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