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 <uapi/linux/if_ether.h>
54 #include <linux/atomic.h>
55 #include <linux/mmu_notifier.h>
56 #include <asm/uaccess.h>
58 extern struct workqueue_struct *ib_wq;
68 extern union ib_gid zgid;
71 struct net_device *ndev;
75 /* IB values map to NodeInfo:NodeType. */
84 enum rdma_transport_type {
88 RDMA_TRANSPORT_USNIC_UDP
91 enum rdma_protocol_type {
95 RDMA_PROTOCOL_USNIC_UDP
98 __attribute_const__ enum rdma_transport_type
99 rdma_node_get_transport(enum rdma_node_type node_type);
101 enum rdma_link_layer {
102 IB_LINK_LAYER_UNSPECIFIED,
103 IB_LINK_LAYER_INFINIBAND,
104 IB_LINK_LAYER_ETHERNET,
107 enum ib_device_cap_flags {
108 IB_DEVICE_RESIZE_MAX_WR = 1,
109 IB_DEVICE_BAD_PKEY_CNTR = (1<<1),
110 IB_DEVICE_BAD_QKEY_CNTR = (1<<2),
111 IB_DEVICE_RAW_MULTI = (1<<3),
112 IB_DEVICE_AUTO_PATH_MIG = (1<<4),
113 IB_DEVICE_CHANGE_PHY_PORT = (1<<5),
114 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6),
115 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7),
116 IB_DEVICE_SHUTDOWN_PORT = (1<<8),
117 IB_DEVICE_INIT_TYPE = (1<<9),
118 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10),
119 IB_DEVICE_SYS_IMAGE_GUID = (1<<11),
120 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12),
121 IB_DEVICE_SRQ_RESIZE = (1<<13),
122 IB_DEVICE_N_NOTIFY_CQ = (1<<14),
123 IB_DEVICE_LOCAL_DMA_LKEY = (1<<15),
124 IB_DEVICE_RESERVED = (1<<16), /* old SEND_W_INV */
125 IB_DEVICE_MEM_WINDOW = (1<<17),
127 * Devices should set IB_DEVICE_UD_IP_SUM if they support
128 * insertion of UDP and TCP checksum on outgoing UD IPoIB
129 * messages and can verify the validity of checksum for
130 * incoming messages. Setting this flag implies that the
131 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
133 IB_DEVICE_UD_IP_CSUM = (1<<18),
134 IB_DEVICE_UD_TSO = (1<<19),
135 IB_DEVICE_XRC = (1<<20),
136 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21),
137 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
138 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1<<23),
139 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1<<24),
140 IB_DEVICE_RC_IP_CSUM = (1<<25),
141 IB_DEVICE_RAW_IP_CSUM = (1<<26),
142 IB_DEVICE_MANAGED_FLOW_STEERING = (1<<29),
143 IB_DEVICE_SIGNATURE_HANDOVER = (1<<30),
144 IB_DEVICE_ON_DEMAND_PAGING = (1<<31),
147 enum ib_signature_prot_cap {
148 IB_PROT_T10DIF_TYPE_1 = 1,
149 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
150 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
153 enum ib_signature_guard_cap {
154 IB_GUARD_T10DIF_CRC = 1,
155 IB_GUARD_T10DIF_CSUM = 1 << 1,
164 enum ib_odp_general_cap_bits {
165 IB_ODP_SUPPORT = 1 << 0,
168 enum ib_odp_transport_cap_bits {
169 IB_ODP_SUPPORT_SEND = 1 << 0,
170 IB_ODP_SUPPORT_RECV = 1 << 1,
171 IB_ODP_SUPPORT_WRITE = 1 << 2,
172 IB_ODP_SUPPORT_READ = 1 << 3,
173 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
177 uint64_t general_caps;
179 uint32_t rc_odp_caps;
180 uint32_t uc_odp_caps;
181 uint32_t ud_odp_caps;
182 } per_transport_caps;
185 enum ib_cq_creation_flags {
186 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
189 struct ib_cq_init_attr {
195 struct ib_device_attr {
197 __be64 sys_image_guid;
205 int device_cap_flags;
215 int max_qp_init_rd_atom;
216 int max_ee_init_rd_atom;
217 enum ib_atomic_cap atomic_cap;
218 enum ib_atomic_cap masked_atomic_cap;
225 int max_mcast_qp_attach;
226 int max_total_mcast_qp_attach;
233 unsigned int max_fast_reg_page_list_len;
235 u8 local_ca_ack_delay;
238 struct ib_odp_caps odp_caps;
239 uint64_t timestamp_mask;
240 uint64_t hca_core_clock; /* in KHZ */
251 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
254 case IB_MTU_256: return 256;
255 case IB_MTU_512: return 512;
256 case IB_MTU_1024: return 1024;
257 case IB_MTU_2048: return 2048;
258 case IB_MTU_4096: return 4096;
269 IB_PORT_ACTIVE_DEFER = 5
272 enum ib_port_cap_flags {
274 IB_PORT_NOTICE_SUP = 1 << 2,
275 IB_PORT_TRAP_SUP = 1 << 3,
276 IB_PORT_OPT_IPD_SUP = 1 << 4,
277 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
278 IB_PORT_SL_MAP_SUP = 1 << 6,
279 IB_PORT_MKEY_NVRAM = 1 << 7,
280 IB_PORT_PKEY_NVRAM = 1 << 8,
281 IB_PORT_LED_INFO_SUP = 1 << 9,
282 IB_PORT_SM_DISABLED = 1 << 10,
283 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
284 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
285 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
286 IB_PORT_CM_SUP = 1 << 16,
287 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
288 IB_PORT_REINIT_SUP = 1 << 18,
289 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
290 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
291 IB_PORT_DR_NOTICE_SUP = 1 << 21,
292 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
293 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
294 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
295 IB_PORT_CLIENT_REG_SUP = 1 << 25,
296 IB_PORT_IP_BASED_GIDS = 1 << 26,
306 static inline int ib_width_enum_to_int(enum ib_port_width width)
309 case IB_WIDTH_1X: return 1;
310 case IB_WIDTH_4X: return 4;
311 case IB_WIDTH_8X: return 8;
312 case IB_WIDTH_12X: return 12;
326 struct ib_protocol_stats {
330 struct iw_protocol_stats {
333 u64 ipInTooBigErrors;
336 u64 ipInUnknownProtos;
337 u64 ipInTruncatedPkts;
340 u64 ipOutForwDatagrams;
372 union rdma_protocol_stats {
373 struct ib_protocol_stats ib;
374 struct iw_protocol_stats iw;
377 /* Define bits for the various functionality this port needs to be supported by
380 /* Management 0x00000FFF */
381 #define RDMA_CORE_CAP_IB_MAD 0x00000001
382 #define RDMA_CORE_CAP_IB_SMI 0x00000002
383 #define RDMA_CORE_CAP_IB_CM 0x00000004
384 #define RDMA_CORE_CAP_IW_CM 0x00000008
385 #define RDMA_CORE_CAP_IB_SA 0x00000010
386 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
388 /* Address format 0x000FF000 */
389 #define RDMA_CORE_CAP_AF_IB 0x00001000
390 #define RDMA_CORE_CAP_ETH_AH 0x00002000
392 /* Protocol 0xFFF00000 */
393 #define RDMA_CORE_CAP_PROT_IB 0x00100000
394 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
395 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
397 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
398 | RDMA_CORE_CAP_IB_MAD \
399 | RDMA_CORE_CAP_IB_SMI \
400 | RDMA_CORE_CAP_IB_CM \
401 | RDMA_CORE_CAP_IB_SA \
402 | RDMA_CORE_CAP_AF_IB)
403 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
404 | RDMA_CORE_CAP_IB_MAD \
405 | RDMA_CORE_CAP_IB_CM \
406 | RDMA_CORE_CAP_AF_IB \
407 | RDMA_CORE_CAP_ETH_AH)
408 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
409 | RDMA_CORE_CAP_IW_CM)
410 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
411 | RDMA_CORE_CAP_OPA_MAD)
413 struct ib_port_attr {
414 enum ib_port_state state;
416 enum ib_mtu active_mtu;
435 enum ib_device_modify_flags {
436 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
437 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
440 struct ib_device_modify {
445 enum ib_port_modify_flags {
446 IB_PORT_SHUTDOWN = 1,
447 IB_PORT_INIT_TYPE = (1<<2),
448 IB_PORT_RESET_QKEY_CNTR = (1<<3)
451 struct ib_port_modify {
452 u32 set_port_cap_mask;
453 u32 clr_port_cap_mask;
461 IB_EVENT_QP_ACCESS_ERR,
465 IB_EVENT_PATH_MIG_ERR,
466 IB_EVENT_DEVICE_FATAL,
467 IB_EVENT_PORT_ACTIVE,
470 IB_EVENT_PKEY_CHANGE,
473 IB_EVENT_SRQ_LIMIT_REACHED,
474 IB_EVENT_QP_LAST_WQE_REACHED,
475 IB_EVENT_CLIENT_REREGISTER,
479 __attribute_const__ const char *ib_event_msg(enum ib_event_type event);
482 struct ib_device *device;
489 enum ib_event_type event;
492 struct ib_event_handler {
493 struct ib_device *device;
494 void (*handler)(struct ib_event_handler *, struct ib_event *);
495 struct list_head list;
498 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
500 (_ptr)->device = _device; \
501 (_ptr)->handler = _handler; \
502 INIT_LIST_HEAD(&(_ptr)->list); \
505 struct ib_global_route {
514 __be32 version_tclass_flow;
523 IB_MULTICAST_QPN = 0xffffff
526 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
533 IB_RATE_PORT_CURRENT = 0,
534 IB_RATE_2_5_GBPS = 2,
542 IB_RATE_120_GBPS = 10,
543 IB_RATE_14_GBPS = 11,
544 IB_RATE_56_GBPS = 12,
545 IB_RATE_112_GBPS = 13,
546 IB_RATE_168_GBPS = 14,
547 IB_RATE_25_GBPS = 15,
548 IB_RATE_100_GBPS = 16,
549 IB_RATE_200_GBPS = 17,
550 IB_RATE_300_GBPS = 18
554 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
555 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
556 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
557 * @rate: rate to convert.
559 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
562 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
563 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
564 * @rate: rate to convert.
566 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
570 * enum ib_mr_type - memory region type
571 * @IB_MR_TYPE_MEM_REG: memory region that is used for
572 * normal registration
573 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
574 * signature operations (data-integrity
579 IB_MR_TYPE_SIGNATURE,
584 * IB_SIG_TYPE_NONE: Unprotected.
585 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
587 enum ib_signature_type {
593 * Signature T10-DIF block-guard types
594 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
595 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
597 enum ib_t10_dif_bg_type {
603 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
605 * @bg_type: T10-DIF block guard type (CRC|CSUM)
606 * @pi_interval: protection information interval.
607 * @bg: seed of guard computation.
608 * @app_tag: application tag of guard block
609 * @ref_tag: initial guard block reference tag.
610 * @ref_remap: Indicate wethear the reftag increments each block
611 * @app_escape: Indicate to skip block check if apptag=0xffff
612 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
613 * @apptag_check_mask: check bitmask of application tag.
615 struct ib_t10_dif_domain {
616 enum ib_t10_dif_bg_type bg_type;
624 u16 apptag_check_mask;
628 * struct ib_sig_domain - Parameters for signature domain
629 * @sig_type: specific signauture type
630 * @sig: union of all signature domain attributes that may
631 * be used to set domain layout.
633 struct ib_sig_domain {
634 enum ib_signature_type sig_type;
636 struct ib_t10_dif_domain dif;
641 * struct ib_sig_attrs - Parameters for signature handover operation
642 * @check_mask: bitmask for signature byte check (8 bytes)
643 * @mem: memory domain layout desciptor.
644 * @wire: wire domain layout desciptor.
646 struct ib_sig_attrs {
648 struct ib_sig_domain mem;
649 struct ib_sig_domain wire;
652 enum ib_sig_err_type {
659 * struct ib_sig_err - signature error descriptor
662 enum ib_sig_err_type err_type;
669 enum ib_mr_status_check {
670 IB_MR_CHECK_SIG_STATUS = 1,
674 * struct ib_mr_status - Memory region status container
676 * @fail_status: Bitmask of MR checks status. For each
677 * failed check a corresponding status bit is set.
678 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
681 struct ib_mr_status {
683 struct ib_sig_err sig_err;
687 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
689 * @mult: multiple to convert.
691 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
694 struct ib_global_route grh;
708 IB_WC_LOC_EEC_OP_ERR,
713 IB_WC_LOC_ACCESS_ERR,
714 IB_WC_REM_INV_REQ_ERR,
715 IB_WC_REM_ACCESS_ERR,
718 IB_WC_RNR_RETRY_EXC_ERR,
719 IB_WC_LOC_RDD_VIOL_ERR,
720 IB_WC_REM_INV_RD_REQ_ERR,
723 IB_WC_INV_EEC_STATE_ERR,
725 IB_WC_RESP_TIMEOUT_ERR,
729 __attribute_const__ const char *ib_wc_status_msg(enum ib_wc_status status);
741 IB_WC_MASKED_COMP_SWAP,
742 IB_WC_MASKED_FETCH_ADD,
744 * Set value of IB_WC_RECV so consumers can test if a completion is a
745 * receive by testing (opcode & IB_WC_RECV).
748 IB_WC_RECV_RDMA_WITH_IMM
753 IB_WC_WITH_IMM = (1<<1),
754 IB_WC_WITH_INVALIDATE = (1<<2),
755 IB_WC_IP_CSUM_OK = (1<<3),
756 IB_WC_WITH_SMAC = (1<<4),
757 IB_WC_WITH_VLAN = (1<<5),
762 enum ib_wc_status status;
763 enum ib_wc_opcode opcode;
777 u8 port_num; /* valid only for DR SMPs on switches */
782 enum ib_cq_notify_flags {
783 IB_CQ_SOLICITED = 1 << 0,
784 IB_CQ_NEXT_COMP = 1 << 1,
785 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
786 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
794 enum ib_srq_attr_mask {
795 IB_SRQ_MAX_WR = 1 << 0,
796 IB_SRQ_LIMIT = 1 << 1,
805 struct ib_srq_init_attr {
806 void (*event_handler)(struct ib_event *, void *);
808 struct ib_srq_attr attr;
809 enum ib_srq_type srq_type;
813 struct ib_xrcd *xrcd;
834 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
835 * here (and in that order) since the MAD layer uses them as
836 * indices into a 2-entry table.
845 IB_QPT_RAW_ETHERTYPE,
846 IB_QPT_RAW_PACKET = 8,
850 /* Reserve a range for qp types internal to the low level driver.
851 * These qp types will not be visible at the IB core layer, so the
852 * IB_QPT_MAX usages should not be affected in the core layer
854 IB_QPT_RESERVED1 = 0x1000,
866 enum ib_qp_create_flags {
867 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
868 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
869 IB_QP_CREATE_NETIF_QP = 1 << 5,
870 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
871 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
872 /* reserve bits 26-31 for low level drivers' internal use */
873 IB_QP_CREATE_RESERVED_START = 1 << 26,
874 IB_QP_CREATE_RESERVED_END = 1 << 31,
878 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
879 * callback to destroy the passed in QP.
882 struct ib_qp_init_attr {
883 void (*event_handler)(struct ib_event *, void *);
885 struct ib_cq *send_cq;
886 struct ib_cq *recv_cq;
888 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
889 struct ib_qp_cap cap;
890 enum ib_sig_type sq_sig_type;
891 enum ib_qp_type qp_type;
892 enum ib_qp_create_flags create_flags;
893 u8 port_num; /* special QP types only */
896 struct ib_qp_open_attr {
897 void (*event_handler)(struct ib_event *, void *);
900 enum ib_qp_type qp_type;
903 enum ib_rnr_timeout {
904 IB_RNR_TIMER_655_36 = 0,
905 IB_RNR_TIMER_000_01 = 1,
906 IB_RNR_TIMER_000_02 = 2,
907 IB_RNR_TIMER_000_03 = 3,
908 IB_RNR_TIMER_000_04 = 4,
909 IB_RNR_TIMER_000_06 = 5,
910 IB_RNR_TIMER_000_08 = 6,
911 IB_RNR_TIMER_000_12 = 7,
912 IB_RNR_TIMER_000_16 = 8,
913 IB_RNR_TIMER_000_24 = 9,
914 IB_RNR_TIMER_000_32 = 10,
915 IB_RNR_TIMER_000_48 = 11,
916 IB_RNR_TIMER_000_64 = 12,
917 IB_RNR_TIMER_000_96 = 13,
918 IB_RNR_TIMER_001_28 = 14,
919 IB_RNR_TIMER_001_92 = 15,
920 IB_RNR_TIMER_002_56 = 16,
921 IB_RNR_TIMER_003_84 = 17,
922 IB_RNR_TIMER_005_12 = 18,
923 IB_RNR_TIMER_007_68 = 19,
924 IB_RNR_TIMER_010_24 = 20,
925 IB_RNR_TIMER_015_36 = 21,
926 IB_RNR_TIMER_020_48 = 22,
927 IB_RNR_TIMER_030_72 = 23,
928 IB_RNR_TIMER_040_96 = 24,
929 IB_RNR_TIMER_061_44 = 25,
930 IB_RNR_TIMER_081_92 = 26,
931 IB_RNR_TIMER_122_88 = 27,
932 IB_RNR_TIMER_163_84 = 28,
933 IB_RNR_TIMER_245_76 = 29,
934 IB_RNR_TIMER_327_68 = 30,
935 IB_RNR_TIMER_491_52 = 31
938 enum ib_qp_attr_mask {
940 IB_QP_CUR_STATE = (1<<1),
941 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
942 IB_QP_ACCESS_FLAGS = (1<<3),
943 IB_QP_PKEY_INDEX = (1<<4),
947 IB_QP_PATH_MTU = (1<<8),
948 IB_QP_TIMEOUT = (1<<9),
949 IB_QP_RETRY_CNT = (1<<10),
950 IB_QP_RNR_RETRY = (1<<11),
951 IB_QP_RQ_PSN = (1<<12),
952 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
953 IB_QP_ALT_PATH = (1<<14),
954 IB_QP_MIN_RNR_TIMER = (1<<15),
955 IB_QP_SQ_PSN = (1<<16),
956 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
957 IB_QP_PATH_MIG_STATE = (1<<18),
959 IB_QP_DEST_QPN = (1<<20),
960 IB_QP_RESERVED1 = (1<<21),
961 IB_QP_RESERVED2 = (1<<22),
962 IB_QP_RESERVED3 = (1<<23),
963 IB_QP_RESERVED4 = (1<<24),
988 enum ib_qp_state qp_state;
989 enum ib_qp_state cur_qp_state;
990 enum ib_mtu path_mtu;
991 enum ib_mig_state path_mig_state;
997 struct ib_qp_cap cap;
998 struct ib_ah_attr ah_attr;
999 struct ib_ah_attr alt_ah_attr;
1002 u8 en_sqd_async_notify;
1005 u8 max_dest_rd_atomic;
1017 IB_WR_RDMA_WRITE_WITH_IMM,
1019 IB_WR_SEND_WITH_IMM,
1021 IB_WR_ATOMIC_CMP_AND_SWP,
1022 IB_WR_ATOMIC_FETCH_AND_ADD,
1024 IB_WR_SEND_WITH_INV,
1025 IB_WR_RDMA_READ_WITH_INV,
1028 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1029 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1032 /* reserve values for low level drivers' internal use.
1033 * These values will not be used at all in the ib core layer.
1035 IB_WR_RESERVED1 = 0xf0,
1047 enum ib_send_flags {
1049 IB_SEND_SIGNALED = (1<<1),
1050 IB_SEND_SOLICITED = (1<<2),
1051 IB_SEND_INLINE = (1<<3),
1052 IB_SEND_IP_CSUM = (1<<4),
1054 /* reserve bits 26-31 for low level drivers' internal use */
1055 IB_SEND_RESERVED_START = (1 << 26),
1056 IB_SEND_RESERVED_END = (1 << 31),
1065 struct ib_fast_reg_page_list {
1066 struct ib_device *device;
1068 unsigned int max_page_list_len;
1072 * struct ib_mw_bind_info - Parameters for a memory window bind operation.
1073 * @mr: A memory region to bind the memory window to.
1074 * @addr: The address where the memory window should begin.
1075 * @length: The length of the memory window, in bytes.
1076 * @mw_access_flags: Access flags from enum ib_access_flags for the window.
1078 * This struct contains the shared parameters for type 1 and type 2
1079 * memory window bind operations.
1081 struct ib_mw_bind_info {
1085 int mw_access_flags;
1089 struct ib_send_wr *next;
1091 struct ib_sge *sg_list;
1093 enum ib_wr_opcode opcode;
1097 u32 invalidate_rkey;
1102 struct ib_send_wr wr;
1107 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1109 return container_of(wr, struct ib_rdma_wr, wr);
1112 struct ib_atomic_wr {
1113 struct ib_send_wr wr;
1117 u64 compare_add_mask;
1122 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1124 return container_of(wr, struct ib_atomic_wr, wr);
1128 struct ib_send_wr wr;
1135 u16 pkey_index; /* valid for GSI only */
1136 u8 port_num; /* valid for DR SMPs on switch only */
1139 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1141 return container_of(wr, struct ib_ud_wr, wr);
1144 struct ib_fast_reg_wr {
1145 struct ib_send_wr wr;
1147 struct ib_fast_reg_page_list *page_list;
1148 unsigned int page_shift;
1149 unsigned int page_list_len;
1155 static inline struct ib_fast_reg_wr *fast_reg_wr(struct ib_send_wr *wr)
1157 return container_of(wr, struct ib_fast_reg_wr, wr);
1160 struct ib_bind_mw_wr {
1161 struct ib_send_wr wr;
1163 /* The new rkey for the memory window. */
1165 struct ib_mw_bind_info bind_info;
1168 static inline struct ib_bind_mw_wr *bind_mw_wr(struct ib_send_wr *wr)
1170 return container_of(wr, struct ib_bind_mw_wr, wr);
1173 struct ib_sig_handover_wr {
1174 struct ib_send_wr wr;
1175 struct ib_sig_attrs *sig_attrs;
1176 struct ib_mr *sig_mr;
1178 struct ib_sge *prot;
1181 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1183 return container_of(wr, struct ib_sig_handover_wr, wr);
1187 struct ib_recv_wr *next;
1189 struct ib_sge *sg_list;
1193 enum ib_access_flags {
1194 IB_ACCESS_LOCAL_WRITE = 1,
1195 IB_ACCESS_REMOTE_WRITE = (1<<1),
1196 IB_ACCESS_REMOTE_READ = (1<<2),
1197 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1198 IB_ACCESS_MW_BIND = (1<<4),
1199 IB_ZERO_BASED = (1<<5),
1200 IB_ACCESS_ON_DEMAND = (1<<6),
1203 struct ib_phys_buf {
1210 u64 device_virt_addr;
1212 int mr_access_flags;
1217 enum ib_mr_rereg_flags {
1218 IB_MR_REREG_TRANS = 1,
1219 IB_MR_REREG_PD = (1<<1),
1220 IB_MR_REREG_ACCESS = (1<<2),
1221 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1225 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
1226 * @wr_id: Work request id.
1227 * @send_flags: Flags from ib_send_flags enum.
1228 * @bind_info: More parameters of the bind operation.
1233 struct ib_mw_bind_info bind_info;
1236 struct ib_fmr_attr {
1244 struct ib_ucontext {
1245 struct ib_device *device;
1246 struct list_head pd_list;
1247 struct list_head mr_list;
1248 struct list_head mw_list;
1249 struct list_head cq_list;
1250 struct list_head qp_list;
1251 struct list_head srq_list;
1252 struct list_head ah_list;
1253 struct list_head xrcd_list;
1254 struct list_head rule_list;
1258 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1259 struct rb_root umem_tree;
1261 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1262 * mmu notifiers registration.
1264 struct rw_semaphore umem_rwsem;
1265 void (*invalidate_range)(struct ib_umem *umem,
1266 unsigned long start, unsigned long end);
1268 struct mmu_notifier mn;
1269 atomic_t notifier_count;
1270 /* A list of umems that don't have private mmu notifier counters yet. */
1271 struct list_head no_private_counters;
1277 u64 user_handle; /* handle given to us by userspace */
1278 struct ib_ucontext *context; /* associated user context */
1279 void *object; /* containing object */
1280 struct list_head list; /* link to context's list */
1281 int id; /* index into kernel idr */
1283 struct rw_semaphore mutex; /* protects .live */
1288 const void __user *inbuf;
1289 void __user *outbuf;
1296 struct ib_device *device;
1297 struct ib_uobject *uobject;
1298 atomic_t usecnt; /* count all resources */
1299 struct ib_mr *local_mr;
1303 struct ib_device *device;
1304 atomic_t usecnt; /* count all exposed resources */
1305 struct inode *inode;
1307 struct mutex tgt_qp_mutex;
1308 struct list_head tgt_qp_list;
1312 struct ib_device *device;
1314 struct ib_uobject *uobject;
1317 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1320 struct ib_device *device;
1321 struct ib_uobject *uobject;
1322 ib_comp_handler comp_handler;
1323 void (*event_handler)(struct ib_event *, void *);
1326 atomic_t usecnt; /* count number of work queues */
1330 struct ib_device *device;
1332 struct ib_uobject *uobject;
1333 void (*event_handler)(struct ib_event *, void *);
1335 enum ib_srq_type srq_type;
1340 struct ib_xrcd *xrcd;
1348 struct ib_device *device;
1350 struct ib_cq *send_cq;
1351 struct ib_cq *recv_cq;
1353 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1354 struct list_head xrcd_list;
1355 /* count times opened, mcast attaches, flow attaches */
1357 struct list_head open_list;
1358 struct ib_qp *real_qp;
1359 struct ib_uobject *uobject;
1360 void (*event_handler)(struct ib_event *, void *);
1363 enum ib_qp_type qp_type;
1367 struct ib_device *device;
1369 struct ib_uobject *uobject;
1372 atomic_t usecnt; /* count number of MWs */
1376 struct ib_device *device;
1378 struct ib_uobject *uobject;
1380 enum ib_mw_type type;
1384 struct ib_device *device;
1386 struct list_head list;
1391 /* Supported steering options */
1392 enum ib_flow_attr_type {
1393 /* steering according to rule specifications */
1394 IB_FLOW_ATTR_NORMAL = 0x0,
1395 /* default unicast and multicast rule -
1396 * receive all Eth traffic which isn't steered to any QP
1398 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1399 /* default multicast rule -
1400 * receive all Eth multicast traffic which isn't steered to any QP
1402 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1403 /* sniffer rule - receive all port traffic */
1404 IB_FLOW_ATTR_SNIFFER = 0x3
1407 /* Supported steering header types */
1408 enum ib_flow_spec_type {
1410 IB_FLOW_SPEC_ETH = 0x20,
1411 IB_FLOW_SPEC_IB = 0x22,
1413 IB_FLOW_SPEC_IPV4 = 0x30,
1415 IB_FLOW_SPEC_TCP = 0x40,
1416 IB_FLOW_SPEC_UDP = 0x41
1418 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1419 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1421 /* Flow steering rule priority is set according to it's domain.
1422 * Lower domain value means higher priority.
1424 enum ib_flow_domain {
1425 IB_FLOW_DOMAIN_USER,
1426 IB_FLOW_DOMAIN_ETHTOOL,
1429 IB_FLOW_DOMAIN_NUM /* Must be last */
1432 struct ib_flow_eth_filter {
1439 struct ib_flow_spec_eth {
1440 enum ib_flow_spec_type type;
1442 struct ib_flow_eth_filter val;
1443 struct ib_flow_eth_filter mask;
1446 struct ib_flow_ib_filter {
1451 struct ib_flow_spec_ib {
1452 enum ib_flow_spec_type type;
1454 struct ib_flow_ib_filter val;
1455 struct ib_flow_ib_filter mask;
1458 struct ib_flow_ipv4_filter {
1463 struct ib_flow_spec_ipv4 {
1464 enum ib_flow_spec_type type;
1466 struct ib_flow_ipv4_filter val;
1467 struct ib_flow_ipv4_filter mask;
1470 struct ib_flow_tcp_udp_filter {
1475 struct ib_flow_spec_tcp_udp {
1476 enum ib_flow_spec_type type;
1478 struct ib_flow_tcp_udp_filter val;
1479 struct ib_flow_tcp_udp_filter mask;
1482 union ib_flow_spec {
1484 enum ib_flow_spec_type type;
1487 struct ib_flow_spec_eth eth;
1488 struct ib_flow_spec_ib ib;
1489 struct ib_flow_spec_ipv4 ipv4;
1490 struct ib_flow_spec_tcp_udp tcp_udp;
1493 struct ib_flow_attr {
1494 enum ib_flow_attr_type type;
1500 /* Following are the optional layers according to user request
1501 * struct ib_flow_spec_xxx
1502 * struct ib_flow_spec_yyy
1508 struct ib_uobject *uobject;
1514 enum ib_process_mad_flags {
1515 IB_MAD_IGNORE_MKEY = 1,
1516 IB_MAD_IGNORE_BKEY = 2,
1517 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1520 enum ib_mad_result {
1521 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1522 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1523 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1524 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1527 #define IB_DEVICE_NAME_MAX 64
1531 struct ib_event_handler event_handler;
1532 struct ib_pkey_cache **pkey_cache;
1533 struct ib_gid_table **gid_cache;
1537 struct ib_dma_mapping_ops {
1538 int (*mapping_error)(struct ib_device *dev,
1540 u64 (*map_single)(struct ib_device *dev,
1541 void *ptr, size_t size,
1542 enum dma_data_direction direction);
1543 void (*unmap_single)(struct ib_device *dev,
1544 u64 addr, size_t size,
1545 enum dma_data_direction direction);
1546 u64 (*map_page)(struct ib_device *dev,
1547 struct page *page, unsigned long offset,
1549 enum dma_data_direction direction);
1550 void (*unmap_page)(struct ib_device *dev,
1551 u64 addr, size_t size,
1552 enum dma_data_direction direction);
1553 int (*map_sg)(struct ib_device *dev,
1554 struct scatterlist *sg, int nents,
1555 enum dma_data_direction direction);
1556 void (*unmap_sg)(struct ib_device *dev,
1557 struct scatterlist *sg, int nents,
1558 enum dma_data_direction direction);
1559 void (*sync_single_for_cpu)(struct ib_device *dev,
1562 enum dma_data_direction dir);
1563 void (*sync_single_for_device)(struct ib_device *dev,
1566 enum dma_data_direction dir);
1567 void *(*alloc_coherent)(struct ib_device *dev,
1571 void (*free_coherent)(struct ib_device *dev,
1572 size_t size, void *cpu_addr,
1578 struct ib_port_immutable {
1586 struct device *dma_device;
1588 char name[IB_DEVICE_NAME_MAX];
1590 struct list_head event_handler_list;
1591 spinlock_t event_handler_lock;
1593 spinlock_t client_data_lock;
1594 struct list_head core_list;
1595 /* Access to the client_data_list is protected by the client_data_lock
1596 * spinlock and the lists_rwsem read-write semaphore */
1597 struct list_head client_data_list;
1599 struct ib_cache cache;
1601 * port_immutable is indexed by port number
1603 struct ib_port_immutable *port_immutable;
1605 int num_comp_vectors;
1607 struct iw_cm_verbs *iwcm;
1609 int (*get_protocol_stats)(struct ib_device *device,
1610 union rdma_protocol_stats *stats);
1611 int (*query_device)(struct ib_device *device,
1612 struct ib_device_attr *device_attr,
1613 struct ib_udata *udata);
1614 int (*query_port)(struct ib_device *device,
1616 struct ib_port_attr *port_attr);
1617 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1619 /* When calling get_netdev, the HW vendor's driver should return the
1620 * net device of device @device at port @port_num or NULL if such
1621 * a net device doesn't exist. The vendor driver should call dev_hold
1622 * on this net device. The HW vendor's device driver must guarantee
1623 * that this function returns NULL before the net device reaches
1624 * NETDEV_UNREGISTER_FINAL state.
1626 struct net_device *(*get_netdev)(struct ib_device *device,
1628 int (*query_gid)(struct ib_device *device,
1629 u8 port_num, int index,
1631 /* When calling add_gid, the HW vendor's driver should
1632 * add the gid of device @device at gid index @index of
1633 * port @port_num to be @gid. Meta-info of that gid (for example,
1634 * the network device related to this gid is available
1635 * at @attr. @context allows the HW vendor driver to store extra
1636 * information together with a GID entry. The HW vendor may allocate
1637 * memory to contain this information and store it in @context when a
1638 * new GID entry is written to. Params are consistent until the next
1639 * call of add_gid or delete_gid. The function should return 0 on
1640 * success or error otherwise. The function could be called
1641 * concurrently for different ports. This function is only called
1642 * when roce_gid_table is used.
1644 int (*add_gid)(struct ib_device *device,
1647 const union ib_gid *gid,
1648 const struct ib_gid_attr *attr,
1650 /* When calling del_gid, the HW vendor's driver should delete the
1651 * gid of device @device at gid index @index of port @port_num.
1652 * Upon the deletion of a GID entry, the HW vendor must free any
1653 * allocated memory. The caller will clear @context afterwards.
1654 * This function is only called when roce_gid_table is used.
1656 int (*del_gid)(struct ib_device *device,
1660 int (*query_pkey)(struct ib_device *device,
1661 u8 port_num, u16 index, u16 *pkey);
1662 int (*modify_device)(struct ib_device *device,
1663 int device_modify_mask,
1664 struct ib_device_modify *device_modify);
1665 int (*modify_port)(struct ib_device *device,
1666 u8 port_num, int port_modify_mask,
1667 struct ib_port_modify *port_modify);
1668 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1669 struct ib_udata *udata);
1670 int (*dealloc_ucontext)(struct ib_ucontext *context);
1671 int (*mmap)(struct ib_ucontext *context,
1672 struct vm_area_struct *vma);
1673 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1674 struct ib_ucontext *context,
1675 struct ib_udata *udata);
1676 int (*dealloc_pd)(struct ib_pd *pd);
1677 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1678 struct ib_ah_attr *ah_attr);
1679 int (*modify_ah)(struct ib_ah *ah,
1680 struct ib_ah_attr *ah_attr);
1681 int (*query_ah)(struct ib_ah *ah,
1682 struct ib_ah_attr *ah_attr);
1683 int (*destroy_ah)(struct ib_ah *ah);
1684 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1685 struct ib_srq_init_attr *srq_init_attr,
1686 struct ib_udata *udata);
1687 int (*modify_srq)(struct ib_srq *srq,
1688 struct ib_srq_attr *srq_attr,
1689 enum ib_srq_attr_mask srq_attr_mask,
1690 struct ib_udata *udata);
1691 int (*query_srq)(struct ib_srq *srq,
1692 struct ib_srq_attr *srq_attr);
1693 int (*destroy_srq)(struct ib_srq *srq);
1694 int (*post_srq_recv)(struct ib_srq *srq,
1695 struct ib_recv_wr *recv_wr,
1696 struct ib_recv_wr **bad_recv_wr);
1697 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1698 struct ib_qp_init_attr *qp_init_attr,
1699 struct ib_udata *udata);
1700 int (*modify_qp)(struct ib_qp *qp,
1701 struct ib_qp_attr *qp_attr,
1703 struct ib_udata *udata);
1704 int (*query_qp)(struct ib_qp *qp,
1705 struct ib_qp_attr *qp_attr,
1707 struct ib_qp_init_attr *qp_init_attr);
1708 int (*destroy_qp)(struct ib_qp *qp);
1709 int (*post_send)(struct ib_qp *qp,
1710 struct ib_send_wr *send_wr,
1711 struct ib_send_wr **bad_send_wr);
1712 int (*post_recv)(struct ib_qp *qp,
1713 struct ib_recv_wr *recv_wr,
1714 struct ib_recv_wr **bad_recv_wr);
1715 struct ib_cq * (*create_cq)(struct ib_device *device,
1716 const struct ib_cq_init_attr *attr,
1717 struct ib_ucontext *context,
1718 struct ib_udata *udata);
1719 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1721 int (*destroy_cq)(struct ib_cq *cq);
1722 int (*resize_cq)(struct ib_cq *cq, int cqe,
1723 struct ib_udata *udata);
1724 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1726 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1727 int (*req_notify_cq)(struct ib_cq *cq,
1728 enum ib_cq_notify_flags flags);
1729 int (*req_ncomp_notif)(struct ib_cq *cq,
1731 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1732 int mr_access_flags);
1733 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
1734 struct ib_phys_buf *phys_buf_array,
1736 int mr_access_flags,
1738 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1739 u64 start, u64 length,
1741 int mr_access_flags,
1742 struct ib_udata *udata);
1743 int (*rereg_user_mr)(struct ib_mr *mr,
1745 u64 start, u64 length,
1747 int mr_access_flags,
1749 struct ib_udata *udata);
1750 int (*query_mr)(struct ib_mr *mr,
1751 struct ib_mr_attr *mr_attr);
1752 int (*dereg_mr)(struct ib_mr *mr);
1753 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
1754 enum ib_mr_type mr_type,
1756 struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1758 void (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1759 int (*rereg_phys_mr)(struct ib_mr *mr,
1762 struct ib_phys_buf *phys_buf_array,
1764 int mr_access_flags,
1766 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1767 enum ib_mw_type type);
1768 int (*bind_mw)(struct ib_qp *qp,
1770 struct ib_mw_bind *mw_bind);
1771 int (*dealloc_mw)(struct ib_mw *mw);
1772 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1773 int mr_access_flags,
1774 struct ib_fmr_attr *fmr_attr);
1775 int (*map_phys_fmr)(struct ib_fmr *fmr,
1776 u64 *page_list, int list_len,
1778 int (*unmap_fmr)(struct list_head *fmr_list);
1779 int (*dealloc_fmr)(struct ib_fmr *fmr);
1780 int (*attach_mcast)(struct ib_qp *qp,
1783 int (*detach_mcast)(struct ib_qp *qp,
1786 int (*process_mad)(struct ib_device *device,
1787 int process_mad_flags,
1789 const struct ib_wc *in_wc,
1790 const struct ib_grh *in_grh,
1791 const struct ib_mad_hdr *in_mad,
1793 struct ib_mad_hdr *out_mad,
1794 size_t *out_mad_size,
1795 u16 *out_mad_pkey_index);
1796 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1797 struct ib_ucontext *ucontext,
1798 struct ib_udata *udata);
1799 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1800 struct ib_flow * (*create_flow)(struct ib_qp *qp,
1804 int (*destroy_flow)(struct ib_flow *flow_id);
1805 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1806 struct ib_mr_status *mr_status);
1807 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
1809 struct ib_dma_mapping_ops *dma_ops;
1811 struct module *owner;
1813 struct kobject *ports_parent;
1814 struct list_head port_list;
1817 IB_DEV_UNINITIALIZED,
1823 u64 uverbs_cmd_mask;
1824 u64 uverbs_ex_cmd_mask;
1834 * The following mandatory functions are used only at device
1835 * registration. Keep functions such as these at the end of this
1836 * structure to avoid cache line misses when accessing struct ib_device
1839 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1844 void (*add) (struct ib_device *);
1845 void (*remove)(struct ib_device *, void *client_data);
1847 /* Returns the net_dev belonging to this ib_client and matching the
1849 * @dev: An RDMA device that the net_dev use for communication.
1850 * @port: A physical port number on the RDMA device.
1851 * @pkey: P_Key that the net_dev uses if applicable.
1852 * @gid: A GID that the net_dev uses to communicate.
1853 * @addr: An IP address the net_dev is configured with.
1854 * @client_data: The device's client data set by ib_set_client_data().
1856 * An ib_client that implements a net_dev on top of RDMA devices
1857 * (such as IP over IB) should implement this callback, allowing the
1858 * rdma_cm module to find the right net_dev for a given request.
1860 * The caller is responsible for calling dev_put on the returned
1862 struct net_device *(*get_net_dev_by_params)(
1863 struct ib_device *dev,
1866 const union ib_gid *gid,
1867 const struct sockaddr *addr,
1869 struct list_head list;
1872 struct ib_device *ib_alloc_device(size_t size);
1873 void ib_dealloc_device(struct ib_device *device);
1875 int ib_register_device(struct ib_device *device,
1876 int (*port_callback)(struct ib_device *,
1877 u8, struct kobject *));
1878 void ib_unregister_device(struct ib_device *device);
1880 int ib_register_client (struct ib_client *client);
1881 void ib_unregister_client(struct ib_client *client);
1883 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1884 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1887 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1889 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1892 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1894 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1898 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1899 * contains all required attributes and no attributes not allowed for
1900 * the given QP state transition.
1901 * @cur_state: Current QP state
1902 * @next_state: Next QP state
1904 * @mask: Mask of supplied QP attributes
1905 * @ll : link layer of port
1907 * This function is a helper function that a low-level driver's
1908 * modify_qp method can use to validate the consumer's input. It
1909 * checks that cur_state and next_state are valid QP states, that a
1910 * transition from cur_state to next_state is allowed by the IB spec,
1911 * and that the attribute mask supplied is allowed for the transition.
1913 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1914 enum ib_qp_type type, enum ib_qp_attr_mask mask,
1915 enum rdma_link_layer ll);
1917 int ib_register_event_handler (struct ib_event_handler *event_handler);
1918 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1919 void ib_dispatch_event(struct ib_event *event);
1921 int ib_query_device(struct ib_device *device,
1922 struct ib_device_attr *device_attr);
1924 int ib_query_port(struct ib_device *device,
1925 u8 port_num, struct ib_port_attr *port_attr);
1927 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1931 * rdma_cap_ib_switch - Check if the device is IB switch
1932 * @device: Device to check
1934 * Device driver is responsible for setting is_switch bit on
1935 * in ib_device structure at init time.
1937 * Return: true if the device is IB switch.
1939 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
1941 return device->is_switch;
1945 * rdma_start_port - Return the first valid port number for the device
1948 * @device: Device to be checked
1950 * Return start port number
1952 static inline u8 rdma_start_port(const struct ib_device *device)
1954 return rdma_cap_ib_switch(device) ? 0 : 1;
1958 * rdma_end_port - Return the last valid port number for the device
1961 * @device: Device to be checked
1963 * Return last port number
1965 static inline u8 rdma_end_port(const struct ib_device *device)
1967 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
1970 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
1972 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
1975 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
1977 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
1980 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
1982 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
1985 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
1987 return device->port_immutable[port_num].core_cap_flags &
1988 (RDMA_CORE_CAP_PROT_IB | RDMA_CORE_CAP_PROT_ROCE);
1992 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
1993 * Management Datagrams.
1994 * @device: Device to check
1995 * @port_num: Port number to check
1997 * Management Datagrams (MAD) are a required part of the InfiniBand
1998 * specification and are supported on all InfiniBand devices. A slightly
1999 * extended version are also supported on OPA interfaces.
2001 * Return: true if the port supports sending/receiving of MAD packets.
2003 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2005 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2009 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2010 * Management Datagrams.
2011 * @device: Device to check
2012 * @port_num: Port number to check
2014 * Intel OmniPath devices extend and/or replace the InfiniBand Management
2015 * datagrams with their own versions. These OPA MADs share many but not all of
2016 * the characteristics of InfiniBand MADs.
2018 * OPA MADs differ in the following ways:
2020 * 1) MADs are variable size up to 2K
2021 * IBTA defined MADs remain fixed at 256 bytes
2022 * 2) OPA SMPs must carry valid PKeys
2023 * 3) OPA SMP packets are a different format
2025 * Return: true if the port supports OPA MAD packet formats.
2027 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2029 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2030 == RDMA_CORE_CAP_OPA_MAD;
2034 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2035 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2036 * @device: Device to check
2037 * @port_num: Port number to check
2039 * Each InfiniBand node is required to provide a Subnet Management Agent
2040 * that the subnet manager can access. Prior to the fabric being fully
2041 * configured by the subnet manager, the SMA is accessed via a well known
2042 * interface called the Subnet Management Interface (SMI). This interface
2043 * uses directed route packets to communicate with the SM to get around the
2044 * chicken and egg problem of the SM needing to know what's on the fabric
2045 * in order to configure the fabric, and needing to configure the fabric in
2046 * order to send packets to the devices on the fabric. These directed
2047 * route packets do not need the fabric fully configured in order to reach
2048 * their destination. The SMI is the only method allowed to send
2049 * directed route packets on an InfiniBand fabric.
2051 * Return: true if the port provides an SMI.
2053 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2055 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2059 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2060 * Communication Manager.
2061 * @device: Device to check
2062 * @port_num: Port number to check
2064 * The InfiniBand Communication Manager is one of many pre-defined General
2065 * Service Agents (GSA) that are accessed via the General Service
2066 * Interface (GSI). It's role is to facilitate establishment of connections
2067 * between nodes as well as other management related tasks for established
2070 * Return: true if the port supports an IB CM (this does not guarantee that
2071 * a CM is actually running however).
2073 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2075 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2079 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2080 * Communication Manager.
2081 * @device: Device to check
2082 * @port_num: Port number to check
2084 * Similar to above, but specific to iWARP connections which have a different
2085 * managment protocol than InfiniBand.
2087 * Return: true if the port supports an iWARP CM (this does not guarantee that
2088 * a CM is actually running however).
2090 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2092 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2096 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2097 * Subnet Administration.
2098 * @device: Device to check
2099 * @port_num: Port number to check
2101 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2102 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2103 * fabrics, devices should resolve routes to other hosts by contacting the
2104 * SA to query the proper route.
2106 * Return: true if the port should act as a client to the fabric Subnet
2107 * Administration interface. This does not imply that the SA service is
2110 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2112 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2116 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2118 * @device: Device to check
2119 * @port_num: Port number to check
2121 * InfiniBand multicast registration is more complex than normal IPv4 or
2122 * IPv6 multicast registration. Each Host Channel Adapter must register
2123 * with the Subnet Manager when it wishes to join a multicast group. It
2124 * should do so only once regardless of how many queue pairs it subscribes
2125 * to this group. And it should leave the group only after all queue pairs
2126 * attached to the group have been detached.
2128 * Return: true if the port must undertake the additional adminstrative
2129 * overhead of registering/unregistering with the SM and tracking of the
2130 * total number of queue pairs attached to the multicast group.
2132 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2134 return rdma_cap_ib_sa(device, port_num);
2138 * rdma_cap_af_ib - Check if the port of device has the capability
2139 * Native Infiniband Address.
2140 * @device: Device to check
2141 * @port_num: Port number to check
2143 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2144 * GID. RoCE uses a different mechanism, but still generates a GID via
2145 * a prescribed mechanism and port specific data.
2147 * Return: true if the port uses a GID address to identify devices on the
2150 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2152 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2156 * rdma_cap_eth_ah - Check if the port of device has the capability
2157 * Ethernet Address Handle.
2158 * @device: Device to check
2159 * @port_num: Port number to check
2161 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2162 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2163 * port. Normally, packet headers are generated by the sending host
2164 * adapter, but when sending connectionless datagrams, we must manually
2165 * inject the proper headers for the fabric we are communicating over.
2167 * Return: true if we are running as a RoCE port and must force the
2168 * addition of a Global Route Header built from our Ethernet Address
2169 * Handle into our header list for connectionless packets.
2171 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2173 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2177 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2180 * @port_num: Port number
2182 * This MAD size includes the MAD headers and MAD payload. No other headers
2185 * Return the max MAD size required by the Port. Will return 0 if the port
2186 * does not support MADs
2188 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2190 return device->port_immutable[port_num].max_mad_size;
2194 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2195 * @device: Device to check
2196 * @port_num: Port number to check
2198 * RoCE GID table mechanism manages the various GIDs for a device.
2200 * NOTE: if allocating the port's GID table has failed, this call will still
2201 * return true, but any RoCE GID table API will fail.
2203 * Return: true if the port uses RoCE GID table mechanism in order to manage
2206 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2209 return rdma_protocol_roce(device, port_num) &&
2210 device->add_gid && device->del_gid;
2213 int ib_query_gid(struct ib_device *device,
2214 u8 port_num, int index, union ib_gid *gid,
2215 struct ib_gid_attr *attr);
2217 int ib_query_pkey(struct ib_device *device,
2218 u8 port_num, u16 index, u16 *pkey);
2220 int ib_modify_device(struct ib_device *device,
2221 int device_modify_mask,
2222 struct ib_device_modify *device_modify);
2224 int ib_modify_port(struct ib_device *device,
2225 u8 port_num, int port_modify_mask,
2226 struct ib_port_modify *port_modify);
2228 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2229 struct net_device *ndev, u8 *port_num, u16 *index);
2231 int ib_find_pkey(struct ib_device *device,
2232 u8 port_num, u16 pkey, u16 *index);
2234 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2236 void ib_dealloc_pd(struct ib_pd *pd);
2239 * ib_create_ah - Creates an address handle for the given address vector.
2240 * @pd: The protection domain associated with the address handle.
2241 * @ah_attr: The attributes of the address vector.
2243 * The address handle is used to reference a local or global destination
2244 * in all UD QP post sends.
2246 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2249 * ib_init_ah_from_wc - Initializes address handle attributes from a
2251 * @device: Device on which the received message arrived.
2252 * @port_num: Port on which the received message arrived.
2253 * @wc: Work completion associated with the received message.
2254 * @grh: References the received global route header. This parameter is
2255 * ignored unless the work completion indicates that the GRH is valid.
2256 * @ah_attr: Returned attributes that can be used when creating an address
2257 * handle for replying to the message.
2259 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2260 const struct ib_wc *wc, const struct ib_grh *grh,
2261 struct ib_ah_attr *ah_attr);
2264 * ib_create_ah_from_wc - Creates an address handle associated with the
2265 * sender of the specified work completion.
2266 * @pd: The protection domain associated with the address handle.
2267 * @wc: Work completion information associated with a received message.
2268 * @grh: References the received global route header. This parameter is
2269 * ignored unless the work completion indicates that the GRH is valid.
2270 * @port_num: The outbound port number to associate with the address.
2272 * The address handle is used to reference a local or global destination
2273 * in all UD QP post sends.
2275 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2276 const struct ib_grh *grh, u8 port_num);
2279 * ib_modify_ah - Modifies the address vector associated with an address
2281 * @ah: The address handle to modify.
2282 * @ah_attr: The new address vector attributes to associate with the
2285 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2288 * ib_query_ah - Queries the address vector associated with an address
2290 * @ah: The address handle to query.
2291 * @ah_attr: The address vector attributes associated with the address
2294 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2297 * ib_destroy_ah - Destroys an address handle.
2298 * @ah: The address handle to destroy.
2300 int ib_destroy_ah(struct ib_ah *ah);
2303 * ib_create_srq - Creates a SRQ associated with the specified protection
2305 * @pd: The protection domain associated with the SRQ.
2306 * @srq_init_attr: A list of initial attributes required to create the
2307 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2308 * the actual capabilities of the created SRQ.
2310 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2311 * requested size of the SRQ, and set to the actual values allocated
2312 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2313 * will always be at least as large as the requested values.
2315 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2316 struct ib_srq_init_attr *srq_init_attr);
2319 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2320 * @srq: The SRQ to modify.
2321 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2322 * the current values of selected SRQ attributes are returned.
2323 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2324 * are being modified.
2326 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2327 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2328 * the number of receives queued drops below the limit.
2330 int ib_modify_srq(struct ib_srq *srq,
2331 struct ib_srq_attr *srq_attr,
2332 enum ib_srq_attr_mask srq_attr_mask);
2335 * ib_query_srq - Returns the attribute list and current values for the
2337 * @srq: The SRQ to query.
2338 * @srq_attr: The attributes of the specified SRQ.
2340 int ib_query_srq(struct ib_srq *srq,
2341 struct ib_srq_attr *srq_attr);
2344 * ib_destroy_srq - Destroys the specified SRQ.
2345 * @srq: The SRQ to destroy.
2347 int ib_destroy_srq(struct ib_srq *srq);
2350 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2351 * @srq: The SRQ to post the work request on.
2352 * @recv_wr: A list of work requests to post on the receive queue.
2353 * @bad_recv_wr: On an immediate failure, this parameter will reference
2354 * the work request that failed to be posted on the QP.
2356 static inline int ib_post_srq_recv(struct ib_srq *srq,
2357 struct ib_recv_wr *recv_wr,
2358 struct ib_recv_wr **bad_recv_wr)
2360 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2364 * ib_create_qp - Creates a QP associated with the specified protection
2366 * @pd: The protection domain associated with the QP.
2367 * @qp_init_attr: A list of initial attributes required to create the
2368 * QP. If QP creation succeeds, then the attributes are updated to
2369 * the actual capabilities of the created QP.
2371 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2372 struct ib_qp_init_attr *qp_init_attr);
2375 * ib_modify_qp - Modifies the attributes for the specified QP and then
2376 * transitions the QP to the given state.
2377 * @qp: The QP to modify.
2378 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2379 * the current values of selected QP attributes are returned.
2380 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2381 * are being modified.
2383 int ib_modify_qp(struct ib_qp *qp,
2384 struct ib_qp_attr *qp_attr,
2388 * ib_query_qp - Returns the attribute list and current values for the
2390 * @qp: The QP to query.
2391 * @qp_attr: The attributes of the specified QP.
2392 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2393 * @qp_init_attr: Additional attributes of the selected QP.
2395 * The qp_attr_mask may be used to limit the query to gathering only the
2396 * selected attributes.
2398 int ib_query_qp(struct ib_qp *qp,
2399 struct ib_qp_attr *qp_attr,
2401 struct ib_qp_init_attr *qp_init_attr);
2404 * ib_destroy_qp - Destroys the specified QP.
2405 * @qp: The QP to destroy.
2407 int ib_destroy_qp(struct ib_qp *qp);
2410 * ib_open_qp - Obtain a reference to an existing sharable QP.
2411 * @xrcd - XRC domain
2412 * @qp_open_attr: Attributes identifying the QP to open.
2414 * Returns a reference to a sharable QP.
2416 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2417 struct ib_qp_open_attr *qp_open_attr);
2420 * ib_close_qp - Release an external reference to a QP.
2421 * @qp: The QP handle to release
2423 * The opened QP handle is released by the caller. The underlying
2424 * shared QP is not destroyed until all internal references are released.
2426 int ib_close_qp(struct ib_qp *qp);
2429 * ib_post_send - Posts a list of work requests to the send queue of
2431 * @qp: The QP to post the work request on.
2432 * @send_wr: A list of work requests to post on the send queue.
2433 * @bad_send_wr: On an immediate failure, this parameter will reference
2434 * the work request that failed to be posted on the QP.
2436 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2437 * error is returned, the QP state shall not be affected,
2438 * ib_post_send() will return an immediate error after queueing any
2439 * earlier work requests in the list.
2441 static inline int ib_post_send(struct ib_qp *qp,
2442 struct ib_send_wr *send_wr,
2443 struct ib_send_wr **bad_send_wr)
2445 return qp->device->post_send(qp, send_wr, bad_send_wr);
2449 * ib_post_recv - Posts a list of work requests to the receive queue of
2451 * @qp: The QP to post the work request on.
2452 * @recv_wr: A list of work requests to post on the receive queue.
2453 * @bad_recv_wr: On an immediate failure, this parameter will reference
2454 * the work request that failed to be posted on the QP.
2456 static inline int ib_post_recv(struct ib_qp *qp,
2457 struct ib_recv_wr *recv_wr,
2458 struct ib_recv_wr **bad_recv_wr)
2460 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2464 * ib_create_cq - Creates a CQ on the specified device.
2465 * @device: The device on which to create the CQ.
2466 * @comp_handler: A user-specified callback that is invoked when a
2467 * completion event occurs on the CQ.
2468 * @event_handler: A user-specified callback that is invoked when an
2469 * asynchronous event not associated with a completion occurs on the CQ.
2470 * @cq_context: Context associated with the CQ returned to the user via
2471 * the associated completion and event handlers.
2472 * @cq_attr: The attributes the CQ should be created upon.
2474 * Users can examine the cq structure to determine the actual CQ size.
2476 struct ib_cq *ib_create_cq(struct ib_device *device,
2477 ib_comp_handler comp_handler,
2478 void (*event_handler)(struct ib_event *, void *),
2480 const struct ib_cq_init_attr *cq_attr);
2483 * ib_resize_cq - Modifies the capacity of the CQ.
2484 * @cq: The CQ to resize.
2485 * @cqe: The minimum size of the CQ.
2487 * Users can examine the cq structure to determine the actual CQ size.
2489 int ib_resize_cq(struct ib_cq *cq, int cqe);
2492 * ib_modify_cq - Modifies moderation params of the CQ
2493 * @cq: The CQ to modify.
2494 * @cq_count: number of CQEs that will trigger an event
2495 * @cq_period: max period of time in usec before triggering an event
2498 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2501 * ib_destroy_cq - Destroys the specified CQ.
2502 * @cq: The CQ to destroy.
2504 int ib_destroy_cq(struct ib_cq *cq);
2507 * ib_poll_cq - poll a CQ for completion(s)
2508 * @cq:the CQ being polled
2509 * @num_entries:maximum number of completions to return
2510 * @wc:array of at least @num_entries &struct ib_wc where completions
2513 * Poll a CQ for (possibly multiple) completions. If the return value
2514 * is < 0, an error occurred. If the return value is >= 0, it is the
2515 * number of completions returned. If the return value is
2516 * non-negative and < num_entries, then the CQ was emptied.
2518 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2521 return cq->device->poll_cq(cq, num_entries, wc);
2525 * ib_peek_cq - Returns the number of unreaped completions currently
2526 * on the specified CQ.
2527 * @cq: The CQ to peek.
2528 * @wc_cnt: A minimum number of unreaped completions to check for.
2530 * If the number of unreaped completions is greater than or equal to wc_cnt,
2531 * this function returns wc_cnt, otherwise, it returns the actual number of
2532 * unreaped completions.
2534 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2537 * ib_req_notify_cq - Request completion notification on a CQ.
2538 * @cq: The CQ to generate an event for.
2540 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2541 * to request an event on the next solicited event or next work
2542 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2543 * may also be |ed in to request a hint about missed events, as
2547 * < 0 means an error occurred while requesting notification
2548 * == 0 means notification was requested successfully, and if
2549 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2550 * were missed and it is safe to wait for another event. In
2551 * this case is it guaranteed that any work completions added
2552 * to the CQ since the last CQ poll will trigger a completion
2553 * notification event.
2554 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2555 * in. It means that the consumer must poll the CQ again to
2556 * make sure it is empty to avoid missing an event because of a
2557 * race between requesting notification and an entry being
2558 * added to the CQ. This return value means it is possible
2559 * (but not guaranteed) that a work completion has been added
2560 * to the CQ since the last poll without triggering a
2561 * completion notification event.
2563 static inline int ib_req_notify_cq(struct ib_cq *cq,
2564 enum ib_cq_notify_flags flags)
2566 return cq->device->req_notify_cq(cq, flags);
2570 * ib_req_ncomp_notif - Request completion notification when there are
2571 * at least the specified number of unreaped completions on the CQ.
2572 * @cq: The CQ to generate an event for.
2573 * @wc_cnt: The number of unreaped completions that should be on the
2574 * CQ before an event is generated.
2576 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2578 return cq->device->req_ncomp_notif ?
2579 cq->device->req_ncomp_notif(cq, wc_cnt) :
2584 * ib_get_dma_mr - Returns a memory region for system memory that is
2586 * @pd: The protection domain associated with the memory region.
2587 * @mr_access_flags: Specifies the memory access rights.
2589 * Note that the ib_dma_*() functions defined below must be used
2590 * to create/destroy addresses used with the Lkey or Rkey returned
2591 * by ib_get_dma_mr().
2593 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2596 * ib_dma_mapping_error - check a DMA addr for error
2597 * @dev: The device for which the dma_addr was created
2598 * @dma_addr: The DMA address to check
2600 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2603 return dev->dma_ops->mapping_error(dev, dma_addr);
2604 return dma_mapping_error(dev->dma_device, dma_addr);
2608 * ib_dma_map_single - Map a kernel virtual address to DMA address
2609 * @dev: The device for which the dma_addr is to be created
2610 * @cpu_addr: The kernel virtual address
2611 * @size: The size of the region in bytes
2612 * @direction: The direction of the DMA
2614 static inline u64 ib_dma_map_single(struct ib_device *dev,
2615 void *cpu_addr, size_t size,
2616 enum dma_data_direction direction)
2619 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2620 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2624 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2625 * @dev: The device for which the DMA address was created
2626 * @addr: The DMA address
2627 * @size: The size of the region in bytes
2628 * @direction: The direction of the DMA
2630 static inline void ib_dma_unmap_single(struct ib_device *dev,
2631 u64 addr, size_t size,
2632 enum dma_data_direction direction)
2635 dev->dma_ops->unmap_single(dev, addr, size, direction);
2637 dma_unmap_single(dev->dma_device, addr, size, direction);
2640 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2641 void *cpu_addr, size_t size,
2642 enum dma_data_direction direction,
2643 struct dma_attrs *attrs)
2645 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2649 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2650 u64 addr, size_t size,
2651 enum dma_data_direction direction,
2652 struct dma_attrs *attrs)
2654 return dma_unmap_single_attrs(dev->dma_device, addr, size,
2659 * ib_dma_map_page - Map a physical page to DMA address
2660 * @dev: The device for which the dma_addr is to be created
2661 * @page: The page to be mapped
2662 * @offset: The offset within the page
2663 * @size: The size of the region in bytes
2664 * @direction: The direction of the DMA
2666 static inline u64 ib_dma_map_page(struct ib_device *dev,
2668 unsigned long offset,
2670 enum dma_data_direction direction)
2673 return dev->dma_ops->map_page(dev, page, offset, size, direction);
2674 return dma_map_page(dev->dma_device, page, offset, size, direction);
2678 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2679 * @dev: The device for which the DMA address was created
2680 * @addr: The DMA address
2681 * @size: The size of the region in bytes
2682 * @direction: The direction of the DMA
2684 static inline void ib_dma_unmap_page(struct ib_device *dev,
2685 u64 addr, size_t size,
2686 enum dma_data_direction direction)
2689 dev->dma_ops->unmap_page(dev, addr, size, direction);
2691 dma_unmap_page(dev->dma_device, addr, size, direction);
2695 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2696 * @dev: The device for which the DMA addresses are to be created
2697 * @sg: The array of scatter/gather entries
2698 * @nents: The number of scatter/gather entries
2699 * @direction: The direction of the DMA
2701 static inline int ib_dma_map_sg(struct ib_device *dev,
2702 struct scatterlist *sg, int nents,
2703 enum dma_data_direction direction)
2706 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2707 return dma_map_sg(dev->dma_device, sg, nents, direction);
2711 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2712 * @dev: The device for which the DMA addresses were created
2713 * @sg: The array of scatter/gather entries
2714 * @nents: The number of scatter/gather entries
2715 * @direction: The direction of the DMA
2717 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2718 struct scatterlist *sg, int nents,
2719 enum dma_data_direction direction)
2722 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2724 dma_unmap_sg(dev->dma_device, sg, nents, direction);
2727 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2728 struct scatterlist *sg, int nents,
2729 enum dma_data_direction direction,
2730 struct dma_attrs *attrs)
2732 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2735 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2736 struct scatterlist *sg, int nents,
2737 enum dma_data_direction direction,
2738 struct dma_attrs *attrs)
2740 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2743 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2744 * @dev: The device for which the DMA addresses were created
2745 * @sg: The scatter/gather entry
2747 * Note: this function is obsolete. To do: change all occurrences of
2748 * ib_sg_dma_address() into sg_dma_address().
2750 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2751 struct scatterlist *sg)
2753 return sg_dma_address(sg);
2757 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2758 * @dev: The device for which the DMA addresses were created
2759 * @sg: The scatter/gather entry
2761 * Note: this function is obsolete. To do: change all occurrences of
2762 * ib_sg_dma_len() into sg_dma_len().
2764 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2765 struct scatterlist *sg)
2767 return sg_dma_len(sg);
2771 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2772 * @dev: The device for which the DMA address was created
2773 * @addr: The DMA address
2774 * @size: The size of the region in bytes
2775 * @dir: The direction of the DMA
2777 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2780 enum dma_data_direction dir)
2783 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2785 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2789 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2790 * @dev: The device for which the DMA address was created
2791 * @addr: The DMA address
2792 * @size: The size of the region in bytes
2793 * @dir: The direction of the DMA
2795 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2798 enum dma_data_direction dir)
2801 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2803 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2807 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2808 * @dev: The device for which the DMA address is requested
2809 * @size: The size of the region to allocate in bytes
2810 * @dma_handle: A pointer for returning the DMA address of the region
2811 * @flag: memory allocator flags
2813 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2819 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2824 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2825 *dma_handle = handle;
2831 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2832 * @dev: The device for which the DMA addresses were allocated
2833 * @size: The size of the region
2834 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2835 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2837 static inline void ib_dma_free_coherent(struct ib_device *dev,
2838 size_t size, void *cpu_addr,
2842 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2844 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2848 * ib_query_mr - Retrieves information about a specific memory region.
2849 * @mr: The memory region to retrieve information about.
2850 * @mr_attr: The attributes of the specified memory region.
2852 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2855 * ib_dereg_mr - Deregisters a memory region and removes it from the
2856 * HCA translation table.
2857 * @mr: The memory region to deregister.
2859 * This function can fail, if the memory region has memory windows bound to it.
2861 int ib_dereg_mr(struct ib_mr *mr);
2863 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
2864 enum ib_mr_type mr_type,
2868 * ib_alloc_fast_reg_page_list - Allocates a page list array
2869 * @device - ib device pointer.
2870 * @page_list_len - size of the page list array to be allocated.
2872 * This allocates and returns a struct ib_fast_reg_page_list * and a
2873 * page_list array that is at least page_list_len in size. The actual
2874 * size is returned in max_page_list_len. The caller is responsible
2875 * for initializing the contents of the page_list array before posting
2876 * a send work request with the IB_WC_FAST_REG_MR opcode.
2878 * The page_list array entries must be translated using one of the
2879 * ib_dma_*() functions just like the addresses passed to
2880 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
2881 * ib_fast_reg_page_list must not be modified by the caller until the
2882 * IB_WC_FAST_REG_MR work request completes.
2884 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2885 struct ib_device *device, int page_list_len);
2888 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2890 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2892 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2895 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2897 * @mr - struct ib_mr pointer to be updated.
2898 * @newkey - new key to be used.
2900 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2902 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2903 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2907 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2908 * for calculating a new rkey for type 2 memory windows.
2909 * @rkey - the rkey to increment.
2911 static inline u32 ib_inc_rkey(u32 rkey)
2913 const u32 mask = 0x000000ff;
2914 return ((rkey + 1) & mask) | (rkey & ~mask);
2918 * ib_alloc_mw - Allocates a memory window.
2919 * @pd: The protection domain associated with the memory window.
2920 * @type: The type of the memory window (1 or 2).
2922 struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
2925 * ib_bind_mw - Posts a work request to the send queue of the specified
2926 * QP, which binds the memory window to the given address range and
2927 * remote access attributes.
2928 * @qp: QP to post the bind work request on.
2929 * @mw: The memory window to bind.
2930 * @mw_bind: Specifies information about the memory window, including
2931 * its address range, remote access rights, and associated memory region.
2933 * If there is no immediate error, the function will update the rkey member
2934 * of the mw parameter to its new value. The bind operation can still fail
2937 static inline int ib_bind_mw(struct ib_qp *qp,
2939 struct ib_mw_bind *mw_bind)
2941 /* XXX reference counting in corresponding MR? */
2942 return mw->device->bind_mw ?
2943 mw->device->bind_mw(qp, mw, mw_bind) :
2948 * ib_dealloc_mw - Deallocates a memory window.
2949 * @mw: The memory window to deallocate.
2951 int ib_dealloc_mw(struct ib_mw *mw);
2954 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2955 * @pd: The protection domain associated with the unmapped region.
2956 * @mr_access_flags: Specifies the memory access rights.
2957 * @fmr_attr: Attributes of the unmapped region.
2959 * A fast memory region must be mapped before it can be used as part of
2962 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2963 int mr_access_flags,
2964 struct ib_fmr_attr *fmr_attr);
2967 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2968 * @fmr: The fast memory region to associate with the pages.
2969 * @page_list: An array of physical pages to map to the fast memory region.
2970 * @list_len: The number of pages in page_list.
2971 * @iova: The I/O virtual address to use with the mapped region.
2973 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2974 u64 *page_list, int list_len,
2977 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2981 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2982 * @fmr_list: A linked list of fast memory regions to unmap.
2984 int ib_unmap_fmr(struct list_head *fmr_list);
2987 * ib_dealloc_fmr - Deallocates a fast memory region.
2988 * @fmr: The fast memory region to deallocate.
2990 int ib_dealloc_fmr(struct ib_fmr *fmr);
2993 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2994 * @qp: QP to attach to the multicast group. The QP must be type
2996 * @gid: Multicast group GID.
2997 * @lid: Multicast group LID in host byte order.
2999 * In order to send and receive multicast packets, subnet
3000 * administration must have created the multicast group and configured
3001 * the fabric appropriately. The port associated with the specified
3002 * QP must also be a member of the multicast group.
3004 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3007 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3008 * @qp: QP to detach from the multicast group.
3009 * @gid: Multicast group GID.
3010 * @lid: Multicast group LID in host byte order.
3012 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3015 * ib_alloc_xrcd - Allocates an XRC domain.
3016 * @device: The device on which to allocate the XRC domain.
3018 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3021 * ib_dealloc_xrcd - Deallocates an XRC domain.
3022 * @xrcd: The XRC domain to deallocate.
3024 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3026 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3027 struct ib_flow_attr *flow_attr, int domain);
3028 int ib_destroy_flow(struct ib_flow *flow_id);
3030 static inline int ib_check_mr_access(int flags)
3033 * Local write permission is required if remote write or
3034 * remote atomic permission is also requested.
3036 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3037 !(flags & IB_ACCESS_LOCAL_WRITE))
3044 * ib_check_mr_status: lightweight check of MR status.
3045 * This routine may provide status checks on a selected
3046 * ib_mr. first use is for signature status check.
3048 * @mr: A memory region.
3049 * @check_mask: Bitmask of which checks to perform from
3050 * ib_mr_status_check enumeration.
3051 * @mr_status: The container of relevant status checks.
3052 * failed checks will be indicated in the status bitmask
3053 * and the relevant info shall be in the error item.
3055 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3056 struct ib_mr_status *mr_status);
3058 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3059 u16 pkey, const union ib_gid *gid,
3060 const struct sockaddr *addr);
3062 #endif /* IB_VERBS_H */