xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision d4f44265a1d74d270da2a6deaec8938b2f1b49d4)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <stdint.h>
35 #include <string.h>
36 #include <stdio.h>
37 #include <errno.h>
38 #include <unistd.h>
39 
40 #include <rte_ethdev.h>
41 #include <rte_ethdev_pci.h>
42 #include <rte_memcpy.h>
43 #include <rte_string_fns.h>
44 #include <rte_memzone.h>
45 #include <rte_malloc.h>
46 #include <rte_atomic.h>
47 #include <rte_branch_prediction.h>
48 #include <rte_pci.h>
49 #include <rte_ether.h>
50 #include <rte_common.h>
51 #include <rte_errno.h>
52 
53 #include <rte_memory.h>
54 #include <rte_eal.h>
55 #include <rte_dev.h>
56 
57 #include "virtio_ethdev.h"
58 #include "virtio_pci.h"
59 #include "virtio_logs.h"
60 #include "virtqueue.h"
61 #include "virtio_rxtx.h"
62 
63 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
64 static int  virtio_dev_configure(struct rte_eth_dev *dev);
65 static int  virtio_dev_start(struct rte_eth_dev *dev);
66 static void virtio_dev_stop(struct rte_eth_dev *dev);
67 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
68 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
69 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
70 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
71 static void virtio_dev_info_get(struct rte_eth_dev *dev,
72 				struct rte_eth_dev_info *dev_info);
73 static int virtio_dev_link_update(struct rte_eth_dev *dev,
74 	__rte_unused int wait_to_complete);
75 
76 static void virtio_set_hwaddr(struct virtio_hw *hw);
77 static void virtio_get_hwaddr(struct virtio_hw *hw);
78 
79 static void virtio_dev_stats_get(struct rte_eth_dev *dev,
80 				 struct rte_eth_stats *stats);
81 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
82 				 struct rte_eth_xstat *xstats, unsigned n);
83 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
84 				       struct rte_eth_xstat_name *xstats_names,
85 				       unsigned limit);
86 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
87 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
88 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
89 				uint16_t vlan_id, int on);
90 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
91 				struct ether_addr *mac_addr,
92 				uint32_t index, uint32_t vmdq __rte_unused);
93 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
94 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
95 				struct ether_addr *mac_addr);
96 
97 static int virtio_dev_queue_stats_mapping_set(
98 	__rte_unused struct rte_eth_dev *eth_dev,
99 	__rte_unused uint16_t queue_id,
100 	__rte_unused uint8_t stat_idx,
101 	__rte_unused uint8_t is_rx);
102 
103 /*
104  * The set of PCI devices this driver supports
105  */
106 static const struct rte_pci_id pci_id_virtio_map[] = {
107 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
108 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
109 	{ .vendor_id = 0, /* sentinel */ },
110 };
111 
112 struct rte_virtio_xstats_name_off {
113 	char name[RTE_ETH_XSTATS_NAME_SIZE];
114 	unsigned offset;
115 };
116 
117 /* [rt]x_qX_ is prepended to the name string here */
118 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
119 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
120 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
121 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
122 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
123 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
124 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
125 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
126 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
127 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
128 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
129 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
130 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
131 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
132 };
133 
134 /* [rt]x_qX_ is prepended to the name string here */
135 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
136 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
137 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
138 	{"errors",                 offsetof(struct virtnet_tx, stats.errors)},
139 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
140 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
141 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
142 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
143 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
144 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
145 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
146 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
147 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
148 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
149 };
150 
151 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
152 			    sizeof(rte_virtio_rxq_stat_strings[0]))
153 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
154 			    sizeof(rte_virtio_txq_stat_strings[0]))
155 
156 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
157 
158 static int
159 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
160 		int *dlen, int pkt_num)
161 {
162 	uint32_t head, i;
163 	int k, sum = 0;
164 	virtio_net_ctrl_ack status = ~0;
165 	struct virtio_pmd_ctrl result;
166 	struct virtqueue *vq;
167 
168 	ctrl->status = status;
169 
170 	if (!cvq || !cvq->vq) {
171 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
172 		return -1;
173 	}
174 	vq = cvq->vq;
175 	head = vq->vq_desc_head_idx;
176 
177 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
178 		"vq->hw->cvq = %p vq = %p",
179 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
180 
181 	if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
182 		return -1;
183 
184 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
185 		sizeof(struct virtio_pmd_ctrl));
186 
187 	/*
188 	 * Format is enforced in qemu code:
189 	 * One TX packet for header;
190 	 * At least one TX packet per argument;
191 	 * One RX packet for ACK.
192 	 */
193 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
194 	vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
195 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
196 	vq->vq_free_cnt--;
197 	i = vq->vq_ring.desc[head].next;
198 
199 	for (k = 0; k < pkt_num; k++) {
200 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
201 		vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
202 			+ sizeof(struct virtio_net_ctrl_hdr)
203 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
204 		vq->vq_ring.desc[i].len = dlen[k];
205 		sum += dlen[k];
206 		vq->vq_free_cnt--;
207 		i = vq->vq_ring.desc[i].next;
208 	}
209 
210 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
211 	vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
212 			+ sizeof(struct virtio_net_ctrl_hdr);
213 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
214 	vq->vq_free_cnt--;
215 
216 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
217 
218 	vq_update_avail_ring(vq, head);
219 	vq_update_avail_idx(vq);
220 
221 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
222 
223 	virtqueue_notify(vq);
224 
225 	rte_rmb();
226 	while (VIRTQUEUE_NUSED(vq) == 0) {
227 		rte_rmb();
228 		usleep(100);
229 	}
230 
231 	while (VIRTQUEUE_NUSED(vq)) {
232 		uint32_t idx, desc_idx, used_idx;
233 		struct vring_used_elem *uep;
234 
235 		used_idx = (uint32_t)(vq->vq_used_cons_idx
236 				& (vq->vq_nentries - 1));
237 		uep = &vq->vq_ring.used->ring[used_idx];
238 		idx = (uint32_t) uep->id;
239 		desc_idx = idx;
240 
241 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
242 			desc_idx = vq->vq_ring.desc[desc_idx].next;
243 			vq->vq_free_cnt++;
244 		}
245 
246 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
247 		vq->vq_desc_head_idx = idx;
248 
249 		vq->vq_used_cons_idx++;
250 		vq->vq_free_cnt++;
251 	}
252 
253 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
254 			vq->vq_free_cnt, vq->vq_desc_head_idx);
255 
256 	memcpy(&result, cvq->virtio_net_hdr_mz->addr,
257 			sizeof(struct virtio_pmd_ctrl));
258 
259 	return result.status;
260 }
261 
262 static int
263 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
264 {
265 	struct virtio_hw *hw = dev->data->dev_private;
266 	struct virtio_pmd_ctrl ctrl;
267 	int dlen[1];
268 	int ret;
269 
270 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
271 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
272 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
273 
274 	dlen[0] = sizeof(uint16_t);
275 
276 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
277 	if (ret) {
278 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
279 			  "failed, this is too late now...");
280 		return -EINVAL;
281 	}
282 
283 	return 0;
284 }
285 
286 static void
287 virtio_dev_queue_release(void *queue __rte_unused)
288 {
289 	/* do nothing */
290 }
291 
292 static int
293 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
294 {
295 	if (vtpci_queue_idx == hw->max_queue_pairs * 2)
296 		return VTNET_CQ;
297 	else if (vtpci_queue_idx % 2 == 0)
298 		return VTNET_RQ;
299 	else
300 		return VTNET_TQ;
301 }
302 
303 static uint16_t
304 virtio_get_nr_vq(struct virtio_hw *hw)
305 {
306 	uint16_t nr_vq = hw->max_queue_pairs * 2;
307 
308 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
309 		nr_vq += 1;
310 
311 	return nr_vq;
312 }
313 
314 static void
315 virtio_init_vring(struct virtqueue *vq)
316 {
317 	int size = vq->vq_nentries;
318 	struct vring *vr = &vq->vq_ring;
319 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
320 
321 	PMD_INIT_FUNC_TRACE();
322 
323 	/*
324 	 * Reinitialise since virtio port might have been stopped and restarted
325 	 */
326 	memset(ring_mem, 0, vq->vq_ring_size);
327 	vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
328 	vq->vq_used_cons_idx = 0;
329 	vq->vq_desc_head_idx = 0;
330 	vq->vq_avail_idx = 0;
331 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
332 	vq->vq_free_cnt = vq->vq_nentries;
333 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
334 
335 	vring_desc_init(vr->desc, size);
336 
337 	/*
338 	 * Disable device(host) interrupting guest
339 	 */
340 	virtqueue_disable_intr(vq);
341 }
342 
343 static int
344 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
345 {
346 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
347 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
348 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
349 	unsigned int vq_size, size;
350 	struct virtio_hw *hw = dev->data->dev_private;
351 	struct virtnet_rx *rxvq = NULL;
352 	struct virtnet_tx *txvq = NULL;
353 	struct virtnet_ctl *cvq = NULL;
354 	struct virtqueue *vq;
355 	size_t sz_hdr_mz = 0;
356 	void *sw_ring = NULL;
357 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
358 	int ret;
359 
360 	PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
361 
362 	/*
363 	 * Read the virtqueue size from the Queue Size field
364 	 * Always power of 2 and if 0 virtqueue does not exist
365 	 */
366 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
367 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
368 	if (vq_size == 0) {
369 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
370 		return -EINVAL;
371 	}
372 
373 	if (!rte_is_power_of_2(vq_size)) {
374 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
375 		return -EINVAL;
376 	}
377 
378 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
379 		 dev->data->port_id, vtpci_queue_idx);
380 
381 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
382 				vq_size * sizeof(struct vq_desc_extra),
383 				RTE_CACHE_LINE_SIZE);
384 	if (queue_type == VTNET_TQ) {
385 		/*
386 		 * For each xmit packet, allocate a virtio_net_hdr
387 		 * and indirect ring elements
388 		 */
389 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
390 	} else if (queue_type == VTNET_CQ) {
391 		/* Allocate a page for control vq command, data and status */
392 		sz_hdr_mz = PAGE_SIZE;
393 	}
394 
395 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
396 				SOCKET_ID_ANY);
397 	if (vq == NULL) {
398 		PMD_INIT_LOG(ERR, "can not allocate vq");
399 		return -ENOMEM;
400 	}
401 	hw->vqs[vtpci_queue_idx] = vq;
402 
403 	vq->hw = hw;
404 	vq->vq_queue_index = vtpci_queue_idx;
405 	vq->vq_nentries = vq_size;
406 
407 	/*
408 	 * Reserve a memzone for vring elements
409 	 */
410 	size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
411 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
412 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
413 		     size, vq->vq_ring_size);
414 
415 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
416 					 SOCKET_ID_ANY,
417 					 0, VIRTIO_PCI_VRING_ALIGN);
418 	if (mz == NULL) {
419 		if (rte_errno == EEXIST)
420 			mz = rte_memzone_lookup(vq_name);
421 		if (mz == NULL) {
422 			ret = -ENOMEM;
423 			goto fail_q_alloc;
424 		}
425 	}
426 
427 	memset(mz->addr, 0, sizeof(mz->len));
428 
429 	vq->vq_ring_mem = mz->phys_addr;
430 	vq->vq_ring_virt_mem = mz->addr;
431 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
432 		     (uint64_t)mz->phys_addr);
433 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
434 		     (uint64_t)(uintptr_t)mz->addr);
435 
436 	virtio_init_vring(vq);
437 
438 	if (sz_hdr_mz) {
439 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
440 			 dev->data->port_id, vtpci_queue_idx);
441 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
442 						     SOCKET_ID_ANY, 0,
443 						     RTE_CACHE_LINE_SIZE);
444 		if (hdr_mz == NULL) {
445 			if (rte_errno == EEXIST)
446 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
447 			if (hdr_mz == NULL) {
448 				ret = -ENOMEM;
449 				goto fail_q_alloc;
450 			}
451 		}
452 	}
453 
454 	if (queue_type == VTNET_RQ) {
455 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
456 			       sizeof(vq->sw_ring[0]);
457 
458 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
459 				RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
460 		if (!sw_ring) {
461 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
462 			ret = -ENOMEM;
463 			goto fail_q_alloc;
464 		}
465 
466 		vq->sw_ring = sw_ring;
467 		rxvq = &vq->rxq;
468 		rxvq->vq = vq;
469 		rxvq->port_id = dev->data->port_id;
470 		rxvq->mz = mz;
471 	} else if (queue_type == VTNET_TQ) {
472 		txvq = &vq->txq;
473 		txvq->vq = vq;
474 		txvq->port_id = dev->data->port_id;
475 		txvq->mz = mz;
476 		txvq->virtio_net_hdr_mz = hdr_mz;
477 		txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
478 	} else if (queue_type == VTNET_CQ) {
479 		cvq = &vq->cq;
480 		cvq->vq = vq;
481 		cvq->mz = mz;
482 		cvq->virtio_net_hdr_mz = hdr_mz;
483 		cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
484 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
485 
486 		hw->cvq = cvq;
487 	}
488 
489 	/* For virtio_user case (that is when hw->dev is NULL), we use
490 	 * virtual address. And we need properly set _offset_, please see
491 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
492 	 */
493 	if (!hw->virtio_user_dev)
494 		vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
495 	else {
496 		vq->vq_ring_mem = (uintptr_t)mz->addr;
497 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
498 		if (queue_type == VTNET_TQ)
499 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
500 		else if (queue_type == VTNET_CQ)
501 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
502 	}
503 
504 	if (queue_type == VTNET_TQ) {
505 		struct virtio_tx_region *txr;
506 		unsigned int i;
507 
508 		txr = hdr_mz->addr;
509 		memset(txr, 0, vq_size * sizeof(*txr));
510 		for (i = 0; i < vq_size; i++) {
511 			struct vring_desc *start_dp = txr[i].tx_indir;
512 
513 			vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
514 
515 			/* first indirect descriptor is always the tx header */
516 			start_dp->addr = txvq->virtio_net_hdr_mem
517 				+ i * sizeof(*txr)
518 				+ offsetof(struct virtio_tx_region, tx_hdr);
519 
520 			start_dp->len = hw->vtnet_hdr_size;
521 			start_dp->flags = VRING_DESC_F_NEXT;
522 		}
523 	}
524 
525 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
526 		PMD_INIT_LOG(ERR, "setup_queue failed");
527 		return -EINVAL;
528 	}
529 
530 	return 0;
531 
532 fail_q_alloc:
533 	rte_free(sw_ring);
534 	rte_memzone_free(hdr_mz);
535 	rte_memzone_free(mz);
536 	rte_free(vq);
537 
538 	return ret;
539 }
540 
541 static void
542 virtio_free_queues(struct virtio_hw *hw)
543 {
544 	uint16_t nr_vq = virtio_get_nr_vq(hw);
545 	struct virtqueue *vq;
546 	int queue_type;
547 	uint16_t i;
548 
549 	if (hw->vqs == NULL)
550 		return;
551 
552 	for (i = 0; i < nr_vq; i++) {
553 		vq = hw->vqs[i];
554 		if (!vq)
555 			continue;
556 
557 		queue_type = virtio_get_queue_type(hw, i);
558 		if (queue_type == VTNET_RQ) {
559 			rte_free(vq->sw_ring);
560 			rte_memzone_free(vq->rxq.mz);
561 		} else if (queue_type == VTNET_TQ) {
562 			rte_memzone_free(vq->txq.mz);
563 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
564 		} else {
565 			rte_memzone_free(vq->cq.mz);
566 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
567 		}
568 
569 		rte_free(vq);
570 		hw->vqs[i] = NULL;
571 	}
572 
573 	rte_free(hw->vqs);
574 	hw->vqs = NULL;
575 }
576 
577 static int
578 virtio_alloc_queues(struct rte_eth_dev *dev)
579 {
580 	struct virtio_hw *hw = dev->data->dev_private;
581 	uint16_t nr_vq = virtio_get_nr_vq(hw);
582 	uint16_t i;
583 	int ret;
584 
585 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
586 	if (!hw->vqs) {
587 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
588 		return -ENOMEM;
589 	}
590 
591 	for (i = 0; i < nr_vq; i++) {
592 		ret = virtio_init_queue(dev, i);
593 		if (ret < 0) {
594 			virtio_free_queues(hw);
595 			return ret;
596 		}
597 	}
598 
599 	return 0;
600 }
601 
602 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
603 
604 static void
605 virtio_dev_close(struct rte_eth_dev *dev)
606 {
607 	struct virtio_hw *hw = dev->data->dev_private;
608 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
609 
610 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
611 
612 	/* reset the NIC */
613 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
614 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
615 	if (intr_conf->rxq)
616 		virtio_queues_unbind_intr(dev);
617 
618 	if (intr_conf->lsc || intr_conf->rxq) {
619 		rte_intr_disable(dev->intr_handle);
620 		rte_intr_efd_disable(dev->intr_handle);
621 		rte_free(dev->intr_handle->intr_vec);
622 		dev->intr_handle->intr_vec = NULL;
623 	}
624 
625 	vtpci_reset(hw);
626 	virtio_dev_free_mbufs(dev);
627 	virtio_free_queues(hw);
628 }
629 
630 static void
631 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
632 {
633 	struct virtio_hw *hw = dev->data->dev_private;
634 	struct virtio_pmd_ctrl ctrl;
635 	int dlen[1];
636 	int ret;
637 
638 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
639 		PMD_INIT_LOG(INFO, "host does not support rx control");
640 		return;
641 	}
642 
643 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
644 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
645 	ctrl.data[0] = 1;
646 	dlen[0] = 1;
647 
648 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
649 	if (ret)
650 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
651 }
652 
653 static void
654 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
655 {
656 	struct virtio_hw *hw = dev->data->dev_private;
657 	struct virtio_pmd_ctrl ctrl;
658 	int dlen[1];
659 	int ret;
660 
661 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
662 		PMD_INIT_LOG(INFO, "host does not support rx control");
663 		return;
664 	}
665 
666 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
667 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
668 	ctrl.data[0] = 0;
669 	dlen[0] = 1;
670 
671 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
672 	if (ret)
673 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
674 }
675 
676 static void
677 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
678 {
679 	struct virtio_hw *hw = dev->data->dev_private;
680 	struct virtio_pmd_ctrl ctrl;
681 	int dlen[1];
682 	int ret;
683 
684 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
685 		PMD_INIT_LOG(INFO, "host does not support rx control");
686 		return;
687 	}
688 
689 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
690 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
691 	ctrl.data[0] = 1;
692 	dlen[0] = 1;
693 
694 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
695 	if (ret)
696 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
697 }
698 
699 static void
700 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
701 {
702 	struct virtio_hw *hw = dev->data->dev_private;
703 	struct virtio_pmd_ctrl ctrl;
704 	int dlen[1];
705 	int ret;
706 
707 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
708 		PMD_INIT_LOG(INFO, "host does not support rx control");
709 		return;
710 	}
711 
712 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
713 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
714 	ctrl.data[0] = 0;
715 	dlen[0] = 1;
716 
717 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
718 	if (ret)
719 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
720 }
721 
722 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
723 static int
724 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
725 {
726 	struct virtio_hw *hw = dev->data->dev_private;
727 	uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
728 				 hw->vtnet_hdr_size;
729 	uint32_t frame_size = mtu + ether_hdr_len;
730 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
731 
732 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
733 
734 	if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
735 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
736 			ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
737 		return -EINVAL;
738 	}
739 	return 0;
740 }
741 
742 static int
743 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
744 {
745 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
746 	struct virtqueue *vq = rxvq->vq;
747 
748 	virtqueue_enable_intr(vq);
749 	return 0;
750 }
751 
752 static int
753 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
754 {
755 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
756 	struct virtqueue *vq = rxvq->vq;
757 
758 	virtqueue_disable_intr(vq);
759 	return 0;
760 }
761 
762 /*
763  * dev_ops for virtio, bare necessities for basic operation
764  */
765 static const struct eth_dev_ops virtio_eth_dev_ops = {
766 	.dev_configure           = virtio_dev_configure,
767 	.dev_start               = virtio_dev_start,
768 	.dev_stop                = virtio_dev_stop,
769 	.dev_close               = virtio_dev_close,
770 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
771 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
772 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
773 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
774 	.mtu_set                 = virtio_mtu_set,
775 	.dev_infos_get           = virtio_dev_info_get,
776 	.stats_get               = virtio_dev_stats_get,
777 	.xstats_get              = virtio_dev_xstats_get,
778 	.xstats_get_names        = virtio_dev_xstats_get_names,
779 	.stats_reset             = virtio_dev_stats_reset,
780 	.xstats_reset            = virtio_dev_stats_reset,
781 	.link_update             = virtio_dev_link_update,
782 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
783 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
784 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
785 	.rx_queue_release        = virtio_dev_queue_release,
786 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
787 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
788 	.tx_queue_release        = virtio_dev_queue_release,
789 	/* collect stats per queue */
790 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
791 	.vlan_filter_set         = virtio_vlan_filter_set,
792 	.mac_addr_add            = virtio_mac_addr_add,
793 	.mac_addr_remove         = virtio_mac_addr_remove,
794 	.mac_addr_set            = virtio_mac_addr_set,
795 };
796 
797 static inline int
798 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
799 				struct rte_eth_link *link)
800 {
801 	struct rte_eth_link *dst = link;
802 	struct rte_eth_link *src = &(dev->data->dev_link);
803 
804 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
805 			*(uint64_t *)src) == 0)
806 		return -1;
807 
808 	return 0;
809 }
810 
811 /**
812  * Atomically writes the link status information into global
813  * structure rte_eth_dev.
814  *
815  * @param dev
816  *   - Pointer to the structure rte_eth_dev to read from.
817  *   - Pointer to the buffer to be saved with the link status.
818  *
819  * @return
820  *   - On success, zero.
821  *   - On failure, negative value.
822  */
823 static inline int
824 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
825 		struct rte_eth_link *link)
826 {
827 	struct rte_eth_link *dst = &(dev->data->dev_link);
828 	struct rte_eth_link *src = link;
829 
830 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
831 					*(uint64_t *)src) == 0)
832 		return -1;
833 
834 	return 0;
835 }
836 
837 static void
838 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
839 {
840 	unsigned i;
841 
842 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
843 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
844 		if (txvq == NULL)
845 			continue;
846 
847 		stats->opackets += txvq->stats.packets;
848 		stats->obytes += txvq->stats.bytes;
849 		stats->oerrors += txvq->stats.errors;
850 
851 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
852 			stats->q_opackets[i] = txvq->stats.packets;
853 			stats->q_obytes[i] = txvq->stats.bytes;
854 		}
855 	}
856 
857 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
858 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
859 		if (rxvq == NULL)
860 			continue;
861 
862 		stats->ipackets += rxvq->stats.packets;
863 		stats->ibytes += rxvq->stats.bytes;
864 		stats->ierrors += rxvq->stats.errors;
865 
866 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
867 			stats->q_ipackets[i] = rxvq->stats.packets;
868 			stats->q_ibytes[i] = rxvq->stats.bytes;
869 		}
870 	}
871 
872 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
873 }
874 
875 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
876 				       struct rte_eth_xstat_name *xstats_names,
877 				       __rte_unused unsigned limit)
878 {
879 	unsigned i;
880 	unsigned count = 0;
881 	unsigned t;
882 
883 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
884 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
885 
886 	if (xstats_names != NULL) {
887 		/* Note: limit checked in rte_eth_xstats_names() */
888 
889 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
890 			struct virtqueue *rxvq = dev->data->rx_queues[i];
891 			if (rxvq == NULL)
892 				continue;
893 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
894 				snprintf(xstats_names[count].name,
895 					sizeof(xstats_names[count].name),
896 					"rx_q%u_%s", i,
897 					rte_virtio_rxq_stat_strings[t].name);
898 				count++;
899 			}
900 		}
901 
902 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
903 			struct virtqueue *txvq = dev->data->tx_queues[i];
904 			if (txvq == NULL)
905 				continue;
906 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
907 				snprintf(xstats_names[count].name,
908 					sizeof(xstats_names[count].name),
909 					"tx_q%u_%s", i,
910 					rte_virtio_txq_stat_strings[t].name);
911 				count++;
912 			}
913 		}
914 		return count;
915 	}
916 	return nstats;
917 }
918 
919 static int
920 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
921 		      unsigned n)
922 {
923 	unsigned i;
924 	unsigned count = 0;
925 
926 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
927 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
928 
929 	if (n < nstats)
930 		return nstats;
931 
932 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
933 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
934 
935 		if (rxvq == NULL)
936 			continue;
937 
938 		unsigned t;
939 
940 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
941 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
942 				rte_virtio_rxq_stat_strings[t].offset);
943 			xstats[count].id = count;
944 			count++;
945 		}
946 	}
947 
948 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
949 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
950 
951 		if (txvq == NULL)
952 			continue;
953 
954 		unsigned t;
955 
956 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
957 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
958 				rte_virtio_txq_stat_strings[t].offset);
959 			xstats[count].id = count;
960 			count++;
961 		}
962 	}
963 
964 	return count;
965 }
966 
967 static void
968 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
969 {
970 	virtio_update_stats(dev, stats);
971 }
972 
973 static void
974 virtio_dev_stats_reset(struct rte_eth_dev *dev)
975 {
976 	unsigned int i;
977 
978 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
979 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
980 		if (txvq == NULL)
981 			continue;
982 
983 		txvq->stats.packets = 0;
984 		txvq->stats.bytes = 0;
985 		txvq->stats.errors = 0;
986 		txvq->stats.multicast = 0;
987 		txvq->stats.broadcast = 0;
988 		memset(txvq->stats.size_bins, 0,
989 		       sizeof(txvq->stats.size_bins[0]) * 8);
990 	}
991 
992 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
993 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
994 		if (rxvq == NULL)
995 			continue;
996 
997 		rxvq->stats.packets = 0;
998 		rxvq->stats.bytes = 0;
999 		rxvq->stats.errors = 0;
1000 		rxvq->stats.multicast = 0;
1001 		rxvq->stats.broadcast = 0;
1002 		memset(rxvq->stats.size_bins, 0,
1003 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1004 	}
1005 }
1006 
1007 static void
1008 virtio_set_hwaddr(struct virtio_hw *hw)
1009 {
1010 	vtpci_write_dev_config(hw,
1011 			offsetof(struct virtio_net_config, mac),
1012 			&hw->mac_addr, ETHER_ADDR_LEN);
1013 }
1014 
1015 static void
1016 virtio_get_hwaddr(struct virtio_hw *hw)
1017 {
1018 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1019 		vtpci_read_dev_config(hw,
1020 			offsetof(struct virtio_net_config, mac),
1021 			&hw->mac_addr, ETHER_ADDR_LEN);
1022 	} else {
1023 		eth_random_addr(&hw->mac_addr[0]);
1024 		virtio_set_hwaddr(hw);
1025 	}
1026 }
1027 
1028 static void
1029 virtio_mac_table_set(struct virtio_hw *hw,
1030 		     const struct virtio_net_ctrl_mac *uc,
1031 		     const struct virtio_net_ctrl_mac *mc)
1032 {
1033 	struct virtio_pmd_ctrl ctrl;
1034 	int err, len[2];
1035 
1036 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1037 		PMD_DRV_LOG(INFO, "host does not support mac table");
1038 		return;
1039 	}
1040 
1041 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1042 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1043 
1044 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1045 	memcpy(ctrl.data, uc, len[0]);
1046 
1047 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1048 	memcpy(ctrl.data + len[0], mc, len[1]);
1049 
1050 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1051 	if (err != 0)
1052 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1053 }
1054 
1055 static void
1056 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1057 		    uint32_t index, uint32_t vmdq __rte_unused)
1058 {
1059 	struct virtio_hw *hw = dev->data->dev_private;
1060 	const struct ether_addr *addrs = dev->data->mac_addrs;
1061 	unsigned int i;
1062 	struct virtio_net_ctrl_mac *uc, *mc;
1063 
1064 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1065 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1066 		return;
1067 	}
1068 
1069 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1070 	uc->entries = 0;
1071 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1072 	mc->entries = 0;
1073 
1074 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1075 		const struct ether_addr *addr
1076 			= (i == index) ? mac_addr : addrs + i;
1077 		struct virtio_net_ctrl_mac *tbl
1078 			= is_multicast_ether_addr(addr) ? mc : uc;
1079 
1080 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1081 	}
1082 
1083 	virtio_mac_table_set(hw, uc, mc);
1084 }
1085 
1086 static void
1087 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1088 {
1089 	struct virtio_hw *hw = dev->data->dev_private;
1090 	struct ether_addr *addrs = dev->data->mac_addrs;
1091 	struct virtio_net_ctrl_mac *uc, *mc;
1092 	unsigned int i;
1093 
1094 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1095 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1096 		return;
1097 	}
1098 
1099 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1100 	uc->entries = 0;
1101 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1102 	mc->entries = 0;
1103 
1104 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1105 		struct virtio_net_ctrl_mac *tbl;
1106 
1107 		if (i == index || is_zero_ether_addr(addrs + i))
1108 			continue;
1109 
1110 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1111 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1112 	}
1113 
1114 	virtio_mac_table_set(hw, uc, mc);
1115 }
1116 
1117 static void
1118 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1119 {
1120 	struct virtio_hw *hw = dev->data->dev_private;
1121 
1122 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1123 
1124 	/* Use atomic update if available */
1125 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1126 		struct virtio_pmd_ctrl ctrl;
1127 		int len = ETHER_ADDR_LEN;
1128 
1129 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1130 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1131 
1132 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1133 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
1134 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1135 		virtio_set_hwaddr(hw);
1136 }
1137 
1138 static int
1139 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1140 {
1141 	struct virtio_hw *hw = dev->data->dev_private;
1142 	struct virtio_pmd_ctrl ctrl;
1143 	int len;
1144 
1145 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1146 		return -ENOTSUP;
1147 
1148 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1149 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1150 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1151 	len = sizeof(vlan_id);
1152 
1153 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1154 }
1155 
1156 static int
1157 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1158 {
1159 	uint64_t host_features;
1160 
1161 	/* Prepare guest_features: feature that driver wants to support */
1162 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1163 		req_features);
1164 
1165 	/* Read device(host) feature bits */
1166 	host_features = VTPCI_OPS(hw)->get_features(hw);
1167 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1168 		host_features);
1169 
1170 	/* If supported, ensure MTU value is valid before acknowledging it. */
1171 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1172 		struct virtio_net_config config;
1173 
1174 		vtpci_read_dev_config(hw,
1175 			offsetof(struct virtio_net_config, mtu),
1176 			&config.mtu, sizeof(config.mtu));
1177 
1178 		if (config.mtu < ETHER_MIN_MTU)
1179 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1180 	}
1181 
1182 	/*
1183 	 * Negotiate features: Subset of device feature bits are written back
1184 	 * guest feature bits.
1185 	 */
1186 	hw->guest_features = req_features;
1187 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1188 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1189 		hw->guest_features);
1190 
1191 	if (hw->modern) {
1192 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1193 			PMD_INIT_LOG(ERR,
1194 				"VIRTIO_F_VERSION_1 features is not enabled.");
1195 			return -1;
1196 		}
1197 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1198 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1199 			PMD_INIT_LOG(ERR,
1200 				"failed to set FEATURES_OK status!");
1201 			return -1;
1202 		}
1203 	}
1204 
1205 	hw->req_guest_features = req_features;
1206 
1207 	return 0;
1208 }
1209 
1210 /*
1211  * Process Virtio Config changed interrupt and call the callback
1212  * if link state changed.
1213  */
1214 void
1215 virtio_interrupt_handler(void *param)
1216 {
1217 	struct rte_eth_dev *dev = param;
1218 	struct virtio_hw *hw = dev->data->dev_private;
1219 	uint8_t isr;
1220 
1221 	/* Read interrupt status which clears interrupt */
1222 	isr = vtpci_isr(hw);
1223 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1224 
1225 	if (rte_intr_enable(dev->intr_handle) < 0)
1226 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1227 
1228 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1229 		if (virtio_dev_link_update(dev, 0) == 0)
1230 			_rte_eth_dev_callback_process(dev,
1231 						      RTE_ETH_EVENT_INTR_LSC, NULL);
1232 	}
1233 
1234 }
1235 
1236 static void
1237 rx_func_get(struct rte_eth_dev *eth_dev)
1238 {
1239 	struct virtio_hw *hw = eth_dev->data->dev_private;
1240 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1241 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1242 	else
1243 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1244 }
1245 
1246 /* Only support 1:1 queue/interrupt mapping so far.
1247  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1248  * interrupt vectors (<N+1).
1249  */
1250 static int
1251 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1252 {
1253 	uint32_t i;
1254 	struct virtio_hw *hw = dev->data->dev_private;
1255 
1256 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1257 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1258 		dev->intr_handle->intr_vec[i] = i + 1;
1259 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1260 						 VIRTIO_MSI_NO_VECTOR) {
1261 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1262 			return -EBUSY;
1263 		}
1264 	}
1265 
1266 	return 0;
1267 }
1268 
1269 static void
1270 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1271 {
1272 	uint32_t i;
1273 	struct virtio_hw *hw = dev->data->dev_private;
1274 
1275 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1276 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1277 		VTPCI_OPS(hw)->set_queue_irq(hw,
1278 					     hw->vqs[i * VTNET_CQ],
1279 					     VIRTIO_MSI_NO_VECTOR);
1280 }
1281 
1282 static int
1283 virtio_configure_intr(struct rte_eth_dev *dev)
1284 {
1285 	struct virtio_hw *hw = dev->data->dev_private;
1286 
1287 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1288 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1289 		return -ENOTSUP;
1290 	}
1291 
1292 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1293 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1294 		return -1;
1295 	}
1296 
1297 	if (!dev->intr_handle->intr_vec) {
1298 		dev->intr_handle->intr_vec =
1299 			rte_zmalloc("intr_vec",
1300 				    hw->max_queue_pairs * sizeof(int), 0);
1301 		if (!dev->intr_handle->intr_vec) {
1302 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1303 				     hw->max_queue_pairs);
1304 			return -ENOMEM;
1305 		}
1306 	}
1307 
1308 	/* Re-register callback to update max_intr */
1309 	rte_intr_callback_unregister(dev->intr_handle,
1310 				     virtio_interrupt_handler,
1311 				     dev);
1312 	rte_intr_callback_register(dev->intr_handle,
1313 				   virtio_interrupt_handler,
1314 				   dev);
1315 
1316 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1317 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1318 	 * And for legacy devices, this should be done before queue/vec binding
1319 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1320 	 * (22) will be ignored.
1321 	 */
1322 	if (rte_intr_enable(dev->intr_handle) < 0) {
1323 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1324 		return -1;
1325 	}
1326 
1327 	if (virtio_queues_bind_intr(dev) < 0) {
1328 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1329 		return -1;
1330 	}
1331 
1332 	return 0;
1333 }
1334 
1335 /* reset device and renegotiate features if needed */
1336 static int
1337 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1338 {
1339 	struct virtio_hw *hw = eth_dev->data->dev_private;
1340 	struct virtio_net_config *config;
1341 	struct virtio_net_config local_config;
1342 	struct rte_pci_device *pci_dev = NULL;
1343 	int ret;
1344 
1345 	/* Reset the device although not necessary at startup */
1346 	vtpci_reset(hw);
1347 
1348 	/* Tell the host we've noticed this device. */
1349 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1350 
1351 	/* Tell the host we've known how to drive the device. */
1352 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1353 	if (virtio_negotiate_features(hw, req_features) < 0)
1354 		return -1;
1355 
1356 	if (!hw->virtio_user_dev) {
1357 		pci_dev = RTE_DEV_TO_PCI(eth_dev->device);
1358 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1359 	}
1360 
1361 	eth_dev->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
1362 	/* If host does not support both status and MSI-X then disable LSC */
1363 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->use_msix)
1364 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1365 	else
1366 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1367 
1368 	rx_func_get(eth_dev);
1369 
1370 	/* Setting up rx_header size for the device */
1371 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1372 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1373 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1374 	else
1375 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1376 
1377 	/* Copy the permanent MAC address to: virtio_hw */
1378 	virtio_get_hwaddr(hw);
1379 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1380 			&eth_dev->data->mac_addrs[0]);
1381 	PMD_INIT_LOG(DEBUG,
1382 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1383 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1384 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1385 
1386 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1387 		config = &local_config;
1388 
1389 		vtpci_read_dev_config(hw,
1390 			offsetof(struct virtio_net_config, mac),
1391 			&config->mac, sizeof(config->mac));
1392 
1393 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1394 			vtpci_read_dev_config(hw,
1395 				offsetof(struct virtio_net_config, status),
1396 				&config->status, sizeof(config->status));
1397 		} else {
1398 			PMD_INIT_LOG(DEBUG,
1399 				     "VIRTIO_NET_F_STATUS is not supported");
1400 			config->status = 0;
1401 		}
1402 
1403 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1404 			vtpci_read_dev_config(hw,
1405 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1406 				&config->max_virtqueue_pairs,
1407 				sizeof(config->max_virtqueue_pairs));
1408 		} else {
1409 			PMD_INIT_LOG(DEBUG,
1410 				     "VIRTIO_NET_F_MQ is not supported");
1411 			config->max_virtqueue_pairs = 1;
1412 		}
1413 
1414 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1415 
1416 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1417 			vtpci_read_dev_config(hw,
1418 				offsetof(struct virtio_net_config, mtu),
1419 				&config->mtu,
1420 				sizeof(config->mtu));
1421 
1422 			/*
1423 			 * MTU value has already been checked at negotiation
1424 			 * time, but check again in case it has changed since
1425 			 * then, which should not happen.
1426 			 */
1427 			if (config->mtu < ETHER_MIN_MTU) {
1428 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1429 						config->mtu);
1430 				return -1;
1431 			}
1432 
1433 			hw->max_mtu = config->mtu;
1434 			/* Set initial MTU to maximum one supported by vhost */
1435 			eth_dev->data->mtu = config->mtu;
1436 
1437 		} else {
1438 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1439 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1440 		}
1441 
1442 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1443 				config->max_virtqueue_pairs);
1444 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1445 		PMD_INIT_LOG(DEBUG,
1446 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1447 				config->mac[0], config->mac[1],
1448 				config->mac[2], config->mac[3],
1449 				config->mac[4], config->mac[5]);
1450 	} else {
1451 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1452 		hw->max_queue_pairs = 1;
1453 	}
1454 
1455 	ret = virtio_alloc_queues(eth_dev);
1456 	if (ret < 0)
1457 		return ret;
1458 
1459 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1460 		if (virtio_configure_intr(eth_dev) < 0) {
1461 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1462 			return -1;
1463 		}
1464 	}
1465 
1466 	vtpci_reinit_complete(hw);
1467 
1468 	if (pci_dev)
1469 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1470 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1471 			pci_dev->id.device_id);
1472 
1473 	return 0;
1474 }
1475 
1476 /*
1477  * Remap the PCI device again (IO port map for legacy device and
1478  * memory map for modern device), so that the secondary process
1479  * could have the PCI initiated correctly.
1480  */
1481 static int
1482 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1483 {
1484 	if (hw->modern) {
1485 		/*
1486 		 * We don't have to re-parse the PCI config space, since
1487 		 * rte_eal_pci_map_device() makes sure the mapped address
1488 		 * in secondary process would equal to the one mapped in
1489 		 * the primary process: error will be returned if that
1490 		 * requirement is not met.
1491 		 *
1492 		 * That said, we could simply reuse all cap pointers
1493 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1494 		 * primary process, which is stored in shared memory.
1495 		 */
1496 		if (rte_eal_pci_map_device(pci_dev)) {
1497 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1498 			return -1;
1499 		}
1500 	} else {
1501 		if (rte_eal_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1502 			return -1;
1503 	}
1504 
1505 	return 0;
1506 }
1507 
1508 static void
1509 virtio_set_vtpci_ops(struct virtio_hw *hw)
1510 {
1511 #ifdef RTE_VIRTIO_USER
1512 	if (hw->virtio_user_dev)
1513 		VTPCI_OPS(hw) = &virtio_user_ops;
1514 	else
1515 #endif
1516 	if (hw->modern)
1517 		VTPCI_OPS(hw) = &modern_ops;
1518 	else
1519 		VTPCI_OPS(hw) = &legacy_ops;
1520 }
1521 
1522 /*
1523  * This function is based on probe() function in virtio_pci.c
1524  * It returns 0 on success.
1525  */
1526 int
1527 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1528 {
1529 	struct virtio_hw *hw = eth_dev->data->dev_private;
1530 	int ret;
1531 
1532 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1533 
1534 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1535 	eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1536 
1537 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1538 		if (!hw->virtio_user_dev) {
1539 			ret = virtio_remap_pci(RTE_DEV_TO_PCI(eth_dev->device),
1540 					       hw);
1541 			if (ret)
1542 				return ret;
1543 		}
1544 
1545 		virtio_set_vtpci_ops(hw);
1546 		if (hw->use_simple_rxtx) {
1547 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1548 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1549 		} else {
1550 			rx_func_get(eth_dev);
1551 		}
1552 		return 0;
1553 	}
1554 
1555 	/* Allocate memory for storing MAC addresses */
1556 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1557 	if (eth_dev->data->mac_addrs == NULL) {
1558 		PMD_INIT_LOG(ERR,
1559 			"Failed to allocate %d bytes needed to store MAC addresses",
1560 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1561 		return -ENOMEM;
1562 	}
1563 
1564 	hw->port_id = eth_dev->data->port_id;
1565 	/* For virtio_user case the hw->virtio_user_dev is populated by
1566 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1567 	 */
1568 	if (!hw->virtio_user_dev) {
1569 		ret = vtpci_init(RTE_DEV_TO_PCI(eth_dev->device), hw);
1570 		if (ret)
1571 			return ret;
1572 	}
1573 
1574 	/* reset device and negotiate default features */
1575 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1576 	if (ret < 0)
1577 		return ret;
1578 
1579 	/* Setup interrupt callback  */
1580 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1581 		rte_intr_callback_register(eth_dev->intr_handle,
1582 			virtio_interrupt_handler, eth_dev);
1583 
1584 	return 0;
1585 }
1586 
1587 static int
1588 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1589 {
1590 	PMD_INIT_FUNC_TRACE();
1591 
1592 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1593 		return -EPERM;
1594 
1595 	virtio_dev_stop(eth_dev);
1596 	virtio_dev_close(eth_dev);
1597 
1598 	eth_dev->dev_ops = NULL;
1599 	eth_dev->tx_pkt_burst = NULL;
1600 	eth_dev->rx_pkt_burst = NULL;
1601 
1602 	rte_free(eth_dev->data->mac_addrs);
1603 	eth_dev->data->mac_addrs = NULL;
1604 
1605 	/* reset interrupt callback  */
1606 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1607 		rte_intr_callback_unregister(eth_dev->intr_handle,
1608 						virtio_interrupt_handler,
1609 						eth_dev);
1610 	if (eth_dev->device)
1611 		rte_eal_pci_unmap_device(RTE_DEV_TO_PCI(eth_dev->device));
1612 
1613 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1614 
1615 	return 0;
1616 }
1617 
1618 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1619 	struct rte_pci_device *pci_dev)
1620 {
1621 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1622 		eth_virtio_dev_init);
1623 }
1624 
1625 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1626 {
1627 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1628 }
1629 
1630 static struct rte_pci_driver rte_virtio_pmd = {
1631 	.driver = {
1632 		.name = "net_virtio",
1633 	},
1634 	.id_table = pci_id_virtio_map,
1635 	.drv_flags = 0,
1636 	.probe = eth_virtio_pci_probe,
1637 	.remove = eth_virtio_pci_remove,
1638 };
1639 
1640 RTE_INIT(rte_virtio_pmd_init);
1641 static void
1642 rte_virtio_pmd_init(void)
1643 {
1644 	if (rte_eal_iopl_init() != 0) {
1645 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1646 		return;
1647 	}
1648 
1649 	rte_eal_pci_register(&rte_virtio_pmd);
1650 }
1651 
1652 /*
1653  * Configure virtio device
1654  * It returns 0 on success.
1655  */
1656 static int
1657 virtio_dev_configure(struct rte_eth_dev *dev)
1658 {
1659 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1660 	struct virtio_hw *hw = dev->data->dev_private;
1661 	uint64_t req_features;
1662 	int ret;
1663 
1664 	PMD_INIT_LOG(DEBUG, "configure");
1665 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1666 	if (rxmode->hw_ip_checksum)
1667 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1668 	if (rxmode->enable_lro)
1669 		req_features |=
1670 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1671 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1672 
1673 	/* if request features changed, reinit the device */
1674 	if (req_features != hw->req_guest_features) {
1675 		ret = virtio_init_device(dev, req_features);
1676 		if (ret < 0)
1677 			return ret;
1678 	}
1679 
1680 	if (rxmode->hw_ip_checksum &&
1681 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1682 		PMD_DRV_LOG(NOTICE,
1683 			"rx ip checksum not available on this host");
1684 		return -ENOTSUP;
1685 	}
1686 
1687 	if (rxmode->enable_lro &&
1688 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1689 			!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1690 		PMD_DRV_LOG(NOTICE,
1691 			"lro not available on this host");
1692 		return -ENOTSUP;
1693 	}
1694 
1695 	/* start control queue */
1696 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1697 		virtio_dev_cq_start(dev);
1698 
1699 	hw->vlan_strip = rxmode->hw_vlan_strip;
1700 
1701 	if (rxmode->hw_vlan_filter
1702 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1703 		PMD_DRV_LOG(NOTICE,
1704 			    "vlan filtering not available on this host");
1705 		return -ENOTSUP;
1706 	}
1707 
1708 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1709 		/* Enable vector (0) for Link State Intrerrupt */
1710 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1711 				VIRTIO_MSI_NO_VECTOR) {
1712 			PMD_DRV_LOG(ERR, "failed to set config vector");
1713 			return -EBUSY;
1714 		}
1715 
1716 	return 0;
1717 }
1718 
1719 
1720 static int
1721 virtio_dev_start(struct rte_eth_dev *dev)
1722 {
1723 	uint16_t nb_queues, i;
1724 	struct virtnet_rx *rxvq;
1725 	struct virtnet_tx *txvq __rte_unused;
1726 	struct virtio_hw *hw = dev->data->dev_private;
1727 
1728 	/* check if lsc interrupt feature is enabled */
1729 	if (dev->data->dev_conf.intr_conf.lsc) {
1730 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1731 			PMD_DRV_LOG(ERR, "link status not supported by host");
1732 			return -ENOTSUP;
1733 		}
1734 	}
1735 
1736 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1737 	 * in device configure, but it could be unmapped  when device is
1738 	 * stopped.
1739 	 */
1740 	if (dev->data->dev_conf.intr_conf.lsc ||
1741 	    dev->data->dev_conf.intr_conf.rxq) {
1742 		rte_intr_disable(dev->intr_handle);
1743 
1744 		if (rte_intr_enable(dev->intr_handle) < 0) {
1745 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1746 			return -EIO;
1747 		}
1748 	}
1749 
1750 	/*Notify the backend
1751 	 *Otherwise the tap backend might already stop its queue due to fullness.
1752 	 *vhost backend will have no chance to be waked up
1753 	 */
1754 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1755 	if (hw->max_queue_pairs > 1) {
1756 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1757 			return -EINVAL;
1758 	}
1759 
1760 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1761 
1762 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1763 		rxvq = dev->data->rx_queues[i];
1764 		virtqueue_notify(rxvq->vq);
1765 	}
1766 
1767 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1768 
1769 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1770 		rxvq = dev->data->rx_queues[i];
1771 		VIRTQUEUE_DUMP(rxvq->vq);
1772 	}
1773 
1774 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1775 		txvq = dev->data->tx_queues[i];
1776 		VIRTQUEUE_DUMP(txvq->vq);
1777 	}
1778 
1779 	hw->started = 1;
1780 
1781 	/* Initialize Link state */
1782 	virtio_dev_link_update(dev, 0);
1783 
1784 	return 0;
1785 }
1786 
1787 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1788 {
1789 	struct rte_mbuf *buf;
1790 	int i, mbuf_num = 0;
1791 
1792 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1793 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1794 
1795 		PMD_INIT_LOG(DEBUG,
1796 			     "Before freeing rxq[%d] used and unused buf", i);
1797 		VIRTQUEUE_DUMP(rxvq->vq);
1798 
1799 		PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1800 		while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1801 			rte_pktmbuf_free(buf);
1802 			mbuf_num++;
1803 		}
1804 
1805 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1806 		PMD_INIT_LOG(DEBUG,
1807 			     "After freeing rxq[%d] used and unused buf", i);
1808 		VIRTQUEUE_DUMP(rxvq->vq);
1809 	}
1810 
1811 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1812 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1813 
1814 		PMD_INIT_LOG(DEBUG,
1815 			     "Before freeing txq[%d] used and unused bufs",
1816 			     i);
1817 		VIRTQUEUE_DUMP(txvq->vq);
1818 
1819 		mbuf_num = 0;
1820 		while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1821 			rte_pktmbuf_free(buf);
1822 			mbuf_num++;
1823 		}
1824 
1825 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1826 		PMD_INIT_LOG(DEBUG,
1827 			     "After freeing txq[%d] used and unused buf", i);
1828 		VIRTQUEUE_DUMP(txvq->vq);
1829 	}
1830 }
1831 
1832 /*
1833  * Stop device: disable interrupt and mark link down
1834  */
1835 static void
1836 virtio_dev_stop(struct rte_eth_dev *dev)
1837 {
1838 	struct virtio_hw *hw = dev->data->dev_private;
1839 	struct rte_eth_link link;
1840 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1841 
1842 	PMD_INIT_LOG(DEBUG, "stop");
1843 
1844 	if (intr_conf->lsc || intr_conf->rxq)
1845 		rte_intr_disable(dev->intr_handle);
1846 
1847 	hw->started = 0;
1848 	memset(&link, 0, sizeof(link));
1849 	virtio_dev_atomic_write_link_status(dev, &link);
1850 }
1851 
1852 static int
1853 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1854 {
1855 	struct rte_eth_link link, old;
1856 	uint16_t status;
1857 	struct virtio_hw *hw = dev->data->dev_private;
1858 	memset(&link, 0, sizeof(link));
1859 	virtio_dev_atomic_read_link_status(dev, &link);
1860 	old = link;
1861 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1862 	link.link_speed  = SPEED_10G;
1863 
1864 	if (hw->started == 0) {
1865 		link.link_status = ETH_LINK_DOWN;
1866 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1867 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
1868 		vtpci_read_dev_config(hw,
1869 				offsetof(struct virtio_net_config, status),
1870 				&status, sizeof(status));
1871 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1872 			link.link_status = ETH_LINK_DOWN;
1873 			PMD_INIT_LOG(DEBUG, "Port %d is down",
1874 				     dev->data->port_id);
1875 		} else {
1876 			link.link_status = ETH_LINK_UP;
1877 			PMD_INIT_LOG(DEBUG, "Port %d is up",
1878 				     dev->data->port_id);
1879 		}
1880 	} else {
1881 		link.link_status = ETH_LINK_UP;
1882 	}
1883 	virtio_dev_atomic_write_link_status(dev, &link);
1884 
1885 	return (old.link_status == link.link_status) ? -1 : 0;
1886 }
1887 
1888 static void
1889 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1890 {
1891 	uint64_t tso_mask, host_features;
1892 	struct virtio_hw *hw = dev->data->dev_private;
1893 
1894 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
1895 
1896 	dev_info->pci_dev = dev->device ? RTE_DEV_TO_PCI(dev->device) : NULL;
1897 	dev_info->max_rx_queues =
1898 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1899 	dev_info->max_tx_queues =
1900 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1901 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1902 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1903 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1904 	dev_info->default_txconf = (struct rte_eth_txconf) {
1905 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1906 	};
1907 
1908 	host_features = VTPCI_OPS(hw)->get_features(hw);
1909 	dev_info->rx_offload_capa = 0;
1910 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
1911 		dev_info->rx_offload_capa |=
1912 			DEV_RX_OFFLOAD_TCP_CKSUM |
1913 			DEV_RX_OFFLOAD_UDP_CKSUM;
1914 	}
1915 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1916 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1917 	if ((host_features & tso_mask) == tso_mask)
1918 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
1919 
1920 	dev_info->tx_offload_capa = 0;
1921 	if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1922 		dev_info->tx_offload_capa |=
1923 			DEV_TX_OFFLOAD_UDP_CKSUM |
1924 			DEV_TX_OFFLOAD_TCP_CKSUM;
1925 	}
1926 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1927 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
1928 	if ((hw->guest_features & tso_mask) == tso_mask)
1929 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1930 }
1931 
1932 /*
1933  * It enables testpmd to collect per queue stats.
1934  */
1935 static int
1936 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1937 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1938 __rte_unused uint8_t is_rx)
1939 {
1940 	return 0;
1941 }
1942 
1943 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
1944 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
1945 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio");
1946