xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 3bb3ebb51b789d4ecb417cbdb1dce5c7211f6f18)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4 
5 #include <stdint.h>
6 #include <string.h>
7 #include <stdio.h>
8 #include <errno.h>
9 #include <unistd.h>
10 
11 #include <ethdev_driver.h>
12 #include <rte_memcpy.h>
13 #include <rte_string_fns.h>
14 #include <rte_memzone.h>
15 #include <rte_malloc.h>
16 #include <rte_branch_prediction.h>
17 #include <rte_ether.h>
18 #include <rte_ip.h>
19 #include <rte_arp.h>
20 #include <rte_common.h>
21 #include <rte_errno.h>
22 #include <rte_cpuflags.h>
23 #include <rte_vect.h>
24 #include <rte_memory.h>
25 #include <rte_eal_paging.h>
26 #include <rte_eal.h>
27 #include <rte_dev.h>
28 #include <rte_cycles.h>
29 #include <rte_kvargs.h>
30 
31 #include "virtio_ethdev.h"
32 #include "virtio.h"
33 #include "virtio_logs.h"
34 #include "virtqueue.h"
35 #include "virtio_rxtx.h"
36 #include "virtio_user/virtio_user_dev.h"
37 
38 static int  virtio_dev_configure(struct rte_eth_dev *dev);
39 static int  virtio_dev_start(struct rte_eth_dev *dev);
40 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
41 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
42 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
43 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
44 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
45 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
46 	uint32_t *speed,
47 	int *vectorized);
48 static int virtio_dev_info_get(struct rte_eth_dev *dev,
49 				struct rte_eth_dev_info *dev_info);
50 static int virtio_dev_link_update(struct rte_eth_dev *dev,
51 	int wait_to_complete);
52 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
53 
54 static void virtio_set_hwaddr(struct virtio_hw *hw);
55 static void virtio_get_hwaddr(struct virtio_hw *hw);
56 
57 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
58 				 struct rte_eth_stats *stats);
59 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
60 				 struct rte_eth_xstat *xstats, unsigned n);
61 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
62 				       struct rte_eth_xstat_name *xstats_names,
63 				       unsigned limit);
64 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
65 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
66 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
67 				uint16_t vlan_id, int on);
68 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
69 				struct rte_ether_addr *mac_addr,
70 				uint32_t index, uint32_t vmdq);
71 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
72 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
73 				struct rte_ether_addr *mac_addr);
74 
75 static int virtio_intr_disable(struct rte_eth_dev *dev);
76 
77 static int virtio_dev_queue_stats_mapping_set(
78 	struct rte_eth_dev *eth_dev,
79 	uint16_t queue_id,
80 	uint8_t stat_idx,
81 	uint8_t is_rx);
82 
83 static void virtio_notify_peers(struct rte_eth_dev *dev);
84 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
85 
86 struct rte_virtio_xstats_name_off {
87 	char name[RTE_ETH_XSTATS_NAME_SIZE];
88 	unsigned offset;
89 };
90 
91 /* [rt]x_qX_ is prepended to the name string here */
92 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
93 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
94 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
95 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
96 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
97 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
98 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
99 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
100 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
101 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
102 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
103 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
104 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
105 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
106 };
107 
108 /* [rt]x_qX_ is prepended to the name string here */
109 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
110 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
111 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
112 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
113 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
114 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
115 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
116 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
117 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
118 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
119 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
120 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
121 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
122 };
123 
124 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
125 			    sizeof(rte_virtio_rxq_stat_strings[0]))
126 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
127 			    sizeof(rte_virtio_txq_stat_strings[0]))
128 
129 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
130 
131 static struct virtio_pmd_ctrl *
132 virtio_send_command_packed(struct virtnet_ctl *cvq,
133 			   struct virtio_pmd_ctrl *ctrl,
134 			   int *dlen, int pkt_num)
135 {
136 	struct virtqueue *vq = cvq->vq;
137 	int head;
138 	struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
139 	struct virtio_pmd_ctrl *result;
140 	uint16_t flags;
141 	int sum = 0;
142 	int nb_descs = 0;
143 	int k;
144 
145 	/*
146 	 * Format is enforced in qemu code:
147 	 * One TX packet for header;
148 	 * At least one TX packet per argument;
149 	 * One RX packet for ACK.
150 	 */
151 	head = vq->vq_avail_idx;
152 	flags = vq->vq_packed.cached_flags;
153 	desc[head].addr = cvq->virtio_net_hdr_mem;
154 	desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
155 	vq->vq_free_cnt--;
156 	nb_descs++;
157 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
158 		vq->vq_avail_idx -= vq->vq_nentries;
159 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
160 	}
161 
162 	for (k = 0; k < pkt_num; k++) {
163 		desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
164 			+ sizeof(struct virtio_net_ctrl_hdr)
165 			+ sizeof(ctrl->status) + sizeof(uint8_t) * sum;
166 		desc[vq->vq_avail_idx].len = dlen[k];
167 		desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
168 			vq->vq_packed.cached_flags;
169 		sum += dlen[k];
170 		vq->vq_free_cnt--;
171 		nb_descs++;
172 		if (++vq->vq_avail_idx >= vq->vq_nentries) {
173 			vq->vq_avail_idx -= vq->vq_nentries;
174 			vq->vq_packed.cached_flags ^=
175 				VRING_PACKED_DESC_F_AVAIL_USED;
176 		}
177 	}
178 
179 	desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
180 		+ sizeof(struct virtio_net_ctrl_hdr);
181 	desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
182 	desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
183 		vq->vq_packed.cached_flags;
184 	vq->vq_free_cnt--;
185 	nb_descs++;
186 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
187 		vq->vq_avail_idx -= vq->vq_nentries;
188 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
189 	}
190 
191 	virtqueue_store_flags_packed(&desc[head], VRING_DESC_F_NEXT | flags,
192 			vq->hw->weak_barriers);
193 
194 	virtio_wmb(vq->hw->weak_barriers);
195 	virtqueue_notify(vq);
196 
197 	/* wait for used desc in virtqueue
198 	 * desc_is_used has a load-acquire or rte_io_rmb inside
199 	 */
200 	while (!desc_is_used(&desc[head], vq))
201 		usleep(100);
202 
203 	/* now get used descriptors */
204 	vq->vq_free_cnt += nb_descs;
205 	vq->vq_used_cons_idx += nb_descs;
206 	if (vq->vq_used_cons_idx >= vq->vq_nentries) {
207 		vq->vq_used_cons_idx -= vq->vq_nentries;
208 		vq->vq_packed.used_wrap_counter ^= 1;
209 	}
210 
211 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
212 			"vq->vq_avail_idx=%d\n"
213 			"vq->vq_used_cons_idx=%d\n"
214 			"vq->vq_packed.cached_flags=0x%x\n"
215 			"vq->vq_packed.used_wrap_counter=%d\n",
216 			vq->vq_free_cnt,
217 			vq->vq_avail_idx,
218 			vq->vq_used_cons_idx,
219 			vq->vq_packed.cached_flags,
220 			vq->vq_packed.used_wrap_counter);
221 
222 	result = cvq->virtio_net_hdr_mz->addr;
223 	return result;
224 }
225 
226 static struct virtio_pmd_ctrl *
227 virtio_send_command_split(struct virtnet_ctl *cvq,
228 			  struct virtio_pmd_ctrl *ctrl,
229 			  int *dlen, int pkt_num)
230 {
231 	struct virtio_pmd_ctrl *result;
232 	struct virtqueue *vq = cvq->vq;
233 	uint32_t head, i;
234 	int k, sum = 0;
235 
236 	head = vq->vq_desc_head_idx;
237 
238 	/*
239 	 * Format is enforced in qemu code:
240 	 * One TX packet for header;
241 	 * At least one TX packet per argument;
242 	 * One RX packet for ACK.
243 	 */
244 	vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
245 	vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
246 	vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
247 	vq->vq_free_cnt--;
248 	i = vq->vq_split.ring.desc[head].next;
249 
250 	for (k = 0; k < pkt_num; k++) {
251 		vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
252 		vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
253 			+ sizeof(struct virtio_net_ctrl_hdr)
254 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
255 		vq->vq_split.ring.desc[i].len = dlen[k];
256 		sum += dlen[k];
257 		vq->vq_free_cnt--;
258 		i = vq->vq_split.ring.desc[i].next;
259 	}
260 
261 	vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
262 	vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
263 			+ sizeof(struct virtio_net_ctrl_hdr);
264 	vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
265 	vq->vq_free_cnt--;
266 
267 	vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
268 
269 	vq_update_avail_ring(vq, head);
270 	vq_update_avail_idx(vq);
271 
272 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
273 
274 	virtqueue_notify(vq);
275 
276 	while (virtqueue_nused(vq) == 0)
277 		usleep(100);
278 
279 	while (virtqueue_nused(vq)) {
280 		uint32_t idx, desc_idx, used_idx;
281 		struct vring_used_elem *uep;
282 
283 		used_idx = (uint32_t)(vq->vq_used_cons_idx
284 				& (vq->vq_nentries - 1));
285 		uep = &vq->vq_split.ring.used->ring[used_idx];
286 		idx = (uint32_t) uep->id;
287 		desc_idx = idx;
288 
289 		while (vq->vq_split.ring.desc[desc_idx].flags &
290 				VRING_DESC_F_NEXT) {
291 			desc_idx = vq->vq_split.ring.desc[desc_idx].next;
292 			vq->vq_free_cnt++;
293 		}
294 
295 		vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
296 		vq->vq_desc_head_idx = idx;
297 
298 		vq->vq_used_cons_idx++;
299 		vq->vq_free_cnt++;
300 	}
301 
302 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
303 			vq->vq_free_cnt, vq->vq_desc_head_idx);
304 
305 	result = cvq->virtio_net_hdr_mz->addr;
306 	return result;
307 }
308 
309 static int
310 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
311 		    int *dlen, int pkt_num)
312 {
313 	virtio_net_ctrl_ack status = ~0;
314 	struct virtio_pmd_ctrl *result;
315 	struct virtqueue *vq;
316 
317 	ctrl->status = status;
318 
319 	if (!cvq || !cvq->vq) {
320 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
321 		return -1;
322 	}
323 
324 	rte_spinlock_lock(&cvq->lock);
325 	vq = cvq->vq;
326 
327 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
328 		"vq->hw->cvq = %p vq = %p",
329 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
330 
331 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
332 		rte_spinlock_unlock(&cvq->lock);
333 		return -1;
334 	}
335 
336 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
337 		sizeof(struct virtio_pmd_ctrl));
338 
339 	if (virtio_with_packed_queue(vq->hw))
340 		result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
341 	else
342 		result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
343 
344 	rte_spinlock_unlock(&cvq->lock);
345 	return result->status;
346 }
347 
348 static int
349 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
350 {
351 	struct virtio_hw *hw = dev->data->dev_private;
352 	struct virtio_pmd_ctrl ctrl;
353 	int dlen[1];
354 	int ret;
355 
356 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
357 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
358 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
359 
360 	dlen[0] = sizeof(uint16_t);
361 
362 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
363 	if (ret) {
364 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
365 			  "failed, this is too late now...");
366 		return -EINVAL;
367 	}
368 
369 	return 0;
370 }
371 
372 static void
373 virtio_dev_queue_release(void *queue __rte_unused)
374 {
375 	/* do nothing */
376 }
377 
378 static uint16_t
379 virtio_get_nr_vq(struct virtio_hw *hw)
380 {
381 	uint16_t nr_vq = hw->max_queue_pairs * 2;
382 
383 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
384 		nr_vq += 1;
385 
386 	return nr_vq;
387 }
388 
389 static void
390 virtio_init_vring(struct virtqueue *vq)
391 {
392 	int size = vq->vq_nentries;
393 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
394 
395 	PMD_INIT_FUNC_TRACE();
396 
397 	memset(ring_mem, 0, vq->vq_ring_size);
398 
399 	vq->vq_used_cons_idx = 0;
400 	vq->vq_desc_head_idx = 0;
401 	vq->vq_avail_idx = 0;
402 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
403 	vq->vq_free_cnt = vq->vq_nentries;
404 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
405 	if (virtio_with_packed_queue(vq->hw)) {
406 		vring_init_packed(&vq->vq_packed.ring, ring_mem,
407 				  VIRTIO_VRING_ALIGN, size);
408 		vring_desc_init_packed(vq, size);
409 	} else {
410 		struct vring *vr = &vq->vq_split.ring;
411 
412 		vring_init_split(vr, ring_mem, VIRTIO_VRING_ALIGN, size);
413 		vring_desc_init_split(vr->desc, size);
414 	}
415 	/*
416 	 * Disable device(host) interrupting guest
417 	 */
418 	virtqueue_disable_intr(vq);
419 }
420 
421 static int
422 virtio_init_queue(struct rte_eth_dev *dev, uint16_t queue_idx)
423 {
424 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
425 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
426 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
427 	unsigned int vq_size, size;
428 	struct virtio_hw *hw = dev->data->dev_private;
429 	struct virtnet_rx *rxvq = NULL;
430 	struct virtnet_tx *txvq = NULL;
431 	struct virtnet_ctl *cvq = NULL;
432 	struct virtqueue *vq;
433 	size_t sz_hdr_mz = 0;
434 	void *sw_ring = NULL;
435 	int queue_type = virtio_get_queue_type(hw, queue_idx);
436 	int ret;
437 	int numa_node = dev->device->numa_node;
438 
439 	PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
440 			queue_idx, numa_node);
441 
442 	/*
443 	 * Read the virtqueue size from the Queue Size field
444 	 * Always power of 2 and if 0 virtqueue does not exist
445 	 */
446 	vq_size = VIRTIO_OPS(hw)->get_queue_num(hw, queue_idx);
447 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
448 	if (vq_size == 0) {
449 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
450 		return -EINVAL;
451 	}
452 
453 	if (!virtio_with_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
454 		PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
455 		return -EINVAL;
456 	}
457 
458 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
459 		 dev->data->port_id, queue_idx);
460 
461 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
462 				vq_size * sizeof(struct vq_desc_extra),
463 				RTE_CACHE_LINE_SIZE);
464 	if (queue_type == VTNET_TQ) {
465 		/*
466 		 * For each xmit packet, allocate a virtio_net_hdr
467 		 * and indirect ring elements
468 		 */
469 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
470 	} else if (queue_type == VTNET_CQ) {
471 		/* Allocate a page for control vq command, data and status */
472 		sz_hdr_mz = rte_mem_page_size();
473 	}
474 
475 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
476 				numa_node);
477 	if (vq == NULL) {
478 		PMD_INIT_LOG(ERR, "can not allocate vq");
479 		return -ENOMEM;
480 	}
481 	hw->vqs[queue_idx] = vq;
482 
483 	vq->hw = hw;
484 	vq->vq_queue_index = queue_idx;
485 	vq->vq_nentries = vq_size;
486 	if (virtio_with_packed_queue(hw)) {
487 		vq->vq_packed.used_wrap_counter = 1;
488 		vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
489 		vq->vq_packed.event_flags_shadow = 0;
490 		if (queue_type == VTNET_RQ)
491 			vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
492 	}
493 
494 	/*
495 	 * Reserve a memzone for vring elements
496 	 */
497 	size = vring_size(hw, vq_size, VIRTIO_VRING_ALIGN);
498 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_VRING_ALIGN);
499 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
500 		     size, vq->vq_ring_size);
501 
502 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
503 			numa_node, RTE_MEMZONE_IOVA_CONTIG,
504 			VIRTIO_VRING_ALIGN);
505 	if (mz == NULL) {
506 		if (rte_errno == EEXIST)
507 			mz = rte_memzone_lookup(vq_name);
508 		if (mz == NULL) {
509 			ret = -ENOMEM;
510 			goto fail_q_alloc;
511 		}
512 	}
513 
514 	memset(mz->addr, 0, mz->len);
515 
516 	vq->vq_ring_mem = mz->iova;
517 	vq->vq_ring_virt_mem = mz->addr;
518 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
519 		     (uint64_t)mz->iova);
520 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
521 		     (uint64_t)(uintptr_t)mz->addr);
522 
523 	virtio_init_vring(vq);
524 
525 	if (sz_hdr_mz) {
526 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
527 			 dev->data->port_id, queue_idx);
528 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
529 				numa_node, RTE_MEMZONE_IOVA_CONTIG,
530 				RTE_CACHE_LINE_SIZE);
531 		if (hdr_mz == NULL) {
532 			if (rte_errno == EEXIST)
533 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
534 			if (hdr_mz == NULL) {
535 				ret = -ENOMEM;
536 				goto fail_q_alloc;
537 			}
538 		}
539 	}
540 
541 	if (queue_type == VTNET_RQ) {
542 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
543 			       sizeof(vq->sw_ring[0]);
544 
545 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
546 				RTE_CACHE_LINE_SIZE, numa_node);
547 		if (!sw_ring) {
548 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
549 			ret = -ENOMEM;
550 			goto fail_q_alloc;
551 		}
552 
553 		vq->sw_ring = sw_ring;
554 		rxvq = &vq->rxq;
555 		rxvq->vq = vq;
556 		rxvq->port_id = dev->data->port_id;
557 		rxvq->mz = mz;
558 	} else if (queue_type == VTNET_TQ) {
559 		txvq = &vq->txq;
560 		txvq->vq = vq;
561 		txvq->port_id = dev->data->port_id;
562 		txvq->mz = mz;
563 		txvq->virtio_net_hdr_mz = hdr_mz;
564 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
565 	} else if (queue_type == VTNET_CQ) {
566 		cvq = &vq->cq;
567 		cvq->vq = vq;
568 		cvq->mz = mz;
569 		cvq->virtio_net_hdr_mz = hdr_mz;
570 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
571 		memset(cvq->virtio_net_hdr_mz->addr, 0, rte_mem_page_size());
572 
573 		hw->cvq = cvq;
574 	}
575 
576 	if (queue_type == VTNET_TQ) {
577 		struct virtio_tx_region *txr;
578 		unsigned int i;
579 
580 		txr = hdr_mz->addr;
581 		memset(txr, 0, vq_size * sizeof(*txr));
582 		for (i = 0; i < vq_size; i++) {
583 			/* first indirect descriptor is always the tx header */
584 			if (!virtio_with_packed_queue(hw)) {
585 				struct vring_desc *start_dp = txr[i].tx_indir;
586 				vring_desc_init_split(start_dp,
587 						      RTE_DIM(txr[i].tx_indir));
588 				start_dp->addr = txvq->virtio_net_hdr_mem
589 					+ i * sizeof(*txr)
590 					+ offsetof(struct virtio_tx_region,
591 						   tx_hdr);
592 				start_dp->len = hw->vtnet_hdr_size;
593 				start_dp->flags = VRING_DESC_F_NEXT;
594 			} else {
595 				struct vring_packed_desc *start_dp =
596 					txr[i].tx_packed_indir;
597 				vring_desc_init_indirect_packed(start_dp,
598 				      RTE_DIM(txr[i].tx_packed_indir));
599 				start_dp->addr = txvq->virtio_net_hdr_mem
600 					+ i * sizeof(*txr)
601 					+ offsetof(struct virtio_tx_region,
602 						   tx_hdr);
603 				start_dp->len = hw->vtnet_hdr_size;
604 			}
605 		}
606 	}
607 
608 	if (VIRTIO_OPS(hw)->setup_queue(hw, vq) < 0) {
609 		PMD_INIT_LOG(ERR, "setup_queue failed");
610 		return -EINVAL;
611 	}
612 
613 	return 0;
614 
615 fail_q_alloc:
616 	rte_free(sw_ring);
617 	rte_memzone_free(hdr_mz);
618 	rte_memzone_free(mz);
619 	rte_free(vq);
620 
621 	return ret;
622 }
623 
624 static void
625 virtio_free_queues(struct virtio_hw *hw)
626 {
627 	uint16_t nr_vq = virtio_get_nr_vq(hw);
628 	struct virtqueue *vq;
629 	int queue_type;
630 	uint16_t i;
631 
632 	if (hw->vqs == NULL)
633 		return;
634 
635 	for (i = 0; i < nr_vq; i++) {
636 		vq = hw->vqs[i];
637 		if (!vq)
638 			continue;
639 
640 		queue_type = virtio_get_queue_type(hw, i);
641 		if (queue_type == VTNET_RQ) {
642 			rte_free(vq->sw_ring);
643 			rte_memzone_free(vq->rxq.mz);
644 		} else if (queue_type == VTNET_TQ) {
645 			rte_memzone_free(vq->txq.mz);
646 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
647 		} else {
648 			rte_memzone_free(vq->cq.mz);
649 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
650 		}
651 
652 		rte_free(vq);
653 		hw->vqs[i] = NULL;
654 	}
655 
656 	rte_free(hw->vqs);
657 	hw->vqs = NULL;
658 }
659 
660 static int
661 virtio_alloc_queues(struct rte_eth_dev *dev)
662 {
663 	struct virtio_hw *hw = dev->data->dev_private;
664 	uint16_t nr_vq = virtio_get_nr_vq(hw);
665 	uint16_t i;
666 	int ret;
667 
668 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
669 	if (!hw->vqs) {
670 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
671 		return -ENOMEM;
672 	}
673 
674 	for (i = 0; i < nr_vq; i++) {
675 		ret = virtio_init_queue(dev, i);
676 		if (ret < 0) {
677 			virtio_free_queues(hw);
678 			return ret;
679 		}
680 	}
681 
682 	return 0;
683 }
684 
685 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
686 
687 int
688 virtio_dev_close(struct rte_eth_dev *dev)
689 {
690 	struct virtio_hw *hw = dev->data->dev_private;
691 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
692 
693 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
694 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
695 		return 0;
696 
697 	if (!hw->opened)
698 		return 0;
699 	hw->opened = 0;
700 
701 	/* reset the NIC */
702 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
703 		VIRTIO_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
704 	if (intr_conf->rxq)
705 		virtio_queues_unbind_intr(dev);
706 
707 	if (intr_conf->lsc || intr_conf->rxq) {
708 		virtio_intr_disable(dev);
709 		rte_intr_efd_disable(dev->intr_handle);
710 		rte_free(dev->intr_handle->intr_vec);
711 		dev->intr_handle->intr_vec = NULL;
712 	}
713 
714 	virtio_reset(hw);
715 	virtio_dev_free_mbufs(dev);
716 	virtio_free_queues(hw);
717 
718 	return VIRTIO_OPS(hw)->dev_close(hw);
719 }
720 
721 static int
722 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
723 {
724 	struct virtio_hw *hw = dev->data->dev_private;
725 	struct virtio_pmd_ctrl ctrl;
726 	int dlen[1];
727 	int ret;
728 
729 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
730 		PMD_INIT_LOG(INFO, "host does not support rx control");
731 		return -ENOTSUP;
732 	}
733 
734 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
735 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
736 	ctrl.data[0] = 1;
737 	dlen[0] = 1;
738 
739 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
740 	if (ret) {
741 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
742 		return -EAGAIN;
743 	}
744 
745 	return 0;
746 }
747 
748 static int
749 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
750 {
751 	struct virtio_hw *hw = dev->data->dev_private;
752 	struct virtio_pmd_ctrl ctrl;
753 	int dlen[1];
754 	int ret;
755 
756 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
757 		PMD_INIT_LOG(INFO, "host does not support rx control");
758 		return -ENOTSUP;
759 	}
760 
761 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
762 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
763 	ctrl.data[0] = 0;
764 	dlen[0] = 1;
765 
766 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
767 	if (ret) {
768 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
769 		return -EAGAIN;
770 	}
771 
772 	return 0;
773 }
774 
775 static int
776 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
777 {
778 	struct virtio_hw *hw = dev->data->dev_private;
779 	struct virtio_pmd_ctrl ctrl;
780 	int dlen[1];
781 	int ret;
782 
783 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
784 		PMD_INIT_LOG(INFO, "host does not support rx control");
785 		return -ENOTSUP;
786 	}
787 
788 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
789 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
790 	ctrl.data[0] = 1;
791 	dlen[0] = 1;
792 
793 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
794 	if (ret) {
795 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
796 		return -EAGAIN;
797 	}
798 
799 	return 0;
800 }
801 
802 static int
803 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
804 {
805 	struct virtio_hw *hw = dev->data->dev_private;
806 	struct virtio_pmd_ctrl ctrl;
807 	int dlen[1];
808 	int ret;
809 
810 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
811 		PMD_INIT_LOG(INFO, "host does not support rx control");
812 		return -ENOTSUP;
813 	}
814 
815 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
816 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
817 	ctrl.data[0] = 0;
818 	dlen[0] = 1;
819 
820 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
821 	if (ret) {
822 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
823 		return -EAGAIN;
824 	}
825 
826 	return 0;
827 }
828 
829 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
830 static int
831 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
832 {
833 	struct virtio_hw *hw = dev->data->dev_private;
834 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
835 				 hw->vtnet_hdr_size;
836 	uint32_t frame_size = mtu + ether_hdr_len;
837 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
838 
839 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
840 
841 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
842 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
843 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
844 		return -EINVAL;
845 	}
846 	return 0;
847 }
848 
849 static int
850 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
851 {
852 	struct virtio_hw *hw = dev->data->dev_private;
853 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
854 	struct virtqueue *vq = rxvq->vq;
855 
856 	virtqueue_enable_intr(vq);
857 	virtio_mb(hw->weak_barriers);
858 	return 0;
859 }
860 
861 static int
862 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
863 {
864 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
865 	struct virtqueue *vq = rxvq->vq;
866 
867 	virtqueue_disable_intr(vq);
868 	return 0;
869 }
870 
871 /*
872  * dev_ops for virtio, bare necessities for basic operation
873  */
874 static const struct eth_dev_ops virtio_eth_dev_ops = {
875 	.dev_configure           = virtio_dev_configure,
876 	.dev_start               = virtio_dev_start,
877 	.dev_stop                = virtio_dev_stop,
878 	.dev_close               = virtio_dev_close,
879 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
880 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
881 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
882 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
883 	.mtu_set                 = virtio_mtu_set,
884 	.dev_infos_get           = virtio_dev_info_get,
885 	.stats_get               = virtio_dev_stats_get,
886 	.xstats_get              = virtio_dev_xstats_get,
887 	.xstats_get_names        = virtio_dev_xstats_get_names,
888 	.stats_reset             = virtio_dev_stats_reset,
889 	.xstats_reset            = virtio_dev_stats_reset,
890 	.link_update             = virtio_dev_link_update,
891 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
892 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
893 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
894 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
895 	.rx_queue_release        = virtio_dev_queue_release,
896 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
897 	.tx_queue_release        = virtio_dev_queue_release,
898 	/* collect stats per queue */
899 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
900 	.vlan_filter_set         = virtio_vlan_filter_set,
901 	.mac_addr_add            = virtio_mac_addr_add,
902 	.mac_addr_remove         = virtio_mac_addr_remove,
903 	.mac_addr_set            = virtio_mac_addr_set,
904 };
905 
906 /*
907  * dev_ops for virtio-user in secondary processes, as we just have
908  * some limited supports currently.
909  */
910 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
911 	.dev_infos_get           = virtio_dev_info_get,
912 	.stats_get               = virtio_dev_stats_get,
913 	.xstats_get              = virtio_dev_xstats_get,
914 	.xstats_get_names        = virtio_dev_xstats_get_names,
915 	.stats_reset             = virtio_dev_stats_reset,
916 	.xstats_reset            = virtio_dev_stats_reset,
917 	/* collect stats per queue */
918 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
919 };
920 
921 static void
922 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
923 {
924 	unsigned i;
925 
926 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
927 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
928 		if (txvq == NULL)
929 			continue;
930 
931 		stats->opackets += txvq->stats.packets;
932 		stats->obytes += txvq->stats.bytes;
933 
934 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
935 			stats->q_opackets[i] = txvq->stats.packets;
936 			stats->q_obytes[i] = txvq->stats.bytes;
937 		}
938 	}
939 
940 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
941 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
942 		if (rxvq == NULL)
943 			continue;
944 
945 		stats->ipackets += rxvq->stats.packets;
946 		stats->ibytes += rxvq->stats.bytes;
947 		stats->ierrors += rxvq->stats.errors;
948 
949 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
950 			stats->q_ipackets[i] = rxvq->stats.packets;
951 			stats->q_ibytes[i] = rxvq->stats.bytes;
952 		}
953 	}
954 
955 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
956 }
957 
958 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
959 				       struct rte_eth_xstat_name *xstats_names,
960 				       __rte_unused unsigned limit)
961 {
962 	unsigned i;
963 	unsigned count = 0;
964 	unsigned t;
965 
966 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
967 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
968 
969 	if (xstats_names != NULL) {
970 		/* Note: limit checked in rte_eth_xstats_names() */
971 
972 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
973 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
974 			if (rxvq == NULL)
975 				continue;
976 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
977 				snprintf(xstats_names[count].name,
978 					sizeof(xstats_names[count].name),
979 					"rx_q%u_%s", i,
980 					rte_virtio_rxq_stat_strings[t].name);
981 				count++;
982 			}
983 		}
984 
985 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
986 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
987 			if (txvq == NULL)
988 				continue;
989 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
990 				snprintf(xstats_names[count].name,
991 					sizeof(xstats_names[count].name),
992 					"tx_q%u_%s", i,
993 					rte_virtio_txq_stat_strings[t].name);
994 				count++;
995 			}
996 		}
997 		return count;
998 	}
999 	return nstats;
1000 }
1001 
1002 static int
1003 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1004 		      unsigned n)
1005 {
1006 	unsigned i;
1007 	unsigned count = 0;
1008 
1009 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1010 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1011 
1012 	if (n < nstats)
1013 		return nstats;
1014 
1015 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1016 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1017 
1018 		if (rxvq == NULL)
1019 			continue;
1020 
1021 		unsigned t;
1022 
1023 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1024 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1025 				rte_virtio_rxq_stat_strings[t].offset);
1026 			xstats[count].id = count;
1027 			count++;
1028 		}
1029 	}
1030 
1031 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1032 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1033 
1034 		if (txvq == NULL)
1035 			continue;
1036 
1037 		unsigned t;
1038 
1039 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1040 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1041 				rte_virtio_txq_stat_strings[t].offset);
1042 			xstats[count].id = count;
1043 			count++;
1044 		}
1045 	}
1046 
1047 	return count;
1048 }
1049 
1050 static int
1051 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1052 {
1053 	virtio_update_stats(dev, stats);
1054 
1055 	return 0;
1056 }
1057 
1058 static int
1059 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1060 {
1061 	unsigned int i;
1062 
1063 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1064 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1065 		if (txvq == NULL)
1066 			continue;
1067 
1068 		txvq->stats.packets = 0;
1069 		txvq->stats.bytes = 0;
1070 		txvq->stats.multicast = 0;
1071 		txvq->stats.broadcast = 0;
1072 		memset(txvq->stats.size_bins, 0,
1073 		       sizeof(txvq->stats.size_bins[0]) * 8);
1074 	}
1075 
1076 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1077 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1078 		if (rxvq == NULL)
1079 			continue;
1080 
1081 		rxvq->stats.packets = 0;
1082 		rxvq->stats.bytes = 0;
1083 		rxvq->stats.errors = 0;
1084 		rxvq->stats.multicast = 0;
1085 		rxvq->stats.broadcast = 0;
1086 		memset(rxvq->stats.size_bins, 0,
1087 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1088 	}
1089 
1090 	return 0;
1091 }
1092 
1093 static void
1094 virtio_set_hwaddr(struct virtio_hw *hw)
1095 {
1096 	virtio_write_dev_config(hw,
1097 			offsetof(struct virtio_net_config, mac),
1098 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1099 }
1100 
1101 static void
1102 virtio_get_hwaddr(struct virtio_hw *hw)
1103 {
1104 	if (virtio_with_feature(hw, VIRTIO_NET_F_MAC)) {
1105 		virtio_read_dev_config(hw,
1106 			offsetof(struct virtio_net_config, mac),
1107 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1108 	} else {
1109 		rte_eth_random_addr(&hw->mac_addr[0]);
1110 		virtio_set_hwaddr(hw);
1111 	}
1112 }
1113 
1114 static int
1115 virtio_mac_table_set(struct virtio_hw *hw,
1116 		     const struct virtio_net_ctrl_mac *uc,
1117 		     const struct virtio_net_ctrl_mac *mc)
1118 {
1119 	struct virtio_pmd_ctrl ctrl;
1120 	int err, len[2];
1121 
1122 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1123 		PMD_DRV_LOG(INFO, "host does not support mac table");
1124 		return -1;
1125 	}
1126 
1127 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1128 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1129 
1130 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1131 	memcpy(ctrl.data, uc, len[0]);
1132 
1133 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1134 	memcpy(ctrl.data + len[0], mc, len[1]);
1135 
1136 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1137 	if (err != 0)
1138 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1139 	return err;
1140 }
1141 
1142 static int
1143 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1144 		    uint32_t index, uint32_t vmdq __rte_unused)
1145 {
1146 	struct virtio_hw *hw = dev->data->dev_private;
1147 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1148 	unsigned int i;
1149 	struct virtio_net_ctrl_mac *uc, *mc;
1150 
1151 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1152 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1153 		return -EINVAL;
1154 	}
1155 
1156 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1157 		sizeof(uc->entries));
1158 	uc->entries = 0;
1159 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1160 		sizeof(mc->entries));
1161 	mc->entries = 0;
1162 
1163 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1164 		const struct rte_ether_addr *addr
1165 			= (i == index) ? mac_addr : addrs + i;
1166 		struct virtio_net_ctrl_mac *tbl
1167 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1168 
1169 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1170 	}
1171 
1172 	return virtio_mac_table_set(hw, uc, mc);
1173 }
1174 
1175 static void
1176 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1177 {
1178 	struct virtio_hw *hw = dev->data->dev_private;
1179 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1180 	struct virtio_net_ctrl_mac *uc, *mc;
1181 	unsigned int i;
1182 
1183 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1184 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1185 		return;
1186 	}
1187 
1188 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1189 		sizeof(uc->entries));
1190 	uc->entries = 0;
1191 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1192 		sizeof(mc->entries));
1193 	mc->entries = 0;
1194 
1195 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1196 		struct virtio_net_ctrl_mac *tbl;
1197 
1198 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1199 			continue;
1200 
1201 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1202 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1203 			RTE_ETHER_ADDR_LEN);
1204 	}
1205 
1206 	virtio_mac_table_set(hw, uc, mc);
1207 }
1208 
1209 static int
1210 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1211 {
1212 	struct virtio_hw *hw = dev->data->dev_private;
1213 
1214 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1215 
1216 	/* Use atomic update if available */
1217 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1218 		struct virtio_pmd_ctrl ctrl;
1219 		int len = RTE_ETHER_ADDR_LEN;
1220 
1221 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1222 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1223 
1224 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1225 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1226 	}
1227 
1228 	if (!virtio_with_feature(hw, VIRTIO_NET_F_MAC))
1229 		return -ENOTSUP;
1230 
1231 	virtio_set_hwaddr(hw);
1232 	return 0;
1233 }
1234 
1235 static int
1236 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1237 {
1238 	struct virtio_hw *hw = dev->data->dev_private;
1239 	struct virtio_pmd_ctrl ctrl;
1240 	int len;
1241 
1242 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1243 		return -ENOTSUP;
1244 
1245 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1246 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1247 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1248 	len = sizeof(vlan_id);
1249 
1250 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1251 }
1252 
1253 static int
1254 virtio_intr_unmask(struct rte_eth_dev *dev)
1255 {
1256 	struct virtio_hw *hw = dev->data->dev_private;
1257 
1258 	if (rte_intr_ack(dev->intr_handle) < 0)
1259 		return -1;
1260 
1261 	if (VIRTIO_OPS(hw)->intr_detect)
1262 		VIRTIO_OPS(hw)->intr_detect(hw);
1263 
1264 	return 0;
1265 }
1266 
1267 static int
1268 virtio_intr_enable(struct rte_eth_dev *dev)
1269 {
1270 	struct virtio_hw *hw = dev->data->dev_private;
1271 
1272 	if (rte_intr_enable(dev->intr_handle) < 0)
1273 		return -1;
1274 
1275 	if (VIRTIO_OPS(hw)->intr_detect)
1276 		VIRTIO_OPS(hw)->intr_detect(hw);
1277 
1278 	return 0;
1279 }
1280 
1281 static int
1282 virtio_intr_disable(struct rte_eth_dev *dev)
1283 {
1284 	struct virtio_hw *hw = dev->data->dev_private;
1285 
1286 	if (rte_intr_disable(dev->intr_handle) < 0)
1287 		return -1;
1288 
1289 	if (VIRTIO_OPS(hw)->intr_detect)
1290 		VIRTIO_OPS(hw)->intr_detect(hw);
1291 
1292 	return 0;
1293 }
1294 
1295 static int
1296 virtio_ethdev_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1297 {
1298 	uint64_t host_features;
1299 
1300 	/* Prepare guest_features: feature that driver wants to support */
1301 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1302 		req_features);
1303 
1304 	/* Read device(host) feature bits */
1305 	host_features = VIRTIO_OPS(hw)->get_features(hw);
1306 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1307 		host_features);
1308 
1309 	/* If supported, ensure MTU value is valid before acknowledging it. */
1310 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1311 		struct virtio_net_config config;
1312 
1313 		virtio_read_dev_config(hw,
1314 			offsetof(struct virtio_net_config, mtu),
1315 			&config.mtu, sizeof(config.mtu));
1316 
1317 		if (config.mtu < RTE_ETHER_MIN_MTU)
1318 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1319 	}
1320 
1321 	/*
1322 	 * Negotiate features: Subset of device feature bits are written back
1323 	 * guest feature bits.
1324 	 */
1325 	hw->guest_features = req_features;
1326 	hw->guest_features = virtio_negotiate_features(hw, host_features);
1327 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1328 		hw->guest_features);
1329 
1330 	if (VIRTIO_OPS(hw)->features_ok(hw) < 0)
1331 		return -1;
1332 
1333 	if (virtio_with_feature(hw, VIRTIO_F_VERSION_1)) {
1334 		virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1335 
1336 		if (!(virtio_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1337 			PMD_INIT_LOG(ERR, "Failed to set FEATURES_OK status!");
1338 			return -1;
1339 		}
1340 	}
1341 
1342 	hw->req_guest_features = req_features;
1343 
1344 	return 0;
1345 }
1346 
1347 int
1348 virtio_dev_pause(struct rte_eth_dev *dev)
1349 {
1350 	struct virtio_hw *hw = dev->data->dev_private;
1351 
1352 	rte_spinlock_lock(&hw->state_lock);
1353 
1354 	if (hw->started == 0) {
1355 		/* Device is just stopped. */
1356 		rte_spinlock_unlock(&hw->state_lock);
1357 		return -1;
1358 	}
1359 	hw->started = 0;
1360 	/*
1361 	 * Prevent the worker threads from touching queues to avoid contention,
1362 	 * 1 ms should be enough for the ongoing Tx function to finish.
1363 	 */
1364 	rte_delay_ms(1);
1365 	return 0;
1366 }
1367 
1368 /*
1369  * Recover hw state to let the worker threads continue.
1370  */
1371 void
1372 virtio_dev_resume(struct rte_eth_dev *dev)
1373 {
1374 	struct virtio_hw *hw = dev->data->dev_private;
1375 
1376 	hw->started = 1;
1377 	rte_spinlock_unlock(&hw->state_lock);
1378 }
1379 
1380 /*
1381  * Should be called only after device is paused.
1382  */
1383 int
1384 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1385 		int nb_pkts)
1386 {
1387 	struct virtio_hw *hw = dev->data->dev_private;
1388 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1389 	int ret;
1390 
1391 	hw->inject_pkts = tx_pkts;
1392 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1393 	hw->inject_pkts = NULL;
1394 
1395 	return ret;
1396 }
1397 
1398 static void
1399 virtio_notify_peers(struct rte_eth_dev *dev)
1400 {
1401 	struct virtio_hw *hw = dev->data->dev_private;
1402 	struct virtnet_rx *rxvq;
1403 	struct rte_mbuf *rarp_mbuf;
1404 
1405 	if (!dev->data->rx_queues)
1406 		return;
1407 
1408 	rxvq = dev->data->rx_queues[0];
1409 	if (!rxvq)
1410 		return;
1411 
1412 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1413 			(struct rte_ether_addr *)hw->mac_addr);
1414 	if (rarp_mbuf == NULL) {
1415 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1416 		return;
1417 	}
1418 
1419 	/* If virtio port just stopped, no need to send RARP */
1420 	if (virtio_dev_pause(dev) < 0) {
1421 		rte_pktmbuf_free(rarp_mbuf);
1422 		return;
1423 	}
1424 
1425 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1426 	virtio_dev_resume(dev);
1427 }
1428 
1429 static void
1430 virtio_ack_link_announce(struct rte_eth_dev *dev)
1431 {
1432 	struct virtio_hw *hw = dev->data->dev_private;
1433 	struct virtio_pmd_ctrl ctrl;
1434 
1435 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1436 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1437 
1438 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1439 }
1440 
1441 /*
1442  * Process virtio config changed interrupt. Call the callback
1443  * if link state changed, generate gratuitous RARP packet if
1444  * the status indicates an ANNOUNCE.
1445  */
1446 void
1447 virtio_interrupt_handler(void *param)
1448 {
1449 	struct rte_eth_dev *dev = param;
1450 	struct virtio_hw *hw = dev->data->dev_private;
1451 	uint8_t isr;
1452 	uint16_t status;
1453 
1454 	/* Read interrupt status which clears interrupt */
1455 	isr = virtio_get_isr(hw);
1456 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1457 
1458 	if (virtio_intr_unmask(dev) < 0)
1459 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1460 
1461 	if (isr & VIRTIO_ISR_CONFIG) {
1462 		if (virtio_dev_link_update(dev, 0) == 0)
1463 			rte_eth_dev_callback_process(dev,
1464 						     RTE_ETH_EVENT_INTR_LSC,
1465 						     NULL);
1466 
1467 		if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1468 			virtio_read_dev_config(hw,
1469 				offsetof(struct virtio_net_config, status),
1470 				&status, sizeof(status));
1471 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1472 				virtio_notify_peers(dev);
1473 				if (hw->cvq)
1474 					virtio_ack_link_announce(dev);
1475 			}
1476 		}
1477 	}
1478 }
1479 
1480 /* set rx and tx handlers according to what is supported */
1481 static void
1482 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1483 {
1484 	struct virtio_hw *hw = eth_dev->data->dev_private;
1485 
1486 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1487 	if (virtio_with_packed_queue(hw)) {
1488 		PMD_INIT_LOG(INFO,
1489 			"virtio: using packed ring %s Tx path on port %u",
1490 			hw->use_vec_tx ? "vectorized" : "standard",
1491 			eth_dev->data->port_id);
1492 		if (hw->use_vec_tx)
1493 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1494 		else
1495 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1496 	} else {
1497 		if (hw->use_inorder_tx) {
1498 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1499 				eth_dev->data->port_id);
1500 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1501 		} else {
1502 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1503 				eth_dev->data->port_id);
1504 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1505 		}
1506 	}
1507 
1508 	if (virtio_with_packed_queue(hw)) {
1509 		if (hw->use_vec_rx) {
1510 			PMD_INIT_LOG(INFO,
1511 				"virtio: using packed ring vectorized Rx path on port %u",
1512 				eth_dev->data->port_id);
1513 			eth_dev->rx_pkt_burst =
1514 				&virtio_recv_pkts_packed_vec;
1515 		} else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1516 			PMD_INIT_LOG(INFO,
1517 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1518 				eth_dev->data->port_id);
1519 			eth_dev->rx_pkt_burst =
1520 				&virtio_recv_mergeable_pkts_packed;
1521 		} else {
1522 			PMD_INIT_LOG(INFO,
1523 				"virtio: using packed ring standard Rx path on port %u",
1524 				eth_dev->data->port_id);
1525 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1526 		}
1527 	} else {
1528 		if (hw->use_vec_rx) {
1529 			PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1530 				eth_dev->data->port_id);
1531 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1532 		} else if (hw->use_inorder_rx) {
1533 			PMD_INIT_LOG(INFO,
1534 				"virtio: using inorder Rx path on port %u",
1535 				eth_dev->data->port_id);
1536 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1537 		} else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1538 			PMD_INIT_LOG(INFO,
1539 				"virtio: using mergeable buffer Rx path on port %u",
1540 				eth_dev->data->port_id);
1541 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1542 		} else {
1543 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1544 				eth_dev->data->port_id);
1545 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1546 		}
1547 	}
1548 
1549 }
1550 
1551 /* Only support 1:1 queue/interrupt mapping so far.
1552  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1553  * interrupt vectors (<N+1).
1554  */
1555 static int
1556 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1557 {
1558 	uint32_t i;
1559 	struct virtio_hw *hw = dev->data->dev_private;
1560 
1561 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1562 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1563 		dev->intr_handle->intr_vec[i] = i + 1;
1564 		if (VIRTIO_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1565 						 VIRTIO_MSI_NO_VECTOR) {
1566 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1567 			return -EBUSY;
1568 		}
1569 	}
1570 
1571 	return 0;
1572 }
1573 
1574 static void
1575 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1576 {
1577 	uint32_t i;
1578 	struct virtio_hw *hw = dev->data->dev_private;
1579 
1580 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1581 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1582 		VIRTIO_OPS(hw)->set_queue_irq(hw,
1583 					     hw->vqs[i * VTNET_CQ],
1584 					     VIRTIO_MSI_NO_VECTOR);
1585 }
1586 
1587 static int
1588 virtio_configure_intr(struct rte_eth_dev *dev)
1589 {
1590 	struct virtio_hw *hw = dev->data->dev_private;
1591 
1592 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1593 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1594 		return -ENOTSUP;
1595 	}
1596 
1597 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1598 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1599 		return -1;
1600 	}
1601 
1602 	if (!dev->intr_handle->intr_vec) {
1603 		dev->intr_handle->intr_vec =
1604 			rte_zmalloc("intr_vec",
1605 				    hw->max_queue_pairs * sizeof(int), 0);
1606 		if (!dev->intr_handle->intr_vec) {
1607 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1608 				     hw->max_queue_pairs);
1609 			return -ENOMEM;
1610 		}
1611 	}
1612 
1613 	/* Re-register callback to update max_intr */
1614 	rte_intr_callback_unregister(dev->intr_handle,
1615 				     virtio_interrupt_handler,
1616 				     dev);
1617 	rte_intr_callback_register(dev->intr_handle,
1618 				   virtio_interrupt_handler,
1619 				   dev);
1620 
1621 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1622 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1623 	 * And for legacy devices, this should be done before queue/vec binding
1624 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1625 	 * (22) will be ignored.
1626 	 */
1627 	if (virtio_intr_enable(dev) < 0) {
1628 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1629 		return -1;
1630 	}
1631 
1632 	if (virtio_queues_bind_intr(dev) < 0) {
1633 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1634 		return -1;
1635 	}
1636 
1637 	return 0;
1638 }
1639 #define DUPLEX_UNKNOWN   0xff
1640 /* reset device and renegotiate features if needed */
1641 static int
1642 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1643 {
1644 	struct virtio_hw *hw = eth_dev->data->dev_private;
1645 	struct virtio_net_config *config;
1646 	struct virtio_net_config local_config;
1647 	int ret;
1648 
1649 	/* Reset the device although not necessary at startup */
1650 	virtio_reset(hw);
1651 
1652 	if (hw->vqs) {
1653 		virtio_dev_free_mbufs(eth_dev);
1654 		virtio_free_queues(hw);
1655 	}
1656 
1657 	/* Tell the host we've noticed this device. */
1658 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1659 
1660 	/* Tell the host we've known how to drive the device. */
1661 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1662 	if (virtio_ethdev_negotiate_features(hw, req_features) < 0)
1663 		return -1;
1664 
1665 	hw->weak_barriers = !virtio_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1666 
1667 	/* If host does not support both status and MSI-X then disable LSC */
1668 	if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->intr_lsc)
1669 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1670 	else
1671 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1672 
1673 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1674 
1675 	/* Setting up rx_header size for the device */
1676 	if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1677 	    virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
1678 	    virtio_with_packed_queue(hw))
1679 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1680 	else
1681 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1682 
1683 	/* Copy the permanent MAC address to: virtio_hw */
1684 	virtio_get_hwaddr(hw);
1685 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1686 			&eth_dev->data->mac_addrs[0]);
1687 	PMD_INIT_LOG(DEBUG,
1688 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1689 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1690 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1691 
1692 	if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1693 		if (virtio_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1694 			config = &local_config;
1695 			virtio_read_dev_config(hw,
1696 				offsetof(struct virtio_net_config, speed),
1697 				&config->speed, sizeof(config->speed));
1698 			virtio_read_dev_config(hw,
1699 				offsetof(struct virtio_net_config, duplex),
1700 				&config->duplex, sizeof(config->duplex));
1701 			hw->speed = config->speed;
1702 			hw->duplex = config->duplex;
1703 		}
1704 	}
1705 	if (hw->duplex == DUPLEX_UNKNOWN)
1706 		hw->duplex = ETH_LINK_FULL_DUPLEX;
1707 	PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1708 		hw->speed, hw->duplex);
1709 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1710 		config = &local_config;
1711 
1712 		virtio_read_dev_config(hw,
1713 			offsetof(struct virtio_net_config, mac),
1714 			&config->mac, sizeof(config->mac));
1715 
1716 		if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1717 			virtio_read_dev_config(hw,
1718 				offsetof(struct virtio_net_config, status),
1719 				&config->status, sizeof(config->status));
1720 		} else {
1721 			PMD_INIT_LOG(DEBUG,
1722 				     "VIRTIO_NET_F_STATUS is not supported");
1723 			config->status = 0;
1724 		}
1725 
1726 		if (virtio_with_feature(hw, VIRTIO_NET_F_MQ)) {
1727 			virtio_read_dev_config(hw,
1728 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1729 				&config->max_virtqueue_pairs,
1730 				sizeof(config->max_virtqueue_pairs));
1731 		} else {
1732 			PMD_INIT_LOG(DEBUG,
1733 				     "VIRTIO_NET_F_MQ is not supported");
1734 			config->max_virtqueue_pairs = 1;
1735 		}
1736 
1737 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1738 
1739 		if (virtio_with_feature(hw, VIRTIO_NET_F_MTU)) {
1740 			virtio_read_dev_config(hw,
1741 				offsetof(struct virtio_net_config, mtu),
1742 				&config->mtu,
1743 				sizeof(config->mtu));
1744 
1745 			/*
1746 			 * MTU value has already been checked at negotiation
1747 			 * time, but check again in case it has changed since
1748 			 * then, which should not happen.
1749 			 */
1750 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1751 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1752 						config->mtu);
1753 				return -1;
1754 			}
1755 
1756 			hw->max_mtu = config->mtu;
1757 			/* Set initial MTU to maximum one supported by vhost */
1758 			eth_dev->data->mtu = config->mtu;
1759 
1760 		} else {
1761 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1762 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1763 		}
1764 
1765 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1766 				config->max_virtqueue_pairs);
1767 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1768 		PMD_INIT_LOG(DEBUG,
1769 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1770 				config->mac[0], config->mac[1],
1771 				config->mac[2], config->mac[3],
1772 				config->mac[4], config->mac[5]);
1773 	} else {
1774 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1775 		hw->max_queue_pairs = 1;
1776 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1777 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1778 	}
1779 
1780 	ret = virtio_alloc_queues(eth_dev);
1781 	if (ret < 0)
1782 		return ret;
1783 
1784 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1785 		if (virtio_configure_intr(eth_dev) < 0) {
1786 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1787 			virtio_free_queues(hw);
1788 			return -1;
1789 		}
1790 	}
1791 
1792 	virtio_reinit_complete(hw);
1793 
1794 	return 0;
1795 }
1796 
1797 /*
1798  * This function is based on probe() function in virtio_pci.c
1799  * It returns 0 on success.
1800  */
1801 int
1802 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1803 {
1804 	struct virtio_hw *hw = eth_dev->data->dev_private;
1805 	uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1806 	int vectorized = 0;
1807 	int ret;
1808 
1809 	if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1810 		PMD_INIT_LOG(ERR,
1811 			"Not sufficient headroom required = %d, avail = %d",
1812 			(int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1813 			RTE_PKTMBUF_HEADROOM);
1814 
1815 		return -1;
1816 	}
1817 
1818 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1819 	eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1820 
1821 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1822 		set_rxtx_funcs(eth_dev);
1823 		return 0;
1824 	}
1825 
1826 	ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1827 	if (ret < 0)
1828 		return ret;
1829 	hw->speed = speed;
1830 
1831 	/* Allocate memory for storing MAC addresses */
1832 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1833 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1834 	if (eth_dev->data->mac_addrs == NULL) {
1835 		PMD_INIT_LOG(ERR,
1836 			"Failed to allocate %d bytes needed to store MAC addresses",
1837 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1838 		return -ENOMEM;
1839 	}
1840 
1841 	rte_spinlock_init(&hw->state_lock);
1842 
1843 	/* reset device and negotiate default features */
1844 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1845 	if (ret < 0)
1846 		goto err_virtio_init;
1847 
1848 	if (vectorized) {
1849 		if (!virtio_with_packed_queue(hw)) {
1850 			hw->use_vec_rx = 1;
1851 		} else {
1852 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1853 			hw->use_vec_rx = 1;
1854 			hw->use_vec_tx = 1;
1855 #else
1856 			PMD_DRV_LOG(INFO,
1857 				"building environment do not support packed ring vectorized");
1858 #endif
1859 		}
1860 	}
1861 
1862 	hw->opened = 1;
1863 
1864 	return 0;
1865 
1866 err_virtio_init:
1867 	rte_free(eth_dev->data->mac_addrs);
1868 	eth_dev->data->mac_addrs = NULL;
1869 	return ret;
1870 }
1871 
1872 static uint32_t
1873 virtio_dev_speed_capa_get(uint32_t speed)
1874 {
1875 	switch (speed) {
1876 	case ETH_SPEED_NUM_10G:
1877 		return ETH_LINK_SPEED_10G;
1878 	case ETH_SPEED_NUM_20G:
1879 		return ETH_LINK_SPEED_20G;
1880 	case ETH_SPEED_NUM_25G:
1881 		return ETH_LINK_SPEED_25G;
1882 	case ETH_SPEED_NUM_40G:
1883 		return ETH_LINK_SPEED_40G;
1884 	case ETH_SPEED_NUM_50G:
1885 		return ETH_LINK_SPEED_50G;
1886 	case ETH_SPEED_NUM_56G:
1887 		return ETH_LINK_SPEED_56G;
1888 	case ETH_SPEED_NUM_100G:
1889 		return ETH_LINK_SPEED_100G;
1890 	case ETH_SPEED_NUM_200G:
1891 		return ETH_LINK_SPEED_200G;
1892 	default:
1893 		return 0;
1894 	}
1895 }
1896 
1897 static int vectorized_check_handler(__rte_unused const char *key,
1898 		const char *value, void *ret_val)
1899 {
1900 	if (strcmp(value, "1") == 0)
1901 		*(int *)ret_val = 1;
1902 	else
1903 		*(int *)ret_val = 0;
1904 
1905 	return 0;
1906 }
1907 
1908 #define VIRTIO_ARG_SPEED      "speed"
1909 #define VIRTIO_ARG_VECTORIZED "vectorized"
1910 
1911 static int
1912 link_speed_handler(const char *key __rte_unused,
1913 		const char *value, void *ret_val)
1914 {
1915 	uint32_t val;
1916 	if (!value || !ret_val)
1917 		return -EINVAL;
1918 	val = strtoul(value, NULL, 0);
1919 	/* validate input */
1920 	if (virtio_dev_speed_capa_get(val) == 0)
1921 		return -EINVAL;
1922 	*(uint32_t *)ret_val = val;
1923 
1924 	return 0;
1925 }
1926 
1927 
1928 static int
1929 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
1930 {
1931 	struct rte_kvargs *kvlist;
1932 	int ret = 0;
1933 
1934 	if (devargs == NULL)
1935 		return 0;
1936 
1937 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1938 	if (kvlist == NULL) {
1939 		PMD_INIT_LOG(ERR, "error when parsing param");
1940 		return 0;
1941 	}
1942 
1943 	if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
1944 		ret = rte_kvargs_process(kvlist,
1945 					VIRTIO_ARG_SPEED,
1946 					link_speed_handler, speed);
1947 		if (ret < 0) {
1948 			PMD_INIT_LOG(ERR, "Failed to parse %s",
1949 					VIRTIO_ARG_SPEED);
1950 			goto exit;
1951 		}
1952 	}
1953 
1954 	if (vectorized &&
1955 		rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
1956 		ret = rte_kvargs_process(kvlist,
1957 				VIRTIO_ARG_VECTORIZED,
1958 				vectorized_check_handler, vectorized);
1959 		if (ret < 0) {
1960 			PMD_INIT_LOG(ERR, "Failed to parse %s",
1961 					VIRTIO_ARG_VECTORIZED);
1962 			goto exit;
1963 		}
1964 	}
1965 
1966 exit:
1967 	rte_kvargs_free(kvlist);
1968 	return ret;
1969 }
1970 
1971 static uint8_t
1972 rx_offload_enabled(struct virtio_hw *hw)
1973 {
1974 	return virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
1975 		virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1976 		virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
1977 }
1978 
1979 static uint8_t
1980 tx_offload_enabled(struct virtio_hw *hw)
1981 {
1982 	return virtio_with_feature(hw, VIRTIO_NET_F_CSUM) ||
1983 		virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
1984 		virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
1985 }
1986 
1987 /*
1988  * Configure virtio device
1989  * It returns 0 on success.
1990  */
1991 static int
1992 virtio_dev_configure(struct rte_eth_dev *dev)
1993 {
1994 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1995 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
1996 	struct virtio_hw *hw = dev->data->dev_private;
1997 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
1998 		hw->vtnet_hdr_size;
1999 	uint64_t rx_offloads = rxmode->offloads;
2000 	uint64_t tx_offloads = txmode->offloads;
2001 	uint64_t req_features;
2002 	int ret;
2003 
2004 	PMD_INIT_LOG(DEBUG, "configure");
2005 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2006 
2007 	if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2008 		PMD_DRV_LOG(ERR,
2009 			"Unsupported Rx multi queue mode %d",
2010 			rxmode->mq_mode);
2011 		return -EINVAL;
2012 	}
2013 
2014 	if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2015 		PMD_DRV_LOG(ERR,
2016 			"Unsupported Tx multi queue mode %d",
2017 			txmode->mq_mode);
2018 		return -EINVAL;
2019 	}
2020 
2021 	if (dev->data->dev_conf.intr_conf.rxq) {
2022 		ret = virtio_init_device(dev, hw->req_guest_features);
2023 		if (ret < 0)
2024 			return ret;
2025 	}
2026 
2027 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2028 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2029 
2030 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2031 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2032 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2033 
2034 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2035 		req_features |=
2036 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2037 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2038 
2039 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2040 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2041 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2042 
2043 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2044 		req_features |=
2045 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2046 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2047 
2048 	/* if request features changed, reinit the device */
2049 	if (req_features != hw->req_guest_features) {
2050 		ret = virtio_init_device(dev, req_features);
2051 		if (ret < 0)
2052 			return ret;
2053 	}
2054 
2055 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2056 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2057 		!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2058 		PMD_DRV_LOG(ERR,
2059 			"rx checksum not available on this host");
2060 		return -ENOTSUP;
2061 	}
2062 
2063 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2064 		(!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2065 		 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2066 		PMD_DRV_LOG(ERR,
2067 			"Large Receive Offload not available on this host");
2068 		return -ENOTSUP;
2069 	}
2070 
2071 	/* start control queue */
2072 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2073 		virtio_dev_cq_start(dev);
2074 
2075 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2076 		hw->vlan_strip = 1;
2077 
2078 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2079 			!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2080 		PMD_DRV_LOG(ERR,
2081 			    "vlan filtering not available on this host");
2082 		return -ENOTSUP;
2083 	}
2084 
2085 	hw->has_tx_offload = tx_offload_enabled(hw);
2086 	hw->has_rx_offload = rx_offload_enabled(hw);
2087 
2088 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2089 		/* Enable vector (0) for Link State Intrerrupt */
2090 		if (VIRTIO_OPS(hw)->set_config_irq(hw, 0) ==
2091 				VIRTIO_MSI_NO_VECTOR) {
2092 			PMD_DRV_LOG(ERR, "failed to set config vector");
2093 			return -EBUSY;
2094 		}
2095 
2096 	if (virtio_with_packed_queue(hw)) {
2097 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2098 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2099 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2100 		     !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2101 		     !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2102 		     rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2103 			PMD_DRV_LOG(INFO,
2104 				"disabled packed ring vectorized path for requirements not met");
2105 			hw->use_vec_rx = 0;
2106 			hw->use_vec_tx = 0;
2107 		}
2108 #elif defined(RTE_ARCH_ARM)
2109 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2110 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2111 		     !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2112 		     !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2113 		     rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2114 			PMD_DRV_LOG(INFO,
2115 				"disabled packed ring vectorized path for requirements not met");
2116 			hw->use_vec_rx = 0;
2117 			hw->use_vec_tx = 0;
2118 		}
2119 #else
2120 		hw->use_vec_rx = 0;
2121 		hw->use_vec_tx = 0;
2122 #endif
2123 
2124 		if (hw->use_vec_rx) {
2125 			if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2126 				PMD_DRV_LOG(INFO,
2127 					"disabled packed ring vectorized rx for mrg_rxbuf enabled");
2128 				hw->use_vec_rx = 0;
2129 			}
2130 
2131 			if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2132 				PMD_DRV_LOG(INFO,
2133 					"disabled packed ring vectorized rx for TCP_LRO enabled");
2134 				hw->use_vec_rx = 0;
2135 			}
2136 		}
2137 	} else {
2138 		if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2139 			hw->use_inorder_tx = 1;
2140 			hw->use_inorder_rx = 1;
2141 			hw->use_vec_rx = 0;
2142 		}
2143 
2144 		if (hw->use_vec_rx) {
2145 #if defined RTE_ARCH_ARM
2146 			if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2147 				PMD_DRV_LOG(INFO,
2148 					"disabled split ring vectorized path for requirement not met");
2149 				hw->use_vec_rx = 0;
2150 			}
2151 #endif
2152 			if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2153 				PMD_DRV_LOG(INFO,
2154 					"disabled split ring vectorized rx for mrg_rxbuf enabled");
2155 				hw->use_vec_rx = 0;
2156 			}
2157 
2158 			if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2159 					   DEV_RX_OFFLOAD_TCP_CKSUM |
2160 					   DEV_RX_OFFLOAD_TCP_LRO |
2161 					   DEV_RX_OFFLOAD_VLAN_STRIP)) {
2162 				PMD_DRV_LOG(INFO,
2163 					"disabled split ring vectorized rx for offloading enabled");
2164 				hw->use_vec_rx = 0;
2165 			}
2166 
2167 			if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2168 				PMD_DRV_LOG(INFO,
2169 					"disabled split ring vectorized rx, max SIMD bitwidth too low");
2170 				hw->use_vec_rx = 0;
2171 			}
2172 		}
2173 	}
2174 
2175 	return 0;
2176 }
2177 
2178 
2179 static int
2180 virtio_dev_start(struct rte_eth_dev *dev)
2181 {
2182 	uint16_t nb_queues, i;
2183 	struct virtnet_rx *rxvq;
2184 	struct virtnet_tx *txvq __rte_unused;
2185 	struct virtio_hw *hw = dev->data->dev_private;
2186 	int ret;
2187 
2188 	/* Finish the initialization of the queues */
2189 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2190 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2191 		if (ret < 0)
2192 			return ret;
2193 	}
2194 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2195 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2196 		if (ret < 0)
2197 			return ret;
2198 	}
2199 
2200 	/* check if lsc interrupt feature is enabled */
2201 	if (dev->data->dev_conf.intr_conf.lsc) {
2202 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2203 			PMD_DRV_LOG(ERR, "link status not supported by host");
2204 			return -ENOTSUP;
2205 		}
2206 	}
2207 
2208 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2209 	 * in device configure, but it could be unmapped  when device is
2210 	 * stopped.
2211 	 */
2212 	if (dev->data->dev_conf.intr_conf.lsc ||
2213 	    dev->data->dev_conf.intr_conf.rxq) {
2214 		virtio_intr_disable(dev);
2215 
2216 		/* Setup interrupt callback  */
2217 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2218 			rte_intr_callback_register(dev->intr_handle,
2219 						   virtio_interrupt_handler,
2220 						   dev);
2221 
2222 		if (virtio_intr_enable(dev) < 0) {
2223 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2224 			return -EIO;
2225 		}
2226 	}
2227 
2228 	/*Notify the backend
2229 	 *Otherwise the tap backend might already stop its queue due to fullness.
2230 	 *vhost backend will have no chance to be waked up
2231 	 */
2232 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2233 	if (hw->max_queue_pairs > 1) {
2234 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2235 			return -EINVAL;
2236 	}
2237 
2238 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2239 
2240 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2241 		rxvq = dev->data->rx_queues[i];
2242 		/* Flush the old packets */
2243 		virtqueue_rxvq_flush(rxvq->vq);
2244 		virtqueue_notify(rxvq->vq);
2245 	}
2246 
2247 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2248 		txvq = dev->data->tx_queues[i];
2249 		virtqueue_notify(txvq->vq);
2250 	}
2251 
2252 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2253 
2254 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2255 		rxvq = dev->data->rx_queues[i];
2256 		VIRTQUEUE_DUMP(rxvq->vq);
2257 	}
2258 
2259 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2260 		txvq = dev->data->tx_queues[i];
2261 		VIRTQUEUE_DUMP(txvq->vq);
2262 	}
2263 
2264 	set_rxtx_funcs(dev);
2265 	hw->started = 1;
2266 
2267 	/* Initialize Link state */
2268 	virtio_dev_link_update(dev, 0);
2269 
2270 	return 0;
2271 }
2272 
2273 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2274 {
2275 	struct virtio_hw *hw = dev->data->dev_private;
2276 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2277 	const char *type __rte_unused;
2278 	unsigned int i, mbuf_num = 0;
2279 	struct virtqueue *vq;
2280 	struct rte_mbuf *buf;
2281 	int queue_type;
2282 
2283 	if (hw->vqs == NULL)
2284 		return;
2285 
2286 	for (i = 0; i < nr_vq; i++) {
2287 		vq = hw->vqs[i];
2288 		if (!vq)
2289 			continue;
2290 
2291 		queue_type = virtio_get_queue_type(hw, i);
2292 		if (queue_type == VTNET_RQ)
2293 			type = "rxq";
2294 		else if (queue_type == VTNET_TQ)
2295 			type = "txq";
2296 		else
2297 			continue;
2298 
2299 		PMD_INIT_LOG(DEBUG,
2300 			"Before freeing %s[%d] used and unused buf",
2301 			type, i);
2302 		VIRTQUEUE_DUMP(vq);
2303 
2304 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2305 			rte_pktmbuf_free(buf);
2306 			mbuf_num++;
2307 		}
2308 
2309 		PMD_INIT_LOG(DEBUG,
2310 			"After freeing %s[%d] used and unused buf",
2311 			type, i);
2312 		VIRTQUEUE_DUMP(vq);
2313 	}
2314 
2315 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2316 }
2317 
2318 /*
2319  * Stop device: disable interrupt and mark link down
2320  */
2321 int
2322 virtio_dev_stop(struct rte_eth_dev *dev)
2323 {
2324 	struct virtio_hw *hw = dev->data->dev_private;
2325 	struct rte_eth_link link;
2326 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2327 
2328 	PMD_INIT_LOG(DEBUG, "stop");
2329 	dev->data->dev_started = 0;
2330 
2331 	rte_spinlock_lock(&hw->state_lock);
2332 	if (!hw->started)
2333 		goto out_unlock;
2334 	hw->started = 0;
2335 
2336 	if (intr_conf->lsc || intr_conf->rxq) {
2337 		virtio_intr_disable(dev);
2338 
2339 		/* Reset interrupt callback  */
2340 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2341 			rte_intr_callback_unregister(dev->intr_handle,
2342 						     virtio_interrupt_handler,
2343 						     dev);
2344 		}
2345 	}
2346 
2347 	memset(&link, 0, sizeof(link));
2348 	rte_eth_linkstatus_set(dev, &link);
2349 out_unlock:
2350 	rte_spinlock_unlock(&hw->state_lock);
2351 
2352 	return 0;
2353 }
2354 
2355 static int
2356 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2357 {
2358 	struct rte_eth_link link;
2359 	uint16_t status;
2360 	struct virtio_hw *hw = dev->data->dev_private;
2361 
2362 	memset(&link, 0, sizeof(link));
2363 	link.link_duplex = hw->duplex;
2364 	link.link_speed  = hw->speed;
2365 	link.link_autoneg = ETH_LINK_AUTONEG;
2366 
2367 	if (!hw->started) {
2368 		link.link_status = ETH_LINK_DOWN;
2369 		link.link_speed = ETH_SPEED_NUM_NONE;
2370 	} else if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2371 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2372 		virtio_read_dev_config(hw,
2373 				offsetof(struct virtio_net_config, status),
2374 				&status, sizeof(status));
2375 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2376 			link.link_status = ETH_LINK_DOWN;
2377 			link.link_speed = ETH_SPEED_NUM_NONE;
2378 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2379 				     dev->data->port_id);
2380 		} else {
2381 			link.link_status = ETH_LINK_UP;
2382 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2383 				     dev->data->port_id);
2384 		}
2385 	} else {
2386 		link.link_status = ETH_LINK_UP;
2387 	}
2388 
2389 	return rte_eth_linkstatus_set(dev, &link);
2390 }
2391 
2392 static int
2393 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2394 {
2395 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2396 	struct virtio_hw *hw = dev->data->dev_private;
2397 	uint64_t offloads = rxmode->offloads;
2398 
2399 	if (mask & ETH_VLAN_FILTER_MASK) {
2400 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2401 				!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2402 
2403 			PMD_DRV_LOG(NOTICE,
2404 				"vlan filtering not available on this host");
2405 
2406 			return -ENOTSUP;
2407 		}
2408 	}
2409 
2410 	if (mask & ETH_VLAN_STRIP_MASK)
2411 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2412 
2413 	return 0;
2414 }
2415 
2416 static int
2417 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2418 {
2419 	uint64_t tso_mask, host_features;
2420 	struct virtio_hw *hw = dev->data->dev_private;
2421 	dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2422 
2423 	dev_info->max_rx_queues =
2424 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2425 	dev_info->max_tx_queues =
2426 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2427 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2428 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2429 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2430 
2431 	host_features = VIRTIO_OPS(hw)->get_features(hw);
2432 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2433 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2434 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2435 		dev_info->rx_offload_capa |=
2436 			DEV_RX_OFFLOAD_TCP_CKSUM |
2437 			DEV_RX_OFFLOAD_UDP_CKSUM;
2438 	}
2439 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2440 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2441 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2442 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2443 	if ((host_features & tso_mask) == tso_mask)
2444 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2445 
2446 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2447 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2448 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2449 		dev_info->tx_offload_capa |=
2450 			DEV_TX_OFFLOAD_UDP_CKSUM |
2451 			DEV_TX_OFFLOAD_TCP_CKSUM;
2452 	}
2453 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2454 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2455 	if ((host_features & tso_mask) == tso_mask)
2456 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2457 
2458 	return 0;
2459 }
2460 
2461 /*
2462  * It enables testpmd to collect per queue stats.
2463  */
2464 static int
2465 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2466 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2467 __rte_unused uint8_t is_rx)
2468 {
2469 	return 0;
2470 }
2471 
2472 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2473 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);
2474