4 * Copyright(c) 2013-2015 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>
42 #include "base/fm10k_api.h"
44 /* Default delay to acquire mailbox lock */
45 #define FM10K_MBXLOCK_DELAY_US 20
46 #define UINT64_LOWER_32BITS_MASK 0x00000000ffffffffULL
48 /* Max try times to acquire switch status */
49 #define MAX_QUERY_SWITCH_STATE_TIMES 10
50 /* Wait interval to get switch status */
51 #define WAIT_SWITCH_MSG_US 100000
52 /* Number of chars per uint32 type */
53 #define CHARS_PER_UINT32 (sizeof(uint32_t))
54 #define BIT_MASK_PER_UINT32 ((1 << CHARS_PER_UINT32) - 1)
56 static void fm10k_close_mbx_service(struct fm10k_hw *hw);
57 static void fm10k_dev_promiscuous_enable(struct rte_eth_dev *dev);
58 static void fm10k_dev_promiscuous_disable(struct rte_eth_dev *dev);
59 static void fm10k_dev_allmulticast_enable(struct rte_eth_dev *dev);
60 static void fm10k_dev_allmulticast_disable(struct rte_eth_dev *dev);
61 static inline int fm10k_glort_valid(struct fm10k_hw *hw);
63 fm10k_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on);
65 fm10k_MAC_filter_set(struct rte_eth_dev *dev, const u8 *mac, bool add);
67 fm10k_MACVLAN_remove_all(struct rte_eth_dev *dev);
70 fm10k_mbx_initlock(struct fm10k_hw *hw)
72 rte_spinlock_init(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back));
76 fm10k_mbx_lock(struct fm10k_hw *hw)
78 while (!rte_spinlock_trylock(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back)))
79 rte_delay_us(FM10K_MBXLOCK_DELAY_US);
83 fm10k_mbx_unlock(struct fm10k_hw *hw)
85 rte_spinlock_unlock(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back));
89 * reset queue to initial state, allocate software buffers used when starting
92 * return -ENOMEM if buffers cannot be allocated
93 * return -EINVAL if buffers do not satisfy alignment condition
96 rx_queue_reset(struct fm10k_rx_queue *q)
100 PMD_INIT_FUNC_TRACE();
102 diag = rte_mempool_get_bulk(q->mp, (void **)q->sw_ring, q->nb_desc);
106 for (i = 0; i < q->nb_desc; ++i) {
107 fm10k_pktmbuf_reset(q->sw_ring[i], q->port_id);
108 if (!fm10k_addr_alignment_valid(q->sw_ring[i])) {
109 rte_mempool_put_bulk(q->mp, (void **)q->sw_ring,
113 dma_addr = MBUF_DMA_ADDR_DEFAULT(q->sw_ring[i]);
114 q->hw_ring[i].q.pkt_addr = dma_addr;
115 q->hw_ring[i].q.hdr_addr = dma_addr;
120 q->next_trigger = q->alloc_thresh - 1;
121 FM10K_PCI_REG_WRITE(q->tail_ptr, q->nb_desc - 1);
126 * clean queue, descriptor rings, free software buffers used when stopping
130 rx_queue_clean(struct fm10k_rx_queue *q)
132 union fm10k_rx_desc zero = {.q = {0, 0, 0, 0} };
134 PMD_INIT_FUNC_TRACE();
136 /* zero descriptor rings */
137 for (i = 0; i < q->nb_desc; ++i)
138 q->hw_ring[i] = zero;
140 /* free software buffers */
141 for (i = 0; i < q->nb_desc; ++i) {
143 rte_pktmbuf_free_seg(q->sw_ring[i]);
144 q->sw_ring[i] = NULL;
150 * free all queue memory used when releasing the queue (i.e. configure)
153 rx_queue_free(struct fm10k_rx_queue *q)
155 PMD_INIT_FUNC_TRACE();
157 PMD_INIT_LOG(DEBUG, "Freeing rx queue %p", q);
160 rte_free(q->sw_ring);
169 * disable RX queue, wait unitl HW finished necessary flush operation
172 rx_queue_disable(struct fm10k_hw *hw, uint16_t qnum)
176 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(qnum));
177 FM10K_WRITE_REG(hw, FM10K_RXQCTL(qnum),
178 reg & ~FM10K_RXQCTL_ENABLE);
180 /* Wait 100us at most */
181 for (i = 0; i < FM10K_QUEUE_DISABLE_TIMEOUT; i++) {
183 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(i));
184 if (!(reg & FM10K_RXQCTL_ENABLE))
188 if (i == FM10K_QUEUE_DISABLE_TIMEOUT)
195 * reset queue to initial state, allocate software buffers used when starting
199 tx_queue_reset(struct fm10k_tx_queue *q)
201 PMD_INIT_FUNC_TRACE();
205 q->nb_free = q->nb_desc - 1;
206 q->free_trigger = q->nb_free - q->free_thresh;
207 fifo_reset(&q->rs_tracker, (q->nb_desc + 1) / q->rs_thresh);
208 FM10K_PCI_REG_WRITE(q->tail_ptr, 0);
212 * clean queue, descriptor rings, free software buffers used when stopping
216 tx_queue_clean(struct fm10k_tx_queue *q)
218 struct fm10k_tx_desc zero = {0, 0, 0, 0, 0, 0};
220 PMD_INIT_FUNC_TRACE();
222 /* zero descriptor rings */
223 for (i = 0; i < q->nb_desc; ++i)
224 q->hw_ring[i] = zero;
226 /* free software buffers */
227 for (i = 0; i < q->nb_desc; ++i) {
229 rte_pktmbuf_free_seg(q->sw_ring[i]);
230 q->sw_ring[i] = NULL;
236 * free all queue memory used when releasing the queue (i.e. configure)
239 tx_queue_free(struct fm10k_tx_queue *q)
241 PMD_INIT_FUNC_TRACE();
243 PMD_INIT_LOG(DEBUG, "Freeing tx queue %p", q);
245 if (q->rs_tracker.list) {
246 rte_free(q->rs_tracker.list);
247 q->rs_tracker.list = NULL;
250 rte_free(q->sw_ring);
259 * disable TX queue, wait unitl HW finished necessary flush operation
262 tx_queue_disable(struct fm10k_hw *hw, uint16_t qnum)
266 reg = FM10K_READ_REG(hw, FM10K_TXDCTL(qnum));
267 FM10K_WRITE_REG(hw, FM10K_TXDCTL(qnum),
268 reg & ~FM10K_TXDCTL_ENABLE);
270 /* Wait 100us at most */
271 for (i = 0; i < FM10K_QUEUE_DISABLE_TIMEOUT; i++) {
273 reg = FM10K_READ_REG(hw, FM10K_TXDCTL(i));
274 if (!(reg & FM10K_TXDCTL_ENABLE))
278 if (i == FM10K_QUEUE_DISABLE_TIMEOUT)
285 fm10k_dev_configure(struct rte_eth_dev *dev)
287 PMD_INIT_FUNC_TRACE();
289 if (dev->data->dev_conf.rxmode.hw_strip_crc == 0)
290 PMD_INIT_LOG(WARNING, "fm10k always strip CRC");
296 fm10k_dev_mq_rx_configure(struct rte_eth_dev *dev)
298 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
299 struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
300 uint32_t mrqc, *key, i, reta, j;
303 #define RSS_KEY_SIZE 40
304 static uint8_t rss_intel_key[RSS_KEY_SIZE] = {
305 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
306 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
307 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
308 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
309 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
312 if (dev->data->nb_rx_queues == 1 ||
313 dev_conf->rxmode.mq_mode != ETH_MQ_RX_RSS ||
314 dev_conf->rx_adv_conf.rss_conf.rss_hf == 0)
317 /* random key is rss_intel_key (default) or user provided (rss_key) */
318 if (dev_conf->rx_adv_conf.rss_conf.rss_key == NULL)
319 key = (uint32_t *)rss_intel_key;
321 key = (uint32_t *)dev_conf->rx_adv_conf.rss_conf.rss_key;
323 /* Now fill our hash function seeds, 4 bytes at a time */
324 for (i = 0; i < RSS_KEY_SIZE / sizeof(*key); ++i)
325 FM10K_WRITE_REG(hw, FM10K_RSSRK(0, i), key[i]);
328 * Fill in redirection table
329 * The byte-swap is needed because NIC registers are in
330 * little-endian order.
333 for (i = 0, j = 0; i < FM10K_RETA_SIZE; i++, j++) {
334 if (j == dev->data->nb_rx_queues)
336 reta = (reta << CHAR_BIT) | j;
338 FM10K_WRITE_REG(hw, FM10K_RETA(0, i >> 2),
343 * Generate RSS hash based on packet types, TCP/UDP
344 * port numbers and/or IPv4/v6 src and dst addresses
346 hf = dev_conf->rx_adv_conf.rss_conf.rss_hf;
348 mrqc |= (hf & ETH_RSS_IPV4) ? FM10K_MRQC_IPV4 : 0;
349 mrqc |= (hf & ETH_RSS_IPV6) ? FM10K_MRQC_IPV6 : 0;
350 mrqc |= (hf & ETH_RSS_IPV6_EX) ? FM10K_MRQC_IPV6 : 0;
351 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_TCP) ? FM10K_MRQC_TCP_IPV4 : 0;
352 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_TCP) ? FM10K_MRQC_TCP_IPV6 : 0;
353 mrqc |= (hf & ETH_RSS_IPV6_TCP_EX) ? FM10K_MRQC_TCP_IPV6 : 0;
354 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_UDP) ? FM10K_MRQC_UDP_IPV4 : 0;
355 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_UDP) ? FM10K_MRQC_UDP_IPV6 : 0;
356 mrqc |= (hf & ETH_RSS_IPV6_UDP_EX) ? FM10K_MRQC_UDP_IPV6 : 0;
359 PMD_INIT_LOG(ERR, "Specified RSS mode 0x%"PRIx64"is not"
364 FM10K_WRITE_REG(hw, FM10K_MRQC(0), mrqc);
368 fm10k_dev_tx_init(struct rte_eth_dev *dev)
370 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
372 struct fm10k_tx_queue *txq;
376 /* Disable TXINT to avoid possible interrupt */
377 for (i = 0; i < hw->mac.max_queues; i++)
378 FM10K_WRITE_REG(hw, FM10K_TXINT(i),
379 3 << FM10K_TXINT_TIMER_SHIFT);
382 for (i = 0; i < dev->data->nb_tx_queues; ++i) {
383 txq = dev->data->tx_queues[i];
384 base_addr = txq->hw_ring_phys_addr;
385 size = txq->nb_desc * sizeof(struct fm10k_tx_desc);
387 /* disable queue to avoid issues while updating state */
388 ret = tx_queue_disable(hw, i);
390 PMD_INIT_LOG(ERR, "failed to disable queue %d", i);
394 /* set location and size for descriptor ring */
395 FM10K_WRITE_REG(hw, FM10K_TDBAL(i),
396 base_addr & UINT64_LOWER_32BITS_MASK);
397 FM10K_WRITE_REG(hw, FM10K_TDBAH(i),
398 base_addr >> (CHAR_BIT * sizeof(uint32_t)));
399 FM10K_WRITE_REG(hw, FM10K_TDLEN(i), size);
405 fm10k_dev_rx_init(struct rte_eth_dev *dev)
407 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
409 struct fm10k_rx_queue *rxq;
412 uint32_t rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY;
415 /* Disable RXINT to avoid possible interrupt */
416 for (i = 0; i < hw->mac.max_queues; i++)
417 FM10K_WRITE_REG(hw, FM10K_RXINT(i),
418 3 << FM10K_RXINT_TIMER_SHIFT);
420 /* Setup RX queues */
421 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
422 rxq = dev->data->rx_queues[i];
423 base_addr = rxq->hw_ring_phys_addr;
424 size = rxq->nb_desc * sizeof(union fm10k_rx_desc);
426 /* disable queue to avoid issues while updating state */
427 ret = rx_queue_disable(hw, i);
429 PMD_INIT_LOG(ERR, "failed to disable queue %d", i);
433 /* Setup the Base and Length of the Rx Descriptor Ring */
434 FM10K_WRITE_REG(hw, FM10K_RDBAL(i),
435 base_addr & UINT64_LOWER_32BITS_MASK);
436 FM10K_WRITE_REG(hw, FM10K_RDBAH(i),
437 base_addr >> (CHAR_BIT * sizeof(uint32_t)));
438 FM10K_WRITE_REG(hw, FM10K_RDLEN(i), size);
440 /* Configure the Rx buffer size for one buff without split */
441 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
442 RTE_PKTMBUF_HEADROOM);
443 /* As RX buffer is aligned to 512B within mbuf, some bytes are
444 * reserved for this purpose, and the worst case could be 511B.
445 * But SRR reg assumes all buffers have the same size. In order
446 * to fill the gap, we'll have to consider the worst case and
447 * assume 512B is reserved. If we don't do so, it's possible
448 * for HW to overwrite data to next mbuf.
450 buf_size -= FM10K_RX_DATABUF_ALIGN;
452 FM10K_WRITE_REG(hw, FM10K_SRRCTL(i),
453 buf_size >> FM10K_SRRCTL_BSIZEPKT_SHIFT);
455 /* It adds dual VLAN length for supporting dual VLAN */
456 if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
457 2 * FM10K_VLAN_TAG_SIZE) > buf_size ||
458 dev->data->dev_conf.rxmode.enable_scatter) {
460 dev->data->scattered_rx = 1;
461 dev->rx_pkt_burst = fm10k_recv_scattered_pkts;
462 reg = FM10K_READ_REG(hw, FM10K_SRRCTL(i));
463 reg |= FM10K_SRRCTL_BUFFER_CHAINING_EN;
464 FM10K_WRITE_REG(hw, FM10K_SRRCTL(i), reg);
467 /* Enable drop on empty, it's RO for VF */
468 if (hw->mac.type == fm10k_mac_pf && rxq->drop_en)
469 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY;
471 FM10K_WRITE_REG(hw, FM10K_RXDCTL(i), rxdctl);
472 FM10K_WRITE_FLUSH(hw);
475 /* Configure RSS if applicable */
476 fm10k_dev_mq_rx_configure(dev);
481 fm10k_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
483 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
486 struct fm10k_rx_queue *rxq;
488 PMD_INIT_FUNC_TRACE();
490 if (rx_queue_id < dev->data->nb_rx_queues) {
491 rxq = dev->data->rx_queues[rx_queue_id];
492 err = rx_queue_reset(rxq);
493 if (err == -ENOMEM) {
494 PMD_INIT_LOG(ERR, "Failed to alloc memory : %d", err);
496 } else if (err == -EINVAL) {
497 PMD_INIT_LOG(ERR, "Invalid buffer address alignment :"
502 /* Setup the HW Rx Head and Tail Descriptor Pointers
503 * Note: this must be done AFTER the queue is enabled on real
504 * hardware, but BEFORE the queue is enabled when using the
505 * emulation platform. Do it in both places for now and remove
506 * this comment and the following two register writes when the
507 * emulation platform is no longer being used.
509 FM10K_WRITE_REG(hw, FM10K_RDH(rx_queue_id), 0);
510 FM10K_WRITE_REG(hw, FM10K_RDT(rx_queue_id), rxq->nb_desc - 1);
512 /* Set PF ownership flag for PF devices */
513 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(rx_queue_id));
514 if (hw->mac.type == fm10k_mac_pf)
515 reg |= FM10K_RXQCTL_PF;
516 reg |= FM10K_RXQCTL_ENABLE;
517 /* enable RX queue */
518 FM10K_WRITE_REG(hw, FM10K_RXQCTL(rx_queue_id), reg);
519 FM10K_WRITE_FLUSH(hw);
521 /* Setup the HW Rx Head and Tail Descriptor Pointers
522 * Note: this must be done AFTER the queue is enabled
524 FM10K_WRITE_REG(hw, FM10K_RDH(rx_queue_id), 0);
525 FM10K_WRITE_REG(hw, FM10K_RDT(rx_queue_id), rxq->nb_desc - 1);
532 fm10k_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
534 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
536 PMD_INIT_FUNC_TRACE();
538 if (rx_queue_id < dev->data->nb_rx_queues) {
539 /* Disable RX queue */
540 rx_queue_disable(hw, rx_queue_id);
542 /* Free mbuf and clean HW ring */
543 rx_queue_clean(dev->data->rx_queues[rx_queue_id]);
550 fm10k_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
552 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
553 /** @todo - this should be defined in the shared code */
554 #define FM10K_TXDCTL_WRITE_BACK_MIN_DELAY 0x00010000
555 uint32_t txdctl = FM10K_TXDCTL_WRITE_BACK_MIN_DELAY;
558 PMD_INIT_FUNC_TRACE();
560 if (tx_queue_id < dev->data->nb_tx_queues) {
561 tx_queue_reset(dev->data->tx_queues[tx_queue_id]);
563 /* reset head and tail pointers */
564 FM10K_WRITE_REG(hw, FM10K_TDH(tx_queue_id), 0);
565 FM10K_WRITE_REG(hw, FM10K_TDT(tx_queue_id), 0);
567 /* enable TX queue */
568 FM10K_WRITE_REG(hw, FM10K_TXDCTL(tx_queue_id),
569 FM10K_TXDCTL_ENABLE | txdctl);
570 FM10K_WRITE_FLUSH(hw);
578 fm10k_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
580 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
582 PMD_INIT_FUNC_TRACE();
584 if (tx_queue_id < dev->data->nb_tx_queues) {
585 tx_queue_disable(hw, tx_queue_id);
586 tx_queue_clean(dev->data->tx_queues[tx_queue_id]);
592 static inline int fm10k_glort_valid(struct fm10k_hw *hw)
594 return ((hw->mac.dglort_map & FM10K_DGLORTMAP_NONE)
595 != FM10K_DGLORTMAP_NONE);
599 fm10k_dev_promiscuous_enable(struct rte_eth_dev *dev)
601 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
604 PMD_INIT_FUNC_TRACE();
606 /* Return if it didn't acquire valid glort range */
607 if (!fm10k_glort_valid(hw))
611 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
612 FM10K_XCAST_MODE_PROMISC);
613 fm10k_mbx_unlock(hw);
615 if (status != FM10K_SUCCESS)
616 PMD_INIT_LOG(ERR, "Failed to enable promiscuous mode");
620 fm10k_dev_promiscuous_disable(struct rte_eth_dev *dev)
622 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
626 PMD_INIT_FUNC_TRACE();
628 /* Return if it didn't acquire valid glort range */
629 if (!fm10k_glort_valid(hw))
632 if (dev->data->all_multicast == 1)
633 mode = FM10K_XCAST_MODE_ALLMULTI;
635 mode = FM10K_XCAST_MODE_NONE;
638 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
640 fm10k_mbx_unlock(hw);
642 if (status != FM10K_SUCCESS)
643 PMD_INIT_LOG(ERR, "Failed to disable promiscuous mode");
647 fm10k_dev_allmulticast_enable(struct rte_eth_dev *dev)
649 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
652 PMD_INIT_FUNC_TRACE();
654 /* Return if it didn't acquire valid glort range */
655 if (!fm10k_glort_valid(hw))
658 /* If promiscuous mode is enabled, it doesn't make sense to enable
659 * allmulticast and disable promiscuous since fm10k only can select
662 if (dev->data->promiscuous) {
663 PMD_INIT_LOG(INFO, "Promiscuous mode is enabled, "\
664 "needn't enable allmulticast");
669 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
670 FM10K_XCAST_MODE_ALLMULTI);
671 fm10k_mbx_unlock(hw);
673 if (status != FM10K_SUCCESS)
674 PMD_INIT_LOG(ERR, "Failed to enable allmulticast mode");
678 fm10k_dev_allmulticast_disable(struct rte_eth_dev *dev)
680 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
683 PMD_INIT_FUNC_TRACE();
685 /* Return if it didn't acquire valid glort range */
686 if (!fm10k_glort_valid(hw))
689 if (dev->data->promiscuous) {
690 PMD_INIT_LOG(ERR, "Failed to disable allmulticast mode "\
691 "since promisc mode is enabled");
696 /* Change mode to unicast mode */
697 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
698 FM10K_XCAST_MODE_NONE);
699 fm10k_mbx_unlock(hw);
701 if (status != FM10K_SUCCESS)
702 PMD_INIT_LOG(ERR, "Failed to disable allmulticast mode");
705 /* fls = find last set bit = 32 minus the number of leading zeros */
707 #define fls(x) (((x) == 0) ? 0 : (32 - __builtin_clz((x))))
709 #define BSIZEPKT_ROUNDUP ((1 << FM10K_SRRCTL_BSIZEPKT_SHIFT) - 1)
711 fm10k_dev_start(struct rte_eth_dev *dev)
713 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
716 PMD_INIT_FUNC_TRACE();
718 /* stop, init, then start the hw */
719 diag = fm10k_stop_hw(hw);
720 if (diag != FM10K_SUCCESS) {
721 PMD_INIT_LOG(ERR, "Hardware stop failed: %d", diag);
725 diag = fm10k_init_hw(hw);
726 if (diag != FM10K_SUCCESS) {
727 PMD_INIT_LOG(ERR, "Hardware init failed: %d", diag);
731 diag = fm10k_start_hw(hw);
732 if (diag != FM10K_SUCCESS) {
733 PMD_INIT_LOG(ERR, "Hardware start failed: %d", diag);
737 diag = fm10k_dev_tx_init(dev);
739 PMD_INIT_LOG(ERR, "TX init failed: %d", diag);
743 diag = fm10k_dev_rx_init(dev);
745 PMD_INIT_LOG(ERR, "RX init failed: %d", diag);
749 if (hw->mac.type == fm10k_mac_pf) {
750 /* Establish only VSI 0 as valid */
751 FM10K_WRITE_REG(hw, FM10K_DGLORTMAP(0), FM10K_DGLORTMAP_ANY);
753 /* Configure RSS bits used in RETA table */
754 FM10K_WRITE_REG(hw, FM10K_DGLORTDEC(0),
755 fls(dev->data->nb_rx_queues - 1) <<
756 FM10K_DGLORTDEC_RSSLENGTH_SHIFT);
758 /* Invalidate all other GLORT entries */
759 for (i = 1; i < FM10K_DGLORT_COUNT; i++)
760 FM10K_WRITE_REG(hw, FM10K_DGLORTMAP(i),
761 FM10K_DGLORTMAP_NONE);
764 for (i = 0; i < dev->data->nb_rx_queues; i++) {
765 struct fm10k_rx_queue *rxq;
766 rxq = dev->data->rx_queues[i];
768 if (rxq->rx_deferred_start)
770 diag = fm10k_dev_rx_queue_start(dev, i);
773 for (j = 0; j < i; ++j)
774 rx_queue_clean(dev->data->rx_queues[j]);
779 for (i = 0; i < dev->data->nb_tx_queues; i++) {
780 struct fm10k_tx_queue *txq;
781 txq = dev->data->tx_queues[i];
783 if (txq->tx_deferred_start)
785 diag = fm10k_dev_tx_queue_start(dev, i);
788 for (j = 0; j < dev->data->nb_rx_queues; ++j)
789 rx_queue_clean(dev->data->rx_queues[j]);
794 if (hw->mac.default_vid && hw->mac.default_vid <= ETHER_MAX_VLAN_ID) {
795 /* Update default vlan */
796 fm10k_vlan_filter_set(dev, hw->mac.default_vid, true);
798 /* Add default mac/vlan filter to PF/Switch manager */
799 fm10k_MAC_filter_set(dev, hw->mac.addr, true);
806 fm10k_dev_stop(struct rte_eth_dev *dev)
810 PMD_INIT_FUNC_TRACE();
812 for (i = 0; i < dev->data->nb_tx_queues; i++)
813 fm10k_dev_tx_queue_stop(dev, i);
815 for (i = 0; i < dev->data->nb_rx_queues; i++)
816 fm10k_dev_rx_queue_stop(dev, i);
820 fm10k_dev_close(struct rte_eth_dev *dev)
822 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
824 PMD_INIT_FUNC_TRACE();
826 fm10k_MACVLAN_remove_all(dev);
828 /* Stop mailbox service first */
829 fm10k_close_mbx_service(hw);
835 fm10k_link_update(struct rte_eth_dev *dev,
836 __rte_unused int wait_to_complete)
838 PMD_INIT_FUNC_TRACE();
840 /* The host-interface link is always up. The speed is ~50Gbps per Gen3
841 * x8 PCIe interface. For now, we leave the speed undefined since there
842 * is no 50Gbps Ethernet. */
843 dev->data->dev_link.link_speed = 0;
844 dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
845 dev->data->dev_link.link_status = 1;
851 fm10k_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
853 uint64_t ipackets, opackets, ibytes, obytes;
854 struct fm10k_hw *hw =
855 FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
856 struct fm10k_hw_stats *hw_stats =
857 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
860 PMD_INIT_FUNC_TRACE();
862 fm10k_update_hw_stats(hw, hw_stats);
864 ipackets = opackets = ibytes = obytes = 0;
865 for (i = 0; (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) &&
866 (i < hw->mac.max_queues); ++i) {
867 stats->q_ipackets[i] = hw_stats->q[i].rx_packets.count;
868 stats->q_opackets[i] = hw_stats->q[i].tx_packets.count;
869 stats->q_ibytes[i] = hw_stats->q[i].rx_bytes.count;
870 stats->q_obytes[i] = hw_stats->q[i].tx_bytes.count;
871 ipackets += stats->q_ipackets[i];
872 opackets += stats->q_opackets[i];
873 ibytes += stats->q_ibytes[i];
874 obytes += stats->q_obytes[i];
876 stats->ipackets = ipackets;
877 stats->opackets = opackets;
878 stats->ibytes = ibytes;
879 stats->obytes = obytes;
883 fm10k_stats_reset(struct rte_eth_dev *dev)
885 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
886 struct fm10k_hw_stats *hw_stats =
887 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
889 PMD_INIT_FUNC_TRACE();
891 memset(hw_stats, 0, sizeof(*hw_stats));
892 fm10k_rebind_hw_stats(hw, hw_stats);
896 fm10k_dev_infos_get(struct rte_eth_dev *dev,
897 struct rte_eth_dev_info *dev_info)
899 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
901 PMD_INIT_FUNC_TRACE();
903 dev_info->min_rx_bufsize = FM10K_MIN_RX_BUF_SIZE;
904 dev_info->max_rx_pktlen = FM10K_MAX_PKT_SIZE;
905 dev_info->max_rx_queues = hw->mac.max_queues;
906 dev_info->max_tx_queues = hw->mac.max_queues;
907 dev_info->max_mac_addrs = FM10K_MAX_MACADDR_NUM;
908 dev_info->max_hash_mac_addrs = 0;
909 dev_info->max_vfs = dev->pci_dev->max_vfs;
910 dev_info->max_vmdq_pools = ETH_64_POOLS;
911 dev_info->rx_offload_capa =
912 DEV_RX_OFFLOAD_VLAN_STRIP |
913 DEV_RX_OFFLOAD_IPV4_CKSUM |
914 DEV_RX_OFFLOAD_UDP_CKSUM |
915 DEV_RX_OFFLOAD_TCP_CKSUM;
916 dev_info->tx_offload_capa =
917 DEV_TX_OFFLOAD_VLAN_INSERT;
918 dev_info->reta_size = FM10K_MAX_RSS_INDICES;
920 dev_info->default_rxconf = (struct rte_eth_rxconf) {
922 .pthresh = FM10K_DEFAULT_RX_PTHRESH,
923 .hthresh = FM10K_DEFAULT_RX_HTHRESH,
924 .wthresh = FM10K_DEFAULT_RX_WTHRESH,
926 .rx_free_thresh = FM10K_RX_FREE_THRESH_DEFAULT(0),
930 dev_info->default_txconf = (struct rte_eth_txconf) {
932 .pthresh = FM10K_DEFAULT_TX_PTHRESH,
933 .hthresh = FM10K_DEFAULT_TX_HTHRESH,
934 .wthresh = FM10K_DEFAULT_TX_WTHRESH,
936 .tx_free_thresh = FM10K_TX_FREE_THRESH_DEFAULT(0),
937 .tx_rs_thresh = FM10K_TX_RS_THRESH_DEFAULT(0),
938 .txq_flags = ETH_TXQ_FLAGS_NOMULTSEGS |
939 ETH_TXQ_FLAGS_NOOFFLOADS,
945 fm10k_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
948 uint16_t mac_num = 0;
949 uint32_t vid_idx, vid_bit, mac_index;
951 struct fm10k_macvlan_filter_info *macvlan;
952 struct rte_eth_dev_data *data = dev->data;
954 hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
955 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
957 if (vlan_id > ETH_VLAN_ID_MAX) {
958 PMD_INIT_LOG(ERR, "Invalid vlan_id: must be < 4096");
962 vid_idx = FM10K_VFTA_IDX(vlan_id);
963 vid_bit = FM10K_VFTA_BIT(vlan_id);
964 /* this VLAN ID is already in the VLAN filter table, return SUCCESS */
965 if (on && (macvlan->vfta[vid_idx] & vid_bit))
967 /* this VLAN ID is NOT in the VLAN filter table, cannot remove */
968 if (!on && !(macvlan->vfta[vid_idx] & vid_bit)) {
969 PMD_INIT_LOG(ERR, "Invalid vlan_id: not existing "
970 "in the VLAN filter table");
975 result = fm10k_update_vlan(hw, vlan_id, 0, on);
976 fm10k_mbx_unlock(hw);
977 if (result != FM10K_SUCCESS) {
978 PMD_INIT_LOG(ERR, "VLAN update failed: %d", result);
982 for (mac_index = 0; (mac_index < FM10K_MAX_MACADDR_NUM) &&
983 (result == FM10K_SUCCESS); mac_index++) {
984 if (is_zero_ether_addr(&data->mac_addrs[mac_index]))
986 if (mac_num > macvlan->mac_num - 1) {
987 PMD_INIT_LOG(ERR, "MAC address number "
992 result = fm10k_update_uc_addr(hw, hw->mac.dglort_map,
993 data->mac_addrs[mac_index].addr_bytes,
995 fm10k_mbx_unlock(hw);
998 if (result != FM10K_SUCCESS) {
999 PMD_INIT_LOG(ERR, "MAC address update failed: %d", result);
1004 macvlan->vlan_num++;
1005 macvlan->vfta[vid_idx] |= vid_bit;
1007 macvlan->vlan_num--;
1008 macvlan->vfta[vid_idx] &= ~vid_bit;
1014 fm10k_vlan_offload_set(__rte_unused struct rte_eth_dev *dev, int mask)
1016 if (mask & ETH_VLAN_STRIP_MASK) {
1017 if (!dev->data->dev_conf.rxmode.hw_vlan_strip)
1018 PMD_INIT_LOG(ERR, "VLAN stripping is "
1019 "always on in fm10k");
1022 if (mask & ETH_VLAN_EXTEND_MASK) {
1023 if (dev->data->dev_conf.rxmode.hw_vlan_extend)
1024 PMD_INIT_LOG(ERR, "VLAN QinQ is not "
1025 "supported in fm10k");
1028 if (mask & ETH_VLAN_FILTER_MASK) {
1029 if (!dev->data->dev_conf.rxmode.hw_vlan_filter)
1030 PMD_INIT_LOG(ERR, "VLAN filter is always on in fm10k");
1034 /* Add/Remove a MAC address, and update filters */
1036 fm10k_MAC_filter_set(struct rte_eth_dev *dev, const u8 *mac, bool add)
1039 struct fm10k_hw *hw;
1040 struct fm10k_macvlan_filter_info *macvlan;
1042 hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1043 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1046 for (j = 0; j < FM10K_VFTA_SIZE; j++) {
1047 if (macvlan->vfta[j]) {
1048 for (k = 0; k < FM10K_UINT32_BIT_SIZE; k++) {
1049 if (macvlan->vfta[j] & (1 << k)) {
1050 if (i + 1 > macvlan->vlan_num) {
1051 PMD_INIT_LOG(ERR, "vlan number "
1056 fm10k_update_uc_addr(hw,
1057 hw->mac.dglort_map, mac,
1058 j * FM10K_UINT32_BIT_SIZE + k,
1060 fm10k_mbx_unlock(hw);
1073 /* Add a MAC address, and update filters */
1075 fm10k_macaddr_add(struct rte_eth_dev *dev,
1076 struct ether_addr *mac_addr,
1077 __rte_unused uint32_t index,
1078 __rte_unused uint32_t pool)
1080 fm10k_MAC_filter_set(dev, mac_addr->addr_bytes, TRUE);
1083 /* Remove a MAC address, and update filters */
1085 fm10k_macaddr_remove(struct rte_eth_dev *dev, uint32_t index)
1087 struct rte_eth_dev_data *data = dev->data;
1089 if (index < FM10K_MAX_MACADDR_NUM)
1090 fm10k_MAC_filter_set(dev, data->mac_addrs[index].addr_bytes,
1094 /* Remove all VLAN and MAC address table entries */
1096 fm10k_MACVLAN_remove_all(struct rte_eth_dev *dev)
1099 struct fm10k_macvlan_filter_info *macvlan;
1101 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1102 for (j = 0; j < FM10K_VFTA_SIZE; j++) {
1103 if (macvlan->vfta[j]) {
1104 for (k = 0; k < FM10K_UINT32_BIT_SIZE; k++) {
1105 if (macvlan->vfta[j] & (1 << k))
1106 fm10k_vlan_filter_set(dev,
1107 j * FM10K_UINT32_BIT_SIZE + k, false);
1114 check_nb_desc(uint16_t min, uint16_t max, uint16_t mult, uint16_t request)
1116 if ((request < min) || (request > max) || ((request % mult) != 0))
1123 * Create a memzone for hardware descriptor rings. Malloc cannot be used since
1124 * the physical address is required. If the memzone is already created, then
1125 * this function returns a pointer to the existing memzone.
1127 static inline const struct rte_memzone *
1128 allocate_hw_ring(const char *driver_name, const char *ring_name,
1129 uint8_t port_id, uint16_t queue_id, int socket_id,
1130 uint32_t size, uint32_t align)
1132 char name[RTE_MEMZONE_NAMESIZE];
1133 const struct rte_memzone *mz;
1135 snprintf(name, sizeof(name), "%s_%s_%d_%d_%d",
1136 driver_name, ring_name, port_id, queue_id, socket_id);
1138 /* return the memzone if it already exists */
1139 mz = rte_memzone_lookup(name);
1143 #ifdef RTE_LIBRTE_XEN_DOM0
1144 return rte_memzone_reserve_bounded(name, size, socket_id, 0, align,
1147 return rte_memzone_reserve_aligned(name, size, socket_id, 0, align);
1152 check_thresh(uint16_t min, uint16_t max, uint16_t div, uint16_t request)
1154 if ((request < min) || (request > max) || ((div % request) != 0))
1161 handle_rxconf(struct fm10k_rx_queue *q, const struct rte_eth_rxconf *conf)
1163 uint16_t rx_free_thresh;
1165 if (conf->rx_free_thresh == 0)
1166 rx_free_thresh = FM10K_RX_FREE_THRESH_DEFAULT(q);
1168 rx_free_thresh = conf->rx_free_thresh;
1170 /* make sure the requested threshold satisfies the constraints */
1171 if (check_thresh(FM10K_RX_FREE_THRESH_MIN(q),
1172 FM10K_RX_FREE_THRESH_MAX(q),
1173 FM10K_RX_FREE_THRESH_DIV(q),
1175 PMD_INIT_LOG(ERR, "rx_free_thresh (%u) must be "
1176 "less than or equal to %u, "
1177 "greater than or equal to %u, "
1178 "and a divisor of %u",
1179 rx_free_thresh, FM10K_RX_FREE_THRESH_MAX(q),
1180 FM10K_RX_FREE_THRESH_MIN(q),
1181 FM10K_RX_FREE_THRESH_DIV(q));
1185 q->alloc_thresh = rx_free_thresh;
1186 q->drop_en = conf->rx_drop_en;
1187 q->rx_deferred_start = conf->rx_deferred_start;
1193 * Hardware requires specific alignment for Rx packet buffers. At
1194 * least one of the following two conditions must be satisfied.
1195 * 1. Address is 512B aligned
1196 * 2. Address is 8B aligned and buffer does not cross 4K boundary.
1198 * As such, the driver may need to adjust the DMA address within the
1199 * buffer by up to 512B.
1201 * return 1 if the element size is valid, otherwise return 0.
1204 mempool_element_size_valid(struct rte_mempool *mp)
1208 /* elt_size includes mbuf header and headroom */
1209 min_size = mp->elt_size - sizeof(struct rte_mbuf) -
1210 RTE_PKTMBUF_HEADROOM;
1212 /* account for up to 512B of alignment */
1213 min_size -= FM10K_RX_DATABUF_ALIGN;
1215 /* sanity check for overflow */
1216 if (min_size > mp->elt_size)
1224 fm10k_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
1225 uint16_t nb_desc, unsigned int socket_id,
1226 const struct rte_eth_rxconf *conf, struct rte_mempool *mp)
1228 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1229 struct fm10k_rx_queue *q;
1230 const struct rte_memzone *mz;
1232 PMD_INIT_FUNC_TRACE();
1234 /* make sure the mempool element size can account for alignment. */
1235 if (!mempool_element_size_valid(mp)) {
1236 PMD_INIT_LOG(ERR, "Error : Mempool element size is too small");
1240 /* make sure a valid number of descriptors have been requested */
1241 if (check_nb_desc(FM10K_MIN_RX_DESC, FM10K_MAX_RX_DESC,
1242 FM10K_MULT_RX_DESC, nb_desc)) {
1243 PMD_INIT_LOG(ERR, "Number of Rx descriptors (%u) must be "
1244 "less than or equal to %"PRIu32", "
1245 "greater than or equal to %u, "
1246 "and a multiple of %u",
1247 nb_desc, (uint32_t)FM10K_MAX_RX_DESC, FM10K_MIN_RX_DESC,
1248 FM10K_MULT_RX_DESC);
1253 * if this queue existed already, free the associated memory. The
1254 * queue cannot be reused in case we need to allocate memory on
1255 * different socket than was previously used.
1257 if (dev->data->rx_queues[queue_id] != NULL) {
1258 rx_queue_free(dev->data->rx_queues[queue_id]);
1259 dev->data->rx_queues[queue_id] = NULL;
1262 /* allocate memory for the queue structure */
1263 q = rte_zmalloc_socket("fm10k", sizeof(*q), RTE_CACHE_LINE_SIZE,
1266 PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
1272 q->nb_desc = nb_desc;
1273 q->port_id = dev->data->port_id;
1274 q->queue_id = queue_id;
1275 q->tail_ptr = (volatile uint32_t *)
1276 &((uint32_t *)hw->hw_addr)[FM10K_RDT(queue_id)];
1277 if (handle_rxconf(q, conf))
1280 /* allocate memory for the software ring */
1281 q->sw_ring = rte_zmalloc_socket("fm10k sw ring",
1282 nb_desc * sizeof(struct rte_mbuf *),
1283 RTE_CACHE_LINE_SIZE, socket_id);
1284 if (q->sw_ring == NULL) {
1285 PMD_INIT_LOG(ERR, "Cannot allocate software ring");
1291 * allocate memory for the hardware descriptor ring. A memzone large
1292 * enough to hold the maximum ring size is requested to allow for
1293 * resizing in later calls to the queue setup function.
1295 mz = allocate_hw_ring(dev->driver->pci_drv.name, "rx_ring",
1296 dev->data->port_id, queue_id, socket_id,
1297 FM10K_MAX_RX_RING_SZ, FM10K_ALIGN_RX_DESC);
1299 PMD_INIT_LOG(ERR, "Cannot allocate hardware ring");
1300 rte_free(q->sw_ring);
1304 q->hw_ring = mz->addr;
1305 q->hw_ring_phys_addr = mz->phys_addr;
1307 dev->data->rx_queues[queue_id] = q;
1312 fm10k_rx_queue_release(void *queue)
1314 PMD_INIT_FUNC_TRACE();
1316 rx_queue_free(queue);
1320 handle_txconf(struct fm10k_tx_queue *q, const struct rte_eth_txconf *conf)
1322 uint16_t tx_free_thresh;
1323 uint16_t tx_rs_thresh;
1325 /* constraint MACROs require that tx_free_thresh is configured
1326 * before tx_rs_thresh */
1327 if (conf->tx_free_thresh == 0)
1328 tx_free_thresh = FM10K_TX_FREE_THRESH_DEFAULT(q);
1330 tx_free_thresh = conf->tx_free_thresh;
1332 /* make sure the requested threshold satisfies the constraints */
1333 if (check_thresh(FM10K_TX_FREE_THRESH_MIN(q),
1334 FM10K_TX_FREE_THRESH_MAX(q),
1335 FM10K_TX_FREE_THRESH_DIV(q),
1337 PMD_INIT_LOG(ERR, "tx_free_thresh (%u) must be "
1338 "less than or equal to %u, "
1339 "greater than or equal to %u, "
1340 "and a divisor of %u",
1341 tx_free_thresh, FM10K_TX_FREE_THRESH_MAX(q),
1342 FM10K_TX_FREE_THRESH_MIN(q),
1343 FM10K_TX_FREE_THRESH_DIV(q));
1347 q->free_thresh = tx_free_thresh;
1349 if (conf->tx_rs_thresh == 0)
1350 tx_rs_thresh = FM10K_TX_RS_THRESH_DEFAULT(q);
1352 tx_rs_thresh = conf->tx_rs_thresh;
1354 q->tx_deferred_start = conf->tx_deferred_start;
1356 /* make sure the requested threshold satisfies the constraints */
1357 if (check_thresh(FM10K_TX_RS_THRESH_MIN(q),
1358 FM10K_TX_RS_THRESH_MAX(q),
1359 FM10K_TX_RS_THRESH_DIV(q),
1361 PMD_INIT_LOG(ERR, "tx_rs_thresh (%u) must be "
1362 "less than or equal to %u, "
1363 "greater than or equal to %u, "
1364 "and a divisor of %u",
1365 tx_rs_thresh, FM10K_TX_RS_THRESH_MAX(q),
1366 FM10K_TX_RS_THRESH_MIN(q),
1367 FM10K_TX_RS_THRESH_DIV(q));
1371 q->rs_thresh = tx_rs_thresh;
1377 fm10k_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
1378 uint16_t nb_desc, unsigned int socket_id,
1379 const struct rte_eth_txconf *conf)
1381 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1382 struct fm10k_tx_queue *q;
1383 const struct rte_memzone *mz;
1385 PMD_INIT_FUNC_TRACE();
1387 /* make sure a valid number of descriptors have been requested */
1388 if (check_nb_desc(FM10K_MIN_TX_DESC, FM10K_MAX_TX_DESC,
1389 FM10K_MULT_TX_DESC, nb_desc)) {
1390 PMD_INIT_LOG(ERR, "Number of Tx descriptors (%u) must be "
1391 "less than or equal to %"PRIu32", "
1392 "greater than or equal to %u, "
1393 "and a multiple of %u",
1394 nb_desc, (uint32_t)FM10K_MAX_TX_DESC, FM10K_MIN_TX_DESC,
1395 FM10K_MULT_TX_DESC);
1400 * if this queue existed already, free the associated memory. The
1401 * queue cannot be reused in case we need to allocate memory on
1402 * different socket than was previously used.
1404 if (dev->data->tx_queues[queue_id] != NULL) {
1405 tx_queue_free(dev->data->tx_queues[queue_id]);
1406 dev->data->tx_queues[queue_id] = NULL;
1409 /* allocate memory for the queue structure */
1410 q = rte_zmalloc_socket("fm10k", sizeof(*q), RTE_CACHE_LINE_SIZE,
1413 PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
1418 q->nb_desc = nb_desc;
1419 q->port_id = dev->data->port_id;
1420 q->queue_id = queue_id;
1421 q->tail_ptr = (volatile uint32_t *)
1422 &((uint32_t *)hw->hw_addr)[FM10K_TDT(queue_id)];
1423 if (handle_txconf(q, conf))
1426 /* allocate memory for the software ring */
1427 q->sw_ring = rte_zmalloc_socket("fm10k sw ring",
1428 nb_desc * sizeof(struct rte_mbuf *),
1429 RTE_CACHE_LINE_SIZE, socket_id);
1430 if (q->sw_ring == NULL) {
1431 PMD_INIT_LOG(ERR, "Cannot allocate software ring");
1437 * allocate memory for the hardware descriptor ring. A memzone large
1438 * enough to hold the maximum ring size is requested to allow for
1439 * resizing in later calls to the queue setup function.
1441 mz = allocate_hw_ring(dev->driver->pci_drv.name, "tx_ring",
1442 dev->data->port_id, queue_id, socket_id,
1443 FM10K_MAX_TX_RING_SZ, FM10K_ALIGN_TX_DESC);
1445 PMD_INIT_LOG(ERR, "Cannot allocate hardware ring");
1446 rte_free(q->sw_ring);
1450 q->hw_ring = mz->addr;
1451 q->hw_ring_phys_addr = mz->phys_addr;
1454 * allocate memory for the RS bit tracker. Enough slots to hold the
1455 * descriptor index for each RS bit needing to be set are required.
1457 q->rs_tracker.list = rte_zmalloc_socket("fm10k rs tracker",
1458 ((nb_desc + 1) / q->rs_thresh) *
1460 RTE_CACHE_LINE_SIZE, socket_id);
1461 if (q->rs_tracker.list == NULL) {
1462 PMD_INIT_LOG(ERR, "Cannot allocate RS bit tracker");
1463 rte_free(q->sw_ring);
1468 dev->data->tx_queues[queue_id] = q;
1473 fm10k_tx_queue_release(void *queue)
1475 PMD_INIT_FUNC_TRACE();
1477 tx_queue_free(queue);
1481 fm10k_reta_update(struct rte_eth_dev *dev,
1482 struct rte_eth_rss_reta_entry64 *reta_conf,
1485 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1486 uint16_t i, j, idx, shift;
1490 PMD_INIT_FUNC_TRACE();
1492 if (reta_size > FM10K_MAX_RSS_INDICES) {
1493 PMD_INIT_LOG(ERR, "The size of hash lookup table configured "
1494 "(%d) doesn't match the number hardware can supported "
1495 "(%d)", reta_size, FM10K_MAX_RSS_INDICES);
1500 * Update Redirection Table RETA[n], n=0..31. The redirection table has
1501 * 128-entries in 32 registers
1503 for (i = 0; i < FM10K_MAX_RSS_INDICES; i += CHARS_PER_UINT32) {
1504 idx = i / RTE_RETA_GROUP_SIZE;
1505 shift = i % RTE_RETA_GROUP_SIZE;
1506 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
1507 BIT_MASK_PER_UINT32);
1512 if (mask != BIT_MASK_PER_UINT32)
1513 reta = FM10K_READ_REG(hw, FM10K_RETA(0, i >> 2));
1515 for (j = 0; j < CHARS_PER_UINT32; j++) {
1516 if (mask & (0x1 << j)) {
1518 reta &= ~(UINT8_MAX << CHAR_BIT * j);
1519 reta |= reta_conf[idx].reta[shift + j] <<
1523 FM10K_WRITE_REG(hw, FM10K_RETA(0, i >> 2), reta);
1530 fm10k_reta_query(struct rte_eth_dev *dev,
1531 struct rte_eth_rss_reta_entry64 *reta_conf,
1534 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1535 uint16_t i, j, idx, shift;
1539 PMD_INIT_FUNC_TRACE();
1541 if (reta_size < FM10K_MAX_RSS_INDICES) {
1542 PMD_INIT_LOG(ERR, "The size of hash lookup table configured "
1543 "(%d) doesn't match the number hardware can supported "
1544 "(%d)", reta_size, FM10K_MAX_RSS_INDICES);
1549 * Read Redirection Table RETA[n], n=0..31. The redirection table has
1550 * 128-entries in 32 registers
1552 for (i = 0; i < FM10K_MAX_RSS_INDICES; i += CHARS_PER_UINT32) {
1553 idx = i / RTE_RETA_GROUP_SIZE;
1554 shift = i % RTE_RETA_GROUP_SIZE;
1555 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
1556 BIT_MASK_PER_UINT32);
1560 reta = FM10K_READ_REG(hw, FM10K_RETA(0, i >> 2));
1561 for (j = 0; j < CHARS_PER_UINT32; j++) {
1562 if (mask & (0x1 << j))
1563 reta_conf[idx].reta[shift + j] = ((reta >>
1564 CHAR_BIT * j) & UINT8_MAX);
1572 fm10k_rss_hash_update(struct rte_eth_dev *dev,
1573 struct rte_eth_rss_conf *rss_conf)
1575 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1576 uint32_t *key = (uint32_t *)rss_conf->rss_key;
1578 uint64_t hf = rss_conf->rss_hf;
1581 PMD_INIT_FUNC_TRACE();
1583 if (rss_conf->rss_key_len < FM10K_RSSRK_SIZE *
1584 FM10K_RSSRK_ENTRIES_PER_REG)
1591 mrqc |= (hf & ETH_RSS_IPV4) ? FM10K_MRQC_IPV4 : 0;
1592 mrqc |= (hf & ETH_RSS_IPV6) ? FM10K_MRQC_IPV6 : 0;
1593 mrqc |= (hf & ETH_RSS_IPV6_EX) ? FM10K_MRQC_IPV6 : 0;
1594 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_TCP) ? FM10K_MRQC_TCP_IPV4 : 0;
1595 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_TCP) ? FM10K_MRQC_TCP_IPV6 : 0;
1596 mrqc |= (hf & ETH_RSS_IPV6_TCP_EX) ? FM10K_MRQC_TCP_IPV6 : 0;
1597 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_UDP) ? FM10K_MRQC_UDP_IPV4 : 0;
1598 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_UDP) ? FM10K_MRQC_UDP_IPV6 : 0;
1599 mrqc |= (hf & ETH_RSS_IPV6_UDP_EX) ? FM10K_MRQC_UDP_IPV6 : 0;
1601 /* If the mapping doesn't fit any supported, return */
1606 for (i = 0; i < FM10K_RSSRK_SIZE; ++i)
1607 FM10K_WRITE_REG(hw, FM10K_RSSRK(0, i), key[i]);
1609 FM10K_WRITE_REG(hw, FM10K_MRQC(0), mrqc);
1615 fm10k_rss_hash_conf_get(struct rte_eth_dev *dev,
1616 struct rte_eth_rss_conf *rss_conf)
1618 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1619 uint32_t *key = (uint32_t *)rss_conf->rss_key;
1624 PMD_INIT_FUNC_TRACE();
1626 if (rss_conf->rss_key_len < FM10K_RSSRK_SIZE *
1627 FM10K_RSSRK_ENTRIES_PER_REG)
1631 for (i = 0; i < FM10K_RSSRK_SIZE; ++i)
1632 key[i] = FM10K_READ_REG(hw, FM10K_RSSRK(0, i));
1634 mrqc = FM10K_READ_REG(hw, FM10K_MRQC(0));
1636 hf |= (mrqc & FM10K_MRQC_IPV4) ? ETH_RSS_IPV4 : 0;
1637 hf |= (mrqc & FM10K_MRQC_IPV6) ? ETH_RSS_IPV6 : 0;
1638 hf |= (mrqc & FM10K_MRQC_IPV6) ? ETH_RSS_IPV6_EX : 0;
1639 hf |= (mrqc & FM10K_MRQC_TCP_IPV4) ? ETH_RSS_NONFRAG_IPV4_TCP : 0;
1640 hf |= (mrqc & FM10K_MRQC_TCP_IPV6) ? ETH_RSS_NONFRAG_IPV6_TCP : 0;
1641 hf |= (mrqc & FM10K_MRQC_TCP_IPV6) ? ETH_RSS_IPV6_TCP_EX : 0;
1642 hf |= (mrqc & FM10K_MRQC_UDP_IPV4) ? ETH_RSS_NONFRAG_IPV4_UDP : 0;
1643 hf |= (mrqc & FM10K_MRQC_UDP_IPV6) ? ETH_RSS_NONFRAG_IPV6_UDP : 0;
1644 hf |= (mrqc & FM10K_MRQC_UDP_IPV6) ? ETH_RSS_IPV6_UDP_EX : 0;
1646 rss_conf->rss_hf = hf;
1652 fm10k_dev_enable_intr_pf(struct rte_eth_dev *dev)
1654 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1655 uint32_t int_map = FM10K_INT_MAP_IMMEDIATE;
1657 /* Bind all local non-queue interrupt to vector 0 */
1660 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_Mailbox), int_map);
1661 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_PCIeFault), int_map);
1662 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_SwitchUpDown), int_map);
1663 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_SwitchEvent), int_map);
1664 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_SRAM), int_map);
1665 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_VFLR), int_map);
1667 /* Enable misc causes */
1668 FM10K_WRITE_REG(hw, FM10K_EIMR, FM10K_EIMR_ENABLE(PCA_FAULT) |
1669 FM10K_EIMR_ENABLE(THI_FAULT) |
1670 FM10K_EIMR_ENABLE(FUM_FAULT) |
1671 FM10K_EIMR_ENABLE(MAILBOX) |
1672 FM10K_EIMR_ENABLE(SWITCHREADY) |
1673 FM10K_EIMR_ENABLE(SWITCHNOTREADY) |
1674 FM10K_EIMR_ENABLE(SRAMERROR) |
1675 FM10K_EIMR_ENABLE(VFLR));
1678 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_AUTOMASK |
1679 FM10K_ITR_MASK_CLEAR);
1680 FM10K_WRITE_FLUSH(hw);
1684 fm10k_dev_enable_intr_vf(struct rte_eth_dev *dev)
1686 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1687 uint32_t int_map = FM10K_INT_MAP_IMMEDIATE;
1689 /* Bind all local non-queue interrupt to vector 0 */
1692 /* Only INT 0 available, other 15 are reserved. */
1693 FM10K_WRITE_REG(hw, FM10K_VFINT_MAP, int_map);
1696 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_AUTOMASK |
1697 FM10K_ITR_MASK_CLEAR);
1698 FM10K_WRITE_FLUSH(hw);
1702 fm10k_dev_handle_fault(struct fm10k_hw *hw, uint32_t eicr)
1704 struct fm10k_fault fault;
1706 const char *estr = "Unknown error";
1708 /* Process PCA fault */
1709 if (eicr & FM10K_EIMR_PCA_FAULT) {
1710 err = fm10k_get_fault(hw, FM10K_PCA_FAULT, &fault);
1713 switch (fault.type) {
1715 estr = "PCA_NO_FAULT"; break;
1716 case PCA_UNMAPPED_ADDR:
1717 estr = "PCA_UNMAPPED_ADDR"; break;
1718 case PCA_BAD_QACCESS_PF:
1719 estr = "PCA_BAD_QACCESS_PF"; break;
1720 case PCA_BAD_QACCESS_VF:
1721 estr = "PCA_BAD_QACCESS_VF"; break;
1722 case PCA_MALICIOUS_REQ:
1723 estr = "PCA_MALICIOUS_REQ"; break;
1724 case PCA_POISONED_TLP:
1725 estr = "PCA_POISONED_TLP"; break;
1727 estr = "PCA_TLP_ABORT"; break;
1731 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
1732 estr, fault.func ? "VF" : "PF", fault.func,
1733 fault.address, fault.specinfo);
1736 /* Process THI fault */
1737 if (eicr & FM10K_EIMR_THI_FAULT) {
1738 err = fm10k_get_fault(hw, FM10K_THI_FAULT, &fault);
1741 switch (fault.type) {
1743 estr = "THI_NO_FAULT"; break;
1744 case THI_MAL_DIS_Q_FAULT:
1745 estr = "THI_MAL_DIS_Q_FAULT"; break;
1749 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
1750 estr, fault.func ? "VF" : "PF", fault.func,
1751 fault.address, fault.specinfo);
1754 /* Process FUM fault */
1755 if (eicr & FM10K_EIMR_FUM_FAULT) {
1756 err = fm10k_get_fault(hw, FM10K_FUM_FAULT, &fault);
1759 switch (fault.type) {
1761 estr = "FUM_NO_FAULT"; break;
1762 case FUM_UNMAPPED_ADDR:
1763 estr = "FUM_UNMAPPED_ADDR"; break;
1764 case FUM_POISONED_TLP:
1765 estr = "FUM_POISONED_TLP"; break;
1766 case FUM_BAD_VF_QACCESS:
1767 estr = "FUM_BAD_VF_QACCESS"; break;
1768 case FUM_ADD_DECODE_ERR:
1769 estr = "FUM_ADD_DECODE_ERR"; break;
1771 estr = "FUM_RO_ERROR"; break;
1772 case FUM_QPRC_CRC_ERROR:
1773 estr = "FUM_QPRC_CRC_ERROR"; break;
1774 case FUM_CSR_TIMEOUT:
1775 estr = "FUM_CSR_TIMEOUT"; break;
1776 case FUM_INVALID_TYPE:
1777 estr = "FUM_INVALID_TYPE"; break;
1778 case FUM_INVALID_LENGTH:
1779 estr = "FUM_INVALID_LENGTH"; break;
1780 case FUM_INVALID_BE:
1781 estr = "FUM_INVALID_BE"; break;
1782 case FUM_INVALID_ALIGN:
1783 estr = "FUM_INVALID_ALIGN"; break;
1787 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
1788 estr, fault.func ? "VF" : "PF", fault.func,
1789 fault.address, fault.specinfo);
1796 PMD_INIT_LOG(ERR, "Failed to handle fault event.");
1801 * PF interrupt handler triggered by NIC for handling specific interrupt.
1804 * Pointer to interrupt handle.
1806 * The address of parameter (struct rte_eth_dev *) regsitered before.
1812 fm10k_dev_interrupt_handler_pf(
1813 __rte_unused struct rte_intr_handle *handle,
1816 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
1817 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1818 uint32_t cause, status;
1820 if (hw->mac.type != fm10k_mac_pf)
1823 cause = FM10K_READ_REG(hw, FM10K_EICR);
1825 /* Handle PCI fault cases */
1826 if (cause & FM10K_EICR_FAULT_MASK) {
1827 PMD_INIT_LOG(ERR, "INT: find fault!");
1828 fm10k_dev_handle_fault(hw, cause);
1831 /* Handle switch up/down */
1832 if (cause & FM10K_EICR_SWITCHNOTREADY)
1833 PMD_INIT_LOG(ERR, "INT: Switch is not ready");
1835 if (cause & FM10K_EICR_SWITCHREADY)
1836 PMD_INIT_LOG(INFO, "INT: Switch is ready");
1838 /* Handle mailbox message */
1840 hw->mbx.ops.process(hw, &hw->mbx);
1841 fm10k_mbx_unlock(hw);
1843 /* Handle SRAM error */
1844 if (cause & FM10K_EICR_SRAMERROR) {
1845 PMD_INIT_LOG(ERR, "INT: SRAM error on PEP");
1847 status = FM10K_READ_REG(hw, FM10K_SRAM_IP);
1848 /* Write to clear pending bits */
1849 FM10K_WRITE_REG(hw, FM10K_SRAM_IP, status);
1851 /* Todo: print out error message after shared code updates */
1854 /* Clear these 3 events if having any */
1855 cause &= FM10K_EICR_SWITCHNOTREADY | FM10K_EICR_MAILBOX |
1856 FM10K_EICR_SWITCHREADY;
1858 FM10K_WRITE_REG(hw, FM10K_EICR, cause);
1860 /* Re-enable interrupt from device side */
1861 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_AUTOMASK |
1862 FM10K_ITR_MASK_CLEAR);
1863 /* Re-enable interrupt from host side */
1864 rte_intr_enable(&(dev->pci_dev->intr_handle));
1868 * VF interrupt handler triggered by NIC for handling specific interrupt.
1871 * Pointer to interrupt handle.
1873 * The address of parameter (struct rte_eth_dev *) regsitered before.
1879 fm10k_dev_interrupt_handler_vf(
1880 __rte_unused struct rte_intr_handle *handle,
1883 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
1884 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1886 if (hw->mac.type != fm10k_mac_vf)
1889 /* Handle mailbox message if lock is acquired */
1891 hw->mbx.ops.process(hw, &hw->mbx);
1892 fm10k_mbx_unlock(hw);
1894 /* Re-enable interrupt from device side */
1895 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_AUTOMASK |
1896 FM10K_ITR_MASK_CLEAR);
1897 /* Re-enable interrupt from host side */
1898 rte_intr_enable(&(dev->pci_dev->intr_handle));
1901 /* Mailbox message handler in VF */
1902 static const struct fm10k_msg_data fm10k_msgdata_vf[] = {
1903 FM10K_TLV_MSG_TEST_HANDLER(fm10k_tlv_msg_test),
1904 FM10K_VF_MSG_MAC_VLAN_HANDLER(fm10k_msg_mac_vlan_vf),
1905 FM10K_VF_MSG_LPORT_STATE_HANDLER(fm10k_msg_lport_state_vf),
1906 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_tlv_msg_error),
1909 /* Mailbox message handler in PF */
1910 static const struct fm10k_msg_data fm10k_msgdata_pf[] = {
1911 FM10K_PF_MSG_ERR_HANDLER(XCAST_MODES, fm10k_msg_err_pf),
1912 FM10K_PF_MSG_ERR_HANDLER(UPDATE_MAC_FWD_RULE, fm10k_msg_err_pf),
1913 FM10K_PF_MSG_LPORT_MAP_HANDLER(fm10k_msg_lport_map_pf),
1914 FM10K_PF_MSG_ERR_HANDLER(LPORT_CREATE, fm10k_msg_err_pf),
1915 FM10K_PF_MSG_ERR_HANDLER(LPORT_DELETE, fm10k_msg_err_pf),
1916 FM10K_PF_MSG_UPDATE_PVID_HANDLER(fm10k_msg_update_pvid_pf),
1917 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_tlv_msg_error),
1921 fm10k_setup_mbx_service(struct fm10k_hw *hw)
1925 /* Initialize mailbox lock */
1926 fm10k_mbx_initlock(hw);
1928 /* Replace default message handler with new ones */
1929 if (hw->mac.type == fm10k_mac_pf)
1930 err = hw->mbx.ops.register_handlers(&hw->mbx, fm10k_msgdata_pf);
1932 err = hw->mbx.ops.register_handlers(&hw->mbx, fm10k_msgdata_vf);
1935 PMD_INIT_LOG(ERR, "Failed to register mailbox handler.err:%d",
1939 /* Connect to SM for PF device or PF for VF device */
1940 return hw->mbx.ops.connect(hw, &hw->mbx);
1944 fm10k_close_mbx_service(struct fm10k_hw *hw)
1946 /* Disconnect from SM for PF device or PF for VF device */
1947 hw->mbx.ops.disconnect(hw, &hw->mbx);
1950 static const struct eth_dev_ops fm10k_eth_dev_ops = {
1951 .dev_configure = fm10k_dev_configure,
1952 .dev_start = fm10k_dev_start,
1953 .dev_stop = fm10k_dev_stop,
1954 .dev_close = fm10k_dev_close,
1955 .promiscuous_enable = fm10k_dev_promiscuous_enable,
1956 .promiscuous_disable = fm10k_dev_promiscuous_disable,
1957 .allmulticast_enable = fm10k_dev_allmulticast_enable,
1958 .allmulticast_disable = fm10k_dev_allmulticast_disable,
1959 .stats_get = fm10k_stats_get,
1960 .stats_reset = fm10k_stats_reset,
1961 .link_update = fm10k_link_update,
1962 .dev_infos_get = fm10k_dev_infos_get,
1963 .vlan_filter_set = fm10k_vlan_filter_set,
1964 .vlan_offload_set = fm10k_vlan_offload_set,
1965 .mac_addr_add = fm10k_macaddr_add,
1966 .mac_addr_remove = fm10k_macaddr_remove,
1967 .rx_queue_start = fm10k_dev_rx_queue_start,
1968 .rx_queue_stop = fm10k_dev_rx_queue_stop,
1969 .tx_queue_start = fm10k_dev_tx_queue_start,
1970 .tx_queue_stop = fm10k_dev_tx_queue_stop,
1971 .rx_queue_setup = fm10k_rx_queue_setup,
1972 .rx_queue_release = fm10k_rx_queue_release,
1973 .tx_queue_setup = fm10k_tx_queue_setup,
1974 .tx_queue_release = fm10k_tx_queue_release,
1975 .reta_update = fm10k_reta_update,
1976 .reta_query = fm10k_reta_query,
1977 .rss_hash_update = fm10k_rss_hash_update,
1978 .rss_hash_conf_get = fm10k_rss_hash_conf_get,
1982 eth_fm10k_dev_init(struct rte_eth_dev *dev)
1984 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1986 struct fm10k_macvlan_filter_info *macvlan;
1988 PMD_INIT_FUNC_TRACE();
1990 dev->dev_ops = &fm10k_eth_dev_ops;
1991 dev->rx_pkt_burst = &fm10k_recv_pkts;
1992 dev->tx_pkt_burst = &fm10k_xmit_pkts;
1994 if (dev->data->scattered_rx)
1995 dev->rx_pkt_burst = &fm10k_recv_scattered_pkts;
1997 /* only initialize in the primary process */
1998 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2001 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
2002 memset(macvlan, 0, sizeof(*macvlan));
2003 /* Vendor and Device ID need to be set before init of shared code */
2004 memset(hw, 0, sizeof(*hw));
2005 hw->device_id = dev->pci_dev->id.device_id;
2006 hw->vendor_id = dev->pci_dev->id.vendor_id;
2007 hw->subsystem_device_id = dev->pci_dev->id.subsystem_device_id;
2008 hw->subsystem_vendor_id = dev->pci_dev->id.subsystem_vendor_id;
2009 hw->revision_id = 0;
2010 hw->hw_addr = (void *)dev->pci_dev->mem_resource[0].addr;
2011 if (hw->hw_addr == NULL) {
2012 PMD_INIT_LOG(ERR, "Bad mem resource."
2013 " Try to blacklist unused devices.");
2017 /* Store fm10k_adapter pointer */
2018 hw->back = dev->data->dev_private;
2020 /* Initialize the shared code */
2021 diag = fm10k_init_shared_code(hw);
2022 if (diag != FM10K_SUCCESS) {
2023 PMD_INIT_LOG(ERR, "Shared code init failed: %d", diag);
2028 * Inialize bus info. Normally we would call fm10k_get_bus_info(), but
2029 * there is no way to get link status without reading BAR4. Until this
2030 * works, assume we have maximum bandwidth.
2031 * @todo - fix bus info
2033 hw->bus_caps.speed = fm10k_bus_speed_8000;
2034 hw->bus_caps.width = fm10k_bus_width_pcie_x8;
2035 hw->bus_caps.payload = fm10k_bus_payload_512;
2036 hw->bus.speed = fm10k_bus_speed_8000;
2037 hw->bus.width = fm10k_bus_width_pcie_x8;
2038 hw->bus.payload = fm10k_bus_payload_256;
2040 /* Initialize the hw */
2041 diag = fm10k_init_hw(hw);
2042 if (diag != FM10K_SUCCESS) {
2043 PMD_INIT_LOG(ERR, "Hardware init failed: %d", diag);
2047 /* Initialize MAC address(es) */
2048 dev->data->mac_addrs = rte_zmalloc("fm10k",
2049 ETHER_ADDR_LEN * FM10K_MAX_MACADDR_NUM, 0);
2050 if (dev->data->mac_addrs == NULL) {
2051 PMD_INIT_LOG(ERR, "Cannot allocate memory for MAC addresses");
2055 diag = fm10k_read_mac_addr(hw);
2057 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
2058 &dev->data->mac_addrs[0]);
2060 if (diag != FM10K_SUCCESS ||
2061 !is_valid_assigned_ether_addr(dev->data->mac_addrs)) {
2063 /* Generate a random addr */
2064 eth_random_addr(hw->mac.addr);
2065 memcpy(hw->mac.perm_addr, hw->mac.addr, ETH_ALEN);
2066 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
2067 &dev->data->mac_addrs[0]);
2070 /* Reset the hw statistics */
2071 fm10k_stats_reset(dev);
2074 diag = fm10k_reset_hw(hw);
2075 if (diag != FM10K_SUCCESS) {
2076 PMD_INIT_LOG(ERR, "Hardware reset failed: %d", diag);
2080 /* Setup mailbox service */
2081 diag = fm10k_setup_mbx_service(hw);
2082 if (diag != FM10K_SUCCESS) {
2083 PMD_INIT_LOG(ERR, "Failed to setup mailbox: %d", diag);
2087 /*PF/VF has different interrupt handling mechanism */
2088 if (hw->mac.type == fm10k_mac_pf) {
2089 /* register callback func to eal lib */
2090 rte_intr_callback_register(&(dev->pci_dev->intr_handle),
2091 fm10k_dev_interrupt_handler_pf, (void *)dev);
2093 /* enable MISC interrupt */
2094 fm10k_dev_enable_intr_pf(dev);
2096 rte_intr_callback_register(&(dev->pci_dev->intr_handle),
2097 fm10k_dev_interrupt_handler_vf, (void *)dev);
2099 fm10k_dev_enable_intr_vf(dev);
2102 /* Enable uio intr after callback registered */
2103 rte_intr_enable(&(dev->pci_dev->intr_handle));
2105 hw->mac.ops.update_int_moderator(hw);
2107 /* Make sure Switch Manager is ready before going forward. */
2108 if (hw->mac.type == fm10k_mac_pf) {
2109 int switch_ready = 0;
2112 for (i = 0; i < MAX_QUERY_SWITCH_STATE_TIMES; i++) {
2114 hw->mac.ops.get_host_state(hw, &switch_ready);
2115 fm10k_mbx_unlock(hw);
2118 /* Delay some time to acquire async LPORT_MAP info. */
2119 rte_delay_us(WAIT_SWITCH_MSG_US);
2122 if (switch_ready == 0) {
2123 PMD_INIT_LOG(ERR, "switch is not ready");
2129 * Below function will trigger operations on mailbox, acquire lock to
2130 * avoid race condition from interrupt handler. Operations on mailbox
2131 * FIFO will trigger interrupt to PF/SM, in which interrupt handler
2132 * will handle and generate an interrupt to our side. Then, FIFO in
2133 * mailbox will be touched.
2136 /* Enable port first */
2137 hw->mac.ops.update_lport_state(hw, hw->mac.dglort_map, 1, 1);
2139 /* Set unicast mode by default. App can change to other mode in other
2142 hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
2143 FM10K_XCAST_MODE_NONE);
2145 fm10k_mbx_unlock(hw);
2152 * The set of PCI devices this driver supports. This driver will enable both PF
2153 * and SRIOV-VF devices.
2155 static const struct rte_pci_id pci_id_fm10k_map[] = {
2156 #define RTE_PCI_DEV_ID_DECL_FM10K(vend, dev) { RTE_PCI_DEVICE(vend, dev) },
2157 #define RTE_PCI_DEV_ID_DECL_FM10KVF(vend, dev) { RTE_PCI_DEVICE(vend, dev) },
2158 #include "rte_pci_dev_ids.h"
2159 { .vendor_id = 0, /* sentinel */ },
2162 static struct eth_driver rte_pmd_fm10k = {
2164 .name = "rte_pmd_fm10k",
2165 .id_table = pci_id_fm10k_map,
2166 .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
2168 .eth_dev_init = eth_fm10k_dev_init,
2169 .dev_private_size = sizeof(struct fm10k_adapter),
2173 * Driver initialization routine.
2174 * Invoked once at EAL init time.
2175 * Register itself as the [Poll Mode] Driver of PCI FM10K devices.
2178 rte_pmd_fm10k_init(__rte_unused const char *name,
2179 __rte_unused const char *params)
2181 PMD_INIT_FUNC_TRACE();
2182 rte_eth_driver_register(&rte_pmd_fm10k);
2186 static struct rte_driver rte_fm10k_driver = {
2188 .init = rte_pmd_fm10k_init,
2191 PMD_REGISTER_DRIVER(rte_fm10k_driver);