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>
52 #include <linux/atomic.h>
53 #include <asm/uaccess.h>
55 extern struct workqueue_struct *ib_wq;
66 /* IB values map to NodeInfo:NodeType. */
74 enum rdma_transport_type {
80 enum rdma_transport_type
81 rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__;
83 enum rdma_link_layer {
84 IB_LINK_LAYER_UNSPECIFIED,
85 IB_LINK_LAYER_INFINIBAND,
86 IB_LINK_LAYER_ETHERNET,
89 enum ib_device_cap_flags {
90 IB_DEVICE_RESIZE_MAX_WR = 1,
91 IB_DEVICE_BAD_PKEY_CNTR = (1<<1),
92 IB_DEVICE_BAD_QKEY_CNTR = (1<<2),
93 IB_DEVICE_RAW_MULTI = (1<<3),
94 IB_DEVICE_AUTO_PATH_MIG = (1<<4),
95 IB_DEVICE_CHANGE_PHY_PORT = (1<<5),
96 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6),
97 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7),
98 IB_DEVICE_SHUTDOWN_PORT = (1<<8),
99 IB_DEVICE_INIT_TYPE = (1<<9),
100 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10),
101 IB_DEVICE_SYS_IMAGE_GUID = (1<<11),
102 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12),
103 IB_DEVICE_SRQ_RESIZE = (1<<13),
104 IB_DEVICE_N_NOTIFY_CQ = (1<<14),
105 IB_DEVICE_LOCAL_DMA_LKEY = (1<<15),
106 IB_DEVICE_RESERVED = (1<<16), /* old SEND_W_INV */
107 IB_DEVICE_MEM_WINDOW = (1<<17),
109 * Devices should set IB_DEVICE_UD_IP_SUM if they support
110 * insertion of UDP and TCP checksum on outgoing UD IPoIB
111 * messages and can verify the validity of checksum for
112 * incoming messages. Setting this flag implies that the
113 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
115 IB_DEVICE_UD_IP_CSUM = (1<<18),
116 IB_DEVICE_UD_TSO = (1<<19),
117 IB_DEVICE_XRC = (1<<20),
118 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21),
119 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
120 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1<<23),
121 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1<<24),
122 IB_DEVICE_MANAGED_FLOW_STEERING = (1<<29)
131 struct ib_device_attr {
133 __be64 sys_image_guid;
141 int device_cap_flags;
151 int max_qp_init_rd_atom;
152 int max_ee_init_rd_atom;
153 enum ib_atomic_cap atomic_cap;
154 enum ib_atomic_cap masked_atomic_cap;
161 int max_mcast_qp_attach;
162 int max_total_mcast_qp_attach;
169 unsigned int max_fast_reg_page_list_len;
171 u8 local_ca_ack_delay;
182 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
185 case IB_MTU_256: return 256;
186 case IB_MTU_512: return 512;
187 case IB_MTU_1024: return 1024;
188 case IB_MTU_2048: return 2048;
189 case IB_MTU_4096: return 4096;
200 IB_PORT_ACTIVE_DEFER = 5
203 enum ib_port_cap_flags {
205 IB_PORT_NOTICE_SUP = 1 << 2,
206 IB_PORT_TRAP_SUP = 1 << 3,
207 IB_PORT_OPT_IPD_SUP = 1 << 4,
208 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
209 IB_PORT_SL_MAP_SUP = 1 << 6,
210 IB_PORT_MKEY_NVRAM = 1 << 7,
211 IB_PORT_PKEY_NVRAM = 1 << 8,
212 IB_PORT_LED_INFO_SUP = 1 << 9,
213 IB_PORT_SM_DISABLED = 1 << 10,
214 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
215 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
216 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
217 IB_PORT_CM_SUP = 1 << 16,
218 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
219 IB_PORT_REINIT_SUP = 1 << 18,
220 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
221 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
222 IB_PORT_DR_NOTICE_SUP = 1 << 21,
223 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
224 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
225 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
226 IB_PORT_CLIENT_REG_SUP = 1 << 25
236 static inline int ib_width_enum_to_int(enum ib_port_width width)
239 case IB_WIDTH_1X: return 1;
240 case IB_WIDTH_4X: return 4;
241 case IB_WIDTH_8X: return 8;
242 case IB_WIDTH_12X: return 12;
256 struct ib_protocol_stats {
260 struct iw_protocol_stats {
263 u64 ipInTooBigErrors;
266 u64 ipInUnknownProtos;
267 u64 ipInTruncatedPkts;
270 u64 ipOutForwDatagrams;
302 union rdma_protocol_stats {
303 struct ib_protocol_stats ib;
304 struct iw_protocol_stats iw;
307 struct ib_port_attr {
308 enum ib_port_state state;
310 enum ib_mtu active_mtu;
329 enum ib_device_modify_flags {
330 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
331 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
334 struct ib_device_modify {
339 enum ib_port_modify_flags {
340 IB_PORT_SHUTDOWN = 1,
341 IB_PORT_INIT_TYPE = (1<<2),
342 IB_PORT_RESET_QKEY_CNTR = (1<<3)
345 struct ib_port_modify {
346 u32 set_port_cap_mask;
347 u32 clr_port_cap_mask;
355 IB_EVENT_QP_ACCESS_ERR,
359 IB_EVENT_PATH_MIG_ERR,
360 IB_EVENT_DEVICE_FATAL,
361 IB_EVENT_PORT_ACTIVE,
364 IB_EVENT_PKEY_CHANGE,
367 IB_EVENT_SRQ_LIMIT_REACHED,
368 IB_EVENT_QP_LAST_WQE_REACHED,
369 IB_EVENT_CLIENT_REREGISTER,
374 struct ib_device *device;
381 enum ib_event_type event;
384 struct ib_event_handler {
385 struct ib_device *device;
386 void (*handler)(struct ib_event_handler *, struct ib_event *);
387 struct list_head list;
390 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
392 (_ptr)->device = _device; \
393 (_ptr)->handler = _handler; \
394 INIT_LIST_HEAD(&(_ptr)->list); \
397 struct ib_global_route {
406 __be32 version_tclass_flow;
415 IB_MULTICAST_QPN = 0xffffff
418 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
425 IB_RATE_PORT_CURRENT = 0,
426 IB_RATE_2_5_GBPS = 2,
434 IB_RATE_120_GBPS = 10,
435 IB_RATE_14_GBPS = 11,
436 IB_RATE_56_GBPS = 12,
437 IB_RATE_112_GBPS = 13,
438 IB_RATE_168_GBPS = 14,
439 IB_RATE_25_GBPS = 15,
440 IB_RATE_100_GBPS = 16,
441 IB_RATE_200_GBPS = 17,
442 IB_RATE_300_GBPS = 18
446 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
447 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
448 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
449 * @rate: rate to convert.
451 int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
454 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
455 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
456 * @rate: rate to convert.
458 int ib_rate_to_mbps(enum ib_rate rate) __attribute_const__;
461 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
463 * @mult: multiple to convert.
465 enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
468 struct ib_global_route grh;
481 IB_WC_LOC_EEC_OP_ERR,
486 IB_WC_LOC_ACCESS_ERR,
487 IB_WC_REM_INV_REQ_ERR,
488 IB_WC_REM_ACCESS_ERR,
491 IB_WC_RNR_RETRY_EXC_ERR,
492 IB_WC_LOC_RDD_VIOL_ERR,
493 IB_WC_REM_INV_RD_REQ_ERR,
496 IB_WC_INV_EEC_STATE_ERR,
498 IB_WC_RESP_TIMEOUT_ERR,
512 IB_WC_MASKED_COMP_SWAP,
513 IB_WC_MASKED_FETCH_ADD,
515 * Set value of IB_WC_RECV so consumers can test if a completion is a
516 * receive by testing (opcode & IB_WC_RECV).
519 IB_WC_RECV_RDMA_WITH_IMM
524 IB_WC_WITH_IMM = (1<<1),
525 IB_WC_WITH_INVALIDATE = (1<<2),
526 IB_WC_IP_CSUM_OK = (1<<3),
531 enum ib_wc_status status;
532 enum ib_wc_opcode opcode;
546 u8 port_num; /* valid only for DR SMPs on switches */
549 enum ib_cq_notify_flags {
550 IB_CQ_SOLICITED = 1 << 0,
551 IB_CQ_NEXT_COMP = 1 << 1,
552 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
553 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
561 enum ib_srq_attr_mask {
562 IB_SRQ_MAX_WR = 1 << 0,
563 IB_SRQ_LIMIT = 1 << 1,
572 struct ib_srq_init_attr {
573 void (*event_handler)(struct ib_event *, void *);
575 struct ib_srq_attr attr;
576 enum ib_srq_type srq_type;
580 struct ib_xrcd *xrcd;
601 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
602 * here (and in that order) since the MAD layer uses them as
603 * indices into a 2-entry table.
612 IB_QPT_RAW_ETHERTYPE,
613 IB_QPT_RAW_PACKET = 8,
617 /* Reserve a range for qp types internal to the low level driver.
618 * These qp types will not be visible at the IB core layer, so the
619 * IB_QPT_MAX usages should not be affected in the core layer
621 IB_QPT_RESERVED1 = 0x1000,
633 enum ib_qp_create_flags {
634 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
635 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
636 /* reserve bits 26-31 for low level drivers' internal use */
637 IB_QP_CREATE_RESERVED_START = 1 << 26,
638 IB_QP_CREATE_RESERVED_END = 1 << 31,
643 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
644 * callback to destroy the passed in QP.
647 struct ib_qp_init_attr {
648 void (*event_handler)(struct ib_event *, void *);
650 struct ib_cq *send_cq;
651 struct ib_cq *recv_cq;
653 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
654 struct ib_qp_cap cap;
655 enum ib_sig_type sq_sig_type;
656 enum ib_qp_type qp_type;
657 enum ib_qp_create_flags create_flags;
658 u8 port_num; /* special QP types only */
661 struct ib_qp_open_attr {
662 void (*event_handler)(struct ib_event *, void *);
665 enum ib_qp_type qp_type;
668 enum ib_rnr_timeout {
669 IB_RNR_TIMER_655_36 = 0,
670 IB_RNR_TIMER_000_01 = 1,
671 IB_RNR_TIMER_000_02 = 2,
672 IB_RNR_TIMER_000_03 = 3,
673 IB_RNR_TIMER_000_04 = 4,
674 IB_RNR_TIMER_000_06 = 5,
675 IB_RNR_TIMER_000_08 = 6,
676 IB_RNR_TIMER_000_12 = 7,
677 IB_RNR_TIMER_000_16 = 8,
678 IB_RNR_TIMER_000_24 = 9,
679 IB_RNR_TIMER_000_32 = 10,
680 IB_RNR_TIMER_000_48 = 11,
681 IB_RNR_TIMER_000_64 = 12,
682 IB_RNR_TIMER_000_96 = 13,
683 IB_RNR_TIMER_001_28 = 14,
684 IB_RNR_TIMER_001_92 = 15,
685 IB_RNR_TIMER_002_56 = 16,
686 IB_RNR_TIMER_003_84 = 17,
687 IB_RNR_TIMER_005_12 = 18,
688 IB_RNR_TIMER_007_68 = 19,
689 IB_RNR_TIMER_010_24 = 20,
690 IB_RNR_TIMER_015_36 = 21,
691 IB_RNR_TIMER_020_48 = 22,
692 IB_RNR_TIMER_030_72 = 23,
693 IB_RNR_TIMER_040_96 = 24,
694 IB_RNR_TIMER_061_44 = 25,
695 IB_RNR_TIMER_081_92 = 26,
696 IB_RNR_TIMER_122_88 = 27,
697 IB_RNR_TIMER_163_84 = 28,
698 IB_RNR_TIMER_245_76 = 29,
699 IB_RNR_TIMER_327_68 = 30,
700 IB_RNR_TIMER_491_52 = 31
703 enum ib_qp_attr_mask {
705 IB_QP_CUR_STATE = (1<<1),
706 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
707 IB_QP_ACCESS_FLAGS = (1<<3),
708 IB_QP_PKEY_INDEX = (1<<4),
712 IB_QP_PATH_MTU = (1<<8),
713 IB_QP_TIMEOUT = (1<<9),
714 IB_QP_RETRY_CNT = (1<<10),
715 IB_QP_RNR_RETRY = (1<<11),
716 IB_QP_RQ_PSN = (1<<12),
717 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
718 IB_QP_ALT_PATH = (1<<14),
719 IB_QP_MIN_RNR_TIMER = (1<<15),
720 IB_QP_SQ_PSN = (1<<16),
721 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
722 IB_QP_PATH_MIG_STATE = (1<<18),
724 IB_QP_DEST_QPN = (1<<20)
749 enum ib_qp_state qp_state;
750 enum ib_qp_state cur_qp_state;
751 enum ib_mtu path_mtu;
752 enum ib_mig_state path_mig_state;
758 struct ib_qp_cap cap;
759 struct ib_ah_attr ah_attr;
760 struct ib_ah_attr alt_ah_attr;
763 u8 en_sqd_async_notify;
766 u8 max_dest_rd_atomic;
778 IB_WR_RDMA_WRITE_WITH_IMM,
782 IB_WR_ATOMIC_CMP_AND_SWP,
783 IB_WR_ATOMIC_FETCH_AND_ADD,
786 IB_WR_RDMA_READ_WITH_INV,
789 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
790 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
792 /* reserve values for low level drivers' internal use.
793 * These values will not be used at all in the ib core layer.
795 IB_WR_RESERVED1 = 0xf0,
809 IB_SEND_SIGNALED = (1<<1),
810 IB_SEND_SOLICITED = (1<<2),
811 IB_SEND_INLINE = (1<<3),
812 IB_SEND_IP_CSUM = (1<<4),
814 /* reserve bits 26-31 for low level drivers' internal use */
815 IB_SEND_RESERVED_START = (1 << 26),
816 IB_SEND_RESERVED_END = (1 << 31),
825 struct ib_fast_reg_page_list {
826 struct ib_device *device;
828 unsigned int max_page_list_len;
832 * struct ib_mw_bind_info - Parameters for a memory window bind operation.
833 * @mr: A memory region to bind the memory window to.
834 * @addr: The address where the memory window should begin.
835 * @length: The length of the memory window, in bytes.
836 * @mw_access_flags: Access flags from enum ib_access_flags for the window.
838 * This struct contains the shared parameters for type 1 and type 2
839 * memory window bind operations.
841 struct ib_mw_bind_info {
849 struct ib_send_wr *next;
851 struct ib_sge *sg_list;
853 enum ib_wr_opcode opcode;
868 u64 compare_add_mask;
879 u16 pkey_index; /* valid for GSI only */
880 u8 port_num; /* valid for DR SMPs on switch only */
884 struct ib_fast_reg_page_list *page_list;
885 unsigned int page_shift;
886 unsigned int page_list_len;
893 /* The new rkey for the memory window. */
895 struct ib_mw_bind_info bind_info;
898 u32 xrc_remote_srq_num; /* XRC TGT QPs only */
902 struct ib_recv_wr *next;
904 struct ib_sge *sg_list;
908 enum ib_access_flags {
909 IB_ACCESS_LOCAL_WRITE = 1,
910 IB_ACCESS_REMOTE_WRITE = (1<<1),
911 IB_ACCESS_REMOTE_READ = (1<<2),
912 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
913 IB_ACCESS_MW_BIND = (1<<4),
914 IB_ZERO_BASED = (1<<5)
924 u64 device_virt_addr;
931 enum ib_mr_rereg_flags {
932 IB_MR_REREG_TRANS = 1,
933 IB_MR_REREG_PD = (1<<1),
934 IB_MR_REREG_ACCESS = (1<<2)
938 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
939 * @wr_id: Work request id.
940 * @send_flags: Flags from ib_send_flags enum.
941 * @bind_info: More parameters of the bind operation.
946 struct ib_mw_bind_info bind_info;
956 struct ib_device *device;
957 struct list_head pd_list;
958 struct list_head mr_list;
959 struct list_head mw_list;
960 struct list_head cq_list;
961 struct list_head qp_list;
962 struct list_head srq_list;
963 struct list_head ah_list;
964 struct list_head xrcd_list;
965 struct list_head rule_list;
970 u64 user_handle; /* handle given to us by userspace */
971 struct ib_ucontext *context; /* associated user context */
972 void *object; /* containing object */
973 struct list_head list; /* link to context's list */
974 int id; /* index into kernel idr */
976 struct rw_semaphore mutex; /* protects .live */
988 struct ib_device *device;
989 struct ib_uobject *uobject;
990 atomic_t usecnt; /* count all resources */
994 struct ib_device *device;
995 atomic_t usecnt; /* count all exposed resources */
998 struct mutex tgt_qp_mutex;
999 struct list_head tgt_qp_list;
1003 struct ib_device *device;
1005 struct ib_uobject *uobject;
1008 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1011 struct ib_device *device;
1012 struct ib_uobject *uobject;
1013 ib_comp_handler comp_handler;
1014 void (*event_handler)(struct ib_event *, void *);
1017 atomic_t usecnt; /* count number of work queues */
1021 struct ib_device *device;
1023 struct ib_uobject *uobject;
1024 void (*event_handler)(struct ib_event *, void *);
1026 enum ib_srq_type srq_type;
1031 struct ib_xrcd *xrcd;
1039 struct ib_device *device;
1041 struct ib_cq *send_cq;
1042 struct ib_cq *recv_cq;
1044 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1045 struct list_head xrcd_list;
1046 /* count times opened, mcast attaches, flow attaches */
1048 struct list_head open_list;
1049 struct ib_qp *real_qp;
1050 struct ib_uobject *uobject;
1051 void (*event_handler)(struct ib_event *, void *);
1054 enum ib_qp_type qp_type;
1058 struct ib_device *device;
1060 struct ib_uobject *uobject;
1063 atomic_t usecnt; /* count number of MWs */
1067 struct ib_device *device;
1069 struct ib_uobject *uobject;
1071 enum ib_mw_type type;
1075 struct ib_device *device;
1077 struct list_head list;
1082 /* Supported steering options */
1083 enum ib_flow_attr_type {
1084 /* steering according to rule specifications */
1085 IB_FLOW_ATTR_NORMAL = 0x0,
1086 /* default unicast and multicast rule -
1087 * receive all Eth traffic which isn't steered to any QP
1089 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1090 /* default multicast rule -
1091 * receive all Eth multicast traffic which isn't steered to any QP
1093 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1094 /* sniffer rule - receive all port traffic */
1095 IB_FLOW_ATTR_SNIFFER = 0x3
1098 /* Supported steering header types */
1099 enum ib_flow_spec_type {
1101 IB_FLOW_SPEC_ETH = 0x20,
1103 IB_FLOW_SPEC_IPV4 = 0x30,
1105 IB_FLOW_SPEC_TCP = 0x40,
1106 IB_FLOW_SPEC_UDP = 0x41
1109 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1111 /* Flow steering rule priority is set according to it's domain.
1112 * Lower domain value means higher priority.
1114 enum ib_flow_domain {
1115 IB_FLOW_DOMAIN_USER,
1116 IB_FLOW_DOMAIN_ETHTOOL,
1119 IB_FLOW_DOMAIN_NUM /* Must be last */
1122 struct ib_flow_eth_filter {
1129 struct ib_flow_spec_eth {
1130 enum ib_flow_spec_type type;
1132 struct ib_flow_eth_filter val;
1133 struct ib_flow_eth_filter mask;
1136 struct ib_flow_ipv4_filter {
1141 struct ib_flow_spec_ipv4 {
1142 enum ib_flow_spec_type type;
1144 struct ib_flow_ipv4_filter val;
1145 struct ib_flow_ipv4_filter mask;
1148 struct ib_flow_tcp_udp_filter {
1153 struct ib_flow_spec_tcp_udp {
1154 enum ib_flow_spec_type type;
1156 struct ib_flow_tcp_udp_filter val;
1157 struct ib_flow_tcp_udp_filter mask;
1160 union ib_flow_spec {
1162 enum ib_flow_spec_type type;
1165 struct ib_flow_spec_eth eth;
1166 struct ib_flow_spec_ipv4 ipv4;
1167 struct ib_flow_spec_tcp_udp tcp_udp;
1170 struct ib_flow_attr {
1171 enum ib_flow_attr_type type;
1177 /* Following are the optional layers according to user request
1178 * struct ib_flow_spec_xxx
1179 * struct ib_flow_spec_yyy
1185 struct ib_uobject *uobject;
1191 enum ib_process_mad_flags {
1192 IB_MAD_IGNORE_MKEY = 1,
1193 IB_MAD_IGNORE_BKEY = 2,
1194 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1197 enum ib_mad_result {
1198 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1199 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1200 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1201 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1204 #define IB_DEVICE_NAME_MAX 64
1208 struct ib_event_handler event_handler;
1209 struct ib_pkey_cache **pkey_cache;
1210 struct ib_gid_cache **gid_cache;
1214 struct ib_dma_mapping_ops {
1215 int (*mapping_error)(struct ib_device *dev,
1217 u64 (*map_single)(struct ib_device *dev,
1218 void *ptr, size_t size,
1219 enum dma_data_direction direction);
1220 void (*unmap_single)(struct ib_device *dev,
1221 u64 addr, size_t size,
1222 enum dma_data_direction direction);
1223 u64 (*map_page)(struct ib_device *dev,
1224 struct page *page, unsigned long offset,
1226 enum dma_data_direction direction);
1227 void (*unmap_page)(struct ib_device *dev,
1228 u64 addr, size_t size,
1229 enum dma_data_direction direction);
1230 int (*map_sg)(struct ib_device *dev,
1231 struct scatterlist *sg, int nents,
1232 enum dma_data_direction direction);
1233 void (*unmap_sg)(struct ib_device *dev,
1234 struct scatterlist *sg, int nents,
1235 enum dma_data_direction direction);
1236 u64 (*dma_address)(struct ib_device *dev,
1237 struct scatterlist *sg);
1238 unsigned int (*dma_len)(struct ib_device *dev,
1239 struct scatterlist *sg);
1240 void (*sync_single_for_cpu)(struct ib_device *dev,
1243 enum dma_data_direction dir);
1244 void (*sync_single_for_device)(struct ib_device *dev,
1247 enum dma_data_direction dir);
1248 void *(*alloc_coherent)(struct ib_device *dev,
1252 void (*free_coherent)(struct ib_device *dev,
1253 size_t size, void *cpu_addr,
1260 struct device *dma_device;
1262 char name[IB_DEVICE_NAME_MAX];
1264 struct list_head event_handler_list;
1265 spinlock_t event_handler_lock;
1267 spinlock_t client_data_lock;
1268 struct list_head core_list;
1269 struct list_head client_data_list;
1271 struct ib_cache cache;
1275 int num_comp_vectors;
1277 struct iw_cm_verbs *iwcm;
1279 int (*get_protocol_stats)(struct ib_device *device,
1280 union rdma_protocol_stats *stats);
1281 int (*query_device)(struct ib_device *device,
1282 struct ib_device_attr *device_attr);
1283 int (*query_port)(struct ib_device *device,
1285 struct ib_port_attr *port_attr);
1286 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1288 int (*query_gid)(struct ib_device *device,
1289 u8 port_num, int index,
1291 int (*query_pkey)(struct ib_device *device,
1292 u8 port_num, u16 index, u16 *pkey);
1293 int (*modify_device)(struct ib_device *device,
1294 int device_modify_mask,
1295 struct ib_device_modify *device_modify);
1296 int (*modify_port)(struct ib_device *device,
1297 u8 port_num, int port_modify_mask,
1298 struct ib_port_modify *port_modify);
1299 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1300 struct ib_udata *udata);
1301 int (*dealloc_ucontext)(struct ib_ucontext *context);
1302 int (*mmap)(struct ib_ucontext *context,
1303 struct vm_area_struct *vma);
1304 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1305 struct ib_ucontext *context,
1306 struct ib_udata *udata);
1307 int (*dealloc_pd)(struct ib_pd *pd);
1308 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1309 struct ib_ah_attr *ah_attr);
1310 int (*modify_ah)(struct ib_ah *ah,
1311 struct ib_ah_attr *ah_attr);
1312 int (*query_ah)(struct ib_ah *ah,
1313 struct ib_ah_attr *ah_attr);
1314 int (*destroy_ah)(struct ib_ah *ah);
1315 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1316 struct ib_srq_init_attr *srq_init_attr,
1317 struct ib_udata *udata);
1318 int (*modify_srq)(struct ib_srq *srq,
1319 struct ib_srq_attr *srq_attr,
1320 enum ib_srq_attr_mask srq_attr_mask,
1321 struct ib_udata *udata);
1322 int (*query_srq)(struct ib_srq *srq,
1323 struct ib_srq_attr *srq_attr);
1324 int (*destroy_srq)(struct ib_srq *srq);
1325 int (*post_srq_recv)(struct ib_srq *srq,
1326 struct ib_recv_wr *recv_wr,
1327 struct ib_recv_wr **bad_recv_wr);
1328 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1329 struct ib_qp_init_attr *qp_init_attr,
1330 struct ib_udata *udata);
1331 int (*modify_qp)(struct ib_qp *qp,
1332 struct ib_qp_attr *qp_attr,
1334 struct ib_udata *udata);
1335 int (*query_qp)(struct ib_qp *qp,
1336 struct ib_qp_attr *qp_attr,
1338 struct ib_qp_init_attr *qp_init_attr);
1339 int (*destroy_qp)(struct ib_qp *qp);
1340 int (*post_send)(struct ib_qp *qp,
1341 struct ib_send_wr *send_wr,
1342 struct ib_send_wr **bad_send_wr);
1343 int (*post_recv)(struct ib_qp *qp,
1344 struct ib_recv_wr *recv_wr,
1345 struct ib_recv_wr **bad_recv_wr);
1346 struct ib_cq * (*create_cq)(struct ib_device *device, int cqe,
1348 struct ib_ucontext *context,
1349 struct ib_udata *udata);
1350 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1352 int (*destroy_cq)(struct ib_cq *cq);
1353 int (*resize_cq)(struct ib_cq *cq, int cqe,
1354 struct ib_udata *udata);
1355 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1357 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1358 int (*req_notify_cq)(struct ib_cq *cq,
1359 enum ib_cq_notify_flags flags);
1360 int (*req_ncomp_notif)(struct ib_cq *cq,
1362 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1363 int mr_access_flags);
1364 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
1365 struct ib_phys_buf *phys_buf_array,
1367 int mr_access_flags,
1369 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1370 u64 start, u64 length,
1372 int mr_access_flags,
1373 struct ib_udata *udata);
1374 int (*query_mr)(struct ib_mr *mr,
1375 struct ib_mr_attr *mr_attr);
1376 int (*dereg_mr)(struct ib_mr *mr);
1377 struct ib_mr * (*alloc_fast_reg_mr)(struct ib_pd *pd,
1378 int max_page_list_len);
1379 struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1381 void (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1382 int (*rereg_phys_mr)(struct ib_mr *mr,
1385 struct ib_phys_buf *phys_buf_array,
1387 int mr_access_flags,
1389 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1390 enum ib_mw_type type);
1391 int (*bind_mw)(struct ib_qp *qp,
1393 struct ib_mw_bind *mw_bind);
1394 int (*dealloc_mw)(struct ib_mw *mw);
1395 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1396 int mr_access_flags,
1397 struct ib_fmr_attr *fmr_attr);
1398 int (*map_phys_fmr)(struct ib_fmr *fmr,
1399 u64 *page_list, int list_len,
1401 int (*unmap_fmr)(struct list_head *fmr_list);
1402 int (*dealloc_fmr)(struct ib_fmr *fmr);
1403 int (*attach_mcast)(struct ib_qp *qp,
1406 int (*detach_mcast)(struct ib_qp *qp,
1409 int (*process_mad)(struct ib_device *device,
1410 int process_mad_flags,
1412 struct ib_wc *in_wc,
1413 struct ib_grh *in_grh,
1414 struct ib_mad *in_mad,
1415 struct ib_mad *out_mad);
1416 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1417 struct ib_ucontext *ucontext,
1418 struct ib_udata *udata);
1419 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1420 struct ib_flow * (*create_flow)(struct ib_qp *qp,
1424 int (*destroy_flow)(struct ib_flow *flow_id);
1426 struct ib_dma_mapping_ops *dma_ops;
1428 struct module *owner;
1430 struct kobject *ports_parent;
1431 struct list_head port_list;
1434 IB_DEV_UNINITIALIZED,
1440 u64 uverbs_cmd_mask;
1451 void (*add) (struct ib_device *);
1452 void (*remove)(struct ib_device *);
1454 struct list_head list;
1457 struct ib_device *ib_alloc_device(size_t size);
1458 void ib_dealloc_device(struct ib_device *device);
1460 int ib_register_device(struct ib_device *device,
1461 int (*port_callback)(struct ib_device *,
1462 u8, struct kobject *));
1463 void ib_unregister_device(struct ib_device *device);
1465 int ib_register_client (struct ib_client *client);
1466 void ib_unregister_client(struct ib_client *client);
1468 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1469 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1472 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1474 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1477 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1479 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1483 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1484 * contains all required attributes and no attributes not allowed for
1485 * the given QP state transition.
1486 * @cur_state: Current QP state
1487 * @next_state: Next QP state
1489 * @mask: Mask of supplied QP attributes
1491 * This function is a helper function that a low-level driver's
1492 * modify_qp method can use to validate the consumer's input. It
1493 * checks that cur_state and next_state are valid QP states, that a
1494 * transition from cur_state to next_state is allowed by the IB spec,
1495 * and that the attribute mask supplied is allowed for the transition.
1497 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1498 enum ib_qp_type type, enum ib_qp_attr_mask mask);
1500 int ib_register_event_handler (struct ib_event_handler *event_handler);
1501 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1502 void ib_dispatch_event(struct ib_event *event);
1504 int ib_query_device(struct ib_device *device,
1505 struct ib_device_attr *device_attr);
1507 int ib_query_port(struct ib_device *device,
1508 u8 port_num, struct ib_port_attr *port_attr);
1510 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1513 int ib_query_gid(struct ib_device *device,
1514 u8 port_num, int index, union ib_gid *gid);
1516 int ib_query_pkey(struct ib_device *device,
1517 u8 port_num, u16 index, u16 *pkey);
1519 int ib_modify_device(struct ib_device *device,
1520 int device_modify_mask,
1521 struct ib_device_modify *device_modify);
1523 int ib_modify_port(struct ib_device *device,
1524 u8 port_num, int port_modify_mask,
1525 struct ib_port_modify *port_modify);
1527 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1528 u8 *port_num, u16 *index);
1530 int ib_find_pkey(struct ib_device *device,
1531 u8 port_num, u16 pkey, u16 *index);
1534 * ib_alloc_pd - Allocates an unused protection domain.
1535 * @device: The device on which to allocate the protection domain.
1537 * A protection domain object provides an association between QPs, shared
1538 * receive queues, address handles, memory regions, and memory windows.
1540 struct ib_pd *ib_alloc_pd(struct ib_device *device);
1543 * ib_dealloc_pd - Deallocates a protection domain.
1544 * @pd: The protection domain to deallocate.
1546 int ib_dealloc_pd(struct ib_pd *pd);
1549 * ib_create_ah - Creates an address handle for the given address vector.
1550 * @pd: The protection domain associated with the address handle.
1551 * @ah_attr: The attributes of the address vector.
1553 * The address handle is used to reference a local or global destination
1554 * in all UD QP post sends.
1556 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1559 * ib_init_ah_from_wc - Initializes address handle attributes from a
1561 * @device: Device on which the received message arrived.
1562 * @port_num: Port on which the received message arrived.
1563 * @wc: Work completion associated with the received message.
1564 * @grh: References the received global route header. This parameter is
1565 * ignored unless the work completion indicates that the GRH is valid.
1566 * @ah_attr: Returned attributes that can be used when creating an address
1567 * handle for replying to the message.
1569 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1570 struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1573 * ib_create_ah_from_wc - Creates an address handle associated with the
1574 * sender of the specified work completion.
1575 * @pd: The protection domain associated with the address handle.
1576 * @wc: Work completion information associated with a received message.
1577 * @grh: References the received global route header. This parameter is
1578 * ignored unless the work completion indicates that the GRH is valid.
1579 * @port_num: The outbound port number to associate with the address.
1581 * The address handle is used to reference a local or global destination
1582 * in all UD QP post sends.
1584 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1585 struct ib_grh *grh, u8 port_num);
1588 * ib_modify_ah - Modifies the address vector associated with an address
1590 * @ah: The address handle to modify.
1591 * @ah_attr: The new address vector attributes to associate with the
1594 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1597 * ib_query_ah - Queries the address vector associated with an address
1599 * @ah: The address handle to query.
1600 * @ah_attr: The address vector attributes associated with the address
1603 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1606 * ib_destroy_ah - Destroys an address handle.
1607 * @ah: The address handle to destroy.
1609 int ib_destroy_ah(struct ib_ah *ah);
1612 * ib_create_srq - Creates a SRQ associated with the specified protection
1614 * @pd: The protection domain associated with the SRQ.
1615 * @srq_init_attr: A list of initial attributes required to create the
1616 * SRQ. If SRQ creation succeeds, then the attributes are updated to
1617 * the actual capabilities of the created SRQ.
1619 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1620 * requested size of the SRQ, and set to the actual values allocated
1621 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
1622 * will always be at least as large as the requested values.
1624 struct ib_srq *ib_create_srq(struct ib_pd *pd,
1625 struct ib_srq_init_attr *srq_init_attr);
1628 * ib_modify_srq - Modifies the attributes for the specified SRQ.
1629 * @srq: The SRQ to modify.
1630 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
1631 * the current values of selected SRQ attributes are returned.
1632 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1633 * are being modified.
1635 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1636 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1637 * the number of receives queued drops below the limit.
1639 int ib_modify_srq(struct ib_srq *srq,
1640 struct ib_srq_attr *srq_attr,
1641 enum ib_srq_attr_mask srq_attr_mask);
1644 * ib_query_srq - Returns the attribute list and current values for the
1646 * @srq: The SRQ to query.
1647 * @srq_attr: The attributes of the specified SRQ.
1649 int ib_query_srq(struct ib_srq *srq,
1650 struct ib_srq_attr *srq_attr);
1653 * ib_destroy_srq - Destroys the specified SRQ.
1654 * @srq: The SRQ to destroy.
1656 int ib_destroy_srq(struct ib_srq *srq);
1659 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1660 * @srq: The SRQ to post the work request on.
1661 * @recv_wr: A list of work requests to post on the receive queue.
1662 * @bad_recv_wr: On an immediate failure, this parameter will reference
1663 * the work request that failed to be posted on the QP.
1665 static inline int ib_post_srq_recv(struct ib_srq *srq,
1666 struct ib_recv_wr *recv_wr,
1667 struct ib_recv_wr **bad_recv_wr)
1669 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1673 * ib_create_qp - Creates a QP associated with the specified protection
1675 * @pd: The protection domain associated with the QP.
1676 * @qp_init_attr: A list of initial attributes required to create the
1677 * QP. If QP creation succeeds, then the attributes are updated to
1678 * the actual capabilities of the created QP.
1680 struct ib_qp *ib_create_qp(struct ib_pd *pd,
1681 struct ib_qp_init_attr *qp_init_attr);
1684 * ib_modify_qp - Modifies the attributes for the specified QP and then
1685 * transitions the QP to the given state.
1686 * @qp: The QP to modify.
1687 * @qp_attr: On input, specifies the QP attributes to modify. On output,
1688 * the current values of selected QP attributes are returned.
1689 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1690 * are being modified.
1692 int ib_modify_qp(struct ib_qp *qp,
1693 struct ib_qp_attr *qp_attr,
1697 * ib_query_qp - Returns the attribute list and current values for the
1699 * @qp: The QP to query.
1700 * @qp_attr: The attributes of the specified QP.
1701 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1702 * @qp_init_attr: Additional attributes of the selected QP.
1704 * The qp_attr_mask may be used to limit the query to gathering only the
1705 * selected attributes.
1707 int ib_query_qp(struct ib_qp *qp,
1708 struct ib_qp_attr *qp_attr,
1710 struct ib_qp_init_attr *qp_init_attr);
1713 * ib_destroy_qp - Destroys the specified QP.
1714 * @qp: The QP to destroy.
1716 int ib_destroy_qp(struct ib_qp *qp);
1719 * ib_open_qp - Obtain a reference to an existing sharable QP.
1720 * @xrcd - XRC domain
1721 * @qp_open_attr: Attributes identifying the QP to open.
1723 * Returns a reference to a sharable QP.
1725 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
1726 struct ib_qp_open_attr *qp_open_attr);
1729 * ib_close_qp - Release an external reference to a QP.
1730 * @qp: The QP handle to release
1732 * The opened QP handle is released by the caller. The underlying
1733 * shared QP is not destroyed until all internal references are released.
1735 int ib_close_qp(struct ib_qp *qp);
1738 * ib_post_send - Posts a list of work requests to the send queue of
1740 * @qp: The QP to post the work request on.
1741 * @send_wr: A list of work requests to post on the send queue.
1742 * @bad_send_wr: On an immediate failure, this parameter will reference
1743 * the work request that failed to be posted on the QP.
1745 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
1746 * error is returned, the QP state shall not be affected,
1747 * ib_post_send() will return an immediate error after queueing any
1748 * earlier work requests in the list.
1750 static inline int ib_post_send(struct ib_qp *qp,
1751 struct ib_send_wr *send_wr,
1752 struct ib_send_wr **bad_send_wr)
1754 return qp->device->post_send(qp, send_wr, bad_send_wr);
1758 * ib_post_recv - Posts a list of work requests to the receive queue of
1760 * @qp: The QP to post the work request on.
1761 * @recv_wr: A list of work requests to post on the receive queue.
1762 * @bad_recv_wr: On an immediate failure, this parameter will reference
1763 * the work request that failed to be posted on the QP.
1765 static inline int ib_post_recv(struct ib_qp *qp,
1766 struct ib_recv_wr *recv_wr,
1767 struct ib_recv_wr **bad_recv_wr)
1769 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1773 * ib_create_cq - Creates a CQ on the specified device.
1774 * @device: The device on which to create the CQ.
1775 * @comp_handler: A user-specified callback that is invoked when a
1776 * completion event occurs on the CQ.
1777 * @event_handler: A user-specified callback that is invoked when an
1778 * asynchronous event not associated with a completion occurs on the CQ.
1779 * @cq_context: Context associated with the CQ returned to the user via
1780 * the associated completion and event handlers.
1781 * @cqe: The minimum size of the CQ.
1782 * @comp_vector - Completion vector used to signal completion events.
1783 * Must be >= 0 and < context->num_comp_vectors.
1785 * Users can examine the cq structure to determine the actual CQ size.
1787 struct ib_cq *ib_create_cq(struct ib_device *device,
1788 ib_comp_handler comp_handler,
1789 void (*event_handler)(struct ib_event *, void *),
1790 void *cq_context, int cqe, int comp_vector);
1793 * ib_resize_cq - Modifies the capacity of the CQ.
1794 * @cq: The CQ to resize.
1795 * @cqe: The minimum size of the CQ.
1797 * Users can examine the cq structure to determine the actual CQ size.
1799 int ib_resize_cq(struct ib_cq *cq, int cqe);
1802 * ib_modify_cq - Modifies moderation params of the CQ
1803 * @cq: The CQ to modify.
1804 * @cq_count: number of CQEs that will trigger an event
1805 * @cq_period: max period of time in usec before triggering an event
1808 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1811 * ib_destroy_cq - Destroys the specified CQ.
1812 * @cq: The CQ to destroy.
1814 int ib_destroy_cq(struct ib_cq *cq);
1817 * ib_poll_cq - poll a CQ for completion(s)
1818 * @cq:the CQ being polled
1819 * @num_entries:maximum number of completions to return
1820 * @wc:array of at least @num_entries &struct ib_wc where completions
1823 * Poll a CQ for (possibly multiple) completions. If the return value
1824 * is < 0, an error occurred. If the return value is >= 0, it is the
1825 * number of completions returned. If the return value is
1826 * non-negative and < num_entries, then the CQ was emptied.
1828 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1831 return cq->device->poll_cq(cq, num_entries, wc);
1835 * ib_peek_cq - Returns the number of unreaped completions currently
1836 * on the specified CQ.
1837 * @cq: The CQ to peek.
1838 * @wc_cnt: A minimum number of unreaped completions to check for.
1840 * If the number of unreaped completions is greater than or equal to wc_cnt,
1841 * this function returns wc_cnt, otherwise, it returns the actual number of
1842 * unreaped completions.
1844 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1847 * ib_req_notify_cq - Request completion notification on a CQ.
1848 * @cq: The CQ to generate an event for.
1850 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1851 * to request an event on the next solicited event or next work
1852 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1853 * may also be |ed in to request a hint about missed events, as
1857 * < 0 means an error occurred while requesting notification
1858 * == 0 means notification was requested successfully, and if
1859 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1860 * were missed and it is safe to wait for another event. In
1861 * this case is it guaranteed that any work completions added
1862 * to the CQ since the last CQ poll will trigger a completion
1863 * notification event.
1864 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1865 * in. It means that the consumer must poll the CQ again to
1866 * make sure it is empty to avoid missing an event because of a
1867 * race between requesting notification and an entry being
1868 * added to the CQ. This return value means it is possible
1869 * (but not guaranteed) that a work completion has been added
1870 * to the CQ since the last poll without triggering a
1871 * completion notification event.
1873 static inline int ib_req_notify_cq(struct ib_cq *cq,
1874 enum ib_cq_notify_flags flags)
1876 return cq->device->req_notify_cq(cq, flags);
1880 * ib_req_ncomp_notif - Request completion notification when there are
1881 * at least the specified number of unreaped completions on the CQ.
1882 * @cq: The CQ to generate an event for.
1883 * @wc_cnt: The number of unreaped completions that should be on the
1884 * CQ before an event is generated.
1886 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1888 return cq->device->req_ncomp_notif ?
1889 cq->device->req_ncomp_notif(cq, wc_cnt) :
1894 * ib_get_dma_mr - Returns a memory region for system memory that is
1896 * @pd: The protection domain associated with the memory region.
1897 * @mr_access_flags: Specifies the memory access rights.
1899 * Note that the ib_dma_*() functions defined below must be used
1900 * to create/destroy addresses used with the Lkey or Rkey returned
1901 * by ib_get_dma_mr().
1903 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1906 * ib_dma_mapping_error - check a DMA addr for error
1907 * @dev: The device for which the dma_addr was created
1908 * @dma_addr: The DMA address to check
1910 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1913 return dev->dma_ops->mapping_error(dev, dma_addr);
1914 return dma_mapping_error(dev->dma_device, dma_addr);
1918 * ib_dma_map_single - Map a kernel virtual address to DMA address
1919 * @dev: The device for which the dma_addr is to be created
1920 * @cpu_addr: The kernel virtual address
1921 * @size: The size of the region in bytes
1922 * @direction: The direction of the DMA
1924 static inline u64 ib_dma_map_single(struct ib_device *dev,
1925 void *cpu_addr, size_t size,
1926 enum dma_data_direction direction)
1929 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1930 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1934 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1935 * @dev: The device for which the DMA address was created
1936 * @addr: The DMA address
1937 * @size: The size of the region in bytes
1938 * @direction: The direction of the DMA
1940 static inline void ib_dma_unmap_single(struct ib_device *dev,
1941 u64 addr, size_t size,
1942 enum dma_data_direction direction)
1945 dev->dma_ops->unmap_single(dev, addr, size, direction);
1947 dma_unmap_single(dev->dma_device, addr, size, direction);
1950 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1951 void *cpu_addr, size_t size,
1952 enum dma_data_direction direction,
1953 struct dma_attrs *attrs)
1955 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1959 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1960 u64 addr, size_t size,
1961 enum dma_data_direction direction,
1962 struct dma_attrs *attrs)
1964 return dma_unmap_single_attrs(dev->dma_device, addr, size,
1969 * ib_dma_map_page - Map a physical page to DMA address
1970 * @dev: The device for which the dma_addr is to be created
1971 * @page: The page to be mapped
1972 * @offset: The offset within the page
1973 * @size: The size of the region in bytes
1974 * @direction: The direction of the DMA
1976 static inline u64 ib_dma_map_page(struct ib_device *dev,
1978 unsigned long offset,
1980 enum dma_data_direction direction)
1983 return dev->dma_ops->map_page(dev, page, offset, size, direction);
1984 return dma_map_page(dev->dma_device, page, offset, size, direction);
1988 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1989 * @dev: The device for which the DMA address was created
1990 * @addr: The DMA address
1991 * @size: The size of the region in bytes
1992 * @direction: The direction of the DMA
1994 static inline void ib_dma_unmap_page(struct ib_device *dev,
1995 u64 addr, size_t size,
1996 enum dma_data_direction direction)
1999 dev->dma_ops->unmap_page(dev, addr, size, direction);
2001 dma_unmap_page(dev->dma_device, addr, size, direction);
2005 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2006 * @dev: The device for which the DMA addresses are to be created
2007 * @sg: The array of scatter/gather entries
2008 * @nents: The number of scatter/gather entries
2009 * @direction: The direction of the DMA
2011 static inline int ib_dma_map_sg(struct ib_device *dev,
2012 struct scatterlist *sg, int nents,
2013 enum dma_data_direction direction)
2016 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2017 return dma_map_sg(dev->dma_device, sg, nents, direction);
2021 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2022 * @dev: The device for which the DMA addresses were created
2023 * @sg: The array of scatter/gather entries
2024 * @nents: The number of scatter/gather entries
2025 * @direction: The direction of the DMA
2027 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2028 struct scatterlist *sg, int nents,
2029 enum dma_data_direction direction)
2032 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2034 dma_unmap_sg(dev->dma_device, sg, nents, direction);
2037 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2038 struct scatterlist *sg, int nents,
2039 enum dma_data_direction direction,
2040 struct dma_attrs *attrs)
2042 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2045 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2046 struct scatterlist *sg, int nents,
2047 enum dma_data_direction direction,
2048 struct dma_attrs *attrs)
2050 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2053 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2054 * @dev: The device for which the DMA addresses were created
2055 * @sg: The scatter/gather entry
2057 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2058 struct scatterlist *sg)
2061 return dev->dma_ops->dma_address(dev, sg);
2062 return sg_dma_address(sg);
2066 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2067 * @dev: The device for which the DMA addresses were created
2068 * @sg: The scatter/gather entry
2070 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2071 struct scatterlist *sg)
2074 return dev->dma_ops->dma_len(dev, sg);
2075 return sg_dma_len(sg);
2079 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2080 * @dev: The device for which the DMA address was created
2081 * @addr: The DMA address
2082 * @size: The size of the region in bytes
2083 * @dir: The direction of the DMA
2085 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2088 enum dma_data_direction dir)
2091 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2093 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2097 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2098 * @dev: The device for which the DMA address was created
2099 * @addr: The DMA address
2100 * @size: The size of the region in bytes
2101 * @dir: The direction of the DMA
2103 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2106 enum dma_data_direction dir)
2109 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2111 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2115 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2116 * @dev: The device for which the DMA address is requested
2117 * @size: The size of the region to allocate in bytes
2118 * @dma_handle: A pointer for returning the DMA address of the region
2119 * @flag: memory allocator flags
2121 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2127 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2132 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2133 *dma_handle = handle;
2139 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2140 * @dev: The device for which the DMA addresses were allocated
2141 * @size: The size of the region
2142 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2143 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2145 static inline void ib_dma_free_coherent(struct ib_device *dev,
2146 size_t size, void *cpu_addr,
2150 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2152 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2156 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
2158 * @pd: The protection domain associated assigned to the registered region.
2159 * @phys_buf_array: Specifies a list of physical buffers to use in the
2161 * @num_phys_buf: Specifies the size of the phys_buf_array.
2162 * @mr_access_flags: Specifies the memory access rights.
2163 * @iova_start: The offset of the region's starting I/O virtual address.
2165 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
2166 struct ib_phys_buf *phys_buf_array,
2168 int mr_access_flags,
2172 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
2173 * Conceptually, this call performs the functions deregister memory region
2174 * followed by register physical memory region. Where possible,
2175 * resources are reused instead of deallocated and reallocated.
2176 * @mr: The memory region to modify.
2177 * @mr_rereg_mask: A bit-mask used to indicate which of the following
2178 * properties of the memory region are being modified.
2179 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
2180 * the new protection domain to associated with the memory region,
2181 * otherwise, this parameter is ignored.
2182 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2183 * field specifies a list of physical buffers to use in the new
2184 * translation, otherwise, this parameter is ignored.
2185 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2186 * field specifies the size of the phys_buf_array, otherwise, this
2187 * parameter is ignored.
2188 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
2189 * field specifies the new memory access rights, otherwise, this
2190 * parameter is ignored.
2191 * @iova_start: The offset of the region's starting I/O virtual address.
2193 int ib_rereg_phys_mr(struct ib_mr *mr,
2196 struct ib_phys_buf *phys_buf_array,
2198 int mr_access_flags,
2202 * ib_query_mr - Retrieves information about a specific memory region.
2203 * @mr: The memory region to retrieve information about.
2204 * @mr_attr: The attributes of the specified memory region.
2206 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2209 * ib_dereg_mr - Deregisters a memory region and removes it from the
2210 * HCA translation table.
2211 * @mr: The memory region to deregister.
2213 * This function can fail, if the memory region has memory windows bound to it.
2215 int ib_dereg_mr(struct ib_mr *mr);
2218 * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2219 * IB_WR_FAST_REG_MR send work request.
2220 * @pd: The protection domain associated with the region.
2221 * @max_page_list_len: requested max physical buffer list length to be
2222 * used with fast register work requests for this MR.
2224 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2227 * ib_alloc_fast_reg_page_list - Allocates a page list array
2228 * @device - ib device pointer.
2229 * @page_list_len - size of the page list array to be allocated.
2231 * This allocates and returns a struct ib_fast_reg_page_list * and a
2232 * page_list array that is at least page_list_len in size. The actual
2233 * size is returned in max_page_list_len. The caller is responsible
2234 * for initializing the contents of the page_list array before posting
2235 * a send work request with the IB_WC_FAST_REG_MR opcode.
2237 * The page_list array entries must be translated using one of the
2238 * ib_dma_*() functions just like the addresses passed to
2239 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
2240 * ib_fast_reg_page_list must not be modified by the caller until the
2241 * IB_WC_FAST_REG_MR work request completes.
2243 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2244 struct ib_device *device, int page_list_len);
2247 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2249 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2251 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2254 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2256 * @mr - struct ib_mr pointer to be updated.
2257 * @newkey - new key to be used.
2259 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2261 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2262 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2266 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2267 * for calculating a new rkey for type 2 memory windows.
2268 * @rkey - the rkey to increment.
2270 static inline u32 ib_inc_rkey(u32 rkey)
2272 const u32 mask = 0x000000ff;
2273 return ((rkey + 1) & mask) | (rkey & ~mask);
2277 * ib_alloc_mw - Allocates a memory window.
2278 * @pd: The protection domain associated with the memory window.
2279 * @type: The type of the memory window (1 or 2).
2281 struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
2284 * ib_bind_mw - Posts a work request to the send queue of the specified
2285 * QP, which binds the memory window to the given address range and
2286 * remote access attributes.
2287 * @qp: QP to post the bind work request on.
2288 * @mw: The memory window to bind.
2289 * @mw_bind: Specifies information about the memory window, including
2290 * its address range, remote access rights, and associated memory region.
2292 * If there is no immediate error, the function will update the rkey member
2293 * of the mw parameter to its new value. The bind operation can still fail
2296 static inline int ib_bind_mw(struct ib_qp *qp,
2298 struct ib_mw_bind *mw_bind)
2300 /* XXX reference counting in corresponding MR? */
2301 return mw->device->bind_mw ?
2302 mw->device->bind_mw(qp, mw, mw_bind) :
2307 * ib_dealloc_mw - Deallocates a memory window.
2308 * @mw: The memory window to deallocate.
2310 int ib_dealloc_mw(struct ib_mw *mw);
2313 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2314 * @pd: The protection domain associated with the unmapped region.
2315 * @mr_access_flags: Specifies the memory access rights.
2316 * @fmr_attr: Attributes of the unmapped region.
2318 * A fast memory region must be mapped before it can be used as part of
2321 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2322 int mr_access_flags,
2323 struct ib_fmr_attr *fmr_attr);
2326 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2327 * @fmr: The fast memory region to associate with the pages.
2328 * @page_list: An array of physical pages to map to the fast memory region.
2329 * @list_len: The number of pages in page_list.
2330 * @iova: The I/O virtual address to use with the mapped region.
2332 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2333 u64 *page_list, int list_len,
2336 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2340 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2341 * @fmr_list: A linked list of fast memory regions to unmap.
2343 int ib_unmap_fmr(struct list_head *fmr_list);
2346 * ib_dealloc_fmr - Deallocates a fast memory region.
2347 * @fmr: The fast memory region to deallocate.
2349 int ib_dealloc_fmr(struct ib_fmr *fmr);
2352 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2353 * @qp: QP to attach to the multicast group. The QP must be type
2355 * @gid: Multicast group GID.
2356 * @lid: Multicast group LID in host byte order.
2358 * In order to send and receive multicast packets, subnet
2359 * administration must have created the multicast group and configured
2360 * the fabric appropriately. The port associated with the specified
2361 * QP must also be a member of the multicast group.
2363 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2366 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2367 * @qp: QP to detach from the multicast group.
2368 * @gid: Multicast group GID.
2369 * @lid: Multicast group LID in host byte order.
2371 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2374 * ib_alloc_xrcd - Allocates an XRC domain.
2375 * @device: The device on which to allocate the XRC domain.
2377 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2380 * ib_dealloc_xrcd - Deallocates an XRC domain.
2381 * @xrcd: The XRC domain to deallocate.
2383 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2385 struct ib_flow *ib_create_flow(struct ib_qp *qp,
2386 struct ib_flow_attr *flow_attr, int domain);
2387 int ib_destroy_flow(struct ib_flow *flow_id);
2389 #endif /* IB_VERBS_H */