1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2016 Cavium, Inc
14 #include <netinet/in.h>
15 #include <sys/queue.h>
17 #include <rte_alarm.h>
18 #include <rte_branch_prediction.h>
19 #include <rte_byteorder.h>
20 #include <rte_common.h>
21 #include <rte_cycles.h>
22 #include <rte_debug.h>
25 #include <rte_ether.h>
26 #include <rte_ethdev_driver.h>
27 #include <rte_ethdev_pci.h>
28 #include <rte_interrupts.h>
30 #include <rte_memory.h>
31 #include <rte_memzone.h>
32 #include <rte_malloc.h>
33 #include <rte_random.h>
35 #include <rte_bus_pci.h>
36 #include <rte_tailq.h>
37 #include <rte_devargs.h>
38 #include <rte_kvargs.h>
40 #include "base/nicvf_plat.h"
42 #include "nicvf_ethdev.h"
43 #include "nicvf_rxtx.h"
44 #include "nicvf_svf.h"
45 #include "nicvf_logs.h"
47 int nicvf_logtype_mbox;
48 int nicvf_logtype_init;
49 int nicvf_logtype_driver;
51 static void nicvf_dev_stop(struct rte_eth_dev *dev);
52 static void nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup);
53 static void nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic,
55 static int nicvf_vlan_offload_config(struct rte_eth_dev *dev, int mask);
56 static int nicvf_vlan_offload_set(struct rte_eth_dev *dev, int mask);
58 RTE_INIT(nicvf_init_log)
60 nicvf_logtype_mbox = rte_log_register("pmd.net.thunderx.mbox");
61 if (nicvf_logtype_mbox >= 0)
62 rte_log_set_level(nicvf_logtype_mbox, RTE_LOG_NOTICE);
64 nicvf_logtype_init = rte_log_register("pmd.net.thunderx.init");
65 if (nicvf_logtype_init >= 0)
66 rte_log_set_level(nicvf_logtype_init, RTE_LOG_NOTICE);
68 nicvf_logtype_driver = rte_log_register("pmd.net.thunderx.driver");
69 if (nicvf_logtype_driver >= 0)
70 rte_log_set_level(nicvf_logtype_driver, RTE_LOG_NOTICE);
74 nicvf_link_status_update(struct nicvf *nic,
75 struct rte_eth_link *link)
77 memset(link, 0, sizeof(*link));
79 link->link_status = nic->link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
81 if (nic->duplex == NICVF_HALF_DUPLEX)
82 link->link_duplex = ETH_LINK_HALF_DUPLEX;
83 else if (nic->duplex == NICVF_FULL_DUPLEX)
84 link->link_duplex = ETH_LINK_FULL_DUPLEX;
85 link->link_speed = nic->speed;
86 link->link_autoneg = ETH_LINK_AUTONEG;
90 nicvf_interrupt(void *arg)
92 struct rte_eth_dev *dev = arg;
93 struct nicvf *nic = nicvf_pmd_priv(dev);
94 struct rte_eth_link link;
96 if (nicvf_reg_poll_interrupts(nic) == NIC_MBOX_MSG_BGX_LINK_CHANGE) {
97 if (dev->data->dev_conf.intr_conf.lsc) {
98 nicvf_link_status_update(nic, &link);
99 rte_eth_linkstatus_set(dev, &link);
101 _rte_eth_dev_callback_process(dev,
102 RTE_ETH_EVENT_INTR_LSC,
107 rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
108 nicvf_interrupt, dev);
112 nicvf_vf_interrupt(void *arg)
114 struct nicvf *nic = arg;
116 nicvf_reg_poll_interrupts(nic);
118 rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
119 nicvf_vf_interrupt, nic);
123 nicvf_periodic_alarm_start(void (fn)(void *), void *arg)
125 return rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000, fn, arg);
129 nicvf_periodic_alarm_stop(void (fn)(void *), void *arg)
131 return rte_eal_alarm_cancel(fn, arg);
135 * Return 0 means link status changed, -1 means not changed
138 nicvf_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
140 #define CHECK_INTERVAL 100 /* 100ms */
141 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
142 struct rte_eth_link link;
143 struct nicvf *nic = nicvf_pmd_priv(dev);
146 PMD_INIT_FUNC_TRACE();
148 if (wait_to_complete) {
149 /* rte_eth_link_get() might need to wait up to 9 seconds */
150 for (i = 0; i < MAX_CHECK_TIME; i++) {
151 nicvf_link_status_update(nic, &link);
152 if (link.link_status == ETH_LINK_UP)
154 rte_delay_ms(CHECK_INTERVAL);
157 nicvf_link_status_update(nic, &link);
160 return rte_eth_linkstatus_set(dev, &link);
164 nicvf_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
166 struct nicvf *nic = nicvf_pmd_priv(dev);
167 uint32_t buffsz, frame_size = mtu + NIC_HW_L2_OVERHEAD;
169 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
171 PMD_INIT_FUNC_TRACE();
173 if (frame_size > NIC_HW_MAX_FRS)
176 if (frame_size < NIC_HW_MIN_FRS)
179 buffsz = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
182 * Refuse mtu that requires the support of scattered packets
183 * when this feature has not been enabled before.
185 if (dev->data->dev_started && !dev->data->scattered_rx &&
186 (frame_size + 2 * VLAN_TAG_SIZE > buffsz))
189 /* check <seg size> * <max_seg> >= max_frame */
190 if (dev->data->scattered_rx &&
191 (frame_size + 2 * VLAN_TAG_SIZE > buffsz * NIC_HW_MAX_SEGS))
194 if (frame_size > ETHER_MAX_LEN)
195 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
197 rxmode->offloads &= ~DEV_RX_OFFLOAD_JUMBO_FRAME;
199 if (nicvf_mbox_update_hw_max_frs(nic, mtu))
202 /* Update max_rx_pkt_len */
203 rxmode->max_rx_pkt_len = mtu + ETHER_HDR_LEN;
206 for (i = 0; i < nic->sqs_count; i++)
207 nic->snicvf[i]->mtu = mtu;
213 nicvf_dev_get_regs(struct rte_eth_dev *dev, struct rte_dev_reg_info *regs)
215 uint64_t *data = regs->data;
216 struct nicvf *nic = nicvf_pmd_priv(dev);
219 regs->length = nicvf_reg_get_count();
220 regs->width = THUNDERX_REG_BYTES;
224 /* Support only full register dump */
225 if ((regs->length == 0) ||
226 (regs->length == (uint32_t)nicvf_reg_get_count())) {
227 regs->version = nic->vendor_id << 16 | nic->device_id;
228 nicvf_reg_dump(nic, data);
235 nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
238 struct nicvf_hw_rx_qstats rx_qstats;
239 struct nicvf_hw_tx_qstats tx_qstats;
240 struct nicvf_hw_stats port_stats;
241 struct nicvf *nic = nicvf_pmd_priv(dev);
242 uint16_t rx_start, rx_end;
243 uint16_t tx_start, tx_end;
246 /* RX queue indices for the first VF */
247 nicvf_rx_range(dev, nic, &rx_start, &rx_end);
249 /* Reading per RX ring stats */
250 for (qidx = rx_start; qidx <= rx_end; qidx++) {
251 if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
254 nicvf_hw_get_rx_qstats(nic, &rx_qstats, qidx);
255 stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
256 stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
259 /* TX queue indices for the first VF */
260 nicvf_tx_range(dev, nic, &tx_start, &tx_end);
262 /* Reading per TX ring stats */
263 for (qidx = tx_start; qidx <= tx_end; qidx++) {
264 if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
267 nicvf_hw_get_tx_qstats(nic, &tx_qstats, qidx);
268 stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
269 stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
272 for (i = 0; i < nic->sqs_count; i++) {
273 struct nicvf *snic = nic->snicvf[i];
278 /* RX queue indices for a secondary VF */
279 nicvf_rx_range(dev, snic, &rx_start, &rx_end);
281 /* Reading per RX ring stats */
282 for (qidx = rx_start; qidx <= rx_end; qidx++) {
283 if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
286 nicvf_hw_get_rx_qstats(snic, &rx_qstats,
287 qidx % MAX_RCV_QUEUES_PER_QS);
288 stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
289 stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
292 /* TX queue indices for a secondary VF */
293 nicvf_tx_range(dev, snic, &tx_start, &tx_end);
294 /* Reading per TX ring stats */
295 for (qidx = tx_start; qidx <= tx_end; qidx++) {
296 if (qidx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)
299 nicvf_hw_get_tx_qstats(snic, &tx_qstats,
300 qidx % MAX_SND_QUEUES_PER_QS);
301 stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
302 stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
306 nicvf_hw_get_stats(nic, &port_stats);
307 stats->ibytes = port_stats.rx_bytes;
308 stats->ipackets = port_stats.rx_ucast_frames;
309 stats->ipackets += port_stats.rx_bcast_frames;
310 stats->ipackets += port_stats.rx_mcast_frames;
311 stats->ierrors = port_stats.rx_l2_errors;
312 stats->imissed = port_stats.rx_drop_red;
313 stats->imissed += port_stats.rx_drop_overrun;
314 stats->imissed += port_stats.rx_drop_bcast;
315 stats->imissed += port_stats.rx_drop_mcast;
316 stats->imissed += port_stats.rx_drop_l3_bcast;
317 stats->imissed += port_stats.rx_drop_l3_mcast;
319 stats->obytes = port_stats.tx_bytes_ok;
320 stats->opackets = port_stats.tx_ucast_frames_ok;
321 stats->opackets += port_stats.tx_bcast_frames_ok;
322 stats->opackets += port_stats.tx_mcast_frames_ok;
323 stats->oerrors = port_stats.tx_drops;
328 static const uint32_t *
329 nicvf_dev_supported_ptypes_get(struct rte_eth_dev *dev)
332 static uint32_t ptypes[32];
333 struct nicvf *nic = nicvf_pmd_priv(dev);
334 static const uint32_t ptypes_common[] = {
336 RTE_PTYPE_L3_IPV4_EXT,
338 RTE_PTYPE_L3_IPV6_EXT,
343 static const uint32_t ptypes_tunnel[] = {
344 RTE_PTYPE_TUNNEL_GRE,
345 RTE_PTYPE_TUNNEL_GENEVE,
346 RTE_PTYPE_TUNNEL_VXLAN,
347 RTE_PTYPE_TUNNEL_NVGRE,
349 static const uint32_t ptypes_end = RTE_PTYPE_UNKNOWN;
351 copied = sizeof(ptypes_common);
352 memcpy(ptypes, ptypes_common, copied);
353 if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
354 memcpy((char *)ptypes + copied, ptypes_tunnel,
355 sizeof(ptypes_tunnel));
356 copied += sizeof(ptypes_tunnel);
359 memcpy((char *)ptypes + copied, &ptypes_end, sizeof(ptypes_end));
361 /* All Ptypes are supported in all Rx functions. */
366 nicvf_dev_stats_reset(struct rte_eth_dev *dev)
369 uint16_t rxqs = 0, txqs = 0;
370 struct nicvf *nic = nicvf_pmd_priv(dev);
371 uint16_t rx_start, rx_end;
372 uint16_t tx_start, tx_end;
374 /* Reset all primary nic counters */
375 nicvf_rx_range(dev, nic, &rx_start, &rx_end);
376 for (i = rx_start; i <= rx_end; i++)
377 rxqs |= (0x3 << (i * 2));
379 nicvf_tx_range(dev, nic, &tx_start, &tx_end);
380 for (i = tx_start; i <= tx_end; i++)
381 txqs |= (0x3 << (i * 2));
383 nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, rxqs, txqs);
385 /* Reset secondary nic queue counters */
386 for (i = 0; i < nic->sqs_count; i++) {
387 struct nicvf *snic = nic->snicvf[i];
391 nicvf_rx_range(dev, snic, &rx_start, &rx_end);
392 for (i = rx_start; i <= rx_end; i++)
393 rxqs |= (0x3 << ((i % MAX_CMP_QUEUES_PER_QS) * 2));
395 nicvf_tx_range(dev, snic, &tx_start, &tx_end);
396 for (i = tx_start; i <= tx_end; i++)
397 txqs |= (0x3 << ((i % MAX_SND_QUEUES_PER_QS) * 2));
399 nicvf_mbox_reset_stat_counters(snic, 0, 0, rxqs, txqs);
403 /* Promiscuous mode enabled by default in LMAC to VF 1:1 map configuration */
405 nicvf_dev_promisc_enable(struct rte_eth_dev *dev __rte_unused)
409 static inline uint64_t
410 nicvf_rss_ethdev_to_nic(struct nicvf *nic, uint64_t ethdev_rss)
412 uint64_t nic_rss = 0;
414 if (ethdev_rss & ETH_RSS_IPV4)
415 nic_rss |= RSS_IP_ENA;
417 if (ethdev_rss & ETH_RSS_IPV6)
418 nic_rss |= RSS_IP_ENA;
420 if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_UDP)
421 nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
423 if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_TCP)
424 nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
426 if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_UDP)
427 nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
429 if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_TCP)
430 nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
432 if (ethdev_rss & ETH_RSS_PORT)
433 nic_rss |= RSS_L2_EXTENDED_HASH_ENA;
435 if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
436 if (ethdev_rss & ETH_RSS_VXLAN)
437 nic_rss |= RSS_TUN_VXLAN_ENA;
439 if (ethdev_rss & ETH_RSS_GENEVE)
440 nic_rss |= RSS_TUN_GENEVE_ENA;
442 if (ethdev_rss & ETH_RSS_NVGRE)
443 nic_rss |= RSS_TUN_NVGRE_ENA;
449 static inline uint64_t
450 nicvf_rss_nic_to_ethdev(struct nicvf *nic, uint64_t nic_rss)
452 uint64_t ethdev_rss = 0;
454 if (nic_rss & RSS_IP_ENA)
455 ethdev_rss |= (ETH_RSS_IPV4 | ETH_RSS_IPV6);
457 if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_TCP_ENA))
458 ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_TCP |
459 ETH_RSS_NONFRAG_IPV6_TCP);
461 if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_UDP_ENA))
462 ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_UDP |
463 ETH_RSS_NONFRAG_IPV6_UDP);
465 if (nic_rss & RSS_L2_EXTENDED_HASH_ENA)
466 ethdev_rss |= ETH_RSS_PORT;
468 if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
469 if (nic_rss & RSS_TUN_VXLAN_ENA)
470 ethdev_rss |= ETH_RSS_VXLAN;
472 if (nic_rss & RSS_TUN_GENEVE_ENA)
473 ethdev_rss |= ETH_RSS_GENEVE;
475 if (nic_rss & RSS_TUN_NVGRE_ENA)
476 ethdev_rss |= ETH_RSS_NVGRE;
482 nicvf_dev_reta_query(struct rte_eth_dev *dev,
483 struct rte_eth_rss_reta_entry64 *reta_conf,
486 struct nicvf *nic = nicvf_pmd_priv(dev);
487 uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
490 if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
491 RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
492 "(%d) doesn't match the number hardware can supported "
493 "(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
497 ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
501 /* Copy RETA table */
502 for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
503 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
504 if ((reta_conf[i].mask >> j) & 0x01)
505 reta_conf[i].reta[j] = tbl[j];
512 nicvf_dev_reta_update(struct rte_eth_dev *dev,
513 struct rte_eth_rss_reta_entry64 *reta_conf,
516 struct nicvf *nic = nicvf_pmd_priv(dev);
517 uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
520 if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
521 RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
522 "(%d) doesn't match the number hardware can supported "
523 "(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
527 ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
531 /* Copy RETA table */
532 for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
533 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
534 if ((reta_conf[i].mask >> j) & 0x01)
535 tbl[j] = reta_conf[i].reta[j];
538 return nicvf_rss_reta_update(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
542 nicvf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
543 struct rte_eth_rss_conf *rss_conf)
545 struct nicvf *nic = nicvf_pmd_priv(dev);
547 if (rss_conf->rss_key)
548 nicvf_rss_get_key(nic, rss_conf->rss_key);
550 rss_conf->rss_key_len = RSS_HASH_KEY_BYTE_SIZE;
551 rss_conf->rss_hf = nicvf_rss_nic_to_ethdev(nic, nicvf_rss_get_cfg(nic));
556 nicvf_dev_rss_hash_update(struct rte_eth_dev *dev,
557 struct rte_eth_rss_conf *rss_conf)
559 struct nicvf *nic = nicvf_pmd_priv(dev);
562 if (rss_conf->rss_key &&
563 rss_conf->rss_key_len != RSS_HASH_KEY_BYTE_SIZE) {
564 RTE_LOG(ERR, PMD, "Hash key size mismatch %d",
565 rss_conf->rss_key_len);
569 if (rss_conf->rss_key)
570 nicvf_rss_set_key(nic, rss_conf->rss_key);
572 nic_rss = nicvf_rss_ethdev_to_nic(nic, rss_conf->rss_hf);
573 nicvf_rss_set_cfg(nic, nic_rss);
578 nicvf_qset_cq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
579 struct nicvf_rxq *rxq, uint16_t qidx, uint32_t desc_cnt)
581 const struct rte_memzone *rz;
582 uint32_t ring_size = CMP_QUEUE_SZ_MAX * sizeof(union cq_entry_t);
584 rz = rte_eth_dma_zone_reserve(dev, "cq_ring",
585 nicvf_netdev_qidx(nic, qidx), ring_size,
586 NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
588 PMD_INIT_LOG(ERR, "Failed to allocate mem for cq hw ring");
592 memset(rz->addr, 0, ring_size);
594 rxq->phys = rz->iova;
595 rxq->desc = rz->addr;
596 rxq->qlen_mask = desc_cnt - 1;
602 nicvf_qset_sq_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
603 struct nicvf_txq *sq, uint16_t qidx, uint32_t desc_cnt)
605 const struct rte_memzone *rz;
606 uint32_t ring_size = SND_QUEUE_SZ_MAX * sizeof(union sq_entry_t);
608 rz = rte_eth_dma_zone_reserve(dev, "sq",
609 nicvf_netdev_qidx(nic, qidx), ring_size,
610 NICVF_SQ_BASE_ALIGN_BYTES, nic->node);
612 PMD_INIT_LOG(ERR, "Failed allocate mem for sq hw ring");
616 memset(rz->addr, 0, ring_size);
620 sq->qlen_mask = desc_cnt - 1;
626 nicvf_qset_rbdr_alloc(struct rte_eth_dev *dev, struct nicvf *nic,
627 uint32_t desc_cnt, uint32_t buffsz)
629 struct nicvf_rbdr *rbdr;
630 const struct rte_memzone *rz;
633 assert(nic->rbdr == NULL);
634 rbdr = rte_zmalloc_socket("rbdr", sizeof(struct nicvf_rbdr),
635 RTE_CACHE_LINE_SIZE, nic->node);
637 PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr");
641 ring_size = sizeof(struct rbdr_entry_t) * RBDR_QUEUE_SZ_MAX;
642 rz = rte_eth_dma_zone_reserve(dev, "rbdr",
643 nicvf_netdev_qidx(nic, 0), ring_size,
644 NICVF_RBDR_BASE_ALIGN_BYTES, nic->node);
646 PMD_INIT_LOG(ERR, "Failed to allocate mem for rbdr desc ring");
650 memset(rz->addr, 0, ring_size);
652 rbdr->phys = rz->iova;
655 rbdr->desc = rz->addr;
656 rbdr->buffsz = buffsz;
657 rbdr->qlen_mask = desc_cnt - 1;
659 nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_STATUS0;
661 nicvf_qset_base(nic, 0) + NIC_QSET_RBDR_0_1_DOOR;
668 nicvf_rbdr_release_mbuf(struct rte_eth_dev *dev, struct nicvf *nic,
669 nicvf_iova_addr_t phy)
673 struct nicvf_rxq *rxq;
674 uint16_t rx_start, rx_end;
676 /* Get queue ranges for this VF */
677 nicvf_rx_range(dev, nic, &rx_start, &rx_end);
679 for (qidx = rx_start; qidx <= rx_end; qidx++) {
680 rxq = dev->data->rx_queues[qidx];
681 if (rxq->precharge_cnt) {
682 obj = (void *)nicvf_mbuff_phy2virt(phy,
684 rte_mempool_put(rxq->pool, obj);
685 rxq->precharge_cnt--;
692 nicvf_rbdr_release_mbufs(struct rte_eth_dev *dev, struct nicvf *nic)
694 uint32_t qlen_mask, head;
695 struct rbdr_entry_t *entry;
696 struct nicvf_rbdr *rbdr = nic->rbdr;
698 qlen_mask = rbdr->qlen_mask;
700 while (head != rbdr->tail) {
701 entry = rbdr->desc + head;
702 nicvf_rbdr_release_mbuf(dev, nic, entry->full_addr);
704 head = head & qlen_mask;
709 nicvf_tx_queue_release_mbufs(struct nicvf_txq *txq)
714 while (head != txq->tail) {
715 if (txq->txbuffs[head]) {
716 rte_pktmbuf_free_seg(txq->txbuffs[head]);
717 txq->txbuffs[head] = NULL;
720 head = head & txq->qlen_mask;
725 nicvf_tx_queue_reset(struct nicvf_txq *txq)
727 uint32_t txq_desc_cnt = txq->qlen_mask + 1;
729 memset(txq->desc, 0, sizeof(union sq_entry_t) * txq_desc_cnt);
730 memset(txq->txbuffs, 0, sizeof(struct rte_mbuf *) * txq_desc_cnt);
737 nicvf_vf_start_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
740 struct nicvf_txq *txq;
743 assert(qidx < MAX_SND_QUEUES_PER_QS);
745 if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
746 RTE_ETH_QUEUE_STATE_STARTED)
749 txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
751 ret = nicvf_qset_sq_config(nic, qidx, txq);
753 PMD_INIT_LOG(ERR, "Failed to configure sq VF%d %d %d",
754 nic->vf_id, qidx, ret);
755 goto config_sq_error;
758 dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
759 RTE_ETH_QUEUE_STATE_STARTED;
763 nicvf_qset_sq_reclaim(nic, qidx);
768 nicvf_vf_stop_tx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
771 struct nicvf_txq *txq;
774 assert(qidx < MAX_SND_QUEUES_PER_QS);
776 if (dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
777 RTE_ETH_QUEUE_STATE_STOPPED)
780 ret = nicvf_qset_sq_reclaim(nic, qidx);
782 PMD_INIT_LOG(ERR, "Failed to reclaim sq VF%d %d %d",
783 nic->vf_id, qidx, ret);
785 txq = dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)];
786 nicvf_tx_queue_release_mbufs(txq);
787 nicvf_tx_queue_reset(txq);
789 dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
790 RTE_ETH_QUEUE_STATE_STOPPED;
795 nicvf_configure_cpi(struct rte_eth_dev *dev)
797 struct nicvf *nic = nicvf_pmd_priv(dev);
801 /* Count started rx queues */
802 for (qidx = qcnt = 0; qidx < dev->data->nb_rx_queues; qidx++)
803 if (dev->data->rx_queue_state[qidx] ==
804 RTE_ETH_QUEUE_STATE_STARTED)
807 nic->cpi_alg = CPI_ALG_NONE;
808 ret = nicvf_mbox_config_cpi(nic, qcnt);
810 PMD_INIT_LOG(ERR, "Failed to configure CPI %d", ret);
816 nicvf_configure_rss(struct rte_eth_dev *dev)
818 struct nicvf *nic = nicvf_pmd_priv(dev);
822 rsshf = nicvf_rss_ethdev_to_nic(nic,
823 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf);
824 PMD_DRV_LOG(INFO, "mode=%d rx_queues=%d loopback=%d rsshf=0x%" PRIx64,
825 dev->data->dev_conf.rxmode.mq_mode,
826 dev->data->nb_rx_queues,
827 dev->data->dev_conf.lpbk_mode, rsshf);
829 if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_NONE)
830 ret = nicvf_rss_term(nic);
831 else if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS)
832 ret = nicvf_rss_config(nic, dev->data->nb_rx_queues, rsshf);
834 PMD_INIT_LOG(ERR, "Failed to configure RSS %d", ret);
840 nicvf_configure_rss_reta(struct rte_eth_dev *dev)
842 struct nicvf *nic = nicvf_pmd_priv(dev);
843 unsigned int idx, qmap_size;
844 uint8_t qmap[RTE_MAX_QUEUES_PER_PORT];
845 uint8_t default_reta[NIC_MAX_RSS_IDR_TBL_SIZE];
847 if (nic->cpi_alg != CPI_ALG_NONE)
850 /* Prepare queue map */
851 for (idx = 0, qmap_size = 0; idx < dev->data->nb_rx_queues; idx++) {
852 if (dev->data->rx_queue_state[idx] ==
853 RTE_ETH_QUEUE_STATE_STARTED)
854 qmap[qmap_size++] = idx;
857 /* Update default RSS RETA */
858 for (idx = 0; idx < NIC_MAX_RSS_IDR_TBL_SIZE; idx++)
859 default_reta[idx] = qmap[idx % qmap_size];
861 return nicvf_rss_reta_update(nic, default_reta,
862 NIC_MAX_RSS_IDR_TBL_SIZE);
866 nicvf_dev_tx_queue_release(void *sq)
868 struct nicvf_txq *txq;
870 PMD_INIT_FUNC_TRACE();
872 txq = (struct nicvf_txq *)sq;
874 if (txq->txbuffs != NULL) {
875 nicvf_tx_queue_release_mbufs(txq);
876 rte_free(txq->txbuffs);
884 nicvf_set_tx_function(struct rte_eth_dev *dev)
886 struct nicvf_txq *txq = NULL;
888 bool multiseg = false;
890 for (i = 0; i < dev->data->nb_tx_queues; i++) {
891 txq = dev->data->tx_queues[i];
892 if (txq->offloads & DEV_TX_OFFLOAD_MULTI_SEGS) {
898 /* Use a simple Tx queue (no offloads, no multi segs) if possible */
900 PMD_DRV_LOG(DEBUG, "Using multi-segment tx callback");
901 dev->tx_pkt_burst = nicvf_xmit_pkts_multiseg;
903 PMD_DRV_LOG(DEBUG, "Using single-segment tx callback");
904 dev->tx_pkt_burst = nicvf_xmit_pkts;
910 if (txq->pool_free == nicvf_single_pool_free_xmited_buffers)
911 PMD_DRV_LOG(DEBUG, "Using single-mempool tx free method");
913 PMD_DRV_LOG(DEBUG, "Using multi-mempool tx free method");
917 nicvf_set_rx_function(struct rte_eth_dev *dev)
919 struct nicvf *nic = nicvf_pmd_priv(dev);
921 const eth_rx_burst_t rx_burst_func[2][2][2] = {
922 /* [NORMAL/SCATTER] [CKSUM/NO_CKSUM] [VLAN_STRIP/NO_VLAN_STRIP] */
923 [0][0][0] = nicvf_recv_pkts_no_offload,
924 [0][0][1] = nicvf_recv_pkts_vlan_strip,
925 [0][1][0] = nicvf_recv_pkts_cksum,
926 [0][1][1] = nicvf_recv_pkts_cksum_vlan_strip,
927 [1][0][0] = nicvf_recv_pkts_multiseg_no_offload,
928 [1][0][1] = nicvf_recv_pkts_multiseg_vlan_strip,
929 [1][1][0] = nicvf_recv_pkts_multiseg_cksum,
930 [1][1][1] = nicvf_recv_pkts_multiseg_cksum_vlan_strip,
934 rx_burst_func[dev->data->scattered_rx]
935 [nic->offload_cksum][nic->vlan_strip];
939 nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
940 uint16_t nb_desc, unsigned int socket_id,
941 const struct rte_eth_txconf *tx_conf)
943 uint16_t tx_free_thresh;
945 struct nicvf_txq *txq;
946 struct nicvf *nic = nicvf_pmd_priv(dev);
949 PMD_INIT_FUNC_TRACE();
951 if (qidx >= MAX_SND_QUEUES_PER_QS)
952 nic = nic->snicvf[qidx / MAX_SND_QUEUES_PER_QS - 1];
954 qidx = qidx % MAX_SND_QUEUES_PER_QS;
956 /* Socket id check */
957 if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
958 PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
959 socket_id, nic->node);
961 /* Tx deferred start is not supported */
962 if (tx_conf->tx_deferred_start) {
963 PMD_INIT_LOG(ERR, "Tx deferred start not supported");
967 /* Roundup nb_desc to available qsize and validate max number of desc */
968 nb_desc = nicvf_qsize_sq_roundup(nb_desc);
970 PMD_INIT_LOG(ERR, "Value of nb_desc beyond available sq qsize");
974 /* Validate tx_free_thresh */
975 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
976 tx_conf->tx_free_thresh :
977 NICVF_DEFAULT_TX_FREE_THRESH);
979 if (tx_free_thresh > (nb_desc) ||
980 tx_free_thresh > NICVF_MAX_TX_FREE_THRESH) {
982 "tx_free_thresh must be less than the number of TX "
983 "descriptors. (tx_free_thresh=%u port=%d "
984 "queue=%d)", (unsigned int)tx_free_thresh,
985 (int)dev->data->port_id, (int)qidx);
989 /* Free memory prior to re-allocation if needed. */
990 if (dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
991 PMD_TX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
992 nicvf_netdev_qidx(nic, qidx));
993 nicvf_dev_tx_queue_release(
994 dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)]);
995 dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
998 /* Allocating tx queue data structure */
999 txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct nicvf_txq),
1000 RTE_CACHE_LINE_SIZE, nic->node);
1002 PMD_INIT_LOG(ERR, "Failed to allocate txq=%d",
1003 nicvf_netdev_qidx(nic, qidx));
1008 txq->queue_id = qidx;
1009 txq->tx_free_thresh = tx_free_thresh;
1010 txq->sq_head = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_HEAD;
1011 txq->sq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_DOOR;
1012 offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1013 txq->offloads = offloads;
1015 is_single_pool = !!(offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE);
1017 /* Choose optimum free threshold value for multipool case */
1018 if (!is_single_pool) {
1019 txq->tx_free_thresh = (uint16_t)
1020 (tx_conf->tx_free_thresh == NICVF_DEFAULT_TX_FREE_THRESH ?
1021 NICVF_TX_FREE_MPOOL_THRESH :
1022 tx_conf->tx_free_thresh);
1023 txq->pool_free = nicvf_multi_pool_free_xmited_buffers;
1025 txq->pool_free = nicvf_single_pool_free_xmited_buffers;
1028 /* Allocate software ring */
1029 txq->txbuffs = rte_zmalloc_socket("txq->txbuffs",
1030 nb_desc * sizeof(struct rte_mbuf *),
1031 RTE_CACHE_LINE_SIZE, nic->node);
1033 if (txq->txbuffs == NULL) {
1034 nicvf_dev_tx_queue_release(txq);
1038 if (nicvf_qset_sq_alloc(dev, nic, txq, qidx, nb_desc)) {
1039 PMD_INIT_LOG(ERR, "Failed to allocate mem for sq %d", qidx);
1040 nicvf_dev_tx_queue_release(txq);
1044 nicvf_tx_queue_reset(txq);
1046 PMD_INIT_LOG(DEBUG, "[%d] txq=%p nb_desc=%d desc=%p"
1047 " phys=0x%" PRIx64 " offloads=0x%" PRIx64,
1048 nicvf_netdev_qidx(nic, qidx), txq, nb_desc, txq->desc,
1049 txq->phys, txq->offloads);
1051 dev->data->tx_queues[nicvf_netdev_qidx(nic, qidx)] = txq;
1052 dev->data->tx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1053 RTE_ETH_QUEUE_STATE_STOPPED;
1058 nicvf_rx_queue_release_mbufs(struct rte_eth_dev *dev, struct nicvf_rxq *rxq)
1061 uint32_t nb_pkts, released_pkts = 0;
1062 uint32_t refill_cnt = 0;
1063 struct rte_mbuf *rx_pkts[NICVF_MAX_RX_FREE_THRESH];
1065 if (dev->rx_pkt_burst == NULL)
1068 while ((rxq_cnt = nicvf_dev_rx_queue_count(dev,
1069 nicvf_netdev_qidx(rxq->nic, rxq->queue_id)))) {
1070 nb_pkts = dev->rx_pkt_burst(rxq, rx_pkts,
1071 NICVF_MAX_RX_FREE_THRESH);
1072 PMD_DRV_LOG(INFO, "nb_pkts=%d rxq_cnt=%d", nb_pkts, rxq_cnt);
1074 rte_pktmbuf_free_seg(rx_pkts[--nb_pkts]);
1080 refill_cnt += nicvf_dev_rbdr_refill(dev,
1081 nicvf_netdev_qidx(rxq->nic, rxq->queue_id));
1083 PMD_DRV_LOG(INFO, "free_cnt=%d refill_cnt=%d",
1084 released_pkts, refill_cnt);
1088 nicvf_rx_queue_reset(struct nicvf_rxq *rxq)
1091 rxq->available_space = 0;
1092 rxq->recv_buffers = 0;
1096 nicvf_vf_start_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
1099 struct nicvf_rxq *rxq;
1102 assert(qidx < MAX_RCV_QUEUES_PER_QS);
1104 if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1105 RTE_ETH_QUEUE_STATE_STARTED)
1108 /* Update rbdr pointer to all rxq */
1109 rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1110 rxq->shared_rbdr = nic->rbdr;
1112 ret = nicvf_qset_rq_config(nic, qidx, rxq);
1114 PMD_INIT_LOG(ERR, "Failed to configure rq VF%d %d %d",
1115 nic->vf_id, qidx, ret);
1116 goto config_rq_error;
1118 ret = nicvf_qset_cq_config(nic, qidx, rxq);
1120 PMD_INIT_LOG(ERR, "Failed to configure cq VF%d %d %d",
1121 nic->vf_id, qidx, ret);
1122 goto config_cq_error;
1125 dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1126 RTE_ETH_QUEUE_STATE_STARTED;
1130 nicvf_qset_cq_reclaim(nic, qidx);
1132 nicvf_qset_rq_reclaim(nic, qidx);
1137 nicvf_vf_stop_rx_queue(struct rte_eth_dev *dev, struct nicvf *nic,
1140 struct nicvf_rxq *rxq;
1141 int ret, other_error;
1143 if (dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] ==
1144 RTE_ETH_QUEUE_STATE_STOPPED)
1147 ret = nicvf_qset_rq_reclaim(nic, qidx);
1149 PMD_INIT_LOG(ERR, "Failed to reclaim rq VF%d %d %d",
1150 nic->vf_id, qidx, ret);
1153 rxq = dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)];
1154 nicvf_rx_queue_release_mbufs(dev, rxq);
1155 nicvf_rx_queue_reset(rxq);
1157 ret = nicvf_qset_cq_reclaim(nic, qidx);
1159 PMD_INIT_LOG(ERR, "Failed to reclaim cq VF%d %d %d",
1160 nic->vf_id, qidx, ret);
1163 dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1164 RTE_ETH_QUEUE_STATE_STOPPED;
1169 nicvf_dev_rx_queue_release(void *rx_queue)
1171 PMD_INIT_FUNC_TRACE();
1177 nicvf_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1179 struct nicvf *nic = nicvf_pmd_priv(dev);
1182 if (qidx >= MAX_RCV_QUEUES_PER_QS)
1183 nic = nic->snicvf[(qidx / MAX_RCV_QUEUES_PER_QS - 1)];
1185 qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1187 ret = nicvf_vf_start_rx_queue(dev, nic, qidx);
1191 ret = nicvf_configure_cpi(dev);
1195 return nicvf_configure_rss_reta(dev);
1199 nicvf_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1202 struct nicvf *nic = nicvf_pmd_priv(dev);
1204 if (qidx >= MAX_SND_QUEUES_PER_QS)
1205 nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1207 qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1209 ret = nicvf_vf_stop_rx_queue(dev, nic, qidx);
1210 ret |= nicvf_configure_cpi(dev);
1211 ret |= nicvf_configure_rss_reta(dev);
1216 nicvf_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t qidx)
1218 struct nicvf *nic = nicvf_pmd_priv(dev);
1220 if (qidx >= MAX_SND_QUEUES_PER_QS)
1221 nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1223 qidx = qidx % MAX_SND_QUEUES_PER_QS;
1225 return nicvf_vf_start_tx_queue(dev, nic, qidx);
1229 nicvf_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t qidx)
1231 struct nicvf *nic = nicvf_pmd_priv(dev);
1233 if (qidx >= MAX_SND_QUEUES_PER_QS)
1234 nic = nic->snicvf[(qidx / MAX_SND_QUEUES_PER_QS - 1)];
1236 qidx = qidx % MAX_SND_QUEUES_PER_QS;
1238 return nicvf_vf_stop_tx_queue(dev, nic, qidx);
1242 nicvf_rxq_mbuf_setup(struct nicvf_rxq *rxq)
1245 struct rte_mbuf mb_def;
1246 struct nicvf *nic = rxq->nic;
1248 RTE_BUILD_BUG_ON(sizeof(union mbuf_initializer) != 8);
1249 RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, data_off) % 8 != 0);
1250 RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, refcnt) -
1251 offsetof(struct rte_mbuf, data_off) != 2);
1252 RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, nb_segs) -
1253 offsetof(struct rte_mbuf, data_off) != 4);
1254 RTE_BUILD_BUG_ON(offsetof(struct rte_mbuf, port) -
1255 offsetof(struct rte_mbuf, data_off) != 6);
1256 RTE_BUILD_BUG_ON(offsetof(struct nicvf_rxq, rxq_fastpath_data_end) -
1257 offsetof(struct nicvf_rxq,
1258 rxq_fastpath_data_start) > 128);
1260 mb_def.data_off = RTE_PKTMBUF_HEADROOM + (nic->skip_bytes);
1261 mb_def.port = rxq->port_id;
1262 rte_mbuf_refcnt_set(&mb_def, 1);
1264 /* Prevent compiler reordering: rearm_data covers previous fields */
1265 rte_compiler_barrier();
1266 p = (uintptr_t)&mb_def.rearm_data;
1267 rxq->mbuf_initializer.value = *(uint64_t *)p;
1271 nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
1272 uint16_t nb_desc, unsigned int socket_id,
1273 const struct rte_eth_rxconf *rx_conf,
1274 struct rte_mempool *mp)
1276 uint16_t rx_free_thresh;
1277 struct nicvf_rxq *rxq;
1278 struct nicvf *nic = nicvf_pmd_priv(dev);
1281 struct rte_pktmbuf_pool_private *mbp_priv;
1283 PMD_INIT_FUNC_TRACE();
1285 /* First skip check */
1286 mbp_priv = rte_mempool_get_priv(mp);
1287 buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
1288 if (buffsz < (uint32_t)(nic->skip_bytes)) {
1289 PMD_INIT_LOG(ERR, "First skip is more than configured buffer size");
1293 if (qidx >= MAX_RCV_QUEUES_PER_QS)
1294 nic = nic->snicvf[qidx / MAX_RCV_QUEUES_PER_QS - 1];
1296 qidx = qidx % MAX_RCV_QUEUES_PER_QS;
1298 /* Socket id check */
1299 if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
1300 PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
1301 socket_id, nic->node);
1303 /* Mempool memory must be contiguous, so must be one memory segment*/
1304 if (mp->nb_mem_chunks != 1) {
1305 PMD_INIT_LOG(ERR, "Non-contiguous mempool, add more huge pages");
1309 /* Mempool memory must be physically contiguous */
1310 if (mp->flags & MEMPOOL_F_NO_IOVA_CONTIG) {
1311 PMD_INIT_LOG(ERR, "Mempool memory must be physically contiguous");
1315 /* Rx deferred start is not supported */
1316 if (rx_conf->rx_deferred_start) {
1317 PMD_INIT_LOG(ERR, "Rx deferred start not supported");
1321 /* Roundup nb_desc to available qsize and validate max number of desc */
1322 nb_desc = nicvf_qsize_cq_roundup(nb_desc);
1324 PMD_INIT_LOG(ERR, "Value nb_desc beyond available hw cq qsize");
1329 /* Check rx_free_thresh upper bound */
1330 rx_free_thresh = (uint16_t)((rx_conf->rx_free_thresh) ?
1331 rx_conf->rx_free_thresh :
1332 NICVF_DEFAULT_RX_FREE_THRESH);
1333 if (rx_free_thresh > NICVF_MAX_RX_FREE_THRESH ||
1334 rx_free_thresh >= nb_desc * .75) {
1335 PMD_INIT_LOG(ERR, "rx_free_thresh greater than expected %d",
1340 /* Free memory prior to re-allocation if needed */
1341 if (dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] != NULL) {
1342 PMD_RX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
1343 nicvf_netdev_qidx(nic, qidx));
1344 nicvf_dev_rx_queue_release(
1345 dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)]);
1346 dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = NULL;
1349 /* Allocate rxq memory */
1350 rxq = rte_zmalloc_socket("ethdev rx queue", sizeof(struct nicvf_rxq),
1351 RTE_CACHE_LINE_SIZE, nic->node);
1353 PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d",
1354 nicvf_netdev_qidx(nic, qidx));
1360 rxq->queue_id = qidx;
1361 rxq->port_id = dev->data->port_id;
1362 rxq->rx_free_thresh = rx_free_thresh;
1363 rxq->rx_drop_en = rx_conf->rx_drop_en;
1364 rxq->cq_status = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_STATUS;
1365 rxq->cq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_DOOR;
1366 rxq->precharge_cnt = 0;
1368 if (nicvf_hw_cap(nic) & NICVF_CAP_CQE_RX2)
1369 rxq->rbptr_offset = NICVF_CQE_RX2_RBPTR_WORD;
1371 rxq->rbptr_offset = NICVF_CQE_RBPTR_WORD;
1373 nicvf_rxq_mbuf_setup(rxq);
1375 /* Alloc completion queue */
1376 if (nicvf_qset_cq_alloc(dev, nic, rxq, rxq->queue_id, nb_desc)) {
1377 PMD_INIT_LOG(ERR, "failed to allocate cq %u", rxq->queue_id);
1378 nicvf_dev_rx_queue_release(rxq);
1382 nicvf_rx_queue_reset(rxq);
1384 offloads = rx_conf->offloads | dev->data->dev_conf.rxmode.offloads;
1385 PMD_INIT_LOG(DEBUG, "[%d] rxq=%p pool=%s nb_desc=(%d/%d)"
1386 " phy=0x%" PRIx64 " offloads=0x%" PRIx64,
1387 nicvf_netdev_qidx(nic, qidx), rxq, mp->name, nb_desc,
1388 rte_mempool_avail_count(mp), rxq->phys, offloads);
1390 dev->data->rx_queues[nicvf_netdev_qidx(nic, qidx)] = rxq;
1391 dev->data->rx_queue_state[nicvf_netdev_qidx(nic, qidx)] =
1392 RTE_ETH_QUEUE_STATE_STOPPED;
1397 nicvf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1399 struct nicvf *nic = nicvf_pmd_priv(dev);
1400 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1402 PMD_INIT_FUNC_TRACE();
1404 /* Autonegotiation may be disabled */
1405 dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
1406 dev_info->speed_capa |= ETH_LINK_SPEED_10M | ETH_LINK_SPEED_100M |
1407 ETH_LINK_SPEED_1G | ETH_LINK_SPEED_10G;
1408 if (nicvf_hw_version(nic) != PCI_SUB_DEVICE_ID_CN81XX_NICVF)
1409 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
1411 dev_info->min_rx_bufsize = ETHER_MIN_MTU;
1412 dev_info->max_rx_pktlen = NIC_HW_MAX_MTU + ETHER_HDR_LEN;
1413 dev_info->max_rx_queues =
1414 (uint16_t)MAX_RCV_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
1415 dev_info->max_tx_queues =
1416 (uint16_t)MAX_SND_QUEUES_PER_QS * (MAX_SQS_PER_VF + 1);
1417 dev_info->max_mac_addrs = 1;
1418 dev_info->max_vfs = pci_dev->max_vfs;
1420 dev_info->rx_offload_capa = NICVF_RX_OFFLOAD_CAPA;
1421 dev_info->tx_offload_capa = NICVF_TX_OFFLOAD_CAPA;
1422 dev_info->rx_queue_offload_capa = NICVF_RX_OFFLOAD_CAPA;
1423 dev_info->tx_queue_offload_capa = NICVF_TX_OFFLOAD_CAPA;
1425 dev_info->reta_size = nic->rss_info.rss_size;
1426 dev_info->hash_key_size = RSS_HASH_KEY_BYTE_SIZE;
1427 dev_info->flow_type_rss_offloads = NICVF_RSS_OFFLOAD_PASS1;
1428 if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING)
1429 dev_info->flow_type_rss_offloads |= NICVF_RSS_OFFLOAD_TUNNEL;
1431 dev_info->default_rxconf = (struct rte_eth_rxconf) {
1432 .rx_free_thresh = NICVF_DEFAULT_RX_FREE_THRESH,
1434 .offloads = DEV_RX_OFFLOAD_CRC_STRIP,
1437 dev_info->default_txconf = (struct rte_eth_txconf) {
1438 .tx_free_thresh = NICVF_DEFAULT_TX_FREE_THRESH,
1439 .offloads = DEV_TX_OFFLOAD_MBUF_FAST_FREE |
1440 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
1441 DEV_TX_OFFLOAD_UDP_CKSUM |
1442 DEV_TX_OFFLOAD_TCP_CKSUM,
1446 static nicvf_iova_addr_t
1447 rbdr_rte_mempool_get(void *dev, void *opaque)
1451 struct nicvf_rxq *rxq;
1452 struct rte_eth_dev *eth_dev = (struct rte_eth_dev *)dev;
1453 struct nicvf *nic = (struct nicvf *)opaque;
1454 uint16_t rx_start, rx_end;
1456 /* Get queue ranges for this VF */
1457 nicvf_rx_range(eth_dev, nic, &rx_start, &rx_end);
1459 for (qidx = rx_start; qidx <= rx_end; qidx++) {
1460 rxq = eth_dev->data->rx_queues[qidx];
1461 /* Maintain equal buffer count across all pools */
1462 if (rxq->precharge_cnt >= rxq->qlen_mask)
1464 rxq->precharge_cnt++;
1465 mbuf = (uintptr_t)rte_pktmbuf_alloc(rxq->pool);
1467 return nicvf_mbuff_virt2phy(mbuf, rxq->mbuf_phys_off);
1473 nicvf_vf_start(struct rte_eth_dev *dev, struct nicvf *nic, uint32_t rbdrsz)
1476 uint16_t qidx, data_off;
1477 uint32_t total_rxq_desc, nb_rbdr_desc, exp_buffs;
1478 uint64_t mbuf_phys_off = 0;
1479 struct nicvf_rxq *rxq;
1480 struct rte_mbuf *mbuf;
1481 uint16_t rx_start, rx_end;
1482 uint16_t tx_start, tx_end;
1485 PMD_INIT_FUNC_TRACE();
1487 /* Userspace process exited without proper shutdown in last run */
1488 if (nicvf_qset_rbdr_active(nic, 0))
1489 nicvf_vf_stop(dev, nic, false);
1491 /* Get queue ranges for this VF */
1492 nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1495 * Thunderx nicvf PMD can support more than one pool per port only when
1496 * 1) Data payload size is same across all the pools in given port
1498 * 2) All mbuffs in the pools are from the same hugepage
1500 * 3) Mbuff metadata size is same across all the pools in given port
1502 * This is to support existing application that uses multiple pool/port.
1503 * But, the purpose of using multipool for QoS will not be addressed.
1507 /* Validate mempool attributes */
1508 for (qidx = rx_start; qidx <= rx_end; qidx++) {
1509 rxq = dev->data->rx_queues[qidx];
1510 rxq->mbuf_phys_off = nicvf_mempool_phy_offset(rxq->pool);
1511 mbuf = rte_pktmbuf_alloc(rxq->pool);
1513 PMD_INIT_LOG(ERR, "Failed allocate mbuf VF%d qid=%d "
1515 nic->vf_id, qidx, rxq->pool->name);
1518 data_off = nicvf_mbuff_meta_length(mbuf);
1519 data_off += RTE_PKTMBUF_HEADROOM;
1520 rte_pktmbuf_free(mbuf);
1522 if (data_off % RTE_CACHE_LINE_SIZE) {
1523 PMD_INIT_LOG(ERR, "%s: unaligned data_off=%d delta=%d",
1524 rxq->pool->name, data_off,
1525 data_off % RTE_CACHE_LINE_SIZE);
1528 rxq->mbuf_phys_off -= data_off;
1529 rxq->mbuf_phys_off -= nic->skip_bytes;
1531 if (mbuf_phys_off == 0)
1532 mbuf_phys_off = rxq->mbuf_phys_off;
1533 if (mbuf_phys_off != rxq->mbuf_phys_off) {
1534 PMD_INIT_LOG(ERR, "pool params not same,%s VF%d %"
1535 PRIx64, rxq->pool->name, nic->vf_id,
1541 /* Check the level of buffers in the pool */
1543 for (qidx = rx_start; qidx <= rx_end; qidx++) {
1544 rxq = dev->data->rx_queues[qidx];
1545 /* Count total numbers of rxq descs */
1546 total_rxq_desc += rxq->qlen_mask + 1;
1547 exp_buffs = RTE_MEMPOOL_CACHE_MAX_SIZE + rxq->rx_free_thresh;
1548 exp_buffs *= dev->data->nb_rx_queues;
1549 if (rte_mempool_avail_count(rxq->pool) < exp_buffs) {
1550 PMD_INIT_LOG(ERR, "Buff shortage in pool=%s (%d/%d)",
1552 rte_mempool_avail_count(rxq->pool),
1558 /* Check RBDR desc overflow */
1559 ret = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1561 PMD_INIT_LOG(ERR, "Reached RBDR desc limit, reduce nr desc "
1562 "VF%d", nic->vf_id);
1567 ret = nicvf_qset_config(nic);
1569 PMD_INIT_LOG(ERR, "Failed to enable qset %d VF%d", ret,
1574 /* Allocate RBDR and RBDR ring desc */
1575 nb_rbdr_desc = nicvf_qsize_rbdr_roundup(total_rxq_desc);
1576 ret = nicvf_qset_rbdr_alloc(dev, nic, nb_rbdr_desc, rbdrsz);
1578 PMD_INIT_LOG(ERR, "Failed to allocate memory for rbdr alloc "
1579 "VF%d", nic->vf_id);
1583 /* Enable and configure RBDR registers */
1584 ret = nicvf_qset_rbdr_config(nic, 0);
1586 PMD_INIT_LOG(ERR, "Failed to configure rbdr %d VF%d", ret,
1588 goto qset_rbdr_free;
1591 /* Fill rte_mempool buffers in RBDR pool and precharge it */
1592 ret = nicvf_qset_rbdr_precharge(dev, nic, 0, rbdr_rte_mempool_get,
1595 PMD_INIT_LOG(ERR, "Failed to fill rbdr %d VF%d", ret,
1597 goto qset_rbdr_reclaim;
1600 PMD_DRV_LOG(INFO, "Filled %d out of %d entries in RBDR VF%d",
1601 nic->rbdr->tail, nb_rbdr_desc, nic->vf_id);
1603 /* Configure VLAN Strip */
1604 mask = ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK |
1605 ETH_VLAN_EXTEND_MASK;
1606 ret = nicvf_vlan_offload_config(dev, mask);
1608 /* Based on the packet type(IPv4 or IPv6), the nicvf HW aligns L3 data
1609 * to the 64bit memory address.
1610 * The alignment creates a hole in mbuf(between the end of headroom and
1611 * packet data start). The new revision of the HW provides an option to
1612 * disable the L3 alignment feature and make mbuf layout looks
1613 * more like other NICs. For better application compatibility, disabling
1614 * l3 alignment feature on the hardware revisions it supports
1616 nicvf_apad_config(nic, false);
1618 /* Get queue ranges for this VF */
1619 nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1621 /* Configure TX queues */
1622 for (qidx = tx_start; qidx <= tx_end; qidx++) {
1623 ret = nicvf_vf_start_tx_queue(dev, nic,
1624 qidx % MAX_SND_QUEUES_PER_QS);
1626 goto start_txq_error;
1629 /* Configure RX queues */
1630 for (qidx = rx_start; qidx <= rx_end; qidx++) {
1631 ret = nicvf_vf_start_rx_queue(dev, nic,
1632 qidx % MAX_RCV_QUEUES_PER_QS);
1634 goto start_rxq_error;
1637 if (!nic->sqs_mode) {
1638 /* Configure CPI algorithm */
1639 ret = nicvf_configure_cpi(dev);
1641 goto start_txq_error;
1643 ret = nicvf_mbox_get_rss_size(nic);
1645 PMD_INIT_LOG(ERR, "Failed to get rss table size");
1646 goto qset_rss_error;
1650 ret = nicvf_configure_rss(dev);
1652 goto qset_rss_error;
1655 /* Done; Let PF make the BGX's RX and TX switches to ON position */
1656 nicvf_mbox_cfg_done(nic);
1660 nicvf_rss_term(nic);
1662 for (qidx = rx_start; qidx <= rx_end; qidx++)
1663 nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1665 for (qidx = tx_start; qidx <= tx_end; qidx++)
1666 nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1668 nicvf_qset_rbdr_reclaim(nic, 0);
1669 nicvf_rbdr_release_mbufs(dev, nic);
1672 rte_free(nic->rbdr);
1676 nicvf_qset_reclaim(nic);
1681 nicvf_dev_start(struct rte_eth_dev *dev)
1686 struct nicvf *nic = nicvf_pmd_priv(dev);
1687 struct rte_eth_rxmode *rx_conf = &dev->data->dev_conf.rxmode;
1689 uint32_t buffsz = 0, rbdrsz = 0;
1690 struct rte_pktmbuf_pool_private *mbp_priv;
1691 struct nicvf_rxq *rxq;
1693 PMD_INIT_FUNC_TRACE();
1695 /* This function must be called for a primary device */
1696 assert_primary(nic);
1698 /* Validate RBDR buff size */
1699 for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
1700 rxq = dev->data->rx_queues[qidx];
1701 mbp_priv = rte_mempool_get_priv(rxq->pool);
1702 buffsz = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM;
1704 PMD_INIT_LOG(ERR, "rxbuf size must be multiply of 128");
1709 if (rbdrsz != buffsz) {
1710 PMD_INIT_LOG(ERR, "buffsz not same, qidx=%d (%d/%d)",
1711 qidx, rbdrsz, buffsz);
1716 /* Configure loopback */
1717 ret = nicvf_loopback_config(nic, dev->data->dev_conf.lpbk_mode);
1719 PMD_INIT_LOG(ERR, "Failed to configure loopback %d", ret);
1723 /* Reset all statistics counters attached to this port */
1724 ret = nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, 0xFFFF, 0xFFFF);
1726 PMD_INIT_LOG(ERR, "Failed to reset stat counters %d", ret);
1730 /* Setup scatter mode if needed by jumbo */
1731 if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
1732 2 * VLAN_TAG_SIZE > buffsz)
1733 dev->data->scattered_rx = 1;
1734 if ((rx_conf->offloads & DEV_RX_OFFLOAD_SCATTER) != 0)
1735 dev->data->scattered_rx = 1;
1737 /* Setup MTU based on max_rx_pkt_len or default */
1738 mtu = dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME ?
1739 dev->data->dev_conf.rxmode.max_rx_pkt_len
1740 - ETHER_HDR_LEN : ETHER_MTU;
1742 if (nicvf_dev_set_mtu(dev, mtu)) {
1743 PMD_INIT_LOG(ERR, "Failed to set default mtu size");
1747 ret = nicvf_vf_start(dev, nic, rbdrsz);
1751 for (i = 0; i < nic->sqs_count; i++) {
1752 assert(nic->snicvf[i]);
1754 ret = nicvf_vf_start(dev, nic->snicvf[i], rbdrsz);
1759 /* Configure callbacks based on offloads */
1760 nicvf_set_tx_function(dev);
1761 nicvf_set_rx_function(dev);
1767 nicvf_dev_stop_cleanup(struct rte_eth_dev *dev, bool cleanup)
1771 struct nicvf *nic = nicvf_pmd_priv(dev);
1773 PMD_INIT_FUNC_TRACE();
1775 /* Teardown secondary vf first */
1776 for (i = 0; i < nic->sqs_count; i++) {
1777 if (!nic->snicvf[i])
1780 nicvf_vf_stop(dev, nic->snicvf[i], cleanup);
1783 /* Stop the primary VF now */
1784 nicvf_vf_stop(dev, nic, cleanup);
1786 /* Disable loopback */
1787 ret = nicvf_loopback_config(nic, 0);
1789 PMD_INIT_LOG(ERR, "Failed to disable loopback %d", ret);
1791 /* Reclaim CPI configuration */
1792 ret = nicvf_mbox_config_cpi(nic, 0);
1794 PMD_INIT_LOG(ERR, "Failed to reclaim CPI config %d", ret);
1798 nicvf_dev_stop(struct rte_eth_dev *dev)
1800 PMD_INIT_FUNC_TRACE();
1802 nicvf_dev_stop_cleanup(dev, false);
1806 nicvf_vf_stop(struct rte_eth_dev *dev, struct nicvf *nic, bool cleanup)
1810 uint16_t tx_start, tx_end;
1811 uint16_t rx_start, rx_end;
1813 PMD_INIT_FUNC_TRACE();
1816 /* Let PF make the BGX's RX and TX switches to OFF position */
1817 nicvf_mbox_shutdown(nic);
1820 /* Disable VLAN Strip */
1821 nicvf_vlan_hw_strip(nic, 0);
1823 /* Get queue ranges for this VF */
1824 nicvf_tx_range(dev, nic, &tx_start, &tx_end);
1826 for (qidx = tx_start; qidx <= tx_end; qidx++)
1827 nicvf_vf_stop_tx_queue(dev, nic, qidx % MAX_SND_QUEUES_PER_QS);
1829 /* Get queue ranges for this VF */
1830 nicvf_rx_range(dev, nic, &rx_start, &rx_end);
1833 for (qidx = rx_start; qidx <= rx_end; qidx++)
1834 nicvf_vf_stop_rx_queue(dev, nic, qidx % MAX_RCV_QUEUES_PER_QS);
1837 ret = nicvf_qset_rbdr_reclaim(nic, 0);
1839 PMD_INIT_LOG(ERR, "Failed to reclaim RBDR %d", ret);
1841 /* Move all charged buffers in RBDR back to pool */
1842 if (nic->rbdr != NULL)
1843 nicvf_rbdr_release_mbufs(dev, nic);
1846 ret = nicvf_qset_reclaim(nic);
1848 PMD_INIT_LOG(ERR, "Failed to disable qset %d", ret);
1850 /* Disable all interrupts */
1851 nicvf_disable_all_interrupts(nic);
1853 /* Free RBDR SW structure */
1855 rte_free(nic->rbdr);
1861 nicvf_dev_close(struct rte_eth_dev *dev)
1864 struct nicvf *nic = nicvf_pmd_priv(dev);
1866 PMD_INIT_FUNC_TRACE();
1868 nicvf_dev_stop_cleanup(dev, true);
1869 nicvf_periodic_alarm_stop(nicvf_interrupt, dev);
1871 for (i = 0; i < nic->sqs_count; i++) {
1872 if (!nic->snicvf[i])
1875 nicvf_periodic_alarm_stop(nicvf_vf_interrupt, nic->snicvf[i]);
1880 nicvf_request_sqs(struct nicvf *nic)
1884 assert_primary(nic);
1885 assert(nic->sqs_count > 0);
1886 assert(nic->sqs_count <= MAX_SQS_PER_VF);
1888 /* Set no of Rx/Tx queues in each of the SQsets */
1889 for (i = 0; i < nic->sqs_count; i++) {
1890 if (nicvf_svf_empty())
1891 rte_panic("Cannot assign sufficient number of "
1892 "secondary queues to primary VF%" PRIu8 "\n",
1895 nic->snicvf[i] = nicvf_svf_pop();
1896 nic->snicvf[i]->sqs_id = i;
1899 return nicvf_mbox_request_sqs(nic);
1903 nicvf_dev_configure(struct rte_eth_dev *dev)
1905 struct rte_eth_dev_data *data = dev->data;
1906 struct rte_eth_conf *conf = &data->dev_conf;
1907 struct rte_eth_rxmode *rxmode = &conf->rxmode;
1908 struct rte_eth_txmode *txmode = &conf->txmode;
1909 struct nicvf *nic = nicvf_pmd_priv(dev);
1912 PMD_INIT_FUNC_TRACE();
1914 if (!rte_eal_has_hugepages()) {
1915 PMD_INIT_LOG(INFO, "Huge page is not configured");
1919 /* KEEP_CRC offload flag is not supported by PMD
1920 * can remove the below block when DEV_RX_OFFLOAD_CRC_STRIP removed
1922 if (rte_eth_dev_must_keep_crc(rxmode->offloads)) {
1923 PMD_INIT_LOG(NOTICE, "Can't disable hw crc strip");
1924 rxmode->offloads |= DEV_RX_OFFLOAD_CRC_STRIP;
1927 if (txmode->mq_mode) {
1928 PMD_INIT_LOG(INFO, "Tx mq_mode DCB or VMDq not supported");
1932 if (rxmode->mq_mode != ETH_MQ_RX_NONE &&
1933 rxmode->mq_mode != ETH_MQ_RX_RSS) {
1934 PMD_INIT_LOG(INFO, "Unsupported rx qmode %d", rxmode->mq_mode);
1938 if (rxmode->split_hdr_size) {
1939 PMD_INIT_LOG(INFO, "Rxmode does not support split header");
1943 if (conf->link_speeds & ETH_LINK_SPEED_FIXED) {
1944 PMD_INIT_LOG(INFO, "Setting link speed/duplex not supported");
1948 if (conf->dcb_capability_en) {
1949 PMD_INIT_LOG(INFO, "DCB enable not supported");
1953 if (conf->fdir_conf.mode != RTE_FDIR_MODE_NONE) {
1954 PMD_INIT_LOG(INFO, "Flow director not supported");
1958 assert_primary(nic);
1959 NICVF_STATIC_ASSERT(MAX_RCV_QUEUES_PER_QS == MAX_SND_QUEUES_PER_QS);
1960 cqcount = RTE_MAX(data->nb_tx_queues, data->nb_rx_queues);
1961 if (cqcount > MAX_RCV_QUEUES_PER_QS) {
1962 nic->sqs_count = RTE_ALIGN_CEIL(cqcount, MAX_RCV_QUEUES_PER_QS);
1963 nic->sqs_count = (nic->sqs_count / MAX_RCV_QUEUES_PER_QS) - 1;
1968 assert(nic->sqs_count <= MAX_SQS_PER_VF);
1970 if (nic->sqs_count > 0) {
1971 if (nicvf_request_sqs(nic)) {
1972 rte_panic("Cannot assign sufficient number of "
1973 "secondary queues to PORT%d VF%" PRIu8 "\n",
1974 dev->data->port_id, nic->vf_id);
1978 if (rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
1979 nic->offload_cksum = 1;
1981 PMD_INIT_LOG(DEBUG, "Configured ethdev port%d hwcap=0x%" PRIx64,
1982 dev->data->port_id, nicvf_hw_cap(nic));
1987 /* Initialize and register driver with DPDK Application */
1988 static const struct eth_dev_ops nicvf_eth_dev_ops = {
1989 .dev_configure = nicvf_dev_configure,
1990 .dev_start = nicvf_dev_start,
1991 .dev_stop = nicvf_dev_stop,
1992 .link_update = nicvf_dev_link_update,
1993 .dev_close = nicvf_dev_close,
1994 .stats_get = nicvf_dev_stats_get,
1995 .stats_reset = nicvf_dev_stats_reset,
1996 .promiscuous_enable = nicvf_dev_promisc_enable,
1997 .dev_infos_get = nicvf_dev_info_get,
1998 .dev_supported_ptypes_get = nicvf_dev_supported_ptypes_get,
1999 .mtu_set = nicvf_dev_set_mtu,
2000 .vlan_offload_set = nicvf_vlan_offload_set,
2001 .reta_update = nicvf_dev_reta_update,
2002 .reta_query = nicvf_dev_reta_query,
2003 .rss_hash_update = nicvf_dev_rss_hash_update,
2004 .rss_hash_conf_get = nicvf_dev_rss_hash_conf_get,
2005 .rx_queue_start = nicvf_dev_rx_queue_start,
2006 .rx_queue_stop = nicvf_dev_rx_queue_stop,
2007 .tx_queue_start = nicvf_dev_tx_queue_start,
2008 .tx_queue_stop = nicvf_dev_tx_queue_stop,
2009 .rx_queue_setup = nicvf_dev_rx_queue_setup,
2010 .rx_queue_release = nicvf_dev_rx_queue_release,
2011 .rx_queue_count = nicvf_dev_rx_queue_count,
2012 .tx_queue_setup = nicvf_dev_tx_queue_setup,
2013 .tx_queue_release = nicvf_dev_tx_queue_release,
2014 .get_reg = nicvf_dev_get_regs,
2018 nicvf_vlan_offload_config(struct rte_eth_dev *dev, int mask)
2020 struct rte_eth_rxmode *rxmode;
2021 struct nicvf *nic = nicvf_pmd_priv(dev);
2022 rxmode = &dev->data->dev_conf.rxmode;
2023 if (mask & ETH_VLAN_STRIP_MASK) {
2024 if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2025 nicvf_vlan_hw_strip(nic, true);
2027 nicvf_vlan_hw_strip(nic, false);
2034 nicvf_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2036 nicvf_vlan_offload_config(dev, mask);
2042 nicvf_set_first_skip(struct rte_eth_dev *dev)
2044 int bytes_to_skip = 0;
2047 struct rte_kvargs *kvlist;
2048 static const char *const skip[] = {
2051 struct nicvf *nic = nicvf_pmd_priv(dev);
2053 if (!dev->device->devargs) {
2054 nicvf_first_skip_config(nic, 0);
2058 kvlist = rte_kvargs_parse(dev->device->devargs->args, skip);
2062 if (kvlist->count == 0)
2065 for (i = 0; i != kvlist->count; ++i) {
2066 const struct rte_kvargs_pair *pair = &kvlist->pairs[i];
2068 if (!strcmp(pair->key, SKIP_DATA_BYTES))
2069 bytes_to_skip = atoi(pair->value);
2072 /*128 bytes amounts to one cache line*/
2073 if (bytes_to_skip >= 0 && bytes_to_skip < 128) {
2074 if (!(bytes_to_skip % 8)) {
2075 nicvf_first_skip_config(nic, (bytes_to_skip / 8));
2076 nic->skip_bytes = bytes_to_skip;
2079 PMD_INIT_LOG(ERR, "skip_data_bytes should be multiple of 8");
2084 PMD_INIT_LOG(ERR, "skip_data_bytes should be less than 128");
2089 nicvf_first_skip_config(nic, 0);
2091 rte_kvargs_free(kvlist);
2095 nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
2098 struct rte_pci_device *pci_dev;
2099 struct nicvf *nic = nicvf_pmd_priv(eth_dev);
2101 PMD_INIT_FUNC_TRACE();
2103 eth_dev->dev_ops = &nicvf_eth_dev_ops;
2105 /* For secondary processes, the primary has done all the work */
2106 if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2108 /* Setup callbacks for secondary process */
2109 nicvf_set_tx_function(eth_dev);
2110 nicvf_set_rx_function(eth_dev);
2113 /* If nic == NULL than it is secondary function
2114 * so ethdev need to be released by caller */
2119 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
2120 rte_eth_copy_pci_info(eth_dev, pci_dev);
2122 nic->device_id = pci_dev->id.device_id;
2123 nic->vendor_id = pci_dev->id.vendor_id;
2124 nic->subsystem_device_id = pci_dev->id.subsystem_device_id;
2125 nic->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
2127 PMD_INIT_LOG(DEBUG, "nicvf: device (%x:%x) %u:%u:%u:%u",
2128 pci_dev->id.vendor_id, pci_dev->id.device_id,
2129 pci_dev->addr.domain, pci_dev->addr.bus,
2130 pci_dev->addr.devid, pci_dev->addr.function);
2132 nic->reg_base = (uintptr_t)pci_dev->mem_resource[0].addr;
2133 if (!nic->reg_base) {
2134 PMD_INIT_LOG(ERR, "Failed to map BAR0");
2139 nicvf_disable_all_interrupts(nic);
2141 ret = nicvf_periodic_alarm_start(nicvf_interrupt, eth_dev);
2143 PMD_INIT_LOG(ERR, "Failed to start period alarm");
2147 ret = nicvf_mbox_check_pf_ready(nic);
2149 PMD_INIT_LOG(ERR, "Failed to get ready message from PF");
2153 "node=%d vf=%d mode=%s sqs=%s loopback_supported=%s",
2154 nic->node, nic->vf_id,
2155 nic->tns_mode == NIC_TNS_MODE ? "tns" : "tns-bypass",
2156 nic->sqs_mode ? "true" : "false",
2157 nic->loopback_supported ? "true" : "false"
2161 ret = nicvf_base_init(nic);
2163 PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
2167 if (nic->sqs_mode) {
2168 /* Push nic to stack of secondary vfs */
2169 nicvf_svf_push(nic);
2171 /* Steal nic pointer from the device for further reuse */
2172 eth_dev->data->dev_private = NULL;
2174 nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
2175 ret = nicvf_periodic_alarm_start(nicvf_vf_interrupt, nic);
2177 PMD_INIT_LOG(ERR, "Failed to start period alarm");
2181 /* Detach port by returning positive error number */
2185 eth_dev->data->mac_addrs = rte_zmalloc("mac_addr", ETHER_ADDR_LEN, 0);
2186 if (eth_dev->data->mac_addrs == NULL) {
2187 PMD_INIT_LOG(ERR, "Failed to allocate memory for mac addr");
2191 if (is_zero_ether_addr((struct ether_addr *)nic->mac_addr))
2192 eth_random_addr(&nic->mac_addr[0]);
2194 ether_addr_copy((struct ether_addr *)nic->mac_addr,
2195 ð_dev->data->mac_addrs[0]);
2197 ret = nicvf_mbox_set_mac_addr(nic, nic->mac_addr);
2199 PMD_INIT_LOG(ERR, "Failed to set mac addr");
2203 ret = nicvf_set_first_skip(eth_dev);
2205 PMD_INIT_LOG(ERR, "Failed to configure first skip");
2208 PMD_INIT_LOG(INFO, "Port %d (%x:%x) mac=%02x:%02x:%02x:%02x:%02x:%02x",
2209 eth_dev->data->port_id, nic->vendor_id, nic->device_id,
2210 nic->mac_addr[0], nic->mac_addr[1], nic->mac_addr[2],
2211 nic->mac_addr[3], nic->mac_addr[4], nic->mac_addr[5]);
2216 rte_free(eth_dev->data->mac_addrs);
2218 nicvf_periodic_alarm_stop(nicvf_interrupt, eth_dev);
2223 static const struct rte_pci_id pci_id_nicvf_map[] = {
2225 .class_id = RTE_CLASS_ANY_ID,
2226 .vendor_id = PCI_VENDOR_ID_CAVIUM,
2227 .device_id = PCI_DEVICE_ID_THUNDERX_CN88XX_PASS1_NICVF,
2228 .subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2229 .subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS1_NICVF,
2232 .class_id = RTE_CLASS_ANY_ID,
2233 .vendor_id = PCI_VENDOR_ID_CAVIUM,
2234 .device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2235 .subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2236 .subsystem_device_id = PCI_SUB_DEVICE_ID_CN88XX_PASS2_NICVF,
2239 .class_id = RTE_CLASS_ANY_ID,
2240 .vendor_id = PCI_VENDOR_ID_CAVIUM,
2241 .device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2242 .subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2243 .subsystem_device_id = PCI_SUB_DEVICE_ID_CN81XX_NICVF,
2246 .class_id = RTE_CLASS_ANY_ID,
2247 .vendor_id = PCI_VENDOR_ID_CAVIUM,
2248 .device_id = PCI_DEVICE_ID_THUNDERX_NICVF,
2249 .subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
2250 .subsystem_device_id = PCI_SUB_DEVICE_ID_CN83XX_NICVF,
2257 static int nicvf_eth_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2258 struct rte_pci_device *pci_dev)
2260 return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct nicvf),
2261 nicvf_eth_dev_init);
2264 static int nicvf_eth_pci_remove(struct rte_pci_device *pci_dev)
2266 return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
2269 static struct rte_pci_driver rte_nicvf_pmd = {
2270 .id_table = pci_id_nicvf_map,
2271 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_KEEP_MAPPED_RES |
2272 RTE_PCI_DRV_INTR_LSC,
2273 .probe = nicvf_eth_pci_probe,
2274 .remove = nicvf_eth_pci_remove,
2277 RTE_PMD_REGISTER_PCI(net_thunderx, rte_nicvf_pmd);
2278 RTE_PMD_REGISTER_PCI_TABLE(net_thunderx, pci_id_nicvf_map);
2279 RTE_PMD_REGISTER_KMOD_DEP(net_thunderx, "* igb_uio | uio_pci_generic | vfio-pci");
2280 RTE_PMD_REGISTER_PARAM_STRING(net_thunderx, SKIP_DATA_BYTES "=<int>");