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 struct ib_device_attr {
178 __be64 sys_image_guid;
186 int device_cap_flags;
196 int max_qp_init_rd_atom;
197 int max_ee_init_rd_atom;
198 enum ib_atomic_cap atomic_cap;
199 enum ib_atomic_cap masked_atomic_cap;
206 int max_mcast_qp_attach;
207 int max_total_mcast_qp_attach;
214 unsigned int max_fast_reg_page_list_len;
216 u8 local_ca_ack_delay;
219 struct ib_odp_caps odp_caps;
230 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
233 case IB_MTU_256: return 256;
234 case IB_MTU_512: return 512;
235 case IB_MTU_1024: return 1024;
236 case IB_MTU_2048: return 2048;
237 case IB_MTU_4096: return 4096;
248 IB_PORT_ACTIVE_DEFER = 5
251 enum ib_port_cap_flags {
253 IB_PORT_NOTICE_SUP = 1 << 2,
254 IB_PORT_TRAP_SUP = 1 << 3,
255 IB_PORT_OPT_IPD_SUP = 1 << 4,
256 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
257 IB_PORT_SL_MAP_SUP = 1 << 6,
258 IB_PORT_MKEY_NVRAM = 1 << 7,
259 IB_PORT_PKEY_NVRAM = 1 << 8,
260 IB_PORT_LED_INFO_SUP = 1 << 9,
261 IB_PORT_SM_DISABLED = 1 << 10,
262 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
263 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
264 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
265 IB_PORT_CM_SUP = 1 << 16,
266 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
267 IB_PORT_REINIT_SUP = 1 << 18,
268 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
269 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
270 IB_PORT_DR_NOTICE_SUP = 1 << 21,
271 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
272 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
273 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
274 IB_PORT_CLIENT_REG_SUP = 1 << 25,
275 IB_PORT_IP_BASED_GIDS = 1 << 26
285 static inline int ib_width_enum_to_int(enum ib_port_width width)
288 case IB_WIDTH_1X: return 1;
289 case IB_WIDTH_4X: return 4;
290 case IB_WIDTH_8X: return 8;
291 case IB_WIDTH_12X: return 12;
305 struct ib_protocol_stats {
309 struct iw_protocol_stats {
312 u64 ipInTooBigErrors;
315 u64 ipInUnknownProtos;
316 u64 ipInTruncatedPkts;
319 u64 ipOutForwDatagrams;
351 union rdma_protocol_stats {
352 struct ib_protocol_stats ib;
353 struct iw_protocol_stats iw;
356 /* Define bits for the various functionality this port needs to be supported by
359 /* Management 0x00000FFF */
360 #define RDMA_CORE_CAP_IB_MAD 0x00000001
361 #define RDMA_CORE_CAP_IB_SMI 0x00000002
362 #define RDMA_CORE_CAP_IB_CM 0x00000004
363 #define RDMA_CORE_CAP_IW_CM 0x00000008
364 #define RDMA_CORE_CAP_IB_SA 0x00000010
366 /* Address format 0x000FF000 */
367 #define RDMA_CORE_CAP_AF_IB 0x00001000
368 #define RDMA_CORE_CAP_ETH_AH 0x00002000
370 /* Protocol 0xFFF00000 */
371 #define RDMA_CORE_CAP_PROT_IB 0x00100000
372 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
373 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
375 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
376 | RDMA_CORE_CAP_IB_MAD \
377 | RDMA_CORE_CAP_IB_SMI \
378 | RDMA_CORE_CAP_IB_CM \
379 | RDMA_CORE_CAP_IB_SA \
380 | RDMA_CORE_CAP_AF_IB)
381 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
382 | RDMA_CORE_CAP_IB_MAD \
383 | RDMA_CORE_CAP_IB_CM \
384 | RDMA_CORE_CAP_IB_SA \
385 | RDMA_CORE_CAP_AF_IB \
386 | RDMA_CORE_CAP_ETH_AH)
387 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
388 | RDMA_CORE_CAP_IW_CM)
390 struct ib_port_attr {
391 enum ib_port_state state;
393 enum ib_mtu active_mtu;
412 enum ib_device_modify_flags {
413 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
414 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
417 struct ib_device_modify {
422 enum ib_port_modify_flags {
423 IB_PORT_SHUTDOWN = 1,
424 IB_PORT_INIT_TYPE = (1<<2),
425 IB_PORT_RESET_QKEY_CNTR = (1<<3)
428 struct ib_port_modify {
429 u32 set_port_cap_mask;
430 u32 clr_port_cap_mask;
438 IB_EVENT_QP_ACCESS_ERR,
442 IB_EVENT_PATH_MIG_ERR,
443 IB_EVENT_DEVICE_FATAL,
444 IB_EVENT_PORT_ACTIVE,
447 IB_EVENT_PKEY_CHANGE,
450 IB_EVENT_SRQ_LIMIT_REACHED,
451 IB_EVENT_QP_LAST_WQE_REACHED,
452 IB_EVENT_CLIENT_REREGISTER,
456 __attribute_const__ const char *ib_event_msg(enum ib_event_type event);
459 struct ib_device *device;
466 enum ib_event_type event;
469 struct ib_event_handler {
470 struct ib_device *device;
471 void (*handler)(struct ib_event_handler *, struct ib_event *);
472 struct list_head list;
475 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
477 (_ptr)->device = _device; \
478 (_ptr)->handler = _handler; \
479 INIT_LIST_HEAD(&(_ptr)->list); \
482 struct ib_global_route {
491 __be32 version_tclass_flow;
500 IB_MULTICAST_QPN = 0xffffff
503 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
510 IB_RATE_PORT_CURRENT = 0,
511 IB_RATE_2_5_GBPS = 2,
519 IB_RATE_120_GBPS = 10,
520 IB_RATE_14_GBPS = 11,
521 IB_RATE_56_GBPS = 12,
522 IB_RATE_112_GBPS = 13,
523 IB_RATE_168_GBPS = 14,
524 IB_RATE_25_GBPS = 15,
525 IB_RATE_100_GBPS = 16,
526 IB_RATE_200_GBPS = 17,
527 IB_RATE_300_GBPS = 18
531 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
532 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
533 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
534 * @rate: rate to convert.
536 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
539 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
540 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
541 * @rate: rate to convert.
543 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
545 enum ib_mr_create_flags {
546 IB_MR_SIGNATURE_EN = 1,
550 * ib_mr_init_attr - Memory region init attributes passed to routine
552 * @max_reg_descriptors: max number of registration descriptors that
553 * may be used with registration work requests.
554 * @flags: MR creation flags bit mask.
556 struct ib_mr_init_attr {
557 int max_reg_descriptors;
563 * IB_SIG_TYPE_NONE: Unprotected.
564 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
566 enum ib_signature_type {
572 * Signature T10-DIF block-guard types
573 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
574 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
576 enum ib_t10_dif_bg_type {
582 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
584 * @bg_type: T10-DIF block guard type (CRC|CSUM)
585 * @pi_interval: protection information interval.
586 * @bg: seed of guard computation.
587 * @app_tag: application tag of guard block
588 * @ref_tag: initial guard block reference tag.
589 * @ref_remap: Indicate wethear the reftag increments each block
590 * @app_escape: Indicate to skip block check if apptag=0xffff
591 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
592 * @apptag_check_mask: check bitmask of application tag.
594 struct ib_t10_dif_domain {
595 enum ib_t10_dif_bg_type bg_type;
603 u16 apptag_check_mask;
607 * struct ib_sig_domain - Parameters for signature domain
608 * @sig_type: specific signauture type
609 * @sig: union of all signature domain attributes that may
610 * be used to set domain layout.
612 struct ib_sig_domain {
613 enum ib_signature_type sig_type;
615 struct ib_t10_dif_domain dif;
620 * struct ib_sig_attrs - Parameters for signature handover operation
621 * @check_mask: bitmask for signature byte check (8 bytes)
622 * @mem: memory domain layout desciptor.
623 * @wire: wire domain layout desciptor.
625 struct ib_sig_attrs {
627 struct ib_sig_domain mem;
628 struct ib_sig_domain wire;
631 enum ib_sig_err_type {
638 * struct ib_sig_err - signature error descriptor
641 enum ib_sig_err_type err_type;
648 enum ib_mr_status_check {
649 IB_MR_CHECK_SIG_STATUS = 1,
653 * struct ib_mr_status - Memory region status container
655 * @fail_status: Bitmask of MR checks status. For each
656 * failed check a corresponding status bit is set.
657 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
660 struct ib_mr_status {
662 struct ib_sig_err sig_err;
666 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
668 * @mult: multiple to convert.
670 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
673 struct ib_global_route grh;
688 IB_WC_LOC_EEC_OP_ERR,
693 IB_WC_LOC_ACCESS_ERR,
694 IB_WC_REM_INV_REQ_ERR,
695 IB_WC_REM_ACCESS_ERR,
698 IB_WC_RNR_RETRY_EXC_ERR,
699 IB_WC_LOC_RDD_VIOL_ERR,
700 IB_WC_REM_INV_RD_REQ_ERR,
703 IB_WC_INV_EEC_STATE_ERR,
705 IB_WC_RESP_TIMEOUT_ERR,
709 __attribute_const__ const char *ib_wc_status_msg(enum ib_wc_status status);
721 IB_WC_MASKED_COMP_SWAP,
722 IB_WC_MASKED_FETCH_ADD,
724 * Set value of IB_WC_RECV so consumers can test if a completion is a
725 * receive by testing (opcode & IB_WC_RECV).
728 IB_WC_RECV_RDMA_WITH_IMM
733 IB_WC_WITH_IMM = (1<<1),
734 IB_WC_WITH_INVALIDATE = (1<<2),
735 IB_WC_IP_CSUM_OK = (1<<3),
736 IB_WC_WITH_SMAC = (1<<4),
737 IB_WC_WITH_VLAN = (1<<5),
742 enum ib_wc_status status;
743 enum ib_wc_opcode opcode;
757 u8 port_num; /* valid only for DR SMPs on switches */
762 enum ib_cq_notify_flags {
763 IB_CQ_SOLICITED = 1 << 0,
764 IB_CQ_NEXT_COMP = 1 << 1,
765 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
766 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
774 enum ib_srq_attr_mask {
775 IB_SRQ_MAX_WR = 1 << 0,
776 IB_SRQ_LIMIT = 1 << 1,
785 struct ib_srq_init_attr {
786 void (*event_handler)(struct ib_event *, void *);
788 struct ib_srq_attr attr;
789 enum ib_srq_type srq_type;
793 struct ib_xrcd *xrcd;
814 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
815 * here (and in that order) since the MAD layer uses them as
816 * indices into a 2-entry table.
825 IB_QPT_RAW_ETHERTYPE,
826 IB_QPT_RAW_PACKET = 8,
830 /* Reserve a range for qp types internal to the low level driver.
831 * These qp types will not be visible at the IB core layer, so the
832 * IB_QPT_MAX usages should not be affected in the core layer
834 IB_QPT_RESERVED1 = 0x1000,
846 enum ib_qp_create_flags {
847 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
848 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
849 IB_QP_CREATE_NETIF_QP = 1 << 5,
850 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
851 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
852 /* reserve bits 26-31 for low level drivers' internal use */
853 IB_QP_CREATE_RESERVED_START = 1 << 26,
854 IB_QP_CREATE_RESERVED_END = 1 << 31,
859 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
860 * callback to destroy the passed in QP.
863 struct ib_qp_init_attr {
864 void (*event_handler)(struct ib_event *, void *);
866 struct ib_cq *send_cq;
867 struct ib_cq *recv_cq;
869 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
870 struct ib_qp_cap cap;
871 enum ib_sig_type sq_sig_type;
872 enum ib_qp_type qp_type;
873 enum ib_qp_create_flags create_flags;
874 u8 port_num; /* special QP types only */
877 struct ib_qp_open_attr {
878 void (*event_handler)(struct ib_event *, void *);
881 enum ib_qp_type qp_type;
884 enum ib_rnr_timeout {
885 IB_RNR_TIMER_655_36 = 0,
886 IB_RNR_TIMER_000_01 = 1,
887 IB_RNR_TIMER_000_02 = 2,
888 IB_RNR_TIMER_000_03 = 3,
889 IB_RNR_TIMER_000_04 = 4,
890 IB_RNR_TIMER_000_06 = 5,
891 IB_RNR_TIMER_000_08 = 6,
892 IB_RNR_TIMER_000_12 = 7,
893 IB_RNR_TIMER_000_16 = 8,
894 IB_RNR_TIMER_000_24 = 9,
895 IB_RNR_TIMER_000_32 = 10,
896 IB_RNR_TIMER_000_48 = 11,
897 IB_RNR_TIMER_000_64 = 12,
898 IB_RNR_TIMER_000_96 = 13,
899 IB_RNR_TIMER_001_28 = 14,
900 IB_RNR_TIMER_001_92 = 15,
901 IB_RNR_TIMER_002_56 = 16,
902 IB_RNR_TIMER_003_84 = 17,
903 IB_RNR_TIMER_005_12 = 18,
904 IB_RNR_TIMER_007_68 = 19,
905 IB_RNR_TIMER_010_24 = 20,
906 IB_RNR_TIMER_015_36 = 21,
907 IB_RNR_TIMER_020_48 = 22,
908 IB_RNR_TIMER_030_72 = 23,
909 IB_RNR_TIMER_040_96 = 24,
910 IB_RNR_TIMER_061_44 = 25,
911 IB_RNR_TIMER_081_92 = 26,
912 IB_RNR_TIMER_122_88 = 27,
913 IB_RNR_TIMER_163_84 = 28,
914 IB_RNR_TIMER_245_76 = 29,
915 IB_RNR_TIMER_327_68 = 30,
916 IB_RNR_TIMER_491_52 = 31
919 enum ib_qp_attr_mask {
921 IB_QP_CUR_STATE = (1<<1),
922 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
923 IB_QP_ACCESS_FLAGS = (1<<3),
924 IB_QP_PKEY_INDEX = (1<<4),
928 IB_QP_PATH_MTU = (1<<8),
929 IB_QP_TIMEOUT = (1<<9),
930 IB_QP_RETRY_CNT = (1<<10),
931 IB_QP_RNR_RETRY = (1<<11),
932 IB_QP_RQ_PSN = (1<<12),
933 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
934 IB_QP_ALT_PATH = (1<<14),
935 IB_QP_MIN_RNR_TIMER = (1<<15),
936 IB_QP_SQ_PSN = (1<<16),
937 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
938 IB_QP_PATH_MIG_STATE = (1<<18),
940 IB_QP_DEST_QPN = (1<<20),
941 IB_QP_SMAC = (1<<21),
942 IB_QP_ALT_SMAC = (1<<22),
944 IB_QP_ALT_VID = (1<<24),
969 enum ib_qp_state qp_state;
970 enum ib_qp_state cur_qp_state;
971 enum ib_mtu path_mtu;
972 enum ib_mig_state path_mig_state;
978 struct ib_qp_cap cap;
979 struct ib_ah_attr ah_attr;
980 struct ib_ah_attr alt_ah_attr;
983 u8 en_sqd_async_notify;
986 u8 max_dest_rd_atomic;
995 u8 alt_smac[ETH_ALEN];
1002 IB_WR_RDMA_WRITE_WITH_IMM,
1004 IB_WR_SEND_WITH_IMM,
1006 IB_WR_ATOMIC_CMP_AND_SWP,
1007 IB_WR_ATOMIC_FETCH_AND_ADD,
1009 IB_WR_SEND_WITH_INV,
1010 IB_WR_RDMA_READ_WITH_INV,
1013 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1014 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1017 /* reserve values for low level drivers' internal use.
1018 * These values will not be used at all in the ib core layer.
1020 IB_WR_RESERVED1 = 0xf0,
1032 enum ib_send_flags {
1034 IB_SEND_SIGNALED = (1<<1),
1035 IB_SEND_SOLICITED = (1<<2),
1036 IB_SEND_INLINE = (1<<3),
1037 IB_SEND_IP_CSUM = (1<<4),
1039 /* reserve bits 26-31 for low level drivers' internal use */
1040 IB_SEND_RESERVED_START = (1 << 26),
1041 IB_SEND_RESERVED_END = (1 << 31),
1050 struct ib_fast_reg_page_list {
1051 struct ib_device *device;
1053 unsigned int max_page_list_len;
1057 * struct ib_mw_bind_info - Parameters for a memory window bind operation.
1058 * @mr: A memory region to bind the memory window to.
1059 * @addr: The address where the memory window should begin.
1060 * @length: The length of the memory window, in bytes.
1061 * @mw_access_flags: Access flags from enum ib_access_flags for the window.
1063 * This struct contains the shared parameters for type 1 and type 2
1064 * memory window bind operations.
1066 struct ib_mw_bind_info {
1070 int mw_access_flags;
1074 struct ib_send_wr *next;
1076 struct ib_sge *sg_list;
1078 enum ib_wr_opcode opcode;
1082 u32 invalidate_rkey;
1093 u64 compare_add_mask;
1104 u16 pkey_index; /* valid for GSI only */
1105 u8 port_num; /* valid for DR SMPs on switch only */
1109 struct ib_fast_reg_page_list *page_list;
1110 unsigned int page_shift;
1111 unsigned int page_list_len;
1118 /* The new rkey for the memory window. */
1120 struct ib_mw_bind_info bind_info;
1123 struct ib_sig_attrs *sig_attrs;
1124 struct ib_mr *sig_mr;
1126 struct ib_sge *prot;
1129 u32 xrc_remote_srq_num; /* XRC TGT QPs only */
1133 struct ib_recv_wr *next;
1135 struct ib_sge *sg_list;
1139 enum ib_access_flags {
1140 IB_ACCESS_LOCAL_WRITE = 1,
1141 IB_ACCESS_REMOTE_WRITE = (1<<1),
1142 IB_ACCESS_REMOTE_READ = (1<<2),
1143 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1144 IB_ACCESS_MW_BIND = (1<<4),
1145 IB_ZERO_BASED = (1<<5),
1146 IB_ACCESS_ON_DEMAND = (1<<6),
1149 struct ib_phys_buf {
1156 u64 device_virt_addr;
1158 int mr_access_flags;
1163 enum ib_mr_rereg_flags {
1164 IB_MR_REREG_TRANS = 1,
1165 IB_MR_REREG_PD = (1<<1),
1166 IB_MR_REREG_ACCESS = (1<<2),
1167 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1171 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
1172 * @wr_id: Work request id.
1173 * @send_flags: Flags from ib_send_flags enum.
1174 * @bind_info: More parameters of the bind operation.
1179 struct ib_mw_bind_info bind_info;
1182 struct ib_fmr_attr {
1190 struct ib_ucontext {
1191 struct ib_device *device;
1192 struct list_head pd_list;
1193 struct list_head mr_list;
1194 struct list_head mw_list;
1195 struct list_head cq_list;
1196 struct list_head qp_list;
1197 struct list_head srq_list;
1198 struct list_head ah_list;
1199 struct list_head xrcd_list;
1200 struct list_head rule_list;
1204 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1205 struct rb_root umem_tree;
1207 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1208 * mmu notifiers registration.
1210 struct rw_semaphore umem_rwsem;
1211 void (*invalidate_range)(struct ib_umem *umem,
1212 unsigned long start, unsigned long end);
1214 struct mmu_notifier mn;
1215 atomic_t notifier_count;
1216 /* A list of umems that don't have private mmu notifier counters yet. */
1217 struct list_head no_private_counters;
1223 u64 user_handle; /* handle given to us by userspace */
1224 struct ib_ucontext *context; /* associated user context */
1225 void *object; /* containing object */
1226 struct list_head list; /* link to context's list */
1227 int id; /* index into kernel idr */
1229 struct rw_semaphore mutex; /* protects .live */
1234 const void __user *inbuf;
1235 void __user *outbuf;
1241 struct ib_device *device;
1242 struct ib_uobject *uobject;
1243 atomic_t usecnt; /* count all resources */
1247 struct ib_device *device;
1248 atomic_t usecnt; /* count all exposed resources */
1249 struct inode *inode;
1251 struct mutex tgt_qp_mutex;
1252 struct list_head tgt_qp_list;
1256 struct ib_device *device;
1258 struct ib_uobject *uobject;
1261 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1264 struct ib_device *device;
1265 struct ib_uobject *uobject;
1266 ib_comp_handler comp_handler;
1267 void (*event_handler)(struct ib_event *, void *);
1270 atomic_t usecnt; /* count number of work queues */
1274 struct ib_device *device;
1276 struct ib_uobject *uobject;
1277 void (*event_handler)(struct ib_event *, void *);
1279 enum ib_srq_type srq_type;
1284 struct ib_xrcd *xrcd;
1292 struct ib_device *device;
1294 struct ib_cq *send_cq;
1295 struct ib_cq *recv_cq;
1297 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1298 struct list_head xrcd_list;
1299 /* count times opened, mcast attaches, flow attaches */
1301 struct list_head open_list;
1302 struct ib_qp *real_qp;
1303 struct ib_uobject *uobject;
1304 void (*event_handler)(struct ib_event *, void *);
1307 enum ib_qp_type qp_type;
1311 struct ib_device *device;
1313 struct ib_uobject *uobject;
1316 atomic_t usecnt; /* count number of MWs */
1320 struct ib_device *device;
1322 struct ib_uobject *uobject;
1324 enum ib_mw_type type;
1328 struct ib_device *device;
1330 struct list_head list;
1335 /* Supported steering options */
1336 enum ib_flow_attr_type {
1337 /* steering according to rule specifications */
1338 IB_FLOW_ATTR_NORMAL = 0x0,
1339 /* default unicast and multicast rule -
1340 * receive all Eth traffic which isn't steered to any QP
1342 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1343 /* default multicast rule -
1344 * receive all Eth multicast traffic which isn't steered to any QP
1346 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1347 /* sniffer rule - receive all port traffic */
1348 IB_FLOW_ATTR_SNIFFER = 0x3
1351 /* Supported steering header types */
1352 enum ib_flow_spec_type {
1354 IB_FLOW_SPEC_ETH = 0x20,
1355 IB_FLOW_SPEC_IB = 0x22,
1357 IB_FLOW_SPEC_IPV4 = 0x30,
1359 IB_FLOW_SPEC_TCP = 0x40,
1360 IB_FLOW_SPEC_UDP = 0x41
1362 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1363 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1365 /* Flow steering rule priority is set according to it's domain.
1366 * Lower domain value means higher priority.
1368 enum ib_flow_domain {
1369 IB_FLOW_DOMAIN_USER,
1370 IB_FLOW_DOMAIN_ETHTOOL,
1373 IB_FLOW_DOMAIN_NUM /* Must be last */
1376 struct ib_flow_eth_filter {
1383 struct ib_flow_spec_eth {
1384 enum ib_flow_spec_type type;
1386 struct ib_flow_eth_filter val;
1387 struct ib_flow_eth_filter mask;
1390 struct ib_flow_ib_filter {
1395 struct ib_flow_spec_ib {
1396 enum ib_flow_spec_type type;
1398 struct ib_flow_ib_filter val;
1399 struct ib_flow_ib_filter mask;
1402 struct ib_flow_ipv4_filter {
1407 struct ib_flow_spec_ipv4 {
1408 enum ib_flow_spec_type type;
1410 struct ib_flow_ipv4_filter val;
1411 struct ib_flow_ipv4_filter mask;
1414 struct ib_flow_tcp_udp_filter {
1419 struct ib_flow_spec_tcp_udp {
1420 enum ib_flow_spec_type type;
1422 struct ib_flow_tcp_udp_filter val;
1423 struct ib_flow_tcp_udp_filter mask;
1426 union ib_flow_spec {
1428 enum ib_flow_spec_type type;
1431 struct ib_flow_spec_eth eth;
1432 struct ib_flow_spec_ib ib;
1433 struct ib_flow_spec_ipv4 ipv4;
1434 struct ib_flow_spec_tcp_udp tcp_udp;
1437 struct ib_flow_attr {
1438 enum ib_flow_attr_type type;
1444 /* Following are the optional layers according to user request
1445 * struct ib_flow_spec_xxx
1446 * struct ib_flow_spec_yyy
1452 struct ib_uobject *uobject;
1458 enum ib_process_mad_flags {
1459 IB_MAD_IGNORE_MKEY = 1,
1460 IB_MAD_IGNORE_BKEY = 2,
1461 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1464 enum ib_mad_result {
1465 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1466 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1467 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1468 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1471 #define IB_DEVICE_NAME_MAX 64
1475 struct ib_event_handler event_handler;
1476 struct ib_pkey_cache **pkey_cache;
1477 struct ib_gid_cache **gid_cache;
1481 struct ib_dma_mapping_ops {
1482 int (*mapping_error)(struct ib_device *dev,
1484 u64 (*map_single)(struct ib_device *dev,
1485 void *ptr, size_t size,
1486 enum dma_data_direction direction);
1487 void (*unmap_single)(struct ib_device *dev,
1488 u64 addr, size_t size,
1489 enum dma_data_direction direction);
1490 u64 (*map_page)(struct ib_device *dev,
1491 struct page *page, unsigned long offset,
1493 enum dma_data_direction direction);
1494 void (*unmap_page)(struct ib_device *dev,
1495 u64 addr, size_t size,
1496 enum dma_data_direction direction);
1497 int (*map_sg)(struct ib_device *dev,
1498 struct scatterlist *sg, int nents,
1499 enum dma_data_direction direction);
1500 void (*unmap_sg)(struct ib_device *dev,
1501 struct scatterlist *sg, int nents,
1502 enum dma_data_direction direction);
1503 void (*sync_single_for_cpu)(struct ib_device *dev,
1506 enum dma_data_direction dir);
1507 void (*sync_single_for_device)(struct ib_device *dev,
1510 enum dma_data_direction dir);
1511 void *(*alloc_coherent)(struct ib_device *dev,
1515 void (*free_coherent)(struct ib_device *dev,
1516 size_t size, void *cpu_addr,
1522 struct ib_port_immutable {
1529 struct device *dma_device;
1531 char name[IB_DEVICE_NAME_MAX];
1533 struct list_head event_handler_list;
1534 spinlock_t event_handler_lock;
1536 spinlock_t client_data_lock;
1537 struct list_head core_list;
1538 struct list_head client_data_list;
1540 struct ib_cache cache;
1542 * port_immutable is indexed by port number
1544 struct ib_port_immutable *port_immutable;
1546 int num_comp_vectors;
1548 struct iw_cm_verbs *iwcm;
1550 int (*get_protocol_stats)(struct ib_device *device,
1551 union rdma_protocol_stats *stats);
1552 int (*query_device)(struct ib_device *device,
1553 struct ib_device_attr *device_attr);
1554 int (*query_port)(struct ib_device *device,
1556 struct ib_port_attr *port_attr);
1557 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1559 int (*query_gid)(struct ib_device *device,
1560 u8 port_num, int index,
1562 int (*query_pkey)(struct ib_device *device,
1563 u8 port_num, u16 index, u16 *pkey);
1564 int (*modify_device)(struct ib_device *device,
1565 int device_modify_mask,
1566 struct ib_device_modify *device_modify);
1567 int (*modify_port)(struct ib_device *device,
1568 u8 port_num, int port_modify_mask,
1569 struct ib_port_modify *port_modify);
1570 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1571 struct ib_udata *udata);
1572 int (*dealloc_ucontext)(struct ib_ucontext *context);
1573 int (*mmap)(struct ib_ucontext *context,
1574 struct vm_area_struct *vma);
1575 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1576 struct ib_ucontext *context,
1577 struct ib_udata *udata);
1578 int (*dealloc_pd)(struct ib_pd *pd);
1579 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1580 struct ib_ah_attr *ah_attr);
1581 int (*modify_ah)(struct ib_ah *ah,
1582 struct ib_ah_attr *ah_attr);
1583 int (*query_ah)(struct ib_ah *ah,
1584 struct ib_ah_attr *ah_attr);
1585 int (*destroy_ah)(struct ib_ah *ah);
1586 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1587 struct ib_srq_init_attr *srq_init_attr,
1588 struct ib_udata *udata);
1589 int (*modify_srq)(struct ib_srq *srq,
1590 struct ib_srq_attr *srq_attr,
1591 enum ib_srq_attr_mask srq_attr_mask,
1592 struct ib_udata *udata);
1593 int (*query_srq)(struct ib_srq *srq,
1594 struct ib_srq_attr *srq_attr);
1595 int (*destroy_srq)(struct ib_srq *srq);
1596 int (*post_srq_recv)(struct ib_srq *srq,
1597 struct ib_recv_wr *recv_wr,
1598 struct ib_recv_wr **bad_recv_wr);
1599 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1600 struct ib_qp_init_attr *qp_init_attr,
1601 struct ib_udata *udata);
1602 int (*modify_qp)(struct ib_qp *qp,
1603 struct ib_qp_attr *qp_attr,
1605 struct ib_udata *udata);
1606 int (*query_qp)(struct ib_qp *qp,
1607 struct ib_qp_attr *qp_attr,
1609 struct ib_qp_init_attr *qp_init_attr);
1610 int (*destroy_qp)(struct ib_qp *qp);
1611 int (*post_send)(struct ib_qp *qp,
1612 struct ib_send_wr *send_wr,
1613 struct ib_send_wr **bad_send_wr);
1614 int (*post_recv)(struct ib_qp *qp,
1615 struct ib_recv_wr *recv_wr,
1616 struct ib_recv_wr **bad_recv_wr);
1617 struct ib_cq * (*create_cq)(struct ib_device *device, int cqe,
1619 struct ib_ucontext *context,
1620 struct ib_udata *udata);
1621 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1623 int (*destroy_cq)(struct ib_cq *cq);
1624 int (*resize_cq)(struct ib_cq *cq, int cqe,
1625 struct ib_udata *udata);
1626 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1628 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1629 int (*req_notify_cq)(struct ib_cq *cq,
1630 enum ib_cq_notify_flags flags);
1631 int (*req_ncomp_notif)(struct ib_cq *cq,
1633 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1634 int mr_access_flags);
1635 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
1636 struct ib_phys_buf *phys_buf_array,
1638 int mr_access_flags,
1640 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1641 u64 start, u64 length,
1643 int mr_access_flags,
1644 struct ib_udata *udata);
1645 int (*rereg_user_mr)(struct ib_mr *mr,
1647 u64 start, u64 length,
1649 int mr_access_flags,
1651 struct ib_udata *udata);
1652 int (*query_mr)(struct ib_mr *mr,
1653 struct ib_mr_attr *mr_attr);
1654 int (*dereg_mr)(struct ib_mr *mr);
1655 int (*destroy_mr)(struct ib_mr *mr);
1656 struct ib_mr * (*create_mr)(struct ib_pd *pd,
1657 struct ib_mr_init_attr *mr_init_attr);
1658 struct ib_mr * (*alloc_fast_reg_mr)(struct ib_pd *pd,
1659 int max_page_list_len);
1660 struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1662 void (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1663 int (*rereg_phys_mr)(struct ib_mr *mr,
1666 struct ib_phys_buf *phys_buf_array,
1668 int mr_access_flags,
1670 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1671 enum ib_mw_type type);
1672 int (*bind_mw)(struct ib_qp *qp,
1674 struct ib_mw_bind *mw_bind);
1675 int (*dealloc_mw)(struct ib_mw *mw);
1676 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1677 int mr_access_flags,
1678 struct ib_fmr_attr *fmr_attr);
1679 int (*map_phys_fmr)(struct ib_fmr *fmr,
1680 u64 *page_list, int list_len,
1682 int (*unmap_fmr)(struct list_head *fmr_list);
1683 int (*dealloc_fmr)(struct ib_fmr *fmr);
1684 int (*attach_mcast)(struct ib_qp *qp,
1687 int (*detach_mcast)(struct ib_qp *qp,
1690 int (*process_mad)(struct ib_device *device,
1691 int process_mad_flags,
1693 const struct ib_wc *in_wc,
1694 const struct ib_grh *in_grh,
1695 const struct ib_mad *in_mad,
1696 struct ib_mad *out_mad);
1697 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1698 struct ib_ucontext *ucontext,
1699 struct ib_udata *udata);
1700 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1701 struct ib_flow * (*create_flow)(struct ib_qp *qp,
1705 int (*destroy_flow)(struct ib_flow *flow_id);
1706 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1707 struct ib_mr_status *mr_status);
1709 struct ib_dma_mapping_ops *dma_ops;
1711 struct module *owner;
1713 struct kobject *ports_parent;
1714 struct list_head port_list;
1717 IB_DEV_UNINITIALIZED,
1723 u64 uverbs_cmd_mask;
1724 u64 uverbs_ex_cmd_mask;
1733 * The following mandatory functions are used only at device
1734 * registration. Keep functions such as these at the end of this
1735 * structure to avoid cache line misses when accessing struct ib_device
1738 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1743 void (*add) (struct ib_device *);
1744 void (*remove)(struct ib_device *);
1746 struct list_head list;
1749 struct ib_device *ib_alloc_device(size_t size);
1750 void ib_dealloc_device(struct ib_device *device);
1752 int ib_register_device(struct ib_device *device,
1753 int (*port_callback)(struct ib_device *,
1754 u8, struct kobject *));
1755 void ib_unregister_device(struct ib_device *device);
1757 int ib_register_client (struct ib_client *client);
1758 void ib_unregister_client(struct ib_client *client);
1760 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1761 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1764 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1766 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1769 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1771 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1775 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1776 * contains all required attributes and no attributes not allowed for
1777 * the given QP state transition.
1778 * @cur_state: Current QP state
1779 * @next_state: Next QP state
1781 * @mask: Mask of supplied QP attributes
1782 * @ll : link layer of port
1784 * This function is a helper function that a low-level driver's
1785 * modify_qp method can use to validate the consumer's input. It
1786 * checks that cur_state and next_state are valid QP states, that a
1787 * transition from cur_state to next_state is allowed by the IB spec,
1788 * and that the attribute mask supplied is allowed for the transition.
1790 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1791 enum ib_qp_type type, enum ib_qp_attr_mask mask,
1792 enum rdma_link_layer ll);
1794 int ib_register_event_handler (struct ib_event_handler *event_handler);
1795 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1796 void ib_dispatch_event(struct ib_event *event);
1798 int ib_query_device(struct ib_device *device,
1799 struct ib_device_attr *device_attr);
1801 int ib_query_port(struct ib_device *device,
1802 u8 port_num, struct ib_port_attr *port_attr);
1804 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1808 * rdma_start_port - Return the first valid port number for the device
1811 * @device: Device to be checked
1813 * Return start port number
1815 static inline u8 rdma_start_port(const struct ib_device *device)
1817 return (device->node_type == RDMA_NODE_IB_SWITCH) ? 0 : 1;
1821 * rdma_end_port - Return the last valid port number for the device
1824 * @device: Device to be checked
1826 * Return last port number
1828 static inline u8 rdma_end_port(const struct ib_device *device)
1830 return (device->node_type == RDMA_NODE_IB_SWITCH) ?
1831 0 : device->phys_port_cnt;
1834 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
1836 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
1839 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
1841 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
1844 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
1846 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
1849 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
1851 return device->port_immutable[port_num].core_cap_flags &
1852 (RDMA_CORE_CAP_PROT_IB | RDMA_CORE_CAP_PROT_ROCE);
1856 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
1857 * Management Datagrams.
1858 * @device: Device to check
1859 * @port_num: Port number to check
1861 * Management Datagrams (MAD) are a required part of the InfiniBand
1862 * specification and are supported on all InfiniBand devices. A slightly
1863 * extended version are also supported on OPA interfaces.
1865 * Return: true if the port supports sending/receiving of MAD packets.
1867 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
1869 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
1873 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
1874 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
1875 * @device: Device to check
1876 * @port_num: Port number to check
1878 * Each InfiniBand node is required to provide a Subnet Management Agent
1879 * that the subnet manager can access. Prior to the fabric being fully
1880 * configured by the subnet manager, the SMA is accessed via a well known
1881 * interface called the Subnet Management Interface (SMI). This interface
1882 * uses directed route packets to communicate with the SM to get around the
1883 * chicken and egg problem of the SM needing to know what's on the fabric
1884 * in order to configure the fabric, and needing to configure the fabric in
1885 * order to send packets to the devices on the fabric. These directed
1886 * route packets do not need the fabric fully configured in order to reach
1887 * their destination. The SMI is the only method allowed to send
1888 * directed route packets on an InfiniBand fabric.
1890 * Return: true if the port provides an SMI.
1892 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
1894 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
1898 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
1899 * Communication Manager.
1900 * @device: Device to check
1901 * @port_num: Port number to check
1903 * The InfiniBand Communication Manager is one of many pre-defined General
1904 * Service Agents (GSA) that are accessed via the General Service
1905 * Interface (GSI). It's role is to facilitate establishment of connections
1906 * between nodes as well as other management related tasks for established
1909 * Return: true if the port supports an IB CM (this does not guarantee that
1910 * a CM is actually running however).
1912 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
1914 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
1918 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
1919 * Communication Manager.
1920 * @device: Device to check
1921 * @port_num: Port number to check
1923 * Similar to above, but specific to iWARP connections which have a different
1924 * managment protocol than InfiniBand.
1926 * Return: true if the port supports an iWARP CM (this does not guarantee that
1927 * a CM is actually running however).
1929 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
1931 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
1935 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
1936 * Subnet Administration.
1937 * @device: Device to check
1938 * @port_num: Port number to check
1940 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
1941 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
1942 * fabrics, devices should resolve routes to other hosts by contacting the
1943 * SA to query the proper route.
1945 * Return: true if the port should act as a client to the fabric Subnet
1946 * Administration interface. This does not imply that the SA service is
1949 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
1951 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
1955 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
1957 * @device: Device to check
1958 * @port_num: Port number to check
1960 * InfiniBand multicast registration is more complex than normal IPv4 or
1961 * IPv6 multicast registration. Each Host Channel Adapter must register
1962 * with the Subnet Manager when it wishes to join a multicast group. It
1963 * should do so only once regardless of how many queue pairs it subscribes
1964 * to this group. And it should leave the group only after all queue pairs
1965 * attached to the group have been detached.
1967 * Return: true if the port must undertake the additional adminstrative
1968 * overhead of registering/unregistering with the SM and tracking of the
1969 * total number of queue pairs attached to the multicast group.
1971 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
1973 return rdma_cap_ib_sa(device, port_num);
1977 * rdma_cap_af_ib - Check if the port of device has the capability
1978 * Native Infiniband Address.
1979 * @device: Device to check
1980 * @port_num: Port number to check
1982 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
1983 * GID. RoCE uses a different mechanism, but still generates a GID via
1984 * a prescribed mechanism and port specific data.
1986 * Return: true if the port uses a GID address to identify devices on the
1989 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
1991 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
1995 * rdma_cap_eth_ah - Check if the port of device has the capability
1996 * Ethernet Address Handle.
1997 * @device: Device to check
1998 * @port_num: Port number to check
2000 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2001 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2002 * port. Normally, packet headers are generated by the sending host
2003 * adapter, but when sending connectionless datagrams, we must manually
2004 * inject the proper headers for the fabric we are communicating over.
2006 * Return: true if we are running as a RoCE port and must force the
2007 * addition of a Global Route Header built from our Ethernet Address
2008 * Handle into our header list for connectionless packets.
2010 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2012 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2016 * rdma_cap_read_multi_sge - Check if the port of device has the capability
2017 * RDMA Read Multiple Scatter-Gather Entries.
2018 * @device: Device to check
2019 * @port_num: Port number to check
2021 * iWARP has a restriction that RDMA READ requests may only have a single
2022 * Scatter/Gather Entry (SGE) in the work request.
2024 * NOTE: although the linux kernel currently assumes all devices are either
2025 * single SGE RDMA READ devices or identical SGE maximums for RDMA READs and
2026 * WRITEs, according to Tom Talpey, this is not accurate. There are some
2027 * devices out there that support more than a single SGE on RDMA READ
2028 * requests, but do not support the same number of SGEs as they do on
2029 * RDMA WRITE requests. The linux kernel would need rearchitecting to
2030 * support these imbalanced READ/WRITE SGEs allowed devices. So, for now,
2031 * suffice with either the device supports the same READ/WRITE SGEs, or
2032 * it only gets one READ sge.
2034 * Return: true for any device that allows more than one SGE in RDMA READ
2037 static inline bool rdma_cap_read_multi_sge(struct ib_device *device,
2040 return !(device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP);
2043 int ib_query_gid(struct ib_device *device,
2044 u8 port_num, int index, union ib_gid *gid);
2046 int ib_query_pkey(struct ib_device *device,
2047 u8 port_num, u16 index, u16 *pkey);
2049 int ib_modify_device(struct ib_device *device,
2050 int device_modify_mask,
2051 struct ib_device_modify *device_modify);
2053 int ib_modify_port(struct ib_device *device,
2054 u8 port_num, int port_modify_mask,
2055 struct ib_port_modify *port_modify);
2057 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2058 u8 *port_num, u16 *index);
2060 int ib_find_pkey(struct ib_device *device,
2061 u8 port_num, u16 pkey, u16 *index);
2064 * ib_alloc_pd - Allocates an unused protection domain.
2065 * @device: The device on which to allocate the protection domain.
2067 * A protection domain object provides an association between QPs, shared
2068 * receive queues, address handles, memory regions, and memory windows.
2070 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2073 * ib_dealloc_pd - Deallocates a protection domain.
2074 * @pd: The protection domain to deallocate.
2076 int ib_dealloc_pd(struct ib_pd *pd);
2079 * ib_create_ah - Creates an address handle for the given address vector.
2080 * @pd: The protection domain associated with the address handle.
2081 * @ah_attr: The attributes of the address vector.
2083 * The address handle is used to reference a local or global destination
2084 * in all UD QP post sends.
2086 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2089 * ib_init_ah_from_wc - Initializes address handle attributes from a
2091 * @device: Device on which the received message arrived.
2092 * @port_num: Port on which the received message arrived.
2093 * @wc: Work completion associated with the received message.
2094 * @grh: References the received global route header. This parameter is
2095 * ignored unless the work completion indicates that the GRH is valid.
2096 * @ah_attr: Returned attributes that can be used when creating an address
2097 * handle for replying to the message.
2099 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
2100 struct ib_grh *grh, struct ib_ah_attr *ah_attr);
2103 * ib_create_ah_from_wc - Creates an address handle associated with the
2104 * sender of the specified work completion.
2105 * @pd: The protection domain associated with the address handle.
2106 * @wc: Work completion information associated with a received message.
2107 * @grh: References the received global route header. This parameter is
2108 * ignored unless the work completion indicates that the GRH is valid.
2109 * @port_num: The outbound port number to associate with the address.
2111 * The address handle is used to reference a local or global destination
2112 * in all UD QP post sends.
2114 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
2115 struct ib_grh *grh, u8 port_num);
2118 * ib_modify_ah - Modifies the address vector associated with an address
2120 * @ah: The address handle to modify.
2121 * @ah_attr: The new address vector attributes to associate with the
2124 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2127 * ib_query_ah - Queries the address vector associated with an address
2129 * @ah: The address handle to query.
2130 * @ah_attr: The address vector attributes associated with the address
2133 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2136 * ib_destroy_ah - Destroys an address handle.
2137 * @ah: The address handle to destroy.
2139 int ib_destroy_ah(struct ib_ah *ah);
2142 * ib_create_srq - Creates a SRQ associated with the specified protection
2144 * @pd: The protection domain associated with the SRQ.
2145 * @srq_init_attr: A list of initial attributes required to create the
2146 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2147 * the actual capabilities of the created SRQ.
2149 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2150 * requested size of the SRQ, and set to the actual values allocated
2151 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2152 * will always be at least as large as the requested values.
2154 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2155 struct ib_srq_init_attr *srq_init_attr);
2158 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2159 * @srq: The SRQ to modify.
2160 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2161 * the current values of selected SRQ attributes are returned.
2162 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2163 * are being modified.
2165 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2166 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2167 * the number of receives queued drops below the limit.
2169 int ib_modify_srq(struct ib_srq *srq,
2170 struct ib_srq_attr *srq_attr,
2171 enum ib_srq_attr_mask srq_attr_mask);
2174 * ib_query_srq - Returns the attribute list and current values for the
2176 * @srq: The SRQ to query.
2177 * @srq_attr: The attributes of the specified SRQ.
2179 int ib_query_srq(struct ib_srq *srq,
2180 struct ib_srq_attr *srq_attr);
2183 * ib_destroy_srq - Destroys the specified SRQ.
2184 * @srq: The SRQ to destroy.
2186 int ib_destroy_srq(struct ib_srq *srq);
2189 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2190 * @srq: The SRQ to post the work request on.
2191 * @recv_wr: A list of work requests to post on the receive queue.
2192 * @bad_recv_wr: On an immediate failure, this parameter will reference
2193 * the work request that failed to be posted on the QP.
2195 static inline int ib_post_srq_recv(struct ib_srq *srq,
2196 struct ib_recv_wr *recv_wr,
2197 struct ib_recv_wr **bad_recv_wr)
2199 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2203 * ib_create_qp - Creates a QP associated with the specified protection
2205 * @pd: The protection domain associated with the QP.
2206 * @qp_init_attr: A list of initial attributes required to create the
2207 * QP. If QP creation succeeds, then the attributes are updated to
2208 * the actual capabilities of the created QP.
2210 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2211 struct ib_qp_init_attr *qp_init_attr);
2214 * ib_modify_qp - Modifies the attributes for the specified QP and then
2215 * transitions the QP to the given state.
2216 * @qp: The QP to modify.
2217 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2218 * the current values of selected QP attributes are returned.
2219 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2220 * are being modified.
2222 int ib_modify_qp(struct ib_qp *qp,
2223 struct ib_qp_attr *qp_attr,
2227 * ib_query_qp - Returns the attribute list and current values for the
2229 * @qp: The QP to query.
2230 * @qp_attr: The attributes of the specified QP.
2231 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2232 * @qp_init_attr: Additional attributes of the selected QP.
2234 * The qp_attr_mask may be used to limit the query to gathering only the
2235 * selected attributes.
2237 int ib_query_qp(struct ib_qp *qp,
2238 struct ib_qp_attr *qp_attr,
2240 struct ib_qp_init_attr *qp_init_attr);
2243 * ib_destroy_qp - Destroys the specified QP.
2244 * @qp: The QP to destroy.
2246 int ib_destroy_qp(struct ib_qp *qp);
2249 * ib_open_qp - Obtain a reference to an existing sharable QP.
2250 * @xrcd - XRC domain
2251 * @qp_open_attr: Attributes identifying the QP to open.
2253 * Returns a reference to a sharable QP.
2255 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2256 struct ib_qp_open_attr *qp_open_attr);
2259 * ib_close_qp - Release an external reference to a QP.
2260 * @qp: The QP handle to release
2262 * The opened QP handle is released by the caller. The underlying
2263 * shared QP is not destroyed until all internal references are released.
2265 int ib_close_qp(struct ib_qp *qp);
2268 * ib_post_send - Posts a list of work requests to the send queue of
2270 * @qp: The QP to post the work request on.
2271 * @send_wr: A list of work requests to post on the send queue.
2272 * @bad_send_wr: On an immediate failure, this parameter will reference
2273 * the work request that failed to be posted on the QP.
2275 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2276 * error is returned, the QP state shall not be affected,
2277 * ib_post_send() will return an immediate error after queueing any
2278 * earlier work requests in the list.
2280 static inline int ib_post_send(struct ib_qp *qp,
2281 struct ib_send_wr *send_wr,
2282 struct ib_send_wr **bad_send_wr)
2284 return qp->device->post_send(qp, send_wr, bad_send_wr);
2288 * ib_post_recv - Posts a list of work requests to the receive queue of
2290 * @qp: The QP to post the work request on.
2291 * @recv_wr: A list of work requests to post on the receive queue.
2292 * @bad_recv_wr: On an immediate failure, this parameter will reference
2293 * the work request that failed to be posted on the QP.
2295 static inline int ib_post_recv(struct ib_qp *qp,
2296 struct ib_recv_wr *recv_wr,
2297 struct ib_recv_wr **bad_recv_wr)
2299 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2303 * ib_create_cq - Creates a CQ on the specified device.
2304 * @device: The device on which to create the CQ.
2305 * @comp_handler: A user-specified callback that is invoked when a
2306 * completion event occurs on the CQ.
2307 * @event_handler: A user-specified callback that is invoked when an
2308 * asynchronous event not associated with a completion occurs on the CQ.
2309 * @cq_context: Context associated with the CQ returned to the user via
2310 * the associated completion and event handlers.
2311 * @cqe: The minimum size of the CQ.
2312 * @comp_vector - Completion vector used to signal completion events.
2313 * Must be >= 0 and < context->num_comp_vectors.
2315 * Users can examine the cq structure to determine the actual CQ size.
2317 struct ib_cq *ib_create_cq(struct ib_device *device,
2318 ib_comp_handler comp_handler,
2319 void (*event_handler)(struct ib_event *, void *),
2320 void *cq_context, int cqe, int comp_vector);
2323 * ib_resize_cq - Modifies the capacity of the CQ.
2324 * @cq: The CQ to resize.
2325 * @cqe: The minimum size of the CQ.
2327 * Users can examine the cq structure to determine the actual CQ size.
2329 int ib_resize_cq(struct ib_cq *cq, int cqe);
2332 * ib_modify_cq - Modifies moderation params of the CQ
2333 * @cq: The CQ to modify.
2334 * @cq_count: number of CQEs that will trigger an event
2335 * @cq_period: max period of time in usec before triggering an event
2338 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2341 * ib_destroy_cq - Destroys the specified CQ.
2342 * @cq: The CQ to destroy.
2344 int ib_destroy_cq(struct ib_cq *cq);
2347 * ib_poll_cq - poll a CQ for completion(s)
2348 * @cq:the CQ being polled
2349 * @num_entries:maximum number of completions to return
2350 * @wc:array of at least @num_entries &struct ib_wc where completions
2353 * Poll a CQ for (possibly multiple) completions. If the return value
2354 * is < 0, an error occurred. If the return value is >= 0, it is the
2355 * number of completions returned. If the return value is
2356 * non-negative and < num_entries, then the CQ was emptied.
2358 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2361 return cq->device->poll_cq(cq, num_entries, wc);
2365 * ib_peek_cq - Returns the number of unreaped completions currently
2366 * on the specified CQ.
2367 * @cq: The CQ to peek.
2368 * @wc_cnt: A minimum number of unreaped completions to check for.
2370 * If the number of unreaped completions is greater than or equal to wc_cnt,
2371 * this function returns wc_cnt, otherwise, it returns the actual number of
2372 * unreaped completions.
2374 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2377 * ib_req_notify_cq - Request completion notification on a CQ.
2378 * @cq: The CQ to generate an event for.
2380 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2381 * to request an event on the next solicited event or next work
2382 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2383 * may also be |ed in to request a hint about missed events, as
2387 * < 0 means an error occurred while requesting notification
2388 * == 0 means notification was requested successfully, and if
2389 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2390 * were missed and it is safe to wait for another event. In
2391 * this case is it guaranteed that any work completions added
2392 * to the CQ since the last CQ poll will trigger a completion
2393 * notification event.
2394 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2395 * in. It means that the consumer must poll the CQ again to
2396 * make sure it is empty to avoid missing an event because of a
2397 * race between requesting notification and an entry being
2398 * added to the CQ. This return value means it is possible
2399 * (but not guaranteed) that a work completion has been added
2400 * to the CQ since the last poll without triggering a
2401 * completion notification event.
2403 static inline int ib_req_notify_cq(struct ib_cq *cq,
2404 enum ib_cq_notify_flags flags)
2406 return cq->device->req_notify_cq(cq, flags);
2410 * ib_req_ncomp_notif - Request completion notification when there are
2411 * at least the specified number of unreaped completions on the CQ.
2412 * @cq: The CQ to generate an event for.
2413 * @wc_cnt: The number of unreaped completions that should be on the
2414 * CQ before an event is generated.
2416 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2418 return cq->device->req_ncomp_notif ?
2419 cq->device->req_ncomp_notif(cq, wc_cnt) :
2424 * ib_get_dma_mr - Returns a memory region for system memory that is
2426 * @pd: The protection domain associated with the memory region.
2427 * @mr_access_flags: Specifies the memory access rights.
2429 * Note that the ib_dma_*() functions defined below must be used
2430 * to create/destroy addresses used with the Lkey or Rkey returned
2431 * by ib_get_dma_mr().
2433 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2436 * ib_dma_mapping_error - check a DMA addr for error
2437 * @dev: The device for which the dma_addr was created
2438 * @dma_addr: The DMA address to check
2440 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2443 return dev->dma_ops->mapping_error(dev, dma_addr);
2444 return dma_mapping_error(dev->dma_device, dma_addr);
2448 * ib_dma_map_single - Map a kernel virtual address to DMA address
2449 * @dev: The device for which the dma_addr is to be created
2450 * @cpu_addr: The kernel virtual address
2451 * @size: The size of the region in bytes
2452 * @direction: The direction of the DMA
2454 static inline u64 ib_dma_map_single(struct ib_device *dev,
2455 void *cpu_addr, size_t size,
2456 enum dma_data_direction direction)
2459 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2460 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2464 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2465 * @dev: The device for which the DMA address was created
2466 * @addr: The DMA address
2467 * @size: The size of the region in bytes
2468 * @direction: The direction of the DMA
2470 static inline void ib_dma_unmap_single(struct ib_device *dev,
2471 u64 addr, size_t size,
2472 enum dma_data_direction direction)
2475 dev->dma_ops->unmap_single(dev, addr, size, direction);
2477 dma_unmap_single(dev->dma_device, addr, size, direction);
2480 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2481 void *cpu_addr, size_t size,
2482 enum dma_data_direction direction,
2483 struct dma_attrs *attrs)
2485 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2489 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2490 u64 addr, size_t size,
2491 enum dma_data_direction direction,
2492 struct dma_attrs *attrs)
2494 return dma_unmap_single_attrs(dev->dma_device, addr, size,
2499 * ib_dma_map_page - Map a physical page to DMA address
2500 * @dev: The device for which the dma_addr is to be created
2501 * @page: The page to be mapped
2502 * @offset: The offset within the page
2503 * @size: The size of the region in bytes
2504 * @direction: The direction of the DMA
2506 static inline u64 ib_dma_map_page(struct ib_device *dev,
2508 unsigned long offset,
2510 enum dma_data_direction direction)
2513 return dev->dma_ops->map_page(dev, page, offset, size, direction);
2514 return dma_map_page(dev->dma_device, page, offset, size, direction);
2518 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2519 * @dev: The device for which the DMA address was created
2520 * @addr: The DMA address
2521 * @size: The size of the region in bytes
2522 * @direction: The direction of the DMA
2524 static inline void ib_dma_unmap_page(struct ib_device *dev,
2525 u64 addr, size_t size,
2526 enum dma_data_direction direction)
2529 dev->dma_ops->unmap_page(dev, addr, size, direction);
2531 dma_unmap_page(dev->dma_device, addr, size, direction);
2535 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2536 * @dev: The device for which the DMA addresses are to be created
2537 * @sg: The array of scatter/gather entries
2538 * @nents: The number of scatter/gather entries
2539 * @direction: The direction of the DMA
2541 static inline int ib_dma_map_sg(struct ib_device *dev,
2542 struct scatterlist *sg, int nents,
2543 enum dma_data_direction direction)
2546 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2547 return dma_map_sg(dev->dma_device, sg, nents, direction);
2551 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2552 * @dev: The device for which the DMA addresses were created
2553 * @sg: The array of scatter/gather entries
2554 * @nents: The number of scatter/gather entries
2555 * @direction: The direction of the DMA
2557 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2558 struct scatterlist *sg, int nents,
2559 enum dma_data_direction direction)
2562 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2564 dma_unmap_sg(dev->dma_device, sg, nents, direction);
2567 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2568 struct scatterlist *sg, int nents,
2569 enum dma_data_direction direction,
2570 struct dma_attrs *attrs)
2572 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2575 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2576 struct scatterlist *sg, int nents,
2577 enum dma_data_direction direction,
2578 struct dma_attrs *attrs)
2580 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2583 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2584 * @dev: The device for which the DMA addresses were created
2585 * @sg: The scatter/gather entry
2587 * Note: this function is obsolete. To do: change all occurrences of
2588 * ib_sg_dma_address() into sg_dma_address().
2590 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2591 struct scatterlist *sg)
2593 return sg_dma_address(sg);
2597 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2598 * @dev: The device for which the DMA addresses were created
2599 * @sg: The scatter/gather entry
2601 * Note: this function is obsolete. To do: change all occurrences of
2602 * ib_sg_dma_len() into sg_dma_len().
2604 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2605 struct scatterlist *sg)
2607 return sg_dma_len(sg);
2611 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2612 * @dev: The device for which the DMA address was created
2613 * @addr: The DMA address
2614 * @size: The size of the region in bytes
2615 * @dir: The direction of the DMA
2617 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2620 enum dma_data_direction dir)
2623 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2625 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2629 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2630 * @dev: The device for which the DMA address was created
2631 * @addr: The DMA address
2632 * @size: The size of the region in bytes
2633 * @dir: The direction of the DMA
2635 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2638 enum dma_data_direction dir)
2641 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2643 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2647 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2648 * @dev: The device for which the DMA address is requested
2649 * @size: The size of the region to allocate in bytes
2650 * @dma_handle: A pointer for returning the DMA address of the region
2651 * @flag: memory allocator flags
2653 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2659 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2664 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2665 *dma_handle = handle;
2671 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2672 * @dev: The device for which the DMA addresses were allocated
2673 * @size: The size of the region
2674 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2675 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2677 static inline void ib_dma_free_coherent(struct ib_device *dev,
2678 size_t size, void *cpu_addr,
2682 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2684 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2688 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
2690 * @pd: The protection domain associated assigned to the registered region.
2691 * @phys_buf_array: Specifies a list of physical buffers to use in the
2693 * @num_phys_buf: Specifies the size of the phys_buf_array.
2694 * @mr_access_flags: Specifies the memory access rights.
2695 * @iova_start: The offset of the region's starting I/O virtual address.
2697 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
2698 struct ib_phys_buf *phys_buf_array,
2700 int mr_access_flags,
2704 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
2705 * Conceptually, this call performs the functions deregister memory region
2706 * followed by register physical memory region. Where possible,
2707 * resources are reused instead of deallocated and reallocated.
2708 * @mr: The memory region to modify.
2709 * @mr_rereg_mask: A bit-mask used to indicate which of the following
2710 * properties of the memory region are being modified.
2711 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
2712 * the new protection domain to associated with the memory region,
2713 * otherwise, this parameter is ignored.
2714 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2715 * field specifies a list of physical buffers to use in the new
2716 * translation, otherwise, this parameter is ignored.
2717 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2718 * field specifies the size of the phys_buf_array, otherwise, this
2719 * parameter is ignored.
2720 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
2721 * field specifies the new memory access rights, otherwise, this
2722 * parameter is ignored.
2723 * @iova_start: The offset of the region's starting I/O virtual address.
2725 int ib_rereg_phys_mr(struct ib_mr *mr,
2728 struct ib_phys_buf *phys_buf_array,
2730 int mr_access_flags,
2734 * ib_query_mr - Retrieves information about a specific memory region.
2735 * @mr: The memory region to retrieve information about.
2736 * @mr_attr: The attributes of the specified memory region.
2738 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2741 * ib_dereg_mr - Deregisters a memory region and removes it from the
2742 * HCA translation table.
2743 * @mr: The memory region to deregister.
2745 * This function can fail, if the memory region has memory windows bound to it.
2747 int ib_dereg_mr(struct ib_mr *mr);
2751 * ib_create_mr - Allocates a memory region that may be used for
2752 * signature handover operations.
2753 * @pd: The protection domain associated with the region.
2754 * @mr_init_attr: memory region init attributes.
2756 struct ib_mr *ib_create_mr(struct ib_pd *pd,
2757 struct ib_mr_init_attr *mr_init_attr);
2760 * ib_destroy_mr - Destroys a memory region that was created using
2761 * ib_create_mr and removes it from HW translation tables.
2762 * @mr: The memory region to destroy.
2764 * This function can fail, if the memory region has memory windows bound to it.
2766 int ib_destroy_mr(struct ib_mr *mr);
2769 * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2770 * IB_WR_FAST_REG_MR send work request.
2771 * @pd: The protection domain associated with the region.
2772 * @max_page_list_len: requested max physical buffer list length to be
2773 * used with fast register work requests for this MR.
2775 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2778 * ib_alloc_fast_reg_page_list - Allocates a page list array
2779 * @device - ib device pointer.
2780 * @page_list_len - size of the page list array to be allocated.
2782 * This allocates and returns a struct ib_fast_reg_page_list * and a
2783 * page_list array that is at least page_list_len in size. The actual
2784 * size is returned in max_page_list_len. The caller is responsible
2785 * for initializing the contents of the page_list array before posting
2786 * a send work request with the IB_WC_FAST_REG_MR opcode.
2788 * The page_list array entries must be translated using one of the
2789 * ib_dma_*() functions just like the addresses passed to
2790 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
2791 * ib_fast_reg_page_list must not be modified by the caller until the
2792 * IB_WC_FAST_REG_MR work request completes.
2794 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2795 struct ib_device *device, int page_list_len);
2798 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2800 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2802 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2805 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2807 * @mr - struct ib_mr pointer to be updated.
2808 * @newkey - new key to be used.
2810 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2812 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2813 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2817 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2818 * for calculating a new rkey for type 2 memory windows.
2819 * @rkey - the rkey to increment.
2821 static inline u32 ib_inc_rkey(u32 rkey)
2823 const u32 mask = 0x000000ff;
2824 return ((rkey + 1) & mask) | (rkey & ~mask);
2828 * ib_alloc_mw - Allocates a memory window.
2829 * @pd: The protection domain associated with the memory window.
2830 * @type: The type of the memory window (1 or 2).
2832 struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
2835 * ib_bind_mw - Posts a work request to the send queue of the specified
2836 * QP, which binds the memory window to the given address range and
2837 * remote access attributes.
2838 * @qp: QP to post the bind work request on.
2839 * @mw: The memory window to bind.
2840 * @mw_bind: Specifies information about the memory window, including
2841 * its address range, remote access rights, and associated memory region.
2843 * If there is no immediate error, the function will update the rkey member
2844 * of the mw parameter to its new value. The bind operation can still fail
2847 static inline int ib_bind_mw(struct ib_qp *qp,
2849 struct ib_mw_bind *mw_bind)
2851 /* XXX reference counting in corresponding MR? */
2852 return mw->device->bind_mw ?
2853 mw->device->bind_mw(qp, mw, mw_bind) :
2858 * ib_dealloc_mw - Deallocates a memory window.
2859 * @mw: The memory window to deallocate.
2861 int ib_dealloc_mw(struct ib_mw *mw);
2864 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2865 * @pd: The protection domain associated with the unmapped region.
2866 * @mr_access_flags: Specifies the memory access rights.
2867 * @fmr_attr: Attributes of the unmapped region.
2869 * A fast memory region must be mapped before it can be used as part of
2872 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2873 int mr_access_flags,
2874 struct ib_fmr_attr *fmr_attr);
2877 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2878 * @fmr: The fast memory region to associate with the pages.
2879 * @page_list: An array of physical pages to map to the fast memory region.
2880 * @list_len: The number of pages in page_list.
2881 * @iova: The I/O virtual address to use with the mapped region.
2883 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2884 u64 *page_list, int list_len,
2887 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2891 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2892 * @fmr_list: A linked list of fast memory regions to unmap.
2894 int ib_unmap_fmr(struct list_head *fmr_list);
2897 * ib_dealloc_fmr - Deallocates a fast memory region.
2898 * @fmr: The fast memory region to deallocate.
2900 int ib_dealloc_fmr(struct ib_fmr *fmr);
2903 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2904 * @qp: QP to attach to the multicast group. The QP must be type
2906 * @gid: Multicast group GID.
2907 * @lid: Multicast group LID in host byte order.
2909 * In order to send and receive multicast packets, subnet
2910 * administration must have created the multicast group and configured
2911 * the fabric appropriately. The port associated with the specified
2912 * QP must also be a member of the multicast group.
2914 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2917 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2918 * @qp: QP to detach from the multicast group.
2919 * @gid: Multicast group GID.
2920 * @lid: Multicast group LID in host byte order.
2922 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2925 * ib_alloc_xrcd - Allocates an XRC domain.
2926 * @device: The device on which to allocate the XRC domain.
2928 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2931 * ib_dealloc_xrcd - Deallocates an XRC domain.
2932 * @xrcd: The XRC domain to deallocate.
2934 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2936 struct ib_flow *ib_create_flow(struct ib_qp *qp,
2937 struct ib_flow_attr *flow_attr, int domain);
2938 int ib_destroy_flow(struct ib_flow *flow_id);
2940 static inline int ib_check_mr_access(int flags)
2943 * Local write permission is required if remote write or
2944 * remote atomic permission is also requested.
2946 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
2947 !(flags & IB_ACCESS_LOCAL_WRITE))
2954 * ib_check_mr_status: lightweight check of MR status.
2955 * This routine may provide status checks on a selected
2956 * ib_mr. first use is for signature status check.
2958 * @mr: A memory region.
2959 * @check_mask: Bitmask of which checks to perform from
2960 * ib_mr_status_check enumeration.
2961 * @mr_status: The container of relevant status checks.
2962 * failed checks will be indicated in the status bitmask
2963 * and the relevant info shall be in the error item.
2965 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
2966 struct ib_mr_status *mr_status);
2968 #endif /* IB_VERBS_H */