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 struct ib_port_attr {
357 enum ib_port_state state;
359 enum ib_mtu active_mtu;
378 enum ib_device_modify_flags {
379 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
380 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
383 struct ib_device_modify {
388 enum ib_port_modify_flags {
389 IB_PORT_SHUTDOWN = 1,
390 IB_PORT_INIT_TYPE = (1<<2),
391 IB_PORT_RESET_QKEY_CNTR = (1<<3)
394 struct ib_port_modify {
395 u32 set_port_cap_mask;
396 u32 clr_port_cap_mask;
404 IB_EVENT_QP_ACCESS_ERR,
408 IB_EVENT_PATH_MIG_ERR,
409 IB_EVENT_DEVICE_FATAL,
410 IB_EVENT_PORT_ACTIVE,
413 IB_EVENT_PKEY_CHANGE,
416 IB_EVENT_SRQ_LIMIT_REACHED,
417 IB_EVENT_QP_LAST_WQE_REACHED,
418 IB_EVENT_CLIENT_REREGISTER,
423 struct ib_device *device;
430 enum ib_event_type event;
433 struct ib_event_handler {
434 struct ib_device *device;
435 void (*handler)(struct ib_event_handler *, struct ib_event *);
436 struct list_head list;
439 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
441 (_ptr)->device = _device; \
442 (_ptr)->handler = _handler; \
443 INIT_LIST_HEAD(&(_ptr)->list); \
446 struct ib_global_route {
455 __be32 version_tclass_flow;
464 IB_MULTICAST_QPN = 0xffffff
467 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
474 IB_RATE_PORT_CURRENT = 0,
475 IB_RATE_2_5_GBPS = 2,
483 IB_RATE_120_GBPS = 10,
484 IB_RATE_14_GBPS = 11,
485 IB_RATE_56_GBPS = 12,
486 IB_RATE_112_GBPS = 13,
487 IB_RATE_168_GBPS = 14,
488 IB_RATE_25_GBPS = 15,
489 IB_RATE_100_GBPS = 16,
490 IB_RATE_200_GBPS = 17,
491 IB_RATE_300_GBPS = 18
495 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
496 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
497 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
498 * @rate: rate to convert.
500 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
503 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
504 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
505 * @rate: rate to convert.
507 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
509 enum ib_mr_create_flags {
510 IB_MR_SIGNATURE_EN = 1,
514 * ib_mr_init_attr - Memory region init attributes passed to routine
516 * @max_reg_descriptors: max number of registration descriptors that
517 * may be used with registration work requests.
518 * @flags: MR creation flags bit mask.
520 struct ib_mr_init_attr {
521 int max_reg_descriptors;
527 * IB_SIG_TYPE_NONE: Unprotected.
528 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
530 enum ib_signature_type {
536 * Signature T10-DIF block-guard types
537 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
538 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
540 enum ib_t10_dif_bg_type {
546 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
548 * @bg_type: T10-DIF block guard type (CRC|CSUM)
549 * @pi_interval: protection information interval.
550 * @bg: seed of guard computation.
551 * @app_tag: application tag of guard block
552 * @ref_tag: initial guard block reference tag.
553 * @ref_remap: Indicate wethear the reftag increments each block
554 * @app_escape: Indicate to skip block check if apptag=0xffff
555 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
556 * @apptag_check_mask: check bitmask of application tag.
558 struct ib_t10_dif_domain {
559 enum ib_t10_dif_bg_type bg_type;
567 u16 apptag_check_mask;
571 * struct ib_sig_domain - Parameters for signature domain
572 * @sig_type: specific signauture type
573 * @sig: union of all signature domain attributes that may
574 * be used to set domain layout.
576 struct ib_sig_domain {
577 enum ib_signature_type sig_type;
579 struct ib_t10_dif_domain dif;
584 * struct ib_sig_attrs - Parameters for signature handover operation
585 * @check_mask: bitmask for signature byte check (8 bytes)
586 * @mem: memory domain layout desciptor.
587 * @wire: wire domain layout desciptor.
589 struct ib_sig_attrs {
591 struct ib_sig_domain mem;
592 struct ib_sig_domain wire;
595 enum ib_sig_err_type {
602 * struct ib_sig_err - signature error descriptor
605 enum ib_sig_err_type err_type;
612 enum ib_mr_status_check {
613 IB_MR_CHECK_SIG_STATUS = 1,
617 * struct ib_mr_status - Memory region status container
619 * @fail_status: Bitmask of MR checks status. For each
620 * failed check a corresponding status bit is set.
621 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
624 struct ib_mr_status {
626 struct ib_sig_err sig_err;
630 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
632 * @mult: multiple to convert.
634 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
637 struct ib_global_route grh;
652 IB_WC_LOC_EEC_OP_ERR,
657 IB_WC_LOC_ACCESS_ERR,
658 IB_WC_REM_INV_REQ_ERR,
659 IB_WC_REM_ACCESS_ERR,
662 IB_WC_RNR_RETRY_EXC_ERR,
663 IB_WC_LOC_RDD_VIOL_ERR,
664 IB_WC_REM_INV_RD_REQ_ERR,
667 IB_WC_INV_EEC_STATE_ERR,
669 IB_WC_RESP_TIMEOUT_ERR,
683 IB_WC_MASKED_COMP_SWAP,
684 IB_WC_MASKED_FETCH_ADD,
686 * Set value of IB_WC_RECV so consumers can test if a completion is a
687 * receive by testing (opcode & IB_WC_RECV).
690 IB_WC_RECV_RDMA_WITH_IMM
695 IB_WC_WITH_IMM = (1<<1),
696 IB_WC_WITH_INVALIDATE = (1<<2),
697 IB_WC_IP_CSUM_OK = (1<<3),
698 IB_WC_WITH_SMAC = (1<<4),
699 IB_WC_WITH_VLAN = (1<<5),
704 enum ib_wc_status status;
705 enum ib_wc_opcode opcode;
719 u8 port_num; /* valid only for DR SMPs on switches */
724 enum ib_cq_notify_flags {
725 IB_CQ_SOLICITED = 1 << 0,
726 IB_CQ_NEXT_COMP = 1 << 1,
727 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
728 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
736 enum ib_srq_attr_mask {
737 IB_SRQ_MAX_WR = 1 << 0,
738 IB_SRQ_LIMIT = 1 << 1,
747 struct ib_srq_init_attr {
748 void (*event_handler)(struct ib_event *, void *);
750 struct ib_srq_attr attr;
751 enum ib_srq_type srq_type;
755 struct ib_xrcd *xrcd;
776 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
777 * here (and in that order) since the MAD layer uses them as
778 * indices into a 2-entry table.
787 IB_QPT_RAW_ETHERTYPE,
788 IB_QPT_RAW_PACKET = 8,
792 /* Reserve a range for qp types internal to the low level driver.
793 * These qp types will not be visible at the IB core layer, so the
794 * IB_QPT_MAX usages should not be affected in the core layer
796 IB_QPT_RESERVED1 = 0x1000,
808 enum ib_qp_create_flags {
809 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
810 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
811 IB_QP_CREATE_NETIF_QP = 1 << 5,
812 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
813 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
814 /* reserve bits 26-31 for low level drivers' internal use */
815 IB_QP_CREATE_RESERVED_START = 1 << 26,
816 IB_QP_CREATE_RESERVED_END = 1 << 31,
821 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
822 * callback to destroy the passed in QP.
825 struct ib_qp_init_attr {
826 void (*event_handler)(struct ib_event *, void *);
828 struct ib_cq *send_cq;
829 struct ib_cq *recv_cq;
831 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
832 struct ib_qp_cap cap;
833 enum ib_sig_type sq_sig_type;
834 enum ib_qp_type qp_type;
835 enum ib_qp_create_flags create_flags;
836 u8 port_num; /* special QP types only */
839 struct ib_qp_open_attr {
840 void (*event_handler)(struct ib_event *, void *);
843 enum ib_qp_type qp_type;
846 enum ib_rnr_timeout {
847 IB_RNR_TIMER_655_36 = 0,
848 IB_RNR_TIMER_000_01 = 1,
849 IB_RNR_TIMER_000_02 = 2,
850 IB_RNR_TIMER_000_03 = 3,
851 IB_RNR_TIMER_000_04 = 4,
852 IB_RNR_TIMER_000_06 = 5,
853 IB_RNR_TIMER_000_08 = 6,
854 IB_RNR_TIMER_000_12 = 7,
855 IB_RNR_TIMER_000_16 = 8,
856 IB_RNR_TIMER_000_24 = 9,
857 IB_RNR_TIMER_000_32 = 10,
858 IB_RNR_TIMER_000_48 = 11,
859 IB_RNR_TIMER_000_64 = 12,
860 IB_RNR_TIMER_000_96 = 13,
861 IB_RNR_TIMER_001_28 = 14,
862 IB_RNR_TIMER_001_92 = 15,
863 IB_RNR_TIMER_002_56 = 16,
864 IB_RNR_TIMER_003_84 = 17,
865 IB_RNR_TIMER_005_12 = 18,
866 IB_RNR_TIMER_007_68 = 19,
867 IB_RNR_TIMER_010_24 = 20,
868 IB_RNR_TIMER_015_36 = 21,
869 IB_RNR_TIMER_020_48 = 22,
870 IB_RNR_TIMER_030_72 = 23,
871 IB_RNR_TIMER_040_96 = 24,
872 IB_RNR_TIMER_061_44 = 25,
873 IB_RNR_TIMER_081_92 = 26,
874 IB_RNR_TIMER_122_88 = 27,
875 IB_RNR_TIMER_163_84 = 28,
876 IB_RNR_TIMER_245_76 = 29,
877 IB_RNR_TIMER_327_68 = 30,
878 IB_RNR_TIMER_491_52 = 31
881 enum ib_qp_attr_mask {
883 IB_QP_CUR_STATE = (1<<1),
884 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
885 IB_QP_ACCESS_FLAGS = (1<<3),
886 IB_QP_PKEY_INDEX = (1<<4),
890 IB_QP_PATH_MTU = (1<<8),
891 IB_QP_TIMEOUT = (1<<9),
892 IB_QP_RETRY_CNT = (1<<10),
893 IB_QP_RNR_RETRY = (1<<11),
894 IB_QP_RQ_PSN = (1<<12),
895 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
896 IB_QP_ALT_PATH = (1<<14),
897 IB_QP_MIN_RNR_TIMER = (1<<15),
898 IB_QP_SQ_PSN = (1<<16),
899 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
900 IB_QP_PATH_MIG_STATE = (1<<18),
902 IB_QP_DEST_QPN = (1<<20),
903 IB_QP_SMAC = (1<<21),
904 IB_QP_ALT_SMAC = (1<<22),
906 IB_QP_ALT_VID = (1<<24),
931 enum ib_qp_state qp_state;
932 enum ib_qp_state cur_qp_state;
933 enum ib_mtu path_mtu;
934 enum ib_mig_state path_mig_state;
940 struct ib_qp_cap cap;
941 struct ib_ah_attr ah_attr;
942 struct ib_ah_attr alt_ah_attr;
945 u8 en_sqd_async_notify;
948 u8 max_dest_rd_atomic;
957 u8 alt_smac[ETH_ALEN];
964 IB_WR_RDMA_WRITE_WITH_IMM,
968 IB_WR_ATOMIC_CMP_AND_SWP,
969 IB_WR_ATOMIC_FETCH_AND_ADD,
972 IB_WR_RDMA_READ_WITH_INV,
975 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
976 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
979 /* reserve values for low level drivers' internal use.
980 * These values will not be used at all in the ib core layer.
982 IB_WR_RESERVED1 = 0xf0,
996 IB_SEND_SIGNALED = (1<<1),
997 IB_SEND_SOLICITED = (1<<2),
998 IB_SEND_INLINE = (1<<3),
999 IB_SEND_IP_CSUM = (1<<4),
1001 /* reserve bits 26-31 for low level drivers' internal use */
1002 IB_SEND_RESERVED_START = (1 << 26),
1003 IB_SEND_RESERVED_END = (1 << 31),
1012 struct ib_fast_reg_page_list {
1013 struct ib_device *device;
1015 unsigned int max_page_list_len;
1019 * struct ib_mw_bind_info - Parameters for a memory window bind operation.
1020 * @mr: A memory region to bind the memory window to.
1021 * @addr: The address where the memory window should begin.
1022 * @length: The length of the memory window, in bytes.
1023 * @mw_access_flags: Access flags from enum ib_access_flags for the window.
1025 * This struct contains the shared parameters for type 1 and type 2
1026 * memory window bind operations.
1028 struct ib_mw_bind_info {
1032 int mw_access_flags;
1036 struct ib_send_wr *next;
1038 struct ib_sge *sg_list;
1040 enum ib_wr_opcode opcode;
1044 u32 invalidate_rkey;
1055 u64 compare_add_mask;
1066 u16 pkey_index; /* valid for GSI only */
1067 u8 port_num; /* valid for DR SMPs on switch only */
1071 struct ib_fast_reg_page_list *page_list;
1072 unsigned int page_shift;
1073 unsigned int page_list_len;
1080 /* The new rkey for the memory window. */
1082 struct ib_mw_bind_info bind_info;
1085 struct ib_sig_attrs *sig_attrs;
1086 struct ib_mr *sig_mr;
1088 struct ib_sge *prot;
1091 u32 xrc_remote_srq_num; /* XRC TGT QPs only */
1095 struct ib_recv_wr *next;
1097 struct ib_sge *sg_list;
1101 enum ib_access_flags {
1102 IB_ACCESS_LOCAL_WRITE = 1,
1103 IB_ACCESS_REMOTE_WRITE = (1<<1),
1104 IB_ACCESS_REMOTE_READ = (1<<2),
1105 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1106 IB_ACCESS_MW_BIND = (1<<4),
1107 IB_ZERO_BASED = (1<<5),
1108 IB_ACCESS_ON_DEMAND = (1<<6),
1111 struct ib_phys_buf {
1118 u64 device_virt_addr;
1120 int mr_access_flags;
1125 enum ib_mr_rereg_flags {
1126 IB_MR_REREG_TRANS = 1,
1127 IB_MR_REREG_PD = (1<<1),
1128 IB_MR_REREG_ACCESS = (1<<2),
1129 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1133 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
1134 * @wr_id: Work request id.
1135 * @send_flags: Flags from ib_send_flags enum.
1136 * @bind_info: More parameters of the bind operation.
1141 struct ib_mw_bind_info bind_info;
1144 struct ib_fmr_attr {
1152 struct ib_ucontext {
1153 struct ib_device *device;
1154 struct list_head pd_list;
1155 struct list_head mr_list;
1156 struct list_head mw_list;
1157 struct list_head cq_list;
1158 struct list_head qp_list;
1159 struct list_head srq_list;
1160 struct list_head ah_list;
1161 struct list_head xrcd_list;
1162 struct list_head rule_list;
1166 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1167 struct rb_root umem_tree;
1169 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1170 * mmu notifiers registration.
1172 struct rw_semaphore umem_rwsem;
1173 void (*invalidate_range)(struct ib_umem *umem,
1174 unsigned long start, unsigned long end);
1176 struct mmu_notifier mn;
1177 atomic_t notifier_count;
1178 /* A list of umems that don't have private mmu notifier counters yet. */
1179 struct list_head no_private_counters;
1185 u64 user_handle; /* handle given to us by userspace */
1186 struct ib_ucontext *context; /* associated user context */
1187 void *object; /* containing object */
1188 struct list_head list; /* link to context's list */
1189 int id; /* index into kernel idr */
1191 struct rw_semaphore mutex; /* protects .live */
1196 const void __user *inbuf;
1197 void __user *outbuf;
1203 struct ib_device *device;
1204 struct ib_uobject *uobject;
1205 atomic_t usecnt; /* count all resources */
1209 struct ib_device *device;
1210 atomic_t usecnt; /* count all exposed resources */
1211 struct inode *inode;
1213 struct mutex tgt_qp_mutex;
1214 struct list_head tgt_qp_list;
1218 struct ib_device *device;
1220 struct ib_uobject *uobject;
1223 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1226 struct ib_device *device;
1227 struct ib_uobject *uobject;
1228 ib_comp_handler comp_handler;
1229 void (*event_handler)(struct ib_event *, void *);
1232 atomic_t usecnt; /* count number of work queues */
1236 struct ib_device *device;
1238 struct ib_uobject *uobject;
1239 void (*event_handler)(struct ib_event *, void *);
1241 enum ib_srq_type srq_type;
1246 struct ib_xrcd *xrcd;
1254 struct ib_device *device;
1256 struct ib_cq *send_cq;
1257 struct ib_cq *recv_cq;
1259 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1260 struct list_head xrcd_list;
1261 /* count times opened, mcast attaches, flow attaches */
1263 struct list_head open_list;
1264 struct ib_qp *real_qp;
1265 struct ib_uobject *uobject;
1266 void (*event_handler)(struct ib_event *, void *);
1269 enum ib_qp_type qp_type;
1273 struct ib_device *device;
1275 struct ib_uobject *uobject;
1278 atomic_t usecnt; /* count number of MWs */
1282 struct ib_device *device;
1284 struct ib_uobject *uobject;
1286 enum ib_mw_type type;
1290 struct ib_device *device;
1292 struct list_head list;
1297 /* Supported steering options */
1298 enum ib_flow_attr_type {
1299 /* steering according to rule specifications */
1300 IB_FLOW_ATTR_NORMAL = 0x0,
1301 /* default unicast and multicast rule -
1302 * receive all Eth traffic which isn't steered to any QP
1304 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1305 /* default multicast rule -
1306 * receive all Eth multicast traffic which isn't steered to any QP
1308 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1309 /* sniffer rule - receive all port traffic */
1310 IB_FLOW_ATTR_SNIFFER = 0x3
1313 /* Supported steering header types */
1314 enum ib_flow_spec_type {
1316 IB_FLOW_SPEC_ETH = 0x20,
1317 IB_FLOW_SPEC_IB = 0x22,
1319 IB_FLOW_SPEC_IPV4 = 0x30,
1321 IB_FLOW_SPEC_TCP = 0x40,
1322 IB_FLOW_SPEC_UDP = 0x41
1324 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1325 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1327 /* Flow steering rule priority is set according to it's domain.
1328 * Lower domain value means higher priority.
1330 enum ib_flow_domain {
1331 IB_FLOW_DOMAIN_USER,
1332 IB_FLOW_DOMAIN_ETHTOOL,
1335 IB_FLOW_DOMAIN_NUM /* Must be last */
1338 struct ib_flow_eth_filter {
1345 struct ib_flow_spec_eth {
1346 enum ib_flow_spec_type type;
1348 struct ib_flow_eth_filter val;
1349 struct ib_flow_eth_filter mask;
1352 struct ib_flow_ib_filter {
1357 struct ib_flow_spec_ib {
1358 enum ib_flow_spec_type type;
1360 struct ib_flow_ib_filter val;
1361 struct ib_flow_ib_filter mask;
1364 struct ib_flow_ipv4_filter {
1369 struct ib_flow_spec_ipv4 {
1370 enum ib_flow_spec_type type;
1372 struct ib_flow_ipv4_filter val;
1373 struct ib_flow_ipv4_filter mask;
1376 struct ib_flow_tcp_udp_filter {
1381 struct ib_flow_spec_tcp_udp {
1382 enum ib_flow_spec_type type;
1384 struct ib_flow_tcp_udp_filter val;
1385 struct ib_flow_tcp_udp_filter mask;
1388 union ib_flow_spec {
1390 enum ib_flow_spec_type type;
1393 struct ib_flow_spec_eth eth;
1394 struct ib_flow_spec_ib ib;
1395 struct ib_flow_spec_ipv4 ipv4;
1396 struct ib_flow_spec_tcp_udp tcp_udp;
1399 struct ib_flow_attr {
1400 enum ib_flow_attr_type type;
1406 /* Following are the optional layers according to user request
1407 * struct ib_flow_spec_xxx
1408 * struct ib_flow_spec_yyy
1414 struct ib_uobject *uobject;
1420 enum ib_process_mad_flags {
1421 IB_MAD_IGNORE_MKEY = 1,
1422 IB_MAD_IGNORE_BKEY = 2,
1423 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1426 enum ib_mad_result {
1427 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1428 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1429 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1430 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1433 #define IB_DEVICE_NAME_MAX 64
1437 struct ib_event_handler event_handler;
1438 struct ib_pkey_cache **pkey_cache;
1439 struct ib_gid_cache **gid_cache;
1443 struct ib_dma_mapping_ops {
1444 int (*mapping_error)(struct ib_device *dev,
1446 u64 (*map_single)(struct ib_device *dev,
1447 void *ptr, size_t size,
1448 enum dma_data_direction direction);
1449 void (*unmap_single)(struct ib_device *dev,
1450 u64 addr, size_t size,
1451 enum dma_data_direction direction);
1452 u64 (*map_page)(struct ib_device *dev,
1453 struct page *page, unsigned long offset,
1455 enum dma_data_direction direction);
1456 void (*unmap_page)(struct ib_device *dev,
1457 u64 addr, size_t size,
1458 enum dma_data_direction direction);
1459 int (*map_sg)(struct ib_device *dev,
1460 struct scatterlist *sg, int nents,
1461 enum dma_data_direction direction);
1462 void (*unmap_sg)(struct ib_device *dev,
1463 struct scatterlist *sg, int nents,
1464 enum dma_data_direction direction);
1465 void (*sync_single_for_cpu)(struct ib_device *dev,
1468 enum dma_data_direction dir);
1469 void (*sync_single_for_device)(struct ib_device *dev,
1472 enum dma_data_direction dir);
1473 void *(*alloc_coherent)(struct ib_device *dev,
1477 void (*free_coherent)(struct ib_device *dev,
1478 size_t size, void *cpu_addr,
1484 struct ib_port_immutable {
1490 struct device *dma_device;
1492 char name[IB_DEVICE_NAME_MAX];
1494 struct list_head event_handler_list;
1495 spinlock_t event_handler_lock;
1497 spinlock_t client_data_lock;
1498 struct list_head core_list;
1499 struct list_head client_data_list;
1501 struct ib_cache cache;
1503 * port_immutable is indexed by port number
1505 struct ib_port_immutable *port_immutable;
1507 int num_comp_vectors;
1509 struct iw_cm_verbs *iwcm;
1511 int (*get_protocol_stats)(struct ib_device *device,
1512 union rdma_protocol_stats *stats);
1513 int (*query_device)(struct ib_device *device,
1514 struct ib_device_attr *device_attr);
1515 int (*query_port)(struct ib_device *device,
1517 struct ib_port_attr *port_attr);
1518 enum rdma_protocol_type (*query_protocol)(struct ib_device *device,
1520 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1522 int (*query_gid)(struct ib_device *device,
1523 u8 port_num, int index,
1525 int (*query_pkey)(struct ib_device *device,
1526 u8 port_num, u16 index, u16 *pkey);
1527 int (*modify_device)(struct ib_device *device,
1528 int device_modify_mask,
1529 struct ib_device_modify *device_modify);
1530 int (*modify_port)(struct ib_device *device,
1531 u8 port_num, int port_modify_mask,
1532 struct ib_port_modify *port_modify);
1533 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1534 struct ib_udata *udata);
1535 int (*dealloc_ucontext)(struct ib_ucontext *context);
1536 int (*mmap)(struct ib_ucontext *context,
1537 struct vm_area_struct *vma);
1538 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1539 struct ib_ucontext *context,
1540 struct ib_udata *udata);
1541 int (*dealloc_pd)(struct ib_pd *pd);
1542 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1543 struct ib_ah_attr *ah_attr);
1544 int (*modify_ah)(struct ib_ah *ah,
1545 struct ib_ah_attr *ah_attr);
1546 int (*query_ah)(struct ib_ah *ah,
1547 struct ib_ah_attr *ah_attr);
1548 int (*destroy_ah)(struct ib_ah *ah);
1549 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1550 struct ib_srq_init_attr *srq_init_attr,
1551 struct ib_udata *udata);
1552 int (*modify_srq)(struct ib_srq *srq,
1553 struct ib_srq_attr *srq_attr,
1554 enum ib_srq_attr_mask srq_attr_mask,
1555 struct ib_udata *udata);
1556 int (*query_srq)(struct ib_srq *srq,
1557 struct ib_srq_attr *srq_attr);
1558 int (*destroy_srq)(struct ib_srq *srq);
1559 int (*post_srq_recv)(struct ib_srq *srq,
1560 struct ib_recv_wr *recv_wr,
1561 struct ib_recv_wr **bad_recv_wr);
1562 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1563 struct ib_qp_init_attr *qp_init_attr,
1564 struct ib_udata *udata);
1565 int (*modify_qp)(struct ib_qp *qp,
1566 struct ib_qp_attr *qp_attr,
1568 struct ib_udata *udata);
1569 int (*query_qp)(struct ib_qp *qp,
1570 struct ib_qp_attr *qp_attr,
1572 struct ib_qp_init_attr *qp_init_attr);
1573 int (*destroy_qp)(struct ib_qp *qp);
1574 int (*post_send)(struct ib_qp *qp,
1575 struct ib_send_wr *send_wr,
1576 struct ib_send_wr **bad_send_wr);
1577 int (*post_recv)(struct ib_qp *qp,
1578 struct ib_recv_wr *recv_wr,
1579 struct ib_recv_wr **bad_recv_wr);
1580 struct ib_cq * (*create_cq)(struct ib_device *device, int cqe,
1582 struct ib_ucontext *context,
1583 struct ib_udata *udata);
1584 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1586 int (*destroy_cq)(struct ib_cq *cq);
1587 int (*resize_cq)(struct ib_cq *cq, int cqe,
1588 struct ib_udata *udata);
1589 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1591 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1592 int (*req_notify_cq)(struct ib_cq *cq,
1593 enum ib_cq_notify_flags flags);
1594 int (*req_ncomp_notif)(struct ib_cq *cq,
1596 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1597 int mr_access_flags);
1598 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
1599 struct ib_phys_buf *phys_buf_array,
1601 int mr_access_flags,
1603 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1604 u64 start, u64 length,
1606 int mr_access_flags,
1607 struct ib_udata *udata);
1608 int (*rereg_user_mr)(struct ib_mr *mr,
1610 u64 start, u64 length,
1612 int mr_access_flags,
1614 struct ib_udata *udata);
1615 int (*query_mr)(struct ib_mr *mr,
1616 struct ib_mr_attr *mr_attr);
1617 int (*dereg_mr)(struct ib_mr *mr);
1618 int (*destroy_mr)(struct ib_mr *mr);
1619 struct ib_mr * (*create_mr)(struct ib_pd *pd,
1620 struct ib_mr_init_attr *mr_init_attr);
1621 struct ib_mr * (*alloc_fast_reg_mr)(struct ib_pd *pd,
1622 int max_page_list_len);
1623 struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1625 void (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1626 int (*rereg_phys_mr)(struct ib_mr *mr,
1629 struct ib_phys_buf *phys_buf_array,
1631 int mr_access_flags,
1633 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1634 enum ib_mw_type type);
1635 int (*bind_mw)(struct ib_qp *qp,
1637 struct ib_mw_bind *mw_bind);
1638 int (*dealloc_mw)(struct ib_mw *mw);
1639 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1640 int mr_access_flags,
1641 struct ib_fmr_attr *fmr_attr);
1642 int (*map_phys_fmr)(struct ib_fmr *fmr,
1643 u64 *page_list, int list_len,
1645 int (*unmap_fmr)(struct list_head *fmr_list);
1646 int (*dealloc_fmr)(struct ib_fmr *fmr);
1647 int (*attach_mcast)(struct ib_qp *qp,
1650 int (*detach_mcast)(struct ib_qp *qp,
1653 int (*process_mad)(struct ib_device *device,
1654 int process_mad_flags,
1656 struct ib_wc *in_wc,
1657 struct ib_grh *in_grh,
1658 struct ib_mad *in_mad,
1659 struct ib_mad *out_mad);
1660 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1661 struct ib_ucontext *ucontext,
1662 struct ib_udata *udata);
1663 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1664 struct ib_flow * (*create_flow)(struct ib_qp *qp,
1668 int (*destroy_flow)(struct ib_flow *flow_id);
1669 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1670 struct ib_mr_status *mr_status);
1672 struct ib_dma_mapping_ops *dma_ops;
1674 struct module *owner;
1676 struct kobject *ports_parent;
1677 struct list_head port_list;
1680 IB_DEV_UNINITIALIZED,
1686 u64 uverbs_cmd_mask;
1687 u64 uverbs_ex_cmd_mask;
1696 * The following mandatory functions are used only at device
1697 * registration. Keep functions such as these at the end of this
1698 * structure to avoid cache line misses when accessing struct ib_device
1701 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1706 void (*add) (struct ib_device *);
1707 void (*remove)(struct ib_device *);
1709 struct list_head list;
1712 struct ib_device *ib_alloc_device(size_t size);
1713 void ib_dealloc_device(struct ib_device *device);
1715 int ib_register_device(struct ib_device *device,
1716 int (*port_callback)(struct ib_device *,
1717 u8, struct kobject *));
1718 void ib_unregister_device(struct ib_device *device);
1720 int ib_register_client (struct ib_client *client);
1721 void ib_unregister_client(struct ib_client *client);
1723 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1724 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1727 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1729 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1732 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1734 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1738 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1739 * contains all required attributes and no attributes not allowed for
1740 * the given QP state transition.
1741 * @cur_state: Current QP state
1742 * @next_state: Next QP state
1744 * @mask: Mask of supplied QP attributes
1745 * @ll : link layer of port
1747 * This function is a helper function that a low-level driver's
1748 * modify_qp method can use to validate the consumer's input. It
1749 * checks that cur_state and next_state are valid QP states, that a
1750 * transition from cur_state to next_state is allowed by the IB spec,
1751 * and that the attribute mask supplied is allowed for the transition.
1753 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1754 enum ib_qp_type type, enum ib_qp_attr_mask mask,
1755 enum rdma_link_layer ll);
1757 int ib_register_event_handler (struct ib_event_handler *event_handler);
1758 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1759 void ib_dispatch_event(struct ib_event *event);
1761 int ib_query_device(struct ib_device *device,
1762 struct ib_device_attr *device_attr);
1764 int ib_query_port(struct ib_device *device,
1765 u8 port_num, struct ib_port_attr *port_attr);
1767 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1771 * rdma_start_port - Return the first valid port number for the device
1774 * @device: Device to be checked
1776 * Return start port number
1778 static inline u8 rdma_start_port(const struct ib_device *device)
1780 return (device->node_type == RDMA_NODE_IB_SWITCH) ? 0 : 1;
1784 * rdma_end_port - Return the last valid port number for the device
1787 * @device: Device to be checked
1789 * Return last port number
1791 static inline u8 rdma_end_port(const struct ib_device *device)
1793 return (device->node_type == RDMA_NODE_IB_SWITCH) ?
1794 0 : device->phys_port_cnt;
1797 static inline bool rdma_protocol_ib(struct ib_device *device, u8 port_num)
1799 return device->query_protocol(device, port_num) == RDMA_PROTOCOL_IB;
1802 static inline bool rdma_protocol_iboe(struct ib_device *device, u8 port_num)
1804 return device->query_protocol(device, port_num) == RDMA_PROTOCOL_IBOE;
1807 static inline bool rdma_protocol_iwarp(struct ib_device *device, u8 port_num)
1809 return device->query_protocol(device, port_num) == RDMA_PROTOCOL_IWARP;
1812 static inline bool rdma_ib_or_iboe(struct ib_device *device, u8 port_num)
1814 enum rdma_protocol_type pt = device->query_protocol(device, port_num);
1816 return (pt == RDMA_PROTOCOL_IB || pt == RDMA_PROTOCOL_IBOE);
1820 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
1821 * Management Datagrams.
1822 * @device: Device to check
1823 * @port_num: Port number to check
1825 * Management Datagrams (MAD) are a required part of the InfiniBand
1826 * specification and are supported on all InfiniBand devices. A slightly
1827 * extended version are also supported on OPA interfaces.
1829 * Return: true if the port supports sending/receiving of MAD packets.
1831 static inline bool rdma_cap_ib_mad(struct ib_device *device, u8 port_num)
1833 return rdma_ib_or_iboe(device, port_num);
1837 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
1838 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
1839 * @device: Device to check
1840 * @port_num: Port number to check
1842 * Each InfiniBand node is required to provide a Subnet Management Agent
1843 * that the subnet manager can access. Prior to the fabric being fully
1844 * configured by the subnet manager, the SMA is accessed via a well known
1845 * interface called the Subnet Management Interface (SMI). This interface
1846 * uses directed route packets to communicate with the SM to get around the
1847 * chicken and egg problem of the SM needing to know what's on the fabric
1848 * in order to configure the fabric, and needing to configure the fabric in
1849 * order to send packets to the devices on the fabric. These directed
1850 * route packets do not need the fabric fully configured in order to reach
1851 * their destination. The SMI is the only method allowed to send
1852 * directed route packets on an InfiniBand fabric.
1854 * Return: true if the port provides an SMI.
1856 static inline bool rdma_cap_ib_smi(struct ib_device *device, u8 port_num)
1858 return rdma_protocol_ib(device, port_num);
1862 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
1863 * Communication Manager.
1864 * @device: Device to check
1865 * @port_num: Port number to check
1867 * The InfiniBand Communication Manager is one of many pre-defined General
1868 * Service Agents (GSA) that are accessed via the General Service
1869 * Interface (GSI). It's role is to facilitate establishment of connections
1870 * between nodes as well as other management related tasks for established
1873 * Return: true if the port supports an IB CM (this does not guarantee that
1874 * a CM is actually running however).
1876 static inline bool rdma_cap_ib_cm(struct ib_device *device, u8 port_num)
1878 return rdma_ib_or_iboe(device, port_num);
1882 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
1883 * Communication Manager.
1884 * @device: Device to check
1885 * @port_num: Port number to check
1887 * Similar to above, but specific to iWARP connections which have a different
1888 * managment protocol than InfiniBand.
1890 * Return: true if the port supports an iWARP CM (this does not guarantee that
1891 * a CM is actually running however).
1893 static inline bool rdma_cap_iw_cm(struct ib_device *device, u8 port_num)
1895 return rdma_protocol_iwarp(device, port_num);
1899 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
1900 * Subnet Administration.
1901 * @device: Device to check
1902 * @port_num: Port number to check
1904 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
1905 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
1906 * fabrics, devices should resolve routes to other hosts by contacting the
1907 * SA to query the proper route.
1909 * Return: true if the port should act as a client to the fabric Subnet
1910 * Administration interface. This does not imply that the SA service is
1913 static inline bool rdma_cap_ib_sa(struct ib_device *device, u8 port_num)
1915 return rdma_protocol_ib(device, port_num);
1919 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
1921 * @device: Device to check
1922 * @port_num: Port number to check
1924 * InfiniBand multicast registration is more complex than normal IPv4 or
1925 * IPv6 multicast registration. Each Host Channel Adapter must register
1926 * with the Subnet Manager when it wishes to join a multicast group. It
1927 * should do so only once regardless of how many queue pairs it subscribes
1928 * to this group. And it should leave the group only after all queue pairs
1929 * attached to the group have been detached.
1931 * Return: true if the port must undertake the additional adminstrative
1932 * overhead of registering/unregistering with the SM and tracking of the
1933 * total number of queue pairs attached to the multicast group.
1935 static inline bool rdma_cap_ib_mcast(struct ib_device *device, u8 port_num)
1937 return rdma_cap_ib_sa(device, port_num);
1941 * rdma_cap_af_ib - Check if the port of device has the capability
1942 * Native Infiniband Address.
1943 * @device: Device to check
1944 * @port_num: Port number to check
1946 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
1947 * GID. RoCE uses a different mechanism, but still generates a GID via
1948 * a prescribed mechanism and port specific data.
1950 * Return: true if the port uses a GID address to identify devices on the
1953 static inline bool rdma_cap_af_ib(struct ib_device *device, u8 port_num)
1955 return rdma_ib_or_iboe(device, port_num);
1959 * rdma_cap_eth_ah - Check if the port of device has the capability
1960 * Ethernet Address Handle.
1961 * @device: Device to check
1962 * @port_num: Port number to check
1964 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
1965 * to fabricate GIDs over Ethernet/IP specific addresses native to the
1966 * port. Normally, packet headers are generated by the sending host
1967 * adapter, but when sending connectionless datagrams, we must manually
1968 * inject the proper headers for the fabric we are communicating over.
1970 * Return: true if we are running as a RoCE port and must force the
1971 * addition of a Global Route Header built from our Ethernet Address
1972 * Handle into our header list for connectionless packets.
1974 static inline bool rdma_cap_eth_ah(struct ib_device *device, u8 port_num)
1976 return rdma_protocol_iboe(device, port_num);
1980 * rdma_cap_read_multi_sge - Check if the port of device has the capability
1981 * RDMA Read Multiple Scatter-Gather Entries.
1982 * @device: Device to check
1983 * @port_num: Port number to check
1985 * iWARP has a restriction that RDMA READ requests may only have a single
1986 * Scatter/Gather Entry (SGE) in the work request.
1988 * NOTE: although the linux kernel currently assumes all devices are either
1989 * single SGE RDMA READ devices or identical SGE maximums for RDMA READs and
1990 * WRITEs, according to Tom Talpey, this is not accurate. There are some
1991 * devices out there that support more than a single SGE on RDMA READ
1992 * requests, but do not support the same number of SGEs as they do on
1993 * RDMA WRITE requests. The linux kernel would need rearchitecting to
1994 * support these imbalanced READ/WRITE SGEs allowed devices. So, for now,
1995 * suffice with either the device supports the same READ/WRITE SGEs, or
1996 * it only gets one READ sge.
1998 * Return: true for any device that allows more than one SGE in RDMA READ
2001 static inline bool rdma_cap_read_multi_sge(struct ib_device *device,
2004 return !rdma_protocol_iwarp(device, port_num);
2007 int ib_query_gid(struct ib_device *device,
2008 u8 port_num, int index, union ib_gid *gid);
2010 int ib_query_pkey(struct ib_device *device,
2011 u8 port_num, u16 index, u16 *pkey);
2013 int ib_modify_device(struct ib_device *device,
2014 int device_modify_mask,
2015 struct ib_device_modify *device_modify);
2017 int ib_modify_port(struct ib_device *device,
2018 u8 port_num, int port_modify_mask,
2019 struct ib_port_modify *port_modify);
2021 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2022 u8 *port_num, u16 *index);
2024 int ib_find_pkey(struct ib_device *device,
2025 u8 port_num, u16 pkey, u16 *index);
2028 * ib_alloc_pd - Allocates an unused protection domain.
2029 * @device: The device on which to allocate the protection domain.
2031 * A protection domain object provides an association between QPs, shared
2032 * receive queues, address handles, memory regions, and memory windows.
2034 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2037 * ib_dealloc_pd - Deallocates a protection domain.
2038 * @pd: The protection domain to deallocate.
2040 int ib_dealloc_pd(struct ib_pd *pd);
2043 * ib_create_ah - Creates an address handle for the given address vector.
2044 * @pd: The protection domain associated with the address handle.
2045 * @ah_attr: The attributes of the address vector.
2047 * The address handle is used to reference a local or global destination
2048 * in all UD QP post sends.
2050 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2053 * ib_init_ah_from_wc - Initializes address handle attributes from a
2055 * @device: Device on which the received message arrived.
2056 * @port_num: Port on which the received message arrived.
2057 * @wc: Work completion associated with the received message.
2058 * @grh: References the received global route header. This parameter is
2059 * ignored unless the work completion indicates that the GRH is valid.
2060 * @ah_attr: Returned attributes that can be used when creating an address
2061 * handle for replying to the message.
2063 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
2064 struct ib_grh *grh, struct ib_ah_attr *ah_attr);
2067 * ib_create_ah_from_wc - Creates an address handle associated with the
2068 * sender of the specified work completion.
2069 * @pd: The protection domain associated with the address handle.
2070 * @wc: Work completion information associated with a received message.
2071 * @grh: References the received global route header. This parameter is
2072 * ignored unless the work completion indicates that the GRH is valid.
2073 * @port_num: The outbound port number to associate with the address.
2075 * The address handle is used to reference a local or global destination
2076 * in all UD QP post sends.
2078 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
2079 struct ib_grh *grh, u8 port_num);
2082 * ib_modify_ah - Modifies the address vector associated with an address
2084 * @ah: The address handle to modify.
2085 * @ah_attr: The new address vector attributes to associate with the
2088 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2091 * ib_query_ah - Queries the address vector associated with an address
2093 * @ah: The address handle to query.
2094 * @ah_attr: The address vector attributes associated with the address
2097 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2100 * ib_destroy_ah - Destroys an address handle.
2101 * @ah: The address handle to destroy.
2103 int ib_destroy_ah(struct ib_ah *ah);
2106 * ib_create_srq - Creates a SRQ associated with the specified protection
2108 * @pd: The protection domain associated with the SRQ.
2109 * @srq_init_attr: A list of initial attributes required to create the
2110 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2111 * the actual capabilities of the created SRQ.
2113 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2114 * requested size of the SRQ, and set to the actual values allocated
2115 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2116 * will always be at least as large as the requested values.
2118 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2119 struct ib_srq_init_attr *srq_init_attr);
2122 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2123 * @srq: The SRQ to modify.
2124 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2125 * the current values of selected SRQ attributes are returned.
2126 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2127 * are being modified.
2129 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2130 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2131 * the number of receives queued drops below the limit.
2133 int ib_modify_srq(struct ib_srq *srq,
2134 struct ib_srq_attr *srq_attr,
2135 enum ib_srq_attr_mask srq_attr_mask);
2138 * ib_query_srq - Returns the attribute list and current values for the
2140 * @srq: The SRQ to query.
2141 * @srq_attr: The attributes of the specified SRQ.
2143 int ib_query_srq(struct ib_srq *srq,
2144 struct ib_srq_attr *srq_attr);
2147 * ib_destroy_srq - Destroys the specified SRQ.
2148 * @srq: The SRQ to destroy.
2150 int ib_destroy_srq(struct ib_srq *srq);
2153 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2154 * @srq: The SRQ to post the work request on.
2155 * @recv_wr: A list of work requests to post on the receive queue.
2156 * @bad_recv_wr: On an immediate failure, this parameter will reference
2157 * the work request that failed to be posted on the QP.
2159 static inline int ib_post_srq_recv(struct ib_srq *srq,
2160 struct ib_recv_wr *recv_wr,
2161 struct ib_recv_wr **bad_recv_wr)
2163 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2167 * ib_create_qp - Creates a QP associated with the specified protection
2169 * @pd: The protection domain associated with the QP.
2170 * @qp_init_attr: A list of initial attributes required to create the
2171 * QP. If QP creation succeeds, then the attributes are updated to
2172 * the actual capabilities of the created QP.
2174 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2175 struct ib_qp_init_attr *qp_init_attr);
2178 * ib_modify_qp - Modifies the attributes for the specified QP and then
2179 * transitions the QP to the given state.
2180 * @qp: The QP to modify.
2181 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2182 * the current values of selected QP attributes are returned.
2183 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2184 * are being modified.
2186 int ib_modify_qp(struct ib_qp *qp,
2187 struct ib_qp_attr *qp_attr,
2191 * ib_query_qp - Returns the attribute list and current values for the
2193 * @qp: The QP to query.
2194 * @qp_attr: The attributes of the specified QP.
2195 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2196 * @qp_init_attr: Additional attributes of the selected QP.
2198 * The qp_attr_mask may be used to limit the query to gathering only the
2199 * selected attributes.
2201 int ib_query_qp(struct ib_qp *qp,
2202 struct ib_qp_attr *qp_attr,
2204 struct ib_qp_init_attr *qp_init_attr);
2207 * ib_destroy_qp - Destroys the specified QP.
2208 * @qp: The QP to destroy.
2210 int ib_destroy_qp(struct ib_qp *qp);
2213 * ib_open_qp - Obtain a reference to an existing sharable QP.
2214 * @xrcd - XRC domain
2215 * @qp_open_attr: Attributes identifying the QP to open.
2217 * Returns a reference to a sharable QP.
2219 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2220 struct ib_qp_open_attr *qp_open_attr);
2223 * ib_close_qp - Release an external reference to a QP.
2224 * @qp: The QP handle to release
2226 * The opened QP handle is released by the caller. The underlying
2227 * shared QP is not destroyed until all internal references are released.
2229 int ib_close_qp(struct ib_qp *qp);
2232 * ib_post_send - Posts a list of work requests to the send queue of
2234 * @qp: The QP to post the work request on.
2235 * @send_wr: A list of work requests to post on the send queue.
2236 * @bad_send_wr: On an immediate failure, this parameter will reference
2237 * the work request that failed to be posted on the QP.
2239 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2240 * error is returned, the QP state shall not be affected,
2241 * ib_post_send() will return an immediate error after queueing any
2242 * earlier work requests in the list.
2244 static inline int ib_post_send(struct ib_qp *qp,
2245 struct ib_send_wr *send_wr,
2246 struct ib_send_wr **bad_send_wr)
2248 return qp->device->post_send(qp, send_wr, bad_send_wr);
2252 * ib_post_recv - Posts a list of work requests to the receive queue of
2254 * @qp: The QP to post the work request on.
2255 * @recv_wr: A list of work requests to post on the receive queue.
2256 * @bad_recv_wr: On an immediate failure, this parameter will reference
2257 * the work request that failed to be posted on the QP.
2259 static inline int ib_post_recv(struct ib_qp *qp,
2260 struct ib_recv_wr *recv_wr,
2261 struct ib_recv_wr **bad_recv_wr)
2263 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2267 * ib_create_cq - Creates a CQ on the specified device.
2268 * @device: The device on which to create the CQ.
2269 * @comp_handler: A user-specified callback that is invoked when a
2270 * completion event occurs on the CQ.
2271 * @event_handler: A user-specified callback that is invoked when an
2272 * asynchronous event not associated with a completion occurs on the CQ.
2273 * @cq_context: Context associated with the CQ returned to the user via
2274 * the associated completion and event handlers.
2275 * @cqe: The minimum size of the CQ.
2276 * @comp_vector - Completion vector used to signal completion events.
2277 * Must be >= 0 and < context->num_comp_vectors.
2279 * Users can examine the cq structure to determine the actual CQ size.
2281 struct ib_cq *ib_create_cq(struct ib_device *device,
2282 ib_comp_handler comp_handler,
2283 void (*event_handler)(struct ib_event *, void *),
2284 void *cq_context, int cqe, int comp_vector);
2287 * ib_resize_cq - Modifies the capacity of the CQ.
2288 * @cq: The CQ to resize.
2289 * @cqe: The minimum size of the CQ.
2291 * Users can examine the cq structure to determine the actual CQ size.
2293 int ib_resize_cq(struct ib_cq *cq, int cqe);
2296 * ib_modify_cq - Modifies moderation params of the CQ
2297 * @cq: The CQ to modify.
2298 * @cq_count: number of CQEs that will trigger an event
2299 * @cq_period: max period of time in usec before triggering an event
2302 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2305 * ib_destroy_cq - Destroys the specified CQ.
2306 * @cq: The CQ to destroy.
2308 int ib_destroy_cq(struct ib_cq *cq);
2311 * ib_poll_cq - poll a CQ for completion(s)
2312 * @cq:the CQ being polled
2313 * @num_entries:maximum number of completions to return
2314 * @wc:array of at least @num_entries &struct ib_wc where completions
2317 * Poll a CQ for (possibly multiple) completions. If the return value
2318 * is < 0, an error occurred. If the return value is >= 0, it is the
2319 * number of completions returned. If the return value is
2320 * non-negative and < num_entries, then the CQ was emptied.
2322 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2325 return cq->device->poll_cq(cq, num_entries, wc);
2329 * ib_peek_cq - Returns the number of unreaped completions currently
2330 * on the specified CQ.
2331 * @cq: The CQ to peek.
2332 * @wc_cnt: A minimum number of unreaped completions to check for.
2334 * If the number of unreaped completions is greater than or equal to wc_cnt,
2335 * this function returns wc_cnt, otherwise, it returns the actual number of
2336 * unreaped completions.
2338 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2341 * ib_req_notify_cq - Request completion notification on a CQ.
2342 * @cq: The CQ to generate an event for.
2344 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2345 * to request an event on the next solicited event or next work
2346 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2347 * may also be |ed in to request a hint about missed events, as
2351 * < 0 means an error occurred while requesting notification
2352 * == 0 means notification was requested successfully, and if
2353 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2354 * were missed and it is safe to wait for another event. In
2355 * this case is it guaranteed that any work completions added
2356 * to the CQ since the last CQ poll will trigger a completion
2357 * notification event.
2358 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2359 * in. It means that the consumer must poll the CQ again to
2360 * make sure it is empty to avoid missing an event because of a
2361 * race between requesting notification and an entry being
2362 * added to the CQ. This return value means it is possible
2363 * (but not guaranteed) that a work completion has been added
2364 * to the CQ since the last poll without triggering a
2365 * completion notification event.
2367 static inline int ib_req_notify_cq(struct ib_cq *cq,
2368 enum ib_cq_notify_flags flags)
2370 return cq->device->req_notify_cq(cq, flags);
2374 * ib_req_ncomp_notif - Request completion notification when there are
2375 * at least the specified number of unreaped completions on the CQ.
2376 * @cq: The CQ to generate an event for.
2377 * @wc_cnt: The number of unreaped completions that should be on the
2378 * CQ before an event is generated.
2380 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2382 return cq->device->req_ncomp_notif ?
2383 cq->device->req_ncomp_notif(cq, wc_cnt) :
2388 * ib_get_dma_mr - Returns a memory region for system memory that is
2390 * @pd: The protection domain associated with the memory region.
2391 * @mr_access_flags: Specifies the memory access rights.
2393 * Note that the ib_dma_*() functions defined below must be used
2394 * to create/destroy addresses used with the Lkey or Rkey returned
2395 * by ib_get_dma_mr().
2397 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2400 * ib_dma_mapping_error - check a DMA addr for error
2401 * @dev: The device for which the dma_addr was created
2402 * @dma_addr: The DMA address to check
2404 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2407 return dev->dma_ops->mapping_error(dev, dma_addr);
2408 return dma_mapping_error(dev->dma_device, dma_addr);
2412 * ib_dma_map_single - Map a kernel virtual address to DMA address
2413 * @dev: The device for which the dma_addr is to be created
2414 * @cpu_addr: The kernel virtual address
2415 * @size: The size of the region in bytes
2416 * @direction: The direction of the DMA
2418 static inline u64 ib_dma_map_single(struct ib_device *dev,
2419 void *cpu_addr, size_t size,
2420 enum dma_data_direction direction)
2423 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2424 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2428 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2429 * @dev: The device for which the DMA address was created
2430 * @addr: The DMA address
2431 * @size: The size of the region in bytes
2432 * @direction: The direction of the DMA
2434 static inline void ib_dma_unmap_single(struct ib_device *dev,
2435 u64 addr, size_t size,
2436 enum dma_data_direction direction)
2439 dev->dma_ops->unmap_single(dev, addr, size, direction);
2441 dma_unmap_single(dev->dma_device, addr, size, direction);
2444 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2445 void *cpu_addr, size_t size,
2446 enum dma_data_direction direction,
2447 struct dma_attrs *attrs)
2449 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2453 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2454 u64 addr, size_t size,
2455 enum dma_data_direction direction,
2456 struct dma_attrs *attrs)
2458 return dma_unmap_single_attrs(dev->dma_device, addr, size,
2463 * ib_dma_map_page - Map a physical page to DMA address
2464 * @dev: The device for which the dma_addr is to be created
2465 * @page: The page to be mapped
2466 * @offset: The offset within the page
2467 * @size: The size of the region in bytes
2468 * @direction: The direction of the DMA
2470 static inline u64 ib_dma_map_page(struct ib_device *dev,
2472 unsigned long offset,
2474 enum dma_data_direction direction)
2477 return dev->dma_ops->map_page(dev, page, offset, size, direction);
2478 return dma_map_page(dev->dma_device, page, offset, size, direction);
2482 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2483 * @dev: The device for which the DMA address was created
2484 * @addr: The DMA address
2485 * @size: The size of the region in bytes
2486 * @direction: The direction of the DMA
2488 static inline void ib_dma_unmap_page(struct ib_device *dev,
2489 u64 addr, size_t size,
2490 enum dma_data_direction direction)
2493 dev->dma_ops->unmap_page(dev, addr, size, direction);
2495 dma_unmap_page(dev->dma_device, addr, size, direction);
2499 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2500 * @dev: The device for which the DMA addresses are to be created
2501 * @sg: The array of scatter/gather entries
2502 * @nents: The number of scatter/gather entries
2503 * @direction: The direction of the DMA
2505 static inline int ib_dma_map_sg(struct ib_device *dev,
2506 struct scatterlist *sg, int nents,
2507 enum dma_data_direction direction)
2510 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2511 return dma_map_sg(dev->dma_device, sg, nents, direction);
2515 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2516 * @dev: The device for which the DMA addresses were created
2517 * @sg: The array of scatter/gather entries
2518 * @nents: The number of scatter/gather entries
2519 * @direction: The direction of the DMA
2521 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2522 struct scatterlist *sg, int nents,
2523 enum dma_data_direction direction)
2526 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2528 dma_unmap_sg(dev->dma_device, sg, nents, direction);
2531 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2532 struct scatterlist *sg, int nents,
2533 enum dma_data_direction direction,
2534 struct dma_attrs *attrs)
2536 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2539 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2540 struct scatterlist *sg, int nents,
2541 enum dma_data_direction direction,
2542 struct dma_attrs *attrs)
2544 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2547 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2548 * @dev: The device for which the DMA addresses were created
2549 * @sg: The scatter/gather entry
2551 * Note: this function is obsolete. To do: change all occurrences of
2552 * ib_sg_dma_address() into sg_dma_address().
2554 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2555 struct scatterlist *sg)
2557 return sg_dma_address(sg);
2561 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2562 * @dev: The device for which the DMA addresses were created
2563 * @sg: The scatter/gather entry
2565 * Note: this function is obsolete. To do: change all occurrences of
2566 * ib_sg_dma_len() into sg_dma_len().
2568 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2569 struct scatterlist *sg)
2571 return sg_dma_len(sg);
2575 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2576 * @dev: The device for which the DMA address was created
2577 * @addr: The DMA address
2578 * @size: The size of the region in bytes
2579 * @dir: The direction of the DMA
2581 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2584 enum dma_data_direction dir)
2587 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2589 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2593 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
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 * @dir: The direction of the DMA
2599 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2602 enum dma_data_direction dir)
2605 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2607 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2611 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2612 * @dev: The device for which the DMA address is requested
2613 * @size: The size of the region to allocate in bytes
2614 * @dma_handle: A pointer for returning the DMA address of the region
2615 * @flag: memory allocator flags
2617 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2623 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2628 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2629 *dma_handle = handle;
2635 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2636 * @dev: The device for which the DMA addresses were allocated
2637 * @size: The size of the region
2638 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2639 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2641 static inline void ib_dma_free_coherent(struct ib_device *dev,
2642 size_t size, void *cpu_addr,
2646 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2648 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2652 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
2654 * @pd: The protection domain associated assigned to the registered region.
2655 * @phys_buf_array: Specifies a list of physical buffers to use in the
2657 * @num_phys_buf: Specifies the size of the phys_buf_array.
2658 * @mr_access_flags: Specifies the memory access rights.
2659 * @iova_start: The offset of the region's starting I/O virtual address.
2661 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
2662 struct ib_phys_buf *phys_buf_array,
2664 int mr_access_flags,
2668 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
2669 * Conceptually, this call performs the functions deregister memory region
2670 * followed by register physical memory region. Where possible,
2671 * resources are reused instead of deallocated and reallocated.
2672 * @mr: The memory region to modify.
2673 * @mr_rereg_mask: A bit-mask used to indicate which of the following
2674 * properties of the memory region are being modified.
2675 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
2676 * the new protection domain to associated with the memory region,
2677 * otherwise, this parameter is ignored.
2678 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2679 * field specifies a list of physical buffers to use in the new
2680 * translation, otherwise, this parameter is ignored.
2681 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2682 * field specifies the size of the phys_buf_array, otherwise, this
2683 * parameter is ignored.
2684 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
2685 * field specifies the new memory access rights, otherwise, this
2686 * parameter is ignored.
2687 * @iova_start: The offset of the region's starting I/O virtual address.
2689 int ib_rereg_phys_mr(struct ib_mr *mr,
2692 struct ib_phys_buf *phys_buf_array,
2694 int mr_access_flags,
2698 * ib_query_mr - Retrieves information about a specific memory region.
2699 * @mr: The memory region to retrieve information about.
2700 * @mr_attr: The attributes of the specified memory region.
2702 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2705 * ib_dereg_mr - Deregisters a memory region and removes it from the
2706 * HCA translation table.
2707 * @mr: The memory region to deregister.
2709 * This function can fail, if the memory region has memory windows bound to it.
2711 int ib_dereg_mr(struct ib_mr *mr);
2715 * ib_create_mr - Allocates a memory region that may be used for
2716 * signature handover operations.
2717 * @pd: The protection domain associated with the region.
2718 * @mr_init_attr: memory region init attributes.
2720 struct ib_mr *ib_create_mr(struct ib_pd *pd,
2721 struct ib_mr_init_attr *mr_init_attr);
2724 * ib_destroy_mr - Destroys a memory region that was created using
2725 * ib_create_mr and removes it from HW translation tables.
2726 * @mr: The memory region to destroy.
2728 * This function can fail, if the memory region has memory windows bound to it.
2730 int ib_destroy_mr(struct ib_mr *mr);
2733 * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2734 * IB_WR_FAST_REG_MR send work request.
2735 * @pd: The protection domain associated with the region.
2736 * @max_page_list_len: requested max physical buffer list length to be
2737 * used with fast register work requests for this MR.
2739 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2742 * ib_alloc_fast_reg_page_list - Allocates a page list array
2743 * @device - ib device pointer.
2744 * @page_list_len - size of the page list array to be allocated.
2746 * This allocates and returns a struct ib_fast_reg_page_list * and a
2747 * page_list array that is at least page_list_len in size. The actual
2748 * size is returned in max_page_list_len. The caller is responsible
2749 * for initializing the contents of the page_list array before posting
2750 * a send work request with the IB_WC_FAST_REG_MR opcode.
2752 * The page_list array entries must be translated using one of the
2753 * ib_dma_*() functions just like the addresses passed to
2754 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
2755 * ib_fast_reg_page_list must not be modified by the caller until the
2756 * IB_WC_FAST_REG_MR work request completes.
2758 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2759 struct ib_device *device, int page_list_len);
2762 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2764 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2766 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2769 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2771 * @mr - struct ib_mr pointer to be updated.
2772 * @newkey - new key to be used.
2774 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2776 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2777 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2781 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2782 * for calculating a new rkey for type 2 memory windows.
2783 * @rkey - the rkey to increment.
2785 static inline u32 ib_inc_rkey(u32 rkey)
2787 const u32 mask = 0x000000ff;
2788 return ((rkey + 1) & mask) | (rkey & ~mask);
2792 * ib_alloc_mw - Allocates a memory window.
2793 * @pd: The protection domain associated with the memory window.
2794 * @type: The type of the memory window (1 or 2).
2796 struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
2799 * ib_bind_mw - Posts a work request to the send queue of the specified
2800 * QP, which binds the memory window to the given address range and
2801 * remote access attributes.
2802 * @qp: QP to post the bind work request on.
2803 * @mw: The memory window to bind.
2804 * @mw_bind: Specifies information about the memory window, including
2805 * its address range, remote access rights, and associated memory region.
2807 * If there is no immediate error, the function will update the rkey member
2808 * of the mw parameter to its new value. The bind operation can still fail
2811 static inline int ib_bind_mw(struct ib_qp *qp,
2813 struct ib_mw_bind *mw_bind)
2815 /* XXX reference counting in corresponding MR? */
2816 return mw->device->bind_mw ?
2817 mw->device->bind_mw(qp, mw, mw_bind) :
2822 * ib_dealloc_mw - Deallocates a memory window.
2823 * @mw: The memory window to deallocate.
2825 int ib_dealloc_mw(struct ib_mw *mw);
2828 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2829 * @pd: The protection domain associated with the unmapped region.
2830 * @mr_access_flags: Specifies the memory access rights.
2831 * @fmr_attr: Attributes of the unmapped region.
2833 * A fast memory region must be mapped before it can be used as part of
2836 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2837 int mr_access_flags,
2838 struct ib_fmr_attr *fmr_attr);
2841 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2842 * @fmr: The fast memory region to associate with the pages.
2843 * @page_list: An array of physical pages to map to the fast memory region.
2844 * @list_len: The number of pages in page_list.
2845 * @iova: The I/O virtual address to use with the mapped region.
2847 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2848 u64 *page_list, int list_len,
2851 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2855 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2856 * @fmr_list: A linked list of fast memory regions to unmap.
2858 int ib_unmap_fmr(struct list_head *fmr_list);
2861 * ib_dealloc_fmr - Deallocates a fast memory region.
2862 * @fmr: The fast memory region to deallocate.
2864 int ib_dealloc_fmr(struct ib_fmr *fmr);
2867 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2868 * @qp: QP to attach to the multicast group. The QP must be type
2870 * @gid: Multicast group GID.
2871 * @lid: Multicast group LID in host byte order.
2873 * In order to send and receive multicast packets, subnet
2874 * administration must have created the multicast group and configured
2875 * the fabric appropriately. The port associated with the specified
2876 * QP must also be a member of the multicast group.
2878 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2881 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2882 * @qp: QP to detach from the multicast group.
2883 * @gid: Multicast group GID.
2884 * @lid: Multicast group LID in host byte order.
2886 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2889 * ib_alloc_xrcd - Allocates an XRC domain.
2890 * @device: The device on which to allocate the XRC domain.
2892 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2895 * ib_dealloc_xrcd - Deallocates an XRC domain.
2896 * @xrcd: The XRC domain to deallocate.
2898 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2900 struct ib_flow *ib_create_flow(struct ib_qp *qp,
2901 struct ib_flow_attr *flow_attr, int domain);
2902 int ib_destroy_flow(struct ib_flow *flow_id);
2904 static inline int ib_check_mr_access(int flags)
2907 * Local write permission is required if remote write or
2908 * remote atomic permission is also requested.
2910 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
2911 !(flags & IB_ACCESS_LOCAL_WRITE))
2918 * ib_check_mr_status: lightweight check of MR status.
2919 * This routine may provide status checks on a selected
2920 * ib_mr. first use is for signature status check.
2922 * @mr: A memory region.
2923 * @check_mask: Bitmask of which checks to perform from
2924 * ib_mr_status_check enumeration.
2925 * @mr_status: The container of relevant status checks.
2926 * failed checks will be indicated in the status bitmask
2927 * and the relevant info shall be in the error item.
2929 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
2930 struct ib_mr_status *mr_status);
2932 #endif /* IB_VERBS_H */