xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 3e0ceb9f17fff027fc6c8f18de35e11719ffa61e)
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_dev_queue_stats_mapping_set(
101 	struct rte_eth_dev *eth_dev,
102 	uint16_t queue_id,
103 	uint8_t stat_idx,
104 	uint8_t is_rx);
105 
106 /*
107  * The set of PCI devices this driver supports
108  */
109 static const struct rte_pci_id pci_id_virtio_map[] = {
110 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
111 	{ RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
112 	{ .vendor_id = 0, /* sentinel */ },
113 };
114 
115 struct rte_virtio_xstats_name_off {
116 	char name[RTE_ETH_XSTATS_NAME_SIZE];
117 	unsigned offset;
118 };
119 
120 /* [rt]x_qX_ is prepended to the name string here */
121 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
122 	{"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
123 	{"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
124 	{"errors",                 offsetof(struct virtnet_rx, stats.errors)},
125 	{"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
126 	{"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
127 	{"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
128 	{"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
129 	{"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
130 	{"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
131 	{"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
132 	{"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
133 	{"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
134 	{"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
135 };
136 
137 /* [rt]x_qX_ is prepended to the name string here */
138 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
139 	{"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
140 	{"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
141 	{"errors",                 offsetof(struct virtnet_tx, stats.errors)},
142 	{"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
143 	{"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
144 	{"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
145 	{"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
146 	{"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
147 	{"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
148 	{"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
149 	{"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
150 	{"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
151 	{"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
152 };
153 
154 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
155 			    sizeof(rte_virtio_rxq_stat_strings[0]))
156 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
157 			    sizeof(rte_virtio_txq_stat_strings[0]))
158 
159 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
160 
161 static int
162 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
163 		int *dlen, int pkt_num)
164 {
165 	uint32_t head, i;
166 	int k, sum = 0;
167 	virtio_net_ctrl_ack status = ~0;
168 	struct virtio_pmd_ctrl *result;
169 	struct virtqueue *vq;
170 
171 	ctrl->status = status;
172 
173 	if (!cvq || !cvq->vq) {
174 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
175 		return -1;
176 	}
177 	vq = cvq->vq;
178 	head = vq->vq_desc_head_idx;
179 
180 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
181 		"vq->hw->cvq = %p vq = %p",
182 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
183 
184 	if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
185 		return -1;
186 
187 	memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
188 		sizeof(struct virtio_pmd_ctrl));
189 
190 	/*
191 	 * Format is enforced in qemu code:
192 	 * One TX packet for header;
193 	 * At least one TX packet per argument;
194 	 * One RX packet for ACK.
195 	 */
196 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
197 	vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
198 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
199 	vq->vq_free_cnt--;
200 	i = vq->vq_ring.desc[head].next;
201 
202 	for (k = 0; k < pkt_num; k++) {
203 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
204 		vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
205 			+ sizeof(struct virtio_net_ctrl_hdr)
206 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
207 		vq->vq_ring.desc[i].len = dlen[k];
208 		sum += dlen[k];
209 		vq->vq_free_cnt--;
210 		i = vq->vq_ring.desc[i].next;
211 	}
212 
213 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
214 	vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
215 			+ sizeof(struct virtio_net_ctrl_hdr);
216 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
217 	vq->vq_free_cnt--;
218 
219 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
220 
221 	vq_update_avail_ring(vq, head);
222 	vq_update_avail_idx(vq);
223 
224 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
225 
226 	virtqueue_notify(vq);
227 
228 	rte_rmb();
229 	while (VIRTQUEUE_NUSED(vq) == 0) {
230 		rte_rmb();
231 		usleep(100);
232 	}
233 
234 	while (VIRTQUEUE_NUSED(vq)) {
235 		uint32_t idx, desc_idx, used_idx;
236 		struct vring_used_elem *uep;
237 
238 		used_idx = (uint32_t)(vq->vq_used_cons_idx
239 				& (vq->vq_nentries - 1));
240 		uep = &vq->vq_ring.used->ring[used_idx];
241 		idx = (uint32_t) uep->id;
242 		desc_idx = idx;
243 
244 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
245 			desc_idx = vq->vq_ring.desc[desc_idx].next;
246 			vq->vq_free_cnt++;
247 		}
248 
249 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
250 		vq->vq_desc_head_idx = idx;
251 
252 		vq->vq_used_cons_idx++;
253 		vq->vq_free_cnt++;
254 	}
255 
256 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
257 			vq->vq_free_cnt, vq->vq_desc_head_idx);
258 
259 	result = cvq->virtio_net_hdr_mz->addr;
260 
261 	return result->status;
262 }
263 
264 static int
265 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
266 {
267 	struct virtio_hw *hw = dev->data->dev_private;
268 	struct virtio_pmd_ctrl ctrl;
269 	int dlen[1];
270 	int ret;
271 
272 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
273 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
274 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
275 
276 	dlen[0] = sizeof(uint16_t);
277 
278 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
279 	if (ret) {
280 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
281 			  "failed, this is too late now...");
282 		return -EINVAL;
283 	}
284 
285 	return 0;
286 }
287 
288 static void
289 virtio_dev_queue_release(void *queue __rte_unused)
290 {
291 	/* do nothing */
292 }
293 
294 static int
295 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
296 {
297 	if (vtpci_queue_idx == hw->max_queue_pairs * 2)
298 		return VTNET_CQ;
299 	else if (vtpci_queue_idx % 2 == 0)
300 		return VTNET_RQ;
301 	else
302 		return VTNET_TQ;
303 }
304 
305 static uint16_t
306 virtio_get_nr_vq(struct virtio_hw *hw)
307 {
308 	uint16_t nr_vq = hw->max_queue_pairs * 2;
309 
310 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
311 		nr_vq += 1;
312 
313 	return nr_vq;
314 }
315 
316 static void
317 virtio_init_vring(struct virtqueue *vq)
318 {
319 	int size = vq->vq_nentries;
320 	struct vring *vr = &vq->vq_ring;
321 	uint8_t *ring_mem = vq->vq_ring_virt_mem;
322 
323 	PMD_INIT_FUNC_TRACE();
324 
325 	/*
326 	 * Reinitialise since virtio port might have been stopped and restarted
327 	 */
328 	memset(ring_mem, 0, vq->vq_ring_size);
329 	vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
330 	vq->vq_used_cons_idx = 0;
331 	vq->vq_desc_head_idx = 0;
332 	vq->vq_avail_idx = 0;
333 	vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
334 	vq->vq_free_cnt = vq->vq_nentries;
335 	memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
336 
337 	vring_desc_init(vr->desc, size);
338 
339 	/*
340 	 * Disable device(host) interrupting guest
341 	 */
342 	virtqueue_disable_intr(vq);
343 }
344 
345 static int
346 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
347 {
348 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
349 	char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
350 	const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
351 	unsigned int vq_size, size;
352 	struct virtio_hw *hw = dev->data->dev_private;
353 	struct virtnet_rx *rxvq = NULL;
354 	struct virtnet_tx *txvq = NULL;
355 	struct virtnet_ctl *cvq = NULL;
356 	struct virtqueue *vq;
357 	size_t sz_hdr_mz = 0;
358 	void *sw_ring = NULL;
359 	int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
360 	int ret;
361 
362 	PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
363 
364 	/*
365 	 * Read the virtqueue size from the Queue Size field
366 	 * Always power of 2 and if 0 virtqueue does not exist
367 	 */
368 	vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
369 	PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
370 	if (vq_size == 0) {
371 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
372 		return -EINVAL;
373 	}
374 
375 	if (!rte_is_power_of_2(vq_size)) {
376 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
377 		return -EINVAL;
378 	}
379 
380 	snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
381 		 dev->data->port_id, vtpci_queue_idx);
382 
383 	size = RTE_ALIGN_CEIL(sizeof(*vq) +
384 				vq_size * sizeof(struct vq_desc_extra),
385 				RTE_CACHE_LINE_SIZE);
386 	if (queue_type == VTNET_TQ) {
387 		/*
388 		 * For each xmit packet, allocate a virtio_net_hdr
389 		 * and indirect ring elements
390 		 */
391 		sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
392 	} else if (queue_type == VTNET_CQ) {
393 		/* Allocate a page for control vq command, data and status */
394 		sz_hdr_mz = PAGE_SIZE;
395 	}
396 
397 	vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
398 				SOCKET_ID_ANY);
399 	if (vq == NULL) {
400 		PMD_INIT_LOG(ERR, "can not allocate vq");
401 		return -ENOMEM;
402 	}
403 	hw->vqs[vtpci_queue_idx] = vq;
404 
405 	vq->hw = hw;
406 	vq->vq_queue_index = vtpci_queue_idx;
407 	vq->vq_nentries = vq_size;
408 
409 	/*
410 	 * Reserve a memzone for vring elements
411 	 */
412 	size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
413 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
414 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
415 		     size, vq->vq_ring_size);
416 
417 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
418 					 SOCKET_ID_ANY,
419 					 0, VIRTIO_PCI_VRING_ALIGN);
420 	if (mz == NULL) {
421 		if (rte_errno == EEXIST)
422 			mz = rte_memzone_lookup(vq_name);
423 		if (mz == NULL) {
424 			ret = -ENOMEM;
425 			goto fail_q_alloc;
426 		}
427 	}
428 
429 	memset(mz->addr, 0, mz->len);
430 
431 	vq->vq_ring_mem = mz->iova;
432 	vq->vq_ring_virt_mem = mz->addr;
433 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
434 		     (uint64_t)mz->iova);
435 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
436 		     (uint64_t)(uintptr_t)mz->addr);
437 
438 	virtio_init_vring(vq);
439 
440 	if (sz_hdr_mz) {
441 		snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
442 			 dev->data->port_id, vtpci_queue_idx);
443 		hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
444 						     SOCKET_ID_ANY, 0,
445 						     RTE_CACHE_LINE_SIZE);
446 		if (hdr_mz == NULL) {
447 			if (rte_errno == EEXIST)
448 				hdr_mz = rte_memzone_lookup(vq_hdr_name);
449 			if (hdr_mz == NULL) {
450 				ret = -ENOMEM;
451 				goto fail_q_alloc;
452 			}
453 		}
454 	}
455 
456 	if (queue_type == VTNET_RQ) {
457 		size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
458 			       sizeof(vq->sw_ring[0]);
459 
460 		sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
461 				RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
462 		if (!sw_ring) {
463 			PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
464 			ret = -ENOMEM;
465 			goto fail_q_alloc;
466 		}
467 
468 		vq->sw_ring = sw_ring;
469 		rxvq = &vq->rxq;
470 		rxvq->vq = vq;
471 		rxvq->port_id = dev->data->port_id;
472 		rxvq->mz = mz;
473 	} else if (queue_type == VTNET_TQ) {
474 		txvq = &vq->txq;
475 		txvq->vq = vq;
476 		txvq->port_id = dev->data->port_id;
477 		txvq->mz = mz;
478 		txvq->virtio_net_hdr_mz = hdr_mz;
479 		txvq->virtio_net_hdr_mem = hdr_mz->iova;
480 	} else if (queue_type == VTNET_CQ) {
481 		cvq = &vq->cq;
482 		cvq->vq = vq;
483 		cvq->mz = mz;
484 		cvq->virtio_net_hdr_mz = hdr_mz;
485 		cvq->virtio_net_hdr_mem = hdr_mz->iova;
486 		memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
487 
488 		hw->cvq = cvq;
489 	}
490 
491 	/* For virtio_user case (that is when hw->dev is NULL), we use
492 	 * virtual address. And we need properly set _offset_, please see
493 	 * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
494 	 */
495 	if (!hw->virtio_user_dev)
496 		vq->offset = offsetof(struct rte_mbuf, buf_iova);
497 	else {
498 		vq->vq_ring_mem = (uintptr_t)mz->addr;
499 		vq->offset = offsetof(struct rte_mbuf, buf_addr);
500 		if (queue_type == VTNET_TQ)
501 			txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
502 		else if (queue_type == VTNET_CQ)
503 			cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
504 	}
505 
506 	if (queue_type == VTNET_TQ) {
507 		struct virtio_tx_region *txr;
508 		unsigned int i;
509 
510 		txr = hdr_mz->addr;
511 		memset(txr, 0, vq_size * sizeof(*txr));
512 		for (i = 0; i < vq_size; i++) {
513 			struct vring_desc *start_dp = txr[i].tx_indir;
514 
515 			vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
516 
517 			/* first indirect descriptor is always the tx header */
518 			start_dp->addr = txvq->virtio_net_hdr_mem
519 				+ i * sizeof(*txr)
520 				+ offsetof(struct virtio_tx_region, tx_hdr);
521 
522 			start_dp->len = hw->vtnet_hdr_size;
523 			start_dp->flags = VRING_DESC_F_NEXT;
524 		}
525 	}
526 
527 	if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
528 		PMD_INIT_LOG(ERR, "setup_queue failed");
529 		return -EINVAL;
530 	}
531 
532 	return 0;
533 
534 fail_q_alloc:
535 	rte_free(sw_ring);
536 	rte_memzone_free(hdr_mz);
537 	rte_memzone_free(mz);
538 	rte_free(vq);
539 
540 	return ret;
541 }
542 
543 static void
544 virtio_free_queues(struct virtio_hw *hw)
545 {
546 	uint16_t nr_vq = virtio_get_nr_vq(hw);
547 	struct virtqueue *vq;
548 	int queue_type;
549 	uint16_t i;
550 
551 	if (hw->vqs == NULL)
552 		return;
553 
554 	for (i = 0; i < nr_vq; i++) {
555 		vq = hw->vqs[i];
556 		if (!vq)
557 			continue;
558 
559 		queue_type = virtio_get_queue_type(hw, i);
560 		if (queue_type == VTNET_RQ) {
561 			rte_free(vq->sw_ring);
562 			rte_memzone_free(vq->rxq.mz);
563 		} else if (queue_type == VTNET_TQ) {
564 			rte_memzone_free(vq->txq.mz);
565 			rte_memzone_free(vq->txq.virtio_net_hdr_mz);
566 		} else {
567 			rte_memzone_free(vq->cq.mz);
568 			rte_memzone_free(vq->cq.virtio_net_hdr_mz);
569 		}
570 
571 		rte_free(vq);
572 		hw->vqs[i] = NULL;
573 	}
574 
575 	rte_free(hw->vqs);
576 	hw->vqs = NULL;
577 }
578 
579 static int
580 virtio_alloc_queues(struct rte_eth_dev *dev)
581 {
582 	struct virtio_hw *hw = dev->data->dev_private;
583 	uint16_t nr_vq = virtio_get_nr_vq(hw);
584 	uint16_t i;
585 	int ret;
586 
587 	hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
588 	if (!hw->vqs) {
589 		PMD_INIT_LOG(ERR, "failed to allocate vqs");
590 		return -ENOMEM;
591 	}
592 
593 	for (i = 0; i < nr_vq; i++) {
594 		ret = virtio_init_queue(dev, i);
595 		if (ret < 0) {
596 			virtio_free_queues(hw);
597 			return ret;
598 		}
599 	}
600 
601 	return 0;
602 }
603 
604 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
605 
606 static void
607 virtio_dev_close(struct rte_eth_dev *dev)
608 {
609 	struct virtio_hw *hw = dev->data->dev_private;
610 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
611 
612 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
613 
614 	/* reset the NIC */
615 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
616 		VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
617 	if (intr_conf->rxq)
618 		virtio_queues_unbind_intr(dev);
619 
620 	if (intr_conf->lsc || intr_conf->rxq) {
621 		rte_intr_disable(dev->intr_handle);
622 		rte_intr_efd_disable(dev->intr_handle);
623 		rte_free(dev->intr_handle->intr_vec);
624 		dev->intr_handle->intr_vec = NULL;
625 	}
626 
627 	vtpci_reset(hw);
628 	virtio_dev_free_mbufs(dev);
629 	virtio_free_queues(hw);
630 }
631 
632 static void
633 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
634 {
635 	struct virtio_hw *hw = dev->data->dev_private;
636 	struct virtio_pmd_ctrl ctrl;
637 	int dlen[1];
638 	int ret;
639 
640 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
641 		PMD_INIT_LOG(INFO, "host does not support rx control");
642 		return;
643 	}
644 
645 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
646 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
647 	ctrl.data[0] = 1;
648 	dlen[0] = 1;
649 
650 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
651 	if (ret)
652 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
653 }
654 
655 static void
656 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
657 {
658 	struct virtio_hw *hw = dev->data->dev_private;
659 	struct virtio_pmd_ctrl ctrl;
660 	int dlen[1];
661 	int ret;
662 
663 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
664 		PMD_INIT_LOG(INFO, "host does not support rx control");
665 		return;
666 	}
667 
668 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
669 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
670 	ctrl.data[0] = 0;
671 	dlen[0] = 1;
672 
673 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
674 	if (ret)
675 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
676 }
677 
678 static void
679 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
680 {
681 	struct virtio_hw *hw = dev->data->dev_private;
682 	struct virtio_pmd_ctrl ctrl;
683 	int dlen[1];
684 	int ret;
685 
686 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
687 		PMD_INIT_LOG(INFO, "host does not support rx control");
688 		return;
689 	}
690 
691 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
692 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
693 	ctrl.data[0] = 1;
694 	dlen[0] = 1;
695 
696 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
697 	if (ret)
698 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
699 }
700 
701 static void
702 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
703 {
704 	struct virtio_hw *hw = dev->data->dev_private;
705 	struct virtio_pmd_ctrl ctrl;
706 	int dlen[1];
707 	int ret;
708 
709 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
710 		PMD_INIT_LOG(INFO, "host does not support rx control");
711 		return;
712 	}
713 
714 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
715 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
716 	ctrl.data[0] = 0;
717 	dlen[0] = 1;
718 
719 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
720 	if (ret)
721 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
722 }
723 
724 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
725 static int
726 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
727 {
728 	struct virtio_hw *hw = dev->data->dev_private;
729 	uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
730 				 hw->vtnet_hdr_size;
731 	uint32_t frame_size = mtu + ether_hdr_len;
732 	uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
733 
734 	max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
735 
736 	if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
737 		PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
738 			ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
739 		return -EINVAL;
740 	}
741 	return 0;
742 }
743 
744 static int
745 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
746 {
747 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
748 	struct virtqueue *vq = rxvq->vq;
749 
750 	virtqueue_enable_intr(vq);
751 	return 0;
752 }
753 
754 static int
755 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
756 {
757 	struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
758 	struct virtqueue *vq = rxvq->vq;
759 
760 	virtqueue_disable_intr(vq);
761 	return 0;
762 }
763 
764 /*
765  * dev_ops for virtio, bare necessities for basic operation
766  */
767 static const struct eth_dev_ops virtio_eth_dev_ops = {
768 	.dev_configure           = virtio_dev_configure,
769 	.dev_start               = virtio_dev_start,
770 	.dev_stop                = virtio_dev_stop,
771 	.dev_close               = virtio_dev_close,
772 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
773 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
774 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
775 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
776 	.mtu_set                 = virtio_mtu_set,
777 	.dev_infos_get           = virtio_dev_info_get,
778 	.stats_get               = virtio_dev_stats_get,
779 	.xstats_get              = virtio_dev_xstats_get,
780 	.xstats_get_names        = virtio_dev_xstats_get_names,
781 	.stats_reset             = virtio_dev_stats_reset,
782 	.xstats_reset            = virtio_dev_stats_reset,
783 	.link_update             = virtio_dev_link_update,
784 	.vlan_offload_set        = virtio_dev_vlan_offload_set,
785 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
786 	.rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
787 	.rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
788 	.rx_queue_release        = virtio_dev_queue_release,
789 	.rx_descriptor_done      = virtio_dev_rx_queue_done,
790 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
791 	.tx_queue_release        = virtio_dev_queue_release,
792 	/* collect stats per queue */
793 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
794 	.vlan_filter_set         = virtio_vlan_filter_set,
795 	.mac_addr_add            = virtio_mac_addr_add,
796 	.mac_addr_remove         = virtio_mac_addr_remove,
797 	.mac_addr_set            = virtio_mac_addr_set,
798 };
799 
800 static inline int
801 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
802 				struct rte_eth_link *link)
803 {
804 	struct rte_eth_link *dst = link;
805 	struct rte_eth_link *src = &(dev->data->dev_link);
806 
807 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
808 			*(uint64_t *)src) == 0)
809 		return -1;
810 
811 	return 0;
812 }
813 
814 /**
815  * Atomically writes the link status information into global
816  * structure rte_eth_dev.
817  *
818  * @param dev
819  *   - Pointer to the structure rte_eth_dev to read from.
820  *   - Pointer to the buffer to be saved with the link status.
821  *
822  * @return
823  *   - On success, zero.
824  *   - On failure, negative value.
825  */
826 static inline int
827 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
828 		struct rte_eth_link *link)
829 {
830 	struct rte_eth_link *dst = &(dev->data->dev_link);
831 	struct rte_eth_link *src = link;
832 
833 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
834 					*(uint64_t *)src) == 0)
835 		return -1;
836 
837 	return 0;
838 }
839 
840 static void
841 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
842 {
843 	unsigned i;
844 
845 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
846 		const struct virtnet_tx *txvq = dev->data->tx_queues[i];
847 		if (txvq == NULL)
848 			continue;
849 
850 		stats->opackets += txvq->stats.packets;
851 		stats->obytes += txvq->stats.bytes;
852 		stats->oerrors += txvq->stats.errors;
853 
854 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
855 			stats->q_opackets[i] = txvq->stats.packets;
856 			stats->q_obytes[i] = txvq->stats.bytes;
857 		}
858 	}
859 
860 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
861 		const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
862 		if (rxvq == NULL)
863 			continue;
864 
865 		stats->ipackets += rxvq->stats.packets;
866 		stats->ibytes += rxvq->stats.bytes;
867 		stats->ierrors += rxvq->stats.errors;
868 
869 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
870 			stats->q_ipackets[i] = rxvq->stats.packets;
871 			stats->q_ibytes[i] = rxvq->stats.bytes;
872 		}
873 	}
874 
875 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
876 }
877 
878 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
879 				       struct rte_eth_xstat_name *xstats_names,
880 				       __rte_unused unsigned limit)
881 {
882 	unsigned i;
883 	unsigned count = 0;
884 	unsigned t;
885 
886 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
887 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
888 
889 	if (xstats_names != NULL) {
890 		/* Note: limit checked in rte_eth_xstats_names() */
891 
892 		for (i = 0; i < dev->data->nb_rx_queues; i++) {
893 			struct virtqueue *rxvq = dev->data->rx_queues[i];
894 			if (rxvq == NULL)
895 				continue;
896 			for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
897 				snprintf(xstats_names[count].name,
898 					sizeof(xstats_names[count].name),
899 					"rx_q%u_%s", i,
900 					rte_virtio_rxq_stat_strings[t].name);
901 				count++;
902 			}
903 		}
904 
905 		for (i = 0; i < dev->data->nb_tx_queues; i++) {
906 			struct virtqueue *txvq = dev->data->tx_queues[i];
907 			if (txvq == NULL)
908 				continue;
909 			for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
910 				snprintf(xstats_names[count].name,
911 					sizeof(xstats_names[count].name),
912 					"tx_q%u_%s", i,
913 					rte_virtio_txq_stat_strings[t].name);
914 				count++;
915 			}
916 		}
917 		return count;
918 	}
919 	return nstats;
920 }
921 
922 static int
923 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
924 		      unsigned n)
925 {
926 	unsigned i;
927 	unsigned count = 0;
928 
929 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
930 		dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
931 
932 	if (n < nstats)
933 		return nstats;
934 
935 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
936 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
937 
938 		if (rxvq == NULL)
939 			continue;
940 
941 		unsigned t;
942 
943 		for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
944 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
945 				rte_virtio_rxq_stat_strings[t].offset);
946 			xstats[count].id = count;
947 			count++;
948 		}
949 	}
950 
951 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
952 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
953 
954 		if (txvq == NULL)
955 			continue;
956 
957 		unsigned t;
958 
959 		for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
960 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
961 				rte_virtio_txq_stat_strings[t].offset);
962 			xstats[count].id = count;
963 			count++;
964 		}
965 	}
966 
967 	return count;
968 }
969 
970 static int
971 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
972 {
973 	virtio_update_stats(dev, stats);
974 
975 	return 0;
976 }
977 
978 static void
979 virtio_dev_stats_reset(struct rte_eth_dev *dev)
980 {
981 	unsigned int i;
982 
983 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
984 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
985 		if (txvq == NULL)
986 			continue;
987 
988 		txvq->stats.packets = 0;
989 		txvq->stats.bytes = 0;
990 		txvq->stats.errors = 0;
991 		txvq->stats.multicast = 0;
992 		txvq->stats.broadcast = 0;
993 		memset(txvq->stats.size_bins, 0,
994 		       sizeof(txvq->stats.size_bins[0]) * 8);
995 	}
996 
997 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
998 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
999 		if (rxvq == NULL)
1000 			continue;
1001 
1002 		rxvq->stats.packets = 0;
1003 		rxvq->stats.bytes = 0;
1004 		rxvq->stats.errors = 0;
1005 		rxvq->stats.multicast = 0;
1006 		rxvq->stats.broadcast = 0;
1007 		memset(rxvq->stats.size_bins, 0,
1008 		       sizeof(rxvq->stats.size_bins[0]) * 8);
1009 	}
1010 }
1011 
1012 static void
1013 virtio_set_hwaddr(struct virtio_hw *hw)
1014 {
1015 	vtpci_write_dev_config(hw,
1016 			offsetof(struct virtio_net_config, mac),
1017 			&hw->mac_addr, ETHER_ADDR_LEN);
1018 }
1019 
1020 static void
1021 virtio_get_hwaddr(struct virtio_hw *hw)
1022 {
1023 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1024 		vtpci_read_dev_config(hw,
1025 			offsetof(struct virtio_net_config, mac),
1026 			&hw->mac_addr, ETHER_ADDR_LEN);
1027 	} else {
1028 		eth_random_addr(&hw->mac_addr[0]);
1029 		virtio_set_hwaddr(hw);
1030 	}
1031 }
1032 
1033 static int
1034 virtio_mac_table_set(struct virtio_hw *hw,
1035 		     const struct virtio_net_ctrl_mac *uc,
1036 		     const struct virtio_net_ctrl_mac *mc)
1037 {
1038 	struct virtio_pmd_ctrl ctrl;
1039 	int err, len[2];
1040 
1041 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1042 		PMD_DRV_LOG(INFO, "host does not support mac table");
1043 		return -1;
1044 	}
1045 
1046 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1047 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1048 
1049 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1050 	memcpy(ctrl.data, uc, len[0]);
1051 
1052 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1053 	memcpy(ctrl.data + len[0], mc, len[1]);
1054 
1055 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1056 	if (err != 0)
1057 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1058 	return err;
1059 }
1060 
1061 static int
1062 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1063 		    uint32_t index, uint32_t vmdq __rte_unused)
1064 {
1065 	struct virtio_hw *hw = dev->data->dev_private;
1066 	const struct ether_addr *addrs = dev->data->mac_addrs;
1067 	unsigned int i;
1068 	struct virtio_net_ctrl_mac *uc, *mc;
1069 
1070 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1071 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1072 		return -EINVAL;
1073 	}
1074 
1075 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1076 	uc->entries = 0;
1077 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1078 	mc->entries = 0;
1079 
1080 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1081 		const struct ether_addr *addr
1082 			= (i == index) ? mac_addr : addrs + i;
1083 		struct virtio_net_ctrl_mac *tbl
1084 			= is_multicast_ether_addr(addr) ? mc : uc;
1085 
1086 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1087 	}
1088 
1089 	return virtio_mac_table_set(hw, uc, mc);
1090 }
1091 
1092 static void
1093 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1094 {
1095 	struct virtio_hw *hw = dev->data->dev_private;
1096 	struct ether_addr *addrs = dev->data->mac_addrs;
1097 	struct virtio_net_ctrl_mac *uc, *mc;
1098 	unsigned int i;
1099 
1100 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
1101 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1102 		return;
1103 	}
1104 
1105 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1106 	uc->entries = 0;
1107 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1108 	mc->entries = 0;
1109 
1110 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1111 		struct virtio_net_ctrl_mac *tbl;
1112 
1113 		if (i == index || is_zero_ether_addr(addrs + i))
1114 			continue;
1115 
1116 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1117 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1118 	}
1119 
1120 	virtio_mac_table_set(hw, uc, mc);
1121 }
1122 
1123 static void
1124 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1125 {
1126 	struct virtio_hw *hw = dev->data->dev_private;
1127 
1128 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1129 
1130 	/* Use atomic update if available */
1131 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1132 		struct virtio_pmd_ctrl ctrl;
1133 		int len = ETHER_ADDR_LEN;
1134 
1135 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1136 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1137 
1138 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1139 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
1140 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1141 		virtio_set_hwaddr(hw);
1142 }
1143 
1144 static int
1145 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1146 {
1147 	struct virtio_hw *hw = dev->data->dev_private;
1148 	struct virtio_pmd_ctrl ctrl;
1149 	int len;
1150 
1151 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1152 		return -ENOTSUP;
1153 
1154 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1155 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1156 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1157 	len = sizeof(vlan_id);
1158 
1159 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1160 }
1161 
1162 static int
1163 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1164 {
1165 	uint64_t host_features;
1166 
1167 	/* Prepare guest_features: feature that driver wants to support */
1168 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1169 		req_features);
1170 
1171 	/* Read device(host) feature bits */
1172 	host_features = VTPCI_OPS(hw)->get_features(hw);
1173 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1174 		host_features);
1175 
1176 	/* If supported, ensure MTU value is valid before acknowledging it. */
1177 	if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1178 		struct virtio_net_config config;
1179 
1180 		vtpci_read_dev_config(hw,
1181 			offsetof(struct virtio_net_config, mtu),
1182 			&config.mtu, sizeof(config.mtu));
1183 
1184 		if (config.mtu < ETHER_MIN_MTU)
1185 			req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1186 	}
1187 
1188 	/*
1189 	 * Negotiate features: Subset of device feature bits are written back
1190 	 * guest feature bits.
1191 	 */
1192 	hw->guest_features = req_features;
1193 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
1194 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1195 		hw->guest_features);
1196 
1197 	if (hw->modern) {
1198 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1199 			PMD_INIT_LOG(ERR,
1200 				"VIRTIO_F_VERSION_1 features is not enabled.");
1201 			return -1;
1202 		}
1203 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1204 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1205 			PMD_INIT_LOG(ERR,
1206 				"failed to set FEATURES_OK status!");
1207 			return -1;
1208 		}
1209 	}
1210 
1211 	hw->req_guest_features = req_features;
1212 
1213 	return 0;
1214 }
1215 
1216 /*
1217  * Process Virtio Config changed interrupt and call the callback
1218  * if link state changed.
1219  */
1220 void
1221 virtio_interrupt_handler(void *param)
1222 {
1223 	struct rte_eth_dev *dev = param;
1224 	struct virtio_hw *hw = dev->data->dev_private;
1225 	uint8_t isr;
1226 
1227 	/* Read interrupt status which clears interrupt */
1228 	isr = vtpci_isr(hw);
1229 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1230 
1231 	if (rte_intr_enable(dev->intr_handle) < 0)
1232 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1233 
1234 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1235 		if (virtio_dev_link_update(dev, 0) == 0)
1236 			_rte_eth_dev_callback_process(dev,
1237 						      RTE_ETH_EVENT_INTR_LSC,
1238 						      NULL, NULL);
1239 	}
1240 
1241 }
1242 
1243 /* set rx and tx handlers according to what is supported */
1244 static void
1245 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1246 {
1247 	struct virtio_hw *hw = eth_dev->data->dev_private;
1248 
1249 	if (hw->use_simple_rx) {
1250 		PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1251 			eth_dev->data->port_id);
1252 		eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1253 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1254 		PMD_INIT_LOG(INFO,
1255 			"virtio: using mergeable buffer Rx path on port %u",
1256 			eth_dev->data->port_id);
1257 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1258 	} else {
1259 		PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1260 			eth_dev->data->port_id);
1261 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1262 	}
1263 
1264 	if (hw->use_simple_tx) {
1265 		PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1266 			eth_dev->data->port_id);
1267 		eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1268 	} else {
1269 		PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1270 			eth_dev->data->port_id);
1271 		eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1272 	}
1273 }
1274 
1275 /* Only support 1:1 queue/interrupt mapping so far.
1276  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1277  * interrupt vectors (<N+1).
1278  */
1279 static int
1280 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1281 {
1282 	uint32_t i;
1283 	struct virtio_hw *hw = dev->data->dev_private;
1284 
1285 	PMD_INIT_LOG(INFO, "queue/interrupt binding");
1286 	for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1287 		dev->intr_handle->intr_vec[i] = i + 1;
1288 		if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1289 						 VIRTIO_MSI_NO_VECTOR) {
1290 			PMD_DRV_LOG(ERR, "failed to set queue vector");
1291 			return -EBUSY;
1292 		}
1293 	}
1294 
1295 	return 0;
1296 }
1297 
1298 static void
1299 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1300 {
1301 	uint32_t i;
1302 	struct virtio_hw *hw = dev->data->dev_private;
1303 
1304 	PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1305 	for (i = 0; i < dev->data->nb_rx_queues; ++i)
1306 		VTPCI_OPS(hw)->set_queue_irq(hw,
1307 					     hw->vqs[i * VTNET_CQ],
1308 					     VIRTIO_MSI_NO_VECTOR);
1309 }
1310 
1311 static int
1312 virtio_configure_intr(struct rte_eth_dev *dev)
1313 {
1314 	struct virtio_hw *hw = dev->data->dev_private;
1315 
1316 	if (!rte_intr_cap_multiple(dev->intr_handle)) {
1317 		PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1318 		return -ENOTSUP;
1319 	}
1320 
1321 	if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1322 		PMD_INIT_LOG(ERR, "Fail to create eventfd");
1323 		return -1;
1324 	}
1325 
1326 	if (!dev->intr_handle->intr_vec) {
1327 		dev->intr_handle->intr_vec =
1328 			rte_zmalloc("intr_vec",
1329 				    hw->max_queue_pairs * sizeof(int), 0);
1330 		if (!dev->intr_handle->intr_vec) {
1331 			PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1332 				     hw->max_queue_pairs);
1333 			return -ENOMEM;
1334 		}
1335 	}
1336 
1337 	/* Re-register callback to update max_intr */
1338 	rte_intr_callback_unregister(dev->intr_handle,
1339 				     virtio_interrupt_handler,
1340 				     dev);
1341 	rte_intr_callback_register(dev->intr_handle,
1342 				   virtio_interrupt_handler,
1343 				   dev);
1344 
1345 	/* DO NOT try to remove this! This function will enable msix, or QEMU
1346 	 * will encounter SIGSEGV when DRIVER_OK is sent.
1347 	 * And for legacy devices, this should be done before queue/vec binding
1348 	 * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1349 	 * (22) will be ignored.
1350 	 */
1351 	if (rte_intr_enable(dev->intr_handle) < 0) {
1352 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1353 		return -1;
1354 	}
1355 
1356 	if (virtio_queues_bind_intr(dev) < 0) {
1357 		PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1358 		return -1;
1359 	}
1360 
1361 	return 0;
1362 }
1363 
1364 /* reset device and renegotiate features if needed */
1365 static int
1366 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1367 {
1368 	struct virtio_hw *hw = eth_dev->data->dev_private;
1369 	struct virtio_net_config *config;
1370 	struct virtio_net_config local_config;
1371 	struct rte_pci_device *pci_dev = NULL;
1372 	int ret;
1373 
1374 	/* Reset the device although not necessary at startup */
1375 	vtpci_reset(hw);
1376 
1377 	/* Tell the host we've noticed this device. */
1378 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1379 
1380 	/* Tell the host we've known how to drive the device. */
1381 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1382 	if (virtio_negotiate_features(hw, req_features) < 0)
1383 		return -1;
1384 
1385 	if (!hw->virtio_user_dev) {
1386 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1387 		rte_eth_copy_pci_info(eth_dev, pci_dev);
1388 	}
1389 
1390 	/* If host does not support both status and MSI-X then disable LSC */
1391 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->use_msix)
1392 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1393 	else
1394 		eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1395 
1396 	/* Setting up rx_header size for the device */
1397 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1398 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1399 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1400 	else
1401 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1402 
1403 	/* Copy the permanent MAC address to: virtio_hw */
1404 	virtio_get_hwaddr(hw);
1405 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1406 			&eth_dev->data->mac_addrs[0]);
1407 	PMD_INIT_LOG(DEBUG,
1408 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1409 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1410 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1411 
1412 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1413 		config = &local_config;
1414 
1415 		vtpci_read_dev_config(hw,
1416 			offsetof(struct virtio_net_config, mac),
1417 			&config->mac, sizeof(config->mac));
1418 
1419 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1420 			vtpci_read_dev_config(hw,
1421 				offsetof(struct virtio_net_config, status),
1422 				&config->status, sizeof(config->status));
1423 		} else {
1424 			PMD_INIT_LOG(DEBUG,
1425 				     "VIRTIO_NET_F_STATUS is not supported");
1426 			config->status = 0;
1427 		}
1428 
1429 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1430 			vtpci_read_dev_config(hw,
1431 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1432 				&config->max_virtqueue_pairs,
1433 				sizeof(config->max_virtqueue_pairs));
1434 		} else {
1435 			PMD_INIT_LOG(DEBUG,
1436 				     "VIRTIO_NET_F_MQ is not supported");
1437 			config->max_virtqueue_pairs = 1;
1438 		}
1439 
1440 		hw->max_queue_pairs = config->max_virtqueue_pairs;
1441 
1442 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1443 			vtpci_read_dev_config(hw,
1444 				offsetof(struct virtio_net_config, mtu),
1445 				&config->mtu,
1446 				sizeof(config->mtu));
1447 
1448 			/*
1449 			 * MTU value has already been checked at negotiation
1450 			 * time, but check again in case it has changed since
1451 			 * then, which should not happen.
1452 			 */
1453 			if (config->mtu < ETHER_MIN_MTU) {
1454 				PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1455 						config->mtu);
1456 				return -1;
1457 			}
1458 
1459 			hw->max_mtu = config->mtu;
1460 			/* Set initial MTU to maximum one supported by vhost */
1461 			eth_dev->data->mtu = config->mtu;
1462 
1463 		} else {
1464 			hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1465 				VLAN_TAG_LEN - hw->vtnet_hdr_size;
1466 		}
1467 
1468 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1469 				config->max_virtqueue_pairs);
1470 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1471 		PMD_INIT_LOG(DEBUG,
1472 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1473 				config->mac[0], config->mac[1],
1474 				config->mac[2], config->mac[3],
1475 				config->mac[4], config->mac[5]);
1476 	} else {
1477 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1478 		hw->max_queue_pairs = 1;
1479 	}
1480 
1481 	ret = virtio_alloc_queues(eth_dev);
1482 	if (ret < 0)
1483 		return ret;
1484 
1485 	if (eth_dev->data->dev_conf.intr_conf.rxq) {
1486 		if (virtio_configure_intr(eth_dev) < 0) {
1487 			PMD_INIT_LOG(ERR, "failed to configure interrupt");
1488 			return -1;
1489 		}
1490 	}
1491 
1492 	vtpci_reinit_complete(hw);
1493 
1494 	if (pci_dev)
1495 		PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1496 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1497 			pci_dev->id.device_id);
1498 
1499 	return 0;
1500 }
1501 
1502 /*
1503  * Remap the PCI device again (IO port map for legacy device and
1504  * memory map for modern device), so that the secondary process
1505  * could have the PCI initiated correctly.
1506  */
1507 static int
1508 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1509 {
1510 	if (hw->modern) {
1511 		/*
1512 		 * We don't have to re-parse the PCI config space, since
1513 		 * rte_pci_map_device() makes sure the mapped address
1514 		 * in secondary process would equal to the one mapped in
1515 		 * the primary process: error will be returned if that
1516 		 * requirement is not met.
1517 		 *
1518 		 * That said, we could simply reuse all cap pointers
1519 		 * (such as dev_cfg, common_cfg, etc.) parsed from the
1520 		 * primary process, which is stored in shared memory.
1521 		 */
1522 		if (rte_pci_map_device(pci_dev)) {
1523 			PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1524 			return -1;
1525 		}
1526 	} else {
1527 		if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1528 			return -1;
1529 	}
1530 
1531 	return 0;
1532 }
1533 
1534 static void
1535 virtio_set_vtpci_ops(struct virtio_hw *hw)
1536 {
1537 #ifdef RTE_VIRTIO_USER
1538 	if (hw->virtio_user_dev)
1539 		VTPCI_OPS(hw) = &virtio_user_ops;
1540 	else
1541 #endif
1542 	if (hw->modern)
1543 		VTPCI_OPS(hw) = &modern_ops;
1544 	else
1545 		VTPCI_OPS(hw) = &legacy_ops;
1546 }
1547 
1548 /*
1549  * This function is based on probe() function in virtio_pci.c
1550  * It returns 0 on success.
1551  */
1552 int
1553 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1554 {
1555 	struct virtio_hw *hw = eth_dev->data->dev_private;
1556 	int ret;
1557 
1558 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1559 
1560 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1561 
1562 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1563 		if (!hw->virtio_user_dev) {
1564 			ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1565 			if (ret)
1566 				return ret;
1567 		}
1568 
1569 		virtio_set_vtpci_ops(hw);
1570 		set_rxtx_funcs(eth_dev);
1571 
1572 		return 0;
1573 	}
1574 
1575 	/* Allocate memory for storing MAC addresses */
1576 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1577 	if (eth_dev->data->mac_addrs == NULL) {
1578 		PMD_INIT_LOG(ERR,
1579 			"Failed to allocate %d bytes needed to store MAC addresses",
1580 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1581 		return -ENOMEM;
1582 	}
1583 
1584 	hw->port_id = eth_dev->data->port_id;
1585 	/* For virtio_user case the hw->virtio_user_dev is populated by
1586 	 * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1587 	 */
1588 	if (!hw->virtio_user_dev) {
1589 		ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1590 		if (ret)
1591 			return ret;
1592 	}
1593 
1594 	/* reset device and negotiate default features */
1595 	ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1596 	if (ret < 0)
1597 		return ret;
1598 
1599 	/* Setup interrupt callback  */
1600 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1601 		rte_intr_callback_register(eth_dev->intr_handle,
1602 			virtio_interrupt_handler, eth_dev);
1603 
1604 	return 0;
1605 }
1606 
1607 static int
1608 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1609 {
1610 	PMD_INIT_FUNC_TRACE();
1611 
1612 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1613 		return -EPERM;
1614 
1615 	virtio_dev_stop(eth_dev);
1616 	virtio_dev_close(eth_dev);
1617 
1618 	eth_dev->dev_ops = NULL;
1619 	eth_dev->tx_pkt_burst = NULL;
1620 	eth_dev->rx_pkt_burst = NULL;
1621 
1622 	rte_free(eth_dev->data->mac_addrs);
1623 	eth_dev->data->mac_addrs = NULL;
1624 
1625 	/* reset interrupt callback  */
1626 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1627 		rte_intr_callback_unregister(eth_dev->intr_handle,
1628 						virtio_interrupt_handler,
1629 						eth_dev);
1630 	if (eth_dev->device)
1631 		rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1632 
1633 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1634 
1635 	return 0;
1636 }
1637 
1638 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1639 	struct rte_pci_device *pci_dev)
1640 {
1641 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1642 		eth_virtio_dev_init);
1643 }
1644 
1645 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1646 {
1647 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1648 }
1649 
1650 static struct rte_pci_driver rte_virtio_pmd = {
1651 	.driver = {
1652 		.name = "net_virtio",
1653 	},
1654 	.id_table = pci_id_virtio_map,
1655 	.drv_flags = 0,
1656 	.probe = eth_virtio_pci_probe,
1657 	.remove = eth_virtio_pci_remove,
1658 };
1659 
1660 RTE_INIT(rte_virtio_pmd_init);
1661 static void
1662 rte_virtio_pmd_init(void)
1663 {
1664 	if (rte_eal_iopl_init() != 0) {
1665 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1666 		return;
1667 	}
1668 
1669 	rte_pci_register(&rte_virtio_pmd);
1670 }
1671 
1672 /*
1673  * Configure virtio device
1674  * It returns 0 on success.
1675  */
1676 static int
1677 virtio_dev_configure(struct rte_eth_dev *dev)
1678 {
1679 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1680 	struct virtio_hw *hw = dev->data->dev_private;
1681 	uint64_t req_features;
1682 	int ret;
1683 
1684 	PMD_INIT_LOG(DEBUG, "configure");
1685 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1686 
1687 	if (dev->data->dev_conf.intr_conf.rxq) {
1688 		ret = virtio_init_device(dev, hw->req_guest_features);
1689 		if (ret < 0)
1690 			return ret;
1691 	}
1692 
1693 	/* The name hw_ip_checksum is a bit confusing since it can be
1694 	 * set by the application to request L3 and/or L4 checksums. In
1695 	 * case of virtio, only L4 checksum is supported.
1696 	 */
1697 	if (rxmode->hw_ip_checksum)
1698 		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1699 
1700 	if (rxmode->enable_lro)
1701 		req_features |=
1702 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1703 			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
1704 
1705 	/* if request features changed, reinit the device */
1706 	if (req_features != hw->req_guest_features) {
1707 		ret = virtio_init_device(dev, req_features);
1708 		if (ret < 0)
1709 			return ret;
1710 	}
1711 
1712 	if (rxmode->hw_ip_checksum &&
1713 		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1714 		PMD_DRV_LOG(ERR,
1715 			"rx checksum not available on this host");
1716 		return -ENOTSUP;
1717 	}
1718 
1719 	if (rxmode->enable_lro &&
1720 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1721 			!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1722 		PMD_DRV_LOG(ERR,
1723 			"Large Receive Offload not available on this host");
1724 		return -ENOTSUP;
1725 	}
1726 
1727 	/* start control queue */
1728 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1729 		virtio_dev_cq_start(dev);
1730 
1731 	hw->vlan_strip = rxmode->hw_vlan_strip;
1732 
1733 	if (rxmode->hw_vlan_filter
1734 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1735 		PMD_DRV_LOG(ERR,
1736 			    "vlan filtering not available on this host");
1737 		return -ENOTSUP;
1738 	}
1739 
1740 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1741 		/* Enable vector (0) for Link State Intrerrupt */
1742 		if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1743 				VIRTIO_MSI_NO_VECTOR) {
1744 			PMD_DRV_LOG(ERR, "failed to set config vector");
1745 			return -EBUSY;
1746 		}
1747 
1748 	hw->use_simple_rx = 1;
1749 	hw->use_simple_tx = 1;
1750 
1751 #if defined RTE_ARCH_ARM64 || defined CONFIG_RTE_ARCH_ARM
1752 	if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1753 		hw->use_simple_rx = 0;
1754 		hw->use_simple_tx = 0;
1755 	}
1756 #endif
1757 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1758 		hw->use_simple_rx = 0;
1759 		hw->use_simple_tx = 0;
1760 	}
1761 
1762 	if (rxmode->hw_ip_checksum)
1763 		hw->use_simple_rx = 0;
1764 
1765 	return 0;
1766 }
1767 
1768 
1769 static int
1770 virtio_dev_start(struct rte_eth_dev *dev)
1771 {
1772 	uint16_t nb_queues, i;
1773 	struct virtnet_rx *rxvq;
1774 	struct virtnet_tx *txvq __rte_unused;
1775 	struct virtio_hw *hw = dev->data->dev_private;
1776 	int ret;
1777 
1778 	/* Finish the initialization of the queues */
1779 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1780 		ret = virtio_dev_rx_queue_setup_finish(dev, i);
1781 		if (ret < 0)
1782 			return ret;
1783 	}
1784 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1785 		ret = virtio_dev_tx_queue_setup_finish(dev, i);
1786 		if (ret < 0)
1787 			return ret;
1788 	}
1789 
1790 	/* check if lsc interrupt feature is enabled */
1791 	if (dev->data->dev_conf.intr_conf.lsc) {
1792 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1793 			PMD_DRV_LOG(ERR, "link status not supported by host");
1794 			return -ENOTSUP;
1795 		}
1796 	}
1797 
1798 	/* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1799 	 * in device configure, but it could be unmapped  when device is
1800 	 * stopped.
1801 	 */
1802 	if (dev->data->dev_conf.intr_conf.lsc ||
1803 	    dev->data->dev_conf.intr_conf.rxq) {
1804 		rte_intr_disable(dev->intr_handle);
1805 
1806 		if (rte_intr_enable(dev->intr_handle) < 0) {
1807 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1808 			return -EIO;
1809 		}
1810 	}
1811 
1812 	/*Notify the backend
1813 	 *Otherwise the tap backend might already stop its queue due to fullness.
1814 	 *vhost backend will have no chance to be waked up
1815 	 */
1816 	nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1817 	if (hw->max_queue_pairs > 1) {
1818 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1819 			return -EINVAL;
1820 	}
1821 
1822 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1823 
1824 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1825 		rxvq = dev->data->rx_queues[i];
1826 		/* Flush the old packets */
1827 		virtqueue_flush(rxvq->vq);
1828 		virtqueue_notify(rxvq->vq);
1829 	}
1830 
1831 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1832 		txvq = dev->data->tx_queues[i];
1833 		virtqueue_notify(txvq->vq);
1834 	}
1835 
1836 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1837 
1838 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1839 		rxvq = dev->data->rx_queues[i];
1840 		VIRTQUEUE_DUMP(rxvq->vq);
1841 	}
1842 
1843 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1844 		txvq = dev->data->tx_queues[i];
1845 		VIRTQUEUE_DUMP(txvq->vq);
1846 	}
1847 
1848 	set_rxtx_funcs(dev);
1849 	hw->started = 1;
1850 
1851 	/* Initialize Link state */
1852 	virtio_dev_link_update(dev, 0);
1853 
1854 	return 0;
1855 }
1856 
1857 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1858 {
1859 	struct rte_mbuf *buf;
1860 	int i, mbuf_num = 0;
1861 
1862 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1863 		struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1864 
1865 		PMD_INIT_LOG(DEBUG,
1866 			     "Before freeing rxq[%d] used and unused buf", i);
1867 		VIRTQUEUE_DUMP(rxvq->vq);
1868 
1869 		PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1870 		while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1871 			rte_pktmbuf_free(buf);
1872 			mbuf_num++;
1873 		}
1874 
1875 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1876 		PMD_INIT_LOG(DEBUG,
1877 			     "After freeing rxq[%d] used and unused buf", i);
1878 		VIRTQUEUE_DUMP(rxvq->vq);
1879 	}
1880 
1881 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1882 		struct virtnet_tx *txvq = dev->data->tx_queues[i];
1883 
1884 		PMD_INIT_LOG(DEBUG,
1885 			     "Before freeing txq[%d] used and unused bufs",
1886 			     i);
1887 		VIRTQUEUE_DUMP(txvq->vq);
1888 
1889 		mbuf_num = 0;
1890 		while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1891 			rte_pktmbuf_free(buf);
1892 			mbuf_num++;
1893 		}
1894 
1895 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1896 		PMD_INIT_LOG(DEBUG,
1897 			     "After freeing txq[%d] used and unused buf", i);
1898 		VIRTQUEUE_DUMP(txvq->vq);
1899 	}
1900 }
1901 
1902 /*
1903  * Stop device: disable interrupt and mark link down
1904  */
1905 static void
1906 virtio_dev_stop(struct rte_eth_dev *dev)
1907 {
1908 	struct virtio_hw *hw = dev->data->dev_private;
1909 	struct rte_eth_link link;
1910 	struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1911 
1912 	PMD_INIT_LOG(DEBUG, "stop");
1913 
1914 	if (intr_conf->lsc || intr_conf->rxq)
1915 		rte_intr_disable(dev->intr_handle);
1916 
1917 	hw->started = 0;
1918 	memset(&link, 0, sizeof(link));
1919 	virtio_dev_atomic_write_link_status(dev, &link);
1920 }
1921 
1922 static int
1923 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1924 {
1925 	struct rte_eth_link link, old;
1926 	uint16_t status;
1927 	struct virtio_hw *hw = dev->data->dev_private;
1928 	memset(&link, 0, sizeof(link));
1929 	virtio_dev_atomic_read_link_status(dev, &link);
1930 	old = link;
1931 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1932 	link.link_speed  = SPEED_10G;
1933 
1934 	if (hw->started == 0) {
1935 		link.link_status = ETH_LINK_DOWN;
1936 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1937 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
1938 		vtpci_read_dev_config(hw,
1939 				offsetof(struct virtio_net_config, status),
1940 				&status, sizeof(status));
1941 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1942 			link.link_status = ETH_LINK_DOWN;
1943 			PMD_INIT_LOG(DEBUG, "Port %d is down",
1944 				     dev->data->port_id);
1945 		} else {
1946 			link.link_status = ETH_LINK_UP;
1947 			PMD_INIT_LOG(DEBUG, "Port %d is up",
1948 				     dev->data->port_id);
1949 		}
1950 	} else {
1951 		link.link_status = ETH_LINK_UP;
1952 	}
1953 	virtio_dev_atomic_write_link_status(dev, &link);
1954 
1955 	return (old.link_status == link.link_status) ? -1 : 0;
1956 }
1957 
1958 static int
1959 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1960 {
1961 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1962 	struct virtio_hw *hw = dev->data->dev_private;
1963 
1964 	if (mask & ETH_VLAN_FILTER_MASK) {
1965 		if (rxmode->hw_vlan_filter &&
1966 				!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1967 
1968 			PMD_DRV_LOG(NOTICE,
1969 				"vlan filtering not available on this host");
1970 
1971 			return -ENOTSUP;
1972 		}
1973 	}
1974 
1975 	if (mask & ETH_VLAN_STRIP_MASK)
1976 		hw->vlan_strip = rxmode->hw_vlan_strip;
1977 
1978 	return 0;
1979 }
1980 
1981 static void
1982 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1983 {
1984 	uint64_t tso_mask, host_features;
1985 	struct virtio_hw *hw = dev->data->dev_private;
1986 
1987 	dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
1988 
1989 	dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
1990 	dev_info->max_rx_queues =
1991 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1992 	dev_info->max_tx_queues =
1993 		RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1994 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1995 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1996 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1997 	dev_info->default_txconf = (struct rte_eth_txconf) {
1998 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1999 	};
2000 
2001 	host_features = VTPCI_OPS(hw)->get_features(hw);
2002 	dev_info->rx_offload_capa = 0;
2003 	if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2004 		dev_info->rx_offload_capa |=
2005 			DEV_RX_OFFLOAD_TCP_CKSUM |
2006 			DEV_RX_OFFLOAD_UDP_CKSUM;
2007 	}
2008 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2009 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
2010 	if ((host_features & tso_mask) == tso_mask)
2011 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2012 
2013 	dev_info->tx_offload_capa = 0;
2014 	if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2015 		dev_info->tx_offload_capa |=
2016 			DEV_TX_OFFLOAD_UDP_CKSUM |
2017 			DEV_TX_OFFLOAD_TCP_CKSUM;
2018 	}
2019 	tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2020 		(1ULL << VIRTIO_NET_F_HOST_TSO6);
2021 	if ((hw->guest_features & tso_mask) == tso_mask)
2022 		dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2023 }
2024 
2025 /*
2026  * It enables testpmd to collect per queue stats.
2027  */
2028 static int
2029 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2030 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2031 __rte_unused uint8_t is_rx)
2032 {
2033 	return 0;
2034 }
2035 
2036 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2037 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2038 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2039