1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
2 /*
3  * Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved.
4  * Copyright (c) 2020, Intel Corporation. All rights reserved.
5  */
6 
7 #ifndef MLX5_IB_H
8 #define MLX5_IB_H
9 
10 #include <linux/kernel.h>
11 #include <linux/sched.h>
12 #include <rdma/ib_verbs.h>
13 #include <rdma/ib_umem.h>
14 #include <rdma/ib_smi.h>
15 #include <linux/mlx5/driver.h>
16 #include <linux/mlx5/cq.h>
17 #include <linux/mlx5/fs.h>
18 #include <linux/mlx5/qp.h>
19 #include <linux/types.h>
20 #include <linux/mlx5/transobj.h>
21 #include <rdma/ib_user_verbs.h>
22 #include <rdma/mlx5-abi.h>
23 #include <rdma/uverbs_ioctl.h>
24 #include <rdma/mlx5_user_ioctl_cmds.h>
25 #include <rdma/mlx5_user_ioctl_verbs.h>
26 
27 #include "srq.h"
28 #include "qp.h"
29 #include "macsec.h"
30 
31 #define mlx5_ib_dbg(_dev, format, arg...)                                      \
32 	dev_dbg(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,      \
33 		__LINE__, current->pid, ##arg)
34 
35 #define mlx5_ib_err(_dev, format, arg...)                                      \
36 	dev_err(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,      \
37 		__LINE__, current->pid, ##arg)
38 
39 #define mlx5_ib_warn(_dev, format, arg...)                                     \
40 	dev_warn(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,     \
41 		 __LINE__, current->pid, ##arg)
42 
43 #define mlx5_ib_log(lvl, _dev, format, arg...)                                 \
44 	dev_printk(lvl, &(_dev)->ib_dev.dev,  "%s:%d:(pid %d): " format,       \
45 		   __func__, __LINE__, current->pid, ##arg)
46 
47 #define MLX5_IB_DEFAULT_UIDX 0xffffff
48 #define MLX5_USER_ASSIGNED_UIDX_MASK __mlx5_mask(qpc, user_index)
49 
50 static __always_inline unsigned long
51 __mlx5_log_page_size_to_bitmap(unsigned int log_pgsz_bits,
52 			       unsigned int pgsz_shift)
53 {
54 	unsigned int largest_pg_shift =
55 		min_t(unsigned long, (1ULL << log_pgsz_bits) - 1 + pgsz_shift,
56 		      BITS_PER_LONG - 1);
57 
58 	/*
59 	 * Despite a command allowing it, the device does not support lower than
60 	 * 4k page size.
61 	 */
62 	pgsz_shift = max_t(unsigned int, MLX5_ADAPTER_PAGE_SHIFT, pgsz_shift);
63 	return GENMASK(largest_pg_shift, pgsz_shift);
64 }
65 
66 /*
67  * For mkc users, instead of a page_offset the command has a start_iova which
68  * specifies both the page_offset and the on-the-wire IOVA
69  */
70 #define mlx5_umem_find_best_pgsz(umem, typ, log_pgsz_fld, pgsz_shift, iova)    \
71 	ib_umem_find_best_pgsz(umem,                                           \
72 			       __mlx5_log_page_size_to_bitmap(                 \
73 				       __mlx5_bit_sz(typ, log_pgsz_fld),       \
74 				       pgsz_shift),                            \
75 			       iova)
76 
77 static __always_inline unsigned long
78 __mlx5_page_offset_to_bitmask(unsigned int page_offset_bits,
79 			      unsigned int offset_shift)
80 {
81 	unsigned int largest_offset_shift =
82 		min_t(unsigned long, page_offset_bits - 1 + offset_shift,
83 		      BITS_PER_LONG - 1);
84 
85 	return GENMASK(largest_offset_shift, offset_shift);
86 }
87 
88 /*
89  * QP/CQ/WQ/etc type commands take a page offset that satisifies:
90  *   page_offset_quantized * (page_size/scale) = page_offset
91  * Which restricts allowed page sizes to ones that satisify the above.
92  */
93 unsigned long __mlx5_umem_find_best_quantized_pgoff(
94 	struct ib_umem *umem, unsigned long pgsz_bitmap,
95 	unsigned int page_offset_bits, u64 pgoff_bitmask, unsigned int scale,
96 	unsigned int *page_offset_quantized);
97 #define mlx5_umem_find_best_quantized_pgoff(umem, typ, log_pgsz_fld,           \
98 					    pgsz_shift, page_offset_fld,       \
99 					    scale, page_offset_quantized)      \
100 	__mlx5_umem_find_best_quantized_pgoff(                                 \
101 		umem,                                                          \
102 		__mlx5_log_page_size_to_bitmap(                                \
103 			__mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift),         \
104 		__mlx5_bit_sz(typ, page_offset_fld),                           \
105 		GENMASK(31, order_base_2(scale)), scale,                       \
106 		page_offset_quantized)
107 
108 #define mlx5_umem_find_best_cq_quantized_pgoff(umem, typ, log_pgsz_fld,        \
109 					       pgsz_shift, page_offset_fld,    \
110 					       scale, page_offset_quantized)   \
111 	__mlx5_umem_find_best_quantized_pgoff(                                 \
112 		umem,                                                          \
113 		__mlx5_log_page_size_to_bitmap(                                \
114 			__mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift),         \
115 		__mlx5_bit_sz(typ, page_offset_fld), 0, scale,                 \
116 		page_offset_quantized)
117 
118 static inline unsigned long
119 mlx5_umem_dmabuf_find_best_pgsz(struct ib_umem_dmabuf *umem_dmabuf)
120 {
121 	/*
122 	 * mkeys used for dmabuf are fixed at PAGE_SIZE because we must be able
123 	 * to hold any sgl after a move operation. Ideally the mkc page size
124 	 * could be changed at runtime to be optimal, but right now the driver
125 	 * cannot do that.
126 	 */
127 	return ib_umem_find_best_pgsz(&umem_dmabuf->umem, PAGE_SIZE,
128 				      umem_dmabuf->umem.iova);
129 }
130 
131 enum {
132 	MLX5_IB_MMAP_OFFSET_START = 9,
133 	MLX5_IB_MMAP_OFFSET_END = 255,
134 };
135 
136 enum {
137 	MLX5_IB_MMAP_CMD_SHIFT	= 8,
138 	MLX5_IB_MMAP_CMD_MASK	= 0xff,
139 };
140 
141 enum {
142 	MLX5_RES_SCAT_DATA32_CQE	= 0x1,
143 	MLX5_RES_SCAT_DATA64_CQE	= 0x2,
144 	MLX5_REQ_SCAT_DATA32_CQE	= 0x11,
145 	MLX5_REQ_SCAT_DATA64_CQE	= 0x22,
146 };
147 
148 enum mlx5_ib_mad_ifc_flags {
149 	MLX5_MAD_IFC_IGNORE_MKEY	= 1,
150 	MLX5_MAD_IFC_IGNORE_BKEY	= 2,
151 	MLX5_MAD_IFC_NET_VIEW		= 4,
152 };
153 
154 enum {
155 	MLX5_CROSS_CHANNEL_BFREG         = 0,
156 };
157 
158 enum {
159 	MLX5_CQE_VERSION_V0,
160 	MLX5_CQE_VERSION_V1,
161 };
162 
163 enum {
164 	MLX5_TM_MAX_RNDV_MSG_SIZE	= 64,
165 	MLX5_TM_MAX_SGE			= 1,
166 };
167 
168 enum {
169 	MLX5_IB_INVALID_UAR_INDEX	= BIT(31),
170 	MLX5_IB_INVALID_BFREG		= BIT(31),
171 };
172 
173 enum {
174 	MLX5_MAX_MEMIC_PAGES = 0x100,
175 	MLX5_MEMIC_ALLOC_SIZE_MASK = 0x3f,
176 };
177 
178 enum {
179 	MLX5_MEMIC_BASE_ALIGN	= 6,
180 	MLX5_MEMIC_BASE_SIZE	= 1 << MLX5_MEMIC_BASE_ALIGN,
181 };
182 
183 enum mlx5_ib_mmap_type {
184 	MLX5_IB_MMAP_TYPE_MEMIC = 1,
185 	MLX5_IB_MMAP_TYPE_VAR = 2,
186 	MLX5_IB_MMAP_TYPE_UAR_WC = 3,
187 	MLX5_IB_MMAP_TYPE_UAR_NC = 4,
188 	MLX5_IB_MMAP_TYPE_MEMIC_OP = 5,
189 };
190 
191 struct mlx5_bfreg_info {
192 	u32 *sys_pages;
193 	int num_low_latency_bfregs;
194 	unsigned int *count;
195 
196 	/*
197 	 * protect bfreg allocation data structs
198 	 */
199 	struct mutex lock;
200 	u32 ver;
201 	u8 lib_uar_4k : 1;
202 	u8 lib_uar_dyn : 1;
203 	u32 num_sys_pages;
204 	u32 num_static_sys_pages;
205 	u32 total_num_bfregs;
206 	u32 num_dyn_bfregs;
207 };
208 
209 struct mlx5_ib_ucontext {
210 	struct ib_ucontext	ibucontext;
211 	struct list_head	db_page_list;
212 
213 	/* protect doorbell record alloc/free
214 	 */
215 	struct mutex		db_page_mutex;
216 	struct mlx5_bfreg_info	bfregi;
217 	u8			cqe_version;
218 	/* Transport Domain number */
219 	u32			tdn;
220 
221 	u64			lib_caps;
222 	u16			devx_uid;
223 	/* For RoCE LAG TX affinity */
224 	atomic_t		tx_port_affinity;
225 };
226 
227 static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
228 {
229 	return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext);
230 }
231 
232 struct mlx5_ib_pd {
233 	struct ib_pd		ibpd;
234 	u32			pdn;
235 	u16			uid;
236 };
237 
238 enum {
239 	MLX5_IB_FLOW_ACTION_MODIFY_HEADER,
240 	MLX5_IB_FLOW_ACTION_PACKET_REFORMAT,
241 	MLX5_IB_FLOW_ACTION_DECAP,
242 };
243 
244 #define MLX5_IB_FLOW_MCAST_PRIO		(MLX5_BY_PASS_NUM_PRIOS - 1)
245 #define MLX5_IB_FLOW_LAST_PRIO		(MLX5_BY_PASS_NUM_REGULAR_PRIOS - 1)
246 #if (MLX5_IB_FLOW_LAST_PRIO <= 0)
247 #error "Invalid number of bypass priorities"
248 #endif
249 #define MLX5_IB_FLOW_LEFTOVERS_PRIO	(MLX5_IB_FLOW_MCAST_PRIO + 1)
250 
251 #define MLX5_IB_NUM_FLOW_FT		(MLX5_IB_FLOW_LEFTOVERS_PRIO + 1)
252 #define MLX5_IB_NUM_SNIFFER_FTS		2
253 #define MLX5_IB_NUM_EGRESS_FTS		1
254 #define MLX5_IB_NUM_FDB_FTS		MLX5_BY_PASS_NUM_REGULAR_PRIOS
255 
256 struct mlx5_ib_anchor {
257 	struct mlx5_flow_table *ft;
258 	struct mlx5_flow_group *fg_goto_table;
259 	struct mlx5_flow_group *fg_drop;
260 	struct mlx5_flow_handle *rule_goto_table;
261 	struct mlx5_flow_handle *rule_drop;
262 	unsigned int rule_goto_table_ref;
263 };
264 
265 struct mlx5_ib_flow_prio {
266 	struct mlx5_flow_table		*flow_table;
267 	struct mlx5_ib_anchor		anchor;
268 	unsigned int			refcount;
269 };
270 
271 struct mlx5_ib_flow_handler {
272 	struct list_head		list;
273 	struct ib_flow			ibflow;
274 	struct mlx5_ib_flow_prio	*prio;
275 	struct mlx5_flow_handle		*rule;
276 	struct ib_counters		*ibcounters;
277 	struct mlx5_ib_dev		*dev;
278 	struct mlx5_ib_flow_matcher	*flow_matcher;
279 };
280 
281 struct mlx5_ib_flow_matcher {
282 	struct mlx5_ib_match_params matcher_mask;
283 	int			mask_len;
284 	enum mlx5_ib_flow_type	flow_type;
285 	enum mlx5_flow_namespace_type ns_type;
286 	u16			priority;
287 	struct mlx5_core_dev	*mdev;
288 	atomic_t		usecnt;
289 	u8			match_criteria_enable;
290 };
291 
292 struct mlx5_ib_steering_anchor {
293 	struct mlx5_ib_flow_prio *ft_prio;
294 	struct mlx5_ib_dev *dev;
295 	atomic_t usecnt;
296 };
297 
298 struct mlx5_ib_pp {
299 	u16 index;
300 	struct mlx5_core_dev *mdev;
301 };
302 
303 enum mlx5_ib_optional_counter_type {
304 	MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS,
305 	MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS,
306 	MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS,
307 
308 	MLX5_IB_OPCOUNTER_MAX,
309 };
310 
311 struct mlx5_ib_flow_db {
312 	struct mlx5_ib_flow_prio	prios[MLX5_IB_NUM_FLOW_FT];
313 	struct mlx5_ib_flow_prio	egress_prios[MLX5_IB_NUM_FLOW_FT];
314 	struct mlx5_ib_flow_prio	sniffer[MLX5_IB_NUM_SNIFFER_FTS];
315 	struct mlx5_ib_flow_prio	egress[MLX5_IB_NUM_EGRESS_FTS];
316 	struct mlx5_ib_flow_prio	fdb[MLX5_IB_NUM_FDB_FTS];
317 	struct mlx5_ib_flow_prio	rdma_rx[MLX5_IB_NUM_FLOW_FT];
318 	struct mlx5_ib_flow_prio	rdma_tx[MLX5_IB_NUM_FLOW_FT];
319 	struct mlx5_ib_flow_prio	opfcs[MLX5_IB_OPCOUNTER_MAX];
320 	struct mlx5_flow_table		*lag_demux_ft;
321 	/* Protect flow steering bypass flow tables
322 	 * when add/del flow rules.
323 	 * only single add/removal of flow steering rule could be done
324 	 * simultaneously.
325 	 */
326 	struct mutex			lock;
327 };
328 
329 /* Use macros here so that don't have to duplicate
330  * enum ib_qp_type for low-level driver
331  */
332 
333 #define MLX5_IB_QPT_REG_UMR	IB_QPT_RESERVED1
334 /*
335  * IB_QPT_GSI creates the software wrapper around GSI, and MLX5_IB_QPT_HW_GSI
336  * creates the actual hardware QP.
337  */
338 #define MLX5_IB_QPT_HW_GSI	IB_QPT_RESERVED2
339 #define MLX5_IB_QPT_DCI		IB_QPT_RESERVED3
340 #define MLX5_IB_QPT_DCT		IB_QPT_RESERVED4
341 #define MLX5_IB_WR_UMR		IB_WR_RESERVED1
342 
343 #define MLX5_IB_UPD_XLT_ZAP	      BIT(0)
344 #define MLX5_IB_UPD_XLT_ENABLE	      BIT(1)
345 #define MLX5_IB_UPD_XLT_ATOMIC	      BIT(2)
346 #define MLX5_IB_UPD_XLT_ADDR	      BIT(3)
347 #define MLX5_IB_UPD_XLT_PD	      BIT(4)
348 #define MLX5_IB_UPD_XLT_ACCESS	      BIT(5)
349 #define MLX5_IB_UPD_XLT_INDIRECT      BIT(6)
350 
351 /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags.
352  *
353  * These flags are intended for internal use by the mlx5_ib driver, and they
354  * rely on the range reserved for that use in the ib_qp_create_flags enum.
355  */
356 #define MLX5_IB_QP_CREATE_SQPN_QP1	IB_QP_CREATE_RESERVED_START
357 #define MLX5_IB_QP_CREATE_WC_TEST	(IB_QP_CREATE_RESERVED_START << 1)
358 
359 struct wr_list {
360 	u16	opcode;
361 	u16	next;
362 };
363 
364 enum mlx5_ib_rq_flags {
365 	MLX5_IB_RQ_CVLAN_STRIPPING	= 1 << 0,
366 	MLX5_IB_RQ_PCI_WRITE_END_PADDING	= 1 << 1,
367 };
368 
369 struct mlx5_ib_wq {
370 	struct mlx5_frag_buf_ctrl fbc;
371 	u64		       *wrid;
372 	u32		       *wr_data;
373 	struct wr_list	       *w_list;
374 	unsigned	       *wqe_head;
375 	u16		        unsig_count;
376 
377 	/* serialize post to the work queue
378 	 */
379 	spinlock_t		lock;
380 	int			wqe_cnt;
381 	int			max_post;
382 	int			max_gs;
383 	int			offset;
384 	int			wqe_shift;
385 	unsigned		head;
386 	unsigned		tail;
387 	u16			cur_post;
388 	u16			last_poll;
389 	void			*cur_edge;
390 };
391 
392 enum mlx5_ib_wq_flags {
393 	MLX5_IB_WQ_FLAGS_DELAY_DROP = 0x1,
394 	MLX5_IB_WQ_FLAGS_STRIDING_RQ = 0x2,
395 };
396 
397 #define MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES 9
398 #define MLX5_MAX_SINGLE_WQE_LOG_NUM_STRIDES 16
399 #define MLX5_MIN_SINGLE_STRIDE_LOG_NUM_BYTES 6
400 #define MLX5_MAX_SINGLE_STRIDE_LOG_NUM_BYTES 13
401 #define MLX5_EXT_MIN_SINGLE_WQE_LOG_NUM_STRIDES 3
402 
403 struct mlx5_ib_rwq {
404 	struct ib_wq		ibwq;
405 	struct mlx5_core_qp	core_qp;
406 	u32			rq_num_pas;
407 	u32			log_rq_stride;
408 	u32			log_rq_size;
409 	u32			rq_page_offset;
410 	u32			log_page_size;
411 	u32			log_num_strides;
412 	u32			two_byte_shift_en;
413 	u32			single_stride_log_num_of_bytes;
414 	struct ib_umem		*umem;
415 	size_t			buf_size;
416 	unsigned int		page_shift;
417 	struct mlx5_db		db;
418 	u32			user_index;
419 	u32			wqe_count;
420 	u32			wqe_shift;
421 	int			wq_sig;
422 	u32			create_flags; /* Use enum mlx5_ib_wq_flags */
423 };
424 
425 struct mlx5_ib_rwq_ind_table {
426 	struct ib_rwq_ind_table ib_rwq_ind_tbl;
427 	u32			rqtn;
428 	u16			uid;
429 };
430 
431 struct mlx5_ib_ubuffer {
432 	struct ib_umem	       *umem;
433 	int			buf_size;
434 	u64			buf_addr;
435 };
436 
437 struct mlx5_ib_qp_base {
438 	struct mlx5_ib_qp	*container_mibqp;
439 	struct mlx5_core_qp	mqp;
440 	struct mlx5_ib_ubuffer	ubuffer;
441 };
442 
443 struct mlx5_ib_qp_trans {
444 	struct mlx5_ib_qp_base	base;
445 	u16			xrcdn;
446 	u32			alt_port;
447 	u8			atomic_rd_en;
448 	u8			resp_depth;
449 };
450 
451 struct mlx5_ib_rss_qp {
452 	u32	tirn;
453 };
454 
455 struct mlx5_ib_rq {
456 	struct mlx5_ib_qp_base base;
457 	struct mlx5_ib_wq	*rq;
458 	struct mlx5_ib_ubuffer	ubuffer;
459 	struct mlx5_db		*doorbell;
460 	u32			tirn;
461 	u8			state;
462 	u32			flags;
463 };
464 
465 struct mlx5_ib_sq {
466 	struct mlx5_ib_qp_base base;
467 	struct mlx5_ib_wq	*sq;
468 	struct mlx5_ib_ubuffer  ubuffer;
469 	struct mlx5_db		*doorbell;
470 	struct mlx5_flow_handle	*flow_rule;
471 	u32			tisn;
472 	u8			state;
473 };
474 
475 struct mlx5_ib_raw_packet_qp {
476 	struct mlx5_ib_sq sq;
477 	struct mlx5_ib_rq rq;
478 };
479 
480 struct mlx5_bf {
481 	int			buf_size;
482 	unsigned long		offset;
483 	struct mlx5_sq_bfreg   *bfreg;
484 };
485 
486 struct mlx5_ib_dct {
487 	struct mlx5_core_dct    mdct;
488 	u32                     *in;
489 };
490 
491 struct mlx5_ib_gsi_qp {
492 	struct ib_qp *rx_qp;
493 	u32 port_num;
494 	struct ib_qp_cap cap;
495 	struct ib_cq *cq;
496 	struct mlx5_ib_gsi_wr *outstanding_wrs;
497 	u32 outstanding_pi, outstanding_ci;
498 	int num_qps;
499 	/* Protects access to the tx_qps. Post send operations synchronize
500 	 * with tx_qp creation in setup_qp(). Also protects the
501 	 * outstanding_wrs array and indices.
502 	 */
503 	spinlock_t lock;
504 	struct ib_qp **tx_qps;
505 };
506 
507 struct mlx5_ib_qp {
508 	struct ib_qp		ibqp;
509 	union {
510 		struct mlx5_ib_qp_trans trans_qp;
511 		struct mlx5_ib_raw_packet_qp raw_packet_qp;
512 		struct mlx5_ib_rss_qp rss_qp;
513 		struct mlx5_ib_dct dct;
514 		struct mlx5_ib_gsi_qp gsi;
515 	};
516 	struct mlx5_frag_buf	buf;
517 
518 	struct mlx5_db		db;
519 	struct mlx5_ib_wq	rq;
520 
521 	u8			sq_signal_bits;
522 	u8			next_fence;
523 	struct mlx5_ib_wq	sq;
524 
525 	/* serialize qp state modifications
526 	 */
527 	struct mutex		mutex;
528 	/* cached variant of create_flags from struct ib_qp_init_attr */
529 	u32			flags;
530 	u32			port;
531 	u8			state;
532 	int			max_inline_data;
533 	struct mlx5_bf	        bf;
534 	u8			has_rq:1;
535 	u8			is_rss:1;
536 
537 	/* only for user space QPs. For kernel
538 	 * we have it from the bf object
539 	 */
540 	int			bfregn;
541 
542 	struct list_head	qps_list;
543 	struct list_head	cq_recv_list;
544 	struct list_head	cq_send_list;
545 	struct mlx5_rate_limit	rl;
546 	u32                     underlay_qpn;
547 	u32			flags_en;
548 	/*
549 	 * IB/core doesn't store low-level QP types, so
550 	 * store both MLX and IBTA types in the field below.
551 	 */
552 	enum ib_qp_type		type;
553 	/* A flag to indicate if there's a new counter is configured
554 	 * but not take effective
555 	 */
556 	u32                     counter_pending;
557 	u16			gsi_lag_port;
558 };
559 
560 struct mlx5_ib_cq_buf {
561 	struct mlx5_frag_buf_ctrl fbc;
562 	struct mlx5_frag_buf    frag_buf;
563 	struct ib_umem		*umem;
564 	int			cqe_size;
565 	int			nent;
566 };
567 
568 enum mlx5_ib_cq_pr_flags {
569 	MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD	= 1 << 0,
570 	MLX5_IB_CQ_PR_FLAGS_REAL_TIME_TS = 1 << 1,
571 };
572 
573 struct mlx5_ib_cq {
574 	struct ib_cq		ibcq;
575 	struct mlx5_core_cq	mcq;
576 	struct mlx5_ib_cq_buf	buf;
577 	struct mlx5_db		db;
578 
579 	/* serialize access to the CQ
580 	 */
581 	spinlock_t		lock;
582 
583 	/* protect resize cq
584 	 */
585 	struct mutex		resize_mutex;
586 	struct mlx5_ib_cq_buf  *resize_buf;
587 	struct ib_umem	       *resize_umem;
588 	int			cqe_size;
589 	struct list_head	list_send_qp;
590 	struct list_head	list_recv_qp;
591 	u32			create_flags;
592 	struct list_head	wc_list;
593 	enum ib_cq_notify_flags notify_flags;
594 	struct work_struct	notify_work;
595 	u16			private_flags; /* Use mlx5_ib_cq_pr_flags */
596 };
597 
598 struct mlx5_ib_wc {
599 	struct ib_wc wc;
600 	struct list_head list;
601 };
602 
603 struct mlx5_ib_srq {
604 	struct ib_srq		ibsrq;
605 	struct mlx5_core_srq	msrq;
606 	struct mlx5_frag_buf	buf;
607 	struct mlx5_db		db;
608 	struct mlx5_frag_buf_ctrl fbc;
609 	u64		       *wrid;
610 	/* protect SRQ hanlding
611 	 */
612 	spinlock_t		lock;
613 	int			head;
614 	int			tail;
615 	u16			wqe_ctr;
616 	struct ib_umem	       *umem;
617 	/* serialize arming a SRQ
618 	 */
619 	struct mutex		mutex;
620 	int			wq_sig;
621 };
622 
623 struct mlx5_ib_xrcd {
624 	struct ib_xrcd		ibxrcd;
625 	u32			xrcdn;
626 };
627 
628 enum mlx5_ib_mtt_access_flags {
629 	MLX5_IB_MTT_READ  = (1 << 0),
630 	MLX5_IB_MTT_WRITE = (1 << 1),
631 };
632 
633 struct mlx5_user_mmap_entry {
634 	struct rdma_user_mmap_entry rdma_entry;
635 	u8 mmap_flag;
636 	u64 address;
637 	u32 page_idx;
638 };
639 
640 enum mlx5_mkey_type {
641 	MLX5_MKEY_MR = 1,
642 	MLX5_MKEY_MW,
643 	MLX5_MKEY_INDIRECT_DEVX,
644 };
645 
646 struct mlx5r_cache_rb_key {
647 	u8 ats:1;
648 	unsigned int access_mode;
649 	unsigned int access_flags;
650 	unsigned int ndescs;
651 };
652 
653 struct mlx5_ib_mkey {
654 	u32 key;
655 	enum mlx5_mkey_type type;
656 	unsigned int ndescs;
657 	struct wait_queue_head wait;
658 	refcount_t usecount;
659 	/* Cacheable user Mkey must hold either a rb_key or a cache_ent. */
660 	struct mlx5r_cache_rb_key rb_key;
661 	struct mlx5_cache_ent *cache_ent;
662 };
663 
664 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
665 
666 #define MLX5_IB_DM_MEMIC_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE   |\
667 					 IB_ACCESS_REMOTE_WRITE  |\
668 					 IB_ACCESS_REMOTE_READ   |\
669 					 IB_ACCESS_REMOTE_ATOMIC |\
670 					 IB_ZERO_BASED)
671 
672 #define MLX5_IB_DM_SW_ICM_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE   |\
673 					  IB_ACCESS_REMOTE_WRITE  |\
674 					  IB_ACCESS_REMOTE_READ   |\
675 					  IB_ZERO_BASED)
676 
677 #define mlx5_update_odp_stats(mr, counter_name, value)		\
678 	atomic64_add(value, &((mr)->odp_stats.counter_name))
679 
680 struct mlx5_ib_mr {
681 	struct ib_mr ibmr;
682 	struct mlx5_ib_mkey mmkey;
683 
684 	struct ib_umem *umem;
685 
686 	union {
687 		/* Used only by kernel MRs (umem == NULL) */
688 		struct {
689 			void *descs;
690 			void *descs_alloc;
691 			dma_addr_t desc_map;
692 			int max_descs;
693 			int desc_size;
694 			int access_mode;
695 
696 			/* For Kernel IB_MR_TYPE_INTEGRITY */
697 			struct mlx5_core_sig_ctx *sig;
698 			struct mlx5_ib_mr *pi_mr;
699 			struct mlx5_ib_mr *klm_mr;
700 			struct mlx5_ib_mr *mtt_mr;
701 			u64 data_iova;
702 			u64 pi_iova;
703 			int meta_ndescs;
704 			int meta_length;
705 			int data_length;
706 		};
707 
708 		/* Used only by User MRs (umem != NULL) */
709 		struct {
710 			unsigned int page_shift;
711 			/* Current access_flags */
712 			int access_flags;
713 
714 			/* For User ODP */
715 			struct mlx5_ib_mr *parent;
716 			struct xarray implicit_children;
717 			union {
718 				struct work_struct work;
719 			} odp_destroy;
720 			struct ib_odp_counters odp_stats;
721 			bool is_odp_implicit;
722 		};
723 	};
724 };
725 
726 static inline bool is_odp_mr(struct mlx5_ib_mr *mr)
727 {
728 	return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
729 	       mr->umem->is_odp;
730 }
731 
732 static inline bool is_dmabuf_mr(struct mlx5_ib_mr *mr)
733 {
734 	return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
735 	       mr->umem->is_dmabuf;
736 }
737 
738 struct mlx5_ib_mw {
739 	struct ib_mw		ibmw;
740 	struct mlx5_ib_mkey	mmkey;
741 };
742 
743 struct mlx5_ib_umr_context {
744 	struct ib_cqe		cqe;
745 	enum ib_wc_status	status;
746 	struct completion	done;
747 };
748 
749 enum {
750 	MLX5_UMR_STATE_UNINIT,
751 	MLX5_UMR_STATE_ACTIVE,
752 	MLX5_UMR_STATE_RECOVER,
753 	MLX5_UMR_STATE_ERR,
754 };
755 
756 struct umr_common {
757 	struct ib_pd	*pd;
758 	struct ib_cq	*cq;
759 	struct ib_qp	*qp;
760 	/* Protects from UMR QP overflow
761 	 */
762 	struct semaphore	sem;
763 	/* Protects from using UMR while the UMR is not active
764 	 */
765 	struct mutex lock;
766 	unsigned int state;
767 };
768 
769 struct mlx5_cache_ent {
770 	struct xarray		mkeys;
771 	unsigned long		stored;
772 	unsigned long		reserved;
773 
774 	char                    name[4];
775 
776 	struct rb_node		node;
777 	struct mlx5r_cache_rb_key rb_key;
778 
779 	u8 is_tmp:1;
780 	u8 disabled:1;
781 	u8 fill_to_high_water:1;
782 
783 	/*
784 	 * - limit is the low water mark for stored mkeys, 2* limit is the
785 	 *   upper water mark.
786 	 */
787 	u32 in_use;
788 	u32 limit;
789 
790 	/* Statistics */
791 	u32                     miss;
792 
793 	struct mlx5_ib_dev     *dev;
794 	struct delayed_work	dwork;
795 };
796 
797 struct mlx5r_async_create_mkey {
798 	union {
799 		u32 in[MLX5_ST_SZ_BYTES(create_mkey_in)];
800 		u32 out[MLX5_ST_SZ_DW(create_mkey_out)];
801 	};
802 	struct mlx5_async_work cb_work;
803 	struct mlx5_cache_ent *ent;
804 	u32 mkey;
805 };
806 
807 struct mlx5_mkey_cache {
808 	struct workqueue_struct *wq;
809 	struct rb_root		rb_root;
810 	struct mutex		rb_lock;
811 	struct dentry		*fs_root;
812 	unsigned long		last_add;
813 	struct delayed_work	remove_ent_dwork;
814 };
815 
816 struct mlx5_ib_port_resources {
817 	struct mlx5_ib_gsi_qp *gsi;
818 	struct work_struct pkey_change_work;
819 };
820 
821 struct mlx5_ib_resources {
822 	struct ib_cq	*c0;
823 	u32 xrcdn0;
824 	u32 xrcdn1;
825 	struct ib_pd	*p0;
826 	struct ib_srq	*s0;
827 	struct ib_srq	*s1;
828 	struct mlx5_ib_port_resources ports[2];
829 };
830 
831 #define MAX_OPFC_RULES 2
832 
833 struct mlx5_ib_op_fc {
834 	struct mlx5_fc *fc;
835 	struct mlx5_flow_handle *rule[MAX_OPFC_RULES];
836 };
837 
838 struct mlx5_ib_counters {
839 	struct rdma_stat_desc *descs;
840 	size_t *offsets;
841 	u32 num_q_counters;
842 	u32 num_cong_counters;
843 	u32 num_ext_ppcnt_counters;
844 	u32 num_op_counters;
845 	u16 set_id;
846 	struct mlx5_ib_op_fc opfcs[MLX5_IB_OPCOUNTER_MAX];
847 };
848 
849 int mlx5_ib_fs_add_op_fc(struct mlx5_ib_dev *dev, u32 port_num,
850 			 struct mlx5_ib_op_fc *opfc,
851 			 enum mlx5_ib_optional_counter_type type);
852 
853 void mlx5_ib_fs_remove_op_fc(struct mlx5_ib_dev *dev,
854 			     struct mlx5_ib_op_fc *opfc,
855 			     enum mlx5_ib_optional_counter_type type);
856 
857 struct mlx5_ib_multiport_info;
858 
859 struct mlx5_ib_multiport {
860 	struct mlx5_ib_multiport_info *mpi;
861 	/* To be held when accessing the multiport info */
862 	spinlock_t mpi_lock;
863 };
864 
865 struct mlx5_roce {
866 	/* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL
867 	 * netdev pointer
868 	 */
869 	rwlock_t		netdev_lock;
870 	struct net_device	*netdev;
871 	struct notifier_block	nb;
872 	struct netdev_net_notifier nn;
873 	struct notifier_block	mdev_nb;
874 	struct net_device	*tracking_netdev;
875 	atomic_t		tx_port_affinity;
876 	enum ib_port_state last_port_state;
877 	struct mlx5_ib_dev	*dev;
878 	u32			native_port_num;
879 };
880 
881 struct mlx5_ib_port {
882 	struct mlx5_ib_counters cnts;
883 	struct mlx5_ib_multiport mp;
884 	struct mlx5_ib_dbg_cc_params *dbg_cc_params;
885 	struct mlx5_roce roce;
886 	struct mlx5_eswitch_rep		*rep;
887 #ifdef CONFIG_MLX5_MACSEC
888 	struct mlx5_reserved_gids *reserved_gids;
889 #endif
890 };
891 
892 struct mlx5_ib_dbg_param {
893 	int			offset;
894 	struct mlx5_ib_dev	*dev;
895 	struct dentry		*dentry;
896 	u32			port_num;
897 };
898 
899 enum mlx5_ib_dbg_cc_types {
900 	MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE,
901 	MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE_ATI,
902 	MLX5_IB_DBG_CC_RP_TIME_RESET,
903 	MLX5_IB_DBG_CC_RP_BYTE_RESET,
904 	MLX5_IB_DBG_CC_RP_THRESHOLD,
905 	MLX5_IB_DBG_CC_RP_AI_RATE,
906 	MLX5_IB_DBG_CC_RP_MAX_RATE,
907 	MLX5_IB_DBG_CC_RP_HAI_RATE,
908 	MLX5_IB_DBG_CC_RP_MIN_DEC_FAC,
909 	MLX5_IB_DBG_CC_RP_MIN_RATE,
910 	MLX5_IB_DBG_CC_RP_RATE_TO_SET_ON_FIRST_CNP,
911 	MLX5_IB_DBG_CC_RP_DCE_TCP_G,
912 	MLX5_IB_DBG_CC_RP_DCE_TCP_RTT,
913 	MLX5_IB_DBG_CC_RP_RATE_REDUCE_MONITOR_PERIOD,
914 	MLX5_IB_DBG_CC_RP_INITIAL_ALPHA_VALUE,
915 	MLX5_IB_DBG_CC_RP_GD,
916 	MLX5_IB_DBG_CC_NP_MIN_TIME_BETWEEN_CNPS,
917 	MLX5_IB_DBG_CC_NP_CNP_DSCP,
918 	MLX5_IB_DBG_CC_NP_CNP_PRIO_MODE,
919 	MLX5_IB_DBG_CC_NP_CNP_PRIO,
920 	MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP_VALID,
921 	MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP,
922 	MLX5_IB_DBG_CC_MAX,
923 };
924 
925 struct mlx5_ib_dbg_cc_params {
926 	struct dentry			*root;
927 	struct mlx5_ib_dbg_param	params[MLX5_IB_DBG_CC_MAX];
928 };
929 
930 enum {
931 	MLX5_MAX_DELAY_DROP_TIMEOUT_MS = 100,
932 };
933 
934 struct mlx5_ib_delay_drop {
935 	struct mlx5_ib_dev     *dev;
936 	struct work_struct	delay_drop_work;
937 	/* serialize setting of delay drop */
938 	struct mutex		lock;
939 	u32			timeout;
940 	bool			activate;
941 	atomic_t		events_cnt;
942 	atomic_t		rqs_cnt;
943 	struct dentry		*dir_debugfs;
944 };
945 
946 enum mlx5_ib_stages {
947 	MLX5_IB_STAGE_INIT,
948 	MLX5_IB_STAGE_FS,
949 	MLX5_IB_STAGE_CAPS,
950 	MLX5_IB_STAGE_NON_DEFAULT_CB,
951 	MLX5_IB_STAGE_ROCE,
952 	MLX5_IB_STAGE_QP,
953 	MLX5_IB_STAGE_SRQ,
954 	MLX5_IB_STAGE_DEVICE_RESOURCES,
955 	MLX5_IB_STAGE_DEVICE_NOTIFIER,
956 	MLX5_IB_STAGE_ODP,
957 	MLX5_IB_STAGE_COUNTERS,
958 	MLX5_IB_STAGE_CONG_DEBUGFS,
959 	MLX5_IB_STAGE_UAR,
960 	MLX5_IB_STAGE_BFREG,
961 	MLX5_IB_STAGE_PRE_IB_REG_UMR,
962 	MLX5_IB_STAGE_WHITELIST_UID,
963 	MLX5_IB_STAGE_IB_REG,
964 	MLX5_IB_STAGE_POST_IB_REG_UMR,
965 	MLX5_IB_STAGE_DELAY_DROP,
966 	MLX5_IB_STAGE_RESTRACK,
967 	MLX5_IB_STAGE_MAX,
968 };
969 
970 struct mlx5_ib_stage {
971 	int (*init)(struct mlx5_ib_dev *dev);
972 	void (*cleanup)(struct mlx5_ib_dev *dev);
973 };
974 
975 #define STAGE_CREATE(_stage, _init, _cleanup) \
976 	.stage[_stage] = {.init = _init, .cleanup = _cleanup}
977 
978 struct mlx5_ib_profile {
979 	struct mlx5_ib_stage stage[MLX5_IB_STAGE_MAX];
980 };
981 
982 struct mlx5_ib_multiport_info {
983 	struct list_head list;
984 	struct mlx5_ib_dev *ibdev;
985 	struct mlx5_core_dev *mdev;
986 	struct notifier_block mdev_events;
987 	struct completion unref_comp;
988 	u64 sys_image_guid;
989 	u32 mdev_refcnt;
990 	bool is_master;
991 	bool unaffiliate;
992 };
993 
994 struct mlx5_ib_flow_action {
995 	struct ib_flow_action		ib_action;
996 	union {
997 		struct {
998 			u64			    ib_flags;
999 			struct mlx5_accel_esp_xfrm *ctx;
1000 		} esp_aes_gcm;
1001 		struct {
1002 			struct mlx5_ib_dev *dev;
1003 			u32 sub_type;
1004 			union {
1005 				struct mlx5_modify_hdr *modify_hdr;
1006 				struct mlx5_pkt_reformat *pkt_reformat;
1007 			};
1008 		} flow_action_raw;
1009 	};
1010 };
1011 
1012 struct mlx5_dm {
1013 	struct mlx5_core_dev *dev;
1014 	/* This lock is used to protect the access to the shared
1015 	 * allocation map when concurrent requests by different
1016 	 * processes are handled.
1017 	 */
1018 	spinlock_t lock;
1019 	DECLARE_BITMAP(memic_alloc_pages, MLX5_MAX_MEMIC_PAGES);
1020 };
1021 
1022 struct mlx5_read_counters_attr {
1023 	struct mlx5_fc *hw_cntrs_hndl;
1024 	u64 *out;
1025 	u32 flags;
1026 };
1027 
1028 enum mlx5_ib_counters_type {
1029 	MLX5_IB_COUNTERS_FLOW,
1030 };
1031 
1032 struct mlx5_ib_mcounters {
1033 	struct ib_counters ibcntrs;
1034 	enum mlx5_ib_counters_type type;
1035 	/* number of counters supported for this counters type */
1036 	u32 counters_num;
1037 	struct mlx5_fc *hw_cntrs_hndl;
1038 	/* read function for this counters type */
1039 	int (*read_counters)(struct ib_device *ibdev,
1040 			     struct mlx5_read_counters_attr *read_attr);
1041 	/* max index set as part of create_flow */
1042 	u32 cntrs_max_index;
1043 	/* number of counters data entries (<description,index> pair) */
1044 	u32 ncounters;
1045 	/* counters data array for descriptions and indexes */
1046 	struct mlx5_ib_flow_counters_desc *counters_data;
1047 	/* protects access to mcounters internal data */
1048 	struct mutex mcntrs_mutex;
1049 };
1050 
1051 static inline struct mlx5_ib_mcounters *
1052 to_mcounters(struct ib_counters *ibcntrs)
1053 {
1054 	return container_of(ibcntrs, struct mlx5_ib_mcounters, ibcntrs);
1055 }
1056 
1057 int parse_flow_flow_action(struct mlx5_ib_flow_action *maction,
1058 			   bool is_egress,
1059 			   struct mlx5_flow_act *action);
1060 struct mlx5_ib_lb_state {
1061 	/* protect the user_td */
1062 	struct mutex		mutex;
1063 	u32			user_td;
1064 	int			qps;
1065 	bool			enabled;
1066 };
1067 
1068 struct mlx5_ib_pf_eq {
1069 	struct notifier_block irq_nb;
1070 	struct mlx5_ib_dev *dev;
1071 	struct mlx5_eq *core;
1072 	struct work_struct work;
1073 	spinlock_t lock; /* Pagefaults spinlock */
1074 	struct workqueue_struct *wq;
1075 	mempool_t *pool;
1076 };
1077 
1078 struct mlx5_devx_event_table {
1079 	struct mlx5_nb devx_nb;
1080 	/* serialize updating the event_xa */
1081 	struct mutex event_xa_lock;
1082 	struct xarray event_xa;
1083 };
1084 
1085 struct mlx5_var_table {
1086 	/* serialize updating the bitmap */
1087 	struct mutex bitmap_lock;
1088 	unsigned long *bitmap;
1089 	u64 hw_start_addr;
1090 	u32 stride_size;
1091 	u64 num_var_hw_entries;
1092 };
1093 
1094 struct mlx5_port_caps {
1095 	bool has_smi;
1096 	u8 ext_port_cap;
1097 };
1098 
1099 
1100 struct mlx5_special_mkeys {
1101 	u32 dump_fill_mkey;
1102 	__be32 null_mkey;
1103 	__be32 terminate_scatter_list_mkey;
1104 };
1105 
1106 struct mlx5_macsec {
1107 	struct mutex lock; /* Protects mlx5_macsec internal contexts */
1108 	struct list_head macsec_devices_list;
1109 	struct notifier_block blocking_events_nb;
1110 };
1111 
1112 struct mlx5_ib_dev {
1113 	struct ib_device		ib_dev;
1114 	struct mlx5_core_dev		*mdev;
1115 	struct notifier_block		mdev_events;
1116 	int				num_ports;
1117 	/* serialize update of capability mask
1118 	 */
1119 	struct mutex			cap_mask_mutex;
1120 	u8				ib_active:1;
1121 	u8				is_rep:1;
1122 	u8				lag_active:1;
1123 	u8				wc_support:1;
1124 	u8				fill_delay;
1125 	struct umr_common		umrc;
1126 	/* sync used page count stats
1127 	 */
1128 	struct mlx5_ib_resources	devr;
1129 
1130 	atomic_t			mkey_var;
1131 	struct mlx5_mkey_cache		cache;
1132 	struct timer_list		delay_timer;
1133 	/* Prevents soft lock on massive reg MRs */
1134 	struct mutex			slow_path_mutex;
1135 	struct ib_odp_caps	odp_caps;
1136 	u64			odp_max_size;
1137 	struct mutex		odp_eq_mutex;
1138 	struct mlx5_ib_pf_eq	odp_pf_eq;
1139 
1140 	struct xarray		odp_mkeys;
1141 
1142 	struct mlx5_ib_flow_db	*flow_db;
1143 	/* protect resources needed as part of reset flow */
1144 	spinlock_t		reset_flow_resource_lock;
1145 	struct list_head	qp_list;
1146 	/* Array with num_ports elements */
1147 	struct mlx5_ib_port	*port;
1148 	struct mlx5_sq_bfreg	bfreg;
1149 	struct mlx5_sq_bfreg	wc_bfreg;
1150 	struct mlx5_sq_bfreg	fp_bfreg;
1151 	struct mlx5_ib_delay_drop	delay_drop;
1152 	const struct mlx5_ib_profile	*profile;
1153 
1154 	struct mlx5_ib_lb_state		lb;
1155 	u8			umr_fence;
1156 	struct list_head	ib_dev_list;
1157 	u64			sys_image_guid;
1158 	struct mlx5_dm		dm;
1159 	u16			devx_whitelist_uid;
1160 	struct mlx5_srq_table   srq_table;
1161 	struct mlx5_qp_table    qp_table;
1162 	struct mlx5_async_ctx   async_ctx;
1163 	struct mlx5_devx_event_table devx_event_table;
1164 	struct mlx5_var_table var_table;
1165 
1166 	struct xarray sig_mrs;
1167 	struct mlx5_port_caps port_caps[MLX5_MAX_PORTS];
1168 	u16 pkey_table_len;
1169 	u8 lag_ports;
1170 	struct mlx5_special_mkeys mkeys;
1171 
1172 #ifdef CONFIG_MLX5_MACSEC
1173 	struct mlx5_macsec macsec;
1174 #endif
1175 };
1176 
1177 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
1178 {
1179 	return container_of(mcq, struct mlx5_ib_cq, mcq);
1180 }
1181 
1182 static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
1183 {
1184 	return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd);
1185 }
1186 
1187 static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev)
1188 {
1189 	return container_of(ibdev, struct mlx5_ib_dev, ib_dev);
1190 }
1191 
1192 static inline struct mlx5_ib_dev *mr_to_mdev(struct mlx5_ib_mr *mr)
1193 {
1194 	return to_mdev(mr->ibmr.device);
1195 }
1196 
1197 static inline struct mlx5_ib_dev *mlx5_udata_to_mdev(struct ib_udata *udata)
1198 {
1199 	struct mlx5_ib_ucontext *context = rdma_udata_to_drv_context(
1200 		udata, struct mlx5_ib_ucontext, ibucontext);
1201 
1202 	return to_mdev(context->ibucontext.device);
1203 }
1204 
1205 static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq)
1206 {
1207 	return container_of(ibcq, struct mlx5_ib_cq, ibcq);
1208 }
1209 
1210 static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp)
1211 {
1212 	return container_of(mqp, struct mlx5_ib_qp_base, mqp)->container_mibqp;
1213 }
1214 
1215 static inline struct mlx5_ib_rwq *to_mibrwq(struct mlx5_core_qp *core_qp)
1216 {
1217 	return container_of(core_qp, struct mlx5_ib_rwq, core_qp);
1218 }
1219 
1220 static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd)
1221 {
1222 	return container_of(ibpd, struct mlx5_ib_pd, ibpd);
1223 }
1224 
1225 static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq)
1226 {
1227 	return container_of(ibsrq, struct mlx5_ib_srq, ibsrq);
1228 }
1229 
1230 static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp)
1231 {
1232 	return container_of(ibqp, struct mlx5_ib_qp, ibqp);
1233 }
1234 
1235 static inline struct mlx5_ib_rwq *to_mrwq(struct ib_wq *ibwq)
1236 {
1237 	return container_of(ibwq, struct mlx5_ib_rwq, ibwq);
1238 }
1239 
1240 static inline struct mlx5_ib_rwq_ind_table *to_mrwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl)
1241 {
1242 	return container_of(ib_rwq_ind_tbl, struct mlx5_ib_rwq_ind_table, ib_rwq_ind_tbl);
1243 }
1244 
1245 static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
1246 {
1247 	return container_of(msrq, struct mlx5_ib_srq, msrq);
1248 }
1249 
1250 static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
1251 {
1252 	return container_of(ibmr, struct mlx5_ib_mr, ibmr);
1253 }
1254 
1255 static inline struct mlx5_ib_mw *to_mmw(struct ib_mw *ibmw)
1256 {
1257 	return container_of(ibmw, struct mlx5_ib_mw, ibmw);
1258 }
1259 
1260 static inline struct mlx5_ib_flow_action *
1261 to_mflow_act(struct ib_flow_action *ibact)
1262 {
1263 	return container_of(ibact, struct mlx5_ib_flow_action, ib_action);
1264 }
1265 
1266 static inline struct mlx5_user_mmap_entry *
1267 to_mmmap(struct rdma_user_mmap_entry *rdma_entry)
1268 {
1269 	return container_of(rdma_entry,
1270 		struct mlx5_user_mmap_entry, rdma_entry);
1271 }
1272 
1273 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
1274 			struct mlx5_db *db);
1275 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);
1276 void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1277 void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1278 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index);
1279 int mlx5_ib_create_ah(struct ib_ah *ah, struct rdma_ah_init_attr *init_attr,
1280 		      struct ib_udata *udata);
1281 int mlx5_ib_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr);
1282 static inline int mlx5_ib_destroy_ah(struct ib_ah *ah, u32 flags)
1283 {
1284 	return 0;
1285 }
1286 int mlx5_ib_create_srq(struct ib_srq *srq, struct ib_srq_init_attr *init_attr,
1287 		       struct ib_udata *udata);
1288 int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
1289 		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
1290 int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr);
1291 int mlx5_ib_destroy_srq(struct ib_srq *srq, struct ib_udata *udata);
1292 int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
1293 			  const struct ib_recv_wr **bad_wr);
1294 int mlx5_ib_enable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1295 void mlx5_ib_disable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1296 int mlx5_ib_create_qp(struct ib_qp *qp, struct ib_qp_init_attr *init_attr,
1297 		      struct ib_udata *udata);
1298 int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1299 		      int attr_mask, struct ib_udata *udata);
1300 int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
1301 		     struct ib_qp_init_attr *qp_init_attr);
1302 int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata);
1303 void mlx5_ib_drain_sq(struct ib_qp *qp);
1304 void mlx5_ib_drain_rq(struct ib_qp *qp);
1305 int mlx5_ib_read_wqe_sq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1306 			size_t buflen, size_t *bc);
1307 int mlx5_ib_read_wqe_rq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1308 			size_t buflen, size_t *bc);
1309 int mlx5_ib_read_wqe_srq(struct mlx5_ib_srq *srq, int wqe_index, void *buffer,
1310 			 size_t buflen, size_t *bc);
1311 int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
1312 		      struct ib_udata *udata);
1313 int mlx5_ib_destroy_cq(struct ib_cq *cq, struct ib_udata *udata);
1314 int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
1315 int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
1316 int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1317 int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata);
1318 struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc);
1319 struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
1320 				  u64 virt_addr, int access_flags,
1321 				  struct ib_udata *udata);
1322 struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start,
1323 					 u64 length, u64 virt_addr,
1324 					 int fd, int access_flags,
1325 					 struct ib_udata *udata);
1326 int mlx5_ib_advise_mr(struct ib_pd *pd,
1327 		      enum ib_uverbs_advise_mr_advice advice,
1328 		      u32 flags,
1329 		      struct ib_sge *sg_list,
1330 		      u32 num_sge,
1331 		      struct uverbs_attr_bundle *attrs);
1332 int mlx5_ib_alloc_mw(struct ib_mw *mw, struct ib_udata *udata);
1333 int mlx5_ib_dealloc_mw(struct ib_mw *mw);
1334 struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd,
1335 					     int access_flags);
1336 void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *mr);
1337 void mlx5_ib_free_odp_mr(struct mlx5_ib_mr *mr);
1338 struct ib_mr *mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start,
1339 				    u64 length, u64 virt_addr, int access_flags,
1340 				    struct ib_pd *pd, struct ib_udata *udata);
1341 int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata);
1342 struct ib_mr *mlx5_ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
1343 			       u32 max_num_sg);
1344 struct ib_mr *mlx5_ib_alloc_mr_integrity(struct ib_pd *pd,
1345 					 u32 max_num_sg,
1346 					 u32 max_num_meta_sg);
1347 int mlx5_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents,
1348 		      unsigned int *sg_offset);
1349 int mlx5_ib_map_mr_sg_pi(struct ib_mr *ibmr, struct scatterlist *data_sg,
1350 			 int data_sg_nents, unsigned int *data_sg_offset,
1351 			 struct scatterlist *meta_sg, int meta_sg_nents,
1352 			 unsigned int *meta_sg_offset);
1353 int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u32 port_num,
1354 			const struct ib_wc *in_wc, const struct ib_grh *in_grh,
1355 			const struct ib_mad *in, struct ib_mad *out,
1356 			size_t *out_mad_size, u16 *out_mad_pkey_index);
1357 int mlx5_ib_alloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1358 int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1359 int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, unsigned int port);
1360 int mlx5_query_mad_ifc_system_image_guid(struct ib_device *ibdev,
1361 					 __be64 *sys_image_guid);
1362 int mlx5_query_mad_ifc_max_pkeys(struct ib_device *ibdev,
1363 				 u16 *max_pkeys);
1364 int mlx5_query_mad_ifc_vendor_id(struct ib_device *ibdev,
1365 				 u32 *vendor_id);
1366 int mlx5_query_mad_ifc_node_desc(struct mlx5_ib_dev *dev, char *node_desc);
1367 int mlx5_query_mad_ifc_node_guid(struct mlx5_ib_dev *dev, __be64 *node_guid);
1368 int mlx5_query_mad_ifc_pkey(struct ib_device *ibdev, u32 port, u16 index,
1369 			    u16 *pkey);
1370 int mlx5_query_mad_ifc_gids(struct ib_device *ibdev, u32 port, int index,
1371 			    union ib_gid *gid);
1372 int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u32 port,
1373 			    struct ib_port_attr *props);
1374 int mlx5_ib_query_port(struct ib_device *ibdev, u32 port,
1375 		       struct ib_port_attr *props);
1376 void mlx5_ib_populate_pas(struct ib_umem *umem, size_t page_size, __be64 *pas,
1377 			  u64 access_flags);
1378 void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num);
1379 int mlx5_ib_get_cqe_size(struct ib_cq *ibcq);
1380 int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev);
1381 void mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev);
1382 struct mlx5_cache_ent *
1383 mlx5r_cache_create_ent_locked(struct mlx5_ib_dev *dev,
1384 			      struct mlx5r_cache_rb_key rb_key,
1385 			      bool persistent_entry);
1386 
1387 struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev,
1388 				       int access_flags, int access_mode,
1389 				       int ndescs);
1390 
1391 int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
1392 			    struct ib_mr_status *mr_status);
1393 struct ib_wq *mlx5_ib_create_wq(struct ib_pd *pd,
1394 				struct ib_wq_init_attr *init_attr,
1395 				struct ib_udata *udata);
1396 int mlx5_ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
1397 int mlx5_ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1398 		      u32 wq_attr_mask, struct ib_udata *udata);
1399 int mlx5_ib_create_rwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_table,
1400 				 struct ib_rwq_ind_table_init_attr *init_attr,
1401 				 struct ib_udata *udata);
1402 int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
1403 struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm,
1404 				struct ib_dm_mr_attr *attr,
1405 				struct uverbs_attr_bundle *attrs);
1406 
1407 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1408 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);
1409 int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq);
1410 void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev);
1411 int __init mlx5_ib_odp_init(void);
1412 void mlx5_ib_odp_cleanup(void);
1413 int mlx5_odp_init_mkey_cache(struct mlx5_ib_dev *dev);
1414 void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1415 			   struct mlx5_ib_mr *mr, int flags);
1416 
1417 int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1418 			       enum ib_uverbs_advise_mr_advice advice,
1419 			       u32 flags, struct ib_sge *sg_list, u32 num_sge);
1420 int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr);
1421 int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr);
1422 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1423 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; }
1424 static inline int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev,
1425 				      struct mlx5_ib_pf_eq *eq)
1426 {
1427 	return 0;
1428 }
1429 static inline void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev) {}
1430 static inline int mlx5_ib_odp_init(void) { return 0; }
1431 static inline void mlx5_ib_odp_cleanup(void)				    {}
1432 static inline int mlx5_odp_init_mkey_cache(struct mlx5_ib_dev *dev)
1433 {
1434 	return 0;
1435 }
1436 static inline void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1437 					 struct mlx5_ib_mr *mr, int flags) {}
1438 
1439 static inline int
1440 mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1441 			   enum ib_uverbs_advise_mr_advice advice, u32 flags,
1442 			   struct ib_sge *sg_list, u32 num_sge)
1443 {
1444 	return -EOPNOTSUPP;
1445 }
1446 static inline int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr)
1447 {
1448 	return -EOPNOTSUPP;
1449 }
1450 static inline int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr)
1451 {
1452 	return -EOPNOTSUPP;
1453 }
1454 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1455 
1456 extern const struct mmu_interval_notifier_ops mlx5_mn_ops;
1457 
1458 /* Needed for rep profile */
1459 void __mlx5_ib_remove(struct mlx5_ib_dev *dev,
1460 		      const struct mlx5_ib_profile *profile,
1461 		      int stage);
1462 int __mlx5_ib_add(struct mlx5_ib_dev *dev,
1463 		  const struct mlx5_ib_profile *profile);
1464 
1465 int mlx5_ib_get_vf_config(struct ib_device *device, int vf,
1466 			  u32 port, struct ifla_vf_info *info);
1467 int mlx5_ib_set_vf_link_state(struct ib_device *device, int vf,
1468 			      u32 port, int state);
1469 int mlx5_ib_get_vf_stats(struct ib_device *device, int vf,
1470 			 u32 port, struct ifla_vf_stats *stats);
1471 int mlx5_ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
1472 			struct ifla_vf_guid *node_guid,
1473 			struct ifla_vf_guid *port_guid);
1474 int mlx5_ib_set_vf_guid(struct ib_device *device, int vf, u32 port,
1475 			u64 guid, int type);
1476 
1477 __be16 mlx5_get_roce_udp_sport_min(const struct mlx5_ib_dev *dev,
1478 				   const struct ib_gid_attr *attr);
1479 
1480 void mlx5_ib_cleanup_cong_debugfs(struct mlx5_ib_dev *dev, u32 port_num);
1481 void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u32 port_num);
1482 
1483 /* GSI QP helper functions */
1484 int mlx5_ib_create_gsi(struct ib_pd *pd, struct mlx5_ib_qp *mqp,
1485 		       struct ib_qp_init_attr *attr);
1486 int mlx5_ib_destroy_gsi(struct mlx5_ib_qp *mqp);
1487 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
1488 			  int attr_mask);
1489 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
1490 			 int qp_attr_mask,
1491 			 struct ib_qp_init_attr *qp_init_attr);
1492 int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
1493 			  const struct ib_send_wr **bad_wr);
1494 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, const struct ib_recv_wr *wr,
1495 			  const struct ib_recv_wr **bad_wr);
1496 void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi);
1497 
1498 int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc);
1499 
1500 void mlx5_ib_free_bfreg(struct mlx5_ib_dev *dev, struct mlx5_bfreg_info *bfregi,
1501 			int bfregn);
1502 struct mlx5_ib_dev *mlx5_ib_get_ibdev_from_mpi(struct mlx5_ib_multiport_info *mpi);
1503 struct mlx5_core_dev *mlx5_ib_get_native_port_mdev(struct mlx5_ib_dev *dev,
1504 						   u32 ib_port_num,
1505 						   u32 *native_port_num);
1506 void mlx5_ib_put_native_port_mdev(struct mlx5_ib_dev *dev,
1507 				  u32 port_num);
1508 
1509 extern const struct uapi_definition mlx5_ib_devx_defs[];
1510 extern const struct uapi_definition mlx5_ib_flow_defs[];
1511 extern const struct uapi_definition mlx5_ib_qos_defs[];
1512 extern const struct uapi_definition mlx5_ib_std_types_defs[];
1513 
1514 static inline int is_qp1(enum ib_qp_type qp_type)
1515 {
1516 	return qp_type == MLX5_IB_QPT_HW_GSI || qp_type == IB_QPT_GSI;
1517 }
1518 
1519 static inline u32 check_cq_create_flags(u32 flags)
1520 {
1521 	/*
1522 	 * It returns non-zero value for unsupported CQ
1523 	 * create flags, otherwise it returns zero.
1524 	 */
1525 	return (flags & ~(IB_UVERBS_CQ_FLAGS_IGNORE_OVERRUN |
1526 			  IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION));
1527 }
1528 
1529 static inline int verify_assign_uidx(u8 cqe_version, u32 cmd_uidx,
1530 				     u32 *user_index)
1531 {
1532 	if (cqe_version) {
1533 		if ((cmd_uidx == MLX5_IB_DEFAULT_UIDX) ||
1534 		    (cmd_uidx & ~MLX5_USER_ASSIGNED_UIDX_MASK))
1535 			return -EINVAL;
1536 		*user_index = cmd_uidx;
1537 	} else {
1538 		*user_index = MLX5_IB_DEFAULT_UIDX;
1539 	}
1540 
1541 	return 0;
1542 }
1543 
1544 static inline int get_qp_user_index(struct mlx5_ib_ucontext *ucontext,
1545 				    struct mlx5_ib_create_qp *ucmd,
1546 				    int inlen,
1547 				    u32 *user_index)
1548 {
1549 	u8 cqe_version = ucontext->cqe_version;
1550 
1551 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1552 	    (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1553 		return 0;
1554 
1555 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1556 		return -EINVAL;
1557 
1558 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1559 }
1560 
1561 static inline int get_srq_user_index(struct mlx5_ib_ucontext *ucontext,
1562 				     struct mlx5_ib_create_srq *ucmd,
1563 				     int inlen,
1564 				     u32 *user_index)
1565 {
1566 	u8 cqe_version = ucontext->cqe_version;
1567 
1568 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1569 	    (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1570 		return 0;
1571 
1572 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1573 		return -EINVAL;
1574 
1575 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1576 }
1577 
1578 static inline int get_uars_per_sys_page(struct mlx5_ib_dev *dev, bool lib_support)
1579 {
1580 	return lib_support && MLX5_CAP_GEN(dev->mdev, uar_4k) ?
1581 				MLX5_UARS_IN_PAGE : 1;
1582 }
1583 
1584 extern void *xlt_emergency_page;
1585 
1586 int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
1587 			struct mlx5_bfreg_info *bfregi, u32 bfregn,
1588 			bool dyn_bfreg);
1589 
1590 static inline int mlx5r_store_odp_mkey(struct mlx5_ib_dev *dev,
1591 				       struct mlx5_ib_mkey *mmkey)
1592 {
1593 	refcount_set(&mmkey->usecount, 1);
1594 
1595 	return xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(mmkey->key),
1596 			       mmkey, GFP_KERNEL));
1597 }
1598 
1599 /* deref an mkey that can participate in ODP flow */
1600 static inline void mlx5r_deref_odp_mkey(struct mlx5_ib_mkey *mmkey)
1601 {
1602 	if (refcount_dec_and_test(&mmkey->usecount))
1603 		wake_up(&mmkey->wait);
1604 }
1605 
1606 /* deref an mkey that can participate in ODP flow and wait for relese */
1607 static inline void mlx5r_deref_wait_odp_mkey(struct mlx5_ib_mkey *mmkey)
1608 {
1609 	mlx5r_deref_odp_mkey(mmkey);
1610 	wait_event(mmkey->wait, refcount_read(&mmkey->usecount) == 0);
1611 }
1612 
1613 int mlx5_ib_test_wc(struct mlx5_ib_dev *dev);
1614 
1615 static inline bool mlx5_ib_lag_should_assign_affinity(struct mlx5_ib_dev *dev)
1616 {
1617 	/*
1618 	 * If the driver is in hash mode and the port_select_flow_table_bypass cap
1619 	 * is supported, it means that the driver no longer needs to assign the port
1620 	 * affinity by default. If a user wants to set the port affinity explicitly,
1621 	 * the user has a dedicated API to do that, so there is no need to assign
1622 	 * the port affinity by default.
1623 	 */
1624 	if (dev->lag_active &&
1625 	    mlx5_lag_mode_is_hash(dev->mdev) &&
1626 	    MLX5_CAP_PORT_SELECTION(dev->mdev, port_select_flow_table_bypass))
1627 		return 0;
1628 
1629 	if (mlx5_lag_is_lacp_owner(dev->mdev) && !dev->lag_active)
1630 		return 0;
1631 
1632 	return dev->lag_active ||
1633 		(MLX5_CAP_GEN(dev->mdev, num_lag_ports) > 1 &&
1634 		 MLX5_CAP_GEN(dev->mdev, lag_tx_port_affinity));
1635 }
1636 
1637 static inline bool rt_supported(int ts_cap)
1638 {
1639 	return ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_REAL_TIME ||
1640 	       ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME;
1641 }
1642 
1643 /*
1644  * PCI Peer to Peer is a trainwreck. If no switch is present then things
1645  * sometimes work, depending on the pci_distance_p2p logic for excluding broken
1646  * root complexes. However if a switch is present in the path, then things get
1647  * really ugly depending on how the switch is setup. This table assumes that the
1648  * root complex is strict and is validating that all req/reps are matches
1649  * perfectly - so any scenario where it sees only half the transaction is a
1650  * failure.
1651  *
1652  * CR/RR/DT  ATS RO P2P
1653  * 00X       X   X  OK
1654  * 010       X   X  fails (request is routed to root but root never sees comp)
1655  * 011       0   X  fails (request is routed to root but root never sees comp)
1656  * 011       1   X  OK
1657  * 10X       X   1  OK
1658  * 101       X   0  fails (completion is routed to root but root didn't see req)
1659  * 110       X   0  SLOW
1660  * 111       0   0  SLOW
1661  * 111       1   0  fails (completion is routed to root but root didn't see req)
1662  * 111       1   1  OK
1663  *
1664  * Unfortunately we cannot reliably know if a switch is present or what the
1665  * CR/RR/DT ACS settings are, as in a VM that is all hidden. Assume that
1666  * CR/RR/DT is 111 if the ATS cap is enabled and follow the last three rows.
1667  *
1668  * For now assume if the umem is a dma_buf then it is P2P.
1669  */
1670 static inline bool mlx5_umem_needs_ats(struct mlx5_ib_dev *dev,
1671 				       struct ib_umem *umem, int access_flags)
1672 {
1673 	if (!MLX5_CAP_GEN(dev->mdev, ats) || !umem->is_dmabuf)
1674 		return false;
1675 	return access_flags & IB_ACCESS_RELAXED_ORDERING;
1676 }
1677 
1678 int set_roce_addr(struct mlx5_ib_dev *dev, u32 port_num,
1679 		  unsigned int index, const union ib_gid *gid,
1680 		  const struct ib_gid_attr *attr);
1681 #endif /* MLX5_IB_H */
1682