2 * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 #include <linux/kernel.h>
37 #include <linux/sched.h>
38 #include <rdma/ib_verbs.h>
39 #include <rdma/ib_smi.h>
40 #include <linux/mlx5/driver.h>
41 #include <linux/mlx5/cq.h>
42 #include <linux/mlx5/qp.h>
43 #include <linux/mlx5/srq.h>
44 #include <linux/types.h>
46 #define mlx5_ib_dbg(dev, format, arg...) \
47 pr_debug("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__, \
48 __LINE__, current->pid, ##arg)
50 #define mlx5_ib_err(dev, format, arg...) \
51 pr_err("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__, \
52 __LINE__, current->pid, ##arg)
54 #define mlx5_ib_warn(dev, format, arg...) \
55 pr_warn("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__, \
56 __LINE__, current->pid, ##arg)
59 MLX5_IB_MMAP_CMD_SHIFT
= 8,
60 MLX5_IB_MMAP_CMD_MASK
= 0xff,
63 enum mlx5_ib_mmap_cmd
{
64 MLX5_IB_MMAP_REGULAR_PAGE
= 0,
65 MLX5_IB_MMAP_GET_CONTIGUOUS_PAGES
= 1, /* always last */
69 MLX5_RES_SCAT_DATA32_CQE
= 0x1,
70 MLX5_RES_SCAT_DATA64_CQE
= 0x2,
71 MLX5_REQ_SCAT_DATA32_CQE
= 0x11,
72 MLX5_REQ_SCAT_DATA64_CQE
= 0x22,
75 enum mlx5_ib_latency_class
{
76 MLX5_IB_LATENCY_CLASS_LOW
,
77 MLX5_IB_LATENCY_CLASS_MEDIUM
,
78 MLX5_IB_LATENCY_CLASS_HIGH
,
79 MLX5_IB_LATENCY_CLASS_FAST_PATH
82 enum mlx5_ib_mad_ifc_flags
{
83 MLX5_MAD_IFC_IGNORE_MKEY
= 1,
84 MLX5_MAD_IFC_IGNORE_BKEY
= 2,
85 MLX5_MAD_IFC_NET_VIEW
= 4,
88 struct mlx5_ib_ucontext
{
89 struct ib_ucontext ibucontext
;
90 struct list_head db_page_list
;
92 /* protect doorbell record alloc/free
94 struct mutex db_page_mutex
;
95 struct mlx5_uuar_info uuari
;
98 static inline struct mlx5_ib_ucontext
*to_mucontext(struct ib_ucontext
*ibucontext
)
100 return container_of(ibucontext
, struct mlx5_ib_ucontext
, ibucontext
);
109 /* Use macros here so that don't have to duplicate
110 * enum ib_send_flags and enum ib_qp_type for low-level driver
113 #define MLX5_IB_SEND_UMR_UNREG IB_SEND_RESERVED_START
114 #define MLX5_IB_SEND_UMR_FAIL_IF_FREE (IB_SEND_RESERVED_START << 1)
115 #define MLX5_IB_SEND_UMR_UPDATE_MTT (IB_SEND_RESERVED_START << 2)
116 #define MLX5_IB_QPT_REG_UMR IB_QPT_RESERVED1
117 #define MLX5_IB_WR_UMR IB_WR_RESERVED1
127 struct wr_list
*w_list
;
131 /* serialize post to the work queue
153 * Connect-IB can trigger up to four concurrent pagefaults
156 enum mlx5_ib_pagefault_context
{
157 MLX5_IB_PAGEFAULT_RESPONDER_READ
,
158 MLX5_IB_PAGEFAULT_REQUESTOR_READ
,
159 MLX5_IB_PAGEFAULT_RESPONDER_WRITE
,
160 MLX5_IB_PAGEFAULT_REQUESTOR_WRITE
,
161 MLX5_IB_PAGEFAULT_CONTEXTS
164 static inline enum mlx5_ib_pagefault_context
165 mlx5_ib_get_pagefault_context(struct mlx5_pagefault
*pagefault
)
167 return pagefault
->flags
& (MLX5_PFAULT_REQUESTOR
| MLX5_PFAULT_WRITE
);
170 struct mlx5_ib_pfault
{
171 struct work_struct work
;
172 struct mlx5_pagefault mpfault
;
177 struct mlx5_core_qp mqp
;
181 struct mlx5_ib_wq rq
;
186 int sq_max_wqes_per_wr
;
188 struct mlx5_ib_wq sq
;
190 struct ib_umem
*umem
;
193 /* serialize qp state modifications
210 /* only for user space QPs. For kernel
211 * we have it from the bf object
218 /* Store signature errors */
221 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
223 * A flag that is true for QP's that are in a state that doesn't
224 * allow page faults, and shouldn't schedule any more faults.
226 int disable_page_faults
;
228 * The disable_page_faults_lock protects a QP's disable_page_faults
229 * field, allowing for a thread to atomically check whether the QP
230 * allows page faults, and if so schedule a page fault.
232 spinlock_t disable_page_faults_lock
;
233 struct mlx5_ib_pfault pagefaults
[MLX5_IB_PAGEFAULT_CONTEXTS
];
237 struct mlx5_ib_cq_buf
{
239 struct ib_umem
*umem
;
244 enum mlx5_ib_qp_flags
{
245 MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK
= 1 << 0,
246 MLX5_IB_QP_SIGNATURE_HANDLING
= 1 << 1,
255 unsigned int page_shift
;
262 struct mlx5_shared_mr_info
{
264 struct ib_umem
*umem
;
269 struct mlx5_core_cq mcq
;
270 struct mlx5_ib_cq_buf buf
;
273 /* serialize access to the CQ
279 struct mutex resize_mutex
;
280 struct mlx5_ib_cq_buf
*resize_buf
;
281 struct ib_umem
*resize_umem
;
287 struct mlx5_core_srq msrq
;
291 /* protect SRQ hanlding
297 struct ib_umem
*umem
;
298 /* serialize arming a SRQ
304 struct mlx5_ib_xrcd
{
305 struct ib_xrcd ibxrcd
;
309 enum mlx5_ib_mtt_access_flags
{
310 MLX5_IB_MTT_READ
= (1 << 0),
311 MLX5_IB_MTT_WRITE
= (1 << 1),
314 #define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
318 struct mlx5_core_mr mmr
;
319 struct ib_umem
*umem
;
320 struct mlx5_shared_mr_info
*smr_info
;
321 struct list_head list
;
325 struct mlx5_ib_dev
*dev
;
326 struct mlx5_create_mkey_mbox_out out
;
327 struct mlx5_core_sig_ctx
*sig
;
331 struct mlx5_ib_fast_reg_page_list
{
332 struct ib_fast_reg_page_list ibfrpl
;
333 __be64
*mapped_page_list
;
337 struct mlx5_ib_umr_context
{
338 enum ib_wc_status status
;
339 struct completion done
;
342 static inline void mlx5_ib_init_umr_context(struct mlx5_ib_umr_context
*context
)
344 context
->status
= -1;
345 init_completion(&context
->done
);
353 /* control access to UMR QP
355 struct semaphore sem
;
366 struct mlx5_core_mr mr
;
373 struct ib_send_wr wr
[2];
375 struct ib_fast_reg_page_list page_list
;
378 struct mlx5_cache_ent
{
379 struct list_head head
;
380 /* sync access to the cahce entry
393 struct dentry
*fsize
;
395 struct dentry
*fmiss
;
396 struct dentry
*flimit
;
398 struct mlx5_ib_dev
*dev
;
399 struct work_struct work
;
400 struct delayed_work dwork
;
404 struct mlx5_mr_cache
{
405 struct workqueue_struct
*wq
;
406 struct mlx5_cache_ent ent
[MAX_MR_CACHE_ENTRIES
];
409 unsigned long last_add
;
412 struct mlx5_ib_resources
{
421 struct ib_device ib_dev
;
422 struct mlx5_core_dev
*mdev
;
423 MLX5_DECLARE_DOORBELL_LOCK(uar_lock
);
425 /* serialize update of capability mask
427 struct mutex cap_mask_mutex
;
429 struct umr_common umrc
;
430 /* sync used page count stats
432 struct mlx5_ib_resources devr
;
433 struct mlx5_mr_cache cache
;
434 struct timer_list delay_timer
;
436 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
437 struct ib_odp_caps odp_caps
;
439 * Sleepable RCU that prevents destruction of MRs while they are still
440 * being used by a page fault handler.
442 struct srcu_struct mr_srcu
;
446 static inline struct mlx5_ib_cq
*to_mibcq(struct mlx5_core_cq
*mcq
)
448 return container_of(mcq
, struct mlx5_ib_cq
, mcq
);
451 static inline struct mlx5_ib_xrcd
*to_mxrcd(struct ib_xrcd
*ibxrcd
)
453 return container_of(ibxrcd
, struct mlx5_ib_xrcd
, ibxrcd
);
456 static inline struct mlx5_ib_dev
*to_mdev(struct ib_device
*ibdev
)
458 return container_of(ibdev
, struct mlx5_ib_dev
, ib_dev
);
461 static inline struct mlx5_ib_fmr
*to_mfmr(struct ib_fmr
*ibfmr
)
463 return container_of(ibfmr
, struct mlx5_ib_fmr
, ibfmr
);
466 static inline struct mlx5_ib_cq
*to_mcq(struct ib_cq
*ibcq
)
468 return container_of(ibcq
, struct mlx5_ib_cq
, ibcq
);
471 static inline struct mlx5_ib_qp
*to_mibqp(struct mlx5_core_qp
*mqp
)
473 return container_of(mqp
, struct mlx5_ib_qp
, mqp
);
476 static inline struct mlx5_ib_mr
*to_mibmr(struct mlx5_core_mr
*mmr
)
478 return container_of(mmr
, struct mlx5_ib_mr
, mmr
);
481 static inline struct mlx5_ib_pd
*to_mpd(struct ib_pd
*ibpd
)
483 return container_of(ibpd
, struct mlx5_ib_pd
, ibpd
);
486 static inline struct mlx5_ib_srq
*to_msrq(struct ib_srq
*ibsrq
)
488 return container_of(ibsrq
, struct mlx5_ib_srq
, ibsrq
);
491 static inline struct mlx5_ib_qp
*to_mqp(struct ib_qp
*ibqp
)
493 return container_of(ibqp
, struct mlx5_ib_qp
, ibqp
);
496 static inline struct mlx5_ib_srq
*to_mibsrq(struct mlx5_core_srq
*msrq
)
498 return container_of(msrq
, struct mlx5_ib_srq
, msrq
);
501 static inline struct mlx5_ib_mr
*to_mmr(struct ib_mr
*ibmr
)
503 return container_of(ibmr
, struct mlx5_ib_mr
, ibmr
);
506 static inline struct mlx5_ib_fast_reg_page_list
*to_mfrpl(struct ib_fast_reg_page_list
*ibfrpl
)
508 return container_of(ibfrpl
, struct mlx5_ib_fast_reg_page_list
, ibfrpl
);
516 static inline struct mlx5_ib_ah
*to_mah(struct ib_ah
*ibah
)
518 return container_of(ibah
, struct mlx5_ib_ah
, ibah
);
521 int mlx5_ib_db_map_user(struct mlx5_ib_ucontext
*context
, unsigned long virt
,
523 void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext
*context
, struct mlx5_db
*db
);
524 void __mlx5_ib_cq_clean(struct mlx5_ib_cq
*cq
, u32 qpn
, struct mlx5_ib_srq
*srq
);
525 void mlx5_ib_cq_clean(struct mlx5_ib_cq
*cq
, u32 qpn
, struct mlx5_ib_srq
*srq
);
526 void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq
*srq
, int wqe_index
);
527 int mlx5_MAD_IFC(struct mlx5_ib_dev
*dev
, int ignore_mkey
, int ignore_bkey
,
528 u8 port
, struct ib_wc
*in_wc
, struct ib_grh
*in_grh
,
529 void *in_mad
, void *response_mad
);
530 struct ib_ah
*create_ib_ah(struct ib_ah_attr
*ah_attr
,
531 struct mlx5_ib_ah
*ah
);
532 struct ib_ah
*mlx5_ib_create_ah(struct ib_pd
*pd
, struct ib_ah_attr
*ah_attr
);
533 int mlx5_ib_query_ah(struct ib_ah
*ibah
, struct ib_ah_attr
*ah_attr
);
534 int mlx5_ib_destroy_ah(struct ib_ah
*ah
);
535 struct ib_srq
*mlx5_ib_create_srq(struct ib_pd
*pd
,
536 struct ib_srq_init_attr
*init_attr
,
537 struct ib_udata
*udata
);
538 int mlx5_ib_modify_srq(struct ib_srq
*ibsrq
, struct ib_srq_attr
*attr
,
539 enum ib_srq_attr_mask attr_mask
, struct ib_udata
*udata
);
540 int mlx5_ib_query_srq(struct ib_srq
*ibsrq
, struct ib_srq_attr
*srq_attr
);
541 int mlx5_ib_destroy_srq(struct ib_srq
*srq
);
542 int mlx5_ib_post_srq_recv(struct ib_srq
*ibsrq
, struct ib_recv_wr
*wr
,
543 struct ib_recv_wr
**bad_wr
);
544 struct ib_qp
*mlx5_ib_create_qp(struct ib_pd
*pd
,
545 struct ib_qp_init_attr
*init_attr
,
546 struct ib_udata
*udata
);
547 int mlx5_ib_modify_qp(struct ib_qp
*ibqp
, struct ib_qp_attr
*attr
,
548 int attr_mask
, struct ib_udata
*udata
);
549 int mlx5_ib_query_qp(struct ib_qp
*ibqp
, struct ib_qp_attr
*qp_attr
, int qp_attr_mask
,
550 struct ib_qp_init_attr
*qp_init_attr
);
551 int mlx5_ib_destroy_qp(struct ib_qp
*qp
);
552 int mlx5_ib_post_send(struct ib_qp
*ibqp
, struct ib_send_wr
*wr
,
553 struct ib_send_wr
**bad_wr
);
554 int mlx5_ib_post_recv(struct ib_qp
*ibqp
, struct ib_recv_wr
*wr
,
555 struct ib_recv_wr
**bad_wr
);
556 void *mlx5_get_send_wqe(struct mlx5_ib_qp
*qp
, int n
);
557 int mlx5_ib_read_user_wqe(struct mlx5_ib_qp
*qp
, int send
, int wqe_index
,
558 void *buffer
, u32 length
);
559 struct ib_cq
*mlx5_ib_create_cq(struct ib_device
*ibdev
, int entries
,
560 int vector
, struct ib_ucontext
*context
,
561 struct ib_udata
*udata
);
562 int mlx5_ib_destroy_cq(struct ib_cq
*cq
);
563 int mlx5_ib_poll_cq(struct ib_cq
*ibcq
, int num_entries
, struct ib_wc
*wc
);
564 int mlx5_ib_arm_cq(struct ib_cq
*ibcq
, enum ib_cq_notify_flags flags
);
565 int mlx5_ib_modify_cq(struct ib_cq
*cq
, u16 cq_count
, u16 cq_period
);
566 int mlx5_ib_resize_cq(struct ib_cq
*ibcq
, int entries
, struct ib_udata
*udata
);
567 struct ib_mr
*mlx5_ib_get_dma_mr(struct ib_pd
*pd
, int acc
);
568 struct ib_mr
*mlx5_ib_reg_user_mr(struct ib_pd
*pd
, u64 start
, u64 length
,
569 u64 virt_addr
, int access_flags
,
570 struct ib_udata
*udata
);
571 int mlx5_ib_update_mtt(struct mlx5_ib_mr
*mr
, u64 start_page_index
,
572 int npages
, int zap
);
573 int mlx5_ib_dereg_mr(struct ib_mr
*ibmr
);
574 int mlx5_ib_destroy_mr(struct ib_mr
*ibmr
);
575 struct ib_mr
*mlx5_ib_create_mr(struct ib_pd
*pd
,
576 struct ib_mr_init_attr
*mr_init_attr
);
577 struct ib_mr
*mlx5_ib_alloc_fast_reg_mr(struct ib_pd
*pd
,
578 int max_page_list_len
);
579 struct ib_fast_reg_page_list
*mlx5_ib_alloc_fast_reg_page_list(struct ib_device
*ibdev
,
581 void mlx5_ib_free_fast_reg_page_list(struct ib_fast_reg_page_list
*page_list
);
582 struct ib_fmr
*mlx5_ib_fmr_alloc(struct ib_pd
*pd
, int acc
,
583 struct ib_fmr_attr
*fmr_attr
);
584 int mlx5_ib_map_phys_fmr(struct ib_fmr
*ibfmr
, u64
*page_list
,
585 int npages
, u64 iova
);
586 int mlx5_ib_unmap_fmr(struct list_head
*fmr_list
);
587 int mlx5_ib_fmr_dealloc(struct ib_fmr
*ibfmr
);
588 int mlx5_ib_process_mad(struct ib_device
*ibdev
, int mad_flags
, u8 port_num
,
589 struct ib_wc
*in_wc
, struct ib_grh
*in_grh
,
590 struct ib_mad
*in_mad
, struct ib_mad
*out_mad
);
591 struct ib_xrcd
*mlx5_ib_alloc_xrcd(struct ib_device
*ibdev
,
592 struct ib_ucontext
*context
,
593 struct ib_udata
*udata
);
594 int mlx5_ib_dealloc_xrcd(struct ib_xrcd
*xrcd
);
595 int mlx5_ib_get_buf_offset(u64 addr
, int page_shift
, u32
*offset
);
596 int mlx5_query_ext_port_caps(struct mlx5_ib_dev
*dev
, u8 port
);
597 int mlx5_ib_query_port(struct ib_device
*ibdev
, u8 port
,
598 struct ib_port_attr
*props
);
599 int mlx5_ib_init_fmr(struct mlx5_ib_dev
*dev
);
600 void mlx5_ib_cleanup_fmr(struct mlx5_ib_dev
*dev
);
601 void mlx5_ib_cont_pages(struct ib_umem
*umem
, u64 addr
, int *count
, int *shift
,
602 int *ncont
, int *order
);
603 void __mlx5_ib_populate_pas(struct mlx5_ib_dev
*dev
, struct ib_umem
*umem
,
604 int page_shift
, size_t offset
, size_t num_pages
,
605 __be64
*pas
, int access_flags
);
606 void mlx5_ib_populate_pas(struct mlx5_ib_dev
*dev
, struct ib_umem
*umem
,
607 int page_shift
, __be64
*pas
, int access_flags
);
608 void mlx5_ib_copy_pas(u64
*old
, u64
*new, int step
, int num
);
609 int mlx5_ib_get_cqe_size(struct mlx5_ib_dev
*dev
, struct ib_cq
*ibcq
);
610 int mlx5_mr_cache_init(struct mlx5_ib_dev
*dev
);
611 int mlx5_mr_cache_cleanup(struct mlx5_ib_dev
*dev
);
612 int mlx5_mr_ib_cont_pages(struct ib_umem
*umem
, u64 addr
, int *count
, int *shift
);
613 void mlx5_umr_cq_handler(struct ib_cq
*cq
, void *cq_context
);
614 int mlx5_ib_check_mr_status(struct ib_mr
*ibmr
, u32 check_mask
,
615 struct ib_mr_status
*mr_status
);
617 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
618 extern struct workqueue_struct
*mlx5_ib_page_fault_wq
;
620 int mlx5_ib_internal_query_odp_caps(struct mlx5_ib_dev
*dev
);
621 void mlx5_ib_mr_pfault_handler(struct mlx5_ib_qp
*qp
,
622 struct mlx5_ib_pfault
*pfault
);
623 void mlx5_ib_odp_create_qp(struct mlx5_ib_qp
*qp
);
624 int mlx5_ib_odp_init_one(struct mlx5_ib_dev
*ibdev
);
625 void mlx5_ib_odp_remove_one(struct mlx5_ib_dev
*ibdev
);
626 int __init
mlx5_ib_odp_init(void);
627 void mlx5_ib_odp_cleanup(void);
628 void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp
*qp
);
629 void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp
*qp
);
630 void mlx5_ib_invalidate_range(struct ib_umem
*umem
, unsigned long start
,
633 #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
634 static inline int mlx5_ib_internal_query_odp_caps(struct mlx5_ib_dev
*dev
)
639 static inline void mlx5_ib_odp_create_qp(struct mlx5_ib_qp
*qp
) {}
640 static inline int mlx5_ib_odp_init_one(struct mlx5_ib_dev
*ibdev
) { return 0; }
641 static inline void mlx5_ib_odp_remove_one(struct mlx5_ib_dev
*ibdev
) {}
642 static inline int mlx5_ib_odp_init(void) { return 0; }
643 static inline void mlx5_ib_odp_cleanup(void) {}
644 static inline void mlx5_ib_qp_disable_pagefaults(struct mlx5_ib_qp
*qp
) {}
645 static inline void mlx5_ib_qp_enable_pagefaults(struct mlx5_ib_qp
*qp
) {}
647 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
649 static inline void init_query_mad(struct ib_smp
*mad
)
651 mad
->base_version
= 1;
652 mad
->mgmt_class
= IB_MGMT_CLASS_SUBN_LID_ROUTED
;
653 mad
->class_version
= 1;
654 mad
->method
= IB_MGMT_METHOD_GET
;
657 static inline u8
convert_access(int acc
)
659 return (acc
& IB_ACCESS_REMOTE_ATOMIC
? MLX5_PERM_ATOMIC
: 0) |
660 (acc
& IB_ACCESS_REMOTE_WRITE
? MLX5_PERM_REMOTE_WRITE
: 0) |
661 (acc
& IB_ACCESS_REMOTE_READ
? MLX5_PERM_REMOTE_READ
: 0) |
662 (acc
& IB_ACCESS_LOCAL_WRITE
? MLX5_PERM_LOCAL_WRITE
: 0) |
663 MLX5_PERM_LOCAL_READ
;
666 #define MLX5_MAX_UMR_SHIFT 16
667 #define MLX5_MAX_UMR_PAGES (1 << MLX5_MAX_UMR_SHIFT)
669 #endif /* MLX5_IB_H */