Loading...
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.
9 *
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:
15 *
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
18 * conditions are met:
19 *
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer.
23 *
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.
28 *
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.
37 */
38
39#if !defined(IB_VERBS_H)
40#define IB_VERBS_H
41
42#include <linux/types.h>
43#include <linux/device.h>
44#include <linux/mm.h>
45#include <linux/dma-mapping.h>
46#include <linux/kref.h>
47#include <linux/list.h>
48#include <linux/rwsem.h>
49#include <linux/scatterlist.h>
50#include <linux/workqueue.h>
51
52#include <linux/atomic.h>
53#include <asm/uaccess.h>
54
55extern struct workqueue_struct *ib_wq;
56
57union ib_gid {
58 u8 raw[16];
59 struct {
60 __be64 subnet_prefix;
61 __be64 interface_id;
62 } global;
63};
64
65enum rdma_node_type {
66 /* IB values map to NodeInfo:NodeType. */
67 RDMA_NODE_IB_CA = 1,
68 RDMA_NODE_IB_SWITCH,
69 RDMA_NODE_IB_ROUTER,
70 RDMA_NODE_RNIC
71};
72
73enum rdma_transport_type {
74 RDMA_TRANSPORT_IB,
75 RDMA_TRANSPORT_IWARP
76};
77
78enum rdma_transport_type
79rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__;
80
81enum rdma_link_layer {
82 IB_LINK_LAYER_UNSPECIFIED,
83 IB_LINK_LAYER_INFINIBAND,
84 IB_LINK_LAYER_ETHERNET,
85};
86
87enum ib_device_cap_flags {
88 IB_DEVICE_RESIZE_MAX_WR = 1,
89 IB_DEVICE_BAD_PKEY_CNTR = (1<<1),
90 IB_DEVICE_BAD_QKEY_CNTR = (1<<2),
91 IB_DEVICE_RAW_MULTI = (1<<3),
92 IB_DEVICE_AUTO_PATH_MIG = (1<<4),
93 IB_DEVICE_CHANGE_PHY_PORT = (1<<5),
94 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6),
95 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7),
96 IB_DEVICE_SHUTDOWN_PORT = (1<<8),
97 IB_DEVICE_INIT_TYPE = (1<<9),
98 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10),
99 IB_DEVICE_SYS_IMAGE_GUID = (1<<11),
100 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12),
101 IB_DEVICE_SRQ_RESIZE = (1<<13),
102 IB_DEVICE_N_NOTIFY_CQ = (1<<14),
103 IB_DEVICE_LOCAL_DMA_LKEY = (1<<15),
104 IB_DEVICE_RESERVED = (1<<16), /* old SEND_W_INV */
105 IB_DEVICE_MEM_WINDOW = (1<<17),
106 /*
107 * Devices should set IB_DEVICE_UD_IP_SUM if they support
108 * insertion of UDP and TCP checksum on outgoing UD IPoIB
109 * messages and can verify the validity of checksum for
110 * incoming messages. Setting this flag implies that the
111 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
112 */
113 IB_DEVICE_UD_IP_CSUM = (1<<18),
114 IB_DEVICE_UD_TSO = (1<<19),
115 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21),
116 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
117};
118
119enum ib_atomic_cap {
120 IB_ATOMIC_NONE,
121 IB_ATOMIC_HCA,
122 IB_ATOMIC_GLOB
123};
124
125struct ib_device_attr {
126 u64 fw_ver;
127 __be64 sys_image_guid;
128 u64 max_mr_size;
129 u64 page_size_cap;
130 u32 vendor_id;
131 u32 vendor_part_id;
132 u32 hw_ver;
133 int max_qp;
134 int max_qp_wr;
135 int device_cap_flags;
136 int max_sge;
137 int max_sge_rd;
138 int max_cq;
139 int max_cqe;
140 int max_mr;
141 int max_pd;
142 int max_qp_rd_atom;
143 int max_ee_rd_atom;
144 int max_res_rd_atom;
145 int max_qp_init_rd_atom;
146 int max_ee_init_rd_atom;
147 enum ib_atomic_cap atomic_cap;
148 enum ib_atomic_cap masked_atomic_cap;
149 int max_ee;
150 int max_rdd;
151 int max_mw;
152 int max_raw_ipv6_qp;
153 int max_raw_ethy_qp;
154 int max_mcast_grp;
155 int max_mcast_qp_attach;
156 int max_total_mcast_qp_attach;
157 int max_ah;
158 int max_fmr;
159 int max_map_per_fmr;
160 int max_srq;
161 int max_srq_wr;
162 int max_srq_sge;
163 unsigned int max_fast_reg_page_list_len;
164 u16 max_pkeys;
165 u8 local_ca_ack_delay;
166};
167
168enum ib_mtu {
169 IB_MTU_256 = 1,
170 IB_MTU_512 = 2,
171 IB_MTU_1024 = 3,
172 IB_MTU_2048 = 4,
173 IB_MTU_4096 = 5
174};
175
176static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
177{
178 switch (mtu) {
179 case IB_MTU_256: return 256;
180 case IB_MTU_512: return 512;
181 case IB_MTU_1024: return 1024;
182 case IB_MTU_2048: return 2048;
183 case IB_MTU_4096: return 4096;
184 default: return -1;
185 }
186}
187
188enum ib_port_state {
189 IB_PORT_NOP = 0,
190 IB_PORT_DOWN = 1,
191 IB_PORT_INIT = 2,
192 IB_PORT_ARMED = 3,
193 IB_PORT_ACTIVE = 4,
194 IB_PORT_ACTIVE_DEFER = 5
195};
196
197enum ib_port_cap_flags {
198 IB_PORT_SM = 1 << 1,
199 IB_PORT_NOTICE_SUP = 1 << 2,
200 IB_PORT_TRAP_SUP = 1 << 3,
201 IB_PORT_OPT_IPD_SUP = 1 << 4,
202 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
203 IB_PORT_SL_MAP_SUP = 1 << 6,
204 IB_PORT_MKEY_NVRAM = 1 << 7,
205 IB_PORT_PKEY_NVRAM = 1 << 8,
206 IB_PORT_LED_INFO_SUP = 1 << 9,
207 IB_PORT_SM_DISABLED = 1 << 10,
208 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
209 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
210 IB_PORT_CM_SUP = 1 << 16,
211 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
212 IB_PORT_REINIT_SUP = 1 << 18,
213 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
214 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
215 IB_PORT_DR_NOTICE_SUP = 1 << 21,
216 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
217 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
218 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
219 IB_PORT_CLIENT_REG_SUP = 1 << 25
220};
221
222enum ib_port_width {
223 IB_WIDTH_1X = 1,
224 IB_WIDTH_4X = 2,
225 IB_WIDTH_8X = 4,
226 IB_WIDTH_12X = 8
227};
228
229static inline int ib_width_enum_to_int(enum ib_port_width width)
230{
231 switch (width) {
232 case IB_WIDTH_1X: return 1;
233 case IB_WIDTH_4X: return 4;
234 case IB_WIDTH_8X: return 8;
235 case IB_WIDTH_12X: return 12;
236 default: return -1;
237 }
238}
239
240struct ib_protocol_stats {
241 /* TBD... */
242};
243
244struct iw_protocol_stats {
245 u64 ipInReceives;
246 u64 ipInHdrErrors;
247 u64 ipInTooBigErrors;
248 u64 ipInNoRoutes;
249 u64 ipInAddrErrors;
250 u64 ipInUnknownProtos;
251 u64 ipInTruncatedPkts;
252 u64 ipInDiscards;
253 u64 ipInDelivers;
254 u64 ipOutForwDatagrams;
255 u64 ipOutRequests;
256 u64 ipOutDiscards;
257 u64 ipOutNoRoutes;
258 u64 ipReasmTimeout;
259 u64 ipReasmReqds;
260 u64 ipReasmOKs;
261 u64 ipReasmFails;
262 u64 ipFragOKs;
263 u64 ipFragFails;
264 u64 ipFragCreates;
265 u64 ipInMcastPkts;
266 u64 ipOutMcastPkts;
267 u64 ipInBcastPkts;
268 u64 ipOutBcastPkts;
269
270 u64 tcpRtoAlgorithm;
271 u64 tcpRtoMin;
272 u64 tcpRtoMax;
273 u64 tcpMaxConn;
274 u64 tcpActiveOpens;
275 u64 tcpPassiveOpens;
276 u64 tcpAttemptFails;
277 u64 tcpEstabResets;
278 u64 tcpCurrEstab;
279 u64 tcpInSegs;
280 u64 tcpOutSegs;
281 u64 tcpRetransSegs;
282 u64 tcpInErrs;
283 u64 tcpOutRsts;
284};
285
286union rdma_protocol_stats {
287 struct ib_protocol_stats ib;
288 struct iw_protocol_stats iw;
289};
290
291struct ib_port_attr {
292 enum ib_port_state state;
293 enum ib_mtu max_mtu;
294 enum ib_mtu active_mtu;
295 int gid_tbl_len;
296 u32 port_cap_flags;
297 u32 max_msg_sz;
298 u32 bad_pkey_cntr;
299 u32 qkey_viol_cntr;
300 u16 pkey_tbl_len;
301 u16 lid;
302 u16 sm_lid;
303 u8 lmc;
304 u8 max_vl_num;
305 u8 sm_sl;
306 u8 subnet_timeout;
307 u8 init_type_reply;
308 u8 active_width;
309 u8 active_speed;
310 u8 phys_state;
311};
312
313enum ib_device_modify_flags {
314 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
315 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
316};
317
318struct ib_device_modify {
319 u64 sys_image_guid;
320 char node_desc[64];
321};
322
323enum ib_port_modify_flags {
324 IB_PORT_SHUTDOWN = 1,
325 IB_PORT_INIT_TYPE = (1<<2),
326 IB_PORT_RESET_QKEY_CNTR = (1<<3)
327};
328
329struct ib_port_modify {
330 u32 set_port_cap_mask;
331 u32 clr_port_cap_mask;
332 u8 init_type;
333};
334
335enum ib_event_type {
336 IB_EVENT_CQ_ERR,
337 IB_EVENT_QP_FATAL,
338 IB_EVENT_QP_REQ_ERR,
339 IB_EVENT_QP_ACCESS_ERR,
340 IB_EVENT_COMM_EST,
341 IB_EVENT_SQ_DRAINED,
342 IB_EVENT_PATH_MIG,
343 IB_EVENT_PATH_MIG_ERR,
344 IB_EVENT_DEVICE_FATAL,
345 IB_EVENT_PORT_ACTIVE,
346 IB_EVENT_PORT_ERR,
347 IB_EVENT_LID_CHANGE,
348 IB_EVENT_PKEY_CHANGE,
349 IB_EVENT_SM_CHANGE,
350 IB_EVENT_SRQ_ERR,
351 IB_EVENT_SRQ_LIMIT_REACHED,
352 IB_EVENT_QP_LAST_WQE_REACHED,
353 IB_EVENT_CLIENT_REREGISTER,
354 IB_EVENT_GID_CHANGE,
355};
356
357struct ib_event {
358 struct ib_device *device;
359 union {
360 struct ib_cq *cq;
361 struct ib_qp *qp;
362 struct ib_srq *srq;
363 u8 port_num;
364 } element;
365 enum ib_event_type event;
366};
367
368struct ib_event_handler {
369 struct ib_device *device;
370 void (*handler)(struct ib_event_handler *, struct ib_event *);
371 struct list_head list;
372};
373
374#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
375 do { \
376 (_ptr)->device = _device; \
377 (_ptr)->handler = _handler; \
378 INIT_LIST_HEAD(&(_ptr)->list); \
379 } while (0)
380
381struct ib_global_route {
382 union ib_gid dgid;
383 u32 flow_label;
384 u8 sgid_index;
385 u8 hop_limit;
386 u8 traffic_class;
387};
388
389struct ib_grh {
390 __be32 version_tclass_flow;
391 __be16 paylen;
392 u8 next_hdr;
393 u8 hop_limit;
394 union ib_gid sgid;
395 union ib_gid dgid;
396};
397
398enum {
399 IB_MULTICAST_QPN = 0xffffff
400};
401
402#define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
403
404enum ib_ah_flags {
405 IB_AH_GRH = 1
406};
407
408enum ib_rate {
409 IB_RATE_PORT_CURRENT = 0,
410 IB_RATE_2_5_GBPS = 2,
411 IB_RATE_5_GBPS = 5,
412 IB_RATE_10_GBPS = 3,
413 IB_RATE_20_GBPS = 6,
414 IB_RATE_30_GBPS = 4,
415 IB_RATE_40_GBPS = 7,
416 IB_RATE_60_GBPS = 8,
417 IB_RATE_80_GBPS = 9,
418 IB_RATE_120_GBPS = 10
419};
420
421/**
422 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
423 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
424 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
425 * @rate: rate to convert.
426 */
427int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
428
429/**
430 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
431 * enum.
432 * @mult: multiple to convert.
433 */
434enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
435
436struct ib_ah_attr {
437 struct ib_global_route grh;
438 u16 dlid;
439 u8 sl;
440 u8 src_path_bits;
441 u8 static_rate;
442 u8 ah_flags;
443 u8 port_num;
444};
445
446enum ib_wc_status {
447 IB_WC_SUCCESS,
448 IB_WC_LOC_LEN_ERR,
449 IB_WC_LOC_QP_OP_ERR,
450 IB_WC_LOC_EEC_OP_ERR,
451 IB_WC_LOC_PROT_ERR,
452 IB_WC_WR_FLUSH_ERR,
453 IB_WC_MW_BIND_ERR,
454 IB_WC_BAD_RESP_ERR,
455 IB_WC_LOC_ACCESS_ERR,
456 IB_WC_REM_INV_REQ_ERR,
457 IB_WC_REM_ACCESS_ERR,
458 IB_WC_REM_OP_ERR,
459 IB_WC_RETRY_EXC_ERR,
460 IB_WC_RNR_RETRY_EXC_ERR,
461 IB_WC_LOC_RDD_VIOL_ERR,
462 IB_WC_REM_INV_RD_REQ_ERR,
463 IB_WC_REM_ABORT_ERR,
464 IB_WC_INV_EECN_ERR,
465 IB_WC_INV_EEC_STATE_ERR,
466 IB_WC_FATAL_ERR,
467 IB_WC_RESP_TIMEOUT_ERR,
468 IB_WC_GENERAL_ERR
469};
470
471enum ib_wc_opcode {
472 IB_WC_SEND,
473 IB_WC_RDMA_WRITE,
474 IB_WC_RDMA_READ,
475 IB_WC_COMP_SWAP,
476 IB_WC_FETCH_ADD,
477 IB_WC_BIND_MW,
478 IB_WC_LSO,
479 IB_WC_LOCAL_INV,
480 IB_WC_FAST_REG_MR,
481 IB_WC_MASKED_COMP_SWAP,
482 IB_WC_MASKED_FETCH_ADD,
483/*
484 * Set value of IB_WC_RECV so consumers can test if a completion is a
485 * receive by testing (opcode & IB_WC_RECV).
486 */
487 IB_WC_RECV = 1 << 7,
488 IB_WC_RECV_RDMA_WITH_IMM
489};
490
491enum ib_wc_flags {
492 IB_WC_GRH = 1,
493 IB_WC_WITH_IMM = (1<<1),
494 IB_WC_WITH_INVALIDATE = (1<<2),
495};
496
497struct ib_wc {
498 u64 wr_id;
499 enum ib_wc_status status;
500 enum ib_wc_opcode opcode;
501 u32 vendor_err;
502 u32 byte_len;
503 struct ib_qp *qp;
504 union {
505 __be32 imm_data;
506 u32 invalidate_rkey;
507 } ex;
508 u32 src_qp;
509 int wc_flags;
510 u16 pkey_index;
511 u16 slid;
512 u8 sl;
513 u8 dlid_path_bits;
514 u8 port_num; /* valid only for DR SMPs on switches */
515 int csum_ok;
516};
517
518enum ib_cq_notify_flags {
519 IB_CQ_SOLICITED = 1 << 0,
520 IB_CQ_NEXT_COMP = 1 << 1,
521 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
522 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
523};
524
525enum ib_srq_attr_mask {
526 IB_SRQ_MAX_WR = 1 << 0,
527 IB_SRQ_LIMIT = 1 << 1,
528};
529
530struct ib_srq_attr {
531 u32 max_wr;
532 u32 max_sge;
533 u32 srq_limit;
534};
535
536struct ib_srq_init_attr {
537 void (*event_handler)(struct ib_event *, void *);
538 void *srq_context;
539 struct ib_srq_attr attr;
540};
541
542struct ib_qp_cap {
543 u32 max_send_wr;
544 u32 max_recv_wr;
545 u32 max_send_sge;
546 u32 max_recv_sge;
547 u32 max_inline_data;
548};
549
550enum ib_sig_type {
551 IB_SIGNAL_ALL_WR,
552 IB_SIGNAL_REQ_WR
553};
554
555enum ib_qp_type {
556 /*
557 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
558 * here (and in that order) since the MAD layer uses them as
559 * indices into a 2-entry table.
560 */
561 IB_QPT_SMI,
562 IB_QPT_GSI,
563
564 IB_QPT_RC,
565 IB_QPT_UC,
566 IB_QPT_UD,
567 IB_QPT_RAW_IPV6,
568 IB_QPT_RAW_ETHERTYPE
569};
570
571enum ib_qp_create_flags {
572 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
573 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
574};
575
576struct ib_qp_init_attr {
577 void (*event_handler)(struct ib_event *, void *);
578 void *qp_context;
579 struct ib_cq *send_cq;
580 struct ib_cq *recv_cq;
581 struct ib_srq *srq;
582 struct ib_qp_cap cap;
583 enum ib_sig_type sq_sig_type;
584 enum ib_qp_type qp_type;
585 enum ib_qp_create_flags create_flags;
586 u8 port_num; /* special QP types only */
587};
588
589enum ib_rnr_timeout {
590 IB_RNR_TIMER_655_36 = 0,
591 IB_RNR_TIMER_000_01 = 1,
592 IB_RNR_TIMER_000_02 = 2,
593 IB_RNR_TIMER_000_03 = 3,
594 IB_RNR_TIMER_000_04 = 4,
595 IB_RNR_TIMER_000_06 = 5,
596 IB_RNR_TIMER_000_08 = 6,
597 IB_RNR_TIMER_000_12 = 7,
598 IB_RNR_TIMER_000_16 = 8,
599 IB_RNR_TIMER_000_24 = 9,
600 IB_RNR_TIMER_000_32 = 10,
601 IB_RNR_TIMER_000_48 = 11,
602 IB_RNR_TIMER_000_64 = 12,
603 IB_RNR_TIMER_000_96 = 13,
604 IB_RNR_TIMER_001_28 = 14,
605 IB_RNR_TIMER_001_92 = 15,
606 IB_RNR_TIMER_002_56 = 16,
607 IB_RNR_TIMER_003_84 = 17,
608 IB_RNR_TIMER_005_12 = 18,
609 IB_RNR_TIMER_007_68 = 19,
610 IB_RNR_TIMER_010_24 = 20,
611 IB_RNR_TIMER_015_36 = 21,
612 IB_RNR_TIMER_020_48 = 22,
613 IB_RNR_TIMER_030_72 = 23,
614 IB_RNR_TIMER_040_96 = 24,
615 IB_RNR_TIMER_061_44 = 25,
616 IB_RNR_TIMER_081_92 = 26,
617 IB_RNR_TIMER_122_88 = 27,
618 IB_RNR_TIMER_163_84 = 28,
619 IB_RNR_TIMER_245_76 = 29,
620 IB_RNR_TIMER_327_68 = 30,
621 IB_RNR_TIMER_491_52 = 31
622};
623
624enum ib_qp_attr_mask {
625 IB_QP_STATE = 1,
626 IB_QP_CUR_STATE = (1<<1),
627 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
628 IB_QP_ACCESS_FLAGS = (1<<3),
629 IB_QP_PKEY_INDEX = (1<<4),
630 IB_QP_PORT = (1<<5),
631 IB_QP_QKEY = (1<<6),
632 IB_QP_AV = (1<<7),
633 IB_QP_PATH_MTU = (1<<8),
634 IB_QP_TIMEOUT = (1<<9),
635 IB_QP_RETRY_CNT = (1<<10),
636 IB_QP_RNR_RETRY = (1<<11),
637 IB_QP_RQ_PSN = (1<<12),
638 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
639 IB_QP_ALT_PATH = (1<<14),
640 IB_QP_MIN_RNR_TIMER = (1<<15),
641 IB_QP_SQ_PSN = (1<<16),
642 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
643 IB_QP_PATH_MIG_STATE = (1<<18),
644 IB_QP_CAP = (1<<19),
645 IB_QP_DEST_QPN = (1<<20)
646};
647
648enum ib_qp_state {
649 IB_QPS_RESET,
650 IB_QPS_INIT,
651 IB_QPS_RTR,
652 IB_QPS_RTS,
653 IB_QPS_SQD,
654 IB_QPS_SQE,
655 IB_QPS_ERR
656};
657
658enum ib_mig_state {
659 IB_MIG_MIGRATED,
660 IB_MIG_REARM,
661 IB_MIG_ARMED
662};
663
664struct ib_qp_attr {
665 enum ib_qp_state qp_state;
666 enum ib_qp_state cur_qp_state;
667 enum ib_mtu path_mtu;
668 enum ib_mig_state path_mig_state;
669 u32 qkey;
670 u32 rq_psn;
671 u32 sq_psn;
672 u32 dest_qp_num;
673 int qp_access_flags;
674 struct ib_qp_cap cap;
675 struct ib_ah_attr ah_attr;
676 struct ib_ah_attr alt_ah_attr;
677 u16 pkey_index;
678 u16 alt_pkey_index;
679 u8 en_sqd_async_notify;
680 u8 sq_draining;
681 u8 max_rd_atomic;
682 u8 max_dest_rd_atomic;
683 u8 min_rnr_timer;
684 u8 port_num;
685 u8 timeout;
686 u8 retry_cnt;
687 u8 rnr_retry;
688 u8 alt_port_num;
689 u8 alt_timeout;
690};
691
692enum ib_wr_opcode {
693 IB_WR_RDMA_WRITE,
694 IB_WR_RDMA_WRITE_WITH_IMM,
695 IB_WR_SEND,
696 IB_WR_SEND_WITH_IMM,
697 IB_WR_RDMA_READ,
698 IB_WR_ATOMIC_CMP_AND_SWP,
699 IB_WR_ATOMIC_FETCH_AND_ADD,
700 IB_WR_LSO,
701 IB_WR_SEND_WITH_INV,
702 IB_WR_RDMA_READ_WITH_INV,
703 IB_WR_LOCAL_INV,
704 IB_WR_FAST_REG_MR,
705 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
706 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
707};
708
709enum ib_send_flags {
710 IB_SEND_FENCE = 1,
711 IB_SEND_SIGNALED = (1<<1),
712 IB_SEND_SOLICITED = (1<<2),
713 IB_SEND_INLINE = (1<<3),
714 IB_SEND_IP_CSUM = (1<<4)
715};
716
717struct ib_sge {
718 u64 addr;
719 u32 length;
720 u32 lkey;
721};
722
723struct ib_fast_reg_page_list {
724 struct ib_device *device;
725 u64 *page_list;
726 unsigned int max_page_list_len;
727};
728
729struct ib_send_wr {
730 struct ib_send_wr *next;
731 u64 wr_id;
732 struct ib_sge *sg_list;
733 int num_sge;
734 enum ib_wr_opcode opcode;
735 int send_flags;
736 union {
737 __be32 imm_data;
738 u32 invalidate_rkey;
739 } ex;
740 union {
741 struct {
742 u64 remote_addr;
743 u32 rkey;
744 } rdma;
745 struct {
746 u64 remote_addr;
747 u64 compare_add;
748 u64 swap;
749 u64 compare_add_mask;
750 u64 swap_mask;
751 u32 rkey;
752 } atomic;
753 struct {
754 struct ib_ah *ah;
755 void *header;
756 int hlen;
757 int mss;
758 u32 remote_qpn;
759 u32 remote_qkey;
760 u16 pkey_index; /* valid for GSI only */
761 u8 port_num; /* valid for DR SMPs on switch only */
762 } ud;
763 struct {
764 u64 iova_start;
765 struct ib_fast_reg_page_list *page_list;
766 unsigned int page_shift;
767 unsigned int page_list_len;
768 u32 length;
769 int access_flags;
770 u32 rkey;
771 } fast_reg;
772 } wr;
773};
774
775struct ib_recv_wr {
776 struct ib_recv_wr *next;
777 u64 wr_id;
778 struct ib_sge *sg_list;
779 int num_sge;
780};
781
782enum ib_access_flags {
783 IB_ACCESS_LOCAL_WRITE = 1,
784 IB_ACCESS_REMOTE_WRITE = (1<<1),
785 IB_ACCESS_REMOTE_READ = (1<<2),
786 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
787 IB_ACCESS_MW_BIND = (1<<4)
788};
789
790struct ib_phys_buf {
791 u64 addr;
792 u64 size;
793};
794
795struct ib_mr_attr {
796 struct ib_pd *pd;
797 u64 device_virt_addr;
798 u64 size;
799 int mr_access_flags;
800 u32 lkey;
801 u32 rkey;
802};
803
804enum ib_mr_rereg_flags {
805 IB_MR_REREG_TRANS = 1,
806 IB_MR_REREG_PD = (1<<1),
807 IB_MR_REREG_ACCESS = (1<<2)
808};
809
810struct ib_mw_bind {
811 struct ib_mr *mr;
812 u64 wr_id;
813 u64 addr;
814 u32 length;
815 int send_flags;
816 int mw_access_flags;
817};
818
819struct ib_fmr_attr {
820 int max_pages;
821 int max_maps;
822 u8 page_shift;
823};
824
825struct ib_ucontext {
826 struct ib_device *device;
827 struct list_head pd_list;
828 struct list_head mr_list;
829 struct list_head mw_list;
830 struct list_head cq_list;
831 struct list_head qp_list;
832 struct list_head srq_list;
833 struct list_head ah_list;
834 int closing;
835};
836
837struct ib_uobject {
838 u64 user_handle; /* handle given to us by userspace */
839 struct ib_ucontext *context; /* associated user context */
840 void *object; /* containing object */
841 struct list_head list; /* link to context's list */
842 int id; /* index into kernel idr */
843 struct kref ref;
844 struct rw_semaphore mutex; /* protects .live */
845 int live;
846};
847
848struct ib_udata {
849 void __user *inbuf;
850 void __user *outbuf;
851 size_t inlen;
852 size_t outlen;
853};
854
855struct ib_pd {
856 struct ib_device *device;
857 struct ib_uobject *uobject;
858 atomic_t usecnt; /* count all resources */
859};
860
861struct ib_ah {
862 struct ib_device *device;
863 struct ib_pd *pd;
864 struct ib_uobject *uobject;
865};
866
867typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
868
869struct ib_cq {
870 struct ib_device *device;
871 struct ib_uobject *uobject;
872 ib_comp_handler comp_handler;
873 void (*event_handler)(struct ib_event *, void *);
874 void *cq_context;
875 int cqe;
876 atomic_t usecnt; /* count number of work queues */
877};
878
879struct ib_srq {
880 struct ib_device *device;
881 struct ib_pd *pd;
882 struct ib_uobject *uobject;
883 void (*event_handler)(struct ib_event *, void *);
884 void *srq_context;
885 atomic_t usecnt;
886};
887
888struct ib_qp {
889 struct ib_device *device;
890 struct ib_pd *pd;
891 struct ib_cq *send_cq;
892 struct ib_cq *recv_cq;
893 struct ib_srq *srq;
894 struct ib_uobject *uobject;
895 void (*event_handler)(struct ib_event *, void *);
896 void *qp_context;
897 u32 qp_num;
898 enum ib_qp_type qp_type;
899};
900
901struct ib_mr {
902 struct ib_device *device;
903 struct ib_pd *pd;
904 struct ib_uobject *uobject;
905 u32 lkey;
906 u32 rkey;
907 atomic_t usecnt; /* count number of MWs */
908};
909
910struct ib_mw {
911 struct ib_device *device;
912 struct ib_pd *pd;
913 struct ib_uobject *uobject;
914 u32 rkey;
915};
916
917struct ib_fmr {
918 struct ib_device *device;
919 struct ib_pd *pd;
920 struct list_head list;
921 u32 lkey;
922 u32 rkey;
923};
924
925struct ib_mad;
926struct ib_grh;
927
928enum ib_process_mad_flags {
929 IB_MAD_IGNORE_MKEY = 1,
930 IB_MAD_IGNORE_BKEY = 2,
931 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
932};
933
934enum ib_mad_result {
935 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
936 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
937 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
938 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
939};
940
941#define IB_DEVICE_NAME_MAX 64
942
943struct ib_cache {
944 rwlock_t lock;
945 struct ib_event_handler event_handler;
946 struct ib_pkey_cache **pkey_cache;
947 struct ib_gid_cache **gid_cache;
948 u8 *lmc_cache;
949};
950
951struct ib_dma_mapping_ops {
952 int (*mapping_error)(struct ib_device *dev,
953 u64 dma_addr);
954 u64 (*map_single)(struct ib_device *dev,
955 void *ptr, size_t size,
956 enum dma_data_direction direction);
957 void (*unmap_single)(struct ib_device *dev,
958 u64 addr, size_t size,
959 enum dma_data_direction direction);
960 u64 (*map_page)(struct ib_device *dev,
961 struct page *page, unsigned long offset,
962 size_t size,
963 enum dma_data_direction direction);
964 void (*unmap_page)(struct ib_device *dev,
965 u64 addr, size_t size,
966 enum dma_data_direction direction);
967 int (*map_sg)(struct ib_device *dev,
968 struct scatterlist *sg, int nents,
969 enum dma_data_direction direction);
970 void (*unmap_sg)(struct ib_device *dev,
971 struct scatterlist *sg, int nents,
972 enum dma_data_direction direction);
973 u64 (*dma_address)(struct ib_device *dev,
974 struct scatterlist *sg);
975 unsigned int (*dma_len)(struct ib_device *dev,
976 struct scatterlist *sg);
977 void (*sync_single_for_cpu)(struct ib_device *dev,
978 u64 dma_handle,
979 size_t size,
980 enum dma_data_direction dir);
981 void (*sync_single_for_device)(struct ib_device *dev,
982 u64 dma_handle,
983 size_t size,
984 enum dma_data_direction dir);
985 void *(*alloc_coherent)(struct ib_device *dev,
986 size_t size,
987 u64 *dma_handle,
988 gfp_t flag);
989 void (*free_coherent)(struct ib_device *dev,
990 size_t size, void *cpu_addr,
991 u64 dma_handle);
992};
993
994struct iw_cm_verbs;
995
996struct ib_device {
997 struct device *dma_device;
998
999 char name[IB_DEVICE_NAME_MAX];
1000
1001 struct list_head event_handler_list;
1002 spinlock_t event_handler_lock;
1003
1004 spinlock_t client_data_lock;
1005 struct list_head core_list;
1006 struct list_head client_data_list;
1007
1008 struct ib_cache cache;
1009 int *pkey_tbl_len;
1010 int *gid_tbl_len;
1011
1012 int num_comp_vectors;
1013
1014 struct iw_cm_verbs *iwcm;
1015
1016 int (*get_protocol_stats)(struct ib_device *device,
1017 union rdma_protocol_stats *stats);
1018 int (*query_device)(struct ib_device *device,
1019 struct ib_device_attr *device_attr);
1020 int (*query_port)(struct ib_device *device,
1021 u8 port_num,
1022 struct ib_port_attr *port_attr);
1023 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1024 u8 port_num);
1025 int (*query_gid)(struct ib_device *device,
1026 u8 port_num, int index,
1027 union ib_gid *gid);
1028 int (*query_pkey)(struct ib_device *device,
1029 u8 port_num, u16 index, u16 *pkey);
1030 int (*modify_device)(struct ib_device *device,
1031 int device_modify_mask,
1032 struct ib_device_modify *device_modify);
1033 int (*modify_port)(struct ib_device *device,
1034 u8 port_num, int port_modify_mask,
1035 struct ib_port_modify *port_modify);
1036 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1037 struct ib_udata *udata);
1038 int (*dealloc_ucontext)(struct ib_ucontext *context);
1039 int (*mmap)(struct ib_ucontext *context,
1040 struct vm_area_struct *vma);
1041 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1042 struct ib_ucontext *context,
1043 struct ib_udata *udata);
1044 int (*dealloc_pd)(struct ib_pd *pd);
1045 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1046 struct ib_ah_attr *ah_attr);
1047 int (*modify_ah)(struct ib_ah *ah,
1048 struct ib_ah_attr *ah_attr);
1049 int (*query_ah)(struct ib_ah *ah,
1050 struct ib_ah_attr *ah_attr);
1051 int (*destroy_ah)(struct ib_ah *ah);
1052 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1053 struct ib_srq_init_attr *srq_init_attr,
1054 struct ib_udata *udata);
1055 int (*modify_srq)(struct ib_srq *srq,
1056 struct ib_srq_attr *srq_attr,
1057 enum ib_srq_attr_mask srq_attr_mask,
1058 struct ib_udata *udata);
1059 int (*query_srq)(struct ib_srq *srq,
1060 struct ib_srq_attr *srq_attr);
1061 int (*destroy_srq)(struct ib_srq *srq);
1062 int (*post_srq_recv)(struct ib_srq *srq,
1063 struct ib_recv_wr *recv_wr,
1064 struct ib_recv_wr **bad_recv_wr);
1065 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1066 struct ib_qp_init_attr *qp_init_attr,
1067 struct ib_udata *udata);
1068 int (*modify_qp)(struct ib_qp *qp,
1069 struct ib_qp_attr *qp_attr,
1070 int qp_attr_mask,
1071 struct ib_udata *udata);
1072 int (*query_qp)(struct ib_qp *qp,
1073 struct ib_qp_attr *qp_attr,
1074 int qp_attr_mask,
1075 struct ib_qp_init_attr *qp_init_attr);
1076 int (*destroy_qp)(struct ib_qp *qp);
1077 int (*post_send)(struct ib_qp *qp,
1078 struct ib_send_wr *send_wr,
1079 struct ib_send_wr **bad_send_wr);
1080 int (*post_recv)(struct ib_qp *qp,
1081 struct ib_recv_wr *recv_wr,
1082 struct ib_recv_wr **bad_recv_wr);
1083 struct ib_cq * (*create_cq)(struct ib_device *device, int cqe,
1084 int comp_vector,
1085 struct ib_ucontext *context,
1086 struct ib_udata *udata);
1087 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1088 u16 cq_period);
1089 int (*destroy_cq)(struct ib_cq *cq);
1090 int (*resize_cq)(struct ib_cq *cq, int cqe,
1091 struct ib_udata *udata);
1092 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1093 struct ib_wc *wc);
1094 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1095 int (*req_notify_cq)(struct ib_cq *cq,
1096 enum ib_cq_notify_flags flags);
1097 int (*req_ncomp_notif)(struct ib_cq *cq,
1098 int wc_cnt);
1099 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1100 int mr_access_flags);
1101 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
1102 struct ib_phys_buf *phys_buf_array,
1103 int num_phys_buf,
1104 int mr_access_flags,
1105 u64 *iova_start);
1106 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1107 u64 start, u64 length,
1108 u64 virt_addr,
1109 int mr_access_flags,
1110 struct ib_udata *udata);
1111 int (*query_mr)(struct ib_mr *mr,
1112 struct ib_mr_attr *mr_attr);
1113 int (*dereg_mr)(struct ib_mr *mr);
1114 struct ib_mr * (*alloc_fast_reg_mr)(struct ib_pd *pd,
1115 int max_page_list_len);
1116 struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1117 int page_list_len);
1118 void (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1119 int (*rereg_phys_mr)(struct ib_mr *mr,
1120 int mr_rereg_mask,
1121 struct ib_pd *pd,
1122 struct ib_phys_buf *phys_buf_array,
1123 int num_phys_buf,
1124 int mr_access_flags,
1125 u64 *iova_start);
1126 struct ib_mw * (*alloc_mw)(struct ib_pd *pd);
1127 int (*bind_mw)(struct ib_qp *qp,
1128 struct ib_mw *mw,
1129 struct ib_mw_bind *mw_bind);
1130 int (*dealloc_mw)(struct ib_mw *mw);
1131 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1132 int mr_access_flags,
1133 struct ib_fmr_attr *fmr_attr);
1134 int (*map_phys_fmr)(struct ib_fmr *fmr,
1135 u64 *page_list, int list_len,
1136 u64 iova);
1137 int (*unmap_fmr)(struct list_head *fmr_list);
1138 int (*dealloc_fmr)(struct ib_fmr *fmr);
1139 int (*attach_mcast)(struct ib_qp *qp,
1140 union ib_gid *gid,
1141 u16 lid);
1142 int (*detach_mcast)(struct ib_qp *qp,
1143 union ib_gid *gid,
1144 u16 lid);
1145 int (*process_mad)(struct ib_device *device,
1146 int process_mad_flags,
1147 u8 port_num,
1148 struct ib_wc *in_wc,
1149 struct ib_grh *in_grh,
1150 struct ib_mad *in_mad,
1151 struct ib_mad *out_mad);
1152
1153 struct ib_dma_mapping_ops *dma_ops;
1154
1155 struct module *owner;
1156 struct device dev;
1157 struct kobject *ports_parent;
1158 struct list_head port_list;
1159
1160 enum {
1161 IB_DEV_UNINITIALIZED,
1162 IB_DEV_REGISTERED,
1163 IB_DEV_UNREGISTERED
1164 } reg_state;
1165
1166 int uverbs_abi_ver;
1167 u64 uverbs_cmd_mask;
1168
1169 char node_desc[64];
1170 __be64 node_guid;
1171 u32 local_dma_lkey;
1172 u8 node_type;
1173 u8 phys_port_cnt;
1174};
1175
1176struct ib_client {
1177 char *name;
1178 void (*add) (struct ib_device *);
1179 void (*remove)(struct ib_device *);
1180
1181 struct list_head list;
1182};
1183
1184struct ib_device *ib_alloc_device(size_t size);
1185void ib_dealloc_device(struct ib_device *device);
1186
1187int ib_register_device(struct ib_device *device,
1188 int (*port_callback)(struct ib_device *,
1189 u8, struct kobject *));
1190void ib_unregister_device(struct ib_device *device);
1191
1192int ib_register_client (struct ib_client *client);
1193void ib_unregister_client(struct ib_client *client);
1194
1195void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1196void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1197 void *data);
1198
1199static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1200{
1201 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1202}
1203
1204static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1205{
1206 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1207}
1208
1209/**
1210 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1211 * contains all required attributes and no attributes not allowed for
1212 * the given QP state transition.
1213 * @cur_state: Current QP state
1214 * @next_state: Next QP state
1215 * @type: QP type
1216 * @mask: Mask of supplied QP attributes
1217 *
1218 * This function is a helper function that a low-level driver's
1219 * modify_qp method can use to validate the consumer's input. It
1220 * checks that cur_state and next_state are valid QP states, that a
1221 * transition from cur_state to next_state is allowed by the IB spec,
1222 * and that the attribute mask supplied is allowed for the transition.
1223 */
1224int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1225 enum ib_qp_type type, enum ib_qp_attr_mask mask);
1226
1227int ib_register_event_handler (struct ib_event_handler *event_handler);
1228int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1229void ib_dispatch_event(struct ib_event *event);
1230
1231int ib_query_device(struct ib_device *device,
1232 struct ib_device_attr *device_attr);
1233
1234int ib_query_port(struct ib_device *device,
1235 u8 port_num, struct ib_port_attr *port_attr);
1236
1237enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1238 u8 port_num);
1239
1240int ib_query_gid(struct ib_device *device,
1241 u8 port_num, int index, union ib_gid *gid);
1242
1243int ib_query_pkey(struct ib_device *device,
1244 u8 port_num, u16 index, u16 *pkey);
1245
1246int ib_modify_device(struct ib_device *device,
1247 int device_modify_mask,
1248 struct ib_device_modify *device_modify);
1249
1250int ib_modify_port(struct ib_device *device,
1251 u8 port_num, int port_modify_mask,
1252 struct ib_port_modify *port_modify);
1253
1254int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1255 u8 *port_num, u16 *index);
1256
1257int ib_find_pkey(struct ib_device *device,
1258 u8 port_num, u16 pkey, u16 *index);
1259
1260/**
1261 * ib_alloc_pd - Allocates an unused protection domain.
1262 * @device: The device on which to allocate the protection domain.
1263 *
1264 * A protection domain object provides an association between QPs, shared
1265 * receive queues, address handles, memory regions, and memory windows.
1266 */
1267struct ib_pd *ib_alloc_pd(struct ib_device *device);
1268
1269/**
1270 * ib_dealloc_pd - Deallocates a protection domain.
1271 * @pd: The protection domain to deallocate.
1272 */
1273int ib_dealloc_pd(struct ib_pd *pd);
1274
1275/**
1276 * ib_create_ah - Creates an address handle for the given address vector.
1277 * @pd: The protection domain associated with the address handle.
1278 * @ah_attr: The attributes of the address vector.
1279 *
1280 * The address handle is used to reference a local or global destination
1281 * in all UD QP post sends.
1282 */
1283struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1284
1285/**
1286 * ib_init_ah_from_wc - Initializes address handle attributes from a
1287 * work completion.
1288 * @device: Device on which the received message arrived.
1289 * @port_num: Port on which the received message arrived.
1290 * @wc: Work completion associated with the received message.
1291 * @grh: References the received global route header. This parameter is
1292 * ignored unless the work completion indicates that the GRH is valid.
1293 * @ah_attr: Returned attributes that can be used when creating an address
1294 * handle for replying to the message.
1295 */
1296int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1297 struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1298
1299/**
1300 * ib_create_ah_from_wc - Creates an address handle associated with the
1301 * sender of the specified work completion.
1302 * @pd: The protection domain associated with the address handle.
1303 * @wc: Work completion information associated with a received message.
1304 * @grh: References the received global route header. This parameter is
1305 * ignored unless the work completion indicates that the GRH is valid.
1306 * @port_num: The outbound port number to associate with the address.
1307 *
1308 * The address handle is used to reference a local or global destination
1309 * in all UD QP post sends.
1310 */
1311struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1312 struct ib_grh *grh, u8 port_num);
1313
1314/**
1315 * ib_modify_ah - Modifies the address vector associated with an address
1316 * handle.
1317 * @ah: The address handle to modify.
1318 * @ah_attr: The new address vector attributes to associate with the
1319 * address handle.
1320 */
1321int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1322
1323/**
1324 * ib_query_ah - Queries the address vector associated with an address
1325 * handle.
1326 * @ah: The address handle to query.
1327 * @ah_attr: The address vector attributes associated with the address
1328 * handle.
1329 */
1330int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1331
1332/**
1333 * ib_destroy_ah - Destroys an address handle.
1334 * @ah: The address handle to destroy.
1335 */
1336int ib_destroy_ah(struct ib_ah *ah);
1337
1338/**
1339 * ib_create_srq - Creates a SRQ associated with the specified protection
1340 * domain.
1341 * @pd: The protection domain associated with the SRQ.
1342 * @srq_init_attr: A list of initial attributes required to create the
1343 * SRQ. If SRQ creation succeeds, then the attributes are updated to
1344 * the actual capabilities of the created SRQ.
1345 *
1346 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1347 * requested size of the SRQ, and set to the actual values allocated
1348 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
1349 * will always be at least as large as the requested values.
1350 */
1351struct ib_srq *ib_create_srq(struct ib_pd *pd,
1352 struct ib_srq_init_attr *srq_init_attr);
1353
1354/**
1355 * ib_modify_srq - Modifies the attributes for the specified SRQ.
1356 * @srq: The SRQ to modify.
1357 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
1358 * the current values of selected SRQ attributes are returned.
1359 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1360 * are being modified.
1361 *
1362 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1363 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1364 * the number of receives queued drops below the limit.
1365 */
1366int ib_modify_srq(struct ib_srq *srq,
1367 struct ib_srq_attr *srq_attr,
1368 enum ib_srq_attr_mask srq_attr_mask);
1369
1370/**
1371 * ib_query_srq - Returns the attribute list and current values for the
1372 * specified SRQ.
1373 * @srq: The SRQ to query.
1374 * @srq_attr: The attributes of the specified SRQ.
1375 */
1376int ib_query_srq(struct ib_srq *srq,
1377 struct ib_srq_attr *srq_attr);
1378
1379/**
1380 * ib_destroy_srq - Destroys the specified SRQ.
1381 * @srq: The SRQ to destroy.
1382 */
1383int ib_destroy_srq(struct ib_srq *srq);
1384
1385/**
1386 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1387 * @srq: The SRQ to post the work request on.
1388 * @recv_wr: A list of work requests to post on the receive queue.
1389 * @bad_recv_wr: On an immediate failure, this parameter will reference
1390 * the work request that failed to be posted on the QP.
1391 */
1392static inline int ib_post_srq_recv(struct ib_srq *srq,
1393 struct ib_recv_wr *recv_wr,
1394 struct ib_recv_wr **bad_recv_wr)
1395{
1396 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1397}
1398
1399/**
1400 * ib_create_qp - Creates a QP associated with the specified protection
1401 * domain.
1402 * @pd: The protection domain associated with the QP.
1403 * @qp_init_attr: A list of initial attributes required to create the
1404 * QP. If QP creation succeeds, then the attributes are updated to
1405 * the actual capabilities of the created QP.
1406 */
1407struct ib_qp *ib_create_qp(struct ib_pd *pd,
1408 struct ib_qp_init_attr *qp_init_attr);
1409
1410/**
1411 * ib_modify_qp - Modifies the attributes for the specified QP and then
1412 * transitions the QP to the given state.
1413 * @qp: The QP to modify.
1414 * @qp_attr: On input, specifies the QP attributes to modify. On output,
1415 * the current values of selected QP attributes are returned.
1416 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1417 * are being modified.
1418 */
1419int ib_modify_qp(struct ib_qp *qp,
1420 struct ib_qp_attr *qp_attr,
1421 int qp_attr_mask);
1422
1423/**
1424 * ib_query_qp - Returns the attribute list and current values for the
1425 * specified QP.
1426 * @qp: The QP to query.
1427 * @qp_attr: The attributes of the specified QP.
1428 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1429 * @qp_init_attr: Additional attributes of the selected QP.
1430 *
1431 * The qp_attr_mask may be used to limit the query to gathering only the
1432 * selected attributes.
1433 */
1434int ib_query_qp(struct ib_qp *qp,
1435 struct ib_qp_attr *qp_attr,
1436 int qp_attr_mask,
1437 struct ib_qp_init_attr *qp_init_attr);
1438
1439/**
1440 * ib_destroy_qp - Destroys the specified QP.
1441 * @qp: The QP to destroy.
1442 */
1443int ib_destroy_qp(struct ib_qp *qp);
1444
1445/**
1446 * ib_post_send - Posts a list of work requests to the send queue of
1447 * the specified QP.
1448 * @qp: The QP to post the work request on.
1449 * @send_wr: A list of work requests to post on the send queue.
1450 * @bad_send_wr: On an immediate failure, this parameter will reference
1451 * the work request that failed to be posted on the QP.
1452 *
1453 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
1454 * error is returned, the QP state shall not be affected,
1455 * ib_post_send() will return an immediate error after queueing any
1456 * earlier work requests in the list.
1457 */
1458static inline int ib_post_send(struct ib_qp *qp,
1459 struct ib_send_wr *send_wr,
1460 struct ib_send_wr **bad_send_wr)
1461{
1462 return qp->device->post_send(qp, send_wr, bad_send_wr);
1463}
1464
1465/**
1466 * ib_post_recv - Posts a list of work requests to the receive queue of
1467 * the specified QP.
1468 * @qp: The QP to post the work request on.
1469 * @recv_wr: A list of work requests to post on the receive queue.
1470 * @bad_recv_wr: On an immediate failure, this parameter will reference
1471 * the work request that failed to be posted on the QP.
1472 */
1473static inline int ib_post_recv(struct ib_qp *qp,
1474 struct ib_recv_wr *recv_wr,
1475 struct ib_recv_wr **bad_recv_wr)
1476{
1477 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1478}
1479
1480/**
1481 * ib_create_cq - Creates a CQ on the specified device.
1482 * @device: The device on which to create the CQ.
1483 * @comp_handler: A user-specified callback that is invoked when a
1484 * completion event occurs on the CQ.
1485 * @event_handler: A user-specified callback that is invoked when an
1486 * asynchronous event not associated with a completion occurs on the CQ.
1487 * @cq_context: Context associated with the CQ returned to the user via
1488 * the associated completion and event handlers.
1489 * @cqe: The minimum size of the CQ.
1490 * @comp_vector - Completion vector used to signal completion events.
1491 * Must be >= 0 and < context->num_comp_vectors.
1492 *
1493 * Users can examine the cq structure to determine the actual CQ size.
1494 */
1495struct ib_cq *ib_create_cq(struct ib_device *device,
1496 ib_comp_handler comp_handler,
1497 void (*event_handler)(struct ib_event *, void *),
1498 void *cq_context, int cqe, int comp_vector);
1499
1500/**
1501 * ib_resize_cq - Modifies the capacity of the CQ.
1502 * @cq: The CQ to resize.
1503 * @cqe: The minimum size of the CQ.
1504 *
1505 * Users can examine the cq structure to determine the actual CQ size.
1506 */
1507int ib_resize_cq(struct ib_cq *cq, int cqe);
1508
1509/**
1510 * ib_modify_cq - Modifies moderation params of the CQ
1511 * @cq: The CQ to modify.
1512 * @cq_count: number of CQEs that will trigger an event
1513 * @cq_period: max period of time in usec before triggering an event
1514 *
1515 */
1516int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1517
1518/**
1519 * ib_destroy_cq - Destroys the specified CQ.
1520 * @cq: The CQ to destroy.
1521 */
1522int ib_destroy_cq(struct ib_cq *cq);
1523
1524/**
1525 * ib_poll_cq - poll a CQ for completion(s)
1526 * @cq:the CQ being polled
1527 * @num_entries:maximum number of completions to return
1528 * @wc:array of at least @num_entries &struct ib_wc where completions
1529 * will be returned
1530 *
1531 * Poll a CQ for (possibly multiple) completions. If the return value
1532 * is < 0, an error occurred. If the return value is >= 0, it is the
1533 * number of completions returned. If the return value is
1534 * non-negative and < num_entries, then the CQ was emptied.
1535 */
1536static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1537 struct ib_wc *wc)
1538{
1539 return cq->device->poll_cq(cq, num_entries, wc);
1540}
1541
1542/**
1543 * ib_peek_cq - Returns the number of unreaped completions currently
1544 * on the specified CQ.
1545 * @cq: The CQ to peek.
1546 * @wc_cnt: A minimum number of unreaped completions to check for.
1547 *
1548 * If the number of unreaped completions is greater than or equal to wc_cnt,
1549 * this function returns wc_cnt, otherwise, it returns the actual number of
1550 * unreaped completions.
1551 */
1552int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1553
1554/**
1555 * ib_req_notify_cq - Request completion notification on a CQ.
1556 * @cq: The CQ to generate an event for.
1557 * @flags:
1558 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1559 * to request an event on the next solicited event or next work
1560 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1561 * may also be |ed in to request a hint about missed events, as
1562 * described below.
1563 *
1564 * Return Value:
1565 * < 0 means an error occurred while requesting notification
1566 * == 0 means notification was requested successfully, and if
1567 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1568 * were missed and it is safe to wait for another event. In
1569 * this case is it guaranteed that any work completions added
1570 * to the CQ since the last CQ poll will trigger a completion
1571 * notification event.
1572 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1573 * in. It means that the consumer must poll the CQ again to
1574 * make sure it is empty to avoid missing an event because of a
1575 * race between requesting notification and an entry being
1576 * added to the CQ. This return value means it is possible
1577 * (but not guaranteed) that a work completion has been added
1578 * to the CQ since the last poll without triggering a
1579 * completion notification event.
1580 */
1581static inline int ib_req_notify_cq(struct ib_cq *cq,
1582 enum ib_cq_notify_flags flags)
1583{
1584 return cq->device->req_notify_cq(cq, flags);
1585}
1586
1587/**
1588 * ib_req_ncomp_notif - Request completion notification when there are
1589 * at least the specified number of unreaped completions on the CQ.
1590 * @cq: The CQ to generate an event for.
1591 * @wc_cnt: The number of unreaped completions that should be on the
1592 * CQ before an event is generated.
1593 */
1594static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1595{
1596 return cq->device->req_ncomp_notif ?
1597 cq->device->req_ncomp_notif(cq, wc_cnt) :
1598 -ENOSYS;
1599}
1600
1601/**
1602 * ib_get_dma_mr - Returns a memory region for system memory that is
1603 * usable for DMA.
1604 * @pd: The protection domain associated with the memory region.
1605 * @mr_access_flags: Specifies the memory access rights.
1606 *
1607 * Note that the ib_dma_*() functions defined below must be used
1608 * to create/destroy addresses used with the Lkey or Rkey returned
1609 * by ib_get_dma_mr().
1610 */
1611struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1612
1613/**
1614 * ib_dma_mapping_error - check a DMA addr for error
1615 * @dev: The device for which the dma_addr was created
1616 * @dma_addr: The DMA address to check
1617 */
1618static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1619{
1620 if (dev->dma_ops)
1621 return dev->dma_ops->mapping_error(dev, dma_addr);
1622 return dma_mapping_error(dev->dma_device, dma_addr);
1623}
1624
1625/**
1626 * ib_dma_map_single - Map a kernel virtual address to DMA address
1627 * @dev: The device for which the dma_addr is to be created
1628 * @cpu_addr: The kernel virtual address
1629 * @size: The size of the region in bytes
1630 * @direction: The direction of the DMA
1631 */
1632static inline u64 ib_dma_map_single(struct ib_device *dev,
1633 void *cpu_addr, size_t size,
1634 enum dma_data_direction direction)
1635{
1636 if (dev->dma_ops)
1637 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1638 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1639}
1640
1641/**
1642 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1643 * @dev: The device for which the DMA address was created
1644 * @addr: The DMA address
1645 * @size: The size of the region in bytes
1646 * @direction: The direction of the DMA
1647 */
1648static inline void ib_dma_unmap_single(struct ib_device *dev,
1649 u64 addr, size_t size,
1650 enum dma_data_direction direction)
1651{
1652 if (dev->dma_ops)
1653 dev->dma_ops->unmap_single(dev, addr, size, direction);
1654 else
1655 dma_unmap_single(dev->dma_device, addr, size, direction);
1656}
1657
1658static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1659 void *cpu_addr, size_t size,
1660 enum dma_data_direction direction,
1661 struct dma_attrs *attrs)
1662{
1663 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1664 direction, attrs);
1665}
1666
1667static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1668 u64 addr, size_t size,
1669 enum dma_data_direction direction,
1670 struct dma_attrs *attrs)
1671{
1672 return dma_unmap_single_attrs(dev->dma_device, addr, size,
1673 direction, attrs);
1674}
1675
1676/**
1677 * ib_dma_map_page - Map a physical page to DMA address
1678 * @dev: The device for which the dma_addr is to be created
1679 * @page: The page to be mapped
1680 * @offset: The offset within the page
1681 * @size: The size of the region in bytes
1682 * @direction: The direction of the DMA
1683 */
1684static inline u64 ib_dma_map_page(struct ib_device *dev,
1685 struct page *page,
1686 unsigned long offset,
1687 size_t size,
1688 enum dma_data_direction direction)
1689{
1690 if (dev->dma_ops)
1691 return dev->dma_ops->map_page(dev, page, offset, size, direction);
1692 return dma_map_page(dev->dma_device, page, offset, size, direction);
1693}
1694
1695/**
1696 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1697 * @dev: The device for which the DMA address was created
1698 * @addr: The DMA address
1699 * @size: The size of the region in bytes
1700 * @direction: The direction of the DMA
1701 */
1702static inline void ib_dma_unmap_page(struct ib_device *dev,
1703 u64 addr, size_t size,
1704 enum dma_data_direction direction)
1705{
1706 if (dev->dma_ops)
1707 dev->dma_ops->unmap_page(dev, addr, size, direction);
1708 else
1709 dma_unmap_page(dev->dma_device, addr, size, direction);
1710}
1711
1712/**
1713 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1714 * @dev: The device for which the DMA addresses are to be created
1715 * @sg: The array of scatter/gather entries
1716 * @nents: The number of scatter/gather entries
1717 * @direction: The direction of the DMA
1718 */
1719static inline int ib_dma_map_sg(struct ib_device *dev,
1720 struct scatterlist *sg, int nents,
1721 enum dma_data_direction direction)
1722{
1723 if (dev->dma_ops)
1724 return dev->dma_ops->map_sg(dev, sg, nents, direction);
1725 return dma_map_sg(dev->dma_device, sg, nents, direction);
1726}
1727
1728/**
1729 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
1730 * @dev: The device for which the DMA addresses were created
1731 * @sg: The array of scatter/gather entries
1732 * @nents: The number of scatter/gather entries
1733 * @direction: The direction of the DMA
1734 */
1735static inline void ib_dma_unmap_sg(struct ib_device *dev,
1736 struct scatterlist *sg, int nents,
1737 enum dma_data_direction direction)
1738{
1739 if (dev->dma_ops)
1740 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
1741 else
1742 dma_unmap_sg(dev->dma_device, sg, nents, direction);
1743}
1744
1745static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
1746 struct scatterlist *sg, int nents,
1747 enum dma_data_direction direction,
1748 struct dma_attrs *attrs)
1749{
1750 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1751}
1752
1753static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
1754 struct scatterlist *sg, int nents,
1755 enum dma_data_direction direction,
1756 struct dma_attrs *attrs)
1757{
1758 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
1759}
1760/**
1761 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
1762 * @dev: The device for which the DMA addresses were created
1763 * @sg: The scatter/gather entry
1764 */
1765static inline u64 ib_sg_dma_address(struct ib_device *dev,
1766 struct scatterlist *sg)
1767{
1768 if (dev->dma_ops)
1769 return dev->dma_ops->dma_address(dev, sg);
1770 return sg_dma_address(sg);
1771}
1772
1773/**
1774 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
1775 * @dev: The device for which the DMA addresses were created
1776 * @sg: The scatter/gather entry
1777 */
1778static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
1779 struct scatterlist *sg)
1780{
1781 if (dev->dma_ops)
1782 return dev->dma_ops->dma_len(dev, sg);
1783 return sg_dma_len(sg);
1784}
1785
1786/**
1787 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
1788 * @dev: The device for which the DMA address was created
1789 * @addr: The DMA address
1790 * @size: The size of the region in bytes
1791 * @dir: The direction of the DMA
1792 */
1793static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
1794 u64 addr,
1795 size_t size,
1796 enum dma_data_direction dir)
1797{
1798 if (dev->dma_ops)
1799 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
1800 else
1801 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
1802}
1803
1804/**
1805 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
1806 * @dev: The device for which the DMA address was created
1807 * @addr: The DMA address
1808 * @size: The size of the region in bytes
1809 * @dir: The direction of the DMA
1810 */
1811static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
1812 u64 addr,
1813 size_t size,
1814 enum dma_data_direction dir)
1815{
1816 if (dev->dma_ops)
1817 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
1818 else
1819 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
1820}
1821
1822/**
1823 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
1824 * @dev: The device for which the DMA address is requested
1825 * @size: The size of the region to allocate in bytes
1826 * @dma_handle: A pointer for returning the DMA address of the region
1827 * @flag: memory allocator flags
1828 */
1829static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
1830 size_t size,
1831 u64 *dma_handle,
1832 gfp_t flag)
1833{
1834 if (dev->dma_ops)
1835 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
1836 else {
1837 dma_addr_t handle;
1838 void *ret;
1839
1840 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
1841 *dma_handle = handle;
1842 return ret;
1843 }
1844}
1845
1846/**
1847 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
1848 * @dev: The device for which the DMA addresses were allocated
1849 * @size: The size of the region
1850 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
1851 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
1852 */
1853static inline void ib_dma_free_coherent(struct ib_device *dev,
1854 size_t size, void *cpu_addr,
1855 u64 dma_handle)
1856{
1857 if (dev->dma_ops)
1858 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
1859 else
1860 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
1861}
1862
1863/**
1864 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
1865 * by an HCA.
1866 * @pd: The protection domain associated assigned to the registered region.
1867 * @phys_buf_array: Specifies a list of physical buffers to use in the
1868 * memory region.
1869 * @num_phys_buf: Specifies the size of the phys_buf_array.
1870 * @mr_access_flags: Specifies the memory access rights.
1871 * @iova_start: The offset of the region's starting I/O virtual address.
1872 */
1873struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
1874 struct ib_phys_buf *phys_buf_array,
1875 int num_phys_buf,
1876 int mr_access_flags,
1877 u64 *iova_start);
1878
1879/**
1880 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
1881 * Conceptually, this call performs the functions deregister memory region
1882 * followed by register physical memory region. Where possible,
1883 * resources are reused instead of deallocated and reallocated.
1884 * @mr: The memory region to modify.
1885 * @mr_rereg_mask: A bit-mask used to indicate which of the following
1886 * properties of the memory region are being modified.
1887 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
1888 * the new protection domain to associated with the memory region,
1889 * otherwise, this parameter is ignored.
1890 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1891 * field specifies a list of physical buffers to use in the new
1892 * translation, otherwise, this parameter is ignored.
1893 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
1894 * field specifies the size of the phys_buf_array, otherwise, this
1895 * parameter is ignored.
1896 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
1897 * field specifies the new memory access rights, otherwise, this
1898 * parameter is ignored.
1899 * @iova_start: The offset of the region's starting I/O virtual address.
1900 */
1901int ib_rereg_phys_mr(struct ib_mr *mr,
1902 int mr_rereg_mask,
1903 struct ib_pd *pd,
1904 struct ib_phys_buf *phys_buf_array,
1905 int num_phys_buf,
1906 int mr_access_flags,
1907 u64 *iova_start);
1908
1909/**
1910 * ib_query_mr - Retrieves information about a specific memory region.
1911 * @mr: The memory region to retrieve information about.
1912 * @mr_attr: The attributes of the specified memory region.
1913 */
1914int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
1915
1916/**
1917 * ib_dereg_mr - Deregisters a memory region and removes it from the
1918 * HCA translation table.
1919 * @mr: The memory region to deregister.
1920 */
1921int ib_dereg_mr(struct ib_mr *mr);
1922
1923/**
1924 * ib_alloc_fast_reg_mr - Allocates memory region usable with the
1925 * IB_WR_FAST_REG_MR send work request.
1926 * @pd: The protection domain associated with the region.
1927 * @max_page_list_len: requested max physical buffer list length to be
1928 * used with fast register work requests for this MR.
1929 */
1930struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
1931
1932/**
1933 * ib_alloc_fast_reg_page_list - Allocates a page list array
1934 * @device - ib device pointer.
1935 * @page_list_len - size of the page list array to be allocated.
1936 *
1937 * This allocates and returns a struct ib_fast_reg_page_list * and a
1938 * page_list array that is at least page_list_len in size. The actual
1939 * size is returned in max_page_list_len. The caller is responsible
1940 * for initializing the contents of the page_list array before posting
1941 * a send work request with the IB_WC_FAST_REG_MR opcode.
1942 *
1943 * The page_list array entries must be translated using one of the
1944 * ib_dma_*() functions just like the addresses passed to
1945 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
1946 * ib_fast_reg_page_list must not be modified by the caller until the
1947 * IB_WC_FAST_REG_MR work request completes.
1948 */
1949struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
1950 struct ib_device *device, int page_list_len);
1951
1952/**
1953 * ib_free_fast_reg_page_list - Deallocates a previously allocated
1954 * page list array.
1955 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
1956 */
1957void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
1958
1959/**
1960 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
1961 * R_Key and L_Key.
1962 * @mr - struct ib_mr pointer to be updated.
1963 * @newkey - new key to be used.
1964 */
1965static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
1966{
1967 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
1968 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
1969}
1970
1971/**
1972 * ib_alloc_mw - Allocates a memory window.
1973 * @pd: The protection domain associated with the memory window.
1974 */
1975struct ib_mw *ib_alloc_mw(struct ib_pd *pd);
1976
1977/**
1978 * ib_bind_mw - Posts a work request to the send queue of the specified
1979 * QP, which binds the memory window to the given address range and
1980 * remote access attributes.
1981 * @qp: QP to post the bind work request on.
1982 * @mw: The memory window to bind.
1983 * @mw_bind: Specifies information about the memory window, including
1984 * its address range, remote access rights, and associated memory region.
1985 */
1986static inline int ib_bind_mw(struct ib_qp *qp,
1987 struct ib_mw *mw,
1988 struct ib_mw_bind *mw_bind)
1989{
1990 /* XXX reference counting in corresponding MR? */
1991 return mw->device->bind_mw ?
1992 mw->device->bind_mw(qp, mw, mw_bind) :
1993 -ENOSYS;
1994}
1995
1996/**
1997 * ib_dealloc_mw - Deallocates a memory window.
1998 * @mw: The memory window to deallocate.
1999 */
2000int ib_dealloc_mw(struct ib_mw *mw);
2001
2002/**
2003 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2004 * @pd: The protection domain associated with the unmapped region.
2005 * @mr_access_flags: Specifies the memory access rights.
2006 * @fmr_attr: Attributes of the unmapped region.
2007 *
2008 * A fast memory region must be mapped before it can be used as part of
2009 * a work request.
2010 */
2011struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2012 int mr_access_flags,
2013 struct ib_fmr_attr *fmr_attr);
2014
2015/**
2016 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2017 * @fmr: The fast memory region to associate with the pages.
2018 * @page_list: An array of physical pages to map to the fast memory region.
2019 * @list_len: The number of pages in page_list.
2020 * @iova: The I/O virtual address to use with the mapped region.
2021 */
2022static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2023 u64 *page_list, int list_len,
2024 u64 iova)
2025{
2026 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2027}
2028
2029/**
2030 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2031 * @fmr_list: A linked list of fast memory regions to unmap.
2032 */
2033int ib_unmap_fmr(struct list_head *fmr_list);
2034
2035/**
2036 * ib_dealloc_fmr - Deallocates a fast memory region.
2037 * @fmr: The fast memory region to deallocate.
2038 */
2039int ib_dealloc_fmr(struct ib_fmr *fmr);
2040
2041/**
2042 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2043 * @qp: QP to attach to the multicast group. The QP must be type
2044 * IB_QPT_UD.
2045 * @gid: Multicast group GID.
2046 * @lid: Multicast group LID in host byte order.
2047 *
2048 * In order to send and receive multicast packets, subnet
2049 * administration must have created the multicast group and configured
2050 * the fabric appropriately. The port associated with the specified
2051 * QP must also be a member of the multicast group.
2052 */
2053int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2054
2055/**
2056 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2057 * @qp: QP to detach from the multicast group.
2058 * @gid: Multicast group GID.
2059 * @lid: Multicast group LID in host byte order.
2060 */
2061int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2062
2063#endif /* IB_VERBS_H */
1/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
2/*
3 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
4 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
5 * Copyright (c) 2004 Intel Corporation. All rights reserved.
6 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
7 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
8 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
9 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 */
11
12#ifndef IB_VERBS_H
13#define IB_VERBS_H
14
15#include <linux/types.h>
16#include <linux/device.h>
17#include <linux/dma-mapping.h>
18#include <linux/kref.h>
19#include <linux/list.h>
20#include <linux/rwsem.h>
21#include <linux/workqueue.h>
22#include <linux/irq_poll.h>
23#include <uapi/linux/if_ether.h>
24#include <net/ipv6.h>
25#include <net/ip.h>
26#include <linux/string.h>
27#include <linux/slab.h>
28#include <linux/netdevice.h>
29#include <linux/refcount.h>
30#include <linux/if_link.h>
31#include <linux/atomic.h>
32#include <linux/mmu_notifier.h>
33#include <linux/uaccess.h>
34#include <linux/cgroup_rdma.h>
35#include <linux/irqflags.h>
36#include <linux/preempt.h>
37#include <linux/dim.h>
38#include <uapi/rdma/ib_user_verbs.h>
39#include <rdma/rdma_counter.h>
40#include <rdma/restrack.h>
41#include <rdma/signature.h>
42#include <uapi/rdma/rdma_user_ioctl.h>
43#include <uapi/rdma/ib_user_ioctl_verbs.h>
44
45#define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
46
47struct ib_umem_odp;
48struct ib_uqp_object;
49struct ib_usrq_object;
50struct ib_uwq_object;
51struct rdma_cm_id;
52
53extern struct workqueue_struct *ib_wq;
54extern struct workqueue_struct *ib_comp_wq;
55extern struct workqueue_struct *ib_comp_unbound_wq;
56
57struct ib_ucq_object;
58
59__printf(3, 4) __cold
60void ibdev_printk(const char *level, const struct ib_device *ibdev,
61 const char *format, ...);
62__printf(2, 3) __cold
63void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
64__printf(2, 3) __cold
65void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
66__printf(2, 3) __cold
67void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
68__printf(2, 3) __cold
69void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
70__printf(2, 3) __cold
71void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
72__printf(2, 3) __cold
73void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
74__printf(2, 3) __cold
75void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
76
77#if defined(CONFIG_DYNAMIC_DEBUG) || \
78 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
79#define ibdev_dbg(__dev, format, args...) \
80 dynamic_ibdev_dbg(__dev, format, ##args)
81#else
82__printf(2, 3) __cold
83static inline
84void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
85#endif
86
87#define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
88do { \
89 static DEFINE_RATELIMIT_STATE(_rs, \
90 DEFAULT_RATELIMIT_INTERVAL, \
91 DEFAULT_RATELIMIT_BURST); \
92 if (__ratelimit(&_rs)) \
93 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
94} while (0)
95
96#define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
97 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
98#define ibdev_alert_ratelimited(ibdev, fmt, ...) \
99 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
100#define ibdev_crit_ratelimited(ibdev, fmt, ...) \
101 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
102#define ibdev_err_ratelimited(ibdev, fmt, ...) \
103 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
104#define ibdev_warn_ratelimited(ibdev, fmt, ...) \
105 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
106#define ibdev_notice_ratelimited(ibdev, fmt, ...) \
107 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
108#define ibdev_info_ratelimited(ibdev, fmt, ...) \
109 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
110
111#if defined(CONFIG_DYNAMIC_DEBUG) || \
112 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
113/* descriptor check is first to prevent flooding with "callbacks suppressed" */
114#define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
115do { \
116 static DEFINE_RATELIMIT_STATE(_rs, \
117 DEFAULT_RATELIMIT_INTERVAL, \
118 DEFAULT_RATELIMIT_BURST); \
119 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
120 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
121 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
122 ##__VA_ARGS__); \
123} while (0)
124#else
125__printf(2, 3) __cold
126static inline
127void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
128#endif
129
130union ib_gid {
131 u8 raw[16];
132 struct {
133 __be64 subnet_prefix;
134 __be64 interface_id;
135 } global;
136};
137
138extern union ib_gid zgid;
139
140enum ib_gid_type {
141 /* If link layer is Ethernet, this is RoCE V1 */
142 IB_GID_TYPE_IB = 0,
143 IB_GID_TYPE_ROCE = 0,
144 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
145 IB_GID_TYPE_SIZE
146};
147
148#define ROCE_V2_UDP_DPORT 4791
149struct ib_gid_attr {
150 struct net_device __rcu *ndev;
151 struct ib_device *device;
152 union ib_gid gid;
153 enum ib_gid_type gid_type;
154 u16 index;
155 u8 port_num;
156};
157
158enum {
159 /* set the local administered indication */
160 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
161};
162
163enum rdma_transport_type {
164 RDMA_TRANSPORT_IB,
165 RDMA_TRANSPORT_IWARP,
166 RDMA_TRANSPORT_USNIC,
167 RDMA_TRANSPORT_USNIC_UDP,
168 RDMA_TRANSPORT_UNSPECIFIED,
169};
170
171enum rdma_protocol_type {
172 RDMA_PROTOCOL_IB,
173 RDMA_PROTOCOL_IBOE,
174 RDMA_PROTOCOL_IWARP,
175 RDMA_PROTOCOL_USNIC_UDP
176};
177
178__attribute_const__ enum rdma_transport_type
179rdma_node_get_transport(unsigned int node_type);
180
181enum rdma_network_type {
182 RDMA_NETWORK_IB,
183 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
184 RDMA_NETWORK_IPV4,
185 RDMA_NETWORK_IPV6
186};
187
188static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
189{
190 if (network_type == RDMA_NETWORK_IPV4 ||
191 network_type == RDMA_NETWORK_IPV6)
192 return IB_GID_TYPE_ROCE_UDP_ENCAP;
193
194 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
195 return IB_GID_TYPE_IB;
196}
197
198static inline enum rdma_network_type
199rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
200{
201 if (attr->gid_type == IB_GID_TYPE_IB)
202 return RDMA_NETWORK_IB;
203
204 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
205 return RDMA_NETWORK_IPV4;
206 else
207 return RDMA_NETWORK_IPV6;
208}
209
210enum rdma_link_layer {
211 IB_LINK_LAYER_UNSPECIFIED,
212 IB_LINK_LAYER_INFINIBAND,
213 IB_LINK_LAYER_ETHERNET,
214};
215
216enum ib_device_cap_flags {
217 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
218 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
219 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
220 IB_DEVICE_RAW_MULTI = (1 << 3),
221 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
222 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
223 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
224 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
225 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
226 /* Not in use, former INIT_TYPE = (1 << 9),*/
227 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
228 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
229 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
230 IB_DEVICE_SRQ_RESIZE = (1 << 13),
231 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
232
233 /*
234 * This device supports a per-device lkey or stag that can be
235 * used without performing a memory registration for the local
236 * memory. Note that ULPs should never check this flag, but
237 * instead of use the local_dma_lkey flag in the ib_pd structure,
238 * which will always contain a usable lkey.
239 */
240 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
241 /* Reserved, old SEND_W_INV = (1 << 16),*/
242 IB_DEVICE_MEM_WINDOW = (1 << 17),
243 /*
244 * Devices should set IB_DEVICE_UD_IP_SUM if they support
245 * insertion of UDP and TCP checksum on outgoing UD IPoIB
246 * messages and can verify the validity of checksum for
247 * incoming messages. Setting this flag implies that the
248 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
249 */
250 IB_DEVICE_UD_IP_CSUM = (1 << 18),
251 IB_DEVICE_UD_TSO = (1 << 19),
252 IB_DEVICE_XRC = (1 << 20),
253
254 /*
255 * This device supports the IB "base memory management extension",
256 * which includes support for fast registrations (IB_WR_REG_MR,
257 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
258 * also be set by any iWarp device which must support FRs to comply
259 * to the iWarp verbs spec. iWarp devices also support the
260 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
261 * stag.
262 */
263 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
264 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
265 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
266 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
267 IB_DEVICE_RC_IP_CSUM = (1 << 25),
268 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
269 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
270 /*
271 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
272 * support execution of WQEs that involve synchronization
273 * of I/O operations with single completion queue managed
274 * by hardware.
275 */
276 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
277 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
278 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
279 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
280 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
281 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
282 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
283 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
284 IB_DEVICE_RDMA_NETDEV_OPA = (1ULL << 35),
285 /* The device supports padding incoming writes to cacheline. */
286 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
287 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
288};
289
290enum ib_atomic_cap {
291 IB_ATOMIC_NONE,
292 IB_ATOMIC_HCA,
293 IB_ATOMIC_GLOB
294};
295
296enum ib_odp_general_cap_bits {
297 IB_ODP_SUPPORT = 1 << 0,
298 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
299};
300
301enum ib_odp_transport_cap_bits {
302 IB_ODP_SUPPORT_SEND = 1 << 0,
303 IB_ODP_SUPPORT_RECV = 1 << 1,
304 IB_ODP_SUPPORT_WRITE = 1 << 2,
305 IB_ODP_SUPPORT_READ = 1 << 3,
306 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
307 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
308};
309
310struct ib_odp_caps {
311 uint64_t general_caps;
312 struct {
313 uint32_t rc_odp_caps;
314 uint32_t uc_odp_caps;
315 uint32_t ud_odp_caps;
316 uint32_t xrc_odp_caps;
317 } per_transport_caps;
318};
319
320struct ib_rss_caps {
321 /* Corresponding bit will be set if qp type from
322 * 'enum ib_qp_type' is supported, e.g.
323 * supported_qpts |= 1 << IB_QPT_UD
324 */
325 u32 supported_qpts;
326 u32 max_rwq_indirection_tables;
327 u32 max_rwq_indirection_table_size;
328};
329
330enum ib_tm_cap_flags {
331 /* Support tag matching with rendezvous offload for RC transport */
332 IB_TM_CAP_RNDV_RC = 1 << 0,
333};
334
335struct ib_tm_caps {
336 /* Max size of RNDV header */
337 u32 max_rndv_hdr_size;
338 /* Max number of entries in tag matching list */
339 u32 max_num_tags;
340 /* From enum ib_tm_cap_flags */
341 u32 flags;
342 /* Max number of outstanding list operations */
343 u32 max_ops;
344 /* Max number of SGE in tag matching entry */
345 u32 max_sge;
346};
347
348struct ib_cq_init_attr {
349 unsigned int cqe;
350 u32 comp_vector;
351 u32 flags;
352};
353
354enum ib_cq_attr_mask {
355 IB_CQ_MODERATE = 1 << 0,
356};
357
358struct ib_cq_caps {
359 u16 max_cq_moderation_count;
360 u16 max_cq_moderation_period;
361};
362
363struct ib_dm_mr_attr {
364 u64 length;
365 u64 offset;
366 u32 access_flags;
367};
368
369struct ib_dm_alloc_attr {
370 u64 length;
371 u32 alignment;
372 u32 flags;
373};
374
375struct ib_device_attr {
376 u64 fw_ver;
377 __be64 sys_image_guid;
378 u64 max_mr_size;
379 u64 page_size_cap;
380 u32 vendor_id;
381 u32 vendor_part_id;
382 u32 hw_ver;
383 int max_qp;
384 int max_qp_wr;
385 u64 device_cap_flags;
386 int max_send_sge;
387 int max_recv_sge;
388 int max_sge_rd;
389 int max_cq;
390 int max_cqe;
391 int max_mr;
392 int max_pd;
393 int max_qp_rd_atom;
394 int max_ee_rd_atom;
395 int max_res_rd_atom;
396 int max_qp_init_rd_atom;
397 int max_ee_init_rd_atom;
398 enum ib_atomic_cap atomic_cap;
399 enum ib_atomic_cap masked_atomic_cap;
400 int max_ee;
401 int max_rdd;
402 int max_mw;
403 int max_raw_ipv6_qp;
404 int max_raw_ethy_qp;
405 int max_mcast_grp;
406 int max_mcast_qp_attach;
407 int max_total_mcast_qp_attach;
408 int max_ah;
409 int max_srq;
410 int max_srq_wr;
411 int max_srq_sge;
412 unsigned int max_fast_reg_page_list_len;
413 unsigned int max_pi_fast_reg_page_list_len;
414 u16 max_pkeys;
415 u8 local_ca_ack_delay;
416 int sig_prot_cap;
417 int sig_guard_cap;
418 struct ib_odp_caps odp_caps;
419 uint64_t timestamp_mask;
420 uint64_t hca_core_clock; /* in KHZ */
421 struct ib_rss_caps rss_caps;
422 u32 max_wq_type_rq;
423 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
424 struct ib_tm_caps tm_caps;
425 struct ib_cq_caps cq_caps;
426 u64 max_dm_size;
427 /* Max entries for sgl for optimized performance per READ */
428 u32 max_sgl_rd;
429};
430
431enum ib_mtu {
432 IB_MTU_256 = 1,
433 IB_MTU_512 = 2,
434 IB_MTU_1024 = 3,
435 IB_MTU_2048 = 4,
436 IB_MTU_4096 = 5
437};
438
439enum opa_mtu {
440 OPA_MTU_8192 = 6,
441 OPA_MTU_10240 = 7
442};
443
444static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
445{
446 switch (mtu) {
447 case IB_MTU_256: return 256;
448 case IB_MTU_512: return 512;
449 case IB_MTU_1024: return 1024;
450 case IB_MTU_2048: return 2048;
451 case IB_MTU_4096: return 4096;
452 default: return -1;
453 }
454}
455
456static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
457{
458 if (mtu >= 4096)
459 return IB_MTU_4096;
460 else if (mtu >= 2048)
461 return IB_MTU_2048;
462 else if (mtu >= 1024)
463 return IB_MTU_1024;
464 else if (mtu >= 512)
465 return IB_MTU_512;
466 else
467 return IB_MTU_256;
468}
469
470static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
471{
472 switch (mtu) {
473 case OPA_MTU_8192:
474 return 8192;
475 case OPA_MTU_10240:
476 return 10240;
477 default:
478 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
479 }
480}
481
482static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
483{
484 if (mtu >= 10240)
485 return OPA_MTU_10240;
486 else if (mtu >= 8192)
487 return OPA_MTU_8192;
488 else
489 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
490}
491
492enum ib_port_state {
493 IB_PORT_NOP = 0,
494 IB_PORT_DOWN = 1,
495 IB_PORT_INIT = 2,
496 IB_PORT_ARMED = 3,
497 IB_PORT_ACTIVE = 4,
498 IB_PORT_ACTIVE_DEFER = 5
499};
500
501enum ib_port_phys_state {
502 IB_PORT_PHYS_STATE_SLEEP = 1,
503 IB_PORT_PHYS_STATE_POLLING = 2,
504 IB_PORT_PHYS_STATE_DISABLED = 3,
505 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
506 IB_PORT_PHYS_STATE_LINK_UP = 5,
507 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
508 IB_PORT_PHYS_STATE_PHY_TEST = 7,
509};
510
511enum ib_port_width {
512 IB_WIDTH_1X = 1,
513 IB_WIDTH_2X = 16,
514 IB_WIDTH_4X = 2,
515 IB_WIDTH_8X = 4,
516 IB_WIDTH_12X = 8
517};
518
519static inline int ib_width_enum_to_int(enum ib_port_width width)
520{
521 switch (width) {
522 case IB_WIDTH_1X: return 1;
523 case IB_WIDTH_2X: return 2;
524 case IB_WIDTH_4X: return 4;
525 case IB_WIDTH_8X: return 8;
526 case IB_WIDTH_12X: return 12;
527 default: return -1;
528 }
529}
530
531enum ib_port_speed {
532 IB_SPEED_SDR = 1,
533 IB_SPEED_DDR = 2,
534 IB_SPEED_QDR = 4,
535 IB_SPEED_FDR10 = 8,
536 IB_SPEED_FDR = 16,
537 IB_SPEED_EDR = 32,
538 IB_SPEED_HDR = 64
539};
540
541/**
542 * struct rdma_hw_stats
543 * @lock - Mutex to protect parallel write access to lifespan and values
544 * of counters, which are 64bits and not guaranteeed to be written
545 * atomicaly on 32bits systems.
546 * @timestamp - Used by the core code to track when the last update was
547 * @lifespan - Used by the core code to determine how old the counters
548 * should be before being updated again. Stored in jiffies, defaults
549 * to 10 milliseconds, drivers can override the default be specifying
550 * their own value during their allocation routine.
551 * @name - Array of pointers to static names used for the counters in
552 * directory.
553 * @num_counters - How many hardware counters there are. If name is
554 * shorter than this number, a kernel oops will result. Driver authors
555 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
556 * in their code to prevent this.
557 * @value - Array of u64 counters that are accessed by the sysfs code and
558 * filled in by the drivers get_stats routine
559 */
560struct rdma_hw_stats {
561 struct mutex lock; /* Protect lifespan and values[] */
562 unsigned long timestamp;
563 unsigned long lifespan;
564 const char * const *names;
565 int num_counters;
566 u64 value[];
567};
568
569#define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
570/**
571 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
572 * for drivers.
573 * @names - Array of static const char *
574 * @num_counters - How many elements in array
575 * @lifespan - How many milliseconds between updates
576 */
577static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
578 const char * const *names, int num_counters,
579 unsigned long lifespan)
580{
581 struct rdma_hw_stats *stats;
582
583 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
584 GFP_KERNEL);
585 if (!stats)
586 return NULL;
587 stats->names = names;
588 stats->num_counters = num_counters;
589 stats->lifespan = msecs_to_jiffies(lifespan);
590
591 return stats;
592}
593
594
595/* Define bits for the various functionality this port needs to be supported by
596 * the core.
597 */
598/* Management 0x00000FFF */
599#define RDMA_CORE_CAP_IB_MAD 0x00000001
600#define RDMA_CORE_CAP_IB_SMI 0x00000002
601#define RDMA_CORE_CAP_IB_CM 0x00000004
602#define RDMA_CORE_CAP_IW_CM 0x00000008
603#define RDMA_CORE_CAP_IB_SA 0x00000010
604#define RDMA_CORE_CAP_OPA_MAD 0x00000020
605
606/* Address format 0x000FF000 */
607#define RDMA_CORE_CAP_AF_IB 0x00001000
608#define RDMA_CORE_CAP_ETH_AH 0x00002000
609#define RDMA_CORE_CAP_OPA_AH 0x00004000
610#define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
611
612/* Protocol 0xFFF00000 */
613#define RDMA_CORE_CAP_PROT_IB 0x00100000
614#define RDMA_CORE_CAP_PROT_ROCE 0x00200000
615#define RDMA_CORE_CAP_PROT_IWARP 0x00400000
616#define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
617#define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
618#define RDMA_CORE_CAP_PROT_USNIC 0x02000000
619
620#define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
621 | RDMA_CORE_CAP_PROT_ROCE \
622 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
623
624#define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
625 | RDMA_CORE_CAP_IB_MAD \
626 | RDMA_CORE_CAP_IB_SMI \
627 | RDMA_CORE_CAP_IB_CM \
628 | RDMA_CORE_CAP_IB_SA \
629 | RDMA_CORE_CAP_AF_IB)
630#define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
631 | RDMA_CORE_CAP_IB_MAD \
632 | RDMA_CORE_CAP_IB_CM \
633 | RDMA_CORE_CAP_AF_IB \
634 | RDMA_CORE_CAP_ETH_AH)
635#define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
636 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
637 | RDMA_CORE_CAP_IB_MAD \
638 | RDMA_CORE_CAP_IB_CM \
639 | RDMA_CORE_CAP_AF_IB \
640 | RDMA_CORE_CAP_ETH_AH)
641#define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
642 | RDMA_CORE_CAP_IW_CM)
643#define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
644 | RDMA_CORE_CAP_OPA_MAD)
645
646#define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
647
648#define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
649
650struct ib_port_attr {
651 u64 subnet_prefix;
652 enum ib_port_state state;
653 enum ib_mtu max_mtu;
654 enum ib_mtu active_mtu;
655 u32 phys_mtu;
656 int gid_tbl_len;
657 unsigned int ip_gids:1;
658 /* This is the value from PortInfo CapabilityMask, defined by IBA */
659 u32 port_cap_flags;
660 u32 max_msg_sz;
661 u32 bad_pkey_cntr;
662 u32 qkey_viol_cntr;
663 u16 pkey_tbl_len;
664 u32 sm_lid;
665 u32 lid;
666 u8 lmc;
667 u8 max_vl_num;
668 u8 sm_sl;
669 u8 subnet_timeout;
670 u8 init_type_reply;
671 u8 active_width;
672 u8 active_speed;
673 u8 phys_state;
674 u16 port_cap_flags2;
675};
676
677enum ib_device_modify_flags {
678 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
679 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
680};
681
682#define IB_DEVICE_NODE_DESC_MAX 64
683
684struct ib_device_modify {
685 u64 sys_image_guid;
686 char node_desc[IB_DEVICE_NODE_DESC_MAX];
687};
688
689enum ib_port_modify_flags {
690 IB_PORT_SHUTDOWN = 1,
691 IB_PORT_INIT_TYPE = (1<<2),
692 IB_PORT_RESET_QKEY_CNTR = (1<<3),
693 IB_PORT_OPA_MASK_CHG = (1<<4)
694};
695
696struct ib_port_modify {
697 u32 set_port_cap_mask;
698 u32 clr_port_cap_mask;
699 u8 init_type;
700};
701
702enum ib_event_type {
703 IB_EVENT_CQ_ERR,
704 IB_EVENT_QP_FATAL,
705 IB_EVENT_QP_REQ_ERR,
706 IB_EVENT_QP_ACCESS_ERR,
707 IB_EVENT_COMM_EST,
708 IB_EVENT_SQ_DRAINED,
709 IB_EVENT_PATH_MIG,
710 IB_EVENT_PATH_MIG_ERR,
711 IB_EVENT_DEVICE_FATAL,
712 IB_EVENT_PORT_ACTIVE,
713 IB_EVENT_PORT_ERR,
714 IB_EVENT_LID_CHANGE,
715 IB_EVENT_PKEY_CHANGE,
716 IB_EVENT_SM_CHANGE,
717 IB_EVENT_SRQ_ERR,
718 IB_EVENT_SRQ_LIMIT_REACHED,
719 IB_EVENT_QP_LAST_WQE_REACHED,
720 IB_EVENT_CLIENT_REREGISTER,
721 IB_EVENT_GID_CHANGE,
722 IB_EVENT_WQ_FATAL,
723};
724
725const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
726
727struct ib_event {
728 struct ib_device *device;
729 union {
730 struct ib_cq *cq;
731 struct ib_qp *qp;
732 struct ib_srq *srq;
733 struct ib_wq *wq;
734 u8 port_num;
735 } element;
736 enum ib_event_type event;
737};
738
739struct ib_event_handler {
740 struct ib_device *device;
741 void (*handler)(struct ib_event_handler *, struct ib_event *);
742 struct list_head list;
743};
744
745#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
746 do { \
747 (_ptr)->device = _device; \
748 (_ptr)->handler = _handler; \
749 INIT_LIST_HEAD(&(_ptr)->list); \
750 } while (0)
751
752struct ib_global_route {
753 const struct ib_gid_attr *sgid_attr;
754 union ib_gid dgid;
755 u32 flow_label;
756 u8 sgid_index;
757 u8 hop_limit;
758 u8 traffic_class;
759};
760
761struct ib_grh {
762 __be32 version_tclass_flow;
763 __be16 paylen;
764 u8 next_hdr;
765 u8 hop_limit;
766 union ib_gid sgid;
767 union ib_gid dgid;
768};
769
770union rdma_network_hdr {
771 struct ib_grh ibgrh;
772 struct {
773 /* The IB spec states that if it's IPv4, the header
774 * is located in the last 20 bytes of the header.
775 */
776 u8 reserved[20];
777 struct iphdr roce4grh;
778 };
779};
780
781#define IB_QPN_MASK 0xFFFFFF
782
783enum {
784 IB_MULTICAST_QPN = 0xffffff
785};
786
787#define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
788#define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
789
790enum ib_ah_flags {
791 IB_AH_GRH = 1
792};
793
794enum ib_rate {
795 IB_RATE_PORT_CURRENT = 0,
796 IB_RATE_2_5_GBPS = 2,
797 IB_RATE_5_GBPS = 5,
798 IB_RATE_10_GBPS = 3,
799 IB_RATE_20_GBPS = 6,
800 IB_RATE_30_GBPS = 4,
801 IB_RATE_40_GBPS = 7,
802 IB_RATE_60_GBPS = 8,
803 IB_RATE_80_GBPS = 9,
804 IB_RATE_120_GBPS = 10,
805 IB_RATE_14_GBPS = 11,
806 IB_RATE_56_GBPS = 12,
807 IB_RATE_112_GBPS = 13,
808 IB_RATE_168_GBPS = 14,
809 IB_RATE_25_GBPS = 15,
810 IB_RATE_100_GBPS = 16,
811 IB_RATE_200_GBPS = 17,
812 IB_RATE_300_GBPS = 18,
813 IB_RATE_28_GBPS = 19,
814 IB_RATE_50_GBPS = 20,
815 IB_RATE_400_GBPS = 21,
816 IB_RATE_600_GBPS = 22,
817};
818
819/**
820 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
821 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
822 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
823 * @rate: rate to convert.
824 */
825__attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
826
827/**
828 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
829 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
830 * @rate: rate to convert.
831 */
832__attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
833
834
835/**
836 * enum ib_mr_type - memory region type
837 * @IB_MR_TYPE_MEM_REG: memory region that is used for
838 * normal registration
839 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
840 * register any arbitrary sg lists (without
841 * the normal mr constraints - see
842 * ib_map_mr_sg)
843 * @IB_MR_TYPE_DM: memory region that is used for device
844 * memory registration
845 * @IB_MR_TYPE_USER: memory region that is used for the user-space
846 * application
847 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
848 * without address translations (VA=PA)
849 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
850 * data integrity operations
851 */
852enum ib_mr_type {
853 IB_MR_TYPE_MEM_REG,
854 IB_MR_TYPE_SG_GAPS,
855 IB_MR_TYPE_DM,
856 IB_MR_TYPE_USER,
857 IB_MR_TYPE_DMA,
858 IB_MR_TYPE_INTEGRITY,
859};
860
861enum ib_mr_status_check {
862 IB_MR_CHECK_SIG_STATUS = 1,
863};
864
865/**
866 * struct ib_mr_status - Memory region status container
867 *
868 * @fail_status: Bitmask of MR checks status. For each
869 * failed check a corresponding status bit is set.
870 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
871 * failure.
872 */
873struct ib_mr_status {
874 u32 fail_status;
875 struct ib_sig_err sig_err;
876};
877
878/**
879 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
880 * enum.
881 * @mult: multiple to convert.
882 */
883__attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
884
885struct rdma_ah_init_attr {
886 struct rdma_ah_attr *ah_attr;
887 u32 flags;
888 struct net_device *xmit_slave;
889};
890
891enum rdma_ah_attr_type {
892 RDMA_AH_ATTR_TYPE_UNDEFINED,
893 RDMA_AH_ATTR_TYPE_IB,
894 RDMA_AH_ATTR_TYPE_ROCE,
895 RDMA_AH_ATTR_TYPE_OPA,
896};
897
898struct ib_ah_attr {
899 u16 dlid;
900 u8 src_path_bits;
901};
902
903struct roce_ah_attr {
904 u8 dmac[ETH_ALEN];
905};
906
907struct opa_ah_attr {
908 u32 dlid;
909 u8 src_path_bits;
910 bool make_grd;
911};
912
913struct rdma_ah_attr {
914 struct ib_global_route grh;
915 u8 sl;
916 u8 static_rate;
917 u8 port_num;
918 u8 ah_flags;
919 enum rdma_ah_attr_type type;
920 union {
921 struct ib_ah_attr ib;
922 struct roce_ah_attr roce;
923 struct opa_ah_attr opa;
924 };
925};
926
927enum ib_wc_status {
928 IB_WC_SUCCESS,
929 IB_WC_LOC_LEN_ERR,
930 IB_WC_LOC_QP_OP_ERR,
931 IB_WC_LOC_EEC_OP_ERR,
932 IB_WC_LOC_PROT_ERR,
933 IB_WC_WR_FLUSH_ERR,
934 IB_WC_MW_BIND_ERR,
935 IB_WC_BAD_RESP_ERR,
936 IB_WC_LOC_ACCESS_ERR,
937 IB_WC_REM_INV_REQ_ERR,
938 IB_WC_REM_ACCESS_ERR,
939 IB_WC_REM_OP_ERR,
940 IB_WC_RETRY_EXC_ERR,
941 IB_WC_RNR_RETRY_EXC_ERR,
942 IB_WC_LOC_RDD_VIOL_ERR,
943 IB_WC_REM_INV_RD_REQ_ERR,
944 IB_WC_REM_ABORT_ERR,
945 IB_WC_INV_EECN_ERR,
946 IB_WC_INV_EEC_STATE_ERR,
947 IB_WC_FATAL_ERR,
948 IB_WC_RESP_TIMEOUT_ERR,
949 IB_WC_GENERAL_ERR
950};
951
952const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
953
954enum ib_wc_opcode {
955 IB_WC_SEND,
956 IB_WC_RDMA_WRITE,
957 IB_WC_RDMA_READ,
958 IB_WC_COMP_SWAP,
959 IB_WC_FETCH_ADD,
960 IB_WC_LSO,
961 IB_WC_LOCAL_INV,
962 IB_WC_REG_MR,
963 IB_WC_MASKED_COMP_SWAP,
964 IB_WC_MASKED_FETCH_ADD,
965/*
966 * Set value of IB_WC_RECV so consumers can test if a completion is a
967 * receive by testing (opcode & IB_WC_RECV).
968 */
969 IB_WC_RECV = 1 << 7,
970 IB_WC_RECV_RDMA_WITH_IMM
971};
972
973enum ib_wc_flags {
974 IB_WC_GRH = 1,
975 IB_WC_WITH_IMM = (1<<1),
976 IB_WC_WITH_INVALIDATE = (1<<2),
977 IB_WC_IP_CSUM_OK = (1<<3),
978 IB_WC_WITH_SMAC = (1<<4),
979 IB_WC_WITH_VLAN = (1<<5),
980 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
981};
982
983struct ib_wc {
984 union {
985 u64 wr_id;
986 struct ib_cqe *wr_cqe;
987 };
988 enum ib_wc_status status;
989 enum ib_wc_opcode opcode;
990 u32 vendor_err;
991 u32 byte_len;
992 struct ib_qp *qp;
993 union {
994 __be32 imm_data;
995 u32 invalidate_rkey;
996 } ex;
997 u32 src_qp;
998 u32 slid;
999 int wc_flags;
1000 u16 pkey_index;
1001 u8 sl;
1002 u8 dlid_path_bits;
1003 u8 port_num; /* valid only for DR SMPs on switches */
1004 u8 smac[ETH_ALEN];
1005 u16 vlan_id;
1006 u8 network_hdr_type;
1007};
1008
1009enum ib_cq_notify_flags {
1010 IB_CQ_SOLICITED = 1 << 0,
1011 IB_CQ_NEXT_COMP = 1 << 1,
1012 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1013 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1014};
1015
1016enum ib_srq_type {
1017 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1018 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1019 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1020};
1021
1022static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1023{
1024 return srq_type == IB_SRQT_XRC ||
1025 srq_type == IB_SRQT_TM;
1026}
1027
1028enum ib_srq_attr_mask {
1029 IB_SRQ_MAX_WR = 1 << 0,
1030 IB_SRQ_LIMIT = 1 << 1,
1031};
1032
1033struct ib_srq_attr {
1034 u32 max_wr;
1035 u32 max_sge;
1036 u32 srq_limit;
1037};
1038
1039struct ib_srq_init_attr {
1040 void (*event_handler)(struct ib_event *, void *);
1041 void *srq_context;
1042 struct ib_srq_attr attr;
1043 enum ib_srq_type srq_type;
1044
1045 struct {
1046 struct ib_cq *cq;
1047 union {
1048 struct {
1049 struct ib_xrcd *xrcd;
1050 } xrc;
1051
1052 struct {
1053 u32 max_num_tags;
1054 } tag_matching;
1055 };
1056 } ext;
1057};
1058
1059struct ib_qp_cap {
1060 u32 max_send_wr;
1061 u32 max_recv_wr;
1062 u32 max_send_sge;
1063 u32 max_recv_sge;
1064 u32 max_inline_data;
1065
1066 /*
1067 * Maximum number of rdma_rw_ctx structures in flight at a time.
1068 * ib_create_qp() will calculate the right amount of neededed WRs
1069 * and MRs based on this.
1070 */
1071 u32 max_rdma_ctxs;
1072};
1073
1074enum ib_sig_type {
1075 IB_SIGNAL_ALL_WR,
1076 IB_SIGNAL_REQ_WR
1077};
1078
1079enum ib_qp_type {
1080 /*
1081 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1082 * here (and in that order) since the MAD layer uses them as
1083 * indices into a 2-entry table.
1084 */
1085 IB_QPT_SMI,
1086 IB_QPT_GSI,
1087
1088 IB_QPT_RC = IB_UVERBS_QPT_RC,
1089 IB_QPT_UC = IB_UVERBS_QPT_UC,
1090 IB_QPT_UD = IB_UVERBS_QPT_UD,
1091 IB_QPT_RAW_IPV6,
1092 IB_QPT_RAW_ETHERTYPE,
1093 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1094 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1095 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1096 IB_QPT_MAX,
1097 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1098 /* Reserve a range for qp types internal to the low level driver.
1099 * These qp types will not be visible at the IB core layer, so the
1100 * IB_QPT_MAX usages should not be affected in the core layer
1101 */
1102 IB_QPT_RESERVED1 = 0x1000,
1103 IB_QPT_RESERVED2,
1104 IB_QPT_RESERVED3,
1105 IB_QPT_RESERVED4,
1106 IB_QPT_RESERVED5,
1107 IB_QPT_RESERVED6,
1108 IB_QPT_RESERVED7,
1109 IB_QPT_RESERVED8,
1110 IB_QPT_RESERVED9,
1111 IB_QPT_RESERVED10,
1112};
1113
1114enum ib_qp_create_flags {
1115 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1116 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1117 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1118 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1119 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1120 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1121 IB_QP_CREATE_NETIF_QP = 1 << 5,
1122 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1123 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1124 IB_QP_CREATE_SCATTER_FCS =
1125 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1126 IB_QP_CREATE_CVLAN_STRIPPING =
1127 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1128 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1129 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1130 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1131 /* reserve bits 26-31 for low level drivers' internal use */
1132 IB_QP_CREATE_RESERVED_START = 1 << 26,
1133 IB_QP_CREATE_RESERVED_END = 1 << 31,
1134};
1135
1136/*
1137 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1138 * callback to destroy the passed in QP.
1139 */
1140
1141struct ib_qp_init_attr {
1142 /* Consumer's event_handler callback must not block */
1143 void (*event_handler)(struct ib_event *, void *);
1144
1145 void *qp_context;
1146 struct ib_cq *send_cq;
1147 struct ib_cq *recv_cq;
1148 struct ib_srq *srq;
1149 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1150 struct ib_qp_cap cap;
1151 enum ib_sig_type sq_sig_type;
1152 enum ib_qp_type qp_type;
1153 u32 create_flags;
1154
1155 /*
1156 * Only needed for special QP types, or when using the RW API.
1157 */
1158 u8 port_num;
1159 struct ib_rwq_ind_table *rwq_ind_tbl;
1160 u32 source_qpn;
1161};
1162
1163struct ib_qp_open_attr {
1164 void (*event_handler)(struct ib_event *, void *);
1165 void *qp_context;
1166 u32 qp_num;
1167 enum ib_qp_type qp_type;
1168};
1169
1170enum ib_rnr_timeout {
1171 IB_RNR_TIMER_655_36 = 0,
1172 IB_RNR_TIMER_000_01 = 1,
1173 IB_RNR_TIMER_000_02 = 2,
1174 IB_RNR_TIMER_000_03 = 3,
1175 IB_RNR_TIMER_000_04 = 4,
1176 IB_RNR_TIMER_000_06 = 5,
1177 IB_RNR_TIMER_000_08 = 6,
1178 IB_RNR_TIMER_000_12 = 7,
1179 IB_RNR_TIMER_000_16 = 8,
1180 IB_RNR_TIMER_000_24 = 9,
1181 IB_RNR_TIMER_000_32 = 10,
1182 IB_RNR_TIMER_000_48 = 11,
1183 IB_RNR_TIMER_000_64 = 12,
1184 IB_RNR_TIMER_000_96 = 13,
1185 IB_RNR_TIMER_001_28 = 14,
1186 IB_RNR_TIMER_001_92 = 15,
1187 IB_RNR_TIMER_002_56 = 16,
1188 IB_RNR_TIMER_003_84 = 17,
1189 IB_RNR_TIMER_005_12 = 18,
1190 IB_RNR_TIMER_007_68 = 19,
1191 IB_RNR_TIMER_010_24 = 20,
1192 IB_RNR_TIMER_015_36 = 21,
1193 IB_RNR_TIMER_020_48 = 22,
1194 IB_RNR_TIMER_030_72 = 23,
1195 IB_RNR_TIMER_040_96 = 24,
1196 IB_RNR_TIMER_061_44 = 25,
1197 IB_RNR_TIMER_081_92 = 26,
1198 IB_RNR_TIMER_122_88 = 27,
1199 IB_RNR_TIMER_163_84 = 28,
1200 IB_RNR_TIMER_245_76 = 29,
1201 IB_RNR_TIMER_327_68 = 30,
1202 IB_RNR_TIMER_491_52 = 31
1203};
1204
1205enum ib_qp_attr_mask {
1206 IB_QP_STATE = 1,
1207 IB_QP_CUR_STATE = (1<<1),
1208 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1209 IB_QP_ACCESS_FLAGS = (1<<3),
1210 IB_QP_PKEY_INDEX = (1<<4),
1211 IB_QP_PORT = (1<<5),
1212 IB_QP_QKEY = (1<<6),
1213 IB_QP_AV = (1<<7),
1214 IB_QP_PATH_MTU = (1<<8),
1215 IB_QP_TIMEOUT = (1<<9),
1216 IB_QP_RETRY_CNT = (1<<10),
1217 IB_QP_RNR_RETRY = (1<<11),
1218 IB_QP_RQ_PSN = (1<<12),
1219 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1220 IB_QP_ALT_PATH = (1<<14),
1221 IB_QP_MIN_RNR_TIMER = (1<<15),
1222 IB_QP_SQ_PSN = (1<<16),
1223 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1224 IB_QP_PATH_MIG_STATE = (1<<18),
1225 IB_QP_CAP = (1<<19),
1226 IB_QP_DEST_QPN = (1<<20),
1227 IB_QP_RESERVED1 = (1<<21),
1228 IB_QP_RESERVED2 = (1<<22),
1229 IB_QP_RESERVED3 = (1<<23),
1230 IB_QP_RESERVED4 = (1<<24),
1231 IB_QP_RATE_LIMIT = (1<<25),
1232};
1233
1234enum ib_qp_state {
1235 IB_QPS_RESET,
1236 IB_QPS_INIT,
1237 IB_QPS_RTR,
1238 IB_QPS_RTS,
1239 IB_QPS_SQD,
1240 IB_QPS_SQE,
1241 IB_QPS_ERR
1242};
1243
1244enum ib_mig_state {
1245 IB_MIG_MIGRATED,
1246 IB_MIG_REARM,
1247 IB_MIG_ARMED
1248};
1249
1250enum ib_mw_type {
1251 IB_MW_TYPE_1 = 1,
1252 IB_MW_TYPE_2 = 2
1253};
1254
1255struct ib_qp_attr {
1256 enum ib_qp_state qp_state;
1257 enum ib_qp_state cur_qp_state;
1258 enum ib_mtu path_mtu;
1259 enum ib_mig_state path_mig_state;
1260 u32 qkey;
1261 u32 rq_psn;
1262 u32 sq_psn;
1263 u32 dest_qp_num;
1264 int qp_access_flags;
1265 struct ib_qp_cap cap;
1266 struct rdma_ah_attr ah_attr;
1267 struct rdma_ah_attr alt_ah_attr;
1268 u16 pkey_index;
1269 u16 alt_pkey_index;
1270 u8 en_sqd_async_notify;
1271 u8 sq_draining;
1272 u8 max_rd_atomic;
1273 u8 max_dest_rd_atomic;
1274 u8 min_rnr_timer;
1275 u8 port_num;
1276 u8 timeout;
1277 u8 retry_cnt;
1278 u8 rnr_retry;
1279 u8 alt_port_num;
1280 u8 alt_timeout;
1281 u32 rate_limit;
1282 struct net_device *xmit_slave;
1283};
1284
1285enum ib_wr_opcode {
1286 /* These are shared with userspace */
1287 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1288 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1289 IB_WR_SEND = IB_UVERBS_WR_SEND,
1290 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1291 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1292 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1293 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1294 IB_WR_LSO = IB_UVERBS_WR_TSO,
1295 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1296 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1297 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1298 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1299 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1300 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1301 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1302
1303 /* These are kernel only and can not be issued by userspace */
1304 IB_WR_REG_MR = 0x20,
1305 IB_WR_REG_MR_INTEGRITY,
1306
1307 /* reserve values for low level drivers' internal use.
1308 * These values will not be used at all in the ib core layer.
1309 */
1310 IB_WR_RESERVED1 = 0xf0,
1311 IB_WR_RESERVED2,
1312 IB_WR_RESERVED3,
1313 IB_WR_RESERVED4,
1314 IB_WR_RESERVED5,
1315 IB_WR_RESERVED6,
1316 IB_WR_RESERVED7,
1317 IB_WR_RESERVED8,
1318 IB_WR_RESERVED9,
1319 IB_WR_RESERVED10,
1320};
1321
1322enum ib_send_flags {
1323 IB_SEND_FENCE = 1,
1324 IB_SEND_SIGNALED = (1<<1),
1325 IB_SEND_SOLICITED = (1<<2),
1326 IB_SEND_INLINE = (1<<3),
1327 IB_SEND_IP_CSUM = (1<<4),
1328
1329 /* reserve bits 26-31 for low level drivers' internal use */
1330 IB_SEND_RESERVED_START = (1 << 26),
1331 IB_SEND_RESERVED_END = (1 << 31),
1332};
1333
1334struct ib_sge {
1335 u64 addr;
1336 u32 length;
1337 u32 lkey;
1338};
1339
1340struct ib_cqe {
1341 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1342};
1343
1344struct ib_send_wr {
1345 struct ib_send_wr *next;
1346 union {
1347 u64 wr_id;
1348 struct ib_cqe *wr_cqe;
1349 };
1350 struct ib_sge *sg_list;
1351 int num_sge;
1352 enum ib_wr_opcode opcode;
1353 int send_flags;
1354 union {
1355 __be32 imm_data;
1356 u32 invalidate_rkey;
1357 } ex;
1358};
1359
1360struct ib_rdma_wr {
1361 struct ib_send_wr wr;
1362 u64 remote_addr;
1363 u32 rkey;
1364};
1365
1366static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1367{
1368 return container_of(wr, struct ib_rdma_wr, wr);
1369}
1370
1371struct ib_atomic_wr {
1372 struct ib_send_wr wr;
1373 u64 remote_addr;
1374 u64 compare_add;
1375 u64 swap;
1376 u64 compare_add_mask;
1377 u64 swap_mask;
1378 u32 rkey;
1379};
1380
1381static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1382{
1383 return container_of(wr, struct ib_atomic_wr, wr);
1384}
1385
1386struct ib_ud_wr {
1387 struct ib_send_wr wr;
1388 struct ib_ah *ah;
1389 void *header;
1390 int hlen;
1391 int mss;
1392 u32 remote_qpn;
1393 u32 remote_qkey;
1394 u16 pkey_index; /* valid for GSI only */
1395 u8 port_num; /* valid for DR SMPs on switch only */
1396};
1397
1398static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1399{
1400 return container_of(wr, struct ib_ud_wr, wr);
1401}
1402
1403struct ib_reg_wr {
1404 struct ib_send_wr wr;
1405 struct ib_mr *mr;
1406 u32 key;
1407 int access;
1408};
1409
1410static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1411{
1412 return container_of(wr, struct ib_reg_wr, wr);
1413}
1414
1415struct ib_recv_wr {
1416 struct ib_recv_wr *next;
1417 union {
1418 u64 wr_id;
1419 struct ib_cqe *wr_cqe;
1420 };
1421 struct ib_sge *sg_list;
1422 int num_sge;
1423};
1424
1425enum ib_access_flags {
1426 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1427 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1428 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1429 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1430 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1431 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1432 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1433 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1434 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1435
1436 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1437 IB_ACCESS_SUPPORTED =
1438 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL,
1439};
1440
1441/*
1442 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1443 * are hidden here instead of a uapi header!
1444 */
1445enum ib_mr_rereg_flags {
1446 IB_MR_REREG_TRANS = 1,
1447 IB_MR_REREG_PD = (1<<1),
1448 IB_MR_REREG_ACCESS = (1<<2),
1449 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1450};
1451
1452struct ib_umem;
1453
1454enum rdma_remove_reason {
1455 /*
1456 * Userspace requested uobject deletion or initial try
1457 * to remove uobject via cleanup. Call could fail
1458 */
1459 RDMA_REMOVE_DESTROY,
1460 /* Context deletion. This call should delete the actual object itself */
1461 RDMA_REMOVE_CLOSE,
1462 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1463 RDMA_REMOVE_DRIVER_REMOVE,
1464 /* uobj is being cleaned-up before being committed */
1465 RDMA_REMOVE_ABORT,
1466 /*
1467 * uobj has been fully created, with the uobj->object set, but is being
1468 * cleaned up before being comitted
1469 */
1470 RDMA_REMOVE_ABORT_HWOBJ,
1471};
1472
1473struct ib_rdmacg_object {
1474#ifdef CONFIG_CGROUP_RDMA
1475 struct rdma_cgroup *cg; /* owner rdma cgroup */
1476#endif
1477};
1478
1479struct ib_ucontext {
1480 struct ib_device *device;
1481 struct ib_uverbs_file *ufile;
1482 /*
1483 * 'closing' can be read by the driver only during a destroy callback,
1484 * it is set when we are closing the file descriptor and indicates
1485 * that mm_sem may be locked.
1486 */
1487 bool closing;
1488
1489 bool cleanup_retryable;
1490
1491 struct ib_rdmacg_object cg_obj;
1492 /*
1493 * Implementation details of the RDMA core, don't use in drivers:
1494 */
1495 struct rdma_restrack_entry res;
1496 struct xarray mmap_xa;
1497};
1498
1499struct ib_uobject {
1500 u64 user_handle; /* handle given to us by userspace */
1501 /* ufile & ucontext owning this object */
1502 struct ib_uverbs_file *ufile;
1503 /* FIXME, save memory: ufile->context == context */
1504 struct ib_ucontext *context; /* associated user context */
1505 void *object; /* containing object */
1506 struct list_head list; /* link to context's list */
1507 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1508 int id; /* index into kernel idr */
1509 struct kref ref;
1510 atomic_t usecnt; /* protects exclusive access */
1511 struct rcu_head rcu; /* kfree_rcu() overhead */
1512
1513 const struct uverbs_api_object *uapi_object;
1514};
1515
1516struct ib_udata {
1517 const void __user *inbuf;
1518 void __user *outbuf;
1519 size_t inlen;
1520 size_t outlen;
1521};
1522
1523struct ib_pd {
1524 u32 local_dma_lkey;
1525 u32 flags;
1526 struct ib_device *device;
1527 struct ib_uobject *uobject;
1528 atomic_t usecnt; /* count all resources */
1529
1530 u32 unsafe_global_rkey;
1531
1532 /*
1533 * Implementation details of the RDMA core, don't use in drivers:
1534 */
1535 struct ib_mr *__internal_mr;
1536 struct rdma_restrack_entry res;
1537};
1538
1539struct ib_xrcd {
1540 struct ib_device *device;
1541 atomic_t usecnt; /* count all exposed resources */
1542 struct inode *inode;
1543 struct rw_semaphore tgt_qps_rwsem;
1544 struct xarray tgt_qps;
1545};
1546
1547struct ib_ah {
1548 struct ib_device *device;
1549 struct ib_pd *pd;
1550 struct ib_uobject *uobject;
1551 const struct ib_gid_attr *sgid_attr;
1552 enum rdma_ah_attr_type type;
1553};
1554
1555typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1556
1557enum ib_poll_context {
1558 IB_POLL_SOFTIRQ, /* poll from softirq context */
1559 IB_POLL_WORKQUEUE, /* poll from workqueue */
1560 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1561 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1562
1563 IB_POLL_DIRECT, /* caller context, no hw completions */
1564};
1565
1566struct ib_cq {
1567 struct ib_device *device;
1568 struct ib_ucq_object *uobject;
1569 ib_comp_handler comp_handler;
1570 void (*event_handler)(struct ib_event *, void *);
1571 void *cq_context;
1572 int cqe;
1573 unsigned int cqe_used;
1574 atomic_t usecnt; /* count number of work queues */
1575 enum ib_poll_context poll_ctx;
1576 struct ib_wc *wc;
1577 struct list_head pool_entry;
1578 union {
1579 struct irq_poll iop;
1580 struct work_struct work;
1581 };
1582 struct workqueue_struct *comp_wq;
1583 struct dim *dim;
1584
1585 /* updated only by trace points */
1586 ktime_t timestamp;
1587 u8 interrupt:1;
1588 u8 shared:1;
1589 unsigned int comp_vector;
1590
1591 /*
1592 * Implementation details of the RDMA core, don't use in drivers:
1593 */
1594 struct rdma_restrack_entry res;
1595};
1596
1597struct ib_srq {
1598 struct ib_device *device;
1599 struct ib_pd *pd;
1600 struct ib_usrq_object *uobject;
1601 void (*event_handler)(struct ib_event *, void *);
1602 void *srq_context;
1603 enum ib_srq_type srq_type;
1604 atomic_t usecnt;
1605
1606 struct {
1607 struct ib_cq *cq;
1608 union {
1609 struct {
1610 struct ib_xrcd *xrcd;
1611 u32 srq_num;
1612 } xrc;
1613 };
1614 } ext;
1615};
1616
1617enum ib_raw_packet_caps {
1618 /* Strip cvlan from incoming packet and report it in the matching work
1619 * completion is supported.
1620 */
1621 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1622 /* Scatter FCS field of an incoming packet to host memory is supported.
1623 */
1624 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1625 /* Checksum offloads are supported (for both send and receive). */
1626 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
1627 /* When a packet is received for an RQ with no receive WQEs, the
1628 * packet processing is delayed.
1629 */
1630 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
1631};
1632
1633enum ib_wq_type {
1634 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1635};
1636
1637enum ib_wq_state {
1638 IB_WQS_RESET,
1639 IB_WQS_RDY,
1640 IB_WQS_ERR
1641};
1642
1643struct ib_wq {
1644 struct ib_device *device;
1645 struct ib_uwq_object *uobject;
1646 void *wq_context;
1647 void (*event_handler)(struct ib_event *, void *);
1648 struct ib_pd *pd;
1649 struct ib_cq *cq;
1650 u32 wq_num;
1651 enum ib_wq_state state;
1652 enum ib_wq_type wq_type;
1653 atomic_t usecnt;
1654};
1655
1656enum ib_wq_flags {
1657 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1658 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1659 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1660 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1661 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1662};
1663
1664struct ib_wq_init_attr {
1665 void *wq_context;
1666 enum ib_wq_type wq_type;
1667 u32 max_wr;
1668 u32 max_sge;
1669 struct ib_cq *cq;
1670 void (*event_handler)(struct ib_event *, void *);
1671 u32 create_flags; /* Use enum ib_wq_flags */
1672};
1673
1674enum ib_wq_attr_mask {
1675 IB_WQ_STATE = 1 << 0,
1676 IB_WQ_CUR_STATE = 1 << 1,
1677 IB_WQ_FLAGS = 1 << 2,
1678};
1679
1680struct ib_wq_attr {
1681 enum ib_wq_state wq_state;
1682 enum ib_wq_state curr_wq_state;
1683 u32 flags; /* Use enum ib_wq_flags */
1684 u32 flags_mask; /* Use enum ib_wq_flags */
1685};
1686
1687struct ib_rwq_ind_table {
1688 struct ib_device *device;
1689 struct ib_uobject *uobject;
1690 atomic_t usecnt;
1691 u32 ind_tbl_num;
1692 u32 log_ind_tbl_size;
1693 struct ib_wq **ind_tbl;
1694};
1695
1696struct ib_rwq_ind_table_init_attr {
1697 u32 log_ind_tbl_size;
1698 /* Each entry is a pointer to Receive Work Queue */
1699 struct ib_wq **ind_tbl;
1700};
1701
1702enum port_pkey_state {
1703 IB_PORT_PKEY_NOT_VALID = 0,
1704 IB_PORT_PKEY_VALID = 1,
1705 IB_PORT_PKEY_LISTED = 2,
1706};
1707
1708struct ib_qp_security;
1709
1710struct ib_port_pkey {
1711 enum port_pkey_state state;
1712 u16 pkey_index;
1713 u8 port_num;
1714 struct list_head qp_list;
1715 struct list_head to_error_list;
1716 struct ib_qp_security *sec;
1717};
1718
1719struct ib_ports_pkeys {
1720 struct ib_port_pkey main;
1721 struct ib_port_pkey alt;
1722};
1723
1724struct ib_qp_security {
1725 struct ib_qp *qp;
1726 struct ib_device *dev;
1727 /* Hold this mutex when changing port and pkey settings. */
1728 struct mutex mutex;
1729 struct ib_ports_pkeys *ports_pkeys;
1730 /* A list of all open shared QP handles. Required to enforce security
1731 * properly for all users of a shared QP.
1732 */
1733 struct list_head shared_qp_list;
1734 void *security;
1735 bool destroying;
1736 atomic_t error_list_count;
1737 struct completion error_complete;
1738 int error_comps_pending;
1739};
1740
1741/*
1742 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1743 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1744 */
1745struct ib_qp {
1746 struct ib_device *device;
1747 struct ib_pd *pd;
1748 struct ib_cq *send_cq;
1749 struct ib_cq *recv_cq;
1750 spinlock_t mr_lock;
1751 int mrs_used;
1752 struct list_head rdma_mrs;
1753 struct list_head sig_mrs;
1754 struct ib_srq *srq;
1755 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1756 struct list_head xrcd_list;
1757
1758 /* count times opened, mcast attaches, flow attaches */
1759 atomic_t usecnt;
1760 struct list_head open_list;
1761 struct ib_qp *real_qp;
1762 struct ib_uqp_object *uobject;
1763 void (*event_handler)(struct ib_event *, void *);
1764 void *qp_context;
1765 /* sgid_attrs associated with the AV's */
1766 const struct ib_gid_attr *av_sgid_attr;
1767 const struct ib_gid_attr *alt_path_sgid_attr;
1768 u32 qp_num;
1769 u32 max_write_sge;
1770 u32 max_read_sge;
1771 enum ib_qp_type qp_type;
1772 struct ib_rwq_ind_table *rwq_ind_tbl;
1773 struct ib_qp_security *qp_sec;
1774 u8 port;
1775
1776 bool integrity_en;
1777 /*
1778 * Implementation details of the RDMA core, don't use in drivers:
1779 */
1780 struct rdma_restrack_entry res;
1781
1782 /* The counter the qp is bind to */
1783 struct rdma_counter *counter;
1784};
1785
1786struct ib_dm {
1787 struct ib_device *device;
1788 u32 length;
1789 u32 flags;
1790 struct ib_uobject *uobject;
1791 atomic_t usecnt;
1792};
1793
1794struct ib_mr {
1795 struct ib_device *device;
1796 struct ib_pd *pd;
1797 u32 lkey;
1798 u32 rkey;
1799 u64 iova;
1800 u64 length;
1801 unsigned int page_size;
1802 enum ib_mr_type type;
1803 bool need_inval;
1804 union {
1805 struct ib_uobject *uobject; /* user */
1806 struct list_head qp_entry; /* FR */
1807 };
1808
1809 struct ib_dm *dm;
1810 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1811 /*
1812 * Implementation details of the RDMA core, don't use in drivers:
1813 */
1814 struct rdma_restrack_entry res;
1815};
1816
1817struct ib_mw {
1818 struct ib_device *device;
1819 struct ib_pd *pd;
1820 struct ib_uobject *uobject;
1821 u32 rkey;
1822 enum ib_mw_type type;
1823};
1824
1825/* Supported steering options */
1826enum ib_flow_attr_type {
1827 /* steering according to rule specifications */
1828 IB_FLOW_ATTR_NORMAL = 0x0,
1829 /* default unicast and multicast rule -
1830 * receive all Eth traffic which isn't steered to any QP
1831 */
1832 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1833 /* default multicast rule -
1834 * receive all Eth multicast traffic which isn't steered to any QP
1835 */
1836 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1837 /* sniffer rule - receive all port traffic */
1838 IB_FLOW_ATTR_SNIFFER = 0x3
1839};
1840
1841/* Supported steering header types */
1842enum ib_flow_spec_type {
1843 /* L2 headers*/
1844 IB_FLOW_SPEC_ETH = 0x20,
1845 IB_FLOW_SPEC_IB = 0x22,
1846 /* L3 header*/
1847 IB_FLOW_SPEC_IPV4 = 0x30,
1848 IB_FLOW_SPEC_IPV6 = 0x31,
1849 IB_FLOW_SPEC_ESP = 0x34,
1850 /* L4 headers*/
1851 IB_FLOW_SPEC_TCP = 0x40,
1852 IB_FLOW_SPEC_UDP = 0x41,
1853 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1854 IB_FLOW_SPEC_GRE = 0x51,
1855 IB_FLOW_SPEC_MPLS = 0x60,
1856 IB_FLOW_SPEC_INNER = 0x100,
1857 /* Actions */
1858 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1859 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1860 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1861 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1862};
1863#define IB_FLOW_SPEC_LAYER_MASK 0xF0
1864#define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1865
1866/* Flow steering rule priority is set according to it's domain.
1867 * Lower domain value means higher priority.
1868 */
1869enum ib_flow_domain {
1870 IB_FLOW_DOMAIN_USER,
1871 IB_FLOW_DOMAIN_ETHTOOL,
1872 IB_FLOW_DOMAIN_RFS,
1873 IB_FLOW_DOMAIN_NIC,
1874 IB_FLOW_DOMAIN_NUM /* Must be last */
1875};
1876
1877enum ib_flow_flags {
1878 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1879 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1880 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1881};
1882
1883struct ib_flow_eth_filter {
1884 u8 dst_mac[6];
1885 u8 src_mac[6];
1886 __be16 ether_type;
1887 __be16 vlan_tag;
1888 /* Must be last */
1889 u8 real_sz[];
1890};
1891
1892struct ib_flow_spec_eth {
1893 u32 type;
1894 u16 size;
1895 struct ib_flow_eth_filter val;
1896 struct ib_flow_eth_filter mask;
1897};
1898
1899struct ib_flow_ib_filter {
1900 __be16 dlid;
1901 __u8 sl;
1902 /* Must be last */
1903 u8 real_sz[];
1904};
1905
1906struct ib_flow_spec_ib {
1907 u32 type;
1908 u16 size;
1909 struct ib_flow_ib_filter val;
1910 struct ib_flow_ib_filter mask;
1911};
1912
1913/* IPv4 header flags */
1914enum ib_ipv4_flags {
1915 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1916 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1917 last have this flag set */
1918};
1919
1920struct ib_flow_ipv4_filter {
1921 __be32 src_ip;
1922 __be32 dst_ip;
1923 u8 proto;
1924 u8 tos;
1925 u8 ttl;
1926 u8 flags;
1927 /* Must be last */
1928 u8 real_sz[];
1929};
1930
1931struct ib_flow_spec_ipv4 {
1932 u32 type;
1933 u16 size;
1934 struct ib_flow_ipv4_filter val;
1935 struct ib_flow_ipv4_filter mask;
1936};
1937
1938struct ib_flow_ipv6_filter {
1939 u8 src_ip[16];
1940 u8 dst_ip[16];
1941 __be32 flow_label;
1942 u8 next_hdr;
1943 u8 traffic_class;
1944 u8 hop_limit;
1945 /* Must be last */
1946 u8 real_sz[];
1947};
1948
1949struct ib_flow_spec_ipv6 {
1950 u32 type;
1951 u16 size;
1952 struct ib_flow_ipv6_filter val;
1953 struct ib_flow_ipv6_filter mask;
1954};
1955
1956struct ib_flow_tcp_udp_filter {
1957 __be16 dst_port;
1958 __be16 src_port;
1959 /* Must be last */
1960 u8 real_sz[];
1961};
1962
1963struct ib_flow_spec_tcp_udp {
1964 u32 type;
1965 u16 size;
1966 struct ib_flow_tcp_udp_filter val;
1967 struct ib_flow_tcp_udp_filter mask;
1968};
1969
1970struct ib_flow_tunnel_filter {
1971 __be32 tunnel_id;
1972 u8 real_sz[];
1973};
1974
1975/* ib_flow_spec_tunnel describes the Vxlan tunnel
1976 * the tunnel_id from val has the vni value
1977 */
1978struct ib_flow_spec_tunnel {
1979 u32 type;
1980 u16 size;
1981 struct ib_flow_tunnel_filter val;
1982 struct ib_flow_tunnel_filter mask;
1983};
1984
1985struct ib_flow_esp_filter {
1986 __be32 spi;
1987 __be32 seq;
1988 /* Must be last */
1989 u8 real_sz[];
1990};
1991
1992struct ib_flow_spec_esp {
1993 u32 type;
1994 u16 size;
1995 struct ib_flow_esp_filter val;
1996 struct ib_flow_esp_filter mask;
1997};
1998
1999struct ib_flow_gre_filter {
2000 __be16 c_ks_res0_ver;
2001 __be16 protocol;
2002 __be32 key;
2003 /* Must be last */
2004 u8 real_sz[];
2005};
2006
2007struct ib_flow_spec_gre {
2008 u32 type;
2009 u16 size;
2010 struct ib_flow_gre_filter val;
2011 struct ib_flow_gre_filter mask;
2012};
2013
2014struct ib_flow_mpls_filter {
2015 __be32 tag;
2016 /* Must be last */
2017 u8 real_sz[];
2018};
2019
2020struct ib_flow_spec_mpls {
2021 u32 type;
2022 u16 size;
2023 struct ib_flow_mpls_filter val;
2024 struct ib_flow_mpls_filter mask;
2025};
2026
2027struct ib_flow_spec_action_tag {
2028 enum ib_flow_spec_type type;
2029 u16 size;
2030 u32 tag_id;
2031};
2032
2033struct ib_flow_spec_action_drop {
2034 enum ib_flow_spec_type type;
2035 u16 size;
2036};
2037
2038struct ib_flow_spec_action_handle {
2039 enum ib_flow_spec_type type;
2040 u16 size;
2041 struct ib_flow_action *act;
2042};
2043
2044enum ib_counters_description {
2045 IB_COUNTER_PACKETS,
2046 IB_COUNTER_BYTES,
2047};
2048
2049struct ib_flow_spec_action_count {
2050 enum ib_flow_spec_type type;
2051 u16 size;
2052 struct ib_counters *counters;
2053};
2054
2055union ib_flow_spec {
2056 struct {
2057 u32 type;
2058 u16 size;
2059 };
2060 struct ib_flow_spec_eth eth;
2061 struct ib_flow_spec_ib ib;
2062 struct ib_flow_spec_ipv4 ipv4;
2063 struct ib_flow_spec_tcp_udp tcp_udp;
2064 struct ib_flow_spec_ipv6 ipv6;
2065 struct ib_flow_spec_tunnel tunnel;
2066 struct ib_flow_spec_esp esp;
2067 struct ib_flow_spec_gre gre;
2068 struct ib_flow_spec_mpls mpls;
2069 struct ib_flow_spec_action_tag flow_tag;
2070 struct ib_flow_spec_action_drop drop;
2071 struct ib_flow_spec_action_handle action;
2072 struct ib_flow_spec_action_count flow_count;
2073};
2074
2075struct ib_flow_attr {
2076 enum ib_flow_attr_type type;
2077 u16 size;
2078 u16 priority;
2079 u32 flags;
2080 u8 num_of_specs;
2081 u8 port;
2082 union ib_flow_spec flows[];
2083};
2084
2085struct ib_flow {
2086 struct ib_qp *qp;
2087 struct ib_device *device;
2088 struct ib_uobject *uobject;
2089};
2090
2091enum ib_flow_action_type {
2092 IB_FLOW_ACTION_UNSPECIFIED,
2093 IB_FLOW_ACTION_ESP = 1,
2094};
2095
2096struct ib_flow_action_attrs_esp_keymats {
2097 enum ib_uverbs_flow_action_esp_keymat protocol;
2098 union {
2099 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2100 } keymat;
2101};
2102
2103struct ib_flow_action_attrs_esp_replays {
2104 enum ib_uverbs_flow_action_esp_replay protocol;
2105 union {
2106 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2107 } replay;
2108};
2109
2110enum ib_flow_action_attrs_esp_flags {
2111 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2112 * This is done in order to share the same flags between user-space and
2113 * kernel and spare an unnecessary translation.
2114 */
2115
2116 /* Kernel flags */
2117 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2118 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2119};
2120
2121struct ib_flow_spec_list {
2122 struct ib_flow_spec_list *next;
2123 union ib_flow_spec spec;
2124};
2125
2126struct ib_flow_action_attrs_esp {
2127 struct ib_flow_action_attrs_esp_keymats *keymat;
2128 struct ib_flow_action_attrs_esp_replays *replay;
2129 struct ib_flow_spec_list *encap;
2130 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2131 * Value of 0 is a valid value.
2132 */
2133 u32 esn;
2134 u32 spi;
2135 u32 seq;
2136 u32 tfc_pad;
2137 /* Use enum ib_flow_action_attrs_esp_flags */
2138 u64 flags;
2139 u64 hard_limit_pkts;
2140};
2141
2142struct ib_flow_action {
2143 struct ib_device *device;
2144 struct ib_uobject *uobject;
2145 enum ib_flow_action_type type;
2146 atomic_t usecnt;
2147};
2148
2149struct ib_mad;
2150struct ib_grh;
2151
2152enum ib_process_mad_flags {
2153 IB_MAD_IGNORE_MKEY = 1,
2154 IB_MAD_IGNORE_BKEY = 2,
2155 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2156};
2157
2158enum ib_mad_result {
2159 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2160 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2161 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2162 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2163};
2164
2165struct ib_port_cache {
2166 u64 subnet_prefix;
2167 struct ib_pkey_cache *pkey;
2168 struct ib_gid_table *gid;
2169 u8 lmc;
2170 enum ib_port_state port_state;
2171};
2172
2173struct ib_port_immutable {
2174 int pkey_tbl_len;
2175 int gid_tbl_len;
2176 u32 core_cap_flags;
2177 u32 max_mad_size;
2178};
2179
2180struct ib_port_data {
2181 struct ib_device *ib_dev;
2182
2183 struct ib_port_immutable immutable;
2184
2185 spinlock_t pkey_list_lock;
2186 struct list_head pkey_list;
2187
2188 struct ib_port_cache cache;
2189
2190 spinlock_t netdev_lock;
2191 struct net_device __rcu *netdev;
2192 struct hlist_node ndev_hash_link;
2193 struct rdma_port_counter port_counter;
2194 struct rdma_hw_stats *hw_stats;
2195};
2196
2197/* rdma netdev type - specifies protocol type */
2198enum rdma_netdev_t {
2199 RDMA_NETDEV_OPA_VNIC,
2200 RDMA_NETDEV_IPOIB,
2201};
2202
2203/**
2204 * struct rdma_netdev - rdma netdev
2205 * For cases where netstack interfacing is required.
2206 */
2207struct rdma_netdev {
2208 void *clnt_priv;
2209 struct ib_device *hca;
2210 u8 port_num;
2211 int mtu;
2212
2213 /*
2214 * cleanup function must be specified.
2215 * FIXME: This is only used for OPA_VNIC and that usage should be
2216 * removed too.
2217 */
2218 void (*free_rdma_netdev)(struct net_device *netdev);
2219
2220 /* control functions */
2221 void (*set_id)(struct net_device *netdev, int id);
2222 /* send packet */
2223 int (*send)(struct net_device *dev, struct sk_buff *skb,
2224 struct ib_ah *address, u32 dqpn);
2225 /* multicast */
2226 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2227 union ib_gid *gid, u16 mlid,
2228 int set_qkey, u32 qkey);
2229 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2230 union ib_gid *gid, u16 mlid);
2231};
2232
2233struct rdma_netdev_alloc_params {
2234 size_t sizeof_priv;
2235 unsigned int txqs;
2236 unsigned int rxqs;
2237 void *param;
2238
2239 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2240 struct net_device *netdev, void *param);
2241};
2242
2243struct ib_odp_counters {
2244 atomic64_t faults;
2245 atomic64_t invalidations;
2246 atomic64_t prefetch;
2247};
2248
2249struct ib_counters {
2250 struct ib_device *device;
2251 struct ib_uobject *uobject;
2252 /* num of objects attached */
2253 atomic_t usecnt;
2254};
2255
2256struct ib_counters_read_attr {
2257 u64 *counters_buff;
2258 u32 ncounters;
2259 u32 flags; /* use enum ib_read_counters_flags */
2260};
2261
2262struct uverbs_attr_bundle;
2263struct iw_cm_id;
2264struct iw_cm_conn_param;
2265
2266#define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2267 .size_##ib_struct = \
2268 (sizeof(struct drv_struct) + \
2269 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2270 BUILD_BUG_ON_ZERO( \
2271 !__same_type(((struct drv_struct *)NULL)->member, \
2272 struct ib_struct)))
2273
2274#define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2275 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2276
2277#define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2278 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2279
2280#define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2281
2282struct rdma_user_mmap_entry {
2283 struct kref ref;
2284 struct ib_ucontext *ucontext;
2285 unsigned long start_pgoff;
2286 size_t npages;
2287 bool driver_removed;
2288};
2289
2290/* Return the offset (in bytes) the user should pass to libc's mmap() */
2291static inline u64
2292rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2293{
2294 return (u64)entry->start_pgoff << PAGE_SHIFT;
2295}
2296
2297/**
2298 * struct ib_device_ops - InfiniBand device operations
2299 * This structure defines all the InfiniBand device operations, providers will
2300 * need to define the supported operations, otherwise they will be set to null.
2301 */
2302struct ib_device_ops {
2303 struct module *owner;
2304 enum rdma_driver_id driver_id;
2305 u32 uverbs_abi_ver;
2306 unsigned int uverbs_no_driver_id_binding:1;
2307
2308 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2309 const struct ib_send_wr **bad_send_wr);
2310 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2311 const struct ib_recv_wr **bad_recv_wr);
2312 void (*drain_rq)(struct ib_qp *qp);
2313 void (*drain_sq)(struct ib_qp *qp);
2314 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2315 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2316 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2317 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2318 int (*post_srq_recv)(struct ib_srq *srq,
2319 const struct ib_recv_wr *recv_wr,
2320 const struct ib_recv_wr **bad_recv_wr);
2321 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2322 u8 port_num, const struct ib_wc *in_wc,
2323 const struct ib_grh *in_grh,
2324 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2325 size_t *out_mad_size, u16 *out_mad_pkey_index);
2326 int (*query_device)(struct ib_device *device,
2327 struct ib_device_attr *device_attr,
2328 struct ib_udata *udata);
2329 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2330 struct ib_device_modify *device_modify);
2331 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2332 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2333 int comp_vector);
2334 int (*query_port)(struct ib_device *device, u8 port_num,
2335 struct ib_port_attr *port_attr);
2336 int (*modify_port)(struct ib_device *device, u8 port_num,
2337 int port_modify_mask,
2338 struct ib_port_modify *port_modify);
2339 /**
2340 * The following mandatory functions are used only at device
2341 * registration. Keep functions such as these at the end of this
2342 * structure to avoid cache line misses when accessing struct ib_device
2343 * in fast paths.
2344 */
2345 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2346 struct ib_port_immutable *immutable);
2347 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2348 u8 port_num);
2349 /**
2350 * When calling get_netdev, the HW vendor's driver should return the
2351 * net device of device @device at port @port_num or NULL if such
2352 * a net device doesn't exist. The vendor driver should call dev_hold
2353 * on this net device. The HW vendor's device driver must guarantee
2354 * that this function returns NULL before the net device has finished
2355 * NETDEV_UNREGISTER state.
2356 */
2357 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2358 /**
2359 * rdma netdev operation
2360 *
2361 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2362 * must return -EOPNOTSUPP if it doesn't support the specified type.
2363 */
2364 struct net_device *(*alloc_rdma_netdev)(
2365 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2366 const char *name, unsigned char name_assign_type,
2367 void (*setup)(struct net_device *));
2368
2369 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2370 enum rdma_netdev_t type,
2371 struct rdma_netdev_alloc_params *params);
2372 /**
2373 * query_gid should be return GID value for @device, when @port_num
2374 * link layer is either IB or iWarp. It is no-op if @port_num port
2375 * is RoCE link layer.
2376 */
2377 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2378 union ib_gid *gid);
2379 /**
2380 * When calling add_gid, the HW vendor's driver should add the gid
2381 * of device of port at gid index available at @attr. Meta-info of
2382 * that gid (for example, the network device related to this gid) is
2383 * available at @attr. @context allows the HW vendor driver to store
2384 * extra information together with a GID entry. The HW vendor driver may
2385 * allocate memory to contain this information and store it in @context
2386 * when a new GID entry is written to. Params are consistent until the
2387 * next call of add_gid or delete_gid. The function should return 0 on
2388 * success or error otherwise. The function could be called
2389 * concurrently for different ports. This function is only called when
2390 * roce_gid_table is used.
2391 */
2392 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2393 /**
2394 * When calling del_gid, the HW vendor's driver should delete the
2395 * gid of device @device at gid index gid_index of port port_num
2396 * available in @attr.
2397 * Upon the deletion of a GID entry, the HW vendor must free any
2398 * allocated memory. The caller will clear @context afterwards.
2399 * This function is only called when roce_gid_table is used.
2400 */
2401 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2402 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2403 u16 *pkey);
2404 int (*alloc_ucontext)(struct ib_ucontext *context,
2405 struct ib_udata *udata);
2406 void (*dealloc_ucontext)(struct ib_ucontext *context);
2407 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2408 /**
2409 * This will be called once refcount of an entry in mmap_xa reaches
2410 * zero. The type of the memory that was mapped may differ between
2411 * entries and is opaque to the rdma_user_mmap interface.
2412 * Therefore needs to be implemented by the driver in mmap_free.
2413 */
2414 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2415 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2416 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2417 void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2418 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2419 struct ib_udata *udata);
2420 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2421 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2422 void (*destroy_ah)(struct ib_ah *ah, u32 flags);
2423 int (*create_srq)(struct ib_srq *srq,
2424 struct ib_srq_init_attr *srq_init_attr,
2425 struct ib_udata *udata);
2426 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2427 enum ib_srq_attr_mask srq_attr_mask,
2428 struct ib_udata *udata);
2429 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2430 void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2431 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2432 struct ib_qp_init_attr *qp_init_attr,
2433 struct ib_udata *udata);
2434 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2435 int qp_attr_mask, struct ib_udata *udata);
2436 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2437 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2438 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2439 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2440 struct ib_udata *udata);
2441 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2442 void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2443 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2444 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2445 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2446 u64 virt_addr, int mr_access_flags,
2447 struct ib_udata *udata);
2448 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2449 u64 virt_addr, int mr_access_flags,
2450 struct ib_pd *pd, struct ib_udata *udata);
2451 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2452 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2453 u32 max_num_sg);
2454 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2455 u32 max_num_data_sg,
2456 u32 max_num_meta_sg);
2457 int (*advise_mr)(struct ib_pd *pd,
2458 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2459 struct ib_sge *sg_list, u32 num_sge,
2460 struct uverbs_attr_bundle *attrs);
2461 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2462 unsigned int *sg_offset);
2463 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2464 struct ib_mr_status *mr_status);
2465 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2466 struct ib_udata *udata);
2467 int (*dealloc_mw)(struct ib_mw *mw);
2468 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2469 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2470 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2471 void (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2472 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2473 struct ib_flow_attr *flow_attr,
2474 int domain, struct ib_udata *udata);
2475 int (*destroy_flow)(struct ib_flow *flow_id);
2476 struct ib_flow_action *(*create_flow_action_esp)(
2477 struct ib_device *device,
2478 const struct ib_flow_action_attrs_esp *attr,
2479 struct uverbs_attr_bundle *attrs);
2480 int (*destroy_flow_action)(struct ib_flow_action *action);
2481 int (*modify_flow_action_esp)(
2482 struct ib_flow_action *action,
2483 const struct ib_flow_action_attrs_esp *attr,
2484 struct uverbs_attr_bundle *attrs);
2485 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2486 int state);
2487 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2488 struct ifla_vf_info *ivf);
2489 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2490 struct ifla_vf_stats *stats);
2491 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port,
2492 struct ifla_vf_guid *node_guid,
2493 struct ifla_vf_guid *port_guid);
2494 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2495 int type);
2496 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2497 struct ib_wq_init_attr *init_attr,
2498 struct ib_udata *udata);
2499 void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2500 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2501 u32 wq_attr_mask, struct ib_udata *udata);
2502 struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2503 struct ib_device *device,
2504 struct ib_rwq_ind_table_init_attr *init_attr,
2505 struct ib_udata *udata);
2506 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2507 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2508 struct ib_ucontext *context,
2509 struct ib_dm_alloc_attr *attr,
2510 struct uverbs_attr_bundle *attrs);
2511 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2512 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2513 struct ib_dm_mr_attr *attr,
2514 struct uverbs_attr_bundle *attrs);
2515 int (*create_counters)(struct ib_counters *counters,
2516 struct uverbs_attr_bundle *attrs);
2517 void (*destroy_counters)(struct ib_counters *counters);
2518 int (*read_counters)(struct ib_counters *counters,
2519 struct ib_counters_read_attr *counters_read_attr,
2520 struct uverbs_attr_bundle *attrs);
2521 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2522 int data_sg_nents, unsigned int *data_sg_offset,
2523 struct scatterlist *meta_sg, int meta_sg_nents,
2524 unsigned int *meta_sg_offset);
2525
2526 /**
2527 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2528 * driver initialized data. The struct is kfree()'ed by the sysfs
2529 * core when the device is removed. A lifespan of -1 in the return
2530 * struct tells the core to set a default lifespan.
2531 */
2532 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2533 u8 port_num);
2534 /**
2535 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2536 * @index - The index in the value array we wish to have updated, or
2537 * num_counters if we want all stats updated
2538 * Return codes -
2539 * < 0 - Error, no counters updated
2540 * index - Updated the single counter pointed to by index
2541 * num_counters - Updated all counters (will reset the timestamp
2542 * and prevent further calls for lifespan milliseconds)
2543 * Drivers are allowed to update all counters in leiu of just the
2544 * one given in index at their option
2545 */
2546 int (*get_hw_stats)(struct ib_device *device,
2547 struct rdma_hw_stats *stats, u8 port, int index);
2548 /*
2549 * This function is called once for each port when a ib device is
2550 * registered.
2551 */
2552 int (*init_port)(struct ib_device *device, u8 port_num,
2553 struct kobject *port_sysfs);
2554 /**
2555 * Allows rdma drivers to add their own restrack attributes.
2556 */
2557 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2558 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2559 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2560 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2561 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2562 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2563 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2564
2565 /* Device lifecycle callbacks */
2566 /*
2567 * Called after the device becomes registered, before clients are
2568 * attached
2569 */
2570 int (*enable_driver)(struct ib_device *dev);
2571 /*
2572 * This is called as part of ib_dealloc_device().
2573 */
2574 void (*dealloc_driver)(struct ib_device *dev);
2575
2576 /* iWarp CM callbacks */
2577 void (*iw_add_ref)(struct ib_qp *qp);
2578 void (*iw_rem_ref)(struct ib_qp *qp);
2579 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2580 int (*iw_connect)(struct iw_cm_id *cm_id,
2581 struct iw_cm_conn_param *conn_param);
2582 int (*iw_accept)(struct iw_cm_id *cm_id,
2583 struct iw_cm_conn_param *conn_param);
2584 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2585 u8 pdata_len);
2586 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2587 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2588 /**
2589 * counter_bind_qp - Bind a QP to a counter.
2590 * @counter - The counter to be bound. If counter->id is zero then
2591 * the driver needs to allocate a new counter and set counter->id
2592 */
2593 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2594 /**
2595 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2596 * counter and bind it onto the default one
2597 */
2598 int (*counter_unbind_qp)(struct ib_qp *qp);
2599 /**
2600 * counter_dealloc -De-allocate the hw counter
2601 */
2602 int (*counter_dealloc)(struct rdma_counter *counter);
2603 /**
2604 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2605 * the driver initialized data.
2606 */
2607 struct rdma_hw_stats *(*counter_alloc_stats)(
2608 struct rdma_counter *counter);
2609 /**
2610 * counter_update_stats - Query the stats value of this counter
2611 */
2612 int (*counter_update_stats)(struct rdma_counter *counter);
2613
2614 /**
2615 * Allows rdma drivers to add their own restrack attributes
2616 * dumped via 'rdma stat' iproute2 command.
2617 */
2618 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2619
2620 /* query driver for its ucontext properties */
2621 int (*query_ucontext)(struct ib_ucontext *context,
2622 struct uverbs_attr_bundle *attrs);
2623
2624 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2625 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2626 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2627 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2628 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2629 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2630 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2631};
2632
2633struct ib_core_device {
2634 /* device must be the first element in structure until,
2635 * union of ib_core_device and device exists in ib_device.
2636 */
2637 struct device dev;
2638 possible_net_t rdma_net;
2639 struct kobject *ports_kobj;
2640 struct list_head port_list;
2641 struct ib_device *owner; /* reach back to owner ib_device */
2642};
2643
2644struct rdma_restrack_root;
2645struct ib_device {
2646 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2647 struct device *dma_device;
2648 struct ib_device_ops ops;
2649 char name[IB_DEVICE_NAME_MAX];
2650 struct rcu_head rcu_head;
2651
2652 struct list_head event_handler_list;
2653 /* Protects event_handler_list */
2654 struct rw_semaphore event_handler_rwsem;
2655
2656 /* Protects QP's event_handler calls and open_qp list */
2657 spinlock_t qp_open_list_lock;
2658
2659 struct rw_semaphore client_data_rwsem;
2660 struct xarray client_data;
2661 struct mutex unregistration_lock;
2662
2663 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2664 rwlock_t cache_lock;
2665 /**
2666 * port_data is indexed by port number
2667 */
2668 struct ib_port_data *port_data;
2669
2670 int num_comp_vectors;
2671
2672 union {
2673 struct device dev;
2674 struct ib_core_device coredev;
2675 };
2676
2677 /* First group for device attributes,
2678 * Second group for driver provided attributes (optional).
2679 * It is NULL terminated array.
2680 */
2681 const struct attribute_group *groups[3];
2682
2683 u64 uverbs_cmd_mask;
2684 u64 uverbs_ex_cmd_mask;
2685
2686 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2687 __be64 node_guid;
2688 u32 local_dma_lkey;
2689 u16 is_switch:1;
2690 /* Indicates kernel verbs support, should not be used in drivers */
2691 u16 kverbs_provider:1;
2692 /* CQ adaptive moderation (RDMA DIM) */
2693 u16 use_cq_dim:1;
2694 u8 node_type;
2695 u8 phys_port_cnt;
2696 struct ib_device_attr attrs;
2697 struct attribute_group *hw_stats_ag;
2698 struct rdma_hw_stats *hw_stats;
2699
2700#ifdef CONFIG_CGROUP_RDMA
2701 struct rdmacg_device cg_device;
2702#endif
2703
2704 u32 index;
2705
2706 spinlock_t cq_pools_lock;
2707 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2708
2709 struct rdma_restrack_root *res;
2710
2711 const struct uapi_definition *driver_def;
2712
2713 /*
2714 * Positive refcount indicates that the device is currently
2715 * registered and cannot be unregistered.
2716 */
2717 refcount_t refcount;
2718 struct completion unreg_completion;
2719 struct work_struct unregistration_work;
2720
2721 const struct rdma_link_ops *link_ops;
2722
2723 /* Protects compat_devs xarray modifications */
2724 struct mutex compat_devs_mutex;
2725 /* Maintains compat devices for each net namespace */
2726 struct xarray compat_devs;
2727
2728 /* Used by iWarp CM */
2729 char iw_ifname[IFNAMSIZ];
2730 u32 iw_driver_flags;
2731 u32 lag_flags;
2732};
2733
2734struct ib_client_nl_info;
2735struct ib_client {
2736 const char *name;
2737 int (*add)(struct ib_device *ibdev);
2738 void (*remove)(struct ib_device *, void *client_data);
2739 void (*rename)(struct ib_device *dev, void *client_data);
2740 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2741 struct ib_client_nl_info *res);
2742 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2743
2744 /* Returns the net_dev belonging to this ib_client and matching the
2745 * given parameters.
2746 * @dev: An RDMA device that the net_dev use for communication.
2747 * @port: A physical port number on the RDMA device.
2748 * @pkey: P_Key that the net_dev uses if applicable.
2749 * @gid: A GID that the net_dev uses to communicate.
2750 * @addr: An IP address the net_dev is configured with.
2751 * @client_data: The device's client data set by ib_set_client_data().
2752 *
2753 * An ib_client that implements a net_dev on top of RDMA devices
2754 * (such as IP over IB) should implement this callback, allowing the
2755 * rdma_cm module to find the right net_dev for a given request.
2756 *
2757 * The caller is responsible for calling dev_put on the returned
2758 * netdev. */
2759 struct net_device *(*get_net_dev_by_params)(
2760 struct ib_device *dev,
2761 u8 port,
2762 u16 pkey,
2763 const union ib_gid *gid,
2764 const struct sockaddr *addr,
2765 void *client_data);
2766
2767 refcount_t uses;
2768 struct completion uses_zero;
2769 u32 client_id;
2770
2771 /* kverbs are not required by the client */
2772 u8 no_kverbs_req:1;
2773};
2774
2775/*
2776 * IB block DMA iterator
2777 *
2778 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2779 * to a HW supported page size.
2780 */
2781struct ib_block_iter {
2782 /* internal states */
2783 struct scatterlist *__sg; /* sg holding the current aligned block */
2784 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2785 unsigned int __sg_nents; /* number of SG entries */
2786 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2787 unsigned int __pg_bit; /* alignment of current block */
2788};
2789
2790struct ib_device *_ib_alloc_device(size_t size);
2791#define ib_alloc_device(drv_struct, member) \
2792 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2793 BUILD_BUG_ON_ZERO(offsetof( \
2794 struct drv_struct, member))), \
2795 struct drv_struct, member)
2796
2797void ib_dealloc_device(struct ib_device *device);
2798
2799void ib_get_device_fw_str(struct ib_device *device, char *str);
2800
2801int ib_register_device(struct ib_device *device, const char *name);
2802void ib_unregister_device(struct ib_device *device);
2803void ib_unregister_driver(enum rdma_driver_id driver_id);
2804void ib_unregister_device_and_put(struct ib_device *device);
2805void ib_unregister_device_queued(struct ib_device *ib_dev);
2806
2807int ib_register_client (struct ib_client *client);
2808void ib_unregister_client(struct ib_client *client);
2809
2810void __rdma_block_iter_start(struct ib_block_iter *biter,
2811 struct scatterlist *sglist,
2812 unsigned int nents,
2813 unsigned long pgsz);
2814bool __rdma_block_iter_next(struct ib_block_iter *biter);
2815
2816/**
2817 * rdma_block_iter_dma_address - get the aligned dma address of the current
2818 * block held by the block iterator.
2819 * @biter: block iterator holding the memory block
2820 */
2821static inline dma_addr_t
2822rdma_block_iter_dma_address(struct ib_block_iter *biter)
2823{
2824 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2825}
2826
2827/**
2828 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2829 * @sglist: sglist to iterate over
2830 * @biter: block iterator holding the memory block
2831 * @nents: maximum number of sg entries to iterate over
2832 * @pgsz: best HW supported page size to use
2833 *
2834 * Callers may use rdma_block_iter_dma_address() to get each
2835 * blocks aligned DMA address.
2836 */
2837#define rdma_for_each_block(sglist, biter, nents, pgsz) \
2838 for (__rdma_block_iter_start(biter, sglist, nents, \
2839 pgsz); \
2840 __rdma_block_iter_next(biter);)
2841
2842/**
2843 * ib_get_client_data - Get IB client context
2844 * @device:Device to get context for
2845 * @client:Client to get context for
2846 *
2847 * ib_get_client_data() returns the client context data set with
2848 * ib_set_client_data(). This can only be called while the client is
2849 * registered to the device, once the ib_client remove() callback returns this
2850 * cannot be called.
2851 */
2852static inline void *ib_get_client_data(struct ib_device *device,
2853 struct ib_client *client)
2854{
2855 return xa_load(&device->client_data, client->client_id);
2856}
2857void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2858 void *data);
2859void ib_set_device_ops(struct ib_device *device,
2860 const struct ib_device_ops *ops);
2861
2862int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2863 unsigned long pfn, unsigned long size, pgprot_t prot,
2864 struct rdma_user_mmap_entry *entry);
2865int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2866 struct rdma_user_mmap_entry *entry,
2867 size_t length);
2868int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2869 struct rdma_user_mmap_entry *entry,
2870 size_t length, u32 min_pgoff,
2871 u32 max_pgoff);
2872
2873struct rdma_user_mmap_entry *
2874rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2875 unsigned long pgoff);
2876struct rdma_user_mmap_entry *
2877rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2878 struct vm_area_struct *vma);
2879void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2880
2881void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2882
2883static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2884{
2885 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2886}
2887
2888static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2889{
2890 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2891}
2892
2893static inline bool ib_is_buffer_cleared(const void __user *p,
2894 size_t len)
2895{
2896 bool ret;
2897 u8 *buf;
2898
2899 if (len > USHRT_MAX)
2900 return false;
2901
2902 buf = memdup_user(p, len);
2903 if (IS_ERR(buf))
2904 return false;
2905
2906 ret = !memchr_inv(buf, 0, len);
2907 kfree(buf);
2908 return ret;
2909}
2910
2911static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2912 size_t offset,
2913 size_t len)
2914{
2915 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2916}
2917
2918/**
2919 * ib_is_destroy_retryable - Check whether the uobject destruction
2920 * is retryable.
2921 * @ret: The initial destruction return code
2922 * @why: remove reason
2923 * @uobj: The uobject that is destroyed
2924 *
2925 * This function is a helper function that IB layer and low-level drivers
2926 * can use to consider whether the destruction of the given uobject is
2927 * retry-able.
2928 * It checks the original return code, if it wasn't success the destruction
2929 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2930 * the remove reason. (i.e. why).
2931 * Must be called with the object locked for destroy.
2932 */
2933static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2934 struct ib_uobject *uobj)
2935{
2936 return ret && (why == RDMA_REMOVE_DESTROY ||
2937 uobj->context->cleanup_retryable);
2938}
2939
2940/**
2941 * ib_destroy_usecnt - Called during destruction to check the usecnt
2942 * @usecnt: The usecnt atomic
2943 * @why: remove reason
2944 * @uobj: The uobject that is destroyed
2945 *
2946 * Non-zero usecnts will block destruction unless destruction was triggered by
2947 * a ucontext cleanup.
2948 */
2949static inline int ib_destroy_usecnt(atomic_t *usecnt,
2950 enum rdma_remove_reason why,
2951 struct ib_uobject *uobj)
2952{
2953 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2954 return -EBUSY;
2955 return 0;
2956}
2957
2958/**
2959 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2960 * contains all required attributes and no attributes not allowed for
2961 * the given QP state transition.
2962 * @cur_state: Current QP state
2963 * @next_state: Next QP state
2964 * @type: QP type
2965 * @mask: Mask of supplied QP attributes
2966 *
2967 * This function is a helper function that a low-level driver's
2968 * modify_qp method can use to validate the consumer's input. It
2969 * checks that cur_state and next_state are valid QP states, that a
2970 * transition from cur_state to next_state is allowed by the IB spec,
2971 * and that the attribute mask supplied is allowed for the transition.
2972 */
2973bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2974 enum ib_qp_type type, enum ib_qp_attr_mask mask);
2975
2976void ib_register_event_handler(struct ib_event_handler *event_handler);
2977void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2978void ib_dispatch_event(const struct ib_event *event);
2979
2980int ib_query_port(struct ib_device *device,
2981 u8 port_num, struct ib_port_attr *port_attr);
2982
2983enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2984 u8 port_num);
2985
2986/**
2987 * rdma_cap_ib_switch - Check if the device is IB switch
2988 * @device: Device to check
2989 *
2990 * Device driver is responsible for setting is_switch bit on
2991 * in ib_device structure at init time.
2992 *
2993 * Return: true if the device is IB switch.
2994 */
2995static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2996{
2997 return device->is_switch;
2998}
2999
3000/**
3001 * rdma_start_port - Return the first valid port number for the device
3002 * specified
3003 *
3004 * @device: Device to be checked
3005 *
3006 * Return start port number
3007 */
3008static inline u8 rdma_start_port(const struct ib_device *device)
3009{
3010 return rdma_cap_ib_switch(device) ? 0 : 1;
3011}
3012
3013/**
3014 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3015 * @device - The struct ib_device * to iterate over
3016 * @iter - The unsigned int to store the port number
3017 */
3018#define rdma_for_each_port(device, iter) \
3019 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
3020 unsigned int, iter))); \
3021 iter <= rdma_end_port(device); (iter)++)
3022
3023/**
3024 * rdma_end_port - Return the last valid port number for the device
3025 * specified
3026 *
3027 * @device: Device to be checked
3028 *
3029 * Return last port number
3030 */
3031static inline u8 rdma_end_port(const struct ib_device *device)
3032{
3033 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3034}
3035
3036static inline int rdma_is_port_valid(const struct ib_device *device,
3037 unsigned int port)
3038{
3039 return (port >= rdma_start_port(device) &&
3040 port <= rdma_end_port(device));
3041}
3042
3043static inline bool rdma_is_grh_required(const struct ib_device *device,
3044 u8 port_num)
3045{
3046 return device->port_data[port_num].immutable.core_cap_flags &
3047 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3048}
3049
3050static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
3051{
3052 return device->port_data[port_num].immutable.core_cap_flags &
3053 RDMA_CORE_CAP_PROT_IB;
3054}
3055
3056static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
3057{
3058 return device->port_data[port_num].immutable.core_cap_flags &
3059 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3060}
3061
3062static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
3063{
3064 return device->port_data[port_num].immutable.core_cap_flags &
3065 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3066}
3067
3068static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
3069{
3070 return device->port_data[port_num].immutable.core_cap_flags &
3071 RDMA_CORE_CAP_PROT_ROCE;
3072}
3073
3074static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
3075{
3076 return device->port_data[port_num].immutable.core_cap_flags &
3077 RDMA_CORE_CAP_PROT_IWARP;
3078}
3079
3080static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
3081{
3082 return rdma_protocol_ib(device, port_num) ||
3083 rdma_protocol_roce(device, port_num);
3084}
3085
3086static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3087{
3088 return device->port_data[port_num].immutable.core_cap_flags &
3089 RDMA_CORE_CAP_PROT_RAW_PACKET;
3090}
3091
3092static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3093{
3094 return device->port_data[port_num].immutable.core_cap_flags &
3095 RDMA_CORE_CAP_PROT_USNIC;
3096}
3097
3098/**
3099 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3100 * Management Datagrams.
3101 * @device: Device to check
3102 * @port_num: Port number to check
3103 *
3104 * Management Datagrams (MAD) are a required part of the InfiniBand
3105 * specification and are supported on all InfiniBand devices. A slightly
3106 * extended version are also supported on OPA interfaces.
3107 *
3108 * Return: true if the port supports sending/receiving of MAD packets.
3109 */
3110static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
3111{
3112 return device->port_data[port_num].immutable.core_cap_flags &
3113 RDMA_CORE_CAP_IB_MAD;
3114}
3115
3116/**
3117 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3118 * Management Datagrams.
3119 * @device: Device to check
3120 * @port_num: Port number to check
3121 *
3122 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3123 * datagrams with their own versions. These OPA MADs share many but not all of
3124 * the characteristics of InfiniBand MADs.
3125 *
3126 * OPA MADs differ in the following ways:
3127 *
3128 * 1) MADs are variable size up to 2K
3129 * IBTA defined MADs remain fixed at 256 bytes
3130 * 2) OPA SMPs must carry valid PKeys
3131 * 3) OPA SMP packets are a different format
3132 *
3133 * Return: true if the port supports OPA MAD packet formats.
3134 */
3135static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3136{
3137 return device->port_data[port_num].immutable.core_cap_flags &
3138 RDMA_CORE_CAP_OPA_MAD;
3139}
3140
3141/**
3142 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3143 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3144 * @device: Device to check
3145 * @port_num: Port number to check
3146 *
3147 * Each InfiniBand node is required to provide a Subnet Management Agent
3148 * that the subnet manager can access. Prior to the fabric being fully
3149 * configured by the subnet manager, the SMA is accessed via a well known
3150 * interface called the Subnet Management Interface (SMI). This interface
3151 * uses directed route packets to communicate with the SM to get around the
3152 * chicken and egg problem of the SM needing to know what's on the fabric
3153 * in order to configure the fabric, and needing to configure the fabric in
3154 * order to send packets to the devices on the fabric. These directed
3155 * route packets do not need the fabric fully configured in order to reach
3156 * their destination. The SMI is the only method allowed to send
3157 * directed route packets on an InfiniBand fabric.
3158 *
3159 * Return: true if the port provides an SMI.
3160 */
3161static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
3162{
3163 return device->port_data[port_num].immutable.core_cap_flags &
3164 RDMA_CORE_CAP_IB_SMI;
3165}
3166
3167/**
3168 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3169 * Communication Manager.
3170 * @device: Device to check
3171 * @port_num: Port number to check
3172 *
3173 * The InfiniBand Communication Manager is one of many pre-defined General
3174 * Service Agents (GSA) that are accessed via the General Service
3175 * Interface (GSI). It's role is to facilitate establishment of connections
3176 * between nodes as well as other management related tasks for established
3177 * connections.
3178 *
3179 * Return: true if the port supports an IB CM (this does not guarantee that
3180 * a CM is actually running however).
3181 */
3182static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
3183{
3184 return device->port_data[port_num].immutable.core_cap_flags &
3185 RDMA_CORE_CAP_IB_CM;
3186}
3187
3188/**
3189 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3190 * Communication Manager.
3191 * @device: Device to check
3192 * @port_num: Port number to check
3193 *
3194 * Similar to above, but specific to iWARP connections which have a different
3195 * managment protocol than InfiniBand.
3196 *
3197 * Return: true if the port supports an iWARP CM (this does not guarantee that
3198 * a CM is actually running however).
3199 */
3200static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
3201{
3202 return device->port_data[port_num].immutable.core_cap_flags &
3203 RDMA_CORE_CAP_IW_CM;
3204}
3205
3206/**
3207 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3208 * Subnet Administration.
3209 * @device: Device to check
3210 * @port_num: Port number to check
3211 *
3212 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3213 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3214 * fabrics, devices should resolve routes to other hosts by contacting the
3215 * SA to query the proper route.
3216 *
3217 * Return: true if the port should act as a client to the fabric Subnet
3218 * Administration interface. This does not imply that the SA service is
3219 * running locally.
3220 */
3221static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
3222{
3223 return device->port_data[port_num].immutable.core_cap_flags &
3224 RDMA_CORE_CAP_IB_SA;
3225}
3226
3227/**
3228 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3229 * Multicast.
3230 * @device: Device to check
3231 * @port_num: Port number to check
3232 *
3233 * InfiniBand multicast registration is more complex than normal IPv4 or
3234 * IPv6 multicast registration. Each Host Channel Adapter must register
3235 * with the Subnet Manager when it wishes to join a multicast group. It
3236 * should do so only once regardless of how many queue pairs it subscribes
3237 * to this group. And it should leave the group only after all queue pairs
3238 * attached to the group have been detached.
3239 *
3240 * Return: true if the port must undertake the additional adminstrative
3241 * overhead of registering/unregistering with the SM and tracking of the
3242 * total number of queue pairs attached to the multicast group.
3243 */
3244static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
3245{
3246 return rdma_cap_ib_sa(device, port_num);
3247}
3248
3249/**
3250 * rdma_cap_af_ib - Check if the port of device has the capability
3251 * Native Infiniband Address.
3252 * @device: Device to check
3253 * @port_num: Port number to check
3254 *
3255 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3256 * GID. RoCE uses a different mechanism, but still generates a GID via
3257 * a prescribed mechanism and port specific data.
3258 *
3259 * Return: true if the port uses a GID address to identify devices on the
3260 * network.
3261 */
3262static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
3263{
3264 return device->port_data[port_num].immutable.core_cap_flags &
3265 RDMA_CORE_CAP_AF_IB;
3266}
3267
3268/**
3269 * rdma_cap_eth_ah - Check if the port of device has the capability
3270 * Ethernet Address Handle.
3271 * @device: Device to check
3272 * @port_num: Port number to check
3273 *
3274 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3275 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3276 * port. Normally, packet headers are generated by the sending host
3277 * adapter, but when sending connectionless datagrams, we must manually
3278 * inject the proper headers for the fabric we are communicating over.
3279 *
3280 * Return: true if we are running as a RoCE port and must force the
3281 * addition of a Global Route Header built from our Ethernet Address
3282 * Handle into our header list for connectionless packets.
3283 */
3284static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
3285{
3286 return device->port_data[port_num].immutable.core_cap_flags &
3287 RDMA_CORE_CAP_ETH_AH;
3288}
3289
3290/**
3291 * rdma_cap_opa_ah - Check if the port of device supports
3292 * OPA Address handles
3293 * @device: Device to check
3294 * @port_num: Port number to check
3295 *
3296 * Return: true if we are running on an OPA device which supports
3297 * the extended OPA addressing.
3298 */
3299static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3300{
3301 return (device->port_data[port_num].immutable.core_cap_flags &
3302 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3303}
3304
3305/**
3306 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3307 *
3308 * @device: Device
3309 * @port_num: Port number
3310 *
3311 * This MAD size includes the MAD headers and MAD payload. No other headers
3312 * are included.
3313 *
3314 * Return the max MAD size required by the Port. Will return 0 if the port
3315 * does not support MADs
3316 */
3317static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3318{
3319 return device->port_data[port_num].immutable.max_mad_size;
3320}
3321
3322/**
3323 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3324 * @device: Device to check
3325 * @port_num: Port number to check
3326 *
3327 * RoCE GID table mechanism manages the various GIDs for a device.
3328 *
3329 * NOTE: if allocating the port's GID table has failed, this call will still
3330 * return true, but any RoCE GID table API will fail.
3331 *
3332 * Return: true if the port uses RoCE GID table mechanism in order to manage
3333 * its GIDs.
3334 */
3335static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3336 u8 port_num)
3337{
3338 return rdma_protocol_roce(device, port_num) &&
3339 device->ops.add_gid && device->ops.del_gid;
3340}
3341
3342/*
3343 * Check if the device supports READ W/ INVALIDATE.
3344 */
3345static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3346{
3347 /*
3348 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3349 * has support for it yet.
3350 */
3351 return rdma_protocol_iwarp(dev, port_num);
3352}
3353
3354/**
3355 * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3356 *
3357 * @addr: address
3358 * @pgsz_bitmap: bitmap of HW supported page sizes
3359 */
3360static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3361 unsigned long pgsz_bitmap)
3362{
3363 unsigned long align;
3364 unsigned long pgsz;
3365
3366 align = addr & -addr;
3367
3368 /* Find page bit such that addr is aligned to the highest supported
3369 * HW page size
3370 */
3371 pgsz = pgsz_bitmap & ~(-align << 1);
3372 if (!pgsz)
3373 return __ffs(pgsz_bitmap);
3374
3375 return __fls(pgsz);
3376}
3377
3378/**
3379 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3380 * @device: Device
3381 * @port_num: 1 based Port number
3382 *
3383 * Return true if port is an Intel OPA port , false if not
3384 */
3385static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3386 u32 port_num)
3387{
3388 return (device->port_data[port_num].immutable.core_cap_flags &
3389 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3390}
3391
3392/**
3393 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3394 * @device: Device
3395 * @port_num: Port number
3396 * @mtu: enum value of MTU
3397 *
3398 * Return the MTU size supported by the port as an integer value. Will return
3399 * -1 if enum value of mtu is not supported.
3400 */
3401static inline int rdma_mtu_enum_to_int(struct ib_device *device, u8 port,
3402 int mtu)
3403{
3404 if (rdma_core_cap_opa_port(device, port))
3405 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3406 else
3407 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3408}
3409
3410/**
3411 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3412 * @device: Device
3413 * @port_num: Port number
3414 * @attr: port attribute
3415 *
3416 * Return the MTU size supported by the port as an integer value.
3417 */
3418static inline int rdma_mtu_from_attr(struct ib_device *device, u8 port,
3419 struct ib_port_attr *attr)
3420{
3421 if (rdma_core_cap_opa_port(device, port))
3422 return attr->phys_mtu;
3423 else
3424 return ib_mtu_enum_to_int(attr->max_mtu);
3425}
3426
3427int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3428 int state);
3429int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3430 struct ifla_vf_info *info);
3431int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3432 struct ifla_vf_stats *stats);
3433int ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
3434 struct ifla_vf_guid *node_guid,
3435 struct ifla_vf_guid *port_guid);
3436int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3437 int type);
3438
3439int ib_query_pkey(struct ib_device *device,
3440 u8 port_num, u16 index, u16 *pkey);
3441
3442int ib_modify_device(struct ib_device *device,
3443 int device_modify_mask,
3444 struct ib_device_modify *device_modify);
3445
3446int ib_modify_port(struct ib_device *device,
3447 u8 port_num, int port_modify_mask,
3448 struct ib_port_modify *port_modify);
3449
3450int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3451 u8 *port_num, u16 *index);
3452
3453int ib_find_pkey(struct ib_device *device,
3454 u8 port_num, u16 pkey, u16 *index);
3455
3456enum ib_pd_flags {
3457 /*
3458 * Create a memory registration for all memory in the system and place
3459 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3460 * ULPs to avoid the overhead of dynamic MRs.
3461 *
3462 * This flag is generally considered unsafe and must only be used in
3463 * extremly trusted environments. Every use of it will log a warning
3464 * in the kernel log.
3465 */
3466 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3467};
3468
3469struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3470 const char *caller);
3471
3472#define ib_alloc_pd(device, flags) \
3473 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3474
3475/**
3476 * ib_dealloc_pd_user - Deallocate kernel/user PD
3477 * @pd: The protection domain
3478 * @udata: Valid user data or NULL for kernel objects
3479 */
3480void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3481
3482/**
3483 * ib_dealloc_pd - Deallocate kernel PD
3484 * @pd: The protection domain
3485 *
3486 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3487 */
3488static inline void ib_dealloc_pd(struct ib_pd *pd)
3489{
3490 ib_dealloc_pd_user(pd, NULL);
3491}
3492
3493enum rdma_create_ah_flags {
3494 /* In a sleepable context */
3495 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3496};
3497
3498/**
3499 * rdma_create_ah - Creates an address handle for the given address vector.
3500 * @pd: The protection domain associated with the address handle.
3501 * @ah_attr: The attributes of the address vector.
3502 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3503 *
3504 * The address handle is used to reference a local or global destination
3505 * in all UD QP post sends.
3506 */
3507struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3508 u32 flags);
3509
3510/**
3511 * rdma_create_user_ah - Creates an address handle for the given address vector.
3512 * It resolves destination mac address for ah attribute of RoCE type.
3513 * @pd: The protection domain associated with the address handle.
3514 * @ah_attr: The attributes of the address vector.
3515 * @udata: pointer to user's input output buffer information need by
3516 * provider driver.
3517 *
3518 * It returns 0 on success and returns appropriate error code on error.
3519 * The address handle is used to reference a local or global destination
3520 * in all UD QP post sends.
3521 */
3522struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3523 struct rdma_ah_attr *ah_attr,
3524 struct ib_udata *udata);
3525/**
3526 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3527 * work completion.
3528 * @hdr: the L3 header to parse
3529 * @net_type: type of header to parse
3530 * @sgid: place to store source gid
3531 * @dgid: place to store destination gid
3532 */
3533int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3534 enum rdma_network_type net_type,
3535 union ib_gid *sgid, union ib_gid *dgid);
3536
3537/**
3538 * ib_get_rdma_header_version - Get the header version
3539 * @hdr: the L3 header to parse
3540 */
3541int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3542
3543/**
3544 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3545 * work completion.
3546 * @device: Device on which the received message arrived.
3547 * @port_num: Port on which the received message arrived.
3548 * @wc: Work completion associated with the received message.
3549 * @grh: References the received global route header. This parameter is
3550 * ignored unless the work completion indicates that the GRH is valid.
3551 * @ah_attr: Returned attributes that can be used when creating an address
3552 * handle for replying to the message.
3553 * When ib_init_ah_attr_from_wc() returns success,
3554 * (a) for IB link layer it optionally contains a reference to SGID attribute
3555 * when GRH is present for IB link layer.
3556 * (b) for RoCE link layer it contains a reference to SGID attribute.
3557 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3558 * attributes which are initialized using ib_init_ah_attr_from_wc().
3559 *
3560 */
3561int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3562 const struct ib_wc *wc, const struct ib_grh *grh,
3563 struct rdma_ah_attr *ah_attr);
3564
3565/**
3566 * ib_create_ah_from_wc - Creates an address handle associated with the
3567 * sender of the specified work completion.
3568 * @pd: The protection domain associated with the address handle.
3569 * @wc: Work completion information associated with a received message.
3570 * @grh: References the received global route header. This parameter is
3571 * ignored unless the work completion indicates that the GRH is valid.
3572 * @port_num: The outbound port number to associate with the address.
3573 *
3574 * The address handle is used to reference a local or global destination
3575 * in all UD QP post sends.
3576 */
3577struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3578 const struct ib_grh *grh, u8 port_num);
3579
3580/**
3581 * rdma_modify_ah - Modifies the address vector associated with an address
3582 * handle.
3583 * @ah: The address handle to modify.
3584 * @ah_attr: The new address vector attributes to associate with the
3585 * address handle.
3586 */
3587int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3588
3589/**
3590 * rdma_query_ah - Queries the address vector associated with an address
3591 * handle.
3592 * @ah: The address handle to query.
3593 * @ah_attr: The address vector attributes associated with the address
3594 * handle.
3595 */
3596int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3597
3598enum rdma_destroy_ah_flags {
3599 /* In a sleepable context */
3600 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3601};
3602
3603/**
3604 * rdma_destroy_ah_user - Destroys an address handle.
3605 * @ah: The address handle to destroy.
3606 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3607 * @udata: Valid user data or NULL for kernel objects
3608 */
3609int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3610
3611/**
3612 * rdma_destroy_ah - Destroys an kernel address handle.
3613 * @ah: The address handle to destroy.
3614 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3615 *
3616 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3617 */
3618static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3619{
3620 return rdma_destroy_ah_user(ah, flags, NULL);
3621}
3622
3623struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3624 struct ib_srq_init_attr *srq_init_attr,
3625 struct ib_usrq_object *uobject,
3626 struct ib_udata *udata);
3627static inline struct ib_srq *
3628ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3629{
3630 if (!pd->device->ops.create_srq)
3631 return ERR_PTR(-EOPNOTSUPP);
3632
3633 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3634}
3635
3636/**
3637 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3638 * @srq: The SRQ to modify.
3639 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3640 * the current values of selected SRQ attributes are returned.
3641 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3642 * are being modified.
3643 *
3644 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3645 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3646 * the number of receives queued drops below the limit.
3647 */
3648int ib_modify_srq(struct ib_srq *srq,
3649 struct ib_srq_attr *srq_attr,
3650 enum ib_srq_attr_mask srq_attr_mask);
3651
3652/**
3653 * ib_query_srq - Returns the attribute list and current values for the
3654 * specified SRQ.
3655 * @srq: The SRQ to query.
3656 * @srq_attr: The attributes of the specified SRQ.
3657 */
3658int ib_query_srq(struct ib_srq *srq,
3659 struct ib_srq_attr *srq_attr);
3660
3661/**
3662 * ib_destroy_srq_user - Destroys the specified SRQ.
3663 * @srq: The SRQ to destroy.
3664 * @udata: Valid user data or NULL for kernel objects
3665 */
3666int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3667
3668/**
3669 * ib_destroy_srq - Destroys the specified kernel SRQ.
3670 * @srq: The SRQ to destroy.
3671 *
3672 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3673 */
3674static inline int ib_destroy_srq(struct ib_srq *srq)
3675{
3676 return ib_destroy_srq_user(srq, NULL);
3677}
3678
3679/**
3680 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3681 * @srq: The SRQ to post the work request on.
3682 * @recv_wr: A list of work requests to post on the receive queue.
3683 * @bad_recv_wr: On an immediate failure, this parameter will reference
3684 * the work request that failed to be posted on the QP.
3685 */
3686static inline int ib_post_srq_recv(struct ib_srq *srq,
3687 const struct ib_recv_wr *recv_wr,
3688 const struct ib_recv_wr **bad_recv_wr)
3689{
3690 const struct ib_recv_wr *dummy;
3691
3692 return srq->device->ops.post_srq_recv(srq, recv_wr,
3693 bad_recv_wr ? : &dummy);
3694}
3695
3696struct ib_qp *ib_create_qp(struct ib_pd *pd,
3697 struct ib_qp_init_attr *qp_init_attr);
3698
3699/**
3700 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3701 * @qp: The QP to modify.
3702 * @attr: On input, specifies the QP attributes to modify. On output,
3703 * the current values of selected QP attributes are returned.
3704 * @attr_mask: A bit-mask used to specify which attributes of the QP
3705 * are being modified.
3706 * @udata: pointer to user's input output buffer information
3707 * are being modified.
3708 * It returns 0 on success and returns appropriate error code on error.
3709 */
3710int ib_modify_qp_with_udata(struct ib_qp *qp,
3711 struct ib_qp_attr *attr,
3712 int attr_mask,
3713 struct ib_udata *udata);
3714
3715/**
3716 * ib_modify_qp - Modifies the attributes for the specified QP and then
3717 * transitions the QP to the given state.
3718 * @qp: The QP to modify.
3719 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3720 * the current values of selected QP attributes are returned.
3721 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3722 * are being modified.
3723 */
3724int ib_modify_qp(struct ib_qp *qp,
3725 struct ib_qp_attr *qp_attr,
3726 int qp_attr_mask);
3727
3728/**
3729 * ib_query_qp - Returns the attribute list and current values for the
3730 * specified QP.
3731 * @qp: The QP to query.
3732 * @qp_attr: The attributes of the specified QP.
3733 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3734 * @qp_init_attr: Additional attributes of the selected QP.
3735 *
3736 * The qp_attr_mask may be used to limit the query to gathering only the
3737 * selected attributes.
3738 */
3739int ib_query_qp(struct ib_qp *qp,
3740 struct ib_qp_attr *qp_attr,
3741 int qp_attr_mask,
3742 struct ib_qp_init_attr *qp_init_attr);
3743
3744/**
3745 * ib_destroy_qp - Destroys the specified QP.
3746 * @qp: The QP to destroy.
3747 * @udata: Valid udata or NULL for kernel objects
3748 */
3749int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3750
3751/**
3752 * ib_destroy_qp - Destroys the specified kernel QP.
3753 * @qp: The QP to destroy.
3754 *
3755 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3756 */
3757static inline int ib_destroy_qp(struct ib_qp *qp)
3758{
3759 return ib_destroy_qp_user(qp, NULL);
3760}
3761
3762/**
3763 * ib_open_qp - Obtain a reference to an existing sharable QP.
3764 * @xrcd - XRC domain
3765 * @qp_open_attr: Attributes identifying the QP to open.
3766 *
3767 * Returns a reference to a sharable QP.
3768 */
3769struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3770 struct ib_qp_open_attr *qp_open_attr);
3771
3772/**
3773 * ib_close_qp - Release an external reference to a QP.
3774 * @qp: The QP handle to release
3775 *
3776 * The opened QP handle is released by the caller. The underlying
3777 * shared QP is not destroyed until all internal references are released.
3778 */
3779int ib_close_qp(struct ib_qp *qp);
3780
3781/**
3782 * ib_post_send - Posts a list of work requests to the send queue of
3783 * the specified QP.
3784 * @qp: The QP to post the work request on.
3785 * @send_wr: A list of work requests to post on the send queue.
3786 * @bad_send_wr: On an immediate failure, this parameter will reference
3787 * the work request that failed to be posted on the QP.
3788 *
3789 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3790 * error is returned, the QP state shall not be affected,
3791 * ib_post_send() will return an immediate error after queueing any
3792 * earlier work requests in the list.
3793 */
3794static inline int ib_post_send(struct ib_qp *qp,
3795 const struct ib_send_wr *send_wr,
3796 const struct ib_send_wr **bad_send_wr)
3797{
3798 const struct ib_send_wr *dummy;
3799
3800 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3801}
3802
3803/**
3804 * ib_post_recv - Posts a list of work requests to the receive queue of
3805 * the specified QP.
3806 * @qp: The QP to post the work request on.
3807 * @recv_wr: A list of work requests to post on the receive queue.
3808 * @bad_recv_wr: On an immediate failure, this parameter will reference
3809 * the work request that failed to be posted on the QP.
3810 */
3811static inline int ib_post_recv(struct ib_qp *qp,
3812 const struct ib_recv_wr *recv_wr,
3813 const struct ib_recv_wr **bad_recv_wr)
3814{
3815 const struct ib_recv_wr *dummy;
3816
3817 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3818}
3819
3820struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3821 int nr_cqe, int comp_vector,
3822 enum ib_poll_context poll_ctx,
3823 const char *caller, struct ib_udata *udata);
3824
3825/**
3826 * ib_alloc_cq_user: Allocate kernel/user CQ
3827 * @dev: The IB device
3828 * @private: Private data attached to the CQE
3829 * @nr_cqe: Number of CQEs in the CQ
3830 * @comp_vector: Completion vector used for the IRQs
3831 * @poll_ctx: Context used for polling the CQ
3832 * @udata: Valid user data or NULL for kernel objects
3833 */
3834static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3835 void *private, int nr_cqe,
3836 int comp_vector,
3837 enum ib_poll_context poll_ctx,
3838 struct ib_udata *udata)
3839{
3840 return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3841 KBUILD_MODNAME, udata);
3842}
3843
3844/**
3845 * ib_alloc_cq: Allocate kernel CQ
3846 * @dev: The IB device
3847 * @private: Private data attached to the CQE
3848 * @nr_cqe: Number of CQEs in the CQ
3849 * @comp_vector: Completion vector used for the IRQs
3850 * @poll_ctx: Context used for polling the CQ
3851 *
3852 * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3853 */
3854static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3855 int nr_cqe, int comp_vector,
3856 enum ib_poll_context poll_ctx)
3857{
3858 return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3859 NULL);
3860}
3861
3862struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3863 int nr_cqe, enum ib_poll_context poll_ctx,
3864 const char *caller);
3865
3866/**
3867 * ib_alloc_cq_any: Allocate kernel CQ
3868 * @dev: The IB device
3869 * @private: Private data attached to the CQE
3870 * @nr_cqe: Number of CQEs in the CQ
3871 * @poll_ctx: Context used for polling the CQ
3872 */
3873static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3874 void *private, int nr_cqe,
3875 enum ib_poll_context poll_ctx)
3876{
3877 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3878 KBUILD_MODNAME);
3879}
3880
3881/**
3882 * ib_free_cq_user - Free kernel/user CQ
3883 * @cq: The CQ to free
3884 * @udata: Valid user data or NULL for kernel objects
3885 *
3886 * NOTE: This function shouldn't be called on shared CQs.
3887 */
3888void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3889
3890/**
3891 * ib_free_cq - Free kernel CQ
3892 * @cq: The CQ to free
3893 *
3894 * NOTE: for user cq use ib_free_cq_user with valid udata!
3895 */
3896static inline void ib_free_cq(struct ib_cq *cq)
3897{
3898 ib_free_cq_user(cq, NULL);
3899}
3900
3901int ib_process_cq_direct(struct ib_cq *cq, int budget);
3902
3903/**
3904 * ib_create_cq - Creates a CQ on the specified device.
3905 * @device: The device on which to create the CQ.
3906 * @comp_handler: A user-specified callback that is invoked when a
3907 * completion event occurs on the CQ.
3908 * @event_handler: A user-specified callback that is invoked when an
3909 * asynchronous event not associated with a completion occurs on the CQ.
3910 * @cq_context: Context associated with the CQ returned to the user via
3911 * the associated completion and event handlers.
3912 * @cq_attr: The attributes the CQ should be created upon.
3913 *
3914 * Users can examine the cq structure to determine the actual CQ size.
3915 */
3916struct ib_cq *__ib_create_cq(struct ib_device *device,
3917 ib_comp_handler comp_handler,
3918 void (*event_handler)(struct ib_event *, void *),
3919 void *cq_context,
3920 const struct ib_cq_init_attr *cq_attr,
3921 const char *caller);
3922#define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3923 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3924
3925/**
3926 * ib_resize_cq - Modifies the capacity of the CQ.
3927 * @cq: The CQ to resize.
3928 * @cqe: The minimum size of the CQ.
3929 *
3930 * Users can examine the cq structure to determine the actual CQ size.
3931 */
3932int ib_resize_cq(struct ib_cq *cq, int cqe);
3933
3934/**
3935 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3936 * @cq: The CQ to modify.
3937 * @cq_count: number of CQEs that will trigger an event
3938 * @cq_period: max period of time in usec before triggering an event
3939 *
3940 */
3941int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3942
3943/**
3944 * ib_destroy_cq_user - Destroys the specified CQ.
3945 * @cq: The CQ to destroy.
3946 * @udata: Valid user data or NULL for kernel objects
3947 */
3948int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3949
3950/**
3951 * ib_destroy_cq - Destroys the specified kernel CQ.
3952 * @cq: The CQ to destroy.
3953 *
3954 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3955 */
3956static inline void ib_destroy_cq(struct ib_cq *cq)
3957{
3958 ib_destroy_cq_user(cq, NULL);
3959}
3960
3961/**
3962 * ib_poll_cq - poll a CQ for completion(s)
3963 * @cq:the CQ being polled
3964 * @num_entries:maximum number of completions to return
3965 * @wc:array of at least @num_entries &struct ib_wc where completions
3966 * will be returned
3967 *
3968 * Poll a CQ for (possibly multiple) completions. If the return value
3969 * is < 0, an error occurred. If the return value is >= 0, it is the
3970 * number of completions returned. If the return value is
3971 * non-negative and < num_entries, then the CQ was emptied.
3972 */
3973static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3974 struct ib_wc *wc)
3975{
3976 return cq->device->ops.poll_cq(cq, num_entries, wc);
3977}
3978
3979/**
3980 * ib_req_notify_cq - Request completion notification on a CQ.
3981 * @cq: The CQ to generate an event for.
3982 * @flags:
3983 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3984 * to request an event on the next solicited event or next work
3985 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3986 * may also be |ed in to request a hint about missed events, as
3987 * described below.
3988 *
3989 * Return Value:
3990 * < 0 means an error occurred while requesting notification
3991 * == 0 means notification was requested successfully, and if
3992 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3993 * were missed and it is safe to wait for another event. In
3994 * this case is it guaranteed that any work completions added
3995 * to the CQ since the last CQ poll will trigger a completion
3996 * notification event.
3997 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3998 * in. It means that the consumer must poll the CQ again to
3999 * make sure it is empty to avoid missing an event because of a
4000 * race between requesting notification and an entry being
4001 * added to the CQ. This return value means it is possible
4002 * (but not guaranteed) that a work completion has been added
4003 * to the CQ since the last poll without triggering a
4004 * completion notification event.
4005 */
4006static inline int ib_req_notify_cq(struct ib_cq *cq,
4007 enum ib_cq_notify_flags flags)
4008{
4009 return cq->device->ops.req_notify_cq(cq, flags);
4010}
4011
4012struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4013 int comp_vector_hint,
4014 enum ib_poll_context poll_ctx);
4015
4016void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4017
4018/**
4019 * ib_req_ncomp_notif - Request completion notification when there are
4020 * at least the specified number of unreaped completions on the CQ.
4021 * @cq: The CQ to generate an event for.
4022 * @wc_cnt: The number of unreaped completions that should be on the
4023 * CQ before an event is generated.
4024 */
4025static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
4026{
4027 return cq->device->ops.req_ncomp_notif ?
4028 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
4029 -ENOSYS;
4030}
4031
4032/**
4033 * ib_dma_mapping_error - check a DMA addr for error
4034 * @dev: The device for which the dma_addr was created
4035 * @dma_addr: The DMA address to check
4036 */
4037static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4038{
4039 return dma_mapping_error(dev->dma_device, dma_addr);
4040}
4041
4042/**
4043 * ib_dma_map_single - Map a kernel virtual address to DMA address
4044 * @dev: The device for which the dma_addr is to be created
4045 * @cpu_addr: The kernel virtual address
4046 * @size: The size of the region in bytes
4047 * @direction: The direction of the DMA
4048 */
4049static inline u64 ib_dma_map_single(struct ib_device *dev,
4050 void *cpu_addr, size_t size,
4051 enum dma_data_direction direction)
4052{
4053 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4054}
4055
4056/**
4057 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4058 * @dev: The device for which the DMA address was created
4059 * @addr: The DMA address
4060 * @size: The size of the region in bytes
4061 * @direction: The direction of the DMA
4062 */
4063static inline void ib_dma_unmap_single(struct ib_device *dev,
4064 u64 addr, size_t size,
4065 enum dma_data_direction direction)
4066{
4067 dma_unmap_single(dev->dma_device, addr, size, direction);
4068}
4069
4070/**
4071 * ib_dma_map_page - Map a physical page to DMA address
4072 * @dev: The device for which the dma_addr is to be created
4073 * @page: The page to be mapped
4074 * @offset: The offset within the page
4075 * @size: The size of the region in bytes
4076 * @direction: The direction of the DMA
4077 */
4078static inline u64 ib_dma_map_page(struct ib_device *dev,
4079 struct page *page,
4080 unsigned long offset,
4081 size_t size,
4082 enum dma_data_direction direction)
4083{
4084 return dma_map_page(dev->dma_device, page, offset, size, direction);
4085}
4086
4087/**
4088 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4089 * @dev: The device for which the DMA address was created
4090 * @addr: The DMA address
4091 * @size: The size of the region in bytes
4092 * @direction: The direction of the DMA
4093 */
4094static inline void ib_dma_unmap_page(struct ib_device *dev,
4095 u64 addr, size_t size,
4096 enum dma_data_direction direction)
4097{
4098 dma_unmap_page(dev->dma_device, addr, size, direction);
4099}
4100
4101/**
4102 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4103 * @dev: The device for which the DMA addresses are to be created
4104 * @sg: The array of scatter/gather entries
4105 * @nents: The number of scatter/gather entries
4106 * @direction: The direction of the DMA
4107 */
4108static inline int ib_dma_map_sg(struct ib_device *dev,
4109 struct scatterlist *sg, int nents,
4110 enum dma_data_direction direction)
4111{
4112 return dma_map_sg(dev->dma_device, sg, nents, direction);
4113}
4114
4115/**
4116 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4117 * @dev: The device for which the DMA addresses were created
4118 * @sg: The array of scatter/gather entries
4119 * @nents: The number of scatter/gather entries
4120 * @direction: The direction of the DMA
4121 */
4122static inline void ib_dma_unmap_sg(struct ib_device *dev,
4123 struct scatterlist *sg, int nents,
4124 enum dma_data_direction direction)
4125{
4126 dma_unmap_sg(dev->dma_device, sg, nents, direction);
4127}
4128
4129static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4130 struct scatterlist *sg, int nents,
4131 enum dma_data_direction direction,
4132 unsigned long dma_attrs)
4133{
4134 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4135 dma_attrs);
4136}
4137
4138static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4139 struct scatterlist *sg, int nents,
4140 enum dma_data_direction direction,
4141 unsigned long dma_attrs)
4142{
4143 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
4144}
4145
4146/**
4147 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4148 * @dev: The device to query
4149 *
4150 * The returned value represents a size in bytes.
4151 */
4152static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4153{
4154 return dma_get_max_seg_size(dev->dma_device);
4155}
4156
4157/**
4158 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4159 * @dev: The device for which the DMA address was created
4160 * @addr: The DMA address
4161 * @size: The size of the region in bytes
4162 * @dir: The direction of the DMA
4163 */
4164static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4165 u64 addr,
4166 size_t size,
4167 enum dma_data_direction dir)
4168{
4169 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4170}
4171
4172/**
4173 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4174 * @dev: The device for which the DMA address was created
4175 * @addr: The DMA address
4176 * @size: The size of the region in bytes
4177 * @dir: The direction of the DMA
4178 */
4179static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4180 u64 addr,
4181 size_t size,
4182 enum dma_data_direction dir)
4183{
4184 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4185}
4186
4187/**
4188 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4189 * @dev: The device for which the DMA address is requested
4190 * @size: The size of the region to allocate in bytes
4191 * @dma_handle: A pointer for returning the DMA address of the region
4192 * @flag: memory allocator flags
4193 */
4194static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4195 size_t size,
4196 dma_addr_t *dma_handle,
4197 gfp_t flag)
4198{
4199 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
4200}
4201
4202/**
4203 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4204 * @dev: The device for which the DMA addresses were allocated
4205 * @size: The size of the region
4206 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4207 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4208 */
4209static inline void ib_dma_free_coherent(struct ib_device *dev,
4210 size_t size, void *cpu_addr,
4211 dma_addr_t dma_handle)
4212{
4213 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
4214}
4215
4216/* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4217 * space. This function should be called when 'current' is the owning MM.
4218 */
4219struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4220 u64 virt_addr, int mr_access_flags);
4221
4222/* ib_advise_mr - give an advice about an address range in a memory region */
4223int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4224 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4225/**
4226 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4227 * HCA translation table.
4228 * @mr: The memory region to deregister.
4229 * @udata: Valid user data or NULL for kernel object
4230 *
4231 * This function can fail, if the memory region has memory windows bound to it.
4232 */
4233int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4234
4235/**
4236 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4237 * HCA translation table.
4238 * @mr: The memory region to deregister.
4239 *
4240 * This function can fail, if the memory region has memory windows bound to it.
4241 *
4242 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4243 */
4244static inline int ib_dereg_mr(struct ib_mr *mr)
4245{
4246 return ib_dereg_mr_user(mr, NULL);
4247}
4248
4249struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4250 u32 max_num_sg);
4251
4252struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4253 u32 max_num_data_sg,
4254 u32 max_num_meta_sg);
4255
4256/**
4257 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4258 * R_Key and L_Key.
4259 * @mr - struct ib_mr pointer to be updated.
4260 * @newkey - new key to be used.
4261 */
4262static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4263{
4264 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4265 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4266}
4267
4268/**
4269 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4270 * for calculating a new rkey for type 2 memory windows.
4271 * @rkey - the rkey to increment.
4272 */
4273static inline u32 ib_inc_rkey(u32 rkey)
4274{
4275 const u32 mask = 0x000000ff;
4276 return ((rkey + 1) & mask) | (rkey & ~mask);
4277}
4278
4279/**
4280 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4281 * @qp: QP to attach to the multicast group. The QP must be type
4282 * IB_QPT_UD.
4283 * @gid: Multicast group GID.
4284 * @lid: Multicast group LID in host byte order.
4285 *
4286 * In order to send and receive multicast packets, subnet
4287 * administration must have created the multicast group and configured
4288 * the fabric appropriately. The port associated with the specified
4289 * QP must also be a member of the multicast group.
4290 */
4291int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4292
4293/**
4294 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4295 * @qp: QP to detach from the multicast group.
4296 * @gid: Multicast group GID.
4297 * @lid: Multicast group LID in host byte order.
4298 */
4299int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4300
4301struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4302 struct inode *inode, struct ib_udata *udata);
4303int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4304
4305static inline int ib_check_mr_access(int flags)
4306{
4307 /*
4308 * Local write permission is required if remote write or
4309 * remote atomic permission is also requested.
4310 */
4311 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4312 !(flags & IB_ACCESS_LOCAL_WRITE))
4313 return -EINVAL;
4314
4315 if (flags & ~IB_ACCESS_SUPPORTED)
4316 return -EINVAL;
4317
4318 return 0;
4319}
4320
4321static inline bool ib_access_writable(int access_flags)
4322{
4323 /*
4324 * We have writable memory backing the MR if any of the following
4325 * access flags are set. "Local write" and "remote write" obviously
4326 * require write access. "Remote atomic" can do things like fetch and
4327 * add, which will modify memory, and "MW bind" can change permissions
4328 * by binding a window.
4329 */
4330 return access_flags &
4331 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4332 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4333}
4334
4335/**
4336 * ib_check_mr_status: lightweight check of MR status.
4337 * This routine may provide status checks on a selected
4338 * ib_mr. first use is for signature status check.
4339 *
4340 * @mr: A memory region.
4341 * @check_mask: Bitmask of which checks to perform from
4342 * ib_mr_status_check enumeration.
4343 * @mr_status: The container of relevant status checks.
4344 * failed checks will be indicated in the status bitmask
4345 * and the relevant info shall be in the error item.
4346 */
4347int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4348 struct ib_mr_status *mr_status);
4349
4350/**
4351 * ib_device_try_get: Hold a registration lock
4352 * device: The device to lock
4353 *
4354 * A device under an active registration lock cannot become unregistered. It
4355 * is only possible to obtain a registration lock on a device that is fully
4356 * registered, otherwise this function returns false.
4357 *
4358 * The registration lock is only necessary for actions which require the
4359 * device to still be registered. Uses that only require the device pointer to
4360 * be valid should use get_device(&ibdev->dev) to hold the memory.
4361 *
4362 */
4363static inline bool ib_device_try_get(struct ib_device *dev)
4364{
4365 return refcount_inc_not_zero(&dev->refcount);
4366}
4367
4368void ib_device_put(struct ib_device *device);
4369struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4370 enum rdma_driver_id driver_id);
4371struct ib_device *ib_device_get_by_name(const char *name,
4372 enum rdma_driver_id driver_id);
4373struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4374 u16 pkey, const union ib_gid *gid,
4375 const struct sockaddr *addr);
4376int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4377 unsigned int port);
4378struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4379
4380struct ib_wq *ib_create_wq(struct ib_pd *pd,
4381 struct ib_wq_init_attr *init_attr);
4382int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
4383int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4384 u32 wq_attr_mask);
4385int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
4386
4387int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4388 unsigned int *sg_offset, unsigned int page_size);
4389int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4390 int data_sg_nents, unsigned int *data_sg_offset,
4391 struct scatterlist *meta_sg, int meta_sg_nents,
4392 unsigned int *meta_sg_offset, unsigned int page_size);
4393
4394static inline int
4395ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4396 unsigned int *sg_offset, unsigned int page_size)
4397{
4398 int n;
4399
4400 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4401 mr->iova = 0;
4402
4403 return n;
4404}
4405
4406int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4407 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4408
4409void ib_drain_rq(struct ib_qp *qp);
4410void ib_drain_sq(struct ib_qp *qp);
4411void ib_drain_qp(struct ib_qp *qp);
4412
4413int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
4414
4415static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4416{
4417 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4418 return attr->roce.dmac;
4419 return NULL;
4420}
4421
4422static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4423{
4424 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4425 attr->ib.dlid = (u16)dlid;
4426 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4427 attr->opa.dlid = dlid;
4428}
4429
4430static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4431{
4432 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4433 return attr->ib.dlid;
4434 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4435 return attr->opa.dlid;
4436 return 0;
4437}
4438
4439static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4440{
4441 attr->sl = sl;
4442}
4443
4444static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4445{
4446 return attr->sl;
4447}
4448
4449static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4450 u8 src_path_bits)
4451{
4452 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4453 attr->ib.src_path_bits = src_path_bits;
4454 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4455 attr->opa.src_path_bits = src_path_bits;
4456}
4457
4458static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4459{
4460 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4461 return attr->ib.src_path_bits;
4462 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4463 return attr->opa.src_path_bits;
4464 return 0;
4465}
4466
4467static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4468 bool make_grd)
4469{
4470 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4471 attr->opa.make_grd = make_grd;
4472}
4473
4474static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4475{
4476 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4477 return attr->opa.make_grd;
4478 return false;
4479}
4480
4481static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4482{
4483 attr->port_num = port_num;
4484}
4485
4486static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4487{
4488 return attr->port_num;
4489}
4490
4491static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4492 u8 static_rate)
4493{
4494 attr->static_rate = static_rate;
4495}
4496
4497static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4498{
4499 return attr->static_rate;
4500}
4501
4502static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4503 enum ib_ah_flags flag)
4504{
4505 attr->ah_flags = flag;
4506}
4507
4508static inline enum ib_ah_flags
4509 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4510{
4511 return attr->ah_flags;
4512}
4513
4514static inline const struct ib_global_route
4515 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4516{
4517 return &attr->grh;
4518}
4519
4520/*To retrieve and modify the grh */
4521static inline struct ib_global_route
4522 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4523{
4524 return &attr->grh;
4525}
4526
4527static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4528{
4529 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4530
4531 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4532}
4533
4534static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4535 __be64 prefix)
4536{
4537 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4538
4539 grh->dgid.global.subnet_prefix = prefix;
4540}
4541
4542static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4543 __be64 if_id)
4544{
4545 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4546
4547 grh->dgid.global.interface_id = if_id;
4548}
4549
4550static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4551 union ib_gid *dgid, u32 flow_label,
4552 u8 sgid_index, u8 hop_limit,
4553 u8 traffic_class)
4554{
4555 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4556
4557 attr->ah_flags = IB_AH_GRH;
4558 if (dgid)
4559 grh->dgid = *dgid;
4560 grh->flow_label = flow_label;
4561 grh->sgid_index = sgid_index;
4562 grh->hop_limit = hop_limit;
4563 grh->traffic_class = traffic_class;
4564 grh->sgid_attr = NULL;
4565}
4566
4567void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4568void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4569 u32 flow_label, u8 hop_limit, u8 traffic_class,
4570 const struct ib_gid_attr *sgid_attr);
4571void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4572 const struct rdma_ah_attr *src);
4573void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4574 const struct rdma_ah_attr *new);
4575void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4576
4577/**
4578 * rdma_ah_find_type - Return address handle type.
4579 *
4580 * @dev: Device to be checked
4581 * @port_num: Port number
4582 */
4583static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4584 u8 port_num)
4585{
4586 if (rdma_protocol_roce(dev, port_num))
4587 return RDMA_AH_ATTR_TYPE_ROCE;
4588 if (rdma_protocol_ib(dev, port_num)) {
4589 if (rdma_cap_opa_ah(dev, port_num))
4590 return RDMA_AH_ATTR_TYPE_OPA;
4591 return RDMA_AH_ATTR_TYPE_IB;
4592 }
4593
4594 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4595}
4596
4597/**
4598 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4599 * In the current implementation the only way to get
4600 * get the 32bit lid is from other sources for OPA.
4601 * For IB, lids will always be 16bits so cast the
4602 * value accordingly.
4603 *
4604 * @lid: A 32bit LID
4605 */
4606static inline u16 ib_lid_cpu16(u32 lid)
4607{
4608 WARN_ON_ONCE(lid & 0xFFFF0000);
4609 return (u16)lid;
4610}
4611
4612/**
4613 * ib_lid_be16 - Return lid in 16bit BE encoding.
4614 *
4615 * @lid: A 32bit LID
4616 */
4617static inline __be16 ib_lid_be16(u32 lid)
4618{
4619 WARN_ON_ONCE(lid & 0xFFFF0000);
4620 return cpu_to_be16((u16)lid);
4621}
4622
4623/**
4624 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4625 * vector
4626 * @device: the rdma device
4627 * @comp_vector: index of completion vector
4628 *
4629 * Returns NULL on failure, otherwise a corresponding cpu map of the
4630 * completion vector (returns all-cpus map if the device driver doesn't
4631 * implement get_vector_affinity).
4632 */
4633static inline const struct cpumask *
4634ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4635{
4636 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4637 !device->ops.get_vector_affinity)
4638 return NULL;
4639
4640 return device->ops.get_vector_affinity(device, comp_vector);
4641
4642}
4643
4644/**
4645 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4646 * and add their gids, as needed, to the relevant RoCE devices.
4647 *
4648 * @device: the rdma device
4649 */
4650void rdma_roce_rescan_device(struct ib_device *ibdev);
4651
4652struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4653
4654int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4655
4656struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4657 enum rdma_netdev_t type, const char *name,
4658 unsigned char name_assign_type,
4659 void (*setup)(struct net_device *));
4660
4661int rdma_init_netdev(struct ib_device *device, u8 port_num,
4662 enum rdma_netdev_t type, const char *name,
4663 unsigned char name_assign_type,
4664 void (*setup)(struct net_device *),
4665 struct net_device *netdev);
4666
4667/**
4668 * rdma_set_device_sysfs_group - Set device attributes group to have
4669 * driver specific sysfs entries at
4670 * for infiniband class.
4671 *
4672 * @device: device pointer for which attributes to be created
4673 * @group: Pointer to group which should be added when device
4674 * is registered with sysfs.
4675 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4676 * group per device to have sysfs attributes.
4677 *
4678 * NOTE: New drivers should not make use of this API; instead new device
4679 * parameter should be exposed via netlink command. This API and mechanism
4680 * exist only for existing drivers.
4681 */
4682static inline void
4683rdma_set_device_sysfs_group(struct ib_device *dev,
4684 const struct attribute_group *group)
4685{
4686 dev->groups[1] = group;
4687}
4688
4689/**
4690 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4691 *
4692 * @device: device pointer for which ib_device pointer to retrieve
4693 *
4694 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4695 *
4696 */
4697static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4698{
4699 struct ib_core_device *coredev =
4700 container_of(device, struct ib_core_device, dev);
4701
4702 return coredev->owner;
4703}
4704
4705/**
4706 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4707 * ib_device holder structure from device pointer.
4708 *
4709 * NOTE: New drivers should not make use of this API; This API is only for
4710 * existing drivers who have exposed sysfs entries using
4711 * rdma_set_device_sysfs_group().
4712 */
4713#define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4714 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4715
4716bool rdma_dev_access_netns(const struct ib_device *device,
4717 const struct net *net);
4718
4719#define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4720#define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4721
4722/**
4723 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4724 * on the flow_label
4725 *
4726 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4727 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4728 * convention.
4729 */
4730static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4731{
4732 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4733
4734 fl_low ^= fl_high >> 14;
4735 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4736}
4737
4738/**
4739 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4740 * local and remote qpn values
4741 *
4742 * This function folded the multiplication results of two qpns, 24 bit each,
4743 * fields, and converts it to a 20 bit results.
4744 *
4745 * This function will create symmetric flow_label value based on the local
4746 * and remote qpn values. this will allow both the requester and responder
4747 * to calculate the same flow_label for a given connection.
4748 *
4749 * This helper function should be used by driver in case the upper layer
4750 * provide a zero flow_label value. This is to improve entropy of RDMA
4751 * traffic in the network.
4752 */
4753static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4754{
4755 u64 v = (u64)lqpn * rqpn;
4756
4757 v ^= v >> 20;
4758 v ^= v >> 40;
4759
4760 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4761}
4762#endif /* IB_VERBS_H */