xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision dc9e658013d9c2be798f9ff43b8cd2c63c844f17)
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 <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_memcpy.h>
14 #include <rte_string_fns.h>
15 #include <rte_memzone.h>
16 #include <rte_malloc.h>
17 #include <rte_branch_prediction.h>
18 #include <rte_pci.h>
19 #include <rte_bus_pci.h>
20 #include <rte_ether.h>
21 #include <rte_ip.h>
22 #include <rte_arp.h>
23 #include <rte_common.h>
24 #include <rte_errno.h>
25 #include <rte_cpuflags.h>
26 
27 #include <rte_memory.h>
28 #include <rte_eal.h>
29 #include <rte_dev.h>
30 #include <rte_cycles.h>
31 #include <rte_kvargs.h>
32 
33 #include "virtio_ethdev.h"
34 #include "virtio_pci.h"
35 #include "virtio_logs.h"
36 #include "virtqueue.h"
37 #include "virtio_rxtx.h"
38 #include "virtio_user/virtio_user_dev.h"
39 
40 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
41 static int  virtio_dev_configure(struct rte_eth_dev *dev);
42 static int  virtio_dev_start(struct rte_eth_dev *dev);
43 static void virtio_dev_stop(struct rte_eth_dev *dev);
44 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
45 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
46 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
47 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
48 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
49 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
50 	int *vdpa,
51 	uint32_t *speed,
52 	int *vectorized);
53 static int virtio_dev_info_get(struct rte_eth_dev *dev,
54 				struct rte_eth_dev_info *dev_info);
55 static int virtio_dev_link_update(struct rte_eth_dev *dev,
56 	int wait_to_complete);
57 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
58 
59 static void virtio_set_hwaddr(struct virtio_hw *hw);
60 static void virtio_get_hwaddr(struct virtio_hw *hw);
61 
62 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
63 				 struct rte_eth_stats *stats);
64 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
65 				 struct rte_eth_xstat *xstats, unsigned n);
66 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
67 				       struct rte_eth_xstat_name *xstats_names,
68 				       unsigned limit);
69 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
70 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
71 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
72 				uint16_t vlan_id, int on);
73 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
74 				struct rte_ether_addr *mac_addr,
75 				uint32_t index, uint32_t vmdq);
76 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
77 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
78 				struct rte_ether_addr *mac_addr);
79 
80 static int virtio_intr_disable(struct rte_eth_dev *dev);
81 
82 static int virtio_dev_queue_stats_mapping_set(
83 	struct rte_eth_dev *eth_dev,
84 	uint16_t queue_id,
85 	uint8_t stat_idx,
86 	uint8_t is_rx);
87 
88 static void virtio_notify_peers(struct rte_eth_dev *dev);
89 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
90 
91 /*
92  * The set of PCI devices this driver supports
93  */
94 static const struct rte_pci_id pci_id_virtio_map[] = {
95 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
96 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
97 	{ .vendor_id = 0, /* sentinel */ },
98 };
99 
100 struct rte_virtio_xstats_name_off {
101 	char name[RTE_ETH_XSTATS_NAME_SIZE];
102 	unsigned offset;
103 };
104 
105 /* [rt]x_qX_ is prepended to the name string here */
106 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
107 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
108 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
109 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
110 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
111 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
112 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
113 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
114 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
115 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
116 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
117 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
118 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
119 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
120 };
121 
122 /* [rt]x_qX_ is prepended to the name string here */
123 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
124 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
125 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
126 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
127 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
128 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
129 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
130 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
131 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
132 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
133 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
134 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
135 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
136 };
137 
138 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
139 			    sizeof(rte_virtio_rxq_stat_strings[0]))
140 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
141 			    sizeof(rte_virtio_txq_stat_strings[0]))
142 
143 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
144 
145 static struct virtio_pmd_ctrl *
146 virtio_send_command_packed(struct virtnet_ctl *cvq,
147 			   struct virtio_pmd_ctrl *ctrl,
148 			   int *dlen, int pkt_num)
149 {
150 	struct virtqueue *vq = cvq->vq;
151 	int head;
152 	struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
153 	struct virtio_pmd_ctrl *result;
154 	uint16_t flags;
155 	int sum = 0;
156 	int nb_descs = 0;
157 	int k;
158 
159 	/*
160 	 * Format is enforced in qemu code:
161 	 * One TX packet for header;
162 	 * At least one TX packet per argument;
163 	 * One RX packet for ACK.
164 	 */
165 	head = vq->vq_avail_idx;
166 	flags = vq->vq_packed.cached_flags;
167 	desc[head].addr = cvq->virtio_net_hdr_mem;
168 	desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
169 	vq->vq_free_cnt--;
170 	nb_descs++;
171 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
172 		vq->vq_avail_idx -= vq->vq_nentries;
173 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
174 	}
175 
176 	for (k = 0; k < pkt_num; k++) {
177 		desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
178 			+ sizeof(struct virtio_net_ctrl_hdr)
179 			+ sizeof(ctrl->status) + sizeof(uint8_t) * sum;
180 		desc[vq->vq_avail_idx].len = dlen[k];
181 		desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
182 			vq->vq_packed.cached_flags;
183 		sum += dlen[k];
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 ^=
189 				VRING_PACKED_DESC_F_AVAIL_USED;
190 		}
191 	}
192 
193 	desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
194 		+ sizeof(struct virtio_net_ctrl_hdr);
195 	desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
196 	desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
197 		vq->vq_packed.cached_flags;
198 	vq->vq_free_cnt--;
199 	nb_descs++;
200 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
201 		vq->vq_avail_idx -= vq->vq_nentries;
202 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
203 	}
204 
205 	virtio_wmb(vq->hw->weak_barriers);
206 	desc[head].flags = VRING_DESC_F_NEXT | flags;
207 
208 	virtio_wmb(vq->hw->weak_barriers);
209 	virtqueue_notify(vq);
210 
211 	/* wait for used descriptors in virtqueue */
212 	while (!desc_is_used(&desc[head], vq))
213 		usleep(100);
214 
215 	virtio_rmb(vq->hw->weak_barriers);
216 
217 	/* now get used descriptors */
218 	vq->vq_free_cnt += nb_descs;
219 	vq->vq_used_cons_idx += nb_descs;
220 	if (vq->vq_used_cons_idx >= vq->vq_nentries) {
221 		vq->vq_used_cons_idx -= vq->vq_nentries;
222 		vq->vq_packed.used_wrap_counter ^= 1;
223 	}
224 
225 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
226 			"vq->vq_avail_idx=%d\n"
227 			"vq->vq_used_cons_idx=%d\n"
228 			"vq->vq_packed.cached_flags=0x%x\n"
229 			"vq->vq_packed.used_wrap_counter=%d\n",
230 			vq->vq_free_cnt,
231 			vq->vq_avail_idx,
232 			vq->vq_used_cons_idx,
233 			vq->vq_packed.cached_flags,
234 			vq->vq_packed.used_wrap_counter);
235 
236 	result = cvq->virtio_net_hdr_mz->addr;
237 	return result;
238 }
239 
240 static struct virtio_pmd_ctrl *
241 virtio_send_command_split(struct virtnet_ctl *cvq,
242 			  struct virtio_pmd_ctrl *ctrl,
243 			  int *dlen, int pkt_num)
244 {
245 	struct virtio_pmd_ctrl *result;
246 	struct virtqueue *vq = cvq->vq;
247 	uint32_t head, i;
248 	int k, sum = 0;
249 
250 	head = vq->vq_desc_head_idx;
251 
252 	/*
253 	 * Format is enforced in qemu code:
254 	 * One TX packet for header;
255 	 * At least one TX packet per argument;
256 	 * One RX packet for ACK.
257 	 */
258 	vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
259 	vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
260 	vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
261 	vq->vq_free_cnt--;
262 	i = vq->vq_split.ring.desc[head].next;
263 
264 	for (k = 0; k < pkt_num; k++) {
265 		vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
266 		vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
267 			+ sizeof(struct virtio_net_ctrl_hdr)
268 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
269 		vq->vq_split.ring.desc[i].len = dlen[k];
270 		sum += dlen[k];
271 		vq->vq_free_cnt--;
272 		i = vq->vq_split.ring.desc[i].next;
273 	}
274 
275 	vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
276 	vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
277 			+ sizeof(struct virtio_net_ctrl_hdr);
278 	vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
279 	vq->vq_free_cnt--;
280 
281 	vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
282 
283 	vq_update_avail_ring(vq, head);
284 	vq_update_avail_idx(vq);
285 
286 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
287 
288 	virtqueue_notify(vq);
289 
290 	while (virtqueue_nused(vq) == 0)
291 		usleep(100);
292 
293 	while (virtqueue_nused(vq)) {
294 		uint32_t idx, desc_idx, used_idx;
295 		struct vring_used_elem *uep;
296 
297 		used_idx = (uint32_t)(vq->vq_used_cons_idx
298 				& (vq->vq_nentries - 1));
299 		uep = &vq->vq_split.ring.used->ring[used_idx];
300 		idx = (uint32_t) uep->id;
301 		desc_idx = idx;
302 
303 		while (vq->vq_split.ring.desc[desc_idx].flags &
304 				VRING_DESC_F_NEXT) {
305 			desc_idx = vq->vq_split.ring.desc[desc_idx].next;
306 			vq->vq_free_cnt++;
307 		}
308 
309 		vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
310 		vq->vq_desc_head_idx = idx;
311 
312 		vq->vq_used_cons_idx++;
313 		vq->vq_free_cnt++;
314 	}
315 
316 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
317 			vq->vq_free_cnt, vq->vq_desc_head_idx);
318 
319 	result = cvq->virtio_net_hdr_mz->addr;
320 	return result;
321 }
322 
323 static int
324 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
325 		    int *dlen, int pkt_num)
326 {
327 	virtio_net_ctrl_ack status = ~0;
328 	struct virtio_pmd_ctrl *result;
329 	struct virtqueue *vq;
330 
331 	ctrl->status = status;
332 
333 	if (!cvq || !cvq->vq) {
334 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
335 		return -1;
336 	}
337 
338 	rte_spinlock_lock(&cvq->lock);
339 	vq = cvq->vq;
340 
341 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
342 		"vq->hw->cvq = %p vq = %p",
343 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
344 
345 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
346 		rte_spinlock_unlock(&cvq->lock);
347 		return -1;
348 	}
349 
350 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
351 		sizeof(struct virtio_pmd_ctrl));
352 
353 	if (vtpci_packed_queue(vq->hw))
354 		result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
355 	else
356 		result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
357 
358 	rte_spinlock_unlock(&cvq->lock);
359 	return result->status;
360 }
361 
362 static int
363 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
364 {
365 	struct virtio_hw *hw = dev->data->dev_private;
366 	struct virtio_pmd_ctrl ctrl;
367 	int dlen[1];
368 	int ret;
369 
370 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
371 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
372 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
373 
374 	dlen[0] = sizeof(uint16_t);
375 
376 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
377 	if (ret) {
378 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
379 			  "failed, this is too late now...");
380 		return -EINVAL;
381 	}
382 
383 	return 0;
384 }
385 
386 static void
387 virtio_dev_queue_release(void *queue __rte_unused)
388 {
389 	/* do nothing */
390 }
391 
392 static uint16_t
393 virtio_get_nr_vq(struct virtio_hw *hw)
394 {
395 	uint16_t nr_vq = hw->max_queue_pairs * 2;
396 
397 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
398 		nr_vq += 1;
399 
400 	return nr_vq;
401 }
402 
403 static void
404 virtio_init_vring(struct virtqueue *vq)
405 {
406 	int size = vq->vq_nentries;
407 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
408 
409 	PMD_INIT_FUNC_TRACE();
410 
411 	memset(ring_mem, 0, vq->vq_ring_size);
412 
413 	vq->vq_used_cons_idx = 0;
414 	vq->vq_desc_head_idx = 0;
415 	vq->vq_avail_idx = 0;
416 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
417 	vq->vq_free_cnt = vq->vq_nentries;
418 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
419 	if (vtpci_packed_queue(vq->hw)) {
420 		vring_init_packed(&vq->vq_packed.ring, ring_mem,
421 				  VIRTIO_PCI_VRING_ALIGN, size);
422 		vring_desc_init_packed(vq, size);
423 	} else {
424 		struct vring *vr = &vq->vq_split.ring;
425 
426 		vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
427 		vring_desc_init_split(vr->desc, size);
428 	}
429 	/*
430 	 * Disable device(host) interrupting guest
431 	 */
432 	virtqueue_disable_intr(vq);
433 }
434 
435 static int
436 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
437 {
438 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
439 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
440 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
441 	unsigned int vq_size, size;
442 	struct virtio_hw *hw = dev->data->dev_private;
443 	struct virtnet_rx *rxvq = NULL;
444 	struct virtnet_tx *txvq = NULL;
445 	struct virtnet_ctl *cvq = NULL;
446 	struct virtqueue *vq;
447 	size_t sz_hdr_mz = 0;
448 	void *sw_ring = NULL;
449 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
450 	int ret;
451 	int numa_node = dev->device->numa_node;
452 
453 	PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
454 			vtpci_queue_idx, numa_node);
455 
456 	/*
457 	 * Read the virtqueue size from the Queue Size field
458 	 * Always power of 2 and if 0 virtqueue does not exist
459 	 */
460 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
461 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
462 	if (vq_size == 0) {
463 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
464 		return -EINVAL;
465 	}
466 
467 	if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
468 		PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
469 		return -EINVAL;
470 	}
471 
472 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
473 		 dev->data->port_id, vtpci_queue_idx);
474 
475 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
476 				vq_size * sizeof(struct vq_desc_extra),
477 				RTE_CACHE_LINE_SIZE);
478 	if (queue_type == VTNET_TQ) {
479 		/*
480 		 * For each xmit packet, allocate a virtio_net_hdr
481 		 * and indirect ring elements
482 		 */
483 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
484 	} else if (queue_type == VTNET_CQ) {
485 		/* Allocate a page for control vq command, data and status */
486 		sz_hdr_mz = PAGE_SIZE;
487 	}
488 
489 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
490 				numa_node);
491 	if (vq == NULL) {
492 		PMD_INIT_LOG(ERR, "can not allocate vq");
493 		return -ENOMEM;
494 	}
495 	hw->vqs[vtpci_queue_idx] = vq;
496 
497 	vq->hw = hw;
498 	vq->vq_queue_index = vtpci_queue_idx;
499 	vq->vq_nentries = vq_size;
500 	if (vtpci_packed_queue(hw)) {
501 		vq->vq_packed.used_wrap_counter = 1;
502 		vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
503 		vq->vq_packed.event_flags_shadow = 0;
504 		if (queue_type == VTNET_RQ)
505 			vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
506 	}
507 
508 	/*
509 	 * Reserve a memzone for vring elements
510 	 */
511 	size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
512 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
513 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
514 		     size, vq->vq_ring_size);
515 
516 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
517 			numa_node, RTE_MEMZONE_IOVA_CONTIG,
518 			VIRTIO_PCI_VRING_ALIGN);
519 	if (mz == NULL) {
520 		if (rte_errno == EEXIST)
521 			mz = rte_memzone_lookup(vq_name);
522 		if (mz == NULL) {
523 			ret = -ENOMEM;
524 			goto fail_q_alloc;
525 		}
526 	}
527 
528 	memset(mz->addr, 0, mz->len);
529 
530 	vq->vq_ring_mem = mz->iova;
531 	vq->vq_ring_virt_mem = mz->addr;
532 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
533 		     (uint64_t)mz->iova);
534 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
535 		     (uint64_t)(uintptr_t)mz->addr);
536 
537 	virtio_init_vring(vq);
538 
539 	if (sz_hdr_mz) {
540 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
541 			 dev->data->port_id, vtpci_queue_idx);
542 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
543 				numa_node, RTE_MEMZONE_IOVA_CONTIG,
544 				RTE_CACHE_LINE_SIZE);
545 		if (hdr_mz == NULL) {
546 			if (rte_errno == EEXIST)
547 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
548 			if (hdr_mz == NULL) {
549 				ret = -ENOMEM;
550 				goto fail_q_alloc;
551 			}
552 		}
553 	}
554 
555 	if (queue_type == VTNET_RQ) {
556 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
557 			       sizeof(vq->sw_ring[0]);
558 
559 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
560 				RTE_CACHE_LINE_SIZE, numa_node);
561 		if (!sw_ring) {
562 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
563 			ret = -ENOMEM;
564 			goto fail_q_alloc;
565 		}
566 
567 		vq->sw_ring = sw_ring;
568 		rxvq = &vq->rxq;
569 		rxvq->vq = vq;
570 		rxvq->port_id = dev->data->port_id;
571 		rxvq->mz = mz;
572 	} else if (queue_type == VTNET_TQ) {
573 		txvq = &vq->txq;
574 		txvq->vq = vq;
575 		txvq->port_id = dev->data->port_id;
576 		txvq->mz = mz;
577 		txvq->virtio_net_hdr_mz = hdr_mz;
578 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
579 	} else if (queue_type == VTNET_CQ) {
580 		cvq = &vq->cq;
581 		cvq->vq = vq;
582 		cvq->mz = mz;
583 		cvq->virtio_net_hdr_mz = hdr_mz;
584 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
585 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
586 
587 		hw->cvq = cvq;
588 	}
589 
590 	/* For virtio_user case (that is when hw->virtio_user_dev is not NULL),
591 	 * we use virtual address. And we need properly set _offset_, please see
592 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
593 	 */
594 	if (!hw->virtio_user_dev)
595 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
596 	else {
597 		vq->vq_ring_mem = (uintptr_t)mz->addr;
598 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
599 		if (queue_type == VTNET_TQ)
600 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
601 		else if (queue_type == VTNET_CQ)
602 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
603 	}
604 
605 	if (queue_type == VTNET_TQ) {
606 		struct virtio_tx_region *txr;
607 		unsigned int i;
608 
609 		txr = hdr_mz->addr;
610 		memset(txr, 0, vq_size * sizeof(*txr));
611 		for (i = 0; i < vq_size; i++) {
612 			struct vring_desc *start_dp = txr[i].tx_indir;
613 
614 			/* first indirect descriptor is always the tx header */
615 			if (!vtpci_packed_queue(hw)) {
616 				vring_desc_init_split(start_dp,
617 						      RTE_DIM(txr[i].tx_indir));
618 				start_dp->addr = txvq->virtio_net_hdr_mem
619 					+ i * sizeof(*txr)
620 					+ offsetof(struct virtio_tx_region,
621 						   tx_hdr);
622 				start_dp->len = hw->vtnet_hdr_size;
623 				start_dp->flags = VRING_DESC_F_NEXT;
624 			}
625 		}
626 	}
627 
628 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
629 		PMD_INIT_LOG(ERR, "setup_queue failed");
630 		return -EINVAL;
631 	}
632 
633 	return 0;
634 
635 fail_q_alloc:
636 	rte_free(sw_ring);
637 	rte_memzone_free(hdr_mz);
638 	rte_memzone_free(mz);
639 	rte_free(vq);
640 
641 	return ret;
642 }
643 
644 static void
645 virtio_free_queues(struct virtio_hw *hw)
646 {
647 	uint16_t nr_vq = virtio_get_nr_vq(hw);
648 	struct virtqueue *vq;
649 	int queue_type;
650 	uint16_t i;
651 
652 	if (hw->vqs == NULL)
653 		return;
654 
655 	for (i = 0; i < nr_vq; i++) {
656 		vq = hw->vqs[i];
657 		if (!vq)
658 			continue;
659 
660 		queue_type = virtio_get_queue_type(hw, i);
661 		if (queue_type == VTNET_RQ) {
662 			rte_free(vq->sw_ring);
663 			rte_memzone_free(vq->rxq.mz);
664 		} else if (queue_type == VTNET_TQ) {
665 			rte_memzone_free(vq->txq.mz);
666 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
667 		} else {
668 			rte_memzone_free(vq->cq.mz);
669 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
670 		}
671 
672 		rte_free(vq);
673 		hw->vqs[i] = NULL;
674 	}
675 
676 	rte_free(hw->vqs);
677 	hw->vqs = NULL;
678 }
679 
680 static int
681 virtio_alloc_queues(struct rte_eth_dev *dev)
682 {
683 	struct virtio_hw *hw = dev->data->dev_private;
684 	uint16_t nr_vq = virtio_get_nr_vq(hw);
685 	uint16_t i;
686 	int ret;
687 
688 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
689 	if (!hw->vqs) {
690 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
691 		return -ENOMEM;
692 	}
693 
694 	for (i = 0; i < nr_vq; i++) {
695 		ret = virtio_init_queue(dev, i);
696 		if (ret < 0) {
697 			virtio_free_queues(hw);
698 			return ret;
699 		}
700 	}
701 
702 	return 0;
703 }
704 
705 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
706 
707 static int
708 virtio_dev_close(struct rte_eth_dev *dev)
709 {
710 	struct virtio_hw *hw = dev->data->dev_private;
711 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
712 
713 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
714 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
715 		return 0;
716 
717 	if (!hw->opened)
718 		return 0;
719 	hw->opened = false;
720 
721 	/* reset the NIC */
722 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
723 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
724 	if (intr_conf->rxq)
725 		virtio_queues_unbind_intr(dev);
726 
727 	if (intr_conf->lsc || intr_conf->rxq) {
728 		virtio_intr_disable(dev);
729 		rte_intr_efd_disable(dev->intr_handle);
730 		rte_free(dev->intr_handle->intr_vec);
731 		dev->intr_handle->intr_vec = NULL;
732 	}
733 
734 	vtpci_reset(hw);
735 	virtio_dev_free_mbufs(dev);
736 	virtio_free_queues(hw);
737 
738 #ifdef RTE_VIRTIO_USER
739 	if (hw->virtio_user_dev)
740 		virtio_user_dev_uninit(hw->virtio_user_dev);
741 	else
742 #endif
743 	if (dev->device) {
744 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
745 		if (!hw->modern)
746 			rte_pci_ioport_unmap(VTPCI_IO(hw));
747 	}
748 
749 	return 0;
750 }
751 
752 static int
753 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
754 {
755 	struct virtio_hw *hw = dev->data->dev_private;
756 	struct virtio_pmd_ctrl ctrl;
757 	int dlen[1];
758 	int ret;
759 
760 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
761 		PMD_INIT_LOG(INFO, "host does not support rx control");
762 		return -ENOTSUP;
763 	}
764 
765 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
766 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
767 	ctrl.data[0] = 1;
768 	dlen[0] = 1;
769 
770 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
771 	if (ret) {
772 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
773 		return -EAGAIN;
774 	}
775 
776 	return 0;
777 }
778 
779 static int
780 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
781 {
782 	struct virtio_hw *hw = dev->data->dev_private;
783 	struct virtio_pmd_ctrl ctrl;
784 	int dlen[1];
785 	int ret;
786 
787 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
788 		PMD_INIT_LOG(INFO, "host does not support rx control");
789 		return -ENOTSUP;
790 	}
791 
792 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
793 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
794 	ctrl.data[0] = 0;
795 	dlen[0] = 1;
796 
797 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
798 	if (ret) {
799 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
800 		return -EAGAIN;
801 	}
802 
803 	return 0;
804 }
805 
806 static int
807 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
808 {
809 	struct virtio_hw *hw = dev->data->dev_private;
810 	struct virtio_pmd_ctrl ctrl;
811 	int dlen[1];
812 	int ret;
813 
814 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
815 		PMD_INIT_LOG(INFO, "host does not support rx control");
816 		return -ENOTSUP;
817 	}
818 
819 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
820 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
821 	ctrl.data[0] = 1;
822 	dlen[0] = 1;
823 
824 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
825 	if (ret) {
826 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
827 		return -EAGAIN;
828 	}
829 
830 	return 0;
831 }
832 
833 static int
834 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
835 {
836 	struct virtio_hw *hw = dev->data->dev_private;
837 	struct virtio_pmd_ctrl ctrl;
838 	int dlen[1];
839 	int ret;
840 
841 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
842 		PMD_INIT_LOG(INFO, "host does not support rx control");
843 		return -ENOTSUP;
844 	}
845 
846 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
847 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
848 	ctrl.data[0] = 0;
849 	dlen[0] = 1;
850 
851 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
852 	if (ret) {
853 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
854 		return -EAGAIN;
855 	}
856 
857 	return 0;
858 }
859 
860 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
861 static int
862 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
863 {
864 	struct virtio_hw *hw = dev->data->dev_private;
865 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
866 				 hw->vtnet_hdr_size;
867 	uint32_t frame_size = mtu + ether_hdr_len;
868 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
869 
870 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
871 
872 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
873 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
874 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
875 		return -EINVAL;
876 	}
877 	return 0;
878 }
879 
880 static int
881 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
882 {
883 	struct virtio_hw *hw = dev->data->dev_private;
884 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
885 	struct virtqueue *vq = rxvq->vq;
886 
887 	virtqueue_enable_intr(vq);
888 	virtio_mb(hw->weak_barriers);
889 	return 0;
890 }
891 
892 static int
893 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
894 {
895 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
896 	struct virtqueue *vq = rxvq->vq;
897 
898 	virtqueue_disable_intr(vq);
899 	return 0;
900 }
901 
902 /*
903  * dev_ops for virtio, bare necessities for basic operation
904  */
905 static const struct eth_dev_ops virtio_eth_dev_ops = {
906 	.dev_configure           = virtio_dev_configure,
907 	.dev_start               = virtio_dev_start,
908 	.dev_stop                = virtio_dev_stop,
909 	.dev_close               = virtio_dev_close,
910 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
911 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
912 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
913 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
914 	.mtu_set                 = virtio_mtu_set,
915 	.dev_infos_get           = virtio_dev_info_get,
916 	.stats_get               = virtio_dev_stats_get,
917 	.xstats_get              = virtio_dev_xstats_get,
918 	.xstats_get_names        = virtio_dev_xstats_get_names,
919 	.stats_reset             = virtio_dev_stats_reset,
920 	.xstats_reset            = virtio_dev_stats_reset,
921 	.link_update             = virtio_dev_link_update,
922 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
923 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
924 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
925 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
926 	.rx_queue_release        = virtio_dev_queue_release,
927 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
928 	.tx_queue_release        = virtio_dev_queue_release,
929 	/* collect stats per queue */
930 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
931 	.vlan_filter_set         = virtio_vlan_filter_set,
932 	.mac_addr_add            = virtio_mac_addr_add,
933 	.mac_addr_remove         = virtio_mac_addr_remove,
934 	.mac_addr_set            = virtio_mac_addr_set,
935 };
936 
937 /*
938  * dev_ops for virtio-user in secondary processes, as we just have
939  * some limited supports currently.
940  */
941 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
942 	.dev_infos_get           = virtio_dev_info_get,
943 	.stats_get               = virtio_dev_stats_get,
944 	.xstats_get              = virtio_dev_xstats_get,
945 	.xstats_get_names        = virtio_dev_xstats_get_names,
946 	.stats_reset             = virtio_dev_stats_reset,
947 	.xstats_reset            = virtio_dev_stats_reset,
948 	/* collect stats per queue */
949 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
950 };
951 
952 static void
953 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
954 {
955 	unsigned i;
956 
957 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
958 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
959 		if (txvq == NULL)
960 			continue;
961 
962 		stats->opackets += txvq->stats.packets;
963 		stats->obytes += txvq->stats.bytes;
964 
965 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
966 			stats->q_opackets[i] = txvq->stats.packets;
967 			stats->q_obytes[i] = txvq->stats.bytes;
968 		}
969 	}
970 
971 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
972 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
973 		if (rxvq == NULL)
974 			continue;
975 
976 		stats->ipackets += rxvq->stats.packets;
977 		stats->ibytes += rxvq->stats.bytes;
978 		stats->ierrors += rxvq->stats.errors;
979 
980 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
981 			stats->q_ipackets[i] = rxvq->stats.packets;
982 			stats->q_ibytes[i] = rxvq->stats.bytes;
983 		}
984 	}
985 
986 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
987 }
988 
989 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
990 				       struct rte_eth_xstat_name *xstats_names,
991 				       __rte_unused unsigned limit)
992 {
993 	unsigned i;
994 	unsigned count = 0;
995 	unsigned t;
996 
997 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
998 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
999 
1000 	if (xstats_names != NULL) {
1001 		/* Note: limit checked in rte_eth_xstats_names() */
1002 
1003 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
1004 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1005 			if (rxvq == NULL)
1006 				continue;
1007 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1008 				snprintf(xstats_names[count].name,
1009 					sizeof(xstats_names[count].name),
1010 					"rx_q%u_%s", i,
1011 					rte_virtio_rxq_stat_strings[t].name);
1012 				count++;
1013 			}
1014 		}
1015 
1016 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
1017 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
1018 			if (txvq == NULL)
1019 				continue;
1020 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1021 				snprintf(xstats_names[count].name,
1022 					sizeof(xstats_names[count].name),
1023 					"tx_q%u_%s", i,
1024 					rte_virtio_txq_stat_strings[t].name);
1025 				count++;
1026 			}
1027 		}
1028 		return count;
1029 	}
1030 	return nstats;
1031 }
1032 
1033 static int
1034 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1035 		      unsigned n)
1036 {
1037 	unsigned i;
1038 	unsigned count = 0;
1039 
1040 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1041 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1042 
1043 	if (n < nstats)
1044 		return nstats;
1045 
1046 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1047 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1048 
1049 		if (rxvq == NULL)
1050 			continue;
1051 
1052 		unsigned t;
1053 
1054 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1055 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1056 				rte_virtio_rxq_stat_strings[t].offset);
1057 			xstats[count].id = count;
1058 			count++;
1059 		}
1060 	}
1061 
1062 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1063 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1064 
1065 		if (txvq == NULL)
1066 			continue;
1067 
1068 		unsigned t;
1069 
1070 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1071 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1072 				rte_virtio_txq_stat_strings[t].offset);
1073 			xstats[count].id = count;
1074 			count++;
1075 		}
1076 	}
1077 
1078 	return count;
1079 }
1080 
1081 static int
1082 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1083 {
1084 	virtio_update_stats(dev, stats);
1085 
1086 	return 0;
1087 }
1088 
1089 static int
1090 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1091 {
1092 	unsigned int i;
1093 
1094 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1095 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1096 		if (txvq == NULL)
1097 			continue;
1098 
1099 		txvq->stats.packets = 0;
1100 		txvq->stats.bytes = 0;
1101 		txvq->stats.multicast = 0;
1102 		txvq->stats.broadcast = 0;
1103 		memset(txvq->stats.size_bins, 0,
1104 		       sizeof(txvq->stats.size_bins[0]) * 8);
1105 	}
1106 
1107 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1108 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1109 		if (rxvq == NULL)
1110 			continue;
1111 
1112 		rxvq->stats.packets = 0;
1113 		rxvq->stats.bytes = 0;
1114 		rxvq->stats.errors = 0;
1115 		rxvq->stats.multicast = 0;
1116 		rxvq->stats.broadcast = 0;
1117 		memset(rxvq->stats.size_bins, 0,
1118 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1119 	}
1120 
1121 	return 0;
1122 }
1123 
1124 static void
1125 virtio_set_hwaddr(struct virtio_hw *hw)
1126 {
1127 	vtpci_write_dev_config(hw,
1128 			offsetof(struct virtio_net_config, mac),
1129 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1130 }
1131 
1132 static void
1133 virtio_get_hwaddr(struct virtio_hw *hw)
1134 {
1135 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1136 		vtpci_read_dev_config(hw,
1137 			offsetof(struct virtio_net_config, mac),
1138 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1139 	} else {
1140 		rte_eth_random_addr(&hw->mac_addr[0]);
1141 		virtio_set_hwaddr(hw);
1142 	}
1143 }
1144 
1145 static int
1146 virtio_mac_table_set(struct virtio_hw *hw,
1147 		     const struct virtio_net_ctrl_mac *uc,
1148 		     const struct virtio_net_ctrl_mac *mc)
1149 {
1150 	struct virtio_pmd_ctrl ctrl;
1151 	int err, len[2];
1152 
1153 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1154 		PMD_DRV_LOG(INFO, "host does not support mac table");
1155 		return -1;
1156 	}
1157 
1158 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1159 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1160 
1161 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1162 	memcpy(ctrl.data, uc, len[0]);
1163 
1164 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1165 	memcpy(ctrl.data + len[0], mc, len[1]);
1166 
1167 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1168 	if (err != 0)
1169 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1170 	return err;
1171 }
1172 
1173 static int
1174 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1175 		    uint32_t index, uint32_t vmdq __rte_unused)
1176 {
1177 	struct virtio_hw *hw = dev->data->dev_private;
1178 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1179 	unsigned int i;
1180 	struct virtio_net_ctrl_mac *uc, *mc;
1181 
1182 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1183 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1184 		return -EINVAL;
1185 	}
1186 
1187 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1188 		sizeof(uc->entries));
1189 	uc->entries = 0;
1190 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1191 		sizeof(mc->entries));
1192 	mc->entries = 0;
1193 
1194 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1195 		const struct rte_ether_addr *addr
1196 			= (i == index) ? mac_addr : addrs + i;
1197 		struct virtio_net_ctrl_mac *tbl
1198 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1199 
1200 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1201 	}
1202 
1203 	return virtio_mac_table_set(hw, uc, mc);
1204 }
1205 
1206 static void
1207 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1208 {
1209 	struct virtio_hw *hw = dev->data->dev_private;
1210 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1211 	struct virtio_net_ctrl_mac *uc, *mc;
1212 	unsigned int i;
1213 
1214 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1215 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1216 		return;
1217 	}
1218 
1219 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1220 		sizeof(uc->entries));
1221 	uc->entries = 0;
1222 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1223 		sizeof(mc->entries));
1224 	mc->entries = 0;
1225 
1226 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1227 		struct virtio_net_ctrl_mac *tbl;
1228 
1229 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1230 			continue;
1231 
1232 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1233 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1234 			RTE_ETHER_ADDR_LEN);
1235 	}
1236 
1237 	virtio_mac_table_set(hw, uc, mc);
1238 }
1239 
1240 static int
1241 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1242 {
1243 	struct virtio_hw *hw = dev->data->dev_private;
1244 
1245 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1246 
1247 	/* Use atomic update if available */
1248 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1249 		struct virtio_pmd_ctrl ctrl;
1250 		int len = RTE_ETHER_ADDR_LEN;
1251 
1252 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1253 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1254 
1255 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1256 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1257 	}
1258 
1259 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1260 		return -ENOTSUP;
1261 
1262 	virtio_set_hwaddr(hw);
1263 	return 0;
1264 }
1265 
1266 static int
1267 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1268 {
1269 	struct virtio_hw *hw = dev->data->dev_private;
1270 	struct virtio_pmd_ctrl ctrl;
1271 	int len;
1272 
1273 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1274 		return -ENOTSUP;
1275 
1276 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1277 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1278 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1279 	len = sizeof(vlan_id);
1280 
1281 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1282 }
1283 
1284 static int
1285 virtio_intr_unmask(struct rte_eth_dev *dev)
1286 {
1287 	struct virtio_hw *hw = dev->data->dev_private;
1288 
1289 	if (rte_intr_ack(dev->intr_handle) < 0)
1290 		return -1;
1291 
1292 	if (!hw->virtio_user_dev)
1293 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1294 
1295 	return 0;
1296 }
1297 
1298 static int
1299 virtio_intr_enable(struct rte_eth_dev *dev)
1300 {
1301 	struct virtio_hw *hw = dev->data->dev_private;
1302 
1303 	if (rte_intr_enable(dev->intr_handle) < 0)
1304 		return -1;
1305 
1306 	if (!hw->virtio_user_dev)
1307 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1308 
1309 	return 0;
1310 }
1311 
1312 static int
1313 virtio_intr_disable(struct rte_eth_dev *dev)
1314 {
1315 	struct virtio_hw *hw = dev->data->dev_private;
1316 
1317 	if (rte_intr_disable(dev->intr_handle) < 0)
1318 		return -1;
1319 
1320 	if (!hw->virtio_user_dev)
1321 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1322 
1323 	return 0;
1324 }
1325 
1326 static int
1327 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1328 {
1329 	uint64_t host_features;
1330 
1331 	/* Prepare guest_features: feature that driver wants to support */
1332 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1333 		req_features);
1334 
1335 	/* Read device(host) feature bits */
1336 	host_features = VTPCI_OPS(hw)->get_features(hw);
1337 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1338 		host_features);
1339 
1340 	/* If supported, ensure MTU value is valid before acknowledging it. */
1341 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1342 		struct virtio_net_config config;
1343 
1344 		vtpci_read_dev_config(hw,
1345 			offsetof(struct virtio_net_config, mtu),
1346 			&config.mtu, sizeof(config.mtu));
1347 
1348 		if (config.mtu < RTE_ETHER_MIN_MTU)
1349 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1350 	}
1351 
1352 	/*
1353 	 * Negotiate features: Subset of device feature bits are written back
1354 	 * guest feature bits.
1355 	 */
1356 	hw->guest_features = req_features;
1357 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1358 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1359 		hw->guest_features);
1360 
1361 	if (hw->modern && !vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1362 		PMD_INIT_LOG(ERR,
1363 			"VIRTIO_F_VERSION_1 features is not enabled.");
1364 		return -1;
1365 	}
1366 
1367 	if (hw->modern || hw->virtio_user_dev) {
1368 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1369 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1370 			PMD_INIT_LOG(ERR,
1371 				"failed to set FEATURES_OK status!");
1372 			return -1;
1373 		}
1374 	}
1375 
1376 	hw->req_guest_features = req_features;
1377 
1378 	return 0;
1379 }
1380 
1381 int
1382 virtio_dev_pause(struct rte_eth_dev *dev)
1383 {
1384 	struct virtio_hw *hw = dev->data->dev_private;
1385 
1386 	rte_spinlock_lock(&hw->state_lock);
1387 
1388 	if (hw->started == 0) {
1389 		/* Device is just stopped. */
1390 		rte_spinlock_unlock(&hw->state_lock);
1391 		return -1;
1392 	}
1393 	hw->started = 0;
1394 	/*
1395 	 * Prevent the worker threads from touching queues to avoid contention,
1396 	 * 1 ms should be enough for the ongoing Tx function to finish.
1397 	 */
1398 	rte_delay_ms(1);
1399 	return 0;
1400 }
1401 
1402 /*
1403  * Recover hw state to let the worker threads continue.
1404  */
1405 void
1406 virtio_dev_resume(struct rte_eth_dev *dev)
1407 {
1408 	struct virtio_hw *hw = dev->data->dev_private;
1409 
1410 	hw->started = 1;
1411 	rte_spinlock_unlock(&hw->state_lock);
1412 }
1413 
1414 /*
1415  * Should be called only after device is paused.
1416  */
1417 int
1418 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1419 		int nb_pkts)
1420 {
1421 	struct virtio_hw *hw = dev->data->dev_private;
1422 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1423 	int ret;
1424 
1425 	hw->inject_pkts = tx_pkts;
1426 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1427 	hw->inject_pkts = NULL;
1428 
1429 	return ret;
1430 }
1431 
1432 static void
1433 virtio_notify_peers(struct rte_eth_dev *dev)
1434 {
1435 	struct virtio_hw *hw = dev->data->dev_private;
1436 	struct virtnet_rx *rxvq;
1437 	struct rte_mbuf *rarp_mbuf;
1438 
1439 	if (!dev->data->rx_queues)
1440 		return;
1441 
1442 	rxvq = dev->data->rx_queues[0];
1443 	if (!rxvq)
1444 		return;
1445 
1446 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1447 			(struct rte_ether_addr *)hw->mac_addr);
1448 	if (rarp_mbuf == NULL) {
1449 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1450 		return;
1451 	}
1452 
1453 	/* If virtio port just stopped, no need to send RARP */
1454 	if (virtio_dev_pause(dev) < 0) {
1455 		rte_pktmbuf_free(rarp_mbuf);
1456 		return;
1457 	}
1458 
1459 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1460 	virtio_dev_resume(dev);
1461 }
1462 
1463 static void
1464 virtio_ack_link_announce(struct rte_eth_dev *dev)
1465 {
1466 	struct virtio_hw *hw = dev->data->dev_private;
1467 	struct virtio_pmd_ctrl ctrl;
1468 
1469 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1470 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1471 
1472 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1473 }
1474 
1475 /*
1476  * Process virtio config changed interrupt. Call the callback
1477  * if link state changed, generate gratuitous RARP packet if
1478  * the status indicates an ANNOUNCE.
1479  */
1480 void
1481 virtio_interrupt_handler(void *param)
1482 {
1483 	struct rte_eth_dev *dev = param;
1484 	struct virtio_hw *hw = dev->data->dev_private;
1485 	uint8_t isr;
1486 	uint16_t status;
1487 
1488 	/* Read interrupt status which clears interrupt */
1489 	isr = vtpci_isr(hw);
1490 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1491 
1492 	if (virtio_intr_unmask(dev) < 0)
1493 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1494 
1495 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1496 		if (virtio_dev_link_update(dev, 0) == 0)
1497 			rte_eth_dev_callback_process(dev,
1498 						     RTE_ETH_EVENT_INTR_LSC,
1499 						     NULL);
1500 
1501 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1502 			vtpci_read_dev_config(hw,
1503 				offsetof(struct virtio_net_config, status),
1504 				&status, sizeof(status));
1505 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1506 				virtio_notify_peers(dev);
1507 				if (hw->cvq)
1508 					virtio_ack_link_announce(dev);
1509 			}
1510 		}
1511 	}
1512 }
1513 
1514 /* set rx and tx handlers according to what is supported */
1515 static void
1516 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1517 {
1518 	struct virtio_hw *hw = eth_dev->data->dev_private;
1519 
1520 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1521 	if (vtpci_packed_queue(hw)) {
1522 		PMD_INIT_LOG(INFO,
1523 			"virtio: using packed ring %s Tx path on port %u",
1524 			hw->use_vec_tx ? "vectorized" : "standard",
1525 			eth_dev->data->port_id);
1526 		if (hw->use_vec_tx)
1527 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1528 		else
1529 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1530 	} else {
1531 		if (hw->use_inorder_tx) {
1532 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1533 				eth_dev->data->port_id);
1534 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1535 		} else {
1536 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1537 				eth_dev->data->port_id);
1538 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1539 		}
1540 	}
1541 
1542 	if (vtpci_packed_queue(hw)) {
1543 		if (hw->use_vec_rx) {
1544 			PMD_INIT_LOG(INFO,
1545 				"virtio: using packed ring vectorized Rx path on port %u",
1546 				eth_dev->data->port_id);
1547 			eth_dev->rx_pkt_burst =
1548 				&virtio_recv_pkts_packed_vec;
1549 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1550 			PMD_INIT_LOG(INFO,
1551 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1552 				eth_dev->data->port_id);
1553 			eth_dev->rx_pkt_burst =
1554 				&virtio_recv_mergeable_pkts_packed;
1555 		} else {
1556 			PMD_INIT_LOG(INFO,
1557 				"virtio: using packed ring standard Rx path on port %u",
1558 				eth_dev->data->port_id);
1559 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1560 		}
1561 	} else {
1562 		if (hw->use_vec_rx) {
1563 			PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1564 				eth_dev->data->port_id);
1565 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1566 		} else if (hw->use_inorder_rx) {
1567 			PMD_INIT_LOG(INFO,
1568 				"virtio: using inorder Rx path on port %u",
1569 				eth_dev->data->port_id);
1570 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1571 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1572 			PMD_INIT_LOG(INFO,
1573 				"virtio: using mergeable buffer Rx path on port %u",
1574 				eth_dev->data->port_id);
1575 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1576 		} else {
1577 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1578 				eth_dev->data->port_id);
1579 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1580 		}
1581 	}
1582 
1583 }
1584 
1585 /* Only support 1:1 queue/interrupt mapping so far.
1586  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1587  * interrupt vectors (<N+1).
1588  */
1589 static int
1590 virtio_queues_bind_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 binding");
1596 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1597 		dev->intr_handle->intr_vec[i] = i + 1;
1598 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1599 						 VIRTIO_MSI_NO_VECTOR) {
1600 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1601 			return -EBUSY;
1602 		}
1603 	}
1604 
1605 	return 0;
1606 }
1607 
1608 static void
1609 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1610 {
1611 	uint32_t i;
1612 	struct virtio_hw *hw = dev->data->dev_private;
1613 
1614 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1615 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1616 		VTPCI_OPS(hw)->set_queue_irq(hw,
1617 					     hw->vqs[i * VTNET_CQ],
1618 					     VIRTIO_MSI_NO_VECTOR);
1619 }
1620 
1621 static int
1622 virtio_configure_intr(struct rte_eth_dev *dev)
1623 {
1624 	struct virtio_hw *hw = dev->data->dev_private;
1625 
1626 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1627 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1628 		return -ENOTSUP;
1629 	}
1630 
1631 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1632 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1633 		return -1;
1634 	}
1635 
1636 	if (!dev->intr_handle->intr_vec) {
1637 		dev->intr_handle->intr_vec =
1638 			rte_zmalloc("intr_vec",
1639 				    hw->max_queue_pairs * sizeof(int), 0);
1640 		if (!dev->intr_handle->intr_vec) {
1641 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1642 				     hw->max_queue_pairs);
1643 			return -ENOMEM;
1644 		}
1645 	}
1646 
1647 	/* Re-register callback to update max_intr */
1648 	rte_intr_callback_unregister(dev->intr_handle,
1649 				     virtio_interrupt_handler,
1650 				     dev);
1651 	rte_intr_callback_register(dev->intr_handle,
1652 				   virtio_interrupt_handler,
1653 				   dev);
1654 
1655 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1656 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1657 	 * And for legacy devices, this should be done before queue/vec binding
1658 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1659 	 * (22) will be ignored.
1660 	 */
1661 	if (virtio_intr_enable(dev) < 0) {
1662 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1663 		return -1;
1664 	}
1665 
1666 	if (virtio_queues_bind_intr(dev) < 0) {
1667 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1668 		return -1;
1669 	}
1670 
1671 	return 0;
1672 }
1673 #define DUPLEX_UNKNOWN   0xff
1674 /* reset device and renegotiate features if needed */
1675 static int
1676 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1677 {
1678 	struct virtio_hw *hw = eth_dev->data->dev_private;
1679 	struct virtio_net_config *config;
1680 	struct virtio_net_config local_config;
1681 	struct rte_pci_device *pci_dev = NULL;
1682 	int ret;
1683 
1684 	/* Reset the device although not necessary at startup */
1685 	vtpci_reset(hw);
1686 
1687 	if (hw->vqs) {
1688 		virtio_dev_free_mbufs(eth_dev);
1689 		virtio_free_queues(hw);
1690 	}
1691 
1692 	/* Tell the host we've noticed this device. */
1693 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1694 
1695 	/* Tell the host we've known how to drive the device. */
1696 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1697 	if (virtio_negotiate_features(hw, req_features) < 0)
1698 		return -1;
1699 
1700 	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1701 
1702 	if (!hw->virtio_user_dev)
1703 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1704 
1705 	/* If host does not support both status and MSI-X then disable LSC */
1706 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1707 	    hw->use_msix != VIRTIO_MSIX_NONE)
1708 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1709 	else
1710 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1711 
1712 	/* Setting up rx_header size for the device */
1713 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1714 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1715 	    vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1716 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1717 	else
1718 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1719 
1720 	/* Copy the permanent MAC address to: virtio_hw */
1721 	virtio_get_hwaddr(hw);
1722 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1723 			&eth_dev->data->mac_addrs[0]);
1724 	PMD_INIT_LOG(DEBUG,
1725 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1726 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1727 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1728 
1729 	if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1730 		if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1731 			config = &local_config;
1732 			vtpci_read_dev_config(hw,
1733 				offsetof(struct virtio_net_config, speed),
1734 				&config->speed, sizeof(config->speed));
1735 			vtpci_read_dev_config(hw,
1736 				offsetof(struct virtio_net_config, duplex),
1737 				&config->duplex, sizeof(config->duplex));
1738 			hw->speed = config->speed;
1739 			hw->duplex = config->duplex;
1740 		}
1741 	}
1742 	if (hw->duplex == DUPLEX_UNKNOWN)
1743 		hw->duplex = ETH_LINK_FULL_DUPLEX;
1744 	PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1745 		hw->speed, hw->duplex);
1746 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1747 		config = &local_config;
1748 
1749 		vtpci_read_dev_config(hw,
1750 			offsetof(struct virtio_net_config, mac),
1751 			&config->mac, sizeof(config->mac));
1752 
1753 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1754 			vtpci_read_dev_config(hw,
1755 				offsetof(struct virtio_net_config, status),
1756 				&config->status, sizeof(config->status));
1757 		} else {
1758 			PMD_INIT_LOG(DEBUG,
1759 				     "VIRTIO_NET_F_STATUS is not supported");
1760 			config->status = 0;
1761 		}
1762 
1763 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1764 			vtpci_read_dev_config(hw,
1765 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1766 				&config->max_virtqueue_pairs,
1767 				sizeof(config->max_virtqueue_pairs));
1768 		} else {
1769 			PMD_INIT_LOG(DEBUG,
1770 				     "VIRTIO_NET_F_MQ is not supported");
1771 			config->max_virtqueue_pairs = 1;
1772 		}
1773 
1774 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1775 
1776 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1777 			vtpci_read_dev_config(hw,
1778 				offsetof(struct virtio_net_config, mtu),
1779 				&config->mtu,
1780 				sizeof(config->mtu));
1781 
1782 			/*
1783 			 * MTU value has already been checked at negotiation
1784 			 * time, but check again in case it has changed since
1785 			 * then, which should not happen.
1786 			 */
1787 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1788 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1789 						config->mtu);
1790 				return -1;
1791 			}
1792 
1793 			hw->max_mtu = config->mtu;
1794 			/* Set initial MTU to maximum one supported by vhost */
1795 			eth_dev->data->mtu = config->mtu;
1796 
1797 		} else {
1798 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1799 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1800 		}
1801 
1802 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1803 				config->max_virtqueue_pairs);
1804 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1805 		PMD_INIT_LOG(DEBUG,
1806 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1807 				config->mac[0], config->mac[1],
1808 				config->mac[2], config->mac[3],
1809 				config->mac[4], config->mac[5]);
1810 	} else {
1811 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1812 		hw->max_queue_pairs = 1;
1813 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1814 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1815 	}
1816 
1817 	ret = virtio_alloc_queues(eth_dev);
1818 	if (ret < 0)
1819 		return ret;
1820 
1821 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1822 		if (virtio_configure_intr(eth_dev) < 0) {
1823 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1824 			virtio_free_queues(hw);
1825 			return -1;
1826 		}
1827 	}
1828 
1829 	vtpci_reinit_complete(hw);
1830 
1831 	if (pci_dev)
1832 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1833 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1834 			pci_dev->id.device_id);
1835 
1836 	return 0;
1837 }
1838 
1839 /*
1840  * Remap the PCI device again (IO port map for legacy device and
1841  * memory map for modern device), so that the secondary process
1842  * could have the PCI initiated correctly.
1843  */
1844 static int
1845 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1846 {
1847 	if (hw->modern) {
1848 		/*
1849 		 * We don't have to re-parse the PCI config space, since
1850 		 * rte_pci_map_device() makes sure the mapped address
1851 		 * in secondary process would equal to the one mapped in
1852 		 * the primary process: error will be returned if that
1853 		 * requirement is not met.
1854 		 *
1855 		 * That said, we could simply reuse all cap pointers
1856 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1857 		 * primary process, which is stored in shared memory.
1858 		 */
1859 		if (rte_pci_map_device(pci_dev)) {
1860 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1861 			return -1;
1862 		}
1863 	} else {
1864 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1865 			return -1;
1866 	}
1867 
1868 	return 0;
1869 }
1870 
1871 static void
1872 virtio_set_vtpci_ops(struct virtio_hw *hw)
1873 {
1874 #ifdef RTE_VIRTIO_USER
1875 	if (hw->virtio_user_dev)
1876 		VTPCI_OPS(hw) = &virtio_user_ops;
1877 	else
1878 #endif
1879 	if (hw->modern)
1880 		VTPCI_OPS(hw) = &modern_ops;
1881 	else
1882 		VTPCI_OPS(hw) = &legacy_ops;
1883 }
1884 
1885 /*
1886  * This function is based on probe() function in virtio_pci.c
1887  * It returns 0 on success.
1888  */
1889 int
1890 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1891 {
1892 	struct virtio_hw *hw = eth_dev->data->dev_private;
1893 	uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1894 	int vectorized = 0;
1895 	int ret;
1896 
1897 	if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1898 		PMD_INIT_LOG(ERR,
1899 			"Not sufficient headroom required = %d, avail = %d",
1900 			(int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1901 			RTE_PKTMBUF_HEADROOM);
1902 
1903 		return -1;
1904 	}
1905 
1906 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1907 	eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1908 
1909 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1910 		if (!hw->virtio_user_dev) {
1911 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1912 			if (ret)
1913 				return ret;
1914 		}
1915 
1916 		virtio_set_vtpci_ops(hw);
1917 		set_rxtx_funcs(eth_dev);
1918 
1919 		return 0;
1920 	}
1921 	ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
1922 		 NULL, &speed, &vectorized);
1923 	if (ret < 0)
1924 		return ret;
1925 	hw->speed = speed;
1926 
1927 	/* Allocate memory for storing MAC addresses */
1928 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1929 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1930 	if (eth_dev->data->mac_addrs == NULL) {
1931 		PMD_INIT_LOG(ERR,
1932 			"Failed to allocate %d bytes needed to store MAC addresses",
1933 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1934 		return -ENOMEM;
1935 	}
1936 
1937 	hw->port_id = eth_dev->data->port_id;
1938 	/* For virtio_user case the hw->virtio_user_dev is populated by
1939 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1940 	 */
1941 	if (!hw->virtio_user_dev) {
1942 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1943 		if (ret)
1944 			goto err_vtpci_init;
1945 	}
1946 
1947 	rte_spinlock_init(&hw->state_lock);
1948 
1949 	/* reset device and negotiate default features */
1950 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1951 	if (ret < 0)
1952 		goto err_virtio_init;
1953 
1954 	if (vectorized) {
1955 		if (!vtpci_packed_queue(hw)) {
1956 			hw->use_vec_rx = 1;
1957 		} else {
1958 #if !defined(CC_AVX512_SUPPORT)
1959 			PMD_DRV_LOG(INFO,
1960 				"building environment do not support packed ring vectorized");
1961 #else
1962 			hw->use_vec_rx = 1;
1963 			hw->use_vec_tx = 1;
1964 #endif
1965 		}
1966 	}
1967 
1968 	hw->opened = true;
1969 
1970 	return 0;
1971 
1972 err_virtio_init:
1973 	if (!hw->virtio_user_dev) {
1974 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1975 		if (!hw->modern)
1976 			rte_pci_ioport_unmap(VTPCI_IO(hw));
1977 	}
1978 err_vtpci_init:
1979 	rte_free(eth_dev->data->mac_addrs);
1980 	eth_dev->data->mac_addrs = NULL;
1981 	return ret;
1982 }
1983 
1984 static int
1985 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1986 {
1987 	PMD_INIT_FUNC_TRACE();
1988 
1989 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1990 		return 0;
1991 
1992 	virtio_dev_stop(eth_dev);
1993 	virtio_dev_close(eth_dev);
1994 
1995 	eth_dev->dev_ops = NULL;
1996 	eth_dev->tx_pkt_burst = NULL;
1997 	eth_dev->rx_pkt_burst = NULL;
1998 
1999 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
2000 
2001 	return 0;
2002 }
2003 
2004 
2005 static int vdpa_check_handler(__rte_unused const char *key,
2006 		const char *value, void *ret_val)
2007 {
2008 	if (strcmp(value, "1") == 0)
2009 		*(int *)ret_val = 1;
2010 	else
2011 		*(int *)ret_val = 0;
2012 
2013 	return 0;
2014 }
2015 
2016 
2017 static uint32_t
2018 virtio_dev_speed_capa_get(uint32_t speed)
2019 {
2020 	switch (speed) {
2021 	case ETH_SPEED_NUM_10G:
2022 		return ETH_LINK_SPEED_10G;
2023 	case ETH_SPEED_NUM_20G:
2024 		return ETH_LINK_SPEED_20G;
2025 	case ETH_SPEED_NUM_25G:
2026 		return ETH_LINK_SPEED_25G;
2027 	case ETH_SPEED_NUM_40G:
2028 		return ETH_LINK_SPEED_40G;
2029 	case ETH_SPEED_NUM_50G:
2030 		return ETH_LINK_SPEED_50G;
2031 	case ETH_SPEED_NUM_56G:
2032 		return ETH_LINK_SPEED_56G;
2033 	case ETH_SPEED_NUM_100G:
2034 		return ETH_LINK_SPEED_100G;
2035 	case ETH_SPEED_NUM_200G:
2036 		return ETH_LINK_SPEED_200G;
2037 	default:
2038 		return 0;
2039 	}
2040 }
2041 
2042 static int vectorized_check_handler(__rte_unused const char *key,
2043 		const char *value, void *ret_val)
2044 {
2045 	if (strcmp(value, "1") == 0)
2046 		*(int *)ret_val = 1;
2047 	else
2048 		*(int *)ret_val = 0;
2049 
2050 	return 0;
2051 }
2052 
2053 #define VIRTIO_ARG_SPEED      "speed"
2054 #define VIRTIO_ARG_VDPA       "vdpa"
2055 #define VIRTIO_ARG_VECTORIZED "vectorized"
2056 
2057 
2058 static int
2059 link_speed_handler(const char *key __rte_unused,
2060 		const char *value, void *ret_val)
2061 {
2062 	uint32_t val;
2063 	if (!value || !ret_val)
2064 		return -EINVAL;
2065 	val = strtoul(value, NULL, 0);
2066 	/* validate input */
2067 	if (virtio_dev_speed_capa_get(val) == 0)
2068 		return -EINVAL;
2069 	*(uint32_t *)ret_val = val;
2070 
2071 	return 0;
2072 }
2073 
2074 
2075 static int
2076 virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
2077 	uint32_t *speed, int *vectorized)
2078 {
2079 	struct rte_kvargs *kvlist;
2080 	int ret = 0;
2081 
2082 	if (devargs == NULL)
2083 		return 0;
2084 
2085 	kvlist = rte_kvargs_parse(devargs->args, NULL);
2086 	if (kvlist == NULL) {
2087 		PMD_INIT_LOG(ERR, "error when parsing param");
2088 		return 0;
2089 	}
2090 	if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
2091 		/* vdpa mode selected when there's a key-value pair:
2092 		 * vdpa=1
2093 		 */
2094 		ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
2095 				vdpa_check_handler, vdpa);
2096 		if (ret < 0) {
2097 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2098 				VIRTIO_ARG_VDPA);
2099 			goto exit;
2100 		}
2101 	}
2102 	if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2103 		ret = rte_kvargs_process(kvlist,
2104 					VIRTIO_ARG_SPEED,
2105 					link_speed_handler, speed);
2106 		if (ret < 0) {
2107 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2108 					VIRTIO_ARG_SPEED);
2109 			goto exit;
2110 		}
2111 	}
2112 
2113 	if (vectorized &&
2114 		rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2115 		ret = rte_kvargs_process(kvlist,
2116 				VIRTIO_ARG_VECTORIZED,
2117 				vectorized_check_handler, vectorized);
2118 		if (ret < 0) {
2119 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2120 					VIRTIO_ARG_VECTORIZED);
2121 			goto exit;
2122 		}
2123 	}
2124 
2125 exit:
2126 	rte_kvargs_free(kvlist);
2127 	return ret;
2128 }
2129 
2130 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2131 	struct rte_pci_device *pci_dev)
2132 {
2133 	int vdpa = 0;
2134 	int ret = 0;
2135 
2136 	ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL,
2137 		NULL);
2138 	if (ret < 0) {
2139 		PMD_INIT_LOG(ERR, "devargs parsing is failed");
2140 		return ret;
2141 	}
2142 	/* virtio pmd skips probe if device needs to work in vdpa mode */
2143 	if (vdpa == 1)
2144 		return 1;
2145 
2146 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2147 		eth_virtio_dev_init);
2148 }
2149 
2150 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2151 {
2152 	int ret;
2153 
2154 	ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2155 	/* Port has already been released by close. */
2156 	if (ret == -ENODEV)
2157 		ret = 0;
2158 	return ret;
2159 }
2160 
2161 static struct rte_pci_driver rte_virtio_pmd = {
2162 	.driver = {
2163 		.name = "net_virtio",
2164 	},
2165 	.id_table = pci_id_virtio_map,
2166 	.drv_flags = 0,
2167 	.probe = eth_virtio_pci_probe,
2168 	.remove = eth_virtio_pci_remove,
2169 };
2170 
2171 RTE_INIT(rte_virtio_pmd_init)
2172 {
2173 	rte_eal_iopl_init();
2174 	rte_pci_register(&rte_virtio_pmd);
2175 }
2176 
2177 static bool
2178 rx_offload_enabled(struct virtio_hw *hw)
2179 {
2180 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2181 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2182 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2183 }
2184 
2185 static bool
2186 tx_offload_enabled(struct virtio_hw *hw)
2187 {
2188 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2189 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2190 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2191 }
2192 
2193 /*
2194  * Configure virtio device
2195  * It returns 0 on success.
2196  */
2197 static int
2198 virtio_dev_configure(struct rte_eth_dev *dev)
2199 {
2200 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2201 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2202 	struct virtio_hw *hw = dev->data->dev_private;
2203 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2204 		hw->vtnet_hdr_size;
2205 	uint64_t rx_offloads = rxmode->offloads;
2206 	uint64_t tx_offloads = txmode->offloads;
2207 	uint64_t req_features;
2208 	int ret;
2209 
2210 	PMD_INIT_LOG(DEBUG, "configure");
2211 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2212 
2213 	if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2214 		PMD_DRV_LOG(ERR,
2215 			"Unsupported Rx multi queue mode %d",
2216 			rxmode->mq_mode);
2217 		return -EINVAL;
2218 	}
2219 
2220 	if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2221 		PMD_DRV_LOG(ERR,
2222 			"Unsupported Tx multi queue mode %d",
2223 			txmode->mq_mode);
2224 		return -EINVAL;
2225 	}
2226 
2227 	if (dev->data->dev_conf.intr_conf.rxq) {
2228 		ret = virtio_init_device(dev, hw->req_guest_features);
2229 		if (ret < 0)
2230 			return ret;
2231 	}
2232 
2233 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2234 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2235 
2236 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2237 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2238 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2239 
2240 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2241 		req_features |=
2242 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2243 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2244 
2245 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2246 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2247 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2248 
2249 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2250 		req_features |=
2251 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2252 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2253 
2254 	/* if request features changed, reinit the device */
2255 	if (req_features != hw->req_guest_features) {
2256 		ret = virtio_init_device(dev, req_features);
2257 		if (ret < 0)
2258 			return ret;
2259 	}
2260 
2261 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2262 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2263 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2264 		PMD_DRV_LOG(ERR,
2265 			"rx checksum not available on this host");
2266 		return -ENOTSUP;
2267 	}
2268 
2269 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2270 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2271 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2272 		PMD_DRV_LOG(ERR,
2273 			"Large Receive Offload not available on this host");
2274 		return -ENOTSUP;
2275 	}
2276 
2277 	/* start control queue */
2278 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2279 		virtio_dev_cq_start(dev);
2280 
2281 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2282 		hw->vlan_strip = 1;
2283 
2284 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2285 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2286 		PMD_DRV_LOG(ERR,
2287 			    "vlan filtering not available on this host");
2288 		return -ENOTSUP;
2289 	}
2290 
2291 	hw->has_tx_offload = tx_offload_enabled(hw);
2292 	hw->has_rx_offload = rx_offload_enabled(hw);
2293 
2294 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2295 		/* Enable vector (0) for Link State Intrerrupt */
2296 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2297 				VIRTIO_MSI_NO_VECTOR) {
2298 			PMD_DRV_LOG(ERR, "failed to set config vector");
2299 			return -EBUSY;
2300 		}
2301 
2302 	if (vtpci_packed_queue(hw)) {
2303 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2304 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2305 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2306 		     !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2307 		     !vtpci_with_feature(hw, VIRTIO_F_VERSION_1))) {
2308 			PMD_DRV_LOG(INFO,
2309 				"disabled packed ring vectorized path for requirements not met");
2310 			hw->use_vec_rx = 0;
2311 			hw->use_vec_tx = 0;
2312 		}
2313 #else
2314 		hw->use_vec_rx = 0;
2315 		hw->use_vec_tx = 0;
2316 #endif
2317 
2318 		if (hw->use_vec_rx) {
2319 			if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2320 				PMD_DRV_LOG(INFO,
2321 					"disabled packed ring vectorized rx for mrg_rxbuf enabled");
2322 				hw->use_vec_rx = 0;
2323 			}
2324 
2325 			if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2326 				PMD_DRV_LOG(INFO,
2327 					"disabled packed ring vectorized rx for TCP_LRO enabled");
2328 				hw->use_vec_rx = 0;
2329 			}
2330 		}
2331 	} else {
2332 		if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2333 			hw->use_inorder_tx = 1;
2334 			hw->use_inorder_rx = 1;
2335 			hw->use_vec_rx = 0;
2336 		}
2337 
2338 		if (hw->use_vec_rx) {
2339 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2340 			if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2341 				PMD_DRV_LOG(INFO,
2342 					"disabled split ring vectorized path for requirement not met");
2343 				hw->use_vec_rx = 0;
2344 			}
2345 #endif
2346 			if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2347 				PMD_DRV_LOG(INFO,
2348 					"disabled split ring vectorized rx for mrg_rxbuf enabled");
2349 				hw->use_vec_rx = 0;
2350 			}
2351 
2352 			if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2353 					   DEV_RX_OFFLOAD_TCP_CKSUM |
2354 					   DEV_RX_OFFLOAD_TCP_LRO |
2355 					   DEV_RX_OFFLOAD_VLAN_STRIP)) {
2356 				PMD_DRV_LOG(INFO,
2357 					"disabled split ring vectorized rx for offloading enabled");
2358 				hw->use_vec_rx = 0;
2359 			}
2360 		}
2361 	}
2362 
2363 	return 0;
2364 }
2365 
2366 
2367 static int
2368 virtio_dev_start(struct rte_eth_dev *dev)
2369 {
2370 	uint16_t nb_queues, i;
2371 	struct virtnet_rx *rxvq;
2372 	struct virtnet_tx *txvq __rte_unused;
2373 	struct virtio_hw *hw = dev->data->dev_private;
2374 	int ret;
2375 
2376 	/* Finish the initialization of the queues */
2377 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2378 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2379 		if (ret < 0)
2380 			return ret;
2381 	}
2382 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2383 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2384 		if (ret < 0)
2385 			return ret;
2386 	}
2387 
2388 	/* check if lsc interrupt feature is enabled */
2389 	if (dev->data->dev_conf.intr_conf.lsc) {
2390 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2391 			PMD_DRV_LOG(ERR, "link status not supported by host");
2392 			return -ENOTSUP;
2393 		}
2394 	}
2395 
2396 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2397 	 * in device configure, but it could be unmapped  when device is
2398 	 * stopped.
2399 	 */
2400 	if (dev->data->dev_conf.intr_conf.lsc ||
2401 	    dev->data->dev_conf.intr_conf.rxq) {
2402 		virtio_intr_disable(dev);
2403 
2404 		/* Setup interrupt callback  */
2405 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2406 			rte_intr_callback_register(dev->intr_handle,
2407 						   virtio_interrupt_handler,
2408 						   dev);
2409 
2410 		if (virtio_intr_enable(dev) < 0) {
2411 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2412 			return -EIO;
2413 		}
2414 	}
2415 
2416 	/*Notify the backend
2417 	 *Otherwise the tap backend might already stop its queue due to fullness.
2418 	 *vhost backend will have no chance to be waked up
2419 	 */
2420 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2421 	if (hw->max_queue_pairs > 1) {
2422 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2423 			return -EINVAL;
2424 	}
2425 
2426 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2427 
2428 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2429 		rxvq = dev->data->rx_queues[i];
2430 		/* Flush the old packets */
2431 		virtqueue_rxvq_flush(rxvq->vq);
2432 		virtqueue_notify(rxvq->vq);
2433 	}
2434 
2435 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2436 		txvq = dev->data->tx_queues[i];
2437 		virtqueue_notify(txvq->vq);
2438 	}
2439 
2440 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2441 
2442 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2443 		rxvq = dev->data->rx_queues[i];
2444 		VIRTQUEUE_DUMP(rxvq->vq);
2445 	}
2446 
2447 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2448 		txvq = dev->data->tx_queues[i];
2449 		VIRTQUEUE_DUMP(txvq->vq);
2450 	}
2451 
2452 	set_rxtx_funcs(dev);
2453 	hw->started = true;
2454 
2455 	/* Initialize Link state */
2456 	virtio_dev_link_update(dev, 0);
2457 
2458 	return 0;
2459 }
2460 
2461 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2462 {
2463 	struct virtio_hw *hw = dev->data->dev_private;
2464 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2465 	const char *type __rte_unused;
2466 	unsigned int i, mbuf_num = 0;
2467 	struct virtqueue *vq;
2468 	struct rte_mbuf *buf;
2469 	int queue_type;
2470 
2471 	if (hw->vqs == NULL)
2472 		return;
2473 
2474 	for (i = 0; i < nr_vq; i++) {
2475 		vq = hw->vqs[i];
2476 		if (!vq)
2477 			continue;
2478 
2479 		queue_type = virtio_get_queue_type(hw, i);
2480 		if (queue_type == VTNET_RQ)
2481 			type = "rxq";
2482 		else if (queue_type == VTNET_TQ)
2483 			type = "txq";
2484 		else
2485 			continue;
2486 
2487 		PMD_INIT_LOG(DEBUG,
2488 			"Before freeing %s[%d] used and unused buf",
2489 			type, i);
2490 		VIRTQUEUE_DUMP(vq);
2491 
2492 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2493 			rte_pktmbuf_free(buf);
2494 			mbuf_num++;
2495 		}
2496 
2497 		PMD_INIT_LOG(DEBUG,
2498 			"After freeing %s[%d] used and unused buf",
2499 			type, i);
2500 		VIRTQUEUE_DUMP(vq);
2501 	}
2502 
2503 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2504 }
2505 
2506 /*
2507  * Stop device: disable interrupt and mark link down
2508  */
2509 static void
2510 virtio_dev_stop(struct rte_eth_dev *dev)
2511 {
2512 	struct virtio_hw *hw = dev->data->dev_private;
2513 	struct rte_eth_link link;
2514 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2515 
2516 	PMD_INIT_LOG(DEBUG, "stop");
2517 
2518 	rte_spinlock_lock(&hw->state_lock);
2519 	if (!hw->started)
2520 		goto out_unlock;
2521 	hw->started = false;
2522 
2523 	if (intr_conf->lsc || intr_conf->rxq) {
2524 		virtio_intr_disable(dev);
2525 
2526 		/* Reset interrupt callback  */
2527 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2528 			rte_intr_callback_unregister(dev->intr_handle,
2529 						     virtio_interrupt_handler,
2530 						     dev);
2531 		}
2532 	}
2533 
2534 	memset(&link, 0, sizeof(link));
2535 	rte_eth_linkstatus_set(dev, &link);
2536 out_unlock:
2537 	rte_spinlock_unlock(&hw->state_lock);
2538 }
2539 
2540 static int
2541 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2542 {
2543 	struct rte_eth_link link;
2544 	uint16_t status;
2545 	struct virtio_hw *hw = dev->data->dev_private;
2546 
2547 	memset(&link, 0, sizeof(link));
2548 	link.link_duplex = hw->duplex;
2549 	link.link_speed  = hw->speed;
2550 	link.link_autoneg = ETH_LINK_AUTONEG;
2551 
2552 	if (!hw->started) {
2553 		link.link_status = ETH_LINK_DOWN;
2554 		link.link_speed = ETH_SPEED_NUM_NONE;
2555 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2556 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2557 		vtpci_read_dev_config(hw,
2558 				offsetof(struct virtio_net_config, status),
2559 				&status, sizeof(status));
2560 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2561 			link.link_status = ETH_LINK_DOWN;
2562 			link.link_speed = ETH_SPEED_NUM_NONE;
2563 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2564 				     dev->data->port_id);
2565 		} else {
2566 			link.link_status = ETH_LINK_UP;
2567 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2568 				     dev->data->port_id);
2569 		}
2570 	} else {
2571 		link.link_status = ETH_LINK_UP;
2572 	}
2573 
2574 	return rte_eth_linkstatus_set(dev, &link);
2575 }
2576 
2577 static int
2578 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2579 {
2580 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2581 	struct virtio_hw *hw = dev->data->dev_private;
2582 	uint64_t offloads = rxmode->offloads;
2583 
2584 	if (mask & ETH_VLAN_FILTER_MASK) {
2585 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2586 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2587 
2588 			PMD_DRV_LOG(NOTICE,
2589 				"vlan filtering not available on this host");
2590 
2591 			return -ENOTSUP;
2592 		}
2593 	}
2594 
2595 	if (mask & ETH_VLAN_STRIP_MASK)
2596 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2597 
2598 	return 0;
2599 }
2600 
2601 static int
2602 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2603 {
2604 	uint64_t tso_mask, host_features;
2605 	struct virtio_hw *hw = dev->data->dev_private;
2606 	dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2607 
2608 	dev_info->max_rx_queues =
2609 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2610 	dev_info->max_tx_queues =
2611 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2612 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2613 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2614 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2615 
2616 	host_features = VTPCI_OPS(hw)->get_features(hw);
2617 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2618 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2619 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2620 		dev_info->rx_offload_capa |=
2621 			DEV_RX_OFFLOAD_TCP_CKSUM |
2622 			DEV_RX_OFFLOAD_UDP_CKSUM;
2623 	}
2624 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2625 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2626 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2627 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2628 	if ((host_features & tso_mask) == tso_mask)
2629 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2630 
2631 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2632 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2633 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2634 		dev_info->tx_offload_capa |=
2635 			DEV_TX_OFFLOAD_UDP_CKSUM |
2636 			DEV_TX_OFFLOAD_TCP_CKSUM;
2637 	}
2638 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2639 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2640 	if ((host_features & tso_mask) == tso_mask)
2641 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2642 
2643 	return 0;
2644 }
2645 
2646 /*
2647  * It enables testpmd to collect per queue stats.
2648  */
2649 static int
2650 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2651 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2652 __rte_unused uint8_t is_rx)
2653 {
2654 	return 0;
2655 }
2656 
2657 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2658 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2659 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2660 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2661 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);
2662