xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 89f0711f9ddfb5822da9d34f384b92f72a61c4dc)
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_atomic.h>
18 #include <rte_branch_prediction.h>
19 #include <rte_pci.h>
20 #include <rte_bus_pci.h>
21 #include <rte_ether.h>
22 #include <rte_ip.h>
23 #include <rte_arp.h>
24 #include <rte_common.h>
25 #include <rte_errno.h>
26 #include <rte_cpuflags.h>
27 
28 #include <rte_memory.h>
29 #include <rte_eal.h>
30 #include <rte_dev.h>
31 #include <rte_cycles.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 
39 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
40 static int  virtio_dev_configure(struct rte_eth_dev *dev);
41 static int  virtio_dev_start(struct rte_eth_dev *dev);
42 static void virtio_dev_stop(struct rte_eth_dev *dev);
43 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
44 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
45 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
46 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
47 static void virtio_dev_info_get(struct rte_eth_dev *dev,
48 				struct rte_eth_dev_info *dev_info);
49 static int virtio_dev_link_update(struct rte_eth_dev *dev,
50 	int wait_to_complete);
51 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
52 
53 static void virtio_set_hwaddr(struct virtio_hw *hw);
54 static void virtio_get_hwaddr(struct virtio_hw *hw);
55 
56 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
57 				 struct rte_eth_stats *stats);
58 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
59 				 struct rte_eth_xstat *xstats, unsigned n);
60 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
61 				       struct rte_eth_xstat_name *xstats_names,
62 				       unsigned limit);
63 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
64 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
65 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
66 				uint16_t vlan_id, int on);
67 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
68 				struct ether_addr *mac_addr,
69 				uint32_t index, uint32_t vmdq);
70 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
71 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
72 				struct ether_addr *mac_addr);
73 
74 static int virtio_intr_enable(struct rte_eth_dev *dev);
75 static int virtio_intr_disable(struct rte_eth_dev *dev);
76 
77 static int virtio_dev_queue_stats_mapping_set(
78 	struct rte_eth_dev *eth_dev,
79 	uint16_t queue_id,
80 	uint8_t stat_idx,
81 	uint8_t is_rx);
82 
83 int virtio_logtype_init;
84 int virtio_logtype_driver;
85 
86 static void virtio_notify_peers(struct rte_eth_dev *dev);
87 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
88 
89 /*
90  * The set of PCI devices this driver supports
91  */
92 static const struct rte_pci_id pci_id_virtio_map[] = {
93 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
94 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
95 	{ .vendor_id = 0, /* sentinel */ },
96 };
97 
98 struct rte_virtio_xstats_name_off {
99 	char name[RTE_ETH_XSTATS_NAME_SIZE];
100 	unsigned offset;
101 };
102 
103 /* [rt]x_qX_ is prepended to the name string here */
104 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
105 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
106 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
107 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
108 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
109 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
110 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
111 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
112 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
113 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
114 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
115 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
116 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
117 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
118 };
119 
120 /* [rt]x_qX_ is prepended to the name string here */
121 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
122 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
123 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
124 	{"errors",                 offsetof(struct virtnet_tx, stats.errors)},
125 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
126 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
127 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
128 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
129 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
130 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
131 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
132 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
133 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
134 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
135 };
136 
137 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
138 			    sizeof(rte_virtio_rxq_stat_strings[0]))
139 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
140 			    sizeof(rte_virtio_txq_stat_strings[0]))
141 
142 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
143 
144 static int
145 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
146 		int *dlen, int pkt_num)
147 {
148 	uint32_t head, i;
149 	int k, sum = 0;
150 	virtio_net_ctrl_ack status = ~0;
151 	struct virtio_pmd_ctrl *result;
152 	struct virtqueue *vq;
153 
154 	ctrl->status = status;
155 
156 	if (!cvq || !cvq->vq) {
157 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
158 		return -1;
159 	}
160 
161 	rte_spinlock_lock(&cvq->lock);
162 	vq = cvq->vq;
163 	head = vq->vq_desc_head_idx;
164 
165 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
166 		"vq->hw->cvq = %p vq = %p",
167 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
168 
169 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
170 		rte_spinlock_unlock(&cvq->lock);
171 		return -1;
172 	}
173 
174 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
175 		sizeof(struct virtio_pmd_ctrl));
176 
177 	/*
178 	 * Format is enforced in qemu code:
179 	 * One TX packet for header;
180 	 * At least one TX packet per argument;
181 	 * One RX packet for ACK.
182 	 */
183 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
184 	vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
185 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
186 	vq->vq_free_cnt--;
187 	i = vq->vq_ring.desc[head].next;
188 
189 	for (k = 0; k < pkt_num; k++) {
190 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
191 		vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
192 			+ sizeof(struct virtio_net_ctrl_hdr)
193 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
194 		vq->vq_ring.desc[i].len = dlen[k];
195 		sum += dlen[k];
196 		vq->vq_free_cnt--;
197 		i = vq->vq_ring.desc[i].next;
198 	}
199 
200 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
201 	vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
202 			+ sizeof(struct virtio_net_ctrl_hdr);
203 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
204 	vq->vq_free_cnt--;
205 
206 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
207 
208 	vq_update_avail_ring(vq, head);
209 	vq_update_avail_idx(vq);
210 
211 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
212 
213 	virtqueue_notify(vq);
214 
215 	rte_rmb();
216 	while (VIRTQUEUE_NUSED(vq) == 0) {
217 		rte_rmb();
218 		usleep(100);
219 	}
220 
221 	while (VIRTQUEUE_NUSED(vq)) {
222 		uint32_t idx, desc_idx, used_idx;
223 		struct vring_used_elem *uep;
224 
225 		used_idx = (uint32_t)(vq->vq_used_cons_idx
226 				& (vq->vq_nentries - 1));
227 		uep = &vq->vq_ring.used->ring[used_idx];
228 		idx = (uint32_t) uep->id;
229 		desc_idx = idx;
230 
231 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
232 			desc_idx = vq->vq_ring.desc[desc_idx].next;
233 			vq->vq_free_cnt++;
234 		}
235 
236 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
237 		vq->vq_desc_head_idx = idx;
238 
239 		vq->vq_used_cons_idx++;
240 		vq->vq_free_cnt++;
241 	}
242 
243 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
244 			vq->vq_free_cnt, vq->vq_desc_head_idx);
245 
246 	result = cvq->virtio_net_hdr_mz->addr;
247 
248 	rte_spinlock_unlock(&cvq->lock);
249 	return result->status;
250 }
251 
252 static int
253 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
254 {
255 	struct virtio_hw *hw = dev->data->dev_private;
256 	struct virtio_pmd_ctrl ctrl;
257 	int dlen[1];
258 	int ret;
259 
260 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
261 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
262 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
263 
264 	dlen[0] = sizeof(uint16_t);
265 
266 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
267 	if (ret) {
268 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
269 			  "failed, this is too late now...");
270 		return -EINVAL;
271 	}
272 
273 	return 0;
274 }
275 
276 static void
277 virtio_dev_queue_release(void *queue __rte_unused)
278 {
279 	/* do nothing */
280 }
281 
282 static uint16_t
283 virtio_get_nr_vq(struct virtio_hw *hw)
284 {
285 	uint16_t nr_vq = hw->max_queue_pairs * 2;
286 
287 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
288 		nr_vq += 1;
289 
290 	return nr_vq;
291 }
292 
293 static void
294 virtio_init_vring(struct virtqueue *vq)
295 {
296 	int size = vq->vq_nentries;
297 	struct vring *vr = &vq->vq_ring;
298 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
299 
300 	PMD_INIT_FUNC_TRACE();
301 
302 	/*
303 	 * Reinitialise since virtio port might have been stopped and restarted
304 	 */
305 	memset(ring_mem, 0, vq->vq_ring_size);
306 	vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
307 	vq->vq_used_cons_idx = 0;
308 	vq->vq_desc_head_idx = 0;
309 	vq->vq_avail_idx = 0;
310 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
311 	vq->vq_free_cnt = vq->vq_nentries;
312 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
313 
314 	vring_desc_init(vr->desc, size);
315 
316 	/*
317 	 * Disable device(host) interrupting guest
318 	 */
319 	virtqueue_disable_intr(vq);
320 }
321 
322 static int
323 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
324 {
325 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
326 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
327 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
328 	unsigned int vq_size, size;
329 	struct virtio_hw *hw = dev->data->dev_private;
330 	struct virtnet_rx *rxvq = NULL;
331 	struct virtnet_tx *txvq = NULL;
332 	struct virtnet_ctl *cvq = NULL;
333 	struct virtqueue *vq;
334 	size_t sz_hdr_mz = 0;
335 	void *sw_ring = NULL;
336 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
337 	int ret;
338 
339 	PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
340 
341 	/*
342 	 * Read the virtqueue size from the Queue Size field
343 	 * Always power of 2 and if 0 virtqueue does not exist
344 	 */
345 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
346 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
347 	if (vq_size == 0) {
348 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
349 		return -EINVAL;
350 	}
351 
352 	if (!rte_is_power_of_2(vq_size)) {
353 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
354 		return -EINVAL;
355 	}
356 
357 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
358 		 dev->data->port_id, vtpci_queue_idx);
359 
360 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
361 				vq_size * sizeof(struct vq_desc_extra),
362 				RTE_CACHE_LINE_SIZE);
363 	if (queue_type == VTNET_TQ) {
364 		/*
365 		 * For each xmit packet, allocate a virtio_net_hdr
366 		 * and indirect ring elements
367 		 */
368 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
369 	} else if (queue_type == VTNET_CQ) {
370 		/* Allocate a page for control vq command, data and status */
371 		sz_hdr_mz = PAGE_SIZE;
372 	}
373 
374 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
375 				SOCKET_ID_ANY);
376 	if (vq == NULL) {
377 		PMD_INIT_LOG(ERR, "can not allocate vq");
378 		return -ENOMEM;
379 	}
380 	hw->vqs[vtpci_queue_idx] = vq;
381 
382 	vq->hw = hw;
383 	vq->vq_queue_index = vtpci_queue_idx;
384 	vq->vq_nentries = vq_size;
385 
386 	/*
387 	 * Reserve a memzone for vring elements
388 	 */
389 	size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
390 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
391 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
392 		     size, vq->vq_ring_size);
393 
394 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
395 					 SOCKET_ID_ANY,
396 					 0, VIRTIO_PCI_VRING_ALIGN);
397 	if (mz == NULL) {
398 		if (rte_errno == EEXIST)
399 			mz = rte_memzone_lookup(vq_name);
400 		if (mz == NULL) {
401 			ret = -ENOMEM;
402 			goto fail_q_alloc;
403 		}
404 	}
405 
406 	memset(mz->addr, 0, mz->len);
407 
408 	vq->vq_ring_mem = mz->iova;
409 	vq->vq_ring_virt_mem = mz->addr;
410 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
411 		     (uint64_t)mz->iova);
412 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
413 		     (uint64_t)(uintptr_t)mz->addr);
414 
415 	virtio_init_vring(vq);
416 
417 	if (sz_hdr_mz) {
418 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
419 			 dev->data->port_id, vtpci_queue_idx);
420 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
421 						     SOCKET_ID_ANY, 0,
422 						     RTE_CACHE_LINE_SIZE);
423 		if (hdr_mz == NULL) {
424 			if (rte_errno == EEXIST)
425 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
426 			if (hdr_mz == NULL) {
427 				ret = -ENOMEM;
428 				goto fail_q_alloc;
429 			}
430 		}
431 	}
432 
433 	if (queue_type == VTNET_RQ) {
434 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
435 			       sizeof(vq->sw_ring[0]);
436 
437 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
438 				RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
439 		if (!sw_ring) {
440 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
441 			ret = -ENOMEM;
442 			goto fail_q_alloc;
443 		}
444 
445 		vq->sw_ring = sw_ring;
446 		rxvq = &vq->rxq;
447 		rxvq->vq = vq;
448 		rxvq->port_id = dev->data->port_id;
449 		rxvq->mz = mz;
450 	} else if (queue_type == VTNET_TQ) {
451 		txvq = &vq->txq;
452 		txvq->vq = vq;
453 		txvq->port_id = dev->data->port_id;
454 		txvq->mz = mz;
455 		txvq->virtio_net_hdr_mz = hdr_mz;
456 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
457 	} else if (queue_type == VTNET_CQ) {
458 		cvq = &vq->cq;
459 		cvq->vq = vq;
460 		cvq->mz = mz;
461 		cvq->virtio_net_hdr_mz = hdr_mz;
462 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
463 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
464 
465 		hw->cvq = cvq;
466 	}
467 
468 	/* For virtio_user case (that is when hw->dev is NULL), we use
469 	 * virtual address. And we need properly set _offset_, please see
470 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
471 	 */
472 	if (!hw->virtio_user_dev)
473 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
474 	else {
475 		vq->vq_ring_mem = (uintptr_t)mz->addr;
476 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
477 		if (queue_type == VTNET_TQ)
478 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
479 		else if (queue_type == VTNET_CQ)
480 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
481 	}
482 
483 	if (queue_type == VTNET_TQ) {
484 		struct virtio_tx_region *txr;
485 		unsigned int i;
486 
487 		txr = hdr_mz->addr;
488 		memset(txr, 0, vq_size * sizeof(*txr));
489 		for (i = 0; i < vq_size; i++) {
490 			struct vring_desc *start_dp = txr[i].tx_indir;
491 
492 			vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
493 
494 			/* first indirect descriptor is always the tx header */
495 			start_dp->addr = txvq->virtio_net_hdr_mem
496 				+ i * sizeof(*txr)
497 				+ offsetof(struct virtio_tx_region, tx_hdr);
498 
499 			start_dp->len = hw->vtnet_hdr_size;
500 			start_dp->flags = VRING_DESC_F_NEXT;
501 		}
502 	}
503 
504 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
505 		PMD_INIT_LOG(ERR, "setup_queue failed");
506 		return -EINVAL;
507 	}
508 
509 	return 0;
510 
511 fail_q_alloc:
512 	rte_free(sw_ring);
513 	rte_memzone_free(hdr_mz);
514 	rte_memzone_free(mz);
515 	rte_free(vq);
516 
517 	return ret;
518 }
519 
520 static void
521 virtio_free_queues(struct virtio_hw *hw)
522 {
523 	uint16_t nr_vq = virtio_get_nr_vq(hw);
524 	struct virtqueue *vq;
525 	int queue_type;
526 	uint16_t i;
527 
528 	if (hw->vqs == NULL)
529 		return;
530 
531 	for (i = 0; i < nr_vq; i++) {
532 		vq = hw->vqs[i];
533 		if (!vq)
534 			continue;
535 
536 		queue_type = virtio_get_queue_type(hw, i);
537 		if (queue_type == VTNET_RQ) {
538 			rte_free(vq->sw_ring);
539 			rte_memzone_free(vq->rxq.mz);
540 		} else if (queue_type == VTNET_TQ) {
541 			rte_memzone_free(vq->txq.mz);
542 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
543 		} else {
544 			rte_memzone_free(vq->cq.mz);
545 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
546 		}
547 
548 		rte_free(vq);
549 		hw->vqs[i] = NULL;
550 	}
551 
552 	rte_free(hw->vqs);
553 	hw->vqs = NULL;
554 }
555 
556 static int
557 virtio_alloc_queues(struct rte_eth_dev *dev)
558 {
559 	struct virtio_hw *hw = dev->data->dev_private;
560 	uint16_t nr_vq = virtio_get_nr_vq(hw);
561 	uint16_t i;
562 	int ret;
563 
564 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
565 	if (!hw->vqs) {
566 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
567 		return -ENOMEM;
568 	}
569 
570 	for (i = 0; i < nr_vq; i++) {
571 		ret = virtio_init_queue(dev, i);
572 		if (ret < 0) {
573 			virtio_free_queues(hw);
574 			return ret;
575 		}
576 	}
577 
578 	return 0;
579 }
580 
581 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
582 
583 static void
584 virtio_dev_close(struct rte_eth_dev *dev)
585 {
586 	struct virtio_hw *hw = dev->data->dev_private;
587 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
588 
589 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
590 
591 	/* reset the NIC */
592 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
593 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
594 	if (intr_conf->rxq)
595 		virtio_queues_unbind_intr(dev);
596 
597 	if (intr_conf->lsc || intr_conf->rxq) {
598 		virtio_intr_disable(dev);
599 		rte_intr_efd_disable(dev->intr_handle);
600 		rte_free(dev->intr_handle->intr_vec);
601 		dev->intr_handle->intr_vec = NULL;
602 	}
603 
604 	vtpci_reset(hw);
605 	virtio_dev_free_mbufs(dev);
606 	virtio_free_queues(hw);
607 }
608 
609 static void
610 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
611 {
612 	struct virtio_hw *hw = dev->data->dev_private;
613 	struct virtio_pmd_ctrl ctrl;
614 	int dlen[1];
615 	int ret;
616 
617 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
618 		PMD_INIT_LOG(INFO, "host does not support rx control");
619 		return;
620 	}
621 
622 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
623 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
624 	ctrl.data[0] = 1;
625 	dlen[0] = 1;
626 
627 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
628 	if (ret)
629 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
630 }
631 
632 static void
633 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
634 {
635 	struct virtio_hw *hw = dev->data->dev_private;
636 	struct virtio_pmd_ctrl ctrl;
637 	int dlen[1];
638 	int ret;
639 
640 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
641 		PMD_INIT_LOG(INFO, "host does not support rx control");
642 		return;
643 	}
644 
645 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
646 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
647 	ctrl.data[0] = 0;
648 	dlen[0] = 1;
649 
650 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
651 	if (ret)
652 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
653 }
654 
655 static void
656 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
657 {
658 	struct virtio_hw *hw = dev->data->dev_private;
659 	struct virtio_pmd_ctrl ctrl;
660 	int dlen[1];
661 	int ret;
662 
663 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
664 		PMD_INIT_LOG(INFO, "host does not support rx control");
665 		return;
666 	}
667 
668 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
669 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
670 	ctrl.data[0] = 1;
671 	dlen[0] = 1;
672 
673 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
674 	if (ret)
675 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
676 }
677 
678 static void
679 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
680 {
681 	struct virtio_hw *hw = dev->data->dev_private;
682 	struct virtio_pmd_ctrl ctrl;
683 	int dlen[1];
684 	int ret;
685 
686 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
687 		PMD_INIT_LOG(INFO, "host does not support rx control");
688 		return;
689 	}
690 
691 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
692 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
693 	ctrl.data[0] = 0;
694 	dlen[0] = 1;
695 
696 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
697 	if (ret)
698 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
699 }
700 
701 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
702 static int
703 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
704 {
705 	struct virtio_hw *hw = dev->data->dev_private;
706 	uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
707 				 hw->vtnet_hdr_size;
708 	uint32_t frame_size = mtu + ether_hdr_len;
709 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
710 
711 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
712 
713 	if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
714 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
715 			ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
716 		return -EINVAL;
717 	}
718 	return 0;
719 }
720 
721 static int
722 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
723 {
724 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
725 	struct virtqueue *vq = rxvq->vq;
726 
727 	virtqueue_enable_intr(vq);
728 	return 0;
729 }
730 
731 static int
732 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
733 {
734 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
735 	struct virtqueue *vq = rxvq->vq;
736 
737 	virtqueue_disable_intr(vq);
738 	return 0;
739 }
740 
741 /*
742  * dev_ops for virtio, bare necessities for basic operation
743  */
744 static const struct eth_dev_ops virtio_eth_dev_ops = {
745 	.dev_configure           = virtio_dev_configure,
746 	.dev_start               = virtio_dev_start,
747 	.dev_stop                = virtio_dev_stop,
748 	.dev_close               = virtio_dev_close,
749 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
750 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
751 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
752 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
753 	.mtu_set                 = virtio_mtu_set,
754 	.dev_infos_get           = virtio_dev_info_get,
755 	.stats_get               = virtio_dev_stats_get,
756 	.xstats_get              = virtio_dev_xstats_get,
757 	.xstats_get_names        = virtio_dev_xstats_get_names,
758 	.stats_reset             = virtio_dev_stats_reset,
759 	.xstats_reset            = virtio_dev_stats_reset,
760 	.link_update             = virtio_dev_link_update,
761 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
762 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
763 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
764 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
765 	.rx_queue_release        = virtio_dev_queue_release,
766 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
767 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
768 	.tx_queue_release        = virtio_dev_queue_release,
769 	/* collect stats per queue */
770 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
771 	.vlan_filter_set         = virtio_vlan_filter_set,
772 	.mac_addr_add            = virtio_mac_addr_add,
773 	.mac_addr_remove         = virtio_mac_addr_remove,
774 	.mac_addr_set            = virtio_mac_addr_set,
775 };
776 
777 static inline int
778 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
779 				struct rte_eth_link *link)
780 {
781 	struct rte_eth_link *dst = link;
782 	struct rte_eth_link *src = &(dev->data->dev_link);
783 
784 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
785 			*(uint64_t *)src) == 0)
786 		return -1;
787 
788 	return 0;
789 }
790 
791 /**
792  * Atomically writes the link status information into global
793  * structure rte_eth_dev.
794  *
795  * @param dev
796  *   - Pointer to the structure rte_eth_dev to read from.
797  *   - Pointer to the buffer to be saved with the link status.
798  *
799  * @return
800  *   - On success, zero.
801  *   - On failure, negative value.
802  */
803 static inline int
804 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
805 		struct rte_eth_link *link)
806 {
807 	struct rte_eth_link *dst = &(dev->data->dev_link);
808 	struct rte_eth_link *src = link;
809 
810 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
811 					*(uint64_t *)src) == 0)
812 		return -1;
813 
814 	return 0;
815 }
816 
817 static void
818 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
819 {
820 	unsigned i;
821 
822 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
823 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
824 		if (txvq == NULL)
825 			continue;
826 
827 		stats->opackets += txvq->stats.packets;
828 		stats->obytes += txvq->stats.bytes;
829 		stats->oerrors += txvq->stats.errors;
830 
831 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
832 			stats->q_opackets[i] = txvq->stats.packets;
833 			stats->q_obytes[i] = txvq->stats.bytes;
834 		}
835 	}
836 
837 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
838 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
839 		if (rxvq == NULL)
840 			continue;
841 
842 		stats->ipackets += rxvq->stats.packets;
843 		stats->ibytes += rxvq->stats.bytes;
844 		stats->ierrors += rxvq->stats.errors;
845 
846 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
847 			stats->q_ipackets[i] = rxvq->stats.packets;
848 			stats->q_ibytes[i] = rxvq->stats.bytes;
849 		}
850 	}
851 
852 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
853 }
854 
855 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
856 				       struct rte_eth_xstat_name *xstats_names,
857 				       __rte_unused unsigned limit)
858 {
859 	unsigned i;
860 	unsigned count = 0;
861 	unsigned t;
862 
863 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
864 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
865 
866 	if (xstats_names != NULL) {
867 		/* Note: limit checked in rte_eth_xstats_names() */
868 
869 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
870 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
871 			if (rxvq == NULL)
872 				continue;
873 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
874 				snprintf(xstats_names[count].name,
875 					sizeof(xstats_names[count].name),
876 					"rx_q%u_%s", i,
877 					rte_virtio_rxq_stat_strings[t].name);
878 				count++;
879 			}
880 		}
881 
882 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
883 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
884 			if (txvq == NULL)
885 				continue;
886 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
887 				snprintf(xstats_names[count].name,
888 					sizeof(xstats_names[count].name),
889 					"tx_q%u_%s", i,
890 					rte_virtio_txq_stat_strings[t].name);
891 				count++;
892 			}
893 		}
894 		return count;
895 	}
896 	return nstats;
897 }
898 
899 static int
900 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
901 		      unsigned n)
902 {
903 	unsigned i;
904 	unsigned count = 0;
905 
906 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
907 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
908 
909 	if (n < nstats)
910 		return nstats;
911 
912 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
913 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
914 
915 		if (rxvq == NULL)
916 			continue;
917 
918 		unsigned t;
919 
920 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
921 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
922 				rte_virtio_rxq_stat_strings[t].offset);
923 			xstats[count].id = count;
924 			count++;
925 		}
926 	}
927 
928 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
929 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
930 
931 		if (txvq == NULL)
932 			continue;
933 
934 		unsigned t;
935 
936 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
937 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
938 				rte_virtio_txq_stat_strings[t].offset);
939 			xstats[count].id = count;
940 			count++;
941 		}
942 	}
943 
944 	return count;
945 }
946 
947 static int
948 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
949 {
950 	virtio_update_stats(dev, stats);
951 
952 	return 0;
953 }
954 
955 static void
956 virtio_dev_stats_reset(struct rte_eth_dev *dev)
957 {
958 	unsigned int i;
959 
960 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
961 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
962 		if (txvq == NULL)
963 			continue;
964 
965 		txvq->stats.packets = 0;
966 		txvq->stats.bytes = 0;
967 		txvq->stats.errors = 0;
968 		txvq->stats.multicast = 0;
969 		txvq->stats.broadcast = 0;
970 		memset(txvq->stats.size_bins, 0,
971 		       sizeof(txvq->stats.size_bins[0]) * 8);
972 	}
973 
974 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
975 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
976 		if (rxvq == NULL)
977 			continue;
978 
979 		rxvq->stats.packets = 0;
980 		rxvq->stats.bytes = 0;
981 		rxvq->stats.errors = 0;
982 		rxvq->stats.multicast = 0;
983 		rxvq->stats.broadcast = 0;
984 		memset(rxvq->stats.size_bins, 0,
985 		       sizeof(rxvq->stats.size_bins[0]) * 8);
986 	}
987 }
988 
989 static void
990 virtio_set_hwaddr(struct virtio_hw *hw)
991 {
992 	vtpci_write_dev_config(hw,
993 			offsetof(struct virtio_net_config, mac),
994 			&hw->mac_addr, ETHER_ADDR_LEN);
995 }
996 
997 static void
998 virtio_get_hwaddr(struct virtio_hw *hw)
999 {
1000 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1001 		vtpci_read_dev_config(hw,
1002 			offsetof(struct virtio_net_config, mac),
1003 			&hw->mac_addr, ETHER_ADDR_LEN);
1004 	} else {
1005 		eth_random_addr(&hw->mac_addr[0]);
1006 		virtio_set_hwaddr(hw);
1007 	}
1008 }
1009 
1010 static int
1011 virtio_mac_table_set(struct virtio_hw *hw,
1012 		     const struct virtio_net_ctrl_mac *uc,
1013 		     const struct virtio_net_ctrl_mac *mc)
1014 {
1015 	struct virtio_pmd_ctrl ctrl;
1016 	int err, len[2];
1017 
1018 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1019 		PMD_DRV_LOG(INFO, "host does not support mac table");
1020 		return -1;
1021 	}
1022 
1023 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1024 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1025 
1026 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1027 	memcpy(ctrl.data, uc, len[0]);
1028 
1029 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1030 	memcpy(ctrl.data + len[0], mc, len[1]);
1031 
1032 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1033 	if (err != 0)
1034 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1035 	return err;
1036 }
1037 
1038 static int
1039 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1040 		    uint32_t index, uint32_t vmdq __rte_unused)
1041 {
1042 	struct virtio_hw *hw = dev->data->dev_private;
1043 	const struct ether_addr *addrs = dev->data->mac_addrs;
1044 	unsigned int i;
1045 	struct virtio_net_ctrl_mac *uc, *mc;
1046 
1047 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1048 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1049 		return -EINVAL;
1050 	}
1051 
1052 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1053 	uc->entries = 0;
1054 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1055 	mc->entries = 0;
1056 
1057 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1058 		const struct ether_addr *addr
1059 			= (i == index) ? mac_addr : addrs + i;
1060 		struct virtio_net_ctrl_mac *tbl
1061 			= is_multicast_ether_addr(addr) ? mc : uc;
1062 
1063 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1064 	}
1065 
1066 	return virtio_mac_table_set(hw, uc, mc);
1067 }
1068 
1069 static void
1070 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1071 {
1072 	struct virtio_hw *hw = dev->data->dev_private;
1073 	struct ether_addr *addrs = dev->data->mac_addrs;
1074 	struct virtio_net_ctrl_mac *uc, *mc;
1075 	unsigned int i;
1076 
1077 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1078 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1079 		return;
1080 	}
1081 
1082 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1083 	uc->entries = 0;
1084 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1085 	mc->entries = 0;
1086 
1087 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1088 		struct virtio_net_ctrl_mac *tbl;
1089 
1090 		if (i == index || is_zero_ether_addr(addrs + i))
1091 			continue;
1092 
1093 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1094 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1095 	}
1096 
1097 	virtio_mac_table_set(hw, uc, mc);
1098 }
1099 
1100 static void
1101 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1102 {
1103 	struct virtio_hw *hw = dev->data->dev_private;
1104 
1105 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1106 
1107 	/* Use atomic update if available */
1108 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1109 		struct virtio_pmd_ctrl ctrl;
1110 		int len = ETHER_ADDR_LEN;
1111 
1112 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1113 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1114 
1115 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1116 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
1117 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1118 		virtio_set_hwaddr(hw);
1119 }
1120 
1121 static int
1122 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1123 {
1124 	struct virtio_hw *hw = dev->data->dev_private;
1125 	struct virtio_pmd_ctrl ctrl;
1126 	int len;
1127 
1128 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1129 		return -ENOTSUP;
1130 
1131 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1132 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1133 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1134 	len = sizeof(vlan_id);
1135 
1136 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1137 }
1138 
1139 static int
1140 virtio_intr_enable(struct rte_eth_dev *dev)
1141 {
1142 	struct virtio_hw *hw = dev->data->dev_private;
1143 
1144 	if (rte_intr_enable(dev->intr_handle) < 0)
1145 		return -1;
1146 
1147 	if (!hw->virtio_user_dev)
1148 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1149 
1150 	return 0;
1151 }
1152 
1153 static int
1154 virtio_intr_disable(struct rte_eth_dev *dev)
1155 {
1156 	struct virtio_hw *hw = dev->data->dev_private;
1157 
1158 	if (rte_intr_disable(dev->intr_handle) < 0)
1159 		return -1;
1160 
1161 	if (!hw->virtio_user_dev)
1162 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1163 
1164 	return 0;
1165 }
1166 
1167 static int
1168 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1169 {
1170 	uint64_t host_features;
1171 
1172 	/* Prepare guest_features: feature that driver wants to support */
1173 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1174 		req_features);
1175 
1176 	/* Read device(host) feature bits */
1177 	host_features = VTPCI_OPS(hw)->get_features(hw);
1178 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1179 		host_features);
1180 
1181 	/* If supported, ensure MTU value is valid before acknowledging it. */
1182 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1183 		struct virtio_net_config config;
1184 
1185 		vtpci_read_dev_config(hw,
1186 			offsetof(struct virtio_net_config, mtu),
1187 			&config.mtu, sizeof(config.mtu));
1188 
1189 		if (config.mtu < ETHER_MIN_MTU)
1190 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1191 	}
1192 
1193 	/*
1194 	 * Negotiate features: Subset of device feature bits are written back
1195 	 * guest feature bits.
1196 	 */
1197 	hw->guest_features = req_features;
1198 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1199 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1200 		hw->guest_features);
1201 
1202 	if (hw->modern) {
1203 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1204 			PMD_INIT_LOG(ERR,
1205 				"VIRTIO_F_VERSION_1 features is not enabled.");
1206 			return -1;
1207 		}
1208 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1209 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1210 			PMD_INIT_LOG(ERR,
1211 				"failed to set FEATURES_OK status!");
1212 			return -1;
1213 		}
1214 	}
1215 
1216 	hw->req_guest_features = req_features;
1217 
1218 	return 0;
1219 }
1220 
1221 int
1222 virtio_dev_pause(struct rte_eth_dev *dev)
1223 {
1224 	struct virtio_hw *hw = dev->data->dev_private;
1225 
1226 	rte_spinlock_lock(&hw->state_lock);
1227 
1228 	if (hw->started == 0) {
1229 		/* Device is just stopped. */
1230 		rte_spinlock_unlock(&hw->state_lock);
1231 		return -1;
1232 	}
1233 	hw->started = 0;
1234 	/*
1235 	 * Prevent the worker threads from touching queues to avoid contention,
1236 	 * 1 ms should be enough for the ongoing Tx function to finish.
1237 	 */
1238 	rte_delay_ms(1);
1239 	return 0;
1240 }
1241 
1242 /*
1243  * Recover hw state to let the worker threads continue.
1244  */
1245 void
1246 virtio_dev_resume(struct rte_eth_dev *dev)
1247 {
1248 	struct virtio_hw *hw = dev->data->dev_private;
1249 
1250 	hw->started = 1;
1251 	rte_spinlock_unlock(&hw->state_lock);
1252 }
1253 
1254 /*
1255  * Should be called only after device is paused.
1256  */
1257 int
1258 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1259 		int nb_pkts)
1260 {
1261 	struct virtio_hw *hw = dev->data->dev_private;
1262 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1263 	int ret;
1264 
1265 	hw->inject_pkts = tx_pkts;
1266 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1267 	hw->inject_pkts = NULL;
1268 
1269 	return ret;
1270 }
1271 
1272 static void
1273 virtio_notify_peers(struct rte_eth_dev *dev)
1274 {
1275 	struct virtio_hw *hw = dev->data->dev_private;
1276 	struct virtnet_rx *rxvq = dev->data->rx_queues[0];
1277 	struct rte_mbuf *rarp_mbuf;
1278 
1279 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1280 			(struct ether_addr *)hw->mac_addr);
1281 	if (rarp_mbuf == NULL) {
1282 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1283 		return;
1284 	}
1285 
1286 	/* If virtio port just stopped, no need to send RARP */
1287 	if (virtio_dev_pause(dev) < 0) {
1288 		rte_pktmbuf_free(rarp_mbuf);
1289 		return;
1290 	}
1291 
1292 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1293 	virtio_dev_resume(dev);
1294 }
1295 
1296 static void
1297 virtio_ack_link_announce(struct rte_eth_dev *dev)
1298 {
1299 	struct virtio_hw *hw = dev->data->dev_private;
1300 	struct virtio_pmd_ctrl ctrl;
1301 
1302 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1303 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1304 
1305 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1306 }
1307 
1308 /*
1309  * Process virtio config changed interrupt. Call the callback
1310  * if link state changed, generate gratuitous RARP packet if
1311  * the status indicates an ANNOUNCE.
1312  */
1313 void
1314 virtio_interrupt_handler(void *param)
1315 {
1316 	struct rte_eth_dev *dev = param;
1317 	struct virtio_hw *hw = dev->data->dev_private;
1318 	uint8_t isr;
1319 
1320 	/* Read interrupt status which clears interrupt */
1321 	isr = vtpci_isr(hw);
1322 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1323 
1324 	if (virtio_intr_enable(dev) < 0)
1325 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1326 
1327 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1328 		if (virtio_dev_link_update(dev, 0) == 0)
1329 			_rte_eth_dev_callback_process(dev,
1330 						      RTE_ETH_EVENT_INTR_LSC,
1331 						      NULL);
1332 	}
1333 
1334 	if (isr & VIRTIO_NET_S_ANNOUNCE) {
1335 		virtio_notify_peers(dev);
1336 		virtio_ack_link_announce(dev);
1337 	}
1338 }
1339 
1340 /* set rx and tx handlers according to what is supported */
1341 static void
1342 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1343 {
1344 	struct virtio_hw *hw = eth_dev->data->dev_private;
1345 
1346 	if (hw->use_simple_rx) {
1347 		PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1348 			eth_dev->data->port_id);
1349 		eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1350 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1351 		PMD_INIT_LOG(INFO,
1352 			"virtio: using mergeable buffer Rx path on port %u",
1353 			eth_dev->data->port_id);
1354 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1355 	} else {
1356 		PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1357 			eth_dev->data->port_id);
1358 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1359 	}
1360 
1361 	if (hw->use_simple_tx) {
1362 		PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1363 			eth_dev->data->port_id);
1364 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1365 	} else {
1366 		PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1367 			eth_dev->data->port_id);
1368 		eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1369 	}
1370 }
1371 
1372 /* Only support 1:1 queue/interrupt mapping so far.
1373  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1374  * interrupt vectors (<N+1).
1375  */
1376 static int
1377 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1378 {
1379 	uint32_t i;
1380 	struct virtio_hw *hw = dev->data->dev_private;
1381 
1382 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1383 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1384 		dev->intr_handle->intr_vec[i] = i + 1;
1385 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1386 						 VIRTIO_MSI_NO_VECTOR) {
1387 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1388 			return -EBUSY;
1389 		}
1390 	}
1391 
1392 	return 0;
1393 }
1394 
1395 static void
1396 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1397 {
1398 	uint32_t i;
1399 	struct virtio_hw *hw = dev->data->dev_private;
1400 
1401 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1402 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1403 		VTPCI_OPS(hw)->set_queue_irq(hw,
1404 					     hw->vqs[i * VTNET_CQ],
1405 					     VIRTIO_MSI_NO_VECTOR);
1406 }
1407 
1408 static int
1409 virtio_configure_intr(struct rte_eth_dev *dev)
1410 {
1411 	struct virtio_hw *hw = dev->data->dev_private;
1412 
1413 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1414 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1415 		return -ENOTSUP;
1416 	}
1417 
1418 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1419 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1420 		return -1;
1421 	}
1422 
1423 	if (!dev->intr_handle->intr_vec) {
1424 		dev->intr_handle->intr_vec =
1425 			rte_zmalloc("intr_vec",
1426 				    hw->max_queue_pairs * sizeof(int), 0);
1427 		if (!dev->intr_handle->intr_vec) {
1428 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1429 				     hw->max_queue_pairs);
1430 			return -ENOMEM;
1431 		}
1432 	}
1433 
1434 	/* Re-register callback to update max_intr */
1435 	rte_intr_callback_unregister(dev->intr_handle,
1436 				     virtio_interrupt_handler,
1437 				     dev);
1438 	rte_intr_callback_register(dev->intr_handle,
1439 				   virtio_interrupt_handler,
1440 				   dev);
1441 
1442 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1443 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1444 	 * And for legacy devices, this should be done before queue/vec binding
1445 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1446 	 * (22) will be ignored.
1447 	 */
1448 	if (virtio_intr_enable(dev) < 0) {
1449 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1450 		return -1;
1451 	}
1452 
1453 	if (virtio_queues_bind_intr(dev) < 0) {
1454 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1455 		return -1;
1456 	}
1457 
1458 	return 0;
1459 }
1460 
1461 /* reset device and renegotiate features if needed */
1462 static int
1463 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1464 {
1465 	struct virtio_hw *hw = eth_dev->data->dev_private;
1466 	struct virtio_net_config *config;
1467 	struct virtio_net_config local_config;
1468 	struct rte_pci_device *pci_dev = NULL;
1469 	int ret;
1470 
1471 	/* Reset the device although not necessary at startup */
1472 	vtpci_reset(hw);
1473 
1474 	if (hw->vqs) {
1475 		virtio_dev_free_mbufs(eth_dev);
1476 		virtio_free_queues(hw);
1477 	}
1478 
1479 	/* Tell the host we've noticed this device. */
1480 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1481 
1482 	/* Tell the host we've known how to drive the device. */
1483 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1484 	if (virtio_negotiate_features(hw, req_features) < 0)
1485 		return -1;
1486 
1487 	if (!hw->virtio_user_dev) {
1488 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1489 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1490 	}
1491 
1492 	/* If host does not support both status and MSI-X then disable LSC */
1493 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1494 	    hw->use_msix != VIRTIO_MSIX_NONE)
1495 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1496 	else
1497 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1498 
1499 	/* Setting up rx_header size for the device */
1500 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1501 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1502 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1503 	else
1504 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1505 
1506 	/* Copy the permanent MAC address to: virtio_hw */
1507 	virtio_get_hwaddr(hw);
1508 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1509 			&eth_dev->data->mac_addrs[0]);
1510 	PMD_INIT_LOG(DEBUG,
1511 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1512 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1513 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1514 
1515 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1516 		config = &local_config;
1517 
1518 		vtpci_read_dev_config(hw,
1519 			offsetof(struct virtio_net_config, mac),
1520 			&config->mac, sizeof(config->mac));
1521 
1522 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1523 			vtpci_read_dev_config(hw,
1524 				offsetof(struct virtio_net_config, status),
1525 				&config->status, sizeof(config->status));
1526 		} else {
1527 			PMD_INIT_LOG(DEBUG,
1528 				     "VIRTIO_NET_F_STATUS is not supported");
1529 			config->status = 0;
1530 		}
1531 
1532 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1533 			vtpci_read_dev_config(hw,
1534 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1535 				&config->max_virtqueue_pairs,
1536 				sizeof(config->max_virtqueue_pairs));
1537 		} else {
1538 			PMD_INIT_LOG(DEBUG,
1539 				     "VIRTIO_NET_F_MQ is not supported");
1540 			config->max_virtqueue_pairs = 1;
1541 		}
1542 
1543 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1544 
1545 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1546 			vtpci_read_dev_config(hw,
1547 				offsetof(struct virtio_net_config, mtu),
1548 				&config->mtu,
1549 				sizeof(config->mtu));
1550 
1551 			/*
1552 			 * MTU value has already been checked at negotiation
1553 			 * time, but check again in case it has changed since
1554 			 * then, which should not happen.
1555 			 */
1556 			if (config->mtu < ETHER_MIN_MTU) {
1557 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1558 						config->mtu);
1559 				return -1;
1560 			}
1561 
1562 			hw->max_mtu = config->mtu;
1563 			/* Set initial MTU to maximum one supported by vhost */
1564 			eth_dev->data->mtu = config->mtu;
1565 
1566 		} else {
1567 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1568 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1569 		}
1570 
1571 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1572 				config->max_virtqueue_pairs);
1573 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1574 		PMD_INIT_LOG(DEBUG,
1575 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1576 				config->mac[0], config->mac[1],
1577 				config->mac[2], config->mac[3],
1578 				config->mac[4], config->mac[5]);
1579 	} else {
1580 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1581 		hw->max_queue_pairs = 1;
1582 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1583 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1584 	}
1585 
1586 	ret = virtio_alloc_queues(eth_dev);
1587 	if (ret < 0)
1588 		return ret;
1589 
1590 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1591 		if (virtio_configure_intr(eth_dev) < 0) {
1592 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1593 			return -1;
1594 		}
1595 	}
1596 
1597 	vtpci_reinit_complete(hw);
1598 
1599 	if (pci_dev)
1600 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1601 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1602 			pci_dev->id.device_id);
1603 
1604 	return 0;
1605 }
1606 
1607 /*
1608  * Remap the PCI device again (IO port map for legacy device and
1609  * memory map for modern device), so that the secondary process
1610  * could have the PCI initiated correctly.
1611  */
1612 static int
1613 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1614 {
1615 	if (hw->modern) {
1616 		/*
1617 		 * We don't have to re-parse the PCI config space, since
1618 		 * rte_pci_map_device() makes sure the mapped address
1619 		 * in secondary process would equal to the one mapped in
1620 		 * the primary process: error will be returned if that
1621 		 * requirement is not met.
1622 		 *
1623 		 * That said, we could simply reuse all cap pointers
1624 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1625 		 * primary process, which is stored in shared memory.
1626 		 */
1627 		if (rte_pci_map_device(pci_dev)) {
1628 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1629 			return -1;
1630 		}
1631 	} else {
1632 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1633 			return -1;
1634 	}
1635 
1636 	return 0;
1637 }
1638 
1639 static void
1640 virtio_set_vtpci_ops(struct virtio_hw *hw)
1641 {
1642 #ifdef RTE_VIRTIO_USER
1643 	if (hw->virtio_user_dev)
1644 		VTPCI_OPS(hw) = &virtio_user_ops;
1645 	else
1646 #endif
1647 	if (hw->modern)
1648 		VTPCI_OPS(hw) = &modern_ops;
1649 	else
1650 		VTPCI_OPS(hw) = &legacy_ops;
1651 }
1652 
1653 /*
1654  * This function is based on probe() function in virtio_pci.c
1655  * It returns 0 on success.
1656  */
1657 int
1658 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1659 {
1660 	struct virtio_hw *hw = eth_dev->data->dev_private;
1661 	int ret;
1662 
1663 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1664 
1665 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1666 
1667 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1668 		if (!hw->virtio_user_dev) {
1669 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1670 			if (ret)
1671 				return ret;
1672 		}
1673 
1674 		virtio_set_vtpci_ops(hw);
1675 		set_rxtx_funcs(eth_dev);
1676 
1677 		return 0;
1678 	}
1679 
1680 	/* Allocate memory for storing MAC addresses */
1681 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1682 	if (eth_dev->data->mac_addrs == NULL) {
1683 		PMD_INIT_LOG(ERR,
1684 			"Failed to allocate %d bytes needed to store MAC addresses",
1685 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1686 		return -ENOMEM;
1687 	}
1688 
1689 	hw->port_id = eth_dev->data->port_id;
1690 	/* For virtio_user case the hw->virtio_user_dev is populated by
1691 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1692 	 */
1693 	if (!hw->virtio_user_dev) {
1694 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1695 		if (ret)
1696 			goto out;
1697 	}
1698 
1699 	/* reset device and negotiate default features */
1700 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1701 	if (ret < 0)
1702 		goto out;
1703 
1704 	/* Setup interrupt callback  */
1705 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1706 		rte_intr_callback_register(eth_dev->intr_handle,
1707 			virtio_interrupt_handler, eth_dev);
1708 
1709 	return 0;
1710 
1711 out:
1712 	rte_free(eth_dev->data->mac_addrs);
1713 	return ret;
1714 }
1715 
1716 static int
1717 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1718 {
1719 	PMD_INIT_FUNC_TRACE();
1720 
1721 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1722 		return -EPERM;
1723 
1724 	virtio_dev_stop(eth_dev);
1725 	virtio_dev_close(eth_dev);
1726 
1727 	eth_dev->dev_ops = NULL;
1728 	eth_dev->tx_pkt_burst = NULL;
1729 	eth_dev->rx_pkt_burst = NULL;
1730 
1731 	rte_free(eth_dev->data->mac_addrs);
1732 	eth_dev->data->mac_addrs = NULL;
1733 
1734 	/* reset interrupt callback  */
1735 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1736 		rte_intr_callback_unregister(eth_dev->intr_handle,
1737 						virtio_interrupt_handler,
1738 						eth_dev);
1739 	if (eth_dev->device)
1740 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1741 
1742 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1743 
1744 	return 0;
1745 }
1746 
1747 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1748 	struct rte_pci_device *pci_dev)
1749 {
1750 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1751 		eth_virtio_dev_init);
1752 }
1753 
1754 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1755 {
1756 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1757 }
1758 
1759 static struct rte_pci_driver rte_virtio_pmd = {
1760 	.driver = {
1761 		.name = "net_virtio",
1762 	},
1763 	.id_table = pci_id_virtio_map,
1764 	.drv_flags = 0,
1765 	.probe = eth_virtio_pci_probe,
1766 	.remove = eth_virtio_pci_remove,
1767 };
1768 
1769 RTE_INIT(rte_virtio_pmd_init);
1770 static void
1771 rte_virtio_pmd_init(void)
1772 {
1773 	if (rte_eal_iopl_init() != 0) {
1774 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1775 		return;
1776 	}
1777 
1778 	rte_pci_register(&rte_virtio_pmd);
1779 }
1780 
1781 /*
1782  * Configure virtio device
1783  * It returns 0 on success.
1784  */
1785 static int
1786 virtio_dev_configure(struct rte_eth_dev *dev)
1787 {
1788 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1789 	struct virtio_hw *hw = dev->data->dev_private;
1790 	uint64_t req_features;
1791 	int ret;
1792 
1793 	PMD_INIT_LOG(DEBUG, "configure");
1794 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1795 
1796 	if (dev->data->dev_conf.intr_conf.rxq) {
1797 		ret = virtio_init_device(dev, hw->req_guest_features);
1798 		if (ret < 0)
1799 			return ret;
1800 	}
1801 
1802 	/* The name hw_ip_checksum is a bit confusing since it can be
1803 	 * set by the application to request L3 and/or L4 checksums. In
1804 	 * case of virtio, only L4 checksum is supported.
1805 	 */
1806 	if (rxmode->hw_ip_checksum)
1807 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1808 
1809 	if (rxmode->enable_lro)
1810 		req_features |=
1811 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1812 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1813 
1814 	/* if request features changed, reinit the device */
1815 	if (req_features != hw->req_guest_features) {
1816 		ret = virtio_init_device(dev, req_features);
1817 		if (ret < 0)
1818 			return ret;
1819 	}
1820 
1821 	if (rxmode->hw_ip_checksum &&
1822 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1823 		PMD_DRV_LOG(ERR,
1824 			"rx checksum not available on this host");
1825 		return -ENOTSUP;
1826 	}
1827 
1828 	if (rxmode->enable_lro &&
1829 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1830 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
1831 		PMD_DRV_LOG(ERR,
1832 			"Large Receive Offload not available on this host");
1833 		return -ENOTSUP;
1834 	}
1835 
1836 	/* start control queue */
1837 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1838 		virtio_dev_cq_start(dev);
1839 
1840 	hw->vlan_strip = rxmode->hw_vlan_strip;
1841 
1842 	if (rxmode->hw_vlan_filter
1843 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1844 		PMD_DRV_LOG(ERR,
1845 			    "vlan filtering not available on this host");
1846 		return -ENOTSUP;
1847 	}
1848 
1849 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1850 		/* Enable vector (0) for Link State Intrerrupt */
1851 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1852 				VIRTIO_MSI_NO_VECTOR) {
1853 			PMD_DRV_LOG(ERR, "failed to set config vector");
1854 			return -EBUSY;
1855 		}
1856 
1857 	rte_spinlock_init(&hw->state_lock);
1858 
1859 	hw->use_simple_rx = 1;
1860 	hw->use_simple_tx = 1;
1861 
1862 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
1863 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1864 		hw->use_simple_rx = 0;
1865 		hw->use_simple_tx = 0;
1866 	}
1867 #endif
1868 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1869 		hw->use_simple_rx = 0;
1870 		hw->use_simple_tx = 0;
1871 	}
1872 
1873 	if (rxmode->hw_ip_checksum)
1874 		hw->use_simple_rx = 0;
1875 
1876 	return 0;
1877 }
1878 
1879 
1880 static int
1881 virtio_dev_start(struct rte_eth_dev *dev)
1882 {
1883 	uint16_t nb_queues, i;
1884 	struct virtnet_rx *rxvq;
1885 	struct virtnet_tx *txvq __rte_unused;
1886 	struct virtio_hw *hw = dev->data->dev_private;
1887 	int ret;
1888 
1889 	/* Finish the initialization of the queues */
1890 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1891 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
1892 		if (ret < 0)
1893 			return ret;
1894 	}
1895 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1896 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
1897 		if (ret < 0)
1898 			return ret;
1899 	}
1900 
1901 	/* check if lsc interrupt feature is enabled */
1902 	if (dev->data->dev_conf.intr_conf.lsc) {
1903 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1904 			PMD_DRV_LOG(ERR, "link status not supported by host");
1905 			return -ENOTSUP;
1906 		}
1907 	}
1908 
1909 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1910 	 * in device configure, but it could be unmapped  when device is
1911 	 * stopped.
1912 	 */
1913 	if (dev->data->dev_conf.intr_conf.lsc ||
1914 	    dev->data->dev_conf.intr_conf.rxq) {
1915 		virtio_intr_disable(dev);
1916 
1917 		if (virtio_intr_enable(dev) < 0) {
1918 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1919 			return -EIO;
1920 		}
1921 	}
1922 
1923 	/*Notify the backend
1924 	 *Otherwise the tap backend might already stop its queue due to fullness.
1925 	 *vhost backend will have no chance to be waked up
1926 	 */
1927 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1928 	if (hw->max_queue_pairs > 1) {
1929 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1930 			return -EINVAL;
1931 	}
1932 
1933 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1934 
1935 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1936 		rxvq = dev->data->rx_queues[i];
1937 		/* Flush the old packets */
1938 		virtqueue_rxvq_flush(rxvq->vq);
1939 		virtqueue_notify(rxvq->vq);
1940 	}
1941 
1942 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1943 		txvq = dev->data->tx_queues[i];
1944 		virtqueue_notify(txvq->vq);
1945 	}
1946 
1947 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1948 
1949 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1950 		rxvq = dev->data->rx_queues[i];
1951 		VIRTQUEUE_DUMP(rxvq->vq);
1952 	}
1953 
1954 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1955 		txvq = dev->data->tx_queues[i];
1956 		VIRTQUEUE_DUMP(txvq->vq);
1957 	}
1958 
1959 	set_rxtx_funcs(dev);
1960 	hw->started = 1;
1961 
1962 	/* Initialize Link state */
1963 	virtio_dev_link_update(dev, 0);
1964 
1965 	return 0;
1966 }
1967 
1968 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1969 {
1970 	struct virtio_hw *hw = dev->data->dev_private;
1971 	uint16_t nr_vq = virtio_get_nr_vq(hw);
1972 	const char *type __rte_unused;
1973 	unsigned int i, mbuf_num = 0;
1974 	struct virtqueue *vq;
1975 	struct rte_mbuf *buf;
1976 	int queue_type;
1977 
1978 	for (i = 0; i < nr_vq; i++) {
1979 		vq = hw->vqs[i];
1980 		if (!vq)
1981 			continue;
1982 
1983 		queue_type = virtio_get_queue_type(hw, i);
1984 		if (queue_type == VTNET_RQ)
1985 			type = "rxq";
1986 		else if (queue_type == VTNET_TQ)
1987 			type = "txq";
1988 		else
1989 			continue;
1990 
1991 		PMD_INIT_LOG(DEBUG,
1992 			"Before freeing %s[%d] used and unused buf",
1993 			type, i);
1994 		VIRTQUEUE_DUMP(vq);
1995 
1996 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
1997 			rte_pktmbuf_free(buf);
1998 			mbuf_num++;
1999 		}
2000 
2001 		PMD_INIT_LOG(DEBUG,
2002 			"After freeing %s[%d] used and unused buf",
2003 			type, i);
2004 		VIRTQUEUE_DUMP(vq);
2005 	}
2006 
2007 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2008 }
2009 
2010 /*
2011  * Stop device: disable interrupt and mark link down
2012  */
2013 static void
2014 virtio_dev_stop(struct rte_eth_dev *dev)
2015 {
2016 	struct virtio_hw *hw = dev->data->dev_private;
2017 	struct rte_eth_link link;
2018 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2019 
2020 	PMD_INIT_LOG(DEBUG, "stop");
2021 
2022 	rte_spinlock_lock(&hw->state_lock);
2023 	if (intr_conf->lsc || intr_conf->rxq)
2024 		virtio_intr_disable(dev);
2025 
2026 	hw->started = 0;
2027 	memset(&link, 0, sizeof(link));
2028 	virtio_dev_atomic_write_link_status(dev, &link);
2029 	rte_spinlock_unlock(&hw->state_lock);
2030 }
2031 
2032 static int
2033 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2034 {
2035 	struct rte_eth_link link, old;
2036 	uint16_t status;
2037 	struct virtio_hw *hw = dev->data->dev_private;
2038 	memset(&link, 0, sizeof(link));
2039 	virtio_dev_atomic_read_link_status(dev, &link);
2040 	old = link;
2041 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
2042 	link.link_speed  = ETH_SPEED_NUM_10G;
2043 
2044 	if (hw->started == 0) {
2045 		link.link_status = ETH_LINK_DOWN;
2046 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2047 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2048 		vtpci_read_dev_config(hw,
2049 				offsetof(struct virtio_net_config, status),
2050 				&status, sizeof(status));
2051 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2052 			link.link_status = ETH_LINK_DOWN;
2053 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2054 				     dev->data->port_id);
2055 		} else {
2056 			link.link_status = ETH_LINK_UP;
2057 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2058 				     dev->data->port_id);
2059 		}
2060 	} else {
2061 		link.link_status = ETH_LINK_UP;
2062 	}
2063 	virtio_dev_atomic_write_link_status(dev, &link);
2064 
2065 	return (old.link_status == link.link_status) ? -1 : 0;
2066 }
2067 
2068 static int
2069 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2070 {
2071 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2072 	struct virtio_hw *hw = dev->data->dev_private;
2073 
2074 	if (mask & ETH_VLAN_FILTER_MASK) {
2075 		if (rxmode->hw_vlan_filter &&
2076 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2077 
2078 			PMD_DRV_LOG(NOTICE,
2079 				"vlan filtering not available on this host");
2080 
2081 			return -ENOTSUP;
2082 		}
2083 	}
2084 
2085 	if (mask & ETH_VLAN_STRIP_MASK)
2086 		hw->vlan_strip = rxmode->hw_vlan_strip;
2087 
2088 	return 0;
2089 }
2090 
2091 static void
2092 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2093 {
2094 	uint64_t tso_mask, host_features;
2095 	struct virtio_hw *hw = dev->data->dev_private;
2096 
2097 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2098 
2099 	dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
2100 	dev_info->max_rx_queues =
2101 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2102 	dev_info->max_tx_queues =
2103 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2104 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2105 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2106 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2107 	dev_info->default_txconf = (struct rte_eth_txconf) {
2108 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
2109 	};
2110 
2111 	host_features = VTPCI_OPS(hw)->get_features(hw);
2112 	dev_info->rx_offload_capa = 0;
2113 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2114 		dev_info->rx_offload_capa |=
2115 			DEV_RX_OFFLOAD_TCP_CKSUM |
2116 			DEV_RX_OFFLOAD_UDP_CKSUM;
2117 	}
2118 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2119 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2120 	if ((host_features & tso_mask) == tso_mask)
2121 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2122 
2123 	dev_info->tx_offload_capa = 0;
2124 	if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2125 		dev_info->tx_offload_capa |=
2126 			DEV_TX_OFFLOAD_UDP_CKSUM |
2127 			DEV_TX_OFFLOAD_TCP_CKSUM;
2128 	}
2129 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2130 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2131 	if ((hw->guest_features & tso_mask) == tso_mask)
2132 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2133 }
2134 
2135 /*
2136  * It enables testpmd to collect per queue stats.
2137  */
2138 static int
2139 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2140 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2141 __rte_unused uint8_t is_rx)
2142 {
2143 	return 0;
2144 }
2145 
2146 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2147 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2148 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2149 
2150 RTE_INIT(virtio_init_log);
2151 static void
2152 virtio_init_log(void)
2153 {
2154 	virtio_logtype_init = rte_log_register("pmd.virtio.init");
2155 	if (virtio_logtype_init >= 0)
2156 		rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2157 	virtio_logtype_driver = rte_log_register("pmd.virtio.driver");
2158 	if (virtio_logtype_driver >= 0)
2159 		rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2160 }
2161