xref: /freebsd/sys/dev/mlx5/mlx5_ib/mlx5_ib.h (revision 5c512b5a76e3dd6cb526c38d99a04d550759cfa4)
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 /*
70  * Reserved mmap command range used to encode rdma_user_mmap entry page
71  * offsets (e.g. dynamically allocated UARs).  Keeping these above the legacy
72  * MLX5_IB_MMAP_* commands ensures mlx5_ib_mmap() routes them to the
73  * rdma_user_mmap offset handler instead of the legacy bfreg uar_mmap() path.
74  */
75 enum {
76 	MLX5_IB_MMAP_OFFSET_START	= 9,
77 	MLX5_IB_MMAP_OFFSET_END		= 255,
78 };
79 
80 enum {
81 	MLX5_RES_SCAT_DATA32_CQE	= 0x1,
82 	MLX5_RES_SCAT_DATA64_CQE	= 0x2,
83 	MLX5_REQ_SCAT_DATA32_CQE	= 0x11,
84 	MLX5_REQ_SCAT_DATA64_CQE	= 0x22,
85 };
86 
87 enum mlx5_ib_latency_class {
88 	MLX5_IB_LATENCY_CLASS_LOW,
89 	MLX5_IB_LATENCY_CLASS_MEDIUM,
90 	MLX5_IB_LATENCY_CLASS_HIGH,
91 	MLX5_IB_LATENCY_CLASS_FAST_PATH
92 };
93 
94 enum mlx5_ib_mad_ifc_flags {
95 	MLX5_MAD_IFC_IGNORE_MKEY	= 1,
96 	MLX5_MAD_IFC_IGNORE_BKEY	= 2,
97 	MLX5_MAD_IFC_NET_VIEW		= 4,
98 };
99 
100 enum {
101 	MLX5_CROSS_CHANNEL_BFREG         = 0,
102 };
103 
104 enum {
105 	MLX5_CQE_VERSION_V0,
106 	MLX5_CQE_VERSION_V1,
107 };
108 
109 enum {
110 	MLX5_IB_INVALID_UAR_INDEX	= BIT(31),
111 	MLX5_IB_INVALID_BFREG		= BIT(31),
112 };
113 
114 enum mlx5_ib_mmap_type {
115 	MLX5_IB_MMAP_TYPE_MEMIC = 1,
116 	MLX5_IB_MMAP_TYPE_VAR = 2,
117 	MLX5_IB_MMAP_TYPE_UAR_WC = 3,
118 	MLX5_IB_MMAP_TYPE_UAR_NC = 4,
119 };
120 
121 struct mlx5_bfreg_info {
122 	u32 *sys_pages;
123 	int num_low_latency_bfregs;
124 	unsigned int *count;
125 
126 	/*
127 	 * protect bfreg allocation data structs
128 	 */
129 	struct mutex lock;
130 	u32 ver;
131 	u8 lib_uar_4k : 1;
132 	u8 lib_uar_dyn : 1;
133 	u32 num_sys_pages;
134 	u32 num_static_sys_pages;
135 	u32 total_num_bfregs;
136 	u32 num_dyn_bfregs;
137 };
138 
139 struct mlx5_ib_ucontext {
140 	struct ib_ucontext	ibucontext;
141 	struct list_head	db_page_list;
142 
143 	/* protect doorbell record alloc/free
144 	 */
145 	struct mutex		db_page_mutex;
146 	struct mlx5_bfreg_info	bfregi;
147 	u8			cqe_version;
148 	/* Transport Domain number */
149 	u32			tdn;
150 
151 	u64			lib_caps;
152 	u16			devx_uid;
153 };
154 
to_mucontext(struct ib_ucontext * ibucontext)155 static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
156 {
157 	return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext);
158 }
159 
160 struct mlx5_ib_pd {
161 	struct ib_pd		ibpd;
162 	u32			pdn;
163 	u16			uid;
164 };
165 
166 #define MLX5_IB_FLOW_MCAST_PRIO		(MLX5_BY_PASS_NUM_PRIOS - 1)
167 #define MLX5_IB_FLOW_LAST_PRIO		(MLX5_BY_PASS_NUM_REGULAR_PRIOS - 1)
168 #if (MLX5_IB_FLOW_LAST_PRIO <= 0)
169 #error "Invalid number of bypass priorities"
170 #endif
171 #define MLX5_IB_FLOW_LEFTOVERS_PRIO	(MLX5_IB_FLOW_MCAST_PRIO + 1)
172 
173 #define MLX5_IB_NUM_FLOW_FT		(MLX5_IB_FLOW_LEFTOVERS_PRIO + 1)
174 #define MLX5_IB_NUM_SNIFFER_FTS		2
175 struct mlx5_ib_flow_prio {
176 	struct mlx5_flow_table		*flow_table;
177 	unsigned int			refcount;
178 };
179 
180 struct mlx5_ib_flow_handler {
181 	struct list_head		list;
182 	struct ib_flow			ibflow;
183 	struct mlx5_ib_flow_prio	*prio;
184 	struct mlx5_flow_handle	*rule;
185 };
186 
187 struct mlx5_ib_flow_db {
188 	struct mlx5_ib_flow_prio	prios[MLX5_IB_NUM_FLOW_FT];
189 	struct mlx5_ib_flow_prio	sniffer[MLX5_IB_NUM_SNIFFER_FTS];
190 	struct mlx5_flow_table		*lag_demux_ft;
191 	/* Protect flow steering bypass flow tables
192 	 * when add/del flow rules.
193 	 * only single add/removal of flow steering rule could be done
194 	 * simultaneously.
195 	 */
196 	struct mutex			lock;
197 };
198 
199 /* Use macros here so that don't have to duplicate
200  * enum ib_send_flags and enum ib_qp_type for low-level driver
201  */
202 
203 #define MLX5_IB_SEND_UMR_UNREG	IB_SEND_RESERVED_START
204 #define MLX5_IB_SEND_UMR_FAIL_IF_FREE (IB_SEND_RESERVED_START << 1)
205 #define MLX5_IB_SEND_UMR_UPDATE_MTT (IB_SEND_RESERVED_START << 2)
206 
207 #define MLX5_IB_SEND_UMR_UPDATE_TRANSLATION	(IB_SEND_RESERVED_START << 3)
208 #define MLX5_IB_SEND_UMR_UPDATE_PD		(IB_SEND_RESERVED_START << 4)
209 #define MLX5_IB_SEND_UMR_UPDATE_ACCESS		IB_SEND_RESERVED_END
210 
211 #define MLX5_IB_QPT_REG_UMR	IB_QPT_RESERVED1
212 /*
213  * IB_QPT_GSI creates the software wrapper around GSI, and MLX5_IB_QPT_HW_GSI
214  * creates the actual hardware QP.
215  */
216 #define MLX5_IB_QPT_HW_GSI	IB_QPT_RESERVED2
217 #define MLX5_IB_QPT_DCI		IB_QPT_RESERVED3
218 #define MLX5_IB_QPT_DCT		IB_QPT_RESERVED4
219 #define MLX5_IB_WR_UMR		IB_WR_RESERVED1
220 
221 /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags.
222  *
223  * These flags are intended for internal use by the mlx5_ib driver, and they
224  * rely on the range reserved for that use in the ib_qp_create_flags enum.
225  */
226 #define MLX5_IB_QP_CREATE_SQPN_QP1	IB_QP_CREATE_RESERVED_START
227 #define MLX5_IB_QP_CREATE_WC_TEST	(IB_QP_CREATE_RESERVED_START << 1)
228 
229 struct wr_list {
230 	u16	opcode;
231 	u16	next;
232 };
233 
234 enum mlx5_ib_wq_flags {
235 	MLX5_IB_WQ_FLAGS_STRIDING_RQ = 0x2,
236 };
237 
238 enum mlx5_ib_rq_flags {
239 	MLX5_IB_RQ_CVLAN_STRIPPING  = 1 << 0,
240 };
241 
242 struct mlx5_ib_wq {
243 	u64		       *wrid;
244 	u32		       *wr_data;
245 	struct wr_list	       *w_list;
246 	unsigned	       *wqe_head;
247 	u16		        unsig_count;
248 
249 	/* serialize post to the work queue
250 	 */
251 	spinlock_t		lock;
252 	int			wqe_cnt;
253 	int			max_post;
254 	int			max_gs;
255 	int			offset;
256 	int			wqe_shift;
257 	unsigned		head;
258 	unsigned		tail;
259 	u16			cur_post;
260 	u16			last_poll;
261 	void		       *qend;
262 };
263 
264 #define MLX5_MIN_SINGLE_WQE_LOG_NUM_STRIDES 9
265 #define MLX5_MAX_SINGLE_WQE_LOG_NUM_STRIDES 16
266 #define MLX5_MIN_SINGLE_STRIDE_LOG_NUM_BYTES 6
267 #define MLX5_MAX_SINGLE_STRIDE_LOG_NUM_BYTES 13
268 
269 struct mlx5_ib_rwq {
270 	struct ib_wq		ibwq;
271 	struct mlx5_core_qp	core_qp;
272 	u32			rq_num_pas;
273 	u32			log_rq_stride;
274 	u32			log_rq_size;
275 	u32			rq_page_offset;
276 	u32			log_page_size;
277 	u32			log_num_strides;
278 	u32			two_byte_shift_en;
279 	u32			single_stride_log_num_of_bytes;
280 	struct ib_umem		*umem;
281 	size_t			buf_size;
282 	unsigned int		page_shift;
283 	int			create_type;
284 	struct mlx5_db		db;
285 	u32			user_index;
286 	u32			wqe_count;
287 	u32			wqe_shift;
288 	int			wq_sig;
289 	u32			create_flags; /* Use enum mlx5_ib_wq_flags */
290 };
291 
292 enum {
293 	MLX5_QP_USER,
294 	MLX5_QP_KERNEL,
295 	MLX5_QP_EMPTY
296 };
297 
298 enum {
299 	MLX5_WQ_USER,
300 	MLX5_WQ_KERNEL
301 };
302 
303 struct mlx5_ib_rwq_ind_table {
304 	struct ib_rwq_ind_table ib_rwq_ind_tbl;
305 	u32			rqtn;
306 	u16			uid;
307 };
308 
309 /*
310  * Connect-IB can trigger up to four concurrent pagefaults
311  * per-QP.
312  */
313 enum mlx5_ib_pagefault_context {
314 	MLX5_IB_PAGEFAULT_RESPONDER_READ,
315 	MLX5_IB_PAGEFAULT_REQUESTOR_READ,
316 	MLX5_IB_PAGEFAULT_RESPONDER_WRITE,
317 	MLX5_IB_PAGEFAULT_REQUESTOR_WRITE,
318 	MLX5_IB_PAGEFAULT_CONTEXTS
319 };
320 
321 static inline enum mlx5_ib_pagefault_context
mlx5_ib_get_pagefault_context(struct mlx5_pagefault * pagefault)322 	mlx5_ib_get_pagefault_context(struct mlx5_pagefault *pagefault)
323 {
324 	return pagefault->flags & (MLX5_PFAULT_REQUESTOR | MLX5_PFAULT_WRITE);
325 }
326 
327 struct mlx5_ib_pfault {
328 	struct work_struct	work;
329 	struct mlx5_pagefault	mpfault;
330 };
331 
332 struct mlx5_ib_ubuffer {
333 	struct ib_umem	       *umem;
334 	int			buf_size;
335 	u64			buf_addr;
336 };
337 
338 struct mlx5_ib_qp_base {
339 	struct mlx5_ib_qp	*container_mibqp;
340 	struct mlx5_core_qp	mqp;
341 	struct mlx5_ib_ubuffer	ubuffer;
342 };
343 
344 struct mlx5_ib_qp_trans {
345 	struct mlx5_ib_qp_base	base;
346 	u16			xrcdn;
347 	u8			alt_port;
348 	u8			atomic_rd_en;
349 	u8			resp_depth;
350 };
351 
352 struct mlx5_ib_rss_qp {
353 	u32	tirn;
354 };
355 
356 struct mlx5_ib_rq {
357 	struct mlx5_ib_qp_base base;
358 	struct mlx5_ib_wq	*rq;
359 	struct mlx5_ib_ubuffer	ubuffer;
360 	struct mlx5_db		*doorbell;
361 	u32			tirn;
362 	u8			state;
363 	u32			flags;
364 };
365 
366 struct mlx5_ib_sq {
367 	struct mlx5_ib_qp_base base;
368 	struct mlx5_ib_wq	*sq;
369 	struct mlx5_ib_ubuffer  ubuffer;
370 	struct mlx5_db		*doorbell;
371 	u32			tisn;
372 	u8			state;
373 };
374 
375 struct mlx5_ib_raw_packet_qp {
376 	struct mlx5_ib_sq sq;
377 	struct mlx5_ib_rq rq;
378 };
379 
380 struct mlx5_bf {
381 	int			buf_size;
382 	unsigned long		offset;
383 	struct mlx5_sq_bfreg   *bfreg;
384 	spinlock_t		lock32;
385 };
386 
387 struct mlx5_ib_dct {
388 	struct mlx5_core_dct    mdct;
389 	u32                     *in;
390 };
391 
392 struct mlx5_ib_qp {
393 	struct ib_qp		ibqp;
394 	union {
395 		struct mlx5_ib_qp_trans trans_qp;
396 		struct mlx5_ib_raw_packet_qp raw_packet_qp;
397 		struct mlx5_ib_rss_qp rss_qp;
398 		struct mlx5_ib_dct dct;
399 	};
400 	struct mlx5_buf		buf;
401 
402 	struct mlx5_db		db;
403 	struct mlx5_ib_wq	rq;
404 
405 	u8			sq_signal_bits;
406 	u8			fm_cache;
407 	struct mlx5_ib_wq	sq;
408 
409 	/* serialize qp state modifications
410 	 */
411 	struct mutex		mutex;
412 	u32			flags;
413 	u8			port;
414 	u8			state;
415 	int			wq_sig;
416 	int			scat_cqe;
417 	int			max_inline_data;
418 	struct mlx5_bf	        bf;
419 	int			has_rq;
420 
421 	/* only for user space QPs. For kernel
422 	 * we have it from the bf object
423 	 */
424 	int			bfregn;
425 
426 	int			create_type;
427 
428 	/* Store signature errors */
429 	bool			signature_en;
430 
431 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
432 	/*
433 	 * A flag that is true for QP's that are in a state that doesn't
434 	 * allow page faults, and shouldn't schedule any more faults.
435 	 */
436 	int                     disable_page_faults;
437 	/*
438 	 * The disable_page_faults_lock protects a QP's disable_page_faults
439 	 * field, allowing for a thread to atomically check whether the QP
440 	 * allows page faults, and if so schedule a page fault.
441 	 */
442 	spinlock_t              disable_page_faults_lock;
443 	struct mlx5_ib_pfault	pagefaults[MLX5_IB_PAGEFAULT_CONTEXTS];
444 #endif
445 	struct list_head	qps_list;
446 	struct list_head	cq_recv_list;
447 	struct list_head	cq_send_list;
448 	bool			tunnel_offload_en;
449 };
450 
451 struct mlx5_ib_cq_buf {
452 	struct mlx5_buf		buf;
453 	struct ib_umem		*umem;
454 	int			cqe_size;
455 	int			nent;
456 };
457 
458 enum mlx5_ib_qp_flags {
459 	MLX5_IB_QP_LSO                          = IB_QP_CREATE_IPOIB_UD_LSO,
460 	MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK     = IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
461 	MLX5_IB_QP_CROSS_CHANNEL            = IB_QP_CREATE_CROSS_CHANNEL,
462 	MLX5_IB_QP_MANAGED_SEND             = IB_QP_CREATE_MANAGED_SEND,
463 	MLX5_IB_QP_MANAGED_RECV             = IB_QP_CREATE_MANAGED_RECV,
464 	MLX5_IB_QP_SIGNATURE_HANDLING           = 1 << 5,
465 	/* QP uses 1 as its source QP number */
466 	MLX5_IB_QP_SQPN_QP1			= 1 << 6,
467 	MLX5_IB_QP_CAP_SCATTER_FCS		= 1 << 7,
468 	MLX5_IB_QP_RSS				= 1 << 8,
469 	MLX5_IB_QP_CVLAN_STRIPPING		= 1 << 9,
470 	MLX5_IB_QP_UNDERLAY			= 1 << 10,
471 };
472 
473 struct mlx5_umr_wr {
474 	struct ib_send_wr		wr;
475 	union {
476 		u64			virt_addr;
477 		u64			offset;
478 	} target;
479 	struct ib_pd		       *pd;
480 	unsigned int			page_shift;
481 	unsigned int			npages;
482 	u64				length;
483 	int				access_flags;
484 	u32				mkey;
485 };
486 
umr_wr(const struct ib_send_wr * wr)487 static inline const struct mlx5_umr_wr *umr_wr(const struct ib_send_wr *wr)
488 {
489 	return container_of(wr, struct mlx5_umr_wr, wr);
490 }
491 
492 struct mlx5_shared_mr_info {
493 	int mr_id;
494 	struct ib_umem		*umem;
495 };
496 
497 struct mlx5_ib_cq {
498 	struct ib_cq		ibcq;
499 	struct mlx5_core_cq	mcq;
500 	struct mlx5_ib_cq_buf	buf;
501 	struct mlx5_db		db;
502 
503 	/* serialize access to the CQ
504 	 */
505 	spinlock_t		lock;
506 
507 	/* protect resize cq
508 	 */
509 	struct mutex		resize_mutex;
510 	struct mlx5_ib_cq_buf  *resize_buf;
511 	struct ib_umem	       *resize_umem;
512 	int			cqe_size;
513 	struct list_head	list_send_qp;
514 	struct list_head	list_recv_qp;
515 	u32			create_flags;
516 	struct list_head	wc_list;
517 	enum ib_cq_notify_flags notify_flags;
518 	struct work_struct	notify_work;
519 };
520 
521 struct mlx5_ib_wc {
522 	struct ib_wc wc;
523 	struct list_head list;
524 };
525 
526 struct mlx5_ib_srq {
527 	struct ib_srq		ibsrq;
528 	struct mlx5_core_srq	msrq;
529 	struct mlx5_buf		buf;
530 	struct mlx5_db		db;
531 	u64		       *wrid;
532 	/* protect SRQ hanlding
533 	 */
534 	spinlock_t		lock;
535 	int			head;
536 	int			tail;
537 	u16			wqe_ctr;
538 	struct ib_umem	       *umem;
539 	/* serialize arming a SRQ
540 	 */
541 	struct mutex		mutex;
542 	int			wq_sig;
543 };
544 
545 struct mlx5_ib_xrcd {
546 	struct ib_xrcd		ibxrcd;
547 	u32			xrcdn;
548 };
549 
550 enum mlx5_ib_mtt_access_flags {
551 	MLX5_IB_MTT_READ  = (1 << 0),
552 	MLX5_IB_MTT_WRITE = (1 << 1),
553 };
554 
555 struct mlx5_user_mmap_entry {
556 	struct rdma_user_mmap_entry rdma_entry;
557 	u8 mmap_flag;
558 	u64 address;
559 	u32 page_idx;
560 };
561 
562 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
563 
564 struct mlx5_ib_mr {
565 	struct ib_mr		ibmr;
566 	void			*descs;
567 	dma_addr_t		desc_map;
568 	int			ndescs;
569 	int			max_descs;
570 	int			desc_size;
571 	int			access_mode;
572 	struct mlx5_core_mkey	mmkey;
573 	struct ib_umem	       *umem;
574 	struct mlx5_shared_mr_info	*smr_info;
575 	struct list_head	list;
576 	int			order;
577 	int			umred;
578 	int			npages;
579 	struct mlx5_ib_dev     *dev;
580 	u32 out[MLX5_ST_SZ_DW(create_mkey_out)];
581 	struct mlx5_core_sig_ctx    *sig;
582 	int			live;
583 	void			*descs_alloc;
584 	int			access_flags; /* Needed for rereg MR */
585 	struct mlx5_async_work	cb_work;
586 };
587 
588 struct mlx5_ib_mw {
589 	struct ib_mw		ibmw;
590 	struct mlx5_core_mkey	mmkey;
591 };
592 
593 struct mlx5_ib_devx_mr {
594 	struct mlx5_core_mkey	mmkey;
595 	int			ndescs;
596 };
597 
598 struct mlx5_ib_umr_context {
599 	struct ib_cqe		cqe;
600 	enum ib_wc_status	status;
601 	struct completion	done;
602 };
603 
604 struct umr_common {
605 	struct ib_pd	*pd;
606 	struct ib_cq	*cq;
607 	struct ib_qp	*qp;
608 	/* control access to UMR QP
609 	 */
610 	struct semaphore	sem;
611 };
612 
613 enum {
614 	MLX5_FMR_INVALID,
615 	MLX5_FMR_VALID,
616 	MLX5_FMR_BUSY,
617 };
618 
619 struct mlx5_cache_ent {
620 	struct list_head	head;
621 	/* sync access to the cahce entry
622 	 */
623 	spinlock_t		lock;
624 
625 
626 	char                    name[4];
627 	u32                     order;
628 	u32			size;
629 	u32                     cur;
630 	u32                     miss;
631 	u32			limit;
632 
633 	struct mlx5_ib_dev     *dev;
634 	struct work_struct	work;
635 	struct delayed_work	dwork;
636 	int			pending;
637 };
638 
639 struct mlx5_mr_cache {
640 	struct workqueue_struct *wq;
641 	struct mlx5_cache_ent	ent[MAX_MR_CACHE_ENTRIES];
642 	int			stopped;
643 	unsigned long		last_add;
644 };
645 
646 struct mlx5_ib_gsi_qp;
647 
648 struct mlx5_ib_port_resources {
649 	struct mlx5_ib_resources *devr;
650 	struct mlx5_ib_gsi_qp *gsi;
651 	struct work_struct pkey_change_work;
652 };
653 
654 struct mlx5_ib_resources {
655 	struct ib_cq	*c0;
656 	struct ib_xrcd	*x0;
657 	struct ib_xrcd	*x1;
658 	struct ib_pd	*p0;
659 	struct ib_srq	*s0;
660 	struct ib_srq	*s1;
661 	struct mlx5_ib_port_resources ports[2];
662 	/* Protects changes to the port resources */
663 	struct mutex	mutex;
664 };
665 
666 struct mlx5_ib_port {
667 	u16 q_cnt_id;
668 };
669 
670 struct mlx5_roce {
671 	/* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL
672 	 * netdev pointer
673 	 */
674 	rwlock_t		netdev_lock;
675 	if_t netdev;
676 	struct notifier_block	nb;
677 	atomic_t		next_port;
678 };
679 
680 #define	MLX5_IB_STATS_COUNT(a,...) a
681 #define	MLX5_IB_STATS_VAR(a,b,c,...) b c;
682 #define	MLX5_IB_STATS_DESC(a,b,c,d,e,...) d, e,
683 
684 #define	MLX5_IB_CONG_PARAMS(m) \
685   /* ECN RP */ \
686   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") \
687   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") \
688   m(+1, u64, rp_time_reset, "rp_time_reset", "Time in microseconds between rate increases if no CNPs are received") \
689   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.") \
690   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") \
691   m(+1, u64, rp_ai_rate, "rp_ai_rate", "The rate, in Mbits per second, used to increase rpTargetRate in the active increase state") \
692   m(+1, u64, rp_hai_rate, "rp_hai_rate", "The rate, in Mbits per second, used to increase rpTargetRate in the hyper increase state") \
693   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]") \
694   m(+1, u64, rp_min_rate, "rp_min_rate", "The minimum value, in Mbps per second, for rate to limit") \
695   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") \
696   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") \
697   m(+1, u64, rp_dce_tcp_rtt, "rp_dce_tcp_rtt", "The time between updates of the aolpha value, in microseconds") \
698   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") \
699   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") \
700   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") \
701   /* ECN NP */ \
702   m(+1, u64, np_cnp_dscp, "np_cnp_dscp", "The DiffServ Code Point of the generated CNP for this port") \
703   m(+1, u64, np_cnp_prio_mode, "np_cnp_prio_mode", "The 802.1p priority value of the generated CNP for this port") \
704   m(+1, u64, np_cnp_prio, "np_cnp_prio", "The 802.1p priority value of the generated CNP for this port")
705 
706 #define	MLX5_IB_CONG_PARAMS_NUM (0 MLX5_IB_CONG_PARAMS(MLX5_IB_STATS_COUNT))
707 
708 #define	MLX5_IB_CONG_STATS(m) \
709   m(+1, u64, syndrome, "syndrome", "Syndrome number") \
710   m(+1, u64, rp_cur_flows, "rp_cur_flows", "Number of flows limited") \
711   m(+1, u64, sum_flows, "sum_flows", "Sum of the number of flows limited over time") \
712   m(+1, u64, rp_cnp_ignored, "rp_cnp_ignored", "Number of CNPs and CNMs ignored") \
713   m(+1, u64, rp_cnp_handled, "rp_cnp_handled", "Number of CNPs and CNMs successfully handled") \
714   m(+1, u64, time_stamp, "time_stamp", "Time stamp in microseconds") \
715   m(+1, u64, accumulators_period, "accumulators_period", "The value of X variable for accumulating counters") \
716   m(+1, u64, np_ecn_marked_roce_packets, "np_ecn_marked_roce_packets", "Number of ECN marked packets seen") \
717   m(+1, u64, np_cnp_sent, "np_cnp_sent", "Number of CNPs sent")
718 
719 #define	MLX5_IB_CONG_STATS_NUM (0 MLX5_IB_CONG_STATS(MLX5_IB_STATS_COUNT))
720 
721 #define	MLX5_IB_CONG_STATUS(m) \
722   /* ECN RP */ \
723   m(+1, u64, rp_0_enable, "rp_0_enable", "Enable reaction point, priority 0", MLX5_IB_RROCE_ECN_RP, 0, enable) \
724   m(+1, u64, rp_1_enable, "rp_1_enable", "Enable reaction point, priority 1", MLX5_IB_RROCE_ECN_RP, 1, enable) \
725   m(+1, u64, rp_2_enable, "rp_2_enable", "Enable reaction point, priority 2", MLX5_IB_RROCE_ECN_RP, 2, enable) \
726   m(+1, u64, rp_3_enable, "rp_3_enable", "Enable reaction point, priority 3", MLX5_IB_RROCE_ECN_RP, 3, enable) \
727   m(+1, u64, rp_4_enable, "rp_4_enable", "Enable reaction point, priority 4", MLX5_IB_RROCE_ECN_RP, 4, enable) \
728   m(+1, u64, rp_5_enable, "rp_5_enable", "Enable reaction point, priority 5", MLX5_IB_RROCE_ECN_RP, 5, enable) \
729   m(+1, u64, rp_6_enable, "rp_6_enable", "Enable reaction point, priority 6", MLX5_IB_RROCE_ECN_RP, 6, enable) \
730   m(+1, u64, rp_7_enable, "rp_7_enable", "Enable reaction point, priority 7", MLX5_IB_RROCE_ECN_RP, 7, enable) \
731   m(+1, u64, rp_8_enable, "rp_8_enable", "Enable reaction point, priority 8", MLX5_IB_RROCE_ECN_RP, 8, enable) \
732   m(+1, u64, rp_9_enable, "rp_9_enable", "Enable reaction point, priority 9", MLX5_IB_RROCE_ECN_RP, 9, enable) \
733   m(+1, u64, rp_10_enable, "rp_10_enable", "Enable reaction point, priority 10", MLX5_IB_RROCE_ECN_RP, 10, enable) \
734   m(+1, u64, rp_11_enable, "rp_11_enable", "Enable reaction point, priority 11", MLX5_IB_RROCE_ECN_RP, 11, enable) \
735   m(+1, u64, rp_12_enable, "rp_12_enable", "Enable reaction point, priority 12", MLX5_IB_RROCE_ECN_RP, 12, enable) \
736   m(+1, u64, rp_13_enable, "rp_13_enable", "Enable reaction point, priority 13", MLX5_IB_RROCE_ECN_RP, 13, enable) \
737   m(+1, u64, rp_14_enable, "rp_14_enable", "Enable reaction point, priority 14", MLX5_IB_RROCE_ECN_RP, 14, enable) \
738   m(+1, u64, rp_15_enable, "rp_15_enable", "Enable reaction point, priority 15", MLX5_IB_RROCE_ECN_RP, 15, enable) \
739   /* ECN NP */ \
740   m(+1, u64, np_0_enable, "np_0_enable", "Enable notification point, priority 0", MLX5_IB_RROCE_ECN_NP, 0, enable) \
741   m(+1, u64, np_1_enable, "np_1_enable", "Enable notification point, priority 1", MLX5_IB_RROCE_ECN_NP, 1, enable) \
742   m(+1, u64, np_2_enable, "np_2_enable", "Enable notification point, priority 2", MLX5_IB_RROCE_ECN_NP, 2, enable) \
743   m(+1, u64, np_3_enable, "np_3_enable", "Enable notification point, priority 3", MLX5_IB_RROCE_ECN_NP, 3, enable) \
744   m(+1, u64, np_4_enable, "np_4_enable", "Enable notification point, priority 4", MLX5_IB_RROCE_ECN_NP, 4, enable) \
745   m(+1, u64, np_5_enable, "np_5_enable", "Enable notification point, priority 5", MLX5_IB_RROCE_ECN_NP, 5, enable) \
746   m(+1, u64, np_6_enable, "np_6_enable", "Enable notification point, priority 6", MLX5_IB_RROCE_ECN_NP, 6, enable) \
747   m(+1, u64, np_7_enable, "np_7_enable", "Enable notification point, priority 7", MLX5_IB_RROCE_ECN_NP, 7, enable) \
748   m(+1, u64, np_8_enable, "np_8_enable", "Enable notification point, priority 8", MLX5_IB_RROCE_ECN_NP, 8, enable) \
749   m(+1, u64, np_9_enable, "np_9_enable", "Enable notification point, priority 9", MLX5_IB_RROCE_ECN_NP, 9, enable) \
750   m(+1, u64, np_10_enable, "np_10_enable", "Enable notification point, priority 10", MLX5_IB_RROCE_ECN_NP, 10, enable) \
751   m(+1, u64, np_11_enable, "np_11_enable", "Enable notification point, priority 11", MLX5_IB_RROCE_ECN_NP, 11, enable) \
752   m(+1, u64, np_12_enable, "np_12_enable", "Enable notification point, priority 12", MLX5_IB_RROCE_ECN_NP, 12, enable) \
753   m(+1, u64, np_13_enable, "np_13_enable", "Enable notification point, priority 13", MLX5_IB_RROCE_ECN_NP, 13, enable) \
754   m(+1, u64, np_14_enable, "np_14_enable", "Enable notification point, priority 14", MLX5_IB_RROCE_ECN_NP, 14, enable) \
755   m(+1, u64, np_15_enable, "np_15_enable", "Enable notification point, priority 15", MLX5_IB_RROCE_ECN_NP, 15, enable) \
756 
757 #define	MLX5_IB_CONG_STATUS_NUM (0 MLX5_IB_CONG_STATUS(MLX5_IB_STATS_COUNT))
758 
759 struct mlx5_ib_congestion {
760 	struct sysctl_ctx_list ctx;
761 	struct sx lock;
762 	struct delayed_work dwork;
763 	union {
764 		u64	arg[1];
765 		struct {
766 			MLX5_IB_CONG_PARAMS(MLX5_IB_STATS_VAR)
767 			MLX5_IB_CONG_STATS(MLX5_IB_STATS_VAR)
768 			MLX5_IB_CONG_STATUS(MLX5_IB_STATS_VAR)
769 		};
770 	};
771 };
772 
773 struct mlx5_devx_event_table {
774 	/* serialize updating the event_xa */
775 	struct mutex event_xa_lock;
776 	struct xarray event_xa;
777 };
778 
779 struct mlx5_ib_dev {
780 	struct ib_device		ib_dev;
781 	struct mlx5_core_dev		*mdev;
782 	struct mlx5_roce		roce;
783 	MLX5_DECLARE_DOORBELL_LOCK(uar_lock);
784 	int				num_ports;
785 	/* serialize update of capability mask
786 	 */
787 	struct mutex			cap_mask_mutex;
788 	u8				ib_active:1;
789 	u8				wc_support:1;
790 	struct umr_common		umrc;
791 	/* sync used page count stats
792 	 */
793 	struct mlx5_ib_resources	devr;
794 	struct mlx5_mr_cache		cache;
795 	struct timer_list		delay_timer;
796 	/* Prevents soft lock on massive reg MRs */
797 	struct mutex			slow_path_mutex;
798 	int				fill_delay;
799 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
800 	struct ib_odp_caps	odp_caps;
801 	/*
802 	 * Sleepable RCU that prevents destruction of MRs while they are still
803 	 * being used by a page fault handler.
804 	 */
805 	struct srcu_struct      mr_srcu;
806 #endif
807 	struct mlx5_ib_flow_db	flow_db;
808 	/* protect resources needed as part of reset flow */
809 	spinlock_t		reset_flow_resource_lock;
810 	struct list_head	qp_list;
811 	/* Array with num_ports elements */
812 	struct mlx5_ib_port	*port;
813 	struct mlx5_sq_bfreg	bfreg;
814 	struct mlx5_sq_bfreg	wc_bfreg;
815 	struct mlx5_sq_bfreg	fp_bfreg;
816 	struct mlx5_devx_event_table devx_event_table;
817 	struct mlx5_ib_congestion congestion;
818 
819 	struct mlx5_async_ctx	async_ctx;
820 
821 	/* protect the user_td */
822 	struct mutex		lb_mutex;
823 	u32			user_td;
824 };
825 
to_mibcq(struct mlx5_core_cq * mcq)826 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
827 {
828 	return container_of(mcq, struct mlx5_ib_cq, mcq);
829 }
830 
to_mxrcd(struct ib_xrcd * ibxrcd)831 static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
832 {
833 	return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd);
834 }
835 
to_mdev(struct ib_device * ibdev)836 static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev)
837 {
838 	return container_of(ibdev, struct mlx5_ib_dev, ib_dev);
839 }
840 
mlx5_udata_to_mdev(struct ib_udata * udata)841 static inline struct mlx5_ib_dev *mlx5_udata_to_mdev(struct ib_udata *udata)
842 {
843 	struct mlx5_ib_ucontext *context = rdma_udata_to_drv_context(
844 		udata, struct mlx5_ib_ucontext, ibucontext);
845 
846 	return to_mdev(context->ibucontext.device);
847 }
848 
to_mcq(struct ib_cq * ibcq)849 static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq)
850 {
851 	return container_of(ibcq, struct mlx5_ib_cq, ibcq);
852 }
853 
to_mibqp(struct mlx5_core_qp * mqp)854 static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp)
855 {
856 	return container_of(mqp, struct mlx5_ib_qp_base, mqp)->container_mibqp;
857 }
858 
to_mibrwq(struct mlx5_core_qp * core_qp)859 static inline struct mlx5_ib_rwq *to_mibrwq(struct mlx5_core_qp *core_qp)
860 {
861 	return container_of(core_qp, struct mlx5_ib_rwq, core_qp);
862 }
863 
to_mibmr(struct mlx5_core_mkey * mmkey)864 static inline struct mlx5_ib_mr *to_mibmr(struct mlx5_core_mkey *mmkey)
865 {
866 	return container_of(mmkey, struct mlx5_ib_mr, mmkey);
867 }
868 
to_mpd(struct ib_pd * ibpd)869 static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd)
870 {
871 	return container_of(ibpd, struct mlx5_ib_pd, ibpd);
872 }
873 
to_msrq(struct ib_srq * ibsrq)874 static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq)
875 {
876 	return container_of(ibsrq, struct mlx5_ib_srq, ibsrq);
877 }
878 
to_mqp(struct ib_qp * ibqp)879 static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp)
880 {
881 	return container_of(ibqp, struct mlx5_ib_qp, ibqp);
882 }
883 
to_mrwq(struct ib_wq * ibwq)884 static inline struct mlx5_ib_rwq *to_mrwq(struct ib_wq *ibwq)
885 {
886 	return container_of(ibwq, struct mlx5_ib_rwq, ibwq);
887 }
888 
to_mrwq_ind_table(struct ib_rwq_ind_table * ib_rwq_ind_tbl)889 static inline struct mlx5_ib_rwq_ind_table *to_mrwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl)
890 {
891 	return container_of(ib_rwq_ind_tbl, struct mlx5_ib_rwq_ind_table, ib_rwq_ind_tbl);
892 }
893 
to_mibsrq(struct mlx5_core_srq * msrq)894 static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
895 {
896 	return container_of(msrq, struct mlx5_ib_srq, msrq);
897 }
898 
to_mmr(struct ib_mr * ibmr)899 static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
900 {
901 	return container_of(ibmr, struct mlx5_ib_mr, ibmr);
902 }
903 
to_mmw(struct ib_mw * ibmw)904 static inline struct mlx5_ib_mw *to_mmw(struct ib_mw *ibmw)
905 {
906 	return container_of(ibmw, struct mlx5_ib_mw, ibmw);
907 }
908 
909 struct mlx5_ib_ah {
910 	struct ib_ah		ibah;
911 	struct mlx5_av		av;
912 };
913 
to_mah(struct ib_ah * ibah)914 static inline struct mlx5_ib_ah *to_mah(struct ib_ah *ibah)
915 {
916 	return container_of(ibah, struct mlx5_ib_ah, ibah);
917 }
918 
919 static inline struct mlx5_user_mmap_entry *
to_mmmap(struct rdma_user_mmap_entry * rdma_entry)920 to_mmmap(struct rdma_user_mmap_entry *rdma_entry)
921 {
922 	return container_of(rdma_entry,
923 		struct mlx5_user_mmap_entry, rdma_entry);
924 }
925 
926 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
927 			struct mlx5_db *db);
928 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);
929 void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
930 void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
931 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index);
932 int mlx5_MAD_IFC(struct mlx5_ib_dev *dev, int ignore_mkey, int ignore_bkey,
933 		 u8 port, const struct ib_wc *in_wc, const struct ib_grh *in_grh,
934 		 const void *in_mad, void *response_mad);
935 int mlx5_ib_create_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr, u32 flags,
936 				struct ib_udata *udata);
937 int mlx5_ib_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr);
938 void mlx5_ib_destroy_ah(struct ib_ah *ah, u32 flags);
939 int mlx5_ib_create_srq(struct ib_srq *srq, struct ib_srq_init_attr *init_attr,
940 		       struct ib_udata *udata);
941 int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
942 		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
943 int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr);
944 void mlx5_ib_destroy_srq(struct ib_srq *srq, struct ib_udata *udata);
945 int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
946 			  const struct ib_recv_wr **bad_wr);
947 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
948 				struct ib_qp_init_attr *init_attr,
949 				struct ib_udata *udata);
950 int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
951 		      int attr_mask, struct ib_udata *udata);
952 int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
953 		     struct ib_qp_init_attr *qp_init_attr);
954 int mlx5_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata);
955 int mlx5_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
956 		      const struct ib_send_wr **bad_wr);
957 int mlx5_ib_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
958 		      const struct ib_recv_wr **bad_wr);
959 void *mlx5_get_send_wqe(struct mlx5_ib_qp *qp, int n);
960 int mlx5_ib_read_user_wqe(struct mlx5_ib_qp *qp, int send, int wqe_index,
961 			  void *buffer, u32 length,
962 			  struct mlx5_ib_qp_base *base);
963 int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
964 		      struct ib_udata *udata);
965 void mlx5_ib_destroy_cq(struct ib_cq *cq, struct ib_udata *udata);
966 int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
967 int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
968 int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
969 int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata);
970 struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc);
971 struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
972 				  u64 virt_addr, int access_flags,
973 				  struct ib_udata *udata);
974 struct ib_mw *mlx5_ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type,
975 			       struct ib_udata *udata);
976 int mlx5_ib_dealloc_mw(struct ib_mw *mw);
977 int mlx5_ib_update_mtt(struct mlx5_ib_mr *mr, u64 start_page_index,
978 		       int npages, int zap);
979 int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start,
980 			  u64 length, u64 virt_addr, int access_flags,
981 			  struct ib_pd *pd, struct ib_udata *udata);
982 int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata);
983 struct ib_mr *mlx5_ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
984 			       u32 max_num_sg, struct ib_udata *udata);
985 int mlx5_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents,
986 		      unsigned int *sg_offset);
987 int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
988 			const struct ib_wc *in_wc, const struct ib_grh *in_grh,
989 			const struct ib_mad_hdr *in, size_t in_mad_size,
990 			struct ib_mad_hdr *out, size_t *out_mad_size,
991 			u16 *out_mad_pkey_index);
992 struct ib_xrcd *mlx5_ib_alloc_xrcd(struct ib_device *ibdev,
993 				   struct ib_udata *udata);
994 int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
995 int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset);
996 int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, u8 port);
997 int mlx5_query_mad_ifc_smp_attr_node_info(struct ib_device *ibdev,
998 					  struct ib_smp *out_mad);
999 int mlx5_query_mad_ifc_system_image_guid(struct ib_device *ibdev,
1000 					 __be64 *sys_image_guid);
1001 int mlx5_query_mad_ifc_max_pkeys(struct ib_device *ibdev,
1002 				 u16 *max_pkeys);
1003 int mlx5_query_mad_ifc_vendor_id(struct ib_device *ibdev,
1004 				 u32 *vendor_id);
1005 int mlx5_query_mad_ifc_node_desc(struct mlx5_ib_dev *dev, char *node_desc);
1006 int mlx5_query_mad_ifc_node_guid(struct mlx5_ib_dev *dev, __be64 *node_guid);
1007 int mlx5_query_mad_ifc_pkey(struct ib_device *ibdev, u8 port, u16 index,
1008 			    u16 *pkey);
1009 int mlx5_query_mad_ifc_gids(struct ib_device *ibdev, u8 port, int index,
1010 			    union ib_gid *gid);
1011 int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u8 port,
1012 			    struct ib_port_attr *props);
1013 int mlx5_ib_query_port(struct ib_device *ibdev, u8 port,
1014 		       struct ib_port_attr *props);
1015 int mlx5_ib_init_fmr(struct mlx5_ib_dev *dev);
1016 void mlx5_ib_cleanup_fmr(struct mlx5_ib_dev *dev);
1017 void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr,
1018 			unsigned long max_page_shift,
1019 			int *count, int *shift,
1020 			int *ncont, int *order);
1021 void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
1022 			    int page_shift, size_t offset, size_t num_pages,
1023 			    __be64 *pas, int access_flags);
1024 void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
1025 			  int page_shift, __be64 *pas, int access_flags);
1026 void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num);
1027 int mlx5_ib_get_cqe_size(struct mlx5_ib_dev *dev, struct ib_cq *ibcq);
1028 int mlx5_mr_cache_init(struct mlx5_ib_dev *dev);
1029 int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev);
1030 int mlx5_mr_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift);
1031 int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
1032 			    struct ib_mr_status *mr_status);
1033 struct ib_wq *mlx5_ib_create_wq(struct ib_pd *pd,
1034 				struct ib_wq_init_attr *init_attr,
1035 				struct ib_udata *udata);
1036 void mlx5_ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
1037 int mlx5_ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1038 		      u32 wq_attr_mask, struct ib_udata *udata);
1039 struct ib_rwq_ind_table *mlx5_ib_create_rwq_ind_table(struct ib_device *device,
1040 						      struct ib_rwq_ind_table_init_attr *init_attr,
1041 						      struct ib_udata *udata);
1042 int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
1043 
1044 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1045 extern struct workqueue_struct *mlx5_ib_page_fault_wq;
1046 
1047 void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev);
1048 void mlx5_ib_mr_pfault_handler(struct mlx5_ib_qp *qp,
1049 			       struct mlx5_ib_pfault *pfault);
1050 void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp);
1051 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);
1052 void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev);
1053 int __init mlx5_ib_odp_init(void);
1054 void mlx5_ib_odp_cleanup(void);
1055 void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp);
1056 void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp);
1057 void mlx5_ib_invalidate_range(struct ib_umem *umem, unsigned long start,
1058 			      unsigned long end);
1059 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev * dev)1060 static inline void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev)
1061 {
1062 	return;
1063 }
1064 
mlx5_ib_odp_create_qp(struct mlx5_ib_qp * qp)1065 static inline void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp)		{}
mlx5_ib_odp_init_one(struct mlx5_ib_dev * ibdev)1066 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; }
mlx5_ib_odp_remove_one(struct mlx5_ib_dev * ibdev)1067 static inline void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev)	{}
mlx5_ib_odp_init(void)1068 static inline int mlx5_ib_odp_init(void) { return 0; }
mlx5_ib_odp_cleanup(void)1069 static inline void mlx5_ib_odp_cleanup(void)				{}
mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp * qp)1070 static inline void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp) {}
mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp * qp)1071 static inline void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp)  {}
1072 
1073 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1074 
1075 int mlx5_ib_get_vf_config(struct ib_device *device, int vf,
1076 			  u8 port, struct ifla_vf_info *info);
1077 int mlx5_ib_set_vf_link_state(struct ib_device *device, int vf,
1078 			      u8 port, int state);
1079 int mlx5_ib_get_vf_stats(struct ib_device *device, int vf,
1080 			 u8 port, struct ifla_vf_stats *stats);
1081 int mlx5_ib_set_vf_guid(struct ib_device *device, int vf, u8 port,
1082 			u64 guid, int type);
1083 
1084 __be16 mlx5_get_roce_udp_sport(struct mlx5_ib_dev *dev,
1085 			       const struct ib_gid_attr *attr);
1086 
1087 /* GSI QP helper functions */
1088 struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
1089 				    struct ib_qp_init_attr *init_attr);
1090 int mlx5_ib_gsi_destroy_qp(struct ib_qp *qp);
1091 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
1092 			  int attr_mask);
1093 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
1094 			 int qp_attr_mask,
1095 			 struct ib_qp_init_attr *qp_init_attr);
1096 int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
1097 			  const struct ib_send_wr **bad_wr);
1098 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, const struct ib_recv_wr *wr,
1099 			  const struct ib_recv_wr **bad_wr);
1100 void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi);
1101 
1102 int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc);
1103 
1104 void mlx5_ib_free_bfreg(struct mlx5_ib_dev *dev, struct mlx5_bfreg_info *bfregi,
1105 			int bfregn);
1106 
1107 #if 1 /* IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS) */
1108 int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user);
1109 void mlx5_ib_devx_destroy(struct mlx5_ib_dev *dev, u16 uid);
1110 void mlx5_ib_devx_init_event_table(struct mlx5_ib_dev *dev);
1111 void mlx5_ib_devx_cleanup_event_table(struct mlx5_ib_dev *dev);
1112 bool mlx5_ib_devx_is_flow_dest(void *obj, int *dest_id, int *dest_type);
1113 bool mlx5_ib_devx_is_flow_counter(void *obj, u32 offset, u32 *counter_id);
1114 extern const struct uapi_definition mlx5_ib_devx_defs[];
1115 #else
1116 static inline int
mlx5_ib_devx_create(struct mlx5_ib_dev * dev,bool is_user)1117 mlx5_ib_devx_create(struct mlx5_ib_dev *dev,
1118 			   bool is_user) { return -EOPNOTSUPP; }
mlx5_ib_devx_destroy(struct mlx5_ib_dev * dev,u16 uid)1119 static inline void mlx5_ib_devx_destroy(struct mlx5_ib_dev *dev, u16 uid) {}
mlx5_ib_devx_init_event_table(struct mlx5_ib_dev * dev)1120 static inline void mlx5_ib_devx_init_event_table(struct mlx5_ib_dev *dev) {}
mlx5_ib_devx_cleanup_event_table(struct mlx5_ib_dev * dev)1121 static inline void mlx5_ib_devx_cleanup_event_table(struct mlx5_ib_dev *dev) {}
mlx5_ib_devx_is_flow_dest(void * obj,int * dest_id,int * dest_type)1122 static inline bool mlx5_ib_devx_is_flow_dest(void *obj, int *dest_id,
1123 					     int *dest_type)
1124 {
1125 	return false;
1126 }
1127 #endif
1128 
init_query_mad(struct ib_smp * mad)1129 static inline void init_query_mad(struct ib_smp *mad)
1130 {
1131 	mad->base_version  = 1;
1132 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
1133 	mad->class_version = 1;
1134 	mad->method	   = IB_MGMT_METHOD_GET;
1135 }
1136 
convert_access(int acc)1137 static inline u8 convert_access(int acc)
1138 {
1139 	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MLX5_PERM_ATOMIC       : 0) |
1140 	       (acc & IB_ACCESS_REMOTE_WRITE  ? MLX5_PERM_REMOTE_WRITE : 0) |
1141 	       (acc & IB_ACCESS_REMOTE_READ   ? MLX5_PERM_REMOTE_READ  : 0) |
1142 	       (acc & IB_ACCESS_LOCAL_WRITE   ? MLX5_PERM_LOCAL_WRITE  : 0) |
1143 	       MLX5_PERM_LOCAL_READ;
1144 }
1145 
is_qp1(enum ib_qp_type qp_type)1146 static inline int is_qp1(enum ib_qp_type qp_type)
1147 {
1148 	return qp_type == MLX5_IB_QPT_HW_GSI;
1149 }
1150 
1151 #define MLX5_MAX_UMR_SHIFT 16
1152 #define MLX5_MAX_UMR_PAGES (1 << MLX5_MAX_UMR_SHIFT)
1153 
check_cq_create_flags(u32 flags)1154 static inline u32 check_cq_create_flags(u32 flags)
1155 {
1156 	/*
1157 	 * It returns non-zero value for unsupported CQ
1158 	 * create flags, otherwise it returns zero.
1159 	 */
1160 	return (flags & ~(IB_CQ_FLAGS_IGNORE_OVERRUN |
1161 			  IB_CQ_FLAGS_TIMESTAMP_COMPLETION));
1162 }
1163 
verify_assign_uidx(u8 cqe_version,u32 cmd_uidx,u32 * user_index)1164 static inline int verify_assign_uidx(u8 cqe_version, u32 cmd_uidx,
1165 				     u32 *user_index)
1166 {
1167 	if (cqe_version) {
1168 		if ((cmd_uidx == MLX5_IB_DEFAULT_UIDX) ||
1169 		    (cmd_uidx & ~MLX5_USER_ASSIGNED_UIDX_MASK))
1170 			return -EINVAL;
1171 		*user_index = cmd_uidx;
1172 	} else {
1173 		*user_index = MLX5_IB_DEFAULT_UIDX;
1174 	}
1175 
1176 	return 0;
1177 }
1178 
get_qp_user_index(struct mlx5_ib_ucontext * ucontext,struct mlx5_ib_create_qp * ucmd,int inlen,u32 * user_index)1179 static inline int get_qp_user_index(struct mlx5_ib_ucontext *ucontext,
1180 				    struct mlx5_ib_create_qp *ucmd,
1181 				    int inlen,
1182 				    u32 *user_index)
1183 {
1184 	u8 cqe_version = ucontext->cqe_version;
1185 
1186 	if (field_avail(struct mlx5_ib_create_qp, uidx, inlen) &&
1187 	    !cqe_version && (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1188 		return 0;
1189 
1190 	if (!!(field_avail(struct mlx5_ib_create_qp, uidx, inlen) !=
1191 	       !!cqe_version))
1192 		return -EINVAL;
1193 
1194 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1195 }
1196 
get_srq_user_index(struct mlx5_ib_ucontext * ucontext,struct mlx5_ib_create_srq * ucmd,int inlen,u32 * user_index)1197 static inline int get_srq_user_index(struct mlx5_ib_ucontext *ucontext,
1198 				     struct mlx5_ib_create_srq *ucmd,
1199 				     int inlen,
1200 				     u32 *user_index)
1201 {
1202 	u8 cqe_version = ucontext->cqe_version;
1203 
1204 	if (field_avail(struct mlx5_ib_create_srq, uidx, inlen) &&
1205 	    !cqe_version && (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1206 		return 0;
1207 
1208 	if (!!(field_avail(struct mlx5_ib_create_srq, uidx, inlen) !=
1209 	       !!cqe_version))
1210 		return -EINVAL;
1211 
1212 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1213 }
1214 
1215 void mlx5_ib_cleanup_congestion(struct mlx5_ib_dev *);
1216 int mlx5_ib_init_congestion(struct mlx5_ib_dev *);
1217 
get_uars_per_sys_page(struct mlx5_ib_dev * dev,bool lib_support)1218 static inline int get_uars_per_sys_page(struct mlx5_ib_dev *dev, bool lib_support)
1219 {
1220 	return lib_support && MLX5_CAP_GEN(dev->mdev, uar_4k) ?
1221 				MLX5_UARS_IN_PAGE : 1;
1222 }
1223 
get_num_static_uars(struct mlx5_ib_dev * dev,struct mlx5_bfreg_info * bfregi)1224 static inline int get_num_static_uars(struct mlx5_ib_dev *dev,
1225 				      struct mlx5_bfreg_info *bfregi)
1226 {
1227 	return get_uars_per_sys_page(dev, bfregi->lib_uar_4k) * bfregi->num_static_sys_pages;
1228 }
1229 
1230 int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
1231 			struct mlx5_bfreg_info *bfregi, u32 bfregn,
1232 			bool dyn_bfreg);
1233 
1234 #endif /* MLX5_IB_H */
1235