xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 8599ed31f2cfad122f6a2096093193f9793f876c)
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 
32 #include "virtio_ethdev.h"
33 #include "virtio_pci.h"
34 #include "virtio_logs.h"
35 #include "virtqueue.h"
36 #include "virtio_rxtx.h"
37 
38 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
39 static int  virtio_dev_configure(struct rte_eth_dev *dev);
40 static int  virtio_dev_start(struct rte_eth_dev *dev);
41 static void virtio_dev_stop(struct rte_eth_dev *dev);
42 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
43 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
44 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
45 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
46 static void virtio_dev_info_get(struct rte_eth_dev *dev,
47 				struct rte_eth_dev_info *dev_info);
48 static int virtio_dev_link_update(struct rte_eth_dev *dev,
49 	int wait_to_complete);
50 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
51 
52 static void virtio_set_hwaddr(struct virtio_hw *hw);
53 static void virtio_get_hwaddr(struct virtio_hw *hw);
54 
55 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
56 				 struct rte_eth_stats *stats);
57 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
58 				 struct rte_eth_xstat *xstats, unsigned n);
59 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
60 				       struct rte_eth_xstat_name *xstats_names,
61 				       unsigned limit);
62 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
63 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
64 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
65 				uint16_t vlan_id, int on);
66 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
67 				struct ether_addr *mac_addr,
68 				uint32_t index, uint32_t vmdq);
69 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
70 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
71 				struct ether_addr *mac_addr);
72 
73 static int virtio_intr_enable(struct rte_eth_dev *dev);
74 static int virtio_intr_disable(struct rte_eth_dev *dev);
75 
76 static int virtio_dev_queue_stats_mapping_set(
77 	struct rte_eth_dev *eth_dev,
78 	uint16_t queue_id,
79 	uint8_t stat_idx,
80 	uint8_t is_rx);
81 
82 int virtio_logtype_init;
83 int virtio_logtype_driver;
84 
85 static void virtio_notify_peers(struct rte_eth_dev *dev);
86 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
87 
88 /*
89  * The set of PCI devices this driver supports
90  */
91 static const struct rte_pci_id pci_id_virtio_map[] = {
92 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
93 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
94 	{ .vendor_id = 0, /* sentinel */ },
95 };
96 
97 struct rte_virtio_xstats_name_off {
98 	char name[RTE_ETH_XSTATS_NAME_SIZE];
99 	unsigned offset;
100 };
101 
102 /* [rt]x_qX_ is prepended to the name string here */
103 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
104 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
105 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
106 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
107 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
108 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
109 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
110 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
111 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
112 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
113 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
114 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
115 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
116 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
117 };
118 
119 /* [rt]x_qX_ is prepended to the name string here */
120 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
121 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
122 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
123 	{"errors",                 offsetof(struct virtnet_tx, stats.errors)},
124 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
125 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
126 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
127 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
128 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
129 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
130 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
131 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
132 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
133 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
134 };
135 
136 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
137 			    sizeof(rte_virtio_rxq_stat_strings[0]))
138 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
139 			    sizeof(rte_virtio_txq_stat_strings[0]))
140 
141 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
142 
143 static int
144 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
145 		int *dlen, int pkt_num)
146 {
147 	uint32_t head, i;
148 	int k, sum = 0;
149 	virtio_net_ctrl_ack status = ~0;
150 	struct virtio_pmd_ctrl *result;
151 	struct virtqueue *vq;
152 
153 	ctrl->status = status;
154 
155 	if (!cvq || !cvq->vq) {
156 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
157 		return -1;
158 	}
159 
160 	rte_spinlock_lock(&cvq->lock);
161 	vq = cvq->vq;
162 	head = vq->vq_desc_head_idx;
163 
164 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
165 		"vq->hw->cvq = %p vq = %p",
166 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
167 
168 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
169 		rte_spinlock_unlock(&cvq->lock);
170 		return -1;
171 	}
172 
173 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
174 		sizeof(struct virtio_pmd_ctrl));
175 
176 	/*
177 	 * Format is enforced in qemu code:
178 	 * One TX packet for header;
179 	 * At least one TX packet per argument;
180 	 * One RX packet for ACK.
181 	 */
182 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
183 	vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
184 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
185 	vq->vq_free_cnt--;
186 	i = vq->vq_ring.desc[head].next;
187 
188 	for (k = 0; k < pkt_num; k++) {
189 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
190 		vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
191 			+ sizeof(struct virtio_net_ctrl_hdr)
192 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
193 		vq->vq_ring.desc[i].len = dlen[k];
194 		sum += dlen[k];
195 		vq->vq_free_cnt--;
196 		i = vq->vq_ring.desc[i].next;
197 	}
198 
199 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
200 	vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
201 			+ sizeof(struct virtio_net_ctrl_hdr);
202 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
203 	vq->vq_free_cnt--;
204 
205 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
206 
207 	vq_update_avail_ring(vq, head);
208 	vq_update_avail_idx(vq);
209 
210 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
211 
212 	virtqueue_notify(vq);
213 
214 	rte_rmb();
215 	while (VIRTQUEUE_NUSED(vq) == 0) {
216 		rte_rmb();
217 		usleep(100);
218 	}
219 
220 	while (VIRTQUEUE_NUSED(vq)) {
221 		uint32_t idx, desc_idx, used_idx;
222 		struct vring_used_elem *uep;
223 
224 		used_idx = (uint32_t)(vq->vq_used_cons_idx
225 				& (vq->vq_nentries - 1));
226 		uep = &vq->vq_ring.used->ring[used_idx];
227 		idx = (uint32_t) uep->id;
228 		desc_idx = idx;
229 
230 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
231 			desc_idx = vq->vq_ring.desc[desc_idx].next;
232 			vq->vq_free_cnt++;
233 		}
234 
235 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
236 		vq->vq_desc_head_idx = idx;
237 
238 		vq->vq_used_cons_idx++;
239 		vq->vq_free_cnt++;
240 	}
241 
242 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
243 			vq->vq_free_cnt, vq->vq_desc_head_idx);
244 
245 	result = cvq->virtio_net_hdr_mz->addr;
246 
247 	rte_spinlock_unlock(&cvq->lock);
248 	return result->status;
249 }
250 
251 static int
252 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
253 {
254 	struct virtio_hw *hw = dev->data->dev_private;
255 	struct virtio_pmd_ctrl ctrl;
256 	int dlen[1];
257 	int ret;
258 
259 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
260 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
261 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
262 
263 	dlen[0] = sizeof(uint16_t);
264 
265 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
266 	if (ret) {
267 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
268 			  "failed, this is too late now...");
269 		return -EINVAL;
270 	}
271 
272 	return 0;
273 }
274 
275 static void
276 virtio_dev_queue_release(void *queue __rte_unused)
277 {
278 	/* do nothing */
279 }
280 
281 static uint16_t
282 virtio_get_nr_vq(struct virtio_hw *hw)
283 {
284 	uint16_t nr_vq = hw->max_queue_pairs * 2;
285 
286 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
287 		nr_vq += 1;
288 
289 	return nr_vq;
290 }
291 
292 static void
293 virtio_init_vring(struct virtqueue *vq)
294 {
295 	int size = vq->vq_nentries;
296 	struct vring *vr = &vq->vq_ring;
297 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
298 
299 	PMD_INIT_FUNC_TRACE();
300 
301 	/*
302 	 * Reinitialise since virtio port might have been stopped and restarted
303 	 */
304 	memset(ring_mem, 0, vq->vq_ring_size);
305 	vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
306 	vq->vq_used_cons_idx = 0;
307 	vq->vq_desc_head_idx = 0;
308 	vq->vq_avail_idx = 0;
309 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
310 	vq->vq_free_cnt = vq->vq_nentries;
311 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
312 
313 	vring_desc_init(vr->desc, size);
314 
315 	/*
316 	 * Disable device(host) interrupting guest
317 	 */
318 	virtqueue_disable_intr(vq);
319 }
320 
321 static int
322 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
323 {
324 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
325 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
326 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
327 	unsigned int vq_size, size;
328 	struct virtio_hw *hw = dev->data->dev_private;
329 	struct virtnet_rx *rxvq = NULL;
330 	struct virtnet_tx *txvq = NULL;
331 	struct virtnet_ctl *cvq = NULL;
332 	struct virtqueue *vq;
333 	size_t sz_hdr_mz = 0;
334 	void *sw_ring = NULL;
335 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
336 	int ret;
337 
338 	PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
339 
340 	/*
341 	 * Read the virtqueue size from the Queue Size field
342 	 * Always power of 2 and if 0 virtqueue does not exist
343 	 */
344 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
345 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
346 	if (vq_size == 0) {
347 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
348 		return -EINVAL;
349 	}
350 
351 	if (!rte_is_power_of_2(vq_size)) {
352 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
353 		return -EINVAL;
354 	}
355 
356 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
357 		 dev->data->port_id, vtpci_queue_idx);
358 
359 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
360 				vq_size * sizeof(struct vq_desc_extra),
361 				RTE_CACHE_LINE_SIZE);
362 	if (queue_type == VTNET_TQ) {
363 		/*
364 		 * For each xmit packet, allocate a virtio_net_hdr
365 		 * and indirect ring elements
366 		 */
367 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
368 	} else if (queue_type == VTNET_CQ) {
369 		/* Allocate a page for control vq command, data and status */
370 		sz_hdr_mz = PAGE_SIZE;
371 	}
372 
373 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
374 				SOCKET_ID_ANY);
375 	if (vq == NULL) {
376 		PMD_INIT_LOG(ERR, "can not allocate vq");
377 		return -ENOMEM;
378 	}
379 	hw->vqs[vtpci_queue_idx] = vq;
380 
381 	vq->hw = hw;
382 	vq->vq_queue_index = vtpci_queue_idx;
383 	vq->vq_nentries = vq_size;
384 
385 	/*
386 	 * Reserve a memzone for vring elements
387 	 */
388 	size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
389 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
390 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
391 		     size, vq->vq_ring_size);
392 
393 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
394 			SOCKET_ID_ANY, RTE_MEMZONE_IOVA_CONTIG,
395 			VIRTIO_PCI_VRING_ALIGN);
396 	if (mz == NULL) {
397 		if (rte_errno == EEXIST)
398 			mz = rte_memzone_lookup(vq_name);
399 		if (mz == NULL) {
400 			ret = -ENOMEM;
401 			goto fail_q_alloc;
402 		}
403 	}
404 
405 	memset(mz->addr, 0, mz->len);
406 
407 	vq->vq_ring_mem = mz->iova;
408 	vq->vq_ring_virt_mem = mz->addr;
409 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
410 		     (uint64_t)mz->iova);
411 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
412 		     (uint64_t)(uintptr_t)mz->addr);
413 
414 	virtio_init_vring(vq);
415 
416 	if (sz_hdr_mz) {
417 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
418 			 dev->data->port_id, vtpci_queue_idx);
419 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
420 				SOCKET_ID_ANY, RTE_MEMZONE_IOVA_CONTIG,
421 				RTE_CACHE_LINE_SIZE);
422 		if (hdr_mz == NULL) {
423 			if (rte_errno == EEXIST)
424 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
425 			if (hdr_mz == NULL) {
426 				ret = -ENOMEM;
427 				goto fail_q_alloc;
428 			}
429 		}
430 	}
431 
432 	if (queue_type == VTNET_RQ) {
433 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
434 			       sizeof(vq->sw_ring[0]);
435 
436 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
437 				RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
438 		if (!sw_ring) {
439 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
440 			ret = -ENOMEM;
441 			goto fail_q_alloc;
442 		}
443 
444 		vq->sw_ring = sw_ring;
445 		rxvq = &vq->rxq;
446 		rxvq->vq = vq;
447 		rxvq->port_id = dev->data->port_id;
448 		rxvq->mz = mz;
449 	} else if (queue_type == VTNET_TQ) {
450 		txvq = &vq->txq;
451 		txvq->vq = vq;
452 		txvq->port_id = dev->data->port_id;
453 		txvq->mz = mz;
454 		txvq->virtio_net_hdr_mz = hdr_mz;
455 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
456 	} else if (queue_type == VTNET_CQ) {
457 		cvq = &vq->cq;
458 		cvq->vq = vq;
459 		cvq->mz = mz;
460 		cvq->virtio_net_hdr_mz = hdr_mz;
461 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
462 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
463 
464 		hw->cvq = cvq;
465 	}
466 
467 	/* For virtio_user case (that is when hw->dev is NULL), we use
468 	 * virtual address. And we need properly set _offset_, please see
469 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
470 	 */
471 	if (!hw->virtio_user_dev)
472 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
473 	else {
474 		vq->vq_ring_mem = (uintptr_t)mz->addr;
475 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
476 		if (queue_type == VTNET_TQ)
477 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
478 		else if (queue_type == VTNET_CQ)
479 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
480 	}
481 
482 	if (queue_type == VTNET_TQ) {
483 		struct virtio_tx_region *txr;
484 		unsigned int i;
485 
486 		txr = hdr_mz->addr;
487 		memset(txr, 0, vq_size * sizeof(*txr));
488 		for (i = 0; i < vq_size; i++) {
489 			struct vring_desc *start_dp = txr[i].tx_indir;
490 
491 			vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
492 
493 			/* first indirect descriptor is always the tx header */
494 			start_dp->addr = txvq->virtio_net_hdr_mem
495 				+ i * sizeof(*txr)
496 				+ offsetof(struct virtio_tx_region, tx_hdr);
497 
498 			start_dp->len = hw->vtnet_hdr_size;
499 			start_dp->flags = VRING_DESC_F_NEXT;
500 		}
501 	}
502 
503 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
504 		PMD_INIT_LOG(ERR, "setup_queue failed");
505 		return -EINVAL;
506 	}
507 
508 	return 0;
509 
510 fail_q_alloc:
511 	rte_free(sw_ring);
512 	rte_memzone_free(hdr_mz);
513 	rte_memzone_free(mz);
514 	rte_free(vq);
515 
516 	return ret;
517 }
518 
519 static void
520 virtio_free_queues(struct virtio_hw *hw)
521 {
522 	uint16_t nr_vq = virtio_get_nr_vq(hw);
523 	struct virtqueue *vq;
524 	int queue_type;
525 	uint16_t i;
526 
527 	if (hw->vqs == NULL)
528 		return;
529 
530 	for (i = 0; i < nr_vq; i++) {
531 		vq = hw->vqs[i];
532 		if (!vq)
533 			continue;
534 
535 		queue_type = virtio_get_queue_type(hw, i);
536 		if (queue_type == VTNET_RQ) {
537 			rte_free(vq->sw_ring);
538 			rte_memzone_free(vq->rxq.mz);
539 		} else if (queue_type == VTNET_TQ) {
540 			rte_memzone_free(vq->txq.mz);
541 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
542 		} else {
543 			rte_memzone_free(vq->cq.mz);
544 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
545 		}
546 
547 		rte_free(vq);
548 		hw->vqs[i] = NULL;
549 	}
550 
551 	rte_free(hw->vqs);
552 	hw->vqs = NULL;
553 }
554 
555 static int
556 virtio_alloc_queues(struct rte_eth_dev *dev)
557 {
558 	struct virtio_hw *hw = dev->data->dev_private;
559 	uint16_t nr_vq = virtio_get_nr_vq(hw);
560 	uint16_t i;
561 	int ret;
562 
563 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
564 	if (!hw->vqs) {
565 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
566 		return -ENOMEM;
567 	}
568 
569 	for (i = 0; i < nr_vq; i++) {
570 		ret = virtio_init_queue(dev, i);
571 		if (ret < 0) {
572 			virtio_free_queues(hw);
573 			return ret;
574 		}
575 	}
576 
577 	return 0;
578 }
579 
580 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
581 
582 static void
583 virtio_dev_close(struct rte_eth_dev *dev)
584 {
585 	struct virtio_hw *hw = dev->data->dev_private;
586 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
587 
588 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
589 
590 	/* reset the NIC */
591 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
592 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
593 	if (intr_conf->rxq)
594 		virtio_queues_unbind_intr(dev);
595 
596 	if (intr_conf->lsc || intr_conf->rxq) {
597 		virtio_intr_disable(dev);
598 		rte_intr_efd_disable(dev->intr_handle);
599 		rte_free(dev->intr_handle->intr_vec);
600 		dev->intr_handle->intr_vec = NULL;
601 	}
602 
603 	vtpci_reset(hw);
604 	virtio_dev_free_mbufs(dev);
605 	virtio_free_queues(hw);
606 }
607 
608 static void
609 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
610 {
611 	struct virtio_hw *hw = dev->data->dev_private;
612 	struct virtio_pmd_ctrl ctrl;
613 	int dlen[1];
614 	int ret;
615 
616 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
617 		PMD_INIT_LOG(INFO, "host does not support rx control");
618 		return;
619 	}
620 
621 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
622 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
623 	ctrl.data[0] = 1;
624 	dlen[0] = 1;
625 
626 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
627 	if (ret)
628 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
629 }
630 
631 static void
632 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
633 {
634 	struct virtio_hw *hw = dev->data->dev_private;
635 	struct virtio_pmd_ctrl ctrl;
636 	int dlen[1];
637 	int ret;
638 
639 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
640 		PMD_INIT_LOG(INFO, "host does not support rx control");
641 		return;
642 	}
643 
644 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
645 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
646 	ctrl.data[0] = 0;
647 	dlen[0] = 1;
648 
649 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
650 	if (ret)
651 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
652 }
653 
654 static void
655 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
656 {
657 	struct virtio_hw *hw = dev->data->dev_private;
658 	struct virtio_pmd_ctrl ctrl;
659 	int dlen[1];
660 	int ret;
661 
662 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
663 		PMD_INIT_LOG(INFO, "host does not support rx control");
664 		return;
665 	}
666 
667 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
668 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
669 	ctrl.data[0] = 1;
670 	dlen[0] = 1;
671 
672 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
673 	if (ret)
674 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
675 }
676 
677 static void
678 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
679 {
680 	struct virtio_hw *hw = dev->data->dev_private;
681 	struct virtio_pmd_ctrl ctrl;
682 	int dlen[1];
683 	int ret;
684 
685 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
686 		PMD_INIT_LOG(INFO, "host does not support rx control");
687 		return;
688 	}
689 
690 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
691 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
692 	ctrl.data[0] = 0;
693 	dlen[0] = 1;
694 
695 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
696 	if (ret)
697 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
698 }
699 
700 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
701 static int
702 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
703 {
704 	struct virtio_hw *hw = dev->data->dev_private;
705 	uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
706 				 hw->vtnet_hdr_size;
707 	uint32_t frame_size = mtu + ether_hdr_len;
708 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
709 
710 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
711 
712 	if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
713 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
714 			ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
715 		return -EINVAL;
716 	}
717 	return 0;
718 }
719 
720 static int
721 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
722 {
723 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
724 	struct virtqueue *vq = rxvq->vq;
725 
726 	virtqueue_enable_intr(vq);
727 	return 0;
728 }
729 
730 static int
731 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
732 {
733 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
734 	struct virtqueue *vq = rxvq->vq;
735 
736 	virtqueue_disable_intr(vq);
737 	return 0;
738 }
739 
740 /*
741  * dev_ops for virtio, bare necessities for basic operation
742  */
743 static const struct eth_dev_ops virtio_eth_dev_ops = {
744 	.dev_configure           = virtio_dev_configure,
745 	.dev_start               = virtio_dev_start,
746 	.dev_stop                = virtio_dev_stop,
747 	.dev_close               = virtio_dev_close,
748 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
749 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
750 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
751 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
752 	.mtu_set                 = virtio_mtu_set,
753 	.dev_infos_get           = virtio_dev_info_get,
754 	.stats_get               = virtio_dev_stats_get,
755 	.xstats_get              = virtio_dev_xstats_get,
756 	.xstats_get_names        = virtio_dev_xstats_get_names,
757 	.stats_reset             = virtio_dev_stats_reset,
758 	.xstats_reset            = virtio_dev_stats_reset,
759 	.link_update             = virtio_dev_link_update,
760 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
761 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
762 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
763 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
764 	.rx_queue_release        = virtio_dev_queue_release,
765 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
766 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
767 	.tx_queue_release        = virtio_dev_queue_release,
768 	/* collect stats per queue */
769 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
770 	.vlan_filter_set         = virtio_vlan_filter_set,
771 	.mac_addr_add            = virtio_mac_addr_add,
772 	.mac_addr_remove         = virtio_mac_addr_remove,
773 	.mac_addr_set            = virtio_mac_addr_set,
774 };
775 
776 static void
777 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
778 {
779 	unsigned i;
780 
781 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
782 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
783 		if (txvq == NULL)
784 			continue;
785 
786 		stats->opackets += txvq->stats.packets;
787 		stats->obytes += txvq->stats.bytes;
788 		stats->oerrors += txvq->stats.errors;
789 
790 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
791 			stats->q_opackets[i] = txvq->stats.packets;
792 			stats->q_obytes[i] = txvq->stats.bytes;
793 		}
794 	}
795 
796 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
797 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
798 		if (rxvq == NULL)
799 			continue;
800 
801 		stats->ipackets += rxvq->stats.packets;
802 		stats->ibytes += rxvq->stats.bytes;
803 		stats->ierrors += rxvq->stats.errors;
804 
805 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
806 			stats->q_ipackets[i] = rxvq->stats.packets;
807 			stats->q_ibytes[i] = rxvq->stats.bytes;
808 		}
809 	}
810 
811 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
812 }
813 
814 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
815 				       struct rte_eth_xstat_name *xstats_names,
816 				       __rte_unused unsigned limit)
817 {
818 	unsigned i;
819 	unsigned count = 0;
820 	unsigned t;
821 
822 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
823 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
824 
825 	if (xstats_names != NULL) {
826 		/* Note: limit checked in rte_eth_xstats_names() */
827 
828 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
829 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
830 			if (rxvq == NULL)
831 				continue;
832 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
833 				snprintf(xstats_names[count].name,
834 					sizeof(xstats_names[count].name),
835 					"rx_q%u_%s", i,
836 					rte_virtio_rxq_stat_strings[t].name);
837 				count++;
838 			}
839 		}
840 
841 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
842 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
843 			if (txvq == NULL)
844 				continue;
845 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
846 				snprintf(xstats_names[count].name,
847 					sizeof(xstats_names[count].name),
848 					"tx_q%u_%s", i,
849 					rte_virtio_txq_stat_strings[t].name);
850 				count++;
851 			}
852 		}
853 		return count;
854 	}
855 	return nstats;
856 }
857 
858 static int
859 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
860 		      unsigned n)
861 {
862 	unsigned i;
863 	unsigned count = 0;
864 
865 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
866 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
867 
868 	if (n < nstats)
869 		return nstats;
870 
871 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
872 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
873 
874 		if (rxvq == NULL)
875 			continue;
876 
877 		unsigned t;
878 
879 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
880 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
881 				rte_virtio_rxq_stat_strings[t].offset);
882 			xstats[count].id = count;
883 			count++;
884 		}
885 	}
886 
887 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
888 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
889 
890 		if (txvq == NULL)
891 			continue;
892 
893 		unsigned t;
894 
895 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
896 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
897 				rte_virtio_txq_stat_strings[t].offset);
898 			xstats[count].id = count;
899 			count++;
900 		}
901 	}
902 
903 	return count;
904 }
905 
906 static int
907 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
908 {
909 	virtio_update_stats(dev, stats);
910 
911 	return 0;
912 }
913 
914 static void
915 virtio_dev_stats_reset(struct rte_eth_dev *dev)
916 {
917 	unsigned int i;
918 
919 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
920 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
921 		if (txvq == NULL)
922 			continue;
923 
924 		txvq->stats.packets = 0;
925 		txvq->stats.bytes = 0;
926 		txvq->stats.errors = 0;
927 		txvq->stats.multicast = 0;
928 		txvq->stats.broadcast = 0;
929 		memset(txvq->stats.size_bins, 0,
930 		       sizeof(txvq->stats.size_bins[0]) * 8);
931 	}
932 
933 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
934 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
935 		if (rxvq == NULL)
936 			continue;
937 
938 		rxvq->stats.packets = 0;
939 		rxvq->stats.bytes = 0;
940 		rxvq->stats.errors = 0;
941 		rxvq->stats.multicast = 0;
942 		rxvq->stats.broadcast = 0;
943 		memset(rxvq->stats.size_bins, 0,
944 		       sizeof(rxvq->stats.size_bins[0]) * 8);
945 	}
946 }
947 
948 static void
949 virtio_set_hwaddr(struct virtio_hw *hw)
950 {
951 	vtpci_write_dev_config(hw,
952 			offsetof(struct virtio_net_config, mac),
953 			&hw->mac_addr, ETHER_ADDR_LEN);
954 }
955 
956 static void
957 virtio_get_hwaddr(struct virtio_hw *hw)
958 {
959 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
960 		vtpci_read_dev_config(hw,
961 			offsetof(struct virtio_net_config, mac),
962 			&hw->mac_addr, ETHER_ADDR_LEN);
963 	} else {
964 		eth_random_addr(&hw->mac_addr[0]);
965 		virtio_set_hwaddr(hw);
966 	}
967 }
968 
969 static int
970 virtio_mac_table_set(struct virtio_hw *hw,
971 		     const struct virtio_net_ctrl_mac *uc,
972 		     const struct virtio_net_ctrl_mac *mc)
973 {
974 	struct virtio_pmd_ctrl ctrl;
975 	int err, len[2];
976 
977 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
978 		PMD_DRV_LOG(INFO, "host does not support mac table");
979 		return -1;
980 	}
981 
982 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
983 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
984 
985 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
986 	memcpy(ctrl.data, uc, len[0]);
987 
988 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
989 	memcpy(ctrl.data + len[0], mc, len[1]);
990 
991 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
992 	if (err != 0)
993 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
994 	return err;
995 }
996 
997 static int
998 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
999 		    uint32_t index, uint32_t vmdq __rte_unused)
1000 {
1001 	struct virtio_hw *hw = dev->data->dev_private;
1002 	const struct ether_addr *addrs = dev->data->mac_addrs;
1003 	unsigned int i;
1004 	struct virtio_net_ctrl_mac *uc, *mc;
1005 
1006 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1007 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1008 		return -EINVAL;
1009 	}
1010 
1011 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1012 	uc->entries = 0;
1013 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1014 	mc->entries = 0;
1015 
1016 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1017 		const struct ether_addr *addr
1018 			= (i == index) ? mac_addr : addrs + i;
1019 		struct virtio_net_ctrl_mac *tbl
1020 			= is_multicast_ether_addr(addr) ? mc : uc;
1021 
1022 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1023 	}
1024 
1025 	return virtio_mac_table_set(hw, uc, mc);
1026 }
1027 
1028 static void
1029 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1030 {
1031 	struct virtio_hw *hw = dev->data->dev_private;
1032 	struct ether_addr *addrs = dev->data->mac_addrs;
1033 	struct virtio_net_ctrl_mac *uc, *mc;
1034 	unsigned int i;
1035 
1036 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1037 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1038 		return;
1039 	}
1040 
1041 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1042 	uc->entries = 0;
1043 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1044 	mc->entries = 0;
1045 
1046 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1047 		struct virtio_net_ctrl_mac *tbl;
1048 
1049 		if (i == index || is_zero_ether_addr(addrs + i))
1050 			continue;
1051 
1052 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1053 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1054 	}
1055 
1056 	virtio_mac_table_set(hw, uc, mc);
1057 }
1058 
1059 static void
1060 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1061 {
1062 	struct virtio_hw *hw = dev->data->dev_private;
1063 
1064 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1065 
1066 	/* Use atomic update if available */
1067 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1068 		struct virtio_pmd_ctrl ctrl;
1069 		int len = ETHER_ADDR_LEN;
1070 
1071 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1072 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1073 
1074 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1075 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
1076 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1077 		virtio_set_hwaddr(hw);
1078 }
1079 
1080 static int
1081 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1082 {
1083 	struct virtio_hw *hw = dev->data->dev_private;
1084 	struct virtio_pmd_ctrl ctrl;
1085 	int len;
1086 
1087 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1088 		return -ENOTSUP;
1089 
1090 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1091 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1092 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1093 	len = sizeof(vlan_id);
1094 
1095 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1096 }
1097 
1098 static int
1099 virtio_intr_enable(struct rte_eth_dev *dev)
1100 {
1101 	struct virtio_hw *hw = dev->data->dev_private;
1102 
1103 	if (rte_intr_enable(dev->intr_handle) < 0)
1104 		return -1;
1105 
1106 	if (!hw->virtio_user_dev)
1107 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1108 
1109 	return 0;
1110 }
1111 
1112 static int
1113 virtio_intr_disable(struct rte_eth_dev *dev)
1114 {
1115 	struct virtio_hw *hw = dev->data->dev_private;
1116 
1117 	if (rte_intr_disable(dev->intr_handle) < 0)
1118 		return -1;
1119 
1120 	if (!hw->virtio_user_dev)
1121 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1122 
1123 	return 0;
1124 }
1125 
1126 static int
1127 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1128 {
1129 	uint64_t host_features;
1130 
1131 	/* Prepare guest_features: feature that driver wants to support */
1132 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1133 		req_features);
1134 
1135 	/* Read device(host) feature bits */
1136 	host_features = VTPCI_OPS(hw)->get_features(hw);
1137 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1138 		host_features);
1139 
1140 	/* If supported, ensure MTU value is valid before acknowledging it. */
1141 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1142 		struct virtio_net_config config;
1143 
1144 		vtpci_read_dev_config(hw,
1145 			offsetof(struct virtio_net_config, mtu),
1146 			&config.mtu, sizeof(config.mtu));
1147 
1148 		if (config.mtu < ETHER_MIN_MTU)
1149 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1150 	}
1151 
1152 	/*
1153 	 * Negotiate features: Subset of device feature bits are written back
1154 	 * guest feature bits.
1155 	 */
1156 	hw->guest_features = req_features;
1157 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1158 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1159 		hw->guest_features);
1160 
1161 	if (hw->modern) {
1162 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1163 			PMD_INIT_LOG(ERR,
1164 				"VIRTIO_F_VERSION_1 features is not enabled.");
1165 			return -1;
1166 		}
1167 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1168 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1169 			PMD_INIT_LOG(ERR,
1170 				"failed to set FEATURES_OK status!");
1171 			return -1;
1172 		}
1173 	}
1174 
1175 	hw->req_guest_features = req_features;
1176 
1177 	return 0;
1178 }
1179 
1180 int
1181 virtio_dev_pause(struct rte_eth_dev *dev)
1182 {
1183 	struct virtio_hw *hw = dev->data->dev_private;
1184 
1185 	rte_spinlock_lock(&hw->state_lock);
1186 
1187 	if (hw->started == 0) {
1188 		/* Device is just stopped. */
1189 		rte_spinlock_unlock(&hw->state_lock);
1190 		return -1;
1191 	}
1192 	hw->started = 0;
1193 	/*
1194 	 * Prevent the worker threads from touching queues to avoid contention,
1195 	 * 1 ms should be enough for the ongoing Tx function to finish.
1196 	 */
1197 	rte_delay_ms(1);
1198 	return 0;
1199 }
1200 
1201 /*
1202  * Recover hw state to let the worker threads continue.
1203  */
1204 void
1205 virtio_dev_resume(struct rte_eth_dev *dev)
1206 {
1207 	struct virtio_hw *hw = dev->data->dev_private;
1208 
1209 	hw->started = 1;
1210 	rte_spinlock_unlock(&hw->state_lock);
1211 }
1212 
1213 /*
1214  * Should be called only after device is paused.
1215  */
1216 int
1217 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1218 		int nb_pkts)
1219 {
1220 	struct virtio_hw *hw = dev->data->dev_private;
1221 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1222 	int ret;
1223 
1224 	hw->inject_pkts = tx_pkts;
1225 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1226 	hw->inject_pkts = NULL;
1227 
1228 	return ret;
1229 }
1230 
1231 static void
1232 virtio_notify_peers(struct rte_eth_dev *dev)
1233 {
1234 	struct virtio_hw *hw = dev->data->dev_private;
1235 	struct virtnet_rx *rxvq;
1236 	struct rte_mbuf *rarp_mbuf;
1237 
1238 	if (!dev->data->rx_queues)
1239 		return;
1240 
1241 	rxvq = dev->data->rx_queues[0];
1242 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1243 			(struct ether_addr *)hw->mac_addr);
1244 	if (rarp_mbuf == NULL) {
1245 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1246 		return;
1247 	}
1248 
1249 	/* If virtio port just stopped, no need to send RARP */
1250 	if (virtio_dev_pause(dev) < 0) {
1251 		rte_pktmbuf_free(rarp_mbuf);
1252 		return;
1253 	}
1254 
1255 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1256 	virtio_dev_resume(dev);
1257 }
1258 
1259 static void
1260 virtio_ack_link_announce(struct rte_eth_dev *dev)
1261 {
1262 	struct virtio_hw *hw = dev->data->dev_private;
1263 	struct virtio_pmd_ctrl ctrl;
1264 
1265 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1266 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1267 
1268 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1269 }
1270 
1271 /*
1272  * Process virtio config changed interrupt. Call the callback
1273  * if link state changed, generate gratuitous RARP packet if
1274  * the status indicates an ANNOUNCE.
1275  */
1276 void
1277 virtio_interrupt_handler(void *param)
1278 {
1279 	struct rte_eth_dev *dev = param;
1280 	struct virtio_hw *hw = dev->data->dev_private;
1281 	uint8_t isr;
1282 
1283 	/* Read interrupt status which clears interrupt */
1284 	isr = vtpci_isr(hw);
1285 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1286 
1287 	if (virtio_intr_enable(dev) < 0)
1288 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1289 
1290 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1291 		if (virtio_dev_link_update(dev, 0) == 0)
1292 			_rte_eth_dev_callback_process(dev,
1293 						      RTE_ETH_EVENT_INTR_LSC,
1294 						      NULL);
1295 	}
1296 
1297 	if (isr & VIRTIO_NET_S_ANNOUNCE) {
1298 		virtio_notify_peers(dev);
1299 		if (hw->cvq)
1300 			virtio_ack_link_announce(dev);
1301 	}
1302 }
1303 
1304 /* set rx and tx handlers according to what is supported */
1305 static void
1306 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1307 {
1308 	struct virtio_hw *hw = eth_dev->data->dev_private;
1309 
1310 	if (hw->use_simple_rx) {
1311 		PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1312 			eth_dev->data->port_id);
1313 		eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1314 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1315 		PMD_INIT_LOG(INFO,
1316 			"virtio: using mergeable buffer Rx path on port %u",
1317 			eth_dev->data->port_id);
1318 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1319 	} else {
1320 		PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1321 			eth_dev->data->port_id);
1322 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1323 	}
1324 
1325 	if (hw->use_simple_tx) {
1326 		PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1327 			eth_dev->data->port_id);
1328 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1329 	} else {
1330 		PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1331 			eth_dev->data->port_id);
1332 		eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1333 	}
1334 }
1335 
1336 /* Only support 1:1 queue/interrupt mapping so far.
1337  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1338  * interrupt vectors (<N+1).
1339  */
1340 static int
1341 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1342 {
1343 	uint32_t i;
1344 	struct virtio_hw *hw = dev->data->dev_private;
1345 
1346 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1347 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1348 		dev->intr_handle->intr_vec[i] = i + 1;
1349 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1350 						 VIRTIO_MSI_NO_VECTOR) {
1351 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1352 			return -EBUSY;
1353 		}
1354 	}
1355 
1356 	return 0;
1357 }
1358 
1359 static void
1360 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1361 {
1362 	uint32_t i;
1363 	struct virtio_hw *hw = dev->data->dev_private;
1364 
1365 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1366 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1367 		VTPCI_OPS(hw)->set_queue_irq(hw,
1368 					     hw->vqs[i * VTNET_CQ],
1369 					     VIRTIO_MSI_NO_VECTOR);
1370 }
1371 
1372 static int
1373 virtio_configure_intr(struct rte_eth_dev *dev)
1374 {
1375 	struct virtio_hw *hw = dev->data->dev_private;
1376 
1377 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1378 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1379 		return -ENOTSUP;
1380 	}
1381 
1382 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1383 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1384 		return -1;
1385 	}
1386 
1387 	if (!dev->intr_handle->intr_vec) {
1388 		dev->intr_handle->intr_vec =
1389 			rte_zmalloc("intr_vec",
1390 				    hw->max_queue_pairs * sizeof(int), 0);
1391 		if (!dev->intr_handle->intr_vec) {
1392 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1393 				     hw->max_queue_pairs);
1394 			return -ENOMEM;
1395 		}
1396 	}
1397 
1398 	/* Re-register callback to update max_intr */
1399 	rte_intr_callback_unregister(dev->intr_handle,
1400 				     virtio_interrupt_handler,
1401 				     dev);
1402 	rte_intr_callback_register(dev->intr_handle,
1403 				   virtio_interrupt_handler,
1404 				   dev);
1405 
1406 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1407 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1408 	 * And for legacy devices, this should be done before queue/vec binding
1409 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1410 	 * (22) will be ignored.
1411 	 */
1412 	if (virtio_intr_enable(dev) < 0) {
1413 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1414 		return -1;
1415 	}
1416 
1417 	if (virtio_queues_bind_intr(dev) < 0) {
1418 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1419 		return -1;
1420 	}
1421 
1422 	return 0;
1423 }
1424 
1425 /* reset device and renegotiate features if needed */
1426 static int
1427 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1428 {
1429 	struct virtio_hw *hw = eth_dev->data->dev_private;
1430 	struct virtio_net_config *config;
1431 	struct virtio_net_config local_config;
1432 	struct rte_pci_device *pci_dev = NULL;
1433 	int ret;
1434 
1435 	/* Reset the device although not necessary at startup */
1436 	vtpci_reset(hw);
1437 
1438 	if (hw->vqs) {
1439 		virtio_dev_free_mbufs(eth_dev);
1440 		virtio_free_queues(hw);
1441 	}
1442 
1443 	/* Tell the host we've noticed this device. */
1444 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1445 
1446 	/* Tell the host we've known how to drive the device. */
1447 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1448 	if (virtio_negotiate_features(hw, req_features) < 0)
1449 		return -1;
1450 
1451 	if (!hw->virtio_user_dev) {
1452 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1453 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1454 	}
1455 
1456 	/* If host does not support both status and MSI-X then disable LSC */
1457 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1458 	    hw->use_msix != VIRTIO_MSIX_NONE)
1459 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1460 	else
1461 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1462 
1463 	/* Setting up rx_header size for the device */
1464 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1465 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1466 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1467 	else
1468 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1469 
1470 	/* Copy the permanent MAC address to: virtio_hw */
1471 	virtio_get_hwaddr(hw);
1472 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1473 			&eth_dev->data->mac_addrs[0]);
1474 	PMD_INIT_LOG(DEBUG,
1475 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1476 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1477 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1478 
1479 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1480 		config = &local_config;
1481 
1482 		vtpci_read_dev_config(hw,
1483 			offsetof(struct virtio_net_config, mac),
1484 			&config->mac, sizeof(config->mac));
1485 
1486 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1487 			vtpci_read_dev_config(hw,
1488 				offsetof(struct virtio_net_config, status),
1489 				&config->status, sizeof(config->status));
1490 		} else {
1491 			PMD_INIT_LOG(DEBUG,
1492 				     "VIRTIO_NET_F_STATUS is not supported");
1493 			config->status = 0;
1494 		}
1495 
1496 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1497 			vtpci_read_dev_config(hw,
1498 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1499 				&config->max_virtqueue_pairs,
1500 				sizeof(config->max_virtqueue_pairs));
1501 		} else {
1502 			PMD_INIT_LOG(DEBUG,
1503 				     "VIRTIO_NET_F_MQ is not supported");
1504 			config->max_virtqueue_pairs = 1;
1505 		}
1506 
1507 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1508 
1509 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1510 			vtpci_read_dev_config(hw,
1511 				offsetof(struct virtio_net_config, mtu),
1512 				&config->mtu,
1513 				sizeof(config->mtu));
1514 
1515 			/*
1516 			 * MTU value has already been checked at negotiation
1517 			 * time, but check again in case it has changed since
1518 			 * then, which should not happen.
1519 			 */
1520 			if (config->mtu < ETHER_MIN_MTU) {
1521 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1522 						config->mtu);
1523 				return -1;
1524 			}
1525 
1526 			hw->max_mtu = config->mtu;
1527 			/* Set initial MTU to maximum one supported by vhost */
1528 			eth_dev->data->mtu = config->mtu;
1529 
1530 		} else {
1531 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1532 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1533 		}
1534 
1535 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1536 				config->max_virtqueue_pairs);
1537 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1538 		PMD_INIT_LOG(DEBUG,
1539 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1540 				config->mac[0], config->mac[1],
1541 				config->mac[2], config->mac[3],
1542 				config->mac[4], config->mac[5]);
1543 	} else {
1544 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1545 		hw->max_queue_pairs = 1;
1546 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1547 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1548 	}
1549 
1550 	ret = virtio_alloc_queues(eth_dev);
1551 	if (ret < 0)
1552 		return ret;
1553 
1554 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1555 		if (virtio_configure_intr(eth_dev) < 0) {
1556 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1557 			return -1;
1558 		}
1559 	}
1560 
1561 	vtpci_reinit_complete(hw);
1562 
1563 	if (pci_dev)
1564 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1565 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1566 			pci_dev->id.device_id);
1567 
1568 	return 0;
1569 }
1570 
1571 /*
1572  * Remap the PCI device again (IO port map for legacy device and
1573  * memory map for modern device), so that the secondary process
1574  * could have the PCI initiated correctly.
1575  */
1576 static int
1577 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1578 {
1579 	if (hw->modern) {
1580 		/*
1581 		 * We don't have to re-parse the PCI config space, since
1582 		 * rte_pci_map_device() makes sure the mapped address
1583 		 * in secondary process would equal to the one mapped in
1584 		 * the primary process: error will be returned if that
1585 		 * requirement is not met.
1586 		 *
1587 		 * That said, we could simply reuse all cap pointers
1588 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1589 		 * primary process, which is stored in shared memory.
1590 		 */
1591 		if (rte_pci_map_device(pci_dev)) {
1592 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1593 			return -1;
1594 		}
1595 	} else {
1596 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1597 			return -1;
1598 	}
1599 
1600 	return 0;
1601 }
1602 
1603 static void
1604 virtio_set_vtpci_ops(struct virtio_hw *hw)
1605 {
1606 #ifdef RTE_VIRTIO_USER
1607 	if (hw->virtio_user_dev)
1608 		VTPCI_OPS(hw) = &virtio_user_ops;
1609 	else
1610 #endif
1611 	if (hw->modern)
1612 		VTPCI_OPS(hw) = &modern_ops;
1613 	else
1614 		VTPCI_OPS(hw) = &legacy_ops;
1615 }
1616 
1617 /*
1618  * This function is based on probe() function in virtio_pci.c
1619  * It returns 0 on success.
1620  */
1621 int
1622 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1623 {
1624 	struct virtio_hw *hw = eth_dev->data->dev_private;
1625 	int ret;
1626 
1627 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1628 
1629 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1630 
1631 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1632 		if (!hw->virtio_user_dev) {
1633 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1634 			if (ret)
1635 				return ret;
1636 		}
1637 
1638 		virtio_set_vtpci_ops(hw);
1639 		set_rxtx_funcs(eth_dev);
1640 
1641 		return 0;
1642 	}
1643 
1644 	/* Allocate memory for storing MAC addresses */
1645 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1646 	if (eth_dev->data->mac_addrs == NULL) {
1647 		PMD_INIT_LOG(ERR,
1648 			"Failed to allocate %d bytes needed to store MAC addresses",
1649 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1650 		return -ENOMEM;
1651 	}
1652 
1653 	hw->port_id = eth_dev->data->port_id;
1654 	/* For virtio_user case the hw->virtio_user_dev is populated by
1655 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1656 	 */
1657 	if (!hw->virtio_user_dev) {
1658 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1659 		if (ret)
1660 			goto out;
1661 	}
1662 
1663 	/* reset device and negotiate default features */
1664 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1665 	if (ret < 0)
1666 		goto out;
1667 
1668 	/* Setup interrupt callback  */
1669 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1670 		rte_intr_callback_register(eth_dev->intr_handle,
1671 			virtio_interrupt_handler, eth_dev);
1672 
1673 	return 0;
1674 
1675 out:
1676 	rte_free(eth_dev->data->mac_addrs);
1677 	return ret;
1678 }
1679 
1680 static int
1681 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1682 {
1683 	PMD_INIT_FUNC_TRACE();
1684 
1685 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1686 		return -EPERM;
1687 
1688 	virtio_dev_stop(eth_dev);
1689 	virtio_dev_close(eth_dev);
1690 
1691 	eth_dev->dev_ops = NULL;
1692 	eth_dev->tx_pkt_burst = NULL;
1693 	eth_dev->rx_pkt_burst = NULL;
1694 
1695 	rte_free(eth_dev->data->mac_addrs);
1696 	eth_dev->data->mac_addrs = NULL;
1697 
1698 	/* reset interrupt callback  */
1699 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1700 		rte_intr_callback_unregister(eth_dev->intr_handle,
1701 						virtio_interrupt_handler,
1702 						eth_dev);
1703 	if (eth_dev->device)
1704 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1705 
1706 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1707 
1708 	return 0;
1709 }
1710 
1711 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1712 	struct rte_pci_device *pci_dev)
1713 {
1714 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1715 		eth_virtio_dev_init);
1716 }
1717 
1718 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1719 {
1720 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1721 }
1722 
1723 static struct rte_pci_driver rte_virtio_pmd = {
1724 	.driver = {
1725 		.name = "net_virtio",
1726 	},
1727 	.id_table = pci_id_virtio_map,
1728 	.drv_flags = 0,
1729 	.probe = eth_virtio_pci_probe,
1730 	.remove = eth_virtio_pci_remove,
1731 };
1732 
1733 RTE_INIT(rte_virtio_pmd_init);
1734 static void
1735 rte_virtio_pmd_init(void)
1736 {
1737 	if (rte_eal_iopl_init() != 0) {
1738 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1739 		return;
1740 	}
1741 
1742 	rte_pci_register(&rte_virtio_pmd);
1743 }
1744 
1745 /*
1746  * Configure virtio device
1747  * It returns 0 on success.
1748  */
1749 static int
1750 virtio_dev_configure(struct rte_eth_dev *dev)
1751 {
1752 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1753 	struct virtio_hw *hw = dev->data->dev_private;
1754 	uint64_t rx_offloads = rxmode->offloads;
1755 	uint64_t req_features;
1756 	int ret;
1757 
1758 	PMD_INIT_LOG(DEBUG, "configure");
1759 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1760 
1761 	if (dev->data->dev_conf.intr_conf.rxq) {
1762 		ret = virtio_init_device(dev, hw->req_guest_features);
1763 		if (ret < 0)
1764 			return ret;
1765 	}
1766 
1767 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
1768 			   DEV_RX_OFFLOAD_TCP_CKSUM))
1769 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1770 
1771 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
1772 		req_features |=
1773 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1774 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1775 
1776 	/* if request features changed, reinit the device */
1777 	if (req_features != hw->req_guest_features) {
1778 		ret = virtio_init_device(dev, req_features);
1779 		if (ret < 0)
1780 			return ret;
1781 	}
1782 
1783 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
1784 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
1785 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1786 		PMD_DRV_LOG(ERR,
1787 			"rx checksum not available on this host");
1788 		return -ENOTSUP;
1789 	}
1790 
1791 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
1792 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1793 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
1794 		PMD_DRV_LOG(ERR,
1795 			"Large Receive Offload not available on this host");
1796 		return -ENOTSUP;
1797 	}
1798 
1799 	/* start control queue */
1800 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1801 		virtio_dev_cq_start(dev);
1802 
1803 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
1804 		hw->vlan_strip = 1;
1805 
1806 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1807 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1808 		PMD_DRV_LOG(ERR,
1809 			    "vlan filtering not available on this host");
1810 		return -ENOTSUP;
1811 	}
1812 
1813 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1814 		/* Enable vector (0) for Link State Intrerrupt */
1815 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1816 				VIRTIO_MSI_NO_VECTOR) {
1817 			PMD_DRV_LOG(ERR, "failed to set config vector");
1818 			return -EBUSY;
1819 		}
1820 
1821 	rte_spinlock_init(&hw->state_lock);
1822 
1823 	hw->use_simple_rx = 1;
1824 	hw->use_simple_tx = 1;
1825 
1826 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
1827 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1828 		hw->use_simple_rx = 0;
1829 		hw->use_simple_tx = 0;
1830 	}
1831 #endif
1832 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1833 		hw->use_simple_rx = 0;
1834 		hw->use_simple_tx = 0;
1835 	}
1836 
1837 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
1838 			   DEV_RX_OFFLOAD_TCP_CKSUM))
1839 		hw->use_simple_rx = 0;
1840 
1841 	return 0;
1842 }
1843 
1844 
1845 static int
1846 virtio_dev_start(struct rte_eth_dev *dev)
1847 {
1848 	uint16_t nb_queues, i;
1849 	struct virtnet_rx *rxvq;
1850 	struct virtnet_tx *txvq __rte_unused;
1851 	struct virtio_hw *hw = dev->data->dev_private;
1852 	int ret;
1853 
1854 	/* Finish the initialization of the queues */
1855 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1856 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
1857 		if (ret < 0)
1858 			return ret;
1859 	}
1860 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1861 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
1862 		if (ret < 0)
1863 			return ret;
1864 	}
1865 
1866 	/* check if lsc interrupt feature is enabled */
1867 	if (dev->data->dev_conf.intr_conf.lsc) {
1868 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1869 			PMD_DRV_LOG(ERR, "link status not supported by host");
1870 			return -ENOTSUP;
1871 		}
1872 	}
1873 
1874 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1875 	 * in device configure, but it could be unmapped  when device is
1876 	 * stopped.
1877 	 */
1878 	if (dev->data->dev_conf.intr_conf.lsc ||
1879 	    dev->data->dev_conf.intr_conf.rxq) {
1880 		virtio_intr_disable(dev);
1881 
1882 		if (virtio_intr_enable(dev) < 0) {
1883 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1884 			return -EIO;
1885 		}
1886 	}
1887 
1888 	/*Notify the backend
1889 	 *Otherwise the tap backend might already stop its queue due to fullness.
1890 	 *vhost backend will have no chance to be waked up
1891 	 */
1892 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1893 	if (hw->max_queue_pairs > 1) {
1894 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1895 			return -EINVAL;
1896 	}
1897 
1898 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1899 
1900 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1901 		rxvq = dev->data->rx_queues[i];
1902 		/* Flush the old packets */
1903 		virtqueue_rxvq_flush(rxvq->vq);
1904 		virtqueue_notify(rxvq->vq);
1905 	}
1906 
1907 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1908 		txvq = dev->data->tx_queues[i];
1909 		virtqueue_notify(txvq->vq);
1910 	}
1911 
1912 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1913 
1914 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1915 		rxvq = dev->data->rx_queues[i];
1916 		VIRTQUEUE_DUMP(rxvq->vq);
1917 	}
1918 
1919 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1920 		txvq = dev->data->tx_queues[i];
1921 		VIRTQUEUE_DUMP(txvq->vq);
1922 	}
1923 
1924 	set_rxtx_funcs(dev);
1925 	hw->started = 1;
1926 
1927 	/* Initialize Link state */
1928 	virtio_dev_link_update(dev, 0);
1929 
1930 	return 0;
1931 }
1932 
1933 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1934 {
1935 	struct virtio_hw *hw = dev->data->dev_private;
1936 	uint16_t nr_vq = virtio_get_nr_vq(hw);
1937 	const char *type __rte_unused;
1938 	unsigned int i, mbuf_num = 0;
1939 	struct virtqueue *vq;
1940 	struct rte_mbuf *buf;
1941 	int queue_type;
1942 
1943 	if (hw->vqs == NULL)
1944 		return;
1945 
1946 	for (i = 0; i < nr_vq; i++) {
1947 		vq = hw->vqs[i];
1948 		if (!vq)
1949 			continue;
1950 
1951 		queue_type = virtio_get_queue_type(hw, i);
1952 		if (queue_type == VTNET_RQ)
1953 			type = "rxq";
1954 		else if (queue_type == VTNET_TQ)
1955 			type = "txq";
1956 		else
1957 			continue;
1958 
1959 		PMD_INIT_LOG(DEBUG,
1960 			"Before freeing %s[%d] used and unused buf",
1961 			type, i);
1962 		VIRTQUEUE_DUMP(vq);
1963 
1964 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
1965 			rte_pktmbuf_free(buf);
1966 			mbuf_num++;
1967 		}
1968 
1969 		PMD_INIT_LOG(DEBUG,
1970 			"After freeing %s[%d] used and unused buf",
1971 			type, i);
1972 		VIRTQUEUE_DUMP(vq);
1973 	}
1974 
1975 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
1976 }
1977 
1978 /*
1979  * Stop device: disable interrupt and mark link down
1980  */
1981 static void
1982 virtio_dev_stop(struct rte_eth_dev *dev)
1983 {
1984 	struct virtio_hw *hw = dev->data->dev_private;
1985 	struct rte_eth_link link;
1986 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1987 
1988 	PMD_INIT_LOG(DEBUG, "stop");
1989 
1990 	rte_spinlock_lock(&hw->state_lock);
1991 	if (intr_conf->lsc || intr_conf->rxq)
1992 		virtio_intr_disable(dev);
1993 
1994 	hw->started = 0;
1995 	memset(&link, 0, sizeof(link));
1996 	rte_eth_linkstatus_set(dev, &link);
1997 	rte_spinlock_unlock(&hw->state_lock);
1998 }
1999 
2000 static int
2001 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2002 {
2003 	struct rte_eth_link link;
2004 	uint16_t status;
2005 	struct virtio_hw *hw = dev->data->dev_private;
2006 
2007 	memset(&link, 0, sizeof(link));
2008 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
2009 	link.link_speed  = ETH_SPEED_NUM_10G;
2010 	link.link_autoneg = ETH_LINK_FIXED;
2011 
2012 	if (hw->started == 0) {
2013 		link.link_status = ETH_LINK_DOWN;
2014 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2015 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2016 		vtpci_read_dev_config(hw,
2017 				offsetof(struct virtio_net_config, status),
2018 				&status, sizeof(status));
2019 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2020 			link.link_status = ETH_LINK_DOWN;
2021 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2022 				     dev->data->port_id);
2023 		} else {
2024 			link.link_status = ETH_LINK_UP;
2025 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2026 				     dev->data->port_id);
2027 		}
2028 	} else {
2029 		link.link_status = ETH_LINK_UP;
2030 	}
2031 
2032 	return rte_eth_linkstatus_set(dev, &link);
2033 }
2034 
2035 static int
2036 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2037 {
2038 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2039 	struct virtio_hw *hw = dev->data->dev_private;
2040 	uint64_t offloads = rxmode->offloads;
2041 
2042 	if (mask & ETH_VLAN_FILTER_MASK) {
2043 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2044 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2045 
2046 			PMD_DRV_LOG(NOTICE,
2047 				"vlan filtering not available on this host");
2048 
2049 			return -ENOTSUP;
2050 		}
2051 	}
2052 
2053 	if (mask & ETH_VLAN_STRIP_MASK)
2054 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2055 
2056 	return 0;
2057 }
2058 
2059 static void
2060 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2061 {
2062 	uint64_t tso_mask, host_features;
2063 	struct virtio_hw *hw = dev->data->dev_private;
2064 
2065 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2066 
2067 	dev_info->max_rx_queues =
2068 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2069 	dev_info->max_tx_queues =
2070 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2071 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2072 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2073 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2074 	dev_info->default_txconf = (struct rte_eth_txconf) {
2075 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
2076 	};
2077 
2078 	host_features = VTPCI_OPS(hw)->get_features(hw);
2079 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2080 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2081 		dev_info->rx_offload_capa |=
2082 			DEV_RX_OFFLOAD_TCP_CKSUM |
2083 			DEV_RX_OFFLOAD_UDP_CKSUM;
2084 	}
2085 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2086 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2087 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2088 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2089 	if ((host_features & tso_mask) == tso_mask)
2090 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2091 
2092 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2093 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2094 	if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2095 		dev_info->tx_offload_capa |=
2096 			DEV_TX_OFFLOAD_UDP_CKSUM |
2097 			DEV_TX_OFFLOAD_TCP_CKSUM;
2098 	}
2099 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2100 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2101 	if ((hw->guest_features & tso_mask) == tso_mask)
2102 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2103 }
2104 
2105 /*
2106  * It enables testpmd to collect per queue stats.
2107  */
2108 static int
2109 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2110 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2111 __rte_unused uint8_t is_rx)
2112 {
2113 	return 0;
2114 }
2115 
2116 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2117 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2118 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2119 
2120 RTE_INIT(virtio_init_log);
2121 static void
2122 virtio_init_log(void)
2123 {
2124 	virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2125 	if (virtio_logtype_init >= 0)
2126 		rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2127 	virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2128 	if (virtio_logtype_driver >= 0)
2129 		rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2130 }
2131