xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 3998e2a07220844d3f3c17f76a781ced3efe0de0)
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.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_common.h>
23 #include <rte_errno.h>
24 #include <rte_cpuflags.h>
25 
26 #include <rte_memory.h>
27 #include <rte_eal.h>
28 #include <rte_dev.h>
29 
30 #include "virtio_ethdev.h"
31 #include "virtio_pci.h"
32 #include "virtio_logs.h"
33 #include "virtqueue.h"
34 #include "virtio_rxtx.h"
35 
36 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
37 static int  virtio_dev_configure(struct rte_eth_dev *dev);
38 static int  virtio_dev_start(struct rte_eth_dev *dev);
39 static void virtio_dev_stop(struct rte_eth_dev *dev);
40 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
41 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
42 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
43 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
44 static void virtio_dev_info_get(struct rte_eth_dev *dev,
45 				struct rte_eth_dev_info *dev_info);
46 static int virtio_dev_link_update(struct rte_eth_dev *dev,
47 	int wait_to_complete);
48 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
49 
50 static void virtio_set_hwaddr(struct virtio_hw *hw);
51 static void virtio_get_hwaddr(struct virtio_hw *hw);
52 
53 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
54 				 struct rte_eth_stats *stats);
55 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
56 				 struct rte_eth_xstat *xstats, unsigned n);
57 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
58 				       struct rte_eth_xstat_name *xstats_names,
59 				       unsigned limit);
60 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
61 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
62 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
63 				uint16_t vlan_id, int on);
64 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
65 				struct ether_addr *mac_addr,
66 				uint32_t index, uint32_t vmdq);
67 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
68 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
69 				struct ether_addr *mac_addr);
70 
71 static int virtio_intr_enable(struct rte_eth_dev *dev);
72 static int virtio_intr_disable(struct rte_eth_dev *dev);
73 
74 static int virtio_dev_queue_stats_mapping_set(
75 	struct rte_eth_dev *eth_dev,
76 	uint16_t queue_id,
77 	uint8_t stat_idx,
78 	uint8_t is_rx);
79 
80 /*
81  * The set of PCI devices this driver supports
82  */
83 static const struct rte_pci_id pci_id_virtio_map[] = {
84 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
85 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
86 	{ .vendor_id = 0, /* sentinel */ },
87 };
88 
89 struct rte_virtio_xstats_name_off {
90 	char name[RTE_ETH_XSTATS_NAME_SIZE];
91 	unsigned offset;
92 };
93 
94 /* [rt]x_qX_ is prepended to the name string here */
95 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
96 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
97 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
98 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
99 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
100 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
101 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
102 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
103 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
104 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
105 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
106 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
107 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
108 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
109 };
110 
111 /* [rt]x_qX_ is prepended to the name string here */
112 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
113 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
114 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
115 	{"errors",                 offsetof(struct virtnet_tx, stats.errors)},
116 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
117 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
118 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
119 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
120 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
121 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
122 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
123 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
124 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
125 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
126 };
127 
128 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
129 			    sizeof(rte_virtio_rxq_stat_strings[0]))
130 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
131 			    sizeof(rte_virtio_txq_stat_strings[0]))
132 
133 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
134 
135 static int
136 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
137 		int *dlen, int pkt_num)
138 {
139 	uint32_t head, i;
140 	int k, sum = 0;
141 	virtio_net_ctrl_ack status = ~0;
142 	struct virtio_pmd_ctrl *result;
143 	struct virtqueue *vq;
144 
145 	ctrl->status = status;
146 
147 	if (!cvq || !cvq->vq) {
148 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
149 		return -1;
150 	}
151 	vq = cvq->vq;
152 	head = vq->vq_desc_head_idx;
153 
154 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
155 		"vq->hw->cvq = %p vq = %p",
156 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
157 
158 	if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
159 		return -1;
160 
161 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
162 		sizeof(struct virtio_pmd_ctrl));
163 
164 	/*
165 	 * Format is enforced in qemu code:
166 	 * One TX packet for header;
167 	 * At least one TX packet per argument;
168 	 * One RX packet for ACK.
169 	 */
170 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
171 	vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
172 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
173 	vq->vq_free_cnt--;
174 	i = vq->vq_ring.desc[head].next;
175 
176 	for (k = 0; k < pkt_num; k++) {
177 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
178 		vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
179 			+ sizeof(struct virtio_net_ctrl_hdr)
180 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
181 		vq->vq_ring.desc[i].len = dlen[k];
182 		sum += dlen[k];
183 		vq->vq_free_cnt--;
184 		i = vq->vq_ring.desc[i].next;
185 	}
186 
187 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
188 	vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
189 			+ sizeof(struct virtio_net_ctrl_hdr);
190 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
191 	vq->vq_free_cnt--;
192 
193 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
194 
195 	vq_update_avail_ring(vq, head);
196 	vq_update_avail_idx(vq);
197 
198 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
199 
200 	virtqueue_notify(vq);
201 
202 	rte_rmb();
203 	while (VIRTQUEUE_NUSED(vq) == 0) {
204 		rte_rmb();
205 		usleep(100);
206 	}
207 
208 	while (VIRTQUEUE_NUSED(vq)) {
209 		uint32_t idx, desc_idx, used_idx;
210 		struct vring_used_elem *uep;
211 
212 		used_idx = (uint32_t)(vq->vq_used_cons_idx
213 				& (vq->vq_nentries - 1));
214 		uep = &vq->vq_ring.used->ring[used_idx];
215 		idx = (uint32_t) uep->id;
216 		desc_idx = idx;
217 
218 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
219 			desc_idx = vq->vq_ring.desc[desc_idx].next;
220 			vq->vq_free_cnt++;
221 		}
222 
223 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
224 		vq->vq_desc_head_idx = idx;
225 
226 		vq->vq_used_cons_idx++;
227 		vq->vq_free_cnt++;
228 	}
229 
230 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
231 			vq->vq_free_cnt, vq->vq_desc_head_idx);
232 
233 	result = cvq->virtio_net_hdr_mz->addr;
234 
235 	return result->status;
236 }
237 
238 static int
239 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
240 {
241 	struct virtio_hw *hw = dev->data->dev_private;
242 	struct virtio_pmd_ctrl ctrl;
243 	int dlen[1];
244 	int ret;
245 
246 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
247 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
248 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
249 
250 	dlen[0] = sizeof(uint16_t);
251 
252 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
253 	if (ret) {
254 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
255 			  "failed, this is too late now...");
256 		return -EINVAL;
257 	}
258 
259 	return 0;
260 }
261 
262 static void
263 virtio_dev_queue_release(void *queue __rte_unused)
264 {
265 	/* do nothing */
266 }
267 
268 static int
269 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
270 {
271 	if (vtpci_queue_idx == hw->max_queue_pairs * 2)
272 		return VTNET_CQ;
273 	else if (vtpci_queue_idx % 2 == 0)
274 		return VTNET_RQ;
275 	else
276 		return VTNET_TQ;
277 }
278 
279 static uint16_t
280 virtio_get_nr_vq(struct virtio_hw *hw)
281 {
282 	uint16_t nr_vq = hw->max_queue_pairs * 2;
283 
284 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
285 		nr_vq += 1;
286 
287 	return nr_vq;
288 }
289 
290 static void
291 virtio_init_vring(struct virtqueue *vq)
292 {
293 	int size = vq->vq_nentries;
294 	struct vring *vr = &vq->vq_ring;
295 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
296 
297 	PMD_INIT_FUNC_TRACE();
298 
299 	/*
300 	 * Reinitialise since virtio port might have been stopped and restarted
301 	 */
302 	memset(ring_mem, 0, vq->vq_ring_size);
303 	vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
304 	vq->vq_used_cons_idx = 0;
305 	vq->vq_desc_head_idx = 0;
306 	vq->vq_avail_idx = 0;
307 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
308 	vq->vq_free_cnt = vq->vq_nentries;
309 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
310 
311 	vring_desc_init(vr->desc, size);
312 
313 	/*
314 	 * Disable device(host) interrupting guest
315 	 */
316 	virtqueue_disable_intr(vq);
317 }
318 
319 static int
320 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
321 {
322 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
323 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
324 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
325 	unsigned int vq_size, size;
326 	struct virtio_hw *hw = dev->data->dev_private;
327 	struct virtnet_rx *rxvq = NULL;
328 	struct virtnet_tx *txvq = NULL;
329 	struct virtnet_ctl *cvq = NULL;
330 	struct virtqueue *vq;
331 	size_t sz_hdr_mz = 0;
332 	void *sw_ring = NULL;
333 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
334 	int ret;
335 
336 	PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
337 
338 	/*
339 	 * Read the virtqueue size from the Queue Size field
340 	 * Always power of 2 and if 0 virtqueue does not exist
341 	 */
342 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
343 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
344 	if (vq_size == 0) {
345 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
346 		return -EINVAL;
347 	}
348 
349 	if (!rte_is_power_of_2(vq_size)) {
350 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
351 		return -EINVAL;
352 	}
353 
354 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
355 		 dev->data->port_id, vtpci_queue_idx);
356 
357 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
358 				vq_size * sizeof(struct vq_desc_extra),
359 				RTE_CACHE_LINE_SIZE);
360 	if (queue_type == VTNET_TQ) {
361 		/*
362 		 * For each xmit packet, allocate a virtio_net_hdr
363 		 * and indirect ring elements
364 		 */
365 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
366 	} else if (queue_type == VTNET_CQ) {
367 		/* Allocate a page for control vq command, data and status */
368 		sz_hdr_mz = PAGE_SIZE;
369 	}
370 
371 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
372 				SOCKET_ID_ANY);
373 	if (vq == NULL) {
374 		PMD_INIT_LOG(ERR, "can not allocate vq");
375 		return -ENOMEM;
376 	}
377 	hw->vqs[vtpci_queue_idx] = vq;
378 
379 	vq->hw = hw;
380 	vq->vq_queue_index = vtpci_queue_idx;
381 	vq->vq_nentries = vq_size;
382 
383 	/*
384 	 * Reserve a memzone for vring elements
385 	 */
386 	size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
387 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
388 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
389 		     size, vq->vq_ring_size);
390 
391 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
392 					 SOCKET_ID_ANY,
393 					 0, VIRTIO_PCI_VRING_ALIGN);
394 	if (mz == NULL) {
395 		if (rte_errno == EEXIST)
396 			mz = rte_memzone_lookup(vq_name);
397 		if (mz == NULL) {
398 			ret = -ENOMEM;
399 			goto fail_q_alloc;
400 		}
401 	}
402 
403 	memset(mz->addr, 0, mz->len);
404 
405 	vq->vq_ring_mem = mz->iova;
406 	vq->vq_ring_virt_mem = mz->addr;
407 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
408 		     (uint64_t)mz->iova);
409 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
410 		     (uint64_t)(uintptr_t)mz->addr);
411 
412 	virtio_init_vring(vq);
413 
414 	if (sz_hdr_mz) {
415 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
416 			 dev->data->port_id, vtpci_queue_idx);
417 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
418 						     SOCKET_ID_ANY, 0,
419 						     RTE_CACHE_LINE_SIZE);
420 		if (hdr_mz == NULL) {
421 			if (rte_errno == EEXIST)
422 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
423 			if (hdr_mz == NULL) {
424 				ret = -ENOMEM;
425 				goto fail_q_alloc;
426 			}
427 		}
428 	}
429 
430 	if (queue_type == VTNET_RQ) {
431 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
432 			       sizeof(vq->sw_ring[0]);
433 
434 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
435 				RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
436 		if (!sw_ring) {
437 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
438 			ret = -ENOMEM;
439 			goto fail_q_alloc;
440 		}
441 
442 		vq->sw_ring = sw_ring;
443 		rxvq = &vq->rxq;
444 		rxvq->vq = vq;
445 		rxvq->port_id = dev->data->port_id;
446 		rxvq->mz = mz;
447 	} else if (queue_type == VTNET_TQ) {
448 		txvq = &vq->txq;
449 		txvq->vq = vq;
450 		txvq->port_id = dev->data->port_id;
451 		txvq->mz = mz;
452 		txvq->virtio_net_hdr_mz = hdr_mz;
453 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
454 	} else if (queue_type == VTNET_CQ) {
455 		cvq = &vq->cq;
456 		cvq->vq = vq;
457 		cvq->mz = mz;
458 		cvq->virtio_net_hdr_mz = hdr_mz;
459 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
460 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
461 
462 		hw->cvq = cvq;
463 	}
464 
465 	/* For virtio_user case (that is when hw->dev is NULL), we use
466 	 * virtual address. And we need properly set _offset_, please see
467 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
468 	 */
469 	if (!hw->virtio_user_dev)
470 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
471 	else {
472 		vq->vq_ring_mem = (uintptr_t)mz->addr;
473 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
474 		if (queue_type == VTNET_TQ)
475 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
476 		else if (queue_type == VTNET_CQ)
477 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
478 	}
479 
480 	if (queue_type == VTNET_TQ) {
481 		struct virtio_tx_region *txr;
482 		unsigned int i;
483 
484 		txr = hdr_mz->addr;
485 		memset(txr, 0, vq_size * sizeof(*txr));
486 		for (i = 0; i < vq_size; i++) {
487 			struct vring_desc *start_dp = txr[i].tx_indir;
488 
489 			vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
490 
491 			/* first indirect descriptor is always the tx header */
492 			start_dp->addr = txvq->virtio_net_hdr_mem
493 				+ i * sizeof(*txr)
494 				+ offsetof(struct virtio_tx_region, tx_hdr);
495 
496 			start_dp->len = hw->vtnet_hdr_size;
497 			start_dp->flags = VRING_DESC_F_NEXT;
498 		}
499 	}
500 
501 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
502 		PMD_INIT_LOG(ERR, "setup_queue failed");
503 		return -EINVAL;
504 	}
505 
506 	return 0;
507 
508 fail_q_alloc:
509 	rte_free(sw_ring);
510 	rte_memzone_free(hdr_mz);
511 	rte_memzone_free(mz);
512 	rte_free(vq);
513 
514 	return ret;
515 }
516 
517 static void
518 virtio_free_queues(struct virtio_hw *hw)
519 {
520 	uint16_t nr_vq = virtio_get_nr_vq(hw);
521 	struct virtqueue *vq;
522 	int queue_type;
523 	uint16_t i;
524 
525 	if (hw->vqs == NULL)
526 		return;
527 
528 	for (i = 0; i < nr_vq; i++) {
529 		vq = hw->vqs[i];
530 		if (!vq)
531 			continue;
532 
533 		queue_type = virtio_get_queue_type(hw, i);
534 		if (queue_type == VTNET_RQ) {
535 			rte_free(vq->sw_ring);
536 			rte_memzone_free(vq->rxq.mz);
537 		} else if (queue_type == VTNET_TQ) {
538 			rte_memzone_free(vq->txq.mz);
539 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
540 		} else {
541 			rte_memzone_free(vq->cq.mz);
542 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
543 		}
544 
545 		rte_free(vq);
546 		hw->vqs[i] = NULL;
547 	}
548 
549 	rte_free(hw->vqs);
550 	hw->vqs = NULL;
551 }
552 
553 static int
554 virtio_alloc_queues(struct rte_eth_dev *dev)
555 {
556 	struct virtio_hw *hw = dev->data->dev_private;
557 	uint16_t nr_vq = virtio_get_nr_vq(hw);
558 	uint16_t i;
559 	int ret;
560 
561 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
562 	if (!hw->vqs) {
563 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
564 		return -ENOMEM;
565 	}
566 
567 	for (i = 0; i < nr_vq; i++) {
568 		ret = virtio_init_queue(dev, i);
569 		if (ret < 0) {
570 			virtio_free_queues(hw);
571 			return ret;
572 		}
573 	}
574 
575 	return 0;
576 }
577 
578 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
579 
580 static void
581 virtio_dev_close(struct rte_eth_dev *dev)
582 {
583 	struct virtio_hw *hw = dev->data->dev_private;
584 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
585 
586 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
587 
588 	/* reset the NIC */
589 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
590 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
591 	if (intr_conf->rxq)
592 		virtio_queues_unbind_intr(dev);
593 
594 	if (intr_conf->lsc || intr_conf->rxq) {
595 		virtio_intr_disable(dev);
596 		rte_intr_efd_disable(dev->intr_handle);
597 		rte_free(dev->intr_handle->intr_vec);
598 		dev->intr_handle->intr_vec = NULL;
599 	}
600 
601 	vtpci_reset(hw);
602 	virtio_dev_free_mbufs(dev);
603 	virtio_free_queues(hw);
604 }
605 
606 static void
607 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
608 {
609 	struct virtio_hw *hw = dev->data->dev_private;
610 	struct virtio_pmd_ctrl ctrl;
611 	int dlen[1];
612 	int ret;
613 
614 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
615 		PMD_INIT_LOG(INFO, "host does not support rx control");
616 		return;
617 	}
618 
619 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
620 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
621 	ctrl.data[0] = 1;
622 	dlen[0] = 1;
623 
624 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
625 	if (ret)
626 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
627 }
628 
629 static void
630 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
631 {
632 	struct virtio_hw *hw = dev->data->dev_private;
633 	struct virtio_pmd_ctrl ctrl;
634 	int dlen[1];
635 	int ret;
636 
637 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
638 		PMD_INIT_LOG(INFO, "host does not support rx control");
639 		return;
640 	}
641 
642 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
643 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
644 	ctrl.data[0] = 0;
645 	dlen[0] = 1;
646 
647 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
648 	if (ret)
649 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
650 }
651 
652 static void
653 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
654 {
655 	struct virtio_hw *hw = dev->data->dev_private;
656 	struct virtio_pmd_ctrl ctrl;
657 	int dlen[1];
658 	int ret;
659 
660 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
661 		PMD_INIT_LOG(INFO, "host does not support rx control");
662 		return;
663 	}
664 
665 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
666 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
667 	ctrl.data[0] = 1;
668 	dlen[0] = 1;
669 
670 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
671 	if (ret)
672 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
673 }
674 
675 static void
676 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
677 {
678 	struct virtio_hw *hw = dev->data->dev_private;
679 	struct virtio_pmd_ctrl ctrl;
680 	int dlen[1];
681 	int ret;
682 
683 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
684 		PMD_INIT_LOG(INFO, "host does not support rx control");
685 		return;
686 	}
687 
688 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
689 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
690 	ctrl.data[0] = 0;
691 	dlen[0] = 1;
692 
693 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
694 	if (ret)
695 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
696 }
697 
698 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
699 static int
700 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
701 {
702 	struct virtio_hw *hw = dev->data->dev_private;
703 	uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
704 				 hw->vtnet_hdr_size;
705 	uint32_t frame_size = mtu + ether_hdr_len;
706 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
707 
708 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
709 
710 	if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
711 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
712 			ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
713 		return -EINVAL;
714 	}
715 	return 0;
716 }
717 
718 static int
719 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
720 {
721 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
722 	struct virtqueue *vq = rxvq->vq;
723 
724 	virtqueue_enable_intr(vq);
725 	return 0;
726 }
727 
728 static int
729 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
730 {
731 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
732 	struct virtqueue *vq = rxvq->vq;
733 
734 	virtqueue_disable_intr(vq);
735 	return 0;
736 }
737 
738 /*
739  * dev_ops for virtio, bare necessities for basic operation
740  */
741 static const struct eth_dev_ops virtio_eth_dev_ops = {
742 	.dev_configure           = virtio_dev_configure,
743 	.dev_start               = virtio_dev_start,
744 	.dev_stop                = virtio_dev_stop,
745 	.dev_close               = virtio_dev_close,
746 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
747 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
748 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
749 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
750 	.mtu_set                 = virtio_mtu_set,
751 	.dev_infos_get           = virtio_dev_info_get,
752 	.stats_get               = virtio_dev_stats_get,
753 	.xstats_get              = virtio_dev_xstats_get,
754 	.xstats_get_names        = virtio_dev_xstats_get_names,
755 	.stats_reset             = virtio_dev_stats_reset,
756 	.xstats_reset            = virtio_dev_stats_reset,
757 	.link_update             = virtio_dev_link_update,
758 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
759 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
760 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
761 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
762 	.rx_queue_release        = virtio_dev_queue_release,
763 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
764 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
765 	.tx_queue_release        = virtio_dev_queue_release,
766 	/* collect stats per queue */
767 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
768 	.vlan_filter_set         = virtio_vlan_filter_set,
769 	.mac_addr_add            = virtio_mac_addr_add,
770 	.mac_addr_remove         = virtio_mac_addr_remove,
771 	.mac_addr_set            = virtio_mac_addr_set,
772 };
773 
774 static inline int
775 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
776 				struct rte_eth_link *link)
777 {
778 	struct rte_eth_link *dst = link;
779 	struct rte_eth_link *src = &(dev->data->dev_link);
780 
781 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
782 			*(uint64_t *)src) == 0)
783 		return -1;
784 
785 	return 0;
786 }
787 
788 /**
789  * Atomically writes the link status information into global
790  * structure rte_eth_dev.
791  *
792  * @param dev
793  *   - Pointer to the structure rte_eth_dev to read from.
794  *   - Pointer to the buffer to be saved with the link status.
795  *
796  * @return
797  *   - On success, zero.
798  *   - On failure, negative value.
799  */
800 static inline int
801 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
802 		struct rte_eth_link *link)
803 {
804 	struct rte_eth_link *dst = &(dev->data->dev_link);
805 	struct rte_eth_link *src = link;
806 
807 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
808 					*(uint64_t *)src) == 0)
809 		return -1;
810 
811 	return 0;
812 }
813 
814 static void
815 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
816 {
817 	unsigned i;
818 
819 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
820 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
821 		if (txvq == NULL)
822 			continue;
823 
824 		stats->opackets += txvq->stats.packets;
825 		stats->obytes += txvq->stats.bytes;
826 		stats->oerrors += txvq->stats.errors;
827 
828 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
829 			stats->q_opackets[i] = txvq->stats.packets;
830 			stats->q_obytes[i] = txvq->stats.bytes;
831 		}
832 	}
833 
834 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
835 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
836 		if (rxvq == NULL)
837 			continue;
838 
839 		stats->ipackets += rxvq->stats.packets;
840 		stats->ibytes += rxvq->stats.bytes;
841 		stats->ierrors += rxvq->stats.errors;
842 
843 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
844 			stats->q_ipackets[i] = rxvq->stats.packets;
845 			stats->q_ibytes[i] = rxvq->stats.bytes;
846 		}
847 	}
848 
849 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
850 }
851 
852 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
853 				       struct rte_eth_xstat_name *xstats_names,
854 				       __rte_unused unsigned limit)
855 {
856 	unsigned i;
857 	unsigned count = 0;
858 	unsigned t;
859 
860 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
861 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
862 
863 	if (xstats_names != NULL) {
864 		/* Note: limit checked in rte_eth_xstats_names() */
865 
866 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
867 			struct virtqueue *rxvq = dev->data->rx_queues[i];
868 			if (rxvq == NULL)
869 				continue;
870 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
871 				snprintf(xstats_names[count].name,
872 					sizeof(xstats_names[count].name),
873 					"rx_q%u_%s", i,
874 					rte_virtio_rxq_stat_strings[t].name);
875 				count++;
876 			}
877 		}
878 
879 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
880 			struct virtqueue *txvq = dev->data->tx_queues[i];
881 			if (txvq == NULL)
882 				continue;
883 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
884 				snprintf(xstats_names[count].name,
885 					sizeof(xstats_names[count].name),
886 					"tx_q%u_%s", i,
887 					rte_virtio_txq_stat_strings[t].name);
888 				count++;
889 			}
890 		}
891 		return count;
892 	}
893 	return nstats;
894 }
895 
896 static int
897 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
898 		      unsigned n)
899 {
900 	unsigned i;
901 	unsigned count = 0;
902 
903 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
904 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
905 
906 	if (n < nstats)
907 		return nstats;
908 
909 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
910 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
911 
912 		if (rxvq == NULL)
913 			continue;
914 
915 		unsigned t;
916 
917 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
918 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
919 				rte_virtio_rxq_stat_strings[t].offset);
920 			xstats[count].id = count;
921 			count++;
922 		}
923 	}
924 
925 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
926 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
927 
928 		if (txvq == NULL)
929 			continue;
930 
931 		unsigned t;
932 
933 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
934 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
935 				rte_virtio_txq_stat_strings[t].offset);
936 			xstats[count].id = count;
937 			count++;
938 		}
939 	}
940 
941 	return count;
942 }
943 
944 static int
945 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
946 {
947 	virtio_update_stats(dev, stats);
948 
949 	return 0;
950 }
951 
952 static void
953 virtio_dev_stats_reset(struct rte_eth_dev *dev)
954 {
955 	unsigned int i;
956 
957 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
958 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
959 		if (txvq == NULL)
960 			continue;
961 
962 		txvq->stats.packets = 0;
963 		txvq->stats.bytes = 0;
964 		txvq->stats.errors = 0;
965 		txvq->stats.multicast = 0;
966 		txvq->stats.broadcast = 0;
967 		memset(txvq->stats.size_bins, 0,
968 		       sizeof(txvq->stats.size_bins[0]) * 8);
969 	}
970 
971 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
972 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
973 		if (rxvq == NULL)
974 			continue;
975 
976 		rxvq->stats.packets = 0;
977 		rxvq->stats.bytes = 0;
978 		rxvq->stats.errors = 0;
979 		rxvq->stats.multicast = 0;
980 		rxvq->stats.broadcast = 0;
981 		memset(rxvq->stats.size_bins, 0,
982 		       sizeof(rxvq->stats.size_bins[0]) * 8);
983 	}
984 }
985 
986 static void
987 virtio_set_hwaddr(struct virtio_hw *hw)
988 {
989 	vtpci_write_dev_config(hw,
990 			offsetof(struct virtio_net_config, mac),
991 			&hw->mac_addr, ETHER_ADDR_LEN);
992 }
993 
994 static void
995 virtio_get_hwaddr(struct virtio_hw *hw)
996 {
997 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
998 		vtpci_read_dev_config(hw,
999 			offsetof(struct virtio_net_config, mac),
1000 			&hw->mac_addr, ETHER_ADDR_LEN);
1001 	} else {
1002 		eth_random_addr(&hw->mac_addr[0]);
1003 		virtio_set_hwaddr(hw);
1004 	}
1005 }
1006 
1007 static int
1008 virtio_mac_table_set(struct virtio_hw *hw,
1009 		     const struct virtio_net_ctrl_mac *uc,
1010 		     const struct virtio_net_ctrl_mac *mc)
1011 {
1012 	struct virtio_pmd_ctrl ctrl;
1013 	int err, len[2];
1014 
1015 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1016 		PMD_DRV_LOG(INFO, "host does not support mac table");
1017 		return -1;
1018 	}
1019 
1020 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1021 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1022 
1023 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1024 	memcpy(ctrl.data, uc, len[0]);
1025 
1026 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1027 	memcpy(ctrl.data + len[0], mc, len[1]);
1028 
1029 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1030 	if (err != 0)
1031 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1032 	return err;
1033 }
1034 
1035 static int
1036 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1037 		    uint32_t index, uint32_t vmdq __rte_unused)
1038 {
1039 	struct virtio_hw *hw = dev->data->dev_private;
1040 	const struct ether_addr *addrs = dev->data->mac_addrs;
1041 	unsigned int i;
1042 	struct virtio_net_ctrl_mac *uc, *mc;
1043 
1044 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1045 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1046 		return -EINVAL;
1047 	}
1048 
1049 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1050 	uc->entries = 0;
1051 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1052 	mc->entries = 0;
1053 
1054 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1055 		const struct ether_addr *addr
1056 			= (i == index) ? mac_addr : addrs + i;
1057 		struct virtio_net_ctrl_mac *tbl
1058 			= is_multicast_ether_addr(addr) ? mc : uc;
1059 
1060 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1061 	}
1062 
1063 	return virtio_mac_table_set(hw, uc, mc);
1064 }
1065 
1066 static void
1067 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1068 {
1069 	struct virtio_hw *hw = dev->data->dev_private;
1070 	struct ether_addr *addrs = dev->data->mac_addrs;
1071 	struct virtio_net_ctrl_mac *uc, *mc;
1072 	unsigned int i;
1073 
1074 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1075 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1076 		return;
1077 	}
1078 
1079 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1080 	uc->entries = 0;
1081 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1082 	mc->entries = 0;
1083 
1084 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1085 		struct virtio_net_ctrl_mac *tbl;
1086 
1087 		if (i == index || is_zero_ether_addr(addrs + i))
1088 			continue;
1089 
1090 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1091 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1092 	}
1093 
1094 	virtio_mac_table_set(hw, uc, mc);
1095 }
1096 
1097 static void
1098 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1099 {
1100 	struct virtio_hw *hw = dev->data->dev_private;
1101 
1102 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1103 
1104 	/* Use atomic update if available */
1105 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1106 		struct virtio_pmd_ctrl ctrl;
1107 		int len = ETHER_ADDR_LEN;
1108 
1109 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1110 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1111 
1112 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1113 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
1114 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1115 		virtio_set_hwaddr(hw);
1116 }
1117 
1118 static int
1119 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1120 {
1121 	struct virtio_hw *hw = dev->data->dev_private;
1122 	struct virtio_pmd_ctrl ctrl;
1123 	int len;
1124 
1125 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1126 		return -ENOTSUP;
1127 
1128 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1129 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1130 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1131 	len = sizeof(vlan_id);
1132 
1133 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1134 }
1135 
1136 static int
1137 virtio_intr_enable(struct rte_eth_dev *dev)
1138 {
1139 	struct virtio_hw *hw = dev->data->dev_private;
1140 
1141 	if (rte_intr_enable(dev->intr_handle) < 0)
1142 		return -1;
1143 
1144 	if (!hw->virtio_user_dev)
1145 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1146 
1147 	return 0;
1148 }
1149 
1150 static int
1151 virtio_intr_disable(struct rte_eth_dev *dev)
1152 {
1153 	struct virtio_hw *hw = dev->data->dev_private;
1154 
1155 	if (rte_intr_disable(dev->intr_handle) < 0)
1156 		return -1;
1157 
1158 	if (!hw->virtio_user_dev)
1159 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1160 
1161 	return 0;
1162 }
1163 
1164 static int
1165 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1166 {
1167 	uint64_t host_features;
1168 
1169 	/* Prepare guest_features: feature that driver wants to support */
1170 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1171 		req_features);
1172 
1173 	/* Read device(host) feature bits */
1174 	host_features = VTPCI_OPS(hw)->get_features(hw);
1175 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1176 		host_features);
1177 
1178 	/* If supported, ensure MTU value is valid before acknowledging it. */
1179 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1180 		struct virtio_net_config config;
1181 
1182 		vtpci_read_dev_config(hw,
1183 			offsetof(struct virtio_net_config, mtu),
1184 			&config.mtu, sizeof(config.mtu));
1185 
1186 		if (config.mtu < ETHER_MIN_MTU)
1187 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1188 	}
1189 
1190 	/*
1191 	 * Negotiate features: Subset of device feature bits are written back
1192 	 * guest feature bits.
1193 	 */
1194 	hw->guest_features = req_features;
1195 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1196 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1197 		hw->guest_features);
1198 
1199 	if (hw->modern) {
1200 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1201 			PMD_INIT_LOG(ERR,
1202 				"VIRTIO_F_VERSION_1 features is not enabled.");
1203 			return -1;
1204 		}
1205 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1206 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1207 			PMD_INIT_LOG(ERR,
1208 				"failed to set FEATURES_OK status!");
1209 			return -1;
1210 		}
1211 	}
1212 
1213 	hw->req_guest_features = req_features;
1214 
1215 	return 0;
1216 }
1217 
1218 /*
1219  * Process Virtio Config changed interrupt and call the callback
1220  * if link state changed.
1221  */
1222 void
1223 virtio_interrupt_handler(void *param)
1224 {
1225 	struct rte_eth_dev *dev = param;
1226 	struct virtio_hw *hw = dev->data->dev_private;
1227 	uint8_t isr;
1228 
1229 	/* Read interrupt status which clears interrupt */
1230 	isr = vtpci_isr(hw);
1231 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1232 
1233 	if (virtio_intr_enable(dev) < 0)
1234 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1235 
1236 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1237 		if (virtio_dev_link_update(dev, 0) == 0)
1238 			_rte_eth_dev_callback_process(dev,
1239 						      RTE_ETH_EVENT_INTR_LSC,
1240 						      NULL, NULL);
1241 	}
1242 
1243 }
1244 
1245 /* set rx and tx handlers according to what is supported */
1246 static void
1247 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1248 {
1249 	struct virtio_hw *hw = eth_dev->data->dev_private;
1250 
1251 	if (hw->use_simple_rx) {
1252 		PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1253 			eth_dev->data->port_id);
1254 		eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1255 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1256 		PMD_INIT_LOG(INFO,
1257 			"virtio: using mergeable buffer Rx path on port %u",
1258 			eth_dev->data->port_id);
1259 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1260 	} else {
1261 		PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1262 			eth_dev->data->port_id);
1263 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1264 	}
1265 
1266 	if (hw->use_simple_tx) {
1267 		PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1268 			eth_dev->data->port_id);
1269 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1270 	} else {
1271 		PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1272 			eth_dev->data->port_id);
1273 		eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1274 	}
1275 }
1276 
1277 /* Only support 1:1 queue/interrupt mapping so far.
1278  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1279  * interrupt vectors (<N+1).
1280  */
1281 static int
1282 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1283 {
1284 	uint32_t i;
1285 	struct virtio_hw *hw = dev->data->dev_private;
1286 
1287 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1288 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1289 		dev->intr_handle->intr_vec[i] = i + 1;
1290 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1291 						 VIRTIO_MSI_NO_VECTOR) {
1292 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1293 			return -EBUSY;
1294 		}
1295 	}
1296 
1297 	return 0;
1298 }
1299 
1300 static void
1301 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1302 {
1303 	uint32_t i;
1304 	struct virtio_hw *hw = dev->data->dev_private;
1305 
1306 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1307 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1308 		VTPCI_OPS(hw)->set_queue_irq(hw,
1309 					     hw->vqs[i * VTNET_CQ],
1310 					     VIRTIO_MSI_NO_VECTOR);
1311 }
1312 
1313 static int
1314 virtio_configure_intr(struct rte_eth_dev *dev)
1315 {
1316 	struct virtio_hw *hw = dev->data->dev_private;
1317 
1318 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1319 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1320 		return -ENOTSUP;
1321 	}
1322 
1323 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1324 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1325 		return -1;
1326 	}
1327 
1328 	if (!dev->intr_handle->intr_vec) {
1329 		dev->intr_handle->intr_vec =
1330 			rte_zmalloc("intr_vec",
1331 				    hw->max_queue_pairs * sizeof(int), 0);
1332 		if (!dev->intr_handle->intr_vec) {
1333 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1334 				     hw->max_queue_pairs);
1335 			return -ENOMEM;
1336 		}
1337 	}
1338 
1339 	/* Re-register callback to update max_intr */
1340 	rte_intr_callback_unregister(dev->intr_handle,
1341 				     virtio_interrupt_handler,
1342 				     dev);
1343 	rte_intr_callback_register(dev->intr_handle,
1344 				   virtio_interrupt_handler,
1345 				   dev);
1346 
1347 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1348 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1349 	 * And for legacy devices, this should be done before queue/vec binding
1350 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1351 	 * (22) will be ignored.
1352 	 */
1353 	if (virtio_intr_enable(dev) < 0) {
1354 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1355 		return -1;
1356 	}
1357 
1358 	if (virtio_queues_bind_intr(dev) < 0) {
1359 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1360 		return -1;
1361 	}
1362 
1363 	return 0;
1364 }
1365 
1366 /* reset device and renegotiate features if needed */
1367 static int
1368 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1369 {
1370 	struct virtio_hw *hw = eth_dev->data->dev_private;
1371 	struct virtio_net_config *config;
1372 	struct virtio_net_config local_config;
1373 	struct rte_pci_device *pci_dev = NULL;
1374 	int ret;
1375 
1376 	/* Reset the device although not necessary at startup */
1377 	vtpci_reset(hw);
1378 
1379 	/* Tell the host we've noticed this device. */
1380 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1381 
1382 	/* Tell the host we've known how to drive the device. */
1383 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1384 	if (virtio_negotiate_features(hw, req_features) < 0)
1385 		return -1;
1386 
1387 	if (!hw->virtio_user_dev) {
1388 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1389 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1390 	}
1391 
1392 	/* If host does not support both status and MSI-X then disable LSC */
1393 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1394 	    hw->use_msix != VIRTIO_MSIX_NONE)
1395 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1396 	else
1397 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1398 
1399 	/* Setting up rx_header size for the device */
1400 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1401 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1402 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1403 	else
1404 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1405 
1406 	/* Copy the permanent MAC address to: virtio_hw */
1407 	virtio_get_hwaddr(hw);
1408 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1409 			&eth_dev->data->mac_addrs[0]);
1410 	PMD_INIT_LOG(DEBUG,
1411 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1412 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1413 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1414 
1415 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1416 		config = &local_config;
1417 
1418 		vtpci_read_dev_config(hw,
1419 			offsetof(struct virtio_net_config, mac),
1420 			&config->mac, sizeof(config->mac));
1421 
1422 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1423 			vtpci_read_dev_config(hw,
1424 				offsetof(struct virtio_net_config, status),
1425 				&config->status, sizeof(config->status));
1426 		} else {
1427 			PMD_INIT_LOG(DEBUG,
1428 				     "VIRTIO_NET_F_STATUS is not supported");
1429 			config->status = 0;
1430 		}
1431 
1432 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1433 			vtpci_read_dev_config(hw,
1434 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1435 				&config->max_virtqueue_pairs,
1436 				sizeof(config->max_virtqueue_pairs));
1437 		} else {
1438 			PMD_INIT_LOG(DEBUG,
1439 				     "VIRTIO_NET_F_MQ is not supported");
1440 			config->max_virtqueue_pairs = 1;
1441 		}
1442 
1443 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1444 
1445 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1446 			vtpci_read_dev_config(hw,
1447 				offsetof(struct virtio_net_config, mtu),
1448 				&config->mtu,
1449 				sizeof(config->mtu));
1450 
1451 			/*
1452 			 * MTU value has already been checked at negotiation
1453 			 * time, but check again in case it has changed since
1454 			 * then, which should not happen.
1455 			 */
1456 			if (config->mtu < ETHER_MIN_MTU) {
1457 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1458 						config->mtu);
1459 				return -1;
1460 			}
1461 
1462 			hw->max_mtu = config->mtu;
1463 			/* Set initial MTU to maximum one supported by vhost */
1464 			eth_dev->data->mtu = config->mtu;
1465 
1466 		} else {
1467 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1468 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1469 		}
1470 
1471 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1472 				config->max_virtqueue_pairs);
1473 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1474 		PMD_INIT_LOG(DEBUG,
1475 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1476 				config->mac[0], config->mac[1],
1477 				config->mac[2], config->mac[3],
1478 				config->mac[4], config->mac[5]);
1479 	} else {
1480 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1481 		hw->max_queue_pairs = 1;
1482 	}
1483 
1484 	ret = virtio_alloc_queues(eth_dev);
1485 	if (ret < 0)
1486 		return ret;
1487 
1488 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1489 		if (virtio_configure_intr(eth_dev) < 0) {
1490 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1491 			return -1;
1492 		}
1493 	}
1494 
1495 	vtpci_reinit_complete(hw);
1496 
1497 	if (pci_dev)
1498 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1499 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1500 			pci_dev->id.device_id);
1501 
1502 	return 0;
1503 }
1504 
1505 /*
1506  * Remap the PCI device again (IO port map for legacy device and
1507  * memory map for modern device), so that the secondary process
1508  * could have the PCI initiated correctly.
1509  */
1510 static int
1511 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1512 {
1513 	if (hw->modern) {
1514 		/*
1515 		 * We don't have to re-parse the PCI config space, since
1516 		 * rte_pci_map_device() makes sure the mapped address
1517 		 * in secondary process would equal to the one mapped in
1518 		 * the primary process: error will be returned if that
1519 		 * requirement is not met.
1520 		 *
1521 		 * That said, we could simply reuse all cap pointers
1522 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1523 		 * primary process, which is stored in shared memory.
1524 		 */
1525 		if (rte_pci_map_device(pci_dev)) {
1526 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1527 			return -1;
1528 		}
1529 	} else {
1530 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1531 			return -1;
1532 	}
1533 
1534 	return 0;
1535 }
1536 
1537 static void
1538 virtio_set_vtpci_ops(struct virtio_hw *hw)
1539 {
1540 #ifdef RTE_VIRTIO_USER
1541 	if (hw->virtio_user_dev)
1542 		VTPCI_OPS(hw) = &virtio_user_ops;
1543 	else
1544 #endif
1545 	if (hw->modern)
1546 		VTPCI_OPS(hw) = &modern_ops;
1547 	else
1548 		VTPCI_OPS(hw) = &legacy_ops;
1549 }
1550 
1551 /*
1552  * This function is based on probe() function in virtio_pci.c
1553  * It returns 0 on success.
1554  */
1555 int
1556 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1557 {
1558 	struct virtio_hw *hw = eth_dev->data->dev_private;
1559 	int ret;
1560 
1561 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1562 
1563 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1564 
1565 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1566 		if (!hw->virtio_user_dev) {
1567 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1568 			if (ret)
1569 				return ret;
1570 		}
1571 
1572 		virtio_set_vtpci_ops(hw);
1573 		set_rxtx_funcs(eth_dev);
1574 
1575 		return 0;
1576 	}
1577 
1578 	/* Allocate memory for storing MAC addresses */
1579 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1580 	if (eth_dev->data->mac_addrs == NULL) {
1581 		PMD_INIT_LOG(ERR,
1582 			"Failed to allocate %d bytes needed to store MAC addresses",
1583 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1584 		return -ENOMEM;
1585 	}
1586 
1587 	hw->port_id = eth_dev->data->port_id;
1588 	/* For virtio_user case the hw->virtio_user_dev is populated by
1589 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1590 	 */
1591 	if (!hw->virtio_user_dev) {
1592 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1593 		if (ret)
1594 			goto out;
1595 	}
1596 
1597 	/* reset device and negotiate default features */
1598 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1599 	if (ret < 0)
1600 		goto out;
1601 
1602 	/* Setup interrupt callback  */
1603 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1604 		rte_intr_callback_register(eth_dev->intr_handle,
1605 			virtio_interrupt_handler, eth_dev);
1606 
1607 	return 0;
1608 
1609 out:
1610 	rte_free(eth_dev->data->mac_addrs);
1611 	return ret;
1612 }
1613 
1614 static int
1615 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1616 {
1617 	PMD_INIT_FUNC_TRACE();
1618 
1619 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1620 		return -EPERM;
1621 
1622 	virtio_dev_stop(eth_dev);
1623 	virtio_dev_close(eth_dev);
1624 
1625 	eth_dev->dev_ops = NULL;
1626 	eth_dev->tx_pkt_burst = NULL;
1627 	eth_dev->rx_pkt_burst = NULL;
1628 
1629 	rte_free(eth_dev->data->mac_addrs);
1630 	eth_dev->data->mac_addrs = NULL;
1631 
1632 	/* reset interrupt callback  */
1633 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1634 		rte_intr_callback_unregister(eth_dev->intr_handle,
1635 						virtio_interrupt_handler,
1636 						eth_dev);
1637 	if (eth_dev->device)
1638 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1639 
1640 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1641 
1642 	return 0;
1643 }
1644 
1645 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1646 	struct rte_pci_device *pci_dev)
1647 {
1648 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1649 		eth_virtio_dev_init);
1650 }
1651 
1652 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1653 {
1654 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1655 }
1656 
1657 static struct rte_pci_driver rte_virtio_pmd = {
1658 	.driver = {
1659 		.name = "net_virtio",
1660 	},
1661 	.id_table = pci_id_virtio_map,
1662 	.drv_flags = 0,
1663 	.probe = eth_virtio_pci_probe,
1664 	.remove = eth_virtio_pci_remove,
1665 };
1666 
1667 RTE_INIT(rte_virtio_pmd_init);
1668 static void
1669 rte_virtio_pmd_init(void)
1670 {
1671 	if (rte_eal_iopl_init() != 0) {
1672 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1673 		return;
1674 	}
1675 
1676 	rte_pci_register(&rte_virtio_pmd);
1677 }
1678 
1679 /*
1680  * Configure virtio device
1681  * It returns 0 on success.
1682  */
1683 static int
1684 virtio_dev_configure(struct rte_eth_dev *dev)
1685 {
1686 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1687 	struct virtio_hw *hw = dev->data->dev_private;
1688 	uint64_t req_features;
1689 	int ret;
1690 
1691 	PMD_INIT_LOG(DEBUG, "configure");
1692 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1693 
1694 	if (dev->data->dev_conf.intr_conf.rxq) {
1695 		ret = virtio_init_device(dev, hw->req_guest_features);
1696 		if (ret < 0)
1697 			return ret;
1698 	}
1699 
1700 	/* The name hw_ip_checksum is a bit confusing since it can be
1701 	 * set by the application to request L3 and/or L4 checksums. In
1702 	 * case of virtio, only L4 checksum is supported.
1703 	 */
1704 	if (rxmode->hw_ip_checksum)
1705 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1706 
1707 	if (rxmode->enable_lro)
1708 		req_features |=
1709 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1710 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1711 
1712 	/* if request features changed, reinit the device */
1713 	if (req_features != hw->req_guest_features) {
1714 		ret = virtio_init_device(dev, req_features);
1715 		if (ret < 0)
1716 			return ret;
1717 	}
1718 
1719 	if (rxmode->hw_ip_checksum &&
1720 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1721 		PMD_DRV_LOG(ERR,
1722 			"rx checksum not available on this host");
1723 		return -ENOTSUP;
1724 	}
1725 
1726 	if (rxmode->enable_lro &&
1727 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1728 			!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1729 		PMD_DRV_LOG(ERR,
1730 			"Large Receive Offload not available on this host");
1731 		return -ENOTSUP;
1732 	}
1733 
1734 	/* start control queue */
1735 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1736 		virtio_dev_cq_start(dev);
1737 
1738 	hw->vlan_strip = rxmode->hw_vlan_strip;
1739 
1740 	if (rxmode->hw_vlan_filter
1741 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1742 		PMD_DRV_LOG(ERR,
1743 			    "vlan filtering not available on this host");
1744 		return -ENOTSUP;
1745 	}
1746 
1747 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1748 		/* Enable vector (0) for Link State Intrerrupt */
1749 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1750 				VIRTIO_MSI_NO_VECTOR) {
1751 			PMD_DRV_LOG(ERR, "failed to set config vector");
1752 			return -EBUSY;
1753 		}
1754 
1755 	hw->use_simple_rx = 1;
1756 	hw->use_simple_tx = 1;
1757 
1758 #if defined RTE_ARCH_ARM64 || defined CONFIG_RTE_ARCH_ARM
1759 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1760 		hw->use_simple_rx = 0;
1761 		hw->use_simple_tx = 0;
1762 	}
1763 #endif
1764 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1765 		hw->use_simple_rx = 0;
1766 		hw->use_simple_tx = 0;
1767 	}
1768 
1769 	if (rxmode->hw_ip_checksum)
1770 		hw->use_simple_rx = 0;
1771 
1772 	return 0;
1773 }
1774 
1775 
1776 static int
1777 virtio_dev_start(struct rte_eth_dev *dev)
1778 {
1779 	uint16_t nb_queues, i;
1780 	struct virtnet_rx *rxvq;
1781 	struct virtnet_tx *txvq __rte_unused;
1782 	struct virtio_hw *hw = dev->data->dev_private;
1783 	int ret;
1784 
1785 	/* Finish the initialization of the queues */
1786 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1787 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
1788 		if (ret < 0)
1789 			return ret;
1790 	}
1791 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1792 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
1793 		if (ret < 0)
1794 			return ret;
1795 	}
1796 
1797 	/* check if lsc interrupt feature is enabled */
1798 	if (dev->data->dev_conf.intr_conf.lsc) {
1799 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1800 			PMD_DRV_LOG(ERR, "link status not supported by host");
1801 			return -ENOTSUP;
1802 		}
1803 	}
1804 
1805 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1806 	 * in device configure, but it could be unmapped  when device is
1807 	 * stopped.
1808 	 */
1809 	if (dev->data->dev_conf.intr_conf.lsc ||
1810 	    dev->data->dev_conf.intr_conf.rxq) {
1811 		virtio_intr_disable(dev);
1812 
1813 		if (virtio_intr_enable(dev) < 0) {
1814 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1815 			return -EIO;
1816 		}
1817 	}
1818 
1819 	/*Notify the backend
1820 	 *Otherwise the tap backend might already stop its queue due to fullness.
1821 	 *vhost backend will have no chance to be waked up
1822 	 */
1823 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1824 	if (hw->max_queue_pairs > 1) {
1825 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1826 			return -EINVAL;
1827 	}
1828 
1829 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1830 
1831 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1832 		rxvq = dev->data->rx_queues[i];
1833 		/* Flush the old packets */
1834 		virtqueue_flush(rxvq->vq);
1835 		virtqueue_notify(rxvq->vq);
1836 	}
1837 
1838 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1839 		txvq = dev->data->tx_queues[i];
1840 		virtqueue_notify(txvq->vq);
1841 	}
1842 
1843 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1844 
1845 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1846 		rxvq = dev->data->rx_queues[i];
1847 		VIRTQUEUE_DUMP(rxvq->vq);
1848 	}
1849 
1850 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1851 		txvq = dev->data->tx_queues[i];
1852 		VIRTQUEUE_DUMP(txvq->vq);
1853 	}
1854 
1855 	set_rxtx_funcs(dev);
1856 	hw->started = 1;
1857 
1858 	/* Initialize Link state */
1859 	virtio_dev_link_update(dev, 0);
1860 
1861 	return 0;
1862 }
1863 
1864 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1865 {
1866 	struct rte_mbuf *buf;
1867 	int i, mbuf_num = 0;
1868 
1869 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1870 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1871 
1872 		PMD_INIT_LOG(DEBUG,
1873 			     "Before freeing rxq[%d] used and unused buf", i);
1874 		VIRTQUEUE_DUMP(rxvq->vq);
1875 
1876 		PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1877 		while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1878 			rte_pktmbuf_free(buf);
1879 			mbuf_num++;
1880 		}
1881 
1882 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1883 		PMD_INIT_LOG(DEBUG,
1884 			     "After freeing rxq[%d] used and unused buf", i);
1885 		VIRTQUEUE_DUMP(rxvq->vq);
1886 	}
1887 
1888 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1889 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1890 
1891 		PMD_INIT_LOG(DEBUG,
1892 			     "Before freeing txq[%d] used and unused bufs",
1893 			     i);
1894 		VIRTQUEUE_DUMP(txvq->vq);
1895 
1896 		mbuf_num = 0;
1897 		while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1898 			rte_pktmbuf_free(buf);
1899 			mbuf_num++;
1900 		}
1901 
1902 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1903 		PMD_INIT_LOG(DEBUG,
1904 			     "After freeing txq[%d] used and unused buf", i);
1905 		VIRTQUEUE_DUMP(txvq->vq);
1906 	}
1907 }
1908 
1909 /*
1910  * Stop device: disable interrupt and mark link down
1911  */
1912 static void
1913 virtio_dev_stop(struct rte_eth_dev *dev)
1914 {
1915 	struct virtio_hw *hw = dev->data->dev_private;
1916 	struct rte_eth_link link;
1917 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1918 
1919 	PMD_INIT_LOG(DEBUG, "stop");
1920 
1921 	if (intr_conf->lsc || intr_conf->rxq)
1922 		virtio_intr_disable(dev);
1923 
1924 	hw->started = 0;
1925 	memset(&link, 0, sizeof(link));
1926 	virtio_dev_atomic_write_link_status(dev, &link);
1927 }
1928 
1929 static int
1930 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1931 {
1932 	struct rte_eth_link link, old;
1933 	uint16_t status;
1934 	struct virtio_hw *hw = dev->data->dev_private;
1935 	memset(&link, 0, sizeof(link));
1936 	virtio_dev_atomic_read_link_status(dev, &link);
1937 	old = link;
1938 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1939 	link.link_speed  = SPEED_10G;
1940 
1941 	if (hw->started == 0) {
1942 		link.link_status = ETH_LINK_DOWN;
1943 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1944 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
1945 		vtpci_read_dev_config(hw,
1946 				offsetof(struct virtio_net_config, status),
1947 				&status, sizeof(status));
1948 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1949 			link.link_status = ETH_LINK_DOWN;
1950 			PMD_INIT_LOG(DEBUG, "Port %d is down",
1951 				     dev->data->port_id);
1952 		} else {
1953 			link.link_status = ETH_LINK_UP;
1954 			PMD_INIT_LOG(DEBUG, "Port %d is up",
1955 				     dev->data->port_id);
1956 		}
1957 	} else {
1958 		link.link_status = ETH_LINK_UP;
1959 	}
1960 	virtio_dev_atomic_write_link_status(dev, &link);
1961 
1962 	return (old.link_status == link.link_status) ? -1 : 0;
1963 }
1964 
1965 static int
1966 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1967 {
1968 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1969 	struct virtio_hw *hw = dev->data->dev_private;
1970 
1971 	if (mask & ETH_VLAN_FILTER_MASK) {
1972 		if (rxmode->hw_vlan_filter &&
1973 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1974 
1975 			PMD_DRV_LOG(NOTICE,
1976 				"vlan filtering not available on this host");
1977 
1978 			return -ENOTSUP;
1979 		}
1980 	}
1981 
1982 	if (mask & ETH_VLAN_STRIP_MASK)
1983 		hw->vlan_strip = rxmode->hw_vlan_strip;
1984 
1985 	return 0;
1986 }
1987 
1988 static void
1989 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1990 {
1991 	uint64_t tso_mask, host_features;
1992 	struct virtio_hw *hw = dev->data->dev_private;
1993 
1994 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
1995 
1996 	dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
1997 	dev_info->max_rx_queues =
1998 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1999 	dev_info->max_tx_queues =
2000 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2001 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2002 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2003 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2004 	dev_info->default_txconf = (struct rte_eth_txconf) {
2005 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
2006 	};
2007 
2008 	host_features = VTPCI_OPS(hw)->get_features(hw);
2009 	dev_info->rx_offload_capa = 0;
2010 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2011 		dev_info->rx_offload_capa |=
2012 			DEV_RX_OFFLOAD_TCP_CKSUM |
2013 			DEV_RX_OFFLOAD_UDP_CKSUM;
2014 	}
2015 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2016 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2017 	if ((host_features & tso_mask) == tso_mask)
2018 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2019 
2020 	dev_info->tx_offload_capa = 0;
2021 	if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2022 		dev_info->tx_offload_capa |=
2023 			DEV_TX_OFFLOAD_UDP_CKSUM |
2024 			DEV_TX_OFFLOAD_TCP_CKSUM;
2025 	}
2026 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2027 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2028 	if ((hw->guest_features & tso_mask) == tso_mask)
2029 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2030 }
2031 
2032 /*
2033  * It enables testpmd to collect per queue stats.
2034  */
2035 static int
2036 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2037 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2038 __rte_unused uint8_t is_rx)
2039 {
2040 	return 0;
2041 }
2042 
2043 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2044 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2045 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2046