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->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 = RTE_DEV_TO_PCI(dev->device);
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 = RTE_DEV_TO_PCI(dev->device);
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);
1320 xstats[count].id = count;
1324 /* PF queue stats */
1325 for (q = 0; q < FM10K_MAX_QUEUES_PF; q++) {
1326 for (i = 0; i < FM10K_NB_RX_Q_XSTATS; i++) {
1327 xstats[count].value =
1328 *(uint64_t *)(((char *)&hw_stats->q[q]) +
1329 fm10k_hw_stats_rx_q_strings[i].offset);
1330 xstats[count].id = count;
1333 for (i = 0; i < FM10K_NB_TX_Q_XSTATS; i++) {
1334 xstats[count].value =
1335 *(uint64_t *)(((char *)&hw_stats->q[q]) +
1336 fm10k_hw_stats_tx_q_strings[i].offset);
1337 xstats[count].id = count;
1342 return FM10K_NB_XSTATS;
1346 fm10k_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1348 uint64_t ipackets, opackets, ibytes, obytes;
1349 struct fm10k_hw *hw =
1350 FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1351 struct fm10k_hw_stats *hw_stats =
1352 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1355 PMD_INIT_FUNC_TRACE();
1357 fm10k_update_hw_stats(hw, hw_stats);
1359 ipackets = opackets = ibytes = obytes = 0;
1360 for (i = 0; (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) &&
1361 (i < hw->mac.max_queues); ++i) {
1362 stats->q_ipackets[i] = hw_stats->q[i].rx_packets.count;
1363 stats->q_opackets[i] = hw_stats->q[i].tx_packets.count;
1364 stats->q_ibytes[i] = hw_stats->q[i].rx_bytes.count;
1365 stats->q_obytes[i] = hw_stats->q[i].tx_bytes.count;
1366 ipackets += stats->q_ipackets[i];
1367 opackets += stats->q_opackets[i];
1368 ibytes += stats->q_ibytes[i];
1369 obytes += stats->q_obytes[i];
1371 stats->ipackets = ipackets;
1372 stats->opackets = opackets;
1373 stats->ibytes = ibytes;
1374 stats->obytes = obytes;
1378 fm10k_stats_reset(struct rte_eth_dev *dev)
1380 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1381 struct fm10k_hw_stats *hw_stats =
1382 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
1384 PMD_INIT_FUNC_TRACE();
1386 memset(hw_stats, 0, sizeof(*hw_stats));
1387 fm10k_rebind_hw_stats(hw, hw_stats);
1391 fm10k_dev_infos_get(struct rte_eth_dev *dev,
1392 struct rte_eth_dev_info *dev_info)
1394 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1395 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev->device);
1397 PMD_INIT_FUNC_TRACE();
1399 dev_info->pci_dev = pdev;
1400 dev_info->min_rx_bufsize = FM10K_MIN_RX_BUF_SIZE;
1401 dev_info->max_rx_pktlen = FM10K_MAX_PKT_SIZE;
1402 dev_info->max_rx_queues = hw->mac.max_queues;
1403 dev_info->max_tx_queues = hw->mac.max_queues;
1404 dev_info->max_mac_addrs = FM10K_MAX_MACADDR_NUM;
1405 dev_info->max_hash_mac_addrs = 0;
1406 dev_info->max_vfs = pdev->max_vfs;
1407 dev_info->vmdq_pool_base = 0;
1408 dev_info->vmdq_queue_base = 0;
1409 dev_info->max_vmdq_pools = ETH_32_POOLS;
1410 dev_info->vmdq_queue_num = FM10K_MAX_QUEUES_PF;
1411 dev_info->rx_offload_capa =
1412 DEV_RX_OFFLOAD_VLAN_STRIP |
1413 DEV_RX_OFFLOAD_IPV4_CKSUM |
1414 DEV_RX_OFFLOAD_UDP_CKSUM |
1415 DEV_RX_OFFLOAD_TCP_CKSUM;
1416 dev_info->tx_offload_capa =
1417 DEV_TX_OFFLOAD_VLAN_INSERT |
1418 DEV_TX_OFFLOAD_IPV4_CKSUM |
1419 DEV_TX_OFFLOAD_UDP_CKSUM |
1420 DEV_TX_OFFLOAD_TCP_CKSUM |
1421 DEV_TX_OFFLOAD_TCP_TSO;
1423 dev_info->hash_key_size = FM10K_RSSRK_SIZE * sizeof(uint32_t);
1424 dev_info->reta_size = FM10K_MAX_RSS_INDICES;
1426 dev_info->default_rxconf = (struct rte_eth_rxconf) {
1428 .pthresh = FM10K_DEFAULT_RX_PTHRESH,
1429 .hthresh = FM10K_DEFAULT_RX_HTHRESH,
1430 .wthresh = FM10K_DEFAULT_RX_WTHRESH,
1432 .rx_free_thresh = FM10K_RX_FREE_THRESH_DEFAULT(0),
1436 dev_info->default_txconf = (struct rte_eth_txconf) {
1438 .pthresh = FM10K_DEFAULT_TX_PTHRESH,
1439 .hthresh = FM10K_DEFAULT_TX_HTHRESH,
1440 .wthresh = FM10K_DEFAULT_TX_WTHRESH,
1442 .tx_free_thresh = FM10K_TX_FREE_THRESH_DEFAULT(0),
1443 .tx_rs_thresh = FM10K_TX_RS_THRESH_DEFAULT(0),
1444 .txq_flags = FM10K_SIMPLE_TX_FLAG,
1447 dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
1448 .nb_max = FM10K_MAX_RX_DESC,
1449 .nb_min = FM10K_MIN_RX_DESC,
1450 .nb_align = FM10K_MULT_RX_DESC,
1453 dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
1454 .nb_max = FM10K_MAX_TX_DESC,
1455 .nb_min = FM10K_MIN_TX_DESC,
1456 .nb_align = FM10K_MULT_TX_DESC,
1457 .nb_seg_max = FM10K_TX_MAX_SEG,
1458 .nb_mtu_seg_max = FM10K_TX_MAX_MTU_SEG,
1461 dev_info->speed_capa = ETH_LINK_SPEED_1G | ETH_LINK_SPEED_2_5G |
1462 ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
1463 ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G;
1466 #ifdef RTE_LIBRTE_FM10K_RX_OLFLAGS_ENABLE
1467 static const uint32_t *
1468 fm10k_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1470 if (dev->rx_pkt_burst == fm10k_recv_pkts ||
1471 dev->rx_pkt_burst == fm10k_recv_scattered_pkts) {
1472 static uint32_t ptypes[] = {
1473 /* refers to rx_desc_to_ol_flags() */
1476 RTE_PTYPE_L3_IPV4_EXT,
1478 RTE_PTYPE_L3_IPV6_EXT,
1485 } else if (dev->rx_pkt_burst == fm10k_recv_pkts_vec ||
1486 dev->rx_pkt_burst == fm10k_recv_scattered_pkts_vec) {
1487 static uint32_t ptypes_vec[] = {
1488 /* refers to fm10k_desc_to_pktype_v() */
1490 RTE_PTYPE_L3_IPV4_EXT,
1492 RTE_PTYPE_L3_IPV6_EXT,
1495 RTE_PTYPE_TUNNEL_GENEVE,
1496 RTE_PTYPE_TUNNEL_NVGRE,
1497 RTE_PTYPE_TUNNEL_VXLAN,
1498 RTE_PTYPE_TUNNEL_GRE,
1508 static const uint32_t *
1509 fm10k_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1516 fm10k_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1519 uint16_t mac_num = 0;
1520 uint32_t vid_idx, vid_bit, mac_index;
1521 struct fm10k_hw *hw;
1522 struct fm10k_macvlan_filter_info *macvlan;
1523 struct rte_eth_dev_data *data = dev->data;
1525 hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1526 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1528 if (macvlan->nb_queue_pools > 0) { /* VMDQ mode */
1529 PMD_INIT_LOG(ERR, "Cannot change VLAN filter in VMDQ mode");
1533 if (vlan_id > ETH_VLAN_ID_MAX) {
1534 PMD_INIT_LOG(ERR, "Invalid vlan_id: must be < 4096");
1538 vid_idx = FM10K_VFTA_IDX(vlan_id);
1539 vid_bit = FM10K_VFTA_BIT(vlan_id);
1540 /* this VLAN ID is already in the VLAN filter table, return SUCCESS */
1541 if (on && (macvlan->vfta[vid_idx] & vid_bit))
1543 /* this VLAN ID is NOT in the VLAN filter table, cannot remove */
1544 if (!on && !(macvlan->vfta[vid_idx] & vid_bit)) {
1545 PMD_INIT_LOG(ERR, "Invalid vlan_id: not existing "
1546 "in the VLAN filter table");
1551 result = fm10k_update_vlan(hw, vlan_id, 0, on);
1552 fm10k_mbx_unlock(hw);
1553 if (result != FM10K_SUCCESS) {
1554 PMD_INIT_LOG(ERR, "VLAN update failed: %d", result);
1558 for (mac_index = 0; (mac_index < FM10K_MAX_MACADDR_NUM) &&
1559 (result == FM10K_SUCCESS); mac_index++) {
1560 if (is_zero_ether_addr(&data->mac_addrs[mac_index]))
1562 if (mac_num > macvlan->mac_num - 1) {
1563 PMD_INIT_LOG(ERR, "MAC address number "
1568 result = fm10k_update_uc_addr(hw, hw->mac.dglort_map,
1569 data->mac_addrs[mac_index].addr_bytes,
1571 fm10k_mbx_unlock(hw);
1574 if (result != FM10K_SUCCESS) {
1575 PMD_INIT_LOG(ERR, "MAC address update failed: %d", result);
1580 macvlan->vlan_num++;
1581 macvlan->vfta[vid_idx] |= vid_bit;
1583 macvlan->vlan_num--;
1584 macvlan->vfta[vid_idx] &= ~vid_bit;
1590 fm10k_vlan_offload_set(__rte_unused struct rte_eth_dev *dev, int mask)
1592 if (mask & ETH_VLAN_STRIP_MASK) {
1593 if (!dev->data->dev_conf.rxmode.hw_vlan_strip)
1594 PMD_INIT_LOG(ERR, "VLAN stripping is "
1595 "always on in fm10k");
1598 if (mask & ETH_VLAN_EXTEND_MASK) {
1599 if (dev->data->dev_conf.rxmode.hw_vlan_extend)
1600 PMD_INIT_LOG(ERR, "VLAN QinQ is not "
1601 "supported in fm10k");
1604 if (mask & ETH_VLAN_FILTER_MASK) {
1605 if (!dev->data->dev_conf.rxmode.hw_vlan_filter)
1606 PMD_INIT_LOG(ERR, "VLAN filter is always on in fm10k");
1610 /* Add/Remove a MAC address, and update filters to main VSI */
1611 static void fm10k_MAC_filter_set_main_vsi(struct rte_eth_dev *dev,
1612 const u8 *mac, bool add, uint32_t pool)
1614 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1615 struct fm10k_macvlan_filter_info *macvlan;
1618 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1620 if (pool != MAIN_VSI_POOL_NUMBER) {
1621 PMD_DRV_LOG(ERR, "VMDQ not enabled, can't set "
1622 "mac to pool %u", pool);
1625 for (i = 0, j = 0; j < FM10K_VFTA_SIZE; j++) {
1626 if (!macvlan->vfta[j])
1628 for (k = 0; k < FM10K_UINT32_BIT_SIZE; k++) {
1629 if (!(macvlan->vfta[j] & (1 << k)))
1631 if (i + 1 > macvlan->vlan_num) {
1632 PMD_INIT_LOG(ERR, "vlan number not match");
1636 fm10k_update_uc_addr(hw, hw->mac.dglort_map, mac,
1637 j * FM10K_UINT32_BIT_SIZE + k, add, 0);
1638 fm10k_mbx_unlock(hw);
1644 /* Add/Remove a MAC address, and update filters to VMDQ */
1645 static void fm10k_MAC_filter_set_vmdq(struct rte_eth_dev *dev,
1646 const u8 *mac, bool add, uint32_t pool)
1648 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1649 struct fm10k_macvlan_filter_info *macvlan;
1650 struct rte_eth_vmdq_rx_conf *vmdq_conf;
1653 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1654 vmdq_conf = &dev->data->dev_conf.rx_adv_conf.vmdq_rx_conf;
1656 if (pool > macvlan->nb_queue_pools) {
1657 PMD_DRV_LOG(ERR, "Pool number %u invalid."
1659 pool, macvlan->nb_queue_pools);
1662 for (i = 0; i < vmdq_conf->nb_pool_maps; i++) {
1663 if (!(vmdq_conf->pool_map[i].pools & (1UL << pool)))
1666 fm10k_update_uc_addr(hw, hw->mac.dglort_map + pool, mac,
1667 vmdq_conf->pool_map[i].vlan_id, add, 0);
1668 fm10k_mbx_unlock(hw);
1672 /* Add/Remove a MAC address, and update filters */
1673 static void fm10k_MAC_filter_set(struct rte_eth_dev *dev,
1674 const u8 *mac, bool add, uint32_t pool)
1676 struct fm10k_macvlan_filter_info *macvlan;
1678 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1680 if (macvlan->nb_queue_pools > 0) /* VMDQ mode */
1681 fm10k_MAC_filter_set_vmdq(dev, mac, add, pool);
1683 fm10k_MAC_filter_set_main_vsi(dev, mac, add, pool);
1691 /* Add a MAC address, and update filters */
1693 fm10k_macaddr_add(struct rte_eth_dev *dev,
1694 struct ether_addr *mac_addr,
1698 struct fm10k_macvlan_filter_info *macvlan;
1700 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1701 fm10k_MAC_filter_set(dev, mac_addr->addr_bytes, TRUE, pool);
1702 macvlan->mac_vmdq_id[index] = pool;
1705 /* Remove a MAC address, and update filters */
1707 fm10k_macaddr_remove(struct rte_eth_dev *dev, uint32_t index)
1709 struct rte_eth_dev_data *data = dev->data;
1710 struct fm10k_macvlan_filter_info *macvlan;
1712 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1713 fm10k_MAC_filter_set(dev, data->mac_addrs[index].addr_bytes,
1714 FALSE, macvlan->mac_vmdq_id[index]);
1715 macvlan->mac_vmdq_id[index] = 0;
1719 check_nb_desc(uint16_t min, uint16_t max, uint16_t mult, uint16_t request)
1721 if ((request < min) || (request > max) || ((request % mult) != 0))
1729 check_thresh(uint16_t min, uint16_t max, uint16_t div, uint16_t request)
1731 if ((request < min) || (request > max) || ((div % request) != 0))
1738 handle_rxconf(struct fm10k_rx_queue *q, const struct rte_eth_rxconf *conf)
1740 uint16_t rx_free_thresh;
1742 if (conf->rx_free_thresh == 0)
1743 rx_free_thresh = FM10K_RX_FREE_THRESH_DEFAULT(q);
1745 rx_free_thresh = conf->rx_free_thresh;
1747 /* make sure the requested threshold satisfies the constraints */
1748 if (check_thresh(FM10K_RX_FREE_THRESH_MIN(q),
1749 FM10K_RX_FREE_THRESH_MAX(q),
1750 FM10K_RX_FREE_THRESH_DIV(q),
1752 PMD_INIT_LOG(ERR, "rx_free_thresh (%u) must be "
1753 "less than or equal to %u, "
1754 "greater than or equal to %u, "
1755 "and a divisor of %u",
1756 rx_free_thresh, FM10K_RX_FREE_THRESH_MAX(q),
1757 FM10K_RX_FREE_THRESH_MIN(q),
1758 FM10K_RX_FREE_THRESH_DIV(q));
1762 q->alloc_thresh = rx_free_thresh;
1763 q->drop_en = conf->rx_drop_en;
1764 q->rx_deferred_start = conf->rx_deferred_start;
1770 * Hardware requires specific alignment for Rx packet buffers. At
1771 * least one of the following two conditions must be satisfied.
1772 * 1. Address is 512B aligned
1773 * 2. Address is 8B aligned and buffer does not cross 4K boundary.
1775 * As such, the driver may need to adjust the DMA address within the
1776 * buffer by up to 512B.
1778 * return 1 if the element size is valid, otherwise return 0.
1781 mempool_element_size_valid(struct rte_mempool *mp)
1785 /* elt_size includes mbuf header and headroom */
1786 min_size = mp->elt_size - sizeof(struct rte_mbuf) -
1787 RTE_PKTMBUF_HEADROOM;
1789 /* account for up to 512B of alignment */
1790 min_size -= FM10K_RX_DATABUF_ALIGN;
1792 /* sanity check for overflow */
1793 if (min_size > mp->elt_size)
1801 fm10k_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
1802 uint16_t nb_desc, unsigned int socket_id,
1803 const struct rte_eth_rxconf *conf, struct rte_mempool *mp)
1805 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1806 struct fm10k_dev_info *dev_info = FM10K_DEV_PRIVATE_TO_INFO(dev);
1807 struct fm10k_rx_queue *q;
1808 const struct rte_memzone *mz;
1810 PMD_INIT_FUNC_TRACE();
1812 /* make sure the mempool element size can account for alignment. */
1813 if (!mempool_element_size_valid(mp)) {
1814 PMD_INIT_LOG(ERR, "Error : Mempool element size is too small");
1818 /* make sure a valid number of descriptors have been requested */
1819 if (check_nb_desc(FM10K_MIN_RX_DESC, FM10K_MAX_RX_DESC,
1820 FM10K_MULT_RX_DESC, nb_desc)) {
1821 PMD_INIT_LOG(ERR, "Number of Rx descriptors (%u) must be "
1822 "less than or equal to %"PRIu32", "
1823 "greater than or equal to %u, "
1824 "and a multiple of %u",
1825 nb_desc, (uint32_t)FM10K_MAX_RX_DESC, FM10K_MIN_RX_DESC,
1826 FM10K_MULT_RX_DESC);
1831 * if this queue existed already, free the associated memory. The
1832 * queue cannot be reused in case we need to allocate memory on
1833 * different socket than was previously used.
1835 if (dev->data->rx_queues[queue_id] != NULL) {
1836 rx_queue_free(dev->data->rx_queues[queue_id]);
1837 dev->data->rx_queues[queue_id] = NULL;
1840 /* allocate memory for the queue structure */
1841 q = rte_zmalloc_socket("fm10k", sizeof(*q), RTE_CACHE_LINE_SIZE,
1844 PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
1850 q->nb_desc = nb_desc;
1851 q->nb_fake_desc = FM10K_MULT_RX_DESC;
1852 q->port_id = dev->data->port_id;
1853 q->queue_id = queue_id;
1854 q->tail_ptr = (volatile uint32_t *)
1855 &((uint32_t *)hw->hw_addr)[FM10K_RDT(queue_id)];
1856 if (handle_rxconf(q, conf))
1859 /* allocate memory for the software ring */
1860 q->sw_ring = rte_zmalloc_socket("fm10k sw ring",
1861 (nb_desc + q->nb_fake_desc) * sizeof(struct rte_mbuf *),
1862 RTE_CACHE_LINE_SIZE, socket_id);
1863 if (q->sw_ring == NULL) {
1864 PMD_INIT_LOG(ERR, "Cannot allocate software ring");
1870 * allocate memory for the hardware descriptor ring. A memzone large
1871 * enough to hold the maximum ring size is requested to allow for
1872 * resizing in later calls to the queue setup function.
1874 mz = rte_eth_dma_zone_reserve(dev, "rx_ring", queue_id,
1875 FM10K_MAX_RX_RING_SZ, FM10K_ALIGN_RX_DESC,
1878 PMD_INIT_LOG(ERR, "Cannot allocate hardware ring");
1879 rte_free(q->sw_ring);
1883 q->hw_ring = mz->addr;
1884 q->hw_ring_phys_addr = rte_mem_phy2mch(mz->memseg_id, mz->phys_addr);
1886 /* Check if number of descs satisfied Vector requirement */
1887 if (!rte_is_power_of_2(nb_desc)) {
1888 PMD_INIT_LOG(DEBUG, "queue[%d] doesn't meet Vector Rx "
1889 "preconditions - canceling the feature for "
1890 "the whole port[%d]",
1891 q->queue_id, q->port_id);
1892 dev_info->rx_vec_allowed = false;
1894 fm10k_rxq_vec_setup(q);
1896 dev->data->rx_queues[queue_id] = q;
1901 fm10k_rx_queue_release(void *queue)
1903 PMD_INIT_FUNC_TRACE();
1905 rx_queue_free(queue);
1909 handle_txconf(struct fm10k_tx_queue *q, const struct rte_eth_txconf *conf)
1911 uint16_t tx_free_thresh;
1912 uint16_t tx_rs_thresh;
1914 /* constraint MACROs require that tx_free_thresh is configured
1915 * before tx_rs_thresh */
1916 if (conf->tx_free_thresh == 0)
1917 tx_free_thresh = FM10K_TX_FREE_THRESH_DEFAULT(q);
1919 tx_free_thresh = conf->tx_free_thresh;
1921 /* make sure the requested threshold satisfies the constraints */
1922 if (check_thresh(FM10K_TX_FREE_THRESH_MIN(q),
1923 FM10K_TX_FREE_THRESH_MAX(q),
1924 FM10K_TX_FREE_THRESH_DIV(q),
1926 PMD_INIT_LOG(ERR, "tx_free_thresh (%u) must be "
1927 "less than or equal to %u, "
1928 "greater than or equal to %u, "
1929 "and a divisor of %u",
1930 tx_free_thresh, FM10K_TX_FREE_THRESH_MAX(q),
1931 FM10K_TX_FREE_THRESH_MIN(q),
1932 FM10K_TX_FREE_THRESH_DIV(q));
1936 q->free_thresh = tx_free_thresh;
1938 if (conf->tx_rs_thresh == 0)
1939 tx_rs_thresh = FM10K_TX_RS_THRESH_DEFAULT(q);
1941 tx_rs_thresh = conf->tx_rs_thresh;
1943 q->tx_deferred_start = conf->tx_deferred_start;
1945 /* make sure the requested threshold satisfies the constraints */
1946 if (check_thresh(FM10K_TX_RS_THRESH_MIN(q),
1947 FM10K_TX_RS_THRESH_MAX(q),
1948 FM10K_TX_RS_THRESH_DIV(q),
1950 PMD_INIT_LOG(ERR, "tx_rs_thresh (%u) must be "
1951 "less than or equal to %u, "
1952 "greater than or equal to %u, "
1953 "and a divisor of %u",
1954 tx_rs_thresh, FM10K_TX_RS_THRESH_MAX(q),
1955 FM10K_TX_RS_THRESH_MIN(q),
1956 FM10K_TX_RS_THRESH_DIV(q));
1960 q->rs_thresh = tx_rs_thresh;
1966 fm10k_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
1967 uint16_t nb_desc, unsigned int socket_id,
1968 const struct rte_eth_txconf *conf)
1970 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1971 struct fm10k_tx_queue *q;
1972 const struct rte_memzone *mz;
1974 PMD_INIT_FUNC_TRACE();
1976 /* make sure a valid number of descriptors have been requested */
1977 if (check_nb_desc(FM10K_MIN_TX_DESC, FM10K_MAX_TX_DESC,
1978 FM10K_MULT_TX_DESC, nb_desc)) {
1979 PMD_INIT_LOG(ERR, "Number of Tx descriptors (%u) must be "
1980 "less than or equal to %"PRIu32", "
1981 "greater than or equal to %u, "
1982 "and a multiple of %u",
1983 nb_desc, (uint32_t)FM10K_MAX_TX_DESC, FM10K_MIN_TX_DESC,
1984 FM10K_MULT_TX_DESC);
1989 * if this queue existed already, free the associated memory. The
1990 * queue cannot be reused in case we need to allocate memory on
1991 * different socket than was previously used.
1993 if (dev->data->tx_queues[queue_id] != NULL) {
1994 struct fm10k_tx_queue *txq = dev->data->tx_queues[queue_id];
1997 dev->data->tx_queues[queue_id] = NULL;
2000 /* allocate memory for the queue structure */
2001 q = rte_zmalloc_socket("fm10k", sizeof(*q), RTE_CACHE_LINE_SIZE,
2004 PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
2009 q->nb_desc = nb_desc;
2010 q->port_id = dev->data->port_id;
2011 q->queue_id = queue_id;
2012 q->txq_flags = conf->txq_flags;
2013 q->ops = &def_txq_ops;
2014 q->tail_ptr = (volatile uint32_t *)
2015 &((uint32_t *)hw->hw_addr)[FM10K_TDT(queue_id)];
2016 if (handle_txconf(q, conf))
2019 /* allocate memory for the software ring */
2020 q->sw_ring = rte_zmalloc_socket("fm10k sw ring",
2021 nb_desc * sizeof(struct rte_mbuf *),
2022 RTE_CACHE_LINE_SIZE, socket_id);
2023 if (q->sw_ring == NULL) {
2024 PMD_INIT_LOG(ERR, "Cannot allocate software ring");
2030 * allocate memory for the hardware descriptor ring. A memzone large
2031 * enough to hold the maximum ring size is requested to allow for
2032 * resizing in later calls to the queue setup function.
2034 mz = rte_eth_dma_zone_reserve(dev, "tx_ring", queue_id,
2035 FM10K_MAX_TX_RING_SZ, FM10K_ALIGN_TX_DESC,
2038 PMD_INIT_LOG(ERR, "Cannot allocate hardware ring");
2039 rte_free(q->sw_ring);
2043 q->hw_ring = mz->addr;
2044 q->hw_ring_phys_addr = rte_mem_phy2mch(mz->memseg_id, mz->phys_addr);
2047 * allocate memory for the RS bit tracker. Enough slots to hold the
2048 * descriptor index for each RS bit needing to be set are required.
2050 q->rs_tracker.list = rte_zmalloc_socket("fm10k rs tracker",
2051 ((nb_desc + 1) / q->rs_thresh) *
2053 RTE_CACHE_LINE_SIZE, socket_id);
2054 if (q->rs_tracker.list == NULL) {
2055 PMD_INIT_LOG(ERR, "Cannot allocate RS bit tracker");
2056 rte_free(q->sw_ring);
2061 dev->data->tx_queues[queue_id] = q;
2066 fm10k_tx_queue_release(void *queue)
2068 struct fm10k_tx_queue *q = queue;
2069 PMD_INIT_FUNC_TRACE();
2075 fm10k_reta_update(struct rte_eth_dev *dev,
2076 struct rte_eth_rss_reta_entry64 *reta_conf,
2079 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2080 uint16_t i, j, idx, shift;
2084 PMD_INIT_FUNC_TRACE();
2086 if (reta_size > FM10K_MAX_RSS_INDICES) {
2087 PMD_INIT_LOG(ERR, "The size of hash lookup table configured "
2088 "(%d) doesn't match the number hardware can supported "
2089 "(%d)", reta_size, FM10K_MAX_RSS_INDICES);
2094 * Update Redirection Table RETA[n], n=0..31. The redirection table has
2095 * 128-entries in 32 registers
2097 for (i = 0; i < FM10K_MAX_RSS_INDICES; i += CHARS_PER_UINT32) {
2098 idx = i / RTE_RETA_GROUP_SIZE;
2099 shift = i % RTE_RETA_GROUP_SIZE;
2100 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2101 BIT_MASK_PER_UINT32);
2106 if (mask != BIT_MASK_PER_UINT32)
2107 reta = FM10K_READ_REG(hw, FM10K_RETA(0, i >> 2));
2109 for (j = 0; j < CHARS_PER_UINT32; j++) {
2110 if (mask & (0x1 << j)) {
2112 reta &= ~(UINT8_MAX << CHAR_BIT * j);
2113 reta |= reta_conf[idx].reta[shift + j] <<
2117 FM10K_WRITE_REG(hw, FM10K_RETA(0, i >> 2), reta);
2124 fm10k_reta_query(struct rte_eth_dev *dev,
2125 struct rte_eth_rss_reta_entry64 *reta_conf,
2128 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2129 uint16_t i, j, idx, shift;
2133 PMD_INIT_FUNC_TRACE();
2135 if (reta_size < FM10K_MAX_RSS_INDICES) {
2136 PMD_INIT_LOG(ERR, "The size of hash lookup table configured "
2137 "(%d) doesn't match the number hardware can supported "
2138 "(%d)", reta_size, FM10K_MAX_RSS_INDICES);
2143 * Read Redirection Table RETA[n], n=0..31. The redirection table has
2144 * 128-entries in 32 registers
2146 for (i = 0; i < FM10K_MAX_RSS_INDICES; i += CHARS_PER_UINT32) {
2147 idx = i / RTE_RETA_GROUP_SIZE;
2148 shift = i % RTE_RETA_GROUP_SIZE;
2149 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2150 BIT_MASK_PER_UINT32);
2154 reta = FM10K_READ_REG(hw, FM10K_RETA(0, i >> 2));
2155 for (j = 0; j < CHARS_PER_UINT32; j++) {
2156 if (mask & (0x1 << j))
2157 reta_conf[idx].reta[shift + j] = ((reta >>
2158 CHAR_BIT * j) & UINT8_MAX);
2166 fm10k_rss_hash_update(struct rte_eth_dev *dev,
2167 struct rte_eth_rss_conf *rss_conf)
2169 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2170 uint32_t *key = (uint32_t *)rss_conf->rss_key;
2172 uint64_t hf = rss_conf->rss_hf;
2175 PMD_INIT_FUNC_TRACE();
2177 if (key && (rss_conf->rss_key_len < FM10K_RSSRK_SIZE *
2178 FM10K_RSSRK_ENTRIES_PER_REG))
2185 mrqc |= (hf & ETH_RSS_IPV4) ? FM10K_MRQC_IPV4 : 0;
2186 mrqc |= (hf & ETH_RSS_IPV6) ? FM10K_MRQC_IPV6 : 0;
2187 mrqc |= (hf & ETH_RSS_IPV6_EX) ? FM10K_MRQC_IPV6 : 0;
2188 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_TCP) ? FM10K_MRQC_TCP_IPV4 : 0;
2189 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_TCP) ? FM10K_MRQC_TCP_IPV6 : 0;
2190 mrqc |= (hf & ETH_RSS_IPV6_TCP_EX) ? FM10K_MRQC_TCP_IPV6 : 0;
2191 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_UDP) ? FM10K_MRQC_UDP_IPV4 : 0;
2192 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_UDP) ? FM10K_MRQC_UDP_IPV6 : 0;
2193 mrqc |= (hf & ETH_RSS_IPV6_UDP_EX) ? FM10K_MRQC_UDP_IPV6 : 0;
2195 /* If the mapping doesn't fit any supported, return */
2200 for (i = 0; i < FM10K_RSSRK_SIZE; ++i)
2201 FM10K_WRITE_REG(hw, FM10K_RSSRK(0, i), key[i]);
2203 FM10K_WRITE_REG(hw, FM10K_MRQC(0), mrqc);
2209 fm10k_rss_hash_conf_get(struct rte_eth_dev *dev,
2210 struct rte_eth_rss_conf *rss_conf)
2212 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2213 uint32_t *key = (uint32_t *)rss_conf->rss_key;
2218 PMD_INIT_FUNC_TRACE();
2220 if (key && (rss_conf->rss_key_len < FM10K_RSSRK_SIZE *
2221 FM10K_RSSRK_ENTRIES_PER_REG))
2225 for (i = 0; i < FM10K_RSSRK_SIZE; ++i)
2226 key[i] = FM10K_READ_REG(hw, FM10K_RSSRK(0, i));
2228 mrqc = FM10K_READ_REG(hw, FM10K_MRQC(0));
2230 hf |= (mrqc & FM10K_MRQC_IPV4) ? ETH_RSS_IPV4 : 0;
2231 hf |= (mrqc & FM10K_MRQC_IPV6) ? ETH_RSS_IPV6 : 0;
2232 hf |= (mrqc & FM10K_MRQC_IPV6) ? ETH_RSS_IPV6_EX : 0;
2233 hf |= (mrqc & FM10K_MRQC_TCP_IPV4) ? ETH_RSS_NONFRAG_IPV4_TCP : 0;
2234 hf |= (mrqc & FM10K_MRQC_TCP_IPV6) ? ETH_RSS_NONFRAG_IPV6_TCP : 0;
2235 hf |= (mrqc & FM10K_MRQC_TCP_IPV6) ? ETH_RSS_IPV6_TCP_EX : 0;
2236 hf |= (mrqc & FM10K_MRQC_UDP_IPV4) ? ETH_RSS_NONFRAG_IPV4_UDP : 0;
2237 hf |= (mrqc & FM10K_MRQC_UDP_IPV6) ? ETH_RSS_NONFRAG_IPV6_UDP : 0;
2238 hf |= (mrqc & FM10K_MRQC_UDP_IPV6) ? ETH_RSS_IPV6_UDP_EX : 0;
2240 rss_conf->rss_hf = hf;
2246 fm10k_dev_enable_intr_pf(struct rte_eth_dev *dev)
2248 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2249 uint32_t int_map = FM10K_INT_MAP_IMMEDIATE;
2251 /* Bind all local non-queue interrupt to vector 0 */
2252 int_map |= FM10K_MISC_VEC_ID;
2254 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_mailbox), int_map);
2255 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_pcie_fault), int_map);
2256 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_up_down), int_map);
2257 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_event), int_map);
2258 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_sram), int_map);
2259 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_vflr), int_map);
2261 /* Enable misc causes */
2262 FM10K_WRITE_REG(hw, FM10K_EIMR, FM10K_EIMR_ENABLE(PCA_FAULT) |
2263 FM10K_EIMR_ENABLE(THI_FAULT) |
2264 FM10K_EIMR_ENABLE(FUM_FAULT) |
2265 FM10K_EIMR_ENABLE(MAILBOX) |
2266 FM10K_EIMR_ENABLE(SWITCHREADY) |
2267 FM10K_EIMR_ENABLE(SWITCHNOTREADY) |
2268 FM10K_EIMR_ENABLE(SRAMERROR) |
2269 FM10K_EIMR_ENABLE(VFLR));
2272 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_AUTOMASK |
2273 FM10K_ITR_MASK_CLEAR);
2274 FM10K_WRITE_FLUSH(hw);
2278 fm10k_dev_disable_intr_pf(struct rte_eth_dev *dev)
2280 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2281 uint32_t int_map = FM10K_INT_MAP_DISABLE;
2283 int_map |= FM10K_MISC_VEC_ID;
2285 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_mailbox), int_map);
2286 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_pcie_fault), int_map);
2287 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_up_down), int_map);
2288 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_switch_event), int_map);
2289 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_sram), int_map);
2290 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_vflr), int_map);
2292 /* Disable misc causes */
2293 FM10K_WRITE_REG(hw, FM10K_EIMR, FM10K_EIMR_DISABLE(PCA_FAULT) |
2294 FM10K_EIMR_DISABLE(THI_FAULT) |
2295 FM10K_EIMR_DISABLE(FUM_FAULT) |
2296 FM10K_EIMR_DISABLE(MAILBOX) |
2297 FM10K_EIMR_DISABLE(SWITCHREADY) |
2298 FM10K_EIMR_DISABLE(SWITCHNOTREADY) |
2299 FM10K_EIMR_DISABLE(SRAMERROR) |
2300 FM10K_EIMR_DISABLE(VFLR));
2303 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_MASK_SET);
2304 FM10K_WRITE_FLUSH(hw);
2308 fm10k_dev_enable_intr_vf(struct rte_eth_dev *dev)
2310 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2311 uint32_t int_map = FM10K_INT_MAP_IMMEDIATE;
2313 /* Bind all local non-queue interrupt to vector 0 */
2314 int_map |= FM10K_MISC_VEC_ID;
2316 /* Only INT 0 available, other 15 are reserved. */
2317 FM10K_WRITE_REG(hw, FM10K_VFINT_MAP, int_map);
2320 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_AUTOMASK |
2321 FM10K_ITR_MASK_CLEAR);
2322 FM10K_WRITE_FLUSH(hw);
2326 fm10k_dev_disable_intr_vf(struct rte_eth_dev *dev)
2328 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2329 uint32_t int_map = FM10K_INT_MAP_DISABLE;
2331 int_map |= FM10K_MISC_VEC_ID;
2333 /* Only INT 0 available, other 15 are reserved. */
2334 FM10K_WRITE_REG(hw, FM10K_VFINT_MAP, int_map);
2337 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_MASK_SET);
2338 FM10K_WRITE_FLUSH(hw);
2342 fm10k_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
2344 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2345 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev->device);
2348 if (hw->mac.type == fm10k_mac_pf)
2349 FM10K_WRITE_REG(hw, FM10K_ITR(Q2V(pdev, queue_id)),
2350 FM10K_ITR_AUTOMASK | FM10K_ITR_MASK_CLEAR);
2352 FM10K_WRITE_REG(hw, FM10K_VFITR(Q2V(pdev, queue_id)),
2353 FM10K_ITR_AUTOMASK | FM10K_ITR_MASK_CLEAR);
2354 rte_intr_enable(&pdev->intr_handle);
2359 fm10k_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
2361 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2362 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev->device);
2365 if (hw->mac.type == fm10k_mac_pf)
2366 FM10K_WRITE_REG(hw, FM10K_ITR(Q2V(pdev, queue_id)),
2367 FM10K_ITR_MASK_SET);
2369 FM10K_WRITE_REG(hw, FM10K_VFITR(Q2V(pdev, queue_id)),
2370 FM10K_ITR_MASK_SET);
2375 fm10k_dev_rxq_interrupt_setup(struct rte_eth_dev *dev)
2377 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2378 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev->device);
2379 struct rte_intr_handle *intr_handle = &pdev->intr_handle;
2380 uint32_t intr_vector, vec;
2384 /* fm10k needs one separate interrupt for mailbox,
2385 * so only drivers which support multiple interrupt vectors
2386 * e.g. vfio-pci can work for fm10k interrupt mode
2388 if (!rte_intr_cap_multiple(intr_handle) ||
2389 dev->data->dev_conf.intr_conf.rxq == 0)
2392 intr_vector = dev->data->nb_rx_queues;
2394 /* disable interrupt first */
2395 rte_intr_disable(intr_handle);
2396 if (hw->mac.type == fm10k_mac_pf)
2397 fm10k_dev_disable_intr_pf(dev);
2399 fm10k_dev_disable_intr_vf(dev);
2401 if (rte_intr_efd_enable(intr_handle, intr_vector)) {
2402 PMD_INIT_LOG(ERR, "Failed to init event fd");
2406 if (rte_intr_dp_is_en(intr_handle) && !result) {
2407 intr_handle->intr_vec = rte_zmalloc("intr_vec",
2408 dev->data->nb_rx_queues * sizeof(int), 0);
2409 if (intr_handle->intr_vec) {
2410 for (queue_id = 0, vec = FM10K_RX_VEC_START;
2411 queue_id < dev->data->nb_rx_queues;
2413 intr_handle->intr_vec[queue_id] = vec;
2414 if (vec < intr_handle->nb_efd - 1
2415 + FM10K_RX_VEC_START)
2419 PMD_INIT_LOG(ERR, "Failed to allocate %d rx_queues"
2420 " intr_vec", dev->data->nb_rx_queues);
2421 rte_intr_efd_disable(intr_handle);
2426 if (hw->mac.type == fm10k_mac_pf)
2427 fm10k_dev_enable_intr_pf(dev);
2429 fm10k_dev_enable_intr_vf(dev);
2430 rte_intr_enable(intr_handle);
2431 hw->mac.ops.update_int_moderator(hw);
2436 fm10k_dev_handle_fault(struct fm10k_hw *hw, uint32_t eicr)
2438 struct fm10k_fault fault;
2440 const char *estr = "Unknown error";
2442 /* Process PCA fault */
2443 if (eicr & FM10K_EICR_PCA_FAULT) {
2444 err = fm10k_get_fault(hw, FM10K_PCA_FAULT, &fault);
2447 switch (fault.type) {
2449 estr = "PCA_NO_FAULT"; break;
2450 case PCA_UNMAPPED_ADDR:
2451 estr = "PCA_UNMAPPED_ADDR"; break;
2452 case PCA_BAD_QACCESS_PF:
2453 estr = "PCA_BAD_QACCESS_PF"; break;
2454 case PCA_BAD_QACCESS_VF:
2455 estr = "PCA_BAD_QACCESS_VF"; break;
2456 case PCA_MALICIOUS_REQ:
2457 estr = "PCA_MALICIOUS_REQ"; break;
2458 case PCA_POISONED_TLP:
2459 estr = "PCA_POISONED_TLP"; break;
2461 estr = "PCA_TLP_ABORT"; break;
2465 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
2466 estr, fault.func ? "VF" : "PF", fault.func,
2467 fault.address, fault.specinfo);
2470 /* Process THI fault */
2471 if (eicr & FM10K_EICR_THI_FAULT) {
2472 err = fm10k_get_fault(hw, FM10K_THI_FAULT, &fault);
2475 switch (fault.type) {
2477 estr = "THI_NO_FAULT"; break;
2478 case THI_MAL_DIS_Q_FAULT:
2479 estr = "THI_MAL_DIS_Q_FAULT"; break;
2483 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
2484 estr, fault.func ? "VF" : "PF", fault.func,
2485 fault.address, fault.specinfo);
2488 /* Process FUM fault */
2489 if (eicr & FM10K_EICR_FUM_FAULT) {
2490 err = fm10k_get_fault(hw, FM10K_FUM_FAULT, &fault);
2493 switch (fault.type) {
2495 estr = "FUM_NO_FAULT"; break;
2496 case FUM_UNMAPPED_ADDR:
2497 estr = "FUM_UNMAPPED_ADDR"; break;
2498 case FUM_POISONED_TLP:
2499 estr = "FUM_POISONED_TLP"; break;
2500 case FUM_BAD_VF_QACCESS:
2501 estr = "FUM_BAD_VF_QACCESS"; break;
2502 case FUM_ADD_DECODE_ERR:
2503 estr = "FUM_ADD_DECODE_ERR"; break;
2505 estr = "FUM_RO_ERROR"; break;
2506 case FUM_QPRC_CRC_ERROR:
2507 estr = "FUM_QPRC_CRC_ERROR"; break;
2508 case FUM_CSR_TIMEOUT:
2509 estr = "FUM_CSR_TIMEOUT"; break;
2510 case FUM_INVALID_TYPE:
2511 estr = "FUM_INVALID_TYPE"; break;
2512 case FUM_INVALID_LENGTH:
2513 estr = "FUM_INVALID_LENGTH"; break;
2514 case FUM_INVALID_BE:
2515 estr = "FUM_INVALID_BE"; break;
2516 case FUM_INVALID_ALIGN:
2517 estr = "FUM_INVALID_ALIGN"; break;
2521 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
2522 estr, fault.func ? "VF" : "PF", fault.func,
2523 fault.address, fault.specinfo);
2528 PMD_INIT_LOG(ERR, "Failed to handle fault event.");
2533 * PF interrupt handler triggered by NIC for handling specific interrupt.
2536 * Pointer to interrupt handle.
2538 * The address of parameter (struct rte_eth_dev *) regsitered before.
2544 fm10k_dev_interrupt_handler_pf(
2545 struct rte_intr_handle *handle,
2548 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2549 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2550 uint32_t cause, status;
2552 if (hw->mac.type != fm10k_mac_pf)
2555 cause = FM10K_READ_REG(hw, FM10K_EICR);
2557 /* Handle PCI fault cases */
2558 if (cause & FM10K_EICR_FAULT_MASK) {
2559 PMD_INIT_LOG(ERR, "INT: find fault!");
2560 fm10k_dev_handle_fault(hw, cause);
2563 /* Handle switch up/down */
2564 if (cause & FM10K_EICR_SWITCHNOTREADY)
2565 PMD_INIT_LOG(ERR, "INT: Switch is not ready");
2567 if (cause & FM10K_EICR_SWITCHREADY)
2568 PMD_INIT_LOG(INFO, "INT: Switch is ready");
2570 /* Handle mailbox message */
2572 hw->mbx.ops.process(hw, &hw->mbx);
2573 fm10k_mbx_unlock(hw);
2575 /* Handle SRAM error */
2576 if (cause & FM10K_EICR_SRAMERROR) {
2577 PMD_INIT_LOG(ERR, "INT: SRAM error on PEP");
2579 status = FM10K_READ_REG(hw, FM10K_SRAM_IP);
2580 /* Write to clear pending bits */
2581 FM10K_WRITE_REG(hw, FM10K_SRAM_IP, status);
2583 /* Todo: print out error message after shared code updates */
2586 /* Clear these 3 events if having any */
2587 cause &= FM10K_EICR_SWITCHNOTREADY | FM10K_EICR_MAILBOX |
2588 FM10K_EICR_SWITCHREADY;
2590 FM10K_WRITE_REG(hw, FM10K_EICR, cause);
2592 /* Re-enable interrupt from device side */
2593 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_AUTOMASK |
2594 FM10K_ITR_MASK_CLEAR);
2595 /* Re-enable interrupt from host side */
2596 rte_intr_enable(handle);
2600 * VF interrupt handler triggered by NIC for handling specific interrupt.
2603 * Pointer to interrupt handle.
2605 * The address of parameter (struct rte_eth_dev *) regsitered before.
2611 fm10k_dev_interrupt_handler_vf(
2612 struct rte_intr_handle *handle,
2615 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
2616 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2618 if (hw->mac.type != fm10k_mac_vf)
2621 /* Handle mailbox message if lock is acquired */
2623 hw->mbx.ops.process(hw, &hw->mbx);
2624 fm10k_mbx_unlock(hw);
2626 /* Re-enable interrupt from device side */
2627 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_AUTOMASK |
2628 FM10K_ITR_MASK_CLEAR);
2629 /* Re-enable interrupt from host side */
2630 rte_intr_enable(handle);
2633 /* Mailbox message handler in VF */
2634 static const struct fm10k_msg_data fm10k_msgdata_vf[] = {
2635 FM10K_TLV_MSG_TEST_HANDLER(fm10k_tlv_msg_test),
2636 FM10K_VF_MSG_MAC_VLAN_HANDLER(fm10k_msg_mac_vlan_vf),
2637 FM10K_VF_MSG_LPORT_STATE_HANDLER(fm10k_msg_lport_state_vf),
2638 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_tlv_msg_error),
2642 fm10k_setup_mbx_service(struct fm10k_hw *hw)
2646 /* Initialize mailbox lock */
2647 fm10k_mbx_initlock(hw);
2649 /* Replace default message handler with new ones */
2650 if (hw->mac.type == fm10k_mac_vf)
2651 err = hw->mbx.ops.register_handlers(&hw->mbx, fm10k_msgdata_vf);
2654 PMD_INIT_LOG(ERR, "Failed to register mailbox handler.err:%d",
2658 /* Connect to SM for PF device or PF for VF device */
2659 return hw->mbx.ops.connect(hw, &hw->mbx);
2663 fm10k_close_mbx_service(struct fm10k_hw *hw)
2665 /* Disconnect from SM for PF device or PF for VF device */
2666 hw->mbx.ops.disconnect(hw, &hw->mbx);
2669 static const struct eth_dev_ops fm10k_eth_dev_ops = {
2670 .dev_configure = fm10k_dev_configure,
2671 .dev_start = fm10k_dev_start,
2672 .dev_stop = fm10k_dev_stop,
2673 .dev_close = fm10k_dev_close,
2674 .promiscuous_enable = fm10k_dev_promiscuous_enable,
2675 .promiscuous_disable = fm10k_dev_promiscuous_disable,
2676 .allmulticast_enable = fm10k_dev_allmulticast_enable,
2677 .allmulticast_disable = fm10k_dev_allmulticast_disable,
2678 .stats_get = fm10k_stats_get,
2679 .xstats_get = fm10k_xstats_get,
2680 .xstats_get_names = fm10k_xstats_get_names,
2681 .stats_reset = fm10k_stats_reset,
2682 .xstats_reset = fm10k_stats_reset,
2683 .link_update = fm10k_link_update,
2684 .dev_infos_get = fm10k_dev_infos_get,
2685 .dev_supported_ptypes_get = fm10k_dev_supported_ptypes_get,
2686 .vlan_filter_set = fm10k_vlan_filter_set,
2687 .vlan_offload_set = fm10k_vlan_offload_set,
2688 .mac_addr_add = fm10k_macaddr_add,
2689 .mac_addr_remove = fm10k_macaddr_remove,
2690 .rx_queue_start = fm10k_dev_rx_queue_start,
2691 .rx_queue_stop = fm10k_dev_rx_queue_stop,
2692 .tx_queue_start = fm10k_dev_tx_queue_start,
2693 .tx_queue_stop = fm10k_dev_tx_queue_stop,
2694 .rx_queue_setup = fm10k_rx_queue_setup,
2695 .rx_queue_release = fm10k_rx_queue_release,
2696 .tx_queue_setup = fm10k_tx_queue_setup,
2697 .tx_queue_release = fm10k_tx_queue_release,
2698 .rx_descriptor_done = fm10k_dev_rx_descriptor_done,
2699 .rx_queue_intr_enable = fm10k_dev_rx_queue_intr_enable,
2700 .rx_queue_intr_disable = fm10k_dev_rx_queue_intr_disable,
2701 .reta_update = fm10k_reta_update,
2702 .reta_query = fm10k_reta_query,
2703 .rss_hash_update = fm10k_rss_hash_update,
2704 .rss_hash_conf_get = fm10k_rss_hash_conf_get,
2707 static int ftag_check_handler(__rte_unused const char *key,
2708 const char *value, __rte_unused void *opaque)
2710 if (strcmp(value, "1"))
2717 fm10k_check_ftag(struct rte_devargs *devargs)
2719 struct rte_kvargs *kvlist;
2720 const char *ftag_key = "enable_ftag";
2722 if (devargs == NULL)
2725 kvlist = rte_kvargs_parse(devargs->args, NULL);
2729 if (!rte_kvargs_count(kvlist, ftag_key)) {
2730 rte_kvargs_free(kvlist);
2733 /* FTAG is enabled when there's key-value pair: enable_ftag=1 */
2734 if (rte_kvargs_process(kvlist, ftag_key,
2735 ftag_check_handler, NULL) < 0) {
2736 rte_kvargs_free(kvlist);
2739 rte_kvargs_free(kvlist);
2744 static void __attribute__((cold))
2745 fm10k_set_tx_function(struct rte_eth_dev *dev)
2747 struct fm10k_tx_queue *txq;
2750 uint16_t tx_ftag_en = 0;
2752 if (fm10k_check_ftag(dev->device->devargs))
2755 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2756 txq = dev->data->tx_queues[i];
2757 txq->tx_ftag_en = tx_ftag_en;
2758 /* Check if Vector Tx is satisfied */
2759 if (fm10k_tx_vec_condition_check(txq))
2764 PMD_INIT_LOG(DEBUG, "Use vector Tx func");
2765 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2766 txq = dev->data->tx_queues[i];
2767 fm10k_txq_vec_setup(txq);
2769 dev->tx_pkt_burst = fm10k_xmit_pkts_vec;
2770 dev->tx_pkt_prepare = NULL;
2772 dev->tx_pkt_burst = fm10k_xmit_pkts;
2773 dev->tx_pkt_prepare = fm10k_prep_pkts;
2774 PMD_INIT_LOG(DEBUG, "Use regular Tx func");
2778 static void __attribute__((cold))
2779 fm10k_set_rx_function(struct rte_eth_dev *dev)
2781 struct fm10k_dev_info *dev_info = FM10K_DEV_PRIVATE_TO_INFO(dev);
2782 uint16_t i, rx_using_sse;
2783 uint16_t rx_ftag_en = 0;
2785 if (fm10k_check_ftag(dev->device->devargs))
2788 /* In order to allow Vector Rx there are a few configuration
2789 * conditions to be met.
2791 if (!fm10k_rx_vec_condition_check(dev) &&
2792 dev_info->rx_vec_allowed && !rx_ftag_en) {
2793 if (dev->data->scattered_rx)
2794 dev->rx_pkt_burst = fm10k_recv_scattered_pkts_vec;
2796 dev->rx_pkt_burst = fm10k_recv_pkts_vec;
2797 } else if (dev->data->scattered_rx)
2798 dev->rx_pkt_burst = fm10k_recv_scattered_pkts;
2800 dev->rx_pkt_burst = fm10k_recv_pkts;
2803 (dev->rx_pkt_burst == fm10k_recv_scattered_pkts_vec ||
2804 dev->rx_pkt_burst == fm10k_recv_pkts_vec);
2807 PMD_INIT_LOG(DEBUG, "Use vector Rx func");
2809 PMD_INIT_LOG(DEBUG, "Use regular Rx func");
2811 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2812 struct fm10k_rx_queue *rxq = dev->data->rx_queues[i];
2814 rxq->rx_using_sse = rx_using_sse;
2815 rxq->rx_ftag_en = rx_ftag_en;
2820 fm10k_params_init(struct rte_eth_dev *dev)
2822 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2823 struct fm10k_dev_info *info = FM10K_DEV_PRIVATE_TO_INFO(dev);
2825 /* Inialize bus info. Normally we would call fm10k_get_bus_info(), but
2826 * there is no way to get link status without reading BAR4. Until this
2827 * works, assume we have maximum bandwidth.
2828 * @todo - fix bus info
2830 hw->bus_caps.speed = fm10k_bus_speed_8000;
2831 hw->bus_caps.width = fm10k_bus_width_pcie_x8;
2832 hw->bus_caps.payload = fm10k_bus_payload_512;
2833 hw->bus.speed = fm10k_bus_speed_8000;
2834 hw->bus.width = fm10k_bus_width_pcie_x8;
2835 hw->bus.payload = fm10k_bus_payload_256;
2837 info->rx_vec_allowed = true;
2841 eth_fm10k_dev_init(struct rte_eth_dev *dev)
2843 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2844 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev->device);
2845 struct rte_intr_handle *intr_handle = &pdev->intr_handle;
2847 struct fm10k_macvlan_filter_info *macvlan;
2849 PMD_INIT_FUNC_TRACE();
2851 dev->dev_ops = &fm10k_eth_dev_ops;
2852 dev->rx_pkt_burst = &fm10k_recv_pkts;
2853 dev->tx_pkt_burst = &fm10k_xmit_pkts;
2854 dev->tx_pkt_prepare = &fm10k_prep_pkts;
2856 /* only initialize in the primary process */
2857 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2860 rte_eth_copy_pci_info(dev, pdev);
2861 dev->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
2863 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
2864 memset(macvlan, 0, sizeof(*macvlan));
2865 /* Vendor and Device ID need to be set before init of shared code */
2866 memset(hw, 0, sizeof(*hw));
2867 hw->device_id = pdev->id.device_id;
2868 hw->vendor_id = pdev->id.vendor_id;
2869 hw->subsystem_device_id = pdev->id.subsystem_device_id;
2870 hw->subsystem_vendor_id = pdev->id.subsystem_vendor_id;
2871 hw->revision_id = 0;
2872 hw->hw_addr = (void *)pdev->mem_resource[0].addr;
2873 if (hw->hw_addr == NULL) {
2874 PMD_INIT_LOG(ERR, "Bad mem resource."
2875 " Try to blacklist unused devices.");
2879 /* Store fm10k_adapter pointer */
2880 hw->back = dev->data->dev_private;
2882 /* Initialize the shared code */
2883 diag = fm10k_init_shared_code(hw);
2884 if (diag != FM10K_SUCCESS) {
2885 PMD_INIT_LOG(ERR, "Shared code init failed: %d", diag);
2889 /* Initialize parameters */
2890 fm10k_params_init(dev);
2892 /* Initialize the hw */
2893 diag = fm10k_init_hw(hw);
2894 if (diag != FM10K_SUCCESS) {
2895 PMD_INIT_LOG(ERR, "Hardware init failed: %d", diag);
2899 /* Initialize MAC address(es) */
2900 dev->data->mac_addrs = rte_zmalloc("fm10k",
2901 ETHER_ADDR_LEN * FM10K_MAX_MACADDR_NUM, 0);
2902 if (dev->data->mac_addrs == NULL) {
2903 PMD_INIT_LOG(ERR, "Cannot allocate memory for MAC addresses");
2907 diag = fm10k_read_mac_addr(hw);
2909 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
2910 &dev->data->mac_addrs[0]);
2912 if (diag != FM10K_SUCCESS ||
2913 !is_valid_assigned_ether_addr(dev->data->mac_addrs)) {
2915 /* Generate a random addr */
2916 eth_random_addr(hw->mac.addr);
2917 memcpy(hw->mac.perm_addr, hw->mac.addr, ETH_ALEN);
2918 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
2919 &dev->data->mac_addrs[0]);
2922 /* Reset the hw statistics */
2923 fm10k_stats_reset(dev);
2926 diag = fm10k_reset_hw(hw);
2927 if (diag != FM10K_SUCCESS) {
2928 PMD_INIT_LOG(ERR, "Hardware reset failed: %d", diag);
2932 /* Setup mailbox service */
2933 diag = fm10k_setup_mbx_service(hw);
2934 if (diag != FM10K_SUCCESS) {
2935 PMD_INIT_LOG(ERR, "Failed to setup mailbox: %d", diag);
2939 /*PF/VF has different interrupt handling mechanism */
2940 if (hw->mac.type == fm10k_mac_pf) {
2941 /* register callback func to eal lib */
2942 rte_intr_callback_register(intr_handle,
2943 fm10k_dev_interrupt_handler_pf, (void *)dev);
2945 /* enable MISC interrupt */
2946 fm10k_dev_enable_intr_pf(dev);
2948 rte_intr_callback_register(intr_handle,
2949 fm10k_dev_interrupt_handler_vf, (void *)dev);
2951 fm10k_dev_enable_intr_vf(dev);
2954 /* Enable intr after callback registered */
2955 rte_intr_enable(intr_handle);
2957 hw->mac.ops.update_int_moderator(hw);
2959 /* Make sure Switch Manager is ready before going forward. */
2960 if (hw->mac.type == fm10k_mac_pf) {
2961 int switch_ready = 0;
2963 for (i = 0; i < MAX_QUERY_SWITCH_STATE_TIMES; i++) {
2965 hw->mac.ops.get_host_state(hw, &switch_ready);
2966 fm10k_mbx_unlock(hw);
2969 /* Delay some time to acquire async LPORT_MAP info. */
2970 rte_delay_us(WAIT_SWITCH_MSG_US);
2973 if (switch_ready == 0) {
2974 PMD_INIT_LOG(ERR, "switch is not ready");
2980 * Below function will trigger operations on mailbox, acquire lock to
2981 * avoid race condition from interrupt handler. Operations on mailbox
2982 * FIFO will trigger interrupt to PF/SM, in which interrupt handler
2983 * will handle and generate an interrupt to our side. Then, FIFO in
2984 * mailbox will be touched.
2987 /* Enable port first */
2988 hw->mac.ops.update_lport_state(hw, hw->mac.dglort_map,
2991 /* Set unicast mode by default. App can change to other mode in other
2994 hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
2995 FM10K_XCAST_MODE_NONE);
2997 fm10k_mbx_unlock(hw);
2999 /* Make sure default VID is ready before going forward. */
3000 if (hw->mac.type == fm10k_mac_pf) {
3001 for (i = 0; i < MAX_QUERY_SWITCH_STATE_TIMES; i++) {
3002 if (hw->mac.default_vid)
3004 /* Delay some time to acquire async port VLAN info. */
3005 rte_delay_us(WAIT_SWITCH_MSG_US);
3008 if (!hw->mac.default_vid) {
3009 PMD_INIT_LOG(ERR, "default VID is not ready");
3014 /* Add default mac address */
3015 fm10k_MAC_filter_set(dev, hw->mac.addr, true,
3016 MAIN_VSI_POOL_NUMBER);
3022 eth_fm10k_dev_uninit(struct rte_eth_dev *dev)
3024 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
3025 struct rte_pci_device *pdev = RTE_DEV_TO_PCI(dev->device);
3026 struct rte_intr_handle *intr_handle = &pdev->intr_handle;
3027 PMD_INIT_FUNC_TRACE();
3029 /* only uninitialize in the primary process */
3030 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3033 /* safe to close dev here */
3034 fm10k_dev_close(dev);
3036 dev->dev_ops = NULL;
3037 dev->rx_pkt_burst = NULL;
3038 dev->tx_pkt_burst = NULL;
3040 /* disable uio/vfio intr */
3041 rte_intr_disable(intr_handle);
3043 /*PF/VF has different interrupt handling mechanism */
3044 if (hw->mac.type == fm10k_mac_pf) {
3045 /* disable interrupt */
3046 fm10k_dev_disable_intr_pf(dev);
3048 /* unregister callback func to eal lib */
3049 rte_intr_callback_unregister(intr_handle,
3050 fm10k_dev_interrupt_handler_pf, (void *)dev);
3052 /* disable interrupt */
3053 fm10k_dev_disable_intr_vf(dev);
3055 rte_intr_callback_unregister(intr_handle,
3056 fm10k_dev_interrupt_handler_vf, (void *)dev);
3059 /* free mac memory */
3060 if (dev->data->mac_addrs) {
3061 rte_free(dev->data->mac_addrs);
3062 dev->data->mac_addrs = NULL;
3065 memset(hw, 0, sizeof(*hw));
3071 * The set of PCI devices this driver supports. This driver will enable both PF
3072 * and SRIOV-VF devices.
3074 static const struct rte_pci_id pci_id_fm10k_map[] = {
3075 { RTE_PCI_DEVICE(FM10K_INTEL_VENDOR_ID, FM10K_DEV_ID_PF) },
3076 { RTE_PCI_DEVICE(FM10K_INTEL_VENDOR_ID, FM10K_DEV_ID_SDI_FM10420_QDA2) },
3077 { RTE_PCI_DEVICE(FM10K_INTEL_VENDOR_ID, FM10K_DEV_ID_VF) },
3078 { .vendor_id = 0, /* sentinel */ },
3081 static struct eth_driver rte_pmd_fm10k = {
3083 .id_table = pci_id_fm10k_map,
3084 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
3085 .probe = rte_eth_dev_pci_probe,
3086 .remove = rte_eth_dev_pci_remove,
3088 .eth_dev_init = eth_fm10k_dev_init,
3089 .eth_dev_uninit = eth_fm10k_dev_uninit,
3090 .dev_private_size = sizeof(struct fm10k_adapter),
3093 RTE_PMD_REGISTER_PCI(net_fm10k, rte_pmd_fm10k.pci_drv);
3094 RTE_PMD_REGISTER_PCI_TABLE(net_fm10k, pci_id_fm10k_map);
3095 RTE_PMD_REGISTER_KMOD_DEP(net_fm10k, "* igb_uio | uio_pci_generic | vfio");