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 (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
469 		PMD_INIT_LOG(ERR, "split virtqueue size is not power of 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->virtio_user_dev is not NULL),
592 	 * we use 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 			/* first indirect descriptor is always the tx header */
614 			if (!vtpci_packed_queue(hw)) {
615 				struct vring_desc *start_dp = txr[i].tx_indir;
616 				vring_desc_init_split(start_dp,
617 						      RTE_DIM(txr[i].tx_indir));
618 				start_dp->addr = txvq->virtio_net_hdr_mem
619 					+ i * sizeof(*txr)
620 					+ offsetof(struct virtio_tx_region,
621 						   tx_hdr);
622 				start_dp->len = hw->vtnet_hdr_size;
623 				start_dp->flags = VRING_DESC_F_NEXT;
624 			} else {
625 				struct vring_packed_desc *start_dp =
626 					txr[i].tx_packed_indir;
627 				vring_desc_init_indirect_packed(start_dp,
628 				      RTE_DIM(txr[i].tx_packed_indir));
629 				start_dp->addr = txvq->virtio_net_hdr_mem
630 					+ i * sizeof(*txr)
631 					+ offsetof(struct virtio_tx_region,
632 						   tx_hdr);
633 				start_dp->len = hw->vtnet_hdr_size;
634 			}
635 		}
636 	}
637 
638 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
639 		PMD_INIT_LOG(ERR, "setup_queue failed");
640 		return -EINVAL;
641 	}
642 
643 	return 0;
644 
645 fail_q_alloc:
646 	rte_free(sw_ring);
647 	rte_memzone_free(hdr_mz);
648 	rte_memzone_free(mz);
649 	rte_free(vq);
650 
651 	return ret;
652 }
653 
654 static void
655 virtio_free_queues(struct virtio_hw *hw)
656 {
657 	uint16_t nr_vq = virtio_get_nr_vq(hw);
658 	struct virtqueue *vq;
659 	int queue_type;
660 	uint16_t i;
661 
662 	if (hw->vqs == NULL)
663 		return;
664 
665 	for (i = 0; i < nr_vq; i++) {
666 		vq = hw->vqs[i];
667 		if (!vq)
668 			continue;
669 
670 		queue_type = virtio_get_queue_type(hw, i);
671 		if (queue_type == VTNET_RQ) {
672 			rte_free(vq->sw_ring);
673 			rte_memzone_free(vq->rxq.mz);
674 		} else if (queue_type == VTNET_TQ) {
675 			rte_memzone_free(vq->txq.mz);
676 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
677 		} else {
678 			rte_memzone_free(vq->cq.mz);
679 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
680 		}
681 
682 		rte_free(vq);
683 		hw->vqs[i] = NULL;
684 	}
685 
686 	rte_free(hw->vqs);
687 	hw->vqs = NULL;
688 }
689 
690 static int
691 virtio_alloc_queues(struct rte_eth_dev *dev)
692 {
693 	struct virtio_hw *hw = dev->data->dev_private;
694 	uint16_t nr_vq = virtio_get_nr_vq(hw);
695 	uint16_t i;
696 	int ret;
697 
698 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
699 	if (!hw->vqs) {
700 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
701 		return -ENOMEM;
702 	}
703 
704 	for (i = 0; i < nr_vq; i++) {
705 		ret = virtio_init_queue(dev, i);
706 		if (ret < 0) {
707 			virtio_free_queues(hw);
708 			return ret;
709 		}
710 	}
711 
712 	return 0;
713 }
714 
715 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
716 
717 static void
718 virtio_dev_close(struct rte_eth_dev *dev)
719 {
720 	struct virtio_hw *hw = dev->data->dev_private;
721 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
722 
723 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
724 
725 	if (!hw->opened)
726 		return;
727 	hw->opened = false;
728 
729 	/* reset the NIC */
730 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
731 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
732 	if (intr_conf->rxq)
733 		virtio_queues_unbind_intr(dev);
734 
735 	if (intr_conf->lsc || intr_conf->rxq) {
736 		virtio_intr_disable(dev);
737 		rte_intr_efd_disable(dev->intr_handle);
738 		rte_free(dev->intr_handle->intr_vec);
739 		dev->intr_handle->intr_vec = NULL;
740 	}
741 
742 	vtpci_reset(hw);
743 	virtio_dev_free_mbufs(dev);
744 	virtio_free_queues(hw);
745 
746 #ifdef RTE_VIRTIO_USER
747 	if (hw->virtio_user_dev)
748 		virtio_user_dev_uninit(hw->virtio_user_dev);
749 	else
750 #endif
751 	if (dev->device) {
752 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
753 		if (!hw->modern)
754 			rte_pci_ioport_unmap(VTPCI_IO(hw));
755 	}
756 }
757 
758 static int
759 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
760 {
761 	struct virtio_hw *hw = dev->data->dev_private;
762 	struct virtio_pmd_ctrl ctrl;
763 	int dlen[1];
764 	int ret;
765 
766 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
767 		PMD_INIT_LOG(INFO, "host does not support rx control");
768 		return -ENOTSUP;
769 	}
770 
771 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
772 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
773 	ctrl.data[0] = 1;
774 	dlen[0] = 1;
775 
776 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
777 	if (ret) {
778 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
779 		return -EAGAIN;
780 	}
781 
782 	return 0;
783 }
784 
785 static int
786 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
787 {
788 	struct virtio_hw *hw = dev->data->dev_private;
789 	struct virtio_pmd_ctrl ctrl;
790 	int dlen[1];
791 	int ret;
792 
793 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
794 		PMD_INIT_LOG(INFO, "host does not support rx control");
795 		return -ENOTSUP;
796 	}
797 
798 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
799 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
800 	ctrl.data[0] = 0;
801 	dlen[0] = 1;
802 
803 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
804 	if (ret) {
805 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
806 		return -EAGAIN;
807 	}
808 
809 	return 0;
810 }
811 
812 static int
813 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
814 {
815 	struct virtio_hw *hw = dev->data->dev_private;
816 	struct virtio_pmd_ctrl ctrl;
817 	int dlen[1];
818 	int ret;
819 
820 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
821 		PMD_INIT_LOG(INFO, "host does not support rx control");
822 		return -ENOTSUP;
823 	}
824 
825 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
826 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
827 	ctrl.data[0] = 1;
828 	dlen[0] = 1;
829 
830 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
831 	if (ret) {
832 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
833 		return -EAGAIN;
834 	}
835 
836 	return 0;
837 }
838 
839 static int
840 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
841 {
842 	struct virtio_hw *hw = dev->data->dev_private;
843 	struct virtio_pmd_ctrl ctrl;
844 	int dlen[1];
845 	int ret;
846 
847 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
848 		PMD_INIT_LOG(INFO, "host does not support rx control");
849 		return -ENOTSUP;
850 	}
851 
852 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
853 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
854 	ctrl.data[0] = 0;
855 	dlen[0] = 1;
856 
857 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
858 	if (ret) {
859 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
860 		return -EAGAIN;
861 	}
862 
863 	return 0;
864 }
865 
866 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
867 static int
868 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
869 {
870 	struct virtio_hw *hw = dev->data->dev_private;
871 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
872 				 hw->vtnet_hdr_size;
873 	uint32_t frame_size = mtu + ether_hdr_len;
874 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
875 
876 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
877 
878 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
879 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
880 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
881 		return -EINVAL;
882 	}
883 	return 0;
884 }
885 
886 static int
887 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
888 {
889 	struct virtio_hw *hw = dev->data->dev_private;
890 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
891 	struct virtqueue *vq = rxvq->vq;
892 
893 	virtqueue_enable_intr(vq);
894 	virtio_mb(hw->weak_barriers);
895 	return 0;
896 }
897 
898 static int
899 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
900 {
901 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
902 	struct virtqueue *vq = rxvq->vq;
903 
904 	virtqueue_disable_intr(vq);
905 	return 0;
906 }
907 
908 /*
909  * dev_ops for virtio, bare necessities for basic operation
910  */
911 static const struct eth_dev_ops virtio_eth_dev_ops = {
912 	.dev_configure           = virtio_dev_configure,
913 	.dev_start               = virtio_dev_start,
914 	.dev_stop                = virtio_dev_stop,
915 	.dev_close               = virtio_dev_close,
916 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
917 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
918 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
919 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
920 	.mtu_set                 = virtio_mtu_set,
921 	.dev_infos_get           = virtio_dev_info_get,
922 	.stats_get               = virtio_dev_stats_get,
923 	.xstats_get              = virtio_dev_xstats_get,
924 	.xstats_get_names        = virtio_dev_xstats_get_names,
925 	.stats_reset             = virtio_dev_stats_reset,
926 	.xstats_reset            = virtio_dev_stats_reset,
927 	.link_update             = virtio_dev_link_update,
928 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
929 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
930 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
931 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
932 	.rx_queue_release        = virtio_dev_queue_release,
933 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
934 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
935 	.tx_queue_release        = virtio_dev_queue_release,
936 	/* collect stats per queue */
937 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
938 	.vlan_filter_set         = virtio_vlan_filter_set,
939 	.mac_addr_add            = virtio_mac_addr_add,
940 	.mac_addr_remove         = virtio_mac_addr_remove,
941 	.mac_addr_set            = virtio_mac_addr_set,
942 };
943 
944 /*
945  * dev_ops for virtio-user in secondary processes, as we just have
946  * some limited supports currently.
947  */
948 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
949 	.dev_infos_get           = virtio_dev_info_get,
950 	.stats_get               = virtio_dev_stats_get,
951 	.xstats_get              = virtio_dev_xstats_get,
952 	.xstats_get_names        = virtio_dev_xstats_get_names,
953 	.stats_reset             = virtio_dev_stats_reset,
954 	.xstats_reset            = virtio_dev_stats_reset,
955 	/* collect stats per queue */
956 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
957 };
958 
959 static void
960 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
961 {
962 	unsigned i;
963 
964 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
965 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
966 		if (txvq == NULL)
967 			continue;
968 
969 		stats->opackets += txvq->stats.packets;
970 		stats->obytes += txvq->stats.bytes;
971 
972 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
973 			stats->q_opackets[i] = txvq->stats.packets;
974 			stats->q_obytes[i] = txvq->stats.bytes;
975 		}
976 	}
977 
978 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
979 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
980 		if (rxvq == NULL)
981 			continue;
982 
983 		stats->ipackets += rxvq->stats.packets;
984 		stats->ibytes += rxvq->stats.bytes;
985 		stats->ierrors += rxvq->stats.errors;
986 
987 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
988 			stats->q_ipackets[i] = rxvq->stats.packets;
989 			stats->q_ibytes[i] = rxvq->stats.bytes;
990 		}
991 	}
992 
993 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
994 }
995 
996 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
997 				       struct rte_eth_xstat_name *xstats_names,
998 				       __rte_unused unsigned limit)
999 {
1000 	unsigned i;
1001 	unsigned count = 0;
1002 	unsigned t;
1003 
1004 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1005 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1006 
1007 	if (xstats_names != NULL) {
1008 		/* Note: limit checked in rte_eth_xstats_names() */
1009 
1010 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
1011 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1012 			if (rxvq == NULL)
1013 				continue;
1014 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1015 				snprintf(xstats_names[count].name,
1016 					sizeof(xstats_names[count].name),
1017 					"rx_q%u_%s", i,
1018 					rte_virtio_rxq_stat_strings[t].name);
1019 				count++;
1020 			}
1021 		}
1022 
1023 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
1024 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
1025 			if (txvq == NULL)
1026 				continue;
1027 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1028 				snprintf(xstats_names[count].name,
1029 					sizeof(xstats_names[count].name),
1030 					"tx_q%u_%s", i,
1031 					rte_virtio_txq_stat_strings[t].name);
1032 				count++;
1033 			}
1034 		}
1035 		return count;
1036 	}
1037 	return nstats;
1038 }
1039 
1040 static int
1041 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1042 		      unsigned n)
1043 {
1044 	unsigned i;
1045 	unsigned count = 0;
1046 
1047 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1048 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1049 
1050 	if (n < nstats)
1051 		return nstats;
1052 
1053 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1054 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1055 
1056 		if (rxvq == NULL)
1057 			continue;
1058 
1059 		unsigned t;
1060 
1061 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1062 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1063 				rte_virtio_rxq_stat_strings[t].offset);
1064 			xstats[count].id = count;
1065 			count++;
1066 		}
1067 	}
1068 
1069 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1070 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1071 
1072 		if (txvq == NULL)
1073 			continue;
1074 
1075 		unsigned t;
1076 
1077 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1078 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1079 				rte_virtio_txq_stat_strings[t].offset);
1080 			xstats[count].id = count;
1081 			count++;
1082 		}
1083 	}
1084 
1085 	return count;
1086 }
1087 
1088 static int
1089 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1090 {
1091 	virtio_update_stats(dev, stats);
1092 
1093 	return 0;
1094 }
1095 
1096 static int
1097 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1098 {
1099 	unsigned int i;
1100 
1101 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1102 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1103 		if (txvq == NULL)
1104 			continue;
1105 
1106 		txvq->stats.packets = 0;
1107 		txvq->stats.bytes = 0;
1108 		txvq->stats.multicast = 0;
1109 		txvq->stats.broadcast = 0;
1110 		memset(txvq->stats.size_bins, 0,
1111 		       sizeof(txvq->stats.size_bins[0]) * 8);
1112 	}
1113 
1114 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1115 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1116 		if (rxvq == NULL)
1117 			continue;
1118 
1119 		rxvq->stats.packets = 0;
1120 		rxvq->stats.bytes = 0;
1121 		rxvq->stats.errors = 0;
1122 		rxvq->stats.multicast = 0;
1123 		rxvq->stats.broadcast = 0;
1124 		memset(rxvq->stats.size_bins, 0,
1125 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1126 	}
1127 
1128 	return 0;
1129 }
1130 
1131 static void
1132 virtio_set_hwaddr(struct virtio_hw *hw)
1133 {
1134 	vtpci_write_dev_config(hw,
1135 			offsetof(struct virtio_net_config, mac),
1136 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1137 }
1138 
1139 static void
1140 virtio_get_hwaddr(struct virtio_hw *hw)
1141 {
1142 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1143 		vtpci_read_dev_config(hw,
1144 			offsetof(struct virtio_net_config, mac),
1145 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1146 	} else {
1147 		rte_eth_random_addr(&hw->mac_addr[0]);
1148 		virtio_set_hwaddr(hw);
1149 	}
1150 }
1151 
1152 static int
1153 virtio_mac_table_set(struct virtio_hw *hw,
1154 		     const struct virtio_net_ctrl_mac *uc,
1155 		     const struct virtio_net_ctrl_mac *mc)
1156 {
1157 	struct virtio_pmd_ctrl ctrl;
1158 	int err, len[2];
1159 
1160 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1161 		PMD_DRV_LOG(INFO, "host does not support mac table");
1162 		return -1;
1163 	}
1164 
1165 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1166 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1167 
1168 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1169 	memcpy(ctrl.data, uc, len[0]);
1170 
1171 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1172 	memcpy(ctrl.data + len[0], mc, len[1]);
1173 
1174 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1175 	if (err != 0)
1176 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1177 	return err;
1178 }
1179 
1180 static int
1181 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1182 		    uint32_t index, uint32_t vmdq __rte_unused)
1183 {
1184 	struct virtio_hw *hw = dev->data->dev_private;
1185 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1186 	unsigned int i;
1187 	struct virtio_net_ctrl_mac *uc, *mc;
1188 
1189 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1190 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1191 		return -EINVAL;
1192 	}
1193 
1194 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1195 		sizeof(uc->entries));
1196 	uc->entries = 0;
1197 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1198 		sizeof(mc->entries));
1199 	mc->entries = 0;
1200 
1201 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1202 		const struct rte_ether_addr *addr
1203 			= (i == index) ? mac_addr : addrs + i;
1204 		struct virtio_net_ctrl_mac *tbl
1205 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1206 
1207 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1208 	}
1209 
1210 	return virtio_mac_table_set(hw, uc, mc);
1211 }
1212 
1213 static void
1214 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1215 {
1216 	struct virtio_hw *hw = dev->data->dev_private;
1217 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1218 	struct virtio_net_ctrl_mac *uc, *mc;
1219 	unsigned int i;
1220 
1221 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1222 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1223 		return;
1224 	}
1225 
1226 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1227 		sizeof(uc->entries));
1228 	uc->entries = 0;
1229 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1230 		sizeof(mc->entries));
1231 	mc->entries = 0;
1232 
1233 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1234 		struct virtio_net_ctrl_mac *tbl;
1235 
1236 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1237 			continue;
1238 
1239 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1240 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1241 			RTE_ETHER_ADDR_LEN);
1242 	}
1243 
1244 	virtio_mac_table_set(hw, uc, mc);
1245 }
1246 
1247 static int
1248 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1249 {
1250 	struct virtio_hw *hw = dev->data->dev_private;
1251 
1252 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1253 
1254 	/* Use atomic update if available */
1255 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1256 		struct virtio_pmd_ctrl ctrl;
1257 		int len = RTE_ETHER_ADDR_LEN;
1258 
1259 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1260 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1261 
1262 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1263 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1264 	}
1265 
1266 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1267 		return -ENOTSUP;
1268 
1269 	virtio_set_hwaddr(hw);
1270 	return 0;
1271 }
1272 
1273 static int
1274 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1275 {
1276 	struct virtio_hw *hw = dev->data->dev_private;
1277 	struct virtio_pmd_ctrl ctrl;
1278 	int len;
1279 
1280 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1281 		return -ENOTSUP;
1282 
1283 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1284 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1285 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1286 	len = sizeof(vlan_id);
1287 
1288 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1289 }
1290 
1291 static int
1292 virtio_intr_unmask(struct rte_eth_dev *dev)
1293 {
1294 	struct virtio_hw *hw = dev->data->dev_private;
1295 
1296 	if (rte_intr_ack(dev->intr_handle) < 0)
1297 		return -1;
1298 
1299 	if (!hw->virtio_user_dev)
1300 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1301 
1302 	return 0;
1303 }
1304 
1305 static int
1306 virtio_intr_enable(struct rte_eth_dev *dev)
1307 {
1308 	struct virtio_hw *hw = dev->data->dev_private;
1309 
1310 	if (rte_intr_enable(dev->intr_handle) < 0)
1311 		return -1;
1312 
1313 	if (!hw->virtio_user_dev)
1314 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1315 
1316 	return 0;
1317 }
1318 
1319 static int
1320 virtio_intr_disable(struct rte_eth_dev *dev)
1321 {
1322 	struct virtio_hw *hw = dev->data->dev_private;
1323 
1324 	if (rte_intr_disable(dev->intr_handle) < 0)
1325 		return -1;
1326 
1327 	if (!hw->virtio_user_dev)
1328 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1329 
1330 	return 0;
1331 }
1332 
1333 static int
1334 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1335 {
1336 	uint64_t host_features;
1337 
1338 	/* Prepare guest_features: feature that driver wants to support */
1339 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1340 		req_features);
1341 
1342 	/* Read device(host) feature bits */
1343 	host_features = VTPCI_OPS(hw)->get_features(hw);
1344 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1345 		host_features);
1346 
1347 	/* If supported, ensure MTU value is valid before acknowledging it. */
1348 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1349 		struct virtio_net_config config;
1350 
1351 		vtpci_read_dev_config(hw,
1352 			offsetof(struct virtio_net_config, mtu),
1353 			&config.mtu, sizeof(config.mtu));
1354 
1355 		if (config.mtu < RTE_ETHER_MIN_MTU)
1356 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1357 	}
1358 
1359 	/*
1360 	 * Negotiate features: Subset of device feature bits are written back
1361 	 * guest feature bits.
1362 	 */
1363 	hw->guest_features = req_features;
1364 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1365 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1366 		hw->guest_features);
1367 
1368 	if (hw->modern) {
1369 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1370 			PMD_INIT_LOG(ERR,
1371 				"VIRTIO_F_VERSION_1 features is not enabled.");
1372 			return -1;
1373 		}
1374 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1375 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1376 			PMD_INIT_LOG(ERR,
1377 				"failed to set FEATURES_OK status!");
1378 			return -1;
1379 		}
1380 	}
1381 
1382 	hw->req_guest_features = req_features;
1383 
1384 	return 0;
1385 }
1386 
1387 int
1388 virtio_dev_pause(struct rte_eth_dev *dev)
1389 {
1390 	struct virtio_hw *hw = dev->data->dev_private;
1391 
1392 	rte_spinlock_lock(&hw->state_lock);
1393 
1394 	if (hw->started == 0) {
1395 		/* Device is just stopped. */
1396 		rte_spinlock_unlock(&hw->state_lock);
1397 		return -1;
1398 	}
1399 	hw->started = 0;
1400 	/*
1401 	 * Prevent the worker threads from touching queues to avoid contention,
1402 	 * 1 ms should be enough for the ongoing Tx function to finish.
1403 	 */
1404 	rte_delay_ms(1);
1405 	return 0;
1406 }
1407 
1408 /*
1409  * Recover hw state to let the worker threads continue.
1410  */
1411 void
1412 virtio_dev_resume(struct rte_eth_dev *dev)
1413 {
1414 	struct virtio_hw *hw = dev->data->dev_private;
1415 
1416 	hw->started = 1;
1417 	rte_spinlock_unlock(&hw->state_lock);
1418 }
1419 
1420 /*
1421  * Should be called only after device is paused.
1422  */
1423 int
1424 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1425 		int nb_pkts)
1426 {
1427 	struct virtio_hw *hw = dev->data->dev_private;
1428 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1429 	int ret;
1430 
1431 	hw->inject_pkts = tx_pkts;
1432 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1433 	hw->inject_pkts = NULL;
1434 
1435 	return ret;
1436 }
1437 
1438 static void
1439 virtio_notify_peers(struct rte_eth_dev *dev)
1440 {
1441 	struct virtio_hw *hw = dev->data->dev_private;
1442 	struct virtnet_rx *rxvq;
1443 	struct rte_mbuf *rarp_mbuf;
1444 
1445 	if (!dev->data->rx_queues)
1446 		return;
1447 
1448 	rxvq = dev->data->rx_queues[0];
1449 	if (!rxvq)
1450 		return;
1451 
1452 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1453 			(struct rte_ether_addr *)hw->mac_addr);
1454 	if (rarp_mbuf == NULL) {
1455 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1456 		return;
1457 	}
1458 
1459 	/* If virtio port just stopped, no need to send RARP */
1460 	if (virtio_dev_pause(dev) < 0) {
1461 		rte_pktmbuf_free(rarp_mbuf);
1462 		return;
1463 	}
1464 
1465 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1466 	virtio_dev_resume(dev);
1467 }
1468 
1469 static void
1470 virtio_ack_link_announce(struct rte_eth_dev *dev)
1471 {
1472 	struct virtio_hw *hw = dev->data->dev_private;
1473 	struct virtio_pmd_ctrl ctrl;
1474 
1475 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1476 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1477 
1478 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1479 }
1480 
1481 /*
1482  * Process virtio config changed interrupt. Call the callback
1483  * if link state changed, generate gratuitous RARP packet if
1484  * the status indicates an ANNOUNCE.
1485  */
1486 void
1487 virtio_interrupt_handler(void *param)
1488 {
1489 	struct rte_eth_dev *dev = param;
1490 	struct virtio_hw *hw = dev->data->dev_private;
1491 	uint8_t isr;
1492 	uint16_t status;
1493 
1494 	/* Read interrupt status which clears interrupt */
1495 	isr = vtpci_isr(hw);
1496 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1497 
1498 	if (virtio_intr_unmask(dev) < 0)
1499 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1500 
1501 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1502 		if (virtio_dev_link_update(dev, 0) == 0)
1503 			_rte_eth_dev_callback_process(dev,
1504 						      RTE_ETH_EVENT_INTR_LSC,
1505 						      NULL);
1506 
1507 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1508 			vtpci_read_dev_config(hw,
1509 				offsetof(struct virtio_net_config, status),
1510 				&status, sizeof(status));
1511 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1512 				virtio_notify_peers(dev);
1513 				if (hw->cvq)
1514 					virtio_ack_link_announce(dev);
1515 			}
1516 		}
1517 	}
1518 }
1519 
1520 /* set rx and tx handlers according to what is supported */
1521 static void
1522 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1523 {
1524 	struct virtio_hw *hw = eth_dev->data->dev_private;
1525 
1526 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1527 	if (vtpci_packed_queue(hw)) {
1528 		PMD_INIT_LOG(INFO,
1529 			"virtio: using packed ring %s Tx path on port %u",
1530 			hw->use_inorder_tx ? "inorder" : "standard",
1531 			eth_dev->data->port_id);
1532 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1533 	} else {
1534 		if (hw->use_inorder_tx) {
1535 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1536 				eth_dev->data->port_id);
1537 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1538 		} else {
1539 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1540 				eth_dev->data->port_id);
1541 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1542 		}
1543 	}
1544 
1545 	if (vtpci_packed_queue(hw)) {
1546 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1547 			PMD_INIT_LOG(INFO,
1548 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1549 				eth_dev->data->port_id);
1550 			eth_dev->rx_pkt_burst =
1551 				&virtio_recv_mergeable_pkts_packed;
1552 		} else {
1553 			PMD_INIT_LOG(INFO,
1554 				"virtio: using packed ring standard Rx path on port %u",
1555 				eth_dev->data->port_id);
1556 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1557 		}
1558 	} else {
1559 		if (hw->use_simple_rx) {
1560 			PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1561 				eth_dev->data->port_id);
1562 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1563 		} else if (hw->use_inorder_rx) {
1564 			PMD_INIT_LOG(INFO,
1565 				"virtio: using inorder Rx path on port %u",
1566 				eth_dev->data->port_id);
1567 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1568 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1569 			PMD_INIT_LOG(INFO,
1570 				"virtio: using mergeable buffer Rx path on port %u",
1571 				eth_dev->data->port_id);
1572 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1573 		} else {
1574 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1575 				eth_dev->data->port_id);
1576 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1577 		}
1578 	}
1579 
1580 }
1581 
1582 /* Only support 1:1 queue/interrupt mapping so far.
1583  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1584  * interrupt vectors (<N+1).
1585  */
1586 static int
1587 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1588 {
1589 	uint32_t i;
1590 	struct virtio_hw *hw = dev->data->dev_private;
1591 
1592 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1593 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1594 		dev->intr_handle->intr_vec[i] = i + 1;
1595 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1596 						 VIRTIO_MSI_NO_VECTOR) {
1597 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1598 			return -EBUSY;
1599 		}
1600 	}
1601 
1602 	return 0;
1603 }
1604 
1605 static void
1606 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1607 {
1608 	uint32_t i;
1609 	struct virtio_hw *hw = dev->data->dev_private;
1610 
1611 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1612 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1613 		VTPCI_OPS(hw)->set_queue_irq(hw,
1614 					     hw->vqs[i * VTNET_CQ],
1615 					     VIRTIO_MSI_NO_VECTOR);
1616 }
1617 
1618 static int
1619 virtio_configure_intr(struct rte_eth_dev *dev)
1620 {
1621 	struct virtio_hw *hw = dev->data->dev_private;
1622 
1623 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1624 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1625 		return -ENOTSUP;
1626 	}
1627 
1628 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1629 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1630 		return -1;
1631 	}
1632 
1633 	if (!dev->intr_handle->intr_vec) {
1634 		dev->intr_handle->intr_vec =
1635 			rte_zmalloc("intr_vec",
1636 				    hw->max_queue_pairs * sizeof(int), 0);
1637 		if (!dev->intr_handle->intr_vec) {
1638 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1639 				     hw->max_queue_pairs);
1640 			return -ENOMEM;
1641 		}
1642 	}
1643 
1644 	/* Re-register callback to update max_intr */
1645 	rte_intr_callback_unregister(dev->intr_handle,
1646 				     virtio_interrupt_handler,
1647 				     dev);
1648 	rte_intr_callback_register(dev->intr_handle,
1649 				   virtio_interrupt_handler,
1650 				   dev);
1651 
1652 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1653 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1654 	 * And for legacy devices, this should be done before queue/vec binding
1655 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1656 	 * (22) will be ignored.
1657 	 */
1658 	if (virtio_intr_enable(dev) < 0) {
1659 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1660 		return -1;
1661 	}
1662 
1663 	if (virtio_queues_bind_intr(dev) < 0) {
1664 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1665 		return -1;
1666 	}
1667 
1668 	return 0;
1669 }
1670 
1671 /* reset device and renegotiate features if needed */
1672 static int
1673 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1674 {
1675 	struct virtio_hw *hw = eth_dev->data->dev_private;
1676 	struct virtio_net_config *config;
1677 	struct virtio_net_config local_config;
1678 	struct rte_pci_device *pci_dev = NULL;
1679 	int ret;
1680 
1681 	/* Reset the device although not necessary at startup */
1682 	vtpci_reset(hw);
1683 
1684 	if (hw->vqs) {
1685 		virtio_dev_free_mbufs(eth_dev);
1686 		virtio_free_queues(hw);
1687 	}
1688 
1689 	/* Tell the host we've noticed this device. */
1690 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1691 
1692 	/* Tell the host we've known how to drive the device. */
1693 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1694 	if (virtio_negotiate_features(hw, req_features) < 0)
1695 		return -1;
1696 
1697 	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1698 
1699 	if (!hw->virtio_user_dev)
1700 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1701 
1702 	/* If host does not support both status and MSI-X then disable LSC */
1703 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1704 	    hw->use_msix != VIRTIO_MSIX_NONE)
1705 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1706 	else
1707 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1708 
1709 	/* Setting up rx_header size for the device */
1710 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1711 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1712 	    vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1713 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1714 	else
1715 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1716 
1717 	/* Copy the permanent MAC address to: virtio_hw */
1718 	virtio_get_hwaddr(hw);
1719 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1720 			&eth_dev->data->mac_addrs[0]);
1721 	PMD_INIT_LOG(DEBUG,
1722 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1723 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1724 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1725 
1726 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1727 		config = &local_config;
1728 
1729 		vtpci_read_dev_config(hw,
1730 			offsetof(struct virtio_net_config, mac),
1731 			&config->mac, sizeof(config->mac));
1732 
1733 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1734 			vtpci_read_dev_config(hw,
1735 				offsetof(struct virtio_net_config, status),
1736 				&config->status, sizeof(config->status));
1737 		} else {
1738 			PMD_INIT_LOG(DEBUG,
1739 				     "VIRTIO_NET_F_STATUS is not supported");
1740 			config->status = 0;
1741 		}
1742 
1743 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1744 			vtpci_read_dev_config(hw,
1745 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1746 				&config->max_virtqueue_pairs,
1747 				sizeof(config->max_virtqueue_pairs));
1748 		} else {
1749 			PMD_INIT_LOG(DEBUG,
1750 				     "VIRTIO_NET_F_MQ is not supported");
1751 			config->max_virtqueue_pairs = 1;
1752 		}
1753 
1754 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1755 
1756 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1757 			vtpci_read_dev_config(hw,
1758 				offsetof(struct virtio_net_config, mtu),
1759 				&config->mtu,
1760 				sizeof(config->mtu));
1761 
1762 			/*
1763 			 * MTU value has already been checked at negotiation
1764 			 * time, but check again in case it has changed since
1765 			 * then, which should not happen.
1766 			 */
1767 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1768 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1769 						config->mtu);
1770 				return -1;
1771 			}
1772 
1773 			hw->max_mtu = config->mtu;
1774 			/* Set initial MTU to maximum one supported by vhost */
1775 			eth_dev->data->mtu = config->mtu;
1776 
1777 		} else {
1778 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1779 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1780 		}
1781 
1782 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1783 				config->max_virtqueue_pairs);
1784 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1785 		PMD_INIT_LOG(DEBUG,
1786 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1787 				config->mac[0], config->mac[1],
1788 				config->mac[2], config->mac[3],
1789 				config->mac[4], config->mac[5]);
1790 	} else {
1791 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1792 		hw->max_queue_pairs = 1;
1793 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1794 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1795 	}
1796 
1797 	ret = virtio_alloc_queues(eth_dev);
1798 	if (ret < 0)
1799 		return ret;
1800 
1801 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1802 		if (virtio_configure_intr(eth_dev) < 0) {
1803 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1804 			virtio_free_queues(hw);
1805 			return -1;
1806 		}
1807 	}
1808 
1809 	vtpci_reinit_complete(hw);
1810 
1811 	if (pci_dev)
1812 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1813 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1814 			pci_dev->id.device_id);
1815 
1816 	return 0;
1817 }
1818 
1819 /*
1820  * Remap the PCI device again (IO port map for legacy device and
1821  * memory map for modern device), so that the secondary process
1822  * could have the PCI initiated correctly.
1823  */
1824 static int
1825 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1826 {
1827 	if (hw->modern) {
1828 		/*
1829 		 * We don't have to re-parse the PCI config space, since
1830 		 * rte_pci_map_device() makes sure the mapped address
1831 		 * in secondary process would equal to the one mapped in
1832 		 * the primary process: error will be returned if that
1833 		 * requirement is not met.
1834 		 *
1835 		 * That said, we could simply reuse all cap pointers
1836 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1837 		 * primary process, which is stored in shared memory.
1838 		 */
1839 		if (rte_pci_map_device(pci_dev)) {
1840 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1841 			return -1;
1842 		}
1843 	} else {
1844 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1845 			return -1;
1846 	}
1847 
1848 	return 0;
1849 }
1850 
1851 static void
1852 virtio_set_vtpci_ops(struct virtio_hw *hw)
1853 {
1854 #ifdef RTE_VIRTIO_USER
1855 	if (hw->virtio_user_dev)
1856 		VTPCI_OPS(hw) = &virtio_user_ops;
1857 	else
1858 #endif
1859 	if (hw->modern)
1860 		VTPCI_OPS(hw) = &modern_ops;
1861 	else
1862 		VTPCI_OPS(hw) = &legacy_ops;
1863 }
1864 
1865 /*
1866  * This function is based on probe() function in virtio_pci.c
1867  * It returns 0 on success.
1868  */
1869 int
1870 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1871 {
1872 	struct virtio_hw *hw = eth_dev->data->dev_private;
1873 	int ret;
1874 
1875 	if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1876 		PMD_INIT_LOG(ERR,
1877 			"Not sufficient headroom required = %d, avail = %d",
1878 			(int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1879 			RTE_PKTMBUF_HEADROOM);
1880 
1881 		return -1;
1882 	}
1883 
1884 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1885 
1886 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1887 		if (!hw->virtio_user_dev) {
1888 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1889 			if (ret)
1890 				return ret;
1891 		}
1892 
1893 		virtio_set_vtpci_ops(hw);
1894 		set_rxtx_funcs(eth_dev);
1895 
1896 		return 0;
1897 	}
1898 
1899 	/*
1900 	 * Pass the information to the rte_eth_dev_close() that it should also
1901 	 * release the private port resources.
1902 	 */
1903 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1904 
1905 	/* Allocate memory for storing MAC addresses */
1906 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1907 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1908 	if (eth_dev->data->mac_addrs == NULL) {
1909 		PMD_INIT_LOG(ERR,
1910 			"Failed to allocate %d bytes needed to store MAC addresses",
1911 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1912 		return -ENOMEM;
1913 	}
1914 
1915 	hw->port_id = eth_dev->data->port_id;
1916 	/* For virtio_user case the hw->virtio_user_dev is populated by
1917 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1918 	 */
1919 	if (!hw->virtio_user_dev) {
1920 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1921 		if (ret)
1922 			goto err_vtpci_init;
1923 	}
1924 
1925 	rte_spinlock_init(&hw->state_lock);
1926 
1927 	/* reset device and negotiate default features */
1928 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1929 	if (ret < 0)
1930 		goto err_virtio_init;
1931 
1932 	hw->opened = true;
1933 
1934 	return 0;
1935 
1936 err_virtio_init:
1937 	if (!hw->virtio_user_dev) {
1938 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1939 		if (!hw->modern)
1940 			rte_pci_ioport_unmap(VTPCI_IO(hw));
1941 	}
1942 err_vtpci_init:
1943 	rte_free(eth_dev->data->mac_addrs);
1944 	eth_dev->data->mac_addrs = NULL;
1945 	return ret;
1946 }
1947 
1948 static int
1949 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1950 {
1951 	PMD_INIT_FUNC_TRACE();
1952 
1953 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1954 		return 0;
1955 
1956 	virtio_dev_stop(eth_dev);
1957 	virtio_dev_close(eth_dev);
1958 
1959 	eth_dev->dev_ops = NULL;
1960 	eth_dev->tx_pkt_burst = NULL;
1961 	eth_dev->rx_pkt_burst = NULL;
1962 
1963 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1964 
1965 	return 0;
1966 }
1967 
1968 static int vdpa_check_handler(__rte_unused const char *key,
1969 		const char *value, __rte_unused void *opaque)
1970 {
1971 	if (strcmp(value, "1"))
1972 		return -1;
1973 
1974 	return 0;
1975 }
1976 
1977 static int
1978 vdpa_mode_selected(struct rte_devargs *devargs)
1979 {
1980 	struct rte_kvargs *kvlist;
1981 	const char *key = "vdpa";
1982 	int ret = 0;
1983 
1984 	if (devargs == NULL)
1985 		return 0;
1986 
1987 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1988 	if (kvlist == NULL)
1989 		return 0;
1990 
1991 	if (!rte_kvargs_count(kvlist, key))
1992 		goto exit;
1993 
1994 	/* vdpa mode selected when there's a key-value pair: vdpa=1 */
1995 	if (rte_kvargs_process(kvlist, key,
1996 				vdpa_check_handler, NULL) < 0) {
1997 		goto exit;
1998 	}
1999 	ret = 1;
2000 
2001 exit:
2002 	rte_kvargs_free(kvlist);
2003 	return ret;
2004 }
2005 
2006 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2007 	struct rte_pci_device *pci_dev)
2008 {
2009 	/* virtio pmd skips probe if device needs to work in vdpa mode */
2010 	if (vdpa_mode_selected(pci_dev->device.devargs))
2011 		return 1;
2012 
2013 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2014 		eth_virtio_dev_init);
2015 }
2016 
2017 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2018 {
2019 	int ret;
2020 
2021 	ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2022 	/* Port has already been released by close. */
2023 	if (ret == -ENODEV)
2024 		ret = 0;
2025 	return ret;
2026 }
2027 
2028 static struct rte_pci_driver rte_virtio_pmd = {
2029 	.driver = {
2030 		.name = "net_virtio",
2031 	},
2032 	.id_table = pci_id_virtio_map,
2033 	.drv_flags = 0,
2034 	.probe = eth_virtio_pci_probe,
2035 	.remove = eth_virtio_pci_remove,
2036 };
2037 
2038 RTE_INIT(rte_virtio_pmd_init)
2039 {
2040 	rte_eal_iopl_init();
2041 	rte_pci_register(&rte_virtio_pmd);
2042 }
2043 
2044 static bool
2045 rx_offload_enabled(struct virtio_hw *hw)
2046 {
2047 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2048 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2049 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2050 }
2051 
2052 static bool
2053 tx_offload_enabled(struct virtio_hw *hw)
2054 {
2055 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2056 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2057 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2058 }
2059 
2060 /*
2061  * Configure virtio device
2062  * It returns 0 on success.
2063  */
2064 static int
2065 virtio_dev_configure(struct rte_eth_dev *dev)
2066 {
2067 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2068 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2069 	struct virtio_hw *hw = dev->data->dev_private;
2070 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2071 		hw->vtnet_hdr_size;
2072 	uint64_t rx_offloads = rxmode->offloads;
2073 	uint64_t tx_offloads = txmode->offloads;
2074 	uint64_t req_features;
2075 	int ret;
2076 
2077 	PMD_INIT_LOG(DEBUG, "configure");
2078 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2079 
2080 	if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2081 		PMD_DRV_LOG(ERR,
2082 			"Unsupported Rx multi queue mode %d",
2083 			rxmode->mq_mode);
2084 		return -EINVAL;
2085 	}
2086 
2087 	if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2088 		PMD_DRV_LOG(ERR,
2089 			"Unsupported Tx multi queue mode %d",
2090 			txmode->mq_mode);
2091 		return -EINVAL;
2092 	}
2093 
2094 	if (dev->data->dev_conf.intr_conf.rxq) {
2095 		ret = virtio_init_device(dev, hw->req_guest_features);
2096 		if (ret < 0)
2097 			return ret;
2098 	}
2099 
2100 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2101 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2102 
2103 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2104 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2105 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2106 
2107 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2108 		req_features |=
2109 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2110 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2111 
2112 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2113 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2114 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2115 
2116 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2117 		req_features |=
2118 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2119 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2120 
2121 	/* if request features changed, reinit the device */
2122 	if (req_features != hw->req_guest_features) {
2123 		ret = virtio_init_device(dev, req_features);
2124 		if (ret < 0)
2125 			return ret;
2126 	}
2127 
2128 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2129 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2130 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2131 		PMD_DRV_LOG(ERR,
2132 			"rx checksum not available on this host");
2133 		return -ENOTSUP;
2134 	}
2135 
2136 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2137 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2138 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2139 		PMD_DRV_LOG(ERR,
2140 			"Large Receive Offload not available on this host");
2141 		return -ENOTSUP;
2142 	}
2143 
2144 	/* start control queue */
2145 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2146 		virtio_dev_cq_start(dev);
2147 
2148 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2149 		hw->vlan_strip = 1;
2150 
2151 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2152 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2153 		PMD_DRV_LOG(ERR,
2154 			    "vlan filtering not available on this host");
2155 		return -ENOTSUP;
2156 	}
2157 
2158 	hw->has_tx_offload = tx_offload_enabled(hw);
2159 	hw->has_rx_offload = rx_offload_enabled(hw);
2160 
2161 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2162 		/* Enable vector (0) for Link State Intrerrupt */
2163 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2164 				VIRTIO_MSI_NO_VECTOR) {
2165 			PMD_DRV_LOG(ERR, "failed to set config vector");
2166 			return -EBUSY;
2167 		}
2168 
2169 	hw->use_simple_rx = 1;
2170 
2171 	if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2172 		hw->use_inorder_tx = 1;
2173 		hw->use_inorder_rx = 1;
2174 		hw->use_simple_rx = 0;
2175 	}
2176 
2177 	if (vtpci_packed_queue(hw)) {
2178 		hw->use_simple_rx = 0;
2179 		hw->use_inorder_rx = 0;
2180 	}
2181 
2182 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2183 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2184 		hw->use_simple_rx = 0;
2185 	}
2186 #endif
2187 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2188 		 hw->use_simple_rx = 0;
2189 	}
2190 
2191 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2192 			   DEV_RX_OFFLOAD_TCP_CKSUM |
2193 			   DEV_RX_OFFLOAD_TCP_LRO |
2194 			   DEV_RX_OFFLOAD_VLAN_STRIP))
2195 		hw->use_simple_rx = 0;
2196 
2197 	return 0;
2198 }
2199 
2200 
2201 static int
2202 virtio_dev_start(struct rte_eth_dev *dev)
2203 {
2204 	uint16_t nb_queues, i;
2205 	struct virtnet_rx *rxvq;
2206 	struct virtnet_tx *txvq __rte_unused;
2207 	struct virtio_hw *hw = dev->data->dev_private;
2208 	int ret;
2209 
2210 	/* Finish the initialization of the queues */
2211 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2212 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2213 		if (ret < 0)
2214 			return ret;
2215 	}
2216 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2217 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2218 		if (ret < 0)
2219 			return ret;
2220 	}
2221 
2222 	/* check if lsc interrupt feature is enabled */
2223 	if (dev->data->dev_conf.intr_conf.lsc) {
2224 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2225 			PMD_DRV_LOG(ERR, "link status not supported by host");
2226 			return -ENOTSUP;
2227 		}
2228 	}
2229 
2230 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2231 	 * in device configure, but it could be unmapped  when device is
2232 	 * stopped.
2233 	 */
2234 	if (dev->data->dev_conf.intr_conf.lsc ||
2235 	    dev->data->dev_conf.intr_conf.rxq) {
2236 		virtio_intr_disable(dev);
2237 
2238 		/* Setup interrupt callback  */
2239 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2240 			rte_intr_callback_register(dev->intr_handle,
2241 						   virtio_interrupt_handler,
2242 						   dev);
2243 
2244 		if (virtio_intr_enable(dev) < 0) {
2245 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2246 			return -EIO;
2247 		}
2248 	}
2249 
2250 	/*Notify the backend
2251 	 *Otherwise the tap backend might already stop its queue due to fullness.
2252 	 *vhost backend will have no chance to be waked up
2253 	 */
2254 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2255 	if (hw->max_queue_pairs > 1) {
2256 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2257 			return -EINVAL;
2258 	}
2259 
2260 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2261 
2262 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2263 		rxvq = dev->data->rx_queues[i];
2264 		/* Flush the old packets */
2265 		virtqueue_rxvq_flush(rxvq->vq);
2266 		virtqueue_notify(rxvq->vq);
2267 	}
2268 
2269 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2270 		txvq = dev->data->tx_queues[i];
2271 		virtqueue_notify(txvq->vq);
2272 	}
2273 
2274 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2275 
2276 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2277 		rxvq = dev->data->rx_queues[i];
2278 		VIRTQUEUE_DUMP(rxvq->vq);
2279 	}
2280 
2281 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2282 		txvq = dev->data->tx_queues[i];
2283 		VIRTQUEUE_DUMP(txvq->vq);
2284 	}
2285 
2286 	set_rxtx_funcs(dev);
2287 	hw->started = true;
2288 
2289 	/* Initialize Link state */
2290 	virtio_dev_link_update(dev, 0);
2291 
2292 	return 0;
2293 }
2294 
2295 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2296 {
2297 	struct virtio_hw *hw = dev->data->dev_private;
2298 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2299 	const char *type __rte_unused;
2300 	unsigned int i, mbuf_num = 0;
2301 	struct virtqueue *vq;
2302 	struct rte_mbuf *buf;
2303 	int queue_type;
2304 
2305 	if (hw->vqs == NULL)
2306 		return;
2307 
2308 	for (i = 0; i < nr_vq; i++) {
2309 		vq = hw->vqs[i];
2310 		if (!vq)
2311 			continue;
2312 
2313 		queue_type = virtio_get_queue_type(hw, i);
2314 		if (queue_type == VTNET_RQ)
2315 			type = "rxq";
2316 		else if (queue_type == VTNET_TQ)
2317 			type = "txq";
2318 		else
2319 			continue;
2320 
2321 		PMD_INIT_LOG(DEBUG,
2322 			"Before freeing %s[%d] used and unused buf",
2323 			type, i);
2324 		VIRTQUEUE_DUMP(vq);
2325 
2326 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2327 			rte_pktmbuf_free(buf);
2328 			mbuf_num++;
2329 		}
2330 
2331 		PMD_INIT_LOG(DEBUG,
2332 			"After freeing %s[%d] used and unused buf",
2333 			type, i);
2334 		VIRTQUEUE_DUMP(vq);
2335 	}
2336 
2337 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2338 }
2339 
2340 /*
2341  * Stop device: disable interrupt and mark link down
2342  */
2343 static void
2344 virtio_dev_stop(struct rte_eth_dev *dev)
2345 {
2346 	struct virtio_hw *hw = dev->data->dev_private;
2347 	struct rte_eth_link link;
2348 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2349 
2350 	PMD_INIT_LOG(DEBUG, "stop");
2351 
2352 	rte_spinlock_lock(&hw->state_lock);
2353 	if (!hw->started)
2354 		goto out_unlock;
2355 	hw->started = false;
2356 
2357 	if (intr_conf->lsc || intr_conf->rxq) {
2358 		virtio_intr_disable(dev);
2359 
2360 		/* Reset interrupt callback  */
2361 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2362 			rte_intr_callback_unregister(dev->intr_handle,
2363 						     virtio_interrupt_handler,
2364 						     dev);
2365 		}
2366 	}
2367 
2368 	memset(&link, 0, sizeof(link));
2369 	rte_eth_linkstatus_set(dev, &link);
2370 out_unlock:
2371 	rte_spinlock_unlock(&hw->state_lock);
2372 }
2373 
2374 static int
2375 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2376 {
2377 	struct rte_eth_link link;
2378 	uint16_t status;
2379 	struct virtio_hw *hw = dev->data->dev_private;
2380 
2381 	memset(&link, 0, sizeof(link));
2382 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
2383 	link.link_speed  = ETH_SPEED_NUM_10G;
2384 	link.link_autoneg = ETH_LINK_FIXED;
2385 
2386 	if (!hw->started) {
2387 		link.link_status = ETH_LINK_DOWN;
2388 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2389 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2390 		vtpci_read_dev_config(hw,
2391 				offsetof(struct virtio_net_config, status),
2392 				&status, sizeof(status));
2393 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2394 			link.link_status = ETH_LINK_DOWN;
2395 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2396 				     dev->data->port_id);
2397 		} else {
2398 			link.link_status = ETH_LINK_UP;
2399 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2400 				     dev->data->port_id);
2401 		}
2402 	} else {
2403 		link.link_status = ETH_LINK_UP;
2404 	}
2405 
2406 	return rte_eth_linkstatus_set(dev, &link);
2407 }
2408 
2409 static int
2410 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2411 {
2412 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2413 	struct virtio_hw *hw = dev->data->dev_private;
2414 	uint64_t offloads = rxmode->offloads;
2415 
2416 	if (mask & ETH_VLAN_FILTER_MASK) {
2417 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2418 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2419 
2420 			PMD_DRV_LOG(NOTICE,
2421 				"vlan filtering not available on this host");
2422 
2423 			return -ENOTSUP;
2424 		}
2425 	}
2426 
2427 	if (mask & ETH_VLAN_STRIP_MASK)
2428 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2429 
2430 	return 0;
2431 }
2432 
2433 static int
2434 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2435 {
2436 	uint64_t tso_mask, host_features;
2437 	struct virtio_hw *hw = dev->data->dev_private;
2438 
2439 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2440 
2441 	dev_info->max_rx_queues =
2442 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2443 	dev_info->max_tx_queues =
2444 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2445 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2446 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2447 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2448 
2449 	host_features = VTPCI_OPS(hw)->get_features(hw);
2450 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2451 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2452 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2453 		dev_info->rx_offload_capa |=
2454 			DEV_RX_OFFLOAD_TCP_CKSUM |
2455 			DEV_RX_OFFLOAD_UDP_CKSUM;
2456 	}
2457 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2458 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2459 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2460 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2461 	if ((host_features & tso_mask) == tso_mask)
2462 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2463 
2464 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2465 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2466 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2467 		dev_info->tx_offload_capa |=
2468 			DEV_TX_OFFLOAD_UDP_CKSUM |
2469 			DEV_TX_OFFLOAD_TCP_CKSUM;
2470 	}
2471 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2472 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2473 	if ((host_features & tso_mask) == tso_mask)
2474 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2475 
2476 	return 0;
2477 }
2478 
2479 /*
2480  * It enables testpmd to collect per queue stats.
2481  */
2482 static int
2483 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2484 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2485 __rte_unused uint8_t is_rx)
2486 {
2487 	return 0;
2488 }
2489 
2490 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2491 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2492 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2493 
2494 RTE_INIT(virtio_init_log)
2495 {
2496 	virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2497 	if (virtio_logtype_init >= 0)
2498 		rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2499 	virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2500 	if (virtio_logtype_driver >= 0)
2501 		rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2502 }
2503