1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2018 6WIND S.A.
3 * Copyright 2018 Mellanox Technologies, Ltd
7 #include <linux/if_link.h>
8 #include <linux/rtnetlink.h>
9 #include <linux/genetlink.h>
11 #include <rdma/rdma_netlink.h>
17 #include <sys/socket.h>
20 #include <rte_errno.h>
23 #include "mlx5_common_utils.h"
24 #include "mlx5_malloc.h"
26 #include <linux/devlink.h>
30 /* Size of the buffer to receive kernel messages */
31 #define MLX5_NL_BUF_SIZE (32 * 1024)
32 /* Send buffer size for the Netlink socket */
33 #define MLX5_SEND_BUF_SIZE 32768
34 /* Receive buffer size for the Netlink socket */
35 #define MLX5_RECV_BUF_SIZE 32768
37 /** Parameters of VLAN devices created by driver. */
38 #define MLX5_VMWA_VLAN_DEVICE_PFX "evmlx"
40 * Define NDA_RTA as defined in iproute2 sources.
42 * see in iproute2 sources file include/libnetlink.h
45 #define MLX5_NDA_RTA(r) \
46 ((struct rtattr *)(((char *)(r)) + NLMSG_ALIGN(sizeof(struct ndmsg))))
49 * Define NLMSG_TAIL as defined in iproute2 sources.
51 * see in iproute2 sources file include/libnetlink.h
54 #define NLMSG_TAIL(nmsg) \
55 ((struct rtattr *)(((char *)(nmsg)) + NLMSG_ALIGN((nmsg)->nlmsg_len)))
58 * The following definitions are normally found in rdma/rdma_netlink.h,
59 * however they are so recent that most systems do not expose them yet.
61 #ifndef HAVE_RDMA_NL_NLDEV
62 #define RDMA_NL_NLDEV 5
64 #ifndef HAVE_RDMA_NLDEV_CMD_GET
65 #define RDMA_NLDEV_CMD_GET 1
67 #ifndef HAVE_RDMA_NLDEV_CMD_PORT_GET
68 #define RDMA_NLDEV_CMD_PORT_GET 5
70 #ifndef HAVE_RDMA_NLDEV_ATTR_DEV_INDEX
71 #define RDMA_NLDEV_ATTR_DEV_INDEX 1
73 #ifndef HAVE_RDMA_NLDEV_ATTR_DEV_NAME
74 #define RDMA_NLDEV_ATTR_DEV_NAME 2
76 #ifndef HAVE_RDMA_NLDEV_ATTR_PORT_INDEX
77 #define RDMA_NLDEV_ATTR_PORT_INDEX 3
79 #ifndef HAVE_RDMA_NLDEV_ATTR_NDEV_INDEX
80 #define RDMA_NLDEV_ATTR_NDEV_INDEX 50
83 /* These are normally found in linux/if_link.h. */
84 #ifndef HAVE_IFLA_NUM_VF
85 #define IFLA_NUM_VF 21
87 #ifndef HAVE_IFLA_EXT_MASK
88 #define IFLA_EXT_MASK 29
90 #ifndef HAVE_IFLA_PHYS_SWITCH_ID
91 #define IFLA_PHYS_SWITCH_ID 36
93 #ifndef HAVE_IFLA_PHYS_PORT_NAME
94 #define IFLA_PHYS_PORT_NAME 38
98 * Some Devlink defines may be missed in old kernel versions,
99 * adjust used defines.
101 #ifndef DEVLINK_GENL_NAME
102 #define DEVLINK_GENL_NAME "devlink"
104 #ifndef DEVLINK_GENL_VERSION
105 #define DEVLINK_GENL_VERSION 1
107 #ifndef DEVLINK_ATTR_BUS_NAME
108 #define DEVLINK_ATTR_BUS_NAME 1
110 #ifndef DEVLINK_ATTR_DEV_NAME
111 #define DEVLINK_ATTR_DEV_NAME 2
113 #ifndef DEVLINK_ATTR_PARAM
114 #define DEVLINK_ATTR_PARAM 80
116 #ifndef DEVLINK_ATTR_PARAM_NAME
117 #define DEVLINK_ATTR_PARAM_NAME 81
119 #ifndef DEVLINK_ATTR_PARAM_TYPE
120 #define DEVLINK_ATTR_PARAM_TYPE 83
122 #ifndef DEVLINK_ATTR_PARAM_VALUES_LIST
123 #define DEVLINK_ATTR_PARAM_VALUES_LIST 84
125 #ifndef DEVLINK_ATTR_PARAM_VALUE
126 #define DEVLINK_ATTR_PARAM_VALUE 85
128 #ifndef DEVLINK_ATTR_PARAM_VALUE_DATA
129 #define DEVLINK_ATTR_PARAM_VALUE_DATA 86
131 #ifndef DEVLINK_ATTR_PARAM_VALUE_CMODE
132 #define DEVLINK_ATTR_PARAM_VALUE_CMODE 87
134 #ifndef DEVLINK_PARAM_CMODE_DRIVERINIT
135 #define DEVLINK_PARAM_CMODE_DRIVERINIT 1
137 #ifndef DEVLINK_CMD_RELOAD
138 #define DEVLINK_CMD_RELOAD 37
140 #ifndef DEVLINK_CMD_PARAM_GET
141 #define DEVLINK_CMD_PARAM_GET 38
143 #ifndef DEVLINK_CMD_PARAM_SET
144 #define DEVLINK_CMD_PARAM_SET 39
150 /* Add/remove MAC address through Netlink */
151 struct mlx5_nl_mac_addr {
152 struct rte_ether_addr (*mac)[];
153 /**< MAC address handled by the device. */
154 int mac_n; /**< Number of addresses in the array. */
157 #define MLX5_NL_CMD_GET_IB_NAME (1 << 0)
158 #define MLX5_NL_CMD_GET_IB_INDEX (1 << 1)
159 #define MLX5_NL_CMD_GET_NET_INDEX (1 << 2)
160 #define MLX5_NL_CMD_GET_PORT_INDEX (1 << 3)
162 /** Data structure used by mlx5_nl_cmdget_cb(). */
163 struct mlx5_nl_ifindex_data {
164 const char *name; /**< IB device name (in). */
165 uint32_t flags; /**< found attribute flags (out). */
166 uint32_t ibindex; /**< IB device index (out). */
167 uint32_t ifindex; /**< Network interface index (out). */
168 uint32_t portnum; /**< IB device max port number (out). */
173 /* Generate Netlink sequence number. */
174 #define MLX5_NL_SN_GENERATE __atomic_add_fetch(&atomic_sn, 1, __ATOMIC_RELAXED)
177 * Opens a Netlink socket.
180 * Netlink protocol (e.g. NETLINK_ROUTE, NETLINK_RDMA).
183 * A file descriptor on success, a negative errno value otherwise and
187 mlx5_nl_init(int protocol)
190 int sndbuf_size = MLX5_SEND_BUF_SIZE;
191 int rcvbuf_size = MLX5_RECV_BUF_SIZE;
192 struct sockaddr_nl local = {
193 .nl_family = AF_NETLINK,
197 fd = socket(AF_NETLINK, SOCK_RAW | SOCK_CLOEXEC, protocol);
202 ret = setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &sndbuf_size, sizeof(int));
207 ret = setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &rcvbuf_size, sizeof(int));
212 ret = bind(fd, (struct sockaddr *)&local, sizeof(local));
224 * Send a request message to the kernel on the Netlink socket.
227 * Netlink socket file descriptor.
229 * The Netlink message send to the kernel.
233 * Pointer to the request structure.
235 * Length of the request in bytes.
238 * The number of sent bytes on success, a negative errno value otherwise and
242 mlx5_nl_request(int nlsk_fd, struct nlmsghdr *nh, uint32_t sn, void *req,
245 struct sockaddr_nl sa = {
246 .nl_family = AF_NETLINK,
248 struct iovec iov[2] = {
249 { .iov_base = nh, .iov_len = sizeof(*nh), },
250 { .iov_base = req, .iov_len = len, },
252 struct msghdr msg = {
254 .msg_namelen = sizeof(sa),
260 nh->nlmsg_pid = 0; /* communication with the kernel uses pid 0 */
262 send_bytes = sendmsg(nlsk_fd, &msg, 0);
263 if (send_bytes < 0) {
271 * Send a message to the kernel on the Netlink socket.
274 * The Netlink socket file descriptor used for communication.
276 * The Netlink message send to the kernel.
281 * The number of sent bytes on success, a negative errno value otherwise and
285 mlx5_nl_send(int nlsk_fd, struct nlmsghdr *nh, uint32_t sn)
287 struct sockaddr_nl sa = {
288 .nl_family = AF_NETLINK,
292 .iov_len = nh->nlmsg_len,
294 struct msghdr msg = {
296 .msg_namelen = sizeof(sa),
302 nh->nlmsg_pid = 0; /* communication with the kernel uses pid 0 */
304 send_bytes = sendmsg(nlsk_fd, &msg, 0);
305 if (send_bytes < 0) {
313 * Receive a message from the kernel on the Netlink socket, following
317 * The Netlink socket file descriptor used for communication.
321 * The callback function to call for each Netlink message received.
322 * @param[in, out] arg
323 * Custom arguments for the callback.
326 * 0 on success, a negative errno value otherwise and rte_errno is set.
329 mlx5_nl_recv(int nlsk_fd, uint32_t sn, int (*cb)(struct nlmsghdr *, void *arg),
332 struct sockaddr_nl sa;
333 void *buf = mlx5_malloc(0, MLX5_RECV_BUF_SIZE, 0, SOCKET_ID_ANY);
336 .iov_len = MLX5_RECV_BUF_SIZE,
338 struct msghdr msg = {
340 .msg_namelen = sizeof(sa),
342 /* One message at a time */
357 recv_bytes = recvmsg(nlsk_fd, &msg, 0);
358 if (recv_bytes == -1) {
363 nh = (struct nlmsghdr *)buf;
364 } while (nh->nlmsg_seq != sn);
366 NLMSG_OK(nh, (unsigned int)recv_bytes);
367 nh = NLMSG_NEXT(nh, recv_bytes)) {
368 if (nh->nlmsg_type == NLMSG_ERROR) {
369 struct nlmsgerr *err_data = NLMSG_DATA(nh);
371 if (err_data->error < 0) {
372 rte_errno = -err_data->error;
380 /* Multi-part msgs and their trailing DONE message. */
381 if (nh->nlmsg_flags & NLM_F_MULTI) {
382 if (nh->nlmsg_type == NLMSG_DONE) {
401 * Parse Netlink message to retrieve the bridge MAC address.
404 * Pointer to Netlink Message Header.
406 * PMD data register with this callback.
409 * 0 on success, a negative errno value otherwise and rte_errno is set.
412 mlx5_nl_mac_addr_cb(struct nlmsghdr *nh, void *arg)
414 struct mlx5_nl_mac_addr *data = arg;
415 struct ndmsg *r = NLMSG_DATA(nh);
416 struct rtattr *attribute;
419 len = nh->nlmsg_len - NLMSG_LENGTH(sizeof(*r));
420 for (attribute = MLX5_NDA_RTA(r);
421 RTA_OK(attribute, len);
422 attribute = RTA_NEXT(attribute, len)) {
423 if (attribute->rta_type == NDA_LLADDR) {
424 if (data->mac_n == MLX5_MAX_MAC_ADDRESSES) {
426 "not enough room to finalize the"
431 #ifdef RTE_LIBRTE_MLX5_DEBUG
432 char m[RTE_ETHER_ADDR_FMT_SIZE];
434 rte_ether_format_addr(m, RTE_ETHER_ADDR_FMT_SIZE,
435 RTA_DATA(attribute));
436 DRV_LOG(DEBUG, "bridge MAC address %s", m);
438 memcpy(&(*data->mac)[data->mac_n++],
439 RTA_DATA(attribute), RTE_ETHER_ADDR_LEN);
446 * Get bridge MAC addresses.
449 * Netlink socket file descriptor.
450 * @param[in] iface_idx
451 * Net device interface index.
453 * Pointer to the array table of MAC addresses to fill.
454 * Its size should be of MLX5_MAX_MAC_ADDRESSES.
456 * Number of entries filled in MAC array.
459 * 0 on success, a negative errno value otherwise and rte_errno is set.
462 mlx5_nl_mac_addr_list(int nlsk_fd, unsigned int iface_idx,
463 struct rte_ether_addr (*mac)[], int *mac_n)
467 struct ifinfomsg ifm;
470 .nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg)),
471 .nlmsg_type = RTM_GETNEIGH,
472 .nlmsg_flags = NLM_F_DUMP | NLM_F_REQUEST,
475 .ifi_family = PF_BRIDGE,
476 .ifi_index = iface_idx,
479 struct mlx5_nl_mac_addr data = {
483 uint32_t sn = MLX5_NL_SN_GENERATE;
488 ret = mlx5_nl_request(nlsk_fd, &req.hdr, sn, &req.ifm,
489 sizeof(struct ifinfomsg));
492 ret = mlx5_nl_recv(nlsk_fd, sn, mlx5_nl_mac_addr_cb, &data);
498 DRV_LOG(DEBUG, "Interface %u cannot retrieve MAC address list %s",
499 iface_idx, strerror(rte_errno));
504 * Modify the MAC address neighbour table with Netlink.
507 * Netlink socket file descriptor.
508 * @param[in] iface_idx
509 * Net device interface index.
511 * MAC address to consider.
513 * 1 to add the MAC address, 0 to remove the MAC address.
516 * 0 on success, a negative errno value otherwise and rte_errno is set.
519 mlx5_nl_mac_addr_modify(int nlsk_fd, unsigned int iface_idx,
520 struct rte_ether_addr *mac, int add)
526 uint8_t buffer[RTE_ETHER_ADDR_LEN];
529 .nlmsg_len = NLMSG_LENGTH(sizeof(struct ndmsg)),
530 .nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE |
531 NLM_F_EXCL | NLM_F_ACK,
532 .nlmsg_type = add ? RTM_NEWNEIGH : RTM_DELNEIGH,
535 .ndm_family = PF_BRIDGE,
536 .ndm_state = NUD_NOARP | NUD_PERMANENT,
537 .ndm_ifindex = iface_idx,
538 .ndm_flags = NTF_SELF,
541 .rta_type = NDA_LLADDR,
542 .rta_len = RTA_LENGTH(RTE_ETHER_ADDR_LEN),
545 uint32_t sn = MLX5_NL_SN_GENERATE;
550 memcpy(RTA_DATA(&req.rta), mac, RTE_ETHER_ADDR_LEN);
551 req.hdr.nlmsg_len = NLMSG_ALIGN(req.hdr.nlmsg_len) +
552 RTA_ALIGN(req.rta.rta_len);
553 ret = mlx5_nl_send(nlsk_fd, &req.hdr, sn);
556 ret = mlx5_nl_recv(nlsk_fd, sn, NULL, NULL);
561 #ifdef RTE_LIBRTE_MLX5_DEBUG
563 char m[RTE_ETHER_ADDR_FMT_SIZE];
565 rte_ether_format_addr(m, RTE_ETHER_ADDR_FMT_SIZE, mac);
567 "Interface %u cannot %s MAC address %s %s",
569 add ? "add" : "remove", m, strerror(rte_errno));
576 * Modify the VF MAC address neighbour table with Netlink.
579 * Netlink socket file descriptor.
580 * @param[in] iface_idx
581 * Net device interface index.
583 * MAC address to consider.
588 * 0 on success, a negative errno value otherwise and rte_errno is set.
591 mlx5_nl_vf_mac_addr_modify(int nlsk_fd, unsigned int iface_idx,
592 struct rte_ether_addr *mac, int vf_index)
597 struct ifinfomsg ifm;
598 struct rtattr vf_list_rta;
599 struct rtattr vf_info_rta;
600 struct rtattr vf_mac_rta;
601 struct ifla_vf_mac ivm;
604 .nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg)),
605 .nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK,
606 .nlmsg_type = RTM_BASE,
609 .ifi_index = iface_idx,
612 .rta_type = IFLA_VFINFO_LIST,
613 .rta_len = RTA_ALIGN(RTA_LENGTH(0)),
616 .rta_type = IFLA_VF_INFO,
617 .rta_len = RTA_ALIGN(RTA_LENGTH(0)),
620 .rta_type = IFLA_VF_MAC,
623 struct ifla_vf_mac ivm = {
626 uint32_t sn = MLX5_NL_SN_GENERATE;
628 memcpy(&ivm.mac, mac, RTE_ETHER_ADDR_LEN);
629 memcpy(RTA_DATA(&req.vf_mac_rta), &ivm, sizeof(ivm));
631 req.vf_mac_rta.rta_len = RTA_LENGTH(sizeof(ivm));
632 req.hdr.nlmsg_len = NLMSG_ALIGN(req.hdr.nlmsg_len) +
633 RTA_ALIGN(req.vf_list_rta.rta_len) +
634 RTA_ALIGN(req.vf_info_rta.rta_len) +
635 RTA_ALIGN(req.vf_mac_rta.rta_len);
636 req.vf_list_rta.rta_len = RTE_PTR_DIFF(NLMSG_TAIL(&req.hdr),
638 req.vf_info_rta.rta_len = RTE_PTR_DIFF(NLMSG_TAIL(&req.hdr),
643 ret = mlx5_nl_send(nlsk_fd, &req.hdr, sn);
646 ret = mlx5_nl_recv(nlsk_fd, sn, NULL, NULL);
652 "representor %u cannot set VF MAC address "
653 "%02X:%02X:%02X:%02X:%02X:%02X : %s",
655 mac->addr_bytes[0], mac->addr_bytes[1],
656 mac->addr_bytes[2], mac->addr_bytes[3],
657 mac->addr_bytes[4], mac->addr_bytes[5],
658 strerror(rte_errno));
666 * Netlink socket file descriptor.
667 * @param[in] iface_idx
668 * Net device interface index.
670 * BITFIELD_DECLARE array to store the mac.
672 * MAC address to register.
677 * 0 on success, a negative errno value otherwise and rte_errno is set.
680 mlx5_nl_mac_addr_add(int nlsk_fd, unsigned int iface_idx,
681 uint64_t *mac_own, struct rte_ether_addr *mac,
686 ret = mlx5_nl_mac_addr_modify(nlsk_fd, iface_idx, mac, 1);
688 MLX5_ASSERT(index < MLX5_MAX_MAC_ADDRESSES);
689 if (index >= MLX5_MAX_MAC_ADDRESSES)
692 BITFIELD_SET(mac_own, index);
700 * Remove a MAC address.
703 * Netlink socket file descriptor.
704 * @param[in] iface_idx
705 * Net device interface index.
707 * BITFIELD_DECLARE array to store the mac.
709 * MAC address to remove.
714 * 0 on success, a negative errno value otherwise and rte_errno is set.
717 mlx5_nl_mac_addr_remove(int nlsk_fd, unsigned int iface_idx, uint64_t *mac_own,
718 struct rte_ether_addr *mac, uint32_t index)
720 MLX5_ASSERT(index < MLX5_MAX_MAC_ADDRESSES);
721 if (index >= MLX5_MAX_MAC_ADDRESSES)
724 BITFIELD_RESET(mac_own, index);
725 return mlx5_nl_mac_addr_modify(nlsk_fd, iface_idx, mac, 0);
729 * Synchronize Netlink bridge table to the internal table.
732 * Netlink socket file descriptor.
733 * @param[in] iface_idx
734 * Net device interface index.
736 * Mac addresses array to sync.
738 * @p mac_addrs array size.
741 mlx5_nl_mac_addr_sync(int nlsk_fd, unsigned int iface_idx,
742 struct rte_ether_addr *mac_addrs, int n)
744 struct rte_ether_addr macs[n];
749 ret = mlx5_nl_mac_addr_list(nlsk_fd, iface_idx, &macs, &macs_n);
752 for (i = 0; i != macs_n; ++i) {
755 /* Verify the address is not in the array yet. */
756 for (j = 0; j != n; ++j)
757 if (rte_is_same_ether_addr(&macs[i], &mac_addrs[j]))
761 if (rte_is_multicast_ether_addr(&macs[i])) {
762 /* Find the first entry available. */
763 for (j = MLX5_MAX_UC_MAC_ADDRESSES; j != n; ++j) {
764 if (rte_is_zero_ether_addr(&mac_addrs[j])) {
765 mac_addrs[j] = macs[i];
770 /* Find the first entry available. */
771 for (j = 0; j != MLX5_MAX_UC_MAC_ADDRESSES; ++j) {
772 if (rte_is_zero_ether_addr(&mac_addrs[j])) {
773 mac_addrs[j] = macs[i];
782 * Flush all added MAC addresses.
785 * Netlink socket file descriptor.
786 * @param[in] iface_idx
787 * Net device interface index.
788 * @param[in] mac_addrs
789 * Mac addresses array to flush.
791 * @p mac_addrs array size.
793 * BITFIELD_DECLARE array to store the mac.
796 mlx5_nl_mac_addr_flush(int nlsk_fd, unsigned int iface_idx,
797 struct rte_ether_addr *mac_addrs, int n,
802 if (n <= 0 || n > MLX5_MAX_MAC_ADDRESSES)
805 for (i = n - 1; i >= 0; --i) {
806 struct rte_ether_addr *m = &mac_addrs[i];
808 if (BITFIELD_ISSET(mac_own, i))
809 mlx5_nl_mac_addr_remove(nlsk_fd, iface_idx, mac_own, m,
815 * Enable promiscuous / all multicast mode through Netlink.
818 * Netlink socket file descriptor.
819 * @param[in] iface_idx
820 * Net device interface index.
822 * IFF_PROMISC for promiscuous, IFF_ALLMULTI for allmulti.
824 * Nonzero to enable, disable otherwise.
827 * 0 on success, a negative errno value otherwise and rte_errno is set.
830 mlx5_nl_device_flags(int nlsk_fd, unsigned int iface_idx, uint32_t flags,
835 struct ifinfomsg ifi;
838 .nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg)),
839 .nlmsg_type = RTM_NEWLINK,
840 .nlmsg_flags = NLM_F_REQUEST,
843 .ifi_flags = enable ? flags : 0,
845 .ifi_index = iface_idx,
848 uint32_t sn = MLX5_NL_SN_GENERATE;
851 MLX5_ASSERT(!(flags & ~(IFF_PROMISC | IFF_ALLMULTI)));
854 ret = mlx5_nl_send(nlsk_fd, &req.hdr, sn);
861 * Enable promiscuous mode through Netlink.
864 * Netlink socket file descriptor.
865 * @param[in] iface_idx
866 * Net device interface index.
868 * Nonzero to enable, disable otherwise.
871 * 0 on success, a negative errno value otherwise and rte_errno is set.
874 mlx5_nl_promisc(int nlsk_fd, unsigned int iface_idx, int enable)
876 int ret = mlx5_nl_device_flags(nlsk_fd, iface_idx, IFF_PROMISC, enable);
880 "Interface %u cannot %s promisc mode: Netlink error %s",
881 iface_idx, enable ? "enable" : "disable",
882 strerror(rte_errno));
887 * Enable all multicast mode through Netlink.
890 * Netlink socket file descriptor.
891 * @param[in] iface_idx
892 * Net device interface index.
894 * Nonzero to enable, disable otherwise.
897 * 0 on success, a negative errno value otherwise and rte_errno is set.
900 mlx5_nl_allmulti(int nlsk_fd, unsigned int iface_idx, int enable)
902 int ret = mlx5_nl_device_flags(nlsk_fd, iface_idx, IFF_ALLMULTI,
907 "Interface %u cannot %s allmulti : Netlink error %s",
908 iface_idx, enable ? "enable" : "disable",
909 strerror(rte_errno));
914 * Process network interface information from Netlink message.
917 * Pointer to Netlink message header.
919 * Opaque data pointer for this callback.
922 * 0 on success, a negative errno value otherwise and rte_errno is set.
925 mlx5_nl_cmdget_cb(struct nlmsghdr *nh, void *arg)
927 struct mlx5_nl_ifindex_data *data = arg;
928 struct mlx5_nl_ifindex_data local = {
931 size_t off = NLMSG_HDRLEN;
933 if (nh->nlmsg_type !=
934 RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_GET) &&
936 RDMA_NL_GET_TYPE(RDMA_NL_NLDEV, RDMA_NLDEV_CMD_PORT_GET))
938 while (off < nh->nlmsg_len) {
939 struct nlattr *na = (void *)((uintptr_t)nh + off);
940 void *payload = (void *)((uintptr_t)na + NLA_HDRLEN);
942 if (na->nla_len > nh->nlmsg_len - off)
944 switch (na->nla_type) {
945 case RDMA_NLDEV_ATTR_DEV_INDEX:
946 local.ibindex = *(uint32_t *)payload;
947 local.flags |= MLX5_NL_CMD_GET_IB_INDEX;
949 case RDMA_NLDEV_ATTR_DEV_NAME:
950 if (!strcmp(payload, data->name))
951 local.flags |= MLX5_NL_CMD_GET_IB_NAME;
953 case RDMA_NLDEV_ATTR_NDEV_INDEX:
954 local.ifindex = *(uint32_t *)payload;
955 local.flags |= MLX5_NL_CMD_GET_NET_INDEX;
957 case RDMA_NLDEV_ATTR_PORT_INDEX:
958 local.portnum = *(uint32_t *)payload;
959 local.flags |= MLX5_NL_CMD_GET_PORT_INDEX;
964 off += NLA_ALIGN(na->nla_len);
967 * It is possible to have multiple messages for all
968 * Infiniband devices in the system with appropriate name.
969 * So we should gather parameters locally and copy to
970 * query context only in case of coinciding device name.
972 if (local.flags & MLX5_NL_CMD_GET_IB_NAME) {
973 data->flags = local.flags;
974 data->ibindex = local.ibindex;
975 data->ifindex = local.ifindex;
976 data->portnum = local.portnum;
985 * Get index of network interface associated with some IB device.
987 * This is the only somewhat safe method to avoid resorting to heuristics
988 * when faced with port representors. Unfortunately it requires at least
992 * Netlink socket of the RDMA kind (NETLINK_RDMA).
996 * IB device port index, starting from 1
998 * A valid (nonzero) interface index on success, 0 otherwise and rte_errno
1002 mlx5_nl_ifindex(int nl, const char *name, uint32_t pindex)
1004 struct mlx5_nl_ifindex_data data = {
1007 .ibindex = 0, /* Determined during first pass. */
1008 .ifindex = 0, /* Determined during second pass. */
1012 uint8_t buf[NLMSG_HDRLEN +
1013 NLA_HDRLEN + NLA_ALIGN(sizeof(data.ibindex)) +
1014 NLA_HDRLEN + NLA_ALIGN(sizeof(pindex))];
1017 .nlmsg_len = NLMSG_LENGTH(0),
1018 .nlmsg_type = RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
1019 RDMA_NLDEV_CMD_GET),
1020 .nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_DUMP,
1024 uint32_t sn = MLX5_NL_SN_GENERATE;
1027 ret = mlx5_nl_send(nl, &req.nh, sn);
1030 ret = mlx5_nl_recv(nl, sn, mlx5_nl_cmdget_cb, &data);
1033 if (!(data.flags & MLX5_NL_CMD_GET_IB_NAME) ||
1034 !(data.flags & MLX5_NL_CMD_GET_IB_INDEX))
1037 sn = MLX5_NL_SN_GENERATE;
1038 req.nh.nlmsg_type = RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
1039 RDMA_NLDEV_CMD_PORT_GET);
1040 req.nh.nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
1041 req.nh.nlmsg_len = NLMSG_LENGTH(sizeof(req.buf) - NLMSG_HDRLEN);
1042 na = (void *)((uintptr_t)req.buf + NLMSG_HDRLEN);
1043 na->nla_len = NLA_HDRLEN + sizeof(data.ibindex);
1044 na->nla_type = RDMA_NLDEV_ATTR_DEV_INDEX;
1045 memcpy((void *)((uintptr_t)na + NLA_HDRLEN),
1046 &data.ibindex, sizeof(data.ibindex));
1047 na = (void *)((uintptr_t)na + NLA_ALIGN(na->nla_len));
1048 na->nla_len = NLA_HDRLEN + sizeof(pindex);
1049 na->nla_type = RDMA_NLDEV_ATTR_PORT_INDEX;
1050 memcpy((void *)((uintptr_t)na + NLA_HDRLEN),
1051 &pindex, sizeof(pindex));
1052 ret = mlx5_nl_send(nl, &req.nh, sn);
1055 ret = mlx5_nl_recv(nl, sn, mlx5_nl_cmdget_cb, &data);
1058 if (!(data.flags & MLX5_NL_CMD_GET_IB_NAME) ||
1059 !(data.flags & MLX5_NL_CMD_GET_IB_INDEX) ||
1060 !(data.flags & MLX5_NL_CMD_GET_NET_INDEX) ||
1063 return data.ifindex;
1070 * Get the number of physical ports of given IB device.
1073 * Netlink socket of the RDMA kind (NETLINK_RDMA).
1078 * A valid (nonzero) number of ports on success, 0 otherwise
1079 * and rte_errno is set.
1082 mlx5_nl_portnum(int nl, const char *name)
1084 struct mlx5_nl_ifindex_data data = {
1090 struct nlmsghdr req = {
1091 .nlmsg_len = NLMSG_LENGTH(0),
1092 .nlmsg_type = RDMA_NL_GET_TYPE(RDMA_NL_NLDEV,
1093 RDMA_NLDEV_CMD_GET),
1094 .nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK | NLM_F_DUMP,
1096 uint32_t sn = MLX5_NL_SN_GENERATE;
1099 ret = mlx5_nl_send(nl, &req, sn);
1102 ret = mlx5_nl_recv(nl, sn, mlx5_nl_cmdget_cb, &data);
1105 if (!(data.flags & MLX5_NL_CMD_GET_IB_NAME) ||
1106 !(data.flags & MLX5_NL_CMD_GET_IB_INDEX) ||
1107 !(data.flags & MLX5_NL_CMD_GET_PORT_INDEX)) {
1113 return data.portnum;
1117 * Analyze gathered port parameters via Netlink to recognize master
1118 * and representor devices for E-Switch configuration.
1120 * @param[in] num_vf_set
1121 * flag of presence of number of VFs port attribute.
1122 * @param[inout] switch_info
1123 * Port information, including port name as a number and port name
1124 * type if recognized
1127 * master and representor flags are set in switch_info according to
1128 * recognized parameters (if any).
1131 mlx5_nl_check_switch_info(bool num_vf_set,
1132 struct mlx5_switch_info *switch_info)
1134 switch (switch_info->name_type) {
1135 case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN:
1137 * Name is not recognized, assume the master,
1138 * check the number of VFs key presence.
1140 switch_info->master = num_vf_set;
1142 case MLX5_PHYS_PORT_NAME_TYPE_NOTSET:
1144 * Name is not set, this assumes the legacy naming
1145 * schema for master, just check if there is a
1146 * number of VFs key.
1148 switch_info->master = num_vf_set;
1150 case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
1151 /* New uplink naming schema recognized. */
1152 switch_info->master = 1;
1154 case MLX5_PHYS_PORT_NAME_TYPE_LEGACY:
1155 /* Legacy representors naming schema. */
1156 switch_info->representor = !num_vf_set;
1158 case MLX5_PHYS_PORT_NAME_TYPE_PFHPF:
1160 case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
1161 /* New representors naming schema. */
1162 switch_info->representor = 1;
1168 * Process switch information from Netlink message.
1171 * Pointer to Netlink message header.
1173 * Opaque data pointer for this callback.
1176 * 0 on success, a negative errno value otherwise and rte_errno is set.
1179 mlx5_nl_switch_info_cb(struct nlmsghdr *nh, void *arg)
1181 struct mlx5_switch_info info = {
1184 .name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET,
1188 size_t off = NLMSG_LENGTH(sizeof(struct ifinfomsg));
1189 bool switch_id_set = false;
1190 bool num_vf_set = false;
1192 if (nh->nlmsg_type != RTM_NEWLINK)
1194 while (off < nh->nlmsg_len) {
1195 struct rtattr *ra = (void *)((uintptr_t)nh + off);
1196 void *payload = RTA_DATA(ra);
1199 if (ra->rta_len > nh->nlmsg_len - off)
1201 switch (ra->rta_type) {
1205 case IFLA_PHYS_PORT_NAME:
1206 mlx5_translate_port_name((char *)payload, &info);
1208 case IFLA_PHYS_SWITCH_ID:
1210 for (i = 0; i < RTA_PAYLOAD(ra); ++i) {
1211 info.switch_id <<= 8;
1212 info.switch_id |= ((uint8_t *)payload)[i];
1214 switch_id_set = true;
1217 off += RTA_ALIGN(ra->rta_len);
1219 if (switch_id_set) {
1220 /* We have some E-Switch configuration. */
1221 mlx5_nl_check_switch_info(num_vf_set, &info);
1223 MLX5_ASSERT(!(info.master && info.representor));
1224 memcpy(arg, &info, sizeof(info));
1232 * Get switch information associated with network interface.
1235 * Netlink socket of the ROUTE kind (NETLINK_ROUTE).
1237 * Network interface index.
1239 * Switch information object, populated in case of success.
1242 * 0 on success, a negative errno value otherwise and rte_errno is set.
1245 mlx5_nl_switch_info(int nl, unsigned int ifindex,
1246 struct mlx5_switch_info *info)
1250 struct ifinfomsg info;
1255 .nlmsg_len = NLMSG_LENGTH
1257 RTA_LENGTH(sizeof(uint32_t))),
1258 .nlmsg_type = RTM_GETLINK,
1259 .nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK,
1262 .ifi_family = AF_UNSPEC,
1263 .ifi_index = ifindex,
1266 .rta_type = IFLA_EXT_MASK,
1267 .rta_len = RTA_LENGTH(sizeof(int32_t)),
1269 .extmask = RTE_LE32(1),
1271 uint32_t sn = MLX5_NL_SN_GENERATE;
1274 ret = mlx5_nl_send(nl, &req.nh, sn);
1276 ret = mlx5_nl_recv(nl, sn, mlx5_nl_switch_info_cb, info);
1277 if (info->master && info->representor) {
1278 DRV_LOG(ERR, "ifindex %u device is recognized as master"
1279 " and as representor", ifindex);
1287 * Delete VLAN network device by ifindex.
1290 * Context object initialized by mlx5_nl_vlan_vmwa_init().
1291 * @param[in] ifindex
1292 * Interface index of network device to delete.
1295 mlx5_nl_vlan_vmwa_delete(struct mlx5_nl_vlan_vmwa_context *vmwa,
1298 uint32_t sn = MLX5_NL_SN_GENERATE;
1302 struct ifinfomsg info;
1305 .nlmsg_len = NLMSG_LENGTH(sizeof(struct ifinfomsg)),
1306 .nlmsg_type = RTM_DELLINK,
1307 .nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK,
1310 .ifi_family = AF_UNSPEC,
1311 .ifi_index = ifindex,
1316 ret = mlx5_nl_send(vmwa->nl_socket, &req.nh, sn);
1318 ret = mlx5_nl_recv(vmwa->nl_socket, sn, NULL, NULL);
1320 DRV_LOG(WARNING, "netlink: error deleting VLAN WA"
1321 " ifindex %u, %d", ifindex, ret);
1325 /* Set of subroutines to build Netlink message. */
1326 static struct nlattr *
1327 nl_msg_tail(struct nlmsghdr *nlh)
1329 return (struct nlattr *)
1330 (((uint8_t *)nlh) + NLMSG_ALIGN(nlh->nlmsg_len));
1334 nl_attr_put(struct nlmsghdr *nlh, int type, const void *data, int alen)
1336 struct nlattr *nla = nl_msg_tail(nlh);
1338 nla->nla_type = type;
1339 nla->nla_len = NLMSG_ALIGN(sizeof(struct nlattr)) + alen;
1340 nlh->nlmsg_len += NLMSG_ALIGN(nla->nla_len);
1343 memcpy((uint8_t *)nla + sizeof(struct nlattr), data, alen);
1346 static struct nlattr *
1347 nl_attr_nest_start(struct nlmsghdr *nlh, int type)
1349 struct nlattr *nest = (struct nlattr *)nl_msg_tail(nlh);
1351 nl_attr_put(nlh, type, NULL, 0);
1356 nl_attr_nest_end(struct nlmsghdr *nlh, struct nlattr *nest)
1358 nest->nla_len = (uint8_t *)nl_msg_tail(nlh) - (uint8_t *)nest;
1362 * Create network VLAN device with specified VLAN tag.
1365 * Context object initialized by mlx5_nl_vlan_vmwa_init().
1366 * @param[in] ifindex
1367 * Base network interface index.
1369 * VLAN tag for VLAN network device to create.
1372 mlx5_nl_vlan_vmwa_create(struct mlx5_nl_vlan_vmwa_context *vmwa,
1373 uint32_t ifindex, uint16_t tag)
1375 struct nlmsghdr *nlh;
1376 struct ifinfomsg *ifm;
1377 char name[sizeof(MLX5_VMWA_VLAN_DEVICE_PFX) + 32];
1380 uint8_t buf[NLMSG_ALIGN(sizeof(struct nlmsghdr)) +
1381 NLMSG_ALIGN(sizeof(struct ifinfomsg)) +
1382 NLMSG_ALIGN(sizeof(struct nlattr)) * 8 +
1383 NLMSG_ALIGN(sizeof(uint32_t)) +
1384 NLMSG_ALIGN(sizeof(name)) +
1385 NLMSG_ALIGN(sizeof("vlan")) +
1386 NLMSG_ALIGN(sizeof(uint32_t)) +
1387 NLMSG_ALIGN(sizeof(uint16_t)) + 16];
1388 struct nlattr *na_info;
1389 struct nlattr *na_vlan;
1390 uint32_t sn = MLX5_NL_SN_GENERATE;
1393 memset(buf, 0, sizeof(buf));
1394 nlh = (struct nlmsghdr *)buf;
1395 nlh->nlmsg_len = sizeof(struct nlmsghdr);
1396 nlh->nlmsg_type = RTM_NEWLINK;
1397 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE |
1398 NLM_F_EXCL | NLM_F_ACK;
1399 ifm = (struct ifinfomsg *)nl_msg_tail(nlh);
1400 nlh->nlmsg_len += sizeof(struct ifinfomsg);
1401 ifm->ifi_family = AF_UNSPEC;
1404 ifm->ifi_flags = IFF_UP;
1405 ifm->ifi_change = 0xffffffff;
1406 nl_attr_put(nlh, IFLA_LINK, &ifindex, sizeof(ifindex));
1407 ret = snprintf(name, sizeof(name), "%s.%u.%u",
1408 MLX5_VMWA_VLAN_DEVICE_PFX, ifindex, tag);
1409 nl_attr_put(nlh, IFLA_IFNAME, name, ret + 1);
1410 na_info = nl_attr_nest_start(nlh, IFLA_LINKINFO);
1411 nl_attr_put(nlh, IFLA_INFO_KIND, "vlan", sizeof("vlan"));
1412 na_vlan = nl_attr_nest_start(nlh, IFLA_INFO_DATA);
1413 nl_attr_put(nlh, IFLA_VLAN_ID, &tag, sizeof(tag));
1414 nl_attr_nest_end(nlh, na_vlan);
1415 nl_attr_nest_end(nlh, na_info);
1416 MLX5_ASSERT(sizeof(buf) >= nlh->nlmsg_len);
1417 ret = mlx5_nl_send(vmwa->nl_socket, nlh, sn);
1419 ret = mlx5_nl_recv(vmwa->nl_socket, sn, NULL, NULL);
1421 DRV_LOG(WARNING, "netlink: VLAN %s create failure (%d)", name,
1424 /* Try to get ifindex of created or pre-existing device. */
1425 ret = if_nametoindex(name);
1427 DRV_LOG(WARNING, "VLAN %s failed to get index (%d)", name,
1435 * Parse Netlink message to retrieve the general family ID.
1438 * Pointer to Netlink Message Header.
1440 * PMD data register with this callback.
1443 * 0 on success, a negative errno value otherwise and rte_errno is set.
1446 mlx5_nl_family_id_cb(struct nlmsghdr *nh, void *arg)
1449 struct nlattr *tail = RTE_PTR_ADD(nh, nh->nlmsg_len);
1450 struct nlattr *nla = RTE_PTR_ADD(nh, NLMSG_ALIGN(sizeof(*nh)) +
1451 NLMSG_ALIGN(sizeof(struct genlmsghdr)));
1453 for (; nla->nla_len && nla < tail;
1454 nla = RTE_PTR_ADD(nla, NLMSG_ALIGN(nla->nla_len))) {
1455 if (nla->nla_type == CTRL_ATTR_FAMILY_ID) {
1456 *(uint16_t *)arg = *(uint16_t *)(nla + 1);
1463 #define MLX5_NL_MAX_ATTR_SIZE 100
1465 * Get generic netlink family ID.
1467 * @param[in] nlsk_fd
1468 * Netlink socket file descriptor.
1473 * ID >= 0 on success and @p enable is updated, a negative errno value
1474 * otherwise and rte_errno is set.
1477 mlx5_nl_generic_family_id_get(int nlsk_fd, const char *name)
1479 struct nlmsghdr *nlh;
1480 struct genlmsghdr *genl;
1481 uint32_t sn = MLX5_NL_SN_GENERATE;
1482 int name_size = strlen(name) + 1;
1485 uint8_t buf[NLMSG_ALIGN(sizeof(struct nlmsghdr)) +
1486 NLMSG_ALIGN(sizeof(struct genlmsghdr)) +
1487 NLMSG_ALIGN(sizeof(struct nlattr)) +
1488 NLMSG_ALIGN(MLX5_NL_MAX_ATTR_SIZE)];
1490 memset(buf, 0, sizeof(buf));
1491 nlh = (struct nlmsghdr *)buf;
1492 nlh->nlmsg_len = sizeof(struct nlmsghdr);
1493 nlh->nlmsg_type = GENL_ID_CTRL;
1494 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
1495 genl = (struct genlmsghdr *)nl_msg_tail(nlh);
1496 nlh->nlmsg_len += sizeof(struct genlmsghdr);
1497 genl->cmd = CTRL_CMD_GETFAMILY;
1499 nl_attr_put(nlh, CTRL_ATTR_FAMILY_NAME, name, name_size);
1500 ret = mlx5_nl_send(nlsk_fd, nlh, sn);
1502 ret = mlx5_nl_recv(nlsk_fd, sn, mlx5_nl_family_id_cb, &id);
1504 DRV_LOG(DEBUG, "Failed to get Netlink %s family ID: %d.", name,
1508 DRV_LOG(DEBUG, "Netlink \"%s\" family ID is %u.", name, id);
1513 * Get Devlink family ID.
1515 * @param[in] nlsk_fd
1516 * Netlink socket file descriptor.
1519 * ID >= 0 on success and @p enable is updated, a negative errno value
1520 * otherwise and rte_errno is set.
1524 mlx5_nl_devlink_family_id_get(int nlsk_fd)
1526 return mlx5_nl_generic_family_id_get(nlsk_fd, DEVLINK_GENL_NAME);
1530 * Parse Netlink message to retrieve the ROCE enable status.
1533 * Pointer to Netlink Message Header.
1535 * PMD data register with this callback.
1538 * 0 on success, a negative errno value otherwise and rte_errno is set.
1541 mlx5_nl_roce_cb(struct nlmsghdr *nh, void *arg)
1546 struct nlattr *tail = RTE_PTR_ADD(nh, nh->nlmsg_len);
1547 struct nlattr *nla = RTE_PTR_ADD(nh, NLMSG_ALIGN(sizeof(*nh)) +
1548 NLMSG_ALIGN(sizeof(struct genlmsghdr)));
1550 while (nla->nla_len && nla < tail) {
1551 switch (nla->nla_type) {
1552 /* Expected nested attributes case. */
1553 case DEVLINK_ATTR_PARAM:
1554 case DEVLINK_ATTR_PARAM_VALUES_LIST:
1555 case DEVLINK_ATTR_PARAM_VALUE:
1559 case DEVLINK_ATTR_PARAM_VALUE_DATA:
1563 nla = RTE_PTR_ADD(nla, NLMSG_ALIGN(nla->nla_len));
1571 * Get ROCE enable status through Netlink.
1573 * @param[in] nlsk_fd
1574 * Netlink socket file descriptor.
1575 * @param[in] family_id
1576 * the Devlink family ID.
1578 * The device PCI address.
1579 * @param[out] enable
1580 * Where to store the enable status.
1583 * 0 on success and @p enable is updated, a negative errno value otherwise
1584 * and rte_errno is set.
1587 mlx5_nl_enable_roce_get(int nlsk_fd, int family_id, const char *pci_addr,
1590 struct nlmsghdr *nlh;
1591 struct genlmsghdr *genl;
1592 uint32_t sn = MLX5_NL_SN_GENERATE;
1595 uint8_t buf[NLMSG_ALIGN(sizeof(struct nlmsghdr)) +
1596 NLMSG_ALIGN(sizeof(struct genlmsghdr)) +
1597 NLMSG_ALIGN(sizeof(struct nlattr)) * 4 +
1598 NLMSG_ALIGN(MLX5_NL_MAX_ATTR_SIZE) * 4];
1600 memset(buf, 0, sizeof(buf));
1601 nlh = (struct nlmsghdr *)buf;
1602 nlh->nlmsg_len = sizeof(struct nlmsghdr);
1603 nlh->nlmsg_type = family_id;
1604 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
1605 genl = (struct genlmsghdr *)nl_msg_tail(nlh);
1606 nlh->nlmsg_len += sizeof(struct genlmsghdr);
1607 genl->cmd = DEVLINK_CMD_PARAM_GET;
1608 genl->version = DEVLINK_GENL_VERSION;
1609 nl_attr_put(nlh, DEVLINK_ATTR_BUS_NAME, "pci", 4);
1610 nl_attr_put(nlh, DEVLINK_ATTR_DEV_NAME, pci_addr, strlen(pci_addr) + 1);
1611 nl_attr_put(nlh, DEVLINK_ATTR_PARAM_NAME, "enable_roce", 12);
1612 ret = mlx5_nl_send(nlsk_fd, nlh, sn);
1614 ret = mlx5_nl_recv(nlsk_fd, sn, mlx5_nl_roce_cb, &cur_en);
1616 DRV_LOG(DEBUG, "Failed to get ROCE enable on device %s: %d.",
1621 DRV_LOG(DEBUG, "ROCE is %sabled for device \"%s\".",
1622 cur_en ? "en" : "dis", pci_addr);
1627 * Reload mlx5 device kernel driver through Netlink.
1629 * @param[in] nlsk_fd
1630 * Netlink socket file descriptor.
1631 * @param[in] family_id
1632 * the Devlink family ID.
1634 * The device PCI address.
1635 * @param[out] enable
1636 * The enable status to set.
1639 * 0 on success, a negative errno value otherwise and rte_errno is set.
1642 mlx5_nl_driver_reload(int nlsk_fd, int family_id, const char *pci_addr)
1644 struct nlmsghdr *nlh;
1645 struct genlmsghdr *genl;
1646 uint32_t sn = MLX5_NL_SN_GENERATE;
1648 uint8_t buf[NLMSG_ALIGN(sizeof(struct nlmsghdr)) +
1649 NLMSG_ALIGN(sizeof(struct genlmsghdr)) +
1650 NLMSG_ALIGN(sizeof(struct nlattr)) * 2 +
1651 NLMSG_ALIGN(MLX5_NL_MAX_ATTR_SIZE) * 2];
1653 memset(buf, 0, sizeof(buf));
1654 nlh = (struct nlmsghdr *)buf;
1655 nlh->nlmsg_len = sizeof(struct nlmsghdr);
1656 nlh->nlmsg_type = family_id;
1657 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
1658 genl = (struct genlmsghdr *)nl_msg_tail(nlh);
1659 nlh->nlmsg_len += sizeof(struct genlmsghdr);
1660 genl->cmd = DEVLINK_CMD_RELOAD;
1661 genl->version = DEVLINK_GENL_VERSION;
1662 nl_attr_put(nlh, DEVLINK_ATTR_BUS_NAME, "pci", 4);
1663 nl_attr_put(nlh, DEVLINK_ATTR_DEV_NAME, pci_addr, strlen(pci_addr) + 1);
1664 ret = mlx5_nl_send(nlsk_fd, nlh, sn);
1666 ret = mlx5_nl_recv(nlsk_fd, sn, NULL, NULL);
1668 DRV_LOG(DEBUG, "Failed to reload %s device by Netlink - %d",
1672 DRV_LOG(DEBUG, "Device \"%s\" was reloaded by Netlink successfully.",
1678 * Set ROCE enable status through Netlink.
1680 * @param[in] nlsk_fd
1681 * Netlink socket file descriptor.
1682 * @param[in] family_id
1683 * the Devlink family ID.
1685 * The device PCI address.
1686 * @param[out] enable
1687 * The enable status to set.
1690 * 0 on success, a negative errno value otherwise and rte_errno is set.
1693 mlx5_nl_enable_roce_set(int nlsk_fd, int family_id, const char *pci_addr,
1696 struct nlmsghdr *nlh;
1697 struct genlmsghdr *genl;
1698 uint32_t sn = MLX5_NL_SN_GENERATE;
1700 uint8_t buf[NLMSG_ALIGN(sizeof(struct nlmsghdr)) +
1701 NLMSG_ALIGN(sizeof(struct genlmsghdr)) +
1702 NLMSG_ALIGN(sizeof(struct nlattr)) * 6 +
1703 NLMSG_ALIGN(MLX5_NL_MAX_ATTR_SIZE) * 6];
1704 uint8_t cmode = DEVLINK_PARAM_CMODE_DRIVERINIT;
1705 uint8_t ptype = NLA_FLAG;
1708 memset(buf, 0, sizeof(buf));
1709 nlh = (struct nlmsghdr *)buf;
1710 nlh->nlmsg_len = sizeof(struct nlmsghdr);
1711 nlh->nlmsg_type = family_id;
1712 nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ACK;
1713 genl = (struct genlmsghdr *)nl_msg_tail(nlh);
1714 nlh->nlmsg_len += sizeof(struct genlmsghdr);
1715 genl->cmd = DEVLINK_CMD_PARAM_SET;
1716 genl->version = DEVLINK_GENL_VERSION;
1717 nl_attr_put(nlh, DEVLINK_ATTR_BUS_NAME, "pci", 4);
1718 nl_attr_put(nlh, DEVLINK_ATTR_DEV_NAME, pci_addr, strlen(pci_addr) + 1);
1719 nl_attr_put(nlh, DEVLINK_ATTR_PARAM_NAME, "enable_roce", 12);
1720 nl_attr_put(nlh, DEVLINK_ATTR_PARAM_VALUE_CMODE, &cmode, sizeof(cmode));
1721 nl_attr_put(nlh, DEVLINK_ATTR_PARAM_TYPE, &ptype, sizeof(ptype));
1723 nl_attr_put(nlh, DEVLINK_ATTR_PARAM_VALUE_DATA, NULL, 0);
1724 ret = mlx5_nl_send(nlsk_fd, nlh, sn);
1726 ret = mlx5_nl_recv(nlsk_fd, sn, NULL, NULL);
1728 DRV_LOG(DEBUG, "Failed to %sable ROCE for device %s by Netlink:"
1729 " %d.", enable ? "en" : "dis", pci_addr, ret);
1732 DRV_LOG(DEBUG, "Device %s ROCE was %sabled by Netlink successfully.",
1733 pci_addr, enable ? "en" : "dis");
1734 /* Now, need to reload the driver. */
1735 return mlx5_nl_driver_reload(nlsk_fd, family_id, pci_addr);