xref: /freebsd/sys/dev/mlx5/mlx5_ib/mlx5_ib.h (revision 5eb61f6c6549f134a4f3bed4c164345d4f616bad)
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  * $FreeBSD$
26  */
27 
28 #ifndef MLX5_IB_H
29 #define MLX5_IB_H
30 
31 #include <linux/kernel.h>
32 #include <linux/sched.h>
33 #include <linux/printk.h>
34 #include <linux/netdevice.h>
35 #include <rdma/ib_verbs.h>
36 #include <rdma/ib_smi.h>
37 #include <dev/mlx5/cq.h>
38 #include <dev/mlx5/qp.h>
39 #include <dev/mlx5/srq.h>
40 #include <linux/types.h>
41 #include <dev/mlx5/mlx5_core/transobj.h>
42 #include <rdma/ib_user_verbs.h>
43 #include <rdma/mlx5-abi.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 mlx5_ib_mmap_cmd {
70 	MLX5_IB_MMAP_REGULAR_PAGE		= 0,
71 	MLX5_IB_MMAP_GET_CONTIGUOUS_PAGES	= 1,
72 	MLX5_IB_MMAP_WC_PAGE			= 2,
73 	MLX5_IB_MMAP_NC_PAGE			= 3,
74 	/* 5 is chosen in order to be compatible with old versions of libmlx5 */
75 	MLX5_IB_MMAP_CORE_CLOCK			= 5,
76 };
77 
78 enum {
79 	MLX5_RES_SCAT_DATA32_CQE	= 0x1,
80 	MLX5_RES_SCAT_DATA64_CQE	= 0x2,
81 	MLX5_REQ_SCAT_DATA32_CQE	= 0x11,
82 	MLX5_REQ_SCAT_DATA64_CQE	= 0x22,
83 };
84 
85 enum mlx5_ib_latency_class {
86 	MLX5_IB_LATENCY_CLASS_LOW,
87 	MLX5_IB_LATENCY_CLASS_MEDIUM,
88 	MLX5_IB_LATENCY_CLASS_HIGH,
89 	MLX5_IB_LATENCY_CLASS_FAST_PATH
90 };
91 
92 enum mlx5_ib_mad_ifc_flags {
93 	MLX5_MAD_IFC_IGNORE_MKEY	= 1,
94 	MLX5_MAD_IFC_IGNORE_BKEY	= 2,
95 	MLX5_MAD_IFC_NET_VIEW		= 4,
96 };
97 
98 enum {
99 	MLX5_CROSS_CHANNEL_BFREG         = 0,
100 };
101 
102 enum {
103 	MLX5_CQE_VERSION_V0,
104 	MLX5_CQE_VERSION_V1,
105 };
106 
107 enum {
108 	MLX5_IB_INVALID_UAR_INDEX	= BIT(31),
109 	MLX5_IB_INVALID_BFREG		= BIT(31),
110 };
111 
112 struct mlx5_ib_vma_private_data {
113 	struct list_head list;
114 	struct vm_area_struct *vma;
115 };
116 
117 struct mlx5_bfreg_info {
118 	u32 *sys_pages;
119 	int num_low_latency_bfregs;
120 	unsigned int *count;
121 
122 	/*
123 	 * protect bfreg allocation data structs
124 	 */
125 	struct mutex lock;
126 	u32 ver;
127 	u8 lib_uar_4k : 1;
128 	u8 lib_uar_dyn : 1;
129 	u32 num_sys_pages;
130 	u32 num_static_sys_pages;
131 	u32 total_num_bfregs;
132 	u32 num_dyn_bfregs;
133 };
134 
135 struct mlx5_ib_ucontext {
136 	struct ib_ucontext	ibucontext;
137 	struct list_head	db_page_list;
138 
139 	/* protect doorbell record alloc/free
140 	 */
141 	struct mutex		db_page_mutex;
142 	struct mlx5_bfreg_info	bfregi;
143 	u8			cqe_version;
144 	/* Transport Domain number */
145 	u32			tdn;
146 	struct list_head	vma_private_list;
147 };
148 
149 static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
150 {
151 	return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext);
152 }
153 
154 struct mlx5_ib_pd {
155 	struct ib_pd		ibpd;
156 	u32			pdn;
157 };
158 
159 #define MLX5_IB_FLOW_MCAST_PRIO		(MLX5_BY_PASS_NUM_PRIOS - 1)
160 #define MLX5_IB_FLOW_LAST_PRIO		(MLX5_BY_PASS_NUM_REGULAR_PRIOS - 1)
161 #if (MLX5_IB_FLOW_LAST_PRIO <= 0)
162 #error "Invalid number of bypass priorities"
163 #endif
164 #define MLX5_IB_FLOW_LEFTOVERS_PRIO	(MLX5_IB_FLOW_MCAST_PRIO + 1)
165 
166 #define MLX5_IB_NUM_FLOW_FT		(MLX5_IB_FLOW_LEFTOVERS_PRIO + 1)
167 #define MLX5_IB_NUM_SNIFFER_FTS		2
168 struct mlx5_ib_flow_prio {
169 	struct mlx5_flow_table		*flow_table;
170 	unsigned int			refcount;
171 };
172 
173 struct mlx5_ib_flow_handler {
174 	struct list_head		list;
175 	struct ib_flow			ibflow;
176 	struct mlx5_ib_flow_prio	*prio;
177 	struct mlx5_flow_rule	*rule;
178 };
179 
180 struct mlx5_ib_flow_db {
181 	struct mlx5_ib_flow_prio	prios[MLX5_IB_NUM_FLOW_FT];
182 	struct mlx5_ib_flow_prio	sniffer[MLX5_IB_NUM_SNIFFER_FTS];
183 	struct mlx5_flow_table		*lag_demux_ft;
184 	/* Protect flow steering bypass flow tables
185 	 * when add/del flow rules.
186 	 * only single add/removal of flow steering rule could be done
187 	 * simultaneously.
188 	 */
189 	struct mutex			lock;
190 };
191 
192 /* Use macros here so that don't have to duplicate
193  * enum ib_send_flags and enum ib_qp_type for low-level driver
194  */
195 
196 #define MLX5_IB_SEND_UMR_UNREG	IB_SEND_RESERVED_START
197 #define MLX5_IB_SEND_UMR_FAIL_IF_FREE (IB_SEND_RESERVED_START << 1)
198 #define MLX5_IB_SEND_UMR_UPDATE_MTT (IB_SEND_RESERVED_START << 2)
199 
200 #define MLX5_IB_SEND_UMR_UPDATE_TRANSLATION	(IB_SEND_RESERVED_START << 3)
201 #define MLX5_IB_SEND_UMR_UPDATE_PD		(IB_SEND_RESERVED_START << 4)
202 #define MLX5_IB_SEND_UMR_UPDATE_ACCESS		IB_SEND_RESERVED_END
203 
204 #define MLX5_IB_QPT_REG_UMR	IB_QPT_RESERVED1
205 /*
206  * IB_QPT_GSI creates the software wrapper around GSI, and MLX5_IB_QPT_HW_GSI
207  * creates the actual hardware QP.
208  */
209 #define MLX5_IB_QPT_HW_GSI	IB_QPT_RESERVED2
210 #define MLX5_IB_QPT_DCI		IB_QPT_RESERVED3
211 #define MLX5_IB_QPT_DCT		IB_QPT_RESERVED4
212 #define MLX5_IB_WR_UMR		IB_WR_RESERVED1
213 
214 /* Private QP creation flags to be passed in ib_qp_init_attr.create_flags.
215  *
216  * These flags are intended for internal use by the mlx5_ib driver, and they
217  * rely on the range reserved for that use in the ib_qp_create_flags enum.
218  */
219 #define MLX5_IB_QP_CREATE_SQPN_QP1	IB_QP_CREATE_RESERVED_START
220 #define MLX5_IB_QP_CREATE_WC_TEST	(IB_QP_CREATE_RESERVED_START << 1)
221 
222 struct wr_list {
223 	u16	opcode;
224 	u16	next;
225 };
226 
227 struct mlx5_ib_wq {
228 	u64		       *wrid;
229 	u32		       *wr_data;
230 	struct wr_list	       *w_list;
231 	unsigned	       *wqe_head;
232 	u16		        unsig_count;
233 
234 	/* serialize post to the work queue
235 	 */
236 	spinlock_t		lock;
237 	int			wqe_cnt;
238 	int			max_post;
239 	int			max_gs;
240 	int			offset;
241 	int			wqe_shift;
242 	unsigned		head;
243 	unsigned		tail;
244 	u16			cur_post;
245 	u16			last_poll;
246 	void		       *qend;
247 };
248 
249 struct mlx5_ib_rwq {
250 	struct ib_wq		ibwq;
251 	struct mlx5_core_qp	core_qp;
252 	u32			rq_num_pas;
253 	u32			log_rq_stride;
254 	u32			log_rq_size;
255 	u32			rq_page_offset;
256 	u32			log_page_size;
257 	struct ib_umem		*umem;
258 	size_t			buf_size;
259 	unsigned int		page_shift;
260 	int			create_type;
261 	struct mlx5_db		db;
262 	u32			user_index;
263 	u32			wqe_count;
264 	u32			wqe_shift;
265 	int			wq_sig;
266 };
267 
268 enum {
269 	MLX5_QP_USER,
270 	MLX5_QP_KERNEL,
271 	MLX5_QP_EMPTY
272 };
273 
274 enum {
275 	MLX5_WQ_USER,
276 	MLX5_WQ_KERNEL
277 };
278 
279 struct mlx5_ib_rwq_ind_table {
280 	struct ib_rwq_ind_table ib_rwq_ind_tbl;
281 	u32			rqtn;
282 };
283 
284 /*
285  * Connect-IB can trigger up to four concurrent pagefaults
286  * per-QP.
287  */
288 enum mlx5_ib_pagefault_context {
289 	MLX5_IB_PAGEFAULT_RESPONDER_READ,
290 	MLX5_IB_PAGEFAULT_REQUESTOR_READ,
291 	MLX5_IB_PAGEFAULT_RESPONDER_WRITE,
292 	MLX5_IB_PAGEFAULT_REQUESTOR_WRITE,
293 	MLX5_IB_PAGEFAULT_CONTEXTS
294 };
295 
296 static inline enum mlx5_ib_pagefault_context
297 	mlx5_ib_get_pagefault_context(struct mlx5_pagefault *pagefault)
298 {
299 	return pagefault->flags & (MLX5_PFAULT_REQUESTOR | MLX5_PFAULT_WRITE);
300 }
301 
302 struct mlx5_ib_pfault {
303 	struct work_struct	work;
304 	struct mlx5_pagefault	mpfault;
305 };
306 
307 struct mlx5_ib_ubuffer {
308 	struct ib_umem	       *umem;
309 	int			buf_size;
310 	u64			buf_addr;
311 };
312 
313 struct mlx5_ib_qp_base {
314 	struct mlx5_ib_qp	*container_mibqp;
315 	struct mlx5_core_qp	mqp;
316 	struct mlx5_ib_ubuffer	ubuffer;
317 };
318 
319 struct mlx5_ib_qp_trans {
320 	struct mlx5_ib_qp_base	base;
321 	u16			xrcdn;
322 	u8			alt_port;
323 	u8			atomic_rd_en;
324 	u8			resp_depth;
325 };
326 
327 struct mlx5_ib_rss_qp {
328 	u32	tirn;
329 };
330 
331 struct mlx5_ib_rq {
332 	struct mlx5_ib_qp_base base;
333 	struct mlx5_ib_wq	*rq;
334 	struct mlx5_ib_ubuffer	ubuffer;
335 	struct mlx5_db		*doorbell;
336 	u32			tirn;
337 	u8			state;
338 };
339 
340 struct mlx5_ib_sq {
341 	struct mlx5_ib_qp_base base;
342 	struct mlx5_ib_wq	*sq;
343 	struct mlx5_ib_ubuffer  ubuffer;
344 	struct mlx5_db		*doorbell;
345 	u32			tisn;
346 	u8			state;
347 };
348 
349 struct mlx5_ib_raw_packet_qp {
350 	struct mlx5_ib_sq sq;
351 	struct mlx5_ib_rq rq;
352 };
353 
354 struct mlx5_bf {
355 	int			buf_size;
356 	unsigned long		offset;
357 	struct mlx5_sq_bfreg   *bfreg;
358 	spinlock_t		lock32;
359 };
360 
361 struct mlx5_ib_qp {
362 	struct ib_qp		ibqp;
363 	union {
364 		struct mlx5_ib_qp_trans trans_qp;
365 		struct mlx5_ib_raw_packet_qp raw_packet_qp;
366 		struct mlx5_ib_rss_qp rss_qp;
367 	};
368 	struct mlx5_buf		buf;
369 
370 	struct mlx5_db		db;
371 	struct mlx5_ib_wq	rq;
372 
373 	u8			sq_signal_bits;
374 	u8			fm_cache;
375 	struct mlx5_ib_wq	sq;
376 
377 	/* serialize qp state modifications
378 	 */
379 	struct mutex		mutex;
380 	u32			flags;
381 	u8			port;
382 	u8			state;
383 	int			wq_sig;
384 	int			scat_cqe;
385 	int			max_inline_data;
386 	struct mlx5_bf	        bf;
387 	int			has_rq;
388 
389 	/* only for user space QPs. For kernel
390 	 * we have it from the bf object
391 	 */
392 	int			bfregn;
393 
394 	int			create_type;
395 
396 	/* Store signature errors */
397 	bool			signature_en;
398 
399 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
400 	/*
401 	 * A flag that is true for QP's that are in a state that doesn't
402 	 * allow page faults, and shouldn't schedule any more faults.
403 	 */
404 	int                     disable_page_faults;
405 	/*
406 	 * The disable_page_faults_lock protects a QP's disable_page_faults
407 	 * field, allowing for a thread to atomically check whether the QP
408 	 * allows page faults, and if so schedule a page fault.
409 	 */
410 	spinlock_t              disable_page_faults_lock;
411 	struct mlx5_ib_pfault	pagefaults[MLX5_IB_PAGEFAULT_CONTEXTS];
412 #endif
413 	struct list_head	qps_list;
414 	struct list_head	cq_recv_list;
415 	struct list_head	cq_send_list;
416 };
417 
418 struct mlx5_ib_cq_buf {
419 	struct mlx5_buf		buf;
420 	struct ib_umem		*umem;
421 	int			cqe_size;
422 	int			nent;
423 };
424 
425 enum mlx5_ib_qp_flags {
426 	MLX5_IB_QP_LSO                          = IB_QP_CREATE_IPOIB_UD_LSO,
427 	MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK     = IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
428 	MLX5_IB_QP_CROSS_CHANNEL            = IB_QP_CREATE_CROSS_CHANNEL,
429 	MLX5_IB_QP_MANAGED_SEND             = IB_QP_CREATE_MANAGED_SEND,
430 	MLX5_IB_QP_MANAGED_RECV             = IB_QP_CREATE_MANAGED_RECV,
431 	MLX5_IB_QP_SIGNATURE_HANDLING           = 1 << 5,
432 	/* QP uses 1 as its source QP number */
433 	MLX5_IB_QP_SQPN_QP1			= 1 << 6,
434 	MLX5_IB_QP_CAP_SCATTER_FCS		= 1 << 7,
435 	MLX5_IB_QP_RSS				= 1 << 8,
436 };
437 
438 struct mlx5_umr_wr {
439 	struct ib_send_wr		wr;
440 	union {
441 		u64			virt_addr;
442 		u64			offset;
443 	} target;
444 	struct ib_pd		       *pd;
445 	unsigned int			page_shift;
446 	unsigned int			npages;
447 	u32				length;
448 	int				access_flags;
449 	u32				mkey;
450 };
451 
452 static inline const struct mlx5_umr_wr *umr_wr(const struct ib_send_wr *wr)
453 {
454 	return container_of(wr, struct mlx5_umr_wr, wr);
455 }
456 
457 struct mlx5_shared_mr_info {
458 	int mr_id;
459 	struct ib_umem		*umem;
460 };
461 
462 struct mlx5_ib_cq {
463 	struct ib_cq		ibcq;
464 	struct mlx5_core_cq	mcq;
465 	struct mlx5_ib_cq_buf	buf;
466 	struct mlx5_db		db;
467 
468 	/* serialize access to the CQ
469 	 */
470 	spinlock_t		lock;
471 
472 	/* protect resize cq
473 	 */
474 	struct mutex		resize_mutex;
475 	struct mlx5_ib_cq_buf  *resize_buf;
476 	struct ib_umem	       *resize_umem;
477 	int			cqe_size;
478 	struct list_head	list_send_qp;
479 	struct list_head	list_recv_qp;
480 	u32			create_flags;
481 	struct list_head	wc_list;
482 	enum ib_cq_notify_flags notify_flags;
483 	struct work_struct	notify_work;
484 };
485 
486 struct mlx5_ib_wc {
487 	struct ib_wc wc;
488 	struct list_head list;
489 };
490 
491 struct mlx5_ib_srq {
492 	struct ib_srq		ibsrq;
493 	struct mlx5_core_srq	msrq;
494 	struct mlx5_buf		buf;
495 	struct mlx5_db		db;
496 	u64		       *wrid;
497 	/* protect SRQ hanlding
498 	 */
499 	spinlock_t		lock;
500 	int			head;
501 	int			tail;
502 	u16			wqe_ctr;
503 	struct ib_umem	       *umem;
504 	/* serialize arming a SRQ
505 	 */
506 	struct mutex		mutex;
507 	int			wq_sig;
508 };
509 
510 struct mlx5_ib_xrcd {
511 	struct ib_xrcd		ibxrcd;
512 	u32			xrcdn;
513 };
514 
515 enum mlx5_ib_mtt_access_flags {
516 	MLX5_IB_MTT_READ  = (1 << 0),
517 	MLX5_IB_MTT_WRITE = (1 << 1),
518 };
519 
520 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
521 
522 struct mlx5_ib_mr {
523 	struct ib_mr		ibmr;
524 	void			*descs;
525 	dma_addr_t		desc_map;
526 	int			ndescs;
527 	int			max_descs;
528 	int			desc_size;
529 	int			access_mode;
530 	struct mlx5_core_mr	mmkey;
531 	struct ib_umem	       *umem;
532 	struct mlx5_shared_mr_info	*smr_info;
533 	struct list_head	list;
534 	int			order;
535 	int			umred;
536 	int			npages;
537 	struct mlx5_ib_dev     *dev;
538 	u32 out[MLX5_ST_SZ_DW(create_mkey_out)];
539 	struct mlx5_core_sig_ctx    *sig;
540 	int			live;
541 	void			*descs_alloc;
542 	int			access_flags; /* Needed for rereg MR */
543 	struct mlx5_async_work	cb_work;
544 };
545 
546 struct mlx5_ib_mw {
547 	struct ib_mw		ibmw;
548 	struct mlx5_core_mr	mmkey;
549 };
550 
551 struct mlx5_ib_umr_context {
552 	struct ib_cqe		cqe;
553 	enum ib_wc_status	status;
554 	struct completion	done;
555 };
556 
557 struct umr_common {
558 	struct ib_pd	*pd;
559 	struct ib_cq	*cq;
560 	struct ib_qp	*qp;
561 	/* control access to UMR QP
562 	 */
563 	struct semaphore	sem;
564 };
565 
566 enum {
567 	MLX5_FMR_INVALID,
568 	MLX5_FMR_VALID,
569 	MLX5_FMR_BUSY,
570 };
571 
572 struct mlx5_cache_ent {
573 	struct list_head	head;
574 	/* sync access to the cahce entry
575 	 */
576 	spinlock_t		lock;
577 
578 
579 	struct dentry	       *dir;
580 	char                    name[4];
581 	u32                     order;
582 	u32			size;
583 	u32                     cur;
584 	u32                     miss;
585 	u32			limit;
586 
587 	struct dentry          *fsize;
588 	struct dentry          *fcur;
589 	struct dentry          *fmiss;
590 	struct dentry          *flimit;
591 
592 	struct mlx5_ib_dev     *dev;
593 	struct work_struct	work;
594 	struct delayed_work	dwork;
595 	int			pending;
596 };
597 
598 struct mlx5_mr_cache {
599 	struct workqueue_struct *wq;
600 	struct mlx5_cache_ent	ent[MAX_MR_CACHE_ENTRIES];
601 	int			stopped;
602 	struct dentry		*root;
603 	unsigned long		last_add;
604 };
605 
606 struct mlx5_ib_gsi_qp;
607 
608 struct mlx5_ib_port_resources {
609 	struct mlx5_ib_resources *devr;
610 	struct mlx5_ib_gsi_qp *gsi;
611 	struct work_struct pkey_change_work;
612 };
613 
614 struct mlx5_ib_resources {
615 	struct ib_cq	*c0;
616 	struct ib_xrcd	*x0;
617 	struct ib_xrcd	*x1;
618 	struct ib_pd	*p0;
619 	struct ib_srq	*s0;
620 	struct ib_srq	*s1;
621 	struct mlx5_ib_port_resources ports[2];
622 	/* Protects changes to the port resources */
623 	struct mutex	mutex;
624 };
625 
626 struct mlx5_ib_port {
627 	u16 q_cnt_id;
628 };
629 
630 struct mlx5_roce {
631 	/* Protect mlx5_ib_get_netdev from invoking dev_hold() with a NULL
632 	 * netdev pointer
633 	 */
634 	rwlock_t		netdev_lock;
635 	struct ifnet		*netdev;
636 	struct notifier_block	nb;
637 	atomic_t		next_port;
638 };
639 
640 #define	MLX5_IB_STATS_COUNT(a,...) a
641 #define	MLX5_IB_STATS_VAR(a,b,c,...) b c;
642 #define	MLX5_IB_STATS_DESC(a,b,c,d,e,...) d, e,
643 
644 #define	MLX5_IB_CONG_PARAMS(m) \
645   /* ECN RP */ \
646   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") \
647   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") \
648   m(+1, u64, rp_time_reset, "rp_time_reset", "Time in microseconds between rate increases if no CNPs are received") \
649   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.") \
650   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") \
651   m(+1, u64, rp_ai_rate, "rp_ai_rate", "The rate, in Mbits per second, used to increase rpTargetRate in the active increase state") \
652   m(+1, u64, rp_hai_rate, "rp_hai_rate", "The rate, in Mbits per second, used to increase rpTargetRate in the hyper increase state") \
653   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]") \
654   m(+1, u64, rp_min_rate, "rp_min_rate", "The minimum value, in Mbps per second, for rate to limit") \
655   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") \
656   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") \
657   m(+1, u64, rp_dce_tcp_rtt, "rp_dce_tcp_rtt", "The time between updates of the aolpha value, in microseconds") \
658   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") \
659   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") \
660   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") \
661   /* ECN NP */ \
662   m(+1, u64, np_cnp_dscp, "np_cnp_dscp", "The DiffServ Code Point of the generated CNP for this port") \
663   m(+1, u64, np_cnp_prio_mode, "np_cnp_prio_mode", "The 802.1p priority value of the generated CNP for this port") \
664   m(+1, u64, np_cnp_prio, "np_cnp_prio", "The 802.1p priority value of the generated CNP for this port")
665 
666 #define	MLX5_IB_CONG_PARAMS_NUM (0 MLX5_IB_CONG_PARAMS(MLX5_IB_STATS_COUNT))
667 
668 #define	MLX5_IB_CONG_STATS(m) \
669   m(+1, u64, syndrome, "syndrome", "Syndrome number") \
670   m(+1, u64, rp_cur_flows, "rp_cur_flows", "Number of flows limited") \
671   m(+1, u64, sum_flows, "sum_flows", "Sum of the number of flows limited over time") \
672   m(+1, u64, rp_cnp_ignored, "rp_cnp_ignored", "Number of CNPs and CNMs ignored") \
673   m(+1, u64, rp_cnp_handled, "rp_cnp_handled", "Number of CNPs and CNMs successfully handled") \
674   m(+1, u64, time_stamp, "time_stamp", "Time stamp in microseconds") \
675   m(+1, u64, accumulators_period, "accumulators_period", "The value of X variable for accumulating counters") \
676   m(+1, u64, np_ecn_marked_roce_packets, "np_ecn_marked_roce_packets", "Number of ECN marked packets seen") \
677   m(+1, u64, np_cnp_sent, "np_cnp_sent", "Number of CNPs sent")
678 
679 #define	MLX5_IB_CONG_STATS_NUM (0 MLX5_IB_CONG_STATS(MLX5_IB_STATS_COUNT))
680 
681 #define	MLX5_IB_CONG_STATUS(m) \
682   /* ECN RP */ \
683   m(+1, u64, rp_0_enable, "rp_0_enable", "Enable reaction point, priority 0", MLX5_IB_RROCE_ECN_RP, 0, enable) \
684   m(+1, u64, rp_1_enable, "rp_1_enable", "Enable reaction point, priority 1", MLX5_IB_RROCE_ECN_RP, 1, enable) \
685   m(+1, u64, rp_2_enable, "rp_2_enable", "Enable reaction point, priority 2", MLX5_IB_RROCE_ECN_RP, 2, enable) \
686   m(+1, u64, rp_3_enable, "rp_3_enable", "Enable reaction point, priority 3", MLX5_IB_RROCE_ECN_RP, 3, enable) \
687   m(+1, u64, rp_4_enable, "rp_4_enable", "Enable reaction point, priority 4", MLX5_IB_RROCE_ECN_RP, 4, enable) \
688   m(+1, u64, rp_5_enable, "rp_5_enable", "Enable reaction point, priority 5", MLX5_IB_RROCE_ECN_RP, 5, enable) \
689   m(+1, u64, rp_6_enable, "rp_6_enable", "Enable reaction point, priority 6", MLX5_IB_RROCE_ECN_RP, 6, enable) \
690   m(+1, u64, rp_7_enable, "rp_7_enable", "Enable reaction point, priority 7", MLX5_IB_RROCE_ECN_RP, 7, enable) \
691   m(+1, u64, rp_8_enable, "rp_8_enable", "Enable reaction point, priority 8", MLX5_IB_RROCE_ECN_RP, 8, enable) \
692   m(+1, u64, rp_9_enable, "rp_9_enable", "Enable reaction point, priority 9", MLX5_IB_RROCE_ECN_RP, 9, enable) \
693   m(+1, u64, rp_10_enable, "rp_10_enable", "Enable reaction point, priority 10", MLX5_IB_RROCE_ECN_RP, 10, enable) \
694   m(+1, u64, rp_11_enable, "rp_11_enable", "Enable reaction point, priority 11", MLX5_IB_RROCE_ECN_RP, 11, enable) \
695   m(+1, u64, rp_12_enable, "rp_12_enable", "Enable reaction point, priority 12", MLX5_IB_RROCE_ECN_RP, 12, enable) \
696   m(+1, u64, rp_13_enable, "rp_13_enable", "Enable reaction point, priority 13", MLX5_IB_RROCE_ECN_RP, 13, enable) \
697   m(+1, u64, rp_14_enable, "rp_14_enable", "Enable reaction point, priority 14", MLX5_IB_RROCE_ECN_RP, 14, enable) \
698   m(+1, u64, rp_15_enable, "rp_15_enable", "Enable reaction point, priority 15", MLX5_IB_RROCE_ECN_RP, 15, enable) \
699   /* ECN NP */ \
700   m(+1, u64, np_0_enable, "np_0_enable", "Enable notification point, priority 0", MLX5_IB_RROCE_ECN_NP, 0, enable) \
701   m(+1, u64, np_1_enable, "np_1_enable", "Enable notification point, priority 1", MLX5_IB_RROCE_ECN_NP, 1, enable) \
702   m(+1, u64, np_2_enable, "np_2_enable", "Enable notification point, priority 2", MLX5_IB_RROCE_ECN_NP, 2, enable) \
703   m(+1, u64, np_3_enable, "np_3_enable", "Enable notification point, priority 3", MLX5_IB_RROCE_ECN_NP, 3, enable) \
704   m(+1, u64, np_4_enable, "np_4_enable", "Enable notification point, priority 4", MLX5_IB_RROCE_ECN_NP, 4, enable) \
705   m(+1, u64, np_5_enable, "np_5_enable", "Enable notification point, priority 5", MLX5_IB_RROCE_ECN_NP, 5, enable) \
706   m(+1, u64, np_6_enable, "np_6_enable", "Enable notification point, priority 6", MLX5_IB_RROCE_ECN_NP, 6, enable) \
707   m(+1, u64, np_7_enable, "np_7_enable", "Enable notification point, priority 7", MLX5_IB_RROCE_ECN_NP, 7, enable) \
708   m(+1, u64, np_8_enable, "np_8_enable", "Enable notification point, priority 8", MLX5_IB_RROCE_ECN_NP, 8, enable) \
709   m(+1, u64, np_9_enable, "np_9_enable", "Enable notification point, priority 9", MLX5_IB_RROCE_ECN_NP, 9, enable) \
710   m(+1, u64, np_10_enable, "np_10_enable", "Enable notification point, priority 10", MLX5_IB_RROCE_ECN_NP, 10, enable) \
711   m(+1, u64, np_11_enable, "np_11_enable", "Enable notification point, priority 11", MLX5_IB_RROCE_ECN_NP, 11, enable) \
712   m(+1, u64, np_12_enable, "np_12_enable", "Enable notification point, priority 12", MLX5_IB_RROCE_ECN_NP, 12, enable) \
713   m(+1, u64, np_13_enable, "np_13_enable", "Enable notification point, priority 13", MLX5_IB_RROCE_ECN_NP, 13, enable) \
714   m(+1, u64, np_14_enable, "np_14_enable", "Enable notification point, priority 14", MLX5_IB_RROCE_ECN_NP, 14, enable) \
715   m(+1, u64, np_15_enable, "np_15_enable", "Enable notification point, priority 15", MLX5_IB_RROCE_ECN_NP, 15, enable) \
716 
717 #define	MLX5_IB_CONG_STATUS_NUM (0 MLX5_IB_CONG_STATUS(MLX5_IB_STATS_COUNT))
718 
719 struct mlx5_ib_congestion {
720 	struct sysctl_ctx_list ctx;
721 	struct sx lock;
722 	struct delayed_work dwork;
723 	union {
724 		u64	arg[1];
725 		struct {
726 			MLX5_IB_CONG_PARAMS(MLX5_IB_STATS_VAR)
727 			MLX5_IB_CONG_STATS(MLX5_IB_STATS_VAR)
728 			MLX5_IB_CONG_STATUS(MLX5_IB_STATS_VAR)
729 		};
730 	};
731 };
732 
733 struct mlx5_ib_dev {
734 	struct ib_device		ib_dev;
735 	struct mlx5_core_dev		*mdev;
736 	struct mlx5_roce		roce;
737 	MLX5_DECLARE_DOORBELL_LOCK(uar_lock);
738 	int				num_ports;
739 	/* serialize update of capability mask
740 	 */
741 	struct mutex			cap_mask_mutex;
742 	bool				ib_active;
743 	struct umr_common		umrc;
744 	/* sync used page count stats
745 	 */
746 	struct mlx5_ib_resources	devr;
747 	struct mlx5_mr_cache		cache;
748 	struct timer_list		delay_timer;
749 	/* Prevents soft lock on massive reg MRs */
750 	struct mutex			slow_path_mutex;
751 	int				fill_delay;
752 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
753 	struct ib_odp_caps	odp_caps;
754 	/*
755 	 * Sleepable RCU that prevents destruction of MRs while they are still
756 	 * being used by a page fault handler.
757 	 */
758 	struct srcu_struct      mr_srcu;
759 #endif
760 	struct mlx5_ib_flow_db	flow_db;
761 	/* protect resources needed as part of reset flow */
762 	spinlock_t		reset_flow_resource_lock;
763 	struct list_head	qp_list;
764 	/* Array with num_ports elements */
765 	struct mlx5_ib_port	*port;
766 	struct mlx5_sq_bfreg	bfreg;
767 	struct mlx5_sq_bfreg	wc_bfreg;
768 	struct mlx5_sq_bfreg	fp_bfreg;
769 	struct mlx5_ib_congestion congestion;
770 
771 	struct mlx5_async_ctx	async_ctx;
772 };
773 
774 static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
775 {
776 	return container_of(mcq, struct mlx5_ib_cq, mcq);
777 }
778 
779 static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
780 {
781 	return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd);
782 }
783 
784 static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev)
785 {
786 	return container_of(ibdev, struct mlx5_ib_dev, ib_dev);
787 }
788 
789 static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq)
790 {
791 	return container_of(ibcq, struct mlx5_ib_cq, ibcq);
792 }
793 
794 static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp)
795 {
796 	return container_of(mqp, struct mlx5_ib_qp_base, mqp)->container_mibqp;
797 }
798 
799 static inline struct mlx5_ib_rwq *to_mibrwq(struct mlx5_core_qp *core_qp)
800 {
801 	return container_of(core_qp, struct mlx5_ib_rwq, core_qp);
802 }
803 
804 static inline struct mlx5_ib_mr *to_mibmr(struct mlx5_core_mr *mmkey)
805 {
806 	return container_of(mmkey, struct mlx5_ib_mr, mmkey);
807 }
808 
809 static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd)
810 {
811 	return container_of(ibpd, struct mlx5_ib_pd, ibpd);
812 }
813 
814 static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq)
815 {
816 	return container_of(ibsrq, struct mlx5_ib_srq, ibsrq);
817 }
818 
819 static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp)
820 {
821 	return container_of(ibqp, struct mlx5_ib_qp, ibqp);
822 }
823 
824 static inline struct mlx5_ib_rwq *to_mrwq(struct ib_wq *ibwq)
825 {
826 	return container_of(ibwq, struct mlx5_ib_rwq, ibwq);
827 }
828 
829 static inline struct mlx5_ib_rwq_ind_table *to_mrwq_ind_table(struct ib_rwq_ind_table *ib_rwq_ind_tbl)
830 {
831 	return container_of(ib_rwq_ind_tbl, struct mlx5_ib_rwq_ind_table, ib_rwq_ind_tbl);
832 }
833 
834 static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
835 {
836 	return container_of(msrq, struct mlx5_ib_srq, msrq);
837 }
838 
839 static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
840 {
841 	return container_of(ibmr, struct mlx5_ib_mr, ibmr);
842 }
843 
844 static inline struct mlx5_ib_mw *to_mmw(struct ib_mw *ibmw)
845 {
846 	return container_of(ibmw, struct mlx5_ib_mw, ibmw);
847 }
848 
849 struct mlx5_ib_ah {
850 	struct ib_ah		ibah;
851 	struct mlx5_av		av;
852 };
853 
854 static inline struct mlx5_ib_ah *to_mah(struct ib_ah *ibah)
855 {
856 	return container_of(ibah, struct mlx5_ib_ah, ibah);
857 }
858 
859 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
860 			struct mlx5_db *db);
861 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);
862 void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
863 void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
864 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index);
865 int mlx5_MAD_IFC(struct mlx5_ib_dev *dev, int ignore_mkey, int ignore_bkey,
866 		 u8 port, const struct ib_wc *in_wc, const struct ib_grh *in_grh,
867 		 const void *in_mad, void *response_mad);
868 struct ib_ah *mlx5_ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr,
869 				struct ib_udata *udata);
870 int mlx5_ib_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr);
871 int mlx5_ib_destroy_ah(struct ib_ah *ah);
872 struct ib_srq *mlx5_ib_create_srq(struct ib_pd *pd,
873 				  struct ib_srq_init_attr *init_attr,
874 				  struct ib_udata *udata);
875 int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
876 		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
877 int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr);
878 int mlx5_ib_destroy_srq(struct ib_srq *srq);
879 int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
880 			  const struct ib_recv_wr **bad_wr);
881 struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
882 				struct ib_qp_init_attr *init_attr,
883 				struct ib_udata *udata);
884 int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
885 		      int attr_mask, struct ib_udata *udata);
886 int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
887 		     struct ib_qp_init_attr *qp_init_attr);
888 int mlx5_ib_destroy_qp(struct ib_qp *qp);
889 int mlx5_ib_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
890 		      const struct ib_send_wr **bad_wr);
891 int mlx5_ib_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
892 		      const struct ib_recv_wr **bad_wr);
893 void *mlx5_get_send_wqe(struct mlx5_ib_qp *qp, int n);
894 int mlx5_ib_read_user_wqe(struct mlx5_ib_qp *qp, int send, int wqe_index,
895 			  void *buffer, u32 length,
896 			  struct mlx5_ib_qp_base *base);
897 struct ib_cq *mlx5_ib_create_cq(struct ib_device *ibdev,
898 				const struct ib_cq_init_attr *attr,
899 				struct ib_ucontext *context,
900 				struct ib_udata *udata);
901 int mlx5_ib_destroy_cq(struct ib_cq *cq);
902 int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
903 int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
904 int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
905 int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata);
906 struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc);
907 struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
908 				  u64 virt_addr, int access_flags,
909 				  struct ib_udata *udata);
910 struct ib_mw *mlx5_ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type,
911 			       struct ib_udata *udata);
912 int mlx5_ib_dealloc_mw(struct ib_mw *mw);
913 int mlx5_ib_update_mtt(struct mlx5_ib_mr *mr, u64 start_page_index,
914 		       int npages, int zap);
915 int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start,
916 			  u64 length, u64 virt_addr, int access_flags,
917 			  struct ib_pd *pd, struct ib_udata *udata);
918 int mlx5_ib_dereg_mr(struct ib_mr *ibmr);
919 struct ib_mr *mlx5_ib_alloc_mr(struct ib_pd *pd,
920 			       enum ib_mr_type mr_type,
921 			       u32 max_num_sg);
922 int mlx5_ib_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents,
923 		      unsigned int *sg_offset);
924 int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
925 			const struct ib_wc *in_wc, const struct ib_grh *in_grh,
926 			const struct ib_mad_hdr *in, size_t in_mad_size,
927 			struct ib_mad_hdr *out, size_t *out_mad_size,
928 			u16 *out_mad_pkey_index);
929 struct ib_xrcd *mlx5_ib_alloc_xrcd(struct ib_device *ibdev,
930 					  struct ib_ucontext *context,
931 					  struct ib_udata *udata);
932 int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd);
933 int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset);
934 int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, u8 port);
935 int mlx5_query_mad_ifc_smp_attr_node_info(struct ib_device *ibdev,
936 					  struct ib_smp *out_mad);
937 int mlx5_query_mad_ifc_system_image_guid(struct ib_device *ibdev,
938 					 __be64 *sys_image_guid);
939 int mlx5_query_mad_ifc_max_pkeys(struct ib_device *ibdev,
940 				 u16 *max_pkeys);
941 int mlx5_query_mad_ifc_vendor_id(struct ib_device *ibdev,
942 				 u32 *vendor_id);
943 int mlx5_query_mad_ifc_node_desc(struct mlx5_ib_dev *dev, char *node_desc);
944 int mlx5_query_mad_ifc_node_guid(struct mlx5_ib_dev *dev, __be64 *node_guid);
945 int mlx5_query_mad_ifc_pkey(struct ib_device *ibdev, u8 port, u16 index,
946 			    u16 *pkey);
947 int mlx5_query_mad_ifc_gids(struct ib_device *ibdev, u8 port, int index,
948 			    union ib_gid *gid);
949 int mlx5_query_mad_ifc_port(struct ib_device *ibdev, u8 port,
950 			    struct ib_port_attr *props);
951 int mlx5_ib_query_port(struct ib_device *ibdev, u8 port,
952 		       struct ib_port_attr *props);
953 int mlx5_ib_init_fmr(struct mlx5_ib_dev *dev);
954 void mlx5_ib_cleanup_fmr(struct mlx5_ib_dev *dev);
955 void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr,
956 			unsigned long max_page_shift,
957 			int *count, int *shift,
958 			int *ncont, int *order);
959 void __mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
960 			    int page_shift, size_t offset, size_t num_pages,
961 			    __be64 *pas, int access_flags);
962 void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
963 			  int page_shift, __be64 *pas, int access_flags);
964 void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num);
965 int mlx5_ib_get_cqe_size(struct mlx5_ib_dev *dev, struct ib_cq *ibcq);
966 int mlx5_mr_cache_init(struct mlx5_ib_dev *dev);
967 int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev);
968 int mlx5_mr_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift);
969 int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
970 			    struct ib_mr_status *mr_status);
971 struct ib_wq *mlx5_ib_create_wq(struct ib_pd *pd,
972 				struct ib_wq_init_attr *init_attr,
973 				struct ib_udata *udata);
974 int mlx5_ib_destroy_wq(struct ib_wq *wq);
975 int mlx5_ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
976 		      u32 wq_attr_mask, struct ib_udata *udata);
977 struct ib_rwq_ind_table *mlx5_ib_create_rwq_ind_table(struct ib_device *device,
978 						      struct ib_rwq_ind_table_init_attr *init_attr,
979 						      struct ib_udata *udata);
980 int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
981 
982 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
983 extern struct workqueue_struct *mlx5_ib_page_fault_wq;
984 
985 void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev);
986 void mlx5_ib_mr_pfault_handler(struct mlx5_ib_qp *qp,
987 			       struct mlx5_ib_pfault *pfault);
988 void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp);
989 int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev);
990 void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev);
991 int __init mlx5_ib_odp_init(void);
992 void mlx5_ib_odp_cleanup(void);
993 void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp);
994 void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp);
995 void mlx5_ib_invalidate_range(struct ib_umem *umem, unsigned long start,
996 			      unsigned long end);
997 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
998 static inline void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev)
999 {
1000 	return;
1001 }
1002 
1003 static inline void mlx5_ib_odp_create_qp(struct mlx5_ib_qp *qp)		{}
1004 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev) { return 0; }
1005 static inline void mlx5_ib_odp_remove_one(struct mlx5_ib_dev *ibdev)	{}
1006 static inline int mlx5_ib_odp_init(void) { return 0; }
1007 static inline void mlx5_ib_odp_cleanup(void)				{}
1008 static inline void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp *qp) {}
1009 static inline void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp *qp)  {}
1010 
1011 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
1012 
1013 int mlx5_ib_get_vf_config(struct ib_device *device, int vf,
1014 			  u8 port, struct ifla_vf_info *info);
1015 int mlx5_ib_set_vf_link_state(struct ib_device *device, int vf,
1016 			      u8 port, int state);
1017 int mlx5_ib_get_vf_stats(struct ib_device *device, int vf,
1018 			 u8 port, struct ifla_vf_stats *stats);
1019 int mlx5_ib_set_vf_guid(struct ib_device *device, int vf, u8 port,
1020 			u64 guid, int type);
1021 
1022 __be16 mlx5_get_roce_udp_sport(struct mlx5_ib_dev *dev, u8 port_num,
1023 			       int index);
1024 int mlx5_get_roce_gid_type(struct mlx5_ib_dev *dev, u8 port_num,
1025 			   int index, enum ib_gid_type *gid_type);
1026 
1027 /* GSI QP helper functions */
1028 struct ib_qp *mlx5_ib_gsi_create_qp(struct ib_pd *pd,
1029 				    struct ib_qp_init_attr *init_attr);
1030 int mlx5_ib_gsi_destroy_qp(struct ib_qp *qp);
1031 int mlx5_ib_gsi_modify_qp(struct ib_qp *qp, struct ib_qp_attr *attr,
1032 			  int attr_mask);
1033 int mlx5_ib_gsi_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
1034 			 int qp_attr_mask,
1035 			 struct ib_qp_init_attr *qp_init_attr);
1036 int mlx5_ib_gsi_post_send(struct ib_qp *qp, const struct ib_send_wr *wr,
1037 			  const struct ib_send_wr **bad_wr);
1038 int mlx5_ib_gsi_post_recv(struct ib_qp *qp, const struct ib_recv_wr *wr,
1039 			  const struct ib_recv_wr **bad_wr);
1040 void mlx5_ib_gsi_pkey_change(struct mlx5_ib_gsi_qp *gsi);
1041 
1042 int mlx5_ib_generate_wc(struct ib_cq *ibcq, struct ib_wc *wc);
1043 
1044 void mlx5_ib_free_bfreg(struct mlx5_ib_dev *dev, struct mlx5_bfreg_info *bfregi,
1045 			int bfregn);
1046 static inline void init_query_mad(struct ib_smp *mad)
1047 {
1048 	mad->base_version  = 1;
1049 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
1050 	mad->class_version = 1;
1051 	mad->method	   = IB_MGMT_METHOD_GET;
1052 }
1053 
1054 static inline u8 convert_access(int acc)
1055 {
1056 	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MLX5_PERM_ATOMIC       : 0) |
1057 	       (acc & IB_ACCESS_REMOTE_WRITE  ? MLX5_PERM_REMOTE_WRITE : 0) |
1058 	       (acc & IB_ACCESS_REMOTE_READ   ? MLX5_PERM_REMOTE_READ  : 0) |
1059 	       (acc & IB_ACCESS_LOCAL_WRITE   ? MLX5_PERM_LOCAL_WRITE  : 0) |
1060 	       MLX5_PERM_LOCAL_READ;
1061 }
1062 
1063 static inline int is_qp1(enum ib_qp_type qp_type)
1064 {
1065 	return qp_type == MLX5_IB_QPT_HW_GSI;
1066 }
1067 
1068 #define MLX5_MAX_UMR_SHIFT 16
1069 #define MLX5_MAX_UMR_PAGES (1 << MLX5_MAX_UMR_SHIFT)
1070 
1071 static inline u32 check_cq_create_flags(u32 flags)
1072 {
1073 	/*
1074 	 * It returns non-zero value for unsupported CQ
1075 	 * create flags, otherwise it returns zero.
1076 	 */
1077 	return (flags & ~(IB_CQ_FLAGS_IGNORE_OVERRUN |
1078 			  IB_CQ_FLAGS_TIMESTAMP_COMPLETION));
1079 }
1080 
1081 static inline int verify_assign_uidx(u8 cqe_version, u32 cmd_uidx,
1082 				     u32 *user_index)
1083 {
1084 	if (cqe_version) {
1085 		if ((cmd_uidx == MLX5_IB_DEFAULT_UIDX) ||
1086 		    (cmd_uidx & ~MLX5_USER_ASSIGNED_UIDX_MASK))
1087 			return -EINVAL;
1088 		*user_index = cmd_uidx;
1089 	} else {
1090 		*user_index = MLX5_IB_DEFAULT_UIDX;
1091 	}
1092 
1093 	return 0;
1094 }
1095 
1096 static inline int get_qp_user_index(struct mlx5_ib_ucontext *ucontext,
1097 				    struct mlx5_ib_create_qp *ucmd,
1098 				    int inlen,
1099 				    u32 *user_index)
1100 {
1101 	u8 cqe_version = ucontext->cqe_version;
1102 
1103 	if (field_avail(struct mlx5_ib_create_qp, uidx, inlen) &&
1104 	    !cqe_version && (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1105 		return 0;
1106 
1107 	if (!!(field_avail(struct mlx5_ib_create_qp, uidx, inlen) !=
1108 	       !!cqe_version))
1109 		return -EINVAL;
1110 
1111 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1112 }
1113 
1114 static inline int get_srq_user_index(struct mlx5_ib_ucontext *ucontext,
1115 				     struct mlx5_ib_create_srq *ucmd,
1116 				     int inlen,
1117 				     u32 *user_index)
1118 {
1119 	u8 cqe_version = ucontext->cqe_version;
1120 
1121 	if (field_avail(struct mlx5_ib_create_srq, uidx, inlen) &&
1122 	    !cqe_version && (ucmd->uidx == MLX5_IB_DEFAULT_UIDX))
1123 		return 0;
1124 
1125 	if (!!(field_avail(struct mlx5_ib_create_srq, uidx, inlen) !=
1126 	       !!cqe_version))
1127 		return -EINVAL;
1128 
1129 	return verify_assign_uidx(cqe_version, ucmd->uidx, user_index);
1130 }
1131 
1132 void mlx5_ib_cleanup_congestion(struct mlx5_ib_dev *);
1133 int mlx5_ib_init_congestion(struct mlx5_ib_dev *);
1134 
1135 static inline int get_uars_per_sys_page(struct mlx5_ib_dev *dev, bool lib_support)
1136 {
1137 	return lib_support && MLX5_CAP_GEN(dev->mdev, uar_4k) ?
1138 				MLX5_UARS_IN_PAGE : 1;
1139 }
1140 
1141 static inline int get_num_static_uars(struct mlx5_ib_dev *dev,
1142 				      struct mlx5_bfreg_info *bfregi)
1143 {
1144 	return get_uars_per_sys_page(dev, bfregi->lib_uar_4k) * bfregi->num_static_sys_pages;
1145 }
1146 
1147 int bfregn_to_uar_index(struct mlx5_ib_dev *dev,
1148 			struct mlx5_bfreg_info *bfregi, u32 bfregn,
1149 			bool dyn_bfreg);
1150 
1151 #endif /* MLX5_IB_H */
1152