xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 25d11a86c56d50947af33d0b79ede622809bd8b9)
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 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 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->ring_packed.desc_packed;
151 	struct virtio_pmd_ctrl *result;
152 	bool avail_wrap_counter;
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 	avail_wrap_counter = vq->avail_wrap_counter;
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->avail_wrap_counter ^= 1;
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 			VRING_DESC_F_AVAIL(vq->avail_wrap_counter) |
181 			VRING_DESC_F_USED(!vq->avail_wrap_counter);
182 		sum += dlen[k];
183 		vq->vq_free_cnt--;
184 		nb_descs++;
185 		if (++vq->vq_avail_idx >= vq->vq_nentries) {
186 			vq->vq_avail_idx -= vq->vq_nentries;
187 			vq->avail_wrap_counter ^= 1;
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 		VRING_DESC_F_AVAIL(vq->avail_wrap_counter) |
196 		VRING_DESC_F_USED(!vq->avail_wrap_counter);
197 	vq->vq_free_cnt--;
198 	nb_descs++;
199 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
200 		vq->vq_avail_idx -= vq->vq_nentries;
201 		vq->avail_wrap_counter ^= 1;
202 	}
203 
204 	virtio_wmb(vq->hw->weak_barriers);
205 	desc[head].flags = VRING_DESC_F_NEXT |
206 		VRING_DESC_F_AVAIL(avail_wrap_counter) |
207 		VRING_DESC_F_USED(!avail_wrap_counter);
208 
209 	virtio_wmb(vq->hw->weak_barriers);
210 	virtqueue_notify(vq);
211 
212 	/* wait for used descriptors in virtqueue */
213 	while (!desc_is_used(&desc[head], vq))
214 		usleep(100);
215 
216 	virtio_rmb(vq->hw->weak_barriers);
217 
218 	/* now get used descriptors */
219 	vq->vq_free_cnt += nb_descs;
220 	vq->vq_used_cons_idx += nb_descs;
221 	if (vq->vq_used_cons_idx >= vq->vq_nentries) {
222 		vq->vq_used_cons_idx -= vq->vq_nentries;
223 		vq->used_wrap_counter ^= 1;
224 	}
225 
226 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
227 			"vq->vq_avail_idx=%d\n"
228 			"vq->vq_used_cons_idx=%d\n"
229 			"vq->avail_wrap_counter=%d\n"
230 			"vq->used_wrap_counter=%d\n",
231 			vq->vq_free_cnt,
232 			vq->vq_avail_idx,
233 			vq->vq_used_cons_idx,
234 			vq->avail_wrap_counter,
235 			vq->used_wrap_counter);
236 
237 	result = cvq->virtio_net_hdr_mz->addr;
238 	return result;
239 }
240 
241 static int
242 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
243 		int *dlen, int pkt_num)
244 {
245 	uint32_t head, i;
246 	int k, sum = 0;
247 	virtio_net_ctrl_ack status = ~0;
248 	struct virtio_pmd_ctrl *result;
249 	struct virtqueue *vq;
250 
251 	ctrl->status = status;
252 
253 	if (!cvq || !cvq->vq) {
254 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
255 		return -1;
256 	}
257 
258 	rte_spinlock_lock(&cvq->lock);
259 	vq = cvq->vq;
260 	head = vq->vq_desc_head_idx;
261 
262 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
263 		"vq->hw->cvq = %p vq = %p",
264 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
265 
266 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
267 		rte_spinlock_unlock(&cvq->lock);
268 		return -1;
269 	}
270 
271 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
272 		sizeof(struct virtio_pmd_ctrl));
273 
274 	if (vtpci_packed_queue(vq->hw)) {
275 		result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
276 		goto out_unlock;
277 	}
278 
279 	/*
280 	 * Format is enforced in qemu code:
281 	 * One TX packet for header;
282 	 * At least one TX packet per argument;
283 	 * One RX packet for ACK.
284 	 */
285 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
286 	vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
287 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
288 	vq->vq_free_cnt--;
289 	i = vq->vq_ring.desc[head].next;
290 
291 	for (k = 0; k < pkt_num; k++) {
292 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
293 		vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
294 			+ sizeof(struct virtio_net_ctrl_hdr)
295 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
296 		vq->vq_ring.desc[i].len = dlen[k];
297 		sum += dlen[k];
298 		vq->vq_free_cnt--;
299 		i = vq->vq_ring.desc[i].next;
300 	}
301 
302 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
303 	vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
304 			+ sizeof(struct virtio_net_ctrl_hdr);
305 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
306 	vq->vq_free_cnt--;
307 
308 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
309 
310 	vq_update_avail_ring(vq, head);
311 	vq_update_avail_idx(vq);
312 
313 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
314 
315 	virtqueue_notify(vq);
316 
317 	rte_rmb();
318 	while (VIRTQUEUE_NUSED(vq) == 0) {
319 		rte_rmb();
320 		usleep(100);
321 	}
322 
323 	while (VIRTQUEUE_NUSED(vq)) {
324 		uint32_t idx, desc_idx, used_idx;
325 		struct vring_used_elem *uep;
326 
327 		used_idx = (uint32_t)(vq->vq_used_cons_idx
328 				& (vq->vq_nentries - 1));
329 		uep = &vq->vq_ring.used->ring[used_idx];
330 		idx = (uint32_t) uep->id;
331 		desc_idx = idx;
332 
333 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
334 			desc_idx = vq->vq_ring.desc[desc_idx].next;
335 			vq->vq_free_cnt++;
336 		}
337 
338 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
339 		vq->vq_desc_head_idx = idx;
340 
341 		vq->vq_used_cons_idx++;
342 		vq->vq_free_cnt++;
343 	}
344 
345 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
346 			vq->vq_free_cnt, vq->vq_desc_head_idx);
347 
348 	result = cvq->virtio_net_hdr_mz->addr;
349 
350 out_unlock:
351 	rte_spinlock_unlock(&cvq->lock);
352 	return result->status;
353 }
354 
355 static int
356 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
357 {
358 	struct virtio_hw *hw = dev->data->dev_private;
359 	struct virtio_pmd_ctrl ctrl;
360 	int dlen[1];
361 	int ret;
362 
363 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
364 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
365 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
366 
367 	dlen[0] = sizeof(uint16_t);
368 
369 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
370 	if (ret) {
371 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
372 			  "failed, this is too late now...");
373 		return -EINVAL;
374 	}
375 
376 	return 0;
377 }
378 
379 static void
380 virtio_dev_queue_release(void *queue __rte_unused)
381 {
382 	/* do nothing */
383 }
384 
385 static uint16_t
386 virtio_get_nr_vq(struct virtio_hw *hw)
387 {
388 	uint16_t nr_vq = hw->max_queue_pairs * 2;
389 
390 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
391 		nr_vq += 1;
392 
393 	return nr_vq;
394 }
395 
396 static void
397 virtio_init_vring(struct virtqueue *vq)
398 {
399 	int size = vq->vq_nentries;
400 	struct vring *vr = &vq->vq_ring;
401 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
402 
403 	PMD_INIT_FUNC_TRACE();
404 
405 	memset(ring_mem, 0, vq->vq_ring_size);
406 
407 	vq->vq_used_cons_idx = 0;
408 	vq->vq_desc_head_idx = 0;
409 	vq->vq_avail_idx = 0;
410 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
411 	vq->vq_free_cnt = vq->vq_nentries;
412 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
413 	if (vtpci_packed_queue(vq->hw)) {
414 		vring_init_packed(&vq->ring_packed, ring_mem,
415 				  VIRTIO_PCI_VRING_ALIGN, size);
416 		vring_desc_init_packed(vq, size);
417 	} else {
418 		vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
419 		vring_desc_init_split(vr->desc, size);
420 	}
421 	/*
422 	 * Disable device(host) interrupting guest
423 	 */
424 	virtqueue_disable_intr(vq);
425 }
426 
427 static int
428 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
429 {
430 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
431 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
432 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
433 	unsigned int vq_size, size;
434 	struct virtio_hw *hw = dev->data->dev_private;
435 	struct virtnet_rx *rxvq = NULL;
436 	struct virtnet_tx *txvq = NULL;
437 	struct virtnet_ctl *cvq = NULL;
438 	struct virtqueue *vq;
439 	size_t sz_hdr_mz = 0;
440 	void *sw_ring = NULL;
441 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
442 	int ret;
443 	int numa_node = dev->device->numa_node;
444 
445 	PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
446 			vtpci_queue_idx, numa_node);
447 
448 	/*
449 	 * Read the virtqueue size from the Queue Size field
450 	 * Always power of 2 and if 0 virtqueue does not exist
451 	 */
452 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
453 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
454 	if (vq_size == 0) {
455 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
456 		return -EINVAL;
457 	}
458 
459 	if (!rte_is_power_of_2(vq_size)) {
460 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
461 		return -EINVAL;
462 	}
463 
464 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
465 		 dev->data->port_id, vtpci_queue_idx);
466 
467 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
468 				vq_size * sizeof(struct vq_desc_extra),
469 				RTE_CACHE_LINE_SIZE);
470 	if (queue_type == VTNET_TQ) {
471 		/*
472 		 * For each xmit packet, allocate a virtio_net_hdr
473 		 * and indirect ring elements
474 		 */
475 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
476 	} else if (queue_type == VTNET_CQ) {
477 		/* Allocate a page for control vq command, data and status */
478 		sz_hdr_mz = PAGE_SIZE;
479 	}
480 
481 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
482 				numa_node);
483 	if (vq == NULL) {
484 		PMD_INIT_LOG(ERR, "can not allocate vq");
485 		return -ENOMEM;
486 	}
487 	hw->vqs[vtpci_queue_idx] = vq;
488 
489 	vq->hw = hw;
490 	vq->vq_queue_index = vtpci_queue_idx;
491 	vq->vq_nentries = vq_size;
492 	vq->event_flags_shadow = 0;
493 	if (vtpci_packed_queue(hw)) {
494 		vq->avail_wrap_counter = 1;
495 		vq->used_wrap_counter = 1;
496 		vq->avail_used_flags =
497 			VRING_DESC_F_AVAIL(vq->avail_wrap_counter) |
498 			VRING_DESC_F_USED(!vq->avail_wrap_counter);
499 	}
500 
501 	/*
502 	 * Reserve a memzone for vring elements
503 	 */
504 	size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
505 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
506 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
507 		     size, vq->vq_ring_size);
508 
509 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
510 			numa_node, RTE_MEMZONE_IOVA_CONTIG,
511 			VIRTIO_PCI_VRING_ALIGN);
512 	if (mz == NULL) {
513 		if (rte_errno == EEXIST)
514 			mz = rte_memzone_lookup(vq_name);
515 		if (mz == NULL) {
516 			ret = -ENOMEM;
517 			goto fail_q_alloc;
518 		}
519 	}
520 
521 	memset(mz->addr, 0, mz->len);
522 
523 	vq->vq_ring_mem = mz->iova;
524 	vq->vq_ring_virt_mem = mz->addr;
525 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
526 		     (uint64_t)mz->iova);
527 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
528 		     (uint64_t)(uintptr_t)mz->addr);
529 
530 	virtio_init_vring(vq);
531 
532 	if (sz_hdr_mz) {
533 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
534 			 dev->data->port_id, vtpci_queue_idx);
535 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
536 				numa_node, RTE_MEMZONE_IOVA_CONTIG,
537 				RTE_CACHE_LINE_SIZE);
538 		if (hdr_mz == NULL) {
539 			if (rte_errno == EEXIST)
540 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
541 			if (hdr_mz == NULL) {
542 				ret = -ENOMEM;
543 				goto fail_q_alloc;
544 			}
545 		}
546 	}
547 
548 	if (queue_type == VTNET_RQ) {
549 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
550 			       sizeof(vq->sw_ring[0]);
551 
552 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
553 				RTE_CACHE_LINE_SIZE, numa_node);
554 		if (!sw_ring) {
555 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
556 			ret = -ENOMEM;
557 			goto fail_q_alloc;
558 		}
559 
560 		vq->sw_ring = sw_ring;
561 		rxvq = &vq->rxq;
562 		rxvq->vq = vq;
563 		rxvq->port_id = dev->data->port_id;
564 		rxvq->mz = mz;
565 	} else if (queue_type == VTNET_TQ) {
566 		txvq = &vq->txq;
567 		txvq->vq = vq;
568 		txvq->port_id = dev->data->port_id;
569 		txvq->mz = mz;
570 		txvq->virtio_net_hdr_mz = hdr_mz;
571 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
572 	} else if (queue_type == VTNET_CQ) {
573 		cvq = &vq->cq;
574 		cvq->vq = vq;
575 		cvq->mz = mz;
576 		cvq->virtio_net_hdr_mz = hdr_mz;
577 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
578 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
579 
580 		hw->cvq = cvq;
581 	}
582 
583 	/* For virtio_user case (that is when hw->dev is NULL), we use
584 	 * virtual address. And we need properly set _offset_, please see
585 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
586 	 */
587 	if (!hw->virtio_user_dev)
588 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
589 	else {
590 		vq->vq_ring_mem = (uintptr_t)mz->addr;
591 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
592 		if (queue_type == VTNET_TQ)
593 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
594 		else if (queue_type == VTNET_CQ)
595 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
596 	}
597 
598 	if (queue_type == VTNET_TQ) {
599 		struct virtio_tx_region *txr;
600 		unsigned int i;
601 
602 		txr = hdr_mz->addr;
603 		memset(txr, 0, vq_size * sizeof(*txr));
604 		for (i = 0; i < vq_size; i++) {
605 			struct vring_desc *start_dp = txr[i].tx_indir;
606 			struct vring_packed_desc *start_dp_packed =
607 				txr[i].tx_indir_pq;
608 
609 			/* first indirect descriptor is always the tx header */
610 			if (vtpci_packed_queue(hw)) {
611 				start_dp_packed->addr = txvq->virtio_net_hdr_mem
612 					+ i * sizeof(*txr)
613 					+ offsetof(struct virtio_tx_region,
614 						   tx_hdr);
615 				start_dp_packed->len = hw->vtnet_hdr_size;
616 			} else {
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 = 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 < ETHER_MIN_MTU || frame_size > max_frame_size) {
843 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
844 			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 virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
854 	struct virtqueue *vq = rxvq->vq;
855 
856 	virtqueue_enable_intr(vq);
857 	return 0;
858 }
859 
860 static int
861 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
862 {
863 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
864 	struct virtqueue *vq = rxvq->vq;
865 
866 	virtqueue_disable_intr(vq);
867 	return 0;
868 }
869 
870 /*
871  * dev_ops for virtio, bare necessities for basic operation
872  */
873 static const struct eth_dev_ops virtio_eth_dev_ops = {
874 	.dev_configure           = virtio_dev_configure,
875 	.dev_start               = virtio_dev_start,
876 	.dev_stop                = virtio_dev_stop,
877 	.dev_close               = virtio_dev_close,
878 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
879 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
880 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
881 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
882 	.mtu_set                 = virtio_mtu_set,
883 	.dev_infos_get           = virtio_dev_info_get,
884 	.stats_get               = virtio_dev_stats_get,
885 	.xstats_get              = virtio_dev_xstats_get,
886 	.xstats_get_names        = virtio_dev_xstats_get_names,
887 	.stats_reset             = virtio_dev_stats_reset,
888 	.xstats_reset            = virtio_dev_stats_reset,
889 	.link_update             = virtio_dev_link_update,
890 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
891 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
892 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
893 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
894 	.rx_queue_release        = virtio_dev_queue_release,
895 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
896 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
897 	.tx_queue_release        = virtio_dev_queue_release,
898 	/* collect stats per queue */
899 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
900 	.vlan_filter_set         = virtio_vlan_filter_set,
901 	.mac_addr_add            = virtio_mac_addr_add,
902 	.mac_addr_remove         = virtio_mac_addr_remove,
903 	.mac_addr_set            = virtio_mac_addr_set,
904 };
905 
906 static void
907 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
908 {
909 	unsigned i;
910 
911 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
912 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
913 		if (txvq == NULL)
914 			continue;
915 
916 		stats->opackets += txvq->stats.packets;
917 		stats->obytes += txvq->stats.bytes;
918 		stats->oerrors += txvq->stats.errors;
919 
920 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
921 			stats->q_opackets[i] = txvq->stats.packets;
922 			stats->q_obytes[i] = txvq->stats.bytes;
923 		}
924 	}
925 
926 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
927 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
928 		if (rxvq == NULL)
929 			continue;
930 
931 		stats->ipackets += rxvq->stats.packets;
932 		stats->ibytes += rxvq->stats.bytes;
933 		stats->ierrors += rxvq->stats.errors;
934 
935 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
936 			stats->q_ipackets[i] = rxvq->stats.packets;
937 			stats->q_ibytes[i] = rxvq->stats.bytes;
938 		}
939 	}
940 
941 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
942 }
943 
944 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
945 				       struct rte_eth_xstat_name *xstats_names,
946 				       __rte_unused unsigned limit)
947 {
948 	unsigned i;
949 	unsigned count = 0;
950 	unsigned t;
951 
952 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
953 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
954 
955 	if (xstats_names != NULL) {
956 		/* Note: limit checked in rte_eth_xstats_names() */
957 
958 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
959 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
960 			if (rxvq == NULL)
961 				continue;
962 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
963 				snprintf(xstats_names[count].name,
964 					sizeof(xstats_names[count].name),
965 					"rx_q%u_%s", i,
966 					rte_virtio_rxq_stat_strings[t].name);
967 				count++;
968 			}
969 		}
970 
971 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
972 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
973 			if (txvq == NULL)
974 				continue;
975 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
976 				snprintf(xstats_names[count].name,
977 					sizeof(xstats_names[count].name),
978 					"tx_q%u_%s", i,
979 					rte_virtio_txq_stat_strings[t].name);
980 				count++;
981 			}
982 		}
983 		return count;
984 	}
985 	return nstats;
986 }
987 
988 static int
989 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
990 		      unsigned n)
991 {
992 	unsigned i;
993 	unsigned count = 0;
994 
995 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
996 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
997 
998 	if (n < nstats)
999 		return nstats;
1000 
1001 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1002 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1003 
1004 		if (rxvq == NULL)
1005 			continue;
1006 
1007 		unsigned t;
1008 
1009 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1010 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1011 				rte_virtio_rxq_stat_strings[t].offset);
1012 			xstats[count].id = count;
1013 			count++;
1014 		}
1015 	}
1016 
1017 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1018 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1019 
1020 		if (txvq == NULL)
1021 			continue;
1022 
1023 		unsigned t;
1024 
1025 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1026 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1027 				rte_virtio_txq_stat_strings[t].offset);
1028 			xstats[count].id = count;
1029 			count++;
1030 		}
1031 	}
1032 
1033 	return count;
1034 }
1035 
1036 static int
1037 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1038 {
1039 	virtio_update_stats(dev, stats);
1040 
1041 	return 0;
1042 }
1043 
1044 static void
1045 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1046 {
1047 	unsigned int i;
1048 
1049 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1050 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1051 		if (txvq == NULL)
1052 			continue;
1053 
1054 		txvq->stats.packets = 0;
1055 		txvq->stats.bytes = 0;
1056 		txvq->stats.errors = 0;
1057 		txvq->stats.multicast = 0;
1058 		txvq->stats.broadcast = 0;
1059 		memset(txvq->stats.size_bins, 0,
1060 		       sizeof(txvq->stats.size_bins[0]) * 8);
1061 	}
1062 
1063 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1064 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1065 		if (rxvq == NULL)
1066 			continue;
1067 
1068 		rxvq->stats.packets = 0;
1069 		rxvq->stats.bytes = 0;
1070 		rxvq->stats.errors = 0;
1071 		rxvq->stats.multicast = 0;
1072 		rxvq->stats.broadcast = 0;
1073 		memset(rxvq->stats.size_bins, 0,
1074 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1075 	}
1076 }
1077 
1078 static void
1079 virtio_set_hwaddr(struct virtio_hw *hw)
1080 {
1081 	vtpci_write_dev_config(hw,
1082 			offsetof(struct virtio_net_config, mac),
1083 			&hw->mac_addr, ETHER_ADDR_LEN);
1084 }
1085 
1086 static void
1087 virtio_get_hwaddr(struct virtio_hw *hw)
1088 {
1089 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1090 		vtpci_read_dev_config(hw,
1091 			offsetof(struct virtio_net_config, mac),
1092 			&hw->mac_addr, ETHER_ADDR_LEN);
1093 	} else {
1094 		eth_random_addr(&hw->mac_addr[0]);
1095 		virtio_set_hwaddr(hw);
1096 	}
1097 }
1098 
1099 static int
1100 virtio_mac_table_set(struct virtio_hw *hw,
1101 		     const struct virtio_net_ctrl_mac *uc,
1102 		     const struct virtio_net_ctrl_mac *mc)
1103 {
1104 	struct virtio_pmd_ctrl ctrl;
1105 	int err, len[2];
1106 
1107 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1108 		PMD_DRV_LOG(INFO, "host does not support mac table");
1109 		return -1;
1110 	}
1111 
1112 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1113 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1114 
1115 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1116 	memcpy(ctrl.data, uc, len[0]);
1117 
1118 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1119 	memcpy(ctrl.data + len[0], mc, len[1]);
1120 
1121 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1122 	if (err != 0)
1123 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1124 	return err;
1125 }
1126 
1127 static int
1128 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1129 		    uint32_t index, uint32_t vmdq __rte_unused)
1130 {
1131 	struct virtio_hw *hw = dev->data->dev_private;
1132 	const struct ether_addr *addrs = dev->data->mac_addrs;
1133 	unsigned int i;
1134 	struct virtio_net_ctrl_mac *uc, *mc;
1135 
1136 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1137 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1138 		return -EINVAL;
1139 	}
1140 
1141 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1142 	uc->entries = 0;
1143 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1144 	mc->entries = 0;
1145 
1146 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1147 		const struct ether_addr *addr
1148 			= (i == index) ? mac_addr : addrs + i;
1149 		struct virtio_net_ctrl_mac *tbl
1150 			= is_multicast_ether_addr(addr) ? mc : uc;
1151 
1152 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1153 	}
1154 
1155 	return virtio_mac_table_set(hw, uc, mc);
1156 }
1157 
1158 static void
1159 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1160 {
1161 	struct virtio_hw *hw = dev->data->dev_private;
1162 	struct ether_addr *addrs = dev->data->mac_addrs;
1163 	struct virtio_net_ctrl_mac *uc, *mc;
1164 	unsigned int i;
1165 
1166 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1167 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1168 		return;
1169 	}
1170 
1171 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1172 	uc->entries = 0;
1173 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1174 	mc->entries = 0;
1175 
1176 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1177 		struct virtio_net_ctrl_mac *tbl;
1178 
1179 		if (i == index || is_zero_ether_addr(addrs + i))
1180 			continue;
1181 
1182 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1183 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1184 	}
1185 
1186 	virtio_mac_table_set(hw, uc, mc);
1187 }
1188 
1189 static int
1190 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1191 {
1192 	struct virtio_hw *hw = dev->data->dev_private;
1193 
1194 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1195 
1196 	/* Use atomic update if available */
1197 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1198 		struct virtio_pmd_ctrl ctrl;
1199 		int len = ETHER_ADDR_LEN;
1200 
1201 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1202 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1203 
1204 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1205 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1206 	}
1207 
1208 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1209 		return -ENOTSUP;
1210 
1211 	virtio_set_hwaddr(hw);
1212 	return 0;
1213 }
1214 
1215 static int
1216 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1217 {
1218 	struct virtio_hw *hw = dev->data->dev_private;
1219 	struct virtio_pmd_ctrl ctrl;
1220 	int len;
1221 
1222 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1223 		return -ENOTSUP;
1224 
1225 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1226 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1227 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1228 	len = sizeof(vlan_id);
1229 
1230 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1231 }
1232 
1233 static int
1234 virtio_intr_enable(struct rte_eth_dev *dev)
1235 {
1236 	struct virtio_hw *hw = dev->data->dev_private;
1237 
1238 	if (rte_intr_enable(dev->intr_handle) < 0)
1239 		return -1;
1240 
1241 	if (!hw->virtio_user_dev)
1242 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1243 
1244 	return 0;
1245 }
1246 
1247 static int
1248 virtio_intr_disable(struct rte_eth_dev *dev)
1249 {
1250 	struct virtio_hw *hw = dev->data->dev_private;
1251 
1252 	if (rte_intr_disable(dev->intr_handle) < 0)
1253 		return -1;
1254 
1255 	if (!hw->virtio_user_dev)
1256 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1257 
1258 	return 0;
1259 }
1260 
1261 static int
1262 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1263 {
1264 	uint64_t host_features;
1265 
1266 	/* Prepare guest_features: feature that driver wants to support */
1267 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1268 		req_features);
1269 
1270 	/* Read device(host) feature bits */
1271 	host_features = VTPCI_OPS(hw)->get_features(hw);
1272 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1273 		host_features);
1274 
1275 	/* If supported, ensure MTU value is valid before acknowledging it. */
1276 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1277 		struct virtio_net_config config;
1278 
1279 		vtpci_read_dev_config(hw,
1280 			offsetof(struct virtio_net_config, mtu),
1281 			&config.mtu, sizeof(config.mtu));
1282 
1283 		if (config.mtu < ETHER_MIN_MTU)
1284 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1285 	}
1286 
1287 	/*
1288 	 * Negotiate features: Subset of device feature bits are written back
1289 	 * guest feature bits.
1290 	 */
1291 	hw->guest_features = req_features;
1292 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1293 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1294 		hw->guest_features);
1295 
1296 	if (hw->modern) {
1297 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1298 			PMD_INIT_LOG(ERR,
1299 				"VIRTIO_F_VERSION_1 features is not enabled.");
1300 			return -1;
1301 		}
1302 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1303 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1304 			PMD_INIT_LOG(ERR,
1305 				"failed to set FEATURES_OK status!");
1306 			return -1;
1307 		}
1308 	}
1309 
1310 	hw->req_guest_features = req_features;
1311 
1312 	return 0;
1313 }
1314 
1315 int
1316 virtio_dev_pause(struct rte_eth_dev *dev)
1317 {
1318 	struct virtio_hw *hw = dev->data->dev_private;
1319 
1320 	rte_spinlock_lock(&hw->state_lock);
1321 
1322 	if (hw->started == 0) {
1323 		/* Device is just stopped. */
1324 		rte_spinlock_unlock(&hw->state_lock);
1325 		return -1;
1326 	}
1327 	hw->started = 0;
1328 	/*
1329 	 * Prevent the worker threads from touching queues to avoid contention,
1330 	 * 1 ms should be enough for the ongoing Tx function to finish.
1331 	 */
1332 	rte_delay_ms(1);
1333 	return 0;
1334 }
1335 
1336 /*
1337  * Recover hw state to let the worker threads continue.
1338  */
1339 void
1340 virtio_dev_resume(struct rte_eth_dev *dev)
1341 {
1342 	struct virtio_hw *hw = dev->data->dev_private;
1343 
1344 	hw->started = 1;
1345 	rte_spinlock_unlock(&hw->state_lock);
1346 }
1347 
1348 /*
1349  * Should be called only after device is paused.
1350  */
1351 int
1352 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1353 		int nb_pkts)
1354 {
1355 	struct virtio_hw *hw = dev->data->dev_private;
1356 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1357 	int ret;
1358 
1359 	hw->inject_pkts = tx_pkts;
1360 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1361 	hw->inject_pkts = NULL;
1362 
1363 	return ret;
1364 }
1365 
1366 static void
1367 virtio_notify_peers(struct rte_eth_dev *dev)
1368 {
1369 	struct virtio_hw *hw = dev->data->dev_private;
1370 	struct virtnet_rx *rxvq;
1371 	struct rte_mbuf *rarp_mbuf;
1372 
1373 	if (!dev->data->rx_queues)
1374 		return;
1375 
1376 	rxvq = dev->data->rx_queues[0];
1377 	if (!rxvq)
1378 		return;
1379 
1380 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1381 			(struct ether_addr *)hw->mac_addr);
1382 	if (rarp_mbuf == NULL) {
1383 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1384 		return;
1385 	}
1386 
1387 	/* If virtio port just stopped, no need to send RARP */
1388 	if (virtio_dev_pause(dev) < 0) {
1389 		rte_pktmbuf_free(rarp_mbuf);
1390 		return;
1391 	}
1392 
1393 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1394 	virtio_dev_resume(dev);
1395 }
1396 
1397 static void
1398 virtio_ack_link_announce(struct rte_eth_dev *dev)
1399 {
1400 	struct virtio_hw *hw = dev->data->dev_private;
1401 	struct virtio_pmd_ctrl ctrl;
1402 
1403 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1404 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1405 
1406 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1407 }
1408 
1409 /*
1410  * Process virtio config changed interrupt. Call the callback
1411  * if link state changed, generate gratuitous RARP packet if
1412  * the status indicates an ANNOUNCE.
1413  */
1414 void
1415 virtio_interrupt_handler(void *param)
1416 {
1417 	struct rte_eth_dev *dev = param;
1418 	struct virtio_hw *hw = dev->data->dev_private;
1419 	uint8_t isr;
1420 	uint16_t status;
1421 
1422 	/* Read interrupt status which clears interrupt */
1423 	isr = vtpci_isr(hw);
1424 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1425 
1426 	if (virtio_intr_enable(dev) < 0)
1427 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1428 
1429 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1430 		if (virtio_dev_link_update(dev, 0) == 0)
1431 			_rte_eth_dev_callback_process(dev,
1432 						      RTE_ETH_EVENT_INTR_LSC,
1433 						      NULL);
1434 
1435 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1436 			vtpci_read_dev_config(hw,
1437 				offsetof(struct virtio_net_config, status),
1438 				&status, sizeof(status));
1439 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1440 				virtio_notify_peers(dev);
1441 				if (hw->cvq)
1442 					virtio_ack_link_announce(dev);
1443 			}
1444 		}
1445 	}
1446 }
1447 
1448 /* set rx and tx handlers according to what is supported */
1449 static void
1450 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1451 {
1452 	struct virtio_hw *hw = eth_dev->data->dev_private;
1453 
1454 	if (vtpci_packed_queue(hw)) {
1455 		PMD_INIT_LOG(INFO,
1456 			"virtio: using packed ring %s Tx path on port %u",
1457 			hw->use_inorder_tx ? "inorder" : "standard",
1458 			eth_dev->data->port_id);
1459 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1460 	} else {
1461 		if (hw->use_inorder_tx) {
1462 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1463 				eth_dev->data->port_id);
1464 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1465 		} else {
1466 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1467 				eth_dev->data->port_id);
1468 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1469 		}
1470 	}
1471 
1472 	if (vtpci_packed_queue(hw)) {
1473 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1474 			PMD_INIT_LOG(INFO,
1475 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1476 				eth_dev->data->port_id);
1477 			eth_dev->rx_pkt_burst =
1478 				&virtio_recv_mergeable_pkts_packed;
1479 		} else {
1480 			PMD_INIT_LOG(INFO,
1481 				"virtio: using packed ring standard Rx path on port %u",
1482 				eth_dev->data->port_id);
1483 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1484 		}
1485 	} else {
1486 		if (hw->use_simple_rx) {
1487 			PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1488 				eth_dev->data->port_id);
1489 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1490 		} else if (hw->use_inorder_rx) {
1491 			PMD_INIT_LOG(INFO,
1492 				"virtio: using inorder Rx path on port %u",
1493 				eth_dev->data->port_id);
1494 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1495 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1496 			PMD_INIT_LOG(INFO,
1497 				"virtio: using mergeable buffer Rx path on port %u",
1498 				eth_dev->data->port_id);
1499 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1500 		} else {
1501 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1502 				eth_dev->data->port_id);
1503 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1504 		}
1505 	}
1506 
1507 }
1508 
1509 /* Only support 1:1 queue/interrupt mapping so far.
1510  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1511  * interrupt vectors (<N+1).
1512  */
1513 static int
1514 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1515 {
1516 	uint32_t i;
1517 	struct virtio_hw *hw = dev->data->dev_private;
1518 
1519 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1520 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1521 		dev->intr_handle->intr_vec[i] = i + 1;
1522 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1523 						 VIRTIO_MSI_NO_VECTOR) {
1524 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1525 			return -EBUSY;
1526 		}
1527 	}
1528 
1529 	return 0;
1530 }
1531 
1532 static void
1533 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1534 {
1535 	uint32_t i;
1536 	struct virtio_hw *hw = dev->data->dev_private;
1537 
1538 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1539 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1540 		VTPCI_OPS(hw)->set_queue_irq(hw,
1541 					     hw->vqs[i * VTNET_CQ],
1542 					     VIRTIO_MSI_NO_VECTOR);
1543 }
1544 
1545 static int
1546 virtio_configure_intr(struct rte_eth_dev *dev)
1547 {
1548 	struct virtio_hw *hw = dev->data->dev_private;
1549 
1550 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1551 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1552 		return -ENOTSUP;
1553 	}
1554 
1555 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1556 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1557 		return -1;
1558 	}
1559 
1560 	if (!dev->intr_handle->intr_vec) {
1561 		dev->intr_handle->intr_vec =
1562 			rte_zmalloc("intr_vec",
1563 				    hw->max_queue_pairs * sizeof(int), 0);
1564 		if (!dev->intr_handle->intr_vec) {
1565 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1566 				     hw->max_queue_pairs);
1567 			return -ENOMEM;
1568 		}
1569 	}
1570 
1571 	/* Re-register callback to update max_intr */
1572 	rte_intr_callback_unregister(dev->intr_handle,
1573 				     virtio_interrupt_handler,
1574 				     dev);
1575 	rte_intr_callback_register(dev->intr_handle,
1576 				   virtio_interrupt_handler,
1577 				   dev);
1578 
1579 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1580 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1581 	 * And for legacy devices, this should be done before queue/vec binding
1582 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1583 	 * (22) will be ignored.
1584 	 */
1585 	if (virtio_intr_enable(dev) < 0) {
1586 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1587 		return -1;
1588 	}
1589 
1590 	if (virtio_queues_bind_intr(dev) < 0) {
1591 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1592 		return -1;
1593 	}
1594 
1595 	return 0;
1596 }
1597 
1598 /* reset device and renegotiate features if needed */
1599 static int
1600 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1601 {
1602 	struct virtio_hw *hw = eth_dev->data->dev_private;
1603 	struct virtio_net_config *config;
1604 	struct virtio_net_config local_config;
1605 	struct rte_pci_device *pci_dev = NULL;
1606 	int ret;
1607 
1608 	/* Reset the device although not necessary at startup */
1609 	vtpci_reset(hw);
1610 
1611 	if (hw->vqs) {
1612 		virtio_dev_free_mbufs(eth_dev);
1613 		virtio_free_queues(hw);
1614 	}
1615 
1616 	/* Tell the host we've noticed this device. */
1617 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1618 
1619 	/* Tell the host we've known how to drive the device. */
1620 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1621 	if (virtio_negotiate_features(hw, req_features) < 0)
1622 		return -1;
1623 
1624 	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1625 
1626 	if (!hw->virtio_user_dev) {
1627 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1628 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1629 	}
1630 
1631 	/* If host does not support both status and MSI-X then disable LSC */
1632 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1633 	    hw->use_msix != VIRTIO_MSIX_NONE)
1634 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1635 	else
1636 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1637 
1638 	/* Setting up rx_header size for the device */
1639 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1640 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1641 	    vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1642 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1643 	else
1644 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1645 
1646 	/* Copy the permanent MAC address to: virtio_hw */
1647 	virtio_get_hwaddr(hw);
1648 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1649 			&eth_dev->data->mac_addrs[0]);
1650 	PMD_INIT_LOG(DEBUG,
1651 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1652 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1653 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1654 
1655 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1656 		config = &local_config;
1657 
1658 		vtpci_read_dev_config(hw,
1659 			offsetof(struct virtio_net_config, mac),
1660 			&config->mac, sizeof(config->mac));
1661 
1662 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1663 			vtpci_read_dev_config(hw,
1664 				offsetof(struct virtio_net_config, status),
1665 				&config->status, sizeof(config->status));
1666 		} else {
1667 			PMD_INIT_LOG(DEBUG,
1668 				     "VIRTIO_NET_F_STATUS is not supported");
1669 			config->status = 0;
1670 		}
1671 
1672 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1673 			vtpci_read_dev_config(hw,
1674 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1675 				&config->max_virtqueue_pairs,
1676 				sizeof(config->max_virtqueue_pairs));
1677 		} else {
1678 			PMD_INIT_LOG(DEBUG,
1679 				     "VIRTIO_NET_F_MQ is not supported");
1680 			config->max_virtqueue_pairs = 1;
1681 		}
1682 
1683 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1684 
1685 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1686 			vtpci_read_dev_config(hw,
1687 				offsetof(struct virtio_net_config, mtu),
1688 				&config->mtu,
1689 				sizeof(config->mtu));
1690 
1691 			/*
1692 			 * MTU value has already been checked at negotiation
1693 			 * time, but check again in case it has changed since
1694 			 * then, which should not happen.
1695 			 */
1696 			if (config->mtu < ETHER_MIN_MTU) {
1697 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1698 						config->mtu);
1699 				return -1;
1700 			}
1701 
1702 			hw->max_mtu = config->mtu;
1703 			/* Set initial MTU to maximum one supported by vhost */
1704 			eth_dev->data->mtu = config->mtu;
1705 
1706 		} else {
1707 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1708 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1709 		}
1710 
1711 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1712 				config->max_virtqueue_pairs);
1713 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1714 		PMD_INIT_LOG(DEBUG,
1715 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1716 				config->mac[0], config->mac[1],
1717 				config->mac[2], config->mac[3],
1718 				config->mac[4], config->mac[5]);
1719 	} else {
1720 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1721 		hw->max_queue_pairs = 1;
1722 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1723 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1724 	}
1725 
1726 	ret = virtio_alloc_queues(eth_dev);
1727 	if (ret < 0)
1728 		return ret;
1729 
1730 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1731 		if (virtio_configure_intr(eth_dev) < 0) {
1732 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1733 			return -1;
1734 		}
1735 	}
1736 
1737 	vtpci_reinit_complete(hw);
1738 
1739 	if (pci_dev)
1740 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1741 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1742 			pci_dev->id.device_id);
1743 
1744 	return 0;
1745 }
1746 
1747 /*
1748  * Remap the PCI device again (IO port map for legacy device and
1749  * memory map for modern device), so that the secondary process
1750  * could have the PCI initiated correctly.
1751  */
1752 static int
1753 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1754 {
1755 	if (hw->modern) {
1756 		/*
1757 		 * We don't have to re-parse the PCI config space, since
1758 		 * rte_pci_map_device() makes sure the mapped address
1759 		 * in secondary process would equal to the one mapped in
1760 		 * the primary process: error will be returned if that
1761 		 * requirement is not met.
1762 		 *
1763 		 * That said, we could simply reuse all cap pointers
1764 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1765 		 * primary process, which is stored in shared memory.
1766 		 */
1767 		if (rte_pci_map_device(pci_dev)) {
1768 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1769 			return -1;
1770 		}
1771 	} else {
1772 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1773 			return -1;
1774 	}
1775 
1776 	return 0;
1777 }
1778 
1779 static void
1780 virtio_set_vtpci_ops(struct virtio_hw *hw)
1781 {
1782 #ifdef RTE_VIRTIO_USER
1783 	if (hw->virtio_user_dev)
1784 		VTPCI_OPS(hw) = &virtio_user_ops;
1785 	else
1786 #endif
1787 	if (hw->modern)
1788 		VTPCI_OPS(hw) = &modern_ops;
1789 	else
1790 		VTPCI_OPS(hw) = &legacy_ops;
1791 }
1792 
1793 /*
1794  * This function is based on probe() function in virtio_pci.c
1795  * It returns 0 on success.
1796  */
1797 int
1798 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1799 {
1800 	struct virtio_hw *hw = eth_dev->data->dev_private;
1801 	int ret;
1802 
1803 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1804 
1805 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1806 
1807 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1808 		if (!hw->virtio_user_dev) {
1809 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1810 			if (ret)
1811 				return ret;
1812 		}
1813 
1814 		virtio_set_vtpci_ops(hw);
1815 		set_rxtx_funcs(eth_dev);
1816 
1817 		return 0;
1818 	}
1819 
1820 	/* Allocate memory for storing MAC addresses */
1821 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1822 	if (eth_dev->data->mac_addrs == NULL) {
1823 		PMD_INIT_LOG(ERR,
1824 			"Failed to allocate %d bytes needed to store MAC addresses",
1825 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1826 		return -ENOMEM;
1827 	}
1828 
1829 	hw->port_id = eth_dev->data->port_id;
1830 	/* For virtio_user case the hw->virtio_user_dev is populated by
1831 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1832 	 */
1833 	if (!hw->virtio_user_dev) {
1834 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1835 		if (ret)
1836 			goto out;
1837 	}
1838 
1839 	/* reset device and negotiate default features */
1840 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1841 	if (ret < 0)
1842 		goto out;
1843 
1844 	return 0;
1845 
1846 out:
1847 	rte_free(eth_dev->data->mac_addrs);
1848 	return ret;
1849 }
1850 
1851 static int
1852 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1853 {
1854 	PMD_INIT_FUNC_TRACE();
1855 
1856 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1857 		return 0;
1858 
1859 	virtio_dev_stop(eth_dev);
1860 	virtio_dev_close(eth_dev);
1861 
1862 	eth_dev->dev_ops = NULL;
1863 	eth_dev->tx_pkt_burst = NULL;
1864 	eth_dev->rx_pkt_burst = NULL;
1865 
1866 	if (eth_dev->device)
1867 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1868 
1869 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1870 
1871 	return 0;
1872 }
1873 
1874 static int vdpa_check_handler(__rte_unused const char *key,
1875 		const char *value, __rte_unused void *opaque)
1876 {
1877 	if (strcmp(value, "1"))
1878 		return -1;
1879 
1880 	return 0;
1881 }
1882 
1883 static int
1884 vdpa_mode_selected(struct rte_devargs *devargs)
1885 {
1886 	struct rte_kvargs *kvlist;
1887 	const char *key = "vdpa";
1888 	int ret = 0;
1889 
1890 	if (devargs == NULL)
1891 		return 0;
1892 
1893 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1894 	if (kvlist == NULL)
1895 		return 0;
1896 
1897 	if (!rte_kvargs_count(kvlist, key))
1898 		goto exit;
1899 
1900 	/* vdpa mode selected when there's a key-value pair: vdpa=1 */
1901 	if (rte_kvargs_process(kvlist, key,
1902 				vdpa_check_handler, NULL) < 0) {
1903 		goto exit;
1904 	}
1905 	ret = 1;
1906 
1907 exit:
1908 	rte_kvargs_free(kvlist);
1909 	return ret;
1910 }
1911 
1912 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1913 	struct rte_pci_device *pci_dev)
1914 {
1915 	if (rte_eal_iopl_init() != 0) {
1916 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1917 		return 1;
1918 	}
1919 
1920 	/* virtio pmd skips probe if device needs to work in vdpa mode */
1921 	if (vdpa_mode_selected(pci_dev->device.devargs))
1922 		return 1;
1923 
1924 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1925 		eth_virtio_dev_init);
1926 }
1927 
1928 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1929 {
1930 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1931 }
1932 
1933 static struct rte_pci_driver rte_virtio_pmd = {
1934 	.driver = {
1935 		.name = "net_virtio",
1936 	},
1937 	.id_table = pci_id_virtio_map,
1938 	.drv_flags = 0,
1939 	.probe = eth_virtio_pci_probe,
1940 	.remove = eth_virtio_pci_remove,
1941 };
1942 
1943 RTE_INIT(rte_virtio_pmd_init)
1944 {
1945 	rte_eal_iopl_init();
1946 	rte_pci_register(&rte_virtio_pmd);
1947 }
1948 
1949 static bool
1950 rx_offload_enabled(struct virtio_hw *hw)
1951 {
1952 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
1953 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1954 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
1955 }
1956 
1957 static bool
1958 tx_offload_enabled(struct virtio_hw *hw)
1959 {
1960 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
1961 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
1962 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
1963 }
1964 
1965 /*
1966  * Configure virtio device
1967  * It returns 0 on success.
1968  */
1969 static int
1970 virtio_dev_configure(struct rte_eth_dev *dev)
1971 {
1972 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1973 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
1974 	struct virtio_hw *hw = dev->data->dev_private;
1975 	uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
1976 		hw->vtnet_hdr_size;
1977 	uint64_t rx_offloads = rxmode->offloads;
1978 	uint64_t tx_offloads = txmode->offloads;
1979 	uint64_t req_features;
1980 	int ret;
1981 
1982 	PMD_INIT_LOG(DEBUG, "configure");
1983 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1984 
1985 	if (dev->data->dev_conf.intr_conf.rxq) {
1986 		ret = virtio_init_device(dev, hw->req_guest_features);
1987 		if (ret < 0)
1988 			return ret;
1989 	}
1990 
1991 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
1992 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1993 
1994 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
1995 			   DEV_RX_OFFLOAD_TCP_CKSUM))
1996 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1997 
1998 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
1999 		req_features |=
2000 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2001 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2002 
2003 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2004 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2005 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2006 
2007 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2008 		req_features |=
2009 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2010 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2011 
2012 	/* if request features changed, reinit the device */
2013 	if (req_features != hw->req_guest_features) {
2014 		ret = virtio_init_device(dev, req_features);
2015 		if (ret < 0)
2016 			return ret;
2017 	}
2018 
2019 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2020 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2021 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2022 		PMD_DRV_LOG(ERR,
2023 			"rx checksum not available on this host");
2024 		return -ENOTSUP;
2025 	}
2026 
2027 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2028 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2029 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2030 		PMD_DRV_LOG(ERR,
2031 			"Large Receive Offload not available on this host");
2032 		return -ENOTSUP;
2033 	}
2034 
2035 	/* start control queue */
2036 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2037 		virtio_dev_cq_start(dev);
2038 
2039 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2040 		hw->vlan_strip = 1;
2041 
2042 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2043 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2044 		PMD_DRV_LOG(ERR,
2045 			    "vlan filtering not available on this host");
2046 		return -ENOTSUP;
2047 	}
2048 
2049 	hw->has_tx_offload = tx_offload_enabled(hw);
2050 	hw->has_rx_offload = rx_offload_enabled(hw);
2051 
2052 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2053 		/* Enable vector (0) for Link State Intrerrupt */
2054 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2055 				VIRTIO_MSI_NO_VECTOR) {
2056 			PMD_DRV_LOG(ERR, "failed to set config vector");
2057 			return -EBUSY;
2058 		}
2059 
2060 	rte_spinlock_init(&hw->state_lock);
2061 
2062 	hw->use_simple_rx = 1;
2063 
2064 	if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2065 		hw->use_inorder_tx = 1;
2066 		hw->use_inorder_rx = 1;
2067 		hw->use_simple_rx = 0;
2068 	}
2069 
2070 	if (vtpci_packed_queue(hw)) {
2071 		hw->use_simple_rx = 0;
2072 		hw->use_inorder_rx = 0;
2073 	}
2074 
2075 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2076 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2077 		hw->use_simple_rx = 0;
2078 	}
2079 #endif
2080 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2081 		 hw->use_simple_rx = 0;
2082 	}
2083 
2084 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2085 			   DEV_RX_OFFLOAD_TCP_CKSUM |
2086 			   DEV_RX_OFFLOAD_TCP_LRO |
2087 			   DEV_RX_OFFLOAD_VLAN_STRIP))
2088 		hw->use_simple_rx = 0;
2089 
2090 	hw->opened = true;
2091 
2092 	return 0;
2093 }
2094 
2095 
2096 static int
2097 virtio_dev_start(struct rte_eth_dev *dev)
2098 {
2099 	uint16_t nb_queues, i;
2100 	struct virtnet_rx *rxvq;
2101 	struct virtnet_tx *txvq __rte_unused;
2102 	struct virtio_hw *hw = dev->data->dev_private;
2103 	int ret;
2104 
2105 	/* Finish the initialization of the queues */
2106 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2107 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2108 		if (ret < 0)
2109 			return ret;
2110 	}
2111 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2112 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2113 		if (ret < 0)
2114 			return ret;
2115 	}
2116 
2117 	/* check if lsc interrupt feature is enabled */
2118 	if (dev->data->dev_conf.intr_conf.lsc) {
2119 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2120 			PMD_DRV_LOG(ERR, "link status not supported by host");
2121 			return -ENOTSUP;
2122 		}
2123 	}
2124 
2125 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2126 	 * in device configure, but it could be unmapped  when device is
2127 	 * stopped.
2128 	 */
2129 	if (dev->data->dev_conf.intr_conf.lsc ||
2130 	    dev->data->dev_conf.intr_conf.rxq) {
2131 		virtio_intr_disable(dev);
2132 
2133 		/* Setup interrupt callback  */
2134 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2135 			rte_intr_callback_register(dev->intr_handle,
2136 						   virtio_interrupt_handler,
2137 						   dev);
2138 
2139 		if (virtio_intr_enable(dev) < 0) {
2140 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2141 			return -EIO;
2142 		}
2143 	}
2144 
2145 	/*Notify the backend
2146 	 *Otherwise the tap backend might already stop its queue due to fullness.
2147 	 *vhost backend will have no chance to be waked up
2148 	 */
2149 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2150 	if (hw->max_queue_pairs > 1) {
2151 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2152 			return -EINVAL;
2153 	}
2154 
2155 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2156 
2157 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2158 		rxvq = dev->data->rx_queues[i];
2159 		/* Flush the old packets */
2160 		virtqueue_rxvq_flush(rxvq->vq);
2161 		virtqueue_notify(rxvq->vq);
2162 	}
2163 
2164 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2165 		txvq = dev->data->tx_queues[i];
2166 		virtqueue_notify(txvq->vq);
2167 	}
2168 
2169 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2170 
2171 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2172 		rxvq = dev->data->rx_queues[i];
2173 		VIRTQUEUE_DUMP(rxvq->vq);
2174 	}
2175 
2176 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2177 		txvq = dev->data->tx_queues[i];
2178 		VIRTQUEUE_DUMP(txvq->vq);
2179 	}
2180 
2181 	set_rxtx_funcs(dev);
2182 	hw->started = true;
2183 
2184 	/* Initialize Link state */
2185 	virtio_dev_link_update(dev, 0);
2186 
2187 	return 0;
2188 }
2189 
2190 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2191 {
2192 	struct virtio_hw *hw = dev->data->dev_private;
2193 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2194 	const char *type __rte_unused;
2195 	unsigned int i, mbuf_num = 0;
2196 	struct virtqueue *vq;
2197 	struct rte_mbuf *buf;
2198 	int queue_type;
2199 
2200 	if (hw->vqs == NULL)
2201 		return;
2202 
2203 	for (i = 0; i < nr_vq; i++) {
2204 		vq = hw->vqs[i];
2205 		if (!vq)
2206 			continue;
2207 
2208 		queue_type = virtio_get_queue_type(hw, i);
2209 		if (queue_type == VTNET_RQ)
2210 			type = "rxq";
2211 		else if (queue_type == VTNET_TQ)
2212 			type = "txq";
2213 		else
2214 			continue;
2215 
2216 		PMD_INIT_LOG(DEBUG,
2217 			"Before freeing %s[%d] used and unused buf",
2218 			type, i);
2219 		VIRTQUEUE_DUMP(vq);
2220 
2221 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2222 			rte_pktmbuf_free(buf);
2223 			mbuf_num++;
2224 		}
2225 
2226 		PMD_INIT_LOG(DEBUG,
2227 			"After freeing %s[%d] used and unused buf",
2228 			type, i);
2229 		VIRTQUEUE_DUMP(vq);
2230 	}
2231 
2232 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2233 }
2234 
2235 /*
2236  * Stop device: disable interrupt and mark link down
2237  */
2238 static void
2239 virtio_dev_stop(struct rte_eth_dev *dev)
2240 {
2241 	struct virtio_hw *hw = dev->data->dev_private;
2242 	struct rte_eth_link link;
2243 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2244 
2245 	PMD_INIT_LOG(DEBUG, "stop");
2246 
2247 	rte_spinlock_lock(&hw->state_lock);
2248 	if (!hw->started)
2249 		goto out_unlock;
2250 	hw->started = false;
2251 
2252 	if (intr_conf->lsc || intr_conf->rxq) {
2253 		virtio_intr_disable(dev);
2254 
2255 		/* Reset interrupt callback  */
2256 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2257 			rte_intr_callback_unregister(dev->intr_handle,
2258 						     virtio_interrupt_handler,
2259 						     dev);
2260 		}
2261 	}
2262 
2263 	memset(&link, 0, sizeof(link));
2264 	rte_eth_linkstatus_set(dev, &link);
2265 out_unlock:
2266 	rte_spinlock_unlock(&hw->state_lock);
2267 }
2268 
2269 static int
2270 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2271 {
2272 	struct rte_eth_link link;
2273 	uint16_t status;
2274 	struct virtio_hw *hw = dev->data->dev_private;
2275 
2276 	memset(&link, 0, sizeof(link));
2277 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
2278 	link.link_speed  = ETH_SPEED_NUM_10G;
2279 	link.link_autoneg = ETH_LINK_FIXED;
2280 
2281 	if (!hw->started) {
2282 		link.link_status = ETH_LINK_DOWN;
2283 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2284 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2285 		vtpci_read_dev_config(hw,
2286 				offsetof(struct virtio_net_config, status),
2287 				&status, sizeof(status));
2288 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2289 			link.link_status = ETH_LINK_DOWN;
2290 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2291 				     dev->data->port_id);
2292 		} else {
2293 			link.link_status = ETH_LINK_UP;
2294 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2295 				     dev->data->port_id);
2296 		}
2297 	} else {
2298 		link.link_status = ETH_LINK_UP;
2299 	}
2300 
2301 	return rte_eth_linkstatus_set(dev, &link);
2302 }
2303 
2304 static int
2305 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2306 {
2307 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2308 	struct virtio_hw *hw = dev->data->dev_private;
2309 	uint64_t offloads = rxmode->offloads;
2310 
2311 	if (mask & ETH_VLAN_FILTER_MASK) {
2312 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2313 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2314 
2315 			PMD_DRV_LOG(NOTICE,
2316 				"vlan filtering not available on this host");
2317 
2318 			return -ENOTSUP;
2319 		}
2320 	}
2321 
2322 	if (mask & ETH_VLAN_STRIP_MASK)
2323 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2324 
2325 	return 0;
2326 }
2327 
2328 static void
2329 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2330 {
2331 	uint64_t tso_mask, host_features;
2332 	struct virtio_hw *hw = dev->data->dev_private;
2333 
2334 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2335 
2336 	dev_info->max_rx_queues =
2337 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2338 	dev_info->max_tx_queues =
2339 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2340 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2341 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2342 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2343 
2344 	host_features = VTPCI_OPS(hw)->get_features(hw);
2345 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2346 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2347 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2348 		dev_info->rx_offload_capa |=
2349 			DEV_RX_OFFLOAD_TCP_CKSUM |
2350 			DEV_RX_OFFLOAD_UDP_CKSUM;
2351 	}
2352 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2353 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2354 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2355 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2356 	if ((host_features & tso_mask) == tso_mask)
2357 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2358 
2359 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2360 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2361 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2362 		dev_info->tx_offload_capa |=
2363 			DEV_TX_OFFLOAD_UDP_CKSUM |
2364 			DEV_TX_OFFLOAD_TCP_CKSUM;
2365 	}
2366 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2367 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2368 	if ((host_features & tso_mask) == tso_mask)
2369 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2370 }
2371 
2372 /*
2373  * It enables testpmd to collect per queue stats.
2374  */
2375 static int
2376 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2377 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2378 __rte_unused uint8_t is_rx)
2379 {
2380 	return 0;
2381 }
2382 
2383 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2384 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2385 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2386 
2387 RTE_INIT(virtio_init_log)
2388 {
2389 	virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2390 	if (virtio_logtype_init >= 0)
2391 		rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2392 	virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2393 	if (virtio_logtype_driver >= 0)
2394 		rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2395 }
2396