1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2008-2017 Cisco Systems, Inc. All rights reserved.
3 * Copyright 2007 Nuova Systems, Inc. All rights reserved.
11 #include <rte_bus_pci.h>
12 #include <rte_ethdev_driver.h>
13 #include <rte_ethdev_pci.h>
14 #include <rte_kvargs.h>
15 #include <rte_string_fns.h>
17 #include "vnic_intr.h"
21 #include "vnic_enet.h"
25 * The set of PCI devices this driver supports
27 #define CISCO_PCI_VENDOR_ID 0x1137
28 static const struct rte_pci_id pci_id_enic_map[] = {
29 {RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET)},
30 {RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_VF)},
31 {RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_SN)},
32 {.vendor_id = 0, /* sentinel */},
35 /* Supported link speeds of production VIC models */
36 static const struct vic_speed_capa {
39 } vic_speed_capa_map[] = {
40 { 0x0043, ETH_LINK_SPEED_10G }, /* VIC */
41 { 0x0047, ETH_LINK_SPEED_10G }, /* P81E PCIe */
42 { 0x0048, ETH_LINK_SPEED_10G }, /* M81KR Mezz */
43 { 0x004f, ETH_LINK_SPEED_10G }, /* 1280 Mezz */
44 { 0x0084, ETH_LINK_SPEED_10G }, /* 1240 MLOM */
45 { 0x0085, ETH_LINK_SPEED_10G }, /* 1225 PCIe */
46 { 0x00cd, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1285 PCIe */
47 { 0x00ce, ETH_LINK_SPEED_10G }, /* 1225T PCIe */
48 { 0x012a, ETH_LINK_SPEED_40G }, /* M4308 */
49 { 0x012c, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1340 MLOM */
50 { 0x012e, ETH_LINK_SPEED_10G }, /* 1227 PCIe */
51 { 0x0137, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1380 Mezz */
52 { 0x014d, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1385 PCIe */
53 { 0x015d, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1387 MLOM */
54 { 0x0215, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
55 ETH_LINK_SPEED_40G }, /* 1440 Mezz */
56 { 0x0216, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
57 ETH_LINK_SPEED_40G }, /* 1480 MLOM */
58 { 0x0217, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G }, /* 1455 PCIe */
59 { 0x0218, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G }, /* 1457 MLOM */
60 { 0x0219, ETH_LINK_SPEED_40G }, /* 1485 PCIe */
61 { 0x021a, ETH_LINK_SPEED_40G }, /* 1487 MLOM */
62 { 0x024a, ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G }, /* 1495 PCIe */
63 { 0x024b, ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G }, /* 1497 MLOM */
64 { 0, 0 }, /* End marker */
67 #define ENIC_DEVARG_DISABLE_OVERLAY "disable-overlay"
68 #define ENIC_DEVARG_ENABLE_AVX2_RX "enable-avx2-rx"
69 #define ENIC_DEVARG_GENEVE_OPT "geneve-opt"
70 #define ENIC_DEVARG_IG_VLAN_REWRITE "ig-vlan-rewrite"
72 RTE_LOG_REGISTER(enic_pmd_logtype, pmd.net.enic, INFO);
75 enicpmd_fdir_ctrl_func(struct rte_eth_dev *eth_dev,
76 enum rte_filter_op filter_op, void *arg)
78 struct enic *enic = pmd_priv(eth_dev);
82 if (filter_op == RTE_ETH_FILTER_NOP)
85 if (arg == NULL && filter_op != RTE_ETH_FILTER_FLUSH)
89 case RTE_ETH_FILTER_ADD:
90 case RTE_ETH_FILTER_UPDATE:
91 ret = enic_fdir_add_fltr(enic,
92 (struct rte_eth_fdir_filter *)arg);
95 case RTE_ETH_FILTER_DELETE:
96 ret = enic_fdir_del_fltr(enic,
97 (struct rte_eth_fdir_filter *)arg);
100 case RTE_ETH_FILTER_STATS:
101 enic_fdir_stats_get(enic, (struct rte_eth_fdir_stats *)arg);
104 case RTE_ETH_FILTER_FLUSH:
105 dev_warning(enic, "unsupported operation %u", filter_op);
108 case RTE_ETH_FILTER_INFO:
109 enic_fdir_info_get(enic, (struct rte_eth_fdir_info *)arg);
112 dev_err(enic, "unknown operation %u", filter_op);
120 enicpmd_dev_filter_ctrl(struct rte_eth_dev *dev,
121 enum rte_filter_type filter_type,
122 enum rte_filter_op filter_op,
125 struct enic *enic = pmd_priv(dev);
128 ENICPMD_FUNC_TRACE();
131 * Currently, when Geneve with options offload is enabled, host
132 * cannot insert match-action rules.
134 if (enic->geneve_opt_enabled)
136 switch (filter_type) {
137 case RTE_ETH_FILTER_GENERIC:
138 if (filter_op != RTE_ETH_FILTER_GET)
140 if (enic->flow_filter_mode == FILTER_FLOWMAN)
141 *(const void **)arg = &enic_fm_flow_ops;
143 *(const void **)arg = &enic_flow_ops;
145 case RTE_ETH_FILTER_FDIR:
146 ret = enicpmd_fdir_ctrl_func(dev, filter_op, arg);
149 dev_warning(enic, "Filter type (%d) not supported",
158 static void enicpmd_dev_tx_queue_release(void *txq)
160 ENICPMD_FUNC_TRACE();
162 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
168 static int enicpmd_dev_setup_intr(struct enic *enic)
173 ENICPMD_FUNC_TRACE();
175 /* Are we done with the init of all the queues? */
176 for (index = 0; index < enic->cq_count; index++) {
177 if (!enic->cq[index].ctrl)
180 if (enic->cq_count != index)
182 for (index = 0; index < enic->wq_count; index++) {
183 if (!enic->wq[index].ctrl)
186 if (enic->wq_count != index)
188 /* check start of packet (SOP) RQs only in case scatter is disabled. */
189 for (index = 0; index < enic->rq_count; index++) {
190 if (!enic->rq[enic_rte_rq_idx_to_sop_idx(index)].ctrl)
193 if (enic->rq_count != index)
196 ret = enic_alloc_intr_resources(enic);
198 dev_err(enic, "alloc intr failed\n");
201 enic_init_vnic_resources(enic);
203 ret = enic_setup_finish(enic);
205 dev_err(enic, "setup could not be finished\n");
210 static int enicpmd_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
213 unsigned int socket_id,
214 const struct rte_eth_txconf *tx_conf)
217 struct enic *enic = pmd_priv(eth_dev);
220 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
221 return -E_RTE_SECONDARY;
223 ENICPMD_FUNC_TRACE();
224 RTE_ASSERT(queue_idx < enic->conf_wq_count);
225 wq = &enic->wq[queue_idx];
226 wq->offloads = tx_conf->offloads |
227 eth_dev->data->dev_conf.txmode.offloads;
228 eth_dev->data->tx_queues[queue_idx] = (void *)wq;
230 ret = enic_alloc_wq(enic, queue_idx, socket_id, nb_desc);
232 dev_err(enic, "error in allocating wq\n");
236 return enicpmd_dev_setup_intr(enic);
239 static int enicpmd_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
242 struct enic *enic = pmd_priv(eth_dev);
244 ENICPMD_FUNC_TRACE();
246 enic_start_wq(enic, queue_idx);
251 static int enicpmd_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
255 struct enic *enic = pmd_priv(eth_dev);
257 ENICPMD_FUNC_TRACE();
259 ret = enic_stop_wq(enic, queue_idx);
261 dev_err(enic, "error in stopping wq %d\n", queue_idx);
266 static int enicpmd_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
269 struct enic *enic = pmd_priv(eth_dev);
271 ENICPMD_FUNC_TRACE();
273 enic_start_rq(enic, queue_idx);
278 static int enicpmd_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
282 struct enic *enic = pmd_priv(eth_dev);
284 ENICPMD_FUNC_TRACE();
286 ret = enic_stop_rq(enic, queue_idx);
288 dev_err(enic, "error in stopping rq %d\n", queue_idx);
293 static void enicpmd_dev_rx_queue_release(void *rxq)
295 ENICPMD_FUNC_TRACE();
297 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
303 static uint32_t enicpmd_dev_rx_queue_count(struct rte_eth_dev *dev,
304 uint16_t rx_queue_id)
306 struct enic *enic = pmd_priv(dev);
307 uint32_t queue_count = 0;
313 rq_num = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
314 cq = &enic->cq[enic_cq_rq(enic, rq_num)];
315 cq_idx = cq->to_clean;
317 cq_tail = ioread32(&cq->ctrl->cq_tail);
319 if (cq_tail < cq_idx)
320 cq_tail += cq->ring.desc_count;
322 queue_count = cq_tail - cq_idx;
327 static int enicpmd_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
330 unsigned int socket_id,
331 const struct rte_eth_rxconf *rx_conf,
332 struct rte_mempool *mp)
335 struct enic *enic = pmd_priv(eth_dev);
337 ENICPMD_FUNC_TRACE();
339 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
340 return -E_RTE_SECONDARY;
341 RTE_ASSERT(enic_rte_rq_idx_to_sop_idx(queue_idx) < enic->conf_rq_count);
342 eth_dev->data->rx_queues[queue_idx] =
343 (void *)&enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
345 ret = enic_alloc_rq(enic, queue_idx, socket_id, mp, nb_desc,
346 rx_conf->rx_free_thresh);
348 dev_err(enic, "error in allocating rq\n");
352 return enicpmd_dev_setup_intr(enic);
355 static int enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
357 struct enic *enic = pmd_priv(eth_dev);
360 ENICPMD_FUNC_TRACE();
362 offloads = eth_dev->data->dev_conf.rxmode.offloads;
363 if (mask & ETH_VLAN_STRIP_MASK) {
364 if (offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
365 enic->ig_vlan_strip_en = 1;
367 enic->ig_vlan_strip_en = 0;
370 return enic_set_vlan_strip(enic);
373 static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
377 struct enic *enic = pmd_priv(eth_dev);
379 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
380 return -E_RTE_SECONDARY;
382 ENICPMD_FUNC_TRACE();
383 ret = enic_set_vnic_res(enic);
385 dev_err(enic, "Set vNIC resource num failed, aborting\n");
389 if (eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
390 eth_dev->data->dev_conf.rxmode.offloads |=
391 DEV_RX_OFFLOAD_RSS_HASH;
394 enic->hw_ip_checksum = !!(eth_dev->data->dev_conf.rxmode.offloads &
395 DEV_RX_OFFLOAD_CHECKSUM);
396 /* All vlan offload masks to apply the current settings */
397 mask = ETH_VLAN_STRIP_MASK |
398 ETH_VLAN_FILTER_MASK |
399 ETH_VLAN_EXTEND_MASK;
400 ret = enicpmd_vlan_offload_set(eth_dev, mask);
402 dev_err(enic, "Failed to configure VLAN offloads\n");
406 * Initialize RSS with the default reta and key. If the user key is
407 * given (rx_adv_conf.rss_conf.rss_key), will use that instead of the
410 return enic_init_rss_nic_cfg(enic);
414 * It returns 0 on success.
416 static int enicpmd_dev_start(struct rte_eth_dev *eth_dev)
418 struct enic *enic = pmd_priv(eth_dev);
420 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
421 return -E_RTE_SECONDARY;
423 ENICPMD_FUNC_TRACE();
424 return enic_enable(enic);
428 * Stop device: disable rx and tx functions to allow for reconfiguring.
430 static void enicpmd_dev_stop(struct rte_eth_dev *eth_dev)
432 struct rte_eth_link link;
433 struct enic *enic = pmd_priv(eth_dev);
435 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
438 ENICPMD_FUNC_TRACE();
441 memset(&link, 0, sizeof(link));
442 rte_eth_linkstatus_set(eth_dev, &link);
448 static void enicpmd_dev_close(struct rte_eth_dev *eth_dev)
450 struct enic *enic = pmd_priv(eth_dev);
452 ENICPMD_FUNC_TRACE();
456 static int enicpmd_dev_link_update(struct rte_eth_dev *eth_dev,
457 __rte_unused int wait_to_complete)
459 ENICPMD_FUNC_TRACE();
460 return enic_link_update(eth_dev);
463 static int enicpmd_dev_stats_get(struct rte_eth_dev *eth_dev,
464 struct rte_eth_stats *stats)
466 struct enic *enic = pmd_priv(eth_dev);
468 ENICPMD_FUNC_TRACE();
469 return enic_dev_stats_get(enic, stats);
472 static int enicpmd_dev_stats_reset(struct rte_eth_dev *eth_dev)
474 struct enic *enic = pmd_priv(eth_dev);
476 ENICPMD_FUNC_TRACE();
477 return enic_dev_stats_clear(enic);
480 static uint32_t speed_capa_from_pci_id(struct rte_eth_dev *eth_dev)
482 const struct vic_speed_capa *m;
483 struct rte_pci_device *pdev;
486 pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
487 id = pdev->id.subsystem_device_id;
488 for (m = vic_speed_capa_map; m->sub_devid != 0; m++) {
489 if (m->sub_devid == id)
492 /* 1300 and later models are at least 40G */
494 return ETH_LINK_SPEED_40G;
495 /* VFs have subsystem id 0, check device id */
497 /* Newer VF implies at least 40G model */
498 if (pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_SN)
499 return ETH_LINK_SPEED_40G;
501 return ETH_LINK_SPEED_10G;
504 static int enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
505 struct rte_eth_dev_info *device_info)
507 struct enic *enic = pmd_priv(eth_dev);
509 ENICPMD_FUNC_TRACE();
510 /* Scattered Rx uses two receive queues per rx queue exposed to dpdk */
511 device_info->max_rx_queues = enic->conf_rq_count / 2;
512 device_info->max_tx_queues = enic->conf_wq_count;
513 device_info->min_rx_bufsize = ENIC_MIN_MTU;
514 /* "Max" mtu is not a typo. HW receives packet sizes up to the
515 * max mtu regardless of the current mtu (vNIC's mtu). vNIC mtu is
516 * a hint to the driver to size receive buffers accordingly so that
517 * larger-than-vnic-mtu packets get truncated.. For DPDK, we let
518 * the user decide the buffer size via rxmode.max_rx_pkt_len, basically
521 device_info->max_rx_pktlen = enic_mtu_to_max_rx_pktlen(enic->max_mtu);
522 device_info->max_mac_addrs = ENIC_UNICAST_PERFECT_FILTERS;
523 device_info->min_mtu = ENIC_MIN_MTU;
524 device_info->max_mtu = enic->max_mtu;
525 device_info->rx_offload_capa = enic->rx_offload_capa;
526 device_info->tx_offload_capa = enic->tx_offload_capa;
527 device_info->tx_queue_offload_capa = enic->tx_queue_offload_capa;
528 device_info->default_rxconf = (struct rte_eth_rxconf) {
529 .rx_free_thresh = ENIC_DEFAULT_RX_FREE_THRESH
531 device_info->reta_size = enic->reta_size;
532 device_info->hash_key_size = enic->hash_key_size;
533 device_info->flow_type_rss_offloads = enic->flow_type_rss_offloads;
534 device_info->rx_desc_lim = (struct rte_eth_desc_lim) {
535 .nb_max = enic->config.rq_desc_count,
536 .nb_min = ENIC_MIN_RQ_DESCS,
537 .nb_align = ENIC_ALIGN_DESCS,
539 device_info->tx_desc_lim = (struct rte_eth_desc_lim) {
540 .nb_max = enic->config.wq_desc_count,
541 .nb_min = ENIC_MIN_WQ_DESCS,
542 .nb_align = ENIC_ALIGN_DESCS,
543 .nb_seg_max = ENIC_TX_XMIT_MAX,
544 .nb_mtu_seg_max = ENIC_NON_TSO_MAX_DESC,
546 device_info->default_rxportconf = (struct rte_eth_dev_portconf) {
547 .burst_size = ENIC_DEFAULT_RX_BURST,
548 .ring_size = RTE_MIN(device_info->rx_desc_lim.nb_max,
549 ENIC_DEFAULT_RX_RING_SIZE),
550 .nb_queues = ENIC_DEFAULT_RX_RINGS,
552 device_info->default_txportconf = (struct rte_eth_dev_portconf) {
553 .burst_size = ENIC_DEFAULT_TX_BURST,
554 .ring_size = RTE_MIN(device_info->tx_desc_lim.nb_max,
555 ENIC_DEFAULT_TX_RING_SIZE),
556 .nb_queues = ENIC_DEFAULT_TX_RINGS,
558 device_info->speed_capa = speed_capa_from_pci_id(eth_dev);
563 static const uint32_t *enicpmd_dev_supported_ptypes_get(struct rte_eth_dev *dev)
565 static const uint32_t ptypes[] = {
567 RTE_PTYPE_L2_ETHER_VLAN,
568 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
569 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
573 RTE_PTYPE_L4_NONFRAG,
576 static const uint32_t ptypes_overlay[] = {
578 RTE_PTYPE_L2_ETHER_VLAN,
579 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
580 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
584 RTE_PTYPE_L4_NONFRAG,
585 RTE_PTYPE_TUNNEL_GRENAT,
586 RTE_PTYPE_INNER_L2_ETHER,
587 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
588 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
589 RTE_PTYPE_INNER_L4_TCP,
590 RTE_PTYPE_INNER_L4_UDP,
591 RTE_PTYPE_INNER_L4_FRAG,
592 RTE_PTYPE_INNER_L4_NONFRAG,
596 if (dev->rx_pkt_burst != enic_dummy_recv_pkts &&
597 dev->rx_pkt_burst != NULL) {
598 struct enic *enic = pmd_priv(dev);
599 if (enic->overlay_offload)
600 return ptypes_overlay;
607 static int enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
609 struct enic *enic = pmd_priv(eth_dev);
612 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
613 return -E_RTE_SECONDARY;
615 ENICPMD_FUNC_TRACE();
618 ret = enic_add_packet_filter(enic);
625 static int enicpmd_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
627 struct enic *enic = pmd_priv(eth_dev);
630 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
631 return -E_RTE_SECONDARY;
633 ENICPMD_FUNC_TRACE();
635 ret = enic_add_packet_filter(enic);
642 static int enicpmd_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
644 struct enic *enic = pmd_priv(eth_dev);
647 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
648 return -E_RTE_SECONDARY;
650 ENICPMD_FUNC_TRACE();
652 ret = enic_add_packet_filter(enic);
659 static int enicpmd_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
661 struct enic *enic = pmd_priv(eth_dev);
664 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
665 return -E_RTE_SECONDARY;
667 ENICPMD_FUNC_TRACE();
669 ret = enic_add_packet_filter(enic);
676 static int enicpmd_add_mac_addr(struct rte_eth_dev *eth_dev,
677 struct rte_ether_addr *mac_addr,
678 __rte_unused uint32_t index, __rte_unused uint32_t pool)
680 struct enic *enic = pmd_priv(eth_dev);
682 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
683 return -E_RTE_SECONDARY;
685 ENICPMD_FUNC_TRACE();
686 return enic_set_mac_address(enic, mac_addr->addr_bytes);
689 static void enicpmd_remove_mac_addr(struct rte_eth_dev *eth_dev, uint32_t index)
691 struct enic *enic = pmd_priv(eth_dev);
693 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
696 ENICPMD_FUNC_TRACE();
697 if (enic_del_mac_address(enic, index))
698 dev_err(enic, "del mac addr failed\n");
701 static int enicpmd_set_mac_addr(struct rte_eth_dev *eth_dev,
702 struct rte_ether_addr *addr)
704 struct enic *enic = pmd_priv(eth_dev);
707 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
708 return -E_RTE_SECONDARY;
710 ENICPMD_FUNC_TRACE();
711 ret = enic_del_mac_address(enic, 0);
714 return enic_set_mac_address(enic, addr->addr_bytes);
717 static void debug_log_add_del_addr(struct rte_ether_addr *addr, bool add)
719 char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
721 rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE, addr);
722 ENICPMD_LOG(DEBUG, " %s address %s\n",
723 add ? "add" : "remove", mac_str);
726 static int enicpmd_set_mc_addr_list(struct rte_eth_dev *eth_dev,
727 struct rte_ether_addr *mc_addr_set,
730 struct enic *enic = pmd_priv(eth_dev);
731 char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
732 struct rte_ether_addr *addr;
736 ENICPMD_FUNC_TRACE();
738 /* Validate the given addresses first */
739 for (i = 0; i < nb_mc_addr && mc_addr_set != NULL; i++) {
740 addr = &mc_addr_set[i];
741 if (!rte_is_multicast_ether_addr(addr) ||
742 rte_is_broadcast_ether_addr(addr)) {
743 rte_ether_format_addr(mac_str,
744 RTE_ETHER_ADDR_FMT_SIZE, addr);
745 ENICPMD_LOG(ERR, " invalid multicast address %s\n",
751 /* Flush all if requested */
752 if (nb_mc_addr == 0 || mc_addr_set == NULL) {
753 ENICPMD_LOG(DEBUG, " flush multicast addresses\n");
754 for (i = 0; i < enic->mc_count; i++) {
755 addr = &enic->mc_addrs[i];
756 debug_log_add_del_addr(addr, false);
757 ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
765 if (nb_mc_addr > ENIC_MULTICAST_PERFECT_FILTERS) {
766 ENICPMD_LOG(ERR, " too many multicast addresses: max=%d\n",
767 ENIC_MULTICAST_PERFECT_FILTERS);
771 * devcmd is slow, so apply the difference instead of flushing and
773 * 1. Delete addresses on the NIC but not on the host
775 for (i = 0; i < enic->mc_count; i++) {
776 addr = &enic->mc_addrs[i];
777 for (j = 0; j < nb_mc_addr; j++) {
778 if (rte_is_same_ether_addr(addr, &mc_addr_set[j]))
783 debug_log_add_del_addr(addr, false);
784 ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
788 /* 2. Add addresses on the host but not on the NIC */
789 for (i = 0; i < nb_mc_addr; i++) {
790 addr = &mc_addr_set[i];
791 for (j = 0; j < enic->mc_count; j++) {
792 if (rte_is_same_ether_addr(addr, &enic->mc_addrs[j]))
795 if (j < enic->mc_count)
797 debug_log_add_del_addr(addr, true);
798 ret = vnic_dev_add_addr(enic->vdev, addr->addr_bytes);
802 /* Keep a copy so we can flush/apply later on.. */
803 memcpy(enic->mc_addrs, mc_addr_set,
804 nb_mc_addr * sizeof(struct rte_ether_addr));
805 enic->mc_count = nb_mc_addr;
809 static int enicpmd_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
811 struct enic *enic = pmd_priv(eth_dev);
813 ENICPMD_FUNC_TRACE();
814 return enic_set_mtu(enic, mtu);
817 static int enicpmd_dev_rss_reta_query(struct rte_eth_dev *dev,
818 struct rte_eth_rss_reta_entry64
822 struct enic *enic = pmd_priv(dev);
823 uint16_t i, idx, shift;
825 ENICPMD_FUNC_TRACE();
826 if (reta_size != ENIC_RSS_RETA_SIZE) {
827 dev_err(enic, "reta_query: wrong reta_size. given=%u expected=%u\n",
828 reta_size, ENIC_RSS_RETA_SIZE);
832 for (i = 0; i < reta_size; i++) {
833 idx = i / RTE_RETA_GROUP_SIZE;
834 shift = i % RTE_RETA_GROUP_SIZE;
835 if (reta_conf[idx].mask & (1ULL << shift))
836 reta_conf[idx].reta[shift] = enic_sop_rq_idx_to_rte_idx(
837 enic->rss_cpu.cpu[i / 4].b[i % 4]);
843 static int enicpmd_dev_rss_reta_update(struct rte_eth_dev *dev,
844 struct rte_eth_rss_reta_entry64
848 struct enic *enic = pmd_priv(dev);
849 union vnic_rss_cpu rss_cpu;
850 uint16_t i, idx, shift;
852 ENICPMD_FUNC_TRACE();
853 if (reta_size != ENIC_RSS_RETA_SIZE) {
854 dev_err(enic, "reta_update: wrong reta_size. given=%u"
856 reta_size, ENIC_RSS_RETA_SIZE);
860 * Start with the current reta and modify it per reta_conf, as we
861 * need to push the entire reta even if we only modify one entry.
863 rss_cpu = enic->rss_cpu;
864 for (i = 0; i < reta_size; i++) {
865 idx = i / RTE_RETA_GROUP_SIZE;
866 shift = i % RTE_RETA_GROUP_SIZE;
867 if (reta_conf[idx].mask & (1ULL << shift))
868 rss_cpu.cpu[i / 4].b[i % 4] =
869 enic_rte_rq_idx_to_sop_idx(
870 reta_conf[idx].reta[shift]);
872 return enic_set_rss_reta(enic, &rss_cpu);
875 static int enicpmd_dev_rss_hash_update(struct rte_eth_dev *dev,
876 struct rte_eth_rss_conf *rss_conf)
878 struct enic *enic = pmd_priv(dev);
880 ENICPMD_FUNC_TRACE();
881 return enic_set_rss_conf(enic, rss_conf);
884 static int enicpmd_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
885 struct rte_eth_rss_conf *rss_conf)
887 struct enic *enic = pmd_priv(dev);
889 ENICPMD_FUNC_TRACE();
890 if (rss_conf == NULL)
892 if (rss_conf->rss_key != NULL &&
893 rss_conf->rss_key_len < ENIC_RSS_HASH_KEY_SIZE) {
894 dev_err(enic, "rss_hash_conf_get: wrong rss_key_len. given=%u"
896 rss_conf->rss_key_len, ENIC_RSS_HASH_KEY_SIZE);
899 rss_conf->rss_hf = enic->rss_hf;
900 if (rss_conf->rss_key != NULL) {
902 for (i = 0; i < ENIC_RSS_HASH_KEY_SIZE; i++) {
903 rss_conf->rss_key[i] =
904 enic->rss_key.key[i / 10].b[i % 10];
906 rss_conf->rss_key_len = ENIC_RSS_HASH_KEY_SIZE;
911 static void enicpmd_dev_rxq_info_get(struct rte_eth_dev *dev,
912 uint16_t rx_queue_id,
913 struct rte_eth_rxq_info *qinfo)
915 struct enic *enic = pmd_priv(dev);
916 struct vnic_rq *rq_sop;
917 struct vnic_rq *rq_data;
918 struct rte_eth_rxconf *conf;
919 uint16_t sop_queue_idx;
920 uint16_t data_queue_idx;
922 ENICPMD_FUNC_TRACE();
923 sop_queue_idx = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
924 data_queue_idx = enic_rte_rq_idx_to_data_idx(rx_queue_id, enic);
925 rq_sop = &enic->rq[sop_queue_idx];
926 rq_data = &enic->rq[data_queue_idx]; /* valid if data_queue_enable */
927 qinfo->mp = rq_sop->mp;
928 qinfo->scattered_rx = rq_sop->data_queue_enable;
929 qinfo->nb_desc = rq_sop->ring.desc_count;
930 if (qinfo->scattered_rx)
931 qinfo->nb_desc += rq_data->ring.desc_count;
933 memset(conf, 0, sizeof(*conf));
934 conf->rx_free_thresh = rq_sop->rx_free_thresh;
935 conf->rx_drop_en = 1;
937 * Except VLAN stripping (port setting), all the checksum offloads
938 * are always enabled.
940 conf->offloads = enic->rx_offload_capa;
941 if (!enic->ig_vlan_strip_en)
942 conf->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
943 /* rx_thresh and other fields are not applicable for enic */
946 static void enicpmd_dev_txq_info_get(struct rte_eth_dev *dev,
947 uint16_t tx_queue_id,
948 struct rte_eth_txq_info *qinfo)
950 struct enic *enic = pmd_priv(dev);
951 struct vnic_wq *wq = &enic->wq[tx_queue_id];
953 ENICPMD_FUNC_TRACE();
954 qinfo->nb_desc = wq->ring.desc_count;
955 memset(&qinfo->conf, 0, sizeof(qinfo->conf));
956 qinfo->conf.offloads = wq->offloads;
957 /* tx_thresh, and all the other fields are not applicable for enic */
960 static int enicpmd_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev,
961 uint16_t rx_queue_id)
963 struct enic *enic = pmd_priv(eth_dev);
965 ENICPMD_FUNC_TRACE();
966 vnic_intr_unmask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
970 static int enicpmd_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev,
971 uint16_t rx_queue_id)
973 struct enic *enic = pmd_priv(eth_dev);
975 ENICPMD_FUNC_TRACE();
976 vnic_intr_mask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
980 static int udp_tunnel_common_check(struct enic *enic,
981 struct rte_eth_udp_tunnel *tnl)
983 if (tnl->prot_type != RTE_TUNNEL_TYPE_VXLAN)
985 if (!enic->overlay_offload) {
986 ENICPMD_LOG(DEBUG, " vxlan (overlay offload) is not "
993 static int update_vxlan_port(struct enic *enic, uint16_t port)
995 if (vnic_dev_overlay_offload_cfg(enic->vdev,
996 OVERLAY_CFG_VXLAN_PORT_UPDATE,
998 ENICPMD_LOG(DEBUG, " failed to update vxlan port\n");
1001 ENICPMD_LOG(DEBUG, " updated vxlan port to %u\n", port);
1002 enic->vxlan_port = port;
1006 static int enicpmd_dev_udp_tunnel_port_add(struct rte_eth_dev *eth_dev,
1007 struct rte_eth_udp_tunnel *tnl)
1009 struct enic *enic = pmd_priv(eth_dev);
1012 ENICPMD_FUNC_TRACE();
1013 ret = udp_tunnel_common_check(enic, tnl);
1017 * The NIC has 1 configurable VXLAN port number. "Adding" a new port
1018 * number replaces it.
1020 if (tnl->udp_port == enic->vxlan_port || tnl->udp_port == 0) {
1021 ENICPMD_LOG(DEBUG, " %u is already configured or invalid\n",
1025 return update_vxlan_port(enic, tnl->udp_port);
1028 static int enicpmd_dev_udp_tunnel_port_del(struct rte_eth_dev *eth_dev,
1029 struct rte_eth_udp_tunnel *tnl)
1031 struct enic *enic = pmd_priv(eth_dev);
1034 ENICPMD_FUNC_TRACE();
1035 ret = udp_tunnel_common_check(enic, tnl);
1039 * Clear the previously set port number and restore the
1040 * hardware default port number. Some drivers disable VXLAN
1041 * offloads when there are no configured port numbers. But
1042 * enic does not do that as VXLAN is part of overlay offload,
1043 * which is tied to inner RSS and TSO.
1045 if (tnl->udp_port != enic->vxlan_port) {
1046 ENICPMD_LOG(DEBUG, " %u is not a configured vxlan port\n",
1050 return update_vxlan_port(enic, RTE_VXLAN_DEFAULT_PORT);
1053 static int enicpmd_dev_fw_version_get(struct rte_eth_dev *eth_dev,
1054 char *fw_version, size_t fw_size)
1056 struct vnic_devcmd_fw_info *info;
1060 ENICPMD_FUNC_TRACE();
1061 if (fw_version == NULL || fw_size <= 0)
1063 enic = pmd_priv(eth_dev);
1064 ret = vnic_dev_fw_info(enic->vdev, &info);
1067 snprintf(fw_version, fw_size, "%s %s",
1068 info->fw_version, info->fw_build);
1069 fw_version[fw_size - 1] = '\0';
1073 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
1074 .dev_configure = enicpmd_dev_configure,
1075 .dev_start = enicpmd_dev_start,
1076 .dev_stop = enicpmd_dev_stop,
1077 .dev_set_link_up = NULL,
1078 .dev_set_link_down = NULL,
1079 .dev_close = enicpmd_dev_close,
1080 .promiscuous_enable = enicpmd_dev_promiscuous_enable,
1081 .promiscuous_disable = enicpmd_dev_promiscuous_disable,
1082 .allmulticast_enable = enicpmd_dev_allmulticast_enable,
1083 .allmulticast_disable = enicpmd_dev_allmulticast_disable,
1084 .link_update = enicpmd_dev_link_update,
1085 .stats_get = enicpmd_dev_stats_get,
1086 .stats_reset = enicpmd_dev_stats_reset,
1087 .queue_stats_mapping_set = NULL,
1088 .dev_infos_get = enicpmd_dev_info_get,
1089 .dev_supported_ptypes_get = enicpmd_dev_supported_ptypes_get,
1090 .mtu_set = enicpmd_mtu_set,
1091 .vlan_filter_set = NULL,
1092 .vlan_tpid_set = NULL,
1093 .vlan_offload_set = enicpmd_vlan_offload_set,
1094 .vlan_strip_queue_set = NULL,
1095 .rx_queue_start = enicpmd_dev_rx_queue_start,
1096 .rx_queue_stop = enicpmd_dev_rx_queue_stop,
1097 .tx_queue_start = enicpmd_dev_tx_queue_start,
1098 .tx_queue_stop = enicpmd_dev_tx_queue_stop,
1099 .rx_queue_setup = enicpmd_dev_rx_queue_setup,
1100 .rx_queue_release = enicpmd_dev_rx_queue_release,
1101 .rx_queue_count = enicpmd_dev_rx_queue_count,
1102 .rx_descriptor_done = NULL,
1103 .tx_queue_setup = enicpmd_dev_tx_queue_setup,
1104 .tx_queue_release = enicpmd_dev_tx_queue_release,
1105 .rx_queue_intr_enable = enicpmd_dev_rx_queue_intr_enable,
1106 .rx_queue_intr_disable = enicpmd_dev_rx_queue_intr_disable,
1107 .rxq_info_get = enicpmd_dev_rxq_info_get,
1108 .txq_info_get = enicpmd_dev_txq_info_get,
1110 .dev_led_off = NULL,
1111 .flow_ctrl_get = NULL,
1112 .flow_ctrl_set = NULL,
1113 .priority_flow_ctrl_set = NULL,
1114 .mac_addr_add = enicpmd_add_mac_addr,
1115 .mac_addr_remove = enicpmd_remove_mac_addr,
1116 .mac_addr_set = enicpmd_set_mac_addr,
1117 .set_mc_addr_list = enicpmd_set_mc_addr_list,
1118 .filter_ctrl = enicpmd_dev_filter_ctrl,
1119 .reta_query = enicpmd_dev_rss_reta_query,
1120 .reta_update = enicpmd_dev_rss_reta_update,
1121 .rss_hash_conf_get = enicpmd_dev_rss_hash_conf_get,
1122 .rss_hash_update = enicpmd_dev_rss_hash_update,
1123 .udp_tunnel_port_add = enicpmd_dev_udp_tunnel_port_add,
1124 .udp_tunnel_port_del = enicpmd_dev_udp_tunnel_port_del,
1125 .fw_version_get = enicpmd_dev_fw_version_get,
1128 static int enic_parse_zero_one(const char *key,
1135 enic = (struct enic *)opaque;
1136 if (strcmp(value, "0") == 0) {
1138 } else if (strcmp(value, "1") == 0) {
1141 dev_err(enic, "Invalid value for %s"
1142 ": expected=0|1 given=%s\n", key, value);
1145 if (strcmp(key, ENIC_DEVARG_DISABLE_OVERLAY) == 0)
1146 enic->disable_overlay = b;
1147 if (strcmp(key, ENIC_DEVARG_ENABLE_AVX2_RX) == 0)
1148 enic->enable_avx2_rx = b;
1149 if (strcmp(key, ENIC_DEVARG_GENEVE_OPT) == 0)
1150 enic->geneve_opt_request = b;
1154 static int enic_parse_ig_vlan_rewrite(__rte_unused const char *key,
1160 enic = (struct enic *)opaque;
1161 if (strcmp(value, "trunk") == 0) {
1162 /* Trunk mode: always tag */
1163 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_DEFAULT_TRUNK;
1164 } else if (strcmp(value, "untag") == 0) {
1165 /* Untag default VLAN mode: untag if VLAN = default VLAN */
1166 enic->ig_vlan_rewrite_mode =
1167 IG_VLAN_REWRITE_MODE_UNTAG_DEFAULT_VLAN;
1168 } else if (strcmp(value, "priority") == 0) {
1170 * Priority-tag default VLAN mode: priority tag (VLAN header
1171 * with ID=0) if VLAN = default
1173 enic->ig_vlan_rewrite_mode =
1174 IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN;
1175 } else if (strcmp(value, "pass") == 0) {
1176 /* Pass through mode: do not touch tags */
1177 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1179 dev_err(enic, "Invalid value for " ENIC_DEVARG_IG_VLAN_REWRITE
1180 ": expected=trunk|untag|priority|pass given=%s\n",
1187 static int enic_check_devargs(struct rte_eth_dev *dev)
1189 static const char *const valid_keys[] = {
1190 ENIC_DEVARG_DISABLE_OVERLAY,
1191 ENIC_DEVARG_ENABLE_AVX2_RX,
1192 ENIC_DEVARG_GENEVE_OPT,
1193 ENIC_DEVARG_IG_VLAN_REWRITE,
1195 struct enic *enic = pmd_priv(dev);
1196 struct rte_kvargs *kvlist;
1198 ENICPMD_FUNC_TRACE();
1200 enic->disable_overlay = false;
1201 enic->enable_avx2_rx = false;
1202 enic->geneve_opt_request = false;
1203 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1204 if (!dev->device->devargs)
1206 kvlist = rte_kvargs_parse(dev->device->devargs->args, valid_keys);
1209 if (rte_kvargs_process(kvlist, ENIC_DEVARG_DISABLE_OVERLAY,
1210 enic_parse_zero_one, enic) < 0 ||
1211 rte_kvargs_process(kvlist, ENIC_DEVARG_ENABLE_AVX2_RX,
1212 enic_parse_zero_one, enic) < 0 ||
1213 rte_kvargs_process(kvlist, ENIC_DEVARG_GENEVE_OPT,
1214 enic_parse_zero_one, enic) < 0 ||
1215 rte_kvargs_process(kvlist, ENIC_DEVARG_IG_VLAN_REWRITE,
1216 enic_parse_ig_vlan_rewrite, enic) < 0) {
1217 rte_kvargs_free(kvlist);
1220 rte_kvargs_free(kvlist);
1224 /* Initialize the driver
1225 * It returns 0 on success.
1227 static int eth_enicpmd_dev_init(struct rte_eth_dev *eth_dev)
1229 struct rte_pci_device *pdev;
1230 struct rte_pci_addr *addr;
1231 struct enic *enic = pmd_priv(eth_dev);
1234 ENICPMD_FUNC_TRACE();
1236 eth_dev->dev_ops = &enicpmd_eth_dev_ops;
1237 eth_dev->rx_pkt_burst = &enic_recv_pkts;
1238 eth_dev->tx_pkt_burst = &enic_xmit_pkts;
1239 eth_dev->tx_pkt_prepare = &enic_prep_pkts;
1240 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1241 enic_pick_tx_handler(eth_dev);
1242 enic_pick_rx_handler(eth_dev);
1245 /* Only the primary sets up adapter and other data in shared memory */
1246 enic->port_id = eth_dev->data->port_id;
1247 enic->rte_dev = eth_dev;
1248 enic->dev_data = eth_dev->data;
1249 /* Let rte_eth_dev_close() release the port resources */
1250 eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1252 pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
1253 rte_eth_copy_pci_info(eth_dev, pdev);
1257 snprintf(enic->bdf_name, ENICPMD_BDF_LENGTH, "%04x:%02x:%02x.%x",
1258 addr->domain, addr->bus, addr->devid, addr->function);
1260 err = enic_check_devargs(eth_dev);
1263 return enic_probe(enic);
1266 static int eth_enic_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1267 struct rte_pci_device *pci_dev)
1269 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct enic),
1270 eth_enicpmd_dev_init);
1273 static int eth_enic_pci_remove(struct rte_pci_device *pci_dev)
1275 return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1278 static struct rte_pci_driver rte_enic_pmd = {
1279 .id_table = pci_id_enic_map,
1280 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1281 .probe = eth_enic_pci_probe,
1282 .remove = eth_enic_pci_remove,
1285 int dev_is_enic(struct rte_eth_dev *dev)
1287 return dev->device->driver == &rte_enic_pmd.driver;
1290 RTE_PMD_REGISTER_PCI(net_enic, rte_enic_pmd);
1291 RTE_PMD_REGISTER_PCI_TABLE(net_enic, pci_id_enic_map);
1292 RTE_PMD_REGISTER_KMOD_DEP(net_enic, "* igb_uio | uio_pci_generic | vfio-pci");
1293 RTE_PMD_REGISTER_PARAM_STRING(net_enic,
1294 ENIC_DEVARG_DISABLE_OVERLAY "=0|1 "
1295 ENIC_DEVARG_ENABLE_AVX2_RX "=0|1 "
1296 ENIC_DEVARG_GENEVE_OPT "=0|1 "
1297 ENIC_DEVARG_IG_VLAN_REWRITE "=trunk|untag|priority|pass");