xref: /dpdk/drivers/net/mlx5/windows/mlx5_os.c (revision daa02b5cddbb8e11b31d41e2bf7bb1ae64dcae2f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2020 Mellanox Technologies, Ltd
3  */
4 
5 #include <errno.h>
6 #include <stdalign.h>
7 #include <stddef.h>
8 #include <stdint.h>
9 #include <stdlib.h>
10 
11 #include <rte_windows.h>
12 #include <ethdev_pci.h>
13 
14 #include <mlx5_glue.h>
15 #include <mlx5_devx_cmds.h>
16 #include <mlx5_common.h>
17 #include <mlx5_common_mp.h>
18 #include <mlx5_common_mr.h>
19 #include <mlx5_malloc.h>
20 
21 #include "mlx5_defs.h"
22 #include "mlx5.h"
23 #include "mlx5_common_os.h"
24 #include "mlx5_utils.h"
25 #include "mlx5_rxtx.h"
26 #include "mlx5_rx.h"
27 #include "mlx5_tx.h"
28 #include "mlx5_autoconf.h"
29 #include "mlx5_flow.h"
30 #include "mlx5_devx.h"
31 
32 static const char *MZ_MLX5_PMD_SHARED_DATA = "mlx5_pmd_shared_data";
33 
34 /* Spinlock for mlx5_shared_data allocation. */
35 static rte_spinlock_t mlx5_shared_data_lock = RTE_SPINLOCK_INITIALIZER;
36 
37 /* rte flow indexed pool configuration. */
38 static struct mlx5_indexed_pool_config icfg[] = {
39 	{
40 		.size = sizeof(struct rte_flow),
41 		.trunk_size = 64,
42 		.need_lock = 1,
43 		.release_mem_en = 0,
44 		.malloc = mlx5_malloc,
45 		.free = mlx5_free,
46 		.per_core_cache = 0,
47 		.type = "ctl_flow_ipool",
48 	},
49 	{
50 		.size = sizeof(struct rte_flow),
51 		.trunk_size = 64,
52 		.grow_trunk = 3,
53 		.grow_shift = 2,
54 		.need_lock = 1,
55 		.release_mem_en = 0,
56 		.malloc = mlx5_malloc,
57 		.free = mlx5_free,
58 		.per_core_cache = 1 << 14,
59 		.type = "rte_flow_ipool",
60 	},
61 	{
62 		.size = sizeof(struct rte_flow),
63 		.trunk_size = 64,
64 		.grow_trunk = 3,
65 		.grow_shift = 2,
66 		.need_lock = 1,
67 		.release_mem_en = 0,
68 		.malloc = mlx5_malloc,
69 		.free = mlx5_free,
70 		.per_core_cache = 0,
71 		.type = "mcp_flow_ipool",
72 	},
73 };
74 
75 /**
76  * Initialize shared data between primary and secondary process.
77  *
78  * A memzone is reserved by primary process and secondary processes attach to
79  * the memzone.
80  *
81  * @return
82  *   0 on success, a negative errno value otherwise and rte_errno is set.
83  */
84 static int
85 mlx5_init_shared_data(void)
86 {
87 	const struct rte_memzone *mz;
88 	int ret = 0;
89 
90 	rte_spinlock_lock(&mlx5_shared_data_lock);
91 	if (mlx5_shared_data == NULL) {
92 		/* Allocate shared memory. */
93 		mz = rte_memzone_reserve(MZ_MLX5_PMD_SHARED_DATA,
94 					 sizeof(*mlx5_shared_data),
95 					 SOCKET_ID_ANY, 0);
96 		if (mz == NULL) {
97 			DRV_LOG(ERR,
98 				"Cannot allocate mlx5 shared data");
99 			ret = -rte_errno;
100 			goto error;
101 		}
102 		mlx5_shared_data = mz->addr;
103 		memset(mlx5_shared_data, 0, sizeof(*mlx5_shared_data));
104 		rte_spinlock_init(&mlx5_shared_data->lock);
105 	}
106 error:
107 	rte_spinlock_unlock(&mlx5_shared_data_lock);
108 	return ret;
109 }
110 
111 /**
112  * PMD global initialization.
113  *
114  * Independent from individual device, this function initializes global
115  * per-PMD data structures distinguishing primary and secondary processes.
116  * Hence, each initialization is called once per a process.
117  *
118  * @return
119  *   0 on success, a negative errno value otherwise and rte_errno is set.
120  */
121 static int
122 mlx5_init_once(void)
123 {
124 	if (mlx5_init_shared_data())
125 		return -rte_errno;
126 	return 0;
127 }
128 
129 /**
130  * Get mlx5 device attributes.
131  *
132  * @param cdev
133  *   Pointer to mlx5 device.
134  *
135  * @param device_attr
136  *   Pointer to mlx5 device attributes.
137  *
138  * @return
139  *   0 on success, non zero error number otherwise.
140  */
141 int
142 mlx5_os_get_dev_attr(struct mlx5_common_device *cdev,
143 		     struct mlx5_dev_attr *device_attr)
144 {
145 	struct mlx5_context *mlx5_ctx;
146 	void *pv_iseg = NULL;
147 	u32 cb_iseg = 0;
148 	int err = 0;
149 
150 	if (!cdev || !cdev->ctx)
151 		return -EINVAL;
152 	mlx5_ctx = (struct mlx5_context *)cdev->ctx;
153 	memset(device_attr, 0, sizeof(*device_attr));
154 	device_attr->max_cq = 1 << cdev->config.hca_attr.log_max_cq;
155 	device_attr->max_qp = 1 << cdev->config.hca_attr.log_max_qp;
156 	device_attr->max_qp_wr = 1 << cdev->config.hca_attr.log_max_qp_sz;
157 	device_attr->max_cqe = 1 << cdev->config.hca_attr.log_max_cq_sz;
158 	device_attr->max_mr = 1 << cdev->config.hca_attr.log_max_mrw_sz;
159 	device_attr->max_pd = 1 << cdev->config.hca_attr.log_max_pd;
160 	device_attr->max_srq = 1 << cdev->config.hca_attr.log_max_srq;
161 	device_attr->max_srq_wr = 1 << cdev->config.hca_attr.log_max_srq_sz;
162 	device_attr->max_tso = 1 << cdev->config.hca_attr.max_lso_cap;
163 	if (cdev->config.hca_attr.rss_ind_tbl_cap) {
164 		device_attr->max_rwq_indirection_table_size =
165 			1 << cdev->config.hca_attr.rss_ind_tbl_cap;
166 	}
167 	device_attr->sw_parsing_offloads =
168 		mlx5_get_supported_sw_parsing_offloads(&cdev->config.hca_attr);
169 	device_attr->tunnel_offloads_caps =
170 		mlx5_get_supported_tunneling_offloads(&cdev->config.hca_attr);
171 	pv_iseg = mlx5_glue->query_hca_iseg(mlx5_ctx, &cb_iseg);
172 	if (pv_iseg == NULL) {
173 		DRV_LOG(ERR, "Failed to get device hca_iseg");
174 		return errno;
175 	}
176 	if (!err) {
177 		snprintf(device_attr->fw_ver, 64, "%x.%x.%04x",
178 			MLX5_GET(initial_seg, pv_iseg, fw_rev_major),
179 			MLX5_GET(initial_seg, pv_iseg, fw_rev_minor),
180 			MLX5_GET(initial_seg, pv_iseg, fw_rev_subminor));
181 	}
182 	return err;
183 }
184 
185 /**
186  * Initialize DR related data within private structure.
187  * Routine checks the reference counter and does actual
188  * resources creation/initialization only if counter is zero.
189  *
190  * @param[in] priv
191  *   Pointer to the private device data structure.
192  *
193  * @return
194  *   Zero on success, positive error code otherwise.
195  */
196 static int
197 mlx5_alloc_shared_dr(struct mlx5_priv *priv)
198 {
199 	struct mlx5_dev_ctx_shared *sh = priv->sh;
200 	int err = 0;
201 
202 	if (!sh->flow_tbls)
203 		err = mlx5_alloc_table_hash_list(priv);
204 	else
205 		DRV_LOG(DEBUG, "sh->flow_tbls[%p] already created, reuse",
206 			(void *)sh->flow_tbls);
207 	return err;
208 }
209 /**
210  * Destroy DR related data within private structure.
211  *
212  * @param[in] priv
213  *   Pointer to the private device data structure.
214  */
215 void
216 mlx5_os_free_shared_dr(struct mlx5_priv *priv)
217 {
218 	mlx5_free_table_hash_list(priv);
219 }
220 
221 /**
222  * Set the completion channel file descriptor interrupt as non-blocking.
223  * Currently it has no support under Windows.
224  *
225  * @param[in] rxq_obj
226  *   Pointer to RQ channel object, which includes the channel fd
227  *
228  * @param[out] fd
229  *   The file descriptor (representing the intetrrupt) used in this channel.
230  *
231  * @return
232  *   0 on successfully setting the fd to non-blocking, non-zero otherwise.
233  */
234 int
235 mlx5_os_set_nonblock_channel_fd(int fd)
236 {
237 	(void)fd;
238 	DRV_LOG(WARNING, "%s: is not supported", __func__);
239 	return -ENOTSUP;
240 }
241 
242 /**
243  * DV flow counter mode detect and config.
244  *
245  * @param dev
246  *   Pointer to rte_eth_dev structure.
247  *
248  */
249 static void
250 mlx5_flow_counter_mode_config(struct rte_eth_dev *dev __rte_unused)
251 {
252 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
253 	struct mlx5_priv *priv = dev->data->dev_private;
254 	struct mlx5_dev_ctx_shared *sh = priv->sh;
255 	bool fallback;
256 
257 #ifndef HAVE_IBV_DEVX_ASYNC
258 	fallback = true;
259 #else
260 	fallback = false;
261 	if (!sh->devx || !priv->config.dv_flow_en ||
262 	    !priv->config.hca_attr.flow_counters_dump ||
263 	    !(priv->config.hca_attr.flow_counter_bulk_alloc_bitmap & 0x4) ||
264 	    (mlx5_flow_dv_discover_counter_offset_support(dev) == -ENOTSUP))
265 		fallback = true;
266 #endif
267 	if (fallback)
268 		DRV_LOG(INFO, "Use fall-back DV counter management. Flow "
269 			"counter dump:%d, bulk_alloc_bitmap:0x%hhx.",
270 			priv->config.hca_attr.flow_counters_dump,
271 			priv->config.hca_attr.flow_counter_bulk_alloc_bitmap);
272 	/* Initialize fallback mode only on the port initializes sh. */
273 	if (sh->refcnt == 1)
274 		sh->cmng.counter_fallback = fallback;
275 	else if (fallback != sh->cmng.counter_fallback)
276 		DRV_LOG(WARNING, "Port %d in sh has different fallback mode "
277 			"with others:%d.", PORT_ID(priv), fallback);
278 #endif
279 }
280 
281 /**
282  * Spawn an Ethernet device from DevX information.
283  *
284  * @param dpdk_dev
285  *   Backing DPDK device.
286  * @param spawn
287  *   Verbs device parameters (name, port, switch_info) to spawn.
288  * @param config
289  *   Device configuration parameters.
290  *
291  * @return
292  *   A valid Ethernet device object on success, NULL otherwise and rte_errno
293  *   is set. The following errors are defined:
294  *
295  *   EEXIST: device is already spawned
296  */
297 static struct rte_eth_dev *
298 mlx5_dev_spawn(struct rte_device *dpdk_dev,
299 	       struct mlx5_dev_spawn_data *spawn,
300 	       struct mlx5_dev_config *config)
301 {
302 	const struct mlx5_switch_info *switch_info = &spawn->info;
303 	struct mlx5_dev_ctx_shared *sh = NULL;
304 	struct mlx5_dev_attr device_attr;
305 	struct rte_eth_dev *eth_dev = NULL;
306 	struct mlx5_priv *priv = NULL;
307 	int err = 0;
308 	unsigned int cqe_comp;
309 	struct rte_ether_addr mac;
310 	char name[RTE_ETH_NAME_MAX_LEN];
311 	int own_domain_id = 0;
312 	uint16_t port_id;
313 	int i;
314 
315 	/* Build device name. */
316 	strlcpy(name, dpdk_dev->name, sizeof(name));
317 	/* check if the device is already spawned */
318 	if (rte_eth_dev_get_port_by_name(name, &port_id) == 0) {
319 		rte_errno = EEXIST;
320 		return NULL;
321 	}
322 	DRV_LOG(DEBUG, "naming Ethernet device \"%s\"", name);
323 	/*
324 	 * Some parameters are needed in advance to create device context. We
325 	 * process the devargs here to get ones, and later process devargs
326 	 * again to override some hardware settings.
327 	 */
328 	err = mlx5_args(config, dpdk_dev->devargs);
329 	if (err) {
330 		err = rte_errno;
331 		DRV_LOG(ERR, "failed to process device arguments: %s",
332 			strerror(rte_errno));
333 		goto error;
334 	}
335 	sh = mlx5_alloc_shared_dev_ctx(spawn, config);
336 	if (!sh)
337 		return NULL;
338 	/* Initialize the shutdown event in mlx5_dev_spawn to
339 	 * support mlx5_is_removed for Windows.
340 	 */
341 	err = mlx5_glue->devx_init_showdown_event(sh->cdev->ctx);
342 	if (err) {
343 		DRV_LOG(ERR, "failed to init showdown event: %s",
344 			strerror(errno));
345 		goto error;
346 	}
347 	DRV_LOG(DEBUG, "MPW isn't supported");
348 	mlx5_os_get_dev_attr(sh->cdev, &device_attr);
349 	config->swp = device_attr.sw_parsing_offloads &
350 		(MLX5_SW_PARSING_CAP | MLX5_SW_PARSING_CSUM_CAP |
351 		 MLX5_SW_PARSING_TSO_CAP);
352 	config->ind_table_max_size =
353 		sh->device_attr.max_rwq_indirection_table_size;
354 	cqe_comp = 0;
355 	config->cqe_comp = cqe_comp;
356 	config->tunnel_en = device_attr.tunnel_offloads_caps &
357 		(MLX5_TUNNELED_OFFLOADS_VXLAN_CAP |
358 		 MLX5_TUNNELED_OFFLOADS_GRE_CAP |
359 		 MLX5_TUNNELED_OFFLOADS_GENEVE_CAP);
360 	if (config->tunnel_en) {
361 		DRV_LOG(DEBUG, "tunnel offloading is supported for %s%s%s",
362 		config->tunnel_en &
363 		MLX5_TUNNELED_OFFLOADS_VXLAN_CAP ? "[VXLAN]" : "",
364 		config->tunnel_en &
365 		MLX5_TUNNELED_OFFLOADS_GRE_CAP ? "[GRE]" : "",
366 		config->tunnel_en &
367 		MLX5_TUNNELED_OFFLOADS_GENEVE_CAP ? "[GENEVE]" : ""
368 		);
369 	} else {
370 		DRV_LOG(DEBUG, "tunnel offloading is not supported");
371 	}
372 	DRV_LOG(DEBUG, "MPLS over GRE/UDP tunnel offloading is no supported");
373 	config->mpls_en = 0;
374 	/* Allocate private eth device data. */
375 	priv = mlx5_malloc(MLX5_MEM_ZERO | MLX5_MEM_RTE,
376 			   sizeof(*priv),
377 			   RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
378 	if (priv == NULL) {
379 		DRV_LOG(ERR, "priv allocation failure");
380 		err = ENOMEM;
381 		goto error;
382 	}
383 	priv->sh = sh;
384 	priv->dev_port = spawn->phys_port;
385 	priv->pci_dev = spawn->pci_dev;
386 	priv->mtu = RTE_ETHER_MTU;
387 	priv->mp_id.port_id = port_id;
388 	strlcpy(priv->mp_id.name, MLX5_MP_NAME, RTE_MP_MAX_NAME_LEN);
389 	priv->representor = !!switch_info->representor;
390 	priv->master = !!switch_info->master;
391 	priv->domain_id = RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID;
392 	priv->vport_meta_tag = 0;
393 	priv->vport_meta_mask = 0;
394 	priv->pf_bond = spawn->pf_bond;
395 	priv->vport_id = -1;
396 	/* representor_id field keeps the unmodified VF index. */
397 	priv->representor_id = -1;
398 	/*
399 	 * Look for sibling devices in order to reuse their switch domain
400 	 * if any, otherwise allocate one.
401 	 */
402 	MLX5_ETH_FOREACH_DEV(port_id, dpdk_dev) {
403 		const struct mlx5_priv *opriv =
404 			rte_eth_devices[port_id].data->dev_private;
405 
406 		if (!opriv ||
407 		    opriv->sh != priv->sh ||
408 			opriv->domain_id ==
409 			RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID)
410 			continue;
411 		priv->domain_id = opriv->domain_id;
412 		break;
413 	}
414 	if (priv->domain_id == RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID) {
415 		err = rte_eth_switch_domain_alloc(&priv->domain_id);
416 		if (err) {
417 			err = rte_errno;
418 			DRV_LOG(ERR, "unable to allocate switch domain: %s",
419 				strerror(rte_errno));
420 			goto error;
421 		}
422 		own_domain_id = 1;
423 	}
424 	/* Override some values set by hardware configuration. */
425 	mlx5_args(config, dpdk_dev->devargs);
426 	err = mlx5_dev_check_sibling_config(priv, config, dpdk_dev);
427 	if (err)
428 		goto error;
429 	DRV_LOG(DEBUG, "counters are not supported");
430 	config->ind_table_max_size =
431 		sh->device_attr.max_rwq_indirection_table_size;
432 	/*
433 	 * Remove this check once DPDK supports larger/variable
434 	 * indirection tables.
435 	 */
436 	if (config->ind_table_max_size > (unsigned int)RTE_ETH_RSS_RETA_SIZE_512)
437 		config->ind_table_max_size = RTE_ETH_RSS_RETA_SIZE_512;
438 	DRV_LOG(DEBUG, "maximum Rx indirection table size is %u",
439 		config->ind_table_max_size);
440 	if (config->hw_padding) {
441 		DRV_LOG(DEBUG, "Rx end alignment padding isn't supported");
442 		config->hw_padding = 0;
443 	}
444 	config->tso = (sh->device_attr.max_tso > 0);
445 	if (config->tso)
446 		config->tso_max_payload_sz = sh->device_attr.max_tso;
447 	DRV_LOG(DEBUG, "%sMPS is %s.",
448 		config->mps == MLX5_MPW_ENHANCED ? "enhanced " :
449 		config->mps == MLX5_MPW ? "legacy " : "",
450 		config->mps != MLX5_MPW_DISABLED ? "enabled" : "disabled");
451 	if (config->cqe_comp && !cqe_comp) {
452 		DRV_LOG(WARNING, "Rx CQE compression isn't supported.");
453 		config->cqe_comp = 0;
454 	}
455 	if (sh->devx) {
456 		config->hca_attr = sh->cdev->config.hca_attr;
457 		config->hw_csum = config->hca_attr.csum_cap;
458 		DRV_LOG(DEBUG, "checksum offloading is %ssupported",
459 		    (config->hw_csum ? "" : "not "));
460 		config->hw_vlan_strip = config->hca_attr.vlan_cap;
461 		DRV_LOG(DEBUG, "VLAN stripping is %ssupported",
462 			(config->hw_vlan_strip ? "" : "not "));
463 		config->hw_fcs_strip = config->hca_attr.scatter_fcs;
464 	}
465 	if (sh->devx) {
466 		uint32_t reg[MLX5_ST_SZ_DW(register_mtutc)];
467 
468 		err = config->hca_attr.access_register_user ?
469 			mlx5_devx_cmd_register_read
470 				(sh->cdev->ctx, MLX5_REGISTER_ID_MTUTC, 0,
471 				reg, MLX5_ST_SZ_DW(register_mtutc)) : ENOTSUP;
472 		if (!err) {
473 			uint32_t ts_mode;
474 
475 			/* MTUTC register is read successfully. */
476 			ts_mode = MLX5_GET(register_mtutc, reg,
477 					   time_stamp_mode);
478 			if (ts_mode == MLX5_MTUTC_TIMESTAMP_MODE_REAL_TIME)
479 				config->rt_timestamp = 1;
480 		} else {
481 			/* Kernel does not support register reading. */
482 			if (config->hca_attr.dev_freq_khz ==
483 						 (NS_PER_S / MS_PER_S))
484 				config->rt_timestamp = 1;
485 		}
486 	}
487 	if (config->mprq.enabled) {
488 		DRV_LOG(WARNING, "Multi-Packet RQ isn't supported");
489 		config->mprq.enabled = 0;
490 	}
491 	if (config->max_dump_files_num == 0)
492 		config->max_dump_files_num = 128;
493 	eth_dev = rte_eth_dev_allocate(name);
494 	if (eth_dev == NULL) {
495 		DRV_LOG(ERR, "can not allocate rte ethdev");
496 		err = ENOMEM;
497 		goto error;
498 	}
499 	if (priv->representor) {
500 		eth_dev->data->dev_flags |= RTE_ETH_DEV_REPRESENTOR;
501 		eth_dev->data->representor_id = priv->representor_id;
502 		MLX5_ETH_FOREACH_DEV(port_id, dpdk_dev) {
503 			struct mlx5_priv *opriv =
504 				rte_eth_devices[port_id].data->dev_private;
505 			if (opriv &&
506 			    opriv->master &&
507 			    opriv->domain_id == priv->domain_id &&
508 			    opriv->sh == priv->sh) {
509 				eth_dev->data->backer_port_id = port_id;
510 				break;
511 			}
512 		}
513 		if (port_id >= RTE_MAX_ETHPORTS)
514 			eth_dev->data->backer_port_id = eth_dev->data->port_id;
515 	}
516 	/*
517 	 * Store associated network device interface index. This index
518 	 * is permanent throughout the lifetime of device. So, we may store
519 	 * the ifindex here and use the cached value further.
520 	 */
521 	MLX5_ASSERT(spawn->ifindex);
522 	priv->if_index = spawn->ifindex;
523 	eth_dev->data->dev_private = priv;
524 	priv->dev_data = eth_dev->data;
525 	eth_dev->data->mac_addrs = priv->mac;
526 	eth_dev->device = dpdk_dev;
527 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
528 	/* Configure the first MAC address by default. */
529 	if (mlx5_get_mac(eth_dev, &mac.addr_bytes)) {
530 		DRV_LOG(ERR,
531 			"port %u cannot get MAC address, is mlx5_en"
532 			" loaded? (errno: %s).",
533 			eth_dev->data->port_id, strerror(rte_errno));
534 		err = ENODEV;
535 		goto error;
536 	}
537 	DRV_LOG(INFO,
538 		"port %u MAC address is " RTE_ETHER_ADDR_PRT_FMT,
539 		eth_dev->data->port_id, RTE_ETHER_ADDR_BYTES(&mac));
540 #ifdef RTE_LIBRTE_MLX5_DEBUG
541 	{
542 		char ifname[MLX5_NAMESIZE];
543 
544 		if (mlx5_get_ifname(eth_dev, &ifname) == 0)
545 			DRV_LOG(DEBUG, "port %u ifname is \"%s\"",
546 				eth_dev->data->port_id, ifname);
547 		else
548 			DRV_LOG(DEBUG, "port %u ifname is unknown.",
549 				eth_dev->data->port_id);
550 	}
551 #endif
552 	/* Get actual MTU if possible. */
553 	err = mlx5_get_mtu(eth_dev, &priv->mtu);
554 	if (err) {
555 		err = rte_errno;
556 		goto error;
557 	}
558 	DRV_LOG(DEBUG, "port %u MTU is %u.", eth_dev->data->port_id,
559 		priv->mtu);
560 	/* Initialize burst functions to prevent crashes before link-up. */
561 	eth_dev->rx_pkt_burst = removed_rx_burst;
562 	eth_dev->tx_pkt_burst = removed_tx_burst;
563 	eth_dev->dev_ops = &mlx5_dev_ops;
564 	eth_dev->rx_descriptor_status = mlx5_rx_descriptor_status;
565 	eth_dev->tx_descriptor_status = mlx5_tx_descriptor_status;
566 	eth_dev->rx_queue_count = mlx5_rx_queue_count;
567 	/* Register MAC address. */
568 	claim_zero(mlx5_mac_addr_add(eth_dev, &mac, 0, 0));
569 	priv->ctrl_flows = 0;
570 	TAILQ_INIT(&priv->flow_meters);
571 	priv->mtr_profile_tbl = mlx5_l3t_create(MLX5_L3T_TYPE_PTR);
572 	if (!priv->mtr_profile_tbl)
573 		goto error;
574 	/* Bring Ethernet device up. */
575 	DRV_LOG(DEBUG, "port %u forcing Ethernet interface up.",
576 		eth_dev->data->port_id);
577 	/* nl calls are unsupported - set to -1 not to fail on release */
578 	priv->nl_socket_rdma = -1;
579 	priv->nl_socket_route = -1;
580 	mlx5_set_link_up(eth_dev);
581 	/*
582 	 * Even though the interrupt handler is not installed yet,
583 	 * interrupts will still trigger on the async_fd from
584 	 * Verbs context returned by ibv_open_device().
585 	 */
586 	mlx5_link_update(eth_dev, 0);
587 	config->dv_esw_en = 0;
588 	/* Detect minimal data bytes to inline. */
589 	mlx5_set_min_inline(spawn, config);
590 	/* Store device configuration on private structure. */
591 	priv->config = *config;
592 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
593 		icfg[i].release_mem_en = !!config->reclaim_mode;
594 		if (config->reclaim_mode)
595 			icfg[i].per_core_cache = 0;
596 		priv->flows[i] = mlx5_ipool_create(&icfg[i]);
597 		if (!priv->flows[i])
598 			goto error;
599 	}
600 	/* Create context for virtual machine VLAN workaround. */
601 	priv->vmwa_context = NULL;
602 	if (config->dv_flow_en) {
603 		err = mlx5_alloc_shared_dr(priv);
604 		if (err)
605 			goto error;
606 	}
607 	/* No supported flow priority number detection. */
608 	priv->config.flow_prio = -1;
609 	if (!priv->config.dv_esw_en &&
610 	    priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
611 		DRV_LOG(WARNING, "metadata mode %u is not supported "
612 				 "(no E-Switch)", priv->config.dv_xmeta_en);
613 		priv->config.dv_xmeta_en = MLX5_XMETA_MODE_LEGACY;
614 	}
615 	mlx5_set_metadata_mask(eth_dev);
616 	if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
617 	    !priv->sh->dv_regc0_mask) {
618 		DRV_LOG(ERR, "metadata mode %u is not supported "
619 			     "(no metadata reg_c[0] is available).",
620 			     priv->config.dv_xmeta_en);
621 			err = ENOTSUP;
622 			goto error;
623 	}
624 	priv->hrxqs = mlx5_list_create("hrxq", eth_dev, true,
625 		mlx5_hrxq_create_cb, mlx5_hrxq_match_cb,
626 		mlx5_hrxq_remove_cb, mlx5_hrxq_clone_cb,
627 		mlx5_hrxq_clone_free_cb);
628 	/* Query availability of metadata reg_c's. */
629 	err = mlx5_flow_discover_mreg_c(eth_dev);
630 	if (err < 0) {
631 		err = -err;
632 		goto error;
633 	}
634 	if (!mlx5_flow_ext_mreg_supported(eth_dev)) {
635 		DRV_LOG(DEBUG,
636 			"port %u extensive metadata register is not supported.",
637 			eth_dev->data->port_id);
638 		if (priv->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
639 			DRV_LOG(ERR, "metadata mode %u is not supported "
640 				     "(no metadata registers available).",
641 				     priv->config.dv_xmeta_en);
642 			err = ENOTSUP;
643 			goto error;
644 		}
645 	}
646 	if (sh->devx && config->dv_flow_en) {
647 		priv->obj_ops = devx_obj_ops;
648 	} else {
649 		DRV_LOG(ERR, "Flow mode %u is not supported "
650 				"(Windows flow must be DevX with DV flow enabled).",
651 				priv->config.dv_flow_en);
652 		err = ENOTSUP;
653 		goto error;
654 	}
655 	mlx5_flow_counter_mode_config(eth_dev);
656 	return eth_dev;
657 error:
658 	if (priv) {
659 		if (priv->mtr_profile_tbl)
660 			mlx5_l3t_destroy(priv->mtr_profile_tbl);
661 		if (own_domain_id)
662 			claim_zero(rte_eth_switch_domain_free(priv->domain_id));
663 		mlx5_free(priv);
664 		if (eth_dev != NULL)
665 			eth_dev->data->dev_private = NULL;
666 	}
667 	if (eth_dev != NULL) {
668 		/* mac_addrs must not be freed alone because part of
669 		 * dev_private
670 		 **/
671 		eth_dev->data->mac_addrs = NULL;
672 		rte_eth_dev_release_port(eth_dev);
673 	}
674 	if (sh)
675 		mlx5_free_shared_dev_ctx(sh);
676 	MLX5_ASSERT(err > 0);
677 	rte_errno = err;
678 	return NULL;
679 }
680 
681 /**
682  * This function should share events between multiple ports of single IB
683  * device.  Currently it has no support under Windows.
684  *
685  * @param sh
686  *   Pointer to mlx5_dev_ctx_shared object.
687  */
688 void
689 mlx5_os_dev_shared_handler_install(struct mlx5_dev_ctx_shared *sh)
690 {
691 	(void)sh;
692 	DRV_LOG(WARNING, "%s: is not supported", __func__);
693 }
694 
695 /**
696  * This function should share events between multiple ports of single IB
697  * device.  Currently it has no support under Windows.
698  *
699  * @param dev
700  *   Pointer to mlx5_dev_ctx_shared object.
701  */
702 void
703 mlx5_os_dev_shared_handler_uninstall(struct mlx5_dev_ctx_shared *sh)
704 {
705 	(void)sh;
706 	DRV_LOG(WARNING, "%s: is not supported", __func__);
707 }
708 
709 /**
710  * Read statistics by a named counter.
711  *
712  * @param[in] priv
713  *   Pointer to the private device data structure.
714  * @param[in] ctr_name
715  *   Pointer to the name of the statistic counter to read
716  * @param[out] stat
717  *   Pointer to read statistic value.
718  * @return
719  *   0 on success and stat is valud, 1 if failed to read the value
720  *   rte_errno is set.
721  *
722  */
723 int
724 mlx5_os_read_dev_stat(struct mlx5_priv *priv, const char *ctr_name,
725 		      uint64_t *stat)
726 {
727 	RTE_SET_USED(priv);
728 	RTE_SET_USED(ctr_name);
729 	RTE_SET_USED(stat);
730 	DRV_LOG(WARNING, "%s: is not supported", __func__);
731 	return -ENOTSUP;
732 }
733 
734 /**
735  * Flush device MAC addresses
736  * Currently it has no support under Windows.
737  *
738  * @param dev
739  *   Pointer to Ethernet device structure.
740  *
741  */
742 void
743 mlx5_os_mac_addr_flush(struct rte_eth_dev *dev)
744 {
745 	(void)dev;
746 	DRV_LOG(WARNING, "%s: is not supported", __func__);
747 }
748 
749 /**
750  * Remove a MAC address from device
751  * Currently it has no support under Windows.
752  *
753  * @param dev
754  *   Pointer to Ethernet device structure.
755  * @param index
756  *   MAC address index.
757  */
758 void
759 mlx5_os_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
760 {
761 	(void)dev;
762 	(void)(index);
763 	DRV_LOG(WARNING, "%s: is not supported", __func__);
764 }
765 
766 /**
767  * Adds a MAC address to the device
768  * Currently it has no support under Windows.
769  *
770  * @param dev
771  *   Pointer to Ethernet device structure.
772  * @param mac_addr
773  *   MAC address to register.
774  * @param index
775  *   MAC address index.
776  *
777  * @return
778  *   0 on success, a negative errno value otherwise
779  */
780 int
781 mlx5_os_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac,
782 		     uint32_t index)
783 {
784 	(void)index;
785 	struct rte_ether_addr lmac;
786 
787 	if (mlx5_get_mac(dev, &lmac.addr_bytes)) {
788 		DRV_LOG(ERR,
789 			"port %u cannot get MAC address, is mlx5_en"
790 			" loaded? (errno: %s)",
791 			dev->data->port_id, strerror(rte_errno));
792 		return rte_errno;
793 	}
794 	if (!rte_is_same_ether_addr(&lmac, mac)) {
795 		DRV_LOG(ERR,
796 			"adding new mac address to device is unsupported");
797 		return -ENOTSUP;
798 	}
799 	return 0;
800 }
801 
802 /**
803  * Modify a VF MAC address
804  * Currently it has no support under Windows.
805  *
806  * @param priv
807  *   Pointer to device private data.
808  * @param mac_addr
809  *   MAC address to modify into.
810  * @param iface_idx
811  *   Net device interface index
812  * @param vf_index
813  *   VF index
814  *
815  * @return
816  *   0 on success, a negative errno value otherwise
817  */
818 int
819 mlx5_os_vf_mac_addr_modify(struct mlx5_priv *priv,
820 			   unsigned int iface_idx,
821 			   struct rte_ether_addr *mac_addr,
822 			   int vf_index)
823 {
824 	(void)priv;
825 	(void)iface_idx;
826 	(void)mac_addr;
827 	(void)vf_index;
828 	DRV_LOG(WARNING, "%s: is not supported", __func__);
829 	return -ENOTSUP;
830 }
831 
832 /**
833  * Set device promiscuous mode
834  * Currently it has no support under Windows.
835  *
836  * @param dev
837  *   Pointer to Ethernet device structure.
838  * @param enable
839  *   0 - promiscuous is disabled, otherwise - enabled
840  *
841  * @return
842  *   0 on success, a negative error value otherwise
843  */
844 int
845 mlx5_os_set_promisc(struct rte_eth_dev *dev, int enable)
846 {
847 	(void)dev;
848 	(void)enable;
849 	DRV_LOG(WARNING, "%s: is not supported", __func__);
850 	return -ENOTSUP;
851 }
852 
853 /**
854  * Set device allmulti mode
855  *
856  * @param dev
857  *   Pointer to Ethernet device structure.
858  * @param enable
859  *   0 - all multicase is disabled, otherwise - enabled
860  *
861  * @return
862  *   0 on success, a negative error value otherwise
863  */
864 int
865 mlx5_os_set_allmulti(struct rte_eth_dev *dev, int enable)
866 {
867 	(void)dev;
868 	(void)enable;
869 	DRV_LOG(WARNING, "%s: is not supported", __func__);
870 	return -ENOTSUP;
871 }
872 
873 /**
874  * DPDK callback to register a PCI device.
875  *
876  * This function spawns Ethernet devices out of a given device.
877  *
878  * @param[in] dev
879  *   Pointer to the common device.
880  *
881  * @return
882  *   0 on success, a negative errno value otherwise and rte_errno is set.
883  */
884 int
885 mlx5_os_net_probe(struct mlx5_common_device *cdev)
886 {
887 	struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(cdev->dev);
888 	struct mlx5_dev_spawn_data spawn = {
889 		.pf_bond = -1,
890 		.max_port = 1,
891 		.phys_port = 1,
892 		.phys_dev_name = mlx5_os_get_ctx_device_name(cdev->ctx),
893 		.pci_dev = pci_dev,
894 		.cdev = cdev,
895 		.ifindex = -1, /* Spawn will assign */
896 		.info = (struct mlx5_switch_info){
897 			.name_type = MLX5_PHYS_PORT_NAME_TYPE_UPLINK,
898 		},
899 	};
900 	struct mlx5_dev_config dev_config = {
901 		.rx_vec_en = 1,
902 		.txq_inline_max = MLX5_ARG_UNSET,
903 		.txq_inline_min = MLX5_ARG_UNSET,
904 		.txq_inline_mpw = MLX5_ARG_UNSET,
905 		.txqs_inline = MLX5_ARG_UNSET,
906 		.mprq = {
907 			.max_memcpy_len = MLX5_MPRQ_MEMCPY_DEFAULT_LEN,
908 			.min_rxqs_num = MLX5_MPRQ_MIN_RXQS,
909 		},
910 		.dv_flow_en = 1,
911 		.log_hp_size = MLX5_ARG_UNSET,
912 	};
913 	int ret;
914 	uint32_t restore;
915 
916 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
917 		DRV_LOG(ERR, "Secondary process is not supported on Windows.");
918 		return -ENOTSUP;
919 	}
920 	ret = mlx5_init_once();
921 	if (ret) {
922 		DRV_LOG(ERR, "unable to init PMD global data: %s",
923 			strerror(rte_errno));
924 		return -rte_errno;
925 	}
926 	/* Device specific configuration. */
927 	switch (pci_dev->id.device_id) {
928 	case PCI_DEVICE_ID_MELLANOX_CONNECTX4VF:
929 	case PCI_DEVICE_ID_MELLANOX_CONNECTX4LXVF:
930 	case PCI_DEVICE_ID_MELLANOX_CONNECTX5VF:
931 	case PCI_DEVICE_ID_MELLANOX_CONNECTX5EXVF:
932 	case PCI_DEVICE_ID_MELLANOX_CONNECTX5BFVF:
933 	case PCI_DEVICE_ID_MELLANOX_CONNECTX6VF:
934 	case PCI_DEVICE_ID_MELLANOX_CONNECTXVF:
935 		dev_config.vf = 1;
936 		break;
937 	default:
938 		dev_config.vf = 0;
939 		break;
940 	}
941 	spawn.eth_dev = mlx5_dev_spawn(cdev->dev, &spawn, &dev_config);
942 	if (!spawn.eth_dev)
943 		return -rte_errno;
944 	restore = spawn.eth_dev->data->dev_flags;
945 	rte_eth_copy_pci_info(spawn.eth_dev, pci_dev);
946 	/* Restore non-PCI flags cleared by the above call. */
947 	spawn.eth_dev->data->dev_flags |= restore;
948 	rte_eth_dev_probing_finish(spawn.eth_dev);
949 	return 0;
950 }
951 
952 /**
953  * Cleanup resources when the last device is closed.
954  */
955 void
956 mlx5_os_net_cleanup(void)
957 {
958 }
959 
960 const struct mlx5_flow_driver_ops mlx5_flow_verbs_drv_ops = {0};
961