]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - include/rdma/ib_verbs.h
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
IB/core: Cut down single member ib_cache structure
[mirror_ubuntu-jammy-kernel.git] / include / rdma / ib_verbs.h
CommitLineData
1da177e4
LT		 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.
2a1d9b7f 7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
f7c6a7b5 8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
1da177e4
LT		 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.
1da177e4
LT		 37 */
38
39#if !defined(IB_VERBS_H)
40#define IB_VERBS_H
41
42#include <linux/types.h>
43#include <linux/device.h>
9b513090 44#include <linux/dma-mapping.h>
459d6e2a 45#include <linux/kref.h>
bfb3ea12
DB		 46#include <linux/list.h>
47#include <linux/rwsem.h>
f0626710 48#include <linux/workqueue.h>
14d3a3b2 49#include <linux/irq_poll.h>
dd5f03be 50#include <uapi/linux/if_ether.h>
c865f246
SK		 51#include <net/ipv6.h>
52#include <net/ip.h>
301a721e
MB		 53#include <linux/string.h>
54#include <linux/slab.h>
2fc77572 55#include <linux/netdevice.h>
01b67117 56#include <linux/refcount.h>
50174a7f 57#include <linux/if_link.h>
60063497 58#include <linux/atomic.h>
882214e2 59#include <linux/mmu_notifier.h>
7c0f6ba6 60#include <linux/uaccess.h>
43579b5f 61#include <linux/cgroup_rdma.h>
f6316032
LR		 62#include <linux/irqflags.h>
63#include <linux/preempt.h>
da662979 64#include <linux/dim.h>
ea6819e1 65#include <uapi/rdma/ib_user_verbs.h>
413d3347 66#include <rdma/rdma_counter.h>
02d8883f 67#include <rdma/restrack.h>
36b1e47f 68#include <rdma/signature.h>
0ede73bc 69#include <uapi/rdma/rdma_user_ioctl.h>
2eb9beae 70#include <uapi/rdma/ib_user_ioctl_verbs.h>
1da177e4 71
9abb0d1b
LR		 72#define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN
73
b5231b01
JG		 74struct ib_umem_odp;
75
f0626710 76extern struct workqueue_struct *ib_wq;
14d3a3b2 77extern struct workqueue_struct *ib_comp_wq;
f794809a 78extern struct workqueue_struct *ib_comp_unbound_wq;
f0626710 79
923abb9d
GP		 80__printf(3, 4) __cold
81void ibdev_printk(const char *level, const struct ib_device *ibdev,
82 const char *format, ...);
83__printf(2, 3) __cold
84void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...);
85__printf(2, 3) __cold
86void ibdev_alert(const struct ib_device *ibdev, const char *format, ...);
87__printf(2, 3) __cold
88void ibdev_crit(const struct ib_device *ibdev, const char *format, ...);
89__printf(2, 3) __cold
90void ibdev_err(const struct ib_device *ibdev, const char *format, ...);
91__printf(2, 3) __cold
92void ibdev_warn(const struct ib_device *ibdev, const char *format, ...);
93__printf(2, 3) __cold
94void ibdev_notice(const struct ib_device *ibdev, const char *format, ...);
95__printf(2, 3) __cold
96void ibdev_info(const struct ib_device *ibdev, const char *format, ...);
97
98#if defined(CONFIG_DYNAMIC_DEBUG)
99#define ibdev_dbg(__dev, format, args...) \
100 dynamic_ibdev_dbg(__dev, format, ##args)
923abb9d
GP		 101#else
102__printf(2, 3) __cold
103static inline
104void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {}
105#endif
106
05bb411a
GP		 107#define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \
108do { \
109 static DEFINE_RATELIMIT_STATE(_rs, \
110 DEFAULT_RATELIMIT_INTERVAL, \
111 DEFAULT_RATELIMIT_BURST); \
112 if (__ratelimit(&_rs)) \
113 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \
114} while (0)
115
116#define ibdev_emerg_ratelimited(ibdev, fmt, ...) \
117 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__)
118#define ibdev_alert_ratelimited(ibdev, fmt, ...) \
119 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__)
120#define ibdev_crit_ratelimited(ibdev, fmt, ...) \
121 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__)
122#define ibdev_err_ratelimited(ibdev, fmt, ...) \
123 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__)
124#define ibdev_warn_ratelimited(ibdev, fmt, ...) \
125 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__)
126#define ibdev_notice_ratelimited(ibdev, fmt, ...) \
127 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__)
128#define ibdev_info_ratelimited(ibdev, fmt, ...) \
129 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__)
130
131#if defined(CONFIG_DYNAMIC_DEBUG)
132/* descriptor check is first to prevent flooding with "callbacks suppressed" */
133#define ibdev_dbg_ratelimited(ibdev, fmt, ...) \
134do { \
135 static DEFINE_RATELIMIT_STATE(_rs, \
136 DEFAULT_RATELIMIT_INTERVAL, \
137 DEFAULT_RATELIMIT_BURST); \
138 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \
139 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \
140 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \
141 ##__VA_ARGS__); \
142} while (0)
143#else
144__printf(2, 3) __cold
145static inline
146void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {}
147#endif
148
1da177e4
LT		 149union ib_gid {
150 u8 raw[16];
151 struct {
97f52eb4
SH		 152 __be64 subnet_prefix;
153 __be64 interface_id;
1da177e4
LT		 154 } global;
155};
156
e26be1bf
MS		 157extern union ib_gid zgid;
158
b39ffa1d
MB		 159enum ib_gid_type {
160 /* If link layer is Ethernet, this is RoCE V1 */
161 IB_GID_TYPE_IB = 0,
162 IB_GID_TYPE_ROCE = 0,
7766a99f 163 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
b39ffa1d
MB		 164 IB_GID_TYPE_SIZE
165};
166
7ead4bcb 167#define ROCE_V2_UDP_DPORT 4791
03db3a2d 168struct ib_gid_attr {
943bd984 169 struct net_device __rcu *ndev;
598ff6ba 170 struct ib_device *device;
b150c386 171 union ib_gid gid;
598ff6ba
PP		 172 enum ib_gid_type gid_type;
173 u16 index;
174 u8 port_num;
03db3a2d
MB		 175};
176
a0c1b2a3
EC		 177enum {
178 /* set the local administered indication */
179 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
180};
181
07ebafba
TT		 182enum rdma_transport_type {
183 RDMA_TRANSPORT_IB,
180771a3 184 RDMA_TRANSPORT_IWARP,
248567f7 185 RDMA_TRANSPORT_USNIC,
f95be3d2
GP		 186 RDMA_TRANSPORT_USNIC_UDP,
187 RDMA_TRANSPORT_UNSPECIFIED,
07ebafba
TT		 188};
189
6b90a6d6
MW		 190enum rdma_protocol_type {
191 RDMA_PROTOCOL_IB,
192 RDMA_PROTOCOL_IBOE,
193 RDMA_PROTOCOL_IWARP,
194 RDMA_PROTOCOL_USNIC_UDP
195};
196
8385fd84 197__attribute_const__ enum rdma_transport_type
5d60c111 198rdma_node_get_transport(unsigned int node_type);
07ebafba 199
c865f246
SK		 200enum rdma_network_type {
201 RDMA_NETWORK_IB,
202 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
203 RDMA_NETWORK_IPV4,
204 RDMA_NETWORK_IPV6
205};
206
207static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
208{
209 if (network_type == RDMA_NETWORK_IPV4 ||
210 network_type == RDMA_NETWORK_IPV6)
211 return IB_GID_TYPE_ROCE_UDP_ENCAP;
212
213 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
214 return IB_GID_TYPE_IB;
215}
216
47ec3866
PP		 217static inline enum rdma_network_type
218rdma_gid_attr_network_type(const struct ib_gid_attr *attr)
c865f246 219{
47ec3866 220 if (attr->gid_type == IB_GID_TYPE_IB)
c865f246
SK		 221 return RDMA_NETWORK_IB;
222
47ec3866 223 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid))
c865f246
SK		 224 return RDMA_NETWORK_IPV4;
225 else
226 return RDMA_NETWORK_IPV6;
227}
228
a3f5adaf
EC		 229enum rdma_link_layer {
230 IB_LINK_LAYER_UNSPECIFIED,
231 IB_LINK_LAYER_INFINIBAND,
232 IB_LINK_LAYER_ETHERNET,
233};
234
1da177e4 235enum ib_device_cap_flags {
7ca0bc53
LR		 236 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
237 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
238 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
239 IB_DEVICE_RAW_MULTI = (1 << 3),
240 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
241 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
242 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
243 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
244 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
78b57f95 245 /* Not in use, former INIT_TYPE = (1 << 9),*/
7ca0bc53
LR		 246 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
247 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
248 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
249 IB_DEVICE_SRQ_RESIZE = (1 << 13),
250 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
b1adc714
CH		 251
252 /*
253 * This device supports a per-device lkey or stag that can be
254 * used without performing a memory registration for the local
255 * memory. Note that ULPs should never check this flag, but
256 * instead of use the local_dma_lkey flag in the ib_pd structure,
257 * which will always contain a usable lkey.
258 */
7ca0bc53 259 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
78b57f95 260 /* Reserved, old SEND_W_INV = (1 << 16),*/
7ca0bc53 261 IB_DEVICE_MEM_WINDOW = (1 << 17),
e0605d91
EC		 262 /*
263 * Devices should set IB_DEVICE_UD_IP_SUM if they support
264 * insertion of UDP and TCP checksum on outgoing UD IPoIB
265 * messages and can verify the validity of checksum for
266 * incoming messages. Setting this flag implies that the
267 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
268 */
7ca0bc53
LR		 269 IB_DEVICE_UD_IP_CSUM = (1 << 18),
270 IB_DEVICE_UD_TSO = (1 << 19),
271 IB_DEVICE_XRC = (1 << 20),
b1adc714
CH		 272
273 /*
274 * This device supports the IB "base memory management extension",
275 * which includes support for fast registrations (IB_WR_REG_MR,
276 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
277 * also be set by any iWarp device which must support FRs to comply
278 * to the iWarp verbs spec. iWarp devices also support the
279 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
280 * stag.
281 */
7ca0bc53
LR		 282 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
283 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
284 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
285 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
286 IB_DEVICE_RC_IP_CSUM = (1 << 25),
ebaaee25 287 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
7ca0bc53 288 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
8a06ce59
LR		 289 /*
290 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
291 * support execution of WQEs that involve synchronization
292 * of I/O operations with single completion queue managed
293 * by hardware.
294 */
78b57f95 295 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
7ca0bc53 296 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
c0a6cbb9 297 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30),
47355b3c 298 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
f5aa9159 299 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
c7e162a4 300 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
ebaaee25 301 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
c7e162a4 302 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
62e45949 303 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35),
e1d2e887
NO		 304 /* The device supports padding incoming writes to cacheline. */
305 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36),
3856ec4b 306 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37),
1b01d335
SG		 307};
308
1da177e4
LT		 309enum ib_atomic_cap {
310 IB_ATOMIC_NONE,
311 IB_ATOMIC_HCA,
312 IB_ATOMIC_GLOB
313};
314
860f10a7 315enum ib_odp_general_cap_bits {
25bf14d6
AK		 316 IB_ODP_SUPPORT = 1 << 0,
317 IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
860f10a7
SG		 318};
319
320enum ib_odp_transport_cap_bits {
321 IB_ODP_SUPPORT_SEND = 1 << 0,
322 IB_ODP_SUPPORT_RECV = 1 << 1,
323 IB_ODP_SUPPORT_WRITE = 1 << 2,
324 IB_ODP_SUPPORT_READ = 1 << 3,
325 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
da823342 326 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5,
860f10a7
SG		 327};
328
329struct ib_odp_caps {
330 uint64_t general_caps;
331 struct {
332 uint32_t rc_odp_caps;
333 uint32_t uc_odp_caps;
334 uint32_t ud_odp_caps;
52a72e2a 335 uint32_t xrc_odp_caps;
860f10a7
SG		 336 } per_transport_caps;
337};
338
ccf20562
YH		 339struct ib_rss_caps {
340 /* Corresponding bit will be set if qp type from
341 * 'enum ib_qp_type' is supported, e.g.
342 * supported_qpts |= 1 << IB_QPT_UD
343 */
344 u32 supported_qpts;
345 u32 max_rwq_indirection_tables;
346 u32 max_rwq_indirection_table_size;
347};
348
6938fc1e 349enum ib_tm_cap_flags {
89705e92
DG		 350 /* Support tag matching with rendezvous offload for RC transport */
351 IB_TM_CAP_RNDV_RC = 1 << 0,
6938fc1e
AK		 352};
353
78b1beb0 354struct ib_tm_caps {
6938fc1e
AK		 355 /* Max size of RNDV header */
356 u32 max_rndv_hdr_size;
357 /* Max number of entries in tag matching list */
358 u32 max_num_tags;
359 /* From enum ib_tm_cap_flags */
360 u32 flags;
361 /* Max number of outstanding list operations */
362 u32 max_ops;
363 /* Max number of SGE in tag matching entry */
364 u32 max_sge;
365};
366
bcf4c1ea
MB		 367struct ib_cq_init_attr {
368 unsigned int cqe;
a9018adf 369 u32 comp_vector;
bcf4c1ea
MB		 370 u32 flags;
371};
372
869ddcf8
YC		 373enum ib_cq_attr_mask {
374 IB_CQ_MODERATE = 1 << 0,
375};
376
18bd9072
YC		 377struct ib_cq_caps {
378 u16 max_cq_moderation_count;
379 u16 max_cq_moderation_period;
380};
381
be934cca
AL		 382struct ib_dm_mr_attr {
383 u64 length;
384 u64 offset;
385 u32 access_flags;
386};
387
bee76d7a
AL		 388struct ib_dm_alloc_attr {
389 u64 length;
390 u32 alignment;
391 u32 flags;
392};
393
1da177e4
LT		 394struct ib_device_attr {
395 u64 fw_ver;
97f52eb4 396 __be64 sys_image_guid;
1da177e4
LT		 397 u64 max_mr_size;
398 u64 page_size_cap;
399 u32 vendor_id;
400 u32 vendor_part_id;
401 u32 hw_ver;
402 int max_qp;
403 int max_qp_wr;
fb532d6a 404 u64 device_cap_flags;
33023fb8
SW		 405 int max_send_sge;
406 int max_recv_sge;
1da177e4
LT		 407 int max_sge_rd;
408 int max_cq;
409 int max_cqe;
410 int max_mr;
411 int max_pd;
412 int max_qp_rd_atom;
413 int max_ee_rd_atom;
414 int max_res_rd_atom;
415 int max_qp_init_rd_atom;
416 int max_ee_init_rd_atom;
417 enum ib_atomic_cap atomic_cap;
5e80ba8f 418 enum ib_atomic_cap masked_atomic_cap;
1da177e4
LT		 419 int max_ee;
420 int max_rdd;
421 int max_mw;
422 int max_raw_ipv6_qp;
423 int max_raw_ethy_qp;
424 int max_mcast_grp;
425 int max_mcast_qp_attach;
426 int max_total_mcast_qp_attach;
427 int max_ah;
428 int max_fmr;
429 int max_map_per_fmr;
430 int max_srq;
431 int max_srq_wr;
432 int max_srq_sge;
00f7ec36 433 unsigned int max_fast_reg_page_list_len;
62e3c379 434 unsigned int max_pi_fast_reg_page_list_len;
1da177e4
LT		 435 u16 max_pkeys;
436 u8 local_ca_ack_delay;
1b01d335
SG		 437 int sig_prot_cap;
438 int sig_guard_cap;
860f10a7 439 struct ib_odp_caps odp_caps;
24306dc6
MB		 440 uint64_t timestamp_mask;
441 uint64_t hca_core_clock; /* in KHZ */
ccf20562
YH		 442 struct ib_rss_caps rss_caps;
443 u32 max_wq_type_rq;
ebaaee25 444 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */
78b1beb0 445 struct ib_tm_caps tm_caps;
18bd9072 446 struct ib_cq_caps cq_caps;
1d8eeb9f 447 u64 max_dm_size;
00bd1439
YF		 448 /* Max entries for sgl for optimized performance per READ */
449 u32 max_sgl_rd;
1da177e4
LT		 450};
451
452enum ib_mtu {
453 IB_MTU_256 = 1,
454 IB_MTU_512 = 2,
455 IB_MTU_1024 = 3,
456 IB_MTU_2048 = 4,
457 IB_MTU_4096 = 5
458};
459
460static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
461{
462 switch (mtu) {
463 case IB_MTU_256: return 256;
464 case IB_MTU_512: return 512;
465 case IB_MTU_1024: return 1024;
466 case IB_MTU_2048: return 2048;
467 case IB_MTU_4096: return 4096;
468 default: return -1;
469 }
470}
471
d3f4aadd
AR		 472static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
473{
474 if (mtu >= 4096)
475 return IB_MTU_4096;
476 else if (mtu >= 2048)
477 return IB_MTU_2048;
478 else if (mtu >= 1024)
479 return IB_MTU_1024;
480 else if (mtu >= 512)
481 return IB_MTU_512;
482 else
483 return IB_MTU_256;
484}
485
1da177e4
LT		 486enum ib_port_state {
487 IB_PORT_NOP = 0,
488 IB_PORT_DOWN = 1,
489 IB_PORT_INIT = 2,
490 IB_PORT_ARMED = 3,
491 IB_PORT_ACTIVE = 4,
492 IB_PORT_ACTIVE_DEFER = 5
493};
494
72a7720f
KH		 495enum ib_port_phys_state {
496 IB_PORT_PHYS_STATE_SLEEP = 1,
497 IB_PORT_PHYS_STATE_POLLING = 2,
498 IB_PORT_PHYS_STATE_DISABLED = 3,
499 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4,
500 IB_PORT_PHYS_STATE_LINK_UP = 5,
501 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6,
502 IB_PORT_PHYS_STATE_PHY_TEST = 7,
503};
504
1da177e4
LT		 505enum ib_port_width {
506 IB_WIDTH_1X = 1,
dbabf685 507 IB_WIDTH_2X = 16,
1da177e4
LT		 508 IB_WIDTH_4X = 2,
509 IB_WIDTH_8X = 4,
510 IB_WIDTH_12X = 8
511};
512
513static inline int ib_width_enum_to_int(enum ib_port_width width)
514{
515 switch (width) {
516 case IB_WIDTH_1X: return 1;
dbabf685 517 case IB_WIDTH_2X: return 2;
1da177e4
LT		 518 case IB_WIDTH_4X: return 4;
519 case IB_WIDTH_8X: return 8;
520 case IB_WIDTH_12X: return 12;
521 default: return -1;
522 }
523}
524
2e96691c
OG		 525enum ib_port_speed {
526 IB_SPEED_SDR = 1,
527 IB_SPEED_DDR = 2,
528 IB_SPEED_QDR = 4,
529 IB_SPEED_FDR10 = 8,
530 IB_SPEED_FDR = 16,
12113a35
NO		 531 IB_SPEED_EDR = 32,
532 IB_SPEED_HDR = 64
2e96691c
OG		 533};
534
b40f4757
CL		 535/**
536 * struct rdma_hw_stats
e945130b
MB		 537 * @lock - Mutex to protect parallel write access to lifespan and values
538 * of counters, which are 64bits and not guaranteeed to be written
539 * atomicaly on 32bits systems.
b40f4757
CL		 540 * @timestamp - Used by the core code to track when the last update was
541 * @lifespan - Used by the core code to determine how old the counters
542 * should be before being updated again. Stored in jiffies, defaults
543 * to 10 milliseconds, drivers can override the default be specifying
544 * their own value during their allocation routine.
545 * @name - Array of pointers to static names used for the counters in
546 * directory.
547 * @num_counters - How many hardware counters there are. If name is
548 * shorter than this number, a kernel oops will result. Driver authors
549 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
550 * in their code to prevent this.
551 * @value - Array of u64 counters that are accessed by the sysfs code and
552 * filled in by the drivers get_stats routine
553 */
554struct rdma_hw_stats {
e945130b 555 struct mutex lock; /* Protect lifespan and values[] */
b40f4757
CL		 556 unsigned long timestamp;
557 unsigned long lifespan;
558 const char * const *names;
559 int num_counters;
560 u64 value[];
7f624d02
SW		 561};
562
b40f4757
CL		 563#define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
564/**
565 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
566 * for drivers.
567 * @names - Array of static const char *
568 * @num_counters - How many elements in array
569 * @lifespan - How many milliseconds between updates
570 */
571static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
572 const char * const *names, int num_counters,
573 unsigned long lifespan)
574{
575 struct rdma_hw_stats *stats;
576
577 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
578 GFP_KERNEL);
579 if (!stats)
580 return NULL;
581 stats->names = names;
582 stats->num_counters = num_counters;
583 stats->lifespan = msecs_to_jiffies(lifespan);
584
585 return stats;
586}
587
588
f9b22e35
IW		 589/* Define bits for the various functionality this port needs to be supported by
590 * the core.
591 */
592/* Management 0x00000FFF */
593#define RDMA_CORE_CAP_IB_MAD 0x00000001
594#define RDMA_CORE_CAP_IB_SMI 0x00000002
595#define RDMA_CORE_CAP_IB_CM 0x00000004
596#define RDMA_CORE_CAP_IW_CM 0x00000008
597#define RDMA_CORE_CAP_IB_SA 0x00000010
65995fee 598#define RDMA_CORE_CAP_OPA_MAD 0x00000020
f9b22e35
IW		 599
600/* Address format 0x000FF000 */
601#define RDMA_CORE_CAP_AF_IB 0x00001000
602#define RDMA_CORE_CAP_ETH_AH 0x00002000
94d595c5 603#define RDMA_CORE_CAP_OPA_AH 0x00004000
b02289b3 604#define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000
f9b22e35
IW		 605
606/* Protocol 0xFFF00000 */
607#define RDMA_CORE_CAP_PROT_IB 0x00100000
608#define RDMA_CORE_CAP_PROT_ROCE 0x00200000
609#define RDMA_CORE_CAP_PROT_IWARP 0x00400000
7766a99f 610#define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
aa773bd4 611#define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000
ce1e055f 612#define RDMA_CORE_CAP_PROT_USNIC 0x02000000
f9b22e35 613
b02289b3
AK		 614#define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \
615 | RDMA_CORE_CAP_PROT_ROCE \
616 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP)
617
f9b22e35
IW		 618#define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
619 | RDMA_CORE_CAP_IB_MAD \
620 | RDMA_CORE_CAP_IB_SMI \
621 | RDMA_CORE_CAP_IB_CM \
622 | RDMA_CORE_CAP_IB_SA \
623 | RDMA_CORE_CAP_AF_IB)
624#define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
625 | RDMA_CORE_CAP_IB_MAD \
626 | RDMA_CORE_CAP_IB_CM \
f9b22e35
IW		 627 | RDMA_CORE_CAP_AF_IB \
628 | RDMA_CORE_CAP_ETH_AH)
7766a99f
MB		 629#define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
630 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
631 | RDMA_CORE_CAP_IB_MAD \
632 | RDMA_CORE_CAP_IB_CM \
633 | RDMA_CORE_CAP_AF_IB \
634 | RDMA_CORE_CAP_ETH_AH)
f9b22e35
IW		 635#define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
636 | RDMA_CORE_CAP_IW_CM)
65995fee
IW		 637#define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
638 | RDMA_CORE_CAP_OPA_MAD)
f9b22e35 639
aa773bd4
OG		 640#define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET)
641
ce1e055f
OG		 642#define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC)
643
1da177e4 644struct ib_port_attr {
fad61ad4 645 u64 subnet_prefix;
1da177e4
LT		 646 enum ib_port_state state;
647 enum ib_mtu max_mtu;
648 enum ib_mtu active_mtu;
649 int gid_tbl_len;
2f944c0f
JG		 650 unsigned int ip_gids:1;
651 /* This is the value from PortInfo CapabilityMask, defined by IBA */
1da177e4
LT		 652 u32 port_cap_flags;
653 u32 max_msg_sz;
654 u32 bad_pkey_cntr;
655 u32 qkey_viol_cntr;
656 u16 pkey_tbl_len;
db58540b 657 u32 sm_lid;
582faf31 658 u32 lid;
1da177e4
LT		 659 u8 lmc;
660 u8 max_vl_num;
661 u8 sm_sl;
662 u8 subnet_timeout;
663 u8 init_type_reply;
664 u8 active_width;
665 u8 active_speed;
666 u8 phys_state;
1e8f43b7 667 u16 port_cap_flags2;
1da177e4
LT		 668};
669
670enum ib_device_modify_flags {
c5bcbbb9
RD		 671 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
672 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
1da177e4
LT		 673};
674
bd99fdea
YS		 675#define IB_DEVICE_NODE_DESC_MAX 64
676
1da177e4
LT		 677struct ib_device_modify {
678 u64 sys_image_guid;
bd99fdea 679 char node_desc[IB_DEVICE_NODE_DESC_MAX];
1da177e4
LT		 680};
681
682enum ib_port_modify_flags {
683 IB_PORT_SHUTDOWN = 1,
684 IB_PORT_INIT_TYPE = (1<<2),
cb49366f
VN		 685 IB_PORT_RESET_QKEY_CNTR = (1<<3),
686 IB_PORT_OPA_MASK_CHG = (1<<4)
1da177e4
LT		 687};
688
689struct ib_port_modify {
690 u32 set_port_cap_mask;
691 u32 clr_port_cap_mask;
692 u8 init_type;
693};
694
695enum ib_event_type {
696 IB_EVENT_CQ_ERR,
697 IB_EVENT_QP_FATAL,
698 IB_EVENT_QP_REQ_ERR,
699 IB_EVENT_QP_ACCESS_ERR,
700 IB_EVENT_COMM_EST,
701 IB_EVENT_SQ_DRAINED,
702 IB_EVENT_PATH_MIG,
703 IB_EVENT_PATH_MIG_ERR,
704 IB_EVENT_DEVICE_FATAL,
705 IB_EVENT_PORT_ACTIVE,
706 IB_EVENT_PORT_ERR,
707 IB_EVENT_LID_CHANGE,
708 IB_EVENT_PKEY_CHANGE,
d41fcc67
RD		 709 IB_EVENT_SM_CHANGE,
710 IB_EVENT_SRQ_ERR,
711 IB_EVENT_SRQ_LIMIT_REACHED,
63942c9a 712 IB_EVENT_QP_LAST_WQE_REACHED,
761d90ed
OG		 713 IB_EVENT_CLIENT_REREGISTER,
714 IB_EVENT_GID_CHANGE,
f213c052 715 IB_EVENT_WQ_FATAL,
1da177e4
LT		 716};
717
db7489e0 718const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
2b1b5b60 719
1da177e4
LT		 720struct ib_event {
721 struct ib_device *device;
722 union {
723 struct ib_cq *cq;
724 struct ib_qp *qp;
d41fcc67 725 struct ib_srq *srq;
f213c052 726 struct ib_wq *wq;
1da177e4
LT		 727 u8 port_num;
728 } element;
729 enum ib_event_type event;
730};
731
732struct ib_event_handler {
733 struct ib_device *device;
734 void (*handler)(struct ib_event_handler *, struct ib_event *);
735 struct list_head list;
736};
737
738#define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
739 do { \
740 (_ptr)->device = _device; \
741 (_ptr)->handler = _handler; \
742 INIT_LIST_HEAD(&(_ptr)->list); \
743 } while (0)
744
745struct ib_global_route {
8d9ec9ad 746 const struct ib_gid_attr *sgid_attr;
1da177e4
LT		 747 union ib_gid dgid;
748 u32 flow_label;
749 u8 sgid_index;
750 u8 hop_limit;
751 u8 traffic_class;
752};
753
513789ed 754struct ib_grh {
97f52eb4
SH		 755 __be32 version_tclass_flow;
756 __be16 paylen;
513789ed
HR		 757 u8 next_hdr;
758 u8 hop_limit;
759 union ib_gid sgid;
760 union ib_gid dgid;
761};
762
c865f246
SK		 763union rdma_network_hdr {
764 struct ib_grh ibgrh;
765 struct {
766 /* The IB spec states that if it's IPv4, the header
767 * is located in the last 20 bytes of the header.
768 */
769 u8 reserved[20];
770 struct iphdr roce4grh;
771 };
772};
773
7dafbab3
DH		 774#define IB_QPN_MASK 0xFFFFFF
775
1da177e4
LT		 776enum {
777 IB_MULTICAST_QPN = 0xffffff
778};
779
f3a7c66b 780#define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
b4e64397 781#define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
97f52eb4 782
1da177e4
LT		 783enum ib_ah_flags {
784 IB_AH_GRH = 1
785};
786
bf6a9e31
JM		 787enum ib_rate {
788 IB_RATE_PORT_CURRENT = 0,
789 IB_RATE_2_5_GBPS = 2,
790 IB_RATE_5_GBPS = 5,
791 IB_RATE_10_GBPS = 3,
792 IB_RATE_20_GBPS = 6,
793 IB_RATE_30_GBPS = 4,
794 IB_RATE_40_GBPS = 7,
795 IB_RATE_60_GBPS = 8,
796 IB_RATE_80_GBPS = 9,
71eeba16
MA		 797 IB_RATE_120_GBPS = 10,
798 IB_RATE_14_GBPS = 11,
799 IB_RATE_56_GBPS = 12,
800 IB_RATE_112_GBPS = 13,
801 IB_RATE_168_GBPS = 14,
802 IB_RATE_25_GBPS = 15,
803 IB_RATE_100_GBPS = 16,
804 IB_RATE_200_GBPS = 17,
a5a5d199
MG		 805 IB_RATE_300_GBPS = 18,
806 IB_RATE_28_GBPS = 19,
807 IB_RATE_50_GBPS = 20,
808 IB_RATE_400_GBPS = 21,
809 IB_RATE_600_GBPS = 22,
bf6a9e31
JM		 810};
811
812/**
813 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
814 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
815 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
816 * @rate: rate to convert.
817 */
8385fd84 818__attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
bf6a9e31 819
71eeba16
MA		 820/**
821 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
822 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
823 * @rate: rate to convert.
824 */
8385fd84 825__attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
71eeba16 826
17cd3a2d
SG		 827
828/**
9bee178b
SG		 829 * enum ib_mr_type - memory region type
830 * @IB_MR_TYPE_MEM_REG: memory region that is used for
831 * normal registration
f5aa9159
SG		 832 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
833 * register any arbitrary sg lists (without
834 * the normal mr constraints - see
835 * ib_map_mr_sg)
a0bc099a
MG		 836 * @IB_MR_TYPE_DM: memory region that is used for device
837 * memory registration
838 * @IB_MR_TYPE_USER: memory region that is used for the user-space
839 * application
840 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations
841 * without address translations (VA=PA)
26bc7eae
IR		 842 * @IB_MR_TYPE_INTEGRITY: memory region that is used for
843 * data integrity operations
17cd3a2d 844 */
9bee178b
SG		 845enum ib_mr_type {
846 IB_MR_TYPE_MEM_REG,
f5aa9159 847 IB_MR_TYPE_SG_GAPS,
a0bc099a
MG		 848 IB_MR_TYPE_DM,
849 IB_MR_TYPE_USER,
850 IB_MR_TYPE_DMA,
26bc7eae 851 IB_MR_TYPE_INTEGRITY,
17cd3a2d
SG		 852};
853
1b01d335
SG		 854enum ib_mr_status_check {
855 IB_MR_CHECK_SIG_STATUS = 1,
856};
857
858/**
859 * struct ib_mr_status - Memory region status container
860 *
861 * @fail_status: Bitmask of MR checks status. For each
862 * failed check a corresponding status bit is set.
863 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
864 * failure.
865 */
866struct ib_mr_status {
867 u32 fail_status;
868 struct ib_sig_err sig_err;
869};
870
bf6a9e31
JM		 871/**
872 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
873 * enum.
874 * @mult: multiple to convert.
875 */
8385fd84 876__attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
bf6a9e31 877
44c58487 878enum rdma_ah_attr_type {
87daac68 879 RDMA_AH_ATTR_TYPE_UNDEFINED,
44c58487
DC		 880 RDMA_AH_ATTR_TYPE_IB,
881 RDMA_AH_ATTR_TYPE_ROCE,
64b4646e 882 RDMA_AH_ATTR_TYPE_OPA,
44c58487
DC		 883};
884
885struct ib_ah_attr {
886 u16 dlid;
887 u8 src_path_bits;
888};
889
890struct roce_ah_attr {
891 u8 dmac[ETH_ALEN];
892};
893
64b4646e
DC		 894struct opa_ah_attr {
895 u32 dlid;
896 u8 src_path_bits;
d98bb7f7 897 bool make_grd;
64b4646e
DC		 898};
899
90898850 900struct rdma_ah_attr {
1da177e4 901 struct ib_global_route grh;
1da177e4 902 u8 sl;
1da177e4 903 u8 static_rate;
1da177e4 904 u8 port_num;
44c58487
DC		 905 u8 ah_flags;
906 enum rdma_ah_attr_type type;
907 union {
908 struct ib_ah_attr ib;
909 struct roce_ah_attr roce;
64b4646e 910 struct opa_ah_attr opa;
44c58487 911 };
1da177e4
LT		 912};
913
914enum ib_wc_status {
915 IB_WC_SUCCESS,
916 IB_WC_LOC_LEN_ERR,
917 IB_WC_LOC_QP_OP_ERR,
918 IB_WC_LOC_EEC_OP_ERR,
919 IB_WC_LOC_PROT_ERR,
920 IB_WC_WR_FLUSH_ERR,
921 IB_WC_MW_BIND_ERR,
922 IB_WC_BAD_RESP_ERR,
923 IB_WC_LOC_ACCESS_ERR,
924 IB_WC_REM_INV_REQ_ERR,
925 IB_WC_REM_ACCESS_ERR,
926 IB_WC_REM_OP_ERR,
927 IB_WC_RETRY_EXC_ERR,
928 IB_WC_RNR_RETRY_EXC_ERR,
929 IB_WC_LOC_RDD_VIOL_ERR,
930 IB_WC_REM_INV_RD_REQ_ERR,
931 IB_WC_REM_ABORT_ERR,
932 IB_WC_INV_EECN_ERR,
933 IB_WC_INV_EEC_STATE_ERR,
934 IB_WC_FATAL_ERR,
935 IB_WC_RESP_TIMEOUT_ERR,
936 IB_WC_GENERAL_ERR
937};
938
db7489e0 939const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
2b1b5b60 940
1da177e4
LT		 941enum ib_wc_opcode {
942 IB_WC_SEND,
943 IB_WC_RDMA_WRITE,
944 IB_WC_RDMA_READ,
945 IB_WC_COMP_SWAP,
946 IB_WC_FETCH_ADD,
c93570f2 947 IB_WC_LSO,
00f7ec36 948 IB_WC_LOCAL_INV,
4c67e2bf 949 IB_WC_REG_MR,
5e80ba8f
VS		 950 IB_WC_MASKED_COMP_SWAP,
951 IB_WC_MASKED_FETCH_ADD,
1da177e4
LT		 952/*
953 * Set value of IB_WC_RECV so consumers can test if a completion is a
954 * receive by testing (opcode & IB_WC_RECV).
955 */
956 IB_WC_RECV = 1 << 7,
957 IB_WC_RECV_RDMA_WITH_IMM
958};
959
960enum ib_wc_flags {
961 IB_WC_GRH = 1,
00f7ec36
SW		 962 IB_WC_WITH_IMM = (1<<1),
963 IB_WC_WITH_INVALIDATE = (1<<2),
d927d505 964 IB_WC_IP_CSUM_OK = (1<<3),
dd5f03be
MB		 965 IB_WC_WITH_SMAC = (1<<4),
966 IB_WC_WITH_VLAN = (1<<5),
c865f246 967 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
1da177e4
LT		 968};
969
970struct ib_wc {
14d3a3b2
CH		 971 union {
972 u64 wr_id;
973 struct ib_cqe *wr_cqe;
974 };
1da177e4
LT		 975 enum ib_wc_status status;
976 enum ib_wc_opcode opcode;
977 u32 vendor_err;
978 u32 byte_len;
062dbb69 979 struct ib_qp *qp;
00f7ec36
SW		 980 union {
981 __be32 imm_data;
982 u32 invalidate_rkey;
983 } ex;
1da177e4 984 u32 src_qp;
cd2a6e7d 985 u32 slid;
1da177e4
LT		 986 int wc_flags;
987 u16 pkey_index;
1da177e4
LT		 988 u8 sl;
989 u8 dlid_path_bits;
990 u8 port_num; /* valid only for DR SMPs on switches */
dd5f03be
MB		 991 u8 smac[ETH_ALEN];
992 u16 vlan_id;
c865f246 993 u8 network_hdr_type;
1da177e4
LT		 994};
995
ed23a727
RD		 996enum ib_cq_notify_flags {
997 IB_CQ_SOLICITED = 1 << 0,
998 IB_CQ_NEXT_COMP = 1 << 1,
999 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
1000 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
1da177e4
LT		 1001};
1002
96104eda 1003enum ib_srq_type {
418d5130 1004 IB_SRQT_BASIC,
9c2c8496
AK		 1005 IB_SRQT_XRC,
1006 IB_SRQT_TM,
96104eda
SH		 1007};
1008
1a56ff6d
AK		 1009static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
1010{
9c2c8496
AK		 1011 return srq_type == IB_SRQT_XRC ||
1012 srq_type == IB_SRQT_TM;
1a56ff6d
AK		 1013}
1014
d41fcc67
RD		 1015enum ib_srq_attr_mask {
1016 IB_SRQ_MAX_WR = 1 << 0,
1017 IB_SRQ_LIMIT = 1 << 1,
1018};
1019
1020struct ib_srq_attr {
1021 u32 max_wr;
1022 u32 max_sge;
1023 u32 srq_limit;
1024};
1025
1026struct ib_srq_init_attr {
1027 void (*event_handler)(struct ib_event *, void *);
1028 void *srq_context;
1029 struct ib_srq_attr attr;
96104eda 1030 enum ib_srq_type srq_type;
418d5130 1031
1a56ff6d
AK		 1032 struct {
1033 struct ib_cq *cq;
1034 union {
1035 struct {
1036 struct ib_xrcd *xrcd;
1037 } xrc;
9c2c8496
AK		 1038
1039 struct {
1040 u32 max_num_tags;
1041 } tag_matching;
1a56ff6d 1042 };
418d5130 1043 } ext;
d41fcc67
RD		 1044};
1045
1da177e4
LT		 1046struct ib_qp_cap {
1047 u32 max_send_wr;
1048 u32 max_recv_wr;
1049 u32 max_send_sge;
1050 u32 max_recv_sge;
1051 u32 max_inline_data;
a060b562
CH		 1052
1053 /*
1054 * Maximum number of rdma_rw_ctx structures in flight at a time.
1055 * ib_create_qp() will calculate the right amount of neededed WRs
1056 * and MRs based on this.
1057 */
1058 u32 max_rdma_ctxs;
1da177e4
LT		 1059};
1060
1061enum ib_sig_type {
1062 IB_SIGNAL_ALL_WR,
1063 IB_SIGNAL_REQ_WR
1064};
1065
1066enum ib_qp_type {
1067 /*
1068 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1069 * here (and in that order) since the MAD layer uses them as
1070 * indices into a 2-entry table.
1071 */
1072 IB_QPT_SMI,
1073 IB_QPT_GSI,
1074
1075 IB_QPT_RC,
1076 IB_QPT_UC,
1077 IB_QPT_UD,
1078 IB_QPT_RAW_IPV6,
b42b63cf 1079 IB_QPT_RAW_ETHERTYPE,
c938a616 1080 IB_QPT_RAW_PACKET = 8,
b42b63cf
SH		 1081 IB_QPT_XRC_INI = 9,
1082 IB_QPT_XRC_TGT,
0134f16b 1083 IB_QPT_MAX,
8011c1e3 1084 IB_QPT_DRIVER = 0xFF,
0134f16b
JM		 1085 /* Reserve a range for qp types internal to the low level driver.
1086 * These qp types will not be visible at the IB core layer, so the
1087 * IB_QPT_MAX usages should not be affected in the core layer
1088 */
1089 IB_QPT_RESERVED1 = 0x1000,
1090 IB_QPT_RESERVED2,
1091 IB_QPT_RESERVED3,
1092 IB_QPT_RESERVED4,
1093 IB_QPT_RESERVED5,
1094 IB_QPT_RESERVED6,
1095 IB_QPT_RESERVED7,
1096 IB_QPT_RESERVED8,
1097 IB_QPT_RESERVED9,
1098 IB_QPT_RESERVED10,
1da177e4
LT		 1099};
1100
b846f25a 1101enum ib_qp_create_flags {
47ee1b9f
RL		 1102 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
1103 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
8a06ce59
LR		 1104 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
1105 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
1106 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
90f1d1b4 1107 IB_QP_CREATE_NETIF_QP = 1 << 5,
c0a6cbb9 1108 IB_QP_CREATE_INTEGRITY_EN = 1 << 6,
7855f584 1109 /* FREE = 1 << 7, */
b531b909 1110 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
9c2b270e 1111 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9,
02984cc7 1112 IB_QP_CREATE_SOURCE_QPN = 1 << 10,
e1d2e887 1113 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11,
d2b57063
JM		 1114 /* reserve bits 26-31 for low level drivers' internal use */
1115 IB_QP_CREATE_RESERVED_START = 1 << 26,
1116 IB_QP_CREATE_RESERVED_END = 1 << 31,
b846f25a
EC		 1117};
1118
73c40c61
YH		 1119/*
1120 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1121 * callback to destroy the passed in QP.
1122 */
1123
1da177e4 1124struct ib_qp_init_attr {
eb93c82e 1125 /* Consumer's event_handler callback must not block */
1da177e4 1126 void (*event_handler)(struct ib_event *, void *);
eb93c82e 1127
1da177e4
LT		 1128 void *qp_context;
1129 struct ib_cq *send_cq;
1130 struct ib_cq *recv_cq;
1131 struct ib_srq *srq;
b42b63cf 1132 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1da177e4
LT		 1133 struct ib_qp_cap cap;
1134 enum ib_sig_type sq_sig_type;
1135 enum ib_qp_type qp_type;
b56511c1 1136 u32 create_flags;
a060b562
CH		 1137
1138 /*
1139 * Only needed for special QP types, or when using the RW API.
1140 */
1141 u8 port_num;
a9017e23 1142 struct ib_rwq_ind_table *rwq_ind_tbl;
02984cc7 1143 u32 source_qpn;
1da177e4
LT		 1144};
1145
0e0ec7e0
SH		 1146struct ib_qp_open_attr {
1147 void (*event_handler)(struct ib_event *, void *);
1148 void *qp_context;
1149 u32 qp_num;
1150 enum ib_qp_type qp_type;
1151};
1152
1da177e4
LT		 1153enum ib_rnr_timeout {
1154 IB_RNR_TIMER_655_36 = 0,
1155 IB_RNR_TIMER_000_01 = 1,
1156 IB_RNR_TIMER_000_02 = 2,
1157 IB_RNR_TIMER_000_03 = 3,
1158 IB_RNR_TIMER_000_04 = 4,
1159 IB_RNR_TIMER_000_06 = 5,
1160 IB_RNR_TIMER_000_08 = 6,
1161 IB_RNR_TIMER_000_12 = 7,
1162 IB_RNR_TIMER_000_16 = 8,
1163 IB_RNR_TIMER_000_24 = 9,
1164 IB_RNR_TIMER_000_32 = 10,
1165 IB_RNR_TIMER_000_48 = 11,
1166 IB_RNR_TIMER_000_64 = 12,
1167 IB_RNR_TIMER_000_96 = 13,
1168 IB_RNR_TIMER_001_28 = 14,
1169 IB_RNR_TIMER_001_92 = 15,
1170 IB_RNR_TIMER_002_56 = 16,
1171 IB_RNR_TIMER_003_84 = 17,
1172 IB_RNR_TIMER_005_12 = 18,
1173 IB_RNR_TIMER_007_68 = 19,
1174 IB_RNR_TIMER_010_24 = 20,
1175 IB_RNR_TIMER_015_36 = 21,
1176 IB_RNR_TIMER_020_48 = 22,
1177 IB_RNR_TIMER_030_72 = 23,
1178 IB_RNR_TIMER_040_96 = 24,
1179 IB_RNR_TIMER_061_44 = 25,
1180 IB_RNR_TIMER_081_92 = 26,
1181 IB_RNR_TIMER_122_88 = 27,
1182 IB_RNR_TIMER_163_84 = 28,
1183 IB_RNR_TIMER_245_76 = 29,
1184 IB_RNR_TIMER_327_68 = 30,
1185 IB_RNR_TIMER_491_52 = 31
1186};
1187
1188enum ib_qp_attr_mask {
1189 IB_QP_STATE = 1,
1190 IB_QP_CUR_STATE = (1<<1),
1191 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1192 IB_QP_ACCESS_FLAGS = (1<<3),
1193 IB_QP_PKEY_INDEX = (1<<4),
1194 IB_QP_PORT = (1<<5),
1195 IB_QP_QKEY = (1<<6),
1196 IB_QP_AV = (1<<7),
1197 IB_QP_PATH_MTU = (1<<8),
1198 IB_QP_TIMEOUT = (1<<9),
1199 IB_QP_RETRY_CNT = (1<<10),
1200 IB_QP_RNR_RETRY = (1<<11),
1201 IB_QP_RQ_PSN = (1<<12),
1202 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1203 IB_QP_ALT_PATH = (1<<14),
1204 IB_QP_MIN_RNR_TIMER = (1<<15),
1205 IB_QP_SQ_PSN = (1<<16),
1206 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1207 IB_QP_PATH_MIG_STATE = (1<<18),
1208 IB_QP_CAP = (1<<19),
dd5f03be 1209 IB_QP_DEST_QPN = (1<<20),
aa744cc0
MB		 1210 IB_QP_RESERVED1 = (1<<21),
1211 IB_QP_RESERVED2 = (1<<22),
1212 IB_QP_RESERVED3 = (1<<23),
1213 IB_QP_RESERVED4 = (1<<24),
528e5a1b 1214 IB_QP_RATE_LIMIT = (1<<25),
1da177e4
LT		 1215};
1216
1217enum ib_qp_state {
1218 IB_QPS_RESET,
1219 IB_QPS_INIT,
1220 IB_QPS_RTR,
1221 IB_QPS_RTS,
1222 IB_QPS_SQD,
1223 IB_QPS_SQE,
1224 IB_QPS_ERR
1225};
1226
1227enum ib_mig_state {
1228 IB_MIG_MIGRATED,
1229 IB_MIG_REARM,
1230 IB_MIG_ARMED
1231};
1232
7083e42e
SM		 1233enum ib_mw_type {
1234 IB_MW_TYPE_1 = 1,
1235 IB_MW_TYPE_2 = 2
1236};
1237
1da177e4
LT		 1238struct ib_qp_attr {
1239 enum ib_qp_state qp_state;
1240 enum ib_qp_state cur_qp_state;
1241 enum ib_mtu path_mtu;
1242 enum ib_mig_state path_mig_state;
1243 u32 qkey;
1244 u32 rq_psn;
1245 u32 sq_psn;
1246 u32 dest_qp_num;
1247 int qp_access_flags;
1248 struct ib_qp_cap cap;
90898850
DC		 1249 struct rdma_ah_attr ah_attr;
1250 struct rdma_ah_attr alt_ah_attr;
1da177e4
LT		 1251 u16 pkey_index;
1252 u16 alt_pkey_index;
1253 u8 en_sqd_async_notify;
1254 u8 sq_draining;
1255 u8 max_rd_atomic;
1256 u8 max_dest_rd_atomic;
1257 u8 min_rnr_timer;
1258 u8 port_num;
1259 u8 timeout;
1260 u8 retry_cnt;
1261 u8 rnr_retry;
1262 u8 alt_port_num;
1263 u8 alt_timeout;
528e5a1b 1264 u32 rate_limit;
1da177e4
LT		 1265};
1266
1267enum ib_wr_opcode {
9a59739b
JG		 1268 /* These are shared with userspace */
1269 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE,
1270 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM,
1271 IB_WR_SEND = IB_UVERBS_WR_SEND,
1272 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM,
1273 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ,
1274 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP,
1275 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD,
1276 IB_WR_LSO = IB_UVERBS_WR_TSO,
1277 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV,
1278 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV,
1279 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV,
1280 IB_WR_MASKED_ATOMIC_CMP_AND_SWP =
1281 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP,
1282 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD =
1283 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1284
1285 /* These are kernel only and can not be issued by userspace */
1286 IB_WR_REG_MR = 0x20,
38ca87c6 1287 IB_WR_REG_MR_INTEGRITY,
9a59739b 1288
0134f16b
JM		 1289 /* reserve values for low level drivers' internal use.
1290 * These values will not be used at all in the ib core layer.
1291 */
1292 IB_WR_RESERVED1 = 0xf0,
1293 IB_WR_RESERVED2,
1294 IB_WR_RESERVED3,
1295 IB_WR_RESERVED4,
1296 IB_WR_RESERVED5,
1297 IB_WR_RESERVED6,
1298 IB_WR_RESERVED7,
1299 IB_WR_RESERVED8,
1300 IB_WR_RESERVED9,
1301 IB_WR_RESERVED10,
1da177e4
LT		 1302};
1303
1304enum ib_send_flags {
1305 IB_SEND_FENCE = 1,
1306 IB_SEND_SIGNALED = (1<<1),
1307 IB_SEND_SOLICITED = (1<<2),
e0605d91 1308 IB_SEND_INLINE = (1<<3),
0134f16b
JM		 1309 IB_SEND_IP_CSUM = (1<<4),
1310
1311 /* reserve bits 26-31 for low level drivers' internal use */
1312 IB_SEND_RESERVED_START = (1 << 26),
1313 IB_SEND_RESERVED_END = (1 << 31),
1da177e4
LT		 1314};
1315
1316struct ib_sge {
1317 u64 addr;
1318 u32 length;
1319 u32 lkey;
1320};
1321
14d3a3b2
CH		 1322struct ib_cqe {
1323 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1324};
1325
1da177e4
LT		 1326struct ib_send_wr {
1327 struct ib_send_wr *next;
14d3a3b2
CH		 1328 union {
1329 u64 wr_id;
1330 struct ib_cqe *wr_cqe;
1331 };
1da177e4
LT		 1332 struct ib_sge *sg_list;
1333 int num_sge;
1334 enum ib_wr_opcode opcode;
1335 int send_flags;
0f39cf3d
RD		 1336 union {
1337 __be32 imm_data;
1338 u32 invalidate_rkey;
1339 } ex;
1da177e4
LT		 1340};
1341
e622f2f4
CH		 1342struct ib_rdma_wr {
1343 struct ib_send_wr wr;
1344 u64 remote_addr;
1345 u32 rkey;
1346};
1347
f696bf6d 1348static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr)
e622f2f4
CH		 1349{
1350 return container_of(wr, struct ib_rdma_wr, wr);
1351}
1352
1353struct ib_atomic_wr {
1354 struct ib_send_wr wr;
1355 u64 remote_addr;
1356 u64 compare_add;
1357 u64 swap;
1358 u64 compare_add_mask;
1359 u64 swap_mask;
1360 u32 rkey;
1361};
1362
f696bf6d 1363static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr)
e622f2f4
CH		 1364{
1365 return container_of(wr, struct ib_atomic_wr, wr);
1366}
1367
1368struct ib_ud_wr {
1369 struct ib_send_wr wr;
1370 struct ib_ah *ah;
1371 void *header;
1372 int hlen;
1373 int mss;
1374 u32 remote_qpn;
1375 u32 remote_qkey;
1376 u16 pkey_index; /* valid for GSI only */
1377 u8 port_num; /* valid for DR SMPs on switch only */
1378};
1379
f696bf6d 1380static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr)
e622f2f4
CH		 1381{
1382 return container_of(wr, struct ib_ud_wr, wr);
1383}
1384
4c67e2bf
SG		 1385struct ib_reg_wr {
1386 struct ib_send_wr wr;
1387 struct ib_mr *mr;
1388 u32 key;
1389 int access;
1390};
1391
f696bf6d 1392static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr)
4c67e2bf
SG		 1393{
1394 return container_of(wr, struct ib_reg_wr, wr);
1395}
1396
1da177e4
LT		 1397struct ib_recv_wr {
1398 struct ib_recv_wr *next;
14d3a3b2
CH		 1399 union {
1400 u64 wr_id;
1401 struct ib_cqe *wr_cqe;
1402 };
1da177e4
LT		 1403 struct ib_sge *sg_list;
1404 int num_sge;
1405};
1406
1407enum ib_access_flags {
4fca0377
JG		 1408 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE,
1409 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE,
1410 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ,
1411 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC,
1412 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND,
1413 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED,
1414 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND,
1415 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB,
1416
1417 IB_ACCESS_SUPPORTED = ((IB_ACCESS_HUGETLB << 1) - 1)
1da177e4
LT		 1418};
1419
b7d3e0a9
CH		 1420/*
1421 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1422 * are hidden here instead of a uapi header!
1423 */
1da177e4
LT		 1424enum ib_mr_rereg_flags {
1425 IB_MR_REREG_TRANS = 1,
1426 IB_MR_REREG_PD = (1<<1),
7e6edb9b
MB		 1427 IB_MR_REREG_ACCESS = (1<<2),
1428 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1da177e4
LT		 1429};
1430
1da177e4
LT		 1431struct ib_fmr_attr {
1432 int max_pages;
1433 int max_maps;
d36f34aa 1434 u8 page_shift;
1da177e4
LT		 1435};
1436
882214e2
HE		 1437struct ib_umem;
1438
38321256 1439enum rdma_remove_reason {
1c77483e
YH		 1440 /*
1441 * Userspace requested uobject deletion or initial try
1442 * to remove uobject via cleanup. Call could fail
1443 */
38321256
MB		 1444 RDMA_REMOVE_DESTROY,
1445 /* Context deletion. This call should delete the actual object itself */
1446 RDMA_REMOVE_CLOSE,
1447 /* Driver is being hot-unplugged. This call should delete the actual object itself */
1448 RDMA_REMOVE_DRIVER_REMOVE,
87ad80ab
JG		 1449 /* uobj is being cleaned-up before being committed */
1450 RDMA_REMOVE_ABORT,
38321256
MB		 1451};
1452
43579b5f
PP		 1453struct ib_rdmacg_object {
1454#ifdef CONFIG_CGROUP_RDMA
1455 struct rdma_cgroup *cg; /* owner rdma cgroup */
1456#endif
1457};
1458
e2773c06
RD		 1459struct ib_ucontext {
1460 struct ib_device *device;
771addf6 1461 struct ib_uverbs_file *ufile;
e951747a
JG		 1462 /*
1463 * 'closing' can be read by the driver only during a destroy callback,
1464 * it is set when we are closing the file descriptor and indicates
1465 * that mm_sem may be locked.
1466 */
6ceb6331 1467 bool closing;
8ada2c1c 1468
1c77483e 1469 bool cleanup_retryable;
38321256 1470
43579b5f 1471 struct ib_rdmacg_object cg_obj;
60615210
LR		 1472 /*
1473 * Implementation details of the RDMA core, don't use in drivers:
1474 */
1475 struct rdma_restrack_entry res;
3411f9f0 1476 struct xarray mmap_xa;
e2773c06
RD		 1477};
1478
1479struct ib_uobject {
1480 u64 user_handle; /* handle given to us by userspace */
6a5e9c88
JG		 1481 /* ufile & ucontext owning this object */
1482 struct ib_uverbs_file *ufile;
1483 /* FIXME, save memory: ufile->context == context */
e2773c06 1484 struct ib_ucontext *context; /* associated user context */
9ead190b 1485 void *object; /* containing object */
e2773c06 1486 struct list_head list; /* link to context's list */
43579b5f 1487 struct ib_rdmacg_object cg_obj; /* rdmacg object */
b3d636b0 1488 int id; /* index into kernel idr */
9ead190b 1489 struct kref ref;
38321256 1490 atomic_t usecnt; /* protects exclusive access */
d144da8c 1491 struct rcu_head rcu; /* kfree_rcu() overhead */
38321256 1492
6b0d08f4 1493 const struct uverbs_api_object *uapi_object;
e2773c06
RD		 1494};
1495
e2773c06 1496struct ib_udata {
309243ec 1497 const void __user *inbuf;
e2773c06
RD		 1498 void __user *outbuf;
1499 size_t inlen;
1500 size_t outlen;
1501};
1502
1da177e4 1503struct ib_pd {
96249d70 1504 u32 local_dma_lkey;
ed082d36 1505 u32 flags;
e2773c06
RD		 1506 struct ib_device *device;
1507 struct ib_uobject *uobject;
1508 atomic_t usecnt; /* count all resources */
50d46335 1509
ed082d36
CH		 1510 u32 unsafe_global_rkey;
1511
50d46335
CH		 1512 /*
1513 * Implementation details of the RDMA core, don't use in drivers:
1514 */
1515 struct ib_mr *__internal_mr;
02d8883f 1516 struct rdma_restrack_entry res;
1da177e4
LT		 1517};
1518
59991f94
SH		 1519struct ib_xrcd {
1520 struct ib_device *device;
d3d72d90 1521 atomic_t usecnt; /* count all exposed resources */
53d0bd1e 1522 struct inode *inode;
d3d72d90
SH		 1523
1524 struct mutex tgt_qp_mutex;
1525 struct list_head tgt_qp_list;
59991f94
SH		 1526};
1527
1da177e4
LT		 1528struct ib_ah {
1529 struct ib_device *device;
1530 struct ib_pd *pd;
e2773c06 1531 struct ib_uobject *uobject;
1a1f460f 1532 const struct ib_gid_attr *sgid_attr;
44c58487 1533 enum rdma_ah_attr_type type;
1da177e4
LT		 1534};
1535
1536typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1537
14d3a3b2 1538enum ib_poll_context {
f794809a
JM		 1539 IB_POLL_DIRECT, /* caller context, no hw completions */
1540 IB_POLL_SOFTIRQ, /* poll from softirq context */
1541 IB_POLL_WORKQUEUE, /* poll from workqueue */
1542 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */
14d3a3b2
CH		 1543};
1544
1da177e4 1545struct ib_cq {
e2773c06
RD		 1546 struct ib_device *device;
1547 struct ib_uobject *uobject;
1548 ib_comp_handler comp_handler;
1549 void (*event_handler)(struct ib_event *, void *);
4deccd6d 1550 void *cq_context;
e2773c06
RD		 1551 int cqe;
1552 atomic_t usecnt; /* count number of work queues */
14d3a3b2
CH		 1553 enum ib_poll_context poll_ctx;
1554 struct ib_wc *wc;
1555 union {
1556 struct irq_poll iop;
1557 struct work_struct work;
1558 };
f794809a 1559 struct workqueue_struct *comp_wq;
da662979 1560 struct dim *dim;
3e5901cb
CL		 1561
1562 /* updated only by trace points */
1563 ktime_t timestamp;
1564 bool interrupt;
1565
02d8883f
LR		 1566 /*
1567 * Implementation details of the RDMA core, don't use in drivers:
1568 */
1569 struct rdma_restrack_entry res;
1da177e4
LT		 1570};
1571
1572struct ib_srq {
d41fcc67
RD		 1573 struct ib_device *device;
1574 struct ib_pd *pd;
1575 struct ib_uobject *uobject;
1576 void (*event_handler)(struct ib_event *, void *);
1577 void *srq_context;
96104eda 1578 enum ib_srq_type srq_type;
1da177e4 1579 atomic_t usecnt;
418d5130 1580
1a56ff6d
AK		 1581 struct {
1582 struct ib_cq *cq;
1583 union {
1584 struct {
1585 struct ib_xrcd *xrcd;
1586 u32 srq_num;
1587 } xrc;
1588 };
418d5130 1589 } ext;
1da177e4
LT		 1590};
1591
ebaaee25
NO		 1592enum ib_raw_packet_caps {
1593 /* Strip cvlan from incoming packet and report it in the matching work
1594 * completion is supported.
1595 */
1596 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0),
1597 /* Scatter FCS field of an incoming packet to host memory is supported.
1598 */
1599 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1),
1600 /* Checksum offloads are supported (for both send and receive). */
1601 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2),
7d9336d8
MG		 1602 /* When a packet is received for an RQ with no receive WQEs, the
1603 * packet processing is delayed.
1604 */
1605 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3),
ebaaee25
NO		 1606};
1607
5fd251c8
YH		 1608enum ib_wq_type {
1609 IB_WQT_RQ
1610};
1611
1612enum ib_wq_state {
1613 IB_WQS_RESET,
1614 IB_WQS_RDY,
1615 IB_WQS_ERR
1616};
1617
1618struct ib_wq {
1619 struct ib_device *device;
1620 struct ib_uobject *uobject;
1621 void *wq_context;
1622 void (*event_handler)(struct ib_event *, void *);
1623 struct ib_pd *pd;
1624 struct ib_cq *cq;
1625 u32 wq_num;
1626 enum ib_wq_state state;
1627 enum ib_wq_type wq_type;
1628 atomic_t usecnt;
1629};
1630
10bac72b
NO		 1631enum ib_wq_flags {
1632 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0,
27b0df11 1633 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1,
7d9336d8 1634 IB_WQ_FLAGS_DELAY_DROP = 1 << 2,
e1d2e887 1635 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3,
10bac72b
NO		 1636};
1637
5fd251c8
YH		 1638struct ib_wq_init_attr {
1639 void *wq_context;
1640 enum ib_wq_type wq_type;
1641 u32 max_wr;
1642 u32 max_sge;
1643 struct ib_cq *cq;
1644 void (*event_handler)(struct ib_event *, void *);
10bac72b 1645 u32 create_flags; /* Use enum ib_wq_flags */
5fd251c8
YH		 1646};
1647
1648enum ib_wq_attr_mask {
10bac72b
NO		 1649 IB_WQ_STATE = 1 << 0,
1650 IB_WQ_CUR_STATE = 1 << 1,
1651 IB_WQ_FLAGS = 1 << 2,
5fd251c8
YH		 1652};
1653
1654struct ib_wq_attr {
1655 enum ib_wq_state wq_state;
1656 enum ib_wq_state curr_wq_state;
10bac72b
NO		 1657 u32 flags; /* Use enum ib_wq_flags */
1658 u32 flags_mask; /* Use enum ib_wq_flags */
5fd251c8
YH		 1659};
1660
6d39786b
YH		 1661struct ib_rwq_ind_table {
1662 struct ib_device *device;
1663 struct ib_uobject *uobject;
1664 atomic_t usecnt;
1665 u32 ind_tbl_num;
1666 u32 log_ind_tbl_size;
1667 struct ib_wq **ind_tbl;
1668};
1669
1670struct ib_rwq_ind_table_init_attr {
1671 u32 log_ind_tbl_size;
1672 /* Each entry is a pointer to Receive Work Queue */
1673 struct ib_wq **ind_tbl;
1674};
1675
d291f1a6
DJ		 1676enum port_pkey_state {
1677 IB_PORT_PKEY_NOT_VALID = 0,
1678 IB_PORT_PKEY_VALID = 1,
1679 IB_PORT_PKEY_LISTED = 2,
1680};
1681
1682struct ib_qp_security;
1683
1684struct ib_port_pkey {
1685 enum port_pkey_state state;
1686 u16 pkey_index;
1687 u8 port_num;
1688 struct list_head qp_list;
1689 struct list_head to_error_list;
1690 struct ib_qp_security *sec;
1691};
1692
1693struct ib_ports_pkeys {
1694 struct ib_port_pkey main;
1695 struct ib_port_pkey alt;
1696};
1697
1698struct ib_qp_security {
1699 struct ib_qp *qp;
1700 struct ib_device *dev;
1701 /* Hold this mutex when changing port and pkey settings. */
1702 struct mutex mutex;
1703 struct ib_ports_pkeys *ports_pkeys;
1704 /* A list of all open shared QP handles. Required to enforce security
1705 * properly for all users of a shared QP.
1706 */
1707 struct list_head shared_qp_list;
1708 void *security;
1709 bool destroying;
1710 atomic_t error_list_count;
1711 struct completion error_complete;
1712 int error_comps_pending;
1713};
1714
632bc3f6
BVA		 1715/*
1716 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1717 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1718 */
1da177e4
LT		 1719struct ib_qp {
1720 struct ib_device *device;
1721 struct ib_pd *pd;
1722 struct ib_cq *send_cq;
1723 struct ib_cq *recv_cq;
fffb0383
CH		 1724 spinlock_t mr_lock;
1725 int mrs_used;
a060b562 1726 struct list_head rdma_mrs;
0e353e34 1727 struct list_head sig_mrs;
1da177e4 1728 struct ib_srq *srq;
b42b63cf 1729 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
d3d72d90 1730 struct list_head xrcd_list;
fffb0383 1731
319a441d
HHZ		 1732 /* count times opened, mcast attaches, flow attaches */
1733 atomic_t usecnt;
0e0ec7e0
SH		 1734 struct list_head open_list;
1735 struct ib_qp *real_qp;
e2773c06 1736 struct ib_uobject *uobject;
1da177e4
LT		 1737 void (*event_handler)(struct ib_event *, void *);
1738 void *qp_context;
1a1f460f
JG		 1739 /* sgid_attrs associated with the AV's */
1740 const struct ib_gid_attr *av_sgid_attr;
1741 const struct ib_gid_attr *alt_path_sgid_attr;
1da177e4 1742 u32 qp_num;
632bc3f6
BVA		 1743 u32 max_write_sge;
1744 u32 max_read_sge;
1da177e4 1745 enum ib_qp_type qp_type;
a9017e23 1746 struct ib_rwq_ind_table *rwq_ind_tbl;
d291f1a6 1747 struct ib_qp_security *qp_sec;
498ca3c8 1748 u8 port;
02d8883f 1749
185eddc4 1750 bool integrity_en;
02d8883f
LR		 1751 /*
1752 * Implementation details of the RDMA core, don't use in drivers:
1753 */
1754 struct rdma_restrack_entry res;
99fa331d
MZ		 1755
1756 /* The counter the qp is bind to */
1757 struct rdma_counter *counter;
1da177e4
LT		 1758};
1759
bee76d7a
AL		 1760struct ib_dm {
1761 struct ib_device *device;
1762 u32 length;
1763 u32 flags;
1764 struct ib_uobject *uobject;
1765 atomic_t usecnt;
1766};
1767
1da177e4 1768struct ib_mr {
e2773c06
RD		 1769 struct ib_device *device;
1770 struct ib_pd *pd;
e2773c06
RD		 1771 u32 lkey;
1772 u32 rkey;
4c67e2bf 1773 u64 iova;
edd31551 1774 u64 length;
4c67e2bf 1775 unsigned int page_size;
a0bc099a 1776 enum ib_mr_type type;
d4a85c30 1777 bool need_inval;
fffb0383
CH		 1778 union {
1779 struct ib_uobject *uobject; /* user */
1780 struct list_head qp_entry; /* FR */
1781 };
fccec5b8 1782
be934cca 1783 struct ib_dm *dm;
7c717d3a 1784 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */
fccec5b8
SW		 1785 /*
1786 * Implementation details of the RDMA core, don't use in drivers:
1787 */
1788 struct rdma_restrack_entry res;
1da177e4
LT		 1789};
1790
1791struct ib_mw {
1792 struct ib_device *device;
1793 struct ib_pd *pd;
e2773c06 1794 struct ib_uobject *uobject;
1da177e4 1795 u32 rkey;
7083e42e 1796 enum ib_mw_type type;
1da177e4
LT		 1797};
1798
1799struct ib_fmr {
1800 struct ib_device *device;
1801 struct ib_pd *pd;
1802 struct list_head list;
1803 u32 lkey;
1804 u32 rkey;
1805};
1806
319a441d
HHZ		 1807/* Supported steering options */
1808enum ib_flow_attr_type {
1809 /* steering according to rule specifications */
1810 IB_FLOW_ATTR_NORMAL = 0x0,
1811 /* default unicast and multicast rule -
1812 * receive all Eth traffic which isn't steered to any QP
1813 */
1814 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1815 /* default multicast rule -
1816 * receive all Eth multicast traffic which isn't steered to any QP
1817 */
1818 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1819 /* sniffer rule - receive all port traffic */
1820 IB_FLOW_ATTR_SNIFFER = 0x3
1821};
1822
1823/* Supported steering header types */
1824enum ib_flow_spec_type {
1825 /* L2 headers*/
76bd23b3
MR		 1826 IB_FLOW_SPEC_ETH = 0x20,
1827 IB_FLOW_SPEC_IB = 0x22,
319a441d 1828 /* L3 header*/
76bd23b3
MR		 1829 IB_FLOW_SPEC_IPV4 = 0x30,
1830 IB_FLOW_SPEC_IPV6 = 0x31,
56ab0b38 1831 IB_FLOW_SPEC_ESP = 0x34,
319a441d 1832 /* L4 headers*/
76bd23b3
MR		 1833 IB_FLOW_SPEC_TCP = 0x40,
1834 IB_FLOW_SPEC_UDP = 0x41,
0dbf3332 1835 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50,
d90e5e50 1836 IB_FLOW_SPEC_GRE = 0x51,
b04f0f03 1837 IB_FLOW_SPEC_MPLS = 0x60,
fbf46860 1838 IB_FLOW_SPEC_INNER = 0x100,
460d0198
MR		 1839 /* Actions */
1840 IB_FLOW_SPEC_ACTION_TAG = 0x1000,
483a3966 1841 IB_FLOW_SPEC_ACTION_DROP = 0x1001,
9b828441 1842 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002,
7eea23a5 1843 IB_FLOW_SPEC_ACTION_COUNT = 0x1003,
319a441d 1844};
240ae00e 1845#define IB_FLOW_SPEC_LAYER_MASK 0xF0
7eea23a5 1846#define IB_FLOW_SPEC_SUPPORT_LAYERS 10
22878dbc 1847
319a441d
HHZ		 1848/* Flow steering rule priority is set according to it's domain.
1849 * Lower domain value means higher priority.
1850 */
1851enum ib_flow_domain {
1852 IB_FLOW_DOMAIN_USER,
1853 IB_FLOW_DOMAIN_ETHTOOL,
1854 IB_FLOW_DOMAIN_RFS,
1855 IB_FLOW_DOMAIN_NIC,
1856 IB_FLOW_DOMAIN_NUM /* Must be last */
1857};
1858
a3100a78
MV		 1859enum ib_flow_flags {
1860 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
21e82d3e
BP		 1861 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */
1862 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */
a3100a78
MV		 1863};
1864
319a441d
HHZ		 1865struct ib_flow_eth_filter {
1866 u8 dst_mac[6];
1867 u8 src_mac[6];
1868 __be16 ether_type;
1869 __be16 vlan_tag;
15dfbd6b
MG		 1870 /* Must be last */
1871 u8 real_sz[0];
319a441d
HHZ		 1872};
1873
1874struct ib_flow_spec_eth {
fbf46860 1875 u32 type;
319a441d
HHZ		 1876 u16 size;
1877 struct ib_flow_eth_filter val;
1878 struct ib_flow_eth_filter mask;
1879};
1880
240ae00e
MB		 1881struct ib_flow_ib_filter {
1882 __be16 dlid;
1883 __u8 sl;
15dfbd6b
MG		 1884 /* Must be last */
1885 u8 real_sz[0];
240ae00e
MB		 1886};
1887
1888struct ib_flow_spec_ib {
fbf46860 1889 u32 type;
240ae00e
MB		 1890 u16 size;
1891 struct ib_flow_ib_filter val;
1892 struct ib_flow_ib_filter mask;
1893};
1894
989a3a8f
MG		 1895/* IPv4 header flags */
1896enum ib_ipv4_flags {
1897 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1898 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the
1899 last have this flag set */
1900};
1901
319a441d
HHZ		 1902struct ib_flow_ipv4_filter {
1903 __be32 src_ip;
1904 __be32 dst_ip;
989a3a8f
MG		 1905 u8 proto;
1906 u8 tos;
1907 u8 ttl;
1908 u8 flags;
15dfbd6b
MG		 1909 /* Must be last */
1910 u8 real_sz[0];
319a441d
HHZ		 1911};
1912
1913struct ib_flow_spec_ipv4 {
fbf46860 1914 u32 type;
319a441d
HHZ		 1915 u16 size;
1916 struct ib_flow_ipv4_filter val;
1917 struct ib_flow_ipv4_filter mask;
1918};
1919
4c2aae71
MG		 1920struct ib_flow_ipv6_filter {
1921 u8 src_ip[16];
1922 u8 dst_ip[16];
a72c6a2b
MG		 1923 __be32 flow_label;
1924 u8 next_hdr;
1925 u8 traffic_class;
1926 u8 hop_limit;
15dfbd6b
MG		 1927 /* Must be last */
1928 u8 real_sz[0];
4c2aae71
MG		 1929};
1930
1931struct ib_flow_spec_ipv6 {
fbf46860 1932 u32 type;
4c2aae71
MG		 1933 u16 size;
1934 struct ib_flow_ipv6_filter val;
1935 struct ib_flow_ipv6_filter mask;
1936};
1937
319a441d
HHZ		 1938struct ib_flow_tcp_udp_filter {
1939 __be16 dst_port;
1940 __be16 src_port;
15dfbd6b
MG		 1941 /* Must be last */
1942 u8 real_sz[0];
319a441d
HHZ		 1943};
1944
1945struct ib_flow_spec_tcp_udp {
fbf46860 1946 u32 type;
319a441d
HHZ		 1947 u16 size;
1948 struct ib_flow_tcp_udp_filter val;
1949 struct ib_flow_tcp_udp_filter mask;
1950};
1951
0dbf3332
MR		 1952struct ib_flow_tunnel_filter {
1953 __be32 tunnel_id;
1954 u8 real_sz[0];
1955};
1956
1957/* ib_flow_spec_tunnel describes the Vxlan tunnel
1958 * the tunnel_id from val has the vni value
1959 */
1960struct ib_flow_spec_tunnel {
fbf46860 1961 u32 type;
0dbf3332
MR		 1962 u16 size;
1963 struct ib_flow_tunnel_filter val;
1964 struct ib_flow_tunnel_filter mask;
1965};
1966
56ab0b38
MB		 1967struct ib_flow_esp_filter {
1968 __be32 spi;
1969 __be32 seq;
1970 /* Must be last */
1971 u8 real_sz[0];
1972};
1973
1974struct ib_flow_spec_esp {
1975 u32 type;
1976 u16 size;
1977 struct ib_flow_esp_filter val;
1978 struct ib_flow_esp_filter mask;
1979};
1980
d90e5e50
AL		 1981struct ib_flow_gre_filter {
1982 __be16 c_ks_res0_ver;
1983 __be16 protocol;
1984 __be32 key;
1985 /* Must be last */
1986 u8 real_sz[0];
1987};
1988
1989struct ib_flow_spec_gre {
1990 u32 type;
1991 u16 size;
1992 struct ib_flow_gre_filter val;
1993 struct ib_flow_gre_filter mask;
1994};
1995
b04f0f03
AL		 1996struct ib_flow_mpls_filter {
1997 __be32 tag;
1998 /* Must be last */
1999 u8 real_sz[0];
2000};
2001
2002struct ib_flow_spec_mpls {
2003 u32 type;
2004 u16 size;
2005 struct ib_flow_mpls_filter val;
2006 struct ib_flow_mpls_filter mask;
2007};
2008
460d0198
MR		 2009struct ib_flow_spec_action_tag {
2010 enum ib_flow_spec_type type;
2011 u16 size;
2012 u32 tag_id;
2013};
2014
483a3966
SS		 2015struct ib_flow_spec_action_drop {
2016 enum ib_flow_spec_type type;
2017 u16 size;
2018};
2019
9b828441
MB		 2020struct ib_flow_spec_action_handle {
2021 enum ib_flow_spec_type type;
2022 u16 size;
2023 struct ib_flow_action *act;
2024};
2025
7eea23a5
RS		 2026enum ib_counters_description {
2027 IB_COUNTER_PACKETS,
2028 IB_COUNTER_BYTES,
2029};
2030
2031struct ib_flow_spec_action_count {
2032 enum ib_flow_spec_type type;
2033 u16 size;
2034 struct ib_counters *counters;
2035};
2036
319a441d
HHZ		 2037union ib_flow_spec {
2038 struct {
fbf46860 2039 u32 type;
319a441d
HHZ		 2040 u16 size;
2041 };
2042 struct ib_flow_spec_eth eth;
240ae00e 2043 struct ib_flow_spec_ib ib;
319a441d
HHZ		 2044 struct ib_flow_spec_ipv4 ipv4;
2045 struct ib_flow_spec_tcp_udp tcp_udp;
4c2aae71 2046 struct ib_flow_spec_ipv6 ipv6;
0dbf3332 2047 struct ib_flow_spec_tunnel tunnel;
56ab0b38 2048 struct ib_flow_spec_esp esp;
d90e5e50 2049 struct ib_flow_spec_gre gre;
b04f0f03 2050 struct ib_flow_spec_mpls mpls;
460d0198 2051 struct ib_flow_spec_action_tag flow_tag;
483a3966 2052 struct ib_flow_spec_action_drop drop;
9b828441 2053 struct ib_flow_spec_action_handle action;
7eea23a5 2054 struct ib_flow_spec_action_count flow_count;
319a441d
HHZ		 2055};
2056
2057struct ib_flow_attr {
2058 enum ib_flow_attr_type type;
2059 u16 size;
2060 u16 priority;
2061 u32 flags;
2062 u8 num_of_specs;
2063 u8 port;
7654cb1b 2064 union ib_flow_spec flows[];
319a441d
HHZ		 2065};
2066
2067struct ib_flow {
2068 struct ib_qp *qp;
6cd080a6 2069 struct ib_device *device;
319a441d
HHZ		 2070 struct ib_uobject *uobject;
2071};
2072
2eb9beae
MB		 2073enum ib_flow_action_type {
2074 IB_FLOW_ACTION_UNSPECIFIED,
2075 IB_FLOW_ACTION_ESP = 1,
2076};
2077
2078struct ib_flow_action_attrs_esp_keymats {
2079 enum ib_uverbs_flow_action_esp_keymat protocol;
2080 union {
2081 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm;
2082 } keymat;
2083};
2084
2085struct ib_flow_action_attrs_esp_replays {
2086 enum ib_uverbs_flow_action_esp_replay protocol;
2087 union {
2088 struct ib_uverbs_flow_action_esp_replay_bmp bmp;
2089 } replay;
2090};
2091
2092enum ib_flow_action_attrs_esp_flags {
2093 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags
2094 * This is done in order to share the same flags between user-space and
2095 * kernel and spare an unnecessary translation.
2096 */
2097
2098 /* Kernel flags */
2099 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32,
7d12f8d5 2100 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33,
2eb9beae
MB		 2101};
2102
2103struct ib_flow_spec_list {
2104 struct ib_flow_spec_list *next;
2105 union ib_flow_spec spec;
2106};
2107
2108struct ib_flow_action_attrs_esp {
2109 struct ib_flow_action_attrs_esp_keymats *keymat;
2110 struct ib_flow_action_attrs_esp_replays *replay;
2111 struct ib_flow_spec_list *encap;
2112 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled.
2113 * Value of 0 is a valid value.
2114 */
2115 u32 esn;
2116 u32 spi;
2117 u32 seq;
2118 u32 tfc_pad;
2119 /* Use enum ib_flow_action_attrs_esp_flags */
2120 u64 flags;
2121 u64 hard_limit_pkts;
2122};
2123
2124struct ib_flow_action {
2125 struct ib_device *device;
2126 struct ib_uobject *uobject;
2127 enum ib_flow_action_type type;
2128 atomic_t usecnt;
2129};
2130
e26e7b88 2131struct ib_mad;
1da177e4
LT		 2132struct ib_grh;
2133
2134enum ib_process_mad_flags {
2135 IB_MAD_IGNORE_MKEY = 1,
2136 IB_MAD_IGNORE_BKEY = 2,
2137 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
2138};
2139
2140enum ib_mad_result {
2141 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
2142 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
2143 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
2144 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
2145};
2146
21d6454a 2147struct ib_port_cache {
883c71fe 2148 u64 subnet_prefix;
21d6454a
JW		 2149 struct ib_pkey_cache *pkey;
2150 struct ib_gid_table *gid;
2151 u8 lmc;
2152 enum ib_port_state port_state;
2153};
2154
7738613e
IW		 2155struct ib_port_immutable {
2156 int pkey_tbl_len;
2157 int gid_tbl_len;
f9b22e35 2158 u32 core_cap_flags;
337877a4 2159 u32 max_mad_size;
7738613e
IW		 2160};
2161
8ceb1357 2162struct ib_port_data {
324e227e
JG		 2163 struct ib_device *ib_dev;
2164
8ceb1357
JG		 2165 struct ib_port_immutable immutable;
2166
2167 spinlock_t pkey_list_lock;
2168 struct list_head pkey_list;
8faea9fd
JG		 2169
2170 struct ib_port_cache cache;
c2261dd7
JG		 2171
2172 spinlock_t netdev_lock;
324e227e
JG		 2173 struct net_device __rcu *netdev;
2174 struct hlist_node ndev_hash_link;
413d3347 2175 struct rdma_port_counter port_counter;
6e7be47a 2176 struct rdma_hw_stats *hw_stats;
8ceb1357
JG		 2177};
2178
2fc77572
VN		 2179/* rdma netdev type - specifies protocol type */
2180enum rdma_netdev_t {
f0ad83ac
NV		 2181 RDMA_NETDEV_OPA_VNIC,
2182 RDMA_NETDEV_IPOIB,
2fc77572
VN		 2183};
2184
2185/**
2186 * struct rdma_netdev - rdma netdev
2187 * For cases where netstack interfacing is required.
2188 */
2189struct rdma_netdev {
2190 void *clnt_priv;
2191 struct ib_device *hca;
2192 u8 port_num;
2193
9f49a5b5
JG		 2194 /*
2195 * cleanup function must be specified.
2196 * FIXME: This is only used for OPA_VNIC and that usage should be
2197 * removed too.
2198 */
8e959601
NV		 2199 void (*free_rdma_netdev)(struct net_device *netdev);
2200
2fc77572
VN		 2201 /* control functions */
2202 void (*set_id)(struct net_device *netdev, int id);
f0ad83ac
NV		 2203 /* send packet */
2204 int (*send)(struct net_device *dev, struct sk_buff *skb,
2205 struct ib_ah *address, u32 dqpn);
2206 /* multicast */
2207 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2208 union ib_gid *gid, u16 mlid,
2209 int set_qkey, u32 qkey);
2210 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2211 union ib_gid *gid, u16 mlid);
2fc77572
VN		 2212};
2213
f6a8a19b
DD		 2214struct rdma_netdev_alloc_params {
2215 size_t sizeof_priv;
2216 unsigned int txqs;
2217 unsigned int rxqs;
2218 void *param;
2219
2220 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num,
2221 struct net_device *netdev, void *param);
2222};
2223
a3de94e3
EA		 2224struct ib_odp_counters {
2225 atomic64_t faults;
2226 atomic64_t invalidations;
2227};
2228
fa9b1802
RS		 2229struct ib_counters {
2230 struct ib_device *device;
2231 struct ib_uobject *uobject;
2232 /* num of objects attached */
2233 atomic_t usecnt;
2234};
2235
51d7a538
RS		 2236struct ib_counters_read_attr {
2237 u64 *counters_buff;
2238 u32 ncounters;
2239 u32 flags; /* use enum ib_read_counters_flags */
2240};
2241
2eb9beae 2242struct uverbs_attr_bundle;
dd05cb82
KH		 2243struct iw_cm_id;
2244struct iw_cm_conn_param;
2eb9beae 2245
30471d4b
LR		 2246#define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \
2247 .size_##ib_struct = \
2248 (sizeof(struct drv_struct) + \
2249 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \
2250 BUILD_BUG_ON_ZERO( \
2251 !__same_type(((struct drv_struct *)NULL)->member, \
2252 struct ib_struct)))
2253
f6316032
LR		 2254#define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \
2255 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp))
2256
30471d4b 2257#define rdma_zalloc_drv_obj(ib_dev, ib_type) \
f6316032 2258 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL)
30471d4b
LR		 2259
2260#define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct
2261
3411f9f0
MK		 2262struct rdma_user_mmap_entry {
2263 struct kref ref;
2264 struct ib_ucontext *ucontext;
2265 unsigned long start_pgoff;
2266 size_t npages;
2267 bool driver_removed;
2268};
2269
2270/* Return the offset (in bytes) the user should pass to libc's mmap() */
2271static inline u64
2272rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry)
2273{
2274 return (u64)entry->start_pgoff << PAGE_SHIFT;
2275}
2276
521ed0d9
KH		 2277/**
2278 * struct ib_device_ops - InfiniBand device operations
2279 * This structure defines all the InfiniBand device operations, providers will
2280 * need to define the supported operations, otherwise they will be set to null.
2281 */
2282struct ib_device_ops {
7a154142 2283 struct module *owner;
b9560a41 2284 enum rdma_driver_id driver_id;
72c6ec18 2285 u32 uverbs_abi_ver;
8f71bb00 2286 unsigned int uverbs_no_driver_id_binding:1;
b9560a41 2287
521ed0d9
KH		 2288 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr,
2289 const struct ib_send_wr **bad_send_wr);
2290 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr,
2291 const struct ib_recv_wr **bad_recv_wr);
2292 void (*drain_rq)(struct ib_qp *qp);
2293 void (*drain_sq)(struct ib_qp *qp);
2294 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc);
2295 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2296 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags);
2297 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt);
2298 int (*post_srq_recv)(struct ib_srq *srq,
2299 const struct ib_recv_wr *recv_wr,
2300 const struct ib_recv_wr **bad_recv_wr);
2301 int (*process_mad)(struct ib_device *device, int process_mad_flags,
2302 u8 port_num, const struct ib_wc *in_wc,
2303 const struct ib_grh *in_grh,
e26e7b88
LR		 2304 const struct ib_mad *in_mad, struct ib_mad *out_mad,
2305 size_t *out_mad_size, u16 *out_mad_pkey_index);
521ed0d9
KH		 2306 int (*query_device)(struct ib_device *device,
2307 struct ib_device_attr *device_attr,
2308 struct ib_udata *udata);
2309 int (*modify_device)(struct ib_device *device, int device_modify_mask,
2310 struct ib_device_modify *device_modify);
2311 void (*get_dev_fw_str)(struct ib_device *device, char *str);
2312 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2313 int comp_vector);
2314 int (*query_port)(struct ib_device *device, u8 port_num,
2315 struct ib_port_attr *port_attr);
2316 int (*modify_port)(struct ib_device *device, u8 port_num,
2317 int port_modify_mask,
2318 struct ib_port_modify *port_modify);
2319 /**
2320 * The following mandatory functions are used only at device
2321 * registration. Keep functions such as these at the end of this
2322 * structure to avoid cache line misses when accessing struct ib_device
2323 * in fast paths.
2324 */
2325 int (*get_port_immutable)(struct ib_device *device, u8 port_num,
2326 struct ib_port_immutable *immutable);
2327 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
2328 u8 port_num);
2329 /**
2330 * When calling get_netdev, the HW vendor's driver should return the
2331 * net device of device @device at port @port_num or NULL if such
2332 * a net device doesn't exist. The vendor driver should call dev_hold
2333 * on this net device. The HW vendor's device driver must guarantee
2334 * that this function returns NULL before the net device has finished
2335 * NETDEV_UNREGISTER state.
2336 */
2337 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num);
2338 /**
2339 * rdma netdev operation
2340 *
2341 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params
2342 * must return -EOPNOTSUPP if it doesn't support the specified type.
2343 */
2344 struct net_device *(*alloc_rdma_netdev)(
2345 struct ib_device *device, u8 port_num, enum rdma_netdev_t type,
2346 const char *name, unsigned char name_assign_type,
2347 void (*setup)(struct net_device *));
2348
2349 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num,
2350 enum rdma_netdev_t type,
2351 struct rdma_netdev_alloc_params *params);
2352 /**
2353 * query_gid should be return GID value for @device, when @port_num
2354 * link layer is either IB or iWarp. It is no-op if @port_num port
2355 * is RoCE link layer.
2356 */
2357 int (*query_gid)(struct ib_device *device, u8 port_num, int index,
2358 union ib_gid *gid);
2359 /**
2360 * When calling add_gid, the HW vendor's driver should add the gid
2361 * of device of port at gid index available at @attr. Meta-info of
2362 * that gid (for example, the network device related to this gid) is
2363 * available at @attr. @context allows the HW vendor driver to store
2364 * extra information together with a GID entry. The HW vendor driver may
2365 * allocate memory to contain this information and store it in @context
2366 * when a new GID entry is written to. Params are consistent until the
2367 * next call of add_gid or delete_gid. The function should return 0 on
2368 * success or error otherwise. The function could be called
2369 * concurrently for different ports. This function is only called when
2370 * roce_gid_table is used.
2371 */
2372 int (*add_gid)(const struct ib_gid_attr *attr, void **context);
2373 /**
2374 * When calling del_gid, the HW vendor's driver should delete the
2375 * gid of device @device at gid index gid_index of port port_num
2376 * available in @attr.
2377 * Upon the deletion of a GID entry, the HW vendor must free any
2378 * allocated memory. The caller will clear @context afterwards.
2379 * This function is only called when roce_gid_table is used.
2380 */
2381 int (*del_gid)(const struct ib_gid_attr *attr, void **context);
2382 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index,
2383 u16 *pkey);
a2a074ef
LR		 2384 int (*alloc_ucontext)(struct ib_ucontext *context,
2385 struct ib_udata *udata);
2386 void (*dealloc_ucontext)(struct ib_ucontext *context);
521ed0d9 2387 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma);
3411f9f0
MK		 2388 /**
2389 * This will be called once refcount of an entry in mmap_xa reaches
2390 * zero. The type of the memory that was mapped may differ between
2391 * entries and is opaque to the rdma_user_mmap interface.
2392 * Therefore needs to be implemented by the driver in mmap_free.
2393 */
2394 void (*mmap_free)(struct rdma_user_mmap_entry *entry);
521ed0d9 2395 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
ff23dfa1 2396 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
c4367a26 2397 void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata);
d3456914
LR		 2398 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr,
2399 u32 flags, struct ib_udata *udata);
521ed0d9
KH		 2400 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2401 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
d3456914 2402 void (*destroy_ah)(struct ib_ah *ah, u32 flags);
68e326de
LR		 2403 int (*create_srq)(struct ib_srq *srq,
2404 struct ib_srq_init_attr *srq_init_attr,
2405 struct ib_udata *udata);
521ed0d9
KH		 2406 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr,
2407 enum ib_srq_attr_mask srq_attr_mask,
2408 struct ib_udata *udata);
2409 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr);
68e326de 2410 void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata);
521ed0d9
KH		 2411 struct ib_qp *(*create_qp)(struct ib_pd *pd,
2412 struct ib_qp_init_attr *qp_init_attr,
2413 struct ib_udata *udata);
2414 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2415 int qp_attr_mask, struct ib_udata *udata);
2416 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
2417 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr);
c4367a26 2418 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata);
e39afe3d
LR		 2419 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr,
2420 struct ib_udata *udata);
521ed0d9 2421 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period);
a52c8e24 2422 void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata);
521ed0d9
KH		 2423 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata);
2424 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags);
2425 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length,
2426 u64 virt_addr, int mr_access_flags,
2427 struct ib_udata *udata);
2428 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length,
2429 u64 virt_addr, int mr_access_flags,
2430 struct ib_pd *pd, struct ib_udata *udata);
c4367a26 2431 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata);
521ed0d9 2432 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type,
c4367a26 2433 u32 max_num_sg, struct ib_udata *udata);
26bc7eae
IR		 2434 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd,
2435 u32 max_num_data_sg,
2436 u32 max_num_meta_sg);
ad8a4496
MS		 2437 int (*advise_mr)(struct ib_pd *pd,
2438 enum ib_uverbs_advise_mr_advice advice, u32 flags,
2439 struct ib_sge *sg_list, u32 num_sge,
2440 struct uverbs_attr_bundle *attrs);
521ed0d9
KH		 2441 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
2442 unsigned int *sg_offset);
2443 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2444 struct ib_mr_status *mr_status);
2445 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type,
2446 struct ib_udata *udata);
2447 int (*dealloc_mw)(struct ib_mw *mw);
2448 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags,
2449 struct ib_fmr_attr *fmr_attr);
2450 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len,
2451 u64 iova);
2452 int (*unmap_fmr)(struct list_head *fmr_list);
2453 int (*dealloc_fmr)(struct ib_fmr *fmr);
2454 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2455 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2456 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device,
521ed0d9 2457 struct ib_udata *udata);
c4367a26 2458 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata);
521ed0d9
KH		 2459 struct ib_flow *(*create_flow)(struct ib_qp *qp,
2460 struct ib_flow_attr *flow_attr,
2461 int domain, struct ib_udata *udata);
2462 int (*destroy_flow)(struct ib_flow *flow_id);
2463 struct ib_flow_action *(*create_flow_action_esp)(
2464 struct ib_device *device,
2465 const struct ib_flow_action_attrs_esp *attr,
2466 struct uverbs_attr_bundle *attrs);
2467 int (*destroy_flow_action)(struct ib_flow_action *action);
2468 int (*modify_flow_action_esp)(
2469 struct ib_flow_action *action,
2470 const struct ib_flow_action_attrs_esp *attr,
2471 struct uverbs_attr_bundle *attrs);
2472 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2473 int state);
2474 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2475 struct ifla_vf_info *ivf);
2476 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2477 struct ifla_vf_stats *stats);
bfcb3c5d
DG		 2478 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port,
2479 struct ifla_vf_guid *node_guid,
2480 struct ifla_vf_guid *port_guid);
521ed0d9
KH		 2481 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2482 int type);
2483 struct ib_wq *(*create_wq)(struct ib_pd *pd,
2484 struct ib_wq_init_attr *init_attr,
2485 struct ib_udata *udata);
a49b1dc7 2486 void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata);
521ed0d9
KH		 2487 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr,
2488 u32 wq_attr_mask, struct ib_udata *udata);
2489 struct ib_rwq_ind_table *(*create_rwq_ind_table)(
2490 struct ib_device *device,
2491 struct ib_rwq_ind_table_init_attr *init_attr,
2492 struct ib_udata *udata);
2493 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2494 struct ib_dm *(*alloc_dm)(struct ib_device *device,
2495 struct ib_ucontext *context,
2496 struct ib_dm_alloc_attr *attr,
2497 struct uverbs_attr_bundle *attrs);
c4367a26 2498 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs);
521ed0d9
KH		 2499 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm,
2500 struct ib_dm_mr_attr *attr,
2501 struct uverbs_attr_bundle *attrs);
2502 struct ib_counters *(*create_counters)(
2503 struct ib_device *device, struct uverbs_attr_bundle *attrs);
2504 int (*destroy_counters)(struct ib_counters *counters);
2505 int (*read_counters)(struct ib_counters *counters,
2506 struct ib_counters_read_attr *counters_read_attr,
2507 struct uverbs_attr_bundle *attrs);
2cdfcdd8
MG		 2508 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg,
2509 int data_sg_nents, unsigned int *data_sg_offset,
2510 struct scatterlist *meta_sg, int meta_sg_nents,
2511 unsigned int *meta_sg_offset);
2512
521ed0d9
KH		 2513 /**
2514 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2515 * driver initialized data. The struct is kfree()'ed by the sysfs
2516 * core when the device is removed. A lifespan of -1 in the return
2517 * struct tells the core to set a default lifespan.
2518 */
2519 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
2520 u8 port_num);
2521 /**
2522 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2523 * @index - The index in the value array we wish to have updated, or
2524 * num_counters if we want all stats updated
2525 * Return codes -
2526 * < 0 - Error, no counters updated
2527 * index - Updated the single counter pointed to by index
2528 * num_counters - Updated all counters (will reset the timestamp
2529 * and prevent further calls for lifespan milliseconds)
2530 * Drivers are allowed to update all counters in leiu of just the
2531 * one given in index at their option
2532 */
2533 int (*get_hw_stats)(struct ib_device *device,
2534 struct rdma_hw_stats *stats, u8 port, int index);
ea4baf7f
PP		 2535 /*
2536 * This function is called once for each port when a ib device is
2537 * registered.
2538 */
2539 int (*init_port)(struct ib_device *device, u8 port_num,
2540 struct kobject *port_sysfs);
02da3750
LR		 2541 /**
2542 * Allows rdma drivers to add their own restrack attributes.
2543 */
2544 int (*fill_res_entry)(struct sk_buff *msg,
2545 struct rdma_restrack_entry *entry);
21a428a0 2546
d0899892 2547 /* Device lifecycle callbacks */
ca22354b
JG		 2548 /*
2549 * Called after the device becomes registered, before clients are
2550 * attached
2551 */
2552 int (*enable_driver)(struct ib_device *dev);
d0899892
JG		 2553 /*
2554 * This is called as part of ib_dealloc_device().
2555 */
2556 void (*dealloc_driver)(struct ib_device *dev);
2557
dd05cb82
KH		 2558 /* iWarp CM callbacks */
2559 void (*iw_add_ref)(struct ib_qp *qp);
2560 void (*iw_rem_ref)(struct ib_qp *qp);
2561 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn);
2562 int (*iw_connect)(struct iw_cm_id *cm_id,
2563 struct iw_cm_conn_param *conn_param);
2564 int (*iw_accept)(struct iw_cm_id *cm_id,
2565 struct iw_cm_conn_param *conn_param);
2566 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata,
2567 u8 pdata_len);
2568 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog);
2569 int (*iw_destroy_listen)(struct iw_cm_id *cm_id);
99fa331d
MZ		 2570 /**
2571 * counter_bind_qp - Bind a QP to a counter.
2572 * @counter - The counter to be bound. If counter->id is zero then
2573 * the driver needs to allocate a new counter and set counter->id
2574 */
2575 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp);
2576 /**
2577 * counter_unbind_qp - Unbind the qp from the dynamically-allocated
2578 * counter and bind it onto the default one
2579 */
2580 int (*counter_unbind_qp)(struct ib_qp *qp);
2581 /**
2582 * counter_dealloc -De-allocate the hw counter
2583 */
2584 int (*counter_dealloc)(struct rdma_counter *counter);
c4ffee7c
MZ		 2585 /**
2586 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in
2587 * the driver initialized data.
2588 */
2589 struct rdma_hw_stats *(*counter_alloc_stats)(
2590 struct rdma_counter *counter);
2591 /**
2592 * counter_update_stats - Query the stats value of this counter
2593 */
2594 int (*counter_update_stats)(struct rdma_counter *counter);
dd05cb82 2595
4061ff7a
EA		 2596 /**
2597 * Allows rdma drivers to add their own restrack attributes
2598 * dumped via 'rdma stat' iproute2 command.
2599 */
2600 int (*fill_stat_entry)(struct sk_buff *msg,
2601 struct rdma_restrack_entry *entry);
2602
d3456914 2603 DECLARE_RDMA_OBJ_SIZE(ib_ah);
e39afe3d 2604 DECLARE_RDMA_OBJ_SIZE(ib_cq);
21a428a0 2605 DECLARE_RDMA_OBJ_SIZE(ib_pd);
68e326de 2606 DECLARE_RDMA_OBJ_SIZE(ib_srq);
a2a074ef 2607 DECLARE_RDMA_OBJ_SIZE(ib_ucontext);
521ed0d9
KH		 2608};
2609
cebe556b
PP		 2610struct ib_core_device {
2611 /* device must be the first element in structure until,
2612 * union of ib_core_device and device exists in ib_device.
2613 */
2614 struct device dev;
4e0f7b90 2615 possible_net_t rdma_net;
cebe556b
PP		 2616 struct kobject *ports_kobj;
2617 struct list_head port_list;
2618 struct ib_device *owner; /* reach back to owner ib_device */
2619};
41eda65c 2620
cebe556b 2621struct rdma_restrack_root;
1da177e4 2622struct ib_device {
0957c29f
BVA		 2623 /* Do not access @dma_device directly from ULP nor from HW drivers. */
2624 struct device *dma_device;
3023a1e9 2625 struct ib_device_ops ops;
1da177e4 2626 char name[IB_DEVICE_NAME_MAX];
324e227e 2627 struct rcu_head rcu_head;
1da177e4
LT		 2628
2629 struct list_head event_handler_list;
6b57cea9
PP		 2630 /* Protects event_handler_list */
2631 struct rw_semaphore event_handler_rwsem;
2632
2633 /* Protects QP's event_handler calls and open_qp list */
2634 spinlock_t event_handler_lock;
1da177e4 2635
921eab11 2636 struct rw_semaphore client_data_rwsem;
0df91bb6 2637 struct xarray client_data;
d0899892 2638 struct mutex unregistration_lock;
1da177e4 2639
17e10646
PP		 2640 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */
2641 rwlock_t cache_lock;
7738613e 2642 /**
8ceb1357 2643 * port_data is indexed by port number
7738613e 2644 */
8ceb1357 2645 struct ib_port_data *port_data;
1da177e4 2646
f4fd0b22
MT		 2647 int num_comp_vectors;
2648
cebe556b
PP		 2649 union {
2650 struct device dev;
2651 struct ib_core_device coredev;
2652 };
2653
d4122f5a
PP		 2654 /* First group for device attributes,
2655 * Second group for driver provided attributes (optional).
2656 * It is NULL terminated array.
2657 */
2658 const struct attribute_group *groups[3];
adee9f3f 2659
17a55f79 2660 u64 uverbs_cmd_mask;
f21519b2 2661 u64 uverbs_ex_cmd_mask;
274c0891 2662
bd99fdea 2663 char node_desc[IB_DEVICE_NODE_DESC_MAX];
cf311cd4 2664 __be64 node_guid;
96f15c03 2665 u32 local_dma_lkey;
4139032b 2666 u16 is_switch:1;
6780c4fa
GP		 2667 /* Indicates kernel verbs support, should not be used in drivers */
2668 u16 kverbs_provider:1;
da662979
YF		 2669 /* CQ adaptive moderation (RDMA DIM) */
2670 u16 use_cq_dim:1;
1da177e4
LT		 2671 u8 node_type;
2672 u8 phys_port_cnt;
3e153a93 2673 struct ib_device_attr attrs;
b40f4757
CL		 2674 struct attribute_group *hw_stats_ag;
2675 struct rdma_hw_stats *hw_stats;
7738613e 2676
43579b5f
PP		 2677#ifdef CONFIG_CGROUP_RDMA
2678 struct rdmacg_device cg_device;
2679#endif
2680
ecc82c53 2681 u32 index;
41eda65c 2682 struct rdma_restrack_root *res;
ecc82c53 2683
0cbf432d 2684 const struct uapi_definition *driver_def;
d79af724 2685
01b67117 2686 /*
d79af724
JG		 2687 * Positive refcount indicates that the device is currently
2688 * registered and cannot be unregistered.
01b67117
PP		 2689 */
2690 refcount_t refcount;
2691 struct completion unreg_completion;
d0899892 2692 struct work_struct unregistration_work;
3856ec4b
SW		 2693
2694 const struct rdma_link_ops *link_ops;
4e0f7b90
PP		 2695
2696 /* Protects compat_devs xarray modifications */
2697 struct mutex compat_devs_mutex;
2698 /* Maintains compat devices for each net namespace */
2699 struct xarray compat_devs;
dd05cb82
KH		 2700
2701 /* Used by iWarp CM */
2702 char iw_ifname[IFNAMSIZ];
2703 u32 iw_driver_flags;
1da177e4
LT		 2704};
2705
0e2d00eb 2706struct ib_client_nl_info;
1da177e4 2707struct ib_client {
e59178d8 2708 const char *name;
1da177e4 2709 void (*add) (struct ib_device *);
7c1eb45a 2710 void (*remove)(struct ib_device *, void *client_data);
dc1435c0 2711 void (*rename)(struct ib_device *dev, void *client_data);
0e2d00eb
JG		 2712 int (*get_nl_info)(struct ib_device *ibdev, void *client_data,
2713 struct ib_client_nl_info *res);
2714 int (*get_global_nl_info)(struct ib_client_nl_info *res);
1da177e4 2715
9268f72d
YK		 2716 /* Returns the net_dev belonging to this ib_client and matching the
2717 * given parameters.
2718 * @dev: An RDMA device that the net_dev use for communication.
2719 * @port: A physical port number on the RDMA device.
2720 * @pkey: P_Key that the net_dev uses if applicable.
2721 * @gid: A GID that the net_dev uses to communicate.
2722 * @addr: An IP address the net_dev is configured with.
2723 * @client_data: The device's client data set by ib_set_client_data().
2724 *
2725 * An ib_client that implements a net_dev on top of RDMA devices
2726 * (such as IP over IB) should implement this callback, allowing the
2727 * rdma_cm module to find the right net_dev for a given request.
2728 *
2729 * The caller is responsible for calling dev_put on the returned
2730 * netdev. */
2731 struct net_device *(*get_net_dev_by_params)(
2732 struct ib_device *dev,
2733 u8 port,
2734 u16 pkey,
2735 const union ib_gid *gid,
2736 const struct sockaddr *addr,
2737 void *client_data);
621e55ff
JG		 2738
2739 refcount_t uses;
2740 struct completion uses_zero;
e59178d8 2741 u32 client_id;
6780c4fa
GP		 2742
2743 /* kverbs are not required by the client */
2744 u8 no_kverbs_req:1;
1da177e4
LT		 2745};
2746
a808273a
SS		 2747/*
2748 * IB block DMA iterator
2749 *
2750 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
2751 * to a HW supported page size.
2752 */
2753struct ib_block_iter {
2754 /* internal states */
2755 struct scatterlist *__sg; /* sg holding the current aligned block */
2756 dma_addr_t __dma_addr; /* unaligned DMA address of this block */
2757 unsigned int __sg_nents; /* number of SG entries */
2758 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */
2759 unsigned int __pg_bit; /* alignment of current block */
2760};
2761
459cc69f
LR		 2762struct ib_device *_ib_alloc_device(size_t size);
2763#define ib_alloc_device(drv_struct, member) \
2764 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \
2765 BUILD_BUG_ON_ZERO(offsetof( \
2766 struct drv_struct, member))), \
2767 struct drv_struct, member)
2768
1da177e4
LT		 2769void ib_dealloc_device(struct ib_device *device);
2770
9abb0d1b 2771void ib_get_device_fw_str(struct ib_device *device, char *str);
5fa76c20 2772
ea4baf7f 2773int ib_register_device(struct ib_device *device, const char *name);
1da177e4 2774void ib_unregister_device(struct ib_device *device);
d0899892
JG		 2775void ib_unregister_driver(enum rdma_driver_id driver_id);
2776void ib_unregister_device_and_put(struct ib_device *device);
2777void ib_unregister_device_queued(struct ib_device *ib_dev);
1da177e4
LT		 2778
2779int ib_register_client (struct ib_client *client);
2780void ib_unregister_client(struct ib_client *client);
2781
a808273a
SS		 2782void __rdma_block_iter_start(struct ib_block_iter *biter,
2783 struct scatterlist *sglist,
2784 unsigned int nents,
2785 unsigned long pgsz);
2786bool __rdma_block_iter_next(struct ib_block_iter *biter);
2787
2788/**
2789 * rdma_block_iter_dma_address - get the aligned dma address of the current
2790 * block held by the block iterator.
2791 * @biter: block iterator holding the memory block
2792 */
2793static inline dma_addr_t
2794rdma_block_iter_dma_address(struct ib_block_iter *biter)
2795{
2796 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
2797}
2798
2799/**
2800 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
2801 * @sglist: sglist to iterate over
2802 * @biter: block iterator holding the memory block
2803 * @nents: maximum number of sg entries to iterate over
2804 * @pgsz: best HW supported page size to use
2805 *
2806 * Callers may use rdma_block_iter_dma_address() to get each
2807 * blocks aligned DMA address.
2808 */
2809#define rdma_for_each_block(sglist, biter, nents, pgsz) \
2810 for (__rdma_block_iter_start(biter, sglist, nents, \
2811 pgsz); \
2812 __rdma_block_iter_next(biter);)
2813
0df91bb6
JG		 2814/**
2815 * ib_get_client_data - Get IB client context
2816 * @device:Device to get context for
2817 * @client:Client to get context for
2818 *
2819 * ib_get_client_data() returns the client context data set with
2820 * ib_set_client_data(). This can only be called while the client is
2821 * registered to the device, once the ib_client remove() callback returns this
2822 * cannot be called.
2823 */
2824static inline void *ib_get_client_data(struct ib_device *device,
2825 struct ib_client *client)
2826{
2827 return xa_load(&device->client_data, client->client_id);
2828}
1da177e4
LT		 2829void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2830 void *data);
521ed0d9
KH		 2831void ib_set_device_ops(struct ib_device *device,
2832 const struct ib_device_ops *ops);
1da177e4 2833
5f9794dc 2834int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma,
c043ff2c
MK		 2835 unsigned long pfn, unsigned long size, pgprot_t prot,
2836 struct rdma_user_mmap_entry *entry);
3411f9f0
MK		 2837int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext,
2838 struct rdma_user_mmap_entry *entry,
2839 size_t length);
7a763d18
YH		 2840int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext,
2841 struct rdma_user_mmap_entry *entry,
2842 size_t length, u32 min_pgoff,
2843 u32 max_pgoff);
2844
3411f9f0
MK		 2845struct rdma_user_mmap_entry *
2846rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext,
2847 unsigned long pgoff);
2848struct rdma_user_mmap_entry *
2849rdma_user_mmap_entry_get(struct ib_ucontext *ucontext,
2850 struct vm_area_struct *vma);
2851void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry);
2852
2853void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry);
5f9794dc 2854
e2773c06
RD		 2855static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2856{
2857 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2858}
2859
2860static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2861{
43c61165 2862 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
e2773c06
RD		 2863}
2864
c66db311
MB		 2865static inline bool ib_is_buffer_cleared(const void __user *p,
2866 size_t len)
301a721e 2867{
92d27ae6 2868 bool ret;
301a721e
MB		 2869 u8 *buf;
2870
2871 if (len > USHRT_MAX)
2872 return false;
2873
92d27ae6
ME		 2874 buf = memdup_user(p, len);
2875 if (IS_ERR(buf))
301a721e
MB		 2876 return false;
2877
301a721e 2878 ret = !memchr_inv(buf, 0, len);
301a721e
MB		 2879 kfree(buf);
2880 return ret;
2881}
2882
c66db311
MB		 2883static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2884 size_t offset,
2885 size_t len)
2886{
2887 return ib_is_buffer_cleared(udata->inbuf + offset, len);
2888}
2889
1c77483e
YH		 2890/**
2891 * ib_is_destroy_retryable - Check whether the uobject destruction
2892 * is retryable.
2893 * @ret: The initial destruction return code
2894 * @why: remove reason
2895 * @uobj: The uobject that is destroyed
2896 *
2897 * This function is a helper function that IB layer and low-level drivers
2898 * can use to consider whether the destruction of the given uobject is
2899 * retry-able.
2900 * It checks the original return code, if it wasn't success the destruction
2901 * is retryable according to the ucontext state (i.e. cleanup_retryable) and
2902 * the remove reason. (i.e. why).
2903 * Must be called with the object locked for destroy.
2904 */
2905static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why,
2906 struct ib_uobject *uobj)
2907{
2908 return ret && (why == RDMA_REMOVE_DESTROY ||
2909 uobj->context->cleanup_retryable);
2910}
2911
2912/**
2913 * ib_destroy_usecnt - Called during destruction to check the usecnt
2914 * @usecnt: The usecnt atomic
2915 * @why: remove reason
2916 * @uobj: The uobject that is destroyed
2917 *
2918 * Non-zero usecnts will block destruction unless destruction was triggered by
2919 * a ucontext cleanup.
2920 */
2921static inline int ib_destroy_usecnt(atomic_t *usecnt,
2922 enum rdma_remove_reason why,
2923 struct ib_uobject *uobj)
2924{
2925 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj))
2926 return -EBUSY;
2927 return 0;
2928}
2929
8a51866f
RD		 2930/**
2931 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2932 * contains all required attributes and no attributes not allowed for
2933 * the given QP state transition.
2934 * @cur_state: Current QP state
2935 * @next_state: Next QP state
2936 * @type: QP type
2937 * @mask: Mask of supplied QP attributes
2938 *
2939 * This function is a helper function that a low-level driver's
2940 * modify_qp method can use to validate the consumer's input. It
2941 * checks that cur_state and next_state are valid QP states, that a
2942 * transition from cur_state to next_state is allowed by the IB spec,
2943 * and that the attribute mask supplied is allowed for the transition.
2944 */
19b1f540 2945bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
d31131bb 2946 enum ib_qp_type type, enum ib_qp_attr_mask mask);
8a51866f 2947
dcc9881e
LR		 2948void ib_register_event_handler(struct ib_event_handler *event_handler);
2949void ib_unregister_event_handler(struct ib_event_handler *event_handler);
6b57cea9 2950void ib_dispatch_event(const struct ib_event *event);
1da177e4 2951
1da177e4
LT		 2952int ib_query_port(struct ib_device *device,
2953 u8 port_num, struct ib_port_attr *port_attr);
2954
a3f5adaf
EC		 2955enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2956 u8 port_num);
2957
4139032b
HR		 2958/**
2959 * rdma_cap_ib_switch - Check if the device is IB switch
2960 * @device: Device to check
2961 *
2962 * Device driver is responsible for setting is_switch bit on
2963 * in ib_device structure at init time.
2964 *
2965 * Return: true if the device is IB switch.
2966 */
2967static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2968{
2969 return device->is_switch;
2970}
2971
0cf18d77
IW		 2972/**
2973 * rdma_start_port - Return the first valid port number for the device
2974 * specified
2975 *
2976 * @device: Device to be checked
2977 *
2978 * Return start port number
2979 */
2980static inline u8 rdma_start_port(const struct ib_device *device)
2981{
4139032b 2982 return rdma_cap_ib_switch(device) ? 0 : 1;
0cf18d77
IW		 2983}
2984
ea1075ed
JG		 2985/**
2986 * rdma_for_each_port - Iterate over all valid port numbers of the IB device
2987 * @device - The struct ib_device * to iterate over
2988 * @iter - The unsigned int to store the port number
2989 */
2990#define rdma_for_each_port(device, iter) \
2991 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \
2992 unsigned int, iter))); \
2993 iter <= rdma_end_port(device); (iter)++)
2994
0cf18d77
IW		 2995/**
2996 * rdma_end_port - Return the last valid port number for the device
2997 * specified
2998 *
2999 * @device: Device to be checked
3000 *
3001 * Return last port number
3002 */
3003static inline u8 rdma_end_port(const struct ib_device *device)
3004{
4139032b 3005 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
0cf18d77
IW		 3006}
3007
24dc831b
YS		 3008static inline int rdma_is_port_valid(const struct ib_device *device,
3009 unsigned int port)
3010{
3011 return (port >= rdma_start_port(device) &&
3012 port <= rdma_end_port(device));
3013}
3014
b02289b3
AK		 3015static inline bool rdma_is_grh_required(const struct ib_device *device,
3016 u8 port_num)
3017{
8ceb1357
JG		 3018 return device->port_data[port_num].immutable.core_cap_flags &
3019 RDMA_CORE_PORT_IB_GRH_REQUIRED;
b02289b3
AK		 3020}
3021
5ede9289 3022static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
de66be94 3023{
8ceb1357
JG		 3024 return device->port_data[port_num].immutable.core_cap_flags &
3025 RDMA_CORE_CAP_PROT_IB;
de66be94
MW		 3026}
3027
5ede9289 3028static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
7766a99f 3029{
8ceb1357
JG		 3030 return device->port_data[port_num].immutable.core_cap_flags &
3031 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
7766a99f
MB		 3032}
3033
3034static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
3035{
8ceb1357
JG		 3036 return device->port_data[port_num].immutable.core_cap_flags &
3037 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
7766a99f
MB		 3038}
3039
3040static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
de66be94 3041{
8ceb1357
JG		 3042 return device->port_data[port_num].immutable.core_cap_flags &
3043 RDMA_CORE_CAP_PROT_ROCE;
de66be94
MW		 3044}
3045
5ede9289 3046static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
de66be94 3047{
8ceb1357
JG		 3048 return device->port_data[port_num].immutable.core_cap_flags &
3049 RDMA_CORE_CAP_PROT_IWARP;
de66be94
MW		 3050}
3051
5ede9289 3052static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
de66be94 3053{
7766a99f
MB		 3054 return rdma_protocol_ib(device, port_num) ||
3055 rdma_protocol_roce(device, port_num);
de66be94
MW		 3056}
3057
aa773bd4
OG		 3058static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
3059{
8ceb1357
JG		 3060 return device->port_data[port_num].immutable.core_cap_flags &
3061 RDMA_CORE_CAP_PROT_RAW_PACKET;
aa773bd4
OG		 3062}
3063
ce1e055f
OG		 3064static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
3065{
8ceb1357
JG		 3066 return device->port_data[port_num].immutable.core_cap_flags &
3067 RDMA_CORE_CAP_PROT_USNIC;
ce1e055f
OG		 3068}
3069
c757dea8 3070/**
296ec009 3071 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
c757dea8 3072 * Management Datagrams.
296ec009
MW		 3073 * @device: Device to check
3074 * @port_num: Port number to check
c757dea8 3075 *
296ec009
MW		 3076 * Management Datagrams (MAD) are a required part of the InfiniBand
3077 * specification and are supported on all InfiniBand devices. A slightly
3078 * extended version are also supported on OPA interfaces.
c757dea8 3079 *
296ec009 3080 * Return: true if the port supports sending/receiving of MAD packets.
c757dea8 3081 */
5ede9289 3082static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
c757dea8 3083{
8ceb1357
JG		 3084 return device->port_data[port_num].immutable.core_cap_flags &
3085 RDMA_CORE_CAP_IB_MAD;
c757dea8
MW		 3086}
3087
65995fee
IW		 3088/**
3089 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
3090 * Management Datagrams.
3091 * @device: Device to check
3092 * @port_num: Port number to check
3093 *
3094 * Intel OmniPath devices extend and/or replace the InfiniBand Management
3095 * datagrams with their own versions. These OPA MADs share many but not all of
3096 * the characteristics of InfiniBand MADs.
3097 *
3098 * OPA MADs differ in the following ways:
3099 *
3100 * 1) MADs are variable size up to 2K
3101 * IBTA defined MADs remain fixed at 256 bytes
3102 * 2) OPA SMPs must carry valid PKeys
3103 * 3) OPA SMP packets are a different format
3104 *
3105 * Return: true if the port supports OPA MAD packet formats.
3106 */
3107static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
3108{
d3243da8
LR		 3109 return device->port_data[port_num].immutable.core_cap_flags &
3110 RDMA_CORE_CAP_OPA_MAD;
65995fee
IW		 3111}
3112
29541e3a 3113/**
296ec009
MW		 3114 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
3115 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
3116 * @device: Device to check
3117 * @port_num: Port number to check
29541e3a 3118 *
296ec009
MW		 3119 * Each InfiniBand node is required to provide a Subnet Management Agent
3120 * that the subnet manager can access. Prior to the fabric being fully
3121 * configured by the subnet manager, the SMA is accessed via a well known
3122 * interface called the Subnet Management Interface (SMI). This interface
3123 * uses directed route packets to communicate with the SM to get around the
3124 * chicken and egg problem of the SM needing to know what's on the fabric
3125 * in order to configure the fabric, and needing to configure the fabric in
3126 * order to send packets to the devices on the fabric. These directed
3127 * route packets do not need the fabric fully configured in order to reach
3128 * their destination. The SMI is the only method allowed to send
3129 * directed route packets on an InfiniBand fabric.
29541e3a 3130 *
296ec009 3131 * Return: true if the port provides an SMI.
29541e3a 3132 */
5ede9289 3133static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
29541e3a 3134{
8ceb1357
JG		 3135 return device->port_data[port_num].immutable.core_cap_flags &
3136 RDMA_CORE_CAP_IB_SMI;
29541e3a
MW		 3137}
3138
72219cea
MW		 3139/**
3140 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
3141 * Communication Manager.
296ec009
MW		 3142 * @device: Device to check
3143 * @port_num: Port number to check
72219cea 3144 *
296ec009
MW		 3145 * The InfiniBand Communication Manager is one of many pre-defined General
3146 * Service Agents (GSA) that are accessed via the General Service
3147 * Interface (GSI). It's role is to facilitate establishment of connections
3148 * between nodes as well as other management related tasks for established
3149 * connections.
72219cea 3150 *
296ec009
MW		 3151 * Return: true if the port supports an IB CM (this does not guarantee that
3152 * a CM is actually running however).
72219cea 3153 */
5ede9289 3154static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
72219cea 3155{
8ceb1357
JG		 3156 return device->port_data[port_num].immutable.core_cap_flags &
3157 RDMA_CORE_CAP_IB_CM;
72219cea
MW		 3158}
3159
04215330
MW		 3160/**
3161 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
3162 * Communication Manager.
296ec009
MW		 3163 * @device: Device to check
3164 * @port_num: Port number to check
04215330 3165 *
296ec009
MW		 3166 * Similar to above, but specific to iWARP connections which have a different
3167 * managment protocol than InfiniBand.
04215330 3168 *
296ec009
MW		 3169 * Return: true if the port supports an iWARP CM (this does not guarantee that
3170 * a CM is actually running however).
04215330 3171 */
5ede9289 3172static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
04215330 3173{
8ceb1357
JG		 3174 return device->port_data[port_num].immutable.core_cap_flags &
3175 RDMA_CORE_CAP_IW_CM;
04215330
MW		 3176}
3177
fe53ba2f
MW		 3178/**
3179 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
3180 * Subnet Administration.
296ec009
MW		 3181 * @device: Device to check
3182 * @port_num: Port number to check
fe53ba2f 3183 *
296ec009
MW		 3184 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
3185 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
3186 * fabrics, devices should resolve routes to other hosts by contacting the
3187 * SA to query the proper route.
fe53ba2f 3188 *
296ec009
MW		 3189 * Return: true if the port should act as a client to the fabric Subnet
3190 * Administration interface. This does not imply that the SA service is
3191 * running locally.
fe53ba2f 3192 */
5ede9289 3193static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
fe53ba2f 3194{
8ceb1357
JG		 3195 return device->port_data[port_num].immutable.core_cap_flags &
3196 RDMA_CORE_CAP_IB_SA;
fe53ba2f
MW		 3197}
3198
a31ad3b0
MW		 3199/**
3200 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
3201 * Multicast.
296ec009
MW		 3202 * @device: Device to check
3203 * @port_num: Port number to check
a31ad3b0 3204 *
296ec009
MW		 3205 * InfiniBand multicast registration is more complex than normal IPv4 or
3206 * IPv6 multicast registration. Each Host Channel Adapter must register
3207 * with the Subnet Manager when it wishes to join a multicast group. It
3208 * should do so only once regardless of how many queue pairs it subscribes
3209 * to this group. And it should leave the group only after all queue pairs
3210 * attached to the group have been detached.
a31ad3b0 3211 *
296ec009
MW		 3212 * Return: true if the port must undertake the additional adminstrative
3213 * overhead of registering/unregistering with the SM and tracking of the
3214 * total number of queue pairs attached to the multicast group.
a31ad3b0 3215 */
5ede9289 3216static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
a31ad3b0
MW		 3217{
3218 return rdma_cap_ib_sa(device, port_num);
3219}
3220
30a74ef4
MW		 3221/**
3222 * rdma_cap_af_ib - Check if the port of device has the capability
3223 * Native Infiniband Address.
296ec009
MW		 3224 * @device: Device to check
3225 * @port_num: Port number to check
30a74ef4 3226 *
296ec009
MW		 3227 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
3228 * GID. RoCE uses a different mechanism, but still generates a GID via
3229 * a prescribed mechanism and port specific data.
30a74ef4 3230 *
296ec009
MW		 3231 * Return: true if the port uses a GID address to identify devices on the
3232 * network.
30a74ef4 3233 */
5ede9289 3234static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
30a74ef4 3235{
8ceb1357
JG		 3236 return device->port_data[port_num].immutable.core_cap_flags &
3237 RDMA_CORE_CAP_AF_IB;
30a74ef4
MW		 3238}
3239
227128fc
MW		 3240/**
3241 * rdma_cap_eth_ah - Check if the port of device has the capability
296ec009
MW		 3242 * Ethernet Address Handle.
3243 * @device: Device to check
3244 * @port_num: Port number to check
227128fc 3245 *
296ec009
MW		 3246 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
3247 * to fabricate GIDs over Ethernet/IP specific addresses native to the
3248 * port. Normally, packet headers are generated by the sending host
3249 * adapter, but when sending connectionless datagrams, we must manually
3250 * inject the proper headers for the fabric we are communicating over.
227128fc 3251 *
296ec009
MW		 3252 * Return: true if we are running as a RoCE port and must force the
3253 * addition of a Global Route Header built from our Ethernet Address
3254 * Handle into our header list for connectionless packets.
227128fc 3255 */
5ede9289 3256static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
227128fc 3257{
8ceb1357
JG		 3258 return device->port_data[port_num].immutable.core_cap_flags &
3259 RDMA_CORE_CAP_ETH_AH;
227128fc
MW		 3260}
3261
94d595c5
DC		 3262/**
3263 * rdma_cap_opa_ah - Check if the port of device supports
3264 * OPA Address handles
3265 * @device: Device to check
3266 * @port_num: Port number to check
3267 *
3268 * Return: true if we are running on an OPA device which supports
3269 * the extended OPA addressing.
3270 */
3271static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
3272{
8ceb1357 3273 return (device->port_data[port_num].immutable.core_cap_flags &
94d595c5
DC		 3274 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
3275}
3276
337877a4
IW		 3277/**
3278 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
3279 *
3280 * @device: Device
3281 * @port_num: Port number
3282 *
3283 * This MAD size includes the MAD headers and MAD payload. No other headers
3284 * are included.
3285 *
3286 * Return the max MAD size required by the Port. Will return 0 if the port
3287 * does not support MADs
3288 */
3289static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
3290{
8ceb1357 3291 return device->port_data[port_num].immutable.max_mad_size;
337877a4
IW		 3292}
3293
03db3a2d
MB		 3294/**
3295 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
3296 * @device: Device to check
3297 * @port_num: Port number to check
3298 *
3299 * RoCE GID table mechanism manages the various GIDs for a device.
3300 *
3301 * NOTE: if allocating the port's GID table has failed, this call will still
3302 * return true, but any RoCE GID table API will fail.
3303 *
3304 * Return: true if the port uses RoCE GID table mechanism in order to manage
3305 * its GIDs.
3306 */
3307static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
3308 u8 port_num)
3309{
3310 return rdma_protocol_roce(device, port_num) &&
3023a1e9 3311 device->ops.add_gid && device->ops.del_gid;
03db3a2d
MB		 3312}
3313
002516ed
CH		 3314/*
3315 * Check if the device supports READ W/ INVALIDATE.
3316 */
3317static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
3318{
3319 /*
3320 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
3321 * has support for it yet.
3322 */
3323 return rdma_protocol_iwarp(dev, port_num);
3324}
3325
4a353399
SS		 3326/**
3327 * rdma_find_pg_bit - Find page bit given address and HW supported page sizes
3328 *
3329 * @addr: address
3330 * @pgsz_bitmap: bitmap of HW supported page sizes
3331 */
3332static inline unsigned int rdma_find_pg_bit(unsigned long addr,
3333 unsigned long pgsz_bitmap)
3334{
3335 unsigned long align;
3336 unsigned long pgsz;
3337
3338 align = addr & -addr;
3339
3340 /* Find page bit such that addr is aligned to the highest supported
3341 * HW page size
3342 */
3343 pgsz = pgsz_bitmap & ~(-align << 1);
3344 if (!pgsz)
3345 return __ffs(pgsz_bitmap);
3346
3347 return __fls(pgsz);
3348}
3349
50174a7f
EC		 3350int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
3351 int state);
3352int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
3353 struct ifla_vf_info *info);
3354int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
3355 struct ifla_vf_stats *stats);
bfcb3c5d
DG		 3356int ib_get_vf_guid(struct ib_device *device, int vf, u8 port,
3357 struct ifla_vf_guid *node_guid,
3358 struct ifla_vf_guid *port_guid);
50174a7f
EC		 3359int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
3360 int type);
3361
1da177e4
LT		 3362int ib_query_pkey(struct ib_device *device,
3363 u8 port_num, u16 index, u16 *pkey);
3364
3365int ib_modify_device(struct ib_device *device,
3366 int device_modify_mask,
3367 struct ib_device_modify *device_modify);
3368
3369int ib_modify_port(struct ib_device *device,
3370 u8 port_num, int port_modify_mask,
3371 struct ib_port_modify *port_modify);
3372
5eb620c8 3373int ib_find_gid(struct ib_device *device, union ib_gid *gid,
b26c4a11 3374 u8 *port_num, u16 *index);
5eb620c8
YE		 3375
3376int ib_find_pkey(struct ib_device *device,
3377 u8 port_num, u16 pkey, u16 *index);
3378
ed082d36
CH		 3379enum ib_pd_flags {
3380 /*
3381 * Create a memory registration for all memory in the system and place
3382 * the rkey for it into pd->unsafe_global_rkey. This can be used by
3383 * ULPs to avoid the overhead of dynamic MRs.
3384 *
3385 * This flag is generally considered unsafe and must only be used in
3386 * extremly trusted environments. Every use of it will log a warning
3387 * in the kernel log.
3388 */
3389 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
3390};
1da177e4 3391
ed082d36
CH		 3392struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
3393 const char *caller);
c4367a26 3394
ed082d36 3395#define ib_alloc_pd(device, flags) \
e4496447 3396 __ib_alloc_pd((device), (flags), KBUILD_MODNAME)
c4367a26
SR		 3397
3398/**
3399 * ib_dealloc_pd_user - Deallocate kernel/user PD
3400 * @pd: The protection domain
3401 * @udata: Valid user data or NULL for kernel objects
3402 */
3403void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata);
3404
3405/**
3406 * ib_dealloc_pd - Deallocate kernel PD
3407 * @pd: The protection domain
3408 *
3409 * NOTE: for user PD use ib_dealloc_pd_user with valid udata!
3410 */
3411static inline void ib_dealloc_pd(struct ib_pd *pd)
3412{
3413 ib_dealloc_pd_user(pd, NULL);
3414}
1da177e4 3415
b090c4e3
GP		 3416enum rdma_create_ah_flags {
3417 /* In a sleepable context */
3418 RDMA_CREATE_AH_SLEEPABLE = BIT(0),
3419};
3420
1da177e4 3421/**
0a18cfe4 3422 * rdma_create_ah - Creates an address handle for the given address vector.
1da177e4
LT		 3423 * @pd: The protection domain associated with the address handle.
3424 * @ah_attr: The attributes of the address vector.
b090c4e3 3425 * @flags: Create address handle flags (see enum rdma_create_ah_flags).
1da177e4
LT		 3426 *
3427 * The address handle is used to reference a local or global destination
3428 * in all UD QP post sends.
3429 */
b090c4e3
GP		 3430struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr,
3431 u32 flags);
1da177e4 3432
5cda6587
PP		 3433/**
3434 * rdma_create_user_ah - Creates an address handle for the given address vector.
3435 * It resolves destination mac address for ah attribute of RoCE type.
3436 * @pd: The protection domain associated with the address handle.
3437 * @ah_attr: The attributes of the address vector.
3438 * @udata: pointer to user's input output buffer information need by
3439 * provider driver.
3440 *
3441 * It returns 0 on success and returns appropriate error code on error.
3442 * The address handle is used to reference a local or global destination
3443 * in all UD QP post sends.
3444 */
3445struct ib_ah *rdma_create_user_ah(struct ib_pd *pd,
3446 struct rdma_ah_attr *ah_attr,
3447 struct ib_udata *udata);
850d8fd7
MS		 3448/**
3449 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
3450 * work completion.
3451 * @hdr: the L3 header to parse
3452 * @net_type: type of header to parse
3453 * @sgid: place to store source gid
3454 * @dgid: place to store destination gid
3455 */
3456int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
3457 enum rdma_network_type net_type,
3458 union ib_gid *sgid, union ib_gid *dgid);
3459
3460/**
3461 * ib_get_rdma_header_version - Get the header version
3462 * @hdr: the L3 header to parse
3463 */
3464int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
3465
4e00d694 3466/**
f6bdb142 3467 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a
4e00d694
SH		 3468 * work completion.
3469 * @device: Device on which the received message arrived.
3470 * @port_num: Port on which the received message arrived.
3471 * @wc: Work completion associated with the received message.
3472 * @grh: References the received global route header. This parameter is
3473 * ignored unless the work completion indicates that the GRH is valid.
3474 * @ah_attr: Returned attributes that can be used when creating an address
3475 * handle for replying to the message.
b7403217
PP		 3476 * When ib_init_ah_attr_from_wc() returns success,
3477 * (a) for IB link layer it optionally contains a reference to SGID attribute
3478 * when GRH is present for IB link layer.
3479 * (b) for RoCE link layer it contains a reference to SGID attribute.
3480 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID
3481 * attributes which are initialized using ib_init_ah_attr_from_wc().
3482 *
4e00d694 3483 */
f6bdb142
PP		 3484int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num,
3485 const struct ib_wc *wc, const struct ib_grh *grh,
3486 struct rdma_ah_attr *ah_attr);
4e00d694 3487
513789ed
HR		 3488/**
3489 * ib_create_ah_from_wc - Creates an address handle associated with the
3490 * sender of the specified work completion.
3491 * @pd: The protection domain associated with the address handle.
3492 * @wc: Work completion information associated with a received message.
3493 * @grh: References the received global route header. This parameter is
3494 * ignored unless the work completion indicates that the GRH is valid.
3495 * @port_num: The outbound port number to associate with the address.
3496 *
3497 * The address handle is used to reference a local or global destination
3498 * in all UD QP post sends.
3499 */
73cdaaee
IW		 3500struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
3501 const struct ib_grh *grh, u8 port_num);
513789ed 3502
1da177e4 3503/**
67b985b6 3504 * rdma_modify_ah - Modifies the address vector associated with an address
1da177e4
LT		 3505 * handle.
3506 * @ah: The address handle to modify.
3507 * @ah_attr: The new address vector attributes to associate with the
3508 * address handle.
3509 */
67b985b6 3510int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
1da177e4
LT		 3511
3512/**
bfbfd661 3513 * rdma_query_ah - Queries the address vector associated with an address
1da177e4
LT		 3514 * handle.
3515 * @ah: The address handle to query.
3516 * @ah_attr: The address vector attributes associated with the address
3517 * handle.
3518 */
bfbfd661 3519int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
1da177e4 3520
2553ba21
GP		 3521enum rdma_destroy_ah_flags {
3522 /* In a sleepable context */
3523 RDMA_DESTROY_AH_SLEEPABLE = BIT(0),
3524};
3525
1da177e4 3526/**
c4367a26 3527 * rdma_destroy_ah_user - Destroys an address handle.
1da177e4 3528 * @ah: The address handle to destroy.
2553ba21 3529 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
c4367a26 3530 * @udata: Valid user data or NULL for kernel objects
1da177e4 3531 */
c4367a26
SR		 3532int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata);
3533
3534/**
3535 * rdma_destroy_ah - Destroys an kernel address handle.
3536 * @ah: The address handle to destroy.
3537 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags).
3538 *
3539 * NOTE: for user ah use rdma_destroy_ah_user with valid udata!
3540 */
3541static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags)
3542{
3543 return rdma_destroy_ah_user(ah, flags, NULL);
3544}
1da177e4 3545
d41fcc67
RD		 3546/**
3547 * ib_create_srq - Creates a SRQ associated with the specified protection
3548 * domain.
3549 * @pd: The protection domain associated with the SRQ.
abb6e9ba
DB		 3550 * @srq_init_attr: A list of initial attributes required to create the
3551 * SRQ. If SRQ creation succeeds, then the attributes are updated to
3552 * the actual capabilities of the created SRQ.
d41fcc67
RD		 3553 *
3554 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
3555 * requested size of the SRQ, and set to the actual values allocated
3556 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
3557 * will always be at least as large as the requested values.
3558 */
3559struct ib_srq *ib_create_srq(struct ib_pd *pd,
3560 struct ib_srq_init_attr *srq_init_attr);
3561
3562/**
3563 * ib_modify_srq - Modifies the attributes for the specified SRQ.
3564 * @srq: The SRQ to modify.
3565 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
3566 * the current values of selected SRQ attributes are returned.
3567 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
3568 * are being modified.
3569 *
3570 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
3571 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
3572 * the number of receives queued drops below the limit.
3573 */
3574int ib_modify_srq(struct ib_srq *srq,
3575 struct ib_srq_attr *srq_attr,
3576 enum ib_srq_attr_mask srq_attr_mask);
3577
3578/**
3579 * ib_query_srq - Returns the attribute list and current values for the
3580 * specified SRQ.
3581 * @srq: The SRQ to query.
3582 * @srq_attr: The attributes of the specified SRQ.
3583 */
3584int ib_query_srq(struct ib_srq *srq,
3585 struct ib_srq_attr *srq_attr);
3586
3587/**
c4367a26
SR		 3588 * ib_destroy_srq_user - Destroys the specified SRQ.
3589 * @srq: The SRQ to destroy.
3590 * @udata: Valid user data or NULL for kernel objects
3591 */
3592int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata);
3593
3594/**
3595 * ib_destroy_srq - Destroys the specified kernel SRQ.
d41fcc67 3596 * @srq: The SRQ to destroy.
c4367a26
SR		 3597 *
3598 * NOTE: for user srq use ib_destroy_srq_user with valid udata!
d41fcc67 3599 */
c4367a26
SR		 3600static inline int ib_destroy_srq(struct ib_srq *srq)
3601{
3602 return ib_destroy_srq_user(srq, NULL);
3603}
d41fcc67
RD		 3604
3605/**
3606 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
3607 * @srq: The SRQ to post the work request on.
3608 * @recv_wr: A list of work requests to post on the receive queue.
3609 * @bad_recv_wr: On an immediate failure, this parameter will reference
3610 * the work request that failed to be posted on the QP.
3611 */
3612static inline int ib_post_srq_recv(struct ib_srq *srq,
d34ac5cd
BVA		 3613 const struct ib_recv_wr *recv_wr,
3614 const struct ib_recv_wr **bad_recv_wr)
d41fcc67 3615{
d34ac5cd 3616 const struct ib_recv_wr *dummy;
bb039a87 3617
3023a1e9
KH		 3618 return srq->device->ops.post_srq_recv(srq, recv_wr,
3619 bad_recv_wr ? : &dummy);
d41fcc67
RD		 3620}
3621
1da177e4 3622/**
c4367a26 3623 * ib_create_qp_user - Creates a QP associated with the specified protection
1da177e4
LT		 3624 * domain.
3625 * @pd: The protection domain associated with the QP.
abb6e9ba
DB		 3626 * @qp_init_attr: A list of initial attributes required to create the
3627 * QP. If QP creation succeeds, then the attributes are updated to
3628 * the actual capabilities of the created QP.
c4367a26 3629 * @udata: Valid user data or NULL for kernel objects
1da177e4 3630 */
c4367a26
SR		 3631struct ib_qp *ib_create_qp_user(struct ib_pd *pd,
3632 struct ib_qp_init_attr *qp_init_attr,
3633 struct ib_udata *udata);
3634
3635/**
3636 * ib_create_qp - Creates a kernel QP associated with the specified protection
3637 * domain.
3638 * @pd: The protection domain associated with the QP.
3639 * @qp_init_attr: A list of initial attributes required to create the
3640 * QP. If QP creation succeeds, then the attributes are updated to
3641 * the actual capabilities of the created QP.
3642 * @udata: Valid user data or NULL for kernel objects
3643 *
3644 * NOTE: for user qp use ib_create_qp_user with valid udata!
3645 */
3646static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
3647 struct ib_qp_init_attr *qp_init_attr)
3648{
3649 return ib_create_qp_user(pd, qp_init_attr, NULL);
3650}
1da177e4 3651
a512c2fb
PP		 3652/**
3653 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3654 * @qp: The QP to modify.
3655 * @attr: On input, specifies the QP attributes to modify. On output,
3656 * the current values of selected QP attributes are returned.
3657 * @attr_mask: A bit-mask used to specify which attributes of the QP
3658 * are being modified.
3659 * @udata: pointer to user's input output buffer information
3660 * are being modified.
3661 * It returns 0 on success and returns appropriate error code on error.
3662 */
3663int ib_modify_qp_with_udata(struct ib_qp *qp,
3664 struct ib_qp_attr *attr,
3665 int attr_mask,
3666 struct ib_udata *udata);
3667
1da177e4
LT		 3668/**
3669 * ib_modify_qp - Modifies the attributes for the specified QP and then
3670 * transitions the QP to the given state.
3671 * @qp: The QP to modify.
3672 * @qp_attr: On input, specifies the QP attributes to modify. On output,
3673 * the current values of selected QP attributes are returned.
3674 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3675 * are being modified.
3676 */
3677int ib_modify_qp(struct ib_qp *qp,
3678 struct ib_qp_attr *qp_attr,
3679 int qp_attr_mask);
3680
3681/**
3682 * ib_query_qp - Returns the attribute list and current values for the
3683 * specified QP.
3684 * @qp: The QP to query.
3685 * @qp_attr: The attributes of the specified QP.
3686 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3687 * @qp_init_attr: Additional attributes of the selected QP.
3688 *
3689 * The qp_attr_mask may be used to limit the query to gathering only the
3690 * selected attributes.
3691 */
3692int ib_query_qp(struct ib_qp *qp,
3693 struct ib_qp_attr *qp_attr,
3694 int qp_attr_mask,
3695 struct ib_qp_init_attr *qp_init_attr);
3696
3697/**
3698 * ib_destroy_qp - Destroys the specified QP.
3699 * @qp: The QP to destroy.
c4367a26 3700 * @udata: Valid udata or NULL for kernel objects
1da177e4 3701 */
c4367a26
SR		 3702int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata);
3703
3704/**
3705 * ib_destroy_qp - Destroys the specified kernel QP.
3706 * @qp: The QP to destroy.
3707 *
3708 * NOTE: for user qp use ib_destroy_qp_user with valid udata!
3709 */
3710static inline int ib_destroy_qp(struct ib_qp *qp)
3711{
3712 return ib_destroy_qp_user(qp, NULL);
3713}
1da177e4 3714
d3d72d90 3715/**
0e0ec7e0
SH		 3716 * ib_open_qp - Obtain a reference to an existing sharable QP.
3717 * @xrcd - XRC domain
3718 * @qp_open_attr: Attributes identifying the QP to open.
3719 *
3720 * Returns a reference to a sharable QP.
3721 */
3722struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3723 struct ib_qp_open_attr *qp_open_attr);
3724
3725/**
3726 * ib_close_qp - Release an external reference to a QP.
d3d72d90
SH		 3727 * @qp: The QP handle to release
3728 *
0e0ec7e0
SH		 3729 * The opened QP handle is released by the caller. The underlying
3730 * shared QP is not destroyed until all internal references are released.
d3d72d90 3731 */
0e0ec7e0 3732int ib_close_qp(struct ib_qp *qp);
d3d72d90 3733
1da177e4
LT		 3734/**
3735 * ib_post_send - Posts a list of work requests to the send queue of
3736 * the specified QP.
3737 * @qp: The QP to post the work request on.
3738 * @send_wr: A list of work requests to post on the send queue.
3739 * @bad_send_wr: On an immediate failure, this parameter will reference
3740 * the work request that failed to be posted on the QP.
55464d46
BVA		 3741 *
3742 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3743 * error is returned, the QP state shall not be affected,
3744 * ib_post_send() will return an immediate error after queueing any
3745 * earlier work requests in the list.
1da177e4
LT		 3746 */
3747static inline int ib_post_send(struct ib_qp *qp,
d34ac5cd
BVA		 3748 const struct ib_send_wr *send_wr,
3749 const struct ib_send_wr **bad_send_wr)
1da177e4 3750{
d34ac5cd 3751 const struct ib_send_wr *dummy;
bb039a87 3752
3023a1e9 3753 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
1da177e4
LT		 3754}
3755
3756/**
3757 * ib_post_recv - Posts a list of work requests to the receive queue of
3758 * the specified QP.
3759 * @qp: The QP to post the work request on.
3760 * @recv_wr: A list of work requests to post on the receive queue.
3761 * @bad_recv_wr: On an immediate failure, this parameter will reference
3762 * the work request that failed to be posted on the QP.
3763 */
3764static inline int ib_post_recv(struct ib_qp *qp,
d34ac5cd
BVA		 3765 const struct ib_recv_wr *recv_wr,
3766 const struct ib_recv_wr **bad_recv_wr)
1da177e4 3767{
d34ac5cd 3768 const struct ib_recv_wr *dummy;
bb039a87 3769
3023a1e9 3770 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
1da177e4
LT		 3771}
3772
c4367a26
SR		 3773struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private,
3774 int nr_cqe, int comp_vector,
3775 enum ib_poll_context poll_ctx,
3776 const char *caller, struct ib_udata *udata);
3777
3778/**
3779 * ib_alloc_cq_user: Allocate kernel/user CQ
3780 * @dev: The IB device
3781 * @private: Private data attached to the CQE
3782 * @nr_cqe: Number of CQEs in the CQ
3783 * @comp_vector: Completion vector used for the IRQs
3784 * @poll_ctx: Context used for polling the CQ
3785 * @udata: Valid user data or NULL for kernel objects
3786 */
3787static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev,
3788 void *private, int nr_cqe,
3789 int comp_vector,
3790 enum ib_poll_context poll_ctx,
3791 struct ib_udata *udata)
3792{
3793 return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3794 KBUILD_MODNAME, udata);
3795}
3796
3797/**
3798 * ib_alloc_cq: Allocate kernel CQ
3799 * @dev: The IB device
3800 * @private: Private data attached to the CQE
3801 * @nr_cqe: Number of CQEs in the CQ
3802 * @comp_vector: Completion vector used for the IRQs
3803 * @poll_ctx: Context used for polling the CQ
3804 *
3805 * NOTE: for user cq use ib_alloc_cq_user with valid udata!
3806 */
3807static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3808 int nr_cqe, int comp_vector,
3809 enum ib_poll_context poll_ctx)
3810{
3811 return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx,
3812 NULL);
3813}
3814
20cf4e02
CL		 3815struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private,
3816 int nr_cqe, enum ib_poll_context poll_ctx,
3817 const char *caller);
3818
3819/**
3820 * ib_alloc_cq_any: Allocate kernel CQ
3821 * @dev: The IB device
3822 * @private: Private data attached to the CQE
3823 * @nr_cqe: Number of CQEs in the CQ
3824 * @poll_ctx: Context used for polling the CQ
3825 */
3826static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev,
3827 void *private, int nr_cqe,
3828 enum ib_poll_context poll_ctx)
3829{
3830 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx,
3831 KBUILD_MODNAME);
3832}
3833
c4367a26
SR		 3834/**
3835 * ib_free_cq_user - Free kernel/user CQ
3836 * @cq: The CQ to free
3837 * @udata: Valid user data or NULL for kernel objects
3838 */
3839void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3840
3841/**
3842 * ib_free_cq - Free kernel CQ
3843 * @cq: The CQ to free
3844 *
3845 * NOTE: for user cq use ib_free_cq_user with valid udata!
3846 */
3847static inline void ib_free_cq(struct ib_cq *cq)
3848{
3849 ib_free_cq_user(cq, NULL);
3850}
f66c8ba4 3851
14d3a3b2
CH		 3852int ib_process_cq_direct(struct ib_cq *cq, int budget);
3853
1da177e4
LT		 3854/**
3855 * ib_create_cq - Creates a CQ on the specified device.
3856 * @device: The device on which to create the CQ.
3857 * @comp_handler: A user-specified callback that is invoked when a
3858 * completion event occurs on the CQ.
3859 * @event_handler: A user-specified callback that is invoked when an
3860 * asynchronous event not associated with a completion occurs on the CQ.
3861 * @cq_context: Context associated with the CQ returned to the user via
3862 * the associated completion and event handlers.
8e37210b 3863 * @cq_attr: The attributes the CQ should be created upon.
1da177e4
LT		 3864 *
3865 * Users can examine the cq structure to determine the actual CQ size.
3866 */
7350cdd0
BP		 3867struct ib_cq *__ib_create_cq(struct ib_device *device,
3868 ib_comp_handler comp_handler,
3869 void (*event_handler)(struct ib_event *, void *),
3870 void *cq_context,
3871 const struct ib_cq_init_attr *cq_attr,
3872 const char *caller);
3873#define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \
3874 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME)
1da177e4
LT		 3875
3876/**
3877 * ib_resize_cq - Modifies the capacity of the CQ.
3878 * @cq: The CQ to resize.
3879 * @cqe: The minimum size of the CQ.
3880 *
3881 * Users can examine the cq structure to determine the actual CQ size.
3882 */
3883int ib_resize_cq(struct ib_cq *cq, int cqe);
3884
2dd57162 3885/**
4190b4e9 3886 * rdma_set_cq_moderation - Modifies moderation params of the CQ
2dd57162
EC		 3887 * @cq: The CQ to modify.
3888 * @cq_count: number of CQEs that will trigger an event
3889 * @cq_period: max period of time in usec before triggering an event
3890 *
3891 */
4190b4e9 3892int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2dd57162 3893
1da177e4 3894/**
c4367a26 3895 * ib_destroy_cq_user - Destroys the specified CQ.
1da177e4 3896 * @cq: The CQ to destroy.
c4367a26 3897 * @udata: Valid user data or NULL for kernel objects
1da177e4 3898 */
c4367a26
SR		 3899int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata);
3900
3901/**
3902 * ib_destroy_cq - Destroys the specified kernel CQ.
3903 * @cq: The CQ to destroy.
3904 *
3905 * NOTE: for user cq use ib_destroy_cq_user with valid udata!
3906 */
890ac8d9 3907static inline void ib_destroy_cq(struct ib_cq *cq)
c4367a26 3908{
890ac8d9 3909 ib_destroy_cq_user(cq, NULL);
c4367a26 3910}
1da177e4
LT		 3911
3912/**
3913 * ib_poll_cq - poll a CQ for completion(s)
3914 * @cq:the CQ being polled
3915 * @num_entries:maximum number of completions to return
3916 * @wc:array of at least @num_entries &struct ib_wc where completions
3917 * will be returned
3918 *
3919 * Poll a CQ for (possibly multiple) completions. If the return value
3920 * is < 0, an error occurred. If the return value is >= 0, it is the
3921 * number of completions returned. If the return value is
3922 * non-negative and < num_entries, then the CQ was emptied.
3923 */
3924static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3925 struct ib_wc *wc)
3926{
3023a1e9 3927 return cq->device->ops.poll_cq(cq, num_entries, wc);
1da177e4
LT		 3928}
3929
1da177e4
LT		 3930/**
3931 * ib_req_notify_cq - Request completion notification on a CQ.
3932 * @cq: The CQ to generate an event for.
ed23a727
RD		 3933 * @flags:
3934 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3935 * to request an event on the next solicited event or next work
3936 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3937 * may also be |ed in to request a hint about missed events, as
3938 * described below.
3939 *
3940 * Return Value:
3941 * < 0 means an error occurred while requesting notification
3942 * == 0 means notification was requested successfully, and if
3943 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3944 * were missed and it is safe to wait for another event. In
3945 * this case is it guaranteed that any work completions added
3946 * to the CQ since the last CQ poll will trigger a completion
3947 * notification event.
3948 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3949 * in. It means that the consumer must poll the CQ again to
3950 * make sure it is empty to avoid missing an event because of a
3951 * race between requesting notification and an entry being
3952 * added to the CQ. This return value means it is possible
3953 * (but not guaranteed) that a work completion has been added
3954 * to the CQ since the last poll without triggering a
3955 * completion notification event.
1da177e4
LT		 3956 */
3957static inline int ib_req_notify_cq(struct ib_cq *cq,
ed23a727 3958 enum ib_cq_notify_flags flags)
1da177e4 3959{
3023a1e9 3960 return cq->device->ops.req_notify_cq(cq, flags);
1da177e4
LT		 3961}
3962
3963/**
3964 * ib_req_ncomp_notif - Request completion notification when there are
3965 * at least the specified number of unreaped completions on the CQ.
3966 * @cq: The CQ to generate an event for.
3967 * @wc_cnt: The number of unreaped completions that should be on the
3968 * CQ before an event is generated.
3969 */
3970static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3971{
3023a1e9
KH		 3972 return cq->device->ops.req_ncomp_notif ?
3973 cq->device->ops.req_ncomp_notif(cq, wc_cnt) :
1da177e4
LT		 3974 -ENOSYS;
3975}
3976
9b513090
RC		 3977/**
3978 * ib_dma_mapping_error - check a DMA addr for error
3979 * @dev: The device for which the dma_addr was created
3980 * @dma_addr: The DMA address to check
3981 */
3982static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3983{
0957c29f 3984 return dma_mapping_error(dev->dma_device, dma_addr);
9b513090
RC		 3985}
3986
3987/**
3988 * ib_dma_map_single - Map a kernel virtual address to DMA address
3989 * @dev: The device for which the dma_addr is to be created
3990 * @cpu_addr: The kernel virtual address
3991 * @size: The size of the region in bytes
3992 * @direction: The direction of the DMA
3993 */
3994static inline u64 ib_dma_map_single(struct ib_device *dev,
3995 void *cpu_addr, size_t size,
3996 enum dma_data_direction direction)
3997{
0957c29f 3998 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
9b513090
RC		 3999}
4000
4001/**
4002 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
4003 * @dev: The device for which the DMA address was created
4004 * @addr: The DMA address
4005 * @size: The size of the region in bytes
4006 * @direction: The direction of the DMA
4007 */
4008static inline void ib_dma_unmap_single(struct ib_device *dev,
4009 u64 addr, size_t size,
4010 enum dma_data_direction direction)
4011{
0957c29f 4012 dma_unmap_single(dev->dma_device, addr, size, direction);
cb9fbc5c
AK		 4013}
4014
9b513090
RC		 4015/**
4016 * ib_dma_map_page - Map a physical page to DMA address
4017 * @dev: The device for which the dma_addr is to be created
4018 * @page: The page to be mapped
4019 * @offset: The offset within the page
4020 * @size: The size of the region in bytes
4021 * @direction: The direction of the DMA
4022 */
4023static inline u64 ib_dma_map_page(struct ib_device *dev,
4024 struct page *page,
4025 unsigned long offset,
4026 size_t size,
4027 enum dma_data_direction direction)
4028{
0957c29f 4029 return dma_map_page(dev->dma_device, page, offset, size, direction);
9b513090
RC		 4030}
4031
4032/**
4033 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
4034 * @dev: The device for which the DMA address was created
4035 * @addr: The DMA address
4036 * @size: The size of the region in bytes
4037 * @direction: The direction of the DMA
4038 */
4039static inline void ib_dma_unmap_page(struct ib_device *dev,
4040 u64 addr, size_t size,
4041 enum dma_data_direction direction)
4042{
0957c29f 4043 dma_unmap_page(dev->dma_device, addr, size, direction);
9b513090
RC		 4044}
4045
4046/**
4047 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
4048 * @dev: The device for which the DMA addresses are to be created
4049 * @sg: The array of scatter/gather entries
4050 * @nents: The number of scatter/gather entries
4051 * @direction: The direction of the DMA
4052 */
4053static inline int ib_dma_map_sg(struct ib_device *dev,
4054 struct scatterlist *sg, int nents,
4055 enum dma_data_direction direction)
4056{
0957c29f 4057 return dma_map_sg(dev->dma_device, sg, nents, direction);
9b513090
RC		 4058}
4059
4060/**
4061 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
4062 * @dev: The device for which the DMA addresses were created
4063 * @sg: The array of scatter/gather entries
4064 * @nents: The number of scatter/gather entries
4065 * @direction: The direction of the DMA
4066 */
4067static inline void ib_dma_unmap_sg(struct ib_device *dev,
4068 struct scatterlist *sg, int nents,
4069 enum dma_data_direction direction)
4070{
0957c29f 4071 dma_unmap_sg(dev->dma_device, sg, nents, direction);
9b513090
RC		 4072}
4073
cb9fbc5c
AK		 4074static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
4075 struct scatterlist *sg, int nents,
4076 enum dma_data_direction direction,
00085f1e 4077 unsigned long dma_attrs)
cb9fbc5c 4078{
0957c29f
BVA		 4079 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
4080 dma_attrs);
cb9fbc5c
AK		 4081}
4082
4083static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
4084 struct scatterlist *sg, int nents,
4085 enum dma_data_direction direction,
00085f1e 4086 unsigned long dma_attrs)
cb9fbc5c 4087{
0957c29f 4088 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
cb9fbc5c 4089}
9b513090 4090
0b5cb330
BVA		 4091/**
4092 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer
4093 * @dev: The device to query
4094 *
4095 * The returned value represents a size in bytes.
4096 */
4097static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev)
4098{
ecdfdfdb 4099 return dma_get_max_seg_size(dev->dma_device);
0b5cb330
BVA		 4100}
4101
9b513090
RC		 4102/**
4103 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
4104 * @dev: The device for which the DMA address was created
4105 * @addr: The DMA address
4106 * @size: The size of the region in bytes
4107 * @dir: The direction of the DMA
4108 */
4109static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
4110 u64 addr,
4111 size_t size,
4112 enum dma_data_direction dir)
4113{
0957c29f 4114 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
9b513090
RC		 4115}
4116
4117/**
4118 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
4119 * @dev: The device for which the DMA address was created
4120 * @addr: The DMA address
4121 * @size: The size of the region in bytes
4122 * @dir: The direction of the DMA
4123 */
4124static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
4125 u64 addr,
4126 size_t size,
4127 enum dma_data_direction dir)
4128{
0957c29f 4129 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
9b513090
RC		 4130}
4131
4132/**
4133 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
4134 * @dev: The device for which the DMA address is requested
4135 * @size: The size of the region to allocate in bytes
4136 * @dma_handle: A pointer for returning the DMA address of the region
4137 * @flag: memory allocator flags
4138 */
4139static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
4140 size_t size,
d43dbacf 4141 dma_addr_t *dma_handle,
9b513090
RC		 4142 gfp_t flag)
4143{
0957c29f 4144 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
9b513090
RC		 4145}
4146
4147/**
4148 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
4149 * @dev: The device for which the DMA addresses were allocated
4150 * @size: The size of the region
4151 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
4152 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
4153 */
4154static inline void ib_dma_free_coherent(struct ib_device *dev,
4155 size_t size, void *cpu_addr,
d43dbacf 4156 dma_addr_t dma_handle)
9b513090 4157{
0957c29f 4158 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
9b513090
RC		 4159}
4160
1da177e4 4161/**
c4367a26
SR		 4162 * ib_dereg_mr_user - Deregisters a memory region and removes it from the
4163 * HCA translation table.
4164 * @mr: The memory region to deregister.
4165 * @udata: Valid user data or NULL for kernel object
4166 *
4167 * This function can fail, if the memory region has memory windows bound to it.
4168 */
4169int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata);
4170
4171/**
4172 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the
1da177e4
LT		 4173 * HCA translation table.
4174 * @mr: The memory region to deregister.
7083e42e
SM		 4175 *
4176 * This function can fail, if the memory region has memory windows bound to it.
c4367a26
SR		 4177 *
4178 * NOTE: for user mr use ib_dereg_mr_user with valid udata!
1da177e4 4179 */
c4367a26
SR		 4180static inline int ib_dereg_mr(struct ib_mr *mr)
4181{
4182 return ib_dereg_mr_user(mr, NULL);
4183}
4184
4185struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type,
4186 u32 max_num_sg, struct ib_udata *udata);
1da177e4 4187
c4367a26
SR		 4188static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
4189 enum ib_mr_type mr_type, u32 max_num_sg)
4190{
4191 return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL);
4192}
00f7ec36 4193
26bc7eae
IR		 4194struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd,
4195 u32 max_num_data_sg,
4196 u32 max_num_meta_sg);
4197
00f7ec36
SW		 4198/**
4199 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
4200 * R_Key and L_Key.
4201 * @mr - struct ib_mr pointer to be updated.
4202 * @newkey - new key to be used.
4203 */
4204static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
4205{
4206 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
4207 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
4208}
4209
7083e42e
SM		 4210/**
4211 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
4212 * for calculating a new rkey for type 2 memory windows.
4213 * @rkey - the rkey to increment.
4214 */
4215static inline u32 ib_inc_rkey(u32 rkey)
4216{
4217 const u32 mask = 0x000000ff;
4218 return ((rkey + 1) & mask) | (rkey & ~mask);
4219}
4220
1da177e4
LT		 4221/**
4222 * ib_alloc_fmr - Allocates a unmapped fast memory region.
4223 * @pd: The protection domain associated with the unmapped region.
4224 * @mr_access_flags: Specifies the memory access rights.
4225 * @fmr_attr: Attributes of the unmapped region.
4226 *
4227 * A fast memory region must be mapped before it can be used as part of
4228 * a work request.
4229 */
4230struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
4231 int mr_access_flags,
4232 struct ib_fmr_attr *fmr_attr);
4233
4234/**
4235 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
4236 * @fmr: The fast memory region to associate with the pages.
4237 * @page_list: An array of physical pages to map to the fast memory region.
4238 * @list_len: The number of pages in page_list.
4239 * @iova: The I/O virtual address to use with the mapped region.
4240 */
4241static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
4242 u64 *page_list, int list_len,
4243 u64 iova)
4244{
3023a1e9 4245 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova);
1da177e4
LT		 4246}
4247
4248/**
4249 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
4250 * @fmr_list: A linked list of fast memory regions to unmap.
4251 */
4252int ib_unmap_fmr(struct list_head *fmr_list);
4253
4254/**
4255 * ib_dealloc_fmr - Deallocates a fast memory region.
4256 * @fmr: The fast memory region to deallocate.
4257 */
4258int ib_dealloc_fmr(struct ib_fmr *fmr);
4259
4260/**
4261 * ib_attach_mcast - Attaches the specified QP to a multicast group.
4262 * @qp: QP to attach to the multicast group. The QP must be type
4263 * IB_QPT_UD.
4264 * @gid: Multicast group GID.
4265 * @lid: Multicast group LID in host byte order.
4266 *
4267 * In order to send and receive multicast packets, subnet
4268 * administration must have created the multicast group and configured
4269 * the fabric appropriately. The port associated with the specified
4270 * QP must also be a member of the multicast group.
4271 */
4272int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4273
4274/**
4275 * ib_detach_mcast - Detaches the specified QP from a multicast group.
4276 * @qp: QP to detach from the multicast group.
4277 * @gid: Multicast group GID.
4278 * @lid: Multicast group LID in host byte order.
4279 */
4280int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
4281
59991f94
SH		 4282/**
4283 * ib_alloc_xrcd - Allocates an XRC domain.
4284 * @device: The device on which to allocate the XRC domain.
f66c8ba4 4285 * @caller: Module name for kernel consumers
59991f94 4286 */
f66c8ba4
LR		 4287struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller);
4288#define ib_alloc_xrcd(device) \
4289 __ib_alloc_xrcd((device), KBUILD_MODNAME)
59991f94
SH		 4290
4291/**
4292 * ib_dealloc_xrcd - Deallocates an XRC domain.
4293 * @xrcd: The XRC domain to deallocate.
c4367a26 4294 * @udata: Valid user data or NULL for kernel object
59991f94 4295 */
c4367a26 4296int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata);
59991f94 4297
1c636f80
EC		 4298static inline int ib_check_mr_access(int flags)
4299{
4300 /*
4301 * Local write permission is required if remote write or
4302 * remote atomic permission is also requested.
4303 */
4304 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
4305 !(flags & IB_ACCESS_LOCAL_WRITE))
4306 return -EINVAL;
4307
4308 return 0;
4309}
4310
08bb558a
JM		 4311static inline bool ib_access_writable(int access_flags)
4312{
4313 /*
4314 * We have writable memory backing the MR if any of the following
4315 * access flags are set. "Local write" and "remote write" obviously
4316 * require write access. "Remote atomic" can do things like fetch and
4317 * add, which will modify memory, and "MW bind" can change permissions
4318 * by binding a window.
4319 */
4320 return access_flags &
4321 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE |
4322 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND);
4323}
4324
1b01d335
SG		 4325/**
4326 * ib_check_mr_status: lightweight check of MR status.
4327 * This routine may provide status checks on a selected
4328 * ib_mr. first use is for signature status check.
4329 *
4330 * @mr: A memory region.
4331 * @check_mask: Bitmask of which checks to perform from
4332 * ib_mr_status_check enumeration.
4333 * @mr_status: The container of relevant status checks.
4334 * failed checks will be indicated in the status bitmask
4335 * and the relevant info shall be in the error item.
4336 */
4337int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
4338 struct ib_mr_status *mr_status);
4339
d79af724
JG		 4340/**
4341 * ib_device_try_get: Hold a registration lock
4342 * device: The device to lock
4343 *
4344 * A device under an active registration lock cannot become unregistered. It
4345 * is only possible to obtain a registration lock on a device that is fully
4346 * registered, otherwise this function returns false.
4347 *
4348 * The registration lock is only necessary for actions which require the
4349 * device to still be registered. Uses that only require the device pointer to
4350 * be valid should use get_device(&ibdev->dev) to hold the memory.
4351 *
4352 */
4353static inline bool ib_device_try_get(struct ib_device *dev)
4354{
4355 return refcount_inc_not_zero(&dev->refcount);
4356}
4357
4358void ib_device_put(struct ib_device *device);
324e227e
JG		 4359struct ib_device *ib_device_get_by_netdev(struct net_device *ndev,
4360 enum rdma_driver_id driver_id);
4361struct ib_device *ib_device_get_by_name(const char *name,
4362 enum rdma_driver_id driver_id);
9268f72d
YK		 4363struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
4364 u16 pkey, const union ib_gid *gid,
4365 const struct sockaddr *addr);
c2261dd7
JG		 4366int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev,
4367 unsigned int port);
4368struct net_device *ib_device_netdev(struct ib_device *dev, u8 port);
4369
5fd251c8
YH		 4370struct ib_wq *ib_create_wq(struct ib_pd *pd,
4371 struct ib_wq_init_attr *init_attr);
c4367a26 4372int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata);
5fd251c8
YH		 4373int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
4374 u32 wq_attr_mask);
6d39786b
YH		 4375struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
4376 struct ib_rwq_ind_table_init_attr*
4377 wq_ind_table_init_attr);
4378int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
9268f72d 4379
ff2ba993 4380int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
9aa8b321 4381 unsigned int *sg_offset, unsigned int page_size);
2cdfcdd8
MG		 4382int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg,
4383 int data_sg_nents, unsigned int *data_sg_offset,
4384 struct scatterlist *meta_sg, int meta_sg_nents,
4385 unsigned int *meta_sg_offset, unsigned int page_size);
4c67e2bf
SG		 4386
4387static inline int
ff2ba993 4388ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
9aa8b321 4389 unsigned int *sg_offset, unsigned int page_size)
4c67e2bf
SG		 4390{
4391 int n;
4392
ff2ba993 4393 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
4c67e2bf
SG		 4394 mr->iova = 0;
4395
4396 return n;
4397}
4398
ff2ba993 4399int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
9aa8b321 4400 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
4c67e2bf 4401
765d6774
SW		 4402void ib_drain_rq(struct ib_qp *qp);
4403void ib_drain_sq(struct ib_qp *qp);
4404void ib_drain_qp(struct ib_qp *qp);
850d8fd7 4405
d4186194 4406int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
2224c47a
DC		 4407
4408static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
4409{
44c58487
DC		 4410 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
4411 return attr->roce.dmac;
4412 return NULL;
2224c47a
DC		 4413}
4414
64b4646e 4415static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
2224c47a 4416{
44c58487 4417 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
64b4646e
DC		 4418 attr->ib.dlid = (u16)dlid;
4419 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4420 attr->opa.dlid = dlid;
2224c47a
DC		 4421}
4422
64b4646e 4423static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
2224c47a 4424{
44c58487
DC		 4425 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4426 return attr->ib.dlid;
64b4646e
DC		 4427 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4428 return attr->opa.dlid;
44c58487 4429 return 0;
2224c47a
DC		 4430}
4431
4432static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
4433{
4434 attr->sl = sl;
4435}
4436
4437static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
4438{
4439 return attr->sl;
4440}
4441
4442static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
4443 u8 src_path_bits)
4444{
44c58487
DC		 4445 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4446 attr->ib.src_path_bits = src_path_bits;
64b4646e
DC		 4447 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4448 attr->opa.src_path_bits = src_path_bits;
2224c47a
DC		 4449}
4450
4451static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
4452{
44c58487
DC		 4453 if (attr->type == RDMA_AH_ATTR_TYPE_IB)
4454 return attr->ib.src_path_bits;
64b4646e
DC		 4455 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4456 return attr->opa.src_path_bits;
44c58487 4457 return 0;
2224c47a
DC		 4458}
4459
d98bb7f7
DH		 4460static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
4461 bool make_grd)
4462{
4463 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4464 attr->opa.make_grd = make_grd;
4465}
4466
4467static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
4468{
4469 if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
4470 return attr->opa.make_grd;
4471 return false;
4472}
4473
2224c47a
DC		 4474static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
4475{
4476 attr->port_num = port_num;
4477}
4478
4479static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
4480{
4481 return attr->port_num;
4482}
4483
4484static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
4485 u8 static_rate)
4486{
4487 attr->static_rate = static_rate;
4488}
4489
4490static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
4491{
4492 return attr->static_rate;
4493}
4494
4495static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
4496 enum ib_ah_flags flag)
4497{
4498 attr->ah_flags = flag;
4499}
4500
4501static inline enum ib_ah_flags
4502 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
4503{
4504 return attr->ah_flags;
4505}
4506
4507static inline const struct ib_global_route
4508 *rdma_ah_read_grh(const struct rdma_ah_attr *attr)
4509{
4510 return &attr->grh;
4511}
4512
4513/*To retrieve and modify the grh */
4514static inline struct ib_global_route
4515 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
4516{
4517 return &attr->grh;
4518}
4519
4520static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
4521{
4522 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4523
4524 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
4525}
4526
4527static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
4528 __be64 prefix)
4529{
4530 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4531
4532 grh->dgid.global.subnet_prefix = prefix;
4533}
4534
4535static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
4536 __be64 if_id)
4537{
4538 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4539
4540 grh->dgid.global.interface_id = if_id;
4541}
4542
4543static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
4544 union ib_gid *dgid, u32 flow_label,
4545 u8 sgid_index, u8 hop_limit,
4546 u8 traffic_class)
4547{
4548 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
4549
4550 attr->ah_flags = IB_AH_GRH;
4551 if (dgid)
4552 grh->dgid = *dgid;
4553 grh->flow_label = flow_label;
4554 grh->sgid_index = sgid_index;
4555 grh->hop_limit = hop_limit;
4556 grh->traffic_class = traffic_class;
8d9ec9ad 4557 grh->sgid_attr = NULL;
2224c47a 4558}
44c58487 4559
8d9ec9ad
JG		 4560void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr);
4561void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid,
4562 u32 flow_label, u8 hop_limit, u8 traffic_class,
4563 const struct ib_gid_attr *sgid_attr);
d97099fe
JG		 4564void rdma_copy_ah_attr(struct rdma_ah_attr *dest,
4565 const struct rdma_ah_attr *src);
4566void rdma_replace_ah_attr(struct rdma_ah_attr *old,
4567 const struct rdma_ah_attr *new);
4568void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src);
8d9ec9ad 4569
87daac68
DH		 4570/**
4571 * rdma_ah_find_type - Return address handle type.
4572 *
4573 * @dev: Device to be checked
4574 * @port_num: Port number
4575 */
44c58487 4576static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
87daac68 4577 u8 port_num)
44c58487 4578{
a6532e71 4579 if (rdma_protocol_roce(dev, port_num))
44c58487 4580 return RDMA_AH_ATTR_TYPE_ROCE;
87daac68
DH		 4581 if (rdma_protocol_ib(dev, port_num)) {
4582 if (rdma_cap_opa_ah(dev, port_num))
4583 return RDMA_AH_ATTR_TYPE_OPA;
44c58487 4584 return RDMA_AH_ATTR_TYPE_IB;
87daac68
DH		 4585 }
4586
4587 return RDMA_AH_ATTR_TYPE_UNDEFINED;
44c58487 4588}
7db20ecd 4589
62ede777
HD		 4590/**
4591 * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
4592 * In the current implementation the only way to get
4593 * get the 32bit lid is from other sources for OPA.
4594 * For IB, lids will always be 16bits so cast the
4595 * value accordingly.
4596 *
4597 * @lid: A 32bit LID
4598 */
4599static inline u16 ib_lid_cpu16(u32 lid)
7db20ecd 4600{
62ede777
HD		 4601 WARN_ON_ONCE(lid & 0xFFFF0000);
4602 return (u16)lid;
7db20ecd
HD		 4603}
4604
62ede777
HD		 4605/**
4606 * ib_lid_be16 - Return lid in 16bit BE encoding.
4607 *
4608 * @lid: A 32bit LID
4609 */
4610static inline __be16 ib_lid_be16(u32 lid)
7db20ecd 4611{
62ede777
HD		 4612 WARN_ON_ONCE(lid & 0xFFFF0000);
4613 return cpu_to_be16((u16)lid);
7db20ecd 4614}
32043830 4615
c66cd353
SG		 4616/**
4617 * ib_get_vector_affinity - Get the affinity mappings of a given completion
4618 * vector
4619 * @device: the rdma device
4620 * @comp_vector: index of completion vector
4621 *
4622 * Returns NULL on failure, otherwise a corresponding cpu map of the
4623 * completion vector (returns all-cpus map if the device driver doesn't
4624 * implement get_vector_affinity).
4625 */
4626static inline const struct cpumask *
4627ib_get_vector_affinity(struct ib_device *device, int comp_vector)
4628{
4629 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
3023a1e9 4630 !device->ops.get_vector_affinity)
c66cd353
SG		 4631 return NULL;
4632
3023a1e9 4633 return device->ops.get_vector_affinity(device, comp_vector);
c66cd353
SG		 4634
4635}
4636
32f69e4b
DJ		 4637/**
4638 * rdma_roce_rescan_device - Rescan all of the network devices in the system
4639 * and add their gids, as needed, to the relevant RoCE devices.
4640 *
4641 * @device: the rdma device
4642 */
4643void rdma_roce_rescan_device(struct ib_device *ibdev);
4644
8313c10f 4645struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile);
7dc08dcf 4646
15a1b4be 4647int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs);
f6a8a19b
DD		 4648
4649struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num,
4650 enum rdma_netdev_t type, const char *name,
4651 unsigned char name_assign_type,
4652 void (*setup)(struct net_device *));
5d6b0cb3
DD		 4653
4654int rdma_init_netdev(struct ib_device *device, u8 port_num,
4655 enum rdma_netdev_t type, const char *name,
4656 unsigned char name_assign_type,
4657 void (*setup)(struct net_device *),
4658 struct net_device *netdev);
4659
d4122f5a
PP		 4660/**
4661 * rdma_set_device_sysfs_group - Set device attributes group to have
4662 * driver specific sysfs entries at
4663 * for infiniband class.
4664 *
4665 * @device: device pointer for which attributes to be created
4666 * @group: Pointer to group which should be added when device
4667 * is registered with sysfs.
4668 * rdma_set_device_sysfs_group() allows existing drivers to expose one
4669 * group per device to have sysfs attributes.
4670 *
4671 * NOTE: New drivers should not make use of this API; instead new device
4672 * parameter should be exposed via netlink command. This API and mechanism
4673 * exist only for existing drivers.
4674 */
4675static inline void
4676rdma_set_device_sysfs_group(struct ib_device *dev,
4677 const struct attribute_group *group)
4678{
4679 dev->groups[1] = group;
4680}
4681
54747231
PP		 4682/**
4683 * rdma_device_to_ibdev - Get ib_device pointer from device pointer
4684 *
4685 * @device: device pointer for which ib_device pointer to retrieve
4686 *
4687 * rdma_device_to_ibdev() retrieves ib_device pointer from device.
4688 *
4689 */
4690static inline struct ib_device *rdma_device_to_ibdev(struct device *device)
4691{
cebe556b
PP		 4692 struct ib_core_device *coredev =
4693 container_of(device, struct ib_core_device, dev);
4694
4695 return coredev->owner;
54747231
PP		 4696}
4697
4698/**
4699 * rdma_device_to_drv_device - Helper macro to reach back to driver's
4700 * ib_device holder structure from device pointer.
4701 *
4702 * NOTE: New drivers should not make use of this API; This API is only for
4703 * existing drivers who have exposed sysfs entries using
4704 * rdma_set_device_sysfs_group().
4705 */
4706#define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \
4707 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member)
41c61401
PP		 4708
4709bool rdma_dev_access_netns(const struct ib_device *device,
4710 const struct net *net);
1da177e4 4711#endif /* IB_VERBS_H */
Ubuntu Jammy Linux kernel mirror