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;
69 /* IB values map to NodeInfo:NodeType. */
78 enum rdma_transport_type {
82 RDMA_TRANSPORT_USNIC_UDP
85 enum rdma_protocol_type {
89 RDMA_PROTOCOL_USNIC_UDP
92 __attribute_const__ enum rdma_transport_type
93 rdma_node_get_transport(enum rdma_node_type node_type);
95 enum rdma_link_layer {
96 IB_LINK_LAYER_UNSPECIFIED,
97 IB_LINK_LAYER_INFINIBAND,
98 IB_LINK_LAYER_ETHERNET,
101 enum ib_device_cap_flags {
102 IB_DEVICE_RESIZE_MAX_WR = 1,
103 IB_DEVICE_BAD_PKEY_CNTR = (1<<1),
104 IB_DEVICE_BAD_QKEY_CNTR = (1<<2),
105 IB_DEVICE_RAW_MULTI = (1<<3),
106 IB_DEVICE_AUTO_PATH_MIG = (1<<4),
107 IB_DEVICE_CHANGE_PHY_PORT = (1<<5),
108 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6),
109 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7),
110 IB_DEVICE_SHUTDOWN_PORT = (1<<8),
111 IB_DEVICE_INIT_TYPE = (1<<9),
112 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10),
113 IB_DEVICE_SYS_IMAGE_GUID = (1<<11),
114 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12),
115 IB_DEVICE_SRQ_RESIZE = (1<<13),
116 IB_DEVICE_N_NOTIFY_CQ = (1<<14),
117 IB_DEVICE_LOCAL_DMA_LKEY = (1<<15),
118 IB_DEVICE_RESERVED = (1<<16), /* old SEND_W_INV */
119 IB_DEVICE_MEM_WINDOW = (1<<17),
121 * Devices should set IB_DEVICE_UD_IP_SUM if they support
122 * insertion of UDP and TCP checksum on outgoing UD IPoIB
123 * messages and can verify the validity of checksum for
124 * incoming messages. Setting this flag implies that the
125 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
127 IB_DEVICE_UD_IP_CSUM = (1<<18),
128 IB_DEVICE_UD_TSO = (1<<19),
129 IB_DEVICE_XRC = (1<<20),
130 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21),
131 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
132 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1<<23),
133 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1<<24),
134 IB_DEVICE_MANAGED_FLOW_STEERING = (1<<29),
135 IB_DEVICE_SIGNATURE_HANDOVER = (1<<30),
136 IB_DEVICE_ON_DEMAND_PAGING = (1<<31),
139 enum ib_signature_prot_cap {
140 IB_PROT_T10DIF_TYPE_1 = 1,
141 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
142 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
145 enum ib_signature_guard_cap {
146 IB_GUARD_T10DIF_CRC = 1,
147 IB_GUARD_T10DIF_CSUM = 1 << 1,
156 enum ib_odp_general_cap_bits {
157 IB_ODP_SUPPORT = 1 << 0,
160 enum ib_odp_transport_cap_bits {
161 IB_ODP_SUPPORT_SEND = 1 << 0,
162 IB_ODP_SUPPORT_RECV = 1 << 1,
163 IB_ODP_SUPPORT_WRITE = 1 << 2,
164 IB_ODP_SUPPORT_READ = 1 << 3,
165 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
169 uint64_t general_caps;
171 uint32_t rc_odp_caps;
172 uint32_t uc_odp_caps;
173 uint32_t ud_odp_caps;
174 } per_transport_caps;
177 enum ib_cq_creation_flags {
178 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
181 struct ib_cq_init_attr {
187 struct ib_device_attr {
189 __be64 sys_image_guid;
197 int device_cap_flags;
207 int max_qp_init_rd_atom;
208 int max_ee_init_rd_atom;
209 enum ib_atomic_cap atomic_cap;
210 enum ib_atomic_cap masked_atomic_cap;
217 int max_mcast_qp_attach;
218 int max_total_mcast_qp_attach;
225 unsigned int max_fast_reg_page_list_len;
227 u8 local_ca_ack_delay;
230 struct ib_odp_caps odp_caps;
231 uint64_t timestamp_mask;
232 uint64_t hca_core_clock; /* in KHZ */
243 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
246 case IB_MTU_256: return 256;
247 case IB_MTU_512: return 512;
248 case IB_MTU_1024: return 1024;
249 case IB_MTU_2048: return 2048;
250 case IB_MTU_4096: return 4096;
261 IB_PORT_ACTIVE_DEFER = 5
264 enum ib_port_cap_flags {
266 IB_PORT_NOTICE_SUP = 1 << 2,
267 IB_PORT_TRAP_SUP = 1 << 3,
268 IB_PORT_OPT_IPD_SUP = 1 << 4,
269 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
270 IB_PORT_SL_MAP_SUP = 1 << 6,
271 IB_PORT_MKEY_NVRAM = 1 << 7,
272 IB_PORT_PKEY_NVRAM = 1 << 8,
273 IB_PORT_LED_INFO_SUP = 1 << 9,
274 IB_PORT_SM_DISABLED = 1 << 10,
275 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
276 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
277 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
278 IB_PORT_CM_SUP = 1 << 16,
279 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
280 IB_PORT_REINIT_SUP = 1 << 18,
281 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
282 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
283 IB_PORT_DR_NOTICE_SUP = 1 << 21,
284 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
285 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
286 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
287 IB_PORT_CLIENT_REG_SUP = 1 << 25,
288 IB_PORT_IP_BASED_GIDS = 1 << 26
298 static inline int ib_width_enum_to_int(enum ib_port_width width)
301 case IB_WIDTH_1X: return 1;
302 case IB_WIDTH_4X: return 4;
303 case IB_WIDTH_8X: return 8;
304 case IB_WIDTH_12X: return 12;
318 struct ib_protocol_stats {
322 struct iw_protocol_stats {
325 u64 ipInTooBigErrors;
328 u64 ipInUnknownProtos;
329 u64 ipInTruncatedPkts;
332 u64 ipOutForwDatagrams;
364 union rdma_protocol_stats {
365 struct ib_protocol_stats ib;
366 struct iw_protocol_stats iw;
369 /* Define bits for the various functionality this port needs to be supported by
372 /* Management 0x00000FFF */
373 #define RDMA_CORE_CAP_IB_MAD 0x00000001
374 #define RDMA_CORE_CAP_IB_SMI 0x00000002
375 #define RDMA_CORE_CAP_IB_CM 0x00000004
376 #define RDMA_CORE_CAP_IW_CM 0x00000008
377 #define RDMA_CORE_CAP_IB_SA 0x00000010
378 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
380 /* Address format 0x000FF000 */
381 #define RDMA_CORE_CAP_AF_IB 0x00001000
382 #define RDMA_CORE_CAP_ETH_AH 0x00002000
384 /* Protocol 0xFFF00000 */
385 #define RDMA_CORE_CAP_PROT_IB 0x00100000
386 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
387 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
389 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
390 | RDMA_CORE_CAP_IB_MAD \
391 | RDMA_CORE_CAP_IB_SMI \
392 | RDMA_CORE_CAP_IB_CM \
393 | RDMA_CORE_CAP_IB_SA \
394 | RDMA_CORE_CAP_AF_IB)
395 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
396 | RDMA_CORE_CAP_IB_MAD \
397 | RDMA_CORE_CAP_IB_CM \
398 | RDMA_CORE_CAP_AF_IB \
399 | RDMA_CORE_CAP_ETH_AH)
400 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
401 | RDMA_CORE_CAP_IW_CM)
402 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
403 | RDMA_CORE_CAP_OPA_MAD)
405 struct ib_port_attr {
406 enum ib_port_state state;
408 enum ib_mtu active_mtu;
427 enum ib_device_modify_flags {
428 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
429 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
432 struct ib_device_modify {
437 enum ib_port_modify_flags {
438 IB_PORT_SHUTDOWN = 1,
439 IB_PORT_INIT_TYPE = (1<<2),
440 IB_PORT_RESET_QKEY_CNTR = (1<<3)
443 struct ib_port_modify {
444 u32 set_port_cap_mask;
445 u32 clr_port_cap_mask;
453 IB_EVENT_QP_ACCESS_ERR,
457 IB_EVENT_PATH_MIG_ERR,
458 IB_EVENT_DEVICE_FATAL,
459 IB_EVENT_PORT_ACTIVE,
462 IB_EVENT_PKEY_CHANGE,
465 IB_EVENT_SRQ_LIMIT_REACHED,
466 IB_EVENT_QP_LAST_WQE_REACHED,
467 IB_EVENT_CLIENT_REREGISTER,
471 __attribute_const__ const char *ib_event_msg(enum ib_event_type event);
474 struct ib_device *device;
481 enum ib_event_type event;
484 struct ib_event_handler {
485 struct ib_device *device;
486 void (*handler)(struct ib_event_handler *, struct ib_event *);
487 struct list_head list;
490 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
492 (_ptr)->device = _device; \
493 (_ptr)->handler = _handler; \
494 INIT_LIST_HEAD(&(_ptr)->list); \
497 struct ib_global_route {
506 __be32 version_tclass_flow;
515 IB_MULTICAST_QPN = 0xffffff
518 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
525 IB_RATE_PORT_CURRENT = 0,
526 IB_RATE_2_5_GBPS = 2,
534 IB_RATE_120_GBPS = 10,
535 IB_RATE_14_GBPS = 11,
536 IB_RATE_56_GBPS = 12,
537 IB_RATE_112_GBPS = 13,
538 IB_RATE_168_GBPS = 14,
539 IB_RATE_25_GBPS = 15,
540 IB_RATE_100_GBPS = 16,
541 IB_RATE_200_GBPS = 17,
542 IB_RATE_300_GBPS = 18
546 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
547 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
548 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
549 * @rate: rate to convert.
551 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
554 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
555 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
556 * @rate: rate to convert.
558 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
562 * enum ib_mr_type - memory region type
563 * @IB_MR_TYPE_MEM_REG: memory region that is used for
564 * normal registration
565 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
566 * signature operations (data-integrity
571 IB_MR_TYPE_SIGNATURE,
576 * IB_SIG_TYPE_NONE: Unprotected.
577 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
579 enum ib_signature_type {
585 * Signature T10-DIF block-guard types
586 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
587 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
589 enum ib_t10_dif_bg_type {
595 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
597 * @bg_type: T10-DIF block guard type (CRC|CSUM)
598 * @pi_interval: protection information interval.
599 * @bg: seed of guard computation.
600 * @app_tag: application tag of guard block
601 * @ref_tag: initial guard block reference tag.
602 * @ref_remap: Indicate wethear the reftag increments each block
603 * @app_escape: Indicate to skip block check if apptag=0xffff
604 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
605 * @apptag_check_mask: check bitmask of application tag.
607 struct ib_t10_dif_domain {
608 enum ib_t10_dif_bg_type bg_type;
616 u16 apptag_check_mask;
620 * struct ib_sig_domain - Parameters for signature domain
621 * @sig_type: specific signauture type
622 * @sig: union of all signature domain attributes that may
623 * be used to set domain layout.
625 struct ib_sig_domain {
626 enum ib_signature_type sig_type;
628 struct ib_t10_dif_domain dif;
633 * struct ib_sig_attrs - Parameters for signature handover operation
634 * @check_mask: bitmask for signature byte check (8 bytes)
635 * @mem: memory domain layout desciptor.
636 * @wire: wire domain layout desciptor.
638 struct ib_sig_attrs {
640 struct ib_sig_domain mem;
641 struct ib_sig_domain wire;
644 enum ib_sig_err_type {
651 * struct ib_sig_err - signature error descriptor
654 enum ib_sig_err_type err_type;
661 enum ib_mr_status_check {
662 IB_MR_CHECK_SIG_STATUS = 1,
666 * struct ib_mr_status - Memory region status container
668 * @fail_status: Bitmask of MR checks status. For each
669 * failed check a corresponding status bit is set.
670 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
673 struct ib_mr_status {
675 struct ib_sig_err sig_err;
679 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
681 * @mult: multiple to convert.
683 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
686 struct ib_global_route grh;
701 IB_WC_LOC_EEC_OP_ERR,
706 IB_WC_LOC_ACCESS_ERR,
707 IB_WC_REM_INV_REQ_ERR,
708 IB_WC_REM_ACCESS_ERR,
711 IB_WC_RNR_RETRY_EXC_ERR,
712 IB_WC_LOC_RDD_VIOL_ERR,
713 IB_WC_REM_INV_RD_REQ_ERR,
716 IB_WC_INV_EEC_STATE_ERR,
718 IB_WC_RESP_TIMEOUT_ERR,
722 __attribute_const__ const char *ib_wc_status_msg(enum ib_wc_status status);
734 IB_WC_MASKED_COMP_SWAP,
735 IB_WC_MASKED_FETCH_ADD,
737 * Set value of IB_WC_RECV so consumers can test if a completion is a
738 * receive by testing (opcode & IB_WC_RECV).
741 IB_WC_RECV_RDMA_WITH_IMM
746 IB_WC_WITH_IMM = (1<<1),
747 IB_WC_WITH_INVALIDATE = (1<<2),
748 IB_WC_IP_CSUM_OK = (1<<3),
749 IB_WC_WITH_SMAC = (1<<4),
750 IB_WC_WITH_VLAN = (1<<5),
755 enum ib_wc_status status;
756 enum ib_wc_opcode opcode;
770 u8 port_num; /* valid only for DR SMPs on switches */
775 enum ib_cq_notify_flags {
776 IB_CQ_SOLICITED = 1 << 0,
777 IB_CQ_NEXT_COMP = 1 << 1,
778 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
779 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
787 enum ib_srq_attr_mask {
788 IB_SRQ_MAX_WR = 1 << 0,
789 IB_SRQ_LIMIT = 1 << 1,
798 struct ib_srq_init_attr {
799 void (*event_handler)(struct ib_event *, void *);
801 struct ib_srq_attr attr;
802 enum ib_srq_type srq_type;
806 struct ib_xrcd *xrcd;
827 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
828 * here (and in that order) since the MAD layer uses them as
829 * indices into a 2-entry table.
838 IB_QPT_RAW_ETHERTYPE,
839 IB_QPT_RAW_PACKET = 8,
843 /* Reserve a range for qp types internal to the low level driver.
844 * These qp types will not be visible at the IB core layer, so the
845 * IB_QPT_MAX usages should not be affected in the core layer
847 IB_QPT_RESERVED1 = 0x1000,
859 enum ib_qp_create_flags {
860 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
861 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
862 IB_QP_CREATE_NETIF_QP = 1 << 5,
863 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
864 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
865 /* reserve bits 26-31 for low level drivers' internal use */
866 IB_QP_CREATE_RESERVED_START = 1 << 26,
867 IB_QP_CREATE_RESERVED_END = 1 << 31,
872 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
873 * callback to destroy the passed in QP.
876 struct ib_qp_init_attr {
877 void (*event_handler)(struct ib_event *, void *);
879 struct ib_cq *send_cq;
880 struct ib_cq *recv_cq;
882 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
883 struct ib_qp_cap cap;
884 enum ib_sig_type sq_sig_type;
885 enum ib_qp_type qp_type;
886 enum ib_qp_create_flags create_flags;
887 u8 port_num; /* special QP types only */
890 struct ib_qp_open_attr {
891 void (*event_handler)(struct ib_event *, void *);
894 enum ib_qp_type qp_type;
897 enum ib_rnr_timeout {
898 IB_RNR_TIMER_655_36 = 0,
899 IB_RNR_TIMER_000_01 = 1,
900 IB_RNR_TIMER_000_02 = 2,
901 IB_RNR_TIMER_000_03 = 3,
902 IB_RNR_TIMER_000_04 = 4,
903 IB_RNR_TIMER_000_06 = 5,
904 IB_RNR_TIMER_000_08 = 6,
905 IB_RNR_TIMER_000_12 = 7,
906 IB_RNR_TIMER_000_16 = 8,
907 IB_RNR_TIMER_000_24 = 9,
908 IB_RNR_TIMER_000_32 = 10,
909 IB_RNR_TIMER_000_48 = 11,
910 IB_RNR_TIMER_000_64 = 12,
911 IB_RNR_TIMER_000_96 = 13,
912 IB_RNR_TIMER_001_28 = 14,
913 IB_RNR_TIMER_001_92 = 15,
914 IB_RNR_TIMER_002_56 = 16,
915 IB_RNR_TIMER_003_84 = 17,
916 IB_RNR_TIMER_005_12 = 18,
917 IB_RNR_TIMER_007_68 = 19,
918 IB_RNR_TIMER_010_24 = 20,
919 IB_RNR_TIMER_015_36 = 21,
920 IB_RNR_TIMER_020_48 = 22,
921 IB_RNR_TIMER_030_72 = 23,
922 IB_RNR_TIMER_040_96 = 24,
923 IB_RNR_TIMER_061_44 = 25,
924 IB_RNR_TIMER_081_92 = 26,
925 IB_RNR_TIMER_122_88 = 27,
926 IB_RNR_TIMER_163_84 = 28,
927 IB_RNR_TIMER_245_76 = 29,
928 IB_RNR_TIMER_327_68 = 30,
929 IB_RNR_TIMER_491_52 = 31
932 enum ib_qp_attr_mask {
934 IB_QP_CUR_STATE = (1<<1),
935 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
936 IB_QP_ACCESS_FLAGS = (1<<3),
937 IB_QP_PKEY_INDEX = (1<<4),
941 IB_QP_PATH_MTU = (1<<8),
942 IB_QP_TIMEOUT = (1<<9),
943 IB_QP_RETRY_CNT = (1<<10),
944 IB_QP_RNR_RETRY = (1<<11),
945 IB_QP_RQ_PSN = (1<<12),
946 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
947 IB_QP_ALT_PATH = (1<<14),
948 IB_QP_MIN_RNR_TIMER = (1<<15),
949 IB_QP_SQ_PSN = (1<<16),
950 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
951 IB_QP_PATH_MIG_STATE = (1<<18),
953 IB_QP_DEST_QPN = (1<<20),
954 IB_QP_SMAC = (1<<21),
955 IB_QP_ALT_SMAC = (1<<22),
957 IB_QP_ALT_VID = (1<<24),
982 enum ib_qp_state qp_state;
983 enum ib_qp_state cur_qp_state;
984 enum ib_mtu path_mtu;
985 enum ib_mig_state path_mig_state;
991 struct ib_qp_cap cap;
992 struct ib_ah_attr ah_attr;
993 struct ib_ah_attr alt_ah_attr;
996 u8 en_sqd_async_notify;
999 u8 max_dest_rd_atomic;
1008 u8 alt_smac[ETH_ALEN];
1015 IB_WR_RDMA_WRITE_WITH_IMM,
1017 IB_WR_SEND_WITH_IMM,
1019 IB_WR_ATOMIC_CMP_AND_SWP,
1020 IB_WR_ATOMIC_FETCH_AND_ADD,
1022 IB_WR_SEND_WITH_INV,
1023 IB_WR_RDMA_READ_WITH_INV,
1026 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1027 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1030 /* reserve values for low level drivers' internal use.
1031 * These values will not be used at all in the ib core layer.
1033 IB_WR_RESERVED1 = 0xf0,
1045 enum ib_send_flags {
1047 IB_SEND_SIGNALED = (1<<1),
1048 IB_SEND_SOLICITED = (1<<2),
1049 IB_SEND_INLINE = (1<<3),
1050 IB_SEND_IP_CSUM = (1<<4),
1052 /* reserve bits 26-31 for low level drivers' internal use */
1053 IB_SEND_RESERVED_START = (1 << 26),
1054 IB_SEND_RESERVED_END = (1 << 31),
1063 struct ib_fast_reg_page_list {
1064 struct ib_device *device;
1066 unsigned int max_page_list_len;
1070 * struct ib_mw_bind_info - Parameters for a memory window bind operation.
1071 * @mr: A memory region to bind the memory window to.
1072 * @addr: The address where the memory window should begin.
1073 * @length: The length of the memory window, in bytes.
1074 * @mw_access_flags: Access flags from enum ib_access_flags for the window.
1076 * This struct contains the shared parameters for type 1 and type 2
1077 * memory window bind operations.
1079 struct ib_mw_bind_info {
1083 int mw_access_flags;
1087 struct ib_send_wr *next;
1089 struct ib_sge *sg_list;
1091 enum ib_wr_opcode opcode;
1095 u32 invalidate_rkey;
1106 u64 compare_add_mask;
1117 u16 pkey_index; /* valid for GSI only */
1118 u8 port_num; /* valid for DR SMPs on switch only */
1122 struct ib_fast_reg_page_list *page_list;
1123 unsigned int page_shift;
1124 unsigned int page_list_len;
1131 /* The new rkey for the memory window. */
1133 struct ib_mw_bind_info bind_info;
1136 struct ib_sig_attrs *sig_attrs;
1137 struct ib_mr *sig_mr;
1139 struct ib_sge *prot;
1142 u32 xrc_remote_srq_num; /* XRC TGT QPs only */
1146 struct ib_recv_wr *next;
1148 struct ib_sge *sg_list;
1152 enum ib_access_flags {
1153 IB_ACCESS_LOCAL_WRITE = 1,
1154 IB_ACCESS_REMOTE_WRITE = (1<<1),
1155 IB_ACCESS_REMOTE_READ = (1<<2),
1156 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1157 IB_ACCESS_MW_BIND = (1<<4),
1158 IB_ZERO_BASED = (1<<5),
1159 IB_ACCESS_ON_DEMAND = (1<<6),
1162 struct ib_phys_buf {
1169 u64 device_virt_addr;
1171 int mr_access_flags;
1176 enum ib_mr_rereg_flags {
1177 IB_MR_REREG_TRANS = 1,
1178 IB_MR_REREG_PD = (1<<1),
1179 IB_MR_REREG_ACCESS = (1<<2),
1180 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1184 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
1185 * @wr_id: Work request id.
1186 * @send_flags: Flags from ib_send_flags enum.
1187 * @bind_info: More parameters of the bind operation.
1192 struct ib_mw_bind_info bind_info;
1195 struct ib_fmr_attr {
1203 struct ib_ucontext {
1204 struct ib_device *device;
1205 struct list_head pd_list;
1206 struct list_head mr_list;
1207 struct list_head mw_list;
1208 struct list_head cq_list;
1209 struct list_head qp_list;
1210 struct list_head srq_list;
1211 struct list_head ah_list;
1212 struct list_head xrcd_list;
1213 struct list_head rule_list;
1217 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1218 struct rb_root umem_tree;
1220 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1221 * mmu notifiers registration.
1223 struct rw_semaphore umem_rwsem;
1224 void (*invalidate_range)(struct ib_umem *umem,
1225 unsigned long start, unsigned long end);
1227 struct mmu_notifier mn;
1228 atomic_t notifier_count;
1229 /* A list of umems that don't have private mmu notifier counters yet. */
1230 struct list_head no_private_counters;
1236 u64 user_handle; /* handle given to us by userspace */
1237 struct ib_ucontext *context; /* associated user context */
1238 void *object; /* containing object */
1239 struct list_head list; /* link to context's list */
1240 int id; /* index into kernel idr */
1242 struct rw_semaphore mutex; /* protects .live */
1247 const void __user *inbuf;
1248 void __user *outbuf;
1254 struct ib_device *device;
1255 struct ib_uobject *uobject;
1256 atomic_t usecnt; /* count all resources */
1260 struct ib_device *device;
1261 atomic_t usecnt; /* count all exposed resources */
1262 struct inode *inode;
1264 struct mutex tgt_qp_mutex;
1265 struct list_head tgt_qp_list;
1269 struct ib_device *device;
1271 struct ib_uobject *uobject;
1274 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1277 struct ib_device *device;
1278 struct ib_uobject *uobject;
1279 ib_comp_handler comp_handler;
1280 void (*event_handler)(struct ib_event *, void *);
1283 atomic_t usecnt; /* count number of work queues */
1287 struct ib_device *device;
1289 struct ib_uobject *uobject;
1290 void (*event_handler)(struct ib_event *, void *);
1292 enum ib_srq_type srq_type;
1297 struct ib_xrcd *xrcd;
1305 struct ib_device *device;
1307 struct ib_cq *send_cq;
1308 struct ib_cq *recv_cq;
1310 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1311 struct list_head xrcd_list;
1312 /* count times opened, mcast attaches, flow attaches */
1314 struct list_head open_list;
1315 struct ib_qp *real_qp;
1316 struct ib_uobject *uobject;
1317 void (*event_handler)(struct ib_event *, void *);
1320 enum ib_qp_type qp_type;
1324 struct ib_device *device;
1326 struct ib_uobject *uobject;
1329 atomic_t usecnt; /* count number of MWs */
1333 struct ib_device *device;
1335 struct ib_uobject *uobject;
1337 enum ib_mw_type type;
1341 struct ib_device *device;
1343 struct list_head list;
1348 /* Supported steering options */
1349 enum ib_flow_attr_type {
1350 /* steering according to rule specifications */
1351 IB_FLOW_ATTR_NORMAL = 0x0,
1352 /* default unicast and multicast rule -
1353 * receive all Eth traffic which isn't steered to any QP
1355 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1356 /* default multicast rule -
1357 * receive all Eth multicast traffic which isn't steered to any QP
1359 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1360 /* sniffer rule - receive all port traffic */
1361 IB_FLOW_ATTR_SNIFFER = 0x3
1364 /* Supported steering header types */
1365 enum ib_flow_spec_type {
1367 IB_FLOW_SPEC_ETH = 0x20,
1368 IB_FLOW_SPEC_IB = 0x22,
1370 IB_FLOW_SPEC_IPV4 = 0x30,
1372 IB_FLOW_SPEC_TCP = 0x40,
1373 IB_FLOW_SPEC_UDP = 0x41
1375 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1376 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1378 /* Flow steering rule priority is set according to it's domain.
1379 * Lower domain value means higher priority.
1381 enum ib_flow_domain {
1382 IB_FLOW_DOMAIN_USER,
1383 IB_FLOW_DOMAIN_ETHTOOL,
1386 IB_FLOW_DOMAIN_NUM /* Must be last */
1389 struct ib_flow_eth_filter {
1396 struct ib_flow_spec_eth {
1397 enum ib_flow_spec_type type;
1399 struct ib_flow_eth_filter val;
1400 struct ib_flow_eth_filter mask;
1403 struct ib_flow_ib_filter {
1408 struct ib_flow_spec_ib {
1409 enum ib_flow_spec_type type;
1411 struct ib_flow_ib_filter val;
1412 struct ib_flow_ib_filter mask;
1415 struct ib_flow_ipv4_filter {
1420 struct ib_flow_spec_ipv4 {
1421 enum ib_flow_spec_type type;
1423 struct ib_flow_ipv4_filter val;
1424 struct ib_flow_ipv4_filter mask;
1427 struct ib_flow_tcp_udp_filter {
1432 struct ib_flow_spec_tcp_udp {
1433 enum ib_flow_spec_type type;
1435 struct ib_flow_tcp_udp_filter val;
1436 struct ib_flow_tcp_udp_filter mask;
1439 union ib_flow_spec {
1441 enum ib_flow_spec_type type;
1444 struct ib_flow_spec_eth eth;
1445 struct ib_flow_spec_ib ib;
1446 struct ib_flow_spec_ipv4 ipv4;
1447 struct ib_flow_spec_tcp_udp tcp_udp;
1450 struct ib_flow_attr {
1451 enum ib_flow_attr_type type;
1457 /* Following are the optional layers according to user request
1458 * struct ib_flow_spec_xxx
1459 * struct ib_flow_spec_yyy
1465 struct ib_uobject *uobject;
1471 enum ib_process_mad_flags {
1472 IB_MAD_IGNORE_MKEY = 1,
1473 IB_MAD_IGNORE_BKEY = 2,
1474 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1477 enum ib_mad_result {
1478 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1479 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1480 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1481 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1484 #define IB_DEVICE_NAME_MAX 64
1488 struct ib_event_handler event_handler;
1489 struct ib_pkey_cache **pkey_cache;
1490 struct ib_gid_cache **gid_cache;
1494 struct ib_dma_mapping_ops {
1495 int (*mapping_error)(struct ib_device *dev,
1497 u64 (*map_single)(struct ib_device *dev,
1498 void *ptr, size_t size,
1499 enum dma_data_direction direction);
1500 void (*unmap_single)(struct ib_device *dev,
1501 u64 addr, size_t size,
1502 enum dma_data_direction direction);
1503 u64 (*map_page)(struct ib_device *dev,
1504 struct page *page, unsigned long offset,
1506 enum dma_data_direction direction);
1507 void (*unmap_page)(struct ib_device *dev,
1508 u64 addr, size_t size,
1509 enum dma_data_direction direction);
1510 int (*map_sg)(struct ib_device *dev,
1511 struct scatterlist *sg, int nents,
1512 enum dma_data_direction direction);
1513 void (*unmap_sg)(struct ib_device *dev,
1514 struct scatterlist *sg, int nents,
1515 enum dma_data_direction direction);
1516 void (*sync_single_for_cpu)(struct ib_device *dev,
1519 enum dma_data_direction dir);
1520 void (*sync_single_for_device)(struct ib_device *dev,
1523 enum dma_data_direction dir);
1524 void *(*alloc_coherent)(struct ib_device *dev,
1528 void (*free_coherent)(struct ib_device *dev,
1529 size_t size, void *cpu_addr,
1535 struct ib_port_immutable {
1543 struct device *dma_device;
1545 char name[IB_DEVICE_NAME_MAX];
1547 struct list_head event_handler_list;
1548 spinlock_t event_handler_lock;
1550 spinlock_t client_data_lock;
1551 struct list_head core_list;
1552 /* Access to the client_data_list is protected by the client_data_lock
1553 * spinlock and the lists_rwsem read-write semaphore */
1554 struct list_head client_data_list;
1556 struct ib_cache cache;
1558 * port_immutable is indexed by port number
1560 struct ib_port_immutable *port_immutable;
1562 int num_comp_vectors;
1564 struct iw_cm_verbs *iwcm;
1566 int (*get_protocol_stats)(struct ib_device *device,
1567 union rdma_protocol_stats *stats);
1568 int (*query_device)(struct ib_device *device,
1569 struct ib_device_attr *device_attr,
1570 struct ib_udata *udata);
1571 int (*query_port)(struct ib_device *device,
1573 struct ib_port_attr *port_attr);
1574 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1576 int (*query_gid)(struct ib_device *device,
1577 u8 port_num, int index,
1579 int (*query_pkey)(struct ib_device *device,
1580 u8 port_num, u16 index, u16 *pkey);
1581 int (*modify_device)(struct ib_device *device,
1582 int device_modify_mask,
1583 struct ib_device_modify *device_modify);
1584 int (*modify_port)(struct ib_device *device,
1585 u8 port_num, int port_modify_mask,
1586 struct ib_port_modify *port_modify);
1587 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1588 struct ib_udata *udata);
1589 int (*dealloc_ucontext)(struct ib_ucontext *context);
1590 int (*mmap)(struct ib_ucontext *context,
1591 struct vm_area_struct *vma);
1592 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1593 struct ib_ucontext *context,
1594 struct ib_udata *udata);
1595 int (*dealloc_pd)(struct ib_pd *pd);
1596 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1597 struct ib_ah_attr *ah_attr);
1598 int (*modify_ah)(struct ib_ah *ah,
1599 struct ib_ah_attr *ah_attr);
1600 int (*query_ah)(struct ib_ah *ah,
1601 struct ib_ah_attr *ah_attr);
1602 int (*destroy_ah)(struct ib_ah *ah);
1603 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1604 struct ib_srq_init_attr *srq_init_attr,
1605 struct ib_udata *udata);
1606 int (*modify_srq)(struct ib_srq *srq,
1607 struct ib_srq_attr *srq_attr,
1608 enum ib_srq_attr_mask srq_attr_mask,
1609 struct ib_udata *udata);
1610 int (*query_srq)(struct ib_srq *srq,
1611 struct ib_srq_attr *srq_attr);
1612 int (*destroy_srq)(struct ib_srq *srq);
1613 int (*post_srq_recv)(struct ib_srq *srq,
1614 struct ib_recv_wr *recv_wr,
1615 struct ib_recv_wr **bad_recv_wr);
1616 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1617 struct ib_qp_init_attr *qp_init_attr,
1618 struct ib_udata *udata);
1619 int (*modify_qp)(struct ib_qp *qp,
1620 struct ib_qp_attr *qp_attr,
1622 struct ib_udata *udata);
1623 int (*query_qp)(struct ib_qp *qp,
1624 struct ib_qp_attr *qp_attr,
1626 struct ib_qp_init_attr *qp_init_attr);
1627 int (*destroy_qp)(struct ib_qp *qp);
1628 int (*post_send)(struct ib_qp *qp,
1629 struct ib_send_wr *send_wr,
1630 struct ib_send_wr **bad_send_wr);
1631 int (*post_recv)(struct ib_qp *qp,
1632 struct ib_recv_wr *recv_wr,
1633 struct ib_recv_wr **bad_recv_wr);
1634 struct ib_cq * (*create_cq)(struct ib_device *device,
1635 const struct ib_cq_init_attr *attr,
1636 struct ib_ucontext *context,
1637 struct ib_udata *udata);
1638 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1640 int (*destroy_cq)(struct ib_cq *cq);
1641 int (*resize_cq)(struct ib_cq *cq, int cqe,
1642 struct ib_udata *udata);
1643 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1645 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1646 int (*req_notify_cq)(struct ib_cq *cq,
1647 enum ib_cq_notify_flags flags);
1648 int (*req_ncomp_notif)(struct ib_cq *cq,
1650 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1651 int mr_access_flags);
1652 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
1653 struct ib_phys_buf *phys_buf_array,
1655 int mr_access_flags,
1657 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1658 u64 start, u64 length,
1660 int mr_access_flags,
1661 struct ib_udata *udata);
1662 int (*rereg_user_mr)(struct ib_mr *mr,
1664 u64 start, u64 length,
1666 int mr_access_flags,
1668 struct ib_udata *udata);
1669 int (*query_mr)(struct ib_mr *mr,
1670 struct ib_mr_attr *mr_attr);
1671 int (*dereg_mr)(struct ib_mr *mr);
1672 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
1673 enum ib_mr_type mr_type,
1675 struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1677 void (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1678 int (*rereg_phys_mr)(struct ib_mr *mr,
1681 struct ib_phys_buf *phys_buf_array,
1683 int mr_access_flags,
1685 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1686 enum ib_mw_type type);
1687 int (*bind_mw)(struct ib_qp *qp,
1689 struct ib_mw_bind *mw_bind);
1690 int (*dealloc_mw)(struct ib_mw *mw);
1691 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1692 int mr_access_flags,
1693 struct ib_fmr_attr *fmr_attr);
1694 int (*map_phys_fmr)(struct ib_fmr *fmr,
1695 u64 *page_list, int list_len,
1697 int (*unmap_fmr)(struct list_head *fmr_list);
1698 int (*dealloc_fmr)(struct ib_fmr *fmr);
1699 int (*attach_mcast)(struct ib_qp *qp,
1702 int (*detach_mcast)(struct ib_qp *qp,
1705 int (*process_mad)(struct ib_device *device,
1706 int process_mad_flags,
1708 const struct ib_wc *in_wc,
1709 const struct ib_grh *in_grh,
1710 const struct ib_mad_hdr *in_mad,
1712 struct ib_mad_hdr *out_mad,
1713 size_t *out_mad_size,
1714 u16 *out_mad_pkey_index);
1715 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1716 struct ib_ucontext *ucontext,
1717 struct ib_udata *udata);
1718 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1719 struct ib_flow * (*create_flow)(struct ib_qp *qp,
1723 int (*destroy_flow)(struct ib_flow *flow_id);
1724 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1725 struct ib_mr_status *mr_status);
1727 struct ib_dma_mapping_ops *dma_ops;
1729 struct module *owner;
1731 struct kobject *ports_parent;
1732 struct list_head port_list;
1735 IB_DEV_UNINITIALIZED,
1741 u64 uverbs_cmd_mask;
1742 u64 uverbs_ex_cmd_mask;
1752 * The following mandatory functions are used only at device
1753 * registration. Keep functions such as these at the end of this
1754 * structure to avoid cache line misses when accessing struct ib_device
1757 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1762 void (*add) (struct ib_device *);
1763 void (*remove)(struct ib_device *, void *client_data);
1765 /* Returns the net_dev belonging to this ib_client and matching the
1767 * @dev: An RDMA device that the net_dev use for communication.
1768 * @port: A physical port number on the RDMA device.
1769 * @pkey: P_Key that the net_dev uses if applicable.
1770 * @gid: A GID that the net_dev uses to communicate.
1771 * @addr: An IP address the net_dev is configured with.
1772 * @client_data: The device's client data set by ib_set_client_data().
1774 * An ib_client that implements a net_dev on top of RDMA devices
1775 * (such as IP over IB) should implement this callback, allowing the
1776 * rdma_cm module to find the right net_dev for a given request.
1778 * The caller is responsible for calling dev_put on the returned
1780 struct net_device *(*get_net_dev_by_params)(
1781 struct ib_device *dev,
1784 const union ib_gid *gid,
1785 const struct sockaddr *addr,
1787 struct list_head list;
1790 struct ib_device *ib_alloc_device(size_t size);
1791 void ib_dealloc_device(struct ib_device *device);
1793 int ib_register_device(struct ib_device *device,
1794 int (*port_callback)(struct ib_device *,
1795 u8, struct kobject *));
1796 void ib_unregister_device(struct ib_device *device);
1798 int ib_register_client (struct ib_client *client);
1799 void ib_unregister_client(struct ib_client *client);
1801 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1802 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1805 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1807 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1810 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1812 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1816 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1817 * contains all required attributes and no attributes not allowed for
1818 * the given QP state transition.
1819 * @cur_state: Current QP state
1820 * @next_state: Next QP state
1822 * @mask: Mask of supplied QP attributes
1823 * @ll : link layer of port
1825 * This function is a helper function that a low-level driver's
1826 * modify_qp method can use to validate the consumer's input. It
1827 * checks that cur_state and next_state are valid QP states, that a
1828 * transition from cur_state to next_state is allowed by the IB spec,
1829 * and that the attribute mask supplied is allowed for the transition.
1831 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1832 enum ib_qp_type type, enum ib_qp_attr_mask mask,
1833 enum rdma_link_layer ll);
1835 int ib_register_event_handler (struct ib_event_handler *event_handler);
1836 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1837 void ib_dispatch_event(struct ib_event *event);
1839 int ib_query_device(struct ib_device *device,
1840 struct ib_device_attr *device_attr);
1842 int ib_query_port(struct ib_device *device,
1843 u8 port_num, struct ib_port_attr *port_attr);
1845 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1849 * rdma_cap_ib_switch - Check if the device is IB switch
1850 * @device: Device to check
1852 * Device driver is responsible for setting is_switch bit on
1853 * in ib_device structure at init time.
1855 * Return: true if the device is IB switch.
1857 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
1859 return device->is_switch;
1863 * rdma_start_port - Return the first valid port number for the device
1866 * @device: Device to be checked
1868 * Return start port number
1870 static inline u8 rdma_start_port(const struct ib_device *device)
1872 return rdma_cap_ib_switch(device) ? 0 : 1;
1876 * rdma_end_port - Return the last valid port number for the device
1879 * @device: Device to be checked
1881 * Return last port number
1883 static inline u8 rdma_end_port(const struct ib_device *device)
1885 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
1888 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
1890 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
1893 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
1895 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
1898 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
1900 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
1903 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
1905 return device->port_immutable[port_num].core_cap_flags &
1906 (RDMA_CORE_CAP_PROT_IB | RDMA_CORE_CAP_PROT_ROCE);
1910 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
1911 * Management Datagrams.
1912 * @device: Device to check
1913 * @port_num: Port number to check
1915 * Management Datagrams (MAD) are a required part of the InfiniBand
1916 * specification and are supported on all InfiniBand devices. A slightly
1917 * extended version are also supported on OPA interfaces.
1919 * Return: true if the port supports sending/receiving of MAD packets.
1921 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
1923 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
1927 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
1928 * Management Datagrams.
1929 * @device: Device to check
1930 * @port_num: Port number to check
1932 * Intel OmniPath devices extend and/or replace the InfiniBand Management
1933 * datagrams with their own versions. These OPA MADs share many but not all of
1934 * the characteristics of InfiniBand MADs.
1936 * OPA MADs differ in the following ways:
1938 * 1) MADs are variable size up to 2K
1939 * IBTA defined MADs remain fixed at 256 bytes
1940 * 2) OPA SMPs must carry valid PKeys
1941 * 3) OPA SMP packets are a different format
1943 * Return: true if the port supports OPA MAD packet formats.
1945 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
1947 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
1948 == RDMA_CORE_CAP_OPA_MAD;
1952 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
1953 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
1954 * @device: Device to check
1955 * @port_num: Port number to check
1957 * Each InfiniBand node is required to provide a Subnet Management Agent
1958 * that the subnet manager can access. Prior to the fabric being fully
1959 * configured by the subnet manager, the SMA is accessed via a well known
1960 * interface called the Subnet Management Interface (SMI). This interface
1961 * uses directed route packets to communicate with the SM to get around the
1962 * chicken and egg problem of the SM needing to know what's on the fabric
1963 * in order to configure the fabric, and needing to configure the fabric in
1964 * order to send packets to the devices on the fabric. These directed
1965 * route packets do not need the fabric fully configured in order to reach
1966 * their destination. The SMI is the only method allowed to send
1967 * directed route packets on an InfiniBand fabric.
1969 * Return: true if the port provides an SMI.
1971 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
1973 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
1977 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
1978 * Communication Manager.
1979 * @device: Device to check
1980 * @port_num: Port number to check
1982 * The InfiniBand Communication Manager is one of many pre-defined General
1983 * Service Agents (GSA) that are accessed via the General Service
1984 * Interface (GSI). It's role is to facilitate establishment of connections
1985 * between nodes as well as other management related tasks for established
1988 * Return: true if the port supports an IB CM (this does not guarantee that
1989 * a CM is actually running however).
1991 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
1993 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
1997 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
1998 * Communication Manager.
1999 * @device: Device to check
2000 * @port_num: Port number to check
2002 * Similar to above, but specific to iWARP connections which have a different
2003 * managment protocol than InfiniBand.
2005 * Return: true if the port supports an iWARP CM (this does not guarantee that
2006 * a CM is actually running however).
2008 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2010 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2014 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2015 * Subnet Administration.
2016 * @device: Device to check
2017 * @port_num: Port number to check
2019 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2020 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2021 * fabrics, devices should resolve routes to other hosts by contacting the
2022 * SA to query the proper route.
2024 * Return: true if the port should act as a client to the fabric Subnet
2025 * Administration interface. This does not imply that the SA service is
2028 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2030 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2034 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2036 * @device: Device to check
2037 * @port_num: Port number to check
2039 * InfiniBand multicast registration is more complex than normal IPv4 or
2040 * IPv6 multicast registration. Each Host Channel Adapter must register
2041 * with the Subnet Manager when it wishes to join a multicast group. It
2042 * should do so only once regardless of how many queue pairs it subscribes
2043 * to this group. And it should leave the group only after all queue pairs
2044 * attached to the group have been detached.
2046 * Return: true if the port must undertake the additional adminstrative
2047 * overhead of registering/unregistering with the SM and tracking of the
2048 * total number of queue pairs attached to the multicast group.
2050 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2052 return rdma_cap_ib_sa(device, port_num);
2056 * rdma_cap_af_ib - Check if the port of device has the capability
2057 * Native Infiniband Address.
2058 * @device: Device to check
2059 * @port_num: Port number to check
2061 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2062 * GID. RoCE uses a different mechanism, but still generates a GID via
2063 * a prescribed mechanism and port specific data.
2065 * Return: true if the port uses a GID address to identify devices on the
2068 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2070 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2074 * rdma_cap_eth_ah - Check if the port of device has the capability
2075 * Ethernet Address Handle.
2076 * @device: Device to check
2077 * @port_num: Port number to check
2079 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2080 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2081 * port. Normally, packet headers are generated by the sending host
2082 * adapter, but when sending connectionless datagrams, we must manually
2083 * inject the proper headers for the fabric we are communicating over.
2085 * Return: true if we are running as a RoCE port and must force the
2086 * addition of a Global Route Header built from our Ethernet Address
2087 * Handle into our header list for connectionless packets.
2089 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2091 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2095 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2098 * @port_num: Port number
2100 * This MAD size includes the MAD headers and MAD payload. No other headers
2103 * Return the max MAD size required by the Port. Will return 0 if the port
2104 * does not support MADs
2106 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2108 return device->port_immutable[port_num].max_mad_size;
2111 int ib_query_gid(struct ib_device *device,
2112 u8 port_num, int index, union ib_gid *gid);
2114 int ib_query_pkey(struct ib_device *device,
2115 u8 port_num, u16 index, u16 *pkey);
2117 int ib_modify_device(struct ib_device *device,
2118 int device_modify_mask,
2119 struct ib_device_modify *device_modify);
2121 int ib_modify_port(struct ib_device *device,
2122 u8 port_num, int port_modify_mask,
2123 struct ib_port_modify *port_modify);
2125 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2126 u8 *port_num, u16 *index);
2128 int ib_find_pkey(struct ib_device *device,
2129 u8 port_num, u16 pkey, u16 *index);
2132 * ib_alloc_pd - Allocates an unused protection domain.
2133 * @device: The device on which to allocate the protection domain.
2135 * A protection domain object provides an association between QPs, shared
2136 * receive queues, address handles, memory regions, and memory windows.
2138 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2141 * ib_dealloc_pd - Deallocates a protection domain.
2142 * @pd: The protection domain to deallocate.
2144 int ib_dealloc_pd(struct ib_pd *pd);
2147 * ib_create_ah - Creates an address handle for the given address vector.
2148 * @pd: The protection domain associated with the address handle.
2149 * @ah_attr: The attributes of the address vector.
2151 * The address handle is used to reference a local or global destination
2152 * in all UD QP post sends.
2154 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2157 * ib_init_ah_from_wc - Initializes address handle attributes from a
2159 * @device: Device on which the received message arrived.
2160 * @port_num: Port on which the received message arrived.
2161 * @wc: Work completion associated with the received message.
2162 * @grh: References the received global route header. This parameter is
2163 * ignored unless the work completion indicates that the GRH is valid.
2164 * @ah_attr: Returned attributes that can be used when creating an address
2165 * handle for replying to the message.
2167 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2168 const struct ib_wc *wc, const struct ib_grh *grh,
2169 struct ib_ah_attr *ah_attr);
2172 * ib_create_ah_from_wc - Creates an address handle associated with the
2173 * sender of the specified work completion.
2174 * @pd: The protection domain associated with the address handle.
2175 * @wc: Work completion information associated with a received message.
2176 * @grh: References the received global route header. This parameter is
2177 * ignored unless the work completion indicates that the GRH is valid.
2178 * @port_num: The outbound port number to associate with the address.
2180 * The address handle is used to reference a local or global destination
2181 * in all UD QP post sends.
2183 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2184 const struct ib_grh *grh, u8 port_num);
2187 * ib_modify_ah - Modifies the address vector associated with an address
2189 * @ah: The address handle to modify.
2190 * @ah_attr: The new address vector attributes to associate with the
2193 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2196 * ib_query_ah - Queries the address vector associated with an address
2198 * @ah: The address handle to query.
2199 * @ah_attr: The address vector attributes associated with the address
2202 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2205 * ib_destroy_ah - Destroys an address handle.
2206 * @ah: The address handle to destroy.
2208 int ib_destroy_ah(struct ib_ah *ah);
2211 * ib_create_srq - Creates a SRQ associated with the specified protection
2213 * @pd: The protection domain associated with the SRQ.
2214 * @srq_init_attr: A list of initial attributes required to create the
2215 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2216 * the actual capabilities of the created SRQ.
2218 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2219 * requested size of the SRQ, and set to the actual values allocated
2220 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2221 * will always be at least as large as the requested values.
2223 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2224 struct ib_srq_init_attr *srq_init_attr);
2227 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2228 * @srq: The SRQ to modify.
2229 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2230 * the current values of selected SRQ attributes are returned.
2231 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2232 * are being modified.
2234 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2235 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2236 * the number of receives queued drops below the limit.
2238 int ib_modify_srq(struct ib_srq *srq,
2239 struct ib_srq_attr *srq_attr,
2240 enum ib_srq_attr_mask srq_attr_mask);
2243 * ib_query_srq - Returns the attribute list and current values for the
2245 * @srq: The SRQ to query.
2246 * @srq_attr: The attributes of the specified SRQ.
2248 int ib_query_srq(struct ib_srq *srq,
2249 struct ib_srq_attr *srq_attr);
2252 * ib_destroy_srq - Destroys the specified SRQ.
2253 * @srq: The SRQ to destroy.
2255 int ib_destroy_srq(struct ib_srq *srq);
2258 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2259 * @srq: The SRQ to post the work request on.
2260 * @recv_wr: A list of work requests to post on the receive queue.
2261 * @bad_recv_wr: On an immediate failure, this parameter will reference
2262 * the work request that failed to be posted on the QP.
2264 static inline int ib_post_srq_recv(struct ib_srq *srq,
2265 struct ib_recv_wr *recv_wr,
2266 struct ib_recv_wr **bad_recv_wr)
2268 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2272 * ib_create_qp - Creates a QP associated with the specified protection
2274 * @pd: The protection domain associated with the QP.
2275 * @qp_init_attr: A list of initial attributes required to create the
2276 * QP. If QP creation succeeds, then the attributes are updated to
2277 * the actual capabilities of the created QP.
2279 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2280 struct ib_qp_init_attr *qp_init_attr);
2283 * ib_modify_qp - Modifies the attributes for the specified QP and then
2284 * transitions the QP to the given state.
2285 * @qp: The QP to modify.
2286 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2287 * the current values of selected QP attributes are returned.
2288 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2289 * are being modified.
2291 int ib_modify_qp(struct ib_qp *qp,
2292 struct ib_qp_attr *qp_attr,
2296 * ib_query_qp - Returns the attribute list and current values for the
2298 * @qp: The QP to query.
2299 * @qp_attr: The attributes of the specified QP.
2300 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2301 * @qp_init_attr: Additional attributes of the selected QP.
2303 * The qp_attr_mask may be used to limit the query to gathering only the
2304 * selected attributes.
2306 int ib_query_qp(struct ib_qp *qp,
2307 struct ib_qp_attr *qp_attr,
2309 struct ib_qp_init_attr *qp_init_attr);
2312 * ib_destroy_qp - Destroys the specified QP.
2313 * @qp: The QP to destroy.
2315 int ib_destroy_qp(struct ib_qp *qp);
2318 * ib_open_qp - Obtain a reference to an existing sharable QP.
2319 * @xrcd - XRC domain
2320 * @qp_open_attr: Attributes identifying the QP to open.
2322 * Returns a reference to a sharable QP.
2324 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2325 struct ib_qp_open_attr *qp_open_attr);
2328 * ib_close_qp - Release an external reference to a QP.
2329 * @qp: The QP handle to release
2331 * The opened QP handle is released by the caller. The underlying
2332 * shared QP is not destroyed until all internal references are released.
2334 int ib_close_qp(struct ib_qp *qp);
2337 * ib_post_send - Posts a list of work requests to the send queue of
2339 * @qp: The QP to post the work request on.
2340 * @send_wr: A list of work requests to post on the send queue.
2341 * @bad_send_wr: On an immediate failure, this parameter will reference
2342 * the work request that failed to be posted on the QP.
2344 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2345 * error is returned, the QP state shall not be affected,
2346 * ib_post_send() will return an immediate error after queueing any
2347 * earlier work requests in the list.
2349 static inline int ib_post_send(struct ib_qp *qp,
2350 struct ib_send_wr *send_wr,
2351 struct ib_send_wr **bad_send_wr)
2353 return qp->device->post_send(qp, send_wr, bad_send_wr);
2357 * ib_post_recv - Posts a list of work requests to the receive queue of
2359 * @qp: The QP to post the work request on.
2360 * @recv_wr: A list of work requests to post on the receive queue.
2361 * @bad_recv_wr: On an immediate failure, this parameter will reference
2362 * the work request that failed to be posted on the QP.
2364 static inline int ib_post_recv(struct ib_qp *qp,
2365 struct ib_recv_wr *recv_wr,
2366 struct ib_recv_wr **bad_recv_wr)
2368 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2372 * ib_create_cq - Creates a CQ on the specified device.
2373 * @device: The device on which to create the CQ.
2374 * @comp_handler: A user-specified callback that is invoked when a
2375 * completion event occurs on the CQ.
2376 * @event_handler: A user-specified callback that is invoked when an
2377 * asynchronous event not associated with a completion occurs on the CQ.
2378 * @cq_context: Context associated with the CQ returned to the user via
2379 * the associated completion and event handlers.
2380 * @cq_attr: The attributes the CQ should be created upon.
2382 * Users can examine the cq structure to determine the actual CQ size.
2384 struct ib_cq *ib_create_cq(struct ib_device *device,
2385 ib_comp_handler comp_handler,
2386 void (*event_handler)(struct ib_event *, void *),
2388 const struct ib_cq_init_attr *cq_attr);
2391 * ib_resize_cq - Modifies the capacity of the CQ.
2392 * @cq: The CQ to resize.
2393 * @cqe: The minimum size of the CQ.
2395 * Users can examine the cq structure to determine the actual CQ size.
2397 int ib_resize_cq(struct ib_cq *cq, int cqe);
2400 * ib_modify_cq - Modifies moderation params of the CQ
2401 * @cq: The CQ to modify.
2402 * @cq_count: number of CQEs that will trigger an event
2403 * @cq_period: max period of time in usec before triggering an event
2406 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2409 * ib_destroy_cq - Destroys the specified CQ.
2410 * @cq: The CQ to destroy.
2412 int ib_destroy_cq(struct ib_cq *cq);
2415 * ib_poll_cq - poll a CQ for completion(s)
2416 * @cq:the CQ being polled
2417 * @num_entries:maximum number of completions to return
2418 * @wc:array of at least @num_entries &struct ib_wc where completions
2421 * Poll a CQ for (possibly multiple) completions. If the return value
2422 * is < 0, an error occurred. If the return value is >= 0, it is the
2423 * number of completions returned. If the return value is
2424 * non-negative and < num_entries, then the CQ was emptied.
2426 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2429 return cq->device->poll_cq(cq, num_entries, wc);
2433 * ib_peek_cq - Returns the number of unreaped completions currently
2434 * on the specified CQ.
2435 * @cq: The CQ to peek.
2436 * @wc_cnt: A minimum number of unreaped completions to check for.
2438 * If the number of unreaped completions is greater than or equal to wc_cnt,
2439 * this function returns wc_cnt, otherwise, it returns the actual number of
2440 * unreaped completions.
2442 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2445 * ib_req_notify_cq - Request completion notification on a CQ.
2446 * @cq: The CQ to generate an event for.
2448 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2449 * to request an event on the next solicited event or next work
2450 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2451 * may also be |ed in to request a hint about missed events, as
2455 * < 0 means an error occurred while requesting notification
2456 * == 0 means notification was requested successfully, and if
2457 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2458 * were missed and it is safe to wait for another event. In
2459 * this case is it guaranteed that any work completions added
2460 * to the CQ since the last CQ poll will trigger a completion
2461 * notification event.
2462 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2463 * in. It means that the consumer must poll the CQ again to
2464 * make sure it is empty to avoid missing an event because of a
2465 * race between requesting notification and an entry being
2466 * added to the CQ. This return value means it is possible
2467 * (but not guaranteed) that a work completion has been added
2468 * to the CQ since the last poll without triggering a
2469 * completion notification event.
2471 static inline int ib_req_notify_cq(struct ib_cq *cq,
2472 enum ib_cq_notify_flags flags)
2474 return cq->device->req_notify_cq(cq, flags);
2478 * ib_req_ncomp_notif - Request completion notification when there are
2479 * at least the specified number of unreaped completions on the CQ.
2480 * @cq: The CQ to generate an event for.
2481 * @wc_cnt: The number of unreaped completions that should be on the
2482 * CQ before an event is generated.
2484 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2486 return cq->device->req_ncomp_notif ?
2487 cq->device->req_ncomp_notif(cq, wc_cnt) :
2492 * ib_get_dma_mr - Returns a memory region for system memory that is
2494 * @pd: The protection domain associated with the memory region.
2495 * @mr_access_flags: Specifies the memory access rights.
2497 * Note that the ib_dma_*() functions defined below must be used
2498 * to create/destroy addresses used with the Lkey or Rkey returned
2499 * by ib_get_dma_mr().
2501 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2504 * ib_dma_mapping_error - check a DMA addr for error
2505 * @dev: The device for which the dma_addr was created
2506 * @dma_addr: The DMA address to check
2508 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2511 return dev->dma_ops->mapping_error(dev, dma_addr);
2512 return dma_mapping_error(dev->dma_device, dma_addr);
2516 * ib_dma_map_single - Map a kernel virtual address to DMA address
2517 * @dev: The device for which the dma_addr is to be created
2518 * @cpu_addr: The kernel virtual address
2519 * @size: The size of the region in bytes
2520 * @direction: The direction of the DMA
2522 static inline u64 ib_dma_map_single(struct ib_device *dev,
2523 void *cpu_addr, size_t size,
2524 enum dma_data_direction direction)
2527 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2528 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2532 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2533 * @dev: The device for which the DMA address was created
2534 * @addr: The DMA address
2535 * @size: The size of the region in bytes
2536 * @direction: The direction of the DMA
2538 static inline void ib_dma_unmap_single(struct ib_device *dev,
2539 u64 addr, size_t size,
2540 enum dma_data_direction direction)
2543 dev->dma_ops->unmap_single(dev, addr, size, direction);
2545 dma_unmap_single(dev->dma_device, addr, size, direction);
2548 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2549 void *cpu_addr, size_t size,
2550 enum dma_data_direction direction,
2551 struct dma_attrs *attrs)
2553 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2557 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2558 u64 addr, size_t size,
2559 enum dma_data_direction direction,
2560 struct dma_attrs *attrs)
2562 return dma_unmap_single_attrs(dev->dma_device, addr, size,
2567 * ib_dma_map_page - Map a physical page to DMA address
2568 * @dev: The device for which the dma_addr is to be created
2569 * @page: The page to be mapped
2570 * @offset: The offset within the page
2571 * @size: The size of the region in bytes
2572 * @direction: The direction of the DMA
2574 static inline u64 ib_dma_map_page(struct ib_device *dev,
2576 unsigned long offset,
2578 enum dma_data_direction direction)
2581 return dev->dma_ops->map_page(dev, page, offset, size, direction);
2582 return dma_map_page(dev->dma_device, page, offset, size, direction);
2586 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2587 * @dev: The device for which the DMA address was created
2588 * @addr: The DMA address
2589 * @size: The size of the region in bytes
2590 * @direction: The direction of the DMA
2592 static inline void ib_dma_unmap_page(struct ib_device *dev,
2593 u64 addr, size_t size,
2594 enum dma_data_direction direction)
2597 dev->dma_ops->unmap_page(dev, addr, size, direction);
2599 dma_unmap_page(dev->dma_device, addr, size, direction);
2603 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2604 * @dev: The device for which the DMA addresses are to be created
2605 * @sg: The array of scatter/gather entries
2606 * @nents: The number of scatter/gather entries
2607 * @direction: The direction of the DMA
2609 static inline int ib_dma_map_sg(struct ib_device *dev,
2610 struct scatterlist *sg, int nents,
2611 enum dma_data_direction direction)
2614 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2615 return dma_map_sg(dev->dma_device, sg, nents, direction);
2619 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2620 * @dev: The device for which the DMA addresses were created
2621 * @sg: The array of scatter/gather entries
2622 * @nents: The number of scatter/gather entries
2623 * @direction: The direction of the DMA
2625 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2626 struct scatterlist *sg, int nents,
2627 enum dma_data_direction direction)
2630 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2632 dma_unmap_sg(dev->dma_device, sg, nents, direction);
2635 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2636 struct scatterlist *sg, int nents,
2637 enum dma_data_direction direction,
2638 struct dma_attrs *attrs)
2640 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2643 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2644 struct scatterlist *sg, int nents,
2645 enum dma_data_direction direction,
2646 struct dma_attrs *attrs)
2648 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2651 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2652 * @dev: The device for which the DMA addresses were created
2653 * @sg: The scatter/gather entry
2655 * Note: this function is obsolete. To do: change all occurrences of
2656 * ib_sg_dma_address() into sg_dma_address().
2658 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2659 struct scatterlist *sg)
2661 return sg_dma_address(sg);
2665 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2666 * @dev: The device for which the DMA addresses were created
2667 * @sg: The scatter/gather entry
2669 * Note: this function is obsolete. To do: change all occurrences of
2670 * ib_sg_dma_len() into sg_dma_len().
2672 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2673 struct scatterlist *sg)
2675 return sg_dma_len(sg);
2679 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2680 * @dev: The device for which the DMA address was created
2681 * @addr: The DMA address
2682 * @size: The size of the region in bytes
2683 * @dir: The direction of the DMA
2685 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2688 enum dma_data_direction dir)
2691 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2693 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2697 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2698 * @dev: The device for which the DMA address was created
2699 * @addr: The DMA address
2700 * @size: The size of the region in bytes
2701 * @dir: The direction of the DMA
2703 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2706 enum dma_data_direction dir)
2709 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2711 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2715 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2716 * @dev: The device for which the DMA address is requested
2717 * @size: The size of the region to allocate in bytes
2718 * @dma_handle: A pointer for returning the DMA address of the region
2719 * @flag: memory allocator flags
2721 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2727 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2732 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2733 *dma_handle = handle;
2739 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2740 * @dev: The device for which the DMA addresses were allocated
2741 * @size: The size of the region
2742 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2743 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2745 static inline void ib_dma_free_coherent(struct ib_device *dev,
2746 size_t size, void *cpu_addr,
2750 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2752 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2756 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
2758 * @pd: The protection domain associated assigned to the registered region.
2759 * @phys_buf_array: Specifies a list of physical buffers to use in the
2761 * @num_phys_buf: Specifies the size of the phys_buf_array.
2762 * @mr_access_flags: Specifies the memory access rights.
2763 * @iova_start: The offset of the region's starting I/O virtual address.
2765 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
2766 struct ib_phys_buf *phys_buf_array,
2768 int mr_access_flags,
2772 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
2773 * Conceptually, this call performs the functions deregister memory region
2774 * followed by register physical memory region. Where possible,
2775 * resources are reused instead of deallocated and reallocated.
2776 * @mr: The memory region to modify.
2777 * @mr_rereg_mask: A bit-mask used to indicate which of the following
2778 * properties of the memory region are being modified.
2779 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
2780 * the new protection domain to associated with the memory region,
2781 * otherwise, this parameter is ignored.
2782 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2783 * field specifies a list of physical buffers to use in the new
2784 * translation, otherwise, this parameter is ignored.
2785 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2786 * field specifies the size of the phys_buf_array, otherwise, this
2787 * parameter is ignored.
2788 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
2789 * field specifies the new memory access rights, otherwise, this
2790 * parameter is ignored.
2791 * @iova_start: The offset of the region's starting I/O virtual address.
2793 int ib_rereg_phys_mr(struct ib_mr *mr,
2796 struct ib_phys_buf *phys_buf_array,
2798 int mr_access_flags,
2802 * ib_query_mr - Retrieves information about a specific memory region.
2803 * @mr: The memory region to retrieve information about.
2804 * @mr_attr: The attributes of the specified memory region.
2806 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2809 * ib_dereg_mr - Deregisters a memory region and removes it from the
2810 * HCA translation table.
2811 * @mr: The memory region to deregister.
2813 * This function can fail, if the memory region has memory windows bound to it.
2815 int ib_dereg_mr(struct ib_mr *mr);
2817 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
2818 enum ib_mr_type mr_type,
2822 * ib_alloc_fast_reg_page_list - Allocates a page list array
2823 * @device - ib device pointer.
2824 * @page_list_len - size of the page list array to be allocated.
2826 * This allocates and returns a struct ib_fast_reg_page_list * and a
2827 * page_list array that is at least page_list_len in size. The actual
2828 * size is returned in max_page_list_len. The caller is responsible
2829 * for initializing the contents of the page_list array before posting
2830 * a send work request with the IB_WC_FAST_REG_MR opcode.
2832 * The page_list array entries must be translated using one of the
2833 * ib_dma_*() functions just like the addresses passed to
2834 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
2835 * ib_fast_reg_page_list must not be modified by the caller until the
2836 * IB_WC_FAST_REG_MR work request completes.
2838 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2839 struct ib_device *device, int page_list_len);
2842 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2844 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2846 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2849 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2851 * @mr - struct ib_mr pointer to be updated.
2852 * @newkey - new key to be used.
2854 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2856 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2857 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2861 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2862 * for calculating a new rkey for type 2 memory windows.
2863 * @rkey - the rkey to increment.
2865 static inline u32 ib_inc_rkey(u32 rkey)
2867 const u32 mask = 0x000000ff;
2868 return ((rkey + 1) & mask) | (rkey & ~mask);
2872 * ib_alloc_mw - Allocates a memory window.
2873 * @pd: The protection domain associated with the memory window.
2874 * @type: The type of the memory window (1 or 2).
2876 struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
2879 * ib_bind_mw - Posts a work request to the send queue of the specified
2880 * QP, which binds the memory window to the given address range and
2881 * remote access attributes.
2882 * @qp: QP to post the bind work request on.
2883 * @mw: The memory window to bind.
2884 * @mw_bind: Specifies information about the memory window, including
2885 * its address range, remote access rights, and associated memory region.
2887 * If there is no immediate error, the function will update the rkey member
2888 * of the mw parameter to its new value. The bind operation can still fail
2891 static inline int ib_bind_mw(struct ib_qp *qp,
2893 struct ib_mw_bind *mw_bind)
2895 /* XXX reference counting in corresponding MR? */
2896 return mw->device->bind_mw ?
2897 mw->device->bind_mw(qp, mw, mw_bind) :
2902 * ib_dealloc_mw - Deallocates a memory window.
2903 * @mw: The memory window to deallocate.
2905 int ib_dealloc_mw(struct ib_mw *mw);
2908 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2909 * @pd: The protection domain associated with the unmapped region.
2910 * @mr_access_flags: Specifies the memory access rights.
2911 * @fmr_attr: Attributes of the unmapped region.
2913 * A fast memory region must be mapped before it can be used as part of
2916 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2917 int mr_access_flags,
2918 struct ib_fmr_attr *fmr_attr);
2921 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2922 * @fmr: The fast memory region to associate with the pages.
2923 * @page_list: An array of physical pages to map to the fast memory region.
2924 * @list_len: The number of pages in page_list.
2925 * @iova: The I/O virtual address to use with the mapped region.
2927 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2928 u64 *page_list, int list_len,
2931 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2935 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2936 * @fmr_list: A linked list of fast memory regions to unmap.
2938 int ib_unmap_fmr(struct list_head *fmr_list);
2941 * ib_dealloc_fmr - Deallocates a fast memory region.
2942 * @fmr: The fast memory region to deallocate.
2944 int ib_dealloc_fmr(struct ib_fmr *fmr);
2947 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2948 * @qp: QP to attach to the multicast group. The QP must be type
2950 * @gid: Multicast group GID.
2951 * @lid: Multicast group LID in host byte order.
2953 * In order to send and receive multicast packets, subnet
2954 * administration must have created the multicast group and configured
2955 * the fabric appropriately. The port associated with the specified
2956 * QP must also be a member of the multicast group.
2958 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2961 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2962 * @qp: QP to detach from the multicast group.
2963 * @gid: Multicast group GID.
2964 * @lid: Multicast group LID in host byte order.
2966 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2969 * ib_alloc_xrcd - Allocates an XRC domain.
2970 * @device: The device on which to allocate the XRC domain.
2972 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2975 * ib_dealloc_xrcd - Deallocates an XRC domain.
2976 * @xrcd: The XRC domain to deallocate.
2978 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2980 struct ib_flow *ib_create_flow(struct ib_qp *qp,
2981 struct ib_flow_attr *flow_attr, int domain);
2982 int ib_destroy_flow(struct ib_flow *flow_id);
2984 static inline int ib_check_mr_access(int flags)
2987 * Local write permission is required if remote write or
2988 * remote atomic permission is also requested.
2990 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
2991 !(flags & IB_ACCESS_LOCAL_WRITE))
2998 * ib_check_mr_status: lightweight check of MR status.
2999 * This routine may provide status checks on a selected
3000 * ib_mr. first use is for signature status check.
3002 * @mr: A memory region.
3003 * @check_mask: Bitmask of which checks to perform from
3004 * ib_mr_status_check enumeration.
3005 * @mr_status: The container of relevant status checks.
3006 * failed checks will be indicated in the status bitmask
3007 * and the relevant info shall be in the error item.
3009 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3010 struct ib_mr_status *mr_status);
3012 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3013 u16 pkey, const union ib_gid *gid,
3014 const struct sockaddr *addr);
3016 #endif /* IB_VERBS_H */