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. */
73 enum rdma_transport_type {
78 enum rdma_transport_type
79 rdma_node_get_transport(enum rdma_node_type node_type) __attribute_const__;
81 enum rdma_link_layer {
82 IB_LINK_LAYER_UNSPECIFIED,
83 IB_LINK_LAYER_INFINIBAND,
84 IB_LINK_LAYER_ETHERNET,
87 enum ib_device_cap_flags {
88 IB_DEVICE_RESIZE_MAX_WR = 1,
89 IB_DEVICE_BAD_PKEY_CNTR = (1<<1),
90 IB_DEVICE_BAD_QKEY_CNTR = (1<<2),
91 IB_DEVICE_RAW_MULTI = (1<<3),
92 IB_DEVICE_AUTO_PATH_MIG = (1<<4),
93 IB_DEVICE_CHANGE_PHY_PORT = (1<<5),
94 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6),
95 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7),
96 IB_DEVICE_SHUTDOWN_PORT = (1<<8),
97 IB_DEVICE_INIT_TYPE = (1<<9),
98 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10),
99 IB_DEVICE_SYS_IMAGE_GUID = (1<<11),
100 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12),
101 IB_DEVICE_SRQ_RESIZE = (1<<13),
102 IB_DEVICE_N_NOTIFY_CQ = (1<<14),
103 IB_DEVICE_LOCAL_DMA_LKEY = (1<<15),
104 IB_DEVICE_RESERVED = (1<<16), /* old SEND_W_INV */
105 IB_DEVICE_MEM_WINDOW = (1<<17),
107 * Devices should set IB_DEVICE_UD_IP_SUM if they support
108 * insertion of UDP and TCP checksum on outgoing UD IPoIB
109 * messages and can verify the validity of checksum for
110 * incoming messages. Setting this flag implies that the
111 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
113 IB_DEVICE_UD_IP_CSUM = (1<<18),
114 IB_DEVICE_UD_TSO = (1<<19),
115 IB_DEVICE_XRC = (1<<20),
116 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21),
117 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
118 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1<<23),
119 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1<<24),
120 IB_DEVICE_MANAGED_FLOW_STEERING = (1<<29)
129 struct ib_device_attr {
131 __be64 sys_image_guid;
139 int device_cap_flags;
149 int max_qp_init_rd_atom;
150 int max_ee_init_rd_atom;
151 enum ib_atomic_cap atomic_cap;
152 enum ib_atomic_cap masked_atomic_cap;
159 int max_mcast_qp_attach;
160 int max_total_mcast_qp_attach;
167 unsigned int max_fast_reg_page_list_len;
169 u8 local_ca_ack_delay;
180 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
183 case IB_MTU_256: return 256;
184 case IB_MTU_512: return 512;
185 case IB_MTU_1024: return 1024;
186 case IB_MTU_2048: return 2048;
187 case IB_MTU_4096: return 4096;
198 IB_PORT_ACTIVE_DEFER = 5
201 enum ib_port_cap_flags {
203 IB_PORT_NOTICE_SUP = 1 << 2,
204 IB_PORT_TRAP_SUP = 1 << 3,
205 IB_PORT_OPT_IPD_SUP = 1 << 4,
206 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
207 IB_PORT_SL_MAP_SUP = 1 << 6,
208 IB_PORT_MKEY_NVRAM = 1 << 7,
209 IB_PORT_PKEY_NVRAM = 1 << 8,
210 IB_PORT_LED_INFO_SUP = 1 << 9,
211 IB_PORT_SM_DISABLED = 1 << 10,
212 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
213 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
214 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
215 IB_PORT_CM_SUP = 1 << 16,
216 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
217 IB_PORT_REINIT_SUP = 1 << 18,
218 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
219 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
220 IB_PORT_DR_NOTICE_SUP = 1 << 21,
221 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
222 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
223 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
224 IB_PORT_CLIENT_REG_SUP = 1 << 25
234 static inline int ib_width_enum_to_int(enum ib_port_width width)
237 case IB_WIDTH_1X: return 1;
238 case IB_WIDTH_4X: return 4;
239 case IB_WIDTH_8X: return 8;
240 case IB_WIDTH_12X: return 12;
254 struct ib_protocol_stats {
258 struct iw_protocol_stats {
261 u64 ipInTooBigErrors;
264 u64 ipInUnknownProtos;
265 u64 ipInTruncatedPkts;
268 u64 ipOutForwDatagrams;
300 union rdma_protocol_stats {
301 struct ib_protocol_stats ib;
302 struct iw_protocol_stats iw;
305 struct ib_port_attr {
306 enum ib_port_state state;
308 enum ib_mtu active_mtu;
327 enum ib_device_modify_flags {
328 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
329 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
332 struct ib_device_modify {
337 enum ib_port_modify_flags {
338 IB_PORT_SHUTDOWN = 1,
339 IB_PORT_INIT_TYPE = (1<<2),
340 IB_PORT_RESET_QKEY_CNTR = (1<<3)
343 struct ib_port_modify {
344 u32 set_port_cap_mask;
345 u32 clr_port_cap_mask;
353 IB_EVENT_QP_ACCESS_ERR,
357 IB_EVENT_PATH_MIG_ERR,
358 IB_EVENT_DEVICE_FATAL,
359 IB_EVENT_PORT_ACTIVE,
362 IB_EVENT_PKEY_CHANGE,
365 IB_EVENT_SRQ_LIMIT_REACHED,
366 IB_EVENT_QP_LAST_WQE_REACHED,
367 IB_EVENT_CLIENT_REREGISTER,
372 struct ib_device *device;
379 enum ib_event_type event;
382 struct ib_event_handler {
383 struct ib_device *device;
384 void (*handler)(struct ib_event_handler *, struct ib_event *);
385 struct list_head list;
388 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
390 (_ptr)->device = _device; \
391 (_ptr)->handler = _handler; \
392 INIT_LIST_HEAD(&(_ptr)->list); \
395 struct ib_global_route {
404 __be32 version_tclass_flow;
413 IB_MULTICAST_QPN = 0xffffff
416 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
423 IB_RATE_PORT_CURRENT = 0,
424 IB_RATE_2_5_GBPS = 2,
432 IB_RATE_120_GBPS = 10,
433 IB_RATE_14_GBPS = 11,
434 IB_RATE_56_GBPS = 12,
435 IB_RATE_112_GBPS = 13,
436 IB_RATE_168_GBPS = 14,
437 IB_RATE_25_GBPS = 15,
438 IB_RATE_100_GBPS = 16,
439 IB_RATE_200_GBPS = 17,
440 IB_RATE_300_GBPS = 18
444 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
445 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
446 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
447 * @rate: rate to convert.
449 int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
452 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
453 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
454 * @rate: rate to convert.
456 int ib_rate_to_mbps(enum ib_rate rate) __attribute_const__;
459 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
461 * @mult: multiple to convert.
463 enum ib_rate mult_to_ib_rate(int mult) __attribute_const__;
466 struct ib_global_route grh;
479 IB_WC_LOC_EEC_OP_ERR,
484 IB_WC_LOC_ACCESS_ERR,
485 IB_WC_REM_INV_REQ_ERR,
486 IB_WC_REM_ACCESS_ERR,
489 IB_WC_RNR_RETRY_EXC_ERR,
490 IB_WC_LOC_RDD_VIOL_ERR,
491 IB_WC_REM_INV_RD_REQ_ERR,
494 IB_WC_INV_EEC_STATE_ERR,
496 IB_WC_RESP_TIMEOUT_ERR,
510 IB_WC_MASKED_COMP_SWAP,
511 IB_WC_MASKED_FETCH_ADD,
513 * Set value of IB_WC_RECV so consumers can test if a completion is a
514 * receive by testing (opcode & IB_WC_RECV).
517 IB_WC_RECV_RDMA_WITH_IMM
522 IB_WC_WITH_IMM = (1<<1),
523 IB_WC_WITH_INVALIDATE = (1<<2),
524 IB_WC_IP_CSUM_OK = (1<<3),
529 enum ib_wc_status status;
530 enum ib_wc_opcode opcode;
544 u8 port_num; /* valid only for DR SMPs on switches */
547 enum ib_cq_notify_flags {
548 IB_CQ_SOLICITED = 1 << 0,
549 IB_CQ_NEXT_COMP = 1 << 1,
550 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
551 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
559 enum ib_srq_attr_mask {
560 IB_SRQ_MAX_WR = 1 << 0,
561 IB_SRQ_LIMIT = 1 << 1,
570 struct ib_srq_init_attr {
571 void (*event_handler)(struct ib_event *, void *);
573 struct ib_srq_attr attr;
574 enum ib_srq_type srq_type;
578 struct ib_xrcd *xrcd;
599 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
600 * here (and in that order) since the MAD layer uses them as
601 * indices into a 2-entry table.
610 IB_QPT_RAW_ETHERTYPE,
611 IB_QPT_RAW_PACKET = 8,
615 /* Reserve a range for qp types internal to the low level driver.
616 * These qp types will not be visible at the IB core layer, so the
617 * IB_QPT_MAX usages should not be affected in the core layer
619 IB_QPT_RESERVED1 = 0x1000,
631 enum ib_qp_create_flags {
632 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
633 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
634 /* reserve bits 26-31 for low level drivers' internal use */
635 IB_QP_CREATE_RESERVED_START = 1 << 26,
636 IB_QP_CREATE_RESERVED_END = 1 << 31,
639 struct ib_qp_init_attr {
640 void (*event_handler)(struct ib_event *, void *);
642 struct ib_cq *send_cq;
643 struct ib_cq *recv_cq;
645 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
646 struct ib_qp_cap cap;
647 enum ib_sig_type sq_sig_type;
648 enum ib_qp_type qp_type;
649 enum ib_qp_create_flags create_flags;
650 u8 port_num; /* special QP types only */
653 struct ib_qp_open_attr {
654 void (*event_handler)(struct ib_event *, void *);
657 enum ib_qp_type qp_type;
660 enum ib_rnr_timeout {
661 IB_RNR_TIMER_655_36 = 0,
662 IB_RNR_TIMER_000_01 = 1,
663 IB_RNR_TIMER_000_02 = 2,
664 IB_RNR_TIMER_000_03 = 3,
665 IB_RNR_TIMER_000_04 = 4,
666 IB_RNR_TIMER_000_06 = 5,
667 IB_RNR_TIMER_000_08 = 6,
668 IB_RNR_TIMER_000_12 = 7,
669 IB_RNR_TIMER_000_16 = 8,
670 IB_RNR_TIMER_000_24 = 9,
671 IB_RNR_TIMER_000_32 = 10,
672 IB_RNR_TIMER_000_48 = 11,
673 IB_RNR_TIMER_000_64 = 12,
674 IB_RNR_TIMER_000_96 = 13,
675 IB_RNR_TIMER_001_28 = 14,
676 IB_RNR_TIMER_001_92 = 15,
677 IB_RNR_TIMER_002_56 = 16,
678 IB_RNR_TIMER_003_84 = 17,
679 IB_RNR_TIMER_005_12 = 18,
680 IB_RNR_TIMER_007_68 = 19,
681 IB_RNR_TIMER_010_24 = 20,
682 IB_RNR_TIMER_015_36 = 21,
683 IB_RNR_TIMER_020_48 = 22,
684 IB_RNR_TIMER_030_72 = 23,
685 IB_RNR_TIMER_040_96 = 24,
686 IB_RNR_TIMER_061_44 = 25,
687 IB_RNR_TIMER_081_92 = 26,
688 IB_RNR_TIMER_122_88 = 27,
689 IB_RNR_TIMER_163_84 = 28,
690 IB_RNR_TIMER_245_76 = 29,
691 IB_RNR_TIMER_327_68 = 30,
692 IB_RNR_TIMER_491_52 = 31
695 enum ib_qp_attr_mask {
697 IB_QP_CUR_STATE = (1<<1),
698 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
699 IB_QP_ACCESS_FLAGS = (1<<3),
700 IB_QP_PKEY_INDEX = (1<<4),
704 IB_QP_PATH_MTU = (1<<8),
705 IB_QP_TIMEOUT = (1<<9),
706 IB_QP_RETRY_CNT = (1<<10),
707 IB_QP_RNR_RETRY = (1<<11),
708 IB_QP_RQ_PSN = (1<<12),
709 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
710 IB_QP_ALT_PATH = (1<<14),
711 IB_QP_MIN_RNR_TIMER = (1<<15),
712 IB_QP_SQ_PSN = (1<<16),
713 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
714 IB_QP_PATH_MIG_STATE = (1<<18),
716 IB_QP_DEST_QPN = (1<<20)
741 enum ib_qp_state qp_state;
742 enum ib_qp_state cur_qp_state;
743 enum ib_mtu path_mtu;
744 enum ib_mig_state path_mig_state;
750 struct ib_qp_cap cap;
751 struct ib_ah_attr ah_attr;
752 struct ib_ah_attr alt_ah_attr;
755 u8 en_sqd_async_notify;
758 u8 max_dest_rd_atomic;
770 IB_WR_RDMA_WRITE_WITH_IMM,
774 IB_WR_ATOMIC_CMP_AND_SWP,
775 IB_WR_ATOMIC_FETCH_AND_ADD,
778 IB_WR_RDMA_READ_WITH_INV,
781 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
782 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
784 /* reserve values for low level drivers' internal use.
785 * These values will not be used at all in the ib core layer.
787 IB_WR_RESERVED1 = 0xf0,
801 IB_SEND_SIGNALED = (1<<1),
802 IB_SEND_SOLICITED = (1<<2),
803 IB_SEND_INLINE = (1<<3),
804 IB_SEND_IP_CSUM = (1<<4),
806 /* reserve bits 26-31 for low level drivers' internal use */
807 IB_SEND_RESERVED_START = (1 << 26),
808 IB_SEND_RESERVED_END = (1 << 31),
817 struct ib_fast_reg_page_list {
818 struct ib_device *device;
820 unsigned int max_page_list_len;
824 * struct ib_mw_bind_info - Parameters for a memory window bind operation.
825 * @mr: A memory region to bind the memory window to.
826 * @addr: The address where the memory window should begin.
827 * @length: The length of the memory window, in bytes.
828 * @mw_access_flags: Access flags from enum ib_access_flags for the window.
830 * This struct contains the shared parameters for type 1 and type 2
831 * memory window bind operations.
833 struct ib_mw_bind_info {
841 struct ib_send_wr *next;
843 struct ib_sge *sg_list;
845 enum ib_wr_opcode opcode;
860 u64 compare_add_mask;
871 u16 pkey_index; /* valid for GSI only */
872 u8 port_num; /* valid for DR SMPs on switch only */
876 struct ib_fast_reg_page_list *page_list;
877 unsigned int page_shift;
878 unsigned int page_list_len;
885 /* The new rkey for the memory window. */
887 struct ib_mw_bind_info bind_info;
890 u32 xrc_remote_srq_num; /* XRC TGT QPs only */
894 struct ib_recv_wr *next;
896 struct ib_sge *sg_list;
900 enum ib_access_flags {
901 IB_ACCESS_LOCAL_WRITE = 1,
902 IB_ACCESS_REMOTE_WRITE = (1<<1),
903 IB_ACCESS_REMOTE_READ = (1<<2),
904 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
905 IB_ACCESS_MW_BIND = (1<<4),
906 IB_ZERO_BASED = (1<<5)
916 u64 device_virt_addr;
923 enum ib_mr_rereg_flags {
924 IB_MR_REREG_TRANS = 1,
925 IB_MR_REREG_PD = (1<<1),
926 IB_MR_REREG_ACCESS = (1<<2)
930 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
931 * @wr_id: Work request id.
932 * @send_flags: Flags from ib_send_flags enum.
933 * @bind_info: More parameters of the bind operation.
938 struct ib_mw_bind_info bind_info;
948 struct ib_device *device;
949 struct list_head pd_list;
950 struct list_head mr_list;
951 struct list_head mw_list;
952 struct list_head cq_list;
953 struct list_head qp_list;
954 struct list_head srq_list;
955 struct list_head ah_list;
956 struct list_head xrcd_list;
961 u64 user_handle; /* handle given to us by userspace */
962 struct ib_ucontext *context; /* associated user context */
963 void *object; /* containing object */
964 struct list_head list; /* link to context's list */
965 int id; /* index into kernel idr */
967 struct rw_semaphore mutex; /* protects .live */
979 struct ib_device *device;
980 struct ib_uobject *uobject;
981 atomic_t usecnt; /* count all resources */
985 struct ib_device *device;
986 atomic_t usecnt; /* count all exposed resources */
989 struct mutex tgt_qp_mutex;
990 struct list_head tgt_qp_list;
994 struct ib_device *device;
996 struct ib_uobject *uobject;
999 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1002 struct ib_device *device;
1003 struct ib_uobject *uobject;
1004 ib_comp_handler comp_handler;
1005 void (*event_handler)(struct ib_event *, void *);
1008 atomic_t usecnt; /* count number of work queues */
1012 struct ib_device *device;
1014 struct ib_uobject *uobject;
1015 void (*event_handler)(struct ib_event *, void *);
1017 enum ib_srq_type srq_type;
1022 struct ib_xrcd *xrcd;
1030 struct ib_device *device;
1032 struct ib_cq *send_cq;
1033 struct ib_cq *recv_cq;
1035 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1036 struct list_head xrcd_list;
1037 /* count times opened, mcast attaches, flow attaches */
1039 struct list_head open_list;
1040 struct ib_qp *real_qp;
1041 struct ib_uobject *uobject;
1042 void (*event_handler)(struct ib_event *, void *);
1045 enum ib_qp_type qp_type;
1049 struct ib_device *device;
1051 struct ib_uobject *uobject;
1054 atomic_t usecnt; /* count number of MWs */
1058 struct ib_device *device;
1060 struct ib_uobject *uobject;
1062 enum ib_mw_type type;
1066 struct ib_device *device;
1068 struct list_head list;
1073 /* Supported steering options */
1074 enum ib_flow_attr_type {
1075 /* steering according to rule specifications */
1076 IB_FLOW_ATTR_NORMAL = 0x0,
1077 /* default unicast and multicast rule -
1078 * receive all Eth traffic which isn't steered to any QP
1080 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1081 /* default multicast rule -
1082 * receive all Eth multicast traffic which isn't steered to any QP
1084 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1085 /* sniffer rule - receive all port traffic */
1086 IB_FLOW_ATTR_SNIFFER = 0x3
1089 /* Supported steering header types */
1090 enum ib_flow_spec_type {
1092 IB_FLOW_SPEC_ETH = 0x20,
1094 IB_FLOW_SPEC_IPV4 = 0x30,
1096 IB_FLOW_SPEC_TCP = 0x40,
1097 IB_FLOW_SPEC_UDP = 0x41
1100 /* Flow steering rule priority is set according to it's domain.
1101 * Lower domain value means higher priority.
1103 enum ib_flow_domain {
1104 IB_FLOW_DOMAIN_USER,
1105 IB_FLOW_DOMAIN_ETHTOOL,
1108 IB_FLOW_DOMAIN_NUM /* Must be last */
1111 struct ib_flow_eth_filter {
1118 struct ib_flow_spec_eth {
1119 enum ib_flow_spec_type type;
1121 struct ib_flow_eth_filter val;
1122 struct ib_flow_eth_filter mask;
1125 struct ib_flow_ipv4_filter {
1130 struct ib_flow_spec_ipv4 {
1131 enum ib_flow_spec_type type;
1133 struct ib_flow_ipv4_filter val;
1134 struct ib_flow_ipv4_filter mask;
1137 struct ib_flow_tcp_udp_filter {
1142 struct ib_flow_spec_tcp_udp {
1143 enum ib_flow_spec_type type;
1145 struct ib_flow_tcp_udp_filter val;
1146 struct ib_flow_tcp_udp_filter mask;
1149 union ib_flow_spec {
1151 enum ib_flow_spec_type type;
1154 struct ib_flow_spec_eth eth;
1155 struct ib_flow_spec_ipv4 ipv4;
1156 struct ib_flow_spec_tcp_udp tcp_udp;
1159 struct ib_flow_attr {
1160 enum ib_flow_attr_type type;
1166 /* Following are the optional layers according to user request
1167 * struct ib_flow_spec_xxx
1168 * struct ib_flow_spec_yyy
1174 struct ib_uobject *uobject;
1180 enum ib_process_mad_flags {
1181 IB_MAD_IGNORE_MKEY = 1,
1182 IB_MAD_IGNORE_BKEY = 2,
1183 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1186 enum ib_mad_result {
1187 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1188 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1189 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1190 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1193 #define IB_DEVICE_NAME_MAX 64
1197 struct ib_event_handler event_handler;
1198 struct ib_pkey_cache **pkey_cache;
1199 struct ib_gid_cache **gid_cache;
1203 struct ib_dma_mapping_ops {
1204 int (*mapping_error)(struct ib_device *dev,
1206 u64 (*map_single)(struct ib_device *dev,
1207 void *ptr, size_t size,
1208 enum dma_data_direction direction);
1209 void (*unmap_single)(struct ib_device *dev,
1210 u64 addr, size_t size,
1211 enum dma_data_direction direction);
1212 u64 (*map_page)(struct ib_device *dev,
1213 struct page *page, unsigned long offset,
1215 enum dma_data_direction direction);
1216 void (*unmap_page)(struct ib_device *dev,
1217 u64 addr, size_t size,
1218 enum dma_data_direction direction);
1219 int (*map_sg)(struct ib_device *dev,
1220 struct scatterlist *sg, int nents,
1221 enum dma_data_direction direction);
1222 void (*unmap_sg)(struct ib_device *dev,
1223 struct scatterlist *sg, int nents,
1224 enum dma_data_direction direction);
1225 u64 (*dma_address)(struct ib_device *dev,
1226 struct scatterlist *sg);
1227 unsigned int (*dma_len)(struct ib_device *dev,
1228 struct scatterlist *sg);
1229 void (*sync_single_for_cpu)(struct ib_device *dev,
1232 enum dma_data_direction dir);
1233 void (*sync_single_for_device)(struct ib_device *dev,
1236 enum dma_data_direction dir);
1237 void *(*alloc_coherent)(struct ib_device *dev,
1241 void (*free_coherent)(struct ib_device *dev,
1242 size_t size, void *cpu_addr,
1249 struct device *dma_device;
1251 char name[IB_DEVICE_NAME_MAX];
1253 struct list_head event_handler_list;
1254 spinlock_t event_handler_lock;
1256 spinlock_t client_data_lock;
1257 struct list_head core_list;
1258 struct list_head client_data_list;
1260 struct ib_cache cache;
1264 int num_comp_vectors;
1266 struct iw_cm_verbs *iwcm;
1268 int (*get_protocol_stats)(struct ib_device *device,
1269 union rdma_protocol_stats *stats);
1270 int (*query_device)(struct ib_device *device,
1271 struct ib_device_attr *device_attr);
1272 int (*query_port)(struct ib_device *device,
1274 struct ib_port_attr *port_attr);
1275 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1277 int (*query_gid)(struct ib_device *device,
1278 u8 port_num, int index,
1280 int (*query_pkey)(struct ib_device *device,
1281 u8 port_num, u16 index, u16 *pkey);
1282 int (*modify_device)(struct ib_device *device,
1283 int device_modify_mask,
1284 struct ib_device_modify *device_modify);
1285 int (*modify_port)(struct ib_device *device,
1286 u8 port_num, int port_modify_mask,
1287 struct ib_port_modify *port_modify);
1288 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1289 struct ib_udata *udata);
1290 int (*dealloc_ucontext)(struct ib_ucontext *context);
1291 int (*mmap)(struct ib_ucontext *context,
1292 struct vm_area_struct *vma);
1293 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1294 struct ib_ucontext *context,
1295 struct ib_udata *udata);
1296 int (*dealloc_pd)(struct ib_pd *pd);
1297 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1298 struct ib_ah_attr *ah_attr);
1299 int (*modify_ah)(struct ib_ah *ah,
1300 struct ib_ah_attr *ah_attr);
1301 int (*query_ah)(struct ib_ah *ah,
1302 struct ib_ah_attr *ah_attr);
1303 int (*destroy_ah)(struct ib_ah *ah);
1304 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1305 struct ib_srq_init_attr *srq_init_attr,
1306 struct ib_udata *udata);
1307 int (*modify_srq)(struct ib_srq *srq,
1308 struct ib_srq_attr *srq_attr,
1309 enum ib_srq_attr_mask srq_attr_mask,
1310 struct ib_udata *udata);
1311 int (*query_srq)(struct ib_srq *srq,
1312 struct ib_srq_attr *srq_attr);
1313 int (*destroy_srq)(struct ib_srq *srq);
1314 int (*post_srq_recv)(struct ib_srq *srq,
1315 struct ib_recv_wr *recv_wr,
1316 struct ib_recv_wr **bad_recv_wr);
1317 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1318 struct ib_qp_init_attr *qp_init_attr,
1319 struct ib_udata *udata);
1320 int (*modify_qp)(struct ib_qp *qp,
1321 struct ib_qp_attr *qp_attr,
1323 struct ib_udata *udata);
1324 int (*query_qp)(struct ib_qp *qp,
1325 struct ib_qp_attr *qp_attr,
1327 struct ib_qp_init_attr *qp_init_attr);
1328 int (*destroy_qp)(struct ib_qp *qp);
1329 int (*post_send)(struct ib_qp *qp,
1330 struct ib_send_wr *send_wr,
1331 struct ib_send_wr **bad_send_wr);
1332 int (*post_recv)(struct ib_qp *qp,
1333 struct ib_recv_wr *recv_wr,
1334 struct ib_recv_wr **bad_recv_wr);
1335 struct ib_cq * (*create_cq)(struct ib_device *device, int cqe,
1337 struct ib_ucontext *context,
1338 struct ib_udata *udata);
1339 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1341 int (*destroy_cq)(struct ib_cq *cq);
1342 int (*resize_cq)(struct ib_cq *cq, int cqe,
1343 struct ib_udata *udata);
1344 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1346 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1347 int (*req_notify_cq)(struct ib_cq *cq,
1348 enum ib_cq_notify_flags flags);
1349 int (*req_ncomp_notif)(struct ib_cq *cq,
1351 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1352 int mr_access_flags);
1353 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
1354 struct ib_phys_buf *phys_buf_array,
1356 int mr_access_flags,
1358 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1359 u64 start, u64 length,
1361 int mr_access_flags,
1362 struct ib_udata *udata);
1363 int (*query_mr)(struct ib_mr *mr,
1364 struct ib_mr_attr *mr_attr);
1365 int (*dereg_mr)(struct ib_mr *mr);
1366 struct ib_mr * (*alloc_fast_reg_mr)(struct ib_pd *pd,
1367 int max_page_list_len);
1368 struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1370 void (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1371 int (*rereg_phys_mr)(struct ib_mr *mr,
1374 struct ib_phys_buf *phys_buf_array,
1376 int mr_access_flags,
1378 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1379 enum ib_mw_type type);
1380 int (*bind_mw)(struct ib_qp *qp,
1382 struct ib_mw_bind *mw_bind);
1383 int (*dealloc_mw)(struct ib_mw *mw);
1384 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1385 int mr_access_flags,
1386 struct ib_fmr_attr *fmr_attr);
1387 int (*map_phys_fmr)(struct ib_fmr *fmr,
1388 u64 *page_list, int list_len,
1390 int (*unmap_fmr)(struct list_head *fmr_list);
1391 int (*dealloc_fmr)(struct ib_fmr *fmr);
1392 int (*attach_mcast)(struct ib_qp *qp,
1395 int (*detach_mcast)(struct ib_qp *qp,
1398 int (*process_mad)(struct ib_device *device,
1399 int process_mad_flags,
1401 struct ib_wc *in_wc,
1402 struct ib_grh *in_grh,
1403 struct ib_mad *in_mad,
1404 struct ib_mad *out_mad);
1405 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1406 struct ib_ucontext *ucontext,
1407 struct ib_udata *udata);
1408 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1409 struct ib_flow * (*create_flow)(struct ib_qp *qp,
1413 int (*destroy_flow)(struct ib_flow *flow_id);
1415 struct ib_dma_mapping_ops *dma_ops;
1417 struct module *owner;
1419 struct kobject *ports_parent;
1420 struct list_head port_list;
1423 IB_DEV_UNINITIALIZED,
1429 u64 uverbs_cmd_mask;
1440 void (*add) (struct ib_device *);
1441 void (*remove)(struct ib_device *);
1443 struct list_head list;
1446 struct ib_device *ib_alloc_device(size_t size);
1447 void ib_dealloc_device(struct ib_device *device);
1449 int ib_register_device(struct ib_device *device,
1450 int (*port_callback)(struct ib_device *,
1451 u8, struct kobject *));
1452 void ib_unregister_device(struct ib_device *device);
1454 int ib_register_client (struct ib_client *client);
1455 void ib_unregister_client(struct ib_client *client);
1457 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1458 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1461 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1463 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1466 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1468 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1472 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1473 * contains all required attributes and no attributes not allowed for
1474 * the given QP state transition.
1475 * @cur_state: Current QP state
1476 * @next_state: Next QP state
1478 * @mask: Mask of supplied QP attributes
1480 * This function is a helper function that a low-level driver's
1481 * modify_qp method can use to validate the consumer's input. It
1482 * checks that cur_state and next_state are valid QP states, that a
1483 * transition from cur_state to next_state is allowed by the IB spec,
1484 * and that the attribute mask supplied is allowed for the transition.
1486 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1487 enum ib_qp_type type, enum ib_qp_attr_mask mask);
1489 int ib_register_event_handler (struct ib_event_handler *event_handler);
1490 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1491 void ib_dispatch_event(struct ib_event *event);
1493 int ib_query_device(struct ib_device *device,
1494 struct ib_device_attr *device_attr);
1496 int ib_query_port(struct ib_device *device,
1497 u8 port_num, struct ib_port_attr *port_attr);
1499 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1502 int ib_query_gid(struct ib_device *device,
1503 u8 port_num, int index, union ib_gid *gid);
1505 int ib_query_pkey(struct ib_device *device,
1506 u8 port_num, u16 index, u16 *pkey);
1508 int ib_modify_device(struct ib_device *device,
1509 int device_modify_mask,
1510 struct ib_device_modify *device_modify);
1512 int ib_modify_port(struct ib_device *device,
1513 u8 port_num, int port_modify_mask,
1514 struct ib_port_modify *port_modify);
1516 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
1517 u8 *port_num, u16 *index);
1519 int ib_find_pkey(struct ib_device *device,
1520 u8 port_num, u16 pkey, u16 *index);
1523 * ib_alloc_pd - Allocates an unused protection domain.
1524 * @device: The device on which to allocate the protection domain.
1526 * A protection domain object provides an association between QPs, shared
1527 * receive queues, address handles, memory regions, and memory windows.
1529 struct ib_pd *ib_alloc_pd(struct ib_device *device);
1532 * ib_dealloc_pd - Deallocates a protection domain.
1533 * @pd: The protection domain to deallocate.
1535 int ib_dealloc_pd(struct ib_pd *pd);
1538 * ib_create_ah - Creates an address handle for the given address vector.
1539 * @pd: The protection domain associated with the address handle.
1540 * @ah_attr: The attributes of the address vector.
1542 * The address handle is used to reference a local or global destination
1543 * in all UD QP post sends.
1545 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
1548 * ib_init_ah_from_wc - Initializes address handle attributes from a
1550 * @device: Device on which the received message arrived.
1551 * @port_num: Port on which the received message arrived.
1552 * @wc: Work completion associated with the received message.
1553 * @grh: References the received global route header. This parameter is
1554 * ignored unless the work completion indicates that the GRH is valid.
1555 * @ah_attr: Returned attributes that can be used when creating an address
1556 * handle for replying to the message.
1558 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, struct ib_wc *wc,
1559 struct ib_grh *grh, struct ib_ah_attr *ah_attr);
1562 * ib_create_ah_from_wc - Creates an address handle associated with the
1563 * sender of the specified work completion.
1564 * @pd: The protection domain associated with the address handle.
1565 * @wc: Work completion information associated with a received message.
1566 * @grh: References the received global route header. This parameter is
1567 * ignored unless the work completion indicates that the GRH is valid.
1568 * @port_num: The outbound port number to associate with the address.
1570 * The address handle is used to reference a local or global destination
1571 * in all UD QP post sends.
1573 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
1574 struct ib_grh *grh, u8 port_num);
1577 * ib_modify_ah - Modifies the address vector associated with an address
1579 * @ah: The address handle to modify.
1580 * @ah_attr: The new address vector attributes to associate with the
1583 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1586 * ib_query_ah - Queries the address vector associated with an address
1588 * @ah: The address handle to query.
1589 * @ah_attr: The address vector attributes associated with the address
1592 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
1595 * ib_destroy_ah - Destroys an address handle.
1596 * @ah: The address handle to destroy.
1598 int ib_destroy_ah(struct ib_ah *ah);
1601 * ib_create_srq - Creates a SRQ associated with the specified protection
1603 * @pd: The protection domain associated with the SRQ.
1604 * @srq_init_attr: A list of initial attributes required to create the
1605 * SRQ. If SRQ creation succeeds, then the attributes are updated to
1606 * the actual capabilities of the created SRQ.
1608 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
1609 * requested size of the SRQ, and set to the actual values allocated
1610 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
1611 * will always be at least as large as the requested values.
1613 struct ib_srq *ib_create_srq(struct ib_pd *pd,
1614 struct ib_srq_init_attr *srq_init_attr);
1617 * ib_modify_srq - Modifies the attributes for the specified SRQ.
1618 * @srq: The SRQ to modify.
1619 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
1620 * the current values of selected SRQ attributes are returned.
1621 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
1622 * are being modified.
1624 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
1625 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
1626 * the number of receives queued drops below the limit.
1628 int ib_modify_srq(struct ib_srq *srq,
1629 struct ib_srq_attr *srq_attr,
1630 enum ib_srq_attr_mask srq_attr_mask);
1633 * ib_query_srq - Returns the attribute list and current values for the
1635 * @srq: The SRQ to query.
1636 * @srq_attr: The attributes of the specified SRQ.
1638 int ib_query_srq(struct ib_srq *srq,
1639 struct ib_srq_attr *srq_attr);
1642 * ib_destroy_srq - Destroys the specified SRQ.
1643 * @srq: The SRQ to destroy.
1645 int ib_destroy_srq(struct ib_srq *srq);
1648 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
1649 * @srq: The SRQ to post the work request on.
1650 * @recv_wr: A list of work requests to post on the receive queue.
1651 * @bad_recv_wr: On an immediate failure, this parameter will reference
1652 * the work request that failed to be posted on the QP.
1654 static inline int ib_post_srq_recv(struct ib_srq *srq,
1655 struct ib_recv_wr *recv_wr,
1656 struct ib_recv_wr **bad_recv_wr)
1658 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
1662 * ib_create_qp - Creates a QP associated with the specified protection
1664 * @pd: The protection domain associated with the QP.
1665 * @qp_init_attr: A list of initial attributes required to create the
1666 * QP. If QP creation succeeds, then the attributes are updated to
1667 * the actual capabilities of the created QP.
1669 struct ib_qp *ib_create_qp(struct ib_pd *pd,
1670 struct ib_qp_init_attr *qp_init_attr);
1673 * ib_modify_qp - Modifies the attributes for the specified QP and then
1674 * transitions the QP to the given state.
1675 * @qp: The QP to modify.
1676 * @qp_attr: On input, specifies the QP attributes to modify. On output,
1677 * the current values of selected QP attributes are returned.
1678 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
1679 * are being modified.
1681 int ib_modify_qp(struct ib_qp *qp,
1682 struct ib_qp_attr *qp_attr,
1686 * ib_query_qp - Returns the attribute list and current values for the
1688 * @qp: The QP to query.
1689 * @qp_attr: The attributes of the specified QP.
1690 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
1691 * @qp_init_attr: Additional attributes of the selected QP.
1693 * The qp_attr_mask may be used to limit the query to gathering only the
1694 * selected attributes.
1696 int ib_query_qp(struct ib_qp *qp,
1697 struct ib_qp_attr *qp_attr,
1699 struct ib_qp_init_attr *qp_init_attr);
1702 * ib_destroy_qp - Destroys the specified QP.
1703 * @qp: The QP to destroy.
1705 int ib_destroy_qp(struct ib_qp *qp);
1708 * ib_open_qp - Obtain a reference to an existing sharable QP.
1709 * @xrcd - XRC domain
1710 * @qp_open_attr: Attributes identifying the QP to open.
1712 * Returns a reference to a sharable QP.
1714 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
1715 struct ib_qp_open_attr *qp_open_attr);
1718 * ib_close_qp - Release an external reference to a QP.
1719 * @qp: The QP handle to release
1721 * The opened QP handle is released by the caller. The underlying
1722 * shared QP is not destroyed until all internal references are released.
1724 int ib_close_qp(struct ib_qp *qp);
1727 * ib_post_send - Posts a list of work requests to the send queue of
1729 * @qp: The QP to post the work request on.
1730 * @send_wr: A list of work requests to post on the send queue.
1731 * @bad_send_wr: On an immediate failure, this parameter will reference
1732 * the work request that failed to be posted on the QP.
1734 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
1735 * error is returned, the QP state shall not be affected,
1736 * ib_post_send() will return an immediate error after queueing any
1737 * earlier work requests in the list.
1739 static inline int ib_post_send(struct ib_qp *qp,
1740 struct ib_send_wr *send_wr,
1741 struct ib_send_wr **bad_send_wr)
1743 return qp->device->post_send(qp, send_wr, bad_send_wr);
1747 * ib_post_recv - Posts a list of work requests to the receive queue of
1749 * @qp: The QP to post the work request on.
1750 * @recv_wr: A list of work requests to post on the receive queue.
1751 * @bad_recv_wr: On an immediate failure, this parameter will reference
1752 * the work request that failed to be posted on the QP.
1754 static inline int ib_post_recv(struct ib_qp *qp,
1755 struct ib_recv_wr *recv_wr,
1756 struct ib_recv_wr **bad_recv_wr)
1758 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
1762 * ib_create_cq - Creates a CQ on the specified device.
1763 * @device: The device on which to create the CQ.
1764 * @comp_handler: A user-specified callback that is invoked when a
1765 * completion event occurs on the CQ.
1766 * @event_handler: A user-specified callback that is invoked when an
1767 * asynchronous event not associated with a completion occurs on the CQ.
1768 * @cq_context: Context associated with the CQ returned to the user via
1769 * the associated completion and event handlers.
1770 * @cqe: The minimum size of the CQ.
1771 * @comp_vector - Completion vector used to signal completion events.
1772 * Must be >= 0 and < context->num_comp_vectors.
1774 * Users can examine the cq structure to determine the actual CQ size.
1776 struct ib_cq *ib_create_cq(struct ib_device *device,
1777 ib_comp_handler comp_handler,
1778 void (*event_handler)(struct ib_event *, void *),
1779 void *cq_context, int cqe, int comp_vector);
1782 * ib_resize_cq - Modifies the capacity of the CQ.
1783 * @cq: The CQ to resize.
1784 * @cqe: The minimum size of the CQ.
1786 * Users can examine the cq structure to determine the actual CQ size.
1788 int ib_resize_cq(struct ib_cq *cq, int cqe);
1791 * ib_modify_cq - Modifies moderation params of the CQ
1792 * @cq: The CQ to modify.
1793 * @cq_count: number of CQEs that will trigger an event
1794 * @cq_period: max period of time in usec before triggering an event
1797 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
1800 * ib_destroy_cq - Destroys the specified CQ.
1801 * @cq: The CQ to destroy.
1803 int ib_destroy_cq(struct ib_cq *cq);
1806 * ib_poll_cq - poll a CQ for completion(s)
1807 * @cq:the CQ being polled
1808 * @num_entries:maximum number of completions to return
1809 * @wc:array of at least @num_entries &struct ib_wc where completions
1812 * Poll a CQ for (possibly multiple) completions. If the return value
1813 * is < 0, an error occurred. If the return value is >= 0, it is the
1814 * number of completions returned. If the return value is
1815 * non-negative and < num_entries, then the CQ was emptied.
1817 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
1820 return cq->device->poll_cq(cq, num_entries, wc);
1824 * ib_peek_cq - Returns the number of unreaped completions currently
1825 * on the specified CQ.
1826 * @cq: The CQ to peek.
1827 * @wc_cnt: A minimum number of unreaped completions to check for.
1829 * If the number of unreaped completions is greater than or equal to wc_cnt,
1830 * this function returns wc_cnt, otherwise, it returns the actual number of
1831 * unreaped completions.
1833 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
1836 * ib_req_notify_cq - Request completion notification on a CQ.
1837 * @cq: The CQ to generate an event for.
1839 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
1840 * to request an event on the next solicited event or next work
1841 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
1842 * may also be |ed in to request a hint about missed events, as
1846 * < 0 means an error occurred while requesting notification
1847 * == 0 means notification was requested successfully, and if
1848 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
1849 * were missed and it is safe to wait for another event. In
1850 * this case is it guaranteed that any work completions added
1851 * to the CQ since the last CQ poll will trigger a completion
1852 * notification event.
1853 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
1854 * in. It means that the consumer must poll the CQ again to
1855 * make sure it is empty to avoid missing an event because of a
1856 * race between requesting notification and an entry being
1857 * added to the CQ. This return value means it is possible
1858 * (but not guaranteed) that a work completion has been added
1859 * to the CQ since the last poll without triggering a
1860 * completion notification event.
1862 static inline int ib_req_notify_cq(struct ib_cq *cq,
1863 enum ib_cq_notify_flags flags)
1865 return cq->device->req_notify_cq(cq, flags);
1869 * ib_req_ncomp_notif - Request completion notification when there are
1870 * at least the specified number of unreaped completions on the CQ.
1871 * @cq: The CQ to generate an event for.
1872 * @wc_cnt: The number of unreaped completions that should be on the
1873 * CQ before an event is generated.
1875 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
1877 return cq->device->req_ncomp_notif ?
1878 cq->device->req_ncomp_notif(cq, wc_cnt) :
1883 * ib_get_dma_mr - Returns a memory region for system memory that is
1885 * @pd: The protection domain associated with the memory region.
1886 * @mr_access_flags: Specifies the memory access rights.
1888 * Note that the ib_dma_*() functions defined below must be used
1889 * to create/destroy addresses used with the Lkey or Rkey returned
1890 * by ib_get_dma_mr().
1892 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
1895 * ib_dma_mapping_error - check a DMA addr for error
1896 * @dev: The device for which the dma_addr was created
1897 * @dma_addr: The DMA address to check
1899 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
1902 return dev->dma_ops->mapping_error(dev, dma_addr);
1903 return dma_mapping_error(dev->dma_device, dma_addr);
1907 * ib_dma_map_single - Map a kernel virtual address to DMA address
1908 * @dev: The device for which the dma_addr is to be created
1909 * @cpu_addr: The kernel virtual address
1910 * @size: The size of the region in bytes
1911 * @direction: The direction of the DMA
1913 static inline u64 ib_dma_map_single(struct ib_device *dev,
1914 void *cpu_addr, size_t size,
1915 enum dma_data_direction direction)
1918 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
1919 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
1923 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
1924 * @dev: The device for which the DMA address was created
1925 * @addr: The DMA address
1926 * @size: The size of the region in bytes
1927 * @direction: The direction of the DMA
1929 static inline void ib_dma_unmap_single(struct ib_device *dev,
1930 u64 addr, size_t size,
1931 enum dma_data_direction direction)
1934 dev->dma_ops->unmap_single(dev, addr, size, direction);
1936 dma_unmap_single(dev->dma_device, addr, size, direction);
1939 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
1940 void *cpu_addr, size_t size,
1941 enum dma_data_direction direction,
1942 struct dma_attrs *attrs)
1944 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
1948 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
1949 u64 addr, size_t size,
1950 enum dma_data_direction direction,
1951 struct dma_attrs *attrs)
1953 return dma_unmap_single_attrs(dev->dma_device, addr, size,
1958 * ib_dma_map_page - Map a physical page to DMA address
1959 * @dev: The device for which the dma_addr is to be created
1960 * @page: The page to be mapped
1961 * @offset: The offset within the page
1962 * @size: The size of the region in bytes
1963 * @direction: The direction of the DMA
1965 static inline u64 ib_dma_map_page(struct ib_device *dev,
1967 unsigned long offset,
1969 enum dma_data_direction direction)
1972 return dev->dma_ops->map_page(dev, page, offset, size, direction);
1973 return dma_map_page(dev->dma_device, page, offset, size, direction);
1977 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
1978 * @dev: The device for which the DMA address was created
1979 * @addr: The DMA address
1980 * @size: The size of the region in bytes
1981 * @direction: The direction of the DMA
1983 static inline void ib_dma_unmap_page(struct ib_device *dev,
1984 u64 addr, size_t size,
1985 enum dma_data_direction direction)
1988 dev->dma_ops->unmap_page(dev, addr, size, direction);
1990 dma_unmap_page(dev->dma_device, addr, size, direction);
1994 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
1995 * @dev: The device for which the DMA addresses are to be created
1996 * @sg: The array of scatter/gather entries
1997 * @nents: The number of scatter/gather entries
1998 * @direction: The direction of the DMA
2000 static inline int ib_dma_map_sg(struct ib_device *dev,
2001 struct scatterlist *sg, int nents,
2002 enum dma_data_direction direction)
2005 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2006 return dma_map_sg(dev->dma_device, sg, nents, direction);
2010 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2011 * @dev: The device for which the DMA addresses were created
2012 * @sg: The array of scatter/gather entries
2013 * @nents: The number of scatter/gather entries
2014 * @direction: The direction of the DMA
2016 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2017 struct scatterlist *sg, int nents,
2018 enum dma_data_direction direction)
2021 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2023 dma_unmap_sg(dev->dma_device, sg, nents, direction);
2026 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2027 struct scatterlist *sg, int nents,
2028 enum dma_data_direction direction,
2029 struct dma_attrs *attrs)
2031 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2034 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2035 struct scatterlist *sg, int nents,
2036 enum dma_data_direction direction,
2037 struct dma_attrs *attrs)
2039 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2042 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2043 * @dev: The device for which the DMA addresses were created
2044 * @sg: The scatter/gather entry
2046 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2047 struct scatterlist *sg)
2050 return dev->dma_ops->dma_address(dev, sg);
2051 return sg_dma_address(sg);
2055 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2056 * @dev: The device for which the DMA addresses were created
2057 * @sg: The scatter/gather entry
2059 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2060 struct scatterlist *sg)
2063 return dev->dma_ops->dma_len(dev, sg);
2064 return sg_dma_len(sg);
2068 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2069 * @dev: The device for which the DMA address was created
2070 * @addr: The DMA address
2071 * @size: The size of the region in bytes
2072 * @dir: The direction of the DMA
2074 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2077 enum dma_data_direction dir)
2080 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2082 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2086 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2087 * @dev: The device for which the DMA address was created
2088 * @addr: The DMA address
2089 * @size: The size of the region in bytes
2090 * @dir: The direction of the DMA
2092 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2095 enum dma_data_direction dir)
2098 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2100 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2104 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2105 * @dev: The device for which the DMA address is requested
2106 * @size: The size of the region to allocate in bytes
2107 * @dma_handle: A pointer for returning the DMA address of the region
2108 * @flag: memory allocator flags
2110 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2116 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2121 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2122 *dma_handle = handle;
2128 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2129 * @dev: The device for which the DMA addresses were allocated
2130 * @size: The size of the region
2131 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2132 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2134 static inline void ib_dma_free_coherent(struct ib_device *dev,
2135 size_t size, void *cpu_addr,
2139 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2141 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2145 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
2147 * @pd: The protection domain associated assigned to the registered region.
2148 * @phys_buf_array: Specifies a list of physical buffers to use in the
2150 * @num_phys_buf: Specifies the size of the phys_buf_array.
2151 * @mr_access_flags: Specifies the memory access rights.
2152 * @iova_start: The offset of the region's starting I/O virtual address.
2154 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
2155 struct ib_phys_buf *phys_buf_array,
2157 int mr_access_flags,
2161 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
2162 * Conceptually, this call performs the functions deregister memory region
2163 * followed by register physical memory region. Where possible,
2164 * resources are reused instead of deallocated and reallocated.
2165 * @mr: The memory region to modify.
2166 * @mr_rereg_mask: A bit-mask used to indicate which of the following
2167 * properties of the memory region are being modified.
2168 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
2169 * the new protection domain to associated with the memory region,
2170 * otherwise, this parameter is ignored.
2171 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2172 * field specifies a list of physical buffers to use in the new
2173 * translation, otherwise, this parameter is ignored.
2174 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2175 * field specifies the size of the phys_buf_array, otherwise, this
2176 * parameter is ignored.
2177 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
2178 * field specifies the new memory access rights, otherwise, this
2179 * parameter is ignored.
2180 * @iova_start: The offset of the region's starting I/O virtual address.
2182 int ib_rereg_phys_mr(struct ib_mr *mr,
2185 struct ib_phys_buf *phys_buf_array,
2187 int mr_access_flags,
2191 * ib_query_mr - Retrieves information about a specific memory region.
2192 * @mr: The memory region to retrieve information about.
2193 * @mr_attr: The attributes of the specified memory region.
2195 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2198 * ib_dereg_mr - Deregisters a memory region and removes it from the
2199 * HCA translation table.
2200 * @mr: The memory region to deregister.
2202 * This function can fail, if the memory region has memory windows bound to it.
2204 int ib_dereg_mr(struct ib_mr *mr);
2207 * ib_alloc_fast_reg_mr - Allocates memory region usable with the
2208 * IB_WR_FAST_REG_MR send work request.
2209 * @pd: The protection domain associated with the region.
2210 * @max_page_list_len: requested max physical buffer list length to be
2211 * used with fast register work requests for this MR.
2213 struct ib_mr *ib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
2216 * ib_alloc_fast_reg_page_list - Allocates a page list array
2217 * @device - ib device pointer.
2218 * @page_list_len - size of the page list array to be allocated.
2220 * This allocates and returns a struct ib_fast_reg_page_list * and a
2221 * page_list array that is at least page_list_len in size. The actual
2222 * size is returned in max_page_list_len. The caller is responsible
2223 * for initializing the contents of the page_list array before posting
2224 * a send work request with the IB_WC_FAST_REG_MR opcode.
2226 * The page_list array entries must be translated using one of the
2227 * ib_dma_*() functions just like the addresses passed to
2228 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
2229 * ib_fast_reg_page_list must not be modified by the caller until the
2230 * IB_WC_FAST_REG_MR work request completes.
2232 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2233 struct ib_device *device, int page_list_len);
2236 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2238 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2240 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2243 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2245 * @mr - struct ib_mr pointer to be updated.
2246 * @newkey - new key to be used.
2248 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2250 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2251 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2255 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2256 * for calculating a new rkey for type 2 memory windows.
2257 * @rkey - the rkey to increment.
2259 static inline u32 ib_inc_rkey(u32 rkey)
2261 const u32 mask = 0x000000ff;
2262 return ((rkey + 1) & mask) | (rkey & ~mask);
2266 * ib_alloc_mw - Allocates a memory window.
2267 * @pd: The protection domain associated with the memory window.
2268 * @type: The type of the memory window (1 or 2).
2270 struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
2273 * ib_bind_mw - Posts a work request to the send queue of the specified
2274 * QP, which binds the memory window to the given address range and
2275 * remote access attributes.
2276 * @qp: QP to post the bind work request on.
2277 * @mw: The memory window to bind.
2278 * @mw_bind: Specifies information about the memory window, including
2279 * its address range, remote access rights, and associated memory region.
2281 * If there is no immediate error, the function will update the rkey member
2282 * of the mw parameter to its new value. The bind operation can still fail
2285 static inline int ib_bind_mw(struct ib_qp *qp,
2287 struct ib_mw_bind *mw_bind)
2289 /* XXX reference counting in corresponding MR? */
2290 return mw->device->bind_mw ?
2291 mw->device->bind_mw(qp, mw, mw_bind) :
2296 * ib_dealloc_mw - Deallocates a memory window.
2297 * @mw: The memory window to deallocate.
2299 int ib_dealloc_mw(struct ib_mw *mw);
2302 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2303 * @pd: The protection domain associated with the unmapped region.
2304 * @mr_access_flags: Specifies the memory access rights.
2305 * @fmr_attr: Attributes of the unmapped region.
2307 * A fast memory region must be mapped before it can be used as part of
2310 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2311 int mr_access_flags,
2312 struct ib_fmr_attr *fmr_attr);
2315 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2316 * @fmr: The fast memory region to associate with the pages.
2317 * @page_list: An array of physical pages to map to the fast memory region.
2318 * @list_len: The number of pages in page_list.
2319 * @iova: The I/O virtual address to use with the mapped region.
2321 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2322 u64 *page_list, int list_len,
2325 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2329 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2330 * @fmr_list: A linked list of fast memory regions to unmap.
2332 int ib_unmap_fmr(struct list_head *fmr_list);
2335 * ib_dealloc_fmr - Deallocates a fast memory region.
2336 * @fmr: The fast memory region to deallocate.
2338 int ib_dealloc_fmr(struct ib_fmr *fmr);
2341 * ib_attach_mcast - Attaches the specified QP to a multicast group.
2342 * @qp: QP to attach to the multicast group. The QP must be type
2344 * @gid: Multicast group GID.
2345 * @lid: Multicast group LID in host byte order.
2347 * In order to send and receive multicast packets, subnet
2348 * administration must have created the multicast group and configured
2349 * the fabric appropriately. The port associated with the specified
2350 * QP must also be a member of the multicast group.
2352 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2355 * ib_detach_mcast - Detaches the specified QP from a multicast group.
2356 * @qp: QP to detach from the multicast group.
2357 * @gid: Multicast group GID.
2358 * @lid: Multicast group LID in host byte order.
2360 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
2363 * ib_alloc_xrcd - Allocates an XRC domain.
2364 * @device: The device on which to allocate the XRC domain.
2366 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
2369 * ib_dealloc_xrcd - Deallocates an XRC domain.
2370 * @xrcd: The XRC domain to deallocate.
2372 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
2374 struct ib_flow *ib_create_flow(struct ib_qp *qp,
2375 struct ib_flow_attr *flow_attr, int domain);
2376 int ib_destroy_flow(struct ib_flow *flow_id);
2378 #endif /* IB_VERBS_H */