xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision 0499793854f52d0a42aa39a6b8036700f3719735)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include <stdint.h>
35 #include <string.h>
36 #include <stdio.h>
37 #include <errno.h>
38 #include <unistd.h>
39 
40 #include <rte_ethdev.h>
41 #include <rte_memcpy.h>
42 #include <rte_string_fns.h>
43 #include <rte_memzone.h>
44 #include <rte_malloc.h>
45 #include <rte_atomic.h>
46 #include <rte_branch_prediction.h>
47 #include <rte_pci.h>
48 #include <rte_ether.h>
49 #include <rte_common.h>
50 #include <rte_errno.h>
51 
52 #include <rte_memory.h>
53 #include <rte_eal.h>
54 #include <rte_dev.h>
55 
56 #include "virtio_ethdev.h"
57 #include "virtio_pci.h"
58 #include "virtio_logs.h"
59 #include "virtqueue.h"
60 #include "virtio_rxtx.h"
61 
62 static int eth_virtio_dev_init(struct rte_eth_dev *eth_dev);
63 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
64 static int  virtio_dev_configure(struct rte_eth_dev *dev);
65 static int  virtio_dev_start(struct rte_eth_dev *dev);
66 static void virtio_dev_stop(struct rte_eth_dev *dev);
67 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
68 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
69 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
70 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
71 static void virtio_dev_info_get(struct rte_eth_dev *dev,
72 				struct rte_eth_dev_info *dev_info);
73 static int virtio_dev_link_update(struct rte_eth_dev *dev,
74 	__rte_unused int wait_to_complete);
75 
76 static void virtio_set_hwaddr(struct virtio_hw *hw);
77 static void virtio_get_hwaddr(struct virtio_hw *hw);
78 
79 static void virtio_dev_stats_get(struct rte_eth_dev *dev,
80 				 struct rte_eth_stats *stats);
81 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
82 				 struct rte_eth_xstats *xstats, unsigned n);
83 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
84 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
85 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
86 				uint16_t vlan_id, int on);
87 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
88 				struct ether_addr *mac_addr,
89 				uint32_t index, uint32_t vmdq __rte_unused);
90 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
91 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
92 				struct ether_addr *mac_addr);
93 
94 static int virtio_dev_queue_stats_mapping_set(
95 	__rte_unused struct rte_eth_dev *eth_dev,
96 	__rte_unused uint16_t queue_id,
97 	__rte_unused uint8_t stat_idx,
98 	__rte_unused uint8_t is_rx);
99 
100 /*
101  * The set of PCI devices this driver supports
102  */
103 static const struct rte_pci_id pci_id_virtio_map[] = {
104 
105 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
106 #include "rte_pci_dev_ids.h"
107 
108 { .vendor_id = 0, /* sentinel */ },
109 };
110 
111 struct rte_virtio_xstats_name_off {
112 	char name[RTE_ETH_XSTATS_NAME_SIZE];
113 	unsigned offset;
114 };
115 
116 /* [rt]x_qX_ is prepended to the name string here */
117 static const struct rte_virtio_xstats_name_off rte_virtio_q_stat_strings[] = {
118 	{"good_packets",           offsetof(struct virtqueue, packets)},
119 	{"good_bytes",             offsetof(struct virtqueue, bytes)},
120 	{"errors",                 offsetof(struct virtqueue, errors)},
121 	{"multicast_packets",      offsetof(struct virtqueue, multicast)},
122 	{"broadcast_packets",      offsetof(struct virtqueue, broadcast)},
123 	{"undersize_packets",      offsetof(struct virtqueue, size_bins[0])},
124 	{"size_64_packets",        offsetof(struct virtqueue, size_bins[1])},
125 	{"size_65_127_packets",    offsetof(struct virtqueue, size_bins[2])},
126 	{"size_128_255_packets",   offsetof(struct virtqueue, size_bins[3])},
127 	{"size_256_511_packets",   offsetof(struct virtqueue, size_bins[4])},
128 	{"size_512_1023_packets",  offsetof(struct virtqueue, size_bins[5])},
129 	{"size_1024_1517_packets", offsetof(struct virtqueue, size_bins[6])},
130 	{"size_1518_max_packets",  offsetof(struct virtqueue, size_bins[7])},
131 };
132 
133 #define VIRTIO_NB_Q_XSTATS (sizeof(rte_virtio_q_stat_strings) / \
134 			    sizeof(rte_virtio_q_stat_strings[0]))
135 
136 static int
137 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
138 		int *dlen, int pkt_num)
139 {
140 	uint32_t head, i;
141 	int k, sum = 0;
142 	virtio_net_ctrl_ack status = ~0;
143 	struct virtio_pmd_ctrl result;
144 
145 	ctrl->status = status;
146 
147 	if (!(vq && vq->hw->cvq)) {
148 		PMD_INIT_LOG(ERR, "Control queue is not supported.");
149 		return -1;
150 	}
151 	head = vq->vq_desc_head_idx;
152 
153 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
154 		"vq->hw->cvq = %p vq = %p",
155 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
156 
157 	if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
158 		return -1;
159 
160 	memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
161 		sizeof(struct virtio_pmd_ctrl));
162 
163 	/*
164 	 * Format is enforced in qemu code:
165 	 * One TX packet for header;
166 	 * At least one TX packet per argument;
167 	 * One RX packet for ACK.
168 	 */
169 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
170 	vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
171 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
172 	vq->vq_free_cnt--;
173 	i = vq->vq_ring.desc[head].next;
174 
175 	for (k = 0; k < pkt_num; k++) {
176 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
177 		vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
178 			+ sizeof(struct virtio_net_ctrl_hdr)
179 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
180 		vq->vq_ring.desc[i].len = dlen[k];
181 		sum += dlen[k];
182 		vq->vq_free_cnt--;
183 		i = vq->vq_ring.desc[i].next;
184 	}
185 
186 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
187 	vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
188 			+ sizeof(struct virtio_net_ctrl_hdr);
189 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
190 	vq->vq_free_cnt--;
191 
192 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
193 
194 	vq_update_avail_ring(vq, head);
195 	vq_update_avail_idx(vq);
196 
197 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
198 
199 	virtqueue_notify(vq);
200 
201 	rte_rmb();
202 	while (vq->vq_used_cons_idx == vq->vq_ring.used->idx) {
203 		rte_rmb();
204 		usleep(100);
205 	}
206 
207 	while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
208 		uint32_t idx, desc_idx, used_idx;
209 		struct vring_used_elem *uep;
210 
211 		used_idx = (uint32_t)(vq->vq_used_cons_idx
212 				& (vq->vq_nentries - 1));
213 		uep = &vq->vq_ring.used->ring[used_idx];
214 		idx = (uint32_t) uep->id;
215 		desc_idx = idx;
216 
217 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
218 			desc_idx = vq->vq_ring.desc[desc_idx].next;
219 			vq->vq_free_cnt++;
220 		}
221 
222 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
223 		vq->vq_desc_head_idx = idx;
224 
225 		vq->vq_used_cons_idx++;
226 		vq->vq_free_cnt++;
227 	}
228 
229 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
230 			vq->vq_free_cnt, vq->vq_desc_head_idx);
231 
232 	memcpy(&result, vq->virtio_net_hdr_mz->addr,
233 			sizeof(struct virtio_pmd_ctrl));
234 
235 	return result.status;
236 }
237 
238 static int
239 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
240 {
241 	struct virtio_hw *hw = dev->data->dev_private;
242 	struct virtio_pmd_ctrl ctrl;
243 	int dlen[1];
244 	int ret;
245 
246 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
247 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
248 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
249 
250 	dlen[0] = sizeof(uint16_t);
251 
252 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
253 	if (ret) {
254 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
255 			  "failed, this is too late now...");
256 		return -EINVAL;
257 	}
258 
259 	return 0;
260 }
261 
262 void
263 virtio_dev_queue_release(struct virtqueue *vq)
264 {
265 	struct virtio_hw *hw;
266 
267 	if (vq) {
268 		hw = vq->hw;
269 		if (vq->configured)
270 			hw->vtpci_ops->del_queue(hw, vq);
271 
272 		rte_memzone_free(vq->mz);
273 		if (vq->virtio_net_hdr_mz)
274 			rte_memzone_free(vq->virtio_net_hdr_mz);
275 
276 		rte_free(vq->sw_ring);
277 		rte_free(vq);
278 	}
279 }
280 
281 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
282 			int queue_type,
283 			uint16_t queue_idx,
284 			uint16_t vtpci_queue_idx,
285 			uint16_t nb_desc,
286 			unsigned int socket_id,
287 			struct virtqueue **pvq)
288 {
289 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
290 	const struct rte_memzone *mz;
291 	unsigned int vq_size, size;
292 	struct virtio_hw *hw = dev->data->dev_private;
293 	struct virtqueue *vq = NULL;
294 	const char *queue_names[] = {"rvq", "txq", "cvq"};
295 
296 	PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
297 
298 	/*
299 	 * Read the virtqueue size from the Queue Size field
300 	 * Always power of 2 and if 0 virtqueue does not exist
301 	 */
302 	vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx);
303 	PMD_INIT_LOG(DEBUG, "vq_size: %u nb_desc:%u", vq_size, nb_desc);
304 	if (vq_size == 0) {
305 		PMD_INIT_LOG(ERR, "virtqueue does not exist");
306 		return -EINVAL;
307 	}
308 
309 	if (!rte_is_power_of_2(vq_size)) {
310 		PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
311 		return -EINVAL;
312 	}
313 
314 	snprintf(vq_name, sizeof(vq_name), "port%d_%s%d",
315 		 dev->data->port_id, queue_names[queue_type], queue_idx);
316 	vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
317 			 vq_size * sizeof(struct vq_desc_extra),
318 			 RTE_CACHE_LINE_SIZE);
319 	if (vq == NULL) {
320 		PMD_INIT_LOG(ERR, "Can not allocate virtqueue");
321 		return -ENOMEM;
322 	}
323 
324 	if (queue_type == VTNET_RQ) {
325 		size_t sz_sw;
326 
327 		sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
328 			sizeof(vq->sw_ring[0]);
329 		vq->sw_ring = rte_zmalloc_socket("rxq->sw_ring", sz_sw,
330 						 RTE_CACHE_LINE_SIZE,
331 						 socket_id);
332 		if (!vq->sw_ring) {
333 			PMD_INIT_LOG(ERR, "Can not allocate RX soft ring");
334 			virtio_dev_queue_release(vq);
335 			return -ENOMEM;
336 		}
337 	}
338 
339 	vq->hw = hw;
340 	vq->port_id = dev->data->port_id;
341 	vq->queue_id = queue_idx;
342 	vq->vq_queue_index = vtpci_queue_idx;
343 	vq->vq_nentries = vq_size;
344 
345 	if (nb_desc == 0 || nb_desc > vq_size)
346 		nb_desc = vq_size;
347 	vq->vq_free_cnt = nb_desc;
348 
349 	/*
350 	 * Reserve a memzone for vring elements
351 	 */
352 	size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
353 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
354 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
355 
356 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
357 		socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
358 	if (mz == NULL) {
359 		if (rte_errno == EEXIST)
360 			mz = rte_memzone_lookup(vq_name);
361 		if (mz == NULL) {
362 			virtio_dev_queue_release(vq);
363 			return -ENOMEM;
364 		}
365 	}
366 
367 	/*
368 	 * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
369 	 * and only accepts 32 bit page frame number.
370 	 * Check if the allocated physical memory exceeds 16TB.
371 	 */
372 	if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
373 		PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
374 		virtio_dev_queue_release(vq);
375 		return -ENOMEM;
376 	}
377 
378 	memset(mz->addr, 0, sizeof(mz->len));
379 	vq->mz = mz;
380 	vq->vq_ring_mem = mz->phys_addr;
381 	vq->vq_ring_virt_mem = mz->addr;
382 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%"PRIx64, (uint64_t)mz->phys_addr);
383 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)(uintptr_t)mz->addr);
384 	vq->virtio_net_hdr_mz  = NULL;
385 	vq->virtio_net_hdr_mem = 0;
386 
387 	if (queue_type == VTNET_TQ) {
388 		const struct rte_memzone *hdr_mz;
389 		struct virtio_tx_region *txr;
390 		unsigned int i;
391 
392 		/*
393 		 * For each xmit packet, allocate a virtio_net_hdr
394 		 * and indirect ring elements
395 		 */
396 		snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
397 			 dev->data->port_id, queue_idx);
398 		hdr_mz = rte_memzone_reserve_aligned(vq_name,
399 						     vq_size * sizeof(*txr),
400 						     socket_id, 0,
401 						     RTE_CACHE_LINE_SIZE);
402 		if (hdr_mz == NULL) {
403 			if (rte_errno == EEXIST)
404 				hdr_mz = rte_memzone_lookup(vq_name);
405 			if (hdr_mz == NULL) {
406 				virtio_dev_queue_release(vq);
407 				return -ENOMEM;
408 			}
409 		}
410 		vq->virtio_net_hdr_mz = hdr_mz;
411 		vq->virtio_net_hdr_mem = hdr_mz->phys_addr;
412 
413 		txr = hdr_mz->addr;
414 		memset(txr, 0, vq_size * sizeof(*txr));
415 		for (i = 0; i < vq_size; i++) {
416 			struct vring_desc *start_dp = txr[i].tx_indir;
417 
418 			vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
419 
420 			/* first indirect descriptor is always the tx header */
421 			start_dp->addr = vq->virtio_net_hdr_mem
422 				+ i * sizeof(*txr)
423 				+ offsetof(struct virtio_tx_region, tx_hdr);
424 
425 			start_dp->len = vq->hw->vtnet_hdr_size;
426 			start_dp->flags = VRING_DESC_F_NEXT;
427 		}
428 
429 	} else if (queue_type == VTNET_CQ) {
430 		/* Allocate a page for control vq command, data and status */
431 		snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
432 			dev->data->port_id);
433 		vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
434 			PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
435 		if (vq->virtio_net_hdr_mz == NULL) {
436 			if (rte_errno == EEXIST)
437 				vq->virtio_net_hdr_mz =
438 					rte_memzone_lookup(vq_name);
439 			if (vq->virtio_net_hdr_mz == NULL) {
440 				virtio_dev_queue_release(vq);
441 				return -ENOMEM;
442 			}
443 		}
444 		vq->virtio_net_hdr_mem =
445 			vq->virtio_net_hdr_mz->phys_addr;
446 		memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
447 	}
448 
449 	hw->vtpci_ops->setup_queue(hw, vq);
450 
451 	vq->configured = 1;
452 	*pvq = vq;
453 	return 0;
454 }
455 
456 static int
457 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
458 		uint32_t socket_id)
459 {
460 	struct virtqueue *vq;
461 	int ret;
462 	struct virtio_hw *hw = dev->data->dev_private;
463 
464 	PMD_INIT_FUNC_TRACE();
465 	ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
466 			vtpci_queue_idx, 0, socket_id, &vq);
467 	if (ret < 0) {
468 		PMD_INIT_LOG(ERR, "control vq initialization failed");
469 		return ret;
470 	}
471 
472 	hw->cvq = vq;
473 	return 0;
474 }
475 
476 static void
477 virtio_free_queues(struct rte_eth_dev *dev)
478 {
479 	unsigned int i;
480 
481 	for (i = 0; i < dev->data->nb_rx_queues; i++)
482 		virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
483 
484 	dev->data->nb_rx_queues = 0;
485 
486 	for (i = 0; i < dev->data->nb_tx_queues; i++)
487 		virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
488 
489 	dev->data->nb_tx_queues = 0;
490 }
491 
492 static void
493 virtio_dev_close(struct rte_eth_dev *dev)
494 {
495 	struct virtio_hw *hw = dev->data->dev_private;
496 
497 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
498 
499 	if (hw->started == 1)
500 		virtio_dev_stop(dev);
501 
502 	/* reset the NIC */
503 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
504 		vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
505 	vtpci_reset(hw);
506 	virtio_dev_free_mbufs(dev);
507 	virtio_free_queues(dev);
508 }
509 
510 static void
511 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
512 {
513 	struct virtio_hw *hw = dev->data->dev_private;
514 	struct virtio_pmd_ctrl ctrl;
515 	int dlen[1];
516 	int ret;
517 
518 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
519 		PMD_INIT_LOG(INFO, "host does not support rx control\n");
520 		return;
521 	}
522 
523 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
524 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
525 	ctrl.data[0] = 1;
526 	dlen[0] = 1;
527 
528 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
529 	if (ret)
530 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
531 }
532 
533 static void
534 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
535 {
536 	struct virtio_hw *hw = dev->data->dev_private;
537 	struct virtio_pmd_ctrl ctrl;
538 	int dlen[1];
539 	int ret;
540 
541 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
542 		PMD_INIT_LOG(INFO, "host does not support rx control\n");
543 		return;
544 	}
545 
546 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
547 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
548 	ctrl.data[0] = 0;
549 	dlen[0] = 1;
550 
551 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
552 	if (ret)
553 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
554 }
555 
556 static void
557 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
558 {
559 	struct virtio_hw *hw = dev->data->dev_private;
560 	struct virtio_pmd_ctrl ctrl;
561 	int dlen[1];
562 	int ret;
563 
564 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
565 		PMD_INIT_LOG(INFO, "host does not support rx control\n");
566 		return;
567 	}
568 
569 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
570 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
571 	ctrl.data[0] = 1;
572 	dlen[0] = 1;
573 
574 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
575 	if (ret)
576 		PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
577 }
578 
579 static void
580 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
581 {
582 	struct virtio_hw *hw = dev->data->dev_private;
583 	struct virtio_pmd_ctrl ctrl;
584 	int dlen[1];
585 	int ret;
586 
587 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
588 		PMD_INIT_LOG(INFO, "host does not support rx control\n");
589 		return;
590 	}
591 
592 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
593 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
594 	ctrl.data[0] = 0;
595 	dlen[0] = 1;
596 
597 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
598 	if (ret)
599 		PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
600 }
601 
602 /*
603  * dev_ops for virtio, bare necessities for basic operation
604  */
605 static const struct eth_dev_ops virtio_eth_dev_ops = {
606 	.dev_configure           = virtio_dev_configure,
607 	.dev_start               = virtio_dev_start,
608 	.dev_stop                = virtio_dev_stop,
609 	.dev_close               = virtio_dev_close,
610 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
611 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
612 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
613 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
614 
615 	.dev_infos_get           = virtio_dev_info_get,
616 	.stats_get               = virtio_dev_stats_get,
617 	.xstats_get              = virtio_dev_xstats_get,
618 	.stats_reset             = virtio_dev_stats_reset,
619 	.xstats_reset            = virtio_dev_stats_reset,
620 	.link_update             = virtio_dev_link_update,
621 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
622 	.rx_queue_release        = virtio_dev_rx_queue_release,
623 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
624 	.tx_queue_release        = virtio_dev_tx_queue_release,
625 	/* collect stats per queue */
626 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
627 	.vlan_filter_set         = virtio_vlan_filter_set,
628 	.mac_addr_add            = virtio_mac_addr_add,
629 	.mac_addr_remove         = virtio_mac_addr_remove,
630 	.mac_addr_set            = virtio_mac_addr_set,
631 };
632 
633 static inline int
634 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
635 				struct rte_eth_link *link)
636 {
637 	struct rte_eth_link *dst = link;
638 	struct rte_eth_link *src = &(dev->data->dev_link);
639 
640 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
641 			*(uint64_t *)src) == 0)
642 		return -1;
643 
644 	return 0;
645 }
646 
647 /**
648  * Atomically writes the link status information into global
649  * structure rte_eth_dev.
650  *
651  * @param dev
652  *   - Pointer to the structure rte_eth_dev to read from.
653  *   - Pointer to the buffer to be saved with the link status.
654  *
655  * @return
656  *   - On success, zero.
657  *   - On failure, negative value.
658  */
659 static inline int
660 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
661 		struct rte_eth_link *link)
662 {
663 	struct rte_eth_link *dst = &(dev->data->dev_link);
664 	struct rte_eth_link *src = link;
665 
666 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
667 					*(uint64_t *)src) == 0)
668 		return -1;
669 
670 	return 0;
671 }
672 
673 static void
674 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
675 {
676 	unsigned i;
677 
678 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
679 		const struct virtqueue *txvq = dev->data->tx_queues[i];
680 		if (txvq == NULL)
681 			continue;
682 
683 		stats->opackets += txvq->packets;
684 		stats->obytes += txvq->bytes;
685 		stats->oerrors += txvq->errors;
686 
687 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
688 			stats->q_opackets[i] = txvq->packets;
689 			stats->q_obytes[i] = txvq->bytes;
690 		}
691 	}
692 
693 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
694 		const struct virtqueue *rxvq = dev->data->rx_queues[i];
695 		if (rxvq == NULL)
696 			continue;
697 
698 		stats->ipackets += rxvq->packets;
699 		stats->ibytes += rxvq->bytes;
700 		stats->ierrors += rxvq->errors;
701 
702 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
703 			stats->q_ipackets[i] = rxvq->packets;
704 			stats->q_ibytes[i] = rxvq->bytes;
705 		}
706 	}
707 
708 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
709 }
710 
711 static int
712 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstats *xstats,
713 		      unsigned n)
714 {
715 	unsigned i;
716 	unsigned count = 0;
717 
718 	unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_Q_XSTATS +
719 		dev->data->nb_rx_queues * VIRTIO_NB_Q_XSTATS;
720 
721 	if (n < nstats)
722 		return nstats;
723 
724 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
725 		struct virtqueue *rxvq = dev->data->rx_queues[i];
726 
727 		if (rxvq == NULL)
728 			continue;
729 
730 		unsigned t;
731 
732 		for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
733 			snprintf(xstats[count].name, sizeof(xstats[count].name),
734 				 "rx_q%u_%s", i,
735 				 rte_virtio_q_stat_strings[t].name);
736 			xstats[count].value = *(uint64_t *)(((char *)rxvq) +
737 				rte_virtio_q_stat_strings[t].offset);
738 			count++;
739 		}
740 	}
741 
742 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
743 		struct virtqueue *txvq = dev->data->tx_queues[i];
744 
745 		if (txvq == NULL)
746 			continue;
747 
748 		unsigned t;
749 
750 		for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
751 			snprintf(xstats[count].name, sizeof(xstats[count].name),
752 				 "tx_q%u_%s", i,
753 				 rte_virtio_q_stat_strings[t].name);
754 			xstats[count].value = *(uint64_t *)(((char *)txvq) +
755 				rte_virtio_q_stat_strings[t].offset);
756 			count++;
757 		}
758 	}
759 
760 	return count;
761 }
762 
763 static void
764 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
765 {
766 	virtio_update_stats(dev, stats);
767 }
768 
769 static void
770 virtio_dev_stats_reset(struct rte_eth_dev *dev)
771 {
772 	unsigned int i;
773 
774 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
775 		struct virtqueue *txvq = dev->data->tx_queues[i];
776 		if (txvq == NULL)
777 			continue;
778 
779 		txvq->packets = 0;
780 		txvq->bytes = 0;
781 		txvq->errors = 0;
782 		txvq->multicast = 0;
783 		txvq->broadcast = 0;
784 		memset(txvq->size_bins, 0, sizeof(txvq->size_bins[0]) * 8);
785 	}
786 
787 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
788 		struct virtqueue *rxvq = dev->data->rx_queues[i];
789 		if (rxvq == NULL)
790 			continue;
791 
792 		rxvq->packets = 0;
793 		rxvq->bytes = 0;
794 		rxvq->errors = 0;
795 		rxvq->multicast = 0;
796 		rxvq->broadcast = 0;
797 		memset(rxvq->size_bins, 0, sizeof(rxvq->size_bins[0]) * 8);
798 	}
799 }
800 
801 static void
802 virtio_set_hwaddr(struct virtio_hw *hw)
803 {
804 	vtpci_write_dev_config(hw,
805 			offsetof(struct virtio_net_config, mac),
806 			&hw->mac_addr, ETHER_ADDR_LEN);
807 }
808 
809 static void
810 virtio_get_hwaddr(struct virtio_hw *hw)
811 {
812 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
813 		vtpci_read_dev_config(hw,
814 			offsetof(struct virtio_net_config, mac),
815 			&hw->mac_addr, ETHER_ADDR_LEN);
816 	} else {
817 		eth_random_addr(&hw->mac_addr[0]);
818 		virtio_set_hwaddr(hw);
819 	}
820 }
821 
822 static void
823 virtio_mac_table_set(struct virtio_hw *hw,
824 		     const struct virtio_net_ctrl_mac *uc,
825 		     const struct virtio_net_ctrl_mac *mc)
826 {
827 	struct virtio_pmd_ctrl ctrl;
828 	int err, len[2];
829 
830 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
831 		PMD_DRV_LOG(INFO, "host does not support mac table");
832 		return;
833 	}
834 
835 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
836 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
837 
838 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
839 	memcpy(ctrl.data, uc, len[0]);
840 
841 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
842 	memcpy(ctrl.data + len[0], mc, len[1]);
843 
844 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
845 	if (err != 0)
846 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
847 }
848 
849 static void
850 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
851 		    uint32_t index, uint32_t vmdq __rte_unused)
852 {
853 	struct virtio_hw *hw = dev->data->dev_private;
854 	const struct ether_addr *addrs = dev->data->mac_addrs;
855 	unsigned int i;
856 	struct virtio_net_ctrl_mac *uc, *mc;
857 
858 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
859 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
860 		return;
861 	}
862 
863 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
864 	uc->entries = 0;
865 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
866 	mc->entries = 0;
867 
868 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
869 		const struct ether_addr *addr
870 			= (i == index) ? mac_addr : addrs + i;
871 		struct virtio_net_ctrl_mac *tbl
872 			= is_multicast_ether_addr(addr) ? mc : uc;
873 
874 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
875 	}
876 
877 	virtio_mac_table_set(hw, uc, mc);
878 }
879 
880 static void
881 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
882 {
883 	struct virtio_hw *hw = dev->data->dev_private;
884 	struct ether_addr *addrs = dev->data->mac_addrs;
885 	struct virtio_net_ctrl_mac *uc, *mc;
886 	unsigned int i;
887 
888 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
889 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
890 		return;
891 	}
892 
893 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
894 	uc->entries = 0;
895 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
896 	mc->entries = 0;
897 
898 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
899 		struct virtio_net_ctrl_mac *tbl;
900 
901 		if (i == index || is_zero_ether_addr(addrs + i))
902 			continue;
903 
904 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
905 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
906 	}
907 
908 	virtio_mac_table_set(hw, uc, mc);
909 }
910 
911 static void
912 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
913 {
914 	struct virtio_hw *hw = dev->data->dev_private;
915 
916 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
917 
918 	/* Use atomic update if available */
919 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
920 		struct virtio_pmd_ctrl ctrl;
921 		int len = ETHER_ADDR_LEN;
922 
923 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
924 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
925 
926 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
927 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
928 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
929 		virtio_set_hwaddr(hw);
930 }
931 
932 static int
933 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
934 {
935 	struct virtio_hw *hw = dev->data->dev_private;
936 	struct virtio_pmd_ctrl ctrl;
937 	int len;
938 
939 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
940 		return -ENOTSUP;
941 
942 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
943 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
944 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
945 	len = sizeof(vlan_id);
946 
947 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
948 }
949 
950 static int
951 virtio_negotiate_features(struct virtio_hw *hw)
952 {
953 	uint64_t host_features;
954 
955 	/* Prepare guest_features: feature that driver wants to support */
956 	hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
957 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
958 		hw->guest_features);
959 
960 	/* Read device(host) feature bits */
961 	host_features = hw->vtpci_ops->get_features(hw);
962 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
963 		host_features);
964 
965 	/*
966 	 * Negotiate features: Subset of device feature bits are written back
967 	 * guest feature bits.
968 	 */
969 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
970 	PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
971 		hw->guest_features);
972 
973 	if (hw->modern) {
974 		if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
975 			PMD_INIT_LOG(ERR,
976 				"VIRTIO_F_VERSION_1 features is not enabled.");
977 			return -1;
978 		}
979 		vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
980 		if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
981 			PMD_INIT_LOG(ERR,
982 				"failed to set FEATURES_OK status!");
983 			return -1;
984 		}
985 	}
986 
987 	return 0;
988 }
989 
990 /*
991  * Process Virtio Config changed interrupt and call the callback
992  * if link state changed.
993  */
994 static void
995 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
996 			 void *param)
997 {
998 	struct rte_eth_dev *dev = param;
999 	struct virtio_hw *hw = dev->data->dev_private;
1000 	uint8_t isr;
1001 
1002 	/* Read interrupt status which clears interrupt */
1003 	isr = vtpci_isr(hw);
1004 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1005 
1006 	if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1007 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1008 
1009 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1010 		if (virtio_dev_link_update(dev, 0) == 0)
1011 			_rte_eth_dev_callback_process(dev,
1012 						      RTE_ETH_EVENT_INTR_LSC);
1013 	}
1014 
1015 }
1016 
1017 static void
1018 rx_func_get(struct rte_eth_dev *eth_dev)
1019 {
1020 	struct virtio_hw *hw = eth_dev->data->dev_private;
1021 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1022 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1023 	else
1024 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1025 }
1026 
1027 /*
1028  * This function is based on probe() function in virtio_pci.c
1029  * It returns 0 on success.
1030  */
1031 static int
1032 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1033 {
1034 	struct virtio_hw *hw = eth_dev->data->dev_private;
1035 	struct virtio_net_config *config;
1036 	struct virtio_net_config local_config;
1037 	struct rte_pci_device *pci_dev;
1038 	uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1039 	int ret;
1040 
1041 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1042 
1043 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1044 	eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1045 
1046 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1047 		rx_func_get(eth_dev);
1048 		return 0;
1049 	}
1050 
1051 	/* Allocate memory for storing MAC addresses */
1052 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1053 	if (eth_dev->data->mac_addrs == NULL) {
1054 		PMD_INIT_LOG(ERR,
1055 			"Failed to allocate %d bytes needed to store MAC addresses",
1056 			VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1057 		return -ENOMEM;
1058 	}
1059 
1060 	pci_dev = eth_dev->pci_dev;
1061 
1062 	ret = vtpci_init(pci_dev, hw, &dev_flags);
1063 	if (ret)
1064 		return ret;
1065 
1066 	/* Reset the device although not necessary at startup */
1067 	vtpci_reset(hw);
1068 
1069 	/* Tell the host we've noticed this device. */
1070 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1071 
1072 	/* Tell the host we've known how to drive the device. */
1073 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1074 	if (virtio_negotiate_features(hw) < 0)
1075 		return -1;
1076 
1077 	/* If host does not support status then disable LSC */
1078 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1079 		dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1080 
1081 	rte_eth_copy_pci_info(eth_dev, pci_dev);
1082 	eth_dev->data->dev_flags = dev_flags;
1083 
1084 	rx_func_get(eth_dev);
1085 
1086 	/* Setting up rx_header size for the device */
1087 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1088 	    vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1089 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1090 	else
1091 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1092 
1093 	/* Copy the permanent MAC address to: virtio_hw */
1094 	virtio_get_hwaddr(hw);
1095 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1096 			&eth_dev->data->mac_addrs[0]);
1097 	PMD_INIT_LOG(DEBUG,
1098 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1099 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1100 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1101 
1102 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1103 		config = &local_config;
1104 
1105 		vtpci_read_dev_config(hw,
1106 			offsetof(struct virtio_net_config, mac),
1107 			&config->mac, sizeof(config->mac));
1108 
1109 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1110 			vtpci_read_dev_config(hw,
1111 				offsetof(struct virtio_net_config, status),
1112 				&config->status, sizeof(config->status));
1113 		} else {
1114 			PMD_INIT_LOG(DEBUG,
1115 				     "VIRTIO_NET_F_STATUS is not supported");
1116 			config->status = 0;
1117 		}
1118 
1119 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1120 			vtpci_read_dev_config(hw,
1121 				offsetof(struct virtio_net_config, max_virtqueue_pairs),
1122 				&config->max_virtqueue_pairs,
1123 				sizeof(config->max_virtqueue_pairs));
1124 		} else {
1125 			PMD_INIT_LOG(DEBUG,
1126 				     "VIRTIO_NET_F_MQ is not supported");
1127 			config->max_virtqueue_pairs = 1;
1128 		}
1129 
1130 		hw->max_rx_queues =
1131 			(VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1132 			VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1133 		hw->max_tx_queues =
1134 			(VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1135 			VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1136 
1137 		virtio_dev_cq_queue_setup(eth_dev,
1138 					config->max_virtqueue_pairs * 2,
1139 					SOCKET_ID_ANY);
1140 
1141 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1142 				config->max_virtqueue_pairs);
1143 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1144 		PMD_INIT_LOG(DEBUG,
1145 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1146 				config->mac[0], config->mac[1],
1147 				config->mac[2], config->mac[3],
1148 				config->mac[4], config->mac[5]);
1149 	} else {
1150 		hw->max_rx_queues = 1;
1151 		hw->max_tx_queues = 1;
1152 	}
1153 
1154 	PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1155 			hw->max_rx_queues, hw->max_tx_queues);
1156 	PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1157 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1158 			pci_dev->id.device_id);
1159 
1160 	/* Setup interrupt callback  */
1161 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1162 		rte_intr_callback_register(&pci_dev->intr_handle,
1163 				   virtio_interrupt_handler, eth_dev);
1164 
1165 	virtio_dev_cq_start(eth_dev);
1166 
1167 	return 0;
1168 }
1169 
1170 static int
1171 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1172 {
1173 	struct rte_pci_device *pci_dev;
1174 	struct virtio_hw *hw = eth_dev->data->dev_private;
1175 
1176 	PMD_INIT_FUNC_TRACE();
1177 
1178 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1179 		return -EPERM;
1180 
1181 	/* Close it anyway since there's no way to know if closed */
1182 	virtio_dev_close(eth_dev);
1183 
1184 	pci_dev = eth_dev->pci_dev;
1185 
1186 	eth_dev->dev_ops = NULL;
1187 	eth_dev->tx_pkt_burst = NULL;
1188 	eth_dev->rx_pkt_burst = NULL;
1189 
1190 	virtio_dev_queue_release(hw->cvq);
1191 
1192 	rte_free(eth_dev->data->mac_addrs);
1193 	eth_dev->data->mac_addrs = NULL;
1194 
1195 	/* reset interrupt callback  */
1196 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1197 		rte_intr_callback_unregister(&pci_dev->intr_handle,
1198 						virtio_interrupt_handler,
1199 						eth_dev);
1200 	rte_eal_pci_unmap_device(pci_dev);
1201 
1202 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1203 
1204 	return 0;
1205 }
1206 
1207 static struct eth_driver rte_virtio_pmd = {
1208 	.pci_drv = {
1209 		.name = "rte_virtio_pmd",
1210 		.id_table = pci_id_virtio_map,
1211 		.drv_flags = RTE_PCI_DRV_DETACHABLE,
1212 	},
1213 	.eth_dev_init = eth_virtio_dev_init,
1214 	.eth_dev_uninit = eth_virtio_dev_uninit,
1215 	.dev_private_size = sizeof(struct virtio_hw),
1216 };
1217 
1218 /*
1219  * Driver initialization routine.
1220  * Invoked once at EAL init time.
1221  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1222  * Returns 0 on success.
1223  */
1224 static int
1225 rte_virtio_pmd_init(const char *name __rte_unused,
1226 		    const char *param __rte_unused)
1227 {
1228 	if (rte_eal_iopl_init() != 0) {
1229 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1230 		return -1;
1231 	}
1232 
1233 	rte_eth_driver_register(&rte_virtio_pmd);
1234 	return 0;
1235 }
1236 
1237 /*
1238  * Configure virtio device
1239  * It returns 0 on success.
1240  */
1241 static int
1242 virtio_dev_configure(struct rte_eth_dev *dev)
1243 {
1244 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1245 	struct virtio_hw *hw = dev->data->dev_private;
1246 
1247 	PMD_INIT_LOG(DEBUG, "configure");
1248 
1249 	if (rxmode->hw_ip_checksum) {
1250 		PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1251 		return -EINVAL;
1252 	}
1253 
1254 	hw->vlan_strip = rxmode->hw_vlan_strip;
1255 
1256 	if (rxmode->hw_vlan_filter
1257 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1258 		PMD_DRV_LOG(NOTICE,
1259 			    "vlan filtering not available on this host");
1260 		return -ENOTSUP;
1261 	}
1262 
1263 	if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1264 		if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1265 			PMD_DRV_LOG(ERR, "failed to set config vector");
1266 			return -EBUSY;
1267 		}
1268 
1269 	return 0;
1270 }
1271 
1272 
1273 static int
1274 virtio_dev_start(struct rte_eth_dev *dev)
1275 {
1276 	uint16_t nb_queues, i;
1277 	struct virtio_hw *hw = dev->data->dev_private;
1278 
1279 	/* check if lsc interrupt feature is enabled */
1280 	if (dev->data->dev_conf.intr_conf.lsc) {
1281 		if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1282 			PMD_DRV_LOG(ERR, "link status not supported by host");
1283 			return -ENOTSUP;
1284 		}
1285 
1286 		if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1287 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1288 			return -EIO;
1289 		}
1290 	}
1291 
1292 	/* Initialize Link state */
1293 	virtio_dev_link_update(dev, 0);
1294 
1295 	/* On restart after stop do not touch queues */
1296 	if (hw->started)
1297 		return 0;
1298 
1299 	/* Do final configuration before rx/tx engine starts */
1300 	virtio_dev_rxtx_start(dev);
1301 	vtpci_reinit_complete(hw);
1302 
1303 	hw->started = 1;
1304 
1305 	/*Notify the backend
1306 	 *Otherwise the tap backend might already stop its queue due to fullness.
1307 	 *vhost backend will have no chance to be waked up
1308 	 */
1309 	nb_queues = dev->data->nb_rx_queues;
1310 	if (nb_queues > 1) {
1311 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1312 			return -EINVAL;
1313 	}
1314 
1315 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1316 
1317 	for (i = 0; i < nb_queues; i++)
1318 		virtqueue_notify(dev->data->rx_queues[i]);
1319 
1320 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1321 
1322 	for (i = 0; i < dev->data->nb_rx_queues; i++)
1323 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1324 
1325 	for (i = 0; i < dev->data->nb_tx_queues; i++)
1326 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1327 
1328 	return 0;
1329 }
1330 
1331 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1332 {
1333 	struct rte_mbuf *buf;
1334 	int i, mbuf_num = 0;
1335 
1336 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1337 		PMD_INIT_LOG(DEBUG,
1338 			     "Before freeing rxq[%d] used and unused buf", i);
1339 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1340 
1341 		PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p",
1342 				i, dev->data->rx_queues[i]);
1343 		while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1344 					dev->data->rx_queues[i])) != NULL) {
1345 			rte_pktmbuf_free(buf);
1346 			mbuf_num++;
1347 		}
1348 
1349 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1350 		PMD_INIT_LOG(DEBUG,
1351 			     "After freeing rxq[%d] used and unused buf", i);
1352 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1353 	}
1354 
1355 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1356 		PMD_INIT_LOG(DEBUG,
1357 			     "Before freeing txq[%d] used and unused bufs",
1358 			     i);
1359 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1360 
1361 		mbuf_num = 0;
1362 		while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1363 					dev->data->tx_queues[i])) != NULL) {
1364 			rte_pktmbuf_free(buf);
1365 
1366 			mbuf_num++;
1367 		}
1368 
1369 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1370 		PMD_INIT_LOG(DEBUG,
1371 			     "After freeing txq[%d] used and unused buf", i);
1372 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1373 	}
1374 }
1375 
1376 /*
1377  * Stop device: disable interrupt and mark link down
1378  */
1379 static void
1380 virtio_dev_stop(struct rte_eth_dev *dev)
1381 {
1382 	struct rte_eth_link link;
1383 	struct virtio_hw *hw = dev->data->dev_private;
1384 
1385 	PMD_INIT_LOG(DEBUG, "stop");
1386 
1387 	hw->started = 0;
1388 
1389 	if (dev->data->dev_conf.intr_conf.lsc)
1390 		rte_intr_disable(&dev->pci_dev->intr_handle);
1391 
1392 	memset(&link, 0, sizeof(link));
1393 	virtio_dev_atomic_write_link_status(dev, &link);
1394 }
1395 
1396 static int
1397 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1398 {
1399 	struct rte_eth_link link, old;
1400 	uint16_t status;
1401 	struct virtio_hw *hw = dev->data->dev_private;
1402 	memset(&link, 0, sizeof(link));
1403 	virtio_dev_atomic_read_link_status(dev, &link);
1404 	old = link;
1405 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1406 	link.link_speed  = SPEED_10G;
1407 
1408 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1409 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
1410 		vtpci_read_dev_config(hw,
1411 				offsetof(struct virtio_net_config, status),
1412 				&status, sizeof(status));
1413 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1414 			link.link_status = ETH_LINK_DOWN;
1415 			PMD_INIT_LOG(DEBUG, "Port %d is down",
1416 				     dev->data->port_id);
1417 		} else {
1418 			link.link_status = ETH_LINK_UP;
1419 			PMD_INIT_LOG(DEBUG, "Port %d is up",
1420 				     dev->data->port_id);
1421 		}
1422 	} else {
1423 		link.link_status = ETH_LINK_UP;
1424 	}
1425 	virtio_dev_atomic_write_link_status(dev, &link);
1426 
1427 	return (old.link_status == link.link_status) ? -1 : 0;
1428 }
1429 
1430 static void
1431 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1432 {
1433 	struct virtio_hw *hw = dev->data->dev_private;
1434 
1435 	dev_info->driver_name = dev->driver->pci_drv.name;
1436 	dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1437 	dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1438 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1439 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1440 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1441 	dev_info->default_txconf = (struct rte_eth_txconf) {
1442 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1443 	};
1444 }
1445 
1446 /*
1447  * It enables testpmd to collect per queue stats.
1448  */
1449 static int
1450 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1451 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1452 __rte_unused uint8_t is_rx)
1453 {
1454 	return 0;
1455 }
1456 
1457 static struct rte_driver rte_virtio_driver = {
1458 	.type = PMD_PDEV,
1459 	.init = rte_virtio_pmd_init,
1460 };
1461 
1462 PMD_REGISTER_DRIVER(rte_virtio_driver);
1463