4 * Copyright (C) Cavium networks Ltd. 2016.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
10 * * Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * * Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in
14 * the documentation and/or other materials provided with the
16 * * Neither the name of Cavium networks nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
23 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
24 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
25 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
26 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
30 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
42 #include <netinet/in.h>
43 #include <sys/queue.h>
44 #include <sys/timerfd.h>
46 #include <rte_alarm.h>
47 #include <rte_atomic.h>
48 #include <rte_branch_prediction.h>
49 #include <rte_byteorder.h>
50 #include <rte_common.h>
51 #include <rte_cycles.h>
52 #include <rte_debug.h>
55 #include <rte_ether.h>
56 #include <rte_ethdev.h>
57 #include <rte_interrupts.h>
59 #include <rte_memory.h>
60 #include <rte_memzone.h>
61 #include <rte_malloc.h>
62 #include <rte_random.h>
64 #include <rte_tailq.h>
66 #include "base/nicvf_plat.h"
68 #include "nicvf_ethdev.h"
70 #include "nicvf_logs.h"
73 nicvf_atomic_write_link_status(struct rte_eth_dev *dev,
74 struct rte_eth_link *link)
76 struct rte_eth_link *dst = &dev->data->dev_link;
77 struct rte_eth_link *src = link;
79 if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
80 *(uint64_t *)src) == 0)
87 nicvf_set_eth_link_status(struct nicvf *nic, struct rte_eth_link *link)
89 link->link_status = nic->link_up;
90 link->link_duplex = ETH_LINK_AUTONEG;
91 if (nic->duplex == NICVF_HALF_DUPLEX)
92 link->link_duplex = ETH_LINK_HALF_DUPLEX;
93 else if (nic->duplex == NICVF_FULL_DUPLEX)
94 link->link_duplex = ETH_LINK_FULL_DUPLEX;
95 link->link_speed = nic->speed;
96 link->link_autoneg = ETH_LINK_SPEED_AUTONEG;
100 nicvf_interrupt(void *arg)
102 struct nicvf *nic = arg;
104 if (nicvf_reg_poll_interrupts(nic) == NIC_MBOX_MSG_BGX_LINK_CHANGE) {
105 if (nic->eth_dev->data->dev_conf.intr_conf.lsc)
106 nicvf_set_eth_link_status(nic,
107 &nic->eth_dev->data->dev_link);
108 _rte_eth_dev_callback_process(nic->eth_dev,
109 RTE_ETH_EVENT_INTR_LSC);
112 rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
113 nicvf_interrupt, nic);
117 nicvf_periodic_alarm_start(struct nicvf *nic)
119 return rte_eal_alarm_set(NICVF_INTR_POLL_INTERVAL_MS * 1000,
120 nicvf_interrupt, nic);
124 nicvf_periodic_alarm_stop(struct nicvf *nic)
126 return rte_eal_alarm_cancel(nicvf_interrupt, nic);
130 * Return 0 means link status changed, -1 means not changed
133 nicvf_dev_link_update(struct rte_eth_dev *dev,
134 int wait_to_complete __rte_unused)
136 struct rte_eth_link link;
137 struct nicvf *nic = nicvf_pmd_priv(dev);
139 PMD_INIT_FUNC_TRACE();
141 memset(&link, 0, sizeof(link));
142 nicvf_set_eth_link_status(nic, &link);
143 return nicvf_atomic_write_link_status(dev, &link);
147 nicvf_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
149 struct nicvf *nic = nicvf_pmd_priv(dev);
150 uint32_t buffsz, frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
152 PMD_INIT_FUNC_TRACE();
154 if (frame_size > NIC_HW_MAX_FRS)
157 if (frame_size < NIC_HW_MIN_FRS)
160 buffsz = dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM;
163 * Refuse mtu that requires the support of scattered packets
164 * when this feature has not been enabled before.
166 if (!dev->data->scattered_rx &&
167 (frame_size + 2 * VLAN_TAG_SIZE > buffsz))
170 /* check <seg size> * <max_seg> >= max_frame */
171 if (dev->data->scattered_rx &&
172 (frame_size + 2 * VLAN_TAG_SIZE > buffsz * NIC_HW_MAX_SEGS))
175 if (frame_size > ETHER_MAX_LEN)
176 dev->data->dev_conf.rxmode.jumbo_frame = 1;
178 dev->data->dev_conf.rxmode.jumbo_frame = 0;
180 if (nicvf_mbox_update_hw_max_frs(nic, frame_size))
183 /* Update max frame size */
184 dev->data->dev_conf.rxmode.max_rx_pkt_len = (uint32_t)frame_size;
190 nicvf_dev_get_reg_length(struct rte_eth_dev *dev __rte_unused)
192 return nicvf_reg_get_count();
196 nicvf_dev_get_regs(struct rte_eth_dev *dev, struct rte_dev_reg_info *regs)
198 uint64_t *data = regs->data;
199 struct nicvf *nic = nicvf_pmd_priv(dev);
204 /* Support only full register dump */
205 if ((regs->length == 0) ||
206 (regs->length == (uint32_t)nicvf_reg_get_count())) {
207 regs->version = nic->vendor_id << 16 | nic->device_id;
208 nicvf_reg_dump(nic, data);
215 nicvf_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
218 struct nicvf_hw_rx_qstats rx_qstats;
219 struct nicvf_hw_tx_qstats tx_qstats;
220 struct nicvf_hw_stats port_stats;
221 struct nicvf *nic = nicvf_pmd_priv(dev);
223 /* Reading per RX ring stats */
224 for (qidx = 0; qidx < dev->data->nb_rx_queues; qidx++) {
225 if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
228 nicvf_hw_get_rx_qstats(nic, &rx_qstats, qidx);
229 stats->q_ibytes[qidx] = rx_qstats.q_rx_bytes;
230 stats->q_ipackets[qidx] = rx_qstats.q_rx_packets;
233 /* Reading per TX ring stats */
234 for (qidx = 0; qidx < dev->data->nb_tx_queues; qidx++) {
235 if (qidx == RTE_ETHDEV_QUEUE_STAT_CNTRS)
238 nicvf_hw_get_tx_qstats(nic, &tx_qstats, qidx);
239 stats->q_obytes[qidx] = tx_qstats.q_tx_bytes;
240 stats->q_opackets[qidx] = tx_qstats.q_tx_packets;
243 nicvf_hw_get_stats(nic, &port_stats);
244 stats->ibytes = port_stats.rx_bytes;
245 stats->ipackets = port_stats.rx_ucast_frames;
246 stats->ipackets += port_stats.rx_bcast_frames;
247 stats->ipackets += port_stats.rx_mcast_frames;
248 stats->ierrors = port_stats.rx_l2_errors;
249 stats->imissed = port_stats.rx_drop_red;
250 stats->imissed += port_stats.rx_drop_overrun;
251 stats->imissed += port_stats.rx_drop_bcast;
252 stats->imissed += port_stats.rx_drop_mcast;
253 stats->imissed += port_stats.rx_drop_l3_bcast;
254 stats->imissed += port_stats.rx_drop_l3_mcast;
256 stats->obytes = port_stats.tx_bytes_ok;
257 stats->opackets = port_stats.tx_ucast_frames_ok;
258 stats->opackets += port_stats.tx_bcast_frames_ok;
259 stats->opackets += port_stats.tx_mcast_frames_ok;
260 stats->oerrors = port_stats.tx_drops;
264 nicvf_dev_stats_reset(struct rte_eth_dev *dev)
267 uint16_t rxqs = 0, txqs = 0;
268 struct nicvf *nic = nicvf_pmd_priv(dev);
270 for (i = 0; i < dev->data->nb_rx_queues; i++)
271 rxqs |= (0x3 << (i * 2));
272 for (i = 0; i < dev->data->nb_tx_queues; i++)
273 txqs |= (0x3 << (i * 2));
275 nicvf_mbox_reset_stat_counters(nic, 0x3FFF, 0x1F, rxqs, txqs);
278 /* Promiscuous mode enabled by default in LMAC to VF 1:1 map configuration */
280 nicvf_dev_promisc_enable(struct rte_eth_dev *dev __rte_unused)
284 static inline uint64_t
285 nicvf_rss_ethdev_to_nic(struct nicvf *nic, uint64_t ethdev_rss)
287 uint64_t nic_rss = 0;
289 if (ethdev_rss & ETH_RSS_IPV4)
290 nic_rss |= RSS_IP_ENA;
292 if (ethdev_rss & ETH_RSS_IPV6)
293 nic_rss |= RSS_IP_ENA;
295 if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_UDP)
296 nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
298 if (ethdev_rss & ETH_RSS_NONFRAG_IPV4_TCP)
299 nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
301 if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_UDP)
302 nic_rss |= (RSS_IP_ENA | RSS_UDP_ENA);
304 if (ethdev_rss & ETH_RSS_NONFRAG_IPV6_TCP)
305 nic_rss |= (RSS_IP_ENA | RSS_TCP_ENA);
307 if (ethdev_rss & ETH_RSS_PORT)
308 nic_rss |= RSS_L2_EXTENDED_HASH_ENA;
310 if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
311 if (ethdev_rss & ETH_RSS_VXLAN)
312 nic_rss |= RSS_TUN_VXLAN_ENA;
314 if (ethdev_rss & ETH_RSS_GENEVE)
315 nic_rss |= RSS_TUN_GENEVE_ENA;
317 if (ethdev_rss & ETH_RSS_NVGRE)
318 nic_rss |= RSS_TUN_NVGRE_ENA;
324 static inline uint64_t
325 nicvf_rss_nic_to_ethdev(struct nicvf *nic, uint64_t nic_rss)
327 uint64_t ethdev_rss = 0;
329 if (nic_rss & RSS_IP_ENA)
330 ethdev_rss |= (ETH_RSS_IPV4 | ETH_RSS_IPV6);
332 if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_TCP_ENA))
333 ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_TCP |
334 ETH_RSS_NONFRAG_IPV6_TCP);
336 if ((nic_rss & RSS_IP_ENA) && (nic_rss & RSS_UDP_ENA))
337 ethdev_rss |= (ETH_RSS_NONFRAG_IPV4_UDP |
338 ETH_RSS_NONFRAG_IPV6_UDP);
340 if (nic_rss & RSS_L2_EXTENDED_HASH_ENA)
341 ethdev_rss |= ETH_RSS_PORT;
343 if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING) {
344 if (nic_rss & RSS_TUN_VXLAN_ENA)
345 ethdev_rss |= ETH_RSS_VXLAN;
347 if (nic_rss & RSS_TUN_GENEVE_ENA)
348 ethdev_rss |= ETH_RSS_GENEVE;
350 if (nic_rss & RSS_TUN_NVGRE_ENA)
351 ethdev_rss |= ETH_RSS_NVGRE;
357 nicvf_dev_reta_query(struct rte_eth_dev *dev,
358 struct rte_eth_rss_reta_entry64 *reta_conf,
361 struct nicvf *nic = nicvf_pmd_priv(dev);
362 uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
365 if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
366 RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
367 "(%d) doesn't match the number hardware can supported "
368 "(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
372 ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
376 /* Copy RETA table */
377 for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
378 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
379 if ((reta_conf[i].mask >> j) & 0x01)
380 reta_conf[i].reta[j] = tbl[j];
387 nicvf_dev_reta_update(struct rte_eth_dev *dev,
388 struct rte_eth_rss_reta_entry64 *reta_conf,
391 struct nicvf *nic = nicvf_pmd_priv(dev);
392 uint8_t tbl[NIC_MAX_RSS_IDR_TBL_SIZE];
395 if (reta_size != NIC_MAX_RSS_IDR_TBL_SIZE) {
396 RTE_LOG(ERR, PMD, "The size of hash lookup table configured "
397 "(%d) doesn't match the number hardware can supported "
398 "(%d)", reta_size, NIC_MAX_RSS_IDR_TBL_SIZE);
402 ret = nicvf_rss_reta_query(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
406 /* Copy RETA table */
407 for (i = 0; i < (NIC_MAX_RSS_IDR_TBL_SIZE / RTE_RETA_GROUP_SIZE); i++) {
408 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
409 if ((reta_conf[i].mask >> j) & 0x01)
410 tbl[j] = reta_conf[i].reta[j];
413 return nicvf_rss_reta_update(nic, tbl, NIC_MAX_RSS_IDR_TBL_SIZE);
417 nicvf_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
418 struct rte_eth_rss_conf *rss_conf)
420 struct nicvf *nic = nicvf_pmd_priv(dev);
422 if (rss_conf->rss_key)
423 nicvf_rss_get_key(nic, rss_conf->rss_key);
425 rss_conf->rss_key_len = RSS_HASH_KEY_BYTE_SIZE;
426 rss_conf->rss_hf = nicvf_rss_nic_to_ethdev(nic, nicvf_rss_get_cfg(nic));
431 nicvf_dev_rss_hash_update(struct rte_eth_dev *dev,
432 struct rte_eth_rss_conf *rss_conf)
434 struct nicvf *nic = nicvf_pmd_priv(dev);
437 if (rss_conf->rss_key &&
438 rss_conf->rss_key_len != RSS_HASH_KEY_BYTE_SIZE) {
439 RTE_LOG(ERR, PMD, "Hash key size mismatch %d",
440 rss_conf->rss_key_len);
444 if (rss_conf->rss_key)
445 nicvf_rss_set_key(nic, rss_conf->rss_key);
447 nic_rss = nicvf_rss_ethdev_to_nic(nic, rss_conf->rss_hf);
448 nicvf_rss_set_cfg(nic, nic_rss);
453 nicvf_qset_cq_alloc(struct nicvf *nic, struct nicvf_rxq *rxq, uint16_t qidx,
456 const struct rte_memzone *rz;
457 uint32_t ring_size = desc_cnt * sizeof(union cq_entry_t);
459 rz = rte_eth_dma_zone_reserve(nic->eth_dev, "cq_ring", qidx, ring_size,
460 NICVF_CQ_BASE_ALIGN_BYTES, nic->node);
462 PMD_INIT_LOG(ERR, "Failed to allocate mem for cq hw ring");
466 memset(rz->addr, 0, ring_size);
468 rxq->phys = rz->phys_addr;
469 rxq->desc = rz->addr;
470 rxq->qlen_mask = desc_cnt - 1;
476 nicvf_qset_sq_alloc(struct nicvf *nic, struct nicvf_txq *sq, uint16_t qidx,
479 const struct rte_memzone *rz;
480 uint32_t ring_size = desc_cnt * sizeof(union sq_entry_t);
482 rz = rte_eth_dma_zone_reserve(nic->eth_dev, "sq", qidx, ring_size,
483 NICVF_SQ_BASE_ALIGN_BYTES, nic->node);
485 PMD_INIT_LOG(ERR, "Failed allocate mem for sq hw ring");
489 memset(rz->addr, 0, ring_size);
491 sq->phys = rz->phys_addr;
493 sq->qlen_mask = desc_cnt - 1;
499 nicvf_tx_queue_release_mbufs(struct nicvf_txq *txq)
504 while (head != txq->tail) {
505 if (txq->txbuffs[head]) {
506 rte_pktmbuf_free_seg(txq->txbuffs[head]);
507 txq->txbuffs[head] = NULL;
510 head = head & txq->qlen_mask;
515 nicvf_tx_queue_reset(struct nicvf_txq *txq)
517 uint32_t txq_desc_cnt = txq->qlen_mask + 1;
519 memset(txq->desc, 0, sizeof(union sq_entry_t) * txq_desc_cnt);
520 memset(txq->txbuffs, 0, sizeof(struct rte_mbuf *) * txq_desc_cnt);
527 nicvf_dev_tx_queue_release(void *sq)
529 struct nicvf_txq *txq;
531 PMD_INIT_FUNC_TRACE();
533 txq = (struct nicvf_txq *)sq;
535 if (txq->txbuffs != NULL) {
536 nicvf_tx_queue_release_mbufs(txq);
537 rte_free(txq->txbuffs);
545 nicvf_dev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
546 uint16_t nb_desc, unsigned int socket_id,
547 const struct rte_eth_txconf *tx_conf)
549 uint16_t tx_free_thresh;
550 uint8_t is_single_pool;
551 struct nicvf_txq *txq;
552 struct nicvf *nic = nicvf_pmd_priv(dev);
554 PMD_INIT_FUNC_TRACE();
556 /* Socket id check */
557 if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
558 PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
559 socket_id, nic->node);
561 /* Tx deferred start is not supported */
562 if (tx_conf->tx_deferred_start) {
563 PMD_INIT_LOG(ERR, "Tx deferred start not supported");
567 /* Roundup nb_desc to available qsize and validate max number of desc */
568 nb_desc = nicvf_qsize_sq_roundup(nb_desc);
570 PMD_INIT_LOG(ERR, "Value of nb_desc beyond available sq qsize");
574 /* Validate tx_free_thresh */
575 tx_free_thresh = (uint16_t)((tx_conf->tx_free_thresh) ?
576 tx_conf->tx_free_thresh :
577 NICVF_DEFAULT_TX_FREE_THRESH);
579 if (tx_free_thresh > (nb_desc) ||
580 tx_free_thresh > NICVF_MAX_TX_FREE_THRESH) {
582 "tx_free_thresh must be less than the number of TX "
583 "descriptors. (tx_free_thresh=%u port=%d "
584 "queue=%d)", (unsigned int)tx_free_thresh,
585 (int)dev->data->port_id, (int)qidx);
589 /* Free memory prior to re-allocation if needed. */
590 if (dev->data->tx_queues[qidx] != NULL) {
591 PMD_TX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
593 nicvf_dev_tx_queue_release(dev->data->tx_queues[qidx]);
594 dev->data->tx_queues[qidx] = NULL;
597 /* Allocating tx queue data structure */
598 txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct nicvf_txq),
599 RTE_CACHE_LINE_SIZE, nic->node);
601 PMD_INIT_LOG(ERR, "Failed to allocate txq=%d", qidx);
606 txq->queue_id = qidx;
607 txq->tx_free_thresh = tx_free_thresh;
608 txq->txq_flags = tx_conf->txq_flags;
609 txq->sq_head = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_HEAD;
610 txq->sq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_SQ_0_7_DOOR;
611 is_single_pool = (txq->txq_flags & ETH_TXQ_FLAGS_NOREFCOUNT &&
612 txq->txq_flags & ETH_TXQ_FLAGS_NOMULTMEMP);
614 /* Choose optimum free threshold value for multipool case */
615 if (!is_single_pool) {
616 txq->tx_free_thresh = (uint16_t)
617 (tx_conf->tx_free_thresh == NICVF_DEFAULT_TX_FREE_THRESH ?
618 NICVF_TX_FREE_MPOOL_THRESH :
619 tx_conf->tx_free_thresh);
622 /* Allocate software ring */
623 txq->txbuffs = rte_zmalloc_socket("txq->txbuffs",
624 nb_desc * sizeof(struct rte_mbuf *),
625 RTE_CACHE_LINE_SIZE, nic->node);
627 if (txq->txbuffs == NULL) {
628 nicvf_dev_tx_queue_release(txq);
632 if (nicvf_qset_sq_alloc(nic, txq, qidx, nb_desc)) {
633 PMD_INIT_LOG(ERR, "Failed to allocate mem for sq %d", qidx);
634 nicvf_dev_tx_queue_release(txq);
638 nicvf_tx_queue_reset(txq);
640 PMD_TX_LOG(DEBUG, "[%d] txq=%p nb_desc=%d desc=%p phys=0x%" PRIx64,
641 qidx, txq, nb_desc, txq->desc, txq->phys);
643 dev->data->tx_queues[qidx] = txq;
644 dev->data->tx_queue_state[qidx] = RTE_ETH_QUEUE_STATE_STOPPED;
649 nicvf_rx_queue_reset(struct nicvf_rxq *rxq)
652 rxq->available_space = 0;
653 rxq->recv_buffers = 0;
657 nicvf_dev_rx_queue_release(void *rx_queue)
659 struct nicvf_rxq *rxq = rx_queue;
661 PMD_INIT_FUNC_TRACE();
668 nicvf_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t qidx,
669 uint16_t nb_desc, unsigned int socket_id,
670 const struct rte_eth_rxconf *rx_conf,
671 struct rte_mempool *mp)
673 uint16_t rx_free_thresh;
674 struct nicvf_rxq *rxq;
675 struct nicvf *nic = nicvf_pmd_priv(dev);
677 PMD_INIT_FUNC_TRACE();
679 /* Socket id check */
680 if (socket_id != (unsigned int)SOCKET_ID_ANY && socket_id != nic->node)
681 PMD_DRV_LOG(WARNING, "socket_id expected %d, configured %d",
682 socket_id, nic->node);
684 /* Mempool memory should be contiguous */
685 if (mp->nb_mem_chunks != 1) {
686 PMD_INIT_LOG(ERR, "Non contiguous mempool, check huge page sz");
690 /* Rx deferred start is not supported */
691 if (rx_conf->rx_deferred_start) {
692 PMD_INIT_LOG(ERR, "Rx deferred start not supported");
696 /* Roundup nb_desc to available qsize and validate max number of desc */
697 nb_desc = nicvf_qsize_cq_roundup(nb_desc);
699 PMD_INIT_LOG(ERR, "Value nb_desc beyond available hw cq qsize");
703 /* Check rx_free_thresh upper bound */
704 rx_free_thresh = (uint16_t)((rx_conf->rx_free_thresh) ?
705 rx_conf->rx_free_thresh :
706 NICVF_DEFAULT_RX_FREE_THRESH);
707 if (rx_free_thresh > NICVF_MAX_RX_FREE_THRESH ||
708 rx_free_thresh >= nb_desc * .75) {
709 PMD_INIT_LOG(ERR, "rx_free_thresh greater than expected %d",
714 /* Free memory prior to re-allocation if needed */
715 if (dev->data->rx_queues[qidx] != NULL) {
716 PMD_RX_LOG(DEBUG, "Freeing memory prior to re-allocation %d",
718 nicvf_dev_rx_queue_release(dev->data->rx_queues[qidx]);
719 dev->data->rx_queues[qidx] = NULL;
722 /* Allocate rxq memory */
723 rxq = rte_zmalloc_socket("ethdev rx queue", sizeof(struct nicvf_rxq),
724 RTE_CACHE_LINE_SIZE, nic->node);
726 PMD_INIT_LOG(ERR, "Failed to allocate rxq=%d", qidx);
732 rxq->queue_id = qidx;
733 rxq->port_id = dev->data->port_id;
734 rxq->rx_free_thresh = rx_free_thresh;
735 rxq->rx_drop_en = rx_conf->rx_drop_en;
736 rxq->cq_status = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_STATUS;
737 rxq->cq_door = nicvf_qset_base(nic, qidx) + NIC_QSET_CQ_0_7_DOOR;
738 rxq->precharge_cnt = 0;
739 rxq->rbptr_offset = NICVF_CQE_RBPTR_WORD;
741 /* Alloc completion queue */
742 if (nicvf_qset_cq_alloc(nic, rxq, rxq->queue_id, nb_desc)) {
743 PMD_INIT_LOG(ERR, "failed to allocate cq %u", rxq->queue_id);
744 nicvf_dev_rx_queue_release(rxq);
748 nicvf_rx_queue_reset(rxq);
750 PMD_RX_LOG(DEBUG, "[%d] rxq=%p pool=%s nb_desc=(%d/%d) phy=%" PRIx64,
751 qidx, rxq, mp->name, nb_desc,
752 rte_mempool_count(mp), rxq->phys);
754 dev->data->rx_queues[qidx] = rxq;
755 dev->data->rx_queue_state[qidx] = RTE_ETH_QUEUE_STATE_STOPPED;
760 nicvf_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
762 struct nicvf *nic = nicvf_pmd_priv(dev);
764 PMD_INIT_FUNC_TRACE();
766 dev_info->min_rx_bufsize = ETHER_MIN_MTU;
767 dev_info->max_rx_pktlen = NIC_HW_MAX_FRS;
768 dev_info->max_rx_queues = (uint16_t)MAX_RCV_QUEUES_PER_QS;
769 dev_info->max_tx_queues = (uint16_t)MAX_SND_QUEUES_PER_QS;
770 dev_info->max_mac_addrs = 1;
771 dev_info->max_vfs = dev->pci_dev->max_vfs;
773 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
774 dev_info->tx_offload_capa =
775 DEV_TX_OFFLOAD_IPV4_CKSUM |
776 DEV_TX_OFFLOAD_UDP_CKSUM |
777 DEV_TX_OFFLOAD_TCP_CKSUM |
778 DEV_TX_OFFLOAD_TCP_TSO |
779 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
781 dev_info->reta_size = nic->rss_info.rss_size;
782 dev_info->hash_key_size = RSS_HASH_KEY_BYTE_SIZE;
783 dev_info->flow_type_rss_offloads = NICVF_RSS_OFFLOAD_PASS1;
784 if (nicvf_hw_cap(nic) & NICVF_CAP_TUNNEL_PARSING)
785 dev_info->flow_type_rss_offloads |= NICVF_RSS_OFFLOAD_TUNNEL;
787 dev_info->default_rxconf = (struct rte_eth_rxconf) {
788 .rx_free_thresh = NICVF_DEFAULT_RX_FREE_THRESH,
792 dev_info->default_txconf = (struct rte_eth_txconf) {
793 .tx_free_thresh = NICVF_DEFAULT_TX_FREE_THRESH,
795 ETH_TXQ_FLAGS_NOMULTSEGS |
796 ETH_TXQ_FLAGS_NOREFCOUNT |
797 ETH_TXQ_FLAGS_NOMULTMEMP |
798 ETH_TXQ_FLAGS_NOVLANOFFL |
799 ETH_TXQ_FLAGS_NOXSUMSCTP,
804 nicvf_dev_configure(struct rte_eth_dev *dev)
806 struct rte_eth_conf *conf = &dev->data->dev_conf;
807 struct rte_eth_rxmode *rxmode = &conf->rxmode;
808 struct rte_eth_txmode *txmode = &conf->txmode;
809 struct nicvf *nic = nicvf_pmd_priv(dev);
811 PMD_INIT_FUNC_TRACE();
813 if (!rte_eal_has_hugepages()) {
814 PMD_INIT_LOG(INFO, "Huge page is not configured");
818 if (txmode->mq_mode) {
819 PMD_INIT_LOG(INFO, "Tx mq_mode DCB or VMDq not supported");
823 if (rxmode->mq_mode != ETH_MQ_RX_NONE &&
824 rxmode->mq_mode != ETH_MQ_RX_RSS) {
825 PMD_INIT_LOG(INFO, "Unsupported rx qmode %d", rxmode->mq_mode);
829 if (!rxmode->hw_strip_crc) {
830 PMD_INIT_LOG(NOTICE, "Can't disable hw crc strip");
831 rxmode->hw_strip_crc = 1;
834 if (rxmode->hw_ip_checksum) {
835 PMD_INIT_LOG(NOTICE, "Rxcksum not supported");
836 rxmode->hw_ip_checksum = 0;
839 if (rxmode->split_hdr_size) {
840 PMD_INIT_LOG(INFO, "Rxmode does not support split header");
844 if (rxmode->hw_vlan_filter) {
845 PMD_INIT_LOG(INFO, "VLAN filter not supported");
849 if (rxmode->hw_vlan_extend) {
850 PMD_INIT_LOG(INFO, "VLAN extended not supported");
854 if (rxmode->enable_lro) {
855 PMD_INIT_LOG(INFO, "LRO not supported");
859 if (conf->link_speeds & ETH_LINK_SPEED_FIXED) {
860 PMD_INIT_LOG(INFO, "Setting link speed/duplex not supported");
864 if (conf->dcb_capability_en) {
865 PMD_INIT_LOG(INFO, "DCB enable not supported");
869 if (conf->fdir_conf.mode != RTE_FDIR_MODE_NONE) {
870 PMD_INIT_LOG(INFO, "Flow director not supported");
874 PMD_INIT_LOG(DEBUG, "Configured ethdev port%d hwcap=0x%" PRIx64,
875 dev->data->port_id, nicvf_hw_cap(nic));
880 /* Initialize and register driver with DPDK Application */
881 static const struct eth_dev_ops nicvf_eth_dev_ops = {
882 .dev_configure = nicvf_dev_configure,
883 .link_update = nicvf_dev_link_update,
884 .stats_get = nicvf_dev_stats_get,
885 .stats_reset = nicvf_dev_stats_reset,
886 .promiscuous_enable = nicvf_dev_promisc_enable,
887 .dev_infos_get = nicvf_dev_info_get,
888 .mtu_set = nicvf_dev_set_mtu,
889 .reta_update = nicvf_dev_reta_update,
890 .reta_query = nicvf_dev_reta_query,
891 .rss_hash_update = nicvf_dev_rss_hash_update,
892 .rss_hash_conf_get = nicvf_dev_rss_hash_conf_get,
893 .rx_queue_setup = nicvf_dev_rx_queue_setup,
894 .rx_queue_release = nicvf_dev_rx_queue_release,
895 .tx_queue_setup = nicvf_dev_tx_queue_setup,
896 .tx_queue_release = nicvf_dev_tx_queue_release,
897 .get_reg_length = nicvf_dev_get_reg_length,
898 .get_reg = nicvf_dev_get_regs,
902 nicvf_eth_dev_init(struct rte_eth_dev *eth_dev)
905 struct rte_pci_device *pci_dev;
906 struct nicvf *nic = nicvf_pmd_priv(eth_dev);
908 PMD_INIT_FUNC_TRACE();
910 eth_dev->dev_ops = &nicvf_eth_dev_ops;
912 pci_dev = eth_dev->pci_dev;
913 rte_eth_copy_pci_info(eth_dev, pci_dev);
915 nic->device_id = pci_dev->id.device_id;
916 nic->vendor_id = pci_dev->id.vendor_id;
917 nic->subsystem_device_id = pci_dev->id.subsystem_device_id;
918 nic->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
919 nic->eth_dev = eth_dev;
921 PMD_INIT_LOG(DEBUG, "nicvf: device (%x:%x) %u:%u:%u:%u",
922 pci_dev->id.vendor_id, pci_dev->id.device_id,
923 pci_dev->addr.domain, pci_dev->addr.bus,
924 pci_dev->addr.devid, pci_dev->addr.function);
926 nic->reg_base = (uintptr_t)pci_dev->mem_resource[0].addr;
927 if (!nic->reg_base) {
928 PMD_INIT_LOG(ERR, "Failed to map BAR0");
933 nicvf_disable_all_interrupts(nic);
935 ret = nicvf_periodic_alarm_start(nic);
937 PMD_INIT_LOG(ERR, "Failed to start period alarm");
941 ret = nicvf_mbox_check_pf_ready(nic);
943 PMD_INIT_LOG(ERR, "Failed to get ready message from PF");
947 "node=%d vf=%d mode=%s sqs=%s loopback_supported=%s",
948 nic->node, nic->vf_id,
949 nic->tns_mode == NIC_TNS_MODE ? "tns" : "tns-bypass",
950 nic->sqs_mode ? "true" : "false",
951 nic->loopback_supported ? "true" : "false"
956 PMD_INIT_LOG(INFO, "Unsupported SQS VF detected, Detaching...");
957 /* Detach port by returning Positive error number */
962 eth_dev->data->mac_addrs = rte_zmalloc("mac_addr", ETHER_ADDR_LEN, 0);
963 if (eth_dev->data->mac_addrs == NULL) {
964 PMD_INIT_LOG(ERR, "Failed to allocate memory for mac addr");
968 if (is_zero_ether_addr((struct ether_addr *)nic->mac_addr))
969 eth_random_addr(&nic->mac_addr[0]);
971 ether_addr_copy((struct ether_addr *)nic->mac_addr,
972 ð_dev->data->mac_addrs[0]);
974 ret = nicvf_mbox_set_mac_addr(nic, nic->mac_addr);
976 PMD_INIT_LOG(ERR, "Failed to set mac addr");
980 ret = nicvf_base_init(nic);
982 PMD_INIT_LOG(ERR, "Failed to execute nicvf_base_init");
986 ret = nicvf_mbox_get_rss_size(nic);
988 PMD_INIT_LOG(ERR, "Failed to get rss table size");
992 PMD_INIT_LOG(INFO, "Port %d (%x:%x) mac=%02x:%02x:%02x:%02x:%02x:%02x",
993 eth_dev->data->port_id, nic->vendor_id, nic->device_id,
994 nic->mac_addr[0], nic->mac_addr[1], nic->mac_addr[2],
995 nic->mac_addr[3], nic->mac_addr[4], nic->mac_addr[5]);
1000 rte_free(eth_dev->data->mac_addrs);
1002 nicvf_periodic_alarm_stop(nic);
1007 static const struct rte_pci_id pci_id_nicvf_map[] = {
1009 .class_id = RTE_CLASS_ANY_ID,
1010 .vendor_id = PCI_VENDOR_ID_CAVIUM,
1011 .device_id = PCI_DEVICE_ID_THUNDERX_PASS1_NICVF,
1012 .subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
1013 .subsystem_device_id = PCI_SUB_DEVICE_ID_THUNDERX_PASS1_NICVF,
1016 .class_id = RTE_CLASS_ANY_ID,
1017 .vendor_id = PCI_VENDOR_ID_CAVIUM,
1018 .device_id = PCI_DEVICE_ID_THUNDERX_PASS2_NICVF,
1019 .subsystem_vendor_id = PCI_VENDOR_ID_CAVIUM,
1020 .subsystem_device_id = PCI_SUB_DEVICE_ID_THUNDERX_PASS2_NICVF,
1027 static struct eth_driver rte_nicvf_pmd = {
1029 .name = "rte_nicvf_pmd",
1030 .id_table = pci_id_nicvf_map,
1031 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1033 .eth_dev_init = nicvf_eth_dev_init,
1034 .dev_private_size = sizeof(struct nicvf),
1038 rte_nicvf_pmd_init(const char *name __rte_unused, const char *para __rte_unused)
1040 PMD_INIT_FUNC_TRACE();
1041 PMD_INIT_LOG(INFO, "librte_pmd_thunderx nicvf version %s",
1042 THUNDERX_NICVF_PMD_VERSION);
1044 rte_eth_driver_register(&rte_nicvf_pmd);
1048 static struct rte_driver rte_nicvf_driver = {
1049 .name = "nicvf_driver",
1051 .init = rte_nicvf_pmd_init,
1054 PMD_REGISTER_DRIVER(rte_nicvf_driver);