xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 91dc9c13ba978fb8147240ed6fa20c41145bf0db)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4 
5 #include <stdint.h>
6 #include <string.h>
7 #include <stdio.h>
8 #include <errno.h>
9 #include <unistd.h>
10 
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_memcpy.h>
14 #include <rte_string_fns.h>
15 #include <rte_memzone.h>
16 #include <rte_malloc.h>
17 #include <rte_branch_prediction.h>
18 #include <rte_pci.h>
19 #include <rte_bus_pci.h>
20 #include <rte_ether.h>
21 #include <rte_ip.h>
22 #include <rte_arp.h>
23 #include <rte_common.h>
24 #include <rte_errno.h>
25 #include <rte_cpuflags.h>
26 
27 #include <rte_memory.h>
28 #include <rte_eal.h>
29 #include <rte_dev.h>
30 #include <rte_cycles.h>
31 #include <rte_kvargs.h>
32 
33 #include "virtio_ethdev.h"
34 #include "virtio_pci.h"
35 #include "virtio_logs.h"
36 #include "virtqueue.h"
37 #include "virtio_rxtx.h"
38 
39 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
40 static int  virtio_dev_configure(struct rte_eth_dev *dev);
41 static int  virtio_dev_start(struct rte_eth_dev *dev);
42 static void virtio_dev_stop(struct rte_eth_dev *dev);
43 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
44 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
45 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
46 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
47 static void virtio_dev_info_get(struct rte_eth_dev *dev,
48 				struct rte_eth_dev_info *dev_info);
49 static int virtio_dev_link_update(struct rte_eth_dev *dev,
50 	int wait_to_complete);
51 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
52 
53 static void virtio_set_hwaddr(struct virtio_hw *hw);
54 static void virtio_get_hwaddr(struct virtio_hw *hw);
55 
56 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
57 				 struct rte_eth_stats *stats);
58 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
59 				 struct rte_eth_xstat *xstats, unsigned n);
60 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
61 				       struct rte_eth_xstat_name *xstats_names,
62 				       unsigned limit);
63 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
64 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
65 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
66 				uint16_t vlan_id, int on);
67 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
68 				struct ether_addr *mac_addr,
69 				uint32_t index, uint32_t vmdq);
70 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
71 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
72 				struct ether_addr *mac_addr);
73 
74 static int virtio_intr_enable(struct rte_eth_dev *dev);
75 static int virtio_intr_disable(struct rte_eth_dev *dev);
76 
77 static int virtio_dev_queue_stats_mapping_set(
78 	struct rte_eth_dev *eth_dev,
79 	uint16_t queue_id,
80 	uint8_t stat_idx,
81 	uint8_t is_rx);
82 
83 int virtio_logtype_init;
84 int virtio_logtype_driver;
85 
86 static void virtio_notify_peers(struct rte_eth_dev *dev);
87 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
88 
89 /*
90  * The set of PCI devices this driver supports
91  */
92 static const struct rte_pci_id pci_id_virtio_map[] = {
93 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
94 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
95 	{ .vendor_id = 0, /* sentinel */ },
96 };
97 
98 struct rte_virtio_xstats_name_off {
99 	char name[RTE_ETH_XSTATS_NAME_SIZE];
100 	unsigned offset;
101 };
102 
103 /* [rt]x_qX_ is prepended to the name string here */
104 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
105 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
106 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
107 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
108 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
109 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
110 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
111 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
112 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
113 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
114 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
115 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
116 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
117 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
118 };
119 
120 /* [rt]x_qX_ is prepended to the name string here */
121 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
122 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
123 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
124 	{"errors",                 offsetof(struct virtnet_tx, stats.errors)},
125 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
126 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
127 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
128 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
129 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
130 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
131 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
132 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
133 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
134 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
135 };
136 
137 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
138 			    sizeof(rte_virtio_rxq_stat_strings[0]))
139 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
140 			    sizeof(rte_virtio_txq_stat_strings[0]))
141 
142 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
143 
144 static int
145 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
146 		int *dlen, int pkt_num)
147 {
148 	uint32_t head, i;
149 	int k, sum = 0;
150 	virtio_net_ctrl_ack status = ~0;
151 	struct virtio_pmd_ctrl *result;
152 	struct virtqueue *vq;
153 
154 	ctrl->status = status;
155 
156 	if (!cvq || !cvq->vq) {
157 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
158 		return -1;
159 	}
160 
161 	rte_spinlock_lock(&cvq->lock);
162 	vq = cvq->vq;
163 	head = vq->vq_desc_head_idx;
164 
165 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
166 		"vq->hw->cvq = %p vq = %p",
167 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
168 
169 	if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
170 		rte_spinlock_unlock(&cvq->lock);
171 		return -1;
172 	}
173 
174 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
175 		sizeof(struct virtio_pmd_ctrl));
176 
177 	/*
178 	 * Format is enforced in qemu code:
179 	 * One TX packet for header;
180 	 * At least one TX packet per argument;
181 	 * One RX packet for ACK.
182 	 */
183 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
184 	vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
185 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
186 	vq->vq_free_cnt--;
187 	i = vq->vq_ring.desc[head].next;
188 
189 	for (k = 0; k < pkt_num; k++) {
190 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
191 		vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
192 			+ sizeof(struct virtio_net_ctrl_hdr)
193 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
194 		vq->vq_ring.desc[i].len = dlen[k];
195 		sum += dlen[k];
196 		vq->vq_free_cnt--;
197 		i = vq->vq_ring.desc[i].next;
198 	}
199 
200 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
201 	vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
202 			+ sizeof(struct virtio_net_ctrl_hdr);
203 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
204 	vq->vq_free_cnt--;
205 
206 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
207 
208 	vq_update_avail_ring(vq, head);
209 	vq_update_avail_idx(vq);
210 
211 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
212 
213 	virtqueue_notify(vq);
214 
215 	rte_rmb();
216 	while (VIRTQUEUE_NUSED(vq) == 0) {
217 		rte_rmb();
218 		usleep(100);
219 	}
220 
221 	while (VIRTQUEUE_NUSED(vq)) {
222 		uint32_t idx, desc_idx, used_idx;
223 		struct vring_used_elem *uep;
224 
225 		used_idx = (uint32_t)(vq->vq_used_cons_idx
226 				& (vq->vq_nentries - 1));
227 		uep = &vq->vq_ring.used->ring[used_idx];
228 		idx = (uint32_t) uep->id;
229 		desc_idx = idx;
230 
231 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
232 			desc_idx = vq->vq_ring.desc[desc_idx].next;
233 			vq->vq_free_cnt++;
234 		}
235 
236 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
237 		vq->vq_desc_head_idx = idx;
238 
239 		vq->vq_used_cons_idx++;
240 		vq->vq_free_cnt++;
241 	}
242 
243 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
244 			vq->vq_free_cnt, vq->vq_desc_head_idx);
245 
246 	result = cvq->virtio_net_hdr_mz->addr;
247 
248 	rte_spinlock_unlock(&cvq->lock);
249 	return result->status;
250 }
251 
252 static int
253 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
254 {
255 	struct virtio_hw *hw = dev->data->dev_private;
256 	struct virtio_pmd_ctrl ctrl;
257 	int dlen[1];
258 	int ret;
259 
260 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
261 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
262 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
263 
264 	dlen[0] = sizeof(uint16_t);
265 
266 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
267 	if (ret) {
268 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
269 			  "failed, this is too late now...");
270 		return -EINVAL;
271 	}
272 
273 	return 0;
274 }
275 
276 static void
277 virtio_dev_queue_release(void *queue __rte_unused)
278 {
279 	/* do nothing */
280 }
281 
282 static uint16_t
283 virtio_get_nr_vq(struct virtio_hw *hw)
284 {
285 	uint16_t nr_vq = hw->max_queue_pairs * 2;
286 
287 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
288 		nr_vq += 1;
289 
290 	return nr_vq;
291 }
292 
293 static void
294 virtio_init_vring(struct virtqueue *vq)
295 {
296 	int size = vq->vq_nentries;
297 	struct vring *vr = &vq->vq_ring;
298 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
299 
300 	PMD_INIT_FUNC_TRACE();
301 
302 	/*
303 	 * Reinitialise since virtio port might have been stopped and restarted
304 	 */
305 	memset(ring_mem, 0, vq->vq_ring_size);
306 	vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
307 	vq->vq_used_cons_idx = 0;
308 	vq->vq_desc_head_idx = 0;
309 	vq->vq_avail_idx = 0;
310 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
311 	vq->vq_free_cnt = vq->vq_nentries;
312 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
313 
314 	vring_desc_init(vr->desc, size);
315 
316 	/*
317 	 * Disable device(host) interrupting guest
318 	 */
319 	virtqueue_disable_intr(vq);
320 }
321 
322 static int
323 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
324 {
325 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
326 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
327 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
328 	unsigned int vq_size, size;
329 	struct virtio_hw *hw = dev->data->dev_private;
330 	struct virtnet_rx *rxvq = NULL;
331 	struct virtnet_tx *txvq = NULL;
332 	struct virtnet_ctl *cvq = NULL;
333 	struct virtqueue *vq;
334 	size_t sz_hdr_mz = 0;
335 	void *sw_ring = NULL;
336 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
337 	int ret;
338 
339 	PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
340 
341 	/*
342 	 * Read the virtqueue size from the Queue Size field
343 	 * Always power of 2 and if 0 virtqueue does not exist
344 	 */
345 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
346 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
347 	if (vq_size == 0) {
348 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
349 		return -EINVAL;
350 	}
351 
352 	if (!rte_is_power_of_2(vq_size)) {
353 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
354 		return -EINVAL;
355 	}
356 
357 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
358 		 dev->data->port_id, vtpci_queue_idx);
359 
360 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
361 				vq_size * sizeof(struct vq_desc_extra),
362 				RTE_CACHE_LINE_SIZE);
363 	if (queue_type == VTNET_TQ) {
364 		/*
365 		 * For each xmit packet, allocate a virtio_net_hdr
366 		 * and indirect ring elements
367 		 */
368 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
369 	} else if (queue_type == VTNET_CQ) {
370 		/* Allocate a page for control vq command, data and status */
371 		sz_hdr_mz = PAGE_SIZE;
372 	}
373 
374 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
375 				SOCKET_ID_ANY);
376 	if (vq == NULL) {
377 		PMD_INIT_LOG(ERR, "can not allocate vq");
378 		return -ENOMEM;
379 	}
380 	hw->vqs[vtpci_queue_idx] = vq;
381 
382 	vq->hw = hw;
383 	vq->vq_queue_index = vtpci_queue_idx;
384 	vq->vq_nentries = vq_size;
385 
386 	/*
387 	 * Reserve a memzone for vring elements
388 	 */
389 	size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
390 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
391 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
392 		     size, vq->vq_ring_size);
393 
394 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
395 			SOCKET_ID_ANY, RTE_MEMZONE_IOVA_CONTIG,
396 			VIRTIO_PCI_VRING_ALIGN);
397 	if (mz == NULL) {
398 		if (rte_errno == EEXIST)
399 			mz = rte_memzone_lookup(vq_name);
400 		if (mz == NULL) {
401 			ret = -ENOMEM;
402 			goto fail_q_alloc;
403 		}
404 	}
405 
406 	memset(mz->addr, 0, mz->len);
407 
408 	vq->vq_ring_mem = mz->iova;
409 	vq->vq_ring_virt_mem = mz->addr;
410 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
411 		     (uint64_t)mz->iova);
412 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
413 		     (uint64_t)(uintptr_t)mz->addr);
414 
415 	virtio_init_vring(vq);
416 
417 	if (sz_hdr_mz) {
418 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
419 			 dev->data->port_id, vtpci_queue_idx);
420 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
421 				SOCKET_ID_ANY, RTE_MEMZONE_IOVA_CONTIG,
422 				RTE_CACHE_LINE_SIZE);
423 		if (hdr_mz == NULL) {
424 			if (rte_errno == EEXIST)
425 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
426 			if (hdr_mz == NULL) {
427 				ret = -ENOMEM;
428 				goto fail_q_alloc;
429 			}
430 		}
431 	}
432 
433 	if (queue_type == VTNET_RQ) {
434 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
435 			       sizeof(vq->sw_ring[0]);
436 
437 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
438 				RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
439 		if (!sw_ring) {
440 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
441 			ret = -ENOMEM;
442 			goto fail_q_alloc;
443 		}
444 
445 		vq->sw_ring = sw_ring;
446 		rxvq = &vq->rxq;
447 		rxvq->vq = vq;
448 		rxvq->port_id = dev->data->port_id;
449 		rxvq->mz = mz;
450 	} else if (queue_type == VTNET_TQ) {
451 		txvq = &vq->txq;
452 		txvq->vq = vq;
453 		txvq->port_id = dev->data->port_id;
454 		txvq->mz = mz;
455 		txvq->virtio_net_hdr_mz = hdr_mz;
456 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
457 	} else if (queue_type == VTNET_CQ) {
458 		cvq = &vq->cq;
459 		cvq->vq = vq;
460 		cvq->mz = mz;
461 		cvq->virtio_net_hdr_mz = hdr_mz;
462 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
463 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
464 
465 		hw->cvq = cvq;
466 	}
467 
468 	/* For virtio_user case (that is when hw->dev is NULL), we use
469 	 * virtual address. And we need properly set _offset_, please see
470 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
471 	 */
472 	if (!hw->virtio_user_dev)
473 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
474 	else {
475 		vq->vq_ring_mem = (uintptr_t)mz->addr;
476 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
477 		if (queue_type == VTNET_TQ)
478 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
479 		else if (queue_type == VTNET_CQ)
480 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
481 	}
482 
483 	if (queue_type == VTNET_TQ) {
484 		struct virtio_tx_region *txr;
485 		unsigned int i;
486 
487 		txr = hdr_mz->addr;
488 		memset(txr, 0, vq_size * sizeof(*txr));
489 		for (i = 0; i < vq_size; i++) {
490 			struct vring_desc *start_dp = txr[i].tx_indir;
491 
492 			vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
493 
494 			/* first indirect descriptor is always the tx header */
495 			start_dp->addr = txvq->virtio_net_hdr_mem
496 				+ i * sizeof(*txr)
497 				+ offsetof(struct virtio_tx_region, tx_hdr);
498 
499 			start_dp->len = hw->vtnet_hdr_size;
500 			start_dp->flags = VRING_DESC_F_NEXT;
501 		}
502 	}
503 
504 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
505 		PMD_INIT_LOG(ERR, "setup_queue failed");
506 		return -EINVAL;
507 	}
508 
509 	return 0;
510 
511 fail_q_alloc:
512 	rte_free(sw_ring);
513 	rte_memzone_free(hdr_mz);
514 	rte_memzone_free(mz);
515 	rte_free(vq);
516 
517 	return ret;
518 }
519 
520 static void
521 virtio_free_queues(struct virtio_hw *hw)
522 {
523 	uint16_t nr_vq = virtio_get_nr_vq(hw);
524 	struct virtqueue *vq;
525 	int queue_type;
526 	uint16_t i;
527 
528 	if (hw->vqs == NULL)
529 		return;
530 
531 	for (i = 0; i < nr_vq; i++) {
532 		vq = hw->vqs[i];
533 		if (!vq)
534 			continue;
535 
536 		queue_type = virtio_get_queue_type(hw, i);
537 		if (queue_type == VTNET_RQ) {
538 			rte_free(vq->sw_ring);
539 			rte_memzone_free(vq->rxq.mz);
540 		} else if (queue_type == VTNET_TQ) {
541 			rte_memzone_free(vq->txq.mz);
542 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
543 		} else {
544 			rte_memzone_free(vq->cq.mz);
545 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
546 		}
547 
548 		rte_free(vq);
549 		hw->vqs[i] = NULL;
550 	}
551 
552 	rte_free(hw->vqs);
553 	hw->vqs = NULL;
554 }
555 
556 static int
557 virtio_alloc_queues(struct rte_eth_dev *dev)
558 {
559 	struct virtio_hw *hw = dev->data->dev_private;
560 	uint16_t nr_vq = virtio_get_nr_vq(hw);
561 	uint16_t i;
562 	int ret;
563 
564 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
565 	if (!hw->vqs) {
566 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
567 		return -ENOMEM;
568 	}
569 
570 	for (i = 0; i < nr_vq; i++) {
571 		ret = virtio_init_queue(dev, i);
572 		if (ret < 0) {
573 			virtio_free_queues(hw);
574 			return ret;
575 		}
576 	}
577 
578 	return 0;
579 }
580 
581 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
582 
583 static void
584 virtio_dev_close(struct rte_eth_dev *dev)
585 {
586 	struct virtio_hw *hw = dev->data->dev_private;
587 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
588 
589 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
590 
591 	if (!hw->opened)
592 		return;
593 	hw->opened = false;
594 
595 	/* reset the NIC */
596 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
597 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
598 	if (intr_conf->rxq)
599 		virtio_queues_unbind_intr(dev);
600 
601 	if (intr_conf->lsc || intr_conf->rxq) {
602 		virtio_intr_disable(dev);
603 		rte_intr_efd_disable(dev->intr_handle);
604 		rte_free(dev->intr_handle->intr_vec);
605 		dev->intr_handle->intr_vec = NULL;
606 	}
607 
608 	vtpci_reset(hw);
609 	virtio_dev_free_mbufs(dev);
610 	virtio_free_queues(hw);
611 }
612 
613 static void
614 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
615 {
616 	struct virtio_hw *hw = dev->data->dev_private;
617 	struct virtio_pmd_ctrl ctrl;
618 	int dlen[1];
619 	int ret;
620 
621 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
622 		PMD_INIT_LOG(INFO, "host does not support rx control");
623 		return;
624 	}
625 
626 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
627 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
628 	ctrl.data[0] = 1;
629 	dlen[0] = 1;
630 
631 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
632 	if (ret)
633 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
634 }
635 
636 static void
637 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
638 {
639 	struct virtio_hw *hw = dev->data->dev_private;
640 	struct virtio_pmd_ctrl ctrl;
641 	int dlen[1];
642 	int ret;
643 
644 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
645 		PMD_INIT_LOG(INFO, "host does not support rx control");
646 		return;
647 	}
648 
649 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
650 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
651 	ctrl.data[0] = 0;
652 	dlen[0] = 1;
653 
654 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
655 	if (ret)
656 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
657 }
658 
659 static void
660 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
661 {
662 	struct virtio_hw *hw = dev->data->dev_private;
663 	struct virtio_pmd_ctrl ctrl;
664 	int dlen[1];
665 	int ret;
666 
667 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
668 		PMD_INIT_LOG(INFO, "host does not support rx control");
669 		return;
670 	}
671 
672 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
673 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
674 	ctrl.data[0] = 1;
675 	dlen[0] = 1;
676 
677 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
678 	if (ret)
679 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
680 }
681 
682 static void
683 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
684 {
685 	struct virtio_hw *hw = dev->data->dev_private;
686 	struct virtio_pmd_ctrl ctrl;
687 	int dlen[1];
688 	int ret;
689 
690 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
691 		PMD_INIT_LOG(INFO, "host does not support rx control");
692 		return;
693 	}
694 
695 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
696 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
697 	ctrl.data[0] = 0;
698 	dlen[0] = 1;
699 
700 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
701 	if (ret)
702 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
703 }
704 
705 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
706 static int
707 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
708 {
709 	struct virtio_hw *hw = dev->data->dev_private;
710 	uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
711 				 hw->vtnet_hdr_size;
712 	uint32_t frame_size = mtu + ether_hdr_len;
713 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
714 
715 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
716 
717 	if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
718 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
719 			ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
720 		return -EINVAL;
721 	}
722 	return 0;
723 }
724 
725 static int
726 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
727 {
728 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
729 	struct virtqueue *vq = rxvq->vq;
730 
731 	virtqueue_enable_intr(vq);
732 	return 0;
733 }
734 
735 static int
736 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
737 {
738 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
739 	struct virtqueue *vq = rxvq->vq;
740 
741 	virtqueue_disable_intr(vq);
742 	return 0;
743 }
744 
745 /*
746  * dev_ops for virtio, bare necessities for basic operation
747  */
748 static const struct eth_dev_ops virtio_eth_dev_ops = {
749 	.dev_configure           = virtio_dev_configure,
750 	.dev_start               = virtio_dev_start,
751 	.dev_stop                = virtio_dev_stop,
752 	.dev_close               = virtio_dev_close,
753 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
754 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
755 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
756 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
757 	.mtu_set                 = virtio_mtu_set,
758 	.dev_infos_get           = virtio_dev_info_get,
759 	.stats_get               = virtio_dev_stats_get,
760 	.xstats_get              = virtio_dev_xstats_get,
761 	.xstats_get_names        = virtio_dev_xstats_get_names,
762 	.stats_reset             = virtio_dev_stats_reset,
763 	.xstats_reset            = virtio_dev_stats_reset,
764 	.link_update             = virtio_dev_link_update,
765 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
766 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
767 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
768 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
769 	.rx_queue_release        = virtio_dev_queue_release,
770 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
771 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
772 	.tx_queue_release        = virtio_dev_queue_release,
773 	/* collect stats per queue */
774 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
775 	.vlan_filter_set         = virtio_vlan_filter_set,
776 	.mac_addr_add            = virtio_mac_addr_add,
777 	.mac_addr_remove         = virtio_mac_addr_remove,
778 	.mac_addr_set            = virtio_mac_addr_set,
779 };
780 
781 static void
782 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
783 {
784 	unsigned i;
785 
786 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
787 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
788 		if (txvq == NULL)
789 			continue;
790 
791 		stats->opackets += txvq->stats.packets;
792 		stats->obytes += txvq->stats.bytes;
793 		stats->oerrors += txvq->stats.errors;
794 
795 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
796 			stats->q_opackets[i] = txvq->stats.packets;
797 			stats->q_obytes[i] = txvq->stats.bytes;
798 		}
799 	}
800 
801 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
802 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
803 		if (rxvq == NULL)
804 			continue;
805 
806 		stats->ipackets += rxvq->stats.packets;
807 		stats->ibytes += rxvq->stats.bytes;
808 		stats->ierrors += rxvq->stats.errors;
809 
810 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
811 			stats->q_ipackets[i] = rxvq->stats.packets;
812 			stats->q_ibytes[i] = rxvq->stats.bytes;
813 		}
814 	}
815 
816 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
817 }
818 
819 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
820 				       struct rte_eth_xstat_name *xstats_names,
821 				       __rte_unused unsigned limit)
822 {
823 	unsigned i;
824 	unsigned count = 0;
825 	unsigned t;
826 
827 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
828 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
829 
830 	if (xstats_names != NULL) {
831 		/* Note: limit checked in rte_eth_xstats_names() */
832 
833 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
834 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
835 			if (rxvq == NULL)
836 				continue;
837 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
838 				snprintf(xstats_names[count].name,
839 					sizeof(xstats_names[count].name),
840 					"rx_q%u_%s", i,
841 					rte_virtio_rxq_stat_strings[t].name);
842 				count++;
843 			}
844 		}
845 
846 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
847 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
848 			if (txvq == NULL)
849 				continue;
850 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
851 				snprintf(xstats_names[count].name,
852 					sizeof(xstats_names[count].name),
853 					"tx_q%u_%s", i,
854 					rte_virtio_txq_stat_strings[t].name);
855 				count++;
856 			}
857 		}
858 		return count;
859 	}
860 	return nstats;
861 }
862 
863 static int
864 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
865 		      unsigned n)
866 {
867 	unsigned i;
868 	unsigned count = 0;
869 
870 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
871 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
872 
873 	if (n < nstats)
874 		return nstats;
875 
876 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
877 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
878 
879 		if (rxvq == NULL)
880 			continue;
881 
882 		unsigned t;
883 
884 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
885 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
886 				rte_virtio_rxq_stat_strings[t].offset);
887 			xstats[count].id = count;
888 			count++;
889 		}
890 	}
891 
892 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
893 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
894 
895 		if (txvq == NULL)
896 			continue;
897 
898 		unsigned t;
899 
900 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
901 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
902 				rte_virtio_txq_stat_strings[t].offset);
903 			xstats[count].id = count;
904 			count++;
905 		}
906 	}
907 
908 	return count;
909 }
910 
911 static int
912 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
913 {
914 	virtio_update_stats(dev, stats);
915 
916 	return 0;
917 }
918 
919 static void
920 virtio_dev_stats_reset(struct rte_eth_dev *dev)
921 {
922 	unsigned int i;
923 
924 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
925 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
926 		if (txvq == NULL)
927 			continue;
928 
929 		txvq->stats.packets = 0;
930 		txvq->stats.bytes = 0;
931 		txvq->stats.errors = 0;
932 		txvq->stats.multicast = 0;
933 		txvq->stats.broadcast = 0;
934 		memset(txvq->stats.size_bins, 0,
935 		       sizeof(txvq->stats.size_bins[0]) * 8);
936 	}
937 
938 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
939 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
940 		if (rxvq == NULL)
941 			continue;
942 
943 		rxvq->stats.packets = 0;
944 		rxvq->stats.bytes = 0;
945 		rxvq->stats.errors = 0;
946 		rxvq->stats.multicast = 0;
947 		rxvq->stats.broadcast = 0;
948 		memset(rxvq->stats.size_bins, 0,
949 		       sizeof(rxvq->stats.size_bins[0]) * 8);
950 	}
951 }
952 
953 static void
954 virtio_set_hwaddr(struct virtio_hw *hw)
955 {
956 	vtpci_write_dev_config(hw,
957 			offsetof(struct virtio_net_config, mac),
958 			&hw->mac_addr, ETHER_ADDR_LEN);
959 }
960 
961 static void
962 virtio_get_hwaddr(struct virtio_hw *hw)
963 {
964 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
965 		vtpci_read_dev_config(hw,
966 			offsetof(struct virtio_net_config, mac),
967 			&hw->mac_addr, ETHER_ADDR_LEN);
968 	} else {
969 		eth_random_addr(&hw->mac_addr[0]);
970 		virtio_set_hwaddr(hw);
971 	}
972 }
973 
974 static int
975 virtio_mac_table_set(struct virtio_hw *hw,
976 		     const struct virtio_net_ctrl_mac *uc,
977 		     const struct virtio_net_ctrl_mac *mc)
978 {
979 	struct virtio_pmd_ctrl ctrl;
980 	int err, len[2];
981 
982 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
983 		PMD_DRV_LOG(INFO, "host does not support mac table");
984 		return -1;
985 	}
986 
987 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
988 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
989 
990 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
991 	memcpy(ctrl.data, uc, len[0]);
992 
993 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
994 	memcpy(ctrl.data + len[0], mc, len[1]);
995 
996 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
997 	if (err != 0)
998 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
999 	return err;
1000 }
1001 
1002 static int
1003 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1004 		    uint32_t index, uint32_t vmdq __rte_unused)
1005 {
1006 	struct virtio_hw *hw = dev->data->dev_private;
1007 	const struct ether_addr *addrs = dev->data->mac_addrs;
1008 	unsigned int i;
1009 	struct virtio_net_ctrl_mac *uc, *mc;
1010 
1011 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1012 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1013 		return -EINVAL;
1014 	}
1015 
1016 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1017 	uc->entries = 0;
1018 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1019 	mc->entries = 0;
1020 
1021 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1022 		const struct ether_addr *addr
1023 			= (i == index) ? mac_addr : addrs + i;
1024 		struct virtio_net_ctrl_mac *tbl
1025 			= is_multicast_ether_addr(addr) ? mc : uc;
1026 
1027 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1028 	}
1029 
1030 	return virtio_mac_table_set(hw, uc, mc);
1031 }
1032 
1033 static void
1034 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1035 {
1036 	struct virtio_hw *hw = dev->data->dev_private;
1037 	struct ether_addr *addrs = dev->data->mac_addrs;
1038 	struct virtio_net_ctrl_mac *uc, *mc;
1039 	unsigned int i;
1040 
1041 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1042 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1043 		return;
1044 	}
1045 
1046 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1047 	uc->entries = 0;
1048 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1049 	mc->entries = 0;
1050 
1051 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1052 		struct virtio_net_ctrl_mac *tbl;
1053 
1054 		if (i == index || is_zero_ether_addr(addrs + i))
1055 			continue;
1056 
1057 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1058 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1059 	}
1060 
1061 	virtio_mac_table_set(hw, uc, mc);
1062 }
1063 
1064 static int
1065 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1066 {
1067 	struct virtio_hw *hw = dev->data->dev_private;
1068 
1069 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1070 
1071 	/* Use atomic update if available */
1072 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1073 		struct virtio_pmd_ctrl ctrl;
1074 		int len = ETHER_ADDR_LEN;
1075 
1076 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1077 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1078 
1079 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1080 		return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1081 	}
1082 
1083 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1084 		return -ENOTSUP;
1085 
1086 	virtio_set_hwaddr(hw);
1087 	return 0;
1088 }
1089 
1090 static int
1091 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1092 {
1093 	struct virtio_hw *hw = dev->data->dev_private;
1094 	struct virtio_pmd_ctrl ctrl;
1095 	int len;
1096 
1097 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1098 		return -ENOTSUP;
1099 
1100 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1101 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1102 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1103 	len = sizeof(vlan_id);
1104 
1105 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1106 }
1107 
1108 static int
1109 virtio_intr_enable(struct rte_eth_dev *dev)
1110 {
1111 	struct virtio_hw *hw = dev->data->dev_private;
1112 
1113 	if (rte_intr_enable(dev->intr_handle) < 0)
1114 		return -1;
1115 
1116 	if (!hw->virtio_user_dev)
1117 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1118 
1119 	return 0;
1120 }
1121 
1122 static int
1123 virtio_intr_disable(struct rte_eth_dev *dev)
1124 {
1125 	struct virtio_hw *hw = dev->data->dev_private;
1126 
1127 	if (rte_intr_disable(dev->intr_handle) < 0)
1128 		return -1;
1129 
1130 	if (!hw->virtio_user_dev)
1131 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1132 
1133 	return 0;
1134 }
1135 
1136 static int
1137 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1138 {
1139 	uint64_t host_features;
1140 
1141 	/* Prepare guest_features: feature that driver wants to support */
1142 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1143 		req_features);
1144 
1145 	/* Read device(host) feature bits */
1146 	host_features = VTPCI_OPS(hw)->get_features(hw);
1147 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1148 		host_features);
1149 
1150 	/* If supported, ensure MTU value is valid before acknowledging it. */
1151 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1152 		struct virtio_net_config config;
1153 
1154 		vtpci_read_dev_config(hw,
1155 			offsetof(struct virtio_net_config, mtu),
1156 			&config.mtu, sizeof(config.mtu));
1157 
1158 		if (config.mtu < ETHER_MIN_MTU)
1159 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1160 	}
1161 
1162 	/*
1163 	 * Negotiate features: Subset of device feature bits are written back
1164 	 * guest feature bits.
1165 	 */
1166 	hw->guest_features = req_features;
1167 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1168 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1169 		hw->guest_features);
1170 
1171 	if (hw->modern) {
1172 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1173 			PMD_INIT_LOG(ERR,
1174 				"VIRTIO_F_VERSION_1 features is not enabled.");
1175 			return -1;
1176 		}
1177 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1178 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1179 			PMD_INIT_LOG(ERR,
1180 				"failed to set FEATURES_OK status!");
1181 			return -1;
1182 		}
1183 	}
1184 
1185 	hw->req_guest_features = req_features;
1186 
1187 	return 0;
1188 }
1189 
1190 int
1191 virtio_dev_pause(struct rte_eth_dev *dev)
1192 {
1193 	struct virtio_hw *hw = dev->data->dev_private;
1194 
1195 	rte_spinlock_lock(&hw->state_lock);
1196 
1197 	if (hw->started == 0) {
1198 		/* Device is just stopped. */
1199 		rte_spinlock_unlock(&hw->state_lock);
1200 		return -1;
1201 	}
1202 	hw->started = 0;
1203 	/*
1204 	 * Prevent the worker threads from touching queues to avoid contention,
1205 	 * 1 ms should be enough for the ongoing Tx function to finish.
1206 	 */
1207 	rte_delay_ms(1);
1208 	return 0;
1209 }
1210 
1211 /*
1212  * Recover hw state to let the worker threads continue.
1213  */
1214 void
1215 virtio_dev_resume(struct rte_eth_dev *dev)
1216 {
1217 	struct virtio_hw *hw = dev->data->dev_private;
1218 
1219 	hw->started = 1;
1220 	rte_spinlock_unlock(&hw->state_lock);
1221 }
1222 
1223 /*
1224  * Should be called only after device is paused.
1225  */
1226 int
1227 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1228 		int nb_pkts)
1229 {
1230 	struct virtio_hw *hw = dev->data->dev_private;
1231 	struct virtnet_tx *txvq = dev->data->tx_queues[0];
1232 	int ret;
1233 
1234 	hw->inject_pkts = tx_pkts;
1235 	ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1236 	hw->inject_pkts = NULL;
1237 
1238 	return ret;
1239 }
1240 
1241 static void
1242 virtio_notify_peers(struct rte_eth_dev *dev)
1243 {
1244 	struct virtio_hw *hw = dev->data->dev_private;
1245 	struct virtnet_rx *rxvq;
1246 	struct rte_mbuf *rarp_mbuf;
1247 
1248 	if (!dev->data->rx_queues)
1249 		return;
1250 
1251 	rxvq = dev->data->rx_queues[0];
1252 	if (!rxvq)
1253 		return;
1254 
1255 	rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1256 			(struct ether_addr *)hw->mac_addr);
1257 	if (rarp_mbuf == NULL) {
1258 		PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1259 		return;
1260 	}
1261 
1262 	/* If virtio port just stopped, no need to send RARP */
1263 	if (virtio_dev_pause(dev) < 0) {
1264 		rte_pktmbuf_free(rarp_mbuf);
1265 		return;
1266 	}
1267 
1268 	virtio_inject_pkts(dev, &rarp_mbuf, 1);
1269 	virtio_dev_resume(dev);
1270 }
1271 
1272 static void
1273 virtio_ack_link_announce(struct rte_eth_dev *dev)
1274 {
1275 	struct virtio_hw *hw = dev->data->dev_private;
1276 	struct virtio_pmd_ctrl ctrl;
1277 
1278 	ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1279 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1280 
1281 	virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1282 }
1283 
1284 /*
1285  * Process virtio config changed interrupt. Call the callback
1286  * if link state changed, generate gratuitous RARP packet if
1287  * the status indicates an ANNOUNCE.
1288  */
1289 void
1290 virtio_interrupt_handler(void *param)
1291 {
1292 	struct rte_eth_dev *dev = param;
1293 	struct virtio_hw *hw = dev->data->dev_private;
1294 	uint8_t isr;
1295 	uint16_t status;
1296 
1297 	/* Read interrupt status which clears interrupt */
1298 	isr = vtpci_isr(hw);
1299 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1300 
1301 	if (virtio_intr_enable(dev) < 0)
1302 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1303 
1304 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1305 		if (virtio_dev_link_update(dev, 0) == 0)
1306 			_rte_eth_dev_callback_process(dev,
1307 						      RTE_ETH_EVENT_INTR_LSC,
1308 						      NULL);
1309 
1310 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1311 			vtpci_read_dev_config(hw,
1312 				offsetof(struct virtio_net_config, status),
1313 				&status, sizeof(status));
1314 			if (status & VIRTIO_NET_S_ANNOUNCE) {
1315 				virtio_notify_peers(dev);
1316 				if (hw->cvq)
1317 					virtio_ack_link_announce(dev);
1318 			}
1319 		}
1320 	}
1321 }
1322 
1323 /* set rx and tx handlers according to what is supported */
1324 static void
1325 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1326 {
1327 	struct virtio_hw *hw = eth_dev->data->dev_private;
1328 
1329 	if (hw->use_simple_rx) {
1330 		PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1331 			eth_dev->data->port_id);
1332 		eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1333 	} else if (hw->use_inorder_rx) {
1334 		PMD_INIT_LOG(INFO,
1335 			"virtio: using inorder mergeable buffer Rx path on port %u",
1336 			eth_dev->data->port_id);
1337 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts_inorder;
1338 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1339 		PMD_INIT_LOG(INFO,
1340 			"virtio: using mergeable buffer Rx path on port %u",
1341 			eth_dev->data->port_id);
1342 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1343 	} else {
1344 		PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1345 			eth_dev->data->port_id);
1346 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1347 	}
1348 
1349 	if (hw->use_inorder_tx) {
1350 		PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1351 			eth_dev->data->port_id);
1352 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1353 	} else {
1354 		PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1355 			eth_dev->data->port_id);
1356 		eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1357 	}
1358 }
1359 
1360 /* Only support 1:1 queue/interrupt mapping so far.
1361  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1362  * interrupt vectors (<N+1).
1363  */
1364 static int
1365 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1366 {
1367 	uint32_t i;
1368 	struct virtio_hw *hw = dev->data->dev_private;
1369 
1370 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1371 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1372 		dev->intr_handle->intr_vec[i] = i + 1;
1373 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1374 						 VIRTIO_MSI_NO_VECTOR) {
1375 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1376 			return -EBUSY;
1377 		}
1378 	}
1379 
1380 	return 0;
1381 }
1382 
1383 static void
1384 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1385 {
1386 	uint32_t i;
1387 	struct virtio_hw *hw = dev->data->dev_private;
1388 
1389 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1390 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1391 		VTPCI_OPS(hw)->set_queue_irq(hw,
1392 					     hw->vqs[i * VTNET_CQ],
1393 					     VIRTIO_MSI_NO_VECTOR);
1394 }
1395 
1396 static int
1397 virtio_configure_intr(struct rte_eth_dev *dev)
1398 {
1399 	struct virtio_hw *hw = dev->data->dev_private;
1400 
1401 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1402 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1403 		return -ENOTSUP;
1404 	}
1405 
1406 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1407 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1408 		return -1;
1409 	}
1410 
1411 	if (!dev->intr_handle->intr_vec) {
1412 		dev->intr_handle->intr_vec =
1413 			rte_zmalloc("intr_vec",
1414 				    hw->max_queue_pairs * sizeof(int), 0);
1415 		if (!dev->intr_handle->intr_vec) {
1416 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1417 				     hw->max_queue_pairs);
1418 			return -ENOMEM;
1419 		}
1420 	}
1421 
1422 	/* Re-register callback to update max_intr */
1423 	rte_intr_callback_unregister(dev->intr_handle,
1424 				     virtio_interrupt_handler,
1425 				     dev);
1426 	rte_intr_callback_register(dev->intr_handle,
1427 				   virtio_interrupt_handler,
1428 				   dev);
1429 
1430 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1431 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1432 	 * And for legacy devices, this should be done before queue/vec binding
1433 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1434 	 * (22) will be ignored.
1435 	 */
1436 	if (virtio_intr_enable(dev) < 0) {
1437 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1438 		return -1;
1439 	}
1440 
1441 	if (virtio_queues_bind_intr(dev) < 0) {
1442 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1443 		return -1;
1444 	}
1445 
1446 	return 0;
1447 }
1448 
1449 /* reset device and renegotiate features if needed */
1450 static int
1451 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1452 {
1453 	struct virtio_hw *hw = eth_dev->data->dev_private;
1454 	struct virtio_net_config *config;
1455 	struct virtio_net_config local_config;
1456 	struct rte_pci_device *pci_dev = NULL;
1457 	int ret;
1458 
1459 	/* Reset the device although not necessary at startup */
1460 	vtpci_reset(hw);
1461 
1462 	if (hw->vqs) {
1463 		virtio_dev_free_mbufs(eth_dev);
1464 		virtio_free_queues(hw);
1465 	}
1466 
1467 	/* Tell the host we've noticed this device. */
1468 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1469 
1470 	/* Tell the host we've known how to drive the device. */
1471 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1472 	if (virtio_negotiate_features(hw, req_features) < 0)
1473 		return -1;
1474 
1475 	if (!hw->virtio_user_dev) {
1476 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1477 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1478 	}
1479 
1480 	/* If host does not support both status and MSI-X then disable LSC */
1481 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1482 	    hw->use_msix != VIRTIO_MSIX_NONE)
1483 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1484 	else
1485 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1486 
1487 	/* Setting up rx_header size for the device */
1488 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1489 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1490 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1491 	else
1492 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1493 
1494 	/* Copy the permanent MAC address to: virtio_hw */
1495 	virtio_get_hwaddr(hw);
1496 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1497 			&eth_dev->data->mac_addrs[0]);
1498 	PMD_INIT_LOG(DEBUG,
1499 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1500 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1501 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1502 
1503 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1504 		config = &local_config;
1505 
1506 		vtpci_read_dev_config(hw,
1507 			offsetof(struct virtio_net_config, mac),
1508 			&config->mac, sizeof(config->mac));
1509 
1510 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1511 			vtpci_read_dev_config(hw,
1512 				offsetof(struct virtio_net_config, status),
1513 				&config->status, sizeof(config->status));
1514 		} else {
1515 			PMD_INIT_LOG(DEBUG,
1516 				     "VIRTIO_NET_F_STATUS is not supported");
1517 			config->status = 0;
1518 		}
1519 
1520 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1521 			vtpci_read_dev_config(hw,
1522 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1523 				&config->max_virtqueue_pairs,
1524 				sizeof(config->max_virtqueue_pairs));
1525 		} else {
1526 			PMD_INIT_LOG(DEBUG,
1527 				     "VIRTIO_NET_F_MQ is not supported");
1528 			config->max_virtqueue_pairs = 1;
1529 		}
1530 
1531 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1532 
1533 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1534 			vtpci_read_dev_config(hw,
1535 				offsetof(struct virtio_net_config, mtu),
1536 				&config->mtu,
1537 				sizeof(config->mtu));
1538 
1539 			/*
1540 			 * MTU value has already been checked at negotiation
1541 			 * time, but check again in case it has changed since
1542 			 * then, which should not happen.
1543 			 */
1544 			if (config->mtu < ETHER_MIN_MTU) {
1545 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1546 						config->mtu);
1547 				return -1;
1548 			}
1549 
1550 			hw->max_mtu = config->mtu;
1551 			/* Set initial MTU to maximum one supported by vhost */
1552 			eth_dev->data->mtu = config->mtu;
1553 
1554 		} else {
1555 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1556 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1557 		}
1558 
1559 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1560 				config->max_virtqueue_pairs);
1561 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1562 		PMD_INIT_LOG(DEBUG,
1563 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1564 				config->mac[0], config->mac[1],
1565 				config->mac[2], config->mac[3],
1566 				config->mac[4], config->mac[5]);
1567 	} else {
1568 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1569 		hw->max_queue_pairs = 1;
1570 		hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1571 			VLAN_TAG_LEN - hw->vtnet_hdr_size;
1572 	}
1573 
1574 	ret = virtio_alloc_queues(eth_dev);
1575 	if (ret < 0)
1576 		return ret;
1577 
1578 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1579 		if (virtio_configure_intr(eth_dev) < 0) {
1580 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1581 			return -1;
1582 		}
1583 	}
1584 
1585 	vtpci_reinit_complete(hw);
1586 
1587 	if (pci_dev)
1588 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1589 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1590 			pci_dev->id.device_id);
1591 
1592 	return 0;
1593 }
1594 
1595 /*
1596  * Remap the PCI device again (IO port map for legacy device and
1597  * memory map for modern device), so that the secondary process
1598  * could have the PCI initiated correctly.
1599  */
1600 static int
1601 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1602 {
1603 	if (hw->modern) {
1604 		/*
1605 		 * We don't have to re-parse the PCI config space, since
1606 		 * rte_pci_map_device() makes sure the mapped address
1607 		 * in secondary process would equal to the one mapped in
1608 		 * the primary process: error will be returned if that
1609 		 * requirement is not met.
1610 		 *
1611 		 * That said, we could simply reuse all cap pointers
1612 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1613 		 * primary process, which is stored in shared memory.
1614 		 */
1615 		if (rte_pci_map_device(pci_dev)) {
1616 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1617 			return -1;
1618 		}
1619 	} else {
1620 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1621 			return -1;
1622 	}
1623 
1624 	return 0;
1625 }
1626 
1627 static void
1628 virtio_set_vtpci_ops(struct virtio_hw *hw)
1629 {
1630 #ifdef RTE_VIRTIO_USER
1631 	if (hw->virtio_user_dev)
1632 		VTPCI_OPS(hw) = &virtio_user_ops;
1633 	else
1634 #endif
1635 	if (hw->modern)
1636 		VTPCI_OPS(hw) = &modern_ops;
1637 	else
1638 		VTPCI_OPS(hw) = &legacy_ops;
1639 }
1640 
1641 /*
1642  * This function is based on probe() function in virtio_pci.c
1643  * It returns 0 on success.
1644  */
1645 int
1646 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1647 {
1648 	struct virtio_hw *hw = eth_dev->data->dev_private;
1649 	int ret;
1650 
1651 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1652 
1653 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1654 
1655 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1656 		if (!hw->virtio_user_dev) {
1657 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1658 			if (ret)
1659 				return ret;
1660 		}
1661 
1662 		virtio_set_vtpci_ops(hw);
1663 		set_rxtx_funcs(eth_dev);
1664 
1665 		return 0;
1666 	}
1667 
1668 	/* Allocate memory for storing MAC addresses */
1669 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1670 	if (eth_dev->data->mac_addrs == NULL) {
1671 		PMD_INIT_LOG(ERR,
1672 			"Failed to allocate %d bytes needed to store MAC addresses",
1673 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1674 		return -ENOMEM;
1675 	}
1676 
1677 	hw->port_id = eth_dev->data->port_id;
1678 	/* For virtio_user case the hw->virtio_user_dev is populated by
1679 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1680 	 */
1681 	if (!hw->virtio_user_dev) {
1682 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1683 		if (ret)
1684 			goto out;
1685 	}
1686 
1687 	/* reset device and negotiate default features */
1688 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1689 	if (ret < 0)
1690 		goto out;
1691 
1692 	return 0;
1693 
1694 out:
1695 	rte_free(eth_dev->data->mac_addrs);
1696 	return ret;
1697 }
1698 
1699 static int
1700 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1701 {
1702 	PMD_INIT_FUNC_TRACE();
1703 
1704 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1705 		return 0;
1706 
1707 	virtio_dev_stop(eth_dev);
1708 	virtio_dev_close(eth_dev);
1709 
1710 	eth_dev->dev_ops = NULL;
1711 	eth_dev->tx_pkt_burst = NULL;
1712 	eth_dev->rx_pkt_burst = NULL;
1713 
1714 	if (eth_dev->device)
1715 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1716 
1717 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1718 
1719 	return 0;
1720 }
1721 
1722 static int vdpa_check_handler(__rte_unused const char *key,
1723 		const char *value, __rte_unused void *opaque)
1724 {
1725 	if (strcmp(value, "1"))
1726 		return -1;
1727 
1728 	return 0;
1729 }
1730 
1731 static int
1732 vdpa_mode_selected(struct rte_devargs *devargs)
1733 {
1734 	struct rte_kvargs *kvlist;
1735 	const char *key = "vdpa";
1736 	int ret = 0;
1737 
1738 	if (devargs == NULL)
1739 		return 0;
1740 
1741 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1742 	if (kvlist == NULL)
1743 		return 0;
1744 
1745 	if (!rte_kvargs_count(kvlist, key))
1746 		goto exit;
1747 
1748 	/* vdpa mode selected when there's a key-value pair: vdpa=1 */
1749 	if (rte_kvargs_process(kvlist, key,
1750 				vdpa_check_handler, NULL) < 0) {
1751 		goto exit;
1752 	}
1753 	ret = 1;
1754 
1755 exit:
1756 	rte_kvargs_free(kvlist);
1757 	return ret;
1758 }
1759 
1760 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1761 	struct rte_pci_device *pci_dev)
1762 {
1763 	/* virtio pmd skips probe if device needs to work in vdpa mode */
1764 	if (vdpa_mode_selected(pci_dev->device.devargs))
1765 		return 1;
1766 
1767 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1768 		eth_virtio_dev_init);
1769 }
1770 
1771 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1772 {
1773 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1774 }
1775 
1776 static struct rte_pci_driver rte_virtio_pmd = {
1777 	.driver = {
1778 		.name = "net_virtio",
1779 	},
1780 	.id_table = pci_id_virtio_map,
1781 	.drv_flags = 0,
1782 	.probe = eth_virtio_pci_probe,
1783 	.remove = eth_virtio_pci_remove,
1784 };
1785 
1786 RTE_INIT(rte_virtio_pmd_init)
1787 {
1788 	if (rte_eal_iopl_init() != 0) {
1789 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1790 		return;
1791 	}
1792 
1793 	rte_pci_register(&rte_virtio_pmd);
1794 }
1795 
1796 static bool
1797 rx_offload_enabled(struct virtio_hw *hw)
1798 {
1799 	return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
1800 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1801 		vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
1802 }
1803 
1804 static bool
1805 tx_offload_enabled(struct virtio_hw *hw)
1806 {
1807 	return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
1808 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
1809 		vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
1810 }
1811 
1812 /*
1813  * Configure virtio device
1814  * It returns 0 on success.
1815  */
1816 static int
1817 virtio_dev_configure(struct rte_eth_dev *dev)
1818 {
1819 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1820 	const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
1821 	struct virtio_hw *hw = dev->data->dev_private;
1822 	uint64_t rx_offloads = rxmode->offloads;
1823 	uint64_t tx_offloads = txmode->offloads;
1824 	uint64_t req_features;
1825 	int ret;
1826 
1827 	PMD_INIT_LOG(DEBUG, "configure");
1828 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1829 
1830 	if (dev->data->dev_conf.intr_conf.rxq) {
1831 		ret = virtio_init_device(dev, hw->req_guest_features);
1832 		if (ret < 0)
1833 			return ret;
1834 	}
1835 
1836 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
1837 			   DEV_RX_OFFLOAD_TCP_CKSUM))
1838 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1839 
1840 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
1841 		req_features |=
1842 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1843 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1844 
1845 	if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
1846 			   DEV_TX_OFFLOAD_TCP_CKSUM))
1847 		req_features |= (1ULL << VIRTIO_NET_F_CSUM);
1848 
1849 	if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
1850 		req_features |=
1851 			(1ULL << VIRTIO_NET_F_HOST_TSO4) |
1852 			(1ULL << VIRTIO_NET_F_HOST_TSO6);
1853 
1854 	/* if request features changed, reinit the device */
1855 	if (req_features != hw->req_guest_features) {
1856 		ret = virtio_init_device(dev, req_features);
1857 		if (ret < 0)
1858 			return ret;
1859 	}
1860 
1861 	if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
1862 			    DEV_RX_OFFLOAD_TCP_CKSUM)) &&
1863 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1864 		PMD_DRV_LOG(ERR,
1865 			"rx checksum not available on this host");
1866 		return -ENOTSUP;
1867 	}
1868 
1869 	if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
1870 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1871 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
1872 		PMD_DRV_LOG(ERR,
1873 			"Large Receive Offload not available on this host");
1874 		return -ENOTSUP;
1875 	}
1876 
1877 	/* start control queue */
1878 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1879 		virtio_dev_cq_start(dev);
1880 
1881 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
1882 		hw->vlan_strip = 1;
1883 
1884 	if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1885 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1886 		PMD_DRV_LOG(ERR,
1887 			    "vlan filtering not available on this host");
1888 		return -ENOTSUP;
1889 	}
1890 
1891 	hw->has_tx_offload = tx_offload_enabled(hw);
1892 	hw->has_rx_offload = rx_offload_enabled(hw);
1893 
1894 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1895 		/* Enable vector (0) for Link State Intrerrupt */
1896 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1897 				VIRTIO_MSI_NO_VECTOR) {
1898 			PMD_DRV_LOG(ERR, "failed to set config vector");
1899 			return -EBUSY;
1900 		}
1901 
1902 	rte_spinlock_init(&hw->state_lock);
1903 
1904 	hw->use_simple_rx = 1;
1905 
1906 	if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
1907 		hw->use_inorder_tx = 1;
1908 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1909 			hw->use_inorder_rx = 1;
1910 			hw->use_simple_rx = 0;
1911 		} else {
1912 			hw->use_inorder_rx = 0;
1913 		}
1914 	}
1915 
1916 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
1917 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1918 		hw->use_simple_rx = 0;
1919 	}
1920 #endif
1921 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1922 		 hw->use_simple_rx = 0;
1923 	}
1924 
1925 	if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
1926 			   DEV_RX_OFFLOAD_TCP_CKSUM |
1927 			   DEV_RX_OFFLOAD_TCP_LRO |
1928 			   DEV_RX_OFFLOAD_VLAN_STRIP))
1929 		hw->use_simple_rx = 0;
1930 
1931 	hw->opened = true;
1932 
1933 	return 0;
1934 }
1935 
1936 
1937 static int
1938 virtio_dev_start(struct rte_eth_dev *dev)
1939 {
1940 	uint16_t nb_queues, i;
1941 	struct virtnet_rx *rxvq;
1942 	struct virtnet_tx *txvq __rte_unused;
1943 	struct virtio_hw *hw = dev->data->dev_private;
1944 	int ret;
1945 
1946 	/* Finish the initialization of the queues */
1947 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1948 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
1949 		if (ret < 0)
1950 			return ret;
1951 	}
1952 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1953 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
1954 		if (ret < 0)
1955 			return ret;
1956 	}
1957 
1958 	/* check if lsc interrupt feature is enabled */
1959 	if (dev->data->dev_conf.intr_conf.lsc) {
1960 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1961 			PMD_DRV_LOG(ERR, "link status not supported by host");
1962 			return -ENOTSUP;
1963 		}
1964 	}
1965 
1966 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1967 	 * in device configure, but it could be unmapped  when device is
1968 	 * stopped.
1969 	 */
1970 	if (dev->data->dev_conf.intr_conf.lsc ||
1971 	    dev->data->dev_conf.intr_conf.rxq) {
1972 		virtio_intr_disable(dev);
1973 
1974 		/* Setup interrupt callback  */
1975 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1976 			rte_intr_callback_register(dev->intr_handle,
1977 						   virtio_interrupt_handler,
1978 						   dev);
1979 
1980 		if (virtio_intr_enable(dev) < 0) {
1981 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1982 			return -EIO;
1983 		}
1984 	}
1985 
1986 	/*Notify the backend
1987 	 *Otherwise the tap backend might already stop its queue due to fullness.
1988 	 *vhost backend will have no chance to be waked up
1989 	 */
1990 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1991 	if (hw->max_queue_pairs > 1) {
1992 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1993 			return -EINVAL;
1994 	}
1995 
1996 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1997 
1998 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1999 		rxvq = dev->data->rx_queues[i];
2000 		/* Flush the old packets */
2001 		virtqueue_rxvq_flush(rxvq->vq);
2002 		virtqueue_notify(rxvq->vq);
2003 	}
2004 
2005 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2006 		txvq = dev->data->tx_queues[i];
2007 		virtqueue_notify(txvq->vq);
2008 	}
2009 
2010 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2011 
2012 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
2013 		rxvq = dev->data->rx_queues[i];
2014 		VIRTQUEUE_DUMP(rxvq->vq);
2015 	}
2016 
2017 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
2018 		txvq = dev->data->tx_queues[i];
2019 		VIRTQUEUE_DUMP(txvq->vq);
2020 	}
2021 
2022 	set_rxtx_funcs(dev);
2023 	hw->started = true;
2024 
2025 	/* Initialize Link state */
2026 	virtio_dev_link_update(dev, 0);
2027 
2028 	return 0;
2029 }
2030 
2031 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2032 {
2033 	struct virtio_hw *hw = dev->data->dev_private;
2034 	uint16_t nr_vq = virtio_get_nr_vq(hw);
2035 	const char *type __rte_unused;
2036 	unsigned int i, mbuf_num = 0;
2037 	struct virtqueue *vq;
2038 	struct rte_mbuf *buf;
2039 	int queue_type;
2040 
2041 	if (hw->vqs == NULL)
2042 		return;
2043 
2044 	for (i = 0; i < nr_vq; i++) {
2045 		vq = hw->vqs[i];
2046 		if (!vq)
2047 			continue;
2048 
2049 		queue_type = virtio_get_queue_type(hw, i);
2050 		if (queue_type == VTNET_RQ)
2051 			type = "rxq";
2052 		else if (queue_type == VTNET_TQ)
2053 			type = "txq";
2054 		else
2055 			continue;
2056 
2057 		PMD_INIT_LOG(DEBUG,
2058 			"Before freeing %s[%d] used and unused buf",
2059 			type, i);
2060 		VIRTQUEUE_DUMP(vq);
2061 
2062 		while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2063 			rte_pktmbuf_free(buf);
2064 			mbuf_num++;
2065 		}
2066 
2067 		PMD_INIT_LOG(DEBUG,
2068 			"After freeing %s[%d] used and unused buf",
2069 			type, i);
2070 		VIRTQUEUE_DUMP(vq);
2071 	}
2072 
2073 	PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2074 }
2075 
2076 /*
2077  * Stop device: disable interrupt and mark link down
2078  */
2079 static void
2080 virtio_dev_stop(struct rte_eth_dev *dev)
2081 {
2082 	struct virtio_hw *hw = dev->data->dev_private;
2083 	struct rte_eth_link link;
2084 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2085 
2086 	PMD_INIT_LOG(DEBUG, "stop");
2087 
2088 	rte_spinlock_lock(&hw->state_lock);
2089 	if (!hw->started)
2090 		goto out_unlock;
2091 	hw->started = false;
2092 
2093 	if (intr_conf->lsc || intr_conf->rxq) {
2094 		virtio_intr_disable(dev);
2095 
2096 		/* Reset interrupt callback  */
2097 		if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2098 			rte_intr_callback_unregister(dev->intr_handle,
2099 						     virtio_interrupt_handler,
2100 						     dev);
2101 		}
2102 	}
2103 
2104 	memset(&link, 0, sizeof(link));
2105 	rte_eth_linkstatus_set(dev, &link);
2106 out_unlock:
2107 	rte_spinlock_unlock(&hw->state_lock);
2108 }
2109 
2110 static int
2111 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2112 {
2113 	struct rte_eth_link link;
2114 	uint16_t status;
2115 	struct virtio_hw *hw = dev->data->dev_private;
2116 
2117 	memset(&link, 0, sizeof(link));
2118 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
2119 	link.link_speed  = ETH_SPEED_NUM_10G;
2120 	link.link_autoneg = ETH_LINK_FIXED;
2121 
2122 	if (!hw->started) {
2123 		link.link_status = ETH_LINK_DOWN;
2124 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2125 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
2126 		vtpci_read_dev_config(hw,
2127 				offsetof(struct virtio_net_config, status),
2128 				&status, sizeof(status));
2129 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2130 			link.link_status = ETH_LINK_DOWN;
2131 			PMD_INIT_LOG(DEBUG, "Port %d is down",
2132 				     dev->data->port_id);
2133 		} else {
2134 			link.link_status = ETH_LINK_UP;
2135 			PMD_INIT_LOG(DEBUG, "Port %d is up",
2136 				     dev->data->port_id);
2137 		}
2138 	} else {
2139 		link.link_status = ETH_LINK_UP;
2140 	}
2141 
2142 	return rte_eth_linkstatus_set(dev, &link);
2143 }
2144 
2145 static int
2146 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2147 {
2148 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2149 	struct virtio_hw *hw = dev->data->dev_private;
2150 	uint64_t offloads = rxmode->offloads;
2151 
2152 	if (mask & ETH_VLAN_FILTER_MASK) {
2153 		if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2154 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2155 
2156 			PMD_DRV_LOG(NOTICE,
2157 				"vlan filtering not available on this host");
2158 
2159 			return -ENOTSUP;
2160 		}
2161 	}
2162 
2163 	if (mask & ETH_VLAN_STRIP_MASK)
2164 		hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2165 
2166 	return 0;
2167 }
2168 
2169 static void
2170 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2171 {
2172 	uint64_t tso_mask, host_features;
2173 	struct virtio_hw *hw = dev->data->dev_private;
2174 
2175 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2176 
2177 	dev_info->max_rx_queues =
2178 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2179 	dev_info->max_tx_queues =
2180 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2181 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2182 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2183 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2184 
2185 	host_features = VTPCI_OPS(hw)->get_features(hw);
2186 	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2187 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2188 		dev_info->rx_offload_capa |=
2189 			DEV_RX_OFFLOAD_TCP_CKSUM |
2190 			DEV_RX_OFFLOAD_UDP_CKSUM;
2191 	}
2192 	if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2193 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2194 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2195 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2196 	if ((host_features & tso_mask) == tso_mask)
2197 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2198 
2199 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2200 				    DEV_TX_OFFLOAD_VLAN_INSERT;
2201 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2202 		dev_info->tx_offload_capa |=
2203 			DEV_TX_OFFLOAD_UDP_CKSUM |
2204 			DEV_TX_OFFLOAD_TCP_CKSUM;
2205 	}
2206 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2207 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2208 	if ((host_features & tso_mask) == tso_mask)
2209 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2210 }
2211 
2212 /*
2213  * It enables testpmd to collect per queue stats.
2214  */
2215 static int
2216 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2217 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2218 __rte_unused uint8_t is_rx)
2219 {
2220 	return 0;
2221 }
2222 
2223 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2224 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2225 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2226 
2227 RTE_INIT(virtio_init_log)
2228 {
2229 	virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2230 	if (virtio_logtype_init >= 0)
2231 		rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2232 	virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2233 	if (virtio_logtype_driver >= 0)
2234 		rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2235 }
2236