xref: /dpdk/drivers/net/virtio/virtio_ethdev.c (revision e76d7a768ce085c140e41f338f45d50118964ae3)
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 #ifdef RTE_EXEC_ENV_LINUXAPP
40 #include <dirent.h>
41 #include <fcntl.h>
42 #endif
43 
44 #include <rte_ethdev.h>
45 #include <rte_memcpy.h>
46 #include <rte_string_fns.h>
47 #include <rte_memzone.h>
48 #include <rte_malloc.h>
49 #include <rte_atomic.h>
50 #include <rte_branch_prediction.h>
51 #include <rte_pci.h>
52 #include <rte_ether.h>
53 #include <rte_common.h>
54 #include <rte_errno.h>
55 
56 #include <rte_memory.h>
57 #include <rte_eal.h>
58 #include <rte_dev.h>
59 
60 #include "virtio_ethdev.h"
61 #include "virtio_pci.h"
62 #include "virtio_logs.h"
63 #include "virtqueue.h"
64 
65 
66 static int eth_virtio_dev_init(struct rte_eth_dev *eth_dev);
67 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
68 static int  virtio_dev_configure(struct rte_eth_dev *dev);
69 static int  virtio_dev_start(struct rte_eth_dev *dev);
70 static void virtio_dev_stop(struct rte_eth_dev *dev);
71 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
72 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
73 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
74 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
75 static void virtio_dev_info_get(struct rte_eth_dev *dev,
76 				struct rte_eth_dev_info *dev_info);
77 static int virtio_dev_link_update(struct rte_eth_dev *dev,
78 	__rte_unused int wait_to_complete);
79 
80 static void virtio_set_hwaddr(struct virtio_hw *hw);
81 static void virtio_get_hwaddr(struct virtio_hw *hw);
82 
83 static void virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
84 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
85 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
86 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
87 				uint16_t vlan_id, int on);
88 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
89 				struct ether_addr *mac_addr,
90 				uint32_t index, uint32_t vmdq __rte_unused);
91 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
92 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
93 				struct ether_addr *mac_addr);
94 
95 static int virtio_dev_queue_stats_mapping_set(
96 	__rte_unused struct rte_eth_dev *eth_dev,
97 	__rte_unused uint16_t queue_id,
98 	__rte_unused uint8_t stat_idx,
99 	__rte_unused uint8_t is_rx);
100 
101 /*
102  * The set of PCI devices this driver supports
103  */
104 static const struct rte_pci_id pci_id_virtio_map[] = {
105 
106 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
107 #include "rte_pci_dev_ids.h"
108 
109 { .vendor_id = 0, /* sentinel */ },
110 };
111 
112 static int
113 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
114 		int *dlen, int pkt_num)
115 {
116 	uint32_t head, i;
117 	int k, sum = 0;
118 	virtio_net_ctrl_ack status = ~0;
119 	struct virtio_pmd_ctrl result;
120 
121 	ctrl->status = status;
122 
123 	if (!(vq && vq->hw->cvq)) {
124 		PMD_INIT_LOG(ERR,
125 			     "%s(): Control queue is not supported.",
126 			     __func__);
127 		return -1;
128 	}
129 	head = vq->vq_desc_head_idx;
130 
131 	PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
132 		"vq->hw->cvq = %p vq = %p",
133 		vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
134 
135 	if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
136 		return -1;
137 
138 	memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
139 		sizeof(struct virtio_pmd_ctrl));
140 
141 	/*
142 	 * Format is enforced in qemu code:
143 	 * One TX packet for header;
144 	 * At least one TX packet per argument;
145 	 * One RX packet for ACK.
146 	 */
147 	vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
148 	vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
149 	vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
150 	vq->vq_free_cnt--;
151 	i = vq->vq_ring.desc[head].next;
152 
153 	for (k = 0; k < pkt_num; k++) {
154 		vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
155 		vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
156 			+ sizeof(struct virtio_net_ctrl_hdr)
157 			+ sizeof(ctrl->status) + sizeof(uint8_t)*sum;
158 		vq->vq_ring.desc[i].len = dlen[k];
159 		sum += dlen[k];
160 		vq->vq_free_cnt--;
161 		i = vq->vq_ring.desc[i].next;
162 	}
163 
164 	vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
165 	vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
166 			+ sizeof(struct virtio_net_ctrl_hdr);
167 	vq->vq_ring.desc[i].len = sizeof(ctrl->status);
168 	vq->vq_free_cnt--;
169 
170 	vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
171 
172 	vq_update_avail_ring(vq, head);
173 	vq_update_avail_idx(vq);
174 
175 	PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
176 
177 	virtqueue_notify(vq);
178 
179 	rte_rmb();
180 	while (vq->vq_used_cons_idx == vq->vq_ring.used->idx) {
181 		rte_rmb();
182 		usleep(100);
183 	}
184 
185 	while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
186 		uint32_t idx, desc_idx, used_idx;
187 		struct vring_used_elem *uep;
188 
189 		used_idx = (uint32_t)(vq->vq_used_cons_idx
190 				& (vq->vq_nentries - 1));
191 		uep = &vq->vq_ring.used->ring[used_idx];
192 		idx = (uint32_t) uep->id;
193 		desc_idx = idx;
194 
195 		while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
196 			desc_idx = vq->vq_ring.desc[desc_idx].next;
197 			vq->vq_free_cnt++;
198 		}
199 
200 		vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
201 		vq->vq_desc_head_idx = idx;
202 
203 		vq->vq_used_cons_idx++;
204 		vq->vq_free_cnt++;
205 	}
206 
207 	PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
208 			vq->vq_free_cnt, vq->vq_desc_head_idx);
209 
210 	memcpy(&result, vq->virtio_net_hdr_mz->addr,
211 			sizeof(struct virtio_pmd_ctrl));
212 
213 	return result.status;
214 }
215 
216 static int
217 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
218 {
219 	struct virtio_hw *hw = dev->data->dev_private;
220 	struct virtio_pmd_ctrl ctrl;
221 	int dlen[1];
222 	int ret;
223 
224 	ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
225 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
226 	memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
227 
228 	dlen[0] = sizeof(uint16_t);
229 
230 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
231 	if (ret) {
232 		PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
233 			  "failed, this is too late now...");
234 		return -EINVAL;
235 	}
236 
237 	return 0;
238 }
239 
240 void
241 virtio_dev_queue_release(struct virtqueue *vq) {
242 	struct virtio_hw *hw;
243 
244 	if (vq) {
245 		hw = vq->hw;
246 		/* Select and deactivate the queue */
247 		VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vq->queue_id);
248 		VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN, 0);
249 
250 		rte_free(vq);
251 		vq = NULL;
252 	}
253 }
254 
255 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
256 			int queue_type,
257 			uint16_t queue_idx,
258 			uint16_t vtpci_queue_idx,
259 			uint16_t nb_desc,
260 			unsigned int socket_id,
261 			struct virtqueue **pvq)
262 {
263 	char vq_name[VIRTQUEUE_MAX_NAME_SZ];
264 	const struct rte_memzone *mz;
265 	unsigned int vq_size, size;
266 	struct virtio_hw *hw = dev->data->dev_private;
267 	struct virtqueue *vq = NULL;
268 
269 	/* Write the virtqueue index to the Queue Select Field */
270 	VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vtpci_queue_idx);
271 	PMD_INIT_LOG(DEBUG, "selecting queue: %u", vtpci_queue_idx);
272 
273 	/*
274 	 * Read the virtqueue size from the Queue Size field
275 	 * Always power of 2 and if 0 virtqueue does not exist
276 	 */
277 	vq_size = VIRTIO_READ_REG_2(hw, VIRTIO_PCI_QUEUE_NUM);
278 	PMD_INIT_LOG(DEBUG, "vq_size: %u nb_desc:%u", vq_size, nb_desc);
279 	if (vq_size == 0) {
280 		PMD_INIT_LOG(ERR, "%s: virtqueue does not exist", __func__);
281 		return -EINVAL;
282 	}
283 
284 	if (!rte_is_power_of_2(vq_size)) {
285 		PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2", __func__);
286 		return -EINVAL;
287 	}
288 
289 	if (queue_type == VTNET_RQ) {
290 		snprintf(vq_name, sizeof(vq_name), "port%d_rvq%d",
291 			dev->data->port_id, queue_idx);
292 		vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
293 			vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
294 	} else if (queue_type == VTNET_TQ) {
295 		snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d",
296 			dev->data->port_id, queue_idx);
297 		vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
298 			vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
299 	} else if (queue_type == VTNET_CQ) {
300 		snprintf(vq_name, sizeof(vq_name), "port%d_cvq",
301 			dev->data->port_id);
302 		vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
303 			vq_size * sizeof(struct vq_desc_extra),
304 			RTE_CACHE_LINE_SIZE);
305 	}
306 	if (vq == NULL) {
307 		PMD_INIT_LOG(ERR, "%s: Can not allocate virtqueue", __func__);
308 		return (-ENOMEM);
309 	}
310 
311 	vq->hw = hw;
312 	vq->port_id = dev->data->port_id;
313 	vq->queue_id = queue_idx;
314 	vq->vq_queue_index = vtpci_queue_idx;
315 	vq->vq_nentries = vq_size;
316 
317 	if (nb_desc == 0 || nb_desc > vq_size)
318 		nb_desc = vq_size;
319 	vq->vq_free_cnt = nb_desc;
320 
321 	/*
322 	 * Reserve a memzone for vring elements
323 	 */
324 	size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
325 	vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
326 	PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
327 
328 	mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
329 		socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
330 	if (mz == NULL) {
331 		if (rte_errno == EEXIST)
332 			mz = rte_memzone_lookup(vq_name);
333 		if (mz == NULL) {
334 			rte_free(vq);
335 			return -ENOMEM;
336 		}
337 	}
338 
339 	/*
340 	 * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
341 	 * and only accepts 32 bit page frame number.
342 	 * Check if the allocated physical memory exceeds 16TB.
343 	 */
344 	if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
345 		PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
346 		rte_free(vq);
347 		return -ENOMEM;
348 	}
349 
350 	memset(mz->addr, 0, sizeof(mz->len));
351 	vq->mz = mz;
352 	vq->vq_ring_mem = mz->phys_addr;
353 	vq->vq_ring_virt_mem = mz->addr;
354 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%"PRIx64, (uint64_t)mz->phys_addr);
355 	PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)(uintptr_t)mz->addr);
356 	vq->virtio_net_hdr_mz  = NULL;
357 	vq->virtio_net_hdr_mem = 0;
358 
359 	if (queue_type == VTNET_TQ) {
360 		/*
361 		 * For each xmit packet, allocate a virtio_net_hdr
362 		 */
363 		snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
364 			dev->data->port_id, queue_idx);
365 		vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
366 			vq_size * hw->vtnet_hdr_size,
367 			socket_id, 0, RTE_CACHE_LINE_SIZE);
368 		if (vq->virtio_net_hdr_mz == NULL) {
369 			if (rte_errno == EEXIST)
370 				vq->virtio_net_hdr_mz =
371 					rte_memzone_lookup(vq_name);
372 			if (vq->virtio_net_hdr_mz == NULL) {
373 				rte_free(vq);
374 				return -ENOMEM;
375 			}
376 		}
377 		vq->virtio_net_hdr_mem =
378 			vq->virtio_net_hdr_mz->phys_addr;
379 		memset(vq->virtio_net_hdr_mz->addr, 0,
380 			vq_size * hw->vtnet_hdr_size);
381 	} else if (queue_type == VTNET_CQ) {
382 		/* Allocate a page for control vq command, data and status */
383 		snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
384 			dev->data->port_id);
385 		vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
386 			PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
387 		if (vq->virtio_net_hdr_mz == NULL) {
388 			if (rte_errno == EEXIST)
389 				vq->virtio_net_hdr_mz =
390 					rte_memzone_lookup(vq_name);
391 			if (vq->virtio_net_hdr_mz == NULL) {
392 				rte_free(vq);
393 				return -ENOMEM;
394 			}
395 		}
396 		vq->virtio_net_hdr_mem =
397 			vq->virtio_net_hdr_mz->phys_addr;
398 		memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
399 	}
400 
401 	/*
402 	 * Set guest physical address of the virtqueue
403 	 * in VIRTIO_PCI_QUEUE_PFN config register of device
404 	 */
405 	VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
406 			mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
407 	*pvq = vq;
408 	return 0;
409 }
410 
411 static int
412 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
413 		uint32_t socket_id)
414 {
415 	struct virtqueue *vq;
416 	int ret;
417 	struct virtio_hw *hw = dev->data->dev_private;
418 
419 	PMD_INIT_FUNC_TRACE();
420 	ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
421 			vtpci_queue_idx, 0, socket_id, &vq);
422 	if (ret < 0) {
423 		PMD_INIT_LOG(ERR, "control vq initialization failed");
424 		return ret;
425 	}
426 
427 	hw->cvq = vq;
428 	return 0;
429 }
430 
431 static void
432 virtio_free_queues(struct rte_eth_dev *dev)
433 {
434 	unsigned int i;
435 
436 	for (i = 0; i < dev->data->nb_rx_queues; i++)
437 		virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
438 
439 	dev->data->nb_rx_queues = 0;
440 
441 	for (i = 0; i < dev->data->nb_tx_queues; i++)
442 		virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
443 
444 	dev->data->nb_tx_queues = 0;
445 }
446 
447 static void
448 virtio_dev_close(struct rte_eth_dev *dev)
449 {
450 	struct virtio_hw *hw = dev->data->dev_private;
451 	struct rte_pci_device *pci_dev = dev->pci_dev;
452 
453 	PMD_INIT_LOG(DEBUG, "virtio_dev_close");
454 
455 	/* reset the NIC */
456 	if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
457 		vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
458 	vtpci_reset(hw);
459 	hw->started = 0;
460 	virtio_dev_free_mbufs(dev);
461 	virtio_free_queues(dev);
462 }
463 
464 static void
465 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
466 {
467 	struct virtio_hw *hw = dev->data->dev_private;
468 	struct virtio_pmd_ctrl ctrl;
469 	int dlen[1];
470 	int ret;
471 
472 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
473 		PMD_INIT_LOG(INFO, "host does not support rx control\n");
474 		return;
475 	}
476 
477 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
478 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
479 	ctrl.data[0] = 1;
480 	dlen[0] = 1;
481 
482 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
483 	if (ret)
484 		PMD_INIT_LOG(ERR, "Failed to enable promisc");
485 }
486 
487 static void
488 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
489 {
490 	struct virtio_hw *hw = dev->data->dev_private;
491 	struct virtio_pmd_ctrl ctrl;
492 	int dlen[1];
493 	int ret;
494 
495 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
496 		PMD_INIT_LOG(INFO, "host does not support rx control\n");
497 		return;
498 	}
499 
500 	ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
501 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
502 	ctrl.data[0] = 0;
503 	dlen[0] = 1;
504 
505 	ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
506 	if (ret)
507 		PMD_INIT_LOG(ERR, "Failed to disable promisc");
508 }
509 
510 static void
511 virtio_dev_allmulticast_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_ALLMULTI;
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 allmulticast");
531 }
532 
533 static void
534 virtio_dev_allmulticast_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_ALLMULTI;
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 allmulticast");
554 }
555 
556 /*
557  * dev_ops for virtio, bare necessities for basic operation
558  */
559 static const struct eth_dev_ops virtio_eth_dev_ops = {
560 	.dev_configure           = virtio_dev_configure,
561 	.dev_start               = virtio_dev_start,
562 	.dev_stop                = virtio_dev_stop,
563 	.dev_close               = virtio_dev_close,
564 	.promiscuous_enable      = virtio_dev_promiscuous_enable,
565 	.promiscuous_disable     = virtio_dev_promiscuous_disable,
566 	.allmulticast_enable     = virtio_dev_allmulticast_enable,
567 	.allmulticast_disable    = virtio_dev_allmulticast_disable,
568 
569 	.dev_infos_get           = virtio_dev_info_get,
570 	.stats_get               = virtio_dev_stats_get,
571 	.stats_reset             = virtio_dev_stats_reset,
572 	.link_update             = virtio_dev_link_update,
573 	.rx_queue_setup          = virtio_dev_rx_queue_setup,
574 	.rx_queue_release        = virtio_dev_rx_queue_release,
575 	.tx_queue_setup          = virtio_dev_tx_queue_setup,
576 	.tx_queue_release        = virtio_dev_tx_queue_release,
577 	/* collect stats per queue */
578 	.queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
579 	.vlan_filter_set         = virtio_vlan_filter_set,
580 	.mac_addr_add            = virtio_mac_addr_add,
581 	.mac_addr_remove         = virtio_mac_addr_remove,
582 	.mac_addr_set            = virtio_mac_addr_set,
583 };
584 
585 static inline int
586 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
587 				struct rte_eth_link *link)
588 {
589 	struct rte_eth_link *dst = link;
590 	struct rte_eth_link *src = &(dev->data->dev_link);
591 
592 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
593 			*(uint64_t *)src) == 0)
594 		return -1;
595 
596 	return 0;
597 }
598 
599 /**
600  * Atomically writes the link status information into global
601  * structure rte_eth_dev.
602  *
603  * @param dev
604  *   - Pointer to the structure rte_eth_dev to read from.
605  *   - Pointer to the buffer to be saved with the link status.
606  *
607  * @return
608  *   - On success, zero.
609  *   - On failure, negative value.
610  */
611 static inline int
612 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
613 		struct rte_eth_link *link)
614 {
615 	struct rte_eth_link *dst = &(dev->data->dev_link);
616 	struct rte_eth_link *src = link;
617 
618 	if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
619 					*(uint64_t *)src) == 0)
620 		return -1;
621 
622 	return 0;
623 }
624 
625 static void
626 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
627 {
628 	unsigned i;
629 
630 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
631 		const struct virtqueue *txvq = dev->data->tx_queues[i];
632 		if (txvq == NULL)
633 			continue;
634 
635 		stats->opackets += txvq->packets;
636 		stats->obytes += txvq->bytes;
637 		stats->oerrors += txvq->errors;
638 
639 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
640 			stats->q_opackets[i] = txvq->packets;
641 			stats->q_obytes[i] = txvq->bytes;
642 		}
643 	}
644 
645 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
646 		const struct virtqueue *rxvq = dev->data->rx_queues[i];
647 		if (rxvq == NULL)
648 			continue;
649 
650 		stats->ipackets += rxvq->packets;
651 		stats->ibytes += rxvq->bytes;
652 		stats->ierrors += rxvq->errors;
653 
654 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
655 			stats->q_ipackets[i] = rxvq->packets;
656 			stats->q_ibytes[i] = rxvq->bytes;
657 		}
658 	}
659 
660 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
661 }
662 
663 static void
664 virtio_dev_stats_reset(struct rte_eth_dev *dev)
665 {
666 	unsigned int i;
667 
668 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
669 		struct virtqueue *txvq = dev->data->tx_queues[i];
670 		if (txvq == NULL)
671 			continue;
672 
673 		txvq->packets = 0;
674 		txvq->bytes = 0;
675 		txvq->errors = 0;
676 	}
677 
678 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
679 		struct virtqueue *rxvq = dev->data->rx_queues[i];
680 		if (rxvq == NULL)
681 			continue;
682 
683 		rxvq->packets = 0;
684 		rxvq->bytes = 0;
685 		rxvq->errors = 0;
686 	}
687 
688 	dev->data->rx_mbuf_alloc_failed = 0;
689 }
690 
691 static void
692 virtio_set_hwaddr(struct virtio_hw *hw)
693 {
694 	vtpci_write_dev_config(hw,
695 			offsetof(struct virtio_net_config, mac),
696 			&hw->mac_addr, ETHER_ADDR_LEN);
697 }
698 
699 static void
700 virtio_get_hwaddr(struct virtio_hw *hw)
701 {
702 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
703 		vtpci_read_dev_config(hw,
704 			offsetof(struct virtio_net_config, mac),
705 			&hw->mac_addr, ETHER_ADDR_LEN);
706 	} else {
707 		eth_random_addr(&hw->mac_addr[0]);
708 		virtio_set_hwaddr(hw);
709 	}
710 }
711 
712 static void
713 virtio_mac_table_set(struct virtio_hw *hw,
714 		     const struct virtio_net_ctrl_mac *uc,
715 		     const struct virtio_net_ctrl_mac *mc)
716 {
717 	struct virtio_pmd_ctrl ctrl;
718 	int err, len[2];
719 
720 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
721 		PMD_DRV_LOG(INFO, "host does not support mac table\n");
722 		return;
723 	}
724 
725 	ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
726 	ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
727 
728 	len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
729 	memcpy(ctrl.data, uc, len[0]);
730 
731 	len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
732 	memcpy(ctrl.data + len[0], mc, len[1]);
733 
734 	err = virtio_send_command(hw->cvq, &ctrl, len, 2);
735 	if (err != 0)
736 		PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
737 }
738 
739 static void
740 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
741 		    uint32_t index, uint32_t vmdq __rte_unused)
742 {
743 	struct virtio_hw *hw = dev->data->dev_private;
744 	const struct ether_addr *addrs = dev->data->mac_addrs;
745 	unsigned int i;
746 	struct virtio_net_ctrl_mac *uc, *mc;
747 
748 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
749 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
750 		return;
751 	}
752 
753 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
754 	uc->entries = 0;
755 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
756 	mc->entries = 0;
757 
758 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
759 		const struct ether_addr *addr
760 			= (i == index) ? mac_addr : addrs + i;
761 		struct virtio_net_ctrl_mac *tbl
762 			= is_multicast_ether_addr(addr) ? mc : uc;
763 
764 		memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
765 	}
766 
767 	virtio_mac_table_set(hw, uc, mc);
768 }
769 
770 static void
771 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
772 {
773 	struct virtio_hw *hw = dev->data->dev_private;
774 	struct ether_addr *addrs = dev->data->mac_addrs;
775 	struct virtio_net_ctrl_mac *uc, *mc;
776 	unsigned int i;
777 
778 	if (index >= VIRTIO_MAX_MAC_ADDRS) {
779 		PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
780 		return;
781 	}
782 
783 	uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
784 	uc->entries = 0;
785 	mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
786 	mc->entries = 0;
787 
788 	for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
789 		struct virtio_net_ctrl_mac *tbl;
790 
791 		if (i == index || is_zero_ether_addr(addrs + i))
792 			continue;
793 
794 		tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
795 		memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
796 	}
797 
798 	virtio_mac_table_set(hw, uc, mc);
799 }
800 
801 static void
802 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
803 {
804 	struct virtio_hw *hw = dev->data->dev_private;
805 
806 	memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
807 
808 	/* Use atomic update if available */
809 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
810 		struct virtio_pmd_ctrl ctrl;
811 		int len = ETHER_ADDR_LEN;
812 
813 		ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
814 		ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
815 
816 		memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
817 		virtio_send_command(hw->cvq, &ctrl, &len, 1);
818 	} else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
819 		virtio_set_hwaddr(hw);
820 }
821 
822 static int
823 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
824 {
825 	struct virtio_hw *hw = dev->data->dev_private;
826 	struct virtio_pmd_ctrl ctrl;
827 	int len;
828 
829 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
830 		return -ENOTSUP;
831 
832 	ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
833 	ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
834 	memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
835 	len = sizeof(vlan_id);
836 
837 	return virtio_send_command(hw->cvq, &ctrl, &len, 1);
838 }
839 
840 static void
841 virtio_negotiate_features(struct virtio_hw *hw)
842 {
843 	uint32_t host_features;
844 
845 	/* Prepare guest_features: feature that driver wants to support */
846 	hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
847 	PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x",
848 		hw->guest_features);
849 
850 	/* Read device(host) feature bits */
851 	host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
852 	PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x",
853 		host_features);
854 
855 	/*
856 	 * Negotiate features: Subset of device feature bits are written back
857 	 * guest feature bits.
858 	 */
859 	hw->guest_features = vtpci_negotiate_features(hw, host_features);
860 	PMD_INIT_LOG(DEBUG, "features after negotiate = %x",
861 		hw->guest_features);
862 }
863 
864 #ifdef RTE_EXEC_ENV_LINUXAPP
865 static int
866 parse_sysfs_value(const char *filename, unsigned long *val)
867 {
868 	FILE *f;
869 	char buf[BUFSIZ];
870 	char *end = NULL;
871 
872 	f = fopen(filename, "r");
873 	if (f == NULL) {
874 		PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s",
875 			     __func__, filename);
876 		return -1;
877 	}
878 
879 	if (fgets(buf, sizeof(buf), f) == NULL) {
880 		PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s",
881 			     __func__, filename);
882 		fclose(f);
883 		return -1;
884 	}
885 	*val = strtoul(buf, &end, 0);
886 	if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
887 		PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s",
888 			     __func__, filename);
889 		fclose(f);
890 		return -1;
891 	}
892 	fclose(f);
893 	return 0;
894 }
895 
896 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen,
897 			unsigned int *uio_num)
898 {
899 	struct dirent *e;
900 	DIR *dir;
901 	char dirname[PATH_MAX];
902 
903 	/* depending on kernel version, uio can be located in uio/uioX
904 	 * or uio:uioX */
905 	snprintf(dirname, sizeof(dirname),
906 		     SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
907 		     loc->domain, loc->bus, loc->devid, loc->function);
908 	dir = opendir(dirname);
909 	if (dir == NULL) {
910 		/* retry with the parent directory */
911 		snprintf(dirname, sizeof(dirname),
912 			     SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
913 			     loc->domain, loc->bus, loc->devid, loc->function);
914 		dir = opendir(dirname);
915 
916 		if (dir == NULL) {
917 			PMD_INIT_LOG(ERR, "Cannot opendir %s", dirname);
918 			return -1;
919 		}
920 	}
921 
922 	/* take the first file starting with "uio" */
923 	while ((e = readdir(dir)) != NULL) {
924 		/* format could be uio%d ...*/
925 		int shortprefix_len = sizeof("uio") - 1;
926 		/* ... or uio:uio%d */
927 		int longprefix_len = sizeof("uio:uio") - 1;
928 		char *endptr;
929 
930 		if (strncmp(e->d_name, "uio", 3) != 0)
931 			continue;
932 
933 		/* first try uio%d */
934 		errno = 0;
935 		*uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
936 		if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
937 			snprintf(buf, buflen, "%s/uio%u", dirname, *uio_num);
938 			break;
939 		}
940 
941 		/* then try uio:uio%d */
942 		errno = 0;
943 		*uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
944 		if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
945 			snprintf(buf, buflen, "%s/uio:uio%u", dirname,
946 				     *uio_num);
947 			break;
948 		}
949 	}
950 	closedir(dir);
951 
952 	/* No uio resource found */
953 	if (e == NULL) {
954 		PMD_INIT_LOG(ERR, "Could not find uio resource");
955 		return -1;
956 	}
957 
958 	return 0;
959 }
960 
961 static int
962 virtio_has_msix(const struct rte_pci_addr *loc)
963 {
964 	DIR *d;
965 	char dirname[PATH_MAX];
966 
967 	snprintf(dirname, sizeof(dirname),
968 		     SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
969 		     loc->domain, loc->bus, loc->devid, loc->function);
970 
971 	d = opendir(dirname);
972 	if (d)
973 		closedir(d);
974 
975 	return (d != NULL);
976 }
977 
978 /* Extract I/O port numbers from sysfs */
979 static int virtio_resource_init_by_uio(struct rte_pci_device *pci_dev)
980 {
981 	char dirname[PATH_MAX];
982 	char filename[PATH_MAX];
983 	unsigned long start, size;
984 	unsigned int uio_num;
985 
986 	if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname), &uio_num) < 0)
987 		return -1;
988 
989 	/* get portio size */
990 	snprintf(filename, sizeof(filename),
991 		     "%s/portio/port0/size", dirname);
992 	if (parse_sysfs_value(filename, &size) < 0) {
993 		PMD_INIT_LOG(ERR, "%s(): cannot parse size",
994 			     __func__);
995 		return -1;
996 	}
997 
998 	/* get portio start */
999 	snprintf(filename, sizeof(filename),
1000 		 "%s/portio/port0/start", dirname);
1001 	if (parse_sysfs_value(filename, &start) < 0) {
1002 		PMD_INIT_LOG(ERR, "%s(): cannot parse portio start",
1003 			     __func__);
1004 		return -1;
1005 	}
1006 	pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
1007 	pci_dev->mem_resource[0].len =  (uint64_t)size;
1008 	PMD_INIT_LOG(DEBUG,
1009 		     "PCI Port IO found start=0x%lx with size=0x%lx",
1010 		     start, size);
1011 
1012 	/* save fd */
1013 	memset(dirname, 0, sizeof(dirname));
1014 	snprintf(dirname, sizeof(dirname), "/dev/uio%u", uio_num);
1015 	pci_dev->intr_handle.fd = open(dirname, O_RDWR);
1016 	if (pci_dev->intr_handle.fd < 0) {
1017 		PMD_INIT_LOG(ERR, "Cannot open %s: %s\n",
1018 			dirname, strerror(errno));
1019 		return -1;
1020 	}
1021 
1022 	pci_dev->intr_handle.type = RTE_INTR_HANDLE_UIO;
1023 	pci_dev->driver->drv_flags |= RTE_PCI_DRV_INTR_LSC;
1024 
1025 	return 0;
1026 }
1027 
1028 /* Extract port I/O numbers from proc/ioports */
1029 static int virtio_resource_init_by_ioports(struct rte_pci_device *pci_dev)
1030 {
1031 	uint16_t start, end;
1032 	int size;
1033 	FILE *fp;
1034 	char *line = NULL;
1035 	char pci_id[16];
1036 	int found = 0;
1037 	size_t linesz;
1038 
1039 	snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
1040 		 pci_dev->addr.domain,
1041 		 pci_dev->addr.bus,
1042 		 pci_dev->addr.devid,
1043 		 pci_dev->addr.function);
1044 
1045 	fp = fopen("/proc/ioports", "r");
1046 	if (fp == NULL) {
1047 		PMD_INIT_LOG(ERR, "%s(): can't open ioports", __func__);
1048 		return -1;
1049 	}
1050 
1051 	while (getdelim(&line, &linesz, '\n', fp) > 0) {
1052 		char *ptr = line;
1053 		char *left;
1054 		int n;
1055 
1056 		n = strcspn(ptr, ":");
1057 		ptr[n] = 0;
1058 		left = &ptr[n+1];
1059 
1060 		while (*left && isspace(*left))
1061 			left++;
1062 
1063 		if (!strncmp(left, pci_id, strlen(pci_id))) {
1064 			found = 1;
1065 
1066 			while (*ptr && isspace(*ptr))
1067 				ptr++;
1068 
1069 			sscanf(ptr, "%04hx-%04hx", &start, &end);
1070 			size = end - start + 1;
1071 
1072 			break;
1073 		}
1074 	}
1075 
1076 	free(line);
1077 	fclose(fp);
1078 
1079 	if (!found)
1080 		return -1;
1081 
1082 	pci_dev->mem_resource[0].addr = (void *)(uintptr_t)(uint32_t)start;
1083 	pci_dev->mem_resource[0].len =  (uint64_t)size;
1084 	PMD_INIT_LOG(DEBUG,
1085 		"PCI Port IO found start=0x%x with size=0x%x",
1086 		start, size);
1087 
1088 	/* can't support lsc interrupt without uio */
1089 	pci_dev->driver->drv_flags &= ~RTE_PCI_DRV_INTR_LSC;
1090 
1091 	return 0;
1092 }
1093 
1094 /* Extract I/O port numbers from sysfs */
1095 static int virtio_resource_init(struct rte_pci_device *pci_dev)
1096 {
1097 	if (virtio_resource_init_by_uio(pci_dev) == 0)
1098 		return 0;
1099 	else
1100 		return virtio_resource_init_by_ioports(pci_dev);
1101 }
1102 
1103 #else
1104 static int
1105 virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
1106 {
1107 	/* nic_uio does not enable interrupts, return 0 (false). */
1108 	return 0;
1109 }
1110 
1111 static int virtio_resource_init(struct rte_pci_device *pci_dev __rte_unused)
1112 {
1113 	/* no setup required */
1114 	return 0;
1115 }
1116 #endif
1117 
1118 /*
1119  * Process Virtio Config changed interrupt and call the callback
1120  * if link state changed.
1121  */
1122 static void
1123 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1124 			 void *param)
1125 {
1126 	struct rte_eth_dev *dev = param;
1127 	struct virtio_hw *hw = dev->data->dev_private;
1128 	uint8_t isr;
1129 
1130 	/* Read interrupt status which clears interrupt */
1131 	isr = vtpci_isr(hw);
1132 	PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1133 
1134 	if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1135 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1136 
1137 	if (isr & VIRTIO_PCI_ISR_CONFIG) {
1138 		if (virtio_dev_link_update(dev, 0) == 0)
1139 			_rte_eth_dev_callback_process(dev,
1140 						      RTE_ETH_EVENT_INTR_LSC);
1141 	}
1142 
1143 }
1144 
1145 static void
1146 rx_func_get(struct rte_eth_dev *eth_dev)
1147 {
1148 	struct virtio_hw *hw = eth_dev->data->dev_private;
1149 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1150 		eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1151 	else
1152 		eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1153 }
1154 
1155 /*
1156  * This function is based on probe() function in virtio_pci.c
1157  * It returns 0 on success.
1158  */
1159 static int
1160 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1161 {
1162 	struct virtio_hw *hw = eth_dev->data->dev_private;
1163 	struct virtio_net_config *config;
1164 	struct virtio_net_config local_config;
1165 	uint32_t offset_conf = sizeof(config->mac);
1166 	struct rte_pci_device *pci_dev;
1167 
1168 	RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr));
1169 
1170 	eth_dev->dev_ops = &virtio_eth_dev_ops;
1171 	eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1172 
1173 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1174 		rx_func_get(eth_dev);
1175 		return 0;
1176 	}
1177 
1178 	/* Allocate memory for storing MAC addresses */
1179 	eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
1180 	if (eth_dev->data->mac_addrs == NULL) {
1181 		PMD_INIT_LOG(ERR,
1182 			"Failed to allocate %d bytes needed to store MAC addresses",
1183 			ETHER_ADDR_LEN);
1184 		return -ENOMEM;
1185 	}
1186 
1187 	pci_dev = eth_dev->pci_dev;
1188 	if (virtio_resource_init(pci_dev) < 0)
1189 		return -1;
1190 
1191 	hw->use_msix = virtio_has_msix(&pci_dev->addr);
1192 	hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
1193 
1194 	/* Reset the device although not necessary at startup */
1195 	vtpci_reset(hw);
1196 
1197 	/* Tell the host we've noticed this device. */
1198 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1199 
1200 	/* Tell the host we've known how to drive the device. */
1201 	vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1202 	virtio_negotiate_features(hw);
1203 
1204 	/* If host does not support status then disable LSC */
1205 	if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1206 		pci_dev->driver->drv_flags &= ~RTE_PCI_DRV_INTR_LSC;
1207 
1208 	rx_func_get(eth_dev);
1209 
1210 	/* Setting up rx_header size for the device */
1211 	if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1212 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1213 	else
1214 		hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1215 
1216 	/* Copy the permanent MAC address to: virtio_hw */
1217 	virtio_get_hwaddr(hw);
1218 	ether_addr_copy((struct ether_addr *) hw->mac_addr,
1219 			&eth_dev->data->mac_addrs[0]);
1220 	PMD_INIT_LOG(DEBUG,
1221 		     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1222 		     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1223 		     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1224 
1225 	if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1226 		config = &local_config;
1227 
1228 		if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1229 			offset_conf += sizeof(config->status);
1230 		} else {
1231 			PMD_INIT_LOG(DEBUG,
1232 				     "VIRTIO_NET_F_STATUS is not supported");
1233 			config->status = 0;
1234 		}
1235 
1236 		if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1237 			offset_conf += sizeof(config->max_virtqueue_pairs);
1238 		} else {
1239 			PMD_INIT_LOG(DEBUG,
1240 				     "VIRTIO_NET_F_MQ is not supported");
1241 			config->max_virtqueue_pairs = 1;
1242 		}
1243 
1244 		vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
1245 
1246 		hw->max_rx_queues =
1247 			(VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1248 			VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1249 		hw->max_tx_queues =
1250 			(VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1251 			VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1252 
1253 		virtio_dev_cq_queue_setup(eth_dev,
1254 					config->max_virtqueue_pairs * 2,
1255 					SOCKET_ID_ANY);
1256 
1257 		PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1258 				config->max_virtqueue_pairs);
1259 		PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1260 		PMD_INIT_LOG(DEBUG,
1261 				"PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1262 				config->mac[0], config->mac[1],
1263 				config->mac[2], config->mac[3],
1264 				config->mac[4], config->mac[5]);
1265 	} else {
1266 		hw->max_rx_queues = 1;
1267 		hw->max_tx_queues = 1;
1268 	}
1269 
1270 	eth_dev->data->nb_rx_queues = hw->max_rx_queues;
1271 	eth_dev->data->nb_tx_queues = hw->max_tx_queues;
1272 
1273 	PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1274 			hw->max_rx_queues, hw->max_tx_queues);
1275 	PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1276 			eth_dev->data->port_id, pci_dev->id.vendor_id,
1277 			pci_dev->id.device_id);
1278 
1279 	/* Setup interrupt callback  */
1280 	if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1281 		rte_intr_callback_register(&pci_dev->intr_handle,
1282 				   virtio_interrupt_handler, eth_dev);
1283 
1284 	virtio_dev_cq_start(eth_dev);
1285 
1286 	return 0;
1287 }
1288 
1289 static int
1290 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1291 {
1292 	struct rte_pci_device *pci_dev;
1293 	struct virtio_hw *hw = eth_dev->data->dev_private;
1294 
1295 	PMD_INIT_FUNC_TRACE();
1296 
1297 	if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1298 		return -EPERM;
1299 
1300 	if (hw->started == 1) {
1301 		virtio_dev_stop(eth_dev);
1302 		virtio_dev_close(eth_dev);
1303 	}
1304 	pci_dev = eth_dev->pci_dev;
1305 
1306 	eth_dev->dev_ops = NULL;
1307 	eth_dev->tx_pkt_burst = NULL;
1308 	eth_dev->rx_pkt_burst = NULL;
1309 
1310 	virtio_dev_queue_release(hw->cvq);
1311 
1312 	rte_free(eth_dev->data->mac_addrs);
1313 	eth_dev->data->mac_addrs = NULL;
1314 
1315 	/* reset interrupt callback  */
1316 	if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1317 		rte_intr_callback_unregister(&pci_dev->intr_handle,
1318 						virtio_interrupt_handler,
1319 						eth_dev);
1320 
1321 	PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1322 
1323 	return 0;
1324 }
1325 
1326 static struct eth_driver rte_virtio_pmd = {
1327 	.pci_drv = {
1328 		.name = "rte_virtio_pmd",
1329 		.id_table = pci_id_virtio_map,
1330 		.drv_flags = RTE_PCI_DRV_DETACHABLE,
1331 	},
1332 	.eth_dev_init = eth_virtio_dev_init,
1333 	.eth_dev_uninit = eth_virtio_dev_uninit,
1334 	.dev_private_size = sizeof(struct virtio_hw),
1335 };
1336 
1337 /*
1338  * Driver initialization routine.
1339  * Invoked once at EAL init time.
1340  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1341  * Returns 0 on success.
1342  */
1343 static int
1344 rte_virtio_pmd_init(const char *name __rte_unused,
1345 		    const char *param __rte_unused)
1346 {
1347 	if (rte_eal_iopl_init() != 0) {
1348 		PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1349 		return -1;
1350 	}
1351 
1352 	rte_eth_driver_register(&rte_virtio_pmd);
1353 	return 0;
1354 }
1355 
1356 /*
1357  * Configure virtio device
1358  * It returns 0 on success.
1359  */
1360 static int
1361 virtio_dev_configure(struct rte_eth_dev *dev)
1362 {
1363 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1364 	struct virtio_hw *hw = dev->data->dev_private;
1365 	struct rte_pci_device *pci_dev = dev->pci_dev;
1366 
1367 	PMD_INIT_LOG(DEBUG, "configure");
1368 
1369 	if (rxmode->hw_ip_checksum) {
1370 		PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1371 		return (-EINVAL);
1372 	}
1373 
1374 	hw->vlan_strip = rxmode->hw_vlan_strip;
1375 
1376 	if (rxmode->hw_vlan_filter
1377 	    && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1378 		PMD_DRV_LOG(NOTICE,
1379 			    "vlan filtering not available on this host");
1380 		return -ENOTSUP;
1381 	}
1382 
1383 	if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1384 		if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1385 			PMD_DRV_LOG(ERR, "failed to set config vector");
1386 			return -EBUSY;
1387 		}
1388 
1389 	return 0;
1390 }
1391 
1392 
1393 static int
1394 virtio_dev_start(struct rte_eth_dev *dev)
1395 {
1396 	uint16_t nb_queues, i;
1397 	struct virtio_hw *hw = dev->data->dev_private;
1398 	struct rte_pci_device *pci_dev = dev->pci_dev;
1399 
1400 	/* check if lsc interrupt feature is enabled */
1401 	if (dev->data->dev_conf.intr_conf.lsc) {
1402 		if (!(pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)) {
1403 			PMD_DRV_LOG(ERR, "link status not supported by host");
1404 			return -ENOTSUP;
1405 		}
1406 
1407 		if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1408 			PMD_DRV_LOG(ERR, "interrupt enable failed");
1409 			return -EIO;
1410 		}
1411 	}
1412 
1413 	/* Initialize Link state */
1414 	virtio_dev_link_update(dev, 0);
1415 
1416 	/* On restart after stop do not touch queues */
1417 	if (hw->started)
1418 		return 0;
1419 
1420 	/* Do final configuration before rx/tx engine starts */
1421 	virtio_dev_rxtx_start(dev);
1422 	vtpci_reinit_complete(hw);
1423 
1424 	hw->started = 1;
1425 
1426 	/*Notify the backend
1427 	 *Otherwise the tap backend might already stop its queue due to fullness.
1428 	 *vhost backend will have no chance to be waked up
1429 	 */
1430 	nb_queues = dev->data->nb_rx_queues;
1431 	if (nb_queues > 1) {
1432 		if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1433 			return -EINVAL;
1434 	}
1435 
1436 	PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1437 
1438 	for (i = 0; i < nb_queues; i++)
1439 		virtqueue_notify(dev->data->rx_queues[i]);
1440 
1441 	PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1442 
1443 	for (i = 0; i < dev->data->nb_rx_queues; i++)
1444 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1445 
1446 	for (i = 0; i < dev->data->nb_tx_queues; i++)
1447 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1448 
1449 	return 0;
1450 }
1451 
1452 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1453 {
1454 	struct rte_mbuf *buf;
1455 	int i, mbuf_num = 0;
1456 
1457 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1458 		PMD_INIT_LOG(DEBUG,
1459 			     "Before freeing rxq[%d] used and unused buf", i);
1460 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1461 
1462 		PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p",
1463 				i, dev->data->rx_queues[i]);
1464 		while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1465 					dev->data->rx_queues[i])) != NULL) {
1466 			rte_pktmbuf_free(buf);
1467 			mbuf_num++;
1468 		}
1469 
1470 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1471 		PMD_INIT_LOG(DEBUG,
1472 			     "After freeing rxq[%d] used and unused buf", i);
1473 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1474 	}
1475 
1476 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1477 		PMD_INIT_LOG(DEBUG,
1478 			     "Before freeing txq[%d] used and unused bufs",
1479 			     i);
1480 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1481 
1482 		mbuf_num = 0;
1483 		while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1484 					dev->data->tx_queues[i])) != NULL) {
1485 			rte_pktmbuf_free(buf);
1486 
1487 			mbuf_num++;
1488 		}
1489 
1490 		PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1491 		PMD_INIT_LOG(DEBUG,
1492 			     "After freeing txq[%d] used and unused buf", i);
1493 		VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1494 	}
1495 }
1496 
1497 /*
1498  * Stop device: disable interrupt and mark link down
1499  */
1500 static void
1501 virtio_dev_stop(struct rte_eth_dev *dev)
1502 {
1503 	struct rte_eth_link link;
1504 
1505 	PMD_INIT_LOG(DEBUG, "stop");
1506 
1507 	if (dev->data->dev_conf.intr_conf.lsc)
1508 		rte_intr_disable(&dev->pci_dev->intr_handle);
1509 
1510 	memset(&link, 0, sizeof(link));
1511 	virtio_dev_atomic_write_link_status(dev, &link);
1512 }
1513 
1514 static int
1515 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1516 {
1517 	struct rte_eth_link link, old;
1518 	uint16_t status;
1519 	struct virtio_hw *hw = dev->data->dev_private;
1520 	memset(&link, 0, sizeof(link));
1521 	virtio_dev_atomic_read_link_status(dev, &link);
1522 	old = link;
1523 	link.link_duplex = FULL_DUPLEX;
1524 	link.link_speed  = SPEED_10G;
1525 
1526 	if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1527 		PMD_INIT_LOG(DEBUG, "Get link status from hw");
1528 		vtpci_read_dev_config(hw,
1529 				offsetof(struct virtio_net_config, status),
1530 				&status, sizeof(status));
1531 		if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1532 			link.link_status = 0;
1533 			PMD_INIT_LOG(DEBUG, "Port %d is down",
1534 				     dev->data->port_id);
1535 		} else {
1536 			link.link_status = 1;
1537 			PMD_INIT_LOG(DEBUG, "Port %d is up",
1538 				     dev->data->port_id);
1539 		}
1540 	} else {
1541 		link.link_status = 1;   /* Link up */
1542 	}
1543 	virtio_dev_atomic_write_link_status(dev, &link);
1544 
1545 	return (old.link_status == link.link_status) ? -1 : 0;
1546 }
1547 
1548 static void
1549 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1550 {
1551 	struct virtio_hw *hw = dev->data->dev_private;
1552 
1553 	dev_info->driver_name = dev->driver->pci_drv.name;
1554 	dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1555 	dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1556 	dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1557 	dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1558 	dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1559 	dev_info->default_txconf = (struct rte_eth_txconf) {
1560 		.txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1561 	};
1562 }
1563 
1564 /*
1565  * It enables testpmd to collect per queue stats.
1566  */
1567 static int
1568 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1569 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1570 __rte_unused uint8_t is_rx)
1571 {
1572 	return 0;
1573 }
1574 
1575 static struct rte_driver rte_virtio_driver = {
1576 	.type = PMD_PDEV,
1577 	.init = rte_virtio_pmd_init,
1578 };
1579 
1580 PMD_REGISTER_DRIVER(rte_virtio_driver);
1581