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 
29 #define mlx5_ib_dbg(_dev, format, arg...)                                      \
30 	dev_dbg(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,      \
31 		__LINE__, current->pid, ##arg)
32 
33 #define mlx5_ib_err(_dev, format, arg...)                                      \
34 	dev_err(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,      \
35 		__LINE__, current->pid, ##arg)
36 
37 #define mlx5_ib_warn(_dev, format, arg...)                                     \
38 	dev_warn(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,     \
39 		 __LINE__, current->pid, ##arg)
40 
41 #define MLX5_IB_DEFAULT_UIDX 0xffffff
42 #define MLX5_USER_ASSIGNED_UIDX_MASK __mlx5_mask(qpc, user_index)
43 
44 static __always_inline unsigned long
45 __mlx5_log_page_size_to_bitmap(unsigned int log_pgsz_bits,
46 			       unsigned int pgsz_shift)
47 {
48 	unsigned int largest_pg_shift =
49 		min_t(unsigned long, (1ULL << log_pgsz_bits) - 1 + pgsz_shift,
50 		      BITS_PER_LONG - 1);
51 
52 	/*
53 	 * Despite a command allowing it, the device does not support lower than
54 	 * 4k page size.
55 	 */
56 	pgsz_shift = max_t(unsigned int, MLX5_ADAPTER_PAGE_SHIFT, pgsz_shift);
57 	return GENMASK(largest_pg_shift, pgsz_shift);
58 }
59 
60 /*
61  * For mkc users, instead of a page_offset the command has a start_iova which
62  * specifies both the page_offset and the on-the-wire IOVA
63  */
64 #define mlx5_umem_find_best_pgsz(umem, typ, log_pgsz_fld, pgsz_shift, iova)    \
65 	ib_umem_find_best_pgsz(umem,                                           \
66 			       __mlx5_log_page_size_to_bitmap(                 \
67 				       __mlx5_bit_sz(typ, log_pgsz_fld),       \
68 				       pgsz_shift),                            \
69 			       iova)
70 
71 static __always_inline unsigned long
72 __mlx5_page_offset_to_bitmask(unsigned int page_offset_bits,
73 			      unsigned int offset_shift)
74 {
75 	unsigned int largest_offset_shift =
76 		min_t(unsigned long, page_offset_bits - 1 + offset_shift,
77 		      BITS_PER_LONG - 1);
78 
79 	return GENMASK(largest_offset_shift, offset_shift);
80 }
81 
82 /*
83  * QP/CQ/WQ/etc type commands take a page offset that satisifies:
84  *   page_offset_quantized * (page_size/scale) = page_offset
85  * Which restricts allowed page sizes to ones that satisify the above.
86  */
87 unsigned long __mlx5_umem_find_best_quantized_pgoff(
88 	struct ib_umem *umem, unsigned long pgsz_bitmap,
89 	unsigned int page_offset_bits, u64 pgoff_bitmask, unsigned int scale,
90 	unsigned int *page_offset_quantized);
91 #define mlx5_umem_find_best_quantized_pgoff(umem, typ, log_pgsz_fld,           \
92 					    pgsz_shift, page_offset_fld,       \
93 					    scale, page_offset_quantized)      \
94 	__mlx5_umem_find_best_quantized_pgoff(                                 \
95 		umem,                                                          \
96 		__mlx5_log_page_size_to_bitmap(                                \
97 			__mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift),         \
98 		__mlx5_bit_sz(typ, page_offset_fld),                           \
99 		GENMASK(31, order_base_2(scale)), scale,                       \
100 		page_offset_quantized)
101 
102 #define mlx5_umem_find_best_cq_quantized_pgoff(umem, typ, log_pgsz_fld,        \
103 					       pgsz_shift, page_offset_fld,    \
104 					       scale, page_offset_quantized)   \
105 	__mlx5_umem_find_best_quantized_pgoff(                                 \
106 		umem,                                                          \
107 		__mlx5_log_page_size_to_bitmap(                                \
108 			__mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift),         \
109 		__mlx5_bit_sz(typ, page_offset_fld), 0, scale,                 \
110 		page_offset_quantized)
111 
112 enum {
113 	MLX5_IB_MMAP_OFFSET_START = 9,
114 	MLX5_IB_MMAP_OFFSET_END = 255,
115 };
116 
117 enum {
118 	MLX5_IB_MMAP_CMD_SHIFT	= 8,
119 	MLX5_IB_MMAP_CMD_MASK	= 0xff,
120 };
121 
122 enum {
123 	MLX5_RES_SCAT_DATA32_CQE	= 0x1,
124 	MLX5_RES_SCAT_DATA64_CQE	= 0x2,
125 	MLX5_REQ_SCAT_DATA32_CQE	= 0x11,
126 	MLX5_REQ_SCAT_DATA64_CQE	= 0x22,
127 };
128 
129 enum mlx5_ib_mad_ifc_flags {
130 	MLX5_MAD_IFC_IGNORE_MKEY	= 1,
131 	MLX5_MAD_IFC_IGNORE_BKEY	= 2,
132 	MLX5_MAD_IFC_NET_VIEW		= 4,
133 };
134 
135 enum {
136 	MLX5_CROSS_CHANNEL_BFREG         = 0,
137 };
138 
139 enum {
140 	MLX5_CQE_VERSION_V0,
141 	MLX5_CQE_VERSION_V1,
142 };
143 
144 enum {
145 	MLX5_TM_MAX_RNDV_MSG_SIZE	= 64,
146 	MLX5_TM_MAX_SGE			= 1,
147 };
148 
149 enum {
150 	MLX5_IB_INVALID_UAR_INDEX	= BIT(31),
151 	MLX5_IB_INVALID_BFREG		= BIT(31),
152 };
153 
154 enum {
155 	MLX5_MAX_MEMIC_PAGES = 0x100,
156 	MLX5_MEMIC_ALLOC_SIZE_MASK = 0x3f,
157 };
158 
159 enum {
160 	MLX5_MEMIC_BASE_ALIGN	= 6,
161 	MLX5_MEMIC_BASE_SIZE	= 1 << MLX5_MEMIC_BASE_ALIGN,
162 };
163 
164 enum mlx5_ib_mmap_type {
165 	MLX5_IB_MMAP_TYPE_MEMIC = 1,
166 	MLX5_IB_MMAP_TYPE_VAR = 2,
167 	MLX5_IB_MMAP_TYPE_UAR_WC = 3,
168 	MLX5_IB_MMAP_TYPE_UAR_NC = 4,
169 	MLX5_IB_MMAP_TYPE_MEMIC_OP = 5,
170 };
171 
172 struct mlx5_bfreg_info {
173 	u32 *sys_pages;
174 	int num_low_latency_bfregs;
175 	unsigned int *count;
176 
177 	/*
178 	 * protect bfreg allocation data structs
179 	 */
180 	struct mutex lock;
181 	u32 ver;
182 	u8 lib_uar_4k : 1;
183 	u8 lib_uar_dyn : 1;
184 	u32 num_sys_pages;
185 	u32 num_static_sys_pages;
186 	u32 total_num_bfregs;
187 	u32 num_dyn_bfregs;
188 };
189 
190 struct mlx5_ib_ucontext {
191 	struct ib_ucontext	ibucontext;
192 	struct list_head	db_page_list;
193 
194 	/* protect doorbell record alloc/free
195 	 */
196 	struct mutex		db_page_mutex;
197 	struct mlx5_bfreg_info	bfregi;
198 	u8			cqe_version;
199 	/* Transport Domain number */
200 	u32			tdn;
201 
202 	u64			lib_caps;
203 	u16			devx_uid;
204 	/* For RoCE LAG TX affinity */
205 	atomic_t		tx_port_affinity;
206 };
207 
208 static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
209 {
210 	return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext);
211 }
212 
213 struct mlx5_ib_pd {
214 	struct ib_pd		ibpd;
215 	u32			pdn;
216 	u16			uid;
217 };
218 
219 enum {
220 	MLX5_IB_FLOW_ACTION_MODIFY_HEADER,
221 	MLX5_IB_FLOW_ACTION_PACKET_REFORMAT,
222 	MLX5_IB_FLOW_ACTION_DECAP,
223 };
224 
225 #define MLX5_IB_FLOW_MCAST_PRIO		(MLX5_BY_PASS_NUM_PRIOS - 1)
226 #define MLX5_IB_FLOW_LAST_PRIO		(MLX5_BY_PASS_NUM_REGULAR_PRIOS - 1)
227 #if (MLX5_IB_FLOW_LAST_PRIO <= 0)
228 #error "Invalid number of bypass priorities"
229 #endif
230 #define MLX5_IB_FLOW_LEFTOVERS_PRIO	(MLX5_IB_FLOW_MCAST_PRIO + 1)
231 
232 #define MLX5_IB_NUM_FLOW_FT		(MLX5_IB_FLOW_LEFTOVERS_PRIO + 1)
233 #define MLX5_IB_NUM_SNIFFER_FTS		2
234 #define MLX5_IB_NUM_EGRESS_FTS		1
235 #define MLX5_IB_NUM_FDB_FTS		MLX5_BY_PASS_NUM_REGULAR_PRIOS
236 struct mlx5_ib_flow_prio {
237 	struct mlx5_flow_table		*flow_table;
238 	unsigned int			refcount;
239 };
240 
241 struct mlx5_ib_flow_handler {
242 	struct list_head		list;
243 	struct ib_flow			ibflow;
244 	struct mlx5_ib_flow_prio	*prio;
245 	struct mlx5_flow_handle		*rule;
246 	struct ib_counters		*ibcounters;
247 	struct mlx5_ib_dev		*dev;
248 	struct mlx5_ib_flow_matcher	*flow_matcher;
249 };
250 
251 struct mlx5_ib_flow_matcher {
252 	struct mlx5_ib_match_params matcher_mask;
253 	int			mask_len;
254 	enum mlx5_ib_flow_type	flow_type;
255 	enum mlx5_flow_namespace_type ns_type;
256 	u16			priority;
257 	struct mlx5_core_dev	*mdev;
258 	atomic_t		usecnt;
259 	u8			match_criteria_enable;
260 };
261 
262 struct mlx5_ib_steering_anchor {
263 	struct mlx5_ib_flow_prio *ft_prio;
264 	struct mlx5_ib_dev *dev;
265 	atomic_t usecnt;
266 };
267 
268 struct mlx5_ib_pp {
269 	u16 index;
270 	struct mlx5_core_dev *mdev;
271 };
272 
273 enum mlx5_ib_optional_counter_type {
274 	MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS,
275 	MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS,
276 	MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS,
277 
278 	MLX5_IB_OPCOUNTER_MAX,
279 };
280 
281 struct mlx5_ib_flow_db {
282 	struct mlx5_ib_flow_prio	prios[MLX5_IB_NUM_FLOW_FT];
283 	struct mlx5_ib_flow_prio	egress_prios[MLX5_IB_NUM_FLOW_FT];
284 	struct mlx5_ib_flow_prio	sniffer[MLX5_IB_NUM_SNIFFER_FTS];
285 	struct mlx5_ib_flow_prio	egress[MLX5_IB_NUM_EGRESS_FTS];
286 	struct mlx5_ib_flow_prio	fdb[MLX5_IB_NUM_FDB_FTS];
287 	struct mlx5_ib_flow_prio	rdma_rx[MLX5_IB_NUM_FLOW_FT];
288 	struct mlx5_ib_flow_prio	rdma_tx[MLX5_IB_NUM_FLOW_FT];
289 	struct mlx5_ib_flow_prio	opfcs[MLX5_IB_OPCOUNTER_MAX];
290 	struct mlx5_flow_table		*lag_demux_ft;
291 	/* Protect flow steering bypass flow tables
292 	 * when add/del flow rules.
293 	 * only single add/removal of flow steering rule could be done
294 	 * simultaneously.
295 	 */
296 	struct mutex			lock;
297 };
298 
299 /* Use macros here so that don't have to duplicate
300  * enum ib_qp_type for low-level driver
301  */
302 
303 #define MLX5_IB_QPT_REG_UMR	IB_QPT_RESERVED1
304 /*
305  * IB_QPT_GSI creates the software wrapper around GSI, and MLX5_IB_QPT_HW_GSI
306  * creates the actual hardware QP.
307  */
308 #define MLX5_IB_QPT_HW_GSI	IB_QPT_RESERVED2
309 #define MLX5_IB_QPT_DCI		IB_QPT_RESERVED3
310 #define MLX5_IB_QPT_DCT		IB_QPT_RESERVED4
311 #define MLX5_IB_WR_UMR		IB_WR_RESERVED1
312 
313 #define MLX5_IB_UPD_XLT_ZAP	      BIT(0)
314 #define MLX5_IB_UPD_XLT_ENABLE	      BIT(1)
315 #define MLX5_IB_UPD_XLT_ATOMIC	      BIT(2)
316 #define MLX5_IB_UPD_XLT_ADDR	      BIT(3)
317 #define MLX5_IB_UPD_XLT_PD	      BIT(4)
318 #define MLX5_IB_UPD_XLT_ACCESS	      BIT(5)
319 #define MLX5_IB_UPD_XLT_INDIRECT      BIT(6)
320 
321 /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags.
322  *
323  * These flags are intended for internal use by the mlx5_ib driver, and they
324  * rely on the range reserved for that use in the ib_qp_create_flags enum.
325  */
326 #define MLX5_IB_QP_CREATE_SQPN_QP1	IB_QP_CREATE_RESERVED_START
327 #define MLX5_IB_QP_CREATE_WC_TEST	(IB_QP_CREATE_RESERVED_START << 1)
328 
329 struct wr_list {
330 	u16	opcode;
331 	u16	next;
332 };
333 
334 enum mlx5_ib_rq_flags {
335 	MLX5_IB_RQ_CVLAN_STRIPPING	= 1 << 0,
336 	MLX5_IB_RQ_PCI_WRITE_END_PADDING	= 1 << 1,
337 };
338 
339 struct mlx5_ib_wq {
340 	struct mlx5_frag_buf_ctrl fbc;
341 	u64		       *wrid;
342 	u32		       *wr_data;
343 	struct wr_list	       *w_list;
344 	unsigned	       *wqe_head;
345 	u16		        unsig_count;
346 
347 	/* serialize post to the work queue
348 	 */
349 	spinlock_t		lock;
350 	int			wqe_cnt;
351 	int			max_post;
352 	int			max_gs;
353 	int			offset;
354 	int			wqe_shift;
355 	unsigned		head;
356 	unsigned		tail;
357 	u16			cur_post;
358 	u16			last_poll;
359 	void			*cur_edge;
360 };
361 
362 enum mlx5_ib_wq_flags {
363 	MLX5_IB_WQ_FLAGS_DELAY_DROP = 0x1,
364 	MLX5_IB_WQ_FLAGS_STRIDING_RQ = 0x2,
365 };
366 
367 #define MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES 9
368 #define MLX5_MAX_SINGLE_WQE_LOG_NUM_STRIDES 16
369 #define MLX5_MIN_SINGLE_STRIDE_LOG_NUM_BYTES 6
370 #define MLX5_MAX_SINGLE_STRIDE_LOG_NUM_BYTES 13
371 #define MLX5_EXT_MIN_SINGLE_WQE_LOG_NUM_STRIDES 3
372 
373 struct mlx5_ib_rwq {
374 	struct ib_wq		ibwq;
375 	struct mlx5_core_qp	core_qp;
376 	u32			rq_num_pas;
377 	u32			log_rq_stride;
378 	u32			log_rq_size;
379 	u32			rq_page_offset;
380 	u32			log_page_size;
381 	u32			log_num_strides;
382 	u32			two_byte_shift_en;
383 	u32			single_stride_log_num_of_bytes;
384 	struct ib_umem		*umem;
385 	size_t			buf_size;
386 	unsigned int		page_shift;
387 	struct mlx5_db		db;
388 	u32			user_index;
389 	u32			wqe_count;
390 	u32			wqe_shift;
391 	int			wq_sig;
392 	u32			create_flags; /* Use enum mlx5_ib_wq_flags */
393 };
394 
395 struct mlx5_ib_rwq_ind_table {
396 	struct ib_rwq_ind_table ib_rwq_ind_tbl;
397 	u32			rqtn;
398 	u16			uid;
399 };
400 
401 struct mlx5_ib_ubuffer {
402 	struct ib_umem	       *umem;
403 	int			buf_size;
404 	u64			buf_addr;
405 };
406 
407 struct mlx5_ib_qp_base {
408 	struct mlx5_ib_qp	*container_mibqp;
409 	struct mlx5_core_qp	mqp;
410 	struct mlx5_ib_ubuffer	ubuffer;
411 };
412 
413 struct mlx5_ib_qp_trans {
414 	struct mlx5_ib_qp_base	base;
415 	u16			xrcdn;
416 	u32			alt_port;
417 	u8			atomic_rd_en;
418 	u8			resp_depth;
419 };
420 
421 struct mlx5_ib_rss_qp {
422 	u32	tirn;
423 };
424 
425 struct mlx5_ib_rq {
426 	struct mlx5_ib_qp_base base;
427 	struct mlx5_ib_wq	*rq;
428 	struct mlx5_ib_ubuffer	ubuffer;
429 	struct mlx5_db		*doorbell;
430 	u32			tirn;
431 	u8			state;
432 	u32			flags;
433 };
434 
435 struct mlx5_ib_sq {
436 	struct mlx5_ib_qp_base base;
437 	struct mlx5_ib_wq	*sq;
438 	struct mlx5_ib_ubuffer  ubuffer;
439 	struct mlx5_db		*doorbell;
440 	struct mlx5_flow_handle	*flow_rule;
441 	u32			tisn;
442 	u8			state;
443 };
444 
445 struct mlx5_ib_raw_packet_qp {
446 	struct mlx5_ib_sq sq;
447 	struct mlx5_ib_rq rq;
448 };
449 
450 struct mlx5_bf {
451 	int			buf_size;
452 	unsigned long		offset;
453 	struct mlx5_sq_bfreg   *bfreg;
454 };
455 
456 struct mlx5_ib_dct {
457 	struct mlx5_core_dct    mdct;
458 	u32                     *in;
459 };
460 
461 struct mlx5_ib_gsi_qp {
462 	struct ib_qp *rx_qp;
463 	u32 port_num;
464 	struct ib_qp_cap cap;
465 	struct ib_cq *cq;
466 	struct mlx5_ib_gsi_wr *outstanding_wrs;
467 	u32 outstanding_pi, outstanding_ci;
468 	int num_qps;
469 	/* Protects access to the tx_qps. Post send operations synchronize
470 	 * with tx_qp creation in setup_qp(). Also protects the
471 	 * outstanding_wrs array and indices.
472 	 */
473 	spinlock_t lock;
474 	struct ib_qp **tx_qps;
475 };
476 
477 struct mlx5_ib_qp {
478 	struct ib_qp		ibqp;
479 	union {
480 		struct mlx5_ib_qp_trans trans_qp;
481 		struct mlx5_ib_raw_packet_qp raw_packet_qp;
482 		struct mlx5_ib_rss_qp rss_qp;
483 		struct mlx5_ib_dct dct;
484 		struct mlx5_ib_gsi_qp gsi;
485 	};
486 	struct mlx5_frag_buf	buf;
487 
488 	struct mlx5_db		db;
489 	struct mlx5_ib_wq	rq;
490 
491 	u8			sq_signal_bits;
492 	u8			next_fence;
493 	struct mlx5_ib_wq	sq;
494 
495 	/* serialize qp state modifications
496 	 */
497 	struct mutex		mutex;
498 	/* cached variant of create_flags from struct ib_qp_init_attr */
499 	u32			flags;
500 	u32			port;
501 	u8			state;
502 	int			max_inline_data;
503 	struct mlx5_bf	        bf;
504 	u8			has_rq:1;
505 	u8			is_rss:1;
506 
507 	/* only for user space QPs. For kernel
508 	 * we have it from the bf object
509 	 */
510 	int			bfregn;
511 
512 	struct list_head	qps_list;
513 	struct list_head	cq_recv_list;
514 	struct list_head	cq_send_list;
515 	struct mlx5_rate_limit	rl;
516 	u32                     underlay_qpn;
517 	u32			flags_en;
518 	/*
519 	 * IB/core doesn't store low-level QP types, so
520 	 * store both MLX and IBTA types in the field below.
521 	 */
522 	enum ib_qp_type		type;
523 	/* A flag to indicate if there's a new counter is configured
524 	 * but not take effective
525 	 */
526 	u32                     counter_pending;
527 	u16			gsi_lag_port;
528 };
529 
530 struct mlx5_ib_cq_buf {
531 	struct mlx5_frag_buf_ctrl fbc;
532 	struct mlx5_frag_buf    frag_buf;
533 	struct ib_umem		*umem;
534 	int			cqe_size;
535 	int			nent;
536 };
537 
538 enum mlx5_ib_cq_pr_flags {
539 	MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD	= 1 << 0,
540 	MLX5_IB_CQ_PR_FLAGS_REAL_TIME_TS = 1 << 1,
541 };
542 
543 struct mlx5_ib_cq {
544 	struct ib_cq		ibcq;
545 	struct mlx5_core_cq	mcq;
546 	struct mlx5_ib_cq_buf	buf;
547 	struct mlx5_db		db;
548 
549 	/* serialize access to the CQ
550 	 */
551 	spinlock_t		lock;
552 
553 	/* protect resize cq
554 	 */
555 	struct mutex		resize_mutex;
556 	struct mlx5_ib_cq_buf  *resize_buf;
557 	struct ib_umem	       *resize_umem;
558 	int			cqe_size;
559 	struct list_head	list_send_qp;
560 	struct list_head	list_recv_qp;
561 	u32			create_flags;
562 	struct list_head	wc_list;
563 	enum ib_cq_notify_flags notify_flags;
564 	struct work_struct	notify_work;
565 	u16			private_flags; /* Use mlx5_ib_cq_pr_flags */
566 };
567 
568 struct mlx5_ib_wc {
569 	struct ib_wc wc;
570 	struct list_head list;
571 };
572 
573 struct mlx5_ib_srq {
574 	struct ib_srq		ibsrq;
575 	struct mlx5_core_srq	msrq;
576 	struct mlx5_frag_buf	buf;
577 	struct mlx5_db		db;
578 	struct mlx5_frag_buf_ctrl fbc;
579 	u64		       *wrid;
580 	/* protect SRQ hanlding
581 	 */
582 	spinlock_t		lock;
583 	int			head;
584 	int			tail;
585 	u16			wqe_ctr;
586 	struct ib_umem	       *umem;
587 	/* serialize arming a SRQ
588 	 */
589 	struct mutex		mutex;
590 	int			wq_sig;
591 };
592 
593 struct mlx5_ib_xrcd {
594 	struct ib_xrcd		ibxrcd;
595 	u32			xrcdn;
596 };
597 
598 enum mlx5_ib_mtt_access_flags {
599 	MLX5_IB_MTT_READ  = (1 << 0),
600 	MLX5_IB_MTT_WRITE = (1 << 1),
601 };
602 
603 struct mlx5_user_mmap_entry {
604 	struct rdma_user_mmap_entry rdma_entry;
605 	u8 mmap_flag;
606 	u64 address;
607 	u32 page_idx;
608 };
609 
610 enum mlx5_mkey_type {
611 	MLX5_MKEY_MR = 1,
612 	MLX5_MKEY_MW,
613 	MLX5_MKEY_INDIRECT_DEVX,
614 };
615 
616 struct mlx5_ib_mkey {
617 	u32 key;
618 	enum mlx5_mkey_type type;
619 	unsigned int ndescs;
620 	struct wait_queue_head wait;
621 	refcount_t usecount;
622 	struct mlx5_cache_ent *cache_ent;
623 };
624 
625 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
626 
627 #define MLX5_IB_DM_MEMIC_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE   |\
628 					 IB_ACCESS_REMOTE_WRITE  |\
629 					 IB_ACCESS_REMOTE_READ   |\
630 					 IB_ACCESS_REMOTE_ATOMIC |\
631 					 IB_ZERO_BASED)
632 
633 #define MLX5_IB_DM_SW_ICM_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE   |\
634 					  IB_ACCESS_REMOTE_WRITE  |\
635 					  IB_ACCESS_REMOTE_READ   |\
636 					  IB_ZERO_BASED)
637 
638 #define mlx5_update_odp_stats(mr, counter_name, value)		\
639 	atomic64_add(value, &((mr)->odp_stats.counter_name))
640 
641 struct mlx5_ib_mr {
642 	struct ib_mr ibmr;
643 	struct mlx5_ib_mkey mmkey;
644 
645 	struct ib_umem *umem;
646 
647 	union {
648 		/* Used only by kernel MRs (umem == NULL) */
649 		struct {
650 			void *descs;
651 			void *descs_alloc;
652 			dma_addr_t desc_map;
653 			int max_descs;
654 			int desc_size;
655 			int access_mode;
656 
657 			/* For Kernel IB_MR_TYPE_INTEGRITY */
658 			struct mlx5_core_sig_ctx *sig;
659 			struct mlx5_ib_mr *pi_mr;
660 			struct mlx5_ib_mr *klm_mr;
661 			struct mlx5_ib_mr *mtt_mr;
662 			u64 data_iova;
663 			u64 pi_iova;
664 			int meta_ndescs;
665 			int meta_length;
666 			int data_length;
667 		};
668 
669 		/* Used only by User MRs (umem != NULL) */
670 		struct {
671 			unsigned int page_shift;
672 			/* Current access_flags */
673 			int access_flags;
674 
675 			/* For User ODP */
676 			struct mlx5_ib_mr *parent;
677 			struct xarray implicit_children;
678 			union {
679 				struct work_struct work;
680 			} odp_destroy;
681 			struct ib_odp_counters odp_stats;
682 			bool is_odp_implicit;
683 		};
684 	};
685 };
686 
687 static inline bool is_odp_mr(struct mlx5_ib_mr *mr)
688 {
689 	return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
690 	       mr->umem->is_odp;
691 }
692 
693 static inline bool is_dmabuf_mr(struct mlx5_ib_mr *mr)
694 {
695 	return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
696 	       mr->umem->is_dmabuf;
697 }
698 
699 struct mlx5_ib_mw {
700 	struct ib_mw		ibmw;
701 	struct mlx5_ib_mkey	mmkey;
702 };
703 
704 struct mlx5_ib_umr_context {
705 	struct ib_cqe		cqe;
706 	enum ib_wc_status	status;
707 	struct completion	done;
708 };
709 
710 enum {
711 	MLX5_UMR_STATE_ACTIVE,
712 	MLX5_UMR_STATE_RECOVER,
713 	MLX5_UMR_STATE_ERR,
714 };
715 
716 struct umr_common {
717 	struct ib_pd	*pd;
718 	struct ib_cq	*cq;
719 	struct ib_qp	*qp;
720 	/* Protects from UMR QP overflow
721 	 */
722 	struct semaphore	sem;
723 	/* Protects from using UMR while the UMR is not active
724 	 */
725 	struct mutex lock;
726 	unsigned int state;
727 };
728 
729 struct mlx5_cache_ent {
730 	struct xarray		mkeys;
731 	unsigned long		stored;
732 	unsigned long		reserved;
733 
734 	char                    name[4];
735 	u32                     order;
736 	u32			access_mode;
737 	u32			page;
738 	unsigned int		ndescs;
739 
740 	u8 disabled:1;
741 	u8 fill_to_high_water:1;
742 
743 	/*
744 	 * - limit is the low water mark for stored mkeys, 2* limit is the
745 	 *   upper water mark.
746 	 */
747 	u32 in_use;
748 	u32 limit;
749 
750 	/* Statistics */
751 	u32                     miss;
752 
753 	struct mlx5_ib_dev     *dev;
754 	struct delayed_work	dwork;
755 };
756 
757 struct mlx5r_async_create_mkey {
758 	union {
759 		u32 in[MLX5_ST_SZ_BYTES(create_mkey_in)];
760 		u32 out[MLX5_ST_SZ_DW(create_mkey_out)];
761 	};
762 	struct mlx5_async_work cb_work;
763 	struct mlx5_cache_ent *ent;
764 	u32 mkey;
765 };
766 
767 struct mlx5_mkey_cache {
768 	struct workqueue_struct *wq;
769 	struct mlx5_cache_ent	ent[MAX_MKEY_CACHE_ENTRIES];
770 	struct dentry		*root;
771 	unsigned long		last_add;
772 };
773 
774 struct mlx5_ib_port_resources {
775 	struct mlx5_ib_gsi_qp *gsi;
776 	struct work_struct pkey_change_work;
777 };
778 
779 struct mlx5_ib_resources {
780 	struct ib_cq	*c0;
781 	u32 xrcdn0;
782 	u32 xrcdn1;
783 	struct ib_pd	*p0;
784 	struct ib_srq	*s0;
785 	struct ib_srq	*s1;
786 	struct mlx5_ib_port_resources ports[2];
787 };
788 
789 #define MAX_OPFC_RULES 2
790 
791 struct mlx5_ib_op_fc {
792 	struct mlx5_fc *fc;
793 	struct mlx5_flow_handle *rule[MAX_OPFC_RULES];
794 };
795 
796 struct mlx5_ib_counters {
797 	struct rdma_stat_desc *descs;
798 	size_t *offsets;
799 	u32 num_q_counters;
800 	u32 num_cong_counters;
801 	u32 num_ext_ppcnt_counters;
802 	u32 num_op_counters;
803 	u16 set_id;
804 	struct mlx5_ib_op_fc opfcs[MLX5_IB_OPCOUNTER_MAX];
805 };
806 
807 int mlx5_ib_fs_add_op_fc(struct mlx5_ib_dev *dev, u32 port_num,
808 			 struct mlx5_ib_op_fc *opfc,
809 			 enum mlx5_ib_optional_counter_type type);
810 
811 void mlx5_ib_fs_remove_op_fc(struct mlx5_ib_dev *dev,
812 			     struct mlx5_ib_op_fc *opfc,
813 			     enum mlx5_ib_optional_counter_type type);
814 
815 struct mlx5_ib_multiport_info;
816 
817 struct mlx5_ib_multiport {
818 	struct mlx5_ib_multiport_info *mpi;
819 	/* To be held when accessing the multiport info */
820 	spinlock_t mpi_lock;
821 };
822 
823 struct mlx5_roce {
824 	/* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL
825 	 * netdev pointer
826 	 */
827 	rwlock_t		netdev_lock;
828 	struct net_device	*netdev;
829 	struct notifier_block	nb;
830 	atomic_t		tx_port_affinity;
831 	enum ib_port_state last_port_state;
832 	struct mlx5_ib_dev	*dev;
833 	u32			native_port_num;
834 };
835 
836 struct mlx5_ib_port {
837 	struct mlx5_ib_counters cnts;
838 	struct mlx5_ib_multiport mp;
839 	struct mlx5_ib_dbg_cc_params *dbg_cc_params;
840 	struct mlx5_roce roce;
841 	struct mlx5_eswitch_rep		*rep;
842 };
843 
844 struct mlx5_ib_dbg_param {
845 	int			offset;
846 	struct mlx5_ib_dev	*dev;
847 	struct dentry		*dentry;
848 	u32			port_num;
849 };
850 
851 enum mlx5_ib_dbg_cc_types {
852 	MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE,
853 	MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE_ATI,
854 	MLX5_IB_DBG_CC_RP_TIME_RESET,
855 	MLX5_IB_DBG_CC_RP_BYTE_RESET,
856 	MLX5_IB_DBG_CC_RP_THRESHOLD,
857 	MLX5_IB_DBG_CC_RP_AI_RATE,
858 	MLX5_IB_DBG_CC_RP_MAX_RATE,
859 	MLX5_IB_DBG_CC_RP_HAI_RATE,
860 	MLX5_IB_DBG_CC_RP_MIN_DEC_FAC,
861 	MLX5_IB_DBG_CC_RP_MIN_RATE,
862 	MLX5_IB_DBG_CC_RP_RATE_TO_SET_ON_FIRST_CNP,
863 	MLX5_IB_DBG_CC_RP_DCE_TCP_G,
864 	MLX5_IB_DBG_CC_RP_DCE_TCP_RTT,
865 	MLX5_IB_DBG_CC_RP_RATE_REDUCE_MONITOR_PERIOD,
866 	MLX5_IB_DBG_CC_RP_INITIAL_ALPHA_VALUE,
867 	MLX5_IB_DBG_CC_RP_GD,
868 	MLX5_IB_DBG_CC_NP_MIN_TIME_BETWEEN_CNPS,
869 	MLX5_IB_DBG_CC_NP_CNP_DSCP,
870 	MLX5_IB_DBG_CC_NP_CNP_PRIO_MODE,
871 	MLX5_IB_DBG_CC_NP_CNP_PRIO,
872 	MLX5_IB_DBG_CC_MAX,
873 };
874 
875 struct mlx5_ib_dbg_cc_params {
876 	struct dentry			*root;
877 	struct mlx5_ib_dbg_param	params[MLX5_IB_DBG_CC_MAX];
878 };
879 
880 enum {
881 	MLX5_MAX_DELAY_DROP_TIMEOUT_MS = 100,
882 };
883 
884 struct mlx5_ib_delay_drop {
885 	struct mlx5_ib_dev     *dev;
886 	struct work_struct	delay_drop_work;
887 	/* serialize setting of delay drop */
888 	struct mutex		lock;
889 	u32			timeout;
890 	bool			activate;
891 	atomic_t		events_cnt;
892 	atomic_t		rqs_cnt;
893 	struct dentry		*dir_debugfs;
894 };
895 
896 enum mlx5_ib_stages {
897 	MLX5_IB_STAGE_INIT,
898 	MLX5_IB_STAGE_FS,
899 	MLX5_IB_STAGE_CAPS,
900 	MLX5_IB_STAGE_NON_DEFAULT_CB,
901 	MLX5_IB_STAGE_ROCE,
902 	MLX5_IB_STAGE_QP,
903 	MLX5_IB_STAGE_SRQ,
904 	MLX5_IB_STAGE_DEVICE_RESOURCES,
905 	MLX5_IB_STAGE_DEVICE_NOTIFIER,
906 	MLX5_IB_STAGE_ODP,
907 	MLX5_IB_STAGE_COUNTERS,
908 	MLX5_IB_STAGE_CONG_DEBUGFS,
909 	MLX5_IB_STAGE_UAR,
910 	MLX5_IB_STAGE_BFREG,
911 	MLX5_IB_STAGE_PRE_IB_REG_UMR,
912 	MLX5_IB_STAGE_WHITELIST_UID,
913 	MLX5_IB_STAGE_IB_REG,
914 	MLX5_IB_STAGE_POST_IB_REG_UMR,
915 	MLX5_IB_STAGE_DELAY_DROP,
916 	MLX5_IB_STAGE_RESTRACK,
917 	MLX5_IB_STAGE_MAX,
918 };
919 
920 struct mlx5_ib_stage {
921 	int (*init)(struct mlx5_ib_dev *dev);
922 	void (*cleanup)(struct mlx5_ib_dev *dev);
923 };
924 
925 #define STAGE_CREATE(_stage, _init, _cleanup) \
926 	.stage[_stage] = {.init = _init, .cleanup = _cleanup}
927 
928 struct mlx5_ib_profile {
929 	struct mlx5_ib_stage stage[MLX5_IB_STAGE_MAX];
930 };
931 
932 struct mlx5_ib_multiport_info {
933 	struct list_head list;
934 	struct mlx5_ib_dev *ibdev;
935 	struct mlx5_core_dev *mdev;
936 	struct notifier_block mdev_events;
937 	struct completion unref_comp;
938 	u64 sys_image_guid;
939 	u32 mdev_refcnt;
940 	bool is_master;
941 	bool unaffiliate;
942 };
943 
944 struct mlx5_ib_flow_action {
945 	struct ib_flow_action		ib_action;
946 	union {
947 		struct {
948 			u64			    ib_flags;
949 			struct mlx5_accel_esp_xfrm *ctx;
950 		} esp_aes_gcm;
951 		struct {
952 			struct mlx5_ib_dev *dev;
953 			u32 sub_type;
954 			union {
955 				struct mlx5_modify_hdr *modify_hdr;
956 				struct mlx5_pkt_reformat *pkt_reformat;
957 			};
958 		} flow_action_raw;
959 	};
960 };
961 
962 struct mlx5_dm {
963 	struct mlx5_core_dev *dev;
964 	/* This lock is used to protect the access to the shared
965 	 * allocation map when concurrent requests by different
966 	 * processes are handled.
967 	 */
968 	spinlock_t lock;
969 	DECLARE_BITMAP(memic_alloc_pages, MLX5_MAX_MEMIC_PAGES);
970 };
971 
972 struct mlx5_read_counters_attr {
973 	struct mlx5_fc *hw_cntrs_hndl;
974 	u64 *out;
975 	u32 flags;
976 };
977 
978 enum mlx5_ib_counters_type {
979 	MLX5_IB_COUNTERS_FLOW,
980 };
981 
982 struct mlx5_ib_mcounters {
983 	struct ib_counters ibcntrs;
984 	enum mlx5_ib_counters_type type;
985 	/* number of counters supported for this counters type */
986 	u32 counters_num;
987 	struct mlx5_fc *hw_cntrs_hndl;
988 	/* read function for this counters type */
989 	int (*read_counters)(struct ib_device *ibdev,
990 			     struct mlx5_read_counters_attr *read_attr);
991 	/* max index set as part of create_flow */
992 	u32 cntrs_max_index;
993 	/* number of counters data entries (<description,index> pair) */
994 	u32 ncounters;
995 	/* counters data array for descriptions and indexes */
996 	struct mlx5_ib_flow_counters_desc *counters_data;
997 	/* protects access to mcounters internal data */
998 	struct mutex mcntrs_mutex;
999 };
1000 
1001 static inline struct mlx5_ib_mcounters *
1002 to_mcounters(struct ib_counters *ibcntrs)
1003 {
1004 	return container_of(ibcntrs, struct mlx5_ib_mcounters, ibcntrs);
1005 }
1006 
1007 int parse_flow_flow_action(struct mlx5_ib_flow_action *maction,
1008 			   bool is_egress,
1009 			   struct mlx5_flow_act *action);
1010 struct mlx5_ib_lb_state {
1011 	/* protect the user_td */
1012 	struct mutex		mutex;
1013 	u32			user_td;
1014 	int			qps;
1015 	bool			enabled;
1016 };
1017 
1018 struct mlx5_ib_pf_eq {
1019 	struct notifier_block irq_nb;
1020 	struct mlx5_ib_dev *dev;
1021 	struct mlx5_eq *core;
1022 	struct work_struct work;
1023 	spinlock_t lock; /* Pagefaults spinlock */
1024 	struct workqueue_struct *wq;
1025 	mempool_t *pool;
1026 };
1027 
1028 struct mlx5_devx_event_table {
1029 	struct mlx5_nb devx_nb;
1030 	/* serialize updating the event_xa */
1031 	struct mutex event_xa_lock;
1032 	struct xarray event_xa;
1033 };
1034 
1035 struct mlx5_var_table {
1036 	/* serialize updating the bitmap */
1037 	struct mutex bitmap_lock;
1038 	unsigned long *bitmap;
1039 	u64 hw_start_addr;
1040 	u32 stride_size;
1041 	u64 num_var_hw_entries;
1042 };
1043 
1044 struct mlx5_port_caps {
1045 	bool has_smi;
1046 	u8 ext_port_cap;
1047 };
1048 
1049 struct mlx5_ib_dev {
1050 	struct ib_device		ib_dev;
1051 	struct mlx5_core_dev		*mdev;
1052 	struct notifier_block		mdev_events;
1053 	int				num_ports;
1054 	/* serialize update of capability mask
1055 	 */
1056 	struct mutex			cap_mask_mutex;
1057 	u8				ib_active:1;
1058 	u8				is_rep:1;
1059 	u8				lag_active:1;
1060 	u8				wc_support:1;
1061 	u8				fill_delay;
1062 	struct umr_common		umrc;
1063 	/* sync used page count stats
1064 	 */
1065 	struct mlx5_ib_resources	devr;
1066 
1067 	atomic_t			mkey_var;
1068 	struct mlx5_mkey_cache		cache;
1069 	struct timer_list		delay_timer;
1070 	/* Prevents soft lock on massive reg MRs */
1071 	struct mutex			slow_path_mutex;
1072 	struct ib_odp_caps	odp_caps;
1073 	u64			odp_max_size;
1074 	struct mutex		odp_eq_mutex;
1075 	struct mlx5_ib_pf_eq	odp_pf_eq;
1076 
1077 	struct xarray		odp_mkeys;
1078 
1079 	u32			null_mkey;
1080 	struct mlx5_ib_flow_db	*flow_db;
1081 	/* protect resources needed as part of reset flow */
1082 	spinlock_t		reset_flow_resource_lock;
1083 	struct list_head	qp_list;
1084 	/* Array with num_ports elements */
1085 	struct mlx5_ib_port	*port;
1086 	struct mlx5_sq_bfreg	bfreg;
1087 	struct mlx5_sq_bfreg	wc_bfreg;
1088 	struct mlx5_sq_bfreg	fp_bfreg;
1089 	struct mlx5_ib_delay_drop	delay_drop;
1090 	const struct mlx5_ib_profile	*profile;
1091 
1092 	struct mlx5_ib_lb_state		lb;
1093 	u8			umr_fence;
1094 	struct list_head	ib_dev_list;
1095 	u64			sys_image_guid;
1096 	struct mlx5_dm		dm;
1097 	u16			devx_whitelist_uid;
1098 	struct mlx5_srq_table   srq_table;
1099 	struct mlx5_qp_table    qp_table;
1100 	struct mlx5_async_ctx   async_ctx;
1101 	struct mlx5_devx_event_table devx_event_table;
1102 	struct mlx5_var_table var_table;
1103 
1104 	struct xarray sig_mrs;
1105 	struct mlx5_port_caps port_caps[MLX5_MAX_PORTS];
1106 	u16 pkey_table_len;
1107 	u8 lag_ports;
1108 };
1109 
1110 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
1111 {
1112 	return container_of(mcq, struct mlx5_ib_cq, mcq);
1113 }
1114 
1115 static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
1116 {
1117 	return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd);
1118 }
1119 
1120 static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev)
1121 {
1122 	return container_of(ibdev, struct mlx5_ib_dev, ib_dev);
1123 }
1124 
1125 static inline struct mlx5_ib_dev *mr_to_mdev(struct mlx5_ib_mr *mr)
1126 {
1127 	return to_mdev(mr->ibmr.device);
1128 }
1129 
1130 static inline struct mlx5_ib_dev *mlx5_udata_to_mdev(struct ib_udata *udata)
1131 {
1132 	struct mlx5_ib_ucontext *context = rdma_udata_to_drv_context(
1133 		udata, struct mlx5_ib_ucontext, ibucontext);
1134 
1135 	return to_mdev(context->ibucontext.device);
1136 }
1137 
1138 static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq)
1139 {
1140 	return container_of(ibcq, struct mlx5_ib_cq, ibcq);
1141 }
1142 
1143 static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp)
1144 {
1145 	return container_of(mqp, struct mlx5_ib_qp_base, mqp)->container_mibqp;
1146 }
1147 
1148 static inline struct mlx5_ib_rwq *to_mibrwq(struct mlx5_core_qp *core_qp)
1149 {
1150 	return container_of(core_qp, struct mlx5_ib_rwq, core_qp);
1151 }
1152 
1153 static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd)
1154 {
1155 	return container_of(ibpd, struct mlx5_ib_pd, ibpd);
1156 }
1157 
1158 static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq)
1159 {
1160 	return container_of(ibsrq, struct mlx5_ib_srq, ibsrq);
1161 }
1162 
1163 static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp)
1164 {
1165 	return container_of(ibqp, struct mlx5_ib_qp, ibqp);
1166 }
1167 
1168 static inline struct mlx5_ib_rwq *to_mrwq(struct ib_wq *ibwq)
1169 {
1170 	return container_of(ibwq, struct mlx5_ib_rwq, ibwq);
1171 }
1172 
1173 static inline struct mlx5_ib_rwq_ind_table *to_mrwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl)
1174 {
1175 	return container_of(ib_rwq_ind_tbl, struct mlx5_ib_rwq_ind_table, ib_rwq_ind_tbl);
1176 }
1177 
1178 static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
1179 {
1180 	return container_of(msrq, struct mlx5_ib_srq, msrq);
1181 }
1182 
1183 static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
1184 {
1185 	return container_of(ibmr, struct mlx5_ib_mr, ibmr);
1186 }
1187 
1188 static inline struct mlx5_ib_mw *to_mmw(struct ib_mw *ibmw)
1189 {
1190 	return container_of(ibmw, struct mlx5_ib_mw, ibmw);
1191 }
1192 
1193 static inline struct mlx5_ib_flow_action *
1194 to_mflow_act(struct ib_flow_action *ibact)
1195 {
1196 	return container_of(ibact, struct mlx5_ib_flow_action, ib_action);
1197 }
1198 
1199 static inline struct mlx5_user_mmap_entry *
1200 to_mmmap(struct rdma_user_mmap_entry *rdma_entry)
1201 {
1202 	return container_of(rdma_entry,
1203 		struct mlx5_user_mmap_entry, rdma_entry);
1204 }
1205 
1206 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
1207 			struct mlx5_db *db);
1208 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);
1209 void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1210 void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1211 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index);
1212 int mlx5_ib_create_ah(struct ib_ah *ah, struct rdma_ah_init_attr *init_attr,
1213 		      struct ib_udata *udata);
1214 int mlx5_ib_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr);
1215 static inline int mlx5_ib_destroy_ah(struct ib_ah *ah, u32 flags)
1216 {
1217 	return 0;
1218 }
1219 int mlx5_ib_create_srq(struct ib_srq *srq, struct ib_srq_init_attr *init_attr,
1220 		       struct ib_udata *udata);
1221 int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
1222 		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
1223 int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr);
1224 int mlx5_ib_destroy_srq(struct ib_srq *srq, struct ib_udata *udata);
1225 int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
1226 			  const struct ib_recv_wr **bad_wr);
1227 int mlx5_ib_enable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1228 void mlx5_ib_disable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1229 int mlx5_ib_create_qp(struct ib_qp *qp, struct ib_qp_init_attr *init_attr,
1230 		      struct ib_udata *udata);
1231 int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1232 		      int attr_mask, struct ib_udata *udata);
1233 int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
1234 		     struct ib_qp_init_attr *qp_init_attr);
1235 int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata);
1236 void mlx5_ib_drain_sq(struct ib_qp *qp);
1237 void mlx5_ib_drain_rq(struct ib_qp *qp);
1238 int mlx5_ib_read_wqe_sq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1239 			size_t buflen, size_t *bc);
1240 int mlx5_ib_read_wqe_rq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1241 			size_t buflen, size_t *bc);
1242 int mlx5_ib_read_wqe_srq(struct mlx5_ib_srq *srq, int wqe_index, void *buffer,
1243 			 size_t buflen, size_t *bc);
1244 int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
1245 		      struct ib_udata *udata);
1246 int mlx5_ib_destroy_cq(struct ib_cq *cq, struct ib_udata *udata);
1247 int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
1248 int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
1249 int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1250 int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata);
1251 struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc);
1252 struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
1253 				  u64 virt_addr, int access_flags,
1254 				  struct ib_udata *udata);
1255 struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start,
1256 					 u64 length, u64 virt_addr,
1257 					 int fd, int access_flags,
1258 					 struct ib_udata *udata);
1259 int mlx5_ib_advise_mr(struct ib_pd *pd,
1260 		      enum ib_uverbs_advise_mr_advice advice,
1261 		      u32 flags,
1262 		      struct ib_sge *sg_list,
1263 		      u32 num_sge,
1264 		      struct uverbs_attr_bundle *attrs);
1265 int mlx5_ib_alloc_mw(struct ib_mw *mw, struct ib_udata *udata);
1266 int mlx5_ib_dealloc_mw(struct ib_mw *mw);
1267 struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd,
1268 					     int access_flags);
1269 void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *mr);
1270 void mlx5_ib_free_odp_mr(struct mlx5_ib_mr *mr);
1271 struct ib_mr *mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start,
1272 				    u64 length, u64 virt_addr, int access_flags,
1273 				    struct ib_pd *pd, struct ib_udata *udata);
1274 int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata);
1275 struct ib_mr *mlx5_ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
1276 			       u32 max_num_sg);
1277 struct ib_mr *mlx5_ib_alloc_mr_integrity(struct ib_pd *pd,
1278 					 u32 max_num_sg,
1279 					 u32 max_num_meta_sg);
1280 int mlx5_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents,
1281 		      unsigned int *sg_offset);
1282 int mlx5_ib_map_mr_sg_pi(struct ib_mr *ibmr, struct scatterlist *data_sg,
1283 			 int data_sg_nents, unsigned int *data_sg_offset,
1284 			 struct scatterlist *meta_sg, int meta_sg_nents,
1285 			 unsigned int *meta_sg_offset);
1286 int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u32 port_num,
1287 			const struct ib_wc *in_wc, const struct ib_grh *in_grh,
1288 			const struct ib_mad *in, struct ib_mad *out,
1289 			size_t *out_mad_size, u16 *out_mad_pkey_index);
1290 int mlx5_ib_alloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1291 int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1292 int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, unsigned int port);
1293 int mlx5_query_mad_ifc_system_image_guid(struct ib_device *ibdev,
1294 					 __be64 *sys_image_guid);
1295 int mlx5_query_mad_ifc_max_pkeys(struct ib_device *ibdev,
1296 				 u16 *max_pkeys);
1297 int mlx5_query_mad_ifc_vendor_id(struct ib_device *ibdev,
1298 				 u32 *vendor_id);
1299 int mlx5_query_mad_ifc_node_desc(struct mlx5_ib_dev *dev, char *node_desc);
1300 int mlx5_query_mad_ifc_node_guid(struct mlx5_ib_dev *dev, __be64 *node_guid);
1301 int mlx5_query_mad_ifc_pkey(struct ib_device *ibdev, u32 port, u16 index,
1302 			    u16 *pkey);
1303 int mlx5_query_mad_ifc_gids(struct ib_device *ibdev, u32 port, int index,
1304 			    union ib_gid *gid);
1305 int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u32 port,
1306 			    struct ib_port_attr *props);
1307 int mlx5_ib_query_port(struct ib_device *ibdev, u32 port,
1308 		       struct ib_port_attr *props);
1309 void mlx5_ib_populate_pas(struct ib_umem *umem, size_t page_size, __be64 *pas,
1310 			  u64 access_flags);
1311 void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num);
1312 int mlx5_ib_get_cqe_size(struct ib_cq *ibcq);
1313 int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev);
1314 int mlx5_mkey_cache_cleanup(struct mlx5_ib_dev *dev);
1315 
1316 struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev,
1317 				       struct mlx5_cache_ent *ent,
1318 				       int access_flags);
1319 
1320 int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
1321 			    struct ib_mr_status *mr_status);
1322 struct ib_wq *mlx5_ib_create_wq(struct ib_pd *pd,
1323 				struct ib_wq_init_attr *init_attr,
1324 				struct ib_udata *udata);
1325 int mlx5_ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
1326 int mlx5_ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1327 		      u32 wq_attr_mask, struct ib_udata *udata);
1328 int mlx5_ib_create_rwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_table,
1329 				 struct ib_rwq_ind_table_init_attr *init_attr,
1330 				 struct ib_udata *udata);
1331 int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
1332 struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm,
1333 				struct ib_dm_mr_attr *attr,
1334 				struct uverbs_attr_bundle *attrs);
1335 
1336 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1337 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);
1338 int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq);
1339 void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev);
1340 int __init mlx5_ib_odp_init(void);
1341 void mlx5_ib_odp_cleanup(void);
1342 void mlx5_odp_init_mkey_cache_entry(struct mlx5_cache_ent *ent);
1343 void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1344 			   struct mlx5_ib_mr *mr, int flags);
1345 
1346 int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1347 			       enum ib_uverbs_advise_mr_advice advice,
1348 			       u32 flags, struct ib_sge *sg_list, u32 num_sge);
1349 int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr);
1350 int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr);
1351 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1352 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; }
1353 static inline int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev,
1354 				      struct mlx5_ib_pf_eq *eq)
1355 {
1356 	return 0;
1357 }
1358 static inline void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev) {}
1359 static inline int mlx5_ib_odp_init(void) { return 0; }
1360 static inline void mlx5_ib_odp_cleanup(void)				    {}
1361 static inline void mlx5_odp_init_mkey_cache_entry(struct mlx5_cache_ent *ent) {}
1362 static inline void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1363 					 struct mlx5_ib_mr *mr, int flags) {}
1364 
1365 static inline int
1366 mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1367 			   enum ib_uverbs_advise_mr_advice advice, u32 flags,
1368 			   struct ib_sge *sg_list, u32 num_sge)
1369 {
1370 	return -EOPNOTSUPP;
1371 }
1372 static inline int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr)
1373 {
1374 	return -EOPNOTSUPP;
1375 }
1376 static inline int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr)
1377 {
1378 	return -EOPNOTSUPP;
1379 }
1380 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1381 
1382 extern const struct mmu_interval_notifier_ops mlx5_mn_ops;
1383 
1384 /* Needed for rep profile */
1385 void __mlx5_ib_remove(struct mlx5_ib_dev *dev,
1386 		      const struct mlx5_ib_profile *profile,
1387 		      int stage);
1388 int __mlx5_ib_add(struct mlx5_ib_dev *dev,
1389 		  const struct mlx5_ib_profile *profile);
1390 
1391 int mlx5_ib_get_vf_config(struct ib_device *device, int vf,
1392 			  u32 port, struct ifla_vf_info *info);
1393 int mlx5_ib_set_vf_link_state(struct ib_device *device, int vf,
1394 			      u32 port, int state);
1395 int mlx5_ib_get_vf_stats(struct ib_device *device, int vf,
1396 			 u32 port, struct ifla_vf_stats *stats);
1397 int mlx5_ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
1398 			struct ifla_vf_guid *node_guid,
1399 			struct ifla_vf_guid *port_guid);
1400 int mlx5_ib_set_vf_guid(struct ib_device *device, int vf, u32 port,
1401 			u64 guid, int type);
1402 
1403 __be16 mlx5_get_roce_udp_sport_min(const struct mlx5_ib_dev *dev,
1404 				   const struct ib_gid_attr *attr);
1405 
1406 void mlx5_ib_cleanup_cong_debugfs(struct mlx5_ib_dev *dev, u32 port_num);
1407 void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u32 port_num);
1408 
1409 /* GSI QP helper functions */
1410 int mlx5_ib_create_gsi(struct ib_pd *pd, struct mlx5_ib_qp *mqp,
1411 		       struct ib_qp_init_attr *attr);
1412 int mlx5_ib_destroy_gsi(struct mlx5_ib_qp *mqp);
1413 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
1414 			  int attr_mask);
1415 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
1416 			 int qp_attr_mask,
1417 			 struct ib_qp_init_attr *qp_init_attr);
1418 int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
1419 			  const struct ib_send_wr **bad_wr);
1420 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, const struct ib_recv_wr *wr,
1421 			  const struct ib_recv_wr **bad_wr);
1422 void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi);
1423 
1424 int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc);
1425 
1426 void mlx5_ib_free_bfreg(struct mlx5_ib_dev *dev, struct mlx5_bfreg_info *bfregi,
1427 			int bfregn);
1428 struct mlx5_ib_dev *mlx5_ib_get_ibdev_from_mpi(struct mlx5_ib_multiport_info *mpi);
1429 struct mlx5_core_dev *mlx5_ib_get_native_port_mdev(struct mlx5_ib_dev *dev,
1430 						   u32 ib_port_num,
1431 						   u32 *native_port_num);
1432 void mlx5_ib_put_native_port_mdev(struct mlx5_ib_dev *dev,
1433 				  u32 port_num);
1434 
1435 extern const struct uapi_definition mlx5_ib_devx_defs[];
1436 extern const struct uapi_definition mlx5_ib_flow_defs[];
1437 extern const struct uapi_definition mlx5_ib_qos_defs[];
1438 extern const struct uapi_definition mlx5_ib_std_types_defs[];
1439 
1440 static inline int is_qp1(enum ib_qp_type qp_type)
1441 {
1442 	return qp_type == MLX5_IB_QPT_HW_GSI || qp_type == IB_QPT_GSI;
1443 }
1444 
1445 static inline u32 check_cq_create_flags(u32 flags)
1446 {
1447 	/*
1448 	 * It returns non-zero value for unsupported CQ
1449 	 * create flags, otherwise it returns zero.
1450 	 */
1451 	return (flags & ~(IB_UVERBS_CQ_FLAGS_IGNORE_OVERRUN |
1452 			  IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION));
1453 }
1454 
1455 static inline int verify_assign_uidx(u8 cqe_version, u32 cmd_uidx,
1456 				     u32 *user_index)
1457 {
1458 	if (cqe_version) {
1459 		if ((cmd_uidx == MLX5_IB_DEFAULT_UIDX) ||
1460 		    (cmd_uidx & ~MLX5_USER_ASSIGNED_UIDX_MASK))
1461 			return -EINVAL;
1462 		*user_index = cmd_uidx;
1463 	} else {
1464 		*user_index = MLX5_IB_DEFAULT_UIDX;
1465 	}
1466 
1467 	return 0;
1468 }
1469 
1470 static inline int get_qp_user_index(struct mlx5_ib_ucontext *ucontext,
1471 				    struct mlx5_ib_create_qp *ucmd,
1472 				    int inlen,
1473 				    u32 *user_index)
1474 {
1475 	u8 cqe_version = ucontext->cqe_version;
1476 
1477 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1478 	    (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1479 		return 0;
1480 
1481 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1482 		return -EINVAL;
1483 
1484 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1485 }
1486 
1487 static inline int get_srq_user_index(struct mlx5_ib_ucontext *ucontext,
1488 				     struct mlx5_ib_create_srq *ucmd,
1489 				     int inlen,
1490 				     u32 *user_index)
1491 {
1492 	u8 cqe_version = ucontext->cqe_version;
1493 
1494 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1495 	    (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1496 		return 0;
1497 
1498 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1499 		return -EINVAL;
1500 
1501 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1502 }
1503 
1504 static inline int get_uars_per_sys_page(struct mlx5_ib_dev *dev, bool lib_support)
1505 {
1506 	return lib_support && MLX5_CAP_GEN(dev->mdev, uar_4k) ?
1507 				MLX5_UARS_IN_PAGE : 1;
1508 }
1509 
1510 extern void *xlt_emergency_page;
1511 
1512 int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
1513 			struct mlx5_bfreg_info *bfregi, u32 bfregn,
1514 			bool dyn_bfreg);
1515 
1516 static inline int mlx5r_store_odp_mkey(struct mlx5_ib_dev *dev,
1517 				       struct mlx5_ib_mkey *mmkey)
1518 {
1519 	refcount_set(&mmkey->usecount, 1);
1520 
1521 	return xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(mmkey->key),
1522 			       mmkey, GFP_KERNEL));
1523 }
1524 
1525 /* deref an mkey that can participate in ODP flow */
1526 static inline void mlx5r_deref_odp_mkey(struct mlx5_ib_mkey *mmkey)
1527 {
1528 	if (refcount_dec_and_test(&mmkey->usecount))
1529 		wake_up(&mmkey->wait);
1530 }
1531 
1532 /* deref an mkey that can participate in ODP flow and wait for relese */
1533 static inline void mlx5r_deref_wait_odp_mkey(struct mlx5_ib_mkey *mmkey)
1534 {
1535 	mlx5r_deref_odp_mkey(mmkey);
1536 	wait_event(mmkey->wait, refcount_read(&mmkey->usecount) == 0);
1537 }
1538 
1539 int mlx5_ib_test_wc(struct mlx5_ib_dev *dev);
1540 
1541 static inline bool mlx5_ib_lag_should_assign_affinity(struct mlx5_ib_dev *dev)
1542 {
1543 	return dev->lag_active ||
1544 		(MLX5_CAP_GEN(dev->mdev, num_lag_ports) > 1 &&
1545 		 MLX5_CAP_GEN(dev->mdev, lag_tx_port_affinity));
1546 }
1547 
1548 static inline bool rt_supported(int ts_cap)
1549 {
1550 	return ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_REAL_TIME ||
1551 	       ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME;
1552 }
1553 #endif /* MLX5_IB_H */
1554