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.
46 #include <sys/ioctl.h>
47 #include <sys/socket.h>
48 #include <sys/utsname.h>
49 #include <netinet/in.h>
50 #include <linux/ethtool.h>
51 #include <linux/sockios.h>
52 #include <linux/version.h>
57 #include <rte_atomic.h>
58 #include <rte_ethdev.h>
60 #include <rte_common.h>
61 #include <rte_interrupts.h>
62 #include <rte_alarm.h>
63 #include <rte_malloc.h>
66 #include "mlx5_rxtx.h"
67 #include "mlx5_utils.h"
69 /* Add defines in case the running kernel is not the same as user headers. */
70 #ifndef ETHTOOL_GLINKSETTINGS
71 struct ethtool_link_settings {
80 uint8_t eth_tp_mdix_ctrl;
81 int8_t link_mode_masks_nwords;
83 uint32_t link_mode_masks[];
86 #define ETHTOOL_GLINKSETTINGS 0x0000004c
87 #define ETHTOOL_LINK_MODE_1000baseT_Full_BIT 5
88 #define ETHTOOL_LINK_MODE_Autoneg_BIT 6
89 #define ETHTOOL_LINK_MODE_1000baseKX_Full_BIT 17
90 #define ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT 18
91 #define ETHTOOL_LINK_MODE_10000baseKR_Full_BIT 19
92 #define ETHTOOL_LINK_MODE_10000baseR_FEC_BIT 20
93 #define ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT 21
94 #define ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT 22
95 #define ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT 23
96 #define ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT 24
97 #define ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT 25
98 #define ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT 26
99 #define ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT 27
100 #define ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT 28
101 #define ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT 29
102 #define ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT 30
104 #ifndef HAVE_ETHTOOL_LINK_MODE_25G
105 #define ETHTOOL_LINK_MODE_25000baseCR_Full_BIT 31
106 #define ETHTOOL_LINK_MODE_25000baseKR_Full_BIT 32
107 #define ETHTOOL_LINK_MODE_25000baseSR_Full_BIT 33
109 #ifndef HAVE_ETHTOOL_LINK_MODE_50G
110 #define ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT 34
111 #define ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT 35
113 #ifndef HAVE_ETHTOOL_LINK_MODE_100G
114 #define ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT 36
115 #define ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT 37
116 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38
117 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39
121 * Return private structure associated with an Ethernet device.
124 * Pointer to Ethernet device structure.
127 * Pointer to private structure.
130 mlx5_get_priv(struct rte_eth_dev *dev)
132 return dev->data->dev_private;
136 * Check if running as a secondary process.
139 * Nonzero if running as a secondary process.
142 mlx5_is_secondary(void)
144 return rte_eal_process_type() == RTE_PROC_SECONDARY;
148 * Get interface name from private structure.
151 * Pointer to private structure.
153 * Interface name output buffer.
156 * 0 on success, -1 on failure and errno is set.
159 priv_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE])
163 unsigned int dev_type = 0;
164 unsigned int dev_port_prev = ~0u;
165 char match[IF_NAMESIZE] = "";
168 MKSTR(path, "%s/device/net", priv->ibdev_path);
174 while ((dent = readdir(dir)) != NULL) {
175 char *name = dent->d_name;
177 unsigned int dev_port;
180 if ((name[0] == '.') &&
181 ((name[1] == '\0') ||
182 ((name[1] == '.') && (name[2] == '\0'))))
185 MKSTR(path, "%s/device/net/%s/%s",
186 priv->ibdev_path, name,
187 (dev_type ? "dev_id" : "dev_port"));
189 file = fopen(path, "rb");
194 * Switch to dev_id when dev_port does not exist as
195 * is the case with Linux kernel versions < 3.15.
206 r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port);
211 * Switch to dev_id when dev_port returns the same value for
212 * all ports. May happen when using a MOFED release older than
213 * 3.0 with a Linux kernel >= 3.15.
215 if (dev_port == dev_port_prev)
217 dev_port_prev = dev_port;
218 if (dev_port == (priv->port - 1u))
219 snprintf(match, sizeof(match), "%s", name);
222 if (match[0] == '\0')
224 strncpy(*ifname, match, sizeof(*ifname));
229 * Check if the counter is located on ib counters file.
235 * 1 if counter is located on ib counters file , 0 otherwise.
238 priv_is_ib_cntr(const char *cntr)
240 if (!strcmp(cntr, "out_of_buffer"))
246 * Read from sysfs entry.
249 * Pointer to private structure.
251 * Entry name relative to sysfs path.
253 * Data output buffer.
258 * 0 on success, -1 on failure and errno is set.
261 priv_sysfs_read(const struct priv *priv, const char *entry,
262 char *buf, size_t size)
264 char ifname[IF_NAMESIZE];
269 if (priv_get_ifname(priv, &ifname))
272 if (priv_is_ib_cntr(entry)) {
273 MKSTR(path, "%s/ports/1/hw_counters/%s",
274 priv->ibdev_path, entry);
275 file = fopen(path, "rb");
277 MKSTR(path, "%s/device/net/%s/%s",
278 priv->ibdev_path, ifname, entry);
279 file = fopen(path, "rb");
283 ret = fread(buf, 1, size, file);
285 if (((size_t)ret < size) && (ferror(file)))
295 * Write to sysfs entry.
298 * Pointer to private structure.
300 * Entry name relative to sysfs path.
307 * 0 on success, -1 on failure and errno is set.
310 priv_sysfs_write(const struct priv *priv, const char *entry,
311 char *buf, size_t size)
313 char ifname[IF_NAMESIZE];
318 if (priv_get_ifname(priv, &ifname))
321 MKSTR(path, "%s/device/net/%s/%s", priv->ibdev_path, ifname, entry);
323 file = fopen(path, "wb");
326 ret = fwrite(buf, 1, size, file);
328 if (((size_t)ret < size) || (ferror(file)))
338 * Get unsigned long sysfs property.
341 * Pointer to private structure.
343 * Entry name relative to sysfs path.
345 * Value output buffer.
348 * 0 on success, -1 on failure and errno is set.
351 priv_get_sysfs_ulong(struct priv *priv, const char *name, unsigned long *value)
354 unsigned long value_ret;
357 ret = priv_sysfs_read(priv, name, value_str, (sizeof(value_str) - 1));
359 DEBUG("cannot read %s value from sysfs: %s",
360 name, strerror(errno));
363 value_str[ret] = '\0';
365 value_ret = strtoul(value_str, NULL, 0);
367 DEBUG("invalid %s value `%s': %s", name, value_str,
376 * Set unsigned long sysfs property.
379 * Pointer to private structure.
381 * Entry name relative to sysfs path.
386 * 0 on success, -1 on failure and errno is set.
389 priv_set_sysfs_ulong(struct priv *priv, const char *name, unsigned long value)
392 MKSTR(value_str, "%lu", value);
394 ret = priv_sysfs_write(priv, name, value_str, (sizeof(value_str) - 1));
396 DEBUG("cannot write %s `%s' (%lu) to sysfs: %s",
397 name, value_str, value, strerror(errno));
404 * Perform ifreq ioctl() on associated Ethernet device.
407 * Pointer to private structure.
409 * Request number to pass to ioctl().
411 * Interface request structure output buffer.
414 * 0 on success, -1 on failure and errno is set.
417 priv_ifreq(const struct priv *priv, int req, struct ifreq *ifr)
419 int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
424 if (priv_get_ifname(priv, &ifr->ifr_name) == 0)
425 ret = ioctl(sock, req, ifr);
431 * Return the number of active VFs for the current device.
434 * Pointer to private structure.
435 * @param[out] num_vfs
436 * Number of active VFs.
439 * 0 on success, -1 on failure and errno is set.
442 priv_get_num_vfs(struct priv *priv, uint16_t *num_vfs)
444 /* The sysfs entry name depends on the operating system. */
445 const char **name = (const char *[]){
446 "device/sriov_numvfs",
447 "device/mlx5_num_vfs",
453 unsigned long ulong_num_vfs;
455 ret = priv_get_sysfs_ulong(priv, *name, &ulong_num_vfs);
457 *num_vfs = ulong_num_vfs;
458 } while (*(++name) && ret);
466 * Pointer to private structure.
468 * MTU value output buffer.
471 * 0 on success, -1 on failure and errno is set.
474 priv_get_mtu(struct priv *priv, uint16_t *mtu)
476 unsigned long ulong_mtu;
478 if (priv_get_sysfs_ulong(priv, "mtu", &ulong_mtu) == -1)
485 * Read device counter from sysfs.
488 * Pointer to private structure.
492 * Counter output buffer.
495 * 0 on success, -1 on failure and errno is set.
498 priv_get_cntr_sysfs(struct priv *priv, const char *name, uint64_t *cntr)
500 unsigned long ulong_ctr;
502 if (priv_get_sysfs_ulong(priv, name, &ulong_ctr) == -1)
512 * Pointer to private structure.
517 * 0 on success, -1 on failure and errno is set.
520 priv_set_mtu(struct priv *priv, uint16_t mtu)
524 if (priv_set_sysfs_ulong(priv, "mtu", mtu) ||
525 priv_get_mtu(priv, &new_mtu))
537 * Pointer to private structure.
539 * Bitmask for flags that must remain untouched.
541 * Bitmask for flags to modify.
544 * 0 on success, -1 on failure and errno is set.
547 priv_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
551 if (priv_get_sysfs_ulong(priv, "flags", &tmp) == -1)
554 tmp |= (flags & (~keep));
555 return priv_set_sysfs_ulong(priv, "flags", tmp);
559 * Ethernet device configuration.
561 * Prepare the driver for a given number of TX and RX queues.
564 * Pointer to Ethernet device structure.
567 * 0 on success, errno value on failure.
570 dev_configure(struct rte_eth_dev *dev)
572 struct priv *priv = dev->data->dev_private;
573 unsigned int rxqs_n = dev->data->nb_rx_queues;
574 unsigned int txqs_n = dev->data->nb_tx_queues;
577 unsigned int reta_idx_n;
579 priv->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
580 priv->rxqs = (void *)dev->data->rx_queues;
581 priv->txqs = (void *)dev->data->tx_queues;
582 if (txqs_n != priv->txqs_n) {
583 INFO("%p: TX queues number update: %u -> %u",
584 (void *)dev, priv->txqs_n, txqs_n);
585 priv->txqs_n = txqs_n;
587 if (rxqs_n > priv->ind_table_max_size) {
588 ERROR("cannot handle this many RX queues (%u)", rxqs_n);
591 if (rxqs_n == priv->rxqs_n)
593 INFO("%p: RX queues number update: %u -> %u",
594 (void *)dev, priv->rxqs_n, rxqs_n);
595 priv->rxqs_n = rxqs_n;
596 /* If the requested number of RX queues is not a power of two, use the
597 * maximum indirection table size for better balancing.
598 * The result is always rounded to the next power of two. */
599 reta_idx_n = (1 << log2above((rxqs_n & (rxqs_n - 1)) ?
600 priv->ind_table_max_size :
602 if (priv_rss_reta_index_resize(priv, reta_idx_n))
604 /* When the number of RX queues is not a power of two, the remaining
605 * table entries are padded with reused WQs and hashes are not spread
607 for (i = 0, j = 0; (i != reta_idx_n); ++i) {
608 (*priv->reta_idx)[i] = j;
616 * DPDK callback for Ethernet device configuration.
619 * Pointer to Ethernet device structure.
622 * 0 on success, negative errno value on failure.
625 mlx5_dev_configure(struct rte_eth_dev *dev)
627 struct priv *priv = dev->data->dev_private;
630 if (mlx5_is_secondary())
631 return -E_RTE_SECONDARY;
634 ret = dev_configure(dev);
641 * DPDK callback to get information about the device.
644 * Pointer to Ethernet device structure.
646 * Info structure output buffer.
649 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
651 struct priv *priv = mlx5_get_priv(dev);
653 char ifname[IF_NAMESIZE];
655 info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
658 /* FIXME: we should ask the device for these values. */
659 info->min_rx_bufsize = 32;
660 info->max_rx_pktlen = 65536;
662 * Since we need one CQ per QP, the limit is the minimum number
663 * between the two values.
665 max = RTE_MIN(priv->device_attr.orig_attr.max_cq,
666 priv->device_attr.orig_attr.max_qp);
667 /* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
670 info->max_rx_queues = max;
671 info->max_tx_queues = max;
672 info->max_mac_addrs = RTE_DIM(priv->mac);
673 info->rx_offload_capa =
675 (DEV_RX_OFFLOAD_IPV4_CKSUM |
676 DEV_RX_OFFLOAD_UDP_CKSUM |
677 DEV_RX_OFFLOAD_TCP_CKSUM) :
679 (priv->hw_vlan_strip ? DEV_RX_OFFLOAD_VLAN_STRIP : 0);
681 info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
683 info->tx_offload_capa |=
684 (DEV_TX_OFFLOAD_IPV4_CKSUM |
685 DEV_TX_OFFLOAD_UDP_CKSUM |
686 DEV_TX_OFFLOAD_TCP_CKSUM);
688 info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
690 info->tx_offload_capa |= (DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
691 DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
692 DEV_TX_OFFLOAD_GRE_TNL_TSO);
693 if (priv_get_ifname(priv, &ifname) == 0)
694 info->if_index = if_nametoindex(ifname);
695 info->reta_size = priv->reta_idx_n ?
696 priv->reta_idx_n : priv->ind_table_max_size;
697 info->hash_key_size = ((*priv->rss_conf) ?
698 (*priv->rss_conf)[0]->rss_key_len :
700 info->speed_capa = priv->link_speed_capa;
705 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev)
707 static const uint32_t ptypes[] = {
708 /* refers to rxq_cq_to_pkt_type() */
710 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
711 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
712 RTE_PTYPE_L4_NONFRAG,
716 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
717 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
718 RTE_PTYPE_INNER_L4_NONFRAG,
719 RTE_PTYPE_INNER_L4_FRAG,
720 RTE_PTYPE_INNER_L4_TCP,
721 RTE_PTYPE_INNER_L4_UDP,
725 if (dev->rx_pkt_burst == mlx5_rx_burst ||
726 dev->rx_pkt_burst == mlx5_rx_burst_vec)
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 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, int wait_to_complete)
742 struct priv *priv = mlx5_get_priv(dev);
743 struct ethtool_cmd edata = {
744 .cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
747 struct rte_eth_link dev_link;
750 /* priv_lock() is not taken to allow concurrent calls. */
752 (void)wait_to_complete;
753 if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
754 WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
757 memset(&dev_link, 0, sizeof(dev_link));
758 dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
759 (ifr.ifr_flags & IFF_RUNNING));
760 ifr.ifr_data = (void *)&edata;
761 if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
762 WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s",
766 link_speed = ethtool_cmd_speed(&edata);
767 if (link_speed == -1)
768 dev_link.link_speed = 0;
770 dev_link.link_speed = link_speed;
771 priv->link_speed_capa = 0;
772 if (edata.supported & SUPPORTED_Autoneg)
773 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
774 if (edata.supported & (SUPPORTED_1000baseT_Full |
775 SUPPORTED_1000baseKX_Full))
776 priv->link_speed_capa |= ETH_LINK_SPEED_1G;
777 if (edata.supported & SUPPORTED_10000baseKR_Full)
778 priv->link_speed_capa |= ETH_LINK_SPEED_10G;
779 if (edata.supported & (SUPPORTED_40000baseKR4_Full |
780 SUPPORTED_40000baseCR4_Full |
781 SUPPORTED_40000baseSR4_Full |
782 SUPPORTED_40000baseLR4_Full))
783 priv->link_speed_capa |= ETH_LINK_SPEED_40G;
784 dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
785 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
786 dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
787 ETH_LINK_SPEED_FIXED);
788 if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
789 /* Link status changed. */
790 dev->data->dev_link = dev_link;
793 /* Link status is still the same. */
798 * Retrieve physical link information (unlocked version using new ioctl).
801 * Pointer to Ethernet device structure.
802 * @param wait_to_complete
803 * Wait for request completion (ignored).
806 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, int wait_to_complete)
808 struct priv *priv = mlx5_get_priv(dev);
809 struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
811 struct rte_eth_link dev_link;
814 (void)wait_to_complete;
815 if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
816 WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
819 memset(&dev_link, 0, sizeof(dev_link));
820 dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
821 (ifr.ifr_flags & IFF_RUNNING));
822 ifr.ifr_data = (void *)&gcmd;
823 if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
824 DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
828 gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
830 alignas(struct ethtool_link_settings)
831 uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
832 sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
833 struct ethtool_link_settings *ecmd = (void *)data;
836 ifr.ifr_data = (void *)ecmd;
837 if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
838 DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
842 dev_link.link_speed = ecmd->speed;
843 sc = ecmd->link_mode_masks[0] |
844 ((uint64_t)ecmd->link_mode_masks[1] << 32);
845 priv->link_speed_capa = 0;
846 if (sc & ETHTOOL_LINK_MODE_Autoneg_BIT)
847 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
848 if (sc & (ETHTOOL_LINK_MODE_1000baseT_Full_BIT |
849 ETHTOOL_LINK_MODE_1000baseKX_Full_BIT))
850 priv->link_speed_capa |= ETH_LINK_SPEED_1G;
851 if (sc & (ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT |
852 ETHTOOL_LINK_MODE_10000baseKR_Full_BIT |
853 ETHTOOL_LINK_MODE_10000baseR_FEC_BIT))
854 priv->link_speed_capa |= ETH_LINK_SPEED_10G;
855 if (sc & (ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT |
856 ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT))
857 priv->link_speed_capa |= ETH_LINK_SPEED_20G;
858 if (sc & (ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT |
859 ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT |
860 ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT |
861 ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT))
862 priv->link_speed_capa |= ETH_LINK_SPEED_40G;
863 if (sc & (ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT |
864 ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT |
865 ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT |
866 ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT))
867 priv->link_speed_capa |= ETH_LINK_SPEED_56G;
868 if (sc & (ETHTOOL_LINK_MODE_25000baseCR_Full_BIT |
869 ETHTOOL_LINK_MODE_25000baseKR_Full_BIT |
870 ETHTOOL_LINK_MODE_25000baseSR_Full_BIT))
871 priv->link_speed_capa |= ETH_LINK_SPEED_25G;
872 if (sc & (ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT |
873 ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT))
874 priv->link_speed_capa |= ETH_LINK_SPEED_50G;
875 if (sc & (ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT |
876 ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT |
877 ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT |
878 ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT))
879 priv->link_speed_capa |= ETH_LINK_SPEED_100G;
880 dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
881 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
882 dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
883 ETH_LINK_SPEED_FIXED);
884 if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
885 /* Link status changed. */
886 dev->data->dev_link = dev_link;
889 /* Link status is still the same. */
894 * DPDK callback to retrieve physical link information.
897 * Pointer to Ethernet device structure.
898 * @param wait_to_complete
899 * Wait for request completion (ignored).
902 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
904 struct utsname utsname;
907 if (uname(&utsname) == -1 ||
908 sscanf(utsname.release, "%d.%d.%d",
909 &ver[0], &ver[1], &ver[2]) != 3 ||
910 KERNEL_VERSION(ver[0], ver[1], ver[2]) < KERNEL_VERSION(4, 9, 0))
911 return mlx5_link_update_unlocked_gset(dev, wait_to_complete);
912 return mlx5_link_update_unlocked_gs(dev, wait_to_complete);
916 * DPDK callback to change the MTU.
919 * Pointer to Ethernet device structure.
924 * 0 on success, negative errno value on failure.
927 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
929 struct priv *priv = dev->data->dev_private;
933 if (mlx5_is_secondary())
934 return -E_RTE_SECONDARY;
937 ret = priv_get_mtu(priv, &kern_mtu);
940 /* Set kernel interface MTU first. */
941 ret = priv_set_mtu(priv, mtu);
944 ret = priv_get_mtu(priv, &kern_mtu);
947 if (kern_mtu == mtu) {
949 DEBUG("adapter port %u MTU set to %u", priv->port, mtu);
955 WARN("cannot set port %u MTU to %u: %s", priv->port, mtu,
963 * DPDK callback to get flow control status.
966 * Pointer to Ethernet device structure.
967 * @param[out] fc_conf
968 * Flow control output buffer.
971 * 0 on success, negative errno value on failure.
974 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
976 struct priv *priv = dev->data->dev_private;
978 struct ethtool_pauseparam ethpause = {
979 .cmd = ETHTOOL_GPAUSEPARAM
983 if (mlx5_is_secondary())
984 return -E_RTE_SECONDARY;
986 ifr.ifr_data = (void *)ðpause;
988 if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
990 WARN("ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM)"
996 fc_conf->autoneg = ethpause.autoneg;
997 if (ethpause.rx_pause && ethpause.tx_pause)
998 fc_conf->mode = RTE_FC_FULL;
999 else if (ethpause.rx_pause)
1000 fc_conf->mode = RTE_FC_RX_PAUSE;
1001 else if (ethpause.tx_pause)
1002 fc_conf->mode = RTE_FC_TX_PAUSE;
1004 fc_conf->mode = RTE_FC_NONE;
1014 * DPDK callback to modify flow control parameters.
1017 * Pointer to Ethernet device structure.
1018 * @param[in] fc_conf
1019 * Flow control parameters.
1022 * 0 on success, negative errno value on failure.
1025 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1027 struct priv *priv = dev->data->dev_private;
1029 struct ethtool_pauseparam ethpause = {
1030 .cmd = ETHTOOL_SPAUSEPARAM
1034 if (mlx5_is_secondary())
1035 return -E_RTE_SECONDARY;
1037 ifr.ifr_data = (void *)ðpause;
1038 ethpause.autoneg = fc_conf->autoneg;
1039 if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1040 (fc_conf->mode & RTE_FC_RX_PAUSE))
1041 ethpause.rx_pause = 1;
1043 ethpause.rx_pause = 0;
1045 if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1046 (fc_conf->mode & RTE_FC_TX_PAUSE))
1047 ethpause.tx_pause = 1;
1049 ethpause.tx_pause = 0;
1052 if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
1054 WARN("ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
1068 * Get PCI information from struct ibv_device.
1071 * Pointer to Ethernet device structure.
1072 * @param[out] pci_addr
1073 * PCI bus address output buffer.
1076 * 0 on success, -1 on failure and errno is set.
1079 mlx5_ibv_device_to_pci_addr(const struct ibv_device *device,
1080 struct rte_pci_addr *pci_addr)
1084 MKSTR(path, "%s/device/uevent", device->ibdev_path);
1086 file = fopen(path, "rb");
1089 while (fgets(line, sizeof(line), file) == line) {
1090 size_t len = strlen(line);
1093 /* Truncate long lines. */
1094 if (len == (sizeof(line) - 1))
1095 while (line[(len - 1)] != '\n') {
1099 line[(len - 1)] = ret;
1101 /* Extract information. */
1104 "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
1108 &pci_addr->function) == 4) {
1118 * Update the link status.
1121 * Pointer to private structure.
1124 * Zero if the callback process can be called immediately.
1127 priv_link_status_update(struct priv *priv)
1129 struct rte_eth_link *link = &priv->dev->data->dev_link;
1131 mlx5_link_update(priv->dev, 0);
1132 if (((link->link_speed == 0) && link->link_status) ||
1133 ((link->link_speed != 0) && !link->link_status)) {
1135 * Inconsistent status. Event likely occurred before the
1136 * kernel netdevice exposes the new status.
1138 if (!priv->pending_alarm) {
1139 priv->pending_alarm = 1;
1140 rte_eal_alarm_set(MLX5_ALARM_TIMEOUT_US,
1141 mlx5_dev_link_status_handler,
1145 } else if (unlikely(priv->pending_alarm)) {
1146 /* Link interrupt occurred while alarm is already scheduled. */
1147 priv->pending_alarm = 0;
1148 rte_eal_alarm_cancel(mlx5_dev_link_status_handler, priv->dev);
1154 * Device status handler.
1157 * Pointer to private structure.
1159 * Pointer to event flags holder.
1162 * Events bitmap of callback process which can be called immediately.
1165 priv_dev_status_handler(struct priv *priv)
1167 struct ibv_async_event event;
1170 /* Read all message and acknowledge them. */
1172 if (ibv_get_async_event(priv->ctx, &event))
1174 if ((event.event_type == IBV_EVENT_PORT_ACTIVE ||
1175 event.event_type == IBV_EVENT_PORT_ERR) &&
1176 (priv->dev->data->dev_conf.intr_conf.lsc == 1))
1177 ret |= (1 << RTE_ETH_EVENT_INTR_LSC);
1178 else if (event.event_type == IBV_EVENT_DEVICE_FATAL &&
1179 priv->dev->data->dev_conf.intr_conf.rmv == 1)
1180 ret |= (1 << RTE_ETH_EVENT_INTR_RMV);
1182 DEBUG("event type %d on port %d not handled",
1183 event.event_type, event.element.port_num);
1184 ibv_ack_async_event(&event);
1186 if (ret & (1 << RTE_ETH_EVENT_INTR_LSC))
1187 if (priv_link_status_update(priv))
1188 ret &= ~(1 << RTE_ETH_EVENT_INTR_LSC);
1193 * Handle delayed link status event.
1196 * Registered argument.
1199 mlx5_dev_link_status_handler(void *arg)
1201 struct rte_eth_dev *dev = arg;
1202 struct priv *priv = dev->data->dev_private;
1206 assert(priv->pending_alarm == 1);
1207 priv->pending_alarm = 0;
1208 ret = priv_link_status_update(priv);
1211 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1216 * Handle interrupts from the NIC.
1218 * @param[in] intr_handle
1219 * Interrupt handler.
1221 * Callback argument.
1224 mlx5_dev_interrupt_handler(void *cb_arg)
1226 struct rte_eth_dev *dev = cb_arg;
1227 struct priv *priv = dev->data->dev_private;
1231 events = priv_dev_status_handler(priv);
1233 if (events & (1 << RTE_ETH_EVENT_INTR_LSC))
1234 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1236 if (events & (1 << RTE_ETH_EVENT_INTR_RMV))
1237 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RMV, NULL,
1242 * Handle interrupts from the socket.
1245 * Callback argument.
1248 mlx5_dev_handler_socket(void *cb_arg)
1250 struct rte_eth_dev *dev = cb_arg;
1251 struct priv *priv = dev->data->dev_private;
1254 priv_socket_handle(priv);
1259 * Uninstall interrupt handler.
1262 * Pointer to private structure.
1264 * Pointer to the rte_eth_dev structure.
1267 priv_dev_interrupt_handler_uninstall(struct priv *priv, struct rte_eth_dev *dev)
1269 if (dev->data->dev_conf.intr_conf.lsc ||
1270 dev->data->dev_conf.intr_conf.rmv)
1271 rte_intr_callback_unregister(&priv->intr_handle,
1272 mlx5_dev_interrupt_handler, dev);
1273 if (priv->primary_socket)
1274 rte_intr_callback_unregister(&priv->intr_handle_socket,
1275 mlx5_dev_handler_socket, dev);
1276 if (priv->pending_alarm)
1277 rte_eal_alarm_cancel(mlx5_dev_link_status_handler, dev);
1278 priv->pending_alarm = 0;
1279 priv->intr_handle.fd = 0;
1280 priv->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
1281 priv->intr_handle_socket.fd = 0;
1282 priv->intr_handle_socket.type = RTE_INTR_HANDLE_UNKNOWN;
1286 * Install interrupt handler.
1289 * Pointer to private structure.
1291 * Pointer to the rte_eth_dev structure.
1294 priv_dev_interrupt_handler_install(struct priv *priv, struct rte_eth_dev *dev)
1298 assert(!mlx5_is_secondary());
1299 assert(priv->ctx->async_fd > 0);
1300 flags = fcntl(priv->ctx->async_fd, F_GETFL);
1301 rc = fcntl(priv->ctx->async_fd, F_SETFL, flags | O_NONBLOCK);
1303 INFO("failed to change file descriptor async event queue");
1304 dev->data->dev_conf.intr_conf.lsc = 0;
1305 dev->data->dev_conf.intr_conf.rmv = 0;
1307 if (dev->data->dev_conf.intr_conf.lsc ||
1308 dev->data->dev_conf.intr_conf.rmv) {
1309 priv->intr_handle.fd = priv->ctx->async_fd;
1310 priv->intr_handle.type = RTE_INTR_HANDLE_EXT;
1311 rte_intr_callback_register(&priv->intr_handle,
1312 mlx5_dev_interrupt_handler, dev);
1315 rc = priv_socket_init(priv);
1316 if (!rc && priv->primary_socket) {
1317 priv->intr_handle_socket.fd = priv->primary_socket;
1318 priv->intr_handle_socket.type = RTE_INTR_HANDLE_EXT;
1319 rte_intr_callback_register(&priv->intr_handle_socket,
1320 mlx5_dev_handler_socket, dev);
1325 * Change the link state (UP / DOWN).
1328 * Pointer to private data structure.
1330 * Pointer to rte_eth_dev structure.
1332 * Nonzero for link up, otherwise link down.
1335 * 0 on success, errno value on failure.
1338 priv_dev_set_link(struct priv *priv, struct rte_eth_dev *dev, int up)
1343 err = priv_set_flags(priv, ~IFF_UP, IFF_UP);
1346 priv_dev_select_tx_function(priv, dev);
1347 priv_dev_select_rx_function(priv, dev);
1349 err = priv_set_flags(priv, ~IFF_UP, ~IFF_UP);
1352 dev->rx_pkt_burst = removed_rx_burst;
1353 dev->tx_pkt_burst = removed_tx_burst;
1359 * DPDK callback to bring the link DOWN.
1362 * Pointer to Ethernet device structure.
1365 * 0 on success, errno value on failure.
1368 mlx5_set_link_down(struct rte_eth_dev *dev)
1370 struct priv *priv = dev->data->dev_private;
1374 err = priv_dev_set_link(priv, dev, 0);
1380 * DPDK callback to bring the link UP.
1383 * Pointer to Ethernet device structure.
1386 * 0 on success, errno value on failure.
1389 mlx5_set_link_up(struct rte_eth_dev *dev)
1391 struct priv *priv = dev->data->dev_private;
1395 err = priv_dev_set_link(priv, dev, 1);
1401 * Configure the TX function to use.
1404 * Pointer to private data structure.
1406 * Pointer to rte_eth_dev structure.
1409 priv_dev_select_tx_function(struct priv *priv, struct rte_eth_dev *dev)
1411 assert(priv != NULL);
1412 assert(dev != NULL);
1413 dev->tx_pkt_burst = mlx5_tx_burst;
1414 /* Select appropriate TX function. */
1415 if (priv->mps == MLX5_MPW_ENHANCED) {
1416 if (priv_check_vec_tx_support(priv) > 0) {
1417 if (priv_check_raw_vec_tx_support(priv) > 0)
1418 dev->tx_pkt_burst = mlx5_tx_burst_raw_vec;
1420 dev->tx_pkt_burst = mlx5_tx_burst_vec;
1421 DEBUG("selected Enhanced MPW TX vectorized function");
1423 dev->tx_pkt_burst = mlx5_tx_burst_empw;
1424 DEBUG("selected Enhanced MPW TX function");
1426 } else if (priv->mps && priv->txq_inline) {
1427 dev->tx_pkt_burst = mlx5_tx_burst_mpw_inline;
1428 DEBUG("selected MPW inline TX function");
1429 } else if (priv->mps) {
1430 dev->tx_pkt_burst = mlx5_tx_burst_mpw;
1431 DEBUG("selected MPW TX function");
1436 * Configure the RX function to use.
1439 * Pointer to private data structure.
1441 * Pointer to rte_eth_dev structure.
1444 priv_dev_select_rx_function(struct priv *priv, struct rte_eth_dev *dev)
1446 assert(priv != NULL);
1447 assert(dev != NULL);
1448 if (priv_check_vec_rx_support(priv) > 0) {
1449 dev->rx_pkt_burst = mlx5_rx_burst_vec;
1450 DEBUG("selected RX vectorized function");
1452 dev->rx_pkt_burst = mlx5_rx_burst;