xref: /dpdk/drivers/net/netvsc/hn_ethdev.c (revision e11bdd37745229bf26b557305c07d118c3dbaad7)
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 	/* free ether device */
138 	rte_eth_dev_release_port(eth_dev);
139 
140 	eth_dev->device = NULL;
141 	eth_dev->intr_handle = NULL;
142 }
143 
144 /* handle "latency=X" from devargs */
145 static int hn_set_latency(const char *key, const char *value, void *opaque)
146 {
147 	struct hn_data *hv = opaque;
148 	char *endp = NULL;
149 	unsigned long lat;
150 
151 	errno = 0;
152 	lat = strtoul(value, &endp, 0);
153 
154 	if (*value == '\0' || *endp != '\0') {
155 		PMD_DRV_LOG(ERR, "invalid parameter %s=%s", key, value);
156 		return -EINVAL;
157 	}
158 
159 	PMD_DRV_LOG(DEBUG, "set latency %lu usec", lat);
160 
161 	hv->latency = lat * 1000;	/* usec to nsec */
162 	return 0;
163 }
164 
165 /* Parse device arguments */
166 static int hn_parse_args(const struct rte_eth_dev *dev)
167 {
168 	struct hn_data *hv = dev->data->dev_private;
169 	struct rte_devargs *devargs = dev->device->devargs;
170 	static const char * const valid_keys[] = {
171 		"latency",
172 		NULL
173 	};
174 	struct rte_kvargs *kvlist;
175 	int ret;
176 
177 	if (!devargs)
178 		return 0;
179 
180 	PMD_INIT_LOG(DEBUG, "device args %s %s",
181 		     devargs->name, devargs->args);
182 
183 	kvlist = rte_kvargs_parse(devargs->args, valid_keys);
184 	if (!kvlist) {
185 		PMD_DRV_LOG(NOTICE, "invalid parameters");
186 		return -EINVAL;
187 	}
188 
189 	ret = rte_kvargs_process(kvlist, "latency", hn_set_latency, hv);
190 	if (ret)
191 		PMD_DRV_LOG(ERR, "Unable to process latency arg\n");
192 
193 	rte_kvargs_free(kvlist);
194 	return ret;
195 }
196 
197 /* Update link status.
198  * Note: the DPDK definition of "wait_to_complete"
199  *   means block this call until link is up.
200  *   which is not worth supporting.
201  */
202 int
203 hn_dev_link_update(struct rte_eth_dev *dev,
204 		   int wait_to_complete)
205 {
206 	struct hn_data *hv = dev->data->dev_private;
207 	struct rte_eth_link link, old;
208 	int error;
209 
210 	old = dev->data->dev_link;
211 
212 	error = hn_rndis_get_linkstatus(hv);
213 	if (error)
214 		return error;
215 
216 	hn_rndis_get_linkspeed(hv);
217 
218 	hn_vf_link_update(dev, wait_to_complete);
219 
220 	link = (struct rte_eth_link) {
221 		.link_duplex = ETH_LINK_FULL_DUPLEX,
222 		.link_autoneg = ETH_LINK_SPEED_FIXED,
223 		.link_speed = hv->link_speed / 10000,
224 	};
225 
226 	if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
227 		link.link_status = ETH_LINK_UP;
228 	else
229 		link.link_status = ETH_LINK_DOWN;
230 
231 	if (old.link_status == link.link_status)
232 		return 0;
233 
234 	PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
235 		     (link.link_status == ETH_LINK_UP) ? "up" : "down");
236 
237 	return rte_eth_linkstatus_set(dev, &link);
238 }
239 
240 static int hn_dev_info_get(struct rte_eth_dev *dev,
241 			   struct rte_eth_dev_info *dev_info)
242 {
243 	struct hn_data *hv = dev->data->dev_private;
244 	int rc;
245 
246 	dev_info->speed_capa = ETH_LINK_SPEED_10G;
247 	dev_info->min_rx_bufsize = HN_MIN_RX_BUF_SIZE;
248 	dev_info->max_rx_pktlen  = HN_MAX_XFER_LEN;
249 	dev_info->max_mac_addrs  = 1;
250 
251 	dev_info->hash_key_size = NDIS_HASH_KEYSIZE_TOEPLITZ;
252 	dev_info->flow_type_rss_offloads = hv->rss_offloads;
253 	dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
254 
255 	dev_info->max_rx_queues = hv->max_queues;
256 	dev_info->max_tx_queues = hv->max_queues;
257 
258 	dev_info->tx_desc_lim.nb_min = 1;
259 	dev_info->tx_desc_lim.nb_max = 4096;
260 
261 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
262 		return 0;
263 
264 	/* fills in rx and tx offload capability */
265 	rc = hn_rndis_get_offload(hv, dev_info);
266 	if (rc != 0)
267 		return rc;
268 
269 	/* merges the offload and queues of vf */
270 	return hn_vf_info_get(hv, dev_info);
271 }
272 
273 static int hn_rss_reta_update(struct rte_eth_dev *dev,
274 			      struct rte_eth_rss_reta_entry64 *reta_conf,
275 			      uint16_t reta_size)
276 {
277 	struct hn_data *hv = dev->data->dev_private;
278 	unsigned int i;
279 	int err;
280 
281 	PMD_INIT_FUNC_TRACE();
282 
283 	if (reta_size != NDIS_HASH_INDCNT) {
284 		PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
285 		return -EINVAL;
286 	}
287 
288 	for (i = 0; i < NDIS_HASH_INDCNT; i++) {
289 		uint16_t idx = i / RTE_RETA_GROUP_SIZE;
290 		uint16_t shift = i % RTE_RETA_GROUP_SIZE;
291 		uint64_t mask = (uint64_t)1 << shift;
292 
293 		if (reta_conf[idx].mask & mask)
294 			hv->rss_ind[i] = reta_conf[idx].reta[shift];
295 	}
296 
297 	err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
298 	if (err) {
299 		PMD_DRV_LOG(NOTICE,
300 			"rss disable failed");
301 		return err;
302 	}
303 
304 	err = hn_rndis_conf_rss(hv, 0);
305 	if (err) {
306 		PMD_DRV_LOG(NOTICE,
307 			    "reta reconfig failed");
308 		return err;
309 	}
310 
311 	return hn_vf_reta_hash_update(dev, reta_conf, reta_size);
312 }
313 
314 static int hn_rss_reta_query(struct rte_eth_dev *dev,
315 			     struct rte_eth_rss_reta_entry64 *reta_conf,
316 			     uint16_t reta_size)
317 {
318 	struct hn_data *hv = dev->data->dev_private;
319 	unsigned int i;
320 
321 	PMD_INIT_FUNC_TRACE();
322 
323 	if (reta_size != NDIS_HASH_INDCNT) {
324 		PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
325 		return -EINVAL;
326 	}
327 
328 	for (i = 0; i < NDIS_HASH_INDCNT; i++) {
329 		uint16_t idx = i / RTE_RETA_GROUP_SIZE;
330 		uint16_t shift = i % RTE_RETA_GROUP_SIZE;
331 		uint64_t mask = (uint64_t)1 << shift;
332 
333 		if (reta_conf[idx].mask & mask)
334 			reta_conf[idx].reta[shift] = hv->rss_ind[i];
335 	}
336 	return 0;
337 }
338 
339 static void hn_rss_hash_init(struct hn_data *hv,
340 			     const struct rte_eth_rss_conf *rss_conf)
341 {
342 	/* Convert from DPDK RSS hash flags to NDIS hash flags */
343 	hv->rss_hash = NDIS_HASH_FUNCTION_TOEPLITZ;
344 
345 	if (rss_conf->rss_hf & ETH_RSS_IPV4)
346 		hv->rss_hash |= NDIS_HASH_IPV4;
347 	if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
348 		hv->rss_hash |= NDIS_HASH_TCP_IPV4;
349 	if (rss_conf->rss_hf & ETH_RSS_IPV6)
350 		hv->rss_hash |=  NDIS_HASH_IPV6;
351 	if (rss_conf->rss_hf & ETH_RSS_IPV6_EX)
352 		hv->rss_hash |=  NDIS_HASH_IPV6_EX;
353 	if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
354 		hv->rss_hash |= NDIS_HASH_TCP_IPV6;
355 	if (rss_conf->rss_hf & ETH_RSS_IPV6_TCP_EX)
356 		hv->rss_hash |= NDIS_HASH_TCP_IPV6_EX;
357 
358 	memcpy(hv->rss_key, rss_conf->rss_key ? : rss_default_key,
359 	       NDIS_HASH_KEYSIZE_TOEPLITZ);
360 }
361 
362 static int hn_rss_hash_update(struct rte_eth_dev *dev,
363 			      struct rte_eth_rss_conf *rss_conf)
364 {
365 	struct hn_data *hv = dev->data->dev_private;
366 	int err;
367 
368 	PMD_INIT_FUNC_TRACE();
369 
370 	err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
371 	if (err) {
372 		PMD_DRV_LOG(NOTICE,
373 			    "rss disable failed");
374 		return err;
375 	}
376 
377 	hn_rss_hash_init(hv, rss_conf);
378 
379 	err = hn_rndis_conf_rss(hv, 0);
380 	if (err) {
381 		PMD_DRV_LOG(NOTICE,
382 			    "rss reconfig failed (RSS disabled)");
383 		return err;
384 	}
385 
386 
387 	return hn_vf_rss_hash_update(dev, rss_conf);
388 }
389 
390 static int hn_rss_hash_conf_get(struct rte_eth_dev *dev,
391 				struct rte_eth_rss_conf *rss_conf)
392 {
393 	struct hn_data *hv = dev->data->dev_private;
394 
395 	PMD_INIT_FUNC_TRACE();
396 
397 	if (hv->ndis_ver < NDIS_VERSION_6_20) {
398 		PMD_DRV_LOG(DEBUG, "RSS not supported on this host");
399 		return -EOPNOTSUPP;
400 	}
401 
402 	rss_conf->rss_key_len = NDIS_HASH_KEYSIZE_TOEPLITZ;
403 	if (rss_conf->rss_key)
404 		memcpy(rss_conf->rss_key, hv->rss_key,
405 		       NDIS_HASH_KEYSIZE_TOEPLITZ);
406 
407 	rss_conf->rss_hf = 0;
408 	if (hv->rss_hash & NDIS_HASH_IPV4)
409 		rss_conf->rss_hf |= ETH_RSS_IPV4;
410 
411 	if (hv->rss_hash & NDIS_HASH_TCP_IPV4)
412 		rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
413 
414 	if (hv->rss_hash & NDIS_HASH_IPV6)
415 		rss_conf->rss_hf |= ETH_RSS_IPV6;
416 
417 	if (hv->rss_hash & NDIS_HASH_IPV6_EX)
418 		rss_conf->rss_hf |= ETH_RSS_IPV6_EX;
419 
420 	if (hv->rss_hash & NDIS_HASH_TCP_IPV6)
421 		rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
422 
423 	if (hv->rss_hash & NDIS_HASH_TCP_IPV6_EX)
424 		rss_conf->rss_hf |= ETH_RSS_IPV6_TCP_EX;
425 
426 	return 0;
427 }
428 
429 static int
430 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
431 {
432 	struct hn_data *hv = dev->data->dev_private;
433 
434 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
435 	return hn_vf_promiscuous_enable(dev);
436 }
437 
438 static int
439 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
440 {
441 	struct hn_data *hv = dev->data->dev_private;
442 	uint32_t filter;
443 
444 	filter = NDIS_PACKET_TYPE_DIRECTED | NDIS_PACKET_TYPE_BROADCAST;
445 	if (dev->data->all_multicast)
446 		filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
447 	hn_rndis_set_rxfilter(hv, filter);
448 	return hn_vf_promiscuous_disable(dev);
449 }
450 
451 static int
452 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
453 {
454 	struct hn_data *hv = dev->data->dev_private;
455 
456 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
457 			      NDIS_PACKET_TYPE_ALL_MULTICAST |
458 			NDIS_PACKET_TYPE_BROADCAST);
459 	return hn_vf_allmulticast_enable(dev);
460 }
461 
462 static int
463 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
464 {
465 	struct hn_data *hv = dev->data->dev_private;
466 
467 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
468 			     NDIS_PACKET_TYPE_BROADCAST);
469 	return hn_vf_allmulticast_disable(dev);
470 }
471 
472 static int
473 hn_dev_mc_addr_list(struct rte_eth_dev *dev,
474 		     struct rte_ether_addr *mc_addr_set,
475 		     uint32_t nb_mc_addr)
476 {
477 	/* No filtering on the synthetic path, but can do it on VF */
478 	return hn_vf_mc_addr_list(dev, mc_addr_set, nb_mc_addr);
479 }
480 
481 /* Setup shared rx/tx queue data */
482 static int hn_subchan_configure(struct hn_data *hv,
483 				uint32_t subchan)
484 {
485 	struct vmbus_channel *primary = hn_primary_chan(hv);
486 	int err;
487 	unsigned int retry = 0;
488 
489 	PMD_DRV_LOG(DEBUG,
490 		    "open %u subchannels", subchan);
491 
492 	/* Send create sub channels command */
493 	err = hn_nvs_alloc_subchans(hv, &subchan);
494 	if (err)
495 		return  err;
496 
497 	while (subchan > 0) {
498 		struct vmbus_channel *new_sc;
499 		uint16_t chn_index;
500 
501 		err = rte_vmbus_subchan_open(primary, &new_sc);
502 		if (err == -ENOENT && ++retry < 1000) {
503 			/* This can happen if not ready yet */
504 			rte_delay_ms(10);
505 			continue;
506 		}
507 
508 		if (err) {
509 			PMD_DRV_LOG(ERR,
510 				    "open subchannel failed: %d", err);
511 			return err;
512 		}
513 
514 		rte_vmbus_set_latency(hv->vmbus, new_sc, hv->latency);
515 
516 		retry = 0;
517 		chn_index = rte_vmbus_sub_channel_index(new_sc);
518 		if (chn_index == 0 || chn_index > hv->max_queues) {
519 			PMD_DRV_LOG(ERR,
520 				    "Invalid subchannel offermsg channel %u",
521 				    chn_index);
522 			return -EIO;
523 		}
524 
525 		PMD_DRV_LOG(DEBUG, "new sub channel %u", chn_index);
526 		hv->channels[chn_index] = new_sc;
527 		--subchan;
528 	}
529 
530 	return err;
531 }
532 
533 static int hn_dev_configure(struct rte_eth_dev *dev)
534 {
535 	struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
536 	struct rte_eth_rss_conf *rss_conf = &dev_conf->rx_adv_conf.rss_conf;
537 	const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
538 	const struct rte_eth_txmode *txmode = &dev_conf->txmode;
539 	struct hn_data *hv = dev->data->dev_private;
540 	uint64_t unsupported;
541 	int i, err, subchan;
542 
543 	PMD_INIT_FUNC_TRACE();
544 
545 	if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
546 		dev_conf->rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
547 
548 	unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
549 	if (unsupported) {
550 		PMD_DRV_LOG(NOTICE,
551 			    "unsupported TX offload: %#" PRIx64,
552 			    unsupported);
553 		return -EINVAL;
554 	}
555 
556 	unsupported = rxmode->offloads & ~HN_RX_OFFLOAD_CAPS;
557 	if (unsupported) {
558 		PMD_DRV_LOG(NOTICE,
559 			    "unsupported RX offload: %#" PRIx64,
560 			    rxmode->offloads);
561 		return -EINVAL;
562 	}
563 
564 	hv->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
565 
566 	err = hn_rndis_conf_offload(hv, txmode->offloads,
567 				    rxmode->offloads);
568 	if (err) {
569 		PMD_DRV_LOG(NOTICE,
570 			    "offload configure failed");
571 		return err;
572 	}
573 
574 	hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
575 				 dev->data->nb_tx_queues);
576 
577 	for (i = 0; i < NDIS_HASH_INDCNT; i++)
578 		hv->rss_ind[i] = i % hv->num_queues;
579 
580 	hn_rss_hash_init(hv, rss_conf);
581 
582 	subchan = hv->num_queues - 1;
583 	if (subchan > 0) {
584 		err = hn_subchan_configure(hv, subchan);
585 		if (err) {
586 			PMD_DRV_LOG(NOTICE,
587 				    "subchannel configuration failed");
588 			return err;
589 		}
590 
591 		err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
592 		if (err) {
593 			PMD_DRV_LOG(NOTICE,
594 				"rss disable failed");
595 			return err;
596 		}
597 
598 		err = hn_rndis_conf_rss(hv, 0);
599 		if (err) {
600 			PMD_DRV_LOG(NOTICE,
601 				    "initial RSS config failed");
602 			return err;
603 		}
604 	}
605 
606 	return hn_vf_configure(dev, dev_conf);
607 }
608 
609 static int hn_dev_stats_get(struct rte_eth_dev *dev,
610 			    struct rte_eth_stats *stats)
611 {
612 	unsigned int i;
613 
614 	hn_vf_stats_get(dev, stats);
615 
616 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
617 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
618 
619 		if (!txq)
620 			continue;
621 
622 		stats->opackets += txq->stats.packets;
623 		stats->obytes += txq->stats.bytes;
624 		stats->oerrors += txq->stats.errors;
625 
626 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
627 			stats->q_opackets[i] = txq->stats.packets;
628 			stats->q_obytes[i] = txq->stats.bytes;
629 		}
630 	}
631 
632 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
633 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
634 
635 		if (!rxq)
636 			continue;
637 
638 		stats->ipackets += rxq->stats.packets;
639 		stats->ibytes += rxq->stats.bytes;
640 		stats->ierrors += rxq->stats.errors;
641 		stats->imissed += rxq->stats.ring_full;
642 
643 		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
644 			stats->q_ipackets[i] = rxq->stats.packets;
645 			stats->q_ibytes[i] = rxq->stats.bytes;
646 		}
647 	}
648 
649 	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
650 	return 0;
651 }
652 
653 static int
654 hn_dev_stats_reset(struct rte_eth_dev *dev)
655 {
656 	unsigned int i;
657 
658 	PMD_INIT_FUNC_TRACE();
659 
660 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
661 		struct hn_tx_queue *txq = dev->data->tx_queues[i];
662 
663 		if (!txq)
664 			continue;
665 		memset(&txq->stats, 0, sizeof(struct hn_stats));
666 	}
667 
668 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
669 		struct hn_rx_queue *rxq = dev->data->rx_queues[i];
670 
671 		if (!rxq)
672 			continue;
673 
674 		memset(&rxq->stats, 0, sizeof(struct hn_stats));
675 	}
676 
677 	return 0;
678 }
679 
680 static int
681 hn_dev_xstats_reset(struct rte_eth_dev *dev)
682 {
683 	int ret;
684 
685 	ret = hn_dev_stats_reset(dev);
686 	if (ret != 0)
687 		return 0;
688 
689 	return hn_vf_xstats_reset(dev);
690 }
691 
692 static int
693 hn_dev_xstats_count(struct rte_eth_dev *dev)
694 {
695 	int ret, count;
696 
697 	count = dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings);
698 	count += dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
699 
700 	ret = hn_vf_xstats_get_names(dev, NULL, 0);
701 	if (ret < 0)
702 		return ret;
703 
704 	return count + ret;
705 }
706 
707 static int
708 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
709 			struct rte_eth_xstat_name *xstats_names,
710 			unsigned int limit)
711 {
712 	unsigned int i, t, count = 0;
713 	int ret;
714 
715 	if (!xstats_names)
716 		return hn_dev_xstats_count(dev);
717 
718 	/* Note: limit checked in rte_eth_xstats_names() */
719 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
720 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
721 
722 		if (!txq)
723 			continue;
724 
725 		if (count >= limit)
726 			break;
727 
728 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
729 			snprintf(xstats_names[count++].name,
730 				 RTE_ETH_XSTATS_NAME_SIZE,
731 				 "tx_q%u_%s", i, hn_stat_strings[t].name);
732 	}
733 
734 	for (i = 0; i < dev->data->nb_rx_queues; i++)  {
735 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
736 
737 		if (!rxq)
738 			continue;
739 
740 		if (count >= limit)
741 			break;
742 
743 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
744 			snprintf(xstats_names[count++].name,
745 				 RTE_ETH_XSTATS_NAME_SIZE,
746 				 "rx_q%u_%s", i,
747 				 hn_stat_strings[t].name);
748 	}
749 
750 	ret = hn_vf_xstats_get_names(dev, xstats_names + count,
751 				     limit - count);
752 	if (ret < 0)
753 		return ret;
754 
755 	return count + ret;
756 }
757 
758 static int
759 hn_dev_xstats_get(struct rte_eth_dev *dev,
760 		  struct rte_eth_xstat *xstats,
761 		  unsigned int n)
762 {
763 	unsigned int i, t, count = 0;
764 	const unsigned int nstats = hn_dev_xstats_count(dev);
765 	const char *stats;
766 	int ret;
767 
768 	PMD_INIT_FUNC_TRACE();
769 
770 	if (n < nstats)
771 		return nstats;
772 
773 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
774 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
775 
776 		if (!txq)
777 			continue;
778 
779 		stats = (const char *)&txq->stats;
780 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
781 			xstats[count].id = count;
782 			xstats[count].value = *(const uint64_t *)
783 				(stats + hn_stat_strings[t].offset);
784 		}
785 	}
786 
787 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
788 		const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
789 
790 		if (!rxq)
791 			continue;
792 
793 		stats = (const char *)&rxq->stats;
794 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
795 			xstats[count].id = count;
796 			xstats[count].value = *(const uint64_t *)
797 				(stats + hn_stat_strings[t].offset);
798 		}
799 	}
800 
801 	ret = hn_vf_xstats_get(dev, xstats, count, n);
802 	if (ret < 0)
803 		return ret;
804 
805 	return count + ret;
806 }
807 
808 static int
809 hn_dev_start(struct rte_eth_dev *dev)
810 {
811 	struct hn_data *hv = dev->data->dev_private;
812 	int error;
813 
814 	PMD_INIT_FUNC_TRACE();
815 
816 	error = hn_rndis_set_rxfilter(hv,
817 				      NDIS_PACKET_TYPE_BROADCAST |
818 				      NDIS_PACKET_TYPE_ALL_MULTICAST |
819 				      NDIS_PACKET_TYPE_DIRECTED);
820 	if (error)
821 		return error;
822 
823 	error = hn_vf_start(dev);
824 	if (error)
825 		hn_rndis_set_rxfilter(hv, 0);
826 
827 	/* Initialize Link state */
828 	if (error == 0)
829 		hn_dev_link_update(dev, 0);
830 
831 	return error;
832 }
833 
834 static void
835 hn_dev_stop(struct rte_eth_dev *dev)
836 {
837 	struct hn_data *hv = dev->data->dev_private;
838 
839 	PMD_INIT_FUNC_TRACE();
840 
841 	hn_rndis_set_rxfilter(hv, 0);
842 	hn_vf_stop(dev);
843 }
844 
845 static void
846 hn_dev_close(struct rte_eth_dev *dev)
847 {
848 	PMD_INIT_FUNC_TRACE();
849 
850 	hn_vf_close(dev);
851 	hn_dev_free_queues(dev);
852 }
853 
854 static const struct eth_dev_ops hn_eth_dev_ops = {
855 	.dev_configure		= hn_dev_configure,
856 	.dev_start		= hn_dev_start,
857 	.dev_stop		= hn_dev_stop,
858 	.dev_close		= hn_dev_close,
859 	.dev_infos_get		= hn_dev_info_get,
860 	.dev_supported_ptypes_get = hn_vf_supported_ptypes,
861 	.promiscuous_enable     = hn_dev_promiscuous_enable,
862 	.promiscuous_disable    = hn_dev_promiscuous_disable,
863 	.allmulticast_enable    = hn_dev_allmulticast_enable,
864 	.allmulticast_disable   = hn_dev_allmulticast_disable,
865 	.set_mc_addr_list	= hn_dev_mc_addr_list,
866 	.reta_update		= hn_rss_reta_update,
867 	.reta_query             = hn_rss_reta_query,
868 	.rss_hash_update	= hn_rss_hash_update,
869 	.rss_hash_conf_get      = hn_rss_hash_conf_get,
870 	.tx_queue_setup		= hn_dev_tx_queue_setup,
871 	.tx_queue_release	= hn_dev_tx_queue_release,
872 	.tx_done_cleanup        = hn_dev_tx_done_cleanup,
873 	.rx_queue_setup		= hn_dev_rx_queue_setup,
874 	.rx_queue_release	= hn_dev_rx_queue_release,
875 	.link_update		= hn_dev_link_update,
876 	.stats_get		= hn_dev_stats_get,
877 	.stats_reset            = hn_dev_stats_reset,
878 	.xstats_get		= hn_dev_xstats_get,
879 	.xstats_get_names	= hn_dev_xstats_get_names,
880 	.xstats_reset		= hn_dev_xstats_reset,
881 };
882 
883 /*
884  * Setup connection between PMD and kernel.
885  */
886 static int
887 hn_attach(struct hn_data *hv, unsigned int mtu)
888 {
889 	int error;
890 
891 	/* Attach NVS */
892 	error = hn_nvs_attach(hv, mtu);
893 	if (error)
894 		goto failed_nvs;
895 
896 	/* Attach RNDIS */
897 	error = hn_rndis_attach(hv);
898 	if (error)
899 		goto failed_rndis;
900 
901 	/*
902 	 * NOTE:
903 	 * Under certain conditions on certain versions of Hyper-V,
904 	 * the RNDIS rxfilter is _not_ zero on the hypervisor side
905 	 * after the successful RNDIS initialization.
906 	 */
907 	hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_NONE);
908 	return 0;
909 failed_rndis:
910 	hn_nvs_detach(hv);
911 failed_nvs:
912 	return error;
913 }
914 
915 static void
916 hn_detach(struct hn_data *hv)
917 {
918 	hn_nvs_detach(hv);
919 	hn_rndis_detach(hv);
920 }
921 
922 static int
923 eth_hn_dev_init(struct rte_eth_dev *eth_dev)
924 {
925 	struct hn_data *hv = eth_dev->data->dev_private;
926 	struct rte_device *device = eth_dev->device;
927 	struct rte_vmbus_device *vmbus;
928 	unsigned int rxr_cnt;
929 	int err, max_chan;
930 
931 	PMD_INIT_FUNC_TRACE();
932 
933 	vmbus = container_of(device, struct rte_vmbus_device, device);
934 	eth_dev->dev_ops = &hn_eth_dev_ops;
935 	eth_dev->tx_pkt_burst = &hn_xmit_pkts;
936 	eth_dev->rx_pkt_burst = &hn_recv_pkts;
937 
938 	/*
939 	 * for secondary processes, we don't initialize any further as primary
940 	 * has already done this work.
941 	 */
942 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
943 		return 0;
944 
945 	/* Since Hyper-V only supports one MAC address */
946 	eth_dev->data->mac_addrs = rte_calloc("hv_mac", HN_MAX_MAC_ADDRS,
947 					      sizeof(struct rte_ether_addr), 0);
948 	if (eth_dev->data->mac_addrs == NULL) {
949 		PMD_INIT_LOG(ERR,
950 			     "Failed to allocate memory store MAC addresses");
951 		return -ENOMEM;
952 	}
953 
954 	hv->vmbus = vmbus;
955 	hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP];
956 	hv->chim_res  = &vmbus->resource[HV_SEND_BUF_MAP];
957 	hv->port_id = eth_dev->data->port_id;
958 	hv->latency = HN_CHAN_LATENCY_NS;
959 	hv->max_queues = 1;
960 	rte_spinlock_init(&hv->vf_lock);
961 	hv->vf_port = HN_INVALID_PORT;
962 
963 	err = hn_parse_args(eth_dev);
964 	if (err)
965 		return err;
966 
967 	strlcpy(hv->owner.name, eth_dev->device->name,
968 		RTE_ETH_MAX_OWNER_NAME_LEN);
969 	err = rte_eth_dev_owner_new(&hv->owner.id);
970 	if (err) {
971 		PMD_INIT_LOG(ERR, "Can not get owner id");
972 		return err;
973 	}
974 
975 	/* Initialize primary channel input for control operations */
976 	err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
977 	if (err)
978 		return err;
979 
980 	rte_vmbus_set_latency(hv->vmbus, hv->channels[0], hv->latency);
981 
982 	hv->primary = hn_rx_queue_alloc(hv, 0,
983 					eth_dev->device->numa_node);
984 
985 	if (!hv->primary)
986 		return -ENOMEM;
987 
988 	err = hn_attach(hv, RTE_ETHER_MTU);
989 	if  (err)
990 		goto failed;
991 
992 	err = hn_chim_init(eth_dev);
993 	if (err)
994 		goto failed;
995 
996 	err = hn_rndis_get_eaddr(hv, eth_dev->data->mac_addrs->addr_bytes);
997 	if (err)
998 		goto failed;
999 
1000 	/* Multi queue requires later versions of windows server */
1001 	if (hv->nvs_ver < NVS_VERSION_5)
1002 		return 0;
1003 
1004 	max_chan = rte_vmbus_max_channels(vmbus);
1005 	PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
1006 	if (max_chan <= 0)
1007 		goto failed;
1008 
1009 	if (hn_rndis_query_rsscaps(hv, &rxr_cnt) != 0)
1010 		rxr_cnt = 1;
1011 
1012 	hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
1013 
1014 	/* If VF was reported but not added, do it now */
1015 	if (hv->vf_present && !hn_vf_attached(hv)) {
1016 		PMD_INIT_LOG(DEBUG, "Adding VF device");
1017 
1018 		err = hn_vf_add(eth_dev, hv);
1019 		if (err)
1020 			hv->vf_present = 0;
1021 	}
1022 
1023 	return 0;
1024 
1025 failed:
1026 	PMD_INIT_LOG(NOTICE, "device init failed");
1027 
1028 	hn_chim_uninit(eth_dev);
1029 	hn_detach(hv);
1030 	return err;
1031 }
1032 
1033 static int
1034 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
1035 {
1036 	struct hn_data *hv = eth_dev->data->dev_private;
1037 	int ret;
1038 
1039 	PMD_INIT_FUNC_TRACE();
1040 
1041 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1042 		return 0;
1043 
1044 	hn_dev_stop(eth_dev);
1045 	hn_dev_close(eth_dev);
1046 
1047 	eth_dev->dev_ops = NULL;
1048 	eth_dev->tx_pkt_burst = NULL;
1049 	eth_dev->rx_pkt_burst = NULL;
1050 
1051 	hn_detach(hv);
1052 	hn_chim_uninit(eth_dev);
1053 	rte_vmbus_chan_close(hv->primary->chan);
1054 	rte_free(hv->primary);
1055 	ret = rte_eth_dev_owner_delete(hv->owner.id);
1056 	if (ret != 0)
1057 		return ret;
1058 
1059 	return 0;
1060 }
1061 
1062 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
1063 			struct rte_vmbus_device *dev)
1064 {
1065 	struct rte_eth_dev *eth_dev;
1066 	int ret;
1067 
1068 	PMD_INIT_FUNC_TRACE();
1069 
1070 	eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
1071 	if (!eth_dev)
1072 		return -ENOMEM;
1073 
1074 	ret = eth_hn_dev_init(eth_dev);
1075 	if (ret)
1076 		eth_dev_vmbus_release(eth_dev);
1077 	else
1078 		rte_eth_dev_probing_finish(eth_dev);
1079 
1080 	return ret;
1081 }
1082 
1083 static int eth_hn_remove(struct rte_vmbus_device *dev)
1084 {
1085 	struct rte_eth_dev *eth_dev;
1086 	int ret;
1087 
1088 	PMD_INIT_FUNC_TRACE();
1089 
1090 	eth_dev = rte_eth_dev_allocated(dev->device.name);
1091 	if (!eth_dev)
1092 		return -ENODEV;
1093 
1094 	ret = eth_hn_dev_uninit(eth_dev);
1095 	if (ret)
1096 		return ret;
1097 
1098 	eth_dev_vmbus_release(eth_dev);
1099 	return 0;
1100 }
1101 
1102 /* Network device GUID */
1103 static const rte_uuid_t hn_net_ids[] = {
1104 	/*  f8615163-df3e-46c5-913f-f2d2f965ed0e */
1105 	RTE_UUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x913f, 0xf2d2f965ed0eULL),
1106 	{ 0 }
1107 };
1108 
1109 static struct rte_vmbus_driver rte_netvsc_pmd = {
1110 	.id_table = hn_net_ids,
1111 	.probe = eth_hn_probe,
1112 	.remove = eth_hn_remove,
1113 };
1114 
1115 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
1116 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
1117 
1118 RTE_INIT(hn_init_log)
1119 {
1120 	hn_logtype_init = rte_log_register("pmd.net.netvsc.init");
1121 	if (hn_logtype_init >= 0)
1122 		rte_log_set_level(hn_logtype_init, RTE_LOG_NOTICE);
1123 	hn_logtype_driver = rte_log_register("pmd.net.netvsc.driver");
1124 	if (hn_logtype_driver >= 0)
1125 		rte_log_set_level(hn_logtype_driver, RTE_LOG_NOTICE);
1126 }
1127