xref: /linux/drivers/infiniband/hw/mlx5/mlx5_ib.h (revision 4fd18fc38757217c746aa063ba9e4729814dc737)
1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
2 /*
3  * Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved.
4  */
5 
6 #ifndef MLX5_IB_H
7 #define MLX5_IB_H
8 
9 #include <linux/kernel.h>
10 #include <linux/sched.h>
11 #include <rdma/ib_verbs.h>
12 #include <rdma/ib_umem.h>
13 #include <rdma/ib_smi.h>
14 #include <linux/mlx5/driver.h>
15 #include <linux/mlx5/cq.h>
16 #include <linux/mlx5/fs.h>
17 #include <linux/mlx5/qp.h>
18 #include <linux/types.h>
19 #include <linux/mlx5/transobj.h>
20 #include <rdma/ib_user_verbs.h>
21 #include <rdma/mlx5-abi.h>
22 #include <rdma/uverbs_ioctl.h>
23 #include <rdma/mlx5_user_ioctl_cmds.h>
24 #include <rdma/mlx5_user_ioctl_verbs.h>
25 
26 #include "srq.h"
27 
28 #define mlx5_ib_dbg(_dev, format, arg...)                                      \
29 	dev_dbg(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,      \
30 		__LINE__, current->pid, ##arg)
31 
32 #define mlx5_ib_err(_dev, format, arg...)                                      \
33 	dev_err(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,      \
34 		__LINE__, current->pid, ##arg)
35 
36 #define mlx5_ib_warn(_dev, format, arg...)                                     \
37 	dev_warn(&(_dev)->ib_dev.dev, "%s:%d:(pid %d): " format, __func__,     \
38 		 __LINE__, current->pid, ##arg)
39 
40 #define MLX5_IB_DEFAULT_UIDX 0xffffff
41 #define MLX5_USER_ASSIGNED_UIDX_MASK __mlx5_mask(qpc, user_index)
42 
43 static __always_inline unsigned long
44 __mlx5_log_page_size_to_bitmap(unsigned int log_pgsz_bits,
45 			       unsigned int pgsz_shift)
46 {
47 	unsigned int largest_pg_shift =
48 		min_t(unsigned long, (1ULL << log_pgsz_bits) - 1 + pgsz_shift,
49 		      BITS_PER_LONG - 1);
50 
51 	/*
52 	 * Despite a command allowing it, the device does not support lower than
53 	 * 4k page size.
54 	 */
55 	pgsz_shift = max_t(unsigned int, MLX5_ADAPTER_PAGE_SHIFT, pgsz_shift);
56 	return GENMASK(largest_pg_shift, pgsz_shift);
57 }
58 
59 /*
60  * For mkc users, instead of a page_offset the command has a start_iova which
61  * specifies both the page_offset and the on-the-wire IOVA
62  */
63 #define mlx5_umem_find_best_pgsz(umem, typ, log_pgsz_fld, pgsz_shift, iova)    \
64 	ib_umem_find_best_pgsz(umem,                                           \
65 			       __mlx5_log_page_size_to_bitmap(                 \
66 				       __mlx5_bit_sz(typ, log_pgsz_fld),       \
67 				       pgsz_shift),                            \
68 			       iova)
69 
70 static __always_inline unsigned long
71 __mlx5_page_offset_to_bitmask(unsigned int page_offset_bits,
72 			      unsigned int offset_shift)
73 {
74 	unsigned int largest_offset_shift =
75 		min_t(unsigned long, page_offset_bits - 1 + offset_shift,
76 		      BITS_PER_LONG - 1);
77 
78 	return GENMASK(largest_offset_shift, offset_shift);
79 }
80 
81 /*
82  * QP/CQ/WQ/etc type commands take a page offset that satisifies:
83  *   page_offset_quantized * (page_size/scale) = page_offset
84  * Which restricts allowed page sizes to ones that satisify the above.
85  */
86 unsigned long __mlx5_umem_find_best_quantized_pgoff(
87 	struct ib_umem *umem, unsigned long pgsz_bitmap,
88 	unsigned int page_offset_bits, u64 pgoff_bitmask, unsigned int scale,
89 	unsigned int *page_offset_quantized);
90 #define mlx5_umem_find_best_quantized_pgoff(umem, typ, log_pgsz_fld,           \
91 					    pgsz_shift, page_offset_fld,       \
92 					    scale, page_offset_quantized)      \
93 	__mlx5_umem_find_best_quantized_pgoff(                                 \
94 		umem,                                                          \
95 		__mlx5_log_page_size_to_bitmap(                                \
96 			__mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift),         \
97 		__mlx5_bit_sz(typ, page_offset_fld),                           \
98 		GENMASK(31, order_base_2(scale)), scale,                       \
99 		page_offset_quantized)
100 
101 #define mlx5_umem_find_best_cq_quantized_pgoff(umem, typ, log_pgsz_fld,        \
102 					       pgsz_shift, page_offset_fld,    \
103 					       scale, page_offset_quantized)   \
104 	__mlx5_umem_find_best_quantized_pgoff(                                 \
105 		umem,                                                          \
106 		__mlx5_log_page_size_to_bitmap(                                \
107 			__mlx5_bit_sz(typ, log_pgsz_fld), pgsz_shift),         \
108 		__mlx5_bit_sz(typ, page_offset_fld), 0, scale,                 \
109 		page_offset_quantized)
110 
111 enum {
112 	MLX5_IB_MMAP_OFFSET_START = 9,
113 	MLX5_IB_MMAP_OFFSET_END = 255,
114 };
115 
116 enum {
117 	MLX5_IB_MMAP_CMD_SHIFT	= 8,
118 	MLX5_IB_MMAP_CMD_MASK	= 0xff,
119 };
120 
121 enum {
122 	MLX5_RES_SCAT_DATA32_CQE	= 0x1,
123 	MLX5_RES_SCAT_DATA64_CQE	= 0x2,
124 	MLX5_REQ_SCAT_DATA32_CQE	= 0x11,
125 	MLX5_REQ_SCAT_DATA64_CQE	= 0x22,
126 };
127 
128 enum mlx5_ib_mad_ifc_flags {
129 	MLX5_MAD_IFC_IGNORE_MKEY	= 1,
130 	MLX5_MAD_IFC_IGNORE_BKEY	= 2,
131 	MLX5_MAD_IFC_NET_VIEW		= 4,
132 };
133 
134 enum {
135 	MLX5_CROSS_CHANNEL_BFREG         = 0,
136 };
137 
138 enum {
139 	MLX5_CQE_VERSION_V0,
140 	MLX5_CQE_VERSION_V1,
141 };
142 
143 enum {
144 	MLX5_TM_MAX_RNDV_MSG_SIZE	= 64,
145 	MLX5_TM_MAX_SGE			= 1,
146 };
147 
148 enum {
149 	MLX5_IB_INVALID_UAR_INDEX	= BIT(31),
150 	MLX5_IB_INVALID_BFREG		= BIT(31),
151 };
152 
153 enum {
154 	MLX5_MAX_MEMIC_PAGES = 0x100,
155 	MLX5_MEMIC_ALLOC_SIZE_MASK = 0x3f,
156 };
157 
158 enum {
159 	MLX5_MEMIC_BASE_ALIGN	= 6,
160 	MLX5_MEMIC_BASE_SIZE	= 1 << MLX5_MEMIC_BASE_ALIGN,
161 };
162 
163 enum mlx5_ib_mmap_type {
164 	MLX5_IB_MMAP_TYPE_MEMIC = 1,
165 	MLX5_IB_MMAP_TYPE_VAR = 2,
166 	MLX5_IB_MMAP_TYPE_UAR_WC = 3,
167 	MLX5_IB_MMAP_TYPE_UAR_NC = 4,
168 };
169 
170 struct mlx5_bfreg_info {
171 	u32 *sys_pages;
172 	int num_low_latency_bfregs;
173 	unsigned int *count;
174 
175 	/*
176 	 * protect bfreg allocation data structs
177 	 */
178 	struct mutex lock;
179 	u32 ver;
180 	u8 lib_uar_4k : 1;
181 	u8 lib_uar_dyn : 1;
182 	u32 num_sys_pages;
183 	u32 num_static_sys_pages;
184 	u32 total_num_bfregs;
185 	u32 num_dyn_bfregs;
186 };
187 
188 struct mlx5_ib_ucontext {
189 	struct ib_ucontext	ibucontext;
190 	struct list_head	db_page_list;
191 
192 	/* protect doorbell record alloc/free
193 	 */
194 	struct mutex		db_page_mutex;
195 	struct mlx5_bfreg_info	bfregi;
196 	u8			cqe_version;
197 	/* Transport Domain number */
198 	u32			tdn;
199 
200 	u64			lib_caps;
201 	u16			devx_uid;
202 	/* For RoCE LAG TX affinity */
203 	atomic_t		tx_port_affinity;
204 };
205 
206 static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
207 {
208 	return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext);
209 }
210 
211 struct mlx5_ib_pd {
212 	struct ib_pd		ibpd;
213 	u32			pdn;
214 	u16			uid;
215 };
216 
217 enum {
218 	MLX5_IB_FLOW_ACTION_MODIFY_HEADER,
219 	MLX5_IB_FLOW_ACTION_PACKET_REFORMAT,
220 	MLX5_IB_FLOW_ACTION_DECAP,
221 };
222 
223 #define MLX5_IB_FLOW_MCAST_PRIO		(MLX5_BY_PASS_NUM_PRIOS - 1)
224 #define MLX5_IB_FLOW_LAST_PRIO		(MLX5_BY_PASS_NUM_REGULAR_PRIOS - 1)
225 #if (MLX5_IB_FLOW_LAST_PRIO <= 0)
226 #error "Invalid number of bypass priorities"
227 #endif
228 #define MLX5_IB_FLOW_LEFTOVERS_PRIO	(MLX5_IB_FLOW_MCAST_PRIO + 1)
229 
230 #define MLX5_IB_NUM_FLOW_FT		(MLX5_IB_FLOW_LEFTOVERS_PRIO + 1)
231 #define MLX5_IB_NUM_SNIFFER_FTS		2
232 #define MLX5_IB_NUM_EGRESS_FTS		1
233 struct mlx5_ib_flow_prio {
234 	struct mlx5_flow_table		*flow_table;
235 	unsigned int			refcount;
236 };
237 
238 struct mlx5_ib_flow_handler {
239 	struct list_head		list;
240 	struct ib_flow			ibflow;
241 	struct mlx5_ib_flow_prio	*prio;
242 	struct mlx5_flow_handle		*rule;
243 	struct ib_counters		*ibcounters;
244 	struct mlx5_ib_dev		*dev;
245 	struct mlx5_ib_flow_matcher	*flow_matcher;
246 };
247 
248 struct mlx5_ib_flow_matcher {
249 	struct mlx5_ib_match_params matcher_mask;
250 	int			mask_len;
251 	enum mlx5_ib_flow_type	flow_type;
252 	enum mlx5_flow_namespace_type ns_type;
253 	u16			priority;
254 	struct mlx5_core_dev	*mdev;
255 	atomic_t		usecnt;
256 	u8			match_criteria_enable;
257 };
258 
259 struct mlx5_ib_pp {
260 	u16 index;
261 	struct mlx5_core_dev *mdev;
262 };
263 
264 struct mlx5_ib_flow_db {
265 	struct mlx5_ib_flow_prio	prios[MLX5_IB_NUM_FLOW_FT];
266 	struct mlx5_ib_flow_prio	egress_prios[MLX5_IB_NUM_FLOW_FT];
267 	struct mlx5_ib_flow_prio	sniffer[MLX5_IB_NUM_SNIFFER_FTS];
268 	struct mlx5_ib_flow_prio	egress[MLX5_IB_NUM_EGRESS_FTS];
269 	struct mlx5_ib_flow_prio	fdb;
270 	struct mlx5_ib_flow_prio	rdma_rx[MLX5_IB_NUM_FLOW_FT];
271 	struct mlx5_ib_flow_prio	rdma_tx[MLX5_IB_NUM_FLOW_FT];
272 	struct mlx5_flow_table		*lag_demux_ft;
273 	/* Protect flow steering bypass flow tables
274 	 * when add/del flow rules.
275 	 * only single add/removal of flow steering rule could be done
276 	 * simultaneously.
277 	 */
278 	struct mutex			lock;
279 };
280 
281 /* Use macros here so that don't have to duplicate
282  * enum ib_send_flags and enum ib_qp_type for low-level driver
283  */
284 
285 #define MLX5_IB_SEND_UMR_ENABLE_MR	       (IB_SEND_RESERVED_START << 0)
286 #define MLX5_IB_SEND_UMR_DISABLE_MR	       (IB_SEND_RESERVED_START << 1)
287 #define MLX5_IB_SEND_UMR_FAIL_IF_FREE	       (IB_SEND_RESERVED_START << 2)
288 #define MLX5_IB_SEND_UMR_UPDATE_XLT	       (IB_SEND_RESERVED_START << 3)
289 #define MLX5_IB_SEND_UMR_UPDATE_TRANSLATION    (IB_SEND_RESERVED_START << 4)
290 #define MLX5_IB_SEND_UMR_UPDATE_PD_ACCESS       IB_SEND_RESERVED_END
291 
292 #define MLX5_IB_QPT_REG_UMR	IB_QPT_RESERVED1
293 /*
294  * IB_QPT_GSI creates the software wrapper around GSI, and MLX5_IB_QPT_HW_GSI
295  * creates the actual hardware QP.
296  */
297 #define MLX5_IB_QPT_HW_GSI	IB_QPT_RESERVED2
298 #define MLX5_IB_QPT_DCI		IB_QPT_RESERVED3
299 #define MLX5_IB_QPT_DCT		IB_QPT_RESERVED4
300 #define MLX5_IB_WR_UMR		IB_WR_RESERVED1
301 
302 #define MLX5_IB_UMR_OCTOWORD	       16
303 #define MLX5_IB_UMR_XLT_ALIGNMENT      64
304 
305 #define MLX5_IB_UPD_XLT_ZAP	      BIT(0)
306 #define MLX5_IB_UPD_XLT_ENABLE	      BIT(1)
307 #define MLX5_IB_UPD_XLT_ATOMIC	      BIT(2)
308 #define MLX5_IB_UPD_XLT_ADDR	      BIT(3)
309 #define MLX5_IB_UPD_XLT_PD	      BIT(4)
310 #define MLX5_IB_UPD_XLT_ACCESS	      BIT(5)
311 #define MLX5_IB_UPD_XLT_INDIRECT      BIT(6)
312 
313 /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags.
314  *
315  * These flags are intended for internal use by the mlx5_ib driver, and they
316  * rely on the range reserved for that use in the ib_qp_create_flags enum.
317  */
318 #define MLX5_IB_QP_CREATE_SQPN_QP1	IB_QP_CREATE_RESERVED_START
319 #define MLX5_IB_QP_CREATE_WC_TEST	(IB_QP_CREATE_RESERVED_START << 1)
320 
321 struct wr_list {
322 	u16	opcode;
323 	u16	next;
324 };
325 
326 enum mlx5_ib_rq_flags {
327 	MLX5_IB_RQ_CVLAN_STRIPPING	= 1 << 0,
328 	MLX5_IB_RQ_PCI_WRITE_END_PADDING	= 1 << 1,
329 };
330 
331 struct mlx5_ib_wq {
332 	struct mlx5_frag_buf_ctrl fbc;
333 	u64		       *wrid;
334 	u32		       *wr_data;
335 	struct wr_list	       *w_list;
336 	unsigned	       *wqe_head;
337 	u16		        unsig_count;
338 
339 	/* serialize post to the work queue
340 	 */
341 	spinlock_t		lock;
342 	int			wqe_cnt;
343 	int			max_post;
344 	int			max_gs;
345 	int			offset;
346 	int			wqe_shift;
347 	unsigned		head;
348 	unsigned		tail;
349 	u16			cur_post;
350 	u16			last_poll;
351 	void			*cur_edge;
352 };
353 
354 enum mlx5_ib_wq_flags {
355 	MLX5_IB_WQ_FLAGS_DELAY_DROP = 0x1,
356 	MLX5_IB_WQ_FLAGS_STRIDING_RQ = 0x2,
357 };
358 
359 #define MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES 9
360 #define MLX5_MAX_SINGLE_WQE_LOG_NUM_STRIDES 16
361 #define MLX5_MIN_SINGLE_STRIDE_LOG_NUM_BYTES 6
362 #define MLX5_MAX_SINGLE_STRIDE_LOG_NUM_BYTES 13
363 #define MLX5_EXT_MIN_SINGLE_WQE_LOG_NUM_STRIDES 3
364 
365 struct mlx5_ib_rwq {
366 	struct ib_wq		ibwq;
367 	struct mlx5_core_qp	core_qp;
368 	u32			rq_num_pas;
369 	u32			log_rq_stride;
370 	u32			log_rq_size;
371 	u32			rq_page_offset;
372 	u32			log_page_size;
373 	u32			log_num_strides;
374 	u32			two_byte_shift_en;
375 	u32			single_stride_log_num_of_bytes;
376 	struct ib_umem		*umem;
377 	size_t			buf_size;
378 	unsigned int		page_shift;
379 	struct mlx5_db		db;
380 	u32			user_index;
381 	u32			wqe_count;
382 	u32			wqe_shift;
383 	int			wq_sig;
384 	u32			create_flags; /* Use enum mlx5_ib_wq_flags */
385 };
386 
387 struct mlx5_ib_rwq_ind_table {
388 	struct ib_rwq_ind_table ib_rwq_ind_tbl;
389 	u32			rqtn;
390 	u16			uid;
391 };
392 
393 struct mlx5_ib_ubuffer {
394 	struct ib_umem	       *umem;
395 	int			buf_size;
396 	u64			buf_addr;
397 };
398 
399 struct mlx5_ib_qp_base {
400 	struct mlx5_ib_qp	*container_mibqp;
401 	struct mlx5_core_qp	mqp;
402 	struct mlx5_ib_ubuffer	ubuffer;
403 };
404 
405 struct mlx5_ib_qp_trans {
406 	struct mlx5_ib_qp_base	base;
407 	u16			xrcdn;
408 	u8			alt_port;
409 	u8			atomic_rd_en;
410 	u8			resp_depth;
411 };
412 
413 struct mlx5_ib_rss_qp {
414 	u32	tirn;
415 };
416 
417 struct mlx5_ib_rq {
418 	struct mlx5_ib_qp_base base;
419 	struct mlx5_ib_wq	*rq;
420 	struct mlx5_ib_ubuffer	ubuffer;
421 	struct mlx5_db		*doorbell;
422 	u32			tirn;
423 	u8			state;
424 	u32			flags;
425 };
426 
427 struct mlx5_ib_sq {
428 	struct mlx5_ib_qp_base base;
429 	struct mlx5_ib_wq	*sq;
430 	struct mlx5_ib_ubuffer  ubuffer;
431 	struct mlx5_db		*doorbell;
432 	struct mlx5_flow_handle	*flow_rule;
433 	u32			tisn;
434 	u8			state;
435 };
436 
437 struct mlx5_ib_raw_packet_qp {
438 	struct mlx5_ib_sq sq;
439 	struct mlx5_ib_rq rq;
440 };
441 
442 struct mlx5_bf {
443 	int			buf_size;
444 	unsigned long		offset;
445 	struct mlx5_sq_bfreg   *bfreg;
446 };
447 
448 struct mlx5_ib_dct {
449 	struct mlx5_core_dct    mdct;
450 	u32                     *in;
451 };
452 
453 struct mlx5_ib_gsi_qp {
454 	struct ib_qp *rx_qp;
455 	u8 port_num;
456 	struct ib_qp_cap cap;
457 	struct ib_cq *cq;
458 	struct mlx5_ib_gsi_wr *outstanding_wrs;
459 	u32 outstanding_pi, outstanding_ci;
460 	int num_qps;
461 	/* Protects access to the tx_qps. Post send operations synchronize
462 	 * with tx_qp creation in setup_qp(). Also protects the
463 	 * outstanding_wrs array and indices.
464 	 */
465 	spinlock_t lock;
466 	struct ib_qp **tx_qps;
467 };
468 
469 struct mlx5_ib_qp {
470 	struct ib_qp		ibqp;
471 	union {
472 		struct mlx5_ib_qp_trans trans_qp;
473 		struct mlx5_ib_raw_packet_qp raw_packet_qp;
474 		struct mlx5_ib_rss_qp rss_qp;
475 		struct mlx5_ib_dct dct;
476 		struct mlx5_ib_gsi_qp gsi;
477 	};
478 	struct mlx5_frag_buf	buf;
479 
480 	struct mlx5_db		db;
481 	struct mlx5_ib_wq	rq;
482 
483 	u8			sq_signal_bits;
484 	u8			next_fence;
485 	struct mlx5_ib_wq	sq;
486 
487 	/* serialize qp state modifications
488 	 */
489 	struct mutex		mutex;
490 	/* cached variant of create_flags from struct ib_qp_init_attr */
491 	u32			flags;
492 	u8			port;
493 	u8			state;
494 	int			max_inline_data;
495 	struct mlx5_bf	        bf;
496 	u8			has_rq:1;
497 	u8			is_rss:1;
498 
499 	/* only for user space QPs. For kernel
500 	 * we have it from the bf object
501 	 */
502 	int			bfregn;
503 
504 	struct list_head	qps_list;
505 	struct list_head	cq_recv_list;
506 	struct list_head	cq_send_list;
507 	struct mlx5_rate_limit	rl;
508 	u32                     underlay_qpn;
509 	u32			flags_en;
510 	/*
511 	 * IB/core doesn't store low-level QP types, so
512 	 * store both MLX and IBTA types in the field below.
513 	 * IB_QPT_DRIVER will be break to DCI/DCT subtypes.
514 	 */
515 	enum ib_qp_type		type;
516 	/* A flag to indicate if there's a new counter is configured
517 	 * but not take effective
518 	 */
519 	u32                     counter_pending;
520 	u16			gsi_lag_port;
521 };
522 
523 struct mlx5_ib_cq_buf {
524 	struct mlx5_frag_buf_ctrl fbc;
525 	struct mlx5_frag_buf    frag_buf;
526 	struct ib_umem		*umem;
527 	int			cqe_size;
528 	int			nent;
529 };
530 
531 struct mlx5_umr_wr {
532 	struct ib_send_wr		wr;
533 	u64				virt_addr;
534 	u64				offset;
535 	struct ib_pd		       *pd;
536 	unsigned int			page_shift;
537 	unsigned int			xlt_size;
538 	u64				length;
539 	int				access_flags;
540 	u32				mkey;
541 	u8				ignore_free_state:1;
542 };
543 
544 static inline const struct mlx5_umr_wr *umr_wr(const struct ib_send_wr *wr)
545 {
546 	return container_of(wr, struct mlx5_umr_wr, wr);
547 }
548 
549 struct mlx5_shared_mr_info {
550 	int mr_id;
551 	struct ib_umem		*umem;
552 };
553 
554 enum mlx5_ib_cq_pr_flags {
555 	MLX5_IB_CQ_PR_FLAGS_CQE_128_PAD	= 1 << 0,
556 };
557 
558 struct mlx5_ib_cq {
559 	struct ib_cq		ibcq;
560 	struct mlx5_core_cq	mcq;
561 	struct mlx5_ib_cq_buf	buf;
562 	struct mlx5_db		db;
563 
564 	/* serialize access to the CQ
565 	 */
566 	spinlock_t		lock;
567 
568 	/* protect resize cq
569 	 */
570 	struct mutex		resize_mutex;
571 	struct mlx5_ib_cq_buf  *resize_buf;
572 	struct ib_umem	       *resize_umem;
573 	int			cqe_size;
574 	struct list_head	list_send_qp;
575 	struct list_head	list_recv_qp;
576 	u32			create_flags;
577 	struct list_head	wc_list;
578 	enum ib_cq_notify_flags notify_flags;
579 	struct work_struct	notify_work;
580 	u16			private_flags; /* Use mlx5_ib_cq_pr_flags */
581 };
582 
583 struct mlx5_ib_wc {
584 	struct ib_wc wc;
585 	struct list_head list;
586 };
587 
588 struct mlx5_ib_srq {
589 	struct ib_srq		ibsrq;
590 	struct mlx5_core_srq	msrq;
591 	struct mlx5_frag_buf	buf;
592 	struct mlx5_db		db;
593 	struct mlx5_frag_buf_ctrl fbc;
594 	u64		       *wrid;
595 	/* protect SRQ hanlding
596 	 */
597 	spinlock_t		lock;
598 	int			head;
599 	int			tail;
600 	u16			wqe_ctr;
601 	struct ib_umem	       *umem;
602 	/* serialize arming a SRQ
603 	 */
604 	struct mutex		mutex;
605 	int			wq_sig;
606 };
607 
608 struct mlx5_ib_xrcd {
609 	struct ib_xrcd		ibxrcd;
610 	u32			xrcdn;
611 };
612 
613 enum mlx5_ib_mtt_access_flags {
614 	MLX5_IB_MTT_READ  = (1 << 0),
615 	MLX5_IB_MTT_WRITE = (1 << 1),
616 };
617 
618 struct mlx5_user_mmap_entry {
619 	struct rdma_user_mmap_entry rdma_entry;
620 	u8 mmap_flag;
621 	u64 address;
622 	u32 page_idx;
623 };
624 
625 struct mlx5_ib_dm {
626 	struct ib_dm		ibdm;
627 	phys_addr_t		dev_addr;
628 	u32			type;
629 	size_t			size;
630 	union {
631 		struct {
632 			u32	obj_id;
633 		} icm_dm;
634 		/* other dm types specific params should be added here */
635 	};
636 	struct mlx5_user_mmap_entry mentry;
637 };
638 
639 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
640 
641 #define MLX5_IB_DM_MEMIC_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE   |\
642 					 IB_ACCESS_REMOTE_WRITE  |\
643 					 IB_ACCESS_REMOTE_READ   |\
644 					 IB_ACCESS_REMOTE_ATOMIC |\
645 					 IB_ZERO_BASED)
646 
647 #define MLX5_IB_DM_SW_ICM_ALLOWED_ACCESS (IB_ACCESS_LOCAL_WRITE   |\
648 					  IB_ACCESS_REMOTE_WRITE  |\
649 					  IB_ACCESS_REMOTE_READ   |\
650 					  IB_ZERO_BASED)
651 
652 #define mlx5_update_odp_stats(mr, counter_name, value)		\
653 	atomic64_add(value, &((mr)->odp_stats.counter_name))
654 
655 struct mlx5_ib_mr {
656 	struct ib_mr		ibmr;
657 	void			*descs;
658 	dma_addr_t		desc_map;
659 	int			ndescs;
660 	int			data_length;
661 	int			meta_ndescs;
662 	int			meta_length;
663 	int			max_descs;
664 	int			desc_size;
665 	int			access_mode;
666 	unsigned int		page_shift;
667 	struct mlx5_core_mkey	mmkey;
668 	struct ib_umem	       *umem;
669 	struct mlx5_shared_mr_info	*smr_info;
670 	struct list_head	list;
671 	struct mlx5_cache_ent  *cache_ent;
672 	u32 out[MLX5_ST_SZ_DW(create_mkey_out)];
673 	struct mlx5_core_sig_ctx    *sig;
674 	void			*descs_alloc;
675 	int			access_flags; /* Needed for rereg MR */
676 
677 	struct mlx5_ib_mr      *parent;
678 	/* Needed for IB_MR_TYPE_INTEGRITY */
679 	struct mlx5_ib_mr      *pi_mr;
680 	struct mlx5_ib_mr      *klm_mr;
681 	struct mlx5_ib_mr      *mtt_mr;
682 	u64			data_iova;
683 	u64			pi_iova;
684 
685 	/* For ODP and implicit */
686 	atomic_t		num_deferred_work;
687 	wait_queue_head_t       q_deferred_work;
688 	struct xarray		implicit_children;
689 	union {
690 		struct rcu_head rcu;
691 		struct list_head elm;
692 		struct work_struct work;
693 	} odp_destroy;
694 	struct ib_odp_counters	odp_stats;
695 	bool			is_odp_implicit;
696 
697 	struct mlx5_async_work  cb_work;
698 };
699 
700 static inline bool is_odp_mr(struct mlx5_ib_mr *mr)
701 {
702 	return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
703 	       mr->umem->is_odp;
704 }
705 
706 struct mlx5_ib_mw {
707 	struct ib_mw		ibmw;
708 	struct mlx5_core_mkey	mmkey;
709 	int			ndescs;
710 };
711 
712 struct mlx5_ib_devx_mr {
713 	struct mlx5_core_mkey	mmkey;
714 	int			ndescs;
715 };
716 
717 struct mlx5_ib_umr_context {
718 	struct ib_cqe		cqe;
719 	enum ib_wc_status	status;
720 	struct completion	done;
721 };
722 
723 struct umr_common {
724 	struct ib_pd	*pd;
725 	struct ib_cq	*cq;
726 	struct ib_qp	*qp;
727 	/* control access to UMR QP
728 	 */
729 	struct semaphore	sem;
730 };
731 
732 struct mlx5_cache_ent {
733 	struct list_head	head;
734 	/* sync access to the cahce entry
735 	 */
736 	spinlock_t		lock;
737 
738 
739 	char                    name[4];
740 	u32                     order;
741 	u32			xlt;
742 	u32			access_mode;
743 	u32			page;
744 
745 	u8 disabled:1;
746 	u8 fill_to_high_water:1;
747 
748 	/*
749 	 * - available_mrs is the length of list head, ie the number of MRs
750 	 *   available for immediate allocation.
751 	 * - total_mrs is available_mrs plus all in use MRs that could be
752 	 *   returned to the cache.
753 	 * - limit is the low water mark for available_mrs, 2* limit is the
754 	 *   upper water mark.
755 	 * - pending is the number of MRs currently being created
756 	 */
757 	u32 total_mrs;
758 	u32 available_mrs;
759 	u32 limit;
760 	u32 pending;
761 
762 	/* Statistics */
763 	u32                     miss;
764 
765 	struct mlx5_ib_dev     *dev;
766 	struct work_struct	work;
767 	struct delayed_work	dwork;
768 };
769 
770 struct mlx5_mr_cache {
771 	struct workqueue_struct *wq;
772 	struct mlx5_cache_ent	ent[MAX_MR_CACHE_ENTRIES];
773 	struct dentry		*root;
774 	unsigned long		last_add;
775 };
776 
777 struct mlx5_ib_port_resources {
778 	struct mlx5_ib_gsi_qp *gsi;
779 	struct work_struct pkey_change_work;
780 };
781 
782 struct mlx5_ib_resources {
783 	struct ib_cq	*c0;
784 	u32 xrcdn0;
785 	u32 xrcdn1;
786 	struct ib_pd	*p0;
787 	struct ib_srq	*s0;
788 	struct ib_srq	*s1;
789 	struct mlx5_ib_port_resources ports[2];
790 	/* Protects changes to the port resources */
791 	struct mutex	mutex;
792 };
793 
794 struct mlx5_ib_counters {
795 	const char **names;
796 	size_t *offsets;
797 	u32 num_q_counters;
798 	u32 num_cong_counters;
799 	u32 num_ext_ppcnt_counters;
800 	u16 set_id;
801 };
802 
803 struct mlx5_ib_multiport_info;
804 
805 struct mlx5_ib_multiport {
806 	struct mlx5_ib_multiport_info *mpi;
807 	/* To be held when accessing the multiport info */
808 	spinlock_t mpi_lock;
809 };
810 
811 struct mlx5_roce {
812 	/* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL
813 	 * netdev pointer
814 	 */
815 	rwlock_t		netdev_lock;
816 	struct net_device	*netdev;
817 	struct notifier_block	nb;
818 	atomic_t		tx_port_affinity;
819 	enum ib_port_state last_port_state;
820 	struct mlx5_ib_dev	*dev;
821 	u8			native_port_num;
822 };
823 
824 struct mlx5_ib_port {
825 	struct mlx5_ib_counters cnts;
826 	struct mlx5_ib_multiport mp;
827 	struct mlx5_ib_dbg_cc_params *dbg_cc_params;
828 	struct mlx5_roce roce;
829 	struct mlx5_eswitch_rep		*rep;
830 };
831 
832 struct mlx5_ib_dbg_param {
833 	int			offset;
834 	struct mlx5_ib_dev	*dev;
835 	struct dentry		*dentry;
836 	u8			port_num;
837 };
838 
839 enum mlx5_ib_dbg_cc_types {
840 	MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE,
841 	MLX5_IB_DBG_CC_RP_CLAMP_TGT_RATE_ATI,
842 	MLX5_IB_DBG_CC_RP_TIME_RESET,
843 	MLX5_IB_DBG_CC_RP_BYTE_RESET,
844 	MLX5_IB_DBG_CC_RP_THRESHOLD,
845 	MLX5_IB_DBG_CC_RP_AI_RATE,
846 	MLX5_IB_DBG_CC_RP_MAX_RATE,
847 	MLX5_IB_DBG_CC_RP_HAI_RATE,
848 	MLX5_IB_DBG_CC_RP_MIN_DEC_FAC,
849 	MLX5_IB_DBG_CC_RP_MIN_RATE,
850 	MLX5_IB_DBG_CC_RP_RATE_TO_SET_ON_FIRST_CNP,
851 	MLX5_IB_DBG_CC_RP_DCE_TCP_G,
852 	MLX5_IB_DBG_CC_RP_DCE_TCP_RTT,
853 	MLX5_IB_DBG_CC_RP_RATE_REDUCE_MONITOR_PERIOD,
854 	MLX5_IB_DBG_CC_RP_INITIAL_ALPHA_VALUE,
855 	MLX5_IB_DBG_CC_RP_GD,
856 	MLX5_IB_DBG_CC_NP_MIN_TIME_BETWEEN_CNPS,
857 	MLX5_IB_DBG_CC_NP_CNP_DSCP,
858 	MLX5_IB_DBG_CC_NP_CNP_PRIO_MODE,
859 	MLX5_IB_DBG_CC_NP_CNP_PRIO,
860 	MLX5_IB_DBG_CC_MAX,
861 };
862 
863 struct mlx5_ib_dbg_cc_params {
864 	struct dentry			*root;
865 	struct mlx5_ib_dbg_param	params[MLX5_IB_DBG_CC_MAX];
866 };
867 
868 enum {
869 	MLX5_MAX_DELAY_DROP_TIMEOUT_MS = 100,
870 };
871 
872 struct mlx5_ib_delay_drop {
873 	struct mlx5_ib_dev     *dev;
874 	struct work_struct	delay_drop_work;
875 	/* serialize setting of delay drop */
876 	struct mutex		lock;
877 	u32			timeout;
878 	bool			activate;
879 	atomic_t		events_cnt;
880 	atomic_t		rqs_cnt;
881 	struct dentry		*dir_debugfs;
882 };
883 
884 enum mlx5_ib_stages {
885 	MLX5_IB_STAGE_INIT,
886 	MLX5_IB_STAGE_FS,
887 	MLX5_IB_STAGE_CAPS,
888 	MLX5_IB_STAGE_NON_DEFAULT_CB,
889 	MLX5_IB_STAGE_ROCE,
890 	MLX5_IB_STAGE_QP,
891 	MLX5_IB_STAGE_SRQ,
892 	MLX5_IB_STAGE_DEVICE_RESOURCES,
893 	MLX5_IB_STAGE_DEVICE_NOTIFIER,
894 	MLX5_IB_STAGE_ODP,
895 	MLX5_IB_STAGE_COUNTERS,
896 	MLX5_IB_STAGE_CONG_DEBUGFS,
897 	MLX5_IB_STAGE_UAR,
898 	MLX5_IB_STAGE_BFREG,
899 	MLX5_IB_STAGE_PRE_IB_REG_UMR,
900 	MLX5_IB_STAGE_WHITELIST_UID,
901 	MLX5_IB_STAGE_IB_REG,
902 	MLX5_IB_STAGE_POST_IB_REG_UMR,
903 	MLX5_IB_STAGE_DELAY_DROP,
904 	MLX5_IB_STAGE_RESTRACK,
905 	MLX5_IB_STAGE_MAX,
906 };
907 
908 struct mlx5_ib_stage {
909 	int (*init)(struct mlx5_ib_dev *dev);
910 	void (*cleanup)(struct mlx5_ib_dev *dev);
911 };
912 
913 #define STAGE_CREATE(_stage, _init, _cleanup) \
914 	.stage[_stage] = {.init = _init, .cleanup = _cleanup}
915 
916 struct mlx5_ib_profile {
917 	struct mlx5_ib_stage stage[MLX5_IB_STAGE_MAX];
918 };
919 
920 struct mlx5_ib_multiport_info {
921 	struct list_head list;
922 	struct mlx5_ib_dev *ibdev;
923 	struct mlx5_core_dev *mdev;
924 	struct notifier_block mdev_events;
925 	struct completion unref_comp;
926 	u64 sys_image_guid;
927 	u32 mdev_refcnt;
928 	bool is_master;
929 	bool unaffiliate;
930 };
931 
932 struct mlx5_ib_flow_action {
933 	struct ib_flow_action		ib_action;
934 	union {
935 		struct {
936 			u64			    ib_flags;
937 			struct mlx5_accel_esp_xfrm *ctx;
938 		} esp_aes_gcm;
939 		struct {
940 			struct mlx5_ib_dev *dev;
941 			u32 sub_type;
942 			union {
943 				struct mlx5_modify_hdr *modify_hdr;
944 				struct mlx5_pkt_reformat *pkt_reformat;
945 			};
946 		} flow_action_raw;
947 	};
948 };
949 
950 struct mlx5_dm {
951 	struct mlx5_core_dev *dev;
952 	/* This lock is used to protect the access to the shared
953 	 * allocation map when concurrent requests by different
954 	 * processes are handled.
955 	 */
956 	spinlock_t lock;
957 	DECLARE_BITMAP(memic_alloc_pages, MLX5_MAX_MEMIC_PAGES);
958 };
959 
960 struct mlx5_read_counters_attr {
961 	struct mlx5_fc *hw_cntrs_hndl;
962 	u64 *out;
963 	u32 flags;
964 };
965 
966 enum mlx5_ib_counters_type {
967 	MLX5_IB_COUNTERS_FLOW,
968 };
969 
970 struct mlx5_ib_mcounters {
971 	struct ib_counters ibcntrs;
972 	enum mlx5_ib_counters_type type;
973 	/* number of counters supported for this counters type */
974 	u32 counters_num;
975 	struct mlx5_fc *hw_cntrs_hndl;
976 	/* read function for this counters type */
977 	int (*read_counters)(struct ib_device *ibdev,
978 			     struct mlx5_read_counters_attr *read_attr);
979 	/* max index set as part of create_flow */
980 	u32 cntrs_max_index;
981 	/* number of counters data entries (<description,index> pair) */
982 	u32 ncounters;
983 	/* counters data array for descriptions and indexes */
984 	struct mlx5_ib_flow_counters_desc *counters_data;
985 	/* protects access to mcounters internal data */
986 	struct mutex mcntrs_mutex;
987 };
988 
989 static inline struct mlx5_ib_mcounters *
990 to_mcounters(struct ib_counters *ibcntrs)
991 {
992 	return container_of(ibcntrs, struct mlx5_ib_mcounters, ibcntrs);
993 }
994 
995 int parse_flow_flow_action(struct mlx5_ib_flow_action *maction,
996 			   bool is_egress,
997 			   struct mlx5_flow_act *action);
998 struct mlx5_ib_lb_state {
999 	/* protect the user_td */
1000 	struct mutex		mutex;
1001 	u32			user_td;
1002 	int			qps;
1003 	bool			enabled;
1004 };
1005 
1006 struct mlx5_ib_pf_eq {
1007 	struct notifier_block irq_nb;
1008 	struct mlx5_ib_dev *dev;
1009 	struct mlx5_eq *core;
1010 	struct work_struct work;
1011 	spinlock_t lock; /* Pagefaults spinlock */
1012 	struct workqueue_struct *wq;
1013 	mempool_t *pool;
1014 };
1015 
1016 struct mlx5_devx_event_table {
1017 	struct mlx5_nb devx_nb;
1018 	/* serialize updating the event_xa */
1019 	struct mutex event_xa_lock;
1020 	struct xarray event_xa;
1021 };
1022 
1023 struct mlx5_var_table {
1024 	/* serialize updating the bitmap */
1025 	struct mutex bitmap_lock;
1026 	unsigned long *bitmap;
1027 	u64 hw_start_addr;
1028 	u32 stride_size;
1029 	u64 num_var_hw_entries;
1030 };
1031 
1032 struct mlx5_ib_dev {
1033 	struct ib_device		ib_dev;
1034 	struct mlx5_core_dev		*mdev;
1035 	struct notifier_block		mdev_events;
1036 	int				num_ports;
1037 	/* serialize update of capability mask
1038 	 */
1039 	struct mutex			cap_mask_mutex;
1040 	u8				ib_active:1;
1041 	u8				is_rep:1;
1042 	u8				lag_active:1;
1043 	u8				wc_support:1;
1044 	u8				fill_delay;
1045 	struct umr_common		umrc;
1046 	/* sync used page count stats
1047 	 */
1048 	struct mlx5_ib_resources	devr;
1049 
1050 	atomic_t			mkey_var;
1051 	struct mlx5_mr_cache		cache;
1052 	struct timer_list		delay_timer;
1053 	/* Prevents soft lock on massive reg MRs */
1054 	struct mutex			slow_path_mutex;
1055 	struct ib_odp_caps	odp_caps;
1056 	u64			odp_max_size;
1057 	struct mlx5_ib_pf_eq	odp_pf_eq;
1058 
1059 	/*
1060 	 * Sleepable RCU that prevents destruction of MRs while they are still
1061 	 * being used by a page fault handler.
1062 	 */
1063 	struct srcu_struct      odp_srcu;
1064 	struct xarray		odp_mkeys;
1065 
1066 	u32			null_mkey;
1067 	struct mlx5_ib_flow_db	*flow_db;
1068 	/* protect resources needed as part of reset flow */
1069 	spinlock_t		reset_flow_resource_lock;
1070 	struct list_head	qp_list;
1071 	/* Array with num_ports elements */
1072 	struct mlx5_ib_port	*port;
1073 	struct mlx5_sq_bfreg	bfreg;
1074 	struct mlx5_sq_bfreg	wc_bfreg;
1075 	struct mlx5_sq_bfreg	fp_bfreg;
1076 	struct mlx5_ib_delay_drop	delay_drop;
1077 	const struct mlx5_ib_profile	*profile;
1078 
1079 	struct mlx5_ib_lb_state		lb;
1080 	u8			umr_fence;
1081 	struct list_head	ib_dev_list;
1082 	u64			sys_image_guid;
1083 	struct mlx5_dm		dm;
1084 	u16			devx_whitelist_uid;
1085 	struct mlx5_srq_table   srq_table;
1086 	struct mlx5_qp_table    qp_table;
1087 	struct mlx5_async_ctx   async_ctx;
1088 	struct mlx5_devx_event_table devx_event_table;
1089 	struct mlx5_var_table var_table;
1090 
1091 	struct xarray sig_mrs;
1092 };
1093 
1094 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
1095 {
1096 	return container_of(mcq, struct mlx5_ib_cq, mcq);
1097 }
1098 
1099 static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
1100 {
1101 	return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd);
1102 }
1103 
1104 static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev)
1105 {
1106 	return container_of(ibdev, struct mlx5_ib_dev, ib_dev);
1107 }
1108 
1109 static inline struct mlx5_ib_dev *mr_to_mdev(struct mlx5_ib_mr *mr)
1110 {
1111 	return to_mdev(mr->ibmr.device);
1112 }
1113 
1114 static inline struct mlx5_ib_dev *mlx5_udata_to_mdev(struct ib_udata *udata)
1115 {
1116 	struct mlx5_ib_ucontext *context = rdma_udata_to_drv_context(
1117 		udata, struct mlx5_ib_ucontext, ibucontext);
1118 
1119 	return to_mdev(context->ibucontext.device);
1120 }
1121 
1122 static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq)
1123 {
1124 	return container_of(ibcq, struct mlx5_ib_cq, ibcq);
1125 }
1126 
1127 static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp)
1128 {
1129 	return container_of(mqp, struct mlx5_ib_qp_base, mqp)->container_mibqp;
1130 }
1131 
1132 static inline struct mlx5_ib_rwq *to_mibrwq(struct mlx5_core_qp *core_qp)
1133 {
1134 	return container_of(core_qp, struct mlx5_ib_rwq, core_qp);
1135 }
1136 
1137 static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd)
1138 {
1139 	return container_of(ibpd, struct mlx5_ib_pd, ibpd);
1140 }
1141 
1142 static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq)
1143 {
1144 	return container_of(ibsrq, struct mlx5_ib_srq, ibsrq);
1145 }
1146 
1147 static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp)
1148 {
1149 	return container_of(ibqp, struct mlx5_ib_qp, ibqp);
1150 }
1151 
1152 static inline struct mlx5_ib_rwq *to_mrwq(struct ib_wq *ibwq)
1153 {
1154 	return container_of(ibwq, struct mlx5_ib_rwq, ibwq);
1155 }
1156 
1157 static inline struct mlx5_ib_rwq_ind_table *to_mrwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl)
1158 {
1159 	return container_of(ib_rwq_ind_tbl, struct mlx5_ib_rwq_ind_table, ib_rwq_ind_tbl);
1160 }
1161 
1162 static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
1163 {
1164 	return container_of(msrq, struct mlx5_ib_srq, msrq);
1165 }
1166 
1167 static inline struct mlx5_ib_dm *to_mdm(struct ib_dm *ibdm)
1168 {
1169 	return container_of(ibdm, struct mlx5_ib_dm, ibdm);
1170 }
1171 
1172 static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
1173 {
1174 	return container_of(ibmr, struct mlx5_ib_mr, ibmr);
1175 }
1176 
1177 static inline struct mlx5_ib_mw *to_mmw(struct ib_mw *ibmw)
1178 {
1179 	return container_of(ibmw, struct mlx5_ib_mw, ibmw);
1180 }
1181 
1182 static inline struct mlx5_ib_flow_action *
1183 to_mflow_act(struct ib_flow_action *ibact)
1184 {
1185 	return container_of(ibact, struct mlx5_ib_flow_action, ib_action);
1186 }
1187 
1188 static inline struct mlx5_user_mmap_entry *
1189 to_mmmap(struct rdma_user_mmap_entry *rdma_entry)
1190 {
1191 	return container_of(rdma_entry,
1192 		struct mlx5_user_mmap_entry, rdma_entry);
1193 }
1194 
1195 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context,
1196 			struct ib_udata *udata, unsigned long virt,
1197 			struct mlx5_db *db);
1198 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);
1199 void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1200 void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
1201 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index);
1202 int mlx5_ib_create_ah(struct ib_ah *ah, struct rdma_ah_init_attr *init_attr,
1203 		      struct ib_udata *udata);
1204 int mlx5_ib_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr);
1205 static inline int mlx5_ib_destroy_ah(struct ib_ah *ah, u32 flags)
1206 {
1207 	return 0;
1208 }
1209 int mlx5_ib_create_srq(struct ib_srq *srq, struct ib_srq_init_attr *init_attr,
1210 		       struct ib_udata *udata);
1211 int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
1212 		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
1213 int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr);
1214 int mlx5_ib_destroy_srq(struct ib_srq *srq, struct ib_udata *udata);
1215 int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
1216 			  const struct ib_recv_wr **bad_wr);
1217 int mlx5_ib_enable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1218 void mlx5_ib_disable_lb(struct mlx5_ib_dev *dev, bool td, bool qp);
1219 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
1220 				struct ib_qp_init_attr *init_attr,
1221 				struct ib_udata *udata);
1222 int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1223 		      int attr_mask, struct ib_udata *udata);
1224 int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
1225 		     struct ib_qp_init_attr *qp_init_attr);
1226 int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata);
1227 void mlx5_ib_drain_sq(struct ib_qp *qp);
1228 void mlx5_ib_drain_rq(struct ib_qp *qp);
1229 int mlx5_ib_read_wqe_sq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1230 			size_t buflen, size_t *bc);
1231 int mlx5_ib_read_wqe_rq(struct mlx5_ib_qp *qp, int wqe_index, void *buffer,
1232 			size_t buflen, size_t *bc);
1233 int mlx5_ib_read_wqe_srq(struct mlx5_ib_srq *srq, int wqe_index, void *buffer,
1234 			 size_t buflen, size_t *bc);
1235 int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
1236 		      struct ib_udata *udata);
1237 int mlx5_ib_destroy_cq(struct ib_cq *cq, struct ib_udata *udata);
1238 int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
1239 int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
1240 int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1241 int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata);
1242 struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc);
1243 struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
1244 				  u64 virt_addr, int access_flags,
1245 				  struct ib_udata *udata);
1246 int mlx5_ib_advise_mr(struct ib_pd *pd,
1247 		      enum ib_uverbs_advise_mr_advice advice,
1248 		      u32 flags,
1249 		      struct ib_sge *sg_list,
1250 		      u32 num_sge,
1251 		      struct uverbs_attr_bundle *attrs);
1252 int mlx5_ib_alloc_mw(struct ib_mw *mw, struct ib_udata *udata);
1253 int mlx5_ib_dealloc_mw(struct ib_mw *mw);
1254 int mlx5_ib_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages,
1255 		       int page_shift, int flags);
1256 struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd,
1257 					     struct ib_udata *udata,
1258 					     int access_flags);
1259 void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *mr);
1260 void mlx5_ib_fence_odp_mr(struct mlx5_ib_mr *mr);
1261 struct ib_mr *mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start,
1262 				    u64 length, u64 virt_addr, int access_flags,
1263 				    struct ib_pd *pd, struct ib_udata *udata);
1264 int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata);
1265 struct ib_mr *mlx5_ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
1266 			       u32 max_num_sg);
1267 struct ib_mr *mlx5_ib_alloc_mr_integrity(struct ib_pd *pd,
1268 					 u32 max_num_sg,
1269 					 u32 max_num_meta_sg);
1270 int mlx5_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents,
1271 		      unsigned int *sg_offset);
1272 int mlx5_ib_map_mr_sg_pi(struct ib_mr *ibmr, struct scatterlist *data_sg,
1273 			 int data_sg_nents, unsigned int *data_sg_offset,
1274 			 struct scatterlist *meta_sg, int meta_sg_nents,
1275 			 unsigned int *meta_sg_offset);
1276 int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
1277 			const struct ib_wc *in_wc, const struct ib_grh *in_grh,
1278 			const struct ib_mad *in, struct ib_mad *out,
1279 			size_t *out_mad_size, u16 *out_mad_pkey_index);
1280 int mlx5_ib_alloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1281 int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
1282 int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, u8 port);
1283 int mlx5_query_mad_ifc_smp_attr_node_info(struct ib_device *ibdev,
1284 					  struct ib_smp *out_mad);
1285 int mlx5_query_mad_ifc_system_image_guid(struct ib_device *ibdev,
1286 					 __be64 *sys_image_guid);
1287 int mlx5_query_mad_ifc_max_pkeys(struct ib_device *ibdev,
1288 				 u16 *max_pkeys);
1289 int mlx5_query_mad_ifc_vendor_id(struct ib_device *ibdev,
1290 				 u32 *vendor_id);
1291 int mlx5_query_mad_ifc_node_desc(struct mlx5_ib_dev *dev, char *node_desc);
1292 int mlx5_query_mad_ifc_node_guid(struct mlx5_ib_dev *dev, __be64 *node_guid);
1293 int mlx5_query_mad_ifc_pkey(struct ib_device *ibdev, u8 port, u16 index,
1294 			    u16 *pkey);
1295 int mlx5_query_mad_ifc_gids(struct ib_device *ibdev, u8 port, int index,
1296 			    union ib_gid *gid);
1297 int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u8 port,
1298 			    struct ib_port_attr *props);
1299 int mlx5_ib_query_port(struct ib_device *ibdev, u8 port,
1300 		       struct ib_port_attr *props);
1301 void mlx5_ib_populate_pas(struct ib_umem *umem, size_t page_size, __be64 *pas,
1302 			  u64 access_flags);
1303 void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num);
1304 int mlx5_ib_get_cqe_size(struct ib_cq *ibcq);
1305 int mlx5_mr_cache_init(struct mlx5_ib_dev *dev);
1306 int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev);
1307 
1308 struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev,
1309 				       unsigned int entry, int access_flags);
1310 void mlx5_mr_cache_free(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr);
1311 int mlx5_mr_cache_invalidate(struct mlx5_ib_mr *mr);
1312 
1313 int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
1314 			    struct ib_mr_status *mr_status);
1315 struct ib_wq *mlx5_ib_create_wq(struct ib_pd *pd,
1316 				struct ib_wq_init_attr *init_attr,
1317 				struct ib_udata *udata);
1318 int mlx5_ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
1319 int mlx5_ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1320 		      u32 wq_attr_mask, struct ib_udata *udata);
1321 int mlx5_ib_create_rwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_table,
1322 				 struct ib_rwq_ind_table_init_attr *init_attr,
1323 				 struct ib_udata *udata);
1324 int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
1325 struct ib_dm *mlx5_ib_alloc_dm(struct ib_device *ibdev,
1326 			       struct ib_ucontext *context,
1327 			       struct ib_dm_alloc_attr *attr,
1328 			       struct uverbs_attr_bundle *attrs);
1329 int mlx5_ib_dealloc_dm(struct ib_dm *ibdm, struct uverbs_attr_bundle *attrs);
1330 struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm,
1331 				struct ib_dm_mr_attr *attr,
1332 				struct uverbs_attr_bundle *attrs);
1333 
1334 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1335 void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev);
1336 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);
1337 void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev);
1338 int __init mlx5_ib_odp_init(void);
1339 void mlx5_ib_odp_cleanup(void);
1340 void mlx5_odp_init_mr_cache_entry(struct mlx5_cache_ent *ent);
1341 void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1342 			   struct mlx5_ib_mr *mr, int flags);
1343 
1344 int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1345 			       enum ib_uverbs_advise_mr_advice advice,
1346 			       u32 flags, struct ib_sge *sg_list, u32 num_sge);
1347 int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr);
1348 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1349 static inline void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev)
1350 {
1351 	return;
1352 }
1353 
1354 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; }
1355 static inline void mlx5_ib_odp_cleanup_one(struct mlx5_ib_dev *ibdev) {}
1356 static inline int mlx5_ib_odp_init(void) { return 0; }
1357 static inline void mlx5_ib_odp_cleanup(void)				    {}
1358 static inline void mlx5_odp_init_mr_cache_entry(struct mlx5_cache_ent *ent) {}
1359 static inline void mlx5_odp_populate_xlt(void *xlt, size_t idx, size_t nentries,
1360 					 struct mlx5_ib_mr *mr, int flags) {}
1361 
1362 static inline int
1363 mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
1364 			   enum ib_uverbs_advise_mr_advice advice, u32 flags,
1365 			   struct ib_sge *sg_list, u32 num_sge)
1366 {
1367 	return -EOPNOTSUPP;
1368 }
1369 static inline int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr)
1370 {
1371 	return -EOPNOTSUPP;
1372 }
1373 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1374 
1375 extern const struct mmu_interval_notifier_ops mlx5_mn_ops;
1376 
1377 /* Needed for rep profile */
1378 void __mlx5_ib_remove(struct mlx5_ib_dev *dev,
1379 		      const struct mlx5_ib_profile *profile,
1380 		      int stage);
1381 int __mlx5_ib_add(struct mlx5_ib_dev *dev,
1382 		  const struct mlx5_ib_profile *profile);
1383 
1384 int mlx5_ib_get_vf_config(struct ib_device *device, int vf,
1385 			  u8 port, struct ifla_vf_info *info);
1386 int mlx5_ib_set_vf_link_state(struct ib_device *device, int vf,
1387 			      u8 port, int state);
1388 int mlx5_ib_get_vf_stats(struct ib_device *device, int vf,
1389 			 u8 port, struct ifla_vf_stats *stats);
1390 int mlx5_ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
1391 			struct ifla_vf_guid *node_guid,
1392 			struct ifla_vf_guid *port_guid);
1393 int mlx5_ib_set_vf_guid(struct ib_device *device, int vf, u8 port,
1394 			u64 guid, int type);
1395 
1396 __be16 mlx5_get_roce_udp_sport_min(const struct mlx5_ib_dev *dev,
1397 				   const struct ib_gid_attr *attr);
1398 
1399 void mlx5_ib_cleanup_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num);
1400 void mlx5_ib_init_cong_debugfs(struct mlx5_ib_dev *dev, u8 port_num);
1401 
1402 /* GSI QP helper functions */
1403 int mlx5_ib_create_gsi(struct ib_pd *pd, struct mlx5_ib_qp *mqp,
1404 		       struct ib_qp_init_attr *attr);
1405 int mlx5_ib_destroy_gsi(struct mlx5_ib_qp *mqp);
1406 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
1407 			  int attr_mask);
1408 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
1409 			 int qp_attr_mask,
1410 			 struct ib_qp_init_attr *qp_init_attr);
1411 int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
1412 			  const struct ib_send_wr **bad_wr);
1413 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, const struct ib_recv_wr *wr,
1414 			  const struct ib_recv_wr **bad_wr);
1415 void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi);
1416 
1417 int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc);
1418 
1419 void mlx5_ib_free_bfreg(struct mlx5_ib_dev *dev, struct mlx5_bfreg_info *bfregi,
1420 			int bfregn);
1421 struct mlx5_ib_dev *mlx5_ib_get_ibdev_from_mpi(struct mlx5_ib_multiport_info *mpi);
1422 struct mlx5_core_dev *mlx5_ib_get_native_port_mdev(struct mlx5_ib_dev *dev,
1423 						   u8 ib_port_num,
1424 						   u8 *native_port_num);
1425 void mlx5_ib_put_native_port_mdev(struct mlx5_ib_dev *dev,
1426 				  u8 port_num);
1427 
1428 extern const struct uapi_definition mlx5_ib_devx_defs[];
1429 extern const struct uapi_definition mlx5_ib_flow_defs[];
1430 extern const struct uapi_definition mlx5_ib_qos_defs[];
1431 extern const struct uapi_definition mlx5_ib_std_types_defs[];
1432 
1433 static inline void init_query_mad(struct ib_smp *mad)
1434 {
1435 	mad->base_version  = 1;
1436 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
1437 	mad->class_version = 1;
1438 	mad->method	   = IB_MGMT_METHOD_GET;
1439 }
1440 
1441 static inline int is_qp1(enum ib_qp_type qp_type)
1442 {
1443 	return qp_type == MLX5_IB_QPT_HW_GSI || qp_type == IB_QPT_GSI;
1444 }
1445 
1446 #define MLX5_MAX_UMR_SHIFT 16
1447 #define MLX5_MAX_UMR_PAGES (1 << MLX5_MAX_UMR_SHIFT)
1448 
1449 static inline u32 check_cq_create_flags(u32 flags)
1450 {
1451 	/*
1452 	 * It returns non-zero value for unsupported CQ
1453 	 * create flags, otherwise it returns zero.
1454 	 */
1455 	return (flags & ~(IB_UVERBS_CQ_FLAGS_IGNORE_OVERRUN |
1456 			  IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION));
1457 }
1458 
1459 static inline int verify_assign_uidx(u8 cqe_version, u32 cmd_uidx,
1460 				     u32 *user_index)
1461 {
1462 	if (cqe_version) {
1463 		if ((cmd_uidx == MLX5_IB_DEFAULT_UIDX) ||
1464 		    (cmd_uidx & ~MLX5_USER_ASSIGNED_UIDX_MASK))
1465 			return -EINVAL;
1466 		*user_index = cmd_uidx;
1467 	} else {
1468 		*user_index = MLX5_IB_DEFAULT_UIDX;
1469 	}
1470 
1471 	return 0;
1472 }
1473 
1474 static inline int get_qp_user_index(struct mlx5_ib_ucontext *ucontext,
1475 				    struct mlx5_ib_create_qp *ucmd,
1476 				    int inlen,
1477 				    u32 *user_index)
1478 {
1479 	u8 cqe_version = ucontext->cqe_version;
1480 
1481 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1482 	    (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1483 		return 0;
1484 
1485 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1486 		return -EINVAL;
1487 
1488 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1489 }
1490 
1491 static inline int get_srq_user_index(struct mlx5_ib_ucontext *ucontext,
1492 				     struct mlx5_ib_create_srq *ucmd,
1493 				     int inlen,
1494 				     u32 *user_index)
1495 {
1496 	u8 cqe_version = ucontext->cqe_version;
1497 
1498 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) && !cqe_version &&
1499 	    (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1500 		return 0;
1501 
1502 	if ((offsetofend(typeof(*ucmd), uidx) <= inlen) != !!cqe_version)
1503 		return -EINVAL;
1504 
1505 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1506 }
1507 
1508 static inline int get_uars_per_sys_page(struct mlx5_ib_dev *dev, bool lib_support)
1509 {
1510 	return lib_support && MLX5_CAP_GEN(dev->mdev, uar_4k) ?
1511 				MLX5_UARS_IN_PAGE : 1;
1512 }
1513 
1514 static inline int get_num_static_uars(struct mlx5_ib_dev *dev,
1515 				      struct mlx5_bfreg_info *bfregi)
1516 {
1517 	return get_uars_per_sys_page(dev, bfregi->lib_uar_4k) * bfregi->num_static_sys_pages;
1518 }
1519 
1520 extern void *xlt_emergency_page;
1521 
1522 int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
1523 			struct mlx5_bfreg_info *bfregi, u32 bfregn,
1524 			bool dyn_bfreg);
1525 
1526 static inline bool mlx5_ib_can_load_pas_with_umr(struct mlx5_ib_dev *dev,
1527 						 size_t length)
1528 {
1529 	/*
1530 	 * umr_check_mkey_mask() rejects MLX5_MKEY_MASK_PAGE_SIZE which is
1531 	 * always set if MLX5_IB_SEND_UMR_UPDATE_TRANSLATION (aka
1532 	 * MLX5_IB_UPD_XLT_ADDR and MLX5_IB_UPD_XLT_ENABLE) is set. Thus, a mkey
1533 	 * can never be enabled without this capability. Simplify this weird
1534 	 * quirky hardware by just saying it can't use PAS lists with UMR at
1535 	 * all.
1536 	 */
1537 	if (MLX5_CAP_GEN(dev->mdev, umr_modify_entity_size_disabled))
1538 		return false;
1539 
1540 	/*
1541 	 * length is the size of the MR in bytes when mlx5_ib_update_xlt() is
1542 	 * used.
1543 	 */
1544 	if (!MLX5_CAP_GEN(dev->mdev, umr_extended_translation_offset) &&
1545 	    length >= MLX5_MAX_UMR_PAGES * PAGE_SIZE)
1546 		return false;
1547 	return true;
1548 }
1549 
1550 /*
1551  * true if an existing MR can be reconfigured to new access_flags using UMR.
1552  * Older HW cannot use UMR to update certain elements of the MKC. See
1553  * umr_check_mkey_mask(), get_umr_update_access_mask() and umr_check_mkey_mask()
1554  */
1555 static inline bool mlx5_ib_can_reconfig_with_umr(struct mlx5_ib_dev *dev,
1556 						 unsigned int current_access_flags,
1557 						 unsigned int target_access_flags)
1558 {
1559 	unsigned int diffs = current_access_flags ^ target_access_flags;
1560 
1561 	if ((diffs & IB_ACCESS_REMOTE_ATOMIC) &&
1562 	    MLX5_CAP_GEN(dev->mdev, atomic) &&
1563 	    MLX5_CAP_GEN(dev->mdev, umr_modify_atomic_disabled))
1564 		return false;
1565 
1566 	if ((diffs & IB_ACCESS_RELAXED_ORDERING) &&
1567 	    MLX5_CAP_GEN(dev->mdev, relaxed_ordering_write) &&
1568 	    !MLX5_CAP_GEN(dev->mdev, relaxed_ordering_write_umr))
1569 		return false;
1570 
1571 	if ((diffs & IB_ACCESS_RELAXED_ORDERING) &&
1572 	    MLX5_CAP_GEN(dev->mdev, relaxed_ordering_read) &&
1573 	    !MLX5_CAP_GEN(dev->mdev, relaxed_ordering_read_umr))
1574 		return false;
1575 
1576 	return true;
1577 }
1578 
1579 int mlx5_ib_test_wc(struct mlx5_ib_dev *dev);
1580 
1581 static inline bool mlx5_ib_lag_should_assign_affinity(struct mlx5_ib_dev *dev)
1582 {
1583 	return dev->lag_active ||
1584 		(MLX5_CAP_GEN(dev->mdev, num_lag_ports) > 1 &&
1585 		 MLX5_CAP_GEN(dev->mdev, lag_tx_port_affinity));
1586 }
1587 #endif /* MLX5_IB_H */
1588