4 * Copyright(c) 2013-2016 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <rte_ethdev.h>
35 #include <rte_malloc.h>
36 #include <rte_memzone.h>
37 #include <rte_string_fns.h>
39 #include <rte_spinlock.h>
40 #include <rte_kvargs.h>
43 #include "base/fm10k_api.h"
45 /* Default delay to acquire mailbox lock */
46 #define FM10K_MBXLOCK_DELAY_US 20
47 #define UINT64_LOWER_32BITS_MASK 0x00000000ffffffffULL
49 #define MAIN_VSI_POOL_NUMBER 0
51 /* Max try times to acquire switch status */
52 #define MAX_QUERY_SWITCH_STATE_TIMES 10
53 /* Wait interval to get switch status */
54 #define WAIT_SWITCH_MSG_US 100000
55 /* Number of chars per uint32 type */
56 #define CHARS_PER_UINT32 (sizeof(uint32_t))
57 #define BIT_MASK_PER_UINT32 ((1 << CHARS_PER_UINT32) - 1)
59 /* default 1:1 map from queue ID to interrupt vector ID */
60 #define Q2V(dev, queue_id) (dev->pci_dev->intr_handle.intr_vec[queue_id])
62 /* First 64 Logical ports for PF/VMDQ, second 64 for Flow director */
63 #define MAX_LPORT_NUM 128
64 #define GLORT_FD_Q_BASE 0x40
65 #define GLORT_PF_MASK 0xFFC0
66 #define GLORT_FD_MASK GLORT_PF_MASK
67 #define GLORT_FD_INDEX GLORT_FD_Q_BASE
69 static void fm10k_close_mbx_service(struct fm10k_hw *hw);
70 static void fm10k_dev_promiscuous_enable(struct rte_eth_dev *dev);
71 static void fm10k_dev_promiscuous_disable(struct rte_eth_dev *dev);
72 static void fm10k_dev_allmulticast_enable(struct rte_eth_dev *dev);
73 static void fm10k_dev_allmulticast_disable(struct rte_eth_dev *dev);
74 static inline int fm10k_glort_valid(struct fm10k_hw *hw);
76 fm10k_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on);
77 static void fm10k_MAC_filter_set(struct rte_eth_dev *dev,
78 const u8 *mac, bool add, uint32_t pool);
79 static void fm10k_tx_queue_release(void *queue);
80 static void fm10k_rx_queue_release(void *queue);
81 static void fm10k_set_rx_function(struct rte_eth_dev *dev);
82 static void fm10k_set_tx_function(struct rte_eth_dev *dev);
83 static int fm10k_check_ftag(struct rte_devargs *devargs);
85 struct fm10k_xstats_name_off {
86 char name[RTE_ETH_XSTATS_NAME_SIZE];
90 struct fm10k_xstats_name_off fm10k_hw_stats_strings[] = {
91 {"completion_timeout_count", offsetof(struct fm10k_hw_stats, timeout)},
92 {"unsupported_requests_count", offsetof(struct fm10k_hw_stats, ur)},
93 {"completer_abort_count", offsetof(struct fm10k_hw_stats, ca)},
94 {"unsupported_message_count", offsetof(struct fm10k_hw_stats, um)},
95 {"checksum_error_count", offsetof(struct fm10k_hw_stats, xec)},
96 {"vlan_dropped", offsetof(struct fm10k_hw_stats, vlan_drop)},
97 {"loopback_dropped", offsetof(struct fm10k_hw_stats, loopback_drop)},
98 {"rx_mbuf_allocation_errors", offsetof(struct fm10k_hw_stats,
102 #define FM10K_NB_HW_XSTATS (sizeof(fm10k_hw_stats_strings) / \
103 sizeof(fm10k_hw_stats_strings[0]))
105 struct fm10k_xstats_name_off fm10k_hw_stats_rx_q_strings[] = {
106 {"packets", offsetof(struct fm10k_hw_stats_q, rx_packets)},
107 {"bytes", offsetof(struct fm10k_hw_stats_q, rx_bytes)},
108 {"dropped", offsetof(struct fm10k_hw_stats_q, rx_drops)},
111 #define FM10K_NB_RX_Q_XSTATS (sizeof(fm10k_hw_stats_rx_q_strings) / \
112 sizeof(fm10k_hw_stats_rx_q_strings[0]))
114 struct fm10k_xstats_name_off fm10k_hw_stats_tx_q_strings[] = {
115 {"packets", offsetof(struct fm10k_hw_stats_q, tx_packets)},
116 {"bytes", offsetof(struct fm10k_hw_stats_q, tx_bytes)},
119 #define FM10K_NB_TX_Q_XSTATS (sizeof(fm10k_hw_stats_tx_q_strings) / \
120 sizeof(fm10k_hw_stats_tx_q_strings[0]))
122 #define FM10K_NB_XSTATS (FM10K_NB_HW_XSTATS + FM10K_MAX_QUEUES_PF * \
123 (FM10K_NB_RX_Q_XSTATS + FM10K_NB_TX_Q_XSTATS))
125 fm10k_dev_rxq_interrupt_setup(struct rte_eth_dev *dev);
128 fm10k_mbx_initlock(struct fm10k_hw *hw)
130 rte_spinlock_init(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back));
134 fm10k_mbx_lock(struct fm10k_hw *hw)
136 while (!rte_spinlock_trylock(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back)))
137 rte_delay_us(FM10K_MBXLOCK_DELAY_US);
141 fm10k_mbx_unlock(struct fm10k_hw *hw)
143 rte_spinlock_unlock(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back));
146 /* Stubs needed for linkage when vPMD is disabled */
147 int __attribute__((weak))
148 fm10k_rx_vec_condition_check(__rte_unused struct rte_eth_dev *dev)
153 uint16_t __attribute__((weak))
155 __rte_unused void *rx_queue,
156 __rte_unused struct rte_mbuf **rx_pkts,
157 __rte_unused uint16_t nb_pkts)
162 uint16_t __attribute__((weak))
163 fm10k_recv_scattered_pkts_vec(
164 __rte_unused void *rx_queue,
165 __rte_unused struct rte_mbuf **rx_pkts,
166 __rte_unused uint16_t nb_pkts)
171 int __attribute__((weak))
172 fm10k_rxq_vec_setup(__rte_unused struct fm10k_rx_queue *rxq)
178 void __attribute__((weak))
179 fm10k_rx_queue_release_mbufs_vec(
180 __rte_unused struct fm10k_rx_queue *rxq)
185 void __attribute__((weak))
186 fm10k_txq_vec_setup(__rte_unused struct fm10k_tx_queue *txq)
191 int __attribute__((weak))
192 fm10k_tx_vec_condition_check(__rte_unused struct fm10k_tx_queue *txq)
197 uint16_t __attribute__((weak))
198 fm10k_xmit_pkts_vec(__rte_unused void *tx_queue,
199 __rte_unused struct rte_mbuf **tx_pkts,
200 __rte_unused uint16_t nb_pkts)
206 * reset queue to initial state, allocate software buffers used when starting
208 * return 0 on success
209 * return -ENOMEM if buffers cannot be allocated
210 * return -EINVAL if buffers do not satisfy alignment condition
213 rx_queue_reset(struct fm10k_rx_queue *q)
215 static const union fm10k_rx_desc zero = {{0} };
218 PMD_INIT_FUNC_TRACE();
220 diag = rte_mempool_get_bulk(q->mp, (void **)q->sw_ring, q->nb_desc);
224 for (i = 0; i < q->nb_desc; ++i) {
225 fm10k_pktmbuf_reset(q->sw_ring[i], q->port_id);
226 if (!fm10k_addr_alignment_valid(q->sw_ring[i])) {
227 rte_mempool_put_bulk(q->mp, (void **)q->sw_ring,
231 dma_addr = MBUF_DMA_ADDR_DEFAULT(q->sw_ring[i]);
232 q->hw_ring[i].q.pkt_addr = dma_addr;
233 q->hw_ring[i].q.hdr_addr = dma_addr;
236 /* initialize extra software ring entries. Space for these extra
237 * entries is always allocated.
239 memset(&q->fake_mbuf, 0x0, sizeof(q->fake_mbuf));
240 for (i = 0; i < q->nb_fake_desc; ++i) {
241 q->sw_ring[q->nb_desc + i] = &q->fake_mbuf;
242 q->hw_ring[q->nb_desc + i] = zero;
247 q->next_trigger = q->alloc_thresh - 1;
248 FM10K_PCI_REG_WRITE(q->tail_ptr, q->nb_desc - 1);
249 q->rxrearm_start = 0;
256 * clean queue, descriptor rings, free software buffers used when stopping
260 rx_queue_clean(struct fm10k_rx_queue *q)
262 union fm10k_rx_desc zero = {.q = {0, 0, 0, 0} };
264 PMD_INIT_FUNC_TRACE();
266 /* zero descriptor rings */
267 for (i = 0; i < q->nb_desc; ++i)
268 q->hw_ring[i] = zero;
270 /* zero faked descriptors */
271 for (i = 0; i < q->nb_fake_desc; ++i)
272 q->hw_ring[q->nb_desc + i] = zero;
274 /* vPMD driver has a different way of releasing mbufs. */
275 if (q->rx_using_sse) {
276 fm10k_rx_queue_release_mbufs_vec(q);
280 /* free software buffers */
281 for (i = 0; i < q->nb_desc; ++i) {
283 rte_pktmbuf_free_seg(q->sw_ring[i]);
284 q->sw_ring[i] = NULL;
290 * free all queue memory used when releasing the queue (i.e. configure)
293 rx_queue_free(struct fm10k_rx_queue *q)
295 PMD_INIT_FUNC_TRACE();
297 PMD_INIT_LOG(DEBUG, "Freeing rx queue %p", q);
300 rte_free(q->sw_ring);
309 * disable RX queue, wait unitl HW finished necessary flush operation
312 rx_queue_disable(struct fm10k_hw *hw, uint16_t qnum)
316 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(qnum));
317 FM10K_WRITE_REG(hw, FM10K_RXQCTL(qnum),
318 reg & ~FM10K_RXQCTL_ENABLE);
320 /* Wait 100us at most */
321 for (i = 0; i < FM10K_QUEUE_DISABLE_TIMEOUT; i++) {
323 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(qnum));
324 if (!(reg & FM10K_RXQCTL_ENABLE))
328 if (i == FM10K_QUEUE_DISABLE_TIMEOUT)
335 * reset queue to initial state, allocate software buffers used when starting
339 tx_queue_reset(struct fm10k_tx_queue *q)
341 PMD_INIT_FUNC_TRACE();
345 q->nb_free = q->nb_desc - 1;
346 fifo_reset(&q->rs_tracker, (q->nb_desc + 1) / q->rs_thresh);
347 FM10K_PCI_REG_WRITE(q->tail_ptr, 0);
351 * clean queue, descriptor rings, free software buffers used when stopping
355 tx_queue_clean(struct fm10k_tx_queue *q)
357 struct fm10k_tx_desc zero = {0, 0, 0, 0, 0, 0};
359 PMD_INIT_FUNC_TRACE();
361 /* zero descriptor rings */
362 for (i = 0; i < q->nb_desc; ++i)
363 q->hw_ring[i] = zero;
365 /* free software buffers */
366 for (i = 0; i < q->nb_desc; ++i) {
368 rte_pktmbuf_free_seg(q->sw_ring[i]);
369 q->sw_ring[i] = NULL;
375 * free all queue memory used when releasing the queue (i.e. configure)
378 tx_queue_free(struct fm10k_tx_queue *q)
380 PMD_INIT_FUNC_TRACE();
382 PMD_INIT_LOG(DEBUG, "Freeing tx queue %p", q);
384 if (q->rs_tracker.list) {
385 rte_free(q->rs_tracker.list);
386 q->rs_tracker.list = NULL;
389 rte_free(q->sw_ring);
398 * disable TX queue, wait unitl HW finished necessary flush operation
401 tx_queue_disable(struct fm10k_hw *hw, uint16_t qnum)
405 reg = FM10K_READ_REG(hw, FM10K_TXDCTL(qnum));
406 FM10K_WRITE_REG(hw, FM10K_TXDCTL(qnum),
407 reg & ~FM10K_TXDCTL_ENABLE);
409 /* Wait 100us at most */
410 for (i = 0; i < FM10K_QUEUE_DISABLE_TIMEOUT; i++) {
412 reg = FM10K_READ_REG(hw, FM10K_TXDCTL(qnum));
413 if (!(reg & FM10K_TXDCTL_ENABLE))
417 if (i == FM10K_QUEUE_DISABLE_TIMEOUT)
424 fm10k_check_mq_mode(struct rte_eth_dev *dev)
426 enum rte_eth_rx_mq_mode rx_mq_mode = dev->data->dev_conf.rxmode.mq_mode;
427 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
428 struct rte_eth_vmdq_rx_conf *vmdq_conf;
429 uint16_t nb_rx_q = dev->data->nb_rx_queues;
431 vmdq_conf = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
433 if (rx_mq_mode & ETH_MQ_RX_DCB_FLAG) {
434 PMD_INIT_LOG(ERR, "DCB mode is not supported.");
438 if (!(rx_mq_mode & ETH_MQ_RX_VMDQ_FLAG))
441 if (hw->mac.type == fm10k_mac_vf) {
442 PMD_INIT_LOG(ERR, "VMDQ mode is not supported in VF.");
446 /* Check VMDQ queue pool number */
447 if (vmdq_conf->nb_queue_pools >
448 sizeof(vmdq_conf->pool_map[0].pools) * CHAR_BIT ||
449 vmdq_conf->nb_queue_pools > nb_rx_q) {
450 PMD_INIT_LOG(ERR, "Too many of queue pools: %d",
451 vmdq_conf->nb_queue_pools);
458 static const struct fm10k_txq_ops def_txq_ops = {
459 .reset = tx_queue_reset,
463 fm10k_dev_configure(struct rte_eth_dev *dev)
467 PMD_INIT_FUNC_TRACE();
469 if (dev->data->dev_conf.rxmode.hw_strip_crc == 0)
470 PMD_INIT_LOG(WARNING, "fm10k always strip CRC");
471 /* multipe queue mode checking */
472 ret = fm10k_check_mq_mode(dev);
474 PMD_DRV_LOG(ERR, "fm10k_check_mq_mode fails with %d.",
482 /* fls = find last set bit = 32 minus the number of leading zeros */
484 #define fls(x) (((x) == 0) ? 0 : (32 - __builtin_clz((x))))
488 fm10k_dev_vmdq_rx_configure(struct rte_eth_dev *dev)
490 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
491 struct rte_eth_vmdq_rx_conf *vmdq_conf;
494 vmdq_conf = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
496 for (i = 0; i < vmdq_conf->nb_pool_maps; i++) {
497 if (!vmdq_conf->pool_map[i].pools)
500 fm10k_update_vlan(hw, vmdq_conf->pool_map[i].vlan_id, 0, true);
501 fm10k_mbx_unlock(hw);
506 fm10k_dev_pf_main_vsi_reset(struct rte_eth_dev *dev)
508 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
510 /* Add default mac address */
511 fm10k_MAC_filter_set(dev, hw->mac.addr, true,
512 MAIN_VSI_POOL_NUMBER);
516 fm10k_dev_rss_configure(struct rte_eth_dev *dev)
518 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
519 struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
520 uint32_t mrqc, *key, i, reta, j;
523 #define RSS_KEY_SIZE 40
524 static uint8_t rss_intel_key[RSS_KEY_SIZE] = {
525 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
526 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
527 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
528 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
529 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
532 if (dev->data->nb_rx_queues == 1 ||
533 dev_conf->rxmode.mq_mode != ETH_MQ_RX_RSS ||
534 dev_conf->rx_adv_conf.rss_conf.rss_hf == 0) {
535 FM10K_WRITE_REG(hw, FM10K_MRQC(0), 0);
539 /* random key is rss_intel_key (default) or user provided (rss_key) */
540 if (dev_conf->rx_adv_conf.rss_conf.rss_key == NULL)
541 key = (uint32_t *)rss_intel_key;
543 key = (uint32_t *)dev_conf->rx_adv_conf.rss_conf.rss_key;
545 /* Now fill our hash function seeds, 4 bytes at a time */
546 for (i = 0; i < RSS_KEY_SIZE / sizeof(*key); ++i)
547 FM10K_WRITE_REG(hw, FM10K_RSSRK(0, i), key[i]);
550 * Fill in redirection table
551 * The byte-swap is needed because NIC registers are in
552 * little-endian order.
555 for (i = 0, j = 0; i < FM10K_MAX_RSS_INDICES; i++, j++) {
556 if (j == dev->data->nb_rx_queues)
558 reta = (reta << CHAR_BIT) | j;
560 FM10K_WRITE_REG(hw, FM10K_RETA(0, i >> 2),
565 * Generate RSS hash based on packet types, TCP/UDP
566 * port numbers and/or IPv4/v6 src and dst addresses
568 hf = dev_conf->rx_adv_conf.rss_conf.rss_hf;
570 mrqc |= (hf & ETH_RSS_IPV4) ? FM10K_MRQC_IPV4 : 0;
571 mrqc |= (hf & ETH_RSS_IPV6) ? FM10K_MRQC_IPV6 : 0;
572 mrqc |= (hf & ETH_RSS_IPV6_EX) ? FM10K_MRQC_IPV6 : 0;
573 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_TCP) ? FM10K_MRQC_TCP_IPV4 : 0;
574 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_TCP) ? FM10K_MRQC_TCP_IPV6 : 0;
575 mrqc |= (hf & ETH_RSS_IPV6_TCP_EX) ? FM10K_MRQC_TCP_IPV6 : 0;
576 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_UDP) ? FM10K_MRQC_UDP_IPV4 : 0;
577 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_UDP) ? FM10K_MRQC_UDP_IPV6 : 0;
578 mrqc |= (hf & ETH_RSS_IPV6_UDP_EX) ? FM10K_MRQC_UDP_IPV6 : 0;
581 PMD_INIT_LOG(ERR, "Specified RSS mode 0x%"PRIx64"is not"
586 FM10K_WRITE_REG(hw, FM10K_MRQC(0), mrqc);
590 fm10k_dev_logic_port_update(struct rte_eth_dev *dev, uint16_t nb_lport_new)
592 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
595 for (i = 0; i < nb_lport_new; i++) {
596 /* Set unicast mode by default. App can change
597 * to other mode in other API func.
600 hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map + i,
601 FM10K_XCAST_MODE_NONE);
602 fm10k_mbx_unlock(hw);
607 fm10k_dev_mq_rx_configure(struct rte_eth_dev *dev)
609 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
610 struct rte_eth_vmdq_rx_conf *vmdq_conf;
611 struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
612 struct fm10k_macvlan_filter_info *macvlan;
613 uint16_t nb_queue_pools = 0; /* pool number in configuration */
614 uint16_t nb_lport_new;
616 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
617 vmdq_conf = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
619 fm10k_dev_rss_configure(dev);
621 /* only PF supports VMDQ */
622 if (hw->mac.type != fm10k_mac_pf)
625 if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_VMDQ_FLAG)
626 nb_queue_pools = vmdq_conf->nb_queue_pools;
628 /* no pool number change, no need to update logic port and VLAN/MAC */
629 if (macvlan->nb_queue_pools == nb_queue_pools)
632 nb_lport_new = nb_queue_pools ? nb_queue_pools : 1;
633 fm10k_dev_logic_port_update(dev, nb_lport_new);
635 /* reset MAC/VLAN as it's based on VMDQ or PF main VSI */
636 memset(dev->data->mac_addrs, 0,
637 ETHER_ADDR_LEN * FM10K_MAX_MACADDR_NUM);
638 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
639 &dev->data->mac_addrs[0]);
640 memset(macvlan, 0, sizeof(*macvlan));
641 macvlan->nb_queue_pools = nb_queue_pools;
644 fm10k_dev_vmdq_rx_configure(dev);
646 fm10k_dev_pf_main_vsi_reset(dev);
650 fm10k_dev_tx_init(struct rte_eth_dev *dev)
652 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
654 struct fm10k_tx_queue *txq;
658 /* Disable TXINT to avoid possible interrupt */
659 for (i = 0; i < hw->mac.max_queues; i++)
660 FM10K_WRITE_REG(hw, FM10K_TXINT(i),
661 3 << FM10K_TXINT_TIMER_SHIFT);
664 for (i = 0; i < dev->data->nb_tx_queues; ++i) {
665 txq = dev->data->tx_queues[i];
666 base_addr = txq->hw_ring_phys_addr;
667 size = txq->nb_desc * sizeof(struct fm10k_tx_desc);
669 /* disable queue to avoid issues while updating state */
670 ret = tx_queue_disable(hw, i);
672 PMD_INIT_LOG(ERR, "failed to disable queue %d", i);
675 /* Enable use of FTAG bit in TX descriptor, PFVTCTL
676 * register is read-only for VF.
678 if (fm10k_check_ftag(dev->pci_dev->devargs)) {
679 if (hw->mac.type == fm10k_mac_pf) {
680 FM10K_WRITE_REG(hw, FM10K_PFVTCTL(i),
681 FM10K_PFVTCTL_FTAG_DESC_ENABLE);
682 PMD_INIT_LOG(DEBUG, "FTAG mode is enabled");
684 PMD_INIT_LOG(ERR, "VF FTAG is not supported.");
689 /* set location and size for descriptor ring */
690 FM10K_WRITE_REG(hw, FM10K_TDBAL(i),
691 base_addr & UINT64_LOWER_32BITS_MASK);
692 FM10K_WRITE_REG(hw, FM10K_TDBAH(i),
693 base_addr >> (CHAR_BIT * sizeof(uint32_t)));
694 FM10K_WRITE_REG(hw, FM10K_TDLEN(i), size);
696 /* assign default SGLORT for each TX queue */
697 FM10K_WRITE_REG(hw, FM10K_TX_SGLORT(i), hw->mac.dglort_map);
700 /* set up vector or scalar TX function as appropriate */
701 fm10k_set_tx_function(dev);
707 fm10k_dev_rx_init(struct rte_eth_dev *dev)
709 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
710 struct fm10k_macvlan_filter_info *macvlan;
711 struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
713 struct fm10k_rx_queue *rxq;
716 uint32_t rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY;
717 uint32_t logic_port = hw->mac.dglort_map;
719 uint16_t queue_stride = 0;
721 /* enable RXINT for interrupt mode */
723 if (rte_intr_dp_is_en(intr_handle)) {
724 for (; i < dev->data->nb_rx_queues; i++) {
725 FM10K_WRITE_REG(hw, FM10K_RXINT(i), Q2V(dev, i));
726 if (hw->mac.type == fm10k_mac_pf)
727 FM10K_WRITE_REG(hw, FM10K_ITR(Q2V(dev, i)),
729 FM10K_ITR_MASK_CLEAR);
731 FM10K_WRITE_REG(hw, FM10K_VFITR(Q2V(dev, i)),
733 FM10K_ITR_MASK_CLEAR);
736 /* Disable other RXINT to avoid possible interrupt */
737 for (; i < hw->mac.max_queues; i++)
738 FM10K_WRITE_REG(hw, FM10K_RXINT(i),
739 3 << FM10K_RXINT_TIMER_SHIFT);
741 /* Setup RX queues */
742 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
743 rxq = dev->data->rx_queues[i];
744 base_addr = rxq->hw_ring_phys_addr;
745 size = rxq->nb_desc * sizeof(union fm10k_rx_desc);
747 /* disable queue to avoid issues while updating state */
748 ret = rx_queue_disable(hw, i);
750 PMD_INIT_LOG(ERR, "failed to disable queue %d", i);
754 /* Setup the Base and Length of the Rx Descriptor Ring */
755 FM10K_WRITE_REG(hw, FM10K_RDBAL(i),
756 base_addr & UINT64_LOWER_32BITS_MASK);
757 FM10K_WRITE_REG(hw, FM10K_RDBAH(i),
758 base_addr >> (CHAR_BIT * sizeof(uint32_t)));
759 FM10K_WRITE_REG(hw, FM10K_RDLEN(i), size);
761 /* Configure the Rx buffer size for one buff without split */
762 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
763 RTE_PKTMBUF_HEADROOM);
764 /* As RX buffer is aligned to 512B within mbuf, some bytes are
765 * reserved for this purpose, and the worst case could be 511B.
766 * But SRR reg assumes all buffers have the same size. In order
767 * to fill the gap, we'll have to consider the worst case and
768 * assume 512B is reserved. If we don't do so, it's possible
769 * for HW to overwrite data to next mbuf.
771 buf_size -= FM10K_RX_DATABUF_ALIGN;
773 FM10K_WRITE_REG(hw, FM10K_SRRCTL(i),
774 (buf_size >> FM10K_SRRCTL_BSIZEPKT_SHIFT) |
775 FM10K_SRRCTL_LOOPBACK_SUPPRESS);
777 /* It adds dual VLAN length for supporting dual VLAN */
778 if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
779 2 * FM10K_VLAN_TAG_SIZE) > buf_size ||
780 dev->data->dev_conf.rxmode.enable_scatter) {
782 dev->data->scattered_rx = 1;
783 reg = FM10K_READ_REG(hw, FM10K_SRRCTL(i));
784 reg |= FM10K_SRRCTL_BUFFER_CHAINING_EN;
785 FM10K_WRITE_REG(hw, FM10K_SRRCTL(i), reg);
788 /* Enable drop on empty, it's RO for VF */
789 if (hw->mac.type == fm10k_mac_pf && rxq->drop_en)
790 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY;
792 FM10K_WRITE_REG(hw, FM10K_RXDCTL(i), rxdctl);
793 FM10K_WRITE_FLUSH(hw);
796 /* Configure VMDQ/RSS if applicable */
797 fm10k_dev_mq_rx_configure(dev);
799 /* Decide the best RX function */
800 fm10k_set_rx_function(dev);
802 /* update RX_SGLORT for loopback suppress*/
803 if (hw->mac.type != fm10k_mac_pf)
805 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
806 if (macvlan->nb_queue_pools)
807 queue_stride = dev->data->nb_rx_queues / macvlan->nb_queue_pools;
808 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
809 if (i && queue_stride && !(i % queue_stride))
811 FM10K_WRITE_REG(hw, FM10K_RX_SGLORT(i), logic_port);
818 fm10k_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
820 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
823 struct fm10k_rx_queue *rxq;
825 PMD_INIT_FUNC_TRACE();
827 if (rx_queue_id < dev->data->nb_rx_queues) {
828 rxq = dev->data->rx_queues[rx_queue_id];
829 err = rx_queue_reset(rxq);
830 if (err == -ENOMEM) {
831 PMD_INIT_LOG(ERR, "Failed to alloc memory : %d", err);
833 } else if (err == -EINVAL) {
834 PMD_INIT_LOG(ERR, "Invalid buffer address alignment :"
839 /* Setup the HW Rx Head and Tail Descriptor Pointers
840 * Note: this must be done AFTER the queue is enabled on real
841 * hardware, but BEFORE the queue is enabled when using the
842 * emulation platform. Do it in both places for now and remove
843 * this comment and the following two register writes when the
844 * emulation platform is no longer being used.
846 FM10K_WRITE_REG(hw, FM10K_RDH(rx_queue_id), 0);
847 FM10K_WRITE_REG(hw, FM10K_RDT(rx_queue_id), rxq->nb_desc - 1);
849 /* Set PF ownership flag for PF devices */
850 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(rx_queue_id));
851 if (hw->mac.type == fm10k_mac_pf)
852 reg |= FM10K_RXQCTL_PF;
853 reg |= FM10K_RXQCTL_ENABLE;
854 /* enable RX queue */
855 FM10K_WRITE_REG(hw, FM10K_RXQCTL(rx_queue_id), reg);
856 FM10K_WRITE_FLUSH(hw);
858 /* Setup the HW Rx Head and Tail Descriptor Pointers
859 * Note: this must be done AFTER the queue is enabled
861 FM10K_WRITE_REG(hw, FM10K_RDH(rx_queue_id), 0);
862 FM10K_WRITE_REG(hw, FM10K_RDT(rx_queue_id), rxq->nb_desc - 1);
863 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
870 fm10k_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
872 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
874 PMD_INIT_FUNC_TRACE();
876 if (rx_queue_id < dev->data->nb_rx_queues) {
877 /* Disable RX queue */
878 rx_queue_disable(hw, rx_queue_id);
880 /* Free mbuf and clean HW ring */
881 rx_queue_clean(dev->data->rx_queues[rx_queue_id]);
882 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
889 fm10k_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
891 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
892 /** @todo - this should be defined in the shared code */
893 #define FM10K_TXDCTL_WRITE_BACK_MIN_DELAY 0x00010000
894 uint32_t txdctl = FM10K_TXDCTL_WRITE_BACK_MIN_DELAY;
897 PMD_INIT_FUNC_TRACE();
899 if (tx_queue_id < dev->data->nb_tx_queues) {
900 struct fm10k_tx_queue *q = dev->data->tx_queues[tx_queue_id];
904 /* reset head and tail pointers */
905 FM10K_WRITE_REG(hw, FM10K_TDH(tx_queue_id), 0);
906 FM10K_WRITE_REG(hw, FM10K_TDT(tx_queue_id), 0);
908 /* enable TX queue */
909 FM10K_WRITE_REG(hw, FM10K_TXDCTL(tx_queue_id),
910 FM10K_TXDCTL_ENABLE | txdctl);
911 FM10K_WRITE_FLUSH(hw);
912 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
920 fm10k_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
922 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
924 PMD_INIT_FUNC_TRACE();
926 if (tx_queue_id < dev->data->nb_tx_queues) {
927 tx_queue_disable(hw, tx_queue_id);
928 tx_queue_clean(dev->data->tx_queues[tx_queue_id]);
929 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
935 static inline int fm10k_glort_valid(struct fm10k_hw *hw)
937 return ((hw->mac.dglort_map & FM10K_DGLORTMAP_NONE)
938 != FM10K_DGLORTMAP_NONE);
942 fm10k_dev_promiscuous_enable(struct rte_eth_dev *dev)
944 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
947 PMD_INIT_FUNC_TRACE();
949 /* Return if it didn't acquire valid glort range */
950 if (!fm10k_glort_valid(hw))
954 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
955 FM10K_XCAST_MODE_PROMISC);
956 fm10k_mbx_unlock(hw);
958 if (status != FM10K_SUCCESS)
959 PMD_INIT_LOG(ERR, "Failed to enable promiscuous mode");
963 fm10k_dev_promiscuous_disable(struct rte_eth_dev *dev)
965 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
969 PMD_INIT_FUNC_TRACE();
971 /* Return if it didn't acquire valid glort range */
972 if (!fm10k_glort_valid(hw))
975 if (dev->data->all_multicast == 1)
976 mode = FM10K_XCAST_MODE_ALLMULTI;
978 mode = FM10K_XCAST_MODE_NONE;
981 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
983 fm10k_mbx_unlock(hw);
985 if (status != FM10K_SUCCESS)
986 PMD_INIT_LOG(ERR, "Failed to disable promiscuous mode");
990 fm10k_dev_allmulticast_enable(struct rte_eth_dev *dev)
992 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
995 PMD_INIT_FUNC_TRACE();
997 /* Return if it didn't acquire valid glort range */
998 if (!fm10k_glort_valid(hw))
1001 /* If promiscuous mode is enabled, it doesn't make sense to enable
1002 * allmulticast and disable promiscuous since fm10k only can select
1005 if (dev->data->promiscuous) {
1006 PMD_INIT_LOG(INFO, "Promiscuous mode is enabled, "\
1007 "needn't enable allmulticast");
1012 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
1013 FM10K_XCAST_MODE_ALLMULTI);
1014 fm10k_mbx_unlock(hw);
1016 if (status != FM10K_SUCCESS)
1017 PMD_INIT_LOG(ERR, "Failed to enable allmulticast mode");
1021 fm10k_dev_allmulticast_disable(struct rte_eth_dev *dev)
1023 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1026 PMD_INIT_FUNC_TRACE();
1028 /* Return if it didn't acquire valid glort range */
1029 if (!fm10k_glort_valid(hw))
1032 if (dev->data->promiscuous) {
1033 PMD_INIT_LOG(ERR, "Failed to disable allmulticast mode "\
1034 "since promisc mode is enabled");
1039 /* Change mode to unicast mode */
1040 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
1041 FM10K_XCAST_MODE_NONE);
1042 fm10k_mbx_unlock(hw);
1044 if (status != FM10K_SUCCESS)
1045 PMD_INIT_LOG(ERR, "Failed to disable allmulticast mode");
1049 fm10k_dev_dglort_map_configure(struct rte_eth_dev *dev)
1051 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1052 uint32_t dglortdec, pool_len, rss_len, i, dglortmask;
1053 uint16_t nb_queue_pools;
1054 struct fm10k_macvlan_filter_info *macvlan;
1056 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1057 nb_queue_pools = macvlan->nb_queue_pools;
1058 pool_len = nb_queue_pools ? fls(nb_queue_pools - 1) : 0;
1059 rss_len = fls(dev->data->nb_rx_queues - 1) - pool_len;
1061 /* GLORT 0x0-0x3F are used by PF and VMDQ, 0x40-0x7F used by FD */
1062 dglortdec = (rss_len << FM10K_DGLORTDEC_RSSLENGTH_SHIFT) | pool_len;
1063 dglortmask = (GLORT_PF_MASK << FM10K_DGLORTMAP_MASK_SHIFT) |
1065 FM10K_WRITE_REG(hw, FM10K_DGLORTMAP(0), dglortmask);
1066 /* Configure VMDQ/RSS DGlort Decoder */
1067 FM10K_WRITE_REG(hw, FM10K_DGLORTDEC(0), dglortdec);
1069 /* Flow Director configurations, only queue number is valid. */
1070 dglortdec = fls(dev->data->nb_rx_queues - 1);
1071 dglortmask = (GLORT_FD_MASK << FM10K_DGLORTMAP_MASK_SHIFT) |
1072 (hw->mac.dglort_map + GLORT_FD_Q_BASE);
1073 FM10K_WRITE_REG(hw, FM10K_DGLORTMAP(1), dglortmask);
1074 FM10K_WRITE_REG(hw, FM10K_DGLORTDEC(1), dglortdec);
1076 /* Invalidate all other GLORT entries */
1077 for (i = 2; i < FM10K_DGLORT_COUNT; i++)
1078 FM10K_WRITE_REG(hw, FM10K_DGLORTMAP(i),
1079 FM10K_DGLORTMAP_NONE);
1082 #define BSIZEPKT_ROUNDUP ((1 << FM10K_SRRCTL_BSIZEPKT_SHIFT) - 1)
1084 fm10k_dev_start(struct rte_eth_dev *dev)
1086 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1089 PMD_INIT_FUNC_TRACE();
1091 /* stop, init, then start the hw */
1092 diag = fm10k_stop_hw(hw);
1093 if (diag != FM10K_SUCCESS) {
1094 PMD_INIT_LOG(ERR, "Hardware stop failed: %d", diag);
1098 diag = fm10k_init_hw(hw);
1099 if (diag != FM10K_SUCCESS) {
1100 PMD_INIT_LOG(ERR, "Hardware init failed: %d", diag);
1104 diag = fm10k_start_hw(hw);
1105 if (diag != FM10K_SUCCESS) {
1106 PMD_INIT_LOG(ERR, "Hardware start failed: %d", diag);
1110 diag = fm10k_dev_tx_init(dev);
1112 PMD_INIT_LOG(ERR, "TX init failed: %d", diag);
1116 if (fm10k_dev_rxq_interrupt_setup(dev))
1119 diag = fm10k_dev_rx_init(dev);
1121 PMD_INIT_LOG(ERR, "RX init failed: %d", diag);
1125 if (hw->mac.type == fm10k_mac_pf)
1126 fm10k_dev_dglort_map_configure(dev);
1128 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1129 struct fm10k_rx_queue *rxq;
1130 rxq = dev->data->rx_queues[i];
1132 if (rxq->rx_deferred_start)
1134 diag = fm10k_dev_rx_queue_start(dev, i);
1137 for (j = 0; j < i; ++j)
1138 rx_queue_clean(dev->data->rx_queues[j]);
1143 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1144 struct fm10k_tx_queue *txq;
1145 txq = dev->data->tx_queues[i];
1147 if (txq->tx_deferred_start)
1149 diag = fm10k_dev_tx_queue_start(dev, i);
1152 for (j = 0; j < i; ++j)
1153 tx_queue_clean(dev->data->tx_queues[j]);
1154 for (j = 0; j < dev->data->nb_rx_queues; ++j)
1155 rx_queue_clean(dev->data->rx_queues[j]);
1160 /* Update default vlan when not in VMDQ mode */
1161 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_VMDQ_FLAG))
1162 fm10k_vlan_filter_set(dev, hw->mac.default_vid, true);
1168 fm10k_dev_stop(struct rte_eth_dev *dev)
1170 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1171 struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
1174 PMD_INIT_FUNC_TRACE();
1176 if (dev->data->tx_queues)
1177 for (i = 0; i < dev->data->nb_tx_queues; i++)
1178 fm10k_dev_tx_queue_stop(dev, i);
1180 if (dev->data->rx_queues)
1181 for (i = 0; i < dev->data->nb_rx_queues; i++)
1182 fm10k_dev_rx_queue_stop(dev, i);
1184 /* Disable datapath event */
1185 if (rte_intr_dp_is_en(intr_handle)) {
1186 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1187 FM10K_WRITE_REG(hw, FM10K_RXINT(i),
1188 3 << FM10K_RXINT_TIMER_SHIFT);
1189 if (hw->mac.type == fm10k_mac_pf)
1190 FM10K_WRITE_REG(hw, FM10K_ITR(Q2V(dev, i)),
1191 FM10K_ITR_MASK_SET);
1193 FM10K_WRITE_REG(hw, FM10K_VFITR(Q2V(dev, i)),
1194 FM10K_ITR_MASK_SET);
1197 /* Clean datapath event and queue/vec mapping */
1198 rte_intr_efd_disable(intr_handle);
1199 rte_free(intr_handle->intr_vec);
1200 intr_handle->intr_vec = NULL;
1204 fm10k_dev_queue_release(struct rte_eth_dev *dev)
1208 PMD_INIT_FUNC_TRACE();
1210 if (dev->data->tx_queues) {
1211 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1212 struct fm10k_tx_queue *txq = dev->data->tx_queues[i];
1218 if (dev->data->rx_queues) {
1219 for (i = 0; i < dev->data->nb_rx_queues; i++)
1220 fm10k_rx_queue_release(dev->data->rx_queues[i]);
1225 fm10k_dev_close(struct rte_eth_dev *dev)
1227 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1229 PMD_INIT_FUNC_TRACE();
1232 hw->mac.ops.update_lport_state(hw, hw->mac.dglort_map,
1233 MAX_LPORT_NUM, false);
1234 fm10k_mbx_unlock(hw);
1236 /* Stop mailbox service first */
1237 fm10k_close_mbx_service(hw);
1238 fm10k_dev_stop(dev);
1239 fm10k_dev_queue_release(dev);
1244 fm10k_link_update(struct rte_eth_dev *dev,
1245 __rte_unused int wait_to_complete)
1247 PMD_INIT_FUNC_TRACE();
1249 /* The host-interface link is always up. The speed is ~50Gbps per Gen3
1250 * x8 PCIe interface. For now, we leave the speed undefined since there
1251 * is no 50Gbps Ethernet. */
1252 dev->data->dev_link.link_speed = 0;
1253 dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
1254 dev->data->dev_link.link_status = ETH_LINK_UP;
1259 static int fm10k_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
1260 struct rte_eth_xstat_name *xstats_names, __rte_unused unsigned limit)
1265 if (xstats_names != NULL) {
1266 /* Note: limit checked in rte_eth_xstats_names() */
1269 for (i = 0; i < FM10K_NB_HW_XSTATS; i++) {
1270 snprintf(xstats_names[count].name,
1271 sizeof(xstats_names[count].name),
1272 "%s", fm10k_hw_stats_strings[count].name);
1273 xstats_names[count].id = count;
1277 /* PF queue stats */
1278 for (q = 0; q < FM10K_MAX_QUEUES_PF; q++) {
1279 for (i = 0; i < FM10K_NB_RX_Q_XSTATS; i++) {
1280 snprintf(xstats_names[count].name,
1281 sizeof(xstats_names[count].name),
1283 fm10k_hw_stats_rx_q_strings[i].name);
1284 xstats_names[count].id = count;
1287 for (i = 0; i < FM10K_NB_TX_Q_XSTATS; i++) {
1288 snprintf(xstats_names[count].name,
1289 sizeof(xstats_names[count].name),
1291 fm10k_hw_stats_tx_q_strings[i].name);
1292 xstats_names[count].id = count;
1297 return FM10K_NB_XSTATS;
1301 fm10k_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstats *xstats,
1304 struct fm10k_hw_stats *hw_stats =
1305 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1306 unsigned i, q, count = 0;
1308 if (n < FM10K_NB_XSTATS)
1309 return FM10K_NB_XSTATS;
1312 for (i = 0; i < FM10K_NB_HW_XSTATS; i++) {
1313 xstats[count].name[0] = '\0';
1314 xstats[count].value = *(uint64_t *)(((char *)hw_stats) +
1315 fm10k_hw_stats_strings[count].offset);
1319 /* PF queue stats */
1320 for (q = 0; q < FM10K_MAX_QUEUES_PF; q++) {
1321 for (i = 0; i < FM10K_NB_RX_Q_XSTATS; i++) {
1322 xstats[count].name[0] = '\0';
1323 xstats[count].value =
1324 *(uint64_t *)(((char *)&hw_stats->q[q]) +
1325 fm10k_hw_stats_rx_q_strings[i].offset);
1328 for (i = 0; i < FM10K_NB_TX_Q_XSTATS; i++) {
1329 xstats[count].name[0] = '\0';
1330 xstats[count].value =
1331 *(uint64_t *)(((char *)&hw_stats->q[q]) +
1332 fm10k_hw_stats_tx_q_strings[i].offset);
1337 return FM10K_NB_XSTATS;
1341 fm10k_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1343 uint64_t ipackets, opackets, ibytes, obytes;
1344 struct fm10k_hw *hw =
1345 FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1346 struct fm10k_hw_stats *hw_stats =
1347 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1350 PMD_INIT_FUNC_TRACE();
1352 fm10k_update_hw_stats(hw, hw_stats);
1354 ipackets = opackets = ibytes = obytes = 0;
1355 for (i = 0; (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) &&
1356 (i < hw->mac.max_queues); ++i) {
1357 stats->q_ipackets[i] = hw_stats->q[i].rx_packets.count;
1358 stats->q_opackets[i] = hw_stats->q[i].tx_packets.count;
1359 stats->q_ibytes[i] = hw_stats->q[i].rx_bytes.count;
1360 stats->q_obytes[i] = hw_stats->q[i].tx_bytes.count;
1361 ipackets += stats->q_ipackets[i];
1362 opackets += stats->q_opackets[i];
1363 ibytes += stats->q_ibytes[i];
1364 obytes += stats->q_obytes[i];
1366 stats->ipackets = ipackets;
1367 stats->opackets = opackets;
1368 stats->ibytes = ibytes;
1369 stats->obytes = obytes;
1373 fm10k_stats_reset(struct rte_eth_dev *dev)
1375 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1376 struct fm10k_hw_stats *hw_stats =
1377 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1379 PMD_INIT_FUNC_TRACE();
1381 memset(hw_stats, 0, sizeof(*hw_stats));
1382 fm10k_rebind_hw_stats(hw, hw_stats);
1386 fm10k_dev_infos_get(struct rte_eth_dev *dev,
1387 struct rte_eth_dev_info *dev_info)
1389 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1391 PMD_INIT_FUNC_TRACE();
1393 dev_info->min_rx_bufsize = FM10K_MIN_RX_BUF_SIZE;
1394 dev_info->max_rx_pktlen = FM10K_MAX_PKT_SIZE;
1395 dev_info->max_rx_queues = hw->mac.max_queues;
1396 dev_info->max_tx_queues = hw->mac.max_queues;
1397 dev_info->max_mac_addrs = FM10K_MAX_MACADDR_NUM;
1398 dev_info->max_hash_mac_addrs = 0;
1399 dev_info->max_vfs = dev->pci_dev->max_vfs;
1400 dev_info->vmdq_pool_base = 0;
1401 dev_info->vmdq_queue_base = 0;
1402 dev_info->max_vmdq_pools = ETH_32_POOLS;
1403 dev_info->vmdq_queue_num = FM10K_MAX_QUEUES_PF;
1404 dev_info->rx_offload_capa =
1405 DEV_RX_OFFLOAD_VLAN_STRIP |
1406 DEV_RX_OFFLOAD_IPV4_CKSUM |
1407 DEV_RX_OFFLOAD_UDP_CKSUM |
1408 DEV_RX_OFFLOAD_TCP_CKSUM;
1409 dev_info->tx_offload_capa =
1410 DEV_TX_OFFLOAD_VLAN_INSERT |
1411 DEV_TX_OFFLOAD_IPV4_CKSUM |
1412 DEV_TX_OFFLOAD_UDP_CKSUM |
1413 DEV_TX_OFFLOAD_TCP_CKSUM |
1414 DEV_TX_OFFLOAD_TCP_TSO;
1416 dev_info->hash_key_size = FM10K_RSSRK_SIZE * sizeof(uint32_t);
1417 dev_info->reta_size = FM10K_MAX_RSS_INDICES;
1419 dev_info->default_rxconf = (struct rte_eth_rxconf) {
1421 .pthresh = FM10K_DEFAULT_RX_PTHRESH,
1422 .hthresh = FM10K_DEFAULT_RX_HTHRESH,
1423 .wthresh = FM10K_DEFAULT_RX_WTHRESH,
1425 .rx_free_thresh = FM10K_RX_FREE_THRESH_DEFAULT(0),
1429 dev_info->default_txconf = (struct rte_eth_txconf) {
1431 .pthresh = FM10K_DEFAULT_TX_PTHRESH,
1432 .hthresh = FM10K_DEFAULT_TX_HTHRESH,
1433 .wthresh = FM10K_DEFAULT_TX_WTHRESH,
1435 .tx_free_thresh = FM10K_TX_FREE_THRESH_DEFAULT(0),
1436 .tx_rs_thresh = FM10K_TX_RS_THRESH_DEFAULT(0),
1437 .txq_flags = FM10K_SIMPLE_TX_FLAG,
1440 dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
1441 .nb_max = FM10K_MAX_RX_DESC,
1442 .nb_min = FM10K_MIN_RX_DESC,
1443 .nb_align = FM10K_MULT_RX_DESC,
1446 dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
1447 .nb_max = FM10K_MAX_TX_DESC,
1448 .nb_min = FM10K_MIN_TX_DESC,
1449 .nb_align = FM10K_MULT_TX_DESC,
1452 dev_info->speed_capa = ETH_LINK_SPEED_1G | ETH_LINK_SPEED_2_5G |
1453 ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
1454 ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G;
1457 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
1458 static const uint32_t *
1459 fm10k_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1461 if (dev->rx_pkt_burst == fm10k_recv_pkts ||
1462 dev->rx_pkt_burst == fm10k_recv_scattered_pkts) {
1463 static uint32_t ptypes[] = {
1464 /* refers to rx_desc_to_ol_flags() */
1467 RTE_PTYPE_L3_IPV4_EXT,
1469 RTE_PTYPE_L3_IPV6_EXT,
1476 } else if (dev->rx_pkt_burst == fm10k_recv_pkts_vec ||
1477 dev->rx_pkt_burst == fm10k_recv_scattered_pkts_vec) {
1478 static uint32_t ptypes_vec[] = {
1479 /* refers to fm10k_desc_to_pktype_v() */
1481 RTE_PTYPE_L3_IPV4_EXT,
1483 RTE_PTYPE_L3_IPV6_EXT,
1486 RTE_PTYPE_TUNNEL_GENEVE,
1487 RTE_PTYPE_TUNNEL_NVGRE,
1488 RTE_PTYPE_TUNNEL_VXLAN,
1489 RTE_PTYPE_TUNNEL_GRE,
1499 static const uint32_t *
1500 fm10k_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1507 fm10k_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1510 uint16_t mac_num = 0;
1511 uint32_t vid_idx, vid_bit, mac_index;
1512 struct fm10k_hw *hw;
1513 struct fm10k_macvlan_filter_info *macvlan;
1514 struct rte_eth_dev_data *data = dev->data;
1516 hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1517 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1519 if (macvlan->nb_queue_pools > 0) { /* VMDQ mode */
1520 PMD_INIT_LOG(ERR, "Cannot change VLAN filter in VMDQ mode");
1524 if (vlan_id > ETH_VLAN_ID_MAX) {
1525 PMD_INIT_LOG(ERR, "Invalid vlan_id: must be < 4096");
1529 vid_idx = FM10K_VFTA_IDX(vlan_id);
1530 vid_bit = FM10K_VFTA_BIT(vlan_id);
1531 /* this VLAN ID is already in the VLAN filter table, return SUCCESS */
1532 if (on && (macvlan->vfta[vid_idx] & vid_bit))
1534 /* this VLAN ID is NOT in the VLAN filter table, cannot remove */
1535 if (!on && !(macvlan->vfta[vid_idx] & vid_bit)) {
1536 PMD_INIT_LOG(ERR, "Invalid vlan_id: not existing "
1537 "in the VLAN filter table");
1542 result = fm10k_update_vlan(hw, vlan_id, 0, on);
1543 fm10k_mbx_unlock(hw);
1544 if (result != FM10K_SUCCESS) {
1545 PMD_INIT_LOG(ERR, "VLAN update failed: %d", result);
1549 for (mac_index = 0; (mac_index < FM10K_MAX_MACADDR_NUM) &&
1550 (result == FM10K_SUCCESS); mac_index++) {
1551 if (is_zero_ether_addr(&data->mac_addrs[mac_index]))
1553 if (mac_num > macvlan->mac_num - 1) {
1554 PMD_INIT_LOG(ERR, "MAC address number "
1559 result = fm10k_update_uc_addr(hw, hw->mac.dglort_map,
1560 data->mac_addrs[mac_index].addr_bytes,
1562 fm10k_mbx_unlock(hw);
1565 if (result != FM10K_SUCCESS) {
1566 PMD_INIT_LOG(ERR, "MAC address update failed: %d", result);
1571 macvlan->vlan_num++;
1572 macvlan->vfta[vid_idx] |= vid_bit;
1574 macvlan->vlan_num--;
1575 macvlan->vfta[vid_idx] &= ~vid_bit;
1581 fm10k_vlan_offload_set(__rte_unused struct rte_eth_dev *dev, int mask)
1583 if (mask & ETH_VLAN_STRIP_MASK) {
1584 if (!dev->data->dev_conf.rxmode.hw_vlan_strip)
1585 PMD_INIT_LOG(ERR, "VLAN stripping is "
1586 "always on in fm10k");
1589 if (mask & ETH_VLAN_EXTEND_MASK) {
1590 if (dev->data->dev_conf.rxmode.hw_vlan_extend)
1591 PMD_INIT_LOG(ERR, "VLAN QinQ is not "
1592 "supported in fm10k");
1595 if (mask & ETH_VLAN_FILTER_MASK) {
1596 if (!dev->data->dev_conf.rxmode.hw_vlan_filter)
1597 PMD_INIT_LOG(ERR, "VLAN filter is always on in fm10k");
1601 /* Add/Remove a MAC address, and update filters to main VSI */
1602 static void fm10k_MAC_filter_set_main_vsi(struct rte_eth_dev *dev,
1603 const u8 *mac, bool add, uint32_t pool)
1605 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1606 struct fm10k_macvlan_filter_info *macvlan;
1609 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1611 if (pool != MAIN_VSI_POOL_NUMBER) {
1612 PMD_DRV_LOG(ERR, "VMDQ not enabled, can't set "
1613 "mac to pool %u", pool);
1616 for (i = 0, j = 0; j < FM10K_VFTA_SIZE; j++) {
1617 if (!macvlan->vfta[j])
1619 for (k = 0; k < FM10K_UINT32_BIT_SIZE; k++) {
1620 if (!(macvlan->vfta[j] & (1 << k)))
1622 if (i + 1 > macvlan->vlan_num) {
1623 PMD_INIT_LOG(ERR, "vlan number not match");
1627 fm10k_update_uc_addr(hw, hw->mac.dglort_map, mac,
1628 j * FM10K_UINT32_BIT_SIZE + k, add, 0);
1629 fm10k_mbx_unlock(hw);
1635 /* Add/Remove a MAC address, and update filters to VMDQ */
1636 static void fm10k_MAC_filter_set_vmdq(struct rte_eth_dev *dev,
1637 const u8 *mac, bool add, uint32_t pool)
1639 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1640 struct fm10k_macvlan_filter_info *macvlan;
1641 struct rte_eth_vmdq_rx_conf *vmdq_conf;
1644 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1645 vmdq_conf = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
1647 if (pool > macvlan->nb_queue_pools) {
1648 PMD_DRV_LOG(ERR, "Pool number %u invalid."
1650 pool, macvlan->nb_queue_pools);
1653 for (i = 0; i < vmdq_conf->nb_pool_maps; i++) {
1654 if (!(vmdq_conf->pool_map[i].pools & (1UL << pool)))
1657 fm10k_update_uc_addr(hw, hw->mac.dglort_map + pool, mac,
1658 vmdq_conf->pool_map[i].vlan_id, add, 0);
1659 fm10k_mbx_unlock(hw);
1663 /* Add/Remove a MAC address, and update filters */
1664 static void fm10k_MAC_filter_set(struct rte_eth_dev *dev,
1665 const u8 *mac, bool add, uint32_t pool)
1667 struct fm10k_macvlan_filter_info *macvlan;
1669 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1671 if (macvlan->nb_queue_pools > 0) /* VMDQ mode */
1672 fm10k_MAC_filter_set_vmdq(dev, mac, add, pool);
1674 fm10k_MAC_filter_set_main_vsi(dev, mac, add, pool);
1682 /* Add a MAC address, and update filters */
1684 fm10k_macaddr_add(struct rte_eth_dev *dev,
1685 struct ether_addr *mac_addr,
1689 struct fm10k_macvlan_filter_info *macvlan;
1691 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1692 fm10k_MAC_filter_set(dev, mac_addr->addr_bytes, TRUE, pool);
1693 macvlan->mac_vmdq_id[index] = pool;
1696 /* Remove a MAC address, and update filters */
1698 fm10k_macaddr_remove(struct rte_eth_dev *dev, uint32_t index)
1700 struct rte_eth_dev_data *data = dev->data;
1701 struct fm10k_macvlan_filter_info *macvlan;
1703 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1704 fm10k_MAC_filter_set(dev, data->mac_addrs[index].addr_bytes,
1705 FALSE, macvlan->mac_vmdq_id[index]);
1706 macvlan->mac_vmdq_id[index] = 0;
1710 check_nb_desc(uint16_t min, uint16_t max, uint16_t mult, uint16_t request)
1712 if ((request < min) || (request > max) || ((request % mult) != 0))
1720 check_thresh(uint16_t min, uint16_t max, uint16_t div, uint16_t request)
1722 if ((request < min) || (request > max) || ((div % request) != 0))
1729 handle_rxconf(struct fm10k_rx_queue *q, const struct rte_eth_rxconf *conf)
1731 uint16_t rx_free_thresh;
1733 if (conf->rx_free_thresh == 0)
1734 rx_free_thresh = FM10K_RX_FREE_THRESH_DEFAULT(q);
1736 rx_free_thresh = conf->rx_free_thresh;
1738 /* make sure the requested threshold satisfies the constraints */
1739 if (check_thresh(FM10K_RX_FREE_THRESH_MIN(q),
1740 FM10K_RX_FREE_THRESH_MAX(q),
1741 FM10K_RX_FREE_THRESH_DIV(q),
1743 PMD_INIT_LOG(ERR, "rx_free_thresh (%u) must be "
1744 "less than or equal to %u, "
1745 "greater than or equal to %u, "
1746 "and a divisor of %u",
1747 rx_free_thresh, FM10K_RX_FREE_THRESH_MAX(q),
1748 FM10K_RX_FREE_THRESH_MIN(q),
1749 FM10K_RX_FREE_THRESH_DIV(q));
1753 q->alloc_thresh = rx_free_thresh;
1754 q->drop_en = conf->rx_drop_en;
1755 q->rx_deferred_start = conf->rx_deferred_start;
1761 * Hardware requires specific alignment for Rx packet buffers. At
1762 * least one of the following two conditions must be satisfied.
1763 * 1. Address is 512B aligned
1764 * 2. Address is 8B aligned and buffer does not cross 4K boundary.
1766 * As such, the driver may need to adjust the DMA address within the
1767 * buffer by up to 512B.
1769 * return 1 if the element size is valid, otherwise return 0.
1772 mempool_element_size_valid(struct rte_mempool *mp)
1776 /* elt_size includes mbuf header and headroom */
1777 min_size = mp->elt_size - sizeof(struct rte_mbuf) -
1778 RTE_PKTMBUF_HEADROOM;
1780 /* account for up to 512B of alignment */
1781 min_size -= FM10K_RX_DATABUF_ALIGN;
1783 /* sanity check for overflow */
1784 if (min_size > mp->elt_size)
1792 fm10k_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
1793 uint16_t nb_desc, unsigned int socket_id,
1794 const struct rte_eth_rxconf *conf, struct rte_mempool *mp)
1796 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1797 struct fm10k_dev_info *dev_info = FM10K_DEV_PRIVATE_TO_INFO(dev);
1798 struct fm10k_rx_queue *q;
1799 const struct rte_memzone *mz;
1801 PMD_INIT_FUNC_TRACE();
1803 /* make sure the mempool element size can account for alignment. */
1804 if (!mempool_element_size_valid(mp)) {
1805 PMD_INIT_LOG(ERR, "Error : Mempool element size is too small");
1809 /* make sure a valid number of descriptors have been requested */
1810 if (check_nb_desc(FM10K_MIN_RX_DESC, FM10K_MAX_RX_DESC,
1811 FM10K_MULT_RX_DESC, nb_desc)) {
1812 PMD_INIT_LOG(ERR, "Number of Rx descriptors (%u) must be "
1813 "less than or equal to %"PRIu32", "
1814 "greater than or equal to %u, "
1815 "and a multiple of %u",
1816 nb_desc, (uint32_t)FM10K_MAX_RX_DESC, FM10K_MIN_RX_DESC,
1817 FM10K_MULT_RX_DESC);
1822 * if this queue existed already, free the associated memory. The
1823 * queue cannot be reused in case we need to allocate memory on
1824 * different socket than was previously used.
1826 if (dev->data->rx_queues[queue_id] != NULL) {
1827 rx_queue_free(dev->data->rx_queues[queue_id]);
1828 dev->data->rx_queues[queue_id] = NULL;
1831 /* allocate memory for the queue structure */
1832 q = rte_zmalloc_socket("fm10k", sizeof(*q), RTE_CACHE_LINE_SIZE,
1835 PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
1841 q->nb_desc = nb_desc;
1842 q->nb_fake_desc = FM10K_MULT_RX_DESC;
1843 q->port_id = dev->data->port_id;
1844 q->queue_id = queue_id;
1845 q->tail_ptr = (volatile uint32_t *)
1846 &((uint32_t *)hw->hw_addr)[FM10K_RDT(queue_id)];
1847 if (handle_rxconf(q, conf))
1850 /* allocate memory for the software ring */
1851 q->sw_ring = rte_zmalloc_socket("fm10k sw ring",
1852 (nb_desc + q->nb_fake_desc) * sizeof(struct rte_mbuf *),
1853 RTE_CACHE_LINE_SIZE, socket_id);
1854 if (q->sw_ring == NULL) {
1855 PMD_INIT_LOG(ERR, "Cannot allocate software ring");
1861 * allocate memory for the hardware descriptor ring. A memzone large
1862 * enough to hold the maximum ring size is requested to allow for
1863 * resizing in later calls to the queue setup function.
1865 mz = rte_eth_dma_zone_reserve(dev, "rx_ring", queue_id,
1866 FM10K_MAX_RX_RING_SZ, FM10K_ALIGN_RX_DESC,
1869 PMD_INIT_LOG(ERR, "Cannot allocate hardware ring");
1870 rte_free(q->sw_ring);
1874 q->hw_ring = mz->addr;
1875 q->hw_ring_phys_addr = rte_mem_phy2mch(mz->memseg_id, mz->phys_addr);
1877 /* Check if number of descs satisfied Vector requirement */
1878 if (!rte_is_power_of_2(nb_desc)) {
1879 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Vector Rx "
1880 "preconditions - canceling the feature for "
1881 "the whole port[%d]",
1882 q->queue_id, q->port_id);
1883 dev_info->rx_vec_allowed = false;
1885 fm10k_rxq_vec_setup(q);
1887 dev->data->rx_queues[queue_id] = q;
1892 fm10k_rx_queue_release(void *queue)
1894 PMD_INIT_FUNC_TRACE();
1896 rx_queue_free(queue);
1900 handle_txconf(struct fm10k_tx_queue *q, const struct rte_eth_txconf *conf)
1902 uint16_t tx_free_thresh;
1903 uint16_t tx_rs_thresh;
1905 /* constraint MACROs require that tx_free_thresh is configured
1906 * before tx_rs_thresh */
1907 if (conf->tx_free_thresh == 0)
1908 tx_free_thresh = FM10K_TX_FREE_THRESH_DEFAULT(q);
1910 tx_free_thresh = conf->tx_free_thresh;
1912 /* make sure the requested threshold satisfies the constraints */
1913 if (check_thresh(FM10K_TX_FREE_THRESH_MIN(q),
1914 FM10K_TX_FREE_THRESH_MAX(q),
1915 FM10K_TX_FREE_THRESH_DIV(q),
1917 PMD_INIT_LOG(ERR, "tx_free_thresh (%u) must be "
1918 "less than or equal to %u, "
1919 "greater than or equal to %u, "
1920 "and a divisor of %u",
1921 tx_free_thresh, FM10K_TX_FREE_THRESH_MAX(q),
1922 FM10K_TX_FREE_THRESH_MIN(q),
1923 FM10K_TX_FREE_THRESH_DIV(q));
1927 q->free_thresh = tx_free_thresh;
1929 if (conf->tx_rs_thresh == 0)
1930 tx_rs_thresh = FM10K_TX_RS_THRESH_DEFAULT(q);
1932 tx_rs_thresh = conf->tx_rs_thresh;
1934 q->tx_deferred_start = conf->tx_deferred_start;
1936 /* make sure the requested threshold satisfies the constraints */
1937 if (check_thresh(FM10K_TX_RS_THRESH_MIN(q),
1938 FM10K_TX_RS_THRESH_MAX(q),
1939 FM10K_TX_RS_THRESH_DIV(q),
1941 PMD_INIT_LOG(ERR, "tx_rs_thresh (%u) must be "
1942 "less than or equal to %u, "
1943 "greater than or equal to %u, "
1944 "and a divisor of %u",
1945 tx_rs_thresh, FM10K_TX_RS_THRESH_MAX(q),
1946 FM10K_TX_RS_THRESH_MIN(q),
1947 FM10K_TX_RS_THRESH_DIV(q));
1951 q->rs_thresh = tx_rs_thresh;
1957 fm10k_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
1958 uint16_t nb_desc, unsigned int socket_id,
1959 const struct rte_eth_txconf *conf)
1961 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1962 struct fm10k_tx_queue *q;
1963 const struct rte_memzone *mz;
1965 PMD_INIT_FUNC_TRACE();
1967 /* make sure a valid number of descriptors have been requested */
1968 if (check_nb_desc(FM10K_MIN_TX_DESC, FM10K_MAX_TX_DESC,
1969 FM10K_MULT_TX_DESC, nb_desc)) {
1970 PMD_INIT_LOG(ERR, "Number of Tx descriptors (%u) must be "
1971 "less than or equal to %"PRIu32", "
1972 "greater than or equal to %u, "
1973 "and a multiple of %u",
1974 nb_desc, (uint32_t)FM10K_MAX_TX_DESC, FM10K_MIN_TX_DESC,
1975 FM10K_MULT_TX_DESC);
1980 * if this queue existed already, free the associated memory. The
1981 * queue cannot be reused in case we need to allocate memory on
1982 * different socket than was previously used.
1984 if (dev->data->tx_queues[queue_id] != NULL) {
1985 struct fm10k_tx_queue *txq = dev->data->tx_queues[queue_id];
1988 dev->data->tx_queues[queue_id] = NULL;
1991 /* allocate memory for the queue structure */
1992 q = rte_zmalloc_socket("fm10k", sizeof(*q), RTE_CACHE_LINE_SIZE,
1995 PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
2000 q->nb_desc = nb_desc;
2001 q->port_id = dev->data->port_id;
2002 q->queue_id = queue_id;
2003 q->txq_flags = conf->txq_flags;
2004 q->ops = &def_txq_ops;
2005 q->tail_ptr = (volatile uint32_t *)
2006 &((uint32_t *)hw->hw_addr)[FM10K_TDT(queue_id)];
2007 if (handle_txconf(q, conf))
2010 /* allocate memory for the software ring */
2011 q->sw_ring = rte_zmalloc_socket("fm10k sw ring",
2012 nb_desc * sizeof(struct rte_mbuf *),
2013 RTE_CACHE_LINE_SIZE, socket_id);
2014 if (q->sw_ring == NULL) {
2015 PMD_INIT_LOG(ERR, "Cannot allocate software ring");
2021 * allocate memory for the hardware descriptor ring. A memzone large
2022 * enough to hold the maximum ring size is requested to allow for
2023 * resizing in later calls to the queue setup function.
2025 mz = rte_eth_dma_zone_reserve(dev, "tx_ring", queue_id,
2026 FM10K_MAX_TX_RING_SZ, FM10K_ALIGN_TX_DESC,
2029 PMD_INIT_LOG(ERR, "Cannot allocate hardware ring");
2030 rte_free(q->sw_ring);
2034 q->hw_ring = mz->addr;
2035 q->hw_ring_phys_addr = rte_mem_phy2mch(mz->memseg_id, mz->phys_addr);
2038 * allocate memory for the RS bit tracker. Enough slots to hold the
2039 * descriptor index for each RS bit needing to be set are required.
2041 q->rs_tracker.list = rte_zmalloc_socket("fm10k rs tracker",
2042 ((nb_desc + 1) / q->rs_thresh) *
2044 RTE_CACHE_LINE_SIZE, socket_id);
2045 if (q->rs_tracker.list == NULL) {
2046 PMD_INIT_LOG(ERR, "Cannot allocate RS bit tracker");
2047 rte_free(q->sw_ring);
2052 dev->data->tx_queues[queue_id] = q;
2057 fm10k_tx_queue_release(void *queue)
2059 struct fm10k_tx_queue *q = queue;
2060 PMD_INIT_FUNC_TRACE();
2066 fm10k_reta_update(struct rte_eth_dev *dev,
2067 struct rte_eth_rss_reta_entry64 *reta_conf,
2070 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2071 uint16_t i, j, idx, shift;
2075 PMD_INIT_FUNC_TRACE();
2077 if (reta_size > FM10K_MAX_RSS_INDICES) {
2078 PMD_INIT_LOG(ERR, "The size of hash lookup table configured "
2079 "(%d) doesn't match the number hardware can supported "
2080 "(%d)", reta_size, FM10K_MAX_RSS_INDICES);
2085 * Update Redirection Table RETA[n], n=0..31. The redirection table has
2086 * 128-entries in 32 registers
2088 for (i = 0; i < FM10K_MAX_RSS_INDICES; i += CHARS_PER_UINT32) {
2089 idx = i / RTE_RETA_GROUP_SIZE;
2090 shift = i % RTE_RETA_GROUP_SIZE;
2091 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2092 BIT_MASK_PER_UINT32);
2097 if (mask != BIT_MASK_PER_UINT32)
2098 reta = FM10K_READ_REG(hw, FM10K_RETA(0, i >> 2));
2100 for (j = 0; j < CHARS_PER_UINT32; j++) {
2101 if (mask & (0x1 << j)) {
2103 reta &= ~(UINT8_MAX << CHAR_BIT * j);
2104 reta |= reta_conf[idx].reta[shift + j] <<
2108 FM10K_WRITE_REG(hw, FM10K_RETA(0, i >> 2), reta);
2115 fm10k_reta_query(struct rte_eth_dev *dev,
2116 struct rte_eth_rss_reta_entry64 *reta_conf,
2119 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2120 uint16_t i, j, idx, shift;
2124 PMD_INIT_FUNC_TRACE();
2126 if (reta_size < FM10K_MAX_RSS_INDICES) {
2127 PMD_INIT_LOG(ERR, "The size of hash lookup table configured "
2128 "(%d) doesn't match the number hardware can supported "
2129 "(%d)", reta_size, FM10K_MAX_RSS_INDICES);
2134 * Read Redirection Table RETA[n], n=0..31. The redirection table has
2135 * 128-entries in 32 registers
2137 for (i = 0; i < FM10K_MAX_RSS_INDICES; i += CHARS_PER_UINT32) {
2138 idx = i / RTE_RETA_GROUP_SIZE;
2139 shift = i % RTE_RETA_GROUP_SIZE;
2140 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2141 BIT_MASK_PER_UINT32);
2145 reta = FM10K_READ_REG(hw, FM10K_RETA(0, i >> 2));
2146 for (j = 0; j < CHARS_PER_UINT32; j++) {
2147 if (mask & (0x1 << j))
2148 reta_conf[idx].reta[shift + j] = ((reta >>
2149 CHAR_BIT * j) & UINT8_MAX);
2157 fm10k_rss_hash_update(struct rte_eth_dev *dev,
2158 struct rte_eth_rss_conf *rss_conf)
2160 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2161 uint32_t *key = (uint32_t *)rss_conf->rss_key;
2163 uint64_t hf = rss_conf->rss_hf;
2166 PMD_INIT_FUNC_TRACE();
2168 if (rss_conf->rss_key_len < FM10K_RSSRK_SIZE *
2169 FM10K_RSSRK_ENTRIES_PER_REG)
2176 mrqc |= (hf & ETH_RSS_IPV4) ? FM10K_MRQC_IPV4 : 0;
2177 mrqc |= (hf & ETH_RSS_IPV6) ? FM10K_MRQC_IPV6 : 0;
2178 mrqc |= (hf & ETH_RSS_IPV6_EX) ? FM10K_MRQC_IPV6 : 0;
2179 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_TCP) ? FM10K_MRQC_TCP_IPV4 : 0;
2180 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_TCP) ? FM10K_MRQC_TCP_IPV6 : 0;
2181 mrqc |= (hf & ETH_RSS_IPV6_TCP_EX) ? FM10K_MRQC_TCP_IPV6 : 0;
2182 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_UDP) ? FM10K_MRQC_UDP_IPV4 : 0;
2183 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_UDP) ? FM10K_MRQC_UDP_IPV6 : 0;
2184 mrqc |= (hf & ETH_RSS_IPV6_UDP_EX) ? FM10K_MRQC_UDP_IPV6 : 0;
2186 /* If the mapping doesn't fit any supported, return */
2191 for (i = 0; i < FM10K_RSSRK_SIZE; ++i)
2192 FM10K_WRITE_REG(hw, FM10K_RSSRK(0, i), key[i]);
2194 FM10K_WRITE_REG(hw, FM10K_MRQC(0), mrqc);
2200 fm10k_rss_hash_conf_get(struct rte_eth_dev *dev,
2201 struct rte_eth_rss_conf *rss_conf)
2203 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2204 uint32_t *key = (uint32_t *)rss_conf->rss_key;
2209 PMD_INIT_FUNC_TRACE();
2211 if (rss_conf->rss_key_len < FM10K_RSSRK_SIZE *
2212 FM10K_RSSRK_ENTRIES_PER_REG)
2216 for (i = 0; i < FM10K_RSSRK_SIZE; ++i)
2217 key[i] = FM10K_READ_REG(hw, FM10K_RSSRK(0, i));
2219 mrqc = FM10K_READ_REG(hw, FM10K_MRQC(0));
2221 hf |= (mrqc & FM10K_MRQC_IPV4) ? ETH_RSS_IPV4 : 0;
2222 hf |= (mrqc & FM10K_MRQC_IPV6) ? ETH_RSS_IPV6 : 0;
2223 hf |= (mrqc & FM10K_MRQC_IPV6) ? ETH_RSS_IPV6_EX : 0;
2224 hf |= (mrqc & FM10K_MRQC_TCP_IPV4) ? ETH_RSS_NONFRAG_IPV4_TCP : 0;
2225 hf |= (mrqc & FM10K_MRQC_TCP_IPV6) ? ETH_RSS_NONFRAG_IPV6_TCP : 0;
2226 hf |= (mrqc & FM10K_MRQC_TCP_IPV6) ? ETH_RSS_IPV6_TCP_EX : 0;
2227 hf |= (mrqc & FM10K_MRQC_UDP_IPV4) ? ETH_RSS_NONFRAG_IPV4_UDP : 0;
2228 hf |= (mrqc & FM10K_MRQC_UDP_IPV6) ? ETH_RSS_NONFRAG_IPV6_UDP : 0;
2229 hf |= (mrqc & FM10K_MRQC_UDP_IPV6) ? ETH_RSS_IPV6_UDP_EX : 0;
2231 rss_conf->rss_hf = hf;
2237 fm10k_dev_enable_intr_pf(struct rte_eth_dev *dev)
2239 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2240 uint32_t int_map = FM10K_INT_MAP_IMMEDIATE;
2242 /* Bind all local non-queue interrupt to vector 0 */
2243 int_map |= FM10K_MISC_VEC_ID;
2245 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_mailbox), int_map);
2246 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_pcie_fault), int_map);
2247 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_up_down), int_map);
2248 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_event), int_map);
2249 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_sram), int_map);
2250 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_vflr), int_map);
2252 /* Enable misc causes */
2253 FM10K_WRITE_REG(hw, FM10K_EIMR, FM10K_EIMR_ENABLE(PCA_FAULT) |
2254 FM10K_EIMR_ENABLE(THI_FAULT) |
2255 FM10K_EIMR_ENABLE(FUM_FAULT) |
2256 FM10K_EIMR_ENABLE(MAILBOX) |
2257 FM10K_EIMR_ENABLE(SWITCHREADY) |
2258 FM10K_EIMR_ENABLE(SWITCHNOTREADY) |
2259 FM10K_EIMR_ENABLE(SRAMERROR) |
2260 FM10K_EIMR_ENABLE(VFLR));
2263 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_AUTOMASK |
2264 FM10K_ITR_MASK_CLEAR);
2265 FM10K_WRITE_FLUSH(hw);
2269 fm10k_dev_disable_intr_pf(struct rte_eth_dev *dev)
2271 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2272 uint32_t int_map = FM10K_INT_MAP_DISABLE;
2274 int_map |= FM10K_MISC_VEC_ID;
2276 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_mailbox), int_map);
2277 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_pcie_fault), int_map);
2278 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_up_down), int_map);
2279 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_event), int_map);
2280 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_sram), int_map);
2281 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_vflr), int_map);
2283 /* Disable misc causes */
2284 FM10K_WRITE_REG(hw, FM10K_EIMR, FM10K_EIMR_DISABLE(PCA_FAULT) |
2285 FM10K_EIMR_DISABLE(THI_FAULT) |
2286 FM10K_EIMR_DISABLE(FUM_FAULT) |
2287 FM10K_EIMR_DISABLE(MAILBOX) |
2288 FM10K_EIMR_DISABLE(SWITCHREADY) |
2289 FM10K_EIMR_DISABLE(SWITCHNOTREADY) |
2290 FM10K_EIMR_DISABLE(SRAMERROR) |
2291 FM10K_EIMR_DISABLE(VFLR));
2294 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_MASK_SET);
2295 FM10K_WRITE_FLUSH(hw);
2299 fm10k_dev_enable_intr_vf(struct rte_eth_dev *dev)
2301 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2302 uint32_t int_map = FM10K_INT_MAP_IMMEDIATE;
2304 /* Bind all local non-queue interrupt to vector 0 */
2305 int_map |= FM10K_MISC_VEC_ID;
2307 /* Only INT 0 available, other 15 are reserved. */
2308 FM10K_WRITE_REG(hw, FM10K_VFINT_MAP, int_map);
2311 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_AUTOMASK |
2312 FM10K_ITR_MASK_CLEAR);
2313 FM10K_WRITE_FLUSH(hw);
2317 fm10k_dev_disable_intr_vf(struct rte_eth_dev *dev)
2319 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2320 uint32_t int_map = FM10K_INT_MAP_DISABLE;
2322 int_map |= FM10K_MISC_VEC_ID;
2324 /* Only INT 0 available, other 15 are reserved. */
2325 FM10K_WRITE_REG(hw, FM10K_VFINT_MAP, int_map);
2328 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_MASK_SET);
2329 FM10K_WRITE_FLUSH(hw);
2333 fm10k_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
2335 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2338 if (hw->mac.type == fm10k_mac_pf)
2339 FM10K_WRITE_REG(hw, FM10K_ITR(Q2V(dev, queue_id)),
2340 FM10K_ITR_AUTOMASK | FM10K_ITR_MASK_CLEAR);
2342 FM10K_WRITE_REG(hw, FM10K_VFITR(Q2V(dev, queue_id)),
2343 FM10K_ITR_AUTOMASK | FM10K_ITR_MASK_CLEAR);
2344 rte_intr_enable(&dev->pci_dev->intr_handle);
2349 fm10k_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
2351 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2354 if (hw->mac.type == fm10k_mac_pf)
2355 FM10K_WRITE_REG(hw, FM10K_ITR(Q2V(dev, queue_id)),
2356 FM10K_ITR_MASK_SET);
2358 FM10K_WRITE_REG(hw, FM10K_VFITR(Q2V(dev, queue_id)),
2359 FM10K_ITR_MASK_SET);
2364 fm10k_dev_rxq_interrupt_setup(struct rte_eth_dev *dev)
2366 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2367 struct rte_intr_handle *intr_handle = &dev->pci_dev->intr_handle;
2368 uint32_t intr_vector, vec;
2372 /* fm10k needs one separate interrupt for mailbox,
2373 * so only drivers which support multiple interrupt vectors
2374 * e.g. vfio-pci can work for fm10k interrupt mode
2376 if (!rte_intr_cap_multiple(intr_handle) ||
2377 dev->data->dev_conf.intr_conf.rxq == 0)
2380 intr_vector = dev->data->nb_rx_queues;
2382 /* disable interrupt first */
2383 rte_intr_disable(&dev->pci_dev->intr_handle);
2384 if (hw->mac.type == fm10k_mac_pf)
2385 fm10k_dev_disable_intr_pf(dev);
2387 fm10k_dev_disable_intr_vf(dev);
2389 if (rte_intr_efd_enable(intr_handle, intr_vector)) {
2390 PMD_INIT_LOG(ERR, "Failed to init event fd");
2394 if (rte_intr_dp_is_en(intr_handle) && !result) {
2395 intr_handle->intr_vec = rte_zmalloc("intr_vec",
2396 dev->data->nb_rx_queues * sizeof(int), 0);
2397 if (intr_handle->intr_vec) {
2398 for (queue_id = 0, vec = FM10K_RX_VEC_START;
2399 queue_id < dev->data->nb_rx_queues;
2401 intr_handle->intr_vec[queue_id] = vec;
2402 if (vec < intr_handle->nb_efd - 1
2403 + FM10K_RX_VEC_START)
2407 PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
2408 " intr_vec", dev->data->nb_rx_queues);
2409 rte_intr_efd_disable(intr_handle);
2414 if (hw->mac.type == fm10k_mac_pf)
2415 fm10k_dev_enable_intr_pf(dev);
2417 fm10k_dev_enable_intr_vf(dev);
2418 rte_intr_enable(&dev->pci_dev->intr_handle);
2419 hw->mac.ops.update_int_moderator(hw);
2424 fm10k_dev_handle_fault(struct fm10k_hw *hw, uint32_t eicr)
2426 struct fm10k_fault fault;
2428 const char *estr = "Unknown error";
2430 /* Process PCA fault */
2431 if (eicr & FM10K_EICR_PCA_FAULT) {
2432 err = fm10k_get_fault(hw, FM10K_PCA_FAULT, &fault);
2435 switch (fault.type) {
2437 estr = "PCA_NO_FAULT"; break;
2438 case PCA_UNMAPPED_ADDR:
2439 estr = "PCA_UNMAPPED_ADDR"; break;
2440 case PCA_BAD_QACCESS_PF:
2441 estr = "PCA_BAD_QACCESS_PF"; break;
2442 case PCA_BAD_QACCESS_VF:
2443 estr = "PCA_BAD_QACCESS_VF"; break;
2444 case PCA_MALICIOUS_REQ:
2445 estr = "PCA_MALICIOUS_REQ"; break;
2446 case PCA_POISONED_TLP:
2447 estr = "PCA_POISONED_TLP"; break;
2449 estr = "PCA_TLP_ABORT"; break;
2453 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
2454 estr, fault.func ? "VF" : "PF", fault.func,
2455 fault.address, fault.specinfo);
2458 /* Process THI fault */
2459 if (eicr & FM10K_EICR_THI_FAULT) {
2460 err = fm10k_get_fault(hw, FM10K_THI_FAULT, &fault);
2463 switch (fault.type) {
2465 estr = "THI_NO_FAULT"; break;
2466 case THI_MAL_DIS_Q_FAULT:
2467 estr = "THI_MAL_DIS_Q_FAULT"; break;
2471 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
2472 estr, fault.func ? "VF" : "PF", fault.func,
2473 fault.address, fault.specinfo);
2476 /* Process FUM fault */
2477 if (eicr & FM10K_EICR_FUM_FAULT) {
2478 err = fm10k_get_fault(hw, FM10K_FUM_FAULT, &fault);
2481 switch (fault.type) {
2483 estr = "FUM_NO_FAULT"; break;
2484 case FUM_UNMAPPED_ADDR:
2485 estr = "FUM_UNMAPPED_ADDR"; break;
2486 case FUM_POISONED_TLP:
2487 estr = "FUM_POISONED_TLP"; break;
2488 case FUM_BAD_VF_QACCESS:
2489 estr = "FUM_BAD_VF_QACCESS"; break;
2490 case FUM_ADD_DECODE_ERR:
2491 estr = "FUM_ADD_DECODE_ERR"; break;
2493 estr = "FUM_RO_ERROR"; break;
2494 case FUM_QPRC_CRC_ERROR:
2495 estr = "FUM_QPRC_CRC_ERROR"; break;
2496 case FUM_CSR_TIMEOUT:
2497 estr = "FUM_CSR_TIMEOUT"; break;
2498 case FUM_INVALID_TYPE:
2499 estr = "FUM_INVALID_TYPE"; break;
2500 case FUM_INVALID_LENGTH:
2501 estr = "FUM_INVALID_LENGTH"; break;
2502 case FUM_INVALID_BE:
2503 estr = "FUM_INVALID_BE"; break;
2504 case FUM_INVALID_ALIGN:
2505 estr = "FUM_INVALID_ALIGN"; break;
2509 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
2510 estr, fault.func ? "VF" : "PF", fault.func,
2511 fault.address, fault.specinfo);
2516 PMD_INIT_LOG(ERR, "Failed to handle fault event.");
2521 * PF interrupt handler triggered by NIC for handling specific interrupt.
2524 * Pointer to interrupt handle.
2526 * The address of parameter (struct rte_eth_dev *) regsitered before.
2532 fm10k_dev_interrupt_handler_pf(
2533 __rte_unused struct rte_intr_handle *handle,
2536 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2537 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2538 uint32_t cause, status;
2540 if (hw->mac.type != fm10k_mac_pf)
2543 cause = FM10K_READ_REG(hw, FM10K_EICR);
2545 /* Handle PCI fault cases */
2546 if (cause & FM10K_EICR_FAULT_MASK) {
2547 PMD_INIT_LOG(ERR, "INT: find fault!");
2548 fm10k_dev_handle_fault(hw, cause);
2551 /* Handle switch up/down */
2552 if (cause & FM10K_EICR_SWITCHNOTREADY)
2553 PMD_INIT_LOG(ERR, "INT: Switch is not ready");
2555 if (cause & FM10K_EICR_SWITCHREADY)
2556 PMD_INIT_LOG(INFO, "INT: Switch is ready");
2558 /* Handle mailbox message */
2560 hw->mbx.ops.process(hw, &hw->mbx);
2561 fm10k_mbx_unlock(hw);
2563 /* Handle SRAM error */
2564 if (cause & FM10K_EICR_SRAMERROR) {
2565 PMD_INIT_LOG(ERR, "INT: SRAM error on PEP");
2567 status = FM10K_READ_REG(hw, FM10K_SRAM_IP);
2568 /* Write to clear pending bits */
2569 FM10K_WRITE_REG(hw, FM10K_SRAM_IP, status);
2571 /* Todo: print out error message after shared code updates */
2574 /* Clear these 3 events if having any */
2575 cause &= FM10K_EICR_SWITCHNOTREADY | FM10K_EICR_MAILBOX |
2576 FM10K_EICR_SWITCHREADY;
2578 FM10K_WRITE_REG(hw, FM10K_EICR, cause);
2580 /* Re-enable interrupt from device side */
2581 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_AUTOMASK |
2582 FM10K_ITR_MASK_CLEAR);
2583 /* Re-enable interrupt from host side */
2584 rte_intr_enable(&(dev->pci_dev->intr_handle));
2588 * VF interrupt handler triggered by NIC for handling specific interrupt.
2591 * Pointer to interrupt handle.
2593 * The address of parameter (struct rte_eth_dev *) regsitered before.
2599 fm10k_dev_interrupt_handler_vf(
2600 __rte_unused struct rte_intr_handle *handle,
2603 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2604 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2606 if (hw->mac.type != fm10k_mac_vf)
2609 /* Handle mailbox message if lock is acquired */
2611 hw->mbx.ops.process(hw, &hw->mbx);
2612 fm10k_mbx_unlock(hw);
2614 /* Re-enable interrupt from device side */
2615 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_AUTOMASK |
2616 FM10K_ITR_MASK_CLEAR);
2617 /* Re-enable interrupt from host side */
2618 rte_intr_enable(&(dev->pci_dev->intr_handle));
2621 /* Mailbox message handler in VF */
2622 static const struct fm10k_msg_data fm10k_msgdata_vf[] = {
2623 FM10K_TLV_MSG_TEST_HANDLER(fm10k_tlv_msg_test),
2624 FM10K_VF_MSG_MAC_VLAN_HANDLER(fm10k_msg_mac_vlan_vf),
2625 FM10K_VF_MSG_LPORT_STATE_HANDLER(fm10k_msg_lport_state_vf),
2626 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_tlv_msg_error),
2630 fm10k_setup_mbx_service(struct fm10k_hw *hw)
2634 /* Initialize mailbox lock */
2635 fm10k_mbx_initlock(hw);
2637 /* Replace default message handler with new ones */
2638 if (hw->mac.type == fm10k_mac_vf)
2639 err = hw->mbx.ops.register_handlers(&hw->mbx, fm10k_msgdata_vf);
2642 PMD_INIT_LOG(ERR, "Failed to register mailbox handler.err:%d",
2646 /* Connect to SM for PF device or PF for VF device */
2647 return hw->mbx.ops.connect(hw, &hw->mbx);
2651 fm10k_close_mbx_service(struct fm10k_hw *hw)
2653 /* Disconnect from SM for PF device or PF for VF device */
2654 hw->mbx.ops.disconnect(hw, &hw->mbx);
2657 static const struct eth_dev_ops fm10k_eth_dev_ops = {
2658 .dev_configure = fm10k_dev_configure,
2659 .dev_start = fm10k_dev_start,
2660 .dev_stop = fm10k_dev_stop,
2661 .dev_close = fm10k_dev_close,
2662 .promiscuous_enable = fm10k_dev_promiscuous_enable,
2663 .promiscuous_disable = fm10k_dev_promiscuous_disable,
2664 .allmulticast_enable = fm10k_dev_allmulticast_enable,
2665 .allmulticast_disable = fm10k_dev_allmulticast_disable,
2666 .stats_get = fm10k_stats_get,
2667 .xstats_get = fm10k_xstats_get,
2668 .xstats_get_names = fm10k_xstats_get_names,
2669 .stats_reset = fm10k_stats_reset,
2670 .xstats_reset = fm10k_stats_reset,
2671 .link_update = fm10k_link_update,
2672 .dev_infos_get = fm10k_dev_infos_get,
2673 .dev_supported_ptypes_get = fm10k_dev_supported_ptypes_get,
2674 .vlan_filter_set = fm10k_vlan_filter_set,
2675 .vlan_offload_set = fm10k_vlan_offload_set,
2676 .mac_addr_add = fm10k_macaddr_add,
2677 .mac_addr_remove = fm10k_macaddr_remove,
2678 .rx_queue_start = fm10k_dev_rx_queue_start,
2679 .rx_queue_stop = fm10k_dev_rx_queue_stop,
2680 .tx_queue_start = fm10k_dev_tx_queue_start,
2681 .tx_queue_stop = fm10k_dev_tx_queue_stop,
2682 .rx_queue_setup = fm10k_rx_queue_setup,
2683 .rx_queue_release = fm10k_rx_queue_release,
2684 .tx_queue_setup = fm10k_tx_queue_setup,
2685 .tx_queue_release = fm10k_tx_queue_release,
2686 .rx_descriptor_done = fm10k_dev_rx_descriptor_done,
2687 .rx_queue_intr_enable = fm10k_dev_rx_queue_intr_enable,
2688 .rx_queue_intr_disable = fm10k_dev_rx_queue_intr_disable,
2689 .reta_update = fm10k_reta_update,
2690 .reta_query = fm10k_reta_query,
2691 .rss_hash_update = fm10k_rss_hash_update,
2692 .rss_hash_conf_get = fm10k_rss_hash_conf_get,
2695 static int ftag_check_handler(__rte_unused const char *key,
2696 const char *value, __rte_unused void *opaque)
2698 if (strcmp(value, "1"))
2705 fm10k_check_ftag(struct rte_devargs *devargs)
2707 struct rte_kvargs *kvlist;
2708 const char *ftag_key = "enable_ftag";
2710 if (devargs == NULL)
2713 kvlist = rte_kvargs_parse(devargs->args, NULL);
2717 if (!rte_kvargs_count(kvlist, ftag_key)) {
2718 rte_kvargs_free(kvlist);
2721 /* FTAG is enabled when there's key-value pair: enable_ftag=1 */
2722 if (rte_kvargs_process(kvlist, ftag_key,
2723 ftag_check_handler, NULL) < 0) {
2724 rte_kvargs_free(kvlist);
2727 rte_kvargs_free(kvlist);
2732 static void __attribute__((cold))
2733 fm10k_set_tx_function(struct rte_eth_dev *dev)
2735 struct fm10k_tx_queue *txq;
2738 uint16_t tx_ftag_en = 0;
2740 if (fm10k_check_ftag(dev->pci_dev->devargs))
2743 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2744 txq = dev->data->tx_queues[i];
2745 txq->tx_ftag_en = tx_ftag_en;
2746 /* Check if Vector Tx is satisfied */
2747 if (fm10k_tx_vec_condition_check(txq)) {
2754 PMD_INIT_LOG(DEBUG, "Use vector Tx func");
2755 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2756 txq = dev->data->tx_queues[i];
2757 fm10k_txq_vec_setup(txq);
2759 dev->tx_pkt_burst = fm10k_xmit_pkts_vec;
2761 dev->tx_pkt_burst = fm10k_xmit_pkts;
2762 PMD_INIT_LOG(DEBUG, "Use regular Tx func");
2766 static void __attribute__((cold))
2767 fm10k_set_rx_function(struct rte_eth_dev *dev)
2769 struct fm10k_dev_info *dev_info = FM10K_DEV_PRIVATE_TO_INFO(dev);
2770 uint16_t i, rx_using_sse;
2771 uint16_t rx_ftag_en = 0;
2773 if (fm10k_check_ftag(dev->pci_dev->devargs))
2776 /* In order to allow Vector Rx there are a few configuration
2777 * conditions to be met.
2779 if (!fm10k_rx_vec_condition_check(dev) &&
2780 dev_info->rx_vec_allowed && !rx_ftag_en) {
2781 if (dev->data->scattered_rx)
2782 dev->rx_pkt_burst = fm10k_recv_scattered_pkts_vec;
2784 dev->rx_pkt_burst = fm10k_recv_pkts_vec;
2785 } else if (dev->data->scattered_rx)
2786 dev->rx_pkt_burst = fm10k_recv_scattered_pkts;
2788 dev->rx_pkt_burst = fm10k_recv_pkts;
2791 (dev->rx_pkt_burst == fm10k_recv_scattered_pkts_vec ||
2792 dev->rx_pkt_burst == fm10k_recv_pkts_vec);
2795 PMD_INIT_LOG(DEBUG, "Use vector Rx func");
2797 PMD_INIT_LOG(DEBUG, "Use regular Rx func");
2799 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2800 struct fm10k_rx_queue *rxq = dev->data->rx_queues[i];
2802 rxq->rx_using_sse = rx_using_sse;
2803 rxq->rx_ftag_en = rx_ftag_en;
2808 fm10k_params_init(struct rte_eth_dev *dev)
2810 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2811 struct fm10k_dev_info *info = FM10K_DEV_PRIVATE_TO_INFO(dev);
2813 /* Inialize bus info. Normally we would call fm10k_get_bus_info(), but
2814 * there is no way to get link status without reading BAR4. Until this
2815 * works, assume we have maximum bandwidth.
2816 * @todo - fix bus info
2818 hw->bus_caps.speed = fm10k_bus_speed_8000;
2819 hw->bus_caps.width = fm10k_bus_width_pcie_x8;
2820 hw->bus_caps.payload = fm10k_bus_payload_512;
2821 hw->bus.speed = fm10k_bus_speed_8000;
2822 hw->bus.width = fm10k_bus_width_pcie_x8;
2823 hw->bus.payload = fm10k_bus_payload_256;
2825 info->rx_vec_allowed = true;
2829 eth_fm10k_dev_init(struct rte_eth_dev *dev)
2831 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2833 struct fm10k_macvlan_filter_info *macvlan;
2835 PMD_INIT_FUNC_TRACE();
2837 dev->dev_ops = &fm10k_eth_dev_ops;
2838 dev->rx_pkt_burst = &fm10k_recv_pkts;
2839 dev->tx_pkt_burst = &fm10k_xmit_pkts;
2841 /* only initialize in the primary process */
2842 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2845 rte_eth_copy_pci_info(dev, dev->pci_dev);
2847 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
2848 memset(macvlan, 0, sizeof(*macvlan));
2849 /* Vendor and Device ID need to be set before init of shared code */
2850 memset(hw, 0, sizeof(*hw));
2851 hw->device_id = dev->pci_dev->id.device_id;
2852 hw->vendor_id = dev->pci_dev->id.vendor_id;
2853 hw->subsystem_device_id = dev->pci_dev->id.subsystem_device_id;
2854 hw->subsystem_vendor_id = dev->pci_dev->id.subsystem_vendor_id;
2855 hw->revision_id = 0;
2856 hw->hw_addr = (void *)dev->pci_dev->mem_resource[0].addr;
2857 if (hw->hw_addr == NULL) {
2858 PMD_INIT_LOG(ERR, "Bad mem resource."
2859 " Try to blacklist unused devices.");
2863 /* Store fm10k_adapter pointer */
2864 hw->back = dev->data->dev_private;
2866 /* Initialize the shared code */
2867 diag = fm10k_init_shared_code(hw);
2868 if (diag != FM10K_SUCCESS) {
2869 PMD_INIT_LOG(ERR, "Shared code init failed: %d", diag);
2873 /* Initialize parameters */
2874 fm10k_params_init(dev);
2876 /* Initialize the hw */
2877 diag = fm10k_init_hw(hw);
2878 if (diag != FM10K_SUCCESS) {
2879 PMD_INIT_LOG(ERR, "Hardware init failed: %d", diag);
2883 /* Initialize MAC address(es) */
2884 dev->data->mac_addrs = rte_zmalloc("fm10k",
2885 ETHER_ADDR_LEN * FM10K_MAX_MACADDR_NUM, 0);
2886 if (dev->data->mac_addrs == NULL) {
2887 PMD_INIT_LOG(ERR, "Cannot allocate memory for MAC addresses");
2891 diag = fm10k_read_mac_addr(hw);
2893 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
2894 &dev->data->mac_addrs[0]);
2896 if (diag != FM10K_SUCCESS ||
2897 !is_valid_assigned_ether_addr(dev->data->mac_addrs)) {
2899 /* Generate a random addr */
2900 eth_random_addr(hw->mac.addr);
2901 memcpy(hw->mac.perm_addr, hw->mac.addr, ETH_ALEN);
2902 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
2903 &dev->data->mac_addrs[0]);
2906 /* Reset the hw statistics */
2907 fm10k_stats_reset(dev);
2910 diag = fm10k_reset_hw(hw);
2911 if (diag != FM10K_SUCCESS) {
2912 PMD_INIT_LOG(ERR, "Hardware reset failed: %d", diag);
2916 /* Setup mailbox service */
2917 diag = fm10k_setup_mbx_service(hw);
2918 if (diag != FM10K_SUCCESS) {
2919 PMD_INIT_LOG(ERR, "Failed to setup mailbox: %d", diag);
2923 /*PF/VF has different interrupt handling mechanism */
2924 if (hw->mac.type == fm10k_mac_pf) {
2925 /* register callback func to eal lib */
2926 rte_intr_callback_register(&(dev->pci_dev->intr_handle),
2927 fm10k_dev_interrupt_handler_pf, (void *)dev);
2929 /* enable MISC interrupt */
2930 fm10k_dev_enable_intr_pf(dev);
2932 rte_intr_callback_register(&(dev->pci_dev->intr_handle),
2933 fm10k_dev_interrupt_handler_vf, (void *)dev);
2935 fm10k_dev_enable_intr_vf(dev);
2938 /* Enable intr after callback registered */
2939 rte_intr_enable(&(dev->pci_dev->intr_handle));
2941 hw->mac.ops.update_int_moderator(hw);
2943 /* Make sure Switch Manager is ready before going forward. */
2944 if (hw->mac.type == fm10k_mac_pf) {
2945 int switch_ready = 0;
2947 for (i = 0; i < MAX_QUERY_SWITCH_STATE_TIMES; i++) {
2949 hw->mac.ops.get_host_state(hw, &switch_ready);
2950 fm10k_mbx_unlock(hw);
2953 /* Delay some time to acquire async LPORT_MAP info. */
2954 rte_delay_us(WAIT_SWITCH_MSG_US);
2957 if (switch_ready == 0) {
2958 PMD_INIT_LOG(ERR, "switch is not ready");
2964 * Below function will trigger operations on mailbox, acquire lock to
2965 * avoid race condition from interrupt handler. Operations on mailbox
2966 * FIFO will trigger interrupt to PF/SM, in which interrupt handler
2967 * will handle and generate an interrupt to our side. Then, FIFO in
2968 * mailbox will be touched.
2971 /* Enable port first */
2972 hw->mac.ops.update_lport_state(hw, hw->mac.dglort_map,
2975 /* Set unicast mode by default. App can change to other mode in other
2978 hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
2979 FM10K_XCAST_MODE_NONE);
2981 fm10k_mbx_unlock(hw);
2983 /* Make sure default VID is ready before going forward. */
2984 if (hw->mac.type == fm10k_mac_pf) {
2985 for (i = 0; i < MAX_QUERY_SWITCH_STATE_TIMES; i++) {
2986 if (hw->mac.default_vid)
2988 /* Delay some time to acquire async port VLAN info. */
2989 rte_delay_us(WAIT_SWITCH_MSG_US);
2992 if (!hw->mac.default_vid) {
2993 PMD_INIT_LOG(ERR, "default VID is not ready");
2998 /* Add default mac address */
2999 fm10k_MAC_filter_set(dev, hw->mac.addr, true,
3000 MAIN_VSI_POOL_NUMBER);
3006 eth_fm10k_dev_uninit(struct rte_eth_dev *dev)
3008 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3010 PMD_INIT_FUNC_TRACE();
3012 /* only uninitialize in the primary process */
3013 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3016 /* safe to close dev here */
3017 fm10k_dev_close(dev);
3019 dev->dev_ops = NULL;
3020 dev->rx_pkt_burst = NULL;
3021 dev->tx_pkt_burst = NULL;
3023 /* disable uio/vfio intr */
3024 rte_intr_disable(&(dev->pci_dev->intr_handle));
3026 /*PF/VF has different interrupt handling mechanism */
3027 if (hw->mac.type == fm10k_mac_pf) {
3028 /* disable interrupt */
3029 fm10k_dev_disable_intr_pf(dev);
3031 /* unregister callback func to eal lib */
3032 rte_intr_callback_unregister(&(dev->pci_dev->intr_handle),
3033 fm10k_dev_interrupt_handler_pf, (void *)dev);
3035 /* disable interrupt */
3036 fm10k_dev_disable_intr_vf(dev);
3038 rte_intr_callback_unregister(&(dev->pci_dev->intr_handle),
3039 fm10k_dev_interrupt_handler_vf, (void *)dev);
3042 /* free mac memory */
3043 if (dev->data->mac_addrs) {
3044 rte_free(dev->data->mac_addrs);
3045 dev->data->mac_addrs = NULL;
3048 memset(hw, 0, sizeof(*hw));
3054 * The set of PCI devices this driver supports. This driver will enable both PF
3055 * and SRIOV-VF devices.
3057 static const struct rte_pci_id pci_id_fm10k_map[] = {
3058 #define RTE_PCI_DEV_ID_DECL_FM10K(vend, dev) { RTE_PCI_DEVICE(vend, dev) },
3059 #define RTE_PCI_DEV_ID_DECL_FM10KVF(vend, dev) { RTE_PCI_DEVICE(vend, dev) },
3060 #include "rte_pci_dev_ids.h"
3061 { .vendor_id = 0, /* sentinel */ },
3064 static struct eth_driver rte_pmd_fm10k = {
3066 .name = "rte_pmd_fm10k",
3067 .id_table = pci_id_fm10k_map,
3068 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
3069 RTE_PCI_DRV_DETACHABLE,
3071 .eth_dev_init = eth_fm10k_dev_init,
3072 .eth_dev_uninit = eth_fm10k_dev_uninit,
3073 .dev_private_size = sizeof(struct fm10k_adapter),
3077 * Driver initialization routine.
3078 * Invoked once at EAL init time.
3079 * Register itself as the [Poll Mode] Driver of PCI FM10K devices.
3082 rte_pmd_fm10k_init(__rte_unused const char *name,
3083 __rte_unused const char *params)
3085 PMD_INIT_FUNC_TRACE();
3086 rte_eth_driver_register(&rte_pmd_fm10k);
3090 static struct rte_driver rte_fm10k_driver = {
3092 .init = rte_pmd_fm10k_init,
3095 PMD_REGISTER_DRIVER(rte_fm10k_driver);