xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 10def6a1070999a8da2b89e8a8e10fa1eb8b866c)
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_memcpy.h>
42 #include <rte_string_fns.h>
43 #include <rte_memzone.h>
44 #include <rte_malloc.h>
45 #include <rte_atomic.h>
46 #include <rte_branch_prediction.h>
47 #include <rte_pci.h>
48 #include <rte_ether.h>
49 #include <rte_common.h>
50 #include <rte_errno.h>
51 
52 #include <rte_memory.h>
53 #include <rte_eal.h>
54 #include <rte_dev.h>
55 
56 #include "virtio_ethdev.h"
57 #include "virtio_pci.h"
58 #include "virtio_logs.h"
59 #include "virtqueue.h"
60 #include "virtio_rxtx.h"
61 
62 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
63 static int  virtio_dev_configure(struct rte_eth_dev *dev);
64 static int  virtio_dev_start(struct rte_eth_dev *dev);
65 static void virtio_dev_stop(struct rte_eth_dev *dev);
66 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
67 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
68 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
69 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
70 static void virtio_dev_info_get(struct rte_eth_dev *dev,
71 				struct rte_eth_dev_info *dev_info);
72 static int virtio_dev_link_update(struct rte_eth_dev *dev,
73 	__rte_unused int wait_to_complete);
74 
75 static void virtio_set_hwaddr(struct virtio_hw *hw);
76 static void virtio_get_hwaddr(struct virtio_hw *hw);
77 
78 static void virtio_dev_stats_get(struct rte_eth_dev *dev,
79 				 struct rte_eth_stats *stats);
80 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
81 				 struct rte_eth_xstat *xstats, unsigned n);
82 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
83 				       struct rte_eth_xstat_name *xstats_names,
84 				       unsigned limit);
85 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
86 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
87 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
88 				uint16_t vlan_id, int on);
89 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
90 				struct ether_addr *mac_addr,
91 				uint32_t index, uint32_t vmdq __rte_unused);
92 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
93 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
94 				struct ether_addr *mac_addr);
95 
96 static int virtio_dev_queue_stats_mapping_set(
97 	__rte_unused struct rte_eth_dev *eth_dev,
98 	__rte_unused uint16_t queue_id,
99 	__rte_unused uint8_t stat_idx,
100 	__rte_unused uint8_t is_rx);
101 
102 /*
103  * The set of PCI devices this driver supports
104  */
105 static const struct rte_pci_id pci_id_virtio_map[] = {
106 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
107 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
108 	{ .vendor_id = 0, /* sentinel */ },
109 };
110 
111 struct rte_virtio_xstats_name_off {
112 	char name[RTE_ETH_XSTATS_NAME_SIZE];
113 	unsigned offset;
114 };
115 
116 /* [rt]x_qX_ is prepended to the name string here */
117 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
118 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
119 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
120 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
121 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
122 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
123 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
124 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
125 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
126 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
127 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
128 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
129 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
130 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
131 };
132 
133 /* [rt]x_qX_ is prepended to the name string here */
134 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
135 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
136 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
137 	{"errors",                 offsetof(struct virtnet_tx, stats.errors)},
138 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
139 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
140 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
141 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
142 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
143 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
144 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
145 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
146 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
147 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
148 };
149 
150 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
151 			    sizeof(rte_virtio_rxq_stat_strings[0]))
152 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
153 			    sizeof(rte_virtio_txq_stat_strings[0]))
154 
155 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
156 
157 static int
158 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
159 		int *dlen, int pkt_num)
160 {
161 	uint32_t head, i;
162 	int k, sum = 0;
163 	virtio_net_ctrl_ack status = ~0;
164 	struct virtio_pmd_ctrl result;
165 	struct virtqueue *vq;
166 
167 	ctrl->status = status;
168 
169 	if (!cvq || !cvq->vq) {
170 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
171 		return -1;
172 	}
173 	vq = cvq->vq;
174 	head = vq->vq_desc_head_idx;
175 
176 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
177 		"vq->hw->cvq = %p vq = %p",
178 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
179 
180 	if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
181 		return -1;
182 
183 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
184 		sizeof(struct virtio_pmd_ctrl));
185 
186 	/*
187 	 * Format is enforced in qemu code:
188 	 * One TX packet for header;
189 	 * At least one TX packet per argument;
190 	 * One RX packet for ACK.
191 	 */
192 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
193 	vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
194 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
195 	vq->vq_free_cnt--;
196 	i = vq->vq_ring.desc[head].next;
197 
198 	for (k = 0; k < pkt_num; k++) {
199 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
200 		vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
201 			+ sizeof(struct virtio_net_ctrl_hdr)
202 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
203 		vq->vq_ring.desc[i].len = dlen[k];
204 		sum += dlen[k];
205 		vq->vq_free_cnt--;
206 		i = vq->vq_ring.desc[i].next;
207 	}
208 
209 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
210 	vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
211 			+ sizeof(struct virtio_net_ctrl_hdr);
212 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
213 	vq->vq_free_cnt--;
214 
215 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
216 
217 	vq_update_avail_ring(vq, head);
218 	vq_update_avail_idx(vq);
219 
220 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
221 
222 	virtqueue_notify(vq);
223 
224 	rte_rmb();
225 	while (VIRTQUEUE_NUSED(vq) == 0) {
226 		rte_rmb();
227 		usleep(100);
228 	}
229 
230 	while (VIRTQUEUE_NUSED(vq)) {
231 		uint32_t idx, desc_idx, used_idx;
232 		struct vring_used_elem *uep;
233 
234 		used_idx = (uint32_t)(vq->vq_used_cons_idx
235 				& (vq->vq_nentries - 1));
236 		uep = &vq->vq_ring.used->ring[used_idx];
237 		idx = (uint32_t) uep->id;
238 		desc_idx = idx;
239 
240 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
241 			desc_idx = vq->vq_ring.desc[desc_idx].next;
242 			vq->vq_free_cnt++;
243 		}
244 
245 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
246 		vq->vq_desc_head_idx = idx;
247 
248 		vq->vq_used_cons_idx++;
249 		vq->vq_free_cnt++;
250 	}
251 
252 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
253 			vq->vq_free_cnt, vq->vq_desc_head_idx);
254 
255 	memcpy(&result, cvq->virtio_net_hdr_mz->addr,
256 			sizeof(struct virtio_pmd_ctrl));
257 
258 	return result.status;
259 }
260 
261 static int
262 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
263 {
264 	struct virtio_hw *hw = dev->data->dev_private;
265 	struct virtio_pmd_ctrl ctrl;
266 	int dlen[1];
267 	int ret;
268 
269 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
270 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
271 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
272 
273 	dlen[0] = sizeof(uint16_t);
274 
275 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
276 	if (ret) {
277 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
278 			  "failed, this is too late now...");
279 		return -EINVAL;
280 	}
281 
282 	return 0;
283 }
284 
285 static void
286 virtio_dev_queue_release(void *queue __rte_unused)
287 {
288 	/* do nothing */
289 }
290 
291 static int
292 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
293 {
294 	if (vtpci_queue_idx == hw->max_queue_pairs * 2)
295 		return VTNET_CQ;
296 	else if (vtpci_queue_idx % 2 == 0)
297 		return VTNET_RQ;
298 	else
299 		return VTNET_TQ;
300 }
301 
302 static uint16_t
303 virtio_get_nr_vq(struct virtio_hw *hw)
304 {
305 	uint16_t nr_vq = hw->max_queue_pairs * 2;
306 
307 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
308 		nr_vq += 1;
309 
310 	return nr_vq;
311 }
312 
313 static void
314 virtio_init_vring(struct virtqueue *vq)
315 {
316 	int size = vq->vq_nentries;
317 	struct vring *vr = &vq->vq_ring;
318 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
319 
320 	PMD_INIT_FUNC_TRACE();
321 
322 	/*
323 	 * Reinitialise since virtio port might have been stopped and restarted
324 	 */
325 	memset(ring_mem, 0, vq->vq_ring_size);
326 	vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
327 	vq->vq_used_cons_idx = 0;
328 	vq->vq_desc_head_idx = 0;
329 	vq->vq_avail_idx = 0;
330 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
331 	vq->vq_free_cnt = vq->vq_nentries;
332 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
333 
334 	vring_desc_init(vr->desc, size);
335 
336 	/*
337 	 * Disable device(host) interrupting guest
338 	 */
339 	virtqueue_disable_intr(vq);
340 }
341 
342 static int
343 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
344 {
345 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
346 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
347 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
348 	unsigned int vq_size, size;
349 	struct virtio_hw *hw = dev->data->dev_private;
350 	struct virtnet_rx *rxvq = NULL;
351 	struct virtnet_tx *txvq = NULL;
352 	struct virtnet_ctl *cvq = NULL;
353 	struct virtqueue *vq;
354 	size_t sz_hdr_mz = 0;
355 	void *sw_ring = NULL;
356 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
357 	int ret;
358 
359 	PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
360 
361 	/*
362 	 * Read the virtqueue size from the Queue Size field
363 	 * Always power of 2 and if 0 virtqueue does not exist
364 	 */
365 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
366 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
367 	if (vq_size == 0) {
368 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
369 		return -EINVAL;
370 	}
371 
372 	if (!rte_is_power_of_2(vq_size)) {
373 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
374 		return -EINVAL;
375 	}
376 
377 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
378 		 dev->data->port_id, vtpci_queue_idx);
379 
380 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
381 				vq_size * sizeof(struct vq_desc_extra),
382 				RTE_CACHE_LINE_SIZE);
383 	if (queue_type == VTNET_TQ) {
384 		/*
385 		 * For each xmit packet, allocate a virtio_net_hdr
386 		 * and indirect ring elements
387 		 */
388 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
389 	} else if (queue_type == VTNET_CQ) {
390 		/* Allocate a page for control vq command, data and status */
391 		sz_hdr_mz = PAGE_SIZE;
392 	}
393 
394 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
395 				SOCKET_ID_ANY);
396 	if (vq == NULL) {
397 		PMD_INIT_LOG(ERR, "can not allocate vq");
398 		return -ENOMEM;
399 	}
400 	hw->vqs[vtpci_queue_idx] = vq;
401 
402 	vq->hw = hw;
403 	vq->vq_queue_index = vtpci_queue_idx;
404 	vq->vq_nentries = vq_size;
405 
406 	/*
407 	 * Reserve a memzone for vring elements
408 	 */
409 	size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
410 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
411 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
412 		     size, vq->vq_ring_size);
413 
414 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
415 					 SOCKET_ID_ANY,
416 					 0, VIRTIO_PCI_VRING_ALIGN);
417 	if (mz == NULL) {
418 		if (rte_errno == EEXIST)
419 			mz = rte_memzone_lookup(vq_name);
420 		if (mz == NULL) {
421 			ret = -ENOMEM;
422 			goto fail_q_alloc;
423 		}
424 	}
425 
426 	memset(mz->addr, 0, sizeof(mz->len));
427 
428 	vq->vq_ring_mem = mz->phys_addr;
429 	vq->vq_ring_virt_mem = mz->addr;
430 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
431 		     (uint64_t)mz->phys_addr);
432 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
433 		     (uint64_t)(uintptr_t)mz->addr);
434 
435 	virtio_init_vring(vq);
436 
437 	if (sz_hdr_mz) {
438 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
439 			 dev->data->port_id, vtpci_queue_idx);
440 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
441 						     SOCKET_ID_ANY, 0,
442 						     RTE_CACHE_LINE_SIZE);
443 		if (hdr_mz == NULL) {
444 			if (rte_errno == EEXIST)
445 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
446 			if (hdr_mz == NULL) {
447 				ret = -ENOMEM;
448 				goto fail_q_alloc;
449 			}
450 		}
451 	}
452 
453 	if (queue_type == VTNET_RQ) {
454 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
455 			       sizeof(vq->sw_ring[0]);
456 
457 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
458 				RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
459 		if (!sw_ring) {
460 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
461 			ret = -ENOMEM;
462 			goto fail_q_alloc;
463 		}
464 
465 		vq->sw_ring = sw_ring;
466 		rxvq = &vq->rxq;
467 		rxvq->vq = vq;
468 		rxvq->port_id = dev->data->port_id;
469 		rxvq->mz = mz;
470 	} else if (queue_type == VTNET_TQ) {
471 		txvq = &vq->txq;
472 		txvq->vq = vq;
473 		txvq->port_id = dev->data->port_id;
474 		txvq->mz = mz;
475 		txvq->virtio_net_hdr_mz = hdr_mz;
476 		txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
477 	} else if (queue_type == VTNET_CQ) {
478 		cvq = &vq->cq;
479 		cvq->vq = vq;
480 		cvq->mz = mz;
481 		cvq->virtio_net_hdr_mz = hdr_mz;
482 		cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
483 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
484 
485 		hw->cvq = cvq;
486 	}
487 
488 	/* For virtio_user case (that is when hw->dev is NULL), we use
489 	 * virtual address. And we need properly set _offset_, please see
490 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
491 	 */
492 	if (!hw->virtio_user_dev)
493 		vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
494 	else {
495 		vq->vq_ring_mem = (uintptr_t)mz->addr;
496 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
497 		if (queue_type == VTNET_TQ)
498 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
499 		else if (queue_type == VTNET_CQ)
500 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
501 	}
502 
503 	if (queue_type == VTNET_TQ) {
504 		struct virtio_tx_region *txr;
505 		unsigned int i;
506 
507 		txr = hdr_mz->addr;
508 		memset(txr, 0, vq_size * sizeof(*txr));
509 		for (i = 0; i < vq_size; i++) {
510 			struct vring_desc *start_dp = txr[i].tx_indir;
511 
512 			vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
513 
514 			/* first indirect descriptor is always the tx header */
515 			start_dp->addr = txvq->virtio_net_hdr_mem
516 				+ i * sizeof(*txr)
517 				+ offsetof(struct virtio_tx_region, tx_hdr);
518 
519 			start_dp->len = hw->vtnet_hdr_size;
520 			start_dp->flags = VRING_DESC_F_NEXT;
521 		}
522 	}
523 
524 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
525 		PMD_INIT_LOG(ERR, "setup_queue failed");
526 		return -EINVAL;
527 	}
528 
529 	return 0;
530 
531 fail_q_alloc:
532 	rte_free(sw_ring);
533 	rte_memzone_free(hdr_mz);
534 	rte_memzone_free(mz);
535 	rte_free(vq);
536 
537 	return ret;
538 }
539 
540 static void
541 virtio_free_queues(struct virtio_hw *hw)
542 {
543 	uint16_t nr_vq = virtio_get_nr_vq(hw);
544 	struct virtqueue *vq;
545 	int queue_type;
546 	uint16_t i;
547 
548 	for (i = 0; i < nr_vq; i++) {
549 		vq = hw->vqs[i];
550 		if (!vq)
551 			continue;
552 
553 		queue_type = virtio_get_queue_type(hw, i);
554 		if (queue_type == VTNET_RQ) {
555 			rte_free(vq->sw_ring);
556 			rte_memzone_free(vq->rxq.mz);
557 		} else if (queue_type == VTNET_TQ) {
558 			rte_memzone_free(vq->txq.mz);
559 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
560 		} else {
561 			rte_memzone_free(vq->cq.mz);
562 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
563 		}
564 
565 		rte_free(vq);
566 	}
567 
568 	rte_free(hw->vqs);
569 }
570 
571 static int
572 virtio_alloc_queues(struct rte_eth_dev *dev)
573 {
574 	struct virtio_hw *hw = dev->data->dev_private;
575 	uint16_t nr_vq = virtio_get_nr_vq(hw);
576 	uint16_t i;
577 	int ret;
578 
579 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
580 	if (!hw->vqs) {
581 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
582 		return -ENOMEM;
583 	}
584 
585 	for (i = 0; i < nr_vq; i++) {
586 		ret = virtio_init_queue(dev, i);
587 		if (ret < 0) {
588 			virtio_free_queues(hw);
589 			return ret;
590 		}
591 	}
592 
593 	return 0;
594 }
595 
596 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
597 
598 static void
599 virtio_dev_close(struct rte_eth_dev *dev)
600 {
601 	struct virtio_hw *hw = dev->data->dev_private;
602 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
603 
604 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
605 
606 	/* reset the NIC */
607 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
608 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
609 	if (intr_conf->rxq)
610 		virtio_queues_unbind_intr(dev);
611 
612 	if (intr_conf->lsc || intr_conf->rxq) {
613 		rte_intr_disable(dev->intr_handle);
614 		rte_intr_efd_disable(dev->intr_handle);
615 		rte_free(dev->intr_handle->intr_vec);
616 		dev->intr_handle->intr_vec = NULL;
617 	}
618 
619 	vtpci_reset(hw);
620 	virtio_dev_free_mbufs(dev);
621 	virtio_free_queues(hw);
622 }
623 
624 static void
625 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
626 {
627 	struct virtio_hw *hw = dev->data->dev_private;
628 	struct virtio_pmd_ctrl ctrl;
629 	int dlen[1];
630 	int ret;
631 
632 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
633 		PMD_INIT_LOG(INFO, "host does not support rx control");
634 		return;
635 	}
636 
637 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
638 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
639 	ctrl.data[0] = 1;
640 	dlen[0] = 1;
641 
642 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
643 	if (ret)
644 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
645 }
646 
647 static void
648 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
649 {
650 	struct virtio_hw *hw = dev->data->dev_private;
651 	struct virtio_pmd_ctrl ctrl;
652 	int dlen[1];
653 	int ret;
654 
655 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
656 		PMD_INIT_LOG(INFO, "host does not support rx control");
657 		return;
658 	}
659 
660 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
661 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
662 	ctrl.data[0] = 0;
663 	dlen[0] = 1;
664 
665 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
666 	if (ret)
667 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
668 }
669 
670 static void
671 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
672 {
673 	struct virtio_hw *hw = dev->data->dev_private;
674 	struct virtio_pmd_ctrl ctrl;
675 	int dlen[1];
676 	int ret;
677 
678 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
679 		PMD_INIT_LOG(INFO, "host does not support rx control");
680 		return;
681 	}
682 
683 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
684 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
685 	ctrl.data[0] = 1;
686 	dlen[0] = 1;
687 
688 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
689 	if (ret)
690 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
691 }
692 
693 static void
694 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
695 {
696 	struct virtio_hw *hw = dev->data->dev_private;
697 	struct virtio_pmd_ctrl ctrl;
698 	int dlen[1];
699 	int ret;
700 
701 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
702 		PMD_INIT_LOG(INFO, "host does not support rx control");
703 		return;
704 	}
705 
706 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
707 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
708 	ctrl.data[0] = 0;
709 	dlen[0] = 1;
710 
711 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
712 	if (ret)
713 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
714 }
715 
716 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
717 static int
718 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
719 {
720 	struct virtio_hw *hw = dev->data->dev_private;
721 	uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
722 				 hw->vtnet_hdr_size;
723 	uint32_t frame_size = mtu + ether_hdr_len;
724 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
725 
726 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
727 
728 	if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
729 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
730 			ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
731 		return -EINVAL;
732 	}
733 	return 0;
734 }
735 
736 static int
737 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
738 {
739 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
740 	struct virtqueue *vq = rxvq->vq;
741 
742 	virtqueue_enable_intr(vq);
743 	return 0;
744 }
745 
746 static int
747 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
748 {
749 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
750 	struct virtqueue *vq = rxvq->vq;
751 
752 	virtqueue_disable_intr(vq);
753 	return 0;
754 }
755 
756 /*
757  * dev_ops for virtio, bare necessities for basic operation
758  */
759 static const struct eth_dev_ops virtio_eth_dev_ops = {
760 	.dev_configure           = virtio_dev_configure,
761 	.dev_start               = virtio_dev_start,
762 	.dev_stop                = virtio_dev_stop,
763 	.dev_close               = virtio_dev_close,
764 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
765 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
766 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
767 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
768 	.mtu_set                 = virtio_mtu_set,
769 	.dev_infos_get           = virtio_dev_info_get,
770 	.stats_get               = virtio_dev_stats_get,
771 	.xstats_get              = virtio_dev_xstats_get,
772 	.xstats_get_names        = virtio_dev_xstats_get_names,
773 	.stats_reset             = virtio_dev_stats_reset,
774 	.xstats_reset            = virtio_dev_stats_reset,
775 	.link_update             = virtio_dev_link_update,
776 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
777 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
778 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
779 	.rx_queue_release        = virtio_dev_queue_release,
780 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
781 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
782 	.tx_queue_release        = virtio_dev_queue_release,
783 	/* collect stats per queue */
784 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
785 	.vlan_filter_set         = virtio_vlan_filter_set,
786 	.mac_addr_add            = virtio_mac_addr_add,
787 	.mac_addr_remove         = virtio_mac_addr_remove,
788 	.mac_addr_set            = virtio_mac_addr_set,
789 };
790 
791 static inline int
792 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
793 				struct rte_eth_link *link)
794 {
795 	struct rte_eth_link *dst = link;
796 	struct rte_eth_link *src = &(dev->data->dev_link);
797 
798 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
799 			*(uint64_t *)src) == 0)
800 		return -1;
801 
802 	return 0;
803 }
804 
805 /**
806  * Atomically writes the link status information into global
807  * structure rte_eth_dev.
808  *
809  * @param dev
810  *   - Pointer to the structure rte_eth_dev to read from.
811  *   - Pointer to the buffer to be saved with the link status.
812  *
813  * @return
814  *   - On success, zero.
815  *   - On failure, negative value.
816  */
817 static inline int
818 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
819 		struct rte_eth_link *link)
820 {
821 	struct rte_eth_link *dst = &(dev->data->dev_link);
822 	struct rte_eth_link *src = link;
823 
824 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
825 					*(uint64_t *)src) == 0)
826 		return -1;
827 
828 	return 0;
829 }
830 
831 static void
832 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
833 {
834 	unsigned i;
835 
836 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
837 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
838 		if (txvq == NULL)
839 			continue;
840 
841 		stats->opackets += txvq->stats.packets;
842 		stats->obytes += txvq->stats.bytes;
843 		stats->oerrors += txvq->stats.errors;
844 
845 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
846 			stats->q_opackets[i] = txvq->stats.packets;
847 			stats->q_obytes[i] = txvq->stats.bytes;
848 		}
849 	}
850 
851 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
852 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
853 		if (rxvq == NULL)
854 			continue;
855 
856 		stats->ipackets += rxvq->stats.packets;
857 		stats->ibytes += rxvq->stats.bytes;
858 		stats->ierrors += rxvq->stats.errors;
859 
860 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
861 			stats->q_ipackets[i] = rxvq->stats.packets;
862 			stats->q_ibytes[i] = rxvq->stats.bytes;
863 		}
864 	}
865 
866 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
867 }
868 
869 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
870 				       struct rte_eth_xstat_name *xstats_names,
871 				       __rte_unused unsigned limit)
872 {
873 	unsigned i;
874 	unsigned count = 0;
875 	unsigned t;
876 
877 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
878 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
879 
880 	if (xstats_names != NULL) {
881 		/* Note: limit checked in rte_eth_xstats_names() */
882 
883 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
884 			struct virtqueue *rxvq = dev->data->rx_queues[i];
885 			if (rxvq == NULL)
886 				continue;
887 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
888 				snprintf(xstats_names[count].name,
889 					sizeof(xstats_names[count].name),
890 					"rx_q%u_%s", i,
891 					rte_virtio_rxq_stat_strings[t].name);
892 				count++;
893 			}
894 		}
895 
896 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
897 			struct virtqueue *txvq = dev->data->tx_queues[i];
898 			if (txvq == NULL)
899 				continue;
900 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
901 				snprintf(xstats_names[count].name,
902 					sizeof(xstats_names[count].name),
903 					"tx_q%u_%s", i,
904 					rte_virtio_txq_stat_strings[t].name);
905 				count++;
906 			}
907 		}
908 		return count;
909 	}
910 	return nstats;
911 }
912 
913 static int
914 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
915 		      unsigned n)
916 {
917 	unsigned i;
918 	unsigned count = 0;
919 
920 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
921 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
922 
923 	if (n < nstats)
924 		return nstats;
925 
926 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
927 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
928 
929 		if (rxvq == NULL)
930 			continue;
931 
932 		unsigned t;
933 
934 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
935 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
936 				rte_virtio_rxq_stat_strings[t].offset);
937 			xstats[count].id = count;
938 			count++;
939 		}
940 	}
941 
942 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
943 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
944 
945 		if (txvq == NULL)
946 			continue;
947 
948 		unsigned t;
949 
950 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
951 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
952 				rte_virtio_txq_stat_strings[t].offset);
953 			xstats[count].id = count;
954 			count++;
955 		}
956 	}
957 
958 	return count;
959 }
960 
961 static void
962 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
963 {
964 	virtio_update_stats(dev, stats);
965 }
966 
967 static void
968 virtio_dev_stats_reset(struct rte_eth_dev *dev)
969 {
970 	unsigned int i;
971 
972 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
973 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
974 		if (txvq == NULL)
975 			continue;
976 
977 		txvq->stats.packets = 0;
978 		txvq->stats.bytes = 0;
979 		txvq->stats.errors = 0;
980 		txvq->stats.multicast = 0;
981 		txvq->stats.broadcast = 0;
982 		memset(txvq->stats.size_bins, 0,
983 		       sizeof(txvq->stats.size_bins[0]) * 8);
984 	}
985 
986 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
987 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
988 		if (rxvq == NULL)
989 			continue;
990 
991 		rxvq->stats.packets = 0;
992 		rxvq->stats.bytes = 0;
993 		rxvq->stats.errors = 0;
994 		rxvq->stats.multicast = 0;
995 		rxvq->stats.broadcast = 0;
996 		memset(rxvq->stats.size_bins, 0,
997 		       sizeof(rxvq->stats.size_bins[0]) * 8);
998 	}
999 }
1000 
1001 static void
1002 virtio_set_hwaddr(struct virtio_hw *hw)
1003 {
1004 	vtpci_write_dev_config(hw,
1005 			offsetof(struct virtio_net_config, mac),
1006 			&hw->mac_addr, ETHER_ADDR_LEN);
1007 }
1008 
1009 static void
1010 virtio_get_hwaddr(struct virtio_hw *hw)
1011 {
1012 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1013 		vtpci_read_dev_config(hw,
1014 			offsetof(struct virtio_net_config, mac),
1015 			&hw->mac_addr, ETHER_ADDR_LEN);
1016 	} else {
1017 		eth_random_addr(&hw->mac_addr[0]);
1018 		virtio_set_hwaddr(hw);
1019 	}
1020 }
1021 
1022 static void
1023 virtio_mac_table_set(struct virtio_hw *hw,
1024 		     const struct virtio_net_ctrl_mac *uc,
1025 		     const struct virtio_net_ctrl_mac *mc)
1026 {
1027 	struct virtio_pmd_ctrl ctrl;
1028 	int err, len[2];
1029 
1030 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1031 		PMD_DRV_LOG(INFO, "host does not support mac table");
1032 		return;
1033 	}
1034 
1035 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1036 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1037 
1038 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1039 	memcpy(ctrl.data, uc, len[0]);
1040 
1041 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1042 	memcpy(ctrl.data + len[0], mc, len[1]);
1043 
1044 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1045 	if (err != 0)
1046 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1047 }
1048 
1049 static void
1050 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1051 		    uint32_t index, uint32_t vmdq __rte_unused)
1052 {
1053 	struct virtio_hw *hw = dev->data->dev_private;
1054 	const struct ether_addr *addrs = dev->data->mac_addrs;
1055 	unsigned int i;
1056 	struct virtio_net_ctrl_mac *uc, *mc;
1057 
1058 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1059 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1060 		return;
1061 	}
1062 
1063 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1064 	uc->entries = 0;
1065 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1066 	mc->entries = 0;
1067 
1068 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1069 		const struct ether_addr *addr
1070 			= (i == index) ? mac_addr : addrs + i;
1071 		struct virtio_net_ctrl_mac *tbl
1072 			= is_multicast_ether_addr(addr) ? mc : uc;
1073 
1074 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1075 	}
1076 
1077 	virtio_mac_table_set(hw, uc, mc);
1078 }
1079 
1080 static void
1081 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1082 {
1083 	struct virtio_hw *hw = dev->data->dev_private;
1084 	struct ether_addr *addrs = dev->data->mac_addrs;
1085 	struct virtio_net_ctrl_mac *uc, *mc;
1086 	unsigned int i;
1087 
1088 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1089 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1090 		return;
1091 	}
1092 
1093 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1094 	uc->entries = 0;
1095 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1096 	mc->entries = 0;
1097 
1098 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1099 		struct virtio_net_ctrl_mac *tbl;
1100 
1101 		if (i == index || is_zero_ether_addr(addrs + i))
1102 			continue;
1103 
1104 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1105 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1106 	}
1107 
1108 	virtio_mac_table_set(hw, uc, mc);
1109 }
1110 
1111 static void
1112 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1113 {
1114 	struct virtio_hw *hw = dev->data->dev_private;
1115 
1116 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1117 
1118 	/* Use atomic update if available */
1119 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1120 		struct virtio_pmd_ctrl ctrl;
1121 		int len = ETHER_ADDR_LEN;
1122 
1123 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1124 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1125 
1126 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1127 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
1128 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1129 		virtio_set_hwaddr(hw);
1130 }
1131 
1132 static int
1133 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1134 {
1135 	struct virtio_hw *hw = dev->data->dev_private;
1136 	struct virtio_pmd_ctrl ctrl;
1137 	int len;
1138 
1139 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1140 		return -ENOTSUP;
1141 
1142 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1143 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1144 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1145 	len = sizeof(vlan_id);
1146 
1147 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1148 }
1149 
1150 static int
1151 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1152 {
1153 	uint64_t host_features;
1154 
1155 	/* Prepare guest_features: feature that driver wants to support */
1156 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1157 		req_features);
1158 
1159 	/* Read device(host) feature bits */
1160 	host_features = VTPCI_OPS(hw)->get_features(hw);
1161 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1162 		host_features);
1163 
1164 	/* If supported, ensure MTU value is valid before acknowledging it. */
1165 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1166 		struct virtio_net_config config;
1167 
1168 		vtpci_read_dev_config(hw,
1169 			offsetof(struct virtio_net_config, mtu),
1170 			&config.mtu, sizeof(config.mtu));
1171 
1172 		if (config.mtu < ETHER_MIN_MTU)
1173 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1174 	}
1175 
1176 	/*
1177 	 * Negotiate features: Subset of device feature bits are written back
1178 	 * guest feature bits.
1179 	 */
1180 	hw->guest_features = req_features;
1181 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1182 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1183 		hw->guest_features);
1184 
1185 	if (hw->modern) {
1186 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1187 			PMD_INIT_LOG(ERR,
1188 				"VIRTIO_F_VERSION_1 features is not enabled.");
1189 			return -1;
1190 		}
1191 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1192 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1193 			PMD_INIT_LOG(ERR,
1194 				"failed to set FEATURES_OK status!");
1195 			return -1;
1196 		}
1197 	}
1198 
1199 	hw->req_guest_features = req_features;
1200 
1201 	return 0;
1202 }
1203 
1204 /*
1205  * Process Virtio Config changed interrupt and call the callback
1206  * if link state changed.
1207  */
1208 void
1209 virtio_interrupt_handler(struct rte_intr_handle *handle,
1210 			 void *param)
1211 {
1212 	struct rte_eth_dev *dev = param;
1213 	struct virtio_hw *hw = dev->data->dev_private;
1214 	uint8_t isr;
1215 
1216 	/* Read interrupt status which clears interrupt */
1217 	isr = vtpci_isr(hw);
1218 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1219 
1220 	if (rte_intr_enable(handle) < 0)
1221 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1222 
1223 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1224 		if (virtio_dev_link_update(dev, 0) == 0)
1225 			_rte_eth_dev_callback_process(dev,
1226 						      RTE_ETH_EVENT_INTR_LSC, NULL);
1227 	}
1228 
1229 }
1230 
1231 static void
1232 rx_func_get(struct rte_eth_dev *eth_dev)
1233 {
1234 	struct virtio_hw *hw = eth_dev->data->dev_private;
1235 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1236 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1237 	else
1238 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1239 }
1240 
1241 /* Only support 1:1 queue/interrupt mapping so far.
1242  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1243  * interrupt vectors (<N+1).
1244  */
1245 static int
1246 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1247 {
1248 	uint32_t i;
1249 	struct virtio_hw *hw = dev->data->dev_private;
1250 
1251 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1252 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1253 		dev->intr_handle->intr_vec[i] = i + 1;
1254 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1255 						 VIRTIO_MSI_NO_VECTOR) {
1256 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1257 			return -EBUSY;
1258 		}
1259 	}
1260 
1261 	return 0;
1262 }
1263 
1264 static void
1265 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1266 {
1267 	uint32_t i;
1268 	struct virtio_hw *hw = dev->data->dev_private;
1269 
1270 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1271 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1272 		VTPCI_OPS(hw)->set_queue_irq(hw,
1273 					     hw->vqs[i * VTNET_CQ],
1274 					     VIRTIO_MSI_NO_VECTOR);
1275 }
1276 
1277 static int
1278 virtio_configure_intr(struct rte_eth_dev *dev)
1279 {
1280 	struct virtio_hw *hw = dev->data->dev_private;
1281 
1282 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1283 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1284 		return -ENOTSUP;
1285 	}
1286 
1287 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1288 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1289 		return -1;
1290 	}
1291 
1292 	if (!dev->intr_handle->intr_vec) {
1293 		dev->intr_handle->intr_vec =
1294 			rte_zmalloc("intr_vec",
1295 				    hw->max_queue_pairs * sizeof(int), 0);
1296 		if (!dev->intr_handle->intr_vec) {
1297 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1298 				     hw->max_queue_pairs);
1299 			return -ENOMEM;
1300 		}
1301 	}
1302 
1303 	/* Re-register callback to update max_intr */
1304 	rte_intr_callback_unregister(dev->intr_handle,
1305 				     virtio_interrupt_handler,
1306 				     dev);
1307 	rte_intr_callback_register(dev->intr_handle,
1308 				   virtio_interrupt_handler,
1309 				   dev);
1310 
1311 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1312 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1313 	 * And for legacy devices, this should be done before queue/vec binding
1314 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1315 	 * (22) will be ignored.
1316 	 */
1317 	if (rte_intr_enable(dev->intr_handle) < 0) {
1318 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1319 		return -1;
1320 	}
1321 
1322 	if (virtio_queues_bind_intr(dev) < 0) {
1323 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1324 		return -1;
1325 	}
1326 
1327 	return 0;
1328 }
1329 
1330 /* reset device and renegotiate features if needed */
1331 static int
1332 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1333 {
1334 	struct virtio_hw *hw = eth_dev->data->dev_private;
1335 	struct virtio_net_config *config;
1336 	struct virtio_net_config local_config;
1337 	struct rte_pci_device *pci_dev = NULL;
1338 	int ret;
1339 
1340 	/* Reset the device although not necessary at startup */
1341 	vtpci_reset(hw);
1342 
1343 	/* Tell the host we've noticed this device. */
1344 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1345 
1346 	/* Tell the host we've known how to drive the device. */
1347 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1348 	if (virtio_negotiate_features(hw, req_features) < 0)
1349 		return -1;
1350 
1351 	if (eth_dev->device) {
1352 		pci_dev = RTE_DEV_TO_PCI(eth_dev->device);
1353 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1354 	}
1355 
1356 	/* If host does not support both status and MSI-X then disable LSC */
1357 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->use_msix)
1358 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1359 	else
1360 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1361 
1362 	rx_func_get(eth_dev);
1363 
1364 	/* Setting up rx_header size for the device */
1365 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1366 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1367 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1368 	else
1369 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1370 
1371 	/* Copy the permanent MAC address to: virtio_hw */
1372 	virtio_get_hwaddr(hw);
1373 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1374 			&eth_dev->data->mac_addrs[0]);
1375 	PMD_INIT_LOG(DEBUG,
1376 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1377 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1378 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1379 
1380 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1381 		config = &local_config;
1382 
1383 		vtpci_read_dev_config(hw,
1384 			offsetof(struct virtio_net_config, mac),
1385 			&config->mac, sizeof(config->mac));
1386 
1387 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1388 			vtpci_read_dev_config(hw,
1389 				offsetof(struct virtio_net_config, status),
1390 				&config->status, sizeof(config->status));
1391 		} else {
1392 			PMD_INIT_LOG(DEBUG,
1393 				     "VIRTIO_NET_F_STATUS is not supported");
1394 			config->status = 0;
1395 		}
1396 
1397 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1398 			vtpci_read_dev_config(hw,
1399 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1400 				&config->max_virtqueue_pairs,
1401 				sizeof(config->max_virtqueue_pairs));
1402 		} else {
1403 			PMD_INIT_LOG(DEBUG,
1404 				     "VIRTIO_NET_F_MQ is not supported");
1405 			config->max_virtqueue_pairs = 1;
1406 		}
1407 
1408 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1409 
1410 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1411 			vtpci_read_dev_config(hw,
1412 				offsetof(struct virtio_net_config, mtu),
1413 				&config->mtu,
1414 				sizeof(config->mtu));
1415 
1416 			/*
1417 			 * MTU value has already been checked at negotiation
1418 			 * time, but check again in case it has changed since
1419 			 * then, which should not happen.
1420 			 */
1421 			if (config->mtu < ETHER_MIN_MTU) {
1422 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1423 						config->mtu);
1424 				return -1;
1425 			}
1426 
1427 			hw->max_mtu = config->mtu;
1428 			/* Set initial MTU to maximum one supported by vhost */
1429 			eth_dev->data->mtu = config->mtu;
1430 
1431 		} else {
1432 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1433 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1434 		}
1435 
1436 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1437 				config->max_virtqueue_pairs);
1438 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1439 		PMD_INIT_LOG(DEBUG,
1440 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1441 				config->mac[0], config->mac[1],
1442 				config->mac[2], config->mac[3],
1443 				config->mac[4], config->mac[5]);
1444 	} else {
1445 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1446 		hw->max_queue_pairs = 1;
1447 	}
1448 
1449 	ret = virtio_alloc_queues(eth_dev);
1450 	if (ret < 0)
1451 		return ret;
1452 
1453 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1454 		if (virtio_configure_intr(eth_dev) < 0) {
1455 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1456 			return -1;
1457 		}
1458 	}
1459 
1460 	vtpci_reinit_complete(hw);
1461 
1462 	if (pci_dev)
1463 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1464 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1465 			pci_dev->id.device_id);
1466 
1467 	return 0;
1468 }
1469 
1470 /*
1471  * Remap the PCI device again (IO port map for legacy device and
1472  * memory map for modern device), so that the secondary process
1473  * could have the PCI initiated correctly.
1474  */
1475 static int
1476 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1477 {
1478 	if (hw->modern) {
1479 		/*
1480 		 * We don't have to re-parse the PCI config space, since
1481 		 * rte_eal_pci_map_device() makes sure the mapped address
1482 		 * in secondary process would equal to the one mapped in
1483 		 * the primary process: error will be returned if that
1484 		 * requirement is not met.
1485 		 *
1486 		 * That said, we could simply reuse all cap pointers
1487 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1488 		 * primary process, which is stored in shared memory.
1489 		 */
1490 		if (rte_eal_pci_map_device(pci_dev)) {
1491 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1492 			return -1;
1493 		}
1494 	} else {
1495 		if (rte_eal_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1496 			return -1;
1497 	}
1498 
1499 	return 0;
1500 }
1501 
1502 static void
1503 virtio_set_vtpci_ops(struct virtio_hw *hw)
1504 {
1505 #ifdef RTE_VIRTIO_USER
1506 	if (hw->virtio_user_dev)
1507 		VTPCI_OPS(hw) = &virtio_user_ops;
1508 	else
1509 #endif
1510 	if (hw->modern)
1511 		VTPCI_OPS(hw) = &modern_ops;
1512 	else
1513 		VTPCI_OPS(hw) = &legacy_ops;
1514 }
1515 
1516 /*
1517  * This function is based on probe() function in virtio_pci.c
1518  * It returns 0 on success.
1519  */
1520 int
1521 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1522 {
1523 	struct virtio_hw *hw = eth_dev->data->dev_private;
1524 	uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1525 	int ret;
1526 
1527 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1528 
1529 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1530 	eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1531 
1532 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1533 		if (!hw->virtio_user_dev) {
1534 			ret = virtio_remap_pci(RTE_DEV_TO_PCI(eth_dev->device),
1535 					       hw);
1536 			if (ret)
1537 				return ret;
1538 		}
1539 
1540 		virtio_set_vtpci_ops(hw);
1541 		if (hw->use_simple_rxtx) {
1542 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1543 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1544 		} else {
1545 			rx_func_get(eth_dev);
1546 		}
1547 		return 0;
1548 	}
1549 
1550 	/* Allocate memory for storing MAC addresses */
1551 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1552 	if (eth_dev->data->mac_addrs == NULL) {
1553 		PMD_INIT_LOG(ERR,
1554 			"Failed to allocate %d bytes needed to store MAC addresses",
1555 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1556 		return -ENOMEM;
1557 	}
1558 
1559 	hw->port_id = eth_dev->data->port_id;
1560 	/* For virtio_user case the hw->virtio_user_dev is populated by
1561 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1562 	 */
1563 	if (!hw->virtio_user_dev) {
1564 		ret = vtpci_init(RTE_DEV_TO_PCI(eth_dev->device), hw,
1565 				 &dev_flags);
1566 		if (ret)
1567 			return ret;
1568 	}
1569 
1570 	eth_dev->data->dev_flags = dev_flags;
1571 
1572 	/* reset device and negotiate default features */
1573 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1574 	if (ret < 0)
1575 		return ret;
1576 
1577 	/* Setup interrupt callback  */
1578 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1579 		rte_intr_callback_register(eth_dev->intr_handle,
1580 			virtio_interrupt_handler, eth_dev);
1581 
1582 	return 0;
1583 }
1584 
1585 static int
1586 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1587 {
1588 	PMD_INIT_FUNC_TRACE();
1589 
1590 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1591 		return -EPERM;
1592 
1593 	virtio_dev_stop(eth_dev);
1594 	virtio_dev_close(eth_dev);
1595 
1596 	eth_dev->dev_ops = NULL;
1597 	eth_dev->tx_pkt_burst = NULL;
1598 	eth_dev->rx_pkt_burst = NULL;
1599 
1600 	rte_free(eth_dev->data->mac_addrs);
1601 	eth_dev->data->mac_addrs = NULL;
1602 
1603 	/* reset interrupt callback  */
1604 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1605 		rte_intr_callback_unregister(eth_dev->intr_handle,
1606 						virtio_interrupt_handler,
1607 						eth_dev);
1608 	if (eth_dev->device)
1609 		rte_eal_pci_unmap_device(RTE_DEV_TO_PCI(eth_dev->device));
1610 
1611 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1612 
1613 	return 0;
1614 }
1615 
1616 static struct eth_driver rte_virtio_pmd = {
1617 	.pci_drv = {
1618 		.driver = {
1619 			.name = "net_virtio",
1620 		},
1621 		.id_table = pci_id_virtio_map,
1622 		.drv_flags = 0,
1623 		.probe = rte_eth_dev_pci_probe,
1624 		.remove = rte_eth_dev_pci_remove,
1625 	},
1626 	.eth_dev_init = eth_virtio_dev_init,
1627 	.eth_dev_uninit = eth_virtio_dev_uninit,
1628 	.dev_private_size = sizeof(struct virtio_hw),
1629 };
1630 
1631 RTE_INIT(rte_virtio_pmd_init);
1632 static void
1633 rte_virtio_pmd_init(void)
1634 {
1635 	if (rte_eal_iopl_init() != 0) {
1636 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1637 		return;
1638 	}
1639 
1640 	rte_eal_pci_register(&rte_virtio_pmd.pci_drv);
1641 }
1642 
1643 /*
1644  * Configure virtio device
1645  * It returns 0 on success.
1646  */
1647 static int
1648 virtio_dev_configure(struct rte_eth_dev *dev)
1649 {
1650 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1651 	struct virtio_hw *hw = dev->data->dev_private;
1652 	uint64_t req_features;
1653 	int ret;
1654 
1655 	PMD_INIT_LOG(DEBUG, "configure");
1656 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1657 	if (rxmode->hw_ip_checksum)
1658 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1659 	if (rxmode->enable_lro)
1660 		req_features |=
1661 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1662 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1663 
1664 	/* if request features changed, reinit the device */
1665 	if (req_features != hw->req_guest_features) {
1666 		ret = virtio_init_device(dev, req_features);
1667 		if (ret < 0)
1668 			return ret;
1669 	}
1670 
1671 	if (rxmode->hw_ip_checksum &&
1672 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1673 		PMD_DRV_LOG(NOTICE,
1674 			"rx ip checksum not available on this host");
1675 		return -ENOTSUP;
1676 	}
1677 
1678 	if (rxmode->enable_lro &&
1679 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1680 			!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1681 		PMD_DRV_LOG(NOTICE,
1682 			"lro not available on this host");
1683 		return -ENOTSUP;
1684 	}
1685 
1686 	/* start control queue */
1687 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1688 		virtio_dev_cq_start(dev);
1689 
1690 	hw->vlan_strip = rxmode->hw_vlan_strip;
1691 
1692 	if (rxmode->hw_vlan_filter
1693 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1694 		PMD_DRV_LOG(NOTICE,
1695 			    "vlan filtering not available on this host");
1696 		return -ENOTSUP;
1697 	}
1698 
1699 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1700 		/* Enable vector (0) for Link State Intrerrupt */
1701 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1702 				VIRTIO_MSI_NO_VECTOR) {
1703 			PMD_DRV_LOG(ERR, "failed to set config vector");
1704 			return -EBUSY;
1705 		}
1706 
1707 	return 0;
1708 }
1709 
1710 
1711 static int
1712 virtio_dev_start(struct rte_eth_dev *dev)
1713 {
1714 	uint16_t nb_queues, i;
1715 	struct virtnet_rx *rxvq;
1716 	struct virtnet_tx *txvq __rte_unused;
1717 	struct virtio_hw *hw = dev->data->dev_private;
1718 
1719 	/* check if lsc interrupt feature is enabled */
1720 	if (dev->data->dev_conf.intr_conf.lsc) {
1721 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1722 			PMD_DRV_LOG(ERR, "link status not supported by host");
1723 			return -ENOTSUP;
1724 		}
1725 	}
1726 
1727 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1728 	 * in device configure, but it could be unmapped  when device is
1729 	 * stopped.
1730 	 */
1731 	if (dev->data->dev_conf.intr_conf.lsc ||
1732 	    dev->data->dev_conf.intr_conf.rxq) {
1733 		rte_intr_disable(dev->intr_handle);
1734 
1735 		if (rte_intr_enable(dev->intr_handle) < 0) {
1736 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1737 			return -EIO;
1738 		}
1739 	}
1740 
1741 	/* Initialize Link state */
1742 	virtio_dev_link_update(dev, 0);
1743 
1744 	/*Notify the backend
1745 	 *Otherwise the tap backend might already stop its queue due to fullness.
1746 	 *vhost backend will have no chance to be waked up
1747 	 */
1748 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1749 	if (hw->max_queue_pairs > 1) {
1750 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1751 			return -EINVAL;
1752 	}
1753 
1754 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1755 
1756 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1757 		rxvq = dev->data->rx_queues[i];
1758 		virtqueue_notify(rxvq->vq);
1759 	}
1760 
1761 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1762 
1763 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1764 		rxvq = dev->data->rx_queues[i];
1765 		VIRTQUEUE_DUMP(rxvq->vq);
1766 	}
1767 
1768 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1769 		txvq = dev->data->tx_queues[i];
1770 		VIRTQUEUE_DUMP(txvq->vq);
1771 	}
1772 
1773 	return 0;
1774 }
1775 
1776 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1777 {
1778 	struct rte_mbuf *buf;
1779 	int i, mbuf_num = 0;
1780 
1781 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1782 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1783 
1784 		PMD_INIT_LOG(DEBUG,
1785 			     "Before freeing rxq[%d] used and unused buf", i);
1786 		VIRTQUEUE_DUMP(rxvq->vq);
1787 
1788 		PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1789 		while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1790 			rte_pktmbuf_free(buf);
1791 			mbuf_num++;
1792 		}
1793 
1794 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1795 		PMD_INIT_LOG(DEBUG,
1796 			     "After freeing rxq[%d] used and unused buf", i);
1797 		VIRTQUEUE_DUMP(rxvq->vq);
1798 	}
1799 
1800 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1801 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1802 
1803 		PMD_INIT_LOG(DEBUG,
1804 			     "Before freeing txq[%d] used and unused bufs",
1805 			     i);
1806 		VIRTQUEUE_DUMP(txvq->vq);
1807 
1808 		mbuf_num = 0;
1809 		while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1810 			rte_pktmbuf_free(buf);
1811 			mbuf_num++;
1812 		}
1813 
1814 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1815 		PMD_INIT_LOG(DEBUG,
1816 			     "After freeing txq[%d] used and unused buf", i);
1817 		VIRTQUEUE_DUMP(txvq->vq);
1818 	}
1819 }
1820 
1821 /*
1822  * Stop device: disable interrupt and mark link down
1823  */
1824 static void
1825 virtio_dev_stop(struct rte_eth_dev *dev)
1826 {
1827 	struct rte_eth_link link;
1828 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1829 
1830 	PMD_INIT_LOG(DEBUG, "stop");
1831 
1832 	if (intr_conf->lsc || intr_conf->rxq)
1833 		rte_intr_disable(dev->intr_handle);
1834 
1835 	memset(&link, 0, sizeof(link));
1836 	virtio_dev_atomic_write_link_status(dev, &link);
1837 }
1838 
1839 static int
1840 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1841 {
1842 	struct rte_eth_link link, old;
1843 	uint16_t status;
1844 	struct virtio_hw *hw = dev->data->dev_private;
1845 	memset(&link, 0, sizeof(link));
1846 	virtio_dev_atomic_read_link_status(dev, &link);
1847 	old = link;
1848 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1849 	link.link_speed  = SPEED_10G;
1850 
1851 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1852 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
1853 		vtpci_read_dev_config(hw,
1854 				offsetof(struct virtio_net_config, status),
1855 				&status, sizeof(status));
1856 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1857 			link.link_status = ETH_LINK_DOWN;
1858 			PMD_INIT_LOG(DEBUG, "Port %d is down",
1859 				     dev->data->port_id);
1860 		} else {
1861 			link.link_status = ETH_LINK_UP;
1862 			PMD_INIT_LOG(DEBUG, "Port %d is up",
1863 				     dev->data->port_id);
1864 		}
1865 	} else {
1866 		link.link_status = ETH_LINK_UP;
1867 	}
1868 	virtio_dev_atomic_write_link_status(dev, &link);
1869 
1870 	return (old.link_status == link.link_status) ? -1 : 0;
1871 }
1872 
1873 static void
1874 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1875 {
1876 	uint64_t tso_mask, host_features;
1877 	struct virtio_hw *hw = dev->data->dev_private;
1878 
1879 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
1880 
1881 	dev_info->pci_dev = dev->device ? RTE_DEV_TO_PCI(dev->device) : NULL;
1882 	dev_info->max_rx_queues =
1883 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1884 	dev_info->max_tx_queues =
1885 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1886 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1887 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1888 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1889 	dev_info->default_txconf = (struct rte_eth_txconf) {
1890 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1891 	};
1892 
1893 	host_features = VTPCI_OPS(hw)->get_features(hw);
1894 	dev_info->rx_offload_capa = 0;
1895 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
1896 		dev_info->rx_offload_capa |=
1897 			DEV_RX_OFFLOAD_TCP_CKSUM |
1898 			DEV_RX_OFFLOAD_UDP_CKSUM;
1899 	}
1900 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1901 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1902 	if ((host_features & tso_mask) == tso_mask)
1903 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
1904 
1905 	dev_info->tx_offload_capa = 0;
1906 	if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1907 		dev_info->tx_offload_capa |=
1908 			DEV_TX_OFFLOAD_UDP_CKSUM |
1909 			DEV_TX_OFFLOAD_TCP_CKSUM;
1910 	}
1911 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1912 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
1913 	if ((hw->guest_features & tso_mask) == tso_mask)
1914 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1915 }
1916 
1917 /*
1918  * It enables testpmd to collect per queue stats.
1919  */
1920 static int
1921 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1922 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1923 __rte_unused uint8_t is_rx)
1924 {
1925 	return 0;
1926 }
1927 
1928 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
1929 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
1930 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio");
1931