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 <uapi/linux/if_ether.h>
53 #include <linux/atomic.h>
54 #include <linux/mmu_notifier.h>
55 #include <asm/uaccess.h>
57 extern struct workqueue_struct *ib_wq;
68 /* IB values map to NodeInfo:NodeType. */
77 enum rdma_transport_type {
81 RDMA_TRANSPORT_USNIC_UDP
84 enum rdma_protocol_type {
88 RDMA_PROTOCOL_USNIC_UDP
91 __attribute_const__ enum rdma_transport_type
92 rdma_node_get_transport(enum rdma_node_type node_type);
94 enum rdma_link_layer {
95 IB_LINK_LAYER_UNSPECIFIED,
96 IB_LINK_LAYER_INFINIBAND,
97 IB_LINK_LAYER_ETHERNET,
100 enum ib_device_cap_flags {
101 IB_DEVICE_RESIZE_MAX_WR = 1,
102 IB_DEVICE_BAD_PKEY_CNTR = (1<<1),
103 IB_DEVICE_BAD_QKEY_CNTR = (1<<2),
104 IB_DEVICE_RAW_MULTI = (1<<3),
105 IB_DEVICE_AUTO_PATH_MIG = (1<<4),
106 IB_DEVICE_CHANGE_PHY_PORT = (1<<5),
107 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6),
108 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7),
109 IB_DEVICE_SHUTDOWN_PORT = (1<<8),
110 IB_DEVICE_INIT_TYPE = (1<<9),
111 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10),
112 IB_DEVICE_SYS_IMAGE_GUID = (1<<11),
113 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12),
114 IB_DEVICE_SRQ_RESIZE = (1<<13),
115 IB_DEVICE_N_NOTIFY_CQ = (1<<14),
116 IB_DEVICE_LOCAL_DMA_LKEY = (1<<15),
117 IB_DEVICE_RESERVED = (1<<16), /* old SEND_W_INV */
118 IB_DEVICE_MEM_WINDOW = (1<<17),
120 * Devices should set IB_DEVICE_UD_IP_SUM if they support
121 * insertion of UDP and TCP checksum on outgoing UD IPoIB
122 * messages and can verify the validity of checksum for
123 * incoming messages. Setting this flag implies that the
124 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
126 IB_DEVICE_UD_IP_CSUM = (1<<18),
127 IB_DEVICE_UD_TSO = (1<<19),
128 IB_DEVICE_XRC = (1<<20),
129 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21),
130 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
131 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1<<23),
132 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1<<24),
133 IB_DEVICE_MANAGED_FLOW_STEERING = (1<<29),
134 IB_DEVICE_SIGNATURE_HANDOVER = (1<<30),
135 IB_DEVICE_ON_DEMAND_PAGING = (1<<31),
138 enum ib_signature_prot_cap {
139 IB_PROT_T10DIF_TYPE_1 = 1,
140 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
141 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
144 enum ib_signature_guard_cap {
145 IB_GUARD_T10DIF_CRC = 1,
146 IB_GUARD_T10DIF_CSUM = 1 << 1,
155 enum ib_odp_general_cap_bits {
156 IB_ODP_SUPPORT = 1 << 0,
159 enum ib_odp_transport_cap_bits {
160 IB_ODP_SUPPORT_SEND = 1 << 0,
161 IB_ODP_SUPPORT_RECV = 1 << 1,
162 IB_ODP_SUPPORT_WRITE = 1 << 2,
163 IB_ODP_SUPPORT_READ = 1 << 3,
164 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
168 uint64_t general_caps;
170 uint32_t rc_odp_caps;
171 uint32_t uc_odp_caps;
172 uint32_t ud_odp_caps;
173 } per_transport_caps;
176 enum ib_cq_creation_flags {
177 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
180 struct ib_cq_init_attr {
186 struct ib_device_attr {
188 __be64 sys_image_guid;
196 int device_cap_flags;
206 int max_qp_init_rd_atom;
207 int max_ee_init_rd_atom;
208 enum ib_atomic_cap atomic_cap;
209 enum ib_atomic_cap masked_atomic_cap;
216 int max_mcast_qp_attach;
217 int max_total_mcast_qp_attach;
224 unsigned int max_fast_reg_page_list_len;
226 u8 local_ca_ack_delay;
229 struct ib_odp_caps odp_caps;
230 uint64_t timestamp_mask;
231 uint64_t hca_core_clock; /* in KHZ */
242 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
245 case IB_MTU_256: return 256;
246 case IB_MTU_512: return 512;
247 case IB_MTU_1024: return 1024;
248 case IB_MTU_2048: return 2048;
249 case IB_MTU_4096: return 4096;
260 IB_PORT_ACTIVE_DEFER = 5
263 enum ib_port_cap_flags {
265 IB_PORT_NOTICE_SUP = 1 << 2,
266 IB_PORT_TRAP_SUP = 1 << 3,
267 IB_PORT_OPT_IPD_SUP = 1 << 4,
268 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
269 IB_PORT_SL_MAP_SUP = 1 << 6,
270 IB_PORT_MKEY_NVRAM = 1 << 7,
271 IB_PORT_PKEY_NVRAM = 1 << 8,
272 IB_PORT_LED_INFO_SUP = 1 << 9,
273 IB_PORT_SM_DISABLED = 1 << 10,
274 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
275 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
276 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
277 IB_PORT_CM_SUP = 1 << 16,
278 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
279 IB_PORT_REINIT_SUP = 1 << 18,
280 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
281 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
282 IB_PORT_DR_NOTICE_SUP = 1 << 21,
283 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
284 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
285 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
286 IB_PORT_CLIENT_REG_SUP = 1 << 25,
287 IB_PORT_IP_BASED_GIDS = 1 << 26
297 static inline int ib_width_enum_to_int(enum ib_port_width width)
300 case IB_WIDTH_1X: return 1;
301 case IB_WIDTH_4X: return 4;
302 case IB_WIDTH_8X: return 8;
303 case IB_WIDTH_12X: return 12;
317 struct ib_protocol_stats {
321 struct iw_protocol_stats {
324 u64 ipInTooBigErrors;
327 u64 ipInUnknownProtos;
328 u64 ipInTruncatedPkts;
331 u64 ipOutForwDatagrams;
363 union rdma_protocol_stats {
364 struct ib_protocol_stats ib;
365 struct iw_protocol_stats iw;
368 /* Define bits for the various functionality this port needs to be supported by
371 /* Management 0x00000FFF */
372 #define RDMA_CORE_CAP_IB_MAD 0x00000001
373 #define RDMA_CORE_CAP_IB_SMI 0x00000002
374 #define RDMA_CORE_CAP_IB_CM 0x00000004
375 #define RDMA_CORE_CAP_IW_CM 0x00000008
376 #define RDMA_CORE_CAP_IB_SA 0x00000010
377 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
379 /* Address format 0x000FF000 */
380 #define RDMA_CORE_CAP_AF_IB 0x00001000
381 #define RDMA_CORE_CAP_ETH_AH 0x00002000
383 /* Protocol 0xFFF00000 */
384 #define RDMA_CORE_CAP_PROT_IB 0x00100000
385 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
386 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
388 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
389 | RDMA_CORE_CAP_IB_MAD \
390 | RDMA_CORE_CAP_IB_SMI \
391 | RDMA_CORE_CAP_IB_CM \
392 | RDMA_CORE_CAP_IB_SA \
393 | RDMA_CORE_CAP_AF_IB)
394 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
395 | RDMA_CORE_CAP_IB_MAD \
396 | RDMA_CORE_CAP_IB_CM \
397 | RDMA_CORE_CAP_AF_IB \
398 | RDMA_CORE_CAP_ETH_AH)
399 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
400 | RDMA_CORE_CAP_IW_CM)
401 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
402 | RDMA_CORE_CAP_OPA_MAD)
404 struct ib_port_attr {
405 enum ib_port_state state;
407 enum ib_mtu active_mtu;
426 enum ib_device_modify_flags {
427 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
428 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
431 struct ib_device_modify {
436 enum ib_port_modify_flags {
437 IB_PORT_SHUTDOWN = 1,
438 IB_PORT_INIT_TYPE = (1<<2),
439 IB_PORT_RESET_QKEY_CNTR = (1<<3)
442 struct ib_port_modify {
443 u32 set_port_cap_mask;
444 u32 clr_port_cap_mask;
452 IB_EVENT_QP_ACCESS_ERR,
456 IB_EVENT_PATH_MIG_ERR,
457 IB_EVENT_DEVICE_FATAL,
458 IB_EVENT_PORT_ACTIVE,
461 IB_EVENT_PKEY_CHANGE,
464 IB_EVENT_SRQ_LIMIT_REACHED,
465 IB_EVENT_QP_LAST_WQE_REACHED,
466 IB_EVENT_CLIENT_REREGISTER,
470 __attribute_const__ const char *ib_event_msg(enum ib_event_type event);
473 struct ib_device *device;
480 enum ib_event_type event;
483 struct ib_event_handler {
484 struct ib_device *device;
485 void (*handler)(struct ib_event_handler *, struct ib_event *);
486 struct list_head list;
489 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
491 (_ptr)->device = _device; \
492 (_ptr)->handler = _handler; \
493 INIT_LIST_HEAD(&(_ptr)->list); \
496 struct ib_global_route {
505 __be32 version_tclass_flow;
514 IB_MULTICAST_QPN = 0xffffff
517 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
524 IB_RATE_PORT_CURRENT = 0,
525 IB_RATE_2_5_GBPS = 2,
533 IB_RATE_120_GBPS = 10,
534 IB_RATE_14_GBPS = 11,
535 IB_RATE_56_GBPS = 12,
536 IB_RATE_112_GBPS = 13,
537 IB_RATE_168_GBPS = 14,
538 IB_RATE_25_GBPS = 15,
539 IB_RATE_100_GBPS = 16,
540 IB_RATE_200_GBPS = 17,
541 IB_RATE_300_GBPS = 18
545 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
546 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
547 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
548 * @rate: rate to convert.
550 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
553 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
554 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
555 * @rate: rate to convert.
557 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
559 enum ib_mr_create_flags {
560 IB_MR_SIGNATURE_EN = 1,
564 * ib_mr_init_attr - Memory region init attributes passed to routine
566 * @max_reg_descriptors: max number of registration descriptors that
567 * may be used with registration work requests.
568 * @flags: MR creation flags bit mask.
570 struct ib_mr_init_attr {
571 int max_reg_descriptors;
577 * IB_SIG_TYPE_NONE: Unprotected.
578 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
580 enum ib_signature_type {
586 * Signature T10-DIF block-guard types
587 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
588 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
590 enum ib_t10_dif_bg_type {
596 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
598 * @bg_type: T10-DIF block guard type (CRC|CSUM)
599 * @pi_interval: protection information interval.
600 * @bg: seed of guard computation.
601 * @app_tag: application tag of guard block
602 * @ref_tag: initial guard block reference tag.
603 * @ref_remap: Indicate wethear the reftag increments each block
604 * @app_escape: Indicate to skip block check if apptag=0xffff
605 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
606 * @apptag_check_mask: check bitmask of application tag.
608 struct ib_t10_dif_domain {
609 enum ib_t10_dif_bg_type bg_type;
617 u16 apptag_check_mask;
621 * struct ib_sig_domain - Parameters for signature domain
622 * @sig_type: specific signauture type
623 * @sig: union of all signature domain attributes that may
624 * be used to set domain layout.
626 struct ib_sig_domain {
627 enum ib_signature_type sig_type;
629 struct ib_t10_dif_domain dif;
634 * struct ib_sig_attrs - Parameters for signature handover operation
635 * @check_mask: bitmask for signature byte check (8 bytes)
636 * @mem: memory domain layout desciptor.
637 * @wire: wire domain layout desciptor.
639 struct ib_sig_attrs {
641 struct ib_sig_domain mem;
642 struct ib_sig_domain wire;
645 enum ib_sig_err_type {
652 * struct ib_sig_err - signature error descriptor
655 enum ib_sig_err_type err_type;
662 enum ib_mr_status_check {
663 IB_MR_CHECK_SIG_STATUS = 1,
667 * struct ib_mr_status - Memory region status container
669 * @fail_status: Bitmask of MR checks status. For each
670 * failed check a corresponding status bit is set.
671 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
674 struct ib_mr_status {
676 struct ib_sig_err sig_err;
680 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
682 * @mult: multiple to convert.
684 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
687 struct ib_global_route grh;
702 IB_WC_LOC_EEC_OP_ERR,
707 IB_WC_LOC_ACCESS_ERR,
708 IB_WC_REM_INV_REQ_ERR,
709 IB_WC_REM_ACCESS_ERR,
712 IB_WC_RNR_RETRY_EXC_ERR,
713 IB_WC_LOC_RDD_VIOL_ERR,
714 IB_WC_REM_INV_RD_REQ_ERR,
717 IB_WC_INV_EEC_STATE_ERR,
719 IB_WC_RESP_TIMEOUT_ERR,
723 __attribute_const__ const char *ib_wc_status_msg(enum ib_wc_status status);
735 IB_WC_MASKED_COMP_SWAP,
736 IB_WC_MASKED_FETCH_ADD,
738 * Set value of IB_WC_RECV so consumers can test if a completion is a
739 * receive by testing (opcode & IB_WC_RECV).
742 IB_WC_RECV_RDMA_WITH_IMM
747 IB_WC_WITH_IMM = (1<<1),
748 IB_WC_WITH_INVALIDATE = (1<<2),
749 IB_WC_IP_CSUM_OK = (1<<3),
750 IB_WC_WITH_SMAC = (1<<4),
751 IB_WC_WITH_VLAN = (1<<5),
756 enum ib_wc_status status;
757 enum ib_wc_opcode opcode;
771 u8 port_num; /* valid only for DR SMPs on switches */
776 enum ib_cq_notify_flags {
777 IB_CQ_SOLICITED = 1 << 0,
778 IB_CQ_NEXT_COMP = 1 << 1,
779 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
780 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
788 enum ib_srq_attr_mask {
789 IB_SRQ_MAX_WR = 1 << 0,
790 IB_SRQ_LIMIT = 1 << 1,
799 struct ib_srq_init_attr {
800 void (*event_handler)(struct ib_event *, void *);
802 struct ib_srq_attr attr;
803 enum ib_srq_type srq_type;
807 struct ib_xrcd *xrcd;
828 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
829 * here (and in that order) since the MAD layer uses them as
830 * indices into a 2-entry table.
839 IB_QPT_RAW_ETHERTYPE,
840 IB_QPT_RAW_PACKET = 8,
844 /* Reserve a range for qp types internal to the low level driver.
845 * These qp types will not be visible at the IB core layer, so the
846 * IB_QPT_MAX usages should not be affected in the core layer
848 IB_QPT_RESERVED1 = 0x1000,
860 enum ib_qp_create_flags {
861 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
862 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
863 IB_QP_CREATE_NETIF_QP = 1 << 5,
864 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
865 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
866 /* reserve bits 26-31 for low level drivers' internal use */
867 IB_QP_CREATE_RESERVED_START = 1 << 26,
868 IB_QP_CREATE_RESERVED_END = 1 << 31,
873 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
874 * callback to destroy the passed in QP.
877 struct ib_qp_init_attr {
878 void (*event_handler)(struct ib_event *, void *);
880 struct ib_cq *send_cq;
881 struct ib_cq *recv_cq;
883 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
884 struct ib_qp_cap cap;
885 enum ib_sig_type sq_sig_type;
886 enum ib_qp_type qp_type;
887 enum ib_qp_create_flags create_flags;
888 u8 port_num; /* special QP types only */
891 struct ib_qp_open_attr {
892 void (*event_handler)(struct ib_event *, void *);
895 enum ib_qp_type qp_type;
898 enum ib_rnr_timeout {
899 IB_RNR_TIMER_655_36 = 0,
900 IB_RNR_TIMER_000_01 = 1,
901 IB_RNR_TIMER_000_02 = 2,
902 IB_RNR_TIMER_000_03 = 3,
903 IB_RNR_TIMER_000_04 = 4,
904 IB_RNR_TIMER_000_06 = 5,
905 IB_RNR_TIMER_000_08 = 6,
906 IB_RNR_TIMER_000_12 = 7,
907 IB_RNR_TIMER_000_16 = 8,
908 IB_RNR_TIMER_000_24 = 9,
909 IB_RNR_TIMER_000_32 = 10,
910 IB_RNR_TIMER_000_48 = 11,
911 IB_RNR_TIMER_000_64 = 12,
912 IB_RNR_TIMER_000_96 = 13,
913 IB_RNR_TIMER_001_28 = 14,
914 IB_RNR_TIMER_001_92 = 15,
915 IB_RNR_TIMER_002_56 = 16,
916 IB_RNR_TIMER_003_84 = 17,
917 IB_RNR_TIMER_005_12 = 18,
918 IB_RNR_TIMER_007_68 = 19,
919 IB_RNR_TIMER_010_24 = 20,
920 IB_RNR_TIMER_015_36 = 21,
921 IB_RNR_TIMER_020_48 = 22,
922 IB_RNR_TIMER_030_72 = 23,
923 IB_RNR_TIMER_040_96 = 24,
924 IB_RNR_TIMER_061_44 = 25,
925 IB_RNR_TIMER_081_92 = 26,
926 IB_RNR_TIMER_122_88 = 27,
927 IB_RNR_TIMER_163_84 = 28,
928 IB_RNR_TIMER_245_76 = 29,
929 IB_RNR_TIMER_327_68 = 30,
930 IB_RNR_TIMER_491_52 = 31
933 enum ib_qp_attr_mask {
935 IB_QP_CUR_STATE = (1<<1),
936 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
937 IB_QP_ACCESS_FLAGS = (1<<3),
938 IB_QP_PKEY_INDEX = (1<<4),
942 IB_QP_PATH_MTU = (1<<8),
943 IB_QP_TIMEOUT = (1<<9),
944 IB_QP_RETRY_CNT = (1<<10),
945 IB_QP_RNR_RETRY = (1<<11),
946 IB_QP_RQ_PSN = (1<<12),
947 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
948 IB_QP_ALT_PATH = (1<<14),
949 IB_QP_MIN_RNR_TIMER = (1<<15),
950 IB_QP_SQ_PSN = (1<<16),
951 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
952 IB_QP_PATH_MIG_STATE = (1<<18),
954 IB_QP_DEST_QPN = (1<<20),
955 IB_QP_SMAC = (1<<21),
956 IB_QP_ALT_SMAC = (1<<22),
958 IB_QP_ALT_VID = (1<<24),
983 enum ib_qp_state qp_state;
984 enum ib_qp_state cur_qp_state;
985 enum ib_mtu path_mtu;
986 enum ib_mig_state path_mig_state;
992 struct ib_qp_cap cap;
993 struct ib_ah_attr ah_attr;
994 struct ib_ah_attr alt_ah_attr;
997 u8 en_sqd_async_notify;
1000 u8 max_dest_rd_atomic;
1009 u8 alt_smac[ETH_ALEN];
1016 IB_WR_RDMA_WRITE_WITH_IMM,
1018 IB_WR_SEND_WITH_IMM,
1020 IB_WR_ATOMIC_CMP_AND_SWP,
1021 IB_WR_ATOMIC_FETCH_AND_ADD,
1023 IB_WR_SEND_WITH_INV,
1024 IB_WR_RDMA_READ_WITH_INV,
1027 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1028 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1031 /* reserve values for low level drivers' internal use.
1032 * These values will not be used at all in the ib core layer.
1034 IB_WR_RESERVED1 = 0xf0,
1046 enum ib_send_flags {
1048 IB_SEND_SIGNALED = (1<<1),
1049 IB_SEND_SOLICITED = (1<<2),
1050 IB_SEND_INLINE = (1<<3),
1051 IB_SEND_IP_CSUM = (1<<4),
1053 /* reserve bits 26-31 for low level drivers' internal use */
1054 IB_SEND_RESERVED_START = (1 << 26),
1055 IB_SEND_RESERVED_END = (1 << 31),
1064 struct ib_fast_reg_page_list {
1065 struct ib_device *device;
1067 unsigned int max_page_list_len;
1071 * struct ib_mw_bind_info - Parameters for a memory window bind operation.
1072 * @mr: A memory region to bind the memory window to.
1073 * @addr: The address where the memory window should begin.
1074 * @length: The length of the memory window, in bytes.
1075 * @mw_access_flags: Access flags from enum ib_access_flags for the window.
1077 * This struct contains the shared parameters for type 1 and type 2
1078 * memory window bind operations.
1080 struct ib_mw_bind_info {
1084 int mw_access_flags;
1088 struct ib_send_wr *next;
1090 struct ib_sge *sg_list;
1092 enum ib_wr_opcode opcode;
1096 u32 invalidate_rkey;
1107 u64 compare_add_mask;
1118 u16 pkey_index; /* valid for GSI only */
1119 u8 port_num; /* valid for DR SMPs on switch only */
1123 struct ib_fast_reg_page_list *page_list;
1124 unsigned int page_shift;
1125 unsigned int page_list_len;
1132 /* The new rkey for the memory window. */
1134 struct ib_mw_bind_info bind_info;
1137 struct ib_sig_attrs *sig_attrs;
1138 struct ib_mr *sig_mr;
1140 struct ib_sge *prot;
1143 u32 xrc_remote_srq_num; /* XRC TGT QPs only */
1147 struct ib_recv_wr *next;
1149 struct ib_sge *sg_list;
1153 enum ib_access_flags {
1154 IB_ACCESS_LOCAL_WRITE = 1,
1155 IB_ACCESS_REMOTE_WRITE = (1<<1),
1156 IB_ACCESS_REMOTE_READ = (1<<2),
1157 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1158 IB_ACCESS_MW_BIND = (1<<4),
1159 IB_ZERO_BASED = (1<<5),
1160 IB_ACCESS_ON_DEMAND = (1<<6),
1163 struct ib_phys_buf {
1170 u64 device_virt_addr;
1172 int mr_access_flags;
1177 enum ib_mr_rereg_flags {
1178 IB_MR_REREG_TRANS = 1,
1179 IB_MR_REREG_PD = (1<<1),
1180 IB_MR_REREG_ACCESS = (1<<2),
1181 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1185 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
1186 * @wr_id: Work request id.
1187 * @send_flags: Flags from ib_send_flags enum.
1188 * @bind_info: More parameters of the bind operation.
1193 struct ib_mw_bind_info bind_info;
1196 struct ib_fmr_attr {
1204 struct ib_ucontext {
1205 struct ib_device *device;
1206 struct list_head pd_list;
1207 struct list_head mr_list;
1208 struct list_head mw_list;
1209 struct list_head cq_list;
1210 struct list_head qp_list;
1211 struct list_head srq_list;
1212 struct list_head ah_list;
1213 struct list_head xrcd_list;
1214 struct list_head rule_list;
1218 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1219 struct rb_root umem_tree;
1221 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1222 * mmu notifiers registration.
1224 struct rw_semaphore umem_rwsem;
1225 void (*invalidate_range)(struct ib_umem *umem,
1226 unsigned long start, unsigned long end);
1228 struct mmu_notifier mn;
1229 atomic_t notifier_count;
1230 /* A list of umems that don't have private mmu notifier counters yet. */
1231 struct list_head no_private_counters;
1237 u64 user_handle; /* handle given to us by userspace */
1238 struct ib_ucontext *context; /* associated user context */
1239 void *object; /* containing object */
1240 struct list_head list; /* link to context's list */
1241 int id; /* index into kernel idr */
1243 struct rw_semaphore mutex; /* protects .live */
1248 const void __user *inbuf;
1249 void __user *outbuf;
1255 struct ib_device *device;
1256 struct ib_uobject *uobject;
1257 atomic_t usecnt; /* count all resources */
1261 struct ib_device *device;
1262 atomic_t usecnt; /* count all exposed resources */
1263 struct inode *inode;
1265 struct mutex tgt_qp_mutex;
1266 struct list_head tgt_qp_list;
1270 struct ib_device *device;
1272 struct ib_uobject *uobject;
1275 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1278 struct ib_device *device;
1279 struct ib_uobject *uobject;
1280 ib_comp_handler comp_handler;
1281 void (*event_handler)(struct ib_event *, void *);
1284 atomic_t usecnt; /* count number of work queues */
1288 struct ib_device *device;
1290 struct ib_uobject *uobject;
1291 void (*event_handler)(struct ib_event *, void *);
1293 enum ib_srq_type srq_type;
1298 struct ib_xrcd *xrcd;
1306 struct ib_device *device;
1308 struct ib_cq *send_cq;
1309 struct ib_cq *recv_cq;
1311 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1312 struct list_head xrcd_list;
1313 /* count times opened, mcast attaches, flow attaches */
1315 struct list_head open_list;
1316 struct ib_qp *real_qp;
1317 struct ib_uobject *uobject;
1318 void (*event_handler)(struct ib_event *, void *);
1321 enum ib_qp_type qp_type;
1325 struct ib_device *device;
1327 struct ib_uobject *uobject;
1330 atomic_t usecnt; /* count number of MWs */
1334 struct ib_device *device;
1336 struct ib_uobject *uobject;
1338 enum ib_mw_type type;
1342 struct ib_device *device;
1344 struct list_head list;
1349 /* Supported steering options */
1350 enum ib_flow_attr_type {
1351 /* steering according to rule specifications */
1352 IB_FLOW_ATTR_NORMAL = 0x0,
1353 /* default unicast and multicast rule -
1354 * receive all Eth traffic which isn't steered to any QP
1356 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1357 /* default multicast rule -
1358 * receive all Eth multicast traffic which isn't steered to any QP
1360 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1361 /* sniffer rule - receive all port traffic */
1362 IB_FLOW_ATTR_SNIFFER = 0x3
1365 /* Supported steering header types */
1366 enum ib_flow_spec_type {
1368 IB_FLOW_SPEC_ETH = 0x20,
1369 IB_FLOW_SPEC_IB = 0x22,
1371 IB_FLOW_SPEC_IPV4 = 0x30,
1373 IB_FLOW_SPEC_TCP = 0x40,
1374 IB_FLOW_SPEC_UDP = 0x41
1376 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1377 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1379 /* Flow steering rule priority is set according to it's domain.
1380 * Lower domain value means higher priority.
1382 enum ib_flow_domain {
1383 IB_FLOW_DOMAIN_USER,
1384 IB_FLOW_DOMAIN_ETHTOOL,
1387 IB_FLOW_DOMAIN_NUM /* Must be last */
1390 struct ib_flow_eth_filter {
1397 struct ib_flow_spec_eth {
1398 enum ib_flow_spec_type type;
1400 struct ib_flow_eth_filter val;
1401 struct ib_flow_eth_filter mask;
1404 struct ib_flow_ib_filter {
1409 struct ib_flow_spec_ib {
1410 enum ib_flow_spec_type type;
1412 struct ib_flow_ib_filter val;
1413 struct ib_flow_ib_filter mask;
1416 struct ib_flow_ipv4_filter {
1421 struct ib_flow_spec_ipv4 {
1422 enum ib_flow_spec_type type;
1424 struct ib_flow_ipv4_filter val;
1425 struct ib_flow_ipv4_filter mask;
1428 struct ib_flow_tcp_udp_filter {
1433 struct ib_flow_spec_tcp_udp {
1434 enum ib_flow_spec_type type;
1436 struct ib_flow_tcp_udp_filter val;
1437 struct ib_flow_tcp_udp_filter mask;
1440 union ib_flow_spec {
1442 enum ib_flow_spec_type type;
1445 struct ib_flow_spec_eth eth;
1446 struct ib_flow_spec_ib ib;
1447 struct ib_flow_spec_ipv4 ipv4;
1448 struct ib_flow_spec_tcp_udp tcp_udp;
1451 struct ib_flow_attr {
1452 enum ib_flow_attr_type type;
1458 /* Following are the optional layers according to user request
1459 * struct ib_flow_spec_xxx
1460 * struct ib_flow_spec_yyy
1466 struct ib_uobject *uobject;
1472 enum ib_process_mad_flags {
1473 IB_MAD_IGNORE_MKEY = 1,
1474 IB_MAD_IGNORE_BKEY = 2,
1475 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1478 enum ib_mad_result {
1479 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1480 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1481 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1482 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1485 #define IB_DEVICE_NAME_MAX 64
1489 struct ib_event_handler event_handler;
1490 struct ib_pkey_cache **pkey_cache;
1491 struct ib_gid_cache **gid_cache;
1495 struct ib_dma_mapping_ops {
1496 int (*mapping_error)(struct ib_device *dev,
1498 u64 (*map_single)(struct ib_device *dev,
1499 void *ptr, size_t size,
1500 enum dma_data_direction direction);
1501 void (*unmap_single)(struct ib_device *dev,
1502 u64 addr, size_t size,
1503 enum dma_data_direction direction);
1504 u64 (*map_page)(struct ib_device *dev,
1505 struct page *page, unsigned long offset,
1507 enum dma_data_direction direction);
1508 void (*unmap_page)(struct ib_device *dev,
1509 u64 addr, size_t size,
1510 enum dma_data_direction direction);
1511 int (*map_sg)(struct ib_device *dev,
1512 struct scatterlist *sg, int nents,
1513 enum dma_data_direction direction);
1514 void (*unmap_sg)(struct ib_device *dev,
1515 struct scatterlist *sg, int nents,
1516 enum dma_data_direction direction);
1517 void (*sync_single_for_cpu)(struct ib_device *dev,
1520 enum dma_data_direction dir);
1521 void (*sync_single_for_device)(struct ib_device *dev,
1524 enum dma_data_direction dir);
1525 void *(*alloc_coherent)(struct ib_device *dev,
1529 void (*free_coherent)(struct ib_device *dev,
1530 size_t size, void *cpu_addr,
1536 struct ib_port_immutable {
1544 struct device *dma_device;
1546 char name[IB_DEVICE_NAME_MAX];
1548 struct list_head event_handler_list;
1549 spinlock_t event_handler_lock;
1551 spinlock_t client_data_lock;
1552 struct list_head core_list;
1553 struct list_head client_data_list;
1555 struct ib_cache cache;
1557 * port_immutable is indexed by port number
1559 struct ib_port_immutable *port_immutable;
1561 int num_comp_vectors;
1563 struct iw_cm_verbs *iwcm;
1565 int (*get_protocol_stats)(struct ib_device *device,
1566 union rdma_protocol_stats *stats);
1567 int (*query_device)(struct ib_device *device,
1568 struct ib_device_attr *device_attr,
1569 struct ib_udata *udata);
1570 int (*query_port)(struct ib_device *device,
1572 struct ib_port_attr *port_attr);
1573 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1575 int (*query_gid)(struct ib_device *device,
1576 u8 port_num, int index,
1578 int (*query_pkey)(struct ib_device *device,
1579 u8 port_num, u16 index, u16 *pkey);
1580 int (*modify_device)(struct ib_device *device,
1581 int device_modify_mask,
1582 struct ib_device_modify *device_modify);
1583 int (*modify_port)(struct ib_device *device,
1584 u8 port_num, int port_modify_mask,
1585 struct ib_port_modify *port_modify);
1586 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1587 struct ib_udata *udata);
1588 int (*dealloc_ucontext)(struct ib_ucontext *context);
1589 int (*mmap)(struct ib_ucontext *context,
1590 struct vm_area_struct *vma);
1591 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1592 struct ib_ucontext *context,
1593 struct ib_udata *udata);
1594 int (*dealloc_pd)(struct ib_pd *pd);
1595 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1596 struct ib_ah_attr *ah_attr);
1597 int (*modify_ah)(struct ib_ah *ah,
1598 struct ib_ah_attr *ah_attr);
1599 int (*query_ah)(struct ib_ah *ah,
1600 struct ib_ah_attr *ah_attr);
1601 int (*destroy_ah)(struct ib_ah *ah);
1602 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1603 struct ib_srq_init_attr *srq_init_attr,
1604 struct ib_udata *udata);
1605 int (*modify_srq)(struct ib_srq *srq,
1606 struct ib_srq_attr *srq_attr,
1607 enum ib_srq_attr_mask srq_attr_mask,
1608 struct ib_udata *udata);
1609 int (*query_srq)(struct ib_srq *srq,
1610 struct ib_srq_attr *srq_attr);
1611 int (*destroy_srq)(struct ib_srq *srq);
1612 int (*post_srq_recv)(struct ib_srq *srq,
1613 struct ib_recv_wr *recv_wr,
1614 struct ib_recv_wr **bad_recv_wr);
1615 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1616 struct ib_qp_init_attr *qp_init_attr,
1617 struct ib_udata *udata);
1618 int (*modify_qp)(struct ib_qp *qp,
1619 struct ib_qp_attr *qp_attr,
1621 struct ib_udata *udata);
1622 int (*query_qp)(struct ib_qp *qp,
1623 struct ib_qp_attr *qp_attr,
1625 struct ib_qp_init_attr *qp_init_attr);
1626 int (*destroy_qp)(struct ib_qp *qp);
1627 int (*post_send)(struct ib_qp *qp,
1628 struct ib_send_wr *send_wr,
1629 struct ib_send_wr **bad_send_wr);
1630 int (*post_recv)(struct ib_qp *qp,
1631 struct ib_recv_wr *recv_wr,
1632 struct ib_recv_wr **bad_recv_wr);
1633 struct ib_cq * (*create_cq)(struct ib_device *device,
1634 const struct ib_cq_init_attr *attr,
1635 struct ib_ucontext *context,
1636 struct ib_udata *udata);
1637 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1639 int (*destroy_cq)(struct ib_cq *cq);
1640 int (*resize_cq)(struct ib_cq *cq, int cqe,
1641 struct ib_udata *udata);
1642 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1644 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1645 int (*req_notify_cq)(struct ib_cq *cq,
1646 enum ib_cq_notify_flags flags);
1647 int (*req_ncomp_notif)(struct ib_cq *cq,
1649 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1650 int mr_access_flags);
1651 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
1652 struct ib_phys_buf *phys_buf_array,
1654 int mr_access_flags,
1656 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1657 u64 start, u64 length,
1659 int mr_access_flags,
1660 struct ib_udata *udata);
1661 int (*rereg_user_mr)(struct ib_mr *mr,
1663 u64 start, u64 length,
1665 int mr_access_flags,
1667 struct ib_udata *udata);
1668 int (*query_mr)(struct ib_mr *mr,
1669 struct ib_mr_attr *mr_attr);
1670 int (*dereg_mr)(struct ib_mr *mr);
1671 int (*destroy_mr)(struct ib_mr *mr);
1672 struct ib_mr * (*create_mr)(struct ib_pd *pd,
1673 struct ib_mr_init_attr *mr_init_attr);
1674 struct ib_mr * (*alloc_fast_reg_mr)(struct ib_pd *pd,
1675 int max_page_list_len);
1676 struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1678 void (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1679 int (*rereg_phys_mr)(struct ib_mr *mr,
1682 struct ib_phys_buf *phys_buf_array,
1684 int mr_access_flags,
1686 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1687 enum ib_mw_type type);
1688 int (*bind_mw)(struct ib_qp *qp,
1690 struct ib_mw_bind *mw_bind);
1691 int (*dealloc_mw)(struct ib_mw *mw);
1692 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1693 int mr_access_flags,
1694 struct ib_fmr_attr *fmr_attr);
1695 int (*map_phys_fmr)(struct ib_fmr *fmr,
1696 u64 *page_list, int list_len,
1698 int (*unmap_fmr)(struct list_head *fmr_list);
1699 int (*dealloc_fmr)(struct ib_fmr *fmr);
1700 int (*attach_mcast)(struct ib_qp *qp,
1703 int (*detach_mcast)(struct ib_qp *qp,
1706 int (*process_mad)(struct ib_device *device,
1707 int process_mad_flags,
1709 const struct ib_wc *in_wc,
1710 const struct ib_grh *in_grh,
1711 const struct ib_mad_hdr *in_mad,
1713 struct ib_mad_hdr *out_mad,
1714 size_t *out_mad_size,
1715 u16 *out_mad_pkey_index);
1716 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1717 struct ib_ucontext *ucontext,
1718 struct ib_udata *udata);
1719 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1720 struct ib_flow * (*create_flow)(struct ib_qp *qp,
1724 int (*destroy_flow)(struct ib_flow *flow_id);
1725 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1726 struct ib_mr_status *mr_status);
1728 struct ib_dma_mapping_ops *dma_ops;
1730 struct module *owner;
1732 struct kobject *ports_parent;
1733 struct list_head port_list;
1736 IB_DEV_UNINITIALIZED,
1742 u64 uverbs_cmd_mask;
1743 u64 uverbs_ex_cmd_mask;
1753 * The following mandatory functions are used only at device
1754 * registration. Keep functions such as these at the end of this
1755 * structure to avoid cache line misses when accessing struct ib_device
1758 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1763 void (*add) (struct ib_device *);
1764 void (*remove)(struct ib_device *);
1766 struct list_head list;
1769 struct ib_device *ib_alloc_device(size_t size);
1770 void ib_dealloc_device(struct ib_device *device);
1772 int ib_register_device(struct ib_device *device,
1773 int (*port_callback)(struct ib_device *,
1774 u8, struct kobject *));
1775 void ib_unregister_device(struct ib_device *device);
1777 int ib_register_client (struct ib_client *client);
1778 void ib_unregister_client(struct ib_client *client);
1780 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1781 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1784 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1786 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1789 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1791 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1795 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1796 * contains all required attributes and no attributes not allowed for
1797 * the given QP state transition.
1798 * @cur_state: Current QP state
1799 * @next_state: Next QP state
1801 * @mask: Mask of supplied QP attributes
1802 * @ll : link layer of port
1804 * This function is a helper function that a low-level driver's
1805 * modify_qp method can use to validate the consumer's input. It
1806 * checks that cur_state and next_state are valid QP states, that a
1807 * transition from cur_state to next_state is allowed by the IB spec,
1808 * and that the attribute mask supplied is allowed for the transition.
1810 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1811 enum ib_qp_type type, enum ib_qp_attr_mask mask,
1812 enum rdma_link_layer ll);
1814 int ib_register_event_handler (struct ib_event_handler *event_handler);
1815 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1816 void ib_dispatch_event(struct ib_event *event);
1818 int ib_query_device(struct ib_device *device,
1819 struct ib_device_attr *device_attr);
1821 int ib_query_port(struct ib_device *device,
1822 u8 port_num, struct ib_port_attr *port_attr);
1824 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1828 * rdma_cap_ib_switch - Check if the device is IB switch
1829 * @device: Device to check
1831 * Device driver is responsible for setting is_switch bit on
1832 * in ib_device structure at init time.
1834 * Return: true if the device is IB switch.
1836 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
1838 return device->is_switch;
1842 * rdma_start_port - Return the first valid port number for the device
1845 * @device: Device to be checked
1847 * Return start port number
1849 static inline u8 rdma_start_port(const struct ib_device *device)
1851 return rdma_cap_ib_switch(device) ? 0 : 1;
1855 * rdma_end_port - Return the last valid port number for the device
1858 * @device: Device to be checked
1860 * Return last port number
1862 static inline u8 rdma_end_port(const struct ib_device *device)
1864 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
1867 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
1869 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
1872 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
1874 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
1877 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
1879 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
1882 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
1884 return device->port_immutable[port_num].core_cap_flags &
1885 (RDMA_CORE_CAP_PROT_IB | RDMA_CORE_CAP_PROT_ROCE);
1889 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
1890 * Management Datagrams.
1891 * @device: Device to check
1892 * @port_num: Port number to check
1894 * Management Datagrams (MAD) are a required part of the InfiniBand
1895 * specification and are supported on all InfiniBand devices. A slightly
1896 * extended version are also supported on OPA interfaces.
1898 * Return: true if the port supports sending/receiving of MAD packets.
1900 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
1902 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
1906 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
1907 * Management Datagrams.
1908 * @device: Device to check
1909 * @port_num: Port number to check
1911 * Intel OmniPath devices extend and/or replace the InfiniBand Management
1912 * datagrams with their own versions. These OPA MADs share many but not all of
1913 * the characteristics of InfiniBand MADs.
1915 * OPA MADs differ in the following ways:
1917 * 1) MADs are variable size up to 2K
1918 * IBTA defined MADs remain fixed at 256 bytes
1919 * 2) OPA SMPs must carry valid PKeys
1920 * 3) OPA SMP packets are a different format
1922 * Return: true if the port supports OPA MAD packet formats.
1924 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
1926 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
1927 == RDMA_CORE_CAP_OPA_MAD;
1931 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
1932 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
1933 * @device: Device to check
1934 * @port_num: Port number to check
1936 * Each InfiniBand node is required to provide a Subnet Management Agent
1937 * that the subnet manager can access. Prior to the fabric being fully
1938 * configured by the subnet manager, the SMA is accessed via a well known
1939 * interface called the Subnet Management Interface (SMI). This interface
1940 * uses directed route packets to communicate with the SM to get around the
1941 * chicken and egg problem of the SM needing to know what's on the fabric
1942 * in order to configure the fabric, and needing to configure the fabric in
1943 * order to send packets to the devices on the fabric. These directed
1944 * route packets do not need the fabric fully configured in order to reach
1945 * their destination. The SMI is the only method allowed to send
1946 * directed route packets on an InfiniBand fabric.
1948 * Return: true if the port provides an SMI.
1950 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
1952 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
1956 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
1957 * Communication Manager.
1958 * @device: Device to check
1959 * @port_num: Port number to check
1961 * The InfiniBand Communication Manager is one of many pre-defined General
1962 * Service Agents (GSA) that are accessed via the General Service
1963 * Interface (GSI). It's role is to facilitate establishment of connections
1964 * between nodes as well as other management related tasks for established
1967 * Return: true if the port supports an IB CM (this does not guarantee that
1968 * a CM is actually running however).
1970 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
1972 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
1976 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
1977 * Communication Manager.
1978 * @device: Device to check
1979 * @port_num: Port number to check
1981 * Similar to above, but specific to iWARP connections which have a different
1982 * managment protocol than InfiniBand.
1984 * Return: true if the port supports an iWARP CM (this does not guarantee that
1985 * a CM is actually running however).
1987 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
1989 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
1993 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
1994 * Subnet Administration.
1995 * @device: Device to check
1996 * @port_num: Port number to check
1998 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
1999 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2000 * fabrics, devices should resolve routes to other hosts by contacting the
2001 * SA to query the proper route.
2003 * Return: true if the port should act as a client to the fabric Subnet
2004 * Administration interface. This does not imply that the SA service is
2007 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2009 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2013 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2015 * @device: Device to check
2016 * @port_num: Port number to check
2018 * InfiniBand multicast registration is more complex than normal IPv4 or
2019 * IPv6 multicast registration. Each Host Channel Adapter must register
2020 * with the Subnet Manager when it wishes to join a multicast group. It
2021 * should do so only once regardless of how many queue pairs it subscribes
2022 * to this group. And it should leave the group only after all queue pairs
2023 * attached to the group have been detached.
2025 * Return: true if the port must undertake the additional adminstrative
2026 * overhead of registering/unregistering with the SM and tracking of the
2027 * total number of queue pairs attached to the multicast group.
2029 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2031 return rdma_cap_ib_sa(device, port_num);
2035 * rdma_cap_af_ib - Check if the port of device has the capability
2036 * Native Infiniband Address.
2037 * @device: Device to check
2038 * @port_num: Port number to check
2040 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2041 * GID. RoCE uses a different mechanism, but still generates a GID via
2042 * a prescribed mechanism and port specific data.
2044 * Return: true if the port uses a GID address to identify devices on the
2047 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2049 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2053 * rdma_cap_eth_ah - Check if the port of device has the capability
2054 * Ethernet Address Handle.
2055 * @device: Device to check
2056 * @port_num: Port number to check
2058 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2059 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2060 * port. Normally, packet headers are generated by the sending host
2061 * adapter, but when sending connectionless datagrams, we must manually
2062 * inject the proper headers for the fabric we are communicating over.
2064 * Return: true if we are running as a RoCE port and must force the
2065 * addition of a Global Route Header built from our Ethernet Address
2066 * Handle into our header list for connectionless packets.
2068 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2070 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2074 * rdma_cap_read_multi_sge - Check if the port of device has the capability
2075 * RDMA Read Multiple Scatter-Gather Entries.
2076 * @device: Device to check
2077 * @port_num: Port number to check
2079 * iWARP has a restriction that RDMA READ requests may only have a single
2080 * Scatter/Gather Entry (SGE) in the work request.
2082 * NOTE: although the linux kernel currently assumes all devices are either
2083 * single SGE RDMA READ devices or identical SGE maximums for RDMA READs and
2084 * WRITEs, according to Tom Talpey, this is not accurate. There are some
2085 * devices out there that support more than a single SGE on RDMA READ
2086 * requests, but do not support the same number of SGEs as they do on
2087 * RDMA WRITE requests. The linux kernel would need rearchitecting to
2088 * support these imbalanced READ/WRITE SGEs allowed devices. So, for now,
2089 * suffice with either the device supports the same READ/WRITE SGEs, or
2090 * it only gets one READ sge.
2092 * Return: true for any device that allows more than one SGE in RDMA READ
2095 static inline bool rdma_cap_read_multi_sge(struct ib_device *device,
2098 return !(device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP);
2102 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2105 * @port_num: Port number
2107 * This MAD size includes the MAD headers and MAD payload. No other headers
2110 * Return the max MAD size required by the Port. Will return 0 if the port
2111 * does not support MADs
2113 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2115 return device->port_immutable[port_num].max_mad_size;
2118 int ib_query_gid(struct ib_device *device,
2119 u8 port_num, int index, union ib_gid *gid);
2121 int ib_query_pkey(struct ib_device *device,
2122 u8 port_num, u16 index, u16 *pkey);
2124 int ib_modify_device(struct ib_device *device,
2125 int device_modify_mask,
2126 struct ib_device_modify *device_modify);
2128 int ib_modify_port(struct ib_device *device,
2129 u8 port_num, int port_modify_mask,
2130 struct ib_port_modify *port_modify);
2132 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2133 u8 *port_num, u16 *index);
2135 int ib_find_pkey(struct ib_device *device,
2136 u8 port_num, u16 pkey, u16 *index);
2139 * ib_alloc_pd - Allocates an unused protection domain.
2140 * @device: The device on which to allocate the protection domain.
2142 * A protection domain object provides an association between QPs, shared
2143 * receive queues, address handles, memory regions, and memory windows.
2145 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2148 * ib_dealloc_pd - Deallocates a protection domain.
2149 * @pd: The protection domain to deallocate.
2151 int ib_dealloc_pd(struct ib_pd *pd);
2154 * ib_create_ah - Creates an address handle for the given address vector.
2155 * @pd: The protection domain associated with the address handle.
2156 * @ah_attr: The attributes of the address vector.
2158 * The address handle is used to reference a local or global destination
2159 * in all UD QP post sends.
2161 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2164 * ib_init_ah_from_wc - Initializes address handle attributes from a
2166 * @device: Device on which the received message arrived.
2167 * @port_num: Port on which the received message arrived.
2168 * @wc: Work completion associated with the received message.
2169 * @grh: References the received global route header. This parameter is
2170 * ignored unless the work completion indicates that the GRH is valid.
2171 * @ah_attr: Returned attributes that can be used when creating an address
2172 * handle for replying to the message.
2174 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2175 const struct ib_wc *wc, const struct ib_grh *grh,
2176 struct ib_ah_attr *ah_attr);
2179 * ib_create_ah_from_wc - Creates an address handle associated with the
2180 * sender of the specified work completion.
2181 * @pd: The protection domain associated with the address handle.
2182 * @wc: Work completion information associated with a received message.
2183 * @grh: References the received global route header. This parameter is
2184 * ignored unless the work completion indicates that the GRH is valid.
2185 * @port_num: The outbound port number to associate with the address.
2187 * The address handle is used to reference a local or global destination
2188 * in all UD QP post sends.
2190 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2191 const struct ib_grh *grh, u8 port_num);
2194 * ib_modify_ah - Modifies the address vector associated with an address
2196 * @ah: The address handle to modify.
2197 * @ah_attr: The new address vector attributes to associate with the
2200 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2203 * ib_query_ah - Queries the address vector associated with an address
2205 * @ah: The address handle to query.
2206 * @ah_attr: The address vector attributes associated with the address
2209 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2212 * ib_destroy_ah - Destroys an address handle.
2213 * @ah: The address handle to destroy.
2215 int ib_destroy_ah(struct ib_ah *ah);
2218 * ib_create_srq - Creates a SRQ associated with the specified protection
2220 * @pd: The protection domain associated with the SRQ.
2221 * @srq_init_attr: A list of initial attributes required to create the
2222 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2223 * the actual capabilities of the created SRQ.
2225 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2226 * requested size of the SRQ, and set to the actual values allocated
2227 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2228 * will always be at least as large as the requested values.
2230 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2231 struct ib_srq_init_attr *srq_init_attr);
2234 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2235 * @srq: The SRQ to modify.
2236 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2237 * the current values of selected SRQ attributes are returned.
2238 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2239 * are being modified.
2241 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2242 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2243 * the number of receives queued drops below the limit.
2245 int ib_modify_srq(struct ib_srq *srq,
2246 struct ib_srq_attr *srq_attr,
2247 enum ib_srq_attr_mask srq_attr_mask);
2250 * ib_query_srq - Returns the attribute list and current values for the
2252 * @srq: The SRQ to query.
2253 * @srq_attr: The attributes of the specified SRQ.
2255 int ib_query_srq(struct ib_srq *srq,
2256 struct ib_srq_attr *srq_attr);
2259 * ib_destroy_srq - Destroys the specified SRQ.
2260 * @srq: The SRQ to destroy.
2262 int ib_destroy_srq(struct ib_srq *srq);
2265 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2266 * @srq: The SRQ to post the work request on.
2267 * @recv_wr: A list of work requests to post on the receive queue.
2268 * @bad_recv_wr: On an immediate failure, this parameter will reference
2269 * the work request that failed to be posted on the QP.
2271 static inline int ib_post_srq_recv(struct ib_srq *srq,
2272 struct ib_recv_wr *recv_wr,
2273 struct ib_recv_wr **bad_recv_wr)
2275 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2279 * ib_create_qp - Creates a QP associated with the specified protection
2281 * @pd: The protection domain associated with the QP.
2282 * @qp_init_attr: A list of initial attributes required to create the
2283 * QP. If QP creation succeeds, then the attributes are updated to
2284 * the actual capabilities of the created QP.
2286 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2287 struct ib_qp_init_attr *qp_init_attr);
2290 * ib_modify_qp - Modifies the attributes for the specified QP and then
2291 * transitions the QP to the given state.
2292 * @qp: The QP to modify.
2293 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2294 * the current values of selected QP attributes are returned.
2295 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2296 * are being modified.
2298 int ib_modify_qp(struct ib_qp *qp,
2299 struct ib_qp_attr *qp_attr,
2303 * ib_query_qp - Returns the attribute list and current values for the
2305 * @qp: The QP to query.
2306 * @qp_attr: The attributes of the specified QP.
2307 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2308 * @qp_init_attr: Additional attributes of the selected QP.
2310 * The qp_attr_mask may be used to limit the query to gathering only the
2311 * selected attributes.
2313 int ib_query_qp(struct ib_qp *qp,
2314 struct ib_qp_attr *qp_attr,
2316 struct ib_qp_init_attr *qp_init_attr);
2319 * ib_destroy_qp - Destroys the specified QP.
2320 * @qp: The QP to destroy.
2322 int ib_destroy_qp(struct ib_qp *qp);
2325 * ib_open_qp - Obtain a reference to an existing sharable QP.
2326 * @xrcd - XRC domain
2327 * @qp_open_attr: Attributes identifying the QP to open.
2329 * Returns a reference to a sharable QP.
2331 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2332 struct ib_qp_open_attr *qp_open_attr);
2335 * ib_close_qp - Release an external reference to a QP.
2336 * @qp: The QP handle to release
2338 * The opened QP handle is released by the caller. The underlying
2339 * shared QP is not destroyed until all internal references are released.
2341 int ib_close_qp(struct ib_qp *qp);
2344 * ib_post_send - Posts a list of work requests to the send queue of
2346 * @qp: The QP to post the work request on.
2347 * @send_wr: A list of work requests to post on the send queue.
2348 * @bad_send_wr: On an immediate failure, this parameter will reference
2349 * the work request that failed to be posted on the QP.
2351 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2352 * error is returned, the QP state shall not be affected,
2353 * ib_post_send() will return an immediate error after queueing any
2354 * earlier work requests in the list.
2356 static inline int ib_post_send(struct ib_qp *qp,
2357 struct ib_send_wr *send_wr,
2358 struct ib_send_wr **bad_send_wr)
2360 return qp->device->post_send(qp, send_wr, bad_send_wr);
2364 * ib_post_recv - Posts a list of work requests to the receive queue of
2366 * @qp: The QP to post the work request on.
2367 * @recv_wr: A list of work requests to post on the receive queue.
2368 * @bad_recv_wr: On an immediate failure, this parameter will reference
2369 * the work request that failed to be posted on the QP.
2371 static inline int ib_post_recv(struct ib_qp *qp,
2372 struct ib_recv_wr *recv_wr,
2373 struct ib_recv_wr **bad_recv_wr)
2375 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2379 * ib_create_cq - Creates a CQ on the specified device.
2380 * @device: The device on which to create the CQ.
2381 * @comp_handler: A user-specified callback that is invoked when a
2382 * completion event occurs on the CQ.
2383 * @event_handler: A user-specified callback that is invoked when an
2384 * asynchronous event not associated with a completion occurs on the CQ.
2385 * @cq_context: Context associated with the CQ returned to the user via
2386 * the associated completion and event handlers.
2387 * @cq_attr: The attributes the CQ should be created upon.
2389 * Users can examine the cq structure to determine the actual CQ size.
2391 struct ib_cq *ib_create_cq(struct ib_device *device,
2392 ib_comp_handler comp_handler,
2393 void (*event_handler)(struct ib_event *, void *),
2395 const struct ib_cq_init_attr *cq_attr);
2398 * ib_resize_cq - Modifies the capacity of the CQ.
2399 * @cq: The CQ to resize.
2400 * @cqe: The minimum size of the CQ.
2402 * Users can examine the cq structure to determine the actual CQ size.
2404 int ib_resize_cq(struct ib_cq *cq, int cqe);
2407 * ib_modify_cq - Modifies moderation params of the CQ
2408 * @cq: The CQ to modify.
2409 * @cq_count: number of CQEs that will trigger an event
2410 * @cq_period: max period of time in usec before triggering an event
2413 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2416 * ib_destroy_cq - Destroys the specified CQ.
2417 * @cq: The CQ to destroy.
2419 int ib_destroy_cq(struct ib_cq *cq);
2422 * ib_poll_cq - poll a CQ for completion(s)
2423 * @cq:the CQ being polled
2424 * @num_entries:maximum number of completions to return
2425 * @wc:array of at least @num_entries &struct ib_wc where completions
2428 * Poll a CQ for (possibly multiple) completions. If the return value
2429 * is < 0, an error occurred. If the return value is >= 0, it is the
2430 * number of completions returned. If the return value is
2431 * non-negative and < num_entries, then the CQ was emptied.
2433 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2436 return cq->device->poll_cq(cq, num_entries, wc);
2440 * ib_peek_cq - Returns the number of unreaped completions currently
2441 * on the specified CQ.
2442 * @cq: The CQ to peek.
2443 * @wc_cnt: A minimum number of unreaped completions to check for.
2445 * If the number of unreaped completions is greater than or equal to wc_cnt,
2446 * this function returns wc_cnt, otherwise, it returns the actual number of
2447 * unreaped completions.
2449 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2452 * ib_req_notify_cq - Request completion notification on a CQ.
2453 * @cq: The CQ to generate an event for.
2455 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2456 * to request an event on the next solicited event or next work
2457 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2458 * may also be |ed in to request a hint about missed events, as
2462 * < 0 means an error occurred while requesting notification
2463 * == 0 means notification was requested successfully, and if
2464 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2465 * were missed and it is safe to wait for another event. In
2466 * this case is it guaranteed that any work completions added
2467 * to the CQ since the last CQ poll will trigger a completion
2468 * notification event.
2469 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2470 * in. It means that the consumer must poll the CQ again to
2471 * make sure it is empty to avoid missing an event because of a
2472 * race between requesting notification and an entry being
2473 * added to the CQ. This return value means it is possible
2474 * (but not guaranteed) that a work completion has been added
2475 * to the CQ since the last poll without triggering a
2476 * completion notification event.
2478 static inline int ib_req_notify_cq(struct ib_cq *cq,
2479 enum ib_cq_notify_flags flags)
2481 return cq->device->req_notify_cq(cq, flags);
2485 * ib_req_ncomp_notif - Request completion notification when there are
2486 * at least the specified number of unreaped completions on the CQ.
2487 * @cq: The CQ to generate an event for.
2488 * @wc_cnt: The number of unreaped completions that should be on the
2489 * CQ before an event is generated.
2491 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2493 return cq->device->req_ncomp_notif ?
2494 cq->device->req_ncomp_notif(cq, wc_cnt) :
2499 * ib_get_dma_mr - Returns a memory region for system memory that is
2501 * @pd: The protection domain associated with the memory region.
2502 * @mr_access_flags: Specifies the memory access rights.
2504 * Note that the ib_dma_*() functions defined below must be used
2505 * to create/destroy addresses used with the Lkey or Rkey returned
2506 * by ib_get_dma_mr().
2508 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2511 * ib_dma_mapping_error - check a DMA addr for error
2512 * @dev: The device for which the dma_addr was created
2513 * @dma_addr: The DMA address to check
2515 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2518 return dev->dma_ops->mapping_error(dev, dma_addr);
2519 return dma_mapping_error(dev->dma_device, dma_addr);
2523 * ib_dma_map_single - Map a kernel virtual address to DMA address
2524 * @dev: The device for which the dma_addr is to be created
2525 * @cpu_addr: The kernel virtual address
2526 * @size: The size of the region in bytes
2527 * @direction: The direction of the DMA
2529 static inline u64 ib_dma_map_single(struct ib_device *dev,
2530 void *cpu_addr, size_t size,
2531 enum dma_data_direction direction)
2534 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2535 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2539 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2540 * @dev: The device for which the DMA address was created
2541 * @addr: The DMA address
2542 * @size: The size of the region in bytes
2543 * @direction: The direction of the DMA
2545 static inline void ib_dma_unmap_single(struct ib_device *dev,
2546 u64 addr, size_t size,
2547 enum dma_data_direction direction)
2550 dev->dma_ops->unmap_single(dev, addr, size, direction);
2552 dma_unmap_single(dev->dma_device, addr, size, direction);
2555 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2556 void *cpu_addr, size_t size,
2557 enum dma_data_direction direction,
2558 struct dma_attrs *attrs)
2560 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2564 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2565 u64 addr, size_t size,
2566 enum dma_data_direction direction,
2567 struct dma_attrs *attrs)
2569 return dma_unmap_single_attrs(dev->dma_device, addr, size,
2574 * ib_dma_map_page - Map a physical page to DMA address
2575 * @dev: The device for which the dma_addr is to be created
2576 * @page: The page to be mapped
2577 * @offset: The offset within the page
2578 * @size: The size of the region in bytes
2579 * @direction: The direction of the DMA
2581 static inline u64 ib_dma_map_page(struct ib_device *dev,
2583 unsigned long offset,
2585 enum dma_data_direction direction)
2588 return dev->dma_ops->map_page(dev, page, offset, size, direction);
2589 return dma_map_page(dev->dma_device, page, offset, size, direction);
2593 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2594 * @dev: The device for which the DMA address was created
2595 * @addr: The DMA address
2596 * @size: The size of the region in bytes
2597 * @direction: The direction of the DMA
2599 static inline void ib_dma_unmap_page(struct ib_device *dev,
2600 u64 addr, size_t size,
2601 enum dma_data_direction direction)
2604 dev->dma_ops->unmap_page(dev, addr, size, direction);
2606 dma_unmap_page(dev->dma_device, addr, size, direction);
2610 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2611 * @dev: The device for which the DMA addresses are to be created
2612 * @sg: The array of scatter/gather entries
2613 * @nents: The number of scatter/gather entries
2614 * @direction: The direction of the DMA
2616 static inline int ib_dma_map_sg(struct ib_device *dev,
2617 struct scatterlist *sg, int nents,
2618 enum dma_data_direction direction)
2621 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2622 return dma_map_sg(dev->dma_device, sg, nents, direction);
2626 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2627 * @dev: The device for which the DMA addresses were created
2628 * @sg: The array of scatter/gather entries
2629 * @nents: The number of scatter/gather entries
2630 * @direction: The direction of the DMA
2632 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2633 struct scatterlist *sg, int nents,
2634 enum dma_data_direction direction)
2637 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2639 dma_unmap_sg(dev->dma_device, sg, nents, direction);
2642 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2643 struct scatterlist *sg, int nents,
2644 enum dma_data_direction direction,
2645 struct dma_attrs *attrs)
2647 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2650 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2651 struct scatterlist *sg, int nents,
2652 enum dma_data_direction direction,
2653 struct dma_attrs *attrs)
2655 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2658 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2659 * @dev: The device for which the DMA addresses were created
2660 * @sg: The scatter/gather entry
2662 * Note: this function is obsolete. To do: change all occurrences of
2663 * ib_sg_dma_address() into sg_dma_address().
2665 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2666 struct scatterlist *sg)
2668 return sg_dma_address(sg);
2672 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2673 * @dev: The device for which the DMA addresses were created
2674 * @sg: The scatter/gather entry
2676 * Note: this function is obsolete. To do: change all occurrences of
2677 * ib_sg_dma_len() into sg_dma_len().
2679 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2680 struct scatterlist *sg)
2682 return sg_dma_len(sg);
2686 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2687 * @dev: The device for which the DMA address was created
2688 * @addr: The DMA address
2689 * @size: The size of the region in bytes
2690 * @dir: The direction of the DMA
2692 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2695 enum dma_data_direction dir)
2698 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2700 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2704 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2705 * @dev: The device for which the DMA address was created
2706 * @addr: The DMA address
2707 * @size: The size of the region in bytes
2708 * @dir: The direction of the DMA
2710 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2713 enum dma_data_direction dir)
2716 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2718 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2722 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2723 * @dev: The device for which the DMA address is requested
2724 * @size: The size of the region to allocate in bytes
2725 * @dma_handle: A pointer for returning the DMA address of the region
2726 * @flag: memory allocator flags
2728 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2734 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2739 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2740 *dma_handle = handle;
2746 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2747 * @dev: The device for which the DMA addresses were allocated
2748 * @size: The size of the region
2749 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2750 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2752 static inline void ib_dma_free_coherent(struct ib_device *dev,
2753 size_t size, void *cpu_addr,
2757 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2759 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2763 * ib_query_mr - Retrieves information about a specific memory region.
2764 * @mr: The memory region to retrieve information about.
2765 * @mr_attr: The attributes of the specified memory region.
2767 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2770 * ib_dereg_mr - Deregisters a memory region and removes it from the
2771 * HCA translation table.
2772 * @mr: The memory region to deregister.
2774 * This function can fail, if the memory region has memory windows bound to it.
2776 int ib_dereg_mr(struct ib_mr *mr);
2780 * ib_create_mr - Allocates a memory region that may be used for
2781 * signature handover operations.
2782 * @pd: The protection domain associated with the region.
2783 * @mr_init_attr: memory region init attributes.
2785 struct ib_mr *ib_create_mr(struct ib_pd *pd,
2786 struct ib_mr_init_attr *mr_init_attr);
2789 * ib_destroy_mr - Destroys a memory region that was created using
2790 * ib_create_mr and removes it from HW translation tables.
2791 * @mr: The memory region to destroy.
2793 * This function can fail, if the memory region has memory windows bound to it.
2795 int ib_destroy_mr(struct ib_mr *mr);
2798 * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2799 * IB_WR_FAST_REG_MR send work request.
2800 * @pd: The protection domain associated with the region.
2801 * @max_page_list_len: requested max physical buffer list length to be
2802 * used with fast register work requests for this MR.
2804 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2807 * ib_alloc_fast_reg_page_list - Allocates a page list array
2808 * @device - ib device pointer.
2809 * @page_list_len - size of the page list array to be allocated.
2811 * This allocates and returns a struct ib_fast_reg_page_list * and a
2812 * page_list array that is at least page_list_len in size. The actual
2813 * size is returned in max_page_list_len. The caller is responsible
2814 * for initializing the contents of the page_list array before posting
2815 * a send work request with the IB_WC_FAST_REG_MR opcode.
2817 * The page_list array entries must be translated using one of the
2818 * ib_dma_*() functions just like the addresses passed to
2819 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
2820 * ib_fast_reg_page_list must not be modified by the caller until the
2821 * IB_WC_FAST_REG_MR work request completes.
2823 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2824 struct ib_device *device, int page_list_len);
2827 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2829 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2831 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2834 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2836 * @mr - struct ib_mr pointer to be updated.
2837 * @newkey - new key to be used.
2839 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2841 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2842 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2846 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2847 * for calculating a new rkey for type 2 memory windows.
2848 * @rkey - the rkey to increment.
2850 static inline u32 ib_inc_rkey(u32 rkey)
2852 const u32 mask = 0x000000ff;
2853 return ((rkey + 1) & mask) | (rkey & ~mask);
2857 * ib_alloc_mw - Allocates a memory window.
2858 * @pd: The protection domain associated with the memory window.
2859 * @type: The type of the memory window (1 or 2).
2861 struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
2864 * ib_bind_mw - Posts a work request to the send queue of the specified
2865 * QP, which binds the memory window to the given address range and
2866 * remote access attributes.
2867 * @qp: QP to post the bind work request on.
2868 * @mw: The memory window to bind.
2869 * @mw_bind: Specifies information about the memory window, including
2870 * its address range, remote access rights, and associated memory region.
2872 * If there is no immediate error, the function will update the rkey member
2873 * of the mw parameter to its new value. The bind operation can still fail
2876 static inline int ib_bind_mw(struct ib_qp *qp,
2878 struct ib_mw_bind *mw_bind)
2880 /* XXX reference counting in corresponding MR? */
2881 return mw->device->bind_mw ?
2882 mw->device->bind_mw(qp, mw, mw_bind) :
2887 * ib_dealloc_mw - Deallocates a memory window.
2888 * @mw: The memory window to deallocate.
2890 int ib_dealloc_mw(struct ib_mw *mw);
2893 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2894 * @pd: The protection domain associated with the unmapped region.
2895 * @mr_access_flags: Specifies the memory access rights.
2896 * @fmr_attr: Attributes of the unmapped region.
2898 * A fast memory region must be mapped before it can be used as part of
2901 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2902 int mr_access_flags,
2903 struct ib_fmr_attr *fmr_attr);
2906 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2907 * @fmr: The fast memory region to associate with the pages.
2908 * @page_list: An array of physical pages to map to the fast memory region.
2909 * @list_len: The number of pages in page_list.
2910 * @iova: The I/O virtual address to use with the mapped region.
2912 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2913 u64 *page_list, int list_len,
2916 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2920 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2921 * @fmr_list: A linked list of fast memory regions to unmap.
2923 int ib_unmap_fmr(struct list_head *fmr_list);
2926 * ib_dealloc_fmr - Deallocates a fast memory region.
2927 * @fmr: The fast memory region to deallocate.
2929 int ib_dealloc_fmr(struct ib_fmr *fmr);
2932 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2933 * @qp: QP to attach to the multicast group. The QP must be type
2935 * @gid: Multicast group GID.
2936 * @lid: Multicast group LID in host byte order.
2938 * In order to send and receive multicast packets, subnet
2939 * administration must have created the multicast group and configured
2940 * the fabric appropriately. The port associated with the specified
2941 * QP must also be a member of the multicast group.
2943 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2946 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2947 * @qp: QP to detach from the multicast group.
2948 * @gid: Multicast group GID.
2949 * @lid: Multicast group LID in host byte order.
2951 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2954 * ib_alloc_xrcd - Allocates an XRC domain.
2955 * @device: The device on which to allocate the XRC domain.
2957 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2960 * ib_dealloc_xrcd - Deallocates an XRC domain.
2961 * @xrcd: The XRC domain to deallocate.
2963 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2965 struct ib_flow *ib_create_flow(struct ib_qp *qp,
2966 struct ib_flow_attr *flow_attr, int domain);
2967 int ib_destroy_flow(struct ib_flow *flow_id);
2969 static inline int ib_check_mr_access(int flags)
2972 * Local write permission is required if remote write or
2973 * remote atomic permission is also requested.
2975 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
2976 !(flags & IB_ACCESS_LOCAL_WRITE))
2983 * ib_check_mr_status: lightweight check of MR status.
2984 * This routine may provide status checks on a selected
2985 * ib_mr. first use is for signature status check.
2987 * @mr: A memory region.
2988 * @check_mask: Bitmask of which checks to perform from
2989 * ib_mr_status_check enumeration.
2990 * @mr_status: The container of relevant status checks.
2991 * failed checks will be indicated in the status bitmask
2992 * and the relevant info shall be in the error item.
2994 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
2995 struct ib_mr_status *mr_status);
2997 #endif /* IB_VERBS_H */