2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
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:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
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.
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
39 #if !defined(IB_VERBS_H)
42 #include <linux/types.h>
43 #include <linux/device.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
59 #include <linux/if_link.h>
60 #include <linux/atomic.h>
61 #include <linux/mmu_notifier.h>
62 #include <asm/uaccess.h>
64 extern struct workqueue_struct *ib_wq;
65 extern struct workqueue_struct *ib_comp_wq;
75 extern union ib_gid zgid;
78 /* If link layer is Ethernet, this is RoCE V1 */
81 IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
85 #define ROCE_V2_UDP_DPORT 4791
87 enum ib_gid_type gid_type;
88 struct net_device *ndev;
92 /* IB values map to NodeInfo:NodeType. */
102 /* set the local administered indication */
103 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2,
106 enum rdma_transport_type {
108 RDMA_TRANSPORT_IWARP,
109 RDMA_TRANSPORT_USNIC,
110 RDMA_TRANSPORT_USNIC_UDP
113 enum rdma_protocol_type {
117 RDMA_PROTOCOL_USNIC_UDP
120 __attribute_const__ enum rdma_transport_type
121 rdma_node_get_transport(enum rdma_node_type node_type);
123 enum rdma_network_type {
125 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
130 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
132 if (network_type == RDMA_NETWORK_IPV4 ||
133 network_type == RDMA_NETWORK_IPV6)
134 return IB_GID_TYPE_ROCE_UDP_ENCAP;
136 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
137 return IB_GID_TYPE_IB;
140 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
143 if (gid_type == IB_GID_TYPE_IB)
144 return RDMA_NETWORK_IB;
146 if (ipv6_addr_v4mapped((struct in6_addr *)gid))
147 return RDMA_NETWORK_IPV4;
149 return RDMA_NETWORK_IPV6;
152 enum rdma_link_layer {
153 IB_LINK_LAYER_UNSPECIFIED,
154 IB_LINK_LAYER_INFINIBAND,
155 IB_LINK_LAYER_ETHERNET,
158 enum ib_device_cap_flags {
159 IB_DEVICE_RESIZE_MAX_WR = (1 << 0),
160 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1),
161 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2),
162 IB_DEVICE_RAW_MULTI = (1 << 3),
163 IB_DEVICE_AUTO_PATH_MIG = (1 << 4),
164 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5),
165 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6),
166 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7),
167 IB_DEVICE_SHUTDOWN_PORT = (1 << 8),
168 IB_DEVICE_INIT_TYPE = (1 << 9),
169 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10),
170 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11),
171 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12),
172 IB_DEVICE_SRQ_RESIZE = (1 << 13),
173 IB_DEVICE_N_NOTIFY_CQ = (1 << 14),
176 * This device supports a per-device lkey or stag that can be
177 * used without performing a memory registration for the local
178 * memory. Note that ULPs should never check this flag, but
179 * instead of use the local_dma_lkey flag in the ib_pd structure,
180 * which will always contain a usable lkey.
182 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15),
183 IB_DEVICE_RESERVED /* old SEND_W_INV */ = (1 << 16),
184 IB_DEVICE_MEM_WINDOW = (1 << 17),
186 * Devices should set IB_DEVICE_UD_IP_SUM if they support
187 * insertion of UDP and TCP checksum on outgoing UD IPoIB
188 * messages and can verify the validity of checksum for
189 * incoming messages. Setting this flag implies that the
190 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
192 IB_DEVICE_UD_IP_CSUM = (1 << 18),
193 IB_DEVICE_UD_TSO = (1 << 19),
194 IB_DEVICE_XRC = (1 << 20),
197 * This device supports the IB "base memory management extension",
198 * which includes support for fast registrations (IB_WR_REG_MR,
199 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should
200 * also be set by any iWarp device which must support FRs to comply
201 * to the iWarp verbs spec. iWarp devices also support the
202 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
205 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21),
206 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22),
207 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23),
208 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24),
209 IB_DEVICE_RC_IP_CSUM = (1 << 25),
210 IB_DEVICE_RAW_IP_CSUM = (1 << 26),
212 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
213 * support execution of WQEs that involve synchronization
214 * of I/O operations with single completion queue managed
217 IB_DEVICE_CROSS_CHANNEL = (1 << 27),
218 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29),
219 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30),
220 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31),
221 IB_DEVICE_SG_GAPS_REG = (1ULL << 32),
222 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33),
223 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34),
226 enum ib_signature_prot_cap {
227 IB_PROT_T10DIF_TYPE_1 = 1,
228 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
229 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
232 enum ib_signature_guard_cap {
233 IB_GUARD_T10DIF_CRC = 1,
234 IB_GUARD_T10DIF_CSUM = 1 << 1,
243 enum ib_odp_general_cap_bits {
244 IB_ODP_SUPPORT = 1 << 0,
247 enum ib_odp_transport_cap_bits {
248 IB_ODP_SUPPORT_SEND = 1 << 0,
249 IB_ODP_SUPPORT_RECV = 1 << 1,
250 IB_ODP_SUPPORT_WRITE = 1 << 2,
251 IB_ODP_SUPPORT_READ = 1 << 3,
252 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
256 uint64_t general_caps;
258 uint32_t rc_odp_caps;
259 uint32_t uc_odp_caps;
260 uint32_t ud_odp_caps;
261 } per_transport_caps;
264 enum ib_cq_creation_flags {
265 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
266 IB_CQ_FLAGS_IGNORE_OVERRUN = 1 << 1,
269 struct ib_cq_init_attr {
275 struct ib_device_attr {
277 __be64 sys_image_guid;
285 u64 device_cap_flags;
295 int max_qp_init_rd_atom;
296 int max_ee_init_rd_atom;
297 enum ib_atomic_cap atomic_cap;
298 enum ib_atomic_cap masked_atomic_cap;
305 int max_mcast_qp_attach;
306 int max_total_mcast_qp_attach;
313 unsigned int max_fast_reg_page_list_len;
315 u8 local_ca_ack_delay;
318 struct ib_odp_caps odp_caps;
319 uint64_t timestamp_mask;
320 uint64_t hca_core_clock; /* in KHZ */
331 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
334 case IB_MTU_256: return 256;
335 case IB_MTU_512: return 512;
336 case IB_MTU_1024: return 1024;
337 case IB_MTU_2048: return 2048;
338 case IB_MTU_4096: return 4096;
349 IB_PORT_ACTIVE_DEFER = 5
352 enum ib_port_cap_flags {
354 IB_PORT_NOTICE_SUP = 1 << 2,
355 IB_PORT_TRAP_SUP = 1 << 3,
356 IB_PORT_OPT_IPD_SUP = 1 << 4,
357 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
358 IB_PORT_SL_MAP_SUP = 1 << 6,
359 IB_PORT_MKEY_NVRAM = 1 << 7,
360 IB_PORT_PKEY_NVRAM = 1 << 8,
361 IB_PORT_LED_INFO_SUP = 1 << 9,
362 IB_PORT_SM_DISABLED = 1 << 10,
363 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
364 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
365 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
366 IB_PORT_CM_SUP = 1 << 16,
367 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
368 IB_PORT_REINIT_SUP = 1 << 18,
369 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
370 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
371 IB_PORT_DR_NOTICE_SUP = 1 << 21,
372 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
373 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
374 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
375 IB_PORT_CLIENT_REG_SUP = 1 << 25,
376 IB_PORT_IP_BASED_GIDS = 1 << 26,
386 static inline int ib_width_enum_to_int(enum ib_port_width width)
389 case IB_WIDTH_1X: return 1;
390 case IB_WIDTH_4X: return 4;
391 case IB_WIDTH_8X: return 8;
392 case IB_WIDTH_12X: return 12;
407 * struct rdma_hw_stats
408 * @timestamp - Used by the core code to track when the last update was
409 * @lifespan - Used by the core code to determine how old the counters
410 * should be before being updated again. Stored in jiffies, defaults
411 * to 10 milliseconds, drivers can override the default be specifying
412 * their own value during their allocation routine.
413 * @name - Array of pointers to static names used for the counters in
415 * @num_counters - How many hardware counters there are. If name is
416 * shorter than this number, a kernel oops will result. Driver authors
417 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
418 * in their code to prevent this.
419 * @value - Array of u64 counters that are accessed by the sysfs code and
420 * filled in by the drivers get_stats routine
422 struct rdma_hw_stats {
423 unsigned long timestamp;
424 unsigned long lifespan;
425 const char * const *names;
430 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
432 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
434 * @names - Array of static const char *
435 * @num_counters - How many elements in array
436 * @lifespan - How many milliseconds between updates
438 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
439 const char * const *names, int num_counters,
440 unsigned long lifespan)
442 struct rdma_hw_stats *stats;
444 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
448 stats->names = names;
449 stats->num_counters = num_counters;
450 stats->lifespan = msecs_to_jiffies(lifespan);
456 /* Define bits for the various functionality this port needs to be supported by
459 /* Management 0x00000FFF */
460 #define RDMA_CORE_CAP_IB_MAD 0x00000001
461 #define RDMA_CORE_CAP_IB_SMI 0x00000002
462 #define RDMA_CORE_CAP_IB_CM 0x00000004
463 #define RDMA_CORE_CAP_IW_CM 0x00000008
464 #define RDMA_CORE_CAP_IB_SA 0x00000010
465 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
467 /* Address format 0x000FF000 */
468 #define RDMA_CORE_CAP_AF_IB 0x00001000
469 #define RDMA_CORE_CAP_ETH_AH 0x00002000
471 /* Protocol 0xFFF00000 */
472 #define RDMA_CORE_CAP_PROT_IB 0x00100000
473 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
474 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
475 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
477 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
478 | RDMA_CORE_CAP_IB_MAD \
479 | RDMA_CORE_CAP_IB_SMI \
480 | RDMA_CORE_CAP_IB_CM \
481 | RDMA_CORE_CAP_IB_SA \
482 | RDMA_CORE_CAP_AF_IB)
483 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
484 | RDMA_CORE_CAP_IB_MAD \
485 | RDMA_CORE_CAP_IB_CM \
486 | RDMA_CORE_CAP_AF_IB \
487 | RDMA_CORE_CAP_ETH_AH)
488 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \
489 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
490 | RDMA_CORE_CAP_IB_MAD \
491 | RDMA_CORE_CAP_IB_CM \
492 | RDMA_CORE_CAP_AF_IB \
493 | RDMA_CORE_CAP_ETH_AH)
494 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
495 | RDMA_CORE_CAP_IW_CM)
496 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
497 | RDMA_CORE_CAP_OPA_MAD)
499 struct ib_port_attr {
501 enum ib_port_state state;
503 enum ib_mtu active_mtu;
523 enum ib_device_modify_flags {
524 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
525 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
528 struct ib_device_modify {
533 enum ib_port_modify_flags {
534 IB_PORT_SHUTDOWN = 1,
535 IB_PORT_INIT_TYPE = (1<<2),
536 IB_PORT_RESET_QKEY_CNTR = (1<<3)
539 struct ib_port_modify {
540 u32 set_port_cap_mask;
541 u32 clr_port_cap_mask;
549 IB_EVENT_QP_ACCESS_ERR,
553 IB_EVENT_PATH_MIG_ERR,
554 IB_EVENT_DEVICE_FATAL,
555 IB_EVENT_PORT_ACTIVE,
558 IB_EVENT_PKEY_CHANGE,
561 IB_EVENT_SRQ_LIMIT_REACHED,
562 IB_EVENT_QP_LAST_WQE_REACHED,
563 IB_EVENT_CLIENT_REREGISTER,
568 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
571 struct ib_device *device;
579 enum ib_event_type event;
582 struct ib_event_handler {
583 struct ib_device *device;
584 void (*handler)(struct ib_event_handler *, struct ib_event *);
585 struct list_head list;
588 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
590 (_ptr)->device = _device; \
591 (_ptr)->handler = _handler; \
592 INIT_LIST_HEAD(&(_ptr)->list); \
595 struct ib_global_route {
604 __be32 version_tclass_flow;
612 union rdma_network_hdr {
615 /* The IB spec states that if it's IPv4, the header
616 * is located in the last 20 bytes of the header.
619 struct iphdr roce4grh;
624 IB_MULTICAST_QPN = 0xffffff
627 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
628 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000)
635 IB_RATE_PORT_CURRENT = 0,
636 IB_RATE_2_5_GBPS = 2,
644 IB_RATE_120_GBPS = 10,
645 IB_RATE_14_GBPS = 11,
646 IB_RATE_56_GBPS = 12,
647 IB_RATE_112_GBPS = 13,
648 IB_RATE_168_GBPS = 14,
649 IB_RATE_25_GBPS = 15,
650 IB_RATE_100_GBPS = 16,
651 IB_RATE_200_GBPS = 17,
652 IB_RATE_300_GBPS = 18
656 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
657 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
658 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
659 * @rate: rate to convert.
661 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
664 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
665 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
666 * @rate: rate to convert.
668 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
672 * enum ib_mr_type - memory region type
673 * @IB_MR_TYPE_MEM_REG: memory region that is used for
674 * normal registration
675 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
676 * signature operations (data-integrity
678 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to
679 * register any arbitrary sg lists (without
680 * the normal mr constraints - see
685 IB_MR_TYPE_SIGNATURE,
691 * IB_SIG_TYPE_NONE: Unprotected.
692 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
694 enum ib_signature_type {
700 * Signature T10-DIF block-guard types
701 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
702 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
704 enum ib_t10_dif_bg_type {
710 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
712 * @bg_type: T10-DIF block guard type (CRC|CSUM)
713 * @pi_interval: protection information interval.
714 * @bg: seed of guard computation.
715 * @app_tag: application tag of guard block
716 * @ref_tag: initial guard block reference tag.
717 * @ref_remap: Indicate wethear the reftag increments each block
718 * @app_escape: Indicate to skip block check if apptag=0xffff
719 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
720 * @apptag_check_mask: check bitmask of application tag.
722 struct ib_t10_dif_domain {
723 enum ib_t10_dif_bg_type bg_type;
731 u16 apptag_check_mask;
735 * struct ib_sig_domain - Parameters for signature domain
736 * @sig_type: specific signauture type
737 * @sig: union of all signature domain attributes that may
738 * be used to set domain layout.
740 struct ib_sig_domain {
741 enum ib_signature_type sig_type;
743 struct ib_t10_dif_domain dif;
748 * struct ib_sig_attrs - Parameters for signature handover operation
749 * @check_mask: bitmask for signature byte check (8 bytes)
750 * @mem: memory domain layout desciptor.
751 * @wire: wire domain layout desciptor.
753 struct ib_sig_attrs {
755 struct ib_sig_domain mem;
756 struct ib_sig_domain wire;
759 enum ib_sig_err_type {
766 * struct ib_sig_err - signature error descriptor
769 enum ib_sig_err_type err_type;
776 enum ib_mr_status_check {
777 IB_MR_CHECK_SIG_STATUS = 1,
781 * struct ib_mr_status - Memory region status container
783 * @fail_status: Bitmask of MR checks status. For each
784 * failed check a corresponding status bit is set.
785 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
788 struct ib_mr_status {
790 struct ib_sig_err sig_err;
794 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
796 * @mult: multiple to convert.
798 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
801 struct ib_global_route grh;
815 IB_WC_LOC_EEC_OP_ERR,
820 IB_WC_LOC_ACCESS_ERR,
821 IB_WC_REM_INV_REQ_ERR,
822 IB_WC_REM_ACCESS_ERR,
825 IB_WC_RNR_RETRY_EXC_ERR,
826 IB_WC_LOC_RDD_VIOL_ERR,
827 IB_WC_REM_INV_RD_REQ_ERR,
830 IB_WC_INV_EEC_STATE_ERR,
832 IB_WC_RESP_TIMEOUT_ERR,
836 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
847 IB_WC_MASKED_COMP_SWAP,
848 IB_WC_MASKED_FETCH_ADD,
850 * Set value of IB_WC_RECV so consumers can test if a completion is a
851 * receive by testing (opcode & IB_WC_RECV).
854 IB_WC_RECV_RDMA_WITH_IMM
859 IB_WC_WITH_IMM = (1<<1),
860 IB_WC_WITH_INVALIDATE = (1<<2),
861 IB_WC_IP_CSUM_OK = (1<<3),
862 IB_WC_WITH_SMAC = (1<<4),
863 IB_WC_WITH_VLAN = (1<<5),
864 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6),
870 struct ib_cqe *wr_cqe;
872 enum ib_wc_status status;
873 enum ib_wc_opcode opcode;
887 u8 port_num; /* valid only for DR SMPs on switches */
893 enum ib_cq_notify_flags {
894 IB_CQ_SOLICITED = 1 << 0,
895 IB_CQ_NEXT_COMP = 1 << 1,
896 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
897 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
905 enum ib_srq_attr_mask {
906 IB_SRQ_MAX_WR = 1 << 0,
907 IB_SRQ_LIMIT = 1 << 1,
916 struct ib_srq_init_attr {
917 void (*event_handler)(struct ib_event *, void *);
919 struct ib_srq_attr attr;
920 enum ib_srq_type srq_type;
924 struct ib_xrcd *xrcd;
938 * Maximum number of rdma_rw_ctx structures in flight at a time.
939 * ib_create_qp() will calculate the right amount of neededed WRs
940 * and MRs based on this.
952 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
953 * here (and in that order) since the MAD layer uses them as
954 * indices into a 2-entry table.
963 IB_QPT_RAW_ETHERTYPE,
964 IB_QPT_RAW_PACKET = 8,
968 /* Reserve a range for qp types internal to the low level driver.
969 * These qp types will not be visible at the IB core layer, so the
970 * IB_QPT_MAX usages should not be affected in the core layer
972 IB_QPT_RESERVED1 = 0x1000,
984 enum ib_qp_create_flags {
985 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
986 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
987 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2,
988 IB_QP_CREATE_MANAGED_SEND = 1 << 3,
989 IB_QP_CREATE_MANAGED_RECV = 1 << 4,
990 IB_QP_CREATE_NETIF_QP = 1 << 5,
991 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
992 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
993 IB_QP_CREATE_SCATTER_FCS = 1 << 8,
994 /* reserve bits 26-31 for low level drivers' internal use */
995 IB_QP_CREATE_RESERVED_START = 1 << 26,
996 IB_QP_CREATE_RESERVED_END = 1 << 31,
1000 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1001 * callback to destroy the passed in QP.
1004 struct ib_qp_init_attr {
1005 void (*event_handler)(struct ib_event *, void *);
1007 struct ib_cq *send_cq;
1008 struct ib_cq *recv_cq;
1010 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1011 struct ib_qp_cap cap;
1012 enum ib_sig_type sq_sig_type;
1013 enum ib_qp_type qp_type;
1014 enum ib_qp_create_flags create_flags;
1017 * Only needed for special QP types, or when using the RW API.
1020 struct ib_rwq_ind_table *rwq_ind_tbl;
1023 struct ib_qp_open_attr {
1024 void (*event_handler)(struct ib_event *, void *);
1027 enum ib_qp_type qp_type;
1030 enum ib_rnr_timeout {
1031 IB_RNR_TIMER_655_36 = 0,
1032 IB_RNR_TIMER_000_01 = 1,
1033 IB_RNR_TIMER_000_02 = 2,
1034 IB_RNR_TIMER_000_03 = 3,
1035 IB_RNR_TIMER_000_04 = 4,
1036 IB_RNR_TIMER_000_06 = 5,
1037 IB_RNR_TIMER_000_08 = 6,
1038 IB_RNR_TIMER_000_12 = 7,
1039 IB_RNR_TIMER_000_16 = 8,
1040 IB_RNR_TIMER_000_24 = 9,
1041 IB_RNR_TIMER_000_32 = 10,
1042 IB_RNR_TIMER_000_48 = 11,
1043 IB_RNR_TIMER_000_64 = 12,
1044 IB_RNR_TIMER_000_96 = 13,
1045 IB_RNR_TIMER_001_28 = 14,
1046 IB_RNR_TIMER_001_92 = 15,
1047 IB_RNR_TIMER_002_56 = 16,
1048 IB_RNR_TIMER_003_84 = 17,
1049 IB_RNR_TIMER_005_12 = 18,
1050 IB_RNR_TIMER_007_68 = 19,
1051 IB_RNR_TIMER_010_24 = 20,
1052 IB_RNR_TIMER_015_36 = 21,
1053 IB_RNR_TIMER_020_48 = 22,
1054 IB_RNR_TIMER_030_72 = 23,
1055 IB_RNR_TIMER_040_96 = 24,
1056 IB_RNR_TIMER_061_44 = 25,
1057 IB_RNR_TIMER_081_92 = 26,
1058 IB_RNR_TIMER_122_88 = 27,
1059 IB_RNR_TIMER_163_84 = 28,
1060 IB_RNR_TIMER_245_76 = 29,
1061 IB_RNR_TIMER_327_68 = 30,
1062 IB_RNR_TIMER_491_52 = 31
1065 enum ib_qp_attr_mask {
1067 IB_QP_CUR_STATE = (1<<1),
1068 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
1069 IB_QP_ACCESS_FLAGS = (1<<3),
1070 IB_QP_PKEY_INDEX = (1<<4),
1071 IB_QP_PORT = (1<<5),
1072 IB_QP_QKEY = (1<<6),
1074 IB_QP_PATH_MTU = (1<<8),
1075 IB_QP_TIMEOUT = (1<<9),
1076 IB_QP_RETRY_CNT = (1<<10),
1077 IB_QP_RNR_RETRY = (1<<11),
1078 IB_QP_RQ_PSN = (1<<12),
1079 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
1080 IB_QP_ALT_PATH = (1<<14),
1081 IB_QP_MIN_RNR_TIMER = (1<<15),
1082 IB_QP_SQ_PSN = (1<<16),
1083 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
1084 IB_QP_PATH_MIG_STATE = (1<<18),
1085 IB_QP_CAP = (1<<19),
1086 IB_QP_DEST_QPN = (1<<20),
1087 IB_QP_RESERVED1 = (1<<21),
1088 IB_QP_RESERVED2 = (1<<22),
1089 IB_QP_RESERVED3 = (1<<23),
1090 IB_QP_RESERVED4 = (1<<24),
1115 enum ib_qp_state qp_state;
1116 enum ib_qp_state cur_qp_state;
1117 enum ib_mtu path_mtu;
1118 enum ib_mig_state path_mig_state;
1123 int qp_access_flags;
1124 struct ib_qp_cap cap;
1125 struct ib_ah_attr ah_attr;
1126 struct ib_ah_attr alt_ah_attr;
1129 u8 en_sqd_async_notify;
1132 u8 max_dest_rd_atomic;
1144 IB_WR_RDMA_WRITE_WITH_IMM,
1146 IB_WR_SEND_WITH_IMM,
1148 IB_WR_ATOMIC_CMP_AND_SWP,
1149 IB_WR_ATOMIC_FETCH_AND_ADD,
1151 IB_WR_SEND_WITH_INV,
1152 IB_WR_RDMA_READ_WITH_INV,
1155 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1156 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1158 /* reserve values for low level drivers' internal use.
1159 * These values will not be used at all in the ib core layer.
1161 IB_WR_RESERVED1 = 0xf0,
1173 enum ib_send_flags {
1175 IB_SEND_SIGNALED = (1<<1),
1176 IB_SEND_SOLICITED = (1<<2),
1177 IB_SEND_INLINE = (1<<3),
1178 IB_SEND_IP_CSUM = (1<<4),
1180 /* reserve bits 26-31 for low level drivers' internal use */
1181 IB_SEND_RESERVED_START = (1 << 26),
1182 IB_SEND_RESERVED_END = (1 << 31),
1192 void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1196 struct ib_send_wr *next;
1199 struct ib_cqe *wr_cqe;
1201 struct ib_sge *sg_list;
1203 enum ib_wr_opcode opcode;
1207 u32 invalidate_rkey;
1212 struct ib_send_wr wr;
1217 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1219 return container_of(wr, struct ib_rdma_wr, wr);
1222 struct ib_atomic_wr {
1223 struct ib_send_wr wr;
1227 u64 compare_add_mask;
1232 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1234 return container_of(wr, struct ib_atomic_wr, wr);
1238 struct ib_send_wr wr;
1245 u16 pkey_index; /* valid for GSI only */
1246 u8 port_num; /* valid for DR SMPs on switch only */
1249 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1251 return container_of(wr, struct ib_ud_wr, wr);
1255 struct ib_send_wr wr;
1261 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1263 return container_of(wr, struct ib_reg_wr, wr);
1266 struct ib_sig_handover_wr {
1267 struct ib_send_wr wr;
1268 struct ib_sig_attrs *sig_attrs;
1269 struct ib_mr *sig_mr;
1271 struct ib_sge *prot;
1274 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1276 return container_of(wr, struct ib_sig_handover_wr, wr);
1280 struct ib_recv_wr *next;
1283 struct ib_cqe *wr_cqe;
1285 struct ib_sge *sg_list;
1289 enum ib_access_flags {
1290 IB_ACCESS_LOCAL_WRITE = 1,
1291 IB_ACCESS_REMOTE_WRITE = (1<<1),
1292 IB_ACCESS_REMOTE_READ = (1<<2),
1293 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1294 IB_ACCESS_MW_BIND = (1<<4),
1295 IB_ZERO_BASED = (1<<5),
1296 IB_ACCESS_ON_DEMAND = (1<<6),
1300 * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1301 * are hidden here instead of a uapi header!
1303 enum ib_mr_rereg_flags {
1304 IB_MR_REREG_TRANS = 1,
1305 IB_MR_REREG_PD = (1<<1),
1306 IB_MR_REREG_ACCESS = (1<<2),
1307 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1310 struct ib_fmr_attr {
1318 struct ib_ucontext {
1319 struct ib_device *device;
1320 struct list_head pd_list;
1321 struct list_head mr_list;
1322 struct list_head mw_list;
1323 struct list_head cq_list;
1324 struct list_head qp_list;
1325 struct list_head srq_list;
1326 struct list_head ah_list;
1327 struct list_head xrcd_list;
1328 struct list_head rule_list;
1329 struct list_head wq_list;
1330 struct list_head rwq_ind_tbl_list;
1334 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1335 struct rb_root umem_tree;
1337 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1338 * mmu notifiers registration.
1340 struct rw_semaphore umem_rwsem;
1341 void (*invalidate_range)(struct ib_umem *umem,
1342 unsigned long start, unsigned long end);
1344 struct mmu_notifier mn;
1345 atomic_t notifier_count;
1346 /* A list of umems that don't have private mmu notifier counters yet. */
1347 struct list_head no_private_counters;
1353 u64 user_handle; /* handle given to us by userspace */
1354 struct ib_ucontext *context; /* associated user context */
1355 void *object; /* containing object */
1356 struct list_head list; /* link to context's list */
1357 int id; /* index into kernel idr */
1359 struct rw_semaphore mutex; /* protects .live */
1360 struct rcu_head rcu; /* kfree_rcu() overhead */
1365 const void __user *inbuf;
1366 void __user *outbuf;
1374 struct ib_device *device;
1375 struct ib_uobject *uobject;
1376 atomic_t usecnt; /* count all resources */
1378 u32 unsafe_global_rkey;
1381 * Implementation details of the RDMA core, don't use in drivers:
1383 struct ib_mr *__internal_mr;
1387 struct ib_device *device;
1388 atomic_t usecnt; /* count all exposed resources */
1389 struct inode *inode;
1391 struct mutex tgt_qp_mutex;
1392 struct list_head tgt_qp_list;
1396 struct ib_device *device;
1398 struct ib_uobject *uobject;
1401 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1403 enum ib_poll_context {
1404 IB_POLL_DIRECT, /* caller context, no hw completions */
1405 IB_POLL_SOFTIRQ, /* poll from softirq context */
1406 IB_POLL_WORKQUEUE, /* poll from workqueue */
1410 struct ib_device *device;
1411 struct ib_uobject *uobject;
1412 ib_comp_handler comp_handler;
1413 void (*event_handler)(struct ib_event *, void *);
1416 atomic_t usecnt; /* count number of work queues */
1417 enum ib_poll_context poll_ctx;
1420 struct irq_poll iop;
1421 struct work_struct work;
1426 struct ib_device *device;
1428 struct ib_uobject *uobject;
1429 void (*event_handler)(struct ib_event *, void *);
1431 enum ib_srq_type srq_type;
1436 struct ib_xrcd *xrcd;
1454 struct ib_device *device;
1455 struct ib_uobject *uobject;
1457 void (*event_handler)(struct ib_event *, void *);
1461 enum ib_wq_state state;
1462 enum ib_wq_type wq_type;
1466 struct ib_wq_init_attr {
1468 enum ib_wq_type wq_type;
1472 void (*event_handler)(struct ib_event *, void *);
1475 enum ib_wq_attr_mask {
1476 IB_WQ_STATE = 1 << 0,
1477 IB_WQ_CUR_STATE = 1 << 1,
1481 enum ib_wq_state wq_state;
1482 enum ib_wq_state curr_wq_state;
1485 struct ib_rwq_ind_table {
1486 struct ib_device *device;
1487 struct ib_uobject *uobject;
1490 u32 log_ind_tbl_size;
1491 struct ib_wq **ind_tbl;
1494 struct ib_rwq_ind_table_init_attr {
1495 u32 log_ind_tbl_size;
1496 /* Each entry is a pointer to Receive Work Queue */
1497 struct ib_wq **ind_tbl;
1501 * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1502 * @max_read_sge: Maximum SGE elements per RDMA READ request.
1505 struct ib_device *device;
1507 struct ib_cq *send_cq;
1508 struct ib_cq *recv_cq;
1511 struct list_head rdma_mrs;
1512 struct list_head sig_mrs;
1514 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1515 struct list_head xrcd_list;
1517 /* count times opened, mcast attaches, flow attaches */
1519 struct list_head open_list;
1520 struct ib_qp *real_qp;
1521 struct ib_uobject *uobject;
1522 void (*event_handler)(struct ib_event *, void *);
1527 enum ib_qp_type qp_type;
1528 struct ib_rwq_ind_table *rwq_ind_tbl;
1532 struct ib_device *device;
1538 unsigned int page_size;
1541 struct ib_uobject *uobject; /* user */
1542 struct list_head qp_entry; /* FR */
1547 struct ib_device *device;
1549 struct ib_uobject *uobject;
1551 enum ib_mw_type type;
1555 struct ib_device *device;
1557 struct list_head list;
1562 /* Supported steering options */
1563 enum ib_flow_attr_type {
1564 /* steering according to rule specifications */
1565 IB_FLOW_ATTR_NORMAL = 0x0,
1566 /* default unicast and multicast rule -
1567 * receive all Eth traffic which isn't steered to any QP
1569 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1570 /* default multicast rule -
1571 * receive all Eth multicast traffic which isn't steered to any QP
1573 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1574 /* sniffer rule - receive all port traffic */
1575 IB_FLOW_ATTR_SNIFFER = 0x3
1578 /* Supported steering header types */
1579 enum ib_flow_spec_type {
1581 IB_FLOW_SPEC_ETH = 0x20,
1582 IB_FLOW_SPEC_IB = 0x22,
1584 IB_FLOW_SPEC_IPV4 = 0x30,
1585 IB_FLOW_SPEC_IPV6 = 0x31,
1587 IB_FLOW_SPEC_TCP = 0x40,
1588 IB_FLOW_SPEC_UDP = 0x41
1590 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1591 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1593 /* Flow steering rule priority is set according to it's domain.
1594 * Lower domain value means higher priority.
1596 enum ib_flow_domain {
1597 IB_FLOW_DOMAIN_USER,
1598 IB_FLOW_DOMAIN_ETHTOOL,
1601 IB_FLOW_DOMAIN_NUM /* Must be last */
1604 enum ib_flow_flags {
1605 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1606 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */
1609 struct ib_flow_eth_filter {
1616 struct ib_flow_spec_eth {
1617 enum ib_flow_spec_type type;
1619 struct ib_flow_eth_filter val;
1620 struct ib_flow_eth_filter mask;
1623 struct ib_flow_ib_filter {
1628 struct ib_flow_spec_ib {
1629 enum ib_flow_spec_type type;
1631 struct ib_flow_ib_filter val;
1632 struct ib_flow_ib_filter mask;
1635 struct ib_flow_ipv4_filter {
1640 struct ib_flow_spec_ipv4 {
1641 enum ib_flow_spec_type type;
1643 struct ib_flow_ipv4_filter val;
1644 struct ib_flow_ipv4_filter mask;
1647 struct ib_flow_ipv6_filter {
1652 struct ib_flow_spec_ipv6 {
1653 enum ib_flow_spec_type type;
1655 struct ib_flow_ipv6_filter val;
1656 struct ib_flow_ipv6_filter mask;
1659 struct ib_flow_tcp_udp_filter {
1664 struct ib_flow_spec_tcp_udp {
1665 enum ib_flow_spec_type type;
1667 struct ib_flow_tcp_udp_filter val;
1668 struct ib_flow_tcp_udp_filter mask;
1671 union ib_flow_spec {
1673 enum ib_flow_spec_type type;
1676 struct ib_flow_spec_eth eth;
1677 struct ib_flow_spec_ib ib;
1678 struct ib_flow_spec_ipv4 ipv4;
1679 struct ib_flow_spec_tcp_udp tcp_udp;
1680 struct ib_flow_spec_ipv6 ipv6;
1683 struct ib_flow_attr {
1684 enum ib_flow_attr_type type;
1690 /* Following are the optional layers according to user request
1691 * struct ib_flow_spec_xxx
1692 * struct ib_flow_spec_yyy
1698 struct ib_uobject *uobject;
1704 enum ib_process_mad_flags {
1705 IB_MAD_IGNORE_MKEY = 1,
1706 IB_MAD_IGNORE_BKEY = 2,
1707 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1710 enum ib_mad_result {
1711 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1712 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1713 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1714 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1717 #define IB_DEVICE_NAME_MAX 64
1721 struct ib_event_handler event_handler;
1722 struct ib_pkey_cache **pkey_cache;
1723 struct ib_gid_table **gid_cache;
1727 struct ib_dma_mapping_ops {
1728 int (*mapping_error)(struct ib_device *dev,
1730 u64 (*map_single)(struct ib_device *dev,
1731 void *ptr, size_t size,
1732 enum dma_data_direction direction);
1733 void (*unmap_single)(struct ib_device *dev,
1734 u64 addr, size_t size,
1735 enum dma_data_direction direction);
1736 u64 (*map_page)(struct ib_device *dev,
1737 struct page *page, unsigned long offset,
1739 enum dma_data_direction direction);
1740 void (*unmap_page)(struct ib_device *dev,
1741 u64 addr, size_t size,
1742 enum dma_data_direction direction);
1743 int (*map_sg)(struct ib_device *dev,
1744 struct scatterlist *sg, int nents,
1745 enum dma_data_direction direction);
1746 void (*unmap_sg)(struct ib_device *dev,
1747 struct scatterlist *sg, int nents,
1748 enum dma_data_direction direction);
1749 void (*sync_single_for_cpu)(struct ib_device *dev,
1752 enum dma_data_direction dir);
1753 void (*sync_single_for_device)(struct ib_device *dev,
1756 enum dma_data_direction dir);
1757 void *(*alloc_coherent)(struct ib_device *dev,
1761 void (*free_coherent)(struct ib_device *dev,
1762 size_t size, void *cpu_addr,
1768 struct ib_port_immutable {
1776 struct device *dma_device;
1778 char name[IB_DEVICE_NAME_MAX];
1780 struct list_head event_handler_list;
1781 spinlock_t event_handler_lock;
1783 spinlock_t client_data_lock;
1784 struct list_head core_list;
1785 /* Access to the client_data_list is protected by the client_data_lock
1786 * spinlock and the lists_rwsem read-write semaphore */
1787 struct list_head client_data_list;
1789 struct ib_cache cache;
1791 * port_immutable is indexed by port number
1793 struct ib_port_immutable *port_immutable;
1795 int num_comp_vectors;
1797 struct iw_cm_verbs *iwcm;
1800 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1801 * driver initialized data. The struct is kfree()'ed by the sysfs
1802 * core when the device is removed. A lifespan of -1 in the return
1803 * struct tells the core to set a default lifespan.
1805 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device,
1808 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1809 * @index - The index in the value array we wish to have updated, or
1810 * num_counters if we want all stats updated
1812 * < 0 - Error, no counters updated
1813 * index - Updated the single counter pointed to by index
1814 * num_counters - Updated all counters (will reset the timestamp
1815 * and prevent further calls for lifespan milliseconds)
1816 * Drivers are allowed to update all counters in leiu of just the
1817 * one given in index at their option
1819 int (*get_hw_stats)(struct ib_device *device,
1820 struct rdma_hw_stats *stats,
1821 u8 port, int index);
1822 int (*query_device)(struct ib_device *device,
1823 struct ib_device_attr *device_attr,
1824 struct ib_udata *udata);
1825 int (*query_port)(struct ib_device *device,
1827 struct ib_port_attr *port_attr);
1828 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1830 /* When calling get_netdev, the HW vendor's driver should return the
1831 * net device of device @device at port @port_num or NULL if such
1832 * a net device doesn't exist. The vendor driver should call dev_hold
1833 * on this net device. The HW vendor's device driver must guarantee
1834 * that this function returns NULL before the net device reaches
1835 * NETDEV_UNREGISTER_FINAL state.
1837 struct net_device *(*get_netdev)(struct ib_device *device,
1839 int (*query_gid)(struct ib_device *device,
1840 u8 port_num, int index,
1842 /* When calling add_gid, the HW vendor's driver should
1843 * add the gid of device @device at gid index @index of
1844 * port @port_num to be @gid. Meta-info of that gid (for example,
1845 * the network device related to this gid is available
1846 * at @attr. @context allows the HW vendor driver to store extra
1847 * information together with a GID entry. The HW vendor may allocate
1848 * memory to contain this information and store it in @context when a
1849 * new GID entry is written to. Params are consistent until the next
1850 * call of add_gid or delete_gid. The function should return 0 on
1851 * success or error otherwise. The function could be called
1852 * concurrently for different ports. This function is only called
1853 * when roce_gid_table is used.
1855 int (*add_gid)(struct ib_device *device,
1858 const union ib_gid *gid,
1859 const struct ib_gid_attr *attr,
1861 /* When calling del_gid, the HW vendor's driver should delete the
1862 * gid of device @device at gid index @index of port @port_num.
1863 * Upon the deletion of a GID entry, the HW vendor must free any
1864 * allocated memory. The caller will clear @context afterwards.
1865 * This function is only called when roce_gid_table is used.
1867 int (*del_gid)(struct ib_device *device,
1871 int (*query_pkey)(struct ib_device *device,
1872 u8 port_num, u16 index, u16 *pkey);
1873 int (*modify_device)(struct ib_device *device,
1874 int device_modify_mask,
1875 struct ib_device_modify *device_modify);
1876 int (*modify_port)(struct ib_device *device,
1877 u8 port_num, int port_modify_mask,
1878 struct ib_port_modify *port_modify);
1879 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1880 struct ib_udata *udata);
1881 int (*dealloc_ucontext)(struct ib_ucontext *context);
1882 int (*mmap)(struct ib_ucontext *context,
1883 struct vm_area_struct *vma);
1884 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1885 struct ib_ucontext *context,
1886 struct ib_udata *udata);
1887 int (*dealloc_pd)(struct ib_pd *pd);
1888 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1889 struct ib_ah_attr *ah_attr);
1890 int (*modify_ah)(struct ib_ah *ah,
1891 struct ib_ah_attr *ah_attr);
1892 int (*query_ah)(struct ib_ah *ah,
1893 struct ib_ah_attr *ah_attr);
1894 int (*destroy_ah)(struct ib_ah *ah);
1895 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1896 struct ib_srq_init_attr *srq_init_attr,
1897 struct ib_udata *udata);
1898 int (*modify_srq)(struct ib_srq *srq,
1899 struct ib_srq_attr *srq_attr,
1900 enum ib_srq_attr_mask srq_attr_mask,
1901 struct ib_udata *udata);
1902 int (*query_srq)(struct ib_srq *srq,
1903 struct ib_srq_attr *srq_attr);
1904 int (*destroy_srq)(struct ib_srq *srq);
1905 int (*post_srq_recv)(struct ib_srq *srq,
1906 struct ib_recv_wr *recv_wr,
1907 struct ib_recv_wr **bad_recv_wr);
1908 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1909 struct ib_qp_init_attr *qp_init_attr,
1910 struct ib_udata *udata);
1911 int (*modify_qp)(struct ib_qp *qp,
1912 struct ib_qp_attr *qp_attr,
1914 struct ib_udata *udata);
1915 int (*query_qp)(struct ib_qp *qp,
1916 struct ib_qp_attr *qp_attr,
1918 struct ib_qp_init_attr *qp_init_attr);
1919 int (*destroy_qp)(struct ib_qp *qp);
1920 int (*post_send)(struct ib_qp *qp,
1921 struct ib_send_wr *send_wr,
1922 struct ib_send_wr **bad_send_wr);
1923 int (*post_recv)(struct ib_qp *qp,
1924 struct ib_recv_wr *recv_wr,
1925 struct ib_recv_wr **bad_recv_wr);
1926 struct ib_cq * (*create_cq)(struct ib_device *device,
1927 const struct ib_cq_init_attr *attr,
1928 struct ib_ucontext *context,
1929 struct ib_udata *udata);
1930 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1932 int (*destroy_cq)(struct ib_cq *cq);
1933 int (*resize_cq)(struct ib_cq *cq, int cqe,
1934 struct ib_udata *udata);
1935 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1937 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1938 int (*req_notify_cq)(struct ib_cq *cq,
1939 enum ib_cq_notify_flags flags);
1940 int (*req_ncomp_notif)(struct ib_cq *cq,
1942 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1943 int mr_access_flags);
1944 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1945 u64 start, u64 length,
1947 int mr_access_flags,
1948 struct ib_udata *udata);
1949 int (*rereg_user_mr)(struct ib_mr *mr,
1951 u64 start, u64 length,
1953 int mr_access_flags,
1955 struct ib_udata *udata);
1956 int (*dereg_mr)(struct ib_mr *mr);
1957 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
1958 enum ib_mr_type mr_type,
1960 int (*map_mr_sg)(struct ib_mr *mr,
1961 struct scatterlist *sg,
1963 unsigned int *sg_offset);
1964 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1965 enum ib_mw_type type,
1966 struct ib_udata *udata);
1967 int (*dealloc_mw)(struct ib_mw *mw);
1968 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1969 int mr_access_flags,
1970 struct ib_fmr_attr *fmr_attr);
1971 int (*map_phys_fmr)(struct ib_fmr *fmr,
1972 u64 *page_list, int list_len,
1974 int (*unmap_fmr)(struct list_head *fmr_list);
1975 int (*dealloc_fmr)(struct ib_fmr *fmr);
1976 int (*attach_mcast)(struct ib_qp *qp,
1979 int (*detach_mcast)(struct ib_qp *qp,
1982 int (*process_mad)(struct ib_device *device,
1983 int process_mad_flags,
1985 const struct ib_wc *in_wc,
1986 const struct ib_grh *in_grh,
1987 const struct ib_mad_hdr *in_mad,
1989 struct ib_mad_hdr *out_mad,
1990 size_t *out_mad_size,
1991 u16 *out_mad_pkey_index);
1992 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1993 struct ib_ucontext *ucontext,
1994 struct ib_udata *udata);
1995 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1996 struct ib_flow * (*create_flow)(struct ib_qp *qp,
2000 int (*destroy_flow)(struct ib_flow *flow_id);
2001 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2002 struct ib_mr_status *mr_status);
2003 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2004 void (*drain_rq)(struct ib_qp *qp);
2005 void (*drain_sq)(struct ib_qp *qp);
2006 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2008 int (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2009 struct ifla_vf_info *ivf);
2010 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2011 struct ifla_vf_stats *stats);
2012 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2014 struct ib_wq * (*create_wq)(struct ib_pd *pd,
2015 struct ib_wq_init_attr *init_attr,
2016 struct ib_udata *udata);
2017 int (*destroy_wq)(struct ib_wq *wq);
2018 int (*modify_wq)(struct ib_wq *wq,
2019 struct ib_wq_attr *attr,
2021 struct ib_udata *udata);
2022 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device,
2023 struct ib_rwq_ind_table_init_attr *init_attr,
2024 struct ib_udata *udata);
2025 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2026 struct ib_dma_mapping_ops *dma_ops;
2028 struct module *owner;
2030 struct kobject *ports_parent;
2031 struct list_head port_list;
2034 IB_DEV_UNINITIALIZED,
2040 u64 uverbs_cmd_mask;
2041 u64 uverbs_ex_cmd_mask;
2049 struct ib_device_attr attrs;
2050 struct attribute_group *hw_stats_ag;
2051 struct rdma_hw_stats *hw_stats;
2054 * The following mandatory functions are used only at device
2055 * registration. Keep functions such as these at the end of this
2056 * structure to avoid cache line misses when accessing struct ib_device
2059 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2060 void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2065 void (*add) (struct ib_device *);
2066 void (*remove)(struct ib_device *, void *client_data);
2068 /* Returns the net_dev belonging to this ib_client and matching the
2070 * @dev: An RDMA device that the net_dev use for communication.
2071 * @port: A physical port number on the RDMA device.
2072 * @pkey: P_Key that the net_dev uses if applicable.
2073 * @gid: A GID that the net_dev uses to communicate.
2074 * @addr: An IP address the net_dev is configured with.
2075 * @client_data: The device's client data set by ib_set_client_data().
2077 * An ib_client that implements a net_dev on top of RDMA devices
2078 * (such as IP over IB) should implement this callback, allowing the
2079 * rdma_cm module to find the right net_dev for a given request.
2081 * The caller is responsible for calling dev_put on the returned
2083 struct net_device *(*get_net_dev_by_params)(
2084 struct ib_device *dev,
2087 const union ib_gid *gid,
2088 const struct sockaddr *addr,
2090 struct list_head list;
2093 struct ib_device *ib_alloc_device(size_t size);
2094 void ib_dealloc_device(struct ib_device *device);
2096 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2098 int ib_register_device(struct ib_device *device,
2099 int (*port_callback)(struct ib_device *,
2100 u8, struct kobject *));
2101 void ib_unregister_device(struct ib_device *device);
2103 int ib_register_client (struct ib_client *client);
2104 void ib_unregister_client(struct ib_client *client);
2106 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2107 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
2110 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2112 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2115 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2117 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2120 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2124 const void __user *p = udata->inbuf + offset;
2128 if (len > USHRT_MAX)
2131 buf = kmalloc(len, GFP_KERNEL);
2135 if (copy_from_user(buf, p, len))
2138 ret = !memchr_inv(buf, 0, len);
2146 * ib_modify_qp_is_ok - Check that the supplied attribute mask
2147 * contains all required attributes and no attributes not allowed for
2148 * the given QP state transition.
2149 * @cur_state: Current QP state
2150 * @next_state: Next QP state
2152 * @mask: Mask of supplied QP attributes
2153 * @ll : link layer of port
2155 * This function is a helper function that a low-level driver's
2156 * modify_qp method can use to validate the consumer's input. It
2157 * checks that cur_state and next_state are valid QP states, that a
2158 * transition from cur_state to next_state is allowed by the IB spec,
2159 * and that the attribute mask supplied is allowed for the transition.
2161 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2162 enum ib_qp_type type, enum ib_qp_attr_mask mask,
2163 enum rdma_link_layer ll);
2165 int ib_register_event_handler (struct ib_event_handler *event_handler);
2166 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2167 void ib_dispatch_event(struct ib_event *event);
2169 int ib_query_port(struct ib_device *device,
2170 u8 port_num, struct ib_port_attr *port_attr);
2172 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2176 * rdma_cap_ib_switch - Check if the device is IB switch
2177 * @device: Device to check
2179 * Device driver is responsible for setting is_switch bit on
2180 * in ib_device structure at init time.
2182 * Return: true if the device is IB switch.
2184 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2186 return device->is_switch;
2190 * rdma_start_port - Return the first valid port number for the device
2193 * @device: Device to be checked
2195 * Return start port number
2197 static inline u8 rdma_start_port(const struct ib_device *device)
2199 return rdma_cap_ib_switch(device) ? 0 : 1;
2203 * rdma_end_port - Return the last valid port number for the device
2206 * @device: Device to be checked
2208 * Return last port number
2210 static inline u8 rdma_end_port(const struct ib_device *device)
2212 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2215 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2217 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2220 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2222 return device->port_immutable[port_num].core_cap_flags &
2223 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2226 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2228 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2231 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2233 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2236 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2238 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2241 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2243 return rdma_protocol_ib(device, port_num) ||
2244 rdma_protocol_roce(device, port_num);
2248 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2249 * Management Datagrams.
2250 * @device: Device to check
2251 * @port_num: Port number to check
2253 * Management Datagrams (MAD) are a required part of the InfiniBand
2254 * specification and are supported on all InfiniBand devices. A slightly
2255 * extended version are also supported on OPA interfaces.
2257 * Return: true if the port supports sending/receiving of MAD packets.
2259 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2261 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2265 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2266 * Management Datagrams.
2267 * @device: Device to check
2268 * @port_num: Port number to check
2270 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2271 * datagrams with their own versions. These OPA MADs share many but not all of
2272 * the characteristics of InfiniBand MADs.
2274 * OPA MADs differ in the following ways:
2276 * 1) MADs are variable size up to 2K
2277 * IBTA defined MADs remain fixed at 256 bytes
2278 * 2) OPA SMPs must carry valid PKeys
2279 * 3) OPA SMP packets are a different format
2281 * Return: true if the port supports OPA MAD packet formats.
2283 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2285 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2286 == RDMA_CORE_CAP_OPA_MAD;
2290 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2291 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2292 * @device: Device to check
2293 * @port_num: Port number to check
2295 * Each InfiniBand node is required to provide a Subnet Management Agent
2296 * that the subnet manager can access. Prior to the fabric being fully
2297 * configured by the subnet manager, the SMA is accessed via a well known
2298 * interface called the Subnet Management Interface (SMI). This interface
2299 * uses directed route packets to communicate with the SM to get around the
2300 * chicken and egg problem of the SM needing to know what's on the fabric
2301 * in order to configure the fabric, and needing to configure the fabric in
2302 * order to send packets to the devices on the fabric. These directed
2303 * route packets do not need the fabric fully configured in order to reach
2304 * their destination. The SMI is the only method allowed to send
2305 * directed route packets on an InfiniBand fabric.
2307 * Return: true if the port provides an SMI.
2309 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2311 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2315 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2316 * Communication Manager.
2317 * @device: Device to check
2318 * @port_num: Port number to check
2320 * The InfiniBand Communication Manager is one of many pre-defined General
2321 * Service Agents (GSA) that are accessed via the General Service
2322 * Interface (GSI). It's role is to facilitate establishment of connections
2323 * between nodes as well as other management related tasks for established
2326 * Return: true if the port supports an IB CM (this does not guarantee that
2327 * a CM is actually running however).
2329 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2331 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2335 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2336 * Communication Manager.
2337 * @device: Device to check
2338 * @port_num: Port number to check
2340 * Similar to above, but specific to iWARP connections which have a different
2341 * managment protocol than InfiniBand.
2343 * Return: true if the port supports an iWARP CM (this does not guarantee that
2344 * a CM is actually running however).
2346 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2348 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2352 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2353 * Subnet Administration.
2354 * @device: Device to check
2355 * @port_num: Port number to check
2357 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2358 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2359 * fabrics, devices should resolve routes to other hosts by contacting the
2360 * SA to query the proper route.
2362 * Return: true if the port should act as a client to the fabric Subnet
2363 * Administration interface. This does not imply that the SA service is
2366 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2368 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2372 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2374 * @device: Device to check
2375 * @port_num: Port number to check
2377 * InfiniBand multicast registration is more complex than normal IPv4 or
2378 * IPv6 multicast registration. Each Host Channel Adapter must register
2379 * with the Subnet Manager when it wishes to join a multicast group. It
2380 * should do so only once regardless of how many queue pairs it subscribes
2381 * to this group. And it should leave the group only after all queue pairs
2382 * attached to the group have been detached.
2384 * Return: true if the port must undertake the additional adminstrative
2385 * overhead of registering/unregistering with the SM and tracking of the
2386 * total number of queue pairs attached to the multicast group.
2388 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2390 return rdma_cap_ib_sa(device, port_num);
2394 * rdma_cap_af_ib - Check if the port of device has the capability
2395 * Native Infiniband Address.
2396 * @device: Device to check
2397 * @port_num: Port number to check
2399 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2400 * GID. RoCE uses a different mechanism, but still generates a GID via
2401 * a prescribed mechanism and port specific data.
2403 * Return: true if the port uses a GID address to identify devices on the
2406 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2408 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2412 * rdma_cap_eth_ah - Check if the port of device has the capability
2413 * Ethernet Address Handle.
2414 * @device: Device to check
2415 * @port_num: Port number to check
2417 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2418 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2419 * port. Normally, packet headers are generated by the sending host
2420 * adapter, but when sending connectionless datagrams, we must manually
2421 * inject the proper headers for the fabric we are communicating over.
2423 * Return: true if we are running as a RoCE port and must force the
2424 * addition of a Global Route Header built from our Ethernet Address
2425 * Handle into our header list for connectionless packets.
2427 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2429 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2433 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2436 * @port_num: Port number
2438 * This MAD size includes the MAD headers and MAD payload. No other headers
2441 * Return the max MAD size required by the Port. Will return 0 if the port
2442 * does not support MADs
2444 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2446 return device->port_immutable[port_num].max_mad_size;
2450 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2451 * @device: Device to check
2452 * @port_num: Port number to check
2454 * RoCE GID table mechanism manages the various GIDs for a device.
2456 * NOTE: if allocating the port's GID table has failed, this call will still
2457 * return true, but any RoCE GID table API will fail.
2459 * Return: true if the port uses RoCE GID table mechanism in order to manage
2462 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2465 return rdma_protocol_roce(device, port_num) &&
2466 device->add_gid && device->del_gid;
2470 * Check if the device supports READ W/ INVALIDATE.
2472 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2475 * iWarp drivers must support READ W/ INVALIDATE. No other protocol
2476 * has support for it yet.
2478 return rdma_protocol_iwarp(dev, port_num);
2481 int ib_query_gid(struct ib_device *device,
2482 u8 port_num, int index, union ib_gid *gid,
2483 struct ib_gid_attr *attr);
2485 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2487 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2488 struct ifla_vf_info *info);
2489 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2490 struct ifla_vf_stats *stats);
2491 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2494 int ib_query_pkey(struct ib_device *device,
2495 u8 port_num, u16 index, u16 *pkey);
2497 int ib_modify_device(struct ib_device *device,
2498 int device_modify_mask,
2499 struct ib_device_modify *device_modify);
2501 int ib_modify_port(struct ib_device *device,
2502 u8 port_num, int port_modify_mask,
2503 struct ib_port_modify *port_modify);
2505 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2506 enum ib_gid_type gid_type, struct net_device *ndev,
2507 u8 *port_num, u16 *index);
2509 int ib_find_pkey(struct ib_device *device,
2510 u8 port_num, u16 pkey, u16 *index);
2514 * Create a memory registration for all memory in the system and place
2515 * the rkey for it into pd->unsafe_global_rkey. This can be used by
2516 * ULPs to avoid the overhead of dynamic MRs.
2518 * This flag is generally considered unsafe and must only be used in
2519 * extremly trusted environments. Every use of it will log a warning
2520 * in the kernel log.
2522 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01,
2525 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2526 const char *caller);
2527 #define ib_alloc_pd(device, flags) \
2528 __ib_alloc_pd((device), (flags), __func__)
2529 void ib_dealloc_pd(struct ib_pd *pd);
2532 * ib_create_ah - Creates an address handle for the given address vector.
2533 * @pd: The protection domain associated with the address handle.
2534 * @ah_attr: The attributes of the address vector.
2536 * The address handle is used to reference a local or global destination
2537 * in all UD QP post sends.
2539 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2542 * ib_init_ah_from_wc - Initializes address handle attributes from a
2544 * @device: Device on which the received message arrived.
2545 * @port_num: Port on which the received message arrived.
2546 * @wc: Work completion associated with the received message.
2547 * @grh: References the received global route header. This parameter is
2548 * ignored unless the work completion indicates that the GRH is valid.
2549 * @ah_attr: Returned attributes that can be used when creating an address
2550 * handle for replying to the message.
2552 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2553 const struct ib_wc *wc, const struct ib_grh *grh,
2554 struct ib_ah_attr *ah_attr);
2557 * ib_create_ah_from_wc - Creates an address handle associated with the
2558 * sender of the specified work completion.
2559 * @pd: The protection domain associated with the address handle.
2560 * @wc: Work completion information associated with a received message.
2561 * @grh: References the received global route header. This parameter is
2562 * ignored unless the work completion indicates that the GRH is valid.
2563 * @port_num: The outbound port number to associate with the address.
2565 * The address handle is used to reference a local or global destination
2566 * in all UD QP post sends.
2568 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2569 const struct ib_grh *grh, u8 port_num);
2572 * ib_modify_ah - Modifies the address vector associated with an address
2574 * @ah: The address handle to modify.
2575 * @ah_attr: The new address vector attributes to associate with the
2578 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2581 * ib_query_ah - Queries the address vector associated with an address
2583 * @ah: The address handle to query.
2584 * @ah_attr: The address vector attributes associated with the address
2587 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2590 * ib_destroy_ah - Destroys an address handle.
2591 * @ah: The address handle to destroy.
2593 int ib_destroy_ah(struct ib_ah *ah);
2596 * ib_create_srq - Creates a SRQ associated with the specified protection
2598 * @pd: The protection domain associated with the SRQ.
2599 * @srq_init_attr: A list of initial attributes required to create the
2600 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2601 * the actual capabilities of the created SRQ.
2603 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2604 * requested size of the SRQ, and set to the actual values allocated
2605 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2606 * will always be at least as large as the requested values.
2608 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2609 struct ib_srq_init_attr *srq_init_attr);
2612 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2613 * @srq: The SRQ to modify.
2614 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2615 * the current values of selected SRQ attributes are returned.
2616 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2617 * are being modified.
2619 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2620 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2621 * the number of receives queued drops below the limit.
2623 int ib_modify_srq(struct ib_srq *srq,
2624 struct ib_srq_attr *srq_attr,
2625 enum ib_srq_attr_mask srq_attr_mask);
2628 * ib_query_srq - Returns the attribute list and current values for the
2630 * @srq: The SRQ to query.
2631 * @srq_attr: The attributes of the specified SRQ.
2633 int ib_query_srq(struct ib_srq *srq,
2634 struct ib_srq_attr *srq_attr);
2637 * ib_destroy_srq - Destroys the specified SRQ.
2638 * @srq: The SRQ to destroy.
2640 int ib_destroy_srq(struct ib_srq *srq);
2643 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2644 * @srq: The SRQ to post the work request on.
2645 * @recv_wr: A list of work requests to post on the receive queue.
2646 * @bad_recv_wr: On an immediate failure, this parameter will reference
2647 * the work request that failed to be posted on the QP.
2649 static inline int ib_post_srq_recv(struct ib_srq *srq,
2650 struct ib_recv_wr *recv_wr,
2651 struct ib_recv_wr **bad_recv_wr)
2653 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2657 * ib_create_qp - Creates a QP associated with the specified protection
2659 * @pd: The protection domain associated with the QP.
2660 * @qp_init_attr: A list of initial attributes required to create the
2661 * QP. If QP creation succeeds, then the attributes are updated to
2662 * the actual capabilities of the created QP.
2664 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2665 struct ib_qp_init_attr *qp_init_attr);
2668 * ib_modify_qp - Modifies the attributes for the specified QP and then
2669 * transitions the QP to the given state.
2670 * @qp: The QP to modify.
2671 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2672 * the current values of selected QP attributes are returned.
2673 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2674 * are being modified.
2676 int ib_modify_qp(struct ib_qp *qp,
2677 struct ib_qp_attr *qp_attr,
2681 * ib_query_qp - Returns the attribute list and current values for the
2683 * @qp: The QP to query.
2684 * @qp_attr: The attributes of the specified QP.
2685 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2686 * @qp_init_attr: Additional attributes of the selected QP.
2688 * The qp_attr_mask may be used to limit the query to gathering only the
2689 * selected attributes.
2691 int ib_query_qp(struct ib_qp *qp,
2692 struct ib_qp_attr *qp_attr,
2694 struct ib_qp_init_attr *qp_init_attr);
2697 * ib_destroy_qp - Destroys the specified QP.
2698 * @qp: The QP to destroy.
2700 int ib_destroy_qp(struct ib_qp *qp);
2703 * ib_open_qp - Obtain a reference to an existing sharable QP.
2704 * @xrcd - XRC domain
2705 * @qp_open_attr: Attributes identifying the QP to open.
2707 * Returns a reference to a sharable QP.
2709 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2710 struct ib_qp_open_attr *qp_open_attr);
2713 * ib_close_qp - Release an external reference to a QP.
2714 * @qp: The QP handle to release
2716 * The opened QP handle is released by the caller. The underlying
2717 * shared QP is not destroyed until all internal references are released.
2719 int ib_close_qp(struct ib_qp *qp);
2722 * ib_post_send - Posts a list of work requests to the send queue of
2724 * @qp: The QP to post the work request on.
2725 * @send_wr: A list of work requests to post on the send queue.
2726 * @bad_send_wr: On an immediate failure, this parameter will reference
2727 * the work request that failed to be posted on the QP.
2729 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2730 * error is returned, the QP state shall not be affected,
2731 * ib_post_send() will return an immediate error after queueing any
2732 * earlier work requests in the list.
2734 static inline int ib_post_send(struct ib_qp *qp,
2735 struct ib_send_wr *send_wr,
2736 struct ib_send_wr **bad_send_wr)
2738 return qp->device->post_send(qp, send_wr, bad_send_wr);
2742 * ib_post_recv - Posts a list of work requests to the receive queue of
2744 * @qp: The QP to post the work request on.
2745 * @recv_wr: A list of work requests to post on the receive queue.
2746 * @bad_recv_wr: On an immediate failure, this parameter will reference
2747 * the work request that failed to be posted on the QP.
2749 static inline int ib_post_recv(struct ib_qp *qp,
2750 struct ib_recv_wr *recv_wr,
2751 struct ib_recv_wr **bad_recv_wr)
2753 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2756 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2757 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2758 void ib_free_cq(struct ib_cq *cq);
2759 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2762 * ib_create_cq - Creates a CQ on the specified device.
2763 * @device: The device on which to create the CQ.
2764 * @comp_handler: A user-specified callback that is invoked when a
2765 * completion event occurs on the CQ.
2766 * @event_handler: A user-specified callback that is invoked when an
2767 * asynchronous event not associated with a completion occurs on the CQ.
2768 * @cq_context: Context associated with the CQ returned to the user via
2769 * the associated completion and event handlers.
2770 * @cq_attr: The attributes the CQ should be created upon.
2772 * Users can examine the cq structure to determine the actual CQ size.
2774 struct ib_cq *ib_create_cq(struct ib_device *device,
2775 ib_comp_handler comp_handler,
2776 void (*event_handler)(struct ib_event *, void *),
2778 const struct ib_cq_init_attr *cq_attr);
2781 * ib_resize_cq - Modifies the capacity of the CQ.
2782 * @cq: The CQ to resize.
2783 * @cqe: The minimum size of the CQ.
2785 * Users can examine the cq structure to determine the actual CQ size.
2787 int ib_resize_cq(struct ib_cq *cq, int cqe);
2790 * ib_modify_cq - Modifies moderation params of the CQ
2791 * @cq: The CQ to modify.
2792 * @cq_count: number of CQEs that will trigger an event
2793 * @cq_period: max period of time in usec before triggering an event
2796 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2799 * ib_destroy_cq - Destroys the specified CQ.
2800 * @cq: The CQ to destroy.
2802 int ib_destroy_cq(struct ib_cq *cq);
2805 * ib_poll_cq - poll a CQ for completion(s)
2806 * @cq:the CQ being polled
2807 * @num_entries:maximum number of completions to return
2808 * @wc:array of at least @num_entries &struct ib_wc where completions
2811 * Poll a CQ for (possibly multiple) completions. If the return value
2812 * is < 0, an error occurred. If the return value is >= 0, it is the
2813 * number of completions returned. If the return value is
2814 * non-negative and < num_entries, then the CQ was emptied.
2816 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2819 return cq->device->poll_cq(cq, num_entries, wc);
2823 * ib_peek_cq - Returns the number of unreaped completions currently
2824 * on the specified CQ.
2825 * @cq: The CQ to peek.
2826 * @wc_cnt: A minimum number of unreaped completions to check for.
2828 * If the number of unreaped completions is greater than or equal to wc_cnt,
2829 * this function returns wc_cnt, otherwise, it returns the actual number of
2830 * unreaped completions.
2832 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2835 * ib_req_notify_cq - Request completion notification on a CQ.
2836 * @cq: The CQ to generate an event for.
2838 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2839 * to request an event on the next solicited event or next work
2840 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2841 * may also be |ed in to request a hint about missed events, as
2845 * < 0 means an error occurred while requesting notification
2846 * == 0 means notification was requested successfully, and if
2847 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2848 * were missed and it is safe to wait for another event. In
2849 * this case is it guaranteed that any work completions added
2850 * to the CQ since the last CQ poll will trigger a completion
2851 * notification event.
2852 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2853 * in. It means that the consumer must poll the CQ again to
2854 * make sure it is empty to avoid missing an event because of a
2855 * race between requesting notification and an entry being
2856 * added to the CQ. This return value means it is possible
2857 * (but not guaranteed) that a work completion has been added
2858 * to the CQ since the last poll without triggering a
2859 * completion notification event.
2861 static inline int ib_req_notify_cq(struct ib_cq *cq,
2862 enum ib_cq_notify_flags flags)
2864 return cq->device->req_notify_cq(cq, flags);
2868 * ib_req_ncomp_notif - Request completion notification when there are
2869 * at least the specified number of unreaped completions on the CQ.
2870 * @cq: The CQ to generate an event for.
2871 * @wc_cnt: The number of unreaped completions that should be on the
2872 * CQ before an event is generated.
2874 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2876 return cq->device->req_ncomp_notif ?
2877 cq->device->req_ncomp_notif(cq, wc_cnt) :
2882 * ib_dma_mapping_error - check a DMA addr for error
2883 * @dev: The device for which the dma_addr was created
2884 * @dma_addr: The DMA address to check
2886 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2889 return dev->dma_ops->mapping_error(dev, dma_addr);
2890 return dma_mapping_error(dev->dma_device, dma_addr);
2894 * ib_dma_map_single - Map a kernel virtual address to DMA address
2895 * @dev: The device for which the dma_addr is to be created
2896 * @cpu_addr: The kernel virtual address
2897 * @size: The size of the region in bytes
2898 * @direction: The direction of the DMA
2900 static inline u64 ib_dma_map_single(struct ib_device *dev,
2901 void *cpu_addr, size_t size,
2902 enum dma_data_direction direction)
2905 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2906 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2910 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2911 * @dev: The device for which the DMA address was created
2912 * @addr: The DMA address
2913 * @size: The size of the region in bytes
2914 * @direction: The direction of the DMA
2916 static inline void ib_dma_unmap_single(struct ib_device *dev,
2917 u64 addr, size_t size,
2918 enum dma_data_direction direction)
2921 dev->dma_ops->unmap_single(dev, addr, size, direction);
2923 dma_unmap_single(dev->dma_device, addr, size, direction);
2926 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2927 void *cpu_addr, size_t size,
2928 enum dma_data_direction direction,
2929 unsigned long dma_attrs)
2931 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2932 direction, dma_attrs);
2935 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2936 u64 addr, size_t size,
2937 enum dma_data_direction direction,
2938 unsigned long dma_attrs)
2940 return dma_unmap_single_attrs(dev->dma_device, addr, size,
2941 direction, dma_attrs);
2945 * ib_dma_map_page - Map a physical page to DMA address
2946 * @dev: The device for which the dma_addr is to be created
2947 * @page: The page to be mapped
2948 * @offset: The offset within the page
2949 * @size: The size of the region in bytes
2950 * @direction: The direction of the DMA
2952 static inline u64 ib_dma_map_page(struct ib_device *dev,
2954 unsigned long offset,
2956 enum dma_data_direction direction)
2959 return dev->dma_ops->map_page(dev, page, offset, size, direction);
2960 return dma_map_page(dev->dma_device, page, offset, size, direction);
2964 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2965 * @dev: The device for which the DMA address was created
2966 * @addr: The DMA address
2967 * @size: The size of the region in bytes
2968 * @direction: The direction of the DMA
2970 static inline void ib_dma_unmap_page(struct ib_device *dev,
2971 u64 addr, size_t size,
2972 enum dma_data_direction direction)
2975 dev->dma_ops->unmap_page(dev, addr, size, direction);
2977 dma_unmap_page(dev->dma_device, addr, size, direction);
2981 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2982 * @dev: The device for which the DMA addresses are to be created
2983 * @sg: The array of scatter/gather entries
2984 * @nents: The number of scatter/gather entries
2985 * @direction: The direction of the DMA
2987 static inline int ib_dma_map_sg(struct ib_device *dev,
2988 struct scatterlist *sg, int nents,
2989 enum dma_data_direction direction)
2992 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2993 return dma_map_sg(dev->dma_device, sg, nents, direction);
2997 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2998 * @dev: The device for which the DMA addresses were created
2999 * @sg: The array of scatter/gather entries
3000 * @nents: The number of scatter/gather entries
3001 * @direction: The direction of the DMA
3003 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3004 struct scatterlist *sg, int nents,
3005 enum dma_data_direction direction)
3008 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
3010 dma_unmap_sg(dev->dma_device, sg, nents, direction);
3013 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3014 struct scatterlist *sg, int nents,
3015 enum dma_data_direction direction,
3016 unsigned long dma_attrs)
3018 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3022 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3023 struct scatterlist *sg, int nents,
3024 enum dma_data_direction direction,
3025 unsigned long dma_attrs)
3027 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3030 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3031 * @dev: The device for which the DMA addresses were created
3032 * @sg: The scatter/gather entry
3034 * Note: this function is obsolete. To do: change all occurrences of
3035 * ib_sg_dma_address() into sg_dma_address().
3037 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3038 struct scatterlist *sg)
3040 return sg_dma_address(sg);
3044 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3045 * @dev: The device for which the DMA addresses were created
3046 * @sg: The scatter/gather entry
3048 * Note: this function is obsolete. To do: change all occurrences of
3049 * ib_sg_dma_len() into sg_dma_len().
3051 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3052 struct scatterlist *sg)
3054 return sg_dma_len(sg);
3058 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3059 * @dev: The device for which the DMA address was created
3060 * @addr: The DMA address
3061 * @size: The size of the region in bytes
3062 * @dir: The direction of the DMA
3064 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3067 enum dma_data_direction dir)
3070 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3072 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3076 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3077 * @dev: The device for which the DMA address was created
3078 * @addr: The DMA address
3079 * @size: The size of the region in bytes
3080 * @dir: The direction of the DMA
3082 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3085 enum dma_data_direction dir)
3088 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3090 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3094 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3095 * @dev: The device for which the DMA address is requested
3096 * @size: The size of the region to allocate in bytes
3097 * @dma_handle: A pointer for returning the DMA address of the region
3098 * @flag: memory allocator flags
3100 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3106 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
3111 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3112 *dma_handle = handle;
3118 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3119 * @dev: The device for which the DMA addresses were allocated
3120 * @size: The size of the region
3121 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3122 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3124 static inline void ib_dma_free_coherent(struct ib_device *dev,
3125 size_t size, void *cpu_addr,
3129 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3131 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3135 * ib_dereg_mr - Deregisters a memory region and removes it from the
3136 * HCA translation table.
3137 * @mr: The memory region to deregister.
3139 * This function can fail, if the memory region has memory windows bound to it.
3141 int ib_dereg_mr(struct ib_mr *mr);
3143 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3144 enum ib_mr_type mr_type,
3148 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3150 * @mr - struct ib_mr pointer to be updated.
3151 * @newkey - new key to be used.
3153 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3155 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3156 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3160 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3161 * for calculating a new rkey for type 2 memory windows.
3162 * @rkey - the rkey to increment.
3164 static inline u32 ib_inc_rkey(u32 rkey)
3166 const u32 mask = 0x000000ff;
3167 return ((rkey + 1) & mask) | (rkey & ~mask);
3171 * ib_alloc_fmr - Allocates a unmapped fast memory region.
3172 * @pd: The protection domain associated with the unmapped region.
3173 * @mr_access_flags: Specifies the memory access rights.
3174 * @fmr_attr: Attributes of the unmapped region.
3176 * A fast memory region must be mapped before it can be used as part of
3179 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3180 int mr_access_flags,
3181 struct ib_fmr_attr *fmr_attr);
3184 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3185 * @fmr: The fast memory region to associate with the pages.
3186 * @page_list: An array of physical pages to map to the fast memory region.
3187 * @list_len: The number of pages in page_list.
3188 * @iova: The I/O virtual address to use with the mapped region.
3190 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3191 u64 *page_list, int list_len,
3194 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3198 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3199 * @fmr_list: A linked list of fast memory regions to unmap.
3201 int ib_unmap_fmr(struct list_head *fmr_list);
3204 * ib_dealloc_fmr - Deallocates a fast memory region.
3205 * @fmr: The fast memory region to deallocate.
3207 int ib_dealloc_fmr(struct ib_fmr *fmr);
3210 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3211 * @qp: QP to attach to the multicast group. The QP must be type
3213 * @gid: Multicast group GID.
3214 * @lid: Multicast group LID in host byte order.
3216 * In order to send and receive multicast packets, subnet
3217 * administration must have created the multicast group and configured
3218 * the fabric appropriately. The port associated with the specified
3219 * QP must also be a member of the multicast group.
3221 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3224 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3225 * @qp: QP to detach from the multicast group.
3226 * @gid: Multicast group GID.
3227 * @lid: Multicast group LID in host byte order.
3229 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3232 * ib_alloc_xrcd - Allocates an XRC domain.
3233 * @device: The device on which to allocate the XRC domain.
3235 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3238 * ib_dealloc_xrcd - Deallocates an XRC domain.
3239 * @xrcd: The XRC domain to deallocate.
3241 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3243 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3244 struct ib_flow_attr *flow_attr, int domain);
3245 int ib_destroy_flow(struct ib_flow *flow_id);
3247 static inline int ib_check_mr_access(int flags)
3250 * Local write permission is required if remote write or
3251 * remote atomic permission is also requested.
3253 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3254 !(flags & IB_ACCESS_LOCAL_WRITE))
3261 * ib_check_mr_status: lightweight check of MR status.
3262 * This routine may provide status checks on a selected
3263 * ib_mr. first use is for signature status check.
3265 * @mr: A memory region.
3266 * @check_mask: Bitmask of which checks to perform from
3267 * ib_mr_status_check enumeration.
3268 * @mr_status: The container of relevant status checks.
3269 * failed checks will be indicated in the status bitmask
3270 * and the relevant info shall be in the error item.
3272 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3273 struct ib_mr_status *mr_status);
3275 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3276 u16 pkey, const union ib_gid *gid,
3277 const struct sockaddr *addr);
3278 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3279 struct ib_wq_init_attr *init_attr);
3280 int ib_destroy_wq(struct ib_wq *wq);
3281 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3283 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3284 struct ib_rwq_ind_table_init_attr*
3285 wq_ind_table_init_attr);
3286 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3288 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3289 unsigned int *sg_offset, unsigned int page_size);
3292 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3293 unsigned int *sg_offset, unsigned int page_size)
3297 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3303 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3304 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3306 void ib_drain_rq(struct ib_qp *qp);
3307 void ib_drain_sq(struct ib_qp *qp);
3308 void ib_drain_qp(struct ib_qp *qp);
3309 #endif /* IB_VERBS_H */