Skip to contentSkip to conversation

net/mlx5e: Add support for devmem and io_uring TCP zero-copy

PATCH Applied Mark Bloch 6 participants 22 messages Jun 16, 2025 – Jun 25, 2025

Subsystems (7)
lore
Share & tools
lore

Mark Bloch

author maintainer
#0 lore raw

v6 cover net/mlx5e: Add support for devmem and io_uring TCP zero-copy

This series adds support for zerocopy rx TCP with devmem and io_uring for ConnectX7 NICs and above. For performance reasons and simplicity HW-GRO will also be turned on when header-data split mode is on.

Performance
===========

Test setup:

* CPU: Intel(R) Xeon(R) Platinum 8380 CPU @ 2.30GHz (single NUMA)
* NIC: ConnectX7
* Benchmarking tool: kperf [0]
* Single TCP flow
* Test duration: 60s

With application thread and interrupts pinned to the *same* core:

|------+-----------+----------|
| MTU  | epoll     | io_uring |
|------+-----------+----------|
| 1500 | 61.6 Gbps | 114 Gbps |
| 4096 | 69.3 Gbps | 151 Gbps |
| 9000 | 67.8 Gbps | 187 Gbps |
|------+-----------+----------|

The CPU usage for io_uring is 95%.

Reproduction steps for io_uring:

server --no-daemon -a 2001:db8::1 --no-memcmp --iou --iou_sendzc \
	--iou_zcrx --iou_dev_name eth2 --iou_zcrx_queue_id 2

server --no-daemon -a 2001:db8::2 --no-memcmp --iou --iou_sendzc

client --src 2001:db8::2 --dst 2001:db8::1 \
	--msg-zerocopy -t 60 --cpu-min=2 --cpu-max=2

Patch overview:
================

First, a netmem API for skb_can_coalesce is added to the core to be able to do skb fragment coalescing on netmems.

The next patches introduce some cleanups in the internal SHAMPO code and improvements to hw gro capability checks in FW.

A separate page_pool is introduced for headers, to be used only when the rxq has a memory provider.

Then the driver is converted to use the netmem API and to allow support for unreadable netmem page pool.

The queue management ops are implemented.

Finally, the tcp-data-split ring parameter is exposed.

Changelog
=========
Changes from v5 [5]:
- Added a TODO in mlx5e_queue_stop regarding queue restart.
- Added Reviewed-by tag.

Changes from v4 [4]:
- Addressed silly return before goto.
- Removed extraneous '\n' and used NL_SET_ERR_MSG_MOD.
- Removed unnecessary netmem_is_net_iov() check.
- Added comment for non HDS packets being dropped when unreadable memory is used.
- Added page_pool_dev_alloc_netmems() helper.
- Added Reviewed-by tags.

Changes from v3 [3]:
- Dropped ethtool stats for hd_page_pool.

Changes from v2 [2]:
- Added support for netmem TX.
- Changed skb_can_coalesce_netmem() based on Mina's suggestion.
- Reworked tcp_data_split setting to no longer change HW-GRO in wanted_features.
- Switched to a single page pool when rxq has no memory providers.

Changes from v1 [1]:
- Added support for skb_can_coalesce_netmem().
- Avoid netmem_to_page() casts in the driver.
- Fixed code to abide 80 char limit with some exceptions to avoid code churn.

References
==========
[0] kperf: git://git.kernel.dk/kperf.git
[1] v1: https://lore.kernel.org/all/[email protected]/
[2] v2: https://lore.kernel.org/all/[email protected]/
[3] v3: https://lore.kernel.org/netdev/[email protected]/
[4] v4: https://lore.kernel.org/all/[email protected]/
[5] v5: https://lore.kernel.org/netdev/[email protected]/

Dragos Tatulea (4):
net: Allow const args for of page_to_netmem()
net: Add skb_can_coalesce for netmem
page_pool: Add page_pool_dev_alloc_netmems helper net/mlx5e: Add TX support for netmems

Saeed Mahameed (8):
net/mlx5e: SHAMPO: Reorganize mlx5_rq_shampo_alloc net/mlx5e: SHAMPO: Remove redundant params net/mlx5e: SHAMPO: Improve hw gro capability checking net/mlx5e: SHAMPO: Separate pool for headers net/mlx5e: Convert over to netmem
net/mlx5e: Add support for UNREADABLE netmem page pools net/mlx5e: Implement queue mgmt ops and single channel swap net/mlx5e: Support ethtool tcp-data-split settings

 drivers/net/ethernet/mellanox/mlx5/core/en.h  |  11 +-
 .../ethernet/mellanox/mlx5/core/en/params.c   |  36 ++-
 .../net/ethernet/mellanox/mlx5/core/en/txrx.h |   3 +-
 .../ethernet/mellanox/mlx5/core/en_ethtool.c  |  33 +-
 .../net/ethernet/mellanox/mlx5/core/en_main.c | 306 +++++++++++++-----
 .../net/ethernet/mellanox/mlx5/core/en_rx.c   | 138 ++++----
 include/linux/skbuff.h                        |  12 +-
 include/net/netmem.h                          |   2 +-
 include/net/page_pool/helpers.h               |   7 +
 9 files changed, 381 insertions(+), 167 deletions(-)

base-commit: 8909f5f4ecd551c2299b28e05254b77424c8c7dc

Signature · 1 line
2.34.1
author maintainer
#1 lore raw

PATCH v6 1/12 net: Allow const args for of page_to_netmem()

From: Dragos Tatulea <[email protected]>

This allows calling page_to_netmem() with a const page * argument.

---
 include/net/netmem.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/net/netmem.h +1 -1
index 386164fb9c18..caf2e773c915 100644
@@ -143,7 +143,7 @@ static inline netmem_ref net_iov_to_netmem(struct net_iov *niov)
 	return (__force netmem_ref)((unsigned long)niov | NET_IOV);
 }
 
-static inline netmem_ref page_to_netmem(struct page *page)
+static inline netmem_ref page_to_netmem(const struct page *page)
 {
 	return (__force netmem_ref)page;
 }
Signature · 1 line
2.34.1
author maintainer
#2 lore raw

PATCH v6 2/12 net: Add skb_can_coalesce for netmem

In reply to Mark Bloch

From: Dragos Tatulea <[email protected]>

Allow drivers that have moved over to netmem to do fragment coalescing.

---
 include/linux/skbuff.h | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)
include/linux/skbuff.h +9 -3
index 5520524c93bf..9508968cb300 100644
@@ -3873,20 +3873,26 @@ static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int l
 bool csum_and_copy_from_iter_full(void *addr, size_t bytes, __wsum *csum, struct iov_iter *i)
 	__must_check;
 
-static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
-				    const struct page *page, int off)
+static inline bool skb_can_coalesce_netmem(struct sk_buff *skb, int i,
+					   netmem_ref netmem, int off)
 {
 	if (skb_zcopy(skb))
 		return false;
 	if (i) {
 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
 
-		return page == skb_frag_page(frag) &&
+		return netmem == skb_frag_netmem(frag) &&
 		       off == skb_frag_off(frag) + skb_frag_size(frag);
 	}
 	return false;
 }
 
+static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
+				    const struct page *page, int off)
+{
+	return skb_can_coalesce_netmem(skb, i, page_to_netmem(page), off);
+}
+
 static inline int __skb_linearize(struct sk_buff *skb)
 {
 	return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
Signature · 1 line
2.34.1
author maintainer
#3 lore raw

PATCH v6 3/12 page_pool: Add page_pool_dev_alloc_netmems helper

In reply to Mark Bloch

From: Dragos Tatulea <[email protected]>

This is the netmem counterpart of page_pool_dev_alloc_pages() which uses the default GFP flags for RX.

---
 include/net/page_pool/helpers.h | 7 +++++++
 1 file changed, 7 insertions(+)
include/net/page_pool/helpers.h +7 -0
index 93f2c31baf9b..773fc65780b5 100644
@@ -153,6 +153,13 @@ static inline netmem_ref page_pool_dev_alloc_netmem(struct page_pool *pool,
 	return page_pool_alloc_netmem(pool, offset, size, gfp);
 }
 
+static inline netmem_ref page_pool_dev_alloc_netmems(struct page_pool *pool)
+{
+	gfp_t gfp = GFP_ATOMIC | __GFP_NOWARN;
+
+	return page_pool_alloc_netmems(pool, gfp);
+}
+
 static inline struct page *page_pool_alloc(struct page_pool *pool,
 					   unsigned int *offset,
 					   unsigned int *size, gfp_t gfp)
Signature · 1 line
2.34.1
author maintainer
#4 lore raw

PATCH v6 4/12 net/mlx5e: SHAMPO: Reorganize mlx5_rq_shampo_alloc

In reply to Mark Bloch

From: Saeed Mahameed <[email protected]>

Drop redundant SHAMPO structure alloc/free functions.

Gather together function calls pertaining to header split info, pass header per WQE (hd_per_wqe) as parameter to those function to avoid use before initialization future mistakes.

Allocate HW GRO related info outside of the header related info scope.

---
 drivers/net/ethernet/mellanox/mlx5/core/en.h  |   1 -
 .../net/ethernet/mellanox/mlx5/core/en_main.c | 135 +++++++++---------
 2 files changed, 66 insertions(+), 70 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en.h +0 -1
index 5b0d03b3efe8..211ea429ea89 100644
@@ -638,7 +638,6 @@ struct mlx5e_shampo_hd {
 	struct mlx5e_frag_page *pages;
 	u32 hd_per_wq;
 	u16 hd_per_wqe;
-	u16 pages_per_wq;
 	unsigned long *bitmap;
 	u16 pi;
 	u16 ci;
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +66 -69

187 lines not on this page · raw message

Signature · 1 line
2.34.1
author maintainer
#5 lore raw

PATCH v6 5/12 net/mlx5e: SHAMPO: Remove redundant params

In reply to Mark Bloch

From: Saeed Mahameed <[email protected]>

Two SHAMPO params are static and always the same, remove them from the global mlx5e_params struct.

---
 drivers/net/ethernet/mellanox/mlx5/core/en.h  |  4 ---
 .../ethernet/mellanox/mlx5/core/en/params.c   | 36 ++++++++++---------
 .../net/ethernet/mellanox/mlx5/core/en_main.c |  4 ---
 3 files changed, 20 insertions(+), 24 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en.h +0 -4
index 211ea429ea89..581eef34f512 100644
@@ -278,10 +278,6 @@ enum packet_merge {
 struct mlx5e_packet_merge_param {
 	enum packet_merge type;
 	u32 timeout;
-	struct {
-		u8 match_criteria_type;
-		u8 alignment_granularity;
-	} shampo;
 };
 
 struct mlx5e_params {
drivers/net/ethernet/mellanox/mlx5/core/en/params.c +20 -16
index 58ec5e44aa7a..fc945bce933a 100644
@@ -901,6 +901,7 @@ int mlx5e_build_rq_param(struct mlx5_core_dev *mdev,
 {
 	void *rqc = param->rqc;
 	void *wq = MLX5_ADDR_OF(rqc, rqc, wq);
+	u32 lro_timeout;
 	int ndsegs = 1;
 	int err;
 
@@ -926,22 +927,25 @@ int mlx5e_build_rq_param(struct mlx5_core_dev *mdev,
 		MLX5_SET(wq, wq, log_wqe_stride_size,
 			 log_wqe_stride_size - MLX5_MPWQE_LOG_STRIDE_SZ_BASE);
 		MLX5_SET(wq, wq, log_wq_sz, mlx5e_mpwqe_get_log_rq_size(mdev, params, xsk));
-		if (params->packet_merge.type == MLX5E_PACKET_MERGE_SHAMPO) {
-			MLX5_SET(wq, wq, shampo_enable, true);
-			MLX5_SET(wq, wq, log_reservation_size,
-				 mlx5e_shampo_get_log_rsrv_size(mdev, params));
-			MLX5_SET(wq, wq,
-				 log_max_num_of_packets_per_reservation,
-				 mlx5e_shampo_get_log_pkt_per_rsrv(mdev, params));
-			MLX5_SET(wq, wq, log_headers_entry_size,
-				 mlx5e_shampo_get_log_hd_entry_size(mdev, params));
-			MLX5_SET(rqc, rqc, reservation_timeout,
-				 mlx5e_choose_lro_timeout(mdev, MLX5E_DEFAULT_SHAMPO_TIMEOUT));
-			MLX5_SET(rqc, rqc, shampo_match_criteria_type,
-				 params->packet_merge.shampo.match_criteria_type);
-			MLX5_SET(rqc, rqc, shampo_no_match_alignment_granularity,
-				 params->packet_merge.shampo.alignment_granularity);
-		}
+		if (params->packet_merge.type != MLX5E_PACKET_MERGE_SHAMPO)
+			break;
+
+		MLX5_SET(wq, wq, shampo_enable, true);
+		MLX5_SET(wq, wq, log_reservation_size,
+			 mlx5e_shampo_get_log_rsrv_size(mdev, params));
+		MLX5_SET(wq, wq,
+			 log_max_num_of_packets_per_reservation,
+			 mlx5e_shampo_get_log_pkt_per_rsrv(mdev, params));
+		MLX5_SET(wq, wq, log_headers_entry_size,
+			 mlx5e_shampo_get_log_hd_entry_size(mdev, params));
+		lro_timeout =
+			mlx5e_choose_lro_timeout(mdev,
+						 MLX5E_DEFAULT_SHAMPO_TIMEOUT);
+		MLX5_SET(rqc, rqc, reservation_timeout, lro_timeout);
+		MLX5_SET(rqc, rqc, shampo_match_criteria_type,
+			 MLX5_RQC_SHAMPO_MATCH_CRITERIA_TYPE_EXTENDED);
+		MLX5_SET(rqc, rqc, shampo_no_match_alignment_granularity,
+			 MLX5_RQC_SHAMPO_NO_MATCH_ALIGNMENT_GRANULARITY_STRIDE);
 		break;
 	}
 	default: /* MLX5_WQ_TYPE_CYCLIC */
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +0 -4
index 3d11c9f87171..e1e44533b744 100644
@@ -4040,10 +4040,6 @@ static int set_feature_hw_gro(struct net_device *netdev, bool enable)
 
 	if (enable) {
 		new_params.packet_merge.type = MLX5E_PACKET_MERGE_SHAMPO;
-		new_params.packet_merge.shampo.match_criteria_type =
-			MLX5_RQC_SHAMPO_MATCH_CRITERIA_TYPE_EXTENDED;
-		new_params.packet_merge.shampo.alignment_granularity =
-			MLX5_RQC_SHAMPO_NO_MATCH_ALIGNMENT_GRANULARITY_STRIDE;
 	} else if (new_params.packet_merge.type == MLX5E_PACKET_MERGE_SHAMPO) {
 		new_params.packet_merge.type = MLX5E_PACKET_MERGE_NONE;
 	} else {
Signature · 1 line
2.34.1
author maintainer
#6 lore raw

PATCH v6 6/12 net/mlx5e: SHAMPO: Improve hw gro capability checking

In reply to Mark Bloch

From: Saeed Mahameed <[email protected]>

Add missing HW capabilities, declare the feature in netdev->vlan_features, similar to other features in mlx5e_build_nic_netdev. No functional change here as all by default disabled features are explicitly disabled at the bottom of the function.

---
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +7 -6
index e1e44533b744..a81d354af7c8 100644
@@ -78,7 +78,8 @@
 
 static bool mlx5e_hw_gro_supported(struct mlx5_core_dev *mdev)
 {
-	if (!MLX5_CAP_GEN(mdev, shampo))
+	if (!MLX5_CAP_GEN(mdev, shampo) ||
+	    !MLX5_CAP_SHAMPO(mdev, shampo_header_split_data_merge))
 		return false;
 
 	/* Our HW-GRO implementation relies on "KSM Mkey" for
@@ -5499,17 +5500,17 @@ static void mlx5e_build_nic_netdev(struct net_device *netdev)
 						   MLX5E_MPWRQ_UMR_MODE_ALIGNED))
 		netdev->vlan_features    |= NETIF_F_LRO;
 
+	if (mlx5e_hw_gro_supported(mdev) &&
+	    mlx5e_check_fragmented_striding_rq_cap(mdev, PAGE_SHIFT,
+						   MLX5E_MPWRQ_UMR_MODE_ALIGNED))
+		netdev->vlan_features |= NETIF_F_GRO_HW;
+
 	netdev->hw_features       = netdev->vlan_features;
 	netdev->hw_features      |= NETIF_F_HW_VLAN_CTAG_TX;
 	netdev->hw_features      |= NETIF_F_HW_VLAN_CTAG_RX;
 	netdev->hw_features      |= NETIF_F_HW_VLAN_CTAG_FILTER;
 	netdev->hw_features      |= NETIF_F_HW_VLAN_STAG_TX;
 
-	if (mlx5e_hw_gro_supported(mdev) &&
-	    mlx5e_check_fragmented_striding_rq_cap(mdev, PAGE_SHIFT,
-						   MLX5E_MPWRQ_UMR_MODE_ALIGNED))
-		netdev->hw_features    |= NETIF_F_GRO_HW;
-
 	if (mlx5e_tunnel_any_tx_proto_supported(mdev)) {
 		netdev->hw_enc_features |= NETIF_F_HW_CSUM;
 		netdev->hw_enc_features |= NETIF_F_TSO;
Signature · 1 line
2.34.1
author maintainer
#7 lore raw

PATCH v6 7/12 net/mlx5e: SHAMPO: Separate pool for headers

In reply to Mark Bloch

From: Saeed Mahameed <[email protected]>

Allow allocating a separate page pool for headers when SHAMPO is on. This will be useful for adding support to zc page pool, which has to be different from the headers page pool.
For now, the pools are the same.

---
 drivers/net/ethernet/mellanox/mlx5/core/en.h  |  4 ++
 .../net/ethernet/mellanox/mlx5/core/en_main.c | 43 ++++++++++++++++++-
 .../net/ethernet/mellanox/mlx5/core/en_rx.c   | 41 ++++++++++--------
 3 files changed, 69 insertions(+), 19 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en.h +4 -0
index 581eef34f512..c329de1d4f0a 100644
@@ -716,7 +716,11 @@ struct mlx5e_rq {
 	struct bpf_prog __rcu *xdp_prog;
 	struct mlx5e_xdpsq    *xdpsq;
 	DECLARE_BITMAP(flags, 8);
+
+	/* page pools */
 	struct page_pool      *page_pool;
+	struct page_pool      *hd_page_pool;
+
 	struct mlx5e_xdp_buff mxbuf;
 
 	/* AF_XDP zero-copy */
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +41 -2

85 lines not on this page · raw message

drivers/net/ethernet/mellanox/mlx5/core/en_rx.c +24 -17

143 lines not on this page · raw message

Signature · 1 line
2.34.1
author maintainer
#8 lore raw

PATCH v6 8/12 net/mlx5e: Convert over to netmem

In reply to Mark Bloch

From: Saeed Mahameed <[email protected]>

mlx5e_page_frag holds the physical page itself, to naturally support zc page pools, remove physical page reference from mlx5 and replace it with netmem_ref, to avoid internal handling in mlx5 for net_iov backed pages.

SHAMPO can issue packets that are not split into header and data. These packets will be dropped if the data part resides in a net_iov as the driver can't read into this area.

No performance degradation observed.

---
 drivers/net/ethernet/mellanox/mlx5/core/en.h  |   2 +-
 .../net/ethernet/mellanox/mlx5/core/en_rx.c   | 105 +++++++++++-------
 2 files changed, 63 insertions(+), 44 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en.h +1 -1
index c329de1d4f0a..65a73913b9a2 100644
@@ -553,7 +553,7 @@ struct mlx5e_icosq {
 } ____cacheline_aligned_in_smp;
 
 struct mlx5e_frag_page {
-	struct page *page;
+	netmem_ref netmem;
 	u16 frags;
 };
 
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c +62 -43

255 lines not on this page · raw message

Signature · 1 line
2.34.1
author maintainer
#9 lore raw

PATCH v6 9/12 net/mlx5e: Add support for UNREADABLE netmem page pools

In reply to Mark Bloch

From: Saeed Mahameed <[email protected]>

On netdev_rx_queue_restart, a special type of page pool maybe expected.

In this patch declare support for UNREADABLE netmem iov pages in the pool params only when header data split shampo RQ mode is enabled, also set the queue index in the page pool params struct.

Shampo mode requirement: Without header split rx needs to peek at the data, we can't do UNREADABLE_NETMEM.

The patch also enables the use of a separate page pool for headers when a memory provider is installed for the queue, otherwise the same common page pool continues to be used.

---
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +8 -1
index 5e649705e35f..a51e204bd364 100644
@@ -749,7 +749,9 @@ static void mlx5e_rq_shampo_hd_info_free(struct mlx5e_rq *rq)
 
 static bool mlx5_rq_needs_separate_hd_pool(struct mlx5e_rq *rq)
 {
-	return false;
+	struct netdev_rx_queue *rxq = __netif_get_rx_queue(rq->netdev, rq->ix);
+
+	return !!rxq->mp_params.mp_ops;
 }
 
 static int mlx5_rq_shampo_alloc(struct mlx5_core_dev *mdev,
@@ -964,6 +966,11 @@ static int mlx5e_alloc_rq(struct mlx5e_params *params,
 		pp_params.netdev    = rq->netdev;
 		pp_params.dma_dir   = rq->buff.map_dir;
 		pp_params.max_len   = PAGE_SIZE;
+		pp_params.queue_idx = rq->ix;
+
+		/* Shampo header data split allow for unreadable netmem */
+		if (test_bit(MLX5E_RQ_STATE_SHAMPO, &rq->state))
+			pp_params.flags |= PP_FLAG_ALLOW_UNREADABLE_NETMEM;
 
 		/* page_pool can be used even when there is no rq->xdp_prog,
 		 * given page_pool does not handle DMA mapping there is no
Signature · 1 line
2.34.1
author maintainer
#10 lore raw

PATCH v6 10/12 net/mlx5e: Implement queue mgmt ops and single channel swap

In reply to Mark Bloch

From: Saeed Mahameed <[email protected]>

The bulk of the work is done in mlx5e_queue_mem_alloc, where we allocate and create the new channel resources, similar to mlx5e_safe_switch_params, but here we do it for a single channel using existing params, sort of a clone channel.
To swap the old channel with the new one, we deactivate and close the old channel then replace it with the new one, since the swap procedure doesn't fail in mlx5, we do it all in one place (mlx5e_queue_start).

---
 .../net/ethernet/mellanox/mlx5/core/en_main.c | 98 +++++++++++++++++++
 1 file changed, 98 insertions(+)
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +98 -0
index a51e204bd364..873a42b4a82d 100644
@@ -5494,6 +5494,103 @@ static const struct netdev_stat_ops mlx5e_stat_ops = {
 	.get_base_stats      = mlx5e_get_base_stats,
 };
 
+struct mlx5_qmgmt_data {
+	struct mlx5e_channel *c;
+	struct mlx5e_channel_param cparam;
+};
+
+static int mlx5e_queue_mem_alloc(struct net_device *dev, void *newq,
+				 int queue_index)
+{
+	struct mlx5_qmgmt_data *new = (struct mlx5_qmgmt_data *)newq;
+	struct mlx5e_priv *priv = netdev_priv(dev);
+	struct mlx5e_channels *chs = &priv->channels;
+	struct mlx5e_params params = chs->params;
+	struct mlx5_core_dev *mdev;
+	int err;
+
+	mutex_lock(&priv->state_lock);
+	if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
+		err = -ENODEV;
+		goto unlock;
+	}
+
+	if (queue_index >= chs->num) {
+		err = -ERANGE;
+		goto unlock;
+	}
+
+	if (MLX5E_GET_PFLAG(&chs->params, MLX5E_PFLAG_TX_PORT_TS) ||
+	    chs->params.ptp_rx   ||
+	    chs->params.xdp_prog ||
+	    priv->htb) {
+		netdev_err(priv->netdev,
+			   "Cloning channels with Port/rx PTP, XDP or HTB is not supported\n");
+		err = -EOPNOTSUPP;
+		goto unlock;
+	}
+
+	mdev = mlx5_sd_ch_ix_get_dev(priv->mdev, queue_index);
+	err = mlx5e_build_channel_param(mdev, &params, &new->cparam);
+	if (err)
+		goto unlock;
+
+	err = mlx5e_open_channel(priv, queue_index, &params, NULL, &new->c);
+unlock:
+	mutex_unlock(&priv->state_lock);
+	return err;
+}
+
+static void mlx5e_queue_mem_free(struct net_device *dev, void *mem)
+{
+	struct mlx5_qmgmt_data *data = (struct mlx5_qmgmt_data *)mem;
+
+	/* not supposed to happen since mlx5e_queue_start never fails
+	 * but this is how this should be implemented just in case
+	 */
+	if (data->c)
+		mlx5e_close_channel(data->c);
+}
+
+static int mlx5e_queue_stop(struct net_device *dev, void *oldq, int queue_index)
+{
+	/* In mlx5 a txq cannot be simply stopped in isolation, only restarted.
+	 * mlx5e_queue_start does not fail, we stop the old queue there.
+	 * TODO: Improve this.
+	 */
+	return 0;
+}
+
+static int mlx5e_queue_start(struct net_device *dev, void *newq,
+			     int queue_index)
+{
+	struct mlx5_qmgmt_data *new = (struct mlx5_qmgmt_data *)newq;
+	struct mlx5e_priv *priv = netdev_priv(dev);
+	struct mlx5e_channel *old;
+
+	mutex_lock(&priv->state_lock);
+
+	/* stop and close the old */
+	old = priv->channels.c[queue_index];
+	mlx5e_deactivate_priv_channels(priv);
+	/* close old before activating new, to avoid napi conflict */
+	mlx5e_close_channel(old);
+
+	/* start the new */
+	priv->channels.c[queue_index] = new->c;
+	mlx5e_activate_priv_channels(priv);
+	mutex_unlock(&priv->state_lock);
+	return 0;
+}
+
+static const struct netdev_queue_mgmt_ops mlx5e_queue_mgmt_ops = {
+	.ndo_queue_mem_size	=	sizeof(struct mlx5_qmgmt_data),
+	.ndo_queue_mem_alloc	=	mlx5e_queue_mem_alloc,
+	.ndo_queue_mem_free	=	mlx5e_queue_mem_free,
+	.ndo_queue_start	=	mlx5e_queue_start,
+	.ndo_queue_stop		=	mlx5e_queue_stop,
+};
+
 static void mlx5e_build_nic_netdev(struct net_device *netdev)
 {
 	struct mlx5e_priv *priv = netdev_priv(netdev);
@@ -5504,6 +5601,7 @@ static void mlx5e_build_nic_netdev(struct net_device *netdev)
 	SET_NETDEV_DEV(netdev, mdev->device);
 
 	netdev->netdev_ops = &mlx5e_netdev_ops;
+	netdev->queue_mgmt_ops = &mlx5e_queue_mgmt_ops;
 	netdev->xdp_metadata_ops = &mlx5e_xdp_metadata_ops;
 	netdev->xsk_tx_metadata_ops = &mlx5e_xsk_tx_metadata_ops;
 	netdev->request_ops_lock = true;
Signature · 1 line
2.34.1
Acked-by on v6 10/12
#11 lore raw

Re: [PATCH net-next v6 10/12] net/mlx5e: Implement queue mgmt ops and single channel swap

Quoted 11 lines · Mark Bloch

From: Saeed Mahameed <[email protected]>

The bulk of the work is done in mlx5e_queue_mem_alloc, where we allocate and create the new channel resources, similar to mlx5e_safe_switch_params, but here we do it for a single channel using existing params, sort of a clone channel.
To swap the old channel with the new one, we deactivate and close the old channel then replace it with the new one, since the swap procedure doesn't fail in mlx5, we do it all in one place (mlx5e_queue_start).

Quoted 4 lines
Quoted 35 lines
---
 .../net/ethernet/mellanox/mlx5/core/en_main.c | 98 +++++++++++++++++++
 1 file changed, 98 insertions(+)
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +23 -0
index a51e204bd364..873a42b4a82d 100644
@@ -5494,6 +5494,103 @@ static const struct netdev_stat_ops mlx5e_stat_ops = {
        .get_base_stats      = mlx5e_get_base_stats,
 };
 
+struct mlx5_qmgmt_data {
+       struct mlx5e_channel *c;
+       struct mlx5e_channel_param cparam;
+};
+
+static int mlx5e_queue_mem_alloc(struct net_device *dev, void *newq,
+                                int queue_index)
+{
+       struct mlx5_qmgmt_data *new = (struct mlx5_qmgmt_data *)newq;
+       struct mlx5e_priv *priv = netdev_priv(dev);
+       struct mlx5e_channels *chs = &priv->channels;
+       struct mlx5e_params params = chs->params;
+       struct mlx5_core_dev *mdev;
+       int err;
+
+       mutex_lock(&priv->state_lock);
+       if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
+               err = -ENODEV;
+               goto unlock;
+       }
+
+       if (queue_index >= chs->num) {
+               err = -ERANGE;

drivers/net/ethernet/mellanox/mlx5/core/en_main.c static const struct netdev_stat_ops mlx5e_stat_ops = { · line 5529

Quoted 10 lines
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +10 -0
+               goto unlock;
+       }
+
+       if (MLX5E_GET_PFLAG(&chs->params, MLX5E_PFLAG_TX_PORT_TS) ||
+           chs->params.ptp_rx   ||
+           chs->params.xdp_prog ||
+           priv->htb) {
+               netdev_err(priv->netdev,
+                          "Cloning channels with Port/rx PTP, XDP or HTB is not supported\n");
+               err = -EOPNOTSUPP;

I would have used a different error code here as EOFNOTSUPP usually means the driver doesn't support queue API at all.

Signature · 2 lines
Thanks,
Mina
#12 lore raw

Re: [PATCH net-next v6 10/12] net/mlx5e: Implement queue mgmt ops and single channel swap

In reply to Mark Bloch
Quoted 28 lines · Mark Bloch

From: Saeed Mahameed <[email protected]>

The bulk of the work is done in mlx5e_queue_mem_alloc, where we allocate and create the new channel resources, similar to mlx5e_safe_switch_params, but here we do it for a single channel using existing params, sort of a clone channel.
To swap the old channel with the new one, we deactivate and close the old channel then replace it with the new one, since the swap procedure doesn't fail in mlx5, we do it all in one place (mlx5e_queue_start).

---
  .../net/ethernet/mellanox/mlx5/core/en_main.c | 98 +++++++++++++++++++
  1 file changed, 98 insertions(+)
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +2 -0
index a51e204bd364..873a42b4a82d 100644
@@ -5494,6 +5494,103 @@ static const struct netdev_stat_ops mlx5e_stat_ops = {
  	.get_base_stats      = mlx5e_get_base_stats,
  };
 
+struct mlx5_qmgmt_data {
+	struct mlx5e_channel *c;

drivers/net/ethernet/mellanox/mlx5/core/en_main.c static const struct netdev_stat_ops mlx5e_stat_ops = { · line 5508

Quoted 10 lines
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +10 -0
+	struct mlx5e_channel_param cparam;
+};
+
+static int mlx5e_queue_mem_alloc(struct net_device *dev, void *newq,
+				 int queue_index)
+{
+	struct mlx5_qmgmt_data *new = (struct mlx5_qmgmt_data *)newq;
+	struct mlx5e_priv *priv = netdev_priv(dev);
+	struct mlx5e_channels *chs = &priv->channels;
+	struct mlx5e_params params = chs->params;

RCT (Reverse Christmas Tree) ?

Yanjun.Zhu

Quoted 96 lines
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +86 -0
+	struct mlx5_core_dev *mdev;
+	int err;
+
+	mutex_lock(&priv->state_lock);
+	if (!test_bit(MLX5E_STATE_OPENED, &priv->state)) {
+		err = -ENODEV;
+		goto unlock;
+	}
+
+	if (queue_index >= chs->num) {
+		err = -ERANGE;
+		goto unlock;
+	}
+
+	if (MLX5E_GET_PFLAG(&chs->params, MLX5E_PFLAG_TX_PORT_TS) ||
+	    chs->params.ptp_rx   ||
+	    chs->params.xdp_prog ||
+	    priv->htb) {
+		netdev_err(priv->netdev,
+			   "Cloning channels with Port/rx PTP, XDP or HTB is not supported\n");
+		err = -EOPNOTSUPP;
+		goto unlock;
+	}
+
+	mdev = mlx5_sd_ch_ix_get_dev(priv->mdev, queue_index);
+	err = mlx5e_build_channel_param(mdev, &params, &new->cparam);
+	if (err)
+		goto unlock;
+
+	err = mlx5e_open_channel(priv, queue_index, &params, NULL, &new->c);
+unlock:
+	mutex_unlock(&priv->state_lock);
+	return err;
+}
+
+static void mlx5e_queue_mem_free(struct net_device *dev, void *mem)
+{
+	struct mlx5_qmgmt_data *data = (struct mlx5_qmgmt_data *)mem;
+
+	/* not supposed to happen since mlx5e_queue_start never fails
+	 * but this is how this should be implemented just in case
+	 */
+	if (data->c)
+		mlx5e_close_channel(data->c);
+}
+
+static int mlx5e_queue_stop(struct net_device *dev, void *oldq, int queue_index)
+{
+	/* In mlx5 a txq cannot be simply stopped in isolation, only restarted.
+	 * mlx5e_queue_start does not fail, we stop the old queue there.
+	 * TODO: Improve this.
+	 */
+	return 0;
+}
+
+static int mlx5e_queue_start(struct net_device *dev, void *newq,
+			     int queue_index)
+{
+	struct mlx5_qmgmt_data *new = (struct mlx5_qmgmt_data *)newq;
+	struct mlx5e_priv *priv = netdev_priv(dev);
+	struct mlx5e_channel *old;
+
+	mutex_lock(&priv->state_lock);
+
+	/* stop and close the old */
+	old = priv->channels.c[queue_index];
+	mlx5e_deactivate_priv_channels(priv);
+	/* close old before activating new, to avoid napi conflict */
+	mlx5e_close_channel(old);
+
+	/* start the new */
+	priv->channels.c[queue_index] = new->c;
+	mlx5e_activate_priv_channels(priv);
+	mutex_unlock(&priv->state_lock);
+	return 0;
+}
+
+static const struct netdev_queue_mgmt_ops mlx5e_queue_mgmt_ops = {
+	.ndo_queue_mem_size	=	sizeof(struct mlx5_qmgmt_data),
+	.ndo_queue_mem_alloc	=	mlx5e_queue_mem_alloc,
+	.ndo_queue_mem_free	=	mlx5e_queue_mem_free,
+	.ndo_queue_start	=	mlx5e_queue_start,
+	.ndo_queue_stop		=	mlx5e_queue_stop,
+};
+
  static void mlx5e_build_nic_netdev(struct net_device *netdev)
  {
  	struct mlx5e_priv *priv = netdev_priv(netdev);
@@ -5504,6 +5601,7 @@ static void mlx5e_build_nic_netdev(struct net_device *netdev)
  	SET_NETDEV_DEV(netdev, mdev->device);
 
  	netdev->netdev_ops = &mlx5e_netdev_ops;
+	netdev->queue_mgmt_ops = &mlx5e_queue_mgmt_ops;
  	netdev->xdp_metadata_ops = &mlx5e_xdp_metadata_ops;
  	netdev->xsk_tx_metadata_ops = &mlx5e_xsk_tx_metadata_ops;
  	netdev->request_ops_lock = true;
author maintainer
#13 lore raw

PATCH v6 11/12 net/mlx5e: Support ethtool tcp-data-split settings

In reply to Mark Bloch

From: Saeed Mahameed <[email protected]>

In mlx5, tcp header-data split requires HW GRO to be on.

Enabling it fails when HW GRO is off.
mlx5e_fix_features now keeps HW GRO on when tcp data split is enabled. Finally, when tcp data split is disabled, features are updated to maybe remove the forced HW GRO.

---
 .../ethernet/mellanox/mlx5/core/en_ethtool.c  | 33 ++++++++++++++++---
 .../net/ethernet/mellanox/mlx5/core/en_main.c |  8 +++++
 2 files changed, 36 insertions(+), 5 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_ethtool.c +28 -5
index 8b9ee8bac674..35479cbf98d5 100644
@@ -32,6 +32,7 @@
 
 #include <linux/dim.h>
 #include <linux/ethtool_netlink.h>
+#include <net/netdev_queues.h>
 
 #include "en.h"
 #include "en/channels.h"
@@ -365,11 +366,6 @@ void mlx5e_ethtool_get_ringparam(struct mlx5e_priv *priv,
 	param->tx_max_pending = 1 << MLX5E_PARAMS_MAXIMUM_LOG_SQ_SIZE;
 	param->rx_pending     = 1 << priv->channels.params.log_rq_mtu_frames;
 	param->tx_pending     = 1 << priv->channels.params.log_sq_size;
-
-	kernel_param->tcp_data_split =
-		(priv->channels.params.packet_merge.type == MLX5E_PACKET_MERGE_SHAMPO) ?
-		ETHTOOL_TCP_DATA_SPLIT_ENABLED :
-		ETHTOOL_TCP_DATA_SPLIT_DISABLED;
 }
 
 static void mlx5e_get_ringparam(struct net_device *dev,
@@ -382,6 +378,27 @@ static void mlx5e_get_ringparam(struct net_device *dev,
 	mlx5e_ethtool_get_ringparam(priv, param, kernel_param);
 }
 
+static bool mlx5e_ethtool_set_tcp_data_split(struct mlx5e_priv *priv,
+					     u8 tcp_data_split,
+					     struct netlink_ext_ack *extack)
+{
+	struct net_device *dev = priv->netdev;
+
+	if (tcp_data_split == ETHTOOL_TCP_DATA_SPLIT_ENABLED &&
+	    !(dev->features & NETIF_F_GRO_HW)) {
+		NL_SET_ERR_MSG_MOD(extack,
+				   "TCP-data-split is not supported when GRO HW is disabled");
+		return false;
+	}
+
+	/* Might need to disable HW-GRO if it was kept on due to hds. */
+	if (tcp_data_split == ETHTOOL_TCP_DATA_SPLIT_DISABLED &&
+	    dev->cfg->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED)
+		netdev_update_features(priv->netdev);
+
+	return true;
+}
+
 int mlx5e_ethtool_set_ringparam(struct mlx5e_priv *priv,
 				struct ethtool_ringparam *param,
 				struct netlink_ext_ack *extack)
@@ -440,6 +457,11 @@ static int mlx5e_set_ringparam(struct net_device *dev,
 {
 	struct mlx5e_priv *priv = netdev_priv(dev);
 
+	if (!mlx5e_ethtool_set_tcp_data_split(priv,
+					      kernel_param->tcp_data_split,
+					      extack))
+		return -EINVAL;
+
 	return mlx5e_ethtool_set_ringparam(priv, param, extack);
 }
 
@@ -2623,6 +2645,7 @@ const struct ethtool_ops mlx5e_ethtool_ops = {
 				     ETHTOOL_COALESCE_USE_ADAPTIVE |
 				     ETHTOOL_COALESCE_USE_CQE,
 	.supported_input_xfrm = RXH_XFRM_SYM_OR_XOR,
+	.supported_ring_params = ETHTOOL_RING_USE_TCP_DATA_SPLIT,
 	.get_drvinfo       = mlx5e_get_drvinfo,
 	.get_link          = ethtool_op_get_link,
 	.get_link_ext_state  = mlx5e_get_link_ext_state,
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +8 -0
index 873a42b4a82d..b4df62b58292 100644
@@ -4413,6 +4413,7 @@ static netdev_features_t mlx5e_fix_uplink_rep_features(struct net_device *netdev
 static netdev_features_t mlx5e_fix_features(struct net_device *netdev,
 					    netdev_features_t features)
 {
+	struct netdev_config *cfg = netdev->cfg_pending;
 	struct mlx5e_priv *priv = netdev_priv(netdev);
 	struct mlx5e_vlan_table *vlan;
 	struct mlx5e_params *params;
@@ -4479,6 +4480,13 @@ static netdev_features_t mlx5e_fix_features(struct net_device *netdev,
 		}
 	}
 
+	/* The header-data split ring param requires HW GRO to stay enabled. */
+	if (cfg && cfg->hds_config == ETHTOOL_TCP_DATA_SPLIT_ENABLED &&
+	    !(features & NETIF_F_GRO_HW)) {
+		netdev_warn(netdev, "Keeping HW-GRO enabled, TCP header-data split depends on it\n");
+		features |= NETIF_F_GRO_HW;
+	}
+
 	if (mlx5e_is_uplink_rep(priv)) {
 		features = mlx5e_fix_uplink_rep_features(netdev, features);
 		netdev->netns_immutable = true;
Signature · 1 line
2.34.1
author maintainer
#14 lore raw

PATCH v6 12/12 net/mlx5e: Add TX support for netmems

In reply to Mark Bloch

From: Dragos Tatulea <[email protected]>

Declare netmem TX support in netdev.

As required, use the netmem aware dma unmapping APIs for unmapping netmems in tx completion path.

---
 drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h | 3 ++-
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 2 ++
 2 files changed, 4 insertions(+), 1 deletion(-)
drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h +2 -1
index e837c21d3d21..6501252359b0 100644
@@ -362,7 +362,8 @@ mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma)
 		dma_unmap_single(pdev, dma->addr, dma->size, DMA_TO_DEVICE);
 		break;
 	case MLX5E_DMA_MAP_PAGE:
-		dma_unmap_page(pdev, dma->addr, dma->size, DMA_TO_DEVICE);
+		netmem_dma_unmap_page_attrs(pdev, dma->addr, dma->size,
+					    DMA_TO_DEVICE, 0);
 		break;
 	default:
 		WARN_ONCE(true, "mlx5e_tx_dma_unmap unknown DMA type!\n");
drivers/net/ethernet/mellanox/mlx5/core/en_main.c +2 -0
index b4df62b58292..24559cbcbfc2 100644
@@ -5741,6 +5741,8 @@ static void mlx5e_build_nic_netdev(struct net_device *netdev)
 
 	netdev->priv_flags       |= IFF_UNICAST_FLT;
 
+	netdev->netmem_tx = true;
+
 	netif_set_tso_max_size(netdev, GSO_MAX_SIZE);
 	mlx5e_set_xdp_feature(netdev);
 	mlx5e_set_netdev_dev_addr(netdev);
Signature · 1 line
2.34.1
#15 lore raw

Re: [PATCH net-next v6 12/12] net/mlx5e: Add TX support for netmems

Quoted 21 lines · Mark Bloch

From: Dragos Tatulea <[email protected]>

Declare netmem TX support in netdev.

As required, use the netmem aware dma unmapping APIs for unmapping netmems in tx completion path.

---
 drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h | 3 ++-
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 2 ++
 2 files changed, 4 insertions(+), 1 deletion(-)
drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h +0 -0
index e837c21d3d21..6501252359b0 100644

drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma) · line 366

Quoted 6 lines
drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h +2 -1
@@ -362,7 +362,8 @@ mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma)
 		dma_unmap_single(pdev, dma->addr, dma->size, DMA_TO_DEVICE);
 		break;
 	case MLX5E_DMA_MAP_PAGE:
-		dma_unmap_page(pdev, dma->addr, dma->size, DMA_TO_DEVICE);
+		netmem_dma_unmap_page_attrs(pdev, dma->addr, dma->size,
+					    DMA_TO_DEVICE, 0);

For this to work, the dma->addr needs to be 0, so the callers of the dma_map() need to be adjusted as well, or am I missing something? There is netmem_dma_unmap_addr_set to handle that, but I don't see anybody calling it. Do we need to add the following (untested)?

drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +3 -1
index 55a8629f0792..fb6465210aed 100644
@@ -210,7 +210,9 @@ mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 		if (unlikely(dma_mapping_error(sq->pdev, dma_addr)))
 			goto dma_unmap_wqe_err;
 
-		dseg->addr       = cpu_to_be64(dma_addr);
+		dseg->addr = 0;
+		if (!netmem_is_net_iov(skb_frag_netmem(frag)))
+			dseg->addr = cpu_to_be64(dma_addr);
 		dseg->lkey       = sq->mkey_be;
 		dseg->byte_count = cpu_to_be32(fsz);
 
#16 lore raw

Re: [PATCH net-next v6 12/12] net/mlx5e: Add TX support for netmems

Quoted 17 lines · Stanislav Fomichev
Quoted 16 lines · Mark Bloch

From: Dragos Tatulea <[email protected]>

Declare netmem TX support in netdev.

As required, use the netmem aware dma unmapping APIs for unmapping netmems in tx completion path.

---
 drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h | 3 ++-
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 2 ++
 2 files changed, 4 insertions(+), 1 deletion(-)
Quoted 17 lines

drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma) · line 366

Quoted 11 lines
drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h +2 -1
index e837c21d3d21..6501252359b0 100644
@@ -362,7 +362,8 @@ mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma)
 		dma_unmap_single(pdev, dma->addr, dma->size, DMA_TO_DEVICE);
 		break;
 	case MLX5E_DMA_MAP_PAGE:
-		dma_unmap_page(pdev, dma->addr, dma->size, DMA_TO_DEVICE);
+		netmem_dma_unmap_page_attrs(pdev, dma->addr, dma->size,
+					    DMA_TO_DEVICE, 0);

For this to work, the dma->addr needs to be 0, so the callers of the dma_map() need to be adjusted as well, or am I missing something? There is netmem_dma_unmap_addr_set to handle that, but I don't see anybody calling it. Do we need to add the following (untested)?

Hmmmm... yes. I figured that skb_frag_dma_map() would do the work but I was wrong, it is not enough.

drivers/net/ethernet/mellanox/mlx5/core/en_tx.c mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb, · line 215

Quoted 12 lines
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +3 -1
index 55a8629f0792..fb6465210aed 100644
@@ -210,7 +210,9 @@ mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb,
 		if (unlikely(dma_mapping_error(sq->pdev, dma_addr)))
 			goto dma_unmap_wqe_err;
 
-		dseg->addr       = cpu_to_be64(dma_addr);
+		dseg->addr = 0;
+		if (!netmem_is_net_iov(skb_frag_netmem(frag)))
+			dseg->addr = cpu_to_be64(dma_addr);

AFAIU we still want to pass the computed dma_address to the data segment to the HW. We only need to make sure in mlx5e_dma_push() to set dma_addr to 0, to avoid calling netmem_dma_unmap_page_attrs() with dma->addr 0. Like in the snippet below. Do you agree?

We will send a fix patch once the above question is answered. Also, is there a way to test this with more confidence? The ncdevmem tx test passed just fine.

Thanks,
Dragos

drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +3 -0
index 55a8629f0792..ecee2e4f678b 100644
@@ -214,6 +214,9 @@ mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb,
                dseg->lkey       = sq->mkey_be;
                dseg->byte_count = cpu_to_be32(fsz);
 
+               if (!netmem_is_net_iov(skb_frag_netmem(frag)))
+                       dma_addr = 0;
+
                mlx5e_dma_push(sq, dma_addr, fsz, MLX5E_DMA_MAP_PAGE);
                num_dma++;
                dseg++;
#17 lore raw

Re: [PATCH net-next v6 12/12] net/mlx5e: Add TX support for netmems

Quoted 47 lines · Dragos Tatulea
Quoted 17 lines · Stanislav Fomichev
Quoted 16 lines · Mark Bloch

From: Dragos Tatulea <[email protected]>

Declare netmem TX support in netdev.

As required, use the netmem aware dma unmapping APIs for unmapping netmems in tx completion path.

---
 drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h | 3 ++-
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 2 ++
 2 files changed, 4 insertions(+), 1 deletion(-)
Quoted 17 lines

drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma) · line 366

Quoted 11 lines
drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h +2 -1
index e837c21d3d21..6501252359b0 100644
@@ -362,7 +362,8 @@ mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma)
            dma_unmap_single(pdev, dma->addr, dma->size, DMA_TO_DEVICE);
            break;
    case MLX5E_DMA_MAP_PAGE:
-           dma_unmap_page(pdev, dma->addr, dma->size, DMA_TO_DEVICE);
+           netmem_dma_unmap_page_attrs(pdev, dma->addr, dma->size,
+                                       DMA_TO_DEVICE, 0);

For this to work, the dma->addr needs to be 0, so the callers of the dma_map() need to be adjusted as well, or am I missing something? There is netmem_dma_unmap_addr_set to handle that, but I don't see anybody calling it. Do we need to add the following (untested)?

Hmmmm... yes. I figured that skb_frag_dma_map() would do the work but I was wrong, it is not enough.

Quoted 8 lines
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +0 -0
index 55a8629f0792..fb6465210aed 100644
@@ -210,7 +210,9 @@ mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb,
              if (unlikely(dma_mapping_error(sq->pdev, dma_addr)))
                      goto dma_unmap_wqe_err;
 
Quoted 7 lines

drivers/net/ethernet/mellanox/mlx5/core/en_tx.c mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb, · line 215

Quoted 4 lines
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +3 -1
-             dseg->addr       = cpu_to_be64(dma_addr);
+             dseg->addr = 0;
+             if (!netmem_is_net_iov(skb_frag_netmem(frag)))
+                     dseg->addr = cpu_to_be64(dma_addr);

AFAIU we still want to pass the computed dma_address to the data segment to the HW. We only need to make sure in mlx5e_dma_push() to set dma_addr to 0,

yes

Quoted 3 lines

to avoid calling netmem_dma_unmap_page_attrs() with dma->addr 0. Like in the snippet below. Do you agree?

the opposite. You want netmem_dma_unmap_page_attrs() to be called with dma->addr == 0, so that is will skip the dma unmapping.

Quoted 4 lines

We will send a fix patch once the above question is answered. Also, is there a way to test this with more confidence? The ncdevmem tx test passed just fine.

You have to test ncdevmem tx on a platform with iommu enabled. Only in this case the netmem_dma_unmap_page_attrs() may cause a problem, and even then it's not a sure thing. It depends on the type of iommu and type of dmabuf i think.

Quoted 8 lines

Thanks,
Dragos

drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +0 -0
index 55a8629f0792..ecee2e4f678b 100644

drivers/net/ethernet/mellanox/mlx5/core/en_tx.c mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb, · line 221

Quoted 8 lines
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +3 -0
@@ -214,6 +214,9 @@ mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb,
                dseg->lkey       = sq->mkey_be;
                dseg->byte_count = cpu_to_be32(fsz);
 
+               if (!netmem_is_net_iov(skb_frag_netmem(frag)))
+                       dma_addr = 0;
+
                mlx5e_dma_push(sq, dma_addr, fsz, MLX5E_DMA_MAP_PAGE);
                num_dma++;

If you can find a way to do this via netmem_dma_unmap_addr_set, I think that would be better, so you're not relying on a manual netmem_is_net_iov check.

The way you'd do that is you'd pass skb_frag_netmem(frag) to mlx5e_dma_push, and then replace the dma->addr = addr with netmem_dma_unmap_addr_set. But up to you.

If you decide to do a net_iov check and dma_addr = 0, add a comment please.

Signature · 2 lines
Thanks,
Mina
#18 lore raw

Re: [PATCH net-next v6 12/12] net/mlx5e: Add TX support for netmems

Quoted 56 lines · Mina Almasry
Quoted 47 lines · Dragos Tatulea
Quoted 17 lines · Stanislav Fomichev
Quoted 16 lines · Mark Bloch

From: Dragos Tatulea <[email protected]>

Declare netmem TX support in netdev.

As required, use the netmem aware dma unmapping APIs for unmapping netmems in tx completion path.

---
 drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h | 3 ++-
 drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 2 ++
 2 files changed, 4 insertions(+), 1 deletion(-)
Quoted 17 lines

drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma) · line 366

Quoted 11 lines
drivers/net/ethernet/mellanox/mlx5/core/en/txrx.h +2 -1
index e837c21d3d21..6501252359b0 100644
@@ -362,7 +362,8 @@ mlx5e_tx_dma_unmap(struct device *pdev, struct mlx5e_sq_dma *dma)
            dma_unmap_single(pdev, dma->addr, dma->size, DMA_TO_DEVICE);
            break;
    case MLX5E_DMA_MAP_PAGE:
-           dma_unmap_page(pdev, dma->addr, dma->size, DMA_TO_DEVICE);
+           netmem_dma_unmap_page_attrs(pdev, dma->addr, dma->size,
+                                       DMA_TO_DEVICE, 0);

For this to work, the dma->addr needs to be 0, so the callers of the dma_map() need to be adjusted as well, or am I missing something? There is netmem_dma_unmap_addr_set to handle that, but I don't see anybody calling it. Do we need to add the following (untested)?

Hmmmm... yes. I figured that skb_frag_dma_map() would do the work but I was wrong, it is not enough.

Quoted 8 lines
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +0 -0
index 55a8629f0792..fb6465210aed 100644
@@ -210,7 +210,9 @@ mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb,
              if (unlikely(dma_mapping_error(sq->pdev, dma_addr)))
                      goto dma_unmap_wqe_err;
 
Quoted 7 lines

drivers/net/ethernet/mellanox/mlx5/core/en_tx.c mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb, · line 215

Quoted 4 lines
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +3 -1
-             dseg->addr       = cpu_to_be64(dma_addr);
+             dseg->addr = 0;
+             if (!netmem_is_net_iov(skb_frag_netmem(frag)))
+                     dseg->addr = cpu_to_be64(dma_addr);

AFAIU we still want to pass the computed dma_address to the data segment to the HW. We only need to make sure in mlx5e_dma_push() to set dma_addr to 0,

Quoted 9 lines

yes

Quoted 3 lines

to avoid calling netmem_dma_unmap_page_attrs() with dma->addr 0. Like in the snippet below. Do you agree?

the opposite. You want netmem_dma_unmap_page_attrs() to be called with dma->addr == 0, so that is will skip the dma unmapping.

Yes sorry, that's what I meant to say.

Quoted 10 lines
Quoted 4 lines

We will send a fix patch once the above question is answered. Also, is there a way to test this with more confidence? The ncdevmem tx test passed just fine.

You have to test ncdevmem tx on a platform with iommu enabled. Only in this case the netmem_dma_unmap_page_attrs() may cause a problem, and even then it's not a sure thing. It depends on the type of iommu and type of dmabuf i think.

Is it worth adding a WARN_ON_ONCE(netmem_is_net_iov())
in netmem_dma_unmap_page_attrs() after addr check to catch these kinds of misuse?

Quoted 19 lines
Quoted 6 lines
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +0 -0
index 55a8629f0792..ecee2e4f678b 100644

drivers/net/ethernet/mellanox/mlx5/core/en_tx.c mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb, · line 221

Quoted 8 lines
drivers/net/ethernet/mellanox/mlx5/core/en_tx.c +3 -0
@@ -214,6 +214,9 @@ mlx5e_txwqe_build_dsegs(struct mlx5e_txqsq *sq, struct sk_buff *skb,
                dseg->lkey       = sq->mkey_be;
                dseg->byte_count = cpu_to_be32(fsz);
 
+               if (!netmem_is_net_iov(skb_frag_netmem(frag)))
+                       dma_addr = 0;
+
                mlx5e_dma_push(sq, dma_addr, fsz, MLX5E_DMA_MAP_PAGE);
                num_dma++;

If you can find a way to do this via netmem_dma_unmap_addr_set, I think that would be better, so you're not relying on a manual netmem_is_net_iov check.

Quoted 4 lines

The way you'd do that is you'd pass skb_frag_netmem(frag) to mlx5e_dma_push, and then replace the dma->addr = addr with netmem_dma_unmap_addr_set. But up to you.

Thanks for the suggestion. This would require some additional refactoring. I need to play with this to see if it requires a lot of rewiring or not.

Quoted 2 lines

If you decide to do a net_iov check and dma_addr = 0, add a comment please.

Ack.

Thanks,
Dragos

#19 lore raw

Re: [PATCH net-next v6 12/12] net/mlx5e: Add TX support for netmems

Quoted 9 lines · Dragos Tatulea
Quoted 5 lines

You have to test ncdevmem tx on a platform with iommu enabled. Only in this case the netmem_dma_unmap_page_attrs() may cause a problem, and even then it's not a sure thing. It depends on the type of iommu and type of dmabuf i think.

Is it worth adding a WARN_ON_ONCE(netmem_is_net_iov())
in netmem_dma_unmap_page_attrs() after addr check to catch these kinds of misuse?

I would say it's worth it, but it's the same challenge you point to in your reply: netmem_dma_unmap_page_attrs currently doesn't take in a netmem, and it may be a big refactor not worth it if it's callers also don't have a reference to the netmem readily available to pass it.

Signature · 2 lines
Thanks,
Mina
#20 lore raw

Re: [PATCH net-next v6 12/12] net/mlx5e: Add TX support for netmems

In reply to Dragos Tatulea
Quoted 4 lines
Quoted 4 lines

If you can find a way to do this via netmem_dma_unmap_addr_set, I think that would be better, so you're not relying on a manual netmem_is_net_iov check.

Quoted 8 lines
Quoted 4 lines

The way you'd do that is you'd pass skb_frag_netmem(frag) to mlx5e_dma_push, and then replace the dma->addr = addr with netmem_dma_unmap_addr_set. But up to you.

Thanks for the suggestion. This would require some additional refactoring. I need to play with this to see if it requires a lot of rewiring or not.

Got around to this. Found a way to use netmem_dma_unmap_addr_set() with a small refactoring that makes sense. We'll send a patch soon.

Thanks,
Dragos

#21 lore raw
In reply to Mark Bloch
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 Mon, 16 Jun 2025 17:14:29 +0300 you wrote:

This series adds support for zerocopy rx TCP with devmem and io_uring for ConnectX7 NICs and above. For performance reasons and simplicity HW-GRO will also be turned on when header-data split mode is on.

Performance
===========

[...]

Here is the summary with links:
- [net-next,v6,01/12] net: Allow const args for of page_to_netmem()

    https://git.kernel.org/netdev/net-next/c/c9e1225352d4
  - [net-next,v6,02/12] net: Add skb_can_coalesce for netmem
    https://git.kernel.org/netdev/net-next/c/1cbb49f85b40
  - [net-next,v6,03/12] page_pool: Add page_pool_dev_alloc_netmems helper
    https://git.kernel.org/netdev/net-next/c/a202f24b0858
  - [net-next,v6,04/12] net/mlx5e: SHAMPO: Reorganize mlx5_rq_shampo_alloc
    https://git.kernel.org/netdev/net-next/c/af4312c4c9c1
  - [net-next,v6,05/12] net/mlx5e: SHAMPO: Remove redundant params
    https://git.kernel.org/netdev/net-next/c/16142defd304
  - [net-next,v6,06/12] net/mlx5e: SHAMPO: Improve hw gro capability checking
    https://git.kernel.org/netdev/net-next/c/d2760abdedde
  - [net-next,v6,07/12] net/mlx5e: SHAMPO: Separate pool for headers
    https://git.kernel.org/netdev/net-next/c/e225d9bd93ed
  - [net-next,v6,08/12] net/mlx5e: Convert over to netmem
    https://git.kernel.org/netdev/net-next/c/d1668f119943
  - [net-next,v6,09/12] net/mlx5e: Add support for UNREADABLE netmem page pools
    https://git.kernel.org/netdev/net-next/c/db3010bb5a01
  - [net-next,v6,10/12] net/mlx5e: Implement queue mgmt ops and single channel swap
    https://git.kernel.org/netdev/net-next/c/b2588ea40ec9
  - [net-next,v6,11/12] net/mlx5e: Support ethtool tcp-data-split settings
    https://git.kernel.org/netdev/net-next/c/46bcce5dfd33
  - [net-next,v6,12/12] net/mlx5e: Add TX support for netmems
    https://git.kernel.org/netdev/net-next/c/5a842c288cfa

You are awesome, thank you!

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