xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision bd03d3f1e4f1734c70bf6be32cdeb5e3ae6fa611)
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_enable(struct rte_eth_dev *dev);
75 static int virtio_intr_disable(struct rte_eth_dev *dev);
76 
77 static int virtio_dev_queue_stats_mapping_set(
78 	struct rte_eth_dev *eth_dev,
79 	uint16_t queue_id,
80 	uint8_t stat_idx,
81 	uint8_t is_rx);
82 
83 int virtio_logtype_init;
84 int virtio_logtype_driver;
85 
86 static void virtio_notify_peers(struct rte_eth_dev *dev);
87 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
88 
89 /*
90  * The set of PCI devices this driver supports
91  */
92 static const struct rte_pci_id pci_id_virtio_map[] = {
93 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
94 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
95 	{ .vendor_id = 0, /* sentinel */ },
96 };
97 
98 struct rte_virtio_xstats_name_off {
99 	char name[RTE_ETH_XSTATS_NAME_SIZE];
100 	unsigned offset;
101 };
102 
103 /* [rt]x_qX_ is prepended to the name string here */
104 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
105 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
106 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
107 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
108 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
109 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
110 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
111 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
112 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
113 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
114 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
115 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
116 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
117 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
118 };
119 
120 /* [rt]x_qX_ is prepended to the name string here */
121 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
122 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
123 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
124 	{"errors",                 offsetof(struct virtnet_tx, stats.errors)},
125 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
126 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
127 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
128 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
129 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
130 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
131 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
132 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
133 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
134 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
135 };
136 
137 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
138 			    sizeof(rte_virtio_rxq_stat_strings[0]))
139 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
140 			    sizeof(rte_virtio_txq_stat_strings[0]))
141 
142 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
143 
144 static struct virtio_pmd_ctrl *
145 virtio_pq_send_command(struct virtnet_ctl *cvq, 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 	int wrap_counter;
153 	uint16_t flags;
154 	int sum = 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 	wrap_counter = vq->avail_wrap_counter;
165 	desc[head].flags = VRING_DESC_F_NEXT;
166 	desc[head].addr = cvq->virtio_net_hdr_mem;
167 	desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
168 	vq->vq_free_cnt--;
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 		flags = VRING_DESC_F_NEXT;
180 		sum += dlen[k];
181 		vq->vq_free_cnt--;
182 		flags |= VRING_DESC_F_AVAIL(vq->avail_wrap_counter) |
183 			 VRING_DESC_F_USED(!vq->avail_wrap_counter);
184 		desc[vq->vq_avail_idx].flags = flags;
185 		rte_smp_wmb();
186 		vq->vq_free_cnt--;
187 		if (++vq->vq_avail_idx >= vq->vq_nentries) {
188 			vq->vq_avail_idx -= vq->vq_nentries;
189 			vq->avail_wrap_counter ^= 1;
190 		}
191 	}
192 
193 
194 	desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
195 		+ sizeof(struct virtio_net_ctrl_hdr);
196 	desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
197 	flags = VRING_DESC_F_WRITE;
198 	flags |= VRING_DESC_F_AVAIL(vq->avail_wrap_counter) |
199 		 VRING_DESC_F_USED(!vq->avail_wrap_counter);
200 	desc[vq->vq_avail_idx].flags = flags;
201 	flags = VRING_DESC_F_NEXT;
202 	flags |= VRING_DESC_F_AVAIL(wrap_counter) |
203 		 VRING_DESC_F_USED(!wrap_counter);
204 	desc[head].flags = flags;
205 	rte_smp_wmb();
206 
207 	vq->vq_free_cnt--;
208 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
209 		vq->vq_avail_idx -= vq->vq_nentries;
210 		vq->avail_wrap_counter ^= 1;
211 	}
212 
213 	virtqueue_notify(vq);
214 
215 	/* wait for used descriptors in virtqueue */
216 	do {
217 		rte_rmb();
218 		usleep(100);
219 	} while (!desc_is_used(&desc[head], vq));
220 
221 	/* now get used descriptors */
222 	while (desc_is_used(&desc[vq->vq_used_cons_idx], vq)) {
223 		vq->vq_free_cnt++;
224 		if (++vq->vq_used_cons_idx >= vq->vq_nentries) {
225 			vq->vq_used_cons_idx -= vq->vq_nentries;
226 			vq->used_wrap_counter ^= 1;
227 		}
228 	}
229 
230 	result = cvq->virtio_net_hdr_mz->addr;
231 	return result;
232 }
233 
234 static int
235 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
236 		int *dlen, int pkt_num)
237 {
238 	uint32_t head, i;
239 	int k, sum = 0;
240 	virtio_net_ctrl_ack status = ~0;
241 	struct virtio_pmd_ctrl *result;
242 	struct virtqueue *vq;
243 
244 	ctrl->status = status;
245 
246 	if (!cvq || !cvq->vq) {
247 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
248 		return -1;
249 	}
250 
251 	rte_spinlock_lock(&cvq->lock);
252 	vq = cvq->vq;
253 	head = vq->vq_desc_head_idx;
254 
255 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
256 		"vq->hw->cvq = %p vq = %p",
257 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
258 
259 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
260 		rte_spinlock_unlock(&cvq->lock);
261 		return -1;
262 	}
263 
264 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
265 		sizeof(struct virtio_pmd_ctrl));
266 
267 	if (vtpci_packed_queue(vq->hw)) {
268 		result = virtio_pq_send_command(cvq, ctrl, dlen, pkt_num);
269 		goto out_unlock;
270 	}
271 
272 	/*
273 	 * Format is enforced in qemu code:
274 	 * One TX packet for header;
275 	 * At least one TX packet per argument;
276 	 * One RX packet for ACK.
277 	 */
278 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
279 	vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
280 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
281 	vq->vq_free_cnt--;
282 	i = vq->vq_ring.desc[head].next;
283 
284 	for (k = 0; k < pkt_num; k++) {
285 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
286 		vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
287 			+ sizeof(struct virtio_net_ctrl_hdr)
288 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
289 		vq->vq_ring.desc[i].len = dlen[k];
290 		sum += dlen[k];
291 		vq->vq_free_cnt--;
292 		i = vq->vq_ring.desc[i].next;
293 	}
294 
295 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
296 	vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
297 			+ sizeof(struct virtio_net_ctrl_hdr);
298 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
299 	vq->vq_free_cnt--;
300 
301 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
302 
303 	vq_update_avail_ring(vq, head);
304 	vq_update_avail_idx(vq);
305 
306 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
307 
308 	virtqueue_notify(vq);
309 
310 	rte_rmb();
311 	while (VIRTQUEUE_NUSED(vq) == 0) {
312 		rte_rmb();
313 		usleep(100);
314 	}
315 
316 	while (VIRTQUEUE_NUSED(vq)) {
317 		uint32_t idx, desc_idx, used_idx;
318 		struct vring_used_elem *uep;
319 
320 		used_idx = (uint32_t)(vq->vq_used_cons_idx
321 				& (vq->vq_nentries - 1));
322 		uep = &vq->vq_ring.used->ring[used_idx];
323 		idx = (uint32_t) uep->id;
324 		desc_idx = idx;
325 
326 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
327 			desc_idx = vq->vq_ring.desc[desc_idx].next;
328 			vq->vq_free_cnt++;
329 		}
330 
331 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
332 		vq->vq_desc_head_idx = idx;
333 
334 		vq->vq_used_cons_idx++;
335 		vq->vq_free_cnt++;
336 	}
337 
338 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
339 			vq->vq_free_cnt, vq->vq_desc_head_idx);
340 
341 	result = cvq->virtio_net_hdr_mz->addr;
342 
343 out_unlock:
344 	rte_spinlock_unlock(&cvq->lock);
345 	return result->status;
346 }
347 
348 static int
349 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
350 {
351 	struct virtio_hw *hw = dev->data->dev_private;
352 	struct virtio_pmd_ctrl ctrl;
353 	int dlen[1];
354 	int ret;
355 
356 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
357 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
358 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
359 
360 	dlen[0] = sizeof(uint16_t);
361 
362 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
363 	if (ret) {
364 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
365 			  "failed, this is too late now...");
366 		return -EINVAL;
367 	}
368 
369 	return 0;
370 }
371 
372 static void
373 virtio_dev_queue_release(void *queue __rte_unused)
374 {
375 	/* do nothing */
376 }
377 
378 static uint16_t
379 virtio_get_nr_vq(struct virtio_hw *hw)
380 {
381 	uint16_t nr_vq = hw->max_queue_pairs * 2;
382 
383 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
384 		nr_vq += 1;
385 
386 	return nr_vq;
387 }
388 
389 static void
390 virtio_init_vring(struct virtqueue *vq)
391 {
392 	int size = vq->vq_nentries;
393 	struct vring *vr = &vq->vq_ring;
394 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
395 
396 	PMD_INIT_FUNC_TRACE();
397 
398 	memset(ring_mem, 0, vq->vq_ring_size);
399 
400 	vq->vq_used_cons_idx = 0;
401 	vq->vq_desc_head_idx = 0;
402 	vq->vq_avail_idx = 0;
403 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
404 	vq->vq_free_cnt = vq->vq_nentries;
405 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
406 	if (vtpci_packed_queue(vq->hw)) {
407 		vring_init_packed(&vq->ring_packed, ring_mem,
408 				  VIRTIO_PCI_VRING_ALIGN, size);
409 		vring_desc_init_packed(vq, size);
410 	} else {
411 		vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
412 		vring_desc_init_split(vr->desc, size);
413 	}
414 	/*
415 	 * Disable device(host) interrupting guest
416 	 */
417 	virtqueue_disable_intr(vq);
418 }
419 
420 static int
421 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
422 {
423 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
424 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
425 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
426 	unsigned int vq_size, size;
427 	struct virtio_hw *hw = dev->data->dev_private;
428 	struct virtnet_rx *rxvq = NULL;
429 	struct virtnet_tx *txvq = NULL;
430 	struct virtnet_ctl *cvq = NULL;
431 	struct virtqueue *vq;
432 	size_t sz_hdr_mz = 0;
433 	void *sw_ring = NULL;
434 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
435 	int ret;
436 	int numa_node = dev->device->numa_node;
437 
438 	PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
439 			vtpci_queue_idx, numa_node);
440 
441 	/*
442 	 * Read the virtqueue size from the Queue Size field
443 	 * Always power of 2 and if 0 virtqueue does not exist
444 	 */
445 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
446 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
447 	if (vq_size == 0) {
448 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
449 		return -EINVAL;
450 	}
451 
452 	if (!rte_is_power_of_2(vq_size)) {
453 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
454 		return -EINVAL;
455 	}
456 
457 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
458 		 dev->data->port_id, vtpci_queue_idx);
459 
460 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
461 				vq_size * sizeof(struct vq_desc_extra),
462 				RTE_CACHE_LINE_SIZE);
463 	if (queue_type == VTNET_TQ) {
464 		/*
465 		 * For each xmit packet, allocate a virtio_net_hdr
466 		 * and indirect ring elements
467 		 */
468 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
469 	} else if (queue_type == VTNET_CQ) {
470 		/* Allocate a page for control vq command, data and status */
471 		sz_hdr_mz = PAGE_SIZE;
472 	}
473 
474 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
475 				numa_node);
476 	if (vq == NULL) {
477 		PMD_INIT_LOG(ERR, "can not allocate vq");
478 		return -ENOMEM;
479 	}
480 	hw->vqs[vtpci_queue_idx] = vq;
481 
482 	vq->hw = hw;
483 	vq->vq_queue_index = vtpci_queue_idx;
484 	vq->vq_nentries = vq_size;
485 	vq->event_flags_shadow = 0;
486 	if (vtpci_packed_queue(hw)) {
487 		vq->avail_wrap_counter = 1;
488 		vq->used_wrap_counter = 1;
489 		vq->avail_used_flags =
490 			VRING_DESC_F_AVAIL(vq->avail_wrap_counter) |
491 			VRING_DESC_F_USED(!vq->avail_wrap_counter);
492 	}
493 
494 	/*
495 	 * Reserve a memzone for vring elements
496 	 */
497 	size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
498 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
499 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
500 		     size, vq->vq_ring_size);
501 
502 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
503 			numa_node, RTE_MEMZONE_IOVA_CONTIG,
504 			VIRTIO_PCI_VRING_ALIGN);
505 	if (mz == NULL) {
506 		if (rte_errno == EEXIST)
507 			mz = rte_memzone_lookup(vq_name);
508 		if (mz == NULL) {
509 			ret = -ENOMEM;
510 			goto fail_q_alloc;
511 		}
512 	}
513 
514 	memset(mz->addr, 0, mz->len);
515 
516 	vq->vq_ring_mem = mz->iova;
517 	vq->vq_ring_virt_mem = mz->addr;
518 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
519 		     (uint64_t)mz->iova);
520 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
521 		     (uint64_t)(uintptr_t)mz->addr);
522 
523 	virtio_init_vring(vq);
524 
525 	if (sz_hdr_mz) {
526 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
527 			 dev->data->port_id, vtpci_queue_idx);
528 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
529 				numa_node, RTE_MEMZONE_IOVA_CONTIG,
530 				RTE_CACHE_LINE_SIZE);
531 		if (hdr_mz == NULL) {
532 			if (rte_errno == EEXIST)
533 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
534 			if (hdr_mz == NULL) {
535 				ret = -ENOMEM;
536 				goto fail_q_alloc;
537 			}
538 		}
539 	}
540 
541 	if (queue_type == VTNET_RQ) {
542 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
543 			       sizeof(vq->sw_ring[0]);
544 
545 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
546 				RTE_CACHE_LINE_SIZE, numa_node);
547 		if (!sw_ring) {
548 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
549 			ret = -ENOMEM;
550 			goto fail_q_alloc;
551 		}
552 
553 		vq->sw_ring = sw_ring;
554 		rxvq = &vq->rxq;
555 		rxvq->vq = vq;
556 		rxvq->port_id = dev->data->port_id;
557 		rxvq->mz = mz;
558 	} else if (queue_type == VTNET_TQ) {
559 		txvq = &vq->txq;
560 		txvq->vq = vq;
561 		txvq->port_id = dev->data->port_id;
562 		txvq->mz = mz;
563 		txvq->virtio_net_hdr_mz = hdr_mz;
564 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
565 	} else if (queue_type == VTNET_CQ) {
566 		cvq = &vq->cq;
567 		cvq->vq = vq;
568 		cvq->mz = mz;
569 		cvq->virtio_net_hdr_mz = hdr_mz;
570 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
571 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
572 
573 		hw->cvq = cvq;
574 	}
575 
576 	/* For virtio_user case (that is when hw->dev is NULL), we use
577 	 * virtual address. And we need properly set _offset_, please see
578 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
579 	 */
580 	if (!hw->virtio_user_dev)
581 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
582 	else {
583 		vq->vq_ring_mem = (uintptr_t)mz->addr;
584 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
585 		if (queue_type == VTNET_TQ)
586 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
587 		else if (queue_type == VTNET_CQ)
588 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
589 	}
590 
591 	if (queue_type == VTNET_TQ) {
592 		struct virtio_tx_region *txr;
593 		unsigned int i;
594 
595 		txr = hdr_mz->addr;
596 		memset(txr, 0, vq_size * sizeof(*txr));
597 		for (i = 0; i < vq_size; i++) {
598 			struct vring_desc *start_dp = txr[i].tx_indir;
599 			struct vring_packed_desc *start_dp_packed =
600 				txr[i].tx_indir_pq;
601 
602 			/* first indirect descriptor is always the tx header */
603 			if (vtpci_packed_queue(hw)) {
604 				start_dp_packed->addr = txvq->virtio_net_hdr_mem
605 					+ i * sizeof(*txr)
606 					+ offsetof(struct virtio_tx_region,
607 						   tx_hdr);
608 				start_dp_packed->len = hw->vtnet_hdr_size;
609 			} else {
610 				vring_desc_init_split(start_dp,
611 						      RTE_DIM(txr[i].tx_indir));
612 				start_dp->addr = txvq->virtio_net_hdr_mem
613 					+ i * sizeof(*txr)
614 					+ offsetof(struct virtio_tx_region,
615 						   tx_hdr);
616 				start_dp->len = hw->vtnet_hdr_size;
617 				start_dp->flags = VRING_DESC_F_NEXT;
618 			}
619 		}
620 	}
621 
622 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
623 		PMD_INIT_LOG(ERR, "setup_queue failed");
624 		return -EINVAL;
625 	}
626 
627 	return 0;
628 
629 fail_q_alloc:
630 	rte_free(sw_ring);
631 	rte_memzone_free(hdr_mz);
632 	rte_memzone_free(mz);
633 	rte_free(vq);
634 
635 	return ret;
636 }
637 
638 static void
639 virtio_free_queues(struct virtio_hw *hw)
640 {
641 	uint16_t nr_vq = virtio_get_nr_vq(hw);
642 	struct virtqueue *vq;
643 	int queue_type;
644 	uint16_t i;
645 
646 	if (hw->vqs == NULL)
647 		return;
648 
649 	for (i = 0; i < nr_vq; i++) {
650 		vq = hw->vqs[i];
651 		if (!vq)
652 			continue;
653 
654 		queue_type = virtio_get_queue_type(hw, i);
655 		if (queue_type == VTNET_RQ) {
656 			rte_free(vq->sw_ring);
657 			rte_memzone_free(vq->rxq.mz);
658 		} else if (queue_type == VTNET_TQ) {
659 			rte_memzone_free(vq->txq.mz);
660 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
661 		} else {
662 			rte_memzone_free(vq->cq.mz);
663 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
664 		}
665 
666 		rte_free(vq);
667 		hw->vqs[i] = NULL;
668 	}
669 
670 	rte_free(hw->vqs);
671 	hw->vqs = NULL;
672 }
673 
674 static int
675 virtio_alloc_queues(struct rte_eth_dev *dev)
676 {
677 	struct virtio_hw *hw = dev->data->dev_private;
678 	uint16_t nr_vq = virtio_get_nr_vq(hw);
679 	uint16_t i;
680 	int ret;
681 
682 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
683 	if (!hw->vqs) {
684 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
685 		return -ENOMEM;
686 	}
687 
688 	for (i = 0; i < nr_vq; i++) {
689 		ret = virtio_init_queue(dev, i);
690 		if (ret < 0) {
691 			virtio_free_queues(hw);
692 			return ret;
693 		}
694 	}
695 
696 	return 0;
697 }
698 
699 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
700 
701 static void
702 virtio_dev_close(struct rte_eth_dev *dev)
703 {
704 	struct virtio_hw *hw = dev->data->dev_private;
705 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
706 
707 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
708 
709 	if (!hw->opened)
710 		return;
711 	hw->opened = false;
712 
713 	/* reset the NIC */
714 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
715 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
716 	if (intr_conf->rxq)
717 		virtio_queues_unbind_intr(dev);
718 
719 	if (intr_conf->lsc || intr_conf->rxq) {
720 		virtio_intr_disable(dev);
721 		rte_intr_efd_disable(dev->intr_handle);
722 		rte_free(dev->intr_handle->intr_vec);
723 		dev->intr_handle->intr_vec = NULL;
724 	}
725 
726 	vtpci_reset(hw);
727 	virtio_dev_free_mbufs(dev);
728 	virtio_free_queues(hw);
729 }
730 
731 static void
732 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
733 {
734 	struct virtio_hw *hw = dev->data->dev_private;
735 	struct virtio_pmd_ctrl ctrl;
736 	int dlen[1];
737 	int ret;
738 
739 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
740 		PMD_INIT_LOG(INFO, "host does not support rx control");
741 		return;
742 	}
743 
744 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
745 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
746 	ctrl.data[0] = 1;
747 	dlen[0] = 1;
748 
749 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
750 	if (ret)
751 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
752 }
753 
754 static void
755 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
756 {
757 	struct virtio_hw *hw = dev->data->dev_private;
758 	struct virtio_pmd_ctrl ctrl;
759 	int dlen[1];
760 	int ret;
761 
762 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
763 		PMD_INIT_LOG(INFO, "host does not support rx control");
764 		return;
765 	}
766 
767 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
768 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
769 	ctrl.data[0] = 0;
770 	dlen[0] = 1;
771 
772 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
773 	if (ret)
774 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
775 }
776 
777 static void
778 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
779 {
780 	struct virtio_hw *hw = dev->data->dev_private;
781 	struct virtio_pmd_ctrl ctrl;
782 	int dlen[1];
783 	int ret;
784 
785 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
786 		PMD_INIT_LOG(INFO, "host does not support rx control");
787 		return;
788 	}
789 
790 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
791 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
792 	ctrl.data[0] = 1;
793 	dlen[0] = 1;
794 
795 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
796 	if (ret)
797 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
798 }
799 
800 static void
801 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
802 {
803 	struct virtio_hw *hw = dev->data->dev_private;
804 	struct virtio_pmd_ctrl ctrl;
805 	int dlen[1];
806 	int ret;
807 
808 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
809 		PMD_INIT_LOG(INFO, "host does not support rx control");
810 		return;
811 	}
812 
813 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
814 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
815 	ctrl.data[0] = 0;
816 	dlen[0] = 1;
817 
818 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
819 	if (ret)
820 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
821 }
822 
823 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
824 static int
825 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
826 {
827 	struct virtio_hw *hw = dev->data->dev_private;
828 	uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
829 				 hw->vtnet_hdr_size;
830 	uint32_t frame_size = mtu + ether_hdr_len;
831 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
832 
833 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
834 
835 	if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
836 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
837 			ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
838 		return -EINVAL;
839 	}
840 	return 0;
841 }
842 
843 static int
844 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
845 {
846 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
847 	struct virtqueue *vq = rxvq->vq;
848 
849 	virtqueue_enable_intr(vq);
850 	return 0;
851 }
852 
853 static int
854 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
855 {
856 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
857 	struct virtqueue *vq = rxvq->vq;
858 
859 	virtqueue_disable_intr(vq);
860 	return 0;
861 }
862 
863 /*
864  * dev_ops for virtio, bare necessities for basic operation
865  */
866 static const struct eth_dev_ops virtio_eth_dev_ops = {
867 	.dev_configure           = virtio_dev_configure,
868 	.dev_start               = virtio_dev_start,
869 	.dev_stop                = virtio_dev_stop,
870 	.dev_close               = virtio_dev_close,
871 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
872 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
873 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
874 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
875 	.mtu_set                 = virtio_mtu_set,
876 	.dev_infos_get           = virtio_dev_info_get,
877 	.stats_get               = virtio_dev_stats_get,
878 	.xstats_get              = virtio_dev_xstats_get,
879 	.xstats_get_names        = virtio_dev_xstats_get_names,
880 	.stats_reset             = virtio_dev_stats_reset,
881 	.xstats_reset            = virtio_dev_stats_reset,
882 	.link_update             = virtio_dev_link_update,
883 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
884 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
885 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
886 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
887 	.rx_queue_release        = virtio_dev_queue_release,
888 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
889 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
890 	.tx_queue_release        = virtio_dev_queue_release,
891 	/* collect stats per queue */
892 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
893 	.vlan_filter_set         = virtio_vlan_filter_set,
894 	.mac_addr_add            = virtio_mac_addr_add,
895 	.mac_addr_remove         = virtio_mac_addr_remove,
896 	.mac_addr_set            = virtio_mac_addr_set,
897 };
898 
899 static void
900 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
901 {
902 	unsigned i;
903 
904 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
905 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
906 		if (txvq == NULL)
907 			continue;
908 
909 		stats->opackets += txvq->stats.packets;
910 		stats->obytes += txvq->stats.bytes;
911 		stats->oerrors += txvq->stats.errors;
912 
913 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
914 			stats->q_opackets[i] = txvq->stats.packets;
915 			stats->q_obytes[i] = txvq->stats.bytes;
916 		}
917 	}
918 
919 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
920 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
921 		if (rxvq == NULL)
922 			continue;
923 
924 		stats->ipackets += rxvq->stats.packets;
925 		stats->ibytes += rxvq->stats.bytes;
926 		stats->ierrors += rxvq->stats.errors;
927 
928 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
929 			stats->q_ipackets[i] = rxvq->stats.packets;
930 			stats->q_ibytes[i] = rxvq->stats.bytes;
931 		}
932 	}
933 
934 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
935 }
936 
937 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
938 				       struct rte_eth_xstat_name *xstats_names,
939 				       __rte_unused unsigned limit)
940 {
941 	unsigned i;
942 	unsigned count = 0;
943 	unsigned t;
944 
945 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
946 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
947 
948 	if (xstats_names != NULL) {
949 		/* Note: limit checked in rte_eth_xstats_names() */
950 
951 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
952 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
953 			if (rxvq == NULL)
954 				continue;
955 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
956 				snprintf(xstats_names[count].name,
957 					sizeof(xstats_names[count].name),
958 					"rx_q%u_%s", i,
959 					rte_virtio_rxq_stat_strings[t].name);
960 				count++;
961 			}
962 		}
963 
964 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
965 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
966 			if (txvq == NULL)
967 				continue;
968 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
969 				snprintf(xstats_names[count].name,
970 					sizeof(xstats_names[count].name),
971 					"tx_q%u_%s", i,
972 					rte_virtio_txq_stat_strings[t].name);
973 				count++;
974 			}
975 		}
976 		return count;
977 	}
978 	return nstats;
979 }
980 
981 static int
982 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
983 		      unsigned n)
984 {
985 	unsigned i;
986 	unsigned count = 0;
987 
988 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
989 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
990 
991 	if (n < nstats)
992 		return nstats;
993 
994 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
995 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
996 
997 		if (rxvq == NULL)
998 			continue;
999 
1000 		unsigned t;
1001 
1002 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1003 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1004 				rte_virtio_rxq_stat_strings[t].offset);
1005 			xstats[count].id = count;
1006 			count++;
1007 		}
1008 	}
1009 
1010 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1011 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1012 
1013 		if (txvq == NULL)
1014 			continue;
1015 
1016 		unsigned t;
1017 
1018 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1019 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1020 				rte_virtio_txq_stat_strings[t].offset);
1021 			xstats[count].id = count;
1022 			count++;
1023 		}
1024 	}
1025 
1026 	return count;
1027 }
1028 
1029 static int
1030 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1031 {
1032 	virtio_update_stats(dev, stats);
1033 
1034 	return 0;
1035 }
1036 
1037 static void
1038 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1039 {
1040 	unsigned int i;
1041 
1042 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1043 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1044 		if (txvq == NULL)
1045 			continue;
1046 
1047 		txvq->stats.packets = 0;
1048 		txvq->stats.bytes = 0;
1049 		txvq->stats.errors = 0;
1050 		txvq->stats.multicast = 0;
1051 		txvq->stats.broadcast = 0;
1052 		memset(txvq->stats.size_bins, 0,
1053 		       sizeof(txvq->stats.size_bins[0]) * 8);
1054 	}
1055 
1056 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1057 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1058 		if (rxvq == NULL)
1059 			continue;
1060 
1061 		rxvq->stats.packets = 0;
1062 		rxvq->stats.bytes = 0;
1063 		rxvq->stats.errors = 0;
1064 		rxvq->stats.multicast = 0;
1065 		rxvq->stats.broadcast = 0;
1066 		memset(rxvq->stats.size_bins, 0,
1067 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1068 	}
1069 }
1070 
1071 static void
1072 virtio_set_hwaddr(struct virtio_hw *hw)
1073 {
1074 	vtpci_write_dev_config(hw,
1075 			offsetof(struct virtio_net_config, mac),
1076 			&hw->mac_addr, ETHER_ADDR_LEN);
1077 }
1078 
1079 static void
1080 virtio_get_hwaddr(struct virtio_hw *hw)
1081 {
1082 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1083 		vtpci_read_dev_config(hw,
1084 			offsetof(struct virtio_net_config, mac),
1085 			&hw->mac_addr, ETHER_ADDR_LEN);
1086 	} else {
1087 		eth_random_addr(&hw->mac_addr[0]);
1088 		virtio_set_hwaddr(hw);
1089 	}
1090 }
1091 
1092 static int
1093 virtio_mac_table_set(struct virtio_hw *hw,
1094 		     const struct virtio_net_ctrl_mac *uc,
1095 		     const struct virtio_net_ctrl_mac *mc)
1096 {
1097 	struct virtio_pmd_ctrl ctrl;
1098 	int err, len[2];
1099 
1100 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1101 		PMD_DRV_LOG(INFO, "host does not support mac table");
1102 		return -1;
1103 	}
1104 
1105 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1106 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1107 
1108 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1109 	memcpy(ctrl.data, uc, len[0]);
1110 
1111 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1112 	memcpy(ctrl.data + len[0], mc, len[1]);
1113 
1114 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1115 	if (err != 0)
1116 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1117 	return err;
1118 }
1119 
1120 static int
1121 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1122 		    uint32_t index, uint32_t vmdq __rte_unused)
1123 {
1124 	struct virtio_hw *hw = dev->data->dev_private;
1125 	const struct ether_addr *addrs = dev->data->mac_addrs;
1126 	unsigned int i;
1127 	struct virtio_net_ctrl_mac *uc, *mc;
1128 
1129 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1130 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1131 		return -EINVAL;
1132 	}
1133 
1134 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1135 	uc->entries = 0;
1136 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1137 	mc->entries = 0;
1138 
1139 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1140 		const struct ether_addr *addr
1141 			= (i == index) ? mac_addr : addrs + i;
1142 		struct virtio_net_ctrl_mac *tbl
1143 			= is_multicast_ether_addr(addr) ? mc : uc;
1144 
1145 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1146 	}
1147 
1148 	return virtio_mac_table_set(hw, uc, mc);
1149 }
1150 
1151 static void
1152 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1153 {
1154 	struct virtio_hw *hw = dev->data->dev_private;
1155 	struct ether_addr *addrs = dev->data->mac_addrs;
1156 	struct virtio_net_ctrl_mac *uc, *mc;
1157 	unsigned int i;
1158 
1159 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1160 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1161 		return;
1162 	}
1163 
1164 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1165 	uc->entries = 0;
1166 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1167 	mc->entries = 0;
1168 
1169 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1170 		struct virtio_net_ctrl_mac *tbl;
1171 
1172 		if (i == index || is_zero_ether_addr(addrs + i))
1173 			continue;
1174 
1175 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1176 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1177 	}
1178 
1179 	virtio_mac_table_set(hw, uc, mc);
1180 }
1181 
1182 static int
1183 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1184 {
1185 	struct virtio_hw *hw = dev->data->dev_private;
1186 
1187 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1188 
1189 	/* Use atomic update if available */
1190 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1191 		struct virtio_pmd_ctrl ctrl;
1192 		int len = ETHER_ADDR_LEN;
1193 
1194 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1195 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1196 
1197 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1198 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1199 	}
1200 
1201 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1202 		return -ENOTSUP;
1203 
1204 	virtio_set_hwaddr(hw);
1205 	return 0;
1206 }
1207 
1208 static int
1209 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1210 {
1211 	struct virtio_hw *hw = dev->data->dev_private;
1212 	struct virtio_pmd_ctrl ctrl;
1213 	int len;
1214 
1215 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1216 		return -ENOTSUP;
1217 
1218 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1219 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1220 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1221 	len = sizeof(vlan_id);
1222 
1223 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1224 }
1225 
1226 static int
1227 virtio_intr_enable(struct rte_eth_dev *dev)
1228 {
1229 	struct virtio_hw *hw = dev->data->dev_private;
1230 
1231 	if (rte_intr_enable(dev->intr_handle) < 0)
1232 		return -1;
1233 
1234 	if (!hw->virtio_user_dev)
1235 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1236 
1237 	return 0;
1238 }
1239 
1240 static int
1241 virtio_intr_disable(struct rte_eth_dev *dev)
1242 {
1243 	struct virtio_hw *hw = dev->data->dev_private;
1244 
1245 	if (rte_intr_disable(dev->intr_handle) < 0)
1246 		return -1;
1247 
1248 	if (!hw->virtio_user_dev)
1249 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1250 
1251 	return 0;
1252 }
1253 
1254 static int
1255 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1256 {
1257 	uint64_t host_features;
1258 
1259 	/* Prepare guest_features: feature that driver wants to support */
1260 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1261 		req_features);
1262 
1263 	/* Read device(host) feature bits */
1264 	host_features = VTPCI_OPS(hw)->get_features(hw);
1265 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1266 		host_features);
1267 
1268 	/* If supported, ensure MTU value is valid before acknowledging it. */
1269 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1270 		struct virtio_net_config config;
1271 
1272 		vtpci_read_dev_config(hw,
1273 			offsetof(struct virtio_net_config, mtu),
1274 			&config.mtu, sizeof(config.mtu));
1275 
1276 		if (config.mtu < ETHER_MIN_MTU)
1277 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1278 	}
1279 
1280 	/*
1281 	 * Negotiate features: Subset of device feature bits are written back
1282 	 * guest feature bits.
1283 	 */
1284 	hw->guest_features = req_features;
1285 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1286 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1287 		hw->guest_features);
1288 
1289 	if (hw->modern) {
1290 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1291 			PMD_INIT_LOG(ERR,
1292 				"VIRTIO_F_VERSION_1 features is not enabled.");
1293 			return -1;
1294 		}
1295 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1296 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1297 			PMD_INIT_LOG(ERR,
1298 				"failed to set FEATURES_OK status!");
1299 			return -1;
1300 		}
1301 	}
1302 
1303 	hw->req_guest_features = req_features;
1304 
1305 	return 0;
1306 }
1307 
1308 int
1309 virtio_dev_pause(struct rte_eth_dev *dev)
1310 {
1311 	struct virtio_hw *hw = dev->data->dev_private;
1312 
1313 	rte_spinlock_lock(&hw->state_lock);
1314 
1315 	if (hw->started == 0) {
1316 		/* Device is just stopped. */
1317 		rte_spinlock_unlock(&hw->state_lock);
1318 		return -1;
1319 	}
1320 	hw->started = 0;
1321 	/*
1322 	 * Prevent the worker threads from touching queues to avoid contention,
1323 	 * 1 ms should be enough for the ongoing Tx function to finish.
1324 	 */
1325 	rte_delay_ms(1);
1326 	return 0;
1327 }
1328 
1329 /*
1330  * Recover hw state to let the worker threads continue.
1331  */
1332 void
1333 virtio_dev_resume(struct rte_eth_dev *dev)
1334 {
1335 	struct virtio_hw *hw = dev->data->dev_private;
1336 
1337 	hw->started = 1;
1338 	rte_spinlock_unlock(&hw->state_lock);
1339 }
1340 
1341 /*
1342  * Should be called only after device is paused.
1343  */
1344 int
1345 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1346 		int nb_pkts)
1347 {
1348 	struct virtio_hw *hw = dev->data->dev_private;
1349 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1350 	int ret;
1351 
1352 	hw->inject_pkts = tx_pkts;
1353 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1354 	hw->inject_pkts = NULL;
1355 
1356 	return ret;
1357 }
1358 
1359 static void
1360 virtio_notify_peers(struct rte_eth_dev *dev)
1361 {
1362 	struct virtio_hw *hw = dev->data->dev_private;
1363 	struct virtnet_rx *rxvq;
1364 	struct rte_mbuf *rarp_mbuf;
1365 
1366 	if (!dev->data->rx_queues)
1367 		return;
1368 
1369 	rxvq = dev->data->rx_queues[0];
1370 	if (!rxvq)
1371 		return;
1372 
1373 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1374 			(struct ether_addr *)hw->mac_addr);
1375 	if (rarp_mbuf == NULL) {
1376 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1377 		return;
1378 	}
1379 
1380 	/* If virtio port just stopped, no need to send RARP */
1381 	if (virtio_dev_pause(dev) < 0) {
1382 		rte_pktmbuf_free(rarp_mbuf);
1383 		return;
1384 	}
1385 
1386 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1387 	virtio_dev_resume(dev);
1388 }
1389 
1390 static void
1391 virtio_ack_link_announce(struct rte_eth_dev *dev)
1392 {
1393 	struct virtio_hw *hw = dev->data->dev_private;
1394 	struct virtio_pmd_ctrl ctrl;
1395 
1396 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1397 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1398 
1399 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1400 }
1401 
1402 /*
1403  * Process virtio config changed interrupt. Call the callback
1404  * if link state changed, generate gratuitous RARP packet if
1405  * the status indicates an ANNOUNCE.
1406  */
1407 void
1408 virtio_interrupt_handler(void *param)
1409 {
1410 	struct rte_eth_dev *dev = param;
1411 	struct virtio_hw *hw = dev->data->dev_private;
1412 	uint8_t isr;
1413 	uint16_t status;
1414 
1415 	/* Read interrupt status which clears interrupt */
1416 	isr = vtpci_isr(hw);
1417 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1418 
1419 	if (virtio_intr_enable(dev) < 0)
1420 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1421 
1422 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1423 		if (virtio_dev_link_update(dev, 0) == 0)
1424 			_rte_eth_dev_callback_process(dev,
1425 						      RTE_ETH_EVENT_INTR_LSC,
1426 						      NULL);
1427 
1428 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1429 			vtpci_read_dev_config(hw,
1430 				offsetof(struct virtio_net_config, status),
1431 				&status, sizeof(status));
1432 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1433 				virtio_notify_peers(dev);
1434 				if (hw->cvq)
1435 					virtio_ack_link_announce(dev);
1436 			}
1437 		}
1438 	}
1439 }
1440 
1441 /* set rx and tx handlers according to what is supported */
1442 static void
1443 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1444 {
1445 	struct virtio_hw *hw = eth_dev->data->dev_private;
1446 
1447 	if (vtpci_packed_queue(hw)) {
1448 		PMD_INIT_LOG(INFO,
1449 			"virtio: using packed ring standard Tx path on port %u",
1450 			eth_dev->data->port_id);
1451 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1452 	} else {
1453 		if (hw->use_inorder_tx) {
1454 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1455 				eth_dev->data->port_id);
1456 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1457 		} else {
1458 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1459 				eth_dev->data->port_id);
1460 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1461 		}
1462 	}
1463 
1464 	if (vtpci_packed_queue(hw)) {
1465 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1466 			PMD_INIT_LOG(INFO,
1467 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1468 				eth_dev->data->port_id);
1469 			eth_dev->rx_pkt_burst =
1470 				&virtio_recv_mergeable_pkts_packed;
1471 		} else {
1472 			PMD_INIT_LOG(INFO,
1473 				"virtio: using packed ring standard Rx path on port %u",
1474 				eth_dev->data->port_id);
1475 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1476 		}
1477 	} else {
1478 		if (hw->use_simple_rx) {
1479 			PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1480 				eth_dev->data->port_id);
1481 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1482 		} else if (hw->use_inorder_rx) {
1483 			PMD_INIT_LOG(INFO,
1484 				"virtio: using inorder Rx path on port %u",
1485 				eth_dev->data->port_id);
1486 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1487 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1488 			PMD_INIT_LOG(INFO,
1489 				"virtio: using mergeable buffer Rx path on port %u",
1490 				eth_dev->data->port_id);
1491 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1492 		} else {
1493 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1494 				eth_dev->data->port_id);
1495 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1496 		}
1497 	}
1498 
1499 }
1500 
1501 /* Only support 1:1 queue/interrupt mapping so far.
1502  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1503  * interrupt vectors (<N+1).
1504  */
1505 static int
1506 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1507 {
1508 	uint32_t i;
1509 	struct virtio_hw *hw = dev->data->dev_private;
1510 
1511 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1512 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1513 		dev->intr_handle->intr_vec[i] = i + 1;
1514 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1515 						 VIRTIO_MSI_NO_VECTOR) {
1516 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1517 			return -EBUSY;
1518 		}
1519 	}
1520 
1521 	return 0;
1522 }
1523 
1524 static void
1525 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1526 {
1527 	uint32_t i;
1528 	struct virtio_hw *hw = dev->data->dev_private;
1529 
1530 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1531 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1532 		VTPCI_OPS(hw)->set_queue_irq(hw,
1533 					     hw->vqs[i * VTNET_CQ],
1534 					     VIRTIO_MSI_NO_VECTOR);
1535 }
1536 
1537 static int
1538 virtio_configure_intr(struct rte_eth_dev *dev)
1539 {
1540 	struct virtio_hw *hw = dev->data->dev_private;
1541 
1542 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1543 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1544 		return -ENOTSUP;
1545 	}
1546 
1547 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1548 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1549 		return -1;
1550 	}
1551 
1552 	if (!dev->intr_handle->intr_vec) {
1553 		dev->intr_handle->intr_vec =
1554 			rte_zmalloc("intr_vec",
1555 				    hw->max_queue_pairs * sizeof(int), 0);
1556 		if (!dev->intr_handle->intr_vec) {
1557 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1558 				     hw->max_queue_pairs);
1559 			return -ENOMEM;
1560 		}
1561 	}
1562 
1563 	/* Re-register callback to update max_intr */
1564 	rte_intr_callback_unregister(dev->intr_handle,
1565 				     virtio_interrupt_handler,
1566 				     dev);
1567 	rte_intr_callback_register(dev->intr_handle,
1568 				   virtio_interrupt_handler,
1569 				   dev);
1570 
1571 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1572 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1573 	 * And for legacy devices, this should be done before queue/vec binding
1574 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1575 	 * (22) will be ignored.
1576 	 */
1577 	if (virtio_intr_enable(dev) < 0) {
1578 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1579 		return -1;
1580 	}
1581 
1582 	if (virtio_queues_bind_intr(dev) < 0) {
1583 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1584 		return -1;
1585 	}
1586 
1587 	return 0;
1588 }
1589 
1590 /* reset device and renegotiate features if needed */
1591 static int
1592 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1593 {
1594 	struct virtio_hw *hw = eth_dev->data->dev_private;
1595 	struct virtio_net_config *config;
1596 	struct virtio_net_config local_config;
1597 	struct rte_pci_device *pci_dev = NULL;
1598 	int ret;
1599 
1600 	/* Reset the device although not necessary at startup */
1601 	vtpci_reset(hw);
1602 
1603 	if (hw->vqs) {
1604 		virtio_dev_free_mbufs(eth_dev);
1605 		virtio_free_queues(hw);
1606 	}
1607 
1608 	/* Tell the host we've noticed this device. */
1609 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1610 
1611 	/* Tell the host we've known how to drive the device. */
1612 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1613 	if (virtio_negotiate_features(hw, req_features) < 0)
1614 		return -1;
1615 
1616 	if (!hw->virtio_user_dev) {
1617 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1618 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1619 	}
1620 
1621 	/* If host does not support both status and MSI-X then disable LSC */
1622 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1623 	    hw->use_msix != VIRTIO_MSIX_NONE)
1624 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1625 	else
1626 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1627 
1628 	/* Setting up rx_header size for the device */
1629 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1630 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1631 	    vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1632 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1633 	else
1634 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1635 
1636 	/* Copy the permanent MAC address to: virtio_hw */
1637 	virtio_get_hwaddr(hw);
1638 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1639 			&eth_dev->data->mac_addrs[0]);
1640 	PMD_INIT_LOG(DEBUG,
1641 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1642 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1643 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1644 
1645 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1646 		config = &local_config;
1647 
1648 		vtpci_read_dev_config(hw,
1649 			offsetof(struct virtio_net_config, mac),
1650 			&config->mac, sizeof(config->mac));
1651 
1652 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1653 			vtpci_read_dev_config(hw,
1654 				offsetof(struct virtio_net_config, status),
1655 				&config->status, sizeof(config->status));
1656 		} else {
1657 			PMD_INIT_LOG(DEBUG,
1658 				     "VIRTIO_NET_F_STATUS is not supported");
1659 			config->status = 0;
1660 		}
1661 
1662 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1663 			vtpci_read_dev_config(hw,
1664 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1665 				&config->max_virtqueue_pairs,
1666 				sizeof(config->max_virtqueue_pairs));
1667 		} else {
1668 			PMD_INIT_LOG(DEBUG,
1669 				     "VIRTIO_NET_F_MQ is not supported");
1670 			config->max_virtqueue_pairs = 1;
1671 		}
1672 
1673 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1674 
1675 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1676 			vtpci_read_dev_config(hw,
1677 				offsetof(struct virtio_net_config, mtu),
1678 				&config->mtu,
1679 				sizeof(config->mtu));
1680 
1681 			/*
1682 			 * MTU value has already been checked at negotiation
1683 			 * time, but check again in case it has changed since
1684 			 * then, which should not happen.
1685 			 */
1686 			if (config->mtu < ETHER_MIN_MTU) {
1687 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1688 						config->mtu);
1689 				return -1;
1690 			}
1691 
1692 			hw->max_mtu = config->mtu;
1693 			/* Set initial MTU to maximum one supported by vhost */
1694 			eth_dev->data->mtu = config->mtu;
1695 
1696 		} else {
1697 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1698 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1699 		}
1700 
1701 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1702 				config->max_virtqueue_pairs);
1703 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1704 		PMD_INIT_LOG(DEBUG,
1705 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1706 				config->mac[0], config->mac[1],
1707 				config->mac[2], config->mac[3],
1708 				config->mac[4], config->mac[5]);
1709 	} else {
1710 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1711 		hw->max_queue_pairs = 1;
1712 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1713 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1714 	}
1715 
1716 	ret = virtio_alloc_queues(eth_dev);
1717 	if (ret < 0)
1718 		return ret;
1719 
1720 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1721 		if (virtio_configure_intr(eth_dev) < 0) {
1722 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1723 			return -1;
1724 		}
1725 	}
1726 
1727 	vtpci_reinit_complete(hw);
1728 
1729 	if (pci_dev)
1730 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1731 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1732 			pci_dev->id.device_id);
1733 
1734 	return 0;
1735 }
1736 
1737 /*
1738  * Remap the PCI device again (IO port map for legacy device and
1739  * memory map for modern device), so that the secondary process
1740  * could have the PCI initiated correctly.
1741  */
1742 static int
1743 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1744 {
1745 	if (hw->modern) {
1746 		/*
1747 		 * We don't have to re-parse the PCI config space, since
1748 		 * rte_pci_map_device() makes sure the mapped address
1749 		 * in secondary process would equal to the one mapped in
1750 		 * the primary process: error will be returned if that
1751 		 * requirement is not met.
1752 		 *
1753 		 * That said, we could simply reuse all cap pointers
1754 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1755 		 * primary process, which is stored in shared memory.
1756 		 */
1757 		if (rte_pci_map_device(pci_dev)) {
1758 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1759 			return -1;
1760 		}
1761 	} else {
1762 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1763 			return -1;
1764 	}
1765 
1766 	return 0;
1767 }
1768 
1769 static void
1770 virtio_set_vtpci_ops(struct virtio_hw *hw)
1771 {
1772 #ifdef RTE_VIRTIO_USER
1773 	if (hw->virtio_user_dev)
1774 		VTPCI_OPS(hw) = &virtio_user_ops;
1775 	else
1776 #endif
1777 	if (hw->modern)
1778 		VTPCI_OPS(hw) = &modern_ops;
1779 	else
1780 		VTPCI_OPS(hw) = &legacy_ops;
1781 }
1782 
1783 /*
1784  * This function is based on probe() function in virtio_pci.c
1785  * It returns 0 on success.
1786  */
1787 int
1788 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1789 {
1790 	struct virtio_hw *hw = eth_dev->data->dev_private;
1791 	int ret;
1792 
1793 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1794 
1795 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1796 
1797 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1798 		if (!hw->virtio_user_dev) {
1799 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1800 			if (ret)
1801 				return ret;
1802 		}
1803 
1804 		virtio_set_vtpci_ops(hw);
1805 		set_rxtx_funcs(eth_dev);
1806 
1807 		return 0;
1808 	}
1809 
1810 	/* Allocate memory for storing MAC addresses */
1811 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1812 	if (eth_dev->data->mac_addrs == NULL) {
1813 		PMD_INIT_LOG(ERR,
1814 			"Failed to allocate %d bytes needed to store MAC addresses",
1815 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1816 		return -ENOMEM;
1817 	}
1818 
1819 	hw->port_id = eth_dev->data->port_id;
1820 	/* For virtio_user case the hw->virtio_user_dev is populated by
1821 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1822 	 */
1823 	if (!hw->virtio_user_dev) {
1824 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1825 		if (ret)
1826 			goto out;
1827 	}
1828 
1829 	/* reset device and negotiate default features */
1830 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1831 	if (ret < 0)
1832 		goto out;
1833 
1834 	return 0;
1835 
1836 out:
1837 	rte_free(eth_dev->data->mac_addrs);
1838 	return ret;
1839 }
1840 
1841 static int
1842 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1843 {
1844 	PMD_INIT_FUNC_TRACE();
1845 
1846 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1847 		return 0;
1848 
1849 	virtio_dev_stop(eth_dev);
1850 	virtio_dev_close(eth_dev);
1851 
1852 	eth_dev->dev_ops = NULL;
1853 	eth_dev->tx_pkt_burst = NULL;
1854 	eth_dev->rx_pkt_burst = NULL;
1855 
1856 	if (eth_dev->device)
1857 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1858 
1859 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1860 
1861 	return 0;
1862 }
1863 
1864 static int vdpa_check_handler(__rte_unused const char *key,
1865 		const char *value, __rte_unused void *opaque)
1866 {
1867 	if (strcmp(value, "1"))
1868 		return -1;
1869 
1870 	return 0;
1871 }
1872 
1873 static int
1874 vdpa_mode_selected(struct rte_devargs *devargs)
1875 {
1876 	struct rte_kvargs *kvlist;
1877 	const char *key = "vdpa";
1878 	int ret = 0;
1879 
1880 	if (devargs == NULL)
1881 		return 0;
1882 
1883 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1884 	if (kvlist == NULL)
1885 		return 0;
1886 
1887 	if (!rte_kvargs_count(kvlist, key))
1888 		goto exit;
1889 
1890 	/* vdpa mode selected when there's a key-value pair: vdpa=1 */
1891 	if (rte_kvargs_process(kvlist, key,
1892 				vdpa_check_handler, NULL) < 0) {
1893 		goto exit;
1894 	}
1895 	ret = 1;
1896 
1897 exit:
1898 	rte_kvargs_free(kvlist);
1899 	return ret;
1900 }
1901 
1902 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1903 	struct rte_pci_device *pci_dev)
1904 {
1905 	if (rte_eal_iopl_init() != 0) {
1906 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1907 		return 1;
1908 	}
1909 
1910 	/* virtio pmd skips probe if device needs to work in vdpa mode */
1911 	if (vdpa_mode_selected(pci_dev->device.devargs))
1912 		return 1;
1913 
1914 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1915 		eth_virtio_dev_init);
1916 }
1917 
1918 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1919 {
1920 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1921 }
1922 
1923 static struct rte_pci_driver rte_virtio_pmd = {
1924 	.driver = {
1925 		.name = "net_virtio",
1926 	},
1927 	.id_table = pci_id_virtio_map,
1928 	.drv_flags = 0,
1929 	.probe = eth_virtio_pci_probe,
1930 	.remove = eth_virtio_pci_remove,
1931 };
1932 
1933 RTE_INIT(rte_virtio_pmd_init)
1934 {
1935 	rte_eal_iopl_init();
1936 	rte_pci_register(&rte_virtio_pmd);
1937 }
1938 
1939 static bool
1940 rx_offload_enabled(struct virtio_hw *hw)
1941 {
1942 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
1943 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1944 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
1945 }
1946 
1947 static bool
1948 tx_offload_enabled(struct virtio_hw *hw)
1949 {
1950 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
1951 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
1952 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
1953 }
1954 
1955 /*
1956  * Configure virtio device
1957  * It returns 0 on success.
1958  */
1959 static int
1960 virtio_dev_configure(struct rte_eth_dev *dev)
1961 {
1962 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1963 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
1964 	struct virtio_hw *hw = dev->data->dev_private;
1965 	uint64_t rx_offloads = rxmode->offloads;
1966 	uint64_t tx_offloads = txmode->offloads;
1967 	uint64_t req_features;
1968 	int ret;
1969 
1970 	PMD_INIT_LOG(DEBUG, "configure");
1971 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1972 
1973 	if (dev->data->dev_conf.intr_conf.rxq) {
1974 		ret = virtio_init_device(dev, hw->req_guest_features);
1975 		if (ret < 0)
1976 			return ret;
1977 	}
1978 
1979 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
1980 			   DEV_RX_OFFLOAD_TCP_CKSUM))
1981 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1982 
1983 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
1984 		req_features |=
1985 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1986 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1987 
1988 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
1989 			   DEV_TX_OFFLOAD_TCP_CKSUM))
1990 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
1991 
1992 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
1993 		req_features |=
1994 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
1995 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
1996 
1997 	/* if request features changed, reinit the device */
1998 	if (req_features != hw->req_guest_features) {
1999 		ret = virtio_init_device(dev, req_features);
2000 		if (ret < 0)
2001 			return ret;
2002 	}
2003 
2004 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2005 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2006 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2007 		PMD_DRV_LOG(ERR,
2008 			"rx checksum not available on this host");
2009 		return -ENOTSUP;
2010 	}
2011 
2012 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2013 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2014 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2015 		PMD_DRV_LOG(ERR,
2016 			"Large Receive Offload not available on this host");
2017 		return -ENOTSUP;
2018 	}
2019 
2020 	/* start control queue */
2021 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2022 		virtio_dev_cq_start(dev);
2023 
2024 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2025 		hw->vlan_strip = 1;
2026 
2027 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2028 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2029 		PMD_DRV_LOG(ERR,
2030 			    "vlan filtering not available on this host");
2031 		return -ENOTSUP;
2032 	}
2033 
2034 	hw->has_tx_offload = tx_offload_enabled(hw);
2035 	hw->has_rx_offload = rx_offload_enabled(hw);
2036 
2037 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2038 		/* Enable vector (0) for Link State Intrerrupt */
2039 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2040 				VIRTIO_MSI_NO_VECTOR) {
2041 			PMD_DRV_LOG(ERR, "failed to set config vector");
2042 			return -EBUSY;
2043 		}
2044 
2045 	rte_spinlock_init(&hw->state_lock);
2046 
2047 	hw->use_simple_rx = 1;
2048 
2049 	if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2050 		hw->use_inorder_tx = 1;
2051 		hw->use_inorder_rx = 1;
2052 		hw->use_simple_rx = 0;
2053 	}
2054 
2055 	if (vtpci_packed_queue(hw)) {
2056 		hw->use_simple_rx = 0;
2057 		hw->use_inorder_rx = 0;
2058 		hw->use_inorder_tx = 0;
2059 	}
2060 
2061 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2062 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2063 		hw->use_simple_rx = 0;
2064 	}
2065 #endif
2066 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2067 		 hw->use_simple_rx = 0;
2068 	}
2069 
2070 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2071 			   DEV_RX_OFFLOAD_TCP_CKSUM |
2072 			   DEV_RX_OFFLOAD_TCP_LRO |
2073 			   DEV_RX_OFFLOAD_VLAN_STRIP))
2074 		hw->use_simple_rx = 0;
2075 
2076 	hw->opened = true;
2077 
2078 	return 0;
2079 }
2080 
2081 
2082 static int
2083 virtio_dev_start(struct rte_eth_dev *dev)
2084 {
2085 	uint16_t nb_queues, i;
2086 	struct virtnet_rx *rxvq;
2087 	struct virtnet_tx *txvq __rte_unused;
2088 	struct virtio_hw *hw = dev->data->dev_private;
2089 	int ret;
2090 
2091 	/* Finish the initialization of the queues */
2092 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2093 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2094 		if (ret < 0)
2095 			return ret;
2096 	}
2097 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2098 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2099 		if (ret < 0)
2100 			return ret;
2101 	}
2102 
2103 	/* check if lsc interrupt feature is enabled */
2104 	if (dev->data->dev_conf.intr_conf.lsc) {
2105 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2106 			PMD_DRV_LOG(ERR, "link status not supported by host");
2107 			return -ENOTSUP;
2108 		}
2109 	}
2110 
2111 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2112 	 * in device configure, but it could be unmapped  when device is
2113 	 * stopped.
2114 	 */
2115 	if (dev->data->dev_conf.intr_conf.lsc ||
2116 	    dev->data->dev_conf.intr_conf.rxq) {
2117 		virtio_intr_disable(dev);
2118 
2119 		/* Setup interrupt callback  */
2120 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2121 			rte_intr_callback_register(dev->intr_handle,
2122 						   virtio_interrupt_handler,
2123 						   dev);
2124 
2125 		if (virtio_intr_enable(dev) < 0) {
2126 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2127 			return -EIO;
2128 		}
2129 	}
2130 
2131 	/*Notify the backend
2132 	 *Otherwise the tap backend might already stop its queue due to fullness.
2133 	 *vhost backend will have no chance to be waked up
2134 	 */
2135 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2136 	if (hw->max_queue_pairs > 1) {
2137 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2138 			return -EINVAL;
2139 	}
2140 
2141 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2142 
2143 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2144 		rxvq = dev->data->rx_queues[i];
2145 		/* Flush the old packets */
2146 		virtqueue_rxvq_flush(rxvq->vq);
2147 		virtqueue_notify(rxvq->vq);
2148 	}
2149 
2150 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2151 		txvq = dev->data->tx_queues[i];
2152 		virtqueue_notify(txvq->vq);
2153 	}
2154 
2155 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2156 
2157 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2158 		rxvq = dev->data->rx_queues[i];
2159 		VIRTQUEUE_DUMP(rxvq->vq);
2160 	}
2161 
2162 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2163 		txvq = dev->data->tx_queues[i];
2164 		VIRTQUEUE_DUMP(txvq->vq);
2165 	}
2166 
2167 	set_rxtx_funcs(dev);
2168 	hw->started = true;
2169 
2170 	/* Initialize Link state */
2171 	virtio_dev_link_update(dev, 0);
2172 
2173 	return 0;
2174 }
2175 
2176 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2177 {
2178 	struct virtio_hw *hw = dev->data->dev_private;
2179 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2180 	const char *type __rte_unused;
2181 	unsigned int i, mbuf_num = 0;
2182 	struct virtqueue *vq;
2183 	struct rte_mbuf *buf;
2184 	int queue_type;
2185 
2186 	if (hw->vqs == NULL)
2187 		return;
2188 
2189 	for (i = 0; i < nr_vq; i++) {
2190 		vq = hw->vqs[i];
2191 		if (!vq)
2192 			continue;
2193 
2194 		queue_type = virtio_get_queue_type(hw, i);
2195 		if (queue_type == VTNET_RQ)
2196 			type = "rxq";
2197 		else if (queue_type == VTNET_TQ)
2198 			type = "txq";
2199 		else
2200 			continue;
2201 
2202 		PMD_INIT_LOG(DEBUG,
2203 			"Before freeing %s[%d] used and unused buf",
2204 			type, i);
2205 		VIRTQUEUE_DUMP(vq);
2206 
2207 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2208 			rte_pktmbuf_free(buf);
2209 			mbuf_num++;
2210 		}
2211 
2212 		PMD_INIT_LOG(DEBUG,
2213 			"After freeing %s[%d] used and unused buf",
2214 			type, i);
2215 		VIRTQUEUE_DUMP(vq);
2216 	}
2217 
2218 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2219 }
2220 
2221 /*
2222  * Stop device: disable interrupt and mark link down
2223  */
2224 static void
2225 virtio_dev_stop(struct rte_eth_dev *dev)
2226 {
2227 	struct virtio_hw *hw = dev->data->dev_private;
2228 	struct rte_eth_link link;
2229 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2230 
2231 	PMD_INIT_LOG(DEBUG, "stop");
2232 
2233 	rte_spinlock_lock(&hw->state_lock);
2234 	if (!hw->started)
2235 		goto out_unlock;
2236 	hw->started = false;
2237 
2238 	if (intr_conf->lsc || intr_conf->rxq) {
2239 		virtio_intr_disable(dev);
2240 
2241 		/* Reset interrupt callback  */
2242 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2243 			rte_intr_callback_unregister(dev->intr_handle,
2244 						     virtio_interrupt_handler,
2245 						     dev);
2246 		}
2247 	}
2248 
2249 	memset(&link, 0, sizeof(link));
2250 	rte_eth_linkstatus_set(dev, &link);
2251 out_unlock:
2252 	rte_spinlock_unlock(&hw->state_lock);
2253 }
2254 
2255 static int
2256 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2257 {
2258 	struct rte_eth_link link;
2259 	uint16_t status;
2260 	struct virtio_hw *hw = dev->data->dev_private;
2261 
2262 	memset(&link, 0, sizeof(link));
2263 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
2264 	link.link_speed  = ETH_SPEED_NUM_10G;
2265 	link.link_autoneg = ETH_LINK_FIXED;
2266 
2267 	if (!hw->started) {
2268 		link.link_status = ETH_LINK_DOWN;
2269 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2270 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2271 		vtpci_read_dev_config(hw,
2272 				offsetof(struct virtio_net_config, status),
2273 				&status, sizeof(status));
2274 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2275 			link.link_status = ETH_LINK_DOWN;
2276 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2277 				     dev->data->port_id);
2278 		} else {
2279 			link.link_status = ETH_LINK_UP;
2280 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2281 				     dev->data->port_id);
2282 		}
2283 	} else {
2284 		link.link_status = ETH_LINK_UP;
2285 	}
2286 
2287 	return rte_eth_linkstatus_set(dev, &link);
2288 }
2289 
2290 static int
2291 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2292 {
2293 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2294 	struct virtio_hw *hw = dev->data->dev_private;
2295 	uint64_t offloads = rxmode->offloads;
2296 
2297 	if (mask & ETH_VLAN_FILTER_MASK) {
2298 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2299 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2300 
2301 			PMD_DRV_LOG(NOTICE,
2302 				"vlan filtering not available on this host");
2303 
2304 			return -ENOTSUP;
2305 		}
2306 	}
2307 
2308 	if (mask & ETH_VLAN_STRIP_MASK)
2309 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2310 
2311 	return 0;
2312 }
2313 
2314 static void
2315 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2316 {
2317 	uint64_t tso_mask, host_features;
2318 	struct virtio_hw *hw = dev->data->dev_private;
2319 
2320 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2321 
2322 	dev_info->max_rx_queues =
2323 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2324 	dev_info->max_tx_queues =
2325 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2326 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2327 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2328 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2329 
2330 	host_features = VTPCI_OPS(hw)->get_features(hw);
2331 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2332 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2333 		dev_info->rx_offload_capa |=
2334 			DEV_RX_OFFLOAD_TCP_CKSUM |
2335 			DEV_RX_OFFLOAD_UDP_CKSUM;
2336 	}
2337 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2338 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2339 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2340 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2341 	if ((host_features & tso_mask) == tso_mask)
2342 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2343 
2344 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2345 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2346 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2347 		dev_info->tx_offload_capa |=
2348 			DEV_TX_OFFLOAD_UDP_CKSUM |
2349 			DEV_TX_OFFLOAD_TCP_CKSUM;
2350 	}
2351 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2352 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2353 	if ((host_features & tso_mask) == tso_mask)
2354 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2355 }
2356 
2357 /*
2358  * It enables testpmd to collect per queue stats.
2359  */
2360 static int
2361 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2362 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2363 __rte_unused uint8_t is_rx)
2364 {
2365 	return 0;
2366 }
2367 
2368 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2369 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2370 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2371 
2372 RTE_INIT(virtio_init_log)
2373 {
2374 	virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2375 	if (virtio_logtype_init >= 0)
2376 		rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2377 	virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2378 	if (virtio_logtype_driver >= 0)
2379 		rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2380 }
2381