
+	do {
+		struct net_device *vlan_dev = *out_dev;
+		struct flow_action_entry vlan_act = {
+			.id = FLOW_ACTION_VLAN_PUSH,
+			.vlan.vid = vlan_dev_vlan_id(vlan_dev),
+			.vlan.proto = vlan_dev_vlan_proto(vlan_dev),
+			.vlan.prio = 0,
+		};
+		int err;
+
+		err = parse_tc_vlan_action(priv, &vlan_act, attr->esw_attr,
+					   &attr->action, extack, NULL);
+		if (err)
+			return err;
 
-	if (is_vlan_dev(*out_dev))
-		err = mlx5e_tc_act_vlan_add_push_action(priv, attr, out_dev, extack);
+		rcu_read_lock();
+		*out_dev = dev_get_by_index_rcu(dev_net(vlan_dev),
+						dev_get_iflink(vlan_dev));
+		rcu_read_unlock();
+		if (!*out_dev)
+			return -ENODEV;
+	} while (is_vlan_dev(*out_dev));
 
-	return err;
+	return 0;
 }
 
 int
Signature · 1 line
2.31.1
reviewer Reviewed-by on v1 3/5
#6 lore raw

Re: [PATCH net-next 3/5] net/mlx5e: Replace recursive VLAN push handling with an iterative loop

Quoted 5 lines · Tariq Toukan

From: Gal Pressman <[email protected]>

mlx5e_tc_act_vlan_add_push_action() uses tail-recursion to walk through a stack of VLAN devices.

Quoted 7 lines

There is no need for a complicated recursion with unnecessary stack consumption and less obvious code flow, rewrite the function so that it uses a do while loop instead.

maintainer
#7 lore raw

Re: [PATCH net-next 3/5] net/mlx5e: Replace recursive VLAN push handling with an iterative loop

In reply to Tariq Toukan

drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/vlan.c mlx5e_tc_act_vlan_add_push_action(struct mlx5e_priv *priv, · line 118

Quoted 7 lines · On Wed, 9 Jul 2025 00:16:25 +0300 Tariq Toukan wrote:
drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/vlan.c +7 -0
+		rcu_read_lock();
+		*out_dev = dev_get_by_index_rcu(dev_net(vlan_dev),
+						dev_get_iflink(vlan_dev));
+		rcu_read_unlock();
+		if (!*out_dev)
+			return -ENODEV;
+	} while (is_vlan_dev(*out_dev));

Would be good adding a comment here to explain why this odd rcu lock / lookup / rcu unlock / keep using the return value outside of rcu protection - code flow is correct :S

author maintainer
#8 lore raw

PATCH 4/5 net/mlx5: Warn when write combining is not supported

In reply to Tariq Toukan

From: Maor Gottlieb <[email protected]>

Warn if write combining is not supported, as it can impact latency. Add the warning message to be printed only when the driver actually run the test and detect unsupported state, rather than when inheriting parent's result for SFs.

---
 drivers/net/ethernet/mellanox/mlx5/core/wc.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/net/ethernet/mellanox/mlx5/core/wc.c +3 -0
index 740b719e7072..2f0316616fa4 100644
@@ -378,6 +378,9 @@ static void mlx5_core_test_wc(struct mlx5_core_dev *mdev)
 	mlx5_free_bfreg(mdev, &sq->bfreg);
 err_alloc_bfreg:
 	kfree(sq);
+
+	if (mdev->wc_state == MLX5_WC_STATE_UNSUPPORTED)
+		mlx5_core_warn(mdev, "Write combining is not supported\n");
 }
 
 bool mlx5_wc_support_get(struct mlx5_core_dev *mdev)
Signature · 1 line
2.31.1
reviewer Reviewed-by on v1 4/5
#9 lore raw

Re: [PATCH net-next 4/5] net/mlx5: Warn when write combining is not supported

Quoted 10 lines · Tariq Toukan

From: Maor Gottlieb <[email protected]>

Warn if write combining is not supported, as it can impact latency. Add the warning message to be printed only when the driver actually run the test and detect unsupported state, rather than when inheriting parent's result for SFs.

author maintainer
#10 lore raw

PATCH 5/5 net/mlx5e: RX, Remove unnecessary RQT redirects

In reply to Tariq Toukan

RQTs (Receive Queue Table) should redirect traffic to the channels' RQs when they're active. Otherwise, redirect to the designated "drop RQ".

RQTs are created in "inactive" state, pointing to the "drop RQ". In activate and de-activate flows, do not "deactivate" the rest of RQTs (beyond the num of channels), as they are already inactive.

This cuts down unnecessary execution of FW commands (MODIFY_RQT), and improves the latency of open/close channels or configuration change.

Perf:
NIC: Connect-X7.
Configuration: 1 combined channel, max num channels 248. Measure time for "interface up + interface down".

Before: 0.313 sec
After:  0.057 sec (5.5x faster)

247 MODIFY_RQT commands saved in interface up.
247 MODIFY_RQT commands saved in interface down.

---
 drivers/net/ethernet/mellanox/mlx5/core/en/rx_res.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en/rx_res.c +1 -3
index 5fcbe47337b0..b3fdb1afa1e6 100644
@@ -579,8 +579,6 @@ void mlx5e_rx_res_channels_activate(struct mlx5e_rx_res *res, struct mlx5e_chann
 
 	for (ix = 0; ix < nch; ix++)
 		mlx5e_rx_res_channel_activate_direct(res, chs, ix);
-	for (ix = nch; ix < res->max_nch; ix++)
-		mlx5e_rx_res_channel_deactivate_direct(res, ix);
 
 	if (res->features & MLX5E_RX_RES_FEATURE_PTP) {
 		u32 rqn;
@@ -603,7 +601,7 @@ void mlx5e_rx_res_channels_deactivate(struct mlx5e_rx_res *res)
 
 	mlx5e_rx_res_rss_disable(res);
 
-	for (ix = 0; ix < res->max_nch; ix++)
+	for (ix = 0; ix < res->rss_nch; ix++)
 		mlx5e_rx_res_channel_deactivate_direct(res, ix);
 
 	if (res->features & MLX5E_RX_RES_FEATURE_PTP) {
Signature · 1 line
2.31.1
reviewer Reviewed-by on v1 5/5
#11 lore raw

Re: [PATCH net-next 5/5] net/mlx5e: RX, Remove unnecessary RQT redirects

Quoted 15 lines · Tariq Toukan

RQTs (Receive Queue Table) should redirect traffic to the channels' RQs when they're active. Otherwise, redirect to the designated "drop RQ".

RQTs are created in "inactive" state, pointing to the "drop RQ". In activate and de-activate flows, do not "deactivate" the rest of RQTs (beyond the num of channels), as they are already inactive.

This cuts down unnecessary execution of FW commands (MODIFY_RQT), and improves the latency of open/close channels or configuration change.

Perf:
NIC: Connect-X7.
Configuration: 1 combined channel, max num channels 248. Measure time for "interface up + interface down".

Quoted 8 lines
Before: 0.313 sec
After:  0.057 sec (5.5x faster)

247 MODIFY_RQT commands saved in interface up.
247 MODIFY_RQT commands saved in interface down.

#12 lore raw
In reply to Tariq Toukan
This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:

Hello:

This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:

Quoted 8 lines · On Wed, 9 Jul 2025 00:16:22 +0300 you wrote:

Hi,

This series contains misc enhancements to the mlx5 driver.

Regards,
Tariq

[...]

Here is the summary with links:
- [net-next,1/5] net/mlx5e: Remove unused VLAN insertion logic in TX path

    https://git.kernel.org/netdev/net-next/c/ade89d1f2486
  - [net-next,2/5] net/mlx5e: CT: extract a memcmp from a spinlock section
    https://git.kernel.org/netdev/net-next/c/122d86aa2a0c
  - [net-next,3/5] net/mlx5e: Replace recursive VLAN push handling with an iterative loop
    https://git.kernel.org/netdev/net-next/c/c0ca344d796c
  - [net-next,4/5] net/mlx5: Warn when write combining is not supported
    https://git.kernel.org/netdev/net-next/c/d980f371b134
  - [net-next,5/5] net/mlx5e: RX, Remove unnecessary RQT redirects
    https://git.kernel.org/netdev/net-next/c/a194be578376

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html