4 * Copyright 2017 6WIND S.A.
5 * Copyright 2017 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.
36 * Miscellaneous control operations for mlx4 driver.
42 #include <linux/ethtool.h>
43 #include <linux/sockios.h>
45 #include <netinet/ip.h>
51 #include <sys/ioctl.h>
52 #include <sys/socket.h>
55 /* Verbs headers do not support -pedantic. */
57 #pragma GCC diagnostic ignored "-Wpedantic"
59 #include <infiniband/verbs.h>
61 #pragma GCC diagnostic error "-Wpedantic"
64 #include <rte_errno.h>
65 #include <rte_ethdev.h>
66 #include <rte_ether.h>
71 #include "mlx4_flow.h"
72 #include "mlx4_rxtx.h"
73 #include "mlx4_utils.h"
76 * Get interface name from private structure.
79 * Pointer to private structure.
81 * Interface name output buffer.
84 * 0 on success, negative errno value otherwise and rte_errno is set.
87 mlx4_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE])
91 unsigned int dev_type = 0;
92 unsigned int dev_port_prev = ~0u;
93 char match[IF_NAMESIZE] = "";
96 MKSTR(path, "%s/device/net", priv->ctx->device->ibdev_path);
104 while ((dent = readdir(dir)) != NULL) {
105 char *name = dent->d_name;
107 unsigned int dev_port;
110 if ((name[0] == '.') &&
111 ((name[1] == '\0') ||
112 ((name[1] == '.') && (name[2] == '\0'))))
115 MKSTR(path, "%s/device/net/%s/%s",
116 priv->ctx->device->ibdev_path, name,
117 (dev_type ? "dev_id" : "dev_port"));
119 file = fopen(path, "rb");
124 * Switch to dev_id when dev_port does not exist as
125 * is the case with Linux kernel versions < 3.15.
136 r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port);
141 * Switch to dev_id when dev_port returns the same value for
142 * all ports. May happen when using a MOFED release older than
143 * 3.0 with a Linux kernel >= 3.15.
145 if (dev_port == dev_port_prev)
147 dev_port_prev = dev_port;
148 if (dev_port == (priv->port - 1u))
149 snprintf(match, sizeof(match), "%s", name);
152 if (match[0] == '\0') {
156 strncpy(*ifname, match, sizeof(*ifname));
161 * Read from sysfs entry.
164 * Pointer to private structure.
166 * Entry name relative to sysfs path.
168 * Data output buffer.
173 * Number of bytes read on success, negative errno value otherwise and
177 mlx4_sysfs_read(const struct priv *priv, const char *entry,
178 char *buf, size_t size)
180 char ifname[IF_NAMESIZE];
184 ret = mlx4_get_ifname(priv, &ifname);
188 MKSTR(path, "%s/device/net/%s/%s", priv->ctx->device->ibdev_path,
191 file = fopen(path, "rb");
196 ret = fread(buf, 1, size, file);
197 if ((size_t)ret < size && ferror(file)) {
208 * Write to sysfs entry.
211 * Pointer to private structure.
213 * Entry name relative to sysfs path.
220 * Number of bytes written on success, negative errno value otherwise and
224 mlx4_sysfs_write(const struct priv *priv, const char *entry,
225 char *buf, size_t size)
227 char ifname[IF_NAMESIZE];
231 ret = mlx4_get_ifname(priv, &ifname);
235 MKSTR(path, "%s/device/net/%s/%s", priv->ctx->device->ibdev_path,
238 file = fopen(path, "wb");
243 ret = fwrite(buf, 1, size, file);
244 if ((size_t)ret < size || ferror(file)) {
255 * Get unsigned long sysfs property.
258 * Pointer to private structure.
260 * Entry name relative to sysfs path.
262 * Value output buffer.
265 * 0 on success, negative errno value otherwise and rte_errno is set.
268 mlx4_get_sysfs_ulong(struct priv *priv, const char *name, unsigned long *value)
271 unsigned long value_ret;
274 ret = mlx4_sysfs_read(priv, name, value_str, (sizeof(value_str) - 1));
276 DEBUG("cannot read %s value from sysfs: %s",
277 name, strerror(rte_errno));
280 value_str[ret] = '\0';
282 value_ret = strtoul(value_str, NULL, 0);
285 DEBUG("invalid %s value `%s': %s", name, value_str,
286 strerror(rte_errno));
294 * Set unsigned long sysfs property.
297 * Pointer to private structure.
299 * Entry name relative to sysfs path.
304 * 0 on success, negative errno value otherwise and rte_errno is set.
307 mlx4_set_sysfs_ulong(struct priv *priv, const char *name, unsigned long value)
310 MKSTR(value_str, "%lu", value);
312 ret = mlx4_sysfs_write(priv, name, value_str, (sizeof(value_str) - 1));
314 DEBUG("cannot write %s `%s' (%lu) to sysfs: %s",
315 name, value_str, value, strerror(rte_errno));
322 * Perform ifreq ioctl() on associated Ethernet device.
325 * Pointer to private structure.
327 * Request number to pass to ioctl().
329 * Interface request structure output buffer.
332 * 0 on success, negative errno value otherwise and rte_errno is set.
335 mlx4_ifreq(const struct priv *priv, int req, struct ifreq *ifr)
337 int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
344 ret = mlx4_get_ifname(priv, &ifr->ifr_name);
345 if (!ret && ioctl(sock, req, ifr) == -1) {
354 * Get MAC address by querying netdevice.
357 * Pointer to private structure.
359 * MAC address output buffer.
362 * 0 on success, negative errno value otherwise and rte_errno is set.
365 mlx4_get_mac(struct priv *priv, uint8_t (*mac)[ETHER_ADDR_LEN])
367 struct ifreq request;
368 int ret = mlx4_ifreq(priv, SIOCGIFHWADDR, &request);
372 memcpy(mac, request.ifr_hwaddr.sa_data, ETHER_ADDR_LEN);
380 * Pointer to private structure.
382 * MTU value output buffer.
385 * 0 on success, negative errno value otherwise and rte_errno is set.
388 mlx4_mtu_get(struct priv *priv, uint16_t *mtu)
390 unsigned long ulong_mtu = 0;
391 int ret = mlx4_get_sysfs_ulong(priv, "mtu", &ulong_mtu);
400 * DPDK callback to change the MTU.
403 * Pointer to Ethernet device structure.
408 * 0 on success, negative errno value otherwise and rte_errno is set.
411 mlx4_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
413 struct priv *priv = dev->data->dev_private;
415 int ret = mlx4_set_sysfs_ulong(priv, "mtu", mtu);
419 ret = mlx4_mtu_get(priv, &new_mtu);
422 if (new_mtu == mtu) {
434 * Pointer to private structure.
436 * Bitmask for flags that must remain untouched.
438 * Bitmask for flags to modify.
441 * 0 on success, negative errno value otherwise and rte_errno is set.
444 mlx4_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
446 unsigned long tmp = 0;
447 int ret = mlx4_get_sysfs_ulong(priv, "flags", &tmp);
452 tmp |= (flags & (~keep));
453 return mlx4_set_sysfs_ulong(priv, "flags", tmp);
457 * Change the link state (UP / DOWN).
460 * Pointer to Ethernet device private data.
462 * Nonzero for link up, otherwise link down.
465 * 0 on success, negative errno value otherwise and rte_errno is set.
468 mlx4_dev_set_link(struct priv *priv, int up)
470 struct rte_eth_dev *dev = priv->dev;
474 err = mlx4_set_flags(priv, ~IFF_UP, IFF_UP);
477 dev->rx_pkt_burst = mlx4_rx_burst;
479 err = mlx4_set_flags(priv, ~IFF_UP, ~IFF_UP);
482 dev->rx_pkt_burst = mlx4_rx_burst_removed;
483 dev->tx_pkt_burst = mlx4_tx_burst_removed;
489 * DPDK callback to bring the link DOWN.
492 * Pointer to Ethernet device structure.
495 * 0 on success, negative errno value otherwise and rte_errno is set.
498 mlx4_dev_set_link_down(struct rte_eth_dev *dev)
500 struct priv *priv = dev->data->dev_private;
502 return mlx4_dev_set_link(priv, 0);
506 * DPDK callback to bring the link UP.
509 * Pointer to Ethernet device structure.
512 * 0 on success, negative errno value otherwise and rte_errno is set.
515 mlx4_dev_set_link_up(struct rte_eth_dev *dev)
517 struct priv *priv = dev->data->dev_private;
519 return mlx4_dev_set_link(priv, 1);
523 * DPDK callback to remove a MAC address.
526 * Pointer to Ethernet device structure.
531 mlx4_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
533 struct priv *priv = dev->data->dev_private;
534 struct rte_flow_error error;
536 if (index >= RTE_DIM(priv->mac)) {
540 memset(&priv->mac[index], 0, sizeof(priv->mac[index]));
541 if (!mlx4_flow_sync(priv, &error))
543 ERROR("failed to synchronize flow rules after removing MAC address"
544 " at index %d (code %d, \"%s\"),"
545 " flow error type %d, cause %p, message: %s",
546 index, rte_errno, strerror(rte_errno), error.type, error.cause,
547 error.message ? error.message : "(unspecified)");
551 * DPDK callback to add a MAC address.
554 * Pointer to Ethernet device structure.
556 * MAC address to register.
560 * VMDq pool index to associate address with (ignored).
563 * 0 on success, negative errno value otherwise and rte_errno is set.
566 mlx4_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
567 uint32_t index, uint32_t vmdq)
569 struct priv *priv = dev->data->dev_private;
570 struct rte_flow_error error;
574 if (index >= RTE_DIM(priv->mac)) {
578 memcpy(&priv->mac[index], mac_addr, sizeof(priv->mac[index]));
579 ret = mlx4_flow_sync(priv, &error);
582 ERROR("failed to synchronize flow rules after adding MAC address"
583 " at index %d (code %d, \"%s\"),"
584 " flow error type %d, cause %p, message: %s",
585 index, rte_errno, strerror(rte_errno), error.type, error.cause,
586 error.message ? error.message : "(unspecified)");
591 * DPDK callback to set the primary MAC address.
594 * Pointer to Ethernet device structure.
596 * MAC address to register.
599 mlx4_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
601 mlx4_mac_addr_add(dev, mac_addr, 0, 0);
605 * DPDK callback to get information about the device.
608 * Pointer to Ethernet device structure.
610 * Info structure output buffer.
613 mlx4_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
615 struct priv *priv = dev->data->dev_private;
617 char ifname[IF_NAMESIZE];
619 info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
622 /* FIXME: we should ask the device for these values. */
623 info->min_rx_bufsize = 32;
624 info->max_rx_pktlen = 65536;
626 * Since we need one CQ per QP, the limit is the minimum number
627 * between the two values.
629 max = ((priv->device_attr.max_cq > priv->device_attr.max_qp) ?
630 priv->device_attr.max_qp : priv->device_attr.max_cq);
631 /* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
634 info->max_rx_queues = max;
635 info->max_tx_queues = max;
636 info->max_mac_addrs = RTE_DIM(priv->mac);
637 info->rx_offload_capa = 0;
638 info->tx_offload_capa = 0;
639 if (mlx4_get_ifname(priv, &ifname) == 0)
640 info->if_index = if_nametoindex(ifname);
650 * DPDK callback to get device statistics.
653 * Pointer to Ethernet device structure.
655 * Stats structure output buffer.
658 mlx4_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
660 struct rte_eth_stats tmp;
664 memset(&tmp, 0, sizeof(tmp));
665 /* Add software counters. */
666 for (i = 0; i != dev->data->nb_rx_queues; ++i) {
667 struct rxq *rxq = dev->data->rx_queues[i];
671 idx = rxq->stats.idx;
672 if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
673 tmp.q_ipackets[idx] += rxq->stats.ipackets;
674 tmp.q_ibytes[idx] += rxq->stats.ibytes;
675 tmp.q_errors[idx] += (rxq->stats.idropped +
676 rxq->stats.rx_nombuf);
678 tmp.ipackets += rxq->stats.ipackets;
679 tmp.ibytes += rxq->stats.ibytes;
680 tmp.ierrors += rxq->stats.idropped;
681 tmp.rx_nombuf += rxq->stats.rx_nombuf;
683 for (i = 0; i != dev->data->nb_tx_queues; ++i) {
684 struct txq *txq = dev->data->tx_queues[i];
688 idx = txq->stats.idx;
689 if (idx < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
690 tmp.q_opackets[idx] += txq->stats.opackets;
691 tmp.q_obytes[idx] += txq->stats.obytes;
692 tmp.q_errors[idx] += txq->stats.odropped;
694 tmp.opackets += txq->stats.opackets;
695 tmp.obytes += txq->stats.obytes;
696 tmp.oerrors += txq->stats.odropped;
703 * DPDK callback to clear device statistics.
706 * Pointer to Ethernet device structure.
709 mlx4_stats_reset(struct rte_eth_dev *dev)
713 for (i = 0; i != dev->data->nb_rx_queues; ++i) {
714 struct rxq *rxq = dev->data->rx_queues[i];
717 rxq->stats = (struct mlx4_rxq_stats){
718 .idx = rxq->stats.idx,
721 for (i = 0; i != dev->data->nb_tx_queues; ++i) {
722 struct txq *txq = dev->data->tx_queues[i];
725 txq->stats = (struct mlx4_txq_stats){
726 .idx = txq->stats.idx,
732 * DPDK callback to retrieve physical link information.
735 * Pointer to Ethernet device structure.
736 * @param wait_to_complete
737 * Wait for request completion (ignored).
740 * 0 on success, negative errno value otherwise and rte_errno is set.
743 mlx4_link_update(struct rte_eth_dev *dev, int wait_to_complete)
745 const struct priv *priv = dev->data->dev_private;
746 struct ethtool_cmd edata = {
750 struct rte_eth_link dev_link;
757 (void)wait_to_complete;
758 if (mlx4_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
759 WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(rte_errno));
762 memset(&dev_link, 0, sizeof(dev_link));
763 dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
764 (ifr.ifr_flags & IFF_RUNNING));
765 ifr.ifr_data = (void *)&edata;
766 if (mlx4_ifreq(priv, SIOCETHTOOL, &ifr)) {
767 WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s",
768 strerror(rte_errno));
771 link_speed = ethtool_cmd_speed(&edata);
772 if (link_speed == -1)
773 dev_link.link_speed = 0;
775 dev_link.link_speed = link_speed;
776 dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
777 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
778 dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
779 ETH_LINK_SPEED_FIXED);
780 dev->data->dev_link = dev_link;
785 * DPDK callback to get flow control status.
788 * Pointer to Ethernet device structure.
789 * @param[out] fc_conf
790 * Flow control output buffer.
793 * 0 on success, negative errno value otherwise and rte_errno is set.
796 mlx4_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
798 struct priv *priv = dev->data->dev_private;
800 struct ethtool_pauseparam ethpause = {
801 .cmd = ETHTOOL_GPAUSEPARAM,
805 ifr.ifr_data = (void *)ðpause;
806 if (mlx4_ifreq(priv, SIOCETHTOOL, &ifr)) {
808 WARN("ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM)"
810 strerror(rte_errno));
813 fc_conf->autoneg = ethpause.autoneg;
814 if (ethpause.rx_pause && ethpause.tx_pause)
815 fc_conf->mode = RTE_FC_FULL;
816 else if (ethpause.rx_pause)
817 fc_conf->mode = RTE_FC_RX_PAUSE;
818 else if (ethpause.tx_pause)
819 fc_conf->mode = RTE_FC_TX_PAUSE;
821 fc_conf->mode = RTE_FC_NONE;
829 * DPDK callback to modify flow control parameters.
832 * Pointer to Ethernet device structure.
834 * Flow control parameters.
837 * 0 on success, negative errno value otherwise and rte_errno is set.
840 mlx4_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
842 struct priv *priv = dev->data->dev_private;
844 struct ethtool_pauseparam ethpause = {
845 .cmd = ETHTOOL_SPAUSEPARAM,
849 ifr.ifr_data = (void *)ðpause;
850 ethpause.autoneg = fc_conf->autoneg;
851 if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
852 (fc_conf->mode & RTE_FC_RX_PAUSE))
853 ethpause.rx_pause = 1;
855 ethpause.rx_pause = 0;
856 if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
857 (fc_conf->mode & RTE_FC_TX_PAUSE))
858 ethpause.tx_pause = 1;
860 ethpause.tx_pause = 0;
861 if (mlx4_ifreq(priv, SIOCETHTOOL, &ifr)) {
863 WARN("ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
865 strerror(rte_errno));