xref: /dpdk/drivers/net/netvsc/hn_ethdev.c (revision 72206323a5dd3182b13f61b25a64abdddfee595c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2016-2018 Microsoft Corporation
3  * Copyright(c) 2013-2016 Brocade Communications Systems, Inc.
4  * All rights reserved.
5  */
6 
7 #include <stdint.h>
8 #include <string.h>
9 #include <stdio.h>
10 #include <errno.h>
11 #include <unistd.h>
12 #include <dirent.h>
13 #include <net/if.h>
14 #include <net/if_arp.h>
15 #include <netinet/in.h>
16 #include <sys/ioctl.h>
17 
18 #include <rte_ethdev.h>
19 #include <rte_memcpy.h>
20 #include <rte_string_fns.h>
21 #include <rte_memzone.h>
22 #include <rte_devargs.h>
23 #include <rte_malloc.h>
24 #include <rte_kvargs.h>
25 #include <rte_atomic.h>
26 #include <rte_branch_prediction.h>
27 #include <rte_ether.h>
28 #include <ethdev_driver.h>
29 #include <rte_cycles.h>
30 #include <rte_errno.h>
31 #include <rte_memory.h>
32 #include <rte_eal.h>
33 #include <rte_dev.h>
34 #include <rte_bus_vmbus.h>
35 #include <rte_alarm.h>
36 
37 #include "hn_logs.h"
38 #include "hn_var.h"
39 #include "hn_rndis.h"
40 #include "hn_nvs.h"
41 #include "ndis.h"
42 
43 #define HN_TX_OFFLOAD_CAPS (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM | \
44 			    RTE_ETH_TX_OFFLOAD_TCP_CKSUM  | \
45 			    RTE_ETH_TX_OFFLOAD_UDP_CKSUM  | \
46 			    RTE_ETH_TX_OFFLOAD_TCP_TSO    | \
47 			    RTE_ETH_TX_OFFLOAD_MULTI_SEGS | \
48 			    RTE_ETH_TX_OFFLOAD_VLAN_INSERT)
49 
50 #define HN_RX_OFFLOAD_CAPS (RTE_ETH_RX_OFFLOAD_CHECKSUM | \
51 			    RTE_ETH_RX_OFFLOAD_VLAN_STRIP | \
52 			    RTE_ETH_RX_OFFLOAD_RSS_HASH)
53 
54 #define NETVSC_ARG_LATENCY "latency"
55 #define NETVSC_ARG_RXBREAK "rx_copybreak"
56 #define NETVSC_ARG_TXBREAK "tx_copybreak"
57 #define NETVSC_ARG_RX_EXTMBUF_ENABLE "rx_extmbuf_enable"
58 
59 /* The max number of retry when hot adding a VF device */
60 #define NETVSC_MAX_HOTADD_RETRY 10
61 
62 struct hn_xstats_name_off {
63 	char name[RTE_ETH_XSTATS_NAME_SIZE];
64 	unsigned int offset;
65 };
66 
67 static const struct hn_xstats_name_off hn_stat_strings[] = {
68 	{ "good_packets",           offsetof(struct hn_stats, packets) },
69 	{ "good_bytes",             offsetof(struct hn_stats, bytes) },
70 	{ "errors",                 offsetof(struct hn_stats, errors) },
71 	{ "ring full",              offsetof(struct hn_stats, ring_full) },
72 	{ "channel full",           offsetof(struct hn_stats, channel_full) },
73 	{ "multicast_packets",      offsetof(struct hn_stats, multicast) },
74 	{ "broadcast_packets",      offsetof(struct hn_stats, broadcast) },
75 	{ "undersize_packets",      offsetof(struct hn_stats, size_bins[0]) },
76 	{ "size_64_packets",        offsetof(struct hn_stats, size_bins[1]) },
77 	{ "size_65_127_packets",    offsetof(struct hn_stats, size_bins[2]) },
78 	{ "size_128_255_packets",   offsetof(struct hn_stats, size_bins[3]) },
79 	{ "size_256_511_packets",   offsetof(struct hn_stats, size_bins[4]) },
80 	{ "size_512_1023_packets",  offsetof(struct hn_stats, size_bins[5]) },
81 	{ "size_1024_1518_packets", offsetof(struct hn_stats, size_bins[6]) },
82 	{ "size_1519_max_packets",  offsetof(struct hn_stats, size_bins[7]) },
83 };
84 
85 /* The default RSS key.
86  * This value is the same as MLX5 so that flows will be
87  * received on same path for both VF and synthetic NIC.
88  */
89 static const uint8_t rss_default_key[NDIS_HASH_KEYSIZE_TOEPLITZ] = {
90 	0x2c, 0xc6, 0x81, 0xd1,	0x5b, 0xdb, 0xf4, 0xf7,
91 	0xfc, 0xa2, 0x83, 0x19,	0xdb, 0x1a, 0x3e, 0x94,
92 	0x6b, 0x9e, 0x38, 0xd9,	0x2c, 0x9c, 0x03, 0xd1,
93 	0xad, 0x99, 0x44, 0xa7,	0xd9, 0x56, 0x3d, 0x59,
94 	0x06, 0x3c, 0x25, 0xf3,	0xfc, 0x1f, 0xdc, 0x2a,
95 };
96 
97 static struct rte_eth_dev *
98 eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
99 {
100 	struct rte_eth_dev *eth_dev;
101 	const char *name;
102 
103 	if (!dev)
104 		return NULL;
105 
106 	name = dev->device.name;
107 
108 	if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
109 		eth_dev = rte_eth_dev_allocate(name);
110 		if (!eth_dev) {
111 			PMD_DRV_LOG(NOTICE, "can not allocate rte ethdev");
112 			return NULL;
113 		}
114 
115 		if (private_data_size) {
116 			eth_dev->data->dev_private =
117 				rte_zmalloc_socket(name, private_data_size,
118 						     RTE_CACHE_LINE_SIZE, dev->device.numa_node);
119 			if (!eth_dev->data->dev_private) {
120 				PMD_DRV_LOG(NOTICE, "can not allocate driver data");
121 				rte_eth_dev_release_port(eth_dev);
122 				return NULL;
123 			}
124 		}
125 	} else {
126 		eth_dev = rte_eth_dev_attach_secondary(name);
127 		if (!eth_dev) {
128 			PMD_DRV_LOG(NOTICE, "can not attach secondary");
129 			return NULL;
130 		}
131 	}
132 
133 	eth_dev->device = &dev->device;
134 
135 	/* interrupt is simulated */
136 	rte_intr_type_set(dev->intr_handle, RTE_INTR_HANDLE_EXT);
137 	eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
138 	eth_dev->intr_handle = dev->intr_handle;
139 
140 	return eth_dev;
141 }
142 
143 static void
144 eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
145 {
146 	/* free ether device */
147 	rte_eth_dev_release_port(eth_dev);
148 
149 	eth_dev->device = NULL;
150 	eth_dev->intr_handle = NULL;
151 }
152 
153 static int hn_set_parameter(const char *key, const char *value, void *opaque)
154 {
155 	struct hn_data *hv = opaque;
156 	char *endp = NULL;
157 	unsigned long v;
158 
159 	v = strtoul(value, &endp, 0);
160 	if (*value == '\0' || *endp != '\0') {
161 		PMD_DRV_LOG(ERR, "invalid parameter %s=%s", key, value);
162 		return -EINVAL;
163 	}
164 
165 	if (!strcmp(key, NETVSC_ARG_LATENCY)) {
166 		/* usec to nsec */
167 		hv->latency = v * 1000;
168 		PMD_DRV_LOG(DEBUG, "set latency %u usec", hv->latency);
169 	} else if (!strcmp(key, NETVSC_ARG_RXBREAK)) {
170 		hv->rx_copybreak = v;
171 		PMD_DRV_LOG(DEBUG, "rx copy break set to %u",
172 			    hv->rx_copybreak);
173 	} else if (!strcmp(key, NETVSC_ARG_TXBREAK)) {
174 		hv->tx_copybreak = v;
175 		PMD_DRV_LOG(DEBUG, "tx copy break set to %u",
176 			    hv->tx_copybreak);
177 	} else if (!strcmp(key, NETVSC_ARG_RX_EXTMBUF_ENABLE)) {
178 		hv->rx_extmbuf_enable = v;
179 		PMD_DRV_LOG(DEBUG, "rx extmbuf enable set to %u",
180 			    hv->rx_extmbuf_enable);
181 	}
182 
183 	return 0;
184 }
185 
186 /* Parse device arguments */
187 static int hn_parse_args(const struct rte_eth_dev *dev)
188 {
189 	struct hn_data *hv = dev->data->dev_private;
190 	struct rte_devargs *devargs = dev->device->devargs;
191 	static const char * const valid_keys[] = {
192 		NETVSC_ARG_LATENCY,
193 		NETVSC_ARG_RXBREAK,
194 		NETVSC_ARG_TXBREAK,
195 		NETVSC_ARG_RX_EXTMBUF_ENABLE,
196 		NULL
197 	};
198 	struct rte_kvargs *kvlist;
199 	int ret;
200 
201 	if (!devargs)
202 		return 0;
203 
204 	PMD_INIT_LOG(DEBUG, "device args %s %s",
205 		     devargs->name, devargs->args);
206 
207 	kvlist = rte_kvargs_parse(devargs->args, valid_keys);
208 	if (!kvlist) {
209 		PMD_DRV_LOG(ERR, "invalid parameters");
210 		return -EINVAL;
211 	}
212 
213 	ret = rte_kvargs_process(kvlist, NULL, hn_set_parameter, hv);
214 	rte_kvargs_free(kvlist);
215 
216 	return ret;
217 }
218 
219 /* Update link status.
220  * Note: the DPDK definition of "wait_to_complete"
221  *   means block this call until link is up.
222  *   which is not worth supporting.
223  */
224 int
225 hn_dev_link_update(struct rte_eth_dev *dev,
226 		   int wait_to_complete __rte_unused)
227 {
228 	struct hn_data *hv = dev->data->dev_private;
229 	struct rte_eth_link link, old;
230 	int error;
231 
232 	old = dev->data->dev_link;
233 
234 	error = hn_rndis_get_linkstatus(hv);
235 	if (error)
236 		return error;
237 
238 	hn_rndis_get_linkspeed(hv);
239 
240 	link = (struct rte_eth_link) {
241 		.link_duplex = RTE_ETH_LINK_FULL_DUPLEX,
242 		.link_autoneg = RTE_ETH_LINK_SPEED_FIXED,
243 		.link_speed = hv->link_speed / 10000,
244 	};
245 
246 	if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
247 		link.link_status = RTE_ETH_LINK_UP;
248 	else
249 		link.link_status = RTE_ETH_LINK_DOWN;
250 
251 	if (old.link_status == link.link_status)
252 		return 0;
253 
254 	PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
255 		     (link.link_status == RTE_ETH_LINK_UP) ? "up" : "down");
256 
257 	return rte_eth_linkstatus_set(dev, &link);
258 }
259 
260 static int hn_dev_info_get(struct rte_eth_dev *dev,
261 			   struct rte_eth_dev_info *dev_info)
262 {
263 	struct hn_data *hv = dev->data->dev_private;
264 	int rc;
265 
266 	dev_info->speed_capa = RTE_ETH_LINK_SPEED_10G;
267 	dev_info->min_rx_bufsize = HN_MIN_RX_BUF_SIZE;
268 	dev_info->max_rx_pktlen  = HN_MAX_XFER_LEN;
269 	dev_info->max_mac_addrs  = 1;
270 
271 	dev_info->hash_key_size = NDIS_HASH_KEYSIZE_TOEPLITZ;
272 	dev_info->flow_type_rss_offloads = hv->rss_offloads;
273 	dev_info->reta_size = RTE_ETH_RSS_RETA_SIZE_128;
274 
275 	dev_info->max_rx_queues = hv->max_queues;
276 	dev_info->max_tx_queues = hv->max_queues;
277 
278 	dev_info->tx_desc_lim.nb_min = 1;
279 	dev_info->tx_desc_lim.nb_max = 4096;
280 
281 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
282 		return 0;
283 
284 	/* fills in rx and tx offload capability */
285 	rc = hn_rndis_get_offload(hv, dev_info);
286 	if (rc != 0)
287 		return rc;
288 
289 	/* merges the offload and queues of vf */
290 	return hn_vf_info_get(hv, dev_info);
291 }
292 
293 static int hn_rss_reta_update(struct rte_eth_dev *dev,
294 			      struct rte_eth_rss_reta_entry64 *reta_conf,
295 			      uint16_t reta_size)
296 {
297 	struct hn_data *hv = dev->data->dev_private;
298 	unsigned int i;
299 	int err;
300 
301 	PMD_INIT_FUNC_TRACE();
302 
303 	if (reta_size != NDIS_HASH_INDCNT) {
304 		PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
305 		return -EINVAL;
306 	}
307 
308 	for (i = 0; i < NDIS_HASH_INDCNT; i++) {
309 		uint16_t idx = i / RTE_ETH_RETA_GROUP_SIZE;
310 		uint16_t shift = i % RTE_ETH_RETA_GROUP_SIZE;
311 		uint64_t mask = (uint64_t)1 << shift;
312 
313 		if (reta_conf[idx].mask & mask)
314 			hv->rss_ind[i] = reta_conf[idx].reta[shift];
315 	}
316 
317 	err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
318 	if (err) {
319 		PMD_DRV_LOG(NOTICE,
320 			"rss disable failed");
321 		return err;
322 	}
323 
324 	err = hn_rndis_conf_rss(hv, 0);
325 	if (err) {
326 		PMD_DRV_LOG(NOTICE,
327 			    "reta reconfig failed");
328 		return err;
329 	}
330 
331 	return hn_vf_reta_hash_update(dev, reta_conf, reta_size);
332 }
333 
334 static int hn_rss_reta_query(struct rte_eth_dev *dev,
335 			     struct rte_eth_rss_reta_entry64 *reta_conf,
336 			     uint16_t reta_size)
337 {
338 	struct hn_data *hv = dev->data->dev_private;
339 	unsigned int i;
340 
341 	PMD_INIT_FUNC_TRACE();
342 
343 	if (reta_size != NDIS_HASH_INDCNT) {
344 		PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
345 		return -EINVAL;
346 	}
347 
348 	for (i = 0; i < NDIS_HASH_INDCNT; i++) {
349 		uint16_t idx = i / RTE_ETH_RETA_GROUP_SIZE;
350 		uint16_t shift = i % RTE_ETH_RETA_GROUP_SIZE;
351 		uint64_t mask = (uint64_t)1 << shift;
352 
353 		if (reta_conf[idx].mask & mask)
354 			reta_conf[idx].reta[shift] = hv->rss_ind[i];
355 	}
356 	return 0;
357 }
358 
359 static void hn_rss_hash_init(struct hn_data *hv,
360 			     const struct rte_eth_rss_conf *rss_conf)
361 {
362 	/* Convert from DPDK RSS hash flags to NDIS hash flags */
363 	hv->rss_hash = NDIS_HASH_FUNCTION_TOEPLITZ;
364 
365 	if (rss_conf->rss_hf & RTE_ETH_RSS_IPV4)
366 		hv->rss_hash |= NDIS_HASH_IPV4;
367 	if (rss_conf->rss_hf & RTE_ETH_RSS_NONFRAG_IPV4_TCP)
368 		hv->rss_hash |= NDIS_HASH_TCP_IPV4;
369 	if (rss_conf->rss_hf & RTE_ETH_RSS_IPV6)
370 		hv->rss_hash |=  NDIS_HASH_IPV6;
371 	if (rss_conf->rss_hf & RTE_ETH_RSS_IPV6_EX)
372 		hv->rss_hash |=  NDIS_HASH_IPV6_EX;
373 	if (rss_conf->rss_hf & RTE_ETH_RSS_NONFRAG_IPV6_TCP)
374 		hv->rss_hash |= NDIS_HASH_TCP_IPV6;
375 	if (rss_conf->rss_hf & RTE_ETH_RSS_IPV6_TCP_EX)
376 		hv->rss_hash |= NDIS_HASH_TCP_IPV6_EX;
377 
378 	memcpy(hv->rss_key, rss_conf->rss_key ? : rss_default_key,
379 	       NDIS_HASH_KEYSIZE_TOEPLITZ);
380 }
381 
382 static int hn_rss_hash_update(struct rte_eth_dev *dev,
383 			      struct rte_eth_rss_conf *rss_conf)
384 {
385 	struct hn_data *hv = dev->data->dev_private;
386 	int err;
387 
388 	PMD_INIT_FUNC_TRACE();
389 
390 	err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
391 	if (err) {
392 		PMD_DRV_LOG(NOTICE,
393 			    "rss disable failed");
394 		return err;
395 	}
396 
397 	hn_rss_hash_init(hv, rss_conf);
398 
399 	if (rss_conf->rss_hf != 0) {
400 		err = hn_rndis_conf_rss(hv, 0);
401 		if (err) {
402 			PMD_DRV_LOG(NOTICE,
403 				    "rss reconfig failed (RSS disabled)");
404 			return err;
405 		}
406 	}
407 
408 	return hn_vf_rss_hash_update(dev, rss_conf);
409 }
410 
411 static int hn_rss_hash_conf_get(struct rte_eth_dev *dev,
412 				struct rte_eth_rss_conf *rss_conf)
413 {
414 	struct hn_data *hv = dev->data->dev_private;
415 
416 	PMD_INIT_FUNC_TRACE();
417 
418 	if (hv->ndis_ver < NDIS_VERSION_6_20) {
419 		PMD_DRV_LOG(DEBUG, "RSS not supported on this host");
420 		return -EOPNOTSUPP;
421 	}
422 
423 	rss_conf->rss_key_len = NDIS_HASH_KEYSIZE_TOEPLITZ;
424 	if (rss_conf->rss_key)
425 		memcpy(rss_conf->rss_key, hv->rss_key,
426 		       NDIS_HASH_KEYSIZE_TOEPLITZ);
427 
428 	rss_conf->rss_hf = 0;
429 	if (hv->rss_hash & NDIS_HASH_IPV4)
430 		rss_conf->rss_hf |= RTE_ETH_RSS_IPV4;
431 
432 	if (hv->rss_hash & NDIS_HASH_TCP_IPV4)
433 		rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_TCP;
434 
435 	if (hv->rss_hash & NDIS_HASH_IPV6)
436 		rss_conf->rss_hf |= RTE_ETH_RSS_IPV6;
437 
438 	if (hv->rss_hash & NDIS_HASH_IPV6_EX)
439 		rss_conf->rss_hf |= RTE_ETH_RSS_IPV6_EX;
440 
441 	if (hv->rss_hash & NDIS_HASH_TCP_IPV6)
442 		rss_conf->rss_hf |= RTE_ETH_RSS_NONFRAG_IPV6_TCP;
443 
444 	if (hv->rss_hash & NDIS_HASH_TCP_IPV6_EX)
445 		rss_conf->rss_hf |= RTE_ETH_RSS_IPV6_TCP_EX;
446 
447 	return 0;
448 }
449 
450 static int
451 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
452 {
453 	struct hn_data *hv = dev->data->dev_private;
454 
455 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
456 	return hn_vf_promiscuous_enable(dev);
457 }
458 
459 static int
460 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
461 {
462 	struct hn_data *hv = dev->data->dev_private;
463 	uint32_t filter;
464 
465 	filter = NDIS_PACKET_TYPE_DIRECTED | NDIS_PACKET_TYPE_BROADCAST;
466 	if (dev->data->all_multicast)
467 		filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
468 	hn_rndis_set_rxfilter(hv, filter);
469 	return hn_vf_promiscuous_disable(dev);
470 }
471 
472 static int
473 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
474 {
475 	struct hn_data *hv = dev->data->dev_private;
476 
477 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
478 			      NDIS_PACKET_TYPE_ALL_MULTICAST |
479 			NDIS_PACKET_TYPE_BROADCAST);
480 	return hn_vf_allmulticast_enable(dev);
481 }
482 
483 static int
484 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
485 {
486 	struct hn_data *hv = dev->data->dev_private;
487 
488 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
489 			     NDIS_PACKET_TYPE_BROADCAST);
490 	return hn_vf_allmulticast_disable(dev);
491 }
492 
493 static int
494 hn_dev_mc_addr_list(struct rte_eth_dev *dev,
495 		     struct rte_ether_addr *mc_addr_set,
496 		     uint32_t nb_mc_addr)
497 {
498 	/* No filtering on the synthetic path, but can do it on VF */
499 	return hn_vf_mc_addr_list(dev, mc_addr_set, nb_mc_addr);
500 }
501 
502 /* Setup shared rx/tx queue data */
503 static int hn_subchan_configure(struct hn_data *hv,
504 				uint32_t subchan)
505 {
506 	struct vmbus_channel *primary = hn_primary_chan(hv);
507 	int err;
508 	unsigned int retry = 0;
509 
510 	PMD_DRV_LOG(DEBUG,
511 		    "open %u subchannels", subchan);
512 
513 	/* Send create sub channels command */
514 	err = hn_nvs_alloc_subchans(hv, &subchan);
515 	if (err)
516 		return  err;
517 
518 	while (subchan > 0) {
519 		struct vmbus_channel *new_sc;
520 		uint16_t chn_index;
521 
522 		err = rte_vmbus_subchan_open(primary, &new_sc);
523 		if (err == -ENOENT && ++retry < 1000) {
524 			/* This can happen if not ready yet */
525 			rte_delay_ms(10);
526 			continue;
527 		}
528 
529 		if (err) {
530 			PMD_DRV_LOG(ERR,
531 				    "open subchannel failed: %d", err);
532 			return err;
533 		}
534 
535 		rte_vmbus_set_latency(hv->vmbus, new_sc, hv->latency);
536 
537 		retry = 0;
538 		chn_index = rte_vmbus_sub_channel_index(new_sc);
539 		if (chn_index == 0 || chn_index > hv->max_queues) {
540 			PMD_DRV_LOG(ERR,
541 				    "Invalid subchannel offermsg channel %u",
542 				    chn_index);
543 			return -EIO;
544 		}
545 
546 		PMD_DRV_LOG(DEBUG, "new sub channel %u", chn_index);
547 		hv->channels[chn_index] = new_sc;
548 		--subchan;
549 	}
550 
551 	return err;
552 }
553 
554 static void netvsc_hotplug_retry(void *args)
555 {
556 	int ret;
557 	struct hv_hotadd_context *hot_ctx = args;
558 	struct hn_data *hv = hot_ctx->hv;
559 	struct rte_eth_dev *dev = &rte_eth_devices[hv->port_id];
560 	struct rte_devargs *d = &hot_ctx->da;
561 	char buf[256];
562 
563 	DIR *di;
564 	struct dirent *dir;
565 	struct ifreq req;
566 	struct rte_ether_addr eth_addr;
567 	int s;
568 
569 	PMD_DRV_LOG(DEBUG, "%s: retry count %d",
570 		    __func__, hot_ctx->eal_hot_plug_retry);
571 
572 	if (hot_ctx->eal_hot_plug_retry++ > NETVSC_MAX_HOTADD_RETRY) {
573 		PMD_DRV_LOG(NOTICE, "Failed to parse PCI device retry=%d",
574 			    hot_ctx->eal_hot_plug_retry);
575 		goto free_hotadd_ctx;
576 	}
577 
578 	snprintf(buf, sizeof(buf), "/sys/bus/pci/devices/%s/net", d->name);
579 	di = opendir(buf);
580 	if (!di) {
581 		PMD_DRV_LOG(DEBUG, "%s: can't open directory %s, "
582 			    "retrying in 1 second", __func__, buf);
583 		goto retry;
584 	}
585 
586 	while ((dir = readdir(di))) {
587 		/* Skip . and .. directories */
588 		if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
589 			continue;
590 
591 		/* trying to get mac address if this is a network device*/
592 		s = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
593 		if (s == -1) {
594 			PMD_DRV_LOG(ERR, "Failed to create socket errno %d",
595 				    errno);
596 			break;
597 		}
598 		strlcpy(req.ifr_name, dir->d_name, sizeof(req.ifr_name));
599 		ret = ioctl(s, SIOCGIFHWADDR, &req);
600 		close(s);
601 		if (ret == -1) {
602 			PMD_DRV_LOG(ERR,
603 				    "Failed to send SIOCGIFHWADDR for device %s",
604 				    dir->d_name);
605 			break;
606 		}
607 		if (req.ifr_hwaddr.sa_family != ARPHRD_ETHER) {
608 			closedir(di);
609 			goto free_hotadd_ctx;
610 		}
611 		memcpy(eth_addr.addr_bytes, req.ifr_hwaddr.sa_data,
612 		       RTE_DIM(eth_addr.addr_bytes));
613 
614 		if (rte_is_same_ether_addr(&eth_addr, dev->data->mac_addrs)) {
615 			PMD_DRV_LOG(NOTICE,
616 				    "Found matching MAC address, adding device %s network name %s",
617 				    d->name, dir->d_name);
618 
619 			/* If this device has been hot removed from this
620 			 * parent device, restore its args.
621 			 */
622 			ret = rte_eal_hotplug_add(d->bus->name, d->name,
623 						  hv->vf_devargs ?
624 						  hv->vf_devargs : "");
625 			if (ret) {
626 				PMD_DRV_LOG(ERR,
627 					    "Failed to add PCI device %s",
628 					    d->name);
629 				break;
630 			}
631 		}
632 		/* When the code reaches here, we either have already added
633 		 * the device, or its MAC address did not match.
634 		 */
635 		closedir(di);
636 		goto free_hotadd_ctx;
637 	}
638 	closedir(di);
639 retry:
640 	/* The device is still being initialized, retry after 1 second */
641 	rte_eal_alarm_set(1000000, netvsc_hotplug_retry, hot_ctx);
642 	return;
643 
644 free_hotadd_ctx:
645 	rte_spinlock_lock(&hv->hotadd_lock);
646 	LIST_REMOVE(hot_ctx, list);
647 	rte_spinlock_unlock(&hv->hotadd_lock);
648 
649 	rte_free(hot_ctx);
650 }
651 
652 static void
653 netvsc_hotadd_callback(const char *device_name, enum rte_dev_event_type type,
654 		       void *arg)
655 {
656 	struct hn_data *hv = arg;
657 	struct hv_hotadd_context *hot_ctx;
658 	struct rte_devargs *d;
659 	int ret;
660 
661 	PMD_DRV_LOG(INFO, "Device notification type=%d device_name=%s",
662 		    type, device_name);
663 
664 	switch (type) {
665 	case RTE_DEV_EVENT_ADD:
666 		/* if we already has a VF, don't check on hot add */
667 		if (hv->vf_ctx.vf_state > vf_removed)
668 			break;
669 
670 		hot_ctx = rte_zmalloc("NETVSC-HOTADD", sizeof(*hot_ctx),
671 				      rte_mem_page_size());
672 
673 		if (!hot_ctx) {
674 			PMD_DRV_LOG(ERR, "Failed to allocate hotadd context");
675 			return;
676 		}
677 
678 		hot_ctx->hv = hv;
679 		d = &hot_ctx->da;
680 
681 		ret = rte_devargs_parse(d, device_name);
682 		if (ret) {
683 			PMD_DRV_LOG(ERR,
684 				    "devargs parsing failed ret=%d", ret);
685 			goto free_ctx;
686 		}
687 
688 		if (!strcmp(d->bus->name, "pci")) {
689 			/* Start the process of figuring out if this
690 			 * PCI device is a VF device
691 			 */
692 			rte_spinlock_lock(&hv->hotadd_lock);
693 			LIST_INSERT_HEAD(&hv->hotadd_list, hot_ctx, list);
694 			rte_spinlock_unlock(&hv->hotadd_lock);
695 			rte_eal_alarm_set(1000000, netvsc_hotplug_retry, hot_ctx);
696 			return;
697 		}
698 
699 		/* We will switch to VF on RDNIS configure message
700 		 * sent from VSP
701 		 */
702 free_ctx:
703 		rte_free(hot_ctx);
704 		break;
705 
706 	default:
707 		break;
708 	}
709 }
710 
711 static int hn_dev_configure(struct rte_eth_dev *dev)
712 {
713 	struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
714 	struct rte_eth_rss_conf *rss_conf = &dev_conf->rx_adv_conf.rss_conf;
715 	const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
716 	const struct rte_eth_txmode *txmode = &dev_conf->txmode;
717 	struct hn_data *hv = dev->data->dev_private;
718 	uint64_t unsupported;
719 	int i, err, subchan;
720 
721 	PMD_INIT_FUNC_TRACE();
722 
723 	if (dev_conf->rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG)
724 		dev_conf->rxmode.offloads |= RTE_ETH_RX_OFFLOAD_RSS_HASH;
725 
726 	unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
727 	if (unsupported) {
728 		PMD_DRV_LOG(NOTICE,
729 			    "unsupported TX offload: %#" PRIx64,
730 			    unsupported);
731 		return -EINVAL;
732 	}
733 
734 	unsupported = rxmode->offloads & ~HN_RX_OFFLOAD_CAPS;
735 	if (unsupported) {
736 		PMD_DRV_LOG(NOTICE,
737 			    "unsupported RX offload: %#" PRIx64,
738 			    rxmode->offloads);
739 		return -EINVAL;
740 	}
741 
742 	hv->vlan_strip = !!(rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP);
743 
744 	err = hn_rndis_conf_offload(hv, txmode->offloads,
745 				    rxmode->offloads);
746 	if (err) {
747 		PMD_DRV_LOG(NOTICE,
748 			    "offload configure failed");
749 		return err;
750 	}
751 
752 	hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
753 				 dev->data->nb_tx_queues);
754 
755 	for (i = 0; i < NDIS_HASH_INDCNT; i++)
756 		hv->rss_ind[i] = i % dev->data->nb_rx_queues;
757 
758 	hn_rss_hash_init(hv, rss_conf);
759 
760 	subchan = hv->num_queues - 1;
761 	if (subchan > 0) {
762 		err = hn_subchan_configure(hv, subchan);
763 		if (err) {
764 			PMD_DRV_LOG(NOTICE,
765 				    "subchannel configuration failed");
766 			return err;
767 		}
768 
769 		err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
770 		if (err) {
771 			PMD_DRV_LOG(NOTICE,
772 				"rss disable failed");
773 			return err;
774 		}
775 
776 		if (rss_conf->rss_hf != 0) {
777 			err = hn_rndis_conf_rss(hv, 0);
778 			if (err) {
779 				PMD_DRV_LOG(NOTICE,
780 					    "initial RSS config failed");
781 				return err;
782 			}
783 		}
784 	}
785 
786 	return hn_vf_configure_locked(dev, dev_conf);
787 }
788 
789 static int hn_dev_stats_get(struct rte_eth_dev *dev,
790 			    struct rte_eth_stats *stats)
791 {
792 	unsigned int i;
793 
794 	hn_vf_stats_get(dev, stats);
795 
796 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
797 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
798 
799 		if (!txq)
800 			continue;
801 
802 		stats->opackets += txq->stats.packets;
803 		stats->obytes += txq->stats.bytes;
804 		stats->oerrors += txq->stats.errors;
805 
806 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
807 			stats->q_opackets[i] = txq->stats.packets;
808 			stats->q_obytes[i] = txq->stats.bytes;
809 		}
810 	}
811 
812 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
813 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
814 
815 		if (!rxq)
816 			continue;
817 
818 		stats->ipackets += rxq->stats.packets;
819 		stats->ibytes += rxq->stats.bytes;
820 		stats->ierrors += rxq->stats.errors;
821 		stats->imissed += rxq->stats.ring_full;
822 
823 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
824 			stats->q_ipackets[i] = rxq->stats.packets;
825 			stats->q_ibytes[i] = rxq->stats.bytes;
826 		}
827 	}
828 
829 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
830 	return 0;
831 }
832 
833 static int
834 hn_dev_stats_reset(struct rte_eth_dev *dev)
835 {
836 	unsigned int i;
837 
838 	PMD_INIT_FUNC_TRACE();
839 
840 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
841 		struct hn_tx_queue *txq = dev->data->tx_queues[i];
842 
843 		if (!txq)
844 			continue;
845 		memset(&txq->stats, 0, sizeof(struct hn_stats));
846 	}
847 
848 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
849 		struct hn_rx_queue *rxq = dev->data->rx_queues[i];
850 
851 		if (!rxq)
852 			continue;
853 
854 		memset(&rxq->stats, 0, sizeof(struct hn_stats));
855 	}
856 
857 	return 0;
858 }
859 
860 static int
861 hn_dev_xstats_reset(struct rte_eth_dev *dev)
862 {
863 	int ret;
864 
865 	ret = hn_dev_stats_reset(dev);
866 	if (ret != 0)
867 		return 0;
868 
869 	return hn_vf_xstats_reset(dev);
870 }
871 
872 static int
873 hn_dev_xstats_count(struct rte_eth_dev *dev)
874 {
875 	int ret, count;
876 
877 	count = dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings);
878 	count += dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
879 
880 	ret = hn_vf_xstats_get_names(dev, NULL, 0);
881 	if (ret < 0)
882 		return ret;
883 
884 	return count + ret;
885 }
886 
887 static int
888 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
889 			struct rte_eth_xstat_name *xstats_names,
890 			unsigned int limit)
891 {
892 	unsigned int i, t, count = 0;
893 	int ret;
894 
895 	if (!xstats_names)
896 		return hn_dev_xstats_count(dev);
897 
898 	/* Note: limit checked in rte_eth_xstats_names() */
899 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
900 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
901 
902 		if (!txq)
903 			continue;
904 
905 		if (count >= limit)
906 			break;
907 
908 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
909 			snprintf(xstats_names[count++].name,
910 				 RTE_ETH_XSTATS_NAME_SIZE,
911 				 "tx_q%u_%s", i, hn_stat_strings[t].name);
912 	}
913 
914 	for (i = 0; i < dev->data->nb_rx_queues; i++)  {
915 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
916 
917 		if (!rxq)
918 			continue;
919 
920 		if (count >= limit)
921 			break;
922 
923 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
924 			snprintf(xstats_names[count++].name,
925 				 RTE_ETH_XSTATS_NAME_SIZE,
926 				 "rx_q%u_%s", i,
927 				 hn_stat_strings[t].name);
928 	}
929 
930 	ret = hn_vf_xstats_get_names(dev, xstats_names + count,
931 				     limit - count);
932 	if (ret < 0)
933 		return ret;
934 
935 	return count + ret;
936 }
937 
938 static int
939 hn_dev_xstats_get(struct rte_eth_dev *dev,
940 		  struct rte_eth_xstat *xstats,
941 		  unsigned int n)
942 {
943 	unsigned int i, t, count = 0;
944 	const unsigned int nstats = hn_dev_xstats_count(dev);
945 	const char *stats;
946 	int ret;
947 
948 	PMD_INIT_FUNC_TRACE();
949 
950 	if (n < nstats)
951 		return nstats;
952 
953 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
954 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
955 
956 		if (!txq)
957 			continue;
958 
959 		stats = (const char *)&txq->stats;
960 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
961 			xstats[count].id = count;
962 			xstats[count].value = *(const uint64_t *)
963 				(stats + hn_stat_strings[t].offset);
964 		}
965 	}
966 
967 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
968 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
969 
970 		if (!rxq)
971 			continue;
972 
973 		stats = (const char *)&rxq->stats;
974 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
975 			xstats[count].id = count;
976 			xstats[count].value = *(const uint64_t *)
977 				(stats + hn_stat_strings[t].offset);
978 		}
979 	}
980 
981 	ret = hn_vf_xstats_get(dev, xstats, count, n);
982 	if (ret < 0)
983 		return ret;
984 
985 	return count + ret;
986 }
987 
988 static int
989 hn_dev_start(struct rte_eth_dev *dev)
990 {
991 	struct hn_data *hv = dev->data->dev_private;
992 	int error;
993 
994 	PMD_INIT_FUNC_TRACE();
995 
996 	/* Register to monitor hot plug events */
997 	error = rte_dev_event_callback_register(NULL, netvsc_hotadd_callback,
998 						hv);
999 	if (error) {
1000 		PMD_DRV_LOG(ERR, "failed to register device event callback");
1001 		return error;
1002 	}
1003 
1004 	error = hn_rndis_set_rxfilter(hv,
1005 				      NDIS_PACKET_TYPE_BROADCAST |
1006 				      NDIS_PACKET_TYPE_ALL_MULTICAST |
1007 				      NDIS_PACKET_TYPE_DIRECTED);
1008 	if (error)
1009 		return error;
1010 
1011 	error = hn_vf_start(dev);
1012 	if (error)
1013 		hn_rndis_set_rxfilter(hv, 0);
1014 
1015 	/* Initialize Link state */
1016 	if (error == 0)
1017 		hn_dev_link_update(dev, 0);
1018 
1019 	return error;
1020 }
1021 
1022 static int
1023 hn_dev_stop(struct rte_eth_dev *dev)
1024 {
1025 	struct hn_data *hv = dev->data->dev_private;
1026 
1027 	PMD_INIT_FUNC_TRACE();
1028 	dev->data->dev_started = 0;
1029 
1030 	rte_dev_event_callback_unregister(NULL, netvsc_hotadd_callback, hv);
1031 	hn_rndis_set_rxfilter(hv, 0);
1032 	return hn_vf_stop(dev);
1033 }
1034 
1035 static int
1036 hn_dev_close(struct rte_eth_dev *dev)
1037 {
1038 	int ret;
1039 	struct hn_data *hv = dev->data->dev_private;
1040 	struct hv_hotadd_context *hot_ctx;
1041 
1042 	PMD_INIT_FUNC_TRACE();
1043 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1044 		return 0;
1045 
1046 	rte_spinlock_lock(&hv->hotadd_lock);
1047 	while (!LIST_EMPTY(&hv->hotadd_list)) {
1048 		hot_ctx = LIST_FIRST(&hv->hotadd_list);
1049 		rte_eal_alarm_cancel(netvsc_hotplug_retry, hot_ctx);
1050 		LIST_REMOVE(hot_ctx, list);
1051 		rte_free(hot_ctx);
1052 	}
1053 	rte_spinlock_unlock(&hv->hotadd_lock);
1054 
1055 	ret = hn_vf_close(dev);
1056 	hn_dev_free_queues(dev);
1057 
1058 	return ret;
1059 }
1060 
1061 static const struct eth_dev_ops hn_eth_dev_ops = {
1062 	.dev_configure		= hn_dev_configure,
1063 	.dev_start		= hn_dev_start,
1064 	.dev_stop		= hn_dev_stop,
1065 	.dev_close		= hn_dev_close,
1066 	.dev_infos_get		= hn_dev_info_get,
1067 	.txq_info_get		= hn_dev_tx_queue_info,
1068 	.rxq_info_get		= hn_dev_rx_queue_info,
1069 	.dev_supported_ptypes_get = hn_vf_supported_ptypes,
1070 	.promiscuous_enable     = hn_dev_promiscuous_enable,
1071 	.promiscuous_disable    = hn_dev_promiscuous_disable,
1072 	.allmulticast_enable    = hn_dev_allmulticast_enable,
1073 	.allmulticast_disable   = hn_dev_allmulticast_disable,
1074 	.set_mc_addr_list	= hn_dev_mc_addr_list,
1075 	.reta_update		= hn_rss_reta_update,
1076 	.reta_query             = hn_rss_reta_query,
1077 	.rss_hash_update	= hn_rss_hash_update,
1078 	.rss_hash_conf_get      = hn_rss_hash_conf_get,
1079 	.tx_queue_setup		= hn_dev_tx_queue_setup,
1080 	.tx_queue_release	= hn_dev_tx_queue_release,
1081 	.tx_done_cleanup        = hn_dev_tx_done_cleanup,
1082 	.rx_queue_setup		= hn_dev_rx_queue_setup,
1083 	.rx_queue_release	= hn_dev_rx_queue_release,
1084 	.link_update		= hn_dev_link_update,
1085 	.stats_get		= hn_dev_stats_get,
1086 	.stats_reset            = hn_dev_stats_reset,
1087 	.xstats_get		= hn_dev_xstats_get,
1088 	.xstats_get_names	= hn_dev_xstats_get_names,
1089 	.xstats_reset		= hn_dev_xstats_reset,
1090 };
1091 
1092 /*
1093  * Setup connection between PMD and kernel.
1094  */
1095 static int
1096 hn_attach(struct hn_data *hv, unsigned int mtu)
1097 {
1098 	int error;
1099 
1100 	/* Attach NVS */
1101 	error = hn_nvs_attach(hv, mtu);
1102 	if (error)
1103 		goto failed_nvs;
1104 
1105 	/* Attach RNDIS */
1106 	error = hn_rndis_attach(hv);
1107 	if (error)
1108 		goto failed_rndis;
1109 
1110 	/*
1111 	 * NOTE:
1112 	 * Under certain conditions on certain versions of Hyper-V,
1113 	 * the RNDIS rxfilter is _not_ zero on the hypervisor side
1114 	 * after the successful RNDIS initialization.
1115 	 */
1116 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_NONE);
1117 	return 0;
1118 failed_rndis:
1119 	hn_nvs_detach(hv);
1120 failed_nvs:
1121 	return error;
1122 }
1123 
1124 static void
1125 hn_detach(struct hn_data *hv)
1126 {
1127 	hn_nvs_detach(hv);
1128 	hn_rndis_detach(hv);
1129 }
1130 
1131 static int
1132 eth_hn_dev_init(struct rte_eth_dev *eth_dev)
1133 {
1134 	struct hn_data *hv = eth_dev->data->dev_private;
1135 	struct rte_device *device = eth_dev->device;
1136 	struct rte_vmbus_device *vmbus;
1137 	unsigned int rxr_cnt;
1138 	int err, max_chan;
1139 
1140 	PMD_INIT_FUNC_TRACE();
1141 
1142 	rte_spinlock_init(&hv->hotadd_lock);
1143 	LIST_INIT(&hv->hotadd_list);
1144 
1145 	vmbus = container_of(device, struct rte_vmbus_device, device);
1146 	eth_dev->dev_ops = &hn_eth_dev_ops;
1147 	eth_dev->rx_queue_count = hn_dev_rx_queue_count;
1148 	eth_dev->rx_descriptor_status = hn_dev_rx_queue_status;
1149 	eth_dev->tx_descriptor_status = hn_dev_tx_descriptor_status;
1150 	eth_dev->tx_pkt_burst = &hn_xmit_pkts;
1151 	eth_dev->rx_pkt_burst = &hn_recv_pkts;
1152 
1153 	/*
1154 	 * for secondary processes, we don't initialize any further as primary
1155 	 * has already done this work.
1156 	 */
1157 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1158 		return 0;
1159 
1160 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1161 
1162 	/* Since Hyper-V only supports one MAC address */
1163 	eth_dev->data->mac_addrs = rte_calloc("hv_mac", HN_MAX_MAC_ADDRS,
1164 					      sizeof(struct rte_ether_addr), 0);
1165 	if (eth_dev->data->mac_addrs == NULL) {
1166 		PMD_INIT_LOG(ERR,
1167 			     "Failed to allocate memory store MAC addresses");
1168 		return -ENOMEM;
1169 	}
1170 
1171 	hv->vmbus = vmbus;
1172 	hv->rxbuf_res = vmbus->resource[HV_RECV_BUF_MAP];
1173 	hv->chim_res  = vmbus->resource[HV_SEND_BUF_MAP];
1174 	hv->port_id = eth_dev->data->port_id;
1175 	hv->latency = HN_CHAN_LATENCY_NS;
1176 	hv->rx_copybreak = HN_RXCOPY_THRESHOLD;
1177 	hv->tx_copybreak = HN_TXCOPY_THRESHOLD;
1178 	hv->rx_extmbuf_enable = HN_RX_EXTMBUF_ENABLE;
1179 	hv->max_queues = 1;
1180 
1181 	rte_rwlock_init(&hv->vf_lock);
1182 	hv->vf_ctx.vf_vsc_switched = false;
1183 	hv->vf_ctx.vf_vsp_reported = false;
1184 	hv->vf_ctx.vf_attached = false;
1185 	hv->vf_ctx.vf_state = vf_unknown;
1186 
1187 	err = hn_parse_args(eth_dev);
1188 	if (err)
1189 		return err;
1190 
1191 	strlcpy(hv->owner.name, eth_dev->device->name,
1192 		RTE_ETH_MAX_OWNER_NAME_LEN);
1193 	err = rte_eth_dev_owner_new(&hv->owner.id);
1194 	if (err) {
1195 		PMD_INIT_LOG(ERR, "Can not get owner id");
1196 		return err;
1197 	}
1198 
1199 	/* Initialize primary channel input for control operations */
1200 	err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
1201 	if (err)
1202 		return err;
1203 
1204 	rte_vmbus_set_latency(hv->vmbus, hv->channels[0], hv->latency);
1205 
1206 	hv->primary = hn_rx_queue_alloc(hv, 0,
1207 					eth_dev->device->numa_node);
1208 
1209 	if (!hv->primary)
1210 		return -ENOMEM;
1211 
1212 	err = hn_attach(hv, RTE_ETHER_MTU);
1213 	if  (err)
1214 		goto failed;
1215 
1216 	err = hn_chim_init(eth_dev);
1217 	if (err)
1218 		goto failed;
1219 
1220 	err = hn_rndis_get_eaddr(hv, eth_dev->data->mac_addrs->addr_bytes);
1221 	if (err)
1222 		goto failed;
1223 
1224 	/* Multi queue requires later versions of windows server */
1225 	if (hv->nvs_ver < NVS_VERSION_5)
1226 		return 0;
1227 
1228 	max_chan = rte_vmbus_max_channels(vmbus);
1229 	PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
1230 	if (max_chan <= 0)
1231 		goto failed;
1232 
1233 	if (hn_rndis_query_rsscaps(hv, &rxr_cnt) != 0)
1234 		rxr_cnt = 1;
1235 
1236 	hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
1237 
1238 	/* If VF was reported but not added, do it now */
1239 	if (hv->vf_ctx.vf_vsp_reported && !hv->vf_ctx.vf_vsc_switched) {
1240 		PMD_INIT_LOG(DEBUG, "Adding VF device");
1241 
1242 		err = hn_vf_add(eth_dev, hv);
1243 	}
1244 
1245 	return 0;
1246 
1247 failed:
1248 	PMD_INIT_LOG(NOTICE, "device init failed");
1249 
1250 	hn_chim_uninit(eth_dev);
1251 	hn_detach(hv);
1252 	return err;
1253 }
1254 
1255 static int
1256 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
1257 {
1258 	struct hn_data *hv = eth_dev->data->dev_private;
1259 	int ret, ret_stop;
1260 
1261 	PMD_INIT_FUNC_TRACE();
1262 
1263 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1264 		return 0;
1265 
1266 	ret_stop = hn_dev_stop(eth_dev);
1267 	hn_dev_close(eth_dev);
1268 
1269 	free(hv->vf_devargs);
1270 	hv->vf_devargs = NULL;
1271 
1272 	hn_detach(hv);
1273 	hn_chim_uninit(eth_dev);
1274 	rte_vmbus_chan_close(hv->primary->chan);
1275 	rte_free(hv->primary);
1276 	ret = rte_eth_dev_owner_delete(hv->owner.id);
1277 	if (ret != 0)
1278 		return ret;
1279 
1280 	return ret_stop;
1281 }
1282 
1283 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
1284 			struct rte_vmbus_device *dev)
1285 {
1286 	struct rte_eth_dev *eth_dev;
1287 	int ret;
1288 
1289 	PMD_INIT_FUNC_TRACE();
1290 
1291 	ret = rte_dev_event_monitor_start();
1292 	if (ret) {
1293 		PMD_DRV_LOG(ERR, "Failed to start device event monitoring");
1294 		return ret;
1295 	}
1296 
1297 	eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
1298 	if (!eth_dev)
1299 		return -ENOMEM;
1300 
1301 	ret = eth_hn_dev_init(eth_dev);
1302 	if (ret) {
1303 		eth_dev_vmbus_release(eth_dev);
1304 		rte_dev_event_monitor_stop();
1305 	} else {
1306 		rte_eth_dev_probing_finish(eth_dev);
1307 	}
1308 
1309 	return ret;
1310 }
1311 
1312 static int eth_hn_remove(struct rte_vmbus_device *dev)
1313 {
1314 	struct rte_eth_dev *eth_dev;
1315 	int ret;
1316 
1317 	PMD_INIT_FUNC_TRACE();
1318 
1319 	eth_dev = rte_eth_dev_allocated(dev->device.name);
1320 	if (!eth_dev)
1321 		return 0; /* port already released */
1322 
1323 	ret = eth_hn_dev_uninit(eth_dev);
1324 	if (ret)
1325 		return ret;
1326 
1327 	eth_dev_vmbus_release(eth_dev);
1328 	rte_dev_event_monitor_stop();
1329 	return 0;
1330 }
1331 
1332 /* Network device GUID */
1333 static const rte_uuid_t hn_net_ids[] = {
1334 	/*  f8615163-df3e-46c5-913f-f2d2f965ed0e */
1335 	RTE_UUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x913f, 0xf2d2f965ed0eULL),
1336 	{ 0 }
1337 };
1338 
1339 static struct rte_vmbus_driver rte_netvsc_pmd = {
1340 	.id_table = hn_net_ids,
1341 	.probe = eth_hn_probe,
1342 	.remove = eth_hn_remove,
1343 };
1344 
1345 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
1346 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
1347 RTE_LOG_REGISTER_SUFFIX(hn_logtype_init, init, NOTICE);
1348 RTE_LOG_REGISTER_SUFFIX(hn_logtype_driver, driver, NOTICE);
1349 RTE_PMD_REGISTER_PARAM_STRING(net_netvsc,
1350 			      NETVSC_ARG_LATENCY "=<uint32> "
1351 			      NETVSC_ARG_RXBREAK "=<uint32> "
1352 			      NETVSC_ARG_TXBREAK "=<uint32> "
1353 			      NETVSC_ARG_RX_EXTMBUF_ENABLE "=<0|1>");
1354