xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 6b72aad61f09b870427f4ebcedae525a985bb37a)
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 
725 	if (mtu < ETHER_MIN_MTU || frame_size > VIRTIO_MAX_RX_PKTLEN) {
726 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
727 			ETHER_MIN_MTU, VIRTIO_MAX_RX_PKTLEN - ether_hdr_len);
728 		return -EINVAL;
729 	}
730 	return 0;
731 }
732 
733 static int
734 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
735 {
736 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
737 	struct virtqueue *vq = rxvq->vq;
738 
739 	virtqueue_enable_intr(vq);
740 	return 0;
741 }
742 
743 static int
744 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
745 {
746 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
747 	struct virtqueue *vq = rxvq->vq;
748 
749 	virtqueue_disable_intr(vq);
750 	return 0;
751 }
752 
753 /*
754  * dev_ops for virtio, bare necessities for basic operation
755  */
756 static const struct eth_dev_ops virtio_eth_dev_ops = {
757 	.dev_configure           = virtio_dev_configure,
758 	.dev_start               = virtio_dev_start,
759 	.dev_stop                = virtio_dev_stop,
760 	.dev_close               = virtio_dev_close,
761 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
762 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
763 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
764 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
765 	.mtu_set                 = virtio_mtu_set,
766 	.dev_infos_get           = virtio_dev_info_get,
767 	.stats_get               = virtio_dev_stats_get,
768 	.xstats_get              = virtio_dev_xstats_get,
769 	.xstats_get_names        = virtio_dev_xstats_get_names,
770 	.stats_reset             = virtio_dev_stats_reset,
771 	.xstats_reset            = virtio_dev_stats_reset,
772 	.link_update             = virtio_dev_link_update,
773 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
774 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
775 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
776 	.rx_queue_release        = virtio_dev_queue_release,
777 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
778 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
779 	.tx_queue_release        = virtio_dev_queue_release,
780 	/* collect stats per queue */
781 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
782 	.vlan_filter_set         = virtio_vlan_filter_set,
783 	.mac_addr_add            = virtio_mac_addr_add,
784 	.mac_addr_remove         = virtio_mac_addr_remove,
785 	.mac_addr_set            = virtio_mac_addr_set,
786 };
787 
788 static inline int
789 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
790 				struct rte_eth_link *link)
791 {
792 	struct rte_eth_link *dst = link;
793 	struct rte_eth_link *src = &(dev->data->dev_link);
794 
795 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
796 			*(uint64_t *)src) == 0)
797 		return -1;
798 
799 	return 0;
800 }
801 
802 /**
803  * Atomically writes the link status information into global
804  * structure rte_eth_dev.
805  *
806  * @param dev
807  *   - Pointer to the structure rte_eth_dev to read from.
808  *   - Pointer to the buffer to be saved with the link status.
809  *
810  * @return
811  *   - On success, zero.
812  *   - On failure, negative value.
813  */
814 static inline int
815 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
816 		struct rte_eth_link *link)
817 {
818 	struct rte_eth_link *dst = &(dev->data->dev_link);
819 	struct rte_eth_link *src = link;
820 
821 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
822 					*(uint64_t *)src) == 0)
823 		return -1;
824 
825 	return 0;
826 }
827 
828 static void
829 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
830 {
831 	unsigned i;
832 
833 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
834 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
835 		if (txvq == NULL)
836 			continue;
837 
838 		stats->opackets += txvq->stats.packets;
839 		stats->obytes += txvq->stats.bytes;
840 		stats->oerrors += txvq->stats.errors;
841 
842 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
843 			stats->q_opackets[i] = txvq->stats.packets;
844 			stats->q_obytes[i] = txvq->stats.bytes;
845 		}
846 	}
847 
848 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
849 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
850 		if (rxvq == NULL)
851 			continue;
852 
853 		stats->ipackets += rxvq->stats.packets;
854 		stats->ibytes += rxvq->stats.bytes;
855 		stats->ierrors += rxvq->stats.errors;
856 
857 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
858 			stats->q_ipackets[i] = rxvq->stats.packets;
859 			stats->q_ibytes[i] = rxvq->stats.bytes;
860 		}
861 	}
862 
863 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
864 }
865 
866 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
867 				       struct rte_eth_xstat_name *xstats_names,
868 				       __rte_unused unsigned limit)
869 {
870 	unsigned i;
871 	unsigned count = 0;
872 	unsigned t;
873 
874 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
875 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
876 
877 	if (xstats_names != NULL) {
878 		/* Note: limit checked in rte_eth_xstats_names() */
879 
880 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
881 			struct virtqueue *rxvq = dev->data->rx_queues[i];
882 			if (rxvq == NULL)
883 				continue;
884 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
885 				snprintf(xstats_names[count].name,
886 					sizeof(xstats_names[count].name),
887 					"rx_q%u_%s", i,
888 					rte_virtio_rxq_stat_strings[t].name);
889 				count++;
890 			}
891 		}
892 
893 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
894 			struct virtqueue *txvq = dev->data->tx_queues[i];
895 			if (txvq == NULL)
896 				continue;
897 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
898 				snprintf(xstats_names[count].name,
899 					sizeof(xstats_names[count].name),
900 					"tx_q%u_%s", i,
901 					rte_virtio_txq_stat_strings[t].name);
902 				count++;
903 			}
904 		}
905 		return count;
906 	}
907 	return nstats;
908 }
909 
910 static int
911 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
912 		      unsigned n)
913 {
914 	unsigned i;
915 	unsigned count = 0;
916 
917 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
918 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
919 
920 	if (n < nstats)
921 		return nstats;
922 
923 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
924 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
925 
926 		if (rxvq == NULL)
927 			continue;
928 
929 		unsigned t;
930 
931 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
932 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
933 				rte_virtio_rxq_stat_strings[t].offset);
934 			xstats[count].id = count;
935 			count++;
936 		}
937 	}
938 
939 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
940 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
941 
942 		if (txvq == NULL)
943 			continue;
944 
945 		unsigned t;
946 
947 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
948 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
949 				rte_virtio_txq_stat_strings[t].offset);
950 			xstats[count].id = count;
951 			count++;
952 		}
953 	}
954 
955 	return count;
956 }
957 
958 static void
959 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
960 {
961 	virtio_update_stats(dev, stats);
962 }
963 
964 static void
965 virtio_dev_stats_reset(struct rte_eth_dev *dev)
966 {
967 	unsigned int i;
968 
969 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
970 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
971 		if (txvq == NULL)
972 			continue;
973 
974 		txvq->stats.packets = 0;
975 		txvq->stats.bytes = 0;
976 		txvq->stats.errors = 0;
977 		txvq->stats.multicast = 0;
978 		txvq->stats.broadcast = 0;
979 		memset(txvq->stats.size_bins, 0,
980 		       sizeof(txvq->stats.size_bins[0]) * 8);
981 	}
982 
983 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
984 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
985 		if (rxvq == NULL)
986 			continue;
987 
988 		rxvq->stats.packets = 0;
989 		rxvq->stats.bytes = 0;
990 		rxvq->stats.errors = 0;
991 		rxvq->stats.multicast = 0;
992 		rxvq->stats.broadcast = 0;
993 		memset(rxvq->stats.size_bins, 0,
994 		       sizeof(rxvq->stats.size_bins[0]) * 8);
995 	}
996 }
997 
998 static void
999 virtio_set_hwaddr(struct virtio_hw *hw)
1000 {
1001 	vtpci_write_dev_config(hw,
1002 			offsetof(struct virtio_net_config, mac),
1003 			&hw->mac_addr, ETHER_ADDR_LEN);
1004 }
1005 
1006 static void
1007 virtio_get_hwaddr(struct virtio_hw *hw)
1008 {
1009 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1010 		vtpci_read_dev_config(hw,
1011 			offsetof(struct virtio_net_config, mac),
1012 			&hw->mac_addr, ETHER_ADDR_LEN);
1013 	} else {
1014 		eth_random_addr(&hw->mac_addr[0]);
1015 		virtio_set_hwaddr(hw);
1016 	}
1017 }
1018 
1019 static void
1020 virtio_mac_table_set(struct virtio_hw *hw,
1021 		     const struct virtio_net_ctrl_mac *uc,
1022 		     const struct virtio_net_ctrl_mac *mc)
1023 {
1024 	struct virtio_pmd_ctrl ctrl;
1025 	int err, len[2];
1026 
1027 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1028 		PMD_DRV_LOG(INFO, "host does not support mac table");
1029 		return;
1030 	}
1031 
1032 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1033 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1034 
1035 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1036 	memcpy(ctrl.data, uc, len[0]);
1037 
1038 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1039 	memcpy(ctrl.data + len[0], mc, len[1]);
1040 
1041 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1042 	if (err != 0)
1043 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1044 }
1045 
1046 static void
1047 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1048 		    uint32_t index, uint32_t vmdq __rte_unused)
1049 {
1050 	struct virtio_hw *hw = dev->data->dev_private;
1051 	const struct ether_addr *addrs = dev->data->mac_addrs;
1052 	unsigned int i;
1053 	struct virtio_net_ctrl_mac *uc, *mc;
1054 
1055 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1056 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1057 		return;
1058 	}
1059 
1060 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1061 	uc->entries = 0;
1062 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1063 	mc->entries = 0;
1064 
1065 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1066 		const struct ether_addr *addr
1067 			= (i == index) ? mac_addr : addrs + i;
1068 		struct virtio_net_ctrl_mac *tbl
1069 			= is_multicast_ether_addr(addr) ? mc : uc;
1070 
1071 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1072 	}
1073 
1074 	virtio_mac_table_set(hw, uc, mc);
1075 }
1076 
1077 static void
1078 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1079 {
1080 	struct virtio_hw *hw = dev->data->dev_private;
1081 	struct ether_addr *addrs = dev->data->mac_addrs;
1082 	struct virtio_net_ctrl_mac *uc, *mc;
1083 	unsigned int i;
1084 
1085 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1086 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1087 		return;
1088 	}
1089 
1090 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1091 	uc->entries = 0;
1092 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1093 	mc->entries = 0;
1094 
1095 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1096 		struct virtio_net_ctrl_mac *tbl;
1097 
1098 		if (i == index || is_zero_ether_addr(addrs + i))
1099 			continue;
1100 
1101 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1102 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1103 	}
1104 
1105 	virtio_mac_table_set(hw, uc, mc);
1106 }
1107 
1108 static void
1109 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1110 {
1111 	struct virtio_hw *hw = dev->data->dev_private;
1112 
1113 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1114 
1115 	/* Use atomic update if available */
1116 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1117 		struct virtio_pmd_ctrl ctrl;
1118 		int len = ETHER_ADDR_LEN;
1119 
1120 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1121 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1122 
1123 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1124 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
1125 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1126 		virtio_set_hwaddr(hw);
1127 }
1128 
1129 static int
1130 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1131 {
1132 	struct virtio_hw *hw = dev->data->dev_private;
1133 	struct virtio_pmd_ctrl ctrl;
1134 	int len;
1135 
1136 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1137 		return -ENOTSUP;
1138 
1139 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1140 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1141 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1142 	len = sizeof(vlan_id);
1143 
1144 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1145 }
1146 
1147 static int
1148 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1149 {
1150 	uint64_t host_features;
1151 
1152 	/* Prepare guest_features: feature that driver wants to support */
1153 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1154 		req_features);
1155 
1156 	/* Read device(host) feature bits */
1157 	host_features = VTPCI_OPS(hw)->get_features(hw);
1158 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1159 		host_features);
1160 
1161 	/*
1162 	 * Negotiate features: Subset of device feature bits are written back
1163 	 * guest feature bits.
1164 	 */
1165 	hw->guest_features = req_features;
1166 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1167 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1168 		hw->guest_features);
1169 
1170 	if (hw->modern) {
1171 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1172 			PMD_INIT_LOG(ERR,
1173 				"VIRTIO_F_VERSION_1 features is not enabled.");
1174 			return -1;
1175 		}
1176 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1177 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1178 			PMD_INIT_LOG(ERR,
1179 				"failed to set FEATURES_OK status!");
1180 			return -1;
1181 		}
1182 	}
1183 
1184 	hw->req_guest_features = req_features;
1185 
1186 	return 0;
1187 }
1188 
1189 /*
1190  * Process Virtio Config changed interrupt and call the callback
1191  * if link state changed.
1192  */
1193 static void
1194 virtio_interrupt_handler(struct rte_intr_handle *handle,
1195 			 void *param)
1196 {
1197 	struct rte_eth_dev *dev = param;
1198 	struct virtio_hw *hw = dev->data->dev_private;
1199 	uint8_t isr;
1200 
1201 	/* Read interrupt status which clears interrupt */
1202 	isr = vtpci_isr(hw);
1203 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1204 
1205 	if (rte_intr_enable(handle) < 0)
1206 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1207 
1208 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1209 		if (virtio_dev_link_update(dev, 0) == 0)
1210 			_rte_eth_dev_callback_process(dev,
1211 						      RTE_ETH_EVENT_INTR_LSC, NULL);
1212 	}
1213 
1214 }
1215 
1216 static void
1217 rx_func_get(struct rte_eth_dev *eth_dev)
1218 {
1219 	struct virtio_hw *hw = eth_dev->data->dev_private;
1220 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1221 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1222 	else
1223 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1224 }
1225 
1226 /* Only support 1:1 queue/interrupt mapping so far.
1227  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1228  * interrupt vectors (<N+1).
1229  */
1230 static int
1231 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1232 {
1233 	uint32_t i;
1234 	struct virtio_hw *hw = dev->data->dev_private;
1235 
1236 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1237 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1238 		dev->intr_handle->intr_vec[i] = i + 1;
1239 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1240 						 VIRTIO_MSI_NO_VECTOR) {
1241 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1242 			return -EBUSY;
1243 		}
1244 	}
1245 
1246 	return 0;
1247 }
1248 
1249 static void
1250 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1251 {
1252 	uint32_t i;
1253 	struct virtio_hw *hw = dev->data->dev_private;
1254 
1255 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1256 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1257 		VTPCI_OPS(hw)->set_queue_irq(hw,
1258 					     hw->vqs[i * VTNET_CQ],
1259 					     VIRTIO_MSI_NO_VECTOR);
1260 }
1261 
1262 static int
1263 virtio_configure_intr(struct rte_eth_dev *dev)
1264 {
1265 	struct virtio_hw *hw = dev->data->dev_private;
1266 
1267 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1268 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1269 		return -ENOTSUP;
1270 	}
1271 
1272 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1273 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1274 		return -1;
1275 	}
1276 
1277 	if (!dev->intr_handle->intr_vec) {
1278 		dev->intr_handle->intr_vec =
1279 			rte_zmalloc("intr_vec",
1280 				    hw->max_queue_pairs * sizeof(int), 0);
1281 		if (!dev->intr_handle->intr_vec) {
1282 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1283 				     hw->max_queue_pairs);
1284 			return -ENOMEM;
1285 		}
1286 	}
1287 
1288 	/* Re-register callback to update max_intr */
1289 	rte_intr_callback_unregister(dev->intr_handle,
1290 				     virtio_interrupt_handler,
1291 				     dev);
1292 	rte_intr_callback_register(dev->intr_handle,
1293 				   virtio_interrupt_handler,
1294 				   dev);
1295 
1296 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1297 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1298 	 * And for legacy devices, this should be done before queue/vec binding
1299 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1300 	 * (22) will be ignored.
1301 	 */
1302 	if (rte_intr_enable(dev->intr_handle) < 0) {
1303 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1304 		return -1;
1305 	}
1306 
1307 	if (virtio_queues_bind_intr(dev) < 0) {
1308 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1309 		return -1;
1310 	}
1311 
1312 	return 0;
1313 }
1314 
1315 /* reset device and renegotiate features if needed */
1316 static int
1317 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1318 {
1319 	struct virtio_hw *hw = eth_dev->data->dev_private;
1320 	struct virtio_net_config *config;
1321 	struct virtio_net_config local_config;
1322 	struct rte_pci_device *pci_dev = NULL;
1323 	int ret;
1324 
1325 	/* Reset the device although not necessary at startup */
1326 	vtpci_reset(hw);
1327 
1328 	/* Tell the host we've noticed this device. */
1329 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1330 
1331 	/* Tell the host we've known how to drive the device. */
1332 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1333 	if (virtio_negotiate_features(hw, req_features) < 0)
1334 		return -1;
1335 
1336 	if (eth_dev->device) {
1337 		pci_dev = RTE_DEV_TO_PCI(eth_dev->device);
1338 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1339 	}
1340 
1341 	/* If host does not support status then disable LSC */
1342 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1343 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1344 	else
1345 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1346 
1347 	rx_func_get(eth_dev);
1348 
1349 	/* Setting up rx_header size for the device */
1350 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1351 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1352 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1353 	else
1354 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1355 
1356 	/* Copy the permanent MAC address to: virtio_hw */
1357 	virtio_get_hwaddr(hw);
1358 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1359 			&eth_dev->data->mac_addrs[0]);
1360 	PMD_INIT_LOG(DEBUG,
1361 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1362 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1363 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1364 
1365 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1366 		config = &local_config;
1367 
1368 		vtpci_read_dev_config(hw,
1369 			offsetof(struct virtio_net_config, mac),
1370 			&config->mac, sizeof(config->mac));
1371 
1372 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1373 			vtpci_read_dev_config(hw,
1374 				offsetof(struct virtio_net_config, status),
1375 				&config->status, sizeof(config->status));
1376 		} else {
1377 			PMD_INIT_LOG(DEBUG,
1378 				     "VIRTIO_NET_F_STATUS is not supported");
1379 			config->status = 0;
1380 		}
1381 
1382 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1383 			vtpci_read_dev_config(hw,
1384 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1385 				&config->max_virtqueue_pairs,
1386 				sizeof(config->max_virtqueue_pairs));
1387 		} else {
1388 			PMD_INIT_LOG(DEBUG,
1389 				     "VIRTIO_NET_F_MQ is not supported");
1390 			config->max_virtqueue_pairs = 1;
1391 		}
1392 
1393 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1394 
1395 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1396 				config->max_virtqueue_pairs);
1397 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1398 		PMD_INIT_LOG(DEBUG,
1399 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1400 				config->mac[0], config->mac[1],
1401 				config->mac[2], config->mac[3],
1402 				config->mac[4], config->mac[5]);
1403 	} else {
1404 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1405 		hw->max_queue_pairs = 1;
1406 	}
1407 
1408 	ret = virtio_alloc_queues(eth_dev);
1409 	if (ret < 0)
1410 		return ret;
1411 
1412 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1413 		if (virtio_configure_intr(eth_dev) < 0) {
1414 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1415 			return -1;
1416 		}
1417 	}
1418 
1419 	vtpci_reinit_complete(hw);
1420 
1421 	if (pci_dev)
1422 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1423 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1424 			pci_dev->id.device_id);
1425 
1426 	return 0;
1427 }
1428 
1429 /*
1430  * Remap the PCI device again (IO port map for legacy device and
1431  * memory map for modern device), so that the secondary process
1432  * could have the PCI initiated correctly.
1433  */
1434 static int
1435 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1436 {
1437 	if (hw->modern) {
1438 		/*
1439 		 * We don't have to re-parse the PCI config space, since
1440 		 * rte_eal_pci_map_device() makes sure the mapped address
1441 		 * in secondary process would equal to the one mapped in
1442 		 * the primary process: error will be returned if that
1443 		 * requirement is not met.
1444 		 *
1445 		 * That said, we could simply reuse all cap pointers
1446 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1447 		 * primary process, which is stored in shared memory.
1448 		 */
1449 		if (rte_eal_pci_map_device(pci_dev)) {
1450 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1451 			return -1;
1452 		}
1453 	} else {
1454 		if (rte_eal_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1455 			return -1;
1456 	}
1457 
1458 	return 0;
1459 }
1460 
1461 static void
1462 virtio_set_vtpci_ops(struct virtio_hw *hw)
1463 {
1464 #ifdef RTE_VIRTIO_USER
1465 	if (hw->virtio_user_dev)
1466 		VTPCI_OPS(hw) = &virtio_user_ops;
1467 	else
1468 #endif
1469 	if (hw->modern)
1470 		VTPCI_OPS(hw) = &modern_ops;
1471 	else
1472 		VTPCI_OPS(hw) = &legacy_ops;
1473 }
1474 
1475 /*
1476  * This function is based on probe() function in virtio_pci.c
1477  * It returns 0 on success.
1478  */
1479 int
1480 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1481 {
1482 	struct virtio_hw *hw = eth_dev->data->dev_private;
1483 	uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1484 	int ret;
1485 
1486 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1487 
1488 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1489 	eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1490 
1491 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1492 		if (!hw->virtio_user_dev) {
1493 			ret = virtio_remap_pci(RTE_DEV_TO_PCI(eth_dev->device),
1494 					       hw);
1495 			if (ret)
1496 				return ret;
1497 		}
1498 
1499 		virtio_set_vtpci_ops(hw);
1500 		if (hw->use_simple_rxtx) {
1501 			eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1502 			eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1503 		} else {
1504 			rx_func_get(eth_dev);
1505 		}
1506 		return 0;
1507 	}
1508 
1509 	/* Allocate memory for storing MAC addresses */
1510 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1511 	if (eth_dev->data->mac_addrs == NULL) {
1512 		PMD_INIT_LOG(ERR,
1513 			"Failed to allocate %d bytes needed to store MAC addresses",
1514 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1515 		return -ENOMEM;
1516 	}
1517 
1518 	hw->port_id = eth_dev->data->port_id;
1519 	/* For virtio_user case the hw->virtio_user_dev is populated by
1520 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1521 	 */
1522 	if (!hw->virtio_user_dev) {
1523 		ret = vtpci_init(RTE_DEV_TO_PCI(eth_dev->device), hw,
1524 				 &dev_flags);
1525 		if (ret)
1526 			return ret;
1527 	}
1528 
1529 	eth_dev->data->dev_flags = dev_flags;
1530 
1531 	/* reset device and negotiate default features */
1532 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1533 	if (ret < 0)
1534 		return ret;
1535 
1536 	/* Setup interrupt callback  */
1537 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1538 		rte_intr_callback_register(eth_dev->intr_handle,
1539 			virtio_interrupt_handler, eth_dev);
1540 
1541 	return 0;
1542 }
1543 
1544 static int
1545 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1546 {
1547 	PMD_INIT_FUNC_TRACE();
1548 
1549 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1550 		return -EPERM;
1551 
1552 	virtio_dev_stop(eth_dev);
1553 	virtio_dev_close(eth_dev);
1554 
1555 	eth_dev->dev_ops = NULL;
1556 	eth_dev->tx_pkt_burst = NULL;
1557 	eth_dev->rx_pkt_burst = NULL;
1558 
1559 	rte_free(eth_dev->data->mac_addrs);
1560 	eth_dev->data->mac_addrs = NULL;
1561 
1562 	/* reset interrupt callback  */
1563 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1564 		rte_intr_callback_unregister(eth_dev->intr_handle,
1565 						virtio_interrupt_handler,
1566 						eth_dev);
1567 	if (eth_dev->device)
1568 		rte_eal_pci_unmap_device(RTE_DEV_TO_PCI(eth_dev->device));
1569 
1570 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1571 
1572 	return 0;
1573 }
1574 
1575 static struct eth_driver rte_virtio_pmd = {
1576 	.pci_drv = {
1577 		.driver = {
1578 			.name = "net_virtio",
1579 		},
1580 		.id_table = pci_id_virtio_map,
1581 		.drv_flags = 0,
1582 		.probe = rte_eth_dev_pci_probe,
1583 		.remove = rte_eth_dev_pci_remove,
1584 	},
1585 	.eth_dev_init = eth_virtio_dev_init,
1586 	.eth_dev_uninit = eth_virtio_dev_uninit,
1587 	.dev_private_size = sizeof(struct virtio_hw),
1588 };
1589 
1590 RTE_INIT(rte_virtio_pmd_init);
1591 static void
1592 rte_virtio_pmd_init(void)
1593 {
1594 	if (rte_eal_iopl_init() != 0) {
1595 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1596 		return;
1597 	}
1598 
1599 	rte_eal_pci_register(&rte_virtio_pmd.pci_drv);
1600 }
1601 
1602 /*
1603  * Configure virtio device
1604  * It returns 0 on success.
1605  */
1606 static int
1607 virtio_dev_configure(struct rte_eth_dev *dev)
1608 {
1609 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1610 	struct virtio_hw *hw = dev->data->dev_private;
1611 	uint64_t req_features;
1612 	int ret;
1613 
1614 	PMD_INIT_LOG(DEBUG, "configure");
1615 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1616 	if (rxmode->hw_ip_checksum)
1617 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1618 	if (rxmode->enable_lro)
1619 		req_features |=
1620 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1621 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1622 
1623 	/* if request features changed, reinit the device */
1624 	if (req_features != hw->req_guest_features) {
1625 		ret = virtio_init_device(dev, req_features);
1626 		if (ret < 0)
1627 			return ret;
1628 	}
1629 
1630 	if (rxmode->hw_ip_checksum &&
1631 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1632 		PMD_DRV_LOG(NOTICE,
1633 			"rx ip checksum not available on this host");
1634 		return -ENOTSUP;
1635 	}
1636 
1637 	if (rxmode->enable_lro &&
1638 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1639 			!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1640 		PMD_DRV_LOG(NOTICE,
1641 			"lro not available on this host");
1642 		return -ENOTSUP;
1643 	}
1644 
1645 	/* start control queue */
1646 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1647 		virtio_dev_cq_start(dev);
1648 
1649 	hw->vlan_strip = rxmode->hw_vlan_strip;
1650 
1651 	if (rxmode->hw_vlan_filter
1652 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1653 		PMD_DRV_LOG(NOTICE,
1654 			    "vlan filtering not available on this host");
1655 		return -ENOTSUP;
1656 	}
1657 
1658 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1659 		/* Enable vector (0) for Link State Intrerrupt */
1660 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1661 				VIRTIO_MSI_NO_VECTOR) {
1662 			PMD_DRV_LOG(ERR, "failed to set config vector");
1663 			return -EBUSY;
1664 		}
1665 
1666 	return 0;
1667 }
1668 
1669 
1670 static int
1671 virtio_dev_start(struct rte_eth_dev *dev)
1672 {
1673 	uint16_t nb_queues, i;
1674 	struct virtnet_rx *rxvq;
1675 	struct virtnet_tx *txvq __rte_unused;
1676 	struct virtio_hw *hw = dev->data->dev_private;
1677 
1678 	/* check if lsc interrupt feature is enabled */
1679 	if (dev->data->dev_conf.intr_conf.lsc) {
1680 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1681 			PMD_DRV_LOG(ERR, "link status not supported by host");
1682 			return -ENOTSUP;
1683 		}
1684 	}
1685 
1686 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1687 	 * in device configure, but it could be unmapped  when device is
1688 	 * stopped.
1689 	 */
1690 	if (dev->data->dev_conf.intr_conf.lsc ||
1691 	    dev->data->dev_conf.intr_conf.rxq) {
1692 		rte_intr_disable(dev->intr_handle);
1693 
1694 		if (rte_intr_enable(dev->intr_handle) < 0) {
1695 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1696 			return -EIO;
1697 		}
1698 	}
1699 
1700 	/* Initialize Link state */
1701 	virtio_dev_link_update(dev, 0);
1702 
1703 	/*Notify the backend
1704 	 *Otherwise the tap backend might already stop its queue due to fullness.
1705 	 *vhost backend will have no chance to be waked up
1706 	 */
1707 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1708 	if (hw->max_queue_pairs > 1) {
1709 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1710 			return -EINVAL;
1711 	}
1712 
1713 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1714 
1715 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1716 		rxvq = dev->data->rx_queues[i];
1717 		virtqueue_notify(rxvq->vq);
1718 	}
1719 
1720 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1721 
1722 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1723 		rxvq = dev->data->rx_queues[i];
1724 		VIRTQUEUE_DUMP(rxvq->vq);
1725 	}
1726 
1727 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1728 		txvq = dev->data->tx_queues[i];
1729 		VIRTQUEUE_DUMP(txvq->vq);
1730 	}
1731 
1732 	return 0;
1733 }
1734 
1735 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1736 {
1737 	struct rte_mbuf *buf;
1738 	int i, mbuf_num = 0;
1739 
1740 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1741 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1742 
1743 		PMD_INIT_LOG(DEBUG,
1744 			     "Before freeing rxq[%d] used and unused buf", i);
1745 		VIRTQUEUE_DUMP(rxvq->vq);
1746 
1747 		PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1748 		while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1749 			rte_pktmbuf_free(buf);
1750 			mbuf_num++;
1751 		}
1752 
1753 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1754 		PMD_INIT_LOG(DEBUG,
1755 			     "After freeing rxq[%d] used and unused buf", i);
1756 		VIRTQUEUE_DUMP(rxvq->vq);
1757 	}
1758 
1759 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1760 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1761 
1762 		PMD_INIT_LOG(DEBUG,
1763 			     "Before freeing txq[%d] used and unused bufs",
1764 			     i);
1765 		VIRTQUEUE_DUMP(txvq->vq);
1766 
1767 		mbuf_num = 0;
1768 		while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1769 			rte_pktmbuf_free(buf);
1770 			mbuf_num++;
1771 		}
1772 
1773 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1774 		PMD_INIT_LOG(DEBUG,
1775 			     "After freeing txq[%d] used and unused buf", i);
1776 		VIRTQUEUE_DUMP(txvq->vq);
1777 	}
1778 }
1779 
1780 /*
1781  * Stop device: disable interrupt and mark link down
1782  */
1783 static void
1784 virtio_dev_stop(struct rte_eth_dev *dev)
1785 {
1786 	struct rte_eth_link link;
1787 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1788 
1789 	PMD_INIT_LOG(DEBUG, "stop");
1790 
1791 	if (intr_conf->lsc || intr_conf->rxq)
1792 		rte_intr_disable(dev->intr_handle);
1793 
1794 	memset(&link, 0, sizeof(link));
1795 	virtio_dev_atomic_write_link_status(dev, &link);
1796 }
1797 
1798 static int
1799 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1800 {
1801 	struct rte_eth_link link, old;
1802 	uint16_t status;
1803 	struct virtio_hw *hw = dev->data->dev_private;
1804 	memset(&link, 0, sizeof(link));
1805 	virtio_dev_atomic_read_link_status(dev, &link);
1806 	old = link;
1807 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1808 	link.link_speed  = SPEED_10G;
1809 
1810 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1811 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
1812 		vtpci_read_dev_config(hw,
1813 				offsetof(struct virtio_net_config, status),
1814 				&status, sizeof(status));
1815 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1816 			link.link_status = ETH_LINK_DOWN;
1817 			PMD_INIT_LOG(DEBUG, "Port %d is down",
1818 				     dev->data->port_id);
1819 		} else {
1820 			link.link_status = ETH_LINK_UP;
1821 			PMD_INIT_LOG(DEBUG, "Port %d is up",
1822 				     dev->data->port_id);
1823 		}
1824 	} else {
1825 		link.link_status = ETH_LINK_UP;
1826 	}
1827 	virtio_dev_atomic_write_link_status(dev, &link);
1828 
1829 	return (old.link_status == link.link_status) ? -1 : 0;
1830 }
1831 
1832 static void
1833 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1834 {
1835 	uint64_t tso_mask, host_features;
1836 	struct virtio_hw *hw = dev->data->dev_private;
1837 
1838 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
1839 
1840 	dev_info->pci_dev = dev->device ? RTE_DEV_TO_PCI(dev->device) : NULL;
1841 	dev_info->max_rx_queues =
1842 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1843 	dev_info->max_tx_queues =
1844 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1845 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1846 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1847 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1848 	dev_info->default_txconf = (struct rte_eth_txconf) {
1849 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1850 	};
1851 
1852 	host_features = VTPCI_OPS(hw)->get_features(hw);
1853 	dev_info->rx_offload_capa = 0;
1854 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
1855 		dev_info->rx_offload_capa |=
1856 			DEV_RX_OFFLOAD_TCP_CKSUM |
1857 			DEV_RX_OFFLOAD_UDP_CKSUM;
1858 	}
1859 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1860 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1861 	if ((host_features & tso_mask) == tso_mask)
1862 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
1863 
1864 	dev_info->tx_offload_capa = 0;
1865 	if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
1866 		dev_info->tx_offload_capa |=
1867 			DEV_TX_OFFLOAD_UDP_CKSUM |
1868 			DEV_TX_OFFLOAD_TCP_CKSUM;
1869 	}
1870 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1871 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
1872 	if ((hw->guest_features & tso_mask) == tso_mask)
1873 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
1874 }
1875 
1876 /*
1877  * It enables testpmd to collect per queue stats.
1878  */
1879 static int
1880 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1881 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1882 __rte_unused uint8_t is_rx)
1883 {
1884 	return 0;
1885 }
1886 
1887 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
1888 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
1889 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio");
1890