xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 03ab51eafda992874a48c392ca66ffb577fe2b71)
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 	dev->data->dev_conf.rxmode.max_rx_pkt_len = hw->max_rx_pkt_len;
928 
929 	return 0;
930 }
931 
932 static int
933 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
934 {
935 	struct virtio_hw *hw = dev->data->dev_private;
936 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
937 	struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
938 
939 	virtqueue_enable_intr(vq);
940 	virtio_mb(hw->weak_barriers);
941 	return 0;
942 }
943 
944 static int
945 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
946 {
947 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
948 	struct virtqueue *vq = virtnet_rxq_to_vq(rxvq);
949 
950 	virtqueue_disable_intr(vq);
951 	return 0;
952 }
953 
954 /*
955  * dev_ops for virtio, bare necessities for basic operation
956  */
957 static const struct eth_dev_ops virtio_eth_dev_ops = {
958 	.dev_configure           = virtio_dev_configure,
959 	.dev_start               = virtio_dev_start,
960 	.dev_stop                = virtio_dev_stop,
961 	.dev_close               = virtio_dev_close,
962 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
963 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
964 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
965 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
966 	.mtu_set                 = virtio_mtu_set,
967 	.dev_infos_get           = virtio_dev_info_get,
968 	.stats_get               = virtio_dev_stats_get,
969 	.xstats_get              = virtio_dev_xstats_get,
970 	.xstats_get_names        = virtio_dev_xstats_get_names,
971 	.stats_reset             = virtio_dev_stats_reset,
972 	.xstats_reset            = virtio_dev_stats_reset,
973 	.link_update             = virtio_dev_link_update,
974 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
975 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
976 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
977 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
978 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
979 	/* collect stats per queue */
980 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
981 	.vlan_filter_set         = virtio_vlan_filter_set,
982 	.mac_addr_add            = virtio_mac_addr_add,
983 	.mac_addr_remove         = virtio_mac_addr_remove,
984 	.mac_addr_set            = virtio_mac_addr_set,
985 };
986 
987 /*
988  * dev_ops for virtio-user in secondary processes, as we just have
989  * some limited supports currently.
990  */
991 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
992 	.dev_infos_get           = virtio_dev_info_get,
993 	.stats_get               = virtio_dev_stats_get,
994 	.xstats_get              = virtio_dev_xstats_get,
995 	.xstats_get_names        = virtio_dev_xstats_get_names,
996 	.stats_reset             = virtio_dev_stats_reset,
997 	.xstats_reset            = virtio_dev_stats_reset,
998 	/* collect stats per queue */
999 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
1000 };
1001 
1002 static void
1003 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1004 {
1005 	unsigned i;
1006 
1007 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1008 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
1009 		if (txvq == NULL)
1010 			continue;
1011 
1012 		stats->opackets += txvq->stats.packets;
1013 		stats->obytes += txvq->stats.bytes;
1014 
1015 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1016 			stats->q_opackets[i] = txvq->stats.packets;
1017 			stats->q_obytes[i] = txvq->stats.bytes;
1018 		}
1019 	}
1020 
1021 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1022 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1023 		if (rxvq == NULL)
1024 			continue;
1025 
1026 		stats->ipackets += rxvq->stats.packets;
1027 		stats->ibytes += rxvq->stats.bytes;
1028 		stats->ierrors += rxvq->stats.errors;
1029 
1030 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
1031 			stats->q_ipackets[i] = rxvq->stats.packets;
1032 			stats->q_ibytes[i] = rxvq->stats.bytes;
1033 		}
1034 	}
1035 
1036 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
1037 }
1038 
1039 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
1040 				       struct rte_eth_xstat_name *xstats_names,
1041 				       __rte_unused unsigned limit)
1042 {
1043 	unsigned i;
1044 	unsigned count = 0;
1045 	unsigned t;
1046 
1047 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1048 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1049 
1050 	if (xstats_names != NULL) {
1051 		/* Note: limit checked in rte_eth_xstats_names() */
1052 
1053 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
1054 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1055 			if (rxvq == NULL)
1056 				continue;
1057 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1058 				snprintf(xstats_names[count].name,
1059 					sizeof(xstats_names[count].name),
1060 					"rx_q%u_%s", i,
1061 					rte_virtio_rxq_stat_strings[t].name);
1062 				count++;
1063 			}
1064 		}
1065 
1066 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
1067 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
1068 			if (txvq == NULL)
1069 				continue;
1070 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1071 				snprintf(xstats_names[count].name,
1072 					sizeof(xstats_names[count].name),
1073 					"tx_q%u_%s", i,
1074 					rte_virtio_txq_stat_strings[t].name);
1075 				count++;
1076 			}
1077 		}
1078 		return count;
1079 	}
1080 	return nstats;
1081 }
1082 
1083 static int
1084 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1085 		      unsigned n)
1086 {
1087 	unsigned i;
1088 	unsigned count = 0;
1089 
1090 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1091 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1092 
1093 	if (n < nstats)
1094 		return nstats;
1095 
1096 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1097 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1098 
1099 		if (rxvq == NULL)
1100 			continue;
1101 
1102 		unsigned t;
1103 
1104 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1105 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1106 				rte_virtio_rxq_stat_strings[t].offset);
1107 			xstats[count].id = count;
1108 			count++;
1109 		}
1110 	}
1111 
1112 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1113 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1114 
1115 		if (txvq == NULL)
1116 			continue;
1117 
1118 		unsigned t;
1119 
1120 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1121 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1122 				rte_virtio_txq_stat_strings[t].offset);
1123 			xstats[count].id = count;
1124 			count++;
1125 		}
1126 	}
1127 
1128 	return count;
1129 }
1130 
1131 static int
1132 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1133 {
1134 	virtio_update_stats(dev, stats);
1135 
1136 	return 0;
1137 }
1138 
1139 static int
1140 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1141 {
1142 	unsigned int i;
1143 
1144 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1145 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1146 		if (txvq == NULL)
1147 			continue;
1148 
1149 		txvq->stats.packets = 0;
1150 		txvq->stats.bytes = 0;
1151 		txvq->stats.multicast = 0;
1152 		txvq->stats.broadcast = 0;
1153 		memset(txvq->stats.size_bins, 0,
1154 		       sizeof(txvq->stats.size_bins[0]) * 8);
1155 	}
1156 
1157 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1158 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1159 		if (rxvq == NULL)
1160 			continue;
1161 
1162 		rxvq->stats.packets = 0;
1163 		rxvq->stats.bytes = 0;
1164 		rxvq->stats.errors = 0;
1165 		rxvq->stats.multicast = 0;
1166 		rxvq->stats.broadcast = 0;
1167 		memset(rxvq->stats.size_bins, 0,
1168 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1169 	}
1170 
1171 	return 0;
1172 }
1173 
1174 static void
1175 virtio_set_hwaddr(struct virtio_hw *hw)
1176 {
1177 	virtio_write_dev_config(hw,
1178 			offsetof(struct virtio_net_config, mac),
1179 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1180 }
1181 
1182 static void
1183 virtio_get_hwaddr(struct virtio_hw *hw)
1184 {
1185 	if (virtio_with_feature(hw, VIRTIO_NET_F_MAC)) {
1186 		virtio_read_dev_config(hw,
1187 			offsetof(struct virtio_net_config, mac),
1188 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1189 	} else {
1190 		rte_eth_random_addr(&hw->mac_addr[0]);
1191 		virtio_set_hwaddr(hw);
1192 	}
1193 }
1194 
1195 static int
1196 virtio_mac_table_set(struct virtio_hw *hw,
1197 		     const struct virtio_net_ctrl_mac *uc,
1198 		     const struct virtio_net_ctrl_mac *mc)
1199 {
1200 	struct virtio_pmd_ctrl ctrl;
1201 	int err, len[2];
1202 
1203 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1204 		PMD_DRV_LOG(INFO, "host does not support mac table");
1205 		return -1;
1206 	}
1207 
1208 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1209 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1210 
1211 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1212 	memcpy(ctrl.data, uc, len[0]);
1213 
1214 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1215 	memcpy(ctrl.data + len[0], mc, len[1]);
1216 
1217 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1218 	if (err != 0)
1219 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1220 	return err;
1221 }
1222 
1223 static int
1224 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1225 		    uint32_t index, uint32_t vmdq __rte_unused)
1226 {
1227 	struct virtio_hw *hw = dev->data->dev_private;
1228 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1229 	unsigned int i;
1230 	struct virtio_net_ctrl_mac *uc, *mc;
1231 
1232 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1233 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1234 		return -EINVAL;
1235 	}
1236 
1237 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1238 		sizeof(uc->entries));
1239 	uc->entries = 0;
1240 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1241 		sizeof(mc->entries));
1242 	mc->entries = 0;
1243 
1244 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1245 		const struct rte_ether_addr *addr
1246 			= (i == index) ? mac_addr : addrs + i;
1247 		struct virtio_net_ctrl_mac *tbl
1248 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1249 
1250 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1251 	}
1252 
1253 	return virtio_mac_table_set(hw, uc, mc);
1254 }
1255 
1256 static void
1257 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1258 {
1259 	struct virtio_hw *hw = dev->data->dev_private;
1260 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1261 	struct virtio_net_ctrl_mac *uc, *mc;
1262 	unsigned int i;
1263 
1264 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1265 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1266 		return;
1267 	}
1268 
1269 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1270 		sizeof(uc->entries));
1271 	uc->entries = 0;
1272 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1273 		sizeof(mc->entries));
1274 	mc->entries = 0;
1275 
1276 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1277 		struct virtio_net_ctrl_mac *tbl;
1278 
1279 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1280 			continue;
1281 
1282 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1283 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1284 			RTE_ETHER_ADDR_LEN);
1285 	}
1286 
1287 	virtio_mac_table_set(hw, uc, mc);
1288 }
1289 
1290 static int
1291 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1292 {
1293 	struct virtio_hw *hw = dev->data->dev_private;
1294 
1295 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1296 
1297 	/* Use atomic update if available */
1298 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1299 		struct virtio_pmd_ctrl ctrl;
1300 		int len = RTE_ETHER_ADDR_LEN;
1301 
1302 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1303 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1304 
1305 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1306 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1307 	}
1308 
1309 	if (!virtio_with_feature(hw, VIRTIO_NET_F_MAC))
1310 		return -ENOTSUP;
1311 
1312 	virtio_set_hwaddr(hw);
1313 	return 0;
1314 }
1315 
1316 static int
1317 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1318 {
1319 	struct virtio_hw *hw = dev->data->dev_private;
1320 	struct virtio_pmd_ctrl ctrl;
1321 	int len;
1322 
1323 	if (!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1324 		return -ENOTSUP;
1325 
1326 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1327 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1328 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1329 	len = sizeof(vlan_id);
1330 
1331 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1332 }
1333 
1334 static int
1335 virtio_intr_unmask(struct rte_eth_dev *dev)
1336 {
1337 	struct virtio_hw *hw = dev->data->dev_private;
1338 
1339 	if (rte_intr_ack(dev->intr_handle) < 0)
1340 		return -1;
1341 
1342 	if (VIRTIO_OPS(hw)->intr_detect)
1343 		VIRTIO_OPS(hw)->intr_detect(hw);
1344 
1345 	return 0;
1346 }
1347 
1348 static int
1349 virtio_intr_enable(struct rte_eth_dev *dev)
1350 {
1351 	struct virtio_hw *hw = dev->data->dev_private;
1352 
1353 	if (rte_intr_enable(dev->intr_handle) < 0)
1354 		return -1;
1355 
1356 	if (VIRTIO_OPS(hw)->intr_detect)
1357 		VIRTIO_OPS(hw)->intr_detect(hw);
1358 
1359 	return 0;
1360 }
1361 
1362 static int
1363 virtio_intr_disable(struct rte_eth_dev *dev)
1364 {
1365 	struct virtio_hw *hw = dev->data->dev_private;
1366 
1367 	if (rte_intr_disable(dev->intr_handle) < 0)
1368 		return -1;
1369 
1370 	if (VIRTIO_OPS(hw)->intr_detect)
1371 		VIRTIO_OPS(hw)->intr_detect(hw);
1372 
1373 	return 0;
1374 }
1375 
1376 static int
1377 virtio_ethdev_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1378 {
1379 	uint64_t host_features;
1380 
1381 	/* Prepare guest_features: feature that driver wants to support */
1382 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1383 		req_features);
1384 
1385 	/* Read device(host) feature bits */
1386 	host_features = VIRTIO_OPS(hw)->get_features(hw);
1387 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1388 		host_features);
1389 
1390 	/* If supported, ensure MTU value is valid before acknowledging it. */
1391 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1392 		struct virtio_net_config config;
1393 
1394 		virtio_read_dev_config(hw,
1395 			offsetof(struct virtio_net_config, mtu),
1396 			&config.mtu, sizeof(config.mtu));
1397 
1398 		if (config.mtu < RTE_ETHER_MIN_MTU)
1399 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1400 	}
1401 
1402 	/*
1403 	 * Negotiate features: Subset of device feature bits are written back
1404 	 * guest feature bits.
1405 	 */
1406 	hw->guest_features = req_features;
1407 	hw->guest_features = virtio_negotiate_features(hw, host_features);
1408 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1409 		hw->guest_features);
1410 
1411 	if (VIRTIO_OPS(hw)->features_ok(hw) < 0)
1412 		return -1;
1413 
1414 	if (virtio_with_feature(hw, VIRTIO_F_VERSION_1)) {
1415 		virtio_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1416 
1417 		if (!(virtio_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1418 			PMD_INIT_LOG(ERR, "Failed to set FEATURES_OK status!");
1419 			return -1;
1420 		}
1421 	}
1422 
1423 	hw->req_guest_features = req_features;
1424 
1425 	return 0;
1426 }
1427 
1428 int
1429 virtio_dev_pause(struct rte_eth_dev *dev)
1430 {
1431 	struct virtio_hw *hw = dev->data->dev_private;
1432 
1433 	rte_spinlock_lock(&hw->state_lock);
1434 
1435 	if (hw->started == 0) {
1436 		/* Device is just stopped. */
1437 		rte_spinlock_unlock(&hw->state_lock);
1438 		return -1;
1439 	}
1440 	hw->started = 0;
1441 	/*
1442 	 * Prevent the worker threads from touching queues to avoid contention,
1443 	 * 1 ms should be enough for the ongoing Tx function to finish.
1444 	 */
1445 	rte_delay_ms(1);
1446 	return 0;
1447 }
1448 
1449 /*
1450  * Recover hw state to let the worker threads continue.
1451  */
1452 void
1453 virtio_dev_resume(struct rte_eth_dev *dev)
1454 {
1455 	struct virtio_hw *hw = dev->data->dev_private;
1456 
1457 	hw->started = 1;
1458 	rte_spinlock_unlock(&hw->state_lock);
1459 }
1460 
1461 /*
1462  * Should be called only after device is paused.
1463  */
1464 int
1465 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1466 		int nb_pkts)
1467 {
1468 	struct virtio_hw *hw = dev->data->dev_private;
1469 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1470 	int ret;
1471 
1472 	hw->inject_pkts = tx_pkts;
1473 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1474 	hw->inject_pkts = NULL;
1475 
1476 	return ret;
1477 }
1478 
1479 static void
1480 virtio_notify_peers(struct rte_eth_dev *dev)
1481 {
1482 	struct virtio_hw *hw = dev->data->dev_private;
1483 	struct virtnet_rx *rxvq;
1484 	struct rte_mbuf *rarp_mbuf;
1485 
1486 	if (!dev->data->rx_queues)
1487 		return;
1488 
1489 	rxvq = dev->data->rx_queues[0];
1490 	if (!rxvq)
1491 		return;
1492 
1493 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1494 			(struct rte_ether_addr *)hw->mac_addr);
1495 	if (rarp_mbuf == NULL) {
1496 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1497 		return;
1498 	}
1499 
1500 	/* If virtio port just stopped, no need to send RARP */
1501 	if (virtio_dev_pause(dev) < 0) {
1502 		rte_pktmbuf_free(rarp_mbuf);
1503 		return;
1504 	}
1505 
1506 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1507 	virtio_dev_resume(dev);
1508 }
1509 
1510 static void
1511 virtio_ack_link_announce(struct rte_eth_dev *dev)
1512 {
1513 	struct virtio_hw *hw = dev->data->dev_private;
1514 	struct virtio_pmd_ctrl ctrl;
1515 
1516 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1517 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1518 
1519 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1520 }
1521 
1522 /*
1523  * Process virtio config changed interrupt. Call the callback
1524  * if link state changed, generate gratuitous RARP packet if
1525  * the status indicates an ANNOUNCE.
1526  */
1527 void
1528 virtio_interrupt_handler(void *param)
1529 {
1530 	struct rte_eth_dev *dev = param;
1531 	struct virtio_hw *hw = dev->data->dev_private;
1532 	uint8_t isr;
1533 	uint16_t status;
1534 
1535 	/* Read interrupt status which clears interrupt */
1536 	isr = virtio_get_isr(hw);
1537 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1538 
1539 	if (virtio_intr_unmask(dev) < 0)
1540 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1541 
1542 	if (isr & VIRTIO_ISR_CONFIG) {
1543 		if (virtio_dev_link_update(dev, 0) == 0)
1544 			rte_eth_dev_callback_process(dev,
1545 						     RTE_ETH_EVENT_INTR_LSC,
1546 						     NULL);
1547 
1548 		if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1549 			virtio_read_dev_config(hw,
1550 				offsetof(struct virtio_net_config, status),
1551 				&status, sizeof(status));
1552 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1553 				virtio_notify_peers(dev);
1554 				if (hw->cvq)
1555 					virtio_ack_link_announce(dev);
1556 			}
1557 		}
1558 	}
1559 }
1560 
1561 /* set rx and tx handlers according to what is supported */
1562 static void
1563 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1564 {
1565 	struct virtio_hw *hw = eth_dev->data->dev_private;
1566 
1567 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1568 	if (virtio_with_packed_queue(hw)) {
1569 		PMD_INIT_LOG(INFO,
1570 			"virtio: using packed ring %s Tx path on port %u",
1571 			hw->use_vec_tx ? "vectorized" : "standard",
1572 			eth_dev->data->port_id);
1573 		if (hw->use_vec_tx)
1574 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1575 		else
1576 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1577 	} else {
1578 		if (hw->use_inorder_tx) {
1579 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1580 				eth_dev->data->port_id);
1581 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1582 		} else {
1583 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1584 				eth_dev->data->port_id);
1585 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1586 		}
1587 	}
1588 
1589 	if (virtio_with_packed_queue(hw)) {
1590 		if (hw->use_vec_rx) {
1591 			PMD_INIT_LOG(INFO,
1592 				"virtio: using packed ring vectorized Rx path on port %u",
1593 				eth_dev->data->port_id);
1594 			eth_dev->rx_pkt_burst =
1595 				&virtio_recv_pkts_packed_vec;
1596 		} else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1597 			PMD_INIT_LOG(INFO,
1598 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1599 				eth_dev->data->port_id);
1600 			eth_dev->rx_pkt_burst =
1601 				&virtio_recv_mergeable_pkts_packed;
1602 		} else {
1603 			PMD_INIT_LOG(INFO,
1604 				"virtio: using packed ring standard Rx path on port %u",
1605 				eth_dev->data->port_id);
1606 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1607 		}
1608 	} else {
1609 		if (hw->use_vec_rx) {
1610 			PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1611 				eth_dev->data->port_id);
1612 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1613 		} else if (hw->use_inorder_rx) {
1614 			PMD_INIT_LOG(INFO,
1615 				"virtio: using inorder Rx path on port %u",
1616 				eth_dev->data->port_id);
1617 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1618 		} else if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1619 			PMD_INIT_LOG(INFO,
1620 				"virtio: using mergeable buffer Rx path on port %u",
1621 				eth_dev->data->port_id);
1622 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1623 		} else {
1624 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1625 				eth_dev->data->port_id);
1626 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1627 		}
1628 	}
1629 
1630 }
1631 
1632 /* Only support 1:1 queue/interrupt mapping so far.
1633  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1634  * interrupt vectors (<N+1).
1635  */
1636 static int
1637 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1638 {
1639 	uint32_t i;
1640 	struct virtio_hw *hw = dev->data->dev_private;
1641 
1642 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1643 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1644 		dev->intr_handle->intr_vec[i] = i + 1;
1645 		if (VIRTIO_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1646 						 VIRTIO_MSI_NO_VECTOR) {
1647 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1648 			return -EBUSY;
1649 		}
1650 	}
1651 
1652 	return 0;
1653 }
1654 
1655 static void
1656 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1657 {
1658 	uint32_t i;
1659 	struct virtio_hw *hw = dev->data->dev_private;
1660 
1661 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1662 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1663 		VIRTIO_OPS(hw)->set_queue_irq(hw,
1664 					     hw->vqs[i * VTNET_CQ],
1665 					     VIRTIO_MSI_NO_VECTOR);
1666 }
1667 
1668 static int
1669 virtio_configure_intr(struct rte_eth_dev *dev)
1670 {
1671 	struct virtio_hw *hw = dev->data->dev_private;
1672 
1673 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1674 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1675 		return -ENOTSUP;
1676 	}
1677 
1678 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1679 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1680 		return -1;
1681 	}
1682 
1683 	if (!dev->intr_handle->intr_vec) {
1684 		dev->intr_handle->intr_vec =
1685 			rte_zmalloc("intr_vec",
1686 				    hw->max_queue_pairs * sizeof(int), 0);
1687 		if (!dev->intr_handle->intr_vec) {
1688 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1689 				     hw->max_queue_pairs);
1690 			return -ENOMEM;
1691 		}
1692 	}
1693 
1694 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1695 		/* Re-register callback to update max_intr */
1696 		rte_intr_callback_unregister(dev->intr_handle,
1697 					     virtio_interrupt_handler,
1698 					     dev);
1699 		rte_intr_callback_register(dev->intr_handle,
1700 					   virtio_interrupt_handler,
1701 					   dev);
1702 	}
1703 
1704 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1705 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1706 	 * And for legacy devices, this should be done before queue/vec binding
1707 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1708 	 * (22) will be ignored.
1709 	 */
1710 	if (virtio_intr_enable(dev) < 0) {
1711 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1712 		return -1;
1713 	}
1714 
1715 	if (virtio_queues_bind_intr(dev) < 0) {
1716 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1717 		return -1;
1718 	}
1719 
1720 	return 0;
1721 }
1722 #define DUPLEX_UNKNOWN   0xff
1723 /* reset device and renegotiate features if needed */
1724 static int
1725 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1726 {
1727 	struct virtio_hw *hw = eth_dev->data->dev_private;
1728 	struct virtio_net_config *config;
1729 	struct virtio_net_config local_config;
1730 	int ret;
1731 
1732 	/* Reset the device although not necessary at startup */
1733 	virtio_reset(hw);
1734 
1735 	if (hw->vqs) {
1736 		virtio_dev_free_mbufs(eth_dev);
1737 		virtio_free_queues(hw);
1738 	}
1739 
1740 	/* Tell the host we've noticed this device. */
1741 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1742 
1743 	/* Tell the host we've known how to drive the device. */
1744 	virtio_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1745 	if (virtio_ethdev_negotiate_features(hw, req_features) < 0)
1746 		return -1;
1747 
1748 	hw->weak_barriers = !virtio_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1749 
1750 	/* If host does not support both status and MSI-X then disable LSC */
1751 	if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->intr_lsc)
1752 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1753 	else
1754 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1755 
1756 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1757 
1758 	/* Setting up rx_header size for the device */
1759 	if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1760 	    virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
1761 	    virtio_with_packed_queue(hw))
1762 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1763 	else
1764 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1765 
1766 	/* Copy the permanent MAC address to: virtio_hw */
1767 	virtio_get_hwaddr(hw);
1768 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1769 			&eth_dev->data->mac_addrs[0]);
1770 	PMD_INIT_LOG(DEBUG,
1771 		     "PORT MAC: " RTE_ETHER_ADDR_PRT_FMT,
1772 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1773 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1774 
1775 	if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1776 		if (virtio_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1777 			config = &local_config;
1778 			virtio_read_dev_config(hw,
1779 				offsetof(struct virtio_net_config, speed),
1780 				&config->speed, sizeof(config->speed));
1781 			virtio_read_dev_config(hw,
1782 				offsetof(struct virtio_net_config, duplex),
1783 				&config->duplex, sizeof(config->duplex));
1784 			hw->speed = config->speed;
1785 			hw->duplex = config->duplex;
1786 		}
1787 	}
1788 	if (hw->duplex == DUPLEX_UNKNOWN)
1789 		hw->duplex = ETH_LINK_FULL_DUPLEX;
1790 	PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1791 		hw->speed, hw->duplex);
1792 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1793 		config = &local_config;
1794 
1795 		virtio_read_dev_config(hw,
1796 			offsetof(struct virtio_net_config, mac),
1797 			&config->mac, sizeof(config->mac));
1798 
1799 		if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1800 			virtio_read_dev_config(hw,
1801 				offsetof(struct virtio_net_config, status),
1802 				&config->status, sizeof(config->status));
1803 		} else {
1804 			PMD_INIT_LOG(DEBUG,
1805 				     "VIRTIO_NET_F_STATUS is not supported");
1806 			config->status = 0;
1807 		}
1808 
1809 		if (virtio_with_feature(hw, VIRTIO_NET_F_MQ)) {
1810 			virtio_read_dev_config(hw,
1811 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1812 				&config->max_virtqueue_pairs,
1813 				sizeof(config->max_virtqueue_pairs));
1814 		} else {
1815 			PMD_INIT_LOG(DEBUG,
1816 				     "VIRTIO_NET_F_MQ is not supported");
1817 			config->max_virtqueue_pairs = 1;
1818 		}
1819 
1820 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1821 
1822 		if (virtio_with_feature(hw, VIRTIO_NET_F_MTU)) {
1823 			virtio_read_dev_config(hw,
1824 				offsetof(struct virtio_net_config, mtu),
1825 				&config->mtu,
1826 				sizeof(config->mtu));
1827 
1828 			/*
1829 			 * MTU value has already been checked at negotiation
1830 			 * time, but check again in case it has changed since
1831 			 * then, which should not happen.
1832 			 */
1833 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1834 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1835 						config->mtu);
1836 				return -1;
1837 			}
1838 
1839 			hw->max_mtu = config->mtu;
1840 			/* Set initial MTU to maximum one supported by vhost */
1841 			eth_dev->data->mtu = config->mtu;
1842 
1843 		} else {
1844 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1845 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1846 		}
1847 
1848 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1849 				config->max_virtqueue_pairs);
1850 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1851 		PMD_INIT_LOG(DEBUG,
1852 				"PORT MAC: " RTE_ETHER_ADDR_PRT_FMT,
1853 				config->mac[0], config->mac[1],
1854 				config->mac[2], config->mac[3],
1855 				config->mac[4], config->mac[5]);
1856 	} else {
1857 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1858 		hw->max_queue_pairs = 1;
1859 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1860 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1861 	}
1862 
1863 	ret = virtio_alloc_queues(eth_dev);
1864 	if (ret < 0)
1865 		return ret;
1866 
1867 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1868 		if (virtio_configure_intr(eth_dev) < 0) {
1869 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1870 			virtio_free_queues(hw);
1871 			return -1;
1872 		}
1873 	}
1874 
1875 	virtio_reinit_complete(hw);
1876 
1877 	return 0;
1878 }
1879 
1880 /*
1881  * This function is based on probe() function in virtio_pci.c
1882  * It returns 0 on success.
1883  */
1884 int
1885 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1886 {
1887 	struct virtio_hw *hw = eth_dev->data->dev_private;
1888 	uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1889 	int vectorized = 0;
1890 	int ret;
1891 
1892 	if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1893 		PMD_INIT_LOG(ERR,
1894 			"Not sufficient headroom required = %d, avail = %d",
1895 			(int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1896 			RTE_PKTMBUF_HEADROOM);
1897 
1898 		return -1;
1899 	}
1900 
1901 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1902 	eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1903 
1904 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1905 		set_rxtx_funcs(eth_dev);
1906 		return 0;
1907 	}
1908 
1909 	ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1910 	if (ret < 0)
1911 		return ret;
1912 	hw->speed = speed;
1913 	hw->duplex = DUPLEX_UNKNOWN;
1914 
1915 	/* Allocate memory for storing MAC addresses */
1916 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1917 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1918 	if (eth_dev->data->mac_addrs == NULL) {
1919 		PMD_INIT_LOG(ERR,
1920 			"Failed to allocate %d bytes needed to store MAC addresses",
1921 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1922 		return -ENOMEM;
1923 	}
1924 
1925 	rte_spinlock_init(&hw->state_lock);
1926 
1927 	/* reset device and negotiate default features */
1928 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1929 	if (ret < 0)
1930 		goto err_virtio_init;
1931 
1932 	if (vectorized) {
1933 		if (!virtio_with_packed_queue(hw)) {
1934 			hw->use_vec_rx = 1;
1935 		} else {
1936 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1937 			hw->use_vec_rx = 1;
1938 			hw->use_vec_tx = 1;
1939 #else
1940 			PMD_DRV_LOG(INFO,
1941 				"building environment do not support packed ring vectorized");
1942 #endif
1943 		}
1944 	}
1945 
1946 	hw->opened = 1;
1947 
1948 	return 0;
1949 
1950 err_virtio_init:
1951 	rte_free(eth_dev->data->mac_addrs);
1952 	eth_dev->data->mac_addrs = NULL;
1953 	return ret;
1954 }
1955 
1956 static uint32_t
1957 virtio_dev_speed_capa_get(uint32_t speed)
1958 {
1959 	switch (speed) {
1960 	case ETH_SPEED_NUM_10G:
1961 		return ETH_LINK_SPEED_10G;
1962 	case ETH_SPEED_NUM_20G:
1963 		return ETH_LINK_SPEED_20G;
1964 	case ETH_SPEED_NUM_25G:
1965 		return ETH_LINK_SPEED_25G;
1966 	case ETH_SPEED_NUM_40G:
1967 		return ETH_LINK_SPEED_40G;
1968 	case ETH_SPEED_NUM_50G:
1969 		return ETH_LINK_SPEED_50G;
1970 	case ETH_SPEED_NUM_56G:
1971 		return ETH_LINK_SPEED_56G;
1972 	case ETH_SPEED_NUM_100G:
1973 		return ETH_LINK_SPEED_100G;
1974 	case ETH_SPEED_NUM_200G:
1975 		return ETH_LINK_SPEED_200G;
1976 	default:
1977 		return 0;
1978 	}
1979 }
1980 
1981 static int vectorized_check_handler(__rte_unused const char *key,
1982 		const char *value, void *ret_val)
1983 {
1984 	if (strcmp(value, "1") == 0)
1985 		*(int *)ret_val = 1;
1986 	else
1987 		*(int *)ret_val = 0;
1988 
1989 	return 0;
1990 }
1991 
1992 #define VIRTIO_ARG_SPEED      "speed"
1993 #define VIRTIO_ARG_VECTORIZED "vectorized"
1994 
1995 static int
1996 link_speed_handler(const char *key __rte_unused,
1997 		const char *value, void *ret_val)
1998 {
1999 	uint32_t val;
2000 	if (!value || !ret_val)
2001 		return -EINVAL;
2002 	val = strtoul(value, NULL, 0);
2003 	/* validate input */
2004 	if (virtio_dev_speed_capa_get(val) == 0)
2005 		return -EINVAL;
2006 	*(uint32_t *)ret_val = val;
2007 
2008 	return 0;
2009 }
2010 
2011 
2012 static int
2013 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
2014 {
2015 	struct rte_kvargs *kvlist;
2016 	int ret = 0;
2017 
2018 	if (devargs == NULL)
2019 		return 0;
2020 
2021 	kvlist = rte_kvargs_parse(devargs->args, NULL);
2022 	if (kvlist == NULL) {
2023 		PMD_INIT_LOG(ERR, "error when parsing param");
2024 		return 0;
2025 	}
2026 
2027 	if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2028 		ret = rte_kvargs_process(kvlist,
2029 					VIRTIO_ARG_SPEED,
2030 					link_speed_handler, speed);
2031 		if (ret < 0) {
2032 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2033 					VIRTIO_ARG_SPEED);
2034 			goto exit;
2035 		}
2036 	}
2037 
2038 	if (vectorized &&
2039 		rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2040 		ret = rte_kvargs_process(kvlist,
2041 				VIRTIO_ARG_VECTORIZED,
2042 				vectorized_check_handler, vectorized);
2043 		if (ret < 0) {
2044 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2045 					VIRTIO_ARG_VECTORIZED);
2046 			goto exit;
2047 		}
2048 	}
2049 
2050 exit:
2051 	rte_kvargs_free(kvlist);
2052 	return ret;
2053 }
2054 
2055 static uint8_t
2056 rx_offload_enabled(struct virtio_hw *hw)
2057 {
2058 	return virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2059 		virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2060 		virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2061 }
2062 
2063 static uint8_t
2064 tx_offload_enabled(struct virtio_hw *hw)
2065 {
2066 	return virtio_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2067 		virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2068 		virtio_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2069 }
2070 
2071 /*
2072  * Configure virtio device
2073  * It returns 0 on success.
2074  */
2075 static int
2076 virtio_dev_configure(struct rte_eth_dev *dev)
2077 {
2078 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2079 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2080 	struct virtio_hw *hw = dev->data->dev_private;
2081 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2082 		hw->vtnet_hdr_size;
2083 	uint64_t rx_offloads = rxmode->offloads;
2084 	uint64_t tx_offloads = txmode->offloads;
2085 	uint64_t req_features;
2086 	int ret;
2087 
2088 	PMD_INIT_LOG(DEBUG, "configure");
2089 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2090 
2091 	if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2092 		PMD_DRV_LOG(ERR,
2093 			"Unsupported Rx multi queue mode %d",
2094 			rxmode->mq_mode);
2095 		return -EINVAL;
2096 	}
2097 
2098 	if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2099 		PMD_DRV_LOG(ERR,
2100 			"Unsupported Tx multi queue mode %d",
2101 			txmode->mq_mode);
2102 		return -EINVAL;
2103 	}
2104 
2105 	if (dev->data->dev_conf.intr_conf.rxq) {
2106 		ret = virtio_init_device(dev, hw->req_guest_features);
2107 		if (ret < 0)
2108 			return ret;
2109 	}
2110 
2111 	if ((rx_offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) &&
2112 	    (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len))
2113 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2114 
2115 	if (rx_offloads & DEV_RX_OFFLOAD_JUMBO_FRAME)
2116 		hw->max_rx_pkt_len = rxmode->max_rx_pkt_len;
2117 	else
2118 		hw->max_rx_pkt_len = ether_hdr_len + dev->data->mtu;
2119 
2120 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2121 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2122 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2123 
2124 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2125 		req_features |=
2126 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2127 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2128 
2129 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2130 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2131 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2132 
2133 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2134 		req_features |=
2135 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2136 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2137 
2138 	/* if request features changed, reinit the device */
2139 	if (req_features != hw->req_guest_features) {
2140 		ret = virtio_init_device(dev, req_features);
2141 		if (ret < 0)
2142 			return ret;
2143 	}
2144 
2145 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2146 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2147 		!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2148 		PMD_DRV_LOG(ERR,
2149 			"rx checksum not available on this host");
2150 		return -ENOTSUP;
2151 	}
2152 
2153 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2154 		(!virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2155 		 !virtio_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2156 		PMD_DRV_LOG(ERR,
2157 			"Large Receive Offload not available on this host");
2158 		return -ENOTSUP;
2159 	}
2160 
2161 	/* start control queue */
2162 	if (virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2163 		virtio_dev_cq_start(dev);
2164 
2165 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2166 		hw->vlan_strip = 1;
2167 
2168 	hw->rx_ol_scatter = (rx_offloads & DEV_RX_OFFLOAD_SCATTER);
2169 
2170 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2171 			!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2172 		PMD_DRV_LOG(ERR,
2173 			    "vlan filtering not available on this host");
2174 		return -ENOTSUP;
2175 	}
2176 
2177 	hw->has_tx_offload = tx_offload_enabled(hw);
2178 	hw->has_rx_offload = rx_offload_enabled(hw);
2179 
2180 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2181 		/* Enable vector (0) for Link State Intrerrupt */
2182 		if (VIRTIO_OPS(hw)->set_config_irq(hw, 0) ==
2183 				VIRTIO_MSI_NO_VECTOR) {
2184 			PMD_DRV_LOG(ERR, "failed to set config vector");
2185 			return -EBUSY;
2186 		}
2187 
2188 	if (virtio_with_packed_queue(hw)) {
2189 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2190 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2191 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2192 		     !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2193 		     !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2194 		     rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2195 			PMD_DRV_LOG(INFO,
2196 				"disabled packed ring vectorized path for requirements not met");
2197 			hw->use_vec_rx = 0;
2198 			hw->use_vec_tx = 0;
2199 		}
2200 #elif defined(RTE_ARCH_ARM)
2201 		if ((hw->use_vec_rx || hw->use_vec_tx) &&
2202 		    (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2203 		     !virtio_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2204 		     !virtio_with_feature(hw, VIRTIO_F_VERSION_1) ||
2205 		     rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2206 			PMD_DRV_LOG(INFO,
2207 				"disabled packed ring vectorized path for requirements not met");
2208 			hw->use_vec_rx = 0;
2209 			hw->use_vec_tx = 0;
2210 		}
2211 #else
2212 		hw->use_vec_rx = 0;
2213 		hw->use_vec_tx = 0;
2214 #endif
2215 
2216 		if (hw->use_vec_rx) {
2217 			if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2218 				PMD_DRV_LOG(INFO,
2219 					"disabled packed ring vectorized rx for mrg_rxbuf enabled");
2220 				hw->use_vec_rx = 0;
2221 			}
2222 
2223 			if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2224 				PMD_DRV_LOG(INFO,
2225 					"disabled packed ring vectorized rx for TCP_LRO enabled");
2226 				hw->use_vec_rx = 0;
2227 			}
2228 		}
2229 	} else {
2230 		if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2231 			hw->use_inorder_tx = 1;
2232 			hw->use_inorder_rx = 1;
2233 			hw->use_vec_rx = 0;
2234 		}
2235 
2236 		if (hw->use_vec_rx) {
2237 #if defined RTE_ARCH_ARM
2238 			if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2239 				PMD_DRV_LOG(INFO,
2240 					"disabled split ring vectorized path for requirement not met");
2241 				hw->use_vec_rx = 0;
2242 			}
2243 #endif
2244 			if (virtio_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2245 				PMD_DRV_LOG(INFO,
2246 					"disabled split ring vectorized rx for mrg_rxbuf enabled");
2247 				hw->use_vec_rx = 0;
2248 			}
2249 
2250 			if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2251 					   DEV_RX_OFFLOAD_TCP_CKSUM |
2252 					   DEV_RX_OFFLOAD_TCP_LRO |
2253 					   DEV_RX_OFFLOAD_VLAN_STRIP)) {
2254 				PMD_DRV_LOG(INFO,
2255 					"disabled split ring vectorized rx for offloading enabled");
2256 				hw->use_vec_rx = 0;
2257 			}
2258 
2259 			if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2260 				PMD_DRV_LOG(INFO,
2261 					"disabled split ring vectorized rx, max SIMD bitwidth too low");
2262 				hw->use_vec_rx = 0;
2263 			}
2264 		}
2265 	}
2266 
2267 	return 0;
2268 }
2269 
2270 
2271 static int
2272 virtio_dev_start(struct rte_eth_dev *dev)
2273 {
2274 	uint16_t nb_queues, i;
2275 	struct virtqueue *vq;
2276 	struct virtio_hw *hw = dev->data->dev_private;
2277 	int ret;
2278 
2279 	/* Finish the initialization of the queues */
2280 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2281 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2282 		if (ret < 0)
2283 			return ret;
2284 	}
2285 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2286 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2287 		if (ret < 0)
2288 			return ret;
2289 	}
2290 
2291 	/* check if lsc interrupt feature is enabled */
2292 	if (dev->data->dev_conf.intr_conf.lsc) {
2293 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2294 			PMD_DRV_LOG(ERR, "link status not supported by host");
2295 			return -ENOTSUP;
2296 		}
2297 	}
2298 
2299 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2300 	 * in device configure, but it could be unmapped  when device is
2301 	 * stopped.
2302 	 */
2303 	if (dev->data->dev_conf.intr_conf.lsc ||
2304 	    dev->data->dev_conf.intr_conf.rxq) {
2305 		virtio_intr_disable(dev);
2306 
2307 		/* Setup interrupt callback  */
2308 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2309 			rte_intr_callback_register(dev->intr_handle,
2310 						   virtio_interrupt_handler,
2311 						   dev);
2312 
2313 		if (virtio_intr_enable(dev) < 0) {
2314 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2315 			return -EIO;
2316 		}
2317 	}
2318 
2319 	/*Notify the backend
2320 	 *Otherwise the tap backend might already stop its queue due to fullness.
2321 	 *vhost backend will have no chance to be waked up
2322 	 */
2323 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2324 	if (hw->max_queue_pairs > 1) {
2325 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2326 			return -EINVAL;
2327 	}
2328 
2329 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2330 
2331 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2332 		vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2333 		/* Flush the old packets */
2334 		virtqueue_rxvq_flush(vq);
2335 		virtqueue_notify(vq);
2336 	}
2337 
2338 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2339 		vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2340 		virtqueue_notify(vq);
2341 	}
2342 
2343 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2344 
2345 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2346 		vq = virtnet_rxq_to_vq(dev->data->rx_queues[i]);
2347 		VIRTQUEUE_DUMP(vq);
2348 	}
2349 
2350 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2351 		vq = virtnet_txq_to_vq(dev->data->tx_queues[i]);
2352 		VIRTQUEUE_DUMP(vq);
2353 	}
2354 
2355 	set_rxtx_funcs(dev);
2356 	hw->started = 1;
2357 
2358 	/* Initialize Link state */
2359 	virtio_dev_link_update(dev, 0);
2360 
2361 	return 0;
2362 }
2363 
2364 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2365 {
2366 	struct virtio_hw *hw = dev->data->dev_private;
2367 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2368 	const char *type __rte_unused;
2369 	unsigned int i, mbuf_num = 0;
2370 	struct virtqueue *vq;
2371 	struct rte_mbuf *buf;
2372 	int queue_type;
2373 
2374 	if (hw->vqs == NULL)
2375 		return;
2376 
2377 	for (i = 0; i < nr_vq; i++) {
2378 		vq = hw->vqs[i];
2379 		if (!vq)
2380 			continue;
2381 
2382 		queue_type = virtio_get_queue_type(hw, i);
2383 		if (queue_type == VTNET_RQ)
2384 			type = "rxq";
2385 		else if (queue_type == VTNET_TQ)
2386 			type = "txq";
2387 		else
2388 			continue;
2389 
2390 		PMD_INIT_LOG(DEBUG,
2391 			"Before freeing %s[%d] used and unused buf",
2392 			type, i);
2393 		VIRTQUEUE_DUMP(vq);
2394 
2395 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2396 			rte_pktmbuf_free(buf);
2397 			mbuf_num++;
2398 		}
2399 
2400 		PMD_INIT_LOG(DEBUG,
2401 			"After freeing %s[%d] used and unused buf",
2402 			type, i);
2403 		VIRTQUEUE_DUMP(vq);
2404 	}
2405 
2406 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2407 }
2408 
2409 static void
2410 virtio_tx_completed_cleanup(struct rte_eth_dev *dev)
2411 {
2412 	struct virtio_hw *hw = dev->data->dev_private;
2413 	struct virtqueue *vq;
2414 	int qidx;
2415 	void (*xmit_cleanup)(struct virtqueue *vq, uint16_t nb_used);
2416 
2417 	if (virtio_with_packed_queue(hw)) {
2418 		if (hw->use_vec_tx)
2419 			xmit_cleanup = &virtio_xmit_cleanup_inorder_packed;
2420 		else if (virtio_with_feature(hw, VIRTIO_F_IN_ORDER))
2421 			xmit_cleanup = &virtio_xmit_cleanup_inorder_packed;
2422 		else
2423 			xmit_cleanup = &virtio_xmit_cleanup_normal_packed;
2424 	} else {
2425 		if (hw->use_inorder_tx)
2426 			xmit_cleanup = &virtio_xmit_cleanup_inorder;
2427 		else
2428 			xmit_cleanup = &virtio_xmit_cleanup;
2429 	}
2430 
2431 	for (qidx = 0; qidx < hw->max_queue_pairs; qidx++) {
2432 		vq = hw->vqs[2 * qidx + VTNET_SQ_TQ_QUEUE_IDX];
2433 		if (vq != NULL)
2434 			xmit_cleanup(vq, virtqueue_nused(vq));
2435 	}
2436 }
2437 
2438 /*
2439  * Stop device: disable interrupt and mark link down
2440  */
2441 int
2442 virtio_dev_stop(struct rte_eth_dev *dev)
2443 {
2444 	struct virtio_hw *hw = dev->data->dev_private;
2445 	struct rte_eth_link link;
2446 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2447 
2448 	PMD_INIT_LOG(DEBUG, "stop");
2449 	dev->data->dev_started = 0;
2450 
2451 	rte_spinlock_lock(&hw->state_lock);
2452 	if (!hw->started)
2453 		goto out_unlock;
2454 	hw->started = 0;
2455 
2456 	virtio_tx_completed_cleanup(dev);
2457 
2458 	if (intr_conf->lsc || intr_conf->rxq) {
2459 		virtio_intr_disable(dev);
2460 
2461 		/* Reset interrupt callback  */
2462 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2463 			rte_intr_callback_unregister(dev->intr_handle,
2464 						     virtio_interrupt_handler,
2465 						     dev);
2466 		}
2467 	}
2468 
2469 	memset(&link, 0, sizeof(link));
2470 	rte_eth_linkstatus_set(dev, &link);
2471 out_unlock:
2472 	rte_spinlock_unlock(&hw->state_lock);
2473 
2474 	return 0;
2475 }
2476 
2477 static int
2478 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2479 {
2480 	struct rte_eth_link link;
2481 	uint16_t status;
2482 	struct virtio_hw *hw = dev->data->dev_private;
2483 
2484 	memset(&link, 0, sizeof(link));
2485 	link.link_duplex = hw->duplex;
2486 	link.link_speed  = hw->speed;
2487 	link.link_autoneg = ETH_LINK_AUTONEG;
2488 
2489 	if (!hw->started) {
2490 		link.link_status = ETH_LINK_DOWN;
2491 		link.link_speed = ETH_SPEED_NUM_NONE;
2492 	} else if (virtio_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2493 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2494 		virtio_read_dev_config(hw,
2495 				offsetof(struct virtio_net_config, status),
2496 				&status, sizeof(status));
2497 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2498 			link.link_status = ETH_LINK_DOWN;
2499 			link.link_speed = ETH_SPEED_NUM_NONE;
2500 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2501 				     dev->data->port_id);
2502 		} else {
2503 			link.link_status = ETH_LINK_UP;
2504 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2505 				     dev->data->port_id);
2506 		}
2507 	} else {
2508 		link.link_status = ETH_LINK_UP;
2509 	}
2510 
2511 	return rte_eth_linkstatus_set(dev, &link);
2512 }
2513 
2514 static int
2515 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2516 {
2517 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2518 	struct virtio_hw *hw = dev->data->dev_private;
2519 	uint64_t offloads = rxmode->offloads;
2520 
2521 	if (mask & ETH_VLAN_FILTER_MASK) {
2522 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2523 				!virtio_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2524 
2525 			PMD_DRV_LOG(NOTICE,
2526 				"vlan filtering not available on this host");
2527 
2528 			return -ENOTSUP;
2529 		}
2530 	}
2531 
2532 	if (mask & ETH_VLAN_STRIP_MASK)
2533 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2534 
2535 	return 0;
2536 }
2537 
2538 static int
2539 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2540 {
2541 	uint64_t tso_mask, host_features;
2542 	struct virtio_hw *hw = dev->data->dev_private;
2543 	dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2544 
2545 	dev_info->max_rx_queues =
2546 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2547 	dev_info->max_tx_queues =
2548 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2549 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2550 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2551 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2552 	dev_info->max_mtu = hw->max_mtu;
2553 
2554 	host_features = VIRTIO_OPS(hw)->get_features(hw);
2555 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2556 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2557 	if (host_features & (1ULL << VIRTIO_NET_F_MRG_RXBUF))
2558 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_SCATTER;
2559 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2560 		dev_info->rx_offload_capa |=
2561 			DEV_RX_OFFLOAD_TCP_CKSUM |
2562 			DEV_RX_OFFLOAD_UDP_CKSUM;
2563 	}
2564 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2565 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2566 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2567 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2568 	if ((host_features & tso_mask) == tso_mask)
2569 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2570 
2571 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2572 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2573 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2574 		dev_info->tx_offload_capa |=
2575 			DEV_TX_OFFLOAD_UDP_CKSUM |
2576 			DEV_TX_OFFLOAD_TCP_CKSUM;
2577 	}
2578 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2579 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2580 	if ((host_features & tso_mask) == tso_mask)
2581 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2582 
2583 	if (host_features & (1ULL << VIRTIO_F_RING_PACKED)) {
2584 		/*
2585 		 * According to 2.7 Packed Virtqueues,
2586 		 * 2.7.10.1 Structure Size and Alignment:
2587 		 * The Queue Size value does not have to be a power of 2.
2588 		 */
2589 		dev_info->rx_desc_lim.nb_max = UINT16_MAX;
2590 		dev_info->tx_desc_lim.nb_max = UINT16_MAX;
2591 	} else {
2592 		/*
2593 		 * According to 2.6 Split Virtqueues:
2594 		 * Queue Size value is always a power of 2. The maximum Queue
2595 		 * Size value is 32768.
2596 		 */
2597 		dev_info->rx_desc_lim.nb_max = 32768;
2598 		dev_info->tx_desc_lim.nb_max = 32768;
2599 	}
2600 	/*
2601 	 * Actual minimum is not the same for virtqueues of different kinds,
2602 	 * but to avoid tangling the code with separate branches, rely on
2603 	 * default thresholds since desc number must be at least of their size.
2604 	 */
2605 	dev_info->rx_desc_lim.nb_min = RTE_MAX(DEFAULT_RX_FREE_THRESH,
2606 					       RTE_VIRTIO_VPMD_RX_REARM_THRESH);
2607 	dev_info->tx_desc_lim.nb_min = DEFAULT_TX_FREE_THRESH;
2608 	dev_info->rx_desc_lim.nb_align = 1;
2609 	dev_info->tx_desc_lim.nb_align = 1;
2610 
2611 	return 0;
2612 }
2613 
2614 /*
2615  * It enables testpmd to collect per queue stats.
2616  */
2617 static int
2618 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2619 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2620 __rte_unused uint8_t is_rx)
2621 {
2622 	return 0;
2623 }
2624 
2625 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_init, init, NOTICE);
2626 RTE_LOG_REGISTER_SUFFIX(virtio_logtype_driver, driver, NOTICE);
2627