xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 200bc52e5aa0d72e70464c9cd22b55cf536ed13c)
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 
39 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
40 static int  virtio_dev_configure(struct rte_eth_dev *dev);
41 static int  virtio_dev_start(struct rte_eth_dev *dev);
42 static void virtio_dev_stop(struct rte_eth_dev *dev);
43 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
44 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
45 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
46 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
47 static void virtio_dev_info_get(struct rte_eth_dev *dev,
48 				struct rte_eth_dev_info *dev_info);
49 static int virtio_dev_link_update(struct rte_eth_dev *dev,
50 	int wait_to_complete);
51 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
52 
53 static void virtio_set_hwaddr(struct virtio_hw *hw);
54 static void virtio_get_hwaddr(struct virtio_hw *hw);
55 
56 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
57 				 struct rte_eth_stats *stats);
58 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
59 				 struct rte_eth_xstat *xstats, unsigned n);
60 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
61 				       struct rte_eth_xstat_name *xstats_names,
62 				       unsigned limit);
63 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
64 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
65 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
66 				uint16_t vlan_id, int on);
67 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
68 				struct rte_ether_addr *mac_addr,
69 				uint32_t index, uint32_t vmdq);
70 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
71 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
72 				struct rte_ether_addr *mac_addr);
73 
74 static int virtio_intr_disable(struct rte_eth_dev *dev);
75 
76 static int virtio_dev_queue_stats_mapping_set(
77 	struct rte_eth_dev *eth_dev,
78 	uint16_t queue_id,
79 	uint8_t stat_idx,
80 	uint8_t is_rx);
81 
82 int virtio_logtype_init;
83 int virtio_logtype_driver;
84 
85 static void virtio_notify_peers(struct rte_eth_dev *dev);
86 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
87 
88 /*
89  * The set of PCI devices this driver supports
90  */
91 static const struct rte_pci_id pci_id_virtio_map[] = {
92 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
93 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
94 	{ .vendor_id = 0, /* sentinel */ },
95 };
96 
97 struct rte_virtio_xstats_name_off {
98 	char name[RTE_ETH_XSTATS_NAME_SIZE];
99 	unsigned offset;
100 };
101 
102 /* [rt]x_qX_ is prepended to the name string here */
103 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
104 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
105 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
106 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
107 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
108 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
109 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
110 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
111 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
112 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
113 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
114 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
115 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
116 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
117 };
118 
119 /* [rt]x_qX_ is prepended to the name string here */
120 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
121 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
122 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
123 	{"errors",                 offsetof(struct virtnet_tx, stats.errors)},
124 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
125 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
126 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
127 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
128 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
129 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
130 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
131 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
132 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
133 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
134 };
135 
136 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
137 			    sizeof(rte_virtio_rxq_stat_strings[0]))
138 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
139 			    sizeof(rte_virtio_txq_stat_strings[0]))
140 
141 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
142 
143 static struct virtio_pmd_ctrl *
144 virtio_send_command_packed(struct virtnet_ctl *cvq,
145 			   struct virtio_pmd_ctrl *ctrl,
146 			   int *dlen, int pkt_num)
147 {
148 	struct virtqueue *vq = cvq->vq;
149 	int head;
150 	struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
151 	struct virtio_pmd_ctrl *result;
152 	uint16_t flags;
153 	int sum = 0;
154 	int nb_descs = 0;
155 	int k;
156 
157 	/*
158 	 * Format is enforced in qemu code:
159 	 * One TX packet for header;
160 	 * At least one TX packet per argument;
161 	 * One RX packet for ACK.
162 	 */
163 	head = vq->vq_avail_idx;
164 	flags = vq->vq_packed.cached_flags;
165 	desc[head].addr = cvq->virtio_net_hdr_mem;
166 	desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
167 	vq->vq_free_cnt--;
168 	nb_descs++;
169 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
170 		vq->vq_avail_idx -= vq->vq_nentries;
171 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
172 	}
173 
174 	for (k = 0; k < pkt_num; k++) {
175 		desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
176 			+ sizeof(struct virtio_net_ctrl_hdr)
177 			+ sizeof(ctrl->status) + sizeof(uint8_t) * sum;
178 		desc[vq->vq_avail_idx].len = dlen[k];
179 		desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
180 			vq->vq_packed.cached_flags;
181 		sum += dlen[k];
182 		vq->vq_free_cnt--;
183 		nb_descs++;
184 		if (++vq->vq_avail_idx >= vq->vq_nentries) {
185 			vq->vq_avail_idx -= vq->vq_nentries;
186 			vq->vq_packed.cached_flags ^=
187 				VRING_PACKED_DESC_F_AVAIL_USED;
188 		}
189 	}
190 
191 	desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
192 		+ sizeof(struct virtio_net_ctrl_hdr);
193 	desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
194 	desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
195 		vq->vq_packed.cached_flags;
196 	vq->vq_free_cnt--;
197 	nb_descs++;
198 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
199 		vq->vq_avail_idx -= vq->vq_nentries;
200 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
201 	}
202 
203 	virtio_wmb(vq->hw->weak_barriers);
204 	desc[head].flags = VRING_DESC_F_NEXT | flags;
205 
206 	virtio_wmb(vq->hw->weak_barriers);
207 	virtqueue_notify(vq);
208 
209 	/* wait for used descriptors in virtqueue */
210 	while (!desc_is_used(&desc[head], vq))
211 		usleep(100);
212 
213 	virtio_rmb(vq->hw->weak_barriers);
214 
215 	/* now get used descriptors */
216 	vq->vq_free_cnt += nb_descs;
217 	vq->vq_used_cons_idx += nb_descs;
218 	if (vq->vq_used_cons_idx >= vq->vq_nentries) {
219 		vq->vq_used_cons_idx -= vq->vq_nentries;
220 		vq->vq_packed.used_wrap_counter ^= 1;
221 	}
222 
223 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
224 			"vq->vq_avail_idx=%d\n"
225 			"vq->vq_used_cons_idx=%d\n"
226 			"vq->vq_packed.cached_flags=0x%x\n"
227 			"vq->vq_packed.used_wrap_counter=%d\n",
228 			vq->vq_free_cnt,
229 			vq->vq_avail_idx,
230 			vq->vq_used_cons_idx,
231 			vq->vq_packed.cached_flags,
232 			vq->vq_packed.used_wrap_counter);
233 
234 	result = cvq->virtio_net_hdr_mz->addr;
235 	return result;
236 }
237 
238 static struct virtio_pmd_ctrl *
239 virtio_send_command_split(struct virtnet_ctl *cvq,
240 			  struct virtio_pmd_ctrl *ctrl,
241 			  int *dlen, int pkt_num)
242 {
243 	struct virtio_pmd_ctrl *result;
244 	struct virtqueue *vq = cvq->vq;
245 	uint32_t head, i;
246 	int k, sum = 0;
247 
248 	head = vq->vq_desc_head_idx;
249 
250 	/*
251 	 * Format is enforced in qemu code:
252 	 * One TX packet for header;
253 	 * At least one TX packet per argument;
254 	 * One RX packet for ACK.
255 	 */
256 	vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
257 	vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
258 	vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
259 	vq->vq_free_cnt--;
260 	i = vq->vq_split.ring.desc[head].next;
261 
262 	for (k = 0; k < pkt_num; k++) {
263 		vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
264 		vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
265 			+ sizeof(struct virtio_net_ctrl_hdr)
266 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
267 		vq->vq_split.ring.desc[i].len = dlen[k];
268 		sum += dlen[k];
269 		vq->vq_free_cnt--;
270 		i = vq->vq_split.ring.desc[i].next;
271 	}
272 
273 	vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
274 	vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
275 			+ sizeof(struct virtio_net_ctrl_hdr);
276 	vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
277 	vq->vq_free_cnt--;
278 
279 	vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
280 
281 	vq_update_avail_ring(vq, head);
282 	vq_update_avail_idx(vq);
283 
284 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
285 
286 	virtqueue_notify(vq);
287 
288 	rte_rmb();
289 	while (VIRTQUEUE_NUSED(vq) == 0) {
290 		rte_rmb();
291 		usleep(100);
292 	}
293 
294 	while (VIRTQUEUE_NUSED(vq)) {
295 		uint32_t idx, desc_idx, used_idx;
296 		struct vring_used_elem *uep;
297 
298 		used_idx = (uint32_t)(vq->vq_used_cons_idx
299 				& (vq->vq_nentries - 1));
300 		uep = &vq->vq_split.ring.used->ring[used_idx];
301 		idx = (uint32_t) uep->id;
302 		desc_idx = idx;
303 
304 		while (vq->vq_split.ring.desc[desc_idx].flags &
305 				VRING_DESC_F_NEXT) {
306 			desc_idx = vq->vq_split.ring.desc[desc_idx].next;
307 			vq->vq_free_cnt++;
308 		}
309 
310 		vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
311 		vq->vq_desc_head_idx = idx;
312 
313 		vq->vq_used_cons_idx++;
314 		vq->vq_free_cnt++;
315 	}
316 
317 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
318 			vq->vq_free_cnt, vq->vq_desc_head_idx);
319 
320 	result = cvq->virtio_net_hdr_mz->addr;
321 	return result;
322 }
323 
324 static int
325 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
326 		    int *dlen, int pkt_num)
327 {
328 	virtio_net_ctrl_ack status = ~0;
329 	struct virtio_pmd_ctrl *result;
330 	struct virtqueue *vq;
331 
332 	ctrl->status = status;
333 
334 	if (!cvq || !cvq->vq) {
335 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
336 		return -1;
337 	}
338 
339 	rte_spinlock_lock(&cvq->lock);
340 	vq = cvq->vq;
341 
342 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
343 		"vq->hw->cvq = %p vq = %p",
344 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
345 
346 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
347 		rte_spinlock_unlock(&cvq->lock);
348 		return -1;
349 	}
350 
351 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
352 		sizeof(struct virtio_pmd_ctrl));
353 
354 	if (vtpci_packed_queue(vq->hw))
355 		result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
356 	else
357 		result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
358 
359 	rte_spinlock_unlock(&cvq->lock);
360 	return result->status;
361 }
362 
363 static int
364 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
365 {
366 	struct virtio_hw *hw = dev->data->dev_private;
367 	struct virtio_pmd_ctrl ctrl;
368 	int dlen[1];
369 	int ret;
370 
371 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
372 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
373 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
374 
375 	dlen[0] = sizeof(uint16_t);
376 
377 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
378 	if (ret) {
379 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
380 			  "failed, this is too late now...");
381 		return -EINVAL;
382 	}
383 
384 	return 0;
385 }
386 
387 static void
388 virtio_dev_queue_release(void *queue __rte_unused)
389 {
390 	/* do nothing */
391 }
392 
393 static uint16_t
394 virtio_get_nr_vq(struct virtio_hw *hw)
395 {
396 	uint16_t nr_vq = hw->max_queue_pairs * 2;
397 
398 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
399 		nr_vq += 1;
400 
401 	return nr_vq;
402 }
403 
404 static void
405 virtio_init_vring(struct virtqueue *vq)
406 {
407 	int size = vq->vq_nentries;
408 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
409 
410 	PMD_INIT_FUNC_TRACE();
411 
412 	memset(ring_mem, 0, vq->vq_ring_size);
413 
414 	vq->vq_used_cons_idx = 0;
415 	vq->vq_desc_head_idx = 0;
416 	vq->vq_avail_idx = 0;
417 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
418 	vq->vq_free_cnt = vq->vq_nentries;
419 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
420 	if (vtpci_packed_queue(vq->hw)) {
421 		vring_init_packed(&vq->vq_packed.ring, ring_mem,
422 				  VIRTIO_PCI_VRING_ALIGN, size);
423 		vring_desc_init_packed(vq, size);
424 	} else {
425 		struct vring *vr = &vq->vq_split.ring;
426 
427 		vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
428 		vring_desc_init_split(vr->desc, size);
429 	}
430 	/*
431 	 * Disable device(host) interrupting guest
432 	 */
433 	virtqueue_disable_intr(vq);
434 }
435 
436 static int
437 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
438 {
439 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
440 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
441 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
442 	unsigned int vq_size, size;
443 	struct virtio_hw *hw = dev->data->dev_private;
444 	struct virtnet_rx *rxvq = NULL;
445 	struct virtnet_tx *txvq = NULL;
446 	struct virtnet_ctl *cvq = NULL;
447 	struct virtqueue *vq;
448 	size_t sz_hdr_mz = 0;
449 	void *sw_ring = NULL;
450 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
451 	int ret;
452 	int numa_node = dev->device->numa_node;
453 
454 	PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
455 			vtpci_queue_idx, numa_node);
456 
457 	/*
458 	 * Read the virtqueue size from the Queue Size field
459 	 * Always power of 2 and if 0 virtqueue does not exist
460 	 */
461 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
462 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
463 	if (vq_size == 0) {
464 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
465 		return -EINVAL;
466 	}
467 
468 	if (!rte_is_power_of_2(vq_size)) {
469 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
470 		return -EINVAL;
471 	}
472 
473 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
474 		 dev->data->port_id, vtpci_queue_idx);
475 
476 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
477 				vq_size * sizeof(struct vq_desc_extra),
478 				RTE_CACHE_LINE_SIZE);
479 	if (queue_type == VTNET_TQ) {
480 		/*
481 		 * For each xmit packet, allocate a virtio_net_hdr
482 		 * and indirect ring elements
483 		 */
484 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
485 	} else if (queue_type == VTNET_CQ) {
486 		/* Allocate a page for control vq command, data and status */
487 		sz_hdr_mz = PAGE_SIZE;
488 	}
489 
490 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
491 				numa_node);
492 	if (vq == NULL) {
493 		PMD_INIT_LOG(ERR, "can not allocate vq");
494 		return -ENOMEM;
495 	}
496 	hw->vqs[vtpci_queue_idx] = vq;
497 
498 	vq->hw = hw;
499 	vq->vq_queue_index = vtpci_queue_idx;
500 	vq->vq_nentries = vq_size;
501 	if (vtpci_packed_queue(hw)) {
502 		vq->vq_packed.used_wrap_counter = 1;
503 		vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
504 		vq->vq_packed.event_flags_shadow = 0;
505 		if (queue_type == VTNET_RQ)
506 			vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
507 	}
508 
509 	/*
510 	 * Reserve a memzone for vring elements
511 	 */
512 	size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
513 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
514 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
515 		     size, vq->vq_ring_size);
516 
517 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
518 			numa_node, RTE_MEMZONE_IOVA_CONTIG,
519 			VIRTIO_PCI_VRING_ALIGN);
520 	if (mz == NULL) {
521 		if (rte_errno == EEXIST)
522 			mz = rte_memzone_lookup(vq_name);
523 		if (mz == NULL) {
524 			ret = -ENOMEM;
525 			goto fail_q_alloc;
526 		}
527 	}
528 
529 	memset(mz->addr, 0, mz->len);
530 
531 	vq->vq_ring_mem = mz->iova;
532 	vq->vq_ring_virt_mem = mz->addr;
533 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
534 		     (uint64_t)mz->iova);
535 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
536 		     (uint64_t)(uintptr_t)mz->addr);
537 
538 	virtio_init_vring(vq);
539 
540 	if (sz_hdr_mz) {
541 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
542 			 dev->data->port_id, vtpci_queue_idx);
543 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
544 				numa_node, RTE_MEMZONE_IOVA_CONTIG,
545 				RTE_CACHE_LINE_SIZE);
546 		if (hdr_mz == NULL) {
547 			if (rte_errno == EEXIST)
548 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
549 			if (hdr_mz == NULL) {
550 				ret = -ENOMEM;
551 				goto fail_q_alloc;
552 			}
553 		}
554 	}
555 
556 	if (queue_type == VTNET_RQ) {
557 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
558 			       sizeof(vq->sw_ring[0]);
559 
560 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
561 				RTE_CACHE_LINE_SIZE, numa_node);
562 		if (!sw_ring) {
563 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
564 			ret = -ENOMEM;
565 			goto fail_q_alloc;
566 		}
567 
568 		vq->sw_ring = sw_ring;
569 		rxvq = &vq->rxq;
570 		rxvq->vq = vq;
571 		rxvq->port_id = dev->data->port_id;
572 		rxvq->mz = mz;
573 	} else if (queue_type == VTNET_TQ) {
574 		txvq = &vq->txq;
575 		txvq->vq = vq;
576 		txvq->port_id = dev->data->port_id;
577 		txvq->mz = mz;
578 		txvq->virtio_net_hdr_mz = hdr_mz;
579 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
580 	} else if (queue_type == VTNET_CQ) {
581 		cvq = &vq->cq;
582 		cvq->vq = vq;
583 		cvq->mz = mz;
584 		cvq->virtio_net_hdr_mz = hdr_mz;
585 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
586 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
587 
588 		hw->cvq = cvq;
589 	}
590 
591 	/* For virtio_user case (that is when hw->dev is NULL), we use
592 	 * virtual address. And we need properly set _offset_, please see
593 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
594 	 */
595 	if (!hw->virtio_user_dev)
596 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
597 	else {
598 		vq->vq_ring_mem = (uintptr_t)mz->addr;
599 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
600 		if (queue_type == VTNET_TQ)
601 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
602 		else if (queue_type == VTNET_CQ)
603 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
604 	}
605 
606 	if (queue_type == VTNET_TQ) {
607 		struct virtio_tx_region *txr;
608 		unsigned int i;
609 
610 		txr = hdr_mz->addr;
611 		memset(txr, 0, vq_size * sizeof(*txr));
612 		for (i = 0; i < vq_size; i++) {
613 			struct vring_desc *start_dp = txr[i].tx_indir;
614 
615 			/* first indirect descriptor is always the tx header */
616 			if (!vtpci_packed_queue(hw)) {
617 				vring_desc_init_split(start_dp,
618 						      RTE_DIM(txr[i].tx_indir));
619 				start_dp->addr = txvq->virtio_net_hdr_mem
620 					+ i * sizeof(*txr)
621 					+ offsetof(struct virtio_tx_region,
622 						   tx_hdr);
623 				start_dp->len = hw->vtnet_hdr_size;
624 				start_dp->flags = VRING_DESC_F_NEXT;
625 			}
626 		}
627 	}
628 
629 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
630 		PMD_INIT_LOG(ERR, "setup_queue failed");
631 		return -EINVAL;
632 	}
633 
634 	return 0;
635 
636 fail_q_alloc:
637 	rte_free(sw_ring);
638 	rte_memzone_free(hdr_mz);
639 	rte_memzone_free(mz);
640 	rte_free(vq);
641 
642 	return ret;
643 }
644 
645 static void
646 virtio_free_queues(struct virtio_hw *hw)
647 {
648 	uint16_t nr_vq = virtio_get_nr_vq(hw);
649 	struct virtqueue *vq;
650 	int queue_type;
651 	uint16_t i;
652 
653 	if (hw->vqs == NULL)
654 		return;
655 
656 	for (i = 0; i < nr_vq; i++) {
657 		vq = hw->vqs[i];
658 		if (!vq)
659 			continue;
660 
661 		queue_type = virtio_get_queue_type(hw, i);
662 		if (queue_type == VTNET_RQ) {
663 			rte_free(vq->sw_ring);
664 			rte_memzone_free(vq->rxq.mz);
665 		} else if (queue_type == VTNET_TQ) {
666 			rte_memzone_free(vq->txq.mz);
667 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
668 		} else {
669 			rte_memzone_free(vq->cq.mz);
670 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
671 		}
672 
673 		rte_free(vq);
674 		hw->vqs[i] = NULL;
675 	}
676 
677 	rte_free(hw->vqs);
678 	hw->vqs = NULL;
679 }
680 
681 static int
682 virtio_alloc_queues(struct rte_eth_dev *dev)
683 {
684 	struct virtio_hw *hw = dev->data->dev_private;
685 	uint16_t nr_vq = virtio_get_nr_vq(hw);
686 	uint16_t i;
687 	int ret;
688 
689 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
690 	if (!hw->vqs) {
691 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
692 		return -ENOMEM;
693 	}
694 
695 	for (i = 0; i < nr_vq; i++) {
696 		ret = virtio_init_queue(dev, i);
697 		if (ret < 0) {
698 			virtio_free_queues(hw);
699 			return ret;
700 		}
701 	}
702 
703 	return 0;
704 }
705 
706 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
707 
708 static void
709 virtio_dev_close(struct rte_eth_dev *dev)
710 {
711 	struct virtio_hw *hw = dev->data->dev_private;
712 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
713 
714 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
715 
716 	if (!hw->opened)
717 		return;
718 	hw->opened = false;
719 
720 	/* reset the NIC */
721 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
722 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
723 	if (intr_conf->rxq)
724 		virtio_queues_unbind_intr(dev);
725 
726 	if (intr_conf->lsc || intr_conf->rxq) {
727 		virtio_intr_disable(dev);
728 		rte_intr_efd_disable(dev->intr_handle);
729 		rte_free(dev->intr_handle->intr_vec);
730 		dev->intr_handle->intr_vec = NULL;
731 	}
732 
733 	vtpci_reset(hw);
734 	virtio_dev_free_mbufs(dev);
735 	virtio_free_queues(hw);
736 }
737 
738 static void
739 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
740 {
741 	struct virtio_hw *hw = dev->data->dev_private;
742 	struct virtio_pmd_ctrl ctrl;
743 	int dlen[1];
744 	int ret;
745 
746 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
747 		PMD_INIT_LOG(INFO, "host does not support rx control");
748 		return;
749 	}
750 
751 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
752 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
753 	ctrl.data[0] = 1;
754 	dlen[0] = 1;
755 
756 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
757 	if (ret)
758 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
759 }
760 
761 static void
762 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
763 {
764 	struct virtio_hw *hw = dev->data->dev_private;
765 	struct virtio_pmd_ctrl ctrl;
766 	int dlen[1];
767 	int ret;
768 
769 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
770 		PMD_INIT_LOG(INFO, "host does not support rx control");
771 		return;
772 	}
773 
774 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
775 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
776 	ctrl.data[0] = 0;
777 	dlen[0] = 1;
778 
779 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
780 	if (ret)
781 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
782 }
783 
784 static void
785 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
786 {
787 	struct virtio_hw *hw = dev->data->dev_private;
788 	struct virtio_pmd_ctrl ctrl;
789 	int dlen[1];
790 	int ret;
791 
792 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
793 		PMD_INIT_LOG(INFO, "host does not support rx control");
794 		return;
795 	}
796 
797 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
798 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
799 	ctrl.data[0] = 1;
800 	dlen[0] = 1;
801 
802 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
803 	if (ret)
804 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
805 }
806 
807 static void
808 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
809 {
810 	struct virtio_hw *hw = dev->data->dev_private;
811 	struct virtio_pmd_ctrl ctrl;
812 	int dlen[1];
813 	int ret;
814 
815 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
816 		PMD_INIT_LOG(INFO, "host does not support rx control");
817 		return;
818 	}
819 
820 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
821 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
822 	ctrl.data[0] = 0;
823 	dlen[0] = 1;
824 
825 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
826 	if (ret)
827 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
828 }
829 
830 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
831 static int
832 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
833 {
834 	struct virtio_hw *hw = dev->data->dev_private;
835 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
836 				 hw->vtnet_hdr_size;
837 	uint32_t frame_size = mtu + ether_hdr_len;
838 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
839 
840 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
841 
842 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
843 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
844 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
845 		return -EINVAL;
846 	}
847 	return 0;
848 }
849 
850 static int
851 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
852 {
853 	struct virtio_hw *hw = dev->data->dev_private;
854 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
855 	struct virtqueue *vq = rxvq->vq;
856 
857 	virtqueue_enable_intr(vq);
858 	virtio_mb(hw->weak_barriers);
859 	return 0;
860 }
861 
862 static int
863 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
864 {
865 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
866 	struct virtqueue *vq = rxvq->vq;
867 
868 	virtqueue_disable_intr(vq);
869 	return 0;
870 }
871 
872 /*
873  * dev_ops for virtio, bare necessities for basic operation
874  */
875 static const struct eth_dev_ops virtio_eth_dev_ops = {
876 	.dev_configure           = virtio_dev_configure,
877 	.dev_start               = virtio_dev_start,
878 	.dev_stop                = virtio_dev_stop,
879 	.dev_close               = virtio_dev_close,
880 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
881 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
882 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
883 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
884 	.mtu_set                 = virtio_mtu_set,
885 	.dev_infos_get           = virtio_dev_info_get,
886 	.stats_get               = virtio_dev_stats_get,
887 	.xstats_get              = virtio_dev_xstats_get,
888 	.xstats_get_names        = virtio_dev_xstats_get_names,
889 	.stats_reset             = virtio_dev_stats_reset,
890 	.xstats_reset            = virtio_dev_stats_reset,
891 	.link_update             = virtio_dev_link_update,
892 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
893 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
894 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
895 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
896 	.rx_queue_release        = virtio_dev_queue_release,
897 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
898 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
899 	.tx_queue_release        = virtio_dev_queue_release,
900 	/* collect stats per queue */
901 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
902 	.vlan_filter_set         = virtio_vlan_filter_set,
903 	.mac_addr_add            = virtio_mac_addr_add,
904 	.mac_addr_remove         = virtio_mac_addr_remove,
905 	.mac_addr_set            = virtio_mac_addr_set,
906 };
907 
908 /*
909  * dev_ops for virtio-user in secondary processes, as we just have
910  * some limited supports currently.
911  */
912 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
913 	.dev_infos_get           = virtio_dev_info_get,
914 	.stats_get               = virtio_dev_stats_get,
915 	.xstats_get              = virtio_dev_xstats_get,
916 	.xstats_get_names        = virtio_dev_xstats_get_names,
917 	.stats_reset             = virtio_dev_stats_reset,
918 	.xstats_reset            = virtio_dev_stats_reset,
919 	/* collect stats per queue */
920 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
921 };
922 
923 static void
924 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
925 {
926 	unsigned i;
927 
928 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
929 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
930 		if (txvq == NULL)
931 			continue;
932 
933 		stats->opackets += txvq->stats.packets;
934 		stats->obytes += txvq->stats.bytes;
935 		stats->oerrors += txvq->stats.errors;
936 
937 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
938 			stats->q_opackets[i] = txvq->stats.packets;
939 			stats->q_obytes[i] = txvq->stats.bytes;
940 		}
941 	}
942 
943 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
944 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
945 		if (rxvq == NULL)
946 			continue;
947 
948 		stats->ipackets += rxvq->stats.packets;
949 		stats->ibytes += rxvq->stats.bytes;
950 		stats->ierrors += rxvq->stats.errors;
951 
952 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
953 			stats->q_ipackets[i] = rxvq->stats.packets;
954 			stats->q_ibytes[i] = rxvq->stats.bytes;
955 		}
956 	}
957 
958 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
959 }
960 
961 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
962 				       struct rte_eth_xstat_name *xstats_names,
963 				       __rte_unused unsigned limit)
964 {
965 	unsigned i;
966 	unsigned count = 0;
967 	unsigned t;
968 
969 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
970 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
971 
972 	if (xstats_names != NULL) {
973 		/* Note: limit checked in rte_eth_xstats_names() */
974 
975 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
976 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
977 			if (rxvq == NULL)
978 				continue;
979 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
980 				snprintf(xstats_names[count].name,
981 					sizeof(xstats_names[count].name),
982 					"rx_q%u_%s", i,
983 					rte_virtio_rxq_stat_strings[t].name);
984 				count++;
985 			}
986 		}
987 
988 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
989 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
990 			if (txvq == NULL)
991 				continue;
992 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
993 				snprintf(xstats_names[count].name,
994 					sizeof(xstats_names[count].name),
995 					"tx_q%u_%s", i,
996 					rte_virtio_txq_stat_strings[t].name);
997 				count++;
998 			}
999 		}
1000 		return count;
1001 	}
1002 	return nstats;
1003 }
1004 
1005 static int
1006 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1007 		      unsigned n)
1008 {
1009 	unsigned i;
1010 	unsigned count = 0;
1011 
1012 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1013 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1014 
1015 	if (n < nstats)
1016 		return nstats;
1017 
1018 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1019 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1020 
1021 		if (rxvq == NULL)
1022 			continue;
1023 
1024 		unsigned t;
1025 
1026 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1027 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1028 				rte_virtio_rxq_stat_strings[t].offset);
1029 			xstats[count].id = count;
1030 			count++;
1031 		}
1032 	}
1033 
1034 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1035 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1036 
1037 		if (txvq == NULL)
1038 			continue;
1039 
1040 		unsigned t;
1041 
1042 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1043 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1044 				rte_virtio_txq_stat_strings[t].offset);
1045 			xstats[count].id = count;
1046 			count++;
1047 		}
1048 	}
1049 
1050 	return count;
1051 }
1052 
1053 static int
1054 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1055 {
1056 	virtio_update_stats(dev, stats);
1057 
1058 	return 0;
1059 }
1060 
1061 static void
1062 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1063 {
1064 	unsigned int i;
1065 
1066 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1067 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1068 		if (txvq == NULL)
1069 			continue;
1070 
1071 		txvq->stats.packets = 0;
1072 		txvq->stats.bytes = 0;
1073 		txvq->stats.errors = 0;
1074 		txvq->stats.multicast = 0;
1075 		txvq->stats.broadcast = 0;
1076 		memset(txvq->stats.size_bins, 0,
1077 		       sizeof(txvq->stats.size_bins[0]) * 8);
1078 	}
1079 
1080 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1081 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1082 		if (rxvq == NULL)
1083 			continue;
1084 
1085 		rxvq->stats.packets = 0;
1086 		rxvq->stats.bytes = 0;
1087 		rxvq->stats.errors = 0;
1088 		rxvq->stats.multicast = 0;
1089 		rxvq->stats.broadcast = 0;
1090 		memset(rxvq->stats.size_bins, 0,
1091 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1092 	}
1093 }
1094 
1095 static void
1096 virtio_set_hwaddr(struct virtio_hw *hw)
1097 {
1098 	vtpci_write_dev_config(hw,
1099 			offsetof(struct virtio_net_config, mac),
1100 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1101 }
1102 
1103 static void
1104 virtio_get_hwaddr(struct virtio_hw *hw)
1105 {
1106 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1107 		vtpci_read_dev_config(hw,
1108 			offsetof(struct virtio_net_config, mac),
1109 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1110 	} else {
1111 		rte_eth_random_addr(&hw->mac_addr[0]);
1112 		virtio_set_hwaddr(hw);
1113 	}
1114 }
1115 
1116 static int
1117 virtio_mac_table_set(struct virtio_hw *hw,
1118 		     const struct virtio_net_ctrl_mac *uc,
1119 		     const struct virtio_net_ctrl_mac *mc)
1120 {
1121 	struct virtio_pmd_ctrl ctrl;
1122 	int err, len[2];
1123 
1124 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1125 		PMD_DRV_LOG(INFO, "host does not support mac table");
1126 		return -1;
1127 	}
1128 
1129 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1130 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1131 
1132 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1133 	memcpy(ctrl.data, uc, len[0]);
1134 
1135 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1136 	memcpy(ctrl.data + len[0], mc, len[1]);
1137 
1138 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1139 	if (err != 0)
1140 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1141 	return err;
1142 }
1143 
1144 static int
1145 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1146 		    uint32_t index, uint32_t vmdq __rte_unused)
1147 {
1148 	struct virtio_hw *hw = dev->data->dev_private;
1149 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1150 	unsigned int i;
1151 	struct virtio_net_ctrl_mac *uc, *mc;
1152 
1153 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1154 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1155 		return -EINVAL;
1156 	}
1157 
1158 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1159 		sizeof(uc->entries));
1160 	uc->entries = 0;
1161 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1162 		sizeof(mc->entries));
1163 	mc->entries = 0;
1164 
1165 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1166 		const struct rte_ether_addr *addr
1167 			= (i == index) ? mac_addr : addrs + i;
1168 		struct virtio_net_ctrl_mac *tbl
1169 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1170 
1171 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1172 	}
1173 
1174 	return virtio_mac_table_set(hw, uc, mc);
1175 }
1176 
1177 static void
1178 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1179 {
1180 	struct virtio_hw *hw = dev->data->dev_private;
1181 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1182 	struct virtio_net_ctrl_mac *uc, *mc;
1183 	unsigned int i;
1184 
1185 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1186 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1187 		return;
1188 	}
1189 
1190 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1191 		sizeof(uc->entries));
1192 	uc->entries = 0;
1193 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1194 		sizeof(mc->entries));
1195 	mc->entries = 0;
1196 
1197 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1198 		struct virtio_net_ctrl_mac *tbl;
1199 
1200 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1201 			continue;
1202 
1203 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1204 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1205 			RTE_ETHER_ADDR_LEN);
1206 	}
1207 
1208 	virtio_mac_table_set(hw, uc, mc);
1209 }
1210 
1211 static int
1212 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1213 {
1214 	struct virtio_hw *hw = dev->data->dev_private;
1215 
1216 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1217 
1218 	/* Use atomic update if available */
1219 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1220 		struct virtio_pmd_ctrl ctrl;
1221 		int len = RTE_ETHER_ADDR_LEN;
1222 
1223 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1224 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1225 
1226 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1227 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1228 	}
1229 
1230 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1231 		return -ENOTSUP;
1232 
1233 	virtio_set_hwaddr(hw);
1234 	return 0;
1235 }
1236 
1237 static int
1238 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1239 {
1240 	struct virtio_hw *hw = dev->data->dev_private;
1241 	struct virtio_pmd_ctrl ctrl;
1242 	int len;
1243 
1244 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1245 		return -ENOTSUP;
1246 
1247 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1248 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1249 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1250 	len = sizeof(vlan_id);
1251 
1252 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1253 }
1254 
1255 static int
1256 virtio_intr_enable(struct rte_eth_dev *dev)
1257 {
1258 	struct virtio_hw *hw = dev->data->dev_private;
1259 
1260 	if (rte_intr_enable(dev->intr_handle) < 0)
1261 		return -1;
1262 
1263 	if (!hw->virtio_user_dev)
1264 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1265 
1266 	return 0;
1267 }
1268 
1269 static int
1270 virtio_intr_disable(struct rte_eth_dev *dev)
1271 {
1272 	struct virtio_hw *hw = dev->data->dev_private;
1273 
1274 	if (rte_intr_disable(dev->intr_handle) < 0)
1275 		return -1;
1276 
1277 	if (!hw->virtio_user_dev)
1278 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1279 
1280 	return 0;
1281 }
1282 
1283 static int
1284 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1285 {
1286 	uint64_t host_features;
1287 
1288 	/* Prepare guest_features: feature that driver wants to support */
1289 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1290 		req_features);
1291 
1292 	/* Read device(host) feature bits */
1293 	host_features = VTPCI_OPS(hw)->get_features(hw);
1294 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1295 		host_features);
1296 
1297 	/* If supported, ensure MTU value is valid before acknowledging it. */
1298 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1299 		struct virtio_net_config config;
1300 
1301 		vtpci_read_dev_config(hw,
1302 			offsetof(struct virtio_net_config, mtu),
1303 			&config.mtu, sizeof(config.mtu));
1304 
1305 		if (config.mtu < RTE_ETHER_MIN_MTU)
1306 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1307 	}
1308 
1309 	/*
1310 	 * Negotiate features: Subset of device feature bits are written back
1311 	 * guest feature bits.
1312 	 */
1313 	hw->guest_features = req_features;
1314 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1315 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1316 		hw->guest_features);
1317 
1318 	if (hw->modern) {
1319 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1320 			PMD_INIT_LOG(ERR,
1321 				"VIRTIO_F_VERSION_1 features is not enabled.");
1322 			return -1;
1323 		}
1324 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1325 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1326 			PMD_INIT_LOG(ERR,
1327 				"failed to set FEATURES_OK status!");
1328 			return -1;
1329 		}
1330 	}
1331 
1332 	hw->req_guest_features = req_features;
1333 
1334 	return 0;
1335 }
1336 
1337 int
1338 virtio_dev_pause(struct rte_eth_dev *dev)
1339 {
1340 	struct virtio_hw *hw = dev->data->dev_private;
1341 
1342 	rte_spinlock_lock(&hw->state_lock);
1343 
1344 	if (hw->started == 0) {
1345 		/* Device is just stopped. */
1346 		rte_spinlock_unlock(&hw->state_lock);
1347 		return -1;
1348 	}
1349 	hw->started = 0;
1350 	/*
1351 	 * Prevent the worker threads from touching queues to avoid contention,
1352 	 * 1 ms should be enough for the ongoing Tx function to finish.
1353 	 */
1354 	rte_delay_ms(1);
1355 	return 0;
1356 }
1357 
1358 /*
1359  * Recover hw state to let the worker threads continue.
1360  */
1361 void
1362 virtio_dev_resume(struct rte_eth_dev *dev)
1363 {
1364 	struct virtio_hw *hw = dev->data->dev_private;
1365 
1366 	hw->started = 1;
1367 	rte_spinlock_unlock(&hw->state_lock);
1368 }
1369 
1370 /*
1371  * Should be called only after device is paused.
1372  */
1373 int
1374 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1375 		int nb_pkts)
1376 {
1377 	struct virtio_hw *hw = dev->data->dev_private;
1378 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1379 	int ret;
1380 
1381 	hw->inject_pkts = tx_pkts;
1382 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1383 	hw->inject_pkts = NULL;
1384 
1385 	return ret;
1386 }
1387 
1388 static void
1389 virtio_notify_peers(struct rte_eth_dev *dev)
1390 {
1391 	struct virtio_hw *hw = dev->data->dev_private;
1392 	struct virtnet_rx *rxvq;
1393 	struct rte_mbuf *rarp_mbuf;
1394 
1395 	if (!dev->data->rx_queues)
1396 		return;
1397 
1398 	rxvq = dev->data->rx_queues[0];
1399 	if (!rxvq)
1400 		return;
1401 
1402 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1403 			(struct rte_ether_addr *)hw->mac_addr);
1404 	if (rarp_mbuf == NULL) {
1405 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1406 		return;
1407 	}
1408 
1409 	/* If virtio port just stopped, no need to send RARP */
1410 	if (virtio_dev_pause(dev) < 0) {
1411 		rte_pktmbuf_free(rarp_mbuf);
1412 		return;
1413 	}
1414 
1415 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1416 	virtio_dev_resume(dev);
1417 }
1418 
1419 static void
1420 virtio_ack_link_announce(struct rte_eth_dev *dev)
1421 {
1422 	struct virtio_hw *hw = dev->data->dev_private;
1423 	struct virtio_pmd_ctrl ctrl;
1424 
1425 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1426 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1427 
1428 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1429 }
1430 
1431 /*
1432  * Process virtio config changed interrupt. Call the callback
1433  * if link state changed, generate gratuitous RARP packet if
1434  * the status indicates an ANNOUNCE.
1435  */
1436 void
1437 virtio_interrupt_handler(void *param)
1438 {
1439 	struct rte_eth_dev *dev = param;
1440 	struct virtio_hw *hw = dev->data->dev_private;
1441 	uint8_t isr;
1442 	uint16_t status;
1443 
1444 	/* Read interrupt status which clears interrupt */
1445 	isr = vtpci_isr(hw);
1446 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1447 
1448 	if (virtio_intr_enable(dev) < 0)
1449 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1450 
1451 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1452 		if (virtio_dev_link_update(dev, 0) == 0)
1453 			_rte_eth_dev_callback_process(dev,
1454 						      RTE_ETH_EVENT_INTR_LSC,
1455 						      NULL);
1456 
1457 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1458 			vtpci_read_dev_config(hw,
1459 				offsetof(struct virtio_net_config, status),
1460 				&status, sizeof(status));
1461 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1462 				virtio_notify_peers(dev);
1463 				if (hw->cvq)
1464 					virtio_ack_link_announce(dev);
1465 			}
1466 		}
1467 	}
1468 }
1469 
1470 /* set rx and tx handlers according to what is supported */
1471 static void
1472 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1473 {
1474 	struct virtio_hw *hw = eth_dev->data->dev_private;
1475 
1476 	if (vtpci_packed_queue(hw)) {
1477 		PMD_INIT_LOG(INFO,
1478 			"virtio: using packed ring %s Tx path on port %u",
1479 			hw->use_inorder_tx ? "inorder" : "standard",
1480 			eth_dev->data->port_id);
1481 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1482 	} else {
1483 		if (hw->use_inorder_tx) {
1484 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1485 				eth_dev->data->port_id);
1486 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1487 		} else {
1488 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1489 				eth_dev->data->port_id);
1490 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1491 		}
1492 	}
1493 
1494 	if (vtpci_packed_queue(hw)) {
1495 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1496 			PMD_INIT_LOG(INFO,
1497 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1498 				eth_dev->data->port_id);
1499 			eth_dev->rx_pkt_burst =
1500 				&virtio_recv_mergeable_pkts_packed;
1501 		} else {
1502 			PMD_INIT_LOG(INFO,
1503 				"virtio: using packed ring standard Rx path on port %u",
1504 				eth_dev->data->port_id);
1505 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1506 		}
1507 	} else {
1508 		if (hw->use_simple_rx) {
1509 			PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1510 				eth_dev->data->port_id);
1511 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1512 		} else if (hw->use_inorder_rx) {
1513 			PMD_INIT_LOG(INFO,
1514 				"virtio: using inorder Rx path on port %u",
1515 				eth_dev->data->port_id);
1516 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1517 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1518 			PMD_INIT_LOG(INFO,
1519 				"virtio: using mergeable buffer Rx path on port %u",
1520 				eth_dev->data->port_id);
1521 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1522 		} else {
1523 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1524 				eth_dev->data->port_id);
1525 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1526 		}
1527 	}
1528 
1529 }
1530 
1531 /* Only support 1:1 queue/interrupt mapping so far.
1532  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1533  * interrupt vectors (<N+1).
1534  */
1535 static int
1536 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1537 {
1538 	uint32_t i;
1539 	struct virtio_hw *hw = dev->data->dev_private;
1540 
1541 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1542 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1543 		dev->intr_handle->intr_vec[i] = i + 1;
1544 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1545 						 VIRTIO_MSI_NO_VECTOR) {
1546 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1547 			return -EBUSY;
1548 		}
1549 	}
1550 
1551 	return 0;
1552 }
1553 
1554 static void
1555 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1556 {
1557 	uint32_t i;
1558 	struct virtio_hw *hw = dev->data->dev_private;
1559 
1560 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1561 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1562 		VTPCI_OPS(hw)->set_queue_irq(hw,
1563 					     hw->vqs[i * VTNET_CQ],
1564 					     VIRTIO_MSI_NO_VECTOR);
1565 }
1566 
1567 static int
1568 virtio_configure_intr(struct rte_eth_dev *dev)
1569 {
1570 	struct virtio_hw *hw = dev->data->dev_private;
1571 
1572 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1573 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1574 		return -ENOTSUP;
1575 	}
1576 
1577 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1578 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1579 		return -1;
1580 	}
1581 
1582 	if (!dev->intr_handle->intr_vec) {
1583 		dev->intr_handle->intr_vec =
1584 			rte_zmalloc("intr_vec",
1585 				    hw->max_queue_pairs * sizeof(int), 0);
1586 		if (!dev->intr_handle->intr_vec) {
1587 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1588 				     hw->max_queue_pairs);
1589 			return -ENOMEM;
1590 		}
1591 	}
1592 
1593 	/* Re-register callback to update max_intr */
1594 	rte_intr_callback_unregister(dev->intr_handle,
1595 				     virtio_interrupt_handler,
1596 				     dev);
1597 	rte_intr_callback_register(dev->intr_handle,
1598 				   virtio_interrupt_handler,
1599 				   dev);
1600 
1601 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1602 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1603 	 * And for legacy devices, this should be done before queue/vec binding
1604 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1605 	 * (22) will be ignored.
1606 	 */
1607 	if (virtio_intr_enable(dev) < 0) {
1608 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1609 		return -1;
1610 	}
1611 
1612 	if (virtio_queues_bind_intr(dev) < 0) {
1613 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1614 		return -1;
1615 	}
1616 
1617 	return 0;
1618 }
1619 
1620 /* reset device and renegotiate features if needed */
1621 static int
1622 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1623 {
1624 	struct virtio_hw *hw = eth_dev->data->dev_private;
1625 	struct virtio_net_config *config;
1626 	struct virtio_net_config local_config;
1627 	struct rte_pci_device *pci_dev = NULL;
1628 	int ret;
1629 
1630 	/* Reset the device although not necessary at startup */
1631 	vtpci_reset(hw);
1632 
1633 	if (hw->vqs) {
1634 		virtio_dev_free_mbufs(eth_dev);
1635 		virtio_free_queues(hw);
1636 	}
1637 
1638 	/* Tell the host we've noticed this device. */
1639 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1640 
1641 	/* Tell the host we've known how to drive the device. */
1642 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1643 	if (virtio_negotiate_features(hw, req_features) < 0)
1644 		return -1;
1645 
1646 	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1647 
1648 	if (!hw->virtio_user_dev) {
1649 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1650 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1651 	}
1652 
1653 	/* If host does not support both status and MSI-X then disable LSC */
1654 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1655 	    hw->use_msix != VIRTIO_MSIX_NONE)
1656 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1657 	else
1658 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1659 
1660 	/* Setting up rx_header size for the device */
1661 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1662 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1663 	    vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1664 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1665 	else
1666 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1667 
1668 	/* Copy the permanent MAC address to: virtio_hw */
1669 	virtio_get_hwaddr(hw);
1670 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1671 			&eth_dev->data->mac_addrs[0]);
1672 	PMD_INIT_LOG(DEBUG,
1673 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1674 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1675 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1676 
1677 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1678 		config = &local_config;
1679 
1680 		vtpci_read_dev_config(hw,
1681 			offsetof(struct virtio_net_config, mac),
1682 			&config->mac, sizeof(config->mac));
1683 
1684 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1685 			vtpci_read_dev_config(hw,
1686 				offsetof(struct virtio_net_config, status),
1687 				&config->status, sizeof(config->status));
1688 		} else {
1689 			PMD_INIT_LOG(DEBUG,
1690 				     "VIRTIO_NET_F_STATUS is not supported");
1691 			config->status = 0;
1692 		}
1693 
1694 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1695 			vtpci_read_dev_config(hw,
1696 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1697 				&config->max_virtqueue_pairs,
1698 				sizeof(config->max_virtqueue_pairs));
1699 		} else {
1700 			PMD_INIT_LOG(DEBUG,
1701 				     "VIRTIO_NET_F_MQ is not supported");
1702 			config->max_virtqueue_pairs = 1;
1703 		}
1704 
1705 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1706 
1707 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1708 			vtpci_read_dev_config(hw,
1709 				offsetof(struct virtio_net_config, mtu),
1710 				&config->mtu,
1711 				sizeof(config->mtu));
1712 
1713 			/*
1714 			 * MTU value has already been checked at negotiation
1715 			 * time, but check again in case it has changed since
1716 			 * then, which should not happen.
1717 			 */
1718 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1719 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1720 						config->mtu);
1721 				return -1;
1722 			}
1723 
1724 			hw->max_mtu = config->mtu;
1725 			/* Set initial MTU to maximum one supported by vhost */
1726 			eth_dev->data->mtu = config->mtu;
1727 
1728 		} else {
1729 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1730 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1731 		}
1732 
1733 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1734 				config->max_virtqueue_pairs);
1735 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1736 		PMD_INIT_LOG(DEBUG,
1737 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1738 				config->mac[0], config->mac[1],
1739 				config->mac[2], config->mac[3],
1740 				config->mac[4], config->mac[5]);
1741 	} else {
1742 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1743 		hw->max_queue_pairs = 1;
1744 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1745 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1746 	}
1747 
1748 	ret = virtio_alloc_queues(eth_dev);
1749 	if (ret < 0)
1750 		return ret;
1751 
1752 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1753 		if (virtio_configure_intr(eth_dev) < 0) {
1754 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1755 			return -1;
1756 		}
1757 	}
1758 
1759 	vtpci_reinit_complete(hw);
1760 
1761 	if (pci_dev)
1762 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1763 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1764 			pci_dev->id.device_id);
1765 
1766 	return 0;
1767 }
1768 
1769 /*
1770  * Remap the PCI device again (IO port map for legacy device and
1771  * memory map for modern device), so that the secondary process
1772  * could have the PCI initiated correctly.
1773  */
1774 static int
1775 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1776 {
1777 	if (hw->modern) {
1778 		/*
1779 		 * We don't have to re-parse the PCI config space, since
1780 		 * rte_pci_map_device() makes sure the mapped address
1781 		 * in secondary process would equal to the one mapped in
1782 		 * the primary process: error will be returned if that
1783 		 * requirement is not met.
1784 		 *
1785 		 * That said, we could simply reuse all cap pointers
1786 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1787 		 * primary process, which is stored in shared memory.
1788 		 */
1789 		if (rte_pci_map_device(pci_dev)) {
1790 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1791 			return -1;
1792 		}
1793 	} else {
1794 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1795 			return -1;
1796 	}
1797 
1798 	return 0;
1799 }
1800 
1801 static void
1802 virtio_set_vtpci_ops(struct virtio_hw *hw)
1803 {
1804 #ifdef RTE_VIRTIO_USER
1805 	if (hw->virtio_user_dev)
1806 		VTPCI_OPS(hw) = &virtio_user_ops;
1807 	else
1808 #endif
1809 	if (hw->modern)
1810 		VTPCI_OPS(hw) = &modern_ops;
1811 	else
1812 		VTPCI_OPS(hw) = &legacy_ops;
1813 }
1814 
1815 /*
1816  * This function is based on probe() function in virtio_pci.c
1817  * It returns 0 on success.
1818  */
1819 int
1820 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1821 {
1822 	struct virtio_hw *hw = eth_dev->data->dev_private;
1823 	int ret;
1824 
1825 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1826 
1827 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1828 
1829 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1830 		if (!hw->virtio_user_dev) {
1831 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1832 			if (ret)
1833 				return ret;
1834 		}
1835 
1836 		virtio_set_vtpci_ops(hw);
1837 		set_rxtx_funcs(eth_dev);
1838 
1839 		return 0;
1840 	}
1841 
1842 	/* Allocate memory for storing MAC addresses */
1843 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1844 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1845 	if (eth_dev->data->mac_addrs == NULL) {
1846 		PMD_INIT_LOG(ERR,
1847 			"Failed to allocate %d bytes needed to store MAC addresses",
1848 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1849 		return -ENOMEM;
1850 	}
1851 
1852 	hw->port_id = eth_dev->data->port_id;
1853 	/* For virtio_user case the hw->virtio_user_dev is populated by
1854 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1855 	 */
1856 	if (!hw->virtio_user_dev) {
1857 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1858 		if (ret)
1859 			goto out;
1860 	}
1861 
1862 	/* reset device and negotiate default features */
1863 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1864 	if (ret < 0)
1865 		goto out;
1866 
1867 	return 0;
1868 
1869 out:
1870 	rte_free(eth_dev->data->mac_addrs);
1871 	eth_dev->data->mac_addrs = NULL;
1872 	return ret;
1873 }
1874 
1875 static int
1876 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1877 {
1878 	PMD_INIT_FUNC_TRACE();
1879 
1880 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1881 		return 0;
1882 
1883 	virtio_dev_stop(eth_dev);
1884 	virtio_dev_close(eth_dev);
1885 
1886 	eth_dev->dev_ops = NULL;
1887 	eth_dev->tx_pkt_burst = NULL;
1888 	eth_dev->rx_pkt_burst = NULL;
1889 
1890 	if (eth_dev->device)
1891 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1892 
1893 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1894 
1895 	return 0;
1896 }
1897 
1898 static int vdpa_check_handler(__rte_unused const char *key,
1899 		const char *value, __rte_unused void *opaque)
1900 {
1901 	if (strcmp(value, "1"))
1902 		return -1;
1903 
1904 	return 0;
1905 }
1906 
1907 static int
1908 vdpa_mode_selected(struct rte_devargs *devargs)
1909 {
1910 	struct rte_kvargs *kvlist;
1911 	const char *key = "vdpa";
1912 	int ret = 0;
1913 
1914 	if (devargs == NULL)
1915 		return 0;
1916 
1917 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1918 	if (kvlist == NULL)
1919 		return 0;
1920 
1921 	if (!rte_kvargs_count(kvlist, key))
1922 		goto exit;
1923 
1924 	/* vdpa mode selected when there's a key-value pair: vdpa=1 */
1925 	if (rte_kvargs_process(kvlist, key,
1926 				vdpa_check_handler, NULL) < 0) {
1927 		goto exit;
1928 	}
1929 	ret = 1;
1930 
1931 exit:
1932 	rte_kvargs_free(kvlist);
1933 	return ret;
1934 }
1935 
1936 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1937 	struct rte_pci_device *pci_dev)
1938 {
1939 	if (rte_eal_iopl_init() != 0) {
1940 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1941 		return 1;
1942 	}
1943 
1944 	/* virtio pmd skips probe if device needs to work in vdpa mode */
1945 	if (vdpa_mode_selected(pci_dev->device.devargs))
1946 		return 1;
1947 
1948 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1949 		eth_virtio_dev_init);
1950 }
1951 
1952 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1953 {
1954 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1955 }
1956 
1957 static struct rte_pci_driver rte_virtio_pmd = {
1958 	.driver = {
1959 		.name = "net_virtio",
1960 	},
1961 	.id_table = pci_id_virtio_map,
1962 	.drv_flags = 0,
1963 	.probe = eth_virtio_pci_probe,
1964 	.remove = eth_virtio_pci_remove,
1965 };
1966 
1967 RTE_INIT(rte_virtio_pmd_init)
1968 {
1969 	rte_eal_iopl_init();
1970 	rte_pci_register(&rte_virtio_pmd);
1971 }
1972 
1973 static bool
1974 rx_offload_enabled(struct virtio_hw *hw)
1975 {
1976 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
1977 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1978 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
1979 }
1980 
1981 static bool
1982 tx_offload_enabled(struct virtio_hw *hw)
1983 {
1984 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
1985 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
1986 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
1987 }
1988 
1989 /*
1990  * Configure virtio device
1991  * It returns 0 on success.
1992  */
1993 static int
1994 virtio_dev_configure(struct rte_eth_dev *dev)
1995 {
1996 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1997 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
1998 	struct virtio_hw *hw = dev->data->dev_private;
1999 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2000 		hw->vtnet_hdr_size;
2001 	uint64_t rx_offloads = rxmode->offloads;
2002 	uint64_t tx_offloads = txmode->offloads;
2003 	uint64_t req_features;
2004 	int ret;
2005 
2006 	PMD_INIT_LOG(DEBUG, "configure");
2007 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2008 
2009 	if (dev->data->dev_conf.intr_conf.rxq) {
2010 		ret = virtio_init_device(dev, hw->req_guest_features);
2011 		if (ret < 0)
2012 			return ret;
2013 	}
2014 
2015 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2016 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2017 
2018 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2019 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2020 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2021 
2022 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2023 		req_features |=
2024 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2025 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2026 
2027 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2028 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2029 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2030 
2031 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2032 		req_features |=
2033 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2034 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2035 
2036 	/* if request features changed, reinit the device */
2037 	if (req_features != hw->req_guest_features) {
2038 		ret = virtio_init_device(dev, req_features);
2039 		if (ret < 0)
2040 			return ret;
2041 	}
2042 
2043 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2044 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2045 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2046 		PMD_DRV_LOG(ERR,
2047 			"rx checksum not available on this host");
2048 		return -ENOTSUP;
2049 	}
2050 
2051 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2052 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2053 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2054 		PMD_DRV_LOG(ERR,
2055 			"Large Receive Offload not available on this host");
2056 		return -ENOTSUP;
2057 	}
2058 
2059 	/* start control queue */
2060 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2061 		virtio_dev_cq_start(dev);
2062 
2063 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2064 		hw->vlan_strip = 1;
2065 
2066 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2067 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2068 		PMD_DRV_LOG(ERR,
2069 			    "vlan filtering not available on this host");
2070 		return -ENOTSUP;
2071 	}
2072 
2073 	hw->has_tx_offload = tx_offload_enabled(hw);
2074 	hw->has_rx_offload = rx_offload_enabled(hw);
2075 
2076 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2077 		/* Enable vector (0) for Link State Intrerrupt */
2078 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2079 				VIRTIO_MSI_NO_VECTOR) {
2080 			PMD_DRV_LOG(ERR, "failed to set config vector");
2081 			return -EBUSY;
2082 		}
2083 
2084 	rte_spinlock_init(&hw->state_lock);
2085 
2086 	hw->use_simple_rx = 1;
2087 
2088 	if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2089 		hw->use_inorder_tx = 1;
2090 		hw->use_inorder_rx = 1;
2091 		hw->use_simple_rx = 0;
2092 	}
2093 
2094 	if (vtpci_packed_queue(hw)) {
2095 		hw->use_simple_rx = 0;
2096 		hw->use_inorder_rx = 0;
2097 	}
2098 
2099 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2100 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2101 		hw->use_simple_rx = 0;
2102 	}
2103 #endif
2104 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2105 		 hw->use_simple_rx = 0;
2106 	}
2107 
2108 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2109 			   DEV_RX_OFFLOAD_TCP_CKSUM |
2110 			   DEV_RX_OFFLOAD_TCP_LRO |
2111 			   DEV_RX_OFFLOAD_VLAN_STRIP))
2112 		hw->use_simple_rx = 0;
2113 
2114 	hw->opened = true;
2115 
2116 	return 0;
2117 }
2118 
2119 
2120 static int
2121 virtio_dev_start(struct rte_eth_dev *dev)
2122 {
2123 	uint16_t nb_queues, i;
2124 	struct virtnet_rx *rxvq;
2125 	struct virtnet_tx *txvq __rte_unused;
2126 	struct virtio_hw *hw = dev->data->dev_private;
2127 	int ret;
2128 
2129 	/* Finish the initialization of the queues */
2130 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2131 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2132 		if (ret < 0)
2133 			return ret;
2134 	}
2135 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2136 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2137 		if (ret < 0)
2138 			return ret;
2139 	}
2140 
2141 	/* check if lsc interrupt feature is enabled */
2142 	if (dev->data->dev_conf.intr_conf.lsc) {
2143 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2144 			PMD_DRV_LOG(ERR, "link status not supported by host");
2145 			return -ENOTSUP;
2146 		}
2147 	}
2148 
2149 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2150 	 * in device configure, but it could be unmapped  when device is
2151 	 * stopped.
2152 	 */
2153 	if (dev->data->dev_conf.intr_conf.lsc ||
2154 	    dev->data->dev_conf.intr_conf.rxq) {
2155 		virtio_intr_disable(dev);
2156 
2157 		/* Setup interrupt callback  */
2158 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2159 			rte_intr_callback_register(dev->intr_handle,
2160 						   virtio_interrupt_handler,
2161 						   dev);
2162 
2163 		if (virtio_intr_enable(dev) < 0) {
2164 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2165 			return -EIO;
2166 		}
2167 	}
2168 
2169 	/*Notify the backend
2170 	 *Otherwise the tap backend might already stop its queue due to fullness.
2171 	 *vhost backend will have no chance to be waked up
2172 	 */
2173 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2174 	if (hw->max_queue_pairs > 1) {
2175 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2176 			return -EINVAL;
2177 	}
2178 
2179 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2180 
2181 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2182 		rxvq = dev->data->rx_queues[i];
2183 		/* Flush the old packets */
2184 		virtqueue_rxvq_flush(rxvq->vq);
2185 		virtqueue_notify(rxvq->vq);
2186 	}
2187 
2188 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2189 		txvq = dev->data->tx_queues[i];
2190 		virtqueue_notify(txvq->vq);
2191 	}
2192 
2193 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2194 
2195 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2196 		rxvq = dev->data->rx_queues[i];
2197 		VIRTQUEUE_DUMP(rxvq->vq);
2198 	}
2199 
2200 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2201 		txvq = dev->data->tx_queues[i];
2202 		VIRTQUEUE_DUMP(txvq->vq);
2203 	}
2204 
2205 	set_rxtx_funcs(dev);
2206 	hw->started = true;
2207 
2208 	/* Initialize Link state */
2209 	virtio_dev_link_update(dev, 0);
2210 
2211 	return 0;
2212 }
2213 
2214 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2215 {
2216 	struct virtio_hw *hw = dev->data->dev_private;
2217 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2218 	const char *type __rte_unused;
2219 	unsigned int i, mbuf_num = 0;
2220 	struct virtqueue *vq;
2221 	struct rte_mbuf *buf;
2222 	int queue_type;
2223 
2224 	if (hw->vqs == NULL)
2225 		return;
2226 
2227 	for (i = 0; i < nr_vq; i++) {
2228 		vq = hw->vqs[i];
2229 		if (!vq)
2230 			continue;
2231 
2232 		queue_type = virtio_get_queue_type(hw, i);
2233 		if (queue_type == VTNET_RQ)
2234 			type = "rxq";
2235 		else if (queue_type == VTNET_TQ)
2236 			type = "txq";
2237 		else
2238 			continue;
2239 
2240 		PMD_INIT_LOG(DEBUG,
2241 			"Before freeing %s[%d] used and unused buf",
2242 			type, i);
2243 		VIRTQUEUE_DUMP(vq);
2244 
2245 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2246 			rte_pktmbuf_free(buf);
2247 			mbuf_num++;
2248 		}
2249 
2250 		PMD_INIT_LOG(DEBUG,
2251 			"After freeing %s[%d] used and unused buf",
2252 			type, i);
2253 		VIRTQUEUE_DUMP(vq);
2254 	}
2255 
2256 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2257 }
2258 
2259 /*
2260  * Stop device: disable interrupt and mark link down
2261  */
2262 static void
2263 virtio_dev_stop(struct rte_eth_dev *dev)
2264 {
2265 	struct virtio_hw *hw = dev->data->dev_private;
2266 	struct rte_eth_link link;
2267 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2268 
2269 	PMD_INIT_LOG(DEBUG, "stop");
2270 
2271 	rte_spinlock_lock(&hw->state_lock);
2272 	if (!hw->started)
2273 		goto out_unlock;
2274 	hw->started = false;
2275 
2276 	if (intr_conf->lsc || intr_conf->rxq) {
2277 		virtio_intr_disable(dev);
2278 
2279 		/* Reset interrupt callback  */
2280 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2281 			rte_intr_callback_unregister(dev->intr_handle,
2282 						     virtio_interrupt_handler,
2283 						     dev);
2284 		}
2285 	}
2286 
2287 	memset(&link, 0, sizeof(link));
2288 	rte_eth_linkstatus_set(dev, &link);
2289 out_unlock:
2290 	rte_spinlock_unlock(&hw->state_lock);
2291 }
2292 
2293 static int
2294 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2295 {
2296 	struct rte_eth_link link;
2297 	uint16_t status;
2298 	struct virtio_hw *hw = dev->data->dev_private;
2299 
2300 	memset(&link, 0, sizeof(link));
2301 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
2302 	link.link_speed  = ETH_SPEED_NUM_10G;
2303 	link.link_autoneg = ETH_LINK_FIXED;
2304 
2305 	if (!hw->started) {
2306 		link.link_status = ETH_LINK_DOWN;
2307 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2308 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2309 		vtpci_read_dev_config(hw,
2310 				offsetof(struct virtio_net_config, status),
2311 				&status, sizeof(status));
2312 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2313 			link.link_status = ETH_LINK_DOWN;
2314 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2315 				     dev->data->port_id);
2316 		} else {
2317 			link.link_status = ETH_LINK_UP;
2318 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2319 				     dev->data->port_id);
2320 		}
2321 	} else {
2322 		link.link_status = ETH_LINK_UP;
2323 	}
2324 
2325 	return rte_eth_linkstatus_set(dev, &link);
2326 }
2327 
2328 static int
2329 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2330 {
2331 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2332 	struct virtio_hw *hw = dev->data->dev_private;
2333 	uint64_t offloads = rxmode->offloads;
2334 
2335 	if (mask & ETH_VLAN_FILTER_MASK) {
2336 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2337 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2338 
2339 			PMD_DRV_LOG(NOTICE,
2340 				"vlan filtering not available on this host");
2341 
2342 			return -ENOTSUP;
2343 		}
2344 	}
2345 
2346 	if (mask & ETH_VLAN_STRIP_MASK)
2347 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2348 
2349 	return 0;
2350 }
2351 
2352 static void
2353 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2354 {
2355 	uint64_t tso_mask, host_features;
2356 	struct virtio_hw *hw = dev->data->dev_private;
2357 
2358 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2359 
2360 	dev_info->max_rx_queues =
2361 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2362 	dev_info->max_tx_queues =
2363 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2364 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2365 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2366 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2367 
2368 	host_features = VTPCI_OPS(hw)->get_features(hw);
2369 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2370 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2371 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2372 		dev_info->rx_offload_capa |=
2373 			DEV_RX_OFFLOAD_TCP_CKSUM |
2374 			DEV_RX_OFFLOAD_UDP_CKSUM;
2375 	}
2376 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2377 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2378 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2379 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2380 	if ((host_features & tso_mask) == tso_mask)
2381 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2382 
2383 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2384 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2385 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2386 		dev_info->tx_offload_capa |=
2387 			DEV_TX_OFFLOAD_UDP_CKSUM |
2388 			DEV_TX_OFFLOAD_TCP_CKSUM;
2389 	}
2390 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2391 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2392 	if ((host_features & tso_mask) == tso_mask)
2393 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2394 }
2395 
2396 /*
2397  * It enables testpmd to collect per queue stats.
2398  */
2399 static int
2400 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2401 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2402 __rte_unused uint8_t is_rx)
2403 {
2404 	return 0;
2405 }
2406 
2407 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2408 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2409 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2410 
2411 RTE_INIT(virtio_init_log)
2412 {
2413 	virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2414 	if (virtio_logtype_init >= 0)
2415 		rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2416 	virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2417 	if (virtio_logtype_driver >= 0)
2418 		rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2419 }
2420