1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2017 6WIND S.A.
3 * Copyright 2017 Mellanox
8 * Miscellaneous control operations for mlx4 driver.
14 #include <linux/ethtool.h>
15 #include <linux/sockios.h>
17 #include <netinet/ip.h>
23 #include <sys/ioctl.h>
24 #include <sys/socket.h>
27 /* Verbs headers do not support -pedantic. */
29 #pragma GCC diagnostic ignored "-Wpedantic"
31 #include <infiniband/verbs.h>
33 #pragma GCC diagnostic error "-Wpedantic"
36 #include <rte_bus_pci.h>
37 #include <rte_errno.h>
38 #include <rte_ethdev_driver.h>
39 #include <rte_ether.h>
44 #include "mlx4_flow.h"
45 #include "mlx4_glue.h"
46 #include "mlx4_rxtx.h"
47 #include "mlx4_utils.h"
50 * Get interface name from private structure.
53 * Pointer to private structure.
55 * Interface name output buffer.
58 * 0 on success, negative errno value otherwise and rte_errno is set.
61 mlx4_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE])
65 unsigned int dev_type = 0;
66 unsigned int dev_port_prev = ~0u;
67 char match[IF_NAMESIZE] = "";
70 MKSTR(path, "%s/device/net", priv->ctx->device->ibdev_path);
78 while ((dent = readdir(dir)) != NULL) {
79 char *name = dent->d_name;
81 unsigned int dev_port;
84 if ((name[0] == '.') &&
86 ((name[1] == '.') && (name[2] == '\0'))))
89 MKSTR(path, "%s/device/net/%s/%s",
90 priv->ctx->device->ibdev_path, name,
91 (dev_type ? "dev_id" : "dev_port"));
93 file = fopen(path, "rb");
98 * Switch to dev_id when dev_port does not exist as
99 * is the case with Linux kernel versions < 3.15.
110 r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port);
115 * Switch to dev_id when dev_port returns the same value for
116 * all ports. May happen when using a MOFED release older than
117 * 3.0 with a Linux kernel >= 3.15.
119 if (dev_port == dev_port_prev)
121 dev_port_prev = dev_port;
122 if (dev_port == (priv->port - 1u))
123 snprintf(match, sizeof(match), "%s", name);
126 if (match[0] == '\0') {
130 strncpy(*ifname, match, sizeof(*ifname));
135 * Perform ifreq ioctl() on associated Ethernet device.
138 * Pointer to private structure.
140 * Request number to pass to ioctl().
142 * Interface request structure output buffer.
145 * 0 on success, negative errno value otherwise and rte_errno is set.
148 mlx4_ifreq(const struct priv *priv, int req, struct ifreq *ifr)
150 int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
157 ret = mlx4_get_ifname(priv, &ifr->ifr_name);
158 if (!ret && ioctl(sock, req, ifr) == -1) {
167 * Get MAC address by querying netdevice.
170 * Pointer to private structure.
172 * MAC address output buffer.
175 * 0 on success, negative errno value otherwise and rte_errno is set.
178 mlx4_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN])
180 struct ifreq request;
181 int ret = mlx4_ifreq(priv, SIOCGIFHWADDR, &request);
185 memcpy(mac, request.ifr_hwaddr.sa_data, ETHER_ADDR_LEN);
193 * Pointer to private structure.
195 * MTU value output buffer.
198 * 0 on success, negative errno value otherwise and rte_errno is set.
201 mlx4_mtu_get(struct priv *priv, uint16_t *mtu)
203 struct ifreq request;
204 int ret = mlx4_ifreq(priv, SIOCGIFMTU, &request);
208 *mtu = request.ifr_mtu;
213 * DPDK callback to change the MTU.
216 * Pointer to Ethernet device structure.
221 * 0 on success, negative errno value otherwise and rte_errno is set.
224 mlx4_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
226 struct priv *priv = dev->data->dev_private;
227 struct ifreq request = { .ifr_mtu = mtu, };
228 int ret = mlx4_ifreq(priv, SIOCSIFMTU, &request);
240 * Pointer to private structure.
242 * Bitmask for flags that must remain untouched.
244 * Bitmask for flags to modify.
247 * 0 on success, negative errno value otherwise and rte_errno is set.
250 mlx4_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
252 struct ifreq request;
253 int ret = mlx4_ifreq(priv, SIOCGIFFLAGS, &request);
257 request.ifr_flags &= keep;
258 request.ifr_flags |= flags & ~keep;
259 return mlx4_ifreq(priv, SIOCSIFFLAGS, &request);
263 * Change the link state (UP / DOWN).
266 * Pointer to Ethernet device private data.
268 * Nonzero for link up, otherwise link down.
271 * 0 on success, negative errno value otherwise and rte_errno is set.
274 mlx4_dev_set_link(struct priv *priv, int up)
279 err = mlx4_set_flags(priv, ~IFF_UP, IFF_UP);
283 err = mlx4_set_flags(priv, ~IFF_UP, ~IFF_UP);
291 * DPDK callback to bring the link DOWN.
294 * Pointer to Ethernet device structure.
297 * 0 on success, negative errno value otherwise and rte_errno is set.
300 mlx4_dev_set_link_down(struct rte_eth_dev *dev)
302 struct priv *priv = dev->data->dev_private;
304 return mlx4_dev_set_link(priv, 0);
308 * DPDK callback to bring the link UP.
311 * Pointer to Ethernet device structure.
314 * 0 on success, negative errno value otherwise and rte_errno is set.
317 mlx4_dev_set_link_up(struct rte_eth_dev *dev)
319 struct priv *priv = dev->data->dev_private;
321 return mlx4_dev_set_link(priv, 1);
325 * Supported Rx mode toggles.
327 * Even and odd values respectively stand for off and on.
330 RXMODE_TOGGLE_PROMISC_OFF,
331 RXMODE_TOGGLE_PROMISC_ON,
332 RXMODE_TOGGLE_ALLMULTI_OFF,
333 RXMODE_TOGGLE_ALLMULTI_ON,
337 * Helper function to toggle promiscuous and all multicast modes.
340 * Pointer to Ethernet device structure.
345 mlx4_rxmode_toggle(struct rte_eth_dev *dev, enum rxmode_toggle toggle)
347 struct priv *priv = dev->data->dev_private;
349 struct rte_flow_error error;
352 case RXMODE_TOGGLE_PROMISC_OFF:
353 case RXMODE_TOGGLE_PROMISC_ON:
354 mode = "promiscuous";
355 dev->data->promiscuous = toggle & 1;
357 case RXMODE_TOGGLE_ALLMULTI_OFF:
358 case RXMODE_TOGGLE_ALLMULTI_ON:
359 mode = "all multicast";
360 dev->data->all_multicast = toggle & 1;
363 if (!mlx4_flow_sync(priv, &error))
365 ERROR("cannot toggle %s mode (code %d, \"%s\"),"
366 " flow error type %d, cause %p, message: %s",
367 mode, rte_errno, strerror(rte_errno), error.type, error.cause,
368 error.message ? error.message : "(unspecified)");
372 * DPDK callback to enable promiscuous mode.
375 * Pointer to Ethernet device structure.
378 mlx4_promiscuous_enable(struct rte_eth_dev *dev)
380 mlx4_rxmode_toggle(dev, RXMODE_TOGGLE_PROMISC_ON);
384 * DPDK callback to disable promiscuous mode.
387 * Pointer to Ethernet device structure.
390 mlx4_promiscuous_disable(struct rte_eth_dev *dev)
392 mlx4_rxmode_toggle(dev, RXMODE_TOGGLE_PROMISC_OFF);
396 * DPDK callback to enable all multicast mode.
399 * Pointer to Ethernet device structure.
402 mlx4_allmulticast_enable(struct rte_eth_dev *dev)
404 mlx4_rxmode_toggle(dev, RXMODE_TOGGLE_ALLMULTI_ON);
408 * DPDK callback to disable all multicast mode.
411 * Pointer to Ethernet device structure.
414 mlx4_allmulticast_disable(struct rte_eth_dev *dev)
416 mlx4_rxmode_toggle(dev, RXMODE_TOGGLE_ALLMULTI_OFF);
420 * DPDK callback to remove a MAC address.
423 * Pointer to Ethernet device structure.
428 mlx4_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
430 struct priv *priv = dev->data->dev_private;
431 struct rte_flow_error error;
433 if (index >= RTE_DIM(priv->mac)) {
437 memset(&priv->mac[index], 0, sizeof(priv->mac[index]));
438 if (!mlx4_flow_sync(priv, &error))
440 ERROR("failed to synchronize flow rules after removing MAC address"
441 " at index %d (code %d, \"%s\"),"
442 " flow error type %d, cause %p, message: %s",
443 index, rte_errno, strerror(rte_errno), error.type, error.cause,
444 error.message ? error.message : "(unspecified)");
448 * DPDK callback to add a MAC address.
451 * Pointer to Ethernet device structure.
453 * MAC address to register.
457 * VMDq pool index to associate address with (ignored).
460 * 0 on success, negative errno value otherwise and rte_errno is set.
463 mlx4_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
464 uint32_t index, uint32_t vmdq)
466 struct priv *priv = dev->data->dev_private;
467 struct rte_flow_error error;
471 if (index >= RTE_DIM(priv->mac)) {
475 memcpy(&priv->mac[index], mac_addr, sizeof(priv->mac[index]));
476 ret = mlx4_flow_sync(priv, &error);
479 ERROR("failed to synchronize flow rules after adding MAC address"
480 " at index %d (code %d, \"%s\"),"
481 " flow error type %d, cause %p, message: %s",
482 index, rte_errno, strerror(rte_errno), error.type, error.cause,
483 error.message ? error.message : "(unspecified)");
488 * DPDK callback to configure a VLAN filter.
491 * Pointer to Ethernet device structure.
498 * 0 on success, negative errno value otherwise and rte_errno is set.
501 mlx4_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
503 struct priv *priv = dev->data->dev_private;
504 struct rte_flow_error error;
505 unsigned int vidx = vlan_id / 64;
506 unsigned int vbit = vlan_id % 64;
510 if (vidx >= RTE_DIM(dev->data->vlan_filter_conf.ids)) {
514 v = &dev->data->vlan_filter_conf.ids[vidx];
515 *v &= ~(UINT64_C(1) << vbit);
516 *v |= (uint64_t)!!on << vbit;
517 ret = mlx4_flow_sync(priv, &error);
520 ERROR("failed to synchronize flow rules after %s VLAN filter on ID %u"
521 " (code %d, \"%s\"), "
522 " flow error type %d, cause %p, message: %s",
523 on ? "enabling" : "disabling", vlan_id,
524 rte_errno, strerror(rte_errno), error.type, error.cause,
525 error.message ? error.message : "(unspecified)");
530 * DPDK callback to set the primary MAC address.
533 * Pointer to Ethernet device structure.
535 * MAC address to register.
538 mlx4_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
540 mlx4_mac_addr_add(dev, mac_addr, 0, 0);
544 * DPDK callback to get information about the device.
547 * Pointer to Ethernet device structure.
549 * Info structure output buffer.
552 mlx4_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
554 struct priv *priv = dev->data->dev_private;
556 char ifname[IF_NAMESIZE];
558 info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
559 /* FIXME: we should ask the device for these values. */
560 info->min_rx_bufsize = 32;
561 info->max_rx_pktlen = 65536;
563 * Since we need one CQ per QP, the limit is the minimum number
564 * between the two values.
566 max = ((priv->device_attr.max_cq > priv->device_attr.max_qp) ?
567 priv->device_attr.max_qp : priv->device_attr.max_cq);
568 /* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
571 info->max_rx_queues = max;
572 info->max_tx_queues = max;
573 info->max_mac_addrs = RTE_DIM(priv->mac);
574 info->tx_offload_capa = mlx4_get_tx_port_offloads(priv);
575 info->rx_queue_offload_capa = mlx4_get_rx_queue_offloads(priv);
576 info->rx_offload_capa = (mlx4_get_rx_port_offloads(priv) |
577 info->rx_queue_offload_capa);
578 if (mlx4_get_ifname(priv, &ifname) == 0)
579 info->if_index = if_nametoindex(ifname);
580 info->hash_key_size = MLX4_RSS_HASH_KEY_SIZE;
590 * DPDK callback to get device statistics.
593 * Pointer to Ethernet device structure.
595 * Stats structure output buffer.
598 mlx4_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
600 struct rte_eth_stats tmp;
604 memset(&tmp, 0, sizeof(tmp));
605 /* Add software counters. */
606 for (i = 0; i != dev->data->nb_rx_queues; ++i) {
607 struct rxq *rxq = dev->data->rx_queues[i];
611 idx = rxq->stats.idx;
612 if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
613 tmp.q_ipackets[idx] += rxq->stats.ipackets;
614 tmp.q_ibytes[idx] += rxq->stats.ibytes;
615 tmp.q_errors[idx] += (rxq->stats.idropped +
616 rxq->stats.rx_nombuf);
618 tmp.ipackets += rxq->stats.ipackets;
619 tmp.ibytes += rxq->stats.ibytes;
620 tmp.ierrors += rxq->stats.idropped;
621 tmp.rx_nombuf += rxq->stats.rx_nombuf;
623 for (i = 0; i != dev->data->nb_tx_queues; ++i) {
624 struct txq *txq = dev->data->tx_queues[i];
628 idx = txq->stats.idx;
629 if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
630 tmp.q_opackets[idx] += txq->stats.opackets;
631 tmp.q_obytes[idx] += txq->stats.obytes;
632 tmp.q_errors[idx] += txq->stats.odropped;
634 tmp.opackets += txq->stats.opackets;
635 tmp.obytes += txq->stats.obytes;
636 tmp.oerrors += txq->stats.odropped;
643 * DPDK callback to clear device statistics.
646 * Pointer to Ethernet device structure.
649 mlx4_stats_reset(struct rte_eth_dev *dev)
653 for (i = 0; i != dev->data->nb_rx_queues; ++i) {
654 struct rxq *rxq = dev->data->rx_queues[i];
657 rxq->stats = (struct mlx4_rxq_stats){
658 .idx = rxq->stats.idx,
661 for (i = 0; i != dev->data->nb_tx_queues; ++i) {
662 struct txq *txq = dev->data->tx_queues[i];
665 txq->stats = (struct mlx4_txq_stats){
666 .idx = txq->stats.idx,
672 * DPDK callback to retrieve physical link information.
675 * Pointer to Ethernet device structure.
676 * @param wait_to_complete
677 * Wait for request completion (ignored).
680 * 0 on success, negative errno value otherwise and rte_errno is set.
683 mlx4_link_update(struct rte_eth_dev *dev, int wait_to_complete)
685 const struct priv *priv = dev->data->dev_private;
686 struct ethtool_cmd edata = {
690 struct rte_eth_link dev_link;
697 (void)wait_to_complete;
698 if (mlx4_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
699 WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(rte_errno));
702 memset(&dev_link, 0, sizeof(dev_link));
703 dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
704 (ifr.ifr_flags & IFF_RUNNING));
705 ifr.ifr_data = (void *)&edata;
706 if (mlx4_ifreq(priv, SIOCETHTOOL, &ifr)) {
707 WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s",
708 strerror(rte_errno));
711 link_speed = ethtool_cmd_speed(&edata);
712 if (link_speed == -1)
713 dev_link.link_speed = 0;
715 dev_link.link_speed = link_speed;
716 dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
717 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
718 dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
719 ETH_LINK_SPEED_FIXED);
720 dev->data->dev_link = dev_link;
725 * DPDK callback to get flow control status.
728 * Pointer to Ethernet device structure.
729 * @param[out] fc_conf
730 * Flow control output buffer.
733 * 0 on success, negative errno value otherwise and rte_errno is set.
736 mlx4_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
738 struct priv *priv = dev->data->dev_private;
740 struct ethtool_pauseparam ethpause = {
741 .cmd = ETHTOOL_GPAUSEPARAM,
745 ifr.ifr_data = (void *)ðpause;
746 if (mlx4_ifreq(priv, SIOCETHTOOL, &ifr)) {
748 WARN("ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM)"
750 strerror(rte_errno));
753 fc_conf->autoneg = ethpause.autoneg;
754 if (ethpause.rx_pause && ethpause.tx_pause)
755 fc_conf->mode = RTE_FC_FULL;
756 else if (ethpause.rx_pause)
757 fc_conf->mode = RTE_FC_RX_PAUSE;
758 else if (ethpause.tx_pause)
759 fc_conf->mode = RTE_FC_TX_PAUSE;
761 fc_conf->mode = RTE_FC_NONE;
769 * DPDK callback to modify flow control parameters.
772 * Pointer to Ethernet device structure.
774 * Flow control parameters.
777 * 0 on success, negative errno value otherwise and rte_errno is set.
780 mlx4_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
782 struct priv *priv = dev->data->dev_private;
784 struct ethtool_pauseparam ethpause = {
785 .cmd = ETHTOOL_SPAUSEPARAM,
789 ifr.ifr_data = (void *)ðpause;
790 ethpause.autoneg = fc_conf->autoneg;
791 if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
792 (fc_conf->mode & RTE_FC_RX_PAUSE))
793 ethpause.rx_pause = 1;
795 ethpause.rx_pause = 0;
796 if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
797 (fc_conf->mode & RTE_FC_TX_PAUSE))
798 ethpause.tx_pause = 1;
800 ethpause.tx_pause = 0;
801 if (mlx4_ifreq(priv, SIOCETHTOOL, &ifr)) {
803 WARN("ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
805 strerror(rte_errno));
815 * DPDK callback to retrieve the received packet types that are recognized
819 * Pointer to Ethernet device structure.
822 * Pointer to an array of recognized packet types if in Rx burst mode,
826 mlx4_dev_supported_ptypes_get(struct rte_eth_dev *dev)
828 static const uint32_t ptypes[] = {
829 /* refers to rxq_cq_to_pkt_type() */
831 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
832 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
838 static const uint32_t ptypes_l2tun[] = {
839 /* refers to rxq_cq_to_pkt_type() */
841 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
842 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
846 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
847 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
850 struct priv *priv = dev->data->dev_private;
852 if (dev->rx_pkt_burst == mlx4_rx_burst) {
853 if (priv->hw_csum_l2tun)
862 * Check if mlx4 device was removed.
865 * Pointer to Ethernet device structure.
868 * 1 when device is removed, otherwise 0.
871 mlx4_is_removed(struct rte_eth_dev *dev)
873 struct ibv_device_attr device_attr;
874 struct priv *priv = dev->data->dev_private;
876 if (mlx4_glue->query_device(priv->ctx, &device_attr) == EIO)