1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright (c) 2014-2018 Netronome Systems, Inc.
5 * Small portions derived from code Copyright(c) 2010-2015 Intel Corporation.
9 * vim:shiftwidth=8:noexpandtab
11 * @file dpdk/pmd/nfp_common.c
13 * Netronome vNIC DPDK Poll-Mode Driver: Common files
16 #include <rte_byteorder.h>
17 #include <rte_common.h>
19 #include <rte_debug.h>
20 #include <ethdev_driver.h>
21 #include <ethdev_pci.h>
23 #include <rte_ether.h>
24 #include <rte_malloc.h>
25 #include <rte_memzone.h>
26 #include <rte_mempool.h>
27 #include <rte_version.h>
28 #include <rte_string_fns.h>
29 #include <rte_alarm.h>
30 #include <rte_spinlock.h>
31 #include <rte_service_component.h>
33 #include "nfpcore/nfp_cpp.h"
34 #include "nfpcore/nfp_nffw.h"
35 #include "nfpcore/nfp_hwinfo.h"
36 #include "nfpcore/nfp_mip.h"
37 #include "nfpcore/nfp_rtsym.h"
38 #include "nfpcore/nfp_nsp.h"
40 #include "nfp_common.h"
44 #include "nfp_cpp_bridge.h"
46 #include <sys/types.h>
47 #include <sys/socket.h>
51 #include <sys/ioctl.h>
55 __nfp_net_reconfig(struct nfp_net_hw *hw, uint32_t update)
61 PMD_DRV_LOG(DEBUG, "Writing to the configuration queue (%p)...",
64 if (hw->qcp_cfg == NULL)
65 rte_panic("Bad configuration queue pointer\n");
67 nfp_qcp_ptr_add(hw->qcp_cfg, NFP_QCP_WRITE_PTR, 1);
70 wait.tv_nsec = 1000000;
72 PMD_DRV_LOG(DEBUG, "Polling for update ack...");
74 /* Poll update field, waiting for NFP to ack the config */
75 for (cnt = 0; ; cnt++) {
76 new = nn_cfg_readl(hw, NFP_NET_CFG_UPDATE);
79 if (new & NFP_NET_CFG_UPDATE_ERR) {
80 PMD_INIT_LOG(ERR, "Reconfig error: 0x%08x", new);
83 if (cnt >= NFP_NET_POLL_TIMEOUT) {
84 PMD_INIT_LOG(ERR, "Reconfig timeout for 0x%08x after"
85 " %dms", update, cnt);
86 rte_panic("Exiting\n");
88 nanosleep(&wait, 0); /* waiting for a 1ms */
90 PMD_DRV_LOG(DEBUG, "Ack DONE");
96 * @nn: device to reconfigure
97 * @ctrl: The value for the ctrl field in the BAR config
98 * @update: The value for the update field in the BAR config
100 * Write the update word to the BAR and ping the reconfig queue. Then poll
101 * until the firmware has acknowledged the update by zeroing the update word.
104 nfp_net_reconfig(struct nfp_net_hw *hw, uint32_t ctrl, uint32_t update)
108 PMD_DRV_LOG(DEBUG, "nfp_net_reconfig: ctrl=%08x update=%08x",
111 rte_spinlock_lock(&hw->reconfig_lock);
113 nn_cfg_writel(hw, NFP_NET_CFG_CTRL, ctrl);
114 nn_cfg_writel(hw, NFP_NET_CFG_UPDATE, update);
118 err = __nfp_net_reconfig(hw, update);
120 rte_spinlock_unlock(&hw->reconfig_lock);
126 * Reconfig errors imply situations where they can be handled.
127 * Otherwise, rte_panic is called inside __nfp_net_reconfig
129 PMD_INIT_LOG(ERR, "Error nfp_net reconfig for ctrl: %x update: %x",
135 * Configure an Ethernet device. This function must be invoked first
136 * before any other function in the Ethernet API. This function can
137 * also be re-invoked when a device is in the stopped state.
140 nfp_net_configure(struct rte_eth_dev *dev)
142 struct rte_eth_conf *dev_conf;
143 struct rte_eth_rxmode *rxmode;
144 struct rte_eth_txmode *txmode;
145 struct nfp_net_hw *hw;
147 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
150 * A DPDK app sends info about how many queues to use and how
151 * those queues need to be configured. This is used by the
152 * DPDK core and it makes sure no more queues than those
153 * advertised by the driver are requested. This function is
154 * called after that internal process
157 PMD_INIT_LOG(DEBUG, "Configure");
159 dev_conf = &dev->data->dev_conf;
160 rxmode = &dev_conf->rxmode;
161 txmode = &dev_conf->txmode;
163 if (rxmode->mq_mode & RTE_ETH_MQ_RX_RSS_FLAG)
164 rxmode->offloads |= RTE_ETH_RX_OFFLOAD_RSS_HASH;
166 /* Checking TX mode */
167 if (txmode->mq_mode) {
168 PMD_INIT_LOG(INFO, "TX mq_mode DCB and VMDq not supported");
172 /* Checking RX mode */
173 if (rxmode->mq_mode & RTE_ETH_MQ_RX_RSS &&
174 !(hw->cap & NFP_NET_CFG_CTRL_RSS)) {
175 PMD_INIT_LOG(INFO, "RSS not supported");
183 nfp_net_enable_queues(struct rte_eth_dev *dev)
185 struct nfp_net_hw *hw;
186 uint64_t enabled_queues = 0;
189 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
191 /* Enabling the required TX queues in the device */
192 for (i = 0; i < dev->data->nb_tx_queues; i++)
193 enabled_queues |= (1 << i);
195 nn_cfg_writeq(hw, NFP_NET_CFG_TXRS_ENABLE, enabled_queues);
199 /* Enabling the required RX queues in the device */
200 for (i = 0; i < dev->data->nb_rx_queues; i++)
201 enabled_queues |= (1 << i);
203 nn_cfg_writeq(hw, NFP_NET_CFG_RXRS_ENABLE, enabled_queues);
207 nfp_net_disable_queues(struct rte_eth_dev *dev)
209 struct nfp_net_hw *hw;
210 uint32_t new_ctrl, update = 0;
212 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
214 nn_cfg_writeq(hw, NFP_NET_CFG_TXRS_ENABLE, 0);
215 nn_cfg_writeq(hw, NFP_NET_CFG_RXRS_ENABLE, 0);
217 new_ctrl = hw->ctrl & ~NFP_NET_CFG_CTRL_ENABLE;
218 update = NFP_NET_CFG_UPDATE_GEN | NFP_NET_CFG_UPDATE_RING |
219 NFP_NET_CFG_UPDATE_MSIX;
221 if (hw->cap & NFP_NET_CFG_CTRL_RINGCFG)
222 new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG;
224 /* If an error when reconfig we avoid to change hw state */
225 if (nfp_net_reconfig(hw, new_ctrl, update) < 0)
232 nfp_net_params_setup(struct nfp_net_hw *hw)
234 nn_cfg_writel(hw, NFP_NET_CFG_MTU, hw->mtu);
235 nn_cfg_writel(hw, NFP_NET_CFG_FLBUFSZ, hw->flbufsz);
239 nfp_net_cfg_queue_setup(struct nfp_net_hw *hw)
241 hw->qcp_cfg = hw->tx_bar + NFP_QCP_QUEUE_ADDR_SZ;
244 #define ETH_ADDR_LEN 6
247 nfp_eth_copy_mac(uint8_t *dst, const uint8_t *src)
251 for (i = 0; i < ETH_ADDR_LEN; i++)
256 nfp_net_write_mac(struct nfp_net_hw *hw, uint8_t *mac)
258 uint32_t mac0 = *(uint32_t *)mac;
261 nn_writel(rte_cpu_to_be_32(mac0), hw->ctrl_bar + NFP_NET_CFG_MACADDR);
264 mac1 = *(uint16_t *)mac;
265 nn_writew(rte_cpu_to_be_16(mac1),
266 hw->ctrl_bar + NFP_NET_CFG_MACADDR + 6);
270 nfp_set_mac_addr(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
272 struct nfp_net_hw *hw;
273 uint32_t update, ctrl;
275 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
276 if ((hw->ctrl & NFP_NET_CFG_CTRL_ENABLE) &&
277 !(hw->cap & NFP_NET_CFG_CTRL_LIVE_ADDR)) {
278 PMD_INIT_LOG(INFO, "MAC address unable to change when"
283 /* Writing new MAC to the specific port BAR address */
284 nfp_net_write_mac(hw, (uint8_t *)mac_addr);
286 /* Signal the NIC about the change */
287 update = NFP_NET_CFG_UPDATE_MACADDR;
289 if ((hw->ctrl & NFP_NET_CFG_CTRL_ENABLE) &&
290 (hw->cap & NFP_NET_CFG_CTRL_LIVE_ADDR))
291 ctrl |= NFP_NET_CFG_CTRL_LIVE_ADDR;
292 if (nfp_net_reconfig(hw, ctrl, update) < 0) {
293 PMD_INIT_LOG(INFO, "MAC address update failed");
300 nfp_configure_rx_interrupt(struct rte_eth_dev *dev,
301 struct rte_intr_handle *intr_handle)
303 struct nfp_net_hw *hw;
306 if (rte_intr_vec_list_alloc(intr_handle, "intr_vec",
307 dev->data->nb_rx_queues)) {
308 PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
309 " intr_vec", dev->data->nb_rx_queues);
313 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
315 if (rte_intr_type_get(intr_handle) == RTE_INTR_HANDLE_UIO) {
316 PMD_INIT_LOG(INFO, "VF: enabling RX interrupt with UIO");
317 /* UIO just supports one queue and no LSC*/
318 nn_cfg_writeb(hw, NFP_NET_CFG_RXR_VEC(0), 0);
319 if (rte_intr_vec_list_index_set(intr_handle, 0, 0))
322 PMD_INIT_LOG(INFO, "VF: enabling RX interrupt with VFIO");
323 for (i = 0; i < dev->data->nb_rx_queues; i++) {
325 * The first msix vector is reserved for non
328 nn_cfg_writeb(hw, NFP_NET_CFG_RXR_VEC(i), i + 1);
329 if (rte_intr_vec_list_index_set(intr_handle, i,
332 PMD_INIT_LOG(DEBUG, "intr_vec[%d]= %d", i,
333 rte_intr_vec_list_index_get(intr_handle,
338 /* Avoiding TX interrupts */
339 hw->ctrl |= NFP_NET_CFG_CTRL_MSIX_TX_OFF;
344 nfp_check_offloads(struct rte_eth_dev *dev)
346 struct nfp_net_hw *hw;
347 struct rte_eth_conf *dev_conf;
348 struct rte_eth_rxmode *rxmode;
349 struct rte_eth_txmode *txmode;
352 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
354 dev_conf = &dev->data->dev_conf;
355 rxmode = &dev_conf->rxmode;
356 txmode = &dev_conf->txmode;
358 if (rxmode->offloads & RTE_ETH_RX_OFFLOAD_IPV4_CKSUM) {
359 if (hw->cap & NFP_NET_CFG_CTRL_RXCSUM)
360 ctrl |= NFP_NET_CFG_CTRL_RXCSUM;
363 if (rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP) {
364 if (hw->cap & NFP_NET_CFG_CTRL_RXVLAN)
365 ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
368 hw->mtu = dev->data->mtu;
370 if (txmode->offloads & RTE_ETH_TX_OFFLOAD_VLAN_INSERT)
371 ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
374 if (hw->cap & NFP_NET_CFG_CTRL_L2BC)
375 ctrl |= NFP_NET_CFG_CTRL_L2BC;
378 if (hw->cap & NFP_NET_CFG_CTRL_L2MC)
379 ctrl |= NFP_NET_CFG_CTRL_L2MC;
381 /* TX checksum offload */
382 if (txmode->offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM ||
383 txmode->offloads & RTE_ETH_TX_OFFLOAD_UDP_CKSUM ||
384 txmode->offloads & RTE_ETH_TX_OFFLOAD_TCP_CKSUM)
385 ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
388 if (txmode->offloads & RTE_ETH_TX_OFFLOAD_TCP_TSO) {
389 if (hw->cap & NFP_NET_CFG_CTRL_LSO)
390 ctrl |= NFP_NET_CFG_CTRL_LSO;
392 ctrl |= NFP_NET_CFG_CTRL_LSO2;
396 if (txmode->offloads & RTE_ETH_TX_OFFLOAD_MULTI_SEGS)
397 ctrl |= NFP_NET_CFG_CTRL_GATHER;
403 nfp_net_promisc_enable(struct rte_eth_dev *dev)
405 uint32_t new_ctrl, update = 0;
406 struct nfp_net_hw *hw;
409 PMD_DRV_LOG(DEBUG, "Promiscuous mode enable");
411 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
413 if (!(hw->cap & NFP_NET_CFG_CTRL_PROMISC)) {
414 PMD_INIT_LOG(INFO, "Promiscuous mode not supported");
418 if (hw->ctrl & NFP_NET_CFG_CTRL_PROMISC) {
419 PMD_DRV_LOG(INFO, "Promiscuous mode already enabled");
423 new_ctrl = hw->ctrl | NFP_NET_CFG_CTRL_PROMISC;
424 update = NFP_NET_CFG_UPDATE_GEN;
427 * DPDK sets promiscuous mode on just after this call assuming
428 * it can not fail ...
430 ret = nfp_net_reconfig(hw, new_ctrl, update);
440 nfp_net_promisc_disable(struct rte_eth_dev *dev)
442 uint32_t new_ctrl, update = 0;
443 struct nfp_net_hw *hw;
446 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
448 if ((hw->ctrl & NFP_NET_CFG_CTRL_PROMISC) == 0) {
449 PMD_DRV_LOG(INFO, "Promiscuous mode already disabled");
453 new_ctrl = hw->ctrl & ~NFP_NET_CFG_CTRL_PROMISC;
454 update = NFP_NET_CFG_UPDATE_GEN;
457 * DPDK sets promiscuous mode off just before this call
458 * assuming it can not fail ...
460 ret = nfp_net_reconfig(hw, new_ctrl, update);
470 * return 0 means link status changed, -1 means not changed
472 * Wait to complete is needed as it can take up to 9 seconds to get the Link
476 nfp_net_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
478 struct nfp_net_hw *hw;
479 struct rte_eth_link link;
480 uint32_t nn_link_status;
483 static const uint32_t ls_to_ethtool[] = {
484 [NFP_NET_CFG_STS_LINK_RATE_UNSUPPORTED] = RTE_ETH_SPEED_NUM_NONE,
485 [NFP_NET_CFG_STS_LINK_RATE_UNKNOWN] = RTE_ETH_SPEED_NUM_NONE,
486 [NFP_NET_CFG_STS_LINK_RATE_1G] = RTE_ETH_SPEED_NUM_1G,
487 [NFP_NET_CFG_STS_LINK_RATE_10G] = RTE_ETH_SPEED_NUM_10G,
488 [NFP_NET_CFG_STS_LINK_RATE_25G] = RTE_ETH_SPEED_NUM_25G,
489 [NFP_NET_CFG_STS_LINK_RATE_40G] = RTE_ETH_SPEED_NUM_40G,
490 [NFP_NET_CFG_STS_LINK_RATE_50G] = RTE_ETH_SPEED_NUM_50G,
491 [NFP_NET_CFG_STS_LINK_RATE_100G] = RTE_ETH_SPEED_NUM_100G,
494 PMD_DRV_LOG(DEBUG, "Link update");
496 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
498 nn_link_status = nn_cfg_readl(hw, NFP_NET_CFG_STS);
500 memset(&link, 0, sizeof(struct rte_eth_link));
502 if (nn_link_status & NFP_NET_CFG_STS_LINK)
503 link.link_status = RTE_ETH_LINK_UP;
505 link.link_duplex = RTE_ETH_LINK_FULL_DUPLEX;
507 nn_link_status = (nn_link_status >> NFP_NET_CFG_STS_LINK_RATE_SHIFT) &
508 NFP_NET_CFG_STS_LINK_RATE_MASK;
510 if (nn_link_status >= RTE_DIM(ls_to_ethtool))
511 link.link_speed = RTE_ETH_SPEED_NUM_NONE;
513 link.link_speed = ls_to_ethtool[nn_link_status];
515 ret = rte_eth_linkstatus_set(dev, &link);
517 if (link.link_status)
518 PMD_DRV_LOG(INFO, "NIC Link is Up");
520 PMD_DRV_LOG(INFO, "NIC Link is Down");
526 nfp_net_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
529 struct nfp_net_hw *hw;
530 struct rte_eth_stats nfp_dev_stats;
532 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
534 /* RTE_ETHDEV_QUEUE_STAT_CNTRS default value is 16 */
536 memset(&nfp_dev_stats, 0, sizeof(nfp_dev_stats));
538 /* reading per RX ring stats */
539 for (i = 0; i < dev->data->nb_rx_queues; i++) {
540 if (i == RTE_ETHDEV_QUEUE_STAT_CNTRS)
543 nfp_dev_stats.q_ipackets[i] =
544 nn_cfg_readq(hw, NFP_NET_CFG_RXR_STATS(i));
546 nfp_dev_stats.q_ipackets[i] -=
547 hw->eth_stats_base.q_ipackets[i];
549 nfp_dev_stats.q_ibytes[i] =
550 nn_cfg_readq(hw, NFP_NET_CFG_RXR_STATS(i) + 0x8);
552 nfp_dev_stats.q_ibytes[i] -=
553 hw->eth_stats_base.q_ibytes[i];
556 /* reading per TX ring stats */
557 for (i = 0; i < dev->data->nb_tx_queues; i++) {
558 if (i == RTE_ETHDEV_QUEUE_STAT_CNTRS)
561 nfp_dev_stats.q_opackets[i] =
562 nn_cfg_readq(hw, NFP_NET_CFG_TXR_STATS(i));
564 nfp_dev_stats.q_opackets[i] -=
565 hw->eth_stats_base.q_opackets[i];
567 nfp_dev_stats.q_obytes[i] =
568 nn_cfg_readq(hw, NFP_NET_CFG_TXR_STATS(i) + 0x8);
570 nfp_dev_stats.q_obytes[i] -=
571 hw->eth_stats_base.q_obytes[i];
574 nfp_dev_stats.ipackets =
575 nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_FRAMES);
577 nfp_dev_stats.ipackets -= hw->eth_stats_base.ipackets;
579 nfp_dev_stats.ibytes =
580 nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_OCTETS);
582 nfp_dev_stats.ibytes -= hw->eth_stats_base.ibytes;
584 nfp_dev_stats.opackets =
585 nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_FRAMES);
587 nfp_dev_stats.opackets -= hw->eth_stats_base.opackets;
589 nfp_dev_stats.obytes =
590 nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_OCTETS);
592 nfp_dev_stats.obytes -= hw->eth_stats_base.obytes;
594 /* reading general device stats */
595 nfp_dev_stats.ierrors =
596 nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_ERRORS);
598 nfp_dev_stats.ierrors -= hw->eth_stats_base.ierrors;
600 nfp_dev_stats.oerrors =
601 nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_ERRORS);
603 nfp_dev_stats.oerrors -= hw->eth_stats_base.oerrors;
605 /* RX ring mbuf allocation failures */
606 nfp_dev_stats.rx_nombuf = dev->data->rx_mbuf_alloc_failed;
608 nfp_dev_stats.imissed =
609 nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_DISCARDS);
611 nfp_dev_stats.imissed -= hw->eth_stats_base.imissed;
614 memcpy(stats, &nfp_dev_stats, sizeof(*stats));
621 nfp_net_stats_reset(struct rte_eth_dev *dev)
624 struct nfp_net_hw *hw;
626 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
629 * hw->eth_stats_base records the per counter starting point.
633 /* reading per RX ring stats */
634 for (i = 0; i < dev->data->nb_rx_queues; i++) {
635 if (i == RTE_ETHDEV_QUEUE_STAT_CNTRS)
638 hw->eth_stats_base.q_ipackets[i] =
639 nn_cfg_readq(hw, NFP_NET_CFG_RXR_STATS(i));
641 hw->eth_stats_base.q_ibytes[i] =
642 nn_cfg_readq(hw, NFP_NET_CFG_RXR_STATS(i) + 0x8);
645 /* reading per TX ring stats */
646 for (i = 0; i < dev->data->nb_tx_queues; i++) {
647 if (i == RTE_ETHDEV_QUEUE_STAT_CNTRS)
650 hw->eth_stats_base.q_opackets[i] =
651 nn_cfg_readq(hw, NFP_NET_CFG_TXR_STATS(i));
653 hw->eth_stats_base.q_obytes[i] =
654 nn_cfg_readq(hw, NFP_NET_CFG_TXR_STATS(i) + 0x8);
657 hw->eth_stats_base.ipackets =
658 nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_FRAMES);
660 hw->eth_stats_base.ibytes =
661 nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_OCTETS);
663 hw->eth_stats_base.opackets =
664 nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_FRAMES);
666 hw->eth_stats_base.obytes =
667 nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_OCTETS);
669 /* reading general device stats */
670 hw->eth_stats_base.ierrors =
671 nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_ERRORS);
673 hw->eth_stats_base.oerrors =
674 nn_cfg_readq(hw, NFP_NET_CFG_STATS_TX_ERRORS);
676 /* RX ring mbuf allocation failures */
677 dev->data->rx_mbuf_alloc_failed = 0;
679 hw->eth_stats_base.imissed =
680 nn_cfg_readq(hw, NFP_NET_CFG_STATS_RX_DISCARDS);
686 nfp_net_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
688 struct nfp_net_hw *hw;
690 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
692 dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
693 dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
694 dev_info->min_rx_bufsize = RTE_ETHER_MIN_MTU;
696 * The maximum rx packet length (max_rx_pktlen) is set to the
697 * maximum supported frame size that the NFP can handle. This
698 * includes layer 2 headers, CRC and other metadata that can
699 * optionally be used.
700 * The maximum layer 3 MTU (max_mtu) is read from hardware,
701 * which was set by the firmware loaded onto the card.
703 dev_info->max_rx_pktlen = NFP_FRAME_SIZE_MAX;
704 dev_info->max_mtu = hw->max_mtu;
705 /* Next should change when PF support is implemented */
706 dev_info->max_mac_addrs = 1;
708 if (hw->cap & NFP_NET_CFG_CTRL_RXVLAN)
709 dev_info->rx_offload_capa = RTE_ETH_RX_OFFLOAD_VLAN_STRIP;
711 if (hw->cap & NFP_NET_CFG_CTRL_RXCSUM)
712 dev_info->rx_offload_capa |= RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
713 RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
714 RTE_ETH_RX_OFFLOAD_TCP_CKSUM;
716 if (hw->cap & NFP_NET_CFG_CTRL_TXVLAN)
717 dev_info->tx_offload_capa = RTE_ETH_TX_OFFLOAD_VLAN_INSERT;
719 if (hw->cap & NFP_NET_CFG_CTRL_TXCSUM)
720 dev_info->tx_offload_capa |= RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
721 RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
722 RTE_ETH_TX_OFFLOAD_TCP_CKSUM;
724 if (hw->cap & NFP_NET_CFG_CTRL_LSO_ANY)
725 dev_info->tx_offload_capa |= RTE_ETH_TX_OFFLOAD_TCP_TSO;
727 if (hw->cap & NFP_NET_CFG_CTRL_GATHER)
728 dev_info->tx_offload_capa |= RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
730 dev_info->default_rxconf = (struct rte_eth_rxconf) {
732 .pthresh = DEFAULT_RX_PTHRESH,
733 .hthresh = DEFAULT_RX_HTHRESH,
734 .wthresh = DEFAULT_RX_WTHRESH,
736 .rx_free_thresh = DEFAULT_RX_FREE_THRESH,
740 dev_info->default_txconf = (struct rte_eth_txconf) {
742 .pthresh = DEFAULT_TX_PTHRESH,
743 .hthresh = DEFAULT_TX_HTHRESH,
744 .wthresh = DEFAULT_TX_WTHRESH,
746 .tx_free_thresh = DEFAULT_TX_FREE_THRESH,
747 .tx_rs_thresh = DEFAULT_TX_RSBIT_THRESH,
750 dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
751 .nb_max = NFP_NET_MAX_RX_DESC,
752 .nb_min = NFP_NET_MIN_RX_DESC,
753 .nb_align = NFP_ALIGN_RING_DESC,
756 dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
757 .nb_max = NFP_NET_MAX_TX_DESC,
758 .nb_min = NFP_NET_MIN_TX_DESC,
759 .nb_align = NFP_ALIGN_RING_DESC,
760 .nb_seg_max = NFP_TX_MAX_SEG,
761 .nb_mtu_seg_max = NFP_TX_MAX_MTU_SEG,
764 if (hw->cap & NFP_NET_CFG_CTRL_RSS) {
765 dev_info->rx_offload_capa |= RTE_ETH_RX_OFFLOAD_RSS_HASH;
767 dev_info->flow_type_rss_offloads = RTE_ETH_RSS_IPV4 |
768 RTE_ETH_RSS_NONFRAG_IPV4_TCP |
769 RTE_ETH_RSS_NONFRAG_IPV4_UDP |
771 RTE_ETH_RSS_NONFRAG_IPV6_TCP |
772 RTE_ETH_RSS_NONFRAG_IPV6_UDP;
774 dev_info->reta_size = NFP_NET_CFG_RSS_ITBL_SZ;
775 dev_info->hash_key_size = NFP_NET_CFG_RSS_KEY_SZ;
778 dev_info->speed_capa = RTE_ETH_LINK_SPEED_1G | RTE_ETH_LINK_SPEED_10G |
779 RTE_ETH_LINK_SPEED_25G | RTE_ETH_LINK_SPEED_40G |
780 RTE_ETH_LINK_SPEED_50G | RTE_ETH_LINK_SPEED_100G;
786 nfp_net_supported_ptypes_get(struct rte_eth_dev *dev)
788 static const uint32_t ptypes[] = {
789 /* refers to nfp_net_set_hash() */
790 RTE_PTYPE_INNER_L3_IPV4,
791 RTE_PTYPE_INNER_L3_IPV6,
792 RTE_PTYPE_INNER_L3_IPV6_EXT,
793 RTE_PTYPE_INNER_L4_MASK,
797 if (dev->rx_pkt_burst == nfp_net_recv_pkts)
803 nfp_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
805 struct rte_pci_device *pci_dev;
806 struct nfp_net_hw *hw;
809 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
810 pci_dev = RTE_ETH_DEV_TO_PCI(dev);
812 if (rte_intr_type_get(pci_dev->intr_handle) !=
816 /* Make sure all updates are written before un-masking */
818 nn_cfg_writeb(hw, NFP_NET_CFG_ICR(base + queue_id),
819 NFP_NET_CFG_ICR_UNMASKED);
824 nfp_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
826 struct rte_pci_device *pci_dev;
827 struct nfp_net_hw *hw;
830 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
831 pci_dev = RTE_ETH_DEV_TO_PCI(dev);
833 if (rte_intr_type_get(pci_dev->intr_handle) !=
837 /* Make sure all updates are written before un-masking */
839 nn_cfg_writeb(hw, NFP_NET_CFG_ICR(base + queue_id), 0x1);
844 nfp_net_dev_link_status_print(struct rte_eth_dev *dev)
846 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
847 struct rte_eth_link link;
849 rte_eth_linkstatus_get(dev, &link);
850 if (link.link_status)
851 PMD_DRV_LOG(INFO, "Port %d: Link Up - speed %u Mbps - %s",
852 dev->data->port_id, link.link_speed,
853 link.link_duplex == RTE_ETH_LINK_FULL_DUPLEX
854 ? "full-duplex" : "half-duplex");
856 PMD_DRV_LOG(INFO, " Port %d: Link Down",
859 PMD_DRV_LOG(INFO, "PCI Address: " PCI_PRI_FMT,
860 pci_dev->addr.domain, pci_dev->addr.bus,
861 pci_dev->addr.devid, pci_dev->addr.function);
864 /* Interrupt configuration and handling */
867 * nfp_net_irq_unmask - Unmask an interrupt
869 * If MSI-X auto-masking is enabled clear the mask bit, otherwise
870 * clear the ICR for the entry.
873 nfp_net_irq_unmask(struct rte_eth_dev *dev)
875 struct nfp_net_hw *hw;
876 struct rte_pci_device *pci_dev;
878 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
879 pci_dev = RTE_ETH_DEV_TO_PCI(dev);
881 if (hw->ctrl & NFP_NET_CFG_CTRL_MSIXAUTO) {
882 /* If MSI-X auto-masking is used, clear the entry */
884 rte_intr_ack(pci_dev->intr_handle);
886 /* Make sure all updates are written before un-masking */
888 nn_cfg_writeb(hw, NFP_NET_CFG_ICR(NFP_NET_IRQ_LSC_IDX),
889 NFP_NET_CFG_ICR_UNMASKED);
894 * Interrupt handler which shall be registered for alarm callback for delayed
895 * handling specific interrupt to wait for the stable nic state. As the NIC
896 * interrupt state is not stable for nfp after link is just down, it needs
897 * to wait 4 seconds to get the stable status.
899 * @param handle Pointer to interrupt handle.
900 * @param param The address of parameter (struct rte_eth_dev *)
905 nfp_net_dev_interrupt_delayed_handler(void *param)
907 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
909 nfp_net_link_update(dev, 0);
910 rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
912 nfp_net_dev_link_status_print(dev);
915 nfp_net_irq_unmask(dev);
919 nfp_net_dev_interrupt_handler(void *param)
922 struct rte_eth_link link;
923 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
925 PMD_DRV_LOG(DEBUG, "We got a LSC interrupt!!!");
927 rte_eth_linkstatus_get(dev, &link);
929 nfp_net_link_update(dev, 0);
932 if (!link.link_status) {
933 /* handle it 1 sec later, wait it being stable */
934 timeout = NFP_NET_LINK_UP_CHECK_TIMEOUT;
937 /* handle it 4 sec later, wait it being stable */
938 timeout = NFP_NET_LINK_DOWN_CHECK_TIMEOUT;
941 if (rte_eal_alarm_set(timeout * 1000,
942 nfp_net_dev_interrupt_delayed_handler,
944 PMD_INIT_LOG(ERR, "Error setting alarm");
946 nfp_net_irq_unmask(dev);
951 nfp_net_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
953 struct nfp_net_hw *hw;
955 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
957 /* mtu setting is forbidden if port is started */
958 if (dev->data->dev_started) {
959 PMD_DRV_LOG(ERR, "port %d must be stopped before configuration",
964 /* writing to configuration space */
965 nn_cfg_writel(hw, NFP_NET_CFG_MTU, mtu);
973 nfp_net_vlan_offload_set(struct rte_eth_dev *dev, int mask)
975 uint32_t new_ctrl, update;
976 struct nfp_net_hw *hw;
979 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
982 /* Enable vlan strip if it is not configured yet */
983 if ((mask & RTE_ETH_VLAN_STRIP_OFFLOAD) &&
984 !(hw->ctrl & NFP_NET_CFG_CTRL_RXVLAN))
985 new_ctrl = hw->ctrl | NFP_NET_CFG_CTRL_RXVLAN;
987 /* Disable vlan strip just if it is configured */
988 if (!(mask & RTE_ETH_VLAN_STRIP_OFFLOAD) &&
989 (hw->ctrl & NFP_NET_CFG_CTRL_RXVLAN))
990 new_ctrl = hw->ctrl & ~NFP_NET_CFG_CTRL_RXVLAN;
995 update = NFP_NET_CFG_UPDATE_GEN;
997 ret = nfp_net_reconfig(hw, new_ctrl, update);
1005 nfp_net_rss_reta_write(struct rte_eth_dev *dev,
1006 struct rte_eth_rss_reta_entry64 *reta_conf,
1009 uint32_t reta, mask;
1012 struct nfp_net_hw *hw =
1013 NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1015 if (reta_size != NFP_NET_CFG_RSS_ITBL_SZ) {
1016 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
1017 "(%d) doesn't match the number hardware can supported "
1018 "(%d)", reta_size, NFP_NET_CFG_RSS_ITBL_SZ);
1023 * Update Redirection Table. There are 128 8bit-entries which can be
1024 * manage as 32 32bit-entries
1026 for (i = 0; i < reta_size; i += 4) {
1027 /* Handling 4 RSS entries per loop */
1028 idx = i / RTE_ETH_RETA_GROUP_SIZE;
1029 shift = i % RTE_ETH_RETA_GROUP_SIZE;
1030 mask = (uint8_t)((reta_conf[idx].mask >> shift) & 0xF);
1036 /* If all 4 entries were set, don't need read RETA register */
1038 reta = nn_cfg_readl(hw, NFP_NET_CFG_RSS_ITBL + i);
1040 for (j = 0; j < 4; j++) {
1041 if (!(mask & (0x1 << j)))
1044 /* Clearing the entry bits */
1045 reta &= ~(0xFF << (8 * j));
1046 reta |= reta_conf[idx].reta[shift + j] << (8 * j);
1048 nn_cfg_writel(hw, NFP_NET_CFG_RSS_ITBL + (idx * 64) + shift,
1054 /* Update Redirection Table(RETA) of Receive Side Scaling of Ethernet device */
1056 nfp_net_reta_update(struct rte_eth_dev *dev,
1057 struct rte_eth_rss_reta_entry64 *reta_conf,
1060 struct nfp_net_hw *hw =
1061 NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1065 if (!(hw->ctrl & NFP_NET_CFG_CTRL_RSS))
1068 ret = nfp_net_rss_reta_write(dev, reta_conf, reta_size);
1072 update = NFP_NET_CFG_UPDATE_RSS;
1074 if (nfp_net_reconfig(hw, hw->ctrl, update) < 0)
1080 /* Query Redirection Table(RETA) of Receive Side Scaling of Ethernet device. */
1082 nfp_net_reta_query(struct rte_eth_dev *dev,
1083 struct rte_eth_rss_reta_entry64 *reta_conf,
1089 struct nfp_net_hw *hw;
1091 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1093 if (!(hw->ctrl & NFP_NET_CFG_CTRL_RSS))
1096 if (reta_size != NFP_NET_CFG_RSS_ITBL_SZ) {
1097 PMD_DRV_LOG(ERR, "The size of hash lookup table configured "
1098 "(%d) doesn't match the number hardware can supported "
1099 "(%d)", reta_size, NFP_NET_CFG_RSS_ITBL_SZ);
1104 * Reading Redirection Table. There are 128 8bit-entries which can be
1105 * manage as 32 32bit-entries
1107 for (i = 0; i < reta_size; i += 4) {
1108 /* Handling 4 RSS entries per loop */
1109 idx = i / RTE_ETH_RETA_GROUP_SIZE;
1110 shift = i % RTE_ETH_RETA_GROUP_SIZE;
1111 mask = (uint8_t)((reta_conf[idx].mask >> shift) & 0xF);
1116 reta = nn_cfg_readl(hw, NFP_NET_CFG_RSS_ITBL + (idx * 64) +
1118 for (j = 0; j < 4; j++) {
1119 if (!(mask & (0x1 << j)))
1121 reta_conf[idx].reta[shift + j] =
1122 (uint8_t)((reta >> (8 * j)) & 0xF);
1129 nfp_net_rss_hash_write(struct rte_eth_dev *dev,
1130 struct rte_eth_rss_conf *rss_conf)
1132 struct nfp_net_hw *hw;
1134 uint32_t cfg_rss_ctrl = 0;
1138 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1140 /* Writing the key byte a byte */
1141 for (i = 0; i < rss_conf->rss_key_len; i++) {
1142 memcpy(&key, &rss_conf->rss_key[i], 1);
1143 nn_cfg_writeb(hw, NFP_NET_CFG_RSS_KEY + i, key);
1146 rss_hf = rss_conf->rss_hf;
1148 if (rss_hf & RTE_ETH_RSS_IPV4)
1149 cfg_rss_ctrl |= NFP_NET_CFG_RSS_IPV4;
1151 if (rss_hf & RTE_ETH_RSS_NONFRAG_IPV4_TCP)
1152 cfg_rss_ctrl |= NFP_NET_CFG_RSS_IPV4_TCP;
1154 if (rss_hf & RTE_ETH_RSS_NONFRAG_IPV4_UDP)
1155 cfg_rss_ctrl |= NFP_NET_CFG_RSS_IPV4_UDP;
1157 if (rss_hf & RTE_ETH_RSS_IPV6)
1158 cfg_rss_ctrl |= NFP_NET_CFG_RSS_IPV6;
1160 if (rss_hf & RTE_ETH_RSS_NONFRAG_IPV6_TCP)
1161 cfg_rss_ctrl |= NFP_NET_CFG_RSS_IPV6_TCP;
1163 if (rss_hf & RTE_ETH_RSS_NONFRAG_IPV6_UDP)
1164 cfg_rss_ctrl |= NFP_NET_CFG_RSS_IPV6_UDP;
1166 cfg_rss_ctrl |= NFP_NET_CFG_RSS_MASK;
1167 cfg_rss_ctrl |= NFP_NET_CFG_RSS_TOEPLITZ;
1169 /* configuring where to apply the RSS hash */
1170 nn_cfg_writel(hw, NFP_NET_CFG_RSS_CTRL, cfg_rss_ctrl);
1172 /* Writing the key size */
1173 nn_cfg_writeb(hw, NFP_NET_CFG_RSS_KEY_SZ, rss_conf->rss_key_len);
1179 nfp_net_rss_hash_update(struct rte_eth_dev *dev,
1180 struct rte_eth_rss_conf *rss_conf)
1184 struct nfp_net_hw *hw;
1186 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1188 rss_hf = rss_conf->rss_hf;
1190 /* Checking if RSS is enabled */
1191 if (!(hw->ctrl & NFP_NET_CFG_CTRL_RSS)) {
1192 if (rss_hf != 0) { /* Enable RSS? */
1193 PMD_DRV_LOG(ERR, "RSS unsupported");
1196 return 0; /* Nothing to do */
1199 if (rss_conf->rss_key_len > NFP_NET_CFG_RSS_KEY_SZ) {
1200 PMD_DRV_LOG(ERR, "hash key too long");
1204 nfp_net_rss_hash_write(dev, rss_conf);
1206 update = NFP_NET_CFG_UPDATE_RSS;
1208 if (nfp_net_reconfig(hw, hw->ctrl, update) < 0)
1215 nfp_net_rss_hash_conf_get(struct rte_eth_dev *dev,
1216 struct rte_eth_rss_conf *rss_conf)
1219 uint32_t cfg_rss_ctrl;
1222 struct nfp_net_hw *hw;
1224 hw = NFP_NET_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1226 if (!(hw->ctrl & NFP_NET_CFG_CTRL_RSS))
1229 rss_hf = rss_conf->rss_hf;
1230 cfg_rss_ctrl = nn_cfg_readl(hw, NFP_NET_CFG_RSS_CTRL);
1232 if (cfg_rss_ctrl & NFP_NET_CFG_RSS_IPV4)
1233 rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_TCP | RTE_ETH_RSS_NONFRAG_IPV4_UDP;
1235 if (cfg_rss_ctrl & NFP_NET_CFG_RSS_IPV4_TCP)
1236 rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_TCP;
1238 if (cfg_rss_ctrl & NFP_NET_CFG_RSS_IPV6_TCP)
1239 rss_hf |= RTE_ETH_RSS_NONFRAG_IPV6_TCP;
1241 if (cfg_rss_ctrl & NFP_NET_CFG_RSS_IPV4_UDP)
1242 rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_UDP;
1244 if (cfg_rss_ctrl & NFP_NET_CFG_RSS_IPV6_UDP)
1245 rss_hf |= RTE_ETH_RSS_NONFRAG_IPV6_UDP;
1247 if (cfg_rss_ctrl & NFP_NET_CFG_RSS_IPV6)
1248 rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_UDP | RTE_ETH_RSS_NONFRAG_IPV6_UDP;
1250 /* Propagate current RSS hash functions to caller */
1251 rss_conf->rss_hf = rss_hf;
1253 /* Reading the key size */
1254 rss_conf->rss_key_len = nn_cfg_readl(hw, NFP_NET_CFG_RSS_KEY_SZ);
1256 /* Reading the key byte a byte */
1257 for (i = 0; i < rss_conf->rss_key_len; i++) {
1258 key = nn_cfg_readb(hw, NFP_NET_CFG_RSS_KEY + i);
1259 memcpy(&rss_conf->rss_key[i], &key, 1);
1266 nfp_net_rss_config_default(struct rte_eth_dev *dev)
1268 struct rte_eth_conf *dev_conf;
1269 struct rte_eth_rss_conf rss_conf;
1270 struct rte_eth_rss_reta_entry64 nfp_reta_conf[2];
1271 uint16_t rx_queues = dev->data->nb_rx_queues;
1275 PMD_DRV_LOG(INFO, "setting default RSS conf for %u queues",
1278 nfp_reta_conf[0].mask = ~0x0;
1279 nfp_reta_conf[1].mask = ~0x0;
1282 for (i = 0; i < 0x40; i += 8) {
1283 for (j = i; j < (i + 8); j++) {
1284 nfp_reta_conf[0].reta[j] = queue;
1285 nfp_reta_conf[1].reta[j] = queue++;
1289 ret = nfp_net_rss_reta_write(dev, nfp_reta_conf, 0x80);
1293 dev_conf = &dev->data->dev_conf;
1295 PMD_DRV_LOG(INFO, "wrong rss conf");
1298 rss_conf = dev_conf->rx_adv_conf.rss_conf;
1300 ret = nfp_net_rss_hash_write(dev, &rss_conf);
1305 RTE_LOG_REGISTER_SUFFIX(nfp_logtype_init, init, NOTICE);
1306 RTE_LOG_REGISTER_SUFFIX(nfp_logtype_driver, driver, NOTICE);
1309 * c-file-style: "Linux"
1310 * indent-tabs-mode: t