xref: /dpdk/drivers/net/enic/enic_main.c (revision 0857b942113874c69dc3db5df11a828ee3cc9b6b)
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 void 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;
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;
218 	}
219 }
220 
221 static void
222 enic_free_rq_buf(struct rte_mbuf **mbuf)
223 {
224 	if (*mbuf == NULL)
225 		return;
226 
227 	rte_pktmbuf_free(*mbuf);
228 	mbuf = NULL;
229 }
230 
231 void enic_init_vnic_resources(struct enic *enic)
232 {
233 	unsigned int error_interrupt_enable = 1;
234 	unsigned int error_interrupt_offset = 0;
235 	unsigned int index = 0;
236 	unsigned int cq_idx;
237 	struct vnic_rq *data_rq;
238 
239 	for (index = 0; index < enic->rq_count; index++) {
240 		cq_idx = enic_cq_rq(enic, enic_rte_rq_idx_to_sop_idx(index));
241 
242 		vnic_rq_init(&enic->rq[enic_rte_rq_idx_to_sop_idx(index)],
243 			cq_idx,
244 			error_interrupt_enable,
245 			error_interrupt_offset);
246 
247 		data_rq = &enic->rq[enic_rte_rq_idx_to_data_idx(index)];
248 		if (data_rq->in_use)
249 			vnic_rq_init(data_rq,
250 				     cq_idx,
251 				     error_interrupt_enable,
252 				     error_interrupt_offset);
253 
254 		vnic_cq_init(&enic->cq[cq_idx],
255 			0 /* flow_control_enable */,
256 			1 /* color_enable */,
257 			0 /* cq_head */,
258 			0 /* cq_tail */,
259 			1 /* cq_tail_color */,
260 			0 /* interrupt_enable */,
261 			1 /* cq_entry_enable */,
262 			0 /* cq_message_enable */,
263 			0 /* interrupt offset */,
264 			0 /* cq_message_addr */);
265 	}
266 
267 	for (index = 0; index < enic->wq_count; index++) {
268 		vnic_wq_init(&enic->wq[index],
269 			enic_cq_wq(enic, index),
270 			error_interrupt_enable,
271 			error_interrupt_offset);
272 
273 		cq_idx = enic_cq_wq(enic, index);
274 		vnic_cq_init(&enic->cq[cq_idx],
275 			0 /* flow_control_enable */,
276 			1 /* color_enable */,
277 			0 /* cq_head */,
278 			0 /* cq_tail */,
279 			1 /* cq_tail_color */,
280 			0 /* interrupt_enable */,
281 			0 /* cq_entry_enable */,
282 			1 /* cq_message_enable */,
283 			0 /* interrupt offset */,
284 			(u64)enic->wq[index].cqmsg_rz->phys_addr);
285 	}
286 
287 	vnic_intr_init(&enic->intr,
288 		enic->config.intr_timer_usec,
289 		enic->config.intr_timer_type,
290 		/*mask_on_assertion*/1);
291 }
292 
293 
294 static int
295 enic_alloc_rx_queue_mbufs(struct enic *enic, struct vnic_rq *rq)
296 {
297 	struct rte_mbuf *mb;
298 	struct rq_enet_desc *rqd = rq->ring.descs;
299 	unsigned i;
300 	dma_addr_t dma_addr;
301 
302 	if (!rq->in_use)
303 		return 0;
304 
305 	dev_debug(enic, "queue %u, allocating %u rx queue mbufs\n", rq->index,
306 		  rq->ring.desc_count);
307 
308 	for (i = 0; i < rq->ring.desc_count; i++, rqd++) {
309 		mb = rte_mbuf_raw_alloc(rq->mp);
310 		if (mb == NULL) {
311 			dev_err(enic, "RX mbuf alloc failed queue_id=%u\n",
312 			(unsigned)rq->index);
313 			return -ENOMEM;
314 		}
315 
316 		mb->data_off = RTE_PKTMBUF_HEADROOM;
317 		dma_addr = (dma_addr_t)(mb->buf_physaddr
318 			   + RTE_PKTMBUF_HEADROOM);
319 		rq_enet_desc_enc(rqd, dma_addr,
320 				(rq->is_sop ? RQ_ENET_TYPE_ONLY_SOP
321 				: RQ_ENET_TYPE_NOT_SOP),
322 				mb->buf_len - RTE_PKTMBUF_HEADROOM);
323 		rq->mbuf_ring[i] = mb;
324 	}
325 
326 	/* make sure all prior writes are complete before doing the PIO write */
327 	rte_rmb();
328 
329 	/* Post all but the last buffer to VIC. */
330 	rq->posted_index = rq->ring.desc_count - 1;
331 
332 	rq->rx_nb_hold = 0;
333 
334 	dev_debug(enic, "port=%u, qidx=%u, Write %u posted idx, %u sw held\n",
335 		enic->port_id, rq->index, rq->posted_index, rq->rx_nb_hold);
336 	iowrite32(rq->posted_index, &rq->ctrl->posted_index);
337 	iowrite32(0, &rq->ctrl->fetch_index);
338 	rte_rmb();
339 
340 	return 0;
341 
342 }
343 
344 static void *
345 enic_alloc_consistent(void *priv, size_t size,
346 	dma_addr_t *dma_handle, u8 *name)
347 {
348 	void *vaddr;
349 	const struct rte_memzone *rz;
350 	*dma_handle = 0;
351 	struct enic *enic = (struct enic *)priv;
352 	struct enic_memzone_entry *mze;
353 
354 	rz = rte_memzone_reserve_aligned((const char *)name,
355 					 size, SOCKET_ID_ANY, 0, ENIC_ALIGN);
356 	if (!rz) {
357 		pr_err("%s : Failed to allocate memory requested for %s\n",
358 			__func__, name);
359 		return NULL;
360 	}
361 
362 	vaddr = rz->addr;
363 	*dma_handle = (dma_addr_t)rz->phys_addr;
364 
365 	mze = rte_malloc("enic memzone entry",
366 			 sizeof(struct enic_memzone_entry), 0);
367 
368 	if (!mze) {
369 		pr_err("%s : Failed to allocate memory for memzone list\n",
370 		       __func__);
371 		rte_memzone_free(rz);
372 	}
373 
374 	mze->rz = rz;
375 
376 	rte_spinlock_lock(&enic->memzone_list_lock);
377 	LIST_INSERT_HEAD(&enic->memzone_list, mze, entries);
378 	rte_spinlock_unlock(&enic->memzone_list_lock);
379 
380 	return vaddr;
381 }
382 
383 static void
384 enic_free_consistent(void *priv,
385 		     __rte_unused size_t size,
386 		     void *vaddr,
387 		     dma_addr_t dma_handle)
388 {
389 	struct enic_memzone_entry *mze;
390 	struct enic *enic = (struct enic *)priv;
391 
392 	rte_spinlock_lock(&enic->memzone_list_lock);
393 	LIST_FOREACH(mze, &enic->memzone_list, entries) {
394 		if (mze->rz->addr == vaddr &&
395 		    mze->rz->phys_addr == dma_handle)
396 			break;
397 	}
398 	if (mze == NULL) {
399 		rte_spinlock_unlock(&enic->memzone_list_lock);
400 		dev_warning(enic,
401 			    "Tried to free memory, but couldn't find it in the memzone list\n");
402 		return;
403 	}
404 	LIST_REMOVE(mze, entries);
405 	rte_spinlock_unlock(&enic->memzone_list_lock);
406 	rte_memzone_free(mze->rz);
407 	rte_free(mze);
408 }
409 
410 int enic_link_update(struct enic *enic)
411 {
412 	struct rte_eth_dev *eth_dev = enic->rte_dev;
413 	int ret;
414 	int link_status = 0;
415 
416 	link_status = enic_get_link_status(enic);
417 	ret = (link_status == enic->link_status);
418 	enic->link_status = link_status;
419 	eth_dev->data->dev_link.link_status = link_status;
420 	eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
421 	eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
422 	return ret;
423 }
424 
425 static void
426 enic_intr_handler(__rte_unused struct rte_intr_handle *handle,
427 	void *arg)
428 {
429 	struct rte_eth_dev *dev = (struct rte_eth_dev *)arg;
430 	struct enic *enic = pmd_priv(dev);
431 
432 	vnic_intr_return_all_credits(&enic->intr);
433 
434 	enic_link_update(enic);
435 	_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
436 	enic_log_q_error(enic);
437 }
438 
439 int enic_enable(struct enic *enic)
440 {
441 	unsigned int index;
442 	int err;
443 	struct rte_eth_dev *eth_dev = enic->rte_dev;
444 
445 	eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
446 	eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
447 
448 	/* vnic notification of link status has already been turned on in
449 	 * enic_dev_init() which is called during probe time.  Here we are
450 	 * just turning on interrupt vector 0 if needed.
451 	 */
452 	if (eth_dev->data->dev_conf.intr_conf.lsc)
453 		vnic_dev_notify_set(enic->vdev, 0);
454 
455 	if (enic_clsf_init(enic))
456 		dev_warning(enic, "Init of hash table for clsf failed."\
457 			"Flow director feature will not work\n");
458 
459 	for (index = 0; index < enic->rq_count; index++) {
460 		err = enic_alloc_rx_queue_mbufs(enic,
461 			&enic->rq[enic_rte_rq_idx_to_sop_idx(index)]);
462 		if (err) {
463 			dev_err(enic, "Failed to alloc sop RX queue mbufs\n");
464 			return err;
465 		}
466 		err = enic_alloc_rx_queue_mbufs(enic,
467 			&enic->rq[enic_rte_rq_idx_to_data_idx(index)]);
468 		if (err) {
469 			/* release the allocated mbufs for the sop rq*/
470 			enic_rxmbuf_queue_release(enic,
471 				&enic->rq[enic_rte_rq_idx_to_sop_idx(index)]);
472 
473 			dev_err(enic, "Failed to alloc data RX queue mbufs\n");
474 			return err;
475 		}
476 	}
477 
478 	for (index = 0; index < enic->wq_count; index++)
479 		enic_start_wq(enic, index);
480 	for (index = 0; index < enic->rq_count; index++)
481 		enic_start_rq(enic, index);
482 
483 	vnic_dev_add_addr(enic->vdev, enic->mac_addr);
484 
485 	vnic_dev_enable_wait(enic->vdev);
486 
487 	/* Register and enable error interrupt */
488 	rte_intr_callback_register(&(enic->pdev->intr_handle),
489 		enic_intr_handler, (void *)enic->rte_dev);
490 
491 	rte_intr_enable(&(enic->pdev->intr_handle));
492 	vnic_intr_unmask(&enic->intr);
493 
494 	return 0;
495 }
496 
497 int enic_alloc_intr_resources(struct enic *enic)
498 {
499 	int err;
500 
501 	dev_info(enic, "vNIC resources used:  "\
502 		"wq %d rq %d cq %d intr %d\n",
503 		enic->wq_count, enic_vnic_rq_count(enic),
504 		enic->cq_count, enic->intr_count);
505 
506 	err = vnic_intr_alloc(enic->vdev, &enic->intr, 0);
507 	if (err)
508 		enic_free_vnic_resources(enic);
509 
510 	return err;
511 }
512 
513 void enic_free_rq(void *rxq)
514 {
515 	struct vnic_rq *rq_sop, *rq_data;
516 	struct enic *enic;
517 
518 	if (rxq == NULL)
519 		return;
520 
521 	rq_sop = (struct vnic_rq *)rxq;
522 	enic = vnic_dev_priv(rq_sop->vdev);
523 	rq_data = &enic->rq[rq_sop->data_queue_idx];
524 
525 	enic_rxmbuf_queue_release(enic, rq_sop);
526 	if (rq_data->in_use)
527 		enic_rxmbuf_queue_release(enic, rq_data);
528 
529 	rte_free(rq_sop->mbuf_ring);
530 	if (rq_data->in_use)
531 		rte_free(rq_data->mbuf_ring);
532 
533 	rq_sop->mbuf_ring = NULL;
534 	rq_data->mbuf_ring = NULL;
535 
536 	vnic_rq_free(rq_sop);
537 	if (rq_data->in_use)
538 		vnic_rq_free(rq_data);
539 
540 	vnic_cq_free(&enic->cq[enic_sop_rq_idx_to_cq_idx(rq_sop->index)]);
541 
542 	rq_sop->in_use = 0;
543 	rq_data->in_use = 0;
544 }
545 
546 void enic_start_wq(struct enic *enic, uint16_t queue_idx)
547 {
548 	struct rte_eth_dev *eth_dev = enic->rte_dev;
549 	vnic_wq_enable(&enic->wq[queue_idx]);
550 	eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
551 }
552 
553 int enic_stop_wq(struct enic *enic, uint16_t queue_idx)
554 {
555 	struct rte_eth_dev *eth_dev = enic->rte_dev;
556 	int ret;
557 
558 	ret = vnic_wq_disable(&enic->wq[queue_idx]);
559 	if (ret)
560 		return ret;
561 
562 	eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
563 	return 0;
564 }
565 
566 void enic_start_rq(struct enic *enic, uint16_t queue_idx)
567 {
568 	struct vnic_rq *rq_sop;
569 	struct vnic_rq *rq_data;
570 	rq_sop = &enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
571 	rq_data = &enic->rq[rq_sop->data_queue_idx];
572 	struct rte_eth_dev *eth_dev = enic->rte_dev;
573 
574 	if (rq_data->in_use)
575 		vnic_rq_enable(rq_data);
576 	rte_mb();
577 	vnic_rq_enable(rq_sop);
578 	eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
579 }
580 
581 int enic_stop_rq(struct enic *enic, uint16_t queue_idx)
582 {
583 	int ret1 = 0, ret2 = 0;
584 	struct rte_eth_dev *eth_dev = enic->rte_dev;
585 	struct vnic_rq *rq_sop;
586 	struct vnic_rq *rq_data;
587 	rq_sop = &enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
588 	rq_data = &enic->rq[rq_sop->data_queue_idx];
589 
590 	ret2 = vnic_rq_disable(rq_sop);
591 	rte_mb();
592 	if (rq_data->in_use)
593 		ret1 = vnic_rq_disable(rq_data);
594 
595 	if (ret2)
596 		return ret2;
597 	else if (ret1)
598 		return ret1;
599 
600 	eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
601 	return 0;
602 }
603 
604 int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
605 	unsigned int socket_id, struct rte_mempool *mp,
606 	uint16_t nb_desc, uint16_t free_thresh)
607 {
608 	int rc;
609 	uint16_t sop_queue_idx = enic_rte_rq_idx_to_sop_idx(queue_idx);
610 	uint16_t data_queue_idx = enic_rte_rq_idx_to_data_idx(queue_idx);
611 	struct vnic_rq *rq_sop = &enic->rq[sop_queue_idx];
612 	struct vnic_rq *rq_data = &enic->rq[data_queue_idx];
613 	unsigned int mbuf_size, mbufs_per_pkt;
614 	unsigned int nb_sop_desc, nb_data_desc;
615 	uint16_t min_sop, max_sop, min_data, max_data;
616 	uint16_t mtu = enic->rte_dev->data->mtu;
617 
618 	rq_sop->is_sop = 1;
619 	rq_sop->data_queue_idx = data_queue_idx;
620 	rq_data->is_sop = 0;
621 	rq_data->data_queue_idx = 0;
622 	rq_sop->socket_id = socket_id;
623 	rq_sop->mp = mp;
624 	rq_data->socket_id = socket_id;
625 	rq_data->mp = mp;
626 	rq_sop->in_use = 1;
627 	rq_sop->rx_free_thresh = free_thresh;
628 	rq_data->rx_free_thresh = free_thresh;
629 	dev_debug(enic, "Set queue_id:%u free thresh:%u\n", queue_idx,
630 		  free_thresh);
631 
632 	mbuf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
633 			       RTE_PKTMBUF_HEADROOM);
634 
635 	if (enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
636 		dev_info(enic, "Rq %u Scatter rx mode enabled\n", queue_idx);
637 		/* ceil((mtu + ETHER_HDR_LEN + 4)/mbuf_size) */
638 		mbufs_per_pkt = ((mtu + ETHER_HDR_LEN + 4) +
639 				 (mbuf_size - 1)) / mbuf_size;
640 	} else {
641 		dev_info(enic, "Scatter rx mode disabled\n");
642 		mbufs_per_pkt = 1;
643 	}
644 
645 	if (mbufs_per_pkt > 1) {
646 		dev_info(enic, "Rq %u Scatter rx mode in use\n", queue_idx);
647 		rq_sop->data_queue_enable = 1;
648 		rq_data->in_use = 1;
649 	} else {
650 		dev_info(enic, "Rq %u Scatter rx mode not being used\n",
651 			 queue_idx);
652 		rq_sop->data_queue_enable = 0;
653 		rq_data->in_use = 0;
654 	}
655 
656 	/* number of descriptors have to be a multiple of 32 */
657 	nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
658 	nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
659 
660 	rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
661 	rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
662 
663 	if (mbufs_per_pkt > 1) {
664 		min_sop = 64;
665 		max_sop = ((enic->config.rq_desc_count /
666 			    (mbufs_per_pkt - 1)) & ~0x1F);
667 		min_data = min_sop * (mbufs_per_pkt - 1);
668 		max_data = enic->config.rq_desc_count;
669 	} else {
670 		min_sop = 64;
671 		max_sop = enic->config.rq_desc_count;
672 		min_data = 0;
673 		max_data = 0;
674 	}
675 
676 	if (nb_desc < (min_sop + min_data)) {
677 		dev_warning(enic,
678 			    "Number of rx descs too low, adjusting to minimum\n");
679 		nb_sop_desc = min_sop;
680 		nb_data_desc = min_data;
681 	} else if (nb_desc > (max_sop + max_data)) {
682 		dev_warning(enic,
683 			    "Number of rx_descs too high, adjusting to maximum\n");
684 		nb_sop_desc = max_sop;
685 		nb_data_desc = max_data;
686 	}
687 	if (mbufs_per_pkt > 1) {
688 		dev_info(enic, "For mtu %d and mbuf size %d valid rx descriptor range is %d to %d\n",
689 			 mtu, mbuf_size, min_sop + min_data,
690 			 max_sop + max_data);
691 	}
692 	dev_info(enic, "Using %d rx descriptors (sop %d, data %d)\n",
693 		 nb_sop_desc + nb_data_desc, nb_sop_desc, nb_data_desc);
694 
695 	/* Allocate sop queue resources */
696 	rc = vnic_rq_alloc(enic->vdev, rq_sop, sop_queue_idx,
697 		nb_sop_desc, sizeof(struct rq_enet_desc));
698 	if (rc) {
699 		dev_err(enic, "error in allocation of sop rq\n");
700 		goto err_exit;
701 	}
702 	nb_sop_desc = rq_sop->ring.desc_count;
703 
704 	if (rq_data->in_use) {
705 		/* Allocate data queue resources */
706 		rc = vnic_rq_alloc(enic->vdev, rq_data, data_queue_idx,
707 				   nb_data_desc,
708 				   sizeof(struct rq_enet_desc));
709 		if (rc) {
710 			dev_err(enic, "error in allocation of data rq\n");
711 			goto err_free_rq_sop;
712 		}
713 		nb_data_desc = rq_data->ring.desc_count;
714 	}
715 	rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
716 			   socket_id, nb_sop_desc + nb_data_desc,
717 			   sizeof(struct cq_enet_rq_desc));
718 	if (rc) {
719 		dev_err(enic, "error in allocation of cq for rq\n");
720 		goto err_free_rq_data;
721 	}
722 
723 	/* Allocate the mbuf rings */
724 	rq_sop->mbuf_ring = (struct rte_mbuf **)
725 		rte_zmalloc_socket("rq->mbuf_ring",
726 				   sizeof(struct rte_mbuf *) * nb_sop_desc,
727 				   RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
728 	if (rq_sop->mbuf_ring == NULL)
729 		goto err_free_cq;
730 
731 	if (rq_data->in_use) {
732 		rq_data->mbuf_ring = (struct rte_mbuf **)
733 			rte_zmalloc_socket("rq->mbuf_ring",
734 				sizeof(struct rte_mbuf *) * nb_data_desc,
735 				RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
736 		if (rq_data->mbuf_ring == NULL)
737 			goto err_free_sop_mbuf;
738 	}
739 
740 	rq_sop->tot_nb_desc = nb_desc; /* squirl away for MTU update function */
741 
742 	return 0;
743 
744 err_free_sop_mbuf:
745 	rte_free(rq_sop->mbuf_ring);
746 err_free_cq:
747 	/* cleanup on error */
748 	vnic_cq_free(&enic->cq[queue_idx]);
749 err_free_rq_data:
750 	if (rq_data->in_use)
751 		vnic_rq_free(rq_data);
752 err_free_rq_sop:
753 	vnic_rq_free(rq_sop);
754 err_exit:
755 	return -ENOMEM;
756 }
757 
758 void enic_free_wq(void *txq)
759 {
760 	struct vnic_wq *wq;
761 	struct enic *enic;
762 
763 	if (txq == NULL)
764 		return;
765 
766 	wq = (struct vnic_wq *)txq;
767 	enic = vnic_dev_priv(wq->vdev);
768 	rte_memzone_free(wq->cqmsg_rz);
769 	vnic_wq_free(wq);
770 	vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
771 }
772 
773 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
774 	unsigned int socket_id, uint16_t nb_desc)
775 {
776 	int err;
777 	struct vnic_wq *wq = &enic->wq[queue_idx];
778 	unsigned int cq_index = enic_cq_wq(enic, queue_idx);
779 	char name[NAME_MAX];
780 	static int instance;
781 
782 	wq->socket_id = socket_id;
783 	if (nb_desc) {
784 		if (nb_desc > enic->config.wq_desc_count) {
785 			dev_warning(enic,
786 				"WQ %d - number of tx desc in cmd line (%d)"\
787 				"is greater than that in the UCSM/CIMC adapter"\
788 				"policy.  Applying the value in the adapter "\
789 				"policy (%d)\n",
790 				queue_idx, nb_desc, enic->config.wq_desc_count);
791 		} else if (nb_desc != enic->config.wq_desc_count) {
792 			enic->config.wq_desc_count = nb_desc;
793 			dev_info(enic,
794 				"TX Queues - effective number of descs:%d\n",
795 				nb_desc);
796 		}
797 	}
798 
799 	/* Allocate queue resources */
800 	err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
801 		enic->config.wq_desc_count,
802 		sizeof(struct wq_enet_desc));
803 	if (err) {
804 		dev_err(enic, "error in allocation of wq\n");
805 		return err;
806 	}
807 
808 	err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
809 		socket_id, enic->config.wq_desc_count,
810 		sizeof(struct cq_enet_wq_desc));
811 	if (err) {
812 		vnic_wq_free(wq);
813 		dev_err(enic, "error in allocation of cq for wq\n");
814 	}
815 
816 	/* setup up CQ message */
817 	snprintf((char *)name, sizeof(name),
818 		 "vnic_cqmsg-%s-%d-%d", enic->bdf_name, queue_idx,
819 		instance++);
820 
821 	wq->cqmsg_rz = rte_memzone_reserve_aligned((const char *)name,
822 						   sizeof(uint32_t),
823 						   SOCKET_ID_ANY, 0,
824 						   ENIC_ALIGN);
825 	if (!wq->cqmsg_rz)
826 		return -ENOMEM;
827 
828 	return err;
829 }
830 
831 int enic_disable(struct enic *enic)
832 {
833 	unsigned int i;
834 	int err;
835 
836 	vnic_intr_mask(&enic->intr);
837 	(void)vnic_intr_masked(&enic->intr); /* flush write */
838 	rte_intr_disable(&enic->pdev->intr_handle);
839 	rte_intr_callback_unregister(&enic->pdev->intr_handle,
840 				     enic_intr_handler,
841 				     (void *)enic->rte_dev);
842 
843 	vnic_dev_disable(enic->vdev);
844 
845 	enic_clsf_destroy(enic);
846 
847 	if (!enic_is_sriov_vf(enic))
848 		vnic_dev_del_addr(enic->vdev, enic->mac_addr);
849 
850 	for (i = 0; i < enic->wq_count; i++) {
851 		err = vnic_wq_disable(&enic->wq[i]);
852 		if (err)
853 			return err;
854 	}
855 	for (i = 0; i < enic_vnic_rq_count(enic); i++) {
856 		if (enic->rq[i].in_use) {
857 			err = vnic_rq_disable(&enic->rq[i]);
858 			if (err)
859 				return err;
860 		}
861 	}
862 
863 	/* If we were using interrupts, set the interrupt vector to -1
864 	 * to disable interrupts.  We are not disabling link notifcations,
865 	 * though, as we want the polling of link status to continue working.
866 	 */
867 	if (enic->rte_dev->data->dev_conf.intr_conf.lsc)
868 		vnic_dev_notify_set(enic->vdev, -1);
869 
870 	vnic_dev_set_reset_flag(enic->vdev, 1);
871 
872 	for (i = 0; i < enic->wq_count; i++)
873 		vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
874 
875 	for (i = 0; i < enic_vnic_rq_count(enic); i++)
876 		if (enic->rq[i].in_use)
877 			vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
878 	for (i = 0; i < enic->cq_count; i++)
879 		vnic_cq_clean(&enic->cq[i]);
880 	vnic_intr_clean(&enic->intr);
881 
882 	return 0;
883 }
884 
885 static int enic_dev_wait(struct vnic_dev *vdev,
886 	int (*start)(struct vnic_dev *, int),
887 	int (*finished)(struct vnic_dev *, int *),
888 	int arg)
889 {
890 	int done;
891 	int err;
892 	int i;
893 
894 	err = start(vdev, arg);
895 	if (err)
896 		return err;
897 
898 	/* Wait for func to complete...2 seconds max */
899 	for (i = 0; i < 2000; i++) {
900 		err = finished(vdev, &done);
901 		if (err)
902 			return err;
903 		if (done)
904 			return 0;
905 		usleep(1000);
906 	}
907 	return -ETIMEDOUT;
908 }
909 
910 static int enic_dev_open(struct enic *enic)
911 {
912 	int err;
913 
914 	err = enic_dev_wait(enic->vdev, vnic_dev_open,
915 		vnic_dev_open_done, 0);
916 	if (err)
917 		dev_err(enic_get_dev(enic),
918 			"vNIC device open failed, err %d\n", err);
919 
920 	return err;
921 }
922 
923 static int enic_set_rsskey(struct enic *enic)
924 {
925 	dma_addr_t rss_key_buf_pa;
926 	union vnic_rss_key *rss_key_buf_va = NULL;
927 	static union vnic_rss_key rss_key = {
928 		.key = {
929 			[0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
930 			[1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
931 			[2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
932 			[3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
933 		}
934 	};
935 	int err;
936 	u8 name[NAME_MAX];
937 
938 	snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
939 	rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
940 		&rss_key_buf_pa, name);
941 	if (!rss_key_buf_va)
942 		return -ENOMEM;
943 
944 	rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
945 
946 	err = enic_set_rss_key(enic,
947 		rss_key_buf_pa,
948 		sizeof(union vnic_rss_key));
949 
950 	enic_free_consistent(enic, sizeof(union vnic_rss_key),
951 		rss_key_buf_va, rss_key_buf_pa);
952 
953 	return err;
954 }
955 
956 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
957 {
958 	dma_addr_t rss_cpu_buf_pa;
959 	union vnic_rss_cpu *rss_cpu_buf_va = NULL;
960 	int i;
961 	int err;
962 	u8 name[NAME_MAX];
963 
964 	snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
965 	rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
966 		&rss_cpu_buf_pa, name);
967 	if (!rss_cpu_buf_va)
968 		return -ENOMEM;
969 
970 	for (i = 0; i < (1 << rss_hash_bits); i++)
971 		(*rss_cpu_buf_va).cpu[i / 4].b[i % 4] =
972 			enic_rte_rq_idx_to_sop_idx(i % enic->rq_count);
973 
974 	err = enic_set_rss_cpu(enic,
975 		rss_cpu_buf_pa,
976 		sizeof(union vnic_rss_cpu));
977 
978 	enic_free_consistent(enic, sizeof(union vnic_rss_cpu),
979 		rss_cpu_buf_va, rss_cpu_buf_pa);
980 
981 	return err;
982 }
983 
984 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
985 	u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
986 {
987 	const u8 tso_ipid_split_en = 0;
988 	int err;
989 
990 	/* Enable VLAN tag stripping */
991 
992 	err = enic_set_nic_cfg(enic,
993 		rss_default_cpu, rss_hash_type,
994 		rss_hash_bits, rss_base_cpu,
995 		rss_enable, tso_ipid_split_en,
996 		enic->ig_vlan_strip_en);
997 
998 	return err;
999 }
1000 
1001 int enic_set_rss_nic_cfg(struct enic *enic)
1002 {
1003 	const u8 rss_default_cpu = 0;
1004 	const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
1005 	    NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
1006 	    NIC_CFG_RSS_HASH_TYPE_IPV6 |
1007 	    NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
1008 	const u8 rss_hash_bits = 7;
1009 	const u8 rss_base_cpu = 0;
1010 	u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
1011 
1012 	if (rss_enable) {
1013 		if (!enic_set_rsskey(enic)) {
1014 			if (enic_set_rsscpu(enic, rss_hash_bits)) {
1015 				rss_enable = 0;
1016 				dev_warning(enic, "RSS disabled, "\
1017 					"Failed to set RSS cpu indirection table.");
1018 			}
1019 		} else {
1020 			rss_enable = 0;
1021 			dev_warning(enic,
1022 				"RSS disabled, Failed to set RSS key.\n");
1023 		}
1024 	}
1025 
1026 	return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
1027 		rss_hash_bits, rss_base_cpu, rss_enable);
1028 }
1029 
1030 int enic_setup_finish(struct enic *enic)
1031 {
1032 	int ret;
1033 
1034 	enic_init_soft_stats(enic);
1035 
1036 	ret = enic_set_rss_nic_cfg(enic);
1037 	if (ret) {
1038 		dev_err(enic, "Failed to config nic, aborting.\n");
1039 		return -1;
1040 	}
1041 
1042 	/* Default conf */
1043 	vnic_dev_packet_filter(enic->vdev,
1044 		1 /* directed  */,
1045 		1 /* multicast */,
1046 		1 /* broadcast */,
1047 		0 /* promisc   */,
1048 		1 /* allmulti  */);
1049 
1050 	enic->promisc = 0;
1051 	enic->allmulti = 1;
1052 
1053 	return 0;
1054 }
1055 
1056 void enic_add_packet_filter(struct enic *enic)
1057 {
1058 	/* Args -> directed, multicast, broadcast, promisc, allmulti */
1059 	vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
1060 		enic->promisc, enic->allmulti);
1061 }
1062 
1063 int enic_get_link_status(struct enic *enic)
1064 {
1065 	return vnic_dev_link_status(enic->vdev);
1066 }
1067 
1068 static void enic_dev_deinit(struct enic *enic)
1069 {
1070 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1071 
1072 	/* stop link status checking */
1073 	vnic_dev_notify_unset(enic->vdev);
1074 
1075 	rte_free(eth_dev->data->mac_addrs);
1076 }
1077 
1078 
1079 int enic_set_vnic_res(struct enic *enic)
1080 {
1081 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1082 	int rc = 0;
1083 
1084 	/* With Rx scatter support, two RQs are now used per RQ used by
1085 	 * the application.
1086 	 */
1087 	if (enic->conf_rq_count < eth_dev->data->nb_rx_queues) {
1088 		dev_err(dev, "Not enough Receive queues. Requested:%u which uses %d RQs on VIC, Configured:%u\n",
1089 			eth_dev->data->nb_rx_queues,
1090 			eth_dev->data->nb_rx_queues * 2, enic->conf_rq_count);
1091 		rc = -EINVAL;
1092 	}
1093 	if (enic->conf_wq_count < eth_dev->data->nb_tx_queues) {
1094 		dev_err(dev, "Not enough Transmit queues. Requested:%u, Configured:%u\n",
1095 			eth_dev->data->nb_tx_queues, enic->conf_wq_count);
1096 		rc = -EINVAL;
1097 	}
1098 
1099 	if (enic->conf_cq_count < (eth_dev->data->nb_rx_queues +
1100 				   eth_dev->data->nb_tx_queues)) {
1101 		dev_err(dev, "Not enough Completion queues. Required:%u, Configured:%u\n",
1102 			(eth_dev->data->nb_rx_queues +
1103 			 eth_dev->data->nb_tx_queues), enic->conf_cq_count);
1104 		rc = -EINVAL;
1105 	}
1106 
1107 	if (rc == 0) {
1108 		enic->rq_count = eth_dev->data->nb_rx_queues;
1109 		enic->wq_count = eth_dev->data->nb_tx_queues;
1110 		enic->cq_count = enic->rq_count + enic->wq_count;
1111 	}
1112 
1113 	return rc;
1114 }
1115 
1116 /* Initialize the completion queue for an RQ */
1117 static int
1118 enic_reinit_rq(struct enic *enic, unsigned int rq_idx)
1119 {
1120 	struct vnic_rq *sop_rq, *data_rq;
1121 	unsigned int cq_idx = enic_cq_rq(enic, rq_idx);
1122 	int rc = 0;
1123 
1124 	sop_rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1125 	data_rq = &enic->rq[enic_rte_rq_idx_to_data_idx(rq_idx)];
1126 
1127 	vnic_cq_clean(&enic->cq[cq_idx]);
1128 	vnic_cq_init(&enic->cq[cq_idx],
1129 		     0 /* flow_control_enable */,
1130 		     1 /* color_enable */,
1131 		     0 /* cq_head */,
1132 		     0 /* cq_tail */,
1133 		     1 /* cq_tail_color */,
1134 		     0 /* interrupt_enable */,
1135 		     1 /* cq_entry_enable */,
1136 		     0 /* cq_message_enable */,
1137 		     0 /* interrupt offset */,
1138 		     0 /* cq_message_addr */);
1139 
1140 
1141 	vnic_rq_init_start(sop_rq, enic_cq_rq(enic,
1142 			   enic_rte_rq_idx_to_sop_idx(rq_idx)), 0,
1143 			   sop_rq->ring.desc_count - 1, 1, 0);
1144 	if (data_rq->in_use) {
1145 		vnic_rq_init_start(data_rq,
1146 				   enic_cq_rq(enic,
1147 				   enic_rte_rq_idx_to_data_idx(rq_idx)), 0,
1148 				   data_rq->ring.desc_count - 1, 1, 0);
1149 	}
1150 
1151 	rc = enic_alloc_rx_queue_mbufs(enic, sop_rq);
1152 	if (rc)
1153 		return rc;
1154 
1155 	if (data_rq->in_use) {
1156 		rc = enic_alloc_rx_queue_mbufs(enic, data_rq);
1157 		if (rc) {
1158 			enic_rxmbuf_queue_release(enic, sop_rq);
1159 			return rc;
1160 		}
1161 	}
1162 
1163 	return 0;
1164 }
1165 
1166 /* The Cisco NIC can send and receive packets up to a max packet size
1167  * determined by the NIC type and firmware. There is also an MTU
1168  * configured into the NIC via the CIMC/UCSM management interface
1169  * which can be overridden by this function (up to the max packet size).
1170  * Depending on the network setup, doing so may cause packet drops
1171  * and unexpected behavior.
1172  */
1173 int enic_set_mtu(struct enic *enic, uint16_t new_mtu)
1174 {
1175 	unsigned int rq_idx;
1176 	struct vnic_rq *rq;
1177 	int rc = 0;
1178 	uint16_t old_mtu;	/* previous setting */
1179 	uint16_t config_mtu;	/* Value configured into NIC via CIMC/UCSM */
1180 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1181 
1182 	old_mtu = eth_dev->data->mtu;
1183 	config_mtu = enic->config.mtu;
1184 
1185 	if (new_mtu > enic->max_mtu) {
1186 		dev_err(enic,
1187 			"MTU not updated: requested (%u) greater than max (%u)\n",
1188 			new_mtu, enic->max_mtu);
1189 		return -EINVAL;
1190 	}
1191 	if (new_mtu < ENIC_MIN_MTU) {
1192 		dev_info(enic,
1193 			"MTU not updated: requested (%u) less than min (%u)\n",
1194 			new_mtu, ENIC_MIN_MTU);
1195 		return -EINVAL;
1196 	}
1197 	if (new_mtu > config_mtu)
1198 		dev_warning(enic,
1199 			"MTU (%u) is greater than value configured in NIC (%u)\n",
1200 			new_mtu, config_mtu);
1201 
1202 	/* The easy case is when scatter is disabled. However if the MTU
1203 	 * becomes greater than the mbuf data size, packet drops will ensue.
1204 	 */
1205 	if (!enic->rte_dev->data->dev_conf.rxmode.enable_scatter) {
1206 		eth_dev->data->mtu = new_mtu;
1207 		goto set_mtu_done;
1208 	}
1209 
1210 	/* Rx scatter is enabled so reconfigure RQ's on the fly. The point is to
1211 	 * change Rx scatter mode if necessary for better performance. I.e. if
1212 	 * MTU was greater than the mbuf size and now it's less, scatter Rx
1213 	 * doesn't have to be used and vice versa.
1214 	  */
1215 	rte_spinlock_lock(&enic->mtu_lock);
1216 
1217 	/* Stop traffic on all RQs */
1218 	for (rq_idx = 0; rq_idx < enic->rq_count * 2; rq_idx++) {
1219 		rq = &enic->rq[rq_idx];
1220 		if (rq->is_sop && rq->in_use) {
1221 			rc = enic_stop_rq(enic,
1222 					  enic_sop_rq_idx_to_rte_idx(rq_idx));
1223 			if (rc) {
1224 				dev_err(enic, "Failed to stop Rq %u\n", rq_idx);
1225 				goto set_mtu_done;
1226 			}
1227 		}
1228 	}
1229 
1230 	/* replace Rx funciton with a no-op to avoid getting stale pkts */
1231 	eth_dev->rx_pkt_burst = enic_dummy_recv_pkts;
1232 	rte_mb();
1233 
1234 	/* Allow time for threads to exit the real Rx function. */
1235 	usleep(100000);
1236 
1237 	/* now it is safe to reconfigure the RQs */
1238 
1239 	/* update the mtu */
1240 	eth_dev->data->mtu = new_mtu;
1241 
1242 	/* free and reallocate RQs with the new MTU */
1243 	for (rq_idx = 0; rq_idx < enic->rq_count; rq_idx++) {
1244 		rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1245 
1246 		enic_free_rq(rq);
1247 		rc = enic_alloc_rq(enic, rq_idx, rq->socket_id, rq->mp,
1248 				   rq->tot_nb_desc, rq->rx_free_thresh);
1249 		if (rc) {
1250 			dev_err(enic,
1251 				"Fatal MTU alloc error- No traffic will pass\n");
1252 			goto set_mtu_done;
1253 		}
1254 
1255 		rc = enic_reinit_rq(enic, rq_idx);
1256 		if (rc) {
1257 			dev_err(enic,
1258 				"Fatal MTU RQ reinit- No traffic will pass\n");
1259 			goto set_mtu_done;
1260 		}
1261 	}
1262 
1263 	/* put back the real receive function */
1264 	rte_mb();
1265 	eth_dev->rx_pkt_burst = enic_recv_pkts;
1266 	rte_mb();
1267 
1268 	/* restart Rx traffic */
1269 	for (rq_idx = 0; rq_idx < enic->rq_count; rq_idx++) {
1270 		rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1271 		if (rq->is_sop && rq->in_use)
1272 			enic_start_rq(enic, rq_idx);
1273 	}
1274 
1275 set_mtu_done:
1276 	dev_info(enic, "MTU changed from %u to %u\n",  old_mtu, new_mtu);
1277 	rte_spinlock_unlock(&enic->mtu_lock);
1278 	return rc;
1279 }
1280 
1281 static int enic_dev_init(struct enic *enic)
1282 {
1283 	int err;
1284 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1285 
1286 	vnic_dev_intr_coal_timer_info_default(enic->vdev);
1287 
1288 	/* Get vNIC configuration
1289 	*/
1290 	err = enic_get_vnic_config(enic);
1291 	if (err) {
1292 		dev_err(dev, "Get vNIC configuration failed, aborting\n");
1293 		return err;
1294 	}
1295 
1296 	/* Get available resource counts */
1297 	enic_get_res_counts(enic);
1298 	if (enic->conf_rq_count == 1) {
1299 		dev_err(enic, "Running with only 1 RQ configured in the vNIC is not supported.\n");
1300 		dev_err(enic, "Please configure 2 RQs in the vNIC for each Rx queue used by DPDK.\n");
1301 		dev_err(enic, "See the ENIC PMD guide for more information.\n");
1302 		return -EINVAL;
1303 	}
1304 
1305 	/* Get the supported filters */
1306 	enic_fdir_info(enic);
1307 
1308 	eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN
1309 						* ENIC_MAX_MAC_ADDR, 0);
1310 	if (!eth_dev->data->mac_addrs) {
1311 		dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1312 		return -1;
1313 	}
1314 	ether_addr_copy((struct ether_addr *) enic->mac_addr,
1315 			eth_dev->data->mac_addrs);
1316 
1317 	vnic_dev_set_reset_flag(enic->vdev, 0);
1318 
1319 	/* set up link status checking */
1320 	vnic_dev_notify_set(enic->vdev, -1); /* No Intr for notify */
1321 
1322 	return 0;
1323 
1324 }
1325 
1326 int enic_probe(struct enic *enic)
1327 {
1328 	struct rte_pci_device *pdev = enic->pdev;
1329 	int err = -1;
1330 
1331 	dev_debug(enic, " Initializing ENIC PMD\n");
1332 
1333 	enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1334 	enic->bar0.len = pdev->mem_resource[0].len;
1335 
1336 	/* Register vNIC device */
1337 	enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1338 	if (!enic->vdev) {
1339 		dev_err(enic, "vNIC registration failed, aborting\n");
1340 		goto err_out;
1341 	}
1342 
1343 	LIST_INIT(&enic->memzone_list);
1344 	rte_spinlock_init(&enic->memzone_list_lock);
1345 
1346 	vnic_register_cbacks(enic->vdev,
1347 		enic_alloc_consistent,
1348 		enic_free_consistent);
1349 
1350 	/* Issue device open to get device in known state */
1351 	err = enic_dev_open(enic);
1352 	if (err) {
1353 		dev_err(enic, "vNIC dev open failed, aborting\n");
1354 		goto err_out_unregister;
1355 	}
1356 
1357 	/* Set ingress vlan rewrite mode before vnic initialization */
1358 	err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1359 		IG_VLAN_REWRITE_MODE_PASS_THRU);
1360 	if (err) {
1361 		dev_err(enic,
1362 			"Failed to set ingress vlan rewrite mode, aborting.\n");
1363 		goto err_out_dev_close;
1364 	}
1365 
1366 	/* Issue device init to initialize the vnic-to-switch link.
1367 	 * We'll start with carrier off and wait for link UP
1368 	 * notification later to turn on carrier.  We don't need
1369 	 * to wait here for the vnic-to-switch link initialization
1370 	 * to complete; link UP notification is the indication that
1371 	 * the process is complete.
1372 	 */
1373 
1374 	err = vnic_dev_init(enic->vdev, 0);
1375 	if (err) {
1376 		dev_err(enic, "vNIC dev init failed, aborting\n");
1377 		goto err_out_dev_close;
1378 	}
1379 
1380 	err = enic_dev_init(enic);
1381 	if (err) {
1382 		dev_err(enic, "Device initialization failed, aborting\n");
1383 		goto err_out_dev_close;
1384 	}
1385 
1386 	return 0;
1387 
1388 err_out_dev_close:
1389 	vnic_dev_close(enic->vdev);
1390 err_out_unregister:
1391 	vnic_dev_unregister(enic->vdev);
1392 err_out:
1393 	return err;
1394 }
1395 
1396 void enic_remove(struct enic *enic)
1397 {
1398 	enic_dev_deinit(enic);
1399 	vnic_dev_close(enic->vdev);
1400 	vnic_dev_unregister(enic->vdev);
1401 }
1402