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);
68 static void fm10k_tx_queue_release(void *queue);
69 static void fm10k_rx_queue_release(void *queue);
72 fm10k_mbx_initlock(struct fm10k_hw *hw)
74 rte_spinlock_init(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back));
78 fm10k_mbx_lock(struct fm10k_hw *hw)
80 while (!rte_spinlock_trylock(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back)))
81 rte_delay_us(FM10K_MBXLOCK_DELAY_US);
85 fm10k_mbx_unlock(struct fm10k_hw *hw)
87 rte_spinlock_unlock(FM10K_DEV_PRIVATE_TO_MBXLOCK(hw->back));
91 * reset queue to initial state, allocate software buffers used when starting
94 * return -ENOMEM if buffers cannot be allocated
95 * return -EINVAL if buffers do not satisfy alignment condition
98 rx_queue_reset(struct fm10k_rx_queue *q)
102 PMD_INIT_FUNC_TRACE();
104 diag = rte_mempool_get_bulk(q->mp, (void **)q->sw_ring, q->nb_desc);
108 for (i = 0; i < q->nb_desc; ++i) {
109 fm10k_pktmbuf_reset(q->sw_ring[i], q->port_id);
110 if (!fm10k_addr_alignment_valid(q->sw_ring[i])) {
111 rte_mempool_put_bulk(q->mp, (void **)q->sw_ring,
115 dma_addr = MBUF_DMA_ADDR_DEFAULT(q->sw_ring[i]);
116 q->hw_ring[i].q.pkt_addr = dma_addr;
117 q->hw_ring[i].q.hdr_addr = dma_addr;
122 q->next_trigger = q->alloc_thresh - 1;
123 FM10K_PCI_REG_WRITE(q->tail_ptr, q->nb_desc - 1);
128 * clean queue, descriptor rings, free software buffers used when stopping
132 rx_queue_clean(struct fm10k_rx_queue *q)
134 union fm10k_rx_desc zero = {.q = {0, 0, 0, 0} };
136 PMD_INIT_FUNC_TRACE();
138 /* zero descriptor rings */
139 for (i = 0; i < q->nb_desc; ++i)
140 q->hw_ring[i] = zero;
142 /* free software buffers */
143 for (i = 0; i < q->nb_desc; ++i) {
145 rte_pktmbuf_free_seg(q->sw_ring[i]);
146 q->sw_ring[i] = NULL;
152 * free all queue memory used when releasing the queue (i.e. configure)
155 rx_queue_free(struct fm10k_rx_queue *q)
157 PMD_INIT_FUNC_TRACE();
159 PMD_INIT_LOG(DEBUG, "Freeing rx queue %p", q);
162 rte_free(q->sw_ring);
171 * disable RX queue, wait unitl HW finished necessary flush operation
174 rx_queue_disable(struct fm10k_hw *hw, uint16_t qnum)
178 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(qnum));
179 FM10K_WRITE_REG(hw, FM10K_RXQCTL(qnum),
180 reg & ~FM10K_RXQCTL_ENABLE);
182 /* Wait 100us at most */
183 for (i = 0; i < FM10K_QUEUE_DISABLE_TIMEOUT; i++) {
185 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(qnum));
186 if (!(reg & FM10K_RXQCTL_ENABLE))
190 if (i == FM10K_QUEUE_DISABLE_TIMEOUT)
197 * reset queue to initial state, allocate software buffers used when starting
201 tx_queue_reset(struct fm10k_tx_queue *q)
203 PMD_INIT_FUNC_TRACE();
207 q->nb_free = q->nb_desc - 1;
208 fifo_reset(&q->rs_tracker, (q->nb_desc + 1) / q->rs_thresh);
209 FM10K_PCI_REG_WRITE(q->tail_ptr, 0);
213 * clean queue, descriptor rings, free software buffers used when stopping
217 tx_queue_clean(struct fm10k_tx_queue *q)
219 struct fm10k_tx_desc zero = {0, 0, 0, 0, 0, 0};
221 PMD_INIT_FUNC_TRACE();
223 /* zero descriptor rings */
224 for (i = 0; i < q->nb_desc; ++i)
225 q->hw_ring[i] = zero;
227 /* free software buffers */
228 for (i = 0; i < q->nb_desc; ++i) {
230 rte_pktmbuf_free_seg(q->sw_ring[i]);
231 q->sw_ring[i] = NULL;
237 * free all queue memory used when releasing the queue (i.e. configure)
240 tx_queue_free(struct fm10k_tx_queue *q)
242 PMD_INIT_FUNC_TRACE();
244 PMD_INIT_LOG(DEBUG, "Freeing tx queue %p", q);
246 if (q->rs_tracker.list) {
247 rte_free(q->rs_tracker.list);
248 q->rs_tracker.list = NULL;
251 rte_free(q->sw_ring);
260 * disable TX queue, wait unitl HW finished necessary flush operation
263 tx_queue_disable(struct fm10k_hw *hw, uint16_t qnum)
267 reg = FM10K_READ_REG(hw, FM10K_TXDCTL(qnum));
268 FM10K_WRITE_REG(hw, FM10K_TXDCTL(qnum),
269 reg & ~FM10K_TXDCTL_ENABLE);
271 /* Wait 100us at most */
272 for (i = 0; i < FM10K_QUEUE_DISABLE_TIMEOUT; i++) {
274 reg = FM10K_READ_REG(hw, FM10K_TXDCTL(qnum));
275 if (!(reg & FM10K_TXDCTL_ENABLE))
279 if (i == FM10K_QUEUE_DISABLE_TIMEOUT)
286 fm10k_dev_configure(struct rte_eth_dev *dev)
288 PMD_INIT_FUNC_TRACE();
290 if (dev->data->dev_conf.rxmode.hw_strip_crc == 0)
291 PMD_INIT_LOG(WARNING, "fm10k always strip CRC");
297 fm10k_dev_mq_rx_configure(struct rte_eth_dev *dev)
299 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
300 struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
301 uint32_t mrqc, *key, i, reta, j;
304 #define RSS_KEY_SIZE 40
305 static uint8_t rss_intel_key[RSS_KEY_SIZE] = {
306 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2,
307 0x41, 0x67, 0x25, 0x3D, 0x43, 0xA3, 0x8F, 0xB0,
308 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
309 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C,
310 0x6A, 0x42, 0xB7, 0x3B, 0xBE, 0xAC, 0x01, 0xFA,
313 if (dev->data->nb_rx_queues == 1 ||
314 dev_conf->rxmode.mq_mode != ETH_MQ_RX_RSS ||
315 dev_conf->rx_adv_conf.rss_conf.rss_hf == 0)
318 /* random key is rss_intel_key (default) or user provided (rss_key) */
319 if (dev_conf->rx_adv_conf.rss_conf.rss_key == NULL)
320 key = (uint32_t *)rss_intel_key;
322 key = (uint32_t *)dev_conf->rx_adv_conf.rss_conf.rss_key;
324 /* Now fill our hash function seeds, 4 bytes at a time */
325 for (i = 0; i < RSS_KEY_SIZE / sizeof(*key); ++i)
326 FM10K_WRITE_REG(hw, FM10K_RSSRK(0, i), key[i]);
329 * Fill in redirection table
330 * The byte-swap is needed because NIC registers are in
331 * little-endian order.
334 for (i = 0, j = 0; i < FM10K_MAX_RSS_INDICES; i++, j++) {
335 if (j == dev->data->nb_rx_queues)
337 reta = (reta << CHAR_BIT) | j;
339 FM10K_WRITE_REG(hw, FM10K_RETA(0, i >> 2),
344 * Generate RSS hash based on packet types, TCP/UDP
345 * port numbers and/or IPv4/v6 src and dst addresses
347 hf = dev_conf->rx_adv_conf.rss_conf.rss_hf;
349 mrqc |= (hf & ETH_RSS_IPV4) ? FM10K_MRQC_IPV4 : 0;
350 mrqc |= (hf & ETH_RSS_IPV6) ? FM10K_MRQC_IPV6 : 0;
351 mrqc |= (hf & ETH_RSS_IPV6_EX) ? FM10K_MRQC_IPV6 : 0;
352 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_TCP) ? FM10K_MRQC_TCP_IPV4 : 0;
353 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_TCP) ? FM10K_MRQC_TCP_IPV6 : 0;
354 mrqc |= (hf & ETH_RSS_IPV6_TCP_EX) ? FM10K_MRQC_TCP_IPV6 : 0;
355 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_UDP) ? FM10K_MRQC_UDP_IPV4 : 0;
356 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_UDP) ? FM10K_MRQC_UDP_IPV6 : 0;
357 mrqc |= (hf & ETH_RSS_IPV6_UDP_EX) ? FM10K_MRQC_UDP_IPV6 : 0;
360 PMD_INIT_LOG(ERR, "Specified RSS mode 0x%"PRIx64"is not"
365 FM10K_WRITE_REG(hw, FM10K_MRQC(0), mrqc);
369 fm10k_dev_tx_init(struct rte_eth_dev *dev)
371 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
373 struct fm10k_tx_queue *txq;
377 /* Disable TXINT to avoid possible interrupt */
378 for (i = 0; i < hw->mac.max_queues; i++)
379 FM10K_WRITE_REG(hw, FM10K_TXINT(i),
380 3 << FM10K_TXINT_TIMER_SHIFT);
383 for (i = 0; i < dev->data->nb_tx_queues; ++i) {
384 txq = dev->data->tx_queues[i];
385 base_addr = txq->hw_ring_phys_addr;
386 size = txq->nb_desc * sizeof(struct fm10k_tx_desc);
388 /* disable queue to avoid issues while updating state */
389 ret = tx_queue_disable(hw, i);
391 PMD_INIT_LOG(ERR, "failed to disable queue %d", i);
395 /* set location and size for descriptor ring */
396 FM10K_WRITE_REG(hw, FM10K_TDBAL(i),
397 base_addr & UINT64_LOWER_32BITS_MASK);
398 FM10K_WRITE_REG(hw, FM10K_TDBAH(i),
399 base_addr >> (CHAR_BIT * sizeof(uint32_t)));
400 FM10K_WRITE_REG(hw, FM10K_TDLEN(i), size);
406 fm10k_dev_rx_init(struct rte_eth_dev *dev)
408 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
410 struct fm10k_rx_queue *rxq;
413 uint32_t rxdctl = FM10K_RXDCTL_WRITE_BACK_MIN_DELAY;
416 /* Disable RXINT to avoid possible interrupt */
417 for (i = 0; i < hw->mac.max_queues; i++)
418 FM10K_WRITE_REG(hw, FM10K_RXINT(i),
419 3 << FM10K_RXINT_TIMER_SHIFT);
421 /* Setup RX queues */
422 for (i = 0; i < dev->data->nb_rx_queues; ++i) {
423 rxq = dev->data->rx_queues[i];
424 base_addr = rxq->hw_ring_phys_addr;
425 size = rxq->nb_desc * sizeof(union fm10k_rx_desc);
427 /* disable queue to avoid issues while updating state */
428 ret = rx_queue_disable(hw, i);
430 PMD_INIT_LOG(ERR, "failed to disable queue %d", i);
434 /* Setup the Base and Length of the Rx Descriptor Ring */
435 FM10K_WRITE_REG(hw, FM10K_RDBAL(i),
436 base_addr & UINT64_LOWER_32BITS_MASK);
437 FM10K_WRITE_REG(hw, FM10K_RDBAH(i),
438 base_addr >> (CHAR_BIT * sizeof(uint32_t)));
439 FM10K_WRITE_REG(hw, FM10K_RDLEN(i), size);
441 /* Configure the Rx buffer size for one buff without split */
442 buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
443 RTE_PKTMBUF_HEADROOM);
444 /* As RX buffer is aligned to 512B within mbuf, some bytes are
445 * reserved for this purpose, and the worst case could be 511B.
446 * But SRR reg assumes all buffers have the same size. In order
447 * to fill the gap, we'll have to consider the worst case and
448 * assume 512B is reserved. If we don't do so, it's possible
449 * for HW to overwrite data to next mbuf.
451 buf_size -= FM10K_RX_DATABUF_ALIGN;
453 FM10K_WRITE_REG(hw, FM10K_SRRCTL(i),
454 buf_size >> FM10K_SRRCTL_BSIZEPKT_SHIFT);
456 /* It adds dual VLAN length for supporting dual VLAN */
457 if ((dev->data->dev_conf.rxmode.max_rx_pkt_len +
458 2 * FM10K_VLAN_TAG_SIZE) > buf_size ||
459 dev->data->dev_conf.rxmode.enable_scatter) {
461 dev->data->scattered_rx = 1;
462 dev->rx_pkt_burst = fm10k_recv_scattered_pkts;
463 reg = FM10K_READ_REG(hw, FM10K_SRRCTL(i));
464 reg |= FM10K_SRRCTL_BUFFER_CHAINING_EN;
465 FM10K_WRITE_REG(hw, FM10K_SRRCTL(i), reg);
468 /* Enable drop on empty, it's RO for VF */
469 if (hw->mac.type == fm10k_mac_pf && rxq->drop_en)
470 rxdctl |= FM10K_RXDCTL_DROP_ON_EMPTY;
472 FM10K_WRITE_REG(hw, FM10K_RXDCTL(i), rxdctl);
473 FM10K_WRITE_FLUSH(hw);
476 /* Configure RSS if applicable */
477 fm10k_dev_mq_rx_configure(dev);
482 fm10k_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
484 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
487 struct fm10k_rx_queue *rxq;
489 PMD_INIT_FUNC_TRACE();
491 if (rx_queue_id < dev->data->nb_rx_queues) {
492 rxq = dev->data->rx_queues[rx_queue_id];
493 err = rx_queue_reset(rxq);
494 if (err == -ENOMEM) {
495 PMD_INIT_LOG(ERR, "Failed to alloc memory : %d", err);
497 } else if (err == -EINVAL) {
498 PMD_INIT_LOG(ERR, "Invalid buffer address alignment :"
503 /* Setup the HW Rx Head and Tail Descriptor Pointers
504 * Note: this must be done AFTER the queue is enabled on real
505 * hardware, but BEFORE the queue is enabled when using the
506 * emulation platform. Do it in both places for now and remove
507 * this comment and the following two register writes when the
508 * emulation platform is no longer being used.
510 FM10K_WRITE_REG(hw, FM10K_RDH(rx_queue_id), 0);
511 FM10K_WRITE_REG(hw, FM10K_RDT(rx_queue_id), rxq->nb_desc - 1);
513 /* Set PF ownership flag for PF devices */
514 reg = FM10K_READ_REG(hw, FM10K_RXQCTL(rx_queue_id));
515 if (hw->mac.type == fm10k_mac_pf)
516 reg |= FM10K_RXQCTL_PF;
517 reg |= FM10K_RXQCTL_ENABLE;
518 /* enable RX queue */
519 FM10K_WRITE_REG(hw, FM10K_RXQCTL(rx_queue_id), reg);
520 FM10K_WRITE_FLUSH(hw);
522 /* Setup the HW Rx Head and Tail Descriptor Pointers
523 * Note: this must be done AFTER the queue is enabled
525 FM10K_WRITE_REG(hw, FM10K_RDH(rx_queue_id), 0);
526 FM10K_WRITE_REG(hw, FM10K_RDT(rx_queue_id), rxq->nb_desc - 1);
533 fm10k_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
535 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
537 PMD_INIT_FUNC_TRACE();
539 if (rx_queue_id < dev->data->nb_rx_queues) {
540 /* Disable RX queue */
541 rx_queue_disable(hw, rx_queue_id);
543 /* Free mbuf and clean HW ring */
544 rx_queue_clean(dev->data->rx_queues[rx_queue_id]);
551 fm10k_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
553 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
554 /** @todo - this should be defined in the shared code */
555 #define FM10K_TXDCTL_WRITE_BACK_MIN_DELAY 0x00010000
556 uint32_t txdctl = FM10K_TXDCTL_WRITE_BACK_MIN_DELAY;
559 PMD_INIT_FUNC_TRACE();
561 if (tx_queue_id < dev->data->nb_tx_queues) {
562 tx_queue_reset(dev->data->tx_queues[tx_queue_id]);
564 /* reset head and tail pointers */
565 FM10K_WRITE_REG(hw, FM10K_TDH(tx_queue_id), 0);
566 FM10K_WRITE_REG(hw, FM10K_TDT(tx_queue_id), 0);
568 /* enable TX queue */
569 FM10K_WRITE_REG(hw, FM10K_TXDCTL(tx_queue_id),
570 FM10K_TXDCTL_ENABLE | txdctl);
571 FM10K_WRITE_FLUSH(hw);
579 fm10k_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
581 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
583 PMD_INIT_FUNC_TRACE();
585 if (tx_queue_id < dev->data->nb_tx_queues) {
586 tx_queue_disable(hw, tx_queue_id);
587 tx_queue_clean(dev->data->tx_queues[tx_queue_id]);
593 static inline int fm10k_glort_valid(struct fm10k_hw *hw)
595 return ((hw->mac.dglort_map & FM10K_DGLORTMAP_NONE)
596 != FM10K_DGLORTMAP_NONE);
600 fm10k_dev_promiscuous_enable(struct rte_eth_dev *dev)
602 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
605 PMD_INIT_FUNC_TRACE();
607 /* Return if it didn't acquire valid glort range */
608 if (!fm10k_glort_valid(hw))
612 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
613 FM10K_XCAST_MODE_PROMISC);
614 fm10k_mbx_unlock(hw);
616 if (status != FM10K_SUCCESS)
617 PMD_INIT_LOG(ERR, "Failed to enable promiscuous mode");
621 fm10k_dev_promiscuous_disable(struct rte_eth_dev *dev)
623 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
627 PMD_INIT_FUNC_TRACE();
629 /* Return if it didn't acquire valid glort range */
630 if (!fm10k_glort_valid(hw))
633 if (dev->data->all_multicast == 1)
634 mode = FM10K_XCAST_MODE_ALLMULTI;
636 mode = FM10K_XCAST_MODE_NONE;
639 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
641 fm10k_mbx_unlock(hw);
643 if (status != FM10K_SUCCESS)
644 PMD_INIT_LOG(ERR, "Failed to disable promiscuous mode");
648 fm10k_dev_allmulticast_enable(struct rte_eth_dev *dev)
650 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
653 PMD_INIT_FUNC_TRACE();
655 /* Return if it didn't acquire valid glort range */
656 if (!fm10k_glort_valid(hw))
659 /* If promiscuous mode is enabled, it doesn't make sense to enable
660 * allmulticast and disable promiscuous since fm10k only can select
663 if (dev->data->promiscuous) {
664 PMD_INIT_LOG(INFO, "Promiscuous mode is enabled, "\
665 "needn't enable allmulticast");
670 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
671 FM10K_XCAST_MODE_ALLMULTI);
672 fm10k_mbx_unlock(hw);
674 if (status != FM10K_SUCCESS)
675 PMD_INIT_LOG(ERR, "Failed to enable allmulticast mode");
679 fm10k_dev_allmulticast_disable(struct rte_eth_dev *dev)
681 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
684 PMD_INIT_FUNC_TRACE();
686 /* Return if it didn't acquire valid glort range */
687 if (!fm10k_glort_valid(hw))
690 if (dev->data->promiscuous) {
691 PMD_INIT_LOG(ERR, "Failed to disable allmulticast mode "\
692 "since promisc mode is enabled");
697 /* Change mode to unicast mode */
698 status = hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
699 FM10K_XCAST_MODE_NONE);
700 fm10k_mbx_unlock(hw);
702 if (status != FM10K_SUCCESS)
703 PMD_INIT_LOG(ERR, "Failed to disable allmulticast mode");
706 /* fls = find last set bit = 32 minus the number of leading zeros */
708 #define fls(x) (((x) == 0) ? 0 : (32 - __builtin_clz((x))))
710 #define BSIZEPKT_ROUNDUP ((1 << FM10K_SRRCTL_BSIZEPKT_SHIFT) - 1)
712 fm10k_dev_start(struct rte_eth_dev *dev)
714 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
717 PMD_INIT_FUNC_TRACE();
719 /* stop, init, then start the hw */
720 diag = fm10k_stop_hw(hw);
721 if (diag != FM10K_SUCCESS) {
722 PMD_INIT_LOG(ERR, "Hardware stop failed: %d", diag);
726 diag = fm10k_init_hw(hw);
727 if (diag != FM10K_SUCCESS) {
728 PMD_INIT_LOG(ERR, "Hardware init failed: %d", diag);
732 diag = fm10k_start_hw(hw);
733 if (diag != FM10K_SUCCESS) {
734 PMD_INIT_LOG(ERR, "Hardware start failed: %d", diag);
738 diag = fm10k_dev_tx_init(dev);
740 PMD_INIT_LOG(ERR, "TX init failed: %d", diag);
744 diag = fm10k_dev_rx_init(dev);
746 PMD_INIT_LOG(ERR, "RX init failed: %d", diag);
750 if (hw->mac.type == fm10k_mac_pf) {
751 /* Establish only VSI 0 as valid */
752 FM10K_WRITE_REG(hw, FM10K_DGLORTMAP(0), FM10K_DGLORTMAP_ANY);
754 /* Configure RSS bits used in RETA table */
755 FM10K_WRITE_REG(hw, FM10K_DGLORTDEC(0),
756 fls(dev->data->nb_rx_queues - 1) <<
757 FM10K_DGLORTDEC_RSSLENGTH_SHIFT);
759 /* Invalidate all other GLORT entries */
760 for (i = 1; i < FM10K_DGLORT_COUNT; i++)
761 FM10K_WRITE_REG(hw, FM10K_DGLORTMAP(i),
762 FM10K_DGLORTMAP_NONE);
765 for (i = 0; i < dev->data->nb_rx_queues; i++) {
766 struct fm10k_rx_queue *rxq;
767 rxq = dev->data->rx_queues[i];
769 if (rxq->rx_deferred_start)
771 diag = fm10k_dev_rx_queue_start(dev, i);
774 for (j = 0; j < i; ++j)
775 rx_queue_clean(dev->data->rx_queues[j]);
780 for (i = 0; i < dev->data->nb_tx_queues; i++) {
781 struct fm10k_tx_queue *txq;
782 txq = dev->data->tx_queues[i];
784 if (txq->tx_deferred_start)
786 diag = fm10k_dev_tx_queue_start(dev, i);
789 for (j = 0; j < i; ++j)
790 tx_queue_clean(dev->data->tx_queues[j]);
791 for (j = 0; j < dev->data->nb_rx_queues; ++j)
792 rx_queue_clean(dev->data->rx_queues[j]);
797 /* Update default vlan */
798 if (hw->mac.default_vid && hw->mac.default_vid <= ETHER_MAX_VLAN_ID)
799 fm10k_vlan_filter_set(dev, hw->mac.default_vid, true);
805 fm10k_dev_stop(struct rte_eth_dev *dev)
809 PMD_INIT_FUNC_TRACE();
811 if (dev->data->tx_queues)
812 for (i = 0; i < dev->data->nb_tx_queues; i++)
813 fm10k_dev_tx_queue_stop(dev, i);
815 if (dev->data->rx_queues)
816 for (i = 0; i < dev->data->nb_rx_queues; i++)
817 fm10k_dev_rx_queue_stop(dev, i);
821 fm10k_dev_queue_release(struct rte_eth_dev *dev)
825 PMD_INIT_FUNC_TRACE();
827 if (dev->data->tx_queues) {
828 for (i = 0; i < dev->data->nb_tx_queues; i++)
829 fm10k_tx_queue_release(dev->data->tx_queues[i]);
832 if (dev->data->rx_queues) {
833 for (i = 0; i < dev->data->nb_rx_queues; i++)
834 fm10k_rx_queue_release(dev->data->rx_queues[i]);
839 fm10k_dev_close(struct rte_eth_dev *dev)
841 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
843 PMD_INIT_FUNC_TRACE();
845 fm10k_MACVLAN_remove_all(dev);
847 /* Stop mailbox service first */
848 fm10k_close_mbx_service(hw);
850 fm10k_dev_queue_release(dev);
855 fm10k_link_update(struct rte_eth_dev *dev,
856 __rte_unused int wait_to_complete)
858 PMD_INIT_FUNC_TRACE();
860 /* The host-interface link is always up. The speed is ~50Gbps per Gen3
861 * x8 PCIe interface. For now, we leave the speed undefined since there
862 * is no 50Gbps Ethernet. */
863 dev->data->dev_link.link_speed = 0;
864 dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
865 dev->data->dev_link.link_status = 1;
871 fm10k_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
873 uint64_t ipackets, opackets, ibytes, obytes;
874 struct fm10k_hw *hw =
875 FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
876 struct fm10k_hw_stats *hw_stats =
877 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
880 PMD_INIT_FUNC_TRACE();
882 fm10k_update_hw_stats(hw, hw_stats);
884 ipackets = opackets = ibytes = obytes = 0;
885 for (i = 0; (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) &&
886 (i < hw->mac.max_queues); ++i) {
887 stats->q_ipackets[i] = hw_stats->q[i].rx_packets.count;
888 stats->q_opackets[i] = hw_stats->q[i].tx_packets.count;
889 stats->q_ibytes[i] = hw_stats->q[i].rx_bytes.count;
890 stats->q_obytes[i] = hw_stats->q[i].tx_bytes.count;
891 ipackets += stats->q_ipackets[i];
892 opackets += stats->q_opackets[i];
893 ibytes += stats->q_ibytes[i];
894 obytes += stats->q_obytes[i];
896 stats->ipackets = ipackets;
897 stats->opackets = opackets;
898 stats->ibytes = ibytes;
899 stats->obytes = obytes;
903 fm10k_stats_reset(struct rte_eth_dev *dev)
905 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
906 struct fm10k_hw_stats *hw_stats =
907 FM10K_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
909 PMD_INIT_FUNC_TRACE();
911 memset(hw_stats, 0, sizeof(*hw_stats));
912 fm10k_rebind_hw_stats(hw, hw_stats);
916 fm10k_dev_infos_get(struct rte_eth_dev *dev,
917 struct rte_eth_dev_info *dev_info)
919 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
921 PMD_INIT_FUNC_TRACE();
923 dev_info->min_rx_bufsize = FM10K_MIN_RX_BUF_SIZE;
924 dev_info->max_rx_pktlen = FM10K_MAX_PKT_SIZE;
925 dev_info->max_rx_queues = hw->mac.max_queues;
926 dev_info->max_tx_queues = hw->mac.max_queues;
927 dev_info->max_mac_addrs = FM10K_MAX_MACADDR_NUM;
928 dev_info->max_hash_mac_addrs = 0;
929 dev_info->max_vfs = dev->pci_dev->max_vfs;
930 dev_info->max_vmdq_pools = ETH_64_POOLS;
931 dev_info->rx_offload_capa =
932 DEV_RX_OFFLOAD_VLAN_STRIP |
933 DEV_RX_OFFLOAD_IPV4_CKSUM |
934 DEV_RX_OFFLOAD_UDP_CKSUM |
935 DEV_RX_OFFLOAD_TCP_CKSUM;
936 dev_info->tx_offload_capa =
937 DEV_TX_OFFLOAD_VLAN_INSERT |
938 DEV_TX_OFFLOAD_IPV4_CKSUM |
939 DEV_TX_OFFLOAD_UDP_CKSUM |
940 DEV_TX_OFFLOAD_TCP_CKSUM |
941 DEV_TX_OFFLOAD_TCP_TSO;
943 dev_info->hash_key_size = FM10K_RSSRK_SIZE * sizeof(uint32_t);
944 dev_info->reta_size = FM10K_MAX_RSS_INDICES;
946 dev_info->default_rxconf = (struct rte_eth_rxconf) {
948 .pthresh = FM10K_DEFAULT_RX_PTHRESH,
949 .hthresh = FM10K_DEFAULT_RX_HTHRESH,
950 .wthresh = FM10K_DEFAULT_RX_WTHRESH,
952 .rx_free_thresh = FM10K_RX_FREE_THRESH_DEFAULT(0),
956 dev_info->default_txconf = (struct rte_eth_txconf) {
958 .pthresh = FM10K_DEFAULT_TX_PTHRESH,
959 .hthresh = FM10K_DEFAULT_TX_HTHRESH,
960 .wthresh = FM10K_DEFAULT_TX_WTHRESH,
962 .tx_free_thresh = FM10K_TX_FREE_THRESH_DEFAULT(0),
963 .tx_rs_thresh = FM10K_TX_RS_THRESH_DEFAULT(0),
964 .txq_flags = ETH_TXQ_FLAGS_NOMULTSEGS |
965 ETH_TXQ_FLAGS_NOOFFLOADS,
971 fm10k_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
974 uint16_t mac_num = 0;
975 uint32_t vid_idx, vid_bit, mac_index;
977 struct fm10k_macvlan_filter_info *macvlan;
978 struct rte_eth_dev_data *data = dev->data;
980 hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
981 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
983 if (vlan_id > ETH_VLAN_ID_MAX) {
984 PMD_INIT_LOG(ERR, "Invalid vlan_id: must be < 4096");
988 vid_idx = FM10K_VFTA_IDX(vlan_id);
989 vid_bit = FM10K_VFTA_BIT(vlan_id);
990 /* this VLAN ID is already in the VLAN filter table, return SUCCESS */
991 if (on && (macvlan->vfta[vid_idx] & vid_bit))
993 /* this VLAN ID is NOT in the VLAN filter table, cannot remove */
994 if (!on && !(macvlan->vfta[vid_idx] & vid_bit)) {
995 PMD_INIT_LOG(ERR, "Invalid vlan_id: not existing "
996 "in the VLAN filter table");
1001 result = fm10k_update_vlan(hw, vlan_id, 0, on);
1002 fm10k_mbx_unlock(hw);
1003 if (result != FM10K_SUCCESS) {
1004 PMD_INIT_LOG(ERR, "VLAN update failed: %d", result);
1008 for (mac_index = 0; (mac_index < FM10K_MAX_MACADDR_NUM) &&
1009 (result == FM10K_SUCCESS); mac_index++) {
1010 if (is_zero_ether_addr(&data->mac_addrs[mac_index]))
1012 if (mac_num > macvlan->mac_num - 1) {
1013 PMD_INIT_LOG(ERR, "MAC address number "
1018 result = fm10k_update_uc_addr(hw, hw->mac.dglort_map,
1019 data->mac_addrs[mac_index].addr_bytes,
1021 fm10k_mbx_unlock(hw);
1024 if (result != FM10K_SUCCESS) {
1025 PMD_INIT_LOG(ERR, "MAC address update failed: %d", result);
1030 macvlan->vlan_num++;
1031 macvlan->vfta[vid_idx] |= vid_bit;
1033 macvlan->vlan_num--;
1034 macvlan->vfta[vid_idx] &= ~vid_bit;
1040 fm10k_vlan_offload_set(__rte_unused struct rte_eth_dev *dev, int mask)
1042 if (mask & ETH_VLAN_STRIP_MASK) {
1043 if (!dev->data->dev_conf.rxmode.hw_vlan_strip)
1044 PMD_INIT_LOG(ERR, "VLAN stripping is "
1045 "always on in fm10k");
1048 if (mask & ETH_VLAN_EXTEND_MASK) {
1049 if (dev->data->dev_conf.rxmode.hw_vlan_extend)
1050 PMD_INIT_LOG(ERR, "VLAN QinQ is not "
1051 "supported in fm10k");
1054 if (mask & ETH_VLAN_FILTER_MASK) {
1055 if (!dev->data->dev_conf.rxmode.hw_vlan_filter)
1056 PMD_INIT_LOG(ERR, "VLAN filter is always on in fm10k");
1060 /* Add/Remove a MAC address, and update filters */
1062 fm10k_MAC_filter_set(struct rte_eth_dev *dev, const u8 *mac, bool add)
1065 struct fm10k_hw *hw;
1066 struct fm10k_macvlan_filter_info *macvlan;
1068 hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1069 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1072 for (j = 0; j < FM10K_VFTA_SIZE; j++) {
1073 if (macvlan->vfta[j]) {
1074 for (k = 0; k < FM10K_UINT32_BIT_SIZE; k++) {
1075 if (macvlan->vfta[j] & (1 << k)) {
1076 if (i + 1 > macvlan->vlan_num) {
1077 PMD_INIT_LOG(ERR, "vlan number "
1082 fm10k_update_uc_addr(hw,
1083 hw->mac.dglort_map, mac,
1084 j * FM10K_UINT32_BIT_SIZE + k,
1086 fm10k_mbx_unlock(hw);
1099 /* Add a MAC address, and update filters */
1101 fm10k_macaddr_add(struct rte_eth_dev *dev,
1102 struct ether_addr *mac_addr,
1103 __rte_unused uint32_t index,
1104 __rte_unused uint32_t pool)
1106 fm10k_MAC_filter_set(dev, mac_addr->addr_bytes, TRUE);
1109 /* Remove a MAC address, and update filters */
1111 fm10k_macaddr_remove(struct rte_eth_dev *dev, uint32_t index)
1113 struct rte_eth_dev_data *data = dev->data;
1115 if (index < FM10K_MAX_MACADDR_NUM)
1116 fm10k_MAC_filter_set(dev, data->mac_addrs[index].addr_bytes,
1120 /* Remove all VLAN and MAC address table entries */
1122 fm10k_MACVLAN_remove_all(struct rte_eth_dev *dev)
1125 struct fm10k_macvlan_filter_info *macvlan;
1127 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
1128 for (j = 0; j < FM10K_VFTA_SIZE; j++) {
1129 if (macvlan->vfta[j]) {
1130 for (k = 0; k < FM10K_UINT32_BIT_SIZE; k++) {
1131 if (macvlan->vfta[j] & (1 << k))
1132 fm10k_vlan_filter_set(dev,
1133 j * FM10K_UINT32_BIT_SIZE + k, false);
1140 check_nb_desc(uint16_t min, uint16_t max, uint16_t mult, uint16_t request)
1142 if ((request < min) || (request > max) || ((request % mult) != 0))
1149 * Create a memzone for hardware descriptor rings. Malloc cannot be used since
1150 * the physical address is required. If the memzone is already created, then
1151 * this function returns a pointer to the existing memzone.
1153 static inline const struct rte_memzone *
1154 allocate_hw_ring(const char *driver_name, const char *ring_name,
1155 uint8_t port_id, uint16_t queue_id, int socket_id,
1156 uint32_t size, uint32_t align)
1158 char name[RTE_MEMZONE_NAMESIZE];
1159 const struct rte_memzone *mz;
1161 snprintf(name, sizeof(name), "%s_%s_%d_%d_%d",
1162 driver_name, ring_name, port_id, queue_id, socket_id);
1164 /* return the memzone if it already exists */
1165 mz = rte_memzone_lookup(name);
1169 #ifdef RTE_LIBRTE_XEN_DOM0
1170 return rte_memzone_reserve_bounded(name, size, socket_id, 0, align,
1173 return rte_memzone_reserve_aligned(name, size, socket_id, 0, align);
1178 check_thresh(uint16_t min, uint16_t max, uint16_t div, uint16_t request)
1180 if ((request < min) || (request > max) || ((div % request) != 0))
1187 handle_rxconf(struct fm10k_rx_queue *q, const struct rte_eth_rxconf *conf)
1189 uint16_t rx_free_thresh;
1191 if (conf->rx_free_thresh == 0)
1192 rx_free_thresh = FM10K_RX_FREE_THRESH_DEFAULT(q);
1194 rx_free_thresh = conf->rx_free_thresh;
1196 /* make sure the requested threshold satisfies the constraints */
1197 if (check_thresh(FM10K_RX_FREE_THRESH_MIN(q),
1198 FM10K_RX_FREE_THRESH_MAX(q),
1199 FM10K_RX_FREE_THRESH_DIV(q),
1201 PMD_INIT_LOG(ERR, "rx_free_thresh (%u) must be "
1202 "less than or equal to %u, "
1203 "greater than or equal to %u, "
1204 "and a divisor of %u",
1205 rx_free_thresh, FM10K_RX_FREE_THRESH_MAX(q),
1206 FM10K_RX_FREE_THRESH_MIN(q),
1207 FM10K_RX_FREE_THRESH_DIV(q));
1211 q->alloc_thresh = rx_free_thresh;
1212 q->drop_en = conf->rx_drop_en;
1213 q->rx_deferred_start = conf->rx_deferred_start;
1219 * Hardware requires specific alignment for Rx packet buffers. At
1220 * least one of the following two conditions must be satisfied.
1221 * 1. Address is 512B aligned
1222 * 2. Address is 8B aligned and buffer does not cross 4K boundary.
1224 * As such, the driver may need to adjust the DMA address within the
1225 * buffer by up to 512B.
1227 * return 1 if the element size is valid, otherwise return 0.
1230 mempool_element_size_valid(struct rte_mempool *mp)
1234 /* elt_size includes mbuf header and headroom */
1235 min_size = mp->elt_size - sizeof(struct rte_mbuf) -
1236 RTE_PKTMBUF_HEADROOM;
1238 /* account for up to 512B of alignment */
1239 min_size -= FM10K_RX_DATABUF_ALIGN;
1241 /* sanity check for overflow */
1242 if (min_size > mp->elt_size)
1250 fm10k_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
1251 uint16_t nb_desc, unsigned int socket_id,
1252 const struct rte_eth_rxconf *conf, struct rte_mempool *mp)
1254 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1255 struct fm10k_rx_queue *q;
1256 const struct rte_memzone *mz;
1258 PMD_INIT_FUNC_TRACE();
1260 /* make sure the mempool element size can account for alignment. */
1261 if (!mempool_element_size_valid(mp)) {
1262 PMD_INIT_LOG(ERR, "Error : Mempool element size is too small");
1266 /* make sure a valid number of descriptors have been requested */
1267 if (check_nb_desc(FM10K_MIN_RX_DESC, FM10K_MAX_RX_DESC,
1268 FM10K_MULT_RX_DESC, nb_desc)) {
1269 PMD_INIT_LOG(ERR, "Number of Rx descriptors (%u) must be "
1270 "less than or equal to %"PRIu32", "
1271 "greater than or equal to %u, "
1272 "and a multiple of %u",
1273 nb_desc, (uint32_t)FM10K_MAX_RX_DESC, FM10K_MIN_RX_DESC,
1274 FM10K_MULT_RX_DESC);
1279 * if this queue existed already, free the associated memory. The
1280 * queue cannot be reused in case we need to allocate memory on
1281 * different socket than was previously used.
1283 if (dev->data->rx_queues[queue_id] != NULL) {
1284 rx_queue_free(dev->data->rx_queues[queue_id]);
1285 dev->data->rx_queues[queue_id] = NULL;
1288 /* allocate memory for the queue structure */
1289 q = rte_zmalloc_socket("fm10k", sizeof(*q), RTE_CACHE_LINE_SIZE,
1292 PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
1298 q->nb_desc = nb_desc;
1299 q->port_id = dev->data->port_id;
1300 q->queue_id = queue_id;
1301 q->tail_ptr = (volatile uint32_t *)
1302 &((uint32_t *)hw->hw_addr)[FM10K_RDT(queue_id)];
1303 if (handle_rxconf(q, conf))
1306 /* allocate memory for the software ring */
1307 q->sw_ring = rte_zmalloc_socket("fm10k sw ring",
1308 nb_desc * sizeof(struct rte_mbuf *),
1309 RTE_CACHE_LINE_SIZE, socket_id);
1310 if (q->sw_ring == NULL) {
1311 PMD_INIT_LOG(ERR, "Cannot allocate software ring");
1317 * allocate memory for the hardware descriptor ring. A memzone large
1318 * enough to hold the maximum ring size is requested to allow for
1319 * resizing in later calls to the queue setup function.
1321 mz = allocate_hw_ring(dev->driver->pci_drv.name, "rx_ring",
1322 dev->data->port_id, queue_id, socket_id,
1323 FM10K_MAX_RX_RING_SZ, FM10K_ALIGN_RX_DESC);
1325 PMD_INIT_LOG(ERR, "Cannot allocate hardware ring");
1326 rte_free(q->sw_ring);
1330 q->hw_ring = mz->addr;
1331 #ifdef RTE_LIBRTE_XEN_DOM0
1332 q->hw_ring_phys_addr = rte_mem_phy2mch(mz->memseg_id, mz->phys_addr);
1334 q->hw_ring_phys_addr = mz->phys_addr;
1337 dev->data->rx_queues[queue_id] = q;
1342 fm10k_rx_queue_release(void *queue)
1344 PMD_INIT_FUNC_TRACE();
1346 rx_queue_free(queue);
1350 handle_txconf(struct fm10k_tx_queue *q, const struct rte_eth_txconf *conf)
1352 uint16_t tx_free_thresh;
1353 uint16_t tx_rs_thresh;
1355 /* constraint MACROs require that tx_free_thresh is configured
1356 * before tx_rs_thresh */
1357 if (conf->tx_free_thresh == 0)
1358 tx_free_thresh = FM10K_TX_FREE_THRESH_DEFAULT(q);
1360 tx_free_thresh = conf->tx_free_thresh;
1362 /* make sure the requested threshold satisfies the constraints */
1363 if (check_thresh(FM10K_TX_FREE_THRESH_MIN(q),
1364 FM10K_TX_FREE_THRESH_MAX(q),
1365 FM10K_TX_FREE_THRESH_DIV(q),
1367 PMD_INIT_LOG(ERR, "tx_free_thresh (%u) must be "
1368 "less than or equal to %u, "
1369 "greater than or equal to %u, "
1370 "and a divisor of %u",
1371 tx_free_thresh, FM10K_TX_FREE_THRESH_MAX(q),
1372 FM10K_TX_FREE_THRESH_MIN(q),
1373 FM10K_TX_FREE_THRESH_DIV(q));
1377 q->free_thresh = tx_free_thresh;
1379 if (conf->tx_rs_thresh == 0)
1380 tx_rs_thresh = FM10K_TX_RS_THRESH_DEFAULT(q);
1382 tx_rs_thresh = conf->tx_rs_thresh;
1384 q->tx_deferred_start = conf->tx_deferred_start;
1386 /* make sure the requested threshold satisfies the constraints */
1387 if (check_thresh(FM10K_TX_RS_THRESH_MIN(q),
1388 FM10K_TX_RS_THRESH_MAX(q),
1389 FM10K_TX_RS_THRESH_DIV(q),
1391 PMD_INIT_LOG(ERR, "tx_rs_thresh (%u) must be "
1392 "less than or equal to %u, "
1393 "greater than or equal to %u, "
1394 "and a divisor of %u",
1395 tx_rs_thresh, FM10K_TX_RS_THRESH_MAX(q),
1396 FM10K_TX_RS_THRESH_MIN(q),
1397 FM10K_TX_RS_THRESH_DIV(q));
1401 q->rs_thresh = tx_rs_thresh;
1407 fm10k_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
1408 uint16_t nb_desc, unsigned int socket_id,
1409 const struct rte_eth_txconf *conf)
1411 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1412 struct fm10k_tx_queue *q;
1413 const struct rte_memzone *mz;
1415 PMD_INIT_FUNC_TRACE();
1417 /* make sure a valid number of descriptors have been requested */
1418 if (check_nb_desc(FM10K_MIN_TX_DESC, FM10K_MAX_TX_DESC,
1419 FM10K_MULT_TX_DESC, nb_desc)) {
1420 PMD_INIT_LOG(ERR, "Number of Tx descriptors (%u) must be "
1421 "less than or equal to %"PRIu32", "
1422 "greater than or equal to %u, "
1423 "and a multiple of %u",
1424 nb_desc, (uint32_t)FM10K_MAX_TX_DESC, FM10K_MIN_TX_DESC,
1425 FM10K_MULT_TX_DESC);
1430 * if this queue existed already, free the associated memory. The
1431 * queue cannot be reused in case we need to allocate memory on
1432 * different socket than was previously used.
1434 if (dev->data->tx_queues[queue_id] != NULL) {
1435 tx_queue_free(dev->data->tx_queues[queue_id]);
1436 dev->data->tx_queues[queue_id] = NULL;
1439 /* allocate memory for the queue structure */
1440 q = rte_zmalloc_socket("fm10k", sizeof(*q), RTE_CACHE_LINE_SIZE,
1443 PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
1448 q->nb_desc = nb_desc;
1449 q->port_id = dev->data->port_id;
1450 q->queue_id = queue_id;
1451 q->tail_ptr = (volatile uint32_t *)
1452 &((uint32_t *)hw->hw_addr)[FM10K_TDT(queue_id)];
1453 if (handle_txconf(q, conf))
1456 /* allocate memory for the software ring */
1457 q->sw_ring = rte_zmalloc_socket("fm10k sw ring",
1458 nb_desc * sizeof(struct rte_mbuf *),
1459 RTE_CACHE_LINE_SIZE, socket_id);
1460 if (q->sw_ring == NULL) {
1461 PMD_INIT_LOG(ERR, "Cannot allocate software ring");
1467 * allocate memory for the hardware descriptor ring. A memzone large
1468 * enough to hold the maximum ring size is requested to allow for
1469 * resizing in later calls to the queue setup function.
1471 mz = allocate_hw_ring(dev->driver->pci_drv.name, "tx_ring",
1472 dev->data->port_id, queue_id, socket_id,
1473 FM10K_MAX_TX_RING_SZ, FM10K_ALIGN_TX_DESC);
1475 PMD_INIT_LOG(ERR, "Cannot allocate hardware ring");
1476 rte_free(q->sw_ring);
1480 q->hw_ring = mz->addr;
1481 #ifdef RTE_LIBRTE_XEN_DOM0
1482 q->hw_ring_phys_addr = rte_mem_phy2mch(mz->memseg_id, mz->phys_addr);
1484 q->hw_ring_phys_addr = mz->phys_addr;
1488 * allocate memory for the RS bit tracker. Enough slots to hold the
1489 * descriptor index for each RS bit needing to be set are required.
1491 q->rs_tracker.list = rte_zmalloc_socket("fm10k rs tracker",
1492 ((nb_desc + 1) / q->rs_thresh) *
1494 RTE_CACHE_LINE_SIZE, socket_id);
1495 if (q->rs_tracker.list == NULL) {
1496 PMD_INIT_LOG(ERR, "Cannot allocate RS bit tracker");
1497 rte_free(q->sw_ring);
1502 dev->data->tx_queues[queue_id] = q;
1507 fm10k_tx_queue_release(void *queue)
1509 PMD_INIT_FUNC_TRACE();
1511 tx_queue_free(queue);
1515 fm10k_reta_update(struct rte_eth_dev *dev,
1516 struct rte_eth_rss_reta_entry64 *reta_conf,
1519 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1520 uint16_t i, j, idx, shift;
1524 PMD_INIT_FUNC_TRACE();
1526 if (reta_size > FM10K_MAX_RSS_INDICES) {
1527 PMD_INIT_LOG(ERR, "The size of hash lookup table configured "
1528 "(%d) doesn't match the number hardware can supported "
1529 "(%d)", reta_size, FM10K_MAX_RSS_INDICES);
1534 * Update Redirection Table RETA[n], n=0..31. The redirection table has
1535 * 128-entries in 32 registers
1537 for (i = 0; i < FM10K_MAX_RSS_INDICES; i += CHARS_PER_UINT32) {
1538 idx = i / RTE_RETA_GROUP_SIZE;
1539 shift = i % RTE_RETA_GROUP_SIZE;
1540 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
1541 BIT_MASK_PER_UINT32);
1546 if (mask != BIT_MASK_PER_UINT32)
1547 reta = FM10K_READ_REG(hw, FM10K_RETA(0, i >> 2));
1549 for (j = 0; j < CHARS_PER_UINT32; j++) {
1550 if (mask & (0x1 << j)) {
1552 reta &= ~(UINT8_MAX << CHAR_BIT * j);
1553 reta |= reta_conf[idx].reta[shift + j] <<
1557 FM10K_WRITE_REG(hw, FM10K_RETA(0, i >> 2), reta);
1564 fm10k_reta_query(struct rte_eth_dev *dev,
1565 struct rte_eth_rss_reta_entry64 *reta_conf,
1568 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1569 uint16_t i, j, idx, shift;
1573 PMD_INIT_FUNC_TRACE();
1575 if (reta_size < FM10K_MAX_RSS_INDICES) {
1576 PMD_INIT_LOG(ERR, "The size of hash lookup table configured "
1577 "(%d) doesn't match the number hardware can supported "
1578 "(%d)", reta_size, FM10K_MAX_RSS_INDICES);
1583 * Read Redirection Table RETA[n], n=0..31. The redirection table has
1584 * 128-entries in 32 registers
1586 for (i = 0; i < FM10K_MAX_RSS_INDICES; i += CHARS_PER_UINT32) {
1587 idx = i / RTE_RETA_GROUP_SIZE;
1588 shift = i % RTE_RETA_GROUP_SIZE;
1589 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
1590 BIT_MASK_PER_UINT32);
1594 reta = FM10K_READ_REG(hw, FM10K_RETA(0, i >> 2));
1595 for (j = 0; j < CHARS_PER_UINT32; j++) {
1596 if (mask & (0x1 << j))
1597 reta_conf[idx].reta[shift + j] = ((reta >>
1598 CHAR_BIT * j) & UINT8_MAX);
1606 fm10k_rss_hash_update(struct rte_eth_dev *dev,
1607 struct rte_eth_rss_conf *rss_conf)
1609 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1610 uint32_t *key = (uint32_t *)rss_conf->rss_key;
1612 uint64_t hf = rss_conf->rss_hf;
1615 PMD_INIT_FUNC_TRACE();
1617 if (rss_conf->rss_key_len < FM10K_RSSRK_SIZE *
1618 FM10K_RSSRK_ENTRIES_PER_REG)
1625 mrqc |= (hf & ETH_RSS_IPV4) ? FM10K_MRQC_IPV4 : 0;
1626 mrqc |= (hf & ETH_RSS_IPV6) ? FM10K_MRQC_IPV6 : 0;
1627 mrqc |= (hf & ETH_RSS_IPV6_EX) ? FM10K_MRQC_IPV6 : 0;
1628 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_TCP) ? FM10K_MRQC_TCP_IPV4 : 0;
1629 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_TCP) ? FM10K_MRQC_TCP_IPV6 : 0;
1630 mrqc |= (hf & ETH_RSS_IPV6_TCP_EX) ? FM10K_MRQC_TCP_IPV6 : 0;
1631 mrqc |= (hf & ETH_RSS_NONFRAG_IPV4_UDP) ? FM10K_MRQC_UDP_IPV4 : 0;
1632 mrqc |= (hf & ETH_RSS_NONFRAG_IPV6_UDP) ? FM10K_MRQC_UDP_IPV6 : 0;
1633 mrqc |= (hf & ETH_RSS_IPV6_UDP_EX) ? FM10K_MRQC_UDP_IPV6 : 0;
1635 /* If the mapping doesn't fit any supported, return */
1640 for (i = 0; i < FM10K_RSSRK_SIZE; ++i)
1641 FM10K_WRITE_REG(hw, FM10K_RSSRK(0, i), key[i]);
1643 FM10K_WRITE_REG(hw, FM10K_MRQC(0), mrqc);
1649 fm10k_rss_hash_conf_get(struct rte_eth_dev *dev,
1650 struct rte_eth_rss_conf *rss_conf)
1652 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1653 uint32_t *key = (uint32_t *)rss_conf->rss_key;
1658 PMD_INIT_FUNC_TRACE();
1660 if (rss_conf->rss_key_len < FM10K_RSSRK_SIZE *
1661 FM10K_RSSRK_ENTRIES_PER_REG)
1665 for (i = 0; i < FM10K_RSSRK_SIZE; ++i)
1666 key[i] = FM10K_READ_REG(hw, FM10K_RSSRK(0, i));
1668 mrqc = FM10K_READ_REG(hw, FM10K_MRQC(0));
1670 hf |= (mrqc & FM10K_MRQC_IPV4) ? ETH_RSS_IPV4 : 0;
1671 hf |= (mrqc & FM10K_MRQC_IPV6) ? ETH_RSS_IPV6 : 0;
1672 hf |= (mrqc & FM10K_MRQC_IPV6) ? ETH_RSS_IPV6_EX : 0;
1673 hf |= (mrqc & FM10K_MRQC_TCP_IPV4) ? ETH_RSS_NONFRAG_IPV4_TCP : 0;
1674 hf |= (mrqc & FM10K_MRQC_TCP_IPV6) ? ETH_RSS_NONFRAG_IPV6_TCP : 0;
1675 hf |= (mrqc & FM10K_MRQC_TCP_IPV6) ? ETH_RSS_IPV6_TCP_EX : 0;
1676 hf |= (mrqc & FM10K_MRQC_UDP_IPV4) ? ETH_RSS_NONFRAG_IPV4_UDP : 0;
1677 hf |= (mrqc & FM10K_MRQC_UDP_IPV6) ? ETH_RSS_NONFRAG_IPV6_UDP : 0;
1678 hf |= (mrqc & FM10K_MRQC_UDP_IPV6) ? ETH_RSS_IPV6_UDP_EX : 0;
1680 rss_conf->rss_hf = hf;
1686 fm10k_dev_enable_intr_pf(struct rte_eth_dev *dev)
1688 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1689 uint32_t int_map = FM10K_INT_MAP_IMMEDIATE;
1691 /* Bind all local non-queue interrupt to vector 0 */
1694 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_Mailbox), int_map);
1695 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_PCIeFault), int_map);
1696 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_SwitchUpDown), int_map);
1697 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_SwitchEvent), int_map);
1698 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_SRAM), int_map);
1699 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_VFLR), int_map);
1701 /* Enable misc causes */
1702 FM10K_WRITE_REG(hw, FM10K_EIMR, FM10K_EIMR_ENABLE(PCA_FAULT) |
1703 FM10K_EIMR_ENABLE(THI_FAULT) |
1704 FM10K_EIMR_ENABLE(FUM_FAULT) |
1705 FM10K_EIMR_ENABLE(MAILBOX) |
1706 FM10K_EIMR_ENABLE(SWITCHREADY) |
1707 FM10K_EIMR_ENABLE(SWITCHNOTREADY) |
1708 FM10K_EIMR_ENABLE(SRAMERROR) |
1709 FM10K_EIMR_ENABLE(VFLR));
1712 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_AUTOMASK |
1713 FM10K_ITR_MASK_CLEAR);
1714 FM10K_WRITE_FLUSH(hw);
1718 fm10k_dev_disable_intr_pf(struct rte_eth_dev *dev)
1720 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1721 uint32_t int_map = FM10K_INT_MAP_DISABLE;
1725 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_Mailbox), int_map);
1726 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_PCIeFault), int_map);
1727 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_SwitchUpDown), int_map);
1728 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_SwitchEvent), int_map);
1729 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_SRAM), int_map);
1730 FM10K_WRITE_REG(hw, FM10K_INT_MAP(fm10k_int_VFLR), int_map);
1732 /* Disable misc causes */
1733 FM10K_WRITE_REG(hw, FM10K_EIMR, FM10K_EIMR_DISABLE(PCA_FAULT) |
1734 FM10K_EIMR_DISABLE(THI_FAULT) |
1735 FM10K_EIMR_DISABLE(FUM_FAULT) |
1736 FM10K_EIMR_DISABLE(MAILBOX) |
1737 FM10K_EIMR_DISABLE(SWITCHREADY) |
1738 FM10K_EIMR_DISABLE(SWITCHNOTREADY) |
1739 FM10K_EIMR_DISABLE(SRAMERROR) |
1740 FM10K_EIMR_DISABLE(VFLR));
1743 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_MASK_SET);
1744 FM10K_WRITE_FLUSH(hw);
1748 fm10k_dev_enable_intr_vf(struct rte_eth_dev *dev)
1750 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1751 uint32_t int_map = FM10K_INT_MAP_IMMEDIATE;
1753 /* Bind all local non-queue interrupt to vector 0 */
1756 /* Only INT 0 available, other 15 are reserved. */
1757 FM10K_WRITE_REG(hw, FM10K_VFINT_MAP, int_map);
1760 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_AUTOMASK |
1761 FM10K_ITR_MASK_CLEAR);
1762 FM10K_WRITE_FLUSH(hw);
1766 fm10k_dev_disable_intr_vf(struct rte_eth_dev *dev)
1768 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1769 uint32_t int_map = FM10K_INT_MAP_DISABLE;
1773 /* Only INT 0 available, other 15 are reserved. */
1774 FM10K_WRITE_REG(hw, FM10K_VFINT_MAP, int_map);
1777 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_MASK_SET);
1778 FM10K_WRITE_FLUSH(hw);
1782 fm10k_dev_handle_fault(struct fm10k_hw *hw, uint32_t eicr)
1784 struct fm10k_fault fault;
1786 const char *estr = "Unknown error";
1788 /* Process PCA fault */
1789 if (eicr & FM10K_EICR_PCA_FAULT) {
1790 err = fm10k_get_fault(hw, FM10K_PCA_FAULT, &fault);
1793 switch (fault.type) {
1795 estr = "PCA_NO_FAULT"; break;
1796 case PCA_UNMAPPED_ADDR:
1797 estr = "PCA_UNMAPPED_ADDR"; break;
1798 case PCA_BAD_QACCESS_PF:
1799 estr = "PCA_BAD_QACCESS_PF"; break;
1800 case PCA_BAD_QACCESS_VF:
1801 estr = "PCA_BAD_QACCESS_VF"; break;
1802 case PCA_MALICIOUS_REQ:
1803 estr = "PCA_MALICIOUS_REQ"; break;
1804 case PCA_POISONED_TLP:
1805 estr = "PCA_POISONED_TLP"; break;
1807 estr = "PCA_TLP_ABORT"; break;
1811 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
1812 estr, fault.func ? "VF" : "PF", fault.func,
1813 fault.address, fault.specinfo);
1816 /* Process THI fault */
1817 if (eicr & FM10K_EICR_THI_FAULT) {
1818 err = fm10k_get_fault(hw, FM10K_THI_FAULT, &fault);
1821 switch (fault.type) {
1823 estr = "THI_NO_FAULT"; break;
1824 case THI_MAL_DIS_Q_FAULT:
1825 estr = "THI_MAL_DIS_Q_FAULT"; break;
1829 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
1830 estr, fault.func ? "VF" : "PF", fault.func,
1831 fault.address, fault.specinfo);
1834 /* Process FUM fault */
1835 if (eicr & FM10K_EICR_FUM_FAULT) {
1836 err = fm10k_get_fault(hw, FM10K_FUM_FAULT, &fault);
1839 switch (fault.type) {
1841 estr = "FUM_NO_FAULT"; break;
1842 case FUM_UNMAPPED_ADDR:
1843 estr = "FUM_UNMAPPED_ADDR"; break;
1844 case FUM_POISONED_TLP:
1845 estr = "FUM_POISONED_TLP"; break;
1846 case FUM_BAD_VF_QACCESS:
1847 estr = "FUM_BAD_VF_QACCESS"; break;
1848 case FUM_ADD_DECODE_ERR:
1849 estr = "FUM_ADD_DECODE_ERR"; break;
1851 estr = "FUM_RO_ERROR"; break;
1852 case FUM_QPRC_CRC_ERROR:
1853 estr = "FUM_QPRC_CRC_ERROR"; break;
1854 case FUM_CSR_TIMEOUT:
1855 estr = "FUM_CSR_TIMEOUT"; break;
1856 case FUM_INVALID_TYPE:
1857 estr = "FUM_INVALID_TYPE"; break;
1858 case FUM_INVALID_LENGTH:
1859 estr = "FUM_INVALID_LENGTH"; break;
1860 case FUM_INVALID_BE:
1861 estr = "FUM_INVALID_BE"; break;
1862 case FUM_INVALID_ALIGN:
1863 estr = "FUM_INVALID_ALIGN"; break;
1867 PMD_INIT_LOG(ERR, "%s: %s(%d) Addr:0x%"PRIx64" Spec: 0x%x",
1868 estr, fault.func ? "VF" : "PF", fault.func,
1869 fault.address, fault.specinfo);
1874 PMD_INIT_LOG(ERR, "Failed to handle fault event.");
1879 * PF interrupt handler triggered by NIC for handling specific interrupt.
1882 * Pointer to interrupt handle.
1884 * The address of parameter (struct rte_eth_dev *) regsitered before.
1890 fm10k_dev_interrupt_handler_pf(
1891 __rte_unused struct rte_intr_handle *handle,
1894 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
1895 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1896 uint32_t cause, status;
1898 if (hw->mac.type != fm10k_mac_pf)
1901 cause = FM10K_READ_REG(hw, FM10K_EICR);
1903 /* Handle PCI fault cases */
1904 if (cause & FM10K_EICR_FAULT_MASK) {
1905 PMD_INIT_LOG(ERR, "INT: find fault!");
1906 fm10k_dev_handle_fault(hw, cause);
1909 /* Handle switch up/down */
1910 if (cause & FM10K_EICR_SWITCHNOTREADY)
1911 PMD_INIT_LOG(ERR, "INT: Switch is not ready");
1913 if (cause & FM10K_EICR_SWITCHREADY)
1914 PMD_INIT_LOG(INFO, "INT: Switch is ready");
1916 /* Handle mailbox message */
1918 hw->mbx.ops.process(hw, &hw->mbx);
1919 fm10k_mbx_unlock(hw);
1921 /* Handle SRAM error */
1922 if (cause & FM10K_EICR_SRAMERROR) {
1923 PMD_INIT_LOG(ERR, "INT: SRAM error on PEP");
1925 status = FM10K_READ_REG(hw, FM10K_SRAM_IP);
1926 /* Write to clear pending bits */
1927 FM10K_WRITE_REG(hw, FM10K_SRAM_IP, status);
1929 /* Todo: print out error message after shared code updates */
1932 /* Clear these 3 events if having any */
1933 cause &= FM10K_EICR_SWITCHNOTREADY | FM10K_EICR_MAILBOX |
1934 FM10K_EICR_SWITCHREADY;
1936 FM10K_WRITE_REG(hw, FM10K_EICR, cause);
1938 /* Re-enable interrupt from device side */
1939 FM10K_WRITE_REG(hw, FM10K_ITR(0), FM10K_ITR_AUTOMASK |
1940 FM10K_ITR_MASK_CLEAR);
1941 /* Re-enable interrupt from host side */
1942 rte_intr_enable(&(dev->pci_dev->intr_handle));
1946 * VF interrupt handler triggered by NIC for handling specific interrupt.
1949 * Pointer to interrupt handle.
1951 * The address of parameter (struct rte_eth_dev *) regsitered before.
1957 fm10k_dev_interrupt_handler_vf(
1958 __rte_unused struct rte_intr_handle *handle,
1961 struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
1962 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
1964 if (hw->mac.type != fm10k_mac_vf)
1967 /* Handle mailbox message if lock is acquired */
1969 hw->mbx.ops.process(hw, &hw->mbx);
1970 fm10k_mbx_unlock(hw);
1972 /* Re-enable interrupt from device side */
1973 FM10K_WRITE_REG(hw, FM10K_VFITR(0), FM10K_ITR_AUTOMASK |
1974 FM10K_ITR_MASK_CLEAR);
1975 /* Re-enable interrupt from host side */
1976 rte_intr_enable(&(dev->pci_dev->intr_handle));
1979 /* Mailbox message handler in VF */
1980 static const struct fm10k_msg_data fm10k_msgdata_vf[] = {
1981 FM10K_TLV_MSG_TEST_HANDLER(fm10k_tlv_msg_test),
1982 FM10K_VF_MSG_MAC_VLAN_HANDLER(fm10k_msg_mac_vlan_vf),
1983 FM10K_VF_MSG_LPORT_STATE_HANDLER(fm10k_msg_lport_state_vf),
1984 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_tlv_msg_error),
1987 /* Mailbox message handler in PF */
1988 static const struct fm10k_msg_data fm10k_msgdata_pf[] = {
1989 FM10K_PF_MSG_ERR_HANDLER(XCAST_MODES, fm10k_msg_err_pf),
1990 FM10K_PF_MSG_ERR_HANDLER(UPDATE_MAC_FWD_RULE, fm10k_msg_err_pf),
1991 FM10K_PF_MSG_LPORT_MAP_HANDLER(fm10k_msg_lport_map_pf),
1992 FM10K_PF_MSG_ERR_HANDLER(LPORT_CREATE, fm10k_msg_err_pf),
1993 FM10K_PF_MSG_ERR_HANDLER(LPORT_DELETE, fm10k_msg_err_pf),
1994 FM10K_PF_MSG_UPDATE_PVID_HANDLER(fm10k_msg_update_pvid_pf),
1995 FM10K_TLV_MSG_ERROR_HANDLER(fm10k_tlv_msg_error),
1999 fm10k_setup_mbx_service(struct fm10k_hw *hw)
2003 /* Initialize mailbox lock */
2004 fm10k_mbx_initlock(hw);
2006 /* Replace default message handler with new ones */
2007 if (hw->mac.type == fm10k_mac_pf)
2008 err = hw->mbx.ops.register_handlers(&hw->mbx, fm10k_msgdata_pf);
2010 err = hw->mbx.ops.register_handlers(&hw->mbx, fm10k_msgdata_vf);
2013 PMD_INIT_LOG(ERR, "Failed to register mailbox handler.err:%d",
2017 /* Connect to SM for PF device or PF for VF device */
2018 return hw->mbx.ops.connect(hw, &hw->mbx);
2022 fm10k_close_mbx_service(struct fm10k_hw *hw)
2024 /* Disconnect from SM for PF device or PF for VF device */
2025 hw->mbx.ops.disconnect(hw, &hw->mbx);
2028 static const struct eth_dev_ops fm10k_eth_dev_ops = {
2029 .dev_configure = fm10k_dev_configure,
2030 .dev_start = fm10k_dev_start,
2031 .dev_stop = fm10k_dev_stop,
2032 .dev_close = fm10k_dev_close,
2033 .promiscuous_enable = fm10k_dev_promiscuous_enable,
2034 .promiscuous_disable = fm10k_dev_promiscuous_disable,
2035 .allmulticast_enable = fm10k_dev_allmulticast_enable,
2036 .allmulticast_disable = fm10k_dev_allmulticast_disable,
2037 .stats_get = fm10k_stats_get,
2038 .stats_reset = fm10k_stats_reset,
2039 .link_update = fm10k_link_update,
2040 .dev_infos_get = fm10k_dev_infos_get,
2041 .vlan_filter_set = fm10k_vlan_filter_set,
2042 .vlan_offload_set = fm10k_vlan_offload_set,
2043 .mac_addr_add = fm10k_macaddr_add,
2044 .mac_addr_remove = fm10k_macaddr_remove,
2045 .rx_queue_start = fm10k_dev_rx_queue_start,
2046 .rx_queue_stop = fm10k_dev_rx_queue_stop,
2047 .tx_queue_start = fm10k_dev_tx_queue_start,
2048 .tx_queue_stop = fm10k_dev_tx_queue_stop,
2049 .rx_queue_setup = fm10k_rx_queue_setup,
2050 .rx_queue_release = fm10k_rx_queue_release,
2051 .tx_queue_setup = fm10k_tx_queue_setup,
2052 .tx_queue_release = fm10k_tx_queue_release,
2053 .reta_update = fm10k_reta_update,
2054 .reta_query = fm10k_reta_query,
2055 .rss_hash_update = fm10k_rss_hash_update,
2056 .rss_hash_conf_get = fm10k_rss_hash_conf_get,
2060 eth_fm10k_dev_init(struct rte_eth_dev *dev)
2062 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2064 struct fm10k_macvlan_filter_info *macvlan;
2066 PMD_INIT_FUNC_TRACE();
2068 dev->dev_ops = &fm10k_eth_dev_ops;
2069 dev->rx_pkt_burst = &fm10k_recv_pkts;
2070 dev->tx_pkt_burst = &fm10k_xmit_pkts;
2072 if (dev->data->scattered_rx)
2073 dev->rx_pkt_burst = &fm10k_recv_scattered_pkts;
2075 /* only initialize in the primary process */
2076 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2079 macvlan = FM10K_DEV_PRIVATE_TO_MACVLAN(dev->data->dev_private);
2080 memset(macvlan, 0, sizeof(*macvlan));
2081 /* Vendor and Device ID need to be set before init of shared code */
2082 memset(hw, 0, sizeof(*hw));
2083 hw->device_id = dev->pci_dev->id.device_id;
2084 hw->vendor_id = dev->pci_dev->id.vendor_id;
2085 hw->subsystem_device_id = dev->pci_dev->id.subsystem_device_id;
2086 hw->subsystem_vendor_id = dev->pci_dev->id.subsystem_vendor_id;
2087 hw->revision_id = 0;
2088 hw->hw_addr = (void *)dev->pci_dev->mem_resource[0].addr;
2089 if (hw->hw_addr == NULL) {
2090 PMD_INIT_LOG(ERR, "Bad mem resource."
2091 " Try to blacklist unused devices.");
2095 /* Store fm10k_adapter pointer */
2096 hw->back = dev->data->dev_private;
2098 /* Initialize the shared code */
2099 diag = fm10k_init_shared_code(hw);
2100 if (diag != FM10K_SUCCESS) {
2101 PMD_INIT_LOG(ERR, "Shared code init failed: %d", diag);
2106 * Inialize bus info. Normally we would call fm10k_get_bus_info(), but
2107 * there is no way to get link status without reading BAR4. Until this
2108 * works, assume we have maximum bandwidth.
2109 * @todo - fix bus info
2111 hw->bus_caps.speed = fm10k_bus_speed_8000;
2112 hw->bus_caps.width = fm10k_bus_width_pcie_x8;
2113 hw->bus_caps.payload = fm10k_bus_payload_512;
2114 hw->bus.speed = fm10k_bus_speed_8000;
2115 hw->bus.width = fm10k_bus_width_pcie_x8;
2116 hw->bus.payload = fm10k_bus_payload_256;
2118 /* Initialize the hw */
2119 diag = fm10k_init_hw(hw);
2120 if (diag != FM10K_SUCCESS) {
2121 PMD_INIT_LOG(ERR, "Hardware init failed: %d", diag);
2125 /* Initialize MAC address(es) */
2126 dev->data->mac_addrs = rte_zmalloc("fm10k",
2127 ETHER_ADDR_LEN * FM10K_MAX_MACADDR_NUM, 0);
2128 if (dev->data->mac_addrs == NULL) {
2129 PMD_INIT_LOG(ERR, "Cannot allocate memory for MAC addresses");
2133 diag = fm10k_read_mac_addr(hw);
2135 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
2136 &dev->data->mac_addrs[0]);
2138 if (diag != FM10K_SUCCESS ||
2139 !is_valid_assigned_ether_addr(dev->data->mac_addrs)) {
2141 /* Generate a random addr */
2142 eth_random_addr(hw->mac.addr);
2143 memcpy(hw->mac.perm_addr, hw->mac.addr, ETH_ALEN);
2144 ether_addr_copy((const struct ether_addr *)hw->mac.addr,
2145 &dev->data->mac_addrs[0]);
2148 /* Reset the hw statistics */
2149 fm10k_stats_reset(dev);
2152 diag = fm10k_reset_hw(hw);
2153 if (diag != FM10K_SUCCESS) {
2154 PMD_INIT_LOG(ERR, "Hardware reset failed: %d", diag);
2158 /* Setup mailbox service */
2159 diag = fm10k_setup_mbx_service(hw);
2160 if (diag != FM10K_SUCCESS) {
2161 PMD_INIT_LOG(ERR, "Failed to setup mailbox: %d", diag);
2165 /*PF/VF has different interrupt handling mechanism */
2166 if (hw->mac.type == fm10k_mac_pf) {
2167 /* register callback func to eal lib */
2168 rte_intr_callback_register(&(dev->pci_dev->intr_handle),
2169 fm10k_dev_interrupt_handler_pf, (void *)dev);
2171 /* enable MISC interrupt */
2172 fm10k_dev_enable_intr_pf(dev);
2174 rte_intr_callback_register(&(dev->pci_dev->intr_handle),
2175 fm10k_dev_interrupt_handler_vf, (void *)dev);
2177 fm10k_dev_enable_intr_vf(dev);
2180 /* Enable uio intr after callback registered */
2181 rte_intr_enable(&(dev->pci_dev->intr_handle));
2183 hw->mac.ops.update_int_moderator(hw);
2185 /* Make sure Switch Manager is ready before going forward. */
2186 if (hw->mac.type == fm10k_mac_pf) {
2187 int switch_ready = 0;
2190 for (i = 0; i < MAX_QUERY_SWITCH_STATE_TIMES; i++) {
2192 hw->mac.ops.get_host_state(hw, &switch_ready);
2193 fm10k_mbx_unlock(hw);
2196 /* Delay some time to acquire async LPORT_MAP info. */
2197 rte_delay_us(WAIT_SWITCH_MSG_US);
2200 if (switch_ready == 0) {
2201 PMD_INIT_LOG(ERR, "switch is not ready");
2207 * Below function will trigger operations on mailbox, acquire lock to
2208 * avoid race condition from interrupt handler. Operations on mailbox
2209 * FIFO will trigger interrupt to PF/SM, in which interrupt handler
2210 * will handle and generate an interrupt to our side. Then, FIFO in
2211 * mailbox will be touched.
2214 /* Enable port first */
2215 hw->mac.ops.update_lport_state(hw, hw->mac.dglort_map, 1, 1);
2217 /* Set unicast mode by default. App can change to other mode in other
2220 hw->mac.ops.update_xcast_mode(hw, hw->mac.dglort_map,
2221 FM10K_XCAST_MODE_NONE);
2223 fm10k_mbx_unlock(hw);
2225 /* Add default mac address */
2226 fm10k_MAC_filter_set(dev, hw->mac.addr, true);
2232 eth_fm10k_dev_uninit(struct rte_eth_dev *dev)
2234 struct fm10k_hw *hw = FM10K_DEV_PRIVATE_TO_HW(dev->data->dev_private);
2236 PMD_INIT_FUNC_TRACE();
2238 /* only uninitialize in the primary process */
2239 if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2242 /* safe to close dev here */
2243 fm10k_dev_close(dev);
2245 dev->dev_ops = NULL;
2246 dev->rx_pkt_burst = NULL;
2247 dev->tx_pkt_burst = NULL;
2249 /* disable uio/vfio intr */
2250 rte_intr_disable(&(dev->pci_dev->intr_handle));
2252 /*PF/VF has different interrupt handling mechanism */
2253 if (hw->mac.type == fm10k_mac_pf) {
2254 /* disable interrupt */
2255 fm10k_dev_disable_intr_pf(dev);
2257 /* unregister callback func to eal lib */
2258 rte_intr_callback_unregister(&(dev->pci_dev->intr_handle),
2259 fm10k_dev_interrupt_handler_pf, (void *)dev);
2261 /* disable interrupt */
2262 fm10k_dev_disable_intr_vf(dev);
2264 rte_intr_callback_unregister(&(dev->pci_dev->intr_handle),
2265 fm10k_dev_interrupt_handler_vf, (void *)dev);
2268 /* free mac memory */
2269 if (dev->data->mac_addrs) {
2270 rte_free(dev->data->mac_addrs);
2271 dev->data->mac_addrs = NULL;
2274 memset(hw, 0, sizeof(*hw));
2280 * The set of PCI devices this driver supports. This driver will enable both PF
2281 * and SRIOV-VF devices.
2283 static const struct rte_pci_id pci_id_fm10k_map[] = {
2284 #define RTE_PCI_DEV_ID_DECL_FM10K(vend, dev) { RTE_PCI_DEVICE(vend, dev) },
2285 #define RTE_PCI_DEV_ID_DECL_FM10KVF(vend, dev) { RTE_PCI_DEVICE(vend, dev) },
2286 #include "rte_pci_dev_ids.h"
2287 { .vendor_id = 0, /* sentinel */ },
2290 static struct eth_driver rte_pmd_fm10k = {
2292 .name = "rte_pmd_fm10k",
2293 .id_table = pci_id_fm10k_map,
2294 .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_DETACHABLE,
2296 .eth_dev_init = eth_fm10k_dev_init,
2297 .eth_dev_uninit = eth_fm10k_dev_uninit,
2298 .dev_private_size = sizeof(struct fm10k_adapter),
2302 * Driver initialization routine.
2303 * Invoked once at EAL init time.
2304 * Register itself as the [Poll Mode] Driver of PCI FM10K devices.
2307 rte_pmd_fm10k_init(__rte_unused const char *name,
2308 __rte_unused const char *params)
2310 PMD_INIT_FUNC_TRACE();
2311 rte_eth_driver_register(&rte_pmd_fm10k);
2315 static struct rte_driver rte_fm10k_driver = {
2317 .init = rte_pmd_fm10k_init,
2320 PMD_REGISTER_DRIVER(rte_fm10k_driver);