xref: /dpdk/drivers/net/netvsc/hn_ethdev.c (revision b1cf08ae095026ec9befe2aeddf4ecd1785a37a6)
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 
13 #include <rte_ethdev.h>
14 #include <rte_memcpy.h>
15 #include <rte_string_fns.h>
16 #include <rte_memzone.h>
17 #include <rte_devargs.h>
18 #include <rte_malloc.h>
19 #include <rte_kvargs.h>
20 #include <rte_atomic.h>
21 #include <rte_branch_prediction.h>
22 #include <rte_ether.h>
23 #include <rte_ethdev_driver.h>
24 #include <rte_cycles.h>
25 #include <rte_errno.h>
26 #include <rte_memory.h>
27 #include <rte_eal.h>
28 #include <rte_dev.h>
29 #include <rte_bus_vmbus.h>
30 
31 #include "hn_logs.h"
32 #include "hn_var.h"
33 #include "hn_rndis.h"
34 #include "hn_nvs.h"
35 #include "ndis.h"
36 
37 #define HN_TX_OFFLOAD_CAPS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
38 			    DEV_TX_OFFLOAD_TCP_CKSUM  | \
39 			    DEV_TX_OFFLOAD_UDP_CKSUM  | \
40 			    DEV_TX_OFFLOAD_TCP_TSO    | \
41 			    DEV_TX_OFFLOAD_MULTI_SEGS | \
42 			    DEV_TX_OFFLOAD_VLAN_INSERT)
43 
44 #define HN_RX_OFFLOAD_CAPS (DEV_RX_OFFLOAD_CHECKSUM | \
45 			    DEV_RX_OFFLOAD_VLAN_STRIP | \
46 			    DEV_RX_OFFLOAD_RSS_HASH)
47 
48 int hn_logtype_init;
49 int hn_logtype_driver;
50 
51 struct hn_xstats_name_off {
52 	char name[RTE_ETH_XSTATS_NAME_SIZE];
53 	unsigned int offset;
54 };
55 
56 static const struct hn_xstats_name_off hn_stat_strings[] = {
57 	{ "good_packets",           offsetof(struct hn_stats, packets) },
58 	{ "good_bytes",             offsetof(struct hn_stats, bytes) },
59 	{ "errors",                 offsetof(struct hn_stats, errors) },
60 	{ "ring full",              offsetof(struct hn_stats, ring_full) },
61 	{ "multicast_packets",      offsetof(struct hn_stats, multicast) },
62 	{ "broadcast_packets",      offsetof(struct hn_stats, broadcast) },
63 	{ "undersize_packets",      offsetof(struct hn_stats, size_bins[0]) },
64 	{ "size_64_packets",        offsetof(struct hn_stats, size_bins[1]) },
65 	{ "size_65_127_packets",    offsetof(struct hn_stats, size_bins[2]) },
66 	{ "size_128_255_packets",   offsetof(struct hn_stats, size_bins[3]) },
67 	{ "size_256_511_packets",   offsetof(struct hn_stats, size_bins[4]) },
68 	{ "size_512_1023_packets",  offsetof(struct hn_stats, size_bins[5]) },
69 	{ "size_1024_1518_packets", offsetof(struct hn_stats, size_bins[6]) },
70 	{ "size_1519_max_packets",  offsetof(struct hn_stats, size_bins[7]) },
71 };
72 
73 /* The default RSS key.
74  * This value is the same as MLX5 so that flows will be
75  * received on same path for both VF ans synthetic NIC.
76  */
77 static const uint8_t rss_default_key[NDIS_HASH_KEYSIZE_TOEPLITZ] = {
78 	0x2c, 0xc6, 0x81, 0xd1,	0x5b, 0xdb, 0xf4, 0xf7,
79 	0xfc, 0xa2, 0x83, 0x19,	0xdb, 0x1a, 0x3e, 0x94,
80 	0x6b, 0x9e, 0x38, 0xd9,	0x2c, 0x9c, 0x03, 0xd1,
81 	0xad, 0x99, 0x44, 0xa7,	0xd9, 0x56, 0x3d, 0x59,
82 	0x06, 0x3c, 0x25, 0xf3,	0xfc, 0x1f, 0xdc, 0x2a,
83 };
84 
85 static struct rte_eth_dev *
86 eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
87 {
88 	struct rte_eth_dev *eth_dev;
89 	const char *name;
90 
91 	if (!dev)
92 		return NULL;
93 
94 	name = dev->device.name;
95 
96 	if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
97 		eth_dev = rte_eth_dev_allocate(name);
98 		if (!eth_dev) {
99 			PMD_DRV_LOG(NOTICE, "can not allocate rte ethdev");
100 			return NULL;
101 		}
102 
103 		if (private_data_size) {
104 			eth_dev->data->dev_private =
105 				rte_zmalloc_socket(name, private_data_size,
106 						     RTE_CACHE_LINE_SIZE, dev->device.numa_node);
107 			if (!eth_dev->data->dev_private) {
108 				PMD_DRV_LOG(NOTICE, "can not allocate driver data");
109 				rte_eth_dev_release_port(eth_dev);
110 				return NULL;
111 			}
112 		}
113 	} else {
114 		eth_dev = rte_eth_dev_attach_secondary(name);
115 		if (!eth_dev) {
116 			PMD_DRV_LOG(NOTICE, "can not attach secondary");
117 			return NULL;
118 		}
119 	}
120 
121 	eth_dev->device = &dev->device;
122 
123 	/* interrupt is simulated */
124 	dev->intr_handle.type = RTE_INTR_HANDLE_EXT;
125 	eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
126 	eth_dev->intr_handle = &dev->intr_handle;
127 
128 	/* allow ethdev to remove on close */
129 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
130 
131 	return eth_dev;
132 }
133 
134 static void
135 eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
136 {
137 	/* mac_addrs must not be freed alone because part of dev_private */
138 	eth_dev->data->mac_addrs = NULL;
139 	/* free ether device */
140 	rte_eth_dev_release_port(eth_dev);
141 
142 	eth_dev->device = NULL;
143 	eth_dev->intr_handle = NULL;
144 }
145 
146 /* handle "latency=X" from devargs */
147 static int hn_set_latency(const char *key, const char *value, void *opaque)
148 {
149 	struct hn_data *hv = opaque;
150 	char *endp = NULL;
151 	unsigned long lat;
152 
153 	errno = 0;
154 	lat = strtoul(value, &endp, 0);
155 
156 	if (*value == '\0' || *endp != '\0') {
157 		PMD_DRV_LOG(ERR, "invalid parameter %s=%s", key, value);
158 		return -EINVAL;
159 	}
160 
161 	PMD_DRV_LOG(DEBUG, "set latency %lu usec", lat);
162 
163 	hv->latency = lat * 1000;	/* usec to nsec */
164 	return 0;
165 }
166 
167 /* Parse device arguments */
168 static int hn_parse_args(const struct rte_eth_dev *dev)
169 {
170 	struct hn_data *hv = dev->data->dev_private;
171 	struct rte_devargs *devargs = dev->device->devargs;
172 	static const char * const valid_keys[] = {
173 		"latency",
174 		NULL
175 	};
176 	struct rte_kvargs *kvlist;
177 	int ret;
178 
179 	if (!devargs)
180 		return 0;
181 
182 	PMD_INIT_LOG(DEBUG, "device args %s %s",
183 		     devargs->name, devargs->args);
184 
185 	kvlist = rte_kvargs_parse(devargs->args, valid_keys);
186 	if (!kvlist) {
187 		PMD_DRV_LOG(NOTICE, "invalid parameters");
188 		return -EINVAL;
189 	}
190 
191 	ret = rte_kvargs_process(kvlist, "latency", hn_set_latency, hv);
192 	if (ret)
193 		PMD_DRV_LOG(ERR, "Unable to process latency arg\n");
194 
195 	rte_kvargs_free(kvlist);
196 	return ret;
197 }
198 
199 /* Update link status.
200  * Note: the DPDK definition of "wait_to_complete"
201  *   means block this call until link is up.
202  *   which is not worth supporting.
203  */
204 int
205 hn_dev_link_update(struct rte_eth_dev *dev,
206 		   int wait_to_complete)
207 {
208 	struct hn_data *hv = dev->data->dev_private;
209 	struct rte_eth_link link, old;
210 	int error;
211 
212 	old = dev->data->dev_link;
213 
214 	error = hn_rndis_get_linkstatus(hv);
215 	if (error)
216 		return error;
217 
218 	hn_rndis_get_linkspeed(hv);
219 
220 	hn_vf_link_update(dev, wait_to_complete);
221 
222 	link = (struct rte_eth_link) {
223 		.link_duplex = ETH_LINK_FULL_DUPLEX,
224 		.link_autoneg = ETH_LINK_SPEED_FIXED,
225 		.link_speed = hv->link_speed / 10000,
226 	};
227 
228 	if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
229 		link.link_status = ETH_LINK_UP;
230 	else
231 		link.link_status = ETH_LINK_DOWN;
232 
233 	if (old.link_status == link.link_status)
234 		return 0;
235 
236 	PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
237 		     (link.link_status == ETH_LINK_UP) ? "up" : "down");
238 
239 	return rte_eth_linkstatus_set(dev, &link);
240 }
241 
242 static int hn_dev_info_get(struct rte_eth_dev *dev,
243 			   struct rte_eth_dev_info *dev_info)
244 {
245 	struct hn_data *hv = dev->data->dev_private;
246 	int rc;
247 
248 	dev_info->speed_capa = ETH_LINK_SPEED_10G;
249 	dev_info->min_rx_bufsize = HN_MIN_RX_BUF_SIZE;
250 	dev_info->max_rx_pktlen  = HN_MAX_XFER_LEN;
251 	dev_info->max_mac_addrs  = 1;
252 
253 	dev_info->hash_key_size = NDIS_HASH_KEYSIZE_TOEPLITZ;
254 	dev_info->flow_type_rss_offloads = hv->rss_offloads;
255 	dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
256 
257 	dev_info->max_rx_queues = hv->max_queues;
258 	dev_info->max_tx_queues = hv->max_queues;
259 
260 	rc = hn_rndis_get_offload(hv, dev_info);
261 	if (rc != 0)
262 		return rc;
263 
264 	rc = hn_vf_info_get(hv, dev_info);
265 	if (rc != 0)
266 		return rc;
267 
268 	return 0;
269 }
270 
271 static int hn_rss_reta_update(struct rte_eth_dev *dev,
272 			      struct rte_eth_rss_reta_entry64 *reta_conf,
273 			      uint16_t reta_size)
274 {
275 	struct hn_data *hv = dev->data->dev_private;
276 	unsigned int i;
277 	int err;
278 
279 	PMD_INIT_FUNC_TRACE();
280 
281 	if (reta_size != NDIS_HASH_INDCNT) {
282 		PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
283 		return -EINVAL;
284 	}
285 
286 	for (i = 0; i < NDIS_HASH_INDCNT; i++) {
287 		uint16_t idx = i / RTE_RETA_GROUP_SIZE;
288 		uint16_t shift = i % RTE_RETA_GROUP_SIZE;
289 		uint64_t mask = (uint64_t)1 << shift;
290 
291 		if (reta_conf[idx].mask & mask)
292 			hv->rss_ind[i] = reta_conf[idx].reta[shift];
293 	}
294 
295 	err = hn_rndis_conf_rss(hv, 0);
296 	if (err) {
297 		PMD_DRV_LOG(NOTICE,
298 			    "reta reconfig failed");
299 		return err;
300 	}
301 
302 	return hn_vf_reta_hash_update(dev, reta_conf, reta_size);
303 }
304 
305 static int hn_rss_reta_query(struct rte_eth_dev *dev,
306 			     struct rte_eth_rss_reta_entry64 *reta_conf,
307 			     uint16_t reta_size)
308 {
309 	struct hn_data *hv = dev->data->dev_private;
310 	unsigned int i;
311 
312 	PMD_INIT_FUNC_TRACE();
313 
314 	if (reta_size != NDIS_HASH_INDCNT) {
315 		PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
316 		return -EINVAL;
317 	}
318 
319 	for (i = 0; i < NDIS_HASH_INDCNT; i++) {
320 		uint16_t idx = i / RTE_RETA_GROUP_SIZE;
321 		uint16_t shift = i % RTE_RETA_GROUP_SIZE;
322 		uint64_t mask = (uint64_t)1 << shift;
323 
324 		if (reta_conf[idx].mask & mask)
325 			reta_conf[idx].reta[shift] = hv->rss_ind[i];
326 	}
327 	return 0;
328 }
329 
330 static void hn_rss_hash_init(struct hn_data *hv,
331 			     const struct rte_eth_rss_conf *rss_conf)
332 {
333 	/* Convert from DPDK RSS hash flags to NDIS hash flags */
334 	hv->rss_hash = NDIS_HASH_FUNCTION_TOEPLITZ;
335 
336 	if (rss_conf->rss_hf & ETH_RSS_IPV4)
337 		hv->rss_hash |= NDIS_HASH_IPV4;
338 	if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
339 		hv->rss_hash |= NDIS_HASH_TCP_IPV4;
340 	if (rss_conf->rss_hf & ETH_RSS_IPV6)
341 		hv->rss_hash |=  NDIS_HASH_IPV6;
342 	if (rss_conf->rss_hf & ETH_RSS_IPV6_EX)
343 		hv->rss_hash |=  NDIS_HASH_IPV6_EX;
344 	if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
345 		hv->rss_hash |= NDIS_HASH_TCP_IPV6;
346 	if (rss_conf->rss_hf & ETH_RSS_IPV6_TCP_EX)
347 		hv->rss_hash |= NDIS_HASH_TCP_IPV6_EX;
348 
349 	memcpy(hv->rss_key, rss_conf->rss_key ? : rss_default_key,
350 	       NDIS_HASH_KEYSIZE_TOEPLITZ);
351 }
352 
353 static int hn_rss_hash_update(struct rte_eth_dev *dev,
354 			      struct rte_eth_rss_conf *rss_conf)
355 {
356 	struct hn_data *hv = dev->data->dev_private;
357 	int err;
358 
359 	PMD_INIT_FUNC_TRACE();
360 
361 	err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
362 	if (err) {
363 		PMD_DRV_LOG(NOTICE,
364 			    "rss disable failed");
365 		return err;
366 	}
367 
368 	hn_rss_hash_init(hv, rss_conf);
369 
370 	err = hn_rndis_conf_rss(hv, 0);
371 	if (err) {
372 		PMD_DRV_LOG(NOTICE,
373 			    "rss reconfig failed (RSS disabled)");
374 		return err;
375 	}
376 
377 
378 	return hn_vf_rss_hash_update(dev, rss_conf);
379 }
380 
381 static int hn_rss_hash_conf_get(struct rte_eth_dev *dev,
382 				struct rte_eth_rss_conf *rss_conf)
383 {
384 	struct hn_data *hv = dev->data->dev_private;
385 
386 	PMD_INIT_FUNC_TRACE();
387 
388 	if (hv->ndis_ver < NDIS_VERSION_6_20) {
389 		PMD_DRV_LOG(DEBUG, "RSS not supported on this host");
390 		return -EOPNOTSUPP;
391 	}
392 
393 	rss_conf->rss_key_len = NDIS_HASH_KEYSIZE_TOEPLITZ;
394 	if (rss_conf->rss_key)
395 		memcpy(rss_conf->rss_key, hv->rss_key,
396 		       NDIS_HASH_KEYSIZE_TOEPLITZ);
397 
398 	rss_conf->rss_hf = 0;
399 	if (hv->rss_hash & NDIS_HASH_IPV4)
400 		rss_conf->rss_hf |= ETH_RSS_IPV4;
401 
402 	if (hv->rss_hash & NDIS_HASH_TCP_IPV4)
403 		rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
404 
405 	if (hv->rss_hash & NDIS_HASH_IPV6)
406 		rss_conf->rss_hf |= ETH_RSS_IPV6;
407 
408 	if (hv->rss_hash & NDIS_HASH_IPV6_EX)
409 		rss_conf->rss_hf |= ETH_RSS_IPV6_EX;
410 
411 	if (hv->rss_hash & NDIS_HASH_TCP_IPV6)
412 		rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
413 
414 	if (hv->rss_hash & NDIS_HASH_TCP_IPV6_EX)
415 		rss_conf->rss_hf |= ETH_RSS_IPV6_TCP_EX;
416 
417 	return 0;
418 }
419 
420 static int
421 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
422 {
423 	struct hn_data *hv = dev->data->dev_private;
424 
425 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
426 	return hn_vf_promiscuous_enable(dev);
427 }
428 
429 static int
430 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
431 {
432 	struct hn_data *hv = dev->data->dev_private;
433 	uint32_t filter;
434 
435 	filter = NDIS_PACKET_TYPE_DIRECTED | NDIS_PACKET_TYPE_BROADCAST;
436 	if (dev->data->all_multicast)
437 		filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
438 	hn_rndis_set_rxfilter(hv, filter);
439 	return hn_vf_promiscuous_disable(dev);
440 }
441 
442 static int
443 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
444 {
445 	struct hn_data *hv = dev->data->dev_private;
446 
447 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
448 			      NDIS_PACKET_TYPE_ALL_MULTICAST |
449 			NDIS_PACKET_TYPE_BROADCAST);
450 	return hn_vf_allmulticast_enable(dev);
451 }
452 
453 static int
454 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
455 {
456 	struct hn_data *hv = dev->data->dev_private;
457 
458 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
459 			     NDIS_PACKET_TYPE_BROADCAST);
460 	return hn_vf_allmulticast_disable(dev);
461 }
462 
463 static int
464 hn_dev_mc_addr_list(struct rte_eth_dev *dev,
465 		     struct rte_ether_addr *mc_addr_set,
466 		     uint32_t nb_mc_addr)
467 {
468 	/* No filtering on the synthetic path, but can do it on VF */
469 	return hn_vf_mc_addr_list(dev, mc_addr_set, nb_mc_addr);
470 }
471 
472 /* Setup shared rx/tx queue data */
473 static int hn_subchan_configure(struct hn_data *hv,
474 				uint32_t subchan)
475 {
476 	struct vmbus_channel *primary = hn_primary_chan(hv);
477 	int err;
478 	unsigned int retry = 0;
479 
480 	PMD_DRV_LOG(DEBUG,
481 		    "open %u subchannels", subchan);
482 
483 	/* Send create sub channels command */
484 	err = hn_nvs_alloc_subchans(hv, &subchan);
485 	if (err)
486 		return  err;
487 
488 	while (subchan > 0) {
489 		struct vmbus_channel *new_sc;
490 		uint16_t chn_index;
491 
492 		err = rte_vmbus_subchan_open(primary, &new_sc);
493 		if (err == -ENOENT && ++retry < 1000) {
494 			/* This can happen if not ready yet */
495 			rte_delay_ms(10);
496 			continue;
497 		}
498 
499 		if (err) {
500 			PMD_DRV_LOG(ERR,
501 				    "open subchannel failed: %d", err);
502 			return err;
503 		}
504 
505 		rte_vmbus_set_latency(hv->vmbus, new_sc, hv->latency);
506 
507 		retry = 0;
508 		chn_index = rte_vmbus_sub_channel_index(new_sc);
509 		if (chn_index == 0 || chn_index > hv->max_queues) {
510 			PMD_DRV_LOG(ERR,
511 				    "Invalid subchannel offermsg channel %u",
512 				    chn_index);
513 			return -EIO;
514 		}
515 
516 		PMD_DRV_LOG(DEBUG, "new sub channel %u", chn_index);
517 		hv->channels[chn_index] = new_sc;
518 		--subchan;
519 	}
520 
521 	return err;
522 }
523 
524 static int hn_dev_configure(struct rte_eth_dev *dev)
525 {
526 	struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
527 	struct rte_eth_rss_conf *rss_conf = &dev_conf->rx_adv_conf.rss_conf;
528 	const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
529 	const struct rte_eth_txmode *txmode = &dev_conf->txmode;
530 	struct hn_data *hv = dev->data->dev_private;
531 	uint64_t unsupported;
532 	int i, err, subchan;
533 
534 	PMD_INIT_FUNC_TRACE();
535 
536 	if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
537 		dev_conf->rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
538 
539 	unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
540 	if (unsupported) {
541 		PMD_DRV_LOG(NOTICE,
542 			    "unsupported TX offload: %#" PRIx64,
543 			    unsupported);
544 		return -EINVAL;
545 	}
546 
547 	unsupported = rxmode->offloads & ~HN_RX_OFFLOAD_CAPS;
548 	if (unsupported) {
549 		PMD_DRV_LOG(NOTICE,
550 			    "unsupported RX offload: %#" PRIx64,
551 			    rxmode->offloads);
552 		return -EINVAL;
553 	}
554 
555 	hv->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
556 
557 	err = hn_rndis_conf_offload(hv, txmode->offloads,
558 				    rxmode->offloads);
559 	if (err) {
560 		PMD_DRV_LOG(NOTICE,
561 			    "offload configure failed");
562 		return err;
563 	}
564 
565 	hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
566 				 dev->data->nb_tx_queues);
567 
568 	for (i = 0; i < NDIS_HASH_INDCNT; i++)
569 		hv->rss_ind[i] = i % hv->num_queues;
570 
571 	hn_rss_hash_init(hv, rss_conf);
572 
573 	subchan = hv->num_queues - 1;
574 	if (subchan > 0) {
575 		err = hn_subchan_configure(hv, subchan);
576 		if (err) {
577 			PMD_DRV_LOG(NOTICE,
578 				    "subchannel configuration failed");
579 			return err;
580 		}
581 
582 		err = hn_rndis_conf_rss(hv, 0);
583 		if (err) {
584 			PMD_DRV_LOG(NOTICE,
585 				    "initial RSS config failed");
586 			return err;
587 		}
588 	}
589 
590 	return hn_vf_configure(dev, dev_conf);
591 }
592 
593 static int hn_dev_stats_get(struct rte_eth_dev *dev,
594 			    struct rte_eth_stats *stats)
595 {
596 	unsigned int i;
597 
598 	hn_vf_stats_get(dev, stats);
599 
600 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
601 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
602 
603 		if (!txq)
604 			continue;
605 
606 		stats->opackets += txq->stats.packets;
607 		stats->obytes += txq->stats.bytes;
608 		stats->oerrors += txq->stats.errors;
609 
610 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
611 			stats->q_opackets[i] = txq->stats.packets;
612 			stats->q_obytes[i] = txq->stats.bytes;
613 		}
614 	}
615 
616 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
617 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
618 
619 		if (!rxq)
620 			continue;
621 
622 		stats->ipackets += rxq->stats.packets;
623 		stats->ibytes += rxq->stats.bytes;
624 		stats->ierrors += rxq->stats.errors;
625 		stats->imissed += rxq->stats.ring_full;
626 
627 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
628 			stats->q_ipackets[i] = rxq->stats.packets;
629 			stats->q_ibytes[i] = rxq->stats.bytes;
630 		}
631 	}
632 
633 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
634 	return 0;
635 }
636 
637 static int
638 hn_dev_stats_reset(struct rte_eth_dev *dev)
639 {
640 	unsigned int i;
641 
642 	PMD_INIT_FUNC_TRACE();
643 
644 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
645 		struct hn_tx_queue *txq = dev->data->tx_queues[i];
646 
647 		if (!txq)
648 			continue;
649 		memset(&txq->stats, 0, sizeof(struct hn_stats));
650 	}
651 
652 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
653 		struct hn_rx_queue *rxq = dev->data->rx_queues[i];
654 
655 		if (!rxq)
656 			continue;
657 
658 		memset(&rxq->stats, 0, sizeof(struct hn_stats));
659 	}
660 
661 	return 0;
662 }
663 
664 static int
665 hn_dev_xstats_reset(struct rte_eth_dev *dev)
666 {
667 	int ret;
668 
669 	ret = hn_dev_stats_reset(dev);
670 	if (ret != 0)
671 		return 0;
672 
673 	return hn_vf_xstats_reset(dev);
674 }
675 
676 static int
677 hn_dev_xstats_count(struct rte_eth_dev *dev)
678 {
679 	int ret, count;
680 
681 	count = dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings);
682 	count += dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
683 
684 	ret = hn_vf_xstats_get_names(dev, NULL, 0);
685 	if (ret < 0)
686 		return ret;
687 
688 	return count + ret;
689 }
690 
691 static int
692 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
693 			struct rte_eth_xstat_name *xstats_names,
694 			unsigned int limit)
695 {
696 	unsigned int i, t, count = 0;
697 	int ret;
698 
699 	if (!xstats_names)
700 		return hn_dev_xstats_count(dev);
701 
702 	/* Note: limit checked in rte_eth_xstats_names() */
703 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
704 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
705 
706 		if (!txq)
707 			continue;
708 
709 		if (count >= limit)
710 			break;
711 
712 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
713 			snprintf(xstats_names[count++].name,
714 				 RTE_ETH_XSTATS_NAME_SIZE,
715 				 "tx_q%u_%s", i, hn_stat_strings[t].name);
716 	}
717 
718 	for (i = 0; i < dev->data->nb_rx_queues; i++)  {
719 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
720 
721 		if (!rxq)
722 			continue;
723 
724 		if (count >= limit)
725 			break;
726 
727 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
728 			snprintf(xstats_names[count++].name,
729 				 RTE_ETH_XSTATS_NAME_SIZE,
730 				 "rx_q%u_%s", i,
731 				 hn_stat_strings[t].name);
732 	}
733 
734 	ret = hn_vf_xstats_get_names(dev, xstats_names + count,
735 				     limit - count);
736 	if (ret < 0)
737 		return ret;
738 
739 	return count + ret;
740 }
741 
742 static int
743 hn_dev_xstats_get(struct rte_eth_dev *dev,
744 		  struct rte_eth_xstat *xstats,
745 		  unsigned int n)
746 {
747 	unsigned int i, t, count = 0;
748 	const unsigned int nstats = hn_dev_xstats_count(dev);
749 	const char *stats;
750 	int ret;
751 
752 	PMD_INIT_FUNC_TRACE();
753 
754 	if (n < nstats)
755 		return nstats;
756 
757 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
758 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
759 
760 		if (!txq)
761 			continue;
762 
763 		stats = (const char *)&txq->stats;
764 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
765 			xstats[count].id = count;
766 			xstats[count].value = *(const uint64_t *)
767 				(stats + hn_stat_strings[t].offset);
768 		}
769 	}
770 
771 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
772 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
773 
774 		if (!rxq)
775 			continue;
776 
777 		stats = (const char *)&rxq->stats;
778 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
779 			xstats[count].id = count;
780 			xstats[count].value = *(const uint64_t *)
781 				(stats + hn_stat_strings[t].offset);
782 		}
783 	}
784 
785 	ret = hn_vf_xstats_get(dev, xstats, count, n);
786 	if (ret < 0)
787 		return ret;
788 
789 	return count + ret;
790 }
791 
792 static int
793 hn_dev_start(struct rte_eth_dev *dev)
794 {
795 	struct hn_data *hv = dev->data->dev_private;
796 	int error;
797 
798 	PMD_INIT_FUNC_TRACE();
799 
800 	error = hn_rndis_set_rxfilter(hv,
801 				      NDIS_PACKET_TYPE_BROADCAST |
802 				      NDIS_PACKET_TYPE_ALL_MULTICAST |
803 				      NDIS_PACKET_TYPE_DIRECTED);
804 	if (error)
805 		return error;
806 
807 	error = hn_vf_start(dev);
808 	if (error)
809 		hn_rndis_set_rxfilter(hv, 0);
810 
811 	return error;
812 }
813 
814 static void
815 hn_dev_stop(struct rte_eth_dev *dev)
816 {
817 	struct hn_data *hv = dev->data->dev_private;
818 
819 	PMD_INIT_FUNC_TRACE();
820 
821 	hn_rndis_set_rxfilter(hv, 0);
822 	hn_vf_stop(dev);
823 }
824 
825 static void
826 hn_dev_close(struct rte_eth_dev *dev)
827 {
828 	PMD_INIT_FUNC_TRACE();
829 
830 	hn_vf_close(dev);
831 	hn_dev_free_queues(dev);
832 }
833 
834 static const struct eth_dev_ops hn_eth_dev_ops = {
835 	.dev_configure		= hn_dev_configure,
836 	.dev_start		= hn_dev_start,
837 	.dev_stop		= hn_dev_stop,
838 	.dev_close		= hn_dev_close,
839 	.dev_infos_get		= hn_dev_info_get,
840 	.dev_supported_ptypes_get = hn_vf_supported_ptypes,
841 	.promiscuous_enable     = hn_dev_promiscuous_enable,
842 	.promiscuous_disable    = hn_dev_promiscuous_disable,
843 	.allmulticast_enable    = hn_dev_allmulticast_enable,
844 	.allmulticast_disable   = hn_dev_allmulticast_disable,
845 	.set_mc_addr_list	= hn_dev_mc_addr_list,
846 	.reta_update		= hn_rss_reta_update,
847 	.reta_query             = hn_rss_reta_query,
848 	.rss_hash_update	= hn_rss_hash_update,
849 	.rss_hash_conf_get      = hn_rss_hash_conf_get,
850 	.tx_queue_setup		= hn_dev_tx_queue_setup,
851 	.tx_queue_release	= hn_dev_tx_queue_release,
852 	.tx_done_cleanup        = hn_dev_tx_done_cleanup,
853 	.rx_queue_setup		= hn_dev_rx_queue_setup,
854 	.rx_queue_release	= hn_dev_rx_queue_release,
855 	.link_update		= hn_dev_link_update,
856 	.stats_get		= hn_dev_stats_get,
857 	.stats_reset            = hn_dev_stats_reset,
858 	.xstats_get		= hn_dev_xstats_get,
859 	.xstats_get_names	= hn_dev_xstats_get_names,
860 	.xstats_reset		= hn_dev_xstats_reset,
861 };
862 
863 /*
864  * Setup connection between PMD and kernel.
865  */
866 static int
867 hn_attach(struct hn_data *hv, unsigned int mtu)
868 {
869 	int error;
870 
871 	/* Attach NVS */
872 	error = hn_nvs_attach(hv, mtu);
873 	if (error)
874 		goto failed_nvs;
875 
876 	/* Attach RNDIS */
877 	error = hn_rndis_attach(hv);
878 	if (error)
879 		goto failed_rndis;
880 
881 	/*
882 	 * NOTE:
883 	 * Under certain conditions on certain versions of Hyper-V,
884 	 * the RNDIS rxfilter is _not_ zero on the hypervisor side
885 	 * after the successful RNDIS initialization.
886 	 */
887 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_NONE);
888 	return 0;
889 failed_rndis:
890 	hn_nvs_detach(hv);
891 failed_nvs:
892 	return error;
893 }
894 
895 static void
896 hn_detach(struct hn_data *hv)
897 {
898 	hn_nvs_detach(hv);
899 	hn_rndis_detach(hv);
900 }
901 
902 static int
903 eth_hn_dev_init(struct rte_eth_dev *eth_dev)
904 {
905 	struct hn_data *hv = eth_dev->data->dev_private;
906 	struct rte_device *device = eth_dev->device;
907 	struct rte_vmbus_device *vmbus;
908 	unsigned int rxr_cnt;
909 	int err, max_chan;
910 
911 	PMD_INIT_FUNC_TRACE();
912 
913 	vmbus = container_of(device, struct rte_vmbus_device, device);
914 	eth_dev->dev_ops = &hn_eth_dev_ops;
915 	eth_dev->tx_pkt_burst = &hn_xmit_pkts;
916 	eth_dev->rx_pkt_burst = &hn_recv_pkts;
917 
918 	/*
919 	 * for secondary processes, we don't initialize any further as primary
920 	 * has already done this work.
921 	 */
922 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
923 		return 0;
924 
925 	/* Since Hyper-V only supports one MAC address, just use local data */
926 	eth_dev->data->mac_addrs = &hv->mac_addr;
927 
928 	hv->vmbus = vmbus;
929 	hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP];
930 	hv->chim_res  = &vmbus->resource[HV_SEND_BUF_MAP];
931 	hv->port_id = eth_dev->data->port_id;
932 	hv->latency = HN_CHAN_LATENCY_NS;
933 	hv->max_queues = 1;
934 	rte_spinlock_init(&hv->vf_lock);
935 	hv->vf_port = HN_INVALID_PORT;
936 
937 	err = hn_parse_args(eth_dev);
938 	if (err)
939 		return err;
940 
941 	strlcpy(hv->owner.name, eth_dev->device->name,
942 		RTE_ETH_MAX_OWNER_NAME_LEN);
943 	err = rte_eth_dev_owner_new(&hv->owner.id);
944 	if (err) {
945 		PMD_INIT_LOG(ERR, "Can not get owner id");
946 		return err;
947 	}
948 
949 	/* Initialize primary channel input for control operations */
950 	err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
951 	if (err)
952 		return err;
953 
954 	rte_vmbus_set_latency(hv->vmbus, hv->channels[0], hv->latency);
955 
956 	hv->primary = hn_rx_queue_alloc(hv, 0,
957 					eth_dev->device->numa_node);
958 
959 	if (!hv->primary)
960 		return -ENOMEM;
961 
962 	err = hn_attach(hv, RTE_ETHER_MTU);
963 	if  (err)
964 		goto failed;
965 
966 	err = hn_tx_pool_init(eth_dev);
967 	if (err)
968 		goto failed;
969 
970 	err = hn_rndis_get_eaddr(hv, hv->mac_addr.addr_bytes);
971 	if (err)
972 		goto failed;
973 
974 	/* Multi queue requires later versions of windows server */
975 	if (hv->nvs_ver < NVS_VERSION_5)
976 		return 0;
977 
978 	max_chan = rte_vmbus_max_channels(vmbus);
979 	PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
980 	if (max_chan <= 0)
981 		goto failed;
982 
983 	if (hn_rndis_query_rsscaps(hv, &rxr_cnt) != 0)
984 		rxr_cnt = 1;
985 
986 	hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
987 
988 	/* If VF was reported but not added, do it now */
989 	if (hv->vf_present && !hn_vf_attached(hv)) {
990 		PMD_INIT_LOG(DEBUG, "Adding VF device");
991 
992 		err = hn_vf_add(eth_dev, hv);
993 		if (err)
994 			hv->vf_present = 0;
995 	}
996 
997 	return 0;
998 
999 failed:
1000 	PMD_INIT_LOG(NOTICE, "device init failed");
1001 
1002 	hn_tx_pool_uninit(eth_dev);
1003 	hn_detach(hv);
1004 	return err;
1005 }
1006 
1007 static int
1008 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
1009 {
1010 	struct hn_data *hv = eth_dev->data->dev_private;
1011 	int ret;
1012 
1013 	PMD_INIT_FUNC_TRACE();
1014 
1015 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1016 		return 0;
1017 
1018 	hn_dev_stop(eth_dev);
1019 	hn_dev_close(eth_dev);
1020 
1021 	eth_dev->dev_ops = NULL;
1022 	eth_dev->tx_pkt_burst = NULL;
1023 	eth_dev->rx_pkt_burst = NULL;
1024 
1025 	hn_detach(hv);
1026 	hn_tx_pool_uninit(eth_dev);
1027 	rte_vmbus_chan_close(hv->primary->chan);
1028 	rte_free(hv->primary);
1029 	ret = rte_eth_dev_owner_delete(hv->owner.id);
1030 	if (ret != 0)
1031 		return ret;
1032 
1033 	return 0;
1034 }
1035 
1036 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
1037 			struct rte_vmbus_device *dev)
1038 {
1039 	struct rte_eth_dev *eth_dev;
1040 	int ret;
1041 
1042 	PMD_INIT_FUNC_TRACE();
1043 
1044 	eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
1045 	if (!eth_dev)
1046 		return -ENOMEM;
1047 
1048 	ret = eth_hn_dev_init(eth_dev);
1049 	if (ret)
1050 		eth_dev_vmbus_release(eth_dev);
1051 	else
1052 		rte_eth_dev_probing_finish(eth_dev);
1053 
1054 	return ret;
1055 }
1056 
1057 static int eth_hn_remove(struct rte_vmbus_device *dev)
1058 {
1059 	struct rte_eth_dev *eth_dev;
1060 	int ret;
1061 
1062 	PMD_INIT_FUNC_TRACE();
1063 
1064 	eth_dev = rte_eth_dev_allocated(dev->device.name);
1065 	if (!eth_dev)
1066 		return -ENODEV;
1067 
1068 	ret = eth_hn_dev_uninit(eth_dev);
1069 	if (ret)
1070 		return ret;
1071 
1072 	eth_dev_vmbus_release(eth_dev);
1073 	return 0;
1074 }
1075 
1076 /* Network device GUID */
1077 static const rte_uuid_t hn_net_ids[] = {
1078 	/*  f8615163-df3e-46c5-913f-f2d2f965ed0e */
1079 	RTE_UUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x913f, 0xf2d2f965ed0eULL),
1080 	{ 0 }
1081 };
1082 
1083 static struct rte_vmbus_driver rte_netvsc_pmd = {
1084 	.id_table = hn_net_ids,
1085 	.probe = eth_hn_probe,
1086 	.remove = eth_hn_remove,
1087 };
1088 
1089 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
1090 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
1091 
1092 RTE_INIT(hn_init_log)
1093 {
1094 	hn_logtype_init = rte_log_register("pmd.net.netvsc.init");
1095 	if (hn_logtype_init >= 0)
1096 		rte_log_set_level(hn_logtype_init, RTE_LOG_NOTICE);
1097 	hn_logtype_driver = rte_log_register("pmd.net.netvsc.driver");
1098 	if (hn_logtype_driver >= 0)
1099 		rte_log_set_level(hn_logtype_driver, RTE_LOG_NOTICE);
1100 }
1101