xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 89813a522e68076e6f50ec18b075fa57cc5ae937)
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 = virtnet_cq_to_vq(cvq);
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 = virtnet_cq_to_vq(cvq);
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) {
320 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
321 		return -1;
322 	}
323 
324 	rte_spinlock_lock(&cvq->lock);
325 	vq = virtnet_cq_to_vq(cvq);
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 	struct rte_mbuf *fake_mbuf = NULL;
439 
440 	PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
441 			queue_idx, numa_node);
442 
443 	/*
444 	 * Read the virtqueue size from the Queue Size field
445 	 * Always power of 2 and if 0 virtqueue does not exist
446 	 */
447 	vq_size = VIRTIO_OPS(hw)->get_queue_num(hw, queue_idx);
448 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
449 	if (vq_size == 0) {
450 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
451 		return -EINVAL;
452 	}
453 
454 	if (!virtio_with_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
455 		PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
456 		return -EINVAL;
457 	}
458 
459 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
460 		 dev->data->port_id, queue_idx);
461 
462 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
463 				vq_size * sizeof(struct vq_desc_extra),
464 				RTE_CACHE_LINE_SIZE);
465 	if (queue_type == VTNET_TQ) {
466 		/*
467 		 * For each xmit packet, allocate a virtio_net_hdr
468 		 * and indirect ring elements
469 		 */
470 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
471 	} else if (queue_type == VTNET_CQ) {
472 		/* Allocate a page for control vq command, data and status */
473 		sz_hdr_mz = rte_mem_page_size();
474 	}
475 
476 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
477 				numa_node);
478 	if (vq == NULL) {
479 		PMD_INIT_LOG(ERR, "can not allocate vq");
480 		return -ENOMEM;
481 	}
482 	hw->vqs[queue_idx] = vq;
483 
484 	vq->hw = hw;
485 	vq->vq_queue_index = queue_idx;
486 	vq->vq_nentries = vq_size;
487 	if (virtio_with_packed_queue(hw)) {
488 		vq->vq_packed.used_wrap_counter = 1;
489 		vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
490 		vq->vq_packed.event_flags_shadow = 0;
491 		if (queue_type == VTNET_RQ)
492 			vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
493 	}
494 
495 	/*
496 	 * Reserve a memzone for vring elements
497 	 */
498 	size = vring_size(hw, vq_size, VIRTIO_VRING_ALIGN);
499 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_VRING_ALIGN);
500 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
501 		     size, vq->vq_ring_size);
502 
503 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
504 			numa_node, RTE_MEMZONE_IOVA_CONTIG,
505 			VIRTIO_VRING_ALIGN);
506 	if (mz == NULL) {
507 		if (rte_errno == EEXIST)
508 			mz = rte_memzone_lookup(vq_name);
509 		if (mz == NULL) {
510 			ret = -ENOMEM;
511 			goto free_vq;
512 		}
513 	}
514 
515 	memset(mz->addr, 0, mz->len);
516 
517 	vq->vq_ring_mem = mz->iova;
518 	vq->vq_ring_virt_mem = mz->addr;
519 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
520 		     (uint64_t)mz->iova);
521 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
522 		     (uint64_t)(uintptr_t)mz->addr);
523 
524 	virtio_init_vring(vq);
525 
526 	if (sz_hdr_mz) {
527 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
528 			 dev->data->port_id, queue_idx);
529 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
530 				numa_node, RTE_MEMZONE_IOVA_CONTIG,
531 				RTE_CACHE_LINE_SIZE);
532 		if (hdr_mz == NULL) {
533 			if (rte_errno == EEXIST)
534 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
535 			if (hdr_mz == NULL) {
536 				ret = -ENOMEM;
537 				goto free_mz;
538 			}
539 		}
540 	}
541 
542 	if (queue_type == VTNET_RQ) {
543 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
544 			       sizeof(vq->sw_ring[0]);
545 
546 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
547 				RTE_CACHE_LINE_SIZE, numa_node);
548 		if (!sw_ring) {
549 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
550 			ret = -ENOMEM;
551 			goto free_hdr_mz;
552 		}
553 
554 		fake_mbuf = rte_zmalloc_socket("sw_ring", sizeof(*fake_mbuf),
555 				RTE_CACHE_LINE_SIZE, numa_node);
556 		if (!fake_mbuf) {
557 			PMD_INIT_LOG(ERR, "can not allocate fake mbuf");
558 			ret = -ENOMEM;
559 			goto free_sw_ring;
560 		}
561 
562 		vq->sw_ring = sw_ring;
563 		rxvq = &vq->rxq;
564 		rxvq->port_id = dev->data->port_id;
565 		rxvq->mz = mz;
566 		rxvq->fake_mbuf = fake_mbuf;
567 	} else if (queue_type == VTNET_TQ) {
568 		txvq = &vq->txq;
569 		txvq->port_id = dev->data->port_id;
570 		txvq->mz = mz;
571 		txvq->virtio_net_hdr_mz = hdr_mz;
572 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
573 	} else if (queue_type == VTNET_CQ) {
574 		cvq = &vq->cq;
575 		cvq->mz = mz;
576 		cvq->virtio_net_hdr_mz = hdr_mz;
577 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
578 		memset(cvq->virtio_net_hdr_mz->addr, 0, rte_mem_page_size());
579 
580 		hw->cvq = cvq;
581 	}
582 
583 	if (queue_type == VTNET_TQ) {
584 		struct virtio_tx_region *txr;
585 		unsigned int i;
586 
587 		txr = hdr_mz->addr;
588 		memset(txr, 0, vq_size * sizeof(*txr));
589 		for (i = 0; i < vq_size; i++) {
590 			/* first indirect descriptor is always the tx header */
591 			if (!virtio_with_packed_queue(hw)) {
592 				struct vring_desc *start_dp = txr[i].tx_indir;
593 				vring_desc_init_split(start_dp,
594 						      RTE_DIM(txr[i].tx_indir));
595 				start_dp->addr = txvq->virtio_net_hdr_mem
596 					+ i * sizeof(*txr)
597 					+ offsetof(struct virtio_tx_region,
598 						   tx_hdr);
599 				start_dp->len = hw->vtnet_hdr_size;
600 				start_dp->flags = VRING_DESC_F_NEXT;
601 			} else {
602 				struct vring_packed_desc *start_dp =
603 					txr[i].tx_packed_indir;
604 				vring_desc_init_indirect_packed(start_dp,
605 				      RTE_DIM(txr[i].tx_packed_indir));
606 				start_dp->addr = txvq->virtio_net_hdr_mem
607 					+ i * sizeof(*txr)
608 					+ offsetof(struct virtio_tx_region,
609 						   tx_hdr);
610 				start_dp->len = hw->vtnet_hdr_size;
611 			}
612 		}
613 	}
614 
615 	if (VIRTIO_OPS(hw)->setup_queue(hw, vq) < 0) {
616 		PMD_INIT_LOG(ERR, "setup_queue failed");
617 		ret = -EINVAL;
618 		goto clean_vq;
619 	}
620 
621 	return 0;
622 
623 clean_vq:
624 	hw->cvq = NULL;
625 	rte_free(fake_mbuf);
626 free_sw_ring:
627 	rte_free(sw_ring);
628 free_hdr_mz:
629 	rte_memzone_free(hdr_mz);
630 free_mz:
631 	rte_memzone_free(mz);
632 free_vq:
633 	rte_free(vq);
634 
635 	return ret;
636 }
637 
638 static void
639 virtio_free_queues(struct virtio_hw *hw)
640 {
641 	uint16_t nr_vq = virtio_get_nr_vq(hw);
642 	struct virtqueue *vq;
643 	int queue_type;
644 	uint16_t i;
645 
646 	if (hw->vqs == NULL)
647 		return;
648 
649 	for (i = 0; i < nr_vq; i++) {
650 		vq = hw->vqs[i];
651 		if (!vq)
652 			continue;
653 
654 		queue_type = virtio_get_queue_type(hw, i);
655 		if (queue_type == VTNET_RQ) {
656 			rte_free(vq->rxq.fake_mbuf);
657 			rte_free(vq->sw_ring);
658 			rte_memzone_free(vq->rxq.mz);
659 		} else if (queue_type == VTNET_TQ) {
660 			rte_memzone_free(vq->txq.mz);
661 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
662 		} else {
663 			rte_memzone_free(vq->cq.mz);
664 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
665 		}
666 
667 		rte_free(vq);
668 		hw->vqs[i] = NULL;
669 	}
670 
671 	rte_free(hw->vqs);
672 	hw->vqs = NULL;
673 }
674 
675 static int
676 virtio_alloc_queues(struct rte_eth_dev *dev)
677 {
678 	struct virtio_hw *hw = dev->data->dev_private;
679 	uint16_t nr_vq = virtio_get_nr_vq(hw);
680 	uint16_t i;
681 	int ret;
682 
683 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
684 	if (!hw->vqs) {
685 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
686 		return -ENOMEM;
687 	}
688 
689 	for (i = 0; i < nr_vq; i++) {
690 		ret = virtio_init_queue(dev, i);
691 		if (ret < 0) {
692 			virtio_free_queues(hw);
693 			return ret;
694 		}
695 	}
696 
697 	return 0;
698 }
699 
700 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
701 
702 int
703 virtio_dev_close(struct rte_eth_dev *dev)
704 {
705 	struct virtio_hw *hw = dev->data->dev_private;
706 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
707 
708 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
709 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
710 		return 0;
711 
712 	if (!hw->opened)
713 		return 0;
714 	hw->opened = 0;
715 
716 	/* reset the NIC */
717 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
718 		VIRTIO_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
719 	if (intr_conf->rxq)
720 		virtio_queues_unbind_intr(dev);
721 
722 	if (intr_conf->lsc || intr_conf->rxq) {
723 		virtio_intr_disable(dev);
724 		rte_intr_efd_disable(dev->intr_handle);
725 		rte_free(dev->intr_handle->intr_vec);
726 		dev->intr_handle->intr_vec = NULL;
727 	}
728 
729 	virtio_reset(hw);
730 	virtio_dev_free_mbufs(dev);
731 	virtio_free_queues(hw);
732 
733 	return VIRTIO_OPS(hw)->dev_close(hw);
734 }
735 
736 static int
737 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
738 {
739 	struct virtio_hw *hw = dev->data->dev_private;
740 	struct virtio_pmd_ctrl ctrl;
741 	int dlen[1];
742 	int ret;
743 
744 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
745 		PMD_INIT_LOG(INFO, "host does not support rx control");
746 		return -ENOTSUP;
747 	}
748 
749 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
750 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
751 	ctrl.data[0] = 1;
752 	dlen[0] = 1;
753 
754 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
755 	if (ret) {
756 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
757 		return -EAGAIN;
758 	}
759 
760 	return 0;
761 }
762 
763 static int
764 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
765 {
766 	struct virtio_hw *hw = dev->data->dev_private;
767 	struct virtio_pmd_ctrl ctrl;
768 	int dlen[1];
769 	int ret;
770 
771 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
772 		PMD_INIT_LOG(INFO, "host does not support rx control");
773 		return -ENOTSUP;
774 	}
775 
776 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
777 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
778 	ctrl.data[0] = 0;
779 	dlen[0] = 1;
780 
781 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
782 	if (ret) {
783 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
784 		return -EAGAIN;
785 	}
786 
787 	return 0;
788 }
789 
790 static int
791 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
792 {
793 	struct virtio_hw *hw = dev->data->dev_private;
794 	struct virtio_pmd_ctrl ctrl;
795 	int dlen[1];
796 	int ret;
797 
798 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
799 		PMD_INIT_LOG(INFO, "host does not support rx control");
800 		return -ENOTSUP;
801 	}
802 
803 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
804 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
805 	ctrl.data[0] = 1;
806 	dlen[0] = 1;
807 
808 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
809 	if (ret) {
810 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
811 		return -EAGAIN;
812 	}
813 
814 	return 0;
815 }
816 
817 static int
818 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
819 {
820 	struct virtio_hw *hw = dev->data->dev_private;
821 	struct virtio_pmd_ctrl ctrl;
822 	int dlen[1];
823 	int ret;
824 
825 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
826 		PMD_INIT_LOG(INFO, "host does not support rx control");
827 		return -ENOTSUP;
828 	}
829 
830 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
831 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
832 	ctrl.data[0] = 0;
833 	dlen[0] = 1;
834 
835 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
836 	if (ret) {
837 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
838 		return -EAGAIN;
839 	}
840 
841 	return 0;
842 }
843 
844 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
845 static int
846 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
847 {
848 	struct virtio_hw *hw = dev->data->dev_private;
849 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
850 				 hw->vtnet_hdr_size;
851 	uint32_t frame_size = mtu + ether_hdr_len;
852 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
853 
854 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
855 
856 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
857 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
858 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
859 		return -EINVAL;
860 	}
861 	return 0;
862 }
863 
864 static int
865 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
866 {
867 	struct virtio_hw *hw = dev->data->dev_private;
868 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
869 	struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
870 
871 	virtqueue_enable_intr(vq);
872 	virtio_mb(hw->weak_barriers);
873 	return 0;
874 }
875 
876 static int
877 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
878 {
879 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
880 	struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
881 
882 	virtqueue_disable_intr(vq);
883 	return 0;
884 }
885 
886 /*
887  * dev_ops for virtio, bare necessities for basic operation
888  */
889 static const struct eth_dev_ops virtio_eth_dev_ops = {
890 	.dev_configure           = virtio_dev_configure,
891 	.dev_start               = virtio_dev_start,
892 	.dev_stop                = virtio_dev_stop,
893 	.dev_close               = virtio_dev_close,
894 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
895 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
896 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
897 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
898 	.mtu_set                 = virtio_mtu_set,
899 	.dev_infos_get           = virtio_dev_info_get,
900 	.stats_get               = virtio_dev_stats_get,
901 	.xstats_get              = virtio_dev_xstats_get,
902 	.xstats_get_names        = virtio_dev_xstats_get_names,
903 	.stats_reset             = virtio_dev_stats_reset,
904 	.xstats_reset            = virtio_dev_stats_reset,
905 	.link_update             = virtio_dev_link_update,
906 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
907 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
908 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
909 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
910 	.rx_queue_release        = virtio_dev_queue_release,
911 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
912 	.tx_queue_release        = virtio_dev_queue_release,
913 	/* collect stats per queue */
914 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
915 	.vlan_filter_set         = virtio_vlan_filter_set,
916 	.mac_addr_add            = virtio_mac_addr_add,
917 	.mac_addr_remove         = virtio_mac_addr_remove,
918 	.mac_addr_set            = virtio_mac_addr_set,
919 };
920 
921 /*
922  * dev_ops for virtio-user in secondary processes, as we just have
923  * some limited supports currently.
924  */
925 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
926 	.dev_infos_get           = virtio_dev_info_get,
927 	.stats_get               = virtio_dev_stats_get,
928 	.xstats_get              = virtio_dev_xstats_get,
929 	.xstats_get_names        = virtio_dev_xstats_get_names,
930 	.stats_reset             = virtio_dev_stats_reset,
931 	.xstats_reset            = virtio_dev_stats_reset,
932 	/* collect stats per queue */
933 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
934 };
935 
936 static void
937 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
938 {
939 	unsigned i;
940 
941 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
942 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
943 		if (txvq == NULL)
944 			continue;
945 
946 		stats->opackets += txvq->stats.packets;
947 		stats->obytes += txvq->stats.bytes;
948 
949 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
950 			stats->q_opackets[i] = txvq->stats.packets;
951 			stats->q_obytes[i] = txvq->stats.bytes;
952 		}
953 	}
954 
955 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
956 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
957 		if (rxvq == NULL)
958 			continue;
959 
960 		stats->ipackets += rxvq->stats.packets;
961 		stats->ibytes += rxvq->stats.bytes;
962 		stats->ierrors += rxvq->stats.errors;
963 
964 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
965 			stats->q_ipackets[i] = rxvq->stats.packets;
966 			stats->q_ibytes[i] = rxvq->stats.bytes;
967 		}
968 	}
969 
970 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
971 }
972 
973 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
974 				       struct rte_eth_xstat_name *xstats_names,
975 				       __rte_unused unsigned limit)
976 {
977 	unsigned i;
978 	unsigned count = 0;
979 	unsigned t;
980 
981 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
982 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
983 
984 	if (xstats_names != NULL) {
985 		/* Note: limit checked in rte_eth_xstats_names() */
986 
987 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
988 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
989 			if (rxvq == NULL)
990 				continue;
991 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
992 				snprintf(xstats_names[count].name,
993 					sizeof(xstats_names[count].name),
994 					"rx_q%u_%s", i,
995 					rte_virtio_rxq_stat_strings[t].name);
996 				count++;
997 			}
998 		}
999 
1000 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
1001 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
1002 			if (txvq == NULL)
1003 				continue;
1004 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1005 				snprintf(xstats_names[count].name,
1006 					sizeof(xstats_names[count].name),
1007 					"tx_q%u_%s", i,
1008 					rte_virtio_txq_stat_strings[t].name);
1009 				count++;
1010 			}
1011 		}
1012 		return count;
1013 	}
1014 	return nstats;
1015 }
1016 
1017 static int
1018 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1019 		      unsigned n)
1020 {
1021 	unsigned i;
1022 	unsigned count = 0;
1023 
1024 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1025 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1026 
1027 	if (n < nstats)
1028 		return nstats;
1029 
1030 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1031 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1032 
1033 		if (rxvq == NULL)
1034 			continue;
1035 
1036 		unsigned t;
1037 
1038 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1039 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1040 				rte_virtio_rxq_stat_strings[t].offset);
1041 			xstats[count].id = count;
1042 			count++;
1043 		}
1044 	}
1045 
1046 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1047 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1048 
1049 		if (txvq == NULL)
1050 			continue;
1051 
1052 		unsigned t;
1053 
1054 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1055 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1056 				rte_virtio_txq_stat_strings[t].offset);
1057 			xstats[count].id = count;
1058 			count++;
1059 		}
1060 	}
1061 
1062 	return count;
1063 }
1064 
1065 static int
1066 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1067 {
1068 	virtio_update_stats(dev, stats);
1069 
1070 	return 0;
1071 }
1072 
1073 static int
1074 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1075 {
1076 	unsigned int i;
1077 
1078 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1079 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1080 		if (txvq == NULL)
1081 			continue;
1082 
1083 		txvq->stats.packets = 0;
1084 		txvq->stats.bytes = 0;
1085 		txvq->stats.multicast = 0;
1086 		txvq->stats.broadcast = 0;
1087 		memset(txvq->stats.size_bins, 0,
1088 		       sizeof(txvq->stats.size_bins[0]) * 8);
1089 	}
1090 
1091 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1092 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1093 		if (rxvq == NULL)
1094 			continue;
1095 
1096 		rxvq->stats.packets = 0;
1097 		rxvq->stats.bytes = 0;
1098 		rxvq->stats.errors = 0;
1099 		rxvq->stats.multicast = 0;
1100 		rxvq->stats.broadcast = 0;
1101 		memset(rxvq->stats.size_bins, 0,
1102 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1103 	}
1104 
1105 	return 0;
1106 }
1107 
1108 static void
1109 virtio_set_hwaddr(struct virtio_hw *hw)
1110 {
1111 	virtio_write_dev_config(hw,
1112 			offsetof(struct virtio_net_config, mac),
1113 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1114 }
1115 
1116 static void
1117 virtio_get_hwaddr(struct virtio_hw *hw)
1118 {
1119 	if (virtio_with_feature(hw, VIRTIO_NET_F_MAC)) {
1120 		virtio_read_dev_config(hw,
1121 			offsetof(struct virtio_net_config, mac),
1122 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1123 	} else {
1124 		rte_eth_random_addr(&hw->mac_addr[0]);
1125 		virtio_set_hwaddr(hw);
1126 	}
1127 }
1128 
1129 static int
1130 virtio_mac_table_set(struct virtio_hw *hw,
1131 		     const struct virtio_net_ctrl_mac *uc,
1132 		     const struct virtio_net_ctrl_mac *mc)
1133 {
1134 	struct virtio_pmd_ctrl ctrl;
1135 	int err, len[2];
1136 
1137 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1138 		PMD_DRV_LOG(INFO, "host does not support mac table");
1139 		return -1;
1140 	}
1141 
1142 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1143 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1144 
1145 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1146 	memcpy(ctrl.data, uc, len[0]);
1147 
1148 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1149 	memcpy(ctrl.data + len[0], mc, len[1]);
1150 
1151 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1152 	if (err != 0)
1153 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1154 	return err;
1155 }
1156 
1157 static int
1158 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1159 		    uint32_t index, uint32_t vmdq __rte_unused)
1160 {
1161 	struct virtio_hw *hw = dev->data->dev_private;
1162 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1163 	unsigned int i;
1164 	struct virtio_net_ctrl_mac *uc, *mc;
1165 
1166 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1167 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1168 		return -EINVAL;
1169 	}
1170 
1171 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1172 		sizeof(uc->entries));
1173 	uc->entries = 0;
1174 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1175 		sizeof(mc->entries));
1176 	mc->entries = 0;
1177 
1178 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1179 		const struct rte_ether_addr *addr
1180 			= (i == index) ? mac_addr : addrs + i;
1181 		struct virtio_net_ctrl_mac *tbl
1182 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1183 
1184 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1185 	}
1186 
1187 	return virtio_mac_table_set(hw, uc, mc);
1188 }
1189 
1190 static void
1191 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1192 {
1193 	struct virtio_hw *hw = dev->data->dev_private;
1194 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1195 	struct virtio_net_ctrl_mac *uc, *mc;
1196 	unsigned int i;
1197 
1198 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1199 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1200 		return;
1201 	}
1202 
1203 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1204 		sizeof(uc->entries));
1205 	uc->entries = 0;
1206 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1207 		sizeof(mc->entries));
1208 	mc->entries = 0;
1209 
1210 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1211 		struct virtio_net_ctrl_mac *tbl;
1212 
1213 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1214 			continue;
1215 
1216 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1217 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1218 			RTE_ETHER_ADDR_LEN);
1219 	}
1220 
1221 	virtio_mac_table_set(hw, uc, mc);
1222 }
1223 
1224 static int
1225 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1226 {
1227 	struct virtio_hw *hw = dev->data->dev_private;
1228 
1229 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1230 
1231 	/* Use atomic update if available */
1232 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1233 		struct virtio_pmd_ctrl ctrl;
1234 		int len = RTE_ETHER_ADDR_LEN;
1235 
1236 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1237 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1238 
1239 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1240 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1241 	}
1242 
1243 	if (!virtio_with_feature(hw, VIRTIO_NET_F_MAC))
1244 		return -ENOTSUP;
1245 
1246 	virtio_set_hwaddr(hw);
1247 	return 0;
1248 }
1249 
1250 static int
1251 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1252 {
1253 	struct virtio_hw *hw = dev->data->dev_private;
1254 	struct virtio_pmd_ctrl ctrl;
1255 	int len;
1256 
1257 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1258 		return -ENOTSUP;
1259 
1260 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1261 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1262 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1263 	len = sizeof(vlan_id);
1264 
1265 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1266 }
1267 
1268 static int
1269 virtio_intr_unmask(struct rte_eth_dev *dev)
1270 {
1271 	struct virtio_hw *hw = dev->data->dev_private;
1272 
1273 	if (rte_intr_ack(dev->intr_handle) < 0)
1274 		return -1;
1275 
1276 	if (VIRTIO_OPS(hw)->intr_detect)
1277 		VIRTIO_OPS(hw)->intr_detect(hw);
1278 
1279 	return 0;
1280 }
1281 
1282 static int
1283 virtio_intr_enable(struct rte_eth_dev *dev)
1284 {
1285 	struct virtio_hw *hw = dev->data->dev_private;
1286 
1287 	if (rte_intr_enable(dev->intr_handle) < 0)
1288 		return -1;
1289 
1290 	if (VIRTIO_OPS(hw)->intr_detect)
1291 		VIRTIO_OPS(hw)->intr_detect(hw);
1292 
1293 	return 0;
1294 }
1295 
1296 static int
1297 virtio_intr_disable(struct rte_eth_dev *dev)
1298 {
1299 	struct virtio_hw *hw = dev->data->dev_private;
1300 
1301 	if (rte_intr_disable(dev->intr_handle) < 0)
1302 		return -1;
1303 
1304 	if (VIRTIO_OPS(hw)->intr_detect)
1305 		VIRTIO_OPS(hw)->intr_detect(hw);
1306 
1307 	return 0;
1308 }
1309 
1310 static int
1311 virtio_ethdev_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1312 {
1313 	uint64_t host_features;
1314 
1315 	/* Prepare guest_features: feature that driver wants to support */
1316 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1317 		req_features);
1318 
1319 	/* Read device(host) feature bits */
1320 	host_features = VIRTIO_OPS(hw)->get_features(hw);
1321 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1322 		host_features);
1323 
1324 	/* If supported, ensure MTU value is valid before acknowledging it. */
1325 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1326 		struct virtio_net_config config;
1327 
1328 		virtio_read_dev_config(hw,
1329 			offsetof(struct virtio_net_config, mtu),
1330 			&config.mtu, sizeof(config.mtu));
1331 
1332 		if (config.mtu < RTE_ETHER_MIN_MTU)
1333 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1334 	}
1335 
1336 	/*
1337 	 * Negotiate features: Subset of device feature bits are written back
1338 	 * guest feature bits.
1339 	 */
1340 	hw->guest_features = req_features;
1341 	hw->guest_features = virtio_negotiate_features(hw, host_features);
1342 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1343 		hw->guest_features);
1344 
1345 	if (VIRTIO_OPS(hw)->features_ok(hw) < 0)
1346 		return -1;
1347 
1348 	if (virtio_with_feature(hw, VIRTIO_F_VERSION_1)) {
1349 		virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1350 
1351 		if (!(virtio_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1352 			PMD_INIT_LOG(ERR, "Failed to set FEATURES_OK status!");
1353 			return -1;
1354 		}
1355 	}
1356 
1357 	hw->req_guest_features = req_features;
1358 
1359 	return 0;
1360 }
1361 
1362 int
1363 virtio_dev_pause(struct rte_eth_dev *dev)
1364 {
1365 	struct virtio_hw *hw = dev->data->dev_private;
1366 
1367 	rte_spinlock_lock(&hw->state_lock);
1368 
1369 	if (hw->started == 0) {
1370 		/* Device is just stopped. */
1371 		rte_spinlock_unlock(&hw->state_lock);
1372 		return -1;
1373 	}
1374 	hw->started = 0;
1375 	/*
1376 	 * Prevent the worker threads from touching queues to avoid contention,
1377 	 * 1 ms should be enough for the ongoing Tx function to finish.
1378 	 */
1379 	rte_delay_ms(1);
1380 	return 0;
1381 }
1382 
1383 /*
1384  * Recover hw state to let the worker threads continue.
1385  */
1386 void
1387 virtio_dev_resume(struct rte_eth_dev *dev)
1388 {
1389 	struct virtio_hw *hw = dev->data->dev_private;
1390 
1391 	hw->started = 1;
1392 	rte_spinlock_unlock(&hw->state_lock);
1393 }
1394 
1395 /*
1396  * Should be called only after device is paused.
1397  */
1398 int
1399 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1400 		int nb_pkts)
1401 {
1402 	struct virtio_hw *hw = dev->data->dev_private;
1403 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1404 	int ret;
1405 
1406 	hw->inject_pkts = tx_pkts;
1407 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1408 	hw->inject_pkts = NULL;
1409 
1410 	return ret;
1411 }
1412 
1413 static void
1414 virtio_notify_peers(struct rte_eth_dev *dev)
1415 {
1416 	struct virtio_hw *hw = dev->data->dev_private;
1417 	struct virtnet_rx *rxvq;
1418 	struct rte_mbuf *rarp_mbuf;
1419 
1420 	if (!dev->data->rx_queues)
1421 		return;
1422 
1423 	rxvq = dev->data->rx_queues[0];
1424 	if (!rxvq)
1425 		return;
1426 
1427 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1428 			(struct rte_ether_addr *)hw->mac_addr);
1429 	if (rarp_mbuf == NULL) {
1430 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1431 		return;
1432 	}
1433 
1434 	/* If virtio port just stopped, no need to send RARP */
1435 	if (virtio_dev_pause(dev) < 0) {
1436 		rte_pktmbuf_free(rarp_mbuf);
1437 		return;
1438 	}
1439 
1440 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1441 	virtio_dev_resume(dev);
1442 }
1443 
1444 static void
1445 virtio_ack_link_announce(struct rte_eth_dev *dev)
1446 {
1447 	struct virtio_hw *hw = dev->data->dev_private;
1448 	struct virtio_pmd_ctrl ctrl;
1449 
1450 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1451 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1452 
1453 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1454 }
1455 
1456 /*
1457  * Process virtio config changed interrupt. Call the callback
1458  * if link state changed, generate gratuitous RARP packet if
1459  * the status indicates an ANNOUNCE.
1460  */
1461 void
1462 virtio_interrupt_handler(void *param)
1463 {
1464 	struct rte_eth_dev *dev = param;
1465 	struct virtio_hw *hw = dev->data->dev_private;
1466 	uint8_t isr;
1467 	uint16_t status;
1468 
1469 	/* Read interrupt status which clears interrupt */
1470 	isr = virtio_get_isr(hw);
1471 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1472 
1473 	if (virtio_intr_unmask(dev) < 0)
1474 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1475 
1476 	if (isr & VIRTIO_ISR_CONFIG) {
1477 		if (virtio_dev_link_update(dev, 0) == 0)
1478 			rte_eth_dev_callback_process(dev,
1479 						     RTE_ETH_EVENT_INTR_LSC,
1480 						     NULL);
1481 
1482 		if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1483 			virtio_read_dev_config(hw,
1484 				offsetof(struct virtio_net_config, status),
1485 				&status, sizeof(status));
1486 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1487 				virtio_notify_peers(dev);
1488 				if (hw->cvq)
1489 					virtio_ack_link_announce(dev);
1490 			}
1491 		}
1492 	}
1493 }
1494 
1495 /* set rx and tx handlers according to what is supported */
1496 static void
1497 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1498 {
1499 	struct virtio_hw *hw = eth_dev->data->dev_private;
1500 
1501 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1502 	if (virtio_with_packed_queue(hw)) {
1503 		PMD_INIT_LOG(INFO,
1504 			"virtio: using packed ring %s Tx path on port %u",
1505 			hw->use_vec_tx ? "vectorized" : "standard",
1506 			eth_dev->data->port_id);
1507 		if (hw->use_vec_tx)
1508 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1509 		else
1510 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1511 	} else {
1512 		if (hw->use_inorder_tx) {
1513 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1514 				eth_dev->data->port_id);
1515 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1516 		} else {
1517 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1518 				eth_dev->data->port_id);
1519 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1520 		}
1521 	}
1522 
1523 	if (virtio_with_packed_queue(hw)) {
1524 		if (hw->use_vec_rx) {
1525 			PMD_INIT_LOG(INFO,
1526 				"virtio: using packed ring vectorized Rx path on port %u",
1527 				eth_dev->data->port_id);
1528 			eth_dev->rx_pkt_burst =
1529 				&virtio_recv_pkts_packed_vec;
1530 		} else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1531 			PMD_INIT_LOG(INFO,
1532 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1533 				eth_dev->data->port_id);
1534 			eth_dev->rx_pkt_burst =
1535 				&virtio_recv_mergeable_pkts_packed;
1536 		} else {
1537 			PMD_INIT_LOG(INFO,
1538 				"virtio: using packed ring standard Rx path on port %u",
1539 				eth_dev->data->port_id);
1540 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1541 		}
1542 	} else {
1543 		if (hw->use_vec_rx) {
1544 			PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1545 				eth_dev->data->port_id);
1546 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1547 		} else if (hw->use_inorder_rx) {
1548 			PMD_INIT_LOG(INFO,
1549 				"virtio: using inorder Rx path on port %u",
1550 				eth_dev->data->port_id);
1551 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1552 		} else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1553 			PMD_INIT_LOG(INFO,
1554 				"virtio: using mergeable buffer Rx path on port %u",
1555 				eth_dev->data->port_id);
1556 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1557 		} else {
1558 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1559 				eth_dev->data->port_id);
1560 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1561 		}
1562 	}
1563 
1564 }
1565 
1566 /* Only support 1:1 queue/interrupt mapping so far.
1567  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1568  * interrupt vectors (<N+1).
1569  */
1570 static int
1571 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1572 {
1573 	uint32_t i;
1574 	struct virtio_hw *hw = dev->data->dev_private;
1575 
1576 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1577 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1578 		dev->intr_handle->intr_vec[i] = i + 1;
1579 		if (VIRTIO_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1580 						 VIRTIO_MSI_NO_VECTOR) {
1581 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1582 			return -EBUSY;
1583 		}
1584 	}
1585 
1586 	return 0;
1587 }
1588 
1589 static void
1590 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1591 {
1592 	uint32_t i;
1593 	struct virtio_hw *hw = dev->data->dev_private;
1594 
1595 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1596 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1597 		VIRTIO_OPS(hw)->set_queue_irq(hw,
1598 					     hw->vqs[i * VTNET_CQ],
1599 					     VIRTIO_MSI_NO_VECTOR);
1600 }
1601 
1602 static int
1603 virtio_configure_intr(struct rte_eth_dev *dev)
1604 {
1605 	struct virtio_hw *hw = dev->data->dev_private;
1606 
1607 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1608 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1609 		return -ENOTSUP;
1610 	}
1611 
1612 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1613 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1614 		return -1;
1615 	}
1616 
1617 	if (!dev->intr_handle->intr_vec) {
1618 		dev->intr_handle->intr_vec =
1619 			rte_zmalloc("intr_vec",
1620 				    hw->max_queue_pairs * sizeof(int), 0);
1621 		if (!dev->intr_handle->intr_vec) {
1622 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1623 				     hw->max_queue_pairs);
1624 			return -ENOMEM;
1625 		}
1626 	}
1627 
1628 	/* Re-register callback to update max_intr */
1629 	rte_intr_callback_unregister(dev->intr_handle,
1630 				     virtio_interrupt_handler,
1631 				     dev);
1632 	rte_intr_callback_register(dev->intr_handle,
1633 				   virtio_interrupt_handler,
1634 				   dev);
1635 
1636 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1637 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1638 	 * And for legacy devices, this should be done before queue/vec binding
1639 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1640 	 * (22) will be ignored.
1641 	 */
1642 	if (virtio_intr_enable(dev) < 0) {
1643 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1644 		return -1;
1645 	}
1646 
1647 	if (virtio_queues_bind_intr(dev) < 0) {
1648 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1649 		return -1;
1650 	}
1651 
1652 	return 0;
1653 }
1654 #define DUPLEX_UNKNOWN   0xff
1655 /* reset device and renegotiate features if needed */
1656 static int
1657 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1658 {
1659 	struct virtio_hw *hw = eth_dev->data->dev_private;
1660 	struct virtio_net_config *config;
1661 	struct virtio_net_config local_config;
1662 	int ret;
1663 
1664 	/* Reset the device although not necessary at startup */
1665 	virtio_reset(hw);
1666 
1667 	if (hw->vqs) {
1668 		virtio_dev_free_mbufs(eth_dev);
1669 		virtio_free_queues(hw);
1670 	}
1671 
1672 	/* Tell the host we've noticed this device. */
1673 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1674 
1675 	/* Tell the host we've known how to drive the device. */
1676 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1677 	if (virtio_ethdev_negotiate_features(hw, req_features) < 0)
1678 		return -1;
1679 
1680 	hw->weak_barriers = !virtio_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1681 
1682 	/* If host does not support both status and MSI-X then disable LSC */
1683 	if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->intr_lsc)
1684 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1685 	else
1686 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1687 
1688 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1689 
1690 	/* Setting up rx_header size for the device */
1691 	if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1692 	    virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
1693 	    virtio_with_packed_queue(hw))
1694 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1695 	else
1696 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1697 
1698 	/* Copy the permanent MAC address to: virtio_hw */
1699 	virtio_get_hwaddr(hw);
1700 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1701 			&eth_dev->data->mac_addrs[0]);
1702 	PMD_INIT_LOG(DEBUG,
1703 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1704 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1705 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1706 
1707 	if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1708 		if (virtio_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1709 			config = &local_config;
1710 			virtio_read_dev_config(hw,
1711 				offsetof(struct virtio_net_config, speed),
1712 				&config->speed, sizeof(config->speed));
1713 			virtio_read_dev_config(hw,
1714 				offsetof(struct virtio_net_config, duplex),
1715 				&config->duplex, sizeof(config->duplex));
1716 			hw->speed = config->speed;
1717 			hw->duplex = config->duplex;
1718 		}
1719 	}
1720 	if (hw->duplex == DUPLEX_UNKNOWN)
1721 		hw->duplex = ETH_LINK_FULL_DUPLEX;
1722 	PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1723 		hw->speed, hw->duplex);
1724 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1725 		config = &local_config;
1726 
1727 		virtio_read_dev_config(hw,
1728 			offsetof(struct virtio_net_config, mac),
1729 			&config->mac, sizeof(config->mac));
1730 
1731 		if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1732 			virtio_read_dev_config(hw,
1733 				offsetof(struct virtio_net_config, status),
1734 				&config->status, sizeof(config->status));
1735 		} else {
1736 			PMD_INIT_LOG(DEBUG,
1737 				     "VIRTIO_NET_F_STATUS is not supported");
1738 			config->status = 0;
1739 		}
1740 
1741 		if (virtio_with_feature(hw, VIRTIO_NET_F_MQ)) {
1742 			virtio_read_dev_config(hw,
1743 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1744 				&config->max_virtqueue_pairs,
1745 				sizeof(config->max_virtqueue_pairs));
1746 		} else {
1747 			PMD_INIT_LOG(DEBUG,
1748 				     "VIRTIO_NET_F_MQ is not supported");
1749 			config->max_virtqueue_pairs = 1;
1750 		}
1751 
1752 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1753 
1754 		if (virtio_with_feature(hw, VIRTIO_NET_F_MTU)) {
1755 			virtio_read_dev_config(hw,
1756 				offsetof(struct virtio_net_config, mtu),
1757 				&config->mtu,
1758 				sizeof(config->mtu));
1759 
1760 			/*
1761 			 * MTU value has already been checked at negotiation
1762 			 * time, but check again in case it has changed since
1763 			 * then, which should not happen.
1764 			 */
1765 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1766 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1767 						config->mtu);
1768 				return -1;
1769 			}
1770 
1771 			hw->max_mtu = config->mtu;
1772 			/* Set initial MTU to maximum one supported by vhost */
1773 			eth_dev->data->mtu = config->mtu;
1774 
1775 		} else {
1776 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1777 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1778 		}
1779 
1780 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1781 				config->max_virtqueue_pairs);
1782 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1783 		PMD_INIT_LOG(DEBUG,
1784 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1785 				config->mac[0], config->mac[1],
1786 				config->mac[2], config->mac[3],
1787 				config->mac[4], config->mac[5]);
1788 	} else {
1789 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1790 		hw->max_queue_pairs = 1;
1791 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1792 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1793 	}
1794 
1795 	ret = virtio_alloc_queues(eth_dev);
1796 	if (ret < 0)
1797 		return ret;
1798 
1799 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1800 		if (virtio_configure_intr(eth_dev) < 0) {
1801 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1802 			virtio_free_queues(hw);
1803 			return -1;
1804 		}
1805 	}
1806 
1807 	virtio_reinit_complete(hw);
1808 
1809 	return 0;
1810 }
1811 
1812 /*
1813  * This function is based on probe() function in virtio_pci.c
1814  * It returns 0 on success.
1815  */
1816 int
1817 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1818 {
1819 	struct virtio_hw *hw = eth_dev->data->dev_private;
1820 	uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1821 	int vectorized = 0;
1822 	int ret;
1823 
1824 	if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1825 		PMD_INIT_LOG(ERR,
1826 			"Not sufficient headroom required = %d, avail = %d",
1827 			(int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1828 			RTE_PKTMBUF_HEADROOM);
1829 
1830 		return -1;
1831 	}
1832 
1833 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1834 	eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1835 
1836 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1837 		set_rxtx_funcs(eth_dev);
1838 		return 0;
1839 	}
1840 
1841 	ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1842 	if (ret < 0)
1843 		return ret;
1844 	hw->speed = speed;
1845 
1846 	/* Allocate memory for storing MAC addresses */
1847 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1848 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1849 	if (eth_dev->data->mac_addrs == NULL) {
1850 		PMD_INIT_LOG(ERR,
1851 			"Failed to allocate %d bytes needed to store MAC addresses",
1852 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1853 		return -ENOMEM;
1854 	}
1855 
1856 	rte_spinlock_init(&hw->state_lock);
1857 
1858 	/* reset device and negotiate default features */
1859 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1860 	if (ret < 0)
1861 		goto err_virtio_init;
1862 
1863 	if (vectorized) {
1864 		if (!virtio_with_packed_queue(hw)) {
1865 			hw->use_vec_rx = 1;
1866 		} else {
1867 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1868 			hw->use_vec_rx = 1;
1869 			hw->use_vec_tx = 1;
1870 #else
1871 			PMD_DRV_LOG(INFO,
1872 				"building environment do not support packed ring vectorized");
1873 #endif
1874 		}
1875 	}
1876 
1877 	hw->opened = 1;
1878 
1879 	return 0;
1880 
1881 err_virtio_init:
1882 	rte_free(eth_dev->data->mac_addrs);
1883 	eth_dev->data->mac_addrs = NULL;
1884 	return ret;
1885 }
1886 
1887 static uint32_t
1888 virtio_dev_speed_capa_get(uint32_t speed)
1889 {
1890 	switch (speed) {
1891 	case ETH_SPEED_NUM_10G:
1892 		return ETH_LINK_SPEED_10G;
1893 	case ETH_SPEED_NUM_20G:
1894 		return ETH_LINK_SPEED_20G;
1895 	case ETH_SPEED_NUM_25G:
1896 		return ETH_LINK_SPEED_25G;
1897 	case ETH_SPEED_NUM_40G:
1898 		return ETH_LINK_SPEED_40G;
1899 	case ETH_SPEED_NUM_50G:
1900 		return ETH_LINK_SPEED_50G;
1901 	case ETH_SPEED_NUM_56G:
1902 		return ETH_LINK_SPEED_56G;
1903 	case ETH_SPEED_NUM_100G:
1904 		return ETH_LINK_SPEED_100G;
1905 	case ETH_SPEED_NUM_200G:
1906 		return ETH_LINK_SPEED_200G;
1907 	default:
1908 		return 0;
1909 	}
1910 }
1911 
1912 static int vectorized_check_handler(__rte_unused const char *key,
1913 		const char *value, void *ret_val)
1914 {
1915 	if (strcmp(value, "1") == 0)
1916 		*(int *)ret_val = 1;
1917 	else
1918 		*(int *)ret_val = 0;
1919 
1920 	return 0;
1921 }
1922 
1923 #define VIRTIO_ARG_SPEED      "speed"
1924 #define VIRTIO_ARG_VECTORIZED "vectorized"
1925 
1926 static int
1927 link_speed_handler(const char *key __rte_unused,
1928 		const char *value, void *ret_val)
1929 {
1930 	uint32_t val;
1931 	if (!value || !ret_val)
1932 		return -EINVAL;
1933 	val = strtoul(value, NULL, 0);
1934 	/* validate input */
1935 	if (virtio_dev_speed_capa_get(val) == 0)
1936 		return -EINVAL;
1937 	*(uint32_t *)ret_val = val;
1938 
1939 	return 0;
1940 }
1941 
1942 
1943 static int
1944 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
1945 {
1946 	struct rte_kvargs *kvlist;
1947 	int ret = 0;
1948 
1949 	if (devargs == NULL)
1950 		return 0;
1951 
1952 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1953 	if (kvlist == NULL) {
1954 		PMD_INIT_LOG(ERR, "error when parsing param");
1955 		return 0;
1956 	}
1957 
1958 	if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
1959 		ret = rte_kvargs_process(kvlist,
1960 					VIRTIO_ARG_SPEED,
1961 					link_speed_handler, speed);
1962 		if (ret < 0) {
1963 			PMD_INIT_LOG(ERR, "Failed to parse %s",
1964 					VIRTIO_ARG_SPEED);
1965 			goto exit;
1966 		}
1967 	}
1968 
1969 	if (vectorized &&
1970 		rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
1971 		ret = rte_kvargs_process(kvlist,
1972 				VIRTIO_ARG_VECTORIZED,
1973 				vectorized_check_handler, vectorized);
1974 		if (ret < 0) {
1975 			PMD_INIT_LOG(ERR, "Failed to parse %s",
1976 					VIRTIO_ARG_VECTORIZED);
1977 			goto exit;
1978 		}
1979 	}
1980 
1981 exit:
1982 	rte_kvargs_free(kvlist);
1983 	return ret;
1984 }
1985 
1986 static uint8_t
1987 rx_offload_enabled(struct virtio_hw *hw)
1988 {
1989 	return virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
1990 		virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1991 		virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
1992 }
1993 
1994 static uint8_t
1995 tx_offload_enabled(struct virtio_hw *hw)
1996 {
1997 	return virtio_with_feature(hw, VIRTIO_NET_F_CSUM) ||
1998 		virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
1999 		virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2000 }
2001 
2002 /*
2003  * Configure virtio device
2004  * It returns 0 on success.
2005  */
2006 static int
2007 virtio_dev_configure(struct rte_eth_dev *dev)
2008 {
2009 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2010 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2011 	struct virtio_hw *hw = dev->data->dev_private;
2012 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2013 		hw->vtnet_hdr_size;
2014 	uint64_t rx_offloads = rxmode->offloads;
2015 	uint64_t tx_offloads = txmode->offloads;
2016 	uint64_t req_features;
2017 	int ret;
2018 
2019 	PMD_INIT_LOG(DEBUG, "configure");
2020 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2021 
2022 	if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2023 		PMD_DRV_LOG(ERR,
2024 			"Unsupported Rx multi queue mode %d",
2025 			rxmode->mq_mode);
2026 		return -EINVAL;
2027 	}
2028 
2029 	if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2030 		PMD_DRV_LOG(ERR,
2031 			"Unsupported Tx multi queue mode %d",
2032 			txmode->mq_mode);
2033 		return -EINVAL;
2034 	}
2035 
2036 	if (dev->data->dev_conf.intr_conf.rxq) {
2037 		ret = virtio_init_device(dev, hw->req_guest_features);
2038 		if (ret < 0)
2039 			return ret;
2040 	}
2041 
2042 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2043 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2044 
2045 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2046 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2047 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2048 
2049 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2050 		req_features |=
2051 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2052 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2053 
2054 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2055 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2056 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2057 
2058 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2059 		req_features |=
2060 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2061 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2062 
2063 	/* if request features changed, reinit the device */
2064 	if (req_features != hw->req_guest_features) {
2065 		ret = virtio_init_device(dev, req_features);
2066 		if (ret < 0)
2067 			return ret;
2068 	}
2069 
2070 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2071 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2072 		!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2073 		PMD_DRV_LOG(ERR,
2074 			"rx checksum not available on this host");
2075 		return -ENOTSUP;
2076 	}
2077 
2078 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2079 		(!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2080 		 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2081 		PMD_DRV_LOG(ERR,
2082 			"Large Receive Offload not available on this host");
2083 		return -ENOTSUP;
2084 	}
2085 
2086 	/* start control queue */
2087 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2088 		virtio_dev_cq_start(dev);
2089 
2090 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2091 		hw->vlan_strip = 1;
2092 
2093 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2094 			!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2095 		PMD_DRV_LOG(ERR,
2096 			    "vlan filtering not available on this host");
2097 		return -ENOTSUP;
2098 	}
2099 
2100 	hw->has_tx_offload = tx_offload_enabled(hw);
2101 	hw->has_rx_offload = rx_offload_enabled(hw);
2102 
2103 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2104 		/* Enable vector (0) for Link State Intrerrupt */
2105 		if (VIRTIO_OPS(hw)->set_config_irq(hw, 0) ==
2106 				VIRTIO_MSI_NO_VECTOR) {
2107 			PMD_DRV_LOG(ERR, "failed to set config vector");
2108 			return -EBUSY;
2109 		}
2110 
2111 	if (virtio_with_packed_queue(hw)) {
2112 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2113 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2114 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2115 		     !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2116 		     !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2117 		     rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2118 			PMD_DRV_LOG(INFO,
2119 				"disabled packed ring vectorized path for requirements not met");
2120 			hw->use_vec_rx = 0;
2121 			hw->use_vec_tx = 0;
2122 		}
2123 #elif defined(RTE_ARCH_ARM)
2124 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2125 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2126 		     !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2127 		     !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2128 		     rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2129 			PMD_DRV_LOG(INFO,
2130 				"disabled packed ring vectorized path for requirements not met");
2131 			hw->use_vec_rx = 0;
2132 			hw->use_vec_tx = 0;
2133 		}
2134 #else
2135 		hw->use_vec_rx = 0;
2136 		hw->use_vec_tx = 0;
2137 #endif
2138 
2139 		if (hw->use_vec_rx) {
2140 			if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2141 				PMD_DRV_LOG(INFO,
2142 					"disabled packed ring vectorized rx for mrg_rxbuf enabled");
2143 				hw->use_vec_rx = 0;
2144 			}
2145 
2146 			if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2147 				PMD_DRV_LOG(INFO,
2148 					"disabled packed ring vectorized rx for TCP_LRO enabled");
2149 				hw->use_vec_rx = 0;
2150 			}
2151 		}
2152 	} else {
2153 		if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2154 			hw->use_inorder_tx = 1;
2155 			hw->use_inorder_rx = 1;
2156 			hw->use_vec_rx = 0;
2157 		}
2158 
2159 		if (hw->use_vec_rx) {
2160 #if defined RTE_ARCH_ARM
2161 			if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2162 				PMD_DRV_LOG(INFO,
2163 					"disabled split ring vectorized path for requirement not met");
2164 				hw->use_vec_rx = 0;
2165 			}
2166 #endif
2167 			if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2168 				PMD_DRV_LOG(INFO,
2169 					"disabled split ring vectorized rx for mrg_rxbuf enabled");
2170 				hw->use_vec_rx = 0;
2171 			}
2172 
2173 			if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2174 					   DEV_RX_OFFLOAD_TCP_CKSUM |
2175 					   DEV_RX_OFFLOAD_TCP_LRO |
2176 					   DEV_RX_OFFLOAD_VLAN_STRIP)) {
2177 				PMD_DRV_LOG(INFO,
2178 					"disabled split ring vectorized rx for offloading enabled");
2179 				hw->use_vec_rx = 0;
2180 			}
2181 
2182 			if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2183 				PMD_DRV_LOG(INFO,
2184 					"disabled split ring vectorized rx, max SIMD bitwidth too low");
2185 				hw->use_vec_rx = 0;
2186 			}
2187 		}
2188 	}
2189 
2190 	return 0;
2191 }
2192 
2193 
2194 static int
2195 virtio_dev_start(struct rte_eth_dev *dev)
2196 {
2197 	uint16_t nb_queues, i;
2198 	struct virtqueue *vq;
2199 	struct virtio_hw *hw = dev->data->dev_private;
2200 	int ret;
2201 
2202 	/* Finish the initialization of the queues */
2203 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2204 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2205 		if (ret < 0)
2206 			return ret;
2207 	}
2208 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2209 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2210 		if (ret < 0)
2211 			return ret;
2212 	}
2213 
2214 	/* check if lsc interrupt feature is enabled */
2215 	if (dev->data->dev_conf.intr_conf.lsc) {
2216 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2217 			PMD_DRV_LOG(ERR, "link status not supported by host");
2218 			return -ENOTSUP;
2219 		}
2220 	}
2221 
2222 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2223 	 * in device configure, but it could be unmapped  when device is
2224 	 * stopped.
2225 	 */
2226 	if (dev->data->dev_conf.intr_conf.lsc ||
2227 	    dev->data->dev_conf.intr_conf.rxq) {
2228 		virtio_intr_disable(dev);
2229 
2230 		/* Setup interrupt callback  */
2231 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2232 			rte_intr_callback_register(dev->intr_handle,
2233 						   virtio_interrupt_handler,
2234 						   dev);
2235 
2236 		if (virtio_intr_enable(dev) < 0) {
2237 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2238 			return -EIO;
2239 		}
2240 	}
2241 
2242 	/*Notify the backend
2243 	 *Otherwise the tap backend might already stop its queue due to fullness.
2244 	 *vhost backend will have no chance to be waked up
2245 	 */
2246 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2247 	if (hw->max_queue_pairs > 1) {
2248 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2249 			return -EINVAL;
2250 	}
2251 
2252 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2253 
2254 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2255 		vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2256 		/* Flush the old packets */
2257 		virtqueue_rxvq_flush(vq);
2258 		virtqueue_notify(vq);
2259 	}
2260 
2261 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2262 		vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2263 		virtqueue_notify(vq);
2264 	}
2265 
2266 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2267 
2268 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2269 		vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2270 		VIRTQUEUE_DUMP(vq);
2271 	}
2272 
2273 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2274 		vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2275 		VIRTQUEUE_DUMP(vq);
2276 	}
2277 
2278 	set_rxtx_funcs(dev);
2279 	hw->started = 1;
2280 
2281 	/* Initialize Link state */
2282 	virtio_dev_link_update(dev, 0);
2283 
2284 	return 0;
2285 }
2286 
2287 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2288 {
2289 	struct virtio_hw *hw = dev->data->dev_private;
2290 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2291 	const char *type __rte_unused;
2292 	unsigned int i, mbuf_num = 0;
2293 	struct virtqueue *vq;
2294 	struct rte_mbuf *buf;
2295 	int queue_type;
2296 
2297 	if (hw->vqs == NULL)
2298 		return;
2299 
2300 	for (i = 0; i < nr_vq; i++) {
2301 		vq = hw->vqs[i];
2302 		if (!vq)
2303 			continue;
2304 
2305 		queue_type = virtio_get_queue_type(hw, i);
2306 		if (queue_type == VTNET_RQ)
2307 			type = "rxq";
2308 		else if (queue_type == VTNET_TQ)
2309 			type = "txq";
2310 		else
2311 			continue;
2312 
2313 		PMD_INIT_LOG(DEBUG,
2314 			"Before freeing %s[%d] used and unused buf",
2315 			type, i);
2316 		VIRTQUEUE_DUMP(vq);
2317 
2318 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2319 			rte_pktmbuf_free(buf);
2320 			mbuf_num++;
2321 		}
2322 
2323 		PMD_INIT_LOG(DEBUG,
2324 			"After freeing %s[%d] used and unused buf",
2325 			type, i);
2326 		VIRTQUEUE_DUMP(vq);
2327 	}
2328 
2329 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2330 }
2331 
2332 /*
2333  * Stop device: disable interrupt and mark link down
2334  */
2335 int
2336 virtio_dev_stop(struct rte_eth_dev *dev)
2337 {
2338 	struct virtio_hw *hw = dev->data->dev_private;
2339 	struct rte_eth_link link;
2340 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2341 
2342 	PMD_INIT_LOG(DEBUG, "stop");
2343 	dev->data->dev_started = 0;
2344 
2345 	rte_spinlock_lock(&hw->state_lock);
2346 	if (!hw->started)
2347 		goto out_unlock;
2348 	hw->started = 0;
2349 
2350 	if (intr_conf->lsc || intr_conf->rxq) {
2351 		virtio_intr_disable(dev);
2352 
2353 		/* Reset interrupt callback  */
2354 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2355 			rte_intr_callback_unregister(dev->intr_handle,
2356 						     virtio_interrupt_handler,
2357 						     dev);
2358 		}
2359 	}
2360 
2361 	memset(&link, 0, sizeof(link));
2362 	rte_eth_linkstatus_set(dev, &link);
2363 out_unlock:
2364 	rte_spinlock_unlock(&hw->state_lock);
2365 
2366 	return 0;
2367 }
2368 
2369 static int
2370 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2371 {
2372 	struct rte_eth_link link;
2373 	uint16_t status;
2374 	struct virtio_hw *hw = dev->data->dev_private;
2375 
2376 	memset(&link, 0, sizeof(link));
2377 	link.link_duplex = hw->duplex;
2378 	link.link_speed  = hw->speed;
2379 	link.link_autoneg = ETH_LINK_AUTONEG;
2380 
2381 	if (!hw->started) {
2382 		link.link_status = ETH_LINK_DOWN;
2383 		link.link_speed = ETH_SPEED_NUM_NONE;
2384 	} else if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2385 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2386 		virtio_read_dev_config(hw,
2387 				offsetof(struct virtio_net_config, status),
2388 				&status, sizeof(status));
2389 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2390 			link.link_status = ETH_LINK_DOWN;
2391 			link.link_speed = ETH_SPEED_NUM_NONE;
2392 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2393 				     dev->data->port_id);
2394 		} else {
2395 			link.link_status = ETH_LINK_UP;
2396 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2397 				     dev->data->port_id);
2398 		}
2399 	} else {
2400 		link.link_status = ETH_LINK_UP;
2401 	}
2402 
2403 	return rte_eth_linkstatus_set(dev, &link);
2404 }
2405 
2406 static int
2407 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2408 {
2409 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2410 	struct virtio_hw *hw = dev->data->dev_private;
2411 	uint64_t offloads = rxmode->offloads;
2412 
2413 	if (mask & ETH_VLAN_FILTER_MASK) {
2414 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2415 				!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2416 
2417 			PMD_DRV_LOG(NOTICE,
2418 				"vlan filtering not available on this host");
2419 
2420 			return -ENOTSUP;
2421 		}
2422 	}
2423 
2424 	if (mask & ETH_VLAN_STRIP_MASK)
2425 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2426 
2427 	return 0;
2428 }
2429 
2430 static int
2431 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2432 {
2433 	uint64_t tso_mask, host_features;
2434 	struct virtio_hw *hw = dev->data->dev_private;
2435 	dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2436 
2437 	dev_info->max_rx_queues =
2438 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2439 	dev_info->max_tx_queues =
2440 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2441 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2442 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2443 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2444 
2445 	host_features = VIRTIO_OPS(hw)->get_features(hw);
2446 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2447 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2448 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2449 		dev_info->rx_offload_capa |=
2450 			DEV_RX_OFFLOAD_TCP_CKSUM |
2451 			DEV_RX_OFFLOAD_UDP_CKSUM;
2452 	}
2453 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2454 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2455 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2456 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2457 	if ((host_features & tso_mask) == tso_mask)
2458 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2459 
2460 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2461 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2462 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2463 		dev_info->tx_offload_capa |=
2464 			DEV_TX_OFFLOAD_UDP_CKSUM |
2465 			DEV_TX_OFFLOAD_TCP_CKSUM;
2466 	}
2467 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2468 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2469 	if ((host_features & tso_mask) == tso_mask)
2470 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2471 
2472 	return 0;
2473 }
2474 
2475 /*
2476  * It enables testpmd to collect per queue stats.
2477  */
2478 static int
2479 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2480 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2481 __rte_unused uint8_t is_rx)
2482 {
2483 	return 0;
2484 }
2485 
2486 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2487 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);
2488