xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision f5057be340e44f3edc0fe90fa875eb89a4c49b4f)
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 uint32_t virtio_dev_speed_capa_get(uint32_t speed);
49 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
50 	int *vdpa,
51 	uint32_t *speed,
52 	int *vectorized);
53 static int virtio_dev_info_get(struct rte_eth_dev *dev,
54 				struct rte_eth_dev_info *dev_info);
55 static int virtio_dev_link_update(struct rte_eth_dev *dev,
56 	int wait_to_complete);
57 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
58 
59 static void virtio_set_hwaddr(struct virtio_hw *hw);
60 static void virtio_get_hwaddr(struct virtio_hw *hw);
61 
62 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
63 				 struct rte_eth_stats *stats);
64 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
65 				 struct rte_eth_xstat *xstats, unsigned n);
66 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
67 				       struct rte_eth_xstat_name *xstats_names,
68 				       unsigned limit);
69 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
70 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
71 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
72 				uint16_t vlan_id, int on);
73 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
74 				struct rte_ether_addr *mac_addr,
75 				uint32_t index, uint32_t vmdq);
76 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
77 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
78 				struct rte_ether_addr *mac_addr);
79 
80 static int virtio_intr_disable(struct rte_eth_dev *dev);
81 
82 static int virtio_dev_queue_stats_mapping_set(
83 	struct rte_eth_dev *eth_dev,
84 	uint16_t queue_id,
85 	uint8_t stat_idx,
86 	uint8_t is_rx);
87 
88 static void virtio_notify_peers(struct rte_eth_dev *dev);
89 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
90 
91 /*
92  * The set of PCI devices this driver supports
93  */
94 static const struct rte_pci_id pci_id_virtio_map[] = {
95 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
96 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
97 	{ .vendor_id = 0, /* sentinel */ },
98 };
99 
100 struct rte_virtio_xstats_name_off {
101 	char name[RTE_ETH_XSTATS_NAME_SIZE];
102 	unsigned offset;
103 };
104 
105 /* [rt]x_qX_ is prepended to the name string here */
106 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
107 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
108 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
109 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
110 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
111 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
112 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
113 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
114 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
115 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
116 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
117 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
118 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
119 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
120 };
121 
122 /* [rt]x_qX_ is prepended to the name string here */
123 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
124 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
125 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
126 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
127 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
128 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
129 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
130 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
131 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
132 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
133 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
134 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
135 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
136 };
137 
138 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
139 			    sizeof(rte_virtio_rxq_stat_strings[0]))
140 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
141 			    sizeof(rte_virtio_txq_stat_strings[0]))
142 
143 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
144 
145 static struct virtio_pmd_ctrl *
146 virtio_send_command_packed(struct virtnet_ctl *cvq,
147 			   struct virtio_pmd_ctrl *ctrl,
148 			   int *dlen, int pkt_num)
149 {
150 	struct virtqueue *vq = cvq->vq;
151 	int head;
152 	struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
153 	struct virtio_pmd_ctrl *result;
154 	uint16_t flags;
155 	int sum = 0;
156 	int nb_descs = 0;
157 	int k;
158 
159 	/*
160 	 * Format is enforced in qemu code:
161 	 * One TX packet for header;
162 	 * At least one TX packet per argument;
163 	 * One RX packet for ACK.
164 	 */
165 	head = vq->vq_avail_idx;
166 	flags = vq->vq_packed.cached_flags;
167 	desc[head].addr = cvq->virtio_net_hdr_mem;
168 	desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
169 	vq->vq_free_cnt--;
170 	nb_descs++;
171 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
172 		vq->vq_avail_idx -= vq->vq_nentries;
173 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
174 	}
175 
176 	for (k = 0; k < pkt_num; k++) {
177 		desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
178 			+ sizeof(struct virtio_net_ctrl_hdr)
179 			+ sizeof(ctrl->status) + sizeof(uint8_t) * sum;
180 		desc[vq->vq_avail_idx].len = dlen[k];
181 		desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
182 			vq->vq_packed.cached_flags;
183 		sum += dlen[k];
184 		vq->vq_free_cnt--;
185 		nb_descs++;
186 		if (++vq->vq_avail_idx >= vq->vq_nentries) {
187 			vq->vq_avail_idx -= vq->vq_nentries;
188 			vq->vq_packed.cached_flags ^=
189 				VRING_PACKED_DESC_F_AVAIL_USED;
190 		}
191 	}
192 
193 	desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
194 		+ sizeof(struct virtio_net_ctrl_hdr);
195 	desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
196 	desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
197 		vq->vq_packed.cached_flags;
198 	vq->vq_free_cnt--;
199 	nb_descs++;
200 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
201 		vq->vq_avail_idx -= vq->vq_nentries;
202 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
203 	}
204 
205 	virtio_wmb(vq->hw->weak_barriers);
206 	desc[head].flags = VRING_DESC_F_NEXT | flags;
207 
208 	virtio_wmb(vq->hw->weak_barriers);
209 	virtqueue_notify(vq);
210 
211 	/* wait for used descriptors in virtqueue */
212 	while (!desc_is_used(&desc[head], vq))
213 		usleep(100);
214 
215 	virtio_rmb(vq->hw->weak_barriers);
216 
217 	/* now get used descriptors */
218 	vq->vq_free_cnt += nb_descs;
219 	vq->vq_used_cons_idx += nb_descs;
220 	if (vq->vq_used_cons_idx >= vq->vq_nentries) {
221 		vq->vq_used_cons_idx -= vq->vq_nentries;
222 		vq->vq_packed.used_wrap_counter ^= 1;
223 	}
224 
225 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
226 			"vq->vq_avail_idx=%d\n"
227 			"vq->vq_used_cons_idx=%d\n"
228 			"vq->vq_packed.cached_flags=0x%x\n"
229 			"vq->vq_packed.used_wrap_counter=%d\n",
230 			vq->vq_free_cnt,
231 			vq->vq_avail_idx,
232 			vq->vq_used_cons_idx,
233 			vq->vq_packed.cached_flags,
234 			vq->vq_packed.used_wrap_counter);
235 
236 	result = cvq->virtio_net_hdr_mz->addr;
237 	return result;
238 }
239 
240 static struct virtio_pmd_ctrl *
241 virtio_send_command_split(struct virtnet_ctl *cvq,
242 			  struct virtio_pmd_ctrl *ctrl,
243 			  int *dlen, int pkt_num)
244 {
245 	struct virtio_pmd_ctrl *result;
246 	struct virtqueue *vq = cvq->vq;
247 	uint32_t head, i;
248 	int k, sum = 0;
249 
250 	head = vq->vq_desc_head_idx;
251 
252 	/*
253 	 * Format is enforced in qemu code:
254 	 * One TX packet for header;
255 	 * At least one TX packet per argument;
256 	 * One RX packet for ACK.
257 	 */
258 	vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
259 	vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
260 	vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
261 	vq->vq_free_cnt--;
262 	i = vq->vq_split.ring.desc[head].next;
263 
264 	for (k = 0; k < pkt_num; k++) {
265 		vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
266 		vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
267 			+ sizeof(struct virtio_net_ctrl_hdr)
268 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
269 		vq->vq_split.ring.desc[i].len = dlen[k];
270 		sum += dlen[k];
271 		vq->vq_free_cnt--;
272 		i = vq->vq_split.ring.desc[i].next;
273 	}
274 
275 	vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
276 	vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
277 			+ sizeof(struct virtio_net_ctrl_hdr);
278 	vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
279 	vq->vq_free_cnt--;
280 
281 	vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
282 
283 	vq_update_avail_ring(vq, head);
284 	vq_update_avail_idx(vq);
285 
286 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
287 
288 	virtqueue_notify(vq);
289 
290 	while (virtqueue_nused(vq) == 0)
291 		usleep(100);
292 
293 	while (virtqueue_nused(vq)) {
294 		uint32_t idx, desc_idx, used_idx;
295 		struct vring_used_elem *uep;
296 
297 		used_idx = (uint32_t)(vq->vq_used_cons_idx
298 				& (vq->vq_nentries - 1));
299 		uep = &vq->vq_split.ring.used->ring[used_idx];
300 		idx = (uint32_t) uep->id;
301 		desc_idx = idx;
302 
303 		while (vq->vq_split.ring.desc[desc_idx].flags &
304 				VRING_DESC_F_NEXT) {
305 			desc_idx = vq->vq_split.ring.desc[desc_idx].next;
306 			vq->vq_free_cnt++;
307 		}
308 
309 		vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
310 		vq->vq_desc_head_idx = idx;
311 
312 		vq->vq_used_cons_idx++;
313 		vq->vq_free_cnt++;
314 	}
315 
316 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
317 			vq->vq_free_cnt, vq->vq_desc_head_idx);
318 
319 	result = cvq->virtio_net_hdr_mz->addr;
320 	return result;
321 }
322 
323 static int
324 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
325 		    int *dlen, int pkt_num)
326 {
327 	virtio_net_ctrl_ack status = ~0;
328 	struct virtio_pmd_ctrl *result;
329 	struct virtqueue *vq;
330 
331 	ctrl->status = status;
332 
333 	if (!cvq || !cvq->vq) {
334 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
335 		return -1;
336 	}
337 
338 	rte_spinlock_lock(&cvq->lock);
339 	vq = cvq->vq;
340 
341 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
342 		"vq->hw->cvq = %p vq = %p",
343 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
344 
345 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
346 		rte_spinlock_unlock(&cvq->lock);
347 		return -1;
348 	}
349 
350 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
351 		sizeof(struct virtio_pmd_ctrl));
352 
353 	if (vtpci_packed_queue(vq->hw))
354 		result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
355 	else
356 		result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
357 
358 	rte_spinlock_unlock(&cvq->lock);
359 	return result->status;
360 }
361 
362 static int
363 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
364 {
365 	struct virtio_hw *hw = dev->data->dev_private;
366 	struct virtio_pmd_ctrl ctrl;
367 	int dlen[1];
368 	int ret;
369 
370 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
371 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
372 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
373 
374 	dlen[0] = sizeof(uint16_t);
375 
376 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
377 	if (ret) {
378 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
379 			  "failed, this is too late now...");
380 		return -EINVAL;
381 	}
382 
383 	return 0;
384 }
385 
386 static void
387 virtio_dev_queue_release(void *queue __rte_unused)
388 {
389 	/* do nothing */
390 }
391 
392 static uint16_t
393 virtio_get_nr_vq(struct virtio_hw *hw)
394 {
395 	uint16_t nr_vq = hw->max_queue_pairs * 2;
396 
397 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
398 		nr_vq += 1;
399 
400 	return nr_vq;
401 }
402 
403 static void
404 virtio_init_vring(struct virtqueue *vq)
405 {
406 	int size = vq->vq_nentries;
407 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
408 
409 	PMD_INIT_FUNC_TRACE();
410 
411 	memset(ring_mem, 0, vq->vq_ring_size);
412 
413 	vq->vq_used_cons_idx = 0;
414 	vq->vq_desc_head_idx = 0;
415 	vq->vq_avail_idx = 0;
416 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
417 	vq->vq_free_cnt = vq->vq_nentries;
418 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
419 	if (vtpci_packed_queue(vq->hw)) {
420 		vring_init_packed(&vq->vq_packed.ring, ring_mem,
421 				  VIRTIO_PCI_VRING_ALIGN, size);
422 		vring_desc_init_packed(vq, size);
423 	} else {
424 		struct vring *vr = &vq->vq_split.ring;
425 
426 		vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
427 		vring_desc_init_split(vr->desc, size);
428 	}
429 	/*
430 	 * Disable device(host) interrupting guest
431 	 */
432 	virtqueue_disable_intr(vq);
433 }
434 
435 static int
436 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
437 {
438 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
439 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
440 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
441 	unsigned int vq_size, size;
442 	struct virtio_hw *hw = dev->data->dev_private;
443 	struct virtnet_rx *rxvq = NULL;
444 	struct virtnet_tx *txvq = NULL;
445 	struct virtnet_ctl *cvq = NULL;
446 	struct virtqueue *vq;
447 	size_t sz_hdr_mz = 0;
448 	void *sw_ring = NULL;
449 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
450 	int ret;
451 	int numa_node = dev->device->numa_node;
452 
453 	PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
454 			vtpci_queue_idx, numa_node);
455 
456 	/*
457 	 * Read the virtqueue size from the Queue Size field
458 	 * Always power of 2 and if 0 virtqueue does not exist
459 	 */
460 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
461 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
462 	if (vq_size == 0) {
463 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
464 		return -EINVAL;
465 	}
466 
467 	if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
468 		PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
469 		return -EINVAL;
470 	}
471 
472 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
473 		 dev->data->port_id, vtpci_queue_idx);
474 
475 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
476 				vq_size * sizeof(struct vq_desc_extra),
477 				RTE_CACHE_LINE_SIZE);
478 	if (queue_type == VTNET_TQ) {
479 		/*
480 		 * For each xmit packet, allocate a virtio_net_hdr
481 		 * and indirect ring elements
482 		 */
483 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
484 	} else if (queue_type == VTNET_CQ) {
485 		/* Allocate a page for control vq command, data and status */
486 		sz_hdr_mz = PAGE_SIZE;
487 	}
488 
489 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
490 				numa_node);
491 	if (vq == NULL) {
492 		PMD_INIT_LOG(ERR, "can not allocate vq");
493 		return -ENOMEM;
494 	}
495 	hw->vqs[vtpci_queue_idx] = vq;
496 
497 	vq->hw = hw;
498 	vq->vq_queue_index = vtpci_queue_idx;
499 	vq->vq_nentries = vq_size;
500 	if (vtpci_packed_queue(hw)) {
501 		vq->vq_packed.used_wrap_counter = 1;
502 		vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
503 		vq->vq_packed.event_flags_shadow = 0;
504 		if (queue_type == VTNET_RQ)
505 			vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
506 	}
507 
508 	/*
509 	 * Reserve a memzone for vring elements
510 	 */
511 	size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
512 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
513 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
514 		     size, vq->vq_ring_size);
515 
516 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
517 			numa_node, RTE_MEMZONE_IOVA_CONTIG,
518 			VIRTIO_PCI_VRING_ALIGN);
519 	if (mz == NULL) {
520 		if (rte_errno == EEXIST)
521 			mz = rte_memzone_lookup(vq_name);
522 		if (mz == NULL) {
523 			ret = -ENOMEM;
524 			goto fail_q_alloc;
525 		}
526 	}
527 
528 	memset(mz->addr, 0, mz->len);
529 
530 	vq->vq_ring_mem = mz->iova;
531 	vq->vq_ring_virt_mem = mz->addr;
532 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
533 		     (uint64_t)mz->iova);
534 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
535 		     (uint64_t)(uintptr_t)mz->addr);
536 
537 	virtio_init_vring(vq);
538 
539 	if (sz_hdr_mz) {
540 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
541 			 dev->data->port_id, vtpci_queue_idx);
542 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
543 				numa_node, RTE_MEMZONE_IOVA_CONTIG,
544 				RTE_CACHE_LINE_SIZE);
545 		if (hdr_mz == NULL) {
546 			if (rte_errno == EEXIST)
547 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
548 			if (hdr_mz == NULL) {
549 				ret = -ENOMEM;
550 				goto fail_q_alloc;
551 			}
552 		}
553 	}
554 
555 	if (queue_type == VTNET_RQ) {
556 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
557 			       sizeof(vq->sw_ring[0]);
558 
559 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
560 				RTE_CACHE_LINE_SIZE, numa_node);
561 		if (!sw_ring) {
562 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
563 			ret = -ENOMEM;
564 			goto fail_q_alloc;
565 		}
566 
567 		vq->sw_ring = sw_ring;
568 		rxvq = &vq->rxq;
569 		rxvq->vq = vq;
570 		rxvq->port_id = dev->data->port_id;
571 		rxvq->mz = mz;
572 	} else if (queue_type == VTNET_TQ) {
573 		txvq = &vq->txq;
574 		txvq->vq = vq;
575 		txvq->port_id = dev->data->port_id;
576 		txvq->mz = mz;
577 		txvq->virtio_net_hdr_mz = hdr_mz;
578 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
579 	} else if (queue_type == VTNET_CQ) {
580 		cvq = &vq->cq;
581 		cvq->vq = vq;
582 		cvq->mz = mz;
583 		cvq->virtio_net_hdr_mz = hdr_mz;
584 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
585 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
586 
587 		hw->cvq = cvq;
588 	}
589 
590 	/* For virtio_user case (that is when hw->virtio_user_dev is not NULL),
591 	 * we use virtual address. And we need properly set _offset_, please see
592 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
593 	 */
594 	if (!hw->virtio_user_dev)
595 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
596 	else {
597 		vq->vq_ring_mem = (uintptr_t)mz->addr;
598 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
599 		if (queue_type == VTNET_TQ)
600 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
601 		else if (queue_type == VTNET_CQ)
602 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
603 	}
604 
605 	if (queue_type == VTNET_TQ) {
606 		struct virtio_tx_region *txr;
607 		unsigned int i;
608 
609 		txr = hdr_mz->addr;
610 		memset(txr, 0, vq_size * sizeof(*txr));
611 		for (i = 0; i < vq_size; i++) {
612 			/* first indirect descriptor is always the tx header */
613 			if (!vtpci_packed_queue(hw)) {
614 				struct vring_desc *start_dp = txr[i].tx_indir;
615 				vring_desc_init_split(start_dp,
616 						      RTE_DIM(txr[i].tx_indir));
617 				start_dp->addr = txvq->virtio_net_hdr_mem
618 					+ i * sizeof(*txr)
619 					+ offsetof(struct virtio_tx_region,
620 						   tx_hdr);
621 				start_dp->len = hw->vtnet_hdr_size;
622 				start_dp->flags = VRING_DESC_F_NEXT;
623 			} else {
624 				struct vring_packed_desc *start_dp =
625 					txr[i].tx_packed_indir;
626 				vring_desc_init_indirect_packed(start_dp,
627 				      RTE_DIM(txr[i].tx_packed_indir));
628 				start_dp->addr = txvq->virtio_net_hdr_mem
629 					+ i * sizeof(*txr)
630 					+ offsetof(struct virtio_tx_region,
631 						   tx_hdr);
632 				start_dp->len = hw->vtnet_hdr_size;
633 			}
634 		}
635 	}
636 
637 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
638 		PMD_INIT_LOG(ERR, "setup_queue failed");
639 		return -EINVAL;
640 	}
641 
642 	return 0;
643 
644 fail_q_alloc:
645 	rte_free(sw_ring);
646 	rte_memzone_free(hdr_mz);
647 	rte_memzone_free(mz);
648 	rte_free(vq);
649 
650 	return ret;
651 }
652 
653 static void
654 virtio_free_queues(struct virtio_hw *hw)
655 {
656 	uint16_t nr_vq = virtio_get_nr_vq(hw);
657 	struct virtqueue *vq;
658 	int queue_type;
659 	uint16_t i;
660 
661 	if (hw->vqs == NULL)
662 		return;
663 
664 	for (i = 0; i < nr_vq; i++) {
665 		vq = hw->vqs[i];
666 		if (!vq)
667 			continue;
668 
669 		queue_type = virtio_get_queue_type(hw, i);
670 		if (queue_type == VTNET_RQ) {
671 			rte_free(vq->sw_ring);
672 			rte_memzone_free(vq->rxq.mz);
673 		} else if (queue_type == VTNET_TQ) {
674 			rte_memzone_free(vq->txq.mz);
675 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
676 		} else {
677 			rte_memzone_free(vq->cq.mz);
678 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
679 		}
680 
681 		rte_free(vq);
682 		hw->vqs[i] = NULL;
683 	}
684 
685 	rte_free(hw->vqs);
686 	hw->vqs = NULL;
687 }
688 
689 static int
690 virtio_alloc_queues(struct rte_eth_dev *dev)
691 {
692 	struct virtio_hw *hw = dev->data->dev_private;
693 	uint16_t nr_vq = virtio_get_nr_vq(hw);
694 	uint16_t i;
695 	int ret;
696 
697 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
698 	if (!hw->vqs) {
699 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
700 		return -ENOMEM;
701 	}
702 
703 	for (i = 0; i < nr_vq; i++) {
704 		ret = virtio_init_queue(dev, i);
705 		if (ret < 0) {
706 			virtio_free_queues(hw);
707 			return ret;
708 		}
709 	}
710 
711 	return 0;
712 }
713 
714 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
715 
716 static int
717 virtio_dev_close(struct rte_eth_dev *dev)
718 {
719 	struct virtio_hw *hw = dev->data->dev_private;
720 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
721 
722 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
723 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
724 		return 0;
725 
726 	if (!hw->opened)
727 		return 0;
728 	hw->opened = false;
729 
730 	/* reset the NIC */
731 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
732 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
733 	if (intr_conf->rxq)
734 		virtio_queues_unbind_intr(dev);
735 
736 	if (intr_conf->lsc || intr_conf->rxq) {
737 		virtio_intr_disable(dev);
738 		rte_intr_efd_disable(dev->intr_handle);
739 		rte_free(dev->intr_handle->intr_vec);
740 		dev->intr_handle->intr_vec = NULL;
741 	}
742 
743 	vtpci_reset(hw);
744 	virtio_dev_free_mbufs(dev);
745 	virtio_free_queues(hw);
746 
747 #ifdef RTE_VIRTIO_USER
748 	if (hw->virtio_user_dev)
749 		virtio_user_dev_uninit(hw->virtio_user_dev);
750 	else
751 #endif
752 	if (dev->device) {
753 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
754 		if (!hw->modern)
755 			rte_pci_ioport_unmap(VTPCI_IO(hw));
756 	}
757 
758 	return 0;
759 }
760 
761 static int
762 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
763 {
764 	struct virtio_hw *hw = dev->data->dev_private;
765 	struct virtio_pmd_ctrl ctrl;
766 	int dlen[1];
767 	int ret;
768 
769 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
770 		PMD_INIT_LOG(INFO, "host does not support rx control");
771 		return -ENOTSUP;
772 	}
773 
774 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
775 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
776 	ctrl.data[0] = 1;
777 	dlen[0] = 1;
778 
779 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
780 	if (ret) {
781 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
782 		return -EAGAIN;
783 	}
784 
785 	return 0;
786 }
787 
788 static int
789 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
790 {
791 	struct virtio_hw *hw = dev->data->dev_private;
792 	struct virtio_pmd_ctrl ctrl;
793 	int dlen[1];
794 	int ret;
795 
796 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
797 		PMD_INIT_LOG(INFO, "host does not support rx control");
798 		return -ENOTSUP;
799 	}
800 
801 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
802 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
803 	ctrl.data[0] = 0;
804 	dlen[0] = 1;
805 
806 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
807 	if (ret) {
808 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
809 		return -EAGAIN;
810 	}
811 
812 	return 0;
813 }
814 
815 static int
816 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
817 {
818 	struct virtio_hw *hw = dev->data->dev_private;
819 	struct virtio_pmd_ctrl ctrl;
820 	int dlen[1];
821 	int ret;
822 
823 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
824 		PMD_INIT_LOG(INFO, "host does not support rx control");
825 		return -ENOTSUP;
826 	}
827 
828 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
829 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
830 	ctrl.data[0] = 1;
831 	dlen[0] = 1;
832 
833 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
834 	if (ret) {
835 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
836 		return -EAGAIN;
837 	}
838 
839 	return 0;
840 }
841 
842 static int
843 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
844 {
845 	struct virtio_hw *hw = dev->data->dev_private;
846 	struct virtio_pmd_ctrl ctrl;
847 	int dlen[1];
848 	int ret;
849 
850 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
851 		PMD_INIT_LOG(INFO, "host does not support rx control");
852 		return -ENOTSUP;
853 	}
854 
855 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
856 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
857 	ctrl.data[0] = 0;
858 	dlen[0] = 1;
859 
860 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
861 	if (ret) {
862 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
863 		return -EAGAIN;
864 	}
865 
866 	return 0;
867 }
868 
869 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
870 static int
871 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
872 {
873 	struct virtio_hw *hw = dev->data->dev_private;
874 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
875 				 hw->vtnet_hdr_size;
876 	uint32_t frame_size = mtu + ether_hdr_len;
877 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
878 
879 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
880 
881 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
882 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
883 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
884 		return -EINVAL;
885 	}
886 	return 0;
887 }
888 
889 static int
890 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
891 {
892 	struct virtio_hw *hw = dev->data->dev_private;
893 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
894 	struct virtqueue *vq = rxvq->vq;
895 
896 	virtqueue_enable_intr(vq);
897 	virtio_mb(hw->weak_barriers);
898 	return 0;
899 }
900 
901 static int
902 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
903 {
904 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
905 	struct virtqueue *vq = rxvq->vq;
906 
907 	virtqueue_disable_intr(vq);
908 	return 0;
909 }
910 
911 /*
912  * dev_ops for virtio, bare necessities for basic operation
913  */
914 static const struct eth_dev_ops virtio_eth_dev_ops = {
915 	.dev_configure           = virtio_dev_configure,
916 	.dev_start               = virtio_dev_start,
917 	.dev_stop                = virtio_dev_stop,
918 	.dev_close               = virtio_dev_close,
919 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
920 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
921 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
922 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
923 	.mtu_set                 = virtio_mtu_set,
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 	.link_update             = virtio_dev_link_update,
931 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
932 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
933 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
934 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
935 	.rx_queue_release        = virtio_dev_queue_release,
936 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
937 	.tx_queue_release        = virtio_dev_queue_release,
938 	/* collect stats per queue */
939 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
940 	.vlan_filter_set         = virtio_vlan_filter_set,
941 	.mac_addr_add            = virtio_mac_addr_add,
942 	.mac_addr_remove         = virtio_mac_addr_remove,
943 	.mac_addr_set            = virtio_mac_addr_set,
944 };
945 
946 /*
947  * dev_ops for virtio-user in secondary processes, as we just have
948  * some limited supports currently.
949  */
950 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
951 	.dev_infos_get           = virtio_dev_info_get,
952 	.stats_get               = virtio_dev_stats_get,
953 	.xstats_get              = virtio_dev_xstats_get,
954 	.xstats_get_names        = virtio_dev_xstats_get_names,
955 	.stats_reset             = virtio_dev_stats_reset,
956 	.xstats_reset            = virtio_dev_stats_reset,
957 	/* collect stats per queue */
958 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
959 };
960 
961 static void
962 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
963 {
964 	unsigned i;
965 
966 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
967 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
968 		if (txvq == NULL)
969 			continue;
970 
971 		stats->opackets += txvq->stats.packets;
972 		stats->obytes += txvq->stats.bytes;
973 
974 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
975 			stats->q_opackets[i] = txvq->stats.packets;
976 			stats->q_obytes[i] = txvq->stats.bytes;
977 		}
978 	}
979 
980 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
981 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
982 		if (rxvq == NULL)
983 			continue;
984 
985 		stats->ipackets += rxvq->stats.packets;
986 		stats->ibytes += rxvq->stats.bytes;
987 		stats->ierrors += rxvq->stats.errors;
988 
989 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
990 			stats->q_ipackets[i] = rxvq->stats.packets;
991 			stats->q_ibytes[i] = rxvq->stats.bytes;
992 		}
993 	}
994 
995 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
996 }
997 
998 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
999 				       struct rte_eth_xstat_name *xstats_names,
1000 				       __rte_unused unsigned limit)
1001 {
1002 	unsigned i;
1003 	unsigned count = 0;
1004 	unsigned t;
1005 
1006 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1007 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1008 
1009 	if (xstats_names != NULL) {
1010 		/* Note: limit checked in rte_eth_xstats_names() */
1011 
1012 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
1013 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1014 			if (rxvq == NULL)
1015 				continue;
1016 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1017 				snprintf(xstats_names[count].name,
1018 					sizeof(xstats_names[count].name),
1019 					"rx_q%u_%s", i,
1020 					rte_virtio_rxq_stat_strings[t].name);
1021 				count++;
1022 			}
1023 		}
1024 
1025 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
1026 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
1027 			if (txvq == NULL)
1028 				continue;
1029 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1030 				snprintf(xstats_names[count].name,
1031 					sizeof(xstats_names[count].name),
1032 					"tx_q%u_%s", i,
1033 					rte_virtio_txq_stat_strings[t].name);
1034 				count++;
1035 			}
1036 		}
1037 		return count;
1038 	}
1039 	return nstats;
1040 }
1041 
1042 static int
1043 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1044 		      unsigned n)
1045 {
1046 	unsigned i;
1047 	unsigned count = 0;
1048 
1049 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1050 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1051 
1052 	if (n < nstats)
1053 		return nstats;
1054 
1055 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1056 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1057 
1058 		if (rxvq == NULL)
1059 			continue;
1060 
1061 		unsigned t;
1062 
1063 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1064 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1065 				rte_virtio_rxq_stat_strings[t].offset);
1066 			xstats[count].id = count;
1067 			count++;
1068 		}
1069 	}
1070 
1071 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1072 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1073 
1074 		if (txvq == NULL)
1075 			continue;
1076 
1077 		unsigned t;
1078 
1079 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1080 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1081 				rte_virtio_txq_stat_strings[t].offset);
1082 			xstats[count].id = count;
1083 			count++;
1084 		}
1085 	}
1086 
1087 	return count;
1088 }
1089 
1090 static int
1091 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1092 {
1093 	virtio_update_stats(dev, stats);
1094 
1095 	return 0;
1096 }
1097 
1098 static int
1099 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1100 {
1101 	unsigned int i;
1102 
1103 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1104 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1105 		if (txvq == NULL)
1106 			continue;
1107 
1108 		txvq->stats.packets = 0;
1109 		txvq->stats.bytes = 0;
1110 		txvq->stats.multicast = 0;
1111 		txvq->stats.broadcast = 0;
1112 		memset(txvq->stats.size_bins, 0,
1113 		       sizeof(txvq->stats.size_bins[0]) * 8);
1114 	}
1115 
1116 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1117 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1118 		if (rxvq == NULL)
1119 			continue;
1120 
1121 		rxvq->stats.packets = 0;
1122 		rxvq->stats.bytes = 0;
1123 		rxvq->stats.errors = 0;
1124 		rxvq->stats.multicast = 0;
1125 		rxvq->stats.broadcast = 0;
1126 		memset(rxvq->stats.size_bins, 0,
1127 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1128 	}
1129 
1130 	return 0;
1131 }
1132 
1133 static void
1134 virtio_set_hwaddr(struct virtio_hw *hw)
1135 {
1136 	vtpci_write_dev_config(hw,
1137 			offsetof(struct virtio_net_config, mac),
1138 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1139 }
1140 
1141 static void
1142 virtio_get_hwaddr(struct virtio_hw *hw)
1143 {
1144 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1145 		vtpci_read_dev_config(hw,
1146 			offsetof(struct virtio_net_config, mac),
1147 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1148 	} else {
1149 		rte_eth_random_addr(&hw->mac_addr[0]);
1150 		virtio_set_hwaddr(hw);
1151 	}
1152 }
1153 
1154 static int
1155 virtio_mac_table_set(struct virtio_hw *hw,
1156 		     const struct virtio_net_ctrl_mac *uc,
1157 		     const struct virtio_net_ctrl_mac *mc)
1158 {
1159 	struct virtio_pmd_ctrl ctrl;
1160 	int err, len[2];
1161 
1162 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1163 		PMD_DRV_LOG(INFO, "host does not support mac table");
1164 		return -1;
1165 	}
1166 
1167 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1168 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1169 
1170 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1171 	memcpy(ctrl.data, uc, len[0]);
1172 
1173 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1174 	memcpy(ctrl.data + len[0], mc, len[1]);
1175 
1176 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1177 	if (err != 0)
1178 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1179 	return err;
1180 }
1181 
1182 static int
1183 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1184 		    uint32_t index, uint32_t vmdq __rte_unused)
1185 {
1186 	struct virtio_hw *hw = dev->data->dev_private;
1187 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1188 	unsigned int i;
1189 	struct virtio_net_ctrl_mac *uc, *mc;
1190 
1191 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1192 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1193 		return -EINVAL;
1194 	}
1195 
1196 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1197 		sizeof(uc->entries));
1198 	uc->entries = 0;
1199 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1200 		sizeof(mc->entries));
1201 	mc->entries = 0;
1202 
1203 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1204 		const struct rte_ether_addr *addr
1205 			= (i == index) ? mac_addr : addrs + i;
1206 		struct virtio_net_ctrl_mac *tbl
1207 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1208 
1209 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1210 	}
1211 
1212 	return virtio_mac_table_set(hw, uc, mc);
1213 }
1214 
1215 static void
1216 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1217 {
1218 	struct virtio_hw *hw = dev->data->dev_private;
1219 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1220 	struct virtio_net_ctrl_mac *uc, *mc;
1221 	unsigned int i;
1222 
1223 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1224 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1225 		return;
1226 	}
1227 
1228 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1229 		sizeof(uc->entries));
1230 	uc->entries = 0;
1231 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1232 		sizeof(mc->entries));
1233 	mc->entries = 0;
1234 
1235 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1236 		struct virtio_net_ctrl_mac *tbl;
1237 
1238 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1239 			continue;
1240 
1241 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1242 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1243 			RTE_ETHER_ADDR_LEN);
1244 	}
1245 
1246 	virtio_mac_table_set(hw, uc, mc);
1247 }
1248 
1249 static int
1250 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1251 {
1252 	struct virtio_hw *hw = dev->data->dev_private;
1253 
1254 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1255 
1256 	/* Use atomic update if available */
1257 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1258 		struct virtio_pmd_ctrl ctrl;
1259 		int len = RTE_ETHER_ADDR_LEN;
1260 
1261 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1262 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1263 
1264 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1265 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1266 	}
1267 
1268 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1269 		return -ENOTSUP;
1270 
1271 	virtio_set_hwaddr(hw);
1272 	return 0;
1273 }
1274 
1275 static int
1276 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1277 {
1278 	struct virtio_hw *hw = dev->data->dev_private;
1279 	struct virtio_pmd_ctrl ctrl;
1280 	int len;
1281 
1282 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1283 		return -ENOTSUP;
1284 
1285 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1286 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1287 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1288 	len = sizeof(vlan_id);
1289 
1290 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1291 }
1292 
1293 static int
1294 virtio_intr_unmask(struct rte_eth_dev *dev)
1295 {
1296 	struct virtio_hw *hw = dev->data->dev_private;
1297 
1298 	if (rte_intr_ack(dev->intr_handle) < 0)
1299 		return -1;
1300 
1301 	if (!hw->virtio_user_dev)
1302 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1303 
1304 	return 0;
1305 }
1306 
1307 static int
1308 virtio_intr_enable(struct rte_eth_dev *dev)
1309 {
1310 	struct virtio_hw *hw = dev->data->dev_private;
1311 
1312 	if (rte_intr_enable(dev->intr_handle) < 0)
1313 		return -1;
1314 
1315 	if (!hw->virtio_user_dev)
1316 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1317 
1318 	return 0;
1319 }
1320 
1321 static int
1322 virtio_intr_disable(struct rte_eth_dev *dev)
1323 {
1324 	struct virtio_hw *hw = dev->data->dev_private;
1325 
1326 	if (rte_intr_disable(dev->intr_handle) < 0)
1327 		return -1;
1328 
1329 	if (!hw->virtio_user_dev)
1330 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1331 
1332 	return 0;
1333 }
1334 
1335 static int
1336 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1337 {
1338 	uint64_t host_features;
1339 
1340 	/* Prepare guest_features: feature that driver wants to support */
1341 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1342 		req_features);
1343 
1344 	/* Read device(host) feature bits */
1345 	host_features = VTPCI_OPS(hw)->get_features(hw);
1346 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1347 		host_features);
1348 
1349 	/* If supported, ensure MTU value is valid before acknowledging it. */
1350 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1351 		struct virtio_net_config config;
1352 
1353 		vtpci_read_dev_config(hw,
1354 			offsetof(struct virtio_net_config, mtu),
1355 			&config.mtu, sizeof(config.mtu));
1356 
1357 		if (config.mtu < RTE_ETHER_MIN_MTU)
1358 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1359 	}
1360 
1361 	/*
1362 	 * Negotiate features: Subset of device feature bits are written back
1363 	 * guest feature bits.
1364 	 */
1365 	hw->guest_features = req_features;
1366 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1367 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1368 		hw->guest_features);
1369 
1370 	if (hw->modern && !vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1371 		PMD_INIT_LOG(ERR,
1372 			"VIRTIO_F_VERSION_1 features is not enabled.");
1373 		return -1;
1374 	}
1375 
1376 	if (hw->modern || hw->virtio_user_dev) {
1377 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1378 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1379 			PMD_INIT_LOG(ERR,
1380 				"failed to set FEATURES_OK status!");
1381 			return -1;
1382 		}
1383 	}
1384 
1385 	hw->req_guest_features = req_features;
1386 
1387 	return 0;
1388 }
1389 
1390 int
1391 virtio_dev_pause(struct rte_eth_dev *dev)
1392 {
1393 	struct virtio_hw *hw = dev->data->dev_private;
1394 
1395 	rte_spinlock_lock(&hw->state_lock);
1396 
1397 	if (hw->started == 0) {
1398 		/* Device is just stopped. */
1399 		rte_spinlock_unlock(&hw->state_lock);
1400 		return -1;
1401 	}
1402 	hw->started = 0;
1403 	/*
1404 	 * Prevent the worker threads from touching queues to avoid contention,
1405 	 * 1 ms should be enough for the ongoing Tx function to finish.
1406 	 */
1407 	rte_delay_ms(1);
1408 	return 0;
1409 }
1410 
1411 /*
1412  * Recover hw state to let the worker threads continue.
1413  */
1414 void
1415 virtio_dev_resume(struct rte_eth_dev *dev)
1416 {
1417 	struct virtio_hw *hw = dev->data->dev_private;
1418 
1419 	hw->started = 1;
1420 	rte_spinlock_unlock(&hw->state_lock);
1421 }
1422 
1423 /*
1424  * Should be called only after device is paused.
1425  */
1426 int
1427 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1428 		int nb_pkts)
1429 {
1430 	struct virtio_hw *hw = dev->data->dev_private;
1431 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1432 	int ret;
1433 
1434 	hw->inject_pkts = tx_pkts;
1435 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1436 	hw->inject_pkts = NULL;
1437 
1438 	return ret;
1439 }
1440 
1441 static void
1442 virtio_notify_peers(struct rte_eth_dev *dev)
1443 {
1444 	struct virtio_hw *hw = dev->data->dev_private;
1445 	struct virtnet_rx *rxvq;
1446 	struct rte_mbuf *rarp_mbuf;
1447 
1448 	if (!dev->data->rx_queues)
1449 		return;
1450 
1451 	rxvq = dev->data->rx_queues[0];
1452 	if (!rxvq)
1453 		return;
1454 
1455 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1456 			(struct rte_ether_addr *)hw->mac_addr);
1457 	if (rarp_mbuf == NULL) {
1458 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1459 		return;
1460 	}
1461 
1462 	/* If virtio port just stopped, no need to send RARP */
1463 	if (virtio_dev_pause(dev) < 0) {
1464 		rte_pktmbuf_free(rarp_mbuf);
1465 		return;
1466 	}
1467 
1468 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1469 	virtio_dev_resume(dev);
1470 }
1471 
1472 static void
1473 virtio_ack_link_announce(struct rte_eth_dev *dev)
1474 {
1475 	struct virtio_hw *hw = dev->data->dev_private;
1476 	struct virtio_pmd_ctrl ctrl;
1477 
1478 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1479 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1480 
1481 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1482 }
1483 
1484 /*
1485  * Process virtio config changed interrupt. Call the callback
1486  * if link state changed, generate gratuitous RARP packet if
1487  * the status indicates an ANNOUNCE.
1488  */
1489 void
1490 virtio_interrupt_handler(void *param)
1491 {
1492 	struct rte_eth_dev *dev = param;
1493 	struct virtio_hw *hw = dev->data->dev_private;
1494 	uint8_t isr;
1495 	uint16_t status;
1496 
1497 	/* Read interrupt status which clears interrupt */
1498 	isr = vtpci_isr(hw);
1499 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1500 
1501 	if (virtio_intr_unmask(dev) < 0)
1502 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1503 
1504 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1505 		if (virtio_dev_link_update(dev, 0) == 0)
1506 			rte_eth_dev_callback_process(dev,
1507 						     RTE_ETH_EVENT_INTR_LSC,
1508 						     NULL);
1509 
1510 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1511 			vtpci_read_dev_config(hw,
1512 				offsetof(struct virtio_net_config, status),
1513 				&status, sizeof(status));
1514 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1515 				virtio_notify_peers(dev);
1516 				if (hw->cvq)
1517 					virtio_ack_link_announce(dev);
1518 			}
1519 		}
1520 	}
1521 }
1522 
1523 /* set rx and tx handlers according to what is supported */
1524 static void
1525 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1526 {
1527 	struct virtio_hw *hw = eth_dev->data->dev_private;
1528 
1529 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1530 	if (vtpci_packed_queue(hw)) {
1531 		PMD_INIT_LOG(INFO,
1532 			"virtio: using packed ring %s Tx path on port %u",
1533 			hw->use_vec_tx ? "vectorized" : "standard",
1534 			eth_dev->data->port_id);
1535 		if (hw->use_vec_tx)
1536 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1537 		else
1538 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1539 	} else {
1540 		if (hw->use_inorder_tx) {
1541 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1542 				eth_dev->data->port_id);
1543 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1544 		} else {
1545 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1546 				eth_dev->data->port_id);
1547 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1548 		}
1549 	}
1550 
1551 	if (vtpci_packed_queue(hw)) {
1552 		if (hw->use_vec_rx) {
1553 			PMD_INIT_LOG(INFO,
1554 				"virtio: using packed ring vectorized Rx path on port %u",
1555 				eth_dev->data->port_id);
1556 			eth_dev->rx_pkt_burst =
1557 				&virtio_recv_pkts_packed_vec;
1558 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1559 			PMD_INIT_LOG(INFO,
1560 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1561 				eth_dev->data->port_id);
1562 			eth_dev->rx_pkt_burst =
1563 				&virtio_recv_mergeable_pkts_packed;
1564 		} else {
1565 			PMD_INIT_LOG(INFO,
1566 				"virtio: using packed ring standard Rx path on port %u",
1567 				eth_dev->data->port_id);
1568 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1569 		}
1570 	} else {
1571 		if (hw->use_vec_rx) {
1572 			PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1573 				eth_dev->data->port_id);
1574 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1575 		} else if (hw->use_inorder_rx) {
1576 			PMD_INIT_LOG(INFO,
1577 				"virtio: using inorder Rx path on port %u",
1578 				eth_dev->data->port_id);
1579 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1580 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1581 			PMD_INIT_LOG(INFO,
1582 				"virtio: using mergeable buffer Rx path on port %u",
1583 				eth_dev->data->port_id);
1584 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1585 		} else {
1586 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1587 				eth_dev->data->port_id);
1588 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1589 		}
1590 	}
1591 
1592 }
1593 
1594 /* Only support 1:1 queue/interrupt mapping so far.
1595  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1596  * interrupt vectors (<N+1).
1597  */
1598 static int
1599 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1600 {
1601 	uint32_t i;
1602 	struct virtio_hw *hw = dev->data->dev_private;
1603 
1604 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1605 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1606 		dev->intr_handle->intr_vec[i] = i + 1;
1607 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1608 						 VIRTIO_MSI_NO_VECTOR) {
1609 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1610 			return -EBUSY;
1611 		}
1612 	}
1613 
1614 	return 0;
1615 }
1616 
1617 static void
1618 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1619 {
1620 	uint32_t i;
1621 	struct virtio_hw *hw = dev->data->dev_private;
1622 
1623 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1624 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1625 		VTPCI_OPS(hw)->set_queue_irq(hw,
1626 					     hw->vqs[i * VTNET_CQ],
1627 					     VIRTIO_MSI_NO_VECTOR);
1628 }
1629 
1630 static int
1631 virtio_configure_intr(struct rte_eth_dev *dev)
1632 {
1633 	struct virtio_hw *hw = dev->data->dev_private;
1634 
1635 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1636 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1637 		return -ENOTSUP;
1638 	}
1639 
1640 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1641 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1642 		return -1;
1643 	}
1644 
1645 	if (!dev->intr_handle->intr_vec) {
1646 		dev->intr_handle->intr_vec =
1647 			rte_zmalloc("intr_vec",
1648 				    hw->max_queue_pairs * sizeof(int), 0);
1649 		if (!dev->intr_handle->intr_vec) {
1650 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1651 				     hw->max_queue_pairs);
1652 			return -ENOMEM;
1653 		}
1654 	}
1655 
1656 	/* Re-register callback to update max_intr */
1657 	rte_intr_callback_unregister(dev->intr_handle,
1658 				     virtio_interrupt_handler,
1659 				     dev);
1660 	rte_intr_callback_register(dev->intr_handle,
1661 				   virtio_interrupt_handler,
1662 				   dev);
1663 
1664 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1665 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1666 	 * And for legacy devices, this should be done before queue/vec binding
1667 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1668 	 * (22) will be ignored.
1669 	 */
1670 	if (virtio_intr_enable(dev) < 0) {
1671 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1672 		return -1;
1673 	}
1674 
1675 	if (virtio_queues_bind_intr(dev) < 0) {
1676 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1677 		return -1;
1678 	}
1679 
1680 	return 0;
1681 }
1682 #define DUPLEX_UNKNOWN   0xff
1683 /* reset device and renegotiate features if needed */
1684 static int
1685 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1686 {
1687 	struct virtio_hw *hw = eth_dev->data->dev_private;
1688 	struct virtio_net_config *config;
1689 	struct virtio_net_config local_config;
1690 	struct rte_pci_device *pci_dev = NULL;
1691 	int ret;
1692 
1693 	/* Reset the device although not necessary at startup */
1694 	vtpci_reset(hw);
1695 
1696 	if (hw->vqs) {
1697 		virtio_dev_free_mbufs(eth_dev);
1698 		virtio_free_queues(hw);
1699 	}
1700 
1701 	/* Tell the host we've noticed this device. */
1702 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1703 
1704 	/* Tell the host we've known how to drive the device. */
1705 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1706 	if (virtio_negotiate_features(hw, req_features) < 0)
1707 		return -1;
1708 
1709 	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1710 
1711 	if (!hw->virtio_user_dev)
1712 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1713 
1714 	/* If host does not support both status and MSI-X then disable LSC */
1715 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1716 	    hw->use_msix != VIRTIO_MSIX_NONE)
1717 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1718 	else
1719 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1720 
1721 	/* Setting up rx_header size for the device */
1722 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1723 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1724 	    vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1725 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1726 	else
1727 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1728 
1729 	/* Copy the permanent MAC address to: virtio_hw */
1730 	virtio_get_hwaddr(hw);
1731 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1732 			&eth_dev->data->mac_addrs[0]);
1733 	PMD_INIT_LOG(DEBUG,
1734 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1735 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1736 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1737 
1738 	if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1739 		if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1740 			config = &local_config;
1741 			vtpci_read_dev_config(hw,
1742 				offsetof(struct virtio_net_config, speed),
1743 				&config->speed, sizeof(config->speed));
1744 			vtpci_read_dev_config(hw,
1745 				offsetof(struct virtio_net_config, duplex),
1746 				&config->duplex, sizeof(config->duplex));
1747 			hw->speed = config->speed;
1748 			hw->duplex = config->duplex;
1749 		}
1750 	}
1751 	if (hw->duplex == DUPLEX_UNKNOWN)
1752 		hw->duplex = ETH_LINK_FULL_DUPLEX;
1753 	PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1754 		hw->speed, hw->duplex);
1755 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1756 		config = &local_config;
1757 
1758 		vtpci_read_dev_config(hw,
1759 			offsetof(struct virtio_net_config, mac),
1760 			&config->mac, sizeof(config->mac));
1761 
1762 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1763 			vtpci_read_dev_config(hw,
1764 				offsetof(struct virtio_net_config, status),
1765 				&config->status, sizeof(config->status));
1766 		} else {
1767 			PMD_INIT_LOG(DEBUG,
1768 				     "VIRTIO_NET_F_STATUS is not supported");
1769 			config->status = 0;
1770 		}
1771 
1772 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1773 			vtpci_read_dev_config(hw,
1774 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1775 				&config->max_virtqueue_pairs,
1776 				sizeof(config->max_virtqueue_pairs));
1777 		} else {
1778 			PMD_INIT_LOG(DEBUG,
1779 				     "VIRTIO_NET_F_MQ is not supported");
1780 			config->max_virtqueue_pairs = 1;
1781 		}
1782 
1783 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1784 
1785 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1786 			vtpci_read_dev_config(hw,
1787 				offsetof(struct virtio_net_config, mtu),
1788 				&config->mtu,
1789 				sizeof(config->mtu));
1790 
1791 			/*
1792 			 * MTU value has already been checked at negotiation
1793 			 * time, but check again in case it has changed since
1794 			 * then, which should not happen.
1795 			 */
1796 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1797 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1798 						config->mtu);
1799 				return -1;
1800 			}
1801 
1802 			hw->max_mtu = config->mtu;
1803 			/* Set initial MTU to maximum one supported by vhost */
1804 			eth_dev->data->mtu = config->mtu;
1805 
1806 		} else {
1807 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1808 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1809 		}
1810 
1811 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1812 				config->max_virtqueue_pairs);
1813 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1814 		PMD_INIT_LOG(DEBUG,
1815 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1816 				config->mac[0], config->mac[1],
1817 				config->mac[2], config->mac[3],
1818 				config->mac[4], config->mac[5]);
1819 	} else {
1820 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1821 		hw->max_queue_pairs = 1;
1822 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1823 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1824 	}
1825 
1826 	ret = virtio_alloc_queues(eth_dev);
1827 	if (ret < 0)
1828 		return ret;
1829 
1830 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1831 		if (virtio_configure_intr(eth_dev) < 0) {
1832 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1833 			virtio_free_queues(hw);
1834 			return -1;
1835 		}
1836 	}
1837 
1838 	vtpci_reinit_complete(hw);
1839 
1840 	if (pci_dev)
1841 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1842 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1843 			pci_dev->id.device_id);
1844 
1845 	return 0;
1846 }
1847 
1848 /*
1849  * Remap the PCI device again (IO port map for legacy device and
1850  * memory map for modern device), so that the secondary process
1851  * could have the PCI initiated correctly.
1852  */
1853 static int
1854 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1855 {
1856 	if (hw->modern) {
1857 		/*
1858 		 * We don't have to re-parse the PCI config space, since
1859 		 * rte_pci_map_device() makes sure the mapped address
1860 		 * in secondary process would equal to the one mapped in
1861 		 * the primary process: error will be returned if that
1862 		 * requirement is not met.
1863 		 *
1864 		 * That said, we could simply reuse all cap pointers
1865 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1866 		 * primary process, which is stored in shared memory.
1867 		 */
1868 		if (rte_pci_map_device(pci_dev)) {
1869 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1870 			return -1;
1871 		}
1872 	} else {
1873 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1874 			return -1;
1875 	}
1876 
1877 	return 0;
1878 }
1879 
1880 static void
1881 virtio_set_vtpci_ops(struct virtio_hw *hw)
1882 {
1883 #ifdef RTE_VIRTIO_USER
1884 	if (hw->virtio_user_dev)
1885 		VTPCI_OPS(hw) = &virtio_user_ops;
1886 	else
1887 #endif
1888 	if (hw->modern)
1889 		VTPCI_OPS(hw) = &modern_ops;
1890 	else
1891 		VTPCI_OPS(hw) = &legacy_ops;
1892 }
1893 
1894 /*
1895  * This function is based on probe() function in virtio_pci.c
1896  * It returns 0 on success.
1897  */
1898 int
1899 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1900 {
1901 	struct virtio_hw *hw = eth_dev->data->dev_private;
1902 	uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1903 	int vectorized = 0;
1904 	int ret;
1905 
1906 	if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1907 		PMD_INIT_LOG(ERR,
1908 			"Not sufficient headroom required = %d, avail = %d",
1909 			(int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1910 			RTE_PKTMBUF_HEADROOM);
1911 
1912 		return -1;
1913 	}
1914 
1915 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1916 	eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1917 
1918 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1919 		if (!hw->virtio_user_dev) {
1920 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1921 			if (ret)
1922 				return ret;
1923 		}
1924 
1925 		virtio_set_vtpci_ops(hw);
1926 		set_rxtx_funcs(eth_dev);
1927 
1928 		return 0;
1929 	}
1930 	ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
1931 		 NULL, &speed, &vectorized);
1932 	if (ret < 0)
1933 		return ret;
1934 	hw->speed = speed;
1935 
1936 	/* Allocate memory for storing MAC addresses */
1937 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1938 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1939 	if (eth_dev->data->mac_addrs == NULL) {
1940 		PMD_INIT_LOG(ERR,
1941 			"Failed to allocate %d bytes needed to store MAC addresses",
1942 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1943 		return -ENOMEM;
1944 	}
1945 
1946 	hw->port_id = eth_dev->data->port_id;
1947 	/* For virtio_user case the hw->virtio_user_dev is populated by
1948 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1949 	 */
1950 	if (!hw->virtio_user_dev) {
1951 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1952 		if (ret)
1953 			goto err_vtpci_init;
1954 	}
1955 
1956 	rte_spinlock_init(&hw->state_lock);
1957 
1958 	/* reset device and negotiate default features */
1959 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1960 	if (ret < 0)
1961 		goto err_virtio_init;
1962 
1963 	if (vectorized) {
1964 		if (!vtpci_packed_queue(hw)) {
1965 			hw->use_vec_rx = 1;
1966 		} else {
1967 #if !defined(CC_AVX512_SUPPORT)
1968 			PMD_DRV_LOG(INFO,
1969 				"building environment do not support packed ring vectorized");
1970 #else
1971 			hw->use_vec_rx = 1;
1972 			hw->use_vec_tx = 1;
1973 #endif
1974 		}
1975 	}
1976 
1977 	hw->opened = true;
1978 
1979 	return 0;
1980 
1981 err_virtio_init:
1982 	if (!hw->virtio_user_dev) {
1983 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1984 		if (!hw->modern)
1985 			rte_pci_ioport_unmap(VTPCI_IO(hw));
1986 	}
1987 err_vtpci_init:
1988 	rte_free(eth_dev->data->mac_addrs);
1989 	eth_dev->data->mac_addrs = NULL;
1990 	return ret;
1991 }
1992 
1993 static int
1994 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1995 {
1996 	PMD_INIT_FUNC_TRACE();
1997 
1998 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1999 		return 0;
2000 
2001 	virtio_dev_stop(eth_dev);
2002 	virtio_dev_close(eth_dev);
2003 
2004 	eth_dev->dev_ops = NULL;
2005 	eth_dev->tx_pkt_burst = NULL;
2006 	eth_dev->rx_pkt_burst = NULL;
2007 
2008 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
2009 
2010 	return 0;
2011 }
2012 
2013 
2014 static int vdpa_check_handler(__rte_unused const char *key,
2015 		const char *value, void *ret_val)
2016 {
2017 	if (strcmp(value, "1") == 0)
2018 		*(int *)ret_val = 1;
2019 	else
2020 		*(int *)ret_val = 0;
2021 
2022 	return 0;
2023 }
2024 
2025 
2026 static uint32_t
2027 virtio_dev_speed_capa_get(uint32_t speed)
2028 {
2029 	switch (speed) {
2030 	case ETH_SPEED_NUM_10G:
2031 		return ETH_LINK_SPEED_10G;
2032 	case ETH_SPEED_NUM_20G:
2033 		return ETH_LINK_SPEED_20G;
2034 	case ETH_SPEED_NUM_25G:
2035 		return ETH_LINK_SPEED_25G;
2036 	case ETH_SPEED_NUM_40G:
2037 		return ETH_LINK_SPEED_40G;
2038 	case ETH_SPEED_NUM_50G:
2039 		return ETH_LINK_SPEED_50G;
2040 	case ETH_SPEED_NUM_56G:
2041 		return ETH_LINK_SPEED_56G;
2042 	case ETH_SPEED_NUM_100G:
2043 		return ETH_LINK_SPEED_100G;
2044 	case ETH_SPEED_NUM_200G:
2045 		return ETH_LINK_SPEED_200G;
2046 	default:
2047 		return 0;
2048 	}
2049 }
2050 
2051 static int vectorized_check_handler(__rte_unused const char *key,
2052 		const char *value, void *ret_val)
2053 {
2054 	if (strcmp(value, "1") == 0)
2055 		*(int *)ret_val = 1;
2056 	else
2057 		*(int *)ret_val = 0;
2058 
2059 	return 0;
2060 }
2061 
2062 #define VIRTIO_ARG_SPEED      "speed"
2063 #define VIRTIO_ARG_VDPA       "vdpa"
2064 #define VIRTIO_ARG_VECTORIZED "vectorized"
2065 
2066 
2067 static int
2068 link_speed_handler(const char *key __rte_unused,
2069 		const char *value, void *ret_val)
2070 {
2071 	uint32_t val;
2072 	if (!value || !ret_val)
2073 		return -EINVAL;
2074 	val = strtoul(value, NULL, 0);
2075 	/* validate input */
2076 	if (virtio_dev_speed_capa_get(val) == 0)
2077 		return -EINVAL;
2078 	*(uint32_t *)ret_val = val;
2079 
2080 	return 0;
2081 }
2082 
2083 
2084 static int
2085 virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
2086 	uint32_t *speed, int *vectorized)
2087 {
2088 	struct rte_kvargs *kvlist;
2089 	int ret = 0;
2090 
2091 	if (devargs == NULL)
2092 		return 0;
2093 
2094 	kvlist = rte_kvargs_parse(devargs->args, NULL);
2095 	if (kvlist == NULL) {
2096 		PMD_INIT_LOG(ERR, "error when parsing param");
2097 		return 0;
2098 	}
2099 	if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
2100 		/* vdpa mode selected when there's a key-value pair:
2101 		 * vdpa=1
2102 		 */
2103 		ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
2104 				vdpa_check_handler, vdpa);
2105 		if (ret < 0) {
2106 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2107 				VIRTIO_ARG_VDPA);
2108 			goto exit;
2109 		}
2110 	}
2111 	if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2112 		ret = rte_kvargs_process(kvlist,
2113 					VIRTIO_ARG_SPEED,
2114 					link_speed_handler, speed);
2115 		if (ret < 0) {
2116 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2117 					VIRTIO_ARG_SPEED);
2118 			goto exit;
2119 		}
2120 	}
2121 
2122 	if (vectorized &&
2123 		rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2124 		ret = rte_kvargs_process(kvlist,
2125 				VIRTIO_ARG_VECTORIZED,
2126 				vectorized_check_handler, vectorized);
2127 		if (ret < 0) {
2128 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2129 					VIRTIO_ARG_VECTORIZED);
2130 			goto exit;
2131 		}
2132 	}
2133 
2134 exit:
2135 	rte_kvargs_free(kvlist);
2136 	return ret;
2137 }
2138 
2139 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2140 	struct rte_pci_device *pci_dev)
2141 {
2142 	int vdpa = 0;
2143 	int ret = 0;
2144 
2145 	ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL,
2146 		NULL);
2147 	if (ret < 0) {
2148 		PMD_INIT_LOG(ERR, "devargs parsing is failed");
2149 		return ret;
2150 	}
2151 	/* virtio pmd skips probe if device needs to work in vdpa mode */
2152 	if (vdpa == 1)
2153 		return 1;
2154 
2155 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2156 		eth_virtio_dev_init);
2157 }
2158 
2159 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2160 {
2161 	int ret;
2162 
2163 	ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2164 	/* Port has already been released by close. */
2165 	if (ret == -ENODEV)
2166 		ret = 0;
2167 	return ret;
2168 }
2169 
2170 static struct rte_pci_driver rte_virtio_pmd = {
2171 	.driver = {
2172 		.name = "net_virtio",
2173 	},
2174 	.id_table = pci_id_virtio_map,
2175 	.drv_flags = 0,
2176 	.probe = eth_virtio_pci_probe,
2177 	.remove = eth_virtio_pci_remove,
2178 };
2179 
2180 RTE_INIT(rte_virtio_pmd_init)
2181 {
2182 	rte_eal_iopl_init();
2183 	rte_pci_register(&rte_virtio_pmd);
2184 }
2185 
2186 static bool
2187 rx_offload_enabled(struct virtio_hw *hw)
2188 {
2189 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2190 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2191 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2192 }
2193 
2194 static bool
2195 tx_offload_enabled(struct virtio_hw *hw)
2196 {
2197 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2198 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2199 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2200 }
2201 
2202 /*
2203  * Configure virtio device
2204  * It returns 0 on success.
2205  */
2206 static int
2207 virtio_dev_configure(struct rte_eth_dev *dev)
2208 {
2209 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2210 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2211 	struct virtio_hw *hw = dev->data->dev_private;
2212 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2213 		hw->vtnet_hdr_size;
2214 	uint64_t rx_offloads = rxmode->offloads;
2215 	uint64_t tx_offloads = txmode->offloads;
2216 	uint64_t req_features;
2217 	int ret;
2218 
2219 	PMD_INIT_LOG(DEBUG, "configure");
2220 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2221 
2222 	if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2223 		PMD_DRV_LOG(ERR,
2224 			"Unsupported Rx multi queue mode %d",
2225 			rxmode->mq_mode);
2226 		return -EINVAL;
2227 	}
2228 
2229 	if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2230 		PMD_DRV_LOG(ERR,
2231 			"Unsupported Tx multi queue mode %d",
2232 			txmode->mq_mode);
2233 		return -EINVAL;
2234 	}
2235 
2236 	if (dev->data->dev_conf.intr_conf.rxq) {
2237 		ret = virtio_init_device(dev, hw->req_guest_features);
2238 		if (ret < 0)
2239 			return ret;
2240 	}
2241 
2242 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2243 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2244 
2245 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2246 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2247 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2248 
2249 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2250 		req_features |=
2251 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2252 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2253 
2254 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2255 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2256 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2257 
2258 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2259 		req_features |=
2260 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2261 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2262 
2263 	/* if request features changed, reinit the device */
2264 	if (req_features != hw->req_guest_features) {
2265 		ret = virtio_init_device(dev, req_features);
2266 		if (ret < 0)
2267 			return ret;
2268 	}
2269 
2270 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2271 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2272 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2273 		PMD_DRV_LOG(ERR,
2274 			"rx checksum not available on this host");
2275 		return -ENOTSUP;
2276 	}
2277 
2278 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2279 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2280 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2281 		PMD_DRV_LOG(ERR,
2282 			"Large Receive Offload not available on this host");
2283 		return -ENOTSUP;
2284 	}
2285 
2286 	/* start control queue */
2287 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2288 		virtio_dev_cq_start(dev);
2289 
2290 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2291 		hw->vlan_strip = 1;
2292 
2293 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2294 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2295 		PMD_DRV_LOG(ERR,
2296 			    "vlan filtering not available on this host");
2297 		return -ENOTSUP;
2298 	}
2299 
2300 	hw->has_tx_offload = tx_offload_enabled(hw);
2301 	hw->has_rx_offload = rx_offload_enabled(hw);
2302 
2303 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2304 		/* Enable vector (0) for Link State Intrerrupt */
2305 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2306 				VIRTIO_MSI_NO_VECTOR) {
2307 			PMD_DRV_LOG(ERR, "failed to set config vector");
2308 			return -EBUSY;
2309 		}
2310 
2311 	if (vtpci_packed_queue(hw)) {
2312 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2313 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2314 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2315 		     !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2316 		     !vtpci_with_feature(hw, VIRTIO_F_VERSION_1))) {
2317 			PMD_DRV_LOG(INFO,
2318 				"disabled packed ring vectorized path for requirements not met");
2319 			hw->use_vec_rx = 0;
2320 			hw->use_vec_tx = 0;
2321 		}
2322 #else
2323 		hw->use_vec_rx = 0;
2324 		hw->use_vec_tx = 0;
2325 #endif
2326 
2327 		if (hw->use_vec_rx) {
2328 			if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2329 				PMD_DRV_LOG(INFO,
2330 					"disabled packed ring vectorized rx for mrg_rxbuf enabled");
2331 				hw->use_vec_rx = 0;
2332 			}
2333 
2334 			if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2335 				PMD_DRV_LOG(INFO,
2336 					"disabled packed ring vectorized rx for TCP_LRO enabled");
2337 				hw->use_vec_rx = 0;
2338 			}
2339 		}
2340 	} else {
2341 		if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2342 			hw->use_inorder_tx = 1;
2343 			hw->use_inorder_rx = 1;
2344 			hw->use_vec_rx = 0;
2345 		}
2346 
2347 		if (hw->use_vec_rx) {
2348 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2349 			if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2350 				PMD_DRV_LOG(INFO,
2351 					"disabled split ring vectorized path for requirement not met");
2352 				hw->use_vec_rx = 0;
2353 			}
2354 #endif
2355 			if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2356 				PMD_DRV_LOG(INFO,
2357 					"disabled split ring vectorized rx for mrg_rxbuf enabled");
2358 				hw->use_vec_rx = 0;
2359 			}
2360 
2361 			if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2362 					   DEV_RX_OFFLOAD_TCP_CKSUM |
2363 					   DEV_RX_OFFLOAD_TCP_LRO |
2364 					   DEV_RX_OFFLOAD_VLAN_STRIP)) {
2365 				PMD_DRV_LOG(INFO,
2366 					"disabled split ring vectorized rx for offloading enabled");
2367 				hw->use_vec_rx = 0;
2368 			}
2369 		}
2370 	}
2371 
2372 	return 0;
2373 }
2374 
2375 
2376 static int
2377 virtio_dev_start(struct rte_eth_dev *dev)
2378 {
2379 	uint16_t nb_queues, i;
2380 	struct virtnet_rx *rxvq;
2381 	struct virtnet_tx *txvq __rte_unused;
2382 	struct virtio_hw *hw = dev->data->dev_private;
2383 	int ret;
2384 
2385 	/* Finish the initialization of the queues */
2386 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2387 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2388 		if (ret < 0)
2389 			return ret;
2390 	}
2391 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2392 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2393 		if (ret < 0)
2394 			return ret;
2395 	}
2396 
2397 	/* check if lsc interrupt feature is enabled */
2398 	if (dev->data->dev_conf.intr_conf.lsc) {
2399 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2400 			PMD_DRV_LOG(ERR, "link status not supported by host");
2401 			return -ENOTSUP;
2402 		}
2403 	}
2404 
2405 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2406 	 * in device configure, but it could be unmapped  when device is
2407 	 * stopped.
2408 	 */
2409 	if (dev->data->dev_conf.intr_conf.lsc ||
2410 	    dev->data->dev_conf.intr_conf.rxq) {
2411 		virtio_intr_disable(dev);
2412 
2413 		/* Setup interrupt callback  */
2414 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2415 			rte_intr_callback_register(dev->intr_handle,
2416 						   virtio_interrupt_handler,
2417 						   dev);
2418 
2419 		if (virtio_intr_enable(dev) < 0) {
2420 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2421 			return -EIO;
2422 		}
2423 	}
2424 
2425 	/*Notify the backend
2426 	 *Otherwise the tap backend might already stop its queue due to fullness.
2427 	 *vhost backend will have no chance to be waked up
2428 	 */
2429 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2430 	if (hw->max_queue_pairs > 1) {
2431 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2432 			return -EINVAL;
2433 	}
2434 
2435 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2436 
2437 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2438 		rxvq = dev->data->rx_queues[i];
2439 		/* Flush the old packets */
2440 		virtqueue_rxvq_flush(rxvq->vq);
2441 		virtqueue_notify(rxvq->vq);
2442 	}
2443 
2444 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2445 		txvq = dev->data->tx_queues[i];
2446 		virtqueue_notify(txvq->vq);
2447 	}
2448 
2449 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2450 
2451 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2452 		rxvq = dev->data->rx_queues[i];
2453 		VIRTQUEUE_DUMP(rxvq->vq);
2454 	}
2455 
2456 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2457 		txvq = dev->data->tx_queues[i];
2458 		VIRTQUEUE_DUMP(txvq->vq);
2459 	}
2460 
2461 	set_rxtx_funcs(dev);
2462 	hw->started = true;
2463 
2464 	/* Initialize Link state */
2465 	virtio_dev_link_update(dev, 0);
2466 
2467 	return 0;
2468 }
2469 
2470 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2471 {
2472 	struct virtio_hw *hw = dev->data->dev_private;
2473 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2474 	const char *type __rte_unused;
2475 	unsigned int i, mbuf_num = 0;
2476 	struct virtqueue *vq;
2477 	struct rte_mbuf *buf;
2478 	int queue_type;
2479 
2480 	if (hw->vqs == NULL)
2481 		return;
2482 
2483 	for (i = 0; i < nr_vq; i++) {
2484 		vq = hw->vqs[i];
2485 		if (!vq)
2486 			continue;
2487 
2488 		queue_type = virtio_get_queue_type(hw, i);
2489 		if (queue_type == VTNET_RQ)
2490 			type = "rxq";
2491 		else if (queue_type == VTNET_TQ)
2492 			type = "txq";
2493 		else
2494 			continue;
2495 
2496 		PMD_INIT_LOG(DEBUG,
2497 			"Before freeing %s[%d] used and unused buf",
2498 			type, i);
2499 		VIRTQUEUE_DUMP(vq);
2500 
2501 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2502 			rte_pktmbuf_free(buf);
2503 			mbuf_num++;
2504 		}
2505 
2506 		PMD_INIT_LOG(DEBUG,
2507 			"After freeing %s[%d] used and unused buf",
2508 			type, i);
2509 		VIRTQUEUE_DUMP(vq);
2510 	}
2511 
2512 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2513 }
2514 
2515 /*
2516  * Stop device: disable interrupt and mark link down
2517  */
2518 static void
2519 virtio_dev_stop(struct rte_eth_dev *dev)
2520 {
2521 	struct virtio_hw *hw = dev->data->dev_private;
2522 	struct rte_eth_link link;
2523 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2524 
2525 	PMD_INIT_LOG(DEBUG, "stop");
2526 
2527 	rte_spinlock_lock(&hw->state_lock);
2528 	if (!hw->started)
2529 		goto out_unlock;
2530 	hw->started = false;
2531 
2532 	if (intr_conf->lsc || intr_conf->rxq) {
2533 		virtio_intr_disable(dev);
2534 
2535 		/* Reset interrupt callback  */
2536 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2537 			rte_intr_callback_unregister(dev->intr_handle,
2538 						     virtio_interrupt_handler,
2539 						     dev);
2540 		}
2541 	}
2542 
2543 	memset(&link, 0, sizeof(link));
2544 	rte_eth_linkstatus_set(dev, &link);
2545 out_unlock:
2546 	rte_spinlock_unlock(&hw->state_lock);
2547 }
2548 
2549 static int
2550 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2551 {
2552 	struct rte_eth_link link;
2553 	uint16_t status;
2554 	struct virtio_hw *hw = dev->data->dev_private;
2555 
2556 	memset(&link, 0, sizeof(link));
2557 	link.link_duplex = hw->duplex;
2558 	link.link_speed  = hw->speed;
2559 	link.link_autoneg = ETH_LINK_AUTONEG;
2560 
2561 	if (!hw->started) {
2562 		link.link_status = ETH_LINK_DOWN;
2563 		link.link_speed = ETH_SPEED_NUM_NONE;
2564 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2565 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2566 		vtpci_read_dev_config(hw,
2567 				offsetof(struct virtio_net_config, status),
2568 				&status, sizeof(status));
2569 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2570 			link.link_status = ETH_LINK_DOWN;
2571 			link.link_speed = ETH_SPEED_NUM_NONE;
2572 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2573 				     dev->data->port_id);
2574 		} else {
2575 			link.link_status = ETH_LINK_UP;
2576 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2577 				     dev->data->port_id);
2578 		}
2579 	} else {
2580 		link.link_status = ETH_LINK_UP;
2581 	}
2582 
2583 	return rte_eth_linkstatus_set(dev, &link);
2584 }
2585 
2586 static int
2587 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2588 {
2589 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2590 	struct virtio_hw *hw = dev->data->dev_private;
2591 	uint64_t offloads = rxmode->offloads;
2592 
2593 	if (mask & ETH_VLAN_FILTER_MASK) {
2594 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2595 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2596 
2597 			PMD_DRV_LOG(NOTICE,
2598 				"vlan filtering not available on this host");
2599 
2600 			return -ENOTSUP;
2601 		}
2602 	}
2603 
2604 	if (mask & ETH_VLAN_STRIP_MASK)
2605 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2606 
2607 	return 0;
2608 }
2609 
2610 static int
2611 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2612 {
2613 	uint64_t tso_mask, host_features;
2614 	struct virtio_hw *hw = dev->data->dev_private;
2615 	dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2616 
2617 	dev_info->max_rx_queues =
2618 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2619 	dev_info->max_tx_queues =
2620 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2621 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2622 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2623 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2624 
2625 	host_features = VTPCI_OPS(hw)->get_features(hw);
2626 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2627 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2628 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2629 		dev_info->rx_offload_capa |=
2630 			DEV_RX_OFFLOAD_TCP_CKSUM |
2631 			DEV_RX_OFFLOAD_UDP_CKSUM;
2632 	}
2633 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2634 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2635 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2636 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2637 	if ((host_features & tso_mask) == tso_mask)
2638 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2639 
2640 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2641 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2642 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2643 		dev_info->tx_offload_capa |=
2644 			DEV_TX_OFFLOAD_UDP_CKSUM |
2645 			DEV_TX_OFFLOAD_TCP_CKSUM;
2646 	}
2647 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2648 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2649 	if ((host_features & tso_mask) == tso_mask)
2650 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2651 
2652 	return 0;
2653 }
2654 
2655 /*
2656  * It enables testpmd to collect per queue stats.
2657  */
2658 static int
2659 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2660 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2661 __rte_unused uint8_t is_rx)
2662 {
2663 	return 0;
2664 }
2665 
2666 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2667 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2668 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2669 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2670 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);
2671