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_pp {
263 	u16 index;
264 	struct mlx5_core_dev *mdev;
265 };
266 
267 enum mlx5_ib_optional_counter_type {
268 	MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS,
269 	MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS,
270 	MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS,
271 
272 	MLX5_IB_OPCOUNTER_MAX,
273 };
274 
275 struct mlx5_ib_flow_db {
276 	struct mlx5_ib_flow_prio	prios[MLX5_IB_NUM_FLOW_FT];
277 	struct mlx5_ib_flow_prio	egress_prios[MLX5_IB_NUM_FLOW_FT];
278 	struct mlx5_ib_flow_prio	sniffer[MLX5_IB_NUM_SNIFFER_FTS];
279 	struct mlx5_ib_flow_prio	egress[MLX5_IB_NUM_EGRESS_FTS];
280 	struct mlx5_ib_flow_prio	fdb[MLX5_IB_NUM_FDB_FTS];
281 	struct mlx5_ib_flow_prio	rdma_rx[MLX5_IB_NUM_FLOW_FT];
282 	struct mlx5_ib_flow_prio	rdma_tx[MLX5_IB_NUM_FLOW_FT];
283 	struct mlx5_ib_flow_prio	opfcs[MLX5_IB_OPCOUNTER_MAX];
284 	struct mlx5_flow_table		*lag_demux_ft;
285 	/* Protect flow steering bypass flow tables
286 	 * when add/del flow rules.
287 	 * only single add/removal of flow steering rule could be done
288 	 * simultaneously.
289 	 */
290 	struct mutex			lock;
291 };
292 
293 /* Use macros here so that don't have to duplicate
294  * enum ib_qp_type for low-level driver
295  */
296 
297 #define MLX5_IB_QPT_REG_UMR	IB_QPT_RESERVED1
298 /*
299  * IB_QPT_GSI creates the software wrapper around GSI, and MLX5_IB_QPT_HW_GSI
300  * creates the actual hardware QP.
301  */
302 #define MLX5_IB_QPT_HW_GSI	IB_QPT_RESERVED2
303 #define MLX5_IB_QPT_DCI		IB_QPT_RESERVED3
304 #define MLX5_IB_QPT_DCT		IB_QPT_RESERVED4
305 #define MLX5_IB_WR_UMR		IB_WR_RESERVED1
306 
307 #define MLX5_IB_UPD_XLT_ZAP	      BIT(0)
308 #define MLX5_IB_UPD_XLT_ENABLE	      BIT(1)
309 #define MLX5_IB_UPD_XLT_ATOMIC	      BIT(2)
310 #define MLX5_IB_UPD_XLT_ADDR	      BIT(3)
311 #define MLX5_IB_UPD_XLT_PD	      BIT(4)
312 #define MLX5_IB_UPD_XLT_ACCESS	      BIT(5)
313 #define MLX5_IB_UPD_XLT_INDIRECT      BIT(6)
314 
315 /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags.
316  *
317  * These flags are intended for internal use by the mlx5_ib driver, and they
318  * rely on the range reserved for that use in the ib_qp_create_flags enum.
319  */
320 #define MLX5_IB_QP_CREATE_SQPN_QP1	IB_QP_CREATE_RESERVED_START
321 #define MLX5_IB_QP_CREATE_WC_TEST	(IB_QP_CREATE_RESERVED_START << 1)
322 
323 struct wr_list {
324 	u16	opcode;
325 	u16	next;
326 };
327 
328 enum mlx5_ib_rq_flags {
329 	MLX5_IB_RQ_CVLAN_STRIPPING	= 1 << 0,
330 	MLX5_IB_RQ_PCI_WRITE_END_PADDING	= 1 << 1,
331 };
332 
333 struct mlx5_ib_wq {
334 	struct mlx5_frag_buf_ctrl fbc;
335 	u64		       *wrid;
336 	u32		       *wr_data;
337 	struct wr_list	       *w_list;
338 	unsigned	       *wqe_head;
339 	u16		        unsig_count;
340 
341 	/* serialize post to the work queue
342 	 */
343 	spinlock_t		lock;
344 	int			wqe_cnt;
345 	int			max_post;
346 	int			max_gs;
347 	int			offset;
348 	int			wqe_shift;
349 	unsigned		head;
350 	unsigned		tail;
351 	u16			cur_post;
352 	u16			last_poll;
353 	void			*cur_edge;
354 };
355 
356 enum mlx5_ib_wq_flags {
357 	MLX5_IB_WQ_FLAGS_DELAY_DROP = 0x1,
358 	MLX5_IB_WQ_FLAGS_STRIDING_RQ = 0x2,
359 };
360 
361 #define MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES 9
362 #define MLX5_MAX_SINGLE_WQE_LOG_NUM_STRIDES 16
363 #define MLX5_MIN_SINGLE_STRIDE_LOG_NUM_BYTES 6
364 #define MLX5_MAX_SINGLE_STRIDE_LOG_NUM_BYTES 13
365 #define MLX5_EXT_MIN_SINGLE_WQE_LOG_NUM_STRIDES 3
366 
367 struct mlx5_ib_rwq {
368 	struct ib_wq		ibwq;
369 	struct mlx5_core_qp	core_qp;
370 	u32			rq_num_pas;
371 	u32			log_rq_stride;
372 	u32			log_rq_size;
373 	u32			rq_page_offset;
374 	u32			log_page_size;
375 	u32			log_num_strides;
376 	u32			two_byte_shift_en;
377 	u32			single_stride_log_num_of_bytes;
378 	struct ib_umem		*umem;
379 	size_t			buf_size;
380 	unsigned int		page_shift;
381 	struct mlx5_db		db;
382 	u32			user_index;
383 	u32			wqe_count;
384 	u32			wqe_shift;
385 	int			wq_sig;
386 	u32			create_flags; /* Use enum mlx5_ib_wq_flags */
387 };
388 
389 struct mlx5_ib_rwq_ind_table {
390 	struct ib_rwq_ind_table ib_rwq_ind_tbl;
391 	u32			rqtn;
392 	u16			uid;
393 };
394 
395 struct mlx5_ib_ubuffer {
396 	struct ib_umem	       *umem;
397 	int			buf_size;
398 	u64			buf_addr;
399 };
400 
401 struct mlx5_ib_qp_base {
402 	struct mlx5_ib_qp	*container_mibqp;
403 	struct mlx5_core_qp	mqp;
404 	struct mlx5_ib_ubuffer	ubuffer;
405 };
406 
407 struct mlx5_ib_qp_trans {
408 	struct mlx5_ib_qp_base	base;
409 	u16			xrcdn;
410 	u32			alt_port;
411 	u8			atomic_rd_en;
412 	u8			resp_depth;
413 };
414 
415 struct mlx5_ib_rss_qp {
416 	u32	tirn;
417 };
418 
419 struct mlx5_ib_rq {
420 	struct mlx5_ib_qp_base base;
421 	struct mlx5_ib_wq	*rq;
422 	struct mlx5_ib_ubuffer	ubuffer;
423 	struct mlx5_db		*doorbell;
424 	u32			tirn;
425 	u8			state;
426 	u32			flags;
427 };
428 
429 struct mlx5_ib_sq {
430 	struct mlx5_ib_qp_base base;
431 	struct mlx5_ib_wq	*sq;
432 	struct mlx5_ib_ubuffer  ubuffer;
433 	struct mlx5_db		*doorbell;
434 	struct mlx5_flow_handle	*flow_rule;
435 	u32			tisn;
436 	u8			state;
437 };
438 
439 struct mlx5_ib_raw_packet_qp {
440 	struct mlx5_ib_sq sq;
441 	struct mlx5_ib_rq rq;
442 };
443 
444 struct mlx5_bf {
445 	int			buf_size;
446 	unsigned long		offset;
447 	struct mlx5_sq_bfreg   *bfreg;
448 };
449 
450 struct mlx5_ib_dct {
451 	struct mlx5_core_dct    mdct;
452 	u32                     *in;
453 };
454 
455 struct mlx5_ib_gsi_qp {
456 	struct ib_qp *rx_qp;
457 	u32 port_num;
458 	struct ib_qp_cap cap;
459 	struct ib_cq *cq;
460 	struct mlx5_ib_gsi_wr *outstanding_wrs;
461 	u32 outstanding_pi, outstanding_ci;
462 	int num_qps;
463 	/* Protects access to the tx_qps. Post send operations synchronize
464 	 * with tx_qp creation in setup_qp(). Also protects the
465 	 * outstanding_wrs array and indices.
466 	 */
467 	spinlock_t lock;
468 	struct ib_qp **tx_qps;
469 };
470 
471 struct mlx5_ib_qp {
472 	struct ib_qp		ibqp;
473 	union {
474 		struct mlx5_ib_qp_trans trans_qp;
475 		struct mlx5_ib_raw_packet_qp raw_packet_qp;
476 		struct mlx5_ib_rss_qp rss_qp;
477 		struct mlx5_ib_dct dct;
478 		struct mlx5_ib_gsi_qp gsi;
479 	};
480 	struct mlx5_frag_buf	buf;
481 
482 	struct mlx5_db		db;
483 	struct mlx5_ib_wq	rq;
484 
485 	u8			sq_signal_bits;
486 	u8			next_fence;
487 	struct mlx5_ib_wq	sq;
488 
489 	/* serialize qp state modifications
490 	 */
491 	struct mutex		mutex;
492 	/* cached variant of create_flags from struct ib_qp_init_attr */
493 	u32			flags;
494 	u32			port;
495 	u8			state;
496 	int			max_inline_data;
497 	struct mlx5_bf	        bf;
498 	u8			has_rq:1;
499 	u8			is_rss:1;
500 
501 	/* only for user space QPs. For kernel
502 	 * we have it from the bf object
503 	 */
504 	int			bfregn;
505 
506 	struct list_head	qps_list;
507 	struct list_head	cq_recv_list;
508 	struct list_head	cq_send_list;
509 	struct mlx5_rate_limit	rl;
510 	u32                     underlay_qpn;
511 	u32			flags_en;
512 	/*
513 	 * IB/core doesn't store low-level QP types, so
514 	 * store both MLX and IBTA types in the field below.
515 	 */
516 	enum ib_qp_type		type;
517 	/* A flag to indicate if there's a new counter is configured
518 	 * but not take effective
519 	 */
520 	u32                     counter_pending;
521 	u16			gsi_lag_port;
522 };
523 
524 struct mlx5_ib_cq_buf {
525 	struct mlx5_frag_buf_ctrl fbc;
526 	struct mlx5_frag_buf    frag_buf;
527 	struct ib_umem		*umem;
528 	int			cqe_size;
529 	int			nent;
530 };
531 
532 enum mlx5_ib_cq_pr_flags {
533 	MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD	= 1 << 0,
534 	MLX5_IB_CQ_PR_FLAGS_REAL_TIME_TS = 1 << 1,
535 };
536 
537 struct mlx5_ib_cq {
538 	struct ib_cq		ibcq;
539 	struct mlx5_core_cq	mcq;
540 	struct mlx5_ib_cq_buf	buf;
541 	struct mlx5_db		db;
542 
543 	/* serialize access to the CQ
544 	 */
545 	spinlock_t		lock;
546 
547 	/* protect resize cq
548 	 */
549 	struct mutex		resize_mutex;
550 	struct mlx5_ib_cq_buf  *resize_buf;
551 	struct ib_umem	       *resize_umem;
552 	int			cqe_size;
553 	struct list_head	list_send_qp;
554 	struct list_head	list_recv_qp;
555 	u32			create_flags;
556 	struct list_head	wc_list;
557 	enum ib_cq_notify_flags notify_flags;
558 	struct work_struct	notify_work;
559 	u16			private_flags; /* Use mlx5_ib_cq_pr_flags */
560 };
561 
562 struct mlx5_ib_wc {
563 	struct ib_wc wc;
564 	struct list_head list;
565 };
566 
567 struct mlx5_ib_srq {
568 	struct ib_srq		ibsrq;
569 	struct mlx5_core_srq	msrq;
570 	struct mlx5_frag_buf	buf;
571 	struct mlx5_db		db;
572 	struct mlx5_frag_buf_ctrl fbc;
573 	u64		       *wrid;
574 	/* protect SRQ hanlding
575 	 */
576 	spinlock_t		lock;
577 	int			head;
578 	int			tail;
579 	u16			wqe_ctr;
580 	struct ib_umem	       *umem;
581 	/* serialize arming a SRQ
582 	 */
583 	struct mutex		mutex;
584 	int			wq_sig;
585 };
586 
587 struct mlx5_ib_xrcd {
588 	struct ib_xrcd		ibxrcd;
589 	u32			xrcdn;
590 };
591 
592 enum mlx5_ib_mtt_access_flags {
593 	MLX5_IB_MTT_READ  = (1 << 0),
594 	MLX5_IB_MTT_WRITE = (1 << 1),
595 };
596 
597 struct mlx5_user_mmap_entry {
598 	struct rdma_user_mmap_entry rdma_entry;
599 	u8 mmap_flag;
600 	u64 address;
601 	u32 page_idx;
602 };
603 
604 enum mlx5_mkey_type {
605 	MLX5_MKEY_MR = 1,
606 	MLX5_MKEY_MW,
607 	MLX5_MKEY_INDIRECT_DEVX,
608 };
609 
610 struct mlx5_ib_mkey {
611 	u32 key;
612 	enum mlx5_mkey_type type;
613 	unsigned int ndescs;
614 	struct wait_queue_head wait;
615 	refcount_t usecount;
616 };
617 
618 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
619 
620 #define MLX5_IB_DM_MEMIC_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE   |\
621 					 IB_ACCESS_REMOTE_WRITE  |\
622 					 IB_ACCESS_REMOTE_READ   |\
623 					 IB_ACCESS_REMOTE_ATOMIC |\
624 					 IB_ZERO_BASED)
625 
626 #define MLX5_IB_DM_SW_ICM_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE   |\
627 					  IB_ACCESS_REMOTE_WRITE  |\
628 					  IB_ACCESS_REMOTE_READ   |\
629 					  IB_ZERO_BASED)
630 
631 #define mlx5_update_odp_stats(mr, counter_name, value)		\
632 	atomic64_add(value, &((mr)->odp_stats.counter_name))
633 
634 struct mlx5_ib_mr {
635 	struct ib_mr ibmr;
636 	struct mlx5_ib_mkey mmkey;
637 
638 	/* User MR data */
639 	struct mlx5_cache_ent *cache_ent;
640 	/* Everything after cache_ent is zero'd when MR allocated */
641 	struct ib_umem *umem;
642 
643 	union {
644 		/* Used only while the MR is in the cache */
645 		struct {
646 			u32 out[MLX5_ST_SZ_DW(create_mkey_out)];
647 			struct mlx5_async_work cb_work;
648 			/* Cache list element */
649 			struct list_head list;
650 		};
651 
652 		/* Used only by kernel MRs (umem == NULL) */
653 		struct {
654 			void *descs;
655 			void *descs_alloc;
656 			dma_addr_t desc_map;
657 			int max_descs;
658 			int desc_size;
659 			int access_mode;
660 
661 			/* For Kernel IB_MR_TYPE_INTEGRITY */
662 			struct mlx5_core_sig_ctx *sig;
663 			struct mlx5_ib_mr *pi_mr;
664 			struct mlx5_ib_mr *klm_mr;
665 			struct mlx5_ib_mr *mtt_mr;
666 			u64 data_iova;
667 			u64 pi_iova;
668 			int meta_ndescs;
669 			int meta_length;
670 			int data_length;
671 		};
672 
673 		/* Used only by User MRs (umem != NULL) */
674 		struct {
675 			unsigned int page_shift;
676 			/* Current access_flags */
677 			int access_flags;
678 
679 			/* For User ODP */
680 			struct mlx5_ib_mr *parent;
681 			struct xarray implicit_children;
682 			union {
683 				struct work_struct work;
684 			} odp_destroy;
685 			struct ib_odp_counters odp_stats;
686 			bool is_odp_implicit;
687 		};
688 	};
689 };
690 
691 /* Zero the fields in the mr that are variant depending on usage */
692 static inline void mlx5_clear_mr(struct mlx5_ib_mr *mr)
693 {
694 	memset_after(mr, 0, cache_ent);
695 }
696 
697 static inline bool is_odp_mr(struct mlx5_ib_mr *mr)
698 {
699 	return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
700 	       mr->umem->is_odp;
701 }
702 
703 static inline bool is_dmabuf_mr(struct mlx5_ib_mr *mr)
704 {
705 	return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
706 	       mr->umem->is_dmabuf;
707 }
708 
709 struct mlx5_ib_mw {
710 	struct ib_mw		ibmw;
711 	struct mlx5_ib_mkey	mmkey;
712 };
713 
714 struct mlx5_ib_umr_context {
715 	struct ib_cqe		cqe;
716 	enum ib_wc_status	status;
717 	struct completion	done;
718 };
719 
720 struct umr_common {
721 	struct ib_pd	*pd;
722 	struct ib_cq	*cq;
723 	struct ib_qp	*qp;
724 	/* control access to UMR QP
725 	 */
726 	struct semaphore	sem;
727 };
728 
729 struct mlx5_cache_ent {
730 	struct list_head	head;
731 	/* sync access to the cahce entry
732 	 */
733 	spinlock_t		lock;
734 
735 
736 	char                    name[4];
737 	u32                     order;
738 	u32			access_mode;
739 	u32			page;
740 	unsigned int		ndescs;
741 
742 	u8 disabled:1;
743 	u8 fill_to_high_water:1;
744 
745 	/*
746 	 * - available_mrs is the length of list head, ie the number of MRs
747 	 *   available for immediate allocation.
748 	 * - total_mrs is available_mrs plus all in use MRs that could be
749 	 *   returned to the cache.
750 	 * - limit is the low water mark for available_mrs, 2* limit is the
751 	 *   upper water mark.
752 	 * - pending is the number of MRs currently being created
753 	 */
754 	u32 total_mrs;
755 	u32 available_mrs;
756 	u32 limit;
757 	u32 pending;
758 
759 	/* Statistics */
760 	u32                     miss;
761 
762 	struct mlx5_ib_dev     *dev;
763 	struct delayed_work	dwork;
764 };
765 
766 struct mlx5_mr_cache {
767 	struct workqueue_struct *wq;
768 	struct mlx5_cache_ent	ent[MAX_MR_CACHE_ENTRIES];
769 	struct dentry		*root;
770 	unsigned long		last_add;
771 };
772 
773 struct mlx5_ib_port_resources {
774 	struct mlx5_ib_gsi_qp *gsi;
775 	struct work_struct pkey_change_work;
776 };
777 
778 struct mlx5_ib_resources {
779 	struct ib_cq	*c0;
780 	u32 xrcdn0;
781 	u32 xrcdn1;
782 	struct ib_pd	*p0;
783 	struct ib_srq	*s0;
784 	struct ib_srq	*s1;
785 	struct mlx5_ib_port_resources ports[2];
786 };
787 
788 #define MAX_OPFC_RULES 2
789 
790 struct mlx5_ib_op_fc {
791 	struct mlx5_fc *fc;
792 	struct mlx5_flow_handle *rule[MAX_OPFC_RULES];
793 };
794 
795 struct mlx5_ib_counters {
796 	struct rdma_stat_desc *descs;
797 	size_t *offsets;
798 	u32 num_q_counters;
799 	u32 num_cong_counters;
800 	u32 num_ext_ppcnt_counters;
801 	u32 num_op_counters;
802 	u16 set_id;
803 	struct mlx5_ib_op_fc opfcs[MLX5_IB_OPCOUNTER_MAX];
804 };
805 
806 int mlx5_ib_fs_add_op_fc(struct mlx5_ib_dev *dev, u32 port_num,
807 			 struct mlx5_ib_op_fc *opfc,
808 			 enum mlx5_ib_optional_counter_type type);
809 
810 void mlx5_ib_fs_remove_op_fc(struct mlx5_ib_dev *dev,
811 			     struct mlx5_ib_op_fc *opfc,
812 			     enum mlx5_ib_optional_counter_type type);
813 
814 struct mlx5_ib_multiport_info;
815 
816 struct mlx5_ib_multiport {
817 	struct mlx5_ib_multiport_info *mpi;
818 	/* To be held when accessing the multiport info */
819 	spinlock_t mpi_lock;
820 };
821 
822 struct mlx5_roce {
823 	/* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL
824 	 * netdev pointer
825 	 */
826 	rwlock_t		netdev_lock;
827 	struct net_device	*netdev;
828 	struct notifier_block	nb;
829 	atomic_t		tx_port_affinity;
830 	enum ib_port_state last_port_state;
831 	struct mlx5_ib_dev	*dev;
832 	u32			native_port_num;
833 };
834 
835 struct mlx5_ib_port {
836 	struct mlx5_ib_counters cnts;
837 	struct mlx5_ib_multiport mp;
838 	struct mlx5_ib_dbg_cc_params *dbg_cc_params;
839 	struct mlx5_roce roce;
840 	struct mlx5_eswitch_rep		*rep;
841 };
842 
843 struct mlx5_ib_dbg_param {
844 	int			offset;
845 	struct mlx5_ib_dev	*dev;
846 	struct dentry		*dentry;
847 	u32			port_num;
848 };
849 
850 enum mlx5_ib_dbg_cc_types {
851 	MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE,
852 	MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE_ATI,
853 	MLX5_IB_DBG_CC_RP_TIME_RESET,
854 	MLX5_IB_DBG_CC_RP_BYTE_RESET,
855 	MLX5_IB_DBG_CC_RP_THRESHOLD,
856 	MLX5_IB_DBG_CC_RP_AI_RATE,
857 	MLX5_IB_DBG_CC_RP_MAX_RATE,
858 	MLX5_IB_DBG_CC_RP_HAI_RATE,
859 	MLX5_IB_DBG_CC_RP_MIN_DEC_FAC,
860 	MLX5_IB_DBG_CC_RP_MIN_RATE,
861 	MLX5_IB_DBG_CC_RP_RATE_TO_SET_ON_FIRST_CNP,
862 	MLX5_IB_DBG_CC_RP_DCE_TCP_G,
863 	MLX5_IB_DBG_CC_RP_DCE_TCP_RTT,
864 	MLX5_IB_DBG_CC_RP_RATE_REDUCE_MONITOR_PERIOD,
865 	MLX5_IB_DBG_CC_RP_INITIAL_ALPHA_VALUE,
866 	MLX5_IB_DBG_CC_RP_GD,
867 	MLX5_IB_DBG_CC_NP_MIN_TIME_BETWEEN_CNPS,
868 	MLX5_IB_DBG_CC_NP_CNP_DSCP,
869 	MLX5_IB_DBG_CC_NP_CNP_PRIO_MODE,
870 	MLX5_IB_DBG_CC_NP_CNP_PRIO,
871 	MLX5_IB_DBG_CC_MAX,
872 };
873 
874 struct mlx5_ib_dbg_cc_params {
875 	struct dentry			*root;
876 	struct mlx5_ib_dbg_param	params[MLX5_IB_DBG_CC_MAX];
877 };
878 
879 enum {
880 	MLX5_MAX_DELAY_DROP_TIMEOUT_MS = 100,
881 };
882 
883 struct mlx5_ib_delay_drop {
884 	struct mlx5_ib_dev     *dev;
885 	struct work_struct	delay_drop_work;
886 	/* serialize setting of delay drop */
887 	struct mutex		lock;
888 	u32			timeout;
889 	bool			activate;
890 	atomic_t		events_cnt;
891 	atomic_t		rqs_cnt;
892 	struct dentry		*dir_debugfs;
893 };
894 
895 enum mlx5_ib_stages {
896 	MLX5_IB_STAGE_INIT,
897 	MLX5_IB_STAGE_FS,
898 	MLX5_IB_STAGE_CAPS,
899 	MLX5_IB_STAGE_NON_DEFAULT_CB,
900 	MLX5_IB_STAGE_ROCE,
901 	MLX5_IB_STAGE_QP,
902 	MLX5_IB_STAGE_SRQ,
903 	MLX5_IB_STAGE_DEVICE_RESOURCES,
904 	MLX5_IB_STAGE_DEVICE_NOTIFIER,
905 	MLX5_IB_STAGE_ODP,
906 	MLX5_IB_STAGE_COUNTERS,
907 	MLX5_IB_STAGE_CONG_DEBUGFS,
908 	MLX5_IB_STAGE_UAR,
909 	MLX5_IB_STAGE_BFREG,
910 	MLX5_IB_STAGE_PRE_IB_REG_UMR,
911 	MLX5_IB_STAGE_WHITELIST_UID,
912 	MLX5_IB_STAGE_IB_REG,
913 	MLX5_IB_STAGE_POST_IB_REG_UMR,
914 	MLX5_IB_STAGE_DELAY_DROP,
915 	MLX5_IB_STAGE_RESTRACK,
916 	MLX5_IB_STAGE_MAX,
917 };
918 
919 struct mlx5_ib_stage {
920 	int (*init)(struct mlx5_ib_dev *dev);
921 	void (*cleanup)(struct mlx5_ib_dev *dev);
922 };
923 
924 #define STAGE_CREATE(_stage, _init, _cleanup) \
925 	.stage[_stage] = {.init = _init, .cleanup = _cleanup}
926 
927 struct mlx5_ib_profile {
928 	struct mlx5_ib_stage stage[MLX5_IB_STAGE_MAX];
929 };
930 
931 struct mlx5_ib_multiport_info {
932 	struct list_head list;
933 	struct mlx5_ib_dev *ibdev;
934 	struct mlx5_core_dev *mdev;
935 	struct notifier_block mdev_events;
936 	struct completion unref_comp;
937 	u64 sys_image_guid;
938 	u32 mdev_refcnt;
939 	bool is_master;
940 	bool unaffiliate;
941 };
942 
943 struct mlx5_ib_flow_action {
944 	struct ib_flow_action		ib_action;
945 	union {
946 		struct {
947 			u64			    ib_flags;
948 			struct mlx5_accel_esp_xfrm *ctx;
949 		} esp_aes_gcm;
950 		struct {
951 			struct mlx5_ib_dev *dev;
952 			u32 sub_type;
953 			union {
954 				struct mlx5_modify_hdr *modify_hdr;
955 				struct mlx5_pkt_reformat *pkt_reformat;
956 			};
957 		} flow_action_raw;
958 	};
959 };
960 
961 struct mlx5_dm {
962 	struct mlx5_core_dev *dev;
963 	/* This lock is used to protect the access to the shared
964 	 * allocation map when concurrent requests by different
965 	 * processes are handled.
966 	 */
967 	spinlock_t lock;
968 	DECLARE_BITMAP(memic_alloc_pages, MLX5_MAX_MEMIC_PAGES);
969 };
970 
971 struct mlx5_read_counters_attr {
972 	struct mlx5_fc *hw_cntrs_hndl;
973 	u64 *out;
974 	u32 flags;
975 };
976 
977 enum mlx5_ib_counters_type {
978 	MLX5_IB_COUNTERS_FLOW,
979 };
980 
981 struct mlx5_ib_mcounters {
982 	struct ib_counters ibcntrs;
983 	enum mlx5_ib_counters_type type;
984 	/* number of counters supported for this counters type */
985 	u32 counters_num;
986 	struct mlx5_fc *hw_cntrs_hndl;
987 	/* read function for this counters type */
988 	int (*read_counters)(struct ib_device *ibdev,
989 			     struct mlx5_read_counters_attr *read_attr);
990 	/* max index set as part of create_flow */
991 	u32 cntrs_max_index;
992 	/* number of counters data entries (<description,index> pair) */
993 	u32 ncounters;
994 	/* counters data array for descriptions and indexes */
995 	struct mlx5_ib_flow_counters_desc *counters_data;
996 	/* protects access to mcounters internal data */
997 	struct mutex mcntrs_mutex;
998 };
999 
1000 static inline struct mlx5_ib_mcounters *
1001 to_mcounters(struct ib_counters *ibcntrs)
1002 {
1003 	return container_of(ibcntrs, struct mlx5_ib_mcounters, ibcntrs);
1004 }
1005 
1006 int parse_flow_flow_action(struct mlx5_ib_flow_action *maction,
1007 			   bool is_egress,
1008 			   struct mlx5_flow_act *action);
1009 struct mlx5_ib_lb_state {
1010 	/* protect the user_td */
1011 	struct mutex		mutex;
1012 	u32			user_td;
1013 	int			qps;
1014 	bool			enabled;
1015 };
1016 
1017 struct mlx5_ib_pf_eq {
1018 	struct notifier_block irq_nb;
1019 	struct mlx5_ib_dev *dev;
1020 	struct mlx5_eq *core;
1021 	struct work_struct work;
1022 	spinlock_t lock; /* Pagefaults spinlock */
1023 	struct workqueue_struct *wq;
1024 	mempool_t *pool;
1025 };
1026 
1027 struct mlx5_devx_event_table {
1028 	struct mlx5_nb devx_nb;
1029 	/* serialize updating the event_xa */
1030 	struct mutex event_xa_lock;
1031 	struct xarray event_xa;
1032 };
1033 
1034 struct mlx5_var_table {
1035 	/* serialize updating the bitmap */
1036 	struct mutex bitmap_lock;
1037 	unsigned long *bitmap;
1038 	u64 hw_start_addr;
1039 	u32 stride_size;
1040 	u64 num_var_hw_entries;
1041 };
1042 
1043 struct mlx5_port_caps {
1044 	bool has_smi;
1045 	u8 ext_port_cap;
1046 };
1047 
1048 struct mlx5_ib_dev {
1049 	struct ib_device		ib_dev;
1050 	struct mlx5_core_dev		*mdev;
1051 	struct notifier_block		mdev_events;
1052 	int				num_ports;
1053 	/* serialize update of capability mask
1054 	 */
1055 	struct mutex			cap_mask_mutex;
1056 	u8				ib_active:1;
1057 	u8				is_rep:1;
1058 	u8				lag_active:1;
1059 	u8				wc_support:1;
1060 	u8				fill_delay;
1061 	struct umr_common		umrc;
1062 	/* sync used page count stats
1063 	 */
1064 	struct mlx5_ib_resources	devr;
1065 
1066 	atomic_t			mkey_var;
1067 	struct mlx5_mr_cache		cache;
1068 	struct timer_list		delay_timer;
1069 	/* Prevents soft lock on massive reg MRs */
1070 	struct mutex			slow_path_mutex;
1071 	struct ib_odp_caps	odp_caps;
1072 	u64			odp_max_size;
1073 	struct mutex		odp_eq_mutex;
1074 	struct mlx5_ib_pf_eq	odp_pf_eq;
1075 
1076 	struct xarray		odp_mkeys;
1077 
1078 	u32			null_mkey;
1079 	struct mlx5_ib_flow_db	*flow_db;
1080 	/* protect resources needed as part of reset flow */
1081 	spinlock_t		reset_flow_resource_lock;
1082 	struct list_head	qp_list;
1083 	/* Array with num_ports elements */
1084 	struct mlx5_ib_port	*port;
1085 	struct mlx5_sq_bfreg	bfreg;
1086 	struct mlx5_sq_bfreg	wc_bfreg;
1087 	struct mlx5_sq_bfreg	fp_bfreg;
1088 	struct mlx5_ib_delay_drop	delay_drop;
1089 	const struct mlx5_ib_profile	*profile;
1090 
1091 	struct mlx5_ib_lb_state		lb;
1092 	u8			umr_fence;
1093 	struct list_head	ib_dev_list;
1094 	u64			sys_image_guid;
1095 	struct mlx5_dm		dm;
1096 	u16			devx_whitelist_uid;
1097 	struct mlx5_srq_table   srq_table;
1098 	struct mlx5_qp_table    qp_table;
1099 	struct mlx5_async_ctx   async_ctx;
1100 	struct mlx5_devx_event_table devx_event_table;
1101 	struct mlx5_var_table var_table;
1102 
1103 	struct xarray sig_mrs;
1104 	struct mlx5_port_caps port_caps[MLX5_MAX_PORTS];
1105 	u16 pkey_table_len;
1106 	u8 lag_ports;
1107 };
1108 
1109 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
1110 {
1111 	return container_of(mcq, struct mlx5_ib_cq, mcq);
1112 }
1113 
1114 static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
1115 {
1116 	return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd);
1117 }
1118 
1119 static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev)
1120 {
1121 	return container_of(ibdev, struct mlx5_ib_dev, ib_dev);
1122 }
1123 
1124 static inline struct mlx5_ib_dev *mr_to_mdev(struct mlx5_ib_mr *mr)
1125 {
1126 	return to_mdev(mr->ibmr.device);
1127 }
1128 
1129 static inline struct mlx5_ib_dev *mlx5_udata_to_mdev(struct ib_udata *udata)
1130 {
1131 	struct mlx5_ib_ucontext *context = rdma_udata_to_drv_context(
1132 		udata, struct mlx5_ib_ucontext, ibucontext);
1133 
1134 	return to_mdev(context->ibucontext.device);
1135 }
1136 
1137 static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq)
1138 {
1139 	return container_of(ibcq, struct mlx5_ib_cq, ibcq);
1140 }
1141 
1142 static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp)
1143 {
1144 	return container_of(mqp, struct mlx5_ib_qp_base, mqp)->container_mibqp;
1145 }
1146 
1147 static inline struct mlx5_ib_rwq *to_mibrwq(struct mlx5_core_qp *core_qp)
1148 {
1149 	return container_of(core_qp, struct mlx5_ib_rwq, core_qp);
1150 }
1151 
1152 static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd)
1153 {
1154 	return container_of(ibpd, struct mlx5_ib_pd, ibpd);
1155 }
1156 
1157 static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq)
1158 {
1159 	return container_of(ibsrq, struct mlx5_ib_srq, ibsrq);
1160 }
1161 
1162 static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp)
1163 {
1164 	return container_of(ibqp, struct mlx5_ib_qp, ibqp);
1165 }
1166 
1167 static inline struct mlx5_ib_rwq *to_mrwq(struct ib_wq *ibwq)
1168 {
1169 	return container_of(ibwq, struct mlx5_ib_rwq, ibwq);
1170 }
1171 
1172 static inline struct mlx5_ib_rwq_ind_table *to_mrwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl)
1173 {
1174 	return container_of(ib_rwq_ind_tbl, struct mlx5_ib_rwq_ind_table, ib_rwq_ind_tbl);
1175 }
1176 
1177 static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
1178 {
1179 	return container_of(msrq, struct mlx5_ib_srq, msrq);
1180 }
1181 
1182 static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
1183 {
1184 	return container_of(ibmr, struct mlx5_ib_mr, ibmr);
1185 }
1186 
1187 static inline struct mlx5_ib_mw *to_mmw(struct ib_mw *ibmw)
1188 {
1189 	return container_of(ibmw, struct mlx5_ib_mw, ibmw);
1190 }
1191 
1192 static inline struct mlx5_ib_flow_action *
1193 to_mflow_act(struct ib_flow_action *ibact)
1194 {
1195 	return container_of(ibact, struct mlx5_ib_flow_action, ib_action);
1196 }
1197 
1198 static inline struct mlx5_user_mmap_entry *
1199 to_mmmap(struct rdma_user_mmap_entry *rdma_entry)
1200 {
1201 	return container_of(rdma_entry,
1202 		struct mlx5_user_mmap_entry, rdma_entry);
1203 }
1204 
1205 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
1206 			struct mlx5_db *db);
1207 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);
1208 void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1209 void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1210 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index);
1211 int mlx5_ib_create_ah(struct ib_ah *ah, struct rdma_ah_init_attr *init_attr,
1212 		      struct ib_udata *udata);
1213 int mlx5_ib_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr);
1214 static inline int mlx5_ib_destroy_ah(struct ib_ah *ah, u32 flags)
1215 {
1216 	return 0;
1217 }
1218 int mlx5_ib_create_srq(struct ib_srq *srq, struct ib_srq_init_attr *init_attr,
1219 		       struct ib_udata *udata);
1220 int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
1221 		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
1222 int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr);
1223 int mlx5_ib_destroy_srq(struct ib_srq *srq, struct ib_udata *udata);
1224 int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
1225 			  const struct ib_recv_wr **bad_wr);
1226 int mlx5_ib_enable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1227 void mlx5_ib_disable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1228 int mlx5_ib_create_qp(struct ib_qp *qp, struct ib_qp_init_attr *init_attr,
1229 		      struct ib_udata *udata);
1230 int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1231 		      int attr_mask, struct ib_udata *udata);
1232 int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
1233 		     struct ib_qp_init_attr *qp_init_attr);
1234 int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata);
1235 void mlx5_ib_drain_sq(struct ib_qp *qp);
1236 void mlx5_ib_drain_rq(struct ib_qp *qp);
1237 int mlx5_ib_read_wqe_sq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1238 			size_t buflen, size_t *bc);
1239 int mlx5_ib_read_wqe_rq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1240 			size_t buflen, size_t *bc);
1241 int mlx5_ib_read_wqe_srq(struct mlx5_ib_srq *srq, int wqe_index, void *buffer,
1242 			 size_t buflen, size_t *bc);
1243 int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
1244 		      struct ib_udata *udata);
1245 int mlx5_ib_destroy_cq(struct ib_cq *cq, struct ib_udata *udata);
1246 int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
1247 int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
1248 int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1249 int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata);
1250 struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc);
1251 struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
1252 				  u64 virt_addr, int access_flags,
1253 				  struct ib_udata *udata);
1254 struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start,
1255 					 u64 length, u64 virt_addr,
1256 					 int fd, int access_flags,
1257 					 struct ib_udata *udata);
1258 int mlx5_ib_advise_mr(struct ib_pd *pd,
1259 		      enum ib_uverbs_advise_mr_advice advice,
1260 		      u32 flags,
1261 		      struct ib_sge *sg_list,
1262 		      u32 num_sge,
1263 		      struct uverbs_attr_bundle *attrs);
1264 int mlx5_ib_alloc_mw(struct ib_mw *mw, struct ib_udata *udata);
1265 int mlx5_ib_dealloc_mw(struct ib_mw *mw);
1266 struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd,
1267 					     int access_flags);
1268 void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *mr);
1269 void mlx5_ib_free_odp_mr(struct mlx5_ib_mr *mr);
1270 struct ib_mr *mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start,
1271 				    u64 length, u64 virt_addr, int access_flags,
1272 				    struct ib_pd *pd, struct ib_udata *udata);
1273 int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata);
1274 struct ib_mr *mlx5_ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
1275 			       u32 max_num_sg);
1276 struct ib_mr *mlx5_ib_alloc_mr_integrity(struct ib_pd *pd,
1277 					 u32 max_num_sg,
1278 					 u32 max_num_meta_sg);
1279 int mlx5_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents,
1280 		      unsigned int *sg_offset);
1281 int mlx5_ib_map_mr_sg_pi(struct ib_mr *ibmr, struct scatterlist *data_sg,
1282 			 int data_sg_nents, unsigned int *data_sg_offset,
1283 			 struct scatterlist *meta_sg, int meta_sg_nents,
1284 			 unsigned int *meta_sg_offset);
1285 int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u32 port_num,
1286 			const struct ib_wc *in_wc, const struct ib_grh *in_grh,
1287 			const struct ib_mad *in, struct ib_mad *out,
1288 			size_t *out_mad_size, u16 *out_mad_pkey_index);
1289 int mlx5_ib_alloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1290 int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1291 int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, unsigned int port);
1292 int mlx5_query_mad_ifc_system_image_guid(struct ib_device *ibdev,
1293 					 __be64 *sys_image_guid);
1294 int mlx5_query_mad_ifc_max_pkeys(struct ib_device *ibdev,
1295 				 u16 *max_pkeys);
1296 int mlx5_query_mad_ifc_vendor_id(struct ib_device *ibdev,
1297 				 u32 *vendor_id);
1298 int mlx5_query_mad_ifc_node_desc(struct mlx5_ib_dev *dev, char *node_desc);
1299 int mlx5_query_mad_ifc_node_guid(struct mlx5_ib_dev *dev, __be64 *node_guid);
1300 int mlx5_query_mad_ifc_pkey(struct ib_device *ibdev, u32 port, u16 index,
1301 			    u16 *pkey);
1302 int mlx5_query_mad_ifc_gids(struct ib_device *ibdev, u32 port, int index,
1303 			    union ib_gid *gid);
1304 int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u32 port,
1305 			    struct ib_port_attr *props);
1306 int mlx5_ib_query_port(struct ib_device *ibdev, u32 port,
1307 		       struct ib_port_attr *props);
1308 void mlx5_ib_populate_pas(struct ib_umem *umem, size_t page_size, __be64 *pas,
1309 			  u64 access_flags);
1310 void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num);
1311 int mlx5_ib_get_cqe_size(struct ib_cq *ibcq);
1312 int mlx5_mr_cache_init(struct mlx5_ib_dev *dev);
1313 int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev);
1314 
1315 struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev,
1316 				       struct mlx5_cache_ent *ent,
1317 				       int access_flags);
1318 
1319 int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
1320 			    struct ib_mr_status *mr_status);
1321 struct ib_wq *mlx5_ib_create_wq(struct ib_pd *pd,
1322 				struct ib_wq_init_attr *init_attr,
1323 				struct ib_udata *udata);
1324 int mlx5_ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
1325 int mlx5_ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1326 		      u32 wq_attr_mask, struct ib_udata *udata);
1327 int mlx5_ib_create_rwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_table,
1328 				 struct ib_rwq_ind_table_init_attr *init_attr,
1329 				 struct ib_udata *udata);
1330 int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
1331 struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm,
1332 				struct ib_dm_mr_attr *attr,
1333 				struct uverbs_attr_bundle *attrs);
1334 
1335 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1336 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);
1337 int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq);
1338 void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev);
1339 int __init mlx5_ib_odp_init(void);
1340 void mlx5_ib_odp_cleanup(void);
1341 void mlx5_odp_init_mr_cache_entry(struct mlx5_cache_ent *ent);
1342 void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1343 			   struct mlx5_ib_mr *mr, int flags);
1344 
1345 int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1346 			       enum ib_uverbs_advise_mr_advice advice,
1347 			       u32 flags, struct ib_sge *sg_list, u32 num_sge);
1348 int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr);
1349 int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr);
1350 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1351 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; }
1352 static inline int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev,
1353 				      struct mlx5_ib_pf_eq *eq)
1354 {
1355 	return 0;
1356 }
1357 static inline void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev) {}
1358 static inline int mlx5_ib_odp_init(void) { return 0; }
1359 static inline void mlx5_ib_odp_cleanup(void)				    {}
1360 static inline void mlx5_odp_init_mr_cache_entry(struct mlx5_cache_ent *ent) {}
1361 static inline void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1362 					 struct mlx5_ib_mr *mr, int flags) {}
1363 
1364 static inline int
1365 mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1366 			   enum ib_uverbs_advise_mr_advice advice, u32 flags,
1367 			   struct ib_sge *sg_list, u32 num_sge)
1368 {
1369 	return -EOPNOTSUPP;
1370 }
1371 static inline int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr)
1372 {
1373 	return -EOPNOTSUPP;
1374 }
1375 static inline int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr)
1376 {
1377 	return -EOPNOTSUPP;
1378 }
1379 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1380 
1381 extern const struct mmu_interval_notifier_ops mlx5_mn_ops;
1382 
1383 /* Needed for rep profile */
1384 void __mlx5_ib_remove(struct mlx5_ib_dev *dev,
1385 		      const struct mlx5_ib_profile *profile,
1386 		      int stage);
1387 int __mlx5_ib_add(struct mlx5_ib_dev *dev,
1388 		  const struct mlx5_ib_profile *profile);
1389 
1390 int mlx5_ib_get_vf_config(struct ib_device *device, int vf,
1391 			  u32 port, struct ifla_vf_info *info);
1392 int mlx5_ib_set_vf_link_state(struct ib_device *device, int vf,
1393 			      u32 port, int state);
1394 int mlx5_ib_get_vf_stats(struct ib_device *device, int vf,
1395 			 u32 port, struct ifla_vf_stats *stats);
1396 int mlx5_ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
1397 			struct ifla_vf_guid *node_guid,
1398 			struct ifla_vf_guid *port_guid);
1399 int mlx5_ib_set_vf_guid(struct ib_device *device, int vf, u32 port,
1400 			u64 guid, int type);
1401 
1402 __be16 mlx5_get_roce_udp_sport_min(const struct mlx5_ib_dev *dev,
1403 				   const struct ib_gid_attr *attr);
1404 
1405 void mlx5_ib_cleanup_cong_debugfs(struct mlx5_ib_dev *dev, u32 port_num);
1406 void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u32 port_num);
1407 
1408 /* GSI QP helper functions */
1409 int mlx5_ib_create_gsi(struct ib_pd *pd, struct mlx5_ib_qp *mqp,
1410 		       struct ib_qp_init_attr *attr);
1411 int mlx5_ib_destroy_gsi(struct mlx5_ib_qp *mqp);
1412 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
1413 			  int attr_mask);
1414 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
1415 			 int qp_attr_mask,
1416 			 struct ib_qp_init_attr *qp_init_attr);
1417 int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
1418 			  const struct ib_send_wr **bad_wr);
1419 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, const struct ib_recv_wr *wr,
1420 			  const struct ib_recv_wr **bad_wr);
1421 void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi);
1422 
1423 int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc);
1424 
1425 void mlx5_ib_free_bfreg(struct mlx5_ib_dev *dev, struct mlx5_bfreg_info *bfregi,
1426 			int bfregn);
1427 struct mlx5_ib_dev *mlx5_ib_get_ibdev_from_mpi(struct mlx5_ib_multiport_info *mpi);
1428 struct mlx5_core_dev *mlx5_ib_get_native_port_mdev(struct mlx5_ib_dev *dev,
1429 						   u32 ib_port_num,
1430 						   u32 *native_port_num);
1431 void mlx5_ib_put_native_port_mdev(struct mlx5_ib_dev *dev,
1432 				  u32 port_num);
1433 
1434 extern const struct uapi_definition mlx5_ib_devx_defs[];
1435 extern const struct uapi_definition mlx5_ib_flow_defs[];
1436 extern const struct uapi_definition mlx5_ib_qos_defs[];
1437 extern const struct uapi_definition mlx5_ib_std_types_defs[];
1438 
1439 static inline int is_qp1(enum ib_qp_type qp_type)
1440 {
1441 	return qp_type == MLX5_IB_QPT_HW_GSI || qp_type == IB_QPT_GSI;
1442 }
1443 
1444 static inline u32 check_cq_create_flags(u32 flags)
1445 {
1446 	/*
1447 	 * It returns non-zero value for unsupported CQ
1448 	 * create flags, otherwise it returns zero.
1449 	 */
1450 	return (flags & ~(IB_UVERBS_CQ_FLAGS_IGNORE_OVERRUN |
1451 			  IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION));
1452 }
1453 
1454 static inline int verify_assign_uidx(u8 cqe_version, u32 cmd_uidx,
1455 				     u32 *user_index)
1456 {
1457 	if (cqe_version) {
1458 		if ((cmd_uidx == MLX5_IB_DEFAULT_UIDX) ||
1459 		    (cmd_uidx & ~MLX5_USER_ASSIGNED_UIDX_MASK))
1460 			return -EINVAL;
1461 		*user_index = cmd_uidx;
1462 	} else {
1463 		*user_index = MLX5_IB_DEFAULT_UIDX;
1464 	}
1465 
1466 	return 0;
1467 }
1468 
1469 static inline int get_qp_user_index(struct mlx5_ib_ucontext *ucontext,
1470 				    struct mlx5_ib_create_qp *ucmd,
1471 				    int inlen,
1472 				    u32 *user_index)
1473 {
1474 	u8 cqe_version = ucontext->cqe_version;
1475 
1476 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1477 	    (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1478 		return 0;
1479 
1480 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1481 		return -EINVAL;
1482 
1483 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1484 }
1485 
1486 static inline int get_srq_user_index(struct mlx5_ib_ucontext *ucontext,
1487 				     struct mlx5_ib_create_srq *ucmd,
1488 				     int inlen,
1489 				     u32 *user_index)
1490 {
1491 	u8 cqe_version = ucontext->cqe_version;
1492 
1493 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1494 	    (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1495 		return 0;
1496 
1497 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1498 		return -EINVAL;
1499 
1500 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1501 }
1502 
1503 static inline int get_uars_per_sys_page(struct mlx5_ib_dev *dev, bool lib_support)
1504 {
1505 	return lib_support && MLX5_CAP_GEN(dev->mdev, uar_4k) ?
1506 				MLX5_UARS_IN_PAGE : 1;
1507 }
1508 
1509 extern void *xlt_emergency_page;
1510 
1511 int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
1512 			struct mlx5_bfreg_info *bfregi, u32 bfregn,
1513 			bool dyn_bfreg);
1514 
1515 static inline int mlx5r_store_odp_mkey(struct mlx5_ib_dev *dev,
1516 				       struct mlx5_ib_mkey *mmkey)
1517 {
1518 	refcount_set(&mmkey->usecount, 1);
1519 
1520 	return xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(mmkey->key),
1521 			       mmkey, GFP_KERNEL));
1522 }
1523 
1524 /* deref an mkey that can participate in ODP flow */
1525 static inline void mlx5r_deref_odp_mkey(struct mlx5_ib_mkey *mmkey)
1526 {
1527 	if (refcount_dec_and_test(&mmkey->usecount))
1528 		wake_up(&mmkey->wait);
1529 }
1530 
1531 /* deref an mkey that can participate in ODP flow and wait for relese */
1532 static inline void mlx5r_deref_wait_odp_mkey(struct mlx5_ib_mkey *mmkey)
1533 {
1534 	mlx5r_deref_odp_mkey(mmkey);
1535 	wait_event(mmkey->wait, refcount_read(&mmkey->usecount) == 0);
1536 }
1537 
1538 int mlx5_ib_test_wc(struct mlx5_ib_dev *dev);
1539 
1540 static inline bool mlx5_ib_lag_should_assign_affinity(struct mlx5_ib_dev *dev)
1541 {
1542 	return dev->lag_active ||
1543 		(MLX5_CAP_GEN(dev->mdev, num_lag_ports) > 1 &&
1544 		 MLX5_CAP_GEN(dev->mdev, lag_tx_port_affinity));
1545 }
1546 
1547 static inline bool rt_supported(int ts_cap)
1548 {
1549 	return ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_REAL_TIME ||
1550 	       ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME;
1551 }
1552 #endif /* MLX5_IB_H */
1553