4 * Copyright(c) 2017 Marvell International Ltd.
5 * Copyright(c) 2017 Semihalf.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * * Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * * Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
18 * * Neither the name of Semihalf nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <rte_ethdev.h>
36 #include <rte_kvargs.h>
38 #include <rte_malloc.h>
39 #include <rte_bus_vdev.h>
41 /* Unluckily, container_of is defined by both DPDK and MUSDK,
42 * we'll declare only one version.
44 * Note that it is not used in this PMD anyway.
51 #include <linux/ethtool.h>
52 #include <linux/sockios.h>
54 #include <net/if_arp.h>
55 #include <sys/ioctl.h>
56 #include <sys/socket.h>
58 #include <sys/types.h>
60 #include "mrvl_ethdev.h"
63 /* bitmask with reserved hifs */
64 #define MRVL_MUSDK_HIFS_RESERVED 0x0F
65 /* bitmask with reserved bpools */
66 #define MRVL_MUSDK_BPOOLS_RESERVED 0x07
67 /* bitmask with reserved kernel RSS tables */
68 #define MRVL_MUSDK_RSS_RESERVED 0x01
69 /* maximum number of available hifs */
70 #define MRVL_MUSDK_HIFS_MAX 9
73 #define MRVL_MUSDK_PREFETCH_SHIFT 2
75 /* TCAM has 25 entries reserved for uc/mc filter entries */
76 #define MRVL_MAC_ADDRS_MAX 25
77 #define MRVL_MATCH_LEN 16
78 #define MRVL_PKT_EFFEC_OFFS (MRVL_PKT_OFFS + MV_MH_SIZE)
79 /* Maximum allowable packet size */
80 #define MRVL_PKT_SIZE_MAX (10240 - MV_MH_SIZE)
82 #define MRVL_IFACE_NAME_ARG "iface"
83 #define MRVL_CFG_ARG "cfg"
85 #define MRVL_BURST_SIZE 64
87 #define MRVL_ARP_LENGTH 28
89 #define MRVL_COOKIE_ADDR_INVALID ~0ULL
91 #define MRVL_COOKIE_HIGH_ADDR_SHIFT (sizeof(pp2_cookie_t) * 8)
92 #define MRVL_COOKIE_HIGH_ADDR_MASK (~0ULL << MRVL_COOKIE_HIGH_ADDR_SHIFT)
94 /* Memory size (in bytes) for MUSDK dma buffers */
95 #define MRVL_MUSDK_DMA_MEMSIZE 41943040
97 static const char * const valid_args[] = {
103 static int used_hifs = MRVL_MUSDK_HIFS_RESERVED;
104 static struct pp2_hif *hifs[RTE_MAX_LCORE];
105 static int used_bpools[PP2_NUM_PKT_PROC] = {
106 MRVL_MUSDK_BPOOLS_RESERVED,
107 MRVL_MUSDK_BPOOLS_RESERVED
110 struct pp2_bpool *mrvl_port_to_bpool_lookup[RTE_MAX_ETHPORTS];
111 int mrvl_port_bpool_size[PP2_NUM_PKT_PROC][PP2_BPOOL_NUM_POOLS][RTE_MAX_LCORE];
112 uint64_t cookie_addr_high = MRVL_COOKIE_ADDR_INVALID;
115 * To use buffer harvesting based on loopback port shadow queue structure
116 * was introduced for buffers information bookkeeping.
118 * Before sending the packet, related buffer information (pp2_buff_inf) is
119 * stored in shadow queue. After packet is transmitted no longer used
120 * packet buffer is released back to it's original hardware pool,
121 * on condition it originated from interface.
122 * In case it was generated by application itself i.e: mbuf->port field is
123 * 0xff then its released to software mempool.
125 struct mrvl_shadow_txq {
126 int head; /* write index - used when sending buffers */
127 int tail; /* read index - used when releasing buffers */
128 u16 size; /* queue occupied size */
129 u16 num_to_release; /* number of buffers sent, that can be released */
130 struct buff_release_entry ent[MRVL_PP2_TX_SHADOWQ_SIZE]; /* q entries */
134 struct mrvl_priv *priv;
135 struct rte_mempool *mp;
144 struct mrvl_priv *priv;
151 * Every tx queue should have dedicated shadow tx queue.
153 * Ports assigned by DPDK might not start at zero or be continuous so
154 * as a workaround define shadow queues for each possible port so that
155 * we eventually fit somewhere.
157 struct mrvl_shadow_txq shadow_txqs[RTE_MAX_ETHPORTS][RTE_MAX_LCORE];
159 /** Number of ports configured. */
161 static int mrvl_lcore_first;
162 static int mrvl_lcore_last;
165 mrvl_get_bpool_size(int pp2_id, int pool_id)
170 for (i = mrvl_lcore_first; i <= mrvl_lcore_last; i++)
171 size += mrvl_port_bpool_size[pp2_id][pool_id][i];
177 mrvl_reserve_bit(int *bitmap, int max)
179 int n = sizeof(*bitmap) * 8 - __builtin_clz(*bitmap);
190 * Configure rss based on dpdk rss configuration.
193 * Pointer to private structure.
195 * Pointer to RSS configuration.
198 * 0 on success, negative error value otherwise.
201 mrvl_configure_rss(struct mrvl_priv *priv, struct rte_eth_rss_conf *rss_conf)
203 if (rss_conf->rss_key)
204 RTE_LOG(WARNING, PMD, "Changing hash key is not supported\n");
206 if (rss_conf->rss_hf == 0) {
207 priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
208 } else if (rss_conf->rss_hf & ETH_RSS_IPV4) {
209 priv->ppio_params.inqs_params.hash_type =
210 PP2_PPIO_HASH_T_2_TUPLE;
211 } else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP) {
212 priv->ppio_params.inqs_params.hash_type =
213 PP2_PPIO_HASH_T_5_TUPLE;
214 priv->rss_hf_tcp = 1;
215 } else if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_UDP) {
216 priv->ppio_params.inqs_params.hash_type =
217 PP2_PPIO_HASH_T_5_TUPLE;
218 priv->rss_hf_tcp = 0;
227 * Ethernet device configuration.
229 * Prepare the driver for a given number of TX and RX queues and
233 * Pointer to Ethernet device structure.
236 * 0 on success, negative error value otherwise.
239 mrvl_dev_configure(struct rte_eth_dev *dev)
241 struct mrvl_priv *priv = dev->data->dev_private;
244 if (dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_NONE &&
245 dev->data->dev_conf.rxmode.mq_mode != ETH_MQ_RX_RSS) {
246 RTE_LOG(INFO, PMD, "Unsupported rx multi queue mode %d\n",
247 dev->data->dev_conf.rxmode.mq_mode);
251 if (!dev->data->dev_conf.rxmode.hw_strip_crc) {
253 "L2 CRC stripping is always enabled in hw\n");
254 dev->data->dev_conf.rxmode.hw_strip_crc = 1;
257 if (dev->data->dev_conf.rxmode.hw_vlan_strip) {
258 RTE_LOG(INFO, PMD, "VLAN stripping not supported\n");
262 if (dev->data->dev_conf.rxmode.split_hdr_size) {
263 RTE_LOG(INFO, PMD, "Split headers not supported\n");
267 if (dev->data->dev_conf.rxmode.enable_scatter) {
268 RTE_LOG(INFO, PMD, "RX Scatter/Gather not supported\n");
272 if (dev->data->dev_conf.rxmode.enable_lro) {
273 RTE_LOG(INFO, PMD, "LRO not supported\n");
277 if (dev->data->dev_conf.rxmode.jumbo_frame)
278 dev->data->mtu = dev->data->dev_conf.rxmode.max_rx_pkt_len -
279 ETHER_HDR_LEN - ETHER_CRC_LEN;
281 ret = mrvl_configure_rxqs(priv, dev->data->port_id,
282 dev->data->nb_rx_queues);
286 priv->ppio_params.outqs_params.num_outqs = dev->data->nb_tx_queues;
287 priv->ppio_params.maintain_stats = 1;
288 priv->nb_rx_queues = dev->data->nb_rx_queues;
290 if (dev->data->nb_rx_queues == 1 &&
291 dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
292 RTE_LOG(WARNING, PMD, "Disabling hash for 1 rx queue\n");
293 priv->ppio_params.inqs_params.hash_type = PP2_PPIO_HASH_T_NONE;
298 return mrvl_configure_rss(priv,
299 &dev->data->dev_conf.rx_adv_conf.rss_conf);
303 * DPDK callback to change the MTU.
305 * Setting the MTU affects hardware MRU (packets larger than the MRU
309 * Pointer to Ethernet device structure.
314 * 0 on success, negative error value otherwise.
317 mrvl_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
319 struct mrvl_priv *priv = dev->data->dev_private;
320 /* extra MV_MH_SIZE bytes are required for Marvell tag */
321 uint16_t mru = mtu + MV_MH_SIZE + ETHER_HDR_LEN + ETHER_CRC_LEN;
324 if (mtu < ETHER_MIN_MTU || mru > MRVL_PKT_SIZE_MAX)
330 ret = pp2_ppio_set_mru(priv->ppio, mru);
334 return pp2_ppio_set_mtu(priv->ppio, mtu);
338 * DPDK callback to bring the link up.
341 * Pointer to Ethernet device structure.
344 * 0 on success, negative error value otherwise.
347 mrvl_dev_set_link_up(struct rte_eth_dev *dev)
349 struct mrvl_priv *priv = dev->data->dev_private;
355 ret = pp2_ppio_enable(priv->ppio);
360 * mtu/mru can be updated if pp2_ppio_enable() was called at least once
361 * as pp2_ppio_enable() changes port->t_mode from default 0 to
362 * PP2_TRAFFIC_INGRESS_EGRESS.
364 * Set mtu to default DPDK value here.
366 ret = mrvl_mtu_set(dev, dev->data->mtu);
368 pp2_ppio_disable(priv->ppio);
374 * DPDK callback to bring the link down.
377 * Pointer to Ethernet device structure.
380 * 0 on success, negative error value otherwise.
383 mrvl_dev_set_link_down(struct rte_eth_dev *dev)
385 struct mrvl_priv *priv = dev->data->dev_private;
390 return pp2_ppio_disable(priv->ppio);
394 * DPDK callback to start the device.
397 * Pointer to Ethernet device structure.
400 * 0 on success, negative errno value on failure.
403 mrvl_dev_start(struct rte_eth_dev *dev)
405 struct mrvl_priv *priv = dev->data->dev_private;
406 char match[MRVL_MATCH_LEN];
409 snprintf(match, sizeof(match), "ppio-%d:%d",
410 priv->pp_id, priv->ppio_id);
411 priv->ppio_params.match = match;
414 * Calculate the maximum bpool size for refill feature to 1.5 of the
415 * configured size. In case the bpool size will exceed this value,
416 * superfluous buffers will be removed
418 priv->bpool_max_size = priv->bpool_init_size +
419 (priv->bpool_init_size >> 1);
421 * Calculate the minimum bpool size for refill feature as follows:
422 * 2 default burst sizes multiply by number of rx queues.
423 * If the bpool size will be below this value, new buffers will
424 * be added to the pool.
426 priv->bpool_min_size = priv->nb_rx_queues * MRVL_BURST_SIZE * 2;
428 ret = pp2_ppio_init(&priv->ppio_params, &priv->ppio);
433 * In case there are some some stale uc/mc mac addresses flush them
434 * here. It cannot be done during mrvl_dev_close() as port information
435 * is already gone at that point (due to pp2_ppio_deinit() in
438 if (!priv->uc_mc_flushed) {
439 ret = pp2_ppio_flush_mac_addrs(priv->ppio, 1, 1);
442 "Failed to flush uc/mc filter list\n");
445 priv->uc_mc_flushed = 1;
448 if (!priv->vlan_flushed) {
449 ret = pp2_ppio_flush_vlan(priv->ppio);
451 RTE_LOG(ERR, PMD, "Failed to flush vlan list\n");
454 * once pp2_ppio_flush_vlan() is supported jump to out
458 priv->vlan_flushed = 1;
461 /* For default QoS config, don't start classifier. */
463 ret = mrvl_start_qos_mapping(priv);
465 pp2_ppio_deinit(priv->ppio);
470 ret = mrvl_dev_set_link_up(dev);
476 pp2_ppio_deinit(priv->ppio);
481 * Flush receive queues.
484 * Pointer to Ethernet device structure.
487 mrvl_flush_rx_queues(struct rte_eth_dev *dev)
491 RTE_LOG(INFO, PMD, "Flushing rx queues\n");
492 for (i = 0; i < dev->data->nb_rx_queues; i++) {
496 struct mrvl_rxq *q = dev->data->rx_queues[i];
497 struct pp2_ppio_desc descs[MRVL_PP2_RXD_MAX];
499 num = MRVL_PP2_RXD_MAX;
500 ret = pp2_ppio_recv(q->priv->ppio,
501 q->priv->rxq_map[q->queue_id].tc,
502 q->priv->rxq_map[q->queue_id].inq,
503 descs, (uint16_t *)&num);
504 } while (ret == 0 && num);
509 * Flush transmit shadow queues.
512 * Pointer to Ethernet device structure.
515 mrvl_flush_tx_shadow_queues(struct rte_eth_dev *dev)
519 RTE_LOG(INFO, PMD, "Flushing tx shadow queues\n");
520 for (i = 0; i < RTE_MAX_LCORE; i++) {
521 struct mrvl_shadow_txq *sq =
522 &shadow_txqs[dev->data->port_id][i];
524 while (sq->tail != sq->head) {
525 uint64_t addr = cookie_addr_high |
526 sq->ent[sq->tail].buff.cookie;
527 rte_pktmbuf_free((struct rte_mbuf *)addr);
528 sq->tail = (sq->tail + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
531 memset(sq, 0, sizeof(*sq));
536 * Flush hardware bpool (buffer-pool).
539 * Pointer to Ethernet device structure.
542 mrvl_flush_bpool(struct rte_eth_dev *dev)
544 struct mrvl_priv *priv = dev->data->dev_private;
548 ret = pp2_bpool_get_num_buffs(priv->bpool, &num);
550 RTE_LOG(ERR, PMD, "Failed to get bpool buffers number\n");
555 struct pp2_buff_inf inf;
558 ret = pp2_bpool_get_buff(hifs[rte_lcore_id()], priv->bpool,
563 addr = cookie_addr_high | inf.cookie;
564 rte_pktmbuf_free((struct rte_mbuf *)addr);
569 * DPDK callback to stop the device.
572 * Pointer to Ethernet device structure.
575 mrvl_dev_stop(struct rte_eth_dev *dev)
577 struct mrvl_priv *priv = dev->data->dev_private;
579 mrvl_dev_set_link_down(dev);
580 mrvl_flush_rx_queues(dev);
581 mrvl_flush_tx_shadow_queues(dev);
583 pp2_cls_qos_tbl_deinit(priv->qos_tbl);
584 pp2_ppio_deinit(priv->ppio);
589 * DPDK callback to close the device.
592 * Pointer to Ethernet device structure.
595 mrvl_dev_close(struct rte_eth_dev *dev)
597 struct mrvl_priv *priv = dev->data->dev_private;
600 for (i = 0; i < priv->ppio_params.inqs_params.num_tcs; ++i) {
601 struct pp2_ppio_tc_params *tc_params =
602 &priv->ppio_params.inqs_params.tcs_params[i];
604 if (tc_params->inqs_params) {
605 rte_free(tc_params->inqs_params);
606 tc_params->inqs_params = NULL;
610 mrvl_flush_bpool(dev);
614 * DPDK callback to retrieve physical link information.
617 * Pointer to Ethernet device structure.
618 * @param wait_to_complete
619 * Wait for request completion (ignored).
622 * 0 on success, negative error value otherwise.
625 mrvl_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
629 * once MUSDK provides necessary API use it here
631 struct mrvl_priv *priv = dev->data->dev_private;
632 struct ethtool_cmd edata;
634 int ret, fd, link_up;
639 edata.cmd = ETHTOOL_GSET;
641 strcpy(req.ifr_name, dev->data->name);
642 req.ifr_data = (void *)&edata;
644 fd = socket(AF_INET, SOCK_DGRAM, 0);
648 ret = ioctl(fd, SIOCETHTOOL, &req);
656 switch (ethtool_cmd_speed(&edata)) {
658 dev->data->dev_link.link_speed = ETH_SPEED_NUM_10M;
661 dev->data->dev_link.link_speed = ETH_SPEED_NUM_100M;
664 dev->data->dev_link.link_speed = ETH_SPEED_NUM_1G;
667 dev->data->dev_link.link_speed = ETH_SPEED_NUM_10G;
670 dev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
673 dev->data->dev_link.link_duplex = edata.duplex ? ETH_LINK_FULL_DUPLEX :
674 ETH_LINK_HALF_DUPLEX;
675 dev->data->dev_link.link_autoneg = edata.autoneg ? ETH_LINK_AUTONEG :
677 pp2_ppio_get_link_state(priv->ppio, &link_up);
678 dev->data->dev_link.link_status = link_up ? ETH_LINK_UP : ETH_LINK_DOWN;
684 * DPDK callback to enable promiscuous mode.
687 * Pointer to Ethernet device structure.
690 mrvl_promiscuous_enable(struct rte_eth_dev *dev)
692 struct mrvl_priv *priv = dev->data->dev_private;
698 ret = pp2_ppio_set_promisc(priv->ppio, 1);
700 RTE_LOG(ERR, PMD, "Failed to enable promiscuous mode\n");
704 * DPDK callback to enable allmulti mode.
707 * Pointer to Ethernet device structure.
710 mrvl_allmulticast_enable(struct rte_eth_dev *dev)
712 struct mrvl_priv *priv = dev->data->dev_private;
718 ret = pp2_ppio_set_mc_promisc(priv->ppio, 1);
720 RTE_LOG(ERR, PMD, "Failed enable all-multicast mode\n");
724 * DPDK callback to disable promiscuous mode.
727 * Pointer to Ethernet device structure.
730 mrvl_promiscuous_disable(struct rte_eth_dev *dev)
732 struct mrvl_priv *priv = dev->data->dev_private;
738 ret = pp2_ppio_set_promisc(priv->ppio, 0);
740 RTE_LOG(ERR, PMD, "Failed to disable promiscuous mode\n");
744 * DPDK callback to disable allmulticast mode.
747 * Pointer to Ethernet device structure.
750 mrvl_allmulticast_disable(struct rte_eth_dev *dev)
752 struct mrvl_priv *priv = dev->data->dev_private;
758 ret = pp2_ppio_set_mc_promisc(priv->ppio, 0);
760 RTE_LOG(ERR, PMD, "Failed to disable all-multicast mode\n");
764 * DPDK callback to remove a MAC address.
767 * Pointer to Ethernet device structure.
772 mrvl_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
774 struct mrvl_priv *priv = dev->data->dev_private;
775 char buf[ETHER_ADDR_FMT_SIZE];
781 ret = pp2_ppio_remove_mac_addr(priv->ppio,
782 dev->data->mac_addrs[index].addr_bytes);
784 ether_format_addr(buf, sizeof(buf),
785 &dev->data->mac_addrs[index]);
786 RTE_LOG(ERR, PMD, "Failed to remove mac %s\n", buf);
791 * DPDK callback to add a MAC address.
794 * Pointer to Ethernet device structure.
796 * MAC address to register.
800 * VMDq pool index to associate address with (unused).
803 * 0 on success, negative error value otherwise.
806 mrvl_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
807 uint32_t index, uint32_t vmdq __rte_unused)
809 struct mrvl_priv *priv = dev->data->dev_private;
810 char buf[ETHER_ADDR_FMT_SIZE];
814 /* For setting index 0, mrvl_mac_addr_set() should be used.*/
821 * Maximum number of uc addresses can be tuned via kernel module mvpp2x
822 * parameter uc_filter_max. Maximum number of mc addresses is then
823 * MRVL_MAC_ADDRS_MAX - uc_filter_max. Currently it defaults to 4 and
826 * If more than uc_filter_max uc addresses were added to filter list
827 * then NIC will switch to promiscuous mode automatically.
829 * If more than MRVL_MAC_ADDRS_MAX - uc_filter_max number mc addresses
830 * were added to filter list then NIC will switch to all-multicast mode
833 ret = pp2_ppio_add_mac_addr(priv->ppio, mac_addr->addr_bytes);
835 ether_format_addr(buf, sizeof(buf), mac_addr);
836 RTE_LOG(ERR, PMD, "Failed to add mac %s\n", buf);
844 * DPDK callback to set the primary MAC address.
847 * Pointer to Ethernet device structure.
849 * MAC address to register.
852 mrvl_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
854 struct mrvl_priv *priv = dev->data->dev_private;
860 ret = pp2_ppio_set_mac_addr(priv->ppio, mac_addr->addr_bytes);
862 char buf[ETHER_ADDR_FMT_SIZE];
863 ether_format_addr(buf, sizeof(buf), mac_addr);
864 RTE_LOG(ERR, PMD, "Failed to set mac to %s\n", buf);
869 * DPDK callback to get device statistics.
872 * Pointer to Ethernet device structure.
874 * Stats structure output buffer.
877 * 0 on success, negative error value otherwise.
880 mrvl_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
882 struct mrvl_priv *priv = dev->data->dev_private;
883 struct pp2_ppio_statistics ppio_stats;
884 uint64_t drop_mac = 0;
885 unsigned int i, idx, ret;
890 for (i = 0; i < dev->data->nb_rx_queues; i++) {
891 struct mrvl_rxq *rxq = dev->data->rx_queues[i];
892 struct pp2_ppio_inq_statistics rx_stats;
898 if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
900 "rx queue %d stats out of range (0 - %d)\n",
901 idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
905 ret = pp2_ppio_inq_get_statistics(priv->ppio,
906 priv->rxq_map[idx].tc,
907 priv->rxq_map[idx].inq,
911 "Failed to update rx queue %d stats\n", idx);
915 stats->q_ibytes[idx] = rxq->bytes_recv;
916 stats->q_ipackets[idx] = rx_stats.enq_desc - rxq->drop_mac;
917 stats->q_errors[idx] = rx_stats.drop_early +
918 rx_stats.drop_fullq +
921 stats->ibytes += rxq->bytes_recv;
922 drop_mac += rxq->drop_mac;
925 for (i = 0; i < dev->data->nb_tx_queues; i++) {
926 struct mrvl_txq *txq = dev->data->tx_queues[i];
927 struct pp2_ppio_outq_statistics tx_stats;
933 if (unlikely(idx >= RTE_ETHDEV_QUEUE_STAT_CNTRS)) {
935 "tx queue %d stats out of range (0 - %d)\n",
936 idx, RTE_ETHDEV_QUEUE_STAT_CNTRS - 1);
939 ret = pp2_ppio_outq_get_statistics(priv->ppio, idx,
943 "Failed to update tx queue %d stats\n", idx);
947 stats->q_opackets[idx] = tx_stats.deq_desc;
948 stats->q_obytes[idx] = txq->bytes_sent;
949 stats->obytes += txq->bytes_sent;
952 ret = pp2_ppio_get_statistics(priv->ppio, &ppio_stats, 0);
954 RTE_LOG(ERR, PMD, "Failed to update port statistics\n");
958 stats->ipackets += ppio_stats.rx_packets - drop_mac;
959 stats->opackets += ppio_stats.tx_packets;
960 stats->imissed += ppio_stats.rx_fullq_dropped +
961 ppio_stats.rx_bm_dropped +
962 ppio_stats.rx_early_dropped +
963 ppio_stats.rx_fifo_dropped +
964 ppio_stats.rx_cls_dropped;
965 stats->ierrors = drop_mac;
971 * DPDK callback to clear device statistics.
974 * Pointer to Ethernet device structure.
977 mrvl_stats_reset(struct rte_eth_dev *dev)
979 struct mrvl_priv *priv = dev->data->dev_private;
985 for (i = 0; i < dev->data->nb_rx_queues; i++) {
986 struct mrvl_rxq *rxq = dev->data->rx_queues[i];
988 pp2_ppio_inq_get_statistics(priv->ppio, priv->rxq_map[i].tc,
989 priv->rxq_map[i].inq, NULL, 1);
994 for (i = 0; i < dev->data->nb_tx_queues; i++) {
995 struct mrvl_txq *txq = dev->data->tx_queues[i];
997 pp2_ppio_outq_get_statistics(priv->ppio, i, NULL, 1);
1001 pp2_ppio_get_statistics(priv->ppio, NULL, 1);
1005 * DPDK callback to get information about the device.
1008 * Pointer to Ethernet device structure (unused).
1010 * Info structure output buffer.
1013 mrvl_dev_infos_get(struct rte_eth_dev *dev __rte_unused,
1014 struct rte_eth_dev_info *info)
1016 info->speed_capa = ETH_LINK_SPEED_10M |
1017 ETH_LINK_SPEED_100M |
1021 info->max_rx_queues = MRVL_PP2_RXQ_MAX;
1022 info->max_tx_queues = MRVL_PP2_TXQ_MAX;
1023 info->max_mac_addrs = MRVL_MAC_ADDRS_MAX;
1025 info->rx_desc_lim.nb_max = MRVL_PP2_RXD_MAX;
1026 info->rx_desc_lim.nb_min = MRVL_PP2_RXD_MIN;
1027 info->rx_desc_lim.nb_align = MRVL_PP2_RXD_ALIGN;
1029 info->tx_desc_lim.nb_max = MRVL_PP2_TXD_MAX;
1030 info->tx_desc_lim.nb_min = MRVL_PP2_TXD_MIN;
1031 info->tx_desc_lim.nb_align = MRVL_PP2_TXD_ALIGN;
1033 info->rx_offload_capa = DEV_RX_OFFLOAD_JUMBO_FRAME |
1034 DEV_RX_OFFLOAD_VLAN_FILTER |
1035 DEV_RX_OFFLOAD_IPV4_CKSUM |
1036 DEV_RX_OFFLOAD_UDP_CKSUM |
1037 DEV_RX_OFFLOAD_TCP_CKSUM;
1039 info->tx_offload_capa = DEV_TX_OFFLOAD_IPV4_CKSUM |
1040 DEV_TX_OFFLOAD_UDP_CKSUM |
1041 DEV_TX_OFFLOAD_TCP_CKSUM;
1043 info->flow_type_rss_offloads = ETH_RSS_IPV4 |
1044 ETH_RSS_NONFRAG_IPV4_TCP |
1045 ETH_RSS_NONFRAG_IPV4_UDP;
1047 /* By default packets are dropped if no descriptors are available */
1048 info->default_rxconf.rx_drop_en = 1;
1050 info->max_rx_pktlen = MRVL_PKT_SIZE_MAX;
1054 * Return supported packet types.
1057 * Pointer to Ethernet device structure (unused).
1060 * Const pointer to the table with supported packet types.
1062 static const uint32_t *
1063 mrvl_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1065 static const uint32_t ptypes[] = {
1068 RTE_PTYPE_L3_IPV4_EXT,
1069 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
1071 RTE_PTYPE_L3_IPV6_EXT,
1072 RTE_PTYPE_L2_ETHER_ARP,
1081 * DPDK callback to get information about specific receive queue.
1084 * Pointer to Ethernet device structure.
1085 * @param rx_queue_id
1086 * Receive queue index.
1088 * Receive queue information structure.
1090 static void mrvl_rxq_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1091 struct rte_eth_rxq_info *qinfo)
1093 struct mrvl_rxq *q = dev->data->rx_queues[rx_queue_id];
1094 struct mrvl_priv *priv = dev->data->dev_private;
1095 int inq = priv->rxq_map[rx_queue_id].inq;
1096 int tc = priv->rxq_map[rx_queue_id].tc;
1097 struct pp2_ppio_tc_params *tc_params =
1098 &priv->ppio_params.inqs_params.tcs_params[tc];
1101 qinfo->nb_desc = tc_params->inqs_params[inq].size;
1105 * DPDK callback to get information about specific transmit queue.
1108 * Pointer to Ethernet device structure.
1109 * @param tx_queue_id
1110 * Transmit queue index.
1112 * Transmit queue information structure.
1114 static void mrvl_txq_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1115 struct rte_eth_txq_info *qinfo)
1117 struct mrvl_priv *priv = dev->data->dev_private;
1120 priv->ppio_params.outqs_params.outqs_params[tx_queue_id].size;
1124 * DPDK callback to Configure a VLAN filter.
1127 * Pointer to Ethernet device structure.
1129 * VLAN ID to filter.
1134 * 0 on success, negative error value otherwise.
1137 mrvl_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1139 struct mrvl_priv *priv = dev->data->dev_private;
1144 return on ? pp2_ppio_add_vlan(priv->ppio, vlan_id) :
1145 pp2_ppio_remove_vlan(priv->ppio, vlan_id);
1149 * Release buffers to hardware bpool (buffer-pool)
1152 * Receive queue pointer.
1154 * Number of buffers to release to bpool.
1157 * 0 on success, negative error value otherwise.
1160 mrvl_fill_bpool(struct mrvl_rxq *rxq, int num)
1162 struct buff_release_entry entries[MRVL_PP2_TXD_MAX];
1163 struct rte_mbuf *mbufs[MRVL_PP2_TXD_MAX];
1165 unsigned int core_id = rte_lcore_id();
1166 struct pp2_hif *hif = hifs[core_id];
1167 struct pp2_bpool *bpool = rxq->priv->bpool;
1169 ret = rte_pktmbuf_alloc_bulk(rxq->mp, mbufs, num);
1173 if (cookie_addr_high == MRVL_COOKIE_ADDR_INVALID)
1175 (uint64_t)mbufs[0] & MRVL_COOKIE_HIGH_ADDR_MASK;
1177 for (i = 0; i < num; i++) {
1178 if (((uint64_t)mbufs[i] & MRVL_COOKIE_HIGH_ADDR_MASK)
1179 != cookie_addr_high) {
1181 "mbuf virtual addr high 0x%lx out of range\n",
1182 (uint64_t)mbufs[i] >> 32);
1186 entries[i].buff.addr =
1187 rte_mbuf_data_iova_default(mbufs[i]);
1188 entries[i].buff.cookie = (pp2_cookie_t)(uint64_t)mbufs[i];
1189 entries[i].bpool = bpool;
1192 pp2_bpool_put_buffs(hif, entries, (uint16_t *)&i);
1193 mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] += i;
1200 for (; i < num; i++)
1201 rte_pktmbuf_free(mbufs[i]);
1207 * DPDK callback to configure the receive queue.
1210 * Pointer to Ethernet device structure.
1214 * Number of descriptors to configure in queue.
1216 * NUMA socket on which memory must be allocated.
1218 * Thresholds parameters (unused_).
1220 * Memory pool for buffer allocations.
1223 * 0 on success, negative error value otherwise.
1226 mrvl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1227 unsigned int socket,
1228 const struct rte_eth_rxconf *conf __rte_unused,
1229 struct rte_mempool *mp)
1231 struct mrvl_priv *priv = dev->data->dev_private;
1232 struct mrvl_rxq *rxq;
1234 max_rx_pkt_len = dev->data->dev_conf.rxmode.max_rx_pkt_len;
1237 if (priv->rxq_map[idx].tc == MRVL_UNKNOWN_TC) {
1239 * Unknown TC mapping, mapping will not have a correct queue.
1241 RTE_LOG(ERR, PMD, "Unknown TC mapping for queue %hu eth%hhu\n",
1242 idx, priv->ppio_id);
1246 min_size = rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM -
1247 MRVL_PKT_EFFEC_OFFS;
1248 if (min_size < max_rx_pkt_len) {
1250 "Mbuf size must be increased to %u bytes to hold up to %u bytes of data.\n",
1251 max_rx_pkt_len + RTE_PKTMBUF_HEADROOM +
1252 MRVL_PKT_EFFEC_OFFS,
1257 if (dev->data->rx_queues[idx]) {
1258 rte_free(dev->data->rx_queues[idx]);
1259 dev->data->rx_queues[idx] = NULL;
1262 rxq = rte_zmalloc_socket("rxq", sizeof(*rxq), 0, socket);
1268 rxq->cksum_enabled = dev->data->dev_conf.rxmode.hw_ip_checksum;
1269 rxq->queue_id = idx;
1270 rxq->port_id = dev->data->port_id;
1271 mrvl_port_to_bpool_lookup[rxq->port_id] = priv->bpool;
1273 tc = priv->rxq_map[rxq->queue_id].tc,
1274 inq = priv->rxq_map[rxq->queue_id].inq;
1275 priv->ppio_params.inqs_params.tcs_params[tc].inqs_params[inq].size =
1278 ret = mrvl_fill_bpool(rxq, desc);
1284 priv->bpool_init_size += desc;
1286 dev->data->rx_queues[idx] = rxq;
1292 * DPDK callback to release the receive queue.
1295 * Generic receive queue pointer.
1298 mrvl_rx_queue_release(void *rxq)
1300 struct mrvl_rxq *q = rxq;
1301 struct pp2_ppio_tc_params *tc_params;
1302 int i, num, tc, inq;
1307 tc = q->priv->rxq_map[q->queue_id].tc;
1308 inq = q->priv->rxq_map[q->queue_id].inq;
1309 tc_params = &q->priv->ppio_params.inqs_params.tcs_params[tc];
1310 num = tc_params->inqs_params[inq].size;
1311 for (i = 0; i < num; i++) {
1312 struct pp2_buff_inf inf;
1315 pp2_bpool_get_buff(hifs[rte_lcore_id()], q->priv->bpool, &inf);
1316 addr = cookie_addr_high | inf.cookie;
1317 rte_pktmbuf_free((struct rte_mbuf *)addr);
1324 * DPDK callback to configure the transmit queue.
1327 * Pointer to Ethernet device structure.
1329 * Transmit queue index.
1331 * Number of descriptors to configure in the queue.
1333 * NUMA socket on which memory must be allocated.
1335 * Thresholds parameters (unused).
1338 * 0 on success, negative error value otherwise.
1341 mrvl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc,
1342 unsigned int socket,
1343 const struct rte_eth_txconf *conf __rte_unused)
1345 struct mrvl_priv *priv = dev->data->dev_private;
1346 struct mrvl_txq *txq;
1348 if (dev->data->tx_queues[idx]) {
1349 rte_free(dev->data->tx_queues[idx]);
1350 dev->data->tx_queues[idx] = NULL;
1353 txq = rte_zmalloc_socket("txq", sizeof(*txq), 0, socket);
1358 txq->queue_id = idx;
1359 txq->port_id = dev->data->port_id;
1360 dev->data->tx_queues[idx] = txq;
1362 priv->ppio_params.outqs_params.outqs_params[idx].size = desc;
1363 priv->ppio_params.outqs_params.outqs_params[idx].weight = 1;
1369 * DPDK callback to release the transmit queue.
1372 * Generic transmit queue pointer.
1375 mrvl_tx_queue_release(void *txq)
1377 struct mrvl_txq *q = txq;
1386 * Update RSS hash configuration
1389 * Pointer to Ethernet device structure.
1391 * Pointer to RSS configuration.
1394 * 0 on success, negative error value otherwise.
1397 mrvl_rss_hash_update(struct rte_eth_dev *dev,
1398 struct rte_eth_rss_conf *rss_conf)
1400 struct mrvl_priv *priv = dev->data->dev_private;
1402 return mrvl_configure_rss(priv, rss_conf);
1406 * DPDK callback to get RSS hash configuration.
1409 * Pointer to Ethernet device structure.
1411 * Pointer to RSS configuration.
1417 mrvl_rss_hash_conf_get(struct rte_eth_dev *dev,
1418 struct rte_eth_rss_conf *rss_conf)
1420 struct mrvl_priv *priv = dev->data->dev_private;
1421 enum pp2_ppio_hash_type hash_type =
1422 priv->ppio_params.inqs_params.hash_type;
1424 rss_conf->rss_key = NULL;
1426 if (hash_type == PP2_PPIO_HASH_T_NONE)
1427 rss_conf->rss_hf = 0;
1428 else if (hash_type == PP2_PPIO_HASH_T_2_TUPLE)
1429 rss_conf->rss_hf = ETH_RSS_IPV4;
1430 else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && priv->rss_hf_tcp)
1431 rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_TCP;
1432 else if (hash_type == PP2_PPIO_HASH_T_5_TUPLE && !priv->rss_hf_tcp)
1433 rss_conf->rss_hf = ETH_RSS_NONFRAG_IPV4_UDP;
1438 static const struct eth_dev_ops mrvl_ops = {
1439 .dev_configure = mrvl_dev_configure,
1440 .dev_start = mrvl_dev_start,
1441 .dev_stop = mrvl_dev_stop,
1442 .dev_set_link_up = mrvl_dev_set_link_up,
1443 .dev_set_link_down = mrvl_dev_set_link_down,
1444 .dev_close = mrvl_dev_close,
1445 .link_update = mrvl_link_update,
1446 .promiscuous_enable = mrvl_promiscuous_enable,
1447 .allmulticast_enable = mrvl_allmulticast_enable,
1448 .promiscuous_disable = mrvl_promiscuous_disable,
1449 .allmulticast_disable = mrvl_allmulticast_disable,
1450 .mac_addr_remove = mrvl_mac_addr_remove,
1451 .mac_addr_add = mrvl_mac_addr_add,
1452 .mac_addr_set = mrvl_mac_addr_set,
1453 .mtu_set = mrvl_mtu_set,
1454 .stats_get = mrvl_stats_get,
1455 .stats_reset = mrvl_stats_reset,
1456 .dev_infos_get = mrvl_dev_infos_get,
1457 .dev_supported_ptypes_get = mrvl_dev_supported_ptypes_get,
1458 .rxq_info_get = mrvl_rxq_info_get,
1459 .txq_info_get = mrvl_txq_info_get,
1460 .vlan_filter_set = mrvl_vlan_filter_set,
1461 .rx_queue_setup = mrvl_rx_queue_setup,
1462 .rx_queue_release = mrvl_rx_queue_release,
1463 .tx_queue_setup = mrvl_tx_queue_setup,
1464 .tx_queue_release = mrvl_tx_queue_release,
1465 .rss_hash_update = mrvl_rss_hash_update,
1466 .rss_hash_conf_get = mrvl_rss_hash_conf_get,
1470 * Return packet type information and l3/l4 offsets.
1473 * Pointer to the received packet descriptor.
1480 * Packet type information.
1482 static inline uint64_t
1483 mrvl_desc_to_packet_type_and_offset(struct pp2_ppio_desc *desc,
1484 uint8_t *l3_offset, uint8_t *l4_offset)
1486 enum pp2_inq_l3_type l3_type;
1487 enum pp2_inq_l4_type l4_type;
1488 uint64_t packet_type;
1490 pp2_ppio_inq_desc_get_l3_info(desc, &l3_type, l3_offset);
1491 pp2_ppio_inq_desc_get_l4_info(desc, &l4_type, l4_offset);
1493 packet_type = RTE_PTYPE_L2_ETHER;
1496 case PP2_INQ_L3_TYPE_IPV4_NO_OPTS:
1497 packet_type |= RTE_PTYPE_L3_IPV4;
1499 case PP2_INQ_L3_TYPE_IPV4_OK:
1500 packet_type |= RTE_PTYPE_L3_IPV4_EXT;
1502 case PP2_INQ_L3_TYPE_IPV4_TTL_ZERO:
1503 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
1505 case PP2_INQ_L3_TYPE_IPV6_NO_EXT:
1506 packet_type |= RTE_PTYPE_L3_IPV6;
1508 case PP2_INQ_L3_TYPE_IPV6_EXT:
1509 packet_type |= RTE_PTYPE_L3_IPV6_EXT;
1511 case PP2_INQ_L3_TYPE_ARP:
1512 packet_type |= RTE_PTYPE_L2_ETHER_ARP;
1514 * In case of ARP l4_offset is set to wrong value.
1515 * Set it to proper one so that later on mbuf->l3_len can be
1516 * calculated subtracting l4_offset and l3_offset.
1518 *l4_offset = *l3_offset + MRVL_ARP_LENGTH;
1521 RTE_LOG(DEBUG, PMD, "Failed to recognise l3 packet type\n");
1526 case PP2_INQ_L4_TYPE_TCP:
1527 packet_type |= RTE_PTYPE_L4_TCP;
1529 case PP2_INQ_L4_TYPE_UDP:
1530 packet_type |= RTE_PTYPE_L4_UDP;
1533 RTE_LOG(DEBUG, PMD, "Failed to recognise l4 packet type\n");
1541 * Get offload information from the received packet descriptor.
1544 * Pointer to the received packet descriptor.
1547 * Mbuf offload flags.
1549 static inline uint64_t
1550 mrvl_desc_to_ol_flags(struct pp2_ppio_desc *desc)
1553 enum pp2_inq_desc_status status;
1555 status = pp2_ppio_inq_desc_get_l3_pkt_error(desc);
1556 if (unlikely(status != PP2_DESC_ERR_OK))
1557 flags = PKT_RX_IP_CKSUM_BAD;
1559 flags = PKT_RX_IP_CKSUM_GOOD;
1561 status = pp2_ppio_inq_desc_get_l4_pkt_error(desc);
1562 if (unlikely(status != PP2_DESC_ERR_OK))
1563 flags |= PKT_RX_L4_CKSUM_BAD;
1565 flags |= PKT_RX_L4_CKSUM_GOOD;
1571 * DPDK callback for receive.
1574 * Generic pointer to the receive queue.
1576 * Array to store received packets.
1578 * Maximum number of packets in array.
1581 * Number of packets successfully received.
1584 mrvl_rx_pkt_burst(void *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
1586 struct mrvl_rxq *q = rxq;
1587 struct pp2_ppio_desc descs[nb_pkts];
1588 struct pp2_bpool *bpool;
1589 int i, ret, rx_done = 0;
1591 unsigned int core_id = rte_lcore_id();
1593 if (unlikely(!q->priv->ppio))
1596 bpool = q->priv->bpool;
1598 ret = pp2_ppio_recv(q->priv->ppio, q->priv->rxq_map[q->queue_id].tc,
1599 q->priv->rxq_map[q->queue_id].inq, descs, &nb_pkts);
1600 if (unlikely(ret < 0)) {
1601 RTE_LOG(ERR, PMD, "Failed to receive packets\n");
1604 mrvl_port_bpool_size[bpool->pp2_id][bpool->id][core_id] -= nb_pkts;
1606 for (i = 0; i < nb_pkts; i++) {
1607 struct rte_mbuf *mbuf;
1608 uint8_t l3_offset, l4_offset;
1609 enum pp2_inq_desc_status status;
1612 if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
1613 struct pp2_ppio_desc *pref_desc;
1616 pref_desc = &descs[i + MRVL_MUSDK_PREFETCH_SHIFT];
1617 pref_addr = cookie_addr_high |
1618 pp2_ppio_inq_desc_get_cookie(pref_desc);
1619 rte_mbuf_prefetch_part1((struct rte_mbuf *)(pref_addr));
1620 rte_mbuf_prefetch_part2((struct rte_mbuf *)(pref_addr));
1623 addr = cookie_addr_high |
1624 pp2_ppio_inq_desc_get_cookie(&descs[i]);
1625 mbuf = (struct rte_mbuf *)addr;
1626 rte_pktmbuf_reset(mbuf);
1628 /* drop packet in case of mac, overrun or resource error */
1629 status = pp2_ppio_inq_desc_get_l2_pkt_error(&descs[i]);
1630 if (unlikely(status != PP2_DESC_ERR_OK)) {
1631 struct pp2_buff_inf binf = {
1632 .addr = rte_mbuf_data_iova_default(mbuf),
1633 .cookie = (pp2_cookie_t)(uint64_t)mbuf,
1636 pp2_bpool_put_buff(hifs[core_id], bpool, &binf);
1637 mrvl_port_bpool_size
1638 [bpool->pp2_id][bpool->id][core_id]++;
1643 mbuf->data_off += MRVL_PKT_EFFEC_OFFS;
1644 mbuf->pkt_len = pp2_ppio_inq_desc_get_pkt_len(&descs[i]);
1645 mbuf->data_len = mbuf->pkt_len;
1646 mbuf->port = q->port_id;
1648 mrvl_desc_to_packet_type_and_offset(&descs[i],
1651 mbuf->l2_len = l3_offset;
1652 mbuf->l3_len = l4_offset - l3_offset;
1654 if (likely(q->cksum_enabled))
1655 mbuf->ol_flags = mrvl_desc_to_ol_flags(&descs[i]);
1657 rx_pkts[rx_done++] = mbuf;
1658 q->bytes_recv += mbuf->pkt_len;
1661 if (rte_spinlock_trylock(&q->priv->lock) == 1) {
1662 num = mrvl_get_bpool_size(bpool->pp2_id, bpool->id);
1664 if (unlikely(num <= q->priv->bpool_min_size ||
1665 (!rx_done && num < q->priv->bpool_init_size))) {
1666 ret = mrvl_fill_bpool(q, MRVL_BURST_SIZE);
1668 RTE_LOG(ERR, PMD, "Failed to fill bpool\n");
1669 } else if (unlikely(num > q->priv->bpool_max_size)) {
1671 int pkt_to_remove = num - q->priv->bpool_init_size;
1672 struct rte_mbuf *mbuf;
1673 struct pp2_buff_inf buff;
1676 "\nport-%d:%d: bpool %d oversize - remove %d buffers (pool size: %d -> %d)\n",
1677 bpool->pp2_id, q->priv->ppio->port_id,
1678 bpool->id, pkt_to_remove, num,
1679 q->priv->bpool_init_size);
1681 for (i = 0; i < pkt_to_remove; i++) {
1682 pp2_bpool_get_buff(hifs[core_id], bpool, &buff);
1683 mbuf = (struct rte_mbuf *)
1684 (cookie_addr_high | buff.cookie);
1685 rte_pktmbuf_free(mbuf);
1687 mrvl_port_bpool_size
1688 [bpool->pp2_id][bpool->id][core_id] -=
1691 rte_spinlock_unlock(&q->priv->lock);
1698 * Prepare offload information.
1702 * @param packet_type
1703 * Packet type bitfield.
1705 * Pointer to the pp2_ouq_l3_type structure.
1707 * Pointer to the pp2_outq_l4_type structure.
1708 * @param gen_l3_cksum
1709 * Will be set to 1 in case l3 checksum is computed.
1711 * Will be set to 1 in case l4 checksum is computed.
1714 * 0 on success, negative error value otherwise.
1717 mrvl_prepare_proto_info(uint64_t ol_flags, uint32_t packet_type,
1718 enum pp2_outq_l3_type *l3_type,
1719 enum pp2_outq_l4_type *l4_type,
1724 * Based on ol_flags prepare information
1725 * for pp2_ppio_outq_desc_set_proto_info() which setups descriptor
1728 if (ol_flags & PKT_TX_IPV4) {
1729 *l3_type = PP2_OUTQ_L3_TYPE_IPV4;
1730 *gen_l3_cksum = ol_flags & PKT_TX_IP_CKSUM ? 1 : 0;
1731 } else if (ol_flags & PKT_TX_IPV6) {
1732 *l3_type = PP2_OUTQ_L3_TYPE_IPV6;
1733 /* no checksum for ipv6 header */
1736 /* if something different then stop processing */
1740 ol_flags &= PKT_TX_L4_MASK;
1741 if ((packet_type & RTE_PTYPE_L4_TCP) &&
1742 ol_flags == PKT_TX_TCP_CKSUM) {
1743 *l4_type = PP2_OUTQ_L4_TYPE_TCP;
1745 } else if ((packet_type & RTE_PTYPE_L4_UDP) &&
1746 ol_flags == PKT_TX_UDP_CKSUM) {
1747 *l4_type = PP2_OUTQ_L4_TYPE_UDP;
1750 *l4_type = PP2_OUTQ_L4_TYPE_OTHER;
1751 /* no checksum for other type */
1759 * Release already sent buffers to bpool (buffer-pool).
1762 * Pointer to the port structure.
1764 * Pointer to the MUSDK hardware interface.
1766 * Pointer to the shadow queue.
1770 * Force releasing packets.
1773 mrvl_free_sent_buffers(struct pp2_ppio *ppio, struct pp2_hif *hif,
1774 struct mrvl_shadow_txq *sq, int qid, int force)
1776 struct buff_release_entry *entry;
1777 uint16_t nb_done = 0, num = 0, skip_bufs = 0;
1778 int i, core_id = rte_lcore_id();
1780 pp2_ppio_get_num_outq_done(ppio, hif, qid, &nb_done);
1782 sq->num_to_release += nb_done;
1784 if (likely(!force &&
1785 sq->num_to_release < MRVL_PP2_BUF_RELEASE_BURST_SIZE))
1788 nb_done = sq->num_to_release;
1789 sq->num_to_release = 0;
1791 for (i = 0; i < nb_done; i++) {
1792 entry = &sq->ent[sq->tail + num];
1793 if (unlikely(!entry->buff.addr)) {
1795 "Shadow memory @%d: cookie(%lx), pa(%lx)!\n",
1796 sq->tail, (u64)entry->buff.cookie,
1797 (u64)entry->buff.addr);
1802 if (unlikely(!entry->bpool)) {
1803 struct rte_mbuf *mbuf;
1805 mbuf = (struct rte_mbuf *)
1806 (cookie_addr_high | entry->buff.cookie);
1807 rte_pktmbuf_free(mbuf);
1812 mrvl_port_bpool_size
1813 [entry->bpool->pp2_id][entry->bpool->id][core_id]++;
1815 if (unlikely(sq->tail + num == MRVL_PP2_TX_SHADOWQ_SIZE))
1820 pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
1822 sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
1828 pp2_bpool_put_buffs(hif, &sq->ent[sq->tail], &num);
1829 sq->tail = (sq->tail + num) & MRVL_PP2_TX_SHADOWQ_MASK;
1835 * DPDK callback for transmit.
1838 * Generic pointer transmit queue.
1840 * Packets to transmit.
1842 * Number of packets in array.
1845 * Number of packets successfully transmitted.
1848 mrvl_tx_pkt_burst(void *txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1850 struct mrvl_txq *q = txq;
1851 struct mrvl_shadow_txq *sq = &shadow_txqs[q->port_id][rte_lcore_id()];
1852 struct pp2_hif *hif = hifs[rte_lcore_id()];
1853 struct pp2_ppio_desc descs[nb_pkts];
1854 int i, ret, bytes_sent = 0;
1855 uint16_t num, sq_free_size;
1858 if (unlikely(!q->priv->ppio))
1862 mrvl_free_sent_buffers(q->priv->ppio, hif, sq, q->queue_id, 0);
1864 sq_free_size = MRVL_PP2_TX_SHADOWQ_SIZE - sq->size - 1;
1865 if (unlikely(nb_pkts > sq_free_size)) {
1867 "No room in shadow queue for %d packets! %d packets will be sent.\n",
1868 nb_pkts, sq_free_size);
1869 nb_pkts = sq_free_size;
1872 for (i = 0; i < nb_pkts; i++) {
1873 struct rte_mbuf *mbuf = tx_pkts[i];
1874 int gen_l3_cksum, gen_l4_cksum;
1875 enum pp2_outq_l3_type l3_type;
1876 enum pp2_outq_l4_type l4_type;
1878 if (likely(nb_pkts - i > MRVL_MUSDK_PREFETCH_SHIFT)) {
1879 struct rte_mbuf *pref_pkt_hdr;
1881 pref_pkt_hdr = tx_pkts[i + MRVL_MUSDK_PREFETCH_SHIFT];
1882 rte_mbuf_prefetch_part1(pref_pkt_hdr);
1883 rte_mbuf_prefetch_part2(pref_pkt_hdr);
1886 sq->ent[sq->head].buff.cookie = (pp2_cookie_t)(uint64_t)mbuf;
1887 sq->ent[sq->head].buff.addr =
1888 rte_mbuf_data_iova_default(mbuf);
1889 sq->ent[sq->head].bpool =
1890 (unlikely(mbuf->port == 0xff || mbuf->refcnt > 1)) ?
1891 NULL : mrvl_port_to_bpool_lookup[mbuf->port];
1892 sq->head = (sq->head + 1) & MRVL_PP2_TX_SHADOWQ_MASK;
1895 pp2_ppio_outq_desc_reset(&descs[i]);
1896 pp2_ppio_outq_desc_set_phys_addr(&descs[i],
1897 rte_pktmbuf_iova(mbuf));
1898 pp2_ppio_outq_desc_set_pkt_offset(&descs[i], 0);
1899 pp2_ppio_outq_desc_set_pkt_len(&descs[i],
1900 rte_pktmbuf_pkt_len(mbuf));
1902 bytes_sent += rte_pktmbuf_pkt_len(mbuf);
1904 * in case unsupported ol_flags were passed
1905 * do not update descriptor offload information
1907 ret = mrvl_prepare_proto_info(mbuf->ol_flags, mbuf->packet_type,
1908 &l3_type, &l4_type, &gen_l3_cksum,
1913 pp2_ppio_outq_desc_set_proto_info(&descs[i], l3_type, l4_type,
1915 mbuf->l2_len + mbuf->l3_len,
1916 gen_l3_cksum, gen_l4_cksum);
1920 pp2_ppio_send(q->priv->ppio, hif, q->queue_id, descs, &nb_pkts);
1921 /* number of packets that were not sent */
1922 if (unlikely(num > nb_pkts)) {
1923 for (i = nb_pkts; i < num; i++) {
1924 sq->head = (MRVL_PP2_TX_SHADOWQ_SIZE + sq->head - 1) &
1925 MRVL_PP2_TX_SHADOWQ_MASK;
1926 addr = cookie_addr_high | sq->ent[sq->head].buff.cookie;
1928 rte_pktmbuf_pkt_len((struct rte_mbuf *)addr);
1930 sq->size -= num - nb_pkts;
1933 q->bytes_sent += bytes_sent;
1939 * Initialize packet processor.
1942 * 0 on success, negative error value otherwise.
1947 struct pp2_init_params init_params;
1949 memset(&init_params, 0, sizeof(init_params));
1950 init_params.hif_reserved_map = MRVL_MUSDK_HIFS_RESERVED;
1951 init_params.bm_pool_reserved_map = MRVL_MUSDK_BPOOLS_RESERVED;
1952 init_params.rss_tbl_reserved_map = MRVL_MUSDK_RSS_RESERVED;
1954 return pp2_init(&init_params);
1958 * Deinitialize packet processor.
1961 * 0 on success, negative error value otherwise.
1964 mrvl_deinit_pp2(void)
1970 * Create private device structure.
1973 * Pointer to the port name passed in the initialization parameters.
1976 * Pointer to the newly allocated private device structure.
1978 static struct mrvl_priv *
1979 mrvl_priv_create(const char *dev_name)
1981 struct pp2_bpool_params bpool_params;
1982 char match[MRVL_MATCH_LEN];
1983 struct mrvl_priv *priv;
1986 priv = rte_zmalloc_socket(dev_name, sizeof(*priv), 0, rte_socket_id());
1990 ret = pp2_netdev_get_ppio_info((char *)(uintptr_t)dev_name,
1991 &priv->pp_id, &priv->ppio_id);
1995 bpool_bit = mrvl_reserve_bit(&used_bpools[priv->pp_id],
1996 PP2_BPOOL_NUM_POOLS);
1999 priv->bpool_bit = bpool_bit;
2001 snprintf(match, sizeof(match), "pool-%d:%d", priv->pp_id,
2003 memset(&bpool_params, 0, sizeof(bpool_params));
2004 bpool_params.match = match;
2005 bpool_params.buff_len = MRVL_PKT_SIZE_MAX + MRVL_PKT_EFFEC_OFFS;
2006 ret = pp2_bpool_init(&bpool_params, &priv->bpool);
2008 goto out_clear_bpool_bit;
2010 priv->ppio_params.type = PP2_PPIO_T_NIC;
2011 rte_spinlock_init(&priv->lock);
2014 out_clear_bpool_bit:
2015 used_bpools[priv->pp_id] &= ~(1 << priv->bpool_bit);
2022 * Create device representing Ethernet port.
2025 * Pointer to the port's name.
2028 * 0 on success, negative error value otherwise.
2031 mrvl_eth_dev_create(struct rte_vdev_device *vdev, const char *name)
2033 int ret, fd = socket(AF_INET, SOCK_DGRAM, 0);
2034 struct rte_eth_dev *eth_dev;
2035 struct mrvl_priv *priv;
2038 eth_dev = rte_eth_dev_allocate(name);
2042 priv = mrvl_priv_create(name);
2048 eth_dev->data->mac_addrs =
2049 rte_zmalloc("mac_addrs",
2050 ETHER_ADDR_LEN * MRVL_MAC_ADDRS_MAX, 0);
2051 if (!eth_dev->data->mac_addrs) {
2052 RTE_LOG(ERR, PMD, "Failed to allocate space for eth addrs\n");
2057 memset(&req, 0, sizeof(req));
2058 strcpy(req.ifr_name, name);
2059 ret = ioctl(fd, SIOCGIFHWADDR, &req);
2063 memcpy(eth_dev->data->mac_addrs[0].addr_bytes,
2064 req.ifr_addr.sa_data, ETHER_ADDR_LEN);
2066 eth_dev->rx_pkt_burst = mrvl_rx_pkt_burst;
2067 eth_dev->tx_pkt_burst = mrvl_tx_pkt_burst;
2068 eth_dev->data->dev_private = priv;
2069 eth_dev->device = &vdev->device;
2070 eth_dev->dev_ops = &mrvl_ops;
2074 rte_free(eth_dev->data->mac_addrs);
2076 rte_eth_dev_release_port(eth_dev);
2084 * Cleanup previously created device representing Ethernet port.
2087 * Pointer to the port name.
2090 mrvl_eth_dev_destroy(const char *name)
2092 struct rte_eth_dev *eth_dev;
2093 struct mrvl_priv *priv;
2095 eth_dev = rte_eth_dev_allocated(name);
2099 priv = eth_dev->data->dev_private;
2100 pp2_bpool_deinit(priv->bpool);
2102 rte_free(eth_dev->data->mac_addrs);
2103 rte_eth_dev_release_port(eth_dev);
2107 * Callback used by rte_kvargs_process() during argument parsing.
2110 * Pointer to the parsed key (unused).
2112 * Pointer to the parsed value.
2114 * Pointer to the extra arguments which contains address of the
2115 * table of pointers to parsed interface names.
2121 mrvl_get_ifnames(const char *key __rte_unused, const char *value,
2124 const char **ifnames = extra_args;
2126 ifnames[mrvl_ports_nb++] = value;
2132 * Initialize per-lcore MUSDK hardware interfaces (hifs).
2135 * 0 on success, negative error value otherwise.
2138 mrvl_init_hifs(void)
2140 struct pp2_hif_params params;
2141 char match[MRVL_MATCH_LEN];
2144 RTE_LCORE_FOREACH(i) {
2145 ret = mrvl_reserve_bit(&used_hifs, MRVL_MUSDK_HIFS_MAX);
2149 snprintf(match, sizeof(match), "hif-%d", ret);
2150 memset(¶ms, 0, sizeof(params));
2151 params.match = match;
2152 params.out_size = MRVL_PP2_AGGR_TXQD_MAX;
2153 ret = pp2_hif_init(¶ms, &hifs[i]);
2155 RTE_LOG(ERR, PMD, "Failed to initialize hif %d\n", i);
2164 * Deinitialize per-lcore MUSDK hardware interfaces (hifs).
2167 mrvl_deinit_hifs(void)
2171 RTE_LCORE_FOREACH(i) {
2173 pp2_hif_deinit(hifs[i]);
2177 static void mrvl_set_first_last_cores(int core_id)
2179 if (core_id < mrvl_lcore_first)
2180 mrvl_lcore_first = core_id;
2182 if (core_id > mrvl_lcore_last)
2183 mrvl_lcore_last = core_id;
2187 * DPDK callback to register the virtual device.
2190 * Pointer to the virtual device.
2193 * 0 on success, negative error value otherwise.
2196 rte_pmd_mrvl_probe(struct rte_vdev_device *vdev)
2198 struct rte_kvargs *kvlist;
2199 const char *ifnames[PP2_NUM_ETH_PPIO * PP2_NUM_PKT_PROC];
2201 uint32_t i, ifnum, cfgnum, core_id;
2204 params = rte_vdev_device_args(vdev);
2208 kvlist = rte_kvargs_parse(params, valid_args);
2212 ifnum = rte_kvargs_count(kvlist, MRVL_IFACE_NAME_ARG);
2213 if (ifnum > RTE_DIM(ifnames))
2214 goto out_free_kvlist;
2216 rte_kvargs_process(kvlist, MRVL_IFACE_NAME_ARG,
2217 mrvl_get_ifnames, &ifnames);
2219 cfgnum = rte_kvargs_count(kvlist, MRVL_CFG_ARG);
2221 RTE_LOG(ERR, PMD, "Cannot handle more than one config file!\n");
2222 goto out_free_kvlist;
2223 } else if (cfgnum == 1) {
2224 rte_kvargs_process(kvlist, MRVL_CFG_ARG,
2225 mrvl_get_qoscfg, &mrvl_qos_cfg);
2229 * ret == -EEXIST is correct, it means DMA
2230 * has been already initialized (by another PMD).
2232 ret = mv_sys_dma_mem_init(MRVL_MUSDK_DMA_MEMSIZE);
2235 goto out_free_kvlist;
2238 "DMA memory has been already initialized by a different driver.\n");
2241 ret = mrvl_init_pp2();
2243 RTE_LOG(ERR, PMD, "Failed to init PP!\n");
2244 goto out_deinit_dma;
2247 ret = mrvl_init_hifs();
2249 goto out_deinit_hifs;
2251 for (i = 0; i < ifnum; i++) {
2252 RTE_LOG(INFO, PMD, "Creating %s\n", ifnames[i]);
2253 ret = mrvl_eth_dev_create(vdev, ifnames[i]);
2258 rte_kvargs_free(kvlist);
2260 memset(mrvl_port_bpool_size, 0, sizeof(mrvl_port_bpool_size));
2262 mrvl_lcore_first = RTE_MAX_LCORE;
2263 mrvl_lcore_last = 0;
2265 RTE_LCORE_FOREACH(core_id) {
2266 mrvl_set_first_last_cores(core_id);
2272 mrvl_eth_dev_destroy(ifnames[i]);
2277 mv_sys_dma_mem_destroy();
2279 rte_kvargs_free(kvlist);
2285 * DPDK callback to remove virtual device.
2288 * Pointer to the removed virtual device.
2291 * 0 on success, negative error value otherwise.
2294 rte_pmd_mrvl_remove(struct rte_vdev_device *vdev)
2299 name = rte_vdev_device_name(vdev);
2303 RTE_LOG(INFO, PMD, "Removing %s\n", name);
2305 for (i = 0; i < rte_eth_dev_count(); i++) {
2306 char ifname[RTE_ETH_NAME_MAX_LEN];
2308 rte_eth_dev_get_name_by_port(i, ifname);
2309 mrvl_eth_dev_destroy(ifname);
2314 mv_sys_dma_mem_destroy();
2319 static struct rte_vdev_driver pmd_mrvl_drv = {
2320 .probe = rte_pmd_mrvl_probe,
2321 .remove = rte_pmd_mrvl_remove,
2324 RTE_PMD_REGISTER_VDEV(net_mrvl, pmd_mrvl_drv);
2325 RTE_PMD_REGISTER_ALIAS(net_mrvl, eth_mrvl);