DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Minggang Li(Gavin)" <gavinl@nvidia.com>
To: <matan@nvidia.com>, <viacheslavo@nvidia.com>, <orika@nvidia.com>,
	<thomas@monjalon.net>, Dariusz Sosnowski <dsosnowski@nvidia.com>,
	Bing Zhao <bingz@nvidia.com>, Suanming Mou <suanmingm@nvidia.com>
Cc: <dev@dpdk.org>, <rasland@nvidia.com>, Rongwei Liu <rongweil@nvidia.com>
Subject: [PATCH 2/7] net/mlx5: optimize device probing
Date: Mon, 23 Dec 2024 12:10:56 +0200	[thread overview]
Message-ID: <20241223101101.677449-3-gavinl@nvidia.com> (raw)
In-Reply-To: <20241223101101.677449-1-gavinl@nvidia.com>

From: Rongwei Liu <rongweil@nvidia.com>

Current DPDK probing logic is:
1. Query IB device index and total port number.
2. Query each port information by traversing the port index and
   get the port's ifindex, name and state information etc.
3. Compare the information with devargs until getting matched.
4. For each probing device, repeat steps 2 and 3.

Step 2 will communicate with kernel via netlink and it's time-consuming.
There is no need to repeat netlink communication for each probing device,
PMD can traverse all ports once and save the information into a caching
structure.

Introduce the device information caching in the mlx5 common device
handle and cache the port number, ibindex, port ifindex.

For dynamic interface changing:
1. New VF by toggling switchdev mode should restart dpdk as sriov
   configuration changed.
2. Changing VF number w/o toggling switchdev mode will trigger
   RTM_DELLINK and RTM_NEWLINK events. All the caching information is
   cleared.
3. New SF triggers RTM_NEWLINK event and no port index information in the
   message. All free entries (ifindex = 0) in the cache are invalidated.
4. Delete SF triggers RTM_DELLINK event. Traverse the cache entries and
   invalidate the one with the same ifindex.

Didn't consider race-condition between probing thread and interrupt
thread.

Signed-off-by: Rongwei Liu <rongweil@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
---
 drivers/common/mlx5/linux/mlx5_common_os.h   |   6 ++
 drivers/common/mlx5/linux/mlx5_nl.c          |  94 +++++++++++++----
 drivers/common/mlx5/linux/mlx5_nl.h          |   8 +-
 drivers/common/mlx5/mlx5_common.c            |   5 +
 drivers/common/mlx5/mlx5_common.h            |  13 +++
 drivers/common/mlx5/windows/mlx5_common_os.h |   5 +
 drivers/net/mlx5/linux/mlx5_ethdev_os.c      |  54 ++++++++++
 drivers/net/mlx5/linux/mlx5_os.c             | 104 ++++++++++++++-----
 drivers/net/mlx5/linux/mlx5_os.h             |   6 --
 drivers/net/mlx5/windows/mlx5_os.h           |   5 -
 10 files changed, 242 insertions(+), 58 deletions(-)

diff --git a/drivers/common/mlx5/linux/mlx5_common_os.h b/drivers/common/mlx5/linux/mlx5_common_os.h
index e8aa1d46ec..2e2c54f1fa 100644
--- a/drivers/common/mlx5/linux/mlx5_common_os.h
+++ b/drivers/common/mlx5/linux/mlx5_common_os.h
@@ -22,6 +22,12 @@
 #include "mlx5_glue.h"
 #include "mlx5_malloc.h"
 
+/* verb enumerations translations to local enums. */
+enum {
+	MLX5_FS_NAME_MAX = IBV_SYSFS_NAME_MAX + 1,
+	MLX5_FS_PATH_MAX = IBV_SYSFS_PATH_MAX + 1
+};
+
 /**
  * Get device name. Given an ibv_device pointer - return a
  * pointer to the corresponding device name.
diff --git a/drivers/common/mlx5/linux/mlx5_nl.c b/drivers/common/mlx5/linux/mlx5_nl.c
index a5ac4dc543..e98073aafe 100644
--- a/drivers/common/mlx5/linux/mlx5_nl.c
+++ b/drivers/common/mlx5/linux/mlx5_nl.c
@@ -1073,16 +1073,18 @@ mlx5_nl_port_info(int nl, uint32_t pindex, struct mlx5_nl_port_info *data)
 	uint32_t sn = MLX5_NL_SN_GENERATE;
 	int ret;
 
-	ret = mlx5_nl_send(nl, &req.nh, sn);
-	if (ret < 0)
-		return ret;
-	ret = mlx5_nl_recv(nl, sn, mlx5_nl_cmdget_cb, data);
-	if (ret < 0)
-		return ret;
-	if (!(data->flags & MLX5_NL_CMD_GET_IB_NAME) ||
-	    !(data->flags & MLX5_NL_CMD_GET_IB_INDEX))
-		goto error;
-	data->flags = 0;
+	if (data->ibindex == UINT32_MAX) {
+		ret = mlx5_nl_send(nl, &req.nh, sn);
+		if (ret < 0)
+			return ret;
+		ret = mlx5_nl_recv(nl, sn, mlx5_nl_cmdget_cb, data);
+		if (ret < 0)
+			return ret;
+		if (!(data->flags & MLX5_NL_CMD_GET_IB_NAME) ||
+		    !(data->flags & MLX5_NL_CMD_GET_IB_INDEX))
+			goto error;
+		data->flags = 0;
+	}
 	sn = MLX5_NL_SN_GENERATE;
 	req.nh.nlmsg_type = RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
 					     RDMA_NLDEV_CMD_PORT_GET);
@@ -1109,7 +1111,7 @@ mlx5_nl_port_info(int nl, uint32_t pindex, struct mlx5_nl_port_info *data)
 	    !(data->flags & MLX5_NL_CMD_GET_NET_INDEX) ||
 	    !data->ifindex)
 		goto error;
-	return 1;
+	return 0;
 error:
 	rte_errno = ENODEV;
 	return -rte_errno;
@@ -1128,21 +1130,48 @@ mlx5_nl_port_info(int nl, uint32_t pindex, struct mlx5_nl_port_info *data)
  *   IB device name.
  * @param[in] pindex
  *   IB device port index, starting from 1
+ * @param[in] dev_info
+ *   Cached mlx5 device information.
  * @return
  *   A valid (nonzero) interface index on success, 0 otherwise and rte_errno
  *   is set.
  */
 unsigned int
-mlx5_nl_ifindex(int nl, const char *name, uint32_t pindex)
+mlx5_nl_ifindex(int nl, const char *name, uint32_t pindex, struct mlx5_dev_info *dev_info)
 {
+	int ret;
+
 	struct mlx5_nl_port_info data = {
 			.ifindex = 0,
 			.name = name,
+			.ibindex = UINT32_MAX,
+			.flags = 0,
 	};
 
-	if (mlx5_nl_port_info(nl, pindex, &data) < 0)
-		return 0;
-	return data.ifindex;
+	if (!strcmp(name, dev_info->ibname)) {
+		if (dev_info->port_info && pindex <= dev_info->port_num &&
+		    dev_info->port_info[pindex].valid) {
+			if (!dev_info->port_info[pindex].ifindex)
+				rte_errno = ENODEV;
+			return dev_info->port_info[pindex].ifindex;
+		}
+		if (dev_info->port_num)
+			data.ibindex = dev_info->ibindex;
+	}
+
+	ret = mlx5_nl_port_info(nl, pindex, &data);
+
+	if (!strcmp(dev_info->ibname, name)) {
+		if ((!ret || ret == -ENODEV) && dev_info->port_info &&
+		    pindex <= dev_info->port_num) {
+			if (!ret)
+				dev_info->port_info[pindex].ifindex = data.ifindex;
+			/* -ENODEV means the pindex is unused but still valid case */
+			dev_info->port_info[pindex].valid = 1;
+		}
+	}
+
+	return ret ? 0 : data.ifindex;
 }
 
 /**
@@ -1157,18 +1186,23 @@ mlx5_nl_ifindex(int nl, const char *name, uint32_t pindex)
  *   IB device name.
  * @param[in] pindex
  *   IB device port index, starting from 1
+ * @param[in] dev_info
+ *   Cached mlx5 device information.
  * @return
  *   Port state (ibv_port_state) on success, negative on error
  *   and rte_errno is set.
  */
 int
-mlx5_nl_port_state(int nl, const char *name, uint32_t pindex)
+mlx5_nl_port_state(int nl, const char *name, uint32_t pindex, struct mlx5_dev_info *dev_info)
 {
 	struct mlx5_nl_port_info data = {
 			.state = 0,
 			.name = name,
+			.ibindex = UINT32_MAX,
 	};
 
+	if (dev_info && !strcmp(name, dev_info->ibname) && dev_info->port_num)
+		data.ibindex = dev_info->ibindex;
 	if (mlx5_nl_port_info(nl, pindex, &data) < 0)
 		return -rte_errno;
 	if ((data.flags & MLX5_NL_CMD_GET_PORT_STATE) == 0) {
@@ -1185,13 +1219,15 @@ mlx5_nl_port_state(int nl, const char *name, uint32_t pindex)
  *   Netlink socket of the RDMA kind (NETLINK_RDMA).
  * @param[in] name
  *   IB device name.
+ * @param[in] dev_info
+ *   Cached mlx5 device info.
  *
  * @return
  *   A valid (nonzero) number of ports on success, 0 otherwise
  *   and rte_errno is set.
  */
 unsigned int
-mlx5_nl_portnum(int nl, const char *name)
+mlx5_nl_portnum(int nl, const char *name, struct mlx5_dev_info *dev_info)
 {
 	struct mlx5_nl_port_info data = {
 		.flags = 0,
@@ -1206,7 +1242,10 @@ mlx5_nl_portnum(int nl, const char *name)
 		.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_DUMP,
 	};
 	uint32_t sn = MLX5_NL_SN_GENERATE;
-	int ret;
+	int ret, size;
+
+	if (dev_info->port_num && !strcmp(name, dev_info->ibname))
+		return dev_info->port_num;
 
 	ret = mlx5_nl_send(nl, &req, sn);
 	if (ret < 0)
@@ -1220,8 +1259,25 @@ mlx5_nl_portnum(int nl, const char *name)
 		rte_errno = ENODEV;
 		return 0;
 	}
-	if (!data.portnum)
+	if (!data.portnum) {
 		rte_errno = EINVAL;
+		return 0;
+	}
+	MLX5_ASSERT(!strlen(dev_info->ibname));
+	dev_info->port_num = data.portnum;
+	dev_info->ibindex = data.ibindex;
+	snprintf(dev_info->ibname, MLX5_FS_NAME_MAX, "%s", name);
+	if (data.portnum > 1) {
+		size = (data.portnum + 1) * sizeof(struct mlx5_port_nl_info);
+		dev_info->port_info = mlx5_malloc(MLX5_MEM_ZERO | MLX5_MEM_RTE, size,
+						  RTE_CACHE_LINE_SIZE,
+						  SOCKET_ID_ANY);
+		if (dev_info->port_info == NULL) {
+			memset(dev_info, 0, sizeof(*dev_info));
+			rte_errno = ENOMEM;
+			return 0;
+		}
+	}
 	return data.portnum;
 }
 
diff --git a/drivers/common/mlx5/linux/mlx5_nl.h b/drivers/common/mlx5/linux/mlx5_nl.h
index 580de3b769..396ffc98ce 100644
--- a/drivers/common/mlx5/linux/mlx5_nl.h
+++ b/drivers/common/mlx5/linux/mlx5_nl.h
@@ -11,6 +11,7 @@
 #include <rte_ether.h>
 
 #include "mlx5_common.h"
+#include "mlx5_common_utils.h"
 
 typedef void (mlx5_nl_event_cb)(struct nlmsghdr *hdr, void *user_data);
 
@@ -52,11 +53,12 @@ int mlx5_nl_promisc(int nlsk_fd, unsigned int iface_idx, int enable);
 __rte_internal
 int mlx5_nl_allmulti(int nlsk_fd, unsigned int iface_idx, int enable);
 __rte_internal
-unsigned int mlx5_nl_portnum(int nl, const char *name);
+unsigned int mlx5_nl_portnum(int nl, const char *name, struct mlx5_dev_info *dev_info);
 __rte_internal
-unsigned int mlx5_nl_ifindex(int nl, const char *name, uint32_t pindex);
+unsigned int mlx5_nl_ifindex(int nl, const char *name, uint32_t pindex,
+			     struct mlx5_dev_info *info);
 __rte_internal
-int mlx5_nl_port_state(int nl, const char *name, uint32_t pindex);
+int mlx5_nl_port_state(int nl, const char *name, uint32_t pindex, struct mlx5_dev_info *dev_info);
 __rte_internal
 int mlx5_nl_vf_mac_addr_modify(int nlsk_fd, unsigned int iface_idx,
 			       struct rte_ether_addr *mac, int vf_index);
diff --git a/drivers/common/mlx5/mlx5_common.c b/drivers/common/mlx5/mlx5_common.c
index ca8543e36e..0aaae91c31 100644
--- a/drivers/common/mlx5/mlx5_common.c
+++ b/drivers/common/mlx5/mlx5_common.c
@@ -735,6 +735,11 @@ mlx5_common_dev_release(struct mlx5_common_device *cdev)
 		if (TAILQ_EMPTY(&devices_list))
 			rte_mem_event_callback_unregister("MLX5_MEM_EVENT_CB",
 							  NULL);
+		if (cdev->dev_info.port_info != NULL) {
+			mlx5_free(cdev->dev_info.port_info);
+			cdev->dev_info.port_info = NULL;
+		}
+		cdev->dev_info.port_num = 0;
 		mlx5_dev_mempool_unsubscribe(cdev);
 		mlx5_mr_release_cache(&cdev->mr_scache);
 		mlx5_dev_hw_global_release(cdev);
diff --git a/drivers/common/mlx5/mlx5_common.h b/drivers/common/mlx5/mlx5_common.h
index 1abd1e8239..6cb40f54dd 100644
--- a/drivers/common/mlx5/mlx5_common.h
+++ b/drivers/common/mlx5/mlx5_common.h
@@ -174,6 +174,18 @@ enum mlx5_nl_phys_port_name_type {
 	MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN, /* Unrecognized. */
 };
 
+struct mlx5_port_nl_info {
+	uint32_t ifindex;
+	uint8_t valid;
+};
+
+struct mlx5_dev_info {
+	uint32_t port_num;
+	uint32_t ibindex;
+	char ibname[MLX5_FS_NAME_MAX];
+	struct mlx5_port_nl_info *port_info;
+};
+
 /** Switch information returned by mlx5_nl_switch_info(). */
 struct mlx5_switch_info {
 	uint32_t master:1; /**< Master device. */
@@ -525,6 +537,7 @@ struct mlx5_common_device {
 	uint32_t classes_loaded;
 	void *ctx; /* Verbs/DV/DevX context. */
 	void *pd; /* Protection Domain. */
+	struct mlx5_dev_info dev_info; /* Device port info queried via netlink. */
 	uint32_t pdn; /* Protection Domain Number. */
 	struct mlx5_mr_share_cache mr_scache; /* Global shared MR cache. */
 	struct mlx5_common_dev_config config; /* Device configuration. */
diff --git a/drivers/common/mlx5/windows/mlx5_common_os.h b/drivers/common/mlx5/windows/mlx5_common_os.h
index acee0c987f..65394035de 100644
--- a/drivers/common/mlx5/windows/mlx5_common_os.h
+++ b/drivers/common/mlx5/windows/mlx5_common_os.h
@@ -20,6 +20,11 @@
 
 #define MLX5_BF_OFFSET 0x800
 
+enum {
+	MLX5_FS_NAME_MAX = MLX5_DEVX_DEVICE_NAME_SIZE + 1,
+	MLX5_FS_PATH_MAX = MLX5_DEVX_DEVICE_PNP_SIZE + 1
+};
+
 /**
  * This API allocates aligned or non-aligned memory.  The free can be on either
  * aligned or nonaligned memory.  To be protected - even though there may be no
diff --git a/drivers/net/mlx5/linux/mlx5_ethdev_os.c b/drivers/net/mlx5/linux/mlx5_ethdev_os.c
index 5d64984022..08ac6dd939 100644
--- a/drivers/net/mlx5/linux/mlx5_ethdev_os.c
+++ b/drivers/net/mlx5/linux/mlx5_ethdev_os.c
@@ -23,6 +23,7 @@
 #include <stdalign.h>
 #include <sys/un.h>
 #include <time.h>
+#include <linux/rtnetlink.h>
 
 #include <ethdev_linux_ethtool.h>
 #include <ethdev_driver.h>
@@ -673,6 +674,57 @@ mlx5_link_update_bond(struct rte_eth_dev *dev)
 		((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING));
 }
 
+static void
+mlx5_handle_port_info_update(struct mlx5_dev_info *dev_info, uint32_t if_index,
+			     uint16_t msg_type)
+{
+	struct mlx5_switch_info info = {
+		.master = 0,
+		.representor = 0,
+		.name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET,
+		.port_name = 0,
+		.switch_id = 0,
+	};
+	uint32_t i;
+	int nl_route;
+
+	if (dev_info->port_num <= 1 || dev_info->port_info == NULL)
+		return;
+
+	for (i = 1; i <= dev_info->port_num; i++) {
+		if (!dev_info->port_info[i].valid)
+			continue;
+		if (dev_info->port_info[i].ifindex == if_index)
+			break;
+	}
+	if (msg_type == RTM_NEWLINK && i > dev_info->port_num) {
+		nl_route = mlx5_nl_init(NETLINK_ROUTE, 0);
+		if  (nl_route < 0)
+			goto flush_all;
+
+		if (mlx5_nl_switch_info(nl_route, if_index, &info)) {
+			if (mlx5_sysfs_switch_info(if_index, &info))
+				goto flush_all;
+		}
+
+		if (info.name_type == MLX5_PHYS_PORT_NAME_TYPE_PFSF ||
+		    info.name_type == MLX5_PHYS_PORT_NAME_TYPE_PFVF)
+			goto flush_all;
+		close(nl_route);
+	} else if (msg_type == RTM_DELLINK && i <= dev_info->port_num) {
+		memset(dev_info->port_info + i, 0, sizeof(struct mlx5_port_nl_info));
+	}
+
+	return;
+flush_all:
+	if (nl_route >= 0)
+		close(nl_route);
+	for (i = 1; i <= dev_info->port_num; i++) {
+		if (!dev_info->port_info[i].ifindex)
+			dev_info->port_info[i].valid = 0;
+	}
+}
+
 static void
 mlx5_dev_interrupt_nl_cb(struct nlmsghdr *hdr, void *cb_arg)
 {
@@ -682,6 +734,8 @@ mlx5_dev_interrupt_nl_cb(struct nlmsghdr *hdr, void *cb_arg)
 
 	if (mlx5_nl_parse_link_status_update(hdr, &if_index) < 0)
 		return;
+	mlx5_handle_port_info_update(&sh->cdev->dev_info, if_index, hdr->nlmsg_type);
+
 	for (i = 0; i < sh->max_port; i++) {
 		struct mlx5_dev_shared_port *port = &sh->port[i];
 		struct rte_eth_dev *dev;
diff --git a/drivers/net/mlx5/linux/mlx5_os.c b/drivers/net/mlx5/linux/mlx5_os.c
index 69a80b9ddc..8f6e584154 100644
--- a/drivers/net/mlx5/linux/mlx5_os.c
+++ b/drivers/net/mlx5/linux/mlx5_os.c
@@ -1268,7 +1268,7 @@ mlx5_dev_spawn(struct rte_device *dpdk_dev,
 		/* IB doesn't allow more than 255 ports, must be Ethernet. */
 		err = mlx5_nl_port_state(nl_rdma,
 			spawn->phys_dev_name,
-			spawn->phys_port);
+			spawn->phys_port, &spawn->cdev->dev_info);
 		if (err < 0) {
 			DRV_LOG(INFO, "Failed to get netlink port state: %s",
 				strerror(rte_errno));
@@ -1897,6 +1897,8 @@ mlx5_dev_spawn_data_cmp(const void *a, const void *b)
  *   Netlink RDMA group socket handle.
  * @param[in] owner
  *   Representor owner PF index.
+ * @param[in] dev_info
+ *   Cached mlx5 device information.
  * @param[out] bond_info
  *   Pointer to bonding information.
  *
@@ -1908,6 +1910,7 @@ static int
 mlx5_device_bond_pci_match(const char *ibdev_name,
 			   const struct rte_pci_addr *pci_dev,
 			   int nl_rdma, uint16_t owner,
+			   struct mlx5_dev_info *dev_info,
 			   struct mlx5_bond_info *bond_info)
 {
 	char ifname[IF_NAMESIZE + 1];
@@ -1928,7 +1931,7 @@ mlx5_device_bond_pci_match(const char *ibdev_name,
 		return -1;
 	if (!strstr(ibdev_name, "bond"))
 		return -1;
-	np = mlx5_nl_portnum(nl_rdma, ibdev_name);
+	np = mlx5_nl_portnum(nl_rdma, ibdev_name, dev_info);
 	if (!np)
 		return -1;
 	if (mlx5_get_device_guid(pci_dev, cur_guid, sizeof(cur_guid)) < 0)
@@ -1940,7 +1943,7 @@ mlx5_device_bond_pci_match(const char *ibdev_name,
 	 */
 	for (i = 1; i <= np; ++i) {
 		/* Check whether Infiniband port is populated. */
-		ifindex = mlx5_nl_ifindex(nl_rdma, ibdev_name, i);
+		ifindex = mlx5_nl_ifindex(nl_rdma, ibdev_name, i, dev_info);
 		if (!ifindex)
 			continue;
 		if (!if_indextoname(ifindex, ifname))
@@ -1978,9 +1981,13 @@ mlx5_device_bond_pci_match(const char *ibdev_name,
 		if (!file)
 			break;
 		info.name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET;
-		if (fscanf(file, "%32s", tmp_str) == 1)
+		if (fscanf(file, "%32s", tmp_str) == 1) {
 			mlx5_translate_port_name(tmp_str, &info);
-		fclose(file);
+			fclose(file);
+		} else {
+			fclose(file);
+			break;
+		}
 		/* Only process PF ports. */
 		if (info.name_type != MLX5_PHYS_PORT_NAME_TYPE_LEGACY &&
 		    info.name_type != MLX5_PHYS_PORT_NAME_TYPE_UPLINK)
@@ -2003,8 +2010,8 @@ mlx5_device_bond_pci_match(const char *ibdev_name,
 		if (ret != 1)
 			break;
 		/* Save bonding info. */
-		strncpy(bond_info->ports[info.port_name].ifname, ifname,
-			sizeof(bond_info->ports[0].ifname));
+		snprintf(bond_info->ports[info.port_name].ifname,
+			 sizeof(bond_info->ports[0].ifname), "%s", ifname);
 		bond_info->ports[info.port_name].pci_addr = pci_addr;
 		bond_info->ports[info.port_name].ifindex = ifindex;
 		bond_info->n_port++;
@@ -2033,6 +2040,7 @@ mlx5_device_bond_pci_match(const char *ibdev_name,
 		      pci_addr.function == owner)))
 			pf = info.port_name;
 	}
+	fclose(bond_file);
 	if (pf >= 0) {
 		/* Get bond interface info */
 		ret = mlx5_sysfs_bond_info(ifindex, &bond_info->ifindex,
@@ -2084,7 +2092,8 @@ mlx5_nl_esw_multiport_get(struct rte_pci_addr *pci_addr, int *enabled)
 #define SYSFS_MPESW_PARAM_MAX_LEN 16
 
 static int
-mlx5_sysfs_esw_multiport_get(struct ibv_device *ibv, struct rte_pci_addr *pci_addr, int *enabled)
+mlx5_sysfs_esw_multiport_get(struct ibv_device *ibv, struct rte_pci_addr *pci_addr, int *enabled,
+			     struct mlx5_dev_info *dev_info)
 {
 	int nl_rdma;
 	unsigned int n_ports;
@@ -2096,7 +2105,7 @@ mlx5_sysfs_esw_multiport_get(struct ibv_device *ibv, struct rte_pci_addr *pci_ad
 	nl_rdma = mlx5_nl_init(NETLINK_RDMA, 0);
 	if (nl_rdma < 0)
 		return nl_rdma;
-	n_ports = mlx5_nl_portnum(nl_rdma, ibv->name);
+	n_ports = mlx5_nl_portnum(nl_rdma, ibv->name, dev_info);
 	if (!n_ports) {
 		ret = -rte_errno;
 		goto close_nl_rdma;
@@ -2104,12 +2113,12 @@ mlx5_sysfs_esw_multiport_get(struct ibv_device *ibv, struct rte_pci_addr *pci_ad
 	for (i = 1; i <= n_ports; ++i) {
 		unsigned int ifindex;
 		char ifname[IF_NAMESIZE + 1];
-		struct rte_pci_addr if_pci_addr;
+		struct rte_pci_addr if_pci_addr = { 0 };
 		char mpesw[SYSFS_MPESW_PARAM_MAX_LEN + 1];
 		FILE *sysfs;
 		int n;
 
-		ifindex = mlx5_nl_ifindex(nl_rdma, ibv->name, i);
+		ifindex = mlx5_nl_ifindex(nl_rdma, ibv->name, i, dev_info);
 		if (!ifindex)
 			continue;
 		if (!if_indextoname(ifindex, ifname))
@@ -2151,7 +2160,8 @@ mlx5_sysfs_esw_multiport_get(struct ibv_device *ibv, struct rte_pci_addr *pci_ad
 }
 
 static int
-mlx5_is_mpesw_enabled(struct ibv_device *ibv, struct rte_pci_addr *ibv_pci_addr, int *enabled)
+mlx5_is_mpesw_enabled(struct ibv_device *ibv, struct rte_pci_addr *ibv_pci_addr, int *enabled,
+		      struct mlx5_dev_info *dev_info)
 {
 	/*
 	 * Try getting Multiport E-Switch state through netlink interface
@@ -2159,7 +2169,7 @@ mlx5_is_mpesw_enabled(struct ibv_device *ibv, struct rte_pci_addr *ibv_pci_addr,
 	 * assume that Multiport E-Switch is disabled and return an error.
 	 */
 	if (mlx5_nl_esw_multiport_get(ibv_pci_addr, enabled) >= 0 ||
-	    mlx5_sysfs_esw_multiport_get(ibv, ibv_pci_addr, enabled) >= 0)
+	    mlx5_sysfs_esw_multiport_get(ibv, ibv_pci_addr, enabled, dev_info) >= 0)
 		return 0;
 	DRV_LOG(DEBUG, "Unable to check MPESW state for IB device %s "
 		       "(PCI: " PCI_PRI_FMT ")",
@@ -2173,7 +2183,7 @@ mlx5_is_mpesw_enabled(struct ibv_device *ibv, struct rte_pci_addr *ibv_pci_addr,
 static int
 mlx5_device_mpesw_pci_match(struct ibv_device *ibv,
 			    const struct rte_pci_addr *owner_pci,
-			    int nl_rdma)
+			    int nl_rdma, struct mlx5_dev_info *dev_info)
 {
 	struct rte_pci_addr ibdev_pci_addr = { 0 };
 	char ifname[IF_NAMESIZE + 1] = { 0 };
@@ -2197,24 +2207,24 @@ mlx5_device_mpesw_pci_match(struct ibv_device *ibv,
 		return -1;
 	}
 	/* Check if IB device has MPESW enabled. */
-	if (mlx5_is_mpesw_enabled(ibv, &ibdev_pci_addr, &enabled))
+	if (mlx5_is_mpesw_enabled(ibv, &ibdev_pci_addr, &enabled, dev_info))
 		return -1;
 	if (!enabled)
 		return -1;
 	/* Iterate through IB ports to find MPESW master uplink port. */
 	if (nl_rdma < 0)
 		return -1;
-	np = mlx5_nl_portnum(nl_rdma, ibv->name);
+	np = mlx5_nl_portnum(nl_rdma, ibv->name, dev_info);
 	if (!np)
 		return -1;
 	for (i = 1; i <= np; ++i) {
-		struct rte_pci_addr pci_addr;
+		struct rte_pci_addr pci_addr = { 0 };
 		FILE *file;
 		char port_name[IF_NAMESIZE + 1];
 		struct mlx5_switch_info	info;
 
 		/* Check whether IB port has a corresponding netdev. */
-		ifindex = mlx5_nl_ifindex(nl_rdma, ibv->name, i);
+		ifindex = mlx5_nl_ifindex(nl_rdma, ibv->name, i, dev_info);
 		if (!ifindex)
 			continue;
 		if (!if_indextoname(ifindex, ifname))
@@ -2321,16 +2331,30 @@ mlx5_os_pci_probe_pf(struct mlx5_common_device *cdev,
 	 * matching ones, gathering into the list.
 	 */
 	struct ibv_device *ibv_match[ret + 1];
+	struct mlx5_dev_info *info, tmp_info[ret];
 	int nl_route = mlx5_nl_init(NETLINK_ROUTE, 0);
 	int nl_rdma = mlx5_nl_init(NETLINK_RDMA, 0);
 	unsigned int i;
 
+	memset(tmp_info, 0, sizeof(tmp_info));
 	while (ret-- > 0) {
 		struct rte_pci_addr pci_addr;
 
+		if (cdev->dev_info.port_num) {
+			if (strcmp(ibv_list[ret]->name, cdev->dev_info.ibname)) {
+				DRV_LOG(INFO, "Unmatched caching device \"%s\" \"%s\"",
+					cdev->dev_info.ibname, ibv_list[ret]->name);
+				continue;
+			}
+			info = &cdev->dev_info;
+		} else {
+			info = &tmp_info[ret];
+		}
 		DRV_LOG(DEBUG, "Checking device \"%s\"", ibv_list[ret]->name);
 		bd = mlx5_device_bond_pci_match(ibv_list[ret]->name, &owner_pci,
-						nl_rdma, owner_id, &bond_info);
+						nl_rdma, owner_id,
+						info,
+						&bond_info);
 		if (bd >= 0) {
 			/*
 			 * Bonding device detected. Only one match is allowed,
@@ -2356,7 +2380,8 @@ mlx5_os_pci_probe_pf(struct mlx5_common_device *cdev,
 			ibv_match[nd++] = ibv_list[ret];
 			break;
 		}
-		mpesw = mlx5_device_mpesw_pci_match(ibv_list[ret], &owner_pci, nl_rdma);
+		mpesw = mlx5_device_mpesw_pci_match(ibv_list[ret], &owner_pci, nl_rdma,
+						    info);
 		if (mpesw >= 0) {
 			/*
 			 * MPESW device detected. Only one matching IB device is allowed,
@@ -2380,10 +2405,18 @@ mlx5_os_pci_probe_pf(struct mlx5_common_device *cdev,
 		}
 		/* Bonding or MPESW device was not found. */
 		if (mlx5_get_pci_addr(ibv_list[ret]->ibdev_path,
-					&pci_addr))
+					&pci_addr)) {
+			if (tmp_info[ret].port_info != NULL)
+				mlx5_free(tmp_info[ret].port_info);
+			memset(&tmp_info[ret], 0, sizeof(tmp_info[0]));
 			continue;
-		if (rte_pci_addr_cmp(&owner_pci, &pci_addr) != 0)
+		}
+		if (rte_pci_addr_cmp(&owner_pci, &pci_addr) != 0) {
+			if (tmp_info[ret].port_info != NULL)
+				mlx5_free(tmp_info[ret].port_info);
+			memset(&tmp_info[ret], 0, sizeof(tmp_info[0]));
 			continue;
+		}
 		DRV_LOG(INFO, "PCI information matches for device \"%s\"",
 			ibv_list[ret]->name);
 		ibv_match[nd++] = ibv_list[ret];
@@ -2401,13 +2434,21 @@ mlx5_os_pci_probe_pf(struct mlx5_common_device *cdev,
 		goto exit;
 	}
 	if (nd == 1) {
+		if (!cdev->dev_info.port_num) {
+			for (i = 0; i < RTE_DIM(tmp_info); i++) {
+				if (tmp_info[i].port_num) {
+					cdev->dev_info = tmp_info[i];
+					break;
+				}
+			}
+		}
 		/*
 		 * Found single matching device may have multiple ports.
 		 * Each port may be representor, we have to check the port
 		 * number and check the representors existence.
 		 */
 		if (nl_rdma >= 0)
-			np = mlx5_nl_portnum(nl_rdma, ibv_match[0]->name);
+			np = mlx5_nl_portnum(nl_rdma, ibv_match[0]->name, &cdev->dev_info);
 		if (!np)
 			DRV_LOG(WARNING,
 				"Cannot get IB device \"%s\" ports number.",
@@ -2424,6 +2465,14 @@ mlx5_os_pci_probe_pf(struct mlx5_common_device *cdev,
 			ret = -rte_errno;
 			goto exit;
 		}
+	} else {
+		/* Can't handle one common device with multiple IB devices caching */
+		for (i = 0; i < RTE_DIM(tmp_info); i++) {
+			if (tmp_info[i].port_info != NULL)
+				mlx5_free(tmp_info[i].port_info);
+			memset(&tmp_info[i], 0, sizeof(tmp_info[0]));
+		}
+		DRV_LOG(INFO, "Cannot handle multiple IB devices info caching in single common device.");
 	}
 	/* Now we can determine the maximal amount of devices to be spawned. */
 	list = mlx5_malloc(MLX5_MEM_ZERO,
@@ -2457,7 +2506,7 @@ mlx5_os_pci_probe_pf(struct mlx5_common_device *cdev,
 			list[ns].mpesw_port = MLX5_MPESW_PORT_INVALID;
 			list[ns].ifindex = mlx5_nl_ifindex(nl_rdma,
 							   ibv_match[0]->name,
-							   i);
+							   i, &cdev->dev_info);
 			if (!list[ns].ifindex) {
 				/*
 				 * No network interface index found for the
@@ -2588,7 +2637,7 @@ mlx5_os_pci_probe_pf(struct mlx5_common_device *cdev,
 				list[ns].ifindex = mlx5_nl_ifindex
 							    (nl_rdma,
 							     ibv_match[i]->name,
-							     1);
+							     1, &cdev->dev_info);
 			if (!list[ns].ifindex) {
 				char ifname[IF_NAMESIZE];
 
@@ -2777,6 +2826,11 @@ mlx5_os_pci_probe_pf(struct mlx5_common_device *cdev,
 		mlx5_free(list);
 	MLX5_ASSERT(ibv_list);
 	mlx5_glue->free_device_list(ibv_list);
+	if (ret) {
+		if (cdev->dev_info.port_info != NULL)
+			mlx5_free(cdev->dev_info.port_info);
+		memset(&cdev->dev_info, 0, sizeof(cdev->dev_info));
+	}
 	return ret;
 }
 
diff --git a/drivers/net/mlx5/linux/mlx5_os.h b/drivers/net/mlx5/linux/mlx5_os.h
index 80c70d713a..4ef0916173 100644
--- a/drivers/net/mlx5/linux/mlx5_os.h
+++ b/drivers/net/mlx5/linux/mlx5_os.h
@@ -8,12 +8,6 @@
 
 #include <net/if.h>
 
-/* verb enumerations translations to local enums. */
-enum {
-	MLX5_FS_NAME_MAX = IBV_SYSFS_NAME_MAX + 1,
-	MLX5_FS_PATH_MAX = IBV_SYSFS_PATH_MAX + 1
-};
-
 /* Maximal data of sendmsg message(in bytes). */
 #define MLX5_SENDMSG_MAX 64
 
diff --git a/drivers/net/mlx5/windows/mlx5_os.h b/drivers/net/mlx5/windows/mlx5_os.h
index 8b58265687..fb7198c244 100644
--- a/drivers/net/mlx5/windows/mlx5_os.h
+++ b/drivers/net/mlx5/windows/mlx5_os.h
@@ -7,11 +7,6 @@
 
 #include "mlx5_win_ext.h"
 
-enum {
-	MLX5_FS_NAME_MAX = MLX5_DEVX_DEVICE_NAME_SIZE + 1,
-	MLX5_FS_PATH_MAX = MLX5_DEVX_DEVICE_PNP_SIZE + 1
-};
-
 #define PCI_DRV_FLAGS 0
 
 #define MLX5_NAMESIZE MLX5_FS_NAME_MAX
-- 
2.34.1


  parent reply	other threads:[~2024-12-23 10:11 UTC|newest]

Thread overview: 8+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-12-23 10:10 [PATCH V1 0/7] port probe time optimization Minggang Li(Gavin)
2024-12-23 10:10 ` [PATCH 1/7] mailmap: update user name Minggang Li(Gavin)
2024-12-23 10:10 ` Minggang Li(Gavin) [this message]
2024-12-23 10:10 ` [PATCH 3/7] net/mlx5: add new devargs to control probe optimization Minggang Li(Gavin)
2024-12-23 10:10 ` [PATCH 4/7] common/mlx5: fix Netlink socket leak Minggang Li(Gavin)
2024-12-23 10:10 ` [PATCH 5/7] common/mlx5: add RDMA monitor event awareness Minggang Li(Gavin)
2024-12-23 10:11 ` [PATCH 6/7] mlx5: use RDMA Netlink to update port information Minggang Li(Gavin)
2024-12-23 10:11 ` [PATCH 7/7] mlx5: add backward compatibility for RDMA monitor Minggang Li(Gavin)

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=20241223101101.677449-3-gavinl@nvidia.com \
    --to=gavinl@nvidia.com \
    --cc=bingz@nvidia.com \
    --cc=dev@dpdk.org \
    --cc=dsosnowski@nvidia.com \
    --cc=matan@nvidia.com \
    --cc=orika@nvidia.com \
    --cc=rasland@nvidia.com \
    --cc=rongweil@nvidia.com \
    --cc=suanmingm@nvidia.com \
    --cc=thomas@monjalon.net \
    --cc=viacheslavo@nvidia.com \
    /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).