xref: /dpdk/drivers/net/mlx5/linux/mlx5_os.c (revision 09442498ef736d0a96632cf8b8c15d8ca78a6468)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright 2015 6WIND S.A.
3  * Copyright 2020 Mellanox Technologies, Ltd
4  */
5 
6 #include <stddef.h>
7 #include <unistd.h>
8 #include <string.h>
9 #include <stdint.h>
10 #include <stdlib.h>
11 #include <errno.h>
12 #include <net/if.h>
13 #include <linux/rtnetlink.h>
14 #include <linux/sockios.h>
15 #include <linux/ethtool.h>
16 #include <fcntl.h>
17 
18 #include <rte_malloc.h>
19 #include <ethdev_driver.h>
20 #include <ethdev_pci.h>
21 #include <rte_pci.h>
22 #include <bus_driver.h>
23 #include <bus_pci_driver.h>
24 #include <bus_auxiliary_driver.h>
25 #include <rte_common.h>
26 #include <rte_kvargs.h>
27 #include <rte_rwlock.h>
28 #include <rte_spinlock.h>
29 #include <rte_string_fns.h>
30 #include <rte_alarm.h>
31 #include <rte_eal_paging.h>
32 
33 #include <mlx5_glue.h>
34 #include <mlx5_devx_cmds.h>
35 #include <mlx5_common.h>
36 #include <mlx5_common_mp.h>
37 #include <mlx5_common_mr.h>
38 #include <mlx5_malloc.h>
39 
40 #include "mlx5_defs.h"
41 #include "mlx5.h"
42 #include "mlx5_common_os.h"
43 #include "mlx5_utils.h"
44 #include "mlx5_rxtx.h"
45 #include "mlx5_rx.h"
46 #include "mlx5_tx.h"
47 #include "mlx5_autoconf.h"
48 #include "mlx5_flow.h"
49 #include "rte_pmd_mlx5.h"
50 #include "mlx5_verbs.h"
51 #include "mlx5_nl.h"
52 #include "mlx5_devx.h"
53 
54 #ifndef HAVE_IBV_MLX5_MOD_MPW
55 #define MLX5DV_CONTEXT_FLAGS_MPW_ALLOWED (1 << 2)
56 #define MLX5DV_CONTEXT_FLAGS_ENHANCED_MPW (1 << 3)
57 #endif
58 
59 #ifndef HAVE_IBV_MLX5_MOD_CQE_128B_COMP
60 #define MLX5DV_CONTEXT_FLAGS_CQE_128B_COMP (1 << 4)
61 #endif
62 
63 static const char *MZ_MLX5_PMD_SHARED_DATA = "mlx5_pmd_shared_data";
64 
65 /* Spinlock for mlx5_shared_data allocation. */
66 static rte_spinlock_t mlx5_shared_data_lock = RTE_SPINLOCK_INITIALIZER;
67 
68 /* Process local data for secondary processes. */
69 static struct mlx5_local_data mlx5_local_data;
70 
71 /* rte flow indexed pool configuration. */
72 static struct mlx5_indexed_pool_config icfg[] = {
73 	{
74 		.size = sizeof(struct rte_flow),
75 		.trunk_size = 64,
76 		.need_lock = 1,
77 		.release_mem_en = 0,
78 		.malloc = mlx5_malloc,
79 		.free = mlx5_free,
80 		.per_core_cache = 0,
81 		.type = "ctl_flow_ipool",
82 	},
83 	{
84 		.size = sizeof(struct rte_flow),
85 		.trunk_size = 64,
86 		.grow_trunk = 3,
87 		.grow_shift = 2,
88 		.need_lock = 1,
89 		.release_mem_en = 0,
90 		.malloc = mlx5_malloc,
91 		.free = mlx5_free,
92 		.per_core_cache = 1 << 14,
93 		.type = "rte_flow_ipool",
94 	},
95 	{
96 		.size = sizeof(struct rte_flow),
97 		.trunk_size = 64,
98 		.grow_trunk = 3,
99 		.grow_shift = 2,
100 		.need_lock = 1,
101 		.release_mem_en = 0,
102 		.malloc = mlx5_malloc,
103 		.free = mlx5_free,
104 		.per_core_cache = 0,
105 		.type = "mcp_flow_ipool",
106 	},
107 };
108 
109 /**
110  * Set the completion channel file descriptor interrupt as non-blocking.
111  *
112  * @param[in] rxq_obj
113  *   Pointer to RQ channel object, which includes the channel fd
114  *
115  * @param[out] fd
116  *   The file descriptor (representing the interrupt) used in this channel.
117  *
118  * @return
119  *   0 on successfully setting the fd to non-blocking, non-zero otherwise.
120  */
121 int
122 mlx5_os_set_nonblock_channel_fd(int fd)
123 {
124 	int flags;
125 
126 	flags = fcntl(fd, F_GETFL);
127 	return fcntl(fd, F_SETFL, flags | O_NONBLOCK);
128 }
129 
130 /**
131  * Get mlx5 device attributes. The glue function query_device_ex() is called
132  * with out parameter of type 'struct ibv_device_attr_ex *'. Then fill in mlx5
133  * device attributes from the glue out parameter.
134  *
135  * @param sh
136  *   Pointer to shared device context.
137  *
138  * @return
139  *   0 on success, a negative errno value otherwise and rte_errno is set.
140  */
141 int
142 mlx5_os_capabilities_prepare(struct mlx5_dev_ctx_shared *sh)
143 {
144 	int err;
145 	struct mlx5_common_device *cdev = sh->cdev;
146 	struct mlx5_hca_attr *hca_attr = &cdev->config.hca_attr;
147 	struct ibv_device_attr_ex attr_ex = { .comp_mask = 0 };
148 	struct mlx5dv_context dv_attr = { .comp_mask = 0 };
149 
150 	err = mlx5_glue->query_device_ex(cdev->ctx, NULL, &attr_ex);
151 	if (err) {
152 		rte_errno = errno;
153 		return -rte_errno;
154 	}
155 #ifdef HAVE_IBV_MLX5_MOD_SWP
156 	dv_attr.comp_mask |= MLX5DV_CONTEXT_MASK_SWP;
157 #endif
158 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
159 	dv_attr.comp_mask |= MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS;
160 #endif
161 #ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT
162 	dv_attr.comp_mask |= MLX5DV_CONTEXT_MASK_STRIDING_RQ;
163 #endif
164 	err = mlx5_glue->dv_query_device(cdev->ctx, &dv_attr);
165 	if (err) {
166 		rte_errno = errno;
167 		return -rte_errno;
168 	}
169 	memset(&sh->dev_cap, 0, sizeof(struct mlx5_dev_cap));
170 	if (mlx5_dev_is_pci(cdev->dev))
171 		sh->dev_cap.vf = mlx5_dev_is_vf_pci(RTE_DEV_TO_PCI(cdev->dev));
172 	else
173 		sh->dev_cap.sf = 1;
174 	sh->dev_cap.max_qp_wr = attr_ex.orig_attr.max_qp_wr;
175 	sh->dev_cap.max_sge = attr_ex.orig_attr.max_sge;
176 	sh->dev_cap.max_cq = attr_ex.orig_attr.max_cq;
177 	sh->dev_cap.max_qp = attr_ex.orig_attr.max_qp;
178 #ifdef HAVE_MLX5DV_DR_ACTION_DEST_DEVX_TIR
179 	sh->dev_cap.dest_tir = 1;
180 #endif
181 #if defined(HAVE_IBV_FLOW_DV_SUPPORT) && defined(HAVE_MLX5DV_DR)
182 	DRV_LOG(DEBUG, "DV flow is supported.");
183 	sh->dev_cap.dv_flow_en = 1;
184 #endif
185 #ifdef HAVE_MLX5DV_DR_ESWITCH
186 	if (hca_attr->eswitch_manager && sh->dev_cap.dv_flow_en && sh->esw_mode)
187 		sh->dev_cap.dv_esw_en = 1;
188 #endif
189 	/*
190 	 * Multi-packet send is supported by ConnectX-4 Lx PF as well
191 	 * as all ConnectX-5 devices.
192 	 */
193 	if (dv_attr.flags & MLX5DV_CONTEXT_FLAGS_MPW_ALLOWED) {
194 		if (dv_attr.flags & MLX5DV_CONTEXT_FLAGS_ENHANCED_MPW) {
195 			DRV_LOG(DEBUG, "Enhanced MPW is supported.");
196 			sh->dev_cap.mps = MLX5_MPW_ENHANCED;
197 		} else {
198 			DRV_LOG(DEBUG, "MPW is supported.");
199 			sh->dev_cap.mps = MLX5_MPW;
200 		}
201 	} else {
202 		DRV_LOG(DEBUG, "MPW isn't supported.");
203 		sh->dev_cap.mps = MLX5_MPW_DISABLED;
204 	}
205 #if (RTE_CACHE_LINE_SIZE == 128)
206 	if (dv_attr.flags & MLX5DV_CONTEXT_FLAGS_CQE_128B_COMP)
207 		sh->dev_cap.cqe_comp = 1;
208 	DRV_LOG(DEBUG, "Rx CQE 128B compression is %ssupported.",
209 		sh->dev_cap.cqe_comp ? "" : "not ");
210 #else
211 	sh->dev_cap.cqe_comp = 1;
212 #endif
213 #ifdef HAVE_IBV_DEVICE_MPLS_SUPPORT
214 	sh->dev_cap.mpls_en =
215 		((dv_attr.tunnel_offloads_caps &
216 		  MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_CW_MPLS_OVER_GRE) &&
217 		 (dv_attr.tunnel_offloads_caps &
218 		  MLX5DV_RAW_PACKET_CAP_TUNNELED_OFFLOAD_CW_MPLS_OVER_UDP));
219 	DRV_LOG(DEBUG, "MPLS over GRE/UDP tunnel offloading is %ssupported.",
220 		sh->dev_cap.mpls_en ? "" : "not ");
221 #else
222 	DRV_LOG(WARNING,
223 		"MPLS over GRE/UDP tunnel offloading disabled due to old OFED/rdma-core version or firmware configuration");
224 #endif
225 #if defined(HAVE_IBV_WQ_FLAG_RX_END_PADDING)
226 	sh->dev_cap.hw_padding = !!attr_ex.rx_pad_end_addr_align;
227 #elif defined(HAVE_IBV_WQ_FLAGS_PCI_WRITE_END_PADDING)
228 	sh->dev_cap.hw_padding = !!(attr_ex.device_cap_flags_ex &
229 				    IBV_DEVICE_PCI_WRITE_END_PADDING);
230 #endif
231 	sh->dev_cap.hw_csum =
232 		!!(attr_ex.device_cap_flags_ex & IBV_DEVICE_RAW_IP_CSUM);
233 	DRV_LOG(DEBUG, "Checksum offloading is %ssupported.",
234 		sh->dev_cap.hw_csum ? "" : "not ");
235 	sh->dev_cap.hw_vlan_strip = !!(attr_ex.raw_packet_caps &
236 				       IBV_RAW_PACKET_CAP_CVLAN_STRIPPING);
237 	DRV_LOG(DEBUG, "VLAN stripping is %ssupported.",
238 		(sh->dev_cap.hw_vlan_strip ? "" : "not "));
239 	sh->dev_cap.hw_fcs_strip = !!(attr_ex.raw_packet_caps &
240 				      IBV_RAW_PACKET_CAP_SCATTER_FCS);
241 #if !defined(HAVE_IBV_DEVICE_COUNTERS_SET_V42) && \
242 	!defined(HAVE_IBV_DEVICE_COUNTERS_SET_V45)
243 	DRV_LOG(DEBUG, "Counters are not supported.");
244 #endif
245 	/*
246 	 * DPDK doesn't support larger/variable indirection tables.
247 	 * Once DPDK supports it, take max size from device attr.
248 	 */
249 	sh->dev_cap.ind_table_max_size =
250 			RTE_MIN(attr_ex.rss_caps.max_rwq_indirection_table_size,
251 				(unsigned int)RTE_ETH_RSS_RETA_SIZE_512);
252 	DRV_LOG(DEBUG, "Maximum Rx indirection table size is %u",
253 		sh->dev_cap.ind_table_max_size);
254 	sh->dev_cap.tso = (attr_ex.tso_caps.max_tso > 0 &&
255 			   (attr_ex.tso_caps.supported_qpts &
256 			    (1 << IBV_QPT_RAW_PACKET)));
257 	if (sh->dev_cap.tso)
258 		sh->dev_cap.tso_max_payload_sz = attr_ex.tso_caps.max_tso;
259 	strlcpy(sh->dev_cap.fw_ver, attr_ex.orig_attr.fw_ver,
260 		sizeof(sh->dev_cap.fw_ver));
261 #ifdef HAVE_IBV_MLX5_MOD_SWP
262 	if (dv_attr.comp_mask & MLX5DV_CONTEXT_MASK_SWP)
263 		sh->dev_cap.swp = dv_attr.sw_parsing_caps.sw_parsing_offloads &
264 				  (MLX5_SW_PARSING_CAP |
265 				   MLX5_SW_PARSING_CSUM_CAP |
266 				   MLX5_SW_PARSING_TSO_CAP);
267 	DRV_LOG(DEBUG, "SWP support: %u", sh->dev_cap.swp);
268 #endif
269 #ifdef HAVE_IBV_DEVICE_STRIDING_RQ_SUPPORT
270 	if (dv_attr.comp_mask & MLX5DV_CONTEXT_MASK_STRIDING_RQ) {
271 		struct mlx5dv_striding_rq_caps *strd_rq_caps =
272 				&dv_attr.striding_rq_caps;
273 
274 		sh->dev_cap.mprq.enabled = 1;
275 		sh->dev_cap.mprq.log_min_stride_size =
276 			strd_rq_caps->min_single_stride_log_num_of_bytes;
277 		sh->dev_cap.mprq.log_max_stride_size =
278 			strd_rq_caps->max_single_stride_log_num_of_bytes;
279 		sh->dev_cap.mprq.log_min_stride_num =
280 			strd_rq_caps->min_single_wqe_log_num_of_strides;
281 		sh->dev_cap.mprq.log_max_stride_num =
282 			strd_rq_caps->max_single_wqe_log_num_of_strides;
283 		sh->dev_cap.mprq.log_min_stride_wqe_size =
284 					cdev->config.devx ?
285 					hca_attr->log_min_stride_wqe_sz :
286 					MLX5_MPRQ_LOG_MIN_STRIDE_WQE_SIZE;
287 		DRV_LOG(DEBUG, "\tmin_single_stride_log_num_of_bytes: %u",
288 			sh->dev_cap.mprq.log_min_stride_size);
289 		DRV_LOG(DEBUG, "\tmax_single_stride_log_num_of_bytes: %u",
290 			sh->dev_cap.mprq.log_max_stride_size);
291 		DRV_LOG(DEBUG, "\tmin_single_wqe_log_num_of_strides: %u",
292 			sh->dev_cap.mprq.log_min_stride_num);
293 		DRV_LOG(DEBUG, "\tmax_single_wqe_log_num_of_strides: %u",
294 			sh->dev_cap.mprq.log_max_stride_num);
295 		DRV_LOG(DEBUG, "\tmin_stride_wqe_log_size: %u",
296 			sh->dev_cap.mprq.log_min_stride_wqe_size);
297 		DRV_LOG(DEBUG, "\tsupported_qpts: %d",
298 			strd_rq_caps->supported_qpts);
299 		DRV_LOG(DEBUG, "Device supports Multi-Packet RQ.");
300 	}
301 #endif
302 #ifdef HAVE_IBV_DEVICE_TUNNEL_SUPPORT
303 	if (dv_attr.comp_mask & MLX5DV_CONTEXT_MASK_TUNNEL_OFFLOADS) {
304 		sh->dev_cap.tunnel_en = dv_attr.tunnel_offloads_caps &
305 					(MLX5_TUNNELED_OFFLOADS_VXLAN_CAP |
306 					 MLX5_TUNNELED_OFFLOADS_GRE_CAP |
307 					 MLX5_TUNNELED_OFFLOADS_GENEVE_CAP);
308 	}
309 	if (sh->dev_cap.tunnel_en) {
310 		DRV_LOG(DEBUG, "Tunnel offloading is supported for %s%s%s",
311 			sh->dev_cap.tunnel_en &
312 			MLX5_TUNNELED_OFFLOADS_VXLAN_CAP ? "[VXLAN]" : "",
313 			sh->dev_cap.tunnel_en &
314 			MLX5_TUNNELED_OFFLOADS_GRE_CAP ? "[GRE]" : "",
315 			sh->dev_cap.tunnel_en &
316 			MLX5_TUNNELED_OFFLOADS_GENEVE_CAP ? "[GENEVE]" : "");
317 	} else {
318 		DRV_LOG(DEBUG, "Tunnel offloading is not supported.");
319 	}
320 #else
321 	DRV_LOG(WARNING,
322 		"Tunnel offloading disabled due to old OFED/rdma-core version");
323 #endif
324 	if (!sh->cdev->config.devx)
325 		return 0;
326 	/* Check capabilities for Packet Pacing. */
327 	DRV_LOG(DEBUG, "Timestamp counter frequency %u kHz.",
328 		hca_attr->dev_freq_khz);
329 	DRV_LOG(DEBUG, "Packet pacing is %ssupported.",
330 		hca_attr->qos.packet_pacing ? "" : "not ");
331 	DRV_LOG(DEBUG, "Cross channel ops are %ssupported.",
332 		hca_attr->cross_channel ? "" : "not ");
333 	DRV_LOG(DEBUG, "WQE index ignore is %ssupported.",
334 		hca_attr->wqe_index_ignore ? "" : "not ");
335 	DRV_LOG(DEBUG, "Non-wire SQ feature is %ssupported.",
336 		hca_attr->non_wire_sq ? "" : "not ");
337 	DRV_LOG(DEBUG, "Static WQE SQ feature is %ssupported (%d)",
338 		hca_attr->log_max_static_sq_wq ? "" : "not ",
339 		hca_attr->log_max_static_sq_wq);
340 	DRV_LOG(DEBUG, "WQE rate PP mode is %ssupported.",
341 		hca_attr->qos.wqe_rate_pp ? "" : "not ");
342 	sh->dev_cap.txpp_en = hca_attr->qos.packet_pacing;
343 	if (!hca_attr->cross_channel) {
344 		DRV_LOG(DEBUG,
345 			"Cross channel operations are required for packet pacing.");
346 		sh->dev_cap.txpp_en = 0;
347 	}
348 	if (!hca_attr->wqe_index_ignore) {
349 		DRV_LOG(DEBUG,
350 			"WQE index ignore feature is required for packet pacing.");
351 		sh->dev_cap.txpp_en = 0;
352 	}
353 	if (!hca_attr->non_wire_sq) {
354 		DRV_LOG(DEBUG,
355 			"Non-wire SQ feature is required for packet pacing.");
356 		sh->dev_cap.txpp_en = 0;
357 	}
358 	if (!hca_attr->log_max_static_sq_wq) {
359 		DRV_LOG(DEBUG,
360 			"Static WQE SQ feature is required for packet pacing.");
361 		sh->dev_cap.txpp_en = 0;
362 	}
363 	if (!hca_attr->qos.wqe_rate_pp) {
364 		DRV_LOG(DEBUG,
365 			"WQE rate mode is required for packet pacing.");
366 		sh->dev_cap.txpp_en = 0;
367 	}
368 #ifndef HAVE_MLX5DV_DEVX_UAR_OFFSET
369 	DRV_LOG(DEBUG,
370 		"DevX does not provide UAR offset, can't create queues for packet pacing.");
371 	sh->dev_cap.txpp_en = 0;
372 #endif
373 	sh->dev_cap.scatter_fcs_w_decap_disable =
374 					hca_attr->scatter_fcs_w_decap_disable;
375 	sh->dev_cap.rq_delay_drop_en = hca_attr->rq_delay_drop;
376 	mlx5_rt_timestamp_config(sh, hca_attr);
377 	return 0;
378 }
379 
380 /**
381  * Detect misc5 support or not
382  *
383  * @param[in] priv
384  *   Device private data pointer
385  */
386 #ifdef HAVE_MLX5DV_DR
387 static void
388 __mlx5_discovery_misc5_cap(struct mlx5_priv *priv)
389 {
390 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
391 	/* Dummy VxLAN matcher to detect rdma-core misc5 cap
392 	 * Case: IPv4--->UDP--->VxLAN--->vni
393 	 */
394 	void *tbl;
395 	struct mlx5_flow_dv_match_params matcher_mask;
396 	void *match_m;
397 	void *matcher;
398 	void *headers_m;
399 	void *misc5_m;
400 	uint32_t *tunnel_header_m;
401 	struct mlx5dv_flow_matcher_attr dv_attr;
402 
403 	memset(&matcher_mask, 0, sizeof(matcher_mask));
404 	matcher_mask.size = sizeof(matcher_mask.buf);
405 	match_m = matcher_mask.buf;
406 	headers_m = MLX5_ADDR_OF(fte_match_param, match_m, outer_headers);
407 	misc5_m = MLX5_ADDR_OF(fte_match_param,
408 			       match_m, misc_parameters_5);
409 	tunnel_header_m = (uint32_t *)
410 				MLX5_ADDR_OF(fte_match_set_misc5,
411 				misc5_m, tunnel_header_1);
412 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_protocol, 0xff);
413 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, ip_version, 4);
414 	MLX5_SET(fte_match_set_lyr_2_4, headers_m, udp_dport, 0xffff);
415 	*tunnel_header_m = 0xffffff;
416 
417 	tbl = mlx5_glue->dr_create_flow_tbl(priv->sh->rx_domain, 1);
418 	if (!tbl) {
419 		DRV_LOG(INFO, "No SW steering support");
420 		return;
421 	}
422 	dv_attr.type = IBV_FLOW_ATTR_NORMAL,
423 	dv_attr.match_mask = (void *)&matcher_mask,
424 	dv_attr.match_criteria_enable =
425 			(1 << MLX5_MATCH_CRITERIA_ENABLE_OUTER_BIT) |
426 			(1 << MLX5_MATCH_CRITERIA_ENABLE_MISC5_BIT);
427 	dv_attr.priority = 3;
428 #ifdef HAVE_MLX5DV_DR_ESWITCH
429 	void *misc2_m;
430 	if (priv->sh->config.dv_esw_en) {
431 		/* FDB enabled reg_c_0 */
432 		dv_attr.match_criteria_enable |=
433 				(1 << MLX5_MATCH_CRITERIA_ENABLE_MISC2_BIT);
434 		misc2_m = MLX5_ADDR_OF(fte_match_param,
435 				       match_m, misc_parameters_2);
436 		MLX5_SET(fte_match_set_misc2, misc2_m,
437 			 metadata_reg_c_0, 0xffff);
438 	}
439 #endif
440 	matcher = mlx5_glue->dv_create_flow_matcher(priv->sh->cdev->ctx,
441 						    &dv_attr, tbl);
442 	if (matcher) {
443 		priv->sh->misc5_cap = 1;
444 		mlx5_glue->dv_destroy_flow_matcher(matcher);
445 	}
446 	mlx5_glue->dr_destroy_flow_tbl(tbl);
447 #else
448 	RTE_SET_USED(priv);
449 #endif
450 }
451 #endif
452 
453 /**
454  * Initialize DR related data within private structure.
455  * Routine checks the reference counter and does actual
456  * resources creation/initialization only if counter is zero.
457  *
458  * @param[in] priv
459  *   Pointer to the private device data structure.
460  *
461  * @return
462  *   Zero on success, positive error code otherwise.
463  */
464 static int
465 mlx5_alloc_shared_dr(struct mlx5_priv *priv)
466 {
467 	struct mlx5_dev_ctx_shared *sh = priv->sh;
468 	char s[MLX5_NAME_SIZE] __rte_unused;
469 	int err;
470 
471 	MLX5_ASSERT(sh && sh->refcnt);
472 	if (sh->refcnt > 1)
473 		return 0;
474 	err = mlx5_alloc_table_hash_list(priv);
475 	if (err)
476 		goto error;
477 	if (priv->sh->config.dv_flow_en == 2)
478 		return 0;
479 	/* The resources below are only valid with DV support. */
480 #ifdef HAVE_IBV_FLOW_DV_SUPPORT
481 	/* Init port id action list. */
482 	snprintf(s, sizeof(s), "%s_port_id_action_list", sh->ibdev_name);
483 	sh->port_id_action_list = mlx5_list_create(s, sh, true,
484 						   flow_dv_port_id_create_cb,
485 						   flow_dv_port_id_match_cb,
486 						   flow_dv_port_id_remove_cb,
487 						   flow_dv_port_id_clone_cb,
488 						 flow_dv_port_id_clone_free_cb);
489 	if (!sh->port_id_action_list)
490 		goto error;
491 	/* Init push vlan action list. */
492 	snprintf(s, sizeof(s), "%s_push_vlan_action_list", sh->ibdev_name);
493 	sh->push_vlan_action_list = mlx5_list_create(s, sh, true,
494 						    flow_dv_push_vlan_create_cb,
495 						    flow_dv_push_vlan_match_cb,
496 						    flow_dv_push_vlan_remove_cb,
497 						    flow_dv_push_vlan_clone_cb,
498 					       flow_dv_push_vlan_clone_free_cb);
499 	if (!sh->push_vlan_action_list)
500 		goto error;
501 	/* Init sample action list. */
502 	snprintf(s, sizeof(s), "%s_sample_action_list", sh->ibdev_name);
503 	sh->sample_action_list = mlx5_list_create(s, sh, true,
504 						  flow_dv_sample_create_cb,
505 						  flow_dv_sample_match_cb,
506 						  flow_dv_sample_remove_cb,
507 						  flow_dv_sample_clone_cb,
508 						  flow_dv_sample_clone_free_cb);
509 	if (!sh->sample_action_list)
510 		goto error;
511 	/* Init dest array action list. */
512 	snprintf(s, sizeof(s), "%s_dest_array_list", sh->ibdev_name);
513 	sh->dest_array_list = mlx5_list_create(s, sh, true,
514 					       flow_dv_dest_array_create_cb,
515 					       flow_dv_dest_array_match_cb,
516 					       flow_dv_dest_array_remove_cb,
517 					       flow_dv_dest_array_clone_cb,
518 					      flow_dv_dest_array_clone_free_cb);
519 	if (!sh->dest_array_list)
520 		goto error;
521 	/* Init shared flex parsers list, no need lcore_share */
522 	snprintf(s, sizeof(s), "%s_flex_parsers_list", sh->ibdev_name);
523 	sh->flex_parsers_dv = mlx5_list_create(s, sh, false,
524 					       mlx5_flex_parser_create_cb,
525 					       mlx5_flex_parser_match_cb,
526 					       mlx5_flex_parser_remove_cb,
527 					       mlx5_flex_parser_clone_cb,
528 					       mlx5_flex_parser_clone_free_cb);
529 	if (!sh->flex_parsers_dv)
530 		goto error;
531 #endif
532 #ifdef HAVE_MLX5DV_DR
533 	void *domain;
534 
535 	/* Reference counter is zero, we should initialize structures. */
536 	domain = mlx5_glue->dr_create_domain(sh->cdev->ctx,
537 					     MLX5DV_DR_DOMAIN_TYPE_NIC_RX);
538 	if (!domain) {
539 		DRV_LOG(ERR, "ingress mlx5dv_dr_create_domain failed");
540 		err = errno;
541 		goto error;
542 	}
543 	sh->rx_domain = domain;
544 	domain = mlx5_glue->dr_create_domain(sh->cdev->ctx,
545 					     MLX5DV_DR_DOMAIN_TYPE_NIC_TX);
546 	if (!domain) {
547 		DRV_LOG(ERR, "egress mlx5dv_dr_create_domain failed");
548 		err = errno;
549 		goto error;
550 	}
551 	sh->tx_domain = domain;
552 #ifdef HAVE_MLX5DV_DR_ESWITCH
553 	if (sh->config.dv_esw_en) {
554 		domain = mlx5_glue->dr_create_domain(sh->cdev->ctx,
555 						     MLX5DV_DR_DOMAIN_TYPE_FDB);
556 		if (!domain) {
557 			DRV_LOG(ERR, "FDB mlx5dv_dr_create_domain failed");
558 			err = errno;
559 			goto error;
560 		}
561 		sh->fdb_domain = domain;
562 	}
563 	/*
564 	 * The drop action is just some dummy placeholder in rdma-core. It
565 	 * does not belong to domains and has no any attributes, and, can be
566 	 * shared by the entire device.
567 	 */
568 	sh->dr_drop_action = mlx5_glue->dr_create_flow_action_drop();
569 	if (!sh->dr_drop_action) {
570 		DRV_LOG(ERR, "FDB mlx5dv_dr_create_flow_action_drop");
571 		err = errno;
572 		goto error;
573 	}
574 #endif
575 	if (!sh->tunnel_hub && sh->config.dv_miss_info)
576 		err = mlx5_alloc_tunnel_hub(sh);
577 	if (err) {
578 		DRV_LOG(ERR, "mlx5_alloc_tunnel_hub failed err=%d", err);
579 		goto error;
580 	}
581 	if (sh->config.reclaim_mode == MLX5_RCM_AGGR) {
582 		mlx5_glue->dr_reclaim_domain_memory(sh->rx_domain, 1);
583 		mlx5_glue->dr_reclaim_domain_memory(sh->tx_domain, 1);
584 		if (sh->fdb_domain)
585 			mlx5_glue->dr_reclaim_domain_memory(sh->fdb_domain, 1);
586 	}
587 	sh->pop_vlan_action = mlx5_glue->dr_create_flow_action_pop_vlan();
588 	if (!sh->config.allow_duplicate_pattern) {
589 #ifndef HAVE_MLX5_DR_ALLOW_DUPLICATE
590 		DRV_LOG(WARNING, "Disallow duplicate pattern is not supported - maybe old rdma-core version?");
591 #endif
592 		mlx5_glue->dr_allow_duplicate_rules(sh->rx_domain, 0);
593 		mlx5_glue->dr_allow_duplicate_rules(sh->tx_domain, 0);
594 		if (sh->fdb_domain)
595 			mlx5_glue->dr_allow_duplicate_rules(sh->fdb_domain, 0);
596 	}
597 
598 	__mlx5_discovery_misc5_cap(priv);
599 #endif /* HAVE_MLX5DV_DR */
600 	sh->default_miss_action =
601 			mlx5_glue->dr_create_flow_action_default_miss();
602 	if (!sh->default_miss_action)
603 		DRV_LOG(WARNING, "Default miss action is not supported.");
604 	LIST_INIT(&sh->shared_rxqs);
605 	return 0;
606 error:
607 	/* Rollback the created objects. */
608 	if (sh->rx_domain) {
609 		mlx5_glue->dr_destroy_domain(sh->rx_domain);
610 		sh->rx_domain = NULL;
611 	}
612 	if (sh->tx_domain) {
613 		mlx5_glue->dr_destroy_domain(sh->tx_domain);
614 		sh->tx_domain = NULL;
615 	}
616 	if (sh->fdb_domain) {
617 		mlx5_glue->dr_destroy_domain(sh->fdb_domain);
618 		sh->fdb_domain = NULL;
619 	}
620 	if (sh->dr_drop_action) {
621 		mlx5_glue->destroy_flow_action(sh->dr_drop_action);
622 		sh->dr_drop_action = NULL;
623 	}
624 	if (sh->pop_vlan_action) {
625 		mlx5_glue->destroy_flow_action(sh->pop_vlan_action);
626 		sh->pop_vlan_action = NULL;
627 	}
628 	if (sh->encaps_decaps) {
629 		mlx5_hlist_destroy(sh->encaps_decaps);
630 		sh->encaps_decaps = NULL;
631 	}
632 	if (sh->modify_cmds) {
633 		mlx5_hlist_destroy(sh->modify_cmds);
634 		sh->modify_cmds = NULL;
635 	}
636 	if (sh->tag_table) {
637 		/* tags should be destroyed with flow before. */
638 		mlx5_hlist_destroy(sh->tag_table);
639 		sh->tag_table = NULL;
640 	}
641 	if (sh->tunnel_hub) {
642 		mlx5_release_tunnel_hub(sh, priv->dev_port);
643 		sh->tunnel_hub = NULL;
644 	}
645 	mlx5_free_table_hash_list(priv);
646 	if (sh->port_id_action_list) {
647 		mlx5_list_destroy(sh->port_id_action_list);
648 		sh->port_id_action_list = NULL;
649 	}
650 	if (sh->push_vlan_action_list) {
651 		mlx5_list_destroy(sh->push_vlan_action_list);
652 		sh->push_vlan_action_list = NULL;
653 	}
654 	if (sh->sample_action_list) {
655 		mlx5_list_destroy(sh->sample_action_list);
656 		sh->sample_action_list = NULL;
657 	}
658 	if (sh->dest_array_list) {
659 		mlx5_list_destroy(sh->dest_array_list);
660 		sh->dest_array_list = NULL;
661 	}
662 	return err;
663 }
664 
665 /**
666  * Destroy DR related data within private structure.
667  *
668  * @param[in] priv
669  *   Pointer to the private device data structure.
670  */
671 void
672 mlx5_os_free_shared_dr(struct mlx5_priv *priv)
673 {
674 	struct mlx5_dev_ctx_shared *sh = priv->sh;
675 
676 	MLX5_ASSERT(sh && sh->refcnt);
677 	if (sh->refcnt > 1)
678 		return;
679 	MLX5_ASSERT(LIST_EMPTY(&sh->shared_rxqs));
680 #ifdef HAVE_MLX5DV_DR
681 	if (sh->rx_domain) {
682 		mlx5_glue->dr_destroy_domain(sh->rx_domain);
683 		sh->rx_domain = NULL;
684 	}
685 	if (sh->tx_domain) {
686 		mlx5_glue->dr_destroy_domain(sh->tx_domain);
687 		sh->tx_domain = NULL;
688 	}
689 #ifdef HAVE_MLX5DV_DR_ESWITCH
690 	if (sh->fdb_domain) {
691 		mlx5_glue->dr_destroy_domain(sh->fdb_domain);
692 		sh->fdb_domain = NULL;
693 	}
694 	if (sh->dr_drop_action) {
695 		mlx5_glue->destroy_flow_action(sh->dr_drop_action);
696 		sh->dr_drop_action = NULL;
697 	}
698 #endif
699 	if (sh->pop_vlan_action) {
700 		mlx5_glue->destroy_flow_action(sh->pop_vlan_action);
701 		sh->pop_vlan_action = NULL;
702 	}
703 	if (sh->send_to_kernel_action.action) {
704 		void *action = sh->send_to_kernel_action.action;
705 
706 		mlx5_glue->destroy_flow_action(action);
707 		sh->send_to_kernel_action.action = NULL;
708 	}
709 	if (sh->send_to_kernel_action.tbl) {
710 		struct mlx5_flow_tbl_resource *tbl =
711 				sh->send_to_kernel_action.tbl;
712 
713 		flow_dv_tbl_resource_release(sh, tbl);
714 		sh->send_to_kernel_action.tbl = NULL;
715 	}
716 #endif /* HAVE_MLX5DV_DR */
717 	if (sh->default_miss_action)
718 		mlx5_glue->destroy_flow_action
719 				(sh->default_miss_action);
720 	if (sh->encaps_decaps) {
721 		mlx5_hlist_destroy(sh->encaps_decaps);
722 		sh->encaps_decaps = NULL;
723 	}
724 	if (sh->modify_cmds) {
725 		mlx5_hlist_destroy(sh->modify_cmds);
726 		sh->modify_cmds = NULL;
727 	}
728 	if (sh->tag_table) {
729 		/* tags should be destroyed with flow before. */
730 		mlx5_hlist_destroy(sh->tag_table);
731 		sh->tag_table = NULL;
732 	}
733 	if (sh->tunnel_hub) {
734 		mlx5_release_tunnel_hub(sh, priv->dev_port);
735 		sh->tunnel_hub = NULL;
736 	}
737 	mlx5_free_table_hash_list(priv);
738 	if (sh->port_id_action_list) {
739 		mlx5_list_destroy(sh->port_id_action_list);
740 		sh->port_id_action_list = NULL;
741 	}
742 	if (sh->push_vlan_action_list) {
743 		mlx5_list_destroy(sh->push_vlan_action_list);
744 		sh->push_vlan_action_list = NULL;
745 	}
746 	if (sh->sample_action_list) {
747 		mlx5_list_destroy(sh->sample_action_list);
748 		sh->sample_action_list = NULL;
749 	}
750 	if (sh->dest_array_list) {
751 		mlx5_list_destroy(sh->dest_array_list);
752 		sh->dest_array_list = NULL;
753 	}
754 }
755 
756 /**
757  * Initialize shared data between primary and secondary process.
758  *
759  * A memzone is reserved by primary process and secondary processes attach to
760  * the memzone.
761  *
762  * @return
763  *   0 on success, a negative errno value otherwise and rte_errno is set.
764  */
765 static int
766 mlx5_init_shared_data(void)
767 {
768 	const struct rte_memzone *mz;
769 	int ret = 0;
770 
771 	rte_spinlock_lock(&mlx5_shared_data_lock);
772 	if (mlx5_shared_data == NULL) {
773 		if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
774 			/* Allocate shared memory. */
775 			mz = rte_memzone_reserve(MZ_MLX5_PMD_SHARED_DATA,
776 						 sizeof(*mlx5_shared_data),
777 						 SOCKET_ID_ANY, 0);
778 			if (mz == NULL) {
779 				DRV_LOG(ERR,
780 					"Cannot allocate mlx5 shared data");
781 				ret = -rte_errno;
782 				goto error;
783 			}
784 			mlx5_shared_data = mz->addr;
785 			memset(mlx5_shared_data, 0, sizeof(*mlx5_shared_data));
786 			rte_spinlock_init(&mlx5_shared_data->lock);
787 		} else {
788 			/* Lookup allocated shared memory. */
789 			mz = rte_memzone_lookup(MZ_MLX5_PMD_SHARED_DATA);
790 			if (mz == NULL) {
791 				DRV_LOG(ERR,
792 					"Cannot attach mlx5 shared data");
793 				ret = -rte_errno;
794 				goto error;
795 			}
796 			mlx5_shared_data = mz->addr;
797 			memset(&mlx5_local_data, 0, sizeof(mlx5_local_data));
798 		}
799 	}
800 error:
801 	rte_spinlock_unlock(&mlx5_shared_data_lock);
802 	return ret;
803 }
804 
805 /**
806  * PMD global initialization.
807  *
808  * Independent from individual device, this function initializes global
809  * per-PMD data structures distinguishing primary and secondary processes.
810  * Hence, each initialization is called once per a process.
811  *
812  * @return
813  *   0 on success, a negative errno value otherwise and rte_errno is set.
814  */
815 static int
816 mlx5_init_once(void)
817 {
818 	struct mlx5_shared_data *sd;
819 	struct mlx5_local_data *ld = &mlx5_local_data;
820 	int ret = 0;
821 
822 	if (mlx5_init_shared_data())
823 		return -rte_errno;
824 	sd = mlx5_shared_data;
825 	MLX5_ASSERT(sd);
826 	rte_spinlock_lock(&sd->lock);
827 	switch (rte_eal_process_type()) {
828 	case RTE_PROC_PRIMARY:
829 		if (sd->init_done)
830 			break;
831 		ret = mlx5_mp_init_primary(MLX5_MP_NAME,
832 					   mlx5_mp_os_primary_handle);
833 		if (ret)
834 			goto out;
835 		sd->init_done = true;
836 		break;
837 	case RTE_PROC_SECONDARY:
838 		if (ld->init_done)
839 			break;
840 		ret = mlx5_mp_init_secondary(MLX5_MP_NAME,
841 					     mlx5_mp_os_secondary_handle);
842 		if (ret)
843 			goto out;
844 		++sd->secondary_cnt;
845 		ld->init_done = true;
846 		break;
847 	default:
848 		break;
849 	}
850 out:
851 	rte_spinlock_unlock(&sd->lock);
852 	return ret;
853 }
854 
855 /**
856  * DR flow drop action support detect.
857  *
858  * @param dev
859  *   Pointer to rte_eth_dev structure.
860  *
861  */
862 static void
863 mlx5_flow_drop_action_config(struct rte_eth_dev *dev __rte_unused)
864 {
865 #ifdef HAVE_MLX5DV_DR
866 	struct mlx5_priv *priv = dev->data->dev_private;
867 
868 	if (!priv->sh->config.dv_flow_en || !priv->sh->dr_drop_action)
869 		return;
870 	/**
871 	 * DR supports drop action placeholder when it is supported;
872 	 * otherwise, use the queue drop action.
873 	 */
874 	if (!priv->sh->drop_action_check_flag) {
875 		if (!mlx5_flow_discover_dr_action_support(dev))
876 			priv->sh->dr_drop_action_en = 1;
877 		priv->sh->drop_action_check_flag = 1;
878 	}
879 	if (priv->sh->dr_drop_action_en)
880 		priv->root_drop_action = priv->sh->dr_drop_action;
881 	else
882 		priv->root_drop_action = priv->drop_queue.hrxq->action;
883 #endif
884 }
885 
886 static void
887 mlx5_queue_counter_id_prepare(struct rte_eth_dev *dev)
888 {
889 	struct mlx5_priv *priv = dev->data->dev_private;
890 	void *ctx = priv->sh->cdev->ctx;
891 
892 	priv->q_counters = mlx5_devx_cmd_queue_counter_alloc(ctx);
893 	if (!priv->q_counters) {
894 		struct ibv_cq *cq = mlx5_glue->create_cq(ctx, 1, NULL, NULL, 0);
895 		struct ibv_wq *wq;
896 
897 		DRV_LOG(DEBUG, "Port %d queue counter object cannot be created "
898 			"by DevX - fall-back to use the kernel driver global "
899 			"queue counter.", dev->data->port_id);
900 		/* Create WQ by kernel and query its queue counter ID. */
901 		if (cq) {
902 			wq = mlx5_glue->create_wq(ctx,
903 						  &(struct ibv_wq_init_attr){
904 						    .wq_type = IBV_WQT_RQ,
905 						    .max_wr = 1,
906 						    .max_sge = 1,
907 						    .pd = priv->sh->cdev->pd,
908 						    .cq = cq,
909 						});
910 			if (wq) {
911 				/* Counter is assigned only on RDY state. */
912 				int ret = mlx5_glue->modify_wq(wq,
913 						 &(struct ibv_wq_attr){
914 						 .attr_mask = IBV_WQ_ATTR_STATE,
915 						 .wq_state = IBV_WQS_RDY,
916 						});
917 
918 				if (ret == 0)
919 					mlx5_devx_cmd_wq_query(wq,
920 							 &priv->counter_set_id);
921 				claim_zero(mlx5_glue->destroy_wq(wq));
922 			}
923 			claim_zero(mlx5_glue->destroy_cq(cq));
924 		}
925 	} else {
926 		priv->counter_set_id = priv->q_counters->id;
927 	}
928 	if (priv->counter_set_id == 0)
929 		DRV_LOG(INFO, "Part of the port %d statistics will not be "
930 			"available.", dev->data->port_id);
931 }
932 
933 /**
934  * Check if representor spawn info match devargs.
935  *
936  * @param spawn
937  *   Verbs device parameters (name, port, switch_info) to spawn.
938  * @param eth_da
939  *   Device devargs to probe.
940  *
941  * @return
942  *   Match result.
943  */
944 static bool
945 mlx5_representor_match(struct mlx5_dev_spawn_data *spawn,
946 		       struct rte_eth_devargs *eth_da)
947 {
948 	struct mlx5_switch_info *switch_info = &spawn->info;
949 	unsigned int p, f;
950 	uint16_t id;
951 	uint16_t repr_id = mlx5_representor_id_encode(switch_info,
952 						      eth_da->type);
953 
954 	switch (eth_da->type) {
955 	case RTE_ETH_REPRESENTOR_SF:
956 		if (!(spawn->info.port_name == -1 &&
957 		      switch_info->name_type ==
958 				MLX5_PHYS_PORT_NAME_TYPE_PFHPF) &&
959 		    switch_info->name_type != MLX5_PHYS_PORT_NAME_TYPE_PFSF) {
960 			rte_errno = EBUSY;
961 			return false;
962 		}
963 		break;
964 	case RTE_ETH_REPRESENTOR_VF:
965 		/* Allows HPF representor index -1 as exception. */
966 		if (!(spawn->info.port_name == -1 &&
967 		      switch_info->name_type ==
968 				MLX5_PHYS_PORT_NAME_TYPE_PFHPF) &&
969 		    switch_info->name_type != MLX5_PHYS_PORT_NAME_TYPE_PFVF) {
970 			rte_errno = EBUSY;
971 			return false;
972 		}
973 		break;
974 	case RTE_ETH_REPRESENTOR_NONE:
975 		rte_errno = EBUSY;
976 		return false;
977 	default:
978 		rte_errno = ENOTSUP;
979 		DRV_LOG(ERR, "unsupported representor type");
980 		return false;
981 	}
982 	/* Check representor ID: */
983 	for (p = 0; p < eth_da->nb_ports; ++p) {
984 		if (spawn->pf_bond < 0) {
985 			/* For non-LAG mode, allow and ignore pf. */
986 			switch_info->pf_num = eth_da->ports[p];
987 			repr_id = mlx5_representor_id_encode(switch_info,
988 							     eth_da->type);
989 		}
990 		for (f = 0; f < eth_da->nb_representor_ports; ++f) {
991 			id = MLX5_REPRESENTOR_ID
992 				(eth_da->ports[p], eth_da->type,
993 				 eth_da->representor_ports[f]);
994 			if (repr_id == id)
995 				return true;
996 		}
997 	}
998 	rte_errno = EBUSY;
999 	return false;
1000 }
1001 
1002 /**
1003  * Spawn an Ethernet device from Verbs information.
1004  *
1005  * @param dpdk_dev
1006  *   Backing DPDK device.
1007  * @param spawn
1008  *   Verbs device parameters (name, port, switch_info) to spawn.
1009  * @param eth_da
1010  *   Device arguments.
1011  * @param mkvlist
1012  *   Pointer to mlx5 kvargs control, can be NULL if there is no devargs.
1013  *
1014  * @return
1015  *   A valid Ethernet device object on success, NULL otherwise and rte_errno
1016  *   is set. The following errors are defined:
1017  *
1018  *   EBUSY: device is not supposed to be spawned.
1019  *   EEXIST: device is already spawned
1020  */
1021 static struct rte_eth_dev *
1022 mlx5_dev_spawn(struct rte_device *dpdk_dev,
1023 	       struct mlx5_dev_spawn_data *spawn,
1024 	       struct rte_eth_devargs *eth_da,
1025 	       struct mlx5_kvargs_ctrl *mkvlist)
1026 {
1027 	const struct mlx5_switch_info *switch_info = &spawn->info;
1028 	struct mlx5_dev_ctx_shared *sh = NULL;
1029 	struct ibv_port_attr port_attr = { .state = IBV_PORT_NOP };
1030 	struct rte_eth_dev *eth_dev = NULL;
1031 	struct mlx5_priv *priv = NULL;
1032 	int err = 0;
1033 	struct rte_ether_addr mac;
1034 	char name[RTE_ETH_NAME_MAX_LEN];
1035 	int own_domain_id = 0;
1036 	uint16_t port_id;
1037 	struct mlx5_port_info vport_info = { .query_flags = 0 };
1038 	int nl_rdma;
1039 	int i;
1040 
1041 	/* Determine if this port representor is supposed to be spawned. */
1042 	if (switch_info->representor && dpdk_dev->devargs &&
1043 	    !mlx5_representor_match(spawn, eth_da))
1044 		return NULL;
1045 	/* Build device name. */
1046 	if (spawn->pf_bond < 0) {
1047 		/* Single device. */
1048 		if (!switch_info->representor)
1049 			strlcpy(name, dpdk_dev->name, sizeof(name));
1050 		else
1051 			err = snprintf(name, sizeof(name), "%s_representor_%s%u",
1052 				 dpdk_dev->name,
1053 				 switch_info->name_type ==
1054 				 MLX5_PHYS_PORT_NAME_TYPE_PFSF ? "sf" : "vf",
1055 				 switch_info->port_name);
1056 	} else {
1057 		/* Bonding device. */
1058 		if (!switch_info->representor) {
1059 			err = snprintf(name, sizeof(name), "%s_%s",
1060 				       dpdk_dev->name, spawn->phys_dev_name);
1061 		} else {
1062 			err = snprintf(name, sizeof(name), "%s_%s_representor_c%dpf%d%s%u",
1063 				dpdk_dev->name, spawn->phys_dev_name,
1064 				switch_info->ctrl_num,
1065 				switch_info->pf_num,
1066 				switch_info->name_type ==
1067 				MLX5_PHYS_PORT_NAME_TYPE_PFSF ? "sf" : "vf",
1068 				switch_info->port_name);
1069 		}
1070 	}
1071 	if (err >= (int)sizeof(name))
1072 		DRV_LOG(WARNING, "device name overflow %s", name);
1073 	/* check if the device is already spawned */
1074 	if (rte_eth_dev_get_port_by_name(name, &port_id) == 0) {
1075 		/*
1076 		 * When device is already spawned, its devargs should be set
1077 		 * as used. otherwise, mlx5_kvargs_validate() will fail.
1078 		 */
1079 		if (mkvlist)
1080 			mlx5_port_args_set_used(name, port_id, mkvlist);
1081 		rte_errno = EEXIST;
1082 		return NULL;
1083 	}
1084 	DRV_LOG(DEBUG, "naming Ethernet device \"%s\"", name);
1085 	if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1086 		struct mlx5_mp_id mp_id;
1087 		int fd;
1088 
1089 		eth_dev = rte_eth_dev_attach_secondary(name);
1090 		if (eth_dev == NULL) {
1091 			DRV_LOG(ERR, "can not attach rte ethdev");
1092 			rte_errno = ENOMEM;
1093 			return NULL;
1094 		}
1095 		eth_dev->device = dpdk_dev;
1096 		eth_dev->dev_ops = &mlx5_dev_sec_ops;
1097 		eth_dev->rx_descriptor_status = mlx5_rx_descriptor_status;
1098 		eth_dev->tx_descriptor_status = mlx5_tx_descriptor_status;
1099 		err = mlx5_proc_priv_init(eth_dev);
1100 		if (err)
1101 			return NULL;
1102 		mlx5_mp_id_init(&mp_id, eth_dev->data->port_id);
1103 		/* Receive command fd from primary process */
1104 		fd = mlx5_mp_req_verbs_cmd_fd(&mp_id);
1105 		if (fd < 0)
1106 			goto err_secondary;
1107 		/* Remap UAR for Tx queues. */
1108 		err = mlx5_tx_uar_init_secondary(eth_dev, fd);
1109 		close(fd);
1110 		if (err)
1111 			goto err_secondary;
1112 		/*
1113 		 * Ethdev pointer is still required as input since
1114 		 * the primary device is not accessible from the
1115 		 * secondary process.
1116 		 */
1117 		eth_dev->rx_pkt_burst = mlx5_select_rx_function(eth_dev);
1118 		eth_dev->tx_pkt_burst = mlx5_select_tx_function(eth_dev);
1119 		return eth_dev;
1120 err_secondary:
1121 		mlx5_dev_close(eth_dev);
1122 		return NULL;
1123 	}
1124 	sh = mlx5_alloc_shared_dev_ctx(spawn, mkvlist);
1125 	if (!sh)
1126 		return NULL;
1127 	nl_rdma = mlx5_nl_init(NETLINK_RDMA, 0);
1128 	/* Check port status. */
1129 	if (spawn->phys_port <= UINT8_MAX) {
1130 		/* Legacy Verbs api only support u8 port number. */
1131 		err = mlx5_glue->query_port(sh->cdev->ctx, spawn->phys_port,
1132 					    &port_attr);
1133 		if (err) {
1134 			DRV_LOG(ERR, "port query failed: %s", strerror(err));
1135 			goto error;
1136 		}
1137 		if (port_attr.link_layer != IBV_LINK_LAYER_ETHERNET) {
1138 			DRV_LOG(ERR, "port is not configured in Ethernet mode");
1139 			err = EINVAL;
1140 			goto error;
1141 		}
1142 	} else if (nl_rdma >= 0) {
1143 		/* IB doesn't allow more than 255 ports, must be Ethernet. */
1144 		err = mlx5_nl_port_state(nl_rdma,
1145 			spawn->phys_dev_name,
1146 			spawn->phys_port);
1147 		if (err < 0) {
1148 			DRV_LOG(INFO, "Failed to get netlink port state: %s",
1149 				strerror(rte_errno));
1150 			err = -rte_errno;
1151 			goto error;
1152 		}
1153 		port_attr.state = (enum ibv_port_state)err;
1154 	}
1155 	if (port_attr.state != IBV_PORT_ACTIVE)
1156 		DRV_LOG(INFO, "port is not active: \"%s\" (%d)",
1157 			mlx5_glue->port_state_str(port_attr.state),
1158 			port_attr.state);
1159 	/* Allocate private eth device data. */
1160 	priv = mlx5_malloc(MLX5_MEM_ZERO | MLX5_MEM_RTE,
1161 			   sizeof(*priv),
1162 			   RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
1163 	if (priv == NULL) {
1164 		DRV_LOG(ERR, "priv allocation failure");
1165 		err = ENOMEM;
1166 		goto error;
1167 	}
1168 	/*
1169 	 * When user configures remote PD and CTX and device creates RxQ by
1170 	 * DevX, external RxQ is both supported and requested.
1171 	 */
1172 	if (mlx5_imported_pd_and_ctx(sh->cdev) && mlx5_devx_obj_ops_en(sh)) {
1173 		priv->ext_rxqs = mlx5_malloc(MLX5_MEM_ZERO | MLX5_MEM_RTE,
1174 					     sizeof(struct mlx5_external_rxq) *
1175 					     MLX5_MAX_EXT_RX_QUEUES, 0,
1176 					     SOCKET_ID_ANY);
1177 		if (priv->ext_rxqs == NULL) {
1178 			DRV_LOG(ERR, "Fail to allocate external RxQ array.");
1179 			err = ENOMEM;
1180 			goto error;
1181 		}
1182 		DRV_LOG(DEBUG, "External RxQ is supported.");
1183 	}
1184 	priv->sh = sh;
1185 	priv->dev_port = spawn->phys_port;
1186 	priv->pci_dev = spawn->pci_dev;
1187 	priv->mtu = RTE_ETHER_MTU;
1188 	/* Some internal functions rely on Netlink sockets, open them now. */
1189 	priv->nl_socket_rdma = nl_rdma;
1190 	priv->nl_socket_route =	mlx5_nl_init(NETLINK_ROUTE, 0);
1191 	priv->representor = !!switch_info->representor;
1192 	priv->master = !!switch_info->master;
1193 	priv->domain_id = RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID;
1194 	priv->vport_meta_tag = 0;
1195 	priv->vport_meta_mask = 0;
1196 	priv->pf_bond = spawn->pf_bond;
1197 
1198 	DRV_LOG(DEBUG,
1199 		"dev_port=%u bus=%s pci=%s master=%d representor=%d pf_bond=%d\n",
1200 		priv->dev_port, dpdk_dev->bus->name,
1201 		priv->pci_dev ? priv->pci_dev->name : "NONE",
1202 		priv->master, priv->representor, priv->pf_bond);
1203 
1204 	/*
1205 	 * If we have E-Switch we should determine the vport attributes.
1206 	 * E-Switch may use either source vport field or reg_c[0] metadata
1207 	 * register to match on vport index. The engaged part of metadata
1208 	 * register is defined by mask.
1209 	 */
1210 	if (sh->esw_mode) {
1211 		err = mlx5_glue->devx_port_query(sh->cdev->ctx,
1212 						 spawn->phys_port,
1213 						 &vport_info);
1214 		if (err) {
1215 			DRV_LOG(WARNING,
1216 				"Cannot query devx port %d on device %s",
1217 				spawn->phys_port, spawn->phys_dev_name);
1218 			vport_info.query_flags = 0;
1219 		}
1220 	}
1221 	if (vport_info.query_flags & MLX5_PORT_QUERY_REG_C0) {
1222 		priv->vport_meta_tag = vport_info.vport_meta_tag;
1223 		priv->vport_meta_mask = vport_info.vport_meta_mask;
1224 		if (!priv->vport_meta_mask) {
1225 			DRV_LOG(ERR,
1226 				"vport zero mask for port %d on bonding device %s",
1227 				spawn->phys_port, spawn->phys_dev_name);
1228 			err = ENOTSUP;
1229 			goto error;
1230 		}
1231 		if (priv->vport_meta_tag & ~priv->vport_meta_mask) {
1232 			DRV_LOG(ERR,
1233 				"Invalid vport tag for port %d on bonding device %s",
1234 				spawn->phys_port, spawn->phys_dev_name);
1235 			err = ENOTSUP;
1236 			goto error;
1237 		}
1238 	}
1239 	if (vport_info.query_flags & MLX5_PORT_QUERY_VPORT) {
1240 		priv->vport_id = vport_info.vport_id;
1241 	} else if (spawn->pf_bond >= 0 && sh->esw_mode) {
1242 		DRV_LOG(ERR,
1243 			"Cannot deduce vport index for port %d on bonding device %s",
1244 			spawn->phys_port, spawn->phys_dev_name);
1245 		err = ENOTSUP;
1246 		goto error;
1247 	} else {
1248 		/*
1249 		 * Suppose vport index in compatible way. Kernel/rdma_core
1250 		 * support single E-Switch per PF configurations only and
1251 		 * vport_id field contains the vport index for associated VF,
1252 		 * which is deduced from representor port name.
1253 		 * For example, let's have the IB device port 10, it has
1254 		 * attached network device eth0, which has port name attribute
1255 		 * pf0vf2, we can deduce the VF number as 2, and set vport index
1256 		 * as 3 (2+1). This assigning schema should be changed if the
1257 		 * multiple E-Switch instances per PF configurations or/and PCI
1258 		 * subfunctions are added.
1259 		 */
1260 		priv->vport_id = switch_info->representor ?
1261 				 switch_info->port_name + 1 : -1;
1262 	}
1263 	priv->representor_id = mlx5_representor_id_encode(switch_info,
1264 							  eth_da->type);
1265 	/*
1266 	 * Look for sibling devices in order to reuse their switch domain
1267 	 * if any, otherwise allocate one.
1268 	 */
1269 	MLX5_ETH_FOREACH_DEV(port_id, dpdk_dev) {
1270 		const struct mlx5_priv *opriv =
1271 			rte_eth_devices[port_id].data->dev_private;
1272 
1273 		if (!opriv ||
1274 		    opriv->sh != priv->sh ||
1275 			opriv->domain_id ==
1276 			RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID)
1277 			continue;
1278 		priv->domain_id = opriv->domain_id;
1279 		DRV_LOG(DEBUG, "dev_port-%u inherit domain_id=%u\n",
1280 			priv->dev_port, priv->domain_id);
1281 		break;
1282 	}
1283 	if (priv->domain_id == RTE_ETH_DEV_SWITCH_DOMAIN_ID_INVALID) {
1284 		err = rte_eth_switch_domain_alloc(&priv->domain_id);
1285 		if (err) {
1286 			err = rte_errno;
1287 			DRV_LOG(ERR, "unable to allocate switch domain: %s",
1288 				strerror(rte_errno));
1289 			goto error;
1290 		}
1291 		own_domain_id = 1;
1292 		DRV_LOG(DEBUG, "dev_port-%u new domain_id=%u\n",
1293 			priv->dev_port, priv->domain_id);
1294 	}
1295 	if (sh->cdev->config.devx) {
1296 		struct mlx5_hca_attr *hca_attr = &sh->cdev->config.hca_attr;
1297 
1298 		sh->steering_format_version = hca_attr->steering_format_version;
1299 #if defined(HAVE_MLX5DV_DR) && \
1300 	(defined(HAVE_MLX5_DR_CREATE_ACTION_FLOW_METER) || \
1301 	 defined(HAVE_MLX5_DR_CREATE_ACTION_ASO))
1302 		if (hca_attr->qos.sup && hca_attr->qos.flow_meter_old &&
1303 		    sh->config.dv_flow_en) {
1304 			uint8_t reg_c_mask = hca_attr->qos.flow_meter_reg_c_ids;
1305 			/*
1306 			 * Meter needs two REG_C's for color match and pre-sfx
1307 			 * flow match. Here get the REG_C for color match.
1308 			 * REG_C_0 and REG_C_1 is reserved for metadata feature.
1309 			 */
1310 			reg_c_mask &= 0xfc;
1311 			if (__builtin_popcount(reg_c_mask) < 1) {
1312 				priv->mtr_en = 0;
1313 				DRV_LOG(WARNING, "No available register for"
1314 					" meter.");
1315 			} else {
1316 				/*
1317 				 * The meter color register is used by the
1318 				 * flow-hit feature as well.
1319 				 * The flow-hit feature must use REG_C_3
1320 				 * Prefer REG_C_3 if it is available.
1321 				 */
1322 				if (reg_c_mask & (1 << (REG_C_3 - REG_C_0)))
1323 					priv->mtr_color_reg = REG_C_3;
1324 				else
1325 					priv->mtr_color_reg = ffs(reg_c_mask)
1326 							      - 1 + REG_C_0;
1327 				priv->mtr_en = 1;
1328 				priv->mtr_reg_share = hca_attr->qos.flow_meter;
1329 				DRV_LOG(DEBUG, "The REG_C meter uses is %d",
1330 					priv->mtr_color_reg);
1331 			}
1332 		}
1333 		if (hca_attr->qos.sup && hca_attr->qos.flow_meter_aso_sup) {
1334 			uint32_t log_obj_size =
1335 				rte_log2_u32(MLX5_ASO_MTRS_PER_POOL >> 1);
1336 			if (log_obj_size >=
1337 			    hca_attr->qos.log_meter_aso_granularity &&
1338 			    log_obj_size <=
1339 			    hca_attr->qos.log_meter_aso_max_alloc)
1340 				sh->meter_aso_en = 1;
1341 		}
1342 		if (priv->mtr_en) {
1343 			err = mlx5_aso_flow_mtrs_mng_init(priv->sh);
1344 			if (err) {
1345 				err = -err;
1346 				goto error;
1347 			}
1348 		}
1349 		if (hca_attr->flow.tunnel_header_0_1)
1350 			sh->tunnel_header_0_1 = 1;
1351 		if (hca_attr->flow.tunnel_header_2_3)
1352 			sh->tunnel_header_2_3 = 1;
1353 #endif
1354 #ifdef HAVE_MLX5_DR_CREATE_ACTION_ASO
1355 		if (hca_attr->flow_hit_aso && priv->mtr_color_reg == REG_C_3) {
1356 			sh->flow_hit_aso_en = 1;
1357 			err = mlx5_flow_aso_age_mng_init(sh);
1358 			if (err) {
1359 				err = -err;
1360 				goto error;
1361 			}
1362 			DRV_LOG(DEBUG, "Flow Hit ASO is supported.");
1363 		}
1364 #endif /* HAVE_MLX5_DR_CREATE_ACTION_ASO */
1365 #if defined (HAVE_MLX5_DR_CREATE_ACTION_ASO) && \
1366     defined (HAVE_MLX5_DR_ACTION_ASO_CT)
1367 		/* HWS create CT ASO SQ based on HWS configure queue number. */
1368 		if (sh->config.dv_flow_en != 2 &&
1369 		    hca_attr->ct_offload && priv->mtr_color_reg == REG_C_3) {
1370 			err = mlx5_flow_aso_ct_mng_init(sh);
1371 			if (err) {
1372 				err = -err;
1373 				goto error;
1374 			}
1375 			DRV_LOG(DEBUG, "CT ASO is supported.");
1376 			sh->ct_aso_en = 1;
1377 		}
1378 #endif /* HAVE_MLX5_DR_CREATE_ACTION_ASO && HAVE_MLX5_DR_ACTION_ASO_CT */
1379 #if defined(HAVE_MLX5DV_DR) && defined(HAVE_MLX5_DR_CREATE_ACTION_FLOW_SAMPLE)
1380 		if (hca_attr->log_max_ft_sampler_num > 0  &&
1381 		    sh->config.dv_flow_en) {
1382 			priv->sampler_en = 1;
1383 			DRV_LOG(DEBUG, "Sampler enabled!");
1384 		} else {
1385 			priv->sampler_en = 0;
1386 			if (!hca_attr->log_max_ft_sampler_num)
1387 				DRV_LOG(WARNING,
1388 					"No available register for sampler.");
1389 			else
1390 				DRV_LOG(DEBUG, "DV flow is not supported!");
1391 		}
1392 #endif
1393 	}
1394 	/* Process parameters and store port configuration on priv structure. */
1395 	err = mlx5_port_args_config(priv, mkvlist, &priv->config);
1396 	if (err) {
1397 		err = rte_errno;
1398 		DRV_LOG(ERR, "Failed to process port configure: %s",
1399 			strerror(rte_errno));
1400 		goto error;
1401 	}
1402 	eth_dev = rte_eth_dev_allocate(name);
1403 	if (eth_dev == NULL) {
1404 		DRV_LOG(ERR, "can not allocate rte ethdev");
1405 		err = ENOMEM;
1406 		goto error;
1407 	}
1408 	if (priv->representor) {
1409 		eth_dev->data->dev_flags |= RTE_ETH_DEV_REPRESENTOR;
1410 		eth_dev->data->representor_id = priv->representor_id;
1411 		MLX5_ETH_FOREACH_DEV(port_id, dpdk_dev) {
1412 			struct mlx5_priv *opriv =
1413 				rte_eth_devices[port_id].data->dev_private;
1414 			if (opriv &&
1415 			    opriv->master &&
1416 			    opriv->domain_id == priv->domain_id &&
1417 			    opriv->sh == priv->sh) {
1418 				eth_dev->data->backer_port_id = port_id;
1419 				break;
1420 			}
1421 		}
1422 		if (port_id >= RTE_MAX_ETHPORTS)
1423 			eth_dev->data->backer_port_id = eth_dev->data->port_id;
1424 	}
1425 	priv->mp_id.port_id = eth_dev->data->port_id;
1426 	strlcpy(priv->mp_id.name, MLX5_MP_NAME, RTE_MP_MAX_NAME_LEN);
1427 	/*
1428 	 * Store associated network device interface index. This index
1429 	 * is permanent throughout the lifetime of device. So, we may store
1430 	 * the ifindex here and use the cached value further.
1431 	 */
1432 	MLX5_ASSERT(spawn->ifindex);
1433 	priv->if_index = spawn->ifindex;
1434 	priv->lag_affinity_idx = sh->refcnt - 1;
1435 	eth_dev->data->dev_private = priv;
1436 	priv->dev_data = eth_dev->data;
1437 	eth_dev->data->mac_addrs = priv->mac;
1438 	eth_dev->device = dpdk_dev;
1439 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1440 	/* Configure the first MAC address by default. */
1441 	if (mlx5_get_mac(eth_dev, &mac.addr_bytes)) {
1442 		DRV_LOG(ERR,
1443 			"port %u cannot get MAC address, is mlx5_en"
1444 			" loaded? (errno: %s)",
1445 			eth_dev->data->port_id, strerror(rte_errno));
1446 		err = ENODEV;
1447 		goto error;
1448 	}
1449 	DRV_LOG(INFO,
1450 		"port %u MAC address is " RTE_ETHER_ADDR_PRT_FMT,
1451 		eth_dev->data->port_id, RTE_ETHER_ADDR_BYTES(&mac));
1452 #ifdef RTE_LIBRTE_MLX5_DEBUG
1453 	{
1454 		char ifname[MLX5_NAMESIZE];
1455 
1456 		if (mlx5_get_ifname(eth_dev, &ifname) == 0)
1457 			DRV_LOG(DEBUG, "port %u ifname is \"%s\"",
1458 				eth_dev->data->port_id, ifname);
1459 		else
1460 			DRV_LOG(DEBUG, "port %u ifname is unknown",
1461 				eth_dev->data->port_id);
1462 	}
1463 #endif
1464 	/* Get actual MTU if possible. */
1465 	err = mlx5_get_mtu(eth_dev, &priv->mtu);
1466 	if (err) {
1467 		err = rte_errno;
1468 		goto error;
1469 	}
1470 	DRV_LOG(DEBUG, "port %u MTU is %u", eth_dev->data->port_id,
1471 		priv->mtu);
1472 	/* Initialize burst functions to prevent crashes before link-up. */
1473 	eth_dev->rx_pkt_burst = rte_eth_pkt_burst_dummy;
1474 	eth_dev->tx_pkt_burst = rte_eth_pkt_burst_dummy;
1475 	eth_dev->dev_ops = &mlx5_dev_ops;
1476 	eth_dev->rx_descriptor_status = mlx5_rx_descriptor_status;
1477 	eth_dev->tx_descriptor_status = mlx5_tx_descriptor_status;
1478 	eth_dev->rx_queue_count = mlx5_rx_queue_count;
1479 	/* Register MAC address. */
1480 	claim_zero(mlx5_mac_addr_add(eth_dev, &mac, 0, 0));
1481 	if (sh->dev_cap.vf && sh->config.vf_nl_en)
1482 		mlx5_nl_mac_addr_sync(priv->nl_socket_route,
1483 				      mlx5_ifindex(eth_dev),
1484 				      eth_dev->data->mac_addrs,
1485 				      MLX5_MAX_MAC_ADDRESSES);
1486 	priv->ctrl_flows = 0;
1487 	rte_spinlock_init(&priv->flow_list_lock);
1488 	TAILQ_INIT(&priv->flow_meters);
1489 	priv->mtr_profile_tbl = mlx5_l3t_create(MLX5_L3T_TYPE_PTR);
1490 	if (!priv->mtr_profile_tbl)
1491 		goto error;
1492 	/* Bring Ethernet device up. */
1493 	DRV_LOG(DEBUG, "port %u forcing Ethernet interface up",
1494 		eth_dev->data->port_id);
1495 	/* Read link status in case it is up and there will be no event. */
1496 	mlx5_link_update(eth_dev, 0);
1497 	/* Watch LSC interrupts between port probe and port start. */
1498 	priv->sh->port[priv->dev_port - 1].nl_ih_port_id =
1499 							eth_dev->data->port_id;
1500 	mlx5_set_link_up(eth_dev);
1501 	for (i = 0; i < MLX5_FLOW_TYPE_MAXI; i++) {
1502 		icfg[i].release_mem_en = !!sh->config.reclaim_mode;
1503 		if (sh->config.reclaim_mode)
1504 			icfg[i].per_core_cache = 0;
1505 		priv->flows[i] = mlx5_ipool_create(&icfg[i]);
1506 		if (!priv->flows[i])
1507 			goto error;
1508 	}
1509 	/* Create context for virtual machine VLAN workaround. */
1510 	priv->vmwa_context = mlx5_vlan_vmwa_init(eth_dev, spawn->ifindex);
1511 	if (sh->config.dv_flow_en) {
1512 		err = mlx5_alloc_shared_dr(priv);
1513 		if (err)
1514 			goto error;
1515 		if (mlx5_flex_item_port_init(eth_dev) < 0)
1516 			goto error;
1517 	}
1518 	if (mlx5_devx_obj_ops_en(sh)) {
1519 		priv->obj_ops = devx_obj_ops;
1520 		mlx5_queue_counter_id_prepare(eth_dev);
1521 		priv->obj_ops.lb_dummy_queue_create =
1522 					mlx5_rxq_ibv_obj_dummy_lb_create;
1523 		priv->obj_ops.lb_dummy_queue_release =
1524 					mlx5_rxq_ibv_obj_dummy_lb_release;
1525 	} else if (spawn->max_port > UINT8_MAX) {
1526 		/* Verbs can't support ports larger than 255 by design. */
1527 		DRV_LOG(ERR, "must enable DV and ESW when RDMA link ports > 255");
1528 		err = ENOTSUP;
1529 		goto error;
1530 	} else {
1531 		priv->obj_ops = ibv_obj_ops;
1532 	}
1533 	if (sh->config.tx_pp &&
1534 	    priv->obj_ops.txq_obj_new != mlx5_txq_devx_obj_new) {
1535 		/*
1536 		 * HAVE_MLX5DV_DEVX_UAR_OFFSET is required to support
1537 		 * packet pacing and already checked above.
1538 		 * Hence, we should only make sure the SQs will be created
1539 		 * with DevX, not with Verbs.
1540 		 * Verbs allocates the SQ UAR on its own and it can't be shared
1541 		 * with Clock Queue UAR as required for Tx scheduling.
1542 		 */
1543 		DRV_LOG(ERR, "Verbs SQs, UAR can't be shared as required for packet pacing");
1544 		err = ENODEV;
1545 		goto error;
1546 	}
1547 	priv->drop_queue.hrxq = mlx5_drop_action_create(eth_dev);
1548 	if (!priv->drop_queue.hrxq)
1549 		goto error;
1550 	priv->hrxqs = mlx5_list_create("hrxq", eth_dev, true,
1551 				       mlx5_hrxq_create_cb,
1552 				       mlx5_hrxq_match_cb,
1553 				       mlx5_hrxq_remove_cb,
1554 				       mlx5_hrxq_clone_cb,
1555 				       mlx5_hrxq_clone_free_cb);
1556 	if (!priv->hrxqs)
1557 		goto error;
1558 	mlx5_set_metadata_mask(eth_dev);
1559 	if (sh->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1560 	    !priv->sh->dv_regc0_mask) {
1561 		DRV_LOG(ERR, "metadata mode %u is not supported "
1562 			     "(no metadata reg_c[0] is available)",
1563 			     sh->config.dv_xmeta_en);
1564 			err = ENOTSUP;
1565 			goto error;
1566 	}
1567 	rte_rwlock_init(&priv->ind_tbls_lock);
1568 	if (priv->sh->config.dv_flow_en == 2) {
1569 #ifdef HAVE_MLX5_HWS_SUPPORT
1570 		if (priv->sh->config.dv_esw_en) {
1571 			uint32_t usable_bits;
1572 			uint32_t required_bits;
1573 
1574 			if (priv->sh->dv_regc0_mask == UINT32_MAX) {
1575 				DRV_LOG(ERR, "E-Switch port metadata is required when using HWS "
1576 					     "but it is disabled (configure it through devlink)");
1577 				err = ENOTSUP;
1578 				goto error;
1579 			}
1580 			if (priv->sh->dv_regc0_mask == 0) {
1581 				DRV_LOG(ERR, "E-Switch with HWS is not supported "
1582 					     "(no available bits in reg_c[0])");
1583 				err = ENOTSUP;
1584 				goto error;
1585 			}
1586 			usable_bits = __builtin_popcount(priv->sh->dv_regc0_mask);
1587 			required_bits = __builtin_popcount(priv->vport_meta_mask);
1588 			if (usable_bits < required_bits) {
1589 				DRV_LOG(ERR, "Not enough bits available in reg_c[0] to provide "
1590 					     "representor matching.");
1591 				err = ENOTSUP;
1592 				goto error;
1593 			}
1594 		}
1595 		if (priv->vport_meta_mask)
1596 			flow_hw_set_port_info(eth_dev);
1597 		if (priv->sh->config.dv_esw_en &&
1598 		    priv->sh->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1599 		    priv->sh->config.dv_xmeta_en != MLX5_XMETA_MODE_META32_HWS) {
1600 			DRV_LOG(ERR,
1601 				"metadata mode %u is not supported in HWS eswitch mode",
1602 				priv->sh->config.dv_xmeta_en);
1603 				err = ENOTSUP;
1604 				goto error;
1605 		}
1606 		/* Only HWS requires this information. */
1607 		flow_hw_init_tags_set(eth_dev);
1608 		flow_hw_init_flow_metadata_config(eth_dev);
1609 		if (priv->sh->config.dv_esw_en &&
1610 		    flow_hw_create_vport_action(eth_dev)) {
1611 			DRV_LOG(ERR, "port %u failed to create vport action",
1612 				eth_dev->data->port_id);
1613 			err = EINVAL;
1614 			goto error;
1615 		}
1616 		return eth_dev;
1617 #else
1618 		DRV_LOG(ERR, "DV support is missing for HWS.");
1619 		goto error;
1620 #endif
1621 	}
1622 	if (!priv->sh->flow_priority_check_flag) {
1623 		/* Supported Verbs flow priority number detection. */
1624 		err = mlx5_flow_discover_priorities(eth_dev);
1625 		priv->sh->flow_max_priority = err;
1626 		priv->sh->flow_priority_check_flag = 1;
1627 	} else {
1628 		err = priv->sh->flow_max_priority;
1629 	}
1630 	if (err < 0) {
1631 		err = -err;
1632 		goto error;
1633 	}
1634 	/* Query availability of metadata reg_c's. */
1635 	if (!priv->sh->metadata_regc_check_flag) {
1636 		err = mlx5_flow_discover_mreg_c(eth_dev);
1637 		if (err < 0) {
1638 			err = -err;
1639 			goto error;
1640 		}
1641 	}
1642 	if (!mlx5_flow_ext_mreg_supported(eth_dev)) {
1643 		DRV_LOG(DEBUG,
1644 			"port %u extensive metadata register is not supported",
1645 			eth_dev->data->port_id);
1646 		if (sh->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY) {
1647 			DRV_LOG(ERR, "metadata mode %u is not supported "
1648 				     "(no metadata registers available)",
1649 				     sh->config.dv_xmeta_en);
1650 			err = ENOTSUP;
1651 			goto error;
1652 		}
1653 	}
1654 	if (sh->config.dv_flow_en &&
1655 	    sh->config.dv_xmeta_en != MLX5_XMETA_MODE_LEGACY &&
1656 	    mlx5_flow_ext_mreg_supported(eth_dev) &&
1657 	    priv->sh->dv_regc0_mask) {
1658 		priv->mreg_cp_tbl = mlx5_hlist_create(MLX5_FLOW_MREG_HNAME,
1659 						      MLX5_FLOW_MREG_HTABLE_SZ,
1660 						      false, true, eth_dev,
1661 						      flow_dv_mreg_create_cb,
1662 						      flow_dv_mreg_match_cb,
1663 						      flow_dv_mreg_remove_cb,
1664 						      flow_dv_mreg_clone_cb,
1665 						    flow_dv_mreg_clone_free_cb);
1666 		if (!priv->mreg_cp_tbl) {
1667 			err = ENOMEM;
1668 			goto error;
1669 		}
1670 	}
1671 	rte_spinlock_init(&priv->shared_act_sl);
1672 	mlx5_flow_counter_mode_config(eth_dev);
1673 	mlx5_flow_drop_action_config(eth_dev);
1674 	if (sh->config.dv_flow_en)
1675 		eth_dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
1676 	return eth_dev;
1677 error:
1678 	if (priv) {
1679 		priv->sh->port[priv->dev_port - 1].nl_ih_port_id =
1680 							       RTE_MAX_ETHPORTS;
1681 		rte_io_wmb();
1682 #ifdef HAVE_MLX5_HWS_SUPPORT
1683 		if (eth_dev &&
1684 		    priv->sh &&
1685 		    priv->sh->config.dv_flow_en == 2 &&
1686 		    priv->sh->config.dv_esw_en)
1687 			flow_hw_destroy_vport_action(eth_dev);
1688 #endif
1689 		if (priv->mreg_cp_tbl)
1690 			mlx5_hlist_destroy(priv->mreg_cp_tbl);
1691 		if (priv->sh)
1692 			mlx5_os_free_shared_dr(priv);
1693 		if (priv->nl_socket_route >= 0)
1694 			close(priv->nl_socket_route);
1695 		if (priv->vmwa_context)
1696 			mlx5_vlan_vmwa_exit(priv->vmwa_context);
1697 		if (eth_dev && priv->drop_queue.hrxq)
1698 			mlx5_drop_action_destroy(eth_dev);
1699 		if (priv->mtr_profile_tbl)
1700 			mlx5_l3t_destroy(priv->mtr_profile_tbl);
1701 		if (own_domain_id)
1702 			claim_zero(rte_eth_switch_domain_free(priv->domain_id));
1703 		if (priv->hrxqs)
1704 			mlx5_list_destroy(priv->hrxqs);
1705 		if (eth_dev && priv->flex_item_map)
1706 			mlx5_flex_item_port_cleanup(eth_dev);
1707 		mlx5_free(priv->ext_rxqs);
1708 		mlx5_free(priv);
1709 		if (eth_dev != NULL)
1710 			eth_dev->data->dev_private = NULL;
1711 	}
1712 	if (eth_dev != NULL) {
1713 		/* mac_addrs must not be freed alone because part of
1714 		 * dev_private
1715 		 **/
1716 		eth_dev->data->mac_addrs = NULL;
1717 		rte_eth_dev_release_port(eth_dev);
1718 	}
1719 	if (sh)
1720 		mlx5_free_shared_dev_ctx(sh);
1721 	if (nl_rdma >= 0)
1722 		close(nl_rdma);
1723 	MLX5_ASSERT(err > 0);
1724 	rte_errno = err;
1725 	return NULL;
1726 }
1727 
1728 /**
1729  * Comparison callback to sort device data.
1730  *
1731  * This is meant to be used with qsort().
1732  *
1733  * @param a[in]
1734  *   Pointer to pointer to first data object.
1735  * @param b[in]
1736  *   Pointer to pointer to second data object.
1737  *
1738  * @return
1739  *   0 if both objects are equal, less than 0 if the first argument is less
1740  *   than the second, greater than 0 otherwise.
1741  */
1742 static int
1743 mlx5_dev_spawn_data_cmp(const void *a, const void *b)
1744 {
1745 	const struct mlx5_switch_info *si_a =
1746 		&((const struct mlx5_dev_spawn_data *)a)->info;
1747 	const struct mlx5_switch_info *si_b =
1748 		&((const struct mlx5_dev_spawn_data *)b)->info;
1749 	int ret;
1750 
1751 	/* Master device first. */
1752 	ret = si_b->master - si_a->master;
1753 	if (ret)
1754 		return ret;
1755 	/* Then representor devices. */
1756 	ret = si_b->representor - si_a->representor;
1757 	if (ret)
1758 		return ret;
1759 	/* Unidentified devices come last in no specific order. */
1760 	if (!si_a->representor)
1761 		return 0;
1762 	/* Order representors by name. */
1763 	return si_a->port_name - si_b->port_name;
1764 }
1765 
1766 /**
1767  * Match PCI information for possible slaves of bonding device.
1768  *
1769  * @param[in] ibdev_name
1770  *   Name of Infiniband device.
1771  * @param[in] pci_dev
1772  *   Pointer to primary PCI address structure to match.
1773  * @param[in] nl_rdma
1774  *   Netlink RDMA group socket handle.
1775  * @param[in] owner
1776  *   Representor owner PF index.
1777  * @param[out] bond_info
1778  *   Pointer to bonding information.
1779  *
1780  * @return
1781  *   negative value if no bonding device found, otherwise
1782  *   positive index of slave PF in bonding.
1783  */
1784 static int
1785 mlx5_device_bond_pci_match(const char *ibdev_name,
1786 			   const struct rte_pci_addr *pci_dev,
1787 			   int nl_rdma, uint16_t owner,
1788 			   struct mlx5_bond_info *bond_info)
1789 {
1790 	char ifname[IF_NAMESIZE + 1];
1791 	unsigned int ifindex;
1792 	unsigned int np, i;
1793 	FILE *bond_file = NULL, *file;
1794 	int pf = -1;
1795 	int ret;
1796 	uint8_t cur_guid[32] = {0};
1797 	uint8_t guid[32] = {0};
1798 
1799 	/*
1800 	 * Try to get master device name. If something goes wrong suppose
1801 	 * the lack of kernel support and no bonding devices.
1802 	 */
1803 	memset(bond_info, 0, sizeof(*bond_info));
1804 	if (nl_rdma < 0)
1805 		return -1;
1806 	if (!strstr(ibdev_name, "bond"))
1807 		return -1;
1808 	np = mlx5_nl_portnum(nl_rdma, ibdev_name);
1809 	if (!np)
1810 		return -1;
1811 	if (mlx5_get_device_guid(pci_dev, cur_guid, sizeof(cur_guid)) < 0)
1812 		return -1;
1813 	/*
1814 	 * The master device might not be on the predefined port(not on port
1815 	 * index 1, it is not guaranteed), we have to scan all Infiniband
1816 	 * device ports and find master.
1817 	 */
1818 	for (i = 1; i <= np; ++i) {
1819 		/* Check whether Infiniband port is populated. */
1820 		ifindex = mlx5_nl_ifindex(nl_rdma, ibdev_name, i);
1821 		if (!ifindex)
1822 			continue;
1823 		if (!if_indextoname(ifindex, ifname))
1824 			continue;
1825 		/* Try to read bonding slave names from sysfs. */
1826 		MKSTR(slaves,
1827 		      "/sys/class/net/%s/master/bonding/slaves", ifname);
1828 		bond_file = fopen(slaves, "r");
1829 		if (bond_file)
1830 			break;
1831 	}
1832 	if (!bond_file)
1833 		return -1;
1834 	/* Use safe format to check maximal buffer length. */
1835 	MLX5_ASSERT(atol(RTE_STR(IF_NAMESIZE)) == IF_NAMESIZE);
1836 	while (fscanf(bond_file, "%" RTE_STR(IF_NAMESIZE) "s", ifname) == 1) {
1837 		char tmp_str[IF_NAMESIZE + 32];
1838 		struct rte_pci_addr pci_addr;
1839 		struct mlx5_switch_info	info;
1840 		int ret;
1841 
1842 		/* Process slave interface names in the loop. */
1843 		snprintf(tmp_str, sizeof(tmp_str),
1844 			 "/sys/class/net/%s", ifname);
1845 		if (mlx5_get_pci_addr(tmp_str, &pci_addr)) {
1846 			DRV_LOG(WARNING,
1847 				"Cannot get PCI address for netdev \"%s\".",
1848 				ifname);
1849 			continue;
1850 		}
1851 		/* Slave interface PCI address match found. */
1852 		snprintf(tmp_str, sizeof(tmp_str),
1853 			 "/sys/class/net/%s/phys_port_name", ifname);
1854 		file = fopen(tmp_str, "rb");
1855 		if (!file)
1856 			break;
1857 		info.name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET;
1858 		if (fscanf(file, "%32s", tmp_str) == 1)
1859 			mlx5_translate_port_name(tmp_str, &info);
1860 		fclose(file);
1861 		/* Only process PF ports. */
1862 		if (info.name_type != MLX5_PHYS_PORT_NAME_TYPE_LEGACY &&
1863 		    info.name_type != MLX5_PHYS_PORT_NAME_TYPE_UPLINK)
1864 			continue;
1865 		/* Check max bonding member. */
1866 		if (info.port_name >= MLX5_BOND_MAX_PORTS) {
1867 			DRV_LOG(WARNING, "bonding index out of range, "
1868 				"please increase MLX5_BOND_MAX_PORTS: %s",
1869 				tmp_str);
1870 			break;
1871 		}
1872 		/* Get ifindex. */
1873 		snprintf(tmp_str, sizeof(tmp_str),
1874 			 "/sys/class/net/%s/ifindex", ifname);
1875 		file = fopen(tmp_str, "rb");
1876 		if (!file)
1877 			break;
1878 		ret = fscanf(file, "%u", &ifindex);
1879 		fclose(file);
1880 		if (ret != 1)
1881 			break;
1882 		/* Save bonding info. */
1883 		strncpy(bond_info->ports[info.port_name].ifname, ifname,
1884 			sizeof(bond_info->ports[0].ifname));
1885 		bond_info->ports[info.port_name].pci_addr = pci_addr;
1886 		bond_info->ports[info.port_name].ifindex = ifindex;
1887 		bond_info->n_port++;
1888 		/*
1889 		 * Under socket direct mode, bonding will use
1890 		 * system_image_guid as identification.
1891 		 * After OFED 5.4, guid is readable (ret >= 0) under sysfs.
1892 		 * All bonding members should have the same guid even if driver
1893 		 * is using PCIe BDF.
1894 		 */
1895 		ret = mlx5_get_device_guid(&pci_addr, guid, sizeof(guid));
1896 		if (ret < 0)
1897 			break;
1898 		else if (ret > 0) {
1899 			if (!memcmp(guid, cur_guid, sizeof(guid)) &&
1900 			    owner == info.port_name &&
1901 			    (owner != 0 || (owner == 0 &&
1902 			    !rte_pci_addr_cmp(pci_dev, &pci_addr))))
1903 				pf = info.port_name;
1904 		} else if (pci_dev->domain == pci_addr.domain &&
1905 		    pci_dev->bus == pci_addr.bus &&
1906 		    pci_dev->devid == pci_addr.devid &&
1907 		    ((pci_dev->function == 0 &&
1908 		      pci_dev->function + owner == pci_addr.function) ||
1909 		     (pci_dev->function == owner &&
1910 		      pci_addr.function == owner)))
1911 			pf = info.port_name;
1912 	}
1913 	if (pf >= 0) {
1914 		/* Get bond interface info */
1915 		ret = mlx5_sysfs_bond_info(ifindex, &bond_info->ifindex,
1916 					   bond_info->ifname);
1917 		if (ret)
1918 			DRV_LOG(ERR, "unable to get bond info: %s",
1919 				strerror(rte_errno));
1920 		else
1921 			DRV_LOG(INFO, "PF device %u, bond device %u(%s)",
1922 				ifindex, bond_info->ifindex, bond_info->ifname);
1923 	}
1924 	if (owner == 0 && pf != 0) {
1925 		DRV_LOG(INFO, "PCIe instance %04x:%02x:%02x.%x isn't bonding owner",
1926 				pci_dev->domain, pci_dev->bus, pci_dev->devid,
1927 				pci_dev->function);
1928 	}
1929 	return pf;
1930 }
1931 
1932 /**
1933  * Register a PCI device within bonding.
1934  *
1935  * This function spawns Ethernet devices out of a given PCI device and
1936  * bonding owner PF index.
1937  *
1938  * @param[in] cdev
1939  *   Pointer to common mlx5 device structure.
1940  * @param[in] req_eth_da
1941  *   Requested ethdev device argument.
1942  * @param[in] owner_id
1943  *   Requested owner PF port ID within bonding device, default to 0.
1944  * @param[in, out] mkvlist
1945  *   Pointer to mlx5 kvargs control, can be NULL if there is no devargs.
1946  *
1947  * @return
1948  *   0 on success, a negative errno value otherwise and rte_errno is set.
1949  */
1950 static int
1951 mlx5_os_pci_probe_pf(struct mlx5_common_device *cdev,
1952 		     struct rte_eth_devargs *req_eth_da,
1953 		     uint16_t owner_id, struct mlx5_kvargs_ctrl *mkvlist)
1954 {
1955 	struct ibv_device **ibv_list;
1956 	/*
1957 	 * Number of found IB Devices matching with requested PCI BDF.
1958 	 * nd != 1 means there are multiple IB devices over the same
1959 	 * PCI device and we have representors and master.
1960 	 */
1961 	unsigned int nd = 0;
1962 	/*
1963 	 * Number of found IB device Ports. nd = 1 and np = 1..n means
1964 	 * we have the single multiport IB device, and there may be
1965 	 * representors attached to some of found ports.
1966 	 */
1967 	unsigned int np = 0;
1968 	/*
1969 	 * Number of DPDK ethernet devices to Spawn - either over
1970 	 * multiple IB devices or multiple ports of single IB device.
1971 	 * Actually this is the number of iterations to spawn.
1972 	 */
1973 	unsigned int ns = 0;
1974 	/*
1975 	 * Bonding device
1976 	 *   < 0 - no bonding device (single one)
1977 	 *  >= 0 - bonding device (value is slave PF index)
1978 	 */
1979 	int bd = -1;
1980 	struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(cdev->dev);
1981 	struct mlx5_dev_spawn_data *list = NULL;
1982 	struct rte_eth_devargs eth_da = *req_eth_da;
1983 	struct rte_pci_addr owner_pci = pci_dev->addr; /* Owner PF. */
1984 	struct mlx5_bond_info bond_info;
1985 	int ret = -1;
1986 
1987 	errno = 0;
1988 	ibv_list = mlx5_glue->get_device_list(&ret);
1989 	if (!ibv_list) {
1990 		rte_errno = errno ? errno : ENOSYS;
1991 		DRV_LOG(ERR, "Cannot list devices, is ib_uverbs loaded?");
1992 		return -rte_errno;
1993 	}
1994 	/*
1995 	 * First scan the list of all Infiniband devices to find
1996 	 * matching ones, gathering into the list.
1997 	 */
1998 	struct ibv_device *ibv_match[ret + 1];
1999 	int nl_route = mlx5_nl_init(NETLINK_ROUTE, 0);
2000 	int nl_rdma = mlx5_nl_init(NETLINK_RDMA, 0);
2001 	unsigned int i;
2002 
2003 	while (ret-- > 0) {
2004 		struct rte_pci_addr pci_addr;
2005 
2006 		DRV_LOG(DEBUG, "Checking device \"%s\"", ibv_list[ret]->name);
2007 		bd = mlx5_device_bond_pci_match(ibv_list[ret]->name, &owner_pci,
2008 						nl_rdma, owner_id, &bond_info);
2009 		if (bd >= 0) {
2010 			/*
2011 			 * Bonding device detected. Only one match is allowed,
2012 			 * the bonding is supported over multi-port IB device,
2013 			 * there should be no matches on representor PCI
2014 			 * functions or non VF LAG bonding devices with
2015 			 * specified address.
2016 			 */
2017 			if (nd) {
2018 				DRV_LOG(ERR,
2019 					"multiple PCI match on bonding device"
2020 					"\"%s\" found", ibv_list[ret]->name);
2021 				rte_errno = ENOENT;
2022 				ret = -rte_errno;
2023 				goto exit;
2024 			}
2025 			/* Amend owner pci address if owner PF ID specified. */
2026 			if (eth_da.nb_representor_ports)
2027 				owner_pci.function += owner_id;
2028 			DRV_LOG(INFO,
2029 				"PCI information matches for slave %d bonding device \"%s\"",
2030 				bd, ibv_list[ret]->name);
2031 			ibv_match[nd++] = ibv_list[ret];
2032 			break;
2033 		} else {
2034 			/* Bonding device not found. */
2035 			if (mlx5_get_pci_addr(ibv_list[ret]->ibdev_path,
2036 					      &pci_addr))
2037 				continue;
2038 			if (owner_pci.domain != pci_addr.domain ||
2039 			    owner_pci.bus != pci_addr.bus ||
2040 			    owner_pci.devid != pci_addr.devid ||
2041 			    owner_pci.function != pci_addr.function)
2042 				continue;
2043 			DRV_LOG(INFO, "PCI information matches for device \"%s\"",
2044 				ibv_list[ret]->name);
2045 			ibv_match[nd++] = ibv_list[ret];
2046 		}
2047 	}
2048 	ibv_match[nd] = NULL;
2049 	if (!nd) {
2050 		/* No device matches, just complain and bail out. */
2051 		DRV_LOG(WARNING,
2052 			"PF %u doesn't have Verbs device matches PCI device " PCI_PRI_FMT ","
2053 			" are kernel drivers loaded?",
2054 			owner_id, owner_pci.domain, owner_pci.bus,
2055 			owner_pci.devid, owner_pci.function);
2056 		rte_errno = ENOENT;
2057 		ret = -rte_errno;
2058 		goto exit;
2059 	}
2060 	if (nd == 1) {
2061 		/*
2062 		 * Found single matching device may have multiple ports.
2063 		 * Each port may be representor, we have to check the port
2064 		 * number and check the representors existence.
2065 		 */
2066 		if (nl_rdma >= 0)
2067 			np = mlx5_nl_portnum(nl_rdma, ibv_match[0]->name);
2068 		if (!np)
2069 			DRV_LOG(WARNING,
2070 				"Cannot get IB device \"%s\" ports number.",
2071 				ibv_match[0]->name);
2072 		if (bd >= 0 && !np) {
2073 			DRV_LOG(ERR, "Cannot get ports for bonding device.");
2074 			rte_errno = ENOENT;
2075 			ret = -rte_errno;
2076 			goto exit;
2077 		}
2078 	}
2079 	/* Now we can determine the maximal amount of devices to be spawned. */
2080 	list = mlx5_malloc(MLX5_MEM_ZERO,
2081 			   sizeof(struct mlx5_dev_spawn_data) * (np ? np : nd),
2082 			   RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
2083 	if (!list) {
2084 		DRV_LOG(ERR, "Spawn data array allocation failure.");
2085 		rte_errno = ENOMEM;
2086 		ret = -rte_errno;
2087 		goto exit;
2088 	}
2089 	if (bd >= 0 || np > 1) {
2090 		/*
2091 		 * Single IB device with multiple ports found,
2092 		 * it may be E-Switch master device and representors.
2093 		 * We have to perform identification through the ports.
2094 		 */
2095 		MLX5_ASSERT(nl_rdma >= 0);
2096 		MLX5_ASSERT(ns == 0);
2097 		MLX5_ASSERT(nd == 1);
2098 		MLX5_ASSERT(np);
2099 		for (i = 1; i <= np; ++i) {
2100 			list[ns].bond_info = &bond_info;
2101 			list[ns].max_port = np;
2102 			list[ns].phys_port = i;
2103 			list[ns].phys_dev_name = ibv_match[0]->name;
2104 			list[ns].eth_dev = NULL;
2105 			list[ns].pci_dev = pci_dev;
2106 			list[ns].cdev = cdev;
2107 			list[ns].pf_bond = bd;
2108 			list[ns].ifindex = mlx5_nl_ifindex(nl_rdma,
2109 							   ibv_match[0]->name,
2110 							   i);
2111 			if (!list[ns].ifindex) {
2112 				/*
2113 				 * No network interface index found for the
2114 				 * specified port, it means there is no
2115 				 * representor on this port. It's OK,
2116 				 * there can be disabled ports, for example
2117 				 * if sriov_numvfs < sriov_totalvfs.
2118 				 */
2119 				continue;
2120 			}
2121 			ret = -1;
2122 			if (nl_route >= 0)
2123 				ret = mlx5_nl_switch_info(nl_route,
2124 							  list[ns].ifindex,
2125 							  &list[ns].info);
2126 			if (ret || (!list[ns].info.representor &&
2127 				    !list[ns].info.master)) {
2128 				/*
2129 				 * We failed to recognize representors with
2130 				 * Netlink, let's try to perform the task
2131 				 * with sysfs.
2132 				 */
2133 				ret = mlx5_sysfs_switch_info(list[ns].ifindex,
2134 							     &list[ns].info);
2135 			}
2136 			if (!ret && bd >= 0) {
2137 				switch (list[ns].info.name_type) {
2138 				case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
2139 					if (np == 1) {
2140 						/*
2141 						 * Force standalone bonding
2142 						 * device for ROCE LAG
2143 						 * configurations.
2144 						 */
2145 						list[ns].info.master = 0;
2146 						list[ns].info.representor = 0;
2147 					}
2148 					if (list[ns].info.port_name == bd)
2149 						ns++;
2150 					break;
2151 				case MLX5_PHYS_PORT_NAME_TYPE_PFHPF:
2152 					/* Fallthrough */
2153 				case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
2154 					/* Fallthrough */
2155 				case MLX5_PHYS_PORT_NAME_TYPE_PFSF:
2156 					if (list[ns].info.pf_num == bd)
2157 						ns++;
2158 					break;
2159 				default:
2160 					break;
2161 				}
2162 				continue;
2163 			}
2164 			if (!ret && (list[ns].info.representor ^
2165 				     list[ns].info.master))
2166 				ns++;
2167 		}
2168 		if (!ns) {
2169 			DRV_LOG(ERR,
2170 				"Unable to recognize master/representors on the IB device with multiple ports.");
2171 			rte_errno = ENOENT;
2172 			ret = -rte_errno;
2173 			goto exit;
2174 		}
2175 	} else {
2176 		/*
2177 		 * The existence of several matching entries (nd > 1) means
2178 		 * port representors have been instantiated. No existing Verbs
2179 		 * call nor sysfs entries can tell them apart, this can only
2180 		 * be done through Netlink calls assuming kernel drivers are
2181 		 * recent enough to support them.
2182 		 *
2183 		 * In the event of identification failure through Netlink,
2184 		 * try again through sysfs, then:
2185 		 *
2186 		 * 1. A single IB device matches (nd == 1) with single
2187 		 *    port (np=0/1) and is not a representor, assume
2188 		 *    no switch support.
2189 		 *
2190 		 * 2. Otherwise no safe assumptions can be made;
2191 		 *    complain louder and bail out.
2192 		 */
2193 		for (i = 0; i != nd; ++i) {
2194 			memset(&list[ns].info, 0, sizeof(list[ns].info));
2195 			list[ns].bond_info = NULL;
2196 			list[ns].max_port = 1;
2197 			list[ns].phys_port = 1;
2198 			list[ns].phys_dev_name = ibv_match[i]->name;
2199 			list[ns].eth_dev = NULL;
2200 			list[ns].pci_dev = pci_dev;
2201 			list[ns].cdev = cdev;
2202 			list[ns].pf_bond = -1;
2203 			list[ns].ifindex = 0;
2204 			if (nl_rdma >= 0)
2205 				list[ns].ifindex = mlx5_nl_ifindex
2206 							    (nl_rdma,
2207 							     ibv_match[i]->name,
2208 							     1);
2209 			if (!list[ns].ifindex) {
2210 				char ifname[IF_NAMESIZE];
2211 
2212 				/*
2213 				 * Netlink failed, it may happen with old
2214 				 * ib_core kernel driver (before 4.16).
2215 				 * We can assume there is old driver because
2216 				 * here we are processing single ports IB
2217 				 * devices. Let's try sysfs to retrieve
2218 				 * the ifindex. The method works for
2219 				 * master device only.
2220 				 */
2221 				if (nd > 1) {
2222 					/*
2223 					 * Multiple devices found, assume
2224 					 * representors, can not distinguish
2225 					 * master/representor and retrieve
2226 					 * ifindex via sysfs.
2227 					 */
2228 					continue;
2229 				}
2230 				ret = mlx5_get_ifname_sysfs
2231 					(ibv_match[i]->ibdev_path, ifname);
2232 				if (!ret)
2233 					list[ns].ifindex =
2234 						if_nametoindex(ifname);
2235 				if (!list[ns].ifindex) {
2236 					/*
2237 					 * No network interface index found
2238 					 * for the specified device, it means
2239 					 * there it is neither representor
2240 					 * nor master.
2241 					 */
2242 					continue;
2243 				}
2244 			}
2245 			ret = -1;
2246 			if (nl_route >= 0)
2247 				ret = mlx5_nl_switch_info(nl_route,
2248 							  list[ns].ifindex,
2249 							  &list[ns].info);
2250 			if (ret || (!list[ns].info.representor &&
2251 				    !list[ns].info.master)) {
2252 				/*
2253 				 * We failed to recognize representors with
2254 				 * Netlink, let's try to perform the task
2255 				 * with sysfs.
2256 				 */
2257 				ret = mlx5_sysfs_switch_info(list[ns].ifindex,
2258 							     &list[ns].info);
2259 			}
2260 			if (!ret && (list[ns].info.representor ^
2261 				     list[ns].info.master)) {
2262 				ns++;
2263 			} else if ((nd == 1) &&
2264 				   !list[ns].info.representor &&
2265 				   !list[ns].info.master) {
2266 				/*
2267 				 * Single IB device with one physical port and
2268 				 * attached network device.
2269 				 * May be SRIOV is not enabled or there is no
2270 				 * representors.
2271 				 */
2272 				DRV_LOG(INFO, "No E-Switch support detected.");
2273 				ns++;
2274 				break;
2275 			}
2276 		}
2277 		if (!ns) {
2278 			DRV_LOG(ERR,
2279 				"Unable to recognize master/representors on the multiple IB devices.");
2280 			rte_errno = ENOENT;
2281 			ret = -rte_errno;
2282 			goto exit;
2283 		}
2284 		/*
2285 		 * New kernels may add the switch_id attribute for the case
2286 		 * there is no E-Switch and we wrongly recognized the only
2287 		 * device as master. Override this if there is the single
2288 		 * device with single port and new device name format present.
2289 		 */
2290 		if (nd == 1 &&
2291 		    list[0].info.name_type == MLX5_PHYS_PORT_NAME_TYPE_UPLINK) {
2292 			list[0].info.master = 0;
2293 			list[0].info.representor = 0;
2294 		}
2295 	}
2296 	MLX5_ASSERT(ns);
2297 	/*
2298 	 * Sort list to probe devices in natural order for users convenience
2299 	 * (i.e. master first, then representors from lowest to highest ID).
2300 	 */
2301 	qsort(list, ns, sizeof(*list), mlx5_dev_spawn_data_cmp);
2302 	if (eth_da.type != RTE_ETH_REPRESENTOR_NONE) {
2303 		/* Set devargs default values. */
2304 		if (eth_da.nb_mh_controllers == 0) {
2305 			eth_da.nb_mh_controllers = 1;
2306 			eth_da.mh_controllers[0] = 0;
2307 		}
2308 		if (eth_da.nb_ports == 0 && ns > 0) {
2309 			if (list[0].pf_bond >= 0 && list[0].info.representor)
2310 				DRV_LOG(WARNING, "Representor on Bonding device should use pf#vf# syntax: %s",
2311 					pci_dev->device.devargs->args);
2312 			eth_da.nb_ports = 1;
2313 			eth_da.ports[0] = list[0].info.pf_num;
2314 		}
2315 		if (eth_da.nb_representor_ports == 0) {
2316 			eth_da.nb_representor_ports = 1;
2317 			eth_da.representor_ports[0] = 0;
2318 		}
2319 	}
2320 	for (i = 0; i != ns; ++i) {
2321 		uint32_t restore;
2322 
2323 		list[i].eth_dev = mlx5_dev_spawn(cdev->dev, &list[i], &eth_da,
2324 						 mkvlist);
2325 		if (!list[i].eth_dev) {
2326 			if (rte_errno != EBUSY && rte_errno != EEXIST)
2327 				break;
2328 			/* Device is disabled or already spawned. Ignore it. */
2329 			continue;
2330 		}
2331 		restore = list[i].eth_dev->data->dev_flags;
2332 		rte_eth_copy_pci_info(list[i].eth_dev, pci_dev);
2333 		/**
2334 		 * Each representor has a dedicated interrupts vector.
2335 		 * rte_eth_copy_pci_info() assigns PF interrupts handle to
2336 		 * representor eth_dev object because representor and PF
2337 		 * share the same PCI address.
2338 		 * Override representor device with a dedicated
2339 		 * interrupts handle here.
2340 		 * Representor interrupts handle is released in mlx5_dev_stop().
2341 		 */
2342 		if (list[i].info.representor) {
2343 			struct rte_intr_handle *intr_handle =
2344 				rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_SHARED);
2345 			if (intr_handle == NULL) {
2346 				DRV_LOG(ERR,
2347 					"port %u failed to allocate memory for interrupt handler "
2348 					"Rx interrupts will not be supported",
2349 					i);
2350 				rte_errno = ENOMEM;
2351 				ret = -rte_errno;
2352 				goto exit;
2353 			}
2354 			list[i].eth_dev->intr_handle = intr_handle;
2355 		}
2356 		/* Restore non-PCI flags cleared by the above call. */
2357 		list[i].eth_dev->data->dev_flags |= restore;
2358 		rte_eth_dev_probing_finish(list[i].eth_dev);
2359 	}
2360 	if (i != ns) {
2361 		DRV_LOG(ERR,
2362 			"probe of PCI device " PCI_PRI_FMT " aborted after"
2363 			" encountering an error: %s",
2364 			owner_pci.domain, owner_pci.bus,
2365 			owner_pci.devid, owner_pci.function,
2366 			strerror(rte_errno));
2367 		ret = -rte_errno;
2368 		/* Roll back. */
2369 		while (i--) {
2370 			if (!list[i].eth_dev)
2371 				continue;
2372 			mlx5_dev_close(list[i].eth_dev);
2373 			/* mac_addrs must not be freed because in dev_private */
2374 			list[i].eth_dev->data->mac_addrs = NULL;
2375 			claim_zero(rte_eth_dev_release_port(list[i].eth_dev));
2376 		}
2377 		/* Restore original error. */
2378 		rte_errno = -ret;
2379 	} else {
2380 		ret = 0;
2381 	}
2382 exit:
2383 	/*
2384 	 * Do the routine cleanup:
2385 	 * - close opened Netlink sockets
2386 	 * - free allocated spawn data array
2387 	 * - free the Infiniband device list
2388 	 */
2389 	if (nl_rdma >= 0)
2390 		close(nl_rdma);
2391 	if (nl_route >= 0)
2392 		close(nl_route);
2393 	if (list)
2394 		mlx5_free(list);
2395 	MLX5_ASSERT(ibv_list);
2396 	mlx5_glue->free_device_list(ibv_list);
2397 	return ret;
2398 }
2399 
2400 static int
2401 mlx5_os_parse_eth_devargs(struct rte_device *dev,
2402 			  struct rte_eth_devargs *eth_da)
2403 {
2404 	int ret = 0;
2405 
2406 	if (dev->devargs == NULL)
2407 		return 0;
2408 	memset(eth_da, 0, sizeof(*eth_da));
2409 	/* Parse representor information first from class argument. */
2410 	if (dev->devargs->cls_str)
2411 		ret = rte_eth_devargs_parse(dev->devargs->cls_str, eth_da);
2412 	if (ret != 0) {
2413 		DRV_LOG(ERR, "failed to parse device arguments: %s",
2414 			dev->devargs->cls_str);
2415 		return -rte_errno;
2416 	}
2417 	if (eth_da->type == RTE_ETH_REPRESENTOR_NONE && dev->devargs->args) {
2418 		/* Parse legacy device argument */
2419 		ret = rte_eth_devargs_parse(dev->devargs->args, eth_da);
2420 		if (ret) {
2421 			DRV_LOG(ERR, "failed to parse device arguments: %s",
2422 				dev->devargs->args);
2423 			return -rte_errno;
2424 		}
2425 	}
2426 	return 0;
2427 }
2428 
2429 /**
2430  * Callback to register a PCI device.
2431  *
2432  * This function spawns Ethernet devices out of a given PCI device.
2433  *
2434  * @param[in] cdev
2435  *   Pointer to common mlx5 device structure.
2436  * @param[in, out] mkvlist
2437  *   Pointer to mlx5 kvargs control, can be NULL if there is no devargs.
2438  *
2439  * @return
2440  *   0 on success, a negative errno value otherwise and rte_errno is set.
2441  */
2442 static int
2443 mlx5_os_pci_probe(struct mlx5_common_device *cdev,
2444 		  struct mlx5_kvargs_ctrl *mkvlist)
2445 {
2446 	struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(cdev->dev);
2447 	struct rte_eth_devargs eth_da = { .nb_ports = 0 };
2448 	int ret = 0;
2449 	uint16_t p;
2450 
2451 	ret = mlx5_os_parse_eth_devargs(cdev->dev, &eth_da);
2452 	if (ret != 0)
2453 		return ret;
2454 
2455 	if (eth_da.nb_ports > 0) {
2456 		/* Iterate all port if devargs pf is range: "pf[0-1]vf[...]". */
2457 		for (p = 0; p < eth_da.nb_ports; p++) {
2458 			ret = mlx5_os_pci_probe_pf(cdev, &eth_da,
2459 						   eth_da.ports[p], mkvlist);
2460 			if (ret) {
2461 				DRV_LOG(INFO, "Probe of PCI device " PCI_PRI_FMT " "
2462 					"aborted due to proding failure of PF %u",
2463 					pci_dev->addr.domain, pci_dev->addr.bus,
2464 					pci_dev->addr.devid, pci_dev->addr.function,
2465 					eth_da.ports[p]);
2466 				mlx5_net_remove(cdev);
2467 				if (p != 0)
2468 					break;
2469 			}
2470 		}
2471 	} else {
2472 		ret = mlx5_os_pci_probe_pf(cdev, &eth_da, 0, mkvlist);
2473 	}
2474 	return ret;
2475 }
2476 
2477 /* Probe a single SF device on auxiliary bus, no representor support. */
2478 static int
2479 mlx5_os_auxiliary_probe(struct mlx5_common_device *cdev,
2480 			struct mlx5_kvargs_ctrl *mkvlist)
2481 {
2482 	struct rte_eth_devargs eth_da = { .nb_ports = 0 };
2483 	struct mlx5_dev_spawn_data spawn = { .pf_bond = -1 };
2484 	struct rte_device *dev = cdev->dev;
2485 	struct rte_auxiliary_device *adev = RTE_DEV_TO_AUXILIARY(dev);
2486 	struct rte_eth_dev *eth_dev;
2487 	int ret = 0;
2488 
2489 	/* Parse ethdev devargs. */
2490 	ret = mlx5_os_parse_eth_devargs(dev, &eth_da);
2491 	if (ret != 0)
2492 		return ret;
2493 	/* Init spawn data. */
2494 	spawn.max_port = 1;
2495 	spawn.phys_port = 1;
2496 	spawn.phys_dev_name = mlx5_os_get_ctx_device_name(cdev->ctx);
2497 	ret = mlx5_auxiliary_get_ifindex(dev->name);
2498 	if (ret < 0) {
2499 		DRV_LOG(ERR, "failed to get ethdev ifindex: %s", dev->name);
2500 		return ret;
2501 	}
2502 	spawn.ifindex = ret;
2503 	spawn.cdev = cdev;
2504 	/* Spawn device. */
2505 	eth_dev = mlx5_dev_spawn(dev, &spawn, &eth_da, mkvlist);
2506 	if (eth_dev == NULL)
2507 		return -rte_errno;
2508 	/* Post create. */
2509 	eth_dev->intr_handle = adev->intr_handle;
2510 	if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
2511 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
2512 		eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_RMV;
2513 		eth_dev->data->numa_node = dev->numa_node;
2514 	}
2515 	rte_eth_dev_probing_finish(eth_dev);
2516 	return 0;
2517 }
2518 
2519 /**
2520  * Net class driver callback to probe a device.
2521  *
2522  * This function probe PCI bus device(s) or a single SF on auxiliary bus.
2523  *
2524  * @param[in] cdev
2525  *   Pointer to the common mlx5 device.
2526  * @param[in, out] mkvlist
2527  *   Pointer to mlx5 kvargs control, can be NULL if there is no devargs.
2528  *
2529  * @return
2530  *   0 on success, a negative errno value otherwise and rte_errno is set.
2531  */
2532 int
2533 mlx5_os_net_probe(struct mlx5_common_device *cdev,
2534 		  struct mlx5_kvargs_ctrl *mkvlist)
2535 {
2536 	int ret;
2537 
2538 	if (rte_eal_process_type() == RTE_PROC_PRIMARY)
2539 		mlx5_pmd_socket_init();
2540 	ret = mlx5_init_once();
2541 	if (ret) {
2542 		DRV_LOG(ERR, "Unable to init PMD global data: %s",
2543 			strerror(rte_errno));
2544 		return -rte_errno;
2545 	}
2546 	ret = mlx5_probe_again_args_validate(cdev, mkvlist);
2547 	if (ret) {
2548 		DRV_LOG(ERR, "Probe again parameters are not compatible : %s",
2549 			strerror(rte_errno));
2550 		return -rte_errno;
2551 	}
2552 	if (mlx5_dev_is_pci(cdev->dev))
2553 		return mlx5_os_pci_probe(cdev, mkvlist);
2554 	else
2555 		return mlx5_os_auxiliary_probe(cdev, mkvlist);
2556 }
2557 
2558 /**
2559  * Cleanup resources when the last device is closed.
2560  */
2561 void
2562 mlx5_os_net_cleanup(void)
2563 {
2564 	mlx5_pmd_socket_uninit();
2565 }
2566 
2567 /**
2568  * Install shared asynchronous device events handler.
2569  * This function is implemented to support event sharing
2570  * between multiple ports of single IB device.
2571  *
2572  * @param sh
2573  *   Pointer to mlx5_dev_ctx_shared object.
2574  */
2575 void
2576 mlx5_os_dev_shared_handler_install(struct mlx5_dev_ctx_shared *sh)
2577 {
2578 	struct ibv_context *ctx = sh->cdev->ctx;
2579 	int nlsk_fd;
2580 
2581 	sh->intr_handle = mlx5_os_interrupt_handler_create
2582 		(RTE_INTR_INSTANCE_F_SHARED, true,
2583 		 ctx->async_fd, mlx5_dev_interrupt_handler, sh);
2584 	if (!sh->intr_handle) {
2585 		DRV_LOG(ERR, "Failed to allocate intr_handle.");
2586 		return;
2587 	}
2588 	nlsk_fd = mlx5_nl_init(NETLINK_ROUTE, RTMGRP_LINK);
2589 	if (nlsk_fd < 0) {
2590 		DRV_LOG(ERR, "Failed to create a socket for Netlink events: %s",
2591 			rte_strerror(rte_errno));
2592 		return;
2593 	}
2594 	sh->intr_handle_nl = mlx5_os_interrupt_handler_create
2595 		(RTE_INTR_INSTANCE_F_SHARED, true,
2596 		 nlsk_fd, mlx5_dev_interrupt_handler_nl, sh);
2597 	if (sh->intr_handle_nl == NULL) {
2598 		DRV_LOG(ERR, "Fail to allocate intr_handle");
2599 		return;
2600 	}
2601 	if (sh->cdev->config.devx) {
2602 #ifdef HAVE_IBV_DEVX_ASYNC
2603 		struct mlx5dv_devx_cmd_comp *devx_comp;
2604 
2605 		sh->devx_comp = (void *)mlx5_glue->devx_create_cmd_comp(ctx);
2606 		devx_comp = sh->devx_comp;
2607 		if (!devx_comp) {
2608 			DRV_LOG(INFO, "failed to allocate devx_comp.");
2609 			return;
2610 		}
2611 		sh->intr_handle_devx = mlx5_os_interrupt_handler_create
2612 			(RTE_INTR_INSTANCE_F_SHARED, true,
2613 			 devx_comp->fd,
2614 			 mlx5_dev_interrupt_handler_devx, sh);
2615 		if (!sh->intr_handle_devx) {
2616 			DRV_LOG(ERR, "Failed to allocate intr_handle.");
2617 			return;
2618 		}
2619 #endif /* HAVE_IBV_DEVX_ASYNC */
2620 	}
2621 }
2622 
2623 /**
2624  * Uninstall shared asynchronous device events handler.
2625  * This function is implemented to support event sharing
2626  * between multiple ports of single IB device.
2627  *
2628  * @param dev
2629  *   Pointer to mlx5_dev_ctx_shared object.
2630  */
2631 void
2632 mlx5_os_dev_shared_handler_uninstall(struct mlx5_dev_ctx_shared *sh)
2633 {
2634 	mlx5_os_interrupt_handler_destroy(sh->intr_handle,
2635 					  mlx5_dev_interrupt_handler, sh);
2636 	mlx5_os_interrupt_handler_destroy(sh->intr_handle_nl,
2637 					  mlx5_dev_interrupt_handler_nl, sh);
2638 #ifdef HAVE_IBV_DEVX_ASYNC
2639 	mlx5_os_interrupt_handler_destroy(sh->intr_handle_devx,
2640 					  mlx5_dev_interrupt_handler_devx, sh);
2641 	if (sh->devx_comp)
2642 		mlx5_glue->devx_destroy_cmd_comp(sh->devx_comp);
2643 #endif
2644 }
2645 
2646 /**
2647  * Read statistics by a named counter.
2648  *
2649  * @param[in] priv
2650  *   Pointer to the private device data structure.
2651  * @param[in] ctr_name
2652  *   Pointer to the name of the statistic counter to read
2653  * @param[out] stat
2654  *   Pointer to read statistic value.
2655  * @return
2656  *   0 on success and stat is valud, 1 if failed to read the value
2657  *   rte_errno is set.
2658  *
2659  */
2660 int
2661 mlx5_os_read_dev_stat(struct mlx5_priv *priv, const char *ctr_name,
2662 		      uint64_t *stat)
2663 {
2664 	int fd;
2665 
2666 	if (priv->sh) {
2667 		if (priv->q_counters != NULL &&
2668 		    strcmp(ctr_name, "out_of_buffer") == 0)
2669 			return mlx5_devx_cmd_queue_counter_query
2670 					(priv->q_counters, 0, (uint32_t *)stat);
2671 		MKSTR(path, "%s/ports/%d/hw_counters/%s",
2672 		      priv->sh->ibdev_path,
2673 		      priv->dev_port,
2674 		      ctr_name);
2675 		fd = open(path, O_RDONLY);
2676 		/*
2677 		 * in switchdev the file location is not per port
2678 		 * but rather in <ibdev_path>/hw_counters/<file_name>.
2679 		 */
2680 		if (fd == -1) {
2681 			MKSTR(path1, "%s/hw_counters/%s",
2682 			      priv->sh->ibdev_path,
2683 			      ctr_name);
2684 			fd = open(path1, O_RDONLY);
2685 		}
2686 		if (fd != -1) {
2687 			char buf[21] = {'\0'};
2688 			ssize_t n = read(fd, buf, sizeof(buf));
2689 
2690 			close(fd);
2691 			if (n != -1) {
2692 				*stat = strtoull(buf, NULL, 10);
2693 				return 0;
2694 			}
2695 		}
2696 	}
2697 	*stat = 0;
2698 	return 1;
2699 }
2700 
2701 /**
2702  * Remove a MAC address from device
2703  *
2704  * @param dev
2705  *   Pointer to Ethernet device structure.
2706  * @param index
2707  *   MAC address index.
2708  */
2709 void
2710 mlx5_os_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
2711 {
2712 	struct mlx5_priv *priv = dev->data->dev_private;
2713 	const int vf = priv->sh->dev_cap.vf;
2714 
2715 	if (vf)
2716 		mlx5_nl_mac_addr_remove(priv->nl_socket_route,
2717 					mlx5_ifindex(dev), priv->mac_own,
2718 					&dev->data->mac_addrs[index], index);
2719 }
2720 
2721 /**
2722  * Adds a MAC address to the device
2723  *
2724  * @param dev
2725  *   Pointer to Ethernet device structure.
2726  * @param mac_addr
2727  *   MAC address to register.
2728  * @param index
2729  *   MAC address index.
2730  *
2731  * @return
2732  *   0 on success, a negative errno value otherwise
2733  */
2734 int
2735 mlx5_os_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac,
2736 		     uint32_t index)
2737 {
2738 	struct mlx5_priv *priv = dev->data->dev_private;
2739 	const int vf = priv->sh->dev_cap.vf;
2740 	int ret = 0;
2741 
2742 	if (vf)
2743 		ret = mlx5_nl_mac_addr_add(priv->nl_socket_route,
2744 					   mlx5_ifindex(dev), priv->mac_own,
2745 					   mac, index);
2746 	return ret;
2747 }
2748 
2749 /**
2750  * Modify a VF MAC address
2751  *
2752  * @param priv
2753  *   Pointer to device private data.
2754  * @param mac_addr
2755  *   MAC address to modify into.
2756  * @param iface_idx
2757  *   Net device interface index
2758  * @param vf_index
2759  *   VF index
2760  *
2761  * @return
2762  *   0 on success, a negative errno value otherwise
2763  */
2764 int
2765 mlx5_os_vf_mac_addr_modify(struct mlx5_priv *priv,
2766 			   unsigned int iface_idx,
2767 			   struct rte_ether_addr *mac_addr,
2768 			   int vf_index)
2769 {
2770 	return mlx5_nl_vf_mac_addr_modify
2771 		(priv->nl_socket_route, iface_idx, mac_addr, vf_index);
2772 }
2773 
2774 /**
2775  * Set device promiscuous mode
2776  *
2777  * @param dev
2778  *   Pointer to Ethernet device structure.
2779  * @param enable
2780  *   0 - promiscuous is disabled, otherwise - enabled
2781  *
2782  * @return
2783  *   0 on success, a negative error value otherwise
2784  */
2785 int
2786 mlx5_os_set_promisc(struct rte_eth_dev *dev, int enable)
2787 {
2788 	struct mlx5_priv *priv = dev->data->dev_private;
2789 
2790 	return mlx5_nl_promisc(priv->nl_socket_route,
2791 			       mlx5_ifindex(dev), !!enable);
2792 }
2793 
2794 /**
2795  * Set device promiscuous mode
2796  *
2797  * @param dev
2798  *   Pointer to Ethernet device structure.
2799  * @param enable
2800  *   0 - all multicase is disabled, otherwise - enabled
2801  *
2802  * @return
2803  *   0 on success, a negative error value otherwise
2804  */
2805 int
2806 mlx5_os_set_allmulti(struct rte_eth_dev *dev, int enable)
2807 {
2808 	struct mlx5_priv *priv = dev->data->dev_private;
2809 
2810 	return mlx5_nl_allmulti(priv->nl_socket_route,
2811 				mlx5_ifindex(dev), !!enable);
2812 }
2813 
2814 /**
2815  * Flush device MAC addresses
2816  *
2817  * @param dev
2818  *   Pointer to Ethernet device structure.
2819  *
2820  */
2821 void
2822 mlx5_os_mac_addr_flush(struct rte_eth_dev *dev)
2823 {
2824 	struct mlx5_priv *priv = dev->data->dev_private;
2825 
2826 	mlx5_nl_mac_addr_flush(priv->nl_socket_route, mlx5_ifindex(dev),
2827 			       dev->data->mac_addrs,
2828 			       MLX5_MAX_MAC_ADDRESSES, priv->mac_own);
2829 }
2830