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.
43 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
45 #pragma GCC diagnostic ignored "-pedantic"
47 #include <infiniband/verbs.h>
49 #pragma GCC diagnostic error "-pedantic"
52 /* DPDK headers don't like -pedantic. */
54 #pragma GCC diagnostic ignored "-pedantic"
56 #include <rte_malloc.h>
57 #include <rte_ethdev.h>
59 #include <rte_common.h>
61 #pragma GCC diagnostic error "-pedantic"
65 #include "mlx5_utils.h"
66 #include "mlx5_rxtx.h"
67 #include "mlx5_autoconf.h"
70 * DPDK callback to close the device.
72 * Destroy all queues and objects, free memory.
75 * Pointer to Ethernet device structure.
78 mlx5_dev_close(struct rte_eth_dev *dev)
80 struct priv *priv = dev->data->dev_private;
85 DEBUG("%p: closing device \"%s\"",
87 ((priv->ctx != NULL) ? priv->ctx->device->name : ""));
88 /* In case mlx5_dev_stop() has not been called. */
89 priv_allmulticast_disable(priv);
90 priv_promiscuous_disable(priv);
91 priv_mac_addrs_disable(priv);
92 priv_destroy_hash_rxqs(priv);
93 /* Prevent crashes when queues are still in use. */
94 dev->rx_pkt_burst = removed_rx_burst;
95 dev->tx_pkt_burst = removed_tx_burst;
96 if (priv->rxqs != NULL) {
97 /* XXX race condition if mlx5_rx_burst() is still running. */
99 for (i = 0; (i != priv->rxqs_n); ++i) {
100 tmp = (*priv->rxqs)[i];
103 (*priv->rxqs)[i] = NULL;
110 if (priv->txqs != NULL) {
111 /* XXX race condition if mlx5_tx_burst() is still running. */
113 for (i = 0; (i != priv->txqs_n); ++i) {
114 tmp = (*priv->txqs)[i];
117 (*priv->txqs)[i] = NULL;
124 if (priv->pd != NULL) {
125 assert(priv->ctx != NULL);
126 claim_zero(ibv_dealloc_pd(priv->pd));
127 claim_zero(ibv_close_device(priv->ctx));
129 assert(priv->ctx == NULL);
131 memset(priv, 0, sizeof(*priv));
134 static const struct eth_dev_ops mlx5_dev_ops = {
135 .dev_configure = mlx5_dev_configure,
136 .dev_start = mlx5_dev_start,
137 .dev_stop = mlx5_dev_stop,
138 .dev_close = mlx5_dev_close,
139 .promiscuous_enable = mlx5_promiscuous_enable,
140 .promiscuous_disable = mlx5_promiscuous_disable,
141 .allmulticast_enable = mlx5_allmulticast_enable,
142 .allmulticast_disable = mlx5_allmulticast_disable,
143 .link_update = mlx5_link_update,
144 .stats_get = mlx5_stats_get,
145 .stats_reset = mlx5_stats_reset,
146 .dev_infos_get = mlx5_dev_infos_get,
147 .vlan_filter_set = mlx5_vlan_filter_set,
148 .rx_queue_setup = mlx5_rx_queue_setup,
149 .tx_queue_setup = mlx5_tx_queue_setup,
150 .rx_queue_release = mlx5_rx_queue_release,
151 .tx_queue_release = mlx5_tx_queue_release,
152 .flow_ctrl_get = mlx5_dev_get_flow_ctrl,
153 .flow_ctrl_set = mlx5_dev_set_flow_ctrl,
154 .mac_addr_remove = mlx5_mac_addr_remove,
155 .mac_addr_add = mlx5_mac_addr_add,
156 .mtu_set = mlx5_dev_set_mtu,
160 struct rte_pci_addr pci_addr; /* associated PCI address */
161 uint32_t ports; /* physical ports bitfield. */
165 * Get device index in mlx5_dev[] from PCI bus address.
167 * @param[in] pci_addr
168 * PCI bus address to look for.
171 * mlx5_dev[] index on success, -1 on failure.
174 mlx5_dev_idx(struct rte_pci_addr *pci_addr)
179 assert(pci_addr != NULL);
180 for (i = 0; (i != RTE_DIM(mlx5_dev)); ++i) {
181 if ((mlx5_dev[i].pci_addr.domain == pci_addr->domain) &&
182 (mlx5_dev[i].pci_addr.bus == pci_addr->bus) &&
183 (mlx5_dev[i].pci_addr.devid == pci_addr->devid) &&
184 (mlx5_dev[i].pci_addr.function == pci_addr->function))
186 if ((mlx5_dev[i].ports == 0) && (ret == -1))
192 static struct eth_driver mlx5_driver;
195 * DPDK callback to register a PCI device.
197 * This function creates an Ethernet device for each port of a given
201 * PCI driver structure (mlx5_driver).
203 * PCI device information.
206 * 0 on success, negative errno value on failure.
209 mlx5_pci_devinit(struct rte_pci_driver *pci_drv, struct rte_pci_device *pci_dev)
211 struct ibv_device **list;
212 struct ibv_device *ibv_dev;
214 struct ibv_context *attr_ctx = NULL;
215 struct ibv_device_attr device_attr;
221 assert(pci_drv == &mlx5_driver.pci_drv);
222 /* Get mlx5_dev[] index. */
223 idx = mlx5_dev_idx(&pci_dev->addr);
225 ERROR("this driver cannot support any more adapters");
228 DEBUG("using driver device index %d", idx);
230 /* Save PCI address. */
231 mlx5_dev[idx].pci_addr = pci_dev->addr;
232 list = ibv_get_device_list(&i);
235 if (errno == ENOSYS) {
236 WARN("cannot list devices, is ib_uverbs loaded?");
243 * For each listed device, check related sysfs entry against
244 * the provided PCI ID.
247 struct rte_pci_addr pci_addr;
250 DEBUG("checking device \"%s\"", list[i]->name);
251 if (mlx5_ibv_device_to_pci_addr(list[i], &pci_addr))
253 if ((pci_dev->addr.domain != pci_addr.domain) ||
254 (pci_dev->addr.bus != pci_addr.bus) ||
255 (pci_dev->addr.devid != pci_addr.devid) ||
256 (pci_dev->addr.function != pci_addr.function))
258 vf = ((pci_dev->id.device_id ==
259 PCI_DEVICE_ID_MELLANOX_CONNECTX4VF) ||
260 (pci_dev->id.device_id ==
261 PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF));
262 INFO("PCI information matches, using device \"%s\" (VF: %s)",
263 list[i]->name, (vf ? "true" : "false"));
264 attr_ctx = ibv_open_device(list[i]);
268 if (attr_ctx == NULL) {
269 ibv_free_device_list(list);
272 WARN("cannot access device, is mlx5_ib loaded?");
275 WARN("cannot use device, are drivers up to date?");
283 DEBUG("device opened");
284 if (ibv_query_device(attr_ctx, &device_attr))
286 INFO("%u port(s) detected", device_attr.phys_port_cnt);
288 for (i = 0; i < device_attr.phys_port_cnt; i++) {
289 uint32_t port = i + 1; /* ports are indexed from one */
290 uint32_t test = (1 << i);
291 struct ibv_context *ctx = NULL;
292 struct ibv_port_attr port_attr;
293 struct ibv_pd *pd = NULL;
294 struct priv *priv = NULL;
295 struct rte_eth_dev *eth_dev;
296 #ifdef HAVE_EXP_QUERY_DEVICE
297 struct ibv_exp_device_attr exp_device_attr;
298 #endif /* HAVE_EXP_QUERY_DEVICE */
299 struct ether_addr mac;
301 #ifdef HAVE_EXP_QUERY_DEVICE
302 exp_device_attr.comp_mask = IBV_EXP_DEVICE_ATTR_EXP_CAP_FLAGS;
303 #endif /* HAVE_EXP_QUERY_DEVICE */
305 DEBUG("using port %u (%08" PRIx32 ")", port, test);
307 ctx = ibv_open_device(ibv_dev);
311 /* Check port status. */
312 err = ibv_query_port(ctx, port, &port_attr);
314 ERROR("port query failed: %s", strerror(err));
317 if (port_attr.state != IBV_PORT_ACTIVE)
318 DEBUG("port %d is not active: \"%s\" (%d)",
319 port, ibv_port_state_str(port_attr.state),
322 /* Allocate protection domain. */
323 pd = ibv_alloc_pd(ctx);
325 ERROR("PD allocation failure");
330 mlx5_dev[idx].ports |= test;
332 /* from rte_ethdev.c */
333 priv = rte_zmalloc("ethdev private structure",
335 RTE_CACHE_LINE_SIZE);
337 ERROR("priv allocation failure");
343 priv->device_attr = device_attr;
346 priv->mtu = ETHER_MTU;
347 #ifdef HAVE_EXP_QUERY_DEVICE
348 if (ibv_exp_query_device(ctx, &exp_device_attr)) {
349 ERROR("ibv_exp_query_device() failed");
354 ((exp_device_attr.exp_device_cap_flags &
355 IBV_EXP_DEVICE_RX_CSUM_TCP_UDP_PKT) &&
356 (exp_device_attr.exp_device_cap_flags &
357 IBV_EXP_DEVICE_RX_CSUM_IP_PKT));
358 DEBUG("checksum offloading is %ssupported",
359 (priv->hw_csum ? "" : "not "));
361 priv->hw_csum_l2tun = !!(exp_device_attr.exp_device_cap_flags &
362 IBV_EXP_DEVICE_VXLAN_SUPPORT);
363 DEBUG("L2 tunnel checksum offloads are %ssupported",
364 (priv->hw_csum_l2tun ? "" : "not "));
366 #endif /* HAVE_EXP_QUERY_DEVICE */
369 /* Configure the first MAC address by default. */
370 if (priv_get_mac(priv, &mac.addr_bytes)) {
371 ERROR("cannot get MAC address, is mlx5_en loaded?"
372 " (errno: %s)", strerror(errno));
375 INFO("port %u MAC address is %02x:%02x:%02x:%02x:%02x:%02x",
377 mac.addr_bytes[0], mac.addr_bytes[1],
378 mac.addr_bytes[2], mac.addr_bytes[3],
379 mac.addr_bytes[4], mac.addr_bytes[5]);
380 /* Register MAC and broadcast addresses. */
381 claim_zero(priv_mac_addr_add(priv, 0,
382 (const uint8_t (*)[ETHER_ADDR_LEN])
384 claim_zero(priv_mac_addr_add(priv, (RTE_DIM(priv->mac) - 1),
385 &(const uint8_t [ETHER_ADDR_LEN])
386 { "\xff\xff\xff\xff\xff\xff" }));
389 char ifname[IF_NAMESIZE];
391 if (priv_get_ifname(priv, &ifname) == 0)
392 DEBUG("port %u ifname is \"%s\"",
395 DEBUG("port %u ifname is unknown", priv->port);
398 /* Get actual MTU if possible. */
399 priv_get_mtu(priv, &priv->mtu);
400 DEBUG("port %u MTU is %u", priv->port, priv->mtu);
402 /* from rte_ethdev.c */
404 char name[RTE_ETH_NAME_MAX_LEN];
406 snprintf(name, sizeof(name), "%s port %u",
407 ibv_get_device_name(ibv_dev), port);
408 eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_PCI);
410 if (eth_dev == NULL) {
411 ERROR("can not allocate rte ethdev");
416 eth_dev->data->dev_private = priv;
417 eth_dev->pci_dev = pci_dev;
418 eth_dev->driver = &mlx5_driver;
419 eth_dev->data->rx_mbuf_alloc_failed = 0;
420 eth_dev->data->mtu = ETHER_MTU;
423 eth_dev->dev_ops = &mlx5_dev_ops;
424 eth_dev->data->mac_addrs = priv->mac;
426 /* Bring Ethernet device up. */
427 DEBUG("forcing Ethernet interface up");
428 priv_set_flags(priv, ~IFF_UP, IFF_UP);
434 claim_zero(ibv_dealloc_pd(pd));
436 claim_zero(ibv_close_device(ctx));
441 * XXX if something went wrong in the loop above, there is a resource
442 * leak (ctx, pd, priv, dpdk ethdev) but we can do nothing about it as
443 * long as the dpdk does not provide a way to deallocate a ethdev and a
444 * way to enumerate the registered ethdevs to free the previous ones.
447 /* no port found, complain */
448 if (!mlx5_dev[idx].ports) {
455 claim_zero(ibv_close_device(attr_ctx));
457 ibv_free_device_list(list);
462 static const struct rte_pci_id mlx5_pci_id_map[] = {
464 .vendor_id = PCI_VENDOR_ID_MELLANOX,
465 .device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4,
466 .subsystem_vendor_id = PCI_ANY_ID,
467 .subsystem_device_id = PCI_ANY_ID
470 .vendor_id = PCI_VENDOR_ID_MELLANOX,
471 .device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4VF,
472 .subsystem_vendor_id = PCI_ANY_ID,
473 .subsystem_device_id = PCI_ANY_ID
476 .vendor_id = PCI_VENDOR_ID_MELLANOX,
477 .device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4LX,
478 .subsystem_vendor_id = PCI_ANY_ID,
479 .subsystem_device_id = PCI_ANY_ID
482 .vendor_id = PCI_VENDOR_ID_MELLANOX,
483 .device_id = PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF,
484 .subsystem_vendor_id = PCI_ANY_ID,
485 .subsystem_device_id = PCI_ANY_ID
492 static struct eth_driver mlx5_driver = {
494 .name = MLX5_DRIVER_NAME,
495 .id_table = mlx5_pci_id_map,
496 .devinit = mlx5_pci_devinit,
498 .dev_private_size = sizeof(struct priv)
502 * Driver initialization routine.
505 rte_mlx5_pmd_init(const char *name, const char *args)
510 * RDMAV_HUGEPAGES_SAFE tells ibv_fork_init() we intend to use
511 * huge pages. Calling ibv_fork_init() during init allows
512 * applications to use fork() safely for purposes other than
513 * using this PMD, which is not supported in forked processes.
515 setenv("RDMAV_HUGEPAGES_SAFE", "1", 1);
517 rte_eal_pci_register(&mlx5_driver.pci_drv);
521 static struct rte_driver rte_mlx5_driver = {
523 .name = MLX5_DRIVER_NAME,
524 .init = rte_mlx5_pmd_init,
527 PMD_REGISTER_DRIVER(rte_mlx5_driver)