xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 10b71caecbe1cddcbb65c050ca775fba575e88db)
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 void
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 
715 	if (!hw->opened)
716 		return;
717 	hw->opened = false;
718 
719 	/* reset the NIC */
720 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
721 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
722 	if (intr_conf->rxq)
723 		virtio_queues_unbind_intr(dev);
724 
725 	if (intr_conf->lsc || intr_conf->rxq) {
726 		virtio_intr_disable(dev);
727 		rte_intr_efd_disable(dev->intr_handle);
728 		rte_free(dev->intr_handle->intr_vec);
729 		dev->intr_handle->intr_vec = NULL;
730 	}
731 
732 	vtpci_reset(hw);
733 	virtio_dev_free_mbufs(dev);
734 	virtio_free_queues(hw);
735 
736 #ifdef RTE_VIRTIO_USER
737 	if (hw->virtio_user_dev)
738 		virtio_user_dev_uninit(hw->virtio_user_dev);
739 	else
740 #endif
741 	if (dev->device) {
742 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
743 		if (!hw->modern)
744 			rte_pci_ioport_unmap(VTPCI_IO(hw));
745 	}
746 }
747 
748 static int
749 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
750 {
751 	struct virtio_hw *hw = dev->data->dev_private;
752 	struct virtio_pmd_ctrl ctrl;
753 	int dlen[1];
754 	int ret;
755 
756 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
757 		PMD_INIT_LOG(INFO, "host does not support rx control");
758 		return -ENOTSUP;
759 	}
760 
761 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
762 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
763 	ctrl.data[0] = 1;
764 	dlen[0] = 1;
765 
766 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
767 	if (ret) {
768 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
769 		return -EAGAIN;
770 	}
771 
772 	return 0;
773 }
774 
775 static int
776 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
777 {
778 	struct virtio_hw *hw = dev->data->dev_private;
779 	struct virtio_pmd_ctrl ctrl;
780 	int dlen[1];
781 	int ret;
782 
783 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
784 		PMD_INIT_LOG(INFO, "host does not support rx control");
785 		return -ENOTSUP;
786 	}
787 
788 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
789 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
790 	ctrl.data[0] = 0;
791 	dlen[0] = 1;
792 
793 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
794 	if (ret) {
795 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
796 		return -EAGAIN;
797 	}
798 
799 	return 0;
800 }
801 
802 static int
803 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
804 {
805 	struct virtio_hw *hw = dev->data->dev_private;
806 	struct virtio_pmd_ctrl ctrl;
807 	int dlen[1];
808 	int ret;
809 
810 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
811 		PMD_INIT_LOG(INFO, "host does not support rx control");
812 		return -ENOTSUP;
813 	}
814 
815 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
816 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
817 	ctrl.data[0] = 1;
818 	dlen[0] = 1;
819 
820 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
821 	if (ret) {
822 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
823 		return -EAGAIN;
824 	}
825 
826 	return 0;
827 }
828 
829 static int
830 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
831 {
832 	struct virtio_hw *hw = dev->data->dev_private;
833 	struct virtio_pmd_ctrl ctrl;
834 	int dlen[1];
835 	int ret;
836 
837 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
838 		PMD_INIT_LOG(INFO, "host does not support rx control");
839 		return -ENOTSUP;
840 	}
841 
842 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
843 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
844 	ctrl.data[0] = 0;
845 	dlen[0] = 1;
846 
847 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
848 	if (ret) {
849 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
850 		return -EAGAIN;
851 	}
852 
853 	return 0;
854 }
855 
856 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
857 static int
858 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
859 {
860 	struct virtio_hw *hw = dev->data->dev_private;
861 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
862 				 hw->vtnet_hdr_size;
863 	uint32_t frame_size = mtu + ether_hdr_len;
864 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
865 
866 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
867 
868 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
869 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
870 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
871 		return -EINVAL;
872 	}
873 	return 0;
874 }
875 
876 static int
877 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
878 {
879 	struct virtio_hw *hw = dev->data->dev_private;
880 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
881 	struct virtqueue *vq = rxvq->vq;
882 
883 	virtqueue_enable_intr(vq);
884 	virtio_mb(hw->weak_barriers);
885 	return 0;
886 }
887 
888 static int
889 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
890 {
891 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
892 	struct virtqueue *vq = rxvq->vq;
893 
894 	virtqueue_disable_intr(vq);
895 	return 0;
896 }
897 
898 /*
899  * dev_ops for virtio, bare necessities for basic operation
900  */
901 static const struct eth_dev_ops virtio_eth_dev_ops = {
902 	.dev_configure           = virtio_dev_configure,
903 	.dev_start               = virtio_dev_start,
904 	.dev_stop                = virtio_dev_stop,
905 	.dev_close               = virtio_dev_close,
906 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
907 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
908 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
909 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
910 	.mtu_set                 = virtio_mtu_set,
911 	.dev_infos_get           = virtio_dev_info_get,
912 	.stats_get               = virtio_dev_stats_get,
913 	.xstats_get              = virtio_dev_xstats_get,
914 	.xstats_get_names        = virtio_dev_xstats_get_names,
915 	.stats_reset             = virtio_dev_stats_reset,
916 	.xstats_reset            = virtio_dev_stats_reset,
917 	.link_update             = virtio_dev_link_update,
918 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
919 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
920 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
921 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
922 	.rx_queue_release        = virtio_dev_queue_release,
923 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
924 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
925 	.tx_queue_release        = virtio_dev_queue_release,
926 	/* collect stats per queue */
927 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
928 	.vlan_filter_set         = virtio_vlan_filter_set,
929 	.mac_addr_add            = virtio_mac_addr_add,
930 	.mac_addr_remove         = virtio_mac_addr_remove,
931 	.mac_addr_set            = virtio_mac_addr_set,
932 };
933 
934 /*
935  * dev_ops for virtio-user in secondary processes, as we just have
936  * some limited supports currently.
937  */
938 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
939 	.dev_infos_get           = virtio_dev_info_get,
940 	.stats_get               = virtio_dev_stats_get,
941 	.xstats_get              = virtio_dev_xstats_get,
942 	.xstats_get_names        = virtio_dev_xstats_get_names,
943 	.stats_reset             = virtio_dev_stats_reset,
944 	.xstats_reset            = virtio_dev_stats_reset,
945 	/* collect stats per queue */
946 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
947 };
948 
949 static void
950 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
951 {
952 	unsigned i;
953 
954 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
955 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
956 		if (txvq == NULL)
957 			continue;
958 
959 		stats->opackets += txvq->stats.packets;
960 		stats->obytes += txvq->stats.bytes;
961 
962 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
963 			stats->q_opackets[i] = txvq->stats.packets;
964 			stats->q_obytes[i] = txvq->stats.bytes;
965 		}
966 	}
967 
968 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
969 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
970 		if (rxvq == NULL)
971 			continue;
972 
973 		stats->ipackets += rxvq->stats.packets;
974 		stats->ibytes += rxvq->stats.bytes;
975 		stats->ierrors += rxvq->stats.errors;
976 
977 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
978 			stats->q_ipackets[i] = rxvq->stats.packets;
979 			stats->q_ibytes[i] = rxvq->stats.bytes;
980 		}
981 	}
982 
983 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
984 }
985 
986 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
987 				       struct rte_eth_xstat_name *xstats_names,
988 				       __rte_unused unsigned limit)
989 {
990 	unsigned i;
991 	unsigned count = 0;
992 	unsigned t;
993 
994 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
995 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
996 
997 	if (xstats_names != NULL) {
998 		/* Note: limit checked in rte_eth_xstats_names() */
999 
1000 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
1001 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1002 			if (rxvq == NULL)
1003 				continue;
1004 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1005 				snprintf(xstats_names[count].name,
1006 					sizeof(xstats_names[count].name),
1007 					"rx_q%u_%s", i,
1008 					rte_virtio_rxq_stat_strings[t].name);
1009 				count++;
1010 			}
1011 		}
1012 
1013 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
1014 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
1015 			if (txvq == NULL)
1016 				continue;
1017 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1018 				snprintf(xstats_names[count].name,
1019 					sizeof(xstats_names[count].name),
1020 					"tx_q%u_%s", i,
1021 					rte_virtio_txq_stat_strings[t].name);
1022 				count++;
1023 			}
1024 		}
1025 		return count;
1026 	}
1027 	return nstats;
1028 }
1029 
1030 static int
1031 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1032 		      unsigned n)
1033 {
1034 	unsigned i;
1035 	unsigned count = 0;
1036 
1037 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1038 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1039 
1040 	if (n < nstats)
1041 		return nstats;
1042 
1043 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1044 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1045 
1046 		if (rxvq == NULL)
1047 			continue;
1048 
1049 		unsigned t;
1050 
1051 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1052 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1053 				rte_virtio_rxq_stat_strings[t].offset);
1054 			xstats[count].id = count;
1055 			count++;
1056 		}
1057 	}
1058 
1059 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1060 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1061 
1062 		if (txvq == NULL)
1063 			continue;
1064 
1065 		unsigned t;
1066 
1067 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1068 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1069 				rte_virtio_txq_stat_strings[t].offset);
1070 			xstats[count].id = count;
1071 			count++;
1072 		}
1073 	}
1074 
1075 	return count;
1076 }
1077 
1078 static int
1079 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1080 {
1081 	virtio_update_stats(dev, stats);
1082 
1083 	return 0;
1084 }
1085 
1086 static int
1087 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1088 {
1089 	unsigned int i;
1090 
1091 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1092 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1093 		if (txvq == NULL)
1094 			continue;
1095 
1096 		txvq->stats.packets = 0;
1097 		txvq->stats.bytes = 0;
1098 		txvq->stats.multicast = 0;
1099 		txvq->stats.broadcast = 0;
1100 		memset(txvq->stats.size_bins, 0,
1101 		       sizeof(txvq->stats.size_bins[0]) * 8);
1102 	}
1103 
1104 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1105 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1106 		if (rxvq == NULL)
1107 			continue;
1108 
1109 		rxvq->stats.packets = 0;
1110 		rxvq->stats.bytes = 0;
1111 		rxvq->stats.errors = 0;
1112 		rxvq->stats.multicast = 0;
1113 		rxvq->stats.broadcast = 0;
1114 		memset(rxvq->stats.size_bins, 0,
1115 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1116 	}
1117 
1118 	return 0;
1119 }
1120 
1121 static void
1122 virtio_set_hwaddr(struct virtio_hw *hw)
1123 {
1124 	vtpci_write_dev_config(hw,
1125 			offsetof(struct virtio_net_config, mac),
1126 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1127 }
1128 
1129 static void
1130 virtio_get_hwaddr(struct virtio_hw *hw)
1131 {
1132 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1133 		vtpci_read_dev_config(hw,
1134 			offsetof(struct virtio_net_config, mac),
1135 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1136 	} else {
1137 		rte_eth_random_addr(&hw->mac_addr[0]);
1138 		virtio_set_hwaddr(hw);
1139 	}
1140 }
1141 
1142 static int
1143 virtio_mac_table_set(struct virtio_hw *hw,
1144 		     const struct virtio_net_ctrl_mac *uc,
1145 		     const struct virtio_net_ctrl_mac *mc)
1146 {
1147 	struct virtio_pmd_ctrl ctrl;
1148 	int err, len[2];
1149 
1150 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1151 		PMD_DRV_LOG(INFO, "host does not support mac table");
1152 		return -1;
1153 	}
1154 
1155 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1156 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1157 
1158 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1159 	memcpy(ctrl.data, uc, len[0]);
1160 
1161 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1162 	memcpy(ctrl.data + len[0], mc, len[1]);
1163 
1164 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1165 	if (err != 0)
1166 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1167 	return err;
1168 }
1169 
1170 static int
1171 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1172 		    uint32_t index, uint32_t vmdq __rte_unused)
1173 {
1174 	struct virtio_hw *hw = dev->data->dev_private;
1175 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1176 	unsigned int i;
1177 	struct virtio_net_ctrl_mac *uc, *mc;
1178 
1179 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1180 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1181 		return -EINVAL;
1182 	}
1183 
1184 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1185 		sizeof(uc->entries));
1186 	uc->entries = 0;
1187 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1188 		sizeof(mc->entries));
1189 	mc->entries = 0;
1190 
1191 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1192 		const struct rte_ether_addr *addr
1193 			= (i == index) ? mac_addr : addrs + i;
1194 		struct virtio_net_ctrl_mac *tbl
1195 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1196 
1197 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1198 	}
1199 
1200 	return virtio_mac_table_set(hw, uc, mc);
1201 }
1202 
1203 static void
1204 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1205 {
1206 	struct virtio_hw *hw = dev->data->dev_private;
1207 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1208 	struct virtio_net_ctrl_mac *uc, *mc;
1209 	unsigned int i;
1210 
1211 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1212 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1213 		return;
1214 	}
1215 
1216 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1217 		sizeof(uc->entries));
1218 	uc->entries = 0;
1219 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1220 		sizeof(mc->entries));
1221 	mc->entries = 0;
1222 
1223 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1224 		struct virtio_net_ctrl_mac *tbl;
1225 
1226 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1227 			continue;
1228 
1229 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1230 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1231 			RTE_ETHER_ADDR_LEN);
1232 	}
1233 
1234 	virtio_mac_table_set(hw, uc, mc);
1235 }
1236 
1237 static int
1238 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1239 {
1240 	struct virtio_hw *hw = dev->data->dev_private;
1241 
1242 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1243 
1244 	/* Use atomic update if available */
1245 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1246 		struct virtio_pmd_ctrl ctrl;
1247 		int len = RTE_ETHER_ADDR_LEN;
1248 
1249 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1250 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1251 
1252 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1253 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1254 	}
1255 
1256 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1257 		return -ENOTSUP;
1258 
1259 	virtio_set_hwaddr(hw);
1260 	return 0;
1261 }
1262 
1263 static int
1264 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1265 {
1266 	struct virtio_hw *hw = dev->data->dev_private;
1267 	struct virtio_pmd_ctrl ctrl;
1268 	int len;
1269 
1270 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1271 		return -ENOTSUP;
1272 
1273 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1274 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1275 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1276 	len = sizeof(vlan_id);
1277 
1278 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1279 }
1280 
1281 static int
1282 virtio_intr_unmask(struct rte_eth_dev *dev)
1283 {
1284 	struct virtio_hw *hw = dev->data->dev_private;
1285 
1286 	if (rte_intr_ack(dev->intr_handle) < 0)
1287 		return -1;
1288 
1289 	if (!hw->virtio_user_dev)
1290 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1291 
1292 	return 0;
1293 }
1294 
1295 static int
1296 virtio_intr_enable(struct rte_eth_dev *dev)
1297 {
1298 	struct virtio_hw *hw = dev->data->dev_private;
1299 
1300 	if (rte_intr_enable(dev->intr_handle) < 0)
1301 		return -1;
1302 
1303 	if (!hw->virtio_user_dev)
1304 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1305 
1306 	return 0;
1307 }
1308 
1309 static int
1310 virtio_intr_disable(struct rte_eth_dev *dev)
1311 {
1312 	struct virtio_hw *hw = dev->data->dev_private;
1313 
1314 	if (rte_intr_disable(dev->intr_handle) < 0)
1315 		return -1;
1316 
1317 	if (!hw->virtio_user_dev)
1318 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1319 
1320 	return 0;
1321 }
1322 
1323 static int
1324 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1325 {
1326 	uint64_t host_features;
1327 
1328 	/* Prepare guest_features: feature that driver wants to support */
1329 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1330 		req_features);
1331 
1332 	/* Read device(host) feature bits */
1333 	host_features = VTPCI_OPS(hw)->get_features(hw);
1334 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1335 		host_features);
1336 
1337 	/* If supported, ensure MTU value is valid before acknowledging it. */
1338 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1339 		struct virtio_net_config config;
1340 
1341 		vtpci_read_dev_config(hw,
1342 			offsetof(struct virtio_net_config, mtu),
1343 			&config.mtu, sizeof(config.mtu));
1344 
1345 		if (config.mtu < RTE_ETHER_MIN_MTU)
1346 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1347 	}
1348 
1349 	/*
1350 	 * Negotiate features: Subset of device feature bits are written back
1351 	 * guest feature bits.
1352 	 */
1353 	hw->guest_features = req_features;
1354 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1355 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1356 		hw->guest_features);
1357 
1358 	if (hw->modern) {
1359 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1360 			PMD_INIT_LOG(ERR,
1361 				"VIRTIO_F_VERSION_1 features is not enabled.");
1362 			return -1;
1363 		}
1364 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1365 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1366 			PMD_INIT_LOG(ERR,
1367 				"failed to set FEATURES_OK status!");
1368 			return -1;
1369 		}
1370 	}
1371 
1372 	hw->req_guest_features = req_features;
1373 
1374 	return 0;
1375 }
1376 
1377 int
1378 virtio_dev_pause(struct rte_eth_dev *dev)
1379 {
1380 	struct virtio_hw *hw = dev->data->dev_private;
1381 
1382 	rte_spinlock_lock(&hw->state_lock);
1383 
1384 	if (hw->started == 0) {
1385 		/* Device is just stopped. */
1386 		rte_spinlock_unlock(&hw->state_lock);
1387 		return -1;
1388 	}
1389 	hw->started = 0;
1390 	/*
1391 	 * Prevent the worker threads from touching queues to avoid contention,
1392 	 * 1 ms should be enough for the ongoing Tx function to finish.
1393 	 */
1394 	rte_delay_ms(1);
1395 	return 0;
1396 }
1397 
1398 /*
1399  * Recover hw state to let the worker threads continue.
1400  */
1401 void
1402 virtio_dev_resume(struct rte_eth_dev *dev)
1403 {
1404 	struct virtio_hw *hw = dev->data->dev_private;
1405 
1406 	hw->started = 1;
1407 	rte_spinlock_unlock(&hw->state_lock);
1408 }
1409 
1410 /*
1411  * Should be called only after device is paused.
1412  */
1413 int
1414 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1415 		int nb_pkts)
1416 {
1417 	struct virtio_hw *hw = dev->data->dev_private;
1418 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1419 	int ret;
1420 
1421 	hw->inject_pkts = tx_pkts;
1422 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1423 	hw->inject_pkts = NULL;
1424 
1425 	return ret;
1426 }
1427 
1428 static void
1429 virtio_notify_peers(struct rte_eth_dev *dev)
1430 {
1431 	struct virtio_hw *hw = dev->data->dev_private;
1432 	struct virtnet_rx *rxvq;
1433 	struct rte_mbuf *rarp_mbuf;
1434 
1435 	if (!dev->data->rx_queues)
1436 		return;
1437 
1438 	rxvq = dev->data->rx_queues[0];
1439 	if (!rxvq)
1440 		return;
1441 
1442 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1443 			(struct rte_ether_addr *)hw->mac_addr);
1444 	if (rarp_mbuf == NULL) {
1445 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1446 		return;
1447 	}
1448 
1449 	/* If virtio port just stopped, no need to send RARP */
1450 	if (virtio_dev_pause(dev) < 0) {
1451 		rte_pktmbuf_free(rarp_mbuf);
1452 		return;
1453 	}
1454 
1455 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1456 	virtio_dev_resume(dev);
1457 }
1458 
1459 static void
1460 virtio_ack_link_announce(struct rte_eth_dev *dev)
1461 {
1462 	struct virtio_hw *hw = dev->data->dev_private;
1463 	struct virtio_pmd_ctrl ctrl;
1464 
1465 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1466 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1467 
1468 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1469 }
1470 
1471 /*
1472  * Process virtio config changed interrupt. Call the callback
1473  * if link state changed, generate gratuitous RARP packet if
1474  * the status indicates an ANNOUNCE.
1475  */
1476 void
1477 virtio_interrupt_handler(void *param)
1478 {
1479 	struct rte_eth_dev *dev = param;
1480 	struct virtio_hw *hw = dev->data->dev_private;
1481 	uint8_t isr;
1482 	uint16_t status;
1483 
1484 	/* Read interrupt status which clears interrupt */
1485 	isr = vtpci_isr(hw);
1486 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1487 
1488 	if (virtio_intr_unmask(dev) < 0)
1489 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1490 
1491 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1492 		if (virtio_dev_link_update(dev, 0) == 0)
1493 			_rte_eth_dev_callback_process(dev,
1494 						      RTE_ETH_EVENT_INTR_LSC,
1495 						      NULL);
1496 
1497 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1498 			vtpci_read_dev_config(hw,
1499 				offsetof(struct virtio_net_config, status),
1500 				&status, sizeof(status));
1501 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1502 				virtio_notify_peers(dev);
1503 				if (hw->cvq)
1504 					virtio_ack_link_announce(dev);
1505 			}
1506 		}
1507 	}
1508 }
1509 
1510 /* set rx and tx handlers according to what is supported */
1511 static void
1512 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1513 {
1514 	struct virtio_hw *hw = eth_dev->data->dev_private;
1515 
1516 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1517 	if (vtpci_packed_queue(hw)) {
1518 		PMD_INIT_LOG(INFO,
1519 			"virtio: using packed ring %s Tx path on port %u",
1520 			hw->use_vec_tx ? "vectorized" : "standard",
1521 			eth_dev->data->port_id);
1522 		if (hw->use_vec_tx)
1523 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1524 		else
1525 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1526 	} else {
1527 		if (hw->use_inorder_tx) {
1528 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1529 				eth_dev->data->port_id);
1530 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1531 		} else {
1532 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1533 				eth_dev->data->port_id);
1534 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1535 		}
1536 	}
1537 
1538 	if (vtpci_packed_queue(hw)) {
1539 		if (hw->use_vec_rx) {
1540 			PMD_INIT_LOG(INFO,
1541 				"virtio: using packed ring vectorized Rx path on port %u",
1542 				eth_dev->data->port_id);
1543 			eth_dev->rx_pkt_burst =
1544 				&virtio_recv_pkts_packed_vec;
1545 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1546 			PMD_INIT_LOG(INFO,
1547 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1548 				eth_dev->data->port_id);
1549 			eth_dev->rx_pkt_burst =
1550 				&virtio_recv_mergeable_pkts_packed;
1551 		} else {
1552 			PMD_INIT_LOG(INFO,
1553 				"virtio: using packed ring standard Rx path on port %u",
1554 				eth_dev->data->port_id);
1555 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1556 		}
1557 	} else {
1558 		if (hw->use_vec_rx) {
1559 			PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1560 				eth_dev->data->port_id);
1561 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1562 		} else if (hw->use_inorder_rx) {
1563 			PMD_INIT_LOG(INFO,
1564 				"virtio: using inorder Rx path on port %u",
1565 				eth_dev->data->port_id);
1566 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1567 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1568 			PMD_INIT_LOG(INFO,
1569 				"virtio: using mergeable buffer Rx path on port %u",
1570 				eth_dev->data->port_id);
1571 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1572 		} else {
1573 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1574 				eth_dev->data->port_id);
1575 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1576 		}
1577 	}
1578 
1579 }
1580 
1581 /* Only support 1:1 queue/interrupt mapping so far.
1582  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1583  * interrupt vectors (<N+1).
1584  */
1585 static int
1586 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1587 {
1588 	uint32_t i;
1589 	struct virtio_hw *hw = dev->data->dev_private;
1590 
1591 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1592 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1593 		dev->intr_handle->intr_vec[i] = i + 1;
1594 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1595 						 VIRTIO_MSI_NO_VECTOR) {
1596 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1597 			return -EBUSY;
1598 		}
1599 	}
1600 
1601 	return 0;
1602 }
1603 
1604 static void
1605 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1606 {
1607 	uint32_t i;
1608 	struct virtio_hw *hw = dev->data->dev_private;
1609 
1610 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1611 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1612 		VTPCI_OPS(hw)->set_queue_irq(hw,
1613 					     hw->vqs[i * VTNET_CQ],
1614 					     VIRTIO_MSI_NO_VECTOR);
1615 }
1616 
1617 static int
1618 virtio_configure_intr(struct rte_eth_dev *dev)
1619 {
1620 	struct virtio_hw *hw = dev->data->dev_private;
1621 
1622 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1623 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1624 		return -ENOTSUP;
1625 	}
1626 
1627 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1628 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1629 		return -1;
1630 	}
1631 
1632 	if (!dev->intr_handle->intr_vec) {
1633 		dev->intr_handle->intr_vec =
1634 			rte_zmalloc("intr_vec",
1635 				    hw->max_queue_pairs * sizeof(int), 0);
1636 		if (!dev->intr_handle->intr_vec) {
1637 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1638 				     hw->max_queue_pairs);
1639 			return -ENOMEM;
1640 		}
1641 	}
1642 
1643 	/* Re-register callback to update max_intr */
1644 	rte_intr_callback_unregister(dev->intr_handle,
1645 				     virtio_interrupt_handler,
1646 				     dev);
1647 	rte_intr_callback_register(dev->intr_handle,
1648 				   virtio_interrupt_handler,
1649 				   dev);
1650 
1651 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1652 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1653 	 * And for legacy devices, this should be done before queue/vec binding
1654 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1655 	 * (22) will be ignored.
1656 	 */
1657 	if (virtio_intr_enable(dev) < 0) {
1658 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1659 		return -1;
1660 	}
1661 
1662 	if (virtio_queues_bind_intr(dev) < 0) {
1663 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1664 		return -1;
1665 	}
1666 
1667 	return 0;
1668 }
1669 #define SPEED_UNKNOWN    0xffffffff
1670 #define DUPLEX_UNKNOWN   0xff
1671 /* reset device and renegotiate features if needed */
1672 static int
1673 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1674 {
1675 	struct virtio_hw *hw = eth_dev->data->dev_private;
1676 	struct virtio_net_config *config;
1677 	struct virtio_net_config local_config;
1678 	struct rte_pci_device *pci_dev = NULL;
1679 	int ret;
1680 
1681 	/* Reset the device although not necessary at startup */
1682 	vtpci_reset(hw);
1683 
1684 	if (hw->vqs) {
1685 		virtio_dev_free_mbufs(eth_dev);
1686 		virtio_free_queues(hw);
1687 	}
1688 
1689 	/* Tell the host we've noticed this device. */
1690 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1691 
1692 	/* Tell the host we've known how to drive the device. */
1693 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1694 	if (virtio_negotiate_features(hw, req_features) < 0)
1695 		return -1;
1696 
1697 	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1698 
1699 	if (!hw->virtio_user_dev)
1700 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1701 
1702 	/* If host does not support both status and MSI-X then disable LSC */
1703 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1704 	    hw->use_msix != VIRTIO_MSIX_NONE)
1705 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1706 	else
1707 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1708 
1709 	/* Setting up rx_header size for the device */
1710 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1711 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1712 	    vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1713 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1714 	else
1715 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1716 
1717 	/* Copy the permanent MAC address to: virtio_hw */
1718 	virtio_get_hwaddr(hw);
1719 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1720 			&eth_dev->data->mac_addrs[0]);
1721 	PMD_INIT_LOG(DEBUG,
1722 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1723 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1724 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1725 
1726 	if (hw->speed == SPEED_UNKNOWN) {
1727 		if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1728 			config = &local_config;
1729 			vtpci_read_dev_config(hw,
1730 				offsetof(struct virtio_net_config, speed),
1731 				&config->speed, sizeof(config->speed));
1732 			vtpci_read_dev_config(hw,
1733 				offsetof(struct virtio_net_config, duplex),
1734 				&config->duplex, sizeof(config->duplex));
1735 			hw->speed = config->speed;
1736 			hw->duplex = config->duplex;
1737 		}
1738 	}
1739 	if (hw->speed == SPEED_UNKNOWN)
1740 		hw->speed = ETH_SPEED_NUM_10G;
1741 	if (hw->duplex == DUPLEX_UNKNOWN)
1742 		hw->duplex = ETH_LINK_FULL_DUPLEX;
1743 	PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1744 		hw->speed, hw->duplex);
1745 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1746 		config = &local_config;
1747 
1748 		vtpci_read_dev_config(hw,
1749 			offsetof(struct virtio_net_config, mac),
1750 			&config->mac, sizeof(config->mac));
1751 
1752 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1753 			vtpci_read_dev_config(hw,
1754 				offsetof(struct virtio_net_config, status),
1755 				&config->status, sizeof(config->status));
1756 		} else {
1757 			PMD_INIT_LOG(DEBUG,
1758 				     "VIRTIO_NET_F_STATUS is not supported");
1759 			config->status = 0;
1760 		}
1761 
1762 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1763 			vtpci_read_dev_config(hw,
1764 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1765 				&config->max_virtqueue_pairs,
1766 				sizeof(config->max_virtqueue_pairs));
1767 		} else {
1768 			PMD_INIT_LOG(DEBUG,
1769 				     "VIRTIO_NET_F_MQ is not supported");
1770 			config->max_virtqueue_pairs = 1;
1771 		}
1772 
1773 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1774 
1775 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1776 			vtpci_read_dev_config(hw,
1777 				offsetof(struct virtio_net_config, mtu),
1778 				&config->mtu,
1779 				sizeof(config->mtu));
1780 
1781 			/*
1782 			 * MTU value has already been checked at negotiation
1783 			 * time, but check again in case it has changed since
1784 			 * then, which should not happen.
1785 			 */
1786 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1787 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1788 						config->mtu);
1789 				return -1;
1790 			}
1791 
1792 			hw->max_mtu = config->mtu;
1793 			/* Set initial MTU to maximum one supported by vhost */
1794 			eth_dev->data->mtu = config->mtu;
1795 
1796 		} else {
1797 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1798 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1799 		}
1800 
1801 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1802 				config->max_virtqueue_pairs);
1803 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1804 		PMD_INIT_LOG(DEBUG,
1805 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1806 				config->mac[0], config->mac[1],
1807 				config->mac[2], config->mac[3],
1808 				config->mac[4], config->mac[5]);
1809 	} else {
1810 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1811 		hw->max_queue_pairs = 1;
1812 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1813 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1814 	}
1815 
1816 	ret = virtio_alloc_queues(eth_dev);
1817 	if (ret < 0)
1818 		return ret;
1819 
1820 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1821 		if (virtio_configure_intr(eth_dev) < 0) {
1822 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1823 			virtio_free_queues(hw);
1824 			return -1;
1825 		}
1826 	}
1827 
1828 	vtpci_reinit_complete(hw);
1829 
1830 	if (pci_dev)
1831 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1832 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1833 			pci_dev->id.device_id);
1834 
1835 	return 0;
1836 }
1837 
1838 /*
1839  * Remap the PCI device again (IO port map for legacy device and
1840  * memory map for modern device), so that the secondary process
1841  * could have the PCI initiated correctly.
1842  */
1843 static int
1844 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1845 {
1846 	if (hw->modern) {
1847 		/*
1848 		 * We don't have to re-parse the PCI config space, since
1849 		 * rte_pci_map_device() makes sure the mapped address
1850 		 * in secondary process would equal to the one mapped in
1851 		 * the primary process: error will be returned if that
1852 		 * requirement is not met.
1853 		 *
1854 		 * That said, we could simply reuse all cap pointers
1855 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1856 		 * primary process, which is stored in shared memory.
1857 		 */
1858 		if (rte_pci_map_device(pci_dev)) {
1859 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1860 			return -1;
1861 		}
1862 	} else {
1863 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1864 			return -1;
1865 	}
1866 
1867 	return 0;
1868 }
1869 
1870 static void
1871 virtio_set_vtpci_ops(struct virtio_hw *hw)
1872 {
1873 #ifdef RTE_VIRTIO_USER
1874 	if (hw->virtio_user_dev)
1875 		VTPCI_OPS(hw) = &virtio_user_ops;
1876 	else
1877 #endif
1878 	if (hw->modern)
1879 		VTPCI_OPS(hw) = &modern_ops;
1880 	else
1881 		VTPCI_OPS(hw) = &legacy_ops;
1882 }
1883 
1884 /*
1885  * This function is based on probe() function in virtio_pci.c
1886  * It returns 0 on success.
1887  */
1888 int
1889 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1890 {
1891 	struct virtio_hw *hw = eth_dev->data->dev_private;
1892 	uint32_t speed = SPEED_UNKNOWN;
1893 	int vectorized = 0;
1894 	int ret;
1895 
1896 	if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1897 		PMD_INIT_LOG(ERR,
1898 			"Not sufficient headroom required = %d, avail = %d",
1899 			(int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1900 			RTE_PKTMBUF_HEADROOM);
1901 
1902 		return -1;
1903 	}
1904 
1905 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1906 
1907 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1908 		if (!hw->virtio_user_dev) {
1909 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1910 			if (ret)
1911 				return ret;
1912 		}
1913 
1914 		virtio_set_vtpci_ops(hw);
1915 		set_rxtx_funcs(eth_dev);
1916 
1917 		return 0;
1918 	}
1919 	ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
1920 		 NULL, &speed, &vectorized);
1921 	if (ret < 0)
1922 		return ret;
1923 	hw->speed = speed;
1924 	/*
1925 	 * Pass the information to the rte_eth_dev_close() that it should also
1926 	 * release the private port resources.
1927 	 */
1928 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1929 
1930 	/* Allocate memory for storing MAC addresses */
1931 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1932 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1933 	if (eth_dev->data->mac_addrs == NULL) {
1934 		PMD_INIT_LOG(ERR,
1935 			"Failed to allocate %d bytes needed to store MAC addresses",
1936 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1937 		return -ENOMEM;
1938 	}
1939 
1940 	hw->port_id = eth_dev->data->port_id;
1941 	/* For virtio_user case the hw->virtio_user_dev is populated by
1942 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1943 	 */
1944 	if (!hw->virtio_user_dev) {
1945 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1946 		if (ret)
1947 			goto err_vtpci_init;
1948 	}
1949 
1950 	rte_spinlock_init(&hw->state_lock);
1951 
1952 	/* reset device and negotiate default features */
1953 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1954 	if (ret < 0)
1955 		goto err_virtio_init;
1956 
1957 	if (vectorized) {
1958 		if (!vtpci_packed_queue(hw)) {
1959 			hw->use_vec_rx = 1;
1960 		} else {
1961 #if !defined(CC_AVX512_SUPPORT)
1962 			PMD_DRV_LOG(INFO,
1963 				"building environment do not support packed ring vectorized");
1964 #else
1965 			hw->use_vec_rx = 1;
1966 			hw->use_vec_tx = 1;
1967 #endif
1968 		}
1969 	}
1970 
1971 	hw->opened = true;
1972 
1973 	return 0;
1974 
1975 err_virtio_init:
1976 	if (!hw->virtio_user_dev) {
1977 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1978 		if (!hw->modern)
1979 			rte_pci_ioport_unmap(VTPCI_IO(hw));
1980 	}
1981 err_vtpci_init:
1982 	rte_free(eth_dev->data->mac_addrs);
1983 	eth_dev->data->mac_addrs = NULL;
1984 	return ret;
1985 }
1986 
1987 static int
1988 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1989 {
1990 	PMD_INIT_FUNC_TRACE();
1991 
1992 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1993 		return 0;
1994 
1995 	virtio_dev_stop(eth_dev);
1996 	virtio_dev_close(eth_dev);
1997 
1998 	eth_dev->dev_ops = NULL;
1999 	eth_dev->tx_pkt_burst = NULL;
2000 	eth_dev->rx_pkt_burst = NULL;
2001 
2002 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
2003 
2004 	return 0;
2005 }
2006 
2007 
2008 static int vdpa_check_handler(__rte_unused const char *key,
2009 		const char *value, void *ret_val)
2010 {
2011 	if (strcmp(value, "1") == 0)
2012 		*(int *)ret_val = 1;
2013 	else
2014 		*(int *)ret_val = 0;
2015 
2016 	return 0;
2017 }
2018 
2019 
2020 static uint32_t
2021 virtio_dev_speed_capa_get(uint32_t speed)
2022 {
2023 	switch (speed) {
2024 	case ETH_SPEED_NUM_10G:
2025 		return ETH_LINK_SPEED_10G;
2026 	case ETH_SPEED_NUM_20G:
2027 		return ETH_LINK_SPEED_20G;
2028 	case ETH_SPEED_NUM_25G:
2029 		return ETH_LINK_SPEED_25G;
2030 	case ETH_SPEED_NUM_40G:
2031 		return ETH_LINK_SPEED_40G;
2032 	case ETH_SPEED_NUM_50G:
2033 		return ETH_LINK_SPEED_50G;
2034 	case ETH_SPEED_NUM_56G:
2035 		return ETH_LINK_SPEED_56G;
2036 	case ETH_SPEED_NUM_100G:
2037 		return ETH_LINK_SPEED_100G;
2038 	default:
2039 		return 0;
2040 	}
2041 }
2042 
2043 static int vectorized_check_handler(__rte_unused const char *key,
2044 		const char *value, void *ret_val)
2045 {
2046 	if (strcmp(value, "1") == 0)
2047 		*(int *)ret_val = 1;
2048 	else
2049 		*(int *)ret_val = 0;
2050 
2051 	return 0;
2052 }
2053 
2054 #define VIRTIO_ARG_SPEED      "speed"
2055 #define VIRTIO_ARG_VDPA       "vdpa"
2056 #define VIRTIO_ARG_VECTORIZED "vectorized"
2057 
2058 
2059 static int
2060 link_speed_handler(const char *key __rte_unused,
2061 		const char *value, void *ret_val)
2062 {
2063 	uint32_t val;
2064 	if (!value || !ret_val)
2065 		return -EINVAL;
2066 	val = strtoul(value, NULL, 0);
2067 	/* validate input */
2068 	if (virtio_dev_speed_capa_get(val) == 0)
2069 		return -EINVAL;
2070 	*(uint32_t *)ret_val = val;
2071 
2072 	return 0;
2073 }
2074 
2075 
2076 static int
2077 virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
2078 	uint32_t *speed, int *vectorized)
2079 {
2080 	struct rte_kvargs *kvlist;
2081 	int ret = 0;
2082 
2083 	if (devargs == NULL)
2084 		return 0;
2085 
2086 	kvlist = rte_kvargs_parse(devargs->args, NULL);
2087 	if (kvlist == NULL) {
2088 		PMD_INIT_LOG(ERR, "error when parsing param");
2089 		return 0;
2090 	}
2091 	if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
2092 		/* vdpa mode selected when there's a key-value pair:
2093 		 * vdpa=1
2094 		 */
2095 		ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
2096 				vdpa_check_handler, vdpa);
2097 		if (ret < 0) {
2098 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2099 				VIRTIO_ARG_VDPA);
2100 			goto exit;
2101 		}
2102 	}
2103 	if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2104 		ret = rte_kvargs_process(kvlist,
2105 					VIRTIO_ARG_SPEED,
2106 					link_speed_handler, speed);
2107 		if (ret < 0) {
2108 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2109 					VIRTIO_ARG_SPEED);
2110 			goto exit;
2111 		}
2112 	}
2113 
2114 	if (vectorized &&
2115 		rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2116 		ret = rte_kvargs_process(kvlist,
2117 				VIRTIO_ARG_VECTORIZED,
2118 				vectorized_check_handler, vectorized);
2119 		if (ret < 0) {
2120 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2121 					VIRTIO_ARG_VECTORIZED);
2122 			goto exit;
2123 		}
2124 	}
2125 
2126 exit:
2127 	rte_kvargs_free(kvlist);
2128 	return ret;
2129 }
2130 
2131 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2132 	struct rte_pci_device *pci_dev)
2133 {
2134 	int vdpa = 0;
2135 	int ret = 0;
2136 
2137 	ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL,
2138 		NULL);
2139 	if (ret < 0) {
2140 		PMD_INIT_LOG(ERR, "devargs parsing is failed");
2141 		return ret;
2142 	}
2143 	/* virtio pmd skips probe if device needs to work in vdpa mode */
2144 	if (vdpa == 1)
2145 		return 1;
2146 
2147 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2148 		eth_virtio_dev_init);
2149 }
2150 
2151 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2152 {
2153 	int ret;
2154 
2155 	ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2156 	/* Port has already been released by close. */
2157 	if (ret == -ENODEV)
2158 		ret = 0;
2159 	return ret;
2160 }
2161 
2162 static struct rte_pci_driver rte_virtio_pmd = {
2163 	.driver = {
2164 		.name = "net_virtio",
2165 	},
2166 	.id_table = pci_id_virtio_map,
2167 	.drv_flags = 0,
2168 	.probe = eth_virtio_pci_probe,
2169 	.remove = eth_virtio_pci_remove,
2170 };
2171 
2172 RTE_INIT(rte_virtio_pmd_init)
2173 {
2174 	rte_eal_iopl_init();
2175 	rte_pci_register(&rte_virtio_pmd);
2176 }
2177 
2178 static bool
2179 rx_offload_enabled(struct virtio_hw *hw)
2180 {
2181 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2182 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2183 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2184 }
2185 
2186 static bool
2187 tx_offload_enabled(struct virtio_hw *hw)
2188 {
2189 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2190 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2191 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2192 }
2193 
2194 /*
2195  * Configure virtio device
2196  * It returns 0 on success.
2197  */
2198 static int
2199 virtio_dev_configure(struct rte_eth_dev *dev)
2200 {
2201 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2202 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2203 	struct virtio_hw *hw = dev->data->dev_private;
2204 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2205 		hw->vtnet_hdr_size;
2206 	uint64_t rx_offloads = rxmode->offloads;
2207 	uint64_t tx_offloads = txmode->offloads;
2208 	uint64_t req_features;
2209 	int ret;
2210 
2211 	PMD_INIT_LOG(DEBUG, "configure");
2212 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2213 
2214 	if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2215 		PMD_DRV_LOG(ERR,
2216 			"Unsupported Rx multi queue mode %d",
2217 			rxmode->mq_mode);
2218 		return -EINVAL;
2219 	}
2220 
2221 	if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2222 		PMD_DRV_LOG(ERR,
2223 			"Unsupported Tx multi queue mode %d",
2224 			txmode->mq_mode);
2225 		return -EINVAL;
2226 	}
2227 
2228 	if (dev->data->dev_conf.intr_conf.rxq) {
2229 		ret = virtio_init_device(dev, hw->req_guest_features);
2230 		if (ret < 0)
2231 			return ret;
2232 	}
2233 
2234 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2235 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2236 
2237 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2238 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2239 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2240 
2241 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2242 		req_features |=
2243 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2244 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2245 
2246 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2247 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2248 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2249 
2250 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2251 		req_features |=
2252 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2253 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2254 
2255 	/* if request features changed, reinit the device */
2256 	if (req_features != hw->req_guest_features) {
2257 		ret = virtio_init_device(dev, req_features);
2258 		if (ret < 0)
2259 			return ret;
2260 	}
2261 
2262 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2263 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2264 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2265 		PMD_DRV_LOG(ERR,
2266 			"rx checksum not available on this host");
2267 		return -ENOTSUP;
2268 	}
2269 
2270 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2271 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2272 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2273 		PMD_DRV_LOG(ERR,
2274 			"Large Receive Offload not available on this host");
2275 		return -ENOTSUP;
2276 	}
2277 
2278 	/* start control queue */
2279 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2280 		virtio_dev_cq_start(dev);
2281 
2282 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2283 		hw->vlan_strip = 1;
2284 
2285 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2286 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2287 		PMD_DRV_LOG(ERR,
2288 			    "vlan filtering not available on this host");
2289 		return -ENOTSUP;
2290 	}
2291 
2292 	hw->has_tx_offload = tx_offload_enabled(hw);
2293 	hw->has_rx_offload = rx_offload_enabled(hw);
2294 
2295 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2296 		/* Enable vector (0) for Link State Intrerrupt */
2297 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2298 				VIRTIO_MSI_NO_VECTOR) {
2299 			PMD_DRV_LOG(ERR, "failed to set config vector");
2300 			return -EBUSY;
2301 		}
2302 
2303 	if (vtpci_packed_queue(hw)) {
2304 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2305 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2306 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2307 		     !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2308 		     !vtpci_with_feature(hw, VIRTIO_F_VERSION_1))) {
2309 			PMD_DRV_LOG(INFO,
2310 				"disabled packed ring vectorized path for requirements not met");
2311 			hw->use_vec_rx = 0;
2312 			hw->use_vec_tx = 0;
2313 		}
2314 #else
2315 		hw->use_vec_rx = 0;
2316 		hw->use_vec_tx = 0;
2317 #endif
2318 
2319 		if (hw->use_vec_rx) {
2320 			if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2321 				PMD_DRV_LOG(INFO,
2322 					"disabled packed ring vectorized rx for mrg_rxbuf enabled");
2323 				hw->use_vec_rx = 0;
2324 			}
2325 
2326 			if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2327 				PMD_DRV_LOG(INFO,
2328 					"disabled packed ring vectorized rx for TCP_LRO enabled");
2329 				hw->use_vec_rx = 0;
2330 			}
2331 		}
2332 	} else {
2333 		if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2334 			hw->use_inorder_tx = 1;
2335 			hw->use_inorder_rx = 1;
2336 			hw->use_vec_rx = 0;
2337 		}
2338 
2339 		if (hw->use_vec_rx) {
2340 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2341 			if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2342 				PMD_DRV_LOG(INFO,
2343 					"disabled split ring vectorized path for requirement not met");
2344 				hw->use_vec_rx = 0;
2345 			}
2346 #endif
2347 			if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2348 				PMD_DRV_LOG(INFO,
2349 					"disabled split ring vectorized rx for mrg_rxbuf enabled");
2350 				hw->use_vec_rx = 0;
2351 			}
2352 
2353 			if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2354 					   DEV_RX_OFFLOAD_TCP_CKSUM |
2355 					   DEV_RX_OFFLOAD_TCP_LRO |
2356 					   DEV_RX_OFFLOAD_VLAN_STRIP)) {
2357 				PMD_DRV_LOG(INFO,
2358 					"disabled split ring vectorized rx for offloading enabled");
2359 				hw->use_vec_rx = 0;
2360 			}
2361 		}
2362 	}
2363 
2364 	return 0;
2365 }
2366 
2367 
2368 static int
2369 virtio_dev_start(struct rte_eth_dev *dev)
2370 {
2371 	uint16_t nb_queues, i;
2372 	struct virtnet_rx *rxvq;
2373 	struct virtnet_tx *txvq __rte_unused;
2374 	struct virtio_hw *hw = dev->data->dev_private;
2375 	int ret;
2376 
2377 	/* Finish the initialization of the queues */
2378 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2379 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2380 		if (ret < 0)
2381 			return ret;
2382 	}
2383 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2384 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2385 		if (ret < 0)
2386 			return ret;
2387 	}
2388 
2389 	/* check if lsc interrupt feature is enabled */
2390 	if (dev->data->dev_conf.intr_conf.lsc) {
2391 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2392 			PMD_DRV_LOG(ERR, "link status not supported by host");
2393 			return -ENOTSUP;
2394 		}
2395 	}
2396 
2397 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2398 	 * in device configure, but it could be unmapped  when device is
2399 	 * stopped.
2400 	 */
2401 	if (dev->data->dev_conf.intr_conf.lsc ||
2402 	    dev->data->dev_conf.intr_conf.rxq) {
2403 		virtio_intr_disable(dev);
2404 
2405 		/* Setup interrupt callback  */
2406 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2407 			rte_intr_callback_register(dev->intr_handle,
2408 						   virtio_interrupt_handler,
2409 						   dev);
2410 
2411 		if (virtio_intr_enable(dev) < 0) {
2412 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2413 			return -EIO;
2414 		}
2415 	}
2416 
2417 	/*Notify the backend
2418 	 *Otherwise the tap backend might already stop its queue due to fullness.
2419 	 *vhost backend will have no chance to be waked up
2420 	 */
2421 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2422 	if (hw->max_queue_pairs > 1) {
2423 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2424 			return -EINVAL;
2425 	}
2426 
2427 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2428 
2429 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2430 		rxvq = dev->data->rx_queues[i];
2431 		/* Flush the old packets */
2432 		virtqueue_rxvq_flush(rxvq->vq);
2433 		virtqueue_notify(rxvq->vq);
2434 	}
2435 
2436 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2437 		txvq = dev->data->tx_queues[i];
2438 		virtqueue_notify(txvq->vq);
2439 	}
2440 
2441 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2442 
2443 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2444 		rxvq = dev->data->rx_queues[i];
2445 		VIRTQUEUE_DUMP(rxvq->vq);
2446 	}
2447 
2448 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2449 		txvq = dev->data->tx_queues[i];
2450 		VIRTQUEUE_DUMP(txvq->vq);
2451 	}
2452 
2453 	set_rxtx_funcs(dev);
2454 	hw->started = true;
2455 
2456 	/* Initialize Link state */
2457 	virtio_dev_link_update(dev, 0);
2458 
2459 	return 0;
2460 }
2461 
2462 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2463 {
2464 	struct virtio_hw *hw = dev->data->dev_private;
2465 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2466 	const char *type __rte_unused;
2467 	unsigned int i, mbuf_num = 0;
2468 	struct virtqueue *vq;
2469 	struct rte_mbuf *buf;
2470 	int queue_type;
2471 
2472 	if (hw->vqs == NULL)
2473 		return;
2474 
2475 	for (i = 0; i < nr_vq; i++) {
2476 		vq = hw->vqs[i];
2477 		if (!vq)
2478 			continue;
2479 
2480 		queue_type = virtio_get_queue_type(hw, i);
2481 		if (queue_type == VTNET_RQ)
2482 			type = "rxq";
2483 		else if (queue_type == VTNET_TQ)
2484 			type = "txq";
2485 		else
2486 			continue;
2487 
2488 		PMD_INIT_LOG(DEBUG,
2489 			"Before freeing %s[%d] used and unused buf",
2490 			type, i);
2491 		VIRTQUEUE_DUMP(vq);
2492 
2493 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2494 			rte_pktmbuf_free(buf);
2495 			mbuf_num++;
2496 		}
2497 
2498 		PMD_INIT_LOG(DEBUG,
2499 			"After freeing %s[%d] used and unused buf",
2500 			type, i);
2501 		VIRTQUEUE_DUMP(vq);
2502 	}
2503 
2504 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2505 }
2506 
2507 /*
2508  * Stop device: disable interrupt and mark link down
2509  */
2510 static void
2511 virtio_dev_stop(struct rte_eth_dev *dev)
2512 {
2513 	struct virtio_hw *hw = dev->data->dev_private;
2514 	struct rte_eth_link link;
2515 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2516 
2517 	PMD_INIT_LOG(DEBUG, "stop");
2518 
2519 	rte_spinlock_lock(&hw->state_lock);
2520 	if (!hw->started)
2521 		goto out_unlock;
2522 	hw->started = false;
2523 
2524 	if (intr_conf->lsc || intr_conf->rxq) {
2525 		virtio_intr_disable(dev);
2526 
2527 		/* Reset interrupt callback  */
2528 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2529 			rte_intr_callback_unregister(dev->intr_handle,
2530 						     virtio_interrupt_handler,
2531 						     dev);
2532 		}
2533 	}
2534 
2535 	memset(&link, 0, sizeof(link));
2536 	rte_eth_linkstatus_set(dev, &link);
2537 out_unlock:
2538 	rte_spinlock_unlock(&hw->state_lock);
2539 }
2540 
2541 static int
2542 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2543 {
2544 	struct rte_eth_link link;
2545 	uint16_t status;
2546 	struct virtio_hw *hw = dev->data->dev_private;
2547 
2548 	memset(&link, 0, sizeof(link));
2549 	link.link_duplex = hw->duplex;
2550 	link.link_speed  = hw->speed;
2551 	link.link_autoneg = ETH_LINK_AUTONEG;
2552 
2553 	if (!hw->started) {
2554 		link.link_status = ETH_LINK_DOWN;
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 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2563 				     dev->data->port_id);
2564 		} else {
2565 			link.link_status = ETH_LINK_UP;
2566 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2567 				     dev->data->port_id);
2568 		}
2569 	} else {
2570 		link.link_status = ETH_LINK_UP;
2571 	}
2572 
2573 	return rte_eth_linkstatus_set(dev, &link);
2574 }
2575 
2576 static int
2577 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2578 {
2579 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2580 	struct virtio_hw *hw = dev->data->dev_private;
2581 	uint64_t offloads = rxmode->offloads;
2582 
2583 	if (mask & ETH_VLAN_FILTER_MASK) {
2584 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2585 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2586 
2587 			PMD_DRV_LOG(NOTICE,
2588 				"vlan filtering not available on this host");
2589 
2590 			return -ENOTSUP;
2591 		}
2592 	}
2593 
2594 	if (mask & ETH_VLAN_STRIP_MASK)
2595 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2596 
2597 	return 0;
2598 }
2599 
2600 static int
2601 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2602 {
2603 	uint64_t tso_mask, host_features;
2604 	struct virtio_hw *hw = dev->data->dev_private;
2605 	dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2606 
2607 	dev_info->max_rx_queues =
2608 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2609 	dev_info->max_tx_queues =
2610 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2611 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2612 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2613 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2614 
2615 	host_features = VTPCI_OPS(hw)->get_features(hw);
2616 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2617 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2618 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2619 		dev_info->rx_offload_capa |=
2620 			DEV_RX_OFFLOAD_TCP_CKSUM |
2621 			DEV_RX_OFFLOAD_UDP_CKSUM;
2622 	}
2623 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2624 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2625 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2626 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2627 	if ((host_features & tso_mask) == tso_mask)
2628 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2629 
2630 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2631 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2632 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2633 		dev_info->tx_offload_capa |=
2634 			DEV_TX_OFFLOAD_UDP_CKSUM |
2635 			DEV_TX_OFFLOAD_TCP_CKSUM;
2636 	}
2637 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2638 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2639 	if ((host_features & tso_mask) == tso_mask)
2640 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2641 
2642 	return 0;
2643 }
2644 
2645 /*
2646  * It enables testpmd to collect per queue stats.
2647  */
2648 static int
2649 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2650 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2651 __rte_unused uint8_t is_rx)
2652 {
2653 	return 0;
2654 }
2655 
2656 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2657 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2658 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2659 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2660 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);
2661