xref: /freebsd/sys/dev/mlx5/mlx5_ib/mlx5_ib.h (revision e23731db48ef9c6568d4768b1f87d48514339faa)
1 /*-
2  * Copyright (c) 2013-2020, Mellanox Technologies, Ltd.  All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  *
13  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND
14  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16  * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
17  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23  * SUCH DAMAGE.
24  */
25 
26 #ifndef MLX5_IB_H
27 #define MLX5_IB_H
28 
29 #include <linux/kernel.h>
30 #include <linux/sched.h>
31 #include <linux/printk.h>
32 #include <linux/netdevice.h>
33 #include <rdma/ib_verbs.h>
34 #include <rdma/ib_umem.h>
35 #include <rdma/ib_smi.h>
36 #include <dev/mlx5/cq.h>
37 #include <dev/mlx5/qp.h>
38 #include <dev/mlx5/srq.h>
39 #include <linux/types.h>
40 #include <dev/mlx5/mlx5_core/transobj.h>
41 #include <rdma/ib_user_verbs.h>
42 #include <rdma/mlx5-abi.h>
43 #include <rdma/uverbs_ioctl.h>
44 
45 #define mlx5_ib_dbg(dev, format, arg...)				\
46 pr_debug("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__,	\
47 	 __LINE__, current->pid, ##arg)
48 
49 #define mlx5_ib_err(dev, format, arg...)				\
50 pr_err("%s: ERR: %s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__,	\
51 	__LINE__, current->pid, ##arg)
52 
53 #define mlx5_ib_warn(dev, format, arg...)				\
54 pr_warn("%s: WARN: %s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__,	\
55 	__LINE__, current->pid, ##arg)
56 
57 #define field_avail(type, fld, sz) (offsetof(type, fld) +		\
58 				    sizeof(((type *)0)->fld) <= (sz))
59 #define MLX5_IB_DEFAULT_UIDX 0xffffff
60 #define MLX5_USER_ASSIGNED_UIDX_MASK __mlx5_mask(qpc, user_index)
61 
62 #define MLX5_MKEY_PAGE_SHIFT_MASK __mlx5_mask(mkc, log_page_size)
63 
64 enum {
65 	MLX5_IB_MMAP_CMD_SHIFT	= 8,
66 	MLX5_IB_MMAP_CMD_MASK	= 0xff,
67 };
68 
69 enum {
70 	MLX5_RES_SCAT_DATA32_CQE	= 0x1,
71 	MLX5_RES_SCAT_DATA64_CQE	= 0x2,
72 	MLX5_REQ_SCAT_DATA32_CQE	= 0x11,
73 	MLX5_REQ_SCAT_DATA64_CQE	= 0x22,
74 };
75 
76 enum mlx5_ib_latency_class {
77 	MLX5_IB_LATENCY_CLASS_LOW,
78 	MLX5_IB_LATENCY_CLASS_MEDIUM,
79 	MLX5_IB_LATENCY_CLASS_HIGH,
80 	MLX5_IB_LATENCY_CLASS_FAST_PATH
81 };
82 
83 enum mlx5_ib_mad_ifc_flags {
84 	MLX5_MAD_IFC_IGNORE_MKEY	= 1,
85 	MLX5_MAD_IFC_IGNORE_BKEY	= 2,
86 	MLX5_MAD_IFC_NET_VIEW		= 4,
87 };
88 
89 enum {
90 	MLX5_CROSS_CHANNEL_BFREG         = 0,
91 };
92 
93 enum {
94 	MLX5_CQE_VERSION_V0,
95 	MLX5_CQE_VERSION_V1,
96 };
97 
98 enum {
99 	MLX5_IB_INVALID_UAR_INDEX	= BIT(31),
100 	MLX5_IB_INVALID_BFREG		= BIT(31),
101 };
102 
103 enum mlx5_ib_mmap_type {
104 	MLX5_IB_MMAP_TYPE_MEMIC = 1,
105 	MLX5_IB_MMAP_TYPE_VAR = 2,
106 	MLX5_IB_MMAP_TYPE_UAR_WC = 3,
107 	MLX5_IB_MMAP_TYPE_UAR_NC = 4,
108 };
109 
110 struct mlx5_bfreg_info {
111 	u32 *sys_pages;
112 	int num_low_latency_bfregs;
113 	unsigned int *count;
114 
115 	/*
116 	 * protect bfreg allocation data structs
117 	 */
118 	struct mutex lock;
119 	u32 ver;
120 	u8 lib_uar_4k : 1;
121 	u8 lib_uar_dyn : 1;
122 	u32 num_sys_pages;
123 	u32 num_static_sys_pages;
124 	u32 total_num_bfregs;
125 	u32 num_dyn_bfregs;
126 };
127 
128 struct mlx5_ib_ucontext {
129 	struct ib_ucontext	ibucontext;
130 	struct list_head	db_page_list;
131 
132 	/* protect doorbell record alloc/free
133 	 */
134 	struct mutex		db_page_mutex;
135 	struct mlx5_bfreg_info	bfregi;
136 	u8			cqe_version;
137 	/* Transport Domain number */
138 	u32			tdn;
139 
140 	u64			lib_caps;
141 	u16			devx_uid;
142 };
143 
144 static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
145 {
146 	return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext);
147 }
148 
149 struct mlx5_ib_pd {
150 	struct ib_pd		ibpd;
151 	u32			pdn;
152 	u16			uid;
153 };
154 
155 #define MLX5_IB_FLOW_MCAST_PRIO		(MLX5_BY_PASS_NUM_PRIOS - 1)
156 #define MLX5_IB_FLOW_LAST_PRIO		(MLX5_BY_PASS_NUM_REGULAR_PRIOS - 1)
157 #if (MLX5_IB_FLOW_LAST_PRIO <= 0)
158 #error "Invalid number of bypass priorities"
159 #endif
160 #define MLX5_IB_FLOW_LEFTOVERS_PRIO	(MLX5_IB_FLOW_MCAST_PRIO + 1)
161 
162 #define MLX5_IB_NUM_FLOW_FT		(MLX5_IB_FLOW_LEFTOVERS_PRIO + 1)
163 #define MLX5_IB_NUM_SNIFFER_FTS		2
164 struct mlx5_ib_flow_prio {
165 	struct mlx5_flow_table		*flow_table;
166 	unsigned int			refcount;
167 };
168 
169 struct mlx5_ib_flow_handler {
170 	struct list_head		list;
171 	struct ib_flow			ibflow;
172 	struct mlx5_ib_flow_prio	*prio;
173 	struct mlx5_flow_handle	*rule;
174 };
175 
176 struct mlx5_ib_flow_db {
177 	struct mlx5_ib_flow_prio	prios[MLX5_IB_NUM_FLOW_FT];
178 	struct mlx5_ib_flow_prio	sniffer[MLX5_IB_NUM_SNIFFER_FTS];
179 	struct mlx5_flow_table		*lag_demux_ft;
180 	/* Protect flow steering bypass flow tables
181 	 * when add/del flow rules.
182 	 * only single add/removal of flow steering rule could be done
183 	 * simultaneously.
184 	 */
185 	struct mutex			lock;
186 };
187 
188 /* Use macros here so that don't have to duplicate
189  * enum ib_send_flags and enum ib_qp_type for low-level driver
190  */
191 
192 #define MLX5_IB_SEND_UMR_UNREG	IB_SEND_RESERVED_START
193 #define MLX5_IB_SEND_UMR_FAIL_IF_FREE (IB_SEND_RESERVED_START << 1)
194 #define MLX5_IB_SEND_UMR_UPDATE_MTT (IB_SEND_RESERVED_START << 2)
195 
196 #define MLX5_IB_SEND_UMR_UPDATE_TRANSLATION	(IB_SEND_RESERVED_START << 3)
197 #define MLX5_IB_SEND_UMR_UPDATE_PD		(IB_SEND_RESERVED_START << 4)
198 #define MLX5_IB_SEND_UMR_UPDATE_ACCESS		IB_SEND_RESERVED_END
199 
200 #define MLX5_IB_QPT_REG_UMR	IB_QPT_RESERVED1
201 /*
202  * IB_QPT_GSI creates the software wrapper around GSI, and MLX5_IB_QPT_HW_GSI
203  * creates the actual hardware QP.
204  */
205 #define MLX5_IB_QPT_HW_GSI	IB_QPT_RESERVED2
206 #define MLX5_IB_QPT_DCI		IB_QPT_RESERVED3
207 #define MLX5_IB_QPT_DCT		IB_QPT_RESERVED4
208 #define MLX5_IB_WR_UMR		IB_WR_RESERVED1
209 
210 /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags.
211  *
212  * These flags are intended for internal use by the mlx5_ib driver, and they
213  * rely on the range reserved for that use in the ib_qp_create_flags enum.
214  */
215 #define MLX5_IB_QP_CREATE_SQPN_QP1	IB_QP_CREATE_RESERVED_START
216 #define MLX5_IB_QP_CREATE_WC_TEST	(IB_QP_CREATE_RESERVED_START << 1)
217 
218 struct wr_list {
219 	u16	opcode;
220 	u16	next;
221 };
222 
223 struct mlx5_ib_wq {
224 	u64		       *wrid;
225 	u32		       *wr_data;
226 	struct wr_list	       *w_list;
227 	unsigned	       *wqe_head;
228 	u16		        unsig_count;
229 
230 	/* serialize post to the work queue
231 	 */
232 	spinlock_t		lock;
233 	int			wqe_cnt;
234 	int			max_post;
235 	int			max_gs;
236 	int			offset;
237 	int			wqe_shift;
238 	unsigned		head;
239 	unsigned		tail;
240 	u16			cur_post;
241 	u16			last_poll;
242 	void		       *qend;
243 };
244 
245 struct mlx5_ib_rwq {
246 	struct ib_wq		ibwq;
247 	struct mlx5_core_qp	core_qp;
248 	u32			rq_num_pas;
249 	u32			log_rq_stride;
250 	u32			log_rq_size;
251 	u32			rq_page_offset;
252 	u32			log_page_size;
253 	struct ib_umem		*umem;
254 	size_t			buf_size;
255 	unsigned int		page_shift;
256 	int			create_type;
257 	struct mlx5_db		db;
258 	u32			user_index;
259 	u32			wqe_count;
260 	u32			wqe_shift;
261 	int			wq_sig;
262 };
263 
264 enum {
265 	MLX5_QP_USER,
266 	MLX5_QP_KERNEL,
267 	MLX5_QP_EMPTY
268 };
269 
270 enum {
271 	MLX5_WQ_USER,
272 	MLX5_WQ_KERNEL
273 };
274 
275 struct mlx5_ib_rwq_ind_table {
276 	struct ib_rwq_ind_table ib_rwq_ind_tbl;
277 	u32			rqtn;
278 	u16			uid;
279 };
280 
281 /*
282  * Connect-IB can trigger up to four concurrent pagefaults
283  * per-QP.
284  */
285 enum mlx5_ib_pagefault_context {
286 	MLX5_IB_PAGEFAULT_RESPONDER_READ,
287 	MLX5_IB_PAGEFAULT_REQUESTOR_READ,
288 	MLX5_IB_PAGEFAULT_RESPONDER_WRITE,
289 	MLX5_IB_PAGEFAULT_REQUESTOR_WRITE,
290 	MLX5_IB_PAGEFAULT_CONTEXTS
291 };
292 
293 static inline enum mlx5_ib_pagefault_context
294 	mlx5_ib_get_pagefault_context(struct mlx5_pagefault *pagefault)
295 {
296 	return pagefault->flags & (MLX5_PFAULT_REQUESTOR | MLX5_PFAULT_WRITE);
297 }
298 
299 struct mlx5_ib_pfault {
300 	struct work_struct	work;
301 	struct mlx5_pagefault	mpfault;
302 };
303 
304 struct mlx5_ib_ubuffer {
305 	struct ib_umem	       *umem;
306 	int			buf_size;
307 	u64			buf_addr;
308 };
309 
310 struct mlx5_ib_qp_base {
311 	struct mlx5_ib_qp	*container_mibqp;
312 	struct mlx5_core_qp	mqp;
313 	struct mlx5_ib_ubuffer	ubuffer;
314 };
315 
316 struct mlx5_ib_qp_trans {
317 	struct mlx5_ib_qp_base	base;
318 	u16			xrcdn;
319 	u8			alt_port;
320 	u8			atomic_rd_en;
321 	u8			resp_depth;
322 };
323 
324 struct mlx5_ib_rss_qp {
325 	u32	tirn;
326 };
327 
328 struct mlx5_ib_rq {
329 	struct mlx5_ib_qp_base base;
330 	struct mlx5_ib_wq	*rq;
331 	struct mlx5_ib_ubuffer	ubuffer;
332 	struct mlx5_db		*doorbell;
333 	u32			tirn;
334 	u8			state;
335 };
336 
337 struct mlx5_ib_sq {
338 	struct mlx5_ib_qp_base base;
339 	struct mlx5_ib_wq	*sq;
340 	struct mlx5_ib_ubuffer  ubuffer;
341 	struct mlx5_db		*doorbell;
342 	u32			tisn;
343 	u8			state;
344 };
345 
346 struct mlx5_ib_raw_packet_qp {
347 	struct mlx5_ib_sq sq;
348 	struct mlx5_ib_rq rq;
349 };
350 
351 struct mlx5_bf {
352 	int			buf_size;
353 	unsigned long		offset;
354 	struct mlx5_sq_bfreg   *bfreg;
355 	spinlock_t		lock32;
356 };
357 
358 struct mlx5_ib_dct {
359 	struct mlx5_core_dct    mdct;
360 	u32                     *in;
361 };
362 
363 struct mlx5_ib_qp {
364 	struct ib_qp		ibqp;
365 	union {
366 		struct mlx5_ib_qp_trans trans_qp;
367 		struct mlx5_ib_raw_packet_qp raw_packet_qp;
368 		struct mlx5_ib_rss_qp rss_qp;
369 		struct mlx5_ib_dct dct;
370 	};
371 	struct mlx5_buf		buf;
372 
373 	struct mlx5_db		db;
374 	struct mlx5_ib_wq	rq;
375 
376 	u8			sq_signal_bits;
377 	u8			fm_cache;
378 	struct mlx5_ib_wq	sq;
379 
380 	/* serialize qp state modifications
381 	 */
382 	struct mutex		mutex;
383 	u32			flags;
384 	u8			port;
385 	u8			state;
386 	int			wq_sig;
387 	int			scat_cqe;
388 	int			max_inline_data;
389 	struct mlx5_bf	        bf;
390 	int			has_rq;
391 
392 	/* only for user space QPs. For kernel
393 	 * we have it from the bf object
394 	 */
395 	int			bfregn;
396 
397 	int			create_type;
398 
399 	/* Store signature errors */
400 	bool			signature_en;
401 
402 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
403 	/*
404 	 * A flag that is true for QP's that are in a state that doesn't
405 	 * allow page faults, and shouldn't schedule any more faults.
406 	 */
407 	int                     disable_page_faults;
408 	/*
409 	 * The disable_page_faults_lock protects a QP's disable_page_faults
410 	 * field, allowing for a thread to atomically check whether the QP
411 	 * allows page faults, and if so schedule a page fault.
412 	 */
413 	spinlock_t              disable_page_faults_lock;
414 	struct mlx5_ib_pfault	pagefaults[MLX5_IB_PAGEFAULT_CONTEXTS];
415 #endif
416 	struct list_head	qps_list;
417 	struct list_head	cq_recv_list;
418 	struct list_head	cq_send_list;
419 };
420 
421 struct mlx5_ib_cq_buf {
422 	struct mlx5_buf		buf;
423 	struct ib_umem		*umem;
424 	int			cqe_size;
425 	int			nent;
426 };
427 
428 enum mlx5_ib_qp_flags {
429 	MLX5_IB_QP_LSO                          = IB_QP_CREATE_IPOIB_UD_LSO,
430 	MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK     = IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
431 	MLX5_IB_QP_CROSS_CHANNEL            = IB_QP_CREATE_CROSS_CHANNEL,
432 	MLX5_IB_QP_MANAGED_SEND             = IB_QP_CREATE_MANAGED_SEND,
433 	MLX5_IB_QP_MANAGED_RECV             = IB_QP_CREATE_MANAGED_RECV,
434 	MLX5_IB_QP_SIGNATURE_HANDLING           = 1 << 5,
435 	/* QP uses 1 as its source QP number */
436 	MLX5_IB_QP_SQPN_QP1			= 1 << 6,
437 	MLX5_IB_QP_CAP_SCATTER_FCS		= 1 << 7,
438 	MLX5_IB_QP_RSS				= 1 << 8,
439 	MLX5_IB_QP_UNDERLAY			= 1 << 10,
440 };
441 
442 struct mlx5_umr_wr {
443 	struct ib_send_wr		wr;
444 	union {
445 		u64			virt_addr;
446 		u64			offset;
447 	} target;
448 	struct ib_pd		       *pd;
449 	unsigned int			page_shift;
450 	unsigned int			npages;
451 	u32				length;
452 	int				access_flags;
453 	u32				mkey;
454 };
455 
456 static inline const struct mlx5_umr_wr *umr_wr(const struct ib_send_wr *wr)
457 {
458 	return container_of(wr, struct mlx5_umr_wr, wr);
459 }
460 
461 struct mlx5_shared_mr_info {
462 	int mr_id;
463 	struct ib_umem		*umem;
464 };
465 
466 struct mlx5_ib_cq {
467 	struct ib_cq		ibcq;
468 	struct mlx5_core_cq	mcq;
469 	struct mlx5_ib_cq_buf	buf;
470 	struct mlx5_db		db;
471 
472 	/* serialize access to the CQ
473 	 */
474 	spinlock_t		lock;
475 
476 	/* protect resize cq
477 	 */
478 	struct mutex		resize_mutex;
479 	struct mlx5_ib_cq_buf  *resize_buf;
480 	struct ib_umem	       *resize_umem;
481 	int			cqe_size;
482 	struct list_head	list_send_qp;
483 	struct list_head	list_recv_qp;
484 	u32			create_flags;
485 	struct list_head	wc_list;
486 	enum ib_cq_notify_flags notify_flags;
487 	struct work_struct	notify_work;
488 };
489 
490 struct mlx5_ib_wc {
491 	struct ib_wc wc;
492 	struct list_head list;
493 };
494 
495 struct mlx5_ib_srq {
496 	struct ib_srq		ibsrq;
497 	struct mlx5_core_srq	msrq;
498 	struct mlx5_buf		buf;
499 	struct mlx5_db		db;
500 	u64		       *wrid;
501 	/* protect SRQ hanlding
502 	 */
503 	spinlock_t		lock;
504 	int			head;
505 	int			tail;
506 	u16			wqe_ctr;
507 	struct ib_umem	       *umem;
508 	/* serialize arming a SRQ
509 	 */
510 	struct mutex		mutex;
511 	int			wq_sig;
512 };
513 
514 struct mlx5_ib_xrcd {
515 	struct ib_xrcd		ibxrcd;
516 	u32			xrcdn;
517 };
518 
519 enum mlx5_ib_mtt_access_flags {
520 	MLX5_IB_MTT_READ  = (1 << 0),
521 	MLX5_IB_MTT_WRITE = (1 << 1),
522 };
523 
524 struct mlx5_user_mmap_entry {
525 	struct rdma_user_mmap_entry rdma_entry;
526 	u8 mmap_flag;
527 	u64 address;
528 	u32 page_idx;
529 };
530 
531 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
532 
533 struct mlx5_ib_mr {
534 	struct ib_mr		ibmr;
535 	void			*descs;
536 	dma_addr_t		desc_map;
537 	int			ndescs;
538 	int			max_descs;
539 	int			desc_size;
540 	int			access_mode;
541 	struct mlx5_core_mkey	mmkey;
542 	struct ib_umem	       *umem;
543 	struct mlx5_shared_mr_info	*smr_info;
544 	struct list_head	list;
545 	int			order;
546 	int			umred;
547 	int			npages;
548 	struct mlx5_ib_dev     *dev;
549 	u32 out[MLX5_ST_SZ_DW(create_mkey_out)];
550 	struct mlx5_core_sig_ctx    *sig;
551 	int			live;
552 	void			*descs_alloc;
553 	int			access_flags; /* Needed for rereg MR */
554 	struct mlx5_async_work	cb_work;
555 };
556 
557 struct mlx5_ib_mw {
558 	struct ib_mw		ibmw;
559 	struct mlx5_core_mkey	mmkey;
560 };
561 
562 struct mlx5_ib_devx_mr {
563 	struct mlx5_core_mkey	mmkey;
564 	int			ndescs;
565 };
566 
567 struct mlx5_ib_umr_context {
568 	struct ib_cqe		cqe;
569 	enum ib_wc_status	status;
570 	struct completion	done;
571 };
572 
573 struct umr_common {
574 	struct ib_pd	*pd;
575 	struct ib_cq	*cq;
576 	struct ib_qp	*qp;
577 	/* control access to UMR QP
578 	 */
579 	struct semaphore	sem;
580 };
581 
582 enum {
583 	MLX5_FMR_INVALID,
584 	MLX5_FMR_VALID,
585 	MLX5_FMR_BUSY,
586 };
587 
588 struct mlx5_cache_ent {
589 	struct list_head	head;
590 	/* sync access to the cahce entry
591 	 */
592 	spinlock_t		lock;
593 
594 
595 	char                    name[4];
596 	u32                     order;
597 	u32			size;
598 	u32                     cur;
599 	u32                     miss;
600 	u32			limit;
601 
602 	struct mlx5_ib_dev     *dev;
603 	struct work_struct	work;
604 	struct delayed_work	dwork;
605 	int			pending;
606 };
607 
608 struct mlx5_mr_cache {
609 	struct workqueue_struct *wq;
610 	struct mlx5_cache_ent	ent[MAX_MR_CACHE_ENTRIES];
611 	int			stopped;
612 	unsigned long		last_add;
613 };
614 
615 struct mlx5_ib_gsi_qp;
616 
617 struct mlx5_ib_port_resources {
618 	struct mlx5_ib_resources *devr;
619 	struct mlx5_ib_gsi_qp *gsi;
620 	struct work_struct pkey_change_work;
621 };
622 
623 struct mlx5_ib_resources {
624 	struct ib_cq	*c0;
625 	struct ib_xrcd	*x0;
626 	struct ib_xrcd	*x1;
627 	struct ib_pd	*p0;
628 	struct ib_srq	*s0;
629 	struct ib_srq	*s1;
630 	struct mlx5_ib_port_resources ports[2];
631 	/* Protects changes to the port resources */
632 	struct mutex	mutex;
633 };
634 
635 struct mlx5_ib_port {
636 	u16 q_cnt_id;
637 };
638 
639 struct mlx5_roce {
640 	/* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL
641 	 * netdev pointer
642 	 */
643 	rwlock_t		netdev_lock;
644 	if_t netdev;
645 	struct notifier_block	nb;
646 	atomic_t		next_port;
647 };
648 
649 #define	MLX5_IB_STATS_COUNT(a,...) a
650 #define	MLX5_IB_STATS_VAR(a,b,c,...) b c;
651 #define	MLX5_IB_STATS_DESC(a,b,c,d,e,...) d, e,
652 
653 #define	MLX5_IB_CONG_PARAMS(m) \
654   /* ECN RP */ \
655   m(+1, u64, rp_clamp_tgt_rate, "rp_clamp_tgt_rate", "If set, whenever a CNP is processed, the target rate is updated to be the current rate") \
656   m(+1, u64, rp_clamp_tgt_rate_ati, "rp_clamp_tgt_rate_ati", "If set, when receiving a CNP, the target rate should be updated if the transission rate was increased due to the timer, and not only due to the byte counter") \
657   m(+1, u64, rp_time_reset, "rp_time_reset", "Time in microseconds between rate increases if no CNPs are received") \
658   m(+1, u64, rp_byte_reset, "rp_byte_reset", "Transmitted data in bytes between rate increases if no CNP's are received. A value of zero means disabled.") \
659   m(+1, u64, rp_threshold, "rp_threshold", "The number of times rpByteStage or rpTimeStage can count before the RP rate control state machine advances states") \
660   m(+1, u64, rp_ai_rate, "rp_ai_rate", "The rate, in Mbits per second, used to increase rpTargetRate in the active increase state") \
661   m(+1, u64, rp_hai_rate, "rp_hai_rate", "The rate, in Mbits per second, used to increase rpTargetRate in the hyper increase state") \
662   m(+1, u64, rp_min_dec_fac, "rp_min_dec_fac", "The minimum factor by which the current transmit rate can be changed when processing a CNP. Value is given as a percentage, [1 .. 100]") \
663   m(+1, u64, rp_min_rate, "rp_min_rate", "The minimum value, in Mbps per second, for rate to limit") \
664   m(+1, u64, rp_rate_to_set_on_first_cnp, "rp_rate_to_set_on_first_cnp", "The rate that is set for the flow when a rate limiter is allocated to it upon first CNP received, in Mbps. A value of zero means use full port speed") \
665   m(+1, u64, rp_dce_tcp_g, "rp_dce_tcp_g", "Used to update the congestion estimator, alpha, once every dce_tcp_rtt once every dce_tcp_rtt microseconds") \
666   m(+1, u64, rp_dce_tcp_rtt, "rp_dce_tcp_rtt", "The time between updates of the aolpha value, in microseconds") \
667   m(+1, u64, rp_rate_reduce_monitor_period, "rp_rate_reduce_monitor_period", "The minimum time between two consecutive rate reductions for a single flow") \
668   m(+1, u64, rp_initial_alpha_value, "rp_initial_alpha_value", "The initial value of alpha to use when receiving the first CNP for a flow") \
669   m(+1, u64, rp_gd, "rp_gd", "If a CNP is received, the flow rate is reduced at the beginning of the next rate_reduce_monitor_period interval") \
670   /* ECN NP */ \
671   m(+1, u64, np_cnp_dscp, "np_cnp_dscp", "The DiffServ Code Point of the generated CNP for this port") \
672   m(+1, u64, np_cnp_prio_mode, "np_cnp_prio_mode", "The 802.1p priority value of the generated CNP for this port") \
673   m(+1, u64, np_cnp_prio, "np_cnp_prio", "The 802.1p priority value of the generated CNP for this port")
674 
675 #define	MLX5_IB_CONG_PARAMS_NUM (0 MLX5_IB_CONG_PARAMS(MLX5_IB_STATS_COUNT))
676 
677 #define	MLX5_IB_CONG_STATS(m) \
678   m(+1, u64, syndrome, "syndrome", "Syndrome number") \
679   m(+1, u64, rp_cur_flows, "rp_cur_flows", "Number of flows limited") \
680   m(+1, u64, sum_flows, "sum_flows", "Sum of the number of flows limited over time") \
681   m(+1, u64, rp_cnp_ignored, "rp_cnp_ignored", "Number of CNPs and CNMs ignored") \
682   m(+1, u64, rp_cnp_handled, "rp_cnp_handled", "Number of CNPs and CNMs successfully handled") \
683   m(+1, u64, time_stamp, "time_stamp", "Time stamp in microseconds") \
684   m(+1, u64, accumulators_period, "accumulators_period", "The value of X variable for accumulating counters") \
685   m(+1, u64, np_ecn_marked_roce_packets, "np_ecn_marked_roce_packets", "Number of ECN marked packets seen") \
686   m(+1, u64, np_cnp_sent, "np_cnp_sent", "Number of CNPs sent")
687 
688 #define	MLX5_IB_CONG_STATS_NUM (0 MLX5_IB_CONG_STATS(MLX5_IB_STATS_COUNT))
689 
690 #define	MLX5_IB_CONG_STATUS(m) \
691   /* ECN RP */ \
692   m(+1, u64, rp_0_enable, "rp_0_enable", "Enable reaction point, priority 0", MLX5_IB_RROCE_ECN_RP, 0, enable) \
693   m(+1, u64, rp_1_enable, "rp_1_enable", "Enable reaction point, priority 1", MLX5_IB_RROCE_ECN_RP, 1, enable) \
694   m(+1, u64, rp_2_enable, "rp_2_enable", "Enable reaction point, priority 2", MLX5_IB_RROCE_ECN_RP, 2, enable) \
695   m(+1, u64, rp_3_enable, "rp_3_enable", "Enable reaction point, priority 3", MLX5_IB_RROCE_ECN_RP, 3, enable) \
696   m(+1, u64, rp_4_enable, "rp_4_enable", "Enable reaction point, priority 4", MLX5_IB_RROCE_ECN_RP, 4, enable) \
697   m(+1, u64, rp_5_enable, "rp_5_enable", "Enable reaction point, priority 5", MLX5_IB_RROCE_ECN_RP, 5, enable) \
698   m(+1, u64, rp_6_enable, "rp_6_enable", "Enable reaction point, priority 6", MLX5_IB_RROCE_ECN_RP, 6, enable) \
699   m(+1, u64, rp_7_enable, "rp_7_enable", "Enable reaction point, priority 7", MLX5_IB_RROCE_ECN_RP, 7, enable) \
700   m(+1, u64, rp_8_enable, "rp_8_enable", "Enable reaction point, priority 8", MLX5_IB_RROCE_ECN_RP, 8, enable) \
701   m(+1, u64, rp_9_enable, "rp_9_enable", "Enable reaction point, priority 9", MLX5_IB_RROCE_ECN_RP, 9, enable) \
702   m(+1, u64, rp_10_enable, "rp_10_enable", "Enable reaction point, priority 10", MLX5_IB_RROCE_ECN_RP, 10, enable) \
703   m(+1, u64, rp_11_enable, "rp_11_enable", "Enable reaction point, priority 11", MLX5_IB_RROCE_ECN_RP, 11, enable) \
704   m(+1, u64, rp_12_enable, "rp_12_enable", "Enable reaction point, priority 12", MLX5_IB_RROCE_ECN_RP, 12, enable) \
705   m(+1, u64, rp_13_enable, "rp_13_enable", "Enable reaction point, priority 13", MLX5_IB_RROCE_ECN_RP, 13, enable) \
706   m(+1, u64, rp_14_enable, "rp_14_enable", "Enable reaction point, priority 14", MLX5_IB_RROCE_ECN_RP, 14, enable) \
707   m(+1, u64, rp_15_enable, "rp_15_enable", "Enable reaction point, priority 15", MLX5_IB_RROCE_ECN_RP, 15, enable) \
708   /* ECN NP */ \
709   m(+1, u64, np_0_enable, "np_0_enable", "Enable notification point, priority 0", MLX5_IB_RROCE_ECN_NP, 0, enable) \
710   m(+1, u64, np_1_enable, "np_1_enable", "Enable notification point, priority 1", MLX5_IB_RROCE_ECN_NP, 1, enable) \
711   m(+1, u64, np_2_enable, "np_2_enable", "Enable notification point, priority 2", MLX5_IB_RROCE_ECN_NP, 2, enable) \
712   m(+1, u64, np_3_enable, "np_3_enable", "Enable notification point, priority 3", MLX5_IB_RROCE_ECN_NP, 3, enable) \
713   m(+1, u64, np_4_enable, "np_4_enable", "Enable notification point, priority 4", MLX5_IB_RROCE_ECN_NP, 4, enable) \
714   m(+1, u64, np_5_enable, "np_5_enable", "Enable notification point, priority 5", MLX5_IB_RROCE_ECN_NP, 5, enable) \
715   m(+1, u64, np_6_enable, "np_6_enable", "Enable notification point, priority 6", MLX5_IB_RROCE_ECN_NP, 6, enable) \
716   m(+1, u64, np_7_enable, "np_7_enable", "Enable notification point, priority 7", MLX5_IB_RROCE_ECN_NP, 7, enable) \
717   m(+1, u64, np_8_enable, "np_8_enable", "Enable notification point, priority 8", MLX5_IB_RROCE_ECN_NP, 8, enable) \
718   m(+1, u64, np_9_enable, "np_9_enable", "Enable notification point, priority 9", MLX5_IB_RROCE_ECN_NP, 9, enable) \
719   m(+1, u64, np_10_enable, "np_10_enable", "Enable notification point, priority 10", MLX5_IB_RROCE_ECN_NP, 10, enable) \
720   m(+1, u64, np_11_enable, "np_11_enable", "Enable notification point, priority 11", MLX5_IB_RROCE_ECN_NP, 11, enable) \
721   m(+1, u64, np_12_enable, "np_12_enable", "Enable notification point, priority 12", MLX5_IB_RROCE_ECN_NP, 12, enable) \
722   m(+1, u64, np_13_enable, "np_13_enable", "Enable notification point, priority 13", MLX5_IB_RROCE_ECN_NP, 13, enable) \
723   m(+1, u64, np_14_enable, "np_14_enable", "Enable notification point, priority 14", MLX5_IB_RROCE_ECN_NP, 14, enable) \
724   m(+1, u64, np_15_enable, "np_15_enable", "Enable notification point, priority 15", MLX5_IB_RROCE_ECN_NP, 15, enable) \
725 
726 #define	MLX5_IB_CONG_STATUS_NUM (0 MLX5_IB_CONG_STATUS(MLX5_IB_STATS_COUNT))
727 
728 struct mlx5_ib_congestion {
729 	struct sysctl_ctx_list ctx;
730 	struct sx lock;
731 	struct delayed_work dwork;
732 	union {
733 		u64	arg[1];
734 		struct {
735 			MLX5_IB_CONG_PARAMS(MLX5_IB_STATS_VAR)
736 			MLX5_IB_CONG_STATS(MLX5_IB_STATS_VAR)
737 			MLX5_IB_CONG_STATUS(MLX5_IB_STATS_VAR)
738 		};
739 	};
740 };
741 
742 struct mlx5_devx_event_table {
743 	/* serialize updating the event_xa */
744 	struct mutex event_xa_lock;
745 	struct xarray event_xa;
746 };
747 
748 struct mlx5_ib_dev {
749 	struct ib_device		ib_dev;
750 	struct mlx5_core_dev		*mdev;
751 	struct mlx5_roce		roce;
752 	MLX5_DECLARE_DOORBELL_LOCK(uar_lock);
753 	int				num_ports;
754 	/* serialize update of capability mask
755 	 */
756 	struct mutex			cap_mask_mutex;
757 	u8				ib_active:1;
758 	u8				wc_support:1;
759 	struct umr_common		umrc;
760 	/* sync used page count stats
761 	 */
762 	struct mlx5_ib_resources	devr;
763 	struct mlx5_mr_cache		cache;
764 	struct timer_list		delay_timer;
765 	/* Prevents soft lock on massive reg MRs */
766 	struct mutex			slow_path_mutex;
767 	int				fill_delay;
768 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
769 	struct ib_odp_caps	odp_caps;
770 	/*
771 	 * Sleepable RCU that prevents destruction of MRs while they are still
772 	 * being used by a page fault handler.
773 	 */
774 	struct srcu_struct      mr_srcu;
775 #endif
776 	struct mlx5_ib_flow_db	flow_db;
777 	/* protect resources needed as part of reset flow */
778 	spinlock_t		reset_flow_resource_lock;
779 	struct list_head	qp_list;
780 	/* Array with num_ports elements */
781 	struct mlx5_ib_port	*port;
782 	struct mlx5_sq_bfreg	bfreg;
783 	struct mlx5_sq_bfreg	wc_bfreg;
784 	struct mlx5_sq_bfreg	fp_bfreg;
785 	struct mlx5_devx_event_table devx_event_table;
786 	struct mlx5_ib_congestion congestion;
787 
788 	struct mlx5_async_ctx	async_ctx;
789 
790 	/* protect the user_td */
791 	struct mutex		lb_mutex;
792 	u32			user_td;
793 };
794 
795 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
796 {
797 	return container_of(mcq, struct mlx5_ib_cq, mcq);
798 }
799 
800 static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
801 {
802 	return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd);
803 }
804 
805 static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev)
806 {
807 	return container_of(ibdev, struct mlx5_ib_dev, ib_dev);
808 }
809 
810 static inline struct mlx5_ib_dev *mlx5_udata_to_mdev(struct ib_udata *udata)
811 {
812 	struct mlx5_ib_ucontext *context = rdma_udata_to_drv_context(
813 		udata, struct mlx5_ib_ucontext, ibucontext);
814 
815 	return to_mdev(context->ibucontext.device);
816 }
817 
818 static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq)
819 {
820 	return container_of(ibcq, struct mlx5_ib_cq, ibcq);
821 }
822 
823 static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp)
824 {
825 	return container_of(mqp, struct mlx5_ib_qp_base, mqp)->container_mibqp;
826 }
827 
828 static inline struct mlx5_ib_rwq *to_mibrwq(struct mlx5_core_qp *core_qp)
829 {
830 	return container_of(core_qp, struct mlx5_ib_rwq, core_qp);
831 }
832 
833 static inline struct mlx5_ib_mr *to_mibmr(struct mlx5_core_mkey *mmkey)
834 {
835 	return container_of(mmkey, struct mlx5_ib_mr, mmkey);
836 }
837 
838 static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd)
839 {
840 	return container_of(ibpd, struct mlx5_ib_pd, ibpd);
841 }
842 
843 static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq)
844 {
845 	return container_of(ibsrq, struct mlx5_ib_srq, ibsrq);
846 }
847 
848 static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp)
849 {
850 	return container_of(ibqp, struct mlx5_ib_qp, ibqp);
851 }
852 
853 static inline struct mlx5_ib_rwq *to_mrwq(struct ib_wq *ibwq)
854 {
855 	return container_of(ibwq, struct mlx5_ib_rwq, ibwq);
856 }
857 
858 static inline struct mlx5_ib_rwq_ind_table *to_mrwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl)
859 {
860 	return container_of(ib_rwq_ind_tbl, struct mlx5_ib_rwq_ind_table, ib_rwq_ind_tbl);
861 }
862 
863 static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
864 {
865 	return container_of(msrq, struct mlx5_ib_srq, msrq);
866 }
867 
868 static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
869 {
870 	return container_of(ibmr, struct mlx5_ib_mr, ibmr);
871 }
872 
873 static inline struct mlx5_ib_mw *to_mmw(struct ib_mw *ibmw)
874 {
875 	return container_of(ibmw, struct mlx5_ib_mw, ibmw);
876 }
877 
878 struct mlx5_ib_ah {
879 	struct ib_ah		ibah;
880 	struct mlx5_av		av;
881 };
882 
883 static inline struct mlx5_ib_ah *to_mah(struct ib_ah *ibah)
884 {
885 	return container_of(ibah, struct mlx5_ib_ah, ibah);
886 }
887 
888 static inline struct mlx5_user_mmap_entry *
889 to_mmmap(struct rdma_user_mmap_entry *rdma_entry)
890 {
891 	return container_of(rdma_entry,
892 		struct mlx5_user_mmap_entry, rdma_entry);
893 }
894 
895 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
896 			struct mlx5_db *db);
897 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);
898 void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
899 void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
900 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index);
901 int mlx5_MAD_IFC(struct mlx5_ib_dev *dev, int ignore_mkey, int ignore_bkey,
902 		 u8 port, const struct ib_wc *in_wc, const struct ib_grh *in_grh,
903 		 const void *in_mad, void *response_mad);
904 int mlx5_ib_create_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr, u32 flags,
905 				struct ib_udata *udata);
906 int mlx5_ib_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr);
907 void mlx5_ib_destroy_ah(struct ib_ah *ah, u32 flags);
908 int mlx5_ib_create_srq(struct ib_srq *srq, struct ib_srq_init_attr *init_attr,
909 		       struct ib_udata *udata);
910 int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
911 		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
912 int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr);
913 void mlx5_ib_destroy_srq(struct ib_srq *srq, struct ib_udata *udata);
914 int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
915 			  const struct ib_recv_wr **bad_wr);
916 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
917 				struct ib_qp_init_attr *init_attr,
918 				struct ib_udata *udata);
919 int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
920 		      int attr_mask, struct ib_udata *udata);
921 int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
922 		     struct ib_qp_init_attr *qp_init_attr);
923 int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata);
924 int mlx5_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
925 		      const struct ib_send_wr **bad_wr);
926 int mlx5_ib_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
927 		      const struct ib_recv_wr **bad_wr);
928 void *mlx5_get_send_wqe(struct mlx5_ib_qp *qp, int n);
929 int mlx5_ib_read_user_wqe(struct mlx5_ib_qp *qp, int send, int wqe_index,
930 			  void *buffer, u32 length,
931 			  struct mlx5_ib_qp_base *base);
932 int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
933 		      struct ib_udata *udata);
934 void mlx5_ib_destroy_cq(struct ib_cq *cq, struct ib_udata *udata);
935 int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
936 int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
937 int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
938 int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata);
939 struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc);
940 struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
941 				  u64 virt_addr, int access_flags,
942 				  struct ib_udata *udata);
943 struct ib_mw *mlx5_ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type,
944 			       struct ib_udata *udata);
945 int mlx5_ib_dealloc_mw(struct ib_mw *mw);
946 int mlx5_ib_update_mtt(struct mlx5_ib_mr *mr, u64 start_page_index,
947 		       int npages, int zap);
948 int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start,
949 			  u64 length, u64 virt_addr, int access_flags,
950 			  struct ib_pd *pd, struct ib_udata *udata);
951 int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata);
952 struct ib_mr *mlx5_ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
953 			       u32 max_num_sg, struct ib_udata *udata);
954 int mlx5_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents,
955 		      unsigned int *sg_offset);
956 int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
957 			const struct ib_wc *in_wc, const struct ib_grh *in_grh,
958 			const struct ib_mad_hdr *in, size_t in_mad_size,
959 			struct ib_mad_hdr *out, size_t *out_mad_size,
960 			u16 *out_mad_pkey_index);
961 struct ib_xrcd *mlx5_ib_alloc_xrcd(struct ib_device *ibdev,
962 				   struct ib_udata *udata);
963 int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
964 int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset);
965 int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, u8 port);
966 int mlx5_query_mad_ifc_smp_attr_node_info(struct ib_device *ibdev,
967 					  struct ib_smp *out_mad);
968 int mlx5_query_mad_ifc_system_image_guid(struct ib_device *ibdev,
969 					 __be64 *sys_image_guid);
970 int mlx5_query_mad_ifc_max_pkeys(struct ib_device *ibdev,
971 				 u16 *max_pkeys);
972 int mlx5_query_mad_ifc_vendor_id(struct ib_device *ibdev,
973 				 u32 *vendor_id);
974 int mlx5_query_mad_ifc_node_desc(struct mlx5_ib_dev *dev, char *node_desc);
975 int mlx5_query_mad_ifc_node_guid(struct mlx5_ib_dev *dev, __be64 *node_guid);
976 int mlx5_query_mad_ifc_pkey(struct ib_device *ibdev, u8 port, u16 index,
977 			    u16 *pkey);
978 int mlx5_query_mad_ifc_gids(struct ib_device *ibdev, u8 port, int index,
979 			    union ib_gid *gid);
980 int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u8 port,
981 			    struct ib_port_attr *props);
982 int mlx5_ib_query_port(struct ib_device *ibdev, u8 port,
983 		       struct ib_port_attr *props);
984 int mlx5_ib_init_fmr(struct mlx5_ib_dev *dev);
985 void mlx5_ib_cleanup_fmr(struct mlx5_ib_dev *dev);
986 void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr,
987 			unsigned long max_page_shift,
988 			int *count, int *shift,
989 			int *ncont, int *order);
990 void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
991 			    int page_shift, size_t offset, size_t num_pages,
992 			    __be64 *pas, int access_flags);
993 void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
994 			  int page_shift, __be64 *pas, int access_flags);
995 void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num);
996 int mlx5_ib_get_cqe_size(struct mlx5_ib_dev *dev, struct ib_cq *ibcq);
997 int mlx5_mr_cache_init(struct mlx5_ib_dev *dev);
998 int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev);
999 int mlx5_mr_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift);
1000 int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
1001 			    struct ib_mr_status *mr_status);
1002 struct ib_wq *mlx5_ib_create_wq(struct ib_pd *pd,
1003 				struct ib_wq_init_attr *init_attr,
1004 				struct ib_udata *udata);
1005 void mlx5_ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
1006 int mlx5_ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1007 		      u32 wq_attr_mask, struct ib_udata *udata);
1008 struct ib_rwq_ind_table *mlx5_ib_create_rwq_ind_table(struct ib_device *device,
1009 						      struct ib_rwq_ind_table_init_attr *init_attr,
1010 						      struct ib_udata *udata);
1011 int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
1012 
1013 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1014 extern struct workqueue_struct *mlx5_ib_page_fault_wq;
1015 
1016 void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev);
1017 void mlx5_ib_mr_pfault_handler(struct mlx5_ib_qp *qp,
1018 			       struct mlx5_ib_pfault *pfault);
1019 void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp);
1020 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);
1021 void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev);
1022 int __init mlx5_ib_odp_init(void);
1023 void mlx5_ib_odp_cleanup(void);
1024 void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp);
1025 void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp);
1026 void mlx5_ib_invalidate_range(struct ib_umem *umem, unsigned long start,
1027 			      unsigned long end);
1028 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1029 static inline void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev)
1030 {
1031 	return;
1032 }
1033 
1034 static inline void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp)		{}
1035 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; }
1036 static inline void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev)	{}
1037 static inline int mlx5_ib_odp_init(void) { return 0; }
1038 static inline void mlx5_ib_odp_cleanup(void)				{}
1039 static inline void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp) {}
1040 static inline void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp)  {}
1041 
1042 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1043 
1044 int mlx5_ib_get_vf_config(struct ib_device *device, int vf,
1045 			  u8 port, struct ifla_vf_info *info);
1046 int mlx5_ib_set_vf_link_state(struct ib_device *device, int vf,
1047 			      u8 port, int state);
1048 int mlx5_ib_get_vf_stats(struct ib_device *device, int vf,
1049 			 u8 port, struct ifla_vf_stats *stats);
1050 int mlx5_ib_set_vf_guid(struct ib_device *device, int vf, u8 port,
1051 			u64 guid, int type);
1052 
1053 __be16 mlx5_get_roce_udp_sport(struct mlx5_ib_dev *dev, u8 port_num,
1054 			       int index);
1055 int mlx5_get_roce_gid_type(struct mlx5_ib_dev *dev, u8 port_num,
1056 			   int index, enum ib_gid_type *gid_type);
1057 
1058 /* GSI QP helper functions */
1059 struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
1060 				    struct ib_qp_init_attr *init_attr);
1061 int mlx5_ib_gsi_destroy_qp(struct ib_qp *qp);
1062 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
1063 			  int attr_mask);
1064 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
1065 			 int qp_attr_mask,
1066 			 struct ib_qp_init_attr *qp_init_attr);
1067 int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
1068 			  const struct ib_send_wr **bad_wr);
1069 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, const struct ib_recv_wr *wr,
1070 			  const struct ib_recv_wr **bad_wr);
1071 void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi);
1072 
1073 int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc);
1074 
1075 void mlx5_ib_free_bfreg(struct mlx5_ib_dev *dev, struct mlx5_bfreg_info *bfregi,
1076 			int bfregn);
1077 
1078 #if 1 /* IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS) */
1079 int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user);
1080 void mlx5_ib_devx_destroy(struct mlx5_ib_dev *dev, u16 uid);
1081 void mlx5_ib_devx_init_event_table(struct mlx5_ib_dev *dev);
1082 void mlx5_ib_devx_cleanup_event_table(struct mlx5_ib_dev *dev);
1083 bool mlx5_ib_devx_is_flow_dest(void *obj, int *dest_id, int *dest_type);
1084 bool mlx5_ib_devx_is_flow_counter(void *obj, u32 offset, u32 *counter_id);
1085 #else
1086 static inline int
1087 mlx5_ib_devx_create(struct mlx5_ib_dev *dev,
1088 			   bool is_user) { return -EOPNOTSUPP; }
1089 static inline void mlx5_ib_devx_destroy(struct mlx5_ib_dev *dev, u16 uid) {}
1090 static inline void mlx5_ib_devx_init_event_table(struct mlx5_ib_dev *dev) {}
1091 static inline void mlx5_ib_devx_cleanup_event_table(struct mlx5_ib_dev *dev) {}
1092 static inline bool mlx5_ib_devx_is_flow_dest(void *obj, int *dest_id,
1093 					     int *dest_type)
1094 {
1095 	return false;
1096 }
1097 #endif
1098 
1099 static inline void init_query_mad(struct ib_smp *mad)
1100 {
1101 	mad->base_version  = 1;
1102 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
1103 	mad->class_version = 1;
1104 	mad->method	   = IB_MGMT_METHOD_GET;
1105 }
1106 
1107 static inline u8 convert_access(int acc)
1108 {
1109 	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MLX5_PERM_ATOMIC       : 0) |
1110 	       (acc & IB_ACCESS_REMOTE_WRITE  ? MLX5_PERM_REMOTE_WRITE : 0) |
1111 	       (acc & IB_ACCESS_REMOTE_READ   ? MLX5_PERM_REMOTE_READ  : 0) |
1112 	       (acc & IB_ACCESS_LOCAL_WRITE   ? MLX5_PERM_LOCAL_WRITE  : 0) |
1113 	       MLX5_PERM_LOCAL_READ;
1114 }
1115 
1116 static inline int is_qp1(enum ib_qp_type qp_type)
1117 {
1118 	return qp_type == MLX5_IB_QPT_HW_GSI;
1119 }
1120 
1121 #define MLX5_MAX_UMR_SHIFT 16
1122 #define MLX5_MAX_UMR_PAGES (1 << MLX5_MAX_UMR_SHIFT)
1123 
1124 static inline u32 check_cq_create_flags(u32 flags)
1125 {
1126 	/*
1127 	 * It returns non-zero value for unsupported CQ
1128 	 * create flags, otherwise it returns zero.
1129 	 */
1130 	return (flags & ~(IB_CQ_FLAGS_IGNORE_OVERRUN |
1131 			  IB_CQ_FLAGS_TIMESTAMP_COMPLETION));
1132 }
1133 
1134 static inline int verify_assign_uidx(u8 cqe_version, u32 cmd_uidx,
1135 				     u32 *user_index)
1136 {
1137 	if (cqe_version) {
1138 		if ((cmd_uidx == MLX5_IB_DEFAULT_UIDX) ||
1139 		    (cmd_uidx & ~MLX5_USER_ASSIGNED_UIDX_MASK))
1140 			return -EINVAL;
1141 		*user_index = cmd_uidx;
1142 	} else {
1143 		*user_index = MLX5_IB_DEFAULT_UIDX;
1144 	}
1145 
1146 	return 0;
1147 }
1148 
1149 static inline int get_qp_user_index(struct mlx5_ib_ucontext *ucontext,
1150 				    struct mlx5_ib_create_qp *ucmd,
1151 				    int inlen,
1152 				    u32 *user_index)
1153 {
1154 	u8 cqe_version = ucontext->cqe_version;
1155 
1156 	if (field_avail(struct mlx5_ib_create_qp, uidx, inlen) &&
1157 	    !cqe_version && (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1158 		return 0;
1159 
1160 	if (!!(field_avail(struct mlx5_ib_create_qp, uidx, inlen) !=
1161 	       !!cqe_version))
1162 		return -EINVAL;
1163 
1164 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1165 }
1166 
1167 static inline int get_srq_user_index(struct mlx5_ib_ucontext *ucontext,
1168 				     struct mlx5_ib_create_srq *ucmd,
1169 				     int inlen,
1170 				     u32 *user_index)
1171 {
1172 	u8 cqe_version = ucontext->cqe_version;
1173 
1174 	if (field_avail(struct mlx5_ib_create_srq, uidx, inlen) &&
1175 	    !cqe_version && (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1176 		return 0;
1177 
1178 	if (!!(field_avail(struct mlx5_ib_create_srq, uidx, inlen) !=
1179 	       !!cqe_version))
1180 		return -EINVAL;
1181 
1182 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1183 }
1184 
1185 void mlx5_ib_cleanup_congestion(struct mlx5_ib_dev *);
1186 int mlx5_ib_init_congestion(struct mlx5_ib_dev *);
1187 
1188 static inline int get_uars_per_sys_page(struct mlx5_ib_dev *dev, bool lib_support)
1189 {
1190 	return lib_support && MLX5_CAP_GEN(dev->mdev, uar_4k) ?
1191 				MLX5_UARS_IN_PAGE : 1;
1192 }
1193 
1194 static inline int get_num_static_uars(struct mlx5_ib_dev *dev,
1195 				      struct mlx5_bfreg_info *bfregi)
1196 {
1197 	return get_uars_per_sys_page(dev, bfregi->lib_uar_4k) * bfregi->num_static_sys_pages;
1198 }
1199 
1200 int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
1201 			struct mlx5_bfreg_info *bfregi, u32 bfregn,
1202 			bool dyn_bfreg);
1203 
1204 #endif /* MLX5_IB_H */
1205