4 * Copyright 2015 6WIND S.A.
5 * Copyright 2015 Mellanox.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of 6WIND S.A. nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
44 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
46 #pragma GCC diagnostic ignored "-Wpedantic"
48 #include <infiniband/verbs.h>
50 #pragma GCC diagnostic error "-Wpedantic"
53 #include <rte_malloc.h>
54 #include <rte_ethdev.h>
55 #include <rte_ethdev_pci.h>
57 #include <rte_common.h>
58 #include <rte_kvargs.h>
61 #include "mlx5_utils.h"
62 #include "mlx5_rxtx.h"
63 #include "mlx5_autoconf.h"
64 #include "mlx5_defs.h"
66 /* Device parameter to enable RX completion queue compression. */
67 #define MLX5_RXQ_CQE_COMP_EN "rxq_cqe_comp_en"
69 /* Device parameter to configure inline send. */
70 #define MLX5_TXQ_INLINE "txq_inline"
73 * Device parameter to configure the number of TX queues threshold for
74 * enabling inline send.
76 #define MLX5_TXQS_MIN_INLINE "txqs_min_inline"
78 /* Device parameter to enable multi-packet send WQEs. */
79 #define MLX5_TXQ_MPW_EN "txq_mpw_en"
81 /* Device parameter to include 2 dsegs in the title WQEBB. */
82 #define MLX5_TXQ_MPW_HDR_DSEG_EN "txq_mpw_hdr_dseg_en"
84 /* Device parameter to limit the size of inlining packet. */
85 #define MLX5_TXQ_MAX_INLINE_LEN "txq_max_inline_len"
87 /* Device parameter to enable hardware TSO offload. */
88 #define MLX5_TSO "tso"
90 /* Device parameter to enable hardware Tx vector. */
91 #define MLX5_TX_VEC_EN "tx_vec_en"
93 /* Device parameter to enable hardware Rx vector. */
94 #define MLX5_RX_VEC_EN "rx_vec_en"
96 /* Default PMD specific parameter value. */
97 #define MLX5_ARG_UNSET (-1)
105 int inline_max_packet_sz;
111 * Retrieve integer value from environment variable.
114 * Environment variable name.
117 * Integer value, 0 if the variable is not set.
120 mlx5_getenv_int(const char *name)
122 const char *val = getenv(name);
130 * DPDK callback to close the device.
132 * Destroy all queues and objects, free memory.
135 * Pointer to Ethernet device structure.
138 mlx5_dev_close(struct rte_eth_dev *dev)
140 struct priv *priv = mlx5_get_priv(dev);
144 DEBUG("%p: closing device \"%s\"",
146 ((priv->ctx != NULL) ? priv->ctx->device->name : ""));
147 /* In case mlx5_dev_stop() has not been called. */
148 priv_dev_interrupt_handler_uninstall(priv, dev);
149 priv_special_flow_disable_all(priv);
150 priv_mac_addrs_disable(priv);
151 priv_destroy_hash_rxqs(priv);
153 /* Remove flow director elements. */
154 priv_fdir_disable(priv);
155 priv_fdir_delete_filters_list(priv);
157 /* Prevent crashes when queues are still in use. */
158 dev->rx_pkt_burst = removed_rx_burst;
159 dev->tx_pkt_burst = removed_tx_burst;
160 if (priv->rxqs != NULL) {
161 /* XXX race condition if mlx5_rx_burst() is still running. */
163 for (i = 0; (i != priv->rxqs_n); ++i) {
164 struct rxq *rxq = (*priv->rxqs)[i];
165 struct rxq_ctrl *rxq_ctrl;
169 rxq_ctrl = container_of(rxq, struct rxq_ctrl, rxq);
170 (*priv->rxqs)[i] = NULL;
171 rxq_cleanup(rxq_ctrl);
177 if (priv->txqs != NULL) {
178 /* XXX race condition if mlx5_tx_burst() is still running. */
180 for (i = 0; (i != priv->txqs_n); ++i) {
181 struct txq *txq = (*priv->txqs)[i];
182 struct txq_ctrl *txq_ctrl;
186 txq_ctrl = container_of(txq, struct txq_ctrl, txq);
187 (*priv->txqs)[i] = NULL;
188 txq_cleanup(txq_ctrl);
194 if (priv->pd != NULL) {
195 assert(priv->ctx != NULL);
196 claim_zero(ibv_dealloc_pd(priv->pd));
197 claim_zero(ibv_close_device(priv->ctx));
199 assert(priv->ctx == NULL);
200 if (priv->rss_conf != NULL) {
201 for (i = 0; (i != hash_rxq_init_n); ++i)
202 rte_free((*priv->rss_conf)[i]);
203 rte_free(priv->rss_conf);
205 if (priv->reta_idx != NULL)
206 rte_free(priv->reta_idx);
208 memset(priv, 0, sizeof(*priv));
211 static const struct eth_dev_ops mlx5_dev_ops = {
212 .dev_configure = mlx5_dev_configure,
213 .dev_start = mlx5_dev_start,
214 .dev_stop = mlx5_dev_stop,
215 .dev_set_link_down = mlx5_set_link_down,
216 .dev_set_link_up = mlx5_set_link_up,
217 .dev_close = mlx5_dev_close,
218 .promiscuous_enable = mlx5_promiscuous_enable,
219 .promiscuous_disable = mlx5_promiscuous_disable,
220 .allmulticast_enable = mlx5_allmulticast_enable,
221 .allmulticast_disable = mlx5_allmulticast_disable,
222 .link_update = mlx5_link_update,
223 .stats_get = mlx5_stats_get,
224 .stats_reset = mlx5_stats_reset,
225 .xstats_get = mlx5_xstats_get,
226 .xstats_reset = mlx5_xstats_reset,
227 .xstats_get_names = mlx5_xstats_get_names,
228 .dev_infos_get = mlx5_dev_infos_get,
229 .dev_supported_ptypes_get = mlx5_dev_supported_ptypes_get,
230 .vlan_filter_set = mlx5_vlan_filter_set,
231 .rx_queue_setup = mlx5_rx_queue_setup,
232 .tx_queue_setup = mlx5_tx_queue_setup,
233 .rx_queue_release = mlx5_rx_queue_release,
234 .tx_queue_release = mlx5_tx_queue_release,
235 .flow_ctrl_get = mlx5_dev_get_flow_ctrl,
236 .flow_ctrl_set = mlx5_dev_set_flow_ctrl,
237 .mac_addr_remove = mlx5_mac_addr_remove,
238 .mac_addr_add = mlx5_mac_addr_add,
239 .mac_addr_set = mlx5_mac_addr_set,
240 .mtu_set = mlx5_dev_set_mtu,
241 .vlan_strip_queue_set = mlx5_vlan_strip_queue_set,
242 .vlan_offload_set = mlx5_vlan_offload_set,
243 .reta_update = mlx5_dev_rss_reta_update,
244 .reta_query = mlx5_dev_rss_reta_query,
245 .rss_hash_update = mlx5_rss_hash_update,
246 .rss_hash_conf_get = mlx5_rss_hash_conf_get,
247 .filter_ctrl = mlx5_dev_filter_ctrl,
248 .rx_descriptor_status = mlx5_rx_descriptor_status,
249 .tx_descriptor_status = mlx5_tx_descriptor_status,
250 #ifdef HAVE_UPDATE_CQ_CI
251 .rx_queue_intr_enable = mlx5_rx_intr_enable,
252 .rx_queue_intr_disable = mlx5_rx_intr_disable,
257 struct rte_pci_addr pci_addr; /* associated PCI address */
258 uint32_t ports; /* physical ports bitfield. */
262 * Get device index in mlx5_dev[] from PCI bus address.
264 * @param[in] pci_addr
265 * PCI bus address to look for.
268 * mlx5_dev[] index on success, -1 on failure.
271 mlx5_dev_idx(struct rte_pci_addr *pci_addr)
276 assert(pci_addr != NULL);
277 for (i = 0; (i != RTE_DIM(mlx5_dev)); ++i) {
278 if ((mlx5_dev[i].pci_addr.domain == pci_addr->domain) &&
279 (mlx5_dev[i].pci_addr.bus == pci_addr->bus) &&
280 (mlx5_dev[i].pci_addr.devid == pci_addr->devid) &&
281 (mlx5_dev[i].pci_addr.function == pci_addr->function))
283 if ((mlx5_dev[i].ports == 0) && (ret == -1))
290 * Verify and store value for device argument.
293 * Key argument to verify.
295 * Value associated with key.
300 * 0 on success, negative errno value on failure.
303 mlx5_args_check(const char *key, const char *val, void *opaque)
305 struct mlx5_args *args = opaque;
309 tmp = strtoul(val, NULL, 0);
311 WARN("%s: \"%s\" is not a valid integer", key, val);
314 if (strcmp(MLX5_RXQ_CQE_COMP_EN, key) == 0) {
315 args->cqe_comp = !!tmp;
316 } else if (strcmp(MLX5_TXQ_INLINE, key) == 0) {
317 args->txq_inline = tmp;
318 } else if (strcmp(MLX5_TXQS_MIN_INLINE, key) == 0) {
319 args->txqs_inline = tmp;
320 } else if (strcmp(MLX5_TXQ_MPW_EN, key) == 0) {
322 } else if (strcmp(MLX5_TXQ_MPW_HDR_DSEG_EN, key) == 0) {
323 args->mpw_hdr_dseg = !!tmp;
324 } else if (strcmp(MLX5_TXQ_MAX_INLINE_LEN, key) == 0) {
325 args->inline_max_packet_sz = tmp;
326 } else if (strcmp(MLX5_TSO, key) == 0) {
328 } else if (strcmp(MLX5_TX_VEC_EN, key) == 0) {
329 args->tx_vec_en = !!tmp;
330 } else if (strcmp(MLX5_RX_VEC_EN, key) == 0) {
331 args->rx_vec_en = !!tmp;
333 WARN("%s: unknown parameter", key);
340 * Parse device parameters.
343 * Pointer to private structure.
345 * Device arguments structure.
348 * 0 on success, errno value on failure.
351 mlx5_args(struct mlx5_args *args, struct rte_devargs *devargs)
353 const char **params = (const char *[]){
354 MLX5_RXQ_CQE_COMP_EN,
356 MLX5_TXQS_MIN_INLINE,
358 MLX5_TXQ_MPW_HDR_DSEG_EN,
359 MLX5_TXQ_MAX_INLINE_LEN,
365 struct rte_kvargs *kvlist;
371 /* Following UGLY cast is done to pass checkpatch. */
372 kvlist = rte_kvargs_parse(devargs->args, params);
375 /* Process parameters. */
376 for (i = 0; (params[i] != NULL); ++i) {
377 if (rte_kvargs_count(kvlist, params[i])) {
378 ret = rte_kvargs_process(kvlist, params[i],
379 mlx5_args_check, args);
381 rte_kvargs_free(kvlist);
386 rte_kvargs_free(kvlist);
390 static struct rte_pci_driver mlx5_driver;
393 * Assign parameters from args into priv, only non default
394 * values are considered.
397 * Pointer to private structure.
399 * Pointer to args values.
402 mlx5_args_assign(struct priv *priv, struct mlx5_args *args)
404 if (args->cqe_comp != MLX5_ARG_UNSET)
405 priv->cqe_comp = args->cqe_comp;
406 if (args->txq_inline != MLX5_ARG_UNSET)
407 priv->txq_inline = args->txq_inline;
408 if (args->txqs_inline != MLX5_ARG_UNSET)
409 priv->txqs_inline = args->txqs_inline;
410 if (args->mps != MLX5_ARG_UNSET)
411 priv->mps = args->mps ? priv->mps : 0;
412 if (args->mpw_hdr_dseg != MLX5_ARG_UNSET)
413 priv->mpw_hdr_dseg = args->mpw_hdr_dseg;
414 if (args->inline_max_packet_sz != MLX5_ARG_UNSET)
415 priv->inline_max_packet_sz = args->inline_max_packet_sz;
416 if (args->tso != MLX5_ARG_UNSET)
417 priv->tso = args->tso;
418 if (args->tx_vec_en != MLX5_ARG_UNSET)
419 priv->tx_vec_en = args->tx_vec_en;
420 if (args->rx_vec_en != MLX5_ARG_UNSET)
421 priv->rx_vec_en = args->rx_vec_en;
425 * DPDK callback to register a PCI device.
427 * This function creates an Ethernet device for each port of a given
431 * PCI driver structure (mlx5_driver).
433 * PCI device information.
436 * 0 on success, negative errno value on failure.
439 mlx5_pci_probe(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
441 struct ibv_device **list;
442 struct ibv_device *ibv_dev;
444 struct ibv_context *attr_ctx = NULL;
445 struct ibv_device_attr device_attr;
448 unsigned int tunnel_en = 0;
453 assert(pci_drv == &mlx5_driver);
454 /* Get mlx5_dev[] index. */
455 idx = mlx5_dev_idx(&pci_dev->addr);
457 ERROR("this driver cannot support any more adapters");
460 DEBUG("using driver device index %d", idx);
462 /* Save PCI address. */
463 mlx5_dev[idx].pci_addr = pci_dev->addr;
464 list = ibv_get_device_list(&i);
468 ERROR("cannot list devices, is ib_uverbs loaded?");
473 * For each listed device, check related sysfs entry against
474 * the provided PCI ID.
477 struct rte_pci_addr pci_addr;
480 DEBUG("checking device \"%s\"", list[i]->name);
481 if (mlx5_ibv_device_to_pci_addr(list[i], &pci_addr))
483 if ((pci_dev->addr.domain != pci_addr.domain) ||
484 (pci_dev->addr.bus != pci_addr.bus) ||
485 (pci_dev->addr.devid != pci_addr.devid) ||
486 (pci_dev->addr.function != pci_addr.function))
488 sriov = ((pci_dev->id.device_id ==
489 PCI_DEVICE_ID_MELLANOX_CONNECTX4VF) ||
490 (pci_dev->id.device_id ==
491 PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF) ||
492 (pci_dev->id.device_id ==
493 PCI_DEVICE_ID_MELLANOX_CONNECTX5VF) ||
494 (pci_dev->id.device_id ==
495 PCI_DEVICE_ID_MELLANOX_CONNECTX5EXVF));
497 * Multi-packet send is supported by ConnectX-4 Lx PF as well
498 * as all ConnectX-5 devices.
500 switch (pci_dev->id.device_id) {
501 case PCI_DEVICE_ID_MELLANOX_CONNECTX4:
503 mps = MLX5_MPW_DISABLED;
505 case PCI_DEVICE_ID_MELLANOX_CONNECTX4LX:
509 case PCI_DEVICE_ID_MELLANOX_CONNECTX5:
510 case PCI_DEVICE_ID_MELLANOX_CONNECTX5VF:
511 case PCI_DEVICE_ID_MELLANOX_CONNECTX5EX:
512 case PCI_DEVICE_ID_MELLANOX_CONNECTX5EXVF:
514 mps = MLX5_MPW_ENHANCED;
517 mps = MLX5_MPW_DISABLED;
519 INFO("PCI information matches, using device \"%s\""
520 " (SR-IOV: %s, %sMPS: %s)",
522 sriov ? "true" : "false",
523 mps == MLX5_MPW_ENHANCED ? "Enhanced " : "",
524 mps != MLX5_MPW_DISABLED ? "true" : "false");
525 attr_ctx = ibv_open_device(list[i]);
529 if (attr_ctx == NULL) {
530 ibv_free_device_list(list);
533 ERROR("cannot access device, is mlx5_ib loaded?");
536 ERROR("cannot use device, are drivers up to date?");
544 DEBUG("device opened");
545 if (ibv_query_device(attr_ctx, &device_attr))
547 INFO("%u port(s) detected", device_attr.phys_port_cnt);
549 for (i = 0; i < device_attr.phys_port_cnt; i++) {
550 uint32_t port = i + 1; /* ports are indexed from one */
551 uint32_t test = (1 << i);
552 struct ibv_context *ctx = NULL;
553 struct ibv_port_attr port_attr;
554 struct ibv_pd *pd = NULL;
555 struct priv *priv = NULL;
556 struct rte_eth_dev *eth_dev;
557 struct ibv_exp_device_attr exp_device_attr;
558 struct ether_addr mac;
559 uint16_t num_vfs = 0;
560 struct mlx5_args args = {
561 .cqe_comp = MLX5_ARG_UNSET,
562 .txq_inline = MLX5_ARG_UNSET,
563 .txqs_inline = MLX5_ARG_UNSET,
564 .mps = MLX5_ARG_UNSET,
565 .mpw_hdr_dseg = MLX5_ARG_UNSET,
566 .inline_max_packet_sz = MLX5_ARG_UNSET,
567 .tso = MLX5_ARG_UNSET,
568 .tx_vec_en = MLX5_ARG_UNSET,
569 .rx_vec_en = MLX5_ARG_UNSET,
572 exp_device_attr.comp_mask =
573 IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS |
574 IBV_EXP_DEVICE_ATTR_RX_HASH |
575 IBV_EXP_DEVICE_ATTR_VLAN_OFFLOADS |
576 IBV_EXP_DEVICE_ATTR_RX_PAD_END_ALIGN |
577 IBV_EXP_DEVICE_ATTR_TSO_CAPS |
580 DEBUG("using port %u (%08" PRIx32 ")", port, test);
582 ctx = ibv_open_device(ibv_dev);
588 /* Check port status. */
589 err = ibv_query_port(ctx, port, &port_attr);
591 ERROR("port query failed: %s", strerror(err));
595 if (port_attr.link_layer != IBV_LINK_LAYER_ETHERNET) {
596 ERROR("port %d is not configured in Ethernet mode",
602 if (port_attr.state != IBV_PORT_ACTIVE)
603 DEBUG("port %d is not active: \"%s\" (%d)",
604 port, ibv_port_state_str(port_attr.state),
607 /* Allocate protection domain. */
608 pd = ibv_alloc_pd(ctx);
610 ERROR("PD allocation failure");
615 mlx5_dev[idx].ports |= test;
617 /* from rte_ethdev.c */
618 priv = rte_zmalloc("ethdev private structure",
620 RTE_CACHE_LINE_SIZE);
622 ERROR("priv allocation failure");
628 priv->device_attr = device_attr;
631 priv->mtu = ETHER_MTU;
632 priv->mps = mps; /* Enable MPW by default if supported. */
633 priv->cqe_comp = 1; /* Enable compression by default. */
634 priv->tunnel_en = tunnel_en;
635 /* Enable vector by default if supported. */
638 err = mlx5_args(&args, pci_dev->device.devargs);
640 ERROR("failed to process device arguments: %s",
644 mlx5_args_assign(priv, &args);
645 if (ibv_exp_query_device(ctx, &exp_device_attr)) {
646 ERROR("ibv_exp_query_device() failed");
652 ((exp_device_attr.exp_device_cap_flags &
653 IBV_EXP_DEVICE_RX_CSUM_TCP_UDP_PKT) &&
654 (exp_device_attr.exp_device_cap_flags &
655 IBV_EXP_DEVICE_RX_CSUM_IP_PKT));
656 DEBUG("checksum offloading is %ssupported",
657 (priv->hw_csum ? "" : "not "));
659 priv->hw_csum_l2tun = !!(exp_device_attr.exp_device_cap_flags &
660 IBV_EXP_DEVICE_VXLAN_SUPPORT);
661 DEBUG("L2 tunnel checksum offloads are %ssupported",
662 (priv->hw_csum_l2tun ? "" : "not "));
664 priv->ind_table_max_size = exp_device_attr.rx_hash_caps.max_rwq_indirection_table_size;
665 /* Remove this check once DPDK supports larger/variable
666 * indirection tables. */
667 if (priv->ind_table_max_size >
668 (unsigned int)ETH_RSS_RETA_SIZE_512)
669 priv->ind_table_max_size = ETH_RSS_RETA_SIZE_512;
670 DEBUG("maximum RX indirection table size is %u",
671 priv->ind_table_max_size);
672 priv->hw_vlan_strip = !!(exp_device_attr.wq_vlan_offloads_cap &
673 IBV_EXP_RECEIVE_WQ_CVLAN_STRIP);
674 DEBUG("VLAN stripping is %ssupported",
675 (priv->hw_vlan_strip ? "" : "not "));
677 priv->hw_fcs_strip = !!(exp_device_attr.exp_device_cap_flags &
678 IBV_EXP_DEVICE_SCATTER_FCS);
679 DEBUG("FCS stripping configuration is %ssupported",
680 (priv->hw_fcs_strip ? "" : "not "));
682 priv->hw_padding = !!exp_device_attr.rx_pad_end_addr_align;
683 DEBUG("hardware RX end alignment padding is %ssupported",
684 (priv->hw_padding ? "" : "not "));
686 priv_get_num_vfs(priv, &num_vfs);
687 priv->sriov = (num_vfs || sriov);
688 priv->tso = ((priv->tso) &&
689 (exp_device_attr.tso_caps.max_tso > 0) &&
690 (exp_device_attr.tso_caps.supported_qpts &
691 (1 << IBV_QPT_RAW_ETH)));
693 priv->max_tso_payload_sz =
694 exp_device_attr.tso_caps.max_tso;
695 if (priv->mps && !mps) {
696 ERROR("multi-packet send not supported on this device"
697 " (" MLX5_TXQ_MPW_EN ")");
700 } else if (priv->mps && priv->tso) {
701 WARN("multi-packet send not supported in conjunction "
702 "with TSO. MPS disabled");
706 priv->mps == MLX5_MPW_ENHANCED ? "Enhanced " : "",
707 priv->mps != MLX5_MPW_DISABLED ? "enabled" : "disabled");
708 /* Set default values for Enhanced MPW, a.k.a MPWv2. */
709 if (priv->mps == MLX5_MPW_ENHANCED) {
710 if (args.txqs_inline == MLX5_ARG_UNSET)
711 priv->txqs_inline = MLX5_EMPW_MIN_TXQS;
712 if (args.inline_max_packet_sz == MLX5_ARG_UNSET)
713 priv->inline_max_packet_sz =
714 MLX5_EMPW_MAX_INLINE_LEN;
715 if (args.txq_inline == MLX5_ARG_UNSET)
716 priv->txq_inline = MLX5_WQE_SIZE_MAX -
719 /* Allocate and register default RSS hash keys. */
720 priv->rss_conf = rte_calloc(__func__, hash_rxq_init_n,
721 sizeof((*priv->rss_conf)[0]), 0);
722 if (priv->rss_conf == NULL) {
726 err = rss_hash_rss_conf_new_key(priv,
727 rss_hash_default_key,
728 rss_hash_default_key_len,
732 /* Configure the first MAC address by default. */
733 if (priv_get_mac(priv, &mac.addr_bytes)) {
734 ERROR("cannot get MAC address, is mlx5_en loaded?"
735 " (errno: %s)", strerror(errno));
739 INFO("port %u MAC address is %02x:%02x:%02x:%02x:%02x:%02x",
741 mac.addr_bytes[0], mac.addr_bytes[1],
742 mac.addr_bytes[2], mac.addr_bytes[3],
743 mac.addr_bytes[4], mac.addr_bytes[5]);
744 /* Register MAC address. */
745 claim_zero(priv_mac_addr_add(priv, 0,
746 (const uint8_t (*)[ETHER_ADDR_LEN])
748 /* Initialize FD filters list. */
749 err = fdir_init_filters_list(priv);
754 char ifname[IF_NAMESIZE];
756 if (priv_get_ifname(priv, &ifname) == 0)
757 DEBUG("port %u ifname is \"%s\"",
760 DEBUG("port %u ifname is unknown", priv->port);
763 /* Get actual MTU if possible. */
764 priv_get_mtu(priv, &priv->mtu);
765 DEBUG("port %u MTU is %u", priv->port, priv->mtu);
767 /* from rte_ethdev.c */
769 char name[RTE_ETH_NAME_MAX_LEN];
771 snprintf(name, sizeof(name), "%s port %u",
772 ibv_get_device_name(ibv_dev), port);
773 eth_dev = rte_eth_dev_allocate(name);
775 if (eth_dev == NULL) {
776 ERROR("can not allocate rte ethdev");
780 eth_dev->data->dev_private = priv;
781 eth_dev->data->mac_addrs = priv->mac;
782 eth_dev->device = &pci_dev->device;
783 rte_eth_copy_pci_info(eth_dev, pci_dev);
784 eth_dev->data->dev_flags |= RTE_ETH_DEV_DETACHABLE;
785 eth_dev->device->driver = &mlx5_driver.driver;
787 eth_dev->dev_ops = &mlx5_dev_ops;
788 TAILQ_INIT(&priv->flows);
790 /* Bring Ethernet device up. */
791 DEBUG("forcing Ethernet interface up");
792 priv_set_flags(priv, ~IFF_UP, IFF_UP);
793 mlx5_link_update(priv->dev, 1);
798 rte_free(priv->rss_conf);
802 claim_zero(ibv_dealloc_pd(pd));
804 claim_zero(ibv_close_device(ctx));
809 * XXX if something went wrong in the loop above, there is a resource
810 * leak (ctx, pd, priv, dpdk ethdev) but we can do nothing about it as
811 * long as the dpdk does not provide a way to deallocate a ethdev and a
812 * way to enumerate the registered ethdevs to free the previous ones.
815 /* no port found, complain */
816 if (!mlx5_dev[idx].ports) {
823 claim_zero(ibv_close_device(attr_ctx));
825 ibv_free_device_list(list);
830 static const struct rte_pci_id mlx5_pci_id_map[] = {
832 RTE_PCI_DEVICE(PCI_VENDOR_ID_MELLANOX,
833 PCI_DEVICE_ID_MELLANOX_CONNECTX4)
836 RTE_PCI_DEVICE(PCI_VENDOR_ID_MELLANOX,
837 PCI_DEVICE_ID_MELLANOX_CONNECTX4VF)
840 RTE_PCI_DEVICE(PCI_VENDOR_ID_MELLANOX,
841 PCI_DEVICE_ID_MELLANOX_CONNECTX4LX)
844 RTE_PCI_DEVICE(PCI_VENDOR_ID_MELLANOX,
845 PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF)
848 RTE_PCI_DEVICE(PCI_VENDOR_ID_MELLANOX,
849 PCI_DEVICE_ID_MELLANOX_CONNECTX5)
852 RTE_PCI_DEVICE(PCI_VENDOR_ID_MELLANOX,
853 PCI_DEVICE_ID_MELLANOX_CONNECTX5VF)
856 RTE_PCI_DEVICE(PCI_VENDOR_ID_MELLANOX,
857 PCI_DEVICE_ID_MELLANOX_CONNECTX5EX)
860 RTE_PCI_DEVICE(PCI_VENDOR_ID_MELLANOX,
861 PCI_DEVICE_ID_MELLANOX_CONNECTX5EXVF)
868 static struct rte_pci_driver mlx5_driver = {
870 .name = MLX5_DRIVER_NAME
872 .id_table = mlx5_pci_id_map,
873 .probe = mlx5_pci_probe,
874 .drv_flags = RTE_PCI_DRV_INTR_LSC | RTE_PCI_DRV_INTR_RMV,
878 * Driver initialization routine.
880 RTE_INIT(rte_mlx5_pmd_init);
882 rte_mlx5_pmd_init(void)
884 /* Build the static table for ptype conversion. */
885 mlx5_set_ptype_table();
887 * RDMAV_HUGEPAGES_SAFE tells ibv_fork_init() we intend to use
888 * huge pages. Calling ibv_fork_init() during init allows
889 * applications to use fork() safely for purposes other than
890 * using this PMD, which is not supported in forked processes.
892 setenv("RDMAV_HUGEPAGES_SAFE", "1", 1);
893 /* Don't map UAR to WC if BlueFlame is not used.*/
894 setenv("MLX5_SHUT_UP_BF", "1", 1);
896 rte_pci_register(&mlx5_driver);
899 RTE_PMD_EXPORT_NAME(net_mlx5, __COUNTER__);
900 RTE_PMD_REGISTER_PCI_TABLE(net_mlx5, mlx5_pci_id_map);
901 RTE_PMD_REGISTER_KMOD_DEP(net_mlx5, "* ib_uverbs & mlx5_core & mlx5_ib");