xref: /dpdk/drivers/net/enic/enic_main.c (revision a3a2e2c8f7de433e10b1548df65b20bf10086d9c)
1 /*
2  * Copyright 2008-2014 Cisco Systems, Inc.  All rights reserved.
3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
4  *
5  * Copyright (c) 2014, Cisco Systems, Inc.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * 1. Redistributions of source code must retain the above copyright
13  * notice, this list of conditions and the following disclaimer.
14  *
15  * 2. Redistributions in binary form must reproduce the above copyright
16  * notice, this list of conditions and the following disclaimer in
17  * the documentation and/or other materials provided with the
18  * distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
28  * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
30  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31  * POSSIBILITY OF SUCH DAMAGE.
32  *
33  */
34 
35 #include <stdio.h>
36 
37 #include <sys/stat.h>
38 #include <sys/mman.h>
39 #include <fcntl.h>
40 #include <libgen.h>
41 
42 #include <rte_pci.h>
43 #include <rte_memzone.h>
44 #include <rte_malloc.h>
45 #include <rte_mbuf.h>
46 #include <rte_string_fns.h>
47 #include <rte_ethdev.h>
48 
49 #include "enic_compat.h"
50 #include "enic.h"
51 #include "wq_enet_desc.h"
52 #include "rq_enet_desc.h"
53 #include "cq_enet_desc.h"
54 #include "vnic_enet.h"
55 #include "vnic_dev.h"
56 #include "vnic_wq.h"
57 #include "vnic_rq.h"
58 #include "vnic_cq.h"
59 #include "vnic_intr.h"
60 #include "vnic_nic.h"
61 
62 static inline int enic_is_sriov_vf(struct enic *enic)
63 {
64 	return enic->pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
65 }
66 
67 static int is_zero_addr(uint8_t *addr)
68 {
69 	return !(addr[0] |  addr[1] | addr[2] | addr[3] | addr[4] | addr[5]);
70 }
71 
72 static int is_mcast_addr(uint8_t *addr)
73 {
74 	return addr[0] & 1;
75 }
76 
77 static int is_eth_addr_valid(uint8_t *addr)
78 {
79 	return !is_mcast_addr(addr) && !is_zero_addr(addr);
80 }
81 
82 static void
83 enic_rxmbuf_queue_release(__rte_unused struct enic *enic, struct vnic_rq *rq)
84 {
85 	uint16_t i;
86 
87 	if (!rq || !rq->mbuf_ring) {
88 		dev_debug(enic, "Pointer to rq or mbuf_ring is NULL");
89 		return;
90 	}
91 
92 	for (i = 0; i < rq->ring.desc_count; i++) {
93 		if (rq->mbuf_ring[i]) {
94 			rte_pktmbuf_free_seg(rq->mbuf_ring[i]);
95 			rq->mbuf_ring[i] = NULL;
96 		}
97 	}
98 }
99 
100 void enic_set_hdr_split_size(struct enic *enic, u16 split_hdr_size)
101 {
102 	vnic_set_hdr_split_size(enic->vdev, split_hdr_size);
103 }
104 
105 static void enic_free_wq_buf(struct vnic_wq_buf *buf)
106 {
107 	struct rte_mbuf *mbuf = (struct rte_mbuf *)buf->mb;
108 
109 	rte_pktmbuf_free_seg(mbuf);
110 	buf->mb = NULL;
111 }
112 
113 static void enic_log_q_error(struct enic *enic)
114 {
115 	unsigned int i;
116 	u32 error_status;
117 
118 	for (i = 0; i < enic->wq_count; i++) {
119 		error_status = vnic_wq_error_status(&enic->wq[i]);
120 		if (error_status)
121 			dev_err(enic, "WQ[%d] error_status %d\n", i,
122 				error_status);
123 	}
124 
125 	for (i = 0; i < enic_vnic_rq_count(enic); i++) {
126 		if (!enic->rq[i].in_use)
127 			continue;
128 		error_status = vnic_rq_error_status(&enic->rq[i]);
129 		if (error_status)
130 			dev_err(enic, "RQ[%d] error_status %d\n", i,
131 				error_status);
132 	}
133 }
134 
135 static void enic_clear_soft_stats(struct enic *enic)
136 {
137 	struct enic_soft_stats *soft_stats = &enic->soft_stats;
138 	rte_atomic64_clear(&soft_stats->rx_nombuf);
139 	rte_atomic64_clear(&soft_stats->rx_packet_errors);
140 	rte_atomic64_clear(&soft_stats->tx_oversized);
141 }
142 
143 static void enic_init_soft_stats(struct enic *enic)
144 {
145 	struct enic_soft_stats *soft_stats = &enic->soft_stats;
146 	rte_atomic64_init(&soft_stats->rx_nombuf);
147 	rte_atomic64_init(&soft_stats->rx_packet_errors);
148 	rte_atomic64_init(&soft_stats->tx_oversized);
149 	enic_clear_soft_stats(enic);
150 }
151 
152 void enic_dev_stats_clear(struct enic *enic)
153 {
154 	if (vnic_dev_stats_clear(enic->vdev))
155 		dev_err(enic, "Error in clearing stats\n");
156 	enic_clear_soft_stats(enic);
157 }
158 
159 void enic_dev_stats_get(struct enic *enic, struct rte_eth_stats *r_stats)
160 {
161 	struct vnic_stats *stats;
162 	struct enic_soft_stats *soft_stats = &enic->soft_stats;
163 	int64_t rx_truncated;
164 	uint64_t rx_packet_errors;
165 
166 	if (vnic_dev_stats_dump(enic->vdev, &stats)) {
167 		dev_err(enic, "Error in getting stats\n");
168 		return;
169 	}
170 
171 	/* The number of truncated packets can only be calculated by
172 	 * subtracting a hardware counter from error packets received by
173 	 * the driver. Note: this causes transient inaccuracies in the
174 	 * ipackets count. Also, the length of truncated packets are
175 	 * counted in ibytes even though truncated packets are dropped
176 	 * which can make ibytes be slightly higher than it should be.
177 	 */
178 	rx_packet_errors = rte_atomic64_read(&soft_stats->rx_packet_errors);
179 	rx_truncated = rx_packet_errors - stats->rx.rx_errors;
180 
181 	r_stats->ipackets = stats->rx.rx_frames_ok - rx_truncated;
182 	r_stats->opackets = stats->tx.tx_frames_ok;
183 
184 	r_stats->ibytes = stats->rx.rx_bytes_ok;
185 	r_stats->obytes = stats->tx.tx_bytes_ok;
186 
187 	r_stats->ierrors = stats->rx.rx_errors + stats->rx.rx_drop;
188 	r_stats->oerrors = stats->tx.tx_errors
189 			   + rte_atomic64_read(&soft_stats->tx_oversized);
190 
191 	r_stats->imissed = stats->rx.rx_no_bufs + rx_truncated;
192 
193 	r_stats->rx_nombuf = rte_atomic64_read(&soft_stats->rx_nombuf);
194 }
195 
196 void enic_del_mac_address(struct enic *enic, int mac_index)
197 {
198 	struct rte_eth_dev *eth_dev = enic->rte_dev;
199 	uint8_t *mac_addr = eth_dev->data->mac_addrs[mac_index].addr_bytes;
200 
201 	if (vnic_dev_del_addr(enic->vdev, mac_addr))
202 		dev_err(enic, "del mac addr failed\n");
203 }
204 
205 int enic_set_mac_address(struct enic *enic, uint8_t *mac_addr)
206 {
207 	int err;
208 
209 	if (!is_eth_addr_valid(mac_addr)) {
210 		dev_err(enic, "invalid mac address\n");
211 		return -EINVAL;
212 	}
213 
214 	err = vnic_dev_add_addr(enic->vdev, mac_addr);
215 	if (err)
216 		dev_err(enic, "add mac addr failed\n");
217 	return err;
218 }
219 
220 static void
221 enic_free_rq_buf(struct rte_mbuf **mbuf)
222 {
223 	if (*mbuf == NULL)
224 		return;
225 
226 	rte_pktmbuf_free(*mbuf);
227 	mbuf = NULL;
228 }
229 
230 void enic_init_vnic_resources(struct enic *enic)
231 {
232 	unsigned int error_interrupt_enable = 1;
233 	unsigned int error_interrupt_offset = 0;
234 	unsigned int index = 0;
235 	unsigned int cq_idx;
236 	struct vnic_rq *data_rq;
237 
238 	for (index = 0; index < enic->rq_count; index++) {
239 		cq_idx = enic_cq_rq(enic, enic_rte_rq_idx_to_sop_idx(index));
240 
241 		vnic_rq_init(&enic->rq[enic_rte_rq_idx_to_sop_idx(index)],
242 			cq_idx,
243 			error_interrupt_enable,
244 			error_interrupt_offset);
245 
246 		data_rq = &enic->rq[enic_rte_rq_idx_to_data_idx(index)];
247 		if (data_rq->in_use)
248 			vnic_rq_init(data_rq,
249 				     cq_idx,
250 				     error_interrupt_enable,
251 				     error_interrupt_offset);
252 
253 		vnic_cq_init(&enic->cq[cq_idx],
254 			0 /* flow_control_enable */,
255 			1 /* color_enable */,
256 			0 /* cq_head */,
257 			0 /* cq_tail */,
258 			1 /* cq_tail_color */,
259 			0 /* interrupt_enable */,
260 			1 /* cq_entry_enable */,
261 			0 /* cq_message_enable */,
262 			0 /* interrupt offset */,
263 			0 /* cq_message_addr */);
264 	}
265 
266 	for (index = 0; index < enic->wq_count; index++) {
267 		vnic_wq_init(&enic->wq[index],
268 			enic_cq_wq(enic, index),
269 			error_interrupt_enable,
270 			error_interrupt_offset);
271 
272 		cq_idx = enic_cq_wq(enic, index);
273 		vnic_cq_init(&enic->cq[cq_idx],
274 			0 /* flow_control_enable */,
275 			1 /* color_enable */,
276 			0 /* cq_head */,
277 			0 /* cq_tail */,
278 			1 /* cq_tail_color */,
279 			0 /* interrupt_enable */,
280 			0 /* cq_entry_enable */,
281 			1 /* cq_message_enable */,
282 			0 /* interrupt offset */,
283 			(u64)enic->wq[index].cqmsg_rz->phys_addr);
284 	}
285 
286 	vnic_intr_init(&enic->intr,
287 		enic->config.intr_timer_usec,
288 		enic->config.intr_timer_type,
289 		/*mask_on_assertion*/1);
290 }
291 
292 
293 static int
294 enic_alloc_rx_queue_mbufs(struct enic *enic, struct vnic_rq *rq)
295 {
296 	struct rte_mbuf *mb;
297 	struct rq_enet_desc *rqd = rq->ring.descs;
298 	unsigned i;
299 	dma_addr_t dma_addr;
300 
301 	if (!rq->in_use)
302 		return 0;
303 
304 	dev_debug(enic, "queue %u, allocating %u rx queue mbufs\n", rq->index,
305 		  rq->ring.desc_count);
306 
307 	for (i = 0; i < rq->ring.desc_count; i++, rqd++) {
308 		mb = rte_mbuf_raw_alloc(rq->mp);
309 		if (mb == NULL) {
310 			dev_err(enic, "RX mbuf alloc failed queue_id=%u\n",
311 			(unsigned)rq->index);
312 			return -ENOMEM;
313 		}
314 
315 		mb->data_off = RTE_PKTMBUF_HEADROOM;
316 		dma_addr = (dma_addr_t)(mb->buf_physaddr
317 			   + RTE_PKTMBUF_HEADROOM);
318 		rq_enet_desc_enc(rqd, dma_addr,
319 				(rq->is_sop ? RQ_ENET_TYPE_ONLY_SOP
320 				: RQ_ENET_TYPE_NOT_SOP),
321 				mb->buf_len - RTE_PKTMBUF_HEADROOM);
322 		rq->mbuf_ring[i] = mb;
323 	}
324 
325 	/* make sure all prior writes are complete before doing the PIO write */
326 	rte_rmb();
327 
328 	/* Post all but the last buffer to VIC. */
329 	rq->posted_index = rq->ring.desc_count - 1;
330 
331 	rq->rx_nb_hold = 0;
332 
333 	dev_debug(enic, "port=%u, qidx=%u, Write %u posted idx, %u sw held\n",
334 		enic->port_id, rq->index, rq->posted_index, rq->rx_nb_hold);
335 	iowrite32(rq->posted_index, &rq->ctrl->posted_index);
336 	iowrite32(0, &rq->ctrl->fetch_index);
337 	rte_rmb();
338 
339 	return 0;
340 
341 }
342 
343 static void *
344 enic_alloc_consistent(void *priv, size_t size,
345 	dma_addr_t *dma_handle, u8 *name)
346 {
347 	void *vaddr;
348 	const struct rte_memzone *rz;
349 	*dma_handle = 0;
350 	struct enic *enic = (struct enic *)priv;
351 	struct enic_memzone_entry *mze;
352 
353 	rz = rte_memzone_reserve_aligned((const char *)name,
354 					 size, SOCKET_ID_ANY, 0, ENIC_ALIGN);
355 	if (!rz) {
356 		pr_err("%s : Failed to allocate memory requested for %s\n",
357 			__func__, name);
358 		return NULL;
359 	}
360 
361 	vaddr = rz->addr;
362 	*dma_handle = (dma_addr_t)rz->phys_addr;
363 
364 	mze = rte_malloc("enic memzone entry",
365 			 sizeof(struct enic_memzone_entry), 0);
366 
367 	if (!mze) {
368 		pr_err("%s : Failed to allocate memory for memzone list\n",
369 		       __func__);
370 		rte_memzone_free(rz);
371 	}
372 
373 	mze->rz = rz;
374 
375 	rte_spinlock_lock(&enic->memzone_list_lock);
376 	LIST_INSERT_HEAD(&enic->memzone_list, mze, entries);
377 	rte_spinlock_unlock(&enic->memzone_list_lock);
378 
379 	return vaddr;
380 }
381 
382 static void
383 enic_free_consistent(void *priv,
384 		     __rte_unused size_t size,
385 		     void *vaddr,
386 		     dma_addr_t dma_handle)
387 {
388 	struct enic_memzone_entry *mze;
389 	struct enic *enic = (struct enic *)priv;
390 
391 	rte_spinlock_lock(&enic->memzone_list_lock);
392 	LIST_FOREACH(mze, &enic->memzone_list, entries) {
393 		if (mze->rz->addr == vaddr &&
394 		    mze->rz->phys_addr == dma_handle)
395 			break;
396 	}
397 	if (mze == NULL) {
398 		rte_spinlock_unlock(&enic->memzone_list_lock);
399 		dev_warning(enic,
400 			    "Tried to free memory, but couldn't find it in the memzone list\n");
401 		return;
402 	}
403 	LIST_REMOVE(mze, entries);
404 	rte_spinlock_unlock(&enic->memzone_list_lock);
405 	rte_memzone_free(mze->rz);
406 	rte_free(mze);
407 }
408 
409 int enic_link_update(struct enic *enic)
410 {
411 	struct rte_eth_dev *eth_dev = enic->rte_dev;
412 	int ret;
413 	int link_status = 0;
414 
415 	link_status = enic_get_link_status(enic);
416 	ret = (link_status == enic->link_status);
417 	enic->link_status = link_status;
418 	eth_dev->data->dev_link.link_status = link_status;
419 	eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
420 	eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
421 	return ret;
422 }
423 
424 static void
425 enic_intr_handler(void *arg)
426 {
427 	struct rte_eth_dev *dev = (struct rte_eth_dev *)arg;
428 	struct enic *enic = pmd_priv(dev);
429 
430 	vnic_intr_return_all_credits(&enic->intr);
431 
432 	enic_link_update(enic);
433 	_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL, NULL);
434 	enic_log_q_error(enic);
435 }
436 
437 int enic_enable(struct enic *enic)
438 {
439 	unsigned int index;
440 	int err;
441 	struct rte_eth_dev *eth_dev = enic->rte_dev;
442 
443 	eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
444 	eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
445 
446 	/* vnic notification of link status has already been turned on in
447 	 * enic_dev_init() which is called during probe time.  Here we are
448 	 * just turning on interrupt vector 0 if needed.
449 	 */
450 	if (eth_dev->data->dev_conf.intr_conf.lsc)
451 		vnic_dev_notify_set(enic->vdev, 0);
452 
453 	if (enic_clsf_init(enic))
454 		dev_warning(enic, "Init of hash table for clsf failed."\
455 			"Flow director feature will not work\n");
456 
457 	for (index = 0; index < enic->rq_count; index++) {
458 		err = enic_alloc_rx_queue_mbufs(enic,
459 			&enic->rq[enic_rte_rq_idx_to_sop_idx(index)]);
460 		if (err) {
461 			dev_err(enic, "Failed to alloc sop RX queue mbufs\n");
462 			return err;
463 		}
464 		err = enic_alloc_rx_queue_mbufs(enic,
465 			&enic->rq[enic_rte_rq_idx_to_data_idx(index)]);
466 		if (err) {
467 			/* release the allocated mbufs for the sop rq*/
468 			enic_rxmbuf_queue_release(enic,
469 				&enic->rq[enic_rte_rq_idx_to_sop_idx(index)]);
470 
471 			dev_err(enic, "Failed to alloc data RX queue mbufs\n");
472 			return err;
473 		}
474 	}
475 
476 	for (index = 0; index < enic->wq_count; index++)
477 		enic_start_wq(enic, index);
478 	for (index = 0; index < enic->rq_count; index++)
479 		enic_start_rq(enic, index);
480 
481 	vnic_dev_add_addr(enic->vdev, enic->mac_addr);
482 
483 	vnic_dev_enable_wait(enic->vdev);
484 
485 	/* Register and enable error interrupt */
486 	rte_intr_callback_register(&(enic->pdev->intr_handle),
487 		enic_intr_handler, (void *)enic->rte_dev);
488 
489 	rte_intr_enable(&(enic->pdev->intr_handle));
490 	vnic_intr_unmask(&enic->intr);
491 
492 	return 0;
493 }
494 
495 int enic_alloc_intr_resources(struct enic *enic)
496 {
497 	int err;
498 
499 	dev_info(enic, "vNIC resources used:  "\
500 		"wq %d rq %d cq %d intr %d\n",
501 		enic->wq_count, enic_vnic_rq_count(enic),
502 		enic->cq_count, enic->intr_count);
503 
504 	err = vnic_intr_alloc(enic->vdev, &enic->intr, 0);
505 	if (err)
506 		enic_free_vnic_resources(enic);
507 
508 	return err;
509 }
510 
511 void enic_free_rq(void *rxq)
512 {
513 	struct vnic_rq *rq_sop, *rq_data;
514 	struct enic *enic;
515 
516 	if (rxq == NULL)
517 		return;
518 
519 	rq_sop = (struct vnic_rq *)rxq;
520 	enic = vnic_dev_priv(rq_sop->vdev);
521 	rq_data = &enic->rq[rq_sop->data_queue_idx];
522 
523 	enic_rxmbuf_queue_release(enic, rq_sop);
524 	if (rq_data->in_use)
525 		enic_rxmbuf_queue_release(enic, rq_data);
526 
527 	rte_free(rq_sop->mbuf_ring);
528 	if (rq_data->in_use)
529 		rte_free(rq_data->mbuf_ring);
530 
531 	rq_sop->mbuf_ring = NULL;
532 	rq_data->mbuf_ring = NULL;
533 
534 	vnic_rq_free(rq_sop);
535 	if (rq_data->in_use)
536 		vnic_rq_free(rq_data);
537 
538 	vnic_cq_free(&enic->cq[enic_sop_rq_idx_to_cq_idx(rq_sop->index)]);
539 
540 	rq_sop->in_use = 0;
541 	rq_data->in_use = 0;
542 }
543 
544 void enic_start_wq(struct enic *enic, uint16_t queue_idx)
545 {
546 	struct rte_eth_dev *eth_dev = enic->rte_dev;
547 	vnic_wq_enable(&enic->wq[queue_idx]);
548 	eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
549 }
550 
551 int enic_stop_wq(struct enic *enic, uint16_t queue_idx)
552 {
553 	struct rte_eth_dev *eth_dev = enic->rte_dev;
554 	int ret;
555 
556 	ret = vnic_wq_disable(&enic->wq[queue_idx]);
557 	if (ret)
558 		return ret;
559 
560 	eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
561 	return 0;
562 }
563 
564 void enic_start_rq(struct enic *enic, uint16_t queue_idx)
565 {
566 	struct vnic_rq *rq_sop;
567 	struct vnic_rq *rq_data;
568 	rq_sop = &enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
569 	rq_data = &enic->rq[rq_sop->data_queue_idx];
570 	struct rte_eth_dev *eth_dev = enic->rte_dev;
571 
572 	if (rq_data->in_use)
573 		vnic_rq_enable(rq_data);
574 	rte_mb();
575 	vnic_rq_enable(rq_sop);
576 	eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
577 }
578 
579 int enic_stop_rq(struct enic *enic, uint16_t queue_idx)
580 {
581 	int ret1 = 0, ret2 = 0;
582 	struct rte_eth_dev *eth_dev = enic->rte_dev;
583 	struct vnic_rq *rq_sop;
584 	struct vnic_rq *rq_data;
585 	rq_sop = &enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
586 	rq_data = &enic->rq[rq_sop->data_queue_idx];
587 
588 	ret2 = vnic_rq_disable(rq_sop);
589 	rte_mb();
590 	if (rq_data->in_use)
591 		ret1 = vnic_rq_disable(rq_data);
592 
593 	if (ret2)
594 		return ret2;
595 	else if (ret1)
596 		return ret1;
597 
598 	eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
599 	return 0;
600 }
601 
602 int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
603 	unsigned int socket_id, struct rte_mempool *mp,
604 	uint16_t nb_desc, uint16_t free_thresh)
605 {
606 	int rc;
607 	uint16_t sop_queue_idx = enic_rte_rq_idx_to_sop_idx(queue_idx);
608 	uint16_t data_queue_idx = enic_rte_rq_idx_to_data_idx(queue_idx);
609 	struct vnic_rq *rq_sop = &enic->rq[sop_queue_idx];
610 	struct vnic_rq *rq_data = &enic->rq[data_queue_idx];
611 	unsigned int mbuf_size, mbufs_per_pkt;
612 	unsigned int nb_sop_desc, nb_data_desc;
613 	uint16_t min_sop, max_sop, min_data, max_data;
614 	uint16_t mtu = enic->rte_dev->data->mtu;
615 
616 	rq_sop->is_sop = 1;
617 	rq_sop->data_queue_idx = data_queue_idx;
618 	rq_data->is_sop = 0;
619 	rq_data->data_queue_idx = 0;
620 	rq_sop->socket_id = socket_id;
621 	rq_sop->mp = mp;
622 	rq_data->socket_id = socket_id;
623 	rq_data->mp = mp;
624 	rq_sop->in_use = 1;
625 	rq_sop->rx_free_thresh = free_thresh;
626 	rq_data->rx_free_thresh = free_thresh;
627 	dev_debug(enic, "Set queue_id:%u free thresh:%u\n", queue_idx,
628 		  free_thresh);
629 
630 	mbuf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
631 			       RTE_PKTMBUF_HEADROOM);
632 
633 	if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
634 		dev_info(enic, "Rq %u Scatter rx mode enabled\n", queue_idx);
635 		/* ceil((mtu + ETHER_HDR_LEN + 4)/mbuf_size) */
636 		mbufs_per_pkt = ((mtu + ETHER_HDR_LEN + 4) +
637 				 (mbuf_size - 1)) / mbuf_size;
638 	} else {
639 		dev_info(enic, "Scatter rx mode disabled\n");
640 		mbufs_per_pkt = 1;
641 	}
642 
643 	if (mbufs_per_pkt > 1) {
644 		dev_info(enic, "Rq %u Scatter rx mode in use\n", queue_idx);
645 		rq_sop->data_queue_enable = 1;
646 		rq_data->in_use = 1;
647 	} else {
648 		dev_info(enic, "Rq %u Scatter rx mode not being used\n",
649 			 queue_idx);
650 		rq_sop->data_queue_enable = 0;
651 		rq_data->in_use = 0;
652 	}
653 
654 	/* number of descriptors have to be a multiple of 32 */
655 	nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
656 	nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
657 
658 	rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
659 	rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
660 
661 	if (mbufs_per_pkt > 1) {
662 		min_sop = 64;
663 		max_sop = ((enic->config.rq_desc_count /
664 			    (mbufs_per_pkt - 1)) & ~0x1F);
665 		min_data = min_sop * (mbufs_per_pkt - 1);
666 		max_data = enic->config.rq_desc_count;
667 	} else {
668 		min_sop = 64;
669 		max_sop = enic->config.rq_desc_count;
670 		min_data = 0;
671 		max_data = 0;
672 	}
673 
674 	if (nb_desc < (min_sop + min_data)) {
675 		dev_warning(enic,
676 			    "Number of rx descs too low, adjusting to minimum\n");
677 		nb_sop_desc = min_sop;
678 		nb_data_desc = min_data;
679 	} else if (nb_desc > (max_sop + max_data)) {
680 		dev_warning(enic,
681 			    "Number of rx_descs too high, adjusting to maximum\n");
682 		nb_sop_desc = max_sop;
683 		nb_data_desc = max_data;
684 	}
685 	if (mbufs_per_pkt > 1) {
686 		dev_info(enic, "For mtu %d and mbuf size %d valid rx descriptor range is %d to %d\n",
687 			 mtu, mbuf_size, min_sop + min_data,
688 			 max_sop + max_data);
689 	}
690 	dev_info(enic, "Using %d rx descriptors (sop %d, data %d)\n",
691 		 nb_sop_desc + nb_data_desc, nb_sop_desc, nb_data_desc);
692 
693 	/* Allocate sop queue resources */
694 	rc = vnic_rq_alloc(enic->vdev, rq_sop, sop_queue_idx,
695 		nb_sop_desc, sizeof(struct rq_enet_desc));
696 	if (rc) {
697 		dev_err(enic, "error in allocation of sop rq\n");
698 		goto err_exit;
699 	}
700 	nb_sop_desc = rq_sop->ring.desc_count;
701 
702 	if (rq_data->in_use) {
703 		/* Allocate data queue resources */
704 		rc = vnic_rq_alloc(enic->vdev, rq_data, data_queue_idx,
705 				   nb_data_desc,
706 				   sizeof(struct rq_enet_desc));
707 		if (rc) {
708 			dev_err(enic, "error in allocation of data rq\n");
709 			goto err_free_rq_sop;
710 		}
711 		nb_data_desc = rq_data->ring.desc_count;
712 	}
713 	rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
714 			   socket_id, nb_sop_desc + nb_data_desc,
715 			   sizeof(struct cq_enet_rq_desc));
716 	if (rc) {
717 		dev_err(enic, "error in allocation of cq for rq\n");
718 		goto err_free_rq_data;
719 	}
720 
721 	/* Allocate the mbuf rings */
722 	rq_sop->mbuf_ring = (struct rte_mbuf **)
723 		rte_zmalloc_socket("rq->mbuf_ring",
724 				   sizeof(struct rte_mbuf *) * nb_sop_desc,
725 				   RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
726 	if (rq_sop->mbuf_ring == NULL)
727 		goto err_free_cq;
728 
729 	if (rq_data->in_use) {
730 		rq_data->mbuf_ring = (struct rte_mbuf **)
731 			rte_zmalloc_socket("rq->mbuf_ring",
732 				sizeof(struct rte_mbuf *) * nb_data_desc,
733 				RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
734 		if (rq_data->mbuf_ring == NULL)
735 			goto err_free_sop_mbuf;
736 	}
737 
738 	rq_sop->tot_nb_desc = nb_desc; /* squirl away for MTU update function */
739 
740 	return 0;
741 
742 err_free_sop_mbuf:
743 	rte_free(rq_sop->mbuf_ring);
744 err_free_cq:
745 	/* cleanup on error */
746 	vnic_cq_free(&enic->cq[queue_idx]);
747 err_free_rq_data:
748 	if (rq_data->in_use)
749 		vnic_rq_free(rq_data);
750 err_free_rq_sop:
751 	vnic_rq_free(rq_sop);
752 err_exit:
753 	return -ENOMEM;
754 }
755 
756 void enic_free_wq(void *txq)
757 {
758 	struct vnic_wq *wq;
759 	struct enic *enic;
760 
761 	if (txq == NULL)
762 		return;
763 
764 	wq = (struct vnic_wq *)txq;
765 	enic = vnic_dev_priv(wq->vdev);
766 	rte_memzone_free(wq->cqmsg_rz);
767 	vnic_wq_free(wq);
768 	vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
769 }
770 
771 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
772 	unsigned int socket_id, uint16_t nb_desc)
773 {
774 	int err;
775 	struct vnic_wq *wq = &enic->wq[queue_idx];
776 	unsigned int cq_index = enic_cq_wq(enic, queue_idx);
777 	char name[NAME_MAX];
778 	static int instance;
779 
780 	wq->socket_id = socket_id;
781 	if (nb_desc) {
782 		if (nb_desc > enic->config.wq_desc_count) {
783 			dev_warning(enic,
784 				"WQ %d - number of tx desc in cmd line (%d)"\
785 				"is greater than that in the UCSM/CIMC adapter"\
786 				"policy.  Applying the value in the adapter "\
787 				"policy (%d)\n",
788 				queue_idx, nb_desc, enic->config.wq_desc_count);
789 		} else if (nb_desc != enic->config.wq_desc_count) {
790 			enic->config.wq_desc_count = nb_desc;
791 			dev_info(enic,
792 				"TX Queues - effective number of descs:%d\n",
793 				nb_desc);
794 		}
795 	}
796 
797 	/* Allocate queue resources */
798 	err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
799 		enic->config.wq_desc_count,
800 		sizeof(struct wq_enet_desc));
801 	if (err) {
802 		dev_err(enic, "error in allocation of wq\n");
803 		return err;
804 	}
805 
806 	err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
807 		socket_id, enic->config.wq_desc_count,
808 		sizeof(struct cq_enet_wq_desc));
809 	if (err) {
810 		vnic_wq_free(wq);
811 		dev_err(enic, "error in allocation of cq for wq\n");
812 	}
813 
814 	/* setup up CQ message */
815 	snprintf((char *)name, sizeof(name),
816 		 "vnic_cqmsg-%s-%d-%d", enic->bdf_name, queue_idx,
817 		instance++);
818 
819 	wq->cqmsg_rz = rte_memzone_reserve_aligned((const char *)name,
820 						   sizeof(uint32_t),
821 						   SOCKET_ID_ANY, 0,
822 						   ENIC_ALIGN);
823 	if (!wq->cqmsg_rz)
824 		return -ENOMEM;
825 
826 	return err;
827 }
828 
829 int enic_disable(struct enic *enic)
830 {
831 	unsigned int i;
832 	int err;
833 
834 	vnic_intr_mask(&enic->intr);
835 	(void)vnic_intr_masked(&enic->intr); /* flush write */
836 	rte_intr_disable(&enic->pdev->intr_handle);
837 	rte_intr_callback_unregister(&enic->pdev->intr_handle,
838 				     enic_intr_handler,
839 				     (void *)enic->rte_dev);
840 
841 	vnic_dev_disable(enic->vdev);
842 
843 	enic_clsf_destroy(enic);
844 
845 	if (!enic_is_sriov_vf(enic))
846 		vnic_dev_del_addr(enic->vdev, enic->mac_addr);
847 
848 	for (i = 0; i < enic->wq_count; i++) {
849 		err = vnic_wq_disable(&enic->wq[i]);
850 		if (err)
851 			return err;
852 	}
853 	for (i = 0; i < enic_vnic_rq_count(enic); i++) {
854 		if (enic->rq[i].in_use) {
855 			err = vnic_rq_disable(&enic->rq[i]);
856 			if (err)
857 				return err;
858 		}
859 	}
860 
861 	/* If we were using interrupts, set the interrupt vector to -1
862 	 * to disable interrupts.  We are not disabling link notifcations,
863 	 * though, as we want the polling of link status to continue working.
864 	 */
865 	if (enic->rte_dev->data->dev_conf.intr_conf.lsc)
866 		vnic_dev_notify_set(enic->vdev, -1);
867 
868 	vnic_dev_set_reset_flag(enic->vdev, 1);
869 
870 	for (i = 0; i < enic->wq_count; i++)
871 		vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
872 
873 	for (i = 0; i < enic_vnic_rq_count(enic); i++)
874 		if (enic->rq[i].in_use)
875 			vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
876 	for (i = 0; i < enic->cq_count; i++)
877 		vnic_cq_clean(&enic->cq[i]);
878 	vnic_intr_clean(&enic->intr);
879 
880 	return 0;
881 }
882 
883 static int enic_dev_wait(struct vnic_dev *vdev,
884 	int (*start)(struct vnic_dev *, int),
885 	int (*finished)(struct vnic_dev *, int *),
886 	int arg)
887 {
888 	int done;
889 	int err;
890 	int i;
891 
892 	err = start(vdev, arg);
893 	if (err)
894 		return err;
895 
896 	/* Wait for func to complete...2 seconds max */
897 	for (i = 0; i < 2000; i++) {
898 		err = finished(vdev, &done);
899 		if (err)
900 			return err;
901 		if (done)
902 			return 0;
903 		usleep(1000);
904 	}
905 	return -ETIMEDOUT;
906 }
907 
908 static int enic_dev_open(struct enic *enic)
909 {
910 	int err;
911 
912 	err = enic_dev_wait(enic->vdev, vnic_dev_open,
913 		vnic_dev_open_done, 0);
914 	if (err)
915 		dev_err(enic_get_dev(enic),
916 			"vNIC device open failed, err %d\n", err);
917 
918 	return err;
919 }
920 
921 static int enic_set_rsskey(struct enic *enic)
922 {
923 	dma_addr_t rss_key_buf_pa;
924 	union vnic_rss_key *rss_key_buf_va = NULL;
925 	static union vnic_rss_key rss_key = {
926 		.key = {
927 			[0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
928 			[1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
929 			[2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
930 			[3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
931 		}
932 	};
933 	int err;
934 	u8 name[NAME_MAX];
935 
936 	snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
937 	rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
938 		&rss_key_buf_pa, name);
939 	if (!rss_key_buf_va)
940 		return -ENOMEM;
941 
942 	rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
943 
944 	err = enic_set_rss_key(enic,
945 		rss_key_buf_pa,
946 		sizeof(union vnic_rss_key));
947 
948 	enic_free_consistent(enic, sizeof(union vnic_rss_key),
949 		rss_key_buf_va, rss_key_buf_pa);
950 
951 	return err;
952 }
953 
954 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
955 {
956 	dma_addr_t rss_cpu_buf_pa;
957 	union vnic_rss_cpu *rss_cpu_buf_va = NULL;
958 	int i;
959 	int err;
960 	u8 name[NAME_MAX];
961 
962 	snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
963 	rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
964 		&rss_cpu_buf_pa, name);
965 	if (!rss_cpu_buf_va)
966 		return -ENOMEM;
967 
968 	for (i = 0; i < (1 << rss_hash_bits); i++)
969 		(*rss_cpu_buf_va).cpu[i / 4].b[i % 4] =
970 			enic_rte_rq_idx_to_sop_idx(i % enic->rq_count);
971 
972 	err = enic_set_rss_cpu(enic,
973 		rss_cpu_buf_pa,
974 		sizeof(union vnic_rss_cpu));
975 
976 	enic_free_consistent(enic, sizeof(union vnic_rss_cpu),
977 		rss_cpu_buf_va, rss_cpu_buf_pa);
978 
979 	return err;
980 }
981 
982 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
983 	u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
984 {
985 	const u8 tso_ipid_split_en = 0;
986 	int err;
987 
988 	/* Enable VLAN tag stripping */
989 
990 	err = enic_set_nic_cfg(enic,
991 		rss_default_cpu, rss_hash_type,
992 		rss_hash_bits, rss_base_cpu,
993 		rss_enable, tso_ipid_split_en,
994 		enic->ig_vlan_strip_en);
995 
996 	return err;
997 }
998 
999 int enic_set_rss_nic_cfg(struct enic *enic)
1000 {
1001 	const u8 rss_default_cpu = 0;
1002 	const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1003 	    NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1004 	    NIC_CFG_RSS_HASH_TYPE_IPV6 |
1005 	    NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1006 	const u8 rss_hash_bits = 7;
1007 	const u8 rss_base_cpu = 0;
1008 	u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1009 
1010 	if (rss_enable) {
1011 		if (!enic_set_rsskey(enic)) {
1012 			if (enic_set_rsscpu(enic, rss_hash_bits)) {
1013 				rss_enable = 0;
1014 				dev_warning(enic, "RSS disabled, "\
1015 					"Failed to set RSS cpu indirection table.");
1016 			}
1017 		} else {
1018 			rss_enable = 0;
1019 			dev_warning(enic,
1020 				"RSS disabled, Failed to set RSS key.\n");
1021 		}
1022 	}
1023 
1024 	return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1025 		rss_hash_bits, rss_base_cpu, rss_enable);
1026 }
1027 
1028 int enic_setup_finish(struct enic *enic)
1029 {
1030 	int ret;
1031 
1032 	enic_init_soft_stats(enic);
1033 
1034 	ret = enic_set_rss_nic_cfg(enic);
1035 	if (ret) {
1036 		dev_err(enic, "Failed to config nic, aborting.\n");
1037 		return -1;
1038 	}
1039 
1040 	/* Default conf */
1041 	vnic_dev_packet_filter(enic->vdev,
1042 		1 /* directed  */,
1043 		1 /* multicast */,
1044 		1 /* broadcast */,
1045 		0 /* promisc   */,
1046 		1 /* allmulti  */);
1047 
1048 	enic->promisc = 0;
1049 	enic->allmulti = 1;
1050 
1051 	return 0;
1052 }
1053 
1054 void enic_add_packet_filter(struct enic *enic)
1055 {
1056 	/* Args -> directed, multicast, broadcast, promisc, allmulti */
1057 	vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
1058 		enic->promisc, enic->allmulti);
1059 }
1060 
1061 int enic_get_link_status(struct enic *enic)
1062 {
1063 	return vnic_dev_link_status(enic->vdev);
1064 }
1065 
1066 static void enic_dev_deinit(struct enic *enic)
1067 {
1068 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1069 
1070 	/* stop link status checking */
1071 	vnic_dev_notify_unset(enic->vdev);
1072 
1073 	rte_free(eth_dev->data->mac_addrs);
1074 }
1075 
1076 
1077 int enic_set_vnic_res(struct enic *enic)
1078 {
1079 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1080 	int rc = 0;
1081 
1082 	/* With Rx scatter support, two RQs are now used per RQ used by
1083 	 * the application.
1084 	 */
1085 	if (enic->conf_rq_count < eth_dev->data->nb_rx_queues) {
1086 		dev_err(dev, "Not enough Receive queues. Requested:%u which uses %d RQs on VIC, Configured:%u\n",
1087 			eth_dev->data->nb_rx_queues,
1088 			eth_dev->data->nb_rx_queues * 2, enic->conf_rq_count);
1089 		rc = -EINVAL;
1090 	}
1091 	if (enic->conf_wq_count < eth_dev->data->nb_tx_queues) {
1092 		dev_err(dev, "Not enough Transmit queues. Requested:%u, Configured:%u\n",
1093 			eth_dev->data->nb_tx_queues, enic->conf_wq_count);
1094 		rc = -EINVAL;
1095 	}
1096 
1097 	if (enic->conf_cq_count < (eth_dev->data->nb_rx_queues +
1098 				   eth_dev->data->nb_tx_queues)) {
1099 		dev_err(dev, "Not enough Completion queues. Required:%u, Configured:%u\n",
1100 			(eth_dev->data->nb_rx_queues +
1101 			 eth_dev->data->nb_tx_queues), enic->conf_cq_count);
1102 		rc = -EINVAL;
1103 	}
1104 
1105 	if (rc == 0) {
1106 		enic->rq_count = eth_dev->data->nb_rx_queues;
1107 		enic->wq_count = eth_dev->data->nb_tx_queues;
1108 		enic->cq_count = enic->rq_count + enic->wq_count;
1109 	}
1110 
1111 	return rc;
1112 }
1113 
1114 /* Initialize the completion queue for an RQ */
1115 static int
1116 enic_reinit_rq(struct enic *enic, unsigned int rq_idx)
1117 {
1118 	struct vnic_rq *sop_rq, *data_rq;
1119 	unsigned int cq_idx = enic_cq_rq(enic, rq_idx);
1120 	int rc = 0;
1121 
1122 	sop_rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1123 	data_rq = &enic->rq[enic_rte_rq_idx_to_data_idx(rq_idx)];
1124 
1125 	vnic_cq_clean(&enic->cq[cq_idx]);
1126 	vnic_cq_init(&enic->cq[cq_idx],
1127 		     0 /* flow_control_enable */,
1128 		     1 /* color_enable */,
1129 		     0 /* cq_head */,
1130 		     0 /* cq_tail */,
1131 		     1 /* cq_tail_color */,
1132 		     0 /* interrupt_enable */,
1133 		     1 /* cq_entry_enable */,
1134 		     0 /* cq_message_enable */,
1135 		     0 /* interrupt offset */,
1136 		     0 /* cq_message_addr */);
1137 
1138 
1139 	vnic_rq_init_start(sop_rq, enic_cq_rq(enic,
1140 			   enic_rte_rq_idx_to_sop_idx(rq_idx)), 0,
1141 			   sop_rq->ring.desc_count - 1, 1, 0);
1142 	if (data_rq->in_use) {
1143 		vnic_rq_init_start(data_rq,
1144 				   enic_cq_rq(enic,
1145 				   enic_rte_rq_idx_to_data_idx(rq_idx)), 0,
1146 				   data_rq->ring.desc_count - 1, 1, 0);
1147 	}
1148 
1149 	rc = enic_alloc_rx_queue_mbufs(enic, sop_rq);
1150 	if (rc)
1151 		return rc;
1152 
1153 	if (data_rq->in_use) {
1154 		rc = enic_alloc_rx_queue_mbufs(enic, data_rq);
1155 		if (rc) {
1156 			enic_rxmbuf_queue_release(enic, sop_rq);
1157 			return rc;
1158 		}
1159 	}
1160 
1161 	return 0;
1162 }
1163 
1164 /* The Cisco NIC can send and receive packets up to a max packet size
1165  * determined by the NIC type and firmware. There is also an MTU
1166  * configured into the NIC via the CIMC/UCSM management interface
1167  * which can be overridden by this function (up to the max packet size).
1168  * Depending on the network setup, doing so may cause packet drops
1169  * and unexpected behavior.
1170  */
1171 int enic_set_mtu(struct enic *enic, uint16_t new_mtu)
1172 {
1173 	unsigned int rq_idx;
1174 	struct vnic_rq *rq;
1175 	int rc = 0;
1176 	uint16_t old_mtu;	/* previous setting */
1177 	uint16_t config_mtu;	/* Value configured into NIC via CIMC/UCSM */
1178 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1179 
1180 	old_mtu = eth_dev->data->mtu;
1181 	config_mtu = enic->config.mtu;
1182 
1183 	if (new_mtu > enic->max_mtu) {
1184 		dev_err(enic,
1185 			"MTU not updated: requested (%u) greater than max (%u)\n",
1186 			new_mtu, enic->max_mtu);
1187 		return -EINVAL;
1188 	}
1189 	if (new_mtu < ENIC_MIN_MTU) {
1190 		dev_info(enic,
1191 			"MTU not updated: requested (%u) less than min (%u)\n",
1192 			new_mtu, ENIC_MIN_MTU);
1193 		return -EINVAL;
1194 	}
1195 	if (new_mtu > config_mtu)
1196 		dev_warning(enic,
1197 			"MTU (%u) is greater than value configured in NIC (%u)\n",
1198 			new_mtu, config_mtu);
1199 
1200 	/* The easy case is when scatter is disabled. However if the MTU
1201 	 * becomes greater than the mbuf data size, packet drops will ensue.
1202 	 */
1203 	if (!enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
1204 		eth_dev->data->mtu = new_mtu;
1205 		goto set_mtu_done;
1206 	}
1207 
1208 	/* Rx scatter is enabled so reconfigure RQ's on the fly. The point is to
1209 	 * change Rx scatter mode if necessary for better performance. I.e. if
1210 	 * MTU was greater than the mbuf size and now it's less, scatter Rx
1211 	 * doesn't have to be used and vice versa.
1212 	  */
1213 	rte_spinlock_lock(&enic->mtu_lock);
1214 
1215 	/* Stop traffic on all RQs */
1216 	for (rq_idx = 0; rq_idx < enic->rq_count * 2; rq_idx++) {
1217 		rq = &enic->rq[rq_idx];
1218 		if (rq->is_sop && rq->in_use) {
1219 			rc = enic_stop_rq(enic,
1220 					  enic_sop_rq_idx_to_rte_idx(rq_idx));
1221 			if (rc) {
1222 				dev_err(enic, "Failed to stop Rq %u\n", rq_idx);
1223 				goto set_mtu_done;
1224 			}
1225 		}
1226 	}
1227 
1228 	/* replace Rx function with a no-op to avoid getting stale pkts */
1229 	eth_dev->rx_pkt_burst = enic_dummy_recv_pkts;
1230 	rte_mb();
1231 
1232 	/* Allow time for threads to exit the real Rx function. */
1233 	usleep(100000);
1234 
1235 	/* now it is safe to reconfigure the RQs */
1236 
1237 	/* update the mtu */
1238 	eth_dev->data->mtu = new_mtu;
1239 
1240 	/* free and reallocate RQs with the new MTU */
1241 	for (rq_idx = 0; rq_idx < enic->rq_count; rq_idx++) {
1242 		rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1243 
1244 		enic_free_rq(rq);
1245 		rc = enic_alloc_rq(enic, rq_idx, rq->socket_id, rq->mp,
1246 				   rq->tot_nb_desc, rq->rx_free_thresh);
1247 		if (rc) {
1248 			dev_err(enic,
1249 				"Fatal MTU alloc error- No traffic will pass\n");
1250 			goto set_mtu_done;
1251 		}
1252 
1253 		rc = enic_reinit_rq(enic, rq_idx);
1254 		if (rc) {
1255 			dev_err(enic,
1256 				"Fatal MTU RQ reinit- No traffic will pass\n");
1257 			goto set_mtu_done;
1258 		}
1259 	}
1260 
1261 	/* put back the real receive function */
1262 	rte_mb();
1263 	eth_dev->rx_pkt_burst = enic_recv_pkts;
1264 	rte_mb();
1265 
1266 	/* restart Rx traffic */
1267 	for (rq_idx = 0; rq_idx < enic->rq_count; rq_idx++) {
1268 		rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1269 		if (rq->is_sop && rq->in_use)
1270 			enic_start_rq(enic, rq_idx);
1271 	}
1272 
1273 set_mtu_done:
1274 	dev_info(enic, "MTU changed from %u to %u\n",  old_mtu, new_mtu);
1275 	rte_spinlock_unlock(&enic->mtu_lock);
1276 	return rc;
1277 }
1278 
1279 static int enic_dev_init(struct enic *enic)
1280 {
1281 	int err;
1282 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1283 
1284 	vnic_dev_intr_coal_timer_info_default(enic->vdev);
1285 
1286 	/* Get vNIC configuration
1287 	*/
1288 	err = enic_get_vnic_config(enic);
1289 	if (err) {
1290 		dev_err(dev, "Get vNIC configuration failed, aborting\n");
1291 		return err;
1292 	}
1293 
1294 	/* Get available resource counts */
1295 	enic_get_res_counts(enic);
1296 	if (enic->conf_rq_count == 1) {
1297 		dev_err(enic, "Running with only 1 RQ configured in the vNIC is not supported.\n");
1298 		dev_err(enic, "Please configure 2 RQs in the vNIC for each Rx queue used by DPDK.\n");
1299 		dev_err(enic, "See the ENIC PMD guide for more information.\n");
1300 		return -EINVAL;
1301 	}
1302 
1303 	/* Get the supported filters */
1304 	enic_fdir_info(enic);
1305 
1306 	eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN
1307 						* ENIC_MAX_MAC_ADDR, 0);
1308 	if (!eth_dev->data->mac_addrs) {
1309 		dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1310 		return -1;
1311 	}
1312 	ether_addr_copy((struct ether_addr *) enic->mac_addr,
1313 			eth_dev->data->mac_addrs);
1314 
1315 	vnic_dev_set_reset_flag(enic->vdev, 0);
1316 
1317 	LIST_INIT(&enic->flows);
1318 	rte_spinlock_init(&enic->flows_lock);
1319 
1320 	/* set up link status checking */
1321 	vnic_dev_notify_set(enic->vdev, -1); /* No Intr for notify */
1322 
1323 	return 0;
1324 
1325 }
1326 
1327 int enic_probe(struct enic *enic)
1328 {
1329 	struct rte_pci_device *pdev = enic->pdev;
1330 	int err = -1;
1331 
1332 	dev_debug(enic, " Initializing ENIC PMD\n");
1333 
1334 	enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1335 	enic->bar0.len = pdev->mem_resource[0].len;
1336 
1337 	/* Register vNIC device */
1338 	enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1339 	if (!enic->vdev) {
1340 		dev_err(enic, "vNIC registration failed, aborting\n");
1341 		goto err_out;
1342 	}
1343 
1344 	LIST_INIT(&enic->memzone_list);
1345 	rte_spinlock_init(&enic->memzone_list_lock);
1346 
1347 	vnic_register_cbacks(enic->vdev,
1348 		enic_alloc_consistent,
1349 		enic_free_consistent);
1350 
1351 	/* Issue device open to get device in known state */
1352 	err = enic_dev_open(enic);
1353 	if (err) {
1354 		dev_err(enic, "vNIC dev open failed, aborting\n");
1355 		goto err_out_unregister;
1356 	}
1357 
1358 	/* Set ingress vlan rewrite mode before vnic initialization */
1359 	err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1360 		IG_VLAN_REWRITE_MODE_PASS_THRU);
1361 	if (err) {
1362 		dev_err(enic,
1363 			"Failed to set ingress vlan rewrite mode, aborting.\n");
1364 		goto err_out_dev_close;
1365 	}
1366 
1367 	/* Issue device init to initialize the vnic-to-switch link.
1368 	 * We'll start with carrier off and wait for link UP
1369 	 * notification later to turn on carrier.  We don't need
1370 	 * to wait here for the vnic-to-switch link initialization
1371 	 * to complete; link UP notification is the indication that
1372 	 * the process is complete.
1373 	 */
1374 
1375 	err = vnic_dev_init(enic->vdev, 0);
1376 	if (err) {
1377 		dev_err(enic, "vNIC dev init failed, aborting\n");
1378 		goto err_out_dev_close;
1379 	}
1380 
1381 	err = enic_dev_init(enic);
1382 	if (err) {
1383 		dev_err(enic, "Device initialization failed, aborting\n");
1384 		goto err_out_dev_close;
1385 	}
1386 
1387 	return 0;
1388 
1389 err_out_dev_close:
1390 	vnic_dev_close(enic->vdev);
1391 err_out_unregister:
1392 	vnic_dev_unregister(enic->vdev);
1393 err_out:
1394 	return err;
1395 }
1396 
1397 void enic_remove(struct enic *enic)
1398 {
1399 	enic_dev_deinit(enic);
1400 	vnic_dev_close(enic->vdev);
1401 	vnic_dev_unregister(enic->vdev);
1402 }
1403