xref: /dpdk/drivers/net/netvsc/hn_ethdev.c (revision 0ecc27f28d202a3356a8601e6762b601ea822c4c)
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 
47 int hn_logtype_init;
48 int hn_logtype_driver;
49 
50 struct hn_xstats_name_off {
51 	char name[RTE_ETH_XSTATS_NAME_SIZE];
52 	unsigned int offset;
53 };
54 
55 static const struct hn_xstats_name_off hn_stat_strings[] = {
56 	{ "good_packets",           offsetof(struct hn_stats, packets) },
57 	{ "good_bytes",             offsetof(struct hn_stats, bytes) },
58 	{ "errors",                 offsetof(struct hn_stats, errors) },
59 	{ "ring full",              offsetof(struct hn_stats, ring_full) },
60 	{ "multicast_packets",      offsetof(struct hn_stats, multicast) },
61 	{ "broadcast_packets",      offsetof(struct hn_stats, broadcast) },
62 	{ "undersize_packets",      offsetof(struct hn_stats, size_bins[0]) },
63 	{ "size_64_packets",        offsetof(struct hn_stats, size_bins[1]) },
64 	{ "size_65_127_packets",    offsetof(struct hn_stats, size_bins[2]) },
65 	{ "size_128_255_packets",   offsetof(struct hn_stats, size_bins[3]) },
66 	{ "size_256_511_packets",   offsetof(struct hn_stats, size_bins[4]) },
67 	{ "size_512_1023_packets",  offsetof(struct hn_stats, size_bins[5]) },
68 	{ "size_1024_1518_packets", offsetof(struct hn_stats, size_bins[6]) },
69 	{ "size_1519_max_packets",  offsetof(struct hn_stats, size_bins[7]) },
70 };
71 
72 /* The default RSS key.
73  * This value is the same as MLX5 so that flows will be
74  * received on same path for both VF ans synthetic NIC.
75  */
76 static const uint8_t rss_default_key[NDIS_HASH_KEYSIZE_TOEPLITZ] = {
77 	0x2c, 0xc6, 0x81, 0xd1,	0x5b, 0xdb, 0xf4, 0xf7,
78 	0xfc, 0xa2, 0x83, 0x19,	0xdb, 0x1a, 0x3e, 0x94,
79 	0x6b, 0x9e, 0x38, 0xd9,	0x2c, 0x9c, 0x03, 0xd1,
80 	0xad, 0x99, 0x44, 0xa7,	0xd9, 0x56, 0x3d, 0x59,
81 	0x06, 0x3c, 0x25, 0xf3,	0xfc, 0x1f, 0xdc, 0x2a,
82 };
83 
84 static struct rte_eth_dev *
85 eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
86 {
87 	struct rte_eth_dev *eth_dev;
88 	const char *name;
89 
90 	if (!dev)
91 		return NULL;
92 
93 	name = dev->device.name;
94 
95 	if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
96 		eth_dev = rte_eth_dev_allocate(name);
97 		if (!eth_dev) {
98 			PMD_DRV_LOG(NOTICE, "can not allocate rte ethdev");
99 			return NULL;
100 		}
101 
102 		if (private_data_size) {
103 			eth_dev->data->dev_private =
104 				rte_zmalloc_socket(name, private_data_size,
105 						     RTE_CACHE_LINE_SIZE, dev->device.numa_node);
106 			if (!eth_dev->data->dev_private) {
107 				PMD_DRV_LOG(NOTICE, "can not allocate driver data");
108 				rte_eth_dev_release_port(eth_dev);
109 				return NULL;
110 			}
111 		}
112 	} else {
113 		eth_dev = rte_eth_dev_attach_secondary(name);
114 		if (!eth_dev) {
115 			PMD_DRV_LOG(NOTICE, "can not attach secondary");
116 			return NULL;
117 		}
118 	}
119 
120 	eth_dev->device = &dev->device;
121 
122 	/* interrupt is simulated */
123 	dev->intr_handle.type = RTE_INTR_HANDLE_EXT;
124 	eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
125 	eth_dev->intr_handle = &dev->intr_handle;
126 
127 	/* allow ethdev to remove on close */
128 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
129 
130 	return eth_dev;
131 }
132 
133 static void
134 eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
135 {
136 	/* mac_addrs must not be freed alone because part of dev_private */
137 	eth_dev->data->mac_addrs = NULL;
138 	/* free ether device */
139 	rte_eth_dev_release_port(eth_dev);
140 
141 	eth_dev->device = NULL;
142 	eth_dev->intr_handle = NULL;
143 }
144 
145 /* handle "latency=X" from devargs */
146 static int hn_set_latency(const char *key, const char *value, void *opaque)
147 {
148 	struct hn_data *hv = opaque;
149 	char *endp = NULL;
150 	unsigned long lat;
151 
152 	errno = 0;
153 	lat = strtoul(value, &endp, 0);
154 
155 	if (*value == '\0' || *endp != '\0') {
156 		PMD_DRV_LOG(ERR, "invalid parameter %s=%s", key, value);
157 		return -EINVAL;
158 	}
159 
160 	PMD_DRV_LOG(DEBUG, "set latency %lu usec", lat);
161 
162 	hv->latency = lat * 1000;	/* usec to nsec */
163 	return 0;
164 }
165 
166 /* Parse device arguments */
167 static int hn_parse_args(const struct rte_eth_dev *dev)
168 {
169 	struct hn_data *hv = dev->data->dev_private;
170 	struct rte_devargs *devargs = dev->device->devargs;
171 	static const char * const valid_keys[] = {
172 		"latency",
173 		NULL
174 	};
175 	struct rte_kvargs *kvlist;
176 	int ret;
177 
178 	if (!devargs)
179 		return 0;
180 
181 	PMD_INIT_LOG(DEBUG, "device args %s %s",
182 		     devargs->name, devargs->args);
183 
184 	kvlist = rte_kvargs_parse(devargs->args, valid_keys);
185 	if (!kvlist) {
186 		PMD_DRV_LOG(NOTICE, "invalid parameters");
187 		return -EINVAL;
188 	}
189 
190 	ret = rte_kvargs_process(kvlist, "latency", hn_set_latency, hv);
191 	if (ret)
192 		PMD_DRV_LOG(ERR, "Unable to process latency arg\n");
193 
194 	rte_kvargs_free(kvlist);
195 	return ret;
196 }
197 
198 /* Update link status.
199  * Note: the DPDK definition of "wait_to_complete"
200  *   means block this call until link is up.
201  *   which is not worth supporting.
202  */
203 int
204 hn_dev_link_update(struct rte_eth_dev *dev,
205 		   int wait_to_complete)
206 {
207 	struct hn_data *hv = dev->data->dev_private;
208 	struct rte_eth_link link, old;
209 	int error;
210 
211 	old = dev->data->dev_link;
212 
213 	error = hn_rndis_get_linkstatus(hv);
214 	if (error)
215 		return error;
216 
217 	hn_rndis_get_linkspeed(hv);
218 
219 	hn_vf_link_update(dev, wait_to_complete);
220 
221 	link = (struct rte_eth_link) {
222 		.link_duplex = ETH_LINK_FULL_DUPLEX,
223 		.link_autoneg = ETH_LINK_SPEED_FIXED,
224 		.link_speed = hv->link_speed / 10000,
225 	};
226 
227 	if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
228 		link.link_status = ETH_LINK_UP;
229 	else
230 		link.link_status = ETH_LINK_DOWN;
231 
232 	if (old.link_status == link.link_status)
233 		return 0;
234 
235 	PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
236 		     (link.link_status == ETH_LINK_UP) ? "up" : "down");
237 
238 	return rte_eth_linkstatus_set(dev, &link);
239 }
240 
241 static int hn_dev_info_get(struct rte_eth_dev *dev,
242 			   struct rte_eth_dev_info *dev_info)
243 {
244 	struct hn_data *hv = dev->data->dev_private;
245 	int rc;
246 
247 	dev_info->speed_capa = ETH_LINK_SPEED_10G;
248 	dev_info->min_rx_bufsize = HN_MIN_RX_BUF_SIZE;
249 	dev_info->max_rx_pktlen  = HN_MAX_XFER_LEN;
250 	dev_info->max_mac_addrs  = 1;
251 
252 	dev_info->hash_key_size = NDIS_HASH_KEYSIZE_TOEPLITZ;
253 	dev_info->flow_type_rss_offloads = hv->rss_offloads;
254 	dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
255 
256 	dev_info->max_rx_queues = hv->max_queues;
257 	dev_info->max_tx_queues = hv->max_queues;
258 
259 	rc = hn_rndis_get_offload(hv, dev_info);
260 	if (rc != 0)
261 		return rc;
262 
263 	rc = hn_vf_info_get(hv, dev_info);
264 	if (rc != 0)
265 		return rc;
266 
267 	return 0;
268 }
269 
270 static int hn_rss_reta_update(struct rte_eth_dev *dev,
271 			      struct rte_eth_rss_reta_entry64 *reta_conf,
272 			      uint16_t reta_size)
273 {
274 	struct hn_data *hv = dev->data->dev_private;
275 	unsigned int i;
276 	int err;
277 
278 	PMD_INIT_FUNC_TRACE();
279 
280 	if (reta_size != NDIS_HASH_INDCNT) {
281 		PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
282 		return -EINVAL;
283 	}
284 
285 	for (i = 0; i < NDIS_HASH_INDCNT; i++) {
286 		uint16_t idx = i / RTE_RETA_GROUP_SIZE;
287 		uint16_t shift = i % RTE_RETA_GROUP_SIZE;
288 		uint64_t mask = (uint64_t)1 << shift;
289 
290 		if (reta_conf[idx].mask & mask)
291 			hv->rss_ind[i] = reta_conf[idx].reta[shift];
292 	}
293 
294 	err = hn_rndis_conf_rss(hv, 0);
295 	if (err) {
296 		PMD_DRV_LOG(NOTICE,
297 			    "reta reconfig failed");
298 		return err;
299 	}
300 
301 	return hn_vf_reta_hash_update(dev, reta_conf, reta_size);
302 }
303 
304 static int hn_rss_reta_query(struct rte_eth_dev *dev,
305 			     struct rte_eth_rss_reta_entry64 *reta_conf,
306 			     uint16_t reta_size)
307 {
308 	struct hn_data *hv = dev->data->dev_private;
309 	unsigned int i;
310 
311 	PMD_INIT_FUNC_TRACE();
312 
313 	if (reta_size != NDIS_HASH_INDCNT) {
314 		PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
315 		return -EINVAL;
316 	}
317 
318 	for (i = 0; i < NDIS_HASH_INDCNT; i++) {
319 		uint16_t idx = i / RTE_RETA_GROUP_SIZE;
320 		uint16_t shift = i % RTE_RETA_GROUP_SIZE;
321 		uint64_t mask = (uint64_t)1 << shift;
322 
323 		if (reta_conf[idx].mask & mask)
324 			reta_conf[idx].reta[shift] = hv->rss_ind[i];
325 	}
326 	return 0;
327 }
328 
329 static void hn_rss_hash_init(struct hn_data *hv,
330 			     const struct rte_eth_rss_conf *rss_conf)
331 {
332 	/* Convert from DPDK RSS hash flags to NDIS hash flags */
333 	hv->rss_hash = NDIS_HASH_FUNCTION_TOEPLITZ;
334 
335 	if (rss_conf->rss_hf & ETH_RSS_IPV4)
336 		hv->rss_hash |= NDIS_HASH_IPV4;
337 	if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
338 		hv->rss_hash |= NDIS_HASH_TCP_IPV4;
339 	if (rss_conf->rss_hf & ETH_RSS_IPV6)
340 		hv->rss_hash |=  NDIS_HASH_IPV6;
341 	if (rss_conf->rss_hf & ETH_RSS_IPV6_EX)
342 		hv->rss_hash |=  NDIS_HASH_IPV6_EX;
343 	if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
344 		hv->rss_hash |= NDIS_HASH_TCP_IPV6;
345 	if (rss_conf->rss_hf & ETH_RSS_IPV6_TCP_EX)
346 		hv->rss_hash |= NDIS_HASH_TCP_IPV6_EX;
347 
348 	memcpy(hv->rss_key, rss_conf->rss_key ? : rss_default_key,
349 	       NDIS_HASH_KEYSIZE_TOEPLITZ);
350 }
351 
352 static int hn_rss_hash_update(struct rte_eth_dev *dev,
353 			      struct rte_eth_rss_conf *rss_conf)
354 {
355 	struct hn_data *hv = dev->data->dev_private;
356 	int err;
357 
358 	PMD_INIT_FUNC_TRACE();
359 
360 	err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
361 	if (err) {
362 		PMD_DRV_LOG(NOTICE,
363 			    "rss disable failed");
364 		return err;
365 	}
366 
367 	hn_rss_hash_init(hv, rss_conf);
368 
369 	err = hn_rndis_conf_rss(hv, 0);
370 	if (err) {
371 		PMD_DRV_LOG(NOTICE,
372 			    "rss reconfig failed (RSS disabled)");
373 		return err;
374 	}
375 
376 
377 	return hn_vf_rss_hash_update(dev, rss_conf);
378 }
379 
380 static int hn_rss_hash_conf_get(struct rte_eth_dev *dev,
381 				struct rte_eth_rss_conf *rss_conf)
382 {
383 	struct hn_data *hv = dev->data->dev_private;
384 
385 	PMD_INIT_FUNC_TRACE();
386 
387 	if (hv->ndis_ver < NDIS_VERSION_6_20) {
388 		PMD_DRV_LOG(DEBUG, "RSS not supported on this host");
389 		return -EOPNOTSUPP;
390 	}
391 
392 	rss_conf->rss_key_len = NDIS_HASH_KEYSIZE_TOEPLITZ;
393 	if (rss_conf->rss_key)
394 		memcpy(rss_conf->rss_key, hv->rss_key,
395 		       NDIS_HASH_KEYSIZE_TOEPLITZ);
396 
397 	rss_conf->rss_hf = 0;
398 	if (hv->rss_hash & NDIS_HASH_IPV4)
399 		rss_conf->rss_hf |= ETH_RSS_IPV4;
400 
401 	if (hv->rss_hash & NDIS_HASH_TCP_IPV4)
402 		rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
403 
404 	if (hv->rss_hash & NDIS_HASH_IPV6)
405 		rss_conf->rss_hf |= ETH_RSS_IPV6;
406 
407 	if (hv->rss_hash & NDIS_HASH_IPV6_EX)
408 		rss_conf->rss_hf |= ETH_RSS_IPV6_EX;
409 
410 	if (hv->rss_hash & NDIS_HASH_TCP_IPV6)
411 		rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
412 
413 	if (hv->rss_hash & NDIS_HASH_TCP_IPV6_EX)
414 		rss_conf->rss_hf |= ETH_RSS_IPV6_TCP_EX;
415 
416 	return 0;
417 }
418 
419 static int
420 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
421 {
422 	struct hn_data *hv = dev->data->dev_private;
423 
424 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
425 	return hn_vf_promiscuous_enable(dev);
426 }
427 
428 static int
429 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
430 {
431 	struct hn_data *hv = dev->data->dev_private;
432 	uint32_t filter;
433 
434 	filter = NDIS_PACKET_TYPE_DIRECTED | NDIS_PACKET_TYPE_BROADCAST;
435 	if (dev->data->all_multicast)
436 		filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
437 	hn_rndis_set_rxfilter(hv, filter);
438 	return hn_vf_promiscuous_disable(dev);
439 }
440 
441 static int
442 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
443 {
444 	struct hn_data *hv = dev->data->dev_private;
445 
446 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
447 			      NDIS_PACKET_TYPE_ALL_MULTICAST |
448 			NDIS_PACKET_TYPE_BROADCAST);
449 	return hn_vf_allmulticast_enable(dev);
450 }
451 
452 static int
453 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
454 {
455 	struct hn_data *hv = dev->data->dev_private;
456 
457 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
458 			     NDIS_PACKET_TYPE_BROADCAST);
459 	return hn_vf_allmulticast_disable(dev);
460 }
461 
462 static int
463 hn_dev_mc_addr_list(struct rte_eth_dev *dev,
464 		     struct rte_ether_addr *mc_addr_set,
465 		     uint32_t nb_mc_addr)
466 {
467 	/* No filtering on the synthetic path, but can do it on VF */
468 	return hn_vf_mc_addr_list(dev, mc_addr_set, nb_mc_addr);
469 }
470 
471 /* Setup shared rx/tx queue data */
472 static int hn_subchan_configure(struct hn_data *hv,
473 				uint32_t subchan)
474 {
475 	struct vmbus_channel *primary = hn_primary_chan(hv);
476 	int err;
477 	unsigned int retry = 0;
478 
479 	PMD_DRV_LOG(DEBUG,
480 		    "open %u subchannels", subchan);
481 
482 	/* Send create sub channels command */
483 	err = hn_nvs_alloc_subchans(hv, &subchan);
484 	if (err)
485 		return  err;
486 
487 	while (subchan > 0) {
488 		struct vmbus_channel *new_sc;
489 		uint16_t chn_index;
490 
491 		err = rte_vmbus_subchan_open(primary, &new_sc);
492 		if (err == -ENOENT && ++retry < 1000) {
493 			/* This can happen if not ready yet */
494 			rte_delay_ms(10);
495 			continue;
496 		}
497 
498 		if (err) {
499 			PMD_DRV_LOG(ERR,
500 				    "open subchannel failed: %d", err);
501 			return err;
502 		}
503 
504 		rte_vmbus_set_latency(hv->vmbus, new_sc, hv->latency);
505 
506 		retry = 0;
507 		chn_index = rte_vmbus_sub_channel_index(new_sc);
508 		if (chn_index == 0 || chn_index > hv->max_queues) {
509 			PMD_DRV_LOG(ERR,
510 				    "Invalid subchannel offermsg channel %u",
511 				    chn_index);
512 			return -EIO;
513 		}
514 
515 		PMD_DRV_LOG(DEBUG, "new sub channel %u", chn_index);
516 		hv->channels[chn_index] = new_sc;
517 		--subchan;
518 	}
519 
520 	return err;
521 }
522 
523 static int hn_dev_configure(struct rte_eth_dev *dev)
524 {
525 	struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
526 	struct rte_eth_rss_conf *rss_conf = &dev_conf->rx_adv_conf.rss_conf;
527 	const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
528 	const struct rte_eth_txmode *txmode = &dev_conf->txmode;
529 	struct hn_data *hv = dev->data->dev_private;
530 	uint64_t unsupported;
531 	int i, err, subchan;
532 
533 	PMD_INIT_FUNC_TRACE();
534 
535 	unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
536 	if (unsupported) {
537 		PMD_DRV_LOG(NOTICE,
538 			    "unsupported TX offload: %#" PRIx64,
539 			    unsupported);
540 		return -EINVAL;
541 	}
542 
543 	unsupported = rxmode->offloads & ~HN_RX_OFFLOAD_CAPS;
544 	if (unsupported) {
545 		PMD_DRV_LOG(NOTICE,
546 			    "unsupported RX offload: %#" PRIx64,
547 			    rxmode->offloads);
548 		return -EINVAL;
549 	}
550 
551 	hv->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
552 
553 	err = hn_rndis_conf_offload(hv, txmode->offloads,
554 				    rxmode->offloads);
555 	if (err) {
556 		PMD_DRV_LOG(NOTICE,
557 			    "offload configure failed");
558 		return err;
559 	}
560 
561 	hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
562 				 dev->data->nb_tx_queues);
563 
564 	for (i = 0; i < NDIS_HASH_INDCNT; i++)
565 		hv->rss_ind[i] = i % hv->num_queues;
566 
567 	hn_rss_hash_init(hv, rss_conf);
568 
569 	subchan = hv->num_queues - 1;
570 	if (subchan > 0) {
571 		err = hn_subchan_configure(hv, subchan);
572 		if (err) {
573 			PMD_DRV_LOG(NOTICE,
574 				    "subchannel configuration failed");
575 			return err;
576 		}
577 
578 		err = hn_rndis_conf_rss(hv, 0);
579 		if (err) {
580 			PMD_DRV_LOG(NOTICE,
581 				    "initial RSS config failed");
582 			return err;
583 		}
584 	}
585 
586 	return hn_vf_configure(dev, dev_conf);
587 }
588 
589 static int hn_dev_stats_get(struct rte_eth_dev *dev,
590 			    struct rte_eth_stats *stats)
591 {
592 	unsigned int i;
593 
594 	hn_vf_stats_get(dev, stats);
595 
596 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
597 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
598 
599 		if (!txq)
600 			continue;
601 
602 		stats->opackets += txq->stats.packets;
603 		stats->obytes += txq->stats.bytes;
604 		stats->oerrors += txq->stats.errors;
605 
606 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
607 			stats->q_opackets[i] = txq->stats.packets;
608 			stats->q_obytes[i] = txq->stats.bytes;
609 		}
610 	}
611 
612 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
613 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
614 
615 		if (!rxq)
616 			continue;
617 
618 		stats->ipackets += rxq->stats.packets;
619 		stats->ibytes += rxq->stats.bytes;
620 		stats->ierrors += rxq->stats.errors;
621 		stats->imissed += rxq->stats.ring_full;
622 
623 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
624 			stats->q_ipackets[i] = rxq->stats.packets;
625 			stats->q_ibytes[i] = rxq->stats.bytes;
626 		}
627 	}
628 
629 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
630 	return 0;
631 }
632 
633 static int
634 hn_dev_stats_reset(struct rte_eth_dev *dev)
635 {
636 	unsigned int i;
637 
638 	PMD_INIT_FUNC_TRACE();
639 
640 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
641 		struct hn_tx_queue *txq = dev->data->tx_queues[i];
642 
643 		if (!txq)
644 			continue;
645 		memset(&txq->stats, 0, sizeof(struct hn_stats));
646 	}
647 
648 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
649 		struct hn_rx_queue *rxq = dev->data->rx_queues[i];
650 
651 		if (!rxq)
652 			continue;
653 
654 		memset(&rxq->stats, 0, sizeof(struct hn_stats));
655 	}
656 
657 	return 0;
658 }
659 
660 static int
661 hn_dev_xstats_reset(struct rte_eth_dev *dev)
662 {
663 	int ret;
664 
665 	ret = hn_dev_stats_reset(dev);
666 	if (ret != 0)
667 		return 0;
668 
669 	return hn_vf_xstats_reset(dev);
670 }
671 
672 static int
673 hn_dev_xstats_count(struct rte_eth_dev *dev)
674 {
675 	int ret, count;
676 
677 	count = dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings);
678 	count += dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
679 
680 	ret = hn_vf_xstats_get_names(dev, NULL, 0);
681 	if (ret < 0)
682 		return ret;
683 
684 	return count + ret;
685 }
686 
687 static int
688 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
689 			struct rte_eth_xstat_name *xstats_names,
690 			unsigned int limit)
691 {
692 	unsigned int i, t, count = 0;
693 	int ret;
694 
695 	if (!xstats_names)
696 		return hn_dev_xstats_count(dev);
697 
698 	/* Note: limit checked in rte_eth_xstats_names() */
699 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
700 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
701 
702 		if (!txq)
703 			continue;
704 
705 		if (count >= limit)
706 			break;
707 
708 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
709 			snprintf(xstats_names[count++].name,
710 				 RTE_ETH_XSTATS_NAME_SIZE,
711 				 "tx_q%u_%s", i, hn_stat_strings[t].name);
712 	}
713 
714 	for (i = 0; i < dev->data->nb_rx_queues; i++)  {
715 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
716 
717 		if (!rxq)
718 			continue;
719 
720 		if (count >= limit)
721 			break;
722 
723 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
724 			snprintf(xstats_names[count++].name,
725 				 RTE_ETH_XSTATS_NAME_SIZE,
726 				 "rx_q%u_%s", i,
727 				 hn_stat_strings[t].name);
728 	}
729 
730 	ret = hn_vf_xstats_get_names(dev, xstats_names + count,
731 				     limit - count);
732 	if (ret < 0)
733 		return ret;
734 
735 	return count + ret;
736 }
737 
738 static int
739 hn_dev_xstats_get(struct rte_eth_dev *dev,
740 		  struct rte_eth_xstat *xstats,
741 		  unsigned int n)
742 {
743 	unsigned int i, t, count = 0;
744 	const unsigned int nstats = hn_dev_xstats_count(dev);
745 	const char *stats;
746 	int ret;
747 
748 	PMD_INIT_FUNC_TRACE();
749 
750 	if (n < nstats)
751 		return nstats;
752 
753 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
754 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
755 
756 		if (!txq)
757 			continue;
758 
759 		stats = (const char *)&txq->stats;
760 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
761 			xstats[count].id = count;
762 			xstats[count].value = *(const uint64_t *)
763 				(stats + hn_stat_strings[t].offset);
764 		}
765 	}
766 
767 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
768 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
769 
770 		if (!rxq)
771 			continue;
772 
773 		stats = (const char *)&rxq->stats;
774 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
775 			xstats[count].id = count;
776 			xstats[count].value = *(const uint64_t *)
777 				(stats + hn_stat_strings[t].offset);
778 		}
779 	}
780 
781 	ret = hn_vf_xstats_get(dev, xstats, count, n);
782 	if (ret < 0)
783 		return ret;
784 
785 	return count + ret;
786 }
787 
788 static int
789 hn_dev_start(struct rte_eth_dev *dev)
790 {
791 	struct hn_data *hv = dev->data->dev_private;
792 	int error;
793 
794 	PMD_INIT_FUNC_TRACE();
795 
796 	error = hn_rndis_set_rxfilter(hv,
797 				      NDIS_PACKET_TYPE_BROADCAST |
798 				      NDIS_PACKET_TYPE_ALL_MULTICAST |
799 				      NDIS_PACKET_TYPE_DIRECTED);
800 	if (error)
801 		return error;
802 
803 	error = hn_vf_start(dev);
804 	if (error)
805 		hn_rndis_set_rxfilter(hv, 0);
806 
807 	return error;
808 }
809 
810 static void
811 hn_dev_stop(struct rte_eth_dev *dev)
812 {
813 	struct hn_data *hv = dev->data->dev_private;
814 
815 	PMD_INIT_FUNC_TRACE();
816 
817 	hn_rndis_set_rxfilter(hv, 0);
818 	hn_vf_stop(dev);
819 }
820 
821 static void
822 hn_dev_close(struct rte_eth_dev *dev)
823 {
824 	PMD_INIT_FUNC_TRACE();
825 
826 	hn_vf_close(dev);
827 	hn_dev_free_queues(dev);
828 }
829 
830 static const struct eth_dev_ops hn_eth_dev_ops = {
831 	.dev_configure		= hn_dev_configure,
832 	.dev_start		= hn_dev_start,
833 	.dev_stop		= hn_dev_stop,
834 	.dev_close		= hn_dev_close,
835 	.dev_infos_get		= hn_dev_info_get,
836 	.dev_supported_ptypes_get = hn_vf_supported_ptypes,
837 	.promiscuous_enable     = hn_dev_promiscuous_enable,
838 	.promiscuous_disable    = hn_dev_promiscuous_disable,
839 	.allmulticast_enable    = hn_dev_allmulticast_enable,
840 	.allmulticast_disable   = hn_dev_allmulticast_disable,
841 	.set_mc_addr_list	= hn_dev_mc_addr_list,
842 	.reta_update		= hn_rss_reta_update,
843 	.reta_query             = hn_rss_reta_query,
844 	.rss_hash_update	= hn_rss_hash_update,
845 	.rss_hash_conf_get      = hn_rss_hash_conf_get,
846 	.tx_queue_setup		= hn_dev_tx_queue_setup,
847 	.tx_queue_release	= hn_dev_tx_queue_release,
848 	.tx_done_cleanup        = hn_dev_tx_done_cleanup,
849 	.rx_queue_setup		= hn_dev_rx_queue_setup,
850 	.rx_queue_release	= hn_dev_rx_queue_release,
851 	.link_update		= hn_dev_link_update,
852 	.stats_get		= hn_dev_stats_get,
853 	.stats_reset            = hn_dev_stats_reset,
854 	.xstats_get		= hn_dev_xstats_get,
855 	.xstats_get_names	= hn_dev_xstats_get_names,
856 	.xstats_reset		= hn_dev_xstats_reset,
857 };
858 
859 /*
860  * Setup connection between PMD and kernel.
861  */
862 static int
863 hn_attach(struct hn_data *hv, unsigned int mtu)
864 {
865 	int error;
866 
867 	/* Attach NVS */
868 	error = hn_nvs_attach(hv, mtu);
869 	if (error)
870 		goto failed_nvs;
871 
872 	/* Attach RNDIS */
873 	error = hn_rndis_attach(hv);
874 	if (error)
875 		goto failed_rndis;
876 
877 	/*
878 	 * NOTE:
879 	 * Under certain conditions on certain versions of Hyper-V,
880 	 * the RNDIS rxfilter is _not_ zero on the hypervisor side
881 	 * after the successful RNDIS initialization.
882 	 */
883 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_NONE);
884 	return 0;
885 failed_rndis:
886 	hn_nvs_detach(hv);
887 failed_nvs:
888 	return error;
889 }
890 
891 static void
892 hn_detach(struct hn_data *hv)
893 {
894 	hn_nvs_detach(hv);
895 	hn_rndis_detach(hv);
896 }
897 
898 static int
899 eth_hn_dev_init(struct rte_eth_dev *eth_dev)
900 {
901 	struct hn_data *hv = eth_dev->data->dev_private;
902 	struct rte_device *device = eth_dev->device;
903 	struct rte_vmbus_device *vmbus;
904 	unsigned int rxr_cnt;
905 	int err, max_chan;
906 
907 	PMD_INIT_FUNC_TRACE();
908 
909 	vmbus = container_of(device, struct rte_vmbus_device, device);
910 	eth_dev->dev_ops = &hn_eth_dev_ops;
911 	eth_dev->tx_pkt_burst = &hn_xmit_pkts;
912 	eth_dev->rx_pkt_burst = &hn_recv_pkts;
913 
914 	/*
915 	 * for secondary processes, we don't initialize any further as primary
916 	 * has already done this work.
917 	 */
918 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
919 		return 0;
920 
921 	/* Since Hyper-V only supports one MAC address, just use local data */
922 	eth_dev->data->mac_addrs = &hv->mac_addr;
923 
924 	hv->vmbus = vmbus;
925 	hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP];
926 	hv->chim_res  = &vmbus->resource[HV_SEND_BUF_MAP];
927 	hv->port_id = eth_dev->data->port_id;
928 	hv->latency = HN_CHAN_LATENCY_NS;
929 	hv->max_queues = 1;
930 	rte_spinlock_init(&hv->vf_lock);
931 	hv->vf_port = HN_INVALID_PORT;
932 
933 	err = hn_parse_args(eth_dev);
934 	if (err)
935 		return err;
936 
937 	strlcpy(hv->owner.name, eth_dev->device->name,
938 		RTE_ETH_MAX_OWNER_NAME_LEN);
939 	err = rte_eth_dev_owner_new(&hv->owner.id);
940 	if (err) {
941 		PMD_INIT_LOG(ERR, "Can not get owner id");
942 		return err;
943 	}
944 
945 	/* Initialize primary channel input for control operations */
946 	err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
947 	if (err)
948 		return err;
949 
950 	rte_vmbus_set_latency(hv->vmbus, hv->channels[0], hv->latency);
951 
952 	hv->primary = hn_rx_queue_alloc(hv, 0,
953 					eth_dev->device->numa_node);
954 
955 	if (!hv->primary)
956 		return -ENOMEM;
957 
958 	err = hn_attach(hv, RTE_ETHER_MTU);
959 	if  (err)
960 		goto failed;
961 
962 	err = hn_tx_pool_init(eth_dev);
963 	if (err)
964 		goto failed;
965 
966 	err = hn_rndis_get_eaddr(hv, hv->mac_addr.addr_bytes);
967 	if (err)
968 		goto failed;
969 
970 	/* Multi queue requires later versions of windows server */
971 	if (hv->nvs_ver < NVS_VERSION_5)
972 		return 0;
973 
974 	max_chan = rte_vmbus_max_channels(vmbus);
975 	PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
976 	if (max_chan <= 0)
977 		goto failed;
978 
979 	if (hn_rndis_query_rsscaps(hv, &rxr_cnt) != 0)
980 		rxr_cnt = 1;
981 
982 	hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
983 
984 	/* If VF was reported but not added, do it now */
985 	if (hv->vf_present && !hn_vf_attached(hv)) {
986 		PMD_INIT_LOG(DEBUG, "Adding VF device");
987 
988 		err = hn_vf_add(eth_dev, hv);
989 		if (err)
990 			hv->vf_present = 0;
991 	}
992 
993 	return 0;
994 
995 failed:
996 	PMD_INIT_LOG(NOTICE, "device init failed");
997 
998 	hn_tx_pool_uninit(eth_dev);
999 	hn_detach(hv);
1000 	return err;
1001 }
1002 
1003 static int
1004 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
1005 {
1006 	struct hn_data *hv = eth_dev->data->dev_private;
1007 	int ret;
1008 
1009 	PMD_INIT_FUNC_TRACE();
1010 
1011 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1012 		return 0;
1013 
1014 	hn_dev_stop(eth_dev);
1015 	hn_dev_close(eth_dev);
1016 
1017 	eth_dev->dev_ops = NULL;
1018 	eth_dev->tx_pkt_burst = NULL;
1019 	eth_dev->rx_pkt_burst = NULL;
1020 
1021 	hn_detach(hv);
1022 	hn_tx_pool_uninit(eth_dev);
1023 	rte_vmbus_chan_close(hv->primary->chan);
1024 	rte_free(hv->primary);
1025 	ret = rte_eth_dev_owner_delete(hv->owner.id);
1026 	if (ret != 0)
1027 		return ret;
1028 
1029 	return 0;
1030 }
1031 
1032 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
1033 			struct rte_vmbus_device *dev)
1034 {
1035 	struct rte_eth_dev *eth_dev;
1036 	int ret;
1037 
1038 	PMD_INIT_FUNC_TRACE();
1039 
1040 	eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
1041 	if (!eth_dev)
1042 		return -ENOMEM;
1043 
1044 	ret = eth_hn_dev_init(eth_dev);
1045 	if (ret)
1046 		eth_dev_vmbus_release(eth_dev);
1047 	else
1048 		rte_eth_dev_probing_finish(eth_dev);
1049 
1050 	return ret;
1051 }
1052 
1053 static int eth_hn_remove(struct rte_vmbus_device *dev)
1054 {
1055 	struct rte_eth_dev *eth_dev;
1056 	int ret;
1057 
1058 	PMD_INIT_FUNC_TRACE();
1059 
1060 	eth_dev = rte_eth_dev_allocated(dev->device.name);
1061 	if (!eth_dev)
1062 		return -ENODEV;
1063 
1064 	ret = eth_hn_dev_uninit(eth_dev);
1065 	if (ret)
1066 		return ret;
1067 
1068 	eth_dev_vmbus_release(eth_dev);
1069 	return 0;
1070 }
1071 
1072 /* Network device GUID */
1073 static const rte_uuid_t hn_net_ids[] = {
1074 	/*  f8615163-df3e-46c5-913f-f2d2f965ed0e */
1075 	RTE_UUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x913f, 0xf2d2f965ed0eULL),
1076 	{ 0 }
1077 };
1078 
1079 static struct rte_vmbus_driver rte_netvsc_pmd = {
1080 	.id_table = hn_net_ids,
1081 	.probe = eth_hn_probe,
1082 	.remove = eth_hn_remove,
1083 };
1084 
1085 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
1086 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
1087 
1088 RTE_INIT(hn_init_log)
1089 {
1090 	hn_logtype_init = rte_log_register("pmd.net.netvsc.init");
1091 	if (hn_logtype_init >= 0)
1092 		rte_log_set_level(hn_logtype_init, RTE_LOG_NOTICE);
1093 	hn_logtype_driver = rte_log_register("pmd.net.netvsc.driver");
1094 	if (hn_logtype_driver >= 0)
1095 		rte_log_set_level(hn_logtype_driver, RTE_LOG_NOTICE);
1096 }
1097