xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 4e30ead5e7ca886535e2b30632b2948d2aac1681)
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 	for (i = 0; i < nr_vq; i++) {
550 		vq = hw->vqs[i];
551 		if (!vq)
552 			continue;
553 
554 		queue_type = virtio_get_queue_type(hw, i);
555 		if (queue_type == VTNET_RQ) {
556 			rte_free(vq->sw_ring);
557 			rte_memzone_free(vq->rxq.mz);
558 		} else if (queue_type == VTNET_TQ) {
559 			rte_memzone_free(vq->txq.mz);
560 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
561 		} else {
562 			rte_memzone_free(vq->cq.mz);
563 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
564 		}
565 
566 		rte_free(vq);
567 	}
568 
569 	rte_free(hw->vqs);
570 }
571 
572 static int
573 virtio_alloc_queues(struct rte_eth_dev *dev)
574 {
575 	struct virtio_hw *hw = dev->data->dev_private;
576 	uint16_t nr_vq = virtio_get_nr_vq(hw);
577 	uint16_t i;
578 	int ret;
579 
580 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
581 	if (!hw->vqs) {
582 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
583 		return -ENOMEM;
584 	}
585 
586 	for (i = 0; i < nr_vq; i++) {
587 		ret = virtio_init_queue(dev, i);
588 		if (ret < 0) {
589 			virtio_free_queues(hw);
590 			return ret;
591 		}
592 	}
593 
594 	return 0;
595 }
596 
597 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
598 
599 static void
600 virtio_dev_close(struct rte_eth_dev *dev)
601 {
602 	struct virtio_hw *hw = dev->data->dev_private;
603 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
604 
605 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
606 
607 	/* reset the NIC */
608 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
609 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
610 	if (intr_conf->rxq)
611 		virtio_queues_unbind_intr(dev);
612 
613 	if (intr_conf->lsc || intr_conf->rxq) {
614 		rte_intr_disable(dev->intr_handle);
615 		rte_intr_efd_disable(dev->intr_handle);
616 		rte_free(dev->intr_handle->intr_vec);
617 		dev->intr_handle->intr_vec = NULL;
618 	}
619 
620 	vtpci_reset(hw);
621 	virtio_dev_free_mbufs(dev);
622 	virtio_free_queues(hw);
623 }
624 
625 static void
626 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
627 {
628 	struct virtio_hw *hw = dev->data->dev_private;
629 	struct virtio_pmd_ctrl ctrl;
630 	int dlen[1];
631 	int ret;
632 
633 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
634 		PMD_INIT_LOG(INFO, "host does not support rx control");
635 		return;
636 	}
637 
638 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
639 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
640 	ctrl.data[0] = 1;
641 	dlen[0] = 1;
642 
643 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
644 	if (ret)
645 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
646 }
647 
648 static void
649 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
650 {
651 	struct virtio_hw *hw = dev->data->dev_private;
652 	struct virtio_pmd_ctrl ctrl;
653 	int dlen[1];
654 	int ret;
655 
656 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
657 		PMD_INIT_LOG(INFO, "host does not support rx control");
658 		return;
659 	}
660 
661 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
662 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
663 	ctrl.data[0] = 0;
664 	dlen[0] = 1;
665 
666 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
667 	if (ret)
668 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
669 }
670 
671 static void
672 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
673 {
674 	struct virtio_hw *hw = dev->data->dev_private;
675 	struct virtio_pmd_ctrl ctrl;
676 	int dlen[1];
677 	int ret;
678 
679 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
680 		PMD_INIT_LOG(INFO, "host does not support rx control");
681 		return;
682 	}
683 
684 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
685 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
686 	ctrl.data[0] = 1;
687 	dlen[0] = 1;
688 
689 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
690 	if (ret)
691 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
692 }
693 
694 static void
695 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
696 {
697 	struct virtio_hw *hw = dev->data->dev_private;
698 	struct virtio_pmd_ctrl ctrl;
699 	int dlen[1];
700 	int ret;
701 
702 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
703 		PMD_INIT_LOG(INFO, "host does not support rx control");
704 		return;
705 	}
706 
707 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
708 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
709 	ctrl.data[0] = 0;
710 	dlen[0] = 1;
711 
712 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
713 	if (ret)
714 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
715 }
716 
717 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
718 static int
719 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
720 {
721 	struct virtio_hw *hw = dev->data->dev_private;
722 	uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
723 				 hw->vtnet_hdr_size;
724 	uint32_t frame_size = mtu + ether_hdr_len;
725 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
726 
727 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
728 
729 	if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
730 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
731 			ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
732 		return -EINVAL;
733 	}
734 	return 0;
735 }
736 
737 static int
738 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
739 {
740 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
741 	struct virtqueue *vq = rxvq->vq;
742 
743 	virtqueue_enable_intr(vq);
744 	return 0;
745 }
746 
747 static int
748 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
749 {
750 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
751 	struct virtqueue *vq = rxvq->vq;
752 
753 	virtqueue_disable_intr(vq);
754 	return 0;
755 }
756 
757 /*
758  * dev_ops for virtio, bare necessities for basic operation
759  */
760 static const struct eth_dev_ops virtio_eth_dev_ops = {
761 	.dev_configure           = virtio_dev_configure,
762 	.dev_start               = virtio_dev_start,
763 	.dev_stop                = virtio_dev_stop,
764 	.dev_close               = virtio_dev_close,
765 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
766 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
767 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
768 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
769 	.mtu_set                 = virtio_mtu_set,
770 	.dev_infos_get           = virtio_dev_info_get,
771 	.stats_get               = virtio_dev_stats_get,
772 	.xstats_get              = virtio_dev_xstats_get,
773 	.xstats_get_names        = virtio_dev_xstats_get_names,
774 	.stats_reset             = virtio_dev_stats_reset,
775 	.xstats_reset            = virtio_dev_stats_reset,
776 	.link_update             = virtio_dev_link_update,
777 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
778 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
779 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
780 	.rx_queue_release        = virtio_dev_queue_release,
781 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
782 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
783 	.tx_queue_release        = virtio_dev_queue_release,
784 	/* collect stats per queue */
785 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
786 	.vlan_filter_set         = virtio_vlan_filter_set,
787 	.mac_addr_add            = virtio_mac_addr_add,
788 	.mac_addr_remove         = virtio_mac_addr_remove,
789 	.mac_addr_set            = virtio_mac_addr_set,
790 };
791 
792 static inline int
793 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
794 				struct rte_eth_link *link)
795 {
796 	struct rte_eth_link *dst = link;
797 	struct rte_eth_link *src = &(dev->data->dev_link);
798 
799 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
800 			*(uint64_t *)src) == 0)
801 		return -1;
802 
803 	return 0;
804 }
805 
806 /**
807  * Atomically writes the link status information into global
808  * structure rte_eth_dev.
809  *
810  * @param dev
811  *   - Pointer to the structure rte_eth_dev to read from.
812  *   - Pointer to the buffer to be saved with the link status.
813  *
814  * @return
815  *   - On success, zero.
816  *   - On failure, negative value.
817  */
818 static inline int
819 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
820 		struct rte_eth_link *link)
821 {
822 	struct rte_eth_link *dst = &(dev->data->dev_link);
823 	struct rte_eth_link *src = link;
824 
825 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
826 					*(uint64_t *)src) == 0)
827 		return -1;
828 
829 	return 0;
830 }
831 
832 static void
833 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
834 {
835 	unsigned i;
836 
837 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
838 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
839 		if (txvq == NULL)
840 			continue;
841 
842 		stats->opackets += txvq->stats.packets;
843 		stats->obytes += txvq->stats.bytes;
844 		stats->oerrors += txvq->stats.errors;
845 
846 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
847 			stats->q_opackets[i] = txvq->stats.packets;
848 			stats->q_obytes[i] = txvq->stats.bytes;
849 		}
850 	}
851 
852 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
853 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
854 		if (rxvq == NULL)
855 			continue;
856 
857 		stats->ipackets += rxvq->stats.packets;
858 		stats->ibytes += rxvq->stats.bytes;
859 		stats->ierrors += rxvq->stats.errors;
860 
861 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
862 			stats->q_ipackets[i] = rxvq->stats.packets;
863 			stats->q_ibytes[i] = rxvq->stats.bytes;
864 		}
865 	}
866 
867 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
868 }
869 
870 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
871 				       struct rte_eth_xstat_name *xstats_names,
872 				       __rte_unused unsigned limit)
873 {
874 	unsigned i;
875 	unsigned count = 0;
876 	unsigned t;
877 
878 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
879 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
880 
881 	if (xstats_names != NULL) {
882 		/* Note: limit checked in rte_eth_xstats_names() */
883 
884 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
885 			struct virtqueue *rxvq = dev->data->rx_queues[i];
886 			if (rxvq == NULL)
887 				continue;
888 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
889 				snprintf(xstats_names[count].name,
890 					sizeof(xstats_names[count].name),
891 					"rx_q%u_%s", i,
892 					rte_virtio_rxq_stat_strings[t].name);
893 				count++;
894 			}
895 		}
896 
897 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
898 			struct virtqueue *txvq = dev->data->tx_queues[i];
899 			if (txvq == NULL)
900 				continue;
901 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
902 				snprintf(xstats_names[count].name,
903 					sizeof(xstats_names[count].name),
904 					"tx_q%u_%s", i,
905 					rte_virtio_txq_stat_strings[t].name);
906 				count++;
907 			}
908 		}
909 		return count;
910 	}
911 	return nstats;
912 }
913 
914 static int
915 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
916 		      unsigned n)
917 {
918 	unsigned i;
919 	unsigned count = 0;
920 
921 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
922 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
923 
924 	if (n < nstats)
925 		return nstats;
926 
927 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
928 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
929 
930 		if (rxvq == NULL)
931 			continue;
932 
933 		unsigned t;
934 
935 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
936 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
937 				rte_virtio_rxq_stat_strings[t].offset);
938 			xstats[count].id = count;
939 			count++;
940 		}
941 	}
942 
943 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
944 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
945 
946 		if (txvq == NULL)
947 			continue;
948 
949 		unsigned t;
950 
951 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
952 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
953 				rte_virtio_txq_stat_strings[t].offset);
954 			xstats[count].id = count;
955 			count++;
956 		}
957 	}
958 
959 	return count;
960 }
961 
962 static void
963 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
964 {
965 	virtio_update_stats(dev, stats);
966 }
967 
968 static void
969 virtio_dev_stats_reset(struct rte_eth_dev *dev)
970 {
971 	unsigned int i;
972 
973 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
974 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
975 		if (txvq == NULL)
976 			continue;
977 
978 		txvq->stats.packets = 0;
979 		txvq->stats.bytes = 0;
980 		txvq->stats.errors = 0;
981 		txvq->stats.multicast = 0;
982 		txvq->stats.broadcast = 0;
983 		memset(txvq->stats.size_bins, 0,
984 		       sizeof(txvq->stats.size_bins[0]) * 8);
985 	}
986 
987 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
988 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
989 		if (rxvq == NULL)
990 			continue;
991 
992 		rxvq->stats.packets = 0;
993 		rxvq->stats.bytes = 0;
994 		rxvq->stats.errors = 0;
995 		rxvq->stats.multicast = 0;
996 		rxvq->stats.broadcast = 0;
997 		memset(rxvq->stats.size_bins, 0,
998 		       sizeof(rxvq->stats.size_bins[0]) * 8);
999 	}
1000 }
1001 
1002 static void
1003 virtio_set_hwaddr(struct virtio_hw *hw)
1004 {
1005 	vtpci_write_dev_config(hw,
1006 			offsetof(struct virtio_net_config, mac),
1007 			&hw->mac_addr, ETHER_ADDR_LEN);
1008 }
1009 
1010 static void
1011 virtio_get_hwaddr(struct virtio_hw *hw)
1012 {
1013 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1014 		vtpci_read_dev_config(hw,
1015 			offsetof(struct virtio_net_config, mac),
1016 			&hw->mac_addr, ETHER_ADDR_LEN);
1017 	} else {
1018 		eth_random_addr(&hw->mac_addr[0]);
1019 		virtio_set_hwaddr(hw);
1020 	}
1021 }
1022 
1023 static void
1024 virtio_mac_table_set(struct virtio_hw *hw,
1025 		     const struct virtio_net_ctrl_mac *uc,
1026 		     const struct virtio_net_ctrl_mac *mc)
1027 {
1028 	struct virtio_pmd_ctrl ctrl;
1029 	int err, len[2];
1030 
1031 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1032 		PMD_DRV_LOG(INFO, "host does not support mac table");
1033 		return;
1034 	}
1035 
1036 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1037 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1038 
1039 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1040 	memcpy(ctrl.data, uc, len[0]);
1041 
1042 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1043 	memcpy(ctrl.data + len[0], mc, len[1]);
1044 
1045 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1046 	if (err != 0)
1047 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1048 }
1049 
1050 static void
1051 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1052 		    uint32_t index, uint32_t vmdq __rte_unused)
1053 {
1054 	struct virtio_hw *hw = dev->data->dev_private;
1055 	const struct ether_addr *addrs = dev->data->mac_addrs;
1056 	unsigned int i;
1057 	struct virtio_net_ctrl_mac *uc, *mc;
1058 
1059 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1060 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1061 		return;
1062 	}
1063 
1064 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1065 	uc->entries = 0;
1066 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1067 	mc->entries = 0;
1068 
1069 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1070 		const struct ether_addr *addr
1071 			= (i == index) ? mac_addr : addrs + i;
1072 		struct virtio_net_ctrl_mac *tbl
1073 			= is_multicast_ether_addr(addr) ? mc : uc;
1074 
1075 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1076 	}
1077 
1078 	virtio_mac_table_set(hw, uc, mc);
1079 }
1080 
1081 static void
1082 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1083 {
1084 	struct virtio_hw *hw = dev->data->dev_private;
1085 	struct ether_addr *addrs = dev->data->mac_addrs;
1086 	struct virtio_net_ctrl_mac *uc, *mc;
1087 	unsigned int i;
1088 
1089 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1090 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1091 		return;
1092 	}
1093 
1094 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1095 	uc->entries = 0;
1096 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1097 	mc->entries = 0;
1098 
1099 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1100 		struct virtio_net_ctrl_mac *tbl;
1101 
1102 		if (i == index || is_zero_ether_addr(addrs + i))
1103 			continue;
1104 
1105 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1106 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1107 	}
1108 
1109 	virtio_mac_table_set(hw, uc, mc);
1110 }
1111 
1112 static void
1113 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1114 {
1115 	struct virtio_hw *hw = dev->data->dev_private;
1116 
1117 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1118 
1119 	/* Use atomic update if available */
1120 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1121 		struct virtio_pmd_ctrl ctrl;
1122 		int len = ETHER_ADDR_LEN;
1123 
1124 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1125 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1126 
1127 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1128 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
1129 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1130 		virtio_set_hwaddr(hw);
1131 }
1132 
1133 static int
1134 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1135 {
1136 	struct virtio_hw *hw = dev->data->dev_private;
1137 	struct virtio_pmd_ctrl ctrl;
1138 	int len;
1139 
1140 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1141 		return -ENOTSUP;
1142 
1143 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1144 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1145 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1146 	len = sizeof(vlan_id);
1147 
1148 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1149 }
1150 
1151 static int
1152 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1153 {
1154 	uint64_t host_features;
1155 
1156 	/* Prepare guest_features: feature that driver wants to support */
1157 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1158 		req_features);
1159 
1160 	/* Read device(host) feature bits */
1161 	host_features = VTPCI_OPS(hw)->get_features(hw);
1162 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1163 		host_features);
1164 
1165 	/* If supported, ensure MTU value is valid before acknowledging it. */
1166 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1167 		struct virtio_net_config config;
1168 
1169 		vtpci_read_dev_config(hw,
1170 			offsetof(struct virtio_net_config, mtu),
1171 			&config.mtu, sizeof(config.mtu));
1172 
1173 		if (config.mtu < ETHER_MIN_MTU)
1174 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1175 	}
1176 
1177 	/*
1178 	 * Negotiate features: Subset of device feature bits are written back
1179 	 * guest feature bits.
1180 	 */
1181 	hw->guest_features = req_features;
1182 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1183 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1184 		hw->guest_features);
1185 
1186 	if (hw->modern) {
1187 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1188 			PMD_INIT_LOG(ERR,
1189 				"VIRTIO_F_VERSION_1 features is not enabled.");
1190 			return -1;
1191 		}
1192 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1193 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1194 			PMD_INIT_LOG(ERR,
1195 				"failed to set FEATURES_OK status!");
1196 			return -1;
1197 		}
1198 	}
1199 
1200 	hw->req_guest_features = req_features;
1201 
1202 	return 0;
1203 }
1204 
1205 /*
1206  * Process Virtio Config changed interrupt and call the callback
1207  * if link state changed.
1208  */
1209 void
1210 virtio_interrupt_handler(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(dev->intr_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 int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1617 	struct rte_pci_device *pci_dev)
1618 {
1619 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1620 		eth_virtio_dev_init);
1621 }
1622 
1623 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1624 {
1625 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1626 }
1627 
1628 static struct rte_pci_driver rte_virtio_pmd = {
1629 	.driver = {
1630 		.name = "net_virtio",
1631 	},
1632 	.id_table = pci_id_virtio_map,
1633 	.drv_flags = 0,
1634 	.probe = eth_virtio_pci_probe,
1635 	.remove = eth_virtio_pci_remove,
1636 };
1637 
1638 RTE_INIT(rte_virtio_pmd_init);
1639 static void
1640 rte_virtio_pmd_init(void)
1641 {
1642 	if (rte_eal_iopl_init() != 0) {
1643 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1644 		return;
1645 	}
1646 
1647 	rte_eal_pci_register(&rte_virtio_pmd);
1648 }
1649 
1650 /*
1651  * Configure virtio device
1652  * It returns 0 on success.
1653  */
1654 static int
1655 virtio_dev_configure(struct rte_eth_dev *dev)
1656 {
1657 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1658 	struct virtio_hw *hw = dev->data->dev_private;
1659 	uint64_t req_features;
1660 	int ret;
1661 
1662 	PMD_INIT_LOG(DEBUG, "configure");
1663 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1664 	if (rxmode->hw_ip_checksum)
1665 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1666 	if (rxmode->enable_lro)
1667 		req_features |=
1668 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1669 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1670 
1671 	/* if request features changed, reinit the device */
1672 	if (req_features != hw->req_guest_features) {
1673 		ret = virtio_init_device(dev, req_features);
1674 		if (ret < 0)
1675 			return ret;
1676 	}
1677 
1678 	if (rxmode->hw_ip_checksum &&
1679 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1680 		PMD_DRV_LOG(NOTICE,
1681 			"rx ip checksum not available on this host");
1682 		return -ENOTSUP;
1683 	}
1684 
1685 	if (rxmode->enable_lro &&
1686 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1687 			!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1688 		PMD_DRV_LOG(NOTICE,
1689 			"lro not available on this host");
1690 		return -ENOTSUP;
1691 	}
1692 
1693 	/* start control queue */
1694 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1695 		virtio_dev_cq_start(dev);
1696 
1697 	hw->vlan_strip = rxmode->hw_vlan_strip;
1698 
1699 	if (rxmode->hw_vlan_filter
1700 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1701 		PMD_DRV_LOG(NOTICE,
1702 			    "vlan filtering not available on this host");
1703 		return -ENOTSUP;
1704 	}
1705 
1706 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1707 		/* Enable vector (0) for Link State Intrerrupt */
1708 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1709 				VIRTIO_MSI_NO_VECTOR) {
1710 			PMD_DRV_LOG(ERR, "failed to set config vector");
1711 			return -EBUSY;
1712 		}
1713 
1714 	return 0;
1715 }
1716 
1717 
1718 static int
1719 virtio_dev_start(struct rte_eth_dev *dev)
1720 {
1721 	uint16_t nb_queues, i;
1722 	struct virtnet_rx *rxvq;
1723 	struct virtnet_tx *txvq __rte_unused;
1724 	struct virtio_hw *hw = dev->data->dev_private;
1725 
1726 	/* check if lsc interrupt feature is enabled */
1727 	if (dev->data->dev_conf.intr_conf.lsc) {
1728 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1729 			PMD_DRV_LOG(ERR, "link status not supported by host");
1730 			return -ENOTSUP;
1731 		}
1732 	}
1733 
1734 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1735 	 * in device configure, but it could be unmapped  when device is
1736 	 * stopped.
1737 	 */
1738 	if (dev->data->dev_conf.intr_conf.lsc ||
1739 	    dev->data->dev_conf.intr_conf.rxq) {
1740 		rte_intr_disable(dev->intr_handle);
1741 
1742 		if (rte_intr_enable(dev->intr_handle) < 0) {
1743 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1744 			return -EIO;
1745 		}
1746 	}
1747 
1748 	/* Initialize Link state */
1749 	virtio_dev_link_update(dev, 0);
1750 
1751 	/*Notify the backend
1752 	 *Otherwise the tap backend might already stop its queue due to fullness.
1753 	 *vhost backend will have no chance to be waked up
1754 	 */
1755 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1756 	if (hw->max_queue_pairs > 1) {
1757 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1758 			return -EINVAL;
1759 	}
1760 
1761 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1762 
1763 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1764 		rxvq = dev->data->rx_queues[i];
1765 		virtqueue_notify(rxvq->vq);
1766 	}
1767 
1768 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1769 
1770 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1771 		rxvq = dev->data->rx_queues[i];
1772 		VIRTQUEUE_DUMP(rxvq->vq);
1773 	}
1774 
1775 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1776 		txvq = dev->data->tx_queues[i];
1777 		VIRTQUEUE_DUMP(txvq->vq);
1778 	}
1779 	hw->started = 1;
1780 
1781 	return 0;
1782 }
1783 
1784 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1785 {
1786 	struct rte_mbuf *buf;
1787 	int i, mbuf_num = 0;
1788 
1789 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1790 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1791 
1792 		PMD_INIT_LOG(DEBUG,
1793 			     "Before freeing rxq[%d] used and unused buf", i);
1794 		VIRTQUEUE_DUMP(rxvq->vq);
1795 
1796 		PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1797 		while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1798 			rte_pktmbuf_free(buf);
1799 			mbuf_num++;
1800 		}
1801 
1802 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1803 		PMD_INIT_LOG(DEBUG,
1804 			     "After freeing rxq[%d] used and unused buf", i);
1805 		VIRTQUEUE_DUMP(rxvq->vq);
1806 	}
1807 
1808 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1809 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1810 
1811 		PMD_INIT_LOG(DEBUG,
1812 			     "Before freeing txq[%d] used and unused bufs",
1813 			     i);
1814 		VIRTQUEUE_DUMP(txvq->vq);
1815 
1816 		mbuf_num = 0;
1817 		while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1818 			rte_pktmbuf_free(buf);
1819 			mbuf_num++;
1820 		}
1821 
1822 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1823 		PMD_INIT_LOG(DEBUG,
1824 			     "After freeing txq[%d] used and unused buf", i);
1825 		VIRTQUEUE_DUMP(txvq->vq);
1826 	}
1827 }
1828 
1829 /*
1830  * Stop device: disable interrupt and mark link down
1831  */
1832 static void
1833 virtio_dev_stop(struct rte_eth_dev *dev)
1834 {
1835 	struct virtio_hw *hw = dev->data->dev_private;
1836 	struct rte_eth_link link;
1837 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1838 
1839 	PMD_INIT_LOG(DEBUG, "stop");
1840 
1841 	if (intr_conf->lsc || intr_conf->rxq)
1842 		rte_intr_disable(dev->intr_handle);
1843 
1844 	hw->started = 0;
1845 	memset(&link, 0, sizeof(link));
1846 	virtio_dev_atomic_write_link_status(dev, &link);
1847 }
1848 
1849 static int
1850 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1851 {
1852 	struct rte_eth_link link, old;
1853 	uint16_t status;
1854 	struct virtio_hw *hw = dev->data->dev_private;
1855 	memset(&link, 0, sizeof(link));
1856 	virtio_dev_atomic_read_link_status(dev, &link);
1857 	old = link;
1858 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1859 	link.link_speed  = SPEED_10G;
1860 
1861 	if (hw->started == 0) {
1862 		link.link_status = ETH_LINK_DOWN;
1863 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1864 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
1865 		vtpci_read_dev_config(hw,
1866 				offsetof(struct virtio_net_config, status),
1867 				&status, sizeof(status));
1868 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1869 			link.link_status = ETH_LINK_DOWN;
1870 			PMD_INIT_LOG(DEBUG, "Port %d is down",
1871 				     dev->data->port_id);
1872 		} else {
1873 			link.link_status = ETH_LINK_UP;
1874 			PMD_INIT_LOG(DEBUG, "Port %d is up",
1875 				     dev->data->port_id);
1876 		}
1877 	} else {
1878 		link.link_status = ETH_LINK_UP;
1879 	}
1880 	virtio_dev_atomic_write_link_status(dev, &link);
1881 
1882 	return (old.link_status == link.link_status) ? -1 : 0;
1883 }
1884 
1885 static void
1886 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1887 {
1888 	uint64_t tso_mask, host_features;
1889 	struct virtio_hw *hw = dev->data->dev_private;
1890 
1891 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
1892 
1893 	dev_info->pci_dev = dev->device ? RTE_DEV_TO_PCI(dev->device) : NULL;
1894 	dev_info->max_rx_queues =
1895 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1896 	dev_info->max_tx_queues =
1897 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1898 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1899 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1900 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1901 	dev_info->default_txconf = (struct rte_eth_txconf) {
1902 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1903 	};
1904 
1905 	host_features = VTPCI_OPS(hw)->get_features(hw);
1906 	dev_info->rx_offload_capa = 0;
1907 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
1908 		dev_info->rx_offload_capa |=
1909 			DEV_RX_OFFLOAD_TCP_CKSUM |
1910 			DEV_RX_OFFLOAD_UDP_CKSUM;
1911 	}
1912 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1913 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1914 	if ((host_features & tso_mask) == tso_mask)
1915 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
1916 
1917 	dev_info->tx_offload_capa = 0;
1918 	if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1919 		dev_info->tx_offload_capa |=
1920 			DEV_TX_OFFLOAD_UDP_CKSUM |
1921 			DEV_TX_OFFLOAD_TCP_CKSUM;
1922 	}
1923 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1924 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
1925 	if ((hw->guest_features & tso_mask) == tso_mask)
1926 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1927 }
1928 
1929 /*
1930  * It enables testpmd to collect per queue stats.
1931  */
1932 static int
1933 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1934 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1935 __rte_unused uint8_t is_rx)
1936 {
1937 	return 0;
1938 }
1939 
1940 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
1941 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
1942 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio");
1943