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"
24 int enicpmd_logtype_init;
25 int enicpmd_logtype_flow;
27 #define ENICPMD_FUNC_TRACE() PMD_INIT_LOG(DEBUG, " >>")
30 * The set of PCI devices this driver supports
32 #define CISCO_PCI_VENDOR_ID 0x1137
33 static const struct rte_pci_id pci_id_enic_map[] = {
34 { RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET) },
35 { RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
36 {.vendor_id = 0, /* sentinel */},
39 #define ENIC_DEVARG_DISABLE_OVERLAY "disable-overlay"
40 #define ENIC_DEVARG_ENABLE_AVX2_RX "enable-avx2-rx"
41 #define ENIC_DEVARG_IG_VLAN_REWRITE "ig-vlan-rewrite"
43 RTE_INIT(enicpmd_init_log)
45 enicpmd_logtype_init = rte_log_register("pmd.net.enic.init");
46 if (enicpmd_logtype_init >= 0)
47 rte_log_set_level(enicpmd_logtype_init, RTE_LOG_NOTICE);
48 enicpmd_logtype_flow = rte_log_register("pmd.net.enic.flow");
49 if (enicpmd_logtype_flow >= 0)
50 rte_log_set_level(enicpmd_logtype_flow, RTE_LOG_NOTICE);
54 enicpmd_fdir_ctrl_func(struct rte_eth_dev *eth_dev,
55 enum rte_filter_op filter_op, void *arg)
57 struct enic *enic = pmd_priv(eth_dev);
61 if (filter_op == RTE_ETH_FILTER_NOP)
64 if (arg == NULL && filter_op != RTE_ETH_FILTER_FLUSH)
68 case RTE_ETH_FILTER_ADD:
69 case RTE_ETH_FILTER_UPDATE:
70 ret = enic_fdir_add_fltr(enic,
71 (struct rte_eth_fdir_filter *)arg);
74 case RTE_ETH_FILTER_DELETE:
75 ret = enic_fdir_del_fltr(enic,
76 (struct rte_eth_fdir_filter *)arg);
79 case RTE_ETH_FILTER_STATS:
80 enic_fdir_stats_get(enic, (struct rte_eth_fdir_stats *)arg);
83 case RTE_ETH_FILTER_FLUSH:
84 dev_warning(enic, "unsupported operation %u", filter_op);
87 case RTE_ETH_FILTER_INFO:
88 enic_fdir_info_get(enic, (struct rte_eth_fdir_info *)arg);
91 dev_err(enic, "unknown operation %u", filter_op);
99 enicpmd_dev_filter_ctrl(struct rte_eth_dev *dev,
100 enum rte_filter_type filter_type,
101 enum rte_filter_op filter_op,
106 ENICPMD_FUNC_TRACE();
108 switch (filter_type) {
109 case RTE_ETH_FILTER_GENERIC:
110 if (filter_op != RTE_ETH_FILTER_GET)
112 *(const void **)arg = &enic_flow_ops;
114 case RTE_ETH_FILTER_FDIR:
115 ret = enicpmd_fdir_ctrl_func(dev, filter_op, arg);
118 dev_warning(enic, "Filter type (%d) not supported",
127 static void enicpmd_dev_tx_queue_release(void *txq)
129 ENICPMD_FUNC_TRACE();
131 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
137 static int enicpmd_dev_setup_intr(struct enic *enic)
142 ENICPMD_FUNC_TRACE();
144 /* Are we done with the init of all the queues? */
145 for (index = 0; index < enic->cq_count; index++) {
146 if (!enic->cq[index].ctrl)
149 if (enic->cq_count != index)
151 for (index = 0; index < enic->wq_count; index++) {
152 if (!enic->wq[index].ctrl)
155 if (enic->wq_count != index)
157 /* check start of packet (SOP) RQs only in case scatter is disabled. */
158 for (index = 0; index < enic->rq_count; index++) {
159 if (!enic->rq[enic_rte_rq_idx_to_sop_idx(index)].ctrl)
162 if (enic->rq_count != index)
165 ret = enic_alloc_intr_resources(enic);
167 dev_err(enic, "alloc intr failed\n");
170 enic_init_vnic_resources(enic);
172 ret = enic_setup_finish(enic);
174 dev_err(enic, "setup could not be finished\n");
179 static int enicpmd_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
182 unsigned int socket_id,
183 const struct rte_eth_txconf *tx_conf)
186 struct enic *enic = pmd_priv(eth_dev);
189 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
190 return -E_RTE_SECONDARY;
192 ENICPMD_FUNC_TRACE();
193 RTE_ASSERT(queue_idx < enic->conf_wq_count);
194 wq = &enic->wq[queue_idx];
195 wq->offloads = tx_conf->offloads |
196 eth_dev->data->dev_conf.txmode.offloads;
197 eth_dev->data->tx_queues[queue_idx] = (void *)wq;
199 ret = enic_alloc_wq(enic, queue_idx, socket_id, nb_desc);
201 dev_err(enic, "error in allocating wq\n");
205 return enicpmd_dev_setup_intr(enic);
208 static int enicpmd_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
211 struct enic *enic = pmd_priv(eth_dev);
213 ENICPMD_FUNC_TRACE();
215 enic_start_wq(enic, queue_idx);
220 static int enicpmd_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
224 struct enic *enic = pmd_priv(eth_dev);
226 ENICPMD_FUNC_TRACE();
228 ret = enic_stop_wq(enic, queue_idx);
230 dev_err(enic, "error in stopping wq %d\n", queue_idx);
235 static int enicpmd_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
238 struct enic *enic = pmd_priv(eth_dev);
240 ENICPMD_FUNC_TRACE();
242 enic_start_rq(enic, queue_idx);
247 static int enicpmd_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
251 struct enic *enic = pmd_priv(eth_dev);
253 ENICPMD_FUNC_TRACE();
255 ret = enic_stop_rq(enic, queue_idx);
257 dev_err(enic, "error in stopping rq %d\n", queue_idx);
262 static void enicpmd_dev_rx_queue_release(void *rxq)
264 ENICPMD_FUNC_TRACE();
266 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
272 static uint32_t enicpmd_dev_rx_queue_count(struct rte_eth_dev *dev,
273 uint16_t rx_queue_id)
275 struct enic *enic = pmd_priv(dev);
276 uint32_t queue_count = 0;
282 rq_num = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
283 cq = &enic->cq[enic_cq_rq(enic, rq_num)];
284 cq_idx = cq->to_clean;
286 cq_tail = ioread32(&cq->ctrl->cq_tail);
288 if (cq_tail < cq_idx)
289 cq_tail += cq->ring.desc_count;
291 queue_count = cq_tail - cq_idx;
296 static int enicpmd_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
299 unsigned int socket_id,
300 const struct rte_eth_rxconf *rx_conf,
301 struct rte_mempool *mp)
304 struct enic *enic = pmd_priv(eth_dev);
306 ENICPMD_FUNC_TRACE();
308 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
309 return -E_RTE_SECONDARY;
310 RTE_ASSERT(enic_rte_rq_idx_to_sop_idx(queue_idx) < enic->conf_rq_count);
311 eth_dev->data->rx_queues[queue_idx] =
312 (void *)&enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
314 ret = enic_alloc_rq(enic, queue_idx, socket_id, mp, nb_desc,
315 rx_conf->rx_free_thresh);
317 dev_err(enic, "error in allocating rq\n");
321 return enicpmd_dev_setup_intr(enic);
324 static int enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
326 struct enic *enic = pmd_priv(eth_dev);
329 ENICPMD_FUNC_TRACE();
331 offloads = eth_dev->data->dev_conf.rxmode.offloads;
332 if (mask & ETH_VLAN_STRIP_MASK) {
333 if (offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
334 enic->ig_vlan_strip_en = 1;
336 enic->ig_vlan_strip_en = 0;
339 if ((mask & ETH_VLAN_FILTER_MASK) &&
340 (offloads & DEV_RX_OFFLOAD_VLAN_FILTER)) {
342 "Configuration of VLAN filter is not supported\n");
345 if ((mask & ETH_VLAN_EXTEND_MASK) &&
346 (offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)) {
348 "Configuration of extended VLAN is not supported\n");
351 return enic_set_vlan_strip(enic);
354 static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
358 struct enic *enic = pmd_priv(eth_dev);
360 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
361 return -E_RTE_SECONDARY;
363 ENICPMD_FUNC_TRACE();
364 ret = enic_set_vnic_res(enic);
366 dev_err(enic, "Set vNIC resource num failed, aborting\n");
371 enic->hw_ip_checksum = !!(eth_dev->data->dev_conf.rxmode.offloads &
372 DEV_RX_OFFLOAD_CHECKSUM);
373 /* All vlan offload masks to apply the current settings */
374 mask = ETH_VLAN_STRIP_MASK |
375 ETH_VLAN_FILTER_MASK |
376 ETH_VLAN_EXTEND_MASK;
377 ret = enicpmd_vlan_offload_set(eth_dev, mask);
379 dev_err(enic, "Failed to configure VLAN offloads\n");
383 * Initialize RSS with the default reta and key. If the user key is
384 * given (rx_adv_conf.rss_conf.rss_key), will use that instead of the
387 return enic_init_rss_nic_cfg(enic);
391 * It returns 0 on success.
393 static int enicpmd_dev_start(struct rte_eth_dev *eth_dev)
395 struct enic *enic = pmd_priv(eth_dev);
397 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
398 return -E_RTE_SECONDARY;
400 ENICPMD_FUNC_TRACE();
401 return enic_enable(enic);
405 * Stop device: disable rx and tx functions to allow for reconfiguring.
407 static void enicpmd_dev_stop(struct rte_eth_dev *eth_dev)
409 struct rte_eth_link link;
410 struct enic *enic = pmd_priv(eth_dev);
412 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
415 ENICPMD_FUNC_TRACE();
418 memset(&link, 0, sizeof(link));
419 rte_eth_linkstatus_set(eth_dev, &link);
425 static void enicpmd_dev_close(struct rte_eth_dev *eth_dev)
427 struct enic *enic = pmd_priv(eth_dev);
429 ENICPMD_FUNC_TRACE();
433 static int enicpmd_dev_link_update(struct rte_eth_dev *eth_dev,
434 __rte_unused int wait_to_complete)
436 struct enic *enic = pmd_priv(eth_dev);
438 ENICPMD_FUNC_TRACE();
439 return enic_link_update(enic);
442 static int enicpmd_dev_stats_get(struct rte_eth_dev *eth_dev,
443 struct rte_eth_stats *stats)
445 struct enic *enic = pmd_priv(eth_dev);
447 ENICPMD_FUNC_TRACE();
448 return enic_dev_stats_get(enic, stats);
451 static void enicpmd_dev_stats_reset(struct rte_eth_dev *eth_dev)
453 struct enic *enic = pmd_priv(eth_dev);
455 ENICPMD_FUNC_TRACE();
456 enic_dev_stats_clear(enic);
459 static void enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
460 struct rte_eth_dev_info *device_info)
462 struct enic *enic = pmd_priv(eth_dev);
464 ENICPMD_FUNC_TRACE();
465 /* Scattered Rx uses two receive queues per rx queue exposed to dpdk */
466 device_info->max_rx_queues = enic->conf_rq_count / 2;
467 device_info->max_tx_queues = enic->conf_wq_count;
468 device_info->min_rx_bufsize = ENIC_MIN_MTU;
469 /* "Max" mtu is not a typo. HW receives packet sizes up to the
470 * max mtu regardless of the current mtu (vNIC's mtu). vNIC mtu is
471 * a hint to the driver to size receive buffers accordingly so that
472 * larger-than-vnic-mtu packets get truncated.. For DPDK, we let
473 * the user decide the buffer size via rxmode.max_rx_pkt_len, basically
476 device_info->max_rx_pktlen = enic_mtu_to_max_rx_pktlen(enic->max_mtu);
477 device_info->max_mac_addrs = ENIC_UNICAST_PERFECT_FILTERS;
478 device_info->min_mtu = ENIC_MIN_MTU;
479 device_info->max_mtu = enic->max_mtu;
480 device_info->rx_offload_capa = enic->rx_offload_capa;
481 device_info->tx_offload_capa = enic->tx_offload_capa;
482 device_info->tx_queue_offload_capa = enic->tx_queue_offload_capa;
483 device_info->default_rxconf = (struct rte_eth_rxconf) {
484 .rx_free_thresh = ENIC_DEFAULT_RX_FREE_THRESH
486 device_info->reta_size = enic->reta_size;
487 device_info->hash_key_size = enic->hash_key_size;
488 device_info->flow_type_rss_offloads = enic->flow_type_rss_offloads;
489 device_info->rx_desc_lim = (struct rte_eth_desc_lim) {
490 .nb_max = enic->config.rq_desc_count,
491 .nb_min = ENIC_MIN_RQ_DESCS,
492 .nb_align = ENIC_ALIGN_DESCS,
494 device_info->tx_desc_lim = (struct rte_eth_desc_lim) {
495 .nb_max = enic->config.wq_desc_count,
496 .nb_min = ENIC_MIN_WQ_DESCS,
497 .nb_align = ENIC_ALIGN_DESCS,
498 .nb_seg_max = ENIC_TX_XMIT_MAX,
499 .nb_mtu_seg_max = ENIC_NON_TSO_MAX_DESC,
501 device_info->default_rxportconf = (struct rte_eth_dev_portconf) {
502 .burst_size = ENIC_DEFAULT_RX_BURST,
503 .ring_size = RTE_MIN(device_info->rx_desc_lim.nb_max,
504 ENIC_DEFAULT_RX_RING_SIZE),
505 .nb_queues = ENIC_DEFAULT_RX_RINGS,
507 device_info->default_txportconf = (struct rte_eth_dev_portconf) {
508 .burst_size = ENIC_DEFAULT_TX_BURST,
509 .ring_size = RTE_MIN(device_info->tx_desc_lim.nb_max,
510 ENIC_DEFAULT_TX_RING_SIZE),
511 .nb_queues = ENIC_DEFAULT_TX_RINGS,
515 static const uint32_t *enicpmd_dev_supported_ptypes_get(struct rte_eth_dev *dev)
517 static const uint32_t ptypes[] = {
519 RTE_PTYPE_L2_ETHER_VLAN,
520 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
521 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
525 RTE_PTYPE_L4_NONFRAG,
528 static const uint32_t ptypes_overlay[] = {
530 RTE_PTYPE_L2_ETHER_VLAN,
531 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
532 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
536 RTE_PTYPE_L4_NONFRAG,
537 RTE_PTYPE_TUNNEL_GRENAT,
538 RTE_PTYPE_INNER_L2_ETHER,
539 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
540 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
541 RTE_PTYPE_INNER_L4_TCP,
542 RTE_PTYPE_INNER_L4_UDP,
543 RTE_PTYPE_INNER_L4_FRAG,
544 RTE_PTYPE_INNER_L4_NONFRAG,
548 if (dev->rx_pkt_burst != enic_dummy_recv_pkts &&
549 dev->rx_pkt_burst != NULL) {
550 struct enic *enic = pmd_priv(dev);
551 if (enic->overlay_offload)
552 return ptypes_overlay;
559 static void enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
561 struct enic *enic = pmd_priv(eth_dev);
563 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
566 ENICPMD_FUNC_TRACE();
569 enic_add_packet_filter(enic);
572 static void enicpmd_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
574 struct enic *enic = pmd_priv(eth_dev);
576 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
579 ENICPMD_FUNC_TRACE();
581 enic_add_packet_filter(enic);
584 static void enicpmd_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
586 struct enic *enic = pmd_priv(eth_dev);
588 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
591 ENICPMD_FUNC_TRACE();
593 enic_add_packet_filter(enic);
596 static void enicpmd_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
598 struct enic *enic = pmd_priv(eth_dev);
600 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
603 ENICPMD_FUNC_TRACE();
605 enic_add_packet_filter(enic);
608 static int enicpmd_add_mac_addr(struct rte_eth_dev *eth_dev,
609 struct rte_ether_addr *mac_addr,
610 __rte_unused uint32_t index, __rte_unused uint32_t pool)
612 struct enic *enic = pmd_priv(eth_dev);
614 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
615 return -E_RTE_SECONDARY;
617 ENICPMD_FUNC_TRACE();
618 return enic_set_mac_address(enic, mac_addr->addr_bytes);
621 static void enicpmd_remove_mac_addr(struct rte_eth_dev *eth_dev, uint32_t index)
623 struct enic *enic = pmd_priv(eth_dev);
625 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
628 ENICPMD_FUNC_TRACE();
629 if (enic_del_mac_address(enic, index))
630 dev_err(enic, "del mac addr failed\n");
633 static int enicpmd_set_mac_addr(struct rte_eth_dev *eth_dev,
634 struct rte_ether_addr *addr)
636 struct enic *enic = pmd_priv(eth_dev);
639 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
640 return -E_RTE_SECONDARY;
642 ENICPMD_FUNC_TRACE();
643 ret = enic_del_mac_address(enic, 0);
646 return enic_set_mac_address(enic, addr->addr_bytes);
649 static void debug_log_add_del_addr(struct rte_ether_addr *addr, bool add)
651 char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
653 rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE, addr);
654 PMD_INIT_LOG(DEBUG, " %s address %s\n",
655 add ? "add" : "remove", mac_str);
658 static int enicpmd_set_mc_addr_list(struct rte_eth_dev *eth_dev,
659 struct rte_ether_addr *mc_addr_set,
662 struct enic *enic = pmd_priv(eth_dev);
663 char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
664 struct rte_ether_addr *addr;
668 ENICPMD_FUNC_TRACE();
670 /* Validate the given addresses first */
671 for (i = 0; i < nb_mc_addr && mc_addr_set != NULL; i++) {
672 addr = &mc_addr_set[i];
673 if (!rte_is_multicast_ether_addr(addr) ||
674 rte_is_broadcast_ether_addr(addr)) {
675 rte_ether_format_addr(mac_str,
676 RTE_ETHER_ADDR_FMT_SIZE, addr);
677 PMD_INIT_LOG(ERR, " invalid multicast address %s\n",
683 /* Flush all if requested */
684 if (nb_mc_addr == 0 || mc_addr_set == NULL) {
685 PMD_INIT_LOG(DEBUG, " flush multicast addresses\n");
686 for (i = 0; i < enic->mc_count; i++) {
687 addr = &enic->mc_addrs[i];
688 debug_log_add_del_addr(addr, false);
689 ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
697 if (nb_mc_addr > ENIC_MULTICAST_PERFECT_FILTERS) {
698 PMD_INIT_LOG(ERR, " too many multicast addresses: max=%d\n",
699 ENIC_MULTICAST_PERFECT_FILTERS);
703 * devcmd is slow, so apply the difference instead of flushing and
705 * 1. Delete addresses on the NIC but not on the host
707 for (i = 0; i < enic->mc_count; i++) {
708 addr = &enic->mc_addrs[i];
709 for (j = 0; j < nb_mc_addr; j++) {
710 if (rte_is_same_ether_addr(addr, &mc_addr_set[j]))
715 debug_log_add_del_addr(addr, false);
716 ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
720 /* 2. Add addresses on the host but not on the NIC */
721 for (i = 0; i < nb_mc_addr; i++) {
722 addr = &mc_addr_set[i];
723 for (j = 0; j < enic->mc_count; j++) {
724 if (rte_is_same_ether_addr(addr, &enic->mc_addrs[j]))
727 if (j < enic->mc_count)
729 debug_log_add_del_addr(addr, true);
730 ret = vnic_dev_add_addr(enic->vdev, addr->addr_bytes);
734 /* Keep a copy so we can flush/apply later on.. */
735 memcpy(enic->mc_addrs, mc_addr_set,
736 nb_mc_addr * sizeof(struct rte_ether_addr));
737 enic->mc_count = nb_mc_addr;
741 static int enicpmd_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
743 struct enic *enic = pmd_priv(eth_dev);
745 ENICPMD_FUNC_TRACE();
746 return enic_set_mtu(enic, mtu);
749 static int enicpmd_dev_rss_reta_query(struct rte_eth_dev *dev,
750 struct rte_eth_rss_reta_entry64
754 struct enic *enic = pmd_priv(dev);
755 uint16_t i, idx, shift;
757 ENICPMD_FUNC_TRACE();
758 if (reta_size != ENIC_RSS_RETA_SIZE) {
759 dev_err(enic, "reta_query: wrong reta_size. given=%u expected=%u\n",
760 reta_size, ENIC_RSS_RETA_SIZE);
764 for (i = 0; i < reta_size; i++) {
765 idx = i / RTE_RETA_GROUP_SIZE;
766 shift = i % RTE_RETA_GROUP_SIZE;
767 if (reta_conf[idx].mask & (1ULL << shift))
768 reta_conf[idx].reta[shift] = enic_sop_rq_idx_to_rte_idx(
769 enic->rss_cpu.cpu[i / 4].b[i % 4]);
775 static int enicpmd_dev_rss_reta_update(struct rte_eth_dev *dev,
776 struct rte_eth_rss_reta_entry64
780 struct enic *enic = pmd_priv(dev);
781 union vnic_rss_cpu rss_cpu;
782 uint16_t i, idx, shift;
784 ENICPMD_FUNC_TRACE();
785 if (reta_size != ENIC_RSS_RETA_SIZE) {
786 dev_err(enic, "reta_update: wrong reta_size. given=%u"
788 reta_size, ENIC_RSS_RETA_SIZE);
792 * Start with the current reta and modify it per reta_conf, as we
793 * need to push the entire reta even if we only modify one entry.
795 rss_cpu = enic->rss_cpu;
796 for (i = 0; i < reta_size; i++) {
797 idx = i / RTE_RETA_GROUP_SIZE;
798 shift = i % RTE_RETA_GROUP_SIZE;
799 if (reta_conf[idx].mask & (1ULL << shift))
800 rss_cpu.cpu[i / 4].b[i % 4] =
801 enic_rte_rq_idx_to_sop_idx(
802 reta_conf[idx].reta[shift]);
804 return enic_set_rss_reta(enic, &rss_cpu);
807 static int enicpmd_dev_rss_hash_update(struct rte_eth_dev *dev,
808 struct rte_eth_rss_conf *rss_conf)
810 struct enic *enic = pmd_priv(dev);
812 ENICPMD_FUNC_TRACE();
813 return enic_set_rss_conf(enic, rss_conf);
816 static int enicpmd_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
817 struct rte_eth_rss_conf *rss_conf)
819 struct enic *enic = pmd_priv(dev);
821 ENICPMD_FUNC_TRACE();
822 if (rss_conf == NULL)
824 if (rss_conf->rss_key != NULL &&
825 rss_conf->rss_key_len < ENIC_RSS_HASH_KEY_SIZE) {
826 dev_err(enic, "rss_hash_conf_get: wrong rss_key_len. given=%u"
828 rss_conf->rss_key_len, ENIC_RSS_HASH_KEY_SIZE);
831 rss_conf->rss_hf = enic->rss_hf;
832 if (rss_conf->rss_key != NULL) {
834 for (i = 0; i < ENIC_RSS_HASH_KEY_SIZE; i++) {
835 rss_conf->rss_key[i] =
836 enic->rss_key.key[i / 10].b[i % 10];
838 rss_conf->rss_key_len = ENIC_RSS_HASH_KEY_SIZE;
843 static void enicpmd_dev_rxq_info_get(struct rte_eth_dev *dev,
844 uint16_t rx_queue_id,
845 struct rte_eth_rxq_info *qinfo)
847 struct enic *enic = pmd_priv(dev);
848 struct vnic_rq *rq_sop;
849 struct vnic_rq *rq_data;
850 struct rte_eth_rxconf *conf;
851 uint16_t sop_queue_idx;
852 uint16_t data_queue_idx;
854 ENICPMD_FUNC_TRACE();
855 sop_queue_idx = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
856 data_queue_idx = enic_rte_rq_idx_to_data_idx(rx_queue_id);
857 rq_sop = &enic->rq[sop_queue_idx];
858 rq_data = &enic->rq[data_queue_idx]; /* valid if data_queue_enable */
859 qinfo->mp = rq_sop->mp;
860 qinfo->scattered_rx = rq_sop->data_queue_enable;
861 qinfo->nb_desc = rq_sop->ring.desc_count;
862 if (qinfo->scattered_rx)
863 qinfo->nb_desc += rq_data->ring.desc_count;
865 memset(conf, 0, sizeof(*conf));
866 conf->rx_free_thresh = rq_sop->rx_free_thresh;
867 conf->rx_drop_en = 1;
869 * Except VLAN stripping (port setting), all the checksum offloads
870 * are always enabled.
872 conf->offloads = enic->rx_offload_capa;
873 if (!enic->ig_vlan_strip_en)
874 conf->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
875 /* rx_thresh and other fields are not applicable for enic */
878 static void enicpmd_dev_txq_info_get(struct rte_eth_dev *dev,
879 uint16_t tx_queue_id,
880 struct rte_eth_txq_info *qinfo)
882 struct enic *enic = pmd_priv(dev);
883 struct vnic_wq *wq = &enic->wq[tx_queue_id];
885 ENICPMD_FUNC_TRACE();
886 qinfo->nb_desc = wq->ring.desc_count;
887 memset(&qinfo->conf, 0, sizeof(qinfo->conf));
888 qinfo->conf.offloads = wq->offloads;
889 /* tx_thresh, and all the other fields are not applicable for enic */
892 static int enicpmd_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev,
893 uint16_t rx_queue_id)
895 struct enic *enic = pmd_priv(eth_dev);
897 ENICPMD_FUNC_TRACE();
898 vnic_intr_unmask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
902 static int enicpmd_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev,
903 uint16_t rx_queue_id)
905 struct enic *enic = pmd_priv(eth_dev);
907 ENICPMD_FUNC_TRACE();
908 vnic_intr_mask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
912 static int udp_tunnel_common_check(struct enic *enic,
913 struct rte_eth_udp_tunnel *tnl)
915 if (tnl->prot_type != RTE_TUNNEL_TYPE_VXLAN)
917 if (!enic->overlay_offload) {
918 PMD_INIT_LOG(DEBUG, " vxlan (overlay offload) is not "
925 static int update_vxlan_port(struct enic *enic, uint16_t port)
927 if (vnic_dev_overlay_offload_cfg(enic->vdev,
928 OVERLAY_CFG_VXLAN_PORT_UPDATE,
930 PMD_INIT_LOG(DEBUG, " failed to update vxlan port\n");
933 PMD_INIT_LOG(DEBUG, " updated vxlan port to %u\n", port);
934 enic->vxlan_port = port;
938 static int enicpmd_dev_udp_tunnel_port_add(struct rte_eth_dev *eth_dev,
939 struct rte_eth_udp_tunnel *tnl)
941 struct enic *enic = pmd_priv(eth_dev);
944 ENICPMD_FUNC_TRACE();
945 ret = udp_tunnel_common_check(enic, tnl);
949 * The NIC has 1 configurable VXLAN port number. "Adding" a new port
950 * number replaces it.
952 if (tnl->udp_port == enic->vxlan_port || tnl->udp_port == 0) {
953 PMD_INIT_LOG(DEBUG, " %u is already configured or invalid\n",
957 return update_vxlan_port(enic, tnl->udp_port);
960 static int enicpmd_dev_udp_tunnel_port_del(struct rte_eth_dev *eth_dev,
961 struct rte_eth_udp_tunnel *tnl)
963 struct enic *enic = pmd_priv(eth_dev);
966 ENICPMD_FUNC_TRACE();
967 ret = udp_tunnel_common_check(enic, tnl);
971 * Clear the previously set port number and restore the
972 * hardware default port number. Some drivers disable VXLAN
973 * offloads when there are no configured port numbers. But
974 * enic does not do that as VXLAN is part of overlay offload,
975 * which is tied to inner RSS and TSO.
977 if (tnl->udp_port != enic->vxlan_port) {
978 PMD_INIT_LOG(DEBUG, " %u is not a configured vxlan port\n",
982 return update_vxlan_port(enic, ENIC_DEFAULT_VXLAN_PORT);
985 static int enicpmd_dev_fw_version_get(struct rte_eth_dev *eth_dev,
986 char *fw_version, size_t fw_size)
988 struct vnic_devcmd_fw_info *info;
992 ENICPMD_FUNC_TRACE();
993 if (fw_version == NULL || fw_size <= 0)
995 enic = pmd_priv(eth_dev);
996 ret = vnic_dev_fw_info(enic->vdev, &info);
999 snprintf(fw_version, fw_size, "%s %s",
1000 info->fw_version, info->fw_build);
1001 fw_version[fw_size - 1] = '\0';
1005 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
1006 .dev_configure = enicpmd_dev_configure,
1007 .dev_start = enicpmd_dev_start,
1008 .dev_stop = enicpmd_dev_stop,
1009 .dev_set_link_up = NULL,
1010 .dev_set_link_down = NULL,
1011 .dev_close = enicpmd_dev_close,
1012 .promiscuous_enable = enicpmd_dev_promiscuous_enable,
1013 .promiscuous_disable = enicpmd_dev_promiscuous_disable,
1014 .allmulticast_enable = enicpmd_dev_allmulticast_enable,
1015 .allmulticast_disable = enicpmd_dev_allmulticast_disable,
1016 .link_update = enicpmd_dev_link_update,
1017 .stats_get = enicpmd_dev_stats_get,
1018 .stats_reset = enicpmd_dev_stats_reset,
1019 .queue_stats_mapping_set = NULL,
1020 .dev_infos_get = enicpmd_dev_info_get,
1021 .dev_supported_ptypes_get = enicpmd_dev_supported_ptypes_get,
1022 .mtu_set = enicpmd_mtu_set,
1023 .vlan_filter_set = NULL,
1024 .vlan_tpid_set = NULL,
1025 .vlan_offload_set = enicpmd_vlan_offload_set,
1026 .vlan_strip_queue_set = NULL,
1027 .rx_queue_start = enicpmd_dev_rx_queue_start,
1028 .rx_queue_stop = enicpmd_dev_rx_queue_stop,
1029 .tx_queue_start = enicpmd_dev_tx_queue_start,
1030 .tx_queue_stop = enicpmd_dev_tx_queue_stop,
1031 .rx_queue_setup = enicpmd_dev_rx_queue_setup,
1032 .rx_queue_release = enicpmd_dev_rx_queue_release,
1033 .rx_queue_count = enicpmd_dev_rx_queue_count,
1034 .rx_descriptor_done = NULL,
1035 .tx_queue_setup = enicpmd_dev_tx_queue_setup,
1036 .tx_queue_release = enicpmd_dev_tx_queue_release,
1037 .rx_queue_intr_enable = enicpmd_dev_rx_queue_intr_enable,
1038 .rx_queue_intr_disable = enicpmd_dev_rx_queue_intr_disable,
1039 .rxq_info_get = enicpmd_dev_rxq_info_get,
1040 .txq_info_get = enicpmd_dev_txq_info_get,
1042 .dev_led_off = NULL,
1043 .flow_ctrl_get = NULL,
1044 .flow_ctrl_set = NULL,
1045 .priority_flow_ctrl_set = NULL,
1046 .mac_addr_add = enicpmd_add_mac_addr,
1047 .mac_addr_remove = enicpmd_remove_mac_addr,
1048 .mac_addr_set = enicpmd_set_mac_addr,
1049 .set_mc_addr_list = enicpmd_set_mc_addr_list,
1050 .filter_ctrl = enicpmd_dev_filter_ctrl,
1051 .reta_query = enicpmd_dev_rss_reta_query,
1052 .reta_update = enicpmd_dev_rss_reta_update,
1053 .rss_hash_conf_get = enicpmd_dev_rss_hash_conf_get,
1054 .rss_hash_update = enicpmd_dev_rss_hash_update,
1055 .udp_tunnel_port_add = enicpmd_dev_udp_tunnel_port_add,
1056 .udp_tunnel_port_del = enicpmd_dev_udp_tunnel_port_del,
1057 .fw_version_get = enicpmd_dev_fw_version_get,
1060 static int enic_parse_zero_one(const char *key,
1067 enic = (struct enic *)opaque;
1068 if (strcmp(value, "0") == 0) {
1070 } else if (strcmp(value, "1") == 0) {
1073 dev_err(enic, "Invalid value for %s"
1074 ": expected=0|1 given=%s\n", key, value);
1077 if (strcmp(key, ENIC_DEVARG_DISABLE_OVERLAY) == 0)
1078 enic->disable_overlay = b;
1079 if (strcmp(key, ENIC_DEVARG_ENABLE_AVX2_RX) == 0)
1080 enic->enable_avx2_rx = b;
1084 static int enic_parse_ig_vlan_rewrite(__rte_unused const char *key,
1090 enic = (struct enic *)opaque;
1091 if (strcmp(value, "trunk") == 0) {
1092 /* Trunk mode: always tag */
1093 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_DEFAULT_TRUNK;
1094 } else if (strcmp(value, "untag") == 0) {
1095 /* Untag default VLAN mode: untag if VLAN = default VLAN */
1096 enic->ig_vlan_rewrite_mode =
1097 IG_VLAN_REWRITE_MODE_UNTAG_DEFAULT_VLAN;
1098 } else if (strcmp(value, "priority") == 0) {
1100 * Priority-tag default VLAN mode: priority tag (VLAN header
1101 * with ID=0) if VLAN = default
1103 enic->ig_vlan_rewrite_mode =
1104 IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN;
1105 } else if (strcmp(value, "pass") == 0) {
1106 /* Pass through mode: do not touch tags */
1107 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1109 dev_err(enic, "Invalid value for " ENIC_DEVARG_IG_VLAN_REWRITE
1110 ": expected=trunk|untag|priority|pass given=%s\n",
1117 static int enic_check_devargs(struct rte_eth_dev *dev)
1119 static const char *const valid_keys[] = {
1120 ENIC_DEVARG_DISABLE_OVERLAY,
1121 ENIC_DEVARG_ENABLE_AVX2_RX,
1122 ENIC_DEVARG_IG_VLAN_REWRITE,
1124 struct enic *enic = pmd_priv(dev);
1125 struct rte_kvargs *kvlist;
1127 ENICPMD_FUNC_TRACE();
1129 enic->disable_overlay = false;
1130 enic->enable_avx2_rx = false;
1131 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1132 if (!dev->device->devargs)
1134 kvlist = rte_kvargs_parse(dev->device->devargs->args, valid_keys);
1137 if (rte_kvargs_process(kvlist, ENIC_DEVARG_DISABLE_OVERLAY,
1138 enic_parse_zero_one, enic) < 0 ||
1139 rte_kvargs_process(kvlist, ENIC_DEVARG_ENABLE_AVX2_RX,
1140 enic_parse_zero_one, enic) < 0 ||
1141 rte_kvargs_process(kvlist, ENIC_DEVARG_IG_VLAN_REWRITE,
1142 enic_parse_ig_vlan_rewrite, enic) < 0) {
1143 rte_kvargs_free(kvlist);
1146 rte_kvargs_free(kvlist);
1150 /* Initialize the driver
1151 * It returns 0 on success.
1153 static int eth_enicpmd_dev_init(struct rte_eth_dev *eth_dev)
1155 struct rte_pci_device *pdev;
1156 struct rte_pci_addr *addr;
1157 struct enic *enic = pmd_priv(eth_dev);
1160 ENICPMD_FUNC_TRACE();
1162 enic->port_id = eth_dev->data->port_id;
1163 enic->rte_dev = eth_dev;
1164 eth_dev->dev_ops = &enicpmd_eth_dev_ops;
1165 eth_dev->rx_pkt_burst = &enic_recv_pkts;
1166 eth_dev->tx_pkt_burst = &enic_xmit_pkts;
1167 eth_dev->tx_pkt_prepare = &enic_prep_pkts;
1168 /* Let rte_eth_dev_close() release the port resources */
1169 eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1171 pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
1172 rte_eth_copy_pci_info(eth_dev, pdev);
1176 snprintf(enic->bdf_name, ENICPMD_BDF_LENGTH, "%04x:%02x:%02x.%x",
1177 addr->domain, addr->bus, addr->devid, addr->function);
1179 err = enic_check_devargs(eth_dev);
1182 return enic_probe(enic);
1185 static int eth_enic_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1186 struct rte_pci_device *pci_dev)
1188 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct enic),
1189 eth_enicpmd_dev_init);
1192 static int eth_enic_pci_remove(struct rte_pci_device *pci_dev)
1194 return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1197 static struct rte_pci_driver rte_enic_pmd = {
1198 .id_table = pci_id_enic_map,
1199 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
1200 RTE_PCI_DRV_IOVA_AS_VA,
1201 .probe = eth_enic_pci_probe,
1202 .remove = eth_enic_pci_remove,
1205 RTE_PMD_REGISTER_PCI(net_enic, rte_enic_pmd);
1206 RTE_PMD_REGISTER_PCI_TABLE(net_enic, pci_id_enic_map);
1207 RTE_PMD_REGISTER_KMOD_DEP(net_enic, "* igb_uio | uio_pci_generic | vfio-pci");
1208 RTE_PMD_REGISTER_PARAM_STRING(net_enic,
1209 ENIC_DEVARG_DISABLE_OVERLAY "=0|1 "
1210 ENIC_DEVARG_ENABLE_AVX2_RX "=0|1 "
1211 ENIC_DEVARG_IG_VLAN_REWRITE "=trunk|untag|priority|pass");