xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 9a710863decb1cdb98efbdd5e11df3ebcfcc37b6)
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 void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
45 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
46 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
47 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
48 static void 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 void 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 void
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;
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 }
771 
772 static void
773 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
774 {
775 	struct virtio_hw *hw = dev->data->dev_private;
776 	struct virtio_pmd_ctrl ctrl;
777 	int dlen[1];
778 	int ret;
779 
780 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
781 		PMD_INIT_LOG(INFO, "host does not support rx control");
782 		return;
783 	}
784 
785 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
786 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
787 	ctrl.data[0] = 0;
788 	dlen[0] = 1;
789 
790 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
791 	if (ret)
792 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
793 }
794 
795 static void
796 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
797 {
798 	struct virtio_hw *hw = dev->data->dev_private;
799 	struct virtio_pmd_ctrl ctrl;
800 	int dlen[1];
801 	int ret;
802 
803 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
804 		PMD_INIT_LOG(INFO, "host does not support rx control");
805 		return;
806 	}
807 
808 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
809 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
810 	ctrl.data[0] = 1;
811 	dlen[0] = 1;
812 
813 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
814 	if (ret)
815 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
816 }
817 
818 static void
819 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
820 {
821 	struct virtio_hw *hw = dev->data->dev_private;
822 	struct virtio_pmd_ctrl ctrl;
823 	int dlen[1];
824 	int ret;
825 
826 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
827 		PMD_INIT_LOG(INFO, "host does not support rx control");
828 		return;
829 	}
830 
831 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
832 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
833 	ctrl.data[0] = 0;
834 	dlen[0] = 1;
835 
836 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
837 	if (ret)
838 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
839 }
840 
841 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
842 static int
843 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
844 {
845 	struct virtio_hw *hw = dev->data->dev_private;
846 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
847 				 hw->vtnet_hdr_size;
848 	uint32_t frame_size = mtu + ether_hdr_len;
849 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
850 
851 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
852 
853 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
854 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
855 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
856 		return -EINVAL;
857 	}
858 	return 0;
859 }
860 
861 static int
862 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
863 {
864 	struct virtio_hw *hw = dev->data->dev_private;
865 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
866 	struct virtqueue *vq = rxvq->vq;
867 
868 	virtqueue_enable_intr(vq);
869 	virtio_mb(hw->weak_barriers);
870 	return 0;
871 }
872 
873 static int
874 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
875 {
876 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
877 	struct virtqueue *vq = rxvq->vq;
878 
879 	virtqueue_disable_intr(vq);
880 	return 0;
881 }
882 
883 /*
884  * dev_ops for virtio, bare necessities for basic operation
885  */
886 static const struct eth_dev_ops virtio_eth_dev_ops = {
887 	.dev_configure           = virtio_dev_configure,
888 	.dev_start               = virtio_dev_start,
889 	.dev_stop                = virtio_dev_stop,
890 	.dev_close               = virtio_dev_close,
891 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
892 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
893 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
894 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
895 	.mtu_set                 = virtio_mtu_set,
896 	.dev_infos_get           = virtio_dev_info_get,
897 	.stats_get               = virtio_dev_stats_get,
898 	.xstats_get              = virtio_dev_xstats_get,
899 	.xstats_get_names        = virtio_dev_xstats_get_names,
900 	.stats_reset             = virtio_dev_stats_reset,
901 	.xstats_reset            = virtio_dev_stats_reset,
902 	.link_update             = virtio_dev_link_update,
903 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
904 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
905 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
906 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
907 	.rx_queue_release        = virtio_dev_queue_release,
908 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
909 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
910 	.tx_queue_release        = virtio_dev_queue_release,
911 	/* collect stats per queue */
912 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
913 	.vlan_filter_set         = virtio_vlan_filter_set,
914 	.mac_addr_add            = virtio_mac_addr_add,
915 	.mac_addr_remove         = virtio_mac_addr_remove,
916 	.mac_addr_set            = virtio_mac_addr_set,
917 };
918 
919 /*
920  * dev_ops for virtio-user in secondary processes, as we just have
921  * some limited supports currently.
922  */
923 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
924 	.dev_infos_get           = virtio_dev_info_get,
925 	.stats_get               = virtio_dev_stats_get,
926 	.xstats_get              = virtio_dev_xstats_get,
927 	.xstats_get_names        = virtio_dev_xstats_get_names,
928 	.stats_reset             = virtio_dev_stats_reset,
929 	.xstats_reset            = virtio_dev_stats_reset,
930 	/* collect stats per queue */
931 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
932 };
933 
934 static void
935 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
936 {
937 	unsigned i;
938 
939 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
940 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
941 		if (txvq == NULL)
942 			continue;
943 
944 		stats->opackets += txvq->stats.packets;
945 		stats->obytes += txvq->stats.bytes;
946 
947 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
948 			stats->q_opackets[i] = txvq->stats.packets;
949 			stats->q_obytes[i] = txvq->stats.bytes;
950 		}
951 	}
952 
953 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
954 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
955 		if (rxvq == NULL)
956 			continue;
957 
958 		stats->ipackets += rxvq->stats.packets;
959 		stats->ibytes += rxvq->stats.bytes;
960 		stats->ierrors += rxvq->stats.errors;
961 
962 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
963 			stats->q_ipackets[i] = rxvq->stats.packets;
964 			stats->q_ibytes[i] = rxvq->stats.bytes;
965 		}
966 	}
967 
968 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
969 }
970 
971 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
972 				       struct rte_eth_xstat_name *xstats_names,
973 				       __rte_unused unsigned limit)
974 {
975 	unsigned i;
976 	unsigned count = 0;
977 	unsigned t;
978 
979 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
980 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
981 
982 	if (xstats_names != NULL) {
983 		/* Note: limit checked in rte_eth_xstats_names() */
984 
985 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
986 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
987 			if (rxvq == NULL)
988 				continue;
989 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
990 				snprintf(xstats_names[count].name,
991 					sizeof(xstats_names[count].name),
992 					"rx_q%u_%s", i,
993 					rte_virtio_rxq_stat_strings[t].name);
994 				count++;
995 			}
996 		}
997 
998 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
999 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
1000 			if (txvq == NULL)
1001 				continue;
1002 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1003 				snprintf(xstats_names[count].name,
1004 					sizeof(xstats_names[count].name),
1005 					"tx_q%u_%s", i,
1006 					rte_virtio_txq_stat_strings[t].name);
1007 				count++;
1008 			}
1009 		}
1010 		return count;
1011 	}
1012 	return nstats;
1013 }
1014 
1015 static int
1016 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1017 		      unsigned n)
1018 {
1019 	unsigned i;
1020 	unsigned count = 0;
1021 
1022 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1023 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1024 
1025 	if (n < nstats)
1026 		return nstats;
1027 
1028 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1029 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1030 
1031 		if (rxvq == NULL)
1032 			continue;
1033 
1034 		unsigned t;
1035 
1036 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1037 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1038 				rte_virtio_rxq_stat_strings[t].offset);
1039 			xstats[count].id = count;
1040 			count++;
1041 		}
1042 	}
1043 
1044 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1045 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1046 
1047 		if (txvq == NULL)
1048 			continue;
1049 
1050 		unsigned t;
1051 
1052 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1053 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1054 				rte_virtio_txq_stat_strings[t].offset);
1055 			xstats[count].id = count;
1056 			count++;
1057 		}
1058 	}
1059 
1060 	return count;
1061 }
1062 
1063 static int
1064 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1065 {
1066 	virtio_update_stats(dev, stats);
1067 
1068 	return 0;
1069 }
1070 
1071 static void
1072 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1073 {
1074 	unsigned int i;
1075 
1076 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1077 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1078 		if (txvq == NULL)
1079 			continue;
1080 
1081 		txvq->stats.packets = 0;
1082 		txvq->stats.bytes = 0;
1083 		txvq->stats.multicast = 0;
1084 		txvq->stats.broadcast = 0;
1085 		memset(txvq->stats.size_bins, 0,
1086 		       sizeof(txvq->stats.size_bins[0]) * 8);
1087 	}
1088 
1089 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1090 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1091 		if (rxvq == NULL)
1092 			continue;
1093 
1094 		rxvq->stats.packets = 0;
1095 		rxvq->stats.bytes = 0;
1096 		rxvq->stats.errors = 0;
1097 		rxvq->stats.multicast = 0;
1098 		rxvq->stats.broadcast = 0;
1099 		memset(rxvq->stats.size_bins, 0,
1100 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1101 	}
1102 }
1103 
1104 static void
1105 virtio_set_hwaddr(struct virtio_hw *hw)
1106 {
1107 	vtpci_write_dev_config(hw,
1108 			offsetof(struct virtio_net_config, mac),
1109 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1110 }
1111 
1112 static void
1113 virtio_get_hwaddr(struct virtio_hw *hw)
1114 {
1115 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1116 		vtpci_read_dev_config(hw,
1117 			offsetof(struct virtio_net_config, mac),
1118 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1119 	} else {
1120 		rte_eth_random_addr(&hw->mac_addr[0]);
1121 		virtio_set_hwaddr(hw);
1122 	}
1123 }
1124 
1125 static int
1126 virtio_mac_table_set(struct virtio_hw *hw,
1127 		     const struct virtio_net_ctrl_mac *uc,
1128 		     const struct virtio_net_ctrl_mac *mc)
1129 {
1130 	struct virtio_pmd_ctrl ctrl;
1131 	int err, len[2];
1132 
1133 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1134 		PMD_DRV_LOG(INFO, "host does not support mac table");
1135 		return -1;
1136 	}
1137 
1138 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1139 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1140 
1141 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1142 	memcpy(ctrl.data, uc, len[0]);
1143 
1144 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1145 	memcpy(ctrl.data + len[0], mc, len[1]);
1146 
1147 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1148 	if (err != 0)
1149 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1150 	return err;
1151 }
1152 
1153 static int
1154 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1155 		    uint32_t index, uint32_t vmdq __rte_unused)
1156 {
1157 	struct virtio_hw *hw = dev->data->dev_private;
1158 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1159 	unsigned int i;
1160 	struct virtio_net_ctrl_mac *uc, *mc;
1161 
1162 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1163 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1164 		return -EINVAL;
1165 	}
1166 
1167 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1168 		sizeof(uc->entries));
1169 	uc->entries = 0;
1170 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1171 		sizeof(mc->entries));
1172 	mc->entries = 0;
1173 
1174 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1175 		const struct rte_ether_addr *addr
1176 			= (i == index) ? mac_addr : addrs + i;
1177 		struct virtio_net_ctrl_mac *tbl
1178 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1179 
1180 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1181 	}
1182 
1183 	return virtio_mac_table_set(hw, uc, mc);
1184 }
1185 
1186 static void
1187 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1188 {
1189 	struct virtio_hw *hw = dev->data->dev_private;
1190 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1191 	struct virtio_net_ctrl_mac *uc, *mc;
1192 	unsigned int i;
1193 
1194 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1195 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1196 		return;
1197 	}
1198 
1199 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1200 		sizeof(uc->entries));
1201 	uc->entries = 0;
1202 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1203 		sizeof(mc->entries));
1204 	mc->entries = 0;
1205 
1206 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1207 		struct virtio_net_ctrl_mac *tbl;
1208 
1209 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1210 			continue;
1211 
1212 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1213 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1214 			RTE_ETHER_ADDR_LEN);
1215 	}
1216 
1217 	virtio_mac_table_set(hw, uc, mc);
1218 }
1219 
1220 static int
1221 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1222 {
1223 	struct virtio_hw *hw = dev->data->dev_private;
1224 
1225 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1226 
1227 	/* Use atomic update if available */
1228 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1229 		struct virtio_pmd_ctrl ctrl;
1230 		int len = RTE_ETHER_ADDR_LEN;
1231 
1232 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1233 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1234 
1235 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1236 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1237 	}
1238 
1239 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1240 		return -ENOTSUP;
1241 
1242 	virtio_set_hwaddr(hw);
1243 	return 0;
1244 }
1245 
1246 static int
1247 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1248 {
1249 	struct virtio_hw *hw = dev->data->dev_private;
1250 	struct virtio_pmd_ctrl ctrl;
1251 	int len;
1252 
1253 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1254 		return -ENOTSUP;
1255 
1256 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1257 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1258 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1259 	len = sizeof(vlan_id);
1260 
1261 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1262 }
1263 
1264 static int
1265 virtio_intr_unmask(struct rte_eth_dev *dev)
1266 {
1267 	struct virtio_hw *hw = dev->data->dev_private;
1268 
1269 	if (rte_intr_ack(dev->intr_handle) < 0)
1270 		return -1;
1271 
1272 	if (!hw->virtio_user_dev)
1273 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1274 
1275 	return 0;
1276 }
1277 
1278 static int
1279 virtio_intr_enable(struct rte_eth_dev *dev)
1280 {
1281 	struct virtio_hw *hw = dev->data->dev_private;
1282 
1283 	if (rte_intr_enable(dev->intr_handle) < 0)
1284 		return -1;
1285 
1286 	if (!hw->virtio_user_dev)
1287 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1288 
1289 	return 0;
1290 }
1291 
1292 static int
1293 virtio_intr_disable(struct rte_eth_dev *dev)
1294 {
1295 	struct virtio_hw *hw = dev->data->dev_private;
1296 
1297 	if (rte_intr_disable(dev->intr_handle) < 0)
1298 		return -1;
1299 
1300 	if (!hw->virtio_user_dev)
1301 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1302 
1303 	return 0;
1304 }
1305 
1306 static int
1307 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1308 {
1309 	uint64_t host_features;
1310 
1311 	/* Prepare guest_features: feature that driver wants to support */
1312 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1313 		req_features);
1314 
1315 	/* Read device(host) feature bits */
1316 	host_features = VTPCI_OPS(hw)->get_features(hw);
1317 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1318 		host_features);
1319 
1320 	/* If supported, ensure MTU value is valid before acknowledging it. */
1321 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1322 		struct virtio_net_config config;
1323 
1324 		vtpci_read_dev_config(hw,
1325 			offsetof(struct virtio_net_config, mtu),
1326 			&config.mtu, sizeof(config.mtu));
1327 
1328 		if (config.mtu < RTE_ETHER_MIN_MTU)
1329 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1330 	}
1331 
1332 	/*
1333 	 * Negotiate features: Subset of device feature bits are written back
1334 	 * guest feature bits.
1335 	 */
1336 	hw->guest_features = req_features;
1337 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1338 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1339 		hw->guest_features);
1340 
1341 	if (hw->modern) {
1342 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1343 			PMD_INIT_LOG(ERR,
1344 				"VIRTIO_F_VERSION_1 features is not enabled.");
1345 			return -1;
1346 		}
1347 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1348 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1349 			PMD_INIT_LOG(ERR,
1350 				"failed to set FEATURES_OK status!");
1351 			return -1;
1352 		}
1353 	}
1354 
1355 	hw->req_guest_features = req_features;
1356 
1357 	return 0;
1358 }
1359 
1360 int
1361 virtio_dev_pause(struct rte_eth_dev *dev)
1362 {
1363 	struct virtio_hw *hw = dev->data->dev_private;
1364 
1365 	rte_spinlock_lock(&hw->state_lock);
1366 
1367 	if (hw->started == 0) {
1368 		/* Device is just stopped. */
1369 		rte_spinlock_unlock(&hw->state_lock);
1370 		return -1;
1371 	}
1372 	hw->started = 0;
1373 	/*
1374 	 * Prevent the worker threads from touching queues to avoid contention,
1375 	 * 1 ms should be enough for the ongoing Tx function to finish.
1376 	 */
1377 	rte_delay_ms(1);
1378 	return 0;
1379 }
1380 
1381 /*
1382  * Recover hw state to let the worker threads continue.
1383  */
1384 void
1385 virtio_dev_resume(struct rte_eth_dev *dev)
1386 {
1387 	struct virtio_hw *hw = dev->data->dev_private;
1388 
1389 	hw->started = 1;
1390 	rte_spinlock_unlock(&hw->state_lock);
1391 }
1392 
1393 /*
1394  * Should be called only after device is paused.
1395  */
1396 int
1397 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1398 		int nb_pkts)
1399 {
1400 	struct virtio_hw *hw = dev->data->dev_private;
1401 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1402 	int ret;
1403 
1404 	hw->inject_pkts = tx_pkts;
1405 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1406 	hw->inject_pkts = NULL;
1407 
1408 	return ret;
1409 }
1410 
1411 static void
1412 virtio_notify_peers(struct rte_eth_dev *dev)
1413 {
1414 	struct virtio_hw *hw = dev->data->dev_private;
1415 	struct virtnet_rx *rxvq;
1416 	struct rte_mbuf *rarp_mbuf;
1417 
1418 	if (!dev->data->rx_queues)
1419 		return;
1420 
1421 	rxvq = dev->data->rx_queues[0];
1422 	if (!rxvq)
1423 		return;
1424 
1425 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1426 			(struct rte_ether_addr *)hw->mac_addr);
1427 	if (rarp_mbuf == NULL) {
1428 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1429 		return;
1430 	}
1431 
1432 	/* If virtio port just stopped, no need to send RARP */
1433 	if (virtio_dev_pause(dev) < 0) {
1434 		rte_pktmbuf_free(rarp_mbuf);
1435 		return;
1436 	}
1437 
1438 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1439 	virtio_dev_resume(dev);
1440 }
1441 
1442 static void
1443 virtio_ack_link_announce(struct rte_eth_dev *dev)
1444 {
1445 	struct virtio_hw *hw = dev->data->dev_private;
1446 	struct virtio_pmd_ctrl ctrl;
1447 
1448 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1449 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1450 
1451 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1452 }
1453 
1454 /*
1455  * Process virtio config changed interrupt. Call the callback
1456  * if link state changed, generate gratuitous RARP packet if
1457  * the status indicates an ANNOUNCE.
1458  */
1459 void
1460 virtio_interrupt_handler(void *param)
1461 {
1462 	struct rte_eth_dev *dev = param;
1463 	struct virtio_hw *hw = dev->data->dev_private;
1464 	uint8_t isr;
1465 	uint16_t status;
1466 
1467 	/* Read interrupt status which clears interrupt */
1468 	isr = vtpci_isr(hw);
1469 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1470 
1471 	if (virtio_intr_unmask(dev) < 0)
1472 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1473 
1474 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1475 		if (virtio_dev_link_update(dev, 0) == 0)
1476 			_rte_eth_dev_callback_process(dev,
1477 						      RTE_ETH_EVENT_INTR_LSC,
1478 						      NULL);
1479 
1480 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1481 			vtpci_read_dev_config(hw,
1482 				offsetof(struct virtio_net_config, status),
1483 				&status, sizeof(status));
1484 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1485 				virtio_notify_peers(dev);
1486 				if (hw->cvq)
1487 					virtio_ack_link_announce(dev);
1488 			}
1489 		}
1490 	}
1491 }
1492 
1493 /* set rx and tx handlers according to what is supported */
1494 static void
1495 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1496 {
1497 	struct virtio_hw *hw = eth_dev->data->dev_private;
1498 
1499 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1500 	if (vtpci_packed_queue(hw)) {
1501 		PMD_INIT_LOG(INFO,
1502 			"virtio: using packed ring %s Tx path on port %u",
1503 			hw->use_inorder_tx ? "inorder" : "standard",
1504 			eth_dev->data->port_id);
1505 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1506 	} else {
1507 		if (hw->use_inorder_tx) {
1508 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1509 				eth_dev->data->port_id);
1510 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1511 		} else {
1512 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1513 				eth_dev->data->port_id);
1514 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1515 		}
1516 	}
1517 
1518 	if (vtpci_packed_queue(hw)) {
1519 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1520 			PMD_INIT_LOG(INFO,
1521 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1522 				eth_dev->data->port_id);
1523 			eth_dev->rx_pkt_burst =
1524 				&virtio_recv_mergeable_pkts_packed;
1525 		} else {
1526 			PMD_INIT_LOG(INFO,
1527 				"virtio: using packed ring standard Rx path on port %u",
1528 				eth_dev->data->port_id);
1529 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1530 		}
1531 	} else {
1532 		if (hw->use_simple_rx) {
1533 			PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1534 				eth_dev->data->port_id);
1535 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1536 		} else if (hw->use_inorder_rx) {
1537 			PMD_INIT_LOG(INFO,
1538 				"virtio: using inorder Rx path on port %u",
1539 				eth_dev->data->port_id);
1540 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1541 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1542 			PMD_INIT_LOG(INFO,
1543 				"virtio: using mergeable buffer Rx path on port %u",
1544 				eth_dev->data->port_id);
1545 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1546 		} else {
1547 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1548 				eth_dev->data->port_id);
1549 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1550 		}
1551 	}
1552 
1553 }
1554 
1555 /* Only support 1:1 queue/interrupt mapping so far.
1556  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1557  * interrupt vectors (<N+1).
1558  */
1559 static int
1560 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1561 {
1562 	uint32_t i;
1563 	struct virtio_hw *hw = dev->data->dev_private;
1564 
1565 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1566 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1567 		dev->intr_handle->intr_vec[i] = i + 1;
1568 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1569 						 VIRTIO_MSI_NO_VECTOR) {
1570 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1571 			return -EBUSY;
1572 		}
1573 	}
1574 
1575 	return 0;
1576 }
1577 
1578 static void
1579 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1580 {
1581 	uint32_t i;
1582 	struct virtio_hw *hw = dev->data->dev_private;
1583 
1584 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1585 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1586 		VTPCI_OPS(hw)->set_queue_irq(hw,
1587 					     hw->vqs[i * VTNET_CQ],
1588 					     VIRTIO_MSI_NO_VECTOR);
1589 }
1590 
1591 static int
1592 virtio_configure_intr(struct rte_eth_dev *dev)
1593 {
1594 	struct virtio_hw *hw = dev->data->dev_private;
1595 
1596 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1597 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1598 		return -ENOTSUP;
1599 	}
1600 
1601 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1602 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1603 		return -1;
1604 	}
1605 
1606 	if (!dev->intr_handle->intr_vec) {
1607 		dev->intr_handle->intr_vec =
1608 			rte_zmalloc("intr_vec",
1609 				    hw->max_queue_pairs * sizeof(int), 0);
1610 		if (!dev->intr_handle->intr_vec) {
1611 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1612 				     hw->max_queue_pairs);
1613 			return -ENOMEM;
1614 		}
1615 	}
1616 
1617 	/* Re-register callback to update max_intr */
1618 	rte_intr_callback_unregister(dev->intr_handle,
1619 				     virtio_interrupt_handler,
1620 				     dev);
1621 	rte_intr_callback_register(dev->intr_handle,
1622 				   virtio_interrupt_handler,
1623 				   dev);
1624 
1625 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1626 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1627 	 * And for legacy devices, this should be done before queue/vec binding
1628 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1629 	 * (22) will be ignored.
1630 	 */
1631 	if (virtio_intr_enable(dev) < 0) {
1632 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1633 		return -1;
1634 	}
1635 
1636 	if (virtio_queues_bind_intr(dev) < 0) {
1637 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1638 		return -1;
1639 	}
1640 
1641 	return 0;
1642 }
1643 
1644 /* reset device and renegotiate features if needed */
1645 static int
1646 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1647 {
1648 	struct virtio_hw *hw = eth_dev->data->dev_private;
1649 	struct virtio_net_config *config;
1650 	struct virtio_net_config local_config;
1651 	struct rte_pci_device *pci_dev = NULL;
1652 	int ret;
1653 
1654 	/* Reset the device although not necessary at startup */
1655 	vtpci_reset(hw);
1656 
1657 	if (hw->vqs) {
1658 		virtio_dev_free_mbufs(eth_dev);
1659 		virtio_free_queues(hw);
1660 	}
1661 
1662 	/* Tell the host we've noticed this device. */
1663 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1664 
1665 	/* Tell the host we've known how to drive the device. */
1666 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1667 	if (virtio_negotiate_features(hw, req_features) < 0)
1668 		return -1;
1669 
1670 	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1671 
1672 	if (!hw->virtio_user_dev)
1673 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1674 
1675 	/* If host does not support both status and MSI-X then disable LSC */
1676 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1677 	    hw->use_msix != VIRTIO_MSIX_NONE)
1678 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1679 	else
1680 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1681 
1682 	/* Setting up rx_header size for the device */
1683 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1684 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1685 	    vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1686 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1687 	else
1688 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1689 
1690 	/* Copy the permanent MAC address to: virtio_hw */
1691 	virtio_get_hwaddr(hw);
1692 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1693 			&eth_dev->data->mac_addrs[0]);
1694 	PMD_INIT_LOG(DEBUG,
1695 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1696 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1697 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1698 
1699 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1700 		config = &local_config;
1701 
1702 		vtpci_read_dev_config(hw,
1703 			offsetof(struct virtio_net_config, mac),
1704 			&config->mac, sizeof(config->mac));
1705 
1706 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1707 			vtpci_read_dev_config(hw,
1708 				offsetof(struct virtio_net_config, status),
1709 				&config->status, sizeof(config->status));
1710 		} else {
1711 			PMD_INIT_LOG(DEBUG,
1712 				     "VIRTIO_NET_F_STATUS is not supported");
1713 			config->status = 0;
1714 		}
1715 
1716 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1717 			vtpci_read_dev_config(hw,
1718 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1719 				&config->max_virtqueue_pairs,
1720 				sizeof(config->max_virtqueue_pairs));
1721 		} else {
1722 			PMD_INIT_LOG(DEBUG,
1723 				     "VIRTIO_NET_F_MQ is not supported");
1724 			config->max_virtqueue_pairs = 1;
1725 		}
1726 
1727 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1728 
1729 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1730 			vtpci_read_dev_config(hw,
1731 				offsetof(struct virtio_net_config, mtu),
1732 				&config->mtu,
1733 				sizeof(config->mtu));
1734 
1735 			/*
1736 			 * MTU value has already been checked at negotiation
1737 			 * time, but check again in case it has changed since
1738 			 * then, which should not happen.
1739 			 */
1740 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1741 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1742 						config->mtu);
1743 				return -1;
1744 			}
1745 
1746 			hw->max_mtu = config->mtu;
1747 			/* Set initial MTU to maximum one supported by vhost */
1748 			eth_dev->data->mtu = config->mtu;
1749 
1750 		} else {
1751 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1752 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1753 		}
1754 
1755 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1756 				config->max_virtqueue_pairs);
1757 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1758 		PMD_INIT_LOG(DEBUG,
1759 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1760 				config->mac[0], config->mac[1],
1761 				config->mac[2], config->mac[3],
1762 				config->mac[4], config->mac[5]);
1763 	} else {
1764 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1765 		hw->max_queue_pairs = 1;
1766 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1767 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1768 	}
1769 
1770 	ret = virtio_alloc_queues(eth_dev);
1771 	if (ret < 0)
1772 		return ret;
1773 
1774 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1775 		if (virtio_configure_intr(eth_dev) < 0) {
1776 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1777 			virtio_free_queues(hw);
1778 			return -1;
1779 		}
1780 	}
1781 
1782 	vtpci_reinit_complete(hw);
1783 
1784 	if (pci_dev)
1785 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1786 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1787 			pci_dev->id.device_id);
1788 
1789 	return 0;
1790 }
1791 
1792 /*
1793  * Remap the PCI device again (IO port map for legacy device and
1794  * memory map for modern device), so that the secondary process
1795  * could have the PCI initiated correctly.
1796  */
1797 static int
1798 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1799 {
1800 	if (hw->modern) {
1801 		/*
1802 		 * We don't have to re-parse the PCI config space, since
1803 		 * rte_pci_map_device() makes sure the mapped address
1804 		 * in secondary process would equal to the one mapped in
1805 		 * the primary process: error will be returned if that
1806 		 * requirement is not met.
1807 		 *
1808 		 * That said, we could simply reuse all cap pointers
1809 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1810 		 * primary process, which is stored in shared memory.
1811 		 */
1812 		if (rte_pci_map_device(pci_dev)) {
1813 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1814 			return -1;
1815 		}
1816 	} else {
1817 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1818 			return -1;
1819 	}
1820 
1821 	return 0;
1822 }
1823 
1824 static void
1825 virtio_set_vtpci_ops(struct virtio_hw *hw)
1826 {
1827 #ifdef RTE_VIRTIO_USER
1828 	if (hw->virtio_user_dev)
1829 		VTPCI_OPS(hw) = &virtio_user_ops;
1830 	else
1831 #endif
1832 	if (hw->modern)
1833 		VTPCI_OPS(hw) = &modern_ops;
1834 	else
1835 		VTPCI_OPS(hw) = &legacy_ops;
1836 }
1837 
1838 /*
1839  * This function is based on probe() function in virtio_pci.c
1840  * It returns 0 on success.
1841  */
1842 int
1843 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1844 {
1845 	struct virtio_hw *hw = eth_dev->data->dev_private;
1846 	int ret;
1847 
1848 	if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1849 		PMD_INIT_LOG(ERR,
1850 			"Not sufficient headroom required = %d, avail = %d",
1851 			(int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1852 			RTE_PKTMBUF_HEADROOM);
1853 
1854 		return -1;
1855 	}
1856 
1857 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1858 
1859 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1860 		if (!hw->virtio_user_dev) {
1861 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1862 			if (ret)
1863 				return ret;
1864 		}
1865 
1866 		virtio_set_vtpci_ops(hw);
1867 		set_rxtx_funcs(eth_dev);
1868 
1869 		return 0;
1870 	}
1871 
1872 	/*
1873 	 * Pass the information to the rte_eth_dev_close() that it should also
1874 	 * release the private port resources.
1875 	 */
1876 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1877 
1878 	/* Allocate memory for storing MAC addresses */
1879 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1880 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1881 	if (eth_dev->data->mac_addrs == NULL) {
1882 		PMD_INIT_LOG(ERR,
1883 			"Failed to allocate %d bytes needed to store MAC addresses",
1884 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1885 		return -ENOMEM;
1886 	}
1887 
1888 	hw->port_id = eth_dev->data->port_id;
1889 	/* For virtio_user case the hw->virtio_user_dev is populated by
1890 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1891 	 */
1892 	if (!hw->virtio_user_dev) {
1893 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1894 		if (ret)
1895 			goto err_vtpci_init;
1896 	}
1897 
1898 	/* reset device and negotiate default features */
1899 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1900 	if (ret < 0)
1901 		goto err_virtio_init;
1902 
1903 	hw->opened = true;
1904 
1905 	return 0;
1906 
1907 err_virtio_init:
1908 	if (!hw->virtio_user_dev) {
1909 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1910 		if (!hw->modern)
1911 			rte_pci_ioport_unmap(VTPCI_IO(hw));
1912 	}
1913 err_vtpci_init:
1914 	rte_free(eth_dev->data->mac_addrs);
1915 	eth_dev->data->mac_addrs = NULL;
1916 	return ret;
1917 }
1918 
1919 static int
1920 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1921 {
1922 	PMD_INIT_FUNC_TRACE();
1923 
1924 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1925 		return 0;
1926 
1927 	virtio_dev_stop(eth_dev);
1928 	virtio_dev_close(eth_dev);
1929 
1930 	eth_dev->dev_ops = NULL;
1931 	eth_dev->tx_pkt_burst = NULL;
1932 	eth_dev->rx_pkt_burst = NULL;
1933 
1934 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1935 
1936 	return 0;
1937 }
1938 
1939 static int vdpa_check_handler(__rte_unused const char *key,
1940 		const char *value, __rte_unused void *opaque)
1941 {
1942 	if (strcmp(value, "1"))
1943 		return -1;
1944 
1945 	return 0;
1946 }
1947 
1948 static int
1949 vdpa_mode_selected(struct rte_devargs *devargs)
1950 {
1951 	struct rte_kvargs *kvlist;
1952 	const char *key = "vdpa";
1953 	int ret = 0;
1954 
1955 	if (devargs == NULL)
1956 		return 0;
1957 
1958 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1959 	if (kvlist == NULL)
1960 		return 0;
1961 
1962 	if (!rte_kvargs_count(kvlist, key))
1963 		goto exit;
1964 
1965 	/* vdpa mode selected when there's a key-value pair: vdpa=1 */
1966 	if (rte_kvargs_process(kvlist, key,
1967 				vdpa_check_handler, NULL) < 0) {
1968 		goto exit;
1969 	}
1970 	ret = 1;
1971 
1972 exit:
1973 	rte_kvargs_free(kvlist);
1974 	return ret;
1975 }
1976 
1977 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1978 	struct rte_pci_device *pci_dev)
1979 {
1980 	if (rte_eal_iopl_init() != 0) {
1981 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1982 		return 1;
1983 	}
1984 
1985 	/* virtio pmd skips probe if device needs to work in vdpa mode */
1986 	if (vdpa_mode_selected(pci_dev->device.devargs))
1987 		return 1;
1988 
1989 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1990 		eth_virtio_dev_init);
1991 }
1992 
1993 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1994 {
1995 	int ret;
1996 
1997 	ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1998 	/* Port has already been released by close. */
1999 	if (ret == -ENODEV)
2000 		ret = 0;
2001 	return ret;
2002 }
2003 
2004 static struct rte_pci_driver rte_virtio_pmd = {
2005 	.driver = {
2006 		.name = "net_virtio",
2007 	},
2008 	.id_table = pci_id_virtio_map,
2009 	.drv_flags = 0,
2010 	.probe = eth_virtio_pci_probe,
2011 	.remove = eth_virtio_pci_remove,
2012 };
2013 
2014 RTE_INIT(rte_virtio_pmd_init)
2015 {
2016 	rte_eal_iopl_init();
2017 	rte_pci_register(&rte_virtio_pmd);
2018 }
2019 
2020 static bool
2021 rx_offload_enabled(struct virtio_hw *hw)
2022 {
2023 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2024 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2025 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2026 }
2027 
2028 static bool
2029 tx_offload_enabled(struct virtio_hw *hw)
2030 {
2031 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2032 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2033 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2034 }
2035 
2036 /*
2037  * Configure virtio device
2038  * It returns 0 on success.
2039  */
2040 static int
2041 virtio_dev_configure(struct rte_eth_dev *dev)
2042 {
2043 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2044 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2045 	struct virtio_hw *hw = dev->data->dev_private;
2046 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2047 		hw->vtnet_hdr_size;
2048 	uint64_t rx_offloads = rxmode->offloads;
2049 	uint64_t tx_offloads = txmode->offloads;
2050 	uint64_t req_features;
2051 	int ret;
2052 
2053 	PMD_INIT_LOG(DEBUG, "configure");
2054 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2055 
2056 	if (dev->data->dev_conf.intr_conf.rxq) {
2057 		ret = virtio_init_device(dev, hw->req_guest_features);
2058 		if (ret < 0)
2059 			return ret;
2060 	}
2061 
2062 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2063 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2064 
2065 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2066 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2067 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2068 
2069 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2070 		req_features |=
2071 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2072 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2073 
2074 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2075 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2076 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2077 
2078 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2079 		req_features |=
2080 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2081 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2082 
2083 	/* if request features changed, reinit the device */
2084 	if (req_features != hw->req_guest_features) {
2085 		ret = virtio_init_device(dev, req_features);
2086 		if (ret < 0)
2087 			return ret;
2088 	}
2089 
2090 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2091 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2092 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2093 		PMD_DRV_LOG(ERR,
2094 			"rx checksum not available on this host");
2095 		return -ENOTSUP;
2096 	}
2097 
2098 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2099 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2100 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2101 		PMD_DRV_LOG(ERR,
2102 			"Large Receive Offload not available on this host");
2103 		return -ENOTSUP;
2104 	}
2105 
2106 	/* start control queue */
2107 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2108 		virtio_dev_cq_start(dev);
2109 
2110 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2111 		hw->vlan_strip = 1;
2112 
2113 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2114 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2115 		PMD_DRV_LOG(ERR,
2116 			    "vlan filtering not available on this host");
2117 		return -ENOTSUP;
2118 	}
2119 
2120 	hw->has_tx_offload = tx_offload_enabled(hw);
2121 	hw->has_rx_offload = rx_offload_enabled(hw);
2122 
2123 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2124 		/* Enable vector (0) for Link State Intrerrupt */
2125 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2126 				VIRTIO_MSI_NO_VECTOR) {
2127 			PMD_DRV_LOG(ERR, "failed to set config vector");
2128 			return -EBUSY;
2129 		}
2130 
2131 	rte_spinlock_init(&hw->state_lock);
2132 
2133 	hw->use_simple_rx = 1;
2134 
2135 	if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2136 		hw->use_inorder_tx = 1;
2137 		hw->use_inorder_rx = 1;
2138 		hw->use_simple_rx = 0;
2139 	}
2140 
2141 	if (vtpci_packed_queue(hw)) {
2142 		hw->use_simple_rx = 0;
2143 		hw->use_inorder_rx = 0;
2144 	}
2145 
2146 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2147 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2148 		hw->use_simple_rx = 0;
2149 	}
2150 #endif
2151 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2152 		 hw->use_simple_rx = 0;
2153 	}
2154 
2155 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2156 			   DEV_RX_OFFLOAD_TCP_CKSUM |
2157 			   DEV_RX_OFFLOAD_TCP_LRO |
2158 			   DEV_RX_OFFLOAD_VLAN_STRIP))
2159 		hw->use_simple_rx = 0;
2160 
2161 	return 0;
2162 }
2163 
2164 
2165 static int
2166 virtio_dev_start(struct rte_eth_dev *dev)
2167 {
2168 	uint16_t nb_queues, i;
2169 	struct virtnet_rx *rxvq;
2170 	struct virtnet_tx *txvq __rte_unused;
2171 	struct virtio_hw *hw = dev->data->dev_private;
2172 	int ret;
2173 
2174 	/* Finish the initialization of the queues */
2175 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2176 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2177 		if (ret < 0)
2178 			return ret;
2179 	}
2180 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2181 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2182 		if (ret < 0)
2183 			return ret;
2184 	}
2185 
2186 	/* check if lsc interrupt feature is enabled */
2187 	if (dev->data->dev_conf.intr_conf.lsc) {
2188 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2189 			PMD_DRV_LOG(ERR, "link status not supported by host");
2190 			return -ENOTSUP;
2191 		}
2192 	}
2193 
2194 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2195 	 * in device configure, but it could be unmapped  when device is
2196 	 * stopped.
2197 	 */
2198 	if (dev->data->dev_conf.intr_conf.lsc ||
2199 	    dev->data->dev_conf.intr_conf.rxq) {
2200 		virtio_intr_disable(dev);
2201 
2202 		/* Setup interrupt callback  */
2203 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2204 			rte_intr_callback_register(dev->intr_handle,
2205 						   virtio_interrupt_handler,
2206 						   dev);
2207 
2208 		if (virtio_intr_enable(dev) < 0) {
2209 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2210 			return -EIO;
2211 		}
2212 	}
2213 
2214 	/*Notify the backend
2215 	 *Otherwise the tap backend might already stop its queue due to fullness.
2216 	 *vhost backend will have no chance to be waked up
2217 	 */
2218 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2219 	if (hw->max_queue_pairs > 1) {
2220 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2221 			return -EINVAL;
2222 	}
2223 
2224 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2225 
2226 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2227 		rxvq = dev->data->rx_queues[i];
2228 		/* Flush the old packets */
2229 		virtqueue_rxvq_flush(rxvq->vq);
2230 		virtqueue_notify(rxvq->vq);
2231 	}
2232 
2233 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2234 		txvq = dev->data->tx_queues[i];
2235 		virtqueue_notify(txvq->vq);
2236 	}
2237 
2238 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2239 
2240 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2241 		rxvq = dev->data->rx_queues[i];
2242 		VIRTQUEUE_DUMP(rxvq->vq);
2243 	}
2244 
2245 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2246 		txvq = dev->data->tx_queues[i];
2247 		VIRTQUEUE_DUMP(txvq->vq);
2248 	}
2249 
2250 	set_rxtx_funcs(dev);
2251 	hw->started = true;
2252 
2253 	/* Initialize Link state */
2254 	virtio_dev_link_update(dev, 0);
2255 
2256 	return 0;
2257 }
2258 
2259 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2260 {
2261 	struct virtio_hw *hw = dev->data->dev_private;
2262 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2263 	const char *type __rte_unused;
2264 	unsigned int i, mbuf_num = 0;
2265 	struct virtqueue *vq;
2266 	struct rte_mbuf *buf;
2267 	int queue_type;
2268 
2269 	if (hw->vqs == NULL)
2270 		return;
2271 
2272 	for (i = 0; i < nr_vq; i++) {
2273 		vq = hw->vqs[i];
2274 		if (!vq)
2275 			continue;
2276 
2277 		queue_type = virtio_get_queue_type(hw, i);
2278 		if (queue_type == VTNET_RQ)
2279 			type = "rxq";
2280 		else if (queue_type == VTNET_TQ)
2281 			type = "txq";
2282 		else
2283 			continue;
2284 
2285 		PMD_INIT_LOG(DEBUG,
2286 			"Before freeing %s[%d] used and unused buf",
2287 			type, i);
2288 		VIRTQUEUE_DUMP(vq);
2289 
2290 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2291 			rte_pktmbuf_free(buf);
2292 			mbuf_num++;
2293 		}
2294 
2295 		PMD_INIT_LOG(DEBUG,
2296 			"After freeing %s[%d] used and unused buf",
2297 			type, i);
2298 		VIRTQUEUE_DUMP(vq);
2299 	}
2300 
2301 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2302 }
2303 
2304 /*
2305  * Stop device: disable interrupt and mark link down
2306  */
2307 static void
2308 virtio_dev_stop(struct rte_eth_dev *dev)
2309 {
2310 	struct virtio_hw *hw = dev->data->dev_private;
2311 	struct rte_eth_link link;
2312 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2313 
2314 	PMD_INIT_LOG(DEBUG, "stop");
2315 
2316 	rte_spinlock_lock(&hw->state_lock);
2317 	if (!hw->started)
2318 		goto out_unlock;
2319 	hw->started = false;
2320 
2321 	if (intr_conf->lsc || intr_conf->rxq) {
2322 		virtio_intr_disable(dev);
2323 
2324 		/* Reset interrupt callback  */
2325 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2326 			rte_intr_callback_unregister(dev->intr_handle,
2327 						     virtio_interrupt_handler,
2328 						     dev);
2329 		}
2330 	}
2331 
2332 	memset(&link, 0, sizeof(link));
2333 	rte_eth_linkstatus_set(dev, &link);
2334 out_unlock:
2335 	rte_spinlock_unlock(&hw->state_lock);
2336 }
2337 
2338 static int
2339 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2340 {
2341 	struct rte_eth_link link;
2342 	uint16_t status;
2343 	struct virtio_hw *hw = dev->data->dev_private;
2344 
2345 	memset(&link, 0, sizeof(link));
2346 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
2347 	link.link_speed  = ETH_SPEED_NUM_10G;
2348 	link.link_autoneg = ETH_LINK_FIXED;
2349 
2350 	if (!hw->started) {
2351 		link.link_status = ETH_LINK_DOWN;
2352 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2353 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2354 		vtpci_read_dev_config(hw,
2355 				offsetof(struct virtio_net_config, status),
2356 				&status, sizeof(status));
2357 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2358 			link.link_status = ETH_LINK_DOWN;
2359 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2360 				     dev->data->port_id);
2361 		} else {
2362 			link.link_status = ETH_LINK_UP;
2363 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2364 				     dev->data->port_id);
2365 		}
2366 	} else {
2367 		link.link_status = ETH_LINK_UP;
2368 	}
2369 
2370 	return rte_eth_linkstatus_set(dev, &link);
2371 }
2372 
2373 static int
2374 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2375 {
2376 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2377 	struct virtio_hw *hw = dev->data->dev_private;
2378 	uint64_t offloads = rxmode->offloads;
2379 
2380 	if (mask & ETH_VLAN_FILTER_MASK) {
2381 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2382 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2383 
2384 			PMD_DRV_LOG(NOTICE,
2385 				"vlan filtering not available on this host");
2386 
2387 			return -ENOTSUP;
2388 		}
2389 	}
2390 
2391 	if (mask & ETH_VLAN_STRIP_MASK)
2392 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2393 
2394 	return 0;
2395 }
2396 
2397 static void
2398 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2399 {
2400 	uint64_t tso_mask, host_features;
2401 	struct virtio_hw *hw = dev->data->dev_private;
2402 
2403 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2404 
2405 	dev_info->max_rx_queues =
2406 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2407 	dev_info->max_tx_queues =
2408 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2409 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2410 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2411 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2412 
2413 	host_features = VTPCI_OPS(hw)->get_features(hw);
2414 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2415 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2416 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2417 		dev_info->rx_offload_capa |=
2418 			DEV_RX_OFFLOAD_TCP_CKSUM |
2419 			DEV_RX_OFFLOAD_UDP_CKSUM;
2420 	}
2421 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2422 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2423 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2424 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2425 	if ((host_features & tso_mask) == tso_mask)
2426 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2427 
2428 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2429 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2430 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2431 		dev_info->tx_offload_capa |=
2432 			DEV_TX_OFFLOAD_UDP_CKSUM |
2433 			DEV_TX_OFFLOAD_TCP_CKSUM;
2434 	}
2435 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2436 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2437 	if ((host_features & tso_mask) == tso_mask)
2438 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2439 }
2440 
2441 /*
2442  * It enables testpmd to collect per queue stats.
2443  */
2444 static int
2445 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2446 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2447 __rte_unused uint8_t is_rx)
2448 {
2449 	return 0;
2450 }
2451 
2452 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2453 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2454 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2455 
2456 RTE_INIT(virtio_init_log)
2457 {
2458 	virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2459 	if (virtio_logtype_init >= 0)
2460 		rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2461 	virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2462 	if (virtio_logtype_driver >= 0)
2463 		rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2464 }
2465