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_DEVICEID_MIN) },
107 	{ .vendor_id = 0, /* sentinel */ },
108 };
109 
110 struct rte_virtio_xstats_name_off {
111 	char name[RTE_ETH_XSTATS_NAME_SIZE];
112 	unsigned offset;
113 };
114 
115 /* [rt]x_qX_ is prepended to the name string here */
116 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
117 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
118 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
119 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
120 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
121 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
122 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
123 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
124 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
125 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
126 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
127 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
128 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
129 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
130 };
131 
132 /* [rt]x_qX_ is prepended to the name string here */
133 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
134 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
135 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
136 	{"errors",                 offsetof(struct virtnet_tx, stats.errors)},
137 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
138 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
139 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
140 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
141 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
142 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
143 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
144 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
145 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
146 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
147 };
148 
149 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
150 			    sizeof(rte_virtio_rxq_stat_strings[0]))
151 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
152 			    sizeof(rte_virtio_txq_stat_strings[0]))
153 
154 static int
155 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
156 		int *dlen, int pkt_num)
157 {
158 	uint32_t head, i;
159 	int k, sum = 0;
160 	virtio_net_ctrl_ack status = ~0;
161 	struct virtio_pmd_ctrl result;
162 	struct virtqueue *vq;
163 
164 	ctrl->status = status;
165 
166 	if (!cvq || !cvq->vq) {
167 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
168 		return -1;
169 	}
170 	vq = cvq->vq;
171 	head = vq->vq_desc_head_idx;
172 
173 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
174 		"vq->hw->cvq = %p vq = %p",
175 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
176 
177 	if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
178 		return -1;
179 
180 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
181 		sizeof(struct virtio_pmd_ctrl));
182 
183 	/*
184 	 * Format is enforced in qemu code:
185 	 * One TX packet for header;
186 	 * At least one TX packet per argument;
187 	 * One RX packet for ACK.
188 	 */
189 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
190 	vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
191 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
192 	vq->vq_free_cnt--;
193 	i = vq->vq_ring.desc[head].next;
194 
195 	for (k = 0; k < pkt_num; k++) {
196 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
197 		vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
198 			+ sizeof(struct virtio_net_ctrl_hdr)
199 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
200 		vq->vq_ring.desc[i].len = dlen[k];
201 		sum += dlen[k];
202 		vq->vq_free_cnt--;
203 		i = vq->vq_ring.desc[i].next;
204 	}
205 
206 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
207 	vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
208 			+ sizeof(struct virtio_net_ctrl_hdr);
209 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
210 	vq->vq_free_cnt--;
211 
212 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
213 
214 	vq_update_avail_ring(vq, head);
215 	vq_update_avail_idx(vq);
216 
217 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
218 
219 	virtqueue_notify(vq);
220 
221 	rte_rmb();
222 	while (VIRTQUEUE_NUSED(vq) == 0) {
223 		rte_rmb();
224 		usleep(100);
225 	}
226 
227 	while (VIRTQUEUE_NUSED(vq)) {
228 		uint32_t idx, desc_idx, used_idx;
229 		struct vring_used_elem *uep;
230 
231 		used_idx = (uint32_t)(vq->vq_used_cons_idx
232 				& (vq->vq_nentries - 1));
233 		uep = &vq->vq_ring.used->ring[used_idx];
234 		idx = (uint32_t) uep->id;
235 		desc_idx = idx;
236 
237 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
238 			desc_idx = vq->vq_ring.desc[desc_idx].next;
239 			vq->vq_free_cnt++;
240 		}
241 
242 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
243 		vq->vq_desc_head_idx = idx;
244 
245 		vq->vq_used_cons_idx++;
246 		vq->vq_free_cnt++;
247 	}
248 
249 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
250 			vq->vq_free_cnt, vq->vq_desc_head_idx);
251 
252 	memcpy(&result, cvq->virtio_net_hdr_mz->addr,
253 			sizeof(struct virtio_pmd_ctrl));
254 
255 	return result.status;
256 }
257 
258 static int
259 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
260 {
261 	struct virtio_hw *hw = dev->data->dev_private;
262 	struct virtio_pmd_ctrl ctrl;
263 	int dlen[1];
264 	int ret;
265 
266 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
267 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
268 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
269 
270 	dlen[0] = sizeof(uint16_t);
271 
272 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
273 	if (ret) {
274 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
275 			  "failed, this is too late now...");
276 		return -EINVAL;
277 	}
278 
279 	return 0;
280 }
281 
282 void
283 virtio_dev_queue_release(struct virtqueue *vq)
284 {
285 	struct virtio_hw *hw;
286 
287 	if (vq) {
288 		hw = vq->hw;
289 		if (vq->configured)
290 			hw->vtpci_ops->del_queue(hw, vq);
291 
292 		rte_free(vq->sw_ring);
293 		rte_free(vq);
294 	}
295 }
296 
297 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
298 			int queue_type,
299 			uint16_t queue_idx,
300 			uint16_t vtpci_queue_idx,
301 			uint16_t nb_desc,
302 			unsigned int socket_id,
303 			void **pvq)
304 {
305 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
306 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
307 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
308 	unsigned int vq_size, size;
309 	struct virtio_hw *hw = dev->data->dev_private;
310 	struct virtnet_rx *rxvq = NULL;
311 	struct virtnet_tx *txvq = NULL;
312 	struct virtnet_ctl *cvq = NULL;
313 	struct virtqueue *vq;
314 	const char *queue_names[] = {"rvq", "txq", "cvq"};
315 	size_t sz_vq, sz_q = 0, sz_hdr_mz = 0;
316 	void *sw_ring = NULL;
317 	int ret;
318 
319 	PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
320 
321 	/*
322 	 * Read the virtqueue size from the Queue Size field
323 	 * Always power of 2 and if 0 virtqueue does not exist
324 	 */
325 	vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx);
326 	PMD_INIT_LOG(DEBUG, "vq_size: %u nb_desc:%u", vq_size, nb_desc);
327 	if (vq_size == 0) {
328 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
329 		return -EINVAL;
330 	}
331 
332 	if (!rte_is_power_of_2(vq_size)) {
333 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
334 		return -EINVAL;
335 	}
336 
337 	snprintf(vq_name, sizeof(vq_name), "port%d_%s%d",
338 		 dev->data->port_id, queue_names[queue_type], queue_idx);
339 
340 	sz_vq = RTE_ALIGN_CEIL(sizeof(*vq) +
341 				vq_size * sizeof(struct vq_desc_extra),
342 				RTE_CACHE_LINE_SIZE);
343 	if (queue_type == VTNET_RQ) {
344 		sz_q = sz_vq + sizeof(*rxvq);
345 	} else if (queue_type == VTNET_TQ) {
346 		sz_q = sz_vq + sizeof(*txvq);
347 		/*
348 		 * For each xmit packet, allocate a virtio_net_hdr
349 		 * and indirect ring elements
350 		 */
351 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
352 	} else if (queue_type == VTNET_CQ) {
353 		sz_q = sz_vq + sizeof(*cvq);
354 		/* Allocate a page for control vq command, data and status */
355 		sz_hdr_mz = PAGE_SIZE;
356 	}
357 
358 	vq = rte_zmalloc_socket(vq_name, sz_q, RTE_CACHE_LINE_SIZE, socket_id);
359 	if (vq == NULL) {
360 		PMD_INIT_LOG(ERR, "can not allocate vq");
361 		return -ENOMEM;
362 	}
363 	vq->hw = hw;
364 	vq->vq_queue_index = vtpci_queue_idx;
365 	vq->vq_nentries = vq_size;
366 
367 	if (nb_desc == 0 || nb_desc > vq_size)
368 		nb_desc = vq_size;
369 	vq->vq_free_cnt = nb_desc;
370 
371 	/*
372 	 * Reserve a memzone for vring elements
373 	 */
374 	size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
375 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
376 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
377 		     size, vq->vq_ring_size);
378 
379 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size, socket_id,
380 					 0, VIRTIO_PCI_VRING_ALIGN);
381 	if (mz == NULL) {
382 		if (rte_errno == EEXIST)
383 			mz = rte_memzone_lookup(vq_name);
384 		if (mz == NULL) {
385 			ret = -ENOMEM;
386 			goto fail_q_alloc;
387 		}
388 	}
389 
390 	memset(mz->addr, 0, sizeof(mz->len));
391 
392 	vq->vq_ring_mem = mz->phys_addr;
393 	vq->vq_ring_virt_mem = mz->addr;
394 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
395 		     (uint64_t)mz->phys_addr);
396 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
397 		     (uint64_t)(uintptr_t)mz->addr);
398 
399 	if (sz_hdr_mz) {
400 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_%s%d_hdr",
401 			 dev->data->port_id, queue_names[queue_type],
402 			 queue_idx);
403 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
404 						     socket_id, 0,
405 						     RTE_CACHE_LINE_SIZE);
406 		if (hdr_mz == NULL) {
407 			if (rte_errno == EEXIST)
408 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
409 			if (hdr_mz == NULL) {
410 				ret = -ENOMEM;
411 				goto fail_q_alloc;
412 			}
413 		}
414 	}
415 
416 	if (queue_type == VTNET_RQ) {
417 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
418 			       sizeof(vq->sw_ring[0]);
419 
420 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
421 					     RTE_CACHE_LINE_SIZE, socket_id);
422 		if (!sw_ring) {
423 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
424 			ret = -ENOMEM;
425 			goto fail_q_alloc;
426 		}
427 
428 		vq->sw_ring = sw_ring;
429 		rxvq = (struct virtnet_rx *)RTE_PTR_ADD(vq, sz_vq);
430 		rxvq->vq = vq;
431 		rxvq->port_id = dev->data->port_id;
432 		rxvq->queue_id = queue_idx;
433 		rxvq->mz = mz;
434 		*pvq = rxvq;
435 	} else if (queue_type == VTNET_TQ) {
436 		txvq = (struct virtnet_tx *)RTE_PTR_ADD(vq, sz_vq);
437 		txvq->vq = vq;
438 		txvq->port_id = dev->data->port_id;
439 		txvq->queue_id = queue_idx;
440 		txvq->mz = mz;
441 		txvq->virtio_net_hdr_mz = hdr_mz;
442 		txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
443 
444 		*pvq = txvq;
445 	} else if (queue_type == VTNET_CQ) {
446 		cvq = (struct virtnet_ctl *)RTE_PTR_ADD(vq, sz_vq);
447 		cvq->vq = vq;
448 		cvq->mz = mz;
449 		cvq->virtio_net_hdr_mz = hdr_mz;
450 		cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
451 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
452 		*pvq = cvq;
453 	}
454 
455 	/* For virtio_user case (that is when dev->pci_dev is NULL), we use
456 	 * virtual address. And we need properly set _offset_, please see
457 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
458 	 */
459 	if (dev->pci_dev)
460 		vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
461 	else {
462 		vq->vq_ring_mem = (uintptr_t)mz->addr;
463 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
464 		if (queue_type == VTNET_TQ)
465 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
466 		else if (queue_type == VTNET_CQ)
467 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
468 	}
469 
470 	if (queue_type == VTNET_TQ) {
471 		struct virtio_tx_region *txr;
472 		unsigned int i;
473 
474 		txr = hdr_mz->addr;
475 		memset(txr, 0, vq_size * sizeof(*txr));
476 		for (i = 0; i < vq_size; i++) {
477 			struct vring_desc *start_dp = txr[i].tx_indir;
478 
479 			vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
480 
481 			/* first indirect descriptor is always the tx header */
482 			start_dp->addr = txvq->virtio_net_hdr_mem
483 				+ i * sizeof(*txr)
484 				+ offsetof(struct virtio_tx_region, tx_hdr);
485 
486 			start_dp->len = hw->vtnet_hdr_size;
487 			start_dp->flags = VRING_DESC_F_NEXT;
488 		}
489 	}
490 
491 	if (hw->vtpci_ops->setup_queue(hw, vq) < 0) {
492 		PMD_INIT_LOG(ERR, "setup_queue failed");
493 		virtio_dev_queue_release(vq);
494 		return -EINVAL;
495 	}
496 
497 	vq->configured = 1;
498 	return 0;
499 
500 fail_q_alloc:
501 	rte_free(sw_ring);
502 	rte_memzone_free(hdr_mz);
503 	rte_memzone_free(mz);
504 	rte_free(vq);
505 
506 	return ret;
507 }
508 
509 static int
510 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
511 		uint32_t socket_id)
512 {
513 	struct virtnet_ctl *cvq;
514 	int ret;
515 	struct virtio_hw *hw = dev->data->dev_private;
516 
517 	PMD_INIT_FUNC_TRACE();
518 	ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
519 			vtpci_queue_idx, 0, socket_id, (void **)&cvq);
520 	if (ret < 0) {
521 		PMD_INIT_LOG(ERR, "control vq initialization failed");
522 		return ret;
523 	}
524 
525 	hw->cvq = cvq;
526 	return 0;
527 }
528 
529 static void
530 virtio_free_queues(struct rte_eth_dev *dev)
531 {
532 	unsigned int i;
533 
534 	for (i = 0; i < dev->data->nb_rx_queues; i++)
535 		virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
536 
537 	dev->data->nb_rx_queues = 0;
538 
539 	for (i = 0; i < dev->data->nb_tx_queues; i++)
540 		virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
541 
542 	dev->data->nb_tx_queues = 0;
543 }
544 
545 static void
546 virtio_dev_close(struct rte_eth_dev *dev)
547 {
548 	struct virtio_hw *hw = dev->data->dev_private;
549 
550 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
551 
552 	/* reset the NIC */
553 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
554 		vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
555 	vtpci_reset(hw);
556 	hw->started = 0;
557 	virtio_dev_free_mbufs(dev);
558 	virtio_free_queues(dev);
559 }
560 
561 static void
562 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
563 {
564 	struct virtio_hw *hw = dev->data->dev_private;
565 	struct virtio_pmd_ctrl ctrl;
566 	int dlen[1];
567 	int ret;
568 
569 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
570 		PMD_INIT_LOG(INFO, "host does not support rx control\n");
571 		return;
572 	}
573 
574 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
575 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
576 	ctrl.data[0] = 1;
577 	dlen[0] = 1;
578 
579 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
580 	if (ret)
581 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
582 }
583 
584 static void
585 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
586 {
587 	struct virtio_hw *hw = dev->data->dev_private;
588 	struct virtio_pmd_ctrl ctrl;
589 	int dlen[1];
590 	int ret;
591 
592 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
593 		PMD_INIT_LOG(INFO, "host does not support rx control\n");
594 		return;
595 	}
596 
597 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
598 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
599 	ctrl.data[0] = 0;
600 	dlen[0] = 1;
601 
602 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
603 	if (ret)
604 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
605 }
606 
607 static void
608 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
609 {
610 	struct virtio_hw *hw = dev->data->dev_private;
611 	struct virtio_pmd_ctrl ctrl;
612 	int dlen[1];
613 	int ret;
614 
615 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
616 		PMD_INIT_LOG(INFO, "host does not support rx control\n");
617 		return;
618 	}
619 
620 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
621 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
622 	ctrl.data[0] = 1;
623 	dlen[0] = 1;
624 
625 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
626 	if (ret)
627 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
628 }
629 
630 static void
631 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
632 {
633 	struct virtio_hw *hw = dev->data->dev_private;
634 	struct virtio_pmd_ctrl ctrl;
635 	int dlen[1];
636 	int ret;
637 
638 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
639 		PMD_INIT_LOG(INFO, "host does not support rx control\n");
640 		return;
641 	}
642 
643 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
644 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
645 	ctrl.data[0] = 0;
646 	dlen[0] = 1;
647 
648 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
649 	if (ret)
650 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
651 }
652 
653 /*
654  * dev_ops for virtio, bare necessities for basic operation
655  */
656 static const struct eth_dev_ops virtio_eth_dev_ops = {
657 	.dev_configure           = virtio_dev_configure,
658 	.dev_start               = virtio_dev_start,
659 	.dev_stop                = virtio_dev_stop,
660 	.dev_close               = virtio_dev_close,
661 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
662 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
663 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
664 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
665 
666 	.dev_infos_get           = virtio_dev_info_get,
667 	.stats_get               = virtio_dev_stats_get,
668 	.xstats_get              = virtio_dev_xstats_get,
669 	.xstats_get_names        = virtio_dev_xstats_get_names,
670 	.stats_reset             = virtio_dev_stats_reset,
671 	.xstats_reset            = virtio_dev_stats_reset,
672 	.link_update             = virtio_dev_link_update,
673 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
674 	.rx_queue_release        = virtio_dev_rx_queue_release,
675 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
676 	.tx_queue_release        = virtio_dev_tx_queue_release,
677 	/* collect stats per queue */
678 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
679 	.vlan_filter_set         = virtio_vlan_filter_set,
680 	.mac_addr_add            = virtio_mac_addr_add,
681 	.mac_addr_remove         = virtio_mac_addr_remove,
682 	.mac_addr_set            = virtio_mac_addr_set,
683 };
684 
685 static inline int
686 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
687 				struct rte_eth_link *link)
688 {
689 	struct rte_eth_link *dst = link;
690 	struct rte_eth_link *src = &(dev->data->dev_link);
691 
692 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
693 			*(uint64_t *)src) == 0)
694 		return -1;
695 
696 	return 0;
697 }
698 
699 /**
700  * Atomically writes the link status information into global
701  * structure rte_eth_dev.
702  *
703  * @param dev
704  *   - Pointer to the structure rte_eth_dev to read from.
705  *   - Pointer to the buffer to be saved with the link status.
706  *
707  * @return
708  *   - On success, zero.
709  *   - On failure, negative value.
710  */
711 static inline int
712 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
713 		struct rte_eth_link *link)
714 {
715 	struct rte_eth_link *dst = &(dev->data->dev_link);
716 	struct rte_eth_link *src = link;
717 
718 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
719 					*(uint64_t *)src) == 0)
720 		return -1;
721 
722 	return 0;
723 }
724 
725 static void
726 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
727 {
728 	unsigned i;
729 
730 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
731 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
732 		if (txvq == NULL)
733 			continue;
734 
735 		stats->opackets += txvq->stats.packets;
736 		stats->obytes += txvq->stats.bytes;
737 		stats->oerrors += txvq->stats.errors;
738 
739 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
740 			stats->q_opackets[i] = txvq->stats.packets;
741 			stats->q_obytes[i] = txvq->stats.bytes;
742 		}
743 	}
744 
745 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
746 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
747 		if (rxvq == NULL)
748 			continue;
749 
750 		stats->ipackets += rxvq->stats.packets;
751 		stats->ibytes += rxvq->stats.bytes;
752 		stats->ierrors += rxvq->stats.errors;
753 
754 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
755 			stats->q_ipackets[i] = rxvq->stats.packets;
756 			stats->q_ibytes[i] = rxvq->stats.bytes;
757 		}
758 	}
759 
760 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
761 }
762 
763 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
764 				       struct rte_eth_xstat_name *xstats_names,
765 				       __rte_unused unsigned limit)
766 {
767 	unsigned i;
768 	unsigned count = 0;
769 	unsigned t;
770 
771 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
772 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
773 
774 	if (xstats_names != NULL) {
775 		/* Note: limit checked in rte_eth_xstats_names() */
776 
777 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
778 			struct virtqueue *rxvq = dev->data->rx_queues[i];
779 			if (rxvq == NULL)
780 				continue;
781 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
782 				snprintf(xstats_names[count].name,
783 					sizeof(xstats_names[count].name),
784 					"rx_q%u_%s", i,
785 					rte_virtio_rxq_stat_strings[t].name);
786 				count++;
787 			}
788 		}
789 
790 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
791 			struct virtqueue *txvq = dev->data->tx_queues[i];
792 			if (txvq == NULL)
793 				continue;
794 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
795 				snprintf(xstats_names[count].name,
796 					sizeof(xstats_names[count].name),
797 					"tx_q%u_%s", i,
798 					rte_virtio_txq_stat_strings[t].name);
799 				count++;
800 			}
801 		}
802 		return count;
803 	}
804 	return nstats;
805 }
806 
807 static int
808 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
809 		      unsigned n)
810 {
811 	unsigned i;
812 	unsigned count = 0;
813 
814 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
815 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
816 
817 	if (n < nstats)
818 		return nstats;
819 
820 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
821 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
822 
823 		if (rxvq == NULL)
824 			continue;
825 
826 		unsigned t;
827 
828 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
829 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
830 				rte_virtio_rxq_stat_strings[t].offset);
831 			count++;
832 		}
833 	}
834 
835 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
836 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
837 
838 		if (txvq == NULL)
839 			continue;
840 
841 		unsigned t;
842 
843 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
844 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
845 				rte_virtio_txq_stat_strings[t].offset);
846 			count++;
847 		}
848 	}
849 
850 	return count;
851 }
852 
853 static void
854 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
855 {
856 	virtio_update_stats(dev, stats);
857 }
858 
859 static void
860 virtio_dev_stats_reset(struct rte_eth_dev *dev)
861 {
862 	unsigned int i;
863 
864 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
865 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
866 		if (txvq == NULL)
867 			continue;
868 
869 		txvq->stats.packets = 0;
870 		txvq->stats.bytes = 0;
871 		txvq->stats.errors = 0;
872 		txvq->stats.multicast = 0;
873 		txvq->stats.broadcast = 0;
874 		memset(txvq->stats.size_bins, 0,
875 		       sizeof(txvq->stats.size_bins[0]) * 8);
876 	}
877 
878 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
879 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
880 		if (rxvq == NULL)
881 			continue;
882 
883 		rxvq->stats.packets = 0;
884 		rxvq->stats.bytes = 0;
885 		rxvq->stats.errors = 0;
886 		rxvq->stats.multicast = 0;
887 		rxvq->stats.broadcast = 0;
888 		memset(rxvq->stats.size_bins, 0,
889 		       sizeof(rxvq->stats.size_bins[0]) * 8);
890 	}
891 }
892 
893 static void
894 virtio_set_hwaddr(struct virtio_hw *hw)
895 {
896 	vtpci_write_dev_config(hw,
897 			offsetof(struct virtio_net_config, mac),
898 			&hw->mac_addr, ETHER_ADDR_LEN);
899 }
900 
901 static void
902 virtio_get_hwaddr(struct virtio_hw *hw)
903 {
904 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
905 		vtpci_read_dev_config(hw,
906 			offsetof(struct virtio_net_config, mac),
907 			&hw->mac_addr, ETHER_ADDR_LEN);
908 	} else {
909 		eth_random_addr(&hw->mac_addr[0]);
910 		virtio_set_hwaddr(hw);
911 	}
912 }
913 
914 static void
915 virtio_mac_table_set(struct virtio_hw *hw,
916 		     const struct virtio_net_ctrl_mac *uc,
917 		     const struct virtio_net_ctrl_mac *mc)
918 {
919 	struct virtio_pmd_ctrl ctrl;
920 	int err, len[2];
921 
922 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
923 		PMD_DRV_LOG(INFO, "host does not support mac table");
924 		return;
925 	}
926 
927 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
928 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
929 
930 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
931 	memcpy(ctrl.data, uc, len[0]);
932 
933 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
934 	memcpy(ctrl.data + len[0], mc, len[1]);
935 
936 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
937 	if (err != 0)
938 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
939 }
940 
941 static void
942 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
943 		    uint32_t index, uint32_t vmdq __rte_unused)
944 {
945 	struct virtio_hw *hw = dev->data->dev_private;
946 	const struct ether_addr *addrs = dev->data->mac_addrs;
947 	unsigned int i;
948 	struct virtio_net_ctrl_mac *uc, *mc;
949 
950 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
951 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
952 		return;
953 	}
954 
955 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
956 	uc->entries = 0;
957 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
958 	mc->entries = 0;
959 
960 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
961 		const struct ether_addr *addr
962 			= (i == index) ? mac_addr : addrs + i;
963 		struct virtio_net_ctrl_mac *tbl
964 			= is_multicast_ether_addr(addr) ? mc : uc;
965 
966 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
967 	}
968 
969 	virtio_mac_table_set(hw, uc, mc);
970 }
971 
972 static void
973 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
974 {
975 	struct virtio_hw *hw = dev->data->dev_private;
976 	struct ether_addr *addrs = dev->data->mac_addrs;
977 	struct virtio_net_ctrl_mac *uc, *mc;
978 	unsigned int i;
979 
980 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
981 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
982 		return;
983 	}
984 
985 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
986 	uc->entries = 0;
987 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
988 	mc->entries = 0;
989 
990 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
991 		struct virtio_net_ctrl_mac *tbl;
992 
993 		if (i == index || is_zero_ether_addr(addrs + i))
994 			continue;
995 
996 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
997 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
998 	}
999 
1000 	virtio_mac_table_set(hw, uc, mc);
1001 }
1002 
1003 static void
1004 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1005 {
1006 	struct virtio_hw *hw = dev->data->dev_private;
1007 
1008 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1009 
1010 	/* Use atomic update if available */
1011 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1012 		struct virtio_pmd_ctrl ctrl;
1013 		int len = ETHER_ADDR_LEN;
1014 
1015 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1016 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1017 
1018 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1019 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
1020 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1021 		virtio_set_hwaddr(hw);
1022 }
1023 
1024 static int
1025 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1026 {
1027 	struct virtio_hw *hw = dev->data->dev_private;
1028 	struct virtio_pmd_ctrl ctrl;
1029 	int len;
1030 
1031 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1032 		return -ENOTSUP;
1033 
1034 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1035 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1036 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1037 	len = sizeof(vlan_id);
1038 
1039 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1040 }
1041 
1042 static int
1043 virtio_negotiate_features(struct virtio_hw *hw)
1044 {
1045 	uint64_t host_features;
1046 
1047 	/* Prepare guest_features: feature that driver wants to support */
1048 	hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
1049 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1050 		hw->guest_features);
1051 
1052 	/* Read device(host) feature bits */
1053 	host_features = hw->vtpci_ops->get_features(hw);
1054 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1055 		host_features);
1056 
1057 	/*
1058 	 * Negotiate features: Subset of device feature bits are written back
1059 	 * guest feature bits.
1060 	 */
1061 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1062 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1063 		hw->guest_features);
1064 
1065 	if (hw->modern) {
1066 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1067 			PMD_INIT_LOG(ERR,
1068 				"VIRTIO_F_VERSION_1 features is not enabled.");
1069 			return -1;
1070 		}
1071 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1072 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1073 			PMD_INIT_LOG(ERR,
1074 				"failed to set FEATURES_OK status!");
1075 			return -1;
1076 		}
1077 	}
1078 
1079 	return 0;
1080 }
1081 
1082 /*
1083  * Process Virtio Config changed interrupt and call the callback
1084  * if link state changed.
1085  */
1086 static void
1087 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1088 			 void *param)
1089 {
1090 	struct rte_eth_dev *dev = param;
1091 	struct virtio_hw *hw = dev->data->dev_private;
1092 	uint8_t isr;
1093 
1094 	/* Read interrupt status which clears interrupt */
1095 	isr = vtpci_isr(hw);
1096 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1097 
1098 	if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1099 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1100 
1101 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1102 		if (virtio_dev_link_update(dev, 0) == 0)
1103 			_rte_eth_dev_callback_process(dev,
1104 						      RTE_ETH_EVENT_INTR_LSC);
1105 	}
1106 
1107 }
1108 
1109 static void
1110 rx_func_get(struct rte_eth_dev *eth_dev)
1111 {
1112 	struct virtio_hw *hw = eth_dev->data->dev_private;
1113 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1114 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1115 	else
1116 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1117 }
1118 
1119 /*
1120  * This function is based on probe() function in virtio_pci.c
1121  * It returns 0 on success.
1122  */
1123 int
1124 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1125 {
1126 	struct virtio_hw *hw = eth_dev->data->dev_private;
1127 	struct virtio_net_config *config;
1128 	struct virtio_net_config local_config;
1129 	struct rte_pci_device *pci_dev;
1130 	uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1131 	int ret;
1132 
1133 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1134 
1135 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1136 	eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1137 
1138 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1139 		rx_func_get(eth_dev);
1140 		return 0;
1141 	}
1142 
1143 	/* Allocate memory for storing MAC addresses */
1144 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1145 	if (eth_dev->data->mac_addrs == NULL) {
1146 		PMD_INIT_LOG(ERR,
1147 			"Failed to allocate %d bytes needed to store MAC addresses",
1148 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1149 		return -ENOMEM;
1150 	}
1151 
1152 	pci_dev = eth_dev->pci_dev;
1153 
1154 	if (pci_dev) {
1155 		ret = vtpci_init(pci_dev, hw, &dev_flags);
1156 		if (ret)
1157 			return ret;
1158 	}
1159 
1160 	/* Reset the device although not necessary at startup */
1161 	vtpci_reset(hw);
1162 
1163 	/* Tell the host we've noticed this device. */
1164 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1165 
1166 	/* Tell the host we've known how to drive the device. */
1167 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1168 	if (virtio_negotiate_features(hw) < 0)
1169 		return -1;
1170 
1171 	/* If host does not support status then disable LSC */
1172 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1173 		dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1174 
1175 	rte_eth_copy_pci_info(eth_dev, pci_dev);
1176 	eth_dev->data->dev_flags = dev_flags;
1177 
1178 	rx_func_get(eth_dev);
1179 
1180 	/* Setting up rx_header size for the device */
1181 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1182 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1183 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1184 	else
1185 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1186 
1187 	/* Copy the permanent MAC address to: virtio_hw */
1188 	virtio_get_hwaddr(hw);
1189 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1190 			&eth_dev->data->mac_addrs[0]);
1191 	PMD_INIT_LOG(DEBUG,
1192 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1193 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1194 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1195 
1196 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1197 		config = &local_config;
1198 
1199 		vtpci_read_dev_config(hw,
1200 			offsetof(struct virtio_net_config, mac),
1201 			&config->mac, sizeof(config->mac));
1202 
1203 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1204 			vtpci_read_dev_config(hw,
1205 				offsetof(struct virtio_net_config, status),
1206 				&config->status, sizeof(config->status));
1207 		} else {
1208 			PMD_INIT_LOG(DEBUG,
1209 				     "VIRTIO_NET_F_STATUS is not supported");
1210 			config->status = 0;
1211 		}
1212 
1213 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1214 			vtpci_read_dev_config(hw,
1215 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1216 				&config->max_virtqueue_pairs,
1217 				sizeof(config->max_virtqueue_pairs));
1218 		} else {
1219 			PMD_INIT_LOG(DEBUG,
1220 				     "VIRTIO_NET_F_MQ is not supported");
1221 			config->max_virtqueue_pairs = 1;
1222 		}
1223 
1224 		hw->max_rx_queues =
1225 			(VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1226 			VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1227 		hw->max_tx_queues =
1228 			(VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1229 			VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1230 
1231 		virtio_dev_cq_queue_setup(eth_dev,
1232 					config->max_virtqueue_pairs * 2,
1233 					SOCKET_ID_ANY);
1234 
1235 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1236 				config->max_virtqueue_pairs);
1237 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1238 		PMD_INIT_LOG(DEBUG,
1239 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1240 				config->mac[0], config->mac[1],
1241 				config->mac[2], config->mac[3],
1242 				config->mac[4], config->mac[5]);
1243 	} else {
1244 		hw->max_rx_queues = 1;
1245 		hw->max_tx_queues = 1;
1246 	}
1247 
1248 	PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1249 			hw->max_rx_queues, hw->max_tx_queues);
1250 	if (pci_dev)
1251 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1252 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1253 			pci_dev->id.device_id);
1254 
1255 	/* Setup interrupt callback  */
1256 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1257 		rte_intr_callback_register(&pci_dev->intr_handle,
1258 				   virtio_interrupt_handler, eth_dev);
1259 
1260 	virtio_dev_cq_start(eth_dev);
1261 
1262 	return 0;
1263 }
1264 
1265 static int
1266 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1267 {
1268 	struct rte_pci_device *pci_dev;
1269 	struct virtio_hw *hw = eth_dev->data->dev_private;
1270 
1271 	PMD_INIT_FUNC_TRACE();
1272 
1273 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1274 		return -EPERM;
1275 
1276 	if (hw->started == 1) {
1277 		virtio_dev_stop(eth_dev);
1278 		virtio_dev_close(eth_dev);
1279 	}
1280 	pci_dev = eth_dev->pci_dev;
1281 
1282 	eth_dev->dev_ops = NULL;
1283 	eth_dev->tx_pkt_burst = NULL;
1284 	eth_dev->rx_pkt_burst = NULL;
1285 
1286 	if (hw->cvq)
1287 		virtio_dev_queue_release(hw->cvq->vq);
1288 
1289 	rte_free(eth_dev->data->mac_addrs);
1290 	eth_dev->data->mac_addrs = NULL;
1291 
1292 	/* reset interrupt callback  */
1293 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1294 		rte_intr_callback_unregister(&pci_dev->intr_handle,
1295 						virtio_interrupt_handler,
1296 						eth_dev);
1297 	rte_eal_pci_unmap_device(pci_dev);
1298 
1299 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1300 
1301 	return 0;
1302 }
1303 
1304 static struct eth_driver rte_virtio_pmd = {
1305 	.pci_drv = {
1306 		.name = "rte_virtio_pmd",
1307 		.id_table = pci_id_virtio_map,
1308 		.drv_flags = RTE_PCI_DRV_DETACHABLE,
1309 	},
1310 	.eth_dev_init = eth_virtio_dev_init,
1311 	.eth_dev_uninit = eth_virtio_dev_uninit,
1312 	.dev_private_size = sizeof(struct virtio_hw),
1313 };
1314 
1315 /*
1316  * Driver initialization routine.
1317  * Invoked once at EAL init time.
1318  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1319  * Returns 0 on success.
1320  */
1321 static int
1322 rte_virtio_pmd_init(const char *name __rte_unused,
1323 		    const char *param __rte_unused)
1324 {
1325 	if (rte_eal_iopl_init() != 0) {
1326 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1327 		return -1;
1328 	}
1329 
1330 	rte_eth_driver_register(&rte_virtio_pmd);
1331 	return 0;
1332 }
1333 
1334 /*
1335  * Configure virtio device
1336  * It returns 0 on success.
1337  */
1338 static int
1339 virtio_dev_configure(struct rte_eth_dev *dev)
1340 {
1341 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1342 	struct virtio_hw *hw = dev->data->dev_private;
1343 
1344 	PMD_INIT_LOG(DEBUG, "configure");
1345 
1346 	if (rxmode->hw_ip_checksum) {
1347 		PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1348 		return -EINVAL;
1349 	}
1350 
1351 	hw->vlan_strip = rxmode->hw_vlan_strip;
1352 
1353 	if (rxmode->hw_vlan_filter
1354 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1355 		PMD_DRV_LOG(NOTICE,
1356 			    "vlan filtering not available on this host");
1357 		return -ENOTSUP;
1358 	}
1359 
1360 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1361 		if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1362 			PMD_DRV_LOG(ERR, "failed to set config vector");
1363 			return -EBUSY;
1364 		}
1365 
1366 	return 0;
1367 }
1368 
1369 
1370 static int
1371 virtio_dev_start(struct rte_eth_dev *dev)
1372 {
1373 	uint16_t nb_queues, i;
1374 	struct virtio_hw *hw = dev->data->dev_private;
1375 	struct virtnet_rx *rxvq;
1376 	struct virtnet_tx *txvq __rte_unused;
1377 
1378 	/* check if lsc interrupt feature is enabled */
1379 	if (dev->data->dev_conf.intr_conf.lsc) {
1380 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1381 			PMD_DRV_LOG(ERR, "link status not supported by host");
1382 			return -ENOTSUP;
1383 		}
1384 
1385 		if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1386 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1387 			return -EIO;
1388 		}
1389 	}
1390 
1391 	/* Initialize Link state */
1392 	virtio_dev_link_update(dev, 0);
1393 
1394 	/* On restart after stop do not touch queues */
1395 	if (hw->started)
1396 		return 0;
1397 
1398 	/* Do final configuration before rx/tx engine starts */
1399 	virtio_dev_rxtx_start(dev);
1400 	vtpci_reinit_complete(hw);
1401 
1402 	hw->started = 1;
1403 
1404 	/*Notify the backend
1405 	 *Otherwise the tap backend might already stop its queue due to fullness.
1406 	 *vhost backend will have no chance to be waked up
1407 	 */
1408 	nb_queues = dev->data->nb_rx_queues;
1409 	if (nb_queues > 1) {
1410 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1411 			return -EINVAL;
1412 	}
1413 
1414 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1415 
1416 	for (i = 0; i < nb_queues; i++) {
1417 		rxvq = dev->data->rx_queues[i];
1418 		virtqueue_notify(rxvq->vq);
1419 	}
1420 
1421 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1422 
1423 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1424 		rxvq = dev->data->rx_queues[i];
1425 		VIRTQUEUE_DUMP(rxvq->vq);
1426 	}
1427 
1428 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1429 		txvq = dev->data->tx_queues[i];
1430 		VIRTQUEUE_DUMP(txvq->vq);
1431 	}
1432 
1433 	return 0;
1434 }
1435 
1436 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1437 {
1438 	struct rte_mbuf *buf;
1439 	int i, mbuf_num = 0;
1440 
1441 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1442 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1443 
1444 		PMD_INIT_LOG(DEBUG,
1445 			     "Before freeing rxq[%d] used and unused buf", i);
1446 		VIRTQUEUE_DUMP(rxvq->vq);
1447 
1448 		PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1449 		while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1450 			rte_pktmbuf_free(buf);
1451 			mbuf_num++;
1452 		}
1453 
1454 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1455 		PMD_INIT_LOG(DEBUG,
1456 			     "After freeing rxq[%d] used and unused buf", i);
1457 		VIRTQUEUE_DUMP(rxvq->vq);
1458 	}
1459 
1460 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1461 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1462 
1463 		PMD_INIT_LOG(DEBUG,
1464 			     "Before freeing txq[%d] used and unused bufs",
1465 			     i);
1466 		VIRTQUEUE_DUMP(txvq->vq);
1467 
1468 		mbuf_num = 0;
1469 		while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1470 			rte_pktmbuf_free(buf);
1471 			mbuf_num++;
1472 		}
1473 
1474 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1475 		PMD_INIT_LOG(DEBUG,
1476 			     "After freeing txq[%d] used and unused buf", i);
1477 		VIRTQUEUE_DUMP(txvq->vq);
1478 	}
1479 }
1480 
1481 /*
1482  * Stop device: disable interrupt and mark link down
1483  */
1484 static void
1485 virtio_dev_stop(struct rte_eth_dev *dev)
1486 {
1487 	struct rte_eth_link link;
1488 
1489 	PMD_INIT_LOG(DEBUG, "stop");
1490 
1491 	if (dev->data->dev_conf.intr_conf.lsc)
1492 		rte_intr_disable(&dev->pci_dev->intr_handle);
1493 
1494 	memset(&link, 0, sizeof(link));
1495 	virtio_dev_atomic_write_link_status(dev, &link);
1496 }
1497 
1498 static int
1499 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1500 {
1501 	struct rte_eth_link link, old;
1502 	uint16_t status;
1503 	struct virtio_hw *hw = dev->data->dev_private;
1504 	memset(&link, 0, sizeof(link));
1505 	virtio_dev_atomic_read_link_status(dev, &link);
1506 	old = link;
1507 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1508 	link.link_speed  = SPEED_10G;
1509 
1510 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1511 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
1512 		vtpci_read_dev_config(hw,
1513 				offsetof(struct virtio_net_config, status),
1514 				&status, sizeof(status));
1515 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1516 			link.link_status = ETH_LINK_DOWN;
1517 			PMD_INIT_LOG(DEBUG, "Port %d is down",
1518 				     dev->data->port_id);
1519 		} else {
1520 			link.link_status = ETH_LINK_UP;
1521 			PMD_INIT_LOG(DEBUG, "Port %d is up",
1522 				     dev->data->port_id);
1523 		}
1524 	} else {
1525 		link.link_status = ETH_LINK_UP;
1526 	}
1527 	virtio_dev_atomic_write_link_status(dev, &link);
1528 
1529 	return (old.link_status == link.link_status) ? -1 : 0;
1530 }
1531 
1532 static void
1533 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1534 {
1535 	struct virtio_hw *hw = dev->data->dev_private;
1536 
1537 	if (dev->pci_dev)
1538 		dev_info->driver_name = dev->driver->pci_drv.name;
1539 	else
1540 		dev_info->driver_name = "virtio_user PMD";
1541 	dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1542 	dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1543 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1544 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1545 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1546 	dev_info->default_txconf = (struct rte_eth_txconf) {
1547 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1548 	};
1549 }
1550 
1551 /*
1552  * It enables testpmd to collect per queue stats.
1553  */
1554 static int
1555 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1556 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1557 __rte_unused uint8_t is_rx)
1558 {
1559 	return 0;
1560 }
1561 
1562 static struct rte_driver rte_virtio_driver = {
1563 	.type = PMD_PDEV,
1564 	.init = rte_virtio_pmd_init,
1565 };
1566 
1567 PMD_REGISTER_DRIVER(rte_virtio_driver, virtio_net);
1568 DRIVER_REGISTER_PCI_TABLE(virtio_net, pci_id_virtio_map);
1569