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 /* A period of quiescence for switch */
56 #define FM10K_SWITCH_QUIESCE_US 10000
57 /* Number of chars per uint32 type */
58 #define CHARS_PER_UINT32 (sizeof(uint32_t))
59 #define BIT_MASK_PER_UINT32 ((1 << CHARS_PER_UINT32) - 1)
61 /* default 1:1 map from queue ID to interrupt vector ID */
62 #define Q2V(pci_dev, queue_id) ((pci_dev)->intr_handle.intr_vec[queue_id])
64 /* First 64 Logical ports for PF/VMDQ, second 64 for Flow director */
65 #define MAX_LPORT_NUM 128
66 #define GLORT_FD_Q_BASE 0x40
67 #define GLORT_PF_MASK 0xFFC0
68 #define GLORT_FD_MASK GLORT_PF_MASK
69 #define GLORT_FD_INDEX GLORT_FD_Q_BASE
71 static void fm10k_close_mbx_service(struct fm10k_hw *hw);
72 static void fm10k_dev_promiscuous_enable(struct rte_eth_dev *dev);
73 static void fm10k_dev_promiscuous_disable(struct rte_eth_dev *dev);
74 static void fm10k_dev_allmulticast_enable(struct rte_eth_dev *dev);
75 static void fm10k_dev_allmulticast_disable(struct rte_eth_dev *dev);
76 static inline int fm10k_glort_valid(struct fm10k_hw *hw);
78 fm10k_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on);
79 static void fm10k_MAC_filter_set(struct rte_eth_dev *dev,
80 const u8 *mac, bool add, uint32_t pool);
81 static void fm10k_tx_queue_release(void *queue);
82 static void fm10k_rx_queue_release(void *queue);
83 static void fm10k_set_rx_function(struct rte_eth_dev *dev);
84 static void fm10k_set_tx_function(struct rte_eth_dev *dev);
85 static int fm10k_check_ftag(struct rte_devargs *devargs);
87 struct fm10k_xstats_name_off {
88 char name[RTE_ETH_XSTATS_NAME_SIZE];
92 struct fm10k_xstats_name_off fm10k_hw_stats_strings[] = {
93 {"completion_timeout_count", offsetof(struct fm10k_hw_stats, timeout)},
94 {"unsupported_requests_count", offsetof(struct fm10k_hw_stats, ur)},
95 {"completer_abort_count", offsetof(struct fm10k_hw_stats, ca)},
96 {"unsupported_message_count", offsetof(struct fm10k_hw_stats, um)},
97 {"checksum_error_count", offsetof(struct fm10k_hw_stats, xec)},
98 {"vlan_dropped", offsetof(struct fm10k_hw_stats, vlan_drop)},
99 {"loopback_dropped", offsetof(struct fm10k_hw_stats, loopback_drop)},
100 {"rx_mbuf_allocation_errors", offsetof(struct fm10k_hw_stats,
104 #define FM10K_NB_HW_XSTATS (sizeof(fm10k_hw_stats_strings) / \
105 sizeof(fm10k_hw_stats_strings[0]))
107 struct fm10k_xstats_name_off fm10k_hw_stats_rx_q_strings[] = {
108 {"packets", offsetof(struct fm10k_hw_stats_q, rx_packets)},
109 {"bytes", offsetof(struct fm10k_hw_stats_q, rx_bytes)},
110 {"dropped", offsetof(struct fm10k_hw_stats_q, rx_drops)},
113 #define FM10K_NB_RX_Q_XSTATS (sizeof(fm10k_hw_stats_rx_q_strings) / \
114 sizeof(fm10k_hw_stats_rx_q_strings[0]))
116 struct fm10k_xstats_name_off fm10k_hw_stats_tx_q_strings[] = {
117 {"packets", offsetof(struct fm10k_hw_stats_q, tx_packets)},
118 {"bytes", offsetof(struct fm10k_hw_stats_q, tx_bytes)},
121 #define FM10K_NB_TX_Q_XSTATS (sizeof(fm10k_hw_stats_tx_q_strings) / \
122 sizeof(fm10k_hw_stats_tx_q_strings[0]))
124 #define FM10K_NB_XSTATS (FM10K_NB_HW_XSTATS + FM10K_MAX_QUEUES_PF * \
125 (FM10K_NB_RX_Q_XSTATS + FM10K_NB_TX_Q_XSTATS))
127 fm10k_dev_rxq_interrupt_setup(struct rte_eth_dev *dev);
130 fm10k_mbx_initlock(struct fm10k_hw *hw)
132 rte_spinlock_init(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back));
136 fm10k_mbx_lock(struct fm10k_hw *hw)
138 while (!rte_spinlock_trylock(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back)))
139 rte_delay_us(FM10K_MBXLOCK_DELAY_US);
143 fm10k_mbx_unlock(struct fm10k_hw *hw)
145 rte_spinlock_unlock(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back));
148 /* Stubs needed for linkage when vPMD is disabled */
149 int __attribute__((weak))
150 fm10k_rx_vec_condition_check(__rte_unused struct rte_eth_dev *dev)
155 uint16_t __attribute__((weak))
157 __rte_unused void *rx_queue,
158 __rte_unused struct rte_mbuf **rx_pkts,
159 __rte_unused uint16_t nb_pkts)
164 uint16_t __attribute__((weak))
165 fm10k_recv_scattered_pkts_vec(
166 __rte_unused void *rx_queue,
167 __rte_unused struct rte_mbuf **rx_pkts,
168 __rte_unused uint16_t nb_pkts)
173 int __attribute__((weak))
174 fm10k_rxq_vec_setup(__rte_unused struct fm10k_rx_queue *rxq)
180 void __attribute__((weak))
181 fm10k_rx_queue_release_mbufs_vec(
182 __rte_unused struct fm10k_rx_queue *rxq)
187 void __attribute__((weak))
188 fm10k_txq_vec_setup(__rte_unused struct fm10k_tx_queue *txq)
193 int __attribute__((weak))
194 fm10k_tx_vec_condition_check(__rte_unused struct fm10k_tx_queue *txq)
199 uint16_t __attribute__((weak))
200 fm10k_xmit_pkts_vec(__rte_unused void *tx_queue,
201 __rte_unused struct rte_mbuf **tx_pkts,
202 __rte_unused uint16_t nb_pkts)
208 * reset queue to initial state, allocate software buffers used when starting
210 * return 0 on success
211 * return -ENOMEM if buffers cannot be allocated
212 * return -EINVAL if buffers do not satisfy alignment condition
215 rx_queue_reset(struct fm10k_rx_queue *q)
217 static const union fm10k_rx_desc zero = {{0} };
220 PMD_INIT_FUNC_TRACE();
222 diag = rte_mempool_get_bulk(q->mp, (void **)q->sw_ring, q->nb_desc);
226 for (i = 0; i < q->nb_desc; ++i) {
227 fm10k_pktmbuf_reset(q->sw_ring[i], q->port_id);
228 if (!fm10k_addr_alignment_valid(q->sw_ring[i])) {
229 rte_mempool_put_bulk(q->mp, (void **)q->sw_ring,
233 dma_addr = MBUF_DMA_ADDR_DEFAULT(q->sw_ring[i]);
234 q->hw_ring[i].q.pkt_addr = dma_addr;
235 q->hw_ring[i].q.hdr_addr = dma_addr;
238 /* initialize extra software ring entries. Space for these extra
239 * entries is always allocated.
241 memset(&q->fake_mbuf, 0x0, sizeof(q->fake_mbuf));
242 for (i = 0; i < q->nb_fake_desc; ++i) {
243 q->sw_ring[q->nb_desc + i] = &q->fake_mbuf;
244 q->hw_ring[q->nb_desc + i] = zero;
249 q->next_trigger = q->alloc_thresh - 1;
250 FM10K_PCI_REG_WRITE(q->tail_ptr, q->nb_desc - 1);
251 q->rxrearm_start = 0;
258 * clean queue, descriptor rings, free software buffers used when stopping
262 rx_queue_clean(struct fm10k_rx_queue *q)
264 union fm10k_rx_desc zero = {.q = {0, 0, 0, 0} };
266 PMD_INIT_FUNC_TRACE();
268 /* zero descriptor rings */
269 for (i = 0; i < q->nb_desc; ++i)
270 q->hw_ring[i] = zero;
272 /* zero faked descriptors */
273 for (i = 0; i < q->nb_fake_desc; ++i)
274 q->hw_ring[q->nb_desc + i] = zero;
276 /* vPMD driver has a different way of releasing mbufs. */
277 if (q->rx_using_sse) {
278 fm10k_rx_queue_release_mbufs_vec(q);
282 /* free software buffers */
283 for (i = 0; i < q->nb_desc; ++i) {
285 rte_pktmbuf_free_seg(q->sw_ring[i]);
286 q->sw_ring[i] = NULL;
292 * free all queue memory used when releasing the queue (i.e. configure)
295 rx_queue_free(struct fm10k_rx_queue *q)
297 PMD_INIT_FUNC_TRACE();
299 PMD_INIT_LOG(DEBUG, "Freeing rx queue %p", q);
302 rte_free(q->sw_ring);
311 * disable RX queue, wait unitl HW finished necessary flush operation
314 rx_queue_disable(struct fm10k_hw *hw, uint16_t qnum)
318 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(qnum));
319 FM10K_WRITE_REG(hw, FM10K_RXQCTL(qnum),
320 reg & ~FM10K_RXQCTL_ENABLE);
322 /* Wait 100us at most */
323 for (i = 0; i < FM10K_QUEUE_DISABLE_TIMEOUT; i++) {
325 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(qnum));
326 if (!(reg & FM10K_RXQCTL_ENABLE))
330 if (i == FM10K_QUEUE_DISABLE_TIMEOUT)
337 * reset queue to initial state, allocate software buffers used when starting
341 tx_queue_reset(struct fm10k_tx_queue *q)
343 PMD_INIT_FUNC_TRACE();
347 q->nb_free = q->nb_desc - 1;
348 fifo_reset(&q->rs_tracker, (q->nb_desc + 1) / q->rs_thresh);
349 FM10K_PCI_REG_WRITE(q->tail_ptr, 0);
353 * clean queue, descriptor rings, free software buffers used when stopping
357 tx_queue_clean(struct fm10k_tx_queue *q)
359 struct fm10k_tx_desc zero = {0, 0, 0, 0, 0, 0};
361 PMD_INIT_FUNC_TRACE();
363 /* zero descriptor rings */
364 for (i = 0; i < q->nb_desc; ++i)
365 q->hw_ring[i] = zero;
367 /* free software buffers */
368 for (i = 0; i < q->nb_desc; ++i) {
370 rte_pktmbuf_free_seg(q->sw_ring[i]);
371 q->sw_ring[i] = NULL;
377 * free all queue memory used when releasing the queue (i.e. configure)
380 tx_queue_free(struct fm10k_tx_queue *q)
382 PMD_INIT_FUNC_TRACE();
384 PMD_INIT_LOG(DEBUG, "Freeing tx queue %p", q);
386 if (q->rs_tracker.list) {
387 rte_free(q->rs_tracker.list);
388 q->rs_tracker.list = NULL;
391 rte_free(q->sw_ring);
400 * disable TX queue, wait unitl HW finished necessary flush operation
403 tx_queue_disable(struct fm10k_hw *hw, uint16_t qnum)
407 reg = FM10K_READ_REG(hw, FM10K_TXDCTL(qnum));
408 FM10K_WRITE_REG(hw, FM10K_TXDCTL(qnum),
409 reg & ~FM10K_TXDCTL_ENABLE);
411 /* Wait 100us at most */
412 for (i = 0; i < FM10K_QUEUE_DISABLE_TIMEOUT; i++) {
414 reg = FM10K_READ_REG(hw, FM10K_TXDCTL(qnum));
415 if (!(reg & FM10K_TXDCTL_ENABLE))
419 if (i == FM10K_QUEUE_DISABLE_TIMEOUT)
426 fm10k_check_mq_mode(struct rte_eth_dev *dev)
428 enum rte_eth_rx_mq_mode rx_mq_mode = dev->data->dev_conf.rxmode.mq_mode;
429 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
430 struct rte_eth_vmdq_rx_conf *vmdq_conf;
431 uint16_t nb_rx_q = dev->data->nb_rx_queues;
433 vmdq_conf = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
435 if (rx_mq_mode & ETH_MQ_RX_DCB_FLAG) {
436 PMD_INIT_LOG(ERR, "DCB mode is not supported.");
440 if (!(rx_mq_mode & ETH_MQ_RX_VMDQ_FLAG))
443 if (hw->mac.type == fm10k_mac_vf) {
444 PMD_INIT_LOG(ERR, "VMDQ mode is not supported in VF.");
448 /* Check VMDQ queue pool number */
449 if (vmdq_conf->nb_queue_pools >
450 sizeof(vmdq_conf->pool_map[0].pools) * CHAR_BIT ||
451 vmdq_conf->nb_queue_pools > nb_rx_q) {
452 PMD_INIT_LOG(ERR, "Too many of queue pools: %d",
453 vmdq_conf->nb_queue_pools);
460 static const struct fm10k_txq_ops def_txq_ops = {
461 .reset = tx_queue_reset,
465 fm10k_dev_configure(struct rte_eth_dev *dev)
469 PMD_INIT_FUNC_TRACE();
471 if (dev->data->dev_conf.rxmode.hw_strip_crc == 0)
472 PMD_INIT_LOG(WARNING, "fm10k always strip CRC");
473 /* multipe queue mode checking */
474 ret = fm10k_check_mq_mode(dev);
476 PMD_DRV_LOG(ERR, "fm10k_check_mq_mode fails with %d.",
484 /* fls = find last set bit = 32 minus the number of leading zeros */
486 #define fls(x) (((x) == 0) ? 0 : (32 - __builtin_clz((x))))
490 fm10k_dev_vmdq_rx_configure(struct rte_eth_dev *dev)
492 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
493 struct rte_eth_vmdq_rx_conf *vmdq_conf;
496 vmdq_conf = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
498 for (i = 0; i < vmdq_conf->nb_pool_maps; i++) {
499 if (!vmdq_conf->pool_map[i].pools)
502 fm10k_update_vlan(hw, vmdq_conf->pool_map[i].vlan_id, 0, true);
503 fm10k_mbx_unlock(hw);
508 fm10k_dev_pf_main_vsi_reset(struct rte_eth_dev *dev)
510 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
512 /* Add default mac address */
513 fm10k_MAC_filter_set(dev, hw->mac.addr, true,
514 MAIN_VSI_POOL_NUMBER);
518 fm10k_dev_rss_configure(struct rte_eth_dev *dev)
520 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
521 struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
522 uint32_t mrqc, *key, i, reta, j;
525 #define RSS_KEY_SIZE 40
526 static uint8_t rss_intel_key[RSS_KEY_SIZE] = {
527 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
528 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
529 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
530 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
531 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
534 if (dev->data->nb_rx_queues == 1 ||
535 dev_conf->rxmode.mq_mode != ETH_MQ_RX_RSS ||
536 dev_conf->rx_adv_conf.rss_conf.rss_hf == 0) {
537 FM10K_WRITE_REG(hw, FM10K_MRQC(0), 0);
541 /* random key is rss_intel_key (default) or user provided (rss_key) */
542 if (dev_conf->rx_adv_conf.rss_conf.rss_key == NULL)
543 key = (uint32_t *)rss_intel_key;
545 key = (uint32_t *)dev_conf->rx_adv_conf.rss_conf.rss_key;
547 /* Now fill our hash function seeds, 4 bytes at a time */
548 for (i = 0; i < RSS_KEY_SIZE / sizeof(*key); ++i)
549 FM10K_WRITE_REG(hw, FM10K_RSSRK(0, i), key[i]);
552 * Fill in redirection table
553 * The byte-swap is needed because NIC registers are in
554 * little-endian order.
557 for (i = 0, j = 0; i < FM10K_MAX_RSS_INDICES; i++, j++) {
558 if (j == dev->data->nb_rx_queues)
560 reta = (reta << CHAR_BIT) | j;
562 FM10K_WRITE_REG(hw, FM10K_RETA(0, i >> 2),
567 * Generate RSS hash based on packet types, TCP/UDP
568 * port numbers and/or IPv4/v6 src and dst addresses
570 hf = dev_conf->rx_adv_conf.rss_conf.rss_hf;
572 mrqc |= (hf & ETH_RSS_IPV4) ? FM10K_MRQC_IPV4 : 0;
573 mrqc |= (hf & ETH_RSS_IPV6) ? FM10K_MRQC_IPV6 : 0;
574 mrqc |= (hf & ETH_RSS_IPV6_EX) ? FM10K_MRQC_IPV6 : 0;
575 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_TCP) ? FM10K_MRQC_TCP_IPV4 : 0;
576 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_TCP) ? FM10K_MRQC_TCP_IPV6 : 0;
577 mrqc |= (hf & ETH_RSS_IPV6_TCP_EX) ? FM10K_MRQC_TCP_IPV6 : 0;
578 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_UDP) ? FM10K_MRQC_UDP_IPV4 : 0;
579 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_UDP) ? FM10K_MRQC_UDP_IPV6 : 0;
580 mrqc |= (hf & ETH_RSS_IPV6_UDP_EX) ? FM10K_MRQC_UDP_IPV6 : 0;
583 PMD_INIT_LOG(ERR, "Specified RSS mode 0x%"PRIx64"is not"
588 FM10K_WRITE_REG(hw, FM10K_MRQC(0), mrqc);
592 fm10k_dev_logic_port_update(struct rte_eth_dev *dev, uint16_t nb_lport_new)
594 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
597 for (i = 0; i < nb_lport_new; i++) {
598 /* Set unicast mode by default. App can change
599 * to other mode in other API func.
602 hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map + i,
603 FM10K_XCAST_MODE_NONE);
604 fm10k_mbx_unlock(hw);
609 fm10k_dev_mq_rx_configure(struct rte_eth_dev *dev)
611 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
612 struct rte_eth_vmdq_rx_conf *vmdq_conf;
613 struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
614 struct fm10k_macvlan_filter_info *macvlan;
615 uint16_t nb_queue_pools = 0; /* pool number in configuration */
616 uint16_t nb_lport_new;
618 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
619 vmdq_conf = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
621 fm10k_dev_rss_configure(dev);
623 /* only PF supports VMDQ */
624 if (hw->mac.type != fm10k_mac_pf)
627 if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_VMDQ_FLAG)
628 nb_queue_pools = vmdq_conf->nb_queue_pools;
630 /* no pool number change, no need to update logic port and VLAN/MAC */
631 if (macvlan->nb_queue_pools == nb_queue_pools)
634 nb_lport_new = nb_queue_pools ? nb_queue_pools : 1;
635 fm10k_dev_logic_port_update(dev, nb_lport_new);
637 /* reset MAC/VLAN as it's based on VMDQ or PF main VSI */
638 memset(dev->data->mac_addrs, 0,
639 ETHER_ADDR_LEN * FM10K_MAX_MACADDR_NUM);
640 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
641 &dev->data->mac_addrs[0]);
642 memset(macvlan, 0, sizeof(*macvlan));
643 macvlan->nb_queue_pools = nb_queue_pools;
646 fm10k_dev_vmdq_rx_configure(dev);
648 fm10k_dev_pf_main_vsi_reset(dev);
652 fm10k_dev_tx_init(struct rte_eth_dev *dev)
654 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
656 struct fm10k_tx_queue *txq;
660 /* Disable TXINT to avoid possible interrupt */
661 for (i = 0; i < hw->mac.max_queues; i++)
662 FM10K_WRITE_REG(hw, FM10K_TXINT(i),
663 3 << FM10K_TXINT_TIMER_SHIFT);
666 for (i = 0; i < dev->data->nb_tx_queues; ++i) {
667 txq = dev->data->tx_queues[i];
668 base_addr = txq->hw_ring_phys_addr;
669 size = txq->nb_desc * sizeof(struct fm10k_tx_desc);
671 /* disable queue to avoid issues while updating state */
672 ret = tx_queue_disable(hw, i);
674 PMD_INIT_LOG(ERR, "failed to disable queue %d", i);
677 /* Enable use of FTAG bit in TX descriptor, PFVTCTL
678 * register is read-only for VF.
680 if (fm10k_check_ftag(dev->pci_dev->device.devargs)) {
681 if (hw->mac.type == fm10k_mac_pf) {
682 FM10K_WRITE_REG(hw, FM10K_PFVTCTL(i),
683 FM10K_PFVTCTL_FTAG_DESC_ENABLE);
684 PMD_INIT_LOG(DEBUG, "FTAG mode is enabled");
686 PMD_INIT_LOG(ERR, "VF FTAG is not supported.");
691 /* set location and size for descriptor ring */
692 FM10K_WRITE_REG(hw, FM10K_TDBAL(i),
693 base_addr & UINT64_LOWER_32BITS_MASK);
694 FM10K_WRITE_REG(hw, FM10K_TDBAH(i),
695 base_addr >> (CHAR_BIT * sizeof(uint32_t)));
696 FM10K_WRITE_REG(hw, FM10K_TDLEN(i), size);
698 /* assign default SGLORT for each TX queue by PF */
699 if (hw->mac.type == fm10k_mac_pf)
700 FM10K_WRITE_REG(hw, FM10K_TX_SGLORT(i), hw->mac.dglort_map);
703 /* set up vector or scalar TX function as appropriate */
704 fm10k_set_tx_function(dev);
710 fm10k_dev_rx_init(struct rte_eth_dev *dev)
712 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
713 struct fm10k_macvlan_filter_info *macvlan;
714 struct rte_pci_device *pdev = dev->pci_dev;
715 struct rte_intr_handle *intr_handle = &pdev->intr_handle;
717 struct fm10k_rx_queue *rxq;
720 uint32_t rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY;
721 uint32_t logic_port = hw->mac.dglort_map;
723 uint16_t queue_stride = 0;
725 /* enable RXINT for interrupt mode */
727 if (rte_intr_dp_is_en(intr_handle)) {
728 for (; i < dev->data->nb_rx_queues; i++) {
729 FM10K_WRITE_REG(hw, FM10K_RXINT(i), Q2V(pdev, i));
730 if (hw->mac.type == fm10k_mac_pf)
731 FM10K_WRITE_REG(hw, FM10K_ITR(Q2V(pdev, i)),
733 FM10K_ITR_MASK_CLEAR);
735 FM10K_WRITE_REG(hw, FM10K_VFITR(Q2V(pdev, i)),
737 FM10K_ITR_MASK_CLEAR);
740 /* Disable other RXINT to avoid possible interrupt */
741 for (; i < hw->mac.max_queues; i++)
742 FM10K_WRITE_REG(hw, FM10K_RXINT(i),
743 3 << FM10K_RXINT_TIMER_SHIFT);
745 /* Setup RX queues */
746 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
747 rxq = dev->data->rx_queues[i];
748 base_addr = rxq->hw_ring_phys_addr;
749 size = rxq->nb_desc * sizeof(union fm10k_rx_desc);
751 /* disable queue to avoid issues while updating state */
752 ret = rx_queue_disable(hw, i);
754 PMD_INIT_LOG(ERR, "failed to disable queue %d", i);
758 /* Setup the Base and Length of the Rx Descriptor Ring */
759 FM10K_WRITE_REG(hw, FM10K_RDBAL(i),
760 base_addr & UINT64_LOWER_32BITS_MASK);
761 FM10K_WRITE_REG(hw, FM10K_RDBAH(i),
762 base_addr >> (CHAR_BIT * sizeof(uint32_t)));
763 FM10K_WRITE_REG(hw, FM10K_RDLEN(i), size);
765 /* Configure the Rx buffer size for one buff without split */
766 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
767 RTE_PKTMBUF_HEADROOM);
768 /* As RX buffer is aligned to 512B within mbuf, some bytes are
769 * reserved for this purpose, and the worst case could be 511B.
770 * But SRR reg assumes all buffers have the same size. In order
771 * to fill the gap, we'll have to consider the worst case and
772 * assume 512B is reserved. If we don't do so, it's possible
773 * for HW to overwrite data to next mbuf.
775 buf_size -= FM10K_RX_DATABUF_ALIGN;
777 FM10K_WRITE_REG(hw, FM10K_SRRCTL(i),
778 (buf_size >> FM10K_SRRCTL_BSIZEPKT_SHIFT) |
779 FM10K_SRRCTL_LOOPBACK_SUPPRESS);
781 /* It adds dual VLAN length for supporting dual VLAN */
782 if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
783 2 * FM10K_VLAN_TAG_SIZE) > buf_size ||
784 dev->data->dev_conf.rxmode.enable_scatter) {
786 dev->data->scattered_rx = 1;
787 reg = FM10K_READ_REG(hw, FM10K_SRRCTL(i));
788 reg |= FM10K_SRRCTL_BUFFER_CHAINING_EN;
789 FM10K_WRITE_REG(hw, FM10K_SRRCTL(i), reg);
792 /* Enable drop on empty, it's RO for VF */
793 if (hw->mac.type == fm10k_mac_pf && rxq->drop_en)
794 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY;
796 FM10K_WRITE_REG(hw, FM10K_RXDCTL(i), rxdctl);
797 FM10K_WRITE_FLUSH(hw);
800 /* Configure VMDQ/RSS if applicable */
801 fm10k_dev_mq_rx_configure(dev);
803 /* Decide the best RX function */
804 fm10k_set_rx_function(dev);
806 /* update RX_SGLORT for loopback suppress*/
807 if (hw->mac.type != fm10k_mac_pf)
809 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
810 if (macvlan->nb_queue_pools)
811 queue_stride = dev->data->nb_rx_queues / macvlan->nb_queue_pools;
812 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
813 if (i && queue_stride && !(i % queue_stride))
815 FM10K_WRITE_REG(hw, FM10K_RX_SGLORT(i), logic_port);
822 fm10k_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
824 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
827 struct fm10k_rx_queue *rxq;
829 PMD_INIT_FUNC_TRACE();
831 if (rx_queue_id < dev->data->nb_rx_queues) {
832 rxq = dev->data->rx_queues[rx_queue_id];
833 err = rx_queue_reset(rxq);
834 if (err == -ENOMEM) {
835 PMD_INIT_LOG(ERR, "Failed to alloc memory : %d", err);
837 } else if (err == -EINVAL) {
838 PMD_INIT_LOG(ERR, "Invalid buffer address alignment :"
843 /* Setup the HW Rx Head and Tail Descriptor Pointers
844 * Note: this must be done AFTER the queue is enabled on real
845 * hardware, but BEFORE the queue is enabled when using the
846 * emulation platform. Do it in both places for now and remove
847 * this comment and the following two register writes when the
848 * emulation platform is no longer being used.
850 FM10K_WRITE_REG(hw, FM10K_RDH(rx_queue_id), 0);
851 FM10K_WRITE_REG(hw, FM10K_RDT(rx_queue_id), rxq->nb_desc - 1);
853 /* Set PF ownership flag for PF devices */
854 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(rx_queue_id));
855 if (hw->mac.type == fm10k_mac_pf)
856 reg |= FM10K_RXQCTL_PF;
857 reg |= FM10K_RXQCTL_ENABLE;
858 /* enable RX queue */
859 FM10K_WRITE_REG(hw, FM10K_RXQCTL(rx_queue_id), reg);
860 FM10K_WRITE_FLUSH(hw);
862 /* Setup the HW Rx Head and Tail Descriptor Pointers
863 * Note: this must be done AFTER the queue is enabled
865 FM10K_WRITE_REG(hw, FM10K_RDH(rx_queue_id), 0);
866 FM10K_WRITE_REG(hw, FM10K_RDT(rx_queue_id), rxq->nb_desc - 1);
867 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
874 fm10k_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
876 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
878 PMD_INIT_FUNC_TRACE();
880 if (rx_queue_id < dev->data->nb_rx_queues) {
881 /* Disable RX queue */
882 rx_queue_disable(hw, rx_queue_id);
884 /* Free mbuf and clean HW ring */
885 rx_queue_clean(dev->data->rx_queues[rx_queue_id]);
886 dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
893 fm10k_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
895 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
896 /** @todo - this should be defined in the shared code */
897 #define FM10K_TXDCTL_WRITE_BACK_MIN_DELAY 0x00010000
898 uint32_t txdctl = FM10K_TXDCTL_WRITE_BACK_MIN_DELAY;
901 PMD_INIT_FUNC_TRACE();
903 if (tx_queue_id < dev->data->nb_tx_queues) {
904 struct fm10k_tx_queue *q = dev->data->tx_queues[tx_queue_id];
908 /* reset head and tail pointers */
909 FM10K_WRITE_REG(hw, FM10K_TDH(tx_queue_id), 0);
910 FM10K_WRITE_REG(hw, FM10K_TDT(tx_queue_id), 0);
912 /* enable TX queue */
913 FM10K_WRITE_REG(hw, FM10K_TXDCTL(tx_queue_id),
914 FM10K_TXDCTL_ENABLE | txdctl);
915 FM10K_WRITE_FLUSH(hw);
916 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
924 fm10k_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
926 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
928 PMD_INIT_FUNC_TRACE();
930 if (tx_queue_id < dev->data->nb_tx_queues) {
931 tx_queue_disable(hw, tx_queue_id);
932 tx_queue_clean(dev->data->tx_queues[tx_queue_id]);
933 dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
939 static inline int fm10k_glort_valid(struct fm10k_hw *hw)
941 return ((hw->mac.dglort_map & FM10K_DGLORTMAP_NONE)
942 != FM10K_DGLORTMAP_NONE);
946 fm10k_dev_promiscuous_enable(struct rte_eth_dev *dev)
948 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
951 PMD_INIT_FUNC_TRACE();
953 /* Return if it didn't acquire valid glort range */
954 if ((hw->mac.type == fm10k_mac_pf) && !fm10k_glort_valid(hw))
958 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
959 FM10K_XCAST_MODE_PROMISC);
960 fm10k_mbx_unlock(hw);
962 if (status != FM10K_SUCCESS)
963 PMD_INIT_LOG(ERR, "Failed to enable promiscuous mode");
967 fm10k_dev_promiscuous_disable(struct rte_eth_dev *dev)
969 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
973 PMD_INIT_FUNC_TRACE();
975 /* Return if it didn't acquire valid glort range */
976 if ((hw->mac.type == fm10k_mac_pf) && !fm10k_glort_valid(hw))
979 if (dev->data->all_multicast == 1)
980 mode = FM10K_XCAST_MODE_ALLMULTI;
982 mode = FM10K_XCAST_MODE_NONE;
985 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
987 fm10k_mbx_unlock(hw);
989 if (status != FM10K_SUCCESS)
990 PMD_INIT_LOG(ERR, "Failed to disable promiscuous mode");
994 fm10k_dev_allmulticast_enable(struct rte_eth_dev *dev)
996 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
999 PMD_INIT_FUNC_TRACE();
1001 /* Return if it didn't acquire valid glort range */
1002 if ((hw->mac.type == fm10k_mac_pf) && !fm10k_glort_valid(hw))
1005 /* If promiscuous mode is enabled, it doesn't make sense to enable
1006 * allmulticast and disable promiscuous since fm10k only can select
1009 if (dev->data->promiscuous) {
1010 PMD_INIT_LOG(INFO, "Promiscuous mode is enabled, "\
1011 "needn't enable allmulticast");
1016 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
1017 FM10K_XCAST_MODE_ALLMULTI);
1018 fm10k_mbx_unlock(hw);
1020 if (status != FM10K_SUCCESS)
1021 PMD_INIT_LOG(ERR, "Failed to enable allmulticast mode");
1025 fm10k_dev_allmulticast_disable(struct rte_eth_dev *dev)
1027 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1030 PMD_INIT_FUNC_TRACE();
1032 /* Return if it didn't acquire valid glort range */
1033 if ((hw->mac.type == fm10k_mac_pf) && !fm10k_glort_valid(hw))
1036 if (dev->data->promiscuous) {
1037 PMD_INIT_LOG(ERR, "Failed to disable allmulticast mode "\
1038 "since promisc mode is enabled");
1043 /* Change mode to unicast mode */
1044 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
1045 FM10K_XCAST_MODE_NONE);
1046 fm10k_mbx_unlock(hw);
1048 if (status != FM10K_SUCCESS)
1049 PMD_INIT_LOG(ERR, "Failed to disable allmulticast mode");
1053 fm10k_dev_dglort_map_configure(struct rte_eth_dev *dev)
1055 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1056 uint32_t dglortdec, pool_len, rss_len, i, dglortmask;
1057 uint16_t nb_queue_pools;
1058 struct fm10k_macvlan_filter_info *macvlan;
1060 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1061 nb_queue_pools = macvlan->nb_queue_pools;
1062 pool_len = nb_queue_pools ? fls(nb_queue_pools - 1) : 0;
1063 rss_len = fls(dev->data->nb_rx_queues - 1) - pool_len;
1065 /* GLORT 0x0-0x3F are used by PF and VMDQ, 0x40-0x7F used by FD */
1066 dglortdec = (rss_len << FM10K_DGLORTDEC_RSSLENGTH_SHIFT) | pool_len;
1067 dglortmask = (GLORT_PF_MASK << FM10K_DGLORTMAP_MASK_SHIFT) |
1069 FM10K_WRITE_REG(hw, FM10K_DGLORTMAP(0), dglortmask);
1070 /* Configure VMDQ/RSS DGlort Decoder */
1071 FM10K_WRITE_REG(hw, FM10K_DGLORTDEC(0), dglortdec);
1073 /* Flow Director configurations, only queue number is valid. */
1074 dglortdec = fls(dev->data->nb_rx_queues - 1);
1075 dglortmask = (GLORT_FD_MASK << FM10K_DGLORTMAP_MASK_SHIFT) |
1076 (hw->mac.dglort_map + GLORT_FD_Q_BASE);
1077 FM10K_WRITE_REG(hw, FM10K_DGLORTMAP(1), dglortmask);
1078 FM10K_WRITE_REG(hw, FM10K_DGLORTDEC(1), dglortdec);
1080 /* Invalidate all other GLORT entries */
1081 for (i = 2; i < FM10K_DGLORT_COUNT; i++)
1082 FM10K_WRITE_REG(hw, FM10K_DGLORTMAP(i),
1083 FM10K_DGLORTMAP_NONE);
1086 #define BSIZEPKT_ROUNDUP ((1 << FM10K_SRRCTL_BSIZEPKT_SHIFT) - 1)
1088 fm10k_dev_start(struct rte_eth_dev *dev)
1090 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1093 PMD_INIT_FUNC_TRACE();
1095 /* stop, init, then start the hw */
1096 diag = fm10k_stop_hw(hw);
1097 if (diag != FM10K_SUCCESS) {
1098 PMD_INIT_LOG(ERR, "Hardware stop failed: %d", diag);
1102 diag = fm10k_init_hw(hw);
1103 if (diag != FM10K_SUCCESS) {
1104 PMD_INIT_LOG(ERR, "Hardware init failed: %d", diag);
1108 diag = fm10k_start_hw(hw);
1109 if (diag != FM10K_SUCCESS) {
1110 PMD_INIT_LOG(ERR, "Hardware start failed: %d", diag);
1114 diag = fm10k_dev_tx_init(dev);
1116 PMD_INIT_LOG(ERR, "TX init failed: %d", diag);
1120 if (fm10k_dev_rxq_interrupt_setup(dev))
1123 diag = fm10k_dev_rx_init(dev);
1125 PMD_INIT_LOG(ERR, "RX init failed: %d", diag);
1129 if (hw->mac.type == fm10k_mac_pf)
1130 fm10k_dev_dglort_map_configure(dev);
1132 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1133 struct fm10k_rx_queue *rxq;
1134 rxq = dev->data->rx_queues[i];
1136 if (rxq->rx_deferred_start)
1138 diag = fm10k_dev_rx_queue_start(dev, i);
1141 for (j = 0; j < i; ++j)
1142 rx_queue_clean(dev->data->rx_queues[j]);
1147 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1148 struct fm10k_tx_queue *txq;
1149 txq = dev->data->tx_queues[i];
1151 if (txq->tx_deferred_start)
1153 diag = fm10k_dev_tx_queue_start(dev, i);
1156 for (j = 0; j < i; ++j)
1157 tx_queue_clean(dev->data->tx_queues[j]);
1158 for (j = 0; j < dev->data->nb_rx_queues; ++j)
1159 rx_queue_clean(dev->data->rx_queues[j]);
1164 /* Update default vlan when not in VMDQ mode */
1165 if (!(dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_VMDQ_FLAG))
1166 fm10k_vlan_filter_set(dev, hw->mac.default_vid, true);
1172 fm10k_dev_stop(struct rte_eth_dev *dev)
1174 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1175 struct rte_pci_device *pdev = dev->pci_dev;
1176 struct rte_intr_handle *intr_handle = &pdev->intr_handle;
1179 PMD_INIT_FUNC_TRACE();
1181 if (dev->data->tx_queues)
1182 for (i = 0; i < dev->data->nb_tx_queues; i++)
1183 fm10k_dev_tx_queue_stop(dev, i);
1185 if (dev->data->rx_queues)
1186 for (i = 0; i < dev->data->nb_rx_queues; i++)
1187 fm10k_dev_rx_queue_stop(dev, i);
1189 /* Disable datapath event */
1190 if (rte_intr_dp_is_en(intr_handle)) {
1191 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1192 FM10K_WRITE_REG(hw, FM10K_RXINT(i),
1193 3 << FM10K_RXINT_TIMER_SHIFT);
1194 if (hw->mac.type == fm10k_mac_pf)
1195 FM10K_WRITE_REG(hw, FM10K_ITR(Q2V(pdev, i)),
1196 FM10K_ITR_MASK_SET);
1198 FM10K_WRITE_REG(hw, FM10K_VFITR(Q2V(pdev, i)),
1199 FM10K_ITR_MASK_SET);
1202 /* Clean datapath event and queue/vec mapping */
1203 rte_intr_efd_disable(intr_handle);
1204 rte_free(intr_handle->intr_vec);
1205 intr_handle->intr_vec = NULL;
1209 fm10k_dev_queue_release(struct rte_eth_dev *dev)
1213 PMD_INIT_FUNC_TRACE();
1215 if (dev->data->tx_queues) {
1216 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1217 struct fm10k_tx_queue *txq = dev->data->tx_queues[i];
1223 if (dev->data->rx_queues) {
1224 for (i = 0; i < dev->data->nb_rx_queues; i++)
1225 fm10k_rx_queue_release(dev->data->rx_queues[i]);
1230 fm10k_dev_close(struct rte_eth_dev *dev)
1232 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1234 PMD_INIT_FUNC_TRACE();
1237 hw->mac.ops.update_lport_state(hw, hw->mac.dglort_map,
1238 MAX_LPORT_NUM, false);
1239 fm10k_mbx_unlock(hw);
1241 /* allow 10ms for device to quiesce */
1242 rte_delay_us(FM10K_SWITCH_QUIESCE_US);
1244 /* Stop mailbox service first */
1245 fm10k_close_mbx_service(hw);
1246 fm10k_dev_stop(dev);
1247 fm10k_dev_queue_release(dev);
1252 fm10k_link_update(struct rte_eth_dev *dev,
1253 __rte_unused int wait_to_complete)
1255 PMD_INIT_FUNC_TRACE();
1257 /* The host-interface link is always up. The speed is ~50Gbps per Gen3
1258 * x8 PCIe interface. For now, we leave the speed undefined since there
1259 * is no 50Gbps Ethernet. */
1260 dev->data->dev_link.link_speed = 0;
1261 dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
1262 dev->data->dev_link.link_status = ETH_LINK_UP;
1267 static int fm10k_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
1268 struct rte_eth_xstat_name *xstats_names, __rte_unused unsigned limit)
1273 if (xstats_names != NULL) {
1274 /* Note: limit checked in rte_eth_xstats_names() */
1277 for (i = 0; i < FM10K_NB_HW_XSTATS; i++) {
1278 snprintf(xstats_names[count].name,
1279 sizeof(xstats_names[count].name),
1280 "%s", fm10k_hw_stats_strings[count].name);
1284 /* PF queue stats */
1285 for (q = 0; q < FM10K_MAX_QUEUES_PF; q++) {
1286 for (i = 0; i < FM10K_NB_RX_Q_XSTATS; i++) {
1287 snprintf(xstats_names[count].name,
1288 sizeof(xstats_names[count].name),
1290 fm10k_hw_stats_rx_q_strings[i].name);
1293 for (i = 0; i < FM10K_NB_TX_Q_XSTATS; i++) {
1294 snprintf(xstats_names[count].name,
1295 sizeof(xstats_names[count].name),
1297 fm10k_hw_stats_tx_q_strings[i].name);
1302 return FM10K_NB_XSTATS;
1306 fm10k_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1309 struct fm10k_hw_stats *hw_stats =
1310 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1311 unsigned i, q, count = 0;
1313 if (n < FM10K_NB_XSTATS)
1314 return FM10K_NB_XSTATS;
1317 for (i = 0; i < FM10K_NB_HW_XSTATS; i++) {
1318 xstats[count].value = *(uint64_t *)(((char *)hw_stats) +
1319 fm10k_hw_stats_strings[count].offset);
1323 /* PF queue stats */
1324 for (q = 0; q < FM10K_MAX_QUEUES_PF; q++) {
1325 for (i = 0; i < FM10K_NB_RX_Q_XSTATS; i++) {
1326 xstats[count].value =
1327 *(uint64_t *)(((char *)&hw_stats->q[q]) +
1328 fm10k_hw_stats_rx_q_strings[i].offset);
1331 for (i = 0; i < FM10K_NB_TX_Q_XSTATS; i++) {
1332 xstats[count].value =
1333 *(uint64_t *)(((char *)&hw_stats->q[q]) +
1334 fm10k_hw_stats_tx_q_strings[i].offset);
1339 return FM10K_NB_XSTATS;
1343 fm10k_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1345 uint64_t ipackets, opackets, ibytes, obytes;
1346 struct fm10k_hw *hw =
1347 FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1348 struct fm10k_hw_stats *hw_stats =
1349 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1352 PMD_INIT_FUNC_TRACE();
1354 fm10k_update_hw_stats(hw, hw_stats);
1356 ipackets = opackets = ibytes = obytes = 0;
1357 for (i = 0; (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) &&
1358 (i < hw->mac.max_queues); ++i) {
1359 stats->q_ipackets[i] = hw_stats->q[i].rx_packets.count;
1360 stats->q_opackets[i] = hw_stats->q[i].tx_packets.count;
1361 stats->q_ibytes[i] = hw_stats->q[i].rx_bytes.count;
1362 stats->q_obytes[i] = hw_stats->q[i].tx_bytes.count;
1363 ipackets += stats->q_ipackets[i];
1364 opackets += stats->q_opackets[i];
1365 ibytes += stats->q_ibytes[i];
1366 obytes += stats->q_obytes[i];
1368 stats->ipackets = ipackets;
1369 stats->opackets = opackets;
1370 stats->ibytes = ibytes;
1371 stats->obytes = obytes;
1375 fm10k_stats_reset(struct rte_eth_dev *dev)
1377 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1378 struct fm10k_hw_stats *hw_stats =
1379 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1381 PMD_INIT_FUNC_TRACE();
1383 memset(hw_stats, 0, sizeof(*hw_stats));
1384 fm10k_rebind_hw_stats(hw, hw_stats);
1388 fm10k_dev_infos_get(struct rte_eth_dev *dev,
1389 struct rte_eth_dev_info *dev_info)
1391 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1392 struct rte_pci_device *pdev = dev->pci_dev;
1394 PMD_INIT_FUNC_TRACE();
1396 dev_info->min_rx_bufsize = FM10K_MIN_RX_BUF_SIZE;
1397 dev_info->max_rx_pktlen = FM10K_MAX_PKT_SIZE;
1398 dev_info->max_rx_queues = hw->mac.max_queues;
1399 dev_info->max_tx_queues = hw->mac.max_queues;
1400 dev_info->max_mac_addrs = FM10K_MAX_MACADDR_NUM;
1401 dev_info->max_hash_mac_addrs = 0;
1402 dev_info->max_vfs = pdev->max_vfs;
1403 dev_info->vmdq_pool_base = 0;
1404 dev_info->vmdq_queue_base = 0;
1405 dev_info->max_vmdq_pools = ETH_32_POOLS;
1406 dev_info->vmdq_queue_num = FM10K_MAX_QUEUES_PF;
1407 dev_info->rx_offload_capa =
1408 DEV_RX_OFFLOAD_VLAN_STRIP |
1409 DEV_RX_OFFLOAD_IPV4_CKSUM |
1410 DEV_RX_OFFLOAD_UDP_CKSUM |
1411 DEV_RX_OFFLOAD_TCP_CKSUM;
1412 dev_info->tx_offload_capa =
1413 DEV_TX_OFFLOAD_VLAN_INSERT |
1414 DEV_TX_OFFLOAD_IPV4_CKSUM |
1415 DEV_TX_OFFLOAD_UDP_CKSUM |
1416 DEV_TX_OFFLOAD_TCP_CKSUM |
1417 DEV_TX_OFFLOAD_TCP_TSO;
1419 dev_info->hash_key_size = FM10K_RSSRK_SIZE * sizeof(uint32_t);
1420 dev_info->reta_size = FM10K_MAX_RSS_INDICES;
1422 dev_info->default_rxconf = (struct rte_eth_rxconf) {
1424 .pthresh = FM10K_DEFAULT_RX_PTHRESH,
1425 .hthresh = FM10K_DEFAULT_RX_HTHRESH,
1426 .wthresh = FM10K_DEFAULT_RX_WTHRESH,
1428 .rx_free_thresh = FM10K_RX_FREE_THRESH_DEFAULT(0),
1432 dev_info->default_txconf = (struct rte_eth_txconf) {
1434 .pthresh = FM10K_DEFAULT_TX_PTHRESH,
1435 .hthresh = FM10K_DEFAULT_TX_HTHRESH,
1436 .wthresh = FM10K_DEFAULT_TX_WTHRESH,
1438 .tx_free_thresh = FM10K_TX_FREE_THRESH_DEFAULT(0),
1439 .tx_rs_thresh = FM10K_TX_RS_THRESH_DEFAULT(0),
1440 .txq_flags = FM10K_SIMPLE_TX_FLAG,
1443 dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
1444 .nb_max = FM10K_MAX_RX_DESC,
1445 .nb_min = FM10K_MIN_RX_DESC,
1446 .nb_align = FM10K_MULT_RX_DESC,
1449 dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
1450 .nb_max = FM10K_MAX_TX_DESC,
1451 .nb_min = FM10K_MIN_TX_DESC,
1452 .nb_align = FM10K_MULT_TX_DESC,
1455 dev_info->speed_capa = ETH_LINK_SPEED_1G | ETH_LINK_SPEED_2_5G |
1456 ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
1457 ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G;
1460 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
1461 static const uint32_t *
1462 fm10k_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1464 if (dev->rx_pkt_burst == fm10k_recv_pkts ||
1465 dev->rx_pkt_burst == fm10k_recv_scattered_pkts) {
1466 static uint32_t ptypes[] = {
1467 /* refers to rx_desc_to_ol_flags() */
1470 RTE_PTYPE_L3_IPV4_EXT,
1472 RTE_PTYPE_L3_IPV6_EXT,
1479 } else if (dev->rx_pkt_burst == fm10k_recv_pkts_vec ||
1480 dev->rx_pkt_burst == fm10k_recv_scattered_pkts_vec) {
1481 static uint32_t ptypes_vec[] = {
1482 /* refers to fm10k_desc_to_pktype_v() */
1484 RTE_PTYPE_L3_IPV4_EXT,
1486 RTE_PTYPE_L3_IPV6_EXT,
1489 RTE_PTYPE_TUNNEL_GENEVE,
1490 RTE_PTYPE_TUNNEL_NVGRE,
1491 RTE_PTYPE_TUNNEL_VXLAN,
1492 RTE_PTYPE_TUNNEL_GRE,
1502 static const uint32_t *
1503 fm10k_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1510 fm10k_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1513 uint16_t mac_num = 0;
1514 uint32_t vid_idx, vid_bit, mac_index;
1515 struct fm10k_hw *hw;
1516 struct fm10k_macvlan_filter_info *macvlan;
1517 struct rte_eth_dev_data *data = dev->data;
1519 hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1520 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1522 if (macvlan->nb_queue_pools > 0) { /* VMDQ mode */
1523 PMD_INIT_LOG(ERR, "Cannot change VLAN filter in VMDQ mode");
1527 if (vlan_id > ETH_VLAN_ID_MAX) {
1528 PMD_INIT_LOG(ERR, "Invalid vlan_id: must be < 4096");
1532 vid_idx = FM10K_VFTA_IDX(vlan_id);
1533 vid_bit = FM10K_VFTA_BIT(vlan_id);
1534 /* this VLAN ID is already in the VLAN filter table, return SUCCESS */
1535 if (on && (macvlan->vfta[vid_idx] & vid_bit))
1537 /* this VLAN ID is NOT in the VLAN filter table, cannot remove */
1538 if (!on && !(macvlan->vfta[vid_idx] & vid_bit)) {
1539 PMD_INIT_LOG(ERR, "Invalid vlan_id: not existing "
1540 "in the VLAN filter table");
1545 result = fm10k_update_vlan(hw, vlan_id, 0, on);
1546 fm10k_mbx_unlock(hw);
1547 if (result != FM10K_SUCCESS) {
1548 PMD_INIT_LOG(ERR, "VLAN update failed: %d", result);
1552 for (mac_index = 0; (mac_index < FM10K_MAX_MACADDR_NUM) &&
1553 (result == FM10K_SUCCESS); mac_index++) {
1554 if (is_zero_ether_addr(&data->mac_addrs[mac_index]))
1556 if (mac_num > macvlan->mac_num - 1) {
1557 PMD_INIT_LOG(ERR, "MAC address number "
1562 result = fm10k_update_uc_addr(hw, hw->mac.dglort_map,
1563 data->mac_addrs[mac_index].addr_bytes,
1565 fm10k_mbx_unlock(hw);
1568 if (result != FM10K_SUCCESS) {
1569 PMD_INIT_LOG(ERR, "MAC address update failed: %d", result);
1574 macvlan->vlan_num++;
1575 macvlan->vfta[vid_idx] |= vid_bit;
1577 macvlan->vlan_num--;
1578 macvlan->vfta[vid_idx] &= ~vid_bit;
1584 fm10k_vlan_offload_set(__rte_unused struct rte_eth_dev *dev, int mask)
1586 if (mask & ETH_VLAN_STRIP_MASK) {
1587 if (!dev->data->dev_conf.rxmode.hw_vlan_strip)
1588 PMD_INIT_LOG(ERR, "VLAN stripping is "
1589 "always on in fm10k");
1592 if (mask & ETH_VLAN_EXTEND_MASK) {
1593 if (dev->data->dev_conf.rxmode.hw_vlan_extend)
1594 PMD_INIT_LOG(ERR, "VLAN QinQ is not "
1595 "supported in fm10k");
1598 if (mask & ETH_VLAN_FILTER_MASK) {
1599 if (!dev->data->dev_conf.rxmode.hw_vlan_filter)
1600 PMD_INIT_LOG(ERR, "VLAN filter is always on in fm10k");
1604 /* Add/Remove a MAC address, and update filters to main VSI */
1605 static void fm10k_MAC_filter_set_main_vsi(struct rte_eth_dev *dev,
1606 const u8 *mac, bool add, uint32_t pool)
1608 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1609 struct fm10k_macvlan_filter_info *macvlan;
1612 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1614 if (pool != MAIN_VSI_POOL_NUMBER) {
1615 PMD_DRV_LOG(ERR, "VMDQ not enabled, can't set "
1616 "mac to pool %u", pool);
1619 for (i = 0, j = 0; j < FM10K_VFTA_SIZE; j++) {
1620 if (!macvlan->vfta[j])
1622 for (k = 0; k < FM10K_UINT32_BIT_SIZE; k++) {
1623 if (!(macvlan->vfta[j] & (1 << k)))
1625 if (i + 1 > macvlan->vlan_num) {
1626 PMD_INIT_LOG(ERR, "vlan number not match");
1630 fm10k_update_uc_addr(hw, hw->mac.dglort_map, mac,
1631 j * FM10K_UINT32_BIT_SIZE + k, add, 0);
1632 fm10k_mbx_unlock(hw);
1638 /* Add/Remove a MAC address, and update filters to VMDQ */
1639 static void fm10k_MAC_filter_set_vmdq(struct rte_eth_dev *dev,
1640 const u8 *mac, bool add, uint32_t pool)
1642 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1643 struct fm10k_macvlan_filter_info *macvlan;
1644 struct rte_eth_vmdq_rx_conf *vmdq_conf;
1647 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1648 vmdq_conf = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
1650 if (pool > macvlan->nb_queue_pools) {
1651 PMD_DRV_LOG(ERR, "Pool number %u invalid."
1653 pool, macvlan->nb_queue_pools);
1656 for (i = 0; i < vmdq_conf->nb_pool_maps; i++) {
1657 if (!(vmdq_conf->pool_map[i].pools & (1UL << pool)))
1660 fm10k_update_uc_addr(hw, hw->mac.dglort_map + pool, mac,
1661 vmdq_conf->pool_map[i].vlan_id, add, 0);
1662 fm10k_mbx_unlock(hw);
1666 /* Add/Remove a MAC address, and update filters */
1667 static void fm10k_MAC_filter_set(struct rte_eth_dev *dev,
1668 const u8 *mac, bool add, uint32_t pool)
1670 struct fm10k_macvlan_filter_info *macvlan;
1672 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1674 if (macvlan->nb_queue_pools > 0) /* VMDQ mode */
1675 fm10k_MAC_filter_set_vmdq(dev, mac, add, pool);
1677 fm10k_MAC_filter_set_main_vsi(dev, mac, add, pool);
1685 /* Add a MAC address, and update filters */
1687 fm10k_macaddr_add(struct rte_eth_dev *dev,
1688 struct ether_addr *mac_addr,
1692 struct fm10k_macvlan_filter_info *macvlan;
1694 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1695 fm10k_MAC_filter_set(dev, mac_addr->addr_bytes, TRUE, pool);
1696 macvlan->mac_vmdq_id[index] = pool;
1699 /* Remove a MAC address, and update filters */
1701 fm10k_macaddr_remove(struct rte_eth_dev *dev, uint32_t index)
1703 struct rte_eth_dev_data *data = dev->data;
1704 struct fm10k_macvlan_filter_info *macvlan;
1706 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1707 fm10k_MAC_filter_set(dev, data->mac_addrs[index].addr_bytes,
1708 FALSE, macvlan->mac_vmdq_id[index]);
1709 macvlan->mac_vmdq_id[index] = 0;
1713 check_nb_desc(uint16_t min, uint16_t max, uint16_t mult, uint16_t request)
1715 if ((request < min) || (request > max) || ((request % mult) != 0))
1723 check_thresh(uint16_t min, uint16_t max, uint16_t div, uint16_t request)
1725 if ((request < min) || (request > max) || ((div % request) != 0))
1732 handle_rxconf(struct fm10k_rx_queue *q, const struct rte_eth_rxconf *conf)
1734 uint16_t rx_free_thresh;
1736 if (conf->rx_free_thresh == 0)
1737 rx_free_thresh = FM10K_RX_FREE_THRESH_DEFAULT(q);
1739 rx_free_thresh = conf->rx_free_thresh;
1741 /* make sure the requested threshold satisfies the constraints */
1742 if (check_thresh(FM10K_RX_FREE_THRESH_MIN(q),
1743 FM10K_RX_FREE_THRESH_MAX(q),
1744 FM10K_RX_FREE_THRESH_DIV(q),
1746 PMD_INIT_LOG(ERR, "rx_free_thresh (%u) must be "
1747 "less than or equal to %u, "
1748 "greater than or equal to %u, "
1749 "and a divisor of %u",
1750 rx_free_thresh, FM10K_RX_FREE_THRESH_MAX(q),
1751 FM10K_RX_FREE_THRESH_MIN(q),
1752 FM10K_RX_FREE_THRESH_DIV(q));
1756 q->alloc_thresh = rx_free_thresh;
1757 q->drop_en = conf->rx_drop_en;
1758 q->rx_deferred_start = conf->rx_deferred_start;
1764 * Hardware requires specific alignment for Rx packet buffers. At
1765 * least one of the following two conditions must be satisfied.
1766 * 1. Address is 512B aligned
1767 * 2. Address is 8B aligned and buffer does not cross 4K boundary.
1769 * As such, the driver may need to adjust the DMA address within the
1770 * buffer by up to 512B.
1772 * return 1 if the element size is valid, otherwise return 0.
1775 mempool_element_size_valid(struct rte_mempool *mp)
1779 /* elt_size includes mbuf header and headroom */
1780 min_size = mp->elt_size - sizeof(struct rte_mbuf) -
1781 RTE_PKTMBUF_HEADROOM;
1783 /* account for up to 512B of alignment */
1784 min_size -= FM10K_RX_DATABUF_ALIGN;
1786 /* sanity check for overflow */
1787 if (min_size > mp->elt_size)
1795 fm10k_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
1796 uint16_t nb_desc, unsigned int socket_id,
1797 const struct rte_eth_rxconf *conf, struct rte_mempool *mp)
1799 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1800 struct fm10k_dev_info *dev_info = FM10K_DEV_PRIVATE_TO_INFO(dev);
1801 struct fm10k_rx_queue *q;
1802 const struct rte_memzone *mz;
1804 PMD_INIT_FUNC_TRACE();
1806 /* make sure the mempool element size can account for alignment. */
1807 if (!mempool_element_size_valid(mp)) {
1808 PMD_INIT_LOG(ERR, "Error : Mempool element size is too small");
1812 /* make sure a valid number of descriptors have been requested */
1813 if (check_nb_desc(FM10K_MIN_RX_DESC, FM10K_MAX_RX_DESC,
1814 FM10K_MULT_RX_DESC, nb_desc)) {
1815 PMD_INIT_LOG(ERR, "Number of Rx descriptors (%u) must be "
1816 "less than or equal to %"PRIu32", "
1817 "greater than or equal to %u, "
1818 "and a multiple of %u",
1819 nb_desc, (uint32_t)FM10K_MAX_RX_DESC, FM10K_MIN_RX_DESC,
1820 FM10K_MULT_RX_DESC);
1825 * if this queue existed already, free the associated memory. The
1826 * queue cannot be reused in case we need to allocate memory on
1827 * different socket than was previously used.
1829 if (dev->data->rx_queues[queue_id] != NULL) {
1830 rx_queue_free(dev->data->rx_queues[queue_id]);
1831 dev->data->rx_queues[queue_id] = NULL;
1834 /* allocate memory for the queue structure */
1835 q = rte_zmalloc_socket("fm10k", sizeof(*q), RTE_CACHE_LINE_SIZE,
1838 PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
1844 q->nb_desc = nb_desc;
1845 q->nb_fake_desc = FM10K_MULT_RX_DESC;
1846 q->port_id = dev->data->port_id;
1847 q->queue_id = queue_id;
1848 q->tail_ptr = (volatile uint32_t *)
1849 &((uint32_t *)hw->hw_addr)[FM10K_RDT(queue_id)];
1850 if (handle_rxconf(q, conf))
1853 /* allocate memory for the software ring */
1854 q->sw_ring = rte_zmalloc_socket("fm10k sw ring",
1855 (nb_desc + q->nb_fake_desc) * sizeof(struct rte_mbuf *),
1856 RTE_CACHE_LINE_SIZE, socket_id);
1857 if (q->sw_ring == NULL) {
1858 PMD_INIT_LOG(ERR, "Cannot allocate software ring");
1864 * allocate memory for the hardware descriptor ring. A memzone large
1865 * enough to hold the maximum ring size is requested to allow for
1866 * resizing in later calls to the queue setup function.
1868 mz = rte_eth_dma_zone_reserve(dev, "rx_ring", queue_id,
1869 FM10K_MAX_RX_RING_SZ, FM10K_ALIGN_RX_DESC,
1872 PMD_INIT_LOG(ERR, "Cannot allocate hardware ring");
1873 rte_free(q->sw_ring);
1877 q->hw_ring = mz->addr;
1878 q->hw_ring_phys_addr = rte_mem_phy2mch(mz->memseg_id, mz->phys_addr);
1880 /* Check if number of descs satisfied Vector requirement */
1881 if (!rte_is_power_of_2(nb_desc)) {
1882 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Vector Rx "
1883 "preconditions - canceling the feature for "
1884 "the whole port[%d]",
1885 q->queue_id, q->port_id);
1886 dev_info->rx_vec_allowed = false;
1888 fm10k_rxq_vec_setup(q);
1890 dev->data->rx_queues[queue_id] = q;
1895 fm10k_rx_queue_release(void *queue)
1897 PMD_INIT_FUNC_TRACE();
1899 rx_queue_free(queue);
1903 handle_txconf(struct fm10k_tx_queue *q, const struct rte_eth_txconf *conf)
1905 uint16_t tx_free_thresh;
1906 uint16_t tx_rs_thresh;
1908 /* constraint MACROs require that tx_free_thresh is configured
1909 * before tx_rs_thresh */
1910 if (conf->tx_free_thresh == 0)
1911 tx_free_thresh = FM10K_TX_FREE_THRESH_DEFAULT(q);
1913 tx_free_thresh = conf->tx_free_thresh;
1915 /* make sure the requested threshold satisfies the constraints */
1916 if (check_thresh(FM10K_TX_FREE_THRESH_MIN(q),
1917 FM10K_TX_FREE_THRESH_MAX(q),
1918 FM10K_TX_FREE_THRESH_DIV(q),
1920 PMD_INIT_LOG(ERR, "tx_free_thresh (%u) must be "
1921 "less than or equal to %u, "
1922 "greater than or equal to %u, "
1923 "and a divisor of %u",
1924 tx_free_thresh, FM10K_TX_FREE_THRESH_MAX(q),
1925 FM10K_TX_FREE_THRESH_MIN(q),
1926 FM10K_TX_FREE_THRESH_DIV(q));
1930 q->free_thresh = tx_free_thresh;
1932 if (conf->tx_rs_thresh == 0)
1933 tx_rs_thresh = FM10K_TX_RS_THRESH_DEFAULT(q);
1935 tx_rs_thresh = conf->tx_rs_thresh;
1937 q->tx_deferred_start = conf->tx_deferred_start;
1939 /* make sure the requested threshold satisfies the constraints */
1940 if (check_thresh(FM10K_TX_RS_THRESH_MIN(q),
1941 FM10K_TX_RS_THRESH_MAX(q),
1942 FM10K_TX_RS_THRESH_DIV(q),
1944 PMD_INIT_LOG(ERR, "tx_rs_thresh (%u) must be "
1945 "less than or equal to %u, "
1946 "greater than or equal to %u, "
1947 "and a divisor of %u",
1948 tx_rs_thresh, FM10K_TX_RS_THRESH_MAX(q),
1949 FM10K_TX_RS_THRESH_MIN(q),
1950 FM10K_TX_RS_THRESH_DIV(q));
1954 q->rs_thresh = tx_rs_thresh;
1960 fm10k_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
1961 uint16_t nb_desc, unsigned int socket_id,
1962 const struct rte_eth_txconf *conf)
1964 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1965 struct fm10k_tx_queue *q;
1966 const struct rte_memzone *mz;
1968 PMD_INIT_FUNC_TRACE();
1970 /* make sure a valid number of descriptors have been requested */
1971 if (check_nb_desc(FM10K_MIN_TX_DESC, FM10K_MAX_TX_DESC,
1972 FM10K_MULT_TX_DESC, nb_desc)) {
1973 PMD_INIT_LOG(ERR, "Number of Tx descriptors (%u) must be "
1974 "less than or equal to %"PRIu32", "
1975 "greater than or equal to %u, "
1976 "and a multiple of %u",
1977 nb_desc, (uint32_t)FM10K_MAX_TX_DESC, FM10K_MIN_TX_DESC,
1978 FM10K_MULT_TX_DESC);
1983 * if this queue existed already, free the associated memory. The
1984 * queue cannot be reused in case we need to allocate memory on
1985 * different socket than was previously used.
1987 if (dev->data->tx_queues[queue_id] != NULL) {
1988 struct fm10k_tx_queue *txq = dev->data->tx_queues[queue_id];
1991 dev->data->tx_queues[queue_id] = NULL;
1994 /* allocate memory for the queue structure */
1995 q = rte_zmalloc_socket("fm10k", sizeof(*q), RTE_CACHE_LINE_SIZE,
1998 PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
2003 q->nb_desc = nb_desc;
2004 q->port_id = dev->data->port_id;
2005 q->queue_id = queue_id;
2006 q->txq_flags = conf->txq_flags;
2007 q->ops = &def_txq_ops;
2008 q->tail_ptr = (volatile uint32_t *)
2009 &((uint32_t *)hw->hw_addr)[FM10K_TDT(queue_id)];
2010 if (handle_txconf(q, conf))
2013 /* allocate memory for the software ring */
2014 q->sw_ring = rte_zmalloc_socket("fm10k sw ring",
2015 nb_desc * sizeof(struct rte_mbuf *),
2016 RTE_CACHE_LINE_SIZE, socket_id);
2017 if (q->sw_ring == NULL) {
2018 PMD_INIT_LOG(ERR, "Cannot allocate software ring");
2024 * allocate memory for the hardware descriptor ring. A memzone large
2025 * enough to hold the maximum ring size is requested to allow for
2026 * resizing in later calls to the queue setup function.
2028 mz = rte_eth_dma_zone_reserve(dev, "tx_ring", queue_id,
2029 FM10K_MAX_TX_RING_SZ, FM10K_ALIGN_TX_DESC,
2032 PMD_INIT_LOG(ERR, "Cannot allocate hardware ring");
2033 rte_free(q->sw_ring);
2037 q->hw_ring = mz->addr;
2038 q->hw_ring_phys_addr = rte_mem_phy2mch(mz->memseg_id, mz->phys_addr);
2041 * allocate memory for the RS bit tracker. Enough slots to hold the
2042 * descriptor index for each RS bit needing to be set are required.
2044 q->rs_tracker.list = rte_zmalloc_socket("fm10k rs tracker",
2045 ((nb_desc + 1) / q->rs_thresh) *
2047 RTE_CACHE_LINE_SIZE, socket_id);
2048 if (q->rs_tracker.list == NULL) {
2049 PMD_INIT_LOG(ERR, "Cannot allocate RS bit tracker");
2050 rte_free(q->sw_ring);
2055 dev->data->tx_queues[queue_id] = q;
2060 fm10k_tx_queue_release(void *queue)
2062 struct fm10k_tx_queue *q = queue;
2063 PMD_INIT_FUNC_TRACE();
2069 fm10k_reta_update(struct rte_eth_dev *dev,
2070 struct rte_eth_rss_reta_entry64 *reta_conf,
2073 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2074 uint16_t i, j, idx, shift;
2078 PMD_INIT_FUNC_TRACE();
2080 if (reta_size > FM10K_MAX_RSS_INDICES) {
2081 PMD_INIT_LOG(ERR, "The size of hash lookup table configured "
2082 "(%d) doesn't match the number hardware can supported "
2083 "(%d)", reta_size, FM10K_MAX_RSS_INDICES);
2088 * Update Redirection Table RETA[n], n=0..31. The redirection table has
2089 * 128-entries in 32 registers
2091 for (i = 0; i < FM10K_MAX_RSS_INDICES; i += CHARS_PER_UINT32) {
2092 idx = i / RTE_RETA_GROUP_SIZE;
2093 shift = i % RTE_RETA_GROUP_SIZE;
2094 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2095 BIT_MASK_PER_UINT32);
2100 if (mask != BIT_MASK_PER_UINT32)
2101 reta = FM10K_READ_REG(hw, FM10K_RETA(0, i >> 2));
2103 for (j = 0; j < CHARS_PER_UINT32; j++) {
2104 if (mask & (0x1 << j)) {
2106 reta &= ~(UINT8_MAX << CHAR_BIT * j);
2107 reta |= reta_conf[idx].reta[shift + j] <<
2111 FM10K_WRITE_REG(hw, FM10K_RETA(0, i >> 2), reta);
2118 fm10k_reta_query(struct rte_eth_dev *dev,
2119 struct rte_eth_rss_reta_entry64 *reta_conf,
2122 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2123 uint16_t i, j, idx, shift;
2127 PMD_INIT_FUNC_TRACE();
2129 if (reta_size < FM10K_MAX_RSS_INDICES) {
2130 PMD_INIT_LOG(ERR, "The size of hash lookup table configured "
2131 "(%d) doesn't match the number hardware can supported "
2132 "(%d)", reta_size, FM10K_MAX_RSS_INDICES);
2137 * Read Redirection Table RETA[n], n=0..31. The redirection table has
2138 * 128-entries in 32 registers
2140 for (i = 0; i < FM10K_MAX_RSS_INDICES; i += CHARS_PER_UINT32) {
2141 idx = i / RTE_RETA_GROUP_SIZE;
2142 shift = i % RTE_RETA_GROUP_SIZE;
2143 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2144 BIT_MASK_PER_UINT32);
2148 reta = FM10K_READ_REG(hw, FM10K_RETA(0, i >> 2));
2149 for (j = 0; j < CHARS_PER_UINT32; j++) {
2150 if (mask & (0x1 << j))
2151 reta_conf[idx].reta[shift + j] = ((reta >>
2152 CHAR_BIT * j) & UINT8_MAX);
2160 fm10k_rss_hash_update(struct rte_eth_dev *dev,
2161 struct rte_eth_rss_conf *rss_conf)
2163 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2164 uint32_t *key = (uint32_t *)rss_conf->rss_key;
2166 uint64_t hf = rss_conf->rss_hf;
2169 PMD_INIT_FUNC_TRACE();
2171 if (key && (rss_conf->rss_key_len < FM10K_RSSRK_SIZE *
2172 FM10K_RSSRK_ENTRIES_PER_REG))
2179 mrqc |= (hf & ETH_RSS_IPV4) ? FM10K_MRQC_IPV4 : 0;
2180 mrqc |= (hf & ETH_RSS_IPV6) ? FM10K_MRQC_IPV6 : 0;
2181 mrqc |= (hf & ETH_RSS_IPV6_EX) ? FM10K_MRQC_IPV6 : 0;
2182 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_TCP) ? FM10K_MRQC_TCP_IPV4 : 0;
2183 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_TCP) ? FM10K_MRQC_TCP_IPV6 : 0;
2184 mrqc |= (hf & ETH_RSS_IPV6_TCP_EX) ? FM10K_MRQC_TCP_IPV6 : 0;
2185 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_UDP) ? FM10K_MRQC_UDP_IPV4 : 0;
2186 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_UDP) ? FM10K_MRQC_UDP_IPV6 : 0;
2187 mrqc |= (hf & ETH_RSS_IPV6_UDP_EX) ? FM10K_MRQC_UDP_IPV6 : 0;
2189 /* If the mapping doesn't fit any supported, return */
2194 for (i = 0; i < FM10K_RSSRK_SIZE; ++i)
2195 FM10K_WRITE_REG(hw, FM10K_RSSRK(0, i), key[i]);
2197 FM10K_WRITE_REG(hw, FM10K_MRQC(0), mrqc);
2203 fm10k_rss_hash_conf_get(struct rte_eth_dev *dev,
2204 struct rte_eth_rss_conf *rss_conf)
2206 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2207 uint32_t *key = (uint32_t *)rss_conf->rss_key;
2212 PMD_INIT_FUNC_TRACE();
2214 if (key && (rss_conf->rss_key_len < FM10K_RSSRK_SIZE *
2215 FM10K_RSSRK_ENTRIES_PER_REG))
2219 for (i = 0; i < FM10K_RSSRK_SIZE; ++i)
2220 key[i] = FM10K_READ_REG(hw, FM10K_RSSRK(0, i));
2222 mrqc = FM10K_READ_REG(hw, FM10K_MRQC(0));
2224 hf |= (mrqc & FM10K_MRQC_IPV4) ? ETH_RSS_IPV4 : 0;
2225 hf |= (mrqc & FM10K_MRQC_IPV6) ? ETH_RSS_IPV6 : 0;
2226 hf |= (mrqc & FM10K_MRQC_IPV6) ? ETH_RSS_IPV6_EX : 0;
2227 hf |= (mrqc & FM10K_MRQC_TCP_IPV4) ? ETH_RSS_NONFRAG_IPV4_TCP : 0;
2228 hf |= (mrqc & FM10K_MRQC_TCP_IPV6) ? ETH_RSS_NONFRAG_IPV6_TCP : 0;
2229 hf |= (mrqc & FM10K_MRQC_TCP_IPV6) ? ETH_RSS_IPV6_TCP_EX : 0;
2230 hf |= (mrqc & FM10K_MRQC_UDP_IPV4) ? ETH_RSS_NONFRAG_IPV4_UDP : 0;
2231 hf |= (mrqc & FM10K_MRQC_UDP_IPV6) ? ETH_RSS_NONFRAG_IPV6_UDP : 0;
2232 hf |= (mrqc & FM10K_MRQC_UDP_IPV6) ? ETH_RSS_IPV6_UDP_EX : 0;
2234 rss_conf->rss_hf = hf;
2240 fm10k_dev_enable_intr_pf(struct rte_eth_dev *dev)
2242 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2243 uint32_t int_map = FM10K_INT_MAP_IMMEDIATE;
2245 /* Bind all local non-queue interrupt to vector 0 */
2246 int_map |= FM10K_MISC_VEC_ID;
2248 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_mailbox), int_map);
2249 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_pcie_fault), int_map);
2250 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_up_down), int_map);
2251 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_event), int_map);
2252 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_sram), int_map);
2253 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_vflr), int_map);
2255 /* Enable misc causes */
2256 FM10K_WRITE_REG(hw, FM10K_EIMR, FM10K_EIMR_ENABLE(PCA_FAULT) |
2257 FM10K_EIMR_ENABLE(THI_FAULT) |
2258 FM10K_EIMR_ENABLE(FUM_FAULT) |
2259 FM10K_EIMR_ENABLE(MAILBOX) |
2260 FM10K_EIMR_ENABLE(SWITCHREADY) |
2261 FM10K_EIMR_ENABLE(SWITCHNOTREADY) |
2262 FM10K_EIMR_ENABLE(SRAMERROR) |
2263 FM10K_EIMR_ENABLE(VFLR));
2266 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_AUTOMASK |
2267 FM10K_ITR_MASK_CLEAR);
2268 FM10K_WRITE_FLUSH(hw);
2272 fm10k_dev_disable_intr_pf(struct rte_eth_dev *dev)
2274 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2275 uint32_t int_map = FM10K_INT_MAP_DISABLE;
2277 int_map |= FM10K_MISC_VEC_ID;
2279 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_mailbox), int_map);
2280 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_pcie_fault), int_map);
2281 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_up_down), int_map);
2282 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_event), int_map);
2283 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_sram), int_map);
2284 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_vflr), int_map);
2286 /* Disable misc causes */
2287 FM10K_WRITE_REG(hw, FM10K_EIMR, FM10K_EIMR_DISABLE(PCA_FAULT) |
2288 FM10K_EIMR_DISABLE(THI_FAULT) |
2289 FM10K_EIMR_DISABLE(FUM_FAULT) |
2290 FM10K_EIMR_DISABLE(MAILBOX) |
2291 FM10K_EIMR_DISABLE(SWITCHREADY) |
2292 FM10K_EIMR_DISABLE(SWITCHNOTREADY) |
2293 FM10K_EIMR_DISABLE(SRAMERROR) |
2294 FM10K_EIMR_DISABLE(VFLR));
2297 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_MASK_SET);
2298 FM10K_WRITE_FLUSH(hw);
2302 fm10k_dev_enable_intr_vf(struct rte_eth_dev *dev)
2304 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2305 uint32_t int_map = FM10K_INT_MAP_IMMEDIATE;
2307 /* Bind all local non-queue interrupt to vector 0 */
2308 int_map |= FM10K_MISC_VEC_ID;
2310 /* Only INT 0 available, other 15 are reserved. */
2311 FM10K_WRITE_REG(hw, FM10K_VFINT_MAP, int_map);
2314 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_AUTOMASK |
2315 FM10K_ITR_MASK_CLEAR);
2316 FM10K_WRITE_FLUSH(hw);
2320 fm10k_dev_disable_intr_vf(struct rte_eth_dev *dev)
2322 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2323 uint32_t int_map = FM10K_INT_MAP_DISABLE;
2325 int_map |= FM10K_MISC_VEC_ID;
2327 /* Only INT 0 available, other 15 are reserved. */
2328 FM10K_WRITE_REG(hw, FM10K_VFINT_MAP, int_map);
2331 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_MASK_SET);
2332 FM10K_WRITE_FLUSH(hw);
2336 fm10k_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
2338 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2339 struct rte_pci_device *pdev = dev->pci_dev;
2342 if (hw->mac.type == fm10k_mac_pf)
2343 FM10K_WRITE_REG(hw, FM10K_ITR(Q2V(pdev, queue_id)),
2344 FM10K_ITR_AUTOMASK | FM10K_ITR_MASK_CLEAR);
2346 FM10K_WRITE_REG(hw, FM10K_VFITR(Q2V(pdev, queue_id)),
2347 FM10K_ITR_AUTOMASK | FM10K_ITR_MASK_CLEAR);
2348 rte_intr_enable(&pdev->intr_handle);
2353 fm10k_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
2355 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2356 struct rte_pci_device *pdev = dev->pci_dev;
2359 if (hw->mac.type == fm10k_mac_pf)
2360 FM10K_WRITE_REG(hw, FM10K_ITR(Q2V(pdev, queue_id)),
2361 FM10K_ITR_MASK_SET);
2363 FM10K_WRITE_REG(hw, FM10K_VFITR(Q2V(pdev, queue_id)),
2364 FM10K_ITR_MASK_SET);
2369 fm10k_dev_rxq_interrupt_setup(struct rte_eth_dev *dev)
2371 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2372 struct rte_pci_device *pdev = dev->pci_dev;
2373 struct rte_intr_handle *intr_handle = &pdev->intr_handle;
2374 uint32_t intr_vector, vec;
2378 /* fm10k needs one separate interrupt for mailbox,
2379 * so only drivers which support multiple interrupt vectors
2380 * e.g. vfio-pci can work for fm10k interrupt mode
2382 if (!rte_intr_cap_multiple(intr_handle) ||
2383 dev->data->dev_conf.intr_conf.rxq == 0)
2386 intr_vector = dev->data->nb_rx_queues;
2388 /* disable interrupt first */
2389 rte_intr_disable(intr_handle);
2390 if (hw->mac.type == fm10k_mac_pf)
2391 fm10k_dev_disable_intr_pf(dev);
2393 fm10k_dev_disable_intr_vf(dev);
2395 if (rte_intr_efd_enable(intr_handle, intr_vector)) {
2396 PMD_INIT_LOG(ERR, "Failed to init event fd");
2400 if (rte_intr_dp_is_en(intr_handle) && !result) {
2401 intr_handle->intr_vec = rte_zmalloc("intr_vec",
2402 dev->data->nb_rx_queues * sizeof(int), 0);
2403 if (intr_handle->intr_vec) {
2404 for (queue_id = 0, vec = FM10K_RX_VEC_START;
2405 queue_id < dev->data->nb_rx_queues;
2407 intr_handle->intr_vec[queue_id] = vec;
2408 if (vec < intr_handle->nb_efd - 1
2409 + FM10K_RX_VEC_START)
2413 PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
2414 " intr_vec", dev->data->nb_rx_queues);
2415 rte_intr_efd_disable(intr_handle);
2420 if (hw->mac.type == fm10k_mac_pf)
2421 fm10k_dev_enable_intr_pf(dev);
2423 fm10k_dev_enable_intr_vf(dev);
2424 rte_intr_enable(intr_handle);
2425 hw->mac.ops.update_int_moderator(hw);
2430 fm10k_dev_handle_fault(struct fm10k_hw *hw, uint32_t eicr)
2432 struct fm10k_fault fault;
2434 const char *estr = "Unknown error";
2436 /* Process PCA fault */
2437 if (eicr & FM10K_EICR_PCA_FAULT) {
2438 err = fm10k_get_fault(hw, FM10K_PCA_FAULT, &fault);
2441 switch (fault.type) {
2443 estr = "PCA_NO_FAULT"; break;
2444 case PCA_UNMAPPED_ADDR:
2445 estr = "PCA_UNMAPPED_ADDR"; break;
2446 case PCA_BAD_QACCESS_PF:
2447 estr = "PCA_BAD_QACCESS_PF"; break;
2448 case PCA_BAD_QACCESS_VF:
2449 estr = "PCA_BAD_QACCESS_VF"; break;
2450 case PCA_MALICIOUS_REQ:
2451 estr = "PCA_MALICIOUS_REQ"; break;
2452 case PCA_POISONED_TLP:
2453 estr = "PCA_POISONED_TLP"; break;
2455 estr = "PCA_TLP_ABORT"; break;
2459 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
2460 estr, fault.func ? "VF" : "PF", fault.func,
2461 fault.address, fault.specinfo);
2464 /* Process THI fault */
2465 if (eicr & FM10K_EICR_THI_FAULT) {
2466 err = fm10k_get_fault(hw, FM10K_THI_FAULT, &fault);
2469 switch (fault.type) {
2471 estr = "THI_NO_FAULT"; break;
2472 case THI_MAL_DIS_Q_FAULT:
2473 estr = "THI_MAL_DIS_Q_FAULT"; break;
2477 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
2478 estr, fault.func ? "VF" : "PF", fault.func,
2479 fault.address, fault.specinfo);
2482 /* Process FUM fault */
2483 if (eicr & FM10K_EICR_FUM_FAULT) {
2484 err = fm10k_get_fault(hw, FM10K_FUM_FAULT, &fault);
2487 switch (fault.type) {
2489 estr = "FUM_NO_FAULT"; break;
2490 case FUM_UNMAPPED_ADDR:
2491 estr = "FUM_UNMAPPED_ADDR"; break;
2492 case FUM_POISONED_TLP:
2493 estr = "FUM_POISONED_TLP"; break;
2494 case FUM_BAD_VF_QACCESS:
2495 estr = "FUM_BAD_VF_QACCESS"; break;
2496 case FUM_ADD_DECODE_ERR:
2497 estr = "FUM_ADD_DECODE_ERR"; break;
2499 estr = "FUM_RO_ERROR"; break;
2500 case FUM_QPRC_CRC_ERROR:
2501 estr = "FUM_QPRC_CRC_ERROR"; break;
2502 case FUM_CSR_TIMEOUT:
2503 estr = "FUM_CSR_TIMEOUT"; break;
2504 case FUM_INVALID_TYPE:
2505 estr = "FUM_INVALID_TYPE"; break;
2506 case FUM_INVALID_LENGTH:
2507 estr = "FUM_INVALID_LENGTH"; break;
2508 case FUM_INVALID_BE:
2509 estr = "FUM_INVALID_BE"; break;
2510 case FUM_INVALID_ALIGN:
2511 estr = "FUM_INVALID_ALIGN"; break;
2515 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
2516 estr, fault.func ? "VF" : "PF", fault.func,
2517 fault.address, fault.specinfo);
2522 PMD_INIT_LOG(ERR, "Failed to handle fault event.");
2527 * PF interrupt handler triggered by NIC for handling specific interrupt.
2530 * Pointer to interrupt handle.
2532 * The address of parameter (struct rte_eth_dev *) regsitered before.
2538 fm10k_dev_interrupt_handler_pf(
2539 struct rte_intr_handle *handle,
2542 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2543 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2544 uint32_t cause, status;
2546 if (hw->mac.type != fm10k_mac_pf)
2549 cause = FM10K_READ_REG(hw, FM10K_EICR);
2551 /* Handle PCI fault cases */
2552 if (cause & FM10K_EICR_FAULT_MASK) {
2553 PMD_INIT_LOG(ERR, "INT: find fault!");
2554 fm10k_dev_handle_fault(hw, cause);
2557 /* Handle switch up/down */
2558 if (cause & FM10K_EICR_SWITCHNOTREADY)
2559 PMD_INIT_LOG(ERR, "INT: Switch is not ready");
2561 if (cause & FM10K_EICR_SWITCHREADY)
2562 PMD_INIT_LOG(INFO, "INT: Switch is ready");
2564 /* Handle mailbox message */
2566 hw->mbx.ops.process(hw, &hw->mbx);
2567 fm10k_mbx_unlock(hw);
2569 /* Handle SRAM error */
2570 if (cause & FM10K_EICR_SRAMERROR) {
2571 PMD_INIT_LOG(ERR, "INT: SRAM error on PEP");
2573 status = FM10K_READ_REG(hw, FM10K_SRAM_IP);
2574 /* Write to clear pending bits */
2575 FM10K_WRITE_REG(hw, FM10K_SRAM_IP, status);
2577 /* Todo: print out error message after shared code updates */
2580 /* Clear these 3 events if having any */
2581 cause &= FM10K_EICR_SWITCHNOTREADY | FM10K_EICR_MAILBOX |
2582 FM10K_EICR_SWITCHREADY;
2584 FM10K_WRITE_REG(hw, FM10K_EICR, cause);
2586 /* Re-enable interrupt from device side */
2587 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_AUTOMASK |
2588 FM10K_ITR_MASK_CLEAR);
2589 /* Re-enable interrupt from host side */
2590 rte_intr_enable(handle);
2594 * VF interrupt handler triggered by NIC for handling specific interrupt.
2597 * Pointer to interrupt handle.
2599 * The address of parameter (struct rte_eth_dev *) regsitered before.
2605 fm10k_dev_interrupt_handler_vf(
2606 struct rte_intr_handle *handle,
2609 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2610 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2612 if (hw->mac.type != fm10k_mac_vf)
2615 /* Handle mailbox message if lock is acquired */
2617 hw->mbx.ops.process(hw, &hw->mbx);
2618 fm10k_mbx_unlock(hw);
2620 /* Re-enable interrupt from device side */
2621 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_AUTOMASK |
2622 FM10K_ITR_MASK_CLEAR);
2623 /* Re-enable interrupt from host side */
2624 rte_intr_enable(handle);
2627 /* Mailbox message handler in VF */
2628 static const struct fm10k_msg_data fm10k_msgdata_vf[] = {
2629 FM10K_TLV_MSG_TEST_HANDLER(fm10k_tlv_msg_test),
2630 FM10K_VF_MSG_MAC_VLAN_HANDLER(fm10k_msg_mac_vlan_vf),
2631 FM10K_VF_MSG_LPORT_STATE_HANDLER(fm10k_msg_lport_state_vf),
2632 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_tlv_msg_error),
2636 fm10k_setup_mbx_service(struct fm10k_hw *hw)
2640 /* Initialize mailbox lock */
2641 fm10k_mbx_initlock(hw);
2643 /* Replace default message handler with new ones */
2644 if (hw->mac.type == fm10k_mac_vf)
2645 err = hw->mbx.ops.register_handlers(&hw->mbx, fm10k_msgdata_vf);
2648 PMD_INIT_LOG(ERR, "Failed to register mailbox handler.err:%d",
2652 /* Connect to SM for PF device or PF for VF device */
2653 return hw->mbx.ops.connect(hw, &hw->mbx);
2657 fm10k_close_mbx_service(struct fm10k_hw *hw)
2659 /* Disconnect from SM for PF device or PF for VF device */
2660 hw->mbx.ops.disconnect(hw, &hw->mbx);
2663 static const struct eth_dev_ops fm10k_eth_dev_ops = {
2664 .dev_configure = fm10k_dev_configure,
2665 .dev_start = fm10k_dev_start,
2666 .dev_stop = fm10k_dev_stop,
2667 .dev_close = fm10k_dev_close,
2668 .promiscuous_enable = fm10k_dev_promiscuous_enable,
2669 .promiscuous_disable = fm10k_dev_promiscuous_disable,
2670 .allmulticast_enable = fm10k_dev_allmulticast_enable,
2671 .allmulticast_disable = fm10k_dev_allmulticast_disable,
2672 .stats_get = fm10k_stats_get,
2673 .xstats_get = fm10k_xstats_get,
2674 .xstats_get_names = fm10k_xstats_get_names,
2675 .stats_reset = fm10k_stats_reset,
2676 .xstats_reset = fm10k_stats_reset,
2677 .link_update = fm10k_link_update,
2678 .dev_infos_get = fm10k_dev_infos_get,
2679 .dev_supported_ptypes_get = fm10k_dev_supported_ptypes_get,
2680 .vlan_filter_set = fm10k_vlan_filter_set,
2681 .vlan_offload_set = fm10k_vlan_offload_set,
2682 .mac_addr_add = fm10k_macaddr_add,
2683 .mac_addr_remove = fm10k_macaddr_remove,
2684 .rx_queue_start = fm10k_dev_rx_queue_start,
2685 .rx_queue_stop = fm10k_dev_rx_queue_stop,
2686 .tx_queue_start = fm10k_dev_tx_queue_start,
2687 .tx_queue_stop = fm10k_dev_tx_queue_stop,
2688 .rx_queue_setup = fm10k_rx_queue_setup,
2689 .rx_queue_release = fm10k_rx_queue_release,
2690 .tx_queue_setup = fm10k_tx_queue_setup,
2691 .tx_queue_release = fm10k_tx_queue_release,
2692 .rx_descriptor_done = fm10k_dev_rx_descriptor_done,
2693 .rx_queue_intr_enable = fm10k_dev_rx_queue_intr_enable,
2694 .rx_queue_intr_disable = fm10k_dev_rx_queue_intr_disable,
2695 .reta_update = fm10k_reta_update,
2696 .reta_query = fm10k_reta_query,
2697 .rss_hash_update = fm10k_rss_hash_update,
2698 .rss_hash_conf_get = fm10k_rss_hash_conf_get,
2701 static int ftag_check_handler(__rte_unused const char *key,
2702 const char *value, __rte_unused void *opaque)
2704 if (strcmp(value, "1"))
2711 fm10k_check_ftag(struct rte_devargs *devargs)
2713 struct rte_kvargs *kvlist;
2714 const char *ftag_key = "enable_ftag";
2716 if (devargs == NULL)
2719 kvlist = rte_kvargs_parse(devargs->args, NULL);
2723 if (!rte_kvargs_count(kvlist, ftag_key)) {
2724 rte_kvargs_free(kvlist);
2727 /* FTAG is enabled when there's key-value pair: enable_ftag=1 */
2728 if (rte_kvargs_process(kvlist, ftag_key,
2729 ftag_check_handler, NULL) < 0) {
2730 rte_kvargs_free(kvlist);
2733 rte_kvargs_free(kvlist);
2738 static void __attribute__((cold))
2739 fm10k_set_tx_function(struct rte_eth_dev *dev)
2741 struct fm10k_tx_queue *txq;
2744 uint16_t tx_ftag_en = 0;
2746 if (fm10k_check_ftag(dev->pci_dev->device.devargs))
2749 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2750 txq = dev->data->tx_queues[i];
2751 txq->tx_ftag_en = tx_ftag_en;
2752 /* Check if Vector Tx is satisfied */
2753 if (fm10k_tx_vec_condition_check(txq))
2758 PMD_INIT_LOG(DEBUG, "Use vector Tx func");
2759 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2760 txq = dev->data->tx_queues[i];
2761 fm10k_txq_vec_setup(txq);
2763 dev->tx_pkt_burst = fm10k_xmit_pkts_vec;
2765 dev->tx_pkt_burst = fm10k_xmit_pkts;
2766 PMD_INIT_LOG(DEBUG, "Use regular Tx func");
2770 static void __attribute__((cold))
2771 fm10k_set_rx_function(struct rte_eth_dev *dev)
2773 struct fm10k_dev_info *dev_info = FM10K_DEV_PRIVATE_TO_INFO(dev);
2774 uint16_t i, rx_using_sse;
2775 uint16_t rx_ftag_en = 0;
2777 if (fm10k_check_ftag(dev->pci_dev->device.devargs))
2780 /* In order to allow Vector Rx there are a few configuration
2781 * conditions to be met.
2783 if (!fm10k_rx_vec_condition_check(dev) &&
2784 dev_info->rx_vec_allowed && !rx_ftag_en) {
2785 if (dev->data->scattered_rx)
2786 dev->rx_pkt_burst = fm10k_recv_scattered_pkts_vec;
2788 dev->rx_pkt_burst = fm10k_recv_pkts_vec;
2789 } else if (dev->data->scattered_rx)
2790 dev->rx_pkt_burst = fm10k_recv_scattered_pkts;
2792 dev->rx_pkt_burst = fm10k_recv_pkts;
2795 (dev->rx_pkt_burst == fm10k_recv_scattered_pkts_vec ||
2796 dev->rx_pkt_burst == fm10k_recv_pkts_vec);
2799 PMD_INIT_LOG(DEBUG, "Use vector Rx func");
2801 PMD_INIT_LOG(DEBUG, "Use regular Rx func");
2803 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2804 struct fm10k_rx_queue *rxq = dev->data->rx_queues[i];
2806 rxq->rx_using_sse = rx_using_sse;
2807 rxq->rx_ftag_en = rx_ftag_en;
2812 fm10k_params_init(struct rte_eth_dev *dev)
2814 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2815 struct fm10k_dev_info *info = FM10K_DEV_PRIVATE_TO_INFO(dev);
2817 /* Inialize bus info. Normally we would call fm10k_get_bus_info(), but
2818 * there is no way to get link status without reading BAR4. Until this
2819 * works, assume we have maximum bandwidth.
2820 * @todo - fix bus info
2822 hw->bus_caps.speed = fm10k_bus_speed_8000;
2823 hw->bus_caps.width = fm10k_bus_width_pcie_x8;
2824 hw->bus_caps.payload = fm10k_bus_payload_512;
2825 hw->bus.speed = fm10k_bus_speed_8000;
2826 hw->bus.width = fm10k_bus_width_pcie_x8;
2827 hw->bus.payload = fm10k_bus_payload_256;
2829 info->rx_vec_allowed = true;
2833 eth_fm10k_dev_init(struct rte_eth_dev *dev)
2835 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2836 struct rte_pci_device *pdev = dev->pci_dev;
2837 struct rte_intr_handle *intr_handle = &pdev->intr_handle;
2839 struct fm10k_macvlan_filter_info *macvlan;
2841 PMD_INIT_FUNC_TRACE();
2843 dev->dev_ops = &fm10k_eth_dev_ops;
2844 dev->rx_pkt_burst = &fm10k_recv_pkts;
2845 dev->tx_pkt_burst = &fm10k_xmit_pkts;
2847 /* only initialize in the primary process */
2848 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2851 rte_eth_copy_pci_info(dev, pdev);
2853 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
2854 memset(macvlan, 0, sizeof(*macvlan));
2855 /* Vendor and Device ID need to be set before init of shared code */
2856 memset(hw, 0, sizeof(*hw));
2857 hw->device_id = pdev->id.device_id;
2858 hw->vendor_id = pdev->id.vendor_id;
2859 hw->subsystem_device_id = pdev->id.subsystem_device_id;
2860 hw->subsystem_vendor_id = pdev->id.subsystem_vendor_id;
2861 hw->revision_id = 0;
2862 hw->hw_addr = (void *)pdev->mem_resource[0].addr;
2863 if (hw->hw_addr == NULL) {
2864 PMD_INIT_LOG(ERR, "Bad mem resource."
2865 " Try to blacklist unused devices.");
2869 /* Store fm10k_adapter pointer */
2870 hw->back = dev->data->dev_private;
2872 /* Initialize the shared code */
2873 diag = fm10k_init_shared_code(hw);
2874 if (diag != FM10K_SUCCESS) {
2875 PMD_INIT_LOG(ERR, "Shared code init failed: %d", diag);
2879 /* Initialize parameters */
2880 fm10k_params_init(dev);
2882 /* Initialize the hw */
2883 diag = fm10k_init_hw(hw);
2884 if (diag != FM10K_SUCCESS) {
2885 PMD_INIT_LOG(ERR, "Hardware init failed: %d", diag);
2889 /* Initialize MAC address(es) */
2890 dev->data->mac_addrs = rte_zmalloc("fm10k",
2891 ETHER_ADDR_LEN * FM10K_MAX_MACADDR_NUM, 0);
2892 if (dev->data->mac_addrs == NULL) {
2893 PMD_INIT_LOG(ERR, "Cannot allocate memory for MAC addresses");
2897 diag = fm10k_read_mac_addr(hw);
2899 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
2900 &dev->data->mac_addrs[0]);
2902 if (diag != FM10K_SUCCESS ||
2903 !is_valid_assigned_ether_addr(dev->data->mac_addrs)) {
2905 /* Generate a random addr */
2906 eth_random_addr(hw->mac.addr);
2907 memcpy(hw->mac.perm_addr, hw->mac.addr, ETH_ALEN);
2908 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
2909 &dev->data->mac_addrs[0]);
2912 /* Reset the hw statistics */
2913 fm10k_stats_reset(dev);
2916 diag = fm10k_reset_hw(hw);
2917 if (diag != FM10K_SUCCESS) {
2918 PMD_INIT_LOG(ERR, "Hardware reset failed: %d", diag);
2922 /* Setup mailbox service */
2923 diag = fm10k_setup_mbx_service(hw);
2924 if (diag != FM10K_SUCCESS) {
2925 PMD_INIT_LOG(ERR, "Failed to setup mailbox: %d", diag);
2929 /*PF/VF has different interrupt handling mechanism */
2930 if (hw->mac.type == fm10k_mac_pf) {
2931 /* register callback func to eal lib */
2932 rte_intr_callback_register(intr_handle,
2933 fm10k_dev_interrupt_handler_pf, (void *)dev);
2935 /* enable MISC interrupt */
2936 fm10k_dev_enable_intr_pf(dev);
2938 rte_intr_callback_register(intr_handle,
2939 fm10k_dev_interrupt_handler_vf, (void *)dev);
2941 fm10k_dev_enable_intr_vf(dev);
2944 /* Enable intr after callback registered */
2945 rte_intr_enable(intr_handle);
2947 hw->mac.ops.update_int_moderator(hw);
2949 /* Make sure Switch Manager is ready before going forward. */
2950 if (hw->mac.type == fm10k_mac_pf) {
2951 int switch_ready = 0;
2953 for (i = 0; i < MAX_QUERY_SWITCH_STATE_TIMES; i++) {
2955 hw->mac.ops.get_host_state(hw, &switch_ready);
2956 fm10k_mbx_unlock(hw);
2959 /* Delay some time to acquire async LPORT_MAP info. */
2960 rte_delay_us(WAIT_SWITCH_MSG_US);
2963 if (switch_ready == 0) {
2964 PMD_INIT_LOG(ERR, "switch is not ready");
2970 * Below function will trigger operations on mailbox, acquire lock to
2971 * avoid race condition from interrupt handler. Operations on mailbox
2972 * FIFO will trigger interrupt to PF/SM, in which interrupt handler
2973 * will handle and generate an interrupt to our side. Then, FIFO in
2974 * mailbox will be touched.
2977 /* Enable port first */
2978 hw->mac.ops.update_lport_state(hw, hw->mac.dglort_map,
2981 /* Set unicast mode by default. App can change to other mode in other
2984 hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
2985 FM10K_XCAST_MODE_NONE);
2987 fm10k_mbx_unlock(hw);
2989 /* Make sure default VID is ready before going forward. */
2990 if (hw->mac.type == fm10k_mac_pf) {
2991 for (i = 0; i < MAX_QUERY_SWITCH_STATE_TIMES; i++) {
2992 if (hw->mac.default_vid)
2994 /* Delay some time to acquire async port VLAN info. */
2995 rte_delay_us(WAIT_SWITCH_MSG_US);
2998 if (!hw->mac.default_vid) {
2999 PMD_INIT_LOG(ERR, "default VID is not ready");
3004 /* Add default mac address */
3005 fm10k_MAC_filter_set(dev, hw->mac.addr, true,
3006 MAIN_VSI_POOL_NUMBER);
3012 eth_fm10k_dev_uninit(struct rte_eth_dev *dev)
3014 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3015 struct rte_pci_device *pdev = dev->pci_dev;
3016 struct rte_intr_handle *intr_handle = &pdev->intr_handle;
3017 PMD_INIT_FUNC_TRACE();
3019 /* only uninitialize in the primary process */
3020 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3023 /* safe to close dev here */
3024 fm10k_dev_close(dev);
3026 dev->dev_ops = NULL;
3027 dev->rx_pkt_burst = NULL;
3028 dev->tx_pkt_burst = NULL;
3030 /* disable uio/vfio intr */
3031 rte_intr_disable(intr_handle);
3033 /*PF/VF has different interrupt handling mechanism */
3034 if (hw->mac.type == fm10k_mac_pf) {
3035 /* disable interrupt */
3036 fm10k_dev_disable_intr_pf(dev);
3038 /* unregister callback func to eal lib */
3039 rte_intr_callback_unregister(intr_handle,
3040 fm10k_dev_interrupt_handler_pf, (void *)dev);
3042 /* disable interrupt */
3043 fm10k_dev_disable_intr_vf(dev);
3045 rte_intr_callback_unregister(intr_handle,
3046 fm10k_dev_interrupt_handler_vf, (void *)dev);
3049 /* free mac memory */
3050 if (dev->data->mac_addrs) {
3051 rte_free(dev->data->mac_addrs);
3052 dev->data->mac_addrs = NULL;
3055 memset(hw, 0, sizeof(*hw));
3061 * The set of PCI devices this driver supports. This driver will enable both PF
3062 * and SRIOV-VF devices.
3064 static const struct rte_pci_id pci_id_fm10k_map[] = {
3065 { RTE_PCI_DEVICE(FM10K_INTEL_VENDOR_ID, FM10K_DEV_ID_PF) },
3066 { RTE_PCI_DEVICE(FM10K_INTEL_VENDOR_ID, FM10K_DEV_ID_SDI_FM10420_QDA2) },
3067 { RTE_PCI_DEVICE(FM10K_INTEL_VENDOR_ID, FM10K_DEV_ID_VF) },
3068 { .vendor_id = 0, /* sentinel */ },
3071 static struct eth_driver rte_pmd_fm10k = {
3073 .id_table = pci_id_fm10k_map,
3074 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
3075 RTE_PCI_DRV_DETACHABLE,
3076 .probe = rte_eth_dev_pci_probe,
3077 .remove = rte_eth_dev_pci_remove,
3079 .eth_dev_init = eth_fm10k_dev_init,
3080 .eth_dev_uninit = eth_fm10k_dev_uninit,
3081 .dev_private_size = sizeof(struct fm10k_adapter),
3084 RTE_PMD_REGISTER_PCI(net_fm10k, rte_pmd_fm10k.pci_drv);
3085 RTE_PMD_REGISTER_PCI_TABLE(net_fm10k, pci_id_fm10k_map);
3086 RTE_PMD_REGISTER_KMOD_DEP(net_fm10k, "* igb_uio | uio_pci_generic | vfio");