1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <stdint.h>
35 #include <string.h>
36 #include <stdio.h>
37 #include <errno.h>
38 #include <unistd.h>
39 
40 #include <rte_ethdev.h>
41 #include <rte_ethdev_pci.h>
42 #include <rte_memcpy.h>
43 #include <rte_string_fns.h>
44 #include <rte_memzone.h>
45 #include <rte_malloc.h>
46 #include <rte_atomic.h>
47 #include <rte_branch_prediction.h>
48 #include <rte_pci.h>
49 #include <rte_bus_pci.h>
50 #include <rte_ether.h>
51 #include <rte_common.h>
52 #include <rte_errno.h>
53 #include <rte_cpuflags.h>
54 
55 #include <rte_memory.h>
56 #include <rte_eal.h>
57 #include <rte_dev.h>
58 
59 #include "virtio_ethdev.h"
60 #include "virtio_pci.h"
61 #include "virtio_logs.h"
62 #include "virtqueue.h"
63 #include "virtio_rxtx.h"
64 
65 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
66 static int  virtio_dev_configure(struct rte_eth_dev *dev);
67 static int  virtio_dev_start(struct rte_eth_dev *dev);
68 static void virtio_dev_stop(struct rte_eth_dev *dev);
69 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
70 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
71 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
72 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
73 static void virtio_dev_info_get(struct rte_eth_dev *dev,
74 				struct rte_eth_dev_info *dev_info);
75 static int virtio_dev_link_update(struct rte_eth_dev *dev,
76 	int wait_to_complete);
77 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
78 
79 static void virtio_set_hwaddr(struct virtio_hw *hw);
80 static void virtio_get_hwaddr(struct virtio_hw *hw);
81 
82 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
83 				 struct rte_eth_stats *stats);
84 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
85 				 struct rte_eth_xstat *xstats, unsigned n);
86 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
87 				       struct rte_eth_xstat_name *xstats_names,
88 				       unsigned limit);
89 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
90 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
91 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
92 				uint16_t vlan_id, int on);
93 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
94 				struct ether_addr *mac_addr,
95 				uint32_t index, uint32_t vmdq);
96 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
97 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
98 				struct ether_addr *mac_addr);
99 
100 static int virtio_intr_enable(struct rte_eth_dev *dev);
101 static int virtio_intr_disable(struct rte_eth_dev *dev);
102 
103 static int virtio_dev_queue_stats_mapping_set(
104 	struct rte_eth_dev *eth_dev,
105 	uint16_t queue_id,
106 	uint8_t stat_idx,
107 	uint8_t is_rx);
108 
109 /*
110  * The set of PCI devices this driver supports
111  */
112 static const struct rte_pci_id pci_id_virtio_map[] = {
113 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
114 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
115 	{ .vendor_id = 0, /* sentinel */ },
116 };
117 
118 struct rte_virtio_xstats_name_off {
119 	char name[RTE_ETH_XSTATS_NAME_SIZE];
120 	unsigned offset;
121 };
122 
123 /* [rt]x_qX_ is prepended to the name string here */
124 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
125 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
126 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
127 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
128 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
129 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
130 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
131 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
132 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
133 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
134 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
135 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
136 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
137 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
138 };
139 
140 /* [rt]x_qX_ is prepended to the name string here */
141 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
142 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
143 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
144 	{"errors",                 offsetof(struct virtnet_tx, stats.errors)},
145 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
146 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
147 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
148 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
149 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
150 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
151 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
152 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
153 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
154 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
155 };
156 
157 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
158 			    sizeof(rte_virtio_rxq_stat_strings[0]))
159 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
160 			    sizeof(rte_virtio_txq_stat_strings[0]))
161 
162 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
163 
164 static int
165 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
166 		int *dlen, int pkt_num)
167 {
168 	uint32_t head, i;
169 	int k, sum = 0;
170 	virtio_net_ctrl_ack status = ~0;
171 	struct virtio_pmd_ctrl *result;
172 	struct virtqueue *vq;
173 
174 	ctrl->status = status;
175 
176 	if (!cvq || !cvq->vq) {
177 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
178 		return -1;
179 	}
180 	vq = cvq->vq;
181 	head = vq->vq_desc_head_idx;
182 
183 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
184 		"vq->hw->cvq = %p vq = %p",
185 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
186 
187 	if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
188 		return -1;
189 
190 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
191 		sizeof(struct virtio_pmd_ctrl));
192 
193 	/*
194 	 * Format is enforced in qemu code:
195 	 * One TX packet for header;
196 	 * At least one TX packet per argument;
197 	 * One RX packet for ACK.
198 	 */
199 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
200 	vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
201 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
202 	vq->vq_free_cnt--;
203 	i = vq->vq_ring.desc[head].next;
204 
205 	for (k = 0; k < pkt_num; k++) {
206 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
207 		vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
208 			+ sizeof(struct virtio_net_ctrl_hdr)
209 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
210 		vq->vq_ring.desc[i].len = dlen[k];
211 		sum += dlen[k];
212 		vq->vq_free_cnt--;
213 		i = vq->vq_ring.desc[i].next;
214 	}
215 
216 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
217 	vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
218 			+ sizeof(struct virtio_net_ctrl_hdr);
219 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
220 	vq->vq_free_cnt--;
221 
222 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
223 
224 	vq_update_avail_ring(vq, head);
225 	vq_update_avail_idx(vq);
226 
227 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
228 
229 	virtqueue_notify(vq);
230 
231 	rte_rmb();
232 	while (VIRTQUEUE_NUSED(vq) == 0) {
233 		rte_rmb();
234 		usleep(100);
235 	}
236 
237 	while (VIRTQUEUE_NUSED(vq)) {
238 		uint32_t idx, desc_idx, used_idx;
239 		struct vring_used_elem *uep;
240 
241 		used_idx = (uint32_t)(vq->vq_used_cons_idx
242 				& (vq->vq_nentries - 1));
243 		uep = &vq->vq_ring.used->ring[used_idx];
244 		idx = (uint32_t) uep->id;
245 		desc_idx = idx;
246 
247 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
248 			desc_idx = vq->vq_ring.desc[desc_idx].next;
249 			vq->vq_free_cnt++;
250 		}
251 
252 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
253 		vq->vq_desc_head_idx = idx;
254 
255 		vq->vq_used_cons_idx++;
256 		vq->vq_free_cnt++;
257 	}
258 
259 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
260 			vq->vq_free_cnt, vq->vq_desc_head_idx);
261 
262 	result = cvq->virtio_net_hdr_mz->addr;
263 
264 	return result->status;
265 }
266 
267 static int
268 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
269 {
270 	struct virtio_hw *hw = dev->data->dev_private;
271 	struct virtio_pmd_ctrl ctrl;
272 	int dlen[1];
273 	int ret;
274 
275 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
276 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
277 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
278 
279 	dlen[0] = sizeof(uint16_t);
280 
281 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
282 	if (ret) {
283 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
284 			  "failed, this is too late now...");
285 		return -EINVAL;
286 	}
287 
288 	return 0;
289 }
290 
291 static void
292 virtio_dev_queue_release(void *queue __rte_unused)
293 {
294 	/* do nothing */
295 }
296 
297 static uint16_t
298 virtio_get_nr_vq(struct virtio_hw *hw)
299 {
300 	uint16_t nr_vq = hw->max_queue_pairs * 2;
301 
302 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
303 		nr_vq += 1;
304 
305 	return nr_vq;
306 }
307 
308 static void
309 virtio_init_vring(struct virtqueue *vq)
310 {
311 	int size = vq->vq_nentries;
312 	struct vring *vr = &vq->vq_ring;
313 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
314 
315 	PMD_INIT_FUNC_TRACE();
316 
317 	/*
318 	 * Reinitialise since virtio port might have been stopped and restarted
319 	 */
320 	memset(ring_mem, 0, vq->vq_ring_size);
321 	vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
322 	vq->vq_used_cons_idx = 0;
323 	vq->vq_desc_head_idx = 0;
324 	vq->vq_avail_idx = 0;
325 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
326 	vq->vq_free_cnt = vq->vq_nentries;
327 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
328 
329 	vring_desc_init(vr->desc, size);
330 
331 	/*
332 	 * Disable device(host) interrupting guest
333 	 */
334 	virtqueue_disable_intr(vq);
335 }
336 
337 static int
338 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
339 {
340 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
341 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
342 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
343 	unsigned int vq_size, size;
344 	struct virtio_hw *hw = dev->data->dev_private;
345 	struct virtnet_rx *rxvq = NULL;
346 	struct virtnet_tx *txvq = NULL;
347 	struct virtnet_ctl *cvq = NULL;
348 	struct virtqueue *vq;
349 	size_t sz_hdr_mz = 0;
350 	void *sw_ring = NULL;
351 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
352 	int ret;
353 
354 	PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
355 
356 	/*
357 	 * Read the virtqueue size from the Queue Size field
358 	 * Always power of 2 and if 0 virtqueue does not exist
359 	 */
360 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
361 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
362 	if (vq_size == 0) {
363 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
364 		return -EINVAL;
365 	}
366 
367 	if (!rte_is_power_of_2(vq_size)) {
368 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
369 		return -EINVAL;
370 	}
371 
372 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
373 		 dev->data->port_id, vtpci_queue_idx);
374 
375 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
376 				vq_size * sizeof(struct vq_desc_extra),
377 				RTE_CACHE_LINE_SIZE);
378 	if (queue_type == VTNET_TQ) {
379 		/*
380 		 * For each xmit packet, allocate a virtio_net_hdr
381 		 * and indirect ring elements
382 		 */
383 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
384 	} else if (queue_type == VTNET_CQ) {
385 		/* Allocate a page for control vq command, data and status */
386 		sz_hdr_mz = PAGE_SIZE;
387 	}
388 
389 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
390 				SOCKET_ID_ANY);
391 	if (vq == NULL) {
392 		PMD_INIT_LOG(ERR, "can not allocate vq");
393 		return -ENOMEM;
394 	}
395 	hw->vqs[vtpci_queue_idx] = vq;
396 
397 	vq->hw = hw;
398 	vq->vq_queue_index = vtpci_queue_idx;
399 	vq->vq_nentries = vq_size;
400 
401 	/*
402 	 * Reserve a memzone for vring elements
403 	 */
404 	size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
405 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
406 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
407 		     size, vq->vq_ring_size);
408 
409 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
410 					 SOCKET_ID_ANY,
411 					 0, VIRTIO_PCI_VRING_ALIGN);
412 	if (mz == NULL) {
413 		if (rte_errno == EEXIST)
414 			mz = rte_memzone_lookup(vq_name);
415 		if (mz == NULL) {
416 			ret = -ENOMEM;
417 			goto fail_q_alloc;
418 		}
419 	}
420 
421 	memset(mz->addr, 0, mz->len);
422 
423 	vq->vq_ring_mem = mz->iova;
424 	vq->vq_ring_virt_mem = mz->addr;
425 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
426 		     (uint64_t)mz->iova);
427 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
428 		     (uint64_t)(uintptr_t)mz->addr);
429 
430 	virtio_init_vring(vq);
431 
432 	if (sz_hdr_mz) {
433 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
434 			 dev->data->port_id, vtpci_queue_idx);
435 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
436 						     SOCKET_ID_ANY, 0,
437 						     RTE_CACHE_LINE_SIZE);
438 		if (hdr_mz == NULL) {
439 			if (rte_errno == EEXIST)
440 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
441 			if (hdr_mz == NULL) {
442 				ret = -ENOMEM;
443 				goto fail_q_alloc;
444 			}
445 		}
446 	}
447 
448 	if (queue_type == VTNET_RQ) {
449 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
450 			       sizeof(vq->sw_ring[0]);
451 
452 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
453 				RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
454 		if (!sw_ring) {
455 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
456 			ret = -ENOMEM;
457 			goto fail_q_alloc;
458 		}
459 
460 		vq->sw_ring = sw_ring;
461 		rxvq = &vq->rxq;
462 		rxvq->vq = vq;
463 		rxvq->port_id = dev->data->port_id;
464 		rxvq->mz = mz;
465 	} else if (queue_type == VTNET_TQ) {
466 		txvq = &vq->txq;
467 		txvq->vq = vq;
468 		txvq->port_id = dev->data->port_id;
469 		txvq->mz = mz;
470 		txvq->virtio_net_hdr_mz = hdr_mz;
471 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
472 	} else if (queue_type == VTNET_CQ) {
473 		cvq = &vq->cq;
474 		cvq->vq = vq;
475 		cvq->mz = mz;
476 		cvq->virtio_net_hdr_mz = hdr_mz;
477 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
478 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
479 
480 		hw->cvq = cvq;
481 	}
482 
483 	/* For virtio_user case (that is when hw->dev is NULL), we use
484 	 * virtual address. And we need properly set _offset_, please see
485 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
486 	 */
487 	if (!hw->virtio_user_dev)
488 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
489 	else {
490 		vq->vq_ring_mem = (uintptr_t)mz->addr;
491 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
492 		if (queue_type == VTNET_TQ)
493 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
494 		else if (queue_type == VTNET_CQ)
495 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
496 	}
497 
498 	if (queue_type == VTNET_TQ) {
499 		struct virtio_tx_region *txr;
500 		unsigned int i;
501 
502 		txr = hdr_mz->addr;
503 		memset(txr, 0, vq_size * sizeof(*txr));
504 		for (i = 0; i < vq_size; i++) {
505 			struct vring_desc *start_dp = txr[i].tx_indir;
506 
507 			vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
508 
509 			/* first indirect descriptor is always the tx header */
510 			start_dp->addr = txvq->virtio_net_hdr_mem
511 				+ i * sizeof(*txr)
512 				+ offsetof(struct virtio_tx_region, tx_hdr);
513 
514 			start_dp->len = hw->vtnet_hdr_size;
515 			start_dp->flags = VRING_DESC_F_NEXT;
516 		}
517 	}
518 
519 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
520 		PMD_INIT_LOG(ERR, "setup_queue failed");
521 		return -EINVAL;
522 	}
523 
524 	return 0;
525 
526 fail_q_alloc:
527 	rte_free(sw_ring);
528 	rte_memzone_free(hdr_mz);
529 	rte_memzone_free(mz);
530 	rte_free(vq);
531 
532 	return ret;
533 }
534 
535 static void
536 virtio_free_queues(struct virtio_hw *hw)
537 {
538 	uint16_t nr_vq = virtio_get_nr_vq(hw);
539 	struct virtqueue *vq;
540 	int queue_type;
541 	uint16_t i;
542 
543 	if (hw->vqs == NULL)
544 		return;
545 
546 	for (i = 0; i < nr_vq; i++) {
547 		vq = hw->vqs[i];
548 		if (!vq)
549 			continue;
550 
551 		queue_type = virtio_get_queue_type(hw, i);
552 		if (queue_type == VTNET_RQ) {
553 			rte_free(vq->sw_ring);
554 			rte_memzone_free(vq->rxq.mz);
555 		} else if (queue_type == VTNET_TQ) {
556 			rte_memzone_free(vq->txq.mz);
557 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
558 		} else {
559 			rte_memzone_free(vq->cq.mz);
560 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
561 		}
562 
563 		rte_free(vq);
564 		hw->vqs[i] = NULL;
565 	}
566 
567 	rte_free(hw->vqs);
568 	hw->vqs = NULL;
569 }
570 
571 static int
572 virtio_alloc_queues(struct rte_eth_dev *dev)
573 {
574 	struct virtio_hw *hw = dev->data->dev_private;
575 	uint16_t nr_vq = virtio_get_nr_vq(hw);
576 	uint16_t i;
577 	int ret;
578 
579 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
580 	if (!hw->vqs) {
581 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
582 		return -ENOMEM;
583 	}
584 
585 	for (i = 0; i < nr_vq; i++) {
586 		ret = virtio_init_queue(dev, i);
587 		if (ret < 0) {
588 			virtio_free_queues(hw);
589 			return ret;
590 		}
591 	}
592 
593 	return 0;
594 }
595 
596 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
597 
598 static void
599 virtio_dev_close(struct rte_eth_dev *dev)
600 {
601 	struct virtio_hw *hw = dev->data->dev_private;
602 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
603 
604 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
605 
606 	/* reset the NIC */
607 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
608 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
609 	if (intr_conf->rxq)
610 		virtio_queues_unbind_intr(dev);
611 
612 	if (intr_conf->lsc || intr_conf->rxq) {
613 		virtio_intr_disable(dev);
614 		rte_intr_efd_disable(dev->intr_handle);
615 		rte_free(dev->intr_handle->intr_vec);
616 		dev->intr_handle->intr_vec = NULL;
617 	}
618 
619 	vtpci_reset(hw);
620 	virtio_dev_free_mbufs(dev);
621 	virtio_free_queues(hw);
622 }
623 
624 static void
625 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
626 {
627 	struct virtio_hw *hw = dev->data->dev_private;
628 	struct virtio_pmd_ctrl ctrl;
629 	int dlen[1];
630 	int ret;
631 
632 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
633 		PMD_INIT_LOG(INFO, "host does not support rx control");
634 		return;
635 	}
636 
637 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
638 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
639 	ctrl.data[0] = 1;
640 	dlen[0] = 1;
641 
642 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
643 	if (ret)
644 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
645 }
646 
647 static void
648 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
649 {
650 	struct virtio_hw *hw = dev->data->dev_private;
651 	struct virtio_pmd_ctrl ctrl;
652 	int dlen[1];
653 	int ret;
654 
655 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
656 		PMD_INIT_LOG(INFO, "host does not support rx control");
657 		return;
658 	}
659 
660 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
661 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
662 	ctrl.data[0] = 0;
663 	dlen[0] = 1;
664 
665 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
666 	if (ret)
667 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
668 }
669 
670 static void
671 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
672 {
673 	struct virtio_hw *hw = dev->data->dev_private;
674 	struct virtio_pmd_ctrl ctrl;
675 	int dlen[1];
676 	int ret;
677 
678 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
679 		PMD_INIT_LOG(INFO, "host does not support rx control");
680 		return;
681 	}
682 
683 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
684 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
685 	ctrl.data[0] = 1;
686 	dlen[0] = 1;
687 
688 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
689 	if (ret)
690 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
691 }
692 
693 static void
694 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
695 {
696 	struct virtio_hw *hw = dev->data->dev_private;
697 	struct virtio_pmd_ctrl ctrl;
698 	int dlen[1];
699 	int ret;
700 
701 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
702 		PMD_INIT_LOG(INFO, "host does not support rx control");
703 		return;
704 	}
705 
706 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
707 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
708 	ctrl.data[0] = 0;
709 	dlen[0] = 1;
710 
711 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
712 	if (ret)
713 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
714 }
715 
716 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
717 static int
718 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
719 {
720 	struct virtio_hw *hw = dev->data->dev_private;
721 	uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
722 				 hw->vtnet_hdr_size;
723 	uint32_t frame_size = mtu + ether_hdr_len;
724 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
725 
726 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
727 
728 	if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
729 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
730 			ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
731 		return -EINVAL;
732 	}
733 	return 0;
734 }
735 
736 static int
737 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
738 {
739 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
740 	struct virtqueue *vq = rxvq->vq;
741 
742 	virtqueue_enable_intr(vq);
743 	return 0;
744 }
745 
746 static int
747 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
748 {
749 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
750 	struct virtqueue *vq = rxvq->vq;
751 
752 	virtqueue_disable_intr(vq);
753 	return 0;
754 }
755 
756 /*
757  * dev_ops for virtio, bare necessities for basic operation
758  */
759 static const struct eth_dev_ops virtio_eth_dev_ops = {
760 	.dev_configure           = virtio_dev_configure,
761 	.dev_start               = virtio_dev_start,
762 	.dev_stop                = virtio_dev_stop,
763 	.dev_close               = virtio_dev_close,
764 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
765 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
766 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
767 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
768 	.mtu_set                 = virtio_mtu_set,
769 	.dev_infos_get           = virtio_dev_info_get,
770 	.stats_get               = virtio_dev_stats_get,
771 	.xstats_get              = virtio_dev_xstats_get,
772 	.xstats_get_names        = virtio_dev_xstats_get_names,
773 	.stats_reset             = virtio_dev_stats_reset,
774 	.xstats_reset            = virtio_dev_stats_reset,
775 	.link_update             = virtio_dev_link_update,
776 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
777 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
778 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
779 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
780 	.rx_queue_release        = virtio_dev_queue_release,
781 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
782 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
783 	.tx_queue_release        = virtio_dev_queue_release,
784 	/* collect stats per queue */
785 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
786 	.vlan_filter_set         = virtio_vlan_filter_set,
787 	.mac_addr_add            = virtio_mac_addr_add,
788 	.mac_addr_remove         = virtio_mac_addr_remove,
789 	.mac_addr_set            = virtio_mac_addr_set,
790 };
791 
792 static inline int
793 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
794 				struct rte_eth_link *link)
795 {
796 	struct rte_eth_link *dst = link;
797 	struct rte_eth_link *src = &(dev->data->dev_link);
798 
799 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
800 			*(uint64_t *)src) == 0)
801 		return -1;
802 
803 	return 0;
804 }
805 
806 /**
807  * Atomically writes the link status information into global
808  * structure rte_eth_dev.
809  *
810  * @param dev
811  *   - Pointer to the structure rte_eth_dev to read from.
812  *   - Pointer to the buffer to be saved with the link status.
813  *
814  * @return
815  *   - On success, zero.
816  *   - On failure, negative value.
817  */
818 static inline int
819 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
820 		struct rte_eth_link *link)
821 {
822 	struct rte_eth_link *dst = &(dev->data->dev_link);
823 	struct rte_eth_link *src = link;
824 
825 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
826 					*(uint64_t *)src) == 0)
827 		return -1;
828 
829 	return 0;
830 }
831 
832 static void
833 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
834 {
835 	unsigned i;
836 
837 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
838 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
839 		if (txvq == NULL)
840 			continue;
841 
842 		stats->opackets += txvq->stats.packets;
843 		stats->obytes += txvq->stats.bytes;
844 		stats->oerrors += txvq->stats.errors;
845 
846 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
847 			stats->q_opackets[i] = txvq->stats.packets;
848 			stats->q_obytes[i] = txvq->stats.bytes;
849 		}
850 	}
851 
852 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
853 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
854 		if (rxvq == NULL)
855 			continue;
856 
857 		stats->ipackets += rxvq->stats.packets;
858 		stats->ibytes += rxvq->stats.bytes;
859 		stats->ierrors += rxvq->stats.errors;
860 
861 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
862 			stats->q_ipackets[i] = rxvq->stats.packets;
863 			stats->q_ibytes[i] = rxvq->stats.bytes;
864 		}
865 	}
866 
867 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
868 }
869 
870 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
871 				       struct rte_eth_xstat_name *xstats_names,
872 				       __rte_unused unsigned limit)
873 {
874 	unsigned i;
875 	unsigned count = 0;
876 	unsigned t;
877 
878 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
879 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
880 
881 	if (xstats_names != NULL) {
882 		/* Note: limit checked in rte_eth_xstats_names() */
883 
884 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
885 			struct virtnet_rx *rxvq = dev->data->rx_queues[i];
886 			if (rxvq == NULL)
887 				continue;
888 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
889 				snprintf(xstats_names[count].name,
890 					sizeof(xstats_names[count].name),
891 					"rx_q%u_%s", i,
892 					rte_virtio_rxq_stat_strings[t].name);
893 				count++;
894 			}
895 		}
896 
897 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
898 			struct virtnet_tx *txvq = dev->data->tx_queues[i];
899 			if (txvq == NULL)
900 				continue;
901 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
902 				snprintf(xstats_names[count].name,
903 					sizeof(xstats_names[count].name),
904 					"tx_q%u_%s", i,
905 					rte_virtio_txq_stat_strings[t].name);
906 				count++;
907 			}
908 		}
909 		return count;
910 	}
911 	return nstats;
912 }
913 
914 static int
915 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
916 		      unsigned n)
917 {
918 	unsigned i;
919 	unsigned count = 0;
920 
921 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
922 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
923 
924 	if (n < nstats)
925 		return nstats;
926 
927 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
928 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
929 
930 		if (rxvq == NULL)
931 			continue;
932 
933 		unsigned t;
934 
935 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
936 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
937 				rte_virtio_rxq_stat_strings[t].offset);
938 			xstats[count].id = count;
939 			count++;
940 		}
941 	}
942 
943 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
944 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
945 
946 		if (txvq == NULL)
947 			continue;
948 
949 		unsigned t;
950 
951 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
952 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
953 				rte_virtio_txq_stat_strings[t].offset);
954 			xstats[count].id = count;
955 			count++;
956 		}
957 	}
958 
959 	return count;
960 }
961 
962 static int
963 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
964 {
965 	virtio_update_stats(dev, stats);
966 
967 	return 0;
968 }
969 
970 static void
971 virtio_dev_stats_reset(struct rte_eth_dev *dev)
972 {
973 	unsigned int i;
974 
975 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
976 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
977 		if (txvq == NULL)
978 			continue;
979 
980 		txvq->stats.packets = 0;
981 		txvq->stats.bytes = 0;
982 		txvq->stats.errors = 0;
983 		txvq->stats.multicast = 0;
984 		txvq->stats.broadcast = 0;
985 		memset(txvq->stats.size_bins, 0,
986 		       sizeof(txvq->stats.size_bins[0]) * 8);
987 	}
988 
989 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
990 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
991 		if (rxvq == NULL)
992 			continue;
993 
994 		rxvq->stats.packets = 0;
995 		rxvq->stats.bytes = 0;
996 		rxvq->stats.errors = 0;
997 		rxvq->stats.multicast = 0;
998 		rxvq->stats.broadcast = 0;
999 		memset(rxvq->stats.size_bins, 0,
1000 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1001 	}
1002 }
1003 
1004 static void
1005 virtio_set_hwaddr(struct virtio_hw *hw)
1006 {
1007 	vtpci_write_dev_config(hw,
1008 			offsetof(struct virtio_net_config, mac),
1009 			&hw->mac_addr, ETHER_ADDR_LEN);
1010 }
1011 
1012 static void
1013 virtio_get_hwaddr(struct virtio_hw *hw)
1014 {
1015 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1016 		vtpci_read_dev_config(hw,
1017 			offsetof(struct virtio_net_config, mac),
1018 			&hw->mac_addr, ETHER_ADDR_LEN);
1019 	} else {
1020 		eth_random_addr(&hw->mac_addr[0]);
1021 		virtio_set_hwaddr(hw);
1022 	}
1023 }
1024 
1025 static int
1026 virtio_mac_table_set(struct virtio_hw *hw,
1027 		     const struct virtio_net_ctrl_mac *uc,
1028 		     const struct virtio_net_ctrl_mac *mc)
1029 {
1030 	struct virtio_pmd_ctrl ctrl;
1031 	int err, len[2];
1032 
1033 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1034 		PMD_DRV_LOG(INFO, "host does not support mac table");
1035 		return -1;
1036 	}
1037 
1038 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1039 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1040 
1041 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1042 	memcpy(ctrl.data, uc, len[0]);
1043 
1044 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1045 	memcpy(ctrl.data + len[0], mc, len[1]);
1046 
1047 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1048 	if (err != 0)
1049 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1050 	return err;
1051 }
1052 
1053 static int
1054 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1055 		    uint32_t index, uint32_t vmdq __rte_unused)
1056 {
1057 	struct virtio_hw *hw = dev->data->dev_private;
1058 	const struct ether_addr *addrs = dev->data->mac_addrs;
1059 	unsigned int i;
1060 	struct virtio_net_ctrl_mac *uc, *mc;
1061 
1062 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1063 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1064 		return -EINVAL;
1065 	}
1066 
1067 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1068 	uc->entries = 0;
1069 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1070 	mc->entries = 0;
1071 
1072 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1073 		const struct ether_addr *addr
1074 			= (i == index) ? mac_addr : addrs + i;
1075 		struct virtio_net_ctrl_mac *tbl
1076 			= is_multicast_ether_addr(addr) ? mc : uc;
1077 
1078 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1079 	}
1080 
1081 	return virtio_mac_table_set(hw, uc, mc);
1082 }
1083 
1084 static void
1085 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1086 {
1087 	struct virtio_hw *hw = dev->data->dev_private;
1088 	struct ether_addr *addrs = dev->data->mac_addrs;
1089 	struct virtio_net_ctrl_mac *uc, *mc;
1090 	unsigned int i;
1091 
1092 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1093 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1094 		return;
1095 	}
1096 
1097 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1098 	uc->entries = 0;
1099 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1100 	mc->entries = 0;
1101 
1102 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1103 		struct virtio_net_ctrl_mac *tbl;
1104 
1105 		if (i == index || is_zero_ether_addr(addrs + i))
1106 			continue;
1107 
1108 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1109 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1110 	}
1111 
1112 	virtio_mac_table_set(hw, uc, mc);
1113 }
1114 
1115 static void
1116 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1117 {
1118 	struct virtio_hw *hw = dev->data->dev_private;
1119 
1120 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1121 
1122 	/* Use atomic update if available */
1123 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1124 		struct virtio_pmd_ctrl ctrl;
1125 		int len = ETHER_ADDR_LEN;
1126 
1127 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1128 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1129 
1130 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1131 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
1132 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1133 		virtio_set_hwaddr(hw);
1134 }
1135 
1136 static int
1137 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1138 {
1139 	struct virtio_hw *hw = dev->data->dev_private;
1140 	struct virtio_pmd_ctrl ctrl;
1141 	int len;
1142 
1143 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1144 		return -ENOTSUP;
1145 
1146 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1147 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1148 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1149 	len = sizeof(vlan_id);
1150 
1151 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1152 }
1153 
1154 static int
1155 virtio_intr_enable(struct rte_eth_dev *dev)
1156 {
1157 	struct virtio_hw *hw = dev->data->dev_private;
1158 
1159 	if (rte_intr_enable(dev->intr_handle) < 0)
1160 		return -1;
1161 
1162 	if (!hw->virtio_user_dev)
1163 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1164 
1165 	return 0;
1166 }
1167 
1168 static int
1169 virtio_intr_disable(struct rte_eth_dev *dev)
1170 {
1171 	struct virtio_hw *hw = dev->data->dev_private;
1172 
1173 	if (rte_intr_disable(dev->intr_handle) < 0)
1174 		return -1;
1175 
1176 	if (!hw->virtio_user_dev)
1177 		hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1178 
1179 	return 0;
1180 }
1181 
1182 static int
1183 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1184 {
1185 	uint64_t host_features;
1186 
1187 	/* Prepare guest_features: feature that driver wants to support */
1188 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1189 		req_features);
1190 
1191 	/* Read device(host) feature bits */
1192 	host_features = VTPCI_OPS(hw)->get_features(hw);
1193 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1194 		host_features);
1195 
1196 	/* If supported, ensure MTU value is valid before acknowledging it. */
1197 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1198 		struct virtio_net_config config;
1199 
1200 		vtpci_read_dev_config(hw,
1201 			offsetof(struct virtio_net_config, mtu),
1202 			&config.mtu, sizeof(config.mtu));
1203 
1204 		if (config.mtu < ETHER_MIN_MTU)
1205 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1206 	}
1207 
1208 	/*
1209 	 * Negotiate features: Subset of device feature bits are written back
1210 	 * guest feature bits.
1211 	 */
1212 	hw->guest_features = req_features;
1213 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1214 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1215 		hw->guest_features);
1216 
1217 	if (hw->modern) {
1218 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1219 			PMD_INIT_LOG(ERR,
1220 				"VIRTIO_F_VERSION_1 features is not enabled.");
1221 			return -1;
1222 		}
1223 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1224 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1225 			PMD_INIT_LOG(ERR,
1226 				"failed to set FEATURES_OK status!");
1227 			return -1;
1228 		}
1229 	}
1230 
1231 	hw->req_guest_features = req_features;
1232 
1233 	return 0;
1234 }
1235 
1236 /*
1237  * Process Virtio Config changed interrupt and call the callback
1238  * if link state changed.
1239  */
1240 void
1241 virtio_interrupt_handler(void *param)
1242 {
1243 	struct rte_eth_dev *dev = param;
1244 	struct virtio_hw *hw = dev->data->dev_private;
1245 	uint8_t isr;
1246 
1247 	/* Read interrupt status which clears interrupt */
1248 	isr = vtpci_isr(hw);
1249 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1250 
1251 	if (virtio_intr_enable(dev) < 0)
1252 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1253 
1254 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1255 		if (virtio_dev_link_update(dev, 0) == 0)
1256 			_rte_eth_dev_callback_process(dev,
1257 						      RTE_ETH_EVENT_INTR_LSC,
1258 						      NULL, NULL);
1259 	}
1260 
1261 }
1262 
1263 /* set rx and tx handlers according to what is supported */
1264 static void
1265 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1266 {
1267 	struct virtio_hw *hw = eth_dev->data->dev_private;
1268 
1269 	if (hw->use_simple_rx) {
1270 		PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1271 			eth_dev->data->port_id);
1272 		eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1273 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1274 		PMD_INIT_LOG(INFO,
1275 			"virtio: using mergeable buffer Rx path on port %u",
1276 			eth_dev->data->port_id);
1277 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1278 	} else {
1279 		PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1280 			eth_dev->data->port_id);
1281 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1282 	}
1283 
1284 	if (hw->use_simple_tx) {
1285 		PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1286 			eth_dev->data->port_id);
1287 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1288 	} else {
1289 		PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1290 			eth_dev->data->port_id);
1291 		eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1292 	}
1293 }
1294 
1295 /* Only support 1:1 queue/interrupt mapping so far.
1296  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1297  * interrupt vectors (<N+1).
1298  */
1299 static int
1300 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1301 {
1302 	uint32_t i;
1303 	struct virtio_hw *hw = dev->data->dev_private;
1304 
1305 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1306 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1307 		dev->intr_handle->intr_vec[i] = i + 1;
1308 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1309 						 VIRTIO_MSI_NO_VECTOR) {
1310 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1311 			return -EBUSY;
1312 		}
1313 	}
1314 
1315 	return 0;
1316 }
1317 
1318 static void
1319 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1320 {
1321 	uint32_t i;
1322 	struct virtio_hw *hw = dev->data->dev_private;
1323 
1324 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1325 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1326 		VTPCI_OPS(hw)->set_queue_irq(hw,
1327 					     hw->vqs[i * VTNET_CQ],
1328 					     VIRTIO_MSI_NO_VECTOR);
1329 }
1330 
1331 static int
1332 virtio_configure_intr(struct rte_eth_dev *dev)
1333 {
1334 	struct virtio_hw *hw = dev->data->dev_private;
1335 
1336 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1337 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1338 		return -ENOTSUP;
1339 	}
1340 
1341 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1342 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1343 		return -1;
1344 	}
1345 
1346 	if (!dev->intr_handle->intr_vec) {
1347 		dev->intr_handle->intr_vec =
1348 			rte_zmalloc("intr_vec",
1349 				    hw->max_queue_pairs * sizeof(int), 0);
1350 		if (!dev->intr_handle->intr_vec) {
1351 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1352 				     hw->max_queue_pairs);
1353 			return -ENOMEM;
1354 		}
1355 	}
1356 
1357 	/* Re-register callback to update max_intr */
1358 	rte_intr_callback_unregister(dev->intr_handle,
1359 				     virtio_interrupt_handler,
1360 				     dev);
1361 	rte_intr_callback_register(dev->intr_handle,
1362 				   virtio_interrupt_handler,
1363 				   dev);
1364 
1365 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1366 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1367 	 * And for legacy devices, this should be done before queue/vec binding
1368 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1369 	 * (22) will be ignored.
1370 	 */
1371 	if (virtio_intr_enable(dev) < 0) {
1372 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1373 		return -1;
1374 	}
1375 
1376 	if (virtio_queues_bind_intr(dev) < 0) {
1377 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1378 		return -1;
1379 	}
1380 
1381 	return 0;
1382 }
1383 
1384 /* reset device and renegotiate features if needed */
1385 static int
1386 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1387 {
1388 	struct virtio_hw *hw = eth_dev->data->dev_private;
1389 	struct virtio_net_config *config;
1390 	struct virtio_net_config local_config;
1391 	struct rte_pci_device *pci_dev = NULL;
1392 	int ret;
1393 
1394 	/* Reset the device although not necessary at startup */
1395 	vtpci_reset(hw);
1396 
1397 	if (hw->vqs) {
1398 		virtio_dev_free_mbufs(eth_dev);
1399 		virtio_free_queues(hw);
1400 	}
1401 
1402 	/* Tell the host we've noticed this device. */
1403 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1404 
1405 	/* Tell the host we've known how to drive the device. */
1406 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1407 	if (virtio_negotiate_features(hw, req_features) < 0)
1408 		return -1;
1409 
1410 	if (!hw->virtio_user_dev) {
1411 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1412 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1413 	}
1414 
1415 	/* If host does not support both status and MSI-X then disable LSC */
1416 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1417 	    hw->use_msix != VIRTIO_MSIX_NONE)
1418 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1419 	else
1420 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1421 
1422 	/* Setting up rx_header size for the device */
1423 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1424 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1425 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1426 	else
1427 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1428 
1429 	/* Copy the permanent MAC address to: virtio_hw */
1430 	virtio_get_hwaddr(hw);
1431 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1432 			&eth_dev->data->mac_addrs[0]);
1433 	PMD_INIT_LOG(DEBUG,
1434 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1435 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1436 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1437 
1438 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1439 		config = &local_config;
1440 
1441 		vtpci_read_dev_config(hw,
1442 			offsetof(struct virtio_net_config, mac),
1443 			&config->mac, sizeof(config->mac));
1444 
1445 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1446 			vtpci_read_dev_config(hw,
1447 				offsetof(struct virtio_net_config, status),
1448 				&config->status, sizeof(config->status));
1449 		} else {
1450 			PMD_INIT_LOG(DEBUG,
1451 				     "VIRTIO_NET_F_STATUS is not supported");
1452 			config->status = 0;
1453 		}
1454 
1455 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1456 			vtpci_read_dev_config(hw,
1457 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1458 				&config->max_virtqueue_pairs,
1459 				sizeof(config->max_virtqueue_pairs));
1460 		} else {
1461 			PMD_INIT_LOG(DEBUG,
1462 				     "VIRTIO_NET_F_MQ is not supported");
1463 			config->max_virtqueue_pairs = 1;
1464 		}
1465 
1466 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1467 
1468 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1469 			vtpci_read_dev_config(hw,
1470 				offsetof(struct virtio_net_config, mtu),
1471 				&config->mtu,
1472 				sizeof(config->mtu));
1473 
1474 			/*
1475 			 * MTU value has already been checked at negotiation
1476 			 * time, but check again in case it has changed since
1477 			 * then, which should not happen.
1478 			 */
1479 			if (config->mtu < ETHER_MIN_MTU) {
1480 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1481 						config->mtu);
1482 				return -1;
1483 			}
1484 
1485 			hw->max_mtu = config->mtu;
1486 			/* Set initial MTU to maximum one supported by vhost */
1487 			eth_dev->data->mtu = config->mtu;
1488 
1489 		} else {
1490 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1491 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1492 		}
1493 
1494 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1495 				config->max_virtqueue_pairs);
1496 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1497 		PMD_INIT_LOG(DEBUG,
1498 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1499 				config->mac[0], config->mac[1],
1500 				config->mac[2], config->mac[3],
1501 				config->mac[4], config->mac[5]);
1502 	} else {
1503 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1504 		hw->max_queue_pairs = 1;
1505 	}
1506 
1507 	ret = virtio_alloc_queues(eth_dev);
1508 	if (ret < 0)
1509 		return ret;
1510 
1511 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1512 		if (virtio_configure_intr(eth_dev) < 0) {
1513 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1514 			return -1;
1515 		}
1516 	}
1517 
1518 	vtpci_reinit_complete(hw);
1519 
1520 	if (pci_dev)
1521 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1522 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1523 			pci_dev->id.device_id);
1524 
1525 	return 0;
1526 }
1527 
1528 /*
1529  * Remap the PCI device again (IO port map for legacy device and
1530  * memory map for modern device), so that the secondary process
1531  * could have the PCI initiated correctly.
1532  */
1533 static int
1534 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1535 {
1536 	if (hw->modern) {
1537 		/*
1538 		 * We don't have to re-parse the PCI config space, since
1539 		 * rte_pci_map_device() makes sure the mapped address
1540 		 * in secondary process would equal to the one mapped in
1541 		 * the primary process: error will be returned if that
1542 		 * requirement is not met.
1543 		 *
1544 		 * That said, we could simply reuse all cap pointers
1545 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1546 		 * primary process, which is stored in shared memory.
1547 		 */
1548 		if (rte_pci_map_device(pci_dev)) {
1549 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1550 			return -1;
1551 		}
1552 	} else {
1553 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1554 			return -1;
1555 	}
1556 
1557 	return 0;
1558 }
1559 
1560 static void
1561 virtio_set_vtpci_ops(struct virtio_hw *hw)
1562 {
1563 #ifdef RTE_VIRTIO_USER
1564 	if (hw->virtio_user_dev)
1565 		VTPCI_OPS(hw) = &virtio_user_ops;
1566 	else
1567 #endif
1568 	if (hw->modern)
1569 		VTPCI_OPS(hw) = &modern_ops;
1570 	else
1571 		VTPCI_OPS(hw) = &legacy_ops;
1572 }
1573 
1574 /*
1575  * This function is based on probe() function in virtio_pci.c
1576  * It returns 0 on success.
1577  */
1578 int
1579 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1580 {
1581 	struct virtio_hw *hw = eth_dev->data->dev_private;
1582 	int ret;
1583 
1584 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1585 
1586 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1587 
1588 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1589 		if (!hw->virtio_user_dev) {
1590 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1591 			if (ret)
1592 				return ret;
1593 		}
1594 
1595 		virtio_set_vtpci_ops(hw);
1596 		set_rxtx_funcs(eth_dev);
1597 
1598 		return 0;
1599 	}
1600 
1601 	/* Allocate memory for storing MAC addresses */
1602 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1603 	if (eth_dev->data->mac_addrs == NULL) {
1604 		PMD_INIT_LOG(ERR,
1605 			"Failed to allocate %d bytes needed to store MAC addresses",
1606 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1607 		return -ENOMEM;
1608 	}
1609 
1610 	hw->port_id = eth_dev->data->port_id;
1611 	/* For virtio_user case the hw->virtio_user_dev is populated by
1612 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1613 	 */
1614 	if (!hw->virtio_user_dev) {
1615 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1616 		if (ret)
1617 			goto out;
1618 	}
1619 
1620 	/* reset device and negotiate default features */
1621 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1622 	if (ret < 0)
1623 		goto out;
1624 
1625 	/* Setup interrupt callback  */
1626 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1627 		rte_intr_callback_register(eth_dev->intr_handle,
1628 			virtio_interrupt_handler, eth_dev);
1629 
1630 	return 0;
1631 
1632 out:
1633 	rte_free(eth_dev->data->mac_addrs);
1634 	return ret;
1635 }
1636 
1637 static int
1638 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1639 {
1640 	PMD_INIT_FUNC_TRACE();
1641 
1642 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1643 		return -EPERM;
1644 
1645 	virtio_dev_stop(eth_dev);
1646 	virtio_dev_close(eth_dev);
1647 
1648 	eth_dev->dev_ops = NULL;
1649 	eth_dev->tx_pkt_burst = NULL;
1650 	eth_dev->rx_pkt_burst = NULL;
1651 
1652 	rte_free(eth_dev->data->mac_addrs);
1653 	eth_dev->data->mac_addrs = NULL;
1654 
1655 	/* reset interrupt callback  */
1656 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1657 		rte_intr_callback_unregister(eth_dev->intr_handle,
1658 						virtio_interrupt_handler,
1659 						eth_dev);
1660 	if (eth_dev->device)
1661 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1662 
1663 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1664 
1665 	return 0;
1666 }
1667 
1668 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1669 	struct rte_pci_device *pci_dev)
1670 {
1671 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1672 		eth_virtio_dev_init);
1673 }
1674 
1675 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1676 {
1677 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1678 }
1679 
1680 static struct rte_pci_driver rte_virtio_pmd = {
1681 	.driver = {
1682 		.name = "net_virtio",
1683 	},
1684 	.id_table = pci_id_virtio_map,
1685 	.drv_flags = 0,
1686 	.probe = eth_virtio_pci_probe,
1687 	.remove = eth_virtio_pci_remove,
1688 };
1689 
1690 RTE_INIT(rte_virtio_pmd_init);
1691 static void
1692 rte_virtio_pmd_init(void)
1693 {
1694 	if (rte_eal_iopl_init() != 0) {
1695 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1696 		return;
1697 	}
1698 
1699 	rte_pci_register(&rte_virtio_pmd);
1700 }
1701 
1702 /*
1703  * Configure virtio device
1704  * It returns 0 on success.
1705  */
1706 static int
1707 virtio_dev_configure(struct rte_eth_dev *dev)
1708 {
1709 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1710 	struct virtio_hw *hw = dev->data->dev_private;
1711 	uint64_t req_features;
1712 	int ret;
1713 
1714 	PMD_INIT_LOG(DEBUG, "configure");
1715 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1716 
1717 	if (dev->data->dev_conf.intr_conf.rxq) {
1718 		ret = virtio_init_device(dev, hw->req_guest_features);
1719 		if (ret < 0)
1720 			return ret;
1721 	}
1722 
1723 	/* The name hw_ip_checksum is a bit confusing since it can be
1724 	 * set by the application to request L3 and/or L4 checksums. In
1725 	 * case of virtio, only L4 checksum is supported.
1726 	 */
1727 	if (rxmode->hw_ip_checksum)
1728 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1729 
1730 	if (rxmode->enable_lro)
1731 		req_features |=
1732 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1733 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1734 
1735 	/* if request features changed, reinit the device */
1736 	if (req_features != hw->req_guest_features) {
1737 		ret = virtio_init_device(dev, req_features);
1738 		if (ret < 0)
1739 			return ret;
1740 	}
1741 
1742 	if (rxmode->hw_ip_checksum &&
1743 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1744 		PMD_DRV_LOG(ERR,
1745 			"rx checksum not available on this host");
1746 		return -ENOTSUP;
1747 	}
1748 
1749 	if (rxmode->enable_lro &&
1750 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1751 		 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
1752 		PMD_DRV_LOG(ERR,
1753 			"Large Receive Offload not available on this host");
1754 		return -ENOTSUP;
1755 	}
1756 
1757 	/* start control queue */
1758 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1759 		virtio_dev_cq_start(dev);
1760 
1761 	hw->vlan_strip = rxmode->hw_vlan_strip;
1762 
1763 	if (rxmode->hw_vlan_filter
1764 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1765 		PMD_DRV_LOG(ERR,
1766 			    "vlan filtering not available on this host");
1767 		return -ENOTSUP;
1768 	}
1769 
1770 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1771 		/* Enable vector (0) for Link State Intrerrupt */
1772 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1773 				VIRTIO_MSI_NO_VECTOR) {
1774 			PMD_DRV_LOG(ERR, "failed to set config vector");
1775 			return -EBUSY;
1776 		}
1777 
1778 	hw->use_simple_rx = 1;
1779 	hw->use_simple_tx = 1;
1780 
1781 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
1782 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1783 		hw->use_simple_rx = 0;
1784 		hw->use_simple_tx = 0;
1785 	}
1786 #endif
1787 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1788 		hw->use_simple_rx = 0;
1789 		hw->use_simple_tx = 0;
1790 	}
1791 
1792 	if (rxmode->hw_ip_checksum)
1793 		hw->use_simple_rx = 0;
1794 
1795 	return 0;
1796 }
1797 
1798 
1799 static int
1800 virtio_dev_start(struct rte_eth_dev *dev)
1801 {
1802 	uint16_t nb_queues, i;
1803 	struct virtnet_rx *rxvq;
1804 	struct virtnet_tx *txvq __rte_unused;
1805 	struct virtio_hw *hw = dev->data->dev_private;
1806 	int ret;
1807 
1808 	/* Finish the initialization of the queues */
1809 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1810 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
1811 		if (ret < 0)
1812 			return ret;
1813 	}
1814 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1815 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
1816 		if (ret < 0)
1817 			return ret;
1818 	}
1819 
1820 	/* check if lsc interrupt feature is enabled */
1821 	if (dev->data->dev_conf.intr_conf.lsc) {
1822 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1823 			PMD_DRV_LOG(ERR, "link status not supported by host");
1824 			return -ENOTSUP;
1825 		}
1826 	}
1827 
1828 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1829 	 * in device configure, but it could be unmapped  when device is
1830 	 * stopped.
1831 	 */
1832 	if (dev->data->dev_conf.intr_conf.lsc ||
1833 	    dev->data->dev_conf.intr_conf.rxq) {
1834 		virtio_intr_disable(dev);
1835 
1836 		if (virtio_intr_enable(dev) < 0) {
1837 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1838 			return -EIO;
1839 		}
1840 	}
1841 
1842 	/*Notify the backend
1843 	 *Otherwise the tap backend might already stop its queue due to fullness.
1844 	 *vhost backend will have no chance to be waked up
1845 	 */
1846 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1847 	if (hw->max_queue_pairs > 1) {
1848 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1849 			return -EINVAL;
1850 	}
1851 
1852 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1853 
1854 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1855 		rxvq = dev->data->rx_queues[i];
1856 		/* Flush the old packets */
1857 		virtqueue_rxvq_flush(rxvq->vq);
1858 		virtqueue_notify(rxvq->vq);
1859 	}
1860 
1861 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1862 		txvq = dev->data->tx_queues[i];
1863 		virtqueue_notify(txvq->vq);
1864 	}
1865 
1866 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1867 
1868 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1869 		rxvq = dev->data->rx_queues[i];
1870 		VIRTQUEUE_DUMP(rxvq->vq);
1871 	}
1872 
1873 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1874 		txvq = dev->data->tx_queues[i];
1875 		VIRTQUEUE_DUMP(txvq->vq);
1876 	}
1877 
1878 	set_rxtx_funcs(dev);
1879 	hw->started = 1;
1880 
1881 	/* Initialize Link state */
1882 	virtio_dev_link_update(dev, 0);
1883 
1884 	return 0;
1885 }
1886 
1887 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1888 {
1889 	struct rte_mbuf *buf;
1890 	int i, mbuf_num = 0;
1891 
1892 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1893 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1894 
1895 		if (rxvq == NULL || rxvq->vq == NULL)
1896 			continue;
1897 
1898 		PMD_INIT_LOG(DEBUG,
1899 			     "Before freeing rxq[%d] used and unused buf", i);
1900 		VIRTQUEUE_DUMP(rxvq->vq);
1901 
1902 		PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1903 		while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1904 			rte_pktmbuf_free(buf);
1905 			mbuf_num++;
1906 		}
1907 
1908 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1909 		PMD_INIT_LOG(DEBUG,
1910 			     "After freeing rxq[%d] used and unused buf", i);
1911 		VIRTQUEUE_DUMP(rxvq->vq);
1912 	}
1913 
1914 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1915 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1916 
1917 		if (txvq == NULL || txvq->vq == NULL)
1918 			continue;
1919 
1920 		PMD_INIT_LOG(DEBUG,
1921 			     "Before freeing txq[%d] used and unused bufs",
1922 			     i);
1923 		VIRTQUEUE_DUMP(txvq->vq);
1924 
1925 		mbuf_num = 0;
1926 		while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1927 			rte_pktmbuf_free(buf);
1928 			mbuf_num++;
1929 		}
1930 
1931 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1932 		PMD_INIT_LOG(DEBUG,
1933 			     "After freeing txq[%d] used and unused buf", i);
1934 		VIRTQUEUE_DUMP(txvq->vq);
1935 	}
1936 }
1937 
1938 /*
1939  * Stop device: disable interrupt and mark link down
1940  */
1941 static void
1942 virtio_dev_stop(struct rte_eth_dev *dev)
1943 {
1944 	struct virtio_hw *hw = dev->data->dev_private;
1945 	struct rte_eth_link link;
1946 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1947 
1948 	PMD_INIT_LOG(DEBUG, "stop");
1949 
1950 	if (intr_conf->lsc || intr_conf->rxq)
1951 		virtio_intr_disable(dev);
1952 
1953 	hw->started = 0;
1954 	memset(&link, 0, sizeof(link));
1955 	virtio_dev_atomic_write_link_status(dev, &link);
1956 }
1957 
1958 static int
1959 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1960 {
1961 	struct rte_eth_link link, old;
1962 	uint16_t status;
1963 	struct virtio_hw *hw = dev->data->dev_private;
1964 	memset(&link, 0, sizeof(link));
1965 	virtio_dev_atomic_read_link_status(dev, &link);
1966 	old = link;
1967 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1968 	link.link_speed  = SPEED_10G;
1969 
1970 	if (hw->started == 0) {
1971 		link.link_status = ETH_LINK_DOWN;
1972 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1973 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
1974 		vtpci_read_dev_config(hw,
1975 				offsetof(struct virtio_net_config, status),
1976 				&status, sizeof(status));
1977 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1978 			link.link_status = ETH_LINK_DOWN;
1979 			PMD_INIT_LOG(DEBUG, "Port %d is down",
1980 				     dev->data->port_id);
1981 		} else {
1982 			link.link_status = ETH_LINK_UP;
1983 			PMD_INIT_LOG(DEBUG, "Port %d is up",
1984 				     dev->data->port_id);
1985 		}
1986 	} else {
1987 		link.link_status = ETH_LINK_UP;
1988 	}
1989 	virtio_dev_atomic_write_link_status(dev, &link);
1990 
1991 	return (old.link_status == link.link_status) ? -1 : 0;
1992 }
1993 
1994 static int
1995 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1996 {
1997 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1998 	struct virtio_hw *hw = dev->data->dev_private;
1999 
2000 	if (mask & ETH_VLAN_FILTER_MASK) {
2001 		if (rxmode->hw_vlan_filter &&
2002 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2003 
2004 			PMD_DRV_LOG(NOTICE,
2005 				"vlan filtering not available on this host");
2006 
2007 			return -ENOTSUP;
2008 		}
2009 	}
2010 
2011 	if (mask & ETH_VLAN_STRIP_MASK)
2012 		hw->vlan_strip = rxmode->hw_vlan_strip;
2013 
2014 	return 0;
2015 }
2016 
2017 static void
2018 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2019 {
2020 	uint64_t tso_mask, host_features;
2021 	struct virtio_hw *hw = dev->data->dev_private;
2022 
2023 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2024 
2025 	dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
2026 	dev_info->max_rx_queues =
2027 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2028 	dev_info->max_tx_queues =
2029 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2030 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2031 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2032 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2033 	dev_info->default_txconf = (struct rte_eth_txconf) {
2034 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
2035 	};
2036 
2037 	host_features = VTPCI_OPS(hw)->get_features(hw);
2038 	dev_info->rx_offload_capa = 0;
2039 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2040 		dev_info->rx_offload_capa |=
2041 			DEV_RX_OFFLOAD_TCP_CKSUM |
2042 			DEV_RX_OFFLOAD_UDP_CKSUM;
2043 	}
2044 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2045 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2046 	if ((host_features & tso_mask) == tso_mask)
2047 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2048 
2049 	dev_info->tx_offload_capa = 0;
2050 	if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2051 		dev_info->tx_offload_capa |=
2052 			DEV_TX_OFFLOAD_UDP_CKSUM |
2053 			DEV_TX_OFFLOAD_TCP_CKSUM;
2054 	}
2055 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2056 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2057 	if ((hw->guest_features & tso_mask) == tso_mask)
2058 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2059 }
2060 
2061 /*
2062  * It enables testpmd to collect per queue stats.
2063  */
2064 static int
2065 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2066 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2067 __rte_unused uint8_t is_rx)
2068 {
2069 	return 0;
2070 }
2071 
2072 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2073 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2074 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2075