2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
4 * Copyright (c) 2004 Intel Corporation. All rights reserved.
5 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved.
10 * This software is available to you under a choice of one of two
11 * licenses. You may choose to be licensed under the terms of the GNU
12 * General Public License (GPL) Version 2, available from the file
13 * COPYING in the main directory of this source tree, or the
14 * OpenIB.org BSD license below:
16 * Redistribution and use in source and binary forms, with or
17 * without modification, are permitted provided that the following
20 * - Redistributions of source code must retain the above
21 * copyright notice, this list of conditions and the following
24 * - Redistributions in binary form must reproduce the above
25 * copyright notice, this list of conditions and the following
26 * disclaimer in the documentation and/or other materials
27 * provided with the distribution.
29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
39 #if !defined(IB_VERBS_H)
42 #include <linux/types.h>
43 #include <linux/device.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <uapi/linux/if_ether.h>
54 #include <linux/atomic.h>
55 #include <linux/mmu_notifier.h>
56 #include <asm/uaccess.h>
58 extern struct workqueue_struct *ib_wq;
68 extern union ib_gid zgid;
71 struct net_device *ndev;
75 /* IB values map to NodeInfo:NodeType. */
84 enum rdma_transport_type {
88 RDMA_TRANSPORT_USNIC_UDP
91 enum rdma_protocol_type {
95 RDMA_PROTOCOL_USNIC_UDP
98 __attribute_const__ enum rdma_transport_type
99 rdma_node_get_transport(enum rdma_node_type node_type);
101 enum rdma_link_layer {
102 IB_LINK_LAYER_UNSPECIFIED,
103 IB_LINK_LAYER_INFINIBAND,
104 IB_LINK_LAYER_ETHERNET,
107 enum ib_device_cap_flags {
108 IB_DEVICE_RESIZE_MAX_WR = 1,
109 IB_DEVICE_BAD_PKEY_CNTR = (1<<1),
110 IB_DEVICE_BAD_QKEY_CNTR = (1<<2),
111 IB_DEVICE_RAW_MULTI = (1<<3),
112 IB_DEVICE_AUTO_PATH_MIG = (1<<4),
113 IB_DEVICE_CHANGE_PHY_PORT = (1<<5),
114 IB_DEVICE_UD_AV_PORT_ENFORCE = (1<<6),
115 IB_DEVICE_CURR_QP_STATE_MOD = (1<<7),
116 IB_DEVICE_SHUTDOWN_PORT = (1<<8),
117 IB_DEVICE_INIT_TYPE = (1<<9),
118 IB_DEVICE_PORT_ACTIVE_EVENT = (1<<10),
119 IB_DEVICE_SYS_IMAGE_GUID = (1<<11),
120 IB_DEVICE_RC_RNR_NAK_GEN = (1<<12),
121 IB_DEVICE_SRQ_RESIZE = (1<<13),
122 IB_DEVICE_N_NOTIFY_CQ = (1<<14),
123 IB_DEVICE_LOCAL_DMA_LKEY = (1<<15),
124 IB_DEVICE_RESERVED = (1<<16), /* old SEND_W_INV */
125 IB_DEVICE_MEM_WINDOW = (1<<17),
127 * Devices should set IB_DEVICE_UD_IP_SUM if they support
128 * insertion of UDP and TCP checksum on outgoing UD IPoIB
129 * messages and can verify the validity of checksum for
130 * incoming messages. Setting this flag implies that the
131 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
133 IB_DEVICE_UD_IP_CSUM = (1<<18),
134 IB_DEVICE_UD_TSO = (1<<19),
135 IB_DEVICE_XRC = (1<<20),
136 IB_DEVICE_MEM_MGT_EXTENSIONS = (1<<21),
137 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
138 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1<<23),
139 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1<<24),
140 IB_DEVICE_MANAGED_FLOW_STEERING = (1<<29),
141 IB_DEVICE_SIGNATURE_HANDOVER = (1<<30),
142 IB_DEVICE_ON_DEMAND_PAGING = (1<<31),
145 enum ib_signature_prot_cap {
146 IB_PROT_T10DIF_TYPE_1 = 1,
147 IB_PROT_T10DIF_TYPE_2 = 1 << 1,
148 IB_PROT_T10DIF_TYPE_3 = 1 << 2,
151 enum ib_signature_guard_cap {
152 IB_GUARD_T10DIF_CRC = 1,
153 IB_GUARD_T10DIF_CSUM = 1 << 1,
162 enum ib_odp_general_cap_bits {
163 IB_ODP_SUPPORT = 1 << 0,
166 enum ib_odp_transport_cap_bits {
167 IB_ODP_SUPPORT_SEND = 1 << 0,
168 IB_ODP_SUPPORT_RECV = 1 << 1,
169 IB_ODP_SUPPORT_WRITE = 1 << 2,
170 IB_ODP_SUPPORT_READ = 1 << 3,
171 IB_ODP_SUPPORT_ATOMIC = 1 << 4,
175 uint64_t general_caps;
177 uint32_t rc_odp_caps;
178 uint32_t uc_odp_caps;
179 uint32_t ud_odp_caps;
180 } per_transport_caps;
183 enum ib_cq_creation_flags {
184 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0,
187 struct ib_cq_init_attr {
193 struct ib_device_attr {
195 __be64 sys_image_guid;
203 int device_cap_flags;
213 int max_qp_init_rd_atom;
214 int max_ee_init_rd_atom;
215 enum ib_atomic_cap atomic_cap;
216 enum ib_atomic_cap masked_atomic_cap;
223 int max_mcast_qp_attach;
224 int max_total_mcast_qp_attach;
231 unsigned int max_fast_reg_page_list_len;
233 u8 local_ca_ack_delay;
236 struct ib_odp_caps odp_caps;
237 uint64_t timestamp_mask;
238 uint64_t hca_core_clock; /* in KHZ */
249 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
252 case IB_MTU_256: return 256;
253 case IB_MTU_512: return 512;
254 case IB_MTU_1024: return 1024;
255 case IB_MTU_2048: return 2048;
256 case IB_MTU_4096: return 4096;
267 IB_PORT_ACTIVE_DEFER = 5
270 enum ib_port_cap_flags {
272 IB_PORT_NOTICE_SUP = 1 << 2,
273 IB_PORT_TRAP_SUP = 1 << 3,
274 IB_PORT_OPT_IPD_SUP = 1 << 4,
275 IB_PORT_AUTO_MIGR_SUP = 1 << 5,
276 IB_PORT_SL_MAP_SUP = 1 << 6,
277 IB_PORT_MKEY_NVRAM = 1 << 7,
278 IB_PORT_PKEY_NVRAM = 1 << 8,
279 IB_PORT_LED_INFO_SUP = 1 << 9,
280 IB_PORT_SM_DISABLED = 1 << 10,
281 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11,
282 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12,
283 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14,
284 IB_PORT_CM_SUP = 1 << 16,
285 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17,
286 IB_PORT_REINIT_SUP = 1 << 18,
287 IB_PORT_DEVICE_MGMT_SUP = 1 << 19,
288 IB_PORT_VENDOR_CLASS_SUP = 1 << 20,
289 IB_PORT_DR_NOTICE_SUP = 1 << 21,
290 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22,
291 IB_PORT_BOOT_MGMT_SUP = 1 << 23,
292 IB_PORT_LINK_LATENCY_SUP = 1 << 24,
293 IB_PORT_CLIENT_REG_SUP = 1 << 25,
294 IB_PORT_IP_BASED_GIDS = 1 << 26,
304 static inline int ib_width_enum_to_int(enum ib_port_width width)
307 case IB_WIDTH_1X: return 1;
308 case IB_WIDTH_4X: return 4;
309 case IB_WIDTH_8X: return 8;
310 case IB_WIDTH_12X: return 12;
324 struct ib_protocol_stats {
328 struct iw_protocol_stats {
331 u64 ipInTooBigErrors;
334 u64 ipInUnknownProtos;
335 u64 ipInTruncatedPkts;
338 u64 ipOutForwDatagrams;
370 union rdma_protocol_stats {
371 struct ib_protocol_stats ib;
372 struct iw_protocol_stats iw;
375 /* Define bits for the various functionality this port needs to be supported by
378 /* Management 0x00000FFF */
379 #define RDMA_CORE_CAP_IB_MAD 0x00000001
380 #define RDMA_CORE_CAP_IB_SMI 0x00000002
381 #define RDMA_CORE_CAP_IB_CM 0x00000004
382 #define RDMA_CORE_CAP_IW_CM 0x00000008
383 #define RDMA_CORE_CAP_IB_SA 0x00000010
384 #define RDMA_CORE_CAP_OPA_MAD 0x00000020
386 /* Address format 0x000FF000 */
387 #define RDMA_CORE_CAP_AF_IB 0x00001000
388 #define RDMA_CORE_CAP_ETH_AH 0x00002000
390 /* Protocol 0xFFF00000 */
391 #define RDMA_CORE_CAP_PROT_IB 0x00100000
392 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000
393 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000
395 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \
396 | RDMA_CORE_CAP_IB_MAD \
397 | RDMA_CORE_CAP_IB_SMI \
398 | RDMA_CORE_CAP_IB_CM \
399 | RDMA_CORE_CAP_IB_SA \
400 | RDMA_CORE_CAP_AF_IB)
401 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \
402 | RDMA_CORE_CAP_IB_MAD \
403 | RDMA_CORE_CAP_IB_CM \
404 | RDMA_CORE_CAP_AF_IB \
405 | RDMA_CORE_CAP_ETH_AH)
406 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \
407 | RDMA_CORE_CAP_IW_CM)
408 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \
409 | RDMA_CORE_CAP_OPA_MAD)
411 struct ib_port_attr {
412 enum ib_port_state state;
414 enum ib_mtu active_mtu;
433 enum ib_device_modify_flags {
434 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0,
435 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1
438 struct ib_device_modify {
443 enum ib_port_modify_flags {
444 IB_PORT_SHUTDOWN = 1,
445 IB_PORT_INIT_TYPE = (1<<2),
446 IB_PORT_RESET_QKEY_CNTR = (1<<3)
449 struct ib_port_modify {
450 u32 set_port_cap_mask;
451 u32 clr_port_cap_mask;
459 IB_EVENT_QP_ACCESS_ERR,
463 IB_EVENT_PATH_MIG_ERR,
464 IB_EVENT_DEVICE_FATAL,
465 IB_EVENT_PORT_ACTIVE,
468 IB_EVENT_PKEY_CHANGE,
471 IB_EVENT_SRQ_LIMIT_REACHED,
472 IB_EVENT_QP_LAST_WQE_REACHED,
473 IB_EVENT_CLIENT_REREGISTER,
477 __attribute_const__ const char *ib_event_msg(enum ib_event_type event);
480 struct ib_device *device;
487 enum ib_event_type event;
490 struct ib_event_handler {
491 struct ib_device *device;
492 void (*handler)(struct ib_event_handler *, struct ib_event *);
493 struct list_head list;
496 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \
498 (_ptr)->device = _device; \
499 (_ptr)->handler = _handler; \
500 INIT_LIST_HEAD(&(_ptr)->list); \
503 struct ib_global_route {
512 __be32 version_tclass_flow;
521 IB_MULTICAST_QPN = 0xffffff
524 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF)
531 IB_RATE_PORT_CURRENT = 0,
532 IB_RATE_2_5_GBPS = 2,
540 IB_RATE_120_GBPS = 10,
541 IB_RATE_14_GBPS = 11,
542 IB_RATE_56_GBPS = 12,
543 IB_RATE_112_GBPS = 13,
544 IB_RATE_168_GBPS = 14,
545 IB_RATE_25_GBPS = 15,
546 IB_RATE_100_GBPS = 16,
547 IB_RATE_200_GBPS = 17,
548 IB_RATE_300_GBPS = 18
552 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
553 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be
554 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
555 * @rate: rate to convert.
557 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
560 * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
561 * For example, IB_RATE_2_5_GBPS will be converted to 2500.
562 * @rate: rate to convert.
564 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
568 * enum ib_mr_type - memory region type
569 * @IB_MR_TYPE_MEM_REG: memory region that is used for
570 * normal registration
571 * @IB_MR_TYPE_SIGNATURE: memory region that is used for
572 * signature operations (data-integrity
577 IB_MR_TYPE_SIGNATURE,
582 * IB_SIG_TYPE_NONE: Unprotected.
583 * IB_SIG_TYPE_T10_DIF: Type T10-DIF
585 enum ib_signature_type {
591 * Signature T10-DIF block-guard types
592 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
593 * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
595 enum ib_t10_dif_bg_type {
601 * struct ib_t10_dif_domain - Parameters specific for T10-DIF
603 * @bg_type: T10-DIF block guard type (CRC|CSUM)
604 * @pi_interval: protection information interval.
605 * @bg: seed of guard computation.
606 * @app_tag: application tag of guard block
607 * @ref_tag: initial guard block reference tag.
608 * @ref_remap: Indicate wethear the reftag increments each block
609 * @app_escape: Indicate to skip block check if apptag=0xffff
610 * @ref_escape: Indicate to skip block check if reftag=0xffffffff
611 * @apptag_check_mask: check bitmask of application tag.
613 struct ib_t10_dif_domain {
614 enum ib_t10_dif_bg_type bg_type;
622 u16 apptag_check_mask;
626 * struct ib_sig_domain - Parameters for signature domain
627 * @sig_type: specific signauture type
628 * @sig: union of all signature domain attributes that may
629 * be used to set domain layout.
631 struct ib_sig_domain {
632 enum ib_signature_type sig_type;
634 struct ib_t10_dif_domain dif;
639 * struct ib_sig_attrs - Parameters for signature handover operation
640 * @check_mask: bitmask for signature byte check (8 bytes)
641 * @mem: memory domain layout desciptor.
642 * @wire: wire domain layout desciptor.
644 struct ib_sig_attrs {
646 struct ib_sig_domain mem;
647 struct ib_sig_domain wire;
650 enum ib_sig_err_type {
657 * struct ib_sig_err - signature error descriptor
660 enum ib_sig_err_type err_type;
667 enum ib_mr_status_check {
668 IB_MR_CHECK_SIG_STATUS = 1,
672 * struct ib_mr_status - Memory region status container
674 * @fail_status: Bitmask of MR checks status. For each
675 * failed check a corresponding status bit is set.
676 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
679 struct ib_mr_status {
681 struct ib_sig_err sig_err;
685 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
687 * @mult: multiple to convert.
689 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
692 struct ib_global_route grh;
707 IB_WC_LOC_EEC_OP_ERR,
712 IB_WC_LOC_ACCESS_ERR,
713 IB_WC_REM_INV_REQ_ERR,
714 IB_WC_REM_ACCESS_ERR,
717 IB_WC_RNR_RETRY_EXC_ERR,
718 IB_WC_LOC_RDD_VIOL_ERR,
719 IB_WC_REM_INV_RD_REQ_ERR,
722 IB_WC_INV_EEC_STATE_ERR,
724 IB_WC_RESP_TIMEOUT_ERR,
728 __attribute_const__ const char *ib_wc_status_msg(enum ib_wc_status status);
740 IB_WC_MASKED_COMP_SWAP,
741 IB_WC_MASKED_FETCH_ADD,
743 * Set value of IB_WC_RECV so consumers can test if a completion is a
744 * receive by testing (opcode & IB_WC_RECV).
747 IB_WC_RECV_RDMA_WITH_IMM
752 IB_WC_WITH_IMM = (1<<1),
753 IB_WC_WITH_INVALIDATE = (1<<2),
754 IB_WC_IP_CSUM_OK = (1<<3),
755 IB_WC_WITH_SMAC = (1<<4),
756 IB_WC_WITH_VLAN = (1<<5),
761 enum ib_wc_status status;
762 enum ib_wc_opcode opcode;
776 u8 port_num; /* valid only for DR SMPs on switches */
781 enum ib_cq_notify_flags {
782 IB_CQ_SOLICITED = 1 << 0,
783 IB_CQ_NEXT_COMP = 1 << 1,
784 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
785 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2,
793 enum ib_srq_attr_mask {
794 IB_SRQ_MAX_WR = 1 << 0,
795 IB_SRQ_LIMIT = 1 << 1,
804 struct ib_srq_init_attr {
805 void (*event_handler)(struct ib_event *, void *);
807 struct ib_srq_attr attr;
808 enum ib_srq_type srq_type;
812 struct ib_xrcd *xrcd;
833 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
834 * here (and in that order) since the MAD layer uses them as
835 * indices into a 2-entry table.
844 IB_QPT_RAW_ETHERTYPE,
845 IB_QPT_RAW_PACKET = 8,
849 /* Reserve a range for qp types internal to the low level driver.
850 * These qp types will not be visible at the IB core layer, so the
851 * IB_QPT_MAX usages should not be affected in the core layer
853 IB_QPT_RESERVED1 = 0x1000,
865 enum ib_qp_create_flags {
866 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0,
867 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
868 IB_QP_CREATE_NETIF_QP = 1 << 5,
869 IB_QP_CREATE_SIGNATURE_EN = 1 << 6,
870 IB_QP_CREATE_USE_GFP_NOIO = 1 << 7,
871 /* reserve bits 26-31 for low level drivers' internal use */
872 IB_QP_CREATE_RESERVED_START = 1 << 26,
873 IB_QP_CREATE_RESERVED_END = 1 << 31,
878 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
879 * callback to destroy the passed in QP.
882 struct ib_qp_init_attr {
883 void (*event_handler)(struct ib_event *, void *);
885 struct ib_cq *send_cq;
886 struct ib_cq *recv_cq;
888 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
889 struct ib_qp_cap cap;
890 enum ib_sig_type sq_sig_type;
891 enum ib_qp_type qp_type;
892 enum ib_qp_create_flags create_flags;
893 u8 port_num; /* special QP types only */
896 struct ib_qp_open_attr {
897 void (*event_handler)(struct ib_event *, void *);
900 enum ib_qp_type qp_type;
903 enum ib_rnr_timeout {
904 IB_RNR_TIMER_655_36 = 0,
905 IB_RNR_TIMER_000_01 = 1,
906 IB_RNR_TIMER_000_02 = 2,
907 IB_RNR_TIMER_000_03 = 3,
908 IB_RNR_TIMER_000_04 = 4,
909 IB_RNR_TIMER_000_06 = 5,
910 IB_RNR_TIMER_000_08 = 6,
911 IB_RNR_TIMER_000_12 = 7,
912 IB_RNR_TIMER_000_16 = 8,
913 IB_RNR_TIMER_000_24 = 9,
914 IB_RNR_TIMER_000_32 = 10,
915 IB_RNR_TIMER_000_48 = 11,
916 IB_RNR_TIMER_000_64 = 12,
917 IB_RNR_TIMER_000_96 = 13,
918 IB_RNR_TIMER_001_28 = 14,
919 IB_RNR_TIMER_001_92 = 15,
920 IB_RNR_TIMER_002_56 = 16,
921 IB_RNR_TIMER_003_84 = 17,
922 IB_RNR_TIMER_005_12 = 18,
923 IB_RNR_TIMER_007_68 = 19,
924 IB_RNR_TIMER_010_24 = 20,
925 IB_RNR_TIMER_015_36 = 21,
926 IB_RNR_TIMER_020_48 = 22,
927 IB_RNR_TIMER_030_72 = 23,
928 IB_RNR_TIMER_040_96 = 24,
929 IB_RNR_TIMER_061_44 = 25,
930 IB_RNR_TIMER_081_92 = 26,
931 IB_RNR_TIMER_122_88 = 27,
932 IB_RNR_TIMER_163_84 = 28,
933 IB_RNR_TIMER_245_76 = 29,
934 IB_RNR_TIMER_327_68 = 30,
935 IB_RNR_TIMER_491_52 = 31
938 enum ib_qp_attr_mask {
940 IB_QP_CUR_STATE = (1<<1),
941 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2),
942 IB_QP_ACCESS_FLAGS = (1<<3),
943 IB_QP_PKEY_INDEX = (1<<4),
947 IB_QP_PATH_MTU = (1<<8),
948 IB_QP_TIMEOUT = (1<<9),
949 IB_QP_RETRY_CNT = (1<<10),
950 IB_QP_RNR_RETRY = (1<<11),
951 IB_QP_RQ_PSN = (1<<12),
952 IB_QP_MAX_QP_RD_ATOMIC = (1<<13),
953 IB_QP_ALT_PATH = (1<<14),
954 IB_QP_MIN_RNR_TIMER = (1<<15),
955 IB_QP_SQ_PSN = (1<<16),
956 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17),
957 IB_QP_PATH_MIG_STATE = (1<<18),
959 IB_QP_DEST_QPN = (1<<20),
960 IB_QP_SMAC = (1<<21),
961 IB_QP_ALT_SMAC = (1<<22),
963 IB_QP_ALT_VID = (1<<24),
988 enum ib_qp_state qp_state;
989 enum ib_qp_state cur_qp_state;
990 enum ib_mtu path_mtu;
991 enum ib_mig_state path_mig_state;
997 struct ib_qp_cap cap;
998 struct ib_ah_attr ah_attr;
999 struct ib_ah_attr alt_ah_attr;
1002 u8 en_sqd_async_notify;
1005 u8 max_dest_rd_atomic;
1014 u8 alt_smac[ETH_ALEN];
1021 IB_WR_RDMA_WRITE_WITH_IMM,
1023 IB_WR_SEND_WITH_IMM,
1025 IB_WR_ATOMIC_CMP_AND_SWP,
1026 IB_WR_ATOMIC_FETCH_AND_ADD,
1028 IB_WR_SEND_WITH_INV,
1029 IB_WR_RDMA_READ_WITH_INV,
1032 IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1033 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1036 /* reserve values for low level drivers' internal use.
1037 * These values will not be used at all in the ib core layer.
1039 IB_WR_RESERVED1 = 0xf0,
1051 enum ib_send_flags {
1053 IB_SEND_SIGNALED = (1<<1),
1054 IB_SEND_SOLICITED = (1<<2),
1055 IB_SEND_INLINE = (1<<3),
1056 IB_SEND_IP_CSUM = (1<<4),
1058 /* reserve bits 26-31 for low level drivers' internal use */
1059 IB_SEND_RESERVED_START = (1 << 26),
1060 IB_SEND_RESERVED_END = (1 << 31),
1069 struct ib_fast_reg_page_list {
1070 struct ib_device *device;
1072 unsigned int max_page_list_len;
1076 * struct ib_mw_bind_info - Parameters for a memory window bind operation.
1077 * @mr: A memory region to bind the memory window to.
1078 * @addr: The address where the memory window should begin.
1079 * @length: The length of the memory window, in bytes.
1080 * @mw_access_flags: Access flags from enum ib_access_flags for the window.
1082 * This struct contains the shared parameters for type 1 and type 2
1083 * memory window bind operations.
1085 struct ib_mw_bind_info {
1089 int mw_access_flags;
1093 struct ib_send_wr *next;
1095 struct ib_sge *sg_list;
1097 enum ib_wr_opcode opcode;
1101 u32 invalidate_rkey;
1112 u64 compare_add_mask;
1123 u16 pkey_index; /* valid for GSI only */
1124 u8 port_num; /* valid for DR SMPs on switch only */
1128 struct ib_fast_reg_page_list *page_list;
1129 unsigned int page_shift;
1130 unsigned int page_list_len;
1137 /* The new rkey for the memory window. */
1139 struct ib_mw_bind_info bind_info;
1142 struct ib_sig_attrs *sig_attrs;
1143 struct ib_mr *sig_mr;
1145 struct ib_sge *prot;
1148 u32 xrc_remote_srq_num; /* XRC TGT QPs only */
1152 struct ib_recv_wr *next;
1154 struct ib_sge *sg_list;
1158 enum ib_access_flags {
1159 IB_ACCESS_LOCAL_WRITE = 1,
1160 IB_ACCESS_REMOTE_WRITE = (1<<1),
1161 IB_ACCESS_REMOTE_READ = (1<<2),
1162 IB_ACCESS_REMOTE_ATOMIC = (1<<3),
1163 IB_ACCESS_MW_BIND = (1<<4),
1164 IB_ZERO_BASED = (1<<5),
1165 IB_ACCESS_ON_DEMAND = (1<<6),
1168 struct ib_phys_buf {
1175 u64 device_virt_addr;
1177 int mr_access_flags;
1182 enum ib_mr_rereg_flags {
1183 IB_MR_REREG_TRANS = 1,
1184 IB_MR_REREG_PD = (1<<1),
1185 IB_MR_REREG_ACCESS = (1<<2),
1186 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1)
1190 * struct ib_mw_bind - Parameters for a type 1 memory window bind operation.
1191 * @wr_id: Work request id.
1192 * @send_flags: Flags from ib_send_flags enum.
1193 * @bind_info: More parameters of the bind operation.
1198 struct ib_mw_bind_info bind_info;
1201 struct ib_fmr_attr {
1209 struct ib_ucontext {
1210 struct ib_device *device;
1211 struct list_head pd_list;
1212 struct list_head mr_list;
1213 struct list_head mw_list;
1214 struct list_head cq_list;
1215 struct list_head qp_list;
1216 struct list_head srq_list;
1217 struct list_head ah_list;
1218 struct list_head xrcd_list;
1219 struct list_head rule_list;
1223 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1224 struct rb_root umem_tree;
1226 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1227 * mmu notifiers registration.
1229 struct rw_semaphore umem_rwsem;
1230 void (*invalidate_range)(struct ib_umem *umem,
1231 unsigned long start, unsigned long end);
1233 struct mmu_notifier mn;
1234 atomic_t notifier_count;
1235 /* A list of umems that don't have private mmu notifier counters yet. */
1236 struct list_head no_private_counters;
1242 u64 user_handle; /* handle given to us by userspace */
1243 struct ib_ucontext *context; /* associated user context */
1244 void *object; /* containing object */
1245 struct list_head list; /* link to context's list */
1246 int id; /* index into kernel idr */
1248 struct rw_semaphore mutex; /* protects .live */
1253 const void __user *inbuf;
1254 void __user *outbuf;
1260 struct ib_device *device;
1261 struct ib_uobject *uobject;
1262 atomic_t usecnt; /* count all resources */
1266 struct ib_device *device;
1267 atomic_t usecnt; /* count all exposed resources */
1268 struct inode *inode;
1270 struct mutex tgt_qp_mutex;
1271 struct list_head tgt_qp_list;
1275 struct ib_device *device;
1277 struct ib_uobject *uobject;
1280 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1283 struct ib_device *device;
1284 struct ib_uobject *uobject;
1285 ib_comp_handler comp_handler;
1286 void (*event_handler)(struct ib_event *, void *);
1289 atomic_t usecnt; /* count number of work queues */
1293 struct ib_device *device;
1295 struct ib_uobject *uobject;
1296 void (*event_handler)(struct ib_event *, void *);
1298 enum ib_srq_type srq_type;
1303 struct ib_xrcd *xrcd;
1311 struct ib_device *device;
1313 struct ib_cq *send_cq;
1314 struct ib_cq *recv_cq;
1316 struct ib_xrcd *xrcd; /* XRC TGT QPs only */
1317 struct list_head xrcd_list;
1318 /* count times opened, mcast attaches, flow attaches */
1320 struct list_head open_list;
1321 struct ib_qp *real_qp;
1322 struct ib_uobject *uobject;
1323 void (*event_handler)(struct ib_event *, void *);
1326 enum ib_qp_type qp_type;
1330 struct ib_device *device;
1332 struct ib_uobject *uobject;
1335 atomic_t usecnt; /* count number of MWs */
1339 struct ib_device *device;
1341 struct ib_uobject *uobject;
1343 enum ib_mw_type type;
1347 struct ib_device *device;
1349 struct list_head list;
1354 /* Supported steering options */
1355 enum ib_flow_attr_type {
1356 /* steering according to rule specifications */
1357 IB_FLOW_ATTR_NORMAL = 0x0,
1358 /* default unicast and multicast rule -
1359 * receive all Eth traffic which isn't steered to any QP
1361 IB_FLOW_ATTR_ALL_DEFAULT = 0x1,
1362 /* default multicast rule -
1363 * receive all Eth multicast traffic which isn't steered to any QP
1365 IB_FLOW_ATTR_MC_DEFAULT = 0x2,
1366 /* sniffer rule - receive all port traffic */
1367 IB_FLOW_ATTR_SNIFFER = 0x3
1370 /* Supported steering header types */
1371 enum ib_flow_spec_type {
1373 IB_FLOW_SPEC_ETH = 0x20,
1374 IB_FLOW_SPEC_IB = 0x22,
1376 IB_FLOW_SPEC_IPV4 = 0x30,
1378 IB_FLOW_SPEC_TCP = 0x40,
1379 IB_FLOW_SPEC_UDP = 0x41
1381 #define IB_FLOW_SPEC_LAYER_MASK 0xF0
1382 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1384 /* Flow steering rule priority is set according to it's domain.
1385 * Lower domain value means higher priority.
1387 enum ib_flow_domain {
1388 IB_FLOW_DOMAIN_USER,
1389 IB_FLOW_DOMAIN_ETHTOOL,
1392 IB_FLOW_DOMAIN_NUM /* Must be last */
1395 struct ib_flow_eth_filter {
1402 struct ib_flow_spec_eth {
1403 enum ib_flow_spec_type type;
1405 struct ib_flow_eth_filter val;
1406 struct ib_flow_eth_filter mask;
1409 struct ib_flow_ib_filter {
1414 struct ib_flow_spec_ib {
1415 enum ib_flow_spec_type type;
1417 struct ib_flow_ib_filter val;
1418 struct ib_flow_ib_filter mask;
1421 struct ib_flow_ipv4_filter {
1426 struct ib_flow_spec_ipv4 {
1427 enum ib_flow_spec_type type;
1429 struct ib_flow_ipv4_filter val;
1430 struct ib_flow_ipv4_filter mask;
1433 struct ib_flow_tcp_udp_filter {
1438 struct ib_flow_spec_tcp_udp {
1439 enum ib_flow_spec_type type;
1441 struct ib_flow_tcp_udp_filter val;
1442 struct ib_flow_tcp_udp_filter mask;
1445 union ib_flow_spec {
1447 enum ib_flow_spec_type type;
1450 struct ib_flow_spec_eth eth;
1451 struct ib_flow_spec_ib ib;
1452 struct ib_flow_spec_ipv4 ipv4;
1453 struct ib_flow_spec_tcp_udp tcp_udp;
1456 struct ib_flow_attr {
1457 enum ib_flow_attr_type type;
1463 /* Following are the optional layers according to user request
1464 * struct ib_flow_spec_xxx
1465 * struct ib_flow_spec_yyy
1471 struct ib_uobject *uobject;
1477 enum ib_process_mad_flags {
1478 IB_MAD_IGNORE_MKEY = 1,
1479 IB_MAD_IGNORE_BKEY = 2,
1480 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1483 enum ib_mad_result {
1484 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */
1485 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */
1486 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */
1487 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */
1490 #define IB_DEVICE_NAME_MAX 64
1494 struct ib_event_handler event_handler;
1495 struct ib_pkey_cache **pkey_cache;
1496 struct ib_gid_table **gid_cache;
1500 struct ib_dma_mapping_ops {
1501 int (*mapping_error)(struct ib_device *dev,
1503 u64 (*map_single)(struct ib_device *dev,
1504 void *ptr, size_t size,
1505 enum dma_data_direction direction);
1506 void (*unmap_single)(struct ib_device *dev,
1507 u64 addr, size_t size,
1508 enum dma_data_direction direction);
1509 u64 (*map_page)(struct ib_device *dev,
1510 struct page *page, unsigned long offset,
1512 enum dma_data_direction direction);
1513 void (*unmap_page)(struct ib_device *dev,
1514 u64 addr, size_t size,
1515 enum dma_data_direction direction);
1516 int (*map_sg)(struct ib_device *dev,
1517 struct scatterlist *sg, int nents,
1518 enum dma_data_direction direction);
1519 void (*unmap_sg)(struct ib_device *dev,
1520 struct scatterlist *sg, int nents,
1521 enum dma_data_direction direction);
1522 void (*sync_single_for_cpu)(struct ib_device *dev,
1525 enum dma_data_direction dir);
1526 void (*sync_single_for_device)(struct ib_device *dev,
1529 enum dma_data_direction dir);
1530 void *(*alloc_coherent)(struct ib_device *dev,
1534 void (*free_coherent)(struct ib_device *dev,
1535 size_t size, void *cpu_addr,
1541 struct ib_port_immutable {
1549 struct device *dma_device;
1551 char name[IB_DEVICE_NAME_MAX];
1553 struct list_head event_handler_list;
1554 spinlock_t event_handler_lock;
1556 spinlock_t client_data_lock;
1557 struct list_head core_list;
1558 /* Access to the client_data_list is protected by the client_data_lock
1559 * spinlock and the lists_rwsem read-write semaphore */
1560 struct list_head client_data_list;
1562 struct ib_cache cache;
1564 * port_immutable is indexed by port number
1566 struct ib_port_immutable *port_immutable;
1568 int num_comp_vectors;
1570 struct iw_cm_verbs *iwcm;
1572 int (*get_protocol_stats)(struct ib_device *device,
1573 union rdma_protocol_stats *stats);
1574 int (*query_device)(struct ib_device *device,
1575 struct ib_device_attr *device_attr,
1576 struct ib_udata *udata);
1577 int (*query_port)(struct ib_device *device,
1579 struct ib_port_attr *port_attr);
1580 enum rdma_link_layer (*get_link_layer)(struct ib_device *device,
1582 /* When calling get_netdev, the HW vendor's driver should return the
1583 * net device of device @device at port @port_num or NULL if such
1584 * a net device doesn't exist. The vendor driver should call dev_hold
1585 * on this net device. The HW vendor's device driver must guarantee
1586 * that this function returns NULL before the net device reaches
1587 * NETDEV_UNREGISTER_FINAL state.
1589 struct net_device *(*get_netdev)(struct ib_device *device,
1591 int (*query_gid)(struct ib_device *device,
1592 u8 port_num, int index,
1594 /* When calling add_gid, the HW vendor's driver should
1595 * add the gid of device @device at gid index @index of
1596 * port @port_num to be @gid. Meta-info of that gid (for example,
1597 * the network device related to this gid is available
1598 * at @attr. @context allows the HW vendor driver to store extra
1599 * information together with a GID entry. The HW vendor may allocate
1600 * memory to contain this information and store it in @context when a
1601 * new GID entry is written to. Params are consistent until the next
1602 * call of add_gid or delete_gid. The function should return 0 on
1603 * success or error otherwise. The function could be called
1604 * concurrently for different ports. This function is only called
1605 * when roce_gid_table is used.
1607 int (*add_gid)(struct ib_device *device,
1610 const union ib_gid *gid,
1611 const struct ib_gid_attr *attr,
1613 /* When calling del_gid, the HW vendor's driver should delete the
1614 * gid of device @device at gid index @index of port @port_num.
1615 * Upon the deletion of a GID entry, the HW vendor must free any
1616 * allocated memory. The caller will clear @context afterwards.
1617 * This function is only called when roce_gid_table is used.
1619 int (*del_gid)(struct ib_device *device,
1623 int (*query_pkey)(struct ib_device *device,
1624 u8 port_num, u16 index, u16 *pkey);
1625 int (*modify_device)(struct ib_device *device,
1626 int device_modify_mask,
1627 struct ib_device_modify *device_modify);
1628 int (*modify_port)(struct ib_device *device,
1629 u8 port_num, int port_modify_mask,
1630 struct ib_port_modify *port_modify);
1631 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device,
1632 struct ib_udata *udata);
1633 int (*dealloc_ucontext)(struct ib_ucontext *context);
1634 int (*mmap)(struct ib_ucontext *context,
1635 struct vm_area_struct *vma);
1636 struct ib_pd * (*alloc_pd)(struct ib_device *device,
1637 struct ib_ucontext *context,
1638 struct ib_udata *udata);
1639 int (*dealloc_pd)(struct ib_pd *pd);
1640 struct ib_ah * (*create_ah)(struct ib_pd *pd,
1641 struct ib_ah_attr *ah_attr);
1642 int (*modify_ah)(struct ib_ah *ah,
1643 struct ib_ah_attr *ah_attr);
1644 int (*query_ah)(struct ib_ah *ah,
1645 struct ib_ah_attr *ah_attr);
1646 int (*destroy_ah)(struct ib_ah *ah);
1647 struct ib_srq * (*create_srq)(struct ib_pd *pd,
1648 struct ib_srq_init_attr *srq_init_attr,
1649 struct ib_udata *udata);
1650 int (*modify_srq)(struct ib_srq *srq,
1651 struct ib_srq_attr *srq_attr,
1652 enum ib_srq_attr_mask srq_attr_mask,
1653 struct ib_udata *udata);
1654 int (*query_srq)(struct ib_srq *srq,
1655 struct ib_srq_attr *srq_attr);
1656 int (*destroy_srq)(struct ib_srq *srq);
1657 int (*post_srq_recv)(struct ib_srq *srq,
1658 struct ib_recv_wr *recv_wr,
1659 struct ib_recv_wr **bad_recv_wr);
1660 struct ib_qp * (*create_qp)(struct ib_pd *pd,
1661 struct ib_qp_init_attr *qp_init_attr,
1662 struct ib_udata *udata);
1663 int (*modify_qp)(struct ib_qp *qp,
1664 struct ib_qp_attr *qp_attr,
1666 struct ib_udata *udata);
1667 int (*query_qp)(struct ib_qp *qp,
1668 struct ib_qp_attr *qp_attr,
1670 struct ib_qp_init_attr *qp_init_attr);
1671 int (*destroy_qp)(struct ib_qp *qp);
1672 int (*post_send)(struct ib_qp *qp,
1673 struct ib_send_wr *send_wr,
1674 struct ib_send_wr **bad_send_wr);
1675 int (*post_recv)(struct ib_qp *qp,
1676 struct ib_recv_wr *recv_wr,
1677 struct ib_recv_wr **bad_recv_wr);
1678 struct ib_cq * (*create_cq)(struct ib_device *device,
1679 const struct ib_cq_init_attr *attr,
1680 struct ib_ucontext *context,
1681 struct ib_udata *udata);
1682 int (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1684 int (*destroy_cq)(struct ib_cq *cq);
1685 int (*resize_cq)(struct ib_cq *cq, int cqe,
1686 struct ib_udata *udata);
1687 int (*poll_cq)(struct ib_cq *cq, int num_entries,
1689 int (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1690 int (*req_notify_cq)(struct ib_cq *cq,
1691 enum ib_cq_notify_flags flags);
1692 int (*req_ncomp_notif)(struct ib_cq *cq,
1694 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd,
1695 int mr_access_flags);
1696 struct ib_mr * (*reg_phys_mr)(struct ib_pd *pd,
1697 struct ib_phys_buf *phys_buf_array,
1699 int mr_access_flags,
1701 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd,
1702 u64 start, u64 length,
1704 int mr_access_flags,
1705 struct ib_udata *udata);
1706 int (*rereg_user_mr)(struct ib_mr *mr,
1708 u64 start, u64 length,
1710 int mr_access_flags,
1712 struct ib_udata *udata);
1713 int (*query_mr)(struct ib_mr *mr,
1714 struct ib_mr_attr *mr_attr);
1715 int (*dereg_mr)(struct ib_mr *mr);
1716 struct ib_mr * (*alloc_mr)(struct ib_pd *pd,
1717 enum ib_mr_type mr_type,
1719 struct ib_fast_reg_page_list * (*alloc_fast_reg_page_list)(struct ib_device *device,
1721 void (*free_fast_reg_page_list)(struct ib_fast_reg_page_list *page_list);
1722 int (*rereg_phys_mr)(struct ib_mr *mr,
1725 struct ib_phys_buf *phys_buf_array,
1727 int mr_access_flags,
1729 struct ib_mw * (*alloc_mw)(struct ib_pd *pd,
1730 enum ib_mw_type type);
1731 int (*bind_mw)(struct ib_qp *qp,
1733 struct ib_mw_bind *mw_bind);
1734 int (*dealloc_mw)(struct ib_mw *mw);
1735 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd,
1736 int mr_access_flags,
1737 struct ib_fmr_attr *fmr_attr);
1738 int (*map_phys_fmr)(struct ib_fmr *fmr,
1739 u64 *page_list, int list_len,
1741 int (*unmap_fmr)(struct list_head *fmr_list);
1742 int (*dealloc_fmr)(struct ib_fmr *fmr);
1743 int (*attach_mcast)(struct ib_qp *qp,
1746 int (*detach_mcast)(struct ib_qp *qp,
1749 int (*process_mad)(struct ib_device *device,
1750 int process_mad_flags,
1752 const struct ib_wc *in_wc,
1753 const struct ib_grh *in_grh,
1754 const struct ib_mad_hdr *in_mad,
1756 struct ib_mad_hdr *out_mad,
1757 size_t *out_mad_size,
1758 u16 *out_mad_pkey_index);
1759 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device,
1760 struct ib_ucontext *ucontext,
1761 struct ib_udata *udata);
1762 int (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1763 struct ib_flow * (*create_flow)(struct ib_qp *qp,
1767 int (*destroy_flow)(struct ib_flow *flow_id);
1768 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1769 struct ib_mr_status *mr_status);
1771 struct ib_dma_mapping_ops *dma_ops;
1773 struct module *owner;
1775 struct kobject *ports_parent;
1776 struct list_head port_list;
1779 IB_DEV_UNINITIALIZED,
1785 u64 uverbs_cmd_mask;
1786 u64 uverbs_ex_cmd_mask;
1796 * The following mandatory functions are used only at device
1797 * registration. Keep functions such as these at the end of this
1798 * structure to avoid cache line misses when accessing struct ib_device
1801 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1806 void (*add) (struct ib_device *);
1807 void (*remove)(struct ib_device *, void *client_data);
1809 /* Returns the net_dev belonging to this ib_client and matching the
1811 * @dev: An RDMA device that the net_dev use for communication.
1812 * @port: A physical port number on the RDMA device.
1813 * @pkey: P_Key that the net_dev uses if applicable.
1814 * @gid: A GID that the net_dev uses to communicate.
1815 * @addr: An IP address the net_dev is configured with.
1816 * @client_data: The device's client data set by ib_set_client_data().
1818 * An ib_client that implements a net_dev on top of RDMA devices
1819 * (such as IP over IB) should implement this callback, allowing the
1820 * rdma_cm module to find the right net_dev for a given request.
1822 * The caller is responsible for calling dev_put on the returned
1824 struct net_device *(*get_net_dev_by_params)(
1825 struct ib_device *dev,
1828 const union ib_gid *gid,
1829 const struct sockaddr *addr,
1831 struct list_head list;
1834 struct ib_device *ib_alloc_device(size_t size);
1835 void ib_dealloc_device(struct ib_device *device);
1837 int ib_register_device(struct ib_device *device,
1838 int (*port_callback)(struct ib_device *,
1839 u8, struct kobject *));
1840 void ib_unregister_device(struct ib_device *device);
1842 int ib_register_client (struct ib_client *client);
1843 void ib_unregister_client(struct ib_client *client);
1845 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1846 void ib_set_client_data(struct ib_device *device, struct ib_client *client,
1849 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1851 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1854 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1856 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1860 * ib_modify_qp_is_ok - Check that the supplied attribute mask
1861 * contains all required attributes and no attributes not allowed for
1862 * the given QP state transition.
1863 * @cur_state: Current QP state
1864 * @next_state: Next QP state
1866 * @mask: Mask of supplied QP attributes
1867 * @ll : link layer of port
1869 * This function is a helper function that a low-level driver's
1870 * modify_qp method can use to validate the consumer's input. It
1871 * checks that cur_state and next_state are valid QP states, that a
1872 * transition from cur_state to next_state is allowed by the IB spec,
1873 * and that the attribute mask supplied is allowed for the transition.
1875 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1876 enum ib_qp_type type, enum ib_qp_attr_mask mask,
1877 enum rdma_link_layer ll);
1879 int ib_register_event_handler (struct ib_event_handler *event_handler);
1880 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1881 void ib_dispatch_event(struct ib_event *event);
1883 int ib_query_device(struct ib_device *device,
1884 struct ib_device_attr *device_attr);
1886 int ib_query_port(struct ib_device *device,
1887 u8 port_num, struct ib_port_attr *port_attr);
1889 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1893 * rdma_cap_ib_switch - Check if the device is IB switch
1894 * @device: Device to check
1896 * Device driver is responsible for setting is_switch bit on
1897 * in ib_device structure at init time.
1899 * Return: true if the device is IB switch.
1901 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
1903 return device->is_switch;
1907 * rdma_start_port - Return the first valid port number for the device
1910 * @device: Device to be checked
1912 * Return start port number
1914 static inline u8 rdma_start_port(const struct ib_device *device)
1916 return rdma_cap_ib_switch(device) ? 0 : 1;
1920 * rdma_end_port - Return the last valid port number for the device
1923 * @device: Device to be checked
1925 * Return last port number
1927 static inline u8 rdma_end_port(const struct ib_device *device)
1929 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
1932 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
1934 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
1937 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
1939 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
1942 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
1944 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
1947 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
1949 return device->port_immutable[port_num].core_cap_flags &
1950 (RDMA_CORE_CAP_PROT_IB | RDMA_CORE_CAP_PROT_ROCE);
1954 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
1955 * Management Datagrams.
1956 * @device: Device to check
1957 * @port_num: Port number to check
1959 * Management Datagrams (MAD) are a required part of the InfiniBand
1960 * specification and are supported on all InfiniBand devices. A slightly
1961 * extended version are also supported on OPA interfaces.
1963 * Return: true if the port supports sending/receiving of MAD packets.
1965 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
1967 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
1971 * rdma_cap_opa_mad - Check if the port of device provides support for OPA
1972 * Management Datagrams.
1973 * @device: Device to check
1974 * @port_num: Port number to check
1976 * Intel OmniPath devices extend and/or replace the InfiniBand Management
1977 * datagrams with their own versions. These OPA MADs share many but not all of
1978 * the characteristics of InfiniBand MADs.
1980 * OPA MADs differ in the following ways:
1982 * 1) MADs are variable size up to 2K
1983 * IBTA defined MADs remain fixed at 256 bytes
1984 * 2) OPA SMPs must carry valid PKeys
1985 * 3) OPA SMP packets are a different format
1987 * Return: true if the port supports OPA MAD packet formats.
1989 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
1991 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
1992 == RDMA_CORE_CAP_OPA_MAD;
1996 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
1997 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
1998 * @device: Device to check
1999 * @port_num: Port number to check
2001 * Each InfiniBand node is required to provide a Subnet Management Agent
2002 * that the subnet manager can access. Prior to the fabric being fully
2003 * configured by the subnet manager, the SMA is accessed via a well known
2004 * interface called the Subnet Management Interface (SMI). This interface
2005 * uses directed route packets to communicate with the SM to get around the
2006 * chicken and egg problem of the SM needing to know what's on the fabric
2007 * in order to configure the fabric, and needing to configure the fabric in
2008 * order to send packets to the devices on the fabric. These directed
2009 * route packets do not need the fabric fully configured in order to reach
2010 * their destination. The SMI is the only method allowed to send
2011 * directed route packets on an InfiniBand fabric.
2013 * Return: true if the port provides an SMI.
2015 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2017 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2021 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2022 * Communication Manager.
2023 * @device: Device to check
2024 * @port_num: Port number to check
2026 * The InfiniBand Communication Manager is one of many pre-defined General
2027 * Service Agents (GSA) that are accessed via the General Service
2028 * Interface (GSI). It's role is to facilitate establishment of connections
2029 * between nodes as well as other management related tasks for established
2032 * Return: true if the port supports an IB CM (this does not guarantee that
2033 * a CM is actually running however).
2035 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2037 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2041 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2042 * Communication Manager.
2043 * @device: Device to check
2044 * @port_num: Port number to check
2046 * Similar to above, but specific to iWARP connections which have a different
2047 * managment protocol than InfiniBand.
2049 * Return: true if the port supports an iWARP CM (this does not guarantee that
2050 * a CM is actually running however).
2052 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2054 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2058 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2059 * Subnet Administration.
2060 * @device: Device to check
2061 * @port_num: Port number to check
2063 * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2064 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand
2065 * fabrics, devices should resolve routes to other hosts by contacting the
2066 * SA to query the proper route.
2068 * Return: true if the port should act as a client to the fabric Subnet
2069 * Administration interface. This does not imply that the SA service is
2072 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2074 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2078 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2080 * @device: Device to check
2081 * @port_num: Port number to check
2083 * InfiniBand multicast registration is more complex than normal IPv4 or
2084 * IPv6 multicast registration. Each Host Channel Adapter must register
2085 * with the Subnet Manager when it wishes to join a multicast group. It
2086 * should do so only once regardless of how many queue pairs it subscribes
2087 * to this group. And it should leave the group only after all queue pairs
2088 * attached to the group have been detached.
2090 * Return: true if the port must undertake the additional adminstrative
2091 * overhead of registering/unregistering with the SM and tracking of the
2092 * total number of queue pairs attached to the multicast group.
2094 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2096 return rdma_cap_ib_sa(device, port_num);
2100 * rdma_cap_af_ib - Check if the port of device has the capability
2101 * Native Infiniband Address.
2102 * @device: Device to check
2103 * @port_num: Port number to check
2105 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2106 * GID. RoCE uses a different mechanism, but still generates a GID via
2107 * a prescribed mechanism and port specific data.
2109 * Return: true if the port uses a GID address to identify devices on the
2112 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2114 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2118 * rdma_cap_eth_ah - Check if the port of device has the capability
2119 * Ethernet Address Handle.
2120 * @device: Device to check
2121 * @port_num: Port number to check
2123 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2124 * to fabricate GIDs over Ethernet/IP specific addresses native to the
2125 * port. Normally, packet headers are generated by the sending host
2126 * adapter, but when sending connectionless datagrams, we must manually
2127 * inject the proper headers for the fabric we are communicating over.
2129 * Return: true if we are running as a RoCE port and must force the
2130 * addition of a Global Route Header built from our Ethernet Address
2131 * Handle into our header list for connectionless packets.
2133 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2135 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2139 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2142 * @port_num: Port number
2144 * This MAD size includes the MAD headers and MAD payload. No other headers
2147 * Return the max MAD size required by the Port. Will return 0 if the port
2148 * does not support MADs
2150 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2152 return device->port_immutable[port_num].max_mad_size;
2156 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2157 * @device: Device to check
2158 * @port_num: Port number to check
2160 * RoCE GID table mechanism manages the various GIDs for a device.
2162 * NOTE: if allocating the port's GID table has failed, this call will still
2163 * return true, but any RoCE GID table API will fail.
2165 * Return: true if the port uses RoCE GID table mechanism in order to manage
2168 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2171 return rdma_protocol_roce(device, port_num) &&
2172 device->add_gid && device->del_gid;
2175 int ib_query_gid(struct ib_device *device,
2176 u8 port_num, int index, union ib_gid *gid);
2178 int ib_query_pkey(struct ib_device *device,
2179 u8 port_num, u16 index, u16 *pkey);
2181 int ib_modify_device(struct ib_device *device,
2182 int device_modify_mask,
2183 struct ib_device_modify *device_modify);
2185 int ib_modify_port(struct ib_device *device,
2186 u8 port_num, int port_modify_mask,
2187 struct ib_port_modify *port_modify);
2189 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2190 u8 *port_num, u16 *index);
2192 int ib_find_pkey(struct ib_device *device,
2193 u8 port_num, u16 pkey, u16 *index);
2196 * ib_alloc_pd - Allocates an unused protection domain.
2197 * @device: The device on which to allocate the protection domain.
2199 * A protection domain object provides an association between QPs, shared
2200 * receive queues, address handles, memory regions, and memory windows.
2202 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2205 * ib_dealloc_pd - Deallocates a protection domain.
2206 * @pd: The protection domain to deallocate.
2208 int ib_dealloc_pd(struct ib_pd *pd);
2211 * ib_create_ah - Creates an address handle for the given address vector.
2212 * @pd: The protection domain associated with the address handle.
2213 * @ah_attr: The attributes of the address vector.
2215 * The address handle is used to reference a local or global destination
2216 * in all UD QP post sends.
2218 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2221 * ib_init_ah_from_wc - Initializes address handle attributes from a
2223 * @device: Device on which the received message arrived.
2224 * @port_num: Port on which the received message arrived.
2225 * @wc: Work completion associated with the received message.
2226 * @grh: References the received global route header. This parameter is
2227 * ignored unless the work completion indicates that the GRH is valid.
2228 * @ah_attr: Returned attributes that can be used when creating an address
2229 * handle for replying to the message.
2231 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2232 const struct ib_wc *wc, const struct ib_grh *grh,
2233 struct ib_ah_attr *ah_attr);
2236 * ib_create_ah_from_wc - Creates an address handle associated with the
2237 * sender of the specified work completion.
2238 * @pd: The protection domain associated with the address handle.
2239 * @wc: Work completion information associated with a received message.
2240 * @grh: References the received global route header. This parameter is
2241 * ignored unless the work completion indicates that the GRH is valid.
2242 * @port_num: The outbound port number to associate with the address.
2244 * The address handle is used to reference a local or global destination
2245 * in all UD QP post sends.
2247 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2248 const struct ib_grh *grh, u8 port_num);
2251 * ib_modify_ah - Modifies the address vector associated with an address
2253 * @ah: The address handle to modify.
2254 * @ah_attr: The new address vector attributes to associate with the
2257 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2260 * ib_query_ah - Queries the address vector associated with an address
2262 * @ah: The address handle to query.
2263 * @ah_attr: The address vector attributes associated with the address
2266 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2269 * ib_destroy_ah - Destroys an address handle.
2270 * @ah: The address handle to destroy.
2272 int ib_destroy_ah(struct ib_ah *ah);
2275 * ib_create_srq - Creates a SRQ associated with the specified protection
2277 * @pd: The protection domain associated with the SRQ.
2278 * @srq_init_attr: A list of initial attributes required to create the
2279 * SRQ. If SRQ creation succeeds, then the attributes are updated to
2280 * the actual capabilities of the created SRQ.
2282 * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2283 * requested size of the SRQ, and set to the actual values allocated
2284 * on return. If ib_create_srq() succeeds, then max_wr and max_sge
2285 * will always be at least as large as the requested values.
2287 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2288 struct ib_srq_init_attr *srq_init_attr);
2291 * ib_modify_srq - Modifies the attributes for the specified SRQ.
2292 * @srq: The SRQ to modify.
2293 * @srq_attr: On input, specifies the SRQ attributes to modify. On output,
2294 * the current values of selected SRQ attributes are returned.
2295 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2296 * are being modified.
2298 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2299 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2300 * the number of receives queued drops below the limit.
2302 int ib_modify_srq(struct ib_srq *srq,
2303 struct ib_srq_attr *srq_attr,
2304 enum ib_srq_attr_mask srq_attr_mask);
2307 * ib_query_srq - Returns the attribute list and current values for the
2309 * @srq: The SRQ to query.
2310 * @srq_attr: The attributes of the specified SRQ.
2312 int ib_query_srq(struct ib_srq *srq,
2313 struct ib_srq_attr *srq_attr);
2316 * ib_destroy_srq - Destroys the specified SRQ.
2317 * @srq: The SRQ to destroy.
2319 int ib_destroy_srq(struct ib_srq *srq);
2322 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2323 * @srq: The SRQ to post the work request on.
2324 * @recv_wr: A list of work requests to post on the receive queue.
2325 * @bad_recv_wr: On an immediate failure, this parameter will reference
2326 * the work request that failed to be posted on the QP.
2328 static inline int ib_post_srq_recv(struct ib_srq *srq,
2329 struct ib_recv_wr *recv_wr,
2330 struct ib_recv_wr **bad_recv_wr)
2332 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2336 * ib_create_qp - Creates a QP associated with the specified protection
2338 * @pd: The protection domain associated with the QP.
2339 * @qp_init_attr: A list of initial attributes required to create the
2340 * QP. If QP creation succeeds, then the attributes are updated to
2341 * the actual capabilities of the created QP.
2343 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2344 struct ib_qp_init_attr *qp_init_attr);
2347 * ib_modify_qp - Modifies the attributes for the specified QP and then
2348 * transitions the QP to the given state.
2349 * @qp: The QP to modify.
2350 * @qp_attr: On input, specifies the QP attributes to modify. On output,
2351 * the current values of selected QP attributes are returned.
2352 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2353 * are being modified.
2355 int ib_modify_qp(struct ib_qp *qp,
2356 struct ib_qp_attr *qp_attr,
2360 * ib_query_qp - Returns the attribute list and current values for the
2362 * @qp: The QP to query.
2363 * @qp_attr: The attributes of the specified QP.
2364 * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2365 * @qp_init_attr: Additional attributes of the selected QP.
2367 * The qp_attr_mask may be used to limit the query to gathering only the
2368 * selected attributes.
2370 int ib_query_qp(struct ib_qp *qp,
2371 struct ib_qp_attr *qp_attr,
2373 struct ib_qp_init_attr *qp_init_attr);
2376 * ib_destroy_qp - Destroys the specified QP.
2377 * @qp: The QP to destroy.
2379 int ib_destroy_qp(struct ib_qp *qp);
2382 * ib_open_qp - Obtain a reference to an existing sharable QP.
2383 * @xrcd - XRC domain
2384 * @qp_open_attr: Attributes identifying the QP to open.
2386 * Returns a reference to a sharable QP.
2388 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2389 struct ib_qp_open_attr *qp_open_attr);
2392 * ib_close_qp - Release an external reference to a QP.
2393 * @qp: The QP handle to release
2395 * The opened QP handle is released by the caller. The underlying
2396 * shared QP is not destroyed until all internal references are released.
2398 int ib_close_qp(struct ib_qp *qp);
2401 * ib_post_send - Posts a list of work requests to the send queue of
2403 * @qp: The QP to post the work request on.
2404 * @send_wr: A list of work requests to post on the send queue.
2405 * @bad_send_wr: On an immediate failure, this parameter will reference
2406 * the work request that failed to be posted on the QP.
2408 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2409 * error is returned, the QP state shall not be affected,
2410 * ib_post_send() will return an immediate error after queueing any
2411 * earlier work requests in the list.
2413 static inline int ib_post_send(struct ib_qp *qp,
2414 struct ib_send_wr *send_wr,
2415 struct ib_send_wr **bad_send_wr)
2417 return qp->device->post_send(qp, send_wr, bad_send_wr);
2421 * ib_post_recv - Posts a list of work requests to the receive queue of
2423 * @qp: The QP to post the work request on.
2424 * @recv_wr: A list of work requests to post on the receive queue.
2425 * @bad_recv_wr: On an immediate failure, this parameter will reference
2426 * the work request that failed to be posted on the QP.
2428 static inline int ib_post_recv(struct ib_qp *qp,
2429 struct ib_recv_wr *recv_wr,
2430 struct ib_recv_wr **bad_recv_wr)
2432 return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2436 * ib_create_cq - Creates a CQ on the specified device.
2437 * @device: The device on which to create the CQ.
2438 * @comp_handler: A user-specified callback that is invoked when a
2439 * completion event occurs on the CQ.
2440 * @event_handler: A user-specified callback that is invoked when an
2441 * asynchronous event not associated with a completion occurs on the CQ.
2442 * @cq_context: Context associated with the CQ returned to the user via
2443 * the associated completion and event handlers.
2444 * @cq_attr: The attributes the CQ should be created upon.
2446 * Users can examine the cq structure to determine the actual CQ size.
2448 struct ib_cq *ib_create_cq(struct ib_device *device,
2449 ib_comp_handler comp_handler,
2450 void (*event_handler)(struct ib_event *, void *),
2452 const struct ib_cq_init_attr *cq_attr);
2455 * ib_resize_cq - Modifies the capacity of the CQ.
2456 * @cq: The CQ to resize.
2457 * @cqe: The minimum size of the CQ.
2459 * Users can examine the cq structure to determine the actual CQ size.
2461 int ib_resize_cq(struct ib_cq *cq, int cqe);
2464 * ib_modify_cq - Modifies moderation params of the CQ
2465 * @cq: The CQ to modify.
2466 * @cq_count: number of CQEs that will trigger an event
2467 * @cq_period: max period of time in usec before triggering an event
2470 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2473 * ib_destroy_cq - Destroys the specified CQ.
2474 * @cq: The CQ to destroy.
2476 int ib_destroy_cq(struct ib_cq *cq);
2479 * ib_poll_cq - poll a CQ for completion(s)
2480 * @cq:the CQ being polled
2481 * @num_entries:maximum number of completions to return
2482 * @wc:array of at least @num_entries &struct ib_wc where completions
2485 * Poll a CQ for (possibly multiple) completions. If the return value
2486 * is < 0, an error occurred. If the return value is >= 0, it is the
2487 * number of completions returned. If the return value is
2488 * non-negative and < num_entries, then the CQ was emptied.
2490 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2493 return cq->device->poll_cq(cq, num_entries, wc);
2497 * ib_peek_cq - Returns the number of unreaped completions currently
2498 * on the specified CQ.
2499 * @cq: The CQ to peek.
2500 * @wc_cnt: A minimum number of unreaped completions to check for.
2502 * If the number of unreaped completions is greater than or equal to wc_cnt,
2503 * this function returns wc_cnt, otherwise, it returns the actual number of
2504 * unreaped completions.
2506 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2509 * ib_req_notify_cq - Request completion notification on a CQ.
2510 * @cq: The CQ to generate an event for.
2512 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2513 * to request an event on the next solicited event or next work
2514 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2515 * may also be |ed in to request a hint about missed events, as
2519 * < 0 means an error occurred while requesting notification
2520 * == 0 means notification was requested successfully, and if
2521 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2522 * were missed and it is safe to wait for another event. In
2523 * this case is it guaranteed that any work completions added
2524 * to the CQ since the last CQ poll will trigger a completion
2525 * notification event.
2526 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2527 * in. It means that the consumer must poll the CQ again to
2528 * make sure it is empty to avoid missing an event because of a
2529 * race between requesting notification and an entry being
2530 * added to the CQ. This return value means it is possible
2531 * (but not guaranteed) that a work completion has been added
2532 * to the CQ since the last poll without triggering a
2533 * completion notification event.
2535 static inline int ib_req_notify_cq(struct ib_cq *cq,
2536 enum ib_cq_notify_flags flags)
2538 return cq->device->req_notify_cq(cq, flags);
2542 * ib_req_ncomp_notif - Request completion notification when there are
2543 * at least the specified number of unreaped completions on the CQ.
2544 * @cq: The CQ to generate an event for.
2545 * @wc_cnt: The number of unreaped completions that should be on the
2546 * CQ before an event is generated.
2548 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2550 return cq->device->req_ncomp_notif ?
2551 cq->device->req_ncomp_notif(cq, wc_cnt) :
2556 * ib_get_dma_mr - Returns a memory region for system memory that is
2558 * @pd: The protection domain associated with the memory region.
2559 * @mr_access_flags: Specifies the memory access rights.
2561 * Note that the ib_dma_*() functions defined below must be used
2562 * to create/destroy addresses used with the Lkey or Rkey returned
2563 * by ib_get_dma_mr().
2565 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2568 * ib_dma_mapping_error - check a DMA addr for error
2569 * @dev: The device for which the dma_addr was created
2570 * @dma_addr: The DMA address to check
2572 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2575 return dev->dma_ops->mapping_error(dev, dma_addr);
2576 return dma_mapping_error(dev->dma_device, dma_addr);
2580 * ib_dma_map_single - Map a kernel virtual address to DMA address
2581 * @dev: The device for which the dma_addr is to be created
2582 * @cpu_addr: The kernel virtual address
2583 * @size: The size of the region in bytes
2584 * @direction: The direction of the DMA
2586 static inline u64 ib_dma_map_single(struct ib_device *dev,
2587 void *cpu_addr, size_t size,
2588 enum dma_data_direction direction)
2591 return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2592 return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2596 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2597 * @dev: The device for which the DMA address was created
2598 * @addr: The DMA address
2599 * @size: The size of the region in bytes
2600 * @direction: The direction of the DMA
2602 static inline void ib_dma_unmap_single(struct ib_device *dev,
2603 u64 addr, size_t size,
2604 enum dma_data_direction direction)
2607 dev->dma_ops->unmap_single(dev, addr, size, direction);
2609 dma_unmap_single(dev->dma_device, addr, size, direction);
2612 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2613 void *cpu_addr, size_t size,
2614 enum dma_data_direction direction,
2615 struct dma_attrs *attrs)
2617 return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2621 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2622 u64 addr, size_t size,
2623 enum dma_data_direction direction,
2624 struct dma_attrs *attrs)
2626 return dma_unmap_single_attrs(dev->dma_device, addr, size,
2631 * ib_dma_map_page - Map a physical page to DMA address
2632 * @dev: The device for which the dma_addr is to be created
2633 * @page: The page to be mapped
2634 * @offset: The offset within the page
2635 * @size: The size of the region in bytes
2636 * @direction: The direction of the DMA
2638 static inline u64 ib_dma_map_page(struct ib_device *dev,
2640 unsigned long offset,
2642 enum dma_data_direction direction)
2645 return dev->dma_ops->map_page(dev, page, offset, size, direction);
2646 return dma_map_page(dev->dma_device, page, offset, size, direction);
2650 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2651 * @dev: The device for which the DMA address was created
2652 * @addr: The DMA address
2653 * @size: The size of the region in bytes
2654 * @direction: The direction of the DMA
2656 static inline void ib_dma_unmap_page(struct ib_device *dev,
2657 u64 addr, size_t size,
2658 enum dma_data_direction direction)
2661 dev->dma_ops->unmap_page(dev, addr, size, direction);
2663 dma_unmap_page(dev->dma_device, addr, size, direction);
2667 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2668 * @dev: The device for which the DMA addresses are to be created
2669 * @sg: The array of scatter/gather entries
2670 * @nents: The number of scatter/gather entries
2671 * @direction: The direction of the DMA
2673 static inline int ib_dma_map_sg(struct ib_device *dev,
2674 struct scatterlist *sg, int nents,
2675 enum dma_data_direction direction)
2678 return dev->dma_ops->map_sg(dev, sg, nents, direction);
2679 return dma_map_sg(dev->dma_device, sg, nents, direction);
2683 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2684 * @dev: The device for which the DMA addresses were created
2685 * @sg: The array of scatter/gather entries
2686 * @nents: The number of scatter/gather entries
2687 * @direction: The direction of the DMA
2689 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2690 struct scatterlist *sg, int nents,
2691 enum dma_data_direction direction)
2694 dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2696 dma_unmap_sg(dev->dma_device, sg, nents, direction);
2699 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2700 struct scatterlist *sg, int nents,
2701 enum dma_data_direction direction,
2702 struct dma_attrs *attrs)
2704 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2707 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2708 struct scatterlist *sg, int nents,
2709 enum dma_data_direction direction,
2710 struct dma_attrs *attrs)
2712 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2715 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2716 * @dev: The device for which the DMA addresses were created
2717 * @sg: The scatter/gather entry
2719 * Note: this function is obsolete. To do: change all occurrences of
2720 * ib_sg_dma_address() into sg_dma_address().
2722 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2723 struct scatterlist *sg)
2725 return sg_dma_address(sg);
2729 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2730 * @dev: The device for which the DMA addresses were created
2731 * @sg: The scatter/gather entry
2733 * Note: this function is obsolete. To do: change all occurrences of
2734 * ib_sg_dma_len() into sg_dma_len().
2736 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2737 struct scatterlist *sg)
2739 return sg_dma_len(sg);
2743 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2744 * @dev: The device for which the DMA address was created
2745 * @addr: The DMA address
2746 * @size: The size of the region in bytes
2747 * @dir: The direction of the DMA
2749 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2752 enum dma_data_direction dir)
2755 dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2757 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2761 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2762 * @dev: The device for which the DMA address was created
2763 * @addr: The DMA address
2764 * @size: The size of the region in bytes
2765 * @dir: The direction of the DMA
2767 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2770 enum dma_data_direction dir)
2773 dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2775 dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2779 * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2780 * @dev: The device for which the DMA address is requested
2781 * @size: The size of the region to allocate in bytes
2782 * @dma_handle: A pointer for returning the DMA address of the region
2783 * @flag: memory allocator flags
2785 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2791 return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2796 ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2797 *dma_handle = handle;
2803 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2804 * @dev: The device for which the DMA addresses were allocated
2805 * @size: The size of the region
2806 * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2807 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2809 static inline void ib_dma_free_coherent(struct ib_device *dev,
2810 size_t size, void *cpu_addr,
2814 dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2816 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2820 * ib_reg_phys_mr - Prepares a virtually addressed memory region for use
2822 * @pd: The protection domain associated assigned to the registered region.
2823 * @phys_buf_array: Specifies a list of physical buffers to use in the
2825 * @num_phys_buf: Specifies the size of the phys_buf_array.
2826 * @mr_access_flags: Specifies the memory access rights.
2827 * @iova_start: The offset of the region's starting I/O virtual address.
2829 struct ib_mr *ib_reg_phys_mr(struct ib_pd *pd,
2830 struct ib_phys_buf *phys_buf_array,
2832 int mr_access_flags,
2836 * ib_rereg_phys_mr - Modifies the attributes of an existing memory region.
2837 * Conceptually, this call performs the functions deregister memory region
2838 * followed by register physical memory region. Where possible,
2839 * resources are reused instead of deallocated and reallocated.
2840 * @mr: The memory region to modify.
2841 * @mr_rereg_mask: A bit-mask used to indicate which of the following
2842 * properties of the memory region are being modified.
2843 * @pd: If %IB_MR_REREG_PD is set in mr_rereg_mask, this field specifies
2844 * the new protection domain to associated with the memory region,
2845 * otherwise, this parameter is ignored.
2846 * @phys_buf_array: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2847 * field specifies a list of physical buffers to use in the new
2848 * translation, otherwise, this parameter is ignored.
2849 * @num_phys_buf: If %IB_MR_REREG_TRANS is set in mr_rereg_mask, this
2850 * field specifies the size of the phys_buf_array, otherwise, this
2851 * parameter is ignored.
2852 * @mr_access_flags: If %IB_MR_REREG_ACCESS is set in mr_rereg_mask, this
2853 * field specifies the new memory access rights, otherwise, this
2854 * parameter is ignored.
2855 * @iova_start: The offset of the region's starting I/O virtual address.
2857 int ib_rereg_phys_mr(struct ib_mr *mr,
2860 struct ib_phys_buf *phys_buf_array,
2862 int mr_access_flags,
2866 * ib_query_mr - Retrieves information about a specific memory region.
2867 * @mr: The memory region to retrieve information about.
2868 * @mr_attr: The attributes of the specified memory region.
2870 int ib_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
2873 * ib_dereg_mr - Deregisters a memory region and removes it from the
2874 * HCA translation table.
2875 * @mr: The memory region to deregister.
2877 * This function can fail, if the memory region has memory windows bound to it.
2879 int ib_dereg_mr(struct ib_mr *mr);
2881 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
2882 enum ib_mr_type mr_type,
2886 * ib_alloc_fast_reg_page_list - Allocates a page list array
2887 * @device - ib device pointer.
2888 * @page_list_len - size of the page list array to be allocated.
2890 * This allocates and returns a struct ib_fast_reg_page_list * and a
2891 * page_list array that is at least page_list_len in size. The actual
2892 * size is returned in max_page_list_len. The caller is responsible
2893 * for initializing the contents of the page_list array before posting
2894 * a send work request with the IB_WC_FAST_REG_MR opcode.
2896 * The page_list array entries must be translated using one of the
2897 * ib_dma_*() functions just like the addresses passed to
2898 * ib_map_phys_fmr(). Once the ib_post_send() is issued, the struct
2899 * ib_fast_reg_page_list must not be modified by the caller until the
2900 * IB_WC_FAST_REG_MR work request completes.
2902 struct ib_fast_reg_page_list *ib_alloc_fast_reg_page_list(
2903 struct ib_device *device, int page_list_len);
2906 * ib_free_fast_reg_page_list - Deallocates a previously allocated
2908 * @page_list - struct ib_fast_reg_page_list pointer to be deallocated.
2910 void ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
2913 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2915 * @mr - struct ib_mr pointer to be updated.
2916 * @newkey - new key to be used.
2918 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2920 mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2921 mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2925 * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2926 * for calculating a new rkey for type 2 memory windows.
2927 * @rkey - the rkey to increment.
2929 static inline u32 ib_inc_rkey(u32 rkey)
2931 const u32 mask = 0x000000ff;
2932 return ((rkey + 1) & mask) | (rkey & ~mask);
2936 * ib_alloc_mw - Allocates a memory window.
2937 * @pd: The protection domain associated with the memory window.
2938 * @type: The type of the memory window (1 or 2).
2940 struct ib_mw *ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
2943 * ib_bind_mw - Posts a work request to the send queue of the specified
2944 * QP, which binds the memory window to the given address range and
2945 * remote access attributes.
2946 * @qp: QP to post the bind work request on.
2947 * @mw: The memory window to bind.
2948 * @mw_bind: Specifies information about the memory window, including
2949 * its address range, remote access rights, and associated memory region.
2951 * If there is no immediate error, the function will update the rkey member
2952 * of the mw parameter to its new value. The bind operation can still fail
2955 static inline int ib_bind_mw(struct ib_qp *qp,
2957 struct ib_mw_bind *mw_bind)
2959 /* XXX reference counting in corresponding MR? */
2960 return mw->device->bind_mw ?
2961 mw->device->bind_mw(qp, mw, mw_bind) :
2966 * ib_dealloc_mw - Deallocates a memory window.
2967 * @mw: The memory window to deallocate.
2969 int ib_dealloc_mw(struct ib_mw *mw);
2972 * ib_alloc_fmr - Allocates a unmapped fast memory region.
2973 * @pd: The protection domain associated with the unmapped region.
2974 * @mr_access_flags: Specifies the memory access rights.
2975 * @fmr_attr: Attributes of the unmapped region.
2977 * A fast memory region must be mapped before it can be used as part of
2980 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2981 int mr_access_flags,
2982 struct ib_fmr_attr *fmr_attr);
2985 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2986 * @fmr: The fast memory region to associate with the pages.
2987 * @page_list: An array of physical pages to map to the fast memory region.
2988 * @list_len: The number of pages in page_list.
2989 * @iova: The I/O virtual address to use with the mapped region.
2991 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2992 u64 *page_list, int list_len,
2995 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2999 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3000 * @fmr_list: A linked list of fast memory regions to unmap.
3002 int ib_unmap_fmr(struct list_head *fmr_list);
3005 * ib_dealloc_fmr - Deallocates a fast memory region.
3006 * @fmr: The fast memory region to deallocate.
3008 int ib_dealloc_fmr(struct ib_fmr *fmr);
3011 * ib_attach_mcast - Attaches the specified QP to a multicast group.
3012 * @qp: QP to attach to the multicast group. The QP must be type
3014 * @gid: Multicast group GID.
3015 * @lid: Multicast group LID in host byte order.
3017 * In order to send and receive multicast packets, subnet
3018 * administration must have created the multicast group and configured
3019 * the fabric appropriately. The port associated with the specified
3020 * QP must also be a member of the multicast group.
3022 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3025 * ib_detach_mcast - Detaches the specified QP from a multicast group.
3026 * @qp: QP to detach from the multicast group.
3027 * @gid: Multicast group GID.
3028 * @lid: Multicast group LID in host byte order.
3030 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3033 * ib_alloc_xrcd - Allocates an XRC domain.
3034 * @device: The device on which to allocate the XRC domain.
3036 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3039 * ib_dealloc_xrcd - Deallocates an XRC domain.
3040 * @xrcd: The XRC domain to deallocate.
3042 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3044 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3045 struct ib_flow_attr *flow_attr, int domain);
3046 int ib_destroy_flow(struct ib_flow *flow_id);
3048 static inline int ib_check_mr_access(int flags)
3051 * Local write permission is required if remote write or
3052 * remote atomic permission is also requested.
3054 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3055 !(flags & IB_ACCESS_LOCAL_WRITE))
3062 * ib_check_mr_status: lightweight check of MR status.
3063 * This routine may provide status checks on a selected
3064 * ib_mr. first use is for signature status check.
3066 * @mr: A memory region.
3067 * @check_mask: Bitmask of which checks to perform from
3068 * ib_mr_status_check enumeration.
3069 * @mr_status: The container of relevant status checks.
3070 * failed checks will be indicated in the status bitmask
3071 * and the relevant info shall be in the error item.
3073 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3074 struct ib_mr_status *mr_status);
3076 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3077 u16 pkey, const union ib_gid *gid,
3078 const struct sockaddr *addr);
3080 #endif /* IB_VERBS_H */