patches for DPDK stable branches
 help / color / mirror / Atom feed
From: Kevin Traynor <ktraynor@redhat.com>
To: Dariusz Sosnowski <dsosnowski@nvidia.com>
Cc: Bing Zhao <bingz@nvidia.com>, dpdk stable <stable@dpdk.org>
Subject: patch 'net/mlx5: fix indirect RSS action hash' has been queued to stable release 24.11.4
Date: Fri, 21 Nov 2025 11:20:31 +0000	[thread overview]
Message-ID: <20251121112128.485623-47-ktraynor@redhat.com> (raw)
In-Reply-To: <20251121112128.485623-1-ktraynor@redhat.com>

Hi,

FYI, your patch has been queued to stable release 24.11.4

Note it hasn't been pushed to http://dpdk.org/browse/dpdk-stable yet.
It will be pushed if I get no objections before 11/26/25. So please
shout if anyone has objections.

Also note that after the patch there's a diff of the upstream commit vs the
patch applied to the branch. This will indicate if there was any rebasing
needed to apply to the stable branch. If there were code changes for rebasing
(ie: not only metadata diffs), please double check that the rebase was
correctly done.

Queued patches are on a temporary branch at:
https://github.com/kevintraynor/dpdk-stable

This queued commit can be viewed at:
https://github.com/kevintraynor/dpdk-stable/commit/64fc7ef57758ad7b358e294df28c902a7e8b8c40

Thanks.

Kevin

---
From 64fc7ef57758ad7b358e294df28c902a7e8b8c40 Mon Sep 17 00:00:00 2001
From: Dariusz Sosnowski <dsosnowski@nvidia.com>
Date: Thu, 30 Oct 2025 18:24:04 +0100
Subject: [PATCH] net/mlx5: fix indirect RSS action hash

[ upstream commit 6b010880a505c5609355180a7f99df940a163385 ]

Whenever indirect RSS flow action is created,
mlx5 PMD creates an hrxq object (abstraction over HW object
used to configure RSS hashing),
for all possible and supported protocols combinations.
For each combination, the hrxq configuration is adjusted
based on RSS hash types provided by the user
(e.g. hash on source L3 address is removed if user passed
RTE_ETH_RSS_L3_SRC_ONLY in hash types).

Function used for adjustment, flow_dv_action_rss_l34_hash_adjust(),
had a bug. If user requested, for example, hashing over both UDP ports
and only IPv6 source address, then RSS hashing was configured
to hash both IPv6 addresses. Adjustment for RTE_ETH_RSS_L3_SRC_ONLY
was skipped.

In HW Steering mode, this resulted in failures to use such indirect
RSS flow action in flow rules created through template flow API.
In this mode, only a single hrxq object is selected during flow rule
creation, based on actual configuration of RSS hash types
in flow action.
Since hrxq was created without applying RTE_ETH_RSS_L3_SRC_ONLY
adjustment and RSS hash types contained RTE_ETH_RSS_L3_SRC_ONLY,
then no matching hrxq could be found, resulting in rule creation failure.

This issue is addressed by the following:

- Missing adjustments are added to flow_dv_action_rss_l34_hash_adjust()
  function.
  This function is reworked to check each protocol type separately,
  instead of using switch case over all combinations.
- Code for setting/looking up hrxq objects based on RSS hash types
  is reworked. Separate switch cases for possible combinations in each
  function are replaced with a single one.
  Additional logging and assertions are added to flag any invalid
  or missing combinations.

Beside that, the existing set of protocols combinations set
did not cover RSS hashing only over UDP or TCP ports,
which is a valid configuration.
These combinations are added in this patch
(new elements in mlx5_rss_hash_fields array).

Fixes: 212d17b6a650 ("net/mlx5: fix missing shared RSS hash types")

Signed-off-by: Dariusz Sosnowski <dsosnowski@nvidia.com>
Acked-by: Bing Zhao <bingz@nvidia.com>
---
 drivers/net/mlx5/mlx5.h         |   2 -
 drivers/net/mlx5/mlx5_flow.c    |  15 ++
 drivers/net/mlx5/mlx5_flow.h    |  37 ++-
 drivers/net/mlx5/mlx5_flow_dv.c | 410 +++++++++++++++++++-------------
 4 files changed, 287 insertions(+), 177 deletions(-)

diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index f2ddb5c7b4..56eb08bd7c 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -1820,6 +1820,4 @@ struct mlx5_obj_ops {
 };
 
-#define MLX5_RSS_HASH_FIELDS_LEN RTE_DIM(mlx5_rss_hash_fields)
-
 enum mlx5_ctrl_flow_type {
 	MLX5_CTRL_FLOW_TYPE_GENERAL,
diff --git a/drivers/net/mlx5/mlx5_flow.c b/drivers/net/mlx5/mlx5_flow.c
index 58ab218d1c..e5f76aae45 100644
--- a/drivers/net/mlx5/mlx5_flow.c
+++ b/drivers/net/mlx5/mlx5_flow.c
@@ -34,4 +34,19 @@
 #include "rte_pmd_mlx5.h"
 
+const uint64_t mlx5_rss_hash_fields[] = {
+	[MLX5_RSS_HASH_IDX_IPV4]     = MLX5_RSS_HASH_IPV4,
+	[MLX5_RSS_HASH_IDX_IPV4_TCP] = MLX5_RSS_HASH_IPV4_TCP,
+	[MLX5_RSS_HASH_IDX_IPV4_UDP] = MLX5_RSS_HASH_IPV4_UDP,
+	[MLX5_RSS_HASH_IDX_IPV4_ESP] = MLX5_RSS_HASH_IPV4_ESP,
+	[MLX5_RSS_HASH_IDX_IPV6]     = MLX5_RSS_HASH_IPV6,
+	[MLX5_RSS_HASH_IDX_IPV6_TCP] = MLX5_RSS_HASH_IPV6_TCP,
+	[MLX5_RSS_HASH_IDX_IPV6_UDP] = MLX5_RSS_HASH_IPV6_UDP,
+	[MLX5_RSS_HASH_IDX_IPV6_ESP] = MLX5_RSS_HASH_IPV6_ESP,
+	[MLX5_RSS_HASH_IDX_TCP]      = MLX5_TCP_IBV_RX_HASH,
+	[MLX5_RSS_HASH_IDX_UDP]      = MLX5_UDP_IBV_RX_HASH,
+	[MLX5_RSS_HASH_IDX_ESP_SPI]  = MLX5_RSS_HASH_ESP_SPI,
+	[MLX5_RSS_HASH_IDX_NONE]     = MLX5_RSS_HASH_NONE,
+};
+
 /*
  * Shared array for quick translation between port_id and vport mask/values
diff --git a/drivers/net/mlx5/mlx5_flow.h b/drivers/net/mlx5/mlx5_flow.h
index aa8ea54c66..c8ab114655 100644
--- a/drivers/net/mlx5/mlx5_flow.h
+++ b/drivers/net/mlx5/mlx5_flow.h
@@ -1865,18 +1865,29 @@ flow_hw_get_reg_id_from_ctx(void *dr_ctx, enum rte_flow_item_type type,
 		 ((func) == RTE_ETH_HASH_FUNCTION_SYMMETRIC_TOEPLITZ_SORT))
 
-/* array of valid combinations of RX Hash fields for RSS */
-static const uint64_t mlx5_rss_hash_fields[] = {
-	MLX5_RSS_HASH_IPV4,
-	MLX5_RSS_HASH_IPV4_TCP,
-	MLX5_RSS_HASH_IPV4_UDP,
-	MLX5_RSS_HASH_IPV4_ESP,
-	MLX5_RSS_HASH_IPV6,
-	MLX5_RSS_HASH_IPV6_TCP,
-	MLX5_RSS_HASH_IPV6_UDP,
-	MLX5_RSS_HASH_IPV6_ESP,
-	MLX5_RSS_HASH_ESP_SPI,
-	MLX5_RSS_HASH_NONE,
+
+/**
+ * Each enum variant corresponds to a single valid protocols combination for hrxq configuration
+ * Each variant serves as an index into #mlx5_rss_hash_fields array containing default
+ * bitmaps of ibv_rx_hash_fields flags for given protocols combination.
+ */
+enum {
+	MLX5_RSS_HASH_IDX_IPV4,
+	MLX5_RSS_HASH_IDX_IPV4_TCP,
+	MLX5_RSS_HASH_IDX_IPV4_UDP,
+	MLX5_RSS_HASH_IDX_IPV4_ESP,
+	MLX5_RSS_HASH_IDX_IPV6,
+	MLX5_RSS_HASH_IDX_IPV6_TCP,
+	MLX5_RSS_HASH_IDX_IPV6_UDP,
+	MLX5_RSS_HASH_IDX_IPV6_ESP,
+	MLX5_RSS_HASH_IDX_TCP,
+	MLX5_RSS_HASH_IDX_UDP,
+	MLX5_RSS_HASH_IDX_ESP_SPI,
+	MLX5_RSS_HASH_IDX_NONE,
+	MLX5_RSS_HASH_IDX_MAX,
 };
 
+/** Array of valid combinations of RX Hash fields for RSS. */
+extern const uint64_t mlx5_rss_hash_fields[];
+
 /* Shared RSS action structure */
 struct mlx5_shared_action_rss {
@@ -1887,5 +1898,5 @@ struct mlx5_shared_action_rss {
 	struct mlx5_ind_table_obj *ind_tbl;
 	/**< Hash RX queues (hrxq, hrxq_tunnel fields) indirection table. */
-	uint32_t hrxq[MLX5_RSS_HASH_FIELDS_LEN];
+	uint32_t hrxq[MLX5_RSS_HASH_IDX_MAX];
 	/**< Hash RX queue indexes mapped to mlx5_rss_hash_fields */
 	rte_spinlock_t action_rss_sl; /**< Shared RSS action spinlock. */
diff --git a/drivers/net/mlx5/mlx5_flow_dv.c b/drivers/net/mlx5/mlx5_flow_dv.c
index 192dfd282d..006684d5bd 100644
--- a/drivers/net/mlx5/mlx5_flow_dv.c
+++ b/drivers/net/mlx5/mlx5_flow_dv.c
@@ -15775,4 +15775,143 @@ flow_dv_translate(struct rte_eth_dev *dev,
 }
 
+/*
+ * Protocol selector bitmap
+ * Each flag is used as an indicator that given protocol is specified in given RSS hash fields.
+ */
+#define RX_HASH_SELECTOR_IPV4 RTE_BIT32(0)
+#define RX_HASH_SELECTOR_IPV6 RTE_BIT32(1)
+#define RX_HASH_SELECTOR_UDP RTE_BIT32(2)
+#define RX_HASH_SELECTOR_TCP RTE_BIT32(3)
+#define RX_HASH_SELECTOR_ESP_SPI RTE_BIT32(4)
+#define RX_HASH_SELECTOR_NONE (0)
+
+#define RX_HASH_SELECTOR_IPV4_TCP (RX_HASH_SELECTOR_IPV4 | RX_HASH_SELECTOR_TCP)
+#define RX_HASH_SELECTOR_IPV4_UDP (RX_HASH_SELECTOR_IPV4 | RX_HASH_SELECTOR_UDP)
+#define RX_HASH_SELECTOR_IPV4_ESP (RX_HASH_SELECTOR_IPV4 | RX_HASH_SELECTOR_ESP_SPI)
+
+#define RX_HASH_SELECTOR_IPV6_TCP (RX_HASH_SELECTOR_IPV6 | RX_HASH_SELECTOR_TCP)
+#define RX_HASH_SELECTOR_IPV6_UDP (RX_HASH_SELECTOR_IPV6 | RX_HASH_SELECTOR_UDP)
+#define RX_HASH_SELECTOR_IPV6_ESP (RX_HASH_SELECTOR_IPV6 | RX_HASH_SELECTOR_ESP_SPI)
+
+static bool
+rx_hash_selector_has_valid_l3(const uint32_t selectors)
+{
+	/* In TIR configuration, RSS hashing on both IPv4 and IPv6 is mutually exclusive. */
+	return !((selectors & RX_HASH_SELECTOR_IPV4) && (selectors & RX_HASH_SELECTOR_IPV6));
+}
+
+static bool
+rx_hash_selector_has_valid_l4(const uint32_t selectors)
+{
+	/* In TIR configuration, RSS hashing on both UDP and TCP is mutually exclusive. */
+	return !((selectors & RX_HASH_SELECTOR_UDP) && (selectors & RX_HASH_SELECTOR_TCP));
+}
+
+static bool
+rx_hash_selector_has_valid_esp(const uint32_t selectors)
+{
+	/* In TIR configuration, RSS hashing on ESP and other L4 protocol is mutually exclusive. */
+	if (selectors & RX_HASH_SELECTOR_ESP_SPI)
+		return !((selectors & RX_HASH_SELECTOR_UDP) || (selectors & RX_HASH_SELECTOR_TCP));
+
+	return true;
+}
+
+/**
+ * Calculate protocol combination based on provided RSS hashing fields.
+ *
+ * @param[in] hash_fields
+ *   Requested RSS hashing fields specified as a flags bitmap, based on ibv_rx_hash_fields.
+ * @param[out] selectors_out
+ *   Calculated protocol combination will be written here.
+ *   Result will be a bitmap of RX_HASH_SELECTOR_* flags.
+ *
+ * @return
+ *   0 if conversion is successful and protocol combination written to @p selectors_out.
+ *   (-EINVAL) otherwise.
+ */
+static int
+rx_hash_calc_selector(const uint64_t hash_fields, uint32_t *selectors_out)
+{
+	const uint64_t filtered_hf = hash_fields & ~IBV_RX_HASH_INNER;
+	uint32_t selectors = 0;
+
+	if (filtered_hf & MLX5_RSS_HASH_IPV4)
+		selectors |= RX_HASH_SELECTOR_IPV4;
+	if (filtered_hf & MLX5_RSS_HASH_IPV6)
+		selectors |= RX_HASH_SELECTOR_IPV6;
+	if (!rx_hash_selector_has_valid_l3(selectors)) {
+		DRV_LOG(NOTICE, "hrxq hashing on both IPv4 and IPv6 is invalid: "
+				"selectors=0x%" PRIx32, selectors);
+		return -EINVAL;
+	}
+
+	if (filtered_hf & MLX5_UDP_IBV_RX_HASH)
+		selectors |= RX_HASH_SELECTOR_UDP;
+	if (filtered_hf & MLX5_TCP_IBV_RX_HASH)
+		selectors |= RX_HASH_SELECTOR_TCP;
+	if (!rx_hash_selector_has_valid_l4(selectors)) {
+		DRV_LOG(NOTICE, "hrxq hashing on both UDP and TCP is invalid: "
+				"selectors=0x%" PRIx32, selectors);
+		return -EINVAL;
+	}
+
+	if (filtered_hf & MLX5_RSS_HASH_ESP_SPI)
+		selectors |= RX_HASH_SELECTOR_ESP_SPI;
+	if (!rx_hash_selector_has_valid_esp(selectors)) {
+		DRV_LOG(NOTICE, "hrxq hashing on ESP SPI and UDP or TCP is mutually exclusive: "
+				"selectors=0x%" PRIx32, selectors);
+		return -EINVAL;
+	}
+
+	*selectors_out = selectors;
+	return 0;
+}
+
+/**
+ * Calculate the hrxq object index based on protocol combination.
+ *
+ * @param[in] selectors
+ *   Protocol combination specified as bitmap of RX_HASH_SELECTOR_* flags.
+ *
+ * @return
+ *   Index into hrxq array in #mlx5_shared_action_rss based on ginve protocol combination.
+ *   (-EINVAL) if given protocol combination is not supported or is invalid.
+ */
+static int
+get_rss_hash_idx(const uint32_t selectors)
+{
+	switch (selectors) {
+	case RX_HASH_SELECTOR_IPV4:
+		return MLX5_RSS_HASH_IDX_IPV4;
+	case RX_HASH_SELECTOR_IPV4_TCP:
+		return MLX5_RSS_HASH_IDX_IPV4_TCP;
+	case RX_HASH_SELECTOR_IPV4_UDP:
+		return MLX5_RSS_HASH_IDX_IPV4_UDP;
+	case RX_HASH_SELECTOR_IPV4_ESP:
+		return MLX5_RSS_HASH_IDX_IPV4_ESP;
+	case RX_HASH_SELECTOR_IPV6:
+		return MLX5_RSS_HASH_IDX_IPV6;
+	case RX_HASH_SELECTOR_IPV6_TCP:
+		return MLX5_RSS_HASH_IDX_IPV6_TCP;
+	case RX_HASH_SELECTOR_IPV6_UDP:
+		return MLX5_RSS_HASH_IDX_IPV6_UDP;
+	case RX_HASH_SELECTOR_IPV6_ESP:
+		return MLX5_RSS_HASH_IDX_IPV6_ESP;
+	case RX_HASH_SELECTOR_TCP:
+		return MLX5_RSS_HASH_IDX_TCP;
+	case RX_HASH_SELECTOR_UDP:
+		return MLX5_RSS_HASH_IDX_UDP;
+	case RX_HASH_SELECTOR_ESP_SPI:
+		return MLX5_RSS_HASH_IDX_ESP_SPI;
+	case RX_HASH_SELECTOR_NONE:
+		return MLX5_RSS_HASH_IDX_NONE;
+	default:
+		DRV_LOG(ERR, "invalid hrxq hash fields combination: "
+			     "selectors=0x%" PRIx32, selectors);
+		return -EINVAL;
+	}
+}
+
 /**
  * Set hash RX queue by hash fields (see enum ibv_rx_hash_fields)
@@ -15782,5 +15921,6 @@ flow_dv_translate(struct rte_eth_dev *dev,
  *   Shred RSS action holding hash RX queue objects.
  * @param[in] hash_fields
- *   Defines combination of packet fields to participate in RX hash.
+ *   Defines combination of packet fields to participate in RX hash,
+ *   specified as a bitmap of #ibv_rx_hash_fields flags.
  * @param[in] tunnel
  *   Tunnel type
@@ -15797,63 +15937,24 @@ __flow_dv_action_rss_hrxq_set(struct mlx5_shared_action_rss *action,
 {
 	uint32_t *hrxqs = action->hrxq;
+	uint32_t selectors = 0;
+	int ret;
 
-	switch (hash_fields & ~IBV_RX_HASH_INNER) {
-	case MLX5_RSS_HASH_IPV4:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_SRC_ONLY:
-		hrxqs[0] = hrxq_idx;
-		return 0;
-	case MLX5_RSS_HASH_IPV4_TCP:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_TCP_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_TCP_SRC_ONLY:
-		hrxqs[1] = hrxq_idx;
-		return 0;
-	case MLX5_RSS_HASH_IPV4_UDP:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_UDP_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_UDP_SRC_ONLY:
-		hrxqs[2] = hrxq_idx;
-		return 0;
-	case MLX5_RSS_HASH_IPV6:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_SRC_ONLY:
-		hrxqs[3] = hrxq_idx;
-		return 0;
-	case MLX5_RSS_HASH_IPV6_TCP:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_TCP_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_TCP_SRC_ONLY:
-		hrxqs[4] = hrxq_idx;
-		return 0;
-	case MLX5_RSS_HASH_IPV6_UDP:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_UDP_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_UDP_SRC_ONLY:
-		hrxqs[5] = hrxq_idx;
-		return 0;
-	case MLX5_RSS_HASH_NONE:
-		hrxqs[6] = hrxq_idx;
-		return 0;
-	case MLX5_RSS_HASH_IPV4_ESP:
-		hrxqs[7] = hrxq_idx;
-		return 0;
-	case MLX5_RSS_HASH_IPV6_ESP:
-		hrxqs[8] = hrxq_idx;
-		return 0;
-	case MLX5_RSS_HASH_ESP_SPI:
-		hrxqs[9] = hrxq_idx;
-		return 0;
-	default:
-		return -1;
-	}
+	ret = rx_hash_calc_selector(hash_fields, &selectors);
+	/*
+	 * Hash fields passed to this function are constructed internally.
+	 * If this fails, then this is a PMD bug.
+	 */
+	MLX5_ASSERT(ret == 0);
+
+	ret = get_rss_hash_idx(selectors);
+	/*
+	 * Based on above assert, selectors should always yield correct index
+	 * in mlx5_rss_hash_fields array.
+	 * If this fails, then this is a PMD bug.
+	 */
+	MLX5_ASSERT(ret >= 0 && ret < MLX5_RSS_HASH_IDX_MAX);
+	hrxqs[ret] = hrxq_idx;
+
+	return 0;
 }
 
@@ -15867,5 +15968,6 @@ __flow_dv_action_rss_hrxq_set(struct mlx5_shared_action_rss *action,
  *   Shared RSS action ID holding hash RX queue objects.
  * @param[in] hash_fields
- *   Defines combination of packet fields to participate in RX hash.
+ *   Defines combination of packet fields to participate in RX hash,
+ *   specified as a bitmap of #ibv_rx_hash_fields flags.
  * @param[in] tunnel
  *   Tunnel type
@@ -15882,54 +15984,24 @@ flow_dv_action_rss_hrxq_lookup(struct rte_eth_dev *dev, uint32_t idx,
 	    mlx5_ipool_get(priv->sh->ipool[MLX5_IPOOL_RSS_SHARED_ACTIONS], idx);
 	const uint32_t *hrxqs = shared_rss->hrxq;
+	uint32_t selectors = 0;
+	int ret;
 
-	switch (hash_fields & ~IBV_RX_HASH_INNER) {
-	case MLX5_RSS_HASH_IPV4:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_SRC_ONLY:
-		return hrxqs[0];
-	case MLX5_RSS_HASH_IPV4_TCP:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_TCP_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_TCP_SRC_ONLY:
-		return hrxqs[1];
-	case MLX5_RSS_HASH_IPV4_UDP:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_UDP_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV4_UDP_SRC_ONLY:
-		return hrxqs[2];
-	case MLX5_RSS_HASH_IPV6:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_SRC_ONLY:
-		return hrxqs[3];
-	case MLX5_RSS_HASH_IPV6_TCP:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_TCP_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_TCP_SRC_ONLY:
-		return hrxqs[4];
-	case MLX5_RSS_HASH_IPV6_UDP:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_UDP_DST_ONLY:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_UDP_SRC_ONLY:
-		return hrxqs[5];
-	case MLX5_RSS_HASH_NONE:
-		return hrxqs[6];
-	case MLX5_RSS_HASH_IPV4_ESP:
-		return hrxqs[7];
-	case MLX5_RSS_HASH_IPV6_ESP:
-		return hrxqs[8];
-	case MLX5_RSS_HASH_ESP_SPI:
-		return hrxqs[9];
-	default:
+	ret = rx_hash_calc_selector(hash_fields, &selectors);
+	if (ret < 0) {
+		DRV_LOG(ERR, "port %u Rx hash selector calculation failed: "
+			     "rss_act_idx=%u hash_fields=0x%" PRIx64 " selectors=0x%" PRIx32,
+			dev->data->port_id, idx, hash_fields, selectors);
 		return 0;
 	}
 
+	ret = get_rss_hash_idx(selectors);
+	if (ret < 0) {
+		DRV_LOG(ERR, "port %u failed hrxq index lookup: "
+			     "rss_act_idx=%u hash_fields=0x%" PRIx64 " selectors=0x%" PRIx32,
+			dev->data->port_id, idx, hash_fields, selectors);
+		return 0;
+	}
+
+	return hrxqs[ret];
 }
 
@@ -16621,5 +16693,5 @@ flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
 static int
 __flow_dv_hrxqs_release(struct rte_eth_dev *dev,
-			uint32_t (*hrxqs)[MLX5_RSS_HASH_FIELDS_LEN])
+			uint32_t (*hrxqs)[MLX5_RSS_HASH_IDX_MAX])
 {
 	size_t i;
@@ -16656,4 +16728,60 @@ __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
 }
 
+static inline void
+filter_ipv4_types(uint64_t rss_types, uint64_t *hash_fields)
+{
+	if (rss_types & MLX5_IPV4_LAYER_TYPES) {
+		*hash_fields &= ~MLX5_RSS_HASH_IPV4;
+		if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
+			*hash_fields |= IBV_RX_HASH_DST_IPV4;
+		else if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
+			*hash_fields |= IBV_RX_HASH_SRC_IPV4;
+		else
+			*hash_fields |= MLX5_RSS_HASH_IPV4;
+	}
+}
+
+static inline void
+filter_ipv6_types(uint64_t rss_types, uint64_t *hash_fields)
+{
+	if (rss_types & MLX5_IPV6_LAYER_TYPES) {
+		*hash_fields &= ~MLX5_RSS_HASH_IPV6;
+		if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
+			*hash_fields |= IBV_RX_HASH_DST_IPV6;
+		else if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
+			*hash_fields |= IBV_RX_HASH_SRC_IPV6;
+		else
+			*hash_fields |= MLX5_RSS_HASH_IPV6;
+	}
+}
+
+static inline void
+filter_udp_types(uint64_t rss_types, uint64_t *hash_fields)
+{
+	if (rss_types & RTE_ETH_RSS_UDP) {
+		*hash_fields &= ~MLX5_UDP_IBV_RX_HASH;
+		if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
+			*hash_fields |= IBV_RX_HASH_DST_PORT_UDP;
+		else if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
+			*hash_fields |= IBV_RX_HASH_SRC_PORT_UDP;
+		else
+			*hash_fields |= MLX5_UDP_IBV_RX_HASH;
+	}
+}
+
+static inline void
+filter_tcp_types(uint64_t rss_types, uint64_t *hash_fields)
+{
+	if (rss_types & RTE_ETH_RSS_TCP) {
+		*hash_fields &= ~MLX5_TCP_IBV_RX_HASH;
+		if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
+			*hash_fields |= IBV_RX_HASH_DST_PORT_TCP;
+		else if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
+			*hash_fields |= IBV_RX_HASH_SRC_PORT_TCP;
+		else
+			*hash_fields |= MLX5_TCP_IBV_RX_HASH;
+	}
+}
+
 /**
  * Adjust L3/L4 hash value of pre-created shared RSS hrxq according to
@@ -16667,7 +16795,7 @@ __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
  *
  * @param[in] orig_rss_types
- *   RSS type as provided in shared RSS action.
+ *   RSS type as provided in shared RSS action, specified as a bitmap of RTE_ETH_RSS_* flags.
  * @param[in, out] hash_field
- *   hash_field variable needed to be adjusted.
+ *   hash_field variable needed to be adjusted, specified as a bitmap of #ibv_rx_hash_fields flags.
  *
  * @return
@@ -16678,58 +16806,16 @@ flow_dv_action_rss_l34_hash_adjust(uint64_t orig_rss_types,
 				   uint64_t *hash_field)
 {
+	uint64_t hash_field_protos = *hash_field & ~IBV_RX_HASH_INNER;
 	uint64_t rss_types = rte_eth_rss_hf_refine(orig_rss_types);
 
-	switch (*hash_field & ~IBV_RX_HASH_INNER) {
-	case MLX5_RSS_HASH_IPV4:
-		if (rss_types & MLX5_IPV4_LAYER_TYPES) {
-			*hash_field &= ~MLX5_RSS_HASH_IPV4;
-			if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
-				*hash_field |= IBV_RX_HASH_DST_IPV4;
-			else if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
-				*hash_field |= IBV_RX_HASH_SRC_IPV4;
-			else
-				*hash_field |= MLX5_RSS_HASH_IPV4;
-		}
-		return;
-	case MLX5_RSS_HASH_IPV6:
-		if (rss_types & MLX5_IPV6_LAYER_TYPES) {
-			*hash_field &= ~MLX5_RSS_HASH_IPV6;
-			if (rss_types & RTE_ETH_RSS_L3_DST_ONLY)
-				*hash_field |= IBV_RX_HASH_DST_IPV6;
-			else if (rss_types & RTE_ETH_RSS_L3_SRC_ONLY)
-				*hash_field |= IBV_RX_HASH_SRC_IPV6;
-			else
-				*hash_field |= MLX5_RSS_HASH_IPV6;
-		}
-		return;
-	case MLX5_RSS_HASH_IPV4_UDP:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_UDP:
-		if (rss_types & RTE_ETH_RSS_UDP) {
-			*hash_field &= ~MLX5_UDP_IBV_RX_HASH;
-			if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
-				*hash_field |= IBV_RX_HASH_DST_PORT_UDP;
-			else if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
-				*hash_field |= IBV_RX_HASH_SRC_PORT_UDP;
-			else
-				*hash_field |= MLX5_UDP_IBV_RX_HASH;
-		}
-		return;
-	case MLX5_RSS_HASH_IPV4_TCP:
-		/* fall-through. */
-	case MLX5_RSS_HASH_IPV6_TCP:
-		if (rss_types & RTE_ETH_RSS_TCP) {
-			*hash_field &= ~MLX5_TCP_IBV_RX_HASH;
-			if (rss_types & RTE_ETH_RSS_L4_DST_ONLY)
-				*hash_field |= IBV_RX_HASH_DST_PORT_TCP;
-			else if (rss_types & RTE_ETH_RSS_L4_SRC_ONLY)
-				*hash_field |= IBV_RX_HASH_SRC_PORT_TCP;
-			else
-				*hash_field |= MLX5_TCP_IBV_RX_HASH;
-		}
-		return;
-	default:
-		return;
-	}
+	if (hash_field_protos & MLX5_RSS_HASH_IPV4)
+		filter_ipv4_types(rss_types, hash_field);
+	else if (hash_field_protos & MLX5_RSS_HASH_IPV6)
+		filter_ipv6_types(rss_types, hash_field);
+
+	if (hash_field_protos & MLX5_UDP_IBV_RX_HASH)
+		filter_udp_types(rss_types, hash_field);
+	else if (hash_field_protos & MLX5_TCP_IBV_RX_HASH)
+		filter_tcp_types(rss_types, hash_field);
 }
 
@@ -16783,5 +16869,5 @@ __flow_dv_action_rss_setup(struct rte_eth_dev *dev,
 	if (priv->sh->config.dv_flow_en == 2)
 		rss_desc.hws_flags = MLX5DR_ACTION_FLAG_HWS_RX;
-	for (i = 0; i < MLX5_RSS_HASH_FIELDS_LEN; i++) {
+	for (i = 0; i < MLX5_RSS_HASH_IDX_MAX; i++) {
 		struct mlx5_hrxq *hrxq;
 		uint64_t hash_fields = mlx5_rss_hash_fields[i];
-- 
2.51.0

---
  Diff of the applied patch vs upstream commit (please double-check if non-empty:
---
--- -	2025-11-21 11:05:11.075027049 +0000
+++ 0047-net-mlx5-fix-indirect-RSS-action-hash.patch	2025-11-21 11:05:09.479201238 +0000
@@ -1 +1 @@
-From 6b010880a505c5609355180a7f99df940a163385 Mon Sep 17 00:00:00 2001
+From 64fc7ef57758ad7b358e294df28c902a7e8b8c40 Mon Sep 17 00:00:00 2001
@@ -5,0 +6,2 @@
+[ upstream commit 6b010880a505c5609355180a7f99df940a163385 ]
+
@@ -49 +50,0 @@
-Cc: stable@dpdk.org
@@ -61 +62 @@
-index 203cf00596..7e4bfacd11 100644
+index f2ddb5c7b4..56eb08bd7c 100644
@@ -64 +65 @@
-@@ -1843,6 +1843,4 @@ struct mlx5_obj_ops {
+@@ -1820,6 +1820,4 @@ struct mlx5_obj_ops {
@@ -72 +73 @@
-index 098ef9d034..ed67a90a22 100644
+index 58ab218d1c..e5f76aae45 100644
@@ -75 +76 @@
-@@ -35,4 +35,19 @@
+@@ -34,4 +34,19 @@
@@ -96 +97 @@
-index ef743fc3cb..2de0f35815 100644
+index aa8ea54c66..c8ab114655 100644
@@ -99 +100 @@
-@@ -1898,18 +1898,29 @@ flow_hw_get_reg_id_from_ctx(void *dr_ctx, enum rte_flow_item_type type,
+@@ -1865,18 +1865,29 @@ flow_hw_get_reg_id_from_ctx(void *dr_ctx, enum rte_flow_item_type type,
@@ -141 +142 @@
-@@ -1920,5 +1931,5 @@ struct mlx5_shared_action_rss {
+@@ -1887,5 +1898,5 @@ struct mlx5_shared_action_rss {
@@ -149 +150 @@
-index 1564bd7cbe..f765f94116 100644
+index 192dfd282d..006684d5bd 100644
@@ -152 +153 @@
-@@ -15789,4 +15789,143 @@ flow_dv_translate(struct rte_eth_dev *dev,
+@@ -15775,4 +15775,143 @@ flow_dv_translate(struct rte_eth_dev *dev,
@@ -296 +297 @@
-@@ -15796,5 +15935,6 @@ flow_dv_translate(struct rte_eth_dev *dev,
+@@ -15782,5 +15921,6 @@ flow_dv_translate(struct rte_eth_dev *dev,
@@ -304 +305 @@
-@@ -15811,63 +15951,24 @@ __flow_dv_action_rss_hrxq_set(struct mlx5_shared_action_rss *action,
+@@ -15797,63 +15937,24 @@ __flow_dv_action_rss_hrxq_set(struct mlx5_shared_action_rss *action,
@@ -387 +388 @@
-@@ -15881,5 +15982,6 @@ __flow_dv_action_rss_hrxq_set(struct mlx5_shared_action_rss *action,
+@@ -15867,5 +15968,6 @@ __flow_dv_action_rss_hrxq_set(struct mlx5_shared_action_rss *action,
@@ -395 +396 @@
-@@ -15896,54 +15998,24 @@ flow_dv_action_rss_hrxq_lookup(struct rte_eth_dev *dev, uint32_t idx,
+@@ -15882,54 +15984,24 @@ flow_dv_action_rss_hrxq_lookup(struct rte_eth_dev *dev, uint32_t idx,
@@ -466 +467 @@
-@@ -16635,5 +16707,5 @@ flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
+@@ -16621,5 +16693,5 @@ flow_dv_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
@@ -473 +474 @@
-@@ -16670,4 +16742,60 @@ __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
+@@ -16656,4 +16728,60 @@ __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
@@ -534 +535 @@
-@@ -16681,7 +16809,7 @@ __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
+@@ -16667,7 +16795,7 @@ __flow_dv_action_rss_hrxqs_release(struct rte_eth_dev *dev,
@@ -544 +545 @@
-@@ -16692,58 +16820,16 @@ flow_dv_action_rss_l34_hash_adjust(uint64_t orig_rss_types,
+@@ -16678,58 +16806,16 @@ flow_dv_action_rss_l34_hash_adjust(uint64_t orig_rss_types,
@@ -613 +614 @@
-@@ -16797,5 +16883,5 @@ __flow_dv_action_rss_setup(struct rte_eth_dev *dev,
+@@ -16783,5 +16869,5 @@ __flow_dv_action_rss_setup(struct rte_eth_dev *dev,


  parent reply	other threads:[~2025-11-21 11:23 UTC|newest]

Thread overview: 100+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-11-21 11:19 patch 'test/hash: check memory allocation' " Kevin Traynor
2025-11-21 11:19 ` patch 'dmadev: fix debug build with tracepoints' " Kevin Traynor
2025-11-21 11:19 ` patch 'bus/cdx: fix device name in probing error message' " Kevin Traynor
2025-11-21 11:19 ` patch 'bus/cdx: fix release in probing for secondary process' " Kevin Traynor
2025-11-21 11:19 ` patch 'buildtools/pmdinfogen: fix warning with python 3.14' " Kevin Traynor
2025-11-21 11:19 ` patch 'net/iavf: fix build with clang 21' " Kevin Traynor
2025-11-21 11:19 ` patch 'test: " Kevin Traynor
2025-11-21 11:19 ` patch 'app/eventdev: " Kevin Traynor
2025-11-21 11:19 ` patch 'eventdev/crypto: " Kevin Traynor
2025-11-21 11:19 ` patch 'rawdev: " Kevin Traynor
2025-11-21 11:19 ` patch 'vdpa/mlx5: remove unused constant' " Kevin Traynor
2025-11-21 11:19 ` patch 'crypto/mlx5: remove unused constants' " Kevin Traynor
2025-11-21 11:19 ` patch 'regex/mlx5: remove useless " Kevin Traynor
2025-11-21 11:19 ` patch 'common/mlx5: " Kevin Traynor
2025-11-21 11:19 ` patch 'net/mlx5: " Kevin Traynor
2025-11-21 11:20 ` patch 'net/mlx5: remove unused macros' " Kevin Traynor
2025-11-21 11:20 ` patch 'doc: fix NVIDIA bifurcated driver presentation link' " Kevin Traynor
2025-11-21 11:20 ` patch 'app/dma-perf: fix use after free' " Kevin Traynor
2025-11-21 11:20 ` patch 'app/dma-perf: check buffer size' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/vmxnet3: disable RSS for single queue for ESX8.0+' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/dpaa: fix resource leak' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: fix checksum error counter' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/ngbe: " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: reduce memory size of ring descriptors' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/ngbe: " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: fix VF Rx buffer size in config register' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: remove duplicate Tx queue assignment' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: add device arguments for FDIR' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: fix maximum number of FDIR filters' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: fix FDIR mode clearing' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: fix FDIR drop action for L4 match packets' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: fix FDIR filter for SCTP tunnel' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: filter FDIR match flex bytes for " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: fix FDIR rule raw relative for L3 packets' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: fix FDIR input mask' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: switch to FDIR when ntuple filter is full' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/txgbe: remove unsupported flow action mark' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/nfp: fix metering cleanup' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/bonding: fix MAC address propagation in 802.3ad mode' " Kevin Traynor
2025-11-21 11:20 ` patch 'app/testpmd: fix DCB Tx port' " Kevin Traynor
2025-11-21 11:20 ` patch 'app/testpmd: fix DCB Rx queues' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/e1000/base: fix crash on init with GCC 13' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/fm10k: fix build with GCC 16' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/mlx4: fix unnecessary comma' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/mlx5: fix unnecessary commas' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/mlx5: store MTU at Rx queue allocation time' " Kevin Traynor
2025-11-21 11:20 ` Kevin Traynor [this message]
2025-11-21 11:20 ` patch 'net/mlx5: remove counter alignment' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/mlx5: fix external queues access' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/mlx5: fix modify field action restriction' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/mlx5: fix meter mark allocation' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/mlx5: fix indirect meter index leak' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/mlx5: fix error reporting on masked indirect actions' " Kevin Traynor
2025-11-21 11:20 ` patch 'vhost: fix external buffer in VDUSE' " Kevin Traynor
2025-11-21 11:20 ` patch 'net: fix L2 length for GRE packets' " Kevin Traynor
2025-11-21 11:20 ` patch 'graph: fix xstats description allocation' " Kevin Traynor
2025-11-21 11:20 ` patch 'graph: fix updating edge with active graph' " Kevin Traynor
2025-11-21 11:20 ` patch 'app/pdump: remove hard-coded memory channels' " Kevin Traynor
2025-11-21 11:20 ` patch 'pdump: handle primary process exit' " Kevin Traynor
2025-11-21 11:20 ` patch 'telemetry: make socket handler typedef private' " Kevin Traynor
2025-11-21 11:20 ` patch 'usertools/telemetry: fix exporter default IP binding' " Kevin Traynor
2025-11-21 11:20 ` patch 'examples/l3fwd-power: fix telemetry command registration' " Kevin Traynor
2025-11-21 11:20 ` patch 'lib: fix backticks matching in Doxygen comments' " Kevin Traynor
2025-11-21 11:20 ` patch 'mcslock: fix memory ordering' " Kevin Traynor
2025-11-21 15:22   ` Wathsala Vithanage
2025-11-21 11:20 ` patch 'net/axgbe: fix build with GCC 16' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/dpaa2: fix duplicate call of close' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/dpaa2: clear active VDQ state when freeing Rx queues' " Kevin Traynor
2025-11-21 11:20 ` patch 'app/testpmd: fix flex item link parsing' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/ixgbe/base: fix PF link state request size' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/ice: fix path selection for QinQ Tx offload' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/ice: fix statistics' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/idpf: fix queue setup with TSO offload' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/iavf: fix check for PF Rx timestamp support' " Kevin Traynor
2025-11-21 11:20 ` patch 'net/iavf: fix Rx timestamp validity check' " Kevin Traynor
2025-11-21 23:41   ` Jacob Keller
2025-11-21 11:20 ` patch 'common/cnxk: fix max number of SQB buffers in clean up' " Kevin Traynor
2025-11-21 11:21 ` patch 'common/cnxk: fix null SQ access' " Kevin Traynor
2025-11-21 11:21 ` patch 'common/cnxk: fix format specifier for bandwidth profile ID' " Kevin Traynor
2025-11-21 11:21 ` patch 'common/cnxk: fix NIX Rx inject enabling' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/cnxk: fix Rx inject LF' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/cnxk: fix default meter pre-color' " Kevin Traynor
2025-11-21 11:21 ` patch 'crypto/qat: fix CCM request descriptor hash state size' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/dpaa2: remove ethdev pointer from bus device' " Kevin Traynor
2025-11-21 11:21 ` patch 'app/flow-perf: fix rules array length' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5: fix spurious CPU wakeups' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5: fix IPv6 DSCP offset in HWS sync API' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5: fix send to kernel action resources release' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5: release representor interrupt handler' " Kevin Traynor
2025-11-21 11:21 ` patch 'common/mlx5: release unused mempool entries' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5/hws: fix buddy memory allocation' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5: fix uninitialized variable' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5: fix flow tag indexes support on root table' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5/hws: fix flow rule hash capability' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5: fix null dereference in modify header' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5: skip Rx control flow tables in isolated mode' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5: fix crash on flow rule destruction' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5: move auxiliary data inline' " Kevin Traynor
2025-11-21 11:21 ` patch 'net/mlx5/windows: fix match criteria in flow creation' " Kevin Traynor
2025-11-21 11:21 ` patch 'net: fix IPv6 link local compliance with RFC 4291' " Kevin Traynor

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20251121112128.485623-47-ktraynor@redhat.com \
    --to=ktraynor@redhat.com \
    --cc=bingz@nvidia.com \
    --cc=dsosnowski@nvidia.com \
    --cc=stable@dpdk.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).