fs: use kmem_cache_zalloc instead
[pv_ops_mirror.git] / include / rdma / ib_verbs.h
blob4bea182d7116c4aa80e863ed2d5a59fed24967e0
1 /*
2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
18 * conditions are met:
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer.
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36 * SOFTWARE.
38 * $Id: ib_verbs.h 1349 2004-12-16 21:09:43Z roland $
41 #if !defined(IB_VERBS_H)
42 #define IB_VERBS_H
44 #include <linux/types.h>
45 #include <linux/device.h>
46 #include <linux/mm.h>
47 #include <linux/dma-mapping.h>
48 #include <linux/kref.h>
49 #include <linux/list.h>
50 #include <linux/rwsem.h>
52 #include <asm/atomic.h>
53 #include <asm/scatterlist.h>
54 #include <asm/uaccess.h>
56 union ib_gid {
57 u8 raw[16];
58 struct {
59 __be64 subnet_prefix;
60 __be64 interface_id;
61 } global;
64 enum rdma_node_type {
65 /* IB values map to NodeInfo:NodeType. */
66 RDMA_NODE_IB_CA = 1,
67 RDMA_NODE_IB_SWITCH,
68 RDMA_NODE_IB_ROUTER,
69 RDMA_NODE_RNIC
72 enum rdma_transport_type {
73 RDMA_TRANSPORT_IB,
74 RDMA_TRANSPORT_IWARP
77 enum rdma_transport_type
78 rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__;
80 enum ib_device_cap_flags {
81 IB_DEVICE_RESIZE_MAX_WR = 1,
82 IB_DEVICE_BAD_PKEY_CNTR = (1<<1),
83 IB_DEVICE_BAD_QKEY_CNTR = (1<<2),
84 IB_DEVICE_RAW_MULTI = (1<<3),
85 IB_DEVICE_AUTO_PATH_MIG = (1<<4),
86 IB_DEVICE_CHANGE_PHY_PORT = (1<<5),
87 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6),
88 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7),
89 IB_DEVICE_SHUTDOWN_PORT = (1<<8),
90 IB_DEVICE_INIT_TYPE = (1<<9),
91 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10),
92 IB_DEVICE_SYS_IMAGE_GUID = (1<<11),
93 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12),
94 IB_DEVICE_SRQ_RESIZE = (1<<13),
95 IB_DEVICE_N_NOTIFY_CQ = (1<<14),
96 IB_DEVICE_ZERO_STAG = (1<<15),
97 IB_DEVICE_SEND_W_INV = (1<<16),
98 IB_DEVICE_MEM_WINDOW = (1<<17)
101 enum ib_atomic_cap {
102 IB_ATOMIC_NONE,
103 IB_ATOMIC_HCA,
104 IB_ATOMIC_GLOB
107 struct ib_device_attr {
108 u64 fw_ver;
109 __be64 sys_image_guid;
110 u64 max_mr_size;
111 u64 page_size_cap;
112 u32 vendor_id;
113 u32 vendor_part_id;
114 u32 hw_ver;
115 int max_qp;
116 int max_qp_wr;
117 int device_cap_flags;
118 int max_sge;
119 int max_sge_rd;
120 int max_cq;
121 int max_cqe;
122 int max_mr;
123 int max_pd;
124 int max_qp_rd_atom;
125 int max_ee_rd_atom;
126 int max_res_rd_atom;
127 int max_qp_init_rd_atom;
128 int max_ee_init_rd_atom;
129 enum ib_atomic_cap atomic_cap;
130 int max_ee;
131 int max_rdd;
132 int max_mw;
133 int max_raw_ipv6_qp;
134 int max_raw_ethy_qp;
135 int max_mcast_grp;
136 int max_mcast_qp_attach;
137 int max_total_mcast_qp_attach;
138 int max_ah;
139 int max_fmr;
140 int max_map_per_fmr;
141 int max_srq;
142 int max_srq_wr;
143 int max_srq_sge;
144 u16 max_pkeys;
145 u8 local_ca_ack_delay;
148 enum ib_mtu {
149 IB_MTU_256 = 1,
150 IB_MTU_512 = 2,
151 IB_MTU_1024 = 3,
152 IB_MTU_2048 = 4,
153 IB_MTU_4096 = 5
156 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
158 switch (mtu) {
159 case IB_MTU_256: return 256;
160 case IB_MTU_512: return 512;
161 case IB_MTU_1024: return 1024;
162 case IB_MTU_2048: return 2048;
163 case IB_MTU_4096: return 4096;
164 default: return -1;
168 enum ib_port_state {
169 IB_PORT_NOP = 0,
170 IB_PORT_DOWN = 1,
171 IB_PORT_INIT = 2,
172 IB_PORT_ARMED = 3,
173 IB_PORT_ACTIVE = 4,
174 IB_PORT_ACTIVE_DEFER = 5
177 enum ib_port_cap_flags {
178 IB_PORT_SM = 1 << 1,
179 IB_PORT_NOTICE_SUP = 1 << 2,
180 IB_PORT_TRAP_SUP = 1 << 3,
181 IB_PORT_OPT_IPD_SUP = 1 << 4,
182 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
183 IB_PORT_SL_MAP_SUP = 1 << 6,
184 IB_PORT_MKEY_NVRAM = 1 << 7,
185 IB_PORT_PKEY_NVRAM = 1 << 8,
186 IB_PORT_LED_INFO_SUP = 1 << 9,
187 IB_PORT_SM_DISABLED = 1 << 10,
188 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
189 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
190 IB_PORT_CM_SUP = 1 << 16,
191 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
192 IB_PORT_REINIT_SUP = 1 << 18,
193 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
194 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
195 IB_PORT_DR_NOTICE_SUP = 1 << 21,
196 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
197 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
198 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
199 IB_PORT_CLIENT_REG_SUP = 1 << 25
202 enum ib_port_width {
203 IB_WIDTH_1X = 1,
204 IB_WIDTH_4X = 2,
205 IB_WIDTH_8X = 4,
206 IB_WIDTH_12X = 8
209 static inline int ib_width_enum_to_int(enum ib_port_width width)
211 switch (width) {
212 case IB_WIDTH_1X: return 1;
213 case IB_WIDTH_4X: return 4;
214 case IB_WIDTH_8X: return 8;
215 case IB_WIDTH_12X: return 12;
216 default: return -1;
220 struct ib_port_attr {
221 enum ib_port_state state;
222 enum ib_mtu max_mtu;
223 enum ib_mtu active_mtu;
224 int gid_tbl_len;
225 u32 port_cap_flags;
226 u32 max_msg_sz;
227 u32 bad_pkey_cntr;
228 u32 qkey_viol_cntr;
229 u16 pkey_tbl_len;
230 u16 lid;
231 u16 sm_lid;
232 u8 lmc;
233 u8 max_vl_num;
234 u8 sm_sl;
235 u8 subnet_timeout;
236 u8 init_type_reply;
237 u8 active_width;
238 u8 active_speed;
239 u8 phys_state;
242 enum ib_device_modify_flags {
243 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
244 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
247 struct ib_device_modify {
248 u64 sys_image_guid;
249 char node_desc[64];
252 enum ib_port_modify_flags {
253 IB_PORT_SHUTDOWN = 1,
254 IB_PORT_INIT_TYPE = (1<<2),
255 IB_PORT_RESET_QKEY_CNTR = (1<<3)
258 struct ib_port_modify {
259 u32 set_port_cap_mask;
260 u32 clr_port_cap_mask;
261 u8 init_type;
264 enum ib_event_type {
265 IB_EVENT_CQ_ERR,
266 IB_EVENT_QP_FATAL,
267 IB_EVENT_QP_REQ_ERR,
268 IB_EVENT_QP_ACCESS_ERR,
269 IB_EVENT_COMM_EST,
270 IB_EVENT_SQ_DRAINED,
271 IB_EVENT_PATH_MIG,
272 IB_EVENT_PATH_MIG_ERR,
273 IB_EVENT_DEVICE_FATAL,
274 IB_EVENT_PORT_ACTIVE,
275 IB_EVENT_PORT_ERR,
276 IB_EVENT_LID_CHANGE,
277 IB_EVENT_PKEY_CHANGE,
278 IB_EVENT_SM_CHANGE,
279 IB_EVENT_SRQ_ERR,
280 IB_EVENT_SRQ_LIMIT_REACHED,
281 IB_EVENT_QP_LAST_WQE_REACHED,
282 IB_EVENT_CLIENT_REREGISTER
285 struct ib_event {
286 struct ib_device *device;
287 union {
288 struct ib_cq *cq;
289 struct ib_qp *qp;
290 struct ib_srq *srq;
291 u8 port_num;
292 } element;
293 enum ib_event_type event;
296 struct ib_event_handler {
297 struct ib_device *device;
298 void (*handler)(struct ib_event_handler *, struct ib_event *);
299 struct list_head list;
302 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
303 do { \
304 (_ptr)->device = _device; \
305 (_ptr)->handler = _handler; \
306 INIT_LIST_HEAD(&(_ptr)->list); \
307 } while (0)
309 struct ib_global_route {
310 union ib_gid dgid;
311 u32 flow_label;
312 u8 sgid_index;
313 u8 hop_limit;
314 u8 traffic_class;
317 struct ib_grh {
318 __be32 version_tclass_flow;
319 __be16 paylen;
320 u8 next_hdr;
321 u8 hop_limit;
322 union ib_gid sgid;
323 union ib_gid dgid;
326 enum {
327 IB_MULTICAST_QPN = 0xffffff
330 #define IB_LID_PERMISSIVE __constant_htons(0xFFFF)
332 enum ib_ah_flags {
333 IB_AH_GRH = 1
336 enum ib_rate {
337 IB_RATE_PORT_CURRENT = 0,
338 IB_RATE_2_5_GBPS = 2,
339 IB_RATE_5_GBPS = 5,
340 IB_RATE_10_GBPS = 3,
341 IB_RATE_20_GBPS = 6,
342 IB_RATE_30_GBPS = 4,
343 IB_RATE_40_GBPS = 7,
344 IB_RATE_60_GBPS = 8,
345 IB_RATE_80_GBPS = 9,
346 IB_RATE_120_GBPS = 10
350 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
351 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
352 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
353 * @rate: rate to convert.
355 int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
358 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
359 * enum.
360 * @mult: multiple to convert.
362 enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
364 struct ib_ah_attr {
365 struct ib_global_route grh;
366 u16 dlid;
367 u8 sl;
368 u8 src_path_bits;
369 u8 static_rate;
370 u8 ah_flags;
371 u8 port_num;
374 enum ib_wc_status {
375 IB_WC_SUCCESS,
376 IB_WC_LOC_LEN_ERR,
377 IB_WC_LOC_QP_OP_ERR,
378 IB_WC_LOC_EEC_OP_ERR,
379 IB_WC_LOC_PROT_ERR,
380 IB_WC_WR_FLUSH_ERR,
381 IB_WC_MW_BIND_ERR,
382 IB_WC_BAD_RESP_ERR,
383 IB_WC_LOC_ACCESS_ERR,
384 IB_WC_REM_INV_REQ_ERR,
385 IB_WC_REM_ACCESS_ERR,
386 IB_WC_REM_OP_ERR,
387 IB_WC_RETRY_EXC_ERR,
388 IB_WC_RNR_RETRY_EXC_ERR,
389 IB_WC_LOC_RDD_VIOL_ERR,
390 IB_WC_REM_INV_RD_REQ_ERR,
391 IB_WC_REM_ABORT_ERR,
392 IB_WC_INV_EECN_ERR,
393 IB_WC_INV_EEC_STATE_ERR,
394 IB_WC_FATAL_ERR,
395 IB_WC_RESP_TIMEOUT_ERR,
396 IB_WC_GENERAL_ERR
399 enum ib_wc_opcode {
400 IB_WC_SEND,
401 IB_WC_RDMA_WRITE,
402 IB_WC_RDMA_READ,
403 IB_WC_COMP_SWAP,
404 IB_WC_FETCH_ADD,
405 IB_WC_BIND_MW,
407 * Set value of IB_WC_RECV so consumers can test if a completion is a
408 * receive by testing (opcode & IB_WC_RECV).
410 IB_WC_RECV = 1 << 7,
411 IB_WC_RECV_RDMA_WITH_IMM
414 enum ib_wc_flags {
415 IB_WC_GRH = 1,
416 IB_WC_WITH_IMM = (1<<1)
419 struct ib_wc {
420 u64 wr_id;
421 enum ib_wc_status status;
422 enum ib_wc_opcode opcode;
423 u32 vendor_err;
424 u32 byte_len;
425 struct ib_qp *qp;
426 __be32 imm_data;
427 u32 src_qp;
428 int wc_flags;
429 u16 pkey_index;
430 u16 slid;
431 u8 sl;
432 u8 dlid_path_bits;
433 u8 port_num; /* valid only for DR SMPs on switches */
436 enum ib_cq_notify_flags {
437 IB_CQ_SOLICITED = 1 << 0,
438 IB_CQ_NEXT_COMP = 1 << 1,
439 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
440 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
443 enum ib_srq_attr_mask {
444 IB_SRQ_MAX_WR = 1 << 0,
445 IB_SRQ_LIMIT = 1 << 1,
448 struct ib_srq_attr {
449 u32 max_wr;
450 u32 max_sge;
451 u32 srq_limit;
454 struct ib_srq_init_attr {
455 void (*event_handler)(struct ib_event *, void *);
456 void *srq_context;
457 struct ib_srq_attr attr;
460 struct ib_qp_cap {
461 u32 max_send_wr;
462 u32 max_recv_wr;
463 u32 max_send_sge;
464 u32 max_recv_sge;
465 u32 max_inline_data;
468 enum ib_sig_type {
469 IB_SIGNAL_ALL_WR,
470 IB_SIGNAL_REQ_WR
473 enum ib_qp_type {
475 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
476 * here (and in that order) since the MAD layer uses them as
477 * indices into a 2-entry table.
479 IB_QPT_SMI,
480 IB_QPT_GSI,
482 IB_QPT_RC,
483 IB_QPT_UC,
484 IB_QPT_UD,
485 IB_QPT_RAW_IPV6,
486 IB_QPT_RAW_ETY
489 struct ib_qp_init_attr {
490 void (*event_handler)(struct ib_event *, void *);
491 void *qp_context;
492 struct ib_cq *send_cq;
493 struct ib_cq *recv_cq;
494 struct ib_srq *srq;
495 struct ib_qp_cap cap;
496 enum ib_sig_type sq_sig_type;
497 enum ib_qp_type qp_type;
498 u8 port_num; /* special QP types only */
501 enum ib_rnr_timeout {
502 IB_RNR_TIMER_655_36 = 0,
503 IB_RNR_TIMER_000_01 = 1,
504 IB_RNR_TIMER_000_02 = 2,
505 IB_RNR_TIMER_000_03 = 3,
506 IB_RNR_TIMER_000_04 = 4,
507 IB_RNR_TIMER_000_06 = 5,
508 IB_RNR_TIMER_000_08 = 6,
509 IB_RNR_TIMER_000_12 = 7,
510 IB_RNR_TIMER_000_16 = 8,
511 IB_RNR_TIMER_000_24 = 9,
512 IB_RNR_TIMER_000_32 = 10,
513 IB_RNR_TIMER_000_48 = 11,
514 IB_RNR_TIMER_000_64 = 12,
515 IB_RNR_TIMER_000_96 = 13,
516 IB_RNR_TIMER_001_28 = 14,
517 IB_RNR_TIMER_001_92 = 15,
518 IB_RNR_TIMER_002_56 = 16,
519 IB_RNR_TIMER_003_84 = 17,
520 IB_RNR_TIMER_005_12 = 18,
521 IB_RNR_TIMER_007_68 = 19,
522 IB_RNR_TIMER_010_24 = 20,
523 IB_RNR_TIMER_015_36 = 21,
524 IB_RNR_TIMER_020_48 = 22,
525 IB_RNR_TIMER_030_72 = 23,
526 IB_RNR_TIMER_040_96 = 24,
527 IB_RNR_TIMER_061_44 = 25,
528 IB_RNR_TIMER_081_92 = 26,
529 IB_RNR_TIMER_122_88 = 27,
530 IB_RNR_TIMER_163_84 = 28,
531 IB_RNR_TIMER_245_76 = 29,
532 IB_RNR_TIMER_327_68 = 30,
533 IB_RNR_TIMER_491_52 = 31
536 enum ib_qp_attr_mask {
537 IB_QP_STATE = 1,
538 IB_QP_CUR_STATE = (1<<1),
539 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
540 IB_QP_ACCESS_FLAGS = (1<<3),
541 IB_QP_PKEY_INDEX = (1<<4),
542 IB_QP_PORT = (1<<5),
543 IB_QP_QKEY = (1<<6),
544 IB_QP_AV = (1<<7),
545 IB_QP_PATH_MTU = (1<<8),
546 IB_QP_TIMEOUT = (1<<9),
547 IB_QP_RETRY_CNT = (1<<10),
548 IB_QP_RNR_RETRY = (1<<11),
549 IB_QP_RQ_PSN = (1<<12),
550 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
551 IB_QP_ALT_PATH = (1<<14),
552 IB_QP_MIN_RNR_TIMER = (1<<15),
553 IB_QP_SQ_PSN = (1<<16),
554 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
555 IB_QP_PATH_MIG_STATE = (1<<18),
556 IB_QP_CAP = (1<<19),
557 IB_QP_DEST_QPN = (1<<20)
560 enum ib_qp_state {
561 IB_QPS_RESET,
562 IB_QPS_INIT,
563 IB_QPS_RTR,
564 IB_QPS_RTS,
565 IB_QPS_SQD,
566 IB_QPS_SQE,
567 IB_QPS_ERR
570 enum ib_mig_state {
571 IB_MIG_MIGRATED,
572 IB_MIG_REARM,
573 IB_MIG_ARMED
576 struct ib_qp_attr {
577 enum ib_qp_state qp_state;
578 enum ib_qp_state cur_qp_state;
579 enum ib_mtu path_mtu;
580 enum ib_mig_state path_mig_state;
581 u32 qkey;
582 u32 rq_psn;
583 u32 sq_psn;
584 u32 dest_qp_num;
585 int qp_access_flags;
586 struct ib_qp_cap cap;
587 struct ib_ah_attr ah_attr;
588 struct ib_ah_attr alt_ah_attr;
589 u16 pkey_index;
590 u16 alt_pkey_index;
591 u8 en_sqd_async_notify;
592 u8 sq_draining;
593 u8 max_rd_atomic;
594 u8 max_dest_rd_atomic;
595 u8 min_rnr_timer;
596 u8 port_num;
597 u8 timeout;
598 u8 retry_cnt;
599 u8 rnr_retry;
600 u8 alt_port_num;
601 u8 alt_timeout;
604 enum ib_wr_opcode {
605 IB_WR_RDMA_WRITE,
606 IB_WR_RDMA_WRITE_WITH_IMM,
607 IB_WR_SEND,
608 IB_WR_SEND_WITH_IMM,
609 IB_WR_RDMA_READ,
610 IB_WR_ATOMIC_CMP_AND_SWP,
611 IB_WR_ATOMIC_FETCH_AND_ADD
614 enum ib_send_flags {
615 IB_SEND_FENCE = 1,
616 IB_SEND_SIGNALED = (1<<1),
617 IB_SEND_SOLICITED = (1<<2),
618 IB_SEND_INLINE = (1<<3)
621 struct ib_sge {
622 u64 addr;
623 u32 length;
624 u32 lkey;
627 struct ib_send_wr {
628 struct ib_send_wr *next;
629 u64 wr_id;
630 struct ib_sge *sg_list;
631 int num_sge;
632 enum ib_wr_opcode opcode;
633 int send_flags;
634 __be32 imm_data;
635 union {
636 struct {
637 u64 remote_addr;
638 u32 rkey;
639 } rdma;
640 struct {
641 u64 remote_addr;
642 u64 compare_add;
643 u64 swap;
644 u32 rkey;
645 } atomic;
646 struct {
647 struct ib_ah *ah;
648 u32 remote_qpn;
649 u32 remote_qkey;
650 u16 pkey_index; /* valid for GSI only */
651 u8 port_num; /* valid for DR SMPs on switch only */
652 } ud;
653 } wr;
656 struct ib_recv_wr {
657 struct ib_recv_wr *next;
658 u64 wr_id;
659 struct ib_sge *sg_list;
660 int num_sge;
663 enum ib_access_flags {
664 IB_ACCESS_LOCAL_WRITE = 1,
665 IB_ACCESS_REMOTE_WRITE = (1<<1),
666 IB_ACCESS_REMOTE_READ = (1<<2),
667 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
668 IB_ACCESS_MW_BIND = (1<<4)
671 struct ib_phys_buf {
672 u64 addr;
673 u64 size;
676 struct ib_mr_attr {
677 struct ib_pd *pd;
678 u64 device_virt_addr;
679 u64 size;
680 int mr_access_flags;
681 u32 lkey;
682 u32 rkey;
685 enum ib_mr_rereg_flags {
686 IB_MR_REREG_TRANS = 1,
687 IB_MR_REREG_PD = (1<<1),
688 IB_MR_REREG_ACCESS = (1<<2)
691 struct ib_mw_bind {
692 struct ib_mr *mr;
693 u64 wr_id;
694 u64 addr;
695 u32 length;
696 int send_flags;
697 int mw_access_flags;
700 struct ib_fmr_attr {
701 int max_pages;
702 int max_maps;
703 u8 page_shift;
706 struct ib_ucontext {
707 struct ib_device *device;
708 struct list_head pd_list;
709 struct list_head mr_list;
710 struct list_head mw_list;
711 struct list_head cq_list;
712 struct list_head qp_list;
713 struct list_head srq_list;
714 struct list_head ah_list;
715 int closing;
718 struct ib_uobject {
719 u64 user_handle; /* handle given to us by userspace */
720 struct ib_ucontext *context; /* associated user context */
721 void *object; /* containing object */
722 struct list_head list; /* link to context's list */
723 u32 id; /* index into kernel idr */
724 struct kref ref;
725 struct rw_semaphore mutex; /* protects .live */
726 int live;
729 struct ib_udata {
730 void __user *inbuf;
731 void __user *outbuf;
732 size_t inlen;
733 size_t outlen;
736 struct ib_pd {
737 struct ib_device *device;
738 struct ib_uobject *uobject;
739 atomic_t usecnt; /* count all resources */
742 struct ib_ah {
743 struct ib_device *device;
744 struct ib_pd *pd;
745 struct ib_uobject *uobject;
748 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
750 struct ib_cq {
751 struct ib_device *device;
752 struct ib_uobject *uobject;
753 ib_comp_handler comp_handler;
754 void (*event_handler)(struct ib_event *, void *);
755 void * cq_context;
756 int cqe;
757 atomic_t usecnt; /* count number of work queues */
760 struct ib_srq {
761 struct ib_device *device;
762 struct ib_pd *pd;
763 struct ib_uobject *uobject;
764 void (*event_handler)(struct ib_event *, void *);
765 void *srq_context;
766 atomic_t usecnt;
769 struct ib_qp {
770 struct ib_device *device;
771 struct ib_pd *pd;
772 struct ib_cq *send_cq;
773 struct ib_cq *recv_cq;
774 struct ib_srq *srq;
775 struct ib_uobject *uobject;
776 void (*event_handler)(struct ib_event *, void *);
777 void *qp_context;
778 u32 qp_num;
779 enum ib_qp_type qp_type;
782 struct ib_mr {
783 struct ib_device *device;
784 struct ib_pd *pd;
785 struct ib_uobject *uobject;
786 u32 lkey;
787 u32 rkey;
788 atomic_t usecnt; /* count number of MWs */
791 struct ib_mw {
792 struct ib_device *device;
793 struct ib_pd *pd;
794 struct ib_uobject *uobject;
795 u32 rkey;
798 struct ib_fmr {
799 struct ib_device *device;
800 struct ib_pd *pd;
801 struct list_head list;
802 u32 lkey;
803 u32 rkey;
806 struct ib_mad;
807 struct ib_grh;
809 enum ib_process_mad_flags {
810 IB_MAD_IGNORE_MKEY = 1,
811 IB_MAD_IGNORE_BKEY = 2,
812 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
815 enum ib_mad_result {
816 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
817 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
818 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
819 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
822 #define IB_DEVICE_NAME_MAX 64
824 struct ib_cache {
825 rwlock_t lock;
826 struct ib_event_handler event_handler;
827 struct ib_pkey_cache **pkey_cache;
828 struct ib_gid_cache **gid_cache;
829 u8 *lmc_cache;
832 struct ib_dma_mapping_ops {
833 int (*mapping_error)(struct ib_device *dev,
834 u64 dma_addr);
835 u64 (*map_single)(struct ib_device *dev,
836 void *ptr, size_t size,
837 enum dma_data_direction direction);
838 void (*unmap_single)(struct ib_device *dev,
839 u64 addr, size_t size,
840 enum dma_data_direction direction);
841 u64 (*map_page)(struct ib_device *dev,
842 struct page *page, unsigned long offset,
843 size_t size,
844 enum dma_data_direction direction);
845 void (*unmap_page)(struct ib_device *dev,
846 u64 addr, size_t size,
847 enum dma_data_direction direction);
848 int (*map_sg)(struct ib_device *dev,
849 struct scatterlist *sg, int nents,
850 enum dma_data_direction direction);
851 void (*unmap_sg)(struct ib_device *dev,
852 struct scatterlist *sg, int nents,
853 enum dma_data_direction direction);
854 u64 (*dma_address)(struct ib_device *dev,
855 struct scatterlist *sg);
856 unsigned int (*dma_len)(struct ib_device *dev,
857 struct scatterlist *sg);
858 void (*sync_single_for_cpu)(struct ib_device *dev,
859 u64 dma_handle,
860 size_t size,
861 enum dma_data_direction dir);
862 void (*sync_single_for_device)(struct ib_device *dev,
863 u64 dma_handle,
864 size_t size,
865 enum dma_data_direction dir);
866 void *(*alloc_coherent)(struct ib_device *dev,
867 size_t size,
868 u64 *dma_handle,
869 gfp_t flag);
870 void (*free_coherent)(struct ib_device *dev,
871 size_t size, void *cpu_addr,
872 u64 dma_handle);
875 struct iw_cm_verbs;
877 struct ib_device {
878 struct device *dma_device;
880 char name[IB_DEVICE_NAME_MAX];
882 struct list_head event_handler_list;
883 spinlock_t event_handler_lock;
885 struct list_head core_list;
886 struct list_head client_data_list;
887 spinlock_t client_data_lock;
889 struct ib_cache cache;
890 int *pkey_tbl_len;
891 int *gid_tbl_len;
893 u32 flags;
895 int num_comp_vectors;
897 struct iw_cm_verbs *iwcm;
899 int (*query_device)(struct ib_device *device,
900 struct ib_device_attr *device_attr);
901 int (*query_port)(struct ib_device *device,
902 u8 port_num,
903 struct ib_port_attr *port_attr);
904 int (*query_gid)(struct ib_device *device,
905 u8 port_num, int index,
906 union ib_gid *gid);
907 int (*query_pkey)(struct ib_device *device,
908 u8 port_num, u16 index, u16 *pkey);
909 int (*modify_device)(struct ib_device *device,
910 int device_modify_mask,
911 struct ib_device_modify *device_modify);
912 int (*modify_port)(struct ib_device *device,
913 u8 port_num, int port_modify_mask,
914 struct ib_port_modify *port_modify);
915 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
916 struct ib_udata *udata);
917 int (*dealloc_ucontext)(struct ib_ucontext *context);
918 int (*mmap)(struct ib_ucontext *context,
919 struct vm_area_struct *vma);
920 struct ib_pd * (*alloc_pd)(struct ib_device *device,
921 struct ib_ucontext *context,
922 struct ib_udata *udata);
923 int (*dealloc_pd)(struct ib_pd *pd);
924 struct ib_ah * (*create_ah)(struct ib_pd *pd,
925 struct ib_ah_attr *ah_attr);
926 int (*modify_ah)(struct ib_ah *ah,
927 struct ib_ah_attr *ah_attr);
928 int (*query_ah)(struct ib_ah *ah,
929 struct ib_ah_attr *ah_attr);
930 int (*destroy_ah)(struct ib_ah *ah);
931 struct ib_srq * (*create_srq)(struct ib_pd *pd,
932 struct ib_srq_init_attr *srq_init_attr,
933 struct ib_udata *udata);
934 int (*modify_srq)(struct ib_srq *srq,
935 struct ib_srq_attr *srq_attr,
936 enum ib_srq_attr_mask srq_attr_mask,
937 struct ib_udata *udata);
938 int (*query_srq)(struct ib_srq *srq,
939 struct ib_srq_attr *srq_attr);
940 int (*destroy_srq)(struct ib_srq *srq);
941 int (*post_srq_recv)(struct ib_srq *srq,
942 struct ib_recv_wr *recv_wr,
943 struct ib_recv_wr **bad_recv_wr);
944 struct ib_qp * (*create_qp)(struct ib_pd *pd,
945 struct ib_qp_init_attr *qp_init_attr,
946 struct ib_udata *udata);
947 int (*modify_qp)(struct ib_qp *qp,
948 struct ib_qp_attr *qp_attr,
949 int qp_attr_mask,
950 struct ib_udata *udata);
951 int (*query_qp)(struct ib_qp *qp,
952 struct ib_qp_attr *qp_attr,
953 int qp_attr_mask,
954 struct ib_qp_init_attr *qp_init_attr);
955 int (*destroy_qp)(struct ib_qp *qp);
956 int (*post_send)(struct ib_qp *qp,
957 struct ib_send_wr *send_wr,
958 struct ib_send_wr **bad_send_wr);
959 int (*post_recv)(struct ib_qp *qp,
960 struct ib_recv_wr *recv_wr,
961 struct ib_recv_wr **bad_recv_wr);
962 struct ib_cq * (*create_cq)(struct ib_device *device, int cqe,
963 int comp_vector,
964 struct ib_ucontext *context,
965 struct ib_udata *udata);
966 int (*destroy_cq)(struct ib_cq *cq);
967 int (*resize_cq)(struct ib_cq *cq, int cqe,
968 struct ib_udata *udata);
969 int (*poll_cq)(struct ib_cq *cq, int num_entries,
970 struct ib_wc *wc);
971 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
972 int (*req_notify_cq)(struct ib_cq *cq,
973 enum ib_cq_notify_flags flags);
974 int (*req_ncomp_notif)(struct ib_cq *cq,
975 int wc_cnt);
976 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
977 int mr_access_flags);
978 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
979 struct ib_phys_buf *phys_buf_array,
980 int num_phys_buf,
981 int mr_access_flags,
982 u64 *iova_start);
983 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
984 u64 start, u64 length,
985 u64 virt_addr,
986 int mr_access_flags,
987 struct ib_udata *udata);
988 int (*query_mr)(struct ib_mr *mr,
989 struct ib_mr_attr *mr_attr);
990 int (*dereg_mr)(struct ib_mr *mr);
991 int (*rereg_phys_mr)(struct ib_mr *mr,
992 int mr_rereg_mask,
993 struct ib_pd *pd,
994 struct ib_phys_buf *phys_buf_array,
995 int num_phys_buf,
996 int mr_access_flags,
997 u64 *iova_start);
998 struct ib_mw * (*alloc_mw)(struct ib_pd *pd);
999 int (*bind_mw)(struct ib_qp *qp,
1000 struct ib_mw *mw,
1001 struct ib_mw_bind *mw_bind);
1002 int (*dealloc_mw)(struct ib_mw *mw);
1003 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1004 int mr_access_flags,
1005 struct ib_fmr_attr *fmr_attr);
1006 int (*map_phys_fmr)(struct ib_fmr *fmr,
1007 u64 *page_list, int list_len,
1008 u64 iova);
1009 int (*unmap_fmr)(struct list_head *fmr_list);
1010 int (*dealloc_fmr)(struct ib_fmr *fmr);
1011 int (*attach_mcast)(struct ib_qp *qp,
1012 union ib_gid *gid,
1013 u16 lid);
1014 int (*detach_mcast)(struct ib_qp *qp,
1015 union ib_gid *gid,
1016 u16 lid);
1017 int (*process_mad)(struct ib_device *device,
1018 int process_mad_flags,
1019 u8 port_num,
1020 struct ib_wc *in_wc,
1021 struct ib_grh *in_grh,
1022 struct ib_mad *in_mad,
1023 struct ib_mad *out_mad);
1025 struct ib_dma_mapping_ops *dma_ops;
1027 struct module *owner;
1028 struct class_device class_dev;
1029 struct kobject ports_parent;
1030 struct list_head port_list;
1032 enum {
1033 IB_DEV_UNINITIALIZED,
1034 IB_DEV_REGISTERED,
1035 IB_DEV_UNREGISTERED
1036 } reg_state;
1038 u64 uverbs_cmd_mask;
1039 int uverbs_abi_ver;
1041 char node_desc[64];
1042 __be64 node_guid;
1043 u8 node_type;
1044 u8 phys_port_cnt;
1047 struct ib_client {
1048 char *name;
1049 void (*add) (struct ib_device *);
1050 void (*remove)(struct ib_device *);
1052 struct list_head list;
1055 struct ib_device *ib_alloc_device(size_t size);
1056 void ib_dealloc_device(struct ib_device *device);
1058 int ib_register_device (struct ib_device *device);
1059 void ib_unregister_device(struct ib_device *device);
1061 int ib_register_client (struct ib_client *client);
1062 void ib_unregister_client(struct ib_client *client);
1064 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1065 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1066 void *data);
1068 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1070 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1073 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1075 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1079 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1080 * contains all required attributes and no attributes not allowed for
1081 * the given QP state transition.
1082 * @cur_state: Current QP state
1083 * @next_state: Next QP state
1084 * @type: QP type
1085 * @mask: Mask of supplied QP attributes
1087 * This function is a helper function that a low-level driver's
1088 * modify_qp method can use to validate the consumer's input. It
1089 * checks that cur_state and next_state are valid QP states, that a
1090 * transition from cur_state to next_state is allowed by the IB spec,
1091 * and that the attribute mask supplied is allowed for the transition.
1093 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1094 enum ib_qp_type type, enum ib_qp_attr_mask mask);
1096 int ib_register_event_handler (struct ib_event_handler *event_handler);
1097 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1098 void ib_dispatch_event(struct ib_event *event);
1100 int ib_query_device(struct ib_device *device,
1101 struct ib_device_attr *device_attr);
1103 int ib_query_port(struct ib_device *device,
1104 u8 port_num, struct ib_port_attr *port_attr);
1106 int ib_query_gid(struct ib_device *device,
1107 u8 port_num, int index, union ib_gid *gid);
1109 int ib_query_pkey(struct ib_device *device,
1110 u8 port_num, u16 index, u16 *pkey);
1112 int ib_modify_device(struct ib_device *device,
1113 int device_modify_mask,
1114 struct ib_device_modify *device_modify);
1116 int ib_modify_port(struct ib_device *device,
1117 u8 port_num, int port_modify_mask,
1118 struct ib_port_modify *port_modify);
1120 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1121 u8 *port_num, u16 *index);
1123 int ib_find_pkey(struct ib_device *device,
1124 u8 port_num, u16 pkey, u16 *index);
1127 * ib_alloc_pd - Allocates an unused protection domain.
1128 * @device: The device on which to allocate the protection domain.
1130 * A protection domain object provides an association between QPs, shared
1131 * receive queues, address handles, memory regions, and memory windows.
1133 struct ib_pd *ib_alloc_pd(struct ib_device *device);
1136 * ib_dealloc_pd - Deallocates a protection domain.
1137 * @pd: The protection domain to deallocate.
1139 int ib_dealloc_pd(struct ib_pd *pd);
1142 * ib_create_ah - Creates an address handle for the given address vector.
1143 * @pd: The protection domain associated with the address handle.
1144 * @ah_attr: The attributes of the address vector.
1146 * The address handle is used to reference a local or global destination
1147 * in all UD QP post sends.
1149 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1152 * ib_init_ah_from_wc - Initializes address handle attributes from a
1153 * work completion.
1154 * @device: Device on which the received message arrived.
1155 * @port_num: Port on which the received message arrived.
1156 * @wc: Work completion associated with the received message.
1157 * @grh: References the received global route header. This parameter is
1158 * ignored unless the work completion indicates that the GRH is valid.
1159 * @ah_attr: Returned attributes that can be used when creating an address
1160 * handle for replying to the message.
1162 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1163 struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1166 * ib_create_ah_from_wc - Creates an address handle associated with the
1167 * sender of the specified work completion.
1168 * @pd: The protection domain associated with the address handle.
1169 * @wc: Work completion information associated with a received message.
1170 * @grh: References the received global route header. This parameter is
1171 * ignored unless the work completion indicates that the GRH is valid.
1172 * @port_num: The outbound port number to associate with the address.
1174 * The address handle is used to reference a local or global destination
1175 * in all UD QP post sends.
1177 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1178 struct ib_grh *grh, u8 port_num);
1181 * ib_modify_ah - Modifies the address vector associated with an address
1182 * handle.
1183 * @ah: The address handle to modify.
1184 * @ah_attr: The new address vector attributes to associate with the
1185 * address handle.
1187 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1190 * ib_query_ah - Queries the address vector associated with an address
1191 * handle.
1192 * @ah: The address handle to query.
1193 * @ah_attr: The address vector attributes associated with the address
1194 * handle.
1196 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1199 * ib_destroy_ah - Destroys an address handle.
1200 * @ah: The address handle to destroy.
1202 int ib_destroy_ah(struct ib_ah *ah);
1205 * ib_create_srq - Creates a SRQ associated with the specified protection
1206 * domain.
1207 * @pd: The protection domain associated with the SRQ.
1208 * @srq_init_attr: A list of initial attributes required to create the
1209 * SRQ. If SRQ creation succeeds, then the attributes are updated to
1210 * the actual capabilities of the created SRQ.
1212 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1213 * requested size of the SRQ, and set to the actual values allocated
1214 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
1215 * will always be at least as large as the requested values.
1217 struct ib_srq *ib_create_srq(struct ib_pd *pd,
1218 struct ib_srq_init_attr *srq_init_attr);
1221 * ib_modify_srq - Modifies the attributes for the specified SRQ.
1222 * @srq: The SRQ to modify.
1223 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
1224 * the current values of selected SRQ attributes are returned.
1225 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1226 * are being modified.
1228 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1229 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1230 * the number of receives queued drops below the limit.
1232 int ib_modify_srq(struct ib_srq *srq,
1233 struct ib_srq_attr *srq_attr,
1234 enum ib_srq_attr_mask srq_attr_mask);
1237 * ib_query_srq - Returns the attribute list and current values for the
1238 * specified SRQ.
1239 * @srq: The SRQ to query.
1240 * @srq_attr: The attributes of the specified SRQ.
1242 int ib_query_srq(struct ib_srq *srq,
1243 struct ib_srq_attr *srq_attr);
1246 * ib_destroy_srq - Destroys the specified SRQ.
1247 * @srq: The SRQ to destroy.
1249 int ib_destroy_srq(struct ib_srq *srq);
1252 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1253 * @srq: The SRQ to post the work request on.
1254 * @recv_wr: A list of work requests to post on the receive queue.
1255 * @bad_recv_wr: On an immediate failure, this parameter will reference
1256 * the work request that failed to be posted on the QP.
1258 static inline int ib_post_srq_recv(struct ib_srq *srq,
1259 struct ib_recv_wr *recv_wr,
1260 struct ib_recv_wr **bad_recv_wr)
1262 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1266 * ib_create_qp - Creates a QP associated with the specified protection
1267 * domain.
1268 * @pd: The protection domain associated with the QP.
1269 * @qp_init_attr: A list of initial attributes required to create the
1270 * QP. If QP creation succeeds, then the attributes are updated to
1271 * the actual capabilities of the created QP.
1273 struct ib_qp *ib_create_qp(struct ib_pd *pd,
1274 struct ib_qp_init_attr *qp_init_attr);
1277 * ib_modify_qp - Modifies the attributes for the specified QP and then
1278 * transitions the QP to the given state.
1279 * @qp: The QP to modify.
1280 * @qp_attr: On input, specifies the QP attributes to modify. On output,
1281 * the current values of selected QP attributes are returned.
1282 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1283 * are being modified.
1285 int ib_modify_qp(struct ib_qp *qp,
1286 struct ib_qp_attr *qp_attr,
1287 int qp_attr_mask);
1290 * ib_query_qp - Returns the attribute list and current values for the
1291 * specified QP.
1292 * @qp: The QP to query.
1293 * @qp_attr: The attributes of the specified QP.
1294 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1295 * @qp_init_attr: Additional attributes of the selected QP.
1297 * The qp_attr_mask may be used to limit the query to gathering only the
1298 * selected attributes.
1300 int ib_query_qp(struct ib_qp *qp,
1301 struct ib_qp_attr *qp_attr,
1302 int qp_attr_mask,
1303 struct ib_qp_init_attr *qp_init_attr);
1306 * ib_destroy_qp - Destroys the specified QP.
1307 * @qp: The QP to destroy.
1309 int ib_destroy_qp(struct ib_qp *qp);
1312 * ib_post_send - Posts a list of work requests to the send queue of
1313 * the specified QP.
1314 * @qp: The QP to post the work request on.
1315 * @send_wr: A list of work requests to post on the send queue.
1316 * @bad_send_wr: On an immediate failure, this parameter will reference
1317 * the work request that failed to be posted on the QP.
1319 static inline int ib_post_send(struct ib_qp *qp,
1320 struct ib_send_wr *send_wr,
1321 struct ib_send_wr **bad_send_wr)
1323 return qp->device->post_send(qp, send_wr, bad_send_wr);
1327 * ib_post_recv - Posts a list of work requests to the receive queue of
1328 * the specified QP.
1329 * @qp: The QP to post the work request on.
1330 * @recv_wr: A list of work requests to post on the receive queue.
1331 * @bad_recv_wr: On an immediate failure, this parameter will reference
1332 * the work request that failed to be posted on the QP.
1334 static inline int ib_post_recv(struct ib_qp *qp,
1335 struct ib_recv_wr *recv_wr,
1336 struct ib_recv_wr **bad_recv_wr)
1338 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1342 * ib_create_cq - Creates a CQ on the specified device.
1343 * @device: The device on which to create the CQ.
1344 * @comp_handler: A user-specified callback that is invoked when a
1345 * completion event occurs on the CQ.
1346 * @event_handler: A user-specified callback that is invoked when an
1347 * asynchronous event not associated with a completion occurs on the CQ.
1348 * @cq_context: Context associated with the CQ returned to the user via
1349 * the associated completion and event handlers.
1350 * @cqe: The minimum size of the CQ.
1351 * @comp_vector - Completion vector used to signal completion events.
1352 * Must be >= 0 and < context->num_comp_vectors.
1354 * Users can examine the cq structure to determine the actual CQ size.
1356 struct ib_cq *ib_create_cq(struct ib_device *device,
1357 ib_comp_handler comp_handler,
1358 void (*event_handler)(struct ib_event *, void *),
1359 void *cq_context, int cqe, int comp_vector);
1362 * ib_resize_cq - Modifies the capacity of the CQ.
1363 * @cq: The CQ to resize.
1364 * @cqe: The minimum size of the CQ.
1366 * Users can examine the cq structure to determine the actual CQ size.
1368 int ib_resize_cq(struct ib_cq *cq, int cqe);
1371 * ib_destroy_cq - Destroys the specified CQ.
1372 * @cq: The CQ to destroy.
1374 int ib_destroy_cq(struct ib_cq *cq);
1377 * ib_poll_cq - poll a CQ for completion(s)
1378 * @cq:the CQ being polled
1379 * @num_entries:maximum number of completions to return
1380 * @wc:array of at least @num_entries &struct ib_wc where completions
1381 * will be returned
1383 * Poll a CQ for (possibly multiple) completions. If the return value
1384 * is < 0, an error occurred. If the return value is >= 0, it is the
1385 * number of completions returned. If the return value is
1386 * non-negative and < num_entries, then the CQ was emptied.
1388 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1389 struct ib_wc *wc)
1391 return cq->device->poll_cq(cq, num_entries, wc);
1395 * ib_peek_cq - Returns the number of unreaped completions currently
1396 * on the specified CQ.
1397 * @cq: The CQ to peek.
1398 * @wc_cnt: A minimum number of unreaped completions to check for.
1400 * If the number of unreaped completions is greater than or equal to wc_cnt,
1401 * this function returns wc_cnt, otherwise, it returns the actual number of
1402 * unreaped completions.
1404 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1407 * ib_req_notify_cq - Request completion notification on a CQ.
1408 * @cq: The CQ to generate an event for.
1409 * @flags:
1410 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1411 * to request an event on the next solicited event or next work
1412 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1413 * may also be |ed in to request a hint about missed events, as
1414 * described below.
1416 * Return Value:
1417 * < 0 means an error occurred while requesting notification
1418 * == 0 means notification was requested successfully, and if
1419 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1420 * were missed and it is safe to wait for another event. In
1421 * this case is it guaranteed that any work completions added
1422 * to the CQ since the last CQ poll will trigger a completion
1423 * notification event.
1424 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1425 * in. It means that the consumer must poll the CQ again to
1426 * make sure it is empty to avoid missing an event because of a
1427 * race between requesting notification and an entry being
1428 * added to the CQ. This return value means it is possible
1429 * (but not guaranteed) that a work completion has been added
1430 * to the CQ since the last poll without triggering a
1431 * completion notification event.
1433 static inline int ib_req_notify_cq(struct ib_cq *cq,
1434 enum ib_cq_notify_flags flags)
1436 return cq->device->req_notify_cq(cq, flags);
1440 * ib_req_ncomp_notif - Request completion notification when there are
1441 * at least the specified number of unreaped completions on the CQ.
1442 * @cq: The CQ to generate an event for.
1443 * @wc_cnt: The number of unreaped completions that should be on the
1444 * CQ before an event is generated.
1446 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1448 return cq->device->req_ncomp_notif ?
1449 cq->device->req_ncomp_notif(cq, wc_cnt) :
1450 -ENOSYS;
1454 * ib_get_dma_mr - Returns a memory region for system memory that is
1455 * usable for DMA.
1456 * @pd: The protection domain associated with the memory region.
1457 * @mr_access_flags: Specifies the memory access rights.
1459 * Note that the ib_dma_*() functions defined below must be used
1460 * to create/destroy addresses used with the Lkey or Rkey returned
1461 * by ib_get_dma_mr().
1463 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1466 * ib_dma_mapping_error - check a DMA addr for error
1467 * @dev: The device for which the dma_addr was created
1468 * @dma_addr: The DMA address to check
1470 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1472 if (dev->dma_ops)
1473 return dev->dma_ops->mapping_error(dev, dma_addr);
1474 return dma_mapping_error(dma_addr);
1478 * ib_dma_map_single - Map a kernel virtual address to DMA address
1479 * @dev: The device for which the dma_addr is to be created
1480 * @cpu_addr: The kernel virtual address
1481 * @size: The size of the region in bytes
1482 * @direction: The direction of the DMA
1484 static inline u64 ib_dma_map_single(struct ib_device *dev,
1485 void *cpu_addr, size_t size,
1486 enum dma_data_direction direction)
1488 if (dev->dma_ops)
1489 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1490 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1494 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1495 * @dev: The device for which the DMA address was created
1496 * @addr: The DMA address
1497 * @size: The size of the region in bytes
1498 * @direction: The direction of the DMA
1500 static inline void ib_dma_unmap_single(struct ib_device *dev,
1501 u64 addr, size_t size,
1502 enum dma_data_direction direction)
1504 if (dev->dma_ops)
1505 dev->dma_ops->unmap_single(dev, addr, size, direction);
1506 else
1507 dma_unmap_single(dev->dma_device, addr, size, direction);
1511 * ib_dma_map_page - Map a physical page to DMA address
1512 * @dev: The device for which the dma_addr is to be created
1513 * @page: The page to be mapped
1514 * @offset: The offset within the page
1515 * @size: The size of the region in bytes
1516 * @direction: The direction of the DMA
1518 static inline u64 ib_dma_map_page(struct ib_device *dev,
1519 struct page *page,
1520 unsigned long offset,
1521 size_t size,
1522 enum dma_data_direction direction)
1524 if (dev->dma_ops)
1525 return dev->dma_ops->map_page(dev, page, offset, size, direction);
1526 return dma_map_page(dev->dma_device, page, offset, size, direction);
1530 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1531 * @dev: The device for which the DMA address was created
1532 * @addr: The DMA address
1533 * @size: The size of the region in bytes
1534 * @direction: The direction of the DMA
1536 static inline void ib_dma_unmap_page(struct ib_device *dev,
1537 u64 addr, size_t size,
1538 enum dma_data_direction direction)
1540 if (dev->dma_ops)
1541 dev->dma_ops->unmap_page(dev, addr, size, direction);
1542 else
1543 dma_unmap_page(dev->dma_device, addr, size, direction);
1547 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1548 * @dev: The device for which the DMA addresses are to be created
1549 * @sg: The array of scatter/gather entries
1550 * @nents: The number of scatter/gather entries
1551 * @direction: The direction of the DMA
1553 static inline int ib_dma_map_sg(struct ib_device *dev,
1554 struct scatterlist *sg, int nents,
1555 enum dma_data_direction direction)
1557 if (dev->dma_ops)
1558 return dev->dma_ops->map_sg(dev, sg, nents, direction);
1559 return dma_map_sg(dev->dma_device, sg, nents, direction);
1563 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
1564 * @dev: The device for which the DMA addresses were created
1565 * @sg: The array of scatter/gather entries
1566 * @nents: The number of scatter/gather entries
1567 * @direction: The direction of the DMA
1569 static inline void ib_dma_unmap_sg(struct ib_device *dev,
1570 struct scatterlist *sg, int nents,
1571 enum dma_data_direction direction)
1573 if (dev->dma_ops)
1574 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
1575 else
1576 dma_unmap_sg(dev->dma_device, sg, nents, direction);
1580 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
1581 * @dev: The device for which the DMA addresses were created
1582 * @sg: The scatter/gather entry
1584 static inline u64 ib_sg_dma_address(struct ib_device *dev,
1585 struct scatterlist *sg)
1587 if (dev->dma_ops)
1588 return dev->dma_ops->dma_address(dev, sg);
1589 return sg_dma_address(sg);
1593 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
1594 * @dev: The device for which the DMA addresses were created
1595 * @sg: The scatter/gather entry
1597 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
1598 struct scatterlist *sg)
1600 if (dev->dma_ops)
1601 return dev->dma_ops->dma_len(dev, sg);
1602 return sg_dma_len(sg);
1606 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
1607 * @dev: The device for which the DMA address was created
1608 * @addr: The DMA address
1609 * @size: The size of the region in bytes
1610 * @dir: The direction of the DMA
1612 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
1613 u64 addr,
1614 size_t size,
1615 enum dma_data_direction dir)
1617 if (dev->dma_ops)
1618 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
1619 else
1620 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
1624 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
1625 * @dev: The device for which the DMA address was created
1626 * @addr: The DMA address
1627 * @size: The size of the region in bytes
1628 * @dir: The direction of the DMA
1630 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
1631 u64 addr,
1632 size_t size,
1633 enum dma_data_direction dir)
1635 if (dev->dma_ops)
1636 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
1637 else
1638 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
1642 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
1643 * @dev: The device for which the DMA address is requested
1644 * @size: The size of the region to allocate in bytes
1645 * @dma_handle: A pointer for returning the DMA address of the region
1646 * @flag: memory allocator flags
1648 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
1649 size_t size,
1650 u64 *dma_handle,
1651 gfp_t flag)
1653 if (dev->dma_ops)
1654 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
1655 else {
1656 dma_addr_t handle;
1657 void *ret;
1659 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
1660 *dma_handle = handle;
1661 return ret;
1666 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
1667 * @dev: The device for which the DMA addresses were allocated
1668 * @size: The size of the region
1669 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
1670 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
1672 static inline void ib_dma_free_coherent(struct ib_device *dev,
1673 size_t size, void *cpu_addr,
1674 u64 dma_handle)
1676 if (dev->dma_ops)
1677 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
1678 else
1679 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
1683 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
1684 * by an HCA.
1685 * @pd: The protection domain associated assigned to the registered region.
1686 * @phys_buf_array: Specifies a list of physical buffers to use in the
1687 * memory region.
1688 * @num_phys_buf: Specifies the size of the phys_buf_array.
1689 * @mr_access_flags: Specifies the memory access rights.
1690 * @iova_start: The offset of the region's starting I/O virtual address.
1692 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
1693 struct ib_phys_buf *phys_buf_array,
1694 int num_phys_buf,
1695 int mr_access_flags,
1696 u64 *iova_start);
1699 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
1700 * Conceptually, this call performs the functions deregister memory region
1701 * followed by register physical memory region. Where possible,
1702 * resources are reused instead of deallocated and reallocated.
1703 * @mr: The memory region to modify.
1704 * @mr_rereg_mask: A bit-mask used to indicate which of the following
1705 * properties of the memory region are being modified.
1706 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
1707 * the new protection domain to associated with the memory region,
1708 * otherwise, this parameter is ignored.
1709 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1710 * field specifies a list of physical buffers to use in the new
1711 * translation, otherwise, this parameter is ignored.
1712 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1713 * field specifies the size of the phys_buf_array, otherwise, this
1714 * parameter is ignored.
1715 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
1716 * field specifies the new memory access rights, otherwise, this
1717 * parameter is ignored.
1718 * @iova_start: The offset of the region's starting I/O virtual address.
1720 int ib_rereg_phys_mr(struct ib_mr *mr,
1721 int mr_rereg_mask,
1722 struct ib_pd *pd,
1723 struct ib_phys_buf *phys_buf_array,
1724 int num_phys_buf,
1725 int mr_access_flags,
1726 u64 *iova_start);
1729 * ib_query_mr - Retrieves information about a specific memory region.
1730 * @mr: The memory region to retrieve information about.
1731 * @mr_attr: The attributes of the specified memory region.
1733 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
1736 * ib_dereg_mr - Deregisters a memory region and removes it from the
1737 * HCA translation table.
1738 * @mr: The memory region to deregister.
1740 int ib_dereg_mr(struct ib_mr *mr);
1743 * ib_alloc_mw - Allocates a memory window.
1744 * @pd: The protection domain associated with the memory window.
1746 struct ib_mw *ib_alloc_mw(struct ib_pd *pd);
1749 * ib_bind_mw - Posts a work request to the send queue of the specified
1750 * QP, which binds the memory window to the given address range and
1751 * remote access attributes.
1752 * @qp: QP to post the bind work request on.
1753 * @mw: The memory window to bind.
1754 * @mw_bind: Specifies information about the memory window, including
1755 * its address range, remote access rights, and associated memory region.
1757 static inline int ib_bind_mw(struct ib_qp *qp,
1758 struct ib_mw *mw,
1759 struct ib_mw_bind *mw_bind)
1761 /* XXX reference counting in corresponding MR? */
1762 return mw->device->bind_mw ?
1763 mw->device->bind_mw(qp, mw, mw_bind) :
1764 -ENOSYS;
1768 * ib_dealloc_mw - Deallocates a memory window.
1769 * @mw: The memory window to deallocate.
1771 int ib_dealloc_mw(struct ib_mw *mw);
1774 * ib_alloc_fmr - Allocates a unmapped fast memory region.
1775 * @pd: The protection domain associated with the unmapped region.
1776 * @mr_access_flags: Specifies the memory access rights.
1777 * @fmr_attr: Attributes of the unmapped region.
1779 * A fast memory region must be mapped before it can be used as part of
1780 * a work request.
1782 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
1783 int mr_access_flags,
1784 struct ib_fmr_attr *fmr_attr);
1787 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
1788 * @fmr: The fast memory region to associate with the pages.
1789 * @page_list: An array of physical pages to map to the fast memory region.
1790 * @list_len: The number of pages in page_list.
1791 * @iova: The I/O virtual address to use with the mapped region.
1793 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
1794 u64 *page_list, int list_len,
1795 u64 iova)
1797 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
1801 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
1802 * @fmr_list: A linked list of fast memory regions to unmap.
1804 int ib_unmap_fmr(struct list_head *fmr_list);
1807 * ib_dealloc_fmr - Deallocates a fast memory region.
1808 * @fmr: The fast memory region to deallocate.
1810 int ib_dealloc_fmr(struct ib_fmr *fmr);
1813 * ib_attach_mcast - Attaches the specified QP to a multicast group.
1814 * @qp: QP to attach to the multicast group. The QP must be type
1815 * IB_QPT_UD.
1816 * @gid: Multicast group GID.
1817 * @lid: Multicast group LID in host byte order.
1819 * In order to send and receive multicast packets, subnet
1820 * administration must have created the multicast group and configured
1821 * the fabric appropriately. The port associated with the specified
1822 * QP must also be a member of the multicast group.
1824 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
1827 * ib_detach_mcast - Detaches the specified QP from a multicast group.
1828 * @qp: QP to detach from the multicast group.
1829 * @gid: Multicast group GID.
1830 * @lid: Multicast group LID in host byte order.
1832 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
1834 #endif /* IB_VERBS_H */