xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision c81e3f21d1ea05e5123278b15d9d5e1257b6ba99)
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 int virtio_logtype_init;
89 int virtio_logtype_driver;
90 
91 static void virtio_notify_peers(struct rte_eth_dev *dev);
92 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
93 
94 /*
95  * The set of PCI devices this driver supports
96  */
97 static const struct rte_pci_id pci_id_virtio_map[] = {
98 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
99 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
100 	{ .vendor_id = 0, /* sentinel */ },
101 };
102 
103 struct rte_virtio_xstats_name_off {
104 	char name[RTE_ETH_XSTATS_NAME_SIZE];
105 	unsigned offset;
106 };
107 
108 /* [rt]x_qX_ is prepended to the name string here */
109 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
110 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
111 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
112 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
113 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
114 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
115 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
116 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
117 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
118 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
119 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
120 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
121 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
122 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
123 };
124 
125 /* [rt]x_qX_ is prepended to the name string here */
126 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
127 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
128 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
129 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
130 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
131 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
132 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
133 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
134 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
135 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
136 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
137 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
138 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
139 };
140 
141 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
142 			    sizeof(rte_virtio_rxq_stat_strings[0]))
143 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
144 			    sizeof(rte_virtio_txq_stat_strings[0]))
145 
146 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
147 
148 static struct virtio_pmd_ctrl *
149 virtio_send_command_packed(struct virtnet_ctl *cvq,
150 			   struct virtio_pmd_ctrl *ctrl,
151 			   int *dlen, int pkt_num)
152 {
153 	struct virtqueue *vq = cvq->vq;
154 	int head;
155 	struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
156 	struct virtio_pmd_ctrl *result;
157 	uint16_t flags;
158 	int sum = 0;
159 	int nb_descs = 0;
160 	int k;
161 
162 	/*
163 	 * Format is enforced in qemu code:
164 	 * One TX packet for header;
165 	 * At least one TX packet per argument;
166 	 * One RX packet for ACK.
167 	 */
168 	head = vq->vq_avail_idx;
169 	flags = vq->vq_packed.cached_flags;
170 	desc[head].addr = cvq->virtio_net_hdr_mem;
171 	desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
172 	vq->vq_free_cnt--;
173 	nb_descs++;
174 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
175 		vq->vq_avail_idx -= vq->vq_nentries;
176 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
177 	}
178 
179 	for (k = 0; k < pkt_num; k++) {
180 		desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
181 			+ sizeof(struct virtio_net_ctrl_hdr)
182 			+ sizeof(ctrl->status) + sizeof(uint8_t) * sum;
183 		desc[vq->vq_avail_idx].len = dlen[k];
184 		desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
185 			vq->vq_packed.cached_flags;
186 		sum += dlen[k];
187 		vq->vq_free_cnt--;
188 		nb_descs++;
189 		if (++vq->vq_avail_idx >= vq->vq_nentries) {
190 			vq->vq_avail_idx -= vq->vq_nentries;
191 			vq->vq_packed.cached_flags ^=
192 				VRING_PACKED_DESC_F_AVAIL_USED;
193 		}
194 	}
195 
196 	desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
197 		+ sizeof(struct virtio_net_ctrl_hdr);
198 	desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
199 	desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
200 		vq->vq_packed.cached_flags;
201 	vq->vq_free_cnt--;
202 	nb_descs++;
203 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
204 		vq->vq_avail_idx -= vq->vq_nentries;
205 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
206 	}
207 
208 	virtio_wmb(vq->hw->weak_barriers);
209 	desc[head].flags = VRING_DESC_F_NEXT | flags;
210 
211 	virtio_wmb(vq->hw->weak_barriers);
212 	virtqueue_notify(vq);
213 
214 	/* wait for used descriptors in virtqueue */
215 	while (!desc_is_used(&desc[head], vq))
216 		usleep(100);
217 
218 	virtio_rmb(vq->hw->weak_barriers);
219 
220 	/* now get used descriptors */
221 	vq->vq_free_cnt += nb_descs;
222 	vq->vq_used_cons_idx += nb_descs;
223 	if (vq->vq_used_cons_idx >= vq->vq_nentries) {
224 		vq->vq_used_cons_idx -= vq->vq_nentries;
225 		vq->vq_packed.used_wrap_counter ^= 1;
226 	}
227 
228 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
229 			"vq->vq_avail_idx=%d\n"
230 			"vq->vq_used_cons_idx=%d\n"
231 			"vq->vq_packed.cached_flags=0x%x\n"
232 			"vq->vq_packed.used_wrap_counter=%d\n",
233 			vq->vq_free_cnt,
234 			vq->vq_avail_idx,
235 			vq->vq_used_cons_idx,
236 			vq->vq_packed.cached_flags,
237 			vq->vq_packed.used_wrap_counter);
238 
239 	result = cvq->virtio_net_hdr_mz->addr;
240 	return result;
241 }
242 
243 static struct virtio_pmd_ctrl *
244 virtio_send_command_split(struct virtnet_ctl *cvq,
245 			  struct virtio_pmd_ctrl *ctrl,
246 			  int *dlen, int pkt_num)
247 {
248 	struct virtio_pmd_ctrl *result;
249 	struct virtqueue *vq = cvq->vq;
250 	uint32_t head, i;
251 	int k, sum = 0;
252 
253 	head = vq->vq_desc_head_idx;
254 
255 	/*
256 	 * Format is enforced in qemu code:
257 	 * One TX packet for header;
258 	 * At least one TX packet per argument;
259 	 * One RX packet for ACK.
260 	 */
261 	vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
262 	vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
263 	vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
264 	vq->vq_free_cnt--;
265 	i = vq->vq_split.ring.desc[head].next;
266 
267 	for (k = 0; k < pkt_num; k++) {
268 		vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
269 		vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
270 			+ sizeof(struct virtio_net_ctrl_hdr)
271 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
272 		vq->vq_split.ring.desc[i].len = dlen[k];
273 		sum += dlen[k];
274 		vq->vq_free_cnt--;
275 		i = vq->vq_split.ring.desc[i].next;
276 	}
277 
278 	vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
279 	vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
280 			+ sizeof(struct virtio_net_ctrl_hdr);
281 	vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
282 	vq->vq_free_cnt--;
283 
284 	vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
285 
286 	vq_update_avail_ring(vq, head);
287 	vq_update_avail_idx(vq);
288 
289 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
290 
291 	virtqueue_notify(vq);
292 
293 	while (virtqueue_nused(vq) == 0)
294 		usleep(100);
295 
296 	while (virtqueue_nused(vq)) {
297 		uint32_t idx, desc_idx, used_idx;
298 		struct vring_used_elem *uep;
299 
300 		used_idx = (uint32_t)(vq->vq_used_cons_idx
301 				& (vq->vq_nentries - 1));
302 		uep = &vq->vq_split.ring.used->ring[used_idx];
303 		idx = (uint32_t) uep->id;
304 		desc_idx = idx;
305 
306 		while (vq->vq_split.ring.desc[desc_idx].flags &
307 				VRING_DESC_F_NEXT) {
308 			desc_idx = vq->vq_split.ring.desc[desc_idx].next;
309 			vq->vq_free_cnt++;
310 		}
311 
312 		vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
313 		vq->vq_desc_head_idx = idx;
314 
315 		vq->vq_used_cons_idx++;
316 		vq->vq_free_cnt++;
317 	}
318 
319 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
320 			vq->vq_free_cnt, vq->vq_desc_head_idx);
321 
322 	result = cvq->virtio_net_hdr_mz->addr;
323 	return result;
324 }
325 
326 static int
327 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
328 		    int *dlen, int pkt_num)
329 {
330 	virtio_net_ctrl_ack status = ~0;
331 	struct virtio_pmd_ctrl *result;
332 	struct virtqueue *vq;
333 
334 	ctrl->status = status;
335 
336 	if (!cvq || !cvq->vq) {
337 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
338 		return -1;
339 	}
340 
341 	rte_spinlock_lock(&cvq->lock);
342 	vq = cvq->vq;
343 
344 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
345 		"vq->hw->cvq = %p vq = %p",
346 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
347 
348 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
349 		rte_spinlock_unlock(&cvq->lock);
350 		return -1;
351 	}
352 
353 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
354 		sizeof(struct virtio_pmd_ctrl));
355 
356 	if (vtpci_packed_queue(vq->hw))
357 		result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
358 	else
359 		result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
360 
361 	rte_spinlock_unlock(&cvq->lock);
362 	return result->status;
363 }
364 
365 static int
366 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
367 {
368 	struct virtio_hw *hw = dev->data->dev_private;
369 	struct virtio_pmd_ctrl ctrl;
370 	int dlen[1];
371 	int ret;
372 
373 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
374 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
375 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
376 
377 	dlen[0] = sizeof(uint16_t);
378 
379 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
380 	if (ret) {
381 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
382 			  "failed, this is too late now...");
383 		return -EINVAL;
384 	}
385 
386 	return 0;
387 }
388 
389 static void
390 virtio_dev_queue_release(void *queue __rte_unused)
391 {
392 	/* do nothing */
393 }
394 
395 static uint16_t
396 virtio_get_nr_vq(struct virtio_hw *hw)
397 {
398 	uint16_t nr_vq = hw->max_queue_pairs * 2;
399 
400 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
401 		nr_vq += 1;
402 
403 	return nr_vq;
404 }
405 
406 static void
407 virtio_init_vring(struct virtqueue *vq)
408 {
409 	int size = vq->vq_nentries;
410 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
411 
412 	PMD_INIT_FUNC_TRACE();
413 
414 	memset(ring_mem, 0, vq->vq_ring_size);
415 
416 	vq->vq_used_cons_idx = 0;
417 	vq->vq_desc_head_idx = 0;
418 	vq->vq_avail_idx = 0;
419 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
420 	vq->vq_free_cnt = vq->vq_nentries;
421 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
422 	if (vtpci_packed_queue(vq->hw)) {
423 		vring_init_packed(&vq->vq_packed.ring, ring_mem,
424 				  VIRTIO_PCI_VRING_ALIGN, size);
425 		vring_desc_init_packed(vq, size);
426 	} else {
427 		struct vring *vr = &vq->vq_split.ring;
428 
429 		vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
430 		vring_desc_init_split(vr->desc, size);
431 	}
432 	/*
433 	 * Disable device(host) interrupting guest
434 	 */
435 	virtqueue_disable_intr(vq);
436 }
437 
438 static int
439 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
440 {
441 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
442 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
443 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
444 	unsigned int vq_size, size;
445 	struct virtio_hw *hw = dev->data->dev_private;
446 	struct virtnet_rx *rxvq = NULL;
447 	struct virtnet_tx *txvq = NULL;
448 	struct virtnet_ctl *cvq = NULL;
449 	struct virtqueue *vq;
450 	size_t sz_hdr_mz = 0;
451 	void *sw_ring = NULL;
452 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
453 	int ret;
454 	int numa_node = dev->device->numa_node;
455 
456 	PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
457 			vtpci_queue_idx, numa_node);
458 
459 	/*
460 	 * Read the virtqueue size from the Queue Size field
461 	 * Always power of 2 and if 0 virtqueue does not exist
462 	 */
463 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
464 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
465 	if (vq_size == 0) {
466 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
467 		return -EINVAL;
468 	}
469 
470 	if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
471 		PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
472 		return -EINVAL;
473 	}
474 
475 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
476 		 dev->data->port_id, vtpci_queue_idx);
477 
478 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
479 				vq_size * sizeof(struct vq_desc_extra),
480 				RTE_CACHE_LINE_SIZE);
481 	if (queue_type == VTNET_TQ) {
482 		/*
483 		 * For each xmit packet, allocate a virtio_net_hdr
484 		 * and indirect ring elements
485 		 */
486 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
487 	} else if (queue_type == VTNET_CQ) {
488 		/* Allocate a page for control vq command, data and status */
489 		sz_hdr_mz = PAGE_SIZE;
490 	}
491 
492 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
493 				numa_node);
494 	if (vq == NULL) {
495 		PMD_INIT_LOG(ERR, "can not allocate vq");
496 		return -ENOMEM;
497 	}
498 	hw->vqs[vtpci_queue_idx] = vq;
499 
500 	vq->hw = hw;
501 	vq->vq_queue_index = vtpci_queue_idx;
502 	vq->vq_nentries = vq_size;
503 	if (vtpci_packed_queue(hw)) {
504 		vq->vq_packed.used_wrap_counter = 1;
505 		vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
506 		vq->vq_packed.event_flags_shadow = 0;
507 		if (queue_type == VTNET_RQ)
508 			vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
509 	}
510 
511 	/*
512 	 * Reserve a memzone for vring elements
513 	 */
514 	size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
515 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
516 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
517 		     size, vq->vq_ring_size);
518 
519 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
520 			numa_node, RTE_MEMZONE_IOVA_CONTIG,
521 			VIRTIO_PCI_VRING_ALIGN);
522 	if (mz == NULL) {
523 		if (rte_errno == EEXIST)
524 			mz = rte_memzone_lookup(vq_name);
525 		if (mz == NULL) {
526 			ret = -ENOMEM;
527 			goto fail_q_alloc;
528 		}
529 	}
530 
531 	memset(mz->addr, 0, mz->len);
532 
533 	vq->vq_ring_mem = mz->iova;
534 	vq->vq_ring_virt_mem = mz->addr;
535 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
536 		     (uint64_t)mz->iova);
537 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
538 		     (uint64_t)(uintptr_t)mz->addr);
539 
540 	virtio_init_vring(vq);
541 
542 	if (sz_hdr_mz) {
543 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
544 			 dev->data->port_id, vtpci_queue_idx);
545 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
546 				numa_node, RTE_MEMZONE_IOVA_CONTIG,
547 				RTE_CACHE_LINE_SIZE);
548 		if (hdr_mz == NULL) {
549 			if (rte_errno == EEXIST)
550 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
551 			if (hdr_mz == NULL) {
552 				ret = -ENOMEM;
553 				goto fail_q_alloc;
554 			}
555 		}
556 	}
557 
558 	if (queue_type == VTNET_RQ) {
559 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
560 			       sizeof(vq->sw_ring[0]);
561 
562 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
563 				RTE_CACHE_LINE_SIZE, numa_node);
564 		if (!sw_ring) {
565 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
566 			ret = -ENOMEM;
567 			goto fail_q_alloc;
568 		}
569 
570 		vq->sw_ring = sw_ring;
571 		rxvq = &vq->rxq;
572 		rxvq->vq = vq;
573 		rxvq->port_id = dev->data->port_id;
574 		rxvq->mz = mz;
575 	} else if (queue_type == VTNET_TQ) {
576 		txvq = &vq->txq;
577 		txvq->vq = vq;
578 		txvq->port_id = dev->data->port_id;
579 		txvq->mz = mz;
580 		txvq->virtio_net_hdr_mz = hdr_mz;
581 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
582 	} else if (queue_type == VTNET_CQ) {
583 		cvq = &vq->cq;
584 		cvq->vq = vq;
585 		cvq->mz = mz;
586 		cvq->virtio_net_hdr_mz = hdr_mz;
587 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
588 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
589 
590 		hw->cvq = cvq;
591 	}
592 
593 	/* For virtio_user case (that is when hw->virtio_user_dev is not NULL),
594 	 * we use virtual address. And we need properly set _offset_, please see
595 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
596 	 */
597 	if (!hw->virtio_user_dev)
598 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
599 	else {
600 		vq->vq_ring_mem = (uintptr_t)mz->addr;
601 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
602 		if (queue_type == VTNET_TQ)
603 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
604 		else if (queue_type == VTNET_CQ)
605 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
606 	}
607 
608 	if (queue_type == VTNET_TQ) {
609 		struct virtio_tx_region *txr;
610 		unsigned int i;
611 
612 		txr = hdr_mz->addr;
613 		memset(txr, 0, vq_size * sizeof(*txr));
614 		for (i = 0; i < vq_size; i++) {
615 			struct vring_desc *start_dp = txr[i].tx_indir;
616 
617 			/* first indirect descriptor is always the tx header */
618 			if (!vtpci_packed_queue(hw)) {
619 				vring_desc_init_split(start_dp,
620 						      RTE_DIM(txr[i].tx_indir));
621 				start_dp->addr = txvq->virtio_net_hdr_mem
622 					+ i * sizeof(*txr)
623 					+ offsetof(struct virtio_tx_region,
624 						   tx_hdr);
625 				start_dp->len = hw->vtnet_hdr_size;
626 				start_dp->flags = VRING_DESC_F_NEXT;
627 			}
628 		}
629 	}
630 
631 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
632 		PMD_INIT_LOG(ERR, "setup_queue failed");
633 		return -EINVAL;
634 	}
635 
636 	return 0;
637 
638 fail_q_alloc:
639 	rte_free(sw_ring);
640 	rte_memzone_free(hdr_mz);
641 	rte_memzone_free(mz);
642 	rte_free(vq);
643 
644 	return ret;
645 }
646 
647 static void
648 virtio_free_queues(struct virtio_hw *hw)
649 {
650 	uint16_t nr_vq = virtio_get_nr_vq(hw);
651 	struct virtqueue *vq;
652 	int queue_type;
653 	uint16_t i;
654 
655 	if (hw->vqs == NULL)
656 		return;
657 
658 	for (i = 0; i < nr_vq; i++) {
659 		vq = hw->vqs[i];
660 		if (!vq)
661 			continue;
662 
663 		queue_type = virtio_get_queue_type(hw, i);
664 		if (queue_type == VTNET_RQ) {
665 			rte_free(vq->sw_ring);
666 			rte_memzone_free(vq->rxq.mz);
667 		} else if (queue_type == VTNET_TQ) {
668 			rte_memzone_free(vq->txq.mz);
669 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
670 		} else {
671 			rte_memzone_free(vq->cq.mz);
672 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
673 		}
674 
675 		rte_free(vq);
676 		hw->vqs[i] = NULL;
677 	}
678 
679 	rte_free(hw->vqs);
680 	hw->vqs = NULL;
681 }
682 
683 static int
684 virtio_alloc_queues(struct rte_eth_dev *dev)
685 {
686 	struct virtio_hw *hw = dev->data->dev_private;
687 	uint16_t nr_vq = virtio_get_nr_vq(hw);
688 	uint16_t i;
689 	int ret;
690 
691 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
692 	if (!hw->vqs) {
693 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
694 		return -ENOMEM;
695 	}
696 
697 	for (i = 0; i < nr_vq; i++) {
698 		ret = virtio_init_queue(dev, i);
699 		if (ret < 0) {
700 			virtio_free_queues(hw);
701 			return ret;
702 		}
703 	}
704 
705 	return 0;
706 }
707 
708 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
709 
710 static void
711 virtio_dev_close(struct rte_eth_dev *dev)
712 {
713 	struct virtio_hw *hw = dev->data->dev_private;
714 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
715 
716 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
717 
718 	if (!hw->opened)
719 		return;
720 	hw->opened = false;
721 
722 	/* reset the NIC */
723 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
724 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
725 	if (intr_conf->rxq)
726 		virtio_queues_unbind_intr(dev);
727 
728 	if (intr_conf->lsc || intr_conf->rxq) {
729 		virtio_intr_disable(dev);
730 		rte_intr_efd_disable(dev->intr_handle);
731 		rte_free(dev->intr_handle->intr_vec);
732 		dev->intr_handle->intr_vec = NULL;
733 	}
734 
735 	vtpci_reset(hw);
736 	virtio_dev_free_mbufs(dev);
737 	virtio_free_queues(hw);
738 
739 #ifdef RTE_VIRTIO_USER
740 	if (hw->virtio_user_dev)
741 		virtio_user_dev_uninit(hw->virtio_user_dev);
742 	else
743 #endif
744 	if (dev->device) {
745 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
746 		if (!hw->modern)
747 			rte_pci_ioport_unmap(VTPCI_IO(hw));
748 	}
749 }
750 
751 static int
752 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
753 {
754 	struct virtio_hw *hw = dev->data->dev_private;
755 	struct virtio_pmd_ctrl ctrl;
756 	int dlen[1];
757 	int ret;
758 
759 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
760 		PMD_INIT_LOG(INFO, "host does not support rx control");
761 		return -ENOTSUP;
762 	}
763 
764 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
765 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
766 	ctrl.data[0] = 1;
767 	dlen[0] = 1;
768 
769 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
770 	if (ret) {
771 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
772 		return -EAGAIN;
773 	}
774 
775 	return 0;
776 }
777 
778 static int
779 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
780 {
781 	struct virtio_hw *hw = dev->data->dev_private;
782 	struct virtio_pmd_ctrl ctrl;
783 	int dlen[1];
784 	int ret;
785 
786 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
787 		PMD_INIT_LOG(INFO, "host does not support rx control");
788 		return -ENOTSUP;
789 	}
790 
791 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
792 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
793 	ctrl.data[0] = 0;
794 	dlen[0] = 1;
795 
796 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
797 	if (ret) {
798 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
799 		return -EAGAIN;
800 	}
801 
802 	return 0;
803 }
804 
805 static int
806 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
807 {
808 	struct virtio_hw *hw = dev->data->dev_private;
809 	struct virtio_pmd_ctrl ctrl;
810 	int dlen[1];
811 	int ret;
812 
813 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
814 		PMD_INIT_LOG(INFO, "host does not support rx control");
815 		return -ENOTSUP;
816 	}
817 
818 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
819 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
820 	ctrl.data[0] = 1;
821 	dlen[0] = 1;
822 
823 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
824 	if (ret) {
825 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
826 		return -EAGAIN;
827 	}
828 
829 	return 0;
830 }
831 
832 static int
833 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
834 {
835 	struct virtio_hw *hw = dev->data->dev_private;
836 	struct virtio_pmd_ctrl ctrl;
837 	int dlen[1];
838 	int ret;
839 
840 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
841 		PMD_INIT_LOG(INFO, "host does not support rx control");
842 		return -ENOTSUP;
843 	}
844 
845 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
846 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
847 	ctrl.data[0] = 0;
848 	dlen[0] = 1;
849 
850 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
851 	if (ret) {
852 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
853 		return -EAGAIN;
854 	}
855 
856 	return 0;
857 }
858 
859 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
860 static int
861 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
862 {
863 	struct virtio_hw *hw = dev->data->dev_private;
864 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
865 				 hw->vtnet_hdr_size;
866 	uint32_t frame_size = mtu + ether_hdr_len;
867 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
868 
869 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
870 
871 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
872 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
873 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
874 		return -EINVAL;
875 	}
876 	return 0;
877 }
878 
879 static int
880 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
881 {
882 	struct virtio_hw *hw = dev->data->dev_private;
883 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
884 	struct virtqueue *vq = rxvq->vq;
885 
886 	virtqueue_enable_intr(vq);
887 	virtio_mb(hw->weak_barriers);
888 	return 0;
889 }
890 
891 static int
892 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
893 {
894 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
895 	struct virtqueue *vq = rxvq->vq;
896 
897 	virtqueue_disable_intr(vq);
898 	return 0;
899 }
900 
901 /*
902  * dev_ops for virtio, bare necessities for basic operation
903  */
904 static const struct eth_dev_ops virtio_eth_dev_ops = {
905 	.dev_configure           = virtio_dev_configure,
906 	.dev_start               = virtio_dev_start,
907 	.dev_stop                = virtio_dev_stop,
908 	.dev_close               = virtio_dev_close,
909 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
910 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
911 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
912 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
913 	.mtu_set                 = virtio_mtu_set,
914 	.dev_infos_get           = virtio_dev_info_get,
915 	.stats_get               = virtio_dev_stats_get,
916 	.xstats_get              = virtio_dev_xstats_get,
917 	.xstats_get_names        = virtio_dev_xstats_get_names,
918 	.stats_reset             = virtio_dev_stats_reset,
919 	.xstats_reset            = virtio_dev_stats_reset,
920 	.link_update             = virtio_dev_link_update,
921 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
922 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
923 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
924 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
925 	.rx_queue_release        = virtio_dev_queue_release,
926 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
927 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
928 	.tx_queue_release        = virtio_dev_queue_release,
929 	/* collect stats per queue */
930 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
931 	.vlan_filter_set         = virtio_vlan_filter_set,
932 	.mac_addr_add            = virtio_mac_addr_add,
933 	.mac_addr_remove         = virtio_mac_addr_remove,
934 	.mac_addr_set            = virtio_mac_addr_set,
935 };
936 
937 /*
938  * dev_ops for virtio-user in secondary processes, as we just have
939  * some limited supports currently.
940  */
941 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
942 	.dev_infos_get           = virtio_dev_info_get,
943 	.stats_get               = virtio_dev_stats_get,
944 	.xstats_get              = virtio_dev_xstats_get,
945 	.xstats_get_names        = virtio_dev_xstats_get_names,
946 	.stats_reset             = virtio_dev_stats_reset,
947 	.xstats_reset            = virtio_dev_stats_reset,
948 	/* collect stats per queue */
949 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
950 };
951 
952 static void
953 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
954 {
955 	unsigned i;
956 
957 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
958 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
959 		if (txvq == NULL)
960 			continue;
961 
962 		stats->opackets += txvq->stats.packets;
963 		stats->obytes += txvq->stats.bytes;
964 
965 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
966 			stats->q_opackets[i] = txvq->stats.packets;
967 			stats->q_obytes[i] = txvq->stats.bytes;
968 		}
969 	}
970 
971 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
972 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
973 		if (rxvq == NULL)
974 			continue;
975 
976 		stats->ipackets += rxvq->stats.packets;
977 		stats->ibytes += rxvq->stats.bytes;
978 		stats->ierrors += rxvq->stats.errors;
979 
980 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
981 			stats->q_ipackets[i] = rxvq->stats.packets;
982 			stats->q_ibytes[i] = rxvq->stats.bytes;
983 		}
984 	}
985 
986 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
987 }
988 
989 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
990 				       struct rte_eth_xstat_name *xstats_names,
991 				       __rte_unused unsigned limit)
992 {
993 	unsigned i;
994 	unsigned count = 0;
995 	unsigned t;
996 
997 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
998 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
999 
1000 	if (xstats_names != NULL) {
1001 		/* Note: limit checked in rte_eth_xstats_names() */
1002 
1003 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
1004 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1005 			if (rxvq == NULL)
1006 				continue;
1007 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1008 				snprintf(xstats_names[count].name,
1009 					sizeof(xstats_names[count].name),
1010 					"rx_q%u_%s", i,
1011 					rte_virtio_rxq_stat_strings[t].name);
1012 				count++;
1013 			}
1014 		}
1015 
1016 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
1017 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
1018 			if (txvq == NULL)
1019 				continue;
1020 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1021 				snprintf(xstats_names[count].name,
1022 					sizeof(xstats_names[count].name),
1023 					"tx_q%u_%s", i,
1024 					rte_virtio_txq_stat_strings[t].name);
1025 				count++;
1026 			}
1027 		}
1028 		return count;
1029 	}
1030 	return nstats;
1031 }
1032 
1033 static int
1034 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1035 		      unsigned n)
1036 {
1037 	unsigned i;
1038 	unsigned count = 0;
1039 
1040 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1041 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1042 
1043 	if (n < nstats)
1044 		return nstats;
1045 
1046 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1047 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1048 
1049 		if (rxvq == NULL)
1050 			continue;
1051 
1052 		unsigned t;
1053 
1054 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1055 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1056 				rte_virtio_rxq_stat_strings[t].offset);
1057 			xstats[count].id = count;
1058 			count++;
1059 		}
1060 	}
1061 
1062 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1063 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1064 
1065 		if (txvq == NULL)
1066 			continue;
1067 
1068 		unsigned t;
1069 
1070 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1071 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1072 				rte_virtio_txq_stat_strings[t].offset);
1073 			xstats[count].id = count;
1074 			count++;
1075 		}
1076 	}
1077 
1078 	return count;
1079 }
1080 
1081 static int
1082 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1083 {
1084 	virtio_update_stats(dev, stats);
1085 
1086 	return 0;
1087 }
1088 
1089 static int
1090 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1091 {
1092 	unsigned int i;
1093 
1094 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1095 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1096 		if (txvq == NULL)
1097 			continue;
1098 
1099 		txvq->stats.packets = 0;
1100 		txvq->stats.bytes = 0;
1101 		txvq->stats.multicast = 0;
1102 		txvq->stats.broadcast = 0;
1103 		memset(txvq->stats.size_bins, 0,
1104 		       sizeof(txvq->stats.size_bins[0]) * 8);
1105 	}
1106 
1107 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1108 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1109 		if (rxvq == NULL)
1110 			continue;
1111 
1112 		rxvq->stats.packets = 0;
1113 		rxvq->stats.bytes = 0;
1114 		rxvq->stats.errors = 0;
1115 		rxvq->stats.multicast = 0;
1116 		rxvq->stats.broadcast = 0;
1117 		memset(rxvq->stats.size_bins, 0,
1118 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1119 	}
1120 
1121 	return 0;
1122 }
1123 
1124 static void
1125 virtio_set_hwaddr(struct virtio_hw *hw)
1126 {
1127 	vtpci_write_dev_config(hw,
1128 			offsetof(struct virtio_net_config, mac),
1129 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1130 }
1131 
1132 static void
1133 virtio_get_hwaddr(struct virtio_hw *hw)
1134 {
1135 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1136 		vtpci_read_dev_config(hw,
1137 			offsetof(struct virtio_net_config, mac),
1138 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1139 	} else {
1140 		rte_eth_random_addr(&hw->mac_addr[0]);
1141 		virtio_set_hwaddr(hw);
1142 	}
1143 }
1144 
1145 static int
1146 virtio_mac_table_set(struct virtio_hw *hw,
1147 		     const struct virtio_net_ctrl_mac *uc,
1148 		     const struct virtio_net_ctrl_mac *mc)
1149 {
1150 	struct virtio_pmd_ctrl ctrl;
1151 	int err, len[2];
1152 
1153 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1154 		PMD_DRV_LOG(INFO, "host does not support mac table");
1155 		return -1;
1156 	}
1157 
1158 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1159 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1160 
1161 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1162 	memcpy(ctrl.data, uc, len[0]);
1163 
1164 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1165 	memcpy(ctrl.data + len[0], mc, len[1]);
1166 
1167 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1168 	if (err != 0)
1169 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1170 	return err;
1171 }
1172 
1173 static int
1174 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1175 		    uint32_t index, uint32_t vmdq __rte_unused)
1176 {
1177 	struct virtio_hw *hw = dev->data->dev_private;
1178 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1179 	unsigned int i;
1180 	struct virtio_net_ctrl_mac *uc, *mc;
1181 
1182 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1183 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1184 		return -EINVAL;
1185 	}
1186 
1187 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1188 		sizeof(uc->entries));
1189 	uc->entries = 0;
1190 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1191 		sizeof(mc->entries));
1192 	mc->entries = 0;
1193 
1194 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1195 		const struct rte_ether_addr *addr
1196 			= (i == index) ? mac_addr : addrs + i;
1197 		struct virtio_net_ctrl_mac *tbl
1198 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1199 
1200 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1201 	}
1202 
1203 	return virtio_mac_table_set(hw, uc, mc);
1204 }
1205 
1206 static void
1207 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1208 {
1209 	struct virtio_hw *hw = dev->data->dev_private;
1210 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1211 	struct virtio_net_ctrl_mac *uc, *mc;
1212 	unsigned int i;
1213 
1214 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1215 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1216 		return;
1217 	}
1218 
1219 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1220 		sizeof(uc->entries));
1221 	uc->entries = 0;
1222 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1223 		sizeof(mc->entries));
1224 	mc->entries = 0;
1225 
1226 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1227 		struct virtio_net_ctrl_mac *tbl;
1228 
1229 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1230 			continue;
1231 
1232 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1233 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1234 			RTE_ETHER_ADDR_LEN);
1235 	}
1236 
1237 	virtio_mac_table_set(hw, uc, mc);
1238 }
1239 
1240 static int
1241 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1242 {
1243 	struct virtio_hw *hw = dev->data->dev_private;
1244 
1245 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1246 
1247 	/* Use atomic update if available */
1248 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1249 		struct virtio_pmd_ctrl ctrl;
1250 		int len = RTE_ETHER_ADDR_LEN;
1251 
1252 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1253 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1254 
1255 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1256 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1257 	}
1258 
1259 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1260 		return -ENOTSUP;
1261 
1262 	virtio_set_hwaddr(hw);
1263 	return 0;
1264 }
1265 
1266 static int
1267 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1268 {
1269 	struct virtio_hw *hw = dev->data->dev_private;
1270 	struct virtio_pmd_ctrl ctrl;
1271 	int len;
1272 
1273 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1274 		return -ENOTSUP;
1275 
1276 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1277 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1278 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1279 	len = sizeof(vlan_id);
1280 
1281 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1282 }
1283 
1284 static int
1285 virtio_intr_unmask(struct rte_eth_dev *dev)
1286 {
1287 	struct virtio_hw *hw = dev->data->dev_private;
1288 
1289 	if (rte_intr_ack(dev->intr_handle) < 0)
1290 		return -1;
1291 
1292 	if (!hw->virtio_user_dev)
1293 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1294 
1295 	return 0;
1296 }
1297 
1298 static int
1299 virtio_intr_enable(struct rte_eth_dev *dev)
1300 {
1301 	struct virtio_hw *hw = dev->data->dev_private;
1302 
1303 	if (rte_intr_enable(dev->intr_handle) < 0)
1304 		return -1;
1305 
1306 	if (!hw->virtio_user_dev)
1307 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1308 
1309 	return 0;
1310 }
1311 
1312 static int
1313 virtio_intr_disable(struct rte_eth_dev *dev)
1314 {
1315 	struct virtio_hw *hw = dev->data->dev_private;
1316 
1317 	if (rte_intr_disable(dev->intr_handle) < 0)
1318 		return -1;
1319 
1320 	if (!hw->virtio_user_dev)
1321 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1322 
1323 	return 0;
1324 }
1325 
1326 static int
1327 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1328 {
1329 	uint64_t host_features;
1330 
1331 	/* Prepare guest_features: feature that driver wants to support */
1332 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1333 		req_features);
1334 
1335 	/* Read device(host) feature bits */
1336 	host_features = VTPCI_OPS(hw)->get_features(hw);
1337 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1338 		host_features);
1339 
1340 	/* If supported, ensure MTU value is valid before acknowledging it. */
1341 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1342 		struct virtio_net_config config;
1343 
1344 		vtpci_read_dev_config(hw,
1345 			offsetof(struct virtio_net_config, mtu),
1346 			&config.mtu, sizeof(config.mtu));
1347 
1348 		if (config.mtu < RTE_ETHER_MIN_MTU)
1349 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1350 	}
1351 
1352 	/*
1353 	 * Negotiate features: Subset of device feature bits are written back
1354 	 * guest feature bits.
1355 	 */
1356 	hw->guest_features = req_features;
1357 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1358 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1359 		hw->guest_features);
1360 
1361 	if (hw->modern) {
1362 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1363 			PMD_INIT_LOG(ERR,
1364 				"VIRTIO_F_VERSION_1 features is not enabled.");
1365 			return -1;
1366 		}
1367 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1368 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1369 			PMD_INIT_LOG(ERR,
1370 				"failed to set FEATURES_OK status!");
1371 			return -1;
1372 		}
1373 	}
1374 
1375 	hw->req_guest_features = req_features;
1376 
1377 	return 0;
1378 }
1379 
1380 int
1381 virtio_dev_pause(struct rte_eth_dev *dev)
1382 {
1383 	struct virtio_hw *hw = dev->data->dev_private;
1384 
1385 	rte_spinlock_lock(&hw->state_lock);
1386 
1387 	if (hw->started == 0) {
1388 		/* Device is just stopped. */
1389 		rte_spinlock_unlock(&hw->state_lock);
1390 		return -1;
1391 	}
1392 	hw->started = 0;
1393 	/*
1394 	 * Prevent the worker threads from touching queues to avoid contention,
1395 	 * 1 ms should be enough for the ongoing Tx function to finish.
1396 	 */
1397 	rte_delay_ms(1);
1398 	return 0;
1399 }
1400 
1401 /*
1402  * Recover hw state to let the worker threads continue.
1403  */
1404 void
1405 virtio_dev_resume(struct rte_eth_dev *dev)
1406 {
1407 	struct virtio_hw *hw = dev->data->dev_private;
1408 
1409 	hw->started = 1;
1410 	rte_spinlock_unlock(&hw->state_lock);
1411 }
1412 
1413 /*
1414  * Should be called only after device is paused.
1415  */
1416 int
1417 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1418 		int nb_pkts)
1419 {
1420 	struct virtio_hw *hw = dev->data->dev_private;
1421 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1422 	int ret;
1423 
1424 	hw->inject_pkts = tx_pkts;
1425 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1426 	hw->inject_pkts = NULL;
1427 
1428 	return ret;
1429 }
1430 
1431 static void
1432 virtio_notify_peers(struct rte_eth_dev *dev)
1433 {
1434 	struct virtio_hw *hw = dev->data->dev_private;
1435 	struct virtnet_rx *rxvq;
1436 	struct rte_mbuf *rarp_mbuf;
1437 
1438 	if (!dev->data->rx_queues)
1439 		return;
1440 
1441 	rxvq = dev->data->rx_queues[0];
1442 	if (!rxvq)
1443 		return;
1444 
1445 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1446 			(struct rte_ether_addr *)hw->mac_addr);
1447 	if (rarp_mbuf == NULL) {
1448 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1449 		return;
1450 	}
1451 
1452 	/* If virtio port just stopped, no need to send RARP */
1453 	if (virtio_dev_pause(dev) < 0) {
1454 		rte_pktmbuf_free(rarp_mbuf);
1455 		return;
1456 	}
1457 
1458 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1459 	virtio_dev_resume(dev);
1460 }
1461 
1462 static void
1463 virtio_ack_link_announce(struct rte_eth_dev *dev)
1464 {
1465 	struct virtio_hw *hw = dev->data->dev_private;
1466 	struct virtio_pmd_ctrl ctrl;
1467 
1468 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1469 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1470 
1471 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1472 }
1473 
1474 /*
1475  * Process virtio config changed interrupt. Call the callback
1476  * if link state changed, generate gratuitous RARP packet if
1477  * the status indicates an ANNOUNCE.
1478  */
1479 void
1480 virtio_interrupt_handler(void *param)
1481 {
1482 	struct rte_eth_dev *dev = param;
1483 	struct virtio_hw *hw = dev->data->dev_private;
1484 	uint8_t isr;
1485 	uint16_t status;
1486 
1487 	/* Read interrupt status which clears interrupt */
1488 	isr = vtpci_isr(hw);
1489 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1490 
1491 	if (virtio_intr_unmask(dev) < 0)
1492 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1493 
1494 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1495 		if (virtio_dev_link_update(dev, 0) == 0)
1496 			_rte_eth_dev_callback_process(dev,
1497 						      RTE_ETH_EVENT_INTR_LSC,
1498 						      NULL);
1499 
1500 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1501 			vtpci_read_dev_config(hw,
1502 				offsetof(struct virtio_net_config, status),
1503 				&status, sizeof(status));
1504 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1505 				virtio_notify_peers(dev);
1506 				if (hw->cvq)
1507 					virtio_ack_link_announce(dev);
1508 			}
1509 		}
1510 	}
1511 }
1512 
1513 /* set rx and tx handlers according to what is supported */
1514 static void
1515 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1516 {
1517 	struct virtio_hw *hw = eth_dev->data->dev_private;
1518 
1519 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1520 	if (vtpci_packed_queue(hw)) {
1521 		PMD_INIT_LOG(INFO,
1522 			"virtio: using packed ring %s Tx path on port %u",
1523 			hw->use_vec_tx ? "vectorized" : "standard",
1524 			eth_dev->data->port_id);
1525 		if (hw->use_vec_tx)
1526 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1527 		else
1528 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1529 	} else {
1530 		if (hw->use_inorder_tx) {
1531 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1532 				eth_dev->data->port_id);
1533 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1534 		} else {
1535 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1536 				eth_dev->data->port_id);
1537 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1538 		}
1539 	}
1540 
1541 	if (vtpci_packed_queue(hw)) {
1542 		if (hw->use_vec_rx) {
1543 			PMD_INIT_LOG(INFO,
1544 				"virtio: using packed ring vectorized Rx path on port %u",
1545 				eth_dev->data->port_id);
1546 			eth_dev->rx_pkt_burst =
1547 				&virtio_recv_pkts_packed_vec;
1548 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1549 			PMD_INIT_LOG(INFO,
1550 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1551 				eth_dev->data->port_id);
1552 			eth_dev->rx_pkt_burst =
1553 				&virtio_recv_mergeable_pkts_packed;
1554 		} else {
1555 			PMD_INIT_LOG(INFO,
1556 				"virtio: using packed ring standard Rx path on port %u",
1557 				eth_dev->data->port_id);
1558 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1559 		}
1560 	} else {
1561 		if (hw->use_vec_rx) {
1562 			PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1563 				eth_dev->data->port_id);
1564 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1565 		} else if (hw->use_inorder_rx) {
1566 			PMD_INIT_LOG(INFO,
1567 				"virtio: using inorder Rx path on port %u",
1568 				eth_dev->data->port_id);
1569 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1570 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1571 			PMD_INIT_LOG(INFO,
1572 				"virtio: using mergeable buffer Rx path on port %u",
1573 				eth_dev->data->port_id);
1574 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1575 		} else {
1576 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1577 				eth_dev->data->port_id);
1578 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1579 		}
1580 	}
1581 
1582 }
1583 
1584 /* Only support 1:1 queue/interrupt mapping so far.
1585  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1586  * interrupt vectors (<N+1).
1587  */
1588 static int
1589 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1590 {
1591 	uint32_t i;
1592 	struct virtio_hw *hw = dev->data->dev_private;
1593 
1594 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1595 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1596 		dev->intr_handle->intr_vec[i] = i + 1;
1597 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1598 						 VIRTIO_MSI_NO_VECTOR) {
1599 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1600 			return -EBUSY;
1601 		}
1602 	}
1603 
1604 	return 0;
1605 }
1606 
1607 static void
1608 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1609 {
1610 	uint32_t i;
1611 	struct virtio_hw *hw = dev->data->dev_private;
1612 
1613 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1614 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1615 		VTPCI_OPS(hw)->set_queue_irq(hw,
1616 					     hw->vqs[i * VTNET_CQ],
1617 					     VIRTIO_MSI_NO_VECTOR);
1618 }
1619 
1620 static int
1621 virtio_configure_intr(struct rte_eth_dev *dev)
1622 {
1623 	struct virtio_hw *hw = dev->data->dev_private;
1624 
1625 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1626 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1627 		return -ENOTSUP;
1628 	}
1629 
1630 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1631 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1632 		return -1;
1633 	}
1634 
1635 	if (!dev->intr_handle->intr_vec) {
1636 		dev->intr_handle->intr_vec =
1637 			rte_zmalloc("intr_vec",
1638 				    hw->max_queue_pairs * sizeof(int), 0);
1639 		if (!dev->intr_handle->intr_vec) {
1640 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1641 				     hw->max_queue_pairs);
1642 			return -ENOMEM;
1643 		}
1644 	}
1645 
1646 	/* Re-register callback to update max_intr */
1647 	rte_intr_callback_unregister(dev->intr_handle,
1648 				     virtio_interrupt_handler,
1649 				     dev);
1650 	rte_intr_callback_register(dev->intr_handle,
1651 				   virtio_interrupt_handler,
1652 				   dev);
1653 
1654 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1655 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1656 	 * And for legacy devices, this should be done before queue/vec binding
1657 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1658 	 * (22) will be ignored.
1659 	 */
1660 	if (virtio_intr_enable(dev) < 0) {
1661 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1662 		return -1;
1663 	}
1664 
1665 	if (virtio_queues_bind_intr(dev) < 0) {
1666 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1667 		return -1;
1668 	}
1669 
1670 	return 0;
1671 }
1672 #define SPEED_UNKNOWN    0xffffffff
1673 #define DUPLEX_UNKNOWN   0xff
1674 /* reset device and renegotiate features if needed */
1675 static int
1676 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1677 {
1678 	struct virtio_hw *hw = eth_dev->data->dev_private;
1679 	struct virtio_net_config *config;
1680 	struct virtio_net_config local_config;
1681 	struct rte_pci_device *pci_dev = NULL;
1682 	int ret;
1683 
1684 	/* Reset the device although not necessary at startup */
1685 	vtpci_reset(hw);
1686 
1687 	if (hw->vqs) {
1688 		virtio_dev_free_mbufs(eth_dev);
1689 		virtio_free_queues(hw);
1690 	}
1691 
1692 	/* Tell the host we've noticed this device. */
1693 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1694 
1695 	/* Tell the host we've known how to drive the device. */
1696 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1697 	if (virtio_negotiate_features(hw, req_features) < 0)
1698 		return -1;
1699 
1700 	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1701 
1702 	if (!hw->virtio_user_dev)
1703 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1704 
1705 	/* If host does not support both status and MSI-X then disable LSC */
1706 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1707 	    hw->use_msix != VIRTIO_MSIX_NONE)
1708 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1709 	else
1710 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1711 
1712 	/* Setting up rx_header size for the device */
1713 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1714 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1715 	    vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1716 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1717 	else
1718 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1719 
1720 	/* Copy the permanent MAC address to: virtio_hw */
1721 	virtio_get_hwaddr(hw);
1722 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1723 			&eth_dev->data->mac_addrs[0]);
1724 	PMD_INIT_LOG(DEBUG,
1725 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1726 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1727 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1728 
1729 	if (hw->speed == SPEED_UNKNOWN) {
1730 		if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1731 			config = &local_config;
1732 			vtpci_read_dev_config(hw,
1733 				offsetof(struct virtio_net_config, speed),
1734 				&config->speed, sizeof(config->speed));
1735 			vtpci_read_dev_config(hw,
1736 				offsetof(struct virtio_net_config, duplex),
1737 				&config->duplex, sizeof(config->duplex));
1738 			hw->speed = config->speed;
1739 			hw->duplex = config->duplex;
1740 		}
1741 	}
1742 	if (hw->speed == SPEED_UNKNOWN)
1743 		hw->speed = ETH_SPEED_NUM_10G;
1744 	if (hw->duplex == DUPLEX_UNKNOWN)
1745 		hw->duplex = ETH_LINK_FULL_DUPLEX;
1746 	PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1747 		hw->speed, hw->duplex);
1748 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1749 		config = &local_config;
1750 
1751 		vtpci_read_dev_config(hw,
1752 			offsetof(struct virtio_net_config, mac),
1753 			&config->mac, sizeof(config->mac));
1754 
1755 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1756 			vtpci_read_dev_config(hw,
1757 				offsetof(struct virtio_net_config, status),
1758 				&config->status, sizeof(config->status));
1759 		} else {
1760 			PMD_INIT_LOG(DEBUG,
1761 				     "VIRTIO_NET_F_STATUS is not supported");
1762 			config->status = 0;
1763 		}
1764 
1765 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1766 			vtpci_read_dev_config(hw,
1767 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1768 				&config->max_virtqueue_pairs,
1769 				sizeof(config->max_virtqueue_pairs));
1770 		} else {
1771 			PMD_INIT_LOG(DEBUG,
1772 				     "VIRTIO_NET_F_MQ is not supported");
1773 			config->max_virtqueue_pairs = 1;
1774 		}
1775 
1776 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1777 
1778 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1779 			vtpci_read_dev_config(hw,
1780 				offsetof(struct virtio_net_config, mtu),
1781 				&config->mtu,
1782 				sizeof(config->mtu));
1783 
1784 			/*
1785 			 * MTU value has already been checked at negotiation
1786 			 * time, but check again in case it has changed since
1787 			 * then, which should not happen.
1788 			 */
1789 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1790 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1791 						config->mtu);
1792 				return -1;
1793 			}
1794 
1795 			hw->max_mtu = config->mtu;
1796 			/* Set initial MTU to maximum one supported by vhost */
1797 			eth_dev->data->mtu = config->mtu;
1798 
1799 		} else {
1800 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1801 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1802 		}
1803 
1804 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1805 				config->max_virtqueue_pairs);
1806 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1807 		PMD_INIT_LOG(DEBUG,
1808 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1809 				config->mac[0], config->mac[1],
1810 				config->mac[2], config->mac[3],
1811 				config->mac[4], config->mac[5]);
1812 	} else {
1813 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1814 		hw->max_queue_pairs = 1;
1815 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1816 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1817 	}
1818 
1819 	ret = virtio_alloc_queues(eth_dev);
1820 	if (ret < 0)
1821 		return ret;
1822 
1823 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1824 		if (virtio_configure_intr(eth_dev) < 0) {
1825 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1826 			virtio_free_queues(hw);
1827 			return -1;
1828 		}
1829 	}
1830 
1831 	vtpci_reinit_complete(hw);
1832 
1833 	if (pci_dev)
1834 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1835 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1836 			pci_dev->id.device_id);
1837 
1838 	return 0;
1839 }
1840 
1841 /*
1842  * Remap the PCI device again (IO port map for legacy device and
1843  * memory map for modern device), so that the secondary process
1844  * could have the PCI initiated correctly.
1845  */
1846 static int
1847 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1848 {
1849 	if (hw->modern) {
1850 		/*
1851 		 * We don't have to re-parse the PCI config space, since
1852 		 * rte_pci_map_device() makes sure the mapped address
1853 		 * in secondary process would equal to the one mapped in
1854 		 * the primary process: error will be returned if that
1855 		 * requirement is not met.
1856 		 *
1857 		 * That said, we could simply reuse all cap pointers
1858 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1859 		 * primary process, which is stored in shared memory.
1860 		 */
1861 		if (rte_pci_map_device(pci_dev)) {
1862 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1863 			return -1;
1864 		}
1865 	} else {
1866 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1867 			return -1;
1868 	}
1869 
1870 	return 0;
1871 }
1872 
1873 static void
1874 virtio_set_vtpci_ops(struct virtio_hw *hw)
1875 {
1876 #ifdef RTE_VIRTIO_USER
1877 	if (hw->virtio_user_dev)
1878 		VTPCI_OPS(hw) = &virtio_user_ops;
1879 	else
1880 #endif
1881 	if (hw->modern)
1882 		VTPCI_OPS(hw) = &modern_ops;
1883 	else
1884 		VTPCI_OPS(hw) = &legacy_ops;
1885 }
1886 
1887 /*
1888  * This function is based on probe() function in virtio_pci.c
1889  * It returns 0 on success.
1890  */
1891 int
1892 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1893 {
1894 	struct virtio_hw *hw = eth_dev->data->dev_private;
1895 	uint32_t speed = SPEED_UNKNOWN;
1896 	int vectorized = 0;
1897 	int ret;
1898 
1899 	if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1900 		PMD_INIT_LOG(ERR,
1901 			"Not sufficient headroom required = %d, avail = %d",
1902 			(int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1903 			RTE_PKTMBUF_HEADROOM);
1904 
1905 		return -1;
1906 	}
1907 
1908 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1909 
1910 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1911 		if (!hw->virtio_user_dev) {
1912 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1913 			if (ret)
1914 				return ret;
1915 		}
1916 
1917 		virtio_set_vtpci_ops(hw);
1918 		set_rxtx_funcs(eth_dev);
1919 
1920 		return 0;
1921 	}
1922 	ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
1923 		 NULL, &speed, &vectorized);
1924 	if (ret < 0)
1925 		return ret;
1926 	hw->speed = speed;
1927 	/*
1928 	 * Pass the information to the rte_eth_dev_close() that it should also
1929 	 * release the private port resources.
1930 	 */
1931 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1932 
1933 	/* Allocate memory for storing MAC addresses */
1934 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1935 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1936 	if (eth_dev->data->mac_addrs == NULL) {
1937 		PMD_INIT_LOG(ERR,
1938 			"Failed to allocate %d bytes needed to store MAC addresses",
1939 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1940 		return -ENOMEM;
1941 	}
1942 
1943 	hw->port_id = eth_dev->data->port_id;
1944 	/* For virtio_user case the hw->virtio_user_dev is populated by
1945 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1946 	 */
1947 	if (!hw->virtio_user_dev) {
1948 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1949 		if (ret)
1950 			goto err_vtpci_init;
1951 	}
1952 
1953 	rte_spinlock_init(&hw->state_lock);
1954 
1955 	/* reset device and negotiate default features */
1956 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1957 	if (ret < 0)
1958 		goto err_virtio_init;
1959 
1960 	if (vectorized) {
1961 		if (!vtpci_packed_queue(hw)) {
1962 			hw->use_vec_rx = 1;
1963 		} else {
1964 #if !defined(CC_AVX512_SUPPORT)
1965 			PMD_DRV_LOG(INFO,
1966 				"building environment do not support packed ring vectorized");
1967 #else
1968 			if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F)) {
1969 				hw->use_vec_rx = 1;
1970 				hw->use_vec_tx = 1;
1971 			}
1972 #endif
1973 		}
1974 	}
1975 
1976 	hw->opened = true;
1977 
1978 	return 0;
1979 
1980 err_virtio_init:
1981 	if (!hw->virtio_user_dev) {
1982 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1983 		if (!hw->modern)
1984 			rte_pci_ioport_unmap(VTPCI_IO(hw));
1985 	}
1986 err_vtpci_init:
1987 	rte_free(eth_dev->data->mac_addrs);
1988 	eth_dev->data->mac_addrs = NULL;
1989 	return ret;
1990 }
1991 
1992 static int
1993 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1994 {
1995 	PMD_INIT_FUNC_TRACE();
1996 
1997 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1998 		return 0;
1999 
2000 	virtio_dev_stop(eth_dev);
2001 	virtio_dev_close(eth_dev);
2002 
2003 	eth_dev->dev_ops = NULL;
2004 	eth_dev->tx_pkt_burst = NULL;
2005 	eth_dev->rx_pkt_burst = NULL;
2006 
2007 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
2008 
2009 	return 0;
2010 }
2011 
2012 
2013 static int vdpa_check_handler(__rte_unused const char *key,
2014 		const char *value, void *ret_val)
2015 {
2016 	if (strcmp(value, "1") == 0)
2017 		*(int *)ret_val = 1;
2018 	else
2019 		*(int *)ret_val = 0;
2020 
2021 	return 0;
2022 }
2023 
2024 
2025 static uint32_t
2026 virtio_dev_speed_capa_get(uint32_t speed)
2027 {
2028 	switch (speed) {
2029 	case ETH_SPEED_NUM_10G:
2030 		return ETH_LINK_SPEED_10G;
2031 	case ETH_SPEED_NUM_20G:
2032 		return ETH_LINK_SPEED_20G;
2033 	case ETH_SPEED_NUM_25G:
2034 		return ETH_LINK_SPEED_25G;
2035 	case ETH_SPEED_NUM_40G:
2036 		return ETH_LINK_SPEED_40G;
2037 	case ETH_SPEED_NUM_50G:
2038 		return ETH_LINK_SPEED_50G;
2039 	case ETH_SPEED_NUM_56G:
2040 		return ETH_LINK_SPEED_56G;
2041 	case ETH_SPEED_NUM_100G:
2042 		return ETH_LINK_SPEED_100G;
2043 	default:
2044 		return 0;
2045 	}
2046 }
2047 
2048 static int vectorized_check_handler(__rte_unused const char *key,
2049 		const char *value, void *ret_val)
2050 {
2051 	if (strcmp(value, "1") == 0)
2052 		*(int *)ret_val = 1;
2053 	else
2054 		*(int *)ret_val = 0;
2055 
2056 	return 0;
2057 }
2058 
2059 #define VIRTIO_ARG_SPEED      "speed"
2060 #define VIRTIO_ARG_VDPA       "vdpa"
2061 #define VIRTIO_ARG_VECTORIZED "vectorized"
2062 
2063 
2064 static int
2065 link_speed_handler(const char *key __rte_unused,
2066 		const char *value, void *ret_val)
2067 {
2068 	uint32_t val;
2069 	if (!value || !ret_val)
2070 		return -EINVAL;
2071 	val = strtoul(value, NULL, 0);
2072 	/* validate input */
2073 	if (virtio_dev_speed_capa_get(val) == 0)
2074 		return -EINVAL;
2075 	*(uint32_t *)ret_val = val;
2076 
2077 	return 0;
2078 }
2079 
2080 
2081 static int
2082 virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
2083 	uint32_t *speed, int *vectorized)
2084 {
2085 	struct rte_kvargs *kvlist;
2086 	int ret = 0;
2087 
2088 	if (devargs == NULL)
2089 		return 0;
2090 
2091 	kvlist = rte_kvargs_parse(devargs->args, NULL);
2092 	if (kvlist == NULL) {
2093 		PMD_INIT_LOG(ERR, "error when parsing param");
2094 		return 0;
2095 	}
2096 	if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
2097 		/* vdpa mode selected when there's a key-value pair:
2098 		 * vdpa=1
2099 		 */
2100 		ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
2101 				vdpa_check_handler, vdpa);
2102 		if (ret < 0) {
2103 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2104 				VIRTIO_ARG_VDPA);
2105 			goto exit;
2106 		}
2107 	}
2108 	if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2109 		ret = rte_kvargs_process(kvlist,
2110 					VIRTIO_ARG_SPEED,
2111 					link_speed_handler, speed);
2112 		if (ret < 0) {
2113 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2114 					VIRTIO_ARG_SPEED);
2115 			goto exit;
2116 		}
2117 	}
2118 
2119 	if (vectorized &&
2120 		rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2121 		ret = rte_kvargs_process(kvlist,
2122 				VIRTIO_ARG_VECTORIZED,
2123 				vectorized_check_handler, vectorized);
2124 		if (ret < 0) {
2125 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2126 					VIRTIO_ARG_VECTORIZED);
2127 			goto exit;
2128 		}
2129 	}
2130 
2131 exit:
2132 	rte_kvargs_free(kvlist);
2133 	return ret;
2134 }
2135 
2136 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2137 	struct rte_pci_device *pci_dev)
2138 {
2139 	int vdpa = 0;
2140 	int ret = 0;
2141 
2142 	ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL,
2143 		NULL);
2144 	if (ret < 0) {
2145 		PMD_INIT_LOG(ERR, "devargs parsing is failed");
2146 		return ret;
2147 	}
2148 	/* virtio pmd skips probe if device needs to work in vdpa mode */
2149 	if (vdpa == 1)
2150 		return 1;
2151 
2152 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2153 		eth_virtio_dev_init);
2154 }
2155 
2156 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2157 {
2158 	int ret;
2159 
2160 	ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2161 	/* Port has already been released by close. */
2162 	if (ret == -ENODEV)
2163 		ret = 0;
2164 	return ret;
2165 }
2166 
2167 static struct rte_pci_driver rte_virtio_pmd = {
2168 	.driver = {
2169 		.name = "net_virtio",
2170 	},
2171 	.id_table = pci_id_virtio_map,
2172 	.drv_flags = 0,
2173 	.probe = eth_virtio_pci_probe,
2174 	.remove = eth_virtio_pci_remove,
2175 };
2176 
2177 RTE_INIT(rte_virtio_pmd_init)
2178 {
2179 	rte_eal_iopl_init();
2180 	rte_pci_register(&rte_virtio_pmd);
2181 }
2182 
2183 static bool
2184 rx_offload_enabled(struct virtio_hw *hw)
2185 {
2186 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2187 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2188 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2189 }
2190 
2191 static bool
2192 tx_offload_enabled(struct virtio_hw *hw)
2193 {
2194 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2195 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2196 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2197 }
2198 
2199 /*
2200  * Configure virtio device
2201  * It returns 0 on success.
2202  */
2203 static int
2204 virtio_dev_configure(struct rte_eth_dev *dev)
2205 {
2206 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2207 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2208 	struct virtio_hw *hw = dev->data->dev_private;
2209 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2210 		hw->vtnet_hdr_size;
2211 	uint64_t rx_offloads = rxmode->offloads;
2212 	uint64_t tx_offloads = txmode->offloads;
2213 	uint64_t req_features;
2214 	int ret;
2215 
2216 	PMD_INIT_LOG(DEBUG, "configure");
2217 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2218 
2219 	if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2220 		PMD_DRV_LOG(ERR,
2221 			"Unsupported Rx multi queue mode %d",
2222 			rxmode->mq_mode);
2223 		return -EINVAL;
2224 	}
2225 
2226 	if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2227 		PMD_DRV_LOG(ERR,
2228 			"Unsupported Tx multi queue mode %d",
2229 			txmode->mq_mode);
2230 		return -EINVAL;
2231 	}
2232 
2233 	if (dev->data->dev_conf.intr_conf.rxq) {
2234 		ret = virtio_init_device(dev, hw->req_guest_features);
2235 		if (ret < 0)
2236 			return ret;
2237 	}
2238 
2239 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2240 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2241 
2242 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2243 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2244 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2245 
2246 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2247 		req_features |=
2248 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2249 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2250 
2251 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2252 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2253 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2254 
2255 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2256 		req_features |=
2257 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2258 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2259 
2260 	/* if request features changed, reinit the device */
2261 	if (req_features != hw->req_guest_features) {
2262 		ret = virtio_init_device(dev, req_features);
2263 		if (ret < 0)
2264 			return ret;
2265 	}
2266 
2267 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2268 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2269 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2270 		PMD_DRV_LOG(ERR,
2271 			"rx checksum not available on this host");
2272 		return -ENOTSUP;
2273 	}
2274 
2275 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2276 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2277 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2278 		PMD_DRV_LOG(ERR,
2279 			"Large Receive Offload not available on this host");
2280 		return -ENOTSUP;
2281 	}
2282 
2283 	/* start control queue */
2284 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2285 		virtio_dev_cq_start(dev);
2286 
2287 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2288 		hw->vlan_strip = 1;
2289 
2290 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2291 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2292 		PMD_DRV_LOG(ERR,
2293 			    "vlan filtering not available on this host");
2294 		return -ENOTSUP;
2295 	}
2296 
2297 	hw->has_tx_offload = tx_offload_enabled(hw);
2298 	hw->has_rx_offload = rx_offload_enabled(hw);
2299 
2300 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2301 		/* Enable vector (0) for Link State Intrerrupt */
2302 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2303 				VIRTIO_MSI_NO_VECTOR) {
2304 			PMD_DRV_LOG(ERR, "failed to set config vector");
2305 			return -EBUSY;
2306 		}
2307 
2308 	if (vtpci_packed_queue(hw)) {
2309 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2310 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2311 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2312 		     !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2313 		     !vtpci_with_feature(hw, VIRTIO_F_VERSION_1))) {
2314 			PMD_DRV_LOG(INFO,
2315 				"disabled packed ring vectorized path for requirements not met");
2316 			hw->use_vec_rx = 0;
2317 			hw->use_vec_tx = 0;
2318 		}
2319 #else
2320 		hw->use_vec_rx = 0;
2321 		hw->use_vec_tx = 0;
2322 #endif
2323 
2324 		if (hw->use_vec_rx) {
2325 			if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2326 				PMD_DRV_LOG(INFO,
2327 					"disabled packed ring vectorized rx for mrg_rxbuf enabled");
2328 				hw->use_vec_rx = 0;
2329 			}
2330 
2331 			if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2332 				PMD_DRV_LOG(INFO,
2333 					"disabled packed ring vectorized rx for TCP_LRO enabled");
2334 				hw->use_vec_rx = 0;
2335 			}
2336 		}
2337 	} else {
2338 		if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2339 			hw->use_inorder_tx = 1;
2340 			hw->use_inorder_rx = 1;
2341 			hw->use_vec_rx = 0;
2342 		}
2343 
2344 		if (hw->use_vec_rx) {
2345 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2346 			if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2347 				PMD_DRV_LOG(INFO,
2348 					"disabled split ring vectorized path for requirement not met");
2349 				hw->use_vec_rx = 0;
2350 			}
2351 #endif
2352 			if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2353 				PMD_DRV_LOG(INFO,
2354 					"disabled split ring vectorized rx for mrg_rxbuf enabled");
2355 				hw->use_vec_rx = 0;
2356 			}
2357 
2358 			if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2359 					   DEV_RX_OFFLOAD_TCP_CKSUM |
2360 					   DEV_RX_OFFLOAD_TCP_LRO |
2361 					   DEV_RX_OFFLOAD_VLAN_STRIP)) {
2362 				PMD_DRV_LOG(INFO,
2363 					"disabled split ring vectorized rx for offloading enabled");
2364 				hw->use_vec_rx = 0;
2365 			}
2366 		}
2367 	}
2368 
2369 	return 0;
2370 }
2371 
2372 
2373 static int
2374 virtio_dev_start(struct rte_eth_dev *dev)
2375 {
2376 	uint16_t nb_queues, i;
2377 	struct virtnet_rx *rxvq;
2378 	struct virtnet_tx *txvq __rte_unused;
2379 	struct virtio_hw *hw = dev->data->dev_private;
2380 	int ret;
2381 
2382 	/* Finish the initialization of the queues */
2383 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2384 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2385 		if (ret < 0)
2386 			return ret;
2387 	}
2388 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2389 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2390 		if (ret < 0)
2391 			return ret;
2392 	}
2393 
2394 	/* check if lsc interrupt feature is enabled */
2395 	if (dev->data->dev_conf.intr_conf.lsc) {
2396 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2397 			PMD_DRV_LOG(ERR, "link status not supported by host");
2398 			return -ENOTSUP;
2399 		}
2400 	}
2401 
2402 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2403 	 * in device configure, but it could be unmapped  when device is
2404 	 * stopped.
2405 	 */
2406 	if (dev->data->dev_conf.intr_conf.lsc ||
2407 	    dev->data->dev_conf.intr_conf.rxq) {
2408 		virtio_intr_disable(dev);
2409 
2410 		/* Setup interrupt callback  */
2411 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2412 			rte_intr_callback_register(dev->intr_handle,
2413 						   virtio_interrupt_handler,
2414 						   dev);
2415 
2416 		if (virtio_intr_enable(dev) < 0) {
2417 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2418 			return -EIO;
2419 		}
2420 	}
2421 
2422 	/*Notify the backend
2423 	 *Otherwise the tap backend might already stop its queue due to fullness.
2424 	 *vhost backend will have no chance to be waked up
2425 	 */
2426 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2427 	if (hw->max_queue_pairs > 1) {
2428 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2429 			return -EINVAL;
2430 	}
2431 
2432 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2433 
2434 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2435 		rxvq = dev->data->rx_queues[i];
2436 		/* Flush the old packets */
2437 		virtqueue_rxvq_flush(rxvq->vq);
2438 		virtqueue_notify(rxvq->vq);
2439 	}
2440 
2441 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2442 		txvq = dev->data->tx_queues[i];
2443 		virtqueue_notify(txvq->vq);
2444 	}
2445 
2446 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2447 
2448 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2449 		rxvq = dev->data->rx_queues[i];
2450 		VIRTQUEUE_DUMP(rxvq->vq);
2451 	}
2452 
2453 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2454 		txvq = dev->data->tx_queues[i];
2455 		VIRTQUEUE_DUMP(txvq->vq);
2456 	}
2457 
2458 	set_rxtx_funcs(dev);
2459 	hw->started = true;
2460 
2461 	/* Initialize Link state */
2462 	virtio_dev_link_update(dev, 0);
2463 
2464 	return 0;
2465 }
2466 
2467 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2468 {
2469 	struct virtio_hw *hw = dev->data->dev_private;
2470 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2471 	const char *type __rte_unused;
2472 	unsigned int i, mbuf_num = 0;
2473 	struct virtqueue *vq;
2474 	struct rte_mbuf *buf;
2475 	int queue_type;
2476 
2477 	if (hw->vqs == NULL)
2478 		return;
2479 
2480 	for (i = 0; i < nr_vq; i++) {
2481 		vq = hw->vqs[i];
2482 		if (!vq)
2483 			continue;
2484 
2485 		queue_type = virtio_get_queue_type(hw, i);
2486 		if (queue_type == VTNET_RQ)
2487 			type = "rxq";
2488 		else if (queue_type == VTNET_TQ)
2489 			type = "txq";
2490 		else
2491 			continue;
2492 
2493 		PMD_INIT_LOG(DEBUG,
2494 			"Before freeing %s[%d] used and unused buf",
2495 			type, i);
2496 		VIRTQUEUE_DUMP(vq);
2497 
2498 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2499 			rte_pktmbuf_free(buf);
2500 			mbuf_num++;
2501 		}
2502 
2503 		PMD_INIT_LOG(DEBUG,
2504 			"After freeing %s[%d] used and unused buf",
2505 			type, i);
2506 		VIRTQUEUE_DUMP(vq);
2507 	}
2508 
2509 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2510 }
2511 
2512 /*
2513  * Stop device: disable interrupt and mark link down
2514  */
2515 static void
2516 virtio_dev_stop(struct rte_eth_dev *dev)
2517 {
2518 	struct virtio_hw *hw = dev->data->dev_private;
2519 	struct rte_eth_link link;
2520 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2521 
2522 	PMD_INIT_LOG(DEBUG, "stop");
2523 
2524 	rte_spinlock_lock(&hw->state_lock);
2525 	if (!hw->started)
2526 		goto out_unlock;
2527 	hw->started = false;
2528 
2529 	if (intr_conf->lsc || intr_conf->rxq) {
2530 		virtio_intr_disable(dev);
2531 
2532 		/* Reset interrupt callback  */
2533 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2534 			rte_intr_callback_unregister(dev->intr_handle,
2535 						     virtio_interrupt_handler,
2536 						     dev);
2537 		}
2538 	}
2539 
2540 	memset(&link, 0, sizeof(link));
2541 	rte_eth_linkstatus_set(dev, &link);
2542 out_unlock:
2543 	rte_spinlock_unlock(&hw->state_lock);
2544 }
2545 
2546 static int
2547 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2548 {
2549 	struct rte_eth_link link;
2550 	uint16_t status;
2551 	struct virtio_hw *hw = dev->data->dev_private;
2552 
2553 	memset(&link, 0, sizeof(link));
2554 	link.link_duplex = hw->duplex;
2555 	link.link_speed  = hw->speed;
2556 	link.link_autoneg = ETH_LINK_AUTONEG;
2557 
2558 	if (!hw->started) {
2559 		link.link_status = ETH_LINK_DOWN;
2560 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2561 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2562 		vtpci_read_dev_config(hw,
2563 				offsetof(struct virtio_net_config, status),
2564 				&status, sizeof(status));
2565 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2566 			link.link_status = ETH_LINK_DOWN;
2567 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2568 				     dev->data->port_id);
2569 		} else {
2570 			link.link_status = ETH_LINK_UP;
2571 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2572 				     dev->data->port_id);
2573 		}
2574 	} else {
2575 		link.link_status = ETH_LINK_UP;
2576 	}
2577 
2578 	return rte_eth_linkstatus_set(dev, &link);
2579 }
2580 
2581 static int
2582 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2583 {
2584 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2585 	struct virtio_hw *hw = dev->data->dev_private;
2586 	uint64_t offloads = rxmode->offloads;
2587 
2588 	if (mask & ETH_VLAN_FILTER_MASK) {
2589 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2590 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2591 
2592 			PMD_DRV_LOG(NOTICE,
2593 				"vlan filtering not available on this host");
2594 
2595 			return -ENOTSUP;
2596 		}
2597 	}
2598 
2599 	if (mask & ETH_VLAN_STRIP_MASK)
2600 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2601 
2602 	return 0;
2603 }
2604 
2605 static int
2606 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2607 {
2608 	uint64_t tso_mask, host_features;
2609 	struct virtio_hw *hw = dev->data->dev_private;
2610 	dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2611 
2612 	dev_info->max_rx_queues =
2613 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2614 	dev_info->max_tx_queues =
2615 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2616 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2617 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2618 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2619 
2620 	host_features = VTPCI_OPS(hw)->get_features(hw);
2621 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2622 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2623 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2624 		dev_info->rx_offload_capa |=
2625 			DEV_RX_OFFLOAD_TCP_CKSUM |
2626 			DEV_RX_OFFLOAD_UDP_CKSUM;
2627 	}
2628 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2629 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2630 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2631 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2632 	if ((host_features & tso_mask) == tso_mask)
2633 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2634 
2635 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2636 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2637 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2638 		dev_info->tx_offload_capa |=
2639 			DEV_TX_OFFLOAD_UDP_CKSUM |
2640 			DEV_TX_OFFLOAD_TCP_CKSUM;
2641 	}
2642 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2643 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2644 	if ((host_features & tso_mask) == tso_mask)
2645 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2646 
2647 	return 0;
2648 }
2649 
2650 /*
2651  * It enables testpmd to collect per queue stats.
2652  */
2653 static int
2654 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2655 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2656 __rte_unused uint8_t is_rx)
2657 {
2658 	return 0;
2659 }
2660 
2661 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2662 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2663 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2664 
2665 RTE_INIT(virtio_init_log)
2666 {
2667 	virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2668 	if (virtio_logtype_init >= 0)
2669 		rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2670 	virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2671 	if (virtio_logtype_driver >= 0)
2672 		rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2673 }
2674