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, 2020 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/ethtool.h>
16#include <linux/types.h>
17#include <linux/device.h>
18#include <linux/dma-mapping.h>
19#include <linux/kref.h>
20#include <linux/list.h>
21#include <linux/rwsem.h>
22#include <linux/workqueue.h>
23#include <linux/irq_poll.h>
24#include <uapi/linux/if_ether.h>
25#include <net/ipv6.h>
26#include <net/ip.h>
27#include <linux/string.h>
28#include <linux/slab.h>
29#include <linux/netdevice.h>
30#include <linux/refcount.h>
31#include <linux/if_link.h>
32#include <linux/atomic.h>
33#include <linux/mmu_notifier.h>
34#include <linux/uaccess.h>
35#include <linux/cgroup_rdma.h>
36#include <linux/irqflags.h>
37#include <linux/preempt.h>
38#include <linux/dim.h>
39#include <uapi/rdma/ib_user_verbs.h>
40#include <rdma/rdma_counter.h>
41#include <rdma/restrack.h>
42#include <rdma/signature.h>
43#include <uapi/rdma/rdma_user_ioctl.h>
44#include <uapi/rdma/ib_user_ioctl_verbs.h>
45
46#define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
47
48struct ib_umem_odp;
49struct ib_uqp_object;
50struct ib_usrq_object;
51struct ib_uwq_object;
52struct rdma_cm_id;
53struct ib_port;
54struct hw_stats_device_data;
55
56extern struct workqueue_struct *ib_wq;
57extern struct workqueue_struct *ib_comp_wq;
58extern struct workqueue_struct *ib_comp_unbound_wq;
59
60struct ib_ucq_object;
61
62__printf(3, 4) __cold
63void ibdev_printk(const char *level, const struct ib_device *ibdev,
64 const char *format, ...);
65__printf(2, 3) __cold
66void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
67__printf(2, 3) __cold
68void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
69__printf(2, 3) __cold
70void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
71__printf(2, 3) __cold
72void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
73__printf(2, 3) __cold
74void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
75__printf(2, 3) __cold
76void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
77__printf(2, 3) __cold
78void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
79
80#if defined(CONFIG_DYNAMIC_DEBUG) || \
81 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
82#define ibdev_dbg(__dev, format, args...) \
83 dynamic_ibdev_dbg(__dev, format, ##args)
84#else
85__printf(2, 3) __cold
86static inline
87void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
88#endif
89
90#define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
91do { \
92 static DEFINE_RATELIMIT_STATE(_rs, \
93 DEFAULT_RATELIMIT_INTERVAL, \
94 DEFAULT_RATELIMIT_BURST); \
95 if (__ratelimit(&_rs)) \
96 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
97} while (0)
98
99#define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
100 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
101#define ibdev_alert_ratelimited(ibdev, fmt, ...) \
102 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
103#define ibdev_crit_ratelimited(ibdev, fmt, ...) \
104 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
105#define ibdev_err_ratelimited(ibdev, fmt, ...) \
106 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
107#define ibdev_warn_ratelimited(ibdev, fmt, ...) \
108 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
109#define ibdev_notice_ratelimited(ibdev, fmt, ...) \
110 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
111#define ibdev_info_ratelimited(ibdev, fmt, ...) \
112 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
113
114#if defined(CONFIG_DYNAMIC_DEBUG) || \
115 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE))
116/* descriptor check is first to prevent flooding with "callbacks suppressed" */
117#define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
118do { \
119 static DEFINE_RATELIMIT_STATE(_rs, \
120 DEFAULT_RATELIMIT_INTERVAL, \
121 DEFAULT_RATELIMIT_BURST); \
122 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
123 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
124 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
125 ##__VA_ARGS__); \
126} while (0)
127#else
128__printf(2, 3) __cold
129static inline
130void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
131#endif
132
133union ib_gid {
134 u8 raw[16];
135 struct {
136 __be64 subnet_prefix;
137 __be64 interface_id;
138 } global;
139};
140
141extern union ib_gid zgid;
142
143enum ib_gid_type {
144 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB,
145 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1,
146 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2,
147 IB_GID_TYPE_SIZE
148};
149
150#define ROCE_V2_UDP_DPORT 4791
151struct ib_gid_attr {
152 struct net_device __rcu *ndev;
153 struct ib_device *device;
154 union ib_gid gid;
155 enum ib_gid_type gid_type;
156 u16 index;
157 u32 port_num;
158};
159
160enum {
161 /* set the local administered indication */
162 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
163};
164
165enum rdma_transport_type {
166 RDMA_TRANSPORT_IB,
167 RDMA_TRANSPORT_IWARP,
168 RDMA_TRANSPORT_USNIC,
169 RDMA_TRANSPORT_USNIC_UDP,
170 RDMA_TRANSPORT_UNSPECIFIED,
171};
172
173enum rdma_protocol_type {
174 RDMA_PROTOCOL_IB,
175 RDMA_PROTOCOL_IBOE,
176 RDMA_PROTOCOL_IWARP,
177 RDMA_PROTOCOL_USNIC_UDP
178};
179
180__attribute_const__ enum rdma_transport_type
181rdma_node_get_transport(unsigned int node_type);
182
183enum rdma_network_type {
184 RDMA_NETWORK_IB,
185 RDMA_NETWORK_ROCE_V1,
186 RDMA_NETWORK_IPV4,
187 RDMA_NETWORK_IPV6
188};
189
190static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
191{
192 if (network_type == RDMA_NETWORK_IPV4 ||
193 network_type == RDMA_NETWORK_IPV6)
194 return IB_GID_TYPE_ROCE_UDP_ENCAP;
195 else if (network_type == RDMA_NETWORK_ROCE_V1)
196 return IB_GID_TYPE_ROCE;
197 else
198 return IB_GID_TYPE_IB;
199}
200
201static inline enum rdma_network_type
202rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
203{
204 if (attr->gid_type == IB_GID_TYPE_IB)
205 return RDMA_NETWORK_IB;
206
207 if (attr->gid_type == IB_GID_TYPE_ROCE)
208 return RDMA_NETWORK_ROCE_V1;
209
210 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
211 return RDMA_NETWORK_IPV4;
212 else
213 return RDMA_NETWORK_IPV6;
214}
215
216enum rdma_link_layer {
217 IB_LINK_LAYER_UNSPECIFIED,
218 IB_LINK_LAYER_INFINIBAND,
219 IB_LINK_LAYER_ETHERNET,
220};
221
222enum ib_device_cap_flags {
223 IB_DEVICE_RESIZE_MAX_WR = IB_UVERBS_DEVICE_RESIZE_MAX_WR,
224 IB_DEVICE_BAD_PKEY_CNTR = IB_UVERBS_DEVICE_BAD_PKEY_CNTR,
225 IB_DEVICE_BAD_QKEY_CNTR = IB_UVERBS_DEVICE_BAD_QKEY_CNTR,
226 IB_DEVICE_RAW_MULTI = IB_UVERBS_DEVICE_RAW_MULTI,
227 IB_DEVICE_AUTO_PATH_MIG = IB_UVERBS_DEVICE_AUTO_PATH_MIG,
228 IB_DEVICE_CHANGE_PHY_PORT = IB_UVERBS_DEVICE_CHANGE_PHY_PORT,
229 IB_DEVICE_UD_AV_PORT_ENFORCE = IB_UVERBS_DEVICE_UD_AV_PORT_ENFORCE,
230 IB_DEVICE_CURR_QP_STATE_MOD = IB_UVERBS_DEVICE_CURR_QP_STATE_MOD,
231 IB_DEVICE_SHUTDOWN_PORT = IB_UVERBS_DEVICE_SHUTDOWN_PORT,
232 /* IB_DEVICE_INIT_TYPE = IB_UVERBS_DEVICE_INIT_TYPE, (not in use) */
233 IB_DEVICE_PORT_ACTIVE_EVENT = IB_UVERBS_DEVICE_PORT_ACTIVE_EVENT,
234 IB_DEVICE_SYS_IMAGE_GUID = IB_UVERBS_DEVICE_SYS_IMAGE_GUID,
235 IB_DEVICE_RC_RNR_NAK_GEN = IB_UVERBS_DEVICE_RC_RNR_NAK_GEN,
236 IB_DEVICE_SRQ_RESIZE = IB_UVERBS_DEVICE_SRQ_RESIZE,
237 IB_DEVICE_N_NOTIFY_CQ = IB_UVERBS_DEVICE_N_NOTIFY_CQ,
238
239 /* Reserved, old SEND_W_INV = 1 << 16,*/
240 IB_DEVICE_MEM_WINDOW = IB_UVERBS_DEVICE_MEM_WINDOW,
241 /*
242 * Devices should set IB_DEVICE_UD_IP_SUM if they support
243 * insertion of UDP and TCP checksum on outgoing UD IPoIB
244 * messages and can verify the validity of checksum for
245 * incoming messages. Setting this flag implies that the
246 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
247 */
248 IB_DEVICE_UD_IP_CSUM = IB_UVERBS_DEVICE_UD_IP_CSUM,
249 IB_DEVICE_XRC = IB_UVERBS_DEVICE_XRC,
250
251 /*
252 * This device supports the IB "base memory management extension",
253 * which includes support for fast registrations (IB_WR_REG_MR,
254 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
255 * also be set by any iWarp device which must support FRs to comply
256 * to the iWarp verbs spec. iWarp devices also support the
257 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
258 * stag.
259 */
260 IB_DEVICE_MEM_MGT_EXTENSIONS = IB_UVERBS_DEVICE_MEM_MGT_EXTENSIONS,
261 IB_DEVICE_MEM_WINDOW_TYPE_2A = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2A,
262 IB_DEVICE_MEM_WINDOW_TYPE_2B = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2B,
263 IB_DEVICE_RC_IP_CSUM = IB_UVERBS_DEVICE_RC_IP_CSUM,
264 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
265 IB_DEVICE_RAW_IP_CSUM = IB_UVERBS_DEVICE_RAW_IP_CSUM,
266 IB_DEVICE_MANAGED_FLOW_STEERING =
267 IB_UVERBS_DEVICE_MANAGED_FLOW_STEERING,
268 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
269 IB_DEVICE_RAW_SCATTER_FCS = IB_UVERBS_DEVICE_RAW_SCATTER_FCS,
270 /* The device supports padding incoming writes to cacheline. */
271 IB_DEVICE_PCI_WRITE_END_PADDING =
272 IB_UVERBS_DEVICE_PCI_WRITE_END_PADDING,
273 /* Placement type attributes */
274 IB_DEVICE_FLUSH_GLOBAL = IB_UVERBS_DEVICE_FLUSH_GLOBAL,
275 IB_DEVICE_FLUSH_PERSISTENT = IB_UVERBS_DEVICE_FLUSH_PERSISTENT,
276 IB_DEVICE_ATOMIC_WRITE = IB_UVERBS_DEVICE_ATOMIC_WRITE,
277};
278
279enum ib_kernel_cap_flags {
280 /*
281 * This device supports a per-device lkey or stag that can be
282 * used without performing a memory registration for the local
283 * memory. Note that ULPs should never check this flag, but
284 * instead of use the local_dma_lkey flag in the ib_pd structure,
285 * which will always contain a usable lkey.
286 */
287 IBK_LOCAL_DMA_LKEY = 1 << 0,
288 /* IB_QP_CREATE_INTEGRITY_EN is supported to implement T10-PI */
289 IBK_INTEGRITY_HANDOVER = 1 << 1,
290 /* IB_ACCESS_ON_DEMAND is supported during reg_user_mr() */
291 IBK_ON_DEMAND_PAGING = 1 << 2,
292 /* IB_MR_TYPE_SG_GAPS is supported */
293 IBK_SG_GAPS_REG = 1 << 3,
294 /* Driver supports RDMA_NLDEV_CMD_DELLINK */
295 IBK_ALLOW_USER_UNREG = 1 << 4,
296
297 /* ipoib will use IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK */
298 IBK_BLOCK_MULTICAST_LOOPBACK = 1 << 5,
299 /* iopib will use IB_QP_CREATE_IPOIB_UD_LSO for its QPs */
300 IBK_UD_TSO = 1 << 6,
301 /* iopib will use the device ops:
302 * get_vf_config
303 * get_vf_guid
304 * get_vf_stats
305 * set_vf_guid
306 * set_vf_link_state
307 */
308 IBK_VIRTUAL_FUNCTION = 1 << 7,
309 /* ipoib will use IB_QP_CREATE_NETDEV_USE for its QPs */
310 IBK_RDMA_NETDEV_OPA = 1 << 8,
311};
312
313enum ib_atomic_cap {
314 IB_ATOMIC_NONE,
315 IB_ATOMIC_HCA,
316 IB_ATOMIC_GLOB
317};
318
319enum ib_odp_general_cap_bits {
320 IB_ODP_SUPPORT = 1 << 0,
321 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
322};
323
324enum ib_odp_transport_cap_bits {
325 IB_ODP_SUPPORT_SEND = 1 << 0,
326 IB_ODP_SUPPORT_RECV = 1 << 1,
327 IB_ODP_SUPPORT_WRITE = 1 << 2,
328 IB_ODP_SUPPORT_READ = 1 << 3,
329 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
330 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
331};
332
333struct ib_odp_caps {
334 uint64_t general_caps;
335 struct {
336 uint32_t rc_odp_caps;
337 uint32_t uc_odp_caps;
338 uint32_t ud_odp_caps;
339 uint32_t xrc_odp_caps;
340 } per_transport_caps;
341};
342
343struct ib_rss_caps {
344 /* Corresponding bit will be set if qp type from
345 * 'enum ib_qp_type' is supported, e.g.
346 * supported_qpts |= 1 << IB_QPT_UD
347 */
348 u32 supported_qpts;
349 u32 max_rwq_indirection_tables;
350 u32 max_rwq_indirection_table_size;
351};
352
353enum ib_tm_cap_flags {
354 /* Support tag matching with rendezvous offload for RC transport */
355 IB_TM_CAP_RNDV_RC = 1 << 0,
356};
357
358struct ib_tm_caps {
359 /* Max size of RNDV header */
360 u32 max_rndv_hdr_size;
361 /* Max number of entries in tag matching list */
362 u32 max_num_tags;
363 /* From enum ib_tm_cap_flags */
364 u32 flags;
365 /* Max number of outstanding list operations */
366 u32 max_ops;
367 /* Max number of SGE in tag matching entry */
368 u32 max_sge;
369};
370
371struct ib_cq_init_attr {
372 unsigned int cqe;
373 u32 comp_vector;
374 u32 flags;
375};
376
377enum ib_cq_attr_mask {
378 IB_CQ_MODERATE = 1 << 0,
379};
380
381struct ib_cq_caps {
382 u16 max_cq_moderation_count;
383 u16 max_cq_moderation_period;
384};
385
386struct ib_dm_mr_attr {
387 u64 length;
388 u64 offset;
389 u32 access_flags;
390};
391
392struct ib_dm_alloc_attr {
393 u64 length;
394 u32 alignment;
395 u32 flags;
396};
397
398struct ib_device_attr {
399 u64 fw_ver;
400 __be64 sys_image_guid;
401 u64 max_mr_size;
402 u64 page_size_cap;
403 u32 vendor_id;
404 u32 vendor_part_id;
405 u32 hw_ver;
406 int max_qp;
407 int max_qp_wr;
408 u64 device_cap_flags;
409 u64 kernel_cap_flags;
410 int max_send_sge;
411 int max_recv_sge;
412 int max_sge_rd;
413 int max_cq;
414 int max_cqe;
415 int max_mr;
416 int max_pd;
417 int max_qp_rd_atom;
418 int max_ee_rd_atom;
419 int max_res_rd_atom;
420 int max_qp_init_rd_atom;
421 int max_ee_init_rd_atom;
422 enum ib_atomic_cap atomic_cap;
423 enum ib_atomic_cap masked_atomic_cap;
424 int max_ee;
425 int max_rdd;
426 int max_mw;
427 int max_raw_ipv6_qp;
428 int max_raw_ethy_qp;
429 int max_mcast_grp;
430 int max_mcast_qp_attach;
431 int max_total_mcast_qp_attach;
432 int max_ah;
433 int max_srq;
434 int max_srq_wr;
435 int max_srq_sge;
436 unsigned int max_fast_reg_page_list_len;
437 unsigned int max_pi_fast_reg_page_list_len;
438 u16 max_pkeys;
439 u8 local_ca_ack_delay;
440 int sig_prot_cap;
441 int sig_guard_cap;
442 struct ib_odp_caps odp_caps;
443 uint64_t timestamp_mask;
444 uint64_t hca_core_clock; /* in KHZ */
445 struct ib_rss_caps rss_caps;
446 u32 max_wq_type_rq;
447 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
448 struct ib_tm_caps tm_caps;
449 struct ib_cq_caps cq_caps;
450 u64 max_dm_size;
451 /* Max entries for sgl for optimized performance per READ */
452 u32 max_sgl_rd;
453};
454
455enum ib_mtu {
456 IB_MTU_256 = 1,
457 IB_MTU_512 = 2,
458 IB_MTU_1024 = 3,
459 IB_MTU_2048 = 4,
460 IB_MTU_4096 = 5
461};
462
463enum opa_mtu {
464 OPA_MTU_8192 = 6,
465 OPA_MTU_10240 = 7
466};
467
468static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
469{
470 switch (mtu) {
471 case IB_MTU_256: return 256;
472 case IB_MTU_512: return 512;
473 case IB_MTU_1024: return 1024;
474 case IB_MTU_2048: return 2048;
475 case IB_MTU_4096: return 4096;
476 default: return -1;
477 }
478}
479
480static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
481{
482 if (mtu >= 4096)
483 return IB_MTU_4096;
484 else if (mtu >= 2048)
485 return IB_MTU_2048;
486 else if (mtu >= 1024)
487 return IB_MTU_1024;
488 else if (mtu >= 512)
489 return IB_MTU_512;
490 else
491 return IB_MTU_256;
492}
493
494static inline int opa_mtu_enum_to_int(enum opa_mtu mtu)
495{
496 switch (mtu) {
497 case OPA_MTU_8192:
498 return 8192;
499 case OPA_MTU_10240:
500 return 10240;
501 default:
502 return(ib_mtu_enum_to_int((enum ib_mtu)mtu));
503 }
504}
505
506static inline enum opa_mtu opa_mtu_int_to_enum(int mtu)
507{
508 if (mtu >= 10240)
509 return OPA_MTU_10240;
510 else if (mtu >= 8192)
511 return OPA_MTU_8192;
512 else
513 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu));
514}
515
516enum ib_port_state {
517 IB_PORT_NOP = 0,
518 IB_PORT_DOWN = 1,
519 IB_PORT_INIT = 2,
520 IB_PORT_ARMED = 3,
521 IB_PORT_ACTIVE = 4,
522 IB_PORT_ACTIVE_DEFER = 5
523};
524
525enum ib_port_phys_state {
526 IB_PORT_PHYS_STATE_SLEEP = 1,
527 IB_PORT_PHYS_STATE_POLLING = 2,
528 IB_PORT_PHYS_STATE_DISABLED = 3,
529 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
530 IB_PORT_PHYS_STATE_LINK_UP = 5,
531 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
532 IB_PORT_PHYS_STATE_PHY_TEST = 7,
533};
534
535enum ib_port_width {
536 IB_WIDTH_1X = 1,
537 IB_WIDTH_2X = 16,
538 IB_WIDTH_4X = 2,
539 IB_WIDTH_8X = 4,
540 IB_WIDTH_12X = 8
541};
542
543static inline int ib_width_enum_to_int(enum ib_port_width width)
544{
545 switch (width) {
546 case IB_WIDTH_1X: return 1;
547 case IB_WIDTH_2X: return 2;
548 case IB_WIDTH_4X: return 4;
549 case IB_WIDTH_8X: return 8;
550 case IB_WIDTH_12X: return 12;
551 default: return -1;
552 }
553}
554
555enum ib_port_speed {
556 IB_SPEED_SDR = 1,
557 IB_SPEED_DDR = 2,
558 IB_SPEED_QDR = 4,
559 IB_SPEED_FDR10 = 8,
560 IB_SPEED_FDR = 16,
561 IB_SPEED_EDR = 32,
562 IB_SPEED_HDR = 64,
563 IB_SPEED_NDR = 128,
564 IB_SPEED_XDR = 256,
565};
566
567enum ib_stat_flag {
568 IB_STAT_FLAG_OPTIONAL = 1 << 0,
569};
570
571/**
572 * struct rdma_stat_desc
573 * @name - The name of the counter
574 * @flags - Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL
575 * @priv - Driver private information; Core code should not use
576 */
577struct rdma_stat_desc {
578 const char *name;
579 unsigned int flags;
580 const void *priv;
581};
582
583/**
584 * struct rdma_hw_stats
585 * @lock - Mutex to protect parallel write access to lifespan and values
586 * of counters, which are 64bits and not guaranteed to be written
587 * atomicaly on 32bits systems.
588 * @timestamp - Used by the core code to track when the last update was
589 * @lifespan - Used by the core code to determine how old the counters
590 * should be before being updated again. Stored in jiffies, defaults
591 * to 10 milliseconds, drivers can override the default be specifying
592 * their own value during their allocation routine.
593 * @descs - Array of pointers to static descriptors used for the counters
594 * in directory.
595 * @is_disabled - A bitmap to indicate each counter is currently disabled
596 * or not.
597 * @num_counters - How many hardware counters there are. If name is
598 * shorter than this number, a kernel oops will result. Driver authors
599 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
600 * in their code to prevent this.
601 * @value - Array of u64 counters that are accessed by the sysfs code and
602 * filled in by the drivers get_stats routine
603 */
604struct rdma_hw_stats {
605 struct mutex lock; /* Protect lifespan and values[] */
606 unsigned long timestamp;
607 unsigned long lifespan;
608 const struct rdma_stat_desc *descs;
609 unsigned long *is_disabled;
610 int num_counters;
611 u64 value[] __counted_by(num_counters);
612};
613
614#define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
615
616struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
617 const struct rdma_stat_desc *descs, int num_counters,
618 unsigned long lifespan);
619
620void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats);
621
622/* Define bits for the various functionality this port needs to be supported by
623 * the core.
624 */
625/* Management 0x00000FFF */
626#define RDMA_CORE_CAP_IB_MAD 0x00000001
627#define RDMA_CORE_CAP_IB_SMI 0x00000002
628#define RDMA_CORE_CAP_IB_CM 0x00000004
629#define RDMA_CORE_CAP_IW_CM 0x00000008
630#define RDMA_CORE_CAP_IB_SA 0x00000010
631#define RDMA_CORE_CAP_OPA_MAD 0x00000020
632
633/* Address format 0x000FF000 */
634#define RDMA_CORE_CAP_AF_IB 0x00001000
635#define RDMA_CORE_CAP_ETH_AH 0x00002000
636#define RDMA_CORE_CAP_OPA_AH 0x00004000
637#define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
638
639/* Protocol 0xFFF00000 */
640#define RDMA_CORE_CAP_PROT_IB 0x00100000
641#define RDMA_CORE_CAP_PROT_ROCE 0x00200000
642#define RDMA_CORE_CAP_PROT_IWARP 0x00400000
643#define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
644#define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
645#define RDMA_CORE_CAP_PROT_USNIC 0x02000000
646
647#define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
648 | RDMA_CORE_CAP_PROT_ROCE \
649 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
650
651#define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
652 | RDMA_CORE_CAP_IB_MAD \
653 | RDMA_CORE_CAP_IB_SMI \
654 | RDMA_CORE_CAP_IB_CM \
655 | RDMA_CORE_CAP_IB_SA \
656 | RDMA_CORE_CAP_AF_IB)
657#define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
658 | RDMA_CORE_CAP_IB_MAD \
659 | RDMA_CORE_CAP_IB_CM \
660 | RDMA_CORE_CAP_AF_IB \
661 | RDMA_CORE_CAP_ETH_AH)
662#define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
663 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
664 | RDMA_CORE_CAP_IB_MAD \
665 | RDMA_CORE_CAP_IB_CM \
666 | RDMA_CORE_CAP_AF_IB \
667 | RDMA_CORE_CAP_ETH_AH)
668#define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
669 | RDMA_CORE_CAP_IW_CM)
670#define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
671 | RDMA_CORE_CAP_OPA_MAD)
672
673#define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
674
675#define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
676
677struct ib_port_attr {
678 u64 subnet_prefix;
679 enum ib_port_state state;
680 enum ib_mtu max_mtu;
681 enum ib_mtu active_mtu;
682 u32 phys_mtu;
683 int gid_tbl_len;
684 unsigned int ip_gids:1;
685 /* This is the value from PortInfo CapabilityMask, defined by IBA */
686 u32 port_cap_flags;
687 u32 max_msg_sz;
688 u32 bad_pkey_cntr;
689 u32 qkey_viol_cntr;
690 u16 pkey_tbl_len;
691 u32 sm_lid;
692 u32 lid;
693 u8 lmc;
694 u8 max_vl_num;
695 u8 sm_sl;
696 u8 subnet_timeout;
697 u8 init_type_reply;
698 u8 active_width;
699 u16 active_speed;
700 u8 phys_state;
701 u16 port_cap_flags2;
702};
703
704enum ib_device_modify_flags {
705 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
706 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
707};
708
709#define IB_DEVICE_NODE_DESC_MAX 64
710
711struct ib_device_modify {
712 u64 sys_image_guid;
713 char node_desc[IB_DEVICE_NODE_DESC_MAX];
714};
715
716enum ib_port_modify_flags {
717 IB_PORT_SHUTDOWN = 1,
718 IB_PORT_INIT_TYPE = (1<<2),
719 IB_PORT_RESET_QKEY_CNTR = (1<<3),
720 IB_PORT_OPA_MASK_CHG = (1<<4)
721};
722
723struct ib_port_modify {
724 u32 set_port_cap_mask;
725 u32 clr_port_cap_mask;
726 u8 init_type;
727};
728
729enum ib_event_type {
730 IB_EVENT_CQ_ERR,
731 IB_EVENT_QP_FATAL,
732 IB_EVENT_QP_REQ_ERR,
733 IB_EVENT_QP_ACCESS_ERR,
734 IB_EVENT_COMM_EST,
735 IB_EVENT_SQ_DRAINED,
736 IB_EVENT_PATH_MIG,
737 IB_EVENT_PATH_MIG_ERR,
738 IB_EVENT_DEVICE_FATAL,
739 IB_EVENT_PORT_ACTIVE,
740 IB_EVENT_PORT_ERR,
741 IB_EVENT_LID_CHANGE,
742 IB_EVENT_PKEY_CHANGE,
743 IB_EVENT_SM_CHANGE,
744 IB_EVENT_SRQ_ERR,
745 IB_EVENT_SRQ_LIMIT_REACHED,
746 IB_EVENT_QP_LAST_WQE_REACHED,
747 IB_EVENT_CLIENT_REREGISTER,
748 IB_EVENT_GID_CHANGE,
749 IB_EVENT_WQ_FATAL,
750};
751
752const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
753
754struct ib_event {
755 struct ib_device *device;
756 union {
757 struct ib_cq *cq;
758 struct ib_qp *qp;
759 struct ib_srq *srq;
760 struct ib_wq *wq;
761 u32 port_num;
762 } element;
763 enum ib_event_type event;
764};
765
766struct ib_event_handler {
767 struct ib_device *device;
768 void (*handler)(struct ib_event_handler *, struct ib_event *);
769 struct list_head list;
770};
771
772#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
773 do { \
774 (_ptr)->device = _device; \
775 (_ptr)->handler = _handler; \
776 INIT_LIST_HEAD(&(_ptr)->list); \
777 } while (0)
778
779struct ib_global_route {
780 const struct ib_gid_attr *sgid_attr;
781 union ib_gid dgid;
782 u32 flow_label;
783 u8 sgid_index;
784 u8 hop_limit;
785 u8 traffic_class;
786};
787
788struct ib_grh {
789 __be32 version_tclass_flow;
790 __be16 paylen;
791 u8 next_hdr;
792 u8 hop_limit;
793 union ib_gid sgid;
794 union ib_gid dgid;
795};
796
797union rdma_network_hdr {
798 struct ib_grh ibgrh;
799 struct {
800 /* The IB spec states that if it's IPv4, the header
801 * is located in the last 20 bytes of the header.
802 */
803 u8 reserved[20];
804 struct iphdr roce4grh;
805 };
806};
807
808#define IB_QPN_MASK 0xFFFFFF
809
810enum {
811 IB_MULTICAST_QPN = 0xffffff
812};
813
814#define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
815#define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
816
817enum ib_ah_flags {
818 IB_AH_GRH = 1
819};
820
821enum ib_rate {
822 IB_RATE_PORT_CURRENT = 0,
823 IB_RATE_2_5_GBPS = 2,
824 IB_RATE_5_GBPS = 5,
825 IB_RATE_10_GBPS = 3,
826 IB_RATE_20_GBPS = 6,
827 IB_RATE_30_GBPS = 4,
828 IB_RATE_40_GBPS = 7,
829 IB_RATE_60_GBPS = 8,
830 IB_RATE_80_GBPS = 9,
831 IB_RATE_120_GBPS = 10,
832 IB_RATE_14_GBPS = 11,
833 IB_RATE_56_GBPS = 12,
834 IB_RATE_112_GBPS = 13,
835 IB_RATE_168_GBPS = 14,
836 IB_RATE_25_GBPS = 15,
837 IB_RATE_100_GBPS = 16,
838 IB_RATE_200_GBPS = 17,
839 IB_RATE_300_GBPS = 18,
840 IB_RATE_28_GBPS = 19,
841 IB_RATE_50_GBPS = 20,
842 IB_RATE_400_GBPS = 21,
843 IB_RATE_600_GBPS = 22,
844 IB_RATE_800_GBPS = 23,
845};
846
847/**
848 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
849 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
850 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
851 * @rate: rate to convert.
852 */
853__attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
854
855/**
856 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
857 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
858 * @rate: rate to convert.
859 */
860__attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
861
862
863/**
864 * enum ib_mr_type - memory region type
865 * @IB_MR_TYPE_MEM_REG: memory region that is used for
866 * normal registration
867 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
868 * register any arbitrary sg lists (without
869 * the normal mr constraints - see
870 * ib_map_mr_sg)
871 * @IB_MR_TYPE_DM: memory region that is used for device
872 * memory registration
873 * @IB_MR_TYPE_USER: memory region that is used for the user-space
874 * application
875 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
876 * without address translations (VA=PA)
877 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
878 * data integrity operations
879 */
880enum ib_mr_type {
881 IB_MR_TYPE_MEM_REG,
882 IB_MR_TYPE_SG_GAPS,
883 IB_MR_TYPE_DM,
884 IB_MR_TYPE_USER,
885 IB_MR_TYPE_DMA,
886 IB_MR_TYPE_INTEGRITY,
887};
888
889enum ib_mr_status_check {
890 IB_MR_CHECK_SIG_STATUS = 1,
891};
892
893/**
894 * struct ib_mr_status - Memory region status container
895 *
896 * @fail_status: Bitmask of MR checks status. For each
897 * failed check a corresponding status bit is set.
898 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
899 * failure.
900 */
901struct ib_mr_status {
902 u32 fail_status;
903 struct ib_sig_err sig_err;
904};
905
906/**
907 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
908 * enum.
909 * @mult: multiple to convert.
910 */
911__attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
912
913struct rdma_ah_init_attr {
914 struct rdma_ah_attr *ah_attr;
915 u32 flags;
916 struct net_device *xmit_slave;
917};
918
919enum rdma_ah_attr_type {
920 RDMA_AH_ATTR_TYPE_UNDEFINED,
921 RDMA_AH_ATTR_TYPE_IB,
922 RDMA_AH_ATTR_TYPE_ROCE,
923 RDMA_AH_ATTR_TYPE_OPA,
924};
925
926struct ib_ah_attr {
927 u16 dlid;
928 u8 src_path_bits;
929};
930
931struct roce_ah_attr {
932 u8 dmac[ETH_ALEN];
933};
934
935struct opa_ah_attr {
936 u32 dlid;
937 u8 src_path_bits;
938 bool make_grd;
939};
940
941struct rdma_ah_attr {
942 struct ib_global_route grh;
943 u8 sl;
944 u8 static_rate;
945 u32 port_num;
946 u8 ah_flags;
947 enum rdma_ah_attr_type type;
948 union {
949 struct ib_ah_attr ib;
950 struct roce_ah_attr roce;
951 struct opa_ah_attr opa;
952 };
953};
954
955enum ib_wc_status {
956 IB_WC_SUCCESS,
957 IB_WC_LOC_LEN_ERR,
958 IB_WC_LOC_QP_OP_ERR,
959 IB_WC_LOC_EEC_OP_ERR,
960 IB_WC_LOC_PROT_ERR,
961 IB_WC_WR_FLUSH_ERR,
962 IB_WC_MW_BIND_ERR,
963 IB_WC_BAD_RESP_ERR,
964 IB_WC_LOC_ACCESS_ERR,
965 IB_WC_REM_INV_REQ_ERR,
966 IB_WC_REM_ACCESS_ERR,
967 IB_WC_REM_OP_ERR,
968 IB_WC_RETRY_EXC_ERR,
969 IB_WC_RNR_RETRY_EXC_ERR,
970 IB_WC_LOC_RDD_VIOL_ERR,
971 IB_WC_REM_INV_RD_REQ_ERR,
972 IB_WC_REM_ABORT_ERR,
973 IB_WC_INV_EECN_ERR,
974 IB_WC_INV_EEC_STATE_ERR,
975 IB_WC_FATAL_ERR,
976 IB_WC_RESP_TIMEOUT_ERR,
977 IB_WC_GENERAL_ERR
978};
979
980const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
981
982enum ib_wc_opcode {
983 IB_WC_SEND = IB_UVERBS_WC_SEND,
984 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE,
985 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ,
986 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP,
987 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD,
988 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW,
989 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV,
990 IB_WC_LSO = IB_UVERBS_WC_TSO,
991 IB_WC_ATOMIC_WRITE = IB_UVERBS_WC_ATOMIC_WRITE,
992 IB_WC_REG_MR,
993 IB_WC_MASKED_COMP_SWAP,
994 IB_WC_MASKED_FETCH_ADD,
995 IB_WC_FLUSH = IB_UVERBS_WC_FLUSH,
996/*
997 * Set value of IB_WC_RECV so consumers can test if a completion is a
998 * receive by testing (opcode & IB_WC_RECV).
999 */
1000 IB_WC_RECV = 1 << 7,
1001 IB_WC_RECV_RDMA_WITH_IMM
1002};
1003
1004enum ib_wc_flags {
1005 IB_WC_GRH = 1,
1006 IB_WC_WITH_IMM = (1<<1),
1007 IB_WC_WITH_INVALIDATE = (1<<2),
1008 IB_WC_IP_CSUM_OK = (1<<3),
1009 IB_WC_WITH_SMAC = (1<<4),
1010 IB_WC_WITH_VLAN = (1<<5),
1011 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
1012};
1013
1014struct ib_wc {
1015 union {
1016 u64 wr_id;
1017 struct ib_cqe *wr_cqe;
1018 };
1019 enum ib_wc_status status;
1020 enum ib_wc_opcode opcode;
1021 u32 vendor_err;
1022 u32 byte_len;
1023 struct ib_qp *qp;
1024 union {
1025 __be32 imm_data;
1026 u32 invalidate_rkey;
1027 } ex;
1028 u32 src_qp;
1029 u32 slid;
1030 int wc_flags;
1031 u16 pkey_index;
1032 u8 sl;
1033 u8 dlid_path_bits;
1034 u32 port_num; /* valid only for DR SMPs on switches */
1035 u8 smac[ETH_ALEN];
1036 u16 vlan_id;
1037 u8 network_hdr_type;
1038};
1039
1040enum ib_cq_notify_flags {
1041 IB_CQ_SOLICITED = 1 << 0,
1042 IB_CQ_NEXT_COMP = 1 << 1,
1043 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1044 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1045};
1046
1047enum ib_srq_type {
1048 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC,
1049 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC,
1050 IB_SRQT_TM = IB_UVERBS_SRQT_TM,
1051};
1052
1053static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1054{
1055 return srq_type == IB_SRQT_XRC ||
1056 srq_type == IB_SRQT_TM;
1057}
1058
1059enum ib_srq_attr_mask {
1060 IB_SRQ_MAX_WR = 1 << 0,
1061 IB_SRQ_LIMIT = 1 << 1,
1062};
1063
1064struct ib_srq_attr {
1065 u32 max_wr;
1066 u32 max_sge;
1067 u32 srq_limit;
1068};
1069
1070struct ib_srq_init_attr {
1071 void (*event_handler)(struct ib_event *, void *);
1072 void *srq_context;
1073 struct ib_srq_attr attr;
1074 enum ib_srq_type srq_type;
1075
1076 struct {
1077 struct ib_cq *cq;
1078 union {
1079 struct {
1080 struct ib_xrcd *xrcd;
1081 } xrc;
1082
1083 struct {
1084 u32 max_num_tags;
1085 } tag_matching;
1086 };
1087 } ext;
1088};
1089
1090struct ib_qp_cap {
1091 u32 max_send_wr;
1092 u32 max_recv_wr;
1093 u32 max_send_sge;
1094 u32 max_recv_sge;
1095 u32 max_inline_data;
1096
1097 /*
1098 * Maximum number of rdma_rw_ctx structures in flight at a time.
1099 * ib_create_qp() will calculate the right amount of needed WRs
1100 * and MRs based on this.
1101 */
1102 u32 max_rdma_ctxs;
1103};
1104
1105enum ib_sig_type {
1106 IB_SIGNAL_ALL_WR,
1107 IB_SIGNAL_REQ_WR
1108};
1109
1110enum ib_qp_type {
1111 /*
1112 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1113 * here (and in that order) since the MAD layer uses them as
1114 * indices into a 2-entry table.
1115 */
1116 IB_QPT_SMI,
1117 IB_QPT_GSI,
1118
1119 IB_QPT_RC = IB_UVERBS_QPT_RC,
1120 IB_QPT_UC = IB_UVERBS_QPT_UC,
1121 IB_QPT_UD = IB_UVERBS_QPT_UD,
1122 IB_QPT_RAW_IPV6,
1123 IB_QPT_RAW_ETHERTYPE,
1124 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET,
1125 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI,
1126 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT,
1127 IB_QPT_MAX,
1128 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER,
1129 /* Reserve a range for qp types internal to the low level driver.
1130 * These qp types will not be visible at the IB core layer, so the
1131 * IB_QPT_MAX usages should not be affected in the core layer
1132 */
1133 IB_QPT_RESERVED1 = 0x1000,
1134 IB_QPT_RESERVED2,
1135 IB_QPT_RESERVED3,
1136 IB_QPT_RESERVED4,
1137 IB_QPT_RESERVED5,
1138 IB_QPT_RESERVED6,
1139 IB_QPT_RESERVED7,
1140 IB_QPT_RESERVED8,
1141 IB_QPT_RESERVED9,
1142 IB_QPT_RESERVED10,
1143};
1144
1145enum ib_qp_create_flags {
1146 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1147 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK =
1148 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
1149 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1150 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1151 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1152 IB_QP_CREATE_NETIF_QP = 1 << 5,
1153 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
1154 IB_QP_CREATE_NETDEV_USE = 1 << 7,
1155 IB_QP_CREATE_SCATTER_FCS =
1156 IB_UVERBS_QP_CREATE_SCATTER_FCS,
1157 IB_QP_CREATE_CVLAN_STRIPPING =
1158 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING,
1159 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
1160 IB_QP_CREATE_PCI_WRITE_END_PADDING =
1161 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING,
1162 /* reserve bits 26-31 for low level drivers' internal use */
1163 IB_QP_CREATE_RESERVED_START = 1 << 26,
1164 IB_QP_CREATE_RESERVED_END = 1 << 31,
1165};
1166
1167/*
1168 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1169 * callback to destroy the passed in QP.
1170 */
1171
1172struct ib_qp_init_attr {
1173 /* This callback occurs in workqueue context */
1174 void (*event_handler)(struct ib_event *, void *);
1175
1176 void *qp_context;
1177 struct ib_cq *send_cq;
1178 struct ib_cq *recv_cq;
1179 struct ib_srq *srq;
1180 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1181 struct ib_qp_cap cap;
1182 enum ib_sig_type sq_sig_type;
1183 enum ib_qp_type qp_type;
1184 u32 create_flags;
1185
1186 /*
1187 * Only needed for special QP types, or when using the RW API.
1188 */
1189 u32 port_num;
1190 struct ib_rwq_ind_table *rwq_ind_tbl;
1191 u32 source_qpn;
1192};
1193
1194struct ib_qp_open_attr {
1195 void (*event_handler)(struct ib_event *, void *);
1196 void *qp_context;
1197 u32 qp_num;
1198 enum ib_qp_type qp_type;
1199};
1200
1201enum ib_rnr_timeout {
1202 IB_RNR_TIMER_655_36 = 0,
1203 IB_RNR_TIMER_000_01 = 1,
1204 IB_RNR_TIMER_000_02 = 2,
1205 IB_RNR_TIMER_000_03 = 3,
1206 IB_RNR_TIMER_000_04 = 4,
1207 IB_RNR_TIMER_000_06 = 5,
1208 IB_RNR_TIMER_000_08 = 6,
1209 IB_RNR_TIMER_000_12 = 7,
1210 IB_RNR_TIMER_000_16 = 8,
1211 IB_RNR_TIMER_000_24 = 9,
1212 IB_RNR_TIMER_000_32 = 10,
1213 IB_RNR_TIMER_000_48 = 11,
1214 IB_RNR_TIMER_000_64 = 12,
1215 IB_RNR_TIMER_000_96 = 13,
1216 IB_RNR_TIMER_001_28 = 14,
1217 IB_RNR_TIMER_001_92 = 15,
1218 IB_RNR_TIMER_002_56 = 16,
1219 IB_RNR_TIMER_003_84 = 17,
1220 IB_RNR_TIMER_005_12 = 18,
1221 IB_RNR_TIMER_007_68 = 19,
1222 IB_RNR_TIMER_010_24 = 20,
1223 IB_RNR_TIMER_015_36 = 21,
1224 IB_RNR_TIMER_020_48 = 22,
1225 IB_RNR_TIMER_030_72 = 23,
1226 IB_RNR_TIMER_040_96 = 24,
1227 IB_RNR_TIMER_061_44 = 25,
1228 IB_RNR_TIMER_081_92 = 26,
1229 IB_RNR_TIMER_122_88 = 27,
1230 IB_RNR_TIMER_163_84 = 28,
1231 IB_RNR_TIMER_245_76 = 29,
1232 IB_RNR_TIMER_327_68 = 30,
1233 IB_RNR_TIMER_491_52 = 31
1234};
1235
1236enum ib_qp_attr_mask {
1237 IB_QP_STATE = 1,
1238 IB_QP_CUR_STATE = (1<<1),
1239 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1240 IB_QP_ACCESS_FLAGS = (1<<3),
1241 IB_QP_PKEY_INDEX = (1<<4),
1242 IB_QP_PORT = (1<<5),
1243 IB_QP_QKEY = (1<<6),
1244 IB_QP_AV = (1<<7),
1245 IB_QP_PATH_MTU = (1<<8),
1246 IB_QP_TIMEOUT = (1<<9),
1247 IB_QP_RETRY_CNT = (1<<10),
1248 IB_QP_RNR_RETRY = (1<<11),
1249 IB_QP_RQ_PSN = (1<<12),
1250 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1251 IB_QP_ALT_PATH = (1<<14),
1252 IB_QP_MIN_RNR_TIMER = (1<<15),
1253 IB_QP_SQ_PSN = (1<<16),
1254 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1255 IB_QP_PATH_MIG_STATE = (1<<18),
1256 IB_QP_CAP = (1<<19),
1257 IB_QP_DEST_QPN = (1<<20),
1258 IB_QP_RESERVED1 = (1<<21),
1259 IB_QP_RESERVED2 = (1<<22),
1260 IB_QP_RESERVED3 = (1<<23),
1261 IB_QP_RESERVED4 = (1<<24),
1262 IB_QP_RATE_LIMIT = (1<<25),
1263
1264 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0),
1265};
1266
1267enum ib_qp_state {
1268 IB_QPS_RESET,
1269 IB_QPS_INIT,
1270 IB_QPS_RTR,
1271 IB_QPS_RTS,
1272 IB_QPS_SQD,
1273 IB_QPS_SQE,
1274 IB_QPS_ERR
1275};
1276
1277enum ib_mig_state {
1278 IB_MIG_MIGRATED,
1279 IB_MIG_REARM,
1280 IB_MIG_ARMED
1281};
1282
1283enum ib_mw_type {
1284 IB_MW_TYPE_1 = 1,
1285 IB_MW_TYPE_2 = 2
1286};
1287
1288struct ib_qp_attr {
1289 enum ib_qp_state qp_state;
1290 enum ib_qp_state cur_qp_state;
1291 enum ib_mtu path_mtu;
1292 enum ib_mig_state path_mig_state;
1293 u32 qkey;
1294 u32 rq_psn;
1295 u32 sq_psn;
1296 u32 dest_qp_num;
1297 int qp_access_flags;
1298 struct ib_qp_cap cap;
1299 struct rdma_ah_attr ah_attr;
1300 struct rdma_ah_attr alt_ah_attr;
1301 u16 pkey_index;
1302 u16 alt_pkey_index;
1303 u8 en_sqd_async_notify;
1304 u8 sq_draining;
1305 u8 max_rd_atomic;
1306 u8 max_dest_rd_atomic;
1307 u8 min_rnr_timer;
1308 u32 port_num;
1309 u8 timeout;
1310 u8 retry_cnt;
1311 u8 rnr_retry;
1312 u32 alt_port_num;
1313 u8 alt_timeout;
1314 u32 rate_limit;
1315 struct net_device *xmit_slave;
1316};
1317
1318enum ib_wr_opcode {
1319 /* These are shared with userspace */
1320 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1321 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1322 IB_WR_SEND = IB_UVERBS_WR_SEND,
1323 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1324 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1325 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1326 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1327 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW,
1328 IB_WR_LSO = IB_UVERBS_WR_TSO,
1329 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1330 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1331 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1332 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1333 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1334 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1335 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1336 IB_WR_FLUSH = IB_UVERBS_WR_FLUSH,
1337 IB_WR_ATOMIC_WRITE = IB_UVERBS_WR_ATOMIC_WRITE,
1338
1339 /* These are kernel only and can not be issued by userspace */
1340 IB_WR_REG_MR = 0x20,
1341 IB_WR_REG_MR_INTEGRITY,
1342
1343 /* reserve values for low level drivers' internal use.
1344 * These values will not be used at all in the ib core layer.
1345 */
1346 IB_WR_RESERVED1 = 0xf0,
1347 IB_WR_RESERVED2,
1348 IB_WR_RESERVED3,
1349 IB_WR_RESERVED4,
1350 IB_WR_RESERVED5,
1351 IB_WR_RESERVED6,
1352 IB_WR_RESERVED7,
1353 IB_WR_RESERVED8,
1354 IB_WR_RESERVED9,
1355 IB_WR_RESERVED10,
1356};
1357
1358enum ib_send_flags {
1359 IB_SEND_FENCE = 1,
1360 IB_SEND_SIGNALED = (1<<1),
1361 IB_SEND_SOLICITED = (1<<2),
1362 IB_SEND_INLINE = (1<<3),
1363 IB_SEND_IP_CSUM = (1<<4),
1364
1365 /* reserve bits 26-31 for low level drivers' internal use */
1366 IB_SEND_RESERVED_START = (1 << 26),
1367 IB_SEND_RESERVED_END = (1 << 31),
1368};
1369
1370struct ib_sge {
1371 u64 addr;
1372 u32 length;
1373 u32 lkey;
1374};
1375
1376struct ib_cqe {
1377 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1378};
1379
1380struct ib_send_wr {
1381 struct ib_send_wr *next;
1382 union {
1383 u64 wr_id;
1384 struct ib_cqe *wr_cqe;
1385 };
1386 struct ib_sge *sg_list;
1387 int num_sge;
1388 enum ib_wr_opcode opcode;
1389 int send_flags;
1390 union {
1391 __be32 imm_data;
1392 u32 invalidate_rkey;
1393 } ex;
1394};
1395
1396struct ib_rdma_wr {
1397 struct ib_send_wr wr;
1398 u64 remote_addr;
1399 u32 rkey;
1400};
1401
1402static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
1403{
1404 return container_of(wr, struct ib_rdma_wr, wr);
1405}
1406
1407struct ib_atomic_wr {
1408 struct ib_send_wr wr;
1409 u64 remote_addr;
1410 u64 compare_add;
1411 u64 swap;
1412 u64 compare_add_mask;
1413 u64 swap_mask;
1414 u32 rkey;
1415};
1416
1417static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
1418{
1419 return container_of(wr, struct ib_atomic_wr, wr);
1420}
1421
1422struct ib_ud_wr {
1423 struct ib_send_wr wr;
1424 struct ib_ah *ah;
1425 void *header;
1426 int hlen;
1427 int mss;
1428 u32 remote_qpn;
1429 u32 remote_qkey;
1430 u16 pkey_index; /* valid for GSI only */
1431 u32 port_num; /* valid for DR SMPs on switch only */
1432};
1433
1434static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
1435{
1436 return container_of(wr, struct ib_ud_wr, wr);
1437}
1438
1439struct ib_reg_wr {
1440 struct ib_send_wr wr;
1441 struct ib_mr *mr;
1442 u32 key;
1443 int access;
1444};
1445
1446static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
1447{
1448 return container_of(wr, struct ib_reg_wr, wr);
1449}
1450
1451struct ib_recv_wr {
1452 struct ib_recv_wr *next;
1453 union {
1454 u64 wr_id;
1455 struct ib_cqe *wr_cqe;
1456 };
1457 struct ib_sge *sg_list;
1458 int num_sge;
1459};
1460
1461enum ib_access_flags {
1462 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1463 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1464 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1465 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1466 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1467 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1468 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1469 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1470 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING,
1471 IB_ACCESS_FLUSH_GLOBAL = IB_UVERBS_ACCESS_FLUSH_GLOBAL,
1472 IB_ACCESS_FLUSH_PERSISTENT = IB_UVERBS_ACCESS_FLUSH_PERSISTENT,
1473
1474 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE,
1475 IB_ACCESS_SUPPORTED =
1476 ((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL,
1477};
1478
1479/*
1480 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1481 * are hidden here instead of a uapi header!
1482 */
1483enum ib_mr_rereg_flags {
1484 IB_MR_REREG_TRANS = 1,
1485 IB_MR_REREG_PD = (1<<1),
1486 IB_MR_REREG_ACCESS = (1<<2),
1487 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1488};
1489
1490struct ib_umem;
1491
1492enum rdma_remove_reason {
1493 /*
1494 * Userspace requested uobject deletion or initial try
1495 * to remove uobject via cleanup. Call could fail
1496 */
1497 RDMA_REMOVE_DESTROY,
1498 /* Context deletion. This call should delete the actual object itself */
1499 RDMA_REMOVE_CLOSE,
1500 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1501 RDMA_REMOVE_DRIVER_REMOVE,
1502 /* uobj is being cleaned-up before being committed */
1503 RDMA_REMOVE_ABORT,
1504 /* The driver failed to destroy the uobject and is being disconnected */
1505 RDMA_REMOVE_DRIVER_FAILURE,
1506};
1507
1508struct ib_rdmacg_object {
1509#ifdef CONFIG_CGROUP_RDMA
1510 struct rdma_cgroup *cg; /* owner rdma cgroup */
1511#endif
1512};
1513
1514struct ib_ucontext {
1515 struct ib_device *device;
1516 struct ib_uverbs_file *ufile;
1517
1518 struct ib_rdmacg_object cg_obj;
1519 /*
1520 * Implementation details of the RDMA core, don't use in drivers:
1521 */
1522 struct rdma_restrack_entry res;
1523 struct xarray mmap_xa;
1524};
1525
1526struct ib_uobject {
1527 u64 user_handle; /* handle given to us by userspace */
1528 /* ufile & ucontext owning this object */
1529 struct ib_uverbs_file *ufile;
1530 /* FIXME, save memory: ufile->context == context */
1531 struct ib_ucontext *context; /* associated user context */
1532 void *object; /* containing object */
1533 struct list_head list; /* link to context's list */
1534 struct ib_rdmacg_object cg_obj; /* rdmacg object */
1535 int id; /* index into kernel idr */
1536 struct kref ref;
1537 atomic_t usecnt; /* protects exclusive access */
1538 struct rcu_head rcu; /* kfree_rcu() overhead */
1539
1540 const struct uverbs_api_object *uapi_object;
1541};
1542
1543struct ib_udata {
1544 const void __user *inbuf;
1545 void __user *outbuf;
1546 size_t inlen;
1547 size_t outlen;
1548};
1549
1550struct ib_pd {
1551 u32 local_dma_lkey;
1552 u32 flags;
1553 struct ib_device *device;
1554 struct ib_uobject *uobject;
1555 atomic_t usecnt; /* count all resources */
1556
1557 u32 unsafe_global_rkey;
1558
1559 /*
1560 * Implementation details of the RDMA core, don't use in drivers:
1561 */
1562 struct ib_mr *__internal_mr;
1563 struct rdma_restrack_entry res;
1564};
1565
1566struct ib_xrcd {
1567 struct ib_device *device;
1568 atomic_t usecnt; /* count all exposed resources */
1569 struct inode *inode;
1570 struct rw_semaphore tgt_qps_rwsem;
1571 struct xarray tgt_qps;
1572};
1573
1574struct ib_ah {
1575 struct ib_device *device;
1576 struct ib_pd *pd;
1577 struct ib_uobject *uobject;
1578 const struct ib_gid_attr *sgid_attr;
1579 enum rdma_ah_attr_type type;
1580};
1581
1582typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1583
1584enum ib_poll_context {
1585 IB_POLL_SOFTIRQ, /* poll from softirq context */
1586 IB_POLL_WORKQUEUE, /* poll from workqueue */
1587 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
1588 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE,
1589
1590 IB_POLL_DIRECT, /* caller context, no hw completions */
1591};
1592
1593struct ib_cq {
1594 struct ib_device *device;
1595 struct ib_ucq_object *uobject;
1596 ib_comp_handler comp_handler;
1597 void (*event_handler)(struct ib_event *, void *);
1598 void *cq_context;
1599 int cqe;
1600 unsigned int cqe_used;
1601 atomic_t usecnt; /* count number of work queues */
1602 enum ib_poll_context poll_ctx;
1603 struct ib_wc *wc;
1604 struct list_head pool_entry;
1605 union {
1606 struct irq_poll iop;
1607 struct work_struct work;
1608 };
1609 struct workqueue_struct *comp_wq;
1610 struct dim *dim;
1611
1612 /* updated only by trace points */
1613 ktime_t timestamp;
1614 u8 interrupt:1;
1615 u8 shared:1;
1616 unsigned int comp_vector;
1617
1618 /*
1619 * Implementation details of the RDMA core, don't use in drivers:
1620 */
1621 struct rdma_restrack_entry res;
1622};
1623
1624struct ib_srq {
1625 struct ib_device *device;
1626 struct ib_pd *pd;
1627 struct ib_usrq_object *uobject;
1628 void (*event_handler)(struct ib_event *, void *);
1629 void *srq_context;
1630 enum ib_srq_type srq_type;
1631 atomic_t usecnt;
1632
1633 struct {
1634 struct ib_cq *cq;
1635 union {
1636 struct {
1637 struct ib_xrcd *xrcd;
1638 u32 srq_num;
1639 } xrc;
1640 };
1641 } ext;
1642
1643 /*
1644 * Implementation details of the RDMA core, don't use in drivers:
1645 */
1646 struct rdma_restrack_entry res;
1647};
1648
1649enum ib_raw_packet_caps {
1650 /*
1651 * Strip cvlan from incoming packet and report it in the matching work
1652 * completion is supported.
1653 */
1654 IB_RAW_PACKET_CAP_CVLAN_STRIPPING =
1655 IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING,
1656 /*
1657 * Scatter FCS field of an incoming packet to host memory is supported.
1658 */
1659 IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS,
1660 /* Checksum offloads are supported (for both send and receive). */
1661 IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM,
1662 /*
1663 * When a packet is received for an RQ with no receive WQEs, the
1664 * packet processing is delayed.
1665 */
1666 IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP,
1667};
1668
1669enum ib_wq_type {
1670 IB_WQT_RQ = IB_UVERBS_WQT_RQ,
1671};
1672
1673enum ib_wq_state {
1674 IB_WQS_RESET,
1675 IB_WQS_RDY,
1676 IB_WQS_ERR
1677};
1678
1679struct ib_wq {
1680 struct ib_device *device;
1681 struct ib_uwq_object *uobject;
1682 void *wq_context;
1683 void (*event_handler)(struct ib_event *, void *);
1684 struct ib_pd *pd;
1685 struct ib_cq *cq;
1686 u32 wq_num;
1687 enum ib_wq_state state;
1688 enum ib_wq_type wq_type;
1689 atomic_t usecnt;
1690};
1691
1692enum ib_wq_flags {
1693 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING,
1694 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS,
1695 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP,
1696 IB_WQ_FLAGS_PCI_WRITE_END_PADDING =
1697 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING,
1698};
1699
1700struct ib_wq_init_attr {
1701 void *wq_context;
1702 enum ib_wq_type wq_type;
1703 u32 max_wr;
1704 u32 max_sge;
1705 struct ib_cq *cq;
1706 void (*event_handler)(struct ib_event *, void *);
1707 u32 create_flags; /* Use enum ib_wq_flags */
1708};
1709
1710enum ib_wq_attr_mask {
1711 IB_WQ_STATE = 1 << 0,
1712 IB_WQ_CUR_STATE = 1 << 1,
1713 IB_WQ_FLAGS = 1 << 2,
1714};
1715
1716struct ib_wq_attr {
1717 enum ib_wq_state wq_state;
1718 enum ib_wq_state curr_wq_state;
1719 u32 flags; /* Use enum ib_wq_flags */
1720 u32 flags_mask; /* Use enum ib_wq_flags */
1721};
1722
1723struct ib_rwq_ind_table {
1724 struct ib_device *device;
1725 struct ib_uobject *uobject;
1726 atomic_t usecnt;
1727 u32 ind_tbl_num;
1728 u32 log_ind_tbl_size;
1729 struct ib_wq **ind_tbl;
1730};
1731
1732struct ib_rwq_ind_table_init_attr {
1733 u32 log_ind_tbl_size;
1734 /* Each entry is a pointer to Receive Work Queue */
1735 struct ib_wq **ind_tbl;
1736};
1737
1738enum port_pkey_state {
1739 IB_PORT_PKEY_NOT_VALID = 0,
1740 IB_PORT_PKEY_VALID = 1,
1741 IB_PORT_PKEY_LISTED = 2,
1742};
1743
1744struct ib_qp_security;
1745
1746struct ib_port_pkey {
1747 enum port_pkey_state state;
1748 u16 pkey_index;
1749 u32 port_num;
1750 struct list_head qp_list;
1751 struct list_head to_error_list;
1752 struct ib_qp_security *sec;
1753};
1754
1755struct ib_ports_pkeys {
1756 struct ib_port_pkey main;
1757 struct ib_port_pkey alt;
1758};
1759
1760struct ib_qp_security {
1761 struct ib_qp *qp;
1762 struct ib_device *dev;
1763 /* Hold this mutex when changing port and pkey settings. */
1764 struct mutex mutex;
1765 struct ib_ports_pkeys *ports_pkeys;
1766 /* A list of all open shared QP handles. Required to enforce security
1767 * properly for all users of a shared QP.
1768 */
1769 struct list_head shared_qp_list;
1770 void *security;
1771 bool destroying;
1772 atomic_t error_list_count;
1773 struct completion error_complete;
1774 int error_comps_pending;
1775};
1776
1777/*
1778 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1779 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1780 */
1781struct ib_qp {
1782 struct ib_device *device;
1783 struct ib_pd *pd;
1784 struct ib_cq *send_cq;
1785 struct ib_cq *recv_cq;
1786 spinlock_t mr_lock;
1787 int mrs_used;
1788 struct list_head rdma_mrs;
1789 struct list_head sig_mrs;
1790 struct ib_srq *srq;
1791 struct completion srq_completion;
1792 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1793 struct list_head xrcd_list;
1794
1795 /* count times opened, mcast attaches, flow attaches */
1796 atomic_t usecnt;
1797 struct list_head open_list;
1798 struct ib_qp *real_qp;
1799 struct ib_uqp_object *uobject;
1800 void (*event_handler)(struct ib_event *, void *);
1801 void (*registered_event_handler)(struct ib_event *, void *);
1802 void *qp_context;
1803 /* sgid_attrs associated with the AV's */
1804 const struct ib_gid_attr *av_sgid_attr;
1805 const struct ib_gid_attr *alt_path_sgid_attr;
1806 u32 qp_num;
1807 u32 max_write_sge;
1808 u32 max_read_sge;
1809 enum ib_qp_type qp_type;
1810 struct ib_rwq_ind_table *rwq_ind_tbl;
1811 struct ib_qp_security *qp_sec;
1812 u32 port;
1813
1814 bool integrity_en;
1815 /*
1816 * Implementation details of the RDMA core, don't use in drivers:
1817 */
1818 struct rdma_restrack_entry res;
1819
1820 /* The counter the qp is bind to */
1821 struct rdma_counter *counter;
1822};
1823
1824struct ib_dm {
1825 struct ib_device *device;
1826 u32 length;
1827 u32 flags;
1828 struct ib_uobject *uobject;
1829 atomic_t usecnt;
1830};
1831
1832struct ib_mr {
1833 struct ib_device *device;
1834 struct ib_pd *pd;
1835 u32 lkey;
1836 u32 rkey;
1837 u64 iova;
1838 u64 length;
1839 unsigned int page_size;
1840 enum ib_mr_type type;
1841 bool need_inval;
1842 union {
1843 struct ib_uobject *uobject; /* user */
1844 struct list_head qp_entry; /* FR */
1845 };
1846
1847 struct ib_dm *dm;
1848 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
1849 /*
1850 * Implementation details of the RDMA core, don't use in drivers:
1851 */
1852 struct rdma_restrack_entry res;
1853};
1854
1855struct ib_mw {
1856 struct ib_device *device;
1857 struct ib_pd *pd;
1858 struct ib_uobject *uobject;
1859 u32 rkey;
1860 enum ib_mw_type type;
1861};
1862
1863/* Supported steering options */
1864enum ib_flow_attr_type {
1865 /* steering according to rule specifications */
1866 IB_FLOW_ATTR_NORMAL = 0x0,
1867 /* default unicast and multicast rule -
1868 * receive all Eth traffic which isn't steered to any QP
1869 */
1870 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1871 /* default multicast rule -
1872 * receive all Eth multicast traffic which isn't steered to any QP
1873 */
1874 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1875 /* sniffer rule - receive all port traffic */
1876 IB_FLOW_ATTR_SNIFFER = 0x3
1877};
1878
1879/* Supported steering header types */
1880enum ib_flow_spec_type {
1881 /* L2 headers*/
1882 IB_FLOW_SPEC_ETH = 0x20,
1883 IB_FLOW_SPEC_IB = 0x22,
1884 /* L3 header*/
1885 IB_FLOW_SPEC_IPV4 = 0x30,
1886 IB_FLOW_SPEC_IPV6 = 0x31,
1887 IB_FLOW_SPEC_ESP = 0x34,
1888 /* L4 headers*/
1889 IB_FLOW_SPEC_TCP = 0x40,
1890 IB_FLOW_SPEC_UDP = 0x41,
1891 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
1892 IB_FLOW_SPEC_GRE = 0x51,
1893 IB_FLOW_SPEC_MPLS = 0x60,
1894 IB_FLOW_SPEC_INNER = 0x100,
1895 /* Actions */
1896 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
1897 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
1898 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
1899 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
1900};
1901#define IB_FLOW_SPEC_LAYER_MASK 0xF0
1902#define IB_FLOW_SPEC_SUPPORT_LAYERS 10
1903
1904enum ib_flow_flags {
1905 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1906 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1907 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
1908};
1909
1910struct ib_flow_eth_filter {
1911 u8 dst_mac[6];
1912 u8 src_mac[6];
1913 __be16 ether_type;
1914 __be16 vlan_tag;
1915};
1916
1917struct ib_flow_spec_eth {
1918 u32 type;
1919 u16 size;
1920 struct ib_flow_eth_filter val;
1921 struct ib_flow_eth_filter mask;
1922};
1923
1924struct ib_flow_ib_filter {
1925 __be16 dlid;
1926 __u8 sl;
1927};
1928
1929struct ib_flow_spec_ib {
1930 u32 type;
1931 u16 size;
1932 struct ib_flow_ib_filter val;
1933 struct ib_flow_ib_filter mask;
1934};
1935
1936/* IPv4 header flags */
1937enum ib_ipv4_flags {
1938 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1939 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1940 last have this flag set */
1941};
1942
1943struct ib_flow_ipv4_filter {
1944 __be32 src_ip;
1945 __be32 dst_ip;
1946 u8 proto;
1947 u8 tos;
1948 u8 ttl;
1949 u8 flags;
1950};
1951
1952struct ib_flow_spec_ipv4 {
1953 u32 type;
1954 u16 size;
1955 struct ib_flow_ipv4_filter val;
1956 struct ib_flow_ipv4_filter mask;
1957};
1958
1959struct ib_flow_ipv6_filter {
1960 u8 src_ip[16];
1961 u8 dst_ip[16];
1962 __be32 flow_label;
1963 u8 next_hdr;
1964 u8 traffic_class;
1965 u8 hop_limit;
1966} __packed;
1967
1968struct ib_flow_spec_ipv6 {
1969 u32 type;
1970 u16 size;
1971 struct ib_flow_ipv6_filter val;
1972 struct ib_flow_ipv6_filter mask;
1973};
1974
1975struct ib_flow_tcp_udp_filter {
1976 __be16 dst_port;
1977 __be16 src_port;
1978};
1979
1980struct ib_flow_spec_tcp_udp {
1981 u32 type;
1982 u16 size;
1983 struct ib_flow_tcp_udp_filter val;
1984 struct ib_flow_tcp_udp_filter mask;
1985};
1986
1987struct ib_flow_tunnel_filter {
1988 __be32 tunnel_id;
1989};
1990
1991/* ib_flow_spec_tunnel describes the Vxlan tunnel
1992 * the tunnel_id from val has the vni value
1993 */
1994struct ib_flow_spec_tunnel {
1995 u32 type;
1996 u16 size;
1997 struct ib_flow_tunnel_filter val;
1998 struct ib_flow_tunnel_filter mask;
1999};
2000
2001struct ib_flow_esp_filter {
2002 __be32 spi;
2003 __be32 seq;
2004};
2005
2006struct ib_flow_spec_esp {
2007 u32 type;
2008 u16 size;
2009 struct ib_flow_esp_filter val;
2010 struct ib_flow_esp_filter mask;
2011};
2012
2013struct ib_flow_gre_filter {
2014 __be16 c_ks_res0_ver;
2015 __be16 protocol;
2016 __be32 key;
2017};
2018
2019struct ib_flow_spec_gre {
2020 u32 type;
2021 u16 size;
2022 struct ib_flow_gre_filter val;
2023 struct ib_flow_gre_filter mask;
2024};
2025
2026struct ib_flow_mpls_filter {
2027 __be32 tag;
2028};
2029
2030struct ib_flow_spec_mpls {
2031 u32 type;
2032 u16 size;
2033 struct ib_flow_mpls_filter val;
2034 struct ib_flow_mpls_filter mask;
2035};
2036
2037struct ib_flow_spec_action_tag {
2038 enum ib_flow_spec_type type;
2039 u16 size;
2040 u32 tag_id;
2041};
2042
2043struct ib_flow_spec_action_drop {
2044 enum ib_flow_spec_type type;
2045 u16 size;
2046};
2047
2048struct ib_flow_spec_action_handle {
2049 enum ib_flow_spec_type type;
2050 u16 size;
2051 struct ib_flow_action *act;
2052};
2053
2054enum ib_counters_description {
2055 IB_COUNTER_PACKETS,
2056 IB_COUNTER_BYTES,
2057};
2058
2059struct ib_flow_spec_action_count {
2060 enum ib_flow_spec_type type;
2061 u16 size;
2062 struct ib_counters *counters;
2063};
2064
2065union ib_flow_spec {
2066 struct {
2067 u32 type;
2068 u16 size;
2069 };
2070 struct ib_flow_spec_eth eth;
2071 struct ib_flow_spec_ib ib;
2072 struct ib_flow_spec_ipv4 ipv4;
2073 struct ib_flow_spec_tcp_udp tcp_udp;
2074 struct ib_flow_spec_ipv6 ipv6;
2075 struct ib_flow_spec_tunnel tunnel;
2076 struct ib_flow_spec_esp esp;
2077 struct ib_flow_spec_gre gre;
2078 struct ib_flow_spec_mpls mpls;
2079 struct ib_flow_spec_action_tag flow_tag;
2080 struct ib_flow_spec_action_drop drop;
2081 struct ib_flow_spec_action_handle action;
2082 struct ib_flow_spec_action_count flow_count;
2083};
2084
2085struct ib_flow_attr {
2086 enum ib_flow_attr_type type;
2087 u16 size;
2088 u16 priority;
2089 u32 flags;
2090 u8 num_of_specs;
2091 u32 port;
2092 union ib_flow_spec flows[];
2093};
2094
2095struct ib_flow {
2096 struct ib_qp *qp;
2097 struct ib_device *device;
2098 struct ib_uobject *uobject;
2099};
2100
2101enum ib_flow_action_type {
2102 IB_FLOW_ACTION_UNSPECIFIED,
2103 IB_FLOW_ACTION_ESP = 1,
2104};
2105
2106struct ib_flow_action_attrs_esp_keymats {
2107 enum ib_uverbs_flow_action_esp_keymat protocol;
2108 union {
2109 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2110 } keymat;
2111};
2112
2113struct ib_flow_action_attrs_esp_replays {
2114 enum ib_uverbs_flow_action_esp_replay protocol;
2115 union {
2116 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2117 } replay;
2118};
2119
2120enum ib_flow_action_attrs_esp_flags {
2121 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2122 * This is done in order to share the same flags between user-space and
2123 * kernel and spare an unnecessary translation.
2124 */
2125
2126 /* Kernel flags */
2127 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
2128 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2129};
2130
2131struct ib_flow_spec_list {
2132 struct ib_flow_spec_list *next;
2133 union ib_flow_spec spec;
2134};
2135
2136struct ib_flow_action_attrs_esp {
2137 struct ib_flow_action_attrs_esp_keymats *keymat;
2138 struct ib_flow_action_attrs_esp_replays *replay;
2139 struct ib_flow_spec_list *encap;
2140 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2141 * Value of 0 is a valid value.
2142 */
2143 u32 esn;
2144 u32 spi;
2145 u32 seq;
2146 u32 tfc_pad;
2147 /* Use enum ib_flow_action_attrs_esp_flags */
2148 u64 flags;
2149 u64 hard_limit_pkts;
2150};
2151
2152struct ib_flow_action {
2153 struct ib_device *device;
2154 struct ib_uobject *uobject;
2155 enum ib_flow_action_type type;
2156 atomic_t usecnt;
2157};
2158
2159struct ib_mad;
2160
2161enum ib_process_mad_flags {
2162 IB_MAD_IGNORE_MKEY = 1,
2163 IB_MAD_IGNORE_BKEY = 2,
2164 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2165};
2166
2167enum ib_mad_result {
2168 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2169 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2170 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2171 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2172};
2173
2174struct ib_port_cache {
2175 u64 subnet_prefix;
2176 struct ib_pkey_cache *pkey;
2177 struct ib_gid_table *gid;
2178 u8 lmc;
2179 enum ib_port_state port_state;
2180};
2181
2182struct ib_port_immutable {
2183 int pkey_tbl_len;
2184 int gid_tbl_len;
2185 u32 core_cap_flags;
2186 u32 max_mad_size;
2187};
2188
2189struct ib_port_data {
2190 struct ib_device *ib_dev;
2191
2192 struct ib_port_immutable immutable;
2193
2194 spinlock_t pkey_list_lock;
2195
2196 spinlock_t netdev_lock;
2197
2198 struct list_head pkey_list;
2199
2200 struct ib_port_cache cache;
2201
2202 struct net_device __rcu *netdev;
2203 netdevice_tracker netdev_tracker;
2204 struct hlist_node ndev_hash_link;
2205 struct rdma_port_counter port_counter;
2206 struct ib_port *sysfs;
2207};
2208
2209/* rdma netdev type - specifies protocol type */
2210enum rdma_netdev_t {
2211 RDMA_NETDEV_OPA_VNIC,
2212 RDMA_NETDEV_IPOIB,
2213};
2214
2215/**
2216 * struct rdma_netdev - rdma netdev
2217 * For cases where netstack interfacing is required.
2218 */
2219struct rdma_netdev {
2220 void *clnt_priv;
2221 struct ib_device *hca;
2222 u32 port_num;
2223 int mtu;
2224
2225 /*
2226 * cleanup function must be specified.
2227 * FIXME: This is only used for OPA_VNIC and that usage should be
2228 * removed too.
2229 */
2230 void (*free_rdma_netdev)(struct net_device *netdev);
2231
2232 /* control functions */
2233 void (*set_id)(struct net_device *netdev, int id);
2234 /* send packet */
2235 int (*send)(struct net_device *dev, struct sk_buff *skb,
2236 struct ib_ah *address, u32 dqpn);
2237 /* multicast */
2238 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2239 union ib_gid *gid, u16 mlid,
2240 int set_qkey, u32 qkey);
2241 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2242 union ib_gid *gid, u16 mlid);
2243 /* timeout */
2244 void (*tx_timeout)(struct net_device *dev, unsigned int txqueue);
2245};
2246
2247struct rdma_netdev_alloc_params {
2248 size_t sizeof_priv;
2249 unsigned int txqs;
2250 unsigned int rxqs;
2251 void *param;
2252
2253 int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num,
2254 struct net_device *netdev, void *param);
2255};
2256
2257struct ib_odp_counters {
2258 atomic64_t faults;
2259 atomic64_t invalidations;
2260 atomic64_t prefetch;
2261};
2262
2263struct ib_counters {
2264 struct ib_device *device;
2265 struct ib_uobject *uobject;
2266 /* num of objects attached */
2267 atomic_t usecnt;
2268};
2269
2270struct ib_counters_read_attr {
2271 u64 *counters_buff;
2272 u32 ncounters;
2273 u32 flags; /* use enum ib_read_counters_flags */
2274};
2275
2276struct uverbs_attr_bundle;
2277struct iw_cm_id;
2278struct iw_cm_conn_param;
2279
2280#define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2281 .size_##ib_struct = \
2282 (sizeof(struct drv_struct) + \
2283 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2284 BUILD_BUG_ON_ZERO( \
2285 !__same_type(((struct drv_struct *)NULL)->member, \
2286 struct ib_struct)))
2287
2288#define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2289 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2290 gfp, false))
2291
2292#define rdma_zalloc_drv_obj_numa(ib_dev, ib_type) \
2293 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \
2294 GFP_KERNEL, true))
2295
2296#define rdma_zalloc_drv_obj(ib_dev, ib_type) \
2297 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
2298
2299#define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2300
2301struct rdma_user_mmap_entry {
2302 struct kref ref;
2303 struct ib_ucontext *ucontext;
2304 unsigned long start_pgoff;
2305 size_t npages;
2306 bool driver_removed;
2307};
2308
2309/* Return the offset (in bytes) the user should pass to libc's mmap() */
2310static inline u64
2311rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2312{
2313 return (u64)entry->start_pgoff << PAGE_SHIFT;
2314}
2315
2316/**
2317 * struct ib_device_ops - InfiniBand device operations
2318 * This structure defines all the InfiniBand device operations, providers will
2319 * need to define the supported operations, otherwise they will be set to null.
2320 */
2321struct ib_device_ops {
2322 struct module *owner;
2323 enum rdma_driver_id driver_id;
2324 u32 uverbs_abi_ver;
2325 unsigned int uverbs_no_driver_id_binding:1;
2326
2327 /*
2328 * NOTE: New drivers should not make use of device_group; instead new
2329 * device parameter should be exposed via netlink command. This
2330 * mechanism exists only for existing drivers.
2331 */
2332 const struct attribute_group *device_group;
2333 const struct attribute_group **port_groups;
2334
2335 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2336 const struct ib_send_wr **bad_send_wr);
2337 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2338 const struct ib_recv_wr **bad_recv_wr);
2339 void (*drain_rq)(struct ib_qp *qp);
2340 void (*drain_sq)(struct ib_qp *qp);
2341 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2342 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2343 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2344 int (*post_srq_recv)(struct ib_srq *srq,
2345 const struct ib_recv_wr *recv_wr,
2346 const struct ib_recv_wr **bad_recv_wr);
2347 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2348 u32 port_num, const struct ib_wc *in_wc,
2349 const struct ib_grh *in_grh,
2350 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2351 size_t *out_mad_size, u16 *out_mad_pkey_index);
2352 int (*query_device)(struct ib_device *device,
2353 struct ib_device_attr *device_attr,
2354 struct ib_udata *udata);
2355 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2356 struct ib_device_modify *device_modify);
2357 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2358 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2359 int comp_vector);
2360 int (*query_port)(struct ib_device *device, u32 port_num,
2361 struct ib_port_attr *port_attr);
2362 int (*modify_port)(struct ib_device *device, u32 port_num,
2363 int port_modify_mask,
2364 struct ib_port_modify *port_modify);
2365 /**
2366 * The following mandatory functions are used only at device
2367 * registration. Keep functions such as these at the end of this
2368 * structure to avoid cache line misses when accessing struct ib_device
2369 * in fast paths.
2370 */
2371 int (*get_port_immutable)(struct ib_device *device, u32 port_num,
2372 struct ib_port_immutable *immutable);
2373 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2374 u32 port_num);
2375 /**
2376 * When calling get_netdev, the HW vendor's driver should return the
2377 * net device of device @device at port @port_num or NULL if such
2378 * a net device doesn't exist. The vendor driver should call dev_hold
2379 * on this net device. The HW vendor's device driver must guarantee
2380 * that this function returns NULL before the net device has finished
2381 * NETDEV_UNREGISTER state.
2382 */
2383 struct net_device *(*get_netdev)(struct ib_device *device,
2384 u32 port_num);
2385 /**
2386 * rdma netdev operation
2387 *
2388 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2389 * must return -EOPNOTSUPP if it doesn't support the specified type.
2390 */
2391 struct net_device *(*alloc_rdma_netdev)(
2392 struct ib_device *device, u32 port_num, enum rdma_netdev_t type,
2393 const char *name, unsigned char name_assign_type,
2394 void (*setup)(struct net_device *));
2395
2396 int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num,
2397 enum rdma_netdev_t type,
2398 struct rdma_netdev_alloc_params *params);
2399 /**
2400 * query_gid should be return GID value for @device, when @port_num
2401 * link layer is either IB or iWarp. It is no-op if @port_num port
2402 * is RoCE link layer.
2403 */
2404 int (*query_gid)(struct ib_device *device, u32 port_num, int index,
2405 union ib_gid *gid);
2406 /**
2407 * When calling add_gid, the HW vendor's driver should add the gid
2408 * of device of port at gid index available at @attr. Meta-info of
2409 * that gid (for example, the network device related to this gid) is
2410 * available at @attr. @context allows the HW vendor driver to store
2411 * extra information together with a GID entry. The HW vendor driver may
2412 * allocate memory to contain this information and store it in @context
2413 * when a new GID entry is written to. Params are consistent until the
2414 * next call of add_gid or delete_gid. The function should return 0 on
2415 * success or error otherwise. The function could be called
2416 * concurrently for different ports. This function is only called when
2417 * roce_gid_table is used.
2418 */
2419 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2420 /**
2421 * When calling del_gid, the HW vendor's driver should delete the
2422 * gid of device @device at gid index gid_index of port port_num
2423 * available in @attr.
2424 * Upon the deletion of a GID entry, the HW vendor must free any
2425 * allocated memory. The caller will clear @context afterwards.
2426 * This function is only called when roce_gid_table is used.
2427 */
2428 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2429 int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index,
2430 u16 *pkey);
2431 int (*alloc_ucontext)(struct ib_ucontext *context,
2432 struct ib_udata *udata);
2433 void (*dealloc_ucontext)(struct ib_ucontext *context);
2434 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
2435 /**
2436 * This will be called once refcount of an entry in mmap_xa reaches
2437 * zero. The type of the memory that was mapped may differ between
2438 * entries and is opaque to the rdma_user_mmap interface.
2439 * Therefore needs to be implemented by the driver in mmap_free.
2440 */
2441 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
2442 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2443 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2444 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
2445 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2446 struct ib_udata *udata);
2447 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr,
2448 struct ib_udata *udata);
2449 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2450 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2451 int (*destroy_ah)(struct ib_ah *ah, u32 flags);
2452 int (*create_srq)(struct ib_srq *srq,
2453 struct ib_srq_init_attr *srq_init_attr,
2454 struct ib_udata *udata);
2455 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2456 enum ib_srq_attr_mask srq_attr_mask,
2457 struct ib_udata *udata);
2458 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
2459 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
2460 int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr,
2461 struct ib_udata *udata);
2462 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2463 int qp_attr_mask, struct ib_udata *udata);
2464 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2465 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
2466 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
2467 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2468 struct uverbs_attr_bundle *attrs);
2469 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2470 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
2471 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2472 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2473 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2474 u64 virt_addr, int mr_access_flags,
2475 struct ib_udata *udata);
2476 struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset,
2477 u64 length, u64 virt_addr, int fd,
2478 int mr_access_flags,
2479 struct uverbs_attr_bundle *attrs);
2480 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start,
2481 u64 length, u64 virt_addr,
2482 int mr_access_flags, struct ib_pd *pd,
2483 struct ib_udata *udata);
2484 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
2485 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
2486 u32 max_num_sg);
2487 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2488 u32 max_num_data_sg,
2489 u32 max_num_meta_sg);
2490 int (*advise_mr)(struct ib_pd *pd,
2491 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2492 struct ib_sge *sg_list, u32 num_sge,
2493 struct uverbs_attr_bundle *attrs);
2494
2495 /*
2496 * Kernel users should universally support relaxed ordering (RO), as
2497 * they are designed to read data only after observing the CQE and use
2498 * the DMA API correctly.
2499 *
2500 * Some drivers implicitly enable RO if platform supports it.
2501 */
2502 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2503 unsigned int *sg_offset);
2504 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2505 struct ib_mr_status *mr_status);
2506 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata);
2507 int (*dealloc_mw)(struct ib_mw *mw);
2508 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2509 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2510 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2511 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
2512 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2513 struct ib_flow_attr *flow_attr,
2514 struct ib_udata *udata);
2515 int (*destroy_flow)(struct ib_flow *flow_id);
2516 int (*destroy_flow_action)(struct ib_flow_action *action);
2517 int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port,
2518 int state);
2519 int (*get_vf_config)(struct ib_device *device, int vf, u32 port,
2520 struct ifla_vf_info *ivf);
2521 int (*get_vf_stats)(struct ib_device *device, int vf, u32 port,
2522 struct ifla_vf_stats *stats);
2523 int (*get_vf_guid)(struct ib_device *device, int vf, u32 port,
2524 struct ifla_vf_guid *node_guid,
2525 struct ifla_vf_guid *port_guid);
2526 int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid,
2527 int type);
2528 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2529 struct ib_wq_init_attr *init_attr,
2530 struct ib_udata *udata);
2531 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
2532 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2533 u32 wq_attr_mask, struct ib_udata *udata);
2534 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table,
2535 struct ib_rwq_ind_table_init_attr *init_attr,
2536 struct ib_udata *udata);
2537 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2538 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2539 struct ib_ucontext *context,
2540 struct ib_dm_alloc_attr *attr,
2541 struct uverbs_attr_bundle *attrs);
2542 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
2543 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2544 struct ib_dm_mr_attr *attr,
2545 struct uverbs_attr_bundle *attrs);
2546 int (*create_counters)(struct ib_counters *counters,
2547 struct uverbs_attr_bundle *attrs);
2548 int (*destroy_counters)(struct ib_counters *counters);
2549 int (*read_counters)(struct ib_counters *counters,
2550 struct ib_counters_read_attr *counters_read_attr,
2551 struct uverbs_attr_bundle *attrs);
2552 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2553 int data_sg_nents, unsigned int *data_sg_offset,
2554 struct scatterlist *meta_sg, int meta_sg_nents,
2555 unsigned int *meta_sg_offset);
2556
2557 /**
2558 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and
2559 * fill in the driver initialized data. The struct is kfree()'ed by
2560 * the sysfs core when the device is removed. A lifespan of -1 in the
2561 * return struct tells the core to set a default lifespan.
2562 */
2563 struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device);
2564 struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device,
2565 u32 port_num);
2566 /**
2567 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2568 * @index - The index in the value array we wish to have updated, or
2569 * num_counters if we want all stats updated
2570 * Return codes -
2571 * < 0 - Error, no counters updated
2572 * index - Updated the single counter pointed to by index
2573 * num_counters - Updated all counters (will reset the timestamp
2574 * and prevent further calls for lifespan milliseconds)
2575 * Drivers are allowed to update all counters in leiu of just the
2576 * one given in index at their option
2577 */
2578 int (*get_hw_stats)(struct ib_device *device,
2579 struct rdma_hw_stats *stats, u32 port, int index);
2580
2581 /**
2582 * modify_hw_stat - Modify the counter configuration
2583 * @enable: true/false when enable/disable a counter
2584 * Return codes - 0 on success or error code otherwise.
2585 */
2586 int (*modify_hw_stat)(struct ib_device *device, u32 port,
2587 unsigned int counter_index, bool enable);
2588 /**
2589 * Allows rdma drivers to add their own restrack attributes.
2590 */
2591 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2592 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr);
2593 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq);
2594 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq);
2595 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp);
2596 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp);
2597 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id);
2598 int (*fill_res_srq_entry)(struct sk_buff *msg, struct ib_srq *ib_srq);
2599 int (*fill_res_srq_entry_raw)(struct sk_buff *msg, struct ib_srq *ib_srq);
2600
2601 /* Device lifecycle callbacks */
2602 /*
2603 * Called after the device becomes registered, before clients are
2604 * attached
2605 */
2606 int (*enable_driver)(struct ib_device *dev);
2607 /*
2608 * This is called as part of ib_dealloc_device().
2609 */
2610 void (*dealloc_driver)(struct ib_device *dev);
2611
2612 /* iWarp CM callbacks */
2613 void (*iw_add_ref)(struct ib_qp *qp);
2614 void (*iw_rem_ref)(struct ib_qp *qp);
2615 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2616 int (*iw_connect)(struct iw_cm_id *cm_id,
2617 struct iw_cm_conn_param *conn_param);
2618 int (*iw_accept)(struct iw_cm_id *cm_id,
2619 struct iw_cm_conn_param *conn_param);
2620 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2621 u8 pdata_len);
2622 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2623 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
2624 /**
2625 * counter_bind_qp - Bind a QP to a counter.
2626 * @counter - The counter to be bound. If counter->id is zero then
2627 * the driver needs to allocate a new counter and set counter->id
2628 */
2629 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2630 /**
2631 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2632 * counter and bind it onto the default one
2633 */
2634 int (*counter_unbind_qp)(struct ib_qp *qp);
2635 /**
2636 * counter_dealloc -De-allocate the hw counter
2637 */
2638 int (*counter_dealloc)(struct rdma_counter *counter);
2639 /**
2640 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2641 * the driver initialized data.
2642 */
2643 struct rdma_hw_stats *(*counter_alloc_stats)(
2644 struct rdma_counter *counter);
2645 /**
2646 * counter_update_stats - Query the stats value of this counter
2647 */
2648 int (*counter_update_stats)(struct rdma_counter *counter);
2649
2650 /**
2651 * Allows rdma drivers to add their own restrack attributes
2652 * dumped via 'rdma stat' iproute2 command.
2653 */
2654 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr);
2655
2656 /* query driver for its ucontext properties */
2657 int (*query_ucontext)(struct ib_ucontext *context,
2658 struct uverbs_attr_bundle *attrs);
2659
2660 /*
2661 * Provide NUMA node. This API exists for rdmavt/hfi1 only.
2662 * Everyone else relies on Linux memory management model.
2663 */
2664 int (*get_numa_node)(struct ib_device *dev);
2665
2666 /**
2667 * add_sub_dev - Add a sub IB device
2668 */
2669 struct ib_device *(*add_sub_dev)(struct ib_device *parent,
2670 enum rdma_nl_dev_type type,
2671 const char *name);
2672
2673 /**
2674 * del_sub_dev - Delete a sub IB device
2675 */
2676 void (*del_sub_dev)(struct ib_device *sub_dev);
2677
2678 /**
2679 * ufile_cleanup - Attempt to cleanup ubojects HW resources inside
2680 * the ufile.
2681 */
2682 void (*ufile_hw_cleanup)(struct ib_uverbs_file *ufile);
2683
2684 DECLARE_RDMA_OBJ_SIZE(ib_ah);
2685 DECLARE_RDMA_OBJ_SIZE(ib_counters);
2686 DECLARE_RDMA_OBJ_SIZE(ib_cq);
2687 DECLARE_RDMA_OBJ_SIZE(ib_mw);
2688 DECLARE_RDMA_OBJ_SIZE(ib_pd);
2689 DECLARE_RDMA_OBJ_SIZE(ib_qp);
2690 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table);
2691 DECLARE_RDMA_OBJ_SIZE(ib_srq);
2692 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
2693 DECLARE_RDMA_OBJ_SIZE(ib_xrcd);
2694};
2695
2696struct ib_core_device {
2697 /* device must be the first element in structure until,
2698 * union of ib_core_device and device exists in ib_device.
2699 */
2700 struct device dev;
2701 possible_net_t rdma_net;
2702 struct kobject *ports_kobj;
2703 struct list_head port_list;
2704 struct ib_device *owner; /* reach back to owner ib_device */
2705};
2706
2707struct rdma_restrack_root;
2708struct ib_device {
2709 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2710 struct device *dma_device;
2711 struct ib_device_ops ops;
2712 char name[IB_DEVICE_NAME_MAX];
2713 struct rcu_head rcu_head;
2714
2715 struct list_head event_handler_list;
2716 /* Protects event_handler_list */
2717 struct rw_semaphore event_handler_rwsem;
2718
2719 /* Protects QP's event_handler calls and open_qp list */
2720 spinlock_t qp_open_list_lock;
2721
2722 struct rw_semaphore client_data_rwsem;
2723 struct xarray client_data;
2724 struct mutex unregistration_lock;
2725
2726 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2727 rwlock_t cache_lock;
2728 /**
2729 * port_data is indexed by port number
2730 */
2731 struct ib_port_data *port_data;
2732
2733 int num_comp_vectors;
2734
2735 union {
2736 struct device dev;
2737 struct ib_core_device coredev;
2738 };
2739
2740 /* First group is for device attributes,
2741 * Second group is for driver provided attributes (optional).
2742 * Third group is for the hw_stats
2743 * It is a NULL terminated array.
2744 */
2745 const struct attribute_group *groups[4];
2746
2747 u64 uverbs_cmd_mask;
2748
2749 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2750 __be64 node_guid;
2751 u32 local_dma_lkey;
2752 u16 is_switch:1;
2753 /* Indicates kernel verbs support, should not be used in drivers */
2754 u16 kverbs_provider:1;
2755 /* CQ adaptive moderation (RDMA DIM) */
2756 u16 use_cq_dim:1;
2757 u8 node_type;
2758 u32 phys_port_cnt;
2759 struct ib_device_attr attrs;
2760 struct hw_stats_device_data *hw_stats_data;
2761
2762#ifdef CONFIG_CGROUP_RDMA
2763 struct rdmacg_device cg_device;
2764#endif
2765
2766 u32 index;
2767
2768 spinlock_t cq_pools_lock;
2769 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1];
2770
2771 struct rdma_restrack_root *res;
2772
2773 const struct uapi_definition *driver_def;
2774
2775 /*
2776 * Positive refcount indicates that the device is currently
2777 * registered and cannot be unregistered.
2778 */
2779 refcount_t refcount;
2780 struct completion unreg_completion;
2781 struct work_struct unregistration_work;
2782
2783 const struct rdma_link_ops *link_ops;
2784
2785 /* Protects compat_devs xarray modifications */
2786 struct mutex compat_devs_mutex;
2787 /* Maintains compat devices for each net namespace */
2788 struct xarray compat_devs;
2789
2790 /* Used by iWarp CM */
2791 char iw_ifname[IFNAMSIZ];
2792 u32 iw_driver_flags;
2793 u32 lag_flags;
2794
2795 /* A parent device has a list of sub-devices */
2796 struct mutex subdev_lock;
2797 struct list_head subdev_list_head;
2798
2799 /* A sub device has a type and a parent */
2800 enum rdma_nl_dev_type type;
2801 struct ib_device *parent;
2802 struct list_head subdev_list;
2803
2804 enum rdma_nl_name_assign_type name_assign_type;
2805};
2806
2807static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size,
2808 gfp_t gfp, bool is_numa_aware)
2809{
2810 if (is_numa_aware && dev->ops.get_numa_node)
2811 return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev));
2812
2813 return kzalloc(size, gfp);
2814}
2815
2816struct ib_client_nl_info;
2817struct ib_client {
2818 const char *name;
2819 int (*add)(struct ib_device *ibdev);
2820 void (*remove)(struct ib_device *, void *client_data);
2821 void (*rename)(struct ib_device *dev, void *client_data);
2822 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2823 struct ib_client_nl_info *res);
2824 int (*get_global_nl_info)(struct ib_client_nl_info *res);
2825
2826 /* Returns the net_dev belonging to this ib_client and matching the
2827 * given parameters.
2828 * @dev: An RDMA device that the net_dev use for communication.
2829 * @port: A physical port number on the RDMA device.
2830 * @pkey: P_Key that the net_dev uses if applicable.
2831 * @gid: A GID that the net_dev uses to communicate.
2832 * @addr: An IP address the net_dev is configured with.
2833 * @client_data: The device's client data set by ib_set_client_data().
2834 *
2835 * An ib_client that implements a net_dev on top of RDMA devices
2836 * (such as IP over IB) should implement this callback, allowing the
2837 * rdma_cm module to find the right net_dev for a given request.
2838 *
2839 * The caller is responsible for calling dev_put on the returned
2840 * netdev. */
2841 struct net_device *(*get_net_dev_by_params)(
2842 struct ib_device *dev,
2843 u32 port,
2844 u16 pkey,
2845 const union ib_gid *gid,
2846 const struct sockaddr *addr,
2847 void *client_data);
2848
2849 refcount_t uses;
2850 struct completion uses_zero;
2851 u32 client_id;
2852
2853 /* kverbs are not required by the client */
2854 u8 no_kverbs_req:1;
2855};
2856
2857/*
2858 * IB block DMA iterator
2859 *
2860 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2861 * to a HW supported page size.
2862 */
2863struct ib_block_iter {
2864 /* internal states */
2865 struct scatterlist *__sg; /* sg holding the current aligned block */
2866 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2867 size_t __sg_numblocks; /* ib_umem_num_dma_blocks() */
2868 unsigned int __sg_nents; /* number of SG entries */
2869 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2870 unsigned int __pg_bit; /* alignment of current block */
2871};
2872
2873struct ib_device *_ib_alloc_device(size_t size);
2874#define ib_alloc_device(drv_struct, member) \
2875 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2876 BUILD_BUG_ON_ZERO(offsetof( \
2877 struct drv_struct, member))), \
2878 struct drv_struct, member)
2879
2880void ib_dealloc_device(struct ib_device *device);
2881
2882void ib_get_device_fw_str(struct ib_device *device, char *str);
2883
2884int ib_register_device(struct ib_device *device, const char *name,
2885 struct device *dma_device);
2886void ib_unregister_device(struct ib_device *device);
2887void ib_unregister_driver(enum rdma_driver_id driver_id);
2888void ib_unregister_device_and_put(struct ib_device *device);
2889void ib_unregister_device_queued(struct ib_device *ib_dev);
2890
2891int ib_register_client (struct ib_client *client);
2892void ib_unregister_client(struct ib_client *client);
2893
2894void __rdma_block_iter_start(struct ib_block_iter *biter,
2895 struct scatterlist *sglist,
2896 unsigned int nents,
2897 unsigned long pgsz);
2898bool __rdma_block_iter_next(struct ib_block_iter *biter);
2899
2900/**
2901 * rdma_block_iter_dma_address - get the aligned dma address of the current
2902 * block held by the block iterator.
2903 * @biter: block iterator holding the memory block
2904 */
2905static inline dma_addr_t
2906rdma_block_iter_dma_address(struct ib_block_iter *biter)
2907{
2908 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2909}
2910
2911/**
2912 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2913 * @sglist: sglist to iterate over
2914 * @biter: block iterator holding the memory block
2915 * @nents: maximum number of sg entries to iterate over
2916 * @pgsz: best HW supported page size to use
2917 *
2918 * Callers may use rdma_block_iter_dma_address() to get each
2919 * blocks aligned DMA address.
2920 */
2921#define rdma_for_each_block(sglist, biter, nents, pgsz) \
2922 for (__rdma_block_iter_start(biter, sglist, nents, \
2923 pgsz); \
2924 __rdma_block_iter_next(biter);)
2925
2926/**
2927 * ib_get_client_data - Get IB client context
2928 * @device:Device to get context for
2929 * @client:Client to get context for
2930 *
2931 * ib_get_client_data() returns the client context data set with
2932 * ib_set_client_data(). This can only be called while the client is
2933 * registered to the device, once the ib_client remove() callback returns this
2934 * cannot be called.
2935 */
2936static inline void *ib_get_client_data(struct ib_device *device,
2937 struct ib_client *client)
2938{
2939 return xa_load(&device->client_data, client->client_id);
2940}
2941void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2942 void *data);
2943void ib_set_device_ops(struct ib_device *device,
2944 const struct ib_device_ops *ops);
2945
2946int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
2947 unsigned long pfn, unsigned long size, pgprot_t prot,
2948 struct rdma_user_mmap_entry *entry);
2949int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2950 struct rdma_user_mmap_entry *entry,
2951 size_t length);
2952int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2953 struct rdma_user_mmap_entry *entry,
2954 size_t length, u32 min_pgoff,
2955 u32 max_pgoff);
2956
2957#if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS)
2958void rdma_user_mmap_disassociate(struct ib_device *device);
2959#else
2960static inline void rdma_user_mmap_disassociate(struct ib_device *device)
2961{
2962}
2963#endif
2964
2965static inline int
2966rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext,
2967 struct rdma_user_mmap_entry *entry,
2968 size_t length, u32 pgoff)
2969{
2970 return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff,
2971 pgoff);
2972}
2973
2974struct rdma_user_mmap_entry *
2975rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2976 unsigned long pgoff);
2977struct rdma_user_mmap_entry *
2978rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2979 struct vm_area_struct *vma);
2980void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2981
2982void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
2983
2984static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2985{
2986 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2987}
2988
2989static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2990{
2991 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2992}
2993
2994static inline bool ib_is_buffer_cleared(const void __user *p,
2995 size_t len)
2996{
2997 bool ret;
2998 u8 *buf;
2999
3000 if (len > USHRT_MAX)
3001 return false;
3002
3003 buf = memdup_user(p, len);
3004 if (IS_ERR(buf))
3005 return false;
3006
3007 ret = !memchr_inv(buf, 0, len);
3008 kfree(buf);
3009 return ret;
3010}
3011
3012static inline bool ib_is_udata_cleared(struct ib_udata *udata,
3013 size_t offset,
3014 size_t len)
3015{
3016 return ib_is_buffer_cleared(udata->inbuf + offset, len);
3017}
3018
3019/**
3020 * ib_modify_qp_is_ok - Check that the supplied attribute mask
3021 * contains all required attributes and no attributes not allowed for
3022 * the given QP state transition.
3023 * @cur_state: Current QP state
3024 * @next_state: Next QP state
3025 * @type: QP type
3026 * @mask: Mask of supplied QP attributes
3027 *
3028 * This function is a helper function that a low-level driver's
3029 * modify_qp method can use to validate the consumer's input. It
3030 * checks that cur_state and next_state are valid QP states, that a
3031 * transition from cur_state to next_state is allowed by the IB spec,
3032 * and that the attribute mask supplied is allowed for the transition.
3033 */
3034bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
3035 enum ib_qp_type type, enum ib_qp_attr_mask mask);
3036
3037void ib_register_event_handler(struct ib_event_handler *event_handler);
3038void ib_unregister_event_handler(struct ib_event_handler *event_handler);
3039void ib_dispatch_event(const struct ib_event *event);
3040
3041int ib_query_port(struct ib_device *device,
3042 u32 port_num, struct ib_port_attr *port_attr);
3043
3044enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
3045 u32 port_num);
3046
3047/**
3048 * rdma_cap_ib_switch - Check if the device is IB switch
3049 * @device: Device to check
3050 *
3051 * Device driver is responsible for setting is_switch bit on
3052 * in ib_device structure at init time.
3053 *
3054 * Return: true if the device is IB switch.
3055 */
3056static inline bool rdma_cap_ib_switch(const struct ib_device *device)
3057{
3058 return device->is_switch;
3059}
3060
3061/**
3062 * rdma_start_port - Return the first valid port number for the device
3063 * specified
3064 *
3065 * @device: Device to be checked
3066 *
3067 * Return start port number
3068 */
3069static inline u32 rdma_start_port(const struct ib_device *device)
3070{
3071 return rdma_cap_ib_switch(device) ? 0 : 1;
3072}
3073
3074/**
3075 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
3076 * @device - The struct ib_device * to iterate over
3077 * @iter - The unsigned int to store the port number
3078 */
3079#define rdma_for_each_port(device, iter) \
3080 for (iter = rdma_start_port(device + \
3081 BUILD_BUG_ON_ZERO(!__same_type(u32, \
3082 iter))); \
3083 iter <= rdma_end_port(device); iter++)
3084
3085/**
3086 * rdma_end_port - Return the last valid port number for the device
3087 * specified
3088 *
3089 * @device: Device to be checked
3090 *
3091 * Return last port number
3092 */
3093static inline u32 rdma_end_port(const struct ib_device *device)
3094{
3095 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
3096}
3097
3098static inline int rdma_is_port_valid(const struct ib_device *device,
3099 unsigned int port)
3100{
3101 return (port >= rdma_start_port(device) &&
3102 port <= rdma_end_port(device));
3103}
3104
3105static inline bool rdma_is_grh_required(const struct ib_device *device,
3106 u32 port_num)
3107{
3108 return device->port_data[port_num].immutable.core_cap_flags &
3109 RDMA_CORE_PORT_IB_GRH_REQUIRED;
3110}
3111
3112static inline bool rdma_protocol_ib(const struct ib_device *device,
3113 u32 port_num)
3114{
3115 return device->port_data[port_num].immutable.core_cap_flags &
3116 RDMA_CORE_CAP_PROT_IB;
3117}
3118
3119static inline bool rdma_protocol_roce(const struct ib_device *device,
3120 u32 port_num)
3121{
3122 return device->port_data[port_num].immutable.core_cap_flags &
3123 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
3124}
3125
3126static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device,
3127 u32 port_num)
3128{
3129 return device->port_data[port_num].immutable.core_cap_flags &
3130 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
3131}
3132
3133static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device,
3134 u32 port_num)
3135{
3136 return device->port_data[port_num].immutable.core_cap_flags &
3137 RDMA_CORE_CAP_PROT_ROCE;
3138}
3139
3140static inline bool rdma_protocol_iwarp(const struct ib_device *device,
3141 u32 port_num)
3142{
3143 return device->port_data[port_num].immutable.core_cap_flags &
3144 RDMA_CORE_CAP_PROT_IWARP;
3145}
3146
3147static inline bool rdma_ib_or_roce(const struct ib_device *device,
3148 u32 port_num)
3149{
3150 return rdma_protocol_ib(device, port_num) ||
3151 rdma_protocol_roce(device, port_num);
3152}
3153
3154static inline bool rdma_protocol_raw_packet(const struct ib_device *device,
3155 u32 port_num)
3156{
3157 return device->port_data[port_num].immutable.core_cap_flags &
3158 RDMA_CORE_CAP_PROT_RAW_PACKET;
3159}
3160
3161static inline bool rdma_protocol_usnic(const struct ib_device *device,
3162 u32 port_num)
3163{
3164 return device->port_data[port_num].immutable.core_cap_flags &
3165 RDMA_CORE_CAP_PROT_USNIC;
3166}
3167
3168/**
3169 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
3170 * Management Datagrams.
3171 * @device: Device to check
3172 * @port_num: Port number to check
3173 *
3174 * Management Datagrams (MAD) are a required part of the InfiniBand
3175 * specification and are supported on all InfiniBand devices. A slightly
3176 * extended version are also supported on OPA interfaces.
3177 *
3178 * Return: true if the port supports sending/receiving of MAD packets.
3179 */
3180static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num)
3181{
3182 return device->port_data[port_num].immutable.core_cap_flags &
3183 RDMA_CORE_CAP_IB_MAD;
3184}
3185
3186/**
3187 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3188 * Management Datagrams.
3189 * @device: Device to check
3190 * @port_num: Port number to check
3191 *
3192 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3193 * datagrams with their own versions. These OPA MADs share many but not all of
3194 * the characteristics of InfiniBand MADs.
3195 *
3196 * OPA MADs differ in the following ways:
3197 *
3198 * 1) MADs are variable size up to 2K
3199 * IBTA defined MADs remain fixed at 256 bytes
3200 * 2) OPA SMPs must carry valid PKeys
3201 * 3) OPA SMP packets are a different format
3202 *
3203 * Return: true if the port supports OPA MAD packet formats.
3204 */
3205static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num)
3206{
3207 return device->port_data[port_num].immutable.core_cap_flags &
3208 RDMA_CORE_CAP_OPA_MAD;
3209}
3210
3211/**
3212 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3213 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3214 * @device: Device to check
3215 * @port_num: Port number to check
3216 *
3217 * Each InfiniBand node is required to provide a Subnet Management Agent
3218 * that the subnet manager can access. Prior to the fabric being fully
3219 * configured by the subnet manager, the SMA is accessed via a well known
3220 * interface called the Subnet Management Interface (SMI). This interface
3221 * uses directed route packets to communicate with the SM to get around the
3222 * chicken and egg problem of the SM needing to know what's on the fabric
3223 * in order to configure the fabric, and needing to configure the fabric in
3224 * order to send packets to the devices on the fabric. These directed
3225 * route packets do not need the fabric fully configured in order to reach
3226 * their destination. The SMI is the only method allowed to send
3227 * directed route packets on an InfiniBand fabric.
3228 *
3229 * Return: true if the port provides an SMI.
3230 */
3231static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num)
3232{
3233 return device->port_data[port_num].immutable.core_cap_flags &
3234 RDMA_CORE_CAP_IB_SMI;
3235}
3236
3237/**
3238 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3239 * Communication Manager.
3240 * @device: Device to check
3241 * @port_num: Port number to check
3242 *
3243 * The InfiniBand Communication Manager is one of many pre-defined General
3244 * Service Agents (GSA) that are accessed via the General Service
3245 * Interface (GSI). It's role is to facilitate establishment of connections
3246 * between nodes as well as other management related tasks for established
3247 * connections.
3248 *
3249 * Return: true if the port supports an IB CM (this does not guarantee that
3250 * a CM is actually running however).
3251 */
3252static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num)
3253{
3254 return device->port_data[port_num].immutable.core_cap_flags &
3255 RDMA_CORE_CAP_IB_CM;
3256}
3257
3258/**
3259 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3260 * Communication Manager.
3261 * @device: Device to check
3262 * @port_num: Port number to check
3263 *
3264 * Similar to above, but specific to iWARP connections which have a different
3265 * managment protocol than InfiniBand.
3266 *
3267 * Return: true if the port supports an iWARP CM (this does not guarantee that
3268 * a CM is actually running however).
3269 */
3270static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num)
3271{
3272 return device->port_data[port_num].immutable.core_cap_flags &
3273 RDMA_CORE_CAP_IW_CM;
3274}
3275
3276/**
3277 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3278 * Subnet Administration.
3279 * @device: Device to check
3280 * @port_num: Port number to check
3281 *
3282 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3283 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3284 * fabrics, devices should resolve routes to other hosts by contacting the
3285 * SA to query the proper route.
3286 *
3287 * Return: true if the port should act as a client to the fabric Subnet
3288 * Administration interface. This does not imply that the SA service is
3289 * running locally.
3290 */
3291static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num)
3292{
3293 return device->port_data[port_num].immutable.core_cap_flags &
3294 RDMA_CORE_CAP_IB_SA;
3295}
3296
3297/**
3298 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3299 * Multicast.
3300 * @device: Device to check
3301 * @port_num: Port number to check
3302 *
3303 * InfiniBand multicast registration is more complex than normal IPv4 or
3304 * IPv6 multicast registration. Each Host Channel Adapter must register
3305 * with the Subnet Manager when it wishes to join a multicast group. It
3306 * should do so only once regardless of how many queue pairs it subscribes
3307 * to this group. And it should leave the group only after all queue pairs
3308 * attached to the group have been detached.
3309 *
3310 * Return: true if the port must undertake the additional adminstrative
3311 * overhead of registering/unregistering with the SM and tracking of the
3312 * total number of queue pairs attached to the multicast group.
3313 */
3314static inline bool rdma_cap_ib_mcast(const struct ib_device *device,
3315 u32 port_num)
3316{
3317 return rdma_cap_ib_sa(device, port_num);
3318}
3319
3320/**
3321 * rdma_cap_af_ib - Check if the port of device has the capability
3322 * Native Infiniband Address.
3323 * @device: Device to check
3324 * @port_num: Port number to check
3325 *
3326 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3327 * GID. RoCE uses a different mechanism, but still generates a GID via
3328 * a prescribed mechanism and port specific data.
3329 *
3330 * Return: true if the port uses a GID address to identify devices on the
3331 * network.
3332 */
3333static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num)
3334{
3335 return device->port_data[port_num].immutable.core_cap_flags &
3336 RDMA_CORE_CAP_AF_IB;
3337}
3338
3339/**
3340 * rdma_cap_eth_ah - Check if the port of device has the capability
3341 * Ethernet Address Handle.
3342 * @device: Device to check
3343 * @port_num: Port number to check
3344 *
3345 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3346 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3347 * port. Normally, packet headers are generated by the sending host
3348 * adapter, but when sending connectionless datagrams, we must manually
3349 * inject the proper headers for the fabric we are communicating over.
3350 *
3351 * Return: true if we are running as a RoCE port and must force the
3352 * addition of a Global Route Header built from our Ethernet Address
3353 * Handle into our header list for connectionless packets.
3354 */
3355static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num)
3356{
3357 return device->port_data[port_num].immutable.core_cap_flags &
3358 RDMA_CORE_CAP_ETH_AH;
3359}
3360
3361/**
3362 * rdma_cap_opa_ah - Check if the port of device supports
3363 * OPA Address handles
3364 * @device: Device to check
3365 * @port_num: Port number to check
3366 *
3367 * Return: true if we are running on an OPA device which supports
3368 * the extended OPA addressing.
3369 */
3370static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num)
3371{
3372 return (device->port_data[port_num].immutable.core_cap_flags &
3373 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3374}
3375
3376/**
3377 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3378 *
3379 * @device: Device
3380 * @port_num: Port number
3381 *
3382 * This MAD size includes the MAD headers and MAD payload. No other headers
3383 * are included.
3384 *
3385 * Return the max MAD size required by the Port. Will return 0 if the port
3386 * does not support MADs
3387 */
3388static inline size_t rdma_max_mad_size(const struct ib_device *device,
3389 u32 port_num)
3390{
3391 return device->port_data[port_num].immutable.max_mad_size;
3392}
3393
3394/**
3395 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3396 * @device: Device to check
3397 * @port_num: Port number to check
3398 *
3399 * RoCE GID table mechanism manages the various GIDs for a device.
3400 *
3401 * NOTE: if allocating the port's GID table has failed, this call will still
3402 * return true, but any RoCE GID table API will fail.
3403 *
3404 * Return: true if the port uses RoCE GID table mechanism in order to manage
3405 * its GIDs.
3406 */
3407static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3408 u32 port_num)
3409{
3410 return rdma_protocol_roce(device, port_num) &&
3411 device->ops.add_gid && device->ops.del_gid;
3412}
3413
3414/*
3415 * Check if the device supports READ W/ INVALIDATE.
3416 */
3417static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3418{
3419 /*
3420 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3421 * has support for it yet.
3422 */
3423 return rdma_protocol_iwarp(dev, port_num);
3424}
3425
3426/**
3427 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not.
3428 * @device: Device
3429 * @port_num: 1 based Port number
3430 *
3431 * Return true if port is an Intel OPA port , false if not
3432 */
3433static inline bool rdma_core_cap_opa_port(struct ib_device *device,
3434 u32 port_num)
3435{
3436 return (device->port_data[port_num].immutable.core_cap_flags &
3437 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA;
3438}
3439
3440/**
3441 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value.
3442 * @device: Device
3443 * @port_num: Port number
3444 * @mtu: enum value of MTU
3445 *
3446 * Return the MTU size supported by the port as an integer value. Will return
3447 * -1 if enum value of mtu is not supported.
3448 */
3449static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port,
3450 int mtu)
3451{
3452 if (rdma_core_cap_opa_port(device, port))
3453 return opa_mtu_enum_to_int((enum opa_mtu)mtu);
3454 else
3455 return ib_mtu_enum_to_int((enum ib_mtu)mtu);
3456}
3457
3458/**
3459 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute.
3460 * @device: Device
3461 * @port_num: Port number
3462 * @attr: port attribute
3463 *
3464 * Return the MTU size supported by the port as an integer value.
3465 */
3466static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port,
3467 struct ib_port_attr *attr)
3468{
3469 if (rdma_core_cap_opa_port(device, port))
3470 return attr->phys_mtu;
3471 else
3472 return ib_mtu_enum_to_int(attr->max_mtu);
3473}
3474
3475int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port,
3476 int state);
3477int ib_get_vf_config(struct ib_device *device, int vf, u32 port,
3478 struct ifla_vf_info *info);
3479int ib_get_vf_stats(struct ib_device *device, int vf, u32 port,
3480 struct ifla_vf_stats *stats);
3481int ib_get_vf_guid(struct ib_device *device, int vf, u32 port,
3482 struct ifla_vf_guid *node_guid,
3483 struct ifla_vf_guid *port_guid);
3484int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid,
3485 int type);
3486
3487int ib_query_pkey(struct ib_device *device,
3488 u32 port_num, u16 index, u16 *pkey);
3489
3490int ib_modify_device(struct ib_device *device,
3491 int device_modify_mask,
3492 struct ib_device_modify *device_modify);
3493
3494int ib_modify_port(struct ib_device *device,
3495 u32 port_num, int port_modify_mask,
3496 struct ib_port_modify *port_modify);
3497
3498int ib_find_gid(struct ib_device *device, union ib_gid *gid,
3499 u32 *port_num, u16 *index);
3500
3501int ib_find_pkey(struct ib_device *device,
3502 u32 port_num, u16 pkey, u16 *index);
3503
3504enum ib_pd_flags {
3505 /*
3506 * Create a memory registration for all memory in the system and place
3507 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3508 * ULPs to avoid the overhead of dynamic MRs.
3509 *
3510 * This flag is generally considered unsafe and must only be used in
3511 * extremly trusted environments. Every use of it will log a warning
3512 * in the kernel log.
3513 */
3514 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3515};
3516
3517struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3518 const char *caller);
3519
3520/**
3521 * ib_alloc_pd - Allocates an unused protection domain.
3522 * @device: The device on which to allocate the protection domain.
3523 * @flags: protection domain flags
3524 *
3525 * A protection domain object provides an association between QPs, shared
3526 * receive queues, address handles, memory regions, and memory windows.
3527 *
3528 * Every PD has a local_dma_lkey which can be used as the lkey value for local
3529 * memory operations.
3530 */
3531#define ib_alloc_pd(device, flags) \
3532 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
3533
3534int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3535
3536/**
3537 * ib_dealloc_pd - Deallocate kernel PD
3538 * @pd: The protection domain
3539 *
3540 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3541 */
3542static inline void ib_dealloc_pd(struct ib_pd *pd)
3543{
3544 int ret = ib_dealloc_pd_user(pd, NULL);
3545
3546 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail");
3547}
3548
3549enum rdma_create_ah_flags {
3550 /* In a sleepable context */
3551 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3552};
3553
3554/**
3555 * rdma_create_ah - Creates an address handle for the given address vector.
3556 * @pd: The protection domain associated with the address handle.
3557 * @ah_attr: The attributes of the address vector.
3558 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
3559 *
3560 * The address handle is used to reference a local or global destination
3561 * in all UD QP post sends.
3562 */
3563struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3564 u32 flags);
3565
3566/**
3567 * rdma_create_user_ah - Creates an address handle for the given address vector.
3568 * It resolves destination mac address for ah attribute of RoCE type.
3569 * @pd: The protection domain associated with the address handle.
3570 * @ah_attr: The attributes of the address vector.
3571 * @udata: pointer to user's input output buffer information need by
3572 * provider driver.
3573 *
3574 * It returns 0 on success and returns appropriate error code on error.
3575 * The address handle is used to reference a local or global destination
3576 * in all UD QP post sends.
3577 */
3578struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3579 struct rdma_ah_attr *ah_attr,
3580 struct ib_udata *udata);
3581/**
3582 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3583 * work completion.
3584 * @hdr: the L3 header to parse
3585 * @net_type: type of header to parse
3586 * @sgid: place to store source gid
3587 * @dgid: place to store destination gid
3588 */
3589int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3590 enum rdma_network_type net_type,
3591 union ib_gid *sgid, union ib_gid *dgid);
3592
3593/**
3594 * ib_get_rdma_header_version - Get the header version
3595 * @hdr: the L3 header to parse
3596 */
3597int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3598
3599/**
3600 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
3601 * work completion.
3602 * @device: Device on which the received message arrived.
3603 * @port_num: Port on which the received message arrived.
3604 * @wc: Work completion associated with the received message.
3605 * @grh: References the received global route header. This parameter is
3606 * ignored unless the work completion indicates that the GRH is valid.
3607 * @ah_attr: Returned attributes that can be used when creating an address
3608 * handle for replying to the message.
3609 * When ib_init_ah_attr_from_wc() returns success,
3610 * (a) for IB link layer it optionally contains a reference to SGID attribute
3611 * when GRH is present for IB link layer.
3612 * (b) for RoCE link layer it contains a reference to SGID attribute.
3613 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3614 * attributes which are initialized using ib_init_ah_attr_from_wc().
3615 *
3616 */
3617int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num,
3618 const struct ib_wc *wc, const struct ib_grh *grh,
3619 struct rdma_ah_attr *ah_attr);
3620
3621/**
3622 * ib_create_ah_from_wc - Creates an address handle associated with the
3623 * sender of the specified work completion.
3624 * @pd: The protection domain associated with the address handle.
3625 * @wc: Work completion information associated with a received message.
3626 * @grh: References the received global route header. This parameter is
3627 * ignored unless the work completion indicates that the GRH is valid.
3628 * @port_num: The outbound port number to associate with the address.
3629 *
3630 * The address handle is used to reference a local or global destination
3631 * in all UD QP post sends.
3632 */
3633struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3634 const struct ib_grh *grh, u32 port_num);
3635
3636/**
3637 * rdma_modify_ah - Modifies the address vector associated with an address
3638 * handle.
3639 * @ah: The address handle to modify.
3640 * @ah_attr: The new address vector attributes to associate with the
3641 * address handle.
3642 */
3643int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3644
3645/**
3646 * rdma_query_ah - Queries the address vector associated with an address
3647 * handle.
3648 * @ah: The address handle to query.
3649 * @ah_attr: The address vector attributes associated with the address
3650 * handle.
3651 */
3652int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
3653
3654enum rdma_destroy_ah_flags {
3655 /* In a sleepable context */
3656 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3657};
3658
3659/**
3660 * rdma_destroy_ah_user - Destroys an address handle.
3661 * @ah: The address handle to destroy.
3662 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3663 * @udata: Valid user data or NULL for kernel objects
3664 */
3665int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3666
3667/**
3668 * rdma_destroy_ah - Destroys an kernel address handle.
3669 * @ah: The address handle to destroy.
3670 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3671 *
3672 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3673 */
3674static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3675{
3676 int ret = rdma_destroy_ah_user(ah, flags, NULL);
3677
3678 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail");
3679}
3680
3681struct ib_srq *ib_create_srq_user(struct ib_pd *pd,
3682 struct ib_srq_init_attr *srq_init_attr,
3683 struct ib_usrq_object *uobject,
3684 struct ib_udata *udata);
3685static inline struct ib_srq *
3686ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr)
3687{
3688 if (!pd->device->ops.create_srq)
3689 return ERR_PTR(-EOPNOTSUPP);
3690
3691 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL);
3692}
3693
3694/**
3695 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3696 * @srq: The SRQ to modify.
3697 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3698 * the current values of selected SRQ attributes are returned.
3699 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3700 * are being modified.
3701 *
3702 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3703 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3704 * the number of receives queued drops below the limit.
3705 */
3706int ib_modify_srq(struct ib_srq *srq,
3707 struct ib_srq_attr *srq_attr,
3708 enum ib_srq_attr_mask srq_attr_mask);
3709
3710/**
3711 * ib_query_srq - Returns the attribute list and current values for the
3712 * specified SRQ.
3713 * @srq: The SRQ to query.
3714 * @srq_attr: The attributes of the specified SRQ.
3715 */
3716int ib_query_srq(struct ib_srq *srq,
3717 struct ib_srq_attr *srq_attr);
3718
3719/**
3720 * ib_destroy_srq_user - Destroys the specified SRQ.
3721 * @srq: The SRQ to destroy.
3722 * @udata: Valid user data or NULL for kernel objects
3723 */
3724int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3725
3726/**
3727 * ib_destroy_srq - Destroys the specified kernel SRQ.
3728 * @srq: The SRQ to destroy.
3729 *
3730 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
3731 */
3732static inline void ib_destroy_srq(struct ib_srq *srq)
3733{
3734 int ret = ib_destroy_srq_user(srq, NULL);
3735
3736 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail");
3737}
3738
3739/**
3740 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3741 * @srq: The SRQ to post the work request on.
3742 * @recv_wr: A list of work requests to post on the receive queue.
3743 * @bad_recv_wr: On an immediate failure, this parameter will reference
3744 * the work request that failed to be posted on the QP.
3745 */
3746static inline int ib_post_srq_recv(struct ib_srq *srq,
3747 const struct ib_recv_wr *recv_wr,
3748 const struct ib_recv_wr **bad_recv_wr)
3749{
3750 const struct ib_recv_wr *dummy;
3751
3752 return srq->device->ops.post_srq_recv(srq, recv_wr,
3753 bad_recv_wr ? : &dummy);
3754}
3755
3756struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
3757 struct ib_qp_init_attr *qp_init_attr,
3758 const char *caller);
3759/**
3760 * ib_create_qp - Creates a kernel QP associated with the specific protection
3761 * domain.
3762 * @pd: The protection domain associated with the QP.
3763 * @init_attr: A list of initial attributes required to create the
3764 * QP. If QP creation succeeds, then the attributes are updated to
3765 * the actual capabilities of the created QP.
3766 */
3767static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3768 struct ib_qp_init_attr *init_attr)
3769{
3770 return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
3771}
3772
3773/**
3774 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3775 * @qp: The QP to modify.
3776 * @attr: On input, specifies the QP attributes to modify. On output,
3777 * the current values of selected QP attributes are returned.
3778 * @attr_mask: A bit-mask used to specify which attributes of the QP
3779 * are being modified.
3780 * @udata: pointer to user's input output buffer information
3781 * are being modified.
3782 * It returns 0 on success and returns appropriate error code on error.
3783 */
3784int ib_modify_qp_with_udata(struct ib_qp *qp,
3785 struct ib_qp_attr *attr,
3786 int attr_mask,
3787 struct ib_udata *udata);
3788
3789/**
3790 * ib_modify_qp - Modifies the attributes for the specified QP and then
3791 * transitions the QP to the given state.
3792 * @qp: The QP to modify.
3793 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3794 * the current values of selected QP attributes are returned.
3795 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3796 * are being modified.
3797 */
3798int ib_modify_qp(struct ib_qp *qp,
3799 struct ib_qp_attr *qp_attr,
3800 int qp_attr_mask);
3801
3802/**
3803 * ib_query_qp - Returns the attribute list and current values for the
3804 * specified QP.
3805 * @qp: The QP to query.
3806 * @qp_attr: The attributes of the specified QP.
3807 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3808 * @qp_init_attr: Additional attributes of the selected QP.
3809 *
3810 * The qp_attr_mask may be used to limit the query to gathering only the
3811 * selected attributes.
3812 */
3813int ib_query_qp(struct ib_qp *qp,
3814 struct ib_qp_attr *qp_attr,
3815 int qp_attr_mask,
3816 struct ib_qp_init_attr *qp_init_attr);
3817
3818/**
3819 * ib_destroy_qp - Destroys the specified QP.
3820 * @qp: The QP to destroy.
3821 * @udata: Valid udata or NULL for kernel objects
3822 */
3823int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3824
3825/**
3826 * ib_destroy_qp - Destroys the specified kernel QP.
3827 * @qp: The QP to destroy.
3828 *
3829 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3830 */
3831static inline int ib_destroy_qp(struct ib_qp *qp)
3832{
3833 return ib_destroy_qp_user(qp, NULL);
3834}
3835
3836/**
3837 * ib_open_qp - Obtain a reference to an existing sharable QP.
3838 * @xrcd - XRC domain
3839 * @qp_open_attr: Attributes identifying the QP to open.
3840 *
3841 * Returns a reference to a sharable QP.
3842 */
3843struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3844 struct ib_qp_open_attr *qp_open_attr);
3845
3846/**
3847 * ib_close_qp - Release an external reference to a QP.
3848 * @qp: The QP handle to release
3849 *
3850 * The opened QP handle is released by the caller. The underlying
3851 * shared QP is not destroyed until all internal references are released.
3852 */
3853int ib_close_qp(struct ib_qp *qp);
3854
3855/**
3856 * ib_post_send - Posts a list of work requests to the send queue of
3857 * the specified QP.
3858 * @qp: The QP to post the work request on.
3859 * @send_wr: A list of work requests to post on the send queue.
3860 * @bad_send_wr: On an immediate failure, this parameter will reference
3861 * the work request that failed to be posted on the QP.
3862 *
3863 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3864 * error is returned, the QP state shall not be affected,
3865 * ib_post_send() will return an immediate error after queueing any
3866 * earlier work requests in the list.
3867 */
3868static inline int ib_post_send(struct ib_qp *qp,
3869 const struct ib_send_wr *send_wr,
3870 const struct ib_send_wr **bad_send_wr)
3871{
3872 const struct ib_send_wr *dummy;
3873
3874 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
3875}
3876
3877/**
3878 * ib_post_recv - Posts a list of work requests to the receive queue of
3879 * the specified QP.
3880 * @qp: The QP to post the work request on.
3881 * @recv_wr: A list of work requests to post on the receive queue.
3882 * @bad_recv_wr: On an immediate failure, this parameter will reference
3883 * the work request that failed to be posted on the QP.
3884 */
3885static inline int ib_post_recv(struct ib_qp *qp,
3886 const struct ib_recv_wr *recv_wr,
3887 const struct ib_recv_wr **bad_recv_wr)
3888{
3889 const struct ib_recv_wr *dummy;
3890
3891 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
3892}
3893
3894struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe,
3895 int comp_vector, enum ib_poll_context poll_ctx,
3896 const char *caller);
3897static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3898 int nr_cqe, int comp_vector,
3899 enum ib_poll_context poll_ctx)
3900{
3901 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx,
3902 KBUILD_MODNAME);
3903}
3904
3905struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3906 int nr_cqe, enum ib_poll_context poll_ctx,
3907 const char *caller);
3908
3909/**
3910 * ib_alloc_cq_any: Allocate kernel CQ
3911 * @dev: The IB device
3912 * @private: Private data attached to the CQE
3913 * @nr_cqe: Number of CQEs in the CQ
3914 * @poll_ctx: Context used for polling the CQ
3915 */
3916static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3917 void *private, int nr_cqe,
3918 enum ib_poll_context poll_ctx)
3919{
3920 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3921 KBUILD_MODNAME);
3922}
3923
3924void ib_free_cq(struct ib_cq *cq);
3925int ib_process_cq_direct(struct ib_cq *cq, int budget);
3926
3927/**
3928 * ib_create_cq - Creates a CQ on the specified device.
3929 * @device: The device on which to create the CQ.
3930 * @comp_handler: A user-specified callback that is invoked when a
3931 * completion event occurs on the CQ.
3932 * @event_handler: A user-specified callback that is invoked when an
3933 * asynchronous event not associated with a completion occurs on the CQ.
3934 * @cq_context: Context associated with the CQ returned to the user via
3935 * the associated completion and event handlers.
3936 * @cq_attr: The attributes the CQ should be created upon.
3937 *
3938 * Users can examine the cq structure to determine the actual CQ size.
3939 */
3940struct ib_cq *__ib_create_cq(struct ib_device *device,
3941 ib_comp_handler comp_handler,
3942 void (*event_handler)(struct ib_event *, void *),
3943 void *cq_context,
3944 const struct ib_cq_init_attr *cq_attr,
3945 const char *caller);
3946#define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3947 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
3948
3949/**
3950 * ib_resize_cq - Modifies the capacity of the CQ.
3951 * @cq: The CQ to resize.
3952 * @cqe: The minimum size of the CQ.
3953 *
3954 * Users can examine the cq structure to determine the actual CQ size.
3955 */
3956int ib_resize_cq(struct ib_cq *cq, int cqe);
3957
3958/**
3959 * rdma_set_cq_moderation - Modifies moderation params of the CQ
3960 * @cq: The CQ to modify.
3961 * @cq_count: number of CQEs that will trigger an event
3962 * @cq_period: max period of time in usec before triggering an event
3963 *
3964 */
3965int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3966
3967/**
3968 * ib_destroy_cq_user - Destroys the specified CQ.
3969 * @cq: The CQ to destroy.
3970 * @udata: Valid user data or NULL for kernel objects
3971 */
3972int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3973
3974/**
3975 * ib_destroy_cq - Destroys the specified kernel CQ.
3976 * @cq: The CQ to destroy.
3977 *
3978 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3979 */
3980static inline void ib_destroy_cq(struct ib_cq *cq)
3981{
3982 int ret = ib_destroy_cq_user(cq, NULL);
3983
3984 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail");
3985}
3986
3987/**
3988 * ib_poll_cq - poll a CQ for completion(s)
3989 * @cq:the CQ being polled
3990 * @num_entries:maximum number of completions to return
3991 * @wc:array of at least @num_entries &struct ib_wc where completions
3992 * will be returned
3993 *
3994 * Poll a CQ for (possibly multiple) completions. If the return value
3995 * is < 0, an error occurred. If the return value is >= 0, it is the
3996 * number of completions returned. If the return value is
3997 * non-negative and < num_entries, then the CQ was emptied.
3998 */
3999static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
4000 struct ib_wc *wc)
4001{
4002 return cq->device->ops.poll_cq(cq, num_entries, wc);
4003}
4004
4005/**
4006 * ib_req_notify_cq - Request completion notification on a CQ.
4007 * @cq: The CQ to generate an event for.
4008 * @flags:
4009 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
4010 * to request an event on the next solicited event or next work
4011 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
4012 * may also be |ed in to request a hint about missed events, as
4013 * described below.
4014 *
4015 * Return Value:
4016 * < 0 means an error occurred while requesting notification
4017 * == 0 means notification was requested successfully, and if
4018 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
4019 * were missed and it is safe to wait for another event. In
4020 * this case is it guaranteed that any work completions added
4021 * to the CQ since the last CQ poll will trigger a completion
4022 * notification event.
4023 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
4024 * in. It means that the consumer must poll the CQ again to
4025 * make sure it is empty to avoid missing an event because of a
4026 * race between requesting notification and an entry being
4027 * added to the CQ. This return value means it is possible
4028 * (but not guaranteed) that a work completion has been added
4029 * to the CQ since the last poll without triggering a
4030 * completion notification event.
4031 */
4032static inline int ib_req_notify_cq(struct ib_cq *cq,
4033 enum ib_cq_notify_flags flags)
4034{
4035 return cq->device->ops.req_notify_cq(cq, flags);
4036}
4037
4038struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe,
4039 int comp_vector_hint,
4040 enum ib_poll_context poll_ctx);
4041
4042void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe);
4043
4044/*
4045 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to
4046 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual
4047 * address into the dma address.
4048 */
4049static inline bool ib_uses_virt_dma(struct ib_device *dev)
4050{
4051 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device;
4052}
4053
4054/*
4055 * Check if a IB device's underlying DMA mapping supports P2PDMA transfers.
4056 */
4057static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev)
4058{
4059 if (ib_uses_virt_dma(dev))
4060 return false;
4061
4062 return dma_pci_p2pdma_supported(dev->dma_device);
4063}
4064
4065/**
4066 * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer
4067 * @dma_addr: The DMA address
4068 *
4069 * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after
4070 * going through the dma_addr marshalling.
4071 */
4072static inline void *ib_virt_dma_to_ptr(u64 dma_addr)
4073{
4074 /* virt_dma mode maps the kvs's directly into the dma addr */
4075 return (void *)(uintptr_t)dma_addr;
4076}
4077
4078/**
4079 * ib_virt_dma_to_page - Convert a dma_addr to a struct page
4080 * @dma_addr: The DMA address
4081 *
4082 * Used by ib_uses_virt_dma() device to get back to the struct page after going
4083 * through the dma_addr marshalling.
4084 */
4085static inline struct page *ib_virt_dma_to_page(u64 dma_addr)
4086{
4087 return virt_to_page(ib_virt_dma_to_ptr(dma_addr));
4088}
4089
4090/**
4091 * ib_dma_mapping_error - check a DMA addr for error
4092 * @dev: The device for which the dma_addr was created
4093 * @dma_addr: The DMA address to check
4094 */
4095static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
4096{
4097 if (ib_uses_virt_dma(dev))
4098 return 0;
4099 return dma_mapping_error(dev->dma_device, dma_addr);
4100}
4101
4102/**
4103 * ib_dma_map_single - Map a kernel virtual address to DMA address
4104 * @dev: The device for which the dma_addr is to be created
4105 * @cpu_addr: The kernel virtual address
4106 * @size: The size of the region in bytes
4107 * @direction: The direction of the DMA
4108 */
4109static inline u64 ib_dma_map_single(struct ib_device *dev,
4110 void *cpu_addr, size_t size,
4111 enum dma_data_direction direction)
4112{
4113 if (ib_uses_virt_dma(dev))
4114 return (uintptr_t)cpu_addr;
4115 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
4116}
4117
4118/**
4119 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4120 * @dev: The device for which the DMA address was created
4121 * @addr: The DMA address
4122 * @size: The size of the region in bytes
4123 * @direction: The direction of the DMA
4124 */
4125static inline void ib_dma_unmap_single(struct ib_device *dev,
4126 u64 addr, size_t size,
4127 enum dma_data_direction direction)
4128{
4129 if (!ib_uses_virt_dma(dev))
4130 dma_unmap_single(dev->dma_device, addr, size, direction);
4131}
4132
4133/**
4134 * ib_dma_map_page - Map a physical page to DMA address
4135 * @dev: The device for which the dma_addr is to be created
4136 * @page: The page to be mapped
4137 * @offset: The offset within the page
4138 * @size: The size of the region in bytes
4139 * @direction: The direction of the DMA
4140 */
4141static inline u64 ib_dma_map_page(struct ib_device *dev,
4142 struct page *page,
4143 unsigned long offset,
4144 size_t size,
4145 enum dma_data_direction direction)
4146{
4147 if (ib_uses_virt_dma(dev))
4148 return (uintptr_t)(page_address(page) + offset);
4149 return dma_map_page(dev->dma_device, page, offset, size, direction);
4150}
4151
4152/**
4153 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4154 * @dev: The device for which the DMA address was created
4155 * @addr: The DMA address
4156 * @size: The size of the region in bytes
4157 * @direction: The direction of the DMA
4158 */
4159static inline void ib_dma_unmap_page(struct ib_device *dev,
4160 u64 addr, size_t size,
4161 enum dma_data_direction direction)
4162{
4163 if (!ib_uses_virt_dma(dev))
4164 dma_unmap_page(dev->dma_device, addr, size, direction);
4165}
4166
4167int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents);
4168static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4169 struct scatterlist *sg, int nents,
4170 enum dma_data_direction direction,
4171 unsigned long dma_attrs)
4172{
4173 if (ib_uses_virt_dma(dev))
4174 return ib_dma_virt_map_sg(dev, sg, nents);
4175 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4176 dma_attrs);
4177}
4178
4179static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4180 struct scatterlist *sg, int nents,
4181 enum dma_data_direction direction,
4182 unsigned long dma_attrs)
4183{
4184 if (!ib_uses_virt_dma(dev))
4185 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
4186 dma_attrs);
4187}
4188
4189/**
4190 * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses
4191 * @dev: The device for which the DMA addresses are to be created
4192 * @sg: The sg_table object describing the buffer
4193 * @direction: The direction of the DMA
4194 * @attrs: Optional DMA attributes for the map operation
4195 */
4196static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev,
4197 struct sg_table *sgt,
4198 enum dma_data_direction direction,
4199 unsigned long dma_attrs)
4200{
4201 int nents;
4202
4203 if (ib_uses_virt_dma(dev)) {
4204 nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents);
4205 if (!nents)
4206 return -EIO;
4207 sgt->nents = nents;
4208 return 0;
4209 }
4210 return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4211}
4212
4213static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev,
4214 struct sg_table *sgt,
4215 enum dma_data_direction direction,
4216 unsigned long dma_attrs)
4217{
4218 if (!ib_uses_virt_dma(dev))
4219 dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs);
4220}
4221
4222/**
4223 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4224 * @dev: The device for which the DMA addresses are to be created
4225 * @sg: The array of scatter/gather entries
4226 * @nents: The number of scatter/gather entries
4227 * @direction: The direction of the DMA
4228 */
4229static inline int ib_dma_map_sg(struct ib_device *dev,
4230 struct scatterlist *sg, int nents,
4231 enum dma_data_direction direction)
4232{
4233 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0);
4234}
4235
4236/**
4237 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4238 * @dev: The device for which the DMA addresses were created
4239 * @sg: The array of scatter/gather entries
4240 * @nents: The number of scatter/gather entries
4241 * @direction: The direction of the DMA
4242 */
4243static inline void ib_dma_unmap_sg(struct ib_device *dev,
4244 struct scatterlist *sg, int nents,
4245 enum dma_data_direction direction)
4246{
4247 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0);
4248}
4249
4250/**
4251 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4252 * @dev: The device to query
4253 *
4254 * The returned value represents a size in bytes.
4255 */
4256static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4257{
4258 if (ib_uses_virt_dma(dev))
4259 return UINT_MAX;
4260 return dma_get_max_seg_size(dev->dma_device);
4261}
4262
4263/**
4264 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4265 * @dev: The device for which the DMA address was created
4266 * @addr: The DMA address
4267 * @size: The size of the region in bytes
4268 * @dir: The direction of the DMA
4269 */
4270static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4271 u64 addr,
4272 size_t size,
4273 enum dma_data_direction dir)
4274{
4275 if (!ib_uses_virt_dma(dev))
4276 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
4277}
4278
4279/**
4280 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4281 * @dev: The device for which the DMA address was created
4282 * @addr: The DMA address
4283 * @size: The size of the region in bytes
4284 * @dir: The direction of the DMA
4285 */
4286static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4287 u64 addr,
4288 size_t size,
4289 enum dma_data_direction dir)
4290{
4291 if (!ib_uses_virt_dma(dev))
4292 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
4293}
4294
4295/* ib_reg_user_mr - register a memory region for virtual addresses from kernel
4296 * space. This function should be called when 'current' is the owning MM.
4297 */
4298struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
4299 u64 virt_addr, int mr_access_flags);
4300
4301/* ib_advise_mr - give an advice about an address range in a memory region */
4302int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice,
4303 u32 flags, struct ib_sge *sg_list, u32 num_sge);
4304/**
4305 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4306 * HCA translation table.
4307 * @mr: The memory region to deregister.
4308 * @udata: Valid user data or NULL for kernel object
4309 *
4310 * This function can fail, if the memory region has memory windows bound to it.
4311 */
4312int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4313
4314/**
4315 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
4316 * HCA translation table.
4317 * @mr: The memory region to deregister.
4318 *
4319 * This function can fail, if the memory region has memory windows bound to it.
4320 *
4321 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
4322 */
4323static inline int ib_dereg_mr(struct ib_mr *mr)
4324{
4325 return ib_dereg_mr_user(mr, NULL);
4326}
4327
4328struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
4329 u32 max_num_sg);
4330
4331struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4332 u32 max_num_data_sg,
4333 u32 max_num_meta_sg);
4334
4335/**
4336 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4337 * R_Key and L_Key.
4338 * @mr - struct ib_mr pointer to be updated.
4339 * @newkey - new key to be used.
4340 */
4341static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4342{
4343 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4344 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4345}
4346
4347/**
4348 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4349 * for calculating a new rkey for type 2 memory windows.
4350 * @rkey - the rkey to increment.
4351 */
4352static inline u32 ib_inc_rkey(u32 rkey)
4353{
4354 const u32 mask = 0x000000ff;
4355 return ((rkey + 1) & mask) | (rkey & ~mask);
4356}
4357
4358/**
4359 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4360 * @qp: QP to attach to the multicast group. The QP must be type
4361 * IB_QPT_UD.
4362 * @gid: Multicast group GID.
4363 * @lid: Multicast group LID in host byte order.
4364 *
4365 * In order to send and receive multicast packets, subnet
4366 * administration must have created the multicast group and configured
4367 * the fabric appropriately. The port associated with the specified
4368 * QP must also be a member of the multicast group.
4369 */
4370int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4371
4372/**
4373 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4374 * @qp: QP to detach from the multicast group.
4375 * @gid: Multicast group GID.
4376 * @lid: Multicast group LID in host byte order.
4377 */
4378int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4379
4380struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device,
4381 struct inode *inode, struct ib_udata *udata);
4382int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata);
4383
4384static inline int ib_check_mr_access(struct ib_device *ib_dev,
4385 unsigned int flags)
4386{
4387 u64 device_cap = ib_dev->attrs.device_cap_flags;
4388
4389 /*
4390 * Local write permission is required if remote write or
4391 * remote atomic permission is also requested.
4392 */
4393 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4394 !(flags & IB_ACCESS_LOCAL_WRITE))
4395 return -EINVAL;
4396
4397 if (flags & ~IB_ACCESS_SUPPORTED)
4398 return -EINVAL;
4399
4400 if (flags & IB_ACCESS_ON_DEMAND &&
4401 !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING))
4402 return -EOPNOTSUPP;
4403
4404 if ((flags & IB_ACCESS_FLUSH_GLOBAL &&
4405 !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) ||
4406 (flags & IB_ACCESS_FLUSH_PERSISTENT &&
4407 !(device_cap & IB_DEVICE_FLUSH_PERSISTENT)))
4408 return -EOPNOTSUPP;
4409
4410 return 0;
4411}
4412
4413static inline bool ib_access_writable(int access_flags)
4414{
4415 /*
4416 * We have writable memory backing the MR if any of the following
4417 * access flags are set. "Local write" and "remote write" obviously
4418 * require write access. "Remote atomic" can do things like fetch and
4419 * add, which will modify memory, and "MW bind" can change permissions
4420 * by binding a window.
4421 */
4422 return access_flags &
4423 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4424 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4425}
4426
4427/**
4428 * ib_check_mr_status: lightweight check of MR status.
4429 * This routine may provide status checks on a selected
4430 * ib_mr. first use is for signature status check.
4431 *
4432 * @mr: A memory region.
4433 * @check_mask: Bitmask of which checks to perform from
4434 * ib_mr_status_check enumeration.
4435 * @mr_status: The container of relevant status checks.
4436 * failed checks will be indicated in the status bitmask
4437 * and the relevant info shall be in the error item.
4438 */
4439int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4440 struct ib_mr_status *mr_status);
4441
4442/**
4443 * ib_device_try_get: Hold a registration lock
4444 * device: The device to lock
4445 *
4446 * A device under an active registration lock cannot become unregistered. It
4447 * is only possible to obtain a registration lock on a device that is fully
4448 * registered, otherwise this function returns false.
4449 *
4450 * The registration lock is only necessary for actions which require the
4451 * device to still be registered. Uses that only require the device pointer to
4452 * be valid should use get_device(&ibdev->dev) to hold the memory.
4453 *
4454 */
4455static inline bool ib_device_try_get(struct ib_device *dev)
4456{
4457 return refcount_inc_not_zero(&dev->refcount);
4458}
4459
4460void ib_device_put(struct ib_device *device);
4461struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4462 enum rdma_driver_id driver_id);
4463struct ib_device *ib_device_get_by_name(const char *name,
4464 enum rdma_driver_id driver_id);
4465struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port,
4466 u16 pkey, const union ib_gid *gid,
4467 const struct sockaddr *addr);
4468int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4469 unsigned int port);
4470struct net_device *ib_device_get_netdev(struct ib_device *ib_dev,
4471 u32 port);
4472struct ib_wq *ib_create_wq(struct ib_pd *pd,
4473 struct ib_wq_init_attr *init_attr);
4474int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata);
4475
4476int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4477 unsigned int *sg_offset, unsigned int page_size);
4478int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4479 int data_sg_nents, unsigned int *data_sg_offset,
4480 struct scatterlist *meta_sg, int meta_sg_nents,
4481 unsigned int *meta_sg_offset, unsigned int page_size);
4482
4483static inline int
4484ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
4485 unsigned int *sg_offset, unsigned int page_size)
4486{
4487 int n;
4488
4489 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4490 mr->iova = 0;
4491
4492 return n;
4493}
4494
4495int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
4496 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4497
4498void ib_drain_rq(struct ib_qp *qp);
4499void ib_drain_sq(struct ib_qp *qp);
4500void ib_drain_qp(struct ib_qp *qp);
4501
4502int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed,
4503 u8 *width);
4504
4505static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4506{
4507 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4508 return attr->roce.dmac;
4509 return NULL;
4510}
4511
4512static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
4513{
4514 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4515 attr->ib.dlid = (u16)dlid;
4516 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4517 attr->opa.dlid = dlid;
4518}
4519
4520static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
4521{
4522 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4523 return attr->ib.dlid;
4524 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4525 return attr->opa.dlid;
4526 return 0;
4527}
4528
4529static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4530{
4531 attr->sl = sl;
4532}
4533
4534static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4535{
4536 return attr->sl;
4537}
4538
4539static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4540 u8 src_path_bits)
4541{
4542 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4543 attr->ib.src_path_bits = src_path_bits;
4544 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4545 attr->opa.src_path_bits = src_path_bits;
4546}
4547
4548static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4549{
4550 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4551 return attr->ib.src_path_bits;
4552 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4553 return attr->opa.src_path_bits;
4554 return 0;
4555}
4556
4557static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4558 bool make_grd)
4559{
4560 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4561 attr->opa.make_grd = make_grd;
4562}
4563
4564static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4565{
4566 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4567 return attr->opa.make_grd;
4568 return false;
4569}
4570
4571static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num)
4572{
4573 attr->port_num = port_num;
4574}
4575
4576static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4577{
4578 return attr->port_num;
4579}
4580
4581static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4582 u8 static_rate)
4583{
4584 attr->static_rate = static_rate;
4585}
4586
4587static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4588{
4589 return attr->static_rate;
4590}
4591
4592static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4593 enum ib_ah_flags flag)
4594{
4595 attr->ah_flags = flag;
4596}
4597
4598static inline enum ib_ah_flags
4599 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4600{
4601 return attr->ah_flags;
4602}
4603
4604static inline const struct ib_global_route
4605 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4606{
4607 return &attr->grh;
4608}
4609
4610/*To retrieve and modify the grh */
4611static inline struct ib_global_route
4612 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4613{
4614 return &attr->grh;
4615}
4616
4617static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4618{
4619 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4620
4621 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4622}
4623
4624static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4625 __be64 prefix)
4626{
4627 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4628
4629 grh->dgid.global.subnet_prefix = prefix;
4630}
4631
4632static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4633 __be64 if_id)
4634{
4635 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4636
4637 grh->dgid.global.interface_id = if_id;
4638}
4639
4640static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4641 union ib_gid *dgid, u32 flow_label,
4642 u8 sgid_index, u8 hop_limit,
4643 u8 traffic_class)
4644{
4645 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4646
4647 attr->ah_flags = IB_AH_GRH;
4648 if (dgid)
4649 grh->dgid = *dgid;
4650 grh->flow_label = flow_label;
4651 grh->sgid_index = sgid_index;
4652 grh->hop_limit = hop_limit;
4653 grh->traffic_class = traffic_class;
4654 grh->sgid_attr = NULL;
4655}
4656
4657void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4658void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4659 u32 flow_label, u8 hop_limit, u8 traffic_class,
4660 const struct ib_gid_attr *sgid_attr);
4661void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4662 const struct rdma_ah_attr *src);
4663void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4664 const struct rdma_ah_attr *new);
4665void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
4666
4667/**
4668 * rdma_ah_find_type - Return address handle type.
4669 *
4670 * @dev: Device to be checked
4671 * @port_num: Port number
4672 */
4673static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
4674 u32 port_num)
4675{
4676 if (rdma_protocol_roce(dev, port_num))
4677 return RDMA_AH_ATTR_TYPE_ROCE;
4678 if (rdma_protocol_ib(dev, port_num)) {
4679 if (rdma_cap_opa_ah(dev, port_num))
4680 return RDMA_AH_ATTR_TYPE_OPA;
4681 return RDMA_AH_ATTR_TYPE_IB;
4682 }
4683 if (dev->type == RDMA_DEVICE_TYPE_SMI)
4684 return RDMA_AH_ATTR_TYPE_IB;
4685
4686 return RDMA_AH_ATTR_TYPE_UNDEFINED;
4687}
4688
4689/**
4690 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4691 * In the current implementation the only way to
4692 * get the 32bit lid is from other sources for OPA.
4693 * For IB, lids will always be 16bits so cast the
4694 * value accordingly.
4695 *
4696 * @lid: A 32bit LID
4697 */
4698static inline u16 ib_lid_cpu16(u32 lid)
4699{
4700 WARN_ON_ONCE(lid & 0xFFFF0000);
4701 return (u16)lid;
4702}
4703
4704/**
4705 * ib_lid_be16 - Return lid in 16bit BE encoding.
4706 *
4707 * @lid: A 32bit LID
4708 */
4709static inline __be16 ib_lid_be16(u32 lid)
4710{
4711 WARN_ON_ONCE(lid & 0xFFFF0000);
4712 return cpu_to_be16((u16)lid);
4713}
4714
4715/**
4716 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4717 * vector
4718 * @device: the rdma device
4719 * @comp_vector: index of completion vector
4720 *
4721 * Returns NULL on failure, otherwise a corresponding cpu map of the
4722 * completion vector (returns all-cpus map if the device driver doesn't
4723 * implement get_vector_affinity).
4724 */
4725static inline const struct cpumask *
4726ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4727{
4728 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
4729 !device->ops.get_vector_affinity)
4730 return NULL;
4731
4732 return device->ops.get_vector_affinity(device, comp_vector);
4733
4734}
4735
4736/**
4737 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4738 * and add their gids, as needed, to the relevant RoCE devices.
4739 *
4740 * @device: the rdma device
4741 */
4742void rdma_roce_rescan_device(struct ib_device *ibdev);
4743void rdma_roce_rescan_port(struct ib_device *ib_dev, u32 port);
4744void roce_del_all_netdev_gids(struct ib_device *ib_dev,
4745 u32 port, struct net_device *ndev);
4746
4747struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
4748
4749int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
4750
4751struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num,
4752 enum rdma_netdev_t type, const char *name,
4753 unsigned char name_assign_type,
4754 void (*setup)(struct net_device *));
4755
4756int rdma_init_netdev(struct ib_device *device, u32 port_num,
4757 enum rdma_netdev_t type, const char *name,
4758 unsigned char name_assign_type,
4759 void (*setup)(struct net_device *),
4760 struct net_device *netdev);
4761
4762/**
4763 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4764 *
4765 * @device: device pointer for which ib_device pointer to retrieve
4766 *
4767 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4768 *
4769 */
4770static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4771{
4772 struct ib_core_device *coredev =
4773 container_of(device, struct ib_core_device, dev);
4774
4775 return coredev->owner;
4776}
4777
4778/**
4779 * ibdev_to_node - return the NUMA node for a given ib_device
4780 * @dev: device to get the NUMA node for.
4781 */
4782static inline int ibdev_to_node(struct ib_device *ibdev)
4783{
4784 struct device *parent = ibdev->dev.parent;
4785
4786 if (!parent)
4787 return NUMA_NO_NODE;
4788 return dev_to_node(parent);
4789}
4790
4791/**
4792 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4793 * ib_device holder structure from device pointer.
4794 *
4795 * NOTE: New drivers should not make use of this API; This API is only for
4796 * existing drivers who have exposed sysfs entries using
4797 * ops->device_group.
4798 */
4799#define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4800 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
4801
4802bool rdma_dev_access_netns(const struct ib_device *device,
4803 const struct net *net);
4804
4805#define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000)
4806#define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF)
4807#define IB_GRH_FLOWLABEL_MASK (0x000FFFFF)
4808
4809/**
4810 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based
4811 * on the flow_label
4812 *
4813 * This function will convert the 20 bit flow_label input to a valid RoCE v2
4814 * UDP src port 14 bit value. All RoCE V2 drivers should use this same
4815 * convention.
4816 */
4817static inline u16 rdma_flow_label_to_udp_sport(u32 fl)
4818{
4819 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000;
4820
4821 fl_low ^= fl_high >> 14;
4822 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN);
4823}
4824
4825/**
4826 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on
4827 * local and remote qpn values
4828 *
4829 * This function folded the multiplication results of two qpns, 24 bit each,
4830 * fields, and converts it to a 20 bit results.
4831 *
4832 * This function will create symmetric flow_label value based on the local
4833 * and remote qpn values. this will allow both the requester and responder
4834 * to calculate the same flow_label for a given connection.
4835 *
4836 * This helper function should be used by driver in case the upper layer
4837 * provide a zero flow_label value. This is to improve entropy of RDMA
4838 * traffic in the network.
4839 */
4840static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn)
4841{
4842 u64 v = (u64)lqpn * rqpn;
4843
4844 v ^= v >> 20;
4845 v ^= v >> 40;
4846
4847 return (u32)(v & IB_GRH_FLOWLABEL_MASK);
4848}
4849
4850/**
4851 * rdma_get_udp_sport - Calculate and set UDP source port based on the flow
4852 * label. If flow label is not defined in GRH then
4853 * calculate it based on lqpn/rqpn.
4854 *
4855 * @fl: flow label from GRH
4856 * @lqpn: local qp number
4857 * @rqpn: remote qp number
4858 */
4859static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn)
4860{
4861 if (!fl)
4862 fl = rdma_calc_flow_label(lqpn, rqpn);
4863
4864 return rdma_flow_label_to_udp_sport(fl);
4865}
4866
4867const struct ib_port_immutable*
4868ib_port_immutable_read(struct ib_device *dev, unsigned int port);
4869
4870/** ib_add_sub_device - Add a sub IB device on an existing one
4871 *
4872 * @parent: The IB device that needs to add a sub device
4873 * @type: The type of the new sub device
4874 * @name: The name of the new sub device
4875 *
4876 *
4877 * Return 0 on success, an error code otherwise
4878 */
4879int ib_add_sub_device(struct ib_device *parent,
4880 enum rdma_nl_dev_type type,
4881 const char *name);
4882
4883
4884/** ib_del_sub_device_and_put - Delect an IB sub device while holding a 'get'
4885 *
4886 * @sub: The sub device that is going to be deleted
4887 *
4888 * Return 0 on success, an error code otherwise
4889 */
4890int ib_del_sub_device_and_put(struct ib_device *sub);
4891
4892static inline void ib_mark_name_assigned_by_user(struct ib_device *ibdev)
4893{
4894 ibdev->name_assign_type = RDMA_NAME_ASSIGN_TYPE_USER;
4895}
4896
4897#endif /* IB_VERBS_H */