xref: /dpdk/drivers/net/enic/enic_main.c (revision 945acb4a0d644d194f1823084a234f9c286dcf8c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2008-2017 Cisco Systems, Inc.  All rights reserved.
3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
4  */
5 
6 #include <stdio.h>
7 
8 #include <sys/stat.h>
9 #include <sys/mman.h>
10 #include <fcntl.h>
11 #include <libgen.h>
12 
13 #include <rte_pci.h>
14 #include <rte_bus_pci.h>
15 #include <rte_memzone.h>
16 #include <rte_malloc.h>
17 #include <rte_mbuf.h>
18 #include <rte_string_fns.h>
19 #include <rte_ethdev.h>
20 
21 #include "enic_compat.h"
22 #include "enic.h"
23 #include "wq_enet_desc.h"
24 #include "rq_enet_desc.h"
25 #include "cq_enet_desc.h"
26 #include "vnic_enet.h"
27 #include "vnic_dev.h"
28 #include "vnic_wq.h"
29 #include "vnic_rq.h"
30 #include "vnic_cq.h"
31 #include "vnic_intr.h"
32 #include "vnic_nic.h"
33 
34 static inline int enic_is_sriov_vf(struct enic *enic)
35 {
36 	return enic->pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_VF;
37 }
38 
39 static int is_zero_addr(uint8_t *addr)
40 {
41 	return !(addr[0] |  addr[1] | addr[2] | addr[3] | addr[4] | addr[5]);
42 }
43 
44 static int is_mcast_addr(uint8_t *addr)
45 {
46 	return addr[0] & 1;
47 }
48 
49 static int is_eth_addr_valid(uint8_t *addr)
50 {
51 	return !is_mcast_addr(addr) && !is_zero_addr(addr);
52 }
53 
54 static void
55 enic_rxmbuf_queue_release(__rte_unused struct enic *enic, struct vnic_rq *rq)
56 {
57 	uint16_t i;
58 
59 	if (!rq || !rq->mbuf_ring) {
60 		dev_debug(enic, "Pointer to rq or mbuf_ring is NULL");
61 		return;
62 	}
63 
64 	for (i = 0; i < rq->ring.desc_count; i++) {
65 		if (rq->mbuf_ring[i]) {
66 			rte_pktmbuf_free_seg(rq->mbuf_ring[i]);
67 			rq->mbuf_ring[i] = NULL;
68 		}
69 	}
70 }
71 
72 static void enic_free_wq_buf(struct vnic_wq_buf *buf)
73 {
74 	struct rte_mbuf *mbuf = (struct rte_mbuf *)buf->mb;
75 
76 	rte_pktmbuf_free_seg(mbuf);
77 	buf->mb = NULL;
78 }
79 
80 static void enic_log_q_error(struct enic *enic)
81 {
82 	unsigned int i;
83 	u32 error_status;
84 
85 	for (i = 0; i < enic->wq_count; i++) {
86 		error_status = vnic_wq_error_status(&enic->wq[i]);
87 		if (error_status)
88 			dev_err(enic, "WQ[%d] error_status %d\n", i,
89 				error_status);
90 	}
91 
92 	for (i = 0; i < enic_vnic_rq_count(enic); i++) {
93 		if (!enic->rq[i].in_use)
94 			continue;
95 		error_status = vnic_rq_error_status(&enic->rq[i]);
96 		if (error_status)
97 			dev_err(enic, "RQ[%d] error_status %d\n", i,
98 				error_status);
99 	}
100 }
101 
102 static void enic_clear_soft_stats(struct enic *enic)
103 {
104 	struct enic_soft_stats *soft_stats = &enic->soft_stats;
105 	rte_atomic64_clear(&soft_stats->rx_nombuf);
106 	rte_atomic64_clear(&soft_stats->rx_packet_errors);
107 	rte_atomic64_clear(&soft_stats->tx_oversized);
108 }
109 
110 static void enic_init_soft_stats(struct enic *enic)
111 {
112 	struct enic_soft_stats *soft_stats = &enic->soft_stats;
113 	rte_atomic64_init(&soft_stats->rx_nombuf);
114 	rte_atomic64_init(&soft_stats->rx_packet_errors);
115 	rte_atomic64_init(&soft_stats->tx_oversized);
116 	enic_clear_soft_stats(enic);
117 }
118 
119 void enic_dev_stats_clear(struct enic *enic)
120 {
121 	if (vnic_dev_stats_clear(enic->vdev))
122 		dev_err(enic, "Error in clearing stats\n");
123 	enic_clear_soft_stats(enic);
124 }
125 
126 int enic_dev_stats_get(struct enic *enic, struct rte_eth_stats *r_stats)
127 {
128 	struct vnic_stats *stats;
129 	struct enic_soft_stats *soft_stats = &enic->soft_stats;
130 	int64_t rx_truncated;
131 	uint64_t rx_packet_errors;
132 	int ret = vnic_dev_stats_dump(enic->vdev, &stats);
133 
134 	if (ret) {
135 		dev_err(enic, "Error in getting stats\n");
136 		return ret;
137 	}
138 
139 	/* The number of truncated packets can only be calculated by
140 	 * subtracting a hardware counter from error packets received by
141 	 * the driver. Note: this causes transient inaccuracies in the
142 	 * ipackets count. Also, the length of truncated packets are
143 	 * counted in ibytes even though truncated packets are dropped
144 	 * which can make ibytes be slightly higher than it should be.
145 	 */
146 	rx_packet_errors = rte_atomic64_read(&soft_stats->rx_packet_errors);
147 	rx_truncated = rx_packet_errors - stats->rx.rx_errors;
148 
149 	r_stats->ipackets = stats->rx.rx_frames_ok - rx_truncated;
150 	r_stats->opackets = stats->tx.tx_frames_ok;
151 
152 	r_stats->ibytes = stats->rx.rx_bytes_ok;
153 	r_stats->obytes = stats->tx.tx_bytes_ok;
154 
155 	r_stats->ierrors = stats->rx.rx_errors + stats->rx.rx_drop;
156 	r_stats->oerrors = stats->tx.tx_errors
157 			   + rte_atomic64_read(&soft_stats->tx_oversized);
158 
159 	r_stats->imissed = stats->rx.rx_no_bufs + rx_truncated;
160 
161 	r_stats->rx_nombuf = rte_atomic64_read(&soft_stats->rx_nombuf);
162 	return 0;
163 }
164 
165 void enic_del_mac_address(struct enic *enic, int mac_index)
166 {
167 	struct rte_eth_dev *eth_dev = enic->rte_dev;
168 	uint8_t *mac_addr = eth_dev->data->mac_addrs[mac_index].addr_bytes;
169 
170 	if (vnic_dev_del_addr(enic->vdev, mac_addr))
171 		dev_err(enic, "del mac addr failed\n");
172 }
173 
174 int enic_set_mac_address(struct enic *enic, uint8_t *mac_addr)
175 {
176 	int err;
177 
178 	if (!is_eth_addr_valid(mac_addr)) {
179 		dev_err(enic, "invalid mac address\n");
180 		return -EINVAL;
181 	}
182 
183 	err = vnic_dev_add_addr(enic->vdev, mac_addr);
184 	if (err)
185 		dev_err(enic, "add mac addr failed\n");
186 	return err;
187 }
188 
189 static void
190 enic_free_rq_buf(struct rte_mbuf **mbuf)
191 {
192 	if (*mbuf == NULL)
193 		return;
194 
195 	rte_pktmbuf_free(*mbuf);
196 	*mbuf = NULL;
197 }
198 
199 void enic_init_vnic_resources(struct enic *enic)
200 {
201 	unsigned int error_interrupt_enable = 1;
202 	unsigned int error_interrupt_offset = 0;
203 	unsigned int index = 0;
204 	unsigned int cq_idx;
205 	struct vnic_rq *data_rq;
206 
207 	for (index = 0; index < enic->rq_count; index++) {
208 		cq_idx = enic_cq_rq(enic, enic_rte_rq_idx_to_sop_idx(index));
209 
210 		vnic_rq_init(&enic->rq[enic_rte_rq_idx_to_sop_idx(index)],
211 			cq_idx,
212 			error_interrupt_enable,
213 			error_interrupt_offset);
214 
215 		data_rq = &enic->rq[enic_rte_rq_idx_to_data_idx(index)];
216 		if (data_rq->in_use)
217 			vnic_rq_init(data_rq,
218 				     cq_idx,
219 				     error_interrupt_enable,
220 				     error_interrupt_offset);
221 
222 		vnic_cq_init(&enic->cq[cq_idx],
223 			0 /* flow_control_enable */,
224 			1 /* color_enable */,
225 			0 /* cq_head */,
226 			0 /* cq_tail */,
227 			1 /* cq_tail_color */,
228 			0 /* interrupt_enable */,
229 			1 /* cq_entry_enable */,
230 			0 /* cq_message_enable */,
231 			0 /* interrupt offset */,
232 			0 /* cq_message_addr */);
233 	}
234 
235 	for (index = 0; index < enic->wq_count; index++) {
236 		vnic_wq_init(&enic->wq[index],
237 			enic_cq_wq(enic, index),
238 			error_interrupt_enable,
239 			error_interrupt_offset);
240 
241 		cq_idx = enic_cq_wq(enic, index);
242 		vnic_cq_init(&enic->cq[cq_idx],
243 			0 /* flow_control_enable */,
244 			1 /* color_enable */,
245 			0 /* cq_head */,
246 			0 /* cq_tail */,
247 			1 /* cq_tail_color */,
248 			0 /* interrupt_enable */,
249 			0 /* cq_entry_enable */,
250 			1 /* cq_message_enable */,
251 			0 /* interrupt offset */,
252 			(u64)enic->wq[index].cqmsg_rz->iova);
253 	}
254 
255 	vnic_intr_init(&enic->intr,
256 		enic->config.intr_timer_usec,
257 		enic->config.intr_timer_type,
258 		/*mask_on_assertion*/1);
259 }
260 
261 
262 static int
263 enic_alloc_rx_queue_mbufs(struct enic *enic, struct vnic_rq *rq)
264 {
265 	struct rte_mbuf *mb;
266 	struct rq_enet_desc *rqd = rq->ring.descs;
267 	unsigned i;
268 	dma_addr_t dma_addr;
269 
270 	if (!rq->in_use)
271 		return 0;
272 
273 	dev_debug(enic, "queue %u, allocating %u rx queue mbufs\n", rq->index,
274 		  rq->ring.desc_count);
275 
276 	for (i = 0; i < rq->ring.desc_count; i++, rqd++) {
277 		mb = rte_mbuf_raw_alloc(rq->mp);
278 		if (mb == NULL) {
279 			dev_err(enic, "RX mbuf alloc failed queue_id=%u\n",
280 			(unsigned)rq->index);
281 			return -ENOMEM;
282 		}
283 
284 		mb->data_off = RTE_PKTMBUF_HEADROOM;
285 		dma_addr = (dma_addr_t)(mb->buf_iova
286 			   + RTE_PKTMBUF_HEADROOM);
287 		rq_enet_desc_enc(rqd, dma_addr,
288 				(rq->is_sop ? RQ_ENET_TYPE_ONLY_SOP
289 				: RQ_ENET_TYPE_NOT_SOP),
290 				mb->buf_len - RTE_PKTMBUF_HEADROOM);
291 		rq->mbuf_ring[i] = mb;
292 	}
293 
294 	/* make sure all prior writes are complete before doing the PIO write */
295 	rte_rmb();
296 
297 	/* Post all but the last buffer to VIC. */
298 	rq->posted_index = rq->ring.desc_count - 1;
299 
300 	rq->rx_nb_hold = 0;
301 
302 	dev_debug(enic, "port=%u, qidx=%u, Write %u posted idx, %u sw held\n",
303 		enic->port_id, rq->index, rq->posted_index, rq->rx_nb_hold);
304 	iowrite32(rq->posted_index, &rq->ctrl->posted_index);
305 	iowrite32(0, &rq->ctrl->fetch_index);
306 	rte_rmb();
307 
308 	return 0;
309 
310 }
311 
312 static void *
313 enic_alloc_consistent(void *priv, size_t size,
314 	dma_addr_t *dma_handle, u8 *name)
315 {
316 	void *vaddr;
317 	const struct rte_memzone *rz;
318 	*dma_handle = 0;
319 	struct enic *enic = (struct enic *)priv;
320 	struct enic_memzone_entry *mze;
321 
322 	rz = rte_memzone_reserve_aligned((const char *)name,
323 					 size, SOCKET_ID_ANY, 0, ENIC_ALIGN);
324 	if (!rz) {
325 		pr_err("%s : Failed to allocate memory requested for %s\n",
326 			__func__, name);
327 		return NULL;
328 	}
329 
330 	vaddr = rz->addr;
331 	*dma_handle = (dma_addr_t)rz->iova;
332 
333 	mze = rte_malloc("enic memzone entry",
334 			 sizeof(struct enic_memzone_entry), 0);
335 
336 	if (!mze) {
337 		pr_err("%s : Failed to allocate memory for memzone list\n",
338 		       __func__);
339 		rte_memzone_free(rz);
340 		return NULL;
341 	}
342 
343 	mze->rz = rz;
344 
345 	rte_spinlock_lock(&enic->memzone_list_lock);
346 	LIST_INSERT_HEAD(&enic->memzone_list, mze, entries);
347 	rte_spinlock_unlock(&enic->memzone_list_lock);
348 
349 	return vaddr;
350 }
351 
352 static void
353 enic_free_consistent(void *priv,
354 		     __rte_unused size_t size,
355 		     void *vaddr,
356 		     dma_addr_t dma_handle)
357 {
358 	struct enic_memzone_entry *mze;
359 	struct enic *enic = (struct enic *)priv;
360 
361 	rte_spinlock_lock(&enic->memzone_list_lock);
362 	LIST_FOREACH(mze, &enic->memzone_list, entries) {
363 		if (mze->rz->addr == vaddr &&
364 		    mze->rz->iova == dma_handle)
365 			break;
366 	}
367 	if (mze == NULL) {
368 		rte_spinlock_unlock(&enic->memzone_list_lock);
369 		dev_warning(enic,
370 			    "Tried to free memory, but couldn't find it in the memzone list\n");
371 		return;
372 	}
373 	LIST_REMOVE(mze, entries);
374 	rte_spinlock_unlock(&enic->memzone_list_lock);
375 	rte_memzone_free(mze->rz);
376 	rte_free(mze);
377 }
378 
379 int enic_link_update(struct enic *enic)
380 {
381 	struct rte_eth_dev *eth_dev = enic->rte_dev;
382 	int ret;
383 	int link_status = 0;
384 
385 	link_status = enic_get_link_status(enic);
386 	ret = (link_status == enic->link_status);
387 	enic->link_status = link_status;
388 	eth_dev->data->dev_link.link_status = link_status;
389 	eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
390 	eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
391 	return ret;
392 }
393 
394 static void
395 enic_intr_handler(void *arg)
396 {
397 	struct rte_eth_dev *dev = (struct rte_eth_dev *)arg;
398 	struct enic *enic = pmd_priv(dev);
399 
400 	vnic_intr_return_all_credits(&enic->intr);
401 
402 	enic_link_update(enic);
403 	_rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
404 	enic_log_q_error(enic);
405 }
406 
407 int enic_enable(struct enic *enic)
408 {
409 	unsigned int index;
410 	int err;
411 	struct rte_eth_dev *eth_dev = enic->rte_dev;
412 
413 	eth_dev->data->dev_link.link_speed = vnic_dev_port_speed(enic->vdev);
414 	eth_dev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
415 
416 	/* vnic notification of link status has already been turned on in
417 	 * enic_dev_init() which is called during probe time.  Here we are
418 	 * just turning on interrupt vector 0 if needed.
419 	 */
420 	if (eth_dev->data->dev_conf.intr_conf.lsc)
421 		vnic_dev_notify_set(enic->vdev, 0);
422 
423 	if (enic_clsf_init(enic))
424 		dev_warning(enic, "Init of hash table for clsf failed."\
425 			"Flow director feature will not work\n");
426 
427 	for (index = 0; index < enic->rq_count; index++) {
428 		err = enic_alloc_rx_queue_mbufs(enic,
429 			&enic->rq[enic_rte_rq_idx_to_sop_idx(index)]);
430 		if (err) {
431 			dev_err(enic, "Failed to alloc sop RX queue mbufs\n");
432 			return err;
433 		}
434 		err = enic_alloc_rx_queue_mbufs(enic,
435 			&enic->rq[enic_rte_rq_idx_to_data_idx(index)]);
436 		if (err) {
437 			/* release the allocated mbufs for the sop rq*/
438 			enic_rxmbuf_queue_release(enic,
439 				&enic->rq[enic_rte_rq_idx_to_sop_idx(index)]);
440 
441 			dev_err(enic, "Failed to alloc data RX queue mbufs\n");
442 			return err;
443 		}
444 	}
445 
446 	for (index = 0; index < enic->wq_count; index++)
447 		enic_start_wq(enic, index);
448 	for (index = 0; index < enic->rq_count; index++)
449 		enic_start_rq(enic, index);
450 
451 	vnic_dev_add_addr(enic->vdev, enic->mac_addr);
452 
453 	vnic_dev_enable_wait(enic->vdev);
454 
455 	/* Register and enable error interrupt */
456 	rte_intr_callback_register(&(enic->pdev->intr_handle),
457 		enic_intr_handler, (void *)enic->rte_dev);
458 
459 	rte_intr_enable(&(enic->pdev->intr_handle));
460 	vnic_intr_unmask(&enic->intr);
461 
462 	return 0;
463 }
464 
465 int enic_alloc_intr_resources(struct enic *enic)
466 {
467 	int err;
468 
469 	dev_info(enic, "vNIC resources used:  "\
470 		"wq %d rq %d cq %d intr %d\n",
471 		enic->wq_count, enic_vnic_rq_count(enic),
472 		enic->cq_count, enic->intr_count);
473 
474 	err = vnic_intr_alloc(enic->vdev, &enic->intr, 0);
475 	if (err)
476 		enic_free_vnic_resources(enic);
477 
478 	return err;
479 }
480 
481 void enic_free_rq(void *rxq)
482 {
483 	struct vnic_rq *rq_sop, *rq_data;
484 	struct enic *enic;
485 
486 	if (rxq == NULL)
487 		return;
488 
489 	rq_sop = (struct vnic_rq *)rxq;
490 	enic = vnic_dev_priv(rq_sop->vdev);
491 	rq_data = &enic->rq[rq_sop->data_queue_idx];
492 
493 	enic_rxmbuf_queue_release(enic, rq_sop);
494 	if (rq_data->in_use)
495 		enic_rxmbuf_queue_release(enic, rq_data);
496 
497 	rte_free(rq_sop->mbuf_ring);
498 	if (rq_data->in_use)
499 		rte_free(rq_data->mbuf_ring);
500 
501 	rq_sop->mbuf_ring = NULL;
502 	rq_data->mbuf_ring = NULL;
503 
504 	vnic_rq_free(rq_sop);
505 	if (rq_data->in_use)
506 		vnic_rq_free(rq_data);
507 
508 	vnic_cq_free(&enic->cq[enic_sop_rq_idx_to_cq_idx(rq_sop->index)]);
509 
510 	rq_sop->in_use = 0;
511 	rq_data->in_use = 0;
512 }
513 
514 void enic_start_wq(struct enic *enic, uint16_t queue_idx)
515 {
516 	struct rte_eth_dev *eth_dev = enic->rte_dev;
517 	vnic_wq_enable(&enic->wq[queue_idx]);
518 	eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
519 }
520 
521 int enic_stop_wq(struct enic *enic, uint16_t queue_idx)
522 {
523 	struct rte_eth_dev *eth_dev = enic->rte_dev;
524 	int ret;
525 
526 	ret = vnic_wq_disable(&enic->wq[queue_idx]);
527 	if (ret)
528 		return ret;
529 
530 	eth_dev->data->tx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
531 	return 0;
532 }
533 
534 void enic_start_rq(struct enic *enic, uint16_t queue_idx)
535 {
536 	struct vnic_rq *rq_sop;
537 	struct vnic_rq *rq_data;
538 	rq_sop = &enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
539 	rq_data = &enic->rq[rq_sop->data_queue_idx];
540 	struct rte_eth_dev *eth_dev = enic->rte_dev;
541 
542 	if (rq_data->in_use)
543 		vnic_rq_enable(rq_data);
544 	rte_mb();
545 	vnic_rq_enable(rq_sop);
546 	eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STARTED;
547 }
548 
549 int enic_stop_rq(struct enic *enic, uint16_t queue_idx)
550 {
551 	int ret1 = 0, ret2 = 0;
552 	struct rte_eth_dev *eth_dev = enic->rte_dev;
553 	struct vnic_rq *rq_sop;
554 	struct vnic_rq *rq_data;
555 	rq_sop = &enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
556 	rq_data = &enic->rq[rq_sop->data_queue_idx];
557 
558 	ret2 = vnic_rq_disable(rq_sop);
559 	rte_mb();
560 	if (rq_data->in_use)
561 		ret1 = vnic_rq_disable(rq_data);
562 
563 	if (ret2)
564 		return ret2;
565 	else if (ret1)
566 		return ret1;
567 
568 	eth_dev->data->rx_queue_state[queue_idx] = RTE_ETH_QUEUE_STATE_STOPPED;
569 	return 0;
570 }
571 
572 int enic_alloc_rq(struct enic *enic, uint16_t queue_idx,
573 	unsigned int socket_id, struct rte_mempool *mp,
574 	uint16_t nb_desc, uint16_t free_thresh)
575 {
576 	int rc;
577 	uint16_t sop_queue_idx = enic_rte_rq_idx_to_sop_idx(queue_idx);
578 	uint16_t data_queue_idx = enic_rte_rq_idx_to_data_idx(queue_idx);
579 	struct vnic_rq *rq_sop = &enic->rq[sop_queue_idx];
580 	struct vnic_rq *rq_data = &enic->rq[data_queue_idx];
581 	unsigned int mbuf_size, mbufs_per_pkt;
582 	unsigned int nb_sop_desc, nb_data_desc;
583 	uint16_t min_sop, max_sop, min_data, max_data;
584 	uint16_t mtu = enic->rte_dev->data->mtu;
585 
586 	rq_sop->is_sop = 1;
587 	rq_sop->data_queue_idx = data_queue_idx;
588 	rq_data->is_sop = 0;
589 	rq_data->data_queue_idx = 0;
590 	rq_sop->socket_id = socket_id;
591 	rq_sop->mp = mp;
592 	rq_data->socket_id = socket_id;
593 	rq_data->mp = mp;
594 	rq_sop->in_use = 1;
595 	rq_sop->rx_free_thresh = free_thresh;
596 	rq_data->rx_free_thresh = free_thresh;
597 	dev_debug(enic, "Set queue_id:%u free thresh:%u\n", queue_idx,
598 		  free_thresh);
599 
600 	mbuf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
601 			       RTE_PKTMBUF_HEADROOM);
602 
603 	if (enic->rte_dev->data->dev_conf.rxmode.offloads &
604 	    DEV_RX_OFFLOAD_SCATTER) {
605 		dev_info(enic, "Rq %u Scatter rx mode enabled\n", queue_idx);
606 		/* ceil((mtu + ETHER_HDR_LEN + 4)/mbuf_size) */
607 		mbufs_per_pkt = ((mtu + ETHER_HDR_LEN + 4) +
608 				 (mbuf_size - 1)) / mbuf_size;
609 	} else {
610 		dev_info(enic, "Scatter rx mode disabled\n");
611 		mbufs_per_pkt = 1;
612 	}
613 
614 	if (mbufs_per_pkt > 1) {
615 		dev_info(enic, "Rq %u Scatter rx mode in use\n", queue_idx);
616 		rq_sop->data_queue_enable = 1;
617 		rq_data->in_use = 1;
618 	} else {
619 		dev_info(enic, "Rq %u Scatter rx mode not being used\n",
620 			 queue_idx);
621 		rq_sop->data_queue_enable = 0;
622 		rq_data->in_use = 0;
623 	}
624 
625 	/* number of descriptors have to be a multiple of 32 */
626 	nb_sop_desc = (nb_desc / mbufs_per_pkt) & ~0x1F;
627 	nb_data_desc = (nb_desc - nb_sop_desc) & ~0x1F;
628 
629 	rq_sop->max_mbufs_per_pkt = mbufs_per_pkt;
630 	rq_data->max_mbufs_per_pkt = mbufs_per_pkt;
631 
632 	if (mbufs_per_pkt > 1) {
633 		min_sop = 64;
634 		max_sop = ((enic->config.rq_desc_count /
635 			    (mbufs_per_pkt - 1)) & ~0x1F);
636 		min_data = min_sop * (mbufs_per_pkt - 1);
637 		max_data = enic->config.rq_desc_count;
638 	} else {
639 		min_sop = 64;
640 		max_sop = enic->config.rq_desc_count;
641 		min_data = 0;
642 		max_data = 0;
643 	}
644 
645 	if (nb_desc < (min_sop + min_data)) {
646 		dev_warning(enic,
647 			    "Number of rx descs too low, adjusting to minimum\n");
648 		nb_sop_desc = min_sop;
649 		nb_data_desc = min_data;
650 	} else if (nb_desc > (max_sop + max_data)) {
651 		dev_warning(enic,
652 			    "Number of rx_descs too high, adjusting to maximum\n");
653 		nb_sop_desc = max_sop;
654 		nb_data_desc = max_data;
655 	}
656 	if (mbufs_per_pkt > 1) {
657 		dev_info(enic, "For mtu %d and mbuf size %d valid rx descriptor range is %d to %d\n",
658 			 mtu, mbuf_size, min_sop + min_data,
659 			 max_sop + max_data);
660 	}
661 	dev_info(enic, "Using %d rx descriptors (sop %d, data %d)\n",
662 		 nb_sop_desc + nb_data_desc, nb_sop_desc, nb_data_desc);
663 
664 	/* Allocate sop queue resources */
665 	rc = vnic_rq_alloc(enic->vdev, rq_sop, sop_queue_idx,
666 		nb_sop_desc, sizeof(struct rq_enet_desc));
667 	if (rc) {
668 		dev_err(enic, "error in allocation of sop rq\n");
669 		goto err_exit;
670 	}
671 	nb_sop_desc = rq_sop->ring.desc_count;
672 
673 	if (rq_data->in_use) {
674 		/* Allocate data queue resources */
675 		rc = vnic_rq_alloc(enic->vdev, rq_data, data_queue_idx,
676 				   nb_data_desc,
677 				   sizeof(struct rq_enet_desc));
678 		if (rc) {
679 			dev_err(enic, "error in allocation of data rq\n");
680 			goto err_free_rq_sop;
681 		}
682 		nb_data_desc = rq_data->ring.desc_count;
683 	}
684 	rc = vnic_cq_alloc(enic->vdev, &enic->cq[queue_idx], queue_idx,
685 			   socket_id, nb_sop_desc + nb_data_desc,
686 			   sizeof(struct cq_enet_rq_desc));
687 	if (rc) {
688 		dev_err(enic, "error in allocation of cq for rq\n");
689 		goto err_free_rq_data;
690 	}
691 
692 	/* Allocate the mbuf rings */
693 	rq_sop->mbuf_ring = (struct rte_mbuf **)
694 		rte_zmalloc_socket("rq->mbuf_ring",
695 				   sizeof(struct rte_mbuf *) * nb_sop_desc,
696 				   RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
697 	if (rq_sop->mbuf_ring == NULL)
698 		goto err_free_cq;
699 
700 	if (rq_data->in_use) {
701 		rq_data->mbuf_ring = (struct rte_mbuf **)
702 			rte_zmalloc_socket("rq->mbuf_ring",
703 				sizeof(struct rte_mbuf *) * nb_data_desc,
704 				RTE_CACHE_LINE_SIZE, rq_sop->socket_id);
705 		if (rq_data->mbuf_ring == NULL)
706 			goto err_free_sop_mbuf;
707 	}
708 
709 	rq_sop->tot_nb_desc = nb_desc; /* squirl away for MTU update function */
710 
711 	return 0;
712 
713 err_free_sop_mbuf:
714 	rte_free(rq_sop->mbuf_ring);
715 err_free_cq:
716 	/* cleanup on error */
717 	vnic_cq_free(&enic->cq[queue_idx]);
718 err_free_rq_data:
719 	if (rq_data->in_use)
720 		vnic_rq_free(rq_data);
721 err_free_rq_sop:
722 	vnic_rq_free(rq_sop);
723 err_exit:
724 	return -ENOMEM;
725 }
726 
727 void enic_free_wq(void *txq)
728 {
729 	struct vnic_wq *wq;
730 	struct enic *enic;
731 
732 	if (txq == NULL)
733 		return;
734 
735 	wq = (struct vnic_wq *)txq;
736 	enic = vnic_dev_priv(wq->vdev);
737 	rte_memzone_free(wq->cqmsg_rz);
738 	vnic_wq_free(wq);
739 	vnic_cq_free(&enic->cq[enic->rq_count + wq->index]);
740 }
741 
742 int enic_alloc_wq(struct enic *enic, uint16_t queue_idx,
743 	unsigned int socket_id, uint16_t nb_desc)
744 {
745 	int err;
746 	struct vnic_wq *wq = &enic->wq[queue_idx];
747 	unsigned int cq_index = enic_cq_wq(enic, queue_idx);
748 	char name[NAME_MAX];
749 	static int instance;
750 
751 	wq->socket_id = socket_id;
752 	if (nb_desc) {
753 		if (nb_desc > enic->config.wq_desc_count) {
754 			dev_warning(enic,
755 				"WQ %d - number of tx desc in cmd line (%d)"\
756 				"is greater than that in the UCSM/CIMC adapter"\
757 				"policy.  Applying the value in the adapter "\
758 				"policy (%d)\n",
759 				queue_idx, nb_desc, enic->config.wq_desc_count);
760 		} else if (nb_desc != enic->config.wq_desc_count) {
761 			enic->config.wq_desc_count = nb_desc;
762 			dev_info(enic,
763 				"TX Queues - effective number of descs:%d\n",
764 				nb_desc);
765 		}
766 	}
767 
768 	/* Allocate queue resources */
769 	err = vnic_wq_alloc(enic->vdev, &enic->wq[queue_idx], queue_idx,
770 		enic->config.wq_desc_count,
771 		sizeof(struct wq_enet_desc));
772 	if (err) {
773 		dev_err(enic, "error in allocation of wq\n");
774 		return err;
775 	}
776 
777 	err = vnic_cq_alloc(enic->vdev, &enic->cq[cq_index], cq_index,
778 		socket_id, enic->config.wq_desc_count,
779 		sizeof(struct cq_enet_wq_desc));
780 	if (err) {
781 		vnic_wq_free(wq);
782 		dev_err(enic, "error in allocation of cq for wq\n");
783 	}
784 
785 	/* setup up CQ message */
786 	snprintf((char *)name, sizeof(name),
787 		 "vnic_cqmsg-%s-%d-%d", enic->bdf_name, queue_idx,
788 		instance++);
789 
790 	wq->cqmsg_rz = rte_memzone_reserve_aligned((const char *)name,
791 						   sizeof(uint32_t),
792 						   SOCKET_ID_ANY, 0,
793 						   ENIC_ALIGN);
794 	if (!wq->cqmsg_rz)
795 		return -ENOMEM;
796 
797 	return err;
798 }
799 
800 int enic_disable(struct enic *enic)
801 {
802 	unsigned int i;
803 	int err;
804 
805 	vnic_intr_mask(&enic->intr);
806 	(void)vnic_intr_masked(&enic->intr); /* flush write */
807 	rte_intr_disable(&enic->pdev->intr_handle);
808 	rte_intr_callback_unregister(&enic->pdev->intr_handle,
809 				     enic_intr_handler,
810 				     (void *)enic->rte_dev);
811 
812 	vnic_dev_disable(enic->vdev);
813 
814 	enic_clsf_destroy(enic);
815 
816 	if (!enic_is_sriov_vf(enic))
817 		vnic_dev_del_addr(enic->vdev, enic->mac_addr);
818 
819 	for (i = 0; i < enic->wq_count; i++) {
820 		err = vnic_wq_disable(&enic->wq[i]);
821 		if (err)
822 			return err;
823 	}
824 	for (i = 0; i < enic_vnic_rq_count(enic); i++) {
825 		if (enic->rq[i].in_use) {
826 			err = vnic_rq_disable(&enic->rq[i]);
827 			if (err)
828 				return err;
829 		}
830 	}
831 
832 	/* If we were using interrupts, set the interrupt vector to -1
833 	 * to disable interrupts.  We are not disabling link notifcations,
834 	 * though, as we want the polling of link status to continue working.
835 	 */
836 	if (enic->rte_dev->data->dev_conf.intr_conf.lsc)
837 		vnic_dev_notify_set(enic->vdev, -1);
838 
839 	vnic_dev_set_reset_flag(enic->vdev, 1);
840 
841 	for (i = 0; i < enic->wq_count; i++)
842 		vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
843 
844 	for (i = 0; i < enic_vnic_rq_count(enic); i++)
845 		if (enic->rq[i].in_use)
846 			vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
847 	for (i = 0; i < enic->cq_count; i++)
848 		vnic_cq_clean(&enic->cq[i]);
849 	vnic_intr_clean(&enic->intr);
850 
851 	return 0;
852 }
853 
854 static int enic_dev_wait(struct vnic_dev *vdev,
855 	int (*start)(struct vnic_dev *, int),
856 	int (*finished)(struct vnic_dev *, int *),
857 	int arg)
858 {
859 	int done;
860 	int err;
861 	int i;
862 
863 	err = start(vdev, arg);
864 	if (err)
865 		return err;
866 
867 	/* Wait for func to complete...2 seconds max */
868 	for (i = 0; i < 2000; i++) {
869 		err = finished(vdev, &done);
870 		if (err)
871 			return err;
872 		if (done)
873 			return 0;
874 		usleep(1000);
875 	}
876 	return -ETIMEDOUT;
877 }
878 
879 static int enic_dev_open(struct enic *enic)
880 {
881 	int err;
882 
883 	err = enic_dev_wait(enic->vdev, vnic_dev_open,
884 		vnic_dev_open_done, 0);
885 	if (err)
886 		dev_err(enic_get_dev(enic),
887 			"vNIC device open failed, err %d\n", err);
888 
889 	return err;
890 }
891 
892 static int enic_set_rsskey(struct enic *enic)
893 {
894 	dma_addr_t rss_key_buf_pa;
895 	union vnic_rss_key *rss_key_buf_va = NULL;
896 	static union vnic_rss_key rss_key = {
897 		.key = {
898 			[0] = {.b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101}},
899 			[1] = {.b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101}},
900 			[2] = {.b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115}},
901 			[3] = {.b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108}},
902 		}
903 	};
904 	int err;
905 	u8 name[NAME_MAX];
906 
907 	snprintf((char *)name, NAME_MAX, "rss_key-%s", enic->bdf_name);
908 	rss_key_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_key),
909 		&rss_key_buf_pa, name);
910 	if (!rss_key_buf_va)
911 		return -ENOMEM;
912 
913 	rte_memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
914 
915 	err = enic_set_rss_key(enic,
916 		rss_key_buf_pa,
917 		sizeof(union vnic_rss_key));
918 
919 	enic_free_consistent(enic, sizeof(union vnic_rss_key),
920 		rss_key_buf_va, rss_key_buf_pa);
921 
922 	return err;
923 }
924 
925 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
926 {
927 	dma_addr_t rss_cpu_buf_pa;
928 	union vnic_rss_cpu *rss_cpu_buf_va = NULL;
929 	int i;
930 	int err;
931 	u8 name[NAME_MAX];
932 
933 	snprintf((char *)name, NAME_MAX, "rss_cpu-%s", enic->bdf_name);
934 	rss_cpu_buf_va = enic_alloc_consistent(enic, sizeof(union vnic_rss_cpu),
935 		&rss_cpu_buf_pa, name);
936 	if (!rss_cpu_buf_va)
937 		return -ENOMEM;
938 
939 	for (i = 0; i < (1 << rss_hash_bits); i++)
940 		(*rss_cpu_buf_va).cpu[i / 4].b[i % 4] =
941 			enic_rte_rq_idx_to_sop_idx(i % enic->rq_count);
942 
943 	err = enic_set_rss_cpu(enic,
944 		rss_cpu_buf_pa,
945 		sizeof(union vnic_rss_cpu));
946 
947 	enic_free_consistent(enic, sizeof(union vnic_rss_cpu),
948 		rss_cpu_buf_va, rss_cpu_buf_pa);
949 
950 	return err;
951 }
952 
953 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
954 	u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
955 {
956 	const u8 tso_ipid_split_en = 0;
957 	int err;
958 
959 	/* Enable VLAN tag stripping */
960 
961 	err = enic_set_nic_cfg(enic,
962 		rss_default_cpu, rss_hash_type,
963 		rss_hash_bits, rss_base_cpu,
964 		rss_enable, tso_ipid_split_en,
965 		enic->ig_vlan_strip_en);
966 
967 	return err;
968 }
969 
970 int enic_set_rss_nic_cfg(struct enic *enic)
971 {
972 	const u8 rss_default_cpu = 0;
973 	const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
974 	    NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
975 	    NIC_CFG_RSS_HASH_TYPE_IPV6 |
976 	    NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
977 	const u8 rss_hash_bits = 7;
978 	const u8 rss_base_cpu = 0;
979 	u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
980 
981 	if (rss_enable) {
982 		if (!enic_set_rsskey(enic)) {
983 			if (enic_set_rsscpu(enic, rss_hash_bits)) {
984 				rss_enable = 0;
985 				dev_warning(enic, "RSS disabled, "\
986 					"Failed to set RSS cpu indirection table.");
987 			}
988 		} else {
989 			rss_enable = 0;
990 			dev_warning(enic,
991 				"RSS disabled, Failed to set RSS key.\n");
992 		}
993 	}
994 
995 	return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
996 		rss_hash_bits, rss_base_cpu, rss_enable);
997 }
998 
999 int enic_setup_finish(struct enic *enic)
1000 {
1001 	int ret;
1002 
1003 	enic_init_soft_stats(enic);
1004 
1005 	ret = enic_set_rss_nic_cfg(enic);
1006 	if (ret) {
1007 		dev_err(enic, "Failed to config nic, aborting.\n");
1008 		return -1;
1009 	}
1010 
1011 	/* Default conf */
1012 	vnic_dev_packet_filter(enic->vdev,
1013 		1 /* directed  */,
1014 		1 /* multicast */,
1015 		1 /* broadcast */,
1016 		0 /* promisc   */,
1017 		1 /* allmulti  */);
1018 
1019 	enic->promisc = 0;
1020 	enic->allmulti = 1;
1021 
1022 	return 0;
1023 }
1024 
1025 void enic_add_packet_filter(struct enic *enic)
1026 {
1027 	/* Args -> directed, multicast, broadcast, promisc, allmulti */
1028 	vnic_dev_packet_filter(enic->vdev, 1, 1, 1,
1029 		enic->promisc, enic->allmulti);
1030 }
1031 
1032 int enic_get_link_status(struct enic *enic)
1033 {
1034 	return vnic_dev_link_status(enic->vdev);
1035 }
1036 
1037 static void enic_dev_deinit(struct enic *enic)
1038 {
1039 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1040 
1041 	/* stop link status checking */
1042 	vnic_dev_notify_unset(enic->vdev);
1043 
1044 	rte_free(eth_dev->data->mac_addrs);
1045 }
1046 
1047 
1048 int enic_set_vnic_res(struct enic *enic)
1049 {
1050 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1051 	int rc = 0;
1052 
1053 	/* With Rx scatter support, two RQs are now used per RQ used by
1054 	 * the application.
1055 	 */
1056 	if (enic->conf_rq_count < eth_dev->data->nb_rx_queues) {
1057 		dev_err(dev, "Not enough Receive queues. Requested:%u which uses %d RQs on VIC, Configured:%u\n",
1058 			eth_dev->data->nb_rx_queues,
1059 			eth_dev->data->nb_rx_queues * 2, enic->conf_rq_count);
1060 		rc = -EINVAL;
1061 	}
1062 	if (enic->conf_wq_count < eth_dev->data->nb_tx_queues) {
1063 		dev_err(dev, "Not enough Transmit queues. Requested:%u, Configured:%u\n",
1064 			eth_dev->data->nb_tx_queues, enic->conf_wq_count);
1065 		rc = -EINVAL;
1066 	}
1067 
1068 	if (enic->conf_cq_count < (eth_dev->data->nb_rx_queues +
1069 				   eth_dev->data->nb_tx_queues)) {
1070 		dev_err(dev, "Not enough Completion queues. Required:%u, Configured:%u\n",
1071 			(eth_dev->data->nb_rx_queues +
1072 			 eth_dev->data->nb_tx_queues), enic->conf_cq_count);
1073 		rc = -EINVAL;
1074 	}
1075 
1076 	if (rc == 0) {
1077 		enic->rq_count = eth_dev->data->nb_rx_queues;
1078 		enic->wq_count = eth_dev->data->nb_tx_queues;
1079 		enic->cq_count = enic->rq_count + enic->wq_count;
1080 	}
1081 
1082 	return rc;
1083 }
1084 
1085 /* Initialize the completion queue for an RQ */
1086 static int
1087 enic_reinit_rq(struct enic *enic, unsigned int rq_idx)
1088 {
1089 	struct vnic_rq *sop_rq, *data_rq;
1090 	unsigned int cq_idx;
1091 	int rc = 0;
1092 
1093 	sop_rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1094 	data_rq = &enic->rq[enic_rte_rq_idx_to_data_idx(rq_idx)];
1095 	cq_idx = rq_idx;
1096 
1097 	vnic_cq_clean(&enic->cq[cq_idx]);
1098 	vnic_cq_init(&enic->cq[cq_idx],
1099 		     0 /* flow_control_enable */,
1100 		     1 /* color_enable */,
1101 		     0 /* cq_head */,
1102 		     0 /* cq_tail */,
1103 		     1 /* cq_tail_color */,
1104 		     0 /* interrupt_enable */,
1105 		     1 /* cq_entry_enable */,
1106 		     0 /* cq_message_enable */,
1107 		     0 /* interrupt offset */,
1108 		     0 /* cq_message_addr */);
1109 
1110 
1111 	vnic_rq_init_start(sop_rq, enic_cq_rq(enic,
1112 			   enic_rte_rq_idx_to_sop_idx(rq_idx)), 0,
1113 			   sop_rq->ring.desc_count - 1, 1, 0);
1114 	if (data_rq->in_use) {
1115 		vnic_rq_init_start(data_rq,
1116 				   enic_cq_rq(enic,
1117 				   enic_rte_rq_idx_to_data_idx(rq_idx)), 0,
1118 				   data_rq->ring.desc_count - 1, 1, 0);
1119 	}
1120 
1121 	rc = enic_alloc_rx_queue_mbufs(enic, sop_rq);
1122 	if (rc)
1123 		return rc;
1124 
1125 	if (data_rq->in_use) {
1126 		rc = enic_alloc_rx_queue_mbufs(enic, data_rq);
1127 		if (rc) {
1128 			enic_rxmbuf_queue_release(enic, sop_rq);
1129 			return rc;
1130 		}
1131 	}
1132 
1133 	return 0;
1134 }
1135 
1136 /* The Cisco NIC can send and receive packets up to a max packet size
1137  * determined by the NIC type and firmware. There is also an MTU
1138  * configured into the NIC via the CIMC/UCSM management interface
1139  * which can be overridden by this function (up to the max packet size).
1140  * Depending on the network setup, doing so may cause packet drops
1141  * and unexpected behavior.
1142  */
1143 int enic_set_mtu(struct enic *enic, uint16_t new_mtu)
1144 {
1145 	unsigned int rq_idx;
1146 	struct vnic_rq *rq;
1147 	int rc = 0;
1148 	uint16_t old_mtu;	/* previous setting */
1149 	uint16_t config_mtu;	/* Value configured into NIC via CIMC/UCSM */
1150 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1151 
1152 	old_mtu = eth_dev->data->mtu;
1153 	config_mtu = enic->config.mtu;
1154 
1155 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1156 		return -E_RTE_SECONDARY;
1157 
1158 	if (new_mtu > enic->max_mtu) {
1159 		dev_err(enic,
1160 			"MTU not updated: requested (%u) greater than max (%u)\n",
1161 			new_mtu, enic->max_mtu);
1162 		return -EINVAL;
1163 	}
1164 	if (new_mtu < ENIC_MIN_MTU) {
1165 		dev_info(enic,
1166 			"MTU not updated: requested (%u) less than min (%u)\n",
1167 			new_mtu, ENIC_MIN_MTU);
1168 		return -EINVAL;
1169 	}
1170 	if (new_mtu > config_mtu)
1171 		dev_warning(enic,
1172 			"MTU (%u) is greater than value configured in NIC (%u)\n",
1173 			new_mtu, config_mtu);
1174 
1175 	/* The easy case is when scatter is disabled. However if the MTU
1176 	 * becomes greater than the mbuf data size, packet drops will ensue.
1177 	 */
1178 	if (!(enic->rte_dev->data->dev_conf.rxmode.offloads &
1179 	      DEV_RX_OFFLOAD_SCATTER)) {
1180 		eth_dev->data->mtu = new_mtu;
1181 		goto set_mtu_done;
1182 	}
1183 
1184 	/* Rx scatter is enabled so reconfigure RQ's on the fly. The point is to
1185 	 * change Rx scatter mode if necessary for better performance. I.e. if
1186 	 * MTU was greater than the mbuf size and now it's less, scatter Rx
1187 	 * doesn't have to be used and vice versa.
1188 	  */
1189 	rte_spinlock_lock(&enic->mtu_lock);
1190 
1191 	/* Stop traffic on all RQs */
1192 	for (rq_idx = 0; rq_idx < enic->rq_count * 2; rq_idx++) {
1193 		rq = &enic->rq[rq_idx];
1194 		if (rq->is_sop && rq->in_use) {
1195 			rc = enic_stop_rq(enic,
1196 					  enic_sop_rq_idx_to_rte_idx(rq_idx));
1197 			if (rc) {
1198 				dev_err(enic, "Failed to stop Rq %u\n", rq_idx);
1199 				goto set_mtu_done;
1200 			}
1201 		}
1202 	}
1203 
1204 	/* replace Rx function with a no-op to avoid getting stale pkts */
1205 	eth_dev->rx_pkt_burst = enic_dummy_recv_pkts;
1206 	rte_mb();
1207 
1208 	/* Allow time for threads to exit the real Rx function. */
1209 	usleep(100000);
1210 
1211 	/* now it is safe to reconfigure the RQs */
1212 
1213 	/* update the mtu */
1214 	eth_dev->data->mtu = new_mtu;
1215 
1216 	/* free and reallocate RQs with the new MTU */
1217 	for (rq_idx = 0; rq_idx < enic->rq_count; rq_idx++) {
1218 		rq = &enic->rq[enic_rte_rq_idx_to_sop_idx(rq_idx)];
1219 
1220 		enic_free_rq(rq);
1221 		rc = enic_alloc_rq(enic, rq_idx, rq->socket_id, rq->mp,
1222 				   rq->tot_nb_desc, rq->rx_free_thresh);
1223 		if (rc) {
1224 			dev_err(enic,
1225 				"Fatal MTU alloc error- No traffic will pass\n");
1226 			goto set_mtu_done;
1227 		}
1228 
1229 		rc = enic_reinit_rq(enic, rq_idx);
1230 		if (rc) {
1231 			dev_err(enic,
1232 				"Fatal MTU RQ reinit- No traffic will pass\n");
1233 			goto set_mtu_done;
1234 		}
1235 	}
1236 
1237 	/* put back the real receive function */
1238 	rte_mb();
1239 	eth_dev->rx_pkt_burst = enic_recv_pkts;
1240 	rte_mb();
1241 
1242 	/* restart Rx traffic */
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 		if (rq->is_sop && rq->in_use)
1246 			enic_start_rq(enic, rq_idx);
1247 	}
1248 
1249 set_mtu_done:
1250 	dev_info(enic, "MTU changed from %u to %u\n",  old_mtu, new_mtu);
1251 	rte_spinlock_unlock(&enic->mtu_lock);
1252 	return rc;
1253 }
1254 
1255 static int enic_dev_init(struct enic *enic)
1256 {
1257 	int err;
1258 	struct rte_eth_dev *eth_dev = enic->rte_dev;
1259 
1260 	vnic_dev_intr_coal_timer_info_default(enic->vdev);
1261 
1262 	/* Get vNIC configuration
1263 	*/
1264 	err = enic_get_vnic_config(enic);
1265 	if (err) {
1266 		dev_err(dev, "Get vNIC configuration failed, aborting\n");
1267 		return err;
1268 	}
1269 
1270 	/* Get available resource counts */
1271 	enic_get_res_counts(enic);
1272 	if (enic->conf_rq_count == 1) {
1273 		dev_err(enic, "Running with only 1 RQ configured in the vNIC is not supported.\n");
1274 		dev_err(enic, "Please configure 2 RQs in the vNIC for each Rx queue used by DPDK.\n");
1275 		dev_err(enic, "See the ENIC PMD guide for more information.\n");
1276 		return -EINVAL;
1277 	}
1278 
1279 	/* Get the supported filters */
1280 	enic_fdir_info(enic);
1281 
1282 	eth_dev->data->mac_addrs = rte_zmalloc("enic_mac_addr", ETH_ALEN
1283 						* ENIC_MAX_MAC_ADDR, 0);
1284 	if (!eth_dev->data->mac_addrs) {
1285 		dev_err(enic, "mac addr storage alloc failed, aborting.\n");
1286 		return -1;
1287 	}
1288 	ether_addr_copy((struct ether_addr *) enic->mac_addr,
1289 			eth_dev->data->mac_addrs);
1290 
1291 	vnic_dev_set_reset_flag(enic->vdev, 0);
1292 
1293 	LIST_INIT(&enic->flows);
1294 	rte_spinlock_init(&enic->flows_lock);
1295 
1296 	/* set up link status checking */
1297 	vnic_dev_notify_set(enic->vdev, -1); /* No Intr for notify */
1298 
1299 	return 0;
1300 
1301 }
1302 
1303 int enic_probe(struct enic *enic)
1304 {
1305 	struct rte_pci_device *pdev = enic->pdev;
1306 	int err = -1;
1307 
1308 	dev_debug(enic, " Initializing ENIC PMD\n");
1309 
1310 	/* if this is a secondary process the hardware is already initialized */
1311 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1312 		return 0;
1313 
1314 	enic->bar0.vaddr = (void *)pdev->mem_resource[0].addr;
1315 	enic->bar0.len = pdev->mem_resource[0].len;
1316 
1317 	/* Register vNIC device */
1318 	enic->vdev = vnic_dev_register(NULL, enic, enic->pdev, &enic->bar0, 1);
1319 	if (!enic->vdev) {
1320 		dev_err(enic, "vNIC registration failed, aborting\n");
1321 		goto err_out;
1322 	}
1323 
1324 	LIST_INIT(&enic->memzone_list);
1325 	rte_spinlock_init(&enic->memzone_list_lock);
1326 
1327 	vnic_register_cbacks(enic->vdev,
1328 		enic_alloc_consistent,
1329 		enic_free_consistent);
1330 
1331 	/* Issue device open to get device in known state */
1332 	err = enic_dev_open(enic);
1333 	if (err) {
1334 		dev_err(enic, "vNIC dev open failed, aborting\n");
1335 		goto err_out_unregister;
1336 	}
1337 
1338 	/* Set ingress vlan rewrite mode before vnic initialization */
1339 	err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1340 		IG_VLAN_REWRITE_MODE_PASS_THRU);
1341 	if (err) {
1342 		dev_err(enic,
1343 			"Failed to set ingress vlan rewrite mode, aborting.\n");
1344 		goto err_out_dev_close;
1345 	}
1346 
1347 	/* Issue device init to initialize the vnic-to-switch link.
1348 	 * We'll start with carrier off and wait for link UP
1349 	 * notification later to turn on carrier.  We don't need
1350 	 * to wait here for the vnic-to-switch link initialization
1351 	 * to complete; link UP notification is the indication that
1352 	 * the process is complete.
1353 	 */
1354 
1355 	err = vnic_dev_init(enic->vdev, 0);
1356 	if (err) {
1357 		dev_err(enic, "vNIC dev init failed, aborting\n");
1358 		goto err_out_dev_close;
1359 	}
1360 
1361 	err = enic_dev_init(enic);
1362 	if (err) {
1363 		dev_err(enic, "Device initialization failed, aborting\n");
1364 		goto err_out_dev_close;
1365 	}
1366 
1367 	return 0;
1368 
1369 err_out_dev_close:
1370 	vnic_dev_close(enic->vdev);
1371 err_out_unregister:
1372 	vnic_dev_unregister(enic->vdev);
1373 err_out:
1374 	return err;
1375 }
1376 
1377 void enic_remove(struct enic *enic)
1378 {
1379 	enic_dev_deinit(enic);
1380 	vnic_dev_close(enic->vdev);
1381 	vnic_dev_unregister(enic->vdev);
1382 }
1383