xref: /linux/drivers/infiniband/hw/mlx5/mlx5_ib.h (revision fc2d791a43d3880496d1c729b8bd74d2c19cb4e7)
1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
2 /*
3  * Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved.
4  * Copyright (c) 2020, Intel Corporation. All rights reserved.
5  */
6 
7 #ifndef MLX5_IB_H
8 #define MLX5_IB_H
9 
10 #include <linux/kernel.h>
11 #include <linux/sched.h>
12 #include <rdma/ib_verbs.h>
13 #include <rdma/ib_umem.h>
14 #include <rdma/ib_smi.h>
15 #include <linux/mlx5/driver.h>
16 #include <linux/mlx5/cq.h>
17 #include <linux/mlx5/fs.h>
18 #include <linux/mlx5/qp.h>
19 #include <linux/types.h>
20 #include <linux/mlx5/transobj.h>
21 #include <rdma/ib_user_verbs.h>
22 #include <rdma/mlx5-abi.h>
23 #include <rdma/uverbs_ioctl.h>
24 #include <rdma/mlx5_user_ioctl_cmds.h>
25 #include <rdma/mlx5_user_ioctl_verbs.h>
26 
27 #include "srq.h"
28 #include "qp.h"
29 #include "macsec.h"
30 
31 #define mlx5_ib_dbg(_dev, format, arg...)                                      \
32 	dev_dbg(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,      \
33 		__LINE__, current->pid, ##arg)
34 
35 #define mlx5_ib_err(_dev, format, arg...)                                      \
36 	dev_err(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,      \
37 		__LINE__, current->pid, ##arg)
38 
39 #define mlx5_ib_warn(_dev, format, arg...)                                     \
40 	dev_warn(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,     \
41 		 __LINE__, current->pid, ##arg)
42 
43 #define mlx5_ib_log(lvl, _dev, format, arg...)                                 \
44 	dev_printk(lvl, &(_dev)->ib_dev.dev,  "%s:%d:(pid %d): " format,       \
45 		   __func__, __LINE__, current->pid, ##arg)
46 
47 #define MLX5_IB_DEFAULT_UIDX 0xffffff
48 #define MLX5_USER_ASSIGNED_UIDX_MASK __mlx5_mask(qpc, user_index)
49 
50 static __always_inline unsigned long
51 __mlx5_log_page_size_to_bitmap(unsigned int log_pgsz_bits,
52 			       unsigned int pgsz_shift)
53 {
54 	unsigned int largest_pg_shift =
55 		min_t(unsigned long, (1ULL << log_pgsz_bits) - 1 + pgsz_shift,
56 		      BITS_PER_LONG - 1);
57 
58 	/*
59 	 * Despite a command allowing it, the device does not support lower than
60 	 * 4k page size.
61 	 */
62 	pgsz_shift = max_t(unsigned int, MLX5_ADAPTER_PAGE_SHIFT, pgsz_shift);
63 	return GENMASK(largest_pg_shift, pgsz_shift);
64 }
65 
66 static __always_inline unsigned long
67 __mlx5_page_offset_to_bitmask(unsigned int page_offset_bits,
68 			      unsigned int offset_shift)
69 {
70 	unsigned int largest_offset_shift =
71 		min_t(unsigned long, page_offset_bits - 1 + offset_shift,
72 		      BITS_PER_LONG - 1);
73 
74 	return GENMASK(largest_offset_shift, offset_shift);
75 }
76 
77 /*
78  * QP/CQ/WQ/etc type commands take a page offset that satisifies:
79  *   page_offset_quantized * (page_size/scale) = page_offset
80  * Which restricts allowed page sizes to ones that satisify the above.
81  */
82 unsigned long __mlx5_umem_find_best_quantized_pgoff(
83 	struct ib_umem *umem, unsigned long pgsz_bitmap,
84 	unsigned int page_offset_bits, u64 pgoff_bitmask, unsigned int scale,
85 	unsigned int *page_offset_quantized);
86 #define mlx5_umem_find_best_quantized_pgoff(umem, typ, log_pgsz_fld,           \
87 					    pgsz_shift, page_offset_fld,       \
88 					    scale, page_offset_quantized)      \
89 	__mlx5_umem_find_best_quantized_pgoff(                                 \
90 		umem,                                                          \
91 		__mlx5_log_page_size_to_bitmap(                                \
92 			__mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift),         \
93 		__mlx5_bit_sz(typ, page_offset_fld),                           \
94 		GENMASK(31, order_base_2(scale)), scale,                       \
95 		page_offset_quantized)
96 
97 #define mlx5_umem_find_best_cq_quantized_pgoff(umem, typ, log_pgsz_fld,        \
98 					       pgsz_shift, page_offset_fld,    \
99 					       scale, page_offset_quantized)   \
100 	__mlx5_umem_find_best_quantized_pgoff(                                 \
101 		umem,                                                          \
102 		__mlx5_log_page_size_to_bitmap(                                \
103 			__mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift),         \
104 		__mlx5_bit_sz(typ, page_offset_fld), 0, scale,                 \
105 		page_offset_quantized)
106 
107 enum {
108 	MLX5_IB_MMAP_OFFSET_START = 9,
109 	MLX5_IB_MMAP_OFFSET_END = 255,
110 };
111 
112 enum {
113 	MLX5_IB_MMAP_CMD_SHIFT	= 8,
114 	MLX5_IB_MMAP_CMD_MASK	= 0xff,
115 };
116 
117 enum {
118 	MLX5_RES_SCAT_DATA32_CQE	= 0x1,
119 	MLX5_RES_SCAT_DATA64_CQE	= 0x2,
120 	MLX5_REQ_SCAT_DATA32_CQE	= 0x11,
121 	MLX5_REQ_SCAT_DATA64_CQE	= 0x22,
122 };
123 
124 enum mlx5_ib_mad_ifc_flags {
125 	MLX5_MAD_IFC_IGNORE_MKEY	= 1,
126 	MLX5_MAD_IFC_IGNORE_BKEY	= 2,
127 	MLX5_MAD_IFC_NET_VIEW		= 4,
128 };
129 
130 enum {
131 	MLX5_CROSS_CHANNEL_BFREG         = 0,
132 };
133 
134 enum {
135 	MLX5_CQE_VERSION_V0,
136 	MLX5_CQE_VERSION_V1,
137 };
138 
139 enum {
140 	MLX5_TM_MAX_RNDV_MSG_SIZE	= 64,
141 	MLX5_TM_MAX_SGE			= 1,
142 };
143 
144 enum {
145 	MLX5_IB_INVALID_UAR_INDEX	= BIT(31),
146 	MLX5_IB_INVALID_BFREG		= BIT(31),
147 };
148 
149 enum {
150 	MLX5_MAX_MEMIC_PAGES = 0x100,
151 	MLX5_MEMIC_ALLOC_SIZE_MASK = 0x3f,
152 };
153 
154 enum {
155 	MLX5_MEMIC_BASE_ALIGN	= 6,
156 	MLX5_MEMIC_BASE_SIZE	= 1 << MLX5_MEMIC_BASE_ALIGN,
157 };
158 
159 enum mlx5_ib_mmap_type {
160 	MLX5_IB_MMAP_TYPE_MEMIC = 1,
161 	MLX5_IB_MMAP_TYPE_VAR = 2,
162 	MLX5_IB_MMAP_TYPE_UAR_WC = 3,
163 	MLX5_IB_MMAP_TYPE_UAR_NC = 4,
164 	MLX5_IB_MMAP_TYPE_MEMIC_OP = 5,
165 	MLX5_IB_MMAP_TYPE_TLP_VAR = 6,
166 };
167 
168 struct mlx5_bfreg_info {
169 	u32 *sys_pages;
170 	int num_low_latency_bfregs;
171 	unsigned int *count;
172 
173 	/*
174 	 * protect bfreg allocation data structs
175 	 */
176 	struct mutex lock;
177 	u32 ver;
178 	u8 lib_uar_4k : 1;
179 	u8 lib_uar_dyn : 1;
180 	u32 num_sys_pages;
181 	u32 num_static_sys_pages;
182 	u32 total_num_bfregs;
183 	u32 num_dyn_bfregs;
184 };
185 
186 struct mlx5_ib_ucontext {
187 	struct ib_ucontext	ibucontext;
188 	struct list_head	db_page_list;
189 
190 	/* protect doorbell record alloc/free
191 	 */
192 	struct mutex		db_page_mutex;
193 	struct mlx5_bfreg_info	bfregi;
194 	u8			cqe_version;
195 	/* Transport Domain number */
196 	u32			tdn;
197 
198 	u64			lib_caps;
199 	u16			devx_uid;
200 	/* For RoCE LAG TX affinity */
201 	atomic_t		tx_port_affinity;
202 };
203 
204 static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
205 {
206 	return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext);
207 }
208 
209 struct mlx5_ib_pd {
210 	struct ib_pd		ibpd;
211 	u32			pdn;
212 	u16			uid;
213 };
214 
215 enum {
216 	MLX5_IB_FLOW_ACTION_MODIFY_HEADER,
217 	MLX5_IB_FLOW_ACTION_PACKET_REFORMAT,
218 	MLX5_IB_FLOW_ACTION_DECAP,
219 };
220 
221 #define MLX5_IB_FLOW_MCAST_PRIO		(MLX5_BY_PASS_NUM_PRIOS - 1)
222 #define MLX5_IB_FLOW_LAST_PRIO		(MLX5_BY_PASS_NUM_REGULAR_PRIOS - 1)
223 #if (MLX5_IB_FLOW_LAST_PRIO <= 0)
224 #error "Invalid number of bypass priorities"
225 #endif
226 #define MLX5_IB_FLOW_LEFTOVERS_PRIO	(MLX5_IB_FLOW_MCAST_PRIO + 1)
227 
228 #define MLX5_IB_NUM_FLOW_FT		(MLX5_IB_FLOW_LEFTOVERS_PRIO + 1)
229 #define MLX5_IB_NUM_SNIFFER_FTS		2
230 #define MLX5_IB_NUM_EGRESS_FTS		1
231 #define MLX5_IB_NUM_FDB_FTS		MLX5_BY_PASS_NUM_REGULAR_PRIOS
232 
233 struct mlx5_ib_anchor {
234 	struct mlx5_flow_table *ft;
235 	struct mlx5_flow_group *fg_goto_table;
236 	struct mlx5_flow_group *fg_drop;
237 	struct mlx5_flow_handle *rule_goto_table;
238 	struct mlx5_flow_handle *rule_drop;
239 	unsigned int rule_goto_table_ref;
240 };
241 
242 struct mlx5_ib_flow_prio {
243 	struct mlx5_flow_table		*flow_table;
244 	struct mlx5_ib_anchor		anchor;
245 	unsigned int			refcount;
246 };
247 
248 struct mlx5_ib_flow_handler {
249 	struct list_head		list;
250 	struct ib_flow			ibflow;
251 	struct mlx5_ib_flow_prio	*prio;
252 	struct mlx5_flow_handle		*rule;
253 	struct ib_counters		*ibcounters;
254 	struct mlx5_ib_dev		*dev;
255 	struct mlx5_ib_flow_matcher	*flow_matcher;
256 };
257 
258 struct mlx5_ib_flow_matcher {
259 	struct mlx5_ib_match_params matcher_mask;
260 	int			mask_len;
261 	enum mlx5_ib_flow_type	flow_type;
262 	enum mlx5_flow_namespace_type ns_type;
263 	u16			priority;
264 	struct mlx5_core_dev	*mdev;
265 	atomic_t		usecnt;
266 	u8			match_criteria_enable;
267 	u32			ib_port;
268 };
269 
270 struct mlx5_ib_steering_anchor {
271 	struct mlx5_ib_flow_prio *ft_prio;
272 	struct mlx5_ib_dev *dev;
273 	atomic_t usecnt;
274 };
275 
276 struct mlx5_ib_pp {
277 	u16 index;
278 	struct mlx5_core_dev *mdev;
279 };
280 
281 enum mlx5_ib_optional_counter_type {
282 	MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS,
283 	MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS,
284 	MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS,
285 	MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS,
286 	MLX5_IB_OPCOUNTER_RDMA_TX_BYTES,
287 	MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS,
288 	MLX5_IB_OPCOUNTER_RDMA_RX_BYTES,
289 
290 	MLX5_IB_OPCOUNTER_CC_RX_CE_PKTS_PER_QP,
291 	MLX5_IB_OPCOUNTER_CC_RX_CNP_PKTS_PER_QP,
292 	MLX5_IB_OPCOUNTER_CC_TX_CNP_PKTS_PER_QP,
293 	MLX5_IB_OPCOUNTER_RDMA_TX_PACKETS_PER_QP,
294 	MLX5_IB_OPCOUNTER_RDMA_TX_BYTES_PER_QP,
295 	MLX5_IB_OPCOUNTER_RDMA_RX_PACKETS_PER_QP,
296 	MLX5_IB_OPCOUNTER_RDMA_RX_BYTES_PER_QP,
297 
298 	MLX5_IB_OPCOUNTER_MAX,
299 };
300 
301 struct mlx5_ib_flow_db {
302 	struct mlx5_ib_flow_prio	prios[MLX5_IB_NUM_FLOW_FT];
303 	struct mlx5_ib_flow_prio	egress_prios[MLX5_IB_NUM_FLOW_FT];
304 	struct mlx5_ib_flow_prio	sniffer[MLX5_IB_NUM_SNIFFER_FTS];
305 	struct mlx5_ib_flow_prio	egress[MLX5_IB_NUM_EGRESS_FTS];
306 	struct mlx5_ib_flow_prio	fdb[MLX5_IB_NUM_FDB_FTS];
307 	struct mlx5_ib_flow_prio	rdma_rx[MLX5_IB_NUM_FLOW_FT];
308 	struct mlx5_ib_flow_prio	rdma_tx[MLX5_IB_NUM_FLOW_FT];
309 	struct mlx5_ib_flow_prio	opfcs[MLX5_IB_OPCOUNTER_MAX];
310 	struct mlx5_ib_flow_prio        *rdma_transport_rx[MLX5_RDMA_TRANSPORT_BYPASS_PRIO];
311 	struct mlx5_ib_flow_prio        *rdma_transport_tx[MLX5_RDMA_TRANSPORT_BYPASS_PRIO];
312 	/* Protect flow steering bypass flow tables
313 	 * when add/del flow rules.
314 	 * only single add/removal of flow steering rule could be done
315 	 * simultaneously.
316 	 */
317 	struct mutex			lock;
318 };
319 
320 /* Use macros here so that don't have to duplicate
321  * enum ib_qp_type for low-level driver
322  */
323 
324 #define MLX5_IB_QPT_REG_UMR	IB_QPT_RESERVED1
325 /*
326  * IB_QPT_GSI creates the software wrapper around GSI, and MLX5_IB_QPT_HW_GSI
327  * creates the actual hardware QP.
328  */
329 #define MLX5_IB_QPT_HW_GSI	IB_QPT_RESERVED2
330 #define MLX5_IB_QPT_DCI		IB_QPT_RESERVED3
331 #define MLX5_IB_QPT_DCT		IB_QPT_RESERVED4
332 #define MLX5_IB_WR_UMR		IB_WR_RESERVED1
333 
334 #define MLX5_IB_UPD_XLT_ZAP	      BIT(0)
335 #define MLX5_IB_UPD_XLT_ENABLE	      BIT(1)
336 #define MLX5_IB_UPD_XLT_ATOMIC	      BIT(2)
337 #define MLX5_IB_UPD_XLT_ADDR	      BIT(3)
338 #define MLX5_IB_UPD_XLT_PD	      BIT(4)
339 #define MLX5_IB_UPD_XLT_ACCESS	      BIT(5)
340 #define MLX5_IB_UPD_XLT_INDIRECT      BIT(6)
341 #define MLX5_IB_UPD_XLT_DOWNGRADE     BIT(7)
342 #define MLX5_IB_UPD_XLT_KEEP_PGSZ     BIT(8)
343 
344 /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags.
345  *
346  * These flags are intended for internal use by the mlx5_ib driver, and they
347  * rely on the range reserved for that use in the ib_qp_create_flags enum.
348  */
349 #define MLX5_IB_QP_CREATE_SQPN_QP1	IB_QP_CREATE_RESERVED_START
350 
351 struct wr_list {
352 	u16	opcode;
353 	u16	next;
354 };
355 
356 enum mlx5_ib_rq_flags {
357 	MLX5_IB_RQ_CVLAN_STRIPPING	= 1 << 0,
358 	MLX5_IB_RQ_PCI_WRITE_END_PADDING	= 1 << 1,
359 };
360 
361 struct mlx5_ib_wq {
362 	struct mlx5_frag_buf_ctrl fbc;
363 	u64		       *wrid;
364 	u32		       *wr_data;
365 	struct wr_list	       *w_list;
366 	unsigned	       *wqe_head;
367 	u16		        unsig_count;
368 
369 	/* serialize post to the work queue
370 	 */
371 	spinlock_t		lock;
372 	int			wqe_cnt;
373 	int			max_post;
374 	int			max_gs;
375 	int			offset;
376 	int			wqe_shift;
377 	unsigned		head;
378 	unsigned		tail;
379 	u16			cur_post;
380 	u16			last_poll;
381 	void			*cur_edge;
382 };
383 
384 enum mlx5_ib_wq_flags {
385 	MLX5_IB_WQ_FLAGS_DELAY_DROP = 0x1,
386 	MLX5_IB_WQ_FLAGS_STRIDING_RQ = 0x2,
387 };
388 
389 #define MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES 9
390 #define MLX5_MAX_SINGLE_WQE_LOG_NUM_STRIDES 16
391 #define MLX5_MIN_SINGLE_STRIDE_LOG_NUM_BYTES 6
392 #define MLX5_MAX_SINGLE_STRIDE_LOG_NUM_BYTES 13
393 #define MLX5_EXT_MIN_SINGLE_WQE_LOG_NUM_STRIDES 3
394 
395 struct mlx5_ib_rwq {
396 	struct ib_wq		ibwq;
397 	struct mlx5_core_qp	core_qp;
398 	u32			rq_num_pas;
399 	u32			log_rq_stride;
400 	u32			log_rq_size;
401 	u32			rq_page_offset;
402 	u32			log_page_size;
403 	u32			log_num_strides;
404 	u32			two_byte_shift_en;
405 	u32			single_stride_log_num_of_bytes;
406 	struct ib_umem		*umem;
407 	size_t			buf_size;
408 	unsigned int		page_shift;
409 	struct mlx5_db		db;
410 	u32			user_index;
411 	u32			wqe_count;
412 	u32			wqe_shift;
413 	int			wq_sig;
414 	u32			create_flags; /* Use enum mlx5_ib_wq_flags */
415 };
416 
417 struct mlx5_ib_rwq_ind_table {
418 	struct ib_rwq_ind_table ib_rwq_ind_tbl;
419 	u32			rqtn;
420 	u16			uid;
421 };
422 
423 struct mlx5_ib_ubuffer {
424 	struct ib_umem	       *umem;
425 	int			buf_size;
426 	u64			buf_addr;
427 };
428 
429 struct mlx5_ib_qp_base {
430 	struct mlx5_ib_qp	*container_mibqp;
431 	struct mlx5_core_qp	mqp;
432 	struct mlx5_ib_ubuffer	ubuffer;
433 };
434 
435 struct mlx5_ib_qp_trans {
436 	struct mlx5_ib_qp_base	base;
437 	u16			xrcdn;
438 	u32			alt_port;
439 	u8			atomic_rd_en;
440 	u8			resp_depth;
441 };
442 
443 struct mlx5_ib_rss_qp {
444 	u32	tirn;
445 };
446 
447 struct mlx5_ib_rq {
448 	struct mlx5_ib_qp_base base;
449 	struct mlx5_ib_wq	*rq;
450 	struct mlx5_ib_ubuffer	ubuffer;
451 	struct mlx5_db		*doorbell;
452 	u32			tirn;
453 	u8			state;
454 	u32			flags;
455 };
456 
457 struct mlx5_ib_sq {
458 	struct mlx5_ib_qp_base base;
459 	struct mlx5_ib_wq	*sq;
460 	struct mlx5_ib_ubuffer  ubuffer;
461 	struct mlx5_db		*doorbell;
462 	struct mlx5_flow_handle	*flow_rule;
463 	u32			tisn;
464 	u8			state;
465 };
466 
467 struct mlx5_ib_raw_packet_qp {
468 	struct mlx5_ib_sq sq;
469 	struct mlx5_ib_rq rq;
470 };
471 
472 struct mlx5_bf {
473 	int			buf_size;
474 	unsigned long		offset;
475 	struct mlx5_sq_bfreg   *bfreg;
476 };
477 
478 struct mlx5_ib_dct {
479 	struct mlx5_core_dct    mdct;
480 	u32                     *in;
481 };
482 
483 struct mlx5_ib_gsi_qp {
484 	struct ib_qp *rx_qp;
485 	u32 port_num;
486 	struct ib_qp_cap cap;
487 	struct ib_cq *cq;
488 	struct mlx5_ib_gsi_wr *outstanding_wrs;
489 	u32 outstanding_pi, outstanding_ci;
490 	int num_qps;
491 	/* Protects access to the tx_qps. Post send operations synchronize
492 	 * with tx_qp creation in setup_qp(). Also protects the
493 	 * outstanding_wrs array and indices.
494 	 */
495 	spinlock_t lock;
496 	struct ib_qp **tx_qps;
497 };
498 
499 struct mlx5_ib_qp {
500 	struct ib_qp		ibqp;
501 	union {
502 		struct mlx5_ib_qp_trans trans_qp;
503 		struct mlx5_ib_raw_packet_qp raw_packet_qp;
504 		struct mlx5_ib_rss_qp rss_qp;
505 		struct mlx5_ib_dct dct;
506 		struct mlx5_ib_gsi_qp gsi;
507 	};
508 	struct mlx5_frag_buf	buf;
509 
510 	struct mlx5_db		db;
511 	struct mlx5_ib_wq	rq;
512 
513 	u8			sq_signal_bits;
514 	u8			next_fence;
515 	struct mlx5_ib_wq	sq;
516 
517 	/* serialize qp state modifications
518 	 */
519 	struct mutex		mutex;
520 	/* cached variant of create_flags from struct ib_qp_init_attr */
521 	u32			flags;
522 	u32			port;
523 	u8			state;
524 	int			max_inline_data;
525 	struct mlx5_bf	        bf;
526 	u8			has_rq:1;
527 	u8			is_rss:1;
528 	u8			is_ooo_rq:1;
529 
530 	/* only for user space QPs. For kernel
531 	 * we have it from the bf object
532 	 */
533 	int			bfregn;
534 
535 	struct list_head	qps_list;
536 	struct list_head	cq_recv_list;
537 	struct list_head	cq_send_list;
538 	struct mlx5_rate_limit	rl;
539 	u32                     underlay_qpn;
540 	u32			flags_en;
541 	/*
542 	 * IB/core doesn't store low-level QP types, so
543 	 * store both MLX and IBTA types in the field below.
544 	 */
545 	enum ib_qp_type		type;
546 	/* A flag to indicate if there's a new counter is configured
547 	 * but not take effective
548 	 */
549 	u32                     counter_pending;
550 	u16			gsi_lag_port;
551 };
552 
553 struct mlx5_ib_cq_buf {
554 	struct mlx5_frag_buf_ctrl fbc;
555 	struct mlx5_frag_buf    frag_buf;
556 	struct ib_umem		*umem;
557 	int			cqe_size;
558 	int			nent;
559 };
560 
561 enum mlx5_ib_cq_pr_flags {
562 	MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD	= 1 << 0,
563 	MLX5_IB_CQ_PR_FLAGS_REAL_TIME_TS = 1 << 1,
564 	MLX5_IB_CQ_PR_TIMESTAMP_COMPLETION = 1 << 2,
565 };
566 
567 struct mlx5_ib_cq {
568 	struct ib_cq		ibcq;
569 	struct mlx5_core_cq	mcq;
570 	struct mlx5_ib_cq_buf	buf;
571 	struct mlx5_db		db;
572 
573 	/* serialize access to the CQ
574 	 */
575 	spinlock_t		lock;
576 
577 	/* protect resize cq
578 	 */
579 	struct mutex		resize_mutex;
580 	struct mlx5_ib_cq_buf  *resize_buf;
581 	struct ib_umem	       *resize_umem;
582 	int			cqe_size;
583 	struct list_head	list_send_qp;
584 	struct list_head	list_recv_qp;
585 	struct list_head	wc_list;
586 	enum ib_cq_notify_flags notify_flags;
587 	struct work_struct	notify_work;
588 	u16			private_flags; /* Use mlx5_ib_cq_pr_flags */
589 };
590 
591 struct mlx5_ib_wc {
592 	struct ib_wc wc;
593 	struct list_head list;
594 };
595 
596 struct mlx5_ib_srq {
597 	struct ib_srq		ibsrq;
598 	struct mlx5_core_srq	msrq;
599 	struct mlx5_frag_buf	buf;
600 	struct mlx5_db		db;
601 	struct mlx5_frag_buf_ctrl fbc;
602 	u64		       *wrid;
603 	/* protect SRQ hanlding
604 	 */
605 	spinlock_t		lock;
606 	int			head;
607 	int			tail;
608 	u16			wqe_ctr;
609 	struct ib_umem	       *umem;
610 	/* serialize arming a SRQ
611 	 */
612 	struct mutex		mutex;
613 	int			wq_sig;
614 };
615 
616 struct mlx5_ib_xrcd {
617 	struct ib_xrcd		ibxrcd;
618 	u32			xrcdn;
619 };
620 
621 enum mlx5_ib_mtt_access_flags {
622 	MLX5_IB_MTT_READ  = (1 << 0),
623 	MLX5_IB_MTT_WRITE = (1 << 1),
624 };
625 
626 struct mlx5_user_mmap_entry {
627 	struct rdma_user_mmap_entry rdma_entry;
628 	u8 mmap_flag;
629 	u64 address;
630 	u32 page_idx;
631 };
632 
633 enum mlx5_mkey_type {
634 	MLX5_MKEY_MR = 1,
635 	MLX5_MKEY_MW,
636 	MLX5_MKEY_INDIRECT_DEVX,
637 	MLX5_MKEY_NULL,
638 	MLX5_MKEY_IMPLICIT_CHILD,
639 };
640 
641 /* Used for non-existent ph value */
642 #define MLX5_IB_NO_PH 0xff
643 
644 struct mlx5_ib_mkey {
645 	u32 key;
646 	enum mlx5_mkey_type type;
647 	unsigned int ndescs;
648 	struct wait_queue_head wait;
649 	refcount_t usecount;
650 };
651 
652 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
653 
654 #define MLX5_IB_DM_MEMIC_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE   |\
655 					 IB_ACCESS_REMOTE_WRITE  |\
656 					 IB_ACCESS_REMOTE_READ   |\
657 					 IB_ACCESS_REMOTE_ATOMIC |\
658 					 IB_ZERO_BASED)
659 
660 #define MLX5_IB_DM_SW_ICM_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE   |\
661 					  IB_ACCESS_REMOTE_WRITE  |\
662 					  IB_ACCESS_REMOTE_READ   |\
663 					  IB_ZERO_BASED)
664 
665 #define mlx5_update_odp_stats(mr, counter_name, value)		\
666 	atomic64_add(value, &((mr)->odp_stats.counter_name))
667 
668 #define mlx5_update_odp_stats_with_handled(mr, counter_name, value)         \
669 	do {                                                                \
670 		mlx5_update_odp_stats(mr, counter_name, value);             \
671 		atomic64_add(1, &((mr)->odp_stats.counter_name##_handled)); \
672 	} while (0)
673 
674 struct mlx5_ib_mr {
675 	struct ib_mr ibmr;
676 	struct mlx5_ib_mkey mmkey;
677 
678 	struct ib_umem *umem;
679 	/* The mr is data direct related */
680 	u8 data_direct :1;
681 
682 	union {
683 		/* Used only by kernel MRs (umem == NULL) */
684 		struct {
685 			void *descs;
686 			void *descs_alloc;
687 			dma_addr_t desc_map;
688 			int max_descs;
689 			int desc_size;
690 			int access_mode;
691 
692 			/* For Kernel IB_MR_TYPE_INTEGRITY */
693 			struct mlx5_core_sig_ctx *sig;
694 			struct mlx5_ib_mr *pi_mr;
695 			struct mlx5_ib_mr *klm_mr;
696 			struct mlx5_ib_mr *mtt_mr;
697 			u64 data_iova;
698 			u64 pi_iova;
699 			int meta_ndescs;
700 			int meta_length;
701 			int data_length;
702 		};
703 
704 		/* Used only by User MRs (umem != NULL) */
705 		struct {
706 			unsigned int page_shift;
707 			/* Current access_flags */
708 			int access_flags;
709 
710 			/* For User ODP */
711 			struct mlx5_ib_mr *parent;
712 			struct xarray implicit_children;
713 			union {
714 				struct work_struct work;
715 			} odp_destroy;
716 			struct ib_odp_counters odp_stats;
717 			bool is_odp_implicit;
718 			/* The affilated data direct crossed mr */
719 			struct mlx5_ib_mr *dd_crossed_mr;
720 			struct list_head dd_node;
721 			u8 revoked :1;
722 			/* Indicates previous dmabuf page fault occurred */
723 			u8 dmabuf_faulted:1;
724 			struct mlx5_ib_mkey null_mmkey;
725 		};
726 	};
727 };
728 
729 static inline bool is_odp_mr(struct mlx5_ib_mr *mr)
730 {
731 	return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
732 	       mr->umem->is_odp;
733 }
734 
735 static inline bool is_dmabuf_mr(struct mlx5_ib_mr *mr)
736 {
737 	return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
738 	       mr->umem->is_dmabuf;
739 }
740 
741 struct mlx5_ib_mw {
742 	struct ib_mw		ibmw;
743 	struct mlx5_ib_mkey	mmkey;
744 };
745 
746 struct mlx5_ib_umr_context {
747 	struct ib_cqe		cqe;
748 	enum ib_wc_status	status;
749 	struct completion	done;
750 };
751 
752 enum {
753 	MLX5_UMR_STATE_UNINIT,
754 	MLX5_UMR_STATE_ACTIVE,
755 	MLX5_UMR_STATE_RECOVER,
756 	MLX5_UMR_STATE_ERR,
757 };
758 
759 struct umr_common {
760 	struct ib_pd	*pd;
761 	struct ib_cq	*cq;
762 	struct ib_qp	*qp;
763 	/* Protects from UMR QP overflow
764 	 */
765 	struct semaphore	sem;
766 	/* Protects from using UMR while the UMR is not active
767 	 */
768 	struct mutex lock;
769 	unsigned int state;
770 	/* Protects from repeat UMR QP creation */
771 	struct mutex init_lock;
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_data_direct_resources {
780 	u32 pdn;
781 	u32 mkey;
782 	u32 mkey_ro;
783 	u8 mkey_ro_valid :1;
784 };
785 
786 struct mlx5_ib_resources {
787 	struct ib_cq	*c0;
788 	struct mutex cq_lock;
789 	u32 xrcdn0;
790 	u32 xrcdn1;
791 	struct ib_pd	*p0;
792 	struct ib_srq	*s0;
793 	struct ib_srq	*s1;
794 	struct mutex srq_lock;
795 	struct mlx5_ib_port_resources ports[2];
796 };
797 
798 #define MAX_OPFC_RULES 2
799 
800 struct mlx5_ib_op_fc {
801 	struct mlx5_fc *fc;
802 	struct mlx5_flow_handle *rule[MAX_OPFC_RULES];
803 };
804 
805 struct mlx5_ib_counters {
806 	struct rdma_stat_desc *descs;
807 	size_t *offsets;
808 	u32 num_q_counters;
809 	u32 num_cong_counters;
810 	u32 num_ext_ppcnt_counters;
811 	u32 num_op_counters;
812 	u16 set_id;
813 	struct mlx5_ib_op_fc opfcs[MLX5_IB_OPCOUNTER_MAX];
814 };
815 
816 int mlx5_ib_fs_add_op_fc(struct mlx5_ib_dev *dev, u32 port_num,
817 			 struct mlx5_ib_op_fc *opfc,
818 			 enum mlx5_ib_optional_counter_type type);
819 
820 void mlx5_ib_fs_remove_op_fc(struct mlx5_ib_dev *dev,
821 			     struct mlx5_ib_op_fc *opfc,
822 			     enum mlx5_ib_optional_counter_type type);
823 
824 int mlx5r_fs_bind_op_fc(struct ib_qp *qp,
825 			struct mlx5_fc *fc_arr[MLX5_IB_OPCOUNTER_MAX],
826 			struct xarray *qpn_opfc_xa, u32 port);
827 
828 void mlx5r_fs_unbind_op_fc(struct ib_qp *qp, struct xarray *qpn_opfc_xa);
829 
830 void mlx5r_fs_destroy_fcs(struct mlx5_ib_dev *dev,
831 			  struct mlx5_fc *fc_arr[MLX5_IB_OPCOUNTER_MAX]);
832 
833 struct mlx5_ib_multiport_info;
834 
835 struct mlx5_ib_multiport {
836 	struct mlx5_ib_multiport_info *mpi;
837 	/* To be held when accessing the multiport info */
838 	spinlock_t mpi_lock;
839 };
840 
841 struct mlx5_roce {
842 	/* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL
843 	 * netdev pointer
844 	 */
845 	struct notifier_block	nb;
846 	struct netdev_net_notifier nn;
847 	struct notifier_block	mdev_nb;
848 	struct net_device	*tracking_netdev;
849 	atomic_t		tx_port_affinity;
850 	enum ib_port_state last_port_state;
851 	struct mlx5_ib_dev	*dev;
852 	u32			native_port_num;
853 };
854 
855 struct mlx5_ib_port {
856 	struct mlx5_ib_counters cnts;
857 	struct mlx5_ib_multiport mp;
858 	struct mlx5_ib_dbg_cc_params *dbg_cc_params;
859 	struct mlx5_roce roce;
860 	struct mlx5_eswitch_rep		*rep;
861 #ifdef CONFIG_MLX5_MACSEC
862 	struct mlx5_reserved_gids *reserved_gids;
863 #endif
864 };
865 
866 struct mlx5_ib_dbg_param {
867 	int			offset;
868 	struct mlx5_ib_dev	*dev;
869 	struct dentry		*dentry;
870 	u32			port_num;
871 };
872 
873 enum mlx5_ib_dbg_cc_types {
874 	MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE,
875 	MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE_ATI,
876 	MLX5_IB_DBG_CC_RP_TIME_RESET,
877 	MLX5_IB_DBG_CC_RP_BYTE_RESET,
878 	MLX5_IB_DBG_CC_RP_THRESHOLD,
879 	MLX5_IB_DBG_CC_RP_AI_RATE,
880 	MLX5_IB_DBG_CC_RP_MAX_RATE,
881 	MLX5_IB_DBG_CC_RP_HAI_RATE,
882 	MLX5_IB_DBG_CC_RP_MIN_DEC_FAC,
883 	MLX5_IB_DBG_CC_RP_MIN_RATE,
884 	MLX5_IB_DBG_CC_RP_RATE_TO_SET_ON_FIRST_CNP,
885 	MLX5_IB_DBG_CC_RP_DCE_TCP_G,
886 	MLX5_IB_DBG_CC_RP_DCE_TCP_RTT,
887 	MLX5_IB_DBG_CC_RP_RATE_REDUCE_MONITOR_PERIOD,
888 	MLX5_IB_DBG_CC_RP_INITIAL_ALPHA_VALUE,
889 	MLX5_IB_DBG_CC_RP_GD,
890 	MLX5_IB_DBG_CC_NP_MIN_TIME_BETWEEN_CNPS,
891 	MLX5_IB_DBG_CC_NP_CNP_DSCP,
892 	MLX5_IB_DBG_CC_NP_CNP_PRIO_MODE,
893 	MLX5_IB_DBG_CC_NP_CNP_PRIO,
894 	MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP_VALID,
895 	MLX5_IB_DBG_CC_GENERAL_RTT_RESP_DSCP,
896 	MLX5_IB_DBG_CC_MAX,
897 };
898 
899 struct mlx5_ib_dbg_cc_params {
900 	struct dentry			*root;
901 	struct mlx5_ib_dbg_param	params[MLX5_IB_DBG_CC_MAX];
902 };
903 
904 enum {
905 	MLX5_MAX_DELAY_DROP_TIMEOUT_MS = 100,
906 };
907 
908 struct mlx5_ib_delay_drop {
909 	struct mlx5_ib_dev     *dev;
910 	struct work_struct	delay_drop_work;
911 	/* serialize setting of delay drop */
912 	struct mutex		lock;
913 	u32			timeout;
914 	bool			activate;
915 	atomic_t		events_cnt;
916 	atomic_t		rqs_cnt;
917 	struct dentry		*dir_debugfs;
918 };
919 
920 enum mlx5_ib_stages {
921 	MLX5_IB_STAGE_INIT,
922 	MLX5_IB_STAGE_FS,
923 	MLX5_IB_STAGE_CAPS,
924 	MLX5_IB_STAGE_NON_DEFAULT_CB,
925 	MLX5_IB_STAGE_ROCE,
926 	MLX5_IB_STAGE_QP,
927 	MLX5_IB_STAGE_SRQ,
928 	MLX5_IB_STAGE_DEVICE_RESOURCES,
929 	MLX5_IB_STAGE_ODP,
930 	MLX5_IB_STAGE_COUNTERS,
931 	MLX5_IB_STAGE_CONG_DEBUGFS,
932 	MLX5_IB_STAGE_BFREG,
933 	MLX5_IB_STAGE_PRE_IB_REG_UMR,
934 	MLX5_IB_STAGE_WHITELIST_UID,
935 	MLX5_IB_STAGE_SYS_ERROR_NOTIFIER,
936 	MLX5_IB_STAGE_IB_REG,
937 	MLX5_IB_STAGE_DEVICE_NOTIFIER,
938 	MLX5_IB_STAGE_POST_IB_REG_UMR,
939 	MLX5_IB_STAGE_DELAY_DROP,
940 	MLX5_IB_STAGE_RESTRACK,
941 	MLX5_IB_STAGE_MAX,
942 };
943 
944 struct mlx5_ib_stage {
945 	int (*init)(struct mlx5_ib_dev *dev);
946 	void (*cleanup)(struct mlx5_ib_dev *dev);
947 };
948 
949 #define STAGE_CREATE(_stage, _init, _cleanup) \
950 	.stage[_stage] = {.init = _init, .cleanup = _cleanup}
951 
952 struct mlx5_ib_profile {
953 	struct mlx5_ib_stage stage[MLX5_IB_STAGE_MAX];
954 };
955 
956 struct mlx5_ib_multiport_info {
957 	struct list_head list;
958 	struct mlx5_ib_dev *ibdev;
959 	struct mlx5_core_dev *mdev;
960 	struct notifier_block mdev_events;
961 	struct completion unref_comp;
962 	u64 sys_image_guid;
963 	u32 mdev_refcnt;
964 	bool is_master;
965 	bool unaffiliate;
966 };
967 
968 struct mlx5_ib_flow_action {
969 	struct ib_flow_action		ib_action;
970 	union {
971 		struct {
972 			u64			    ib_flags;
973 			struct mlx5_accel_esp_xfrm *ctx;
974 		} esp_aes_gcm;
975 		struct {
976 			struct mlx5_ib_dev *dev;
977 			u32 sub_type;
978 			union {
979 				struct mlx5_modify_hdr *modify_hdr;
980 				struct mlx5_pkt_reformat *pkt_reformat;
981 			};
982 		} flow_action_raw;
983 	};
984 };
985 
986 struct mlx5_dm {
987 	struct mlx5_core_dev *dev;
988 	/* This lock is used to protect the access to the shared
989 	 * allocation map when concurrent requests by different
990 	 * processes are handled.
991 	 */
992 	spinlock_t lock;
993 	DECLARE_BITMAP(memic_alloc_pages, MLX5_MAX_MEMIC_PAGES);
994 };
995 
996 struct mlx5_read_counters_attr {
997 	struct mlx5_fc *hw_cntrs_hndl;
998 	u64 *out;
999 	u32 flags;
1000 };
1001 
1002 enum mlx5_ib_counters_type {
1003 	MLX5_IB_COUNTERS_FLOW,
1004 };
1005 
1006 struct mlx5_ib_mcounters {
1007 	struct ib_counters ibcntrs;
1008 	enum mlx5_ib_counters_type type;
1009 	/* number of counters supported for this counters type */
1010 	u32 counters_num;
1011 	struct mlx5_fc *hw_cntrs_hndl;
1012 	/* read function for this counters type */
1013 	int (*read_counters)(struct ib_device *ibdev,
1014 			     struct mlx5_read_counters_attr *read_attr);
1015 	/* max index set as part of create_flow */
1016 	u32 cntrs_max_index;
1017 	/* number of counters data entries (<description,index> pair) */
1018 	u32 ncounters;
1019 	/* counters data array for descriptions and indexes */
1020 	struct mlx5_ib_flow_counters_desc *counters_data;
1021 	/* protects access to mcounters internal data */
1022 	struct mutex mcntrs_mutex;
1023 };
1024 
1025 static inline struct mlx5_ib_mcounters *
1026 to_mcounters(struct ib_counters *ibcntrs)
1027 {
1028 	return container_of(ibcntrs, struct mlx5_ib_mcounters, ibcntrs);
1029 }
1030 
1031 int parse_flow_flow_action(struct mlx5_ib_flow_action *maction,
1032 			   bool is_egress,
1033 			   struct mlx5_flow_act *action);
1034 struct mlx5_ib_lb_state {
1035 	/* protect the user_td */
1036 	struct mutex		mutex;
1037 	u32			user_td;
1038 	int			qps;
1039 	bool			enabled;
1040 	bool			force_enable;
1041 };
1042 
1043 struct mlx5_ib_pf_eq {
1044 	struct notifier_block irq_nb;
1045 	struct mlx5_ib_dev *dev;
1046 	struct mlx5_eq *core;
1047 	struct work_struct work;
1048 	spinlock_t lock; /* Pagefaults spinlock */
1049 	struct workqueue_struct *wq;
1050 	mempool_t *pool;
1051 };
1052 
1053 struct mlx5_devx_event_table {
1054 	struct mlx5_nb devx_nb;
1055 	/* serialize updating the event_xa */
1056 	struct mutex event_xa_lock;
1057 	struct xarray event_xa;
1058 };
1059 
1060 struct mlx5_var_region {
1061 	/* serialize updating the bitmap */
1062 	struct mutex bitmap_lock;
1063 	unsigned long *bitmap;
1064 	u64 hw_start_addr;
1065 	u32 stride_size;
1066 	u64 num_var_hw_entries;
1067 };
1068 
1069 struct mlx5_var_table {
1070 	struct mlx5_var_region var_region;
1071 	struct mlx5_var_region tlp_var_region;
1072 };
1073 
1074 struct mlx5_port_caps {
1075 	bool has_smi;
1076 	u8 ext_port_cap;
1077 };
1078 
1079 
1080 struct mlx5_special_mkeys {
1081 	u32 dump_fill_mkey;
1082 	__be32 null_mkey;
1083 	__be32 terminate_scatter_list_mkey;
1084 };
1085 
1086 struct mlx5_macsec {
1087 	struct mutex lock; /* Protects mlx5_macsec internal contexts */
1088 	struct list_head macsec_devices_list;
1089 	struct notifier_block blocking_events_nb;
1090 };
1091 
1092 struct mlx5_ib_dev {
1093 	struct ib_device		ib_dev;
1094 	struct mlx5_core_dev		*mdev;
1095 	struct mlx5_data_direct_dev	*data_direct_dev;
1096 	/* protect accessing data_direct_dev */
1097 	struct mutex			data_direct_lock;
1098 	struct notifier_block		mdev_events;
1099 	struct notifier_block		sys_error_events;
1100 	struct notifier_block           lag_events;
1101 	int				num_ports;
1102 	/* serialize update of capability mask
1103 	 */
1104 	struct mutex			cap_mask_mutex;
1105 	u8				ib_active:1;
1106 	u8				is_rep:1;
1107 	u8				lag_active:1;
1108 	u8				fill_delay;
1109 	struct umr_common		umrc;
1110 	/* sync used page count stats
1111 	 */
1112 	struct mlx5_ib_resources	devr;
1113 
1114 	atomic_t			mkey_var;
1115 	/* Prevents soft lock on massive reg MRs */
1116 	struct mutex			slow_path_mutex;
1117 	struct ib_odp_caps	odp_caps;
1118 	u64			odp_max_size;
1119 	struct mutex		odp_eq_mutex;
1120 	struct mlx5_ib_pf_eq	odp_pf_eq;
1121 
1122 	struct xarray		odp_mkeys;
1123 
1124 	struct mlx5_ib_flow_db	*flow_db;
1125 	/* protect resources needed as part of reset flow */
1126 	spinlock_t		reset_flow_resource_lock;
1127 	struct list_head	qp_list;
1128 	struct list_head data_direct_mr_list;
1129 	/* Array with num_ports elements */
1130 	struct mlx5_ib_port	*port;
1131 	struct mlx5_sq_bfreg	bfreg;
1132 	struct mlx5_sq_bfreg	fp_bfreg;
1133 	struct mlx5_ib_delay_drop	delay_drop;
1134 	const struct mlx5_ib_profile	*profile;
1135 
1136 	struct mlx5_ib_lb_state		lb;
1137 	u8			umr_fence;
1138 	struct list_head	ib_dev_list;
1139 	u64			sys_image_guid;
1140 	struct mlx5_dm		dm;
1141 	u16			devx_whitelist_uid;
1142 	struct mlx5_srq_table   srq_table;
1143 	struct mlx5_qp_table    qp_table;
1144 	struct mlx5_async_ctx   async_ctx;
1145 	struct mlx5_devx_event_table devx_event_table;
1146 	struct mlx5_var_table var_table;
1147 
1148 	struct xarray sig_mrs;
1149 	struct mlx5_port_caps port_caps[MLX5_MAX_PORTS];
1150 	u16 pkey_table_len;
1151 	u8 lag_ports;
1152 	struct mlx5_special_mkeys mkeys;
1153 	struct mlx5_data_direct_resources ddr;
1154 
1155 #ifdef CONFIG_MLX5_MACSEC
1156 	struct mlx5_macsec macsec;
1157 #endif
1158 
1159 	u8 num_plane;
1160 	struct mlx5_ib_dev *smi_dev;
1161 	const char *sub_dev_name;
1162 };
1163 
1164 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
1165 {
1166 	return container_of(mcq, struct mlx5_ib_cq, mcq);
1167 }
1168 
1169 static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
1170 {
1171 	return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd);
1172 }
1173 
1174 static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev)
1175 {
1176 	return container_of(ibdev, struct mlx5_ib_dev, ib_dev);
1177 }
1178 
1179 static inline struct mlx5_ib_dev *mr_to_mdev(struct mlx5_ib_mr *mr)
1180 {
1181 	return to_mdev(mr->ibmr.device);
1182 }
1183 
1184 static inline struct mlx5_ib_dev *mlx5_udata_to_mdev(struct ib_udata *udata)
1185 {
1186 	struct mlx5_ib_ucontext *context = rdma_udata_to_drv_context(
1187 		udata, struct mlx5_ib_ucontext, ibucontext);
1188 
1189 	return to_mdev(context->ibucontext.device);
1190 }
1191 
1192 static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq)
1193 {
1194 	return container_of(ibcq, struct mlx5_ib_cq, ibcq);
1195 }
1196 
1197 static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp)
1198 {
1199 	return container_of(mqp, struct mlx5_ib_qp_base, mqp)->container_mibqp;
1200 }
1201 
1202 static inline struct mlx5_ib_rwq *to_mibrwq(struct mlx5_core_qp *core_qp)
1203 {
1204 	return container_of(core_qp, struct mlx5_ib_rwq, core_qp);
1205 }
1206 
1207 static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd)
1208 {
1209 	return container_of(ibpd, struct mlx5_ib_pd, ibpd);
1210 }
1211 
1212 static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq)
1213 {
1214 	return container_of(ibsrq, struct mlx5_ib_srq, ibsrq);
1215 }
1216 
1217 static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp)
1218 {
1219 	return container_of(ibqp, struct mlx5_ib_qp, ibqp);
1220 }
1221 
1222 static inline struct mlx5_ib_rwq *to_mrwq(struct ib_wq *ibwq)
1223 {
1224 	return container_of(ibwq, struct mlx5_ib_rwq, ibwq);
1225 }
1226 
1227 static inline struct mlx5_ib_rwq_ind_table *to_mrwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl)
1228 {
1229 	return container_of(ib_rwq_ind_tbl, struct mlx5_ib_rwq_ind_table, ib_rwq_ind_tbl);
1230 }
1231 
1232 static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
1233 {
1234 	return container_of(msrq, struct mlx5_ib_srq, msrq);
1235 }
1236 
1237 static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
1238 {
1239 	return container_of(ibmr, struct mlx5_ib_mr, ibmr);
1240 }
1241 
1242 static inline struct mlx5_ib_mw *to_mmw(struct ib_mw *ibmw)
1243 {
1244 	return container_of(ibmw, struct mlx5_ib_mw, ibmw);
1245 }
1246 
1247 static inline struct mlx5_ib_flow_action *
1248 to_mflow_act(struct ib_flow_action *ibact)
1249 {
1250 	return container_of(ibact, struct mlx5_ib_flow_action, ib_action);
1251 }
1252 
1253 static inline struct mlx5_user_mmap_entry *
1254 to_mmmap(struct rdma_user_mmap_entry *rdma_entry)
1255 {
1256 	return container_of(rdma_entry,
1257 		struct mlx5_user_mmap_entry, rdma_entry);
1258 }
1259 
1260 int mlx5_ib_dev_res_cq_init(struct mlx5_ib_dev *dev);
1261 int mlx5_ib_dev_res_srq_init(struct mlx5_ib_dev *dev);
1262 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
1263 			struct mlx5_db *db);
1264 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);
1265 void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1266 void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1267 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index);
1268 int mlx5_ib_create_ah(struct ib_ah *ah, struct rdma_ah_init_attr *init_attr,
1269 		      struct ib_udata *udata);
1270 int mlx5_ib_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr);
1271 static inline int mlx5_ib_destroy_ah(struct ib_ah *ah, u32 flags)
1272 {
1273 	return 0;
1274 }
1275 int mlx5_ib_create_srq(struct ib_srq *srq, struct ib_srq_init_attr *init_attr,
1276 		       struct ib_udata *udata);
1277 int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
1278 		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
1279 int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr);
1280 int mlx5_ib_destroy_srq(struct ib_srq *srq, struct ib_udata *udata);
1281 int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
1282 			  const struct ib_recv_wr **bad_wr);
1283 int mlx5_ib_enable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1284 void mlx5_ib_disable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1285 int mlx5_ib_create_qp(struct ib_qp *qp, struct ib_qp_init_attr *init_attr,
1286 		      struct ib_udata *udata);
1287 int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1288 		      int attr_mask, struct ib_udata *udata);
1289 int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
1290 		     struct ib_qp_init_attr *qp_init_attr);
1291 int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata);
1292 void mlx5_ib_drain_sq(struct ib_qp *qp);
1293 void mlx5_ib_drain_rq(struct ib_qp *qp);
1294 int mlx5_ib_read_wqe_sq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1295 			size_t buflen, size_t *bc);
1296 int mlx5_ib_read_wqe_rq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1297 			size_t buflen, size_t *bc);
1298 int mlx5_ib_read_wqe_srq(struct mlx5_ib_srq *srq, int wqe_index, void *buffer,
1299 			 size_t buflen, size_t *bc);
1300 int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
1301 		      struct uverbs_attr_bundle *attrs);
1302 int mlx5_ib_create_user_cq(struct ib_cq *ibcq,
1303 			   const struct ib_cq_init_attr *attr,
1304 			   struct uverbs_attr_bundle *attrs);
1305 int mlx5_ib_destroy_cq(struct ib_cq *cq, struct ib_udata *udata);
1306 int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
1307 int mlx5_ib_pre_destroy_cq(struct ib_cq *cq);
1308 void mlx5_ib_post_destroy_cq(struct ib_cq *cq);
1309 int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
1310 int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1311 int mlx5_ib_resize_cq(struct ib_cq *ibcq, unsigned int entries,
1312 		      struct ib_udata *udata);
1313 struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc);
1314 struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
1315 				  u64 virt_addr, int access_flags,
1316 				  struct ib_dmah *dmah,
1317 				  struct ib_udata *udata);
1318 struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start,
1319 					 u64 length, u64 virt_addr,
1320 					 int fd, int access_flags,
1321 					 struct ib_dmah *dmah,
1322 					 struct uverbs_attr_bundle *attrs);
1323 int mlx5_ib_advise_mr(struct ib_pd *pd,
1324 		      enum ib_uverbs_advise_mr_advice advice,
1325 		      u32 flags,
1326 		      struct ib_sge *sg_list,
1327 		      u32 num_sge,
1328 		      struct uverbs_attr_bundle *attrs);
1329 int mlx5_ib_alloc_mw(struct ib_mw *mw, struct ib_udata *udata);
1330 int mlx5_ib_dealloc_mw(struct ib_mw *mw);
1331 struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd,
1332 					     int access_flags);
1333 void mlx5_ib_free_odp_mr(struct mlx5_ib_mr *mr);
1334 struct ib_mr *mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start,
1335 				    u64 length, u64 virt_addr, int access_flags,
1336 				    struct ib_pd *pd, struct ib_udata *udata);
1337 int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata);
1338 struct ib_mr *mlx5_ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
1339 			       u32 max_num_sg);
1340 struct ib_mr *mlx5_ib_alloc_mr_integrity(struct ib_pd *pd,
1341 					 u32 max_num_sg,
1342 					 u32 max_num_meta_sg);
1343 int mlx5_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents,
1344 		      unsigned int *sg_offset);
1345 int mlx5_ib_map_mr_sg_pi(struct ib_mr *ibmr, struct scatterlist *data_sg,
1346 			 int data_sg_nents, unsigned int *data_sg_offset,
1347 			 struct scatterlist *meta_sg, int meta_sg_nents,
1348 			 unsigned int *meta_sg_offset);
1349 int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u32 port_num,
1350 			const struct ib_wc *in_wc, const struct ib_grh *in_grh,
1351 			const struct ib_mad *in, struct ib_mad *out,
1352 			size_t *out_mad_size, u16 *out_mad_pkey_index);
1353 int mlx5_ib_alloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1354 int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1355 int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, unsigned int port);
1356 int mlx5_query_mad_ifc_system_image_guid(struct ib_device *ibdev,
1357 					 __be64 *sys_image_guid);
1358 int mlx5_query_mad_ifc_max_pkeys(struct ib_device *ibdev,
1359 				 u16 *max_pkeys);
1360 int mlx5_query_mad_ifc_vendor_id(struct ib_device *ibdev,
1361 				 u32 *vendor_id);
1362 int mlx5_query_mad_ifc_node_desc(struct mlx5_ib_dev *dev, char *node_desc);
1363 int mlx5_query_mad_ifc_node_guid(struct mlx5_ib_dev *dev, __be64 *node_guid);
1364 int mlx5_query_mad_ifc_pkey(struct ib_device *ibdev, u32 port, u16 index,
1365 			    u16 *pkey);
1366 int mlx5_query_mad_ifc_gids(struct ib_device *ibdev, u32 port, int index,
1367 			    union ib_gid *gid);
1368 int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u32 port,
1369 			    struct ib_port_attr *props);
1370 int mlx5_ib_query_port(struct ib_device *ibdev, u32 port,
1371 		       struct ib_port_attr *props);
1372 int mlx5_ib_query_port_speed(struct ib_device *ibdev, u32 port_num,
1373 			      u64 *speed);
1374 void mlx5_ib_populate_pas(struct ib_umem *umem, size_t page_size, __be64 *pas,
1375 			  u64 access_flags);
1376 int mlx5_ib_get_cqe_size(struct ib_cq *ibcq);
1377 int mlx5r_frmr_pools_init(struct ib_device *device);
1378 void mlx5r_frmr_pools_cleanup(struct ib_device *device);
1379 struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev,
1380 				       int access_flags, int access_mode,
1381 				       int ndescs);
1382 
1383 int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
1384 			    struct ib_mr_status *mr_status);
1385 struct ib_wq *mlx5_ib_create_wq(struct ib_pd *pd,
1386 				struct ib_wq_init_attr *init_attr,
1387 				struct ib_udata *udata);
1388 int mlx5_ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
1389 int mlx5_ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1390 		      u32 wq_attr_mask, struct ib_udata *udata);
1391 int mlx5_ib_create_rwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_table,
1392 				 struct ib_rwq_ind_table_init_attr *init_attr,
1393 				 struct ib_udata *udata);
1394 int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
1395 struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm,
1396 				struct ib_dm_mr_attr *attr,
1397 				struct uverbs_attr_bundle *attrs);
1398 void mlx5_ib_data_direct_bind(struct mlx5_ib_dev *ibdev,
1399 			      struct mlx5_data_direct_dev *dev);
1400 void mlx5_ib_data_direct_unbind(struct mlx5_ib_dev *ibdev);
1401 void mlx5_ib_revoke_data_direct_mrs(struct mlx5_ib_dev *dev);
1402 
1403 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1404 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);
1405 int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq);
1406 void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev);
1407 int __init mlx5_ib_odp_init(void);
1408 void mlx5_ib_odp_cleanup(void);
1409 int mlx5_odp_init_mkey_cache(struct mlx5_ib_dev *dev);
1410 int mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1411 			  struct mlx5_ib_mr *mr, int flags);
1412 
1413 int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1414 			       enum ib_uverbs_advise_mr_advice advice,
1415 			       u32 flags, struct ib_sge *sg_list, u32 num_sge);
1416 int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr);
1417 int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr);
1418 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1419 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; }
1420 static inline int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev,
1421 				      struct mlx5_ib_pf_eq *eq)
1422 {
1423 	return 0;
1424 }
1425 static inline void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev) {}
1426 static inline int mlx5_ib_odp_init(void) { return 0; }
1427 static inline void mlx5_ib_odp_cleanup(void)				    {}
1428 static inline int mlx5_odp_init_mkey_cache(struct mlx5_ib_dev *dev)
1429 {
1430 	return 0;
1431 }
1432 static inline int mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1433 					struct mlx5_ib_mr *mr, int flags)
1434 {
1435 	return -EOPNOTSUPP;
1436 }
1437 
1438 static inline int
1439 mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1440 			   enum ib_uverbs_advise_mr_advice advice, u32 flags,
1441 			   struct ib_sge *sg_list, u32 num_sge)
1442 {
1443 	return -EOPNOTSUPP;
1444 }
1445 static inline int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr)
1446 {
1447 	return -EOPNOTSUPP;
1448 }
1449 static inline int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr)
1450 {
1451 	return -EOPNOTSUPP;
1452 }
1453 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1454 
1455 extern const struct mmu_interval_notifier_ops mlx5_mn_ops;
1456 
1457 /* Needed for rep profile */
1458 void __mlx5_ib_remove(struct mlx5_ib_dev *dev,
1459 		      const struct mlx5_ib_profile *profile,
1460 		      int stage);
1461 int __mlx5_ib_add(struct mlx5_ib_dev *dev,
1462 		  const struct mlx5_ib_profile *profile);
1463 
1464 int mlx5_ib_get_vf_config(struct ib_device *device, int vf,
1465 			  u32 port, struct ifla_vf_info *info);
1466 int mlx5_ib_set_vf_link_state(struct ib_device *device, int vf,
1467 			      u32 port, int state);
1468 int mlx5_ib_get_vf_stats(struct ib_device *device, int vf,
1469 			 u32 port, struct ifla_vf_stats *stats);
1470 int mlx5_ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
1471 			struct ifla_vf_guid *node_guid,
1472 			struct ifla_vf_guid *port_guid);
1473 int mlx5_ib_set_vf_guid(struct ib_device *device, int vf, u32 port,
1474 			u64 guid, int type);
1475 
1476 __be16 mlx5_get_roce_udp_sport_min(const struct mlx5_ib_dev *dev,
1477 				   const struct ib_gid_attr *attr);
1478 
1479 void mlx5_ib_cleanup_cong_debugfs(struct mlx5_ib_dev *dev, u32 port_num);
1480 void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u32 port_num);
1481 
1482 /* GSI QP helper functions */
1483 int mlx5_ib_create_gsi(struct ib_pd *pd, struct mlx5_ib_qp *mqp,
1484 		       struct ib_qp_init_attr *attr);
1485 int mlx5_ib_destroy_gsi(struct mlx5_ib_qp *mqp);
1486 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
1487 			  int attr_mask);
1488 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
1489 			 int qp_attr_mask,
1490 			 struct ib_qp_init_attr *qp_init_attr);
1491 int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
1492 			  const struct ib_send_wr **bad_wr);
1493 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, const struct ib_recv_wr *wr,
1494 			  const struct ib_recv_wr **bad_wr);
1495 void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi);
1496 
1497 int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc);
1498 
1499 void mlx5_ib_free_bfreg(struct mlx5_ib_dev *dev, struct mlx5_bfreg_info *bfregi,
1500 			int bfregn);
1501 struct mlx5_ib_dev *mlx5_ib_get_ibdev_from_mpi(struct mlx5_ib_multiport_info *mpi);
1502 struct mlx5_core_dev *mlx5_ib_get_native_port_mdev(struct mlx5_ib_dev *dev,
1503 						   u32 ib_port_num,
1504 						   u32 *native_port_num);
1505 void mlx5_ib_put_native_port_mdev(struct mlx5_ib_dev *dev,
1506 				  u32 port_num);
1507 
1508 extern const struct uapi_definition mlx5_ib_devx_defs[];
1509 extern const struct uapi_definition mlx5_ib_flow_defs[];
1510 extern const struct uapi_definition mlx5_ib_qos_defs[];
1511 extern const struct uapi_definition mlx5_ib_std_types_defs[];
1512 extern const struct uapi_definition mlx5_ib_create_cq_defs[];
1513 
1514 static inline int is_qp1(enum ib_qp_type qp_type)
1515 {
1516 	return qp_type == MLX5_IB_QPT_HW_GSI || qp_type == IB_QPT_GSI;
1517 }
1518 
1519 static inline u32 check_cq_create_flags(u32 flags)
1520 {
1521 	/*
1522 	 * It returns non-zero value for unsupported CQ
1523 	 * create flags, otherwise it returns zero.
1524 	 */
1525 	return (flags & ~(IB_UVERBS_CQ_FLAGS_IGNORE_OVERRUN |
1526 			  IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION));
1527 }
1528 
1529 static inline int verify_assign_uidx(u8 cqe_version, u32 cmd_uidx,
1530 				     u32 *user_index)
1531 {
1532 	if (cqe_version) {
1533 		if ((cmd_uidx == MLX5_IB_DEFAULT_UIDX) ||
1534 		    (cmd_uidx & ~MLX5_USER_ASSIGNED_UIDX_MASK))
1535 			return -EINVAL;
1536 		*user_index = cmd_uidx;
1537 	} else {
1538 		*user_index = MLX5_IB_DEFAULT_UIDX;
1539 	}
1540 
1541 	return 0;
1542 }
1543 
1544 static inline int get_qp_user_index(struct mlx5_ib_ucontext *ucontext,
1545 				    struct mlx5_ib_create_qp *ucmd,
1546 				    int inlen,
1547 				    u32 *user_index)
1548 {
1549 	u8 cqe_version = ucontext->cqe_version;
1550 
1551 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1552 	    (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1553 		return 0;
1554 
1555 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1556 		return -EINVAL;
1557 
1558 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1559 }
1560 
1561 static inline int get_srq_user_index(struct mlx5_ib_ucontext *ucontext,
1562 				     struct mlx5_ib_create_srq *ucmd,
1563 				     int inlen,
1564 				     u32 *user_index)
1565 {
1566 	u8 cqe_version = ucontext->cqe_version;
1567 
1568 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1569 	    (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1570 		return 0;
1571 
1572 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1573 		return -EINVAL;
1574 
1575 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1576 }
1577 
1578 static inline int get_uars_per_sys_page(struct mlx5_ib_dev *dev, bool lib_support)
1579 {
1580 	return lib_support && MLX5_CAP_GEN(dev->mdev, uar_4k) ?
1581 				MLX5_UARS_IN_PAGE : 1;
1582 }
1583 
1584 extern void *xlt_emergency_page;
1585 
1586 int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
1587 			struct mlx5_bfreg_info *bfregi, u32 bfregn,
1588 			bool dyn_bfreg);
1589 
1590 static inline int mlx5r_store_odp_mkey(struct mlx5_ib_dev *dev,
1591 				       struct mlx5_ib_mkey *mmkey)
1592 {
1593 	refcount_set(&mmkey->usecount, 1);
1594 
1595 	return xa_err(xa_store(&dev->odp_mkeys, mlx5_base_mkey(mmkey->key),
1596 			       mmkey, GFP_KERNEL));
1597 }
1598 
1599 /* deref an mkey that can participate in ODP flow */
1600 static inline void mlx5r_deref_odp_mkey(struct mlx5_ib_mkey *mmkey)
1601 {
1602 	if (refcount_dec_and_test(&mmkey->usecount))
1603 		wake_up(&mmkey->wait);
1604 }
1605 
1606 /* deref an mkey that can participate in ODP flow and wait for relese */
1607 static inline void mlx5r_deref_wait_odp_mkey(struct mlx5_ib_mkey *mmkey)
1608 {
1609 	mlx5r_deref_odp_mkey(mmkey);
1610 	wait_event(mmkey->wait, refcount_read(&mmkey->usecount) == 0);
1611 }
1612 
1613 static inline bool mlx5_ib_lag_should_assign_affinity(struct mlx5_ib_dev *dev)
1614 {
1615 	/*
1616 	 * If the driver is in hash mode and the port_select_flow_table_bypass cap
1617 	 * is supported, it means that the driver no longer needs to assign the port
1618 	 * affinity by default. If a user wants to set the port affinity explicitly,
1619 	 * the user has a dedicated API to do that, so there is no need to assign
1620 	 * the port affinity by default.
1621 	 */
1622 	if (dev->lag_active &&
1623 	    mlx5_lag_mode_is_hash(dev->mdev) &&
1624 	    MLX5_CAP_PORT_SELECTION(dev->mdev, port_select_flow_table_bypass))
1625 		return 0;
1626 
1627 	if (mlx5_lag_is_lacp_owner(dev->mdev) && !dev->lag_active)
1628 		return 0;
1629 
1630 	return dev->lag_active ||
1631 		(MLX5_CAP_GEN(dev->mdev, num_lag_ports) > 1 &&
1632 		 MLX5_CAP_GEN(dev->mdev, lag_tx_port_affinity));
1633 }
1634 
1635 static inline bool rt_supported(int ts_cap)
1636 {
1637 	return ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_REAL_TIME ||
1638 	       ts_cap == MLX5_TIMESTAMP_FORMAT_CAP_FREE_RUNNING_AND_REAL_TIME;
1639 }
1640 
1641 /*
1642  * PCI Peer to Peer is a trainwreck. If no switch is present then things
1643  * sometimes work, depending on the pci_distance_p2p logic for excluding broken
1644  * root complexes. However if a switch is present in the path, then things get
1645  * really ugly depending on how the switch is setup. This table assumes that the
1646  * root complex is strict and is validating that all req/reps are matches
1647  * perfectly - so any scenario where it sees only half the transaction is a
1648  * failure.
1649  *
1650  * CR/RR/DT  ATS RO P2P
1651  * 00X       X   X  OK
1652  * 010       X   X  fails (request is routed to root but root never sees comp)
1653  * 011       0   X  fails (request is routed to root but root never sees comp)
1654  * 011       1   X  OK
1655  * 10X       X   1  OK
1656  * 101       X   0  fails (completion is routed to root but root didn't see req)
1657  * 110       X   0  SLOW
1658  * 111       0   0  SLOW
1659  * 111       1   0  fails (completion is routed to root but root didn't see req)
1660  * 111       1   1  OK
1661  *
1662  * Unfortunately we cannot reliably know if a switch is present or what the
1663  * CR/RR/DT ACS settings are, as in a VM that is all hidden. Assume that
1664  * CR/RR/DT is 111 if the ATS cap is enabled and follow the last three rows.
1665  *
1666  * For now assume if the umem is a dma_buf then it is P2P.
1667  */
1668 static inline bool mlx5_umem_needs_ats(struct mlx5_ib_dev *dev,
1669 				       struct ib_umem *umem, int access_flags)
1670 {
1671 	if (!MLX5_CAP_GEN(dev->mdev, ats) || !umem->is_dmabuf)
1672 		return false;
1673 	return access_flags & IB_ACCESS_RELAXED_ORDERING;
1674 }
1675 
1676 int set_roce_addr(struct mlx5_ib_dev *dev, u32 port_num,
1677 		  unsigned int index, const union ib_gid *gid,
1678 		  const struct ib_gid_attr *attr);
1679 
1680 static inline u32 smi_to_native_portnum(struct mlx5_ib_dev *dev, u32 port)
1681 {
1682 	return (port - 1) / dev->num_ports + 1;
1683 }
1684 
1685 static inline unsigned int get_max_log_entity_size_cap(struct mlx5_ib_dev *dev,
1686 						       int access_mode)
1687 {
1688 	int max_log_size = 0;
1689 
1690 	if (access_mode == MLX5_MKC_ACCESS_MODE_MTT)
1691 		max_log_size =
1692 			MLX5_CAP_GEN_2(dev->mdev, max_mkey_log_entity_size_mtt);
1693 	else if (access_mode == MLX5_MKC_ACCESS_MODE_KSM)
1694 		max_log_size = MLX5_CAP_GEN_2(
1695 			dev->mdev, max_mkey_log_entity_size_fixed_buffer);
1696 
1697 	if (!max_log_size ||
1698 	    (max_log_size > 31 &&
1699 	     !MLX5_CAP_GEN_2(dev->mdev, umr_log_entity_size_5)))
1700 		max_log_size = 31;
1701 
1702 	return max_log_size;
1703 }
1704 
1705 static inline unsigned int get_min_log_entity_size_cap(struct mlx5_ib_dev *dev,
1706 						       int access_mode)
1707 {
1708 	int min_log_size = 0;
1709 
1710 	if (access_mode == MLX5_MKC_ACCESS_MODE_KSM &&
1711 	    MLX5_CAP_GEN_2(dev->mdev,
1712 			   min_mkey_log_entity_size_fixed_buffer_valid))
1713 		min_log_size = MLX5_CAP_GEN_2(
1714 			dev->mdev, min_mkey_log_entity_size_fixed_buffer);
1715 	else
1716 		min_log_size =
1717 			MLX5_CAP_GEN_2(dev->mdev, log_min_mkey_entity_size);
1718 
1719 	min_log_size = max(min_log_size, MLX5_ADAPTER_PAGE_SHIFT);
1720 	return min_log_size;
1721 }
1722 
1723 /*
1724  * For mkc users, instead of a page_offset the command has a start_iova which
1725  * specifies both the page_offset and the on-the-wire IOVA
1726  */
1727 static __always_inline unsigned long
1728 mlx5_umem_mkc_find_best_pgsz(struct mlx5_ib_dev *dev, struct ib_umem *umem,
1729 			     u64 iova, int access_mode)
1730 {
1731 	unsigned int max_log_entity_size_cap, min_log_entity_size_cap;
1732 	unsigned long bitmap;
1733 
1734 	max_log_entity_size_cap = get_max_log_entity_size_cap(dev, access_mode);
1735 	min_log_entity_size_cap = get_min_log_entity_size_cap(dev, access_mode);
1736 
1737 	bitmap = GENMASK_ULL(max_log_entity_size_cap, min_log_entity_size_cap);
1738 
1739 	/* In KSM mode HW requires IOVA and mkey's page size to be aligned */
1740 	if (access_mode == MLX5_MKC_ACCESS_MODE_KSM && iova)
1741 		bitmap &= GENMASK_ULL(__ffs64(iova), 0);
1742 
1743 	return ib_umem_find_best_pgsz(umem, bitmap, iova);
1744 }
1745 
1746 static inline unsigned long
1747 mlx5_umem_dmabuf_find_best_pgsz(struct ib_umem_dmabuf *umem_dmabuf,
1748 				int access_mode)
1749 {
1750 	return mlx5_umem_mkc_find_best_pgsz(to_mdev(umem_dmabuf->umem.ibdev),
1751 					    &umem_dmabuf->umem,
1752 					    umem_dmabuf->umem.iova,
1753 					    access_mode);
1754 }
1755 
1756 #endif /* MLX5_IB_H */
1757