Skip to contentSkip to conversation

net/mlx5: avoid wildcard drop flow in priority discovery

PATCH Superseded Gavin Hu 4 participants 5 messages Jul 13 – Aug 3 Network drivers

lore
Share & tools
lore

Gavin Hu

author
#0 lore raw

PATCH 1/1 net/mlx5: avoid wildcard drop flow in priority discovery

From: Wei Yan <[email protected]>

Priority discovery installs temporary drop flows to probe supported flow priorities. These flows match a wildcard Ethernet pattern, so they can also match real traffic while discovery is running. On shared devices this may temporarily drop kernel traffic and cause TCP connection timeouts.

Constrain the probe flows to match both source and destination MAC addresses as ff:ff:ff:ff:ff:ff. This keeps the rule valid for discovery while avoiding matches on normal Ethernet traffic.

Fixes 3eca5f8a610e ("net/mlx5: move flow priority discovery to Verbs file")
Fixes c5042f93a425 ("net/mlx5: discover max flow priority using DevX")
---
 drivers/net/mlx5/mlx5_flow_dv.c    | 18 ++++++++++++++++++
 drivers/net/mlx5/mlx5_flow_verbs.c |  8 ++++++++
 2 files changed, 26 insertions(+)
drivers/net/mlx5/mlx5_flow_dv.c +18 -0
index 58ebcf87eb..9cb7c59b83 100644
@@ -19794,6 +19794,22 @@ mlx5_flow_discover_dr_action_support(struct rte_eth_dev *dev)
 	void *matcher = NULL;
 	void *flow = NULL;
 	int ret = -1;
+	struct rte_flow_item_eth eth;
+	struct rte_flow_item item = {
+		.type = RTE_FLOW_ITEM_TYPE_ETH,
+		.spec = &eth,
+		.mask = &eth,
+	};
+
+	memset(&eth, 0, sizeof(eth));
+	memset(&eth.hdr.dst_addr, 0xff, sizeof(eth.hdr.dst_addr));
+	memset(&eth.hdr.src_addr, 0xff, sizeof(eth.hdr.src_addr));
+	flow_dv_translate_item_eth(mask.buf, &item,
+				   /* inner */ false, /* group */ 0,
+				   MLX5_SET_MATCHER_SW_M);
+	flow_dv_translate_item_eth(value.buf, &item,
+				   /* inner */ false, /* group */ 0,
+				   MLX5_SET_MATCHER_SW_V);
 
 	tbl = mlx5_flow_dv_tbl_resource_get(dev, 0, 0, 0, false, NULL,
 					    0, 0, 0, NULL);
@@ -20683,6 +20699,8 @@ flow_dv_discover_priorities(struct rte_eth_dev *dev,
 	flow.dv.actions[0] = action;
 	flow.dv.actions_n = 1;
 	memset(&eth, 0, sizeof(eth));
+	memset(&eth.hdr.dst_addr, 0xff, sizeof(eth.hdr.dst_addr));
+	memset(&eth.hdr.src_addr, 0xff, sizeof(eth.hdr.src_addr));
 	flow_dv_translate_item_eth(matcher.mask.buf, &item,
 				   /* inner */ false, /* group */ 0,
 				   MLX5_SET_MATCHER_SW_M);
drivers/net/mlx5/mlx5_flow_verbs.c +8 -0
index bb240d38d9..f92258cb5a 100644
@@ -65,6 +65,14 @@ flow_verbs_discover_priorities(struct rte_eth_dev *dev,
 		.eth = {
 			.type = IBV_FLOW_SPEC_ETH,
 			.size = sizeof(struct ibv_flow_spec_eth),
+			.val = {
+				.dst_mac = "\xff\xff\xff\xff\xff\xff",
+				.src_mac = "\xff\xff\xff\xff\xff\xff",
+			},
+			.mask = {
+				.dst_mac = "\xff\xff\xff\xff\xff\xff",
+				.src_mac = "\xff\xff\xff\xff\xff\xff",
+			},
 		},
 		.drop = {
 			.size = sizeof(struct ibv_flow_spec_action_drop),
Signature · 1 line
2.27.0
#1 lore raw

13/07/2026 06:18, Gavin Hu:

Quoted 10 lines

From: Wei Yan <[email protected]>

Priority discovery installs temporary drop flows to probe supported flow priorities. These flows match a wildcard Ethernet pattern, so they can also match real traffic while discovery is running. On shared devices this may temporarily drop kernel traffic and cause TCP connection timeouts.

Constrain the probe flows to match both source and destination MAC addresses as ff:ff:ff:ff:ff:ff. This keeps the rule valid for discovery while avoiding matches on normal Ethernet traffic.

It does not compile in all conditions.
Please check CI results.

author
#2 lore raw

PATCH v2 1/1 net/mlx5: avoid wildcard drop flow in priority discovery

In reply to Gavin Hu

From: Wei Yan <[email protected]>

Priority discovery installs temporary drop flows to probe supported flow priorities. These flows match a wildcard Ethernet pattern, so they can also match real traffic while discovery is running. On shared devices this may temporarily drop kernel traffic and cause TCP connection timeouts.

Constrain the probe flows to match both source and destination MAC addresses as ff:ff:ff:ff:ff:ff. This keeps the rule valid for discovery while avoiding matches on normal Ethernet traffic.

Fixes 3eca5f8a610e ("net/mlx5: move flow priority discovery to Verbs file")
Fixes c5042f93a425 ("net/mlx5: discover max flow priority using DevX")

---
+v2:
+* Fixed the compiling issue by some compilers in mlx5_flow_verbs.c.

 drivers/net/mlx5/mlx5_flow_dv.c    | 18 ++++++++++++++++++
 drivers/net/mlx5/mlx5_flow_verbs.c |  8 ++++++++
 2 files changed, 26 insertions(+)
drivers/net/mlx5/mlx5_flow_dv.c +18 -0
index 58ebcf87eb..9cb7c59b83 100644
@@ -19794,6 +19794,22 @@ mlx5_flow_discover_dr_action_support(struct rte_eth_dev *dev)
 	void *matcher = NULL;
 	void *flow = NULL;
 	int ret = -1;
+	struct rte_flow_item_eth eth;
+	struct rte_flow_item item = {
+		.type = RTE_FLOW_ITEM_TYPE_ETH,
+		.spec = &eth,
+		.mask = &eth,
+	};
+
+	memset(&eth, 0, sizeof(eth));
+	memset(&eth.hdr.dst_addr, 0xff, sizeof(eth.hdr.dst_addr));
+	memset(&eth.hdr.src_addr, 0xff, sizeof(eth.hdr.src_addr));
+	flow_dv_translate_item_eth(mask.buf, &item,
+				   /* inner */ false, /* group */ 0,
+				   MLX5_SET_MATCHER_SW_M);
+	flow_dv_translate_item_eth(value.buf, &item,
+				   /* inner */ false, /* group */ 0,
+				   MLX5_SET_MATCHER_SW_V);
 
 	tbl = mlx5_flow_dv_tbl_resource_get(dev, 0, 0, 0, false, NULL,
 					    0, 0, 0, NULL);
@@ -20683,6 +20699,8 @@ flow_dv_discover_priorities(struct rte_eth_dev *dev,
 	flow.dv.actions[0] = action;
 	flow.dv.actions_n = 1;
 	memset(&eth, 0, sizeof(eth));
+	memset(&eth.hdr.dst_addr, 0xff, sizeof(eth.hdr.dst_addr));
+	memset(&eth.hdr.src_addr, 0xff, sizeof(eth.hdr.src_addr));
 	flow_dv_translate_item_eth(matcher.mask.buf, &item,
 				   /* inner */ false, /* group */ 0,
 				   MLX5_SET_MATCHER_SW_M);
drivers/net/mlx5/mlx5_flow_verbs.c +8 -0
index bb240d38d9..eb576f73d4 100644
@@ -65,6 +65,14 @@ flow_verbs_discover_priorities(struct rte_eth_dev *dev,
 		.eth = {
 			.type = IBV_FLOW_SPEC_ETH,
 			.size = sizeof(struct ibv_flow_spec_eth),
+			.val = {
+				.dst_mac = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff},
+				.src_mac = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff},
+			},
+			.mask = {
+				.dst_mac = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff},
+				.src_mac = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff},
+			},
 		},
 		.drop = {
 			.size = sizeof(struct ibv_flow_spec_action_drop),
Signature · 1 line
2.27.0
Acked-by on v2 1/1
#3 lore raw
Quoted 25 lines

-----Original Message-----
From: Gavin Hu <[email protected]>
Sent: Tuesday, July 14, 2026 2:08 PM
To: [email protected]
Cc: Wei Yan <[email protected]>; [email protected]; Dariusz Sosnowski <[email protected]>; Slava Ovsiienko <[email protected]>; Bing Zhao <[email protected]>; Ori Kam <[email protected]>; Suanming Mou <[email protected]>; Matan Azrad <[email protected]>; Ophir Munk <[email protected]>; Dmitry Kozlyuk <[email protected]>
Subject: [PATCH v2] net/mlx5: avoid wildcard drop flow in priority discovery

From: Wei Yan <[email protected]>

Priority discovery installs temporary drop flows to probe supported flow priorities. These flows match a wildcard Ethernet pattern, so they can also match real traffic while discovery is running. On shared devices this may temporarily drop kernel traffic and cause TCP connection timeouts.

Constrain the probe flows to match both source and destination MAC addresses as ff:ff:ff:ff:ff:ff. This keeps the rule valid for discovery while avoiding matches on normal Ethernet traffic.

Quoted 6 lines
Fixes 3eca5f8a610e ("net/mlx5: move flow priority discovery to Verbs file")
Fixes c5042f93a425 ("net/mlx5: discover max flow priority using DevX")

Thanks

#4 lore raw
In reply to Gavin Hu

Hi,

Quoted 19 lines · Gavin Hu

From: Wei Yan <[email protected]>

Priority discovery installs temporary drop flows to probe supported flow priorities. These flows match a wildcard Ethernet pattern, so they can also match real traffic while discovery is running. On shared devices this may temporarily drop kernel traffic and cause TCP connection timeouts.

Constrain the probe flows to match both source and destination MAC addresses as ff:ff:ff:ff:ff:ff. This keeps the rule valid for discovery while avoiding matches on normal Ethernet traffic.

Fixes 3eca5f8a610e ("net/mlx5: move flow priority discovery to Verbs file")
Fixes c5042f93a425 ("net/mlx5: discover max flow priority using DevX")

---

Review

Quoted 2 lines
(quoted diff) +2 -0
+v2:
+* Fixed the compiling issue by some compilers in mlx5_flow_verbs.c.

Patch applied to next-net-mlx,

Kindest regards
Raslan Darawsheh