[PATCH V2] net/mlx5: add support for flows targeting multicast MAC addresses

[PATCH V2] net/mlx5: add support for flows targeting multicast MAC addresses

[Gavin Li]

Rules for multicast MAC addresses are intended to filter multicast traffic
and are managed through multicast MAC add/remove APIs. In mlx5_dev_spawn
function, devices (PF, VFs, and SFs) retrieve the netdev-configured MAC
addresses via netlink and store them in the PMD device data, which
includes multicast MAC addresses.

To update multicast MAC address rules, create them within
mlx5_traffic_enable.

Fixes: 2d0665a7f771 ("net/mlx5: align PF and VF/SF MAC address handling")
Cc: stable@dpdk.org

Signed-off-by: Gavin Li <gavinl@nvidia.com>
---
 drivers/net/mlx5/linux/mlx5_os.c | 3 +++
 drivers/net/mlx5/mlx5.h          | 1 +
 drivers/net/mlx5/mlx5_trigger.c  | 5 ++++-
 3 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 696a3e12c7..b8d75a97e7 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -3216,6 +3216,7 @@ mlx5_os_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 		mlx5_nl_mac_addr_remove(priv->nl_socket_route,
 					mlx5_ifindex(dev), priv->mac_own,
 					&dev->data->mac_addrs[index], index);
+	BITFIELD_RESET(priv->mac_pmd, index);
 }
 
 /**
@@ -3243,6 +3244,8 @@ mlx5_os_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac,
 		ret = mlx5_nl_mac_addr_add(priv->nl_socket_route,
 					   mlx5_ifindex(dev), priv->mac_own,
 					   mac, index);
+	if (!ret)
+		BITFIELD_SET(priv->mac_pmd, index);
 	return ret;
 }
 
diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h
index c08894cd03..50ed46540c 100644
--- a/drivers/net/mlx5/mlx5.h
+++ b/drivers/net/mlx5/mlx5.h
@@ -1976,6 +1976,7 @@ struct mlx5_priv {
 	struct rte_pci_device *pci_dev; /* Backend PCI device. */
 	struct rte_ether_addr mac[MLX5_MAX_MAC_ADDRESSES]; /* MAC addresses. */
 	BITFIELD_DECLARE(mac_own, uint64_t, MLX5_MAX_MAC_ADDRESSES);
+	BITFIELD_DECLARE(mac_pmd, uint64_t, MLX5_MAX_MAC_ADDRESSES);
 	/* Bit-field of MAC addresses owned by the PMD. */
 	uint16_t vlan_filter[MLX5_MAX_VLAN_IDS]; /* VLAN filters table. */
 	unsigned int vlan_filter_n; /* Number of configured VLAN filters. */
diff --git a/drivers/net/mlx5/mlx5_trigger.c b/drivers/net/mlx5/mlx5_trigger.c
index 6c6f228afd..cbc91be3b1 100644
--- a/drivers/net/mlx5/mlx5_trigger.c
+++ b/drivers/net/mlx5/mlx5_trigger.c
@@ -1813,7 +1813,10 @@ mlx5_traffic_enable(struct rte_eth_dev *dev)
 	for (i = 0; i != MLX5_MAX_MAC_ADDRESSES; ++i) {
 		struct rte_ether_addr *mac = &dev->data->mac_addrs[i];
 
-		if (!memcmp(mac, &cmp, sizeof(*mac)) || rte_is_multicast_ether_addr(mac))
+		/* Add flows for unicast and multicast mac addresses added by API. */
+		if (!memcmp(mac, &cmp, sizeof(*mac)) ||
+		    !BITFIELD_ISSET(priv->mac_pmd, i) ||
+		    (dev->data->all_multicast && rte_is_multicast_ether_addr(mac)))
 			continue;
 		memcpy(&unicast.hdr.dst_addr.addr_bytes,
 		       mac->addr_bytes,
-- 
2.34.1

Re: [PATCH V2] net/mlx5: add support for flows targeting multicast MAC addresses

[Thomas Monjalon]

14/08/2025 12:08, Gavin Li:
> Rules for multicast MAC addresses are intended to filter multicast traffic
> and are managed through multicast MAC add/remove APIs. In mlx5_dev_spawn
> function, devices (PF, VFs, and SFs) retrieve the netdev-configured MAC
> addresses via netlink and store them in the PMD device data, which
> includes multicast MAC addresses.
> 
> To update multicast MAC address rules, create them within
> mlx5_traffic_enable.

Sorry this is not clear.
Please explain what was the previous behaviour with a past tense,
and what is the new changed behaviour.

>  	BITFIELD_DECLARE(mac_own, uint64_t, MLX5_MAX_MAC_ADDRESSES);
> +	BITFIELD_DECLARE(mac_pmd, uint64_t, MLX5_MAX_MAC_ADDRESSES);

Not sure about the naming.
What is the difference between mac_own and mac_pmd?

Re: [PATCH V2] net/mlx5: add support for flows targeting multicast MAC addresses

[Minggang(Gavin) Li]

[-- Attachment #1: Type: text/plain, Size: 1232 bytes --]


On 8/14/2025 10:05 PM, Thomas Monjalon wrote:
> 14/08/2025 12:08, Gavin Li:
>> Rules for multicast MAC addresses are intended to filter multicast traffic
>> and are managed through multicast MAC add/remove APIs. In mlx5_dev_spawn
>> function, devices (PF, VFs, and SFs) retrieve the netdev-configured MAC
>> addresses via netlink and store them in the PMD device data, which
>> includes multicast MAC addresses.
>>
>> To update multicast MAC address rules, create them within
>> mlx5_traffic_enable.
> Sorry this is not clear.
> Please explain what was the previous behaviour with a past tense,
> and what is the new changed behaviour.
ACK
>>   	BITFIELD_DECLARE(mac_own, uint64_t, MLX5_MAX_MAC_ADDRESSES);
>> +	BITFIELD_DECLARE(mac_pmd, uint64_t, MLX5_MAX_MAC_ADDRESSES);
> Not sure about the naming.
> What is the difference between mac_own and mac_pmd?

mac_own serves as a flag for MAC addresses, indicating that they were 
added for VF by pmd. This flag is useful for the flush API, which can 
clear flows associated with VF MAC addresses. Similarly, mac_pmd is used 
to show that a MAC address was added by pmd for PF, VF, and SFs. It 
helps differentiate MAC addresses that have been synchronized from the 
kernel driver.

>

[-- Attachment #2: Type: text/html, Size: 2810 bytes --]