xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 0ecc27f28d202a3356a8601e6762b601ea822c4c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4 
5 #include <stdint.h>
6 #include <string.h>
7 #include <stdio.h>
8 #include <errno.h>
9 #include <unistd.h>
10 
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_memcpy.h>
14 #include <rte_string_fns.h>
15 #include <rte_memzone.h>
16 #include <rte_malloc.h>
17 #include <rte_branch_prediction.h>
18 #include <rte_pci.h>
19 #include <rte_bus_pci.h>
20 #include <rte_ether.h>
21 #include <rte_ip.h>
22 #include <rte_arp.h>
23 #include <rte_common.h>
24 #include <rte_errno.h>
25 #include <rte_cpuflags.h>
26 
27 #include <rte_memory.h>
28 #include <rte_eal.h>
29 #include <rte_dev.h>
30 #include <rte_cycles.h>
31 #include <rte_kvargs.h>
32 
33 #include "virtio_ethdev.h"
34 #include "virtio_pci.h"
35 #include "virtio_logs.h"
36 #include "virtqueue.h"
37 #include "virtio_rxtx.h"
38 #include "virtio_user/virtio_user_dev.h"
39 
40 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
41 static int  virtio_dev_configure(struct rte_eth_dev *dev);
42 static int  virtio_dev_start(struct rte_eth_dev *dev);
43 static void virtio_dev_stop(struct rte_eth_dev *dev);
44 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
45 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
46 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
47 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
48 static int virtio_dev_info_get(struct rte_eth_dev *dev,
49 				struct rte_eth_dev_info *dev_info);
50 static int virtio_dev_link_update(struct rte_eth_dev *dev,
51 	int wait_to_complete);
52 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
53 
54 static void virtio_set_hwaddr(struct virtio_hw *hw);
55 static void virtio_get_hwaddr(struct virtio_hw *hw);
56 
57 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
58 				 struct rte_eth_stats *stats);
59 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
60 				 struct rte_eth_xstat *xstats, unsigned n);
61 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
62 				       struct rte_eth_xstat_name *xstats_names,
63 				       unsigned limit);
64 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
65 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
66 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
67 				uint16_t vlan_id, int on);
68 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
69 				struct rte_ether_addr *mac_addr,
70 				uint32_t index, uint32_t vmdq);
71 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
72 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
73 				struct rte_ether_addr *mac_addr);
74 
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 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
125 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
126 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
127 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
128 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
129 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
130 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
131 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
132 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
133 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
134 };
135 
136 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
137 			    sizeof(rte_virtio_rxq_stat_strings[0]))
138 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
139 			    sizeof(rte_virtio_txq_stat_strings[0]))
140 
141 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
142 
143 static struct virtio_pmd_ctrl *
144 virtio_send_command_packed(struct virtnet_ctl *cvq,
145 			   struct virtio_pmd_ctrl *ctrl,
146 			   int *dlen, int pkt_num)
147 {
148 	struct virtqueue *vq = cvq->vq;
149 	int head;
150 	struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
151 	struct virtio_pmd_ctrl *result;
152 	uint16_t flags;
153 	int sum = 0;
154 	int nb_descs = 0;
155 	int k;
156 
157 	/*
158 	 * Format is enforced in qemu code:
159 	 * One TX packet for header;
160 	 * At least one TX packet per argument;
161 	 * One RX packet for ACK.
162 	 */
163 	head = vq->vq_avail_idx;
164 	flags = vq->vq_packed.cached_flags;
165 	desc[head].addr = cvq->virtio_net_hdr_mem;
166 	desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
167 	vq->vq_free_cnt--;
168 	nb_descs++;
169 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
170 		vq->vq_avail_idx -= vq->vq_nentries;
171 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
172 	}
173 
174 	for (k = 0; k < pkt_num; k++) {
175 		desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
176 			+ sizeof(struct virtio_net_ctrl_hdr)
177 			+ sizeof(ctrl->status) + sizeof(uint8_t) * sum;
178 		desc[vq->vq_avail_idx].len = dlen[k];
179 		desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
180 			vq->vq_packed.cached_flags;
181 		sum += dlen[k];
182 		vq->vq_free_cnt--;
183 		nb_descs++;
184 		if (++vq->vq_avail_idx >= vq->vq_nentries) {
185 			vq->vq_avail_idx -= vq->vq_nentries;
186 			vq->vq_packed.cached_flags ^=
187 				VRING_PACKED_DESC_F_AVAIL_USED;
188 		}
189 	}
190 
191 	desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
192 		+ sizeof(struct virtio_net_ctrl_hdr);
193 	desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
194 	desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
195 		vq->vq_packed.cached_flags;
196 	vq->vq_free_cnt--;
197 	nb_descs++;
198 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
199 		vq->vq_avail_idx -= vq->vq_nentries;
200 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
201 	}
202 
203 	virtio_wmb(vq->hw->weak_barriers);
204 	desc[head].flags = VRING_DESC_F_NEXT | flags;
205 
206 	virtio_wmb(vq->hw->weak_barriers);
207 	virtqueue_notify(vq);
208 
209 	/* wait for used descriptors in virtqueue */
210 	while (!desc_is_used(&desc[head], vq))
211 		usleep(100);
212 
213 	virtio_rmb(vq->hw->weak_barriers);
214 
215 	/* now get used descriptors */
216 	vq->vq_free_cnt += nb_descs;
217 	vq->vq_used_cons_idx += nb_descs;
218 	if (vq->vq_used_cons_idx >= vq->vq_nentries) {
219 		vq->vq_used_cons_idx -= vq->vq_nentries;
220 		vq->vq_packed.used_wrap_counter ^= 1;
221 	}
222 
223 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
224 			"vq->vq_avail_idx=%d\n"
225 			"vq->vq_used_cons_idx=%d\n"
226 			"vq->vq_packed.cached_flags=0x%x\n"
227 			"vq->vq_packed.used_wrap_counter=%d\n",
228 			vq->vq_free_cnt,
229 			vq->vq_avail_idx,
230 			vq->vq_used_cons_idx,
231 			vq->vq_packed.cached_flags,
232 			vq->vq_packed.used_wrap_counter);
233 
234 	result = cvq->virtio_net_hdr_mz->addr;
235 	return result;
236 }
237 
238 static struct virtio_pmd_ctrl *
239 virtio_send_command_split(struct virtnet_ctl *cvq,
240 			  struct virtio_pmd_ctrl *ctrl,
241 			  int *dlen, int pkt_num)
242 {
243 	struct virtio_pmd_ctrl *result;
244 	struct virtqueue *vq = cvq->vq;
245 	uint32_t head, i;
246 	int k, sum = 0;
247 
248 	head = vq->vq_desc_head_idx;
249 
250 	/*
251 	 * Format is enforced in qemu code:
252 	 * One TX packet for header;
253 	 * At least one TX packet per argument;
254 	 * One RX packet for ACK.
255 	 */
256 	vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
257 	vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
258 	vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
259 	vq->vq_free_cnt--;
260 	i = vq->vq_split.ring.desc[head].next;
261 
262 	for (k = 0; k < pkt_num; k++) {
263 		vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
264 		vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
265 			+ sizeof(struct virtio_net_ctrl_hdr)
266 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
267 		vq->vq_split.ring.desc[i].len = dlen[k];
268 		sum += dlen[k];
269 		vq->vq_free_cnt--;
270 		i = vq->vq_split.ring.desc[i].next;
271 	}
272 
273 	vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
274 	vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
275 			+ sizeof(struct virtio_net_ctrl_hdr);
276 	vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
277 	vq->vq_free_cnt--;
278 
279 	vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
280 
281 	vq_update_avail_ring(vq, head);
282 	vq_update_avail_idx(vq);
283 
284 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
285 
286 	virtqueue_notify(vq);
287 
288 	rte_rmb();
289 	while (VIRTQUEUE_NUSED(vq) == 0) {
290 		rte_rmb();
291 		usleep(100);
292 	}
293 
294 	while (VIRTQUEUE_NUSED(vq)) {
295 		uint32_t idx, desc_idx, used_idx;
296 		struct vring_used_elem *uep;
297 
298 		used_idx = (uint32_t)(vq->vq_used_cons_idx
299 				& (vq->vq_nentries - 1));
300 		uep = &vq->vq_split.ring.used->ring[used_idx];
301 		idx = (uint32_t) uep->id;
302 		desc_idx = idx;
303 
304 		while (vq->vq_split.ring.desc[desc_idx].flags &
305 				VRING_DESC_F_NEXT) {
306 			desc_idx = vq->vq_split.ring.desc[desc_idx].next;
307 			vq->vq_free_cnt++;
308 		}
309 
310 		vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
311 		vq->vq_desc_head_idx = idx;
312 
313 		vq->vq_used_cons_idx++;
314 		vq->vq_free_cnt++;
315 	}
316 
317 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
318 			vq->vq_free_cnt, vq->vq_desc_head_idx);
319 
320 	result = cvq->virtio_net_hdr_mz->addr;
321 	return result;
322 }
323 
324 static int
325 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
326 		    int *dlen, int pkt_num)
327 {
328 	virtio_net_ctrl_ack status = ~0;
329 	struct virtio_pmd_ctrl *result;
330 	struct virtqueue *vq;
331 
332 	ctrl->status = status;
333 
334 	if (!cvq || !cvq->vq) {
335 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
336 		return -1;
337 	}
338 
339 	rte_spinlock_lock(&cvq->lock);
340 	vq = cvq->vq;
341 
342 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
343 		"vq->hw->cvq = %p vq = %p",
344 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
345 
346 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
347 		rte_spinlock_unlock(&cvq->lock);
348 		return -1;
349 	}
350 
351 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
352 		sizeof(struct virtio_pmd_ctrl));
353 
354 	if (vtpci_packed_queue(vq->hw))
355 		result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
356 	else
357 		result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
358 
359 	rte_spinlock_unlock(&cvq->lock);
360 	return result->status;
361 }
362 
363 static int
364 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
365 {
366 	struct virtio_hw *hw = dev->data->dev_private;
367 	struct virtio_pmd_ctrl ctrl;
368 	int dlen[1];
369 	int ret;
370 
371 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
372 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
373 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
374 
375 	dlen[0] = sizeof(uint16_t);
376 
377 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
378 	if (ret) {
379 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
380 			  "failed, this is too late now...");
381 		return -EINVAL;
382 	}
383 
384 	return 0;
385 }
386 
387 static void
388 virtio_dev_queue_release(void *queue __rte_unused)
389 {
390 	/* do nothing */
391 }
392 
393 static uint16_t
394 virtio_get_nr_vq(struct virtio_hw *hw)
395 {
396 	uint16_t nr_vq = hw->max_queue_pairs * 2;
397 
398 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
399 		nr_vq += 1;
400 
401 	return nr_vq;
402 }
403 
404 static void
405 virtio_init_vring(struct virtqueue *vq)
406 {
407 	int size = vq->vq_nentries;
408 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
409 
410 	PMD_INIT_FUNC_TRACE();
411 
412 	memset(ring_mem, 0, vq->vq_ring_size);
413 
414 	vq->vq_used_cons_idx = 0;
415 	vq->vq_desc_head_idx = 0;
416 	vq->vq_avail_idx = 0;
417 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
418 	vq->vq_free_cnt = vq->vq_nentries;
419 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
420 	if (vtpci_packed_queue(vq->hw)) {
421 		vring_init_packed(&vq->vq_packed.ring, ring_mem,
422 				  VIRTIO_PCI_VRING_ALIGN, size);
423 		vring_desc_init_packed(vq, size);
424 	} else {
425 		struct vring *vr = &vq->vq_split.ring;
426 
427 		vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
428 		vring_desc_init_split(vr->desc, size);
429 	}
430 	/*
431 	 * Disable device(host) interrupting guest
432 	 */
433 	virtqueue_disable_intr(vq);
434 }
435 
436 static int
437 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
438 {
439 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
440 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
441 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
442 	unsigned int vq_size, size;
443 	struct virtio_hw *hw = dev->data->dev_private;
444 	struct virtnet_rx *rxvq = NULL;
445 	struct virtnet_tx *txvq = NULL;
446 	struct virtnet_ctl *cvq = NULL;
447 	struct virtqueue *vq;
448 	size_t sz_hdr_mz = 0;
449 	void *sw_ring = NULL;
450 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
451 	int ret;
452 	int numa_node = dev->device->numa_node;
453 
454 	PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
455 			vtpci_queue_idx, numa_node);
456 
457 	/*
458 	 * Read the virtqueue size from the Queue Size field
459 	 * Always power of 2 and if 0 virtqueue does not exist
460 	 */
461 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
462 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
463 	if (vq_size == 0) {
464 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
465 		return -EINVAL;
466 	}
467 
468 	if (!rte_is_power_of_2(vq_size)) {
469 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
470 		return -EINVAL;
471 	}
472 
473 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
474 		 dev->data->port_id, vtpci_queue_idx);
475 
476 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
477 				vq_size * sizeof(struct vq_desc_extra),
478 				RTE_CACHE_LINE_SIZE);
479 	if (queue_type == VTNET_TQ) {
480 		/*
481 		 * For each xmit packet, allocate a virtio_net_hdr
482 		 * and indirect ring elements
483 		 */
484 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
485 	} else if (queue_type == VTNET_CQ) {
486 		/* Allocate a page for control vq command, data and status */
487 		sz_hdr_mz = PAGE_SIZE;
488 	}
489 
490 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
491 				numa_node);
492 	if (vq == NULL) {
493 		PMD_INIT_LOG(ERR, "can not allocate vq");
494 		return -ENOMEM;
495 	}
496 	hw->vqs[vtpci_queue_idx] = vq;
497 
498 	vq->hw = hw;
499 	vq->vq_queue_index = vtpci_queue_idx;
500 	vq->vq_nentries = vq_size;
501 	if (vtpci_packed_queue(hw)) {
502 		vq->vq_packed.used_wrap_counter = 1;
503 		vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
504 		vq->vq_packed.event_flags_shadow = 0;
505 		if (queue_type == VTNET_RQ)
506 			vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
507 	}
508 
509 	/*
510 	 * Reserve a memzone for vring elements
511 	 */
512 	size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
513 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
514 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
515 		     size, vq->vq_ring_size);
516 
517 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
518 			numa_node, RTE_MEMZONE_IOVA_CONTIG,
519 			VIRTIO_PCI_VRING_ALIGN);
520 	if (mz == NULL) {
521 		if (rte_errno == EEXIST)
522 			mz = rte_memzone_lookup(vq_name);
523 		if (mz == NULL) {
524 			ret = -ENOMEM;
525 			goto fail_q_alloc;
526 		}
527 	}
528 
529 	memset(mz->addr, 0, mz->len);
530 
531 	vq->vq_ring_mem = mz->iova;
532 	vq->vq_ring_virt_mem = mz->addr;
533 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
534 		     (uint64_t)mz->iova);
535 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
536 		     (uint64_t)(uintptr_t)mz->addr);
537 
538 	virtio_init_vring(vq);
539 
540 	if (sz_hdr_mz) {
541 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
542 			 dev->data->port_id, vtpci_queue_idx);
543 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
544 				numa_node, RTE_MEMZONE_IOVA_CONTIG,
545 				RTE_CACHE_LINE_SIZE);
546 		if (hdr_mz == NULL) {
547 			if (rte_errno == EEXIST)
548 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
549 			if (hdr_mz == NULL) {
550 				ret = -ENOMEM;
551 				goto fail_q_alloc;
552 			}
553 		}
554 	}
555 
556 	if (queue_type == VTNET_RQ) {
557 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
558 			       sizeof(vq->sw_ring[0]);
559 
560 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
561 				RTE_CACHE_LINE_SIZE, numa_node);
562 		if (!sw_ring) {
563 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
564 			ret = -ENOMEM;
565 			goto fail_q_alloc;
566 		}
567 
568 		vq->sw_ring = sw_ring;
569 		rxvq = &vq->rxq;
570 		rxvq->vq = vq;
571 		rxvq->port_id = dev->data->port_id;
572 		rxvq->mz = mz;
573 	} else if (queue_type == VTNET_TQ) {
574 		txvq = &vq->txq;
575 		txvq->vq = vq;
576 		txvq->port_id = dev->data->port_id;
577 		txvq->mz = mz;
578 		txvq->virtio_net_hdr_mz = hdr_mz;
579 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
580 	} else if (queue_type == VTNET_CQ) {
581 		cvq = &vq->cq;
582 		cvq->vq = vq;
583 		cvq->mz = mz;
584 		cvq->virtio_net_hdr_mz = hdr_mz;
585 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
586 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
587 
588 		hw->cvq = cvq;
589 	}
590 
591 	/* For virtio_user case (that is when hw->dev is NULL), we use
592 	 * virtual address. And we need properly set _offset_, please see
593 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
594 	 */
595 	if (!hw->virtio_user_dev)
596 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
597 	else {
598 		vq->vq_ring_mem = (uintptr_t)mz->addr;
599 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
600 		if (queue_type == VTNET_TQ)
601 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
602 		else if (queue_type == VTNET_CQ)
603 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
604 	}
605 
606 	if (queue_type == VTNET_TQ) {
607 		struct virtio_tx_region *txr;
608 		unsigned int i;
609 
610 		txr = hdr_mz->addr;
611 		memset(txr, 0, vq_size * sizeof(*txr));
612 		for (i = 0; i < vq_size; i++) {
613 			struct vring_desc *start_dp = txr[i].tx_indir;
614 
615 			/* first indirect descriptor is always the tx header */
616 			if (!vtpci_packed_queue(hw)) {
617 				vring_desc_init_split(start_dp,
618 						      RTE_DIM(txr[i].tx_indir));
619 				start_dp->addr = txvq->virtio_net_hdr_mem
620 					+ i * sizeof(*txr)
621 					+ offsetof(struct virtio_tx_region,
622 						   tx_hdr);
623 				start_dp->len = hw->vtnet_hdr_size;
624 				start_dp->flags = VRING_DESC_F_NEXT;
625 			}
626 		}
627 	}
628 
629 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
630 		PMD_INIT_LOG(ERR, "setup_queue failed");
631 		return -EINVAL;
632 	}
633 
634 	return 0;
635 
636 fail_q_alloc:
637 	rte_free(sw_ring);
638 	rte_memzone_free(hdr_mz);
639 	rte_memzone_free(mz);
640 	rte_free(vq);
641 
642 	return ret;
643 }
644 
645 static void
646 virtio_free_queues(struct virtio_hw *hw)
647 {
648 	uint16_t nr_vq = virtio_get_nr_vq(hw);
649 	struct virtqueue *vq;
650 	int queue_type;
651 	uint16_t i;
652 
653 	if (hw->vqs == NULL)
654 		return;
655 
656 	for (i = 0; i < nr_vq; i++) {
657 		vq = hw->vqs[i];
658 		if (!vq)
659 			continue;
660 
661 		queue_type = virtio_get_queue_type(hw, i);
662 		if (queue_type == VTNET_RQ) {
663 			rte_free(vq->sw_ring);
664 			rte_memzone_free(vq->rxq.mz);
665 		} else if (queue_type == VTNET_TQ) {
666 			rte_memzone_free(vq->txq.mz);
667 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
668 		} else {
669 			rte_memzone_free(vq->cq.mz);
670 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
671 		}
672 
673 		rte_free(vq);
674 		hw->vqs[i] = NULL;
675 	}
676 
677 	rte_free(hw->vqs);
678 	hw->vqs = NULL;
679 }
680 
681 static int
682 virtio_alloc_queues(struct rte_eth_dev *dev)
683 {
684 	struct virtio_hw *hw = dev->data->dev_private;
685 	uint16_t nr_vq = virtio_get_nr_vq(hw);
686 	uint16_t i;
687 	int ret;
688 
689 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
690 	if (!hw->vqs) {
691 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
692 		return -ENOMEM;
693 	}
694 
695 	for (i = 0; i < nr_vq; i++) {
696 		ret = virtio_init_queue(dev, i);
697 		if (ret < 0) {
698 			virtio_free_queues(hw);
699 			return ret;
700 		}
701 	}
702 
703 	return 0;
704 }
705 
706 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
707 
708 static void
709 virtio_dev_close(struct rte_eth_dev *dev)
710 {
711 	struct virtio_hw *hw = dev->data->dev_private;
712 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
713 
714 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
715 
716 	if (!hw->opened)
717 		return;
718 	hw->opened = false;
719 
720 	/* reset the NIC */
721 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
722 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
723 	if (intr_conf->rxq)
724 		virtio_queues_unbind_intr(dev);
725 
726 	if (intr_conf->lsc || intr_conf->rxq) {
727 		virtio_intr_disable(dev);
728 		rte_intr_efd_disable(dev->intr_handle);
729 		rte_free(dev->intr_handle->intr_vec);
730 		dev->intr_handle->intr_vec = NULL;
731 	}
732 
733 	vtpci_reset(hw);
734 	virtio_dev_free_mbufs(dev);
735 	virtio_free_queues(hw);
736 
737 #ifdef RTE_VIRTIO_USER
738 	if (hw->virtio_user_dev)
739 		virtio_user_dev_uninit(hw->virtio_user_dev);
740 	else
741 #endif
742 	if (dev->device) {
743 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
744 		if (!hw->modern)
745 			rte_pci_ioport_unmap(VTPCI_IO(hw));
746 	}
747 }
748 
749 static int
750 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
751 {
752 	struct virtio_hw *hw = dev->data->dev_private;
753 	struct virtio_pmd_ctrl ctrl;
754 	int dlen[1];
755 	int ret;
756 
757 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
758 		PMD_INIT_LOG(INFO, "host does not support rx control");
759 		return -ENOTSUP;
760 	}
761 
762 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
763 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
764 	ctrl.data[0] = 1;
765 	dlen[0] = 1;
766 
767 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
768 	if (ret) {
769 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
770 		return -EAGAIN;
771 	}
772 
773 	return 0;
774 }
775 
776 static int
777 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
778 {
779 	struct virtio_hw *hw = dev->data->dev_private;
780 	struct virtio_pmd_ctrl ctrl;
781 	int dlen[1];
782 	int ret;
783 
784 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
785 		PMD_INIT_LOG(INFO, "host does not support rx control");
786 		return -ENOTSUP;
787 	}
788 
789 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
790 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
791 	ctrl.data[0] = 0;
792 	dlen[0] = 1;
793 
794 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
795 	if (ret) {
796 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
797 		return -EAGAIN;
798 	}
799 
800 	return 0;
801 }
802 
803 static int
804 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
805 {
806 	struct virtio_hw *hw = dev->data->dev_private;
807 	struct virtio_pmd_ctrl ctrl;
808 	int dlen[1];
809 	int ret;
810 
811 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
812 		PMD_INIT_LOG(INFO, "host does not support rx control");
813 		return -ENOTSUP;
814 	}
815 
816 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
817 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
818 	ctrl.data[0] = 1;
819 	dlen[0] = 1;
820 
821 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
822 	if (ret) {
823 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
824 		return -EAGAIN;
825 	}
826 
827 	return 0;
828 }
829 
830 static int
831 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
832 {
833 	struct virtio_hw *hw = dev->data->dev_private;
834 	struct virtio_pmd_ctrl ctrl;
835 	int dlen[1];
836 	int ret;
837 
838 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
839 		PMD_INIT_LOG(INFO, "host does not support rx control");
840 		return -ENOTSUP;
841 	}
842 
843 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
844 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
845 	ctrl.data[0] = 0;
846 	dlen[0] = 1;
847 
848 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
849 	if (ret) {
850 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
851 		return -EAGAIN;
852 	}
853 
854 	return 0;
855 }
856 
857 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
858 static int
859 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
860 {
861 	struct virtio_hw *hw = dev->data->dev_private;
862 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
863 				 hw->vtnet_hdr_size;
864 	uint32_t frame_size = mtu + ether_hdr_len;
865 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
866 
867 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
868 
869 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
870 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
871 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
872 		return -EINVAL;
873 	}
874 	return 0;
875 }
876 
877 static int
878 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
879 {
880 	struct virtio_hw *hw = dev->data->dev_private;
881 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
882 	struct virtqueue *vq = rxvq->vq;
883 
884 	virtqueue_enable_intr(vq);
885 	virtio_mb(hw->weak_barriers);
886 	return 0;
887 }
888 
889 static int
890 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
891 {
892 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
893 	struct virtqueue *vq = rxvq->vq;
894 
895 	virtqueue_disable_intr(vq);
896 	return 0;
897 }
898 
899 /*
900  * dev_ops for virtio, bare necessities for basic operation
901  */
902 static const struct eth_dev_ops virtio_eth_dev_ops = {
903 	.dev_configure           = virtio_dev_configure,
904 	.dev_start               = virtio_dev_start,
905 	.dev_stop                = virtio_dev_stop,
906 	.dev_close               = virtio_dev_close,
907 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
908 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
909 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
910 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
911 	.mtu_set                 = virtio_mtu_set,
912 	.dev_infos_get           = virtio_dev_info_get,
913 	.stats_get               = virtio_dev_stats_get,
914 	.xstats_get              = virtio_dev_xstats_get,
915 	.xstats_get_names        = virtio_dev_xstats_get_names,
916 	.stats_reset             = virtio_dev_stats_reset,
917 	.xstats_reset            = virtio_dev_stats_reset,
918 	.link_update             = virtio_dev_link_update,
919 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
920 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
921 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
922 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
923 	.rx_queue_release        = virtio_dev_queue_release,
924 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
925 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
926 	.tx_queue_release        = virtio_dev_queue_release,
927 	/* collect stats per queue */
928 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
929 	.vlan_filter_set         = virtio_vlan_filter_set,
930 	.mac_addr_add            = virtio_mac_addr_add,
931 	.mac_addr_remove         = virtio_mac_addr_remove,
932 	.mac_addr_set            = virtio_mac_addr_set,
933 };
934 
935 /*
936  * dev_ops for virtio-user in secondary processes, as we just have
937  * some limited supports currently.
938  */
939 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
940 	.dev_infos_get           = virtio_dev_info_get,
941 	.stats_get               = virtio_dev_stats_get,
942 	.xstats_get              = virtio_dev_xstats_get,
943 	.xstats_get_names        = virtio_dev_xstats_get_names,
944 	.stats_reset             = virtio_dev_stats_reset,
945 	.xstats_reset            = virtio_dev_stats_reset,
946 	/* collect stats per queue */
947 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
948 };
949 
950 static void
951 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
952 {
953 	unsigned i;
954 
955 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
956 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
957 		if (txvq == NULL)
958 			continue;
959 
960 		stats->opackets += txvq->stats.packets;
961 		stats->obytes += txvq->stats.bytes;
962 
963 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
964 			stats->q_opackets[i] = txvq->stats.packets;
965 			stats->q_obytes[i] = txvq->stats.bytes;
966 		}
967 	}
968 
969 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
970 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
971 		if (rxvq == NULL)
972 			continue;
973 
974 		stats->ipackets += rxvq->stats.packets;
975 		stats->ibytes += rxvq->stats.bytes;
976 		stats->ierrors += rxvq->stats.errors;
977 
978 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
979 			stats->q_ipackets[i] = rxvq->stats.packets;
980 			stats->q_ibytes[i] = rxvq->stats.bytes;
981 		}
982 	}
983 
984 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
985 }
986 
987 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
988 				       struct rte_eth_xstat_name *xstats_names,
989 				       __rte_unused unsigned limit)
990 {
991 	unsigned i;
992 	unsigned count = 0;
993 	unsigned t;
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 (xstats_names != NULL) {
999 		/* Note: limit checked in rte_eth_xstats_names() */
1000 
1001 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
1002 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1003 			if (rxvq == NULL)
1004 				continue;
1005 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1006 				snprintf(xstats_names[count].name,
1007 					sizeof(xstats_names[count].name),
1008 					"rx_q%u_%s", i,
1009 					rte_virtio_rxq_stat_strings[t].name);
1010 				count++;
1011 			}
1012 		}
1013 
1014 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
1015 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
1016 			if (txvq == NULL)
1017 				continue;
1018 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1019 				snprintf(xstats_names[count].name,
1020 					sizeof(xstats_names[count].name),
1021 					"tx_q%u_%s", i,
1022 					rte_virtio_txq_stat_strings[t].name);
1023 				count++;
1024 			}
1025 		}
1026 		return count;
1027 	}
1028 	return nstats;
1029 }
1030 
1031 static int
1032 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1033 		      unsigned n)
1034 {
1035 	unsigned i;
1036 	unsigned count = 0;
1037 
1038 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1039 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1040 
1041 	if (n < nstats)
1042 		return nstats;
1043 
1044 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1045 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1046 
1047 		if (rxvq == NULL)
1048 			continue;
1049 
1050 		unsigned t;
1051 
1052 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1053 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1054 				rte_virtio_rxq_stat_strings[t].offset);
1055 			xstats[count].id = count;
1056 			count++;
1057 		}
1058 	}
1059 
1060 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1061 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1062 
1063 		if (txvq == NULL)
1064 			continue;
1065 
1066 		unsigned t;
1067 
1068 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1069 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1070 				rte_virtio_txq_stat_strings[t].offset);
1071 			xstats[count].id = count;
1072 			count++;
1073 		}
1074 	}
1075 
1076 	return count;
1077 }
1078 
1079 static int
1080 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1081 {
1082 	virtio_update_stats(dev, stats);
1083 
1084 	return 0;
1085 }
1086 
1087 static int
1088 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1089 {
1090 	unsigned int i;
1091 
1092 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1093 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1094 		if (txvq == NULL)
1095 			continue;
1096 
1097 		txvq->stats.packets = 0;
1098 		txvq->stats.bytes = 0;
1099 		txvq->stats.multicast = 0;
1100 		txvq->stats.broadcast = 0;
1101 		memset(txvq->stats.size_bins, 0,
1102 		       sizeof(txvq->stats.size_bins[0]) * 8);
1103 	}
1104 
1105 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1106 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1107 		if (rxvq == NULL)
1108 			continue;
1109 
1110 		rxvq->stats.packets = 0;
1111 		rxvq->stats.bytes = 0;
1112 		rxvq->stats.errors = 0;
1113 		rxvq->stats.multicast = 0;
1114 		rxvq->stats.broadcast = 0;
1115 		memset(rxvq->stats.size_bins, 0,
1116 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1117 	}
1118 
1119 	return 0;
1120 }
1121 
1122 static void
1123 virtio_set_hwaddr(struct virtio_hw *hw)
1124 {
1125 	vtpci_write_dev_config(hw,
1126 			offsetof(struct virtio_net_config, mac),
1127 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1128 }
1129 
1130 static void
1131 virtio_get_hwaddr(struct virtio_hw *hw)
1132 {
1133 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1134 		vtpci_read_dev_config(hw,
1135 			offsetof(struct virtio_net_config, mac),
1136 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1137 	} else {
1138 		rte_eth_random_addr(&hw->mac_addr[0]);
1139 		virtio_set_hwaddr(hw);
1140 	}
1141 }
1142 
1143 static int
1144 virtio_mac_table_set(struct virtio_hw *hw,
1145 		     const struct virtio_net_ctrl_mac *uc,
1146 		     const struct virtio_net_ctrl_mac *mc)
1147 {
1148 	struct virtio_pmd_ctrl ctrl;
1149 	int err, len[2];
1150 
1151 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1152 		PMD_DRV_LOG(INFO, "host does not support mac table");
1153 		return -1;
1154 	}
1155 
1156 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1157 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1158 
1159 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1160 	memcpy(ctrl.data, uc, len[0]);
1161 
1162 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1163 	memcpy(ctrl.data + len[0], mc, len[1]);
1164 
1165 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1166 	if (err != 0)
1167 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1168 	return err;
1169 }
1170 
1171 static int
1172 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1173 		    uint32_t index, uint32_t vmdq __rte_unused)
1174 {
1175 	struct virtio_hw *hw = dev->data->dev_private;
1176 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1177 	unsigned int i;
1178 	struct virtio_net_ctrl_mac *uc, *mc;
1179 
1180 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1181 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1182 		return -EINVAL;
1183 	}
1184 
1185 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1186 		sizeof(uc->entries));
1187 	uc->entries = 0;
1188 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1189 		sizeof(mc->entries));
1190 	mc->entries = 0;
1191 
1192 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1193 		const struct rte_ether_addr *addr
1194 			= (i == index) ? mac_addr : addrs + i;
1195 		struct virtio_net_ctrl_mac *tbl
1196 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1197 
1198 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1199 	}
1200 
1201 	return virtio_mac_table_set(hw, uc, mc);
1202 }
1203 
1204 static void
1205 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1206 {
1207 	struct virtio_hw *hw = dev->data->dev_private;
1208 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1209 	struct virtio_net_ctrl_mac *uc, *mc;
1210 	unsigned int i;
1211 
1212 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1213 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1214 		return;
1215 	}
1216 
1217 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1218 		sizeof(uc->entries));
1219 	uc->entries = 0;
1220 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1221 		sizeof(mc->entries));
1222 	mc->entries = 0;
1223 
1224 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1225 		struct virtio_net_ctrl_mac *tbl;
1226 
1227 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1228 			continue;
1229 
1230 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1231 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1232 			RTE_ETHER_ADDR_LEN);
1233 	}
1234 
1235 	virtio_mac_table_set(hw, uc, mc);
1236 }
1237 
1238 static int
1239 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1240 {
1241 	struct virtio_hw *hw = dev->data->dev_private;
1242 
1243 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1244 
1245 	/* Use atomic update if available */
1246 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1247 		struct virtio_pmd_ctrl ctrl;
1248 		int len = RTE_ETHER_ADDR_LEN;
1249 
1250 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1251 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1252 
1253 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1254 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1255 	}
1256 
1257 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1258 		return -ENOTSUP;
1259 
1260 	virtio_set_hwaddr(hw);
1261 	return 0;
1262 }
1263 
1264 static int
1265 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1266 {
1267 	struct virtio_hw *hw = dev->data->dev_private;
1268 	struct virtio_pmd_ctrl ctrl;
1269 	int len;
1270 
1271 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1272 		return -ENOTSUP;
1273 
1274 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1275 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1276 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1277 	len = sizeof(vlan_id);
1278 
1279 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1280 }
1281 
1282 static int
1283 virtio_intr_unmask(struct rte_eth_dev *dev)
1284 {
1285 	struct virtio_hw *hw = dev->data->dev_private;
1286 
1287 	if (rte_intr_ack(dev->intr_handle) < 0)
1288 		return -1;
1289 
1290 	if (!hw->virtio_user_dev)
1291 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1292 
1293 	return 0;
1294 }
1295 
1296 static int
1297 virtio_intr_enable(struct rte_eth_dev *dev)
1298 {
1299 	struct virtio_hw *hw = dev->data->dev_private;
1300 
1301 	if (rte_intr_enable(dev->intr_handle) < 0)
1302 		return -1;
1303 
1304 	if (!hw->virtio_user_dev)
1305 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1306 
1307 	return 0;
1308 }
1309 
1310 static int
1311 virtio_intr_disable(struct rte_eth_dev *dev)
1312 {
1313 	struct virtio_hw *hw = dev->data->dev_private;
1314 
1315 	if (rte_intr_disable(dev->intr_handle) < 0)
1316 		return -1;
1317 
1318 	if (!hw->virtio_user_dev)
1319 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1320 
1321 	return 0;
1322 }
1323 
1324 static int
1325 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1326 {
1327 	uint64_t host_features;
1328 
1329 	/* Prepare guest_features: feature that driver wants to support */
1330 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1331 		req_features);
1332 
1333 	/* Read device(host) feature bits */
1334 	host_features = VTPCI_OPS(hw)->get_features(hw);
1335 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1336 		host_features);
1337 
1338 	/* If supported, ensure MTU value is valid before acknowledging it. */
1339 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1340 		struct virtio_net_config config;
1341 
1342 		vtpci_read_dev_config(hw,
1343 			offsetof(struct virtio_net_config, mtu),
1344 			&config.mtu, sizeof(config.mtu));
1345 
1346 		if (config.mtu < RTE_ETHER_MIN_MTU)
1347 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1348 	}
1349 
1350 	/*
1351 	 * Negotiate features: Subset of device feature bits are written back
1352 	 * guest feature bits.
1353 	 */
1354 	hw->guest_features = req_features;
1355 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1356 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1357 		hw->guest_features);
1358 
1359 	if (hw->modern) {
1360 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1361 			PMD_INIT_LOG(ERR,
1362 				"VIRTIO_F_VERSION_1 features is not enabled.");
1363 			return -1;
1364 		}
1365 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1366 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1367 			PMD_INIT_LOG(ERR,
1368 				"failed to set FEATURES_OK status!");
1369 			return -1;
1370 		}
1371 	}
1372 
1373 	hw->req_guest_features = req_features;
1374 
1375 	return 0;
1376 }
1377 
1378 int
1379 virtio_dev_pause(struct rte_eth_dev *dev)
1380 {
1381 	struct virtio_hw *hw = dev->data->dev_private;
1382 
1383 	rte_spinlock_lock(&hw->state_lock);
1384 
1385 	if (hw->started == 0) {
1386 		/* Device is just stopped. */
1387 		rte_spinlock_unlock(&hw->state_lock);
1388 		return -1;
1389 	}
1390 	hw->started = 0;
1391 	/*
1392 	 * Prevent the worker threads from touching queues to avoid contention,
1393 	 * 1 ms should be enough for the ongoing Tx function to finish.
1394 	 */
1395 	rte_delay_ms(1);
1396 	return 0;
1397 }
1398 
1399 /*
1400  * Recover hw state to let the worker threads continue.
1401  */
1402 void
1403 virtio_dev_resume(struct rte_eth_dev *dev)
1404 {
1405 	struct virtio_hw *hw = dev->data->dev_private;
1406 
1407 	hw->started = 1;
1408 	rte_spinlock_unlock(&hw->state_lock);
1409 }
1410 
1411 /*
1412  * Should be called only after device is paused.
1413  */
1414 int
1415 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1416 		int nb_pkts)
1417 {
1418 	struct virtio_hw *hw = dev->data->dev_private;
1419 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1420 	int ret;
1421 
1422 	hw->inject_pkts = tx_pkts;
1423 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1424 	hw->inject_pkts = NULL;
1425 
1426 	return ret;
1427 }
1428 
1429 static void
1430 virtio_notify_peers(struct rte_eth_dev *dev)
1431 {
1432 	struct virtio_hw *hw = dev->data->dev_private;
1433 	struct virtnet_rx *rxvq;
1434 	struct rte_mbuf *rarp_mbuf;
1435 
1436 	if (!dev->data->rx_queues)
1437 		return;
1438 
1439 	rxvq = dev->data->rx_queues[0];
1440 	if (!rxvq)
1441 		return;
1442 
1443 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1444 			(struct rte_ether_addr *)hw->mac_addr);
1445 	if (rarp_mbuf == NULL) {
1446 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1447 		return;
1448 	}
1449 
1450 	/* If virtio port just stopped, no need to send RARP */
1451 	if (virtio_dev_pause(dev) < 0) {
1452 		rte_pktmbuf_free(rarp_mbuf);
1453 		return;
1454 	}
1455 
1456 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1457 	virtio_dev_resume(dev);
1458 }
1459 
1460 static void
1461 virtio_ack_link_announce(struct rte_eth_dev *dev)
1462 {
1463 	struct virtio_hw *hw = dev->data->dev_private;
1464 	struct virtio_pmd_ctrl ctrl;
1465 
1466 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1467 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1468 
1469 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1470 }
1471 
1472 /*
1473  * Process virtio config changed interrupt. Call the callback
1474  * if link state changed, generate gratuitous RARP packet if
1475  * the status indicates an ANNOUNCE.
1476  */
1477 void
1478 virtio_interrupt_handler(void *param)
1479 {
1480 	struct rte_eth_dev *dev = param;
1481 	struct virtio_hw *hw = dev->data->dev_private;
1482 	uint8_t isr;
1483 	uint16_t status;
1484 
1485 	/* Read interrupt status which clears interrupt */
1486 	isr = vtpci_isr(hw);
1487 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1488 
1489 	if (virtio_intr_unmask(dev) < 0)
1490 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1491 
1492 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1493 		if (virtio_dev_link_update(dev, 0) == 0)
1494 			_rte_eth_dev_callback_process(dev,
1495 						      RTE_ETH_EVENT_INTR_LSC,
1496 						      NULL);
1497 
1498 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1499 			vtpci_read_dev_config(hw,
1500 				offsetof(struct virtio_net_config, status),
1501 				&status, sizeof(status));
1502 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1503 				virtio_notify_peers(dev);
1504 				if (hw->cvq)
1505 					virtio_ack_link_announce(dev);
1506 			}
1507 		}
1508 	}
1509 }
1510 
1511 /* set rx and tx handlers according to what is supported */
1512 static void
1513 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1514 {
1515 	struct virtio_hw *hw = eth_dev->data->dev_private;
1516 
1517 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1518 	if (vtpci_packed_queue(hw)) {
1519 		PMD_INIT_LOG(INFO,
1520 			"virtio: using packed ring %s Tx path on port %u",
1521 			hw->use_inorder_tx ? "inorder" : "standard",
1522 			eth_dev->data->port_id);
1523 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1524 	} else {
1525 		if (hw->use_inorder_tx) {
1526 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1527 				eth_dev->data->port_id);
1528 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1529 		} else {
1530 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1531 				eth_dev->data->port_id);
1532 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1533 		}
1534 	}
1535 
1536 	if (vtpci_packed_queue(hw)) {
1537 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1538 			PMD_INIT_LOG(INFO,
1539 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1540 				eth_dev->data->port_id);
1541 			eth_dev->rx_pkt_burst =
1542 				&virtio_recv_mergeable_pkts_packed;
1543 		} else {
1544 			PMD_INIT_LOG(INFO,
1545 				"virtio: using packed ring standard Rx path on port %u",
1546 				eth_dev->data->port_id);
1547 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1548 		}
1549 	} else {
1550 		if (hw->use_simple_rx) {
1551 			PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1552 				eth_dev->data->port_id);
1553 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1554 		} else if (hw->use_inorder_rx) {
1555 			PMD_INIT_LOG(INFO,
1556 				"virtio: using inorder Rx path on port %u",
1557 				eth_dev->data->port_id);
1558 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1559 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1560 			PMD_INIT_LOG(INFO,
1561 				"virtio: using mergeable buffer Rx path on port %u",
1562 				eth_dev->data->port_id);
1563 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1564 		} else {
1565 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1566 				eth_dev->data->port_id);
1567 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1568 		}
1569 	}
1570 
1571 }
1572 
1573 /* Only support 1:1 queue/interrupt mapping so far.
1574  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1575  * interrupt vectors (<N+1).
1576  */
1577 static int
1578 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1579 {
1580 	uint32_t i;
1581 	struct virtio_hw *hw = dev->data->dev_private;
1582 
1583 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1584 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1585 		dev->intr_handle->intr_vec[i] = i + 1;
1586 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1587 						 VIRTIO_MSI_NO_VECTOR) {
1588 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1589 			return -EBUSY;
1590 		}
1591 	}
1592 
1593 	return 0;
1594 }
1595 
1596 static void
1597 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1598 {
1599 	uint32_t i;
1600 	struct virtio_hw *hw = dev->data->dev_private;
1601 
1602 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1603 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1604 		VTPCI_OPS(hw)->set_queue_irq(hw,
1605 					     hw->vqs[i * VTNET_CQ],
1606 					     VIRTIO_MSI_NO_VECTOR);
1607 }
1608 
1609 static int
1610 virtio_configure_intr(struct rte_eth_dev *dev)
1611 {
1612 	struct virtio_hw *hw = dev->data->dev_private;
1613 
1614 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1615 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1616 		return -ENOTSUP;
1617 	}
1618 
1619 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1620 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1621 		return -1;
1622 	}
1623 
1624 	if (!dev->intr_handle->intr_vec) {
1625 		dev->intr_handle->intr_vec =
1626 			rte_zmalloc("intr_vec",
1627 				    hw->max_queue_pairs * sizeof(int), 0);
1628 		if (!dev->intr_handle->intr_vec) {
1629 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1630 				     hw->max_queue_pairs);
1631 			return -ENOMEM;
1632 		}
1633 	}
1634 
1635 	/* Re-register callback to update max_intr */
1636 	rte_intr_callback_unregister(dev->intr_handle,
1637 				     virtio_interrupt_handler,
1638 				     dev);
1639 	rte_intr_callback_register(dev->intr_handle,
1640 				   virtio_interrupt_handler,
1641 				   dev);
1642 
1643 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1644 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1645 	 * And for legacy devices, this should be done before queue/vec binding
1646 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1647 	 * (22) will be ignored.
1648 	 */
1649 	if (virtio_intr_enable(dev) < 0) {
1650 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1651 		return -1;
1652 	}
1653 
1654 	if (virtio_queues_bind_intr(dev) < 0) {
1655 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1656 		return -1;
1657 	}
1658 
1659 	return 0;
1660 }
1661 
1662 /* reset device and renegotiate features if needed */
1663 static int
1664 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1665 {
1666 	struct virtio_hw *hw = eth_dev->data->dev_private;
1667 	struct virtio_net_config *config;
1668 	struct virtio_net_config local_config;
1669 	struct rte_pci_device *pci_dev = NULL;
1670 	int ret;
1671 
1672 	/* Reset the device although not necessary at startup */
1673 	vtpci_reset(hw);
1674 
1675 	if (hw->vqs) {
1676 		virtio_dev_free_mbufs(eth_dev);
1677 		virtio_free_queues(hw);
1678 	}
1679 
1680 	/* Tell the host we've noticed this device. */
1681 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1682 
1683 	/* Tell the host we've known how to drive the device. */
1684 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1685 	if (virtio_negotiate_features(hw, req_features) < 0)
1686 		return -1;
1687 
1688 	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1689 
1690 	if (!hw->virtio_user_dev)
1691 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1692 
1693 	/* If host does not support both status and MSI-X then disable LSC */
1694 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1695 	    hw->use_msix != VIRTIO_MSIX_NONE)
1696 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1697 	else
1698 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1699 
1700 	/* Setting up rx_header size for the device */
1701 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1702 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1703 	    vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1704 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1705 	else
1706 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1707 
1708 	/* Copy the permanent MAC address to: virtio_hw */
1709 	virtio_get_hwaddr(hw);
1710 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1711 			&eth_dev->data->mac_addrs[0]);
1712 	PMD_INIT_LOG(DEBUG,
1713 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1714 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1715 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1716 
1717 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1718 		config = &local_config;
1719 
1720 		vtpci_read_dev_config(hw,
1721 			offsetof(struct virtio_net_config, mac),
1722 			&config->mac, sizeof(config->mac));
1723 
1724 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1725 			vtpci_read_dev_config(hw,
1726 				offsetof(struct virtio_net_config, status),
1727 				&config->status, sizeof(config->status));
1728 		} else {
1729 			PMD_INIT_LOG(DEBUG,
1730 				     "VIRTIO_NET_F_STATUS is not supported");
1731 			config->status = 0;
1732 		}
1733 
1734 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1735 			vtpci_read_dev_config(hw,
1736 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1737 				&config->max_virtqueue_pairs,
1738 				sizeof(config->max_virtqueue_pairs));
1739 		} else {
1740 			PMD_INIT_LOG(DEBUG,
1741 				     "VIRTIO_NET_F_MQ is not supported");
1742 			config->max_virtqueue_pairs = 1;
1743 		}
1744 
1745 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1746 
1747 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1748 			vtpci_read_dev_config(hw,
1749 				offsetof(struct virtio_net_config, mtu),
1750 				&config->mtu,
1751 				sizeof(config->mtu));
1752 
1753 			/*
1754 			 * MTU value has already been checked at negotiation
1755 			 * time, but check again in case it has changed since
1756 			 * then, which should not happen.
1757 			 */
1758 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1759 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1760 						config->mtu);
1761 				return -1;
1762 			}
1763 
1764 			hw->max_mtu = config->mtu;
1765 			/* Set initial MTU to maximum one supported by vhost */
1766 			eth_dev->data->mtu = config->mtu;
1767 
1768 		} else {
1769 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1770 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1771 		}
1772 
1773 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1774 				config->max_virtqueue_pairs);
1775 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1776 		PMD_INIT_LOG(DEBUG,
1777 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1778 				config->mac[0], config->mac[1],
1779 				config->mac[2], config->mac[3],
1780 				config->mac[4], config->mac[5]);
1781 	} else {
1782 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1783 		hw->max_queue_pairs = 1;
1784 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1785 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1786 	}
1787 
1788 	ret = virtio_alloc_queues(eth_dev);
1789 	if (ret < 0)
1790 		return ret;
1791 
1792 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1793 		if (virtio_configure_intr(eth_dev) < 0) {
1794 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1795 			virtio_free_queues(hw);
1796 			return -1;
1797 		}
1798 	}
1799 
1800 	vtpci_reinit_complete(hw);
1801 
1802 	if (pci_dev)
1803 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1804 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1805 			pci_dev->id.device_id);
1806 
1807 	return 0;
1808 }
1809 
1810 /*
1811  * Remap the PCI device again (IO port map for legacy device and
1812  * memory map for modern device), so that the secondary process
1813  * could have the PCI initiated correctly.
1814  */
1815 static int
1816 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1817 {
1818 	if (hw->modern) {
1819 		/*
1820 		 * We don't have to re-parse the PCI config space, since
1821 		 * rte_pci_map_device() makes sure the mapped address
1822 		 * in secondary process would equal to the one mapped in
1823 		 * the primary process: error will be returned if that
1824 		 * requirement is not met.
1825 		 *
1826 		 * That said, we could simply reuse all cap pointers
1827 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1828 		 * primary process, which is stored in shared memory.
1829 		 */
1830 		if (rte_pci_map_device(pci_dev)) {
1831 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1832 			return -1;
1833 		}
1834 	} else {
1835 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1836 			return -1;
1837 	}
1838 
1839 	return 0;
1840 }
1841 
1842 static void
1843 virtio_set_vtpci_ops(struct virtio_hw *hw)
1844 {
1845 #ifdef RTE_VIRTIO_USER
1846 	if (hw->virtio_user_dev)
1847 		VTPCI_OPS(hw) = &virtio_user_ops;
1848 	else
1849 #endif
1850 	if (hw->modern)
1851 		VTPCI_OPS(hw) = &modern_ops;
1852 	else
1853 		VTPCI_OPS(hw) = &legacy_ops;
1854 }
1855 
1856 /*
1857  * This function is based on probe() function in virtio_pci.c
1858  * It returns 0 on success.
1859  */
1860 int
1861 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1862 {
1863 	struct virtio_hw *hw = eth_dev->data->dev_private;
1864 	int ret;
1865 
1866 	if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1867 		PMD_INIT_LOG(ERR,
1868 			"Not sufficient headroom required = %d, avail = %d",
1869 			(int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1870 			RTE_PKTMBUF_HEADROOM);
1871 
1872 		return -1;
1873 	}
1874 
1875 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1876 
1877 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1878 		if (!hw->virtio_user_dev) {
1879 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1880 			if (ret)
1881 				return ret;
1882 		}
1883 
1884 		virtio_set_vtpci_ops(hw);
1885 		set_rxtx_funcs(eth_dev);
1886 
1887 		return 0;
1888 	}
1889 
1890 	/*
1891 	 * Pass the information to the rte_eth_dev_close() that it should also
1892 	 * release the private port resources.
1893 	 */
1894 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1895 
1896 	/* Allocate memory for storing MAC addresses */
1897 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1898 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1899 	if (eth_dev->data->mac_addrs == NULL) {
1900 		PMD_INIT_LOG(ERR,
1901 			"Failed to allocate %d bytes needed to store MAC addresses",
1902 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1903 		return -ENOMEM;
1904 	}
1905 
1906 	hw->port_id = eth_dev->data->port_id;
1907 	/* For virtio_user case the hw->virtio_user_dev is populated by
1908 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1909 	 */
1910 	if (!hw->virtio_user_dev) {
1911 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1912 		if (ret)
1913 			goto err_vtpci_init;
1914 	}
1915 
1916 	/* reset device and negotiate default features */
1917 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1918 	if (ret < 0)
1919 		goto err_virtio_init;
1920 
1921 	hw->opened = true;
1922 
1923 	return 0;
1924 
1925 err_virtio_init:
1926 	if (!hw->virtio_user_dev) {
1927 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1928 		if (!hw->modern)
1929 			rte_pci_ioport_unmap(VTPCI_IO(hw));
1930 	}
1931 err_vtpci_init:
1932 	rte_free(eth_dev->data->mac_addrs);
1933 	eth_dev->data->mac_addrs = NULL;
1934 	return ret;
1935 }
1936 
1937 static int
1938 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1939 {
1940 	PMD_INIT_FUNC_TRACE();
1941 
1942 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1943 		return 0;
1944 
1945 	virtio_dev_stop(eth_dev);
1946 	virtio_dev_close(eth_dev);
1947 
1948 	eth_dev->dev_ops = NULL;
1949 	eth_dev->tx_pkt_burst = NULL;
1950 	eth_dev->rx_pkt_burst = NULL;
1951 
1952 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1953 
1954 	return 0;
1955 }
1956 
1957 static int vdpa_check_handler(__rte_unused const char *key,
1958 		const char *value, __rte_unused void *opaque)
1959 {
1960 	if (strcmp(value, "1"))
1961 		return -1;
1962 
1963 	return 0;
1964 }
1965 
1966 static int
1967 vdpa_mode_selected(struct rte_devargs *devargs)
1968 {
1969 	struct rte_kvargs *kvlist;
1970 	const char *key = "vdpa";
1971 	int ret = 0;
1972 
1973 	if (devargs == NULL)
1974 		return 0;
1975 
1976 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1977 	if (kvlist == NULL)
1978 		return 0;
1979 
1980 	if (!rte_kvargs_count(kvlist, key))
1981 		goto exit;
1982 
1983 	/* vdpa mode selected when there's a key-value pair: vdpa=1 */
1984 	if (rte_kvargs_process(kvlist, key,
1985 				vdpa_check_handler, NULL) < 0) {
1986 		goto exit;
1987 	}
1988 	ret = 1;
1989 
1990 exit:
1991 	rte_kvargs_free(kvlist);
1992 	return ret;
1993 }
1994 
1995 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1996 	struct rte_pci_device *pci_dev)
1997 {
1998 	if (rte_eal_iopl_init() != 0) {
1999 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
2000 		return 1;
2001 	}
2002 
2003 	/* virtio pmd skips probe if device needs to work in vdpa mode */
2004 	if (vdpa_mode_selected(pci_dev->device.devargs))
2005 		return 1;
2006 
2007 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2008 		eth_virtio_dev_init);
2009 }
2010 
2011 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2012 {
2013 	int ret;
2014 
2015 	ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2016 	/* Port has already been released by close. */
2017 	if (ret == -ENODEV)
2018 		ret = 0;
2019 	return ret;
2020 }
2021 
2022 static struct rte_pci_driver rte_virtio_pmd = {
2023 	.driver = {
2024 		.name = "net_virtio",
2025 	},
2026 	.id_table = pci_id_virtio_map,
2027 	.drv_flags = 0,
2028 	.probe = eth_virtio_pci_probe,
2029 	.remove = eth_virtio_pci_remove,
2030 };
2031 
2032 RTE_INIT(rte_virtio_pmd_init)
2033 {
2034 	rte_eal_iopl_init();
2035 	rte_pci_register(&rte_virtio_pmd);
2036 }
2037 
2038 static bool
2039 rx_offload_enabled(struct virtio_hw *hw)
2040 {
2041 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2042 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2043 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2044 }
2045 
2046 static bool
2047 tx_offload_enabled(struct virtio_hw *hw)
2048 {
2049 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2050 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2051 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2052 }
2053 
2054 /*
2055  * Configure virtio device
2056  * It returns 0 on success.
2057  */
2058 static int
2059 virtio_dev_configure(struct rte_eth_dev *dev)
2060 {
2061 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2062 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2063 	struct virtio_hw *hw = dev->data->dev_private;
2064 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2065 		hw->vtnet_hdr_size;
2066 	uint64_t rx_offloads = rxmode->offloads;
2067 	uint64_t tx_offloads = txmode->offloads;
2068 	uint64_t req_features;
2069 	int ret;
2070 
2071 	PMD_INIT_LOG(DEBUG, "configure");
2072 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2073 
2074 	if (dev->data->dev_conf.intr_conf.rxq) {
2075 		ret = virtio_init_device(dev, hw->req_guest_features);
2076 		if (ret < 0)
2077 			return ret;
2078 	}
2079 
2080 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2081 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2082 
2083 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2084 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2085 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2086 
2087 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2088 		req_features |=
2089 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2090 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2091 
2092 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2093 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2094 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2095 
2096 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2097 		req_features |=
2098 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2099 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2100 
2101 	/* if request features changed, reinit the device */
2102 	if (req_features != hw->req_guest_features) {
2103 		ret = virtio_init_device(dev, req_features);
2104 		if (ret < 0)
2105 			return ret;
2106 	}
2107 
2108 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2109 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2110 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2111 		PMD_DRV_LOG(ERR,
2112 			"rx checksum not available on this host");
2113 		return -ENOTSUP;
2114 	}
2115 
2116 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2117 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2118 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2119 		PMD_DRV_LOG(ERR,
2120 			"Large Receive Offload not available on this host");
2121 		return -ENOTSUP;
2122 	}
2123 
2124 	/* start control queue */
2125 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2126 		virtio_dev_cq_start(dev);
2127 
2128 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2129 		hw->vlan_strip = 1;
2130 
2131 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2132 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2133 		PMD_DRV_LOG(ERR,
2134 			    "vlan filtering not available on this host");
2135 		return -ENOTSUP;
2136 	}
2137 
2138 	hw->has_tx_offload = tx_offload_enabled(hw);
2139 	hw->has_rx_offload = rx_offload_enabled(hw);
2140 
2141 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2142 		/* Enable vector (0) for Link State Intrerrupt */
2143 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2144 				VIRTIO_MSI_NO_VECTOR) {
2145 			PMD_DRV_LOG(ERR, "failed to set config vector");
2146 			return -EBUSY;
2147 		}
2148 
2149 	rte_spinlock_init(&hw->state_lock);
2150 
2151 	hw->use_simple_rx = 1;
2152 
2153 	if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2154 		hw->use_inorder_tx = 1;
2155 		hw->use_inorder_rx = 1;
2156 		hw->use_simple_rx = 0;
2157 	}
2158 
2159 	if (vtpci_packed_queue(hw)) {
2160 		hw->use_simple_rx = 0;
2161 		hw->use_inorder_rx = 0;
2162 	}
2163 
2164 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2165 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2166 		hw->use_simple_rx = 0;
2167 	}
2168 #endif
2169 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2170 		 hw->use_simple_rx = 0;
2171 	}
2172 
2173 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2174 			   DEV_RX_OFFLOAD_TCP_CKSUM |
2175 			   DEV_RX_OFFLOAD_TCP_LRO |
2176 			   DEV_RX_OFFLOAD_VLAN_STRIP))
2177 		hw->use_simple_rx = 0;
2178 
2179 	return 0;
2180 }
2181 
2182 
2183 static int
2184 virtio_dev_start(struct rte_eth_dev *dev)
2185 {
2186 	uint16_t nb_queues, i;
2187 	struct virtnet_rx *rxvq;
2188 	struct virtnet_tx *txvq __rte_unused;
2189 	struct virtio_hw *hw = dev->data->dev_private;
2190 	int ret;
2191 
2192 	/* Finish the initialization of the queues */
2193 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2194 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2195 		if (ret < 0)
2196 			return ret;
2197 	}
2198 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2199 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2200 		if (ret < 0)
2201 			return ret;
2202 	}
2203 
2204 	/* check if lsc interrupt feature is enabled */
2205 	if (dev->data->dev_conf.intr_conf.lsc) {
2206 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2207 			PMD_DRV_LOG(ERR, "link status not supported by host");
2208 			return -ENOTSUP;
2209 		}
2210 	}
2211 
2212 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2213 	 * in device configure, but it could be unmapped  when device is
2214 	 * stopped.
2215 	 */
2216 	if (dev->data->dev_conf.intr_conf.lsc ||
2217 	    dev->data->dev_conf.intr_conf.rxq) {
2218 		virtio_intr_disable(dev);
2219 
2220 		/* Setup interrupt callback  */
2221 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2222 			rte_intr_callback_register(dev->intr_handle,
2223 						   virtio_interrupt_handler,
2224 						   dev);
2225 
2226 		if (virtio_intr_enable(dev) < 0) {
2227 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2228 			return -EIO;
2229 		}
2230 	}
2231 
2232 	/*Notify the backend
2233 	 *Otherwise the tap backend might already stop its queue due to fullness.
2234 	 *vhost backend will have no chance to be waked up
2235 	 */
2236 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2237 	if (hw->max_queue_pairs > 1) {
2238 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2239 			return -EINVAL;
2240 	}
2241 
2242 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2243 
2244 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2245 		rxvq = dev->data->rx_queues[i];
2246 		/* Flush the old packets */
2247 		virtqueue_rxvq_flush(rxvq->vq);
2248 		virtqueue_notify(rxvq->vq);
2249 	}
2250 
2251 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2252 		txvq = dev->data->tx_queues[i];
2253 		virtqueue_notify(txvq->vq);
2254 	}
2255 
2256 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2257 
2258 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2259 		rxvq = dev->data->rx_queues[i];
2260 		VIRTQUEUE_DUMP(rxvq->vq);
2261 	}
2262 
2263 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2264 		txvq = dev->data->tx_queues[i];
2265 		VIRTQUEUE_DUMP(txvq->vq);
2266 	}
2267 
2268 	set_rxtx_funcs(dev);
2269 	hw->started = true;
2270 
2271 	/* Initialize Link state */
2272 	virtio_dev_link_update(dev, 0);
2273 
2274 	return 0;
2275 }
2276 
2277 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2278 {
2279 	struct virtio_hw *hw = dev->data->dev_private;
2280 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2281 	const char *type __rte_unused;
2282 	unsigned int i, mbuf_num = 0;
2283 	struct virtqueue *vq;
2284 	struct rte_mbuf *buf;
2285 	int queue_type;
2286 
2287 	if (hw->vqs == NULL)
2288 		return;
2289 
2290 	for (i = 0; i < nr_vq; i++) {
2291 		vq = hw->vqs[i];
2292 		if (!vq)
2293 			continue;
2294 
2295 		queue_type = virtio_get_queue_type(hw, i);
2296 		if (queue_type == VTNET_RQ)
2297 			type = "rxq";
2298 		else if (queue_type == VTNET_TQ)
2299 			type = "txq";
2300 		else
2301 			continue;
2302 
2303 		PMD_INIT_LOG(DEBUG,
2304 			"Before freeing %s[%d] used and unused buf",
2305 			type, i);
2306 		VIRTQUEUE_DUMP(vq);
2307 
2308 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2309 			rte_pktmbuf_free(buf);
2310 			mbuf_num++;
2311 		}
2312 
2313 		PMD_INIT_LOG(DEBUG,
2314 			"After freeing %s[%d] used and unused buf",
2315 			type, i);
2316 		VIRTQUEUE_DUMP(vq);
2317 	}
2318 
2319 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2320 }
2321 
2322 /*
2323  * Stop device: disable interrupt and mark link down
2324  */
2325 static void
2326 virtio_dev_stop(struct rte_eth_dev *dev)
2327 {
2328 	struct virtio_hw *hw = dev->data->dev_private;
2329 	struct rte_eth_link link;
2330 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2331 
2332 	PMD_INIT_LOG(DEBUG, "stop");
2333 
2334 	rte_spinlock_lock(&hw->state_lock);
2335 	if (!hw->started)
2336 		goto out_unlock;
2337 	hw->started = false;
2338 
2339 	if (intr_conf->lsc || intr_conf->rxq) {
2340 		virtio_intr_disable(dev);
2341 
2342 		/* Reset interrupt callback  */
2343 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2344 			rte_intr_callback_unregister(dev->intr_handle,
2345 						     virtio_interrupt_handler,
2346 						     dev);
2347 		}
2348 	}
2349 
2350 	memset(&link, 0, sizeof(link));
2351 	rte_eth_linkstatus_set(dev, &link);
2352 out_unlock:
2353 	rte_spinlock_unlock(&hw->state_lock);
2354 }
2355 
2356 static int
2357 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2358 {
2359 	struct rte_eth_link link;
2360 	uint16_t status;
2361 	struct virtio_hw *hw = dev->data->dev_private;
2362 
2363 	memset(&link, 0, sizeof(link));
2364 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
2365 	link.link_speed  = ETH_SPEED_NUM_10G;
2366 	link.link_autoneg = ETH_LINK_FIXED;
2367 
2368 	if (!hw->started) {
2369 		link.link_status = ETH_LINK_DOWN;
2370 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2371 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2372 		vtpci_read_dev_config(hw,
2373 				offsetof(struct virtio_net_config, status),
2374 				&status, sizeof(status));
2375 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2376 			link.link_status = ETH_LINK_DOWN;
2377 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2378 				     dev->data->port_id);
2379 		} else {
2380 			link.link_status = ETH_LINK_UP;
2381 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2382 				     dev->data->port_id);
2383 		}
2384 	} else {
2385 		link.link_status = ETH_LINK_UP;
2386 	}
2387 
2388 	return rte_eth_linkstatus_set(dev, &link);
2389 }
2390 
2391 static int
2392 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2393 {
2394 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2395 	struct virtio_hw *hw = dev->data->dev_private;
2396 	uint64_t offloads = rxmode->offloads;
2397 
2398 	if (mask & ETH_VLAN_FILTER_MASK) {
2399 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2400 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2401 
2402 			PMD_DRV_LOG(NOTICE,
2403 				"vlan filtering not available on this host");
2404 
2405 			return -ENOTSUP;
2406 		}
2407 	}
2408 
2409 	if (mask & ETH_VLAN_STRIP_MASK)
2410 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2411 
2412 	return 0;
2413 }
2414 
2415 static int
2416 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2417 {
2418 	uint64_t tso_mask, host_features;
2419 	struct virtio_hw *hw = dev->data->dev_private;
2420 
2421 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2422 
2423 	dev_info->max_rx_queues =
2424 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2425 	dev_info->max_tx_queues =
2426 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2427 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2428 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2429 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2430 
2431 	host_features = VTPCI_OPS(hw)->get_features(hw);
2432 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2433 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2434 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2435 		dev_info->rx_offload_capa |=
2436 			DEV_RX_OFFLOAD_TCP_CKSUM |
2437 			DEV_RX_OFFLOAD_UDP_CKSUM;
2438 	}
2439 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2440 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2441 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2442 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2443 	if ((host_features & tso_mask) == tso_mask)
2444 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2445 
2446 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2447 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2448 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2449 		dev_info->tx_offload_capa |=
2450 			DEV_TX_OFFLOAD_UDP_CKSUM |
2451 			DEV_TX_OFFLOAD_TCP_CKSUM;
2452 	}
2453 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2454 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2455 	if ((host_features & tso_mask) == tso_mask)
2456 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2457 
2458 	return 0;
2459 }
2460 
2461 /*
2462  * It enables testpmd to collect per queue stats.
2463  */
2464 static int
2465 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2466 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2467 __rte_unused uint8_t is_rx)
2468 {
2469 	return 0;
2470 }
2471 
2472 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2473 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2474 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2475 
2476 RTE_INIT(virtio_init_log)
2477 {
2478 	virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2479 	if (virtio_logtype_init >= 0)
2480 		rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2481 	virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2482 	if (virtio_logtype_driver >= 0)
2483 		rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2484 }
2485