xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision a8a544890fe00054f05ac05ccb78e6fdc1ed4ba9)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4 
5 #include <stdint.h>
6 #include <string.h>
7 #include <stdio.h>
8 #include <errno.h>
9 #include <unistd.h>
10 
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_memcpy.h>
14 #include <rte_string_fns.h>
15 #include <rte_memzone.h>
16 #include <rte_malloc.h>
17 #include <rte_branch_prediction.h>
18 #include <rte_pci.h>
19 #include <rte_bus_pci.h>
20 #include <rte_ether.h>
21 #include <rte_ip.h>
22 #include <rte_arp.h>
23 #include <rte_common.h>
24 #include <rte_errno.h>
25 #include <rte_cpuflags.h>
26 
27 #include <rte_memory.h>
28 #include <rte_eal.h>
29 #include <rte_dev.h>
30 #include <rte_cycles.h>
31 #include <rte_kvargs.h>
32 
33 #include "virtio_ethdev.h"
34 #include "virtio_pci.h"
35 #include "virtio_logs.h"
36 #include "virtqueue.h"
37 #include "virtio_rxtx.h"
38 #include "virtio_user/virtio_user_dev.h"
39 
40 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
41 static int  virtio_dev_configure(struct rte_eth_dev *dev);
42 static int  virtio_dev_start(struct rte_eth_dev *dev);
43 static void virtio_dev_stop(struct rte_eth_dev *dev);
44 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
45 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
46 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
47 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
48 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
49 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
50 	int *vdpa,
51 	uint32_t *speed);
52 static int virtio_dev_info_get(struct rte_eth_dev *dev,
53 				struct rte_eth_dev_info *dev_info);
54 static int virtio_dev_link_update(struct rte_eth_dev *dev,
55 	int wait_to_complete);
56 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
57 
58 static void virtio_set_hwaddr(struct virtio_hw *hw);
59 static void virtio_get_hwaddr(struct virtio_hw *hw);
60 
61 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
62 				 struct rte_eth_stats *stats);
63 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
64 				 struct rte_eth_xstat *xstats, unsigned n);
65 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
66 				       struct rte_eth_xstat_name *xstats_names,
67 				       unsigned limit);
68 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
69 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
70 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
71 				uint16_t vlan_id, int on);
72 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
73 				struct rte_ether_addr *mac_addr,
74 				uint32_t index, uint32_t vmdq);
75 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
76 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
77 				struct rte_ether_addr *mac_addr);
78 
79 static int virtio_intr_disable(struct rte_eth_dev *dev);
80 
81 static int virtio_dev_queue_stats_mapping_set(
82 	struct rte_eth_dev *eth_dev,
83 	uint16_t queue_id,
84 	uint8_t stat_idx,
85 	uint8_t is_rx);
86 
87 int virtio_logtype_init;
88 int virtio_logtype_driver;
89 
90 static void virtio_notify_peers(struct rte_eth_dev *dev);
91 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
92 
93 /*
94  * The set of PCI devices this driver supports
95  */
96 static const struct rte_pci_id pci_id_virtio_map[] = {
97 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
98 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
99 	{ .vendor_id = 0, /* sentinel */ },
100 };
101 
102 struct rte_virtio_xstats_name_off {
103 	char name[RTE_ETH_XSTATS_NAME_SIZE];
104 	unsigned offset;
105 };
106 
107 /* [rt]x_qX_ is prepended to the name string here */
108 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
109 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
110 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
111 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
112 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
113 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
114 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
115 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
116 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
117 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
118 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
119 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
120 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
121 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
122 };
123 
124 /* [rt]x_qX_ is prepended to the name string here */
125 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
126 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
127 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
128 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
129 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
130 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
131 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
132 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
133 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
134 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
135 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
136 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
137 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
138 };
139 
140 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
141 			    sizeof(rte_virtio_rxq_stat_strings[0]))
142 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
143 			    sizeof(rte_virtio_txq_stat_strings[0]))
144 
145 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
146 
147 static struct virtio_pmd_ctrl *
148 virtio_send_command_packed(struct virtnet_ctl *cvq,
149 			   struct virtio_pmd_ctrl *ctrl,
150 			   int *dlen, int pkt_num)
151 {
152 	struct virtqueue *vq = cvq->vq;
153 	int head;
154 	struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
155 	struct virtio_pmd_ctrl *result;
156 	uint16_t flags;
157 	int sum = 0;
158 	int nb_descs = 0;
159 	int k;
160 
161 	/*
162 	 * Format is enforced in qemu code:
163 	 * One TX packet for header;
164 	 * At least one TX packet per argument;
165 	 * One RX packet for ACK.
166 	 */
167 	head = vq->vq_avail_idx;
168 	flags = vq->vq_packed.cached_flags;
169 	desc[head].addr = cvq->virtio_net_hdr_mem;
170 	desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
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 ^= VRING_PACKED_DESC_F_AVAIL_USED;
176 	}
177 
178 	for (k = 0; k < pkt_num; k++) {
179 		desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
180 			+ sizeof(struct virtio_net_ctrl_hdr)
181 			+ sizeof(ctrl->status) + sizeof(uint8_t) * sum;
182 		desc[vq->vq_avail_idx].len = dlen[k];
183 		desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
184 			vq->vq_packed.cached_flags;
185 		sum += dlen[k];
186 		vq->vq_free_cnt--;
187 		nb_descs++;
188 		if (++vq->vq_avail_idx >= vq->vq_nentries) {
189 			vq->vq_avail_idx -= vq->vq_nentries;
190 			vq->vq_packed.cached_flags ^=
191 				VRING_PACKED_DESC_F_AVAIL_USED;
192 		}
193 	}
194 
195 	desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
196 		+ sizeof(struct virtio_net_ctrl_hdr);
197 	desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
198 	desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
199 		vq->vq_packed.cached_flags;
200 	vq->vq_free_cnt--;
201 	nb_descs++;
202 	if (++vq->vq_avail_idx >= vq->vq_nentries) {
203 		vq->vq_avail_idx -= vq->vq_nentries;
204 		vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
205 	}
206 
207 	virtio_wmb(vq->hw->weak_barriers);
208 	desc[head].flags = VRING_DESC_F_NEXT | flags;
209 
210 	virtio_wmb(vq->hw->weak_barriers);
211 	virtqueue_notify(vq);
212 
213 	/* wait for used descriptors in virtqueue */
214 	while (!desc_is_used(&desc[head], vq))
215 		usleep(100);
216 
217 	virtio_rmb(vq->hw->weak_barriers);
218 
219 	/* now get used descriptors */
220 	vq->vq_free_cnt += nb_descs;
221 	vq->vq_used_cons_idx += nb_descs;
222 	if (vq->vq_used_cons_idx >= vq->vq_nentries) {
223 		vq->vq_used_cons_idx -= vq->vq_nentries;
224 		vq->vq_packed.used_wrap_counter ^= 1;
225 	}
226 
227 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
228 			"vq->vq_avail_idx=%d\n"
229 			"vq->vq_used_cons_idx=%d\n"
230 			"vq->vq_packed.cached_flags=0x%x\n"
231 			"vq->vq_packed.used_wrap_counter=%d\n",
232 			vq->vq_free_cnt,
233 			vq->vq_avail_idx,
234 			vq->vq_used_cons_idx,
235 			vq->vq_packed.cached_flags,
236 			vq->vq_packed.used_wrap_counter);
237 
238 	result = cvq->virtio_net_hdr_mz->addr;
239 	return result;
240 }
241 
242 static struct virtio_pmd_ctrl *
243 virtio_send_command_split(struct virtnet_ctl *cvq,
244 			  struct virtio_pmd_ctrl *ctrl,
245 			  int *dlen, int pkt_num)
246 {
247 	struct virtio_pmd_ctrl *result;
248 	struct virtqueue *vq = cvq->vq;
249 	uint32_t head, i;
250 	int k, sum = 0;
251 
252 	head = vq->vq_desc_head_idx;
253 
254 	/*
255 	 * Format is enforced in qemu code:
256 	 * One TX packet for header;
257 	 * At least one TX packet per argument;
258 	 * One RX packet for ACK.
259 	 */
260 	vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
261 	vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
262 	vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
263 	vq->vq_free_cnt--;
264 	i = vq->vq_split.ring.desc[head].next;
265 
266 	for (k = 0; k < pkt_num; k++) {
267 		vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
268 		vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
269 			+ sizeof(struct virtio_net_ctrl_hdr)
270 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
271 		vq->vq_split.ring.desc[i].len = dlen[k];
272 		sum += dlen[k];
273 		vq->vq_free_cnt--;
274 		i = vq->vq_split.ring.desc[i].next;
275 	}
276 
277 	vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
278 	vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
279 			+ sizeof(struct virtio_net_ctrl_hdr);
280 	vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
281 	vq->vq_free_cnt--;
282 
283 	vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
284 
285 	vq_update_avail_ring(vq, head);
286 	vq_update_avail_idx(vq);
287 
288 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
289 
290 	virtqueue_notify(vq);
291 
292 	rte_rmb();
293 	while (VIRTQUEUE_NUSED(vq) == 0) {
294 		rte_rmb();
295 		usleep(100);
296 	}
297 
298 	while (VIRTQUEUE_NUSED(vq)) {
299 		uint32_t idx, desc_idx, used_idx;
300 		struct vring_used_elem *uep;
301 
302 		used_idx = (uint32_t)(vq->vq_used_cons_idx
303 				& (vq->vq_nentries - 1));
304 		uep = &vq->vq_split.ring.used->ring[used_idx];
305 		idx = (uint32_t) uep->id;
306 		desc_idx = idx;
307 
308 		while (vq->vq_split.ring.desc[desc_idx].flags &
309 				VRING_DESC_F_NEXT) {
310 			desc_idx = vq->vq_split.ring.desc[desc_idx].next;
311 			vq->vq_free_cnt++;
312 		}
313 
314 		vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
315 		vq->vq_desc_head_idx = idx;
316 
317 		vq->vq_used_cons_idx++;
318 		vq->vq_free_cnt++;
319 	}
320 
321 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
322 			vq->vq_free_cnt, vq->vq_desc_head_idx);
323 
324 	result = cvq->virtio_net_hdr_mz->addr;
325 	return result;
326 }
327 
328 static int
329 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
330 		    int *dlen, int pkt_num)
331 {
332 	virtio_net_ctrl_ack status = ~0;
333 	struct virtio_pmd_ctrl *result;
334 	struct virtqueue *vq;
335 
336 	ctrl->status = status;
337 
338 	if (!cvq || !cvq->vq) {
339 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
340 		return -1;
341 	}
342 
343 	rte_spinlock_lock(&cvq->lock);
344 	vq = cvq->vq;
345 
346 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
347 		"vq->hw->cvq = %p vq = %p",
348 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
349 
350 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
351 		rte_spinlock_unlock(&cvq->lock);
352 		return -1;
353 	}
354 
355 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
356 		sizeof(struct virtio_pmd_ctrl));
357 
358 	if (vtpci_packed_queue(vq->hw))
359 		result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
360 	else
361 		result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
362 
363 	rte_spinlock_unlock(&cvq->lock);
364 	return result->status;
365 }
366 
367 static int
368 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
369 {
370 	struct virtio_hw *hw = dev->data->dev_private;
371 	struct virtio_pmd_ctrl ctrl;
372 	int dlen[1];
373 	int ret;
374 
375 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
376 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
377 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
378 
379 	dlen[0] = sizeof(uint16_t);
380 
381 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
382 	if (ret) {
383 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
384 			  "failed, this is too late now...");
385 		return -EINVAL;
386 	}
387 
388 	return 0;
389 }
390 
391 static void
392 virtio_dev_queue_release(void *queue __rte_unused)
393 {
394 	/* do nothing */
395 }
396 
397 static uint16_t
398 virtio_get_nr_vq(struct virtio_hw *hw)
399 {
400 	uint16_t nr_vq = hw->max_queue_pairs * 2;
401 
402 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
403 		nr_vq += 1;
404 
405 	return nr_vq;
406 }
407 
408 static void
409 virtio_init_vring(struct virtqueue *vq)
410 {
411 	int size = vq->vq_nentries;
412 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
413 
414 	PMD_INIT_FUNC_TRACE();
415 
416 	memset(ring_mem, 0, vq->vq_ring_size);
417 
418 	vq->vq_used_cons_idx = 0;
419 	vq->vq_desc_head_idx = 0;
420 	vq->vq_avail_idx = 0;
421 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
422 	vq->vq_free_cnt = vq->vq_nentries;
423 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
424 	if (vtpci_packed_queue(vq->hw)) {
425 		vring_init_packed(&vq->vq_packed.ring, ring_mem,
426 				  VIRTIO_PCI_VRING_ALIGN, size);
427 		vring_desc_init_packed(vq, size);
428 	} else {
429 		struct vring *vr = &vq->vq_split.ring;
430 
431 		vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
432 		vring_desc_init_split(vr->desc, size);
433 	}
434 	/*
435 	 * Disable device(host) interrupting guest
436 	 */
437 	virtqueue_disable_intr(vq);
438 }
439 
440 static int
441 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
442 {
443 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
444 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
445 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
446 	unsigned int vq_size, size;
447 	struct virtio_hw *hw = dev->data->dev_private;
448 	struct virtnet_rx *rxvq = NULL;
449 	struct virtnet_tx *txvq = NULL;
450 	struct virtnet_ctl *cvq = NULL;
451 	struct virtqueue *vq;
452 	size_t sz_hdr_mz = 0;
453 	void *sw_ring = NULL;
454 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
455 	int ret;
456 	int numa_node = dev->device->numa_node;
457 
458 	PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
459 			vtpci_queue_idx, numa_node);
460 
461 	/*
462 	 * Read the virtqueue size from the Queue Size field
463 	 * Always power of 2 and if 0 virtqueue does not exist
464 	 */
465 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
466 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
467 	if (vq_size == 0) {
468 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
469 		return -EINVAL;
470 	}
471 
472 	if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
473 		PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
474 		return -EINVAL;
475 	}
476 
477 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
478 		 dev->data->port_id, vtpci_queue_idx);
479 
480 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
481 				vq_size * sizeof(struct vq_desc_extra),
482 				RTE_CACHE_LINE_SIZE);
483 	if (queue_type == VTNET_TQ) {
484 		/*
485 		 * For each xmit packet, allocate a virtio_net_hdr
486 		 * and indirect ring elements
487 		 */
488 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
489 	} else if (queue_type == VTNET_CQ) {
490 		/* Allocate a page for control vq command, data and status */
491 		sz_hdr_mz = PAGE_SIZE;
492 	}
493 
494 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
495 				numa_node);
496 	if (vq == NULL) {
497 		PMD_INIT_LOG(ERR, "can not allocate vq");
498 		return -ENOMEM;
499 	}
500 	hw->vqs[vtpci_queue_idx] = vq;
501 
502 	vq->hw = hw;
503 	vq->vq_queue_index = vtpci_queue_idx;
504 	vq->vq_nentries = vq_size;
505 	if (vtpci_packed_queue(hw)) {
506 		vq->vq_packed.used_wrap_counter = 1;
507 		vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
508 		vq->vq_packed.event_flags_shadow = 0;
509 		if (queue_type == VTNET_RQ)
510 			vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
511 	}
512 
513 	/*
514 	 * Reserve a memzone for vring elements
515 	 */
516 	size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
517 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
518 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
519 		     size, vq->vq_ring_size);
520 
521 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
522 			numa_node, RTE_MEMZONE_IOVA_CONTIG,
523 			VIRTIO_PCI_VRING_ALIGN);
524 	if (mz == NULL) {
525 		if (rte_errno == EEXIST)
526 			mz = rte_memzone_lookup(vq_name);
527 		if (mz == NULL) {
528 			ret = -ENOMEM;
529 			goto fail_q_alloc;
530 		}
531 	}
532 
533 	memset(mz->addr, 0, mz->len);
534 
535 	vq->vq_ring_mem = mz->iova;
536 	vq->vq_ring_virt_mem = mz->addr;
537 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
538 		     (uint64_t)mz->iova);
539 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
540 		     (uint64_t)(uintptr_t)mz->addr);
541 
542 	virtio_init_vring(vq);
543 
544 	if (sz_hdr_mz) {
545 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
546 			 dev->data->port_id, vtpci_queue_idx);
547 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
548 				numa_node, RTE_MEMZONE_IOVA_CONTIG,
549 				RTE_CACHE_LINE_SIZE);
550 		if (hdr_mz == NULL) {
551 			if (rte_errno == EEXIST)
552 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
553 			if (hdr_mz == NULL) {
554 				ret = -ENOMEM;
555 				goto fail_q_alloc;
556 			}
557 		}
558 	}
559 
560 	if (queue_type == VTNET_RQ) {
561 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
562 			       sizeof(vq->sw_ring[0]);
563 
564 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
565 				RTE_CACHE_LINE_SIZE, numa_node);
566 		if (!sw_ring) {
567 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
568 			ret = -ENOMEM;
569 			goto fail_q_alloc;
570 		}
571 
572 		vq->sw_ring = sw_ring;
573 		rxvq = &vq->rxq;
574 		rxvq->vq = vq;
575 		rxvq->port_id = dev->data->port_id;
576 		rxvq->mz = mz;
577 	} else if (queue_type == VTNET_TQ) {
578 		txvq = &vq->txq;
579 		txvq->vq = vq;
580 		txvq->port_id = dev->data->port_id;
581 		txvq->mz = mz;
582 		txvq->virtio_net_hdr_mz = hdr_mz;
583 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
584 	} else if (queue_type == VTNET_CQ) {
585 		cvq = &vq->cq;
586 		cvq->vq = vq;
587 		cvq->mz = mz;
588 		cvq->virtio_net_hdr_mz = hdr_mz;
589 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
590 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
591 
592 		hw->cvq = cvq;
593 	}
594 
595 	/* For virtio_user case (that is when hw->virtio_user_dev is not NULL),
596 	 * we use virtual address. And we need properly set _offset_, please see
597 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
598 	 */
599 	if (!hw->virtio_user_dev)
600 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
601 	else {
602 		vq->vq_ring_mem = (uintptr_t)mz->addr;
603 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
604 		if (queue_type == VTNET_TQ)
605 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
606 		else if (queue_type == VTNET_CQ)
607 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
608 	}
609 
610 	if (queue_type == VTNET_TQ) {
611 		struct virtio_tx_region *txr;
612 		unsigned int i;
613 
614 		txr = hdr_mz->addr;
615 		memset(txr, 0, vq_size * sizeof(*txr));
616 		for (i = 0; i < vq_size; i++) {
617 			struct vring_desc *start_dp = txr[i].tx_indir;
618 
619 			/* first indirect descriptor is always the tx header */
620 			if (!vtpci_packed_queue(hw)) {
621 				vring_desc_init_split(start_dp,
622 						      RTE_DIM(txr[i].tx_indir));
623 				start_dp->addr = txvq->virtio_net_hdr_mem
624 					+ i * sizeof(*txr)
625 					+ offsetof(struct virtio_tx_region,
626 						   tx_hdr);
627 				start_dp->len = hw->vtnet_hdr_size;
628 				start_dp->flags = VRING_DESC_F_NEXT;
629 			}
630 		}
631 	}
632 
633 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
634 		PMD_INIT_LOG(ERR, "setup_queue failed");
635 		return -EINVAL;
636 	}
637 
638 	return 0;
639 
640 fail_q_alloc:
641 	rte_free(sw_ring);
642 	rte_memzone_free(hdr_mz);
643 	rte_memzone_free(mz);
644 	rte_free(vq);
645 
646 	return ret;
647 }
648 
649 static void
650 virtio_free_queues(struct virtio_hw *hw)
651 {
652 	uint16_t nr_vq = virtio_get_nr_vq(hw);
653 	struct virtqueue *vq;
654 	int queue_type;
655 	uint16_t i;
656 
657 	if (hw->vqs == NULL)
658 		return;
659 
660 	for (i = 0; i < nr_vq; i++) {
661 		vq = hw->vqs[i];
662 		if (!vq)
663 			continue;
664 
665 		queue_type = virtio_get_queue_type(hw, i);
666 		if (queue_type == VTNET_RQ) {
667 			rte_free(vq->sw_ring);
668 			rte_memzone_free(vq->rxq.mz);
669 		} else if (queue_type == VTNET_TQ) {
670 			rte_memzone_free(vq->txq.mz);
671 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
672 		} else {
673 			rte_memzone_free(vq->cq.mz);
674 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
675 		}
676 
677 		rte_free(vq);
678 		hw->vqs[i] = NULL;
679 	}
680 
681 	rte_free(hw->vqs);
682 	hw->vqs = NULL;
683 }
684 
685 static int
686 virtio_alloc_queues(struct rte_eth_dev *dev)
687 {
688 	struct virtio_hw *hw = dev->data->dev_private;
689 	uint16_t nr_vq = virtio_get_nr_vq(hw);
690 	uint16_t i;
691 	int ret;
692 
693 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
694 	if (!hw->vqs) {
695 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
696 		return -ENOMEM;
697 	}
698 
699 	for (i = 0; i < nr_vq; i++) {
700 		ret = virtio_init_queue(dev, i);
701 		if (ret < 0) {
702 			virtio_free_queues(hw);
703 			return ret;
704 		}
705 	}
706 
707 	return 0;
708 }
709 
710 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
711 
712 static void
713 virtio_dev_close(struct rte_eth_dev *dev)
714 {
715 	struct virtio_hw *hw = dev->data->dev_private;
716 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
717 
718 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
719 
720 	if (!hw->opened)
721 		return;
722 	hw->opened = false;
723 
724 	/* reset the NIC */
725 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
726 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
727 	if (intr_conf->rxq)
728 		virtio_queues_unbind_intr(dev);
729 
730 	if (intr_conf->lsc || intr_conf->rxq) {
731 		virtio_intr_disable(dev);
732 		rte_intr_efd_disable(dev->intr_handle);
733 		rte_free(dev->intr_handle->intr_vec);
734 		dev->intr_handle->intr_vec = NULL;
735 	}
736 
737 	vtpci_reset(hw);
738 	virtio_dev_free_mbufs(dev);
739 	virtio_free_queues(hw);
740 
741 #ifdef RTE_VIRTIO_USER
742 	if (hw->virtio_user_dev)
743 		virtio_user_dev_uninit(hw->virtio_user_dev);
744 	else
745 #endif
746 	if (dev->device) {
747 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
748 		if (!hw->modern)
749 			rte_pci_ioport_unmap(VTPCI_IO(hw));
750 	}
751 }
752 
753 static int
754 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
755 {
756 	struct virtio_hw *hw = dev->data->dev_private;
757 	struct virtio_pmd_ctrl ctrl;
758 	int dlen[1];
759 	int ret;
760 
761 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
762 		PMD_INIT_LOG(INFO, "host does not support rx control");
763 		return -ENOTSUP;
764 	}
765 
766 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
767 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
768 	ctrl.data[0] = 1;
769 	dlen[0] = 1;
770 
771 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
772 	if (ret) {
773 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
774 		return -EAGAIN;
775 	}
776 
777 	return 0;
778 }
779 
780 static int
781 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
782 {
783 	struct virtio_hw *hw = dev->data->dev_private;
784 	struct virtio_pmd_ctrl ctrl;
785 	int dlen[1];
786 	int ret;
787 
788 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
789 		PMD_INIT_LOG(INFO, "host does not support rx control");
790 		return -ENOTSUP;
791 	}
792 
793 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
794 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
795 	ctrl.data[0] = 0;
796 	dlen[0] = 1;
797 
798 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
799 	if (ret) {
800 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
801 		return -EAGAIN;
802 	}
803 
804 	return 0;
805 }
806 
807 static int
808 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
809 {
810 	struct virtio_hw *hw = dev->data->dev_private;
811 	struct virtio_pmd_ctrl ctrl;
812 	int dlen[1];
813 	int ret;
814 
815 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
816 		PMD_INIT_LOG(INFO, "host does not support rx control");
817 		return -ENOTSUP;
818 	}
819 
820 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
821 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
822 	ctrl.data[0] = 1;
823 	dlen[0] = 1;
824 
825 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
826 	if (ret) {
827 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
828 		return -EAGAIN;
829 	}
830 
831 	return 0;
832 }
833 
834 static int
835 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
836 {
837 	struct virtio_hw *hw = dev->data->dev_private;
838 	struct virtio_pmd_ctrl ctrl;
839 	int dlen[1];
840 	int ret;
841 
842 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
843 		PMD_INIT_LOG(INFO, "host does not support rx control");
844 		return -ENOTSUP;
845 	}
846 
847 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
848 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
849 	ctrl.data[0] = 0;
850 	dlen[0] = 1;
851 
852 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
853 	if (ret) {
854 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
855 		return -EAGAIN;
856 	}
857 
858 	return 0;
859 }
860 
861 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
862 static int
863 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
864 {
865 	struct virtio_hw *hw = dev->data->dev_private;
866 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
867 				 hw->vtnet_hdr_size;
868 	uint32_t frame_size = mtu + ether_hdr_len;
869 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
870 
871 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
872 
873 	if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
874 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
875 			RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
876 		return -EINVAL;
877 	}
878 	return 0;
879 }
880 
881 static int
882 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
883 {
884 	struct virtio_hw *hw = dev->data->dev_private;
885 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
886 	struct virtqueue *vq = rxvq->vq;
887 
888 	virtqueue_enable_intr(vq);
889 	virtio_mb(hw->weak_barriers);
890 	return 0;
891 }
892 
893 static int
894 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
895 {
896 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
897 	struct virtqueue *vq = rxvq->vq;
898 
899 	virtqueue_disable_intr(vq);
900 	return 0;
901 }
902 
903 /*
904  * dev_ops for virtio, bare necessities for basic operation
905  */
906 static const struct eth_dev_ops virtio_eth_dev_ops = {
907 	.dev_configure           = virtio_dev_configure,
908 	.dev_start               = virtio_dev_start,
909 	.dev_stop                = virtio_dev_stop,
910 	.dev_close               = virtio_dev_close,
911 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
912 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
913 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
914 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
915 	.mtu_set                 = virtio_mtu_set,
916 	.dev_infos_get           = virtio_dev_info_get,
917 	.stats_get               = virtio_dev_stats_get,
918 	.xstats_get              = virtio_dev_xstats_get,
919 	.xstats_get_names        = virtio_dev_xstats_get_names,
920 	.stats_reset             = virtio_dev_stats_reset,
921 	.xstats_reset            = virtio_dev_stats_reset,
922 	.link_update             = virtio_dev_link_update,
923 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
924 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
925 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
926 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
927 	.rx_queue_release        = virtio_dev_queue_release,
928 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
929 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
930 	.tx_queue_release        = virtio_dev_queue_release,
931 	/* collect stats per queue */
932 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
933 	.vlan_filter_set         = virtio_vlan_filter_set,
934 	.mac_addr_add            = virtio_mac_addr_add,
935 	.mac_addr_remove         = virtio_mac_addr_remove,
936 	.mac_addr_set            = virtio_mac_addr_set,
937 };
938 
939 /*
940  * dev_ops for virtio-user in secondary processes, as we just have
941  * some limited supports currently.
942  */
943 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
944 	.dev_infos_get           = virtio_dev_info_get,
945 	.stats_get               = virtio_dev_stats_get,
946 	.xstats_get              = virtio_dev_xstats_get,
947 	.xstats_get_names        = virtio_dev_xstats_get_names,
948 	.stats_reset             = virtio_dev_stats_reset,
949 	.xstats_reset            = virtio_dev_stats_reset,
950 	/* collect stats per queue */
951 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
952 };
953 
954 static void
955 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
956 {
957 	unsigned i;
958 
959 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
960 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
961 		if (txvq == NULL)
962 			continue;
963 
964 		stats->opackets += txvq->stats.packets;
965 		stats->obytes += txvq->stats.bytes;
966 
967 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
968 			stats->q_opackets[i] = txvq->stats.packets;
969 			stats->q_obytes[i] = txvq->stats.bytes;
970 		}
971 	}
972 
973 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
974 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
975 		if (rxvq == NULL)
976 			continue;
977 
978 		stats->ipackets += rxvq->stats.packets;
979 		stats->ibytes += rxvq->stats.bytes;
980 		stats->ierrors += rxvq->stats.errors;
981 
982 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
983 			stats->q_ipackets[i] = rxvq->stats.packets;
984 			stats->q_ibytes[i] = rxvq->stats.bytes;
985 		}
986 	}
987 
988 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
989 }
990 
991 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
992 				       struct rte_eth_xstat_name *xstats_names,
993 				       __rte_unused unsigned limit)
994 {
995 	unsigned i;
996 	unsigned count = 0;
997 	unsigned t;
998 
999 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1000 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1001 
1002 	if (xstats_names != NULL) {
1003 		/* Note: limit checked in rte_eth_xstats_names() */
1004 
1005 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
1006 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1007 			if (rxvq == NULL)
1008 				continue;
1009 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1010 				snprintf(xstats_names[count].name,
1011 					sizeof(xstats_names[count].name),
1012 					"rx_q%u_%s", i,
1013 					rte_virtio_rxq_stat_strings[t].name);
1014 				count++;
1015 			}
1016 		}
1017 
1018 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
1019 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
1020 			if (txvq == NULL)
1021 				continue;
1022 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1023 				snprintf(xstats_names[count].name,
1024 					sizeof(xstats_names[count].name),
1025 					"tx_q%u_%s", i,
1026 					rte_virtio_txq_stat_strings[t].name);
1027 				count++;
1028 			}
1029 		}
1030 		return count;
1031 	}
1032 	return nstats;
1033 }
1034 
1035 static int
1036 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1037 		      unsigned n)
1038 {
1039 	unsigned i;
1040 	unsigned count = 0;
1041 
1042 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1043 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1044 
1045 	if (n < nstats)
1046 		return nstats;
1047 
1048 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1049 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1050 
1051 		if (rxvq == NULL)
1052 			continue;
1053 
1054 		unsigned t;
1055 
1056 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1057 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1058 				rte_virtio_rxq_stat_strings[t].offset);
1059 			xstats[count].id = count;
1060 			count++;
1061 		}
1062 	}
1063 
1064 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1065 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1066 
1067 		if (txvq == NULL)
1068 			continue;
1069 
1070 		unsigned t;
1071 
1072 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1073 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
1074 				rte_virtio_txq_stat_strings[t].offset);
1075 			xstats[count].id = count;
1076 			count++;
1077 		}
1078 	}
1079 
1080 	return count;
1081 }
1082 
1083 static int
1084 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1085 {
1086 	virtio_update_stats(dev, stats);
1087 
1088 	return 0;
1089 }
1090 
1091 static int
1092 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1093 {
1094 	unsigned int i;
1095 
1096 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1097 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1098 		if (txvq == NULL)
1099 			continue;
1100 
1101 		txvq->stats.packets = 0;
1102 		txvq->stats.bytes = 0;
1103 		txvq->stats.multicast = 0;
1104 		txvq->stats.broadcast = 0;
1105 		memset(txvq->stats.size_bins, 0,
1106 		       sizeof(txvq->stats.size_bins[0]) * 8);
1107 	}
1108 
1109 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1110 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1111 		if (rxvq == NULL)
1112 			continue;
1113 
1114 		rxvq->stats.packets = 0;
1115 		rxvq->stats.bytes = 0;
1116 		rxvq->stats.errors = 0;
1117 		rxvq->stats.multicast = 0;
1118 		rxvq->stats.broadcast = 0;
1119 		memset(rxvq->stats.size_bins, 0,
1120 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1121 	}
1122 
1123 	return 0;
1124 }
1125 
1126 static void
1127 virtio_set_hwaddr(struct virtio_hw *hw)
1128 {
1129 	vtpci_write_dev_config(hw,
1130 			offsetof(struct virtio_net_config, mac),
1131 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1132 }
1133 
1134 static void
1135 virtio_get_hwaddr(struct virtio_hw *hw)
1136 {
1137 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1138 		vtpci_read_dev_config(hw,
1139 			offsetof(struct virtio_net_config, mac),
1140 			&hw->mac_addr, RTE_ETHER_ADDR_LEN);
1141 	} else {
1142 		rte_eth_random_addr(&hw->mac_addr[0]);
1143 		virtio_set_hwaddr(hw);
1144 	}
1145 }
1146 
1147 static int
1148 virtio_mac_table_set(struct virtio_hw *hw,
1149 		     const struct virtio_net_ctrl_mac *uc,
1150 		     const struct virtio_net_ctrl_mac *mc)
1151 {
1152 	struct virtio_pmd_ctrl ctrl;
1153 	int err, len[2];
1154 
1155 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1156 		PMD_DRV_LOG(INFO, "host does not support mac table");
1157 		return -1;
1158 	}
1159 
1160 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1161 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1162 
1163 	len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1164 	memcpy(ctrl.data, uc, len[0]);
1165 
1166 	len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1167 	memcpy(ctrl.data + len[0], mc, len[1]);
1168 
1169 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1170 	if (err != 0)
1171 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1172 	return err;
1173 }
1174 
1175 static int
1176 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1177 		    uint32_t index, uint32_t vmdq __rte_unused)
1178 {
1179 	struct virtio_hw *hw = dev->data->dev_private;
1180 	const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1181 	unsigned int i;
1182 	struct virtio_net_ctrl_mac *uc, *mc;
1183 
1184 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1185 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1186 		return -EINVAL;
1187 	}
1188 
1189 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1190 		sizeof(uc->entries));
1191 	uc->entries = 0;
1192 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1193 		sizeof(mc->entries));
1194 	mc->entries = 0;
1195 
1196 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1197 		const struct rte_ether_addr *addr
1198 			= (i == index) ? mac_addr : addrs + i;
1199 		struct virtio_net_ctrl_mac *tbl
1200 			= rte_is_multicast_ether_addr(addr) ? mc : uc;
1201 
1202 		memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1203 	}
1204 
1205 	return virtio_mac_table_set(hw, uc, mc);
1206 }
1207 
1208 static void
1209 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1210 {
1211 	struct virtio_hw *hw = dev->data->dev_private;
1212 	struct rte_ether_addr *addrs = dev->data->mac_addrs;
1213 	struct virtio_net_ctrl_mac *uc, *mc;
1214 	unsigned int i;
1215 
1216 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1217 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1218 		return;
1219 	}
1220 
1221 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1222 		sizeof(uc->entries));
1223 	uc->entries = 0;
1224 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1225 		sizeof(mc->entries));
1226 	mc->entries = 0;
1227 
1228 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1229 		struct virtio_net_ctrl_mac *tbl;
1230 
1231 		if (i == index || rte_is_zero_ether_addr(addrs + i))
1232 			continue;
1233 
1234 		tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1235 		memcpy(&tbl->macs[tbl->entries++], addrs + i,
1236 			RTE_ETHER_ADDR_LEN);
1237 	}
1238 
1239 	virtio_mac_table_set(hw, uc, mc);
1240 }
1241 
1242 static int
1243 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1244 {
1245 	struct virtio_hw *hw = dev->data->dev_private;
1246 
1247 	memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1248 
1249 	/* Use atomic update if available */
1250 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1251 		struct virtio_pmd_ctrl ctrl;
1252 		int len = RTE_ETHER_ADDR_LEN;
1253 
1254 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1255 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1256 
1257 		memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1258 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1259 	}
1260 
1261 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1262 		return -ENOTSUP;
1263 
1264 	virtio_set_hwaddr(hw);
1265 	return 0;
1266 }
1267 
1268 static int
1269 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1270 {
1271 	struct virtio_hw *hw = dev->data->dev_private;
1272 	struct virtio_pmd_ctrl ctrl;
1273 	int len;
1274 
1275 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1276 		return -ENOTSUP;
1277 
1278 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1279 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1280 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1281 	len = sizeof(vlan_id);
1282 
1283 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1284 }
1285 
1286 static int
1287 virtio_intr_unmask(struct rte_eth_dev *dev)
1288 {
1289 	struct virtio_hw *hw = dev->data->dev_private;
1290 
1291 	if (rte_intr_ack(dev->intr_handle) < 0)
1292 		return -1;
1293 
1294 	if (!hw->virtio_user_dev)
1295 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1296 
1297 	return 0;
1298 }
1299 
1300 static int
1301 virtio_intr_enable(struct rte_eth_dev *dev)
1302 {
1303 	struct virtio_hw *hw = dev->data->dev_private;
1304 
1305 	if (rte_intr_enable(dev->intr_handle) < 0)
1306 		return -1;
1307 
1308 	if (!hw->virtio_user_dev)
1309 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1310 
1311 	return 0;
1312 }
1313 
1314 static int
1315 virtio_intr_disable(struct rte_eth_dev *dev)
1316 {
1317 	struct virtio_hw *hw = dev->data->dev_private;
1318 
1319 	if (rte_intr_disable(dev->intr_handle) < 0)
1320 		return -1;
1321 
1322 	if (!hw->virtio_user_dev)
1323 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1324 
1325 	return 0;
1326 }
1327 
1328 static int
1329 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1330 {
1331 	uint64_t host_features;
1332 
1333 	/* Prepare guest_features: feature that driver wants to support */
1334 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1335 		req_features);
1336 
1337 	/* Read device(host) feature bits */
1338 	host_features = VTPCI_OPS(hw)->get_features(hw);
1339 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1340 		host_features);
1341 
1342 	/* If supported, ensure MTU value is valid before acknowledging it. */
1343 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1344 		struct virtio_net_config config;
1345 
1346 		vtpci_read_dev_config(hw,
1347 			offsetof(struct virtio_net_config, mtu),
1348 			&config.mtu, sizeof(config.mtu));
1349 
1350 		if (config.mtu < RTE_ETHER_MIN_MTU)
1351 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1352 	}
1353 
1354 	/*
1355 	 * Negotiate features: Subset of device feature bits are written back
1356 	 * guest feature bits.
1357 	 */
1358 	hw->guest_features = req_features;
1359 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1360 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1361 		hw->guest_features);
1362 
1363 	if (hw->modern) {
1364 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1365 			PMD_INIT_LOG(ERR,
1366 				"VIRTIO_F_VERSION_1 features is not enabled.");
1367 			return -1;
1368 		}
1369 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1370 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1371 			PMD_INIT_LOG(ERR,
1372 				"failed to set FEATURES_OK status!");
1373 			return -1;
1374 		}
1375 	}
1376 
1377 	hw->req_guest_features = req_features;
1378 
1379 	return 0;
1380 }
1381 
1382 int
1383 virtio_dev_pause(struct rte_eth_dev *dev)
1384 {
1385 	struct virtio_hw *hw = dev->data->dev_private;
1386 
1387 	rte_spinlock_lock(&hw->state_lock);
1388 
1389 	if (hw->started == 0) {
1390 		/* Device is just stopped. */
1391 		rte_spinlock_unlock(&hw->state_lock);
1392 		return -1;
1393 	}
1394 	hw->started = 0;
1395 	/*
1396 	 * Prevent the worker threads from touching queues to avoid contention,
1397 	 * 1 ms should be enough for the ongoing Tx function to finish.
1398 	 */
1399 	rte_delay_ms(1);
1400 	return 0;
1401 }
1402 
1403 /*
1404  * Recover hw state to let the worker threads continue.
1405  */
1406 void
1407 virtio_dev_resume(struct rte_eth_dev *dev)
1408 {
1409 	struct virtio_hw *hw = dev->data->dev_private;
1410 
1411 	hw->started = 1;
1412 	rte_spinlock_unlock(&hw->state_lock);
1413 }
1414 
1415 /*
1416  * Should be called only after device is paused.
1417  */
1418 int
1419 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1420 		int nb_pkts)
1421 {
1422 	struct virtio_hw *hw = dev->data->dev_private;
1423 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1424 	int ret;
1425 
1426 	hw->inject_pkts = tx_pkts;
1427 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1428 	hw->inject_pkts = NULL;
1429 
1430 	return ret;
1431 }
1432 
1433 static void
1434 virtio_notify_peers(struct rte_eth_dev *dev)
1435 {
1436 	struct virtio_hw *hw = dev->data->dev_private;
1437 	struct virtnet_rx *rxvq;
1438 	struct rte_mbuf *rarp_mbuf;
1439 
1440 	if (!dev->data->rx_queues)
1441 		return;
1442 
1443 	rxvq = dev->data->rx_queues[0];
1444 	if (!rxvq)
1445 		return;
1446 
1447 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1448 			(struct rte_ether_addr *)hw->mac_addr);
1449 	if (rarp_mbuf == NULL) {
1450 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1451 		return;
1452 	}
1453 
1454 	/* If virtio port just stopped, no need to send RARP */
1455 	if (virtio_dev_pause(dev) < 0) {
1456 		rte_pktmbuf_free(rarp_mbuf);
1457 		return;
1458 	}
1459 
1460 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1461 	virtio_dev_resume(dev);
1462 }
1463 
1464 static void
1465 virtio_ack_link_announce(struct rte_eth_dev *dev)
1466 {
1467 	struct virtio_hw *hw = dev->data->dev_private;
1468 	struct virtio_pmd_ctrl ctrl;
1469 
1470 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1471 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1472 
1473 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1474 }
1475 
1476 /*
1477  * Process virtio config changed interrupt. Call the callback
1478  * if link state changed, generate gratuitous RARP packet if
1479  * the status indicates an ANNOUNCE.
1480  */
1481 void
1482 virtio_interrupt_handler(void *param)
1483 {
1484 	struct rte_eth_dev *dev = param;
1485 	struct virtio_hw *hw = dev->data->dev_private;
1486 	uint8_t isr;
1487 	uint16_t status;
1488 
1489 	/* Read interrupt status which clears interrupt */
1490 	isr = vtpci_isr(hw);
1491 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1492 
1493 	if (virtio_intr_unmask(dev) < 0)
1494 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1495 
1496 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1497 		if (virtio_dev_link_update(dev, 0) == 0)
1498 			_rte_eth_dev_callback_process(dev,
1499 						      RTE_ETH_EVENT_INTR_LSC,
1500 						      NULL);
1501 
1502 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1503 			vtpci_read_dev_config(hw,
1504 				offsetof(struct virtio_net_config, status),
1505 				&status, sizeof(status));
1506 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1507 				virtio_notify_peers(dev);
1508 				if (hw->cvq)
1509 					virtio_ack_link_announce(dev);
1510 			}
1511 		}
1512 	}
1513 }
1514 
1515 /* set rx and tx handlers according to what is supported */
1516 static void
1517 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1518 {
1519 	struct virtio_hw *hw = eth_dev->data->dev_private;
1520 
1521 	eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1522 	if (vtpci_packed_queue(hw)) {
1523 		PMD_INIT_LOG(INFO,
1524 			"virtio: using packed ring %s Tx path on port %u",
1525 			hw->use_inorder_tx ? "inorder" : "standard",
1526 			eth_dev->data->port_id);
1527 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1528 	} else {
1529 		if (hw->use_inorder_tx) {
1530 			PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1531 				eth_dev->data->port_id);
1532 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1533 		} else {
1534 			PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1535 				eth_dev->data->port_id);
1536 			eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1537 		}
1538 	}
1539 
1540 	if (vtpci_packed_queue(hw)) {
1541 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1542 			PMD_INIT_LOG(INFO,
1543 				"virtio: using packed ring mergeable buffer Rx path on port %u",
1544 				eth_dev->data->port_id);
1545 			eth_dev->rx_pkt_burst =
1546 				&virtio_recv_mergeable_pkts_packed;
1547 		} else {
1548 			PMD_INIT_LOG(INFO,
1549 				"virtio: using packed ring standard Rx path on port %u",
1550 				eth_dev->data->port_id);
1551 			eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1552 		}
1553 	} else {
1554 		if (hw->use_simple_rx) {
1555 			PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1556 				eth_dev->data->port_id);
1557 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1558 		} else if (hw->use_inorder_rx) {
1559 			PMD_INIT_LOG(INFO,
1560 				"virtio: using inorder Rx path on port %u",
1561 				eth_dev->data->port_id);
1562 			eth_dev->rx_pkt_burst =	&virtio_recv_pkts_inorder;
1563 		} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1564 			PMD_INIT_LOG(INFO,
1565 				"virtio: using mergeable buffer Rx path on port %u",
1566 				eth_dev->data->port_id);
1567 			eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1568 		} else {
1569 			PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1570 				eth_dev->data->port_id);
1571 			eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1572 		}
1573 	}
1574 
1575 }
1576 
1577 /* Only support 1:1 queue/interrupt mapping so far.
1578  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1579  * interrupt vectors (<N+1).
1580  */
1581 static int
1582 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1583 {
1584 	uint32_t i;
1585 	struct virtio_hw *hw = dev->data->dev_private;
1586 
1587 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1588 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1589 		dev->intr_handle->intr_vec[i] = i + 1;
1590 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1591 						 VIRTIO_MSI_NO_VECTOR) {
1592 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1593 			return -EBUSY;
1594 		}
1595 	}
1596 
1597 	return 0;
1598 }
1599 
1600 static void
1601 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1602 {
1603 	uint32_t i;
1604 	struct virtio_hw *hw = dev->data->dev_private;
1605 
1606 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1607 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1608 		VTPCI_OPS(hw)->set_queue_irq(hw,
1609 					     hw->vqs[i * VTNET_CQ],
1610 					     VIRTIO_MSI_NO_VECTOR);
1611 }
1612 
1613 static int
1614 virtio_configure_intr(struct rte_eth_dev *dev)
1615 {
1616 	struct virtio_hw *hw = dev->data->dev_private;
1617 
1618 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1619 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1620 		return -ENOTSUP;
1621 	}
1622 
1623 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1624 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1625 		return -1;
1626 	}
1627 
1628 	if (!dev->intr_handle->intr_vec) {
1629 		dev->intr_handle->intr_vec =
1630 			rte_zmalloc("intr_vec",
1631 				    hw->max_queue_pairs * sizeof(int), 0);
1632 		if (!dev->intr_handle->intr_vec) {
1633 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1634 				     hw->max_queue_pairs);
1635 			return -ENOMEM;
1636 		}
1637 	}
1638 
1639 	/* Re-register callback to update max_intr */
1640 	rte_intr_callback_unregister(dev->intr_handle,
1641 				     virtio_interrupt_handler,
1642 				     dev);
1643 	rte_intr_callback_register(dev->intr_handle,
1644 				   virtio_interrupt_handler,
1645 				   dev);
1646 
1647 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1648 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1649 	 * And for legacy devices, this should be done before queue/vec binding
1650 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1651 	 * (22) will be ignored.
1652 	 */
1653 	if (virtio_intr_enable(dev) < 0) {
1654 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1655 		return -1;
1656 	}
1657 
1658 	if (virtio_queues_bind_intr(dev) < 0) {
1659 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1660 		return -1;
1661 	}
1662 
1663 	return 0;
1664 }
1665 #define SPEED_UNKNOWN    0xffffffff
1666 #define DUPLEX_UNKNOWN   0xff
1667 /* reset device and renegotiate features if needed */
1668 static int
1669 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1670 {
1671 	struct virtio_hw *hw = eth_dev->data->dev_private;
1672 	struct virtio_net_config *config;
1673 	struct virtio_net_config local_config;
1674 	struct rte_pci_device *pci_dev = NULL;
1675 	int ret;
1676 
1677 	/* Reset the device although not necessary at startup */
1678 	vtpci_reset(hw);
1679 
1680 	if (hw->vqs) {
1681 		virtio_dev_free_mbufs(eth_dev);
1682 		virtio_free_queues(hw);
1683 	}
1684 
1685 	/* Tell the host we've noticed this device. */
1686 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1687 
1688 	/* Tell the host we've known how to drive the device. */
1689 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1690 	if (virtio_negotiate_features(hw, req_features) < 0)
1691 		return -1;
1692 
1693 	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1694 
1695 	if (!hw->virtio_user_dev)
1696 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1697 
1698 	/* If host does not support both status and MSI-X then disable LSC */
1699 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1700 	    hw->use_msix != VIRTIO_MSIX_NONE)
1701 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1702 	else
1703 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1704 
1705 	/* Setting up rx_header size for the device */
1706 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1707 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1708 	    vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1709 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1710 	else
1711 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1712 
1713 	/* Copy the permanent MAC address to: virtio_hw */
1714 	virtio_get_hwaddr(hw);
1715 	rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1716 			&eth_dev->data->mac_addrs[0]);
1717 	PMD_INIT_LOG(DEBUG,
1718 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1719 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1720 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1721 
1722 	if (hw->speed == SPEED_UNKNOWN) {
1723 		if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1724 			config = &local_config;
1725 			vtpci_read_dev_config(hw,
1726 				offsetof(struct virtio_net_config, speed),
1727 				&config->speed, sizeof(config->speed));
1728 			vtpci_read_dev_config(hw,
1729 				offsetof(struct virtio_net_config, duplex),
1730 				&config->duplex, sizeof(config->duplex));
1731 			hw->speed = config->speed;
1732 			hw->duplex = config->duplex;
1733 		}
1734 	}
1735 	if (hw->speed == SPEED_UNKNOWN)
1736 		hw->speed = ETH_SPEED_NUM_10G;
1737 	if (hw->duplex == DUPLEX_UNKNOWN)
1738 		hw->duplex = ETH_LINK_FULL_DUPLEX;
1739 	PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1740 		hw->speed, hw->duplex);
1741 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1742 		config = &local_config;
1743 
1744 		vtpci_read_dev_config(hw,
1745 			offsetof(struct virtio_net_config, mac),
1746 			&config->mac, sizeof(config->mac));
1747 
1748 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1749 			vtpci_read_dev_config(hw,
1750 				offsetof(struct virtio_net_config, status),
1751 				&config->status, sizeof(config->status));
1752 		} else {
1753 			PMD_INIT_LOG(DEBUG,
1754 				     "VIRTIO_NET_F_STATUS is not supported");
1755 			config->status = 0;
1756 		}
1757 
1758 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1759 			vtpci_read_dev_config(hw,
1760 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1761 				&config->max_virtqueue_pairs,
1762 				sizeof(config->max_virtqueue_pairs));
1763 		} else {
1764 			PMD_INIT_LOG(DEBUG,
1765 				     "VIRTIO_NET_F_MQ is not supported");
1766 			config->max_virtqueue_pairs = 1;
1767 		}
1768 
1769 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1770 
1771 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1772 			vtpci_read_dev_config(hw,
1773 				offsetof(struct virtio_net_config, mtu),
1774 				&config->mtu,
1775 				sizeof(config->mtu));
1776 
1777 			/*
1778 			 * MTU value has already been checked at negotiation
1779 			 * time, but check again in case it has changed since
1780 			 * then, which should not happen.
1781 			 */
1782 			if (config->mtu < RTE_ETHER_MIN_MTU) {
1783 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1784 						config->mtu);
1785 				return -1;
1786 			}
1787 
1788 			hw->max_mtu = config->mtu;
1789 			/* Set initial MTU to maximum one supported by vhost */
1790 			eth_dev->data->mtu = config->mtu;
1791 
1792 		} else {
1793 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1794 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1795 		}
1796 
1797 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1798 				config->max_virtqueue_pairs);
1799 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1800 		PMD_INIT_LOG(DEBUG,
1801 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1802 				config->mac[0], config->mac[1],
1803 				config->mac[2], config->mac[3],
1804 				config->mac[4], config->mac[5]);
1805 	} else {
1806 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1807 		hw->max_queue_pairs = 1;
1808 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1809 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1810 	}
1811 
1812 	ret = virtio_alloc_queues(eth_dev);
1813 	if (ret < 0)
1814 		return ret;
1815 
1816 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1817 		if (virtio_configure_intr(eth_dev) < 0) {
1818 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1819 			virtio_free_queues(hw);
1820 			return -1;
1821 		}
1822 	}
1823 
1824 	vtpci_reinit_complete(hw);
1825 
1826 	if (pci_dev)
1827 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1828 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1829 			pci_dev->id.device_id);
1830 
1831 	return 0;
1832 }
1833 
1834 /*
1835  * Remap the PCI device again (IO port map for legacy device and
1836  * memory map for modern device), so that the secondary process
1837  * could have the PCI initiated correctly.
1838  */
1839 static int
1840 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1841 {
1842 	if (hw->modern) {
1843 		/*
1844 		 * We don't have to re-parse the PCI config space, since
1845 		 * rte_pci_map_device() makes sure the mapped address
1846 		 * in secondary process would equal to the one mapped in
1847 		 * the primary process: error will be returned if that
1848 		 * requirement is not met.
1849 		 *
1850 		 * That said, we could simply reuse all cap pointers
1851 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1852 		 * primary process, which is stored in shared memory.
1853 		 */
1854 		if (rte_pci_map_device(pci_dev)) {
1855 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1856 			return -1;
1857 		}
1858 	} else {
1859 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1860 			return -1;
1861 	}
1862 
1863 	return 0;
1864 }
1865 
1866 static void
1867 virtio_set_vtpci_ops(struct virtio_hw *hw)
1868 {
1869 #ifdef RTE_VIRTIO_USER
1870 	if (hw->virtio_user_dev)
1871 		VTPCI_OPS(hw) = &virtio_user_ops;
1872 	else
1873 #endif
1874 	if (hw->modern)
1875 		VTPCI_OPS(hw) = &modern_ops;
1876 	else
1877 		VTPCI_OPS(hw) = &legacy_ops;
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 = SPEED_UNKNOWN;
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 		if (!hw->virtio_user_dev) {
1904 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1905 			if (ret)
1906 				return ret;
1907 		}
1908 
1909 		virtio_set_vtpci_ops(hw);
1910 		set_rxtx_funcs(eth_dev);
1911 
1912 		return 0;
1913 	}
1914 	ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
1915 		 NULL, &speed);
1916 	if (ret < 0)
1917 		return ret;
1918 	hw->speed = speed;
1919 	/*
1920 	 * Pass the information to the rte_eth_dev_close() that it should also
1921 	 * release the private port resources.
1922 	 */
1923 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1924 
1925 	/* Allocate memory for storing MAC addresses */
1926 	eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1927 				VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1928 	if (eth_dev->data->mac_addrs == NULL) {
1929 		PMD_INIT_LOG(ERR,
1930 			"Failed to allocate %d bytes needed to store MAC addresses",
1931 			VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1932 		return -ENOMEM;
1933 	}
1934 
1935 	hw->port_id = eth_dev->data->port_id;
1936 	/* For virtio_user case the hw->virtio_user_dev is populated by
1937 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1938 	 */
1939 	if (!hw->virtio_user_dev) {
1940 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1941 		if (ret)
1942 			goto err_vtpci_init;
1943 	}
1944 
1945 	rte_spinlock_init(&hw->state_lock);
1946 
1947 	/* reset device and negotiate default features */
1948 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1949 	if (ret < 0)
1950 		goto err_virtio_init;
1951 
1952 	hw->opened = true;
1953 
1954 	return 0;
1955 
1956 err_virtio_init:
1957 	if (!hw->virtio_user_dev) {
1958 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1959 		if (!hw->modern)
1960 			rte_pci_ioport_unmap(VTPCI_IO(hw));
1961 	}
1962 err_vtpci_init:
1963 	rte_free(eth_dev->data->mac_addrs);
1964 	eth_dev->data->mac_addrs = NULL;
1965 	return ret;
1966 }
1967 
1968 static int
1969 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1970 {
1971 	PMD_INIT_FUNC_TRACE();
1972 
1973 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1974 		return 0;
1975 
1976 	virtio_dev_stop(eth_dev);
1977 	virtio_dev_close(eth_dev);
1978 
1979 	eth_dev->dev_ops = NULL;
1980 	eth_dev->tx_pkt_burst = NULL;
1981 	eth_dev->rx_pkt_burst = NULL;
1982 
1983 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1984 
1985 	return 0;
1986 }
1987 
1988 
1989 static int vdpa_check_handler(__rte_unused const char *key,
1990 		const char *value, void *ret_val)
1991 {
1992 	if (strcmp(value, "1") == 0)
1993 		*(int *)ret_val = 1;
1994 	else
1995 		*(int *)ret_val = 0;
1996 
1997 	return 0;
1998 }
1999 
2000 
2001 static uint32_t
2002 virtio_dev_speed_capa_get(uint32_t speed)
2003 {
2004 	switch (speed) {
2005 	case ETH_SPEED_NUM_10G:
2006 		return ETH_LINK_SPEED_10G;
2007 	case ETH_SPEED_NUM_20G:
2008 		return ETH_LINK_SPEED_20G;
2009 	case ETH_SPEED_NUM_25G:
2010 		return ETH_LINK_SPEED_25G;
2011 	case ETH_SPEED_NUM_40G:
2012 		return ETH_LINK_SPEED_40G;
2013 	case ETH_SPEED_NUM_50G:
2014 		return ETH_LINK_SPEED_50G;
2015 	case ETH_SPEED_NUM_56G:
2016 		return ETH_LINK_SPEED_56G;
2017 	case ETH_SPEED_NUM_100G:
2018 		return ETH_LINK_SPEED_100G;
2019 	default:
2020 		return 0;
2021 	}
2022 }
2023 
2024 
2025 #define VIRTIO_ARG_SPEED      "speed"
2026 #define VIRTIO_ARG_VDPA       "vdpa"
2027 
2028 
2029 static int
2030 link_speed_handler(const char *key __rte_unused,
2031 		const char *value, void *ret_val)
2032 {
2033 	uint32_t val;
2034 	if (!value || !ret_val)
2035 		return -EINVAL;
2036 	val = strtoul(value, NULL, 0);
2037 	/* validate input */
2038 	if (virtio_dev_speed_capa_get(val) == 0)
2039 		return -EINVAL;
2040 	*(uint32_t *)ret_val = val;
2041 
2042 	return 0;
2043 }
2044 
2045 
2046 static int
2047 virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
2048 	uint32_t *speed)
2049 {
2050 	struct rte_kvargs *kvlist;
2051 	int ret = 0;
2052 
2053 	if (devargs == NULL)
2054 		return 0;
2055 
2056 	kvlist = rte_kvargs_parse(devargs->args, NULL);
2057 	if (kvlist == NULL) {
2058 		PMD_INIT_LOG(ERR, "error when parsing param");
2059 		return 0;
2060 	}
2061 	if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
2062 		/* vdpa mode selected when there's a key-value pair:
2063 		 * vdpa=1
2064 		 */
2065 		ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
2066 				vdpa_check_handler, vdpa);
2067 		if (ret < 0) {
2068 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2069 				VIRTIO_ARG_VDPA);
2070 			goto exit;
2071 		}
2072 	}
2073 	if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2074 		ret = rte_kvargs_process(kvlist,
2075 					VIRTIO_ARG_SPEED,
2076 					link_speed_handler, speed);
2077 		if (ret < 0) {
2078 			PMD_INIT_LOG(ERR, "Failed to parse %s",
2079 					VIRTIO_ARG_SPEED);
2080 			goto exit;
2081 		}
2082 	}
2083 
2084 exit:
2085 	rte_kvargs_free(kvlist);
2086 	return ret;
2087 }
2088 
2089 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2090 	struct rte_pci_device *pci_dev)
2091 {
2092 	int vdpa = 0;
2093 	int ret = 0;
2094 
2095 	ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL);
2096 	if (ret < 0) {
2097 		PMD_INIT_LOG(ERR, "devargs parsing is failed");
2098 		return ret;
2099 	}
2100 	/* virtio pmd skips probe if device needs to work in vdpa mode */
2101 	if (vdpa == 1)
2102 		return 1;
2103 
2104 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2105 		eth_virtio_dev_init);
2106 }
2107 
2108 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2109 {
2110 	int ret;
2111 
2112 	ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2113 	/* Port has already been released by close. */
2114 	if (ret == -ENODEV)
2115 		ret = 0;
2116 	return ret;
2117 }
2118 
2119 static struct rte_pci_driver rte_virtio_pmd = {
2120 	.driver = {
2121 		.name = "net_virtio",
2122 	},
2123 	.id_table = pci_id_virtio_map,
2124 	.drv_flags = 0,
2125 	.probe = eth_virtio_pci_probe,
2126 	.remove = eth_virtio_pci_remove,
2127 };
2128 
2129 RTE_INIT(rte_virtio_pmd_init)
2130 {
2131 	rte_eal_iopl_init();
2132 	rte_pci_register(&rte_virtio_pmd);
2133 }
2134 
2135 static bool
2136 rx_offload_enabled(struct virtio_hw *hw)
2137 {
2138 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2139 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2140 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2141 }
2142 
2143 static bool
2144 tx_offload_enabled(struct virtio_hw *hw)
2145 {
2146 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2147 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2148 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2149 }
2150 
2151 /*
2152  * Configure virtio device
2153  * It returns 0 on success.
2154  */
2155 static int
2156 virtio_dev_configure(struct rte_eth_dev *dev)
2157 {
2158 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2159 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2160 	struct virtio_hw *hw = dev->data->dev_private;
2161 	uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2162 		hw->vtnet_hdr_size;
2163 	uint64_t rx_offloads = rxmode->offloads;
2164 	uint64_t tx_offloads = txmode->offloads;
2165 	uint64_t req_features;
2166 	int ret;
2167 
2168 	PMD_INIT_LOG(DEBUG, "configure");
2169 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2170 
2171 	if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2172 		PMD_DRV_LOG(ERR,
2173 			"Unsupported Rx multi queue mode %d",
2174 			rxmode->mq_mode);
2175 		return -EINVAL;
2176 	}
2177 
2178 	if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2179 		PMD_DRV_LOG(ERR,
2180 			"Unsupported Tx multi queue mode %d",
2181 			txmode->mq_mode);
2182 		return -EINVAL;
2183 	}
2184 
2185 	if (dev->data->dev_conf.intr_conf.rxq) {
2186 		ret = virtio_init_device(dev, hw->req_guest_features);
2187 		if (ret < 0)
2188 			return ret;
2189 	}
2190 
2191 	if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2192 		req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2193 
2194 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2195 			   DEV_RX_OFFLOAD_TCP_CKSUM))
2196 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2197 
2198 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2199 		req_features |=
2200 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2201 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2202 
2203 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2204 			   DEV_TX_OFFLOAD_TCP_CKSUM))
2205 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2206 
2207 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2208 		req_features |=
2209 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
2210 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
2211 
2212 	/* if request features changed, reinit the device */
2213 	if (req_features != hw->req_guest_features) {
2214 		ret = virtio_init_device(dev, req_features);
2215 		if (ret < 0)
2216 			return ret;
2217 	}
2218 
2219 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2220 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2221 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2222 		PMD_DRV_LOG(ERR,
2223 			"rx checksum not available on this host");
2224 		return -ENOTSUP;
2225 	}
2226 
2227 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2228 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2229 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2230 		PMD_DRV_LOG(ERR,
2231 			"Large Receive Offload not available on this host");
2232 		return -ENOTSUP;
2233 	}
2234 
2235 	/* start control queue */
2236 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2237 		virtio_dev_cq_start(dev);
2238 
2239 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2240 		hw->vlan_strip = 1;
2241 
2242 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2243 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2244 		PMD_DRV_LOG(ERR,
2245 			    "vlan filtering not available on this host");
2246 		return -ENOTSUP;
2247 	}
2248 
2249 	hw->has_tx_offload = tx_offload_enabled(hw);
2250 	hw->has_rx_offload = rx_offload_enabled(hw);
2251 
2252 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2253 		/* Enable vector (0) for Link State Intrerrupt */
2254 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2255 				VIRTIO_MSI_NO_VECTOR) {
2256 			PMD_DRV_LOG(ERR, "failed to set config vector");
2257 			return -EBUSY;
2258 		}
2259 
2260 	hw->use_simple_rx = 1;
2261 
2262 	if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2263 		hw->use_inorder_tx = 1;
2264 		hw->use_inorder_rx = 1;
2265 		hw->use_simple_rx = 0;
2266 	}
2267 
2268 	if (vtpci_packed_queue(hw)) {
2269 		hw->use_simple_rx = 0;
2270 		hw->use_inorder_rx = 0;
2271 	}
2272 
2273 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2274 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2275 		hw->use_simple_rx = 0;
2276 	}
2277 #endif
2278 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2279 		 hw->use_simple_rx = 0;
2280 	}
2281 
2282 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2283 			   DEV_RX_OFFLOAD_TCP_CKSUM |
2284 			   DEV_RX_OFFLOAD_TCP_LRO |
2285 			   DEV_RX_OFFLOAD_VLAN_STRIP))
2286 		hw->use_simple_rx = 0;
2287 
2288 	return 0;
2289 }
2290 
2291 
2292 static int
2293 virtio_dev_start(struct rte_eth_dev *dev)
2294 {
2295 	uint16_t nb_queues, i;
2296 	struct virtnet_rx *rxvq;
2297 	struct virtnet_tx *txvq __rte_unused;
2298 	struct virtio_hw *hw = dev->data->dev_private;
2299 	int ret;
2300 
2301 	/* Finish the initialization of the queues */
2302 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2303 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
2304 		if (ret < 0)
2305 			return ret;
2306 	}
2307 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2308 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
2309 		if (ret < 0)
2310 			return ret;
2311 	}
2312 
2313 	/* check if lsc interrupt feature is enabled */
2314 	if (dev->data->dev_conf.intr_conf.lsc) {
2315 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2316 			PMD_DRV_LOG(ERR, "link status not supported by host");
2317 			return -ENOTSUP;
2318 		}
2319 	}
2320 
2321 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2322 	 * in device configure, but it could be unmapped  when device is
2323 	 * stopped.
2324 	 */
2325 	if (dev->data->dev_conf.intr_conf.lsc ||
2326 	    dev->data->dev_conf.intr_conf.rxq) {
2327 		virtio_intr_disable(dev);
2328 
2329 		/* Setup interrupt callback  */
2330 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2331 			rte_intr_callback_register(dev->intr_handle,
2332 						   virtio_interrupt_handler,
2333 						   dev);
2334 
2335 		if (virtio_intr_enable(dev) < 0) {
2336 			PMD_DRV_LOG(ERR, "interrupt enable failed");
2337 			return -EIO;
2338 		}
2339 	}
2340 
2341 	/*Notify the backend
2342 	 *Otherwise the tap backend might already stop its queue due to fullness.
2343 	 *vhost backend will have no chance to be waked up
2344 	 */
2345 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2346 	if (hw->max_queue_pairs > 1) {
2347 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2348 			return -EINVAL;
2349 	}
2350 
2351 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2352 
2353 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2354 		rxvq = dev->data->rx_queues[i];
2355 		/* Flush the old packets */
2356 		virtqueue_rxvq_flush(rxvq->vq);
2357 		virtqueue_notify(rxvq->vq);
2358 	}
2359 
2360 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2361 		txvq = dev->data->tx_queues[i];
2362 		virtqueue_notify(txvq->vq);
2363 	}
2364 
2365 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2366 
2367 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2368 		rxvq = dev->data->rx_queues[i];
2369 		VIRTQUEUE_DUMP(rxvq->vq);
2370 	}
2371 
2372 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2373 		txvq = dev->data->tx_queues[i];
2374 		VIRTQUEUE_DUMP(txvq->vq);
2375 	}
2376 
2377 	set_rxtx_funcs(dev);
2378 	hw->started = true;
2379 
2380 	/* Initialize Link state */
2381 	virtio_dev_link_update(dev, 0);
2382 
2383 	return 0;
2384 }
2385 
2386 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2387 {
2388 	struct virtio_hw *hw = dev->data->dev_private;
2389 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2390 	const char *type __rte_unused;
2391 	unsigned int i, mbuf_num = 0;
2392 	struct virtqueue *vq;
2393 	struct rte_mbuf *buf;
2394 	int queue_type;
2395 
2396 	if (hw->vqs == NULL)
2397 		return;
2398 
2399 	for (i = 0; i < nr_vq; i++) {
2400 		vq = hw->vqs[i];
2401 		if (!vq)
2402 			continue;
2403 
2404 		queue_type = virtio_get_queue_type(hw, i);
2405 		if (queue_type == VTNET_RQ)
2406 			type = "rxq";
2407 		else if (queue_type == VTNET_TQ)
2408 			type = "txq";
2409 		else
2410 			continue;
2411 
2412 		PMD_INIT_LOG(DEBUG,
2413 			"Before freeing %s[%d] used and unused buf",
2414 			type, i);
2415 		VIRTQUEUE_DUMP(vq);
2416 
2417 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2418 			rte_pktmbuf_free(buf);
2419 			mbuf_num++;
2420 		}
2421 
2422 		PMD_INIT_LOG(DEBUG,
2423 			"After freeing %s[%d] used and unused buf",
2424 			type, i);
2425 		VIRTQUEUE_DUMP(vq);
2426 	}
2427 
2428 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2429 }
2430 
2431 /*
2432  * Stop device: disable interrupt and mark link down
2433  */
2434 static void
2435 virtio_dev_stop(struct rte_eth_dev *dev)
2436 {
2437 	struct virtio_hw *hw = dev->data->dev_private;
2438 	struct rte_eth_link link;
2439 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2440 
2441 	PMD_INIT_LOG(DEBUG, "stop");
2442 
2443 	rte_spinlock_lock(&hw->state_lock);
2444 	if (!hw->started)
2445 		goto out_unlock;
2446 	hw->started = false;
2447 
2448 	if (intr_conf->lsc || intr_conf->rxq) {
2449 		virtio_intr_disable(dev);
2450 
2451 		/* Reset interrupt callback  */
2452 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2453 			rte_intr_callback_unregister(dev->intr_handle,
2454 						     virtio_interrupt_handler,
2455 						     dev);
2456 		}
2457 	}
2458 
2459 	memset(&link, 0, sizeof(link));
2460 	rte_eth_linkstatus_set(dev, &link);
2461 out_unlock:
2462 	rte_spinlock_unlock(&hw->state_lock);
2463 }
2464 
2465 static int
2466 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2467 {
2468 	struct rte_eth_link link;
2469 	uint16_t status;
2470 	struct virtio_hw *hw = dev->data->dev_private;
2471 
2472 	memset(&link, 0, sizeof(link));
2473 	link.link_duplex = hw->duplex;
2474 	link.link_speed  = hw->speed;
2475 	link.link_autoneg = ETH_LINK_AUTONEG;
2476 
2477 	if (!hw->started) {
2478 		link.link_status = ETH_LINK_DOWN;
2479 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2480 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2481 		vtpci_read_dev_config(hw,
2482 				offsetof(struct virtio_net_config, status),
2483 				&status, sizeof(status));
2484 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2485 			link.link_status = ETH_LINK_DOWN;
2486 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2487 				     dev->data->port_id);
2488 		} else {
2489 			link.link_status = ETH_LINK_UP;
2490 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2491 				     dev->data->port_id);
2492 		}
2493 	} else {
2494 		link.link_status = ETH_LINK_UP;
2495 	}
2496 
2497 	return rte_eth_linkstatus_set(dev, &link);
2498 }
2499 
2500 static int
2501 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2502 {
2503 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2504 	struct virtio_hw *hw = dev->data->dev_private;
2505 	uint64_t offloads = rxmode->offloads;
2506 
2507 	if (mask & ETH_VLAN_FILTER_MASK) {
2508 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2509 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2510 
2511 			PMD_DRV_LOG(NOTICE,
2512 				"vlan filtering not available on this host");
2513 
2514 			return -ENOTSUP;
2515 		}
2516 	}
2517 
2518 	if (mask & ETH_VLAN_STRIP_MASK)
2519 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2520 
2521 	return 0;
2522 }
2523 
2524 static int
2525 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2526 {
2527 	uint64_t tso_mask, host_features;
2528 	struct virtio_hw *hw = dev->data->dev_private;
2529 	dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2530 
2531 	dev_info->max_rx_queues =
2532 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2533 	dev_info->max_tx_queues =
2534 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2535 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2536 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2537 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2538 
2539 	host_features = VTPCI_OPS(hw)->get_features(hw);
2540 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2541 	dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2542 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2543 		dev_info->rx_offload_capa |=
2544 			DEV_RX_OFFLOAD_TCP_CKSUM |
2545 			DEV_RX_OFFLOAD_UDP_CKSUM;
2546 	}
2547 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2548 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2549 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2550 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2551 	if ((host_features & tso_mask) == tso_mask)
2552 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2553 
2554 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2555 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2556 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2557 		dev_info->tx_offload_capa |=
2558 			DEV_TX_OFFLOAD_UDP_CKSUM |
2559 			DEV_TX_OFFLOAD_TCP_CKSUM;
2560 	}
2561 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2562 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2563 	if ((host_features & tso_mask) == tso_mask)
2564 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2565 
2566 	return 0;
2567 }
2568 
2569 /*
2570  * It enables testpmd to collect per queue stats.
2571  */
2572 static int
2573 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2574 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2575 __rte_unused uint8_t is_rx)
2576 {
2577 	return 0;
2578 }
2579 
2580 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2581 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2582 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2583 
2584 RTE_INIT(virtio_init_log)
2585 {
2586 	virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2587 	if (virtio_logtype_init >= 0)
2588 		rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2589 	virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2590 	if (virtio_logtype_driver >= 0)
2591 		rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2592 }
2593