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 * $FreeBSD$
39 */
40
41 #if !defined(IB_VERBS_H)
42 #define IB_VERBS_H
43
44 #include <linux/types.h>
45 #include <linux/device.h>
46 #include <linux/mm.h>
47 #include <linux/dma-mapping.h>
48 #include <linux/kref.h>
49 #include <linux/list.h>
50 #include <linux/rwsem.h>
51 #include <linux/scatterlist.h>
52 #include <linux/workqueue.h>
53 #include <linux/socket.h>
54 #include <linux/if_ether.h>
55 #include <net/ipv6.h>
56 #include <net/ip.h>
57 #include <linux/string.h>
58 #include <linux/slab.h>
59 #include <linux/rcupdate.h>
60 #include <linux/netdevice.h>
61 #include <netinet/ip.h>
62
63 #include <asm/atomic.h>
64 #include <asm/uaccess.h>
65
66 struct ifla_vf_info;
67 struct ifla_vf_stats;
68
69 extern struct workqueue_struct *ib_wq;
70 extern struct workqueue_struct *ib_comp_wq;
71
72 union ib_gid {
73 u8 raw[16];
74 struct {
75 __be64 subnet_prefix;
76 __be64 interface_id;
77 } global;
78 };
79
80 extern union ib_gid zgid;
81
82 enum ib_gid_type {
83 /* If link layer is Ethernet, this is RoCE V1 */
84 IB_GID_TYPE_IB = 0,
85 IB_GID_TYPE_ROCE = 0,
86 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
87 IB_GID_TYPE_SIZE
88 };
89
90 #define ROCE_V2_UDP_DPORT 4791
91 struct ib_gid_attr {
92 enum ib_gid_type gid_type;
93 struct net_device *ndev;
94 };
95
96 enum rdma_node_type {
97 /* IB values map to NodeInfo:NodeType. */
98 RDMA_NODE_IB_CA = 1,
99 RDMA_NODE_IB_SWITCH,
100 RDMA_NODE_IB_ROUTER,
101 RDMA_NODE_RNIC,
102 RDMA_NODE_USNIC,
103 RDMA_NODE_USNIC_UDP,
104 };
105
106 enum {
107 /* set the local administered indication */
108 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
109 };
110
111 enum rdma_transport_type {
112 RDMA_TRANSPORT_IB,
113 RDMA_TRANSPORT_IWARP,
114 RDMA_TRANSPORT_USNIC,
115 RDMA_TRANSPORT_USNIC_UDP
116 };
117
118 enum rdma_protocol_type {
119 RDMA_PROTOCOL_IB,
120 RDMA_PROTOCOL_IBOE,
121 RDMA_PROTOCOL_IWARP,
122 RDMA_PROTOCOL_USNIC_UDP
123 };
124
125 __attribute_const__ enum rdma_transport_type
126 rdma_node_get_transport(enum rdma_node_type node_type);
127
128 enum rdma_network_type {
129 RDMA_NETWORK_IB,
130 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
131 RDMA_NETWORK_IPV4,
132 RDMA_NETWORK_IPV6
133 };
134
ib_network_to_gid_type(enum rdma_network_type network_type)135 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
136 {
137 if (network_type == RDMA_NETWORK_IPV4 ||
138 network_type == RDMA_NETWORK_IPV6)
139 return IB_GID_TYPE_ROCE_UDP_ENCAP;
140
141 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
142 return IB_GID_TYPE_IB;
143 }
144
ib_gid_to_network_type(enum ib_gid_type gid_type,union ib_gid * gid)145 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
146 union ib_gid *gid)
147 {
148 if (gid_type == IB_GID_TYPE_IB)
149 return RDMA_NETWORK_IB;
150
151 if (ipv6_addr_v4mapped((struct in6_addr *)gid))
152 return RDMA_NETWORK_IPV4;
153 else
154 return RDMA_NETWORK_IPV6;
155 }
156
157 enum rdma_link_layer {
158 IB_LINK_LAYER_UNSPECIFIED,
159 IB_LINK_LAYER_INFINIBAND,
160 IB_LINK_LAYER_ETHERNET,
161 };
162
163 enum ib_device_cap_flags {
164 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
165 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
166 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
167 IB_DEVICE_RAW_MULTI = (1 << 3),
168 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
169 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
170 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
171 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
172 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
173 IB_DEVICE_INIT_TYPE = (1 << 9),
174 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
175 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
176 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
177 IB_DEVICE_SRQ_RESIZE = (1 << 13),
178 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
179
180 /*
181 * This device supports a per-device lkey or stag that can be
182 * used without performing a memory registration for the local
183 * memory. Note that ULPs should never check this flag, but
184 * instead of use the local_dma_lkey flag in the ib_pd structure,
185 * which will always contain a usable lkey.
186 */
187 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
188 IB_DEVICE_RESERVED /* old SEND_W_INV */ = (1 << 16),
189 IB_DEVICE_MEM_WINDOW = (1 << 17),
190 /*
191 * Devices should set IB_DEVICE_UD_IP_SUM if they support
192 * insertion of UDP and TCP checksum on outgoing UD IPoIB
193 * messages and can verify the validity of checksum for
194 * incoming messages. Setting this flag implies that the
195 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
196 */
197 IB_DEVICE_UD_IP_CSUM = (1 << 18),
198 IB_DEVICE_UD_TSO = (1 << 19),
199 IB_DEVICE_XRC = (1 << 20),
200
201 /*
202 * This device supports the IB "base memory management extension",
203 * which includes support for fast registrations (IB_WR_REG_MR,
204 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
205 * also be set by any iWarp device which must support FRs to comply
206 * to the iWarp verbs spec. iWarp devices also support the
207 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
208 * stag.
209 */
210 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
211 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
212 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
213 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
214 IB_DEVICE_RC_IP_CSUM = (1 << 25),
215 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
216 /*
217 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
218 * support execution of WQEs that involve synchronization
219 * of I/O operations with single completion queue managed
220 * by hardware.
221 */
222 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
223 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
224 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
225 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
226 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
227 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
228 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
229 };
230
231 enum ib_signature_prot_cap {
232 IB_PROT_T10DIF_TYPE_1 = 1,
233 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
234 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
235 };
236
237 enum ib_signature_guard_cap {
238 IB_GUARD_T10DIF_CRC = 1,
239 IB_GUARD_T10DIF_CSUM = 1 << 1,
240 };
241
242 enum ib_atomic_cap {
243 IB_ATOMIC_NONE,
244 IB_ATOMIC_HCA,
245 IB_ATOMIC_GLOB
246 };
247
248 enum ib_odp_general_cap_bits {
249 IB_ODP_SUPPORT = 1 << 0,
250 };
251
252 enum ib_odp_transport_cap_bits {
253 IB_ODP_SUPPORT_SEND = 1 << 0,
254 IB_ODP_SUPPORT_RECV = 1 << 1,
255 IB_ODP_SUPPORT_WRITE = 1 << 2,
256 IB_ODP_SUPPORT_READ = 1 << 3,
257 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
258 };
259
260 struct ib_odp_caps {
261 uint64_t general_caps;
262 struct {
263 uint32_t rc_odp_caps;
264 uint32_t uc_odp_caps;
265 uint32_t ud_odp_caps;
266 } per_transport_caps;
267 };
268
269 struct ib_rss_caps {
270 /* Corresponding bit will be set if qp type from
271 * 'enum ib_qp_type' is supported, e.g.
272 * supported_qpts |= 1 << IB_QPT_UD
273 */
274 u32 supported_qpts;
275 u32 max_rwq_indirection_tables;
276 u32 max_rwq_indirection_table_size;
277 };
278
279 enum ib_cq_creation_flags {
280 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
281 IB_CQ_FLAGS_IGNORE_OVERRUN = 1 << 1,
282 };
283
284 struct ib_cq_init_attr {
285 unsigned int cqe;
286 u32 comp_vector;
287 u32 flags;
288 };
289
290 struct ib_device_attr {
291 u64 fw_ver;
292 __be64 sys_image_guid;
293 u64 max_mr_size;
294 u64 page_size_cap;
295 u32 vendor_id;
296 u32 vendor_part_id;
297 u32 hw_ver;
298 int max_qp;
299 int max_qp_wr;
300 u64 device_cap_flags;
301 int max_sge;
302 int max_sge_rd;
303 int max_cq;
304 int max_cqe;
305 int max_mr;
306 int max_pd;
307 int max_qp_rd_atom;
308 int max_ee_rd_atom;
309 int max_res_rd_atom;
310 int max_qp_init_rd_atom;
311 int max_ee_init_rd_atom;
312 enum ib_atomic_cap atomic_cap;
313 enum ib_atomic_cap masked_atomic_cap;
314 int max_ee;
315 int max_rdd;
316 int max_mw;
317 int max_raw_ipv6_qp;
318 int max_raw_ethy_qp;
319 int max_mcast_grp;
320 int max_mcast_qp_attach;
321 int max_total_mcast_qp_attach;
322 int max_ah;
323 int max_fmr;
324 int max_map_per_fmr;
325 int max_srq;
326 int max_srq_wr;
327 int max_srq_sge;
328 unsigned int max_fast_reg_page_list_len;
329 u16 max_pkeys;
330 u8 local_ca_ack_delay;
331 int sig_prot_cap;
332 int sig_guard_cap;
333 struct ib_odp_caps odp_caps;
334 uint64_t timestamp_mask;
335 uint64_t hca_core_clock; /* in KHZ */
336 struct ib_rss_caps rss_caps;
337 u32 max_wq_type_rq;
338 };
339
340 enum ib_mtu {
341 IB_MTU_256 = 1,
342 IB_MTU_512 = 2,
343 IB_MTU_1024 = 3,
344 IB_MTU_2048 = 4,
345 IB_MTU_4096 = 5
346 };
347
ib_mtu_enum_to_int(enum ib_mtu mtu)348 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
349 {
350 switch (mtu) {
351 case IB_MTU_256: return 256;
352 case IB_MTU_512: return 512;
353 case IB_MTU_1024: return 1024;
354 case IB_MTU_2048: return 2048;
355 case IB_MTU_4096: return 4096;
356 default: return -1;
357 }
358 }
359
360 enum ib_port_state {
361 IB_PORT_NOP = 0,
362 IB_PORT_DOWN = 1,
363 IB_PORT_INIT = 2,
364 IB_PORT_ARMED = 3,
365 IB_PORT_ACTIVE = 4,
366 IB_PORT_ACTIVE_DEFER = 5,
367 IB_PORT_DUMMY = -1, /* force enum signed */
368 };
369
370 enum ib_port_cap_flags {
371 IB_PORT_SM = 1 << 1,
372 IB_PORT_NOTICE_SUP = 1 << 2,
373 IB_PORT_TRAP_SUP = 1 << 3,
374 IB_PORT_OPT_IPD_SUP = 1 << 4,
375 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
376 IB_PORT_SL_MAP_SUP = 1 << 6,
377 IB_PORT_MKEY_NVRAM = 1 << 7,
378 IB_PORT_PKEY_NVRAM = 1 << 8,
379 IB_PORT_LED_INFO_SUP = 1 << 9,
380 IB_PORT_SM_DISABLED = 1 << 10,
381 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
382 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
383 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
384 IB_PORT_CM_SUP = 1 << 16,
385 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
386 IB_PORT_REINIT_SUP = 1 << 18,
387 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
388 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
389 IB_PORT_DR_NOTICE_SUP = 1 << 21,
390 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
391 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
392 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
393 IB_PORT_CLIENT_REG_SUP = 1 << 25,
394 IB_PORT_IP_BASED_GIDS = 1 << 26,
395 };
396
397 enum ib_port_width {
398 IB_WIDTH_1X = 1,
399 IB_WIDTH_2X = 16,
400 IB_WIDTH_4X = 2,
401 IB_WIDTH_8X = 4,
402 IB_WIDTH_12X = 8
403 };
404
ib_width_enum_to_int(enum ib_port_width width)405 static inline int ib_width_enum_to_int(enum ib_port_width width)
406 {
407 switch (width) {
408 case IB_WIDTH_1X: return 1;
409 case IB_WIDTH_2X: return 2;
410 case IB_WIDTH_4X: return 4;
411 case IB_WIDTH_8X: return 8;
412 case IB_WIDTH_12X: return 12;
413 default: return -1;
414 }
415 }
416
417 enum ib_port_speed {
418 IB_SPEED_SDR = 1,
419 IB_SPEED_DDR = 2,
420 IB_SPEED_QDR = 4,
421 IB_SPEED_FDR10 = 8,
422 IB_SPEED_FDR = 16,
423 IB_SPEED_EDR = 32,
424 IB_SPEED_HDR = 64
425 };
426
427 /**
428 * struct rdma_hw_stats
429 * @timestamp - Used by the core code to track when the last update was
430 * @lifespan - Used by the core code to determine how old the counters
431 * should be before being updated again. Stored in jiffies, defaults
432 * to 10 milliseconds, drivers can override the default be specifying
433 * their own value during their allocation routine.
434 * @name - Array of pointers to static names used for the counters in
435 * directory.
436 * @num_counters - How many hardware counters there are. If name is
437 * shorter than this number, a kernel oops will result. Driver authors
438 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
439 * in their code to prevent this.
440 * @value - Array of u64 counters that are accessed by the sysfs code and
441 * filled in by the drivers get_stats routine
442 */
443 struct rdma_hw_stats {
444 unsigned long timestamp;
445 unsigned long lifespan;
446 const char * const *names;
447 int num_counters;
448 u64 value[];
449 };
450
451 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
452 /**
453 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
454 * for drivers.
455 * @names - Array of static const char *
456 * @num_counters - How many elements in array
457 * @lifespan - How many milliseconds between updates
458 */
rdma_alloc_hw_stats_struct(const char * const * names,int num_counters,unsigned long lifespan)459 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
460 const char * const *names, int num_counters,
461 unsigned long lifespan)
462 {
463 struct rdma_hw_stats *stats;
464
465 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
466 GFP_KERNEL);
467 if (!stats)
468 return NULL;
469 stats->names = names;
470 stats->num_counters = num_counters;
471 stats->lifespan = msecs_to_jiffies(lifespan);
472
473 return stats;
474 }
475
476
477 /* Define bits for the various functionality this port needs to be supported by
478 * the core.
479 */
480 /* Management 0x00000FFF */
481 #define RDMA_CORE_CAP_IB_MAD 0x00000001
482 #define RDMA_CORE_CAP_IB_SMI 0x00000002
483 #define RDMA_CORE_CAP_IB_CM 0x00000004
484 #define RDMA_CORE_CAP_IW_CM 0x00000008
485 #define RDMA_CORE_CAP_IB_SA 0x00000010
486 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
487
488 /* Address format 0x000FF000 */
489 #define RDMA_CORE_CAP_AF_IB 0x00001000
490 #define RDMA_CORE_CAP_ETH_AH 0x00002000
491
492 /* Protocol 0xFFF00000 */
493 #define RDMA_CORE_CAP_PROT_IB 0x00100000
494 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
495 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
496 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
497
498 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
499 | RDMA_CORE_CAP_IB_MAD \
500 | RDMA_CORE_CAP_IB_SMI \
501 | RDMA_CORE_CAP_IB_CM \
502 | RDMA_CORE_CAP_IB_SA \
503 | RDMA_CORE_CAP_AF_IB)
504 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
505 | RDMA_CORE_CAP_IB_MAD \
506 | RDMA_CORE_CAP_IB_CM \
507 | RDMA_CORE_CAP_AF_IB \
508 | RDMA_CORE_CAP_ETH_AH)
509 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
510 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
511 | RDMA_CORE_CAP_IB_MAD \
512 | RDMA_CORE_CAP_IB_CM \
513 | RDMA_CORE_CAP_AF_IB \
514 | RDMA_CORE_CAP_ETH_AH)
515 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
516 | RDMA_CORE_CAP_IW_CM)
517 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
518 | RDMA_CORE_CAP_OPA_MAD)
519
520 struct ib_port_attr {
521 u64 subnet_prefix;
522 enum ib_port_state state;
523 enum ib_mtu max_mtu;
524 enum ib_mtu active_mtu;
525 int gid_tbl_len;
526 u32 port_cap_flags;
527 u32 max_msg_sz;
528 u32 bad_pkey_cntr;
529 u32 qkey_viol_cntr;
530 u16 pkey_tbl_len;
531 u16 lid;
532 u16 sm_lid;
533 u8 lmc;
534 u8 max_vl_num;
535 u8 sm_sl;
536 u8 subnet_timeout;
537 u8 init_type_reply;
538 u8 active_width;
539 u8 active_speed;
540 u8 phys_state;
541 bool grh_required;
542 };
543
544 enum ib_device_modify_flags {
545 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
546 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
547 };
548
549 #define IB_DEVICE_NODE_DESC_MAX 64
550
551 struct ib_device_modify {
552 u64 sys_image_guid;
553 char node_desc[IB_DEVICE_NODE_DESC_MAX];
554 };
555
556 enum ib_port_modify_flags {
557 IB_PORT_SHUTDOWN = 1,
558 IB_PORT_INIT_TYPE = (1<<2),
559 IB_PORT_RESET_QKEY_CNTR = (1<<3)
560 };
561
562 struct ib_port_modify {
563 u32 set_port_cap_mask;
564 u32 clr_port_cap_mask;
565 u8 init_type;
566 };
567
568 enum ib_event_type {
569 IB_EVENT_CQ_ERR,
570 IB_EVENT_QP_FATAL,
571 IB_EVENT_QP_REQ_ERR,
572 IB_EVENT_QP_ACCESS_ERR,
573 IB_EVENT_COMM_EST,
574 IB_EVENT_SQ_DRAINED,
575 IB_EVENT_PATH_MIG,
576 IB_EVENT_PATH_MIG_ERR,
577 IB_EVENT_DEVICE_FATAL,
578 IB_EVENT_PORT_ACTIVE,
579 IB_EVENT_PORT_ERR,
580 IB_EVENT_LID_CHANGE,
581 IB_EVENT_PKEY_CHANGE,
582 IB_EVENT_SM_CHANGE,
583 IB_EVENT_SRQ_ERR,
584 IB_EVENT_SRQ_LIMIT_REACHED,
585 IB_EVENT_QP_LAST_WQE_REACHED,
586 IB_EVENT_CLIENT_REREGISTER,
587 IB_EVENT_GID_CHANGE,
588 IB_EVENT_WQ_FATAL,
589 };
590
591 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
592
593 struct ib_event {
594 struct ib_device *device;
595 union {
596 struct ib_cq *cq;
597 struct ib_qp *qp;
598 struct ib_srq *srq;
599 struct ib_wq *wq;
600 u8 port_num;
601 } element;
602 enum ib_event_type event;
603 };
604
605 struct ib_event_handler {
606 struct ib_device *device;
607 void (*handler)(struct ib_event_handler *, struct ib_event *);
608 struct list_head list;
609 };
610
611 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
612 do { \
613 (_ptr)->device = _device; \
614 (_ptr)->handler = _handler; \
615 INIT_LIST_HEAD(&(_ptr)->list); \
616 } while (0)
617
618 struct ib_global_route {
619 union ib_gid dgid;
620 u32 flow_label;
621 u8 sgid_index;
622 u8 hop_limit;
623 u8 traffic_class;
624 };
625
626 struct ib_grh {
627 __be32 version_tclass_flow;
628 __be16 paylen;
629 u8 next_hdr;
630 u8 hop_limit;
631 union ib_gid sgid;
632 union ib_gid dgid;
633 };
634
635 union rdma_network_hdr {
636 struct ib_grh ibgrh;
637 struct {
638 /* The IB spec states that if it's IPv4, the header
639 * is located in the last 20 bytes of the header.
640 */
641 u8 reserved[20];
642 struct ip roce4grh;
643 };
644 };
645
646 enum {
647 IB_MULTICAST_QPN = 0xffffff
648 };
649
650 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
651 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
652
653 enum ib_ah_flags {
654 IB_AH_GRH = 1
655 };
656
657 enum ib_rate {
658 IB_RATE_PORT_CURRENT = 0,
659 IB_RATE_2_5_GBPS = 2,
660 IB_RATE_5_GBPS = 5,
661 IB_RATE_10_GBPS = 3,
662 IB_RATE_20_GBPS = 6,
663 IB_RATE_30_GBPS = 4,
664 IB_RATE_40_GBPS = 7,
665 IB_RATE_60_GBPS = 8,
666 IB_RATE_80_GBPS = 9,
667 IB_RATE_120_GBPS = 10,
668 IB_RATE_14_GBPS = 11,
669 IB_RATE_56_GBPS = 12,
670 IB_RATE_112_GBPS = 13,
671 IB_RATE_168_GBPS = 14,
672 IB_RATE_25_GBPS = 15,
673 IB_RATE_100_GBPS = 16,
674 IB_RATE_200_GBPS = 17,
675 IB_RATE_300_GBPS = 18,
676 IB_RATE_28_GBPS = 19,
677 IB_RATE_50_GBPS = 20,
678 IB_RATE_400_GBPS = 21,
679 IB_RATE_600_GBPS = 22,
680 };
681
682 /**
683 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
684 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
685 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
686 * @rate: rate to convert.
687 */
688 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
689
690 /**
691 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
692 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
693 * @rate: rate to convert.
694 */
695 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
696
697
698 /**
699 * enum ib_mr_type - memory region type
700 * @IB_MR_TYPE_MEM_REG: memory region that is used for
701 * normal registration
702 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
703 * signature operations (data-integrity
704 * capable regions)
705 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
706 * register any arbitrary sg lists (without
707 * the normal mr constraints - see
708 * ib_map_mr_sg)
709 */
710 enum ib_mr_type {
711 IB_MR_TYPE_MEM_REG,
712 IB_MR_TYPE_SIGNATURE,
713 IB_MR_TYPE_SG_GAPS,
714 };
715
716 /**
717 * Signature types
718 * IB_SIG_TYPE_NONE: Unprotected.
719 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
720 */
721 enum ib_signature_type {
722 IB_SIG_TYPE_NONE,
723 IB_SIG_TYPE_T10_DIF,
724 };
725
726 /**
727 * Signature T10-DIF block-guard types
728 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
729 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
730 */
731 enum ib_t10_dif_bg_type {
732 IB_T10DIF_CRC,
733 IB_T10DIF_CSUM
734 };
735
736 /**
737 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
738 * domain.
739 * @bg_type: T10-DIF block guard type (CRC|CSUM)
740 * @pi_interval: protection information interval.
741 * @bg: seed of guard computation.
742 * @app_tag: application tag of guard block
743 * @ref_tag: initial guard block reference tag.
744 * @ref_remap: Indicate wethear the reftag increments each block
745 * @app_escape: Indicate to skip block check if apptag=0xffff
746 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
747 * @apptag_check_mask: check bitmask of application tag.
748 */
749 struct ib_t10_dif_domain {
750 enum ib_t10_dif_bg_type bg_type;
751 u16 pi_interval;
752 u16 bg;
753 u16 app_tag;
754 u32 ref_tag;
755 bool ref_remap;
756 bool app_escape;
757 bool ref_escape;
758 u16 apptag_check_mask;
759 };
760
761 /**
762 * struct ib_sig_domain - Parameters for signature domain
763 * @sig_type: specific signauture type
764 * @sig: union of all signature domain attributes that may
765 * be used to set domain layout.
766 */
767 struct ib_sig_domain {
768 enum ib_signature_type sig_type;
769 union {
770 struct ib_t10_dif_domain dif;
771 } sig;
772 };
773
774 /**
775 * struct ib_sig_attrs - Parameters for signature handover operation
776 * @check_mask: bitmask for signature byte check (8 bytes)
777 * @mem: memory domain layout desciptor.
778 * @wire: wire domain layout desciptor.
779 */
780 struct ib_sig_attrs {
781 u8 check_mask;
782 struct ib_sig_domain mem;
783 struct ib_sig_domain wire;
784 };
785
786 enum ib_sig_err_type {
787 IB_SIG_BAD_GUARD,
788 IB_SIG_BAD_REFTAG,
789 IB_SIG_BAD_APPTAG,
790 };
791
792 /**
793 * struct ib_sig_err - signature error descriptor
794 */
795 struct ib_sig_err {
796 enum ib_sig_err_type err_type;
797 u32 expected;
798 u32 actual;
799 u64 sig_err_offset;
800 u32 key;
801 };
802
803 enum ib_mr_status_check {
804 IB_MR_CHECK_SIG_STATUS = 1,
805 };
806
807 /**
808 * struct ib_mr_status - Memory region status container
809 *
810 * @fail_status: Bitmask of MR checks status. For each
811 * failed check a corresponding status bit is set.
812 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
813 * failure.
814 */
815 struct ib_mr_status {
816 u32 fail_status;
817 struct ib_sig_err sig_err;
818 };
819
820 /**
821 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
822 * enum.
823 * @mult: multiple to convert.
824 */
825 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
826
827 struct ib_ah_attr {
828 struct ib_global_route grh;
829 u16 dlid;
830 u8 sl;
831 u8 src_path_bits;
832 u8 static_rate;
833 u8 ah_flags;
834 u8 port_num;
835 u8 dmac[ETH_ALEN];
836 };
837
838 enum ib_wc_status {
839 IB_WC_SUCCESS,
840 IB_WC_LOC_LEN_ERR,
841 IB_WC_LOC_QP_OP_ERR,
842 IB_WC_LOC_EEC_OP_ERR,
843 IB_WC_LOC_PROT_ERR,
844 IB_WC_WR_FLUSH_ERR,
845 IB_WC_MW_BIND_ERR,
846 IB_WC_BAD_RESP_ERR,
847 IB_WC_LOC_ACCESS_ERR,
848 IB_WC_REM_INV_REQ_ERR,
849 IB_WC_REM_ACCESS_ERR,
850 IB_WC_REM_OP_ERR,
851 IB_WC_RETRY_EXC_ERR,
852 IB_WC_RNR_RETRY_EXC_ERR,
853 IB_WC_LOC_RDD_VIOL_ERR,
854 IB_WC_REM_INV_RD_REQ_ERR,
855 IB_WC_REM_ABORT_ERR,
856 IB_WC_INV_EECN_ERR,
857 IB_WC_INV_EEC_STATE_ERR,
858 IB_WC_FATAL_ERR,
859 IB_WC_RESP_TIMEOUT_ERR,
860 IB_WC_GENERAL_ERR
861 };
862
863 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
864
865 enum ib_wc_opcode {
866 IB_WC_SEND,
867 IB_WC_RDMA_WRITE,
868 IB_WC_RDMA_READ,
869 IB_WC_COMP_SWAP,
870 IB_WC_FETCH_ADD,
871 IB_WC_LSO,
872 IB_WC_LOCAL_INV,
873 IB_WC_REG_MR,
874 IB_WC_MASKED_COMP_SWAP,
875 IB_WC_MASKED_FETCH_ADD,
876 /*
877 * Set value of IB_WC_RECV so consumers can test if a completion is a
878 * receive by testing (opcode & IB_WC_RECV).
879 */
880 IB_WC_RECV = 1 << 7,
881 IB_WC_RECV_RDMA_WITH_IMM,
882 IB_WC_DUMMY = -1, /* force enum signed */
883 };
884
885 enum ib_wc_flags {
886 IB_WC_GRH = 1,
887 IB_WC_WITH_IMM = (1<<1),
888 IB_WC_WITH_INVALIDATE = (1<<2),
889 IB_WC_IP_CSUM_OK = (1<<3),
890 IB_WC_WITH_SMAC = (1<<4),
891 IB_WC_WITH_VLAN = (1<<5),
892 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
893 };
894
895 struct ib_wc {
896 union {
897 u64 wr_id;
898 struct ib_cqe *wr_cqe;
899 };
900 enum ib_wc_status status;
901 enum ib_wc_opcode opcode;
902 u32 vendor_err;
903 u32 byte_len;
904 struct ib_qp *qp;
905 union {
906 __be32 imm_data;
907 u32 invalidate_rkey;
908 } ex;
909 u32 src_qp;
910 int wc_flags;
911 u16 pkey_index;
912 u16 slid;
913 u8 sl;
914 u8 dlid_path_bits;
915 u8 port_num; /* valid only for DR SMPs on switches */
916 u8 smac[ETH_ALEN];
917 u16 vlan_id;
918 u8 network_hdr_type;
919 };
920
921 enum ib_cq_notify_flags {
922 IB_CQ_SOLICITED = 1 << 0,
923 IB_CQ_NEXT_COMP = 1 << 1,
924 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
925 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
926 };
927
928 enum ib_srq_type {
929 IB_SRQT_BASIC,
930 IB_SRQT_XRC
931 };
932
933 enum ib_srq_attr_mask {
934 IB_SRQ_MAX_WR = 1 << 0,
935 IB_SRQ_LIMIT = 1 << 1,
936 };
937
938 struct ib_srq_attr {
939 u32 max_wr;
940 u32 max_sge;
941 u32 srq_limit;
942 };
943
944 struct ib_srq_init_attr {
945 void (*event_handler)(struct ib_event *, void *);
946 void *srq_context;
947 struct ib_srq_attr attr;
948 enum ib_srq_type srq_type;
949
950 union {
951 struct {
952 struct ib_xrcd *xrcd;
953 struct ib_cq *cq;
954 } xrc;
955 } ext;
956 };
957
958 struct ib_qp_cap {
959 u32 max_send_wr;
960 u32 max_recv_wr;
961 u32 max_send_sge;
962 u32 max_recv_sge;
963 u32 max_inline_data;
964
965 /*
966 * Maximum number of rdma_rw_ctx structures in flight at a time.
967 * ib_create_qp() will calculate the right amount of neededed WRs
968 * and MRs based on this.
969 */
970 u32 max_rdma_ctxs;
971 };
972
973 enum ib_sig_type {
974 IB_SIGNAL_ALL_WR,
975 IB_SIGNAL_REQ_WR
976 };
977
978 enum ib_qp_type {
979 /*
980 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
981 * here (and in that order) since the MAD layer uses them as
982 * indices into a 2-entry table.
983 */
984 IB_QPT_SMI,
985 IB_QPT_GSI,
986
987 IB_QPT_RC,
988 IB_QPT_UC,
989 IB_QPT_UD,
990 IB_QPT_RAW_IPV6,
991 IB_QPT_RAW_ETHERTYPE,
992 IB_QPT_RAW_PACKET = 8,
993 IB_QPT_XRC_INI = 9,
994 IB_QPT_XRC_TGT,
995 IB_QPT_MAX,
996 /* Reserve a range for qp types internal to the low level driver.
997 * These qp types will not be visible at the IB core layer, so the
998 * IB_QPT_MAX usages should not be affected in the core layer
999 */
1000 IB_QPT_RESERVED1 = 0x1000,
1001 IB_QPT_RESERVED2,
1002 IB_QPT_RESERVED3,
1003 IB_QPT_RESERVED4,
1004 IB_QPT_RESERVED5,
1005 IB_QPT_RESERVED6,
1006 IB_QPT_RESERVED7,
1007 IB_QPT_RESERVED8,
1008 IB_QPT_RESERVED9,
1009 IB_QPT_RESERVED10,
1010 };
1011
1012 enum ib_qp_create_flags {
1013 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1014 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
1015 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1016 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1017 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
1018 IB_QP_CREATE_NETIF_QP = 1 << 5,
1019 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
1020 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
1021 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
1022 /* reserve bits 26-31 for low level drivers' internal use */
1023 IB_QP_CREATE_RESERVED_START = 1 << 26,
1024 IB_QP_CREATE_RESERVED_END = 1 << 31,
1025 };
1026
1027 /*
1028 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1029 * callback to destroy the passed in QP.
1030 */
1031
1032 struct ib_qp_init_attr {
1033 void (*event_handler)(struct ib_event *, void *);
1034 void *qp_context;
1035 struct ib_cq *send_cq;
1036 struct ib_cq *recv_cq;
1037 struct ib_srq *srq;
1038 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1039 struct ib_qp_cap cap;
1040 enum ib_sig_type sq_sig_type;
1041 enum ib_qp_type qp_type;
1042 enum ib_qp_create_flags create_flags;
1043
1044 /*
1045 * Only needed for special QP types, or when using the RW API.
1046 */
1047 u8 port_num;
1048 struct ib_rwq_ind_table *rwq_ind_tbl;
1049 };
1050
1051 struct ib_qp_open_attr {
1052 void (*event_handler)(struct ib_event *, void *);
1053 void *qp_context;
1054 u32 qp_num;
1055 enum ib_qp_type qp_type;
1056 };
1057
1058 enum ib_rnr_timeout {
1059 IB_RNR_TIMER_655_36 = 0,
1060 IB_RNR_TIMER_000_01 = 1,
1061 IB_RNR_TIMER_000_02 = 2,
1062 IB_RNR_TIMER_000_03 = 3,
1063 IB_RNR_TIMER_000_04 = 4,
1064 IB_RNR_TIMER_000_06 = 5,
1065 IB_RNR_TIMER_000_08 = 6,
1066 IB_RNR_TIMER_000_12 = 7,
1067 IB_RNR_TIMER_000_16 = 8,
1068 IB_RNR_TIMER_000_24 = 9,
1069 IB_RNR_TIMER_000_32 = 10,
1070 IB_RNR_TIMER_000_48 = 11,
1071 IB_RNR_TIMER_000_64 = 12,
1072 IB_RNR_TIMER_000_96 = 13,
1073 IB_RNR_TIMER_001_28 = 14,
1074 IB_RNR_TIMER_001_92 = 15,
1075 IB_RNR_TIMER_002_56 = 16,
1076 IB_RNR_TIMER_003_84 = 17,
1077 IB_RNR_TIMER_005_12 = 18,
1078 IB_RNR_TIMER_007_68 = 19,
1079 IB_RNR_TIMER_010_24 = 20,
1080 IB_RNR_TIMER_015_36 = 21,
1081 IB_RNR_TIMER_020_48 = 22,
1082 IB_RNR_TIMER_030_72 = 23,
1083 IB_RNR_TIMER_040_96 = 24,
1084 IB_RNR_TIMER_061_44 = 25,
1085 IB_RNR_TIMER_081_92 = 26,
1086 IB_RNR_TIMER_122_88 = 27,
1087 IB_RNR_TIMER_163_84 = 28,
1088 IB_RNR_TIMER_245_76 = 29,
1089 IB_RNR_TIMER_327_68 = 30,
1090 IB_RNR_TIMER_491_52 = 31
1091 };
1092
1093 enum ib_qp_attr_mask {
1094 IB_QP_STATE = 1,
1095 IB_QP_CUR_STATE = (1<<1),
1096 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1097 IB_QP_ACCESS_FLAGS = (1<<3),
1098 IB_QP_PKEY_INDEX = (1<<4),
1099 IB_QP_PORT = (1<<5),
1100 IB_QP_QKEY = (1<<6),
1101 IB_QP_AV = (1<<7),
1102 IB_QP_PATH_MTU = (1<<8),
1103 IB_QP_TIMEOUT = (1<<9),
1104 IB_QP_RETRY_CNT = (1<<10),
1105 IB_QP_RNR_RETRY = (1<<11),
1106 IB_QP_RQ_PSN = (1<<12),
1107 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1108 IB_QP_ALT_PATH = (1<<14),
1109 IB_QP_MIN_RNR_TIMER = (1<<15),
1110 IB_QP_SQ_PSN = (1<<16),
1111 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1112 IB_QP_PATH_MIG_STATE = (1<<18),
1113 IB_QP_CAP = (1<<19),
1114 IB_QP_DEST_QPN = (1<<20),
1115 IB_QP_RESERVED1 = (1<<21),
1116 IB_QP_RESERVED2 = (1<<22),
1117 IB_QP_RESERVED3 = (1<<23),
1118 IB_QP_RESERVED4 = (1<<24),
1119 };
1120
1121 enum ib_qp_state {
1122 IB_QPS_RESET,
1123 IB_QPS_INIT,
1124 IB_QPS_RTR,
1125 IB_QPS_RTS,
1126 IB_QPS_SQD,
1127 IB_QPS_SQE,
1128 IB_QPS_ERR,
1129 IB_QPS_DUMMY = -1, /* force enum signed */
1130 };
1131
1132 enum ib_mig_state {
1133 IB_MIG_MIGRATED,
1134 IB_MIG_REARM,
1135 IB_MIG_ARMED
1136 };
1137
1138 enum ib_mw_type {
1139 IB_MW_TYPE_1 = 1,
1140 IB_MW_TYPE_2 = 2
1141 };
1142
1143 struct ib_qp_attr {
1144 enum ib_qp_state qp_state;
1145 enum ib_qp_state cur_qp_state;
1146 enum ib_mtu path_mtu;
1147 enum ib_mig_state path_mig_state;
1148 u32 qkey;
1149 u32 rq_psn;
1150 u32 sq_psn;
1151 u32 dest_qp_num;
1152 int qp_access_flags;
1153 struct ib_qp_cap cap;
1154 struct ib_ah_attr ah_attr;
1155 struct ib_ah_attr alt_ah_attr;
1156 u16 pkey_index;
1157 u16 alt_pkey_index;
1158 u8 en_sqd_async_notify;
1159 u8 sq_draining;
1160 u8 max_rd_atomic;
1161 u8 max_dest_rd_atomic;
1162 u8 min_rnr_timer;
1163 u8 port_num;
1164 u8 timeout;
1165 u8 retry_cnt;
1166 u8 rnr_retry;
1167 u8 alt_port_num;
1168 u8 alt_timeout;
1169 };
1170
1171 enum ib_wr_opcode {
1172 IB_WR_RDMA_WRITE,
1173 IB_WR_RDMA_WRITE_WITH_IMM,
1174 IB_WR_SEND,
1175 IB_WR_SEND_WITH_IMM,
1176 IB_WR_RDMA_READ,
1177 IB_WR_ATOMIC_CMP_AND_SWP,
1178 IB_WR_ATOMIC_FETCH_AND_ADD,
1179 IB_WR_LSO,
1180 IB_WR_SEND_WITH_INV,
1181 IB_WR_RDMA_READ_WITH_INV,
1182 IB_WR_LOCAL_INV,
1183 IB_WR_REG_MR,
1184 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1185 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1186 IB_WR_REG_SIG_MR,
1187 /* reserve values for low level drivers' internal use.
1188 * These values will not be used at all in the ib core layer.
1189 */
1190 IB_WR_RESERVED1 = 0xf0,
1191 IB_WR_RESERVED2,
1192 IB_WR_RESERVED3,
1193 IB_WR_RESERVED4,
1194 IB_WR_RESERVED5,
1195 IB_WR_RESERVED6,
1196 IB_WR_RESERVED7,
1197 IB_WR_RESERVED8,
1198 IB_WR_RESERVED9,
1199 IB_WR_RESERVED10,
1200 IB_WR_DUMMY = -1, /* force enum signed */
1201 };
1202
1203 enum ib_send_flags {
1204 IB_SEND_FENCE = 1,
1205 IB_SEND_SIGNALED = (1<<1),
1206 IB_SEND_SOLICITED = (1<<2),
1207 IB_SEND_INLINE = (1<<3),
1208 IB_SEND_IP_CSUM = (1<<4),
1209
1210 /* reserve bits 26-31 for low level drivers' internal use */
1211 IB_SEND_RESERVED_START = (1 << 26),
1212 IB_SEND_RESERVED_END = (1 << 31),
1213 };
1214
1215 struct ib_sge {
1216 u64 addr;
1217 u32 length;
1218 u32 lkey;
1219 };
1220
1221 struct ib_cqe {
1222 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1223 };
1224
1225 struct ib_send_wr {
1226 struct ib_send_wr *next;
1227 union {
1228 u64 wr_id;
1229 struct ib_cqe *wr_cqe;
1230 };
1231 struct ib_sge *sg_list;
1232 int num_sge;
1233 enum ib_wr_opcode opcode;
1234 int send_flags;
1235 union {
1236 __be32 imm_data;
1237 u32 invalidate_rkey;
1238 } ex;
1239 };
1240
1241 struct ib_rdma_wr {
1242 struct ib_send_wr wr;
1243 u64 remote_addr;
1244 u32 rkey;
1245 };
1246
rdma_wr(struct ib_send_wr * wr)1247 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1248 {
1249 return container_of(wr, struct ib_rdma_wr, wr);
1250 }
1251
1252 struct ib_atomic_wr {
1253 struct ib_send_wr wr;
1254 u64 remote_addr;
1255 u64 compare_add;
1256 u64 swap;
1257 u64 compare_add_mask;
1258 u64 swap_mask;
1259 u32 rkey;
1260 };
1261
atomic_wr(struct ib_send_wr * wr)1262 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1263 {
1264 return container_of(wr, struct ib_atomic_wr, wr);
1265 }
1266
1267 struct ib_ud_wr {
1268 struct ib_send_wr wr;
1269 struct ib_ah *ah;
1270 void *header;
1271 int hlen;
1272 int mss;
1273 u32 remote_qpn;
1274 u32 remote_qkey;
1275 u16 pkey_index; /* valid for GSI only */
1276 u8 port_num; /* valid for DR SMPs on switch only */
1277 };
1278
ud_wr(struct ib_send_wr * wr)1279 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1280 {
1281 return container_of(wr, struct ib_ud_wr, wr);
1282 }
1283
1284 struct ib_reg_wr {
1285 struct ib_send_wr wr;
1286 struct ib_mr *mr;
1287 u32 key;
1288 int access;
1289 };
1290
reg_wr(struct ib_send_wr * wr)1291 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1292 {
1293 return container_of(wr, struct ib_reg_wr, wr);
1294 }
1295
1296 struct ib_sig_handover_wr {
1297 struct ib_send_wr wr;
1298 struct ib_sig_attrs *sig_attrs;
1299 struct ib_mr *sig_mr;
1300 int access_flags;
1301 struct ib_sge *prot;
1302 };
1303
sig_handover_wr(struct ib_send_wr * wr)1304 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1305 {
1306 return container_of(wr, struct ib_sig_handover_wr, wr);
1307 }
1308
1309 struct ib_recv_wr {
1310 struct ib_recv_wr *next;
1311 union {
1312 u64 wr_id;
1313 struct ib_cqe *wr_cqe;
1314 };
1315 struct ib_sge *sg_list;
1316 int num_sge;
1317 };
1318
1319 enum ib_access_flags {
1320 IB_ACCESS_LOCAL_WRITE = 1,
1321 IB_ACCESS_REMOTE_WRITE = (1<<1),
1322 IB_ACCESS_REMOTE_READ = (1<<2),
1323 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1324 IB_ACCESS_MW_BIND = (1<<4),
1325 IB_ZERO_BASED = (1<<5),
1326 IB_ACCESS_ON_DEMAND = (1<<6),
1327 };
1328
1329 struct ib_phys_buf {
1330 u64 addr;
1331 u64 size;
1332 };
1333
1334 /*
1335 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1336 * are hidden here instead of a uapi header!
1337 */
1338 enum ib_mr_rereg_flags {
1339 IB_MR_REREG_TRANS = 1,
1340 IB_MR_REREG_PD = (1<<1),
1341 IB_MR_REREG_ACCESS = (1<<2),
1342 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1343 };
1344
1345 struct ib_fmr_attr {
1346 int max_pages;
1347 int max_maps;
1348 u8 page_shift;
1349 };
1350
1351 struct ib_umem;
1352
1353 struct ib_ucontext {
1354 struct ib_device *device;
1355 struct list_head pd_list;
1356 struct list_head mr_list;
1357 struct list_head mw_list;
1358 struct list_head cq_list;
1359 struct list_head qp_list;
1360 struct list_head srq_list;
1361 struct list_head ah_list;
1362 struct list_head xrcd_list;
1363 struct list_head rule_list;
1364 struct list_head wq_list;
1365 struct list_head rwq_ind_tbl_list;
1366 int closing;
1367
1368 pid_t tgid;
1369 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1370 struct rb_root umem_tree;
1371 /*
1372 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1373 * mmu notifiers registration.
1374 */
1375 struct rw_semaphore umem_rwsem;
1376 void (*invalidate_range)(struct ib_umem *umem,
1377 unsigned long start, unsigned long end);
1378
1379 struct mmu_notifier mn;
1380 atomic_t notifier_count;
1381 /* A list of umems that don't have private mmu notifier counters yet. */
1382 struct list_head no_private_counters;
1383 int odp_mrs_count;
1384 #endif
1385 };
1386
1387 struct ib_uobject {
1388 u64 user_handle; /* handle given to us by userspace */
1389 struct ib_ucontext *context; /* associated user context */
1390 void *object; /* containing object */
1391 struct list_head list; /* link to context's list */
1392 int id; /* index into kernel idr */
1393 struct kref ref;
1394 struct rw_semaphore mutex; /* protects .live */
1395 struct rcu_head rcu; /* kfree_rcu() overhead */
1396 int live;
1397 };
1398
1399 struct ib_udata {
1400 const void __user *inbuf;
1401 void __user *outbuf;
1402 size_t inlen;
1403 size_t outlen;
1404 };
1405
1406 struct ib_pd {
1407 u32 local_dma_lkey;
1408 u32 flags;
1409 struct ib_device *device;
1410 struct ib_uobject *uobject;
1411 atomic_t usecnt; /* count all resources */
1412
1413 u32 unsafe_global_rkey;
1414
1415 /*
1416 * Implementation details of the RDMA core, don't use in drivers:
1417 */
1418 struct ib_mr *__internal_mr;
1419 };
1420
1421 struct ib_xrcd {
1422 struct ib_device *device;
1423 atomic_t usecnt; /* count all exposed resources */
1424 struct inode *inode;
1425
1426 struct mutex tgt_qp_mutex;
1427 struct list_head tgt_qp_list;
1428 };
1429
1430 struct ib_ah {
1431 struct ib_device *device;
1432 struct ib_pd *pd;
1433 struct ib_uobject *uobject;
1434 };
1435
1436 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1437
1438 enum ib_poll_context {
1439 IB_POLL_DIRECT, /* caller context, no hw completions */
1440 IB_POLL_SOFTIRQ, /* poll from softirq context */
1441 IB_POLL_WORKQUEUE, /* poll from workqueue */
1442 };
1443
1444 struct ib_cq {
1445 struct ib_device *device;
1446 struct ib_uobject *uobject;
1447 ib_comp_handler comp_handler;
1448 void (*event_handler)(struct ib_event *, void *);
1449 void *cq_context;
1450 int cqe;
1451 atomic_t usecnt; /* count number of work queues */
1452 enum ib_poll_context poll_ctx;
1453 struct work_struct work;
1454 };
1455
1456 struct ib_srq {
1457 struct ib_device *device;
1458 struct ib_pd *pd;
1459 struct ib_uobject *uobject;
1460 void (*event_handler)(struct ib_event *, void *);
1461 void *srq_context;
1462 enum ib_srq_type srq_type;
1463 atomic_t usecnt;
1464
1465 union {
1466 struct {
1467 struct ib_xrcd *xrcd;
1468 struct ib_cq *cq;
1469 u32 srq_num;
1470 } xrc;
1471 } ext;
1472 };
1473
1474 enum ib_wq_type {
1475 IB_WQT_RQ
1476 };
1477
1478 enum ib_wq_state {
1479 IB_WQS_RESET,
1480 IB_WQS_RDY,
1481 IB_WQS_ERR
1482 };
1483
1484 struct ib_wq {
1485 struct ib_device *device;
1486 struct ib_uobject *uobject;
1487 void *wq_context;
1488 void (*event_handler)(struct ib_event *, void *);
1489 struct ib_pd *pd;
1490 struct ib_cq *cq;
1491 u32 wq_num;
1492 enum ib_wq_state state;
1493 enum ib_wq_type wq_type;
1494 atomic_t usecnt;
1495 };
1496
1497 struct ib_wq_init_attr {
1498 void *wq_context;
1499 enum ib_wq_type wq_type;
1500 u32 max_wr;
1501 u32 max_sge;
1502 struct ib_cq *cq;
1503 void (*event_handler)(struct ib_event *, void *);
1504 };
1505
1506 enum ib_wq_attr_mask {
1507 IB_WQ_STATE = 1 << 0,
1508 IB_WQ_CUR_STATE = 1 << 1,
1509 };
1510
1511 struct ib_wq_attr {
1512 enum ib_wq_state wq_state;
1513 enum ib_wq_state curr_wq_state;
1514 };
1515
1516 struct ib_rwq_ind_table {
1517 struct ib_device *device;
1518 struct ib_uobject *uobject;
1519 atomic_t usecnt;
1520 u32 ind_tbl_num;
1521 u32 log_ind_tbl_size;
1522 struct ib_wq **ind_tbl;
1523 };
1524
1525 struct ib_rwq_ind_table_init_attr {
1526 u32 log_ind_tbl_size;
1527 /* Each entry is a pointer to Receive Work Queue */
1528 struct ib_wq **ind_tbl;
1529 };
1530
1531 /*
1532 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1533 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1534 */
1535 struct ib_qp {
1536 struct ib_device *device;
1537 struct ib_pd *pd;
1538 struct ib_cq *send_cq;
1539 struct ib_cq *recv_cq;
1540 spinlock_t mr_lock;
1541 struct ib_srq *srq;
1542 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1543 struct list_head xrcd_list;
1544
1545 /* count times opened, mcast attaches, flow attaches */
1546 atomic_t usecnt;
1547 struct list_head open_list;
1548 struct ib_qp *real_qp;
1549 struct ib_uobject *uobject;
1550 void (*event_handler)(struct ib_event *, void *);
1551 void *qp_context;
1552 u32 qp_num;
1553 u32 max_write_sge;
1554 u32 max_read_sge;
1555 enum ib_qp_type qp_type;
1556 struct ib_rwq_ind_table *rwq_ind_tbl;
1557 };
1558
1559 struct ib_mr {
1560 struct ib_device *device;
1561 struct ib_pd *pd;
1562 u32 lkey;
1563 u32 rkey;
1564 u64 iova;
1565 u64 length;
1566 unsigned int page_size;
1567 bool need_inval;
1568 union {
1569 struct ib_uobject *uobject; /* user */
1570 struct list_head qp_entry; /* FR */
1571 };
1572 };
1573
1574 struct ib_mw {
1575 struct ib_device *device;
1576 struct ib_pd *pd;
1577 struct ib_uobject *uobject;
1578 u32 rkey;
1579 enum ib_mw_type type;
1580 };
1581
1582 struct ib_fmr {
1583 struct ib_device *device;
1584 struct ib_pd *pd;
1585 struct list_head list;
1586 u32 lkey;
1587 u32 rkey;
1588 };
1589
1590 /* Supported steering options */
1591 enum ib_flow_attr_type {
1592 /* steering according to rule specifications */
1593 IB_FLOW_ATTR_NORMAL = 0x0,
1594 /* default unicast and multicast rule -
1595 * receive all Eth traffic which isn't steered to any QP
1596 */
1597 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1598 /* default multicast rule -
1599 * receive all Eth multicast traffic which isn't steered to any QP
1600 */
1601 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1602 /* sniffer rule - receive all port traffic */
1603 IB_FLOW_ATTR_SNIFFER = 0x3
1604 };
1605
1606 /* Supported steering header types */
1607 enum ib_flow_spec_type {
1608 /* L2 headers*/
1609 IB_FLOW_SPEC_ETH = 0x20,
1610 IB_FLOW_SPEC_IB = 0x22,
1611 /* L3 header*/
1612 IB_FLOW_SPEC_IPV4 = 0x30,
1613 IB_FLOW_SPEC_IPV6 = 0x31,
1614 /* L4 headers*/
1615 IB_FLOW_SPEC_TCP = 0x40,
1616 IB_FLOW_SPEC_UDP = 0x41
1617 };
1618 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1619 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1620
1621 /* Flow steering rule priority is set according to it's domain.
1622 * Lower domain value means higher priority.
1623 */
1624 enum ib_flow_domain {
1625 IB_FLOW_DOMAIN_USER,
1626 IB_FLOW_DOMAIN_ETHTOOL,
1627 IB_FLOW_DOMAIN_RFS,
1628 IB_FLOW_DOMAIN_NIC,
1629 IB_FLOW_DOMAIN_NUM /* Must be last */
1630 };
1631
1632 enum ib_flow_flags {
1633 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1634 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1635 };
1636
1637 struct ib_flow_eth_filter {
1638 u8 dst_mac[6];
1639 u8 src_mac[6];
1640 __be16 ether_type;
1641 __be16 vlan_tag;
1642 /* Must be last */
1643 u8 real_sz[0];
1644 };
1645
1646 struct ib_flow_spec_eth {
1647 enum ib_flow_spec_type type;
1648 u16 size;
1649 struct ib_flow_eth_filter val;
1650 struct ib_flow_eth_filter mask;
1651 };
1652
1653 struct ib_flow_ib_filter {
1654 __be16 dlid;
1655 __u8 sl;
1656 /* Must be last */
1657 u8 real_sz[0];
1658 };
1659
1660 struct ib_flow_spec_ib {
1661 enum ib_flow_spec_type type;
1662 u16 size;
1663 struct ib_flow_ib_filter val;
1664 struct ib_flow_ib_filter mask;
1665 };
1666
1667 /* IPv4 header flags */
1668 enum ib_ipv4_flags {
1669 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1670 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1671 last have this flag set */
1672 };
1673
1674 struct ib_flow_ipv4_filter {
1675 __be32 src_ip;
1676 __be32 dst_ip;
1677 u8 proto;
1678 u8 tos;
1679 u8 ttl;
1680 u8 flags;
1681 /* Must be last */
1682 u8 real_sz[0];
1683 };
1684
1685 struct ib_flow_spec_ipv4 {
1686 enum ib_flow_spec_type type;
1687 u16 size;
1688 struct ib_flow_ipv4_filter val;
1689 struct ib_flow_ipv4_filter mask;
1690 };
1691
1692 struct ib_flow_ipv6_filter {
1693 u8 src_ip[16];
1694 u8 dst_ip[16];
1695 __be32 flow_label;
1696 u8 next_hdr;
1697 u8 traffic_class;
1698 u8 hop_limit;
1699 /* Must be last */
1700 u8 real_sz[0];
1701 };
1702
1703 struct ib_flow_spec_ipv6 {
1704 enum ib_flow_spec_type type;
1705 u16 size;
1706 struct ib_flow_ipv6_filter val;
1707 struct ib_flow_ipv6_filter mask;
1708 };
1709
1710 struct ib_flow_tcp_udp_filter {
1711 __be16 dst_port;
1712 __be16 src_port;
1713 /* Must be last */
1714 u8 real_sz[0];
1715 };
1716
1717 struct ib_flow_spec_tcp_udp {
1718 enum ib_flow_spec_type type;
1719 u16 size;
1720 struct ib_flow_tcp_udp_filter val;
1721 struct ib_flow_tcp_udp_filter mask;
1722 };
1723
1724 union ib_flow_spec {
1725 struct {
1726 enum ib_flow_spec_type type;
1727 u16 size;
1728 };
1729 struct ib_flow_spec_eth eth;
1730 struct ib_flow_spec_ib ib;
1731 struct ib_flow_spec_ipv4 ipv4;
1732 struct ib_flow_spec_tcp_udp tcp_udp;
1733 struct ib_flow_spec_ipv6 ipv6;
1734 };
1735
1736 struct ib_flow_attr {
1737 enum ib_flow_attr_type type;
1738 u16 size;
1739 u16 priority;
1740 u32 flags;
1741 u8 num_of_specs;
1742 u8 port;
1743 /* Following are the optional layers according to user request
1744 * struct ib_flow_spec_xxx
1745 * struct ib_flow_spec_yyy
1746 */
1747 };
1748
1749 struct ib_flow {
1750 struct ib_qp *qp;
1751 struct ib_uobject *uobject;
1752 };
1753
1754 struct ib_mad_hdr;
1755 struct ib_grh;
1756
1757 enum ib_process_mad_flags {
1758 IB_MAD_IGNORE_MKEY = 1,
1759 IB_MAD_IGNORE_BKEY = 2,
1760 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1761 };
1762
1763 enum ib_mad_result {
1764 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1765 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1766 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1767 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1768 };
1769
1770 #define IB_DEVICE_NAME_MAX 64
1771
1772 struct ib_cache {
1773 rwlock_t lock;
1774 struct ib_event_handler event_handler;
1775 struct ib_pkey_cache **pkey_cache;
1776 struct ib_gid_table **gid_cache;
1777 u8 *lmc_cache;
1778 };
1779
1780 struct ib_dma_mapping_ops {
1781 int (*mapping_error)(struct ib_device *dev,
1782 u64 dma_addr);
1783 u64 (*map_single)(struct ib_device *dev,
1784 void *ptr, size_t size,
1785 enum dma_data_direction direction);
1786 void (*unmap_single)(struct ib_device *dev,
1787 u64 addr, size_t size,
1788 enum dma_data_direction direction);
1789 u64 (*map_page)(struct ib_device *dev,
1790 struct page *page, unsigned long offset,
1791 size_t size,
1792 enum dma_data_direction direction);
1793 void (*unmap_page)(struct ib_device *dev,
1794 u64 addr, size_t size,
1795 enum dma_data_direction direction);
1796 int (*map_sg)(struct ib_device *dev,
1797 struct scatterlist *sg, int nents,
1798 enum dma_data_direction direction);
1799 void (*unmap_sg)(struct ib_device *dev,
1800 struct scatterlist *sg, int nents,
1801 enum dma_data_direction direction);
1802 int (*map_sg_attrs)(struct ib_device *dev,
1803 struct scatterlist *sg, int nents,
1804 enum dma_data_direction direction,
1805 struct dma_attrs *attrs);
1806 void (*unmap_sg_attrs)(struct ib_device *dev,
1807 struct scatterlist *sg, int nents,
1808 enum dma_data_direction direction,
1809 struct dma_attrs *attrs);
1810 void (*sync_single_for_cpu)(struct ib_device *dev,
1811 u64 dma_handle,
1812 size_t size,
1813 enum dma_data_direction dir);
1814 void (*sync_single_for_device)(struct ib_device *dev,
1815 u64 dma_handle,
1816 size_t size,
1817 enum dma_data_direction dir);
1818 void *(*alloc_coherent)(struct ib_device *dev,
1819 size_t size,
1820 u64 *dma_handle,
1821 gfp_t flag);
1822 void (*free_coherent)(struct ib_device *dev,
1823 size_t size, void *cpu_addr,
1824 u64 dma_handle);
1825 };
1826
1827 struct iw_cm_verbs;
1828
1829 struct ib_port_immutable {
1830 int pkey_tbl_len;
1831 int gid_tbl_len;
1832 u32 core_cap_flags;
1833 u32 max_mad_size;
1834 };
1835
1836 struct ib_device {
1837 struct device *dma_device;
1838
1839 char name[IB_DEVICE_NAME_MAX];
1840
1841 struct list_head event_handler_list;
1842 spinlock_t event_handler_lock;
1843
1844 spinlock_t client_data_lock;
1845 struct list_head core_list;
1846 /* Access to the client_data_list is protected by the client_data_lock
1847 * spinlock and the lists_rwsem read-write semaphore */
1848 struct list_head client_data_list;
1849
1850 struct ib_cache cache;
1851 /**
1852 * port_immutable is indexed by port number
1853 */
1854 struct ib_port_immutable *port_immutable;
1855
1856 int num_comp_vectors;
1857
1858 struct iw_cm_verbs *iwcm;
1859
1860 /**
1861 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1862 * driver initialized data. The struct is kfree()'ed by the sysfs
1863 * core when the device is removed. A lifespan of -1 in the return
1864 * struct tells the core to set a default lifespan.
1865 */
1866 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
1867 u8 port_num);
1868 /**
1869 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1870 * @index - The index in the value array we wish to have updated, or
1871 * num_counters if we want all stats updated
1872 * Return codes -
1873 * < 0 - Error, no counters updated
1874 * index - Updated the single counter pointed to by index
1875 * num_counters - Updated all counters (will reset the timestamp
1876 * and prevent further calls for lifespan milliseconds)
1877 * Drivers are allowed to update all counters in leiu of just the
1878 * one given in index at their option
1879 */
1880 int (*get_hw_stats)(struct ib_device *device,
1881 struct rdma_hw_stats *stats,
1882 u8 port, int index);
1883 int (*query_device)(struct ib_device *device,
1884 struct ib_device_attr *device_attr,
1885 struct ib_udata *udata);
1886 int (*query_port)(struct ib_device *device,
1887 u8 port_num,
1888 struct ib_port_attr *port_attr);
1889 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1890 u8 port_num);
1891 /* When calling get_netdev, the HW vendor's driver should return the
1892 * net device of device @device at port @port_num or NULL if such
1893 * a net device doesn't exist. The vendor driver should call dev_hold
1894 * on this net device. The HW vendor's device driver must guarantee
1895 * that this function returns NULL before the net device reaches
1896 * NETDEV_UNREGISTER_FINAL state.
1897 */
1898 struct net_device *(*get_netdev)(struct ib_device *device,
1899 u8 port_num);
1900 int (*query_gid)(struct ib_device *device,
1901 u8 port_num, int index,
1902 union ib_gid *gid);
1903 /* When calling add_gid, the HW vendor's driver should
1904 * add the gid of device @device at gid index @index of
1905 * port @port_num to be @gid. Meta-info of that gid (for example,
1906 * the network device related to this gid is available
1907 * at @attr. @context allows the HW vendor driver to store extra
1908 * information together with a GID entry. The HW vendor may allocate
1909 * memory to contain this information and store it in @context when a
1910 * new GID entry is written to. Params are consistent until the next
1911 * call of add_gid or delete_gid. The function should return 0 on
1912 * success or error otherwise. The function could be called
1913 * concurrently for different ports. This function is only called
1914 * when roce_gid_table is used.
1915 */
1916 int (*add_gid)(struct ib_device *device,
1917 u8 port_num,
1918 unsigned int index,
1919 const union ib_gid *gid,
1920 const struct ib_gid_attr *attr,
1921 void **context);
1922 /* When calling del_gid, the HW vendor's driver should delete the
1923 * gid of device @device at gid index @index of port @port_num.
1924 * Upon the deletion of a GID entry, the HW vendor must free any
1925 * allocated memory. The caller will clear @context afterwards.
1926 * This function is only called when roce_gid_table is used.
1927 */
1928 int (*del_gid)(struct ib_device *device,
1929 u8 port_num,
1930 unsigned int index,
1931 void **context);
1932 int (*query_pkey)(struct ib_device *device,
1933 u8 port_num, u16 index, u16 *pkey);
1934 int (*modify_device)(struct ib_device *device,
1935 int device_modify_mask,
1936 struct ib_device_modify *device_modify);
1937 int (*modify_port)(struct ib_device *device,
1938 u8 port_num, int port_modify_mask,
1939 struct ib_port_modify *port_modify);
1940 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1941 struct ib_udata *udata);
1942 int (*dealloc_ucontext)(struct ib_ucontext *context);
1943 int (*mmap)(struct ib_ucontext *context,
1944 struct vm_area_struct *vma);
1945 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1946 struct ib_ucontext *context,
1947 struct ib_udata *udata);
1948 int (*dealloc_pd)(struct ib_pd *pd);
1949 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1950 struct ib_ah_attr *ah_attr,
1951 struct ib_udata *udata);
1952 int (*modify_ah)(struct ib_ah *ah,
1953 struct ib_ah_attr *ah_attr);
1954 int (*query_ah)(struct ib_ah *ah,
1955 struct ib_ah_attr *ah_attr);
1956 int (*destroy_ah)(struct ib_ah *ah);
1957 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1958 struct ib_srq_init_attr *srq_init_attr,
1959 struct ib_udata *udata);
1960 int (*modify_srq)(struct ib_srq *srq,
1961 struct ib_srq_attr *srq_attr,
1962 enum ib_srq_attr_mask srq_attr_mask,
1963 struct ib_udata *udata);
1964 int (*query_srq)(struct ib_srq *srq,
1965 struct ib_srq_attr *srq_attr);
1966 int (*destroy_srq)(struct ib_srq *srq);
1967 int (*post_srq_recv)(struct ib_srq *srq,
1968 struct ib_recv_wr *recv_wr,
1969 struct ib_recv_wr **bad_recv_wr);
1970 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1971 struct ib_qp_init_attr *qp_init_attr,
1972 struct ib_udata *udata);
1973 int (*modify_qp)(struct ib_qp *qp,
1974 struct ib_qp_attr *qp_attr,
1975 int qp_attr_mask,
1976 struct ib_udata *udata);
1977 int (*query_qp)(struct ib_qp *qp,
1978 struct ib_qp_attr *qp_attr,
1979 int qp_attr_mask,
1980 struct ib_qp_init_attr *qp_init_attr);
1981 int (*destroy_qp)(struct ib_qp *qp);
1982 int (*post_send)(struct ib_qp *qp,
1983 struct ib_send_wr *send_wr,
1984 struct ib_send_wr **bad_send_wr);
1985 int (*post_recv)(struct ib_qp *qp,
1986 struct ib_recv_wr *recv_wr,
1987 struct ib_recv_wr **bad_recv_wr);
1988 struct ib_cq * (*create_cq)(struct ib_device *device,
1989 const struct ib_cq_init_attr *attr,
1990 struct ib_ucontext *context,
1991 struct ib_udata *udata);
1992 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1993 u16 cq_period);
1994 int (*destroy_cq)(struct ib_cq *cq);
1995 int (*resize_cq)(struct ib_cq *cq, int cqe,
1996 struct ib_udata *udata);
1997 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1998 struct ib_wc *wc);
1999 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2000 int (*req_notify_cq)(struct ib_cq *cq,
2001 enum ib_cq_notify_flags flags);
2002 int (*req_ncomp_notif)(struct ib_cq *cq,
2003 int wc_cnt);
2004 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
2005 int mr_access_flags);
2006 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
2007 struct ib_phys_buf *phys_buf_array,
2008 int num_phys_buf,
2009 int mr_access_flags,
2010 u64 *iova_start);
2011 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
2012 u64 start, u64 length,
2013 u64 virt_addr,
2014 int mr_access_flags,
2015 struct ib_udata *udata);
2016 int (*rereg_user_mr)(struct ib_mr *mr,
2017 int flags,
2018 u64 start, u64 length,
2019 u64 virt_addr,
2020 int mr_access_flags,
2021 struct ib_pd *pd,
2022 struct ib_udata *udata);
2023 int (*dereg_mr)(struct ib_mr *mr);
2024 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
2025 enum ib_mr_type mr_type,
2026 u32 max_num_sg);
2027 int (*map_mr_sg)(struct ib_mr *mr,
2028 struct scatterlist *sg,
2029 int sg_nents,
2030 unsigned int *sg_offset);
2031 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
2032 enum ib_mw_type type,
2033 struct ib_udata *udata);
2034 int (*dealloc_mw)(struct ib_mw *mw);
2035 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
2036 int mr_access_flags,
2037 struct ib_fmr_attr *fmr_attr);
2038 int (*map_phys_fmr)(struct ib_fmr *fmr,
2039 u64 *page_list, int list_len,
2040 u64 iova);
2041 int (*unmap_fmr)(struct list_head *fmr_list);
2042 int (*dealloc_fmr)(struct ib_fmr *fmr);
2043 int (*attach_mcast)(struct ib_qp *qp,
2044 union ib_gid *gid,
2045 u16 lid);
2046 int (*detach_mcast)(struct ib_qp *qp,
2047 union ib_gid *gid,
2048 u16 lid);
2049 int (*process_mad)(struct ib_device *device,
2050 int process_mad_flags,
2051 u8 port_num,
2052 const struct ib_wc *in_wc,
2053 const struct ib_grh *in_grh,
2054 const struct ib_mad_hdr *in_mad,
2055 size_t in_mad_size,
2056 struct ib_mad_hdr *out_mad,
2057 size_t *out_mad_size,
2058 u16 *out_mad_pkey_index);
2059 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
2060 struct ib_ucontext *ucontext,
2061 struct ib_udata *udata);
2062 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2063 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2064 struct ib_flow_attr
2065 *flow_attr,
2066 int domain);
2067 int (*destroy_flow)(struct ib_flow *flow_id);
2068 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2069 struct ib_mr_status *mr_status);
2070 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2071 void (*drain_rq)(struct ib_qp *qp);
2072 void (*drain_sq)(struct ib_qp *qp);
2073 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2074 int state);
2075 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2076 struct ifla_vf_info *ivf);
2077 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2078 struct ifla_vf_stats *stats);
2079 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2080 int type);
2081 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2082 struct ib_wq_init_attr *init_attr,
2083 struct ib_udata *udata);
2084 int (*destroy_wq)(struct ib_wq *wq);
2085 int (*modify_wq)(struct ib_wq *wq,
2086 struct ib_wq_attr *attr,
2087 u32 wq_attr_mask,
2088 struct ib_udata *udata);
2089 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2090 struct ib_rwq_ind_table_init_attr *init_attr,
2091 struct ib_udata *udata);
2092 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2093 struct ib_dma_mapping_ops *dma_ops;
2094
2095 struct module *owner;
2096 struct device dev;
2097 struct kobject *ports_parent;
2098 struct list_head port_list;
2099
2100 enum {
2101 IB_DEV_UNINITIALIZED,
2102 IB_DEV_REGISTERED,
2103 IB_DEV_UNREGISTERED
2104 } reg_state;
2105
2106 int uverbs_abi_ver;
2107 u64 uverbs_cmd_mask;
2108 u64 uverbs_ex_cmd_mask;
2109
2110 char node_desc[IB_DEVICE_NODE_DESC_MAX];
2111 __be64 node_guid;
2112 u32 local_dma_lkey;
2113 u16 is_switch:1;
2114 u8 node_type;
2115 u8 phys_port_cnt;
2116 struct ib_device_attr attrs;
2117 struct attribute_group *hw_stats_ag;
2118 struct rdma_hw_stats *hw_stats;
2119
2120 /**
2121 * The following mandatory functions are used only at device
2122 * registration. Keep functions such as these at the end of this
2123 * structure to avoid cache line misses when accessing struct ib_device
2124 * in fast paths.
2125 */
2126 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2127 void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2128 };
2129
2130 struct ib_client {
2131 char *name;
2132 void (*add) (struct ib_device *);
2133 void (*remove)(struct ib_device *, void *client_data);
2134
2135 /* Returns the net_dev belonging to this ib_client and matching the
2136 * given parameters.
2137 * @dev: An RDMA device that the net_dev use for communication.
2138 * @port: A physical port number on the RDMA device.
2139 * @pkey: P_Key that the net_dev uses if applicable.
2140 * @gid: A GID that the net_dev uses to communicate.
2141 * @addr: An IP address the net_dev is configured with.
2142 * @client_data: The device's client data set by ib_set_client_data().
2143 *
2144 * An ib_client that implements a net_dev on top of RDMA devices
2145 * (such as IP over IB) should implement this callback, allowing the
2146 * rdma_cm module to find the right net_dev for a given request.
2147 *
2148 * The caller is responsible for calling dev_put on the returned
2149 * netdev. */
2150 struct net_device *(*get_net_dev_by_params)(
2151 struct ib_device *dev,
2152 u8 port,
2153 u16 pkey,
2154 const union ib_gid *gid,
2155 const struct sockaddr *addr,
2156 void *client_data);
2157 struct list_head list;
2158 };
2159
2160 struct ib_device *ib_alloc_device(size_t size);
2161 void ib_dealloc_device(struct ib_device *device);
2162
2163 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2164
2165 int ib_register_device(struct ib_device *device,
2166 int (*port_callback)(struct ib_device *,
2167 u8, struct kobject *));
2168 void ib_unregister_device(struct ib_device *device);
2169
2170 int ib_register_client (struct ib_client *client);
2171 void ib_unregister_client(struct ib_client *client);
2172
2173 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2174 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2175 void *data);
2176
ib_copy_from_udata(void * dest,struct ib_udata * udata,size_t len)2177 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2178 {
2179 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2180 }
2181
ib_copy_to_udata(struct ib_udata * udata,void * src,size_t len)2182 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2183 {
2184 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2185 }
2186
ib_is_udata_cleared(struct ib_udata * udata,size_t offset,size_t len)2187 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2188 size_t offset,
2189 size_t len)
2190 {
2191 const void __user *p = (const char __user *)udata->inbuf + offset;
2192 bool ret;
2193 u8 *buf;
2194
2195 if (len > USHRT_MAX)
2196 return false;
2197
2198 buf = memdup_user(p, len);
2199 if (IS_ERR(buf))
2200 return false;
2201
2202 ret = !memchr_inv(buf, 0, len);
2203 kfree(buf);
2204 return ret;
2205 }
2206
2207 /**
2208 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2209 * contains all required attributes and no attributes not allowed for
2210 * the given QP state transition.
2211 * @cur_state: Current QP state
2212 * @next_state: Next QP state
2213 * @type: QP type
2214 * @mask: Mask of supplied QP attributes
2215 * @ll : link layer of port
2216 *
2217 * This function is a helper function that a low-level driver's
2218 * modify_qp method can use to validate the consumer's input. It
2219 * checks that cur_state and next_state are valid QP states, that a
2220 * transition from cur_state to next_state is allowed by the IB spec,
2221 * and that the attribute mask supplied is allowed for the transition.
2222 */
2223 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2224 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2225 enum rdma_link_layer ll);
2226
2227 int ib_register_event_handler (struct ib_event_handler *event_handler);
2228 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2229 void ib_dispatch_event(struct ib_event *event);
2230
2231 int ib_query_port(struct ib_device *device,
2232 u8 port_num, struct ib_port_attr *port_attr);
2233
2234 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2235 u8 port_num);
2236
2237 /**
2238 * rdma_cap_ib_switch - Check if the device is IB switch
2239 * @device: Device to check
2240 *
2241 * Device driver is responsible for setting is_switch bit on
2242 * in ib_device structure at init time.
2243 *
2244 * Return: true if the device is IB switch.
2245 */
rdma_cap_ib_switch(const struct ib_device * device)2246 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2247 {
2248 return device->is_switch;
2249 }
2250
2251 /**
2252 * rdma_start_port - Return the first valid port number for the device
2253 * specified
2254 *
2255 * @device: Device to be checked
2256 *
2257 * Return start port number
2258 */
rdma_start_port(const struct ib_device * device)2259 static inline u8 rdma_start_port(const struct ib_device *device)
2260 {
2261 return rdma_cap_ib_switch(device) ? 0 : 1;
2262 }
2263
2264 /**
2265 * rdma_end_port - Return the last valid port number for the device
2266 * specified
2267 *
2268 * @device: Device to be checked
2269 *
2270 * Return last port number
2271 */
rdma_end_port(const struct ib_device * device)2272 static inline u8 rdma_end_port(const struct ib_device *device)
2273 {
2274 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2275 }
2276
rdma_is_port_valid(const struct ib_device * device,unsigned int port)2277 static inline int rdma_is_port_valid(const struct ib_device *device,
2278 unsigned int port)
2279 {
2280 return (port >= rdma_start_port(device) &&
2281 port <= rdma_end_port(device));
2282 }
2283
rdma_protocol_ib(const struct ib_device * device,u8 port_num)2284 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2285 {
2286 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2287 }
2288
rdma_protocol_roce(const struct ib_device * device,u8 port_num)2289 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2290 {
2291 return device->port_immutable[port_num].core_cap_flags &
2292 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2293 }
2294
rdma_protocol_roce_udp_encap(const struct ib_device * device,u8 port_num)2295 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2296 {
2297 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2298 }
2299
rdma_protocol_roce_eth_encap(const struct ib_device * device,u8 port_num)2300 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2301 {
2302 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2303 }
2304
rdma_protocol_iwarp(const struct ib_device * device,u8 port_num)2305 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2306 {
2307 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2308 }
2309
rdma_ib_or_roce(const struct ib_device * device,u8 port_num)2310 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2311 {
2312 return rdma_protocol_ib(device, port_num) ||
2313 rdma_protocol_roce(device, port_num);
2314 }
2315
2316 /**
2317 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2318 * Management Datagrams.
2319 * @device: Device to check
2320 * @port_num: Port number to check
2321 *
2322 * Management Datagrams (MAD) are a required part of the InfiniBand
2323 * specification and are supported on all InfiniBand devices. A slightly
2324 * extended version are also supported on OPA interfaces.
2325 *
2326 * Return: true if the port supports sending/receiving of MAD packets.
2327 */
rdma_cap_ib_mad(const struct ib_device * device,u8 port_num)2328 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2329 {
2330 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2331 }
2332
2333 /**
2334 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2335 * Management Datagrams.
2336 * @device: Device to check
2337 * @port_num: Port number to check
2338 *
2339 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2340 * datagrams with their own versions. These OPA MADs share many but not all of
2341 * the characteristics of InfiniBand MADs.
2342 *
2343 * OPA MADs differ in the following ways:
2344 *
2345 * 1) MADs are variable size up to 2K
2346 * IBTA defined MADs remain fixed at 256 bytes
2347 * 2) OPA SMPs must carry valid PKeys
2348 * 3) OPA SMP packets are a different format
2349 *
2350 * Return: true if the port supports OPA MAD packet formats.
2351 */
rdma_cap_opa_mad(struct ib_device * device,u8 port_num)2352 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2353 {
2354 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2355 == RDMA_CORE_CAP_OPA_MAD;
2356 }
2357
2358 /**
2359 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2360 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2361 * @device: Device to check
2362 * @port_num: Port number to check
2363 *
2364 * Each InfiniBand node is required to provide a Subnet Management Agent
2365 * that the subnet manager can access. Prior to the fabric being fully
2366 * configured by the subnet manager, the SMA is accessed via a well known
2367 * interface called the Subnet Management Interface (SMI). This interface
2368 * uses directed route packets to communicate with the SM to get around the
2369 * chicken and egg problem of the SM needing to know what's on the fabric
2370 * in order to configure the fabric, and needing to configure the fabric in
2371 * order to send packets to the devices on the fabric. These directed
2372 * route packets do not need the fabric fully configured in order to reach
2373 * their destination. The SMI is the only method allowed to send
2374 * directed route packets on an InfiniBand fabric.
2375 *
2376 * Return: true if the port provides an SMI.
2377 */
rdma_cap_ib_smi(const struct ib_device * device,u8 port_num)2378 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2379 {
2380 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2381 }
2382
2383 /**
2384 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2385 * Communication Manager.
2386 * @device: Device to check
2387 * @port_num: Port number to check
2388 *
2389 * The InfiniBand Communication Manager is one of many pre-defined General
2390 * Service Agents (GSA) that are accessed via the General Service
2391 * Interface (GSI). It's role is to facilitate establishment of connections
2392 * between nodes as well as other management related tasks for established
2393 * connections.
2394 *
2395 * Return: true if the port supports an IB CM (this does not guarantee that
2396 * a CM is actually running however).
2397 */
rdma_cap_ib_cm(const struct ib_device * device,u8 port_num)2398 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2399 {
2400 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2401 }
2402
2403 /**
2404 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2405 * Communication Manager.
2406 * @device: Device to check
2407 * @port_num: Port number to check
2408 *
2409 * Similar to above, but specific to iWARP connections which have a different
2410 * managment protocol than InfiniBand.
2411 *
2412 * Return: true if the port supports an iWARP CM (this does not guarantee that
2413 * a CM is actually running however).
2414 */
rdma_cap_iw_cm(const struct ib_device * device,u8 port_num)2415 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2416 {
2417 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2418 }
2419
2420 /**
2421 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2422 * Subnet Administration.
2423 * @device: Device to check
2424 * @port_num: Port number to check
2425 *
2426 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2427 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2428 * fabrics, devices should resolve routes to other hosts by contacting the
2429 * SA to query the proper route.
2430 *
2431 * Return: true if the port should act as a client to the fabric Subnet
2432 * Administration interface. This does not imply that the SA service is
2433 * running locally.
2434 */
rdma_cap_ib_sa(const struct ib_device * device,u8 port_num)2435 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2436 {
2437 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2438 }
2439
2440 /**
2441 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2442 * Multicast.
2443 * @device: Device to check
2444 * @port_num: Port number to check
2445 *
2446 * InfiniBand multicast registration is more complex than normal IPv4 or
2447 * IPv6 multicast registration. Each Host Channel Adapter must register
2448 * with the Subnet Manager when it wishes to join a multicast group. It
2449 * should do so only once regardless of how many queue pairs it subscribes
2450 * to this group. And it should leave the group only after all queue pairs
2451 * attached to the group have been detached.
2452 *
2453 * Return: true if the port must undertake the additional adminstrative
2454 * overhead of registering/unregistering with the SM and tracking of the
2455 * total number of queue pairs attached to the multicast group.
2456 */
rdma_cap_ib_mcast(const struct ib_device * device,u8 port_num)2457 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2458 {
2459 return rdma_cap_ib_sa(device, port_num);
2460 }
2461
2462 /**
2463 * rdma_cap_af_ib - Check if the port of device has the capability
2464 * Native Infiniband Address.
2465 * @device: Device to check
2466 * @port_num: Port number to check
2467 *
2468 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2469 * GID. RoCE uses a different mechanism, but still generates a GID via
2470 * a prescribed mechanism and port specific data.
2471 *
2472 * Return: true if the port uses a GID address to identify devices on the
2473 * network.
2474 */
rdma_cap_af_ib(const struct ib_device * device,u8 port_num)2475 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2476 {
2477 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2478 }
2479
2480 /**
2481 * rdma_cap_eth_ah - Check if the port of device has the capability
2482 * Ethernet Address Handle.
2483 * @device: Device to check
2484 * @port_num: Port number to check
2485 *
2486 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2487 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2488 * port. Normally, packet headers are generated by the sending host
2489 * adapter, but when sending connectionless datagrams, we must manually
2490 * inject the proper headers for the fabric we are communicating over.
2491 *
2492 * Return: true if we are running as a RoCE port and must force the
2493 * addition of a Global Route Header built from our Ethernet Address
2494 * Handle into our header list for connectionless packets.
2495 */
rdma_cap_eth_ah(const struct ib_device * device,u8 port_num)2496 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2497 {
2498 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2499 }
2500
2501 /**
2502 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2503 *
2504 * @device: Device
2505 * @port_num: Port number
2506 *
2507 * This MAD size includes the MAD headers and MAD payload. No other headers
2508 * are included.
2509 *
2510 * Return the max MAD size required by the Port. Will return 0 if the port
2511 * does not support MADs
2512 */
rdma_max_mad_size(const struct ib_device * device,u8 port_num)2513 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2514 {
2515 return device->port_immutable[port_num].max_mad_size;
2516 }
2517
2518 /**
2519 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2520 * @device: Device to check
2521 * @port_num: Port number to check
2522 *
2523 * RoCE GID table mechanism manages the various GIDs for a device.
2524 *
2525 * NOTE: if allocating the port's GID table has failed, this call will still
2526 * return true, but any RoCE GID table API will fail.
2527 *
2528 * Return: true if the port uses RoCE GID table mechanism in order to manage
2529 * its GIDs.
2530 */
rdma_cap_roce_gid_table(const struct ib_device * device,u8 port_num)2531 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2532 u8 port_num)
2533 {
2534 return rdma_protocol_roce(device, port_num) &&
2535 device->add_gid && device->del_gid;
2536 }
2537
2538 /*
2539 * Check if the device supports READ W/ INVALIDATE.
2540 */
rdma_cap_read_inv(struct ib_device * dev,u32 port_num)2541 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2542 {
2543 /*
2544 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2545 * has support for it yet.
2546 */
2547 return rdma_protocol_iwarp(dev, port_num);
2548 }
2549
2550 int ib_query_gid(struct ib_device *device,
2551 u8 port_num, int index, union ib_gid *gid,
2552 struct ib_gid_attr *attr);
2553
2554 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2555 int state);
2556 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2557 struct ifla_vf_info *info);
2558 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2559 struct ifla_vf_stats *stats);
2560 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2561 int type);
2562
2563 int ib_query_pkey(struct ib_device *device,
2564 u8 port_num, u16 index, u16 *pkey);
2565
2566 int ib_modify_device(struct ib_device *device,
2567 int device_modify_mask,
2568 struct ib_device_modify *device_modify);
2569
2570 int ib_modify_port(struct ib_device *device,
2571 u8 port_num, int port_modify_mask,
2572 struct ib_port_modify *port_modify);
2573
2574 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2575 enum ib_gid_type gid_type, struct net_device *ndev,
2576 u8 *port_num, u16 *index);
2577
2578 int ib_find_pkey(struct ib_device *device,
2579 u8 port_num, u16 pkey, u16 *index);
2580
2581 enum ib_pd_flags {
2582 /*
2583 * Create a memory registration for all memory in the system and place
2584 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2585 * ULPs to avoid the overhead of dynamic MRs.
2586 *
2587 * This flag is generally considered unsafe and must only be used in
2588 * extremly trusted environments. Every use of it will log a warning
2589 * in the kernel log.
2590 */
2591 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
2592 };
2593
2594 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2595 const char *caller);
2596 #define ib_alloc_pd(device, flags) \
2597 __ib_alloc_pd((device), (flags), __func__)
2598 void ib_dealloc_pd(struct ib_pd *pd);
2599
2600 /**
2601 * ib_create_ah - Creates an address handle for the given address vector.
2602 * @pd: The protection domain associated with the address handle.
2603 * @ah_attr: The attributes of the address vector.
2604 *
2605 * The address handle is used to reference a local or global destination
2606 * in all UD QP post sends.
2607 */
2608 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2609
2610 /**
2611 * ib_init_ah_from_wc - Initializes address handle attributes from a
2612 * work completion.
2613 * @device: Device on which the received message arrived.
2614 * @port_num: Port on which the received message arrived.
2615 * @wc: Work completion associated with the received message.
2616 * @grh: References the received global route header. This parameter is
2617 * ignored unless the work completion indicates that the GRH is valid.
2618 * @ah_attr: Returned attributes that can be used when creating an address
2619 * handle for replying to the message.
2620 */
2621 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2622 const struct ib_wc *wc, const struct ib_grh *grh,
2623 struct ib_ah_attr *ah_attr);
2624
2625 /**
2626 * ib_create_ah_from_wc - Creates an address handle associated with the
2627 * sender of the specified work completion.
2628 * @pd: The protection domain associated with the address handle.
2629 * @wc: Work completion information associated with a received message.
2630 * @grh: References the received global route header. This parameter is
2631 * ignored unless the work completion indicates that the GRH is valid.
2632 * @port_num: The outbound port number to associate with the address.
2633 *
2634 * The address handle is used to reference a local or global destination
2635 * in all UD QP post sends.
2636 */
2637 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2638 const struct ib_grh *grh, u8 port_num);
2639
2640 /**
2641 * ib_modify_ah - Modifies the address vector associated with an address
2642 * handle.
2643 * @ah: The address handle to modify.
2644 * @ah_attr: The new address vector attributes to associate with the
2645 * address handle.
2646 */
2647 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2648
2649 /**
2650 * ib_query_ah - Queries the address vector associated with an address
2651 * handle.
2652 * @ah: The address handle to query.
2653 * @ah_attr: The address vector attributes associated with the address
2654 * handle.
2655 */
2656 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2657
2658 /**
2659 * ib_destroy_ah - Destroys an address handle.
2660 * @ah: The address handle to destroy.
2661 */
2662 int ib_destroy_ah(struct ib_ah *ah);
2663
2664 /**
2665 * ib_create_srq - Creates a SRQ associated with the specified protection
2666 * domain.
2667 * @pd: The protection domain associated with the SRQ.
2668 * @srq_init_attr: A list of initial attributes required to create the
2669 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2670 * the actual capabilities of the created SRQ.
2671 *
2672 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2673 * requested size of the SRQ, and set to the actual values allocated
2674 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2675 * will always be at least as large as the requested values.
2676 */
2677 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2678 struct ib_srq_init_attr *srq_init_attr);
2679
2680 /**
2681 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2682 * @srq: The SRQ to modify.
2683 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2684 * the current values of selected SRQ attributes are returned.
2685 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2686 * are being modified.
2687 *
2688 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2689 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2690 * the number of receives queued drops below the limit.
2691 */
2692 int ib_modify_srq(struct ib_srq *srq,
2693 struct ib_srq_attr *srq_attr,
2694 enum ib_srq_attr_mask srq_attr_mask);
2695
2696 /**
2697 * ib_query_srq - Returns the attribute list and current values for the
2698 * specified SRQ.
2699 * @srq: The SRQ to query.
2700 * @srq_attr: The attributes of the specified SRQ.
2701 */
2702 int ib_query_srq(struct ib_srq *srq,
2703 struct ib_srq_attr *srq_attr);
2704
2705 /**
2706 * ib_destroy_srq - Destroys the specified SRQ.
2707 * @srq: The SRQ to destroy.
2708 */
2709 int ib_destroy_srq(struct ib_srq *srq);
2710
2711 /**
2712 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2713 * @srq: The SRQ to post the work request on.
2714 * @recv_wr: A list of work requests to post on the receive queue.
2715 * @bad_recv_wr: On an immediate failure, this parameter will reference
2716 * the work request that failed to be posted on the QP.
2717 */
ib_post_srq_recv(struct ib_srq * srq,struct ib_recv_wr * recv_wr,struct ib_recv_wr ** bad_recv_wr)2718 static inline int ib_post_srq_recv(struct ib_srq *srq,
2719 struct ib_recv_wr *recv_wr,
2720 struct ib_recv_wr **bad_recv_wr)
2721 {
2722 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2723 }
2724
2725 /**
2726 * ib_create_qp - Creates a QP associated with the specified protection
2727 * domain.
2728 * @pd: The protection domain associated with the QP.
2729 * @qp_init_attr: A list of initial attributes required to create the
2730 * QP. If QP creation succeeds, then the attributes are updated to
2731 * the actual capabilities of the created QP.
2732 */
2733 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2734 struct ib_qp_init_attr *qp_init_attr);
2735
2736 /**
2737 * ib_modify_qp - Modifies the attributes for the specified QP and then
2738 * transitions the QP to the given state.
2739 * @qp: The QP to modify.
2740 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2741 * the current values of selected QP attributes are returned.
2742 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2743 * are being modified.
2744 */
2745 int ib_modify_qp(struct ib_qp *qp,
2746 struct ib_qp_attr *qp_attr,
2747 int qp_attr_mask);
2748
2749 /**
2750 * ib_query_qp - Returns the attribute list and current values for the
2751 * specified QP.
2752 * @qp: The QP to query.
2753 * @qp_attr: The attributes of the specified QP.
2754 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2755 * @qp_init_attr: Additional attributes of the selected QP.
2756 *
2757 * The qp_attr_mask may be used to limit the query to gathering only the
2758 * selected attributes.
2759 */
2760 int ib_query_qp(struct ib_qp *qp,
2761 struct ib_qp_attr *qp_attr,
2762 int qp_attr_mask,
2763 struct ib_qp_init_attr *qp_init_attr);
2764
2765 /**
2766 * ib_destroy_qp - Destroys the specified QP.
2767 * @qp: The QP to destroy.
2768 */
2769 int ib_destroy_qp(struct ib_qp *qp);
2770
2771 /**
2772 * ib_open_qp - Obtain a reference to an existing sharable QP.
2773 * @xrcd - XRC domain
2774 * @qp_open_attr: Attributes identifying the QP to open.
2775 *
2776 * Returns a reference to a sharable QP.
2777 */
2778 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2779 struct ib_qp_open_attr *qp_open_attr);
2780
2781 /**
2782 * ib_close_qp - Release an external reference to a QP.
2783 * @qp: The QP handle to release
2784 *
2785 * The opened QP handle is released by the caller. The underlying
2786 * shared QP is not destroyed until all internal references are released.
2787 */
2788 int ib_close_qp(struct ib_qp *qp);
2789
2790 /**
2791 * ib_post_send - Posts a list of work requests to the send queue of
2792 * the specified QP.
2793 * @qp: The QP to post the work request on.
2794 * @send_wr: A list of work requests to post on the send queue.
2795 * @bad_send_wr: On an immediate failure, this parameter will reference
2796 * the work request that failed to be posted on the QP.
2797 *
2798 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2799 * error is returned, the QP state shall not be affected,
2800 * ib_post_send() will return an immediate error after queueing any
2801 * earlier work requests in the list.
2802 */
ib_post_send(struct ib_qp * qp,struct ib_send_wr * send_wr,struct ib_send_wr ** bad_send_wr)2803 static inline int ib_post_send(struct ib_qp *qp,
2804 struct ib_send_wr *send_wr,
2805 struct ib_send_wr **bad_send_wr)
2806 {
2807 return qp->device->post_send(qp, send_wr, bad_send_wr);
2808 }
2809
2810 /**
2811 * ib_post_recv - Posts a list of work requests to the receive queue of
2812 * the specified QP.
2813 * @qp: The QP to post the work request on.
2814 * @recv_wr: A list of work requests to post on the receive queue.
2815 * @bad_recv_wr: On an immediate failure, this parameter will reference
2816 * the work request that failed to be posted on the QP.
2817 */
ib_post_recv(struct ib_qp * qp,struct ib_recv_wr * recv_wr,struct ib_recv_wr ** bad_recv_wr)2818 static inline int ib_post_recv(struct ib_qp *qp,
2819 struct ib_recv_wr *recv_wr,
2820 struct ib_recv_wr **bad_recv_wr)
2821 {
2822 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2823 }
2824
2825 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2826 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2827 void ib_free_cq(struct ib_cq *cq);
2828
2829 /**
2830 * ib_create_cq - Creates a CQ on the specified device.
2831 * @device: The device on which to create the CQ.
2832 * @comp_handler: A user-specified callback that is invoked when a
2833 * completion event occurs on the CQ.
2834 * @event_handler: A user-specified callback that is invoked when an
2835 * asynchronous event not associated with a completion occurs on the CQ.
2836 * @cq_context: Context associated with the CQ returned to the user via
2837 * the associated completion and event handlers.
2838 * @cq_attr: The attributes the CQ should be created upon.
2839 *
2840 * Users can examine the cq structure to determine the actual CQ size.
2841 */
2842 struct ib_cq *ib_create_cq(struct ib_device *device,
2843 ib_comp_handler comp_handler,
2844 void (*event_handler)(struct ib_event *, void *),
2845 void *cq_context,
2846 const struct ib_cq_init_attr *cq_attr);
2847
2848 /**
2849 * ib_resize_cq - Modifies the capacity of the CQ.
2850 * @cq: The CQ to resize.
2851 * @cqe: The minimum size of the CQ.
2852 *
2853 * Users can examine the cq structure to determine the actual CQ size.
2854 */
2855 int ib_resize_cq(struct ib_cq *cq, int cqe);
2856
2857 /**
2858 * ib_modify_cq - Modifies moderation params of the CQ
2859 * @cq: The CQ to modify.
2860 * @cq_count: number of CQEs that will trigger an event
2861 * @cq_period: max period of time in usec before triggering an event
2862 *
2863 */
2864 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2865
2866 /**
2867 * ib_destroy_cq - Destroys the specified CQ.
2868 * @cq: The CQ to destroy.
2869 */
2870 int ib_destroy_cq(struct ib_cq *cq);
2871
2872 /**
2873 * ib_poll_cq - poll a CQ for completion(s)
2874 * @cq:the CQ being polled
2875 * @num_entries:maximum number of completions to return
2876 * @wc:array of at least @num_entries &struct ib_wc where completions
2877 * will be returned
2878 *
2879 * Poll a CQ for (possibly multiple) completions. If the return value
2880 * is < 0, an error occurred. If the return value is >= 0, it is the
2881 * number of completions returned. If the return value is
2882 * non-negative and < num_entries, then the CQ was emptied.
2883 */
ib_poll_cq(struct ib_cq * cq,int num_entries,struct ib_wc * wc)2884 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2885 struct ib_wc *wc)
2886 {
2887 return cq->device->poll_cq(cq, num_entries, wc);
2888 }
2889
2890 /**
2891 * ib_peek_cq - Returns the number of unreaped completions currently
2892 * on the specified CQ.
2893 * @cq: The CQ to peek.
2894 * @wc_cnt: A minimum number of unreaped completions to check for.
2895 *
2896 * If the number of unreaped completions is greater than or equal to wc_cnt,
2897 * this function returns wc_cnt, otherwise, it returns the actual number of
2898 * unreaped completions.
2899 */
2900 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2901
2902 /**
2903 * ib_req_notify_cq - Request completion notification on a CQ.
2904 * @cq: The CQ to generate an event for.
2905 * @flags:
2906 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2907 * to request an event on the next solicited event or next work
2908 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2909 * may also be |ed in to request a hint about missed events, as
2910 * described below.
2911 *
2912 * Return Value:
2913 * < 0 means an error occurred while requesting notification
2914 * == 0 means notification was requested successfully, and if
2915 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2916 * were missed and it is safe to wait for another event. In
2917 * this case is it guaranteed that any work completions added
2918 * to the CQ since the last CQ poll will trigger a completion
2919 * notification event.
2920 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2921 * in. It means that the consumer must poll the CQ again to
2922 * make sure it is empty to avoid missing an event because of a
2923 * race between requesting notification and an entry being
2924 * added to the CQ. This return value means it is possible
2925 * (but not guaranteed) that a work completion has been added
2926 * to the CQ since the last poll without triggering a
2927 * completion notification event.
2928 */
ib_req_notify_cq(struct ib_cq * cq,enum ib_cq_notify_flags flags)2929 static inline int ib_req_notify_cq(struct ib_cq *cq,
2930 enum ib_cq_notify_flags flags)
2931 {
2932 return cq->device->req_notify_cq(cq, flags);
2933 }
2934
2935 /**
2936 * ib_req_ncomp_notif - Request completion notification when there are
2937 * at least the specified number of unreaped completions on the CQ.
2938 * @cq: The CQ to generate an event for.
2939 * @wc_cnt: The number of unreaped completions that should be on the
2940 * CQ before an event is generated.
2941 */
ib_req_ncomp_notif(struct ib_cq * cq,int wc_cnt)2942 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2943 {
2944 return cq->device->req_ncomp_notif ?
2945 cq->device->req_ncomp_notif(cq, wc_cnt) :
2946 -ENOSYS;
2947 }
2948
2949 /**
2950 * ib_dma_mapping_error - check a DMA addr for error
2951 * @dev: The device for which the dma_addr was created
2952 * @dma_addr: The DMA address to check
2953 */
ib_dma_mapping_error(struct ib_device * dev,u64 dma_addr)2954 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2955 {
2956 if (dev->dma_ops)
2957 return dev->dma_ops->mapping_error(dev, dma_addr);
2958 return dma_mapping_error(dev->dma_device, dma_addr);
2959 }
2960
2961 /**
2962 * ib_dma_map_single - Map a kernel virtual address to DMA address
2963 * @dev: The device for which the dma_addr is to be created
2964 * @cpu_addr: The kernel virtual address
2965 * @size: The size of the region in bytes
2966 * @direction: The direction of the DMA
2967 */
ib_dma_map_single(struct ib_device * dev,void * cpu_addr,size_t size,enum dma_data_direction direction)2968 static inline u64 ib_dma_map_single(struct ib_device *dev,
2969 void *cpu_addr, size_t size,
2970 enum dma_data_direction direction)
2971 {
2972 if (dev->dma_ops)
2973 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2974 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2975 }
2976
2977 /**
2978 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2979 * @dev: The device for which the DMA address was created
2980 * @addr: The DMA address
2981 * @size: The size of the region in bytes
2982 * @direction: The direction of the DMA
2983 */
ib_dma_unmap_single(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction direction)2984 static inline void ib_dma_unmap_single(struct ib_device *dev,
2985 u64 addr, size_t size,
2986 enum dma_data_direction direction)
2987 {
2988 if (dev->dma_ops)
2989 dev->dma_ops->unmap_single(dev, addr, size, direction);
2990 else
2991 dma_unmap_single(dev->dma_device, addr, size, direction);
2992 }
2993
ib_dma_map_single_attrs(struct ib_device * dev,void * cpu_addr,size_t size,enum dma_data_direction direction,struct dma_attrs * dma_attrs)2994 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2995 void *cpu_addr, size_t size,
2996 enum dma_data_direction direction,
2997 struct dma_attrs *dma_attrs)
2998 {
2999 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
3000 direction, dma_attrs);
3001 }
3002
ib_dma_unmap_single_attrs(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction direction,struct dma_attrs * dma_attrs)3003 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
3004 u64 addr, size_t size,
3005 enum dma_data_direction direction,
3006 struct dma_attrs *dma_attrs)
3007 {
3008 return dma_unmap_single_attrs(dev->dma_device, addr, size,
3009 direction, dma_attrs);
3010 }
3011
3012 /**
3013 * ib_dma_map_page - Map a physical page to DMA address
3014 * @dev: The device for which the dma_addr is to be created
3015 * @page: The page to be mapped
3016 * @offset: The offset within the page
3017 * @size: The size of the region in bytes
3018 * @direction: The direction of the DMA
3019 */
ib_dma_map_page(struct ib_device * dev,struct page * page,unsigned long offset,size_t size,enum dma_data_direction direction)3020 static inline u64 ib_dma_map_page(struct ib_device *dev,
3021 struct page *page,
3022 unsigned long offset,
3023 size_t size,
3024 enum dma_data_direction direction)
3025 {
3026 if (dev->dma_ops)
3027 return dev->dma_ops->map_page(dev, page, offset, size, direction);
3028 return dma_map_page(dev->dma_device, page, offset, size, direction);
3029 }
3030
3031 /**
3032 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3033 * @dev: The device for which the DMA address was created
3034 * @addr: The DMA address
3035 * @size: The size of the region in bytes
3036 * @direction: The direction of the DMA
3037 */
ib_dma_unmap_page(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction direction)3038 static inline void ib_dma_unmap_page(struct ib_device *dev,
3039 u64 addr, size_t size,
3040 enum dma_data_direction direction)
3041 {
3042 if (dev->dma_ops)
3043 dev->dma_ops->unmap_page(dev, addr, size, direction);
3044 else
3045 dma_unmap_page(dev->dma_device, addr, size, direction);
3046 }
3047
3048 /**
3049 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3050 * @dev: The device for which the DMA addresses are to be created
3051 * @sg: The array of scatter/gather entries
3052 * @nents: The number of scatter/gather entries
3053 * @direction: The direction of the DMA
3054 */
ib_dma_map_sg(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction)3055 static inline int ib_dma_map_sg(struct ib_device *dev,
3056 struct scatterlist *sg, int nents,
3057 enum dma_data_direction direction)
3058 {
3059 if (dev->dma_ops)
3060 return dev->dma_ops->map_sg(dev, sg, nents, direction);
3061 return dma_map_sg(dev->dma_device, sg, nents, direction);
3062 }
3063
3064 /**
3065 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3066 * @dev: The device for which the DMA addresses were created
3067 * @sg: The array of scatter/gather entries
3068 * @nents: The number of scatter/gather entries
3069 * @direction: The direction of the DMA
3070 */
ib_dma_unmap_sg(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction)3071 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3072 struct scatterlist *sg, int nents,
3073 enum dma_data_direction direction)
3074 {
3075 if (dev->dma_ops)
3076 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
3077 else
3078 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3079 }
3080
ib_dma_map_sg_attrs(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction,struct dma_attrs * dma_attrs)3081 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3082 struct scatterlist *sg, int nents,
3083 enum dma_data_direction direction,
3084 struct dma_attrs *dma_attrs)
3085 {
3086 if (dev->dma_ops)
3087 return dev->dma_ops->map_sg_attrs(dev, sg, nents, direction,
3088 dma_attrs);
3089 else
3090 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3091 dma_attrs);
3092 }
3093
ib_dma_unmap_sg_attrs(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction,struct dma_attrs * dma_attrs)3094 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3095 struct scatterlist *sg, int nents,
3096 enum dma_data_direction direction,
3097 struct dma_attrs *dma_attrs)
3098 {
3099 if (dev->dma_ops)
3100 return dev->dma_ops->unmap_sg_attrs(dev, sg, nents, direction,
3101 dma_attrs);
3102 else
3103 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
3104 dma_attrs);
3105 }
3106 /**
3107 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3108 * @dev: The device for which the DMA addresses were created
3109 * @sg: The scatter/gather entry
3110 *
3111 * Note: this function is obsolete. To do: change all occurrences of
3112 * ib_sg_dma_address() into sg_dma_address().
3113 */
ib_sg_dma_address(struct ib_device * dev,struct scatterlist * sg)3114 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3115 struct scatterlist *sg)
3116 {
3117 return sg_dma_address(sg);
3118 }
3119
3120 /**
3121 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3122 * @dev: The device for which the DMA addresses were created
3123 * @sg: The scatter/gather entry
3124 *
3125 * Note: this function is obsolete. To do: change all occurrences of
3126 * ib_sg_dma_len() into sg_dma_len().
3127 */
ib_sg_dma_len(struct ib_device * dev,struct scatterlist * sg)3128 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3129 struct scatterlist *sg)
3130 {
3131 return sg_dma_len(sg);
3132 }
3133
3134 /**
3135 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3136 * @dev: The device for which the DMA address was created
3137 * @addr: The DMA address
3138 * @size: The size of the region in bytes
3139 * @dir: The direction of the DMA
3140 */
ib_dma_sync_single_for_cpu(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction dir)3141 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3142 u64 addr,
3143 size_t size,
3144 enum dma_data_direction dir)
3145 {
3146 if (dev->dma_ops)
3147 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3148 else
3149 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3150 }
3151
3152 /**
3153 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3154 * @dev: The device for which the DMA address was created
3155 * @addr: The DMA address
3156 * @size: The size of the region in bytes
3157 * @dir: The direction of the DMA
3158 */
ib_dma_sync_single_for_device(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction dir)3159 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3160 u64 addr,
3161 size_t size,
3162 enum dma_data_direction dir)
3163 {
3164 if (dev->dma_ops)
3165 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3166 else
3167 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3168 }
3169
3170 /**
3171 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3172 * @dev: The device for which the DMA address is requested
3173 * @size: The size of the region to allocate in bytes
3174 * @dma_handle: A pointer for returning the DMA address of the region
3175 * @flag: memory allocator flags
3176 */
ib_dma_alloc_coherent(struct ib_device * dev,size_t size,u64 * dma_handle,gfp_t flag)3177 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3178 size_t size,
3179 u64 *dma_handle,
3180 gfp_t flag)
3181 {
3182 if (dev->dma_ops)
3183 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
3184 else {
3185 dma_addr_t handle;
3186 void *ret;
3187
3188 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3189 *dma_handle = handle;
3190 return ret;
3191 }
3192 }
3193
3194 /**
3195 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3196 * @dev: The device for which the DMA addresses were allocated
3197 * @size: The size of the region
3198 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3199 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3200 */
ib_dma_free_coherent(struct ib_device * dev,size_t size,void * cpu_addr,u64 dma_handle)3201 static inline void ib_dma_free_coherent(struct ib_device *dev,
3202 size_t size, void *cpu_addr,
3203 u64 dma_handle)
3204 {
3205 if (dev->dma_ops)
3206 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3207 else
3208 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3209 }
3210
3211 /**
3212 * ib_dereg_mr - Deregisters a memory region and removes it from the
3213 * HCA translation table.
3214 * @mr: The memory region to deregister.
3215 *
3216 * This function can fail, if the memory region has memory windows bound to it.
3217 */
3218 int ib_dereg_mr(struct ib_mr *mr);
3219
3220 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3221 enum ib_mr_type mr_type,
3222 u32 max_num_sg);
3223
3224 /**
3225 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3226 * R_Key and L_Key.
3227 * @mr - struct ib_mr pointer to be updated.
3228 * @newkey - new key to be used.
3229 */
ib_update_fast_reg_key(struct ib_mr * mr,u8 newkey)3230 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3231 {
3232 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3233 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3234 }
3235
3236 /**
3237 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3238 * for calculating a new rkey for type 2 memory windows.
3239 * @rkey - the rkey to increment.
3240 */
ib_inc_rkey(u32 rkey)3241 static inline u32 ib_inc_rkey(u32 rkey)
3242 {
3243 const u32 mask = 0x000000ff;
3244 return ((rkey + 1) & mask) | (rkey & ~mask);
3245 }
3246
3247 /**
3248 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3249 * @pd: The protection domain associated with the unmapped region.
3250 * @mr_access_flags: Specifies the memory access rights.
3251 * @fmr_attr: Attributes of the unmapped region.
3252 *
3253 * A fast memory region must be mapped before it can be used as part of
3254 * a work request.
3255 */
3256 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3257 int mr_access_flags,
3258 struct ib_fmr_attr *fmr_attr);
3259
3260 /**
3261 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3262 * @fmr: The fast memory region to associate with the pages.
3263 * @page_list: An array of physical pages to map to the fast memory region.
3264 * @list_len: The number of pages in page_list.
3265 * @iova: The I/O virtual address to use with the mapped region.
3266 */
ib_map_phys_fmr(struct ib_fmr * fmr,u64 * page_list,int list_len,u64 iova)3267 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3268 u64 *page_list, int list_len,
3269 u64 iova)
3270 {
3271 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3272 }
3273
3274 /**
3275 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3276 * @fmr_list: A linked list of fast memory regions to unmap.
3277 */
3278 int ib_unmap_fmr(struct list_head *fmr_list);
3279
3280 /**
3281 * ib_dealloc_fmr - Deallocates a fast memory region.
3282 * @fmr: The fast memory region to deallocate.
3283 */
3284 int ib_dealloc_fmr(struct ib_fmr *fmr);
3285
3286 /**
3287 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3288 * @qp: QP to attach to the multicast group. The QP must be type
3289 * IB_QPT_UD.
3290 * @gid: Multicast group GID.
3291 * @lid: Multicast group LID in host byte order.
3292 *
3293 * In order to send and receive multicast packets, subnet
3294 * administration must have created the multicast group and configured
3295 * the fabric appropriately. The port associated with the specified
3296 * QP must also be a member of the multicast group.
3297 */
3298 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3299
3300 /**
3301 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3302 * @qp: QP to detach from the multicast group.
3303 * @gid: Multicast group GID.
3304 * @lid: Multicast group LID in host byte order.
3305 */
3306 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3307
3308 /**
3309 * ib_alloc_xrcd - Allocates an XRC domain.
3310 * @device: The device on which to allocate the XRC domain.
3311 */
3312 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3313
3314 /**
3315 * ib_dealloc_xrcd - Deallocates an XRC domain.
3316 * @xrcd: The XRC domain to deallocate.
3317 */
3318 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3319
3320 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3321 struct ib_flow_attr *flow_attr, int domain);
3322 int ib_destroy_flow(struct ib_flow *flow_id);
3323
ib_check_mr_access(int flags)3324 static inline int ib_check_mr_access(int flags)
3325 {
3326 /*
3327 * Local write permission is required if remote write or
3328 * remote atomic permission is also requested.
3329 */
3330 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3331 !(flags & IB_ACCESS_LOCAL_WRITE))
3332 return -EINVAL;
3333
3334 return 0;
3335 }
3336
3337 /**
3338 * ib_check_mr_status: lightweight check of MR status.
3339 * This routine may provide status checks on a selected
3340 * ib_mr. first use is for signature status check.
3341 *
3342 * @mr: A memory region.
3343 * @check_mask: Bitmask of which checks to perform from
3344 * ib_mr_status_check enumeration.
3345 * @mr_status: The container of relevant status checks.
3346 * failed checks will be indicated in the status bitmask
3347 * and the relevant info shall be in the error item.
3348 */
3349 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3350 struct ib_mr_status *mr_status);
3351
3352 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3353 u16 pkey, const union ib_gid *gid,
3354 const struct sockaddr *addr);
3355 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3356 struct ib_wq_init_attr *init_attr);
3357 int ib_destroy_wq(struct ib_wq *wq);
3358 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3359 u32 wq_attr_mask);
3360 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3361 struct ib_rwq_ind_table_init_attr*
3362 wq_ind_table_init_attr);
3363 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3364
3365 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3366 unsigned int *sg_offset, unsigned int page_size);
3367
3368 static inline int
ib_map_mr_sg_zbva(struct ib_mr * mr,struct scatterlist * sg,int sg_nents,unsigned int * sg_offset,unsigned int page_size)3369 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3370 unsigned int *sg_offset, unsigned int page_size)
3371 {
3372 int n;
3373
3374 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3375 mr->iova = 0;
3376
3377 return n;
3378 }
3379
3380 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3381 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3382
3383 void ib_drain_rq(struct ib_qp *qp);
3384 void ib_drain_sq(struct ib_qp *qp);
3385 void ib_drain_qp(struct ib_qp *qp);
3386
3387 int ib_resolve_eth_dmac(struct ib_device *device,
3388 struct ib_ah_attr *ah_attr);
3389 #endif /* IB_VERBS_H */
3390