xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 59f3a8acbcdbafeebe816a26d76dfb06e6450f31)
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 <ethdev_driver.h>
12 #include <rte_memcpy.h>
13 #include <rte_string_fns.h>
14 #include <rte_memzone.h>
15 #include <rte_malloc.h>
16 #include <rte_branch_prediction.h>
17 #include <rte_ether.h>
18 #include <rte_ip.h>
19 #include <rte_arp.h>
20 #include <rte_common.h>
21 #include <rte_errno.h>
22 #include <rte_cpuflags.h>
23 #include <rte_vect.h>
24 #include <rte_memory.h>
25 #include <rte_eal_paging.h>
26 #include <rte_eal.h>
27 #include <rte_dev.h>
28 #include <rte_cycles.h>
29 #include <rte_kvargs.h>
30 
31 #include "virtio_ethdev.h"
32 #include "virtio.h"
33 #include "virtio_logs.h"
34 #include "virtqueue.h"
35 #include "virtio_rxtx.h"
36 #include "virtio_rxtx_simple.h"
37 #include "virtio_user/virtio_user_dev.h"
38 
39 static int  virtio_dev_configure(struct rte_eth_dev *dev);
40 static int  virtio_dev_start(struct rte_eth_dev *dev);
41 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
42 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
43 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
44 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
45 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
46 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
47 	uint32_t *speed,
48 	int *vectorized);
49 static int virtio_dev_info_get(struct rte_eth_dev *dev,
50 				struct rte_eth_dev_info *dev_info);
51 static int virtio_dev_link_update(struct rte_eth_dev *dev,
52 	int wait_to_complete);
53 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
54 
55 static void virtio_set_hwaddr(struct virtio_hw *hw);
56 static void virtio_get_hwaddr(struct virtio_hw *hw);
57 
58 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
59 				 struct rte_eth_stats *stats);
60 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
61 				 struct rte_eth_xstat *xstats, unsigned n);
62 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
63 				       struct rte_eth_xstat_name *xstats_names,
64 				       unsigned limit);
65 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
66 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
67 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
68 				uint16_t vlan_id, int on);
69 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
70 				struct rte_ether_addr *mac_addr,
71 				uint32_t index, uint32_t vmdq);
72 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
73 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
74 				struct rte_ether_addr *mac_addr);
75 
76 static int virtio_intr_disable(struct rte_eth_dev *dev);
77 
78 static int virtio_dev_queue_stats_mapping_set(
79 	struct rte_eth_dev *eth_dev,
80 	uint16_t queue_id,
81 	uint8_t stat_idx,
82 	uint8_t is_rx);
83 
84 static void virtio_notify_peers(struct rte_eth_dev *dev);
85 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
86 
87 struct rte_virtio_xstats_name_off {
88 	char name[RTE_ETH_XSTATS_NAME_SIZE];
89 	unsigned offset;
90 };
91 
92 /* [rt]x_qX_ is prepended to the name string here */
93 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
94 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
95 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
96 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
97 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
98 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
99 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
100 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
101 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
102 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
103 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
104 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
105 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
106 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
107 };
108 
109 /* [rt]x_qX_ is prepended to the name string here */
110 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
111 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
112 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
113 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
114 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
115 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
116 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
117 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
118 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
119 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
120 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
121 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
122 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
123 };
124 
125 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
126 			    sizeof(rte_virtio_rxq_stat_strings[0]))
127 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
128 			    sizeof(rte_virtio_txq_stat_strings[0]))
129 
130 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
131 
132 static struct virtio_pmd_ctrl *
133 virtio_send_command_packed(struct virtnet_ctl *cvq,
134 			   struct virtio_pmd_ctrl *ctrl,
135 			   int *dlen, int pkt_num)
136 {
137 	struct virtqueue *vq = virtnet_cq_to_vq(cvq);
138 	int head;
139 	struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
140 	struct virtio_pmd_ctrl *result;
141 	uint16_t flags;
142 	int sum = 0;
143 	int nb_descs = 0;
144 	int k;
145 
146 	/*
147 	 * Format is enforced in qemu code:
148 	 * One TX packet for header;
149 	 * At least one TX packet per argument;
150 	 * One RX packet for ACK.
151 	 */
152 	head = vq->vq_avail_idx;
153 	flags = vq->vq_packed.cached_flags;
154 	desc[head].addr = cvq->virtio_net_hdr_mem;
155 	desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
156 	vq->vq_free_cnt--;
157 	nb_descs++;
158 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
159 		vq->vq_avail_idx -= vq->vq_nentries;
160 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
161 	}
162 
163 	for (k = 0; k < pkt_num; k++) {
164 		desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
165 			+ sizeof(struct virtio_net_ctrl_hdr)
166 			+ sizeof(ctrl->status) + sizeof(uint8_t) * sum;
167 		desc[vq->vq_avail_idx].len = dlen[k];
168 		desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
169 			vq->vq_packed.cached_flags;
170 		sum += dlen[k];
171 		vq->vq_free_cnt--;
172 		nb_descs++;
173 		if (++vq->vq_avail_idx >= vq->vq_nentries) {
174 			vq->vq_avail_idx -= vq->vq_nentries;
175 			vq->vq_packed.cached_flags ^=
176 				VRING_PACKED_DESC_F_AVAIL_USED;
177 		}
178 	}
179 
180 	desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
181 		+ sizeof(struct virtio_net_ctrl_hdr);
182 	desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
183 	desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
184 		vq->vq_packed.cached_flags;
185 	vq->vq_free_cnt--;
186 	nb_descs++;
187 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
188 		vq->vq_avail_idx -= vq->vq_nentries;
189 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
190 	}
191 
192 	virtqueue_store_flags_packed(&desc[head], VRING_DESC_F_NEXT | flags,
193 			vq->hw->weak_barriers);
194 
195 	virtio_wmb(vq->hw->weak_barriers);
196 	virtqueue_notify(vq);
197 
198 	/* wait for used desc in virtqueue
199 	 * desc_is_used has a load-acquire or rte_io_rmb inside
200 	 */
201 	while (!desc_is_used(&desc[head], vq))
202 		usleep(100);
203 
204 	/* now get used descriptors */
205 	vq->vq_free_cnt += nb_descs;
206 	vq->vq_used_cons_idx += nb_descs;
207 	if (vq->vq_used_cons_idx >= vq->vq_nentries) {
208 		vq->vq_used_cons_idx -= vq->vq_nentries;
209 		vq->vq_packed.used_wrap_counter ^= 1;
210 	}
211 
212 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
213 			"vq->vq_avail_idx=%d\n"
214 			"vq->vq_used_cons_idx=%d\n"
215 			"vq->vq_packed.cached_flags=0x%x\n"
216 			"vq->vq_packed.used_wrap_counter=%d",
217 			vq->vq_free_cnt,
218 			vq->vq_avail_idx,
219 			vq->vq_used_cons_idx,
220 			vq->vq_packed.cached_flags,
221 			vq->vq_packed.used_wrap_counter);
222 
223 	result = cvq->virtio_net_hdr_mz->addr;
224 	return result;
225 }
226 
227 static struct virtio_pmd_ctrl *
228 virtio_send_command_split(struct virtnet_ctl *cvq,
229 			  struct virtio_pmd_ctrl *ctrl,
230 			  int *dlen, int pkt_num)
231 {
232 	struct virtio_pmd_ctrl *result;
233 	struct virtqueue *vq = virtnet_cq_to_vq(cvq);
234 	uint32_t head, i;
235 	int k, sum = 0;
236 
237 	head = vq->vq_desc_head_idx;
238 
239 	/*
240 	 * Format is enforced in qemu code:
241 	 * One TX packet for header;
242 	 * At least one TX packet per argument;
243 	 * One RX packet for ACK.
244 	 */
245 	vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
246 	vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
247 	vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
248 	vq->vq_free_cnt--;
249 	i = vq->vq_split.ring.desc[head].next;
250 
251 	for (k = 0; k < pkt_num; k++) {
252 		vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
253 		vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
254 			+ sizeof(struct virtio_net_ctrl_hdr)
255 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
256 		vq->vq_split.ring.desc[i].len = dlen[k];
257 		sum += dlen[k];
258 		vq->vq_free_cnt--;
259 		i = vq->vq_split.ring.desc[i].next;
260 	}
261 
262 	vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
263 	vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
264 			+ sizeof(struct virtio_net_ctrl_hdr);
265 	vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
266 	vq->vq_free_cnt--;
267 
268 	vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
269 
270 	vq_update_avail_ring(vq, head);
271 	vq_update_avail_idx(vq);
272 
273 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
274 
275 	virtqueue_notify(vq);
276 
277 	while (virtqueue_nused(vq) == 0)
278 		usleep(100);
279 
280 	while (virtqueue_nused(vq)) {
281 		uint32_t idx, desc_idx, used_idx;
282 		struct vring_used_elem *uep;
283 
284 		used_idx = (uint32_t)(vq->vq_used_cons_idx
285 				& (vq->vq_nentries - 1));
286 		uep = &vq->vq_split.ring.used->ring[used_idx];
287 		idx = (uint32_t) uep->id;
288 		desc_idx = idx;
289 
290 		while (vq->vq_split.ring.desc[desc_idx].flags &
291 				VRING_DESC_F_NEXT) {
292 			desc_idx = vq->vq_split.ring.desc[desc_idx].next;
293 			vq->vq_free_cnt++;
294 		}
295 
296 		vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
297 		vq->vq_desc_head_idx = idx;
298 
299 		vq->vq_used_cons_idx++;
300 		vq->vq_free_cnt++;
301 	}
302 
303 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
304 			vq->vq_free_cnt, vq->vq_desc_head_idx);
305 
306 	result = cvq->virtio_net_hdr_mz->addr;
307 	return result;
308 }
309 
310 static int
311 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
312 		    int *dlen, int pkt_num)
313 {
314 	virtio_net_ctrl_ack status = ~0;
315 	struct virtio_pmd_ctrl *result;
316 	struct virtqueue *vq;
317 
318 	ctrl->status = status;
319 
320 	if (!cvq) {
321 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
322 		return -1;
323 	}
324 
325 	rte_spinlock_lock(&cvq->lock);
326 	vq = virtnet_cq_to_vq(cvq);
327 
328 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
329 		"vq->hw->cvq = %p vq = %p",
330 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
331 
332 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
333 		rte_spinlock_unlock(&cvq->lock);
334 		return -1;
335 	}
336 
337 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
338 		sizeof(struct virtio_pmd_ctrl));
339 
340 	if (virtio_with_packed_queue(vq->hw))
341 		result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
342 	else
343 		result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
344 
345 	rte_spinlock_unlock(&cvq->lock);
346 	return result->status;
347 }
348 
349 static int
350 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
351 {
352 	struct virtio_hw *hw = dev->data->dev_private;
353 	struct virtio_pmd_ctrl ctrl;
354 	int dlen[1];
355 	int ret;
356 
357 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
358 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
359 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
360 
361 	dlen[0] = sizeof(uint16_t);
362 
363 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
364 	if (ret) {
365 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
366 			  "failed, this is too late now...");
367 		return -EINVAL;
368 	}
369 
370 	return 0;
371 }
372 
373 static uint16_t
374 virtio_get_nr_vq(struct virtio_hw *hw)
375 {
376 	uint16_t nr_vq = hw->max_queue_pairs * 2;
377 
378 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
379 		nr_vq += 1;
380 
381 	return nr_vq;
382 }
383 
384 static void
385 virtio_init_vring(struct virtqueue *vq)
386 {
387 	int size = vq->vq_nentries;
388 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
389 
390 	PMD_INIT_FUNC_TRACE();
391 
392 	memset(ring_mem, 0, vq->vq_ring_size);
393 
394 	vq->vq_used_cons_idx = 0;
395 	vq->vq_desc_head_idx = 0;
396 	vq->vq_avail_idx = 0;
397 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
398 	vq->vq_free_cnt = vq->vq_nentries;
399 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
400 	if (virtio_with_packed_queue(vq->hw)) {
401 		vring_init_packed(&vq->vq_packed.ring, ring_mem,
402 				  VIRTIO_VRING_ALIGN, size);
403 		vring_desc_init_packed(vq, size);
404 	} else {
405 		struct vring *vr = &vq->vq_split.ring;
406 
407 		vring_init_split(vr, ring_mem, VIRTIO_VRING_ALIGN, size);
408 		vring_desc_init_split(vr->desc, size);
409 	}
410 	/*
411 	 * Disable device(host) interrupting guest
412 	 */
413 	virtqueue_disable_intr(vq);
414 }
415 
416 static int
417 virtio_init_queue(struct rte_eth_dev *dev, uint16_t queue_idx)
418 {
419 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
420 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
421 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
422 	unsigned int vq_size, size;
423 	struct virtio_hw *hw = dev->data->dev_private;
424 	struct virtnet_rx *rxvq = NULL;
425 	struct virtnet_tx *txvq = NULL;
426 	struct virtnet_ctl *cvq = NULL;
427 	struct virtqueue *vq;
428 	size_t sz_hdr_mz = 0;
429 	void *sw_ring = NULL;
430 	int queue_type = virtio_get_queue_type(hw, queue_idx);
431 	int ret;
432 	int numa_node = dev->device->numa_node;
433 	struct rte_mbuf *fake_mbuf = NULL;
434 
435 	PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
436 			queue_idx, numa_node);
437 
438 	/*
439 	 * Read the virtqueue size from the Queue Size field
440 	 * Always power of 2 and if 0 virtqueue does not exist
441 	 */
442 	vq_size = VIRTIO_OPS(hw)->get_queue_num(hw, queue_idx);
443 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
444 	if (vq_size == 0) {
445 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
446 		return -EINVAL;
447 	}
448 
449 	if (!virtio_with_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
450 		PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
451 		return -EINVAL;
452 	}
453 
454 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
455 		 dev->data->port_id, queue_idx);
456 
457 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
458 				vq_size * sizeof(struct vq_desc_extra),
459 				RTE_CACHE_LINE_SIZE);
460 	if (queue_type == VTNET_TQ) {
461 		/*
462 		 * For each xmit packet, allocate a virtio_net_hdr
463 		 * and indirect ring elements
464 		 */
465 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
466 	} else if (queue_type == VTNET_CQ) {
467 		/* Allocate a page for control vq command, data and status */
468 		sz_hdr_mz = rte_mem_page_size();
469 	}
470 
471 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
472 				numa_node);
473 	if (vq == NULL) {
474 		PMD_INIT_LOG(ERR, "can not allocate vq");
475 		return -ENOMEM;
476 	}
477 	hw->vqs[queue_idx] = vq;
478 
479 	vq->hw = hw;
480 	vq->vq_queue_index = queue_idx;
481 	vq->vq_nentries = vq_size;
482 	if (virtio_with_packed_queue(hw)) {
483 		vq->vq_packed.used_wrap_counter = 1;
484 		vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
485 		vq->vq_packed.event_flags_shadow = 0;
486 		if (queue_type == VTNET_RQ)
487 			vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
488 	}
489 
490 	/*
491 	 * Reserve a memzone for vring elements
492 	 */
493 	size = vring_size(hw, vq_size, VIRTIO_VRING_ALIGN);
494 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_VRING_ALIGN);
495 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
496 		     size, vq->vq_ring_size);
497 
498 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
499 			numa_node, RTE_MEMZONE_IOVA_CONTIG,
500 			VIRTIO_VRING_ALIGN);
501 	if (mz == NULL) {
502 		if (rte_errno == EEXIST)
503 			mz = rte_memzone_lookup(vq_name);
504 		if (mz == NULL) {
505 			ret = -ENOMEM;
506 			goto free_vq;
507 		}
508 	}
509 
510 	memset(mz->addr, 0, mz->len);
511 
512 	if (hw->use_va)
513 		vq->vq_ring_mem = (uintptr_t)mz->addr;
514 	else
515 		vq->vq_ring_mem = mz->iova;
516 
517 	vq->vq_ring_virt_mem = mz->addr;
518 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem: 0x%" PRIx64, vq->vq_ring_mem);
519 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: %p", vq->vq_ring_virt_mem);
520 
521 	virtio_init_vring(vq);
522 
523 	if (sz_hdr_mz) {
524 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
525 			 dev->data->port_id, queue_idx);
526 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
527 				numa_node, RTE_MEMZONE_IOVA_CONTIG,
528 				RTE_CACHE_LINE_SIZE);
529 		if (hdr_mz == NULL) {
530 			if (rte_errno == EEXIST)
531 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
532 			if (hdr_mz == NULL) {
533 				ret = -ENOMEM;
534 				goto free_mz;
535 			}
536 		}
537 	}
538 
539 	if (queue_type == VTNET_RQ) {
540 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
541 			       sizeof(vq->sw_ring[0]);
542 
543 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
544 				RTE_CACHE_LINE_SIZE, numa_node);
545 		if (!sw_ring) {
546 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
547 			ret = -ENOMEM;
548 			goto free_hdr_mz;
549 		}
550 
551 		fake_mbuf = rte_zmalloc_socket("sw_ring", sizeof(*fake_mbuf),
552 				RTE_CACHE_LINE_SIZE, numa_node);
553 		if (!fake_mbuf) {
554 			PMD_INIT_LOG(ERR, "can not allocate fake mbuf");
555 			ret = -ENOMEM;
556 			goto free_sw_ring;
557 		}
558 
559 		vq->sw_ring = sw_ring;
560 		rxvq = &vq->rxq;
561 		rxvq->port_id = dev->data->port_id;
562 		rxvq->mz = mz;
563 		rxvq->fake_mbuf = fake_mbuf;
564 	} else if (queue_type == VTNET_TQ) {
565 		txvq = &vq->txq;
566 		txvq->port_id = dev->data->port_id;
567 		txvq->mz = mz;
568 		txvq->virtio_net_hdr_mz = hdr_mz;
569 		if (hw->use_va)
570 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
571 		else
572 			txvq->virtio_net_hdr_mem = hdr_mz->iova;
573 	} else if (queue_type == VTNET_CQ) {
574 		cvq = &vq->cq;
575 		cvq->mz = mz;
576 		cvq->virtio_net_hdr_mz = hdr_mz;
577 		if (hw->use_va)
578 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
579 		else
580 			cvq->virtio_net_hdr_mem = hdr_mz->iova;
581 		memset(cvq->virtio_net_hdr_mz->addr, 0, rte_mem_page_size());
582 
583 		hw->cvq = cvq;
584 	}
585 
586 	if (hw->use_va)
587 		vq->mbuf_addr_offset = offsetof(struct rte_mbuf, buf_addr);
588 	else
589 		vq->mbuf_addr_offset = offsetof(struct rte_mbuf, buf_iova);
590 
591 	if (queue_type == VTNET_TQ) {
592 		struct virtio_tx_region *txr;
593 		unsigned int i;
594 
595 		txr = hdr_mz->addr;
596 		memset(txr, 0, vq_size * sizeof(*txr));
597 		for (i = 0; i < vq_size; i++) {
598 			/* first indirect descriptor is always the tx header */
599 			if (!virtio_with_packed_queue(hw)) {
600 				struct vring_desc *start_dp = txr[i].tx_indir;
601 				vring_desc_init_split(start_dp,
602 						      RTE_DIM(txr[i].tx_indir));
603 				start_dp->addr = txvq->virtio_net_hdr_mem
604 					+ i * sizeof(*txr)
605 					+ offsetof(struct virtio_tx_region,
606 						   tx_hdr);
607 				start_dp->len = hw->vtnet_hdr_size;
608 				start_dp->flags = VRING_DESC_F_NEXT;
609 			} else {
610 				struct vring_packed_desc *start_dp =
611 					txr[i].tx_packed_indir;
612 				vring_desc_init_indirect_packed(start_dp,
613 				      RTE_DIM(txr[i].tx_packed_indir));
614 				start_dp->addr = txvq->virtio_net_hdr_mem
615 					+ i * sizeof(*txr)
616 					+ offsetof(struct virtio_tx_region,
617 						   tx_hdr);
618 				start_dp->len = hw->vtnet_hdr_size;
619 			}
620 		}
621 	}
622 
623 	if (VIRTIO_OPS(hw)->setup_queue(hw, vq) < 0) {
624 		PMD_INIT_LOG(ERR, "setup_queue failed");
625 		ret = -EINVAL;
626 		goto clean_vq;
627 	}
628 
629 	return 0;
630 
631 clean_vq:
632 	hw->cvq = NULL;
633 	rte_free(fake_mbuf);
634 free_sw_ring:
635 	rte_free(sw_ring);
636 free_hdr_mz:
637 	rte_memzone_free(hdr_mz);
638 free_mz:
639 	rte_memzone_free(mz);
640 free_vq:
641 	rte_free(vq);
642 	hw->vqs[queue_idx] = NULL;
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->rxq.fake_mbuf);
666 			rte_free(vq->sw_ring);
667 			rte_memzone_free(vq->rxq.mz);
668 		} else if (queue_type == VTNET_TQ) {
669 			rte_memzone_free(vq->txq.mz);
670 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
671 		} else {
672 			rte_memzone_free(vq->cq.mz);
673 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
674 		}
675 
676 		rte_free(vq);
677 		hw->vqs[i] = NULL;
678 	}
679 
680 	rte_free(hw->vqs);
681 	hw->vqs = NULL;
682 }
683 
684 static int
685 virtio_alloc_queues(struct rte_eth_dev *dev)
686 {
687 	struct virtio_hw *hw = dev->data->dev_private;
688 	uint16_t nr_vq = virtio_get_nr_vq(hw);
689 	uint16_t i;
690 	int ret;
691 
692 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
693 	if (!hw->vqs) {
694 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
695 		return -ENOMEM;
696 	}
697 
698 	for (i = 0; i < nr_vq; i++) {
699 		ret = virtio_init_queue(dev, i);
700 		if (ret < 0) {
701 			virtio_free_queues(hw);
702 			return ret;
703 		}
704 	}
705 
706 	return 0;
707 }
708 
709 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
710 
711 int
712 virtio_dev_close(struct rte_eth_dev *dev)
713 {
714 	struct virtio_hw *hw = dev->data->dev_private;
715 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
716 
717 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
718 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
719 		return 0;
720 
721 	if (!hw->opened)
722 		return 0;
723 	hw->opened = 0;
724 
725 	/* reset the NIC */
726 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
727 		VIRTIO_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
728 	if (intr_conf->rxq)
729 		virtio_queues_unbind_intr(dev);
730 
731 	if (intr_conf->lsc || intr_conf->rxq) {
732 		virtio_intr_disable(dev);
733 		rte_intr_efd_disable(dev->intr_handle);
734 		rte_free(dev->intr_handle->intr_vec);
735 		dev->intr_handle->intr_vec = NULL;
736 	}
737 
738 	virtio_reset(hw);
739 	virtio_dev_free_mbufs(dev);
740 	virtio_free_queues(hw);
741 
742 	return VIRTIO_OPS(hw)->dev_close(hw);
743 }
744 
745 static int
746 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
747 {
748 	struct virtio_hw *hw = dev->data->dev_private;
749 	struct virtio_pmd_ctrl ctrl;
750 	int dlen[1];
751 	int ret;
752 
753 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
754 		PMD_INIT_LOG(INFO, "host does not support rx control");
755 		return -ENOTSUP;
756 	}
757 
758 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
759 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
760 	ctrl.data[0] = 1;
761 	dlen[0] = 1;
762 
763 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
764 	if (ret) {
765 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
766 		return -EAGAIN;
767 	}
768 
769 	return 0;
770 }
771 
772 static int
773 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
774 {
775 	struct virtio_hw *hw = dev->data->dev_private;
776 	struct virtio_pmd_ctrl ctrl;
777 	int dlen[1];
778 	int ret;
779 
780 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
781 		PMD_INIT_LOG(INFO, "host does not support rx control");
782 		return -ENOTSUP;
783 	}
784 
785 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
786 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
787 	ctrl.data[0] = 0;
788 	dlen[0] = 1;
789 
790 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
791 	if (ret) {
792 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
793 		return -EAGAIN;
794 	}
795 
796 	return 0;
797 }
798 
799 static int
800 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
801 {
802 	struct virtio_hw *hw = dev->data->dev_private;
803 	struct virtio_pmd_ctrl ctrl;
804 	int dlen[1];
805 	int ret;
806 
807 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
808 		PMD_INIT_LOG(INFO, "host does not support rx control");
809 		return -ENOTSUP;
810 	}
811 
812 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
813 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
814 	ctrl.data[0] = 1;
815 	dlen[0] = 1;
816 
817 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
818 	if (ret) {
819 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
820 		return -EAGAIN;
821 	}
822 
823 	return 0;
824 }
825 
826 static int
827 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
828 {
829 	struct virtio_hw *hw = dev->data->dev_private;
830 	struct virtio_pmd_ctrl ctrl;
831 	int dlen[1];
832 	int ret;
833 
834 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
835 		PMD_INIT_LOG(INFO, "host does not support rx control");
836 		return -ENOTSUP;
837 	}
838 
839 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
840 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
841 	ctrl.data[0] = 0;
842 	dlen[0] = 1;
843 
844 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
845 	if (ret) {
846 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
847 		return -EAGAIN;
848 	}
849 
850 	return 0;
851 }
852 
853 uint16_t
854 virtio_rx_mem_pool_buf_size(struct rte_mempool *mp)
855 {
856 	return rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
857 }
858 
859 bool
860 virtio_rx_check_scatter(uint16_t max_rx_pkt_len, uint16_t rx_buf_size,
861 			bool rx_scatter_enabled, const char **error)
862 {
863 	if (!rx_scatter_enabled && max_rx_pkt_len > rx_buf_size) {
864 		*error = "Rx scatter is disabled and RxQ mbuf pool object size is too small";
865 		return false;
866 	}
867 
868 	return true;
869 }
870 
871 static bool
872 virtio_check_scatter_on_all_rx_queues(struct rte_eth_dev *dev,
873 				      uint16_t frame_size)
874 {
875 	struct virtio_hw *hw = dev->data->dev_private;
876 	struct virtnet_rx *rxvq;
877 	struct virtqueue *vq;
878 	unsigned int qidx;
879 	uint16_t buf_size;
880 	const char *error;
881 
882 	if (hw->vqs == NULL)
883 		return true;
884 
885 	for (qidx = 0; (vq = hw->vqs[2 * qidx + VTNET_SQ_RQ_QUEUE_IDX]) != NULL;
886 	     qidx++) {
887 		rxvq = &vq->rxq;
888 		if (rxvq->mpool == NULL)
889 			continue;
890 		buf_size = virtio_rx_mem_pool_buf_size(rxvq->mpool);
891 
892 		if (!virtio_rx_check_scatter(frame_size, buf_size,
893 					     hw->rx_ol_scatter, &error)) {
894 			PMD_INIT_LOG(ERR, "MTU check for RxQ %u failed: %s",
895 				     qidx, error);
896 			return false;
897 		}
898 	}
899 
900 	return true;
901 }
902 
903 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
904 static int
905 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
906 {
907 	struct virtio_hw *hw = dev->data->dev_private;
908 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
909 				 hw->vtnet_hdr_size;
910 	uint32_t frame_size = mtu + ether_hdr_len;
911 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
912 
913 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
914 
915 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
916 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
917 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
918 		return -EINVAL;
919 	}
920 
921 	if (!virtio_check_scatter_on_all_rx_queues(dev, frame_size)) {
922 		PMD_INIT_LOG(ERR, "MTU vs Rx scatter and Rx buffers check failed");
923 		return -EINVAL;
924 	}
925 
926 	hw->max_rx_pkt_len = frame_size;
927 
928 	return 0;
929 }
930 
931 static int
932 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
933 {
934 	struct virtio_hw *hw = dev->data->dev_private;
935 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
936 	struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
937 
938 	virtqueue_enable_intr(vq);
939 	virtio_mb(hw->weak_barriers);
940 	return 0;
941 }
942 
943 static int
944 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
945 {
946 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
947 	struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
948 
949 	virtqueue_disable_intr(vq);
950 	return 0;
951 }
952 
953 /*
954  * dev_ops for virtio, bare necessities for basic operation
955  */
956 static const struct eth_dev_ops virtio_eth_dev_ops = {
957 	.dev_configure           = virtio_dev_configure,
958 	.dev_start               = virtio_dev_start,
959 	.dev_stop                = virtio_dev_stop,
960 	.dev_close               = virtio_dev_close,
961 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
962 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
963 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
964 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
965 	.mtu_set                 = virtio_mtu_set,
966 	.dev_infos_get           = virtio_dev_info_get,
967 	.stats_get               = virtio_dev_stats_get,
968 	.xstats_get              = virtio_dev_xstats_get,
969 	.xstats_get_names        = virtio_dev_xstats_get_names,
970 	.stats_reset             = virtio_dev_stats_reset,
971 	.xstats_reset            = virtio_dev_stats_reset,
972 	.link_update             = virtio_dev_link_update,
973 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
974 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
975 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
976 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
977 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
978 	/* collect stats per queue */
979 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
980 	.vlan_filter_set         = virtio_vlan_filter_set,
981 	.mac_addr_add            = virtio_mac_addr_add,
982 	.mac_addr_remove         = virtio_mac_addr_remove,
983 	.mac_addr_set            = virtio_mac_addr_set,
984 };
985 
986 /*
987  * dev_ops for virtio-user in secondary processes, as we just have
988  * some limited supports currently.
989  */
990 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
991 	.dev_infos_get           = virtio_dev_info_get,
992 	.stats_get               = virtio_dev_stats_get,
993 	.xstats_get              = virtio_dev_xstats_get,
994 	.xstats_get_names        = virtio_dev_xstats_get_names,
995 	.stats_reset             = virtio_dev_stats_reset,
996 	.xstats_reset            = virtio_dev_stats_reset,
997 	/* collect stats per queue */
998 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
999 };
1000 
1001 static void
1002 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1003 {
1004 	unsigned i;
1005 
1006 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1007 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
1008 		if (txvq == NULL)
1009 			continue;
1010 
1011 		stats->opackets += txvq->stats.packets;
1012 		stats->obytes += txvq->stats.bytes;
1013 
1014 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1015 			stats->q_opackets[i] = txvq->stats.packets;
1016 			stats->q_obytes[i] = txvq->stats.bytes;
1017 		}
1018 	}
1019 
1020 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1021 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1022 		if (rxvq == NULL)
1023 			continue;
1024 
1025 		stats->ipackets += rxvq->stats.packets;
1026 		stats->ibytes += rxvq->stats.bytes;
1027 		stats->ierrors += rxvq->stats.errors;
1028 
1029 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1030 			stats->q_ipackets[i] = rxvq->stats.packets;
1031 			stats->q_ibytes[i] = rxvq->stats.bytes;
1032 		}
1033 	}
1034 
1035 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
1036 }
1037 
1038 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
1039 				       struct rte_eth_xstat_name *xstats_names,
1040 				       __rte_unused unsigned limit)
1041 {
1042 	unsigned i;
1043 	unsigned count = 0;
1044 	unsigned t;
1045 
1046 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1047 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1048 
1049 	if (xstats_names != NULL) {
1050 		/* Note: limit checked in rte_eth_xstats_names() */
1051 
1052 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
1053 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1054 			if (rxvq == NULL)
1055 				continue;
1056 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1057 				snprintf(xstats_names[count].name,
1058 					sizeof(xstats_names[count].name),
1059 					"rx_q%u_%s", i,
1060 					rte_virtio_rxq_stat_strings[t].name);
1061 				count++;
1062 			}
1063 		}
1064 
1065 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
1066 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
1067 			if (txvq == NULL)
1068 				continue;
1069 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1070 				snprintf(xstats_names[count].name,
1071 					sizeof(xstats_names[count].name),
1072 					"tx_q%u_%s", i,
1073 					rte_virtio_txq_stat_strings[t].name);
1074 				count++;
1075 			}
1076 		}
1077 		return count;
1078 	}
1079 	return nstats;
1080 }
1081 
1082 static int
1083 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1084 		      unsigned n)
1085 {
1086 	unsigned i;
1087 	unsigned count = 0;
1088 
1089 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1090 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1091 
1092 	if (n < nstats)
1093 		return nstats;
1094 
1095 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1096 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1097 
1098 		if (rxvq == NULL)
1099 			continue;
1100 
1101 		unsigned t;
1102 
1103 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1104 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1105 				rte_virtio_rxq_stat_strings[t].offset);
1106 			xstats[count].id = count;
1107 			count++;
1108 		}
1109 	}
1110 
1111 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1112 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1113 
1114 		if (txvq == NULL)
1115 			continue;
1116 
1117 		unsigned t;
1118 
1119 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1120 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1121 				rte_virtio_txq_stat_strings[t].offset);
1122 			xstats[count].id = count;
1123 			count++;
1124 		}
1125 	}
1126 
1127 	return count;
1128 }
1129 
1130 static int
1131 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1132 {
1133 	virtio_update_stats(dev, stats);
1134 
1135 	return 0;
1136 }
1137 
1138 static int
1139 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1140 {
1141 	unsigned int i;
1142 
1143 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1144 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1145 		if (txvq == NULL)
1146 			continue;
1147 
1148 		txvq->stats.packets = 0;
1149 		txvq->stats.bytes = 0;
1150 		txvq->stats.multicast = 0;
1151 		txvq->stats.broadcast = 0;
1152 		memset(txvq->stats.size_bins, 0,
1153 		       sizeof(txvq->stats.size_bins[0]) * 8);
1154 	}
1155 
1156 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1157 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1158 		if (rxvq == NULL)
1159 			continue;
1160 
1161 		rxvq->stats.packets = 0;
1162 		rxvq->stats.bytes = 0;
1163 		rxvq->stats.errors = 0;
1164 		rxvq->stats.multicast = 0;
1165 		rxvq->stats.broadcast = 0;
1166 		memset(rxvq->stats.size_bins, 0,
1167 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1168 	}
1169 
1170 	return 0;
1171 }
1172 
1173 static void
1174 virtio_set_hwaddr(struct virtio_hw *hw)
1175 {
1176 	virtio_write_dev_config(hw,
1177 			offsetof(struct virtio_net_config, mac),
1178 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1179 }
1180 
1181 static void
1182 virtio_get_hwaddr(struct virtio_hw *hw)
1183 {
1184 	if (virtio_with_feature(hw, VIRTIO_NET_F_MAC)) {
1185 		virtio_read_dev_config(hw,
1186 			offsetof(struct virtio_net_config, mac),
1187 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1188 	} else {
1189 		rte_eth_random_addr(&hw->mac_addr[0]);
1190 		virtio_set_hwaddr(hw);
1191 	}
1192 }
1193 
1194 static int
1195 virtio_mac_table_set(struct virtio_hw *hw,
1196 		     const struct virtio_net_ctrl_mac *uc,
1197 		     const struct virtio_net_ctrl_mac *mc)
1198 {
1199 	struct virtio_pmd_ctrl ctrl;
1200 	int err, len[2];
1201 
1202 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1203 		PMD_DRV_LOG(INFO, "host does not support mac table");
1204 		return -1;
1205 	}
1206 
1207 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1208 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1209 
1210 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1211 	memcpy(ctrl.data, uc, len[0]);
1212 
1213 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1214 	memcpy(ctrl.data + len[0], mc, len[1]);
1215 
1216 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1217 	if (err != 0)
1218 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1219 	return err;
1220 }
1221 
1222 static int
1223 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1224 		    uint32_t index, uint32_t vmdq __rte_unused)
1225 {
1226 	struct virtio_hw *hw = dev->data->dev_private;
1227 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1228 	unsigned int i;
1229 	struct virtio_net_ctrl_mac *uc, *mc;
1230 
1231 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1232 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1233 		return -EINVAL;
1234 	}
1235 
1236 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1237 		sizeof(uc->entries));
1238 	uc->entries = 0;
1239 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1240 		sizeof(mc->entries));
1241 	mc->entries = 0;
1242 
1243 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1244 		const struct rte_ether_addr *addr
1245 			= (i == index) ? mac_addr : addrs + i;
1246 		struct virtio_net_ctrl_mac *tbl
1247 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1248 
1249 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1250 	}
1251 
1252 	return virtio_mac_table_set(hw, uc, mc);
1253 }
1254 
1255 static void
1256 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1257 {
1258 	struct virtio_hw *hw = dev->data->dev_private;
1259 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1260 	struct virtio_net_ctrl_mac *uc, *mc;
1261 	unsigned int i;
1262 
1263 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1264 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1265 		return;
1266 	}
1267 
1268 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1269 		sizeof(uc->entries));
1270 	uc->entries = 0;
1271 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1272 		sizeof(mc->entries));
1273 	mc->entries = 0;
1274 
1275 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1276 		struct virtio_net_ctrl_mac *tbl;
1277 
1278 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1279 			continue;
1280 
1281 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1282 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1283 			RTE_ETHER_ADDR_LEN);
1284 	}
1285 
1286 	virtio_mac_table_set(hw, uc, mc);
1287 }
1288 
1289 static int
1290 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1291 {
1292 	struct virtio_hw *hw = dev->data->dev_private;
1293 
1294 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1295 
1296 	/* Use atomic update if available */
1297 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1298 		struct virtio_pmd_ctrl ctrl;
1299 		int len = RTE_ETHER_ADDR_LEN;
1300 
1301 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1302 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1303 
1304 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1305 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1306 	}
1307 
1308 	if (!virtio_with_feature(hw, VIRTIO_NET_F_MAC))
1309 		return -ENOTSUP;
1310 
1311 	virtio_set_hwaddr(hw);
1312 	return 0;
1313 }
1314 
1315 static int
1316 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1317 {
1318 	struct virtio_hw *hw = dev->data->dev_private;
1319 	struct virtio_pmd_ctrl ctrl;
1320 	int len;
1321 
1322 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1323 		return -ENOTSUP;
1324 
1325 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1326 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1327 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1328 	len = sizeof(vlan_id);
1329 
1330 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1331 }
1332 
1333 static int
1334 virtio_intr_unmask(struct rte_eth_dev *dev)
1335 {
1336 	struct virtio_hw *hw = dev->data->dev_private;
1337 
1338 	if (rte_intr_ack(dev->intr_handle) < 0)
1339 		return -1;
1340 
1341 	if (VIRTIO_OPS(hw)->intr_detect)
1342 		VIRTIO_OPS(hw)->intr_detect(hw);
1343 
1344 	return 0;
1345 }
1346 
1347 static int
1348 virtio_intr_enable(struct rte_eth_dev *dev)
1349 {
1350 	struct virtio_hw *hw = dev->data->dev_private;
1351 
1352 	if (rte_intr_enable(dev->intr_handle) < 0)
1353 		return -1;
1354 
1355 	if (VIRTIO_OPS(hw)->intr_detect)
1356 		VIRTIO_OPS(hw)->intr_detect(hw);
1357 
1358 	return 0;
1359 }
1360 
1361 static int
1362 virtio_intr_disable(struct rte_eth_dev *dev)
1363 {
1364 	struct virtio_hw *hw = dev->data->dev_private;
1365 
1366 	if (rte_intr_disable(dev->intr_handle) < 0)
1367 		return -1;
1368 
1369 	if (VIRTIO_OPS(hw)->intr_detect)
1370 		VIRTIO_OPS(hw)->intr_detect(hw);
1371 
1372 	return 0;
1373 }
1374 
1375 static int
1376 virtio_ethdev_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1377 {
1378 	uint64_t host_features;
1379 
1380 	/* Prepare guest_features: feature that driver wants to support */
1381 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1382 		req_features);
1383 
1384 	/* Read device(host) feature bits */
1385 	host_features = VIRTIO_OPS(hw)->get_features(hw);
1386 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1387 		host_features);
1388 
1389 	/* If supported, ensure MTU value is valid before acknowledging it. */
1390 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1391 		struct virtio_net_config config;
1392 
1393 		virtio_read_dev_config(hw,
1394 			offsetof(struct virtio_net_config, mtu),
1395 			&config.mtu, sizeof(config.mtu));
1396 
1397 		if (config.mtu < RTE_ETHER_MIN_MTU)
1398 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1399 	}
1400 
1401 	/*
1402 	 * Negotiate features: Subset of device feature bits are written back
1403 	 * guest feature bits.
1404 	 */
1405 	hw->guest_features = req_features;
1406 	hw->guest_features = virtio_negotiate_features(hw, host_features);
1407 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1408 		hw->guest_features);
1409 
1410 	if (VIRTIO_OPS(hw)->features_ok(hw) < 0)
1411 		return -1;
1412 
1413 	if (virtio_with_feature(hw, VIRTIO_F_VERSION_1)) {
1414 		virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1415 
1416 		if (!(virtio_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1417 			PMD_INIT_LOG(ERR, "Failed to set FEATURES_OK status!");
1418 			return -1;
1419 		}
1420 	}
1421 
1422 	hw->req_guest_features = req_features;
1423 
1424 	return 0;
1425 }
1426 
1427 int
1428 virtio_dev_pause(struct rte_eth_dev *dev)
1429 {
1430 	struct virtio_hw *hw = dev->data->dev_private;
1431 
1432 	rte_spinlock_lock(&hw->state_lock);
1433 
1434 	if (hw->started == 0) {
1435 		/* Device is just stopped. */
1436 		rte_spinlock_unlock(&hw->state_lock);
1437 		return -1;
1438 	}
1439 	hw->started = 0;
1440 	/*
1441 	 * Prevent the worker threads from touching queues to avoid contention,
1442 	 * 1 ms should be enough for the ongoing Tx function to finish.
1443 	 */
1444 	rte_delay_ms(1);
1445 	return 0;
1446 }
1447 
1448 /*
1449  * Recover hw state to let the worker threads continue.
1450  */
1451 void
1452 virtio_dev_resume(struct rte_eth_dev *dev)
1453 {
1454 	struct virtio_hw *hw = dev->data->dev_private;
1455 
1456 	hw->started = 1;
1457 	rte_spinlock_unlock(&hw->state_lock);
1458 }
1459 
1460 /*
1461  * Should be called only after device is paused.
1462  */
1463 int
1464 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1465 		int nb_pkts)
1466 {
1467 	struct virtio_hw *hw = dev->data->dev_private;
1468 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1469 	int ret;
1470 
1471 	hw->inject_pkts = tx_pkts;
1472 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1473 	hw->inject_pkts = NULL;
1474 
1475 	return ret;
1476 }
1477 
1478 static void
1479 virtio_notify_peers(struct rte_eth_dev *dev)
1480 {
1481 	struct virtio_hw *hw = dev->data->dev_private;
1482 	struct virtnet_rx *rxvq;
1483 	struct rte_mbuf *rarp_mbuf;
1484 
1485 	if (!dev->data->rx_queues)
1486 		return;
1487 
1488 	rxvq = dev->data->rx_queues[0];
1489 	if (!rxvq)
1490 		return;
1491 
1492 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1493 			(struct rte_ether_addr *)hw->mac_addr);
1494 	if (rarp_mbuf == NULL) {
1495 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1496 		return;
1497 	}
1498 
1499 	/* If virtio port just stopped, no need to send RARP */
1500 	if (virtio_dev_pause(dev) < 0) {
1501 		rte_pktmbuf_free(rarp_mbuf);
1502 		return;
1503 	}
1504 
1505 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1506 	virtio_dev_resume(dev);
1507 }
1508 
1509 static void
1510 virtio_ack_link_announce(struct rte_eth_dev *dev)
1511 {
1512 	struct virtio_hw *hw = dev->data->dev_private;
1513 	struct virtio_pmd_ctrl ctrl;
1514 
1515 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1516 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1517 
1518 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1519 }
1520 
1521 /*
1522  * Process virtio config changed interrupt. Call the callback
1523  * if link state changed, generate gratuitous RARP packet if
1524  * the status indicates an ANNOUNCE.
1525  */
1526 void
1527 virtio_interrupt_handler(void *param)
1528 {
1529 	struct rte_eth_dev *dev = param;
1530 	struct virtio_hw *hw = dev->data->dev_private;
1531 	uint8_t isr;
1532 	uint16_t status;
1533 
1534 	/* Read interrupt status which clears interrupt */
1535 	isr = virtio_get_isr(hw);
1536 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1537 
1538 	if (virtio_intr_unmask(dev) < 0)
1539 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1540 
1541 	if (isr & VIRTIO_ISR_CONFIG) {
1542 		if (virtio_dev_link_update(dev, 0) == 0)
1543 			rte_eth_dev_callback_process(dev,
1544 						     RTE_ETH_EVENT_INTR_LSC,
1545 						     NULL);
1546 
1547 		if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1548 			virtio_read_dev_config(hw,
1549 				offsetof(struct virtio_net_config, status),
1550 				&status, sizeof(status));
1551 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1552 				virtio_notify_peers(dev);
1553 				if (hw->cvq)
1554 					virtio_ack_link_announce(dev);
1555 			}
1556 		}
1557 	}
1558 }
1559 
1560 /* set rx and tx handlers according to what is supported */
1561 static void
1562 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1563 {
1564 	struct virtio_hw *hw = eth_dev->data->dev_private;
1565 
1566 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1567 	if (virtio_with_packed_queue(hw)) {
1568 		PMD_INIT_LOG(INFO,
1569 			"virtio: using packed ring %s Tx path on port %u",
1570 			hw->use_vec_tx ? "vectorized" : "standard",
1571 			eth_dev->data->port_id);
1572 		if (hw->use_vec_tx)
1573 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1574 		else
1575 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1576 	} else {
1577 		if (hw->use_inorder_tx) {
1578 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1579 				eth_dev->data->port_id);
1580 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1581 		} else {
1582 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1583 				eth_dev->data->port_id);
1584 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1585 		}
1586 	}
1587 
1588 	if (virtio_with_packed_queue(hw)) {
1589 		if (hw->use_vec_rx) {
1590 			PMD_INIT_LOG(INFO,
1591 				"virtio: using packed ring vectorized Rx path on port %u",
1592 				eth_dev->data->port_id);
1593 			eth_dev->rx_pkt_burst =
1594 				&virtio_recv_pkts_packed_vec;
1595 		} else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1596 			PMD_INIT_LOG(INFO,
1597 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1598 				eth_dev->data->port_id);
1599 			eth_dev->rx_pkt_burst =
1600 				&virtio_recv_mergeable_pkts_packed;
1601 		} else {
1602 			PMD_INIT_LOG(INFO,
1603 				"virtio: using packed ring standard Rx path on port %u",
1604 				eth_dev->data->port_id);
1605 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1606 		}
1607 	} else {
1608 		if (hw->use_vec_rx) {
1609 			PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1610 				eth_dev->data->port_id);
1611 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1612 		} else if (hw->use_inorder_rx) {
1613 			PMD_INIT_LOG(INFO,
1614 				"virtio: using inorder Rx path on port %u",
1615 				eth_dev->data->port_id);
1616 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1617 		} else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1618 			PMD_INIT_LOG(INFO,
1619 				"virtio: using mergeable buffer Rx path on port %u",
1620 				eth_dev->data->port_id);
1621 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1622 		} else {
1623 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1624 				eth_dev->data->port_id);
1625 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1626 		}
1627 	}
1628 
1629 }
1630 
1631 /* Only support 1:1 queue/interrupt mapping so far.
1632  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1633  * interrupt vectors (<N+1).
1634  */
1635 static int
1636 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1637 {
1638 	uint32_t i;
1639 	struct virtio_hw *hw = dev->data->dev_private;
1640 
1641 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1642 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1643 		dev->intr_handle->intr_vec[i] = i + 1;
1644 		if (VIRTIO_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1645 						 VIRTIO_MSI_NO_VECTOR) {
1646 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1647 			return -EBUSY;
1648 		}
1649 	}
1650 
1651 	return 0;
1652 }
1653 
1654 static void
1655 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1656 {
1657 	uint32_t i;
1658 	struct virtio_hw *hw = dev->data->dev_private;
1659 
1660 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1661 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1662 		VIRTIO_OPS(hw)->set_queue_irq(hw,
1663 					     hw->vqs[i * VTNET_CQ],
1664 					     VIRTIO_MSI_NO_VECTOR);
1665 }
1666 
1667 static int
1668 virtio_configure_intr(struct rte_eth_dev *dev)
1669 {
1670 	struct virtio_hw *hw = dev->data->dev_private;
1671 
1672 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1673 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1674 		return -ENOTSUP;
1675 	}
1676 
1677 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1678 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1679 		return -1;
1680 	}
1681 
1682 	if (!dev->intr_handle->intr_vec) {
1683 		dev->intr_handle->intr_vec =
1684 			rte_zmalloc("intr_vec",
1685 				    hw->max_queue_pairs * sizeof(int), 0);
1686 		if (!dev->intr_handle->intr_vec) {
1687 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1688 				     hw->max_queue_pairs);
1689 			return -ENOMEM;
1690 		}
1691 	}
1692 
1693 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1694 		/* Re-register callback to update max_intr */
1695 		rte_intr_callback_unregister(dev->intr_handle,
1696 					     virtio_interrupt_handler,
1697 					     dev);
1698 		rte_intr_callback_register(dev->intr_handle,
1699 					   virtio_interrupt_handler,
1700 					   dev);
1701 	}
1702 
1703 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1704 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1705 	 * And for legacy devices, this should be done before queue/vec binding
1706 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1707 	 * (22) will be ignored.
1708 	 */
1709 	if (virtio_intr_enable(dev) < 0) {
1710 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1711 		return -1;
1712 	}
1713 
1714 	if (virtio_queues_bind_intr(dev) < 0) {
1715 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1716 		return -1;
1717 	}
1718 
1719 	return 0;
1720 }
1721 #define DUPLEX_UNKNOWN   0xff
1722 /* reset device and renegotiate features if needed */
1723 static int
1724 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1725 {
1726 	struct virtio_hw *hw = eth_dev->data->dev_private;
1727 	struct virtio_net_config *config;
1728 	struct virtio_net_config local_config;
1729 	int ret;
1730 
1731 	/* Reset the device although not necessary at startup */
1732 	virtio_reset(hw);
1733 
1734 	if (hw->vqs) {
1735 		virtio_dev_free_mbufs(eth_dev);
1736 		virtio_free_queues(hw);
1737 	}
1738 
1739 	/* Tell the host we've noticed this device. */
1740 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1741 
1742 	/* Tell the host we've known how to drive the device. */
1743 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1744 	if (virtio_ethdev_negotiate_features(hw, req_features) < 0)
1745 		return -1;
1746 
1747 	hw->weak_barriers = !virtio_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1748 
1749 	/* If host does not support both status and MSI-X then disable LSC */
1750 	if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->intr_lsc)
1751 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1752 	else
1753 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1754 
1755 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1756 
1757 	/* Setting up rx_header size for the device */
1758 	if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1759 	    virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
1760 	    virtio_with_packed_queue(hw))
1761 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1762 	else
1763 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1764 
1765 	/* Copy the permanent MAC address to: virtio_hw */
1766 	virtio_get_hwaddr(hw);
1767 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1768 			&eth_dev->data->mac_addrs[0]);
1769 	PMD_INIT_LOG(DEBUG,
1770 		     "PORT MAC: " RTE_ETHER_ADDR_PRT_FMT,
1771 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1772 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1773 
1774 	if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1775 		if (virtio_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1776 			config = &local_config;
1777 			virtio_read_dev_config(hw,
1778 				offsetof(struct virtio_net_config, speed),
1779 				&config->speed, sizeof(config->speed));
1780 			virtio_read_dev_config(hw,
1781 				offsetof(struct virtio_net_config, duplex),
1782 				&config->duplex, sizeof(config->duplex));
1783 			hw->speed = config->speed;
1784 			hw->duplex = config->duplex;
1785 		}
1786 	}
1787 	if (hw->duplex == DUPLEX_UNKNOWN)
1788 		hw->duplex = ETH_LINK_FULL_DUPLEX;
1789 	PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1790 		hw->speed, hw->duplex);
1791 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1792 		config = &local_config;
1793 
1794 		virtio_read_dev_config(hw,
1795 			offsetof(struct virtio_net_config, mac),
1796 			&config->mac, sizeof(config->mac));
1797 
1798 		if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1799 			virtio_read_dev_config(hw,
1800 				offsetof(struct virtio_net_config, status),
1801 				&config->status, sizeof(config->status));
1802 		} else {
1803 			PMD_INIT_LOG(DEBUG,
1804 				     "VIRTIO_NET_F_STATUS is not supported");
1805 			config->status = 0;
1806 		}
1807 
1808 		if (virtio_with_feature(hw, VIRTIO_NET_F_MQ)) {
1809 			virtio_read_dev_config(hw,
1810 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1811 				&config->max_virtqueue_pairs,
1812 				sizeof(config->max_virtqueue_pairs));
1813 		} else {
1814 			PMD_INIT_LOG(DEBUG,
1815 				     "VIRTIO_NET_F_MQ is not supported");
1816 			config->max_virtqueue_pairs = 1;
1817 		}
1818 
1819 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1820 
1821 		if (virtio_with_feature(hw, VIRTIO_NET_F_MTU)) {
1822 			virtio_read_dev_config(hw,
1823 				offsetof(struct virtio_net_config, mtu),
1824 				&config->mtu,
1825 				sizeof(config->mtu));
1826 
1827 			/*
1828 			 * MTU value has already been checked at negotiation
1829 			 * time, but check again in case it has changed since
1830 			 * then, which should not happen.
1831 			 */
1832 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1833 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1834 						config->mtu);
1835 				return -1;
1836 			}
1837 
1838 			hw->max_mtu = config->mtu;
1839 			/* Set initial MTU to maximum one supported by vhost */
1840 			eth_dev->data->mtu = config->mtu;
1841 
1842 		} else {
1843 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1844 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1845 		}
1846 
1847 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1848 				config->max_virtqueue_pairs);
1849 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1850 		PMD_INIT_LOG(DEBUG,
1851 				"PORT MAC: " RTE_ETHER_ADDR_PRT_FMT,
1852 				config->mac[0], config->mac[1],
1853 				config->mac[2], config->mac[3],
1854 				config->mac[4], config->mac[5]);
1855 	} else {
1856 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1857 		hw->max_queue_pairs = 1;
1858 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1859 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1860 	}
1861 
1862 	ret = virtio_alloc_queues(eth_dev);
1863 	if (ret < 0)
1864 		return ret;
1865 
1866 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1867 		if (virtio_configure_intr(eth_dev) < 0) {
1868 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1869 			virtio_free_queues(hw);
1870 			return -1;
1871 		}
1872 	}
1873 
1874 	virtio_reinit_complete(hw);
1875 
1876 	return 0;
1877 }
1878 
1879 /*
1880  * This function is based on probe() function in virtio_pci.c
1881  * It returns 0 on success.
1882  */
1883 int
1884 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1885 {
1886 	struct virtio_hw *hw = eth_dev->data->dev_private;
1887 	uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1888 	int vectorized = 0;
1889 	int ret;
1890 
1891 	if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1892 		PMD_INIT_LOG(ERR,
1893 			"Not sufficient headroom required = %d, avail = %d",
1894 			(int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1895 			RTE_PKTMBUF_HEADROOM);
1896 
1897 		return -1;
1898 	}
1899 
1900 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1901 
1902 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1903 		set_rxtx_funcs(eth_dev);
1904 		return 0;
1905 	}
1906 
1907 	ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1908 	if (ret < 0)
1909 		return ret;
1910 	hw->speed = speed;
1911 	hw->duplex = DUPLEX_UNKNOWN;
1912 
1913 	/* Allocate memory for storing MAC addresses */
1914 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1915 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1916 	if (eth_dev->data->mac_addrs == NULL) {
1917 		PMD_INIT_LOG(ERR,
1918 			"Failed to allocate %d bytes needed to store MAC addresses",
1919 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1920 		return -ENOMEM;
1921 	}
1922 
1923 	rte_spinlock_init(&hw->state_lock);
1924 
1925 	/* reset device and negotiate default features */
1926 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1927 	if (ret < 0)
1928 		goto err_virtio_init;
1929 
1930 	if (vectorized) {
1931 		if (!virtio_with_packed_queue(hw)) {
1932 			hw->use_vec_rx = 1;
1933 		} else {
1934 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1935 			hw->use_vec_rx = 1;
1936 			hw->use_vec_tx = 1;
1937 #else
1938 			PMD_DRV_LOG(INFO,
1939 				"building environment do not support packed ring vectorized");
1940 #endif
1941 		}
1942 	}
1943 
1944 	hw->opened = 1;
1945 
1946 	return 0;
1947 
1948 err_virtio_init:
1949 	rte_free(eth_dev->data->mac_addrs);
1950 	eth_dev->data->mac_addrs = NULL;
1951 	return ret;
1952 }
1953 
1954 static uint32_t
1955 virtio_dev_speed_capa_get(uint32_t speed)
1956 {
1957 	switch (speed) {
1958 	case ETH_SPEED_NUM_10G:
1959 		return ETH_LINK_SPEED_10G;
1960 	case ETH_SPEED_NUM_20G:
1961 		return ETH_LINK_SPEED_20G;
1962 	case ETH_SPEED_NUM_25G:
1963 		return ETH_LINK_SPEED_25G;
1964 	case ETH_SPEED_NUM_40G:
1965 		return ETH_LINK_SPEED_40G;
1966 	case ETH_SPEED_NUM_50G:
1967 		return ETH_LINK_SPEED_50G;
1968 	case ETH_SPEED_NUM_56G:
1969 		return ETH_LINK_SPEED_56G;
1970 	case ETH_SPEED_NUM_100G:
1971 		return ETH_LINK_SPEED_100G;
1972 	case ETH_SPEED_NUM_200G:
1973 		return ETH_LINK_SPEED_200G;
1974 	default:
1975 		return 0;
1976 	}
1977 }
1978 
1979 static int vectorized_check_handler(__rte_unused const char *key,
1980 		const char *value, void *ret_val)
1981 {
1982 	if (strcmp(value, "1") == 0)
1983 		*(int *)ret_val = 1;
1984 	else
1985 		*(int *)ret_val = 0;
1986 
1987 	return 0;
1988 }
1989 
1990 #define VIRTIO_ARG_SPEED      "speed"
1991 #define VIRTIO_ARG_VECTORIZED "vectorized"
1992 
1993 static int
1994 link_speed_handler(const char *key __rte_unused,
1995 		const char *value, void *ret_val)
1996 {
1997 	uint32_t val;
1998 	if (!value || !ret_val)
1999 		return -EINVAL;
2000 	val = strtoul(value, NULL, 0);
2001 	/* validate input */
2002 	if (virtio_dev_speed_capa_get(val) == 0)
2003 		return -EINVAL;
2004 	*(uint32_t *)ret_val = val;
2005 
2006 	return 0;
2007 }
2008 
2009 
2010 static int
2011 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
2012 {
2013 	struct rte_kvargs *kvlist;
2014 	int ret = 0;
2015 
2016 	if (devargs == NULL)
2017 		return 0;
2018 
2019 	kvlist = rte_kvargs_parse(devargs->args, NULL);
2020 	if (kvlist == NULL) {
2021 		PMD_INIT_LOG(ERR, "error when parsing param");
2022 		return 0;
2023 	}
2024 
2025 	if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2026 		ret = rte_kvargs_process(kvlist,
2027 					VIRTIO_ARG_SPEED,
2028 					link_speed_handler, speed);
2029 		if (ret < 0) {
2030 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2031 					VIRTIO_ARG_SPEED);
2032 			goto exit;
2033 		}
2034 	}
2035 
2036 	if (vectorized &&
2037 		rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2038 		ret = rte_kvargs_process(kvlist,
2039 				VIRTIO_ARG_VECTORIZED,
2040 				vectorized_check_handler, vectorized);
2041 		if (ret < 0) {
2042 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2043 					VIRTIO_ARG_VECTORIZED);
2044 			goto exit;
2045 		}
2046 	}
2047 
2048 exit:
2049 	rte_kvargs_free(kvlist);
2050 	return ret;
2051 }
2052 
2053 static uint8_t
2054 rx_offload_enabled(struct virtio_hw *hw)
2055 {
2056 	return virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2057 		virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2058 		virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2059 }
2060 
2061 static uint8_t
2062 tx_offload_enabled(struct virtio_hw *hw)
2063 {
2064 	return virtio_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2065 		virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2066 		virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2067 }
2068 
2069 /*
2070  * Configure virtio device
2071  * It returns 0 on success.
2072  */
2073 static int
2074 virtio_dev_configure(struct rte_eth_dev *dev)
2075 {
2076 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2077 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2078 	struct virtio_hw *hw = dev->data->dev_private;
2079 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2080 		hw->vtnet_hdr_size;
2081 	uint64_t rx_offloads = rxmode->offloads;
2082 	uint64_t tx_offloads = txmode->offloads;
2083 	uint64_t req_features;
2084 	int ret;
2085 
2086 	PMD_INIT_LOG(DEBUG, "configure");
2087 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2088 
2089 	if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2090 		PMD_DRV_LOG(ERR,
2091 			"Unsupported Rx multi queue mode %d",
2092 			rxmode->mq_mode);
2093 		return -EINVAL;
2094 	}
2095 
2096 	if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2097 		PMD_DRV_LOG(ERR,
2098 			"Unsupported Tx multi queue mode %d",
2099 			txmode->mq_mode);
2100 		return -EINVAL;
2101 	}
2102 
2103 	if (dev->data->dev_conf.intr_conf.rxq) {
2104 		ret = virtio_init_device(dev, hw->req_guest_features);
2105 		if (ret < 0)
2106 			return ret;
2107 	}
2108 
2109 	if (rxmode->mtu > hw->max_mtu)
2110 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2111 
2112 	hw->max_rx_pkt_len = ether_hdr_len + rxmode->mtu;
2113 
2114 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2115 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2116 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2117 
2118 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2119 		req_features |=
2120 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2121 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2122 
2123 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2124 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2125 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2126 
2127 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2128 		req_features |=
2129 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2130 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2131 
2132 	/* if request features changed, reinit the device */
2133 	if (req_features != hw->req_guest_features) {
2134 		ret = virtio_init_device(dev, req_features);
2135 		if (ret < 0)
2136 			return ret;
2137 	}
2138 
2139 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2140 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2141 		!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2142 		PMD_DRV_LOG(ERR,
2143 			"rx checksum not available on this host");
2144 		return -ENOTSUP;
2145 	}
2146 
2147 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2148 		(!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2149 		 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2150 		PMD_DRV_LOG(ERR,
2151 			"Large Receive Offload not available on this host");
2152 		return -ENOTSUP;
2153 	}
2154 
2155 	/* start control queue */
2156 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2157 		virtio_dev_cq_start(dev);
2158 
2159 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2160 		hw->vlan_strip = 1;
2161 
2162 	hw->rx_ol_scatter = (rx_offloads & DEV_RX_OFFLOAD_SCATTER);
2163 
2164 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2165 			!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2166 		PMD_DRV_LOG(ERR,
2167 			    "vlan filtering not available on this host");
2168 		return -ENOTSUP;
2169 	}
2170 
2171 	hw->has_tx_offload = tx_offload_enabled(hw);
2172 	hw->has_rx_offload = rx_offload_enabled(hw);
2173 
2174 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2175 		/* Enable vector (0) for Link State Intrerrupt */
2176 		if (VIRTIO_OPS(hw)->set_config_irq(hw, 0) ==
2177 				VIRTIO_MSI_NO_VECTOR) {
2178 			PMD_DRV_LOG(ERR, "failed to set config vector");
2179 			return -EBUSY;
2180 		}
2181 
2182 	if (virtio_with_packed_queue(hw)) {
2183 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2184 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2185 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2186 		     !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2187 		     !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2188 		     rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2189 			PMD_DRV_LOG(INFO,
2190 				"disabled packed ring vectorized path for requirements not met");
2191 			hw->use_vec_rx = 0;
2192 			hw->use_vec_tx = 0;
2193 		}
2194 #elif defined(RTE_ARCH_ARM)
2195 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2196 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2197 		     !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2198 		     !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2199 		     rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2200 			PMD_DRV_LOG(INFO,
2201 				"disabled packed ring vectorized path for requirements not met");
2202 			hw->use_vec_rx = 0;
2203 			hw->use_vec_tx = 0;
2204 		}
2205 #else
2206 		hw->use_vec_rx = 0;
2207 		hw->use_vec_tx = 0;
2208 #endif
2209 
2210 		if (hw->use_vec_rx) {
2211 			if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2212 				PMD_DRV_LOG(INFO,
2213 					"disabled packed ring vectorized rx for mrg_rxbuf enabled");
2214 				hw->use_vec_rx = 0;
2215 			}
2216 
2217 			if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2218 				PMD_DRV_LOG(INFO,
2219 					"disabled packed ring vectorized rx for TCP_LRO enabled");
2220 				hw->use_vec_rx = 0;
2221 			}
2222 		}
2223 	} else {
2224 		if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2225 			hw->use_inorder_tx = 1;
2226 			hw->use_inorder_rx = 1;
2227 			hw->use_vec_rx = 0;
2228 		}
2229 
2230 		if (hw->use_vec_rx) {
2231 #if defined RTE_ARCH_ARM
2232 			if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2233 				PMD_DRV_LOG(INFO,
2234 					"disabled split ring vectorized path for requirement not met");
2235 				hw->use_vec_rx = 0;
2236 			}
2237 #endif
2238 			if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2239 				PMD_DRV_LOG(INFO,
2240 					"disabled split ring vectorized rx for mrg_rxbuf enabled");
2241 				hw->use_vec_rx = 0;
2242 			}
2243 
2244 			if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2245 					   DEV_RX_OFFLOAD_TCP_CKSUM |
2246 					   DEV_RX_OFFLOAD_TCP_LRO |
2247 					   DEV_RX_OFFLOAD_VLAN_STRIP)) {
2248 				PMD_DRV_LOG(INFO,
2249 					"disabled split ring vectorized rx for offloading enabled");
2250 				hw->use_vec_rx = 0;
2251 			}
2252 
2253 			if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2254 				PMD_DRV_LOG(INFO,
2255 					"disabled split ring vectorized rx, max SIMD bitwidth too low");
2256 				hw->use_vec_rx = 0;
2257 			}
2258 		}
2259 	}
2260 
2261 	return 0;
2262 }
2263 
2264 
2265 static int
2266 virtio_dev_start(struct rte_eth_dev *dev)
2267 {
2268 	uint16_t nb_queues, i;
2269 	struct virtqueue *vq;
2270 	struct virtio_hw *hw = dev->data->dev_private;
2271 	int ret;
2272 
2273 	/* Finish the initialization of the queues */
2274 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2275 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2276 		if (ret < 0)
2277 			return ret;
2278 	}
2279 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2280 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2281 		if (ret < 0)
2282 			return ret;
2283 	}
2284 
2285 	/* check if lsc interrupt feature is enabled */
2286 	if (dev->data->dev_conf.intr_conf.lsc) {
2287 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2288 			PMD_DRV_LOG(ERR, "link status not supported by host");
2289 			return -ENOTSUP;
2290 		}
2291 	}
2292 
2293 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2294 	 * in device configure, but it could be unmapped  when device is
2295 	 * stopped.
2296 	 */
2297 	if (dev->data->dev_conf.intr_conf.lsc ||
2298 	    dev->data->dev_conf.intr_conf.rxq) {
2299 		virtio_intr_disable(dev);
2300 
2301 		/* Setup interrupt callback  */
2302 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2303 			rte_intr_callback_register(dev->intr_handle,
2304 						   virtio_interrupt_handler,
2305 						   dev);
2306 
2307 		if (virtio_intr_enable(dev) < 0) {
2308 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2309 			return -EIO;
2310 		}
2311 	}
2312 
2313 	/*Notify the backend
2314 	 *Otherwise the tap backend might already stop its queue due to fullness.
2315 	 *vhost backend will have no chance to be waked up
2316 	 */
2317 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2318 	if (hw->max_queue_pairs > 1) {
2319 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2320 			return -EINVAL;
2321 	}
2322 
2323 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2324 
2325 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2326 		vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2327 		/* Flush the old packets */
2328 		virtqueue_rxvq_flush(vq);
2329 		virtqueue_notify(vq);
2330 	}
2331 
2332 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2333 		vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2334 		virtqueue_notify(vq);
2335 	}
2336 
2337 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2338 
2339 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2340 		vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2341 		VIRTQUEUE_DUMP(vq);
2342 	}
2343 
2344 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2345 		vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2346 		VIRTQUEUE_DUMP(vq);
2347 	}
2348 
2349 	set_rxtx_funcs(dev);
2350 	hw->started = 1;
2351 
2352 	/* Initialize Link state */
2353 	virtio_dev_link_update(dev, 0);
2354 
2355 	return 0;
2356 }
2357 
2358 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2359 {
2360 	struct virtio_hw *hw = dev->data->dev_private;
2361 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2362 	const char *type __rte_unused;
2363 	unsigned int i, mbuf_num = 0;
2364 	struct virtqueue *vq;
2365 	struct rte_mbuf *buf;
2366 	int queue_type;
2367 
2368 	if (hw->vqs == NULL)
2369 		return;
2370 
2371 	for (i = 0; i < nr_vq; i++) {
2372 		vq = hw->vqs[i];
2373 		if (!vq)
2374 			continue;
2375 
2376 		queue_type = virtio_get_queue_type(hw, i);
2377 		if (queue_type == VTNET_RQ)
2378 			type = "rxq";
2379 		else if (queue_type == VTNET_TQ)
2380 			type = "txq";
2381 		else
2382 			continue;
2383 
2384 		PMD_INIT_LOG(DEBUG,
2385 			"Before freeing %s[%d] used and unused buf",
2386 			type, i);
2387 		VIRTQUEUE_DUMP(vq);
2388 
2389 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2390 			rte_pktmbuf_free(buf);
2391 			mbuf_num++;
2392 		}
2393 
2394 		PMD_INIT_LOG(DEBUG,
2395 			"After freeing %s[%d] used and unused buf",
2396 			type, i);
2397 		VIRTQUEUE_DUMP(vq);
2398 	}
2399 
2400 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2401 }
2402 
2403 static void
2404 virtio_tx_completed_cleanup(struct rte_eth_dev *dev)
2405 {
2406 	struct virtio_hw *hw = dev->data->dev_private;
2407 	struct virtqueue *vq;
2408 	int qidx;
2409 	void (*xmit_cleanup)(struct virtqueue *vq, uint16_t nb_used);
2410 
2411 	if (virtio_with_packed_queue(hw)) {
2412 		if (hw->use_vec_tx)
2413 			xmit_cleanup = &virtio_xmit_cleanup_inorder_packed;
2414 		else if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER))
2415 			xmit_cleanup = &virtio_xmit_cleanup_inorder_packed;
2416 		else
2417 			xmit_cleanup = &virtio_xmit_cleanup_normal_packed;
2418 	} else {
2419 		if (hw->use_inorder_tx)
2420 			xmit_cleanup = &virtio_xmit_cleanup_inorder;
2421 		else
2422 			xmit_cleanup = &virtio_xmit_cleanup;
2423 	}
2424 
2425 	for (qidx = 0; qidx < hw->max_queue_pairs; qidx++) {
2426 		vq = hw->vqs[2 * qidx + VTNET_SQ_TQ_QUEUE_IDX];
2427 		if (vq != NULL)
2428 			xmit_cleanup(vq, virtqueue_nused(vq));
2429 	}
2430 }
2431 
2432 /*
2433  * Stop device: disable interrupt and mark link down
2434  */
2435 int
2436 virtio_dev_stop(struct rte_eth_dev *dev)
2437 {
2438 	struct virtio_hw *hw = dev->data->dev_private;
2439 	struct rte_eth_link link;
2440 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2441 
2442 	PMD_INIT_LOG(DEBUG, "stop");
2443 	dev->data->dev_started = 0;
2444 
2445 	rte_spinlock_lock(&hw->state_lock);
2446 	if (!hw->started)
2447 		goto out_unlock;
2448 	hw->started = 0;
2449 
2450 	virtio_tx_completed_cleanup(dev);
2451 
2452 	if (intr_conf->lsc || intr_conf->rxq) {
2453 		virtio_intr_disable(dev);
2454 
2455 		/* Reset interrupt callback  */
2456 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2457 			rte_intr_callback_unregister(dev->intr_handle,
2458 						     virtio_interrupt_handler,
2459 						     dev);
2460 		}
2461 	}
2462 
2463 	memset(&link, 0, sizeof(link));
2464 	rte_eth_linkstatus_set(dev, &link);
2465 out_unlock:
2466 	rte_spinlock_unlock(&hw->state_lock);
2467 
2468 	return 0;
2469 }
2470 
2471 static int
2472 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2473 {
2474 	struct rte_eth_link link;
2475 	uint16_t status;
2476 	struct virtio_hw *hw = dev->data->dev_private;
2477 
2478 	memset(&link, 0, sizeof(link));
2479 	link.link_duplex = hw->duplex;
2480 	link.link_speed  = hw->speed;
2481 	link.link_autoneg = ETH_LINK_AUTONEG;
2482 
2483 	if (!hw->started) {
2484 		link.link_status = ETH_LINK_DOWN;
2485 		link.link_speed = ETH_SPEED_NUM_NONE;
2486 	} else if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2487 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2488 		virtio_read_dev_config(hw,
2489 				offsetof(struct virtio_net_config, status),
2490 				&status, sizeof(status));
2491 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2492 			link.link_status = ETH_LINK_DOWN;
2493 			link.link_speed = ETH_SPEED_NUM_NONE;
2494 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2495 				     dev->data->port_id);
2496 		} else {
2497 			link.link_status = ETH_LINK_UP;
2498 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2499 				     dev->data->port_id);
2500 		}
2501 	} else {
2502 		link.link_status = ETH_LINK_UP;
2503 	}
2504 
2505 	return rte_eth_linkstatus_set(dev, &link);
2506 }
2507 
2508 static int
2509 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2510 {
2511 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2512 	struct virtio_hw *hw = dev->data->dev_private;
2513 	uint64_t offloads = rxmode->offloads;
2514 
2515 	if (mask & ETH_VLAN_FILTER_MASK) {
2516 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2517 				!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2518 
2519 			PMD_DRV_LOG(NOTICE,
2520 				"vlan filtering not available on this host");
2521 
2522 			return -ENOTSUP;
2523 		}
2524 	}
2525 
2526 	if (mask & ETH_VLAN_STRIP_MASK)
2527 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2528 
2529 	return 0;
2530 }
2531 
2532 static int
2533 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2534 {
2535 	uint64_t tso_mask, host_features;
2536 	struct virtio_hw *hw = dev->data->dev_private;
2537 	dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2538 
2539 	dev_info->max_rx_queues =
2540 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2541 	dev_info->max_tx_queues =
2542 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2543 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2544 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2545 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2546 	dev_info->max_mtu = hw->max_mtu;
2547 
2548 	host_features = VIRTIO_OPS(hw)->get_features(hw);
2549 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2550 	if (host_features & (1ULL << VIRTIO_NET_F_MRG_RXBUF))
2551 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_SCATTER;
2552 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2553 		dev_info->rx_offload_capa |=
2554 			DEV_RX_OFFLOAD_TCP_CKSUM |
2555 			DEV_RX_OFFLOAD_UDP_CKSUM;
2556 	}
2557 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2558 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2559 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2560 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2561 	if ((host_features & tso_mask) == tso_mask)
2562 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2563 
2564 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2565 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2566 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2567 		dev_info->tx_offload_capa |=
2568 			DEV_TX_OFFLOAD_UDP_CKSUM |
2569 			DEV_TX_OFFLOAD_TCP_CKSUM;
2570 	}
2571 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2572 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2573 	if ((host_features & tso_mask) == tso_mask)
2574 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2575 
2576 	if (host_features & (1ULL << VIRTIO_F_RING_PACKED)) {
2577 		/*
2578 		 * According to 2.7 Packed Virtqueues,
2579 		 * 2.7.10.1 Structure Size and Alignment:
2580 		 * The Queue Size value does not have to be a power of 2.
2581 		 */
2582 		dev_info->rx_desc_lim.nb_max = UINT16_MAX;
2583 		dev_info->tx_desc_lim.nb_max = UINT16_MAX;
2584 	} else {
2585 		/*
2586 		 * According to 2.6 Split Virtqueues:
2587 		 * Queue Size value is always a power of 2. The maximum Queue
2588 		 * Size value is 32768.
2589 		 */
2590 		dev_info->rx_desc_lim.nb_max = 32768;
2591 		dev_info->tx_desc_lim.nb_max = 32768;
2592 	}
2593 	/*
2594 	 * Actual minimum is not the same for virtqueues of different kinds,
2595 	 * but to avoid tangling the code with separate branches, rely on
2596 	 * default thresholds since desc number must be at least of their size.
2597 	 */
2598 	dev_info->rx_desc_lim.nb_min = RTE_MAX(DEFAULT_RX_FREE_THRESH,
2599 					       RTE_VIRTIO_VPMD_RX_REARM_THRESH);
2600 	dev_info->tx_desc_lim.nb_min = DEFAULT_TX_FREE_THRESH;
2601 	dev_info->rx_desc_lim.nb_align = 1;
2602 	dev_info->tx_desc_lim.nb_align = 1;
2603 
2604 	return 0;
2605 }
2606 
2607 /*
2608  * It enables testpmd to collect per queue stats.
2609  */
2610 static int
2611 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2612 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2613 __rte_unused uint8_t is_rx)
2614 {
2615 	return 0;
2616 }
2617 
2618 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_init, init, NOTICE);
2619 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_driver, driver, NOTICE);
2620